xref: /netbsd-src/sys/dev/pci/if_bge.c (revision ccd9df534e375a4366c5b55f23782053c7a98d82)
1 /*	$NetBSD: if_bge.c,v 1.393 2024/07/05 04:31:51 rin Exp $	*/
2 
3 /*
4  * Copyright (c) 2001 Wind River Systems
5  * Copyright (c) 1997, 1998, 1999, 2001
6  *	Bill Paul <wpaul@windriver.com>.  All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. All advertising materials mentioning features or use of this software
17  *    must display the following acknowledgement:
18  *	This product includes software developed by Bill Paul.
19  * 4. Neither the name of the author nor the names of any co-contributors
20  *    may be used to endorse or promote products derived from this software
21  *    without specific prior written permission.
22  *
23  * THIS SOFTWARE IS PROVIDED BY Bill Paul AND CONTRIBUTORS ``AS IS'' AND
24  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26  * ARE DISCLAIMED.  IN NO EVENT SHALL Bill Paul OR THE VOICES IN HIS HEAD
27  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
28  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
29  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
30  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
31  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
32  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
33  * THE POSSIBILITY OF SUCH DAMAGE.
34  *
35  * $FreeBSD: if_bge.c,v 1.13 2002/04/04 06:01:31 wpaul Exp $
36  */
37 
38 /*
39  * Broadcom BCM570x family gigabit ethernet driver for NetBSD.
40  *
41  * NetBSD version by:
42  *
43  *	Frank van der Linden <fvdl@wasabisystems.com>
44  *	Jason Thorpe <thorpej@wasabisystems.com>
45  *	Jonathan Stone <jonathan@dsg.stanford.edu>
46  *
47  * Originally written for FreeBSD by Bill Paul <wpaul@windriver.com>
48  * Senior Engineer, Wind River Systems
49  */
50 
51 /*
52  * The Broadcom BCM5700 is based on technology originally developed by
53  * Alteon Networks as part of the Tigon I and Tigon II gigabit ethernet
54  * MAC chips. The BCM5700, sometimes referred to as the Tigon III, has
55  * two on-board MIPS R4000 CPUs and can have as much as 16MB of external
56  * SSRAM. The BCM5700 supports TCP, UDP and IP checksum offload, jumbo
57  * frames, highly configurable RX filtering, and 16 RX and TX queues
58  * (which, along with RX filter rules, can be used for QOS applications).
59  * Other features, such as TCP segmentation, may be available as part
60  * of value-added firmware updates. Unlike the Tigon I and Tigon II,
61  * firmware images can be stored in hardware and need not be compiled
62  * into the driver.
63  *
64  * The BCM5700 supports the PCI v2.2 and PCI-X v1.0 standards, and will
65  * function in a 32-bit/64-bit 33/66MHz bus, or a 64-bit/133MHz bus.
66  *
67  * The BCM5701 is a single-chip solution incorporating both the BCM5700
68  * MAC and a BCM5401 10/100/1000 PHY. Unlike the BCM5700, the BCM5701
69  * does not support external SSRAM.
70  *
71  * Broadcom also produces a variation of the BCM5700 under the "Altima"
72  * brand name, which is functionally similar but lacks PCI-X support.
73  *
74  * Without external SSRAM, you can only have at most 4 TX rings,
75  * and the use of the mini RX ring is disabled. This seems to imply
76  * that these features are simply not available on the BCM5701. As a
77  * result, this driver does not implement any support for the mini RX
78  * ring.
79  */
80 
81 #include <sys/cdefs.h>
82 __KERNEL_RCSID(0, "$NetBSD: if_bge.c,v 1.393 2024/07/05 04:31:51 rin Exp $");
83 
84 #include <sys/param.h>
85 #include <sys/types.h>
86 
87 #include <sys/callout.h>
88 #include <sys/device.h>
89 #include <sys/kernel.h>
90 #include <sys/kmem.h>
91 #include <sys/mbuf.h>
92 #include <sys/rndsource.h>
93 #include <sys/socket.h>
94 #include <sys/sockio.h>
95 #include <sys/sysctl.h>
96 #include <sys/systm.h>
97 
98 #include <net/if.h>
99 #include <net/if_dl.h>
100 #include <net/if_media.h>
101 #include <net/if_ether.h>
102 #include <net/bpf.h>
103 
104 /* Headers for TCP Segmentation Offload (TSO) */
105 #include <netinet/in_systm.h>		/* n_time for <netinet/ip.h>... */
106 #include <netinet/in.h>			/* ip_{src,dst}, for <netinet/ip.h> */
107 #include <netinet/ip.h>			/* for struct ip */
108 #include <netinet/tcp.h>		/* for struct tcphdr */
109 
110 #include <dev/pci/pcireg.h>
111 #include <dev/pci/pcivar.h>
112 #include <dev/pci/pcidevs.h>
113 
114 #include <dev/mii/mii.h>
115 #include <dev/mii/miivar.h>
116 #include <dev/mii/miidevs.h>
117 #include <dev/mii/brgphyreg.h>
118 
119 #include <dev/pci/if_bgereg.h>
120 #include <dev/pci/if_bgevar.h>
121 
122 #include <prop/proplib.h>
123 
124 #define ETHER_MIN_NOPAD (ETHER_MIN_LEN - ETHER_CRC_LEN) /* i.e., 60 */
125 
126 
127 /*
128  * Tunable thresholds for rx-side bge interrupt mitigation.
129  */
130 
131 /*
132  * The pairs of values below were obtained from empirical measurement
133  * on bcm5700 rev B2; they ar designed to give roughly 1 receive
134  * interrupt for every N packets received, where N is, approximately,
135  * the second value (rx_max_bds) in each pair.  The values are chosen
136  * such that moving from one pair to the succeeding pair was observed
137  * to roughly halve interrupt rate under sustained input packet load.
138  * The values were empirically chosen to avoid overflowing internal
139  * limits on the  bcm5700: increasing rx_ticks much beyond 600
140  * results in internal wrapping and higher interrupt rates.
141  * The limit of 46 frames was chosen to match NFS workloads.
142  *
143  * These values also work well on bcm5701, bcm5704C, and (less
144  * tested) bcm5703.  On other chipsets, (including the Altima chip
145  * family), the larger values may overflow internal chip limits,
146  * leading to increasing interrupt rates rather than lower interrupt
147  * rates.
148  *
149  * Applications using heavy interrupt mitigation (interrupting every
150  * 32 or 46 frames) in both directions may need to increase the TCP
151  * windowsize to above 131072 bytes (e.g., to 199608 bytes) to sustain
152  * full link bandwidth, due to ACKs and window updates lingering
153  * in the RX queue during the 30-to-40-frame interrupt-mitigation window.
154  */
155 static const struct bge_load_rx_thresh {
156 	int rx_ticks;
157 	int rx_max_bds; }
158 bge_rx_threshes[] = {
159 	{ 16,	1 },	/* rx_max_bds = 1 disables interrupt mitigation */
160 	{ 32,	2 },
161 	{ 50,	4 },
162 	{ 100,	8 },
163 	{ 192, 16 },
164 	{ 416, 32 },
165 	{ 598, 46 }
166 };
167 #define NBGE_RX_THRESH (sizeof(bge_rx_threshes) / sizeof(bge_rx_threshes[0]))
168 
169 /* XXX patchable; should be sysctl'able */
170 static int bge_auto_thresh = 1;
171 static int bge_rx_thresh_lvl;
172 
173 static int bge_rxthresh_nodenum;
174 
175 typedef int (*bge_eaddr_fcn_t)(struct bge_softc *, uint8_t[]);
176 
177 static uint32_t bge_chipid(const struct pci_attach_args *);
178 static int bge_can_use_msi(struct bge_softc *);
179 static int bge_probe(device_t, cfdata_t, void *);
180 static void bge_attach(device_t, device_t, void *);
181 static int bge_detach(device_t, int);
182 static void bge_release_resources(struct bge_softc *);
183 
184 static int bge_get_eaddr_fw(struct bge_softc *, uint8_t[]);
185 static int bge_get_eaddr_mem(struct bge_softc *, uint8_t[]);
186 static int bge_get_eaddr_nvram(struct bge_softc *, uint8_t[]);
187 static int bge_get_eaddr_eeprom(struct bge_softc *, uint8_t[]);
188 static int bge_get_eaddr(struct bge_softc *, uint8_t[]);
189 
190 static void bge_txeof(struct bge_softc *);
191 static void bge_rxcsum(struct bge_softc *, struct bge_rx_bd *, struct mbuf *);
192 static void bge_rxeof(struct bge_softc *);
193 
194 static void bge_asf_driver_up (struct bge_softc *);
195 static void bge_tick(void *);
196 static void bge_stats_update(struct bge_softc *);
197 static void bge_stats_update_regs(struct bge_softc *);
198 static int bge_encap(struct bge_softc *, struct mbuf *, uint32_t *);
199 
200 static int bge_intr(void *);
201 static void bge_start(struct ifnet *);
202 static void bge_start_locked(struct ifnet *);
203 static int bge_ifflags_cb(struct ethercom *);
204 static int bge_ioctl(struct ifnet *, u_long, void *);
205 static int bge_init(struct ifnet *);
206 static int bge_init_locked(struct ifnet *);
207 static void bge_stop(struct ifnet *, int);
208 static void bge_stop_locked(struct ifnet *, bool);
209 static bool bge_watchdog_tick(struct ifnet *);
210 static int bge_ifmedia_upd(struct ifnet *);
211 static void bge_ifmedia_sts(struct ifnet *, struct ifmediareq *);
212 static void bge_handle_reset_work(struct work *, void *);
213 
214 static uint8_t bge_nvram_getbyte(struct bge_softc *, int, uint8_t *);
215 static int bge_read_nvram(struct bge_softc *, uint8_t *, int, int);
216 
217 static uint8_t bge_eeprom_getbyte(struct bge_softc *, int, uint8_t *);
218 static int bge_read_eeprom(struct bge_softc *, void *, int, int);
219 static void bge_setmulti(struct bge_softc *);
220 
221 static void bge_handle_events(struct bge_softc *);
222 static int bge_alloc_jumbo_mem(struct bge_softc *);
223 static void bge_free_jumbo_mem(struct bge_softc *);
224 static void *bge_jalloc(struct bge_softc *);
225 static void bge_jfree(struct mbuf *, void *, size_t, void *);
226 static int bge_newbuf_jumbo(struct bge_softc *, int, struct mbuf *);
227 static int bge_init_rx_ring_jumbo(struct bge_softc *);
228 static void bge_free_rx_ring_jumbo(struct bge_softc *);
229 
230 static int bge_newbuf_std(struct bge_softc *, int);
231 static int bge_init_rx_ring_std(struct bge_softc *);
232 static void bge_fill_rx_ring_std(struct bge_softc *);
233 static void bge_free_rx_ring_std(struct bge_softc *m);
234 
235 static void bge_free_tx_ring(struct bge_softc *m, bool);
236 static int bge_init_tx_ring(struct bge_softc *);
237 
238 static int bge_chipinit(struct bge_softc *);
239 static int bge_blockinit(struct bge_softc *);
240 static int bge_phy_addr(struct bge_softc *);
241 static uint32_t bge_readmem_ind(struct bge_softc *, int);
242 static void bge_writemem_ind(struct bge_softc *, int, int);
243 static void bge_writembx(struct bge_softc *, int, int);
244 static void bge_writembx_flush(struct bge_softc *, int, int);
245 static void bge_writemem_direct(struct bge_softc *, int, int);
246 static void bge_writereg_ind(struct bge_softc *, int, int);
247 static void bge_set_max_readrq(struct bge_softc *);
248 
249 static int bge_miibus_readreg(device_t, int, int, uint16_t *);
250 static int bge_miibus_writereg(device_t, int, int, uint16_t);
251 static void bge_miibus_statchg(struct ifnet *);
252 
253 #define BGE_RESET_SHUTDOWN	0
254 #define	BGE_RESET_START		1
255 #define	BGE_RESET_SUSPEND	2
256 static void bge_sig_post_reset(struct bge_softc *, int);
257 static void bge_sig_legacy(struct bge_softc *, int);
258 static void bge_sig_pre_reset(struct bge_softc *, int);
259 static void bge_wait_for_event_ack(struct bge_softc *);
260 static void bge_stop_fw(struct bge_softc *);
261 static int bge_reset(struct bge_softc *);
262 static void bge_link_upd(struct bge_softc *);
263 static void bge_sysctl_init(struct bge_softc *);
264 static int bge_sysctl_verify(SYSCTLFN_PROTO);
265 
266 static void bge_ape_lock_init(struct bge_softc *);
267 static void bge_ape_read_fw_ver(struct bge_softc *);
268 static int bge_ape_lock(struct bge_softc *, int);
269 static void bge_ape_unlock(struct bge_softc *, int);
270 static void bge_ape_send_event(struct bge_softc *, uint32_t);
271 static void bge_ape_driver_state_change(struct bge_softc *, int);
272 
273 #ifdef BGE_DEBUG
274 #define DPRINTF(x)	if (bgedebug) printf x
275 #define DPRINTFN(n, x)	if (bgedebug >= (n)) printf x
276 #define BGE_TSO_PRINTF(x)  do { if (bge_tso_debug) printf x ;} while (0)
277 int	bgedebug = 0;
278 int	bge_tso_debug = 0;
279 void	bge_debug_info(struct bge_softc *);
280 #else
281 #define DPRINTF(x)
282 #define DPRINTFN(n, x)
283 #define BGE_TSO_PRINTF(x)
284 #endif
285 
286 #ifdef BGE_EVENT_COUNTERS
287 #define	BGE_EVCNT_INCR(ev)	(ev).ev_count++
288 #define	BGE_EVCNT_ADD(ev, val)	(ev).ev_count += (val)
289 #define	BGE_EVCNT_UPD(ev, val)	(ev).ev_count = (val)
290 #else
291 #define	BGE_EVCNT_INCR(ev)	/* nothing */
292 #define	BGE_EVCNT_ADD(ev, val)	/* nothing */
293 #define	BGE_EVCNT_UPD(ev, val)	/* nothing */
294 #endif
295 
296 #define VIDDID(a, b) PCI_VENDOR_ ## a, PCI_PRODUCT_ ## a ## _ ## b
297 /*
298  * The BCM5700 documentation seems to indicate that the hardware still has the
299  * Alteon vendor ID burned into it, though it should always be overridden by
300  * the value in the EEPROM.  We'll check for it anyway.
301  */
302 static const struct bge_product {
303 	pci_vendor_id_t		bp_vendor;
304 	pci_product_id_t	bp_product;
305 	const char		*bp_name;
306 } bge_products[] = {
307 	{ VIDDID(ALTEON,   BCM5700),	"Broadcom BCM5700 Gigabit" },
308 	{ VIDDID(ALTEON,   BCM5701),	"Broadcom BCM5701 Gigabit" },
309 	{ VIDDID(ALTIMA,   AC1000),	"Altima AC1000 Gigabit" },
310 	{ VIDDID(ALTIMA,   AC1001),	"Altima AC1001 Gigabit" },
311 	{ VIDDID(ALTIMA,   AC1003),	"Altima AC1003 Gigabit" },
312 	{ VIDDID(ALTIMA,   AC9100),	"Altima AC9100 Gigabit" },
313 	{ VIDDID(APPLE,	   BCM5701),	"APPLE BCM5701 Gigabit" },
314 	{ VIDDID(BROADCOM, BCM5700),	"Broadcom BCM5700 Gigabit" },
315 	{ VIDDID(BROADCOM, BCM5701),	"Broadcom BCM5701 Gigabit" },
316 	{ VIDDID(BROADCOM, BCM5702),	"Broadcom BCM5702 Gigabit" },
317 	{ VIDDID(BROADCOM, BCM5702FE),	"Broadcom BCM5702FE Fast" },
318 	{ VIDDID(BROADCOM, BCM5702X),	"Broadcom BCM5702X Gigabit" },
319 	{ VIDDID(BROADCOM, BCM5703),	"Broadcom BCM5703 Gigabit" },
320 	{ VIDDID(BROADCOM, BCM5703X),	"Broadcom BCM5703X Gigabit" },
321 	{ VIDDID(BROADCOM, BCM5703_ALT),"Broadcom BCM5703 Gigabit" },
322 	{ VIDDID(BROADCOM, BCM5704C),	"Broadcom BCM5704C Dual Gigabit" },
323 	{ VIDDID(BROADCOM, BCM5704S),	"Broadcom BCM5704S Dual Gigabit" },
324 	{ VIDDID(BROADCOM, BCM5704S_ALT),"Broadcom BCM5704S Dual Gigabit" },
325 	{ VIDDID(BROADCOM, BCM5705),	"Broadcom BCM5705 Gigabit" },
326 	{ VIDDID(BROADCOM, BCM5705F),	"Broadcom BCM5705F Gigabit" },
327 	{ VIDDID(BROADCOM, BCM5705K),	"Broadcom BCM5705K Gigabit" },
328 	{ VIDDID(BROADCOM, BCM5705M),	"Broadcom BCM5705M Gigabit" },
329 	{ VIDDID(BROADCOM, BCM5705M_ALT),"Broadcom BCM5705M Gigabit" },
330 	{ VIDDID(BROADCOM, BCM5714),	"Broadcom BCM5714 Gigabit" },
331 	{ VIDDID(BROADCOM, BCM5714S),	"Broadcom BCM5714S Gigabit" },
332 	{ VIDDID(BROADCOM, BCM5715),	"Broadcom BCM5715 Gigabit" },
333 	{ VIDDID(BROADCOM, BCM5715S),	"Broadcom BCM5715S Gigabit" },
334 	{ VIDDID(BROADCOM, BCM5717),	"Broadcom BCM5717 Gigabit" },
335 	{ VIDDID(BROADCOM, BCM5717C),	"Broadcom BCM5717 Gigabit" },
336 	{ VIDDID(BROADCOM, BCM5718),	"Broadcom BCM5718 Gigabit" },
337 	{ VIDDID(BROADCOM, BCM5719),	"Broadcom BCM5719 Gigabit" },
338 	{ VIDDID(BROADCOM, BCM5720),	"Broadcom BCM5720 Gigabit" },
339 	{ VIDDID(BROADCOM, BCM5721),	"Broadcom BCM5721 Gigabit" },
340 	{ VIDDID(BROADCOM, BCM5722),	"Broadcom BCM5722 Gigabit" },
341 	{ VIDDID(BROADCOM, BCM5723),	"Broadcom BCM5723 Gigabit" },
342 	{ VIDDID(BROADCOM, BCM5725),	"Broadcom BCM5725 Gigabit" },
343 	{ VIDDID(BROADCOM, BCM5727),	"Broadcom BCM5727 Gigabit" },
344 	{ VIDDID(BROADCOM, BCM5750),	"Broadcom BCM5750 Gigabit" },
345 	{ VIDDID(BROADCOM, BCM5751),	"Broadcom BCM5751 Gigabit" },
346 	{ VIDDID(BROADCOM, BCM5751F),	"Broadcom BCM5751F Gigabit" },
347 	{ VIDDID(BROADCOM, BCM5751M),	"Broadcom BCM5751M Gigabit" },
348 	{ VIDDID(BROADCOM, BCM5752),	"Broadcom BCM5752 Gigabit" },
349 	{ VIDDID(BROADCOM, BCM5752M),	"Broadcom BCM5752M Gigabit" },
350 	{ VIDDID(BROADCOM, BCM5753),	"Broadcom BCM5753 Gigabit" },
351 	{ VIDDID(BROADCOM, BCM5753F),	"Broadcom BCM5753F Gigabit" },
352 	{ VIDDID(BROADCOM, BCM5753M),	"Broadcom BCM5753M Gigabit" },
353 	{ VIDDID(BROADCOM, BCM5754),	"Broadcom BCM5754 Gigabit" },
354 	{ VIDDID(BROADCOM, BCM5754M),	"Broadcom BCM5754M Gigabit" },
355 	{ VIDDID(BROADCOM, BCM5755),	"Broadcom BCM5755 Gigabit" },
356 	{ VIDDID(BROADCOM, BCM5755M),	"Broadcom BCM5755M Gigabit" },
357 	{ VIDDID(BROADCOM, BCM5756),	"Broadcom BCM5756 Gigabit" },
358 	{ VIDDID(BROADCOM, BCM5761),	"Broadcom BCM5761 Gigabit" },
359 	{ VIDDID(BROADCOM, BCM5761E),	"Broadcom BCM5761E Gigabit" },
360 	{ VIDDID(BROADCOM, BCM5761S),	"Broadcom BCM5761S Gigabit" },
361 	{ VIDDID(BROADCOM, BCM5761SE),	"Broadcom BCM5761SE Gigabit" },
362 	{ VIDDID(BROADCOM, BCM5762),	"Broadcom BCM5762 Gigabit" },
363 	{ VIDDID(BROADCOM, BCM5764),	"Broadcom BCM5764 Gigabit" },
364 	{ VIDDID(BROADCOM, BCM5780),	"Broadcom BCM5780 Gigabit" },
365 	{ VIDDID(BROADCOM, BCM5780S),	"Broadcom BCM5780S Gigabit" },
366 	{ VIDDID(BROADCOM, BCM5781),	"Broadcom BCM5781 Gigabit" },
367 	{ VIDDID(BROADCOM, BCM5782),	"Broadcom BCM5782 Gigabit" },
368 	{ VIDDID(BROADCOM, BCM5784M),	"BCM5784M NetLink 1000baseT" },
369 	{ VIDDID(BROADCOM, BCM5785F),	"BCM5785F NetLink 10/100" },
370 	{ VIDDID(BROADCOM, BCM5785G),	"BCM5785G NetLink 1000baseT" },
371 	{ VIDDID(BROADCOM, BCM5786),	"Broadcom BCM5786 Gigabit" },
372 	{ VIDDID(BROADCOM, BCM5787),	"Broadcom BCM5787 Gigabit" },
373 	{ VIDDID(BROADCOM, BCM5787F),	"Broadcom BCM5787F 10/100" },
374 	{ VIDDID(BROADCOM, BCM5787M),	"Broadcom BCM5787M Gigabit" },
375 	{ VIDDID(BROADCOM, BCM5788),	"Broadcom BCM5788 Gigabit" },
376 	{ VIDDID(BROADCOM, BCM5789),	"Broadcom BCM5789 Gigabit" },
377 	{ VIDDID(BROADCOM, BCM5901),	"Broadcom BCM5901 Fast" },
378 	{ VIDDID(BROADCOM, BCM5901A2),	"Broadcom BCM5901A2 Fast" },
379 	{ VIDDID(BROADCOM, BCM5903M),	"Broadcom BCM5903M Fast" },
380 	{ VIDDID(BROADCOM, BCM5906),	"Broadcom BCM5906 Fast" },
381 	{ VIDDID(BROADCOM, BCM5906M),	"Broadcom BCM5906M Fast" },
382 	{ VIDDID(BROADCOM, BCM57760),	"Broadcom BCM57760 Gigabit" },
383 	{ VIDDID(BROADCOM, BCM57761),	"Broadcom BCM57761 Gigabit" },
384 	{ VIDDID(BROADCOM, BCM57762),	"Broadcom BCM57762 Gigabit" },
385 	{ VIDDID(BROADCOM, BCM57764),	"Broadcom BCM57764 Gigabit" },
386 	{ VIDDID(BROADCOM, BCM57765),	"Broadcom BCM57765 Gigabit" },
387 	{ VIDDID(BROADCOM, BCM57766),	"Broadcom BCM57766 Gigabit" },
388 	{ VIDDID(BROADCOM, BCM57767),	"Broadcom BCM57767 Gigabit" },
389 	{ VIDDID(BROADCOM, BCM57780),	"Broadcom BCM57780 Gigabit" },
390 	{ VIDDID(BROADCOM, BCM57781),	"Broadcom BCM57781 Gigabit" },
391 	{ VIDDID(BROADCOM, BCM57782),	"Broadcom BCM57782 Gigabit" },
392 	{ VIDDID(BROADCOM, BCM57785),	"Broadcom BCM57785 Gigabit" },
393 	{ VIDDID(BROADCOM, BCM57786),	"Broadcom BCM57786 Gigabit" },
394 	{ VIDDID(BROADCOM, BCM57787),	"Broadcom BCM57787 Gigabit" },
395 	{ VIDDID(BROADCOM, BCM57788),	"Broadcom BCM57788 Gigabit" },
396 	{ VIDDID(BROADCOM, BCM57790),	"Broadcom BCM57790 Gigabit" },
397 	{ VIDDID(BROADCOM, BCM57791),	"Broadcom BCM57791 Gigabit" },
398 	{ VIDDID(BROADCOM, BCM57795),	"Broadcom BCM57795 Gigabit" },
399 	{ VIDDID(SCHNEIDERKOCH, SK_9DX1),"SysKonnect SK-9Dx1 Gigabit" },
400 	{ VIDDID(SCHNEIDERKOCH, SK_9MXX),"SysKonnect SK-9Mxx Gigabit" },
401 	{ VIDDID(3COM, 3C996),		"3Com 3c996 Gigabit" },
402 	{ VIDDID(FUJITSU4, PW008GE4),	"Fujitsu PW008GE4 Gigabit" },
403 	{ VIDDID(FUJITSU4, PW008GE5),	"Fujitsu PW008GE5 Gigabit" },
404 	{ VIDDID(FUJITSU4, PP250_450_LAN),"Fujitsu Primepower 250/450 Gigabit" },
405 	{ 0, 0, NULL },
406 };
407 
408 #define BGE_IS_JUMBO_CAPABLE(sc)	((sc)->bge_flags & BGEF_JUMBO_CAPABLE)
409 #define BGE_IS_5700_FAMILY(sc)		((sc)->bge_flags & BGEF_5700_FAMILY)
410 #define BGE_IS_5705_PLUS(sc)		((sc)->bge_flags & BGEF_5705_PLUS)
411 #define BGE_IS_5714_FAMILY(sc)		((sc)->bge_flags & BGEF_5714_FAMILY)
412 #define BGE_IS_575X_PLUS(sc)		((sc)->bge_flags & BGEF_575X_PLUS)
413 #define BGE_IS_5755_PLUS(sc)		((sc)->bge_flags & BGEF_5755_PLUS)
414 #define BGE_IS_57765_FAMILY(sc)		((sc)->bge_flags & BGEF_57765_FAMILY)
415 #define BGE_IS_57765_PLUS(sc)		((sc)->bge_flags & BGEF_57765_PLUS)
416 #define BGE_IS_5717_PLUS(sc)		((sc)->bge_flags & BGEF_5717_PLUS)
417 
418 static const struct bge_revision {
419 	uint32_t		br_chipid;
420 	const char		*br_name;
421 } bge_revisions[] = {
422 	{ BGE_CHIPID_BCM5700_A0, "BCM5700 A0" },
423 	{ BGE_CHIPID_BCM5700_A1, "BCM5700 A1" },
424 	{ BGE_CHIPID_BCM5700_B0, "BCM5700 B0" },
425 	{ BGE_CHIPID_BCM5700_B1, "BCM5700 B1" },
426 	{ BGE_CHIPID_BCM5700_B2, "BCM5700 B2" },
427 	{ BGE_CHIPID_BCM5700_B3, "BCM5700 B3" },
428 	{ BGE_CHIPID_BCM5700_ALTIMA, "BCM5700 Altima" },
429 	{ BGE_CHIPID_BCM5700_C0, "BCM5700 C0" },
430 	{ BGE_CHIPID_BCM5701_A0, "BCM5701 A0" },
431 	{ BGE_CHIPID_BCM5701_B0, "BCM5701 B0" },
432 	{ BGE_CHIPID_BCM5701_B2, "BCM5701 B2" },
433 	{ BGE_CHIPID_BCM5701_B5, "BCM5701 B5" },
434 	{ BGE_CHIPID_BCM5703_A0, "BCM5702/5703 A0" },
435 	{ BGE_CHIPID_BCM5703_A1, "BCM5702/5703 A1" },
436 	{ BGE_CHIPID_BCM5703_A2, "BCM5702/5703 A2" },
437 	{ BGE_CHIPID_BCM5703_A3, "BCM5702/5703 A3" },
438 	{ BGE_CHIPID_BCM5703_B0, "BCM5702/5703 B0" },
439 	{ BGE_CHIPID_BCM5704_A0, "BCM5704 A0" },
440 	{ BGE_CHIPID_BCM5704_A1, "BCM5704 A1" },
441 	{ BGE_CHIPID_BCM5704_A2, "BCM5704 A2" },
442 	{ BGE_CHIPID_BCM5704_A3, "BCM5704 A3" },
443 	{ BGE_CHIPID_BCM5704_B0, "BCM5704 B0" },
444 	{ BGE_CHIPID_BCM5705_A0, "BCM5705 A0" },
445 	{ BGE_CHIPID_BCM5705_A1, "BCM5705 A1" },
446 	{ BGE_CHIPID_BCM5705_A2, "BCM5705 A2" },
447 	{ BGE_CHIPID_BCM5705_A3, "BCM5705 A3" },
448 	{ BGE_CHIPID_BCM5750_A0, "BCM5750 A0" },
449 	{ BGE_CHIPID_BCM5750_A1, "BCM5750 A1" },
450 	{ BGE_CHIPID_BCM5750_A3, "BCM5750 A3" },
451 	{ BGE_CHIPID_BCM5750_B0, "BCM5750 B0" },
452 	{ BGE_CHIPID_BCM5750_B1, "BCM5750 B1" },
453 	{ BGE_CHIPID_BCM5750_C0, "BCM5750 C0" },
454 	{ BGE_CHIPID_BCM5750_C1, "BCM5750 C1" },
455 	{ BGE_CHIPID_BCM5750_C2, "BCM5750 C2" },
456 	{ BGE_CHIPID_BCM5752_A0, "BCM5752 A0" },
457 	{ BGE_CHIPID_BCM5752_A1, "BCM5752 A1" },
458 	{ BGE_CHIPID_BCM5752_A2, "BCM5752 A2" },
459 	{ BGE_CHIPID_BCM5714_A0, "BCM5714 A0" },
460 	{ BGE_CHIPID_BCM5714_B0, "BCM5714 B0" },
461 	{ BGE_CHIPID_BCM5714_B3, "BCM5714 B3" },
462 	{ BGE_CHIPID_BCM5715_A0, "BCM5715 A0" },
463 	{ BGE_CHIPID_BCM5715_A1, "BCM5715 A1" },
464 	{ BGE_CHIPID_BCM5715_A3, "BCM5715 A3" },
465 	{ BGE_CHIPID_BCM5717_A0, "BCM5717 A0" },
466 	{ BGE_CHIPID_BCM5717_B0, "BCM5717 B0" },
467 	{ BGE_CHIPID_BCM5719_A0, "BCM5719 A0" },
468 	{ BGE_CHIPID_BCM5720_A0, "BCM5720 A0" },
469 	{ BGE_CHIPID_BCM5755_A0, "BCM5755 A0" },
470 	{ BGE_CHIPID_BCM5755_A1, "BCM5755 A1" },
471 	{ BGE_CHIPID_BCM5755_A2, "BCM5755 A2" },
472 	{ BGE_CHIPID_BCM5755_C0, "BCM5755 C0" },
473 	{ BGE_CHIPID_BCM5761_A0, "BCM5761 A0" },
474 	{ BGE_CHIPID_BCM5761_A1, "BCM5761 A1" },
475 	{ BGE_CHIPID_BCM5762_A0, "BCM5762 A0" },
476 	{ BGE_CHIPID_BCM5762_B0, "BCM5762 B0" },
477 	{ BGE_CHIPID_BCM5784_A0, "BCM5784 A0" },
478 	{ BGE_CHIPID_BCM5784_A1, "BCM5784 A1" },
479 	{ BGE_CHIPID_BCM5784_B0, "BCM5784 B0" },
480 	/* 5754 and 5787 share the same ASIC ID */
481 	{ BGE_CHIPID_BCM5787_A0, "BCM5754/5787 A0" },
482 	{ BGE_CHIPID_BCM5787_A1, "BCM5754/5787 A1" },
483 	{ BGE_CHIPID_BCM5787_A2, "BCM5754/5787 A2" },
484 	{ BGE_CHIPID_BCM5906_A0, "BCM5906 A0" },
485 	{ BGE_CHIPID_BCM5906_A1, "BCM5906 A1" },
486 	{ BGE_CHIPID_BCM5906_A2, "BCM5906 A2" },
487 	{ BGE_CHIPID_BCM57765_A0, "BCM57765 A0" },
488 	{ BGE_CHIPID_BCM57765_B0, "BCM57765 B0" },
489 	{ BGE_CHIPID_BCM57766_A0, "BCM57766 A0" },
490 	{ BGE_CHIPID_BCM57780_A0, "BCM57780 A0" },
491 	{ BGE_CHIPID_BCM57780_A1, "BCM57780 A1" },
492 
493 	{ 0, NULL }
494 };
495 
496 /*
497  * Some defaults for major revisions, so that newer steppings
498  * that we don't know about have a shot at working.
499  */
500 static const struct bge_revision bge_majorrevs[] = {
501 	{ BGE_ASICREV_BCM5700, "unknown BCM5700" },
502 	{ BGE_ASICREV_BCM5701, "unknown BCM5701" },
503 	{ BGE_ASICREV_BCM5703, "unknown BCM5703" },
504 	{ BGE_ASICREV_BCM5704, "unknown BCM5704" },
505 	{ BGE_ASICREV_BCM5705, "unknown BCM5705" },
506 	{ BGE_ASICREV_BCM5750, "unknown BCM5750" },
507 	{ BGE_ASICREV_BCM5714, "unknown BCM5714" },
508 	{ BGE_ASICREV_BCM5714_A0, "unknown BCM5714" },
509 	{ BGE_ASICREV_BCM5752, "unknown BCM5752" },
510 	{ BGE_ASICREV_BCM5780, "unknown BCM5780" },
511 	{ BGE_ASICREV_BCM5755, "unknown BCM5755" },
512 	{ BGE_ASICREV_BCM5761, "unknown BCM5761" },
513 	{ BGE_ASICREV_BCM5784, "unknown BCM5784" },
514 	{ BGE_ASICREV_BCM5785, "unknown BCM5785" },
515 	/* 5754 and 5787 share the same ASIC ID */
516 	{ BGE_ASICREV_BCM5787, "unknown BCM5754/5787" },
517 	{ BGE_ASICREV_BCM5906, "unknown BCM5906" },
518 	{ BGE_ASICREV_BCM57765, "unknown BCM57765" },
519 	{ BGE_ASICREV_BCM57766, "unknown BCM57766" },
520 	{ BGE_ASICREV_BCM57780, "unknown BCM57780" },
521 	{ BGE_ASICREV_BCM5717, "unknown BCM5717" },
522 	{ BGE_ASICREV_BCM5719, "unknown BCM5719" },
523 	{ BGE_ASICREV_BCM5720, "unknown BCM5720" },
524 	{ BGE_ASICREV_BCM5762, "unknown BCM5762" },
525 
526 	{ 0, NULL }
527 };
528 
529 static int bge_allow_asf = 1;
530 
531 #ifndef BGE_WATCHDOG_TIMEOUT
532 #define BGE_WATCHDOG_TIMEOUT 5
533 #endif
534 static int bge_watchdog_timeout = BGE_WATCHDOG_TIMEOUT;
535 
536 
537 CFATTACH_DECL3_NEW(bge, sizeof(struct bge_softc),
538     bge_probe, bge_attach, bge_detach, NULL, NULL, NULL, DVF_DETACH_SHUTDOWN);
539 
540 static uint32_t
541 bge_readmem_ind(struct bge_softc *sc, int off)
542 {
543 	pcireg_t val;
544 
545 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906 &&
546 	    off >= BGE_STATS_BLOCK && off < BGE_SEND_RING_1_TO_4)
547 		return 0;
548 
549 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MEMWIN_BASEADDR, off);
550 	val = pci_conf_read(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MEMWIN_DATA);
551 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MEMWIN_BASEADDR, 0);
552 	return val;
553 }
554 
555 static void
556 bge_writemem_ind(struct bge_softc *sc, int off, int val)
557 {
558 
559 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MEMWIN_BASEADDR, off);
560 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MEMWIN_DATA, val);
561 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MEMWIN_BASEADDR, 0);
562 }
563 
564 /*
565  * PCI Express only
566  */
567 static void
568 bge_set_max_readrq(struct bge_softc *sc)
569 {
570 	pcireg_t val;
571 
572 	val = pci_conf_read(sc->sc_pc, sc->sc_pcitag, sc->bge_pciecap
573 	    + PCIE_DCSR);
574 	val &= ~PCIE_DCSR_MAX_READ_REQ;
575 	switch (sc->bge_expmrq) {
576 	case 2048:
577 		val |= BGE_PCIE_DEVCTL_MAX_READRQ_2048;
578 		break;
579 	case 4096:
580 		val |= BGE_PCIE_DEVCTL_MAX_READRQ_4096;
581 		break;
582 	default:
583 		panic("incorrect expmrq value(%d)", sc->bge_expmrq);
584 		break;
585 	}
586 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, sc->bge_pciecap
587 	    + PCIE_DCSR, val);
588 }
589 
590 #ifdef notdef
591 static uint32_t
592 bge_readreg_ind(struct bge_softc *sc, int off)
593 {
594 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_REG_BASEADDR, off);
595 	return pci_conf_read(sc->sc_pc, sc->sc_pcitag, BGE_PCI_REG_DATA);
596 }
597 #endif
598 
599 static void
600 bge_writereg_ind(struct bge_softc *sc, int off, int val)
601 {
602 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_REG_BASEADDR, off);
603 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_REG_DATA, val);
604 }
605 
606 static void
607 bge_writemem_direct(struct bge_softc *sc, int off, int val)
608 {
609 	CSR_WRITE_4(sc, off, val);
610 }
611 
612 static void
613 bge_writembx(struct bge_softc *sc, int off, int val)
614 {
615 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906)
616 		off += BGE_LPMBX_IRQ0_HI - BGE_MBX_IRQ0_HI;
617 
618 	CSR_WRITE_4(sc, off, val);
619 }
620 
621 static void
622 bge_writembx_flush(struct bge_softc *sc, int off, int val)
623 {
624 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906)
625 		off += BGE_LPMBX_IRQ0_HI - BGE_MBX_IRQ0_HI;
626 
627 	CSR_WRITE_4_FLUSH(sc, off, val);
628 }
629 
630 /*
631  * Clear all stale locks and select the lock for this driver instance.
632  */
633 void
634 bge_ape_lock_init(struct bge_softc *sc)
635 {
636 	struct pci_attach_args *pa = &(sc->bge_pa);
637 	uint32_t bit, regbase;
638 	int i;
639 
640 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5761)
641 		regbase = BGE_APE_LOCK_GRANT;
642 	else
643 		regbase = BGE_APE_PER_LOCK_GRANT;
644 
645 	/* Clear any stale locks. */
646 	for (i = BGE_APE_LOCK_PHY0; i <= BGE_APE_LOCK_GPIO; i++) {
647 		switch (i) {
648 		case BGE_APE_LOCK_PHY0:
649 		case BGE_APE_LOCK_PHY1:
650 		case BGE_APE_LOCK_PHY2:
651 		case BGE_APE_LOCK_PHY3:
652 			bit = BGE_APE_LOCK_GRANT_DRIVER0;
653 			break;
654 		default:
655 			if (pa->pa_function == 0)
656 				bit = BGE_APE_LOCK_GRANT_DRIVER0;
657 			else
658 				bit = (1 << pa->pa_function);
659 		}
660 		APE_WRITE_4(sc, regbase + 4 * i, bit);
661 	}
662 
663 	/* Select the PHY lock based on the device's function number. */
664 	switch (pa->pa_function) {
665 	case 0:
666 		sc->bge_phy_ape_lock = BGE_APE_LOCK_PHY0;
667 		break;
668 	case 1:
669 		sc->bge_phy_ape_lock = BGE_APE_LOCK_PHY1;
670 		break;
671 	case 2:
672 		sc->bge_phy_ape_lock = BGE_APE_LOCK_PHY2;
673 		break;
674 	case 3:
675 		sc->bge_phy_ape_lock = BGE_APE_LOCK_PHY3;
676 		break;
677 	default:
678 		printf("%s: PHY lock not supported on function\n",
679 		    device_xname(sc->bge_dev));
680 		break;
681 	}
682 }
683 
684 /*
685  * Check for APE firmware, set flags, and print version info.
686  */
687 void
688 bge_ape_read_fw_ver(struct bge_softc *sc)
689 {
690 	const char *fwtype;
691 	uint32_t apedata, features;
692 
693 	/* Check for a valid APE signature in shared memory. */
694 	apedata = APE_READ_4(sc, BGE_APE_SEG_SIG);
695 	if (apedata != BGE_APE_SEG_SIG_MAGIC) {
696 		sc->bge_mfw_flags &= ~ BGE_MFW_ON_APE;
697 		return;
698 	}
699 
700 	/* Check if APE firmware is running. */
701 	apedata = APE_READ_4(sc, BGE_APE_FW_STATUS);
702 	if ((apedata & BGE_APE_FW_STATUS_READY) == 0) {
703 		printf("%s: APE signature found but FW status not ready! "
704 		    "0x%08x\n", device_xname(sc->bge_dev), apedata);
705 		return;
706 	}
707 
708 	sc->bge_mfw_flags |= BGE_MFW_ON_APE;
709 
710 	/* Fetch the APE firmware type and version. */
711 	apedata = APE_READ_4(sc, BGE_APE_FW_VERSION);
712 	features = APE_READ_4(sc, BGE_APE_FW_FEATURES);
713 	if ((features & BGE_APE_FW_FEATURE_NCSI) != 0) {
714 		sc->bge_mfw_flags |= BGE_MFW_TYPE_NCSI;
715 		fwtype = "NCSI";
716 	} else if ((features & BGE_APE_FW_FEATURE_DASH) != 0) {
717 		sc->bge_mfw_flags |= BGE_MFW_TYPE_DASH;
718 		fwtype = "DASH";
719 	} else
720 		fwtype = "UNKN";
721 
722 	/* Print the APE firmware version. */
723 	aprint_normal_dev(sc->bge_dev, "APE firmware %s %d.%d.%d.%d\n", fwtype,
724 	    (apedata & BGE_APE_FW_VERSION_MAJMSK) >> BGE_APE_FW_VERSION_MAJSFT,
725 	    (apedata & BGE_APE_FW_VERSION_MINMSK) >> BGE_APE_FW_VERSION_MINSFT,
726 	    (apedata & BGE_APE_FW_VERSION_REVMSK) >> BGE_APE_FW_VERSION_REVSFT,
727 	    (apedata & BGE_APE_FW_VERSION_BLDMSK));
728 }
729 
730 int
731 bge_ape_lock(struct bge_softc *sc, int locknum)
732 {
733 	struct pci_attach_args *pa = &(sc->bge_pa);
734 	uint32_t bit, gnt, req, status;
735 	int i, off;
736 
737 	if ((sc->bge_mfw_flags & BGE_MFW_ON_APE) == 0)
738 		return 0;
739 
740 	/* Lock request/grant registers have different bases. */
741 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5761) {
742 		req = BGE_APE_LOCK_REQ;
743 		gnt = BGE_APE_LOCK_GRANT;
744 	} else {
745 		req = BGE_APE_PER_LOCK_REQ;
746 		gnt = BGE_APE_PER_LOCK_GRANT;
747 	}
748 
749 	off = 4 * locknum;
750 
751 	switch (locknum) {
752 	case BGE_APE_LOCK_GPIO:
753 		/* Lock required when using GPIO. */
754 		if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5761)
755 			return 0;
756 		if (pa->pa_function == 0)
757 			bit = BGE_APE_LOCK_REQ_DRIVER0;
758 		else
759 			bit = (1 << pa->pa_function);
760 		break;
761 	case BGE_APE_LOCK_GRC:
762 		/* Lock required to reset the device. */
763 		if (pa->pa_function == 0)
764 			bit = BGE_APE_LOCK_REQ_DRIVER0;
765 		else
766 			bit = (1 << pa->pa_function);
767 		break;
768 	case BGE_APE_LOCK_MEM:
769 		/* Lock required when accessing certain APE memory. */
770 		if (pa->pa_function == 0)
771 			bit = BGE_APE_LOCK_REQ_DRIVER0;
772 		else
773 			bit = (1 << pa->pa_function);
774 		break;
775 	case BGE_APE_LOCK_PHY0:
776 	case BGE_APE_LOCK_PHY1:
777 	case BGE_APE_LOCK_PHY2:
778 	case BGE_APE_LOCK_PHY3:
779 		/* Lock required when accessing PHYs. */
780 		bit = BGE_APE_LOCK_REQ_DRIVER0;
781 		break;
782 	default:
783 		return EINVAL;
784 	}
785 
786 	/* Request a lock. */
787 	APE_WRITE_4_FLUSH(sc, req + off, bit);
788 
789 	/* Wait up to 1 second to acquire lock. */
790 	for (i = 0; i < 20000; i++) {
791 		status = APE_READ_4(sc, gnt + off);
792 		if (status == bit)
793 			break;
794 		DELAY(50);
795 	}
796 
797 	/* Handle any errors. */
798 	if (status != bit) {
799 		printf("%s: APE lock %d request failed! "
800 		    "request = 0x%04x[0x%04x], status = 0x%04x[0x%04x]\n",
801 		    device_xname(sc->bge_dev),
802 		    locknum, req + off, bit & 0xFFFF, gnt + off,
803 		    status & 0xFFFF);
804 		/* Revoke the lock request. */
805 		APE_WRITE_4(sc, gnt + off, bit);
806 		return EBUSY;
807 	}
808 
809 	return 0;
810 }
811 
812 void
813 bge_ape_unlock(struct bge_softc *sc, int locknum)
814 {
815 	struct pci_attach_args *pa = &(sc->bge_pa);
816 	uint32_t bit, gnt;
817 	int off;
818 
819 	if ((sc->bge_mfw_flags & BGE_MFW_ON_APE) == 0)
820 		return;
821 
822 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5761)
823 		gnt = BGE_APE_LOCK_GRANT;
824 	else
825 		gnt = BGE_APE_PER_LOCK_GRANT;
826 
827 	off = 4 * locknum;
828 
829 	switch (locknum) {
830 	case BGE_APE_LOCK_GPIO:
831 		if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5761)
832 			return;
833 		if (pa->pa_function == 0)
834 			bit = BGE_APE_LOCK_GRANT_DRIVER0;
835 		else
836 			bit = (1 << pa->pa_function);
837 		break;
838 	case BGE_APE_LOCK_GRC:
839 		if (pa->pa_function == 0)
840 			bit = BGE_APE_LOCK_GRANT_DRIVER0;
841 		else
842 			bit = (1 << pa->pa_function);
843 		break;
844 	case BGE_APE_LOCK_MEM:
845 		if (pa->pa_function == 0)
846 			bit = BGE_APE_LOCK_GRANT_DRIVER0;
847 		else
848 			bit = (1 << pa->pa_function);
849 		break;
850 	case BGE_APE_LOCK_PHY0:
851 	case BGE_APE_LOCK_PHY1:
852 	case BGE_APE_LOCK_PHY2:
853 	case BGE_APE_LOCK_PHY3:
854 		bit = BGE_APE_LOCK_GRANT_DRIVER0;
855 		break;
856 	default:
857 		return;
858 	}
859 
860 	/* Write and flush for consecutive bge_ape_lock() */
861 	APE_WRITE_4_FLUSH(sc, gnt + off, bit);
862 }
863 
864 /*
865  * Send an event to the APE firmware.
866  */
867 void
868 bge_ape_send_event(struct bge_softc *sc, uint32_t event)
869 {
870 	uint32_t apedata;
871 	int i;
872 
873 	/* NCSI does not support APE events. */
874 	if ((sc->bge_mfw_flags & BGE_MFW_ON_APE) == 0)
875 		return;
876 
877 	/* Wait up to 1ms for APE to service previous event. */
878 	for (i = 10; i > 0; i--) {
879 		if (bge_ape_lock(sc, BGE_APE_LOCK_MEM) != 0)
880 			break;
881 		apedata = APE_READ_4(sc, BGE_APE_EVENT_STATUS);
882 		if ((apedata & BGE_APE_EVENT_STATUS_EVENT_PENDING) == 0) {
883 			APE_WRITE_4(sc, BGE_APE_EVENT_STATUS, event |
884 			    BGE_APE_EVENT_STATUS_EVENT_PENDING);
885 			bge_ape_unlock(sc, BGE_APE_LOCK_MEM);
886 			APE_WRITE_4(sc, BGE_APE_EVENT, BGE_APE_EVENT_1);
887 			break;
888 		}
889 		bge_ape_unlock(sc, BGE_APE_LOCK_MEM);
890 		DELAY(100);
891 	}
892 	if (i == 0) {
893 		printf("%s: APE event 0x%08x send timed out\n",
894 		    device_xname(sc->bge_dev), event);
895 	}
896 }
897 
898 void
899 bge_ape_driver_state_change(struct bge_softc *sc, int kind)
900 {
901 	uint32_t apedata, event;
902 
903 	if ((sc->bge_mfw_flags & BGE_MFW_ON_APE) == 0)
904 		return;
905 
906 	switch (kind) {
907 	case BGE_RESET_START:
908 		/* If this is the first load, clear the load counter. */
909 		apedata = APE_READ_4(sc, BGE_APE_HOST_SEG_SIG);
910 		if (apedata != BGE_APE_HOST_SEG_SIG_MAGIC)
911 			APE_WRITE_4(sc, BGE_APE_HOST_INIT_COUNT, 0);
912 		else {
913 			apedata = APE_READ_4(sc, BGE_APE_HOST_INIT_COUNT);
914 			APE_WRITE_4(sc, BGE_APE_HOST_INIT_COUNT, ++apedata);
915 		}
916 		APE_WRITE_4(sc, BGE_APE_HOST_SEG_SIG,
917 		    BGE_APE_HOST_SEG_SIG_MAGIC);
918 		APE_WRITE_4(sc, BGE_APE_HOST_SEG_LEN,
919 		    BGE_APE_HOST_SEG_LEN_MAGIC);
920 
921 		/* Add some version info if bge(4) supports it. */
922 		APE_WRITE_4(sc, BGE_APE_HOST_DRIVER_ID,
923 		    BGE_APE_HOST_DRIVER_ID_MAGIC(1, 0));
924 		APE_WRITE_4(sc, BGE_APE_HOST_BEHAVIOR,
925 		    BGE_APE_HOST_BEHAV_NO_PHYLOCK);
926 		APE_WRITE_4(sc, BGE_APE_HOST_HEARTBEAT_INT_MS,
927 		    BGE_APE_HOST_HEARTBEAT_INT_DISABLE);
928 		APE_WRITE_4(sc, BGE_APE_HOST_DRVR_STATE,
929 		    BGE_APE_HOST_DRVR_STATE_START);
930 		event = BGE_APE_EVENT_STATUS_STATE_START;
931 		break;
932 	case BGE_RESET_SHUTDOWN:
933 		APE_WRITE_4(sc, BGE_APE_HOST_DRVR_STATE,
934 		    BGE_APE_HOST_DRVR_STATE_UNLOAD);
935 		event = BGE_APE_EVENT_STATUS_STATE_UNLOAD;
936 		break;
937 	case BGE_RESET_SUSPEND:
938 		event = BGE_APE_EVENT_STATUS_STATE_SUSPEND;
939 		break;
940 	default:
941 		return;
942 	}
943 
944 	bge_ape_send_event(sc, event | BGE_APE_EVENT_STATUS_DRIVER_EVNT |
945 	    BGE_APE_EVENT_STATUS_STATE_CHNGE);
946 }
947 
948 static uint8_t
949 bge_nvram_getbyte(struct bge_softc *sc, int addr, uint8_t *dest)
950 {
951 	uint32_t access, byte = 0;
952 	int i;
953 
954 	/* Lock. */
955 	CSR_WRITE_4(sc, BGE_NVRAM_SWARB, BGE_NVRAMSWARB_SET1);
956 	for (i = 0; i < 8000; i++) {
957 		if (CSR_READ_4(sc, BGE_NVRAM_SWARB) & BGE_NVRAMSWARB_GNT1)
958 			break;
959 		DELAY(20);
960 	}
961 	if (i == 8000)
962 		return 1;
963 
964 	/* Enable access. */
965 	access = CSR_READ_4(sc, BGE_NVRAM_ACCESS);
966 	CSR_WRITE_4(sc, BGE_NVRAM_ACCESS, access | BGE_NVRAMACC_ENABLE);
967 
968 	CSR_WRITE_4(sc, BGE_NVRAM_ADDR, addr & 0xfffffffc);
969 	CSR_WRITE_4(sc, BGE_NVRAM_CMD, BGE_NVRAM_READCMD);
970 	for (i = 0; i < BGE_TIMEOUT * 10; i++) {
971 		DELAY(10);
972 		if (CSR_READ_4(sc, BGE_NVRAM_CMD) & BGE_NVRAMCMD_DONE) {
973 			DELAY(10);
974 			break;
975 		}
976 	}
977 
978 	if (i == BGE_TIMEOUT * 10) {
979 		aprint_error_dev(sc->bge_dev, "nvram read timed out\n");
980 		return 1;
981 	}
982 
983 	/* Get result. */
984 	byte = CSR_READ_4(sc, BGE_NVRAM_RDDATA);
985 
986 	*dest = (bswap32(byte) >> ((addr % 4) * 8)) & 0xFF;
987 
988 	/* Disable access. */
989 	CSR_WRITE_4(sc, BGE_NVRAM_ACCESS, access);
990 
991 	/* Unlock. */
992 	CSR_WRITE_4_FLUSH(sc, BGE_NVRAM_SWARB, BGE_NVRAMSWARB_CLR1);
993 
994 	return 0;
995 }
996 
997 /*
998  * Read a sequence of bytes from NVRAM.
999  */
1000 static int
1001 bge_read_nvram(struct bge_softc *sc, uint8_t *dest, int off, int cnt)
1002 {
1003 	int error = 0, i;
1004 	uint8_t byte = 0;
1005 
1006 	if (BGE_ASICREV(sc->bge_chipid) != BGE_ASICREV_BCM5906)
1007 		return 1;
1008 
1009 	for (i = 0; i < cnt; i++) {
1010 		error = bge_nvram_getbyte(sc, off + i, &byte);
1011 		if (error)
1012 			break;
1013 		*(dest + i) = byte;
1014 	}
1015 
1016 	return error ? 1 : 0;
1017 }
1018 
1019 /*
1020  * Read a byte of data stored in the EEPROM at address 'addr.' The
1021  * BCM570x supports both the traditional bitbang interface and an
1022  * auto access interface for reading the EEPROM. We use the auto
1023  * access method.
1024  */
1025 static uint8_t
1026 bge_eeprom_getbyte(struct bge_softc *sc, int addr, uint8_t *dest)
1027 {
1028 	int i;
1029 	uint32_t byte = 0;
1030 
1031 	/*
1032 	 * Enable use of auto EEPROM access so we can avoid
1033 	 * having to use the bitbang method.
1034 	 */
1035 	BGE_SETBIT_FLUSH(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_AUTO_EEPROM);
1036 
1037 	/* Reset the EEPROM, load the clock period. */
1038 	CSR_WRITE_4_FLUSH(sc, BGE_EE_ADDR,
1039 	    BGE_EEADDR_RESET | BGE_EEHALFCLK(BGE_HALFCLK_384SCL));
1040 	DELAY(20);
1041 
1042 	/* Issue the read EEPROM command. */
1043 	CSR_WRITE_4(sc, BGE_EE_ADDR, BGE_EE_READCMD | addr);
1044 
1045 	/* Wait for completion */
1046 	for (i = 0; i < BGE_TIMEOUT * 10; i++) {
1047 		DELAY(10);
1048 		if (CSR_READ_4(sc, BGE_EE_ADDR) & BGE_EEADDR_DONE)
1049 			break;
1050 	}
1051 
1052 	if (i == BGE_TIMEOUT * 10) {
1053 		aprint_error_dev(sc->bge_dev, "eeprom read timed out\n");
1054 		return 1;
1055 	}
1056 
1057 	/* Get result. */
1058 	byte = CSR_READ_4(sc, BGE_EE_DATA);
1059 
1060 	*dest = (byte >> ((addr % 4) * 8)) & 0xFF;
1061 
1062 	return 0;
1063 }
1064 
1065 /*
1066  * Read a sequence of bytes from the EEPROM.
1067  */
1068 static int
1069 bge_read_eeprom(struct bge_softc *sc, void *destv, int off, int cnt)
1070 {
1071 	int error = 0, i;
1072 	uint8_t byte = 0;
1073 	char *dest = destv;
1074 
1075 	for (i = 0; i < cnt; i++) {
1076 		error = bge_eeprom_getbyte(sc, off + i, &byte);
1077 		if (error)
1078 			break;
1079 		*(dest + i) = byte;
1080 	}
1081 
1082 	return error ? 1 : 0;
1083 }
1084 
1085 static int
1086 bge_miibus_readreg(device_t dev, int phy, int reg, uint16_t *val)
1087 {
1088 	struct bge_softc * const sc = device_private(dev);
1089 	uint32_t data;
1090 	uint32_t autopoll;
1091 	int rv = 0;
1092 	int i;
1093 
1094 	if (bge_ape_lock(sc, sc->bge_phy_ape_lock) != 0)
1095 		return -1;
1096 
1097 	/* Reading with autopolling on may trigger PCI errors */
1098 	autopoll = CSR_READ_4(sc, BGE_MI_MODE);
1099 	if (autopoll & BGE_MIMODE_AUTOPOLL) {
1100 		BGE_STS_CLRBIT(sc, BGE_STS_AUTOPOLL);
1101 		BGE_CLRBIT_FLUSH(sc, BGE_MI_MODE, BGE_MIMODE_AUTOPOLL);
1102 		DELAY(80);
1103 	}
1104 
1105 	CSR_WRITE_4_FLUSH(sc, BGE_MI_COMM, BGE_MICMD_READ | BGE_MICOMM_BUSY |
1106 	    BGE_MIPHY(phy) | BGE_MIREG(reg));
1107 
1108 	for (i = 0; i < BGE_TIMEOUT; i++) {
1109 		delay(10);
1110 		data = CSR_READ_4(sc, BGE_MI_COMM);
1111 		if (!(data & BGE_MICOMM_BUSY)) {
1112 			DELAY(5);
1113 			data = CSR_READ_4(sc, BGE_MI_COMM);
1114 			break;
1115 		}
1116 	}
1117 
1118 	if (i == BGE_TIMEOUT) {
1119 		aprint_error_dev(sc->bge_dev, "PHY read timed out\n");
1120 		rv = ETIMEDOUT;
1121 	} else if ((data & BGE_MICOMM_READFAIL) != 0) {
1122 		/* XXX This error occurs on some devices while attaching. */
1123 		aprint_debug_dev(sc->bge_dev, "PHY read I/O error\n");
1124 		rv = EIO;
1125 	} else
1126 		*val = data & BGE_MICOMM_DATA;
1127 
1128 	if (autopoll & BGE_MIMODE_AUTOPOLL) {
1129 		BGE_STS_SETBIT(sc, BGE_STS_AUTOPOLL);
1130 		BGE_SETBIT_FLUSH(sc, BGE_MI_MODE, BGE_MIMODE_AUTOPOLL);
1131 		DELAY(80);
1132 	}
1133 
1134 	bge_ape_unlock(sc, sc->bge_phy_ape_lock);
1135 
1136 	return rv;
1137 }
1138 
1139 static int
1140 bge_miibus_writereg(device_t dev, int phy, int reg, uint16_t val)
1141 {
1142 	struct bge_softc * const sc = device_private(dev);
1143 	uint32_t data, autopoll;
1144 	int rv = 0;
1145 	int i;
1146 
1147 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906 &&
1148 	    (reg == MII_GTCR || reg == BRGPHY_MII_AUXCTL))
1149 		return 0;
1150 
1151 	if (bge_ape_lock(sc, sc->bge_phy_ape_lock) != 0)
1152 		return -1;
1153 
1154 	/* Reading with autopolling on may trigger PCI errors */
1155 	autopoll = CSR_READ_4(sc, BGE_MI_MODE);
1156 	if (autopoll & BGE_MIMODE_AUTOPOLL) {
1157 		BGE_STS_CLRBIT(sc, BGE_STS_AUTOPOLL);
1158 		BGE_CLRBIT_FLUSH(sc, BGE_MI_MODE, BGE_MIMODE_AUTOPOLL);
1159 		DELAY(80);
1160 	}
1161 
1162 	CSR_WRITE_4_FLUSH(sc, BGE_MI_COMM, BGE_MICMD_WRITE | BGE_MICOMM_BUSY |
1163 	    BGE_MIPHY(phy) | BGE_MIREG(reg) | val);
1164 
1165 	for (i = 0; i < BGE_TIMEOUT; i++) {
1166 		delay(10);
1167 		data = CSR_READ_4(sc, BGE_MI_COMM);
1168 		if (!(data & BGE_MICOMM_BUSY)) {
1169 			delay(5);
1170 			data = CSR_READ_4(sc, BGE_MI_COMM);
1171 			break;
1172 		}
1173 	}
1174 
1175 	if (i == BGE_TIMEOUT) {
1176 		aprint_error_dev(sc->bge_dev, "PHY write timed out\n");
1177 		rv = ETIMEDOUT;
1178 	} else if ((data & BGE_MICOMM_READFAIL) != 0) {
1179 		aprint_error_dev(sc->bge_dev, "PHY write I/O error\n");
1180 		rv = EIO;
1181 	}
1182 
1183 	if (autopoll & BGE_MIMODE_AUTOPOLL) {
1184 		BGE_STS_SETBIT(sc, BGE_STS_AUTOPOLL);
1185 		BGE_SETBIT_FLUSH(sc, BGE_MI_MODE, BGE_MIMODE_AUTOPOLL);
1186 		delay(80);
1187 	}
1188 
1189 	bge_ape_unlock(sc, sc->bge_phy_ape_lock);
1190 
1191 	return rv;
1192 }
1193 
1194 static void
1195 bge_miibus_statchg(struct ifnet *ifp)
1196 {
1197 	struct bge_softc * const sc = ifp->if_softc;
1198 	struct mii_data *mii = &sc->bge_mii;
1199 	uint32_t mac_mode, rx_mode, tx_mode;
1200 
1201 	/*
1202 	 * Get flow control negotiation result.
1203 	 */
1204 	if (IFM_SUBTYPE(mii->mii_media.ifm_cur->ifm_media) == IFM_AUTO &&
1205 	    (mii->mii_media_active & IFM_ETH_FMASK) != sc->bge_flowflags)
1206 		sc->bge_flowflags = mii->mii_media_active & IFM_ETH_FMASK;
1207 
1208 	if (!BGE_STS_BIT(sc, BGE_STS_LINK) &&
1209 	    mii->mii_media_status & IFM_ACTIVE &&
1210 	    IFM_SUBTYPE(mii->mii_media_active) != IFM_NONE)
1211 		BGE_STS_SETBIT(sc, BGE_STS_LINK);
1212 	else if (BGE_STS_BIT(sc, BGE_STS_LINK) &&
1213 	    (!(mii->mii_media_status & IFM_ACTIVE) ||
1214 	    IFM_SUBTYPE(mii->mii_media_active) == IFM_NONE))
1215 		BGE_STS_CLRBIT(sc, BGE_STS_LINK);
1216 
1217 	if (!BGE_STS_BIT(sc, BGE_STS_LINK))
1218 		return;
1219 
1220 	/* Set the port mode (MII/GMII) to match the link speed. */
1221 	mac_mode = CSR_READ_4(sc, BGE_MAC_MODE) &
1222 	    ~(BGE_MACMODE_PORTMODE | BGE_MACMODE_HALF_DUPLEX);
1223 	tx_mode = CSR_READ_4(sc, BGE_TX_MODE);
1224 	rx_mode = CSR_READ_4(sc, BGE_RX_MODE);
1225 	if (IFM_SUBTYPE(mii->mii_media_active) == IFM_1000_T ||
1226 	    IFM_SUBTYPE(mii->mii_media_active) == IFM_1000_SX)
1227 		mac_mode |= BGE_PORTMODE_GMII;
1228 	else
1229 		mac_mode |= BGE_PORTMODE_MII;
1230 
1231 	tx_mode &= ~BGE_TXMODE_FLOWCTL_ENABLE;
1232 	rx_mode &= ~BGE_RXMODE_FLOWCTL_ENABLE;
1233 	if ((mii->mii_media_active & IFM_FDX) != 0) {
1234 		if (sc->bge_flowflags & IFM_ETH_TXPAUSE)
1235 			tx_mode |= BGE_TXMODE_FLOWCTL_ENABLE;
1236 		if (sc->bge_flowflags & IFM_ETH_RXPAUSE)
1237 			rx_mode |= BGE_RXMODE_FLOWCTL_ENABLE;
1238 	} else
1239 		mac_mode |= BGE_MACMODE_HALF_DUPLEX;
1240 
1241 	CSR_WRITE_4_FLUSH(sc, BGE_MAC_MODE, mac_mode);
1242 	DELAY(40);
1243 	CSR_WRITE_4(sc, BGE_TX_MODE, tx_mode);
1244 	CSR_WRITE_4(sc, BGE_RX_MODE, rx_mode);
1245 }
1246 
1247 /*
1248  * Update rx threshold levels to values in a particular slot
1249  * of the interrupt-mitigation table bge_rx_threshes.
1250  */
1251 static void
1252 bge_set_thresh(struct ifnet *ifp, int lvl)
1253 {
1254 	struct bge_softc * const sc = ifp->if_softc;
1255 
1256 	/*
1257 	 * For now, just save the new Rx-intr thresholds and record
1258 	 * that a threshold update is pending.  Updating the hardware
1259 	 * registers here (even at splhigh()) is observed to
1260 	 * occasionally cause glitches where Rx-interrupts are not
1261 	 * honoured for up to 10 seconds. jonathan@NetBSD.org, 2003-04-05
1262 	 */
1263 	mutex_enter(sc->sc_intr_lock);
1264 	sc->bge_rx_coal_ticks = bge_rx_threshes[lvl].rx_ticks;
1265 	sc->bge_rx_max_coal_bds = bge_rx_threshes[lvl].rx_max_bds;
1266 	sc->bge_pending_rxintr_change = true;
1267 	mutex_exit(sc->sc_intr_lock);
1268 }
1269 
1270 
1271 /*
1272  * Update Rx thresholds of all bge devices
1273  */
1274 static void
1275 bge_update_all_threshes(int lvl)
1276 {
1277 	const char * const namebuf = "bge";
1278 	const size_t namelen = strlen(namebuf);
1279 	struct ifnet *ifp;
1280 
1281 	if (lvl < 0)
1282 		lvl = 0;
1283 	else if (lvl >= NBGE_RX_THRESH)
1284 		lvl = NBGE_RX_THRESH - 1;
1285 
1286 	/*
1287 	 * Now search all the interfaces for this name/number
1288 	 */
1289 	int s = pserialize_read_enter();
1290 	IFNET_READER_FOREACH(ifp) {
1291 		if (strncmp(ifp->if_xname, namebuf, namelen) != 0)
1292 			continue;
1293 		/* We got a match: update if doing auto-threshold-tuning */
1294 		if (bge_auto_thresh)
1295 			bge_set_thresh(ifp, lvl);
1296 	}
1297 	pserialize_read_exit(s);
1298 }
1299 
1300 /*
1301  * Handle events that have triggered interrupts.
1302  */
1303 static void
1304 bge_handle_events(struct bge_softc *sc)
1305 {
1306 
1307 	return;
1308 }
1309 
1310 /*
1311  * Memory management for jumbo frames.
1312  */
1313 
1314 static int
1315 bge_alloc_jumbo_mem(struct bge_softc *sc)
1316 {
1317 	char *ptr, *kva;
1318 	int i, rseg, state, error;
1319 	struct bge_jpool_entry *entry;
1320 
1321 	state = error = 0;
1322 
1323 	/* Grab a big chunk o' storage. */
1324 	if (bus_dmamem_alloc(sc->bge_dmatag, BGE_JMEM, PAGE_SIZE, 0,
1325 	    &sc->bge_cdata.bge_rx_jumbo_seg, 1, &rseg, BUS_DMA_WAITOK)) {
1326 		aprint_error_dev(sc->bge_dev, "can't alloc rx buffers\n");
1327 		return ENOBUFS;
1328 	}
1329 
1330 	state = 1;
1331 	if (bus_dmamem_map(sc->bge_dmatag, &sc->bge_cdata.bge_rx_jumbo_seg,
1332 	    rseg, BGE_JMEM, (void **)&kva, BUS_DMA_WAITOK)) {
1333 		aprint_error_dev(sc->bge_dev,
1334 		    "can't map DMA buffers (%d bytes)\n", (int)BGE_JMEM);
1335 		error = ENOBUFS;
1336 		goto out;
1337 	}
1338 
1339 	state = 2;
1340 	if (bus_dmamap_create(sc->bge_dmatag, BGE_JMEM, 1, BGE_JMEM, 0,
1341 	    BUS_DMA_WAITOK, &sc->bge_cdata.bge_rx_jumbo_map)) {
1342 		aprint_error_dev(sc->bge_dev, "can't create DMA map\n");
1343 		error = ENOBUFS;
1344 		goto out;
1345 	}
1346 
1347 	state = 3;
1348 	if (bus_dmamap_load(sc->bge_dmatag, sc->bge_cdata.bge_rx_jumbo_map,
1349 	    kva, BGE_JMEM, NULL, BUS_DMA_WAITOK)) {
1350 		aprint_error_dev(sc->bge_dev, "can't load DMA map\n");
1351 		error = ENOBUFS;
1352 		goto out;
1353 	}
1354 
1355 	state = 4;
1356 	sc->bge_cdata.bge_jumbo_buf = (void *)kva;
1357 	DPRINTFN(1,("bge_jumbo_buf = %p\n", sc->bge_cdata.bge_jumbo_buf));
1358 
1359 	SLIST_INIT(&sc->bge_jfree_listhead);
1360 	SLIST_INIT(&sc->bge_jinuse_listhead);
1361 
1362 	/*
1363 	 * Now divide it up into 9K pieces and save the addresses
1364 	 * in an array.
1365 	 */
1366 	ptr = sc->bge_cdata.bge_jumbo_buf;
1367 	for (i = 0; i < BGE_JSLOTS; i++) {
1368 		sc->bge_cdata.bge_jslots[i] = ptr;
1369 		ptr += BGE_JLEN;
1370 		entry = kmem_alloc(sizeof(*entry), KM_SLEEP);
1371 		entry->slot = i;
1372 		SLIST_INSERT_HEAD(&sc->bge_jfree_listhead,
1373 				 entry, jpool_entries);
1374 	}
1375 out:
1376 	if (error != 0) {
1377 		switch (state) {
1378 		case 4:
1379 			bus_dmamap_unload(sc->bge_dmatag,
1380 			    sc->bge_cdata.bge_rx_jumbo_map);
1381 			/* FALLTHROUGH */
1382 		case 3:
1383 			bus_dmamap_destroy(sc->bge_dmatag,
1384 			    sc->bge_cdata.bge_rx_jumbo_map);
1385 			/* FALLTHROUGH */
1386 		case 2:
1387 			bus_dmamem_unmap(sc->bge_dmatag, kva, BGE_JMEM);
1388 			/* FALLTHROUGH */
1389 		case 1:
1390 			bus_dmamem_free(sc->bge_dmatag,
1391 			    &sc->bge_cdata.bge_rx_jumbo_seg, rseg);
1392 			break;
1393 		default:
1394 			break;
1395 		}
1396 	}
1397 
1398 	return error;
1399 }
1400 
1401 static void
1402 bge_free_jumbo_mem(struct bge_softc *sc)
1403 {
1404 	struct bge_jpool_entry *entry, *tmp;
1405 
1406 	KASSERT(SLIST_EMPTY(&sc->bge_jinuse_listhead));
1407 
1408 	SLIST_FOREACH_SAFE(entry, &sc->bge_jfree_listhead, jpool_entries, tmp) {
1409 		kmem_free(entry, sizeof(*entry));
1410 	}
1411 
1412 	bus_dmamap_unload(sc->bge_dmatag, sc->bge_cdata.bge_rx_jumbo_map);
1413 
1414 	bus_dmamap_destroy(sc->bge_dmatag, sc->bge_cdata.bge_rx_jumbo_map);
1415 
1416 	bus_dmamem_unmap(sc->bge_dmatag, sc->bge_cdata.bge_jumbo_buf, BGE_JMEM);
1417 
1418 	bus_dmamem_free(sc->bge_dmatag, &sc->bge_cdata.bge_rx_jumbo_seg, 1);
1419 }
1420 
1421 /*
1422  * Allocate a jumbo buffer.
1423  */
1424 static void *
1425 bge_jalloc(struct bge_softc *sc)
1426 {
1427 	struct bge_jpool_entry	 *entry;
1428 
1429 	entry = SLIST_FIRST(&sc->bge_jfree_listhead);
1430 
1431 	if (entry == NULL) {
1432 		aprint_error_dev(sc->bge_dev, "no free jumbo buffers\n");
1433 		return NULL;
1434 	}
1435 
1436 	SLIST_REMOVE_HEAD(&sc->bge_jfree_listhead, jpool_entries);
1437 	SLIST_INSERT_HEAD(&sc->bge_jinuse_listhead, entry, jpool_entries);
1438 	return sc->bge_cdata.bge_jslots[entry->slot];
1439 }
1440 
1441 /*
1442  * Release a jumbo buffer.
1443  */
1444 static void
1445 bge_jfree(struct mbuf *m, void *buf, size_t size, void *arg)
1446 {
1447 	struct bge_jpool_entry *entry;
1448 	struct bge_softc * const sc = arg;
1449 
1450 	if (sc == NULL)
1451 		panic("bge_jfree: can't find softc pointer!");
1452 
1453 	/* calculate the slot this buffer belongs to */
1454 	int i = ((char *)buf - (char *)sc->bge_cdata.bge_jumbo_buf) / BGE_JLEN;
1455 
1456 	if (i < 0 || i >= BGE_JSLOTS)
1457 		panic("bge_jfree: asked to free buffer that we don't manage!");
1458 
1459 	mutex_enter(sc->sc_intr_lock);
1460 	entry = SLIST_FIRST(&sc->bge_jinuse_listhead);
1461 	if (entry == NULL)
1462 		panic("bge_jfree: buffer not in use!");
1463 	entry->slot = i;
1464 	SLIST_REMOVE_HEAD(&sc->bge_jinuse_listhead, jpool_entries);
1465 	SLIST_INSERT_HEAD(&sc->bge_jfree_listhead, entry, jpool_entries);
1466 	mutex_exit(sc->sc_intr_lock);
1467 
1468 	if (__predict_true(m != NULL))
1469 		pool_cache_put(mb_cache, m);
1470 }
1471 
1472 
1473 /*
1474  * Initialize a standard receive ring descriptor.
1475  */
1476 static int
1477 bge_newbuf_std(struct bge_softc *sc, int i)
1478 {
1479 	const bus_dmamap_t dmamap = sc->bge_cdata.bge_rx_std_map[i];
1480 	struct mbuf *m;
1481 
1482 	MGETHDR(m, M_DONTWAIT, MT_DATA);
1483 	if (m == NULL)
1484 		return ENOBUFS;
1485 
1486 	MCLGET(m, M_DONTWAIT);
1487 	if (!(m->m_flags & M_EXT)) {
1488 		m_freem(m);
1489 		return ENOBUFS;
1490 	}
1491 	m->m_len = m->m_pkthdr.len = MCLBYTES;
1492 
1493 	if (!(sc->bge_flags & BGEF_RX_ALIGNBUG))
1494 	    m_adj(m, ETHER_ALIGN);
1495 	if (bus_dmamap_load_mbuf(sc->bge_dmatag, dmamap, m,
1496 	    BUS_DMA_READ | BUS_DMA_NOWAIT)) {
1497 		m_freem(m);
1498 		return ENOBUFS;
1499 	}
1500 	bus_dmamap_sync(sc->bge_dmatag, dmamap, 0, dmamap->dm_mapsize,
1501 	    BUS_DMASYNC_PREREAD);
1502 	sc->bge_cdata.bge_rx_std_chain[i] = m;
1503 
1504 	bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
1505 	    offsetof(struct bge_ring_data, bge_rx_std_ring) +
1506 		i * sizeof(struct bge_rx_bd),
1507 	    sizeof(struct bge_rx_bd),
1508 	    BUS_DMASYNC_POSTWRITE);
1509 
1510 	struct bge_rx_bd * const r = &sc->bge_rdata->bge_rx_std_ring[i];
1511 	BGE_HOSTADDR(r->bge_addr, dmamap->dm_segs[0].ds_addr);
1512 	r->bge_flags = BGE_RXBDFLAG_END;
1513 	r->bge_len = m->m_len;
1514 	r->bge_idx = i;
1515 
1516 	bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
1517 	    offsetof(struct bge_ring_data, bge_rx_std_ring) +
1518 		i * sizeof(struct bge_rx_bd),
1519 	    sizeof(struct bge_rx_bd),
1520 	    BUS_DMASYNC_PREWRITE);
1521 
1522 	sc->bge_std_cnt++;
1523 
1524 	return 0;
1525 }
1526 
1527 /*
1528  * Initialize a jumbo receive ring descriptor. This allocates
1529  * a jumbo buffer from the pool managed internally by the driver.
1530  */
1531 static int
1532 bge_newbuf_jumbo(struct bge_softc *sc, int i, struct mbuf *m)
1533 {
1534 	struct mbuf *m_new = NULL;
1535 	struct bge_rx_bd *r;
1536 	void *buf = NULL;
1537 
1538 	if (m == NULL) {
1539 
1540 		/* Allocate the mbuf. */
1541 		MGETHDR(m_new, M_DONTWAIT, MT_DATA);
1542 		if (m_new == NULL)
1543 			return ENOBUFS;
1544 
1545 		/* Allocate the jumbo buffer */
1546 		buf = bge_jalloc(sc);
1547 		if (buf == NULL) {
1548 			m_freem(m_new);
1549 			aprint_error_dev(sc->bge_dev,
1550 			    "jumbo allocation failed -- packet dropped!\n");
1551 			return ENOBUFS;
1552 		}
1553 
1554 		/* Attach the buffer to the mbuf. */
1555 		m_new->m_len = m_new->m_pkthdr.len = BGE_JUMBO_FRAMELEN;
1556 		MEXTADD(m_new, buf, BGE_JUMBO_FRAMELEN, M_DEVBUF,
1557 		    bge_jfree, sc);
1558 		m_new->m_flags |= M_EXT_RW;
1559 	} else {
1560 		m_new = m;
1561 		buf = m_new->m_data = m_new->m_ext.ext_buf;
1562 		m_new->m_ext.ext_size = BGE_JUMBO_FRAMELEN;
1563 	}
1564 	if (!(sc->bge_flags & BGEF_RX_ALIGNBUG))
1565 	    m_adj(m_new, ETHER_ALIGN);
1566 	bus_dmamap_sync(sc->bge_dmatag, sc->bge_cdata.bge_rx_jumbo_map,
1567 	    mtod(m_new, char *) - (char *)sc->bge_cdata.bge_jumbo_buf,
1568 	    BGE_JLEN, BUS_DMASYNC_PREREAD);
1569 
1570 	/* Set up the descriptor. */
1571 	r = &sc->bge_rdata->bge_rx_jumbo_ring[i];
1572 	sc->bge_cdata.bge_rx_jumbo_chain[i] = m_new;
1573 	BGE_HOSTADDR(r->bge_addr, BGE_JUMBO_DMA_ADDR(sc, m_new));
1574 	r->bge_flags = BGE_RXBDFLAG_END | BGE_RXBDFLAG_JUMBO_RING;
1575 	r->bge_len = m_new->m_len;
1576 	r->bge_idx = i;
1577 
1578 	bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
1579 	    offsetof(struct bge_ring_data, bge_rx_jumbo_ring) +
1580 		i * sizeof(struct bge_rx_bd),
1581 	    sizeof(struct bge_rx_bd),
1582 	    BUS_DMASYNC_PREWRITE | BUS_DMASYNC_PREREAD);
1583 
1584 	return 0;
1585 }
1586 
1587 static int
1588 bge_init_rx_ring_std(struct bge_softc *sc)
1589 {
1590 	bus_dmamap_t dmamap;
1591 	int error = 0;
1592 	u_int i;
1593 
1594 	if (sc->bge_flags & BGEF_RXRING_VALID)
1595 		return 0;
1596 
1597 	for (i = 0; i < BGE_STD_RX_RING_CNT; i++) {
1598 		error = bus_dmamap_create(sc->bge_dmatag, MCLBYTES, 1,
1599 		    MCLBYTES, 0, BUS_DMA_WAITOK | BUS_DMA_ALLOCNOW, &dmamap);
1600 		if (error)
1601 			goto uncreate;
1602 
1603 		sc->bge_cdata.bge_rx_std_map[i] = dmamap;
1604 		memset(&sc->bge_rdata->bge_rx_std_ring[i], 0,
1605 		    sizeof(struct bge_rx_bd));
1606 	}
1607 
1608 	sc->bge_std = i - 1;
1609 	sc->bge_std_cnt = 0;
1610 	bge_fill_rx_ring_std(sc);
1611 
1612 	sc->bge_flags |= BGEF_RXRING_VALID;
1613 
1614 	return 0;
1615 
1616 uncreate:
1617 	while (--i) {
1618 		bus_dmamap_destroy(sc->bge_dmatag,
1619 		    sc->bge_cdata.bge_rx_std_map[i]);
1620 	}
1621 	return error;
1622 }
1623 
1624 static void
1625 bge_fill_rx_ring_std(struct bge_softc *sc)
1626 {
1627 	int i = sc->bge_std;
1628 	bool post = false;
1629 
1630 	while (sc->bge_std_cnt < BGE_STD_RX_RING_CNT) {
1631 		BGE_INC(i, BGE_STD_RX_RING_CNT);
1632 
1633 		if (bge_newbuf_std(sc, i) != 0)
1634 			break;
1635 
1636 		sc->bge_std = i;
1637 		post = true;
1638 	}
1639 
1640 	if (post)
1641 		bge_writembx(sc, BGE_MBX_RX_STD_PROD_LO, sc->bge_std);
1642 }
1643 
1644 
1645 static void
1646 bge_free_rx_ring_std(struct bge_softc *sc)
1647 {
1648 
1649 	if (!(sc->bge_flags & BGEF_RXRING_VALID))
1650 		return;
1651 
1652 	for (u_int i = 0; i < BGE_STD_RX_RING_CNT; i++) {
1653 		const bus_dmamap_t dmap = sc->bge_cdata.bge_rx_std_map[i];
1654 		struct mbuf * const m = sc->bge_cdata.bge_rx_std_chain[i];
1655 		if (m != NULL) {
1656 			bus_dmamap_sync(sc->bge_dmatag, dmap, 0,
1657 			    dmap->dm_mapsize, BUS_DMASYNC_POSTREAD);
1658 			bus_dmamap_unload(sc->bge_dmatag, dmap);
1659 			m_freem(m);
1660 			sc->bge_cdata.bge_rx_std_chain[i] = NULL;
1661 		}
1662 		bus_dmamap_destroy(sc->bge_dmatag,
1663 		    sc->bge_cdata.bge_rx_std_map[i]);
1664 		sc->bge_cdata.bge_rx_std_map[i] = NULL;
1665 		memset((char *)&sc->bge_rdata->bge_rx_std_ring[i], 0,
1666 		    sizeof(struct bge_rx_bd));
1667 	}
1668 
1669 	sc->bge_flags &= ~BGEF_RXRING_VALID;
1670 }
1671 
1672 static int
1673 bge_init_rx_ring_jumbo(struct bge_softc *sc)
1674 {
1675 	int i;
1676 	volatile struct bge_rcb *rcb;
1677 
1678 	if (sc->bge_flags & BGEF_JUMBO_RXRING_VALID)
1679 		return 0;
1680 
1681 	for (i = 0; i < BGE_JUMBO_RX_RING_CNT; i++) {
1682 		if (bge_newbuf_jumbo(sc, i, NULL) == ENOBUFS)
1683 			return ENOBUFS;
1684 	}
1685 
1686 	sc->bge_jumbo = i - 1;
1687 	sc->bge_flags |= BGEF_JUMBO_RXRING_VALID;
1688 
1689 	rcb = &sc->bge_rdata->bge_info.bge_jumbo_rx_rcb;
1690 	rcb->bge_maxlen_flags = 0;
1691 	CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_MAXLEN_FLAGS, rcb->bge_maxlen_flags);
1692 
1693 	bge_writembx(sc, BGE_MBX_RX_JUMBO_PROD_LO, sc->bge_jumbo);
1694 
1695 	return 0;
1696 }
1697 
1698 static void
1699 bge_free_rx_ring_jumbo(struct bge_softc *sc)
1700 {
1701 	int i;
1702 
1703 	if (!(sc->bge_flags & BGEF_JUMBO_RXRING_VALID))
1704 		return;
1705 
1706 	for (i = 0; i < BGE_JUMBO_RX_RING_CNT; i++) {
1707 		m_freem(sc->bge_cdata.bge_rx_jumbo_chain[i]);
1708 		sc->bge_cdata.bge_rx_jumbo_chain[i] = NULL;
1709 		memset((char *)&sc->bge_rdata->bge_rx_jumbo_ring[i], 0,
1710 		    sizeof(struct bge_rx_bd));
1711 	}
1712 
1713 	sc->bge_flags &= ~BGEF_JUMBO_RXRING_VALID;
1714 }
1715 
1716 static void
1717 bge_free_tx_ring(struct bge_softc *sc, bool disable)
1718 {
1719 	int i;
1720 	struct txdmamap_pool_entry *dma;
1721 
1722 	if (!(sc->bge_flags & BGEF_TXRING_VALID))
1723 		return;
1724 
1725 	for (i = 0; i < BGE_TX_RING_CNT; i++) {
1726 		if (sc->bge_cdata.bge_tx_chain[i] != NULL) {
1727 			m_freem(sc->bge_cdata.bge_tx_chain[i]);
1728 			sc->bge_cdata.bge_tx_chain[i] = NULL;
1729 			SLIST_INSERT_HEAD(&sc->txdma_list, sc->txdma[i],
1730 					    link);
1731 			sc->txdma[i] = 0;
1732 		}
1733 		memset((char *)&sc->bge_rdata->bge_tx_ring[i], 0,
1734 		    sizeof(struct bge_tx_bd));
1735 	}
1736 
1737 	if (disable) {
1738 		while ((dma = SLIST_FIRST(&sc->txdma_list))) {
1739 			SLIST_REMOVE_HEAD(&sc->txdma_list, link);
1740 			bus_dmamap_destroy(sc->bge_dmatag, dma->dmamap);
1741 			if (sc->bge_dma64) {
1742 				bus_dmamap_destroy(sc->bge_dmatag32,
1743 				    dma->dmamap32);
1744 			}
1745 			kmem_free(dma, sizeof(*dma));
1746 		}
1747 		SLIST_INIT(&sc->txdma_list);
1748 	}
1749 
1750 	sc->bge_flags &= ~BGEF_TXRING_VALID;
1751 }
1752 
1753 static int
1754 bge_init_tx_ring(struct bge_softc *sc)
1755 {
1756 	struct ifnet * const ifp = &sc->ethercom.ec_if;
1757 	int i;
1758 	bus_dmamap_t dmamap, dmamap32;
1759 	bus_size_t maxsegsz;
1760 	struct txdmamap_pool_entry *dma;
1761 
1762 	if (sc->bge_flags & BGEF_TXRING_VALID)
1763 		return 0;
1764 
1765 	sc->bge_txcnt = 0;
1766 	sc->bge_tx_saved_considx = 0;
1767 
1768 	/* Initialize transmit producer index for host-memory send ring. */
1769 	sc->bge_tx_prodidx = 0;
1770 	bge_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, sc->bge_tx_prodidx);
1771 	/* 5700 b2 errata */
1772 	if (BGE_CHIPREV(sc->bge_chipid) == BGE_CHIPREV_5700_BX)
1773 		bge_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, sc->bge_tx_prodidx);
1774 
1775 	/* NIC-memory send ring not used; initialize to zero. */
1776 	bge_writembx(sc, BGE_MBX_TX_NIC_PROD0_LO, 0);
1777 	/* 5700 b2 errata */
1778 	if (BGE_CHIPREV(sc->bge_chipid) == BGE_CHIPREV_5700_BX)
1779 		bge_writembx(sc, BGE_MBX_TX_NIC_PROD0_LO, 0);
1780 
1781 	/* Limit DMA segment size for some chips */
1782 	if ((BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM57766) &&
1783 	    (ifp->if_mtu <= ETHERMTU))
1784 		maxsegsz = 2048;
1785 	else if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5719)
1786 		maxsegsz = 4096;
1787 	else
1788 		maxsegsz = ETHER_MAX_LEN_JUMBO;
1789 
1790 	if (SLIST_FIRST(&sc->txdma_list) != NULL)
1791 		goto alloc_done;
1792 
1793 	for (i = 0; i < BGE_TX_RING_CNT; i++) {
1794 		if (bus_dmamap_create(sc->bge_dmatag, BGE_TXDMA_MAX,
1795 		    BGE_NTXSEG, maxsegsz, 0, BUS_DMA_WAITOK | BUS_DMA_ALLOCNOW,
1796 		    &dmamap))
1797 			return ENOBUFS;
1798 		if (dmamap == NULL)
1799 			panic("dmamap NULL in bge_init_tx_ring");
1800 		if (sc->bge_dma64) {
1801 			if (bus_dmamap_create(sc->bge_dmatag32, BGE_TXDMA_MAX,
1802 			    BGE_NTXSEG, maxsegsz, 0,
1803 			    BUS_DMA_WAITOK | BUS_DMA_ALLOCNOW,
1804 			    &dmamap32)) {
1805 				bus_dmamap_destroy(sc->bge_dmatag, dmamap);
1806 				return ENOBUFS;
1807 			}
1808 			if (dmamap32 == NULL)
1809 				panic("dmamap32 NULL in bge_init_tx_ring");
1810 		} else
1811 			dmamap32 = dmamap;
1812 		dma = kmem_alloc(sizeof(*dma), KM_NOSLEEP);
1813 		if (dma == NULL) {
1814 			aprint_error_dev(sc->bge_dev,
1815 			    "can't alloc txdmamap_pool_entry\n");
1816 			bus_dmamap_destroy(sc->bge_dmatag, dmamap);
1817 			if (sc->bge_dma64)
1818 				bus_dmamap_destroy(sc->bge_dmatag32, dmamap32);
1819 			return ENOMEM;
1820 		}
1821 		dma->dmamap = dmamap;
1822 		dma->dmamap32 = dmamap32;
1823 		SLIST_INSERT_HEAD(&sc->txdma_list, dma, link);
1824 	}
1825 alloc_done:
1826 	sc->bge_flags |= BGEF_TXRING_VALID;
1827 
1828 	return 0;
1829 }
1830 
1831 static void
1832 bge_setmulti(struct bge_softc *sc)
1833 {
1834 	struct ethercom * const ec = &sc->ethercom;
1835 	struct ether_multi	*enm;
1836 	struct ether_multistep	step;
1837 	uint32_t		hashes[4] = { 0, 0, 0, 0 };
1838 	uint32_t		h;
1839 	int			i;
1840 
1841 	KASSERT(mutex_owned(sc->sc_core_lock));
1842 	if (sc->bge_if_flags & IFF_PROMISC)
1843 		goto allmulti;
1844 
1845 	/* Now program new ones. */
1846 	ETHER_LOCK(ec);
1847 	ETHER_FIRST_MULTI(step, ec, enm);
1848 	while (enm != NULL) {
1849 		if (memcmp(enm->enm_addrlo, enm->enm_addrhi, ETHER_ADDR_LEN)) {
1850 			/*
1851 			 * We must listen to a range of multicast addresses.
1852 			 * For now, just accept all multicasts, rather than
1853 			 * trying to set only those filter bits needed to match
1854 			 * the range.  (At this time, the only use of address
1855 			 * ranges is for IP multicast routing, for which the
1856 			 * range is big enough to require all bits set.)
1857 			 */
1858 			ETHER_UNLOCK(ec);
1859 			goto allmulti;
1860 		}
1861 
1862 		h = ether_crc32_le(enm->enm_addrlo, ETHER_ADDR_LEN);
1863 
1864 		/* Just want the 7 least-significant bits. */
1865 		h &= 0x7f;
1866 
1867 		hashes[(h & 0x60) >> 5] |= 1U << (h & 0x1F);
1868 		ETHER_NEXT_MULTI(step, enm);
1869 	}
1870 	ec->ec_flags &= ~ETHER_F_ALLMULTI;
1871 	ETHER_UNLOCK(ec);
1872 
1873 	goto setit;
1874 
1875  allmulti:
1876 	ETHER_LOCK(ec);
1877 	ec->ec_flags |= ETHER_F_ALLMULTI;
1878 	ETHER_UNLOCK(ec);
1879 	hashes[0] = hashes[1] = hashes[2] = hashes[3] = 0xffffffff;
1880 
1881  setit:
1882 	for (i = 0; i < 4; i++)
1883 		CSR_WRITE_4(sc, BGE_MAR0 + (i * 4), hashes[i]);
1884 }
1885 
1886 static void
1887 bge_sig_pre_reset(struct bge_softc *sc, int type)
1888 {
1889 
1890 	/*
1891 	 * Some chips don't like this so only do this if ASF is enabled
1892 	 */
1893 	if (sc->bge_asf_mode)
1894 		bge_writemem_ind(sc, BGE_SRAM_FW_MB, BGE_SRAM_FW_MB_MAGIC);
1895 
1896 	if (sc->bge_asf_mode & ASF_NEW_HANDSHAKE) {
1897 		switch (type) {
1898 		case BGE_RESET_START:
1899 			bge_writemem_ind(sc, BGE_SRAM_FW_DRV_STATE_MB,
1900 			    BGE_FW_DRV_STATE_START);
1901 			break;
1902 		case BGE_RESET_SHUTDOWN:
1903 			bge_writemem_ind(sc, BGE_SRAM_FW_DRV_STATE_MB,
1904 			    BGE_FW_DRV_STATE_UNLOAD);
1905 			break;
1906 		case BGE_RESET_SUSPEND:
1907 			bge_writemem_ind(sc, BGE_SRAM_FW_DRV_STATE_MB,
1908 			    BGE_FW_DRV_STATE_SUSPEND);
1909 			break;
1910 		}
1911 	}
1912 
1913 	if (type == BGE_RESET_START || type == BGE_RESET_SUSPEND)
1914 		bge_ape_driver_state_change(sc, type);
1915 }
1916 
1917 static void
1918 bge_sig_post_reset(struct bge_softc *sc, int type)
1919 {
1920 
1921 	if (sc->bge_asf_mode & ASF_NEW_HANDSHAKE) {
1922 		switch (type) {
1923 		case BGE_RESET_START:
1924 			bge_writemem_ind(sc, BGE_SRAM_FW_DRV_STATE_MB,
1925 			    BGE_FW_DRV_STATE_START_DONE);
1926 			/* START DONE */
1927 			break;
1928 		case BGE_RESET_SHUTDOWN:
1929 			bge_writemem_ind(sc, BGE_SRAM_FW_DRV_STATE_MB,
1930 			    BGE_FW_DRV_STATE_UNLOAD_DONE);
1931 			break;
1932 		}
1933 	}
1934 
1935 	if (type == BGE_RESET_SHUTDOWN)
1936 		bge_ape_driver_state_change(sc, type);
1937 }
1938 
1939 static void
1940 bge_sig_legacy(struct bge_softc *sc, int type)
1941 {
1942 
1943 	if (sc->bge_asf_mode) {
1944 		switch (type) {
1945 		case BGE_RESET_START:
1946 			bge_writemem_ind(sc, BGE_SRAM_FW_DRV_STATE_MB,
1947 			    BGE_FW_DRV_STATE_START);
1948 			break;
1949 		case BGE_RESET_SHUTDOWN:
1950 			bge_writemem_ind(sc, BGE_SRAM_FW_DRV_STATE_MB,
1951 			    BGE_FW_DRV_STATE_UNLOAD);
1952 			break;
1953 		}
1954 	}
1955 }
1956 
1957 static void
1958 bge_wait_for_event_ack(struct bge_softc *sc)
1959 {
1960 	int i;
1961 
1962 	/* wait up to 2500usec */
1963 	for (i = 0; i < 250; i++) {
1964 		if (!(CSR_READ_4(sc, BGE_RX_CPU_EVENT) &
1965 			BGE_RX_CPU_DRV_EVENT))
1966 			break;
1967 		DELAY(10);
1968 	}
1969 }
1970 
1971 static void
1972 bge_stop_fw(struct bge_softc *sc)
1973 {
1974 
1975 	if (sc->bge_asf_mode) {
1976 		bge_wait_for_event_ack(sc);
1977 
1978 		bge_writemem_ind(sc, BGE_SRAM_FW_CMD_MB, BGE_FW_CMD_PAUSE);
1979 		CSR_WRITE_4_FLUSH(sc, BGE_RX_CPU_EVENT,
1980 		    CSR_READ_4(sc, BGE_RX_CPU_EVENT) | BGE_RX_CPU_DRV_EVENT);
1981 
1982 		bge_wait_for_event_ack(sc);
1983 	}
1984 }
1985 
1986 static int
1987 bge_poll_fw(struct bge_softc *sc)
1988 {
1989 	uint32_t val;
1990 	int i;
1991 
1992 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906) {
1993 		for (i = 0; i < BGE_TIMEOUT; i++) {
1994 			val = CSR_READ_4(sc, BGE_VCPU_STATUS);
1995 			if (val & BGE_VCPU_STATUS_INIT_DONE)
1996 				break;
1997 			DELAY(100);
1998 		}
1999 		if (i >= BGE_TIMEOUT) {
2000 			aprint_error_dev(sc->bge_dev, "reset timed out\n");
2001 			return -1;
2002 		}
2003 	} else {
2004 		/*
2005 		 * Poll the value location we just wrote until
2006 		 * we see the 1's complement of the magic number.
2007 		 * This indicates that the firmware initialization
2008 		 * is complete.
2009 		 * XXX 1000ms for Flash and 10000ms for SEEPROM.
2010 		 */
2011 		for (i = 0; i < BGE_TIMEOUT; i++) {
2012 			val = bge_readmem_ind(sc, BGE_SRAM_FW_MB);
2013 			if (val == ~BGE_SRAM_FW_MB_MAGIC)
2014 				break;
2015 			DELAY(10);
2016 		}
2017 
2018 		if ((i >= BGE_TIMEOUT)
2019 		    && ((sc->bge_flags & BGEF_NO_EEPROM) == 0)) {
2020 			aprint_error_dev(sc->bge_dev,
2021 			    "firmware handshake timed out, val = %x\n", val);
2022 			return -1;
2023 		}
2024 	}
2025 
2026 	if (sc->bge_chipid == BGE_CHIPID_BCM57765_A0) {
2027 		/* tg3 says we have to wait extra time */
2028 		delay(10 * 1000);
2029 	}
2030 
2031 	return 0;
2032 }
2033 
2034 int
2035 bge_phy_addr(struct bge_softc *sc)
2036 {
2037 	struct pci_attach_args *pa = &(sc->bge_pa);
2038 	int phy_addr = 1;
2039 
2040 	/*
2041 	 * PHY address mapping for various devices.
2042 	 *
2043 	 *	    | F0 Cu | F0 Sr | F1 Cu | F1 Sr |
2044 	 * ---------+-------+-------+-------+-------+
2045 	 * BCM57XX  |	1   |	X   |	X   |	X   |
2046 	 * BCM5704  |	1   |	X   |	1   |	X   |
2047 	 * BCM5717  |	1   |	8   |	2   |	9   |
2048 	 * BCM5719  |	1   |	8   |	2   |	9   |
2049 	 * BCM5720  |	1   |	8   |	2   |	9   |
2050 	 *
2051 	 *	    | F2 Cu | F2 Sr | F3 Cu | F3 Sr |
2052 	 * ---------+-------+-------+-------+-------+
2053 	 * BCM57XX  |	X   |	X   |	X   |	X   |
2054 	 * BCM5704  |	X   |	X   |	X   |	X   |
2055 	 * BCM5717  |	X   |	X   |	X   |	X   |
2056 	 * BCM5719  |	3   |	10  |	4   |	11  |
2057 	 * BCM5720  |	X   |	X   |	X   |	X   |
2058 	 *
2059 	 * Other addresses may respond but they are not
2060 	 * IEEE compliant PHYs and should be ignored.
2061 	 */
2062 	switch (BGE_ASICREV(sc->bge_chipid)) {
2063 	case BGE_ASICREV_BCM5717:
2064 	case BGE_ASICREV_BCM5719:
2065 	case BGE_ASICREV_BCM5720:
2066 		phy_addr = pa->pa_function;
2067 		if (sc->bge_chipid != BGE_CHIPID_BCM5717_A0) {
2068 			phy_addr += (CSR_READ_4(sc, BGE_SGDIG_STS) &
2069 			    BGE_SGDIGSTS_IS_SERDES) ? 8 : 1;
2070 		} else {
2071 			phy_addr += (CSR_READ_4(sc, BGE_CPMU_PHY_STRAP) &
2072 			    BGE_CPMU_PHY_STRAP_IS_SERDES) ? 8 : 1;
2073 		}
2074 	}
2075 
2076 	return phy_addr;
2077 }
2078 
2079 /*
2080  * Do endian, PCI and DMA initialization. Also check the on-board ROM
2081  * self-test results.
2082  */
2083 static int
2084 bge_chipinit(struct bge_softc *sc)
2085 {
2086 	uint32_t dma_rw_ctl, misc_ctl, mode_ctl, reg;
2087 	int i;
2088 
2089 	/* Set endianness before we access any non-PCI registers. */
2090 	misc_ctl = BGE_INIT;
2091 	if (sc->bge_flags & BGEF_TAGGED_STATUS)
2092 		misc_ctl |= BGE_PCIMISCCTL_TAGGED_STATUS;
2093 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MISC_CTL,
2094 	    misc_ctl);
2095 
2096 	/*
2097 	 * Clear the MAC statistics block in the NIC's
2098 	 * internal memory.
2099 	 */
2100 	for (i = BGE_STATS_BLOCK;
2101 	    i < BGE_STATS_BLOCK_END + 1; i += sizeof(uint32_t))
2102 		BGE_MEMWIN_WRITE(sc->sc_pc, sc->sc_pcitag, i, 0);
2103 
2104 	for (i = BGE_STATUS_BLOCK;
2105 	    i < BGE_STATUS_BLOCK_END + 1; i += sizeof(uint32_t))
2106 		BGE_MEMWIN_WRITE(sc->sc_pc, sc->sc_pcitag, i, 0);
2107 
2108 	/* 5717 workaround from tg3 */
2109 	if (sc->bge_chipid == BGE_CHIPID_BCM5717_A0) {
2110 		/* Save */
2111 		mode_ctl = CSR_READ_4(sc, BGE_MODE_CTL);
2112 
2113 		/* Temporary modify MODE_CTL to control TLP */
2114 		reg = mode_ctl & ~BGE_MODECTL_PCIE_TLPADDRMASK;
2115 		CSR_WRITE_4(sc, BGE_MODE_CTL, reg | BGE_MODECTL_PCIE_TLPADDR1);
2116 
2117 		/* Control TLP */
2118 		reg = CSR_READ_4(sc, BGE_TLP_CONTROL_REG +
2119 		    BGE_TLP_PHYCTL1);
2120 		CSR_WRITE_4(sc, BGE_TLP_CONTROL_REG + BGE_TLP_PHYCTL1,
2121 		    reg | BGE_TLP_PHYCTL1_EN_L1PLLPD);
2122 
2123 		/* Restore */
2124 		CSR_WRITE_4(sc, BGE_MODE_CTL, mode_ctl);
2125 	}
2126 
2127 	if (BGE_IS_57765_FAMILY(sc)) {
2128 		if (sc->bge_chipid == BGE_CHIPID_BCM57765_A0) {
2129 			/* Save */
2130 			mode_ctl = CSR_READ_4(sc, BGE_MODE_CTL);
2131 
2132 			/* Temporary modify MODE_CTL to control TLP */
2133 			reg = mode_ctl & ~BGE_MODECTL_PCIE_TLPADDRMASK;
2134 			CSR_WRITE_4(sc, BGE_MODE_CTL,
2135 			    reg | BGE_MODECTL_PCIE_TLPADDR1);
2136 
2137 			/* Control TLP */
2138 			reg = CSR_READ_4(sc, BGE_TLP_CONTROL_REG +
2139 			    BGE_TLP_PHYCTL5);
2140 			CSR_WRITE_4(sc, BGE_TLP_CONTROL_REG + BGE_TLP_PHYCTL5,
2141 			    reg | BGE_TLP_PHYCTL5_DIS_L2CLKREQ);
2142 
2143 			/* Restore */
2144 			CSR_WRITE_4(sc, BGE_MODE_CTL, mode_ctl);
2145 		}
2146 		if (BGE_CHIPREV(sc->bge_chipid) != BGE_CHIPREV_57765_AX) {
2147 			/*
2148 			 * For the 57766 and non Ax versions of 57765, bootcode
2149 			 * needs to setup the PCIE Fast Training Sequence (FTS)
2150 			 * value to prevent transmit hangs.
2151 			 */
2152 			reg = CSR_READ_4(sc, BGE_CPMU_PADRNG_CTL);
2153 			CSR_WRITE_4(sc, BGE_CPMU_PADRNG_CTL,
2154 			    reg | BGE_CPMU_PADRNG_CTL_RDIV2);
2155 
2156 			/* Save */
2157 			mode_ctl = CSR_READ_4(sc, BGE_MODE_CTL);
2158 
2159 			/* Temporary modify MODE_CTL to control TLP */
2160 			reg = mode_ctl & ~BGE_MODECTL_PCIE_TLPADDRMASK;
2161 			CSR_WRITE_4(sc, BGE_MODE_CTL,
2162 			    reg | BGE_MODECTL_PCIE_TLPADDR0);
2163 
2164 			/* Control TLP */
2165 			reg = CSR_READ_4(sc, BGE_TLP_CONTROL_REG +
2166 			    BGE_TLP_FTSMAX);
2167 			reg &= ~BGE_TLP_FTSMAX_MSK;
2168 			CSR_WRITE_4(sc, BGE_TLP_CONTROL_REG + BGE_TLP_FTSMAX,
2169 			    reg | BGE_TLP_FTSMAX_VAL);
2170 
2171 			/* Restore */
2172 			CSR_WRITE_4(sc, BGE_MODE_CTL, mode_ctl);
2173 		}
2174 
2175 		reg = CSR_READ_4(sc, BGE_CPMU_LSPD_10MB_CLK);
2176 		reg &= ~BGE_CPMU_LSPD_10MB_MACCLK_MASK;
2177 		reg |= BGE_CPMU_LSPD_10MB_MACCLK_6_25;
2178 		CSR_WRITE_4(sc, BGE_CPMU_LSPD_10MB_CLK, reg);
2179 	}
2180 
2181 	/* Set up the PCI DMA control register. */
2182 	dma_rw_ctl = BGE_PCI_READ_CMD | BGE_PCI_WRITE_CMD;
2183 	if (sc->bge_flags & BGEF_PCIE) {
2184 		/* Read watermark not used, 128 bytes for write. */
2185 		DPRINTFN(4, ("(%s: PCI-Express DMA setting)\n",
2186 		    device_xname(sc->bge_dev)));
2187 		if (sc->bge_mps >= 256)
2188 			dma_rw_ctl |= BGE_PCIDMARWCTL_WR_WAT_SHIFT(7);
2189 		else
2190 			dma_rw_ctl |= BGE_PCIDMARWCTL_WR_WAT_SHIFT(3);
2191 	} else if (sc->bge_flags & BGEF_PCIX) {
2192 		DPRINTFN(4, ("(:%s: PCI-X DMA setting)\n",
2193 		    device_xname(sc->bge_dev)));
2194 		/* PCI-X bus */
2195 		if (BGE_IS_5714_FAMILY(sc)) {
2196 			/* 256 bytes for read and write. */
2197 			dma_rw_ctl |= BGE_PCIDMARWCTL_RD_WAT_SHIFT(2) |
2198 			    BGE_PCIDMARWCTL_WR_WAT_SHIFT(2);
2199 
2200 			if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5780)
2201 				dma_rw_ctl |= BGE_PCIDMARWCTL_ONEDMA_ATONCE_GLOBAL;
2202 			else
2203 				dma_rw_ctl |= BGE_PCIDMARWCTL_ONEDMA_ATONCE_LOCAL;
2204 		} else if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5703) {
2205 			/*
2206 			 * In the BCM5703, the DMA read watermark should
2207 			 * be set to less than or equal to the maximum
2208 			 * memory read byte count of the PCI-X command
2209 			 * register.
2210 			 */
2211 			dma_rw_ctl |= BGE_PCIDMARWCTL_RD_WAT_SHIFT(4) |
2212 			    BGE_PCIDMARWCTL_WR_WAT_SHIFT(3);
2213 		} else if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5704) {
2214 			/* 1536 bytes for read, 384 bytes for write. */
2215 			dma_rw_ctl |= BGE_PCIDMARWCTL_RD_WAT_SHIFT(7) |
2216 			    BGE_PCIDMARWCTL_WR_WAT_SHIFT(3);
2217 		} else {
2218 			/* 384 bytes for read and write. */
2219 			dma_rw_ctl |= BGE_PCIDMARWCTL_RD_WAT_SHIFT(3) |
2220 			    BGE_PCIDMARWCTL_WR_WAT_SHIFT(3) |
2221 			    (0x0F);
2222 		}
2223 
2224 		if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5703 ||
2225 		    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5704) {
2226 			uint32_t tmp;
2227 
2228 			/* Set ONEDMA_ATONCE for hardware workaround. */
2229 			tmp = CSR_READ_4(sc, BGE_PCI_CLKCTL) & 0x1f;
2230 			if (tmp == 6 || tmp == 7)
2231 				dma_rw_ctl |=
2232 				    BGE_PCIDMARWCTL_ONEDMA_ATONCE_GLOBAL;
2233 
2234 			/* Set PCI-X DMA write workaround. */
2235 			dma_rw_ctl |= BGE_PCIDMARWCTL_ASRT_ALL_BE;
2236 		}
2237 	} else {
2238 		/* Conventional PCI bus: 256 bytes for read and write. */
2239 		DPRINTFN(4, ("(%s: PCI 2.2 DMA setting)\n",
2240 		    device_xname(sc->bge_dev)));
2241 		dma_rw_ctl |= BGE_PCIDMARWCTL_RD_WAT_SHIFT(7) |
2242 		    BGE_PCIDMARWCTL_WR_WAT_SHIFT(7);
2243 
2244 		if (BGE_ASICREV(sc->bge_chipid) != BGE_ASICREV_BCM5705 &&
2245 		    BGE_ASICREV(sc->bge_chipid) != BGE_ASICREV_BCM5750)
2246 			dma_rw_ctl |= 0x0F;
2247 	}
2248 
2249 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5700 ||
2250 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5701)
2251 		dma_rw_ctl |= BGE_PCIDMARWCTL_USE_MRM |
2252 		    BGE_PCIDMARWCTL_ASRT_ALL_BE;
2253 
2254 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5703 ||
2255 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5704)
2256 		dma_rw_ctl &= ~BGE_PCIDMARWCTL_MINDMA;
2257 
2258 	if (BGE_IS_57765_PLUS(sc)) {
2259 		dma_rw_ctl &= ~BGE_PCIDMARWCTL_DIS_CACHE_ALIGNMENT;
2260 		if (sc->bge_chipid == BGE_CHIPID_BCM57765_A0)
2261 			dma_rw_ctl &= ~BGE_PCIDMARWCTL_CRDRDR_RDMA_MRRS_MSK;
2262 
2263 		/*
2264 		 * Enable HW workaround for controllers that misinterpret
2265 		 * a status tag update and leave interrupts permanently
2266 		 * disabled.
2267 		 */
2268 		if (!BGE_IS_57765_FAMILY(sc) &&
2269 		    BGE_ASICREV(sc->bge_chipid) != BGE_ASICREV_BCM5717 &&
2270 		    BGE_ASICREV(sc->bge_chipid) != BGE_ASICREV_BCM5762)
2271 			dma_rw_ctl |= BGE_PCIDMARWCTL_TAGGED_STATUS_WA;
2272 	}
2273 
2274 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_DMA_RW_CTL,
2275 	    dma_rw_ctl);
2276 
2277 	/*
2278 	 * Set up general mode register.
2279 	 */
2280 	mode_ctl = BGE_DMA_SWAP_OPTIONS;
2281 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5720 ||
2282 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5762) {
2283 		/* Retain Host-2-BMC settings written by APE firmware. */
2284 		mode_ctl |= CSR_READ_4(sc, BGE_MODE_CTL) &
2285 		    (BGE_MODECTL_BYTESWAP_B2HRX_DATA |
2286 		    BGE_MODECTL_WORDSWAP_B2HRX_DATA |
2287 		    BGE_MODECTL_B2HRX_ENABLE | BGE_MODECTL_HTX2B_ENABLE);
2288 	}
2289 	mode_ctl |= BGE_MODECTL_MAC_ATTN_INTR | BGE_MODECTL_HOST_SEND_BDS |
2290 	    BGE_MODECTL_TX_NO_PHDR_CSUM;
2291 
2292 	/*
2293 	 * BCM5701 B5 have a bug causing data corruption when using
2294 	 * 64-bit DMA reads, which can be terminated early and then
2295 	 * completed later as 32-bit accesses, in combination with
2296 	 * certain bridges.
2297 	 */
2298 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5701 &&
2299 	    sc->bge_chipid == BGE_CHIPID_BCM5701_B5)
2300 		mode_ctl |= BGE_MODECTL_FORCE_PCI32;
2301 
2302 	/*
2303 	 * Tell the firmware the driver is running
2304 	 */
2305 	if (sc->bge_asf_mode & ASF_STACKUP)
2306 		mode_ctl |= BGE_MODECTL_STACKUP;
2307 
2308 	CSR_WRITE_4(sc, BGE_MODE_CTL, mode_ctl);
2309 
2310 	/*
2311 	 * Disable memory write invalidate.  Apparently it is not supported
2312 	 * properly by these devices.
2313 	 */
2314 	PCI_CLRBIT(sc->sc_pc, sc->sc_pcitag, PCI_COMMAND_STATUS_REG,
2315 		   PCI_COMMAND_INVALIDATE_ENABLE);
2316 
2317 #ifdef __brokenalpha__
2318 	/*
2319 	 * Must insure that we do not cross an 8K (bytes) boundary
2320 	 * for DMA reads.  Our highest limit is 1K bytes.  This is a
2321 	 * restriction on some ALPHA platforms with early revision
2322 	 * 21174 PCI chipsets, such as the AlphaPC 164lx
2323 	 */
2324 	PCI_SETBIT(sc, BGE_PCI_DMA_RW_CTL, BGE_PCI_READ_BNDRY_1024, 4);
2325 #endif
2326 
2327 	/* Set the timer prescaler (always 66MHz) */
2328 	CSR_WRITE_4_FLUSH(sc, BGE_MISC_CFG, BGE_32BITTIME_66MHZ);
2329 
2330 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906) {
2331 		DELAY(40);	/* XXX */
2332 
2333 		/* Put PHY into ready state */
2334 		BGE_CLRBIT_FLUSH(sc, BGE_MISC_CFG, BGE_MISCCFG_EPHY_IDDQ);
2335 		DELAY(40);
2336 	}
2337 
2338 	return 0;
2339 }
2340 
2341 static int
2342 bge_blockinit(struct bge_softc *sc)
2343 {
2344 	volatile struct bge_rcb	 *rcb;
2345 	bus_size_t rcb_addr;
2346 	struct ifnet * const ifp = &sc->ethercom.ec_if;
2347 	bge_hostaddr taddr;
2348 	uint32_t	dmactl, rdmareg, mimode, val;
2349 	int		i, limit;
2350 
2351 	/*
2352 	 * Initialize the memory window pointer register so that
2353 	 * we can access the first 32K of internal NIC RAM. This will
2354 	 * allow us to set up the TX send ring RCBs and the RX return
2355 	 * ring RCBs, plus other things which live in NIC memory.
2356 	 */
2357 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MEMWIN_BASEADDR, 0);
2358 
2359 	if (!BGE_IS_5705_PLUS(sc)) {
2360 		/* 57XX step 33 */
2361 		/* Configure mbuf memory pool */
2362 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_BASEADDR, BGE_BUFFPOOL_1);
2363 
2364 		if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5704)
2365 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_LEN, 0x10000);
2366 		else
2367 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_LEN, 0x18000);
2368 
2369 		/* 57XX step 34 */
2370 		/* Configure DMA resource pool */
2371 		CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_BASEADDR,
2372 		    BGE_DMA_DESCRIPTORS);
2373 		CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_LEN, 0x2000);
2374 	}
2375 
2376 	/* 5718 step 11, 57XX step 35 */
2377 	/*
2378 	 * Configure mbuf pool watermarks. New broadcom docs strongly
2379 	 * recommend these.
2380 	 */
2381 	if (BGE_IS_5717_PLUS(sc)) {
2382 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 0x0);
2383 		if (ifp->if_mtu > ETHERMTU) {
2384 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x7e);
2385 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0xea);
2386 		} else {
2387 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x2a);
2388 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0xa0);
2389 		}
2390 	} else if (BGE_IS_5705_PLUS(sc)) {
2391 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 0x0);
2392 
2393 		if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906) {
2394 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x04);
2395 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0x10);
2396 		} else {
2397 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x10);
2398 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0x60);
2399 		}
2400 	} else {
2401 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 0x50);
2402 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x20);
2403 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0x60);
2404 	}
2405 
2406 	/* 57XX step 36 */
2407 	/* Configure DMA resource watermarks */
2408 	CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_LOWAT, 5);
2409 	CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_HIWAT, 10);
2410 
2411 	/* 5718 step 13, 57XX step 38 */
2412 	/* Enable buffer manager */
2413 	val = BGE_BMANMODE_ENABLE | BGE_BMANMODE_ATTN;
2414 	/*
2415 	 * Change the arbitration algorithm of TXMBUF read request to
2416 	 * round-robin instead of priority based for BCM5719.  When
2417 	 * TXFIFO is almost empty, RDMA will hold its request until
2418 	 * TXFIFO is not almost empty.
2419 	 */
2420 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5719)
2421 		val |= BGE_BMANMODE_NO_TX_UNDERRUN;
2422 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5717 ||
2423 		sc->bge_chipid == BGE_CHIPID_BCM5719_A0 ||
2424 		sc->bge_chipid == BGE_CHIPID_BCM5720_A0)
2425 		val |= BGE_BMANMODE_LOMBUF_ATTN;
2426 	CSR_WRITE_4(sc, BGE_BMAN_MODE, val);
2427 
2428 	/* 57XX step 39 */
2429 	/* Poll for buffer manager start indication */
2430 	for (i = 0; i < BGE_TIMEOUT * 2; i++) {
2431 		DELAY(10);
2432 		if (CSR_READ_4(sc, BGE_BMAN_MODE) & BGE_BMANMODE_ENABLE)
2433 			break;
2434 	}
2435 
2436 	if (i == BGE_TIMEOUT * 2) {
2437 		aprint_error_dev(sc->bge_dev,
2438 		    "buffer manager failed to start\n");
2439 		return ENXIO;
2440 	}
2441 
2442 	/* 57XX step 40 */
2443 	/* Enable flow-through queues */
2444 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0xFFFFFFFF);
2445 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0);
2446 
2447 	/* Wait until queue initialization is complete */
2448 	for (i = 0; i < BGE_TIMEOUT * 2; i++) {
2449 		if (CSR_READ_4(sc, BGE_FTQ_RESET) == 0)
2450 			break;
2451 		DELAY(10);
2452 	}
2453 
2454 	if (i == BGE_TIMEOUT * 2) {
2455 		aprint_error_dev(sc->bge_dev,
2456 		    "flow-through queue init failed\n");
2457 		return ENXIO;
2458 	}
2459 
2460 	/*
2461 	 * Summary of rings supported by the controller:
2462 	 *
2463 	 * Standard Receive Producer Ring
2464 	 * - This ring is used to feed receive buffers for "standard"
2465 	 *   sized frames (typically 1536 bytes) to the controller.
2466 	 *
2467 	 * Jumbo Receive Producer Ring
2468 	 * - This ring is used to feed receive buffers for jumbo sized
2469 	 *   frames (i.e. anything bigger than the "standard" frames)
2470 	 *   to the controller.
2471 	 *
2472 	 * Mini Receive Producer Ring
2473 	 * - This ring is used to feed receive buffers for "mini"
2474 	 *   sized frames to the controller.
2475 	 * - This feature required external memory for the controller
2476 	 *   but was never used in a production system.  Should always
2477 	 *   be disabled.
2478 	 *
2479 	 * Receive Return Ring
2480 	 * - After the controller has placed an incoming frame into a
2481 	 *   receive buffer that buffer is moved into a receive return
2482 	 *   ring.  The driver is then responsible to passing the
2483 	 *   buffer up to the stack.  Many versions of the controller
2484 	 *   support multiple RR rings.
2485 	 *
2486 	 * Send Ring
2487 	 * - This ring is used for outgoing frames.  Many versions of
2488 	 *   the controller support multiple send rings.
2489 	 */
2490 
2491 	/* 5718 step 15, 57XX step 41 */
2492 	/* Initialize the standard RX ring control block */
2493 	rcb = &sc->bge_rdata->bge_info.bge_std_rx_rcb;
2494 	BGE_HOSTADDR(rcb->bge_hostaddr, BGE_RING_DMA_ADDR(sc, bge_rx_std_ring));
2495 	/* 5718 step 16 */
2496 	if (BGE_IS_57765_PLUS(sc)) {
2497 		/*
2498 		 * Bits 31-16: Programmable ring size (2048, 1024, 512, .., 32)
2499 		 * Bits 15-2 : Maximum RX frame size
2500 		 * Bit 1     : 1 = Ring Disabled, 0 = Ring Enabled
2501 		 * Bit 0     : Reserved
2502 		 */
2503 		rcb->bge_maxlen_flags =
2504 		    BGE_RCB_MAXLEN_FLAGS(512, BGE_MAX_FRAMELEN << 2);
2505 	} else if (BGE_IS_5705_PLUS(sc)) {
2506 		/*
2507 		 * Bits 31-16: Programmable ring size (512, 256, 128, 64, 32)
2508 		 * Bits 15-2 : Reserved (should be 0)
2509 		 * Bit 1     : 1 = Ring Disabled, 0 = Ring Enabled
2510 		 * Bit 0     : Reserved
2511 		 */
2512 		rcb->bge_maxlen_flags = BGE_RCB_MAXLEN_FLAGS(512, 0);
2513 	} else {
2514 		/*
2515 		 * Ring size is always XXX entries
2516 		 * Bits 31-16: Maximum RX frame size
2517 		 * Bits 15-2 : Reserved (should be 0)
2518 		 * Bit 1     : 1 = Ring Disabled, 0 = Ring Enabled
2519 		 * Bit 0     : Reserved
2520 		 */
2521 		rcb->bge_maxlen_flags =
2522 		    BGE_RCB_MAXLEN_FLAGS(BGE_MAX_FRAMELEN, 0);
2523 	}
2524 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5717 ||
2525 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5719 ||
2526 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5720)
2527 		rcb->bge_nicaddr = BGE_STD_RX_RINGS_5717;
2528 	else
2529 		rcb->bge_nicaddr = BGE_STD_RX_RINGS;
2530 	/* Write the standard receive producer ring control block. */
2531 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_HADDR_HI, rcb->bge_hostaddr.bge_addr_hi);
2532 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_HADDR_LO, rcb->bge_hostaddr.bge_addr_lo);
2533 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_MAXLEN_FLAGS, rcb->bge_maxlen_flags);
2534 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_NICADDR, rcb->bge_nicaddr);
2535 
2536 	/* Reset the standard receive producer ring producer index. */
2537 	bge_writembx(sc, BGE_MBX_RX_STD_PROD_LO, 0);
2538 
2539 	/* 57XX step 42 */
2540 	/*
2541 	 * Initialize the jumbo RX ring control block
2542 	 * We set the 'ring disabled' bit in the flags
2543 	 * field until we're actually ready to start
2544 	 * using this ring (i.e. once we set the MTU
2545 	 * high enough to require it).
2546 	 */
2547 	if (BGE_IS_JUMBO_CAPABLE(sc)) {
2548 		rcb = &sc->bge_rdata->bge_info.bge_jumbo_rx_rcb;
2549 		BGE_HOSTADDR(rcb->bge_hostaddr,
2550 		    BGE_RING_DMA_ADDR(sc, bge_rx_jumbo_ring));
2551 		rcb->bge_maxlen_flags = BGE_RCB_MAXLEN_FLAGS(0,
2552 		    BGE_RCB_FLAG_USE_EXT_RX_BD | BGE_RCB_FLAG_RING_DISABLED);
2553 		if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5717 ||
2554 		    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5719 ||
2555 		    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5720)
2556 			rcb->bge_nicaddr = BGE_JUMBO_RX_RINGS_5717;
2557 		else
2558 			rcb->bge_nicaddr = BGE_JUMBO_RX_RINGS;
2559 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_HADDR_HI,
2560 		    rcb->bge_hostaddr.bge_addr_hi);
2561 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_HADDR_LO,
2562 		    rcb->bge_hostaddr.bge_addr_lo);
2563 		/* Program the jumbo receive producer ring RCB parameters. */
2564 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_MAXLEN_FLAGS,
2565 		    rcb->bge_maxlen_flags);
2566 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_NICADDR, rcb->bge_nicaddr);
2567 		/* Reset the jumbo receive producer ring producer index. */
2568 		bge_writembx(sc, BGE_MBX_RX_JUMBO_PROD_LO, 0);
2569 	}
2570 
2571 	/* 57XX step 43 */
2572 	/* Disable the mini receive producer ring RCB. */
2573 	if (BGE_IS_5700_FAMILY(sc)) {
2574 		/* Set up dummy disabled mini ring RCB */
2575 		rcb = &sc->bge_rdata->bge_info.bge_mini_rx_rcb;
2576 		rcb->bge_maxlen_flags =
2577 		    BGE_RCB_MAXLEN_FLAGS(0, BGE_RCB_FLAG_RING_DISABLED);
2578 		CSR_WRITE_4(sc, BGE_RX_MINI_RCB_MAXLEN_FLAGS,
2579 		    rcb->bge_maxlen_flags);
2580 		/* Reset the mini receive producer ring producer index. */
2581 		bge_writembx(sc, BGE_MBX_RX_MINI_PROD_LO, 0);
2582 
2583 		bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
2584 		    offsetof(struct bge_ring_data, bge_info),
2585 		    sizeof(struct bge_gib),
2586 		    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
2587 	}
2588 
2589 	/* Choose de-pipeline mode for BCM5906 A0, A1 and A2. */
2590 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906) {
2591 		if (sc->bge_chipid == BGE_CHIPID_BCM5906_A0 ||
2592 		    sc->bge_chipid == BGE_CHIPID_BCM5906_A1 ||
2593 		    sc->bge_chipid == BGE_CHIPID_BCM5906_A2)
2594 			CSR_WRITE_4(sc, BGE_ISO_PKT_TX,
2595 			    (CSR_READ_4(sc, BGE_ISO_PKT_TX) & ~3) | 2);
2596 	}
2597 	/* 5718 step 14, 57XX step 44 */
2598 	/*
2599 	 * The BD ring replenish thresholds control how often the
2600 	 * hardware fetches new BD's from the producer rings in host
2601 	 * memory.  Setting the value too low on a busy system can
2602 	 * starve the hardware and reduce the throughput.
2603 	 *
2604 	 * Set the BD ring replenish thresholds. The recommended
2605 	 * values are 1/8th the number of descriptors allocated to
2606 	 * each ring, but since we try to avoid filling the entire
2607 	 * ring we set these to the minimal value of 8.  This needs to
2608 	 * be done on several of the supported chip revisions anyway,
2609 	 * to work around HW bugs.
2610 	 */
2611 	CSR_WRITE_4(sc, BGE_RBDI_STD_REPL_THRESH, 8);
2612 	if (BGE_IS_JUMBO_CAPABLE(sc))
2613 		CSR_WRITE_4(sc, BGE_RBDI_JUMBO_REPL_THRESH, 8);
2614 
2615 	/* 5718 step 18 */
2616 	if (BGE_IS_5717_PLUS(sc)) {
2617 		CSR_WRITE_4(sc, BGE_STD_REPL_LWM, 4);
2618 		CSR_WRITE_4(sc, BGE_JUMBO_REPL_LWM, 4);
2619 	}
2620 
2621 	/* 57XX step 45 */
2622 	/*
2623 	 * Disable all send rings by setting the 'ring disabled' bit
2624 	 * in the flags field of all the TX send ring control blocks,
2625 	 * located in NIC memory.
2626 	 */
2627 	if (BGE_IS_5700_FAMILY(sc)) {
2628 		/* 5700 to 5704 had 16 send rings. */
2629 		limit = BGE_TX_RINGS_EXTSSRAM_MAX;
2630 	} else if (BGE_IS_5717_PLUS(sc)) {
2631 		limit = BGE_TX_RINGS_5717_MAX;
2632 	} else if (BGE_IS_57765_FAMILY(sc) ||
2633 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5762) {
2634 		limit = BGE_TX_RINGS_57765_MAX;
2635 	} else
2636 		limit = 1;
2637 	rcb_addr = BGE_MEMWIN_START + BGE_SEND_RING_RCB;
2638 	for (i = 0; i < limit; i++) {
2639 		RCB_WRITE_4(sc, rcb_addr, bge_maxlen_flags,
2640 		    BGE_RCB_MAXLEN_FLAGS(0, BGE_RCB_FLAG_RING_DISABLED));
2641 		RCB_WRITE_4(sc, rcb_addr, bge_nicaddr, 0);
2642 		rcb_addr += sizeof(struct bge_rcb);
2643 	}
2644 
2645 	/* 57XX step 46 and 47 */
2646 	/* Configure send ring RCB 0 (we use only the first ring) */
2647 	rcb_addr = BGE_MEMWIN_START + BGE_SEND_RING_RCB;
2648 	BGE_HOSTADDR(taddr, BGE_RING_DMA_ADDR(sc, bge_tx_ring));
2649 	RCB_WRITE_4(sc, rcb_addr, bge_hostaddr.bge_addr_hi, taddr.bge_addr_hi);
2650 	RCB_WRITE_4(sc, rcb_addr, bge_hostaddr.bge_addr_lo, taddr.bge_addr_lo);
2651 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5717 ||
2652 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5719 ||
2653 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5720)
2654 		RCB_WRITE_4(sc, rcb_addr, bge_nicaddr, BGE_SEND_RING_5717);
2655 	else
2656 		RCB_WRITE_4(sc, rcb_addr, bge_nicaddr,
2657 		    BGE_NIC_TXRING_ADDR(0, BGE_TX_RING_CNT));
2658 	RCB_WRITE_4(sc, rcb_addr, bge_maxlen_flags,
2659 	    BGE_RCB_MAXLEN_FLAGS(BGE_TX_RING_CNT, 0));
2660 
2661 	/* 57XX step 48 */
2662 	/*
2663 	 * Disable all receive return rings by setting the
2664 	 * 'ring disabled' bit in the flags field of all the receive
2665 	 * return ring control blocks, located in NIC memory.
2666 	 */
2667 	if (BGE_IS_5717_PLUS(sc)) {
2668 		/* Should be 17, use 16 until we get an SRAM map. */
2669 		limit = 16;
2670 	} else if (BGE_IS_5700_FAMILY(sc))
2671 		limit = BGE_RX_RINGS_MAX;
2672 	else if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5755 ||
2673 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5762 ||
2674 	    BGE_IS_57765_FAMILY(sc))
2675 		limit = 4;
2676 	else
2677 		limit = 1;
2678 	/* Disable all receive return rings */
2679 	rcb_addr = BGE_MEMWIN_START + BGE_RX_RETURN_RING_RCB;
2680 	for (i = 0; i < limit; i++) {
2681 		RCB_WRITE_4(sc, rcb_addr, bge_hostaddr.bge_addr_hi, 0);
2682 		RCB_WRITE_4(sc, rcb_addr, bge_hostaddr.bge_addr_lo, 0);
2683 		RCB_WRITE_4(sc, rcb_addr, bge_maxlen_flags,
2684 		    BGE_RCB_MAXLEN_FLAGS(sc->bge_return_ring_cnt,
2685 			BGE_RCB_FLAG_RING_DISABLED));
2686 		RCB_WRITE_4(sc, rcb_addr, bge_nicaddr, 0);
2687 		bge_writembx(sc, BGE_MBX_RX_CONS0_LO +
2688 		    (i * (sizeof(uint64_t))), 0);
2689 		rcb_addr += sizeof(struct bge_rcb);
2690 	}
2691 
2692 	/* 57XX step 49 */
2693 	/*
2694 	 * Set up receive return ring 0.  Note that the NIC address
2695 	 * for RX return rings is 0x0.  The return rings live entirely
2696 	 * within the host, so the nicaddr field in the RCB isn't used.
2697 	 */
2698 	rcb_addr = BGE_MEMWIN_START + BGE_RX_RETURN_RING_RCB;
2699 	BGE_HOSTADDR(taddr, BGE_RING_DMA_ADDR(sc, bge_rx_return_ring));
2700 	RCB_WRITE_4(sc, rcb_addr, bge_hostaddr.bge_addr_hi, taddr.bge_addr_hi);
2701 	RCB_WRITE_4(sc, rcb_addr, bge_hostaddr.bge_addr_lo, taddr.bge_addr_lo);
2702 	RCB_WRITE_4(sc, rcb_addr, bge_nicaddr, 0x00000000);
2703 	RCB_WRITE_4(sc, rcb_addr, bge_maxlen_flags,
2704 	    BGE_RCB_MAXLEN_FLAGS(sc->bge_return_ring_cnt, 0));
2705 
2706 	/* 5718 step 24, 57XX step 53 */
2707 	/* Set random backoff seed for TX */
2708 	CSR_WRITE_4(sc, BGE_TX_RANDOM_BACKOFF,
2709 	    (CLLADDR(ifp->if_sadl)[0] + CLLADDR(ifp->if_sadl)[1] +
2710 		CLLADDR(ifp->if_sadl)[2] + CLLADDR(ifp->if_sadl)[3] +
2711 		CLLADDR(ifp->if_sadl)[4] + CLLADDR(ifp->if_sadl)[5]) &
2712 	    BGE_TX_BACKOFF_SEED_MASK);
2713 
2714 	/* 5718 step 26, 57XX step 55 */
2715 	/* Set inter-packet gap */
2716 	val = 0x2620;
2717 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5720 ||
2718 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5762)
2719 		val |= CSR_READ_4(sc, BGE_TX_LENGTHS) &
2720 		    (BGE_TXLEN_JMB_FRM_LEN_MSK | BGE_TXLEN_CNT_DN_VAL_MSK);
2721 	CSR_WRITE_4(sc, BGE_TX_LENGTHS, val);
2722 
2723 	/* 5718 step 27, 57XX step 56 */
2724 	/*
2725 	 * Specify which ring to use for packets that don't match
2726 	 * any RX rules.
2727 	 */
2728 	CSR_WRITE_4(sc, BGE_RX_RULES_CFG, 0x08);
2729 
2730 	/* 5718 step 28, 57XX step 57 */
2731 	/*
2732 	 * Configure number of RX lists. One interrupt distribution
2733 	 * list, sixteen active lists, one bad frames class.
2734 	 */
2735 	CSR_WRITE_4(sc, BGE_RXLP_CFG, 0x181);
2736 
2737 	/* 5718 step 29, 57XX step 58 */
2738 	/* Initialize RX list placement stats mask. */
2739 	if (BGE_IS_575X_PLUS(sc)) {
2740 		val = CSR_READ_4(sc, BGE_RXLP_STATS_ENABLE_MASK);
2741 		val &= ~BGE_RXLPSTATCONTROL_DACK_FIX;
2742 		CSR_WRITE_4(sc, BGE_RXLP_STATS_ENABLE_MASK, val);
2743 	} else
2744 		CSR_WRITE_4(sc, BGE_RXLP_STATS_ENABLE_MASK, 0x007FFFFF);
2745 
2746 	/* 5718 step 30, 57XX step 59 */
2747 	CSR_WRITE_4(sc, BGE_RXLP_STATS_CTL, 0x1);
2748 
2749 	/* 5718 step 33, 57XX step 62 */
2750 	/* Disable host coalescing until we get it set up */
2751 	CSR_WRITE_4(sc, BGE_HCC_MODE, 0x00000000);
2752 
2753 	/* 5718 step 34, 57XX step 63 */
2754 	/* Poll to make sure it's shut down. */
2755 	for (i = 0; i < BGE_TIMEOUT * 2; i++) {
2756 		DELAY(10);
2757 		if (!(CSR_READ_4(sc, BGE_HCC_MODE) & BGE_HCCMODE_ENABLE))
2758 			break;
2759 	}
2760 
2761 	if (i == BGE_TIMEOUT * 2) {
2762 		aprint_error_dev(sc->bge_dev,
2763 		    "host coalescing engine failed to idle\n");
2764 		return ENXIO;
2765 	}
2766 
2767 	/* 5718 step 35, 36, 37 */
2768 	/* Set up host coalescing defaults */
2769 	CSR_WRITE_4(sc, BGE_HCC_RX_COAL_TICKS, sc->bge_rx_coal_ticks);
2770 	CSR_WRITE_4(sc, BGE_HCC_TX_COAL_TICKS, sc->bge_tx_coal_ticks);
2771 	CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS, sc->bge_rx_max_coal_bds);
2772 	CSR_WRITE_4(sc, BGE_HCC_TX_MAX_COAL_BDS, sc->bge_tx_max_coal_bds);
2773 	if (!(BGE_IS_5705_PLUS(sc))) {
2774 		CSR_WRITE_4(sc, BGE_HCC_RX_COAL_TICKS_INT, 0);
2775 		CSR_WRITE_4(sc, BGE_HCC_TX_COAL_TICKS_INT, 0);
2776 	}
2777 	CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS_INT, 0);
2778 	CSR_WRITE_4(sc, BGE_HCC_TX_MAX_COAL_BDS_INT, 0);
2779 
2780 	/* Set up address of statistics block */
2781 	if (BGE_IS_5700_FAMILY(sc)) {
2782 		BGE_HOSTADDR(taddr, BGE_RING_DMA_ADDR(sc, bge_info.bge_stats));
2783 		CSR_WRITE_4(sc, BGE_HCC_STATS_TICKS, sc->bge_stat_ticks);
2784 		CSR_WRITE_4(sc, BGE_HCC_STATS_BASEADDR, BGE_STATS_BLOCK);
2785 		CSR_WRITE_4(sc, BGE_HCC_STATS_ADDR_HI, taddr.bge_addr_hi);
2786 		CSR_WRITE_4(sc, BGE_HCC_STATS_ADDR_LO, taddr.bge_addr_lo);
2787 	}
2788 
2789 	/* 5718 step 38 */
2790 	/* Set up address of status block */
2791 	BGE_HOSTADDR(taddr, BGE_RING_DMA_ADDR(sc, bge_status_block));
2792 	CSR_WRITE_4(sc, BGE_HCC_STATUSBLK_BASEADDR, BGE_STATUS_BLOCK);
2793 	CSR_WRITE_4(sc, BGE_HCC_STATUSBLK_ADDR_HI, taddr.bge_addr_hi);
2794 	CSR_WRITE_4(sc, BGE_HCC_STATUSBLK_ADDR_LO, taddr.bge_addr_lo);
2795 	sc->bge_rdata->bge_status_block.bge_idx[0].bge_rx_prod_idx = 0;
2796 	sc->bge_rdata->bge_status_block.bge_idx[0].bge_tx_cons_idx = 0;
2797 
2798 	/* Set up status block size. */
2799 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5700 &&
2800 	    sc->bge_chipid != BGE_CHIPID_BCM5700_C0) {
2801 		val = BGE_STATBLKSZ_FULL;
2802 		bzero(&sc->bge_rdata->bge_status_block, BGE_STATUS_BLK_SZ);
2803 	} else {
2804 		val = BGE_STATBLKSZ_32BYTE;
2805 		bzero(&sc->bge_rdata->bge_status_block, 32);
2806 	}
2807 
2808 	/* 5718 step 39, 57XX step 73 */
2809 	/* Turn on host coalescing state machine */
2810 	CSR_WRITE_4(sc, BGE_HCC_MODE, val | BGE_HCCMODE_ENABLE);
2811 
2812 	/* 5718 step 40, 57XX step 74 */
2813 	/* Turn on RX BD completion state machine and enable attentions */
2814 	CSR_WRITE_4(sc, BGE_RBDC_MODE,
2815 	    BGE_RBDCMODE_ENABLE | BGE_RBDCMODE_ATTN);
2816 
2817 	/* 5718 step 41, 57XX step 75 */
2818 	/* Turn on RX list placement state machine */
2819 	CSR_WRITE_4(sc, BGE_RXLP_MODE, BGE_RXLPMODE_ENABLE);
2820 
2821 	/* 57XX step 76 */
2822 	/* Turn on RX list selector state machine. */
2823 	if (!(BGE_IS_5705_PLUS(sc)))
2824 		CSR_WRITE_4(sc, BGE_RXLS_MODE, BGE_RXLSMODE_ENABLE);
2825 
2826 	val = BGE_MACMODE_TXDMA_ENB | BGE_MACMODE_RXDMA_ENB |
2827 	    BGE_MACMODE_RX_STATS_CLEAR | BGE_MACMODE_TX_STATS_CLEAR |
2828 	    BGE_MACMODE_RX_STATS_ENB | BGE_MACMODE_TX_STATS_ENB |
2829 	    BGE_MACMODE_FRMHDR_DMA_ENB;
2830 
2831 	if (sc->bge_flags & BGEF_FIBER_TBI)
2832 		val |= BGE_PORTMODE_TBI;
2833 	else if (sc->bge_flags & BGEF_FIBER_MII)
2834 		val |= BGE_PORTMODE_GMII;
2835 	else
2836 		val |= BGE_PORTMODE_MII;
2837 
2838 	/* 5718 step 42 and 43, 57XX step 77 and 78 */
2839 	/* Allow APE to send/receive frames. */
2840 	if ((sc->bge_mfw_flags & BGE_MFW_ON_APE) != 0)
2841 		val |= BGE_MACMODE_APE_RX_EN | BGE_MACMODE_APE_TX_EN;
2842 
2843 	/* Turn on DMA, clear stats */
2844 	CSR_WRITE_4_FLUSH(sc, BGE_MAC_MODE, val);
2845 	/* 5718 step 44 */
2846 	DELAY(40);
2847 
2848 	/* 5718 step 45, 57XX step 79 */
2849 	/* Set misc. local control, enable interrupts on attentions */
2850 	BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_INTR_ONATTN);
2851 	if (BGE_IS_5717_PLUS(sc)) {
2852 		CSR_READ_4(sc, BGE_MISC_LOCAL_CTL); /* Flush */
2853 		/* 5718 step 46 */
2854 		DELAY(100);
2855 	}
2856 
2857 	/* 57XX step 81 */
2858 	/* Turn on DMA completion state machine */
2859 	if (!(BGE_IS_5705_PLUS(sc)))
2860 		CSR_WRITE_4(sc, BGE_DMAC_MODE, BGE_DMACMODE_ENABLE);
2861 
2862 	/* 5718 step 47, 57XX step 82 */
2863 	val = BGE_WDMAMODE_ENABLE | BGE_WDMAMODE_ALL_ATTNS;
2864 
2865 	/* 5718 step 48 */
2866 	/* Enable host coalescing bug fix. */
2867 	if (BGE_IS_5755_PLUS(sc))
2868 		val |= BGE_WDMAMODE_STATUS_TAG_FIX;
2869 
2870 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5785)
2871 		val |= BGE_WDMAMODE_BURST_ALL_DATA;
2872 
2873 	/* Turn on write DMA state machine */
2874 	CSR_WRITE_4_FLUSH(sc, BGE_WDMA_MODE, val);
2875 	/* 5718 step 49 */
2876 	DELAY(40);
2877 
2878 	val = BGE_RDMAMODE_ENABLE | BGE_RDMAMODE_ALL_ATTNS;
2879 
2880 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5717)
2881 		val |= BGE_RDMAMODE_MULT_DMA_RD_DIS;
2882 
2883 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5784 ||
2884 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5785 ||
2885 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM57780)
2886 		val |= BGE_RDMAMODE_BD_SBD_CRPT_ATTN |
2887 		    BGE_RDMAMODE_MBUF_RBD_CRPT_ATTN |
2888 		    BGE_RDMAMODE_MBUF_SBD_CRPT_ATTN;
2889 
2890 	if (sc->bge_flags & BGEF_PCIE)
2891 		val |= BGE_RDMAMODE_FIFO_LONG_BURST;
2892 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM57766) {
2893 		if (ifp->if_mtu <= ETHERMTU)
2894 			val |= BGE_RDMAMODE_JMB_2K_MMRR;
2895 	}
2896 	if (sc->bge_flags & BGEF_TSO) {
2897 		val |= BGE_RDMAMODE_TSO4_ENABLE;
2898 		if (BGE_IS_5717_PLUS(sc))
2899 			val |= BGE_RDMAMODE_TSO6_ENABLE;
2900 	}
2901 
2902 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5720 ||
2903 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5762) {
2904 		val |= CSR_READ_4(sc, BGE_RDMA_MODE) &
2905 		    BGE_RDMAMODE_H2BNC_VLAN_DET;
2906 		/*
2907 		 * Allow multiple outstanding read requests from
2908 		 * non-LSO read DMA engine.
2909 		 */
2910 		val &= ~BGE_RDMAMODE_MULT_DMA_RD_DIS;
2911 	}
2912 
2913 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5761 ||
2914 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5784 ||
2915 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5785 ||
2916 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM57780 ||
2917 	    BGE_IS_57765_PLUS(sc)) {
2918 		if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5762)
2919 			rdmareg = BGE_RDMA_RSRVCTRL_REG2;
2920 		else
2921 			rdmareg = BGE_RDMA_RSRVCTRL;
2922 		dmactl = CSR_READ_4(sc, rdmareg);
2923 		/*
2924 		 * Adjust tx margin to prevent TX data corruption and
2925 		 * fix internal FIFO overflow.
2926 		 */
2927 		if (sc->bge_chipid == BGE_CHIPID_BCM5719_A0 ||
2928 		    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5762) {
2929 			dmactl &= ~(BGE_RDMA_RSRVCTRL_FIFO_LWM_MASK |
2930 			    BGE_RDMA_RSRVCTRL_FIFO_HWM_MASK |
2931 			    BGE_RDMA_RSRVCTRL_TXMRGN_MASK);
2932 			dmactl |= BGE_RDMA_RSRVCTRL_FIFO_LWM_1_5K |
2933 			    BGE_RDMA_RSRVCTRL_FIFO_HWM_1_5K |
2934 			    BGE_RDMA_RSRVCTRL_TXMRGN_320B;
2935 		}
2936 		/*
2937 		 * Enable fix for read DMA FIFO overruns.
2938 		 * The fix is to limit the number of RX BDs
2939 		 * the hardware would fetch at a time.
2940 		 */
2941 		CSR_WRITE_4(sc, rdmareg, dmactl |
2942 		    BGE_RDMA_RSRVCTRL_FIFO_OFLW_FIX);
2943 	}
2944 
2945 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5719) {
2946 		CSR_WRITE_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL,
2947 		    CSR_READ_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL) |
2948 		    BGE_RDMA_LSO_CRPTEN_CTRL_BLEN_BD_4K |
2949 		    BGE_RDMA_LSO_CRPTEN_CTRL_BLEN_LSO_4K);
2950 	} else if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5720) {
2951 		/*
2952 		 * Allow 4KB burst length reads for non-LSO frames.
2953 		 * Enable 512B burst length reads for buffer descriptors.
2954 		 */
2955 		CSR_WRITE_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL,
2956 		    CSR_READ_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL) |
2957 		    BGE_RDMA_LSO_CRPTEN_CTRL_BLEN_BD_512 |
2958 		    BGE_RDMA_LSO_CRPTEN_CTRL_BLEN_LSO_4K);
2959 	} else if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5762) {
2960 		CSR_WRITE_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL_REG2,
2961 		    CSR_READ_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL_REG2) |
2962 		    BGE_RDMA_LSO_CRPTEN_CTRL_BLEN_BD_4K |
2963 		    BGE_RDMA_LSO_CRPTEN_CTRL_BLEN_LSO_4K);
2964 	}
2965 	/* Turn on read DMA state machine */
2966 	CSR_WRITE_4_FLUSH(sc, BGE_RDMA_MODE, val);
2967 	/* 5718 step 52 */
2968 	delay(40);
2969 
2970 	if (sc->bge_flags & BGEF_RDMA_BUG) {
2971 		for (i = 0; i < BGE_NUM_RDMA_CHANNELS / 2; i++) {
2972 			val = CSR_READ_4(sc, BGE_RDMA_LENGTH + i * 4);
2973 			if ((val & 0xFFFF) > BGE_FRAMELEN)
2974 				break;
2975 			if (((val >> 16) & 0xFFFF) > BGE_FRAMELEN)
2976 				break;
2977 		}
2978 		if (i != BGE_NUM_RDMA_CHANNELS / 2) {
2979 			val = CSR_READ_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL);
2980 			if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5719)
2981 				val |= BGE_RDMA_TX_LENGTH_WA_5719;
2982 			else
2983 				val |= BGE_RDMA_TX_LENGTH_WA_5720;
2984 			CSR_WRITE_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL, val);
2985 		}
2986 	}
2987 
2988 	/* 5718 step 56, 57XX step 84 */
2989 	/* Turn on RX data completion state machine */
2990 	CSR_WRITE_4(sc, BGE_RDC_MODE, BGE_RDCMODE_ENABLE);
2991 
2992 	/* Turn on RX data and RX BD initiator state machine */
2993 	CSR_WRITE_4(sc, BGE_RDBDI_MODE, BGE_RDBDIMODE_ENABLE);
2994 
2995 	/* 57XX step 85 */
2996 	/* Turn on Mbuf cluster free state machine */
2997 	if (!BGE_IS_5705_PLUS(sc))
2998 		CSR_WRITE_4(sc, BGE_MBCF_MODE, BGE_MBCFMODE_ENABLE);
2999 
3000 	/* 5718 step 57, 57XX step 86 */
3001 	/* Turn on send data completion state machine */
3002 	val = BGE_SDCMODE_ENABLE;
3003 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5761)
3004 		val |= BGE_SDCMODE_CDELAY;
3005 	CSR_WRITE_4(sc, BGE_SDC_MODE, val);
3006 
3007 	/* 5718 step 58 */
3008 	/* Turn on send BD completion state machine */
3009 	CSR_WRITE_4(sc, BGE_SBDC_MODE, BGE_SBDCMODE_ENABLE);
3010 
3011 	/* 57XX step 88 */
3012 	/* Turn on RX BD initiator state machine */
3013 	CSR_WRITE_4(sc, BGE_RBDI_MODE, BGE_RBDIMODE_ENABLE);
3014 
3015 	/* 5718 step 60, 57XX step 90 */
3016 	/* Turn on send data initiator state machine */
3017 	if (sc->bge_flags & BGEF_TSO) {
3018 		/* XXX: magic value from Linux driver */
3019 		CSR_WRITE_4(sc, BGE_SDI_MODE, BGE_SDIMODE_ENABLE |
3020 		    BGE_SDIMODE_HW_LSO_PRE_DMA);
3021 	} else
3022 		CSR_WRITE_4(sc, BGE_SDI_MODE, BGE_SDIMODE_ENABLE);
3023 
3024 	/* 5718 step 61, 57XX step 91 */
3025 	/* Turn on send BD initiator state machine */
3026 	CSR_WRITE_4(sc, BGE_SBDI_MODE, BGE_SBDIMODE_ENABLE);
3027 
3028 	/* 5718 step 62, 57XX step 92 */
3029 	/* Turn on send BD selector state machine */
3030 	CSR_WRITE_4(sc, BGE_SRS_MODE, BGE_SRSMODE_ENABLE);
3031 
3032 	/* 5718 step 31, 57XX step 60 */
3033 	CSR_WRITE_4(sc, BGE_SDI_STATS_ENABLE_MASK, 0x007FFFFF);
3034 	/* 5718 step 32, 57XX step 61 */
3035 	CSR_WRITE_4(sc, BGE_SDI_STATS_CTL,
3036 	    BGE_SDISTATSCTL_ENABLE | BGE_SDISTATSCTL_FASTER);
3037 
3038 	/* ack/clear link change events */
3039 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED |
3040 	    BGE_MACSTAT_CFG_CHANGED | BGE_MACSTAT_MI_COMPLETE |
3041 	    BGE_MACSTAT_LINK_CHANGED);
3042 	CSR_WRITE_4(sc, BGE_MI_STS, 0);
3043 
3044 	/*
3045 	 * Enable attention when the link has changed state for
3046 	 * devices that use auto polling.
3047 	 */
3048 	if (sc->bge_flags & BGEF_FIBER_TBI) {
3049 		CSR_WRITE_4(sc, BGE_MI_STS, BGE_MISTS_LINK);
3050 	} else {
3051 		if ((sc->bge_flags & BGEF_CPMU_PRESENT) != 0)
3052 			mimode = BGE_MIMODE_500KHZ_CONST;
3053 		else
3054 			mimode = BGE_MIMODE_BASE;
3055 		/* 5718 step 68. 5718 step 69 (optionally). */
3056 		if (BGE_IS_5700_FAMILY(sc) ||
3057 		    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5705) {
3058 			mimode |= BGE_MIMODE_AUTOPOLL;
3059 			BGE_STS_SETBIT(sc, BGE_STS_AUTOPOLL);
3060 		}
3061 		mimode |= BGE_MIMODE_PHYADDR(sc->bge_phy_addr);
3062 		CSR_WRITE_4(sc, BGE_MI_MODE, mimode);
3063 		if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5700)
3064 			CSR_WRITE_4(sc, BGE_MAC_EVT_ENB,
3065 			    BGE_EVTENB_MI_INTERRUPT);
3066 	}
3067 
3068 	/*
3069 	 * Clear any pending link state attention.
3070 	 * Otherwise some link state change events may be lost until attention
3071 	 * is cleared by bge_intr() -> bge_link_upd() sequence.
3072 	 * It's not necessary on newer BCM chips - perhaps enabling link
3073 	 * state change attentions implies clearing pending attention.
3074 	 */
3075 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED |
3076 	    BGE_MACSTAT_CFG_CHANGED | BGE_MACSTAT_MI_COMPLETE |
3077 	    BGE_MACSTAT_LINK_CHANGED);
3078 
3079 	/* Enable link state change attentions. */
3080 	BGE_SETBIT(sc, BGE_MAC_EVT_ENB, BGE_EVTENB_LINK_CHANGED);
3081 
3082 	return 0;
3083 }
3084 
3085 static const struct bge_revision *
3086 bge_lookup_rev(uint32_t chipid)
3087 {
3088 	const struct bge_revision *br;
3089 
3090 	for (br = bge_revisions; br->br_name != NULL; br++) {
3091 		if (br->br_chipid == chipid)
3092 			return br;
3093 	}
3094 
3095 	for (br = bge_majorrevs; br->br_name != NULL; br++) {
3096 		if (br->br_chipid == BGE_ASICREV(chipid))
3097 			return br;
3098 	}
3099 
3100 	return NULL;
3101 }
3102 
3103 static const struct bge_product *
3104 bge_lookup(const struct pci_attach_args *pa)
3105 {
3106 	const struct bge_product *bp;
3107 
3108 	for (bp = bge_products; bp->bp_name != NULL; bp++) {
3109 		if (PCI_VENDOR(pa->pa_id) == bp->bp_vendor &&
3110 		    PCI_PRODUCT(pa->pa_id) == bp->bp_product)
3111 			return bp;
3112 	}
3113 
3114 	return NULL;
3115 }
3116 
3117 static uint32_t
3118 bge_chipid(const struct pci_attach_args *pa)
3119 {
3120 	uint32_t id;
3121 
3122 	id = pci_conf_read(pa->pa_pc, pa->pa_tag, BGE_PCI_MISC_CTL)
3123 		>> BGE_PCIMISCCTL_ASICREV_SHIFT;
3124 
3125 	if (BGE_ASICREV(id) == BGE_ASICREV_USE_PRODID_REG) {
3126 		switch (PCI_PRODUCT(pa->pa_id)) {
3127 		case PCI_PRODUCT_BROADCOM_BCM5717:
3128 		case PCI_PRODUCT_BROADCOM_BCM5718:
3129 		case PCI_PRODUCT_BROADCOM_BCM5719:
3130 		case PCI_PRODUCT_BROADCOM_BCM5720:
3131 		case PCI_PRODUCT_BROADCOM_BCM5725:
3132 		case PCI_PRODUCT_BROADCOM_BCM5727:
3133 		case PCI_PRODUCT_BROADCOM_BCM5762:
3134 		case PCI_PRODUCT_BROADCOM_BCM57764:
3135 		case PCI_PRODUCT_BROADCOM_BCM57767:
3136 		case PCI_PRODUCT_BROADCOM_BCM57787:
3137 			id = pci_conf_read(pa->pa_pc, pa->pa_tag,
3138 			    BGE_PCI_GEN2_PRODID_ASICREV);
3139 			break;
3140 		case PCI_PRODUCT_BROADCOM_BCM57761:
3141 		case PCI_PRODUCT_BROADCOM_BCM57762:
3142 		case PCI_PRODUCT_BROADCOM_BCM57765:
3143 		case PCI_PRODUCT_BROADCOM_BCM57766:
3144 		case PCI_PRODUCT_BROADCOM_BCM57781:
3145 		case PCI_PRODUCT_BROADCOM_BCM57782:
3146 		case PCI_PRODUCT_BROADCOM_BCM57785:
3147 		case PCI_PRODUCT_BROADCOM_BCM57786:
3148 		case PCI_PRODUCT_BROADCOM_BCM57791:
3149 		case PCI_PRODUCT_BROADCOM_BCM57795:
3150 			id = pci_conf_read(pa->pa_pc, pa->pa_tag,
3151 			    BGE_PCI_GEN15_PRODID_ASICREV);
3152 			break;
3153 		default:
3154 			id = pci_conf_read(pa->pa_pc, pa->pa_tag,
3155 			    BGE_PCI_PRODID_ASICREV);
3156 			break;
3157 		}
3158 	}
3159 
3160 	return id;
3161 }
3162 
3163 /*
3164  * Return true if MSI can be used with this device.
3165  */
3166 static int
3167 bge_can_use_msi(struct bge_softc *sc)
3168 {
3169 	int can_use_msi = 0;
3170 
3171 	switch (BGE_ASICREV(sc->bge_chipid)) {
3172 	case BGE_ASICREV_BCM5714_A0:
3173 	case BGE_ASICREV_BCM5714:
3174 		/*
3175 		 * Apparently, MSI doesn't work when these chips are
3176 		 * configured in single-port mode.
3177 		 */
3178 		break;
3179 	case BGE_ASICREV_BCM5750:
3180 		if (BGE_CHIPREV(sc->bge_chipid) != BGE_CHIPREV_5750_AX &&
3181 		    BGE_CHIPREV(sc->bge_chipid) != BGE_CHIPREV_5750_BX)
3182 			can_use_msi = 1;
3183 		break;
3184 	default:
3185 		if (BGE_IS_575X_PLUS(sc))
3186 			can_use_msi = 1;
3187 	}
3188 	return can_use_msi;
3189 }
3190 
3191 /*
3192  * Probe for a Broadcom chip. Check the PCI vendor and device IDs
3193  * against our list and return its name if we find a match. Note
3194  * that since the Broadcom controller contains VPD support, we
3195  * can get the device name string from the controller itself instead
3196  * of the compiled-in string. This is a little slow, but it guarantees
3197  * we'll always announce the right product name.
3198  */
3199 static int
3200 bge_probe(device_t parent, cfdata_t match, void *aux)
3201 {
3202 	struct pci_attach_args *pa = (struct pci_attach_args *)aux;
3203 
3204 	if (bge_lookup(pa) != NULL)
3205 		return 1;
3206 
3207 	return 0;
3208 }
3209 
3210 static void
3211 bge_attach(device_t parent, device_t self, void *aux)
3212 {
3213 	struct bge_softc * const sc = device_private(self);
3214 	struct pci_attach_args * const pa = aux;
3215 	prop_dictionary_t dict;
3216 	const struct bge_product *bp;
3217 	const struct bge_revision *br;
3218 	pci_chipset_tag_t	pc;
3219 	const char		*intrstr = NULL;
3220 	uint32_t		hwcfg, hwcfg2, hwcfg3, hwcfg4, hwcfg5;
3221 	uint32_t		command;
3222 	struct ifnet		*ifp;
3223 	struct mii_data * const mii = &sc->bge_mii;
3224 	uint32_t		misccfg, mimode, macmode;
3225 	void *			kva;
3226 	u_char			eaddr[ETHER_ADDR_LEN];
3227 	pcireg_t		memtype, subid, reg;
3228 	bus_addr_t		memaddr;
3229 	uint32_t		pm_ctl;
3230 	bool			no_seeprom;
3231 	int			capmask, trys;
3232 	int			mii_flags;
3233 	int			map_flags;
3234 	char intrbuf[PCI_INTRSTR_LEN];
3235 
3236 	bp = bge_lookup(pa);
3237 	KASSERT(bp != NULL);
3238 
3239 	sc->sc_pc = pa->pa_pc;
3240 	sc->sc_pcitag = pa->pa_tag;
3241 	sc->bge_dev = self;
3242 
3243 	sc->bge_pa = *pa;
3244 	pc = sc->sc_pc;
3245 	subid = pci_conf_read(pc, sc->sc_pcitag, PCI_SUBSYS_ID_REG);
3246 
3247 	aprint_naive(": Ethernet controller\n");
3248 	aprint_normal(": %s Ethernet\n", bp->bp_name);
3249 
3250 	/*
3251 	 * Map control/status registers.
3252 	 */
3253 	DPRINTFN(5, ("Map control/status regs\n"));
3254 	command = pci_conf_read(pc, sc->sc_pcitag, PCI_COMMAND_STATUS_REG);
3255 	command |= PCI_COMMAND_MEM_ENABLE | PCI_COMMAND_MASTER_ENABLE;
3256 	pci_conf_write(pc, sc->sc_pcitag, PCI_COMMAND_STATUS_REG, command);
3257 	command = pci_conf_read(pc, sc->sc_pcitag, PCI_COMMAND_STATUS_REG);
3258 
3259 	if (!(command & PCI_COMMAND_MEM_ENABLE)) {
3260 		aprint_error_dev(sc->bge_dev,
3261 		    "failed to enable memory mapping!\n");
3262 		return;
3263 	}
3264 
3265 	DPRINTFN(5, ("pci_mem_find\n"));
3266 	memtype = pci_mapreg_type(sc->sc_pc, sc->sc_pcitag, BGE_PCI_BAR0);
3267 	switch (memtype) {
3268 	case PCI_MAPREG_TYPE_MEM | PCI_MAPREG_MEM_TYPE_32BIT:
3269 	case PCI_MAPREG_TYPE_MEM | PCI_MAPREG_MEM_TYPE_64BIT:
3270 #if 0
3271 		if (pci_mapreg_map(pa, BGE_PCI_BAR0,
3272 		    memtype, 0, &sc->bge_btag, &sc->bge_bhandle,
3273 		    &memaddr, &sc->bge_bsize) == 0)
3274 			break;
3275 #else
3276 		/*
3277 		 * Workaround for PCI prefetchable bit. Some BCM5717-5720 based
3278 		 * system get NMI on boot (PR#48451). This problem might not be
3279 		 * the driver's bug but our PCI common part's bug. Until we
3280 		 * find a real reason, we ignore the prefetchable bit.
3281 		 */
3282 		if (pci_mapreg_info(pa->pa_pc, pa->pa_tag, BGE_PCI_BAR0,
3283 		    memtype, &memaddr, &sc->bge_bsize, &map_flags) == 0) {
3284 			map_flags &= ~BUS_SPACE_MAP_PREFETCHABLE;
3285 			if (bus_space_map(pa->pa_memt, memaddr, sc->bge_bsize,
3286 			    map_flags, &sc->bge_bhandle) == 0) {
3287 				sc->bge_btag = pa->pa_memt;
3288 				break;
3289 			}
3290 		}
3291 #endif
3292 		/* FALLTHROUGH */
3293 	default:
3294 		aprint_error_dev(sc->bge_dev, "can't find mem space\n");
3295 		return;
3296 	}
3297 
3298 	sc->bge_stopping = false;
3299 	sc->bge_txrx_stopping = false;
3300 
3301 	/* Save various chip information. */
3302 	sc->bge_chipid = bge_chipid(pa);
3303 	sc->bge_phy_addr = bge_phy_addr(sc);
3304 
3305 	if (pci_get_capability(sc->sc_pc, sc->sc_pcitag, PCI_CAP_PCIEXPRESS,
3306 	    &sc->bge_pciecap, NULL) != 0) {
3307 		/* PCIe */
3308 		sc->bge_flags |= BGEF_PCIE;
3309 		/* Extract supported maximum payload size. */
3310 		reg = pci_conf_read(sc->sc_pc, sc->sc_pcitag,
3311 		    sc->bge_pciecap + PCIE_DCAP);
3312 		sc->bge_mps = 128 << (reg & PCIE_DCAP_MAX_PAYLOAD);
3313 		if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5719 ||
3314 		    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5720)
3315 			sc->bge_expmrq = 2048;
3316 		else
3317 			sc->bge_expmrq = 4096;
3318 		bge_set_max_readrq(sc);
3319 	} else if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5785) {
3320 		/* PCIe without PCIe cap */
3321 		sc->bge_flags |= BGEF_PCIE;
3322 	} else if ((pci_conf_read(sc->sc_pc, sc->sc_pcitag, BGE_PCI_PCISTATE) &
3323 		BGE_PCISTATE_PCI_BUSMODE) == 0) {
3324 		/* PCI-X */
3325 		sc->bge_flags |= BGEF_PCIX;
3326 		if (pci_get_capability(pa->pa_pc, pa->pa_tag, PCI_CAP_PCIX,
3327 			&sc->bge_pcixcap, NULL) == 0)
3328 			aprint_error_dev(sc->bge_dev,
3329 			    "unable to find PCIX capability\n");
3330 	}
3331 
3332 	if (BGE_CHIPREV(sc->bge_chipid) == BGE_CHIPREV_5700_BX) {
3333 		/*
3334 		 * Kludge for 5700 Bx bug: a hardware bug (PCIX byte enable?)
3335 		 * can clobber the chip's PCI config-space power control
3336 		 * registers, leaving the card in D3 powersave state. We do
3337 		 * not have memory-mapped registers in this state, so force
3338 		 * device into D0 state before starting initialization.
3339 		 */
3340 		pm_ctl = pci_conf_read(pc, sc->sc_pcitag, BGE_PCI_PWRMGMT_CMD);
3341 		pm_ctl &= ~(PCI_PWR_D0 | PCI_PWR_D1 | PCI_PWR_D2 | PCI_PWR_D3);
3342 		pm_ctl |= (1 << 8) | PCI_PWR_D0 ; /* D0 state */
3343 		pci_conf_write(pc, sc->sc_pcitag, BGE_PCI_PWRMGMT_CMD, pm_ctl);
3344 		DELAY(1000);	/* 27 usec is allegedly sufficient */
3345 	}
3346 
3347 	/* Save chipset family. */
3348 	switch (BGE_ASICREV(sc->bge_chipid)) {
3349 	case BGE_ASICREV_BCM5717:
3350 	case BGE_ASICREV_BCM5719:
3351 	case BGE_ASICREV_BCM5720:
3352 		sc->bge_flags |= BGEF_5717_PLUS;
3353 		/* FALLTHROUGH */
3354 	case BGE_ASICREV_BCM5762:
3355 	case BGE_ASICREV_BCM57765:
3356 	case BGE_ASICREV_BCM57766:
3357 		if (!BGE_IS_5717_PLUS(sc))
3358 			sc->bge_flags |= BGEF_57765_FAMILY;
3359 		sc->bge_flags |= BGEF_57765_PLUS | BGEF_5755_PLUS |
3360 		    BGEF_575X_PLUS | BGEF_5705_PLUS | BGEF_JUMBO_CAPABLE;
3361 		if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5719 ||
3362 		    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5720) {
3363 			/*
3364 			 * Enable work around for DMA engine miscalculation
3365 			 * of TXMBUF available space.
3366 			 */
3367 			sc->bge_flags |= BGEF_RDMA_BUG;
3368 
3369 			if ((BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5719) &&
3370 			    (sc->bge_chipid == BGE_CHIPID_BCM5719_A0)) {
3371 				/* Jumbo frame on BCM5719 A0 does not work. */
3372 				sc->bge_flags &= ~BGEF_JUMBO_CAPABLE;
3373 			}
3374 		}
3375 		break;
3376 	case BGE_ASICREV_BCM5755:
3377 	case BGE_ASICREV_BCM5761:
3378 	case BGE_ASICREV_BCM5784:
3379 	case BGE_ASICREV_BCM5785:
3380 	case BGE_ASICREV_BCM5787:
3381 	case BGE_ASICREV_BCM57780:
3382 		sc->bge_flags |= BGEF_5755_PLUS | BGEF_575X_PLUS | BGEF_5705_PLUS;
3383 		break;
3384 	case BGE_ASICREV_BCM5700:
3385 	case BGE_ASICREV_BCM5701:
3386 	case BGE_ASICREV_BCM5703:
3387 	case BGE_ASICREV_BCM5704:
3388 		sc->bge_flags |= BGEF_5700_FAMILY | BGEF_JUMBO_CAPABLE;
3389 		break;
3390 	case BGE_ASICREV_BCM5714_A0:
3391 	case BGE_ASICREV_BCM5780:
3392 	case BGE_ASICREV_BCM5714:
3393 		sc->bge_flags |= BGEF_5714_FAMILY | BGEF_JUMBO_CAPABLE;
3394 		/* FALLTHROUGH */
3395 	case BGE_ASICREV_BCM5750:
3396 	case BGE_ASICREV_BCM5752:
3397 	case BGE_ASICREV_BCM5906:
3398 		sc->bge_flags |= BGEF_575X_PLUS;
3399 		/* FALLTHROUGH */
3400 	case BGE_ASICREV_BCM5705:
3401 		sc->bge_flags |= BGEF_5705_PLUS;
3402 		break;
3403 	}
3404 
3405 	/* Identify chips with APE processor. */
3406 	switch (BGE_ASICREV(sc->bge_chipid)) {
3407 	case BGE_ASICREV_BCM5717:
3408 	case BGE_ASICREV_BCM5719:
3409 	case BGE_ASICREV_BCM5720:
3410 	case BGE_ASICREV_BCM5761:
3411 	case BGE_ASICREV_BCM5762:
3412 		sc->bge_flags |= BGEF_APE;
3413 		break;
3414 	}
3415 
3416 	/*
3417 	 * The 40bit DMA bug applies to the 5714/5715 controllers and is
3418 	 * not actually a MAC controller bug but an issue with the embedded
3419 	 * PCIe to PCI-X bridge in the device. Use 40bit DMA workaround.
3420 	 */
3421 	if (BGE_IS_5714_FAMILY(sc) && ((sc->bge_flags & BGEF_PCIX) != 0))
3422 		sc->bge_flags |= BGEF_40BIT_BUG;
3423 
3424 	/* Chips with APE need BAR2 access for APE registers/memory. */
3425 	if ((sc->bge_flags & BGEF_APE) != 0) {
3426 		memtype = pci_mapreg_type(pa->pa_pc, pa->pa_tag, BGE_PCI_BAR2);
3427 #if 0
3428 		if (pci_mapreg_map(pa, BGE_PCI_BAR2, memtype, 0,
3429 			&sc->bge_apetag, &sc->bge_apehandle, NULL,
3430 			&sc->bge_apesize)) {
3431 			aprint_error_dev(sc->bge_dev,
3432 			    "couldn't map BAR2 memory\n");
3433 			return;
3434 		}
3435 #else
3436 		/*
3437 		 * Workaround for PCI prefetchable bit. Some BCM5717-5720 based
3438 		 * system get NMI on boot (PR#48451). This problem might not be
3439 		 * the driver's bug but our PCI common part's bug. Until we
3440 		 * find a real reason, we ignore the prefetchable bit.
3441 		 */
3442 		if (pci_mapreg_info(pa->pa_pc, pa->pa_tag, BGE_PCI_BAR2,
3443 		    memtype, &memaddr, &sc->bge_apesize, &map_flags) != 0) {
3444 			aprint_error_dev(sc->bge_dev,
3445 			    "couldn't map BAR2 memory\n");
3446 			return;
3447 		}
3448 
3449 		map_flags &= ~BUS_SPACE_MAP_PREFETCHABLE;
3450 		if (bus_space_map(pa->pa_memt, memaddr,
3451 		    sc->bge_apesize, map_flags, &sc->bge_apehandle) != 0) {
3452 			aprint_error_dev(sc->bge_dev,
3453 			    "couldn't map BAR2 memory\n");
3454 			return;
3455 		}
3456 		sc->bge_apetag = pa->pa_memt;
3457 #endif
3458 
3459 		/* Enable APE register/memory access by host driver. */
3460 		reg = pci_conf_read(pa->pa_pc, pa->pa_tag, BGE_PCI_PCISTATE);
3461 		reg |= BGE_PCISTATE_ALLOW_APE_CTLSPC_WR |
3462 		    BGE_PCISTATE_ALLOW_APE_SHMEM_WR |
3463 		    BGE_PCISTATE_ALLOW_APE_PSPACE_WR;
3464 		pci_conf_write(pa->pa_pc, pa->pa_tag, BGE_PCI_PCISTATE, reg);
3465 
3466 		bge_ape_lock_init(sc);
3467 		bge_ape_read_fw_ver(sc);
3468 	}
3469 
3470 	/* Identify the chips that use an CPMU. */
3471 	if (BGE_IS_5717_PLUS(sc) ||
3472 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5784 ||
3473 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5761 ||
3474 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5785 ||
3475 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM57780)
3476 		sc->bge_flags |= BGEF_CPMU_PRESENT;
3477 
3478 	/*
3479 	 * When using the BCM5701 in PCI-X mode, data corruption has
3480 	 * been observed in the first few bytes of some received packets.
3481 	 * Aligning the packet buffer in memory eliminates the corruption.
3482 	 * Unfortunately, this misaligns the packet payloads.  On platforms
3483 	 * which do not support unaligned accesses, we will realign the
3484 	 * payloads by copying the received packets.
3485 	 */
3486 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5701 &&
3487 	    sc->bge_flags & BGEF_PCIX)
3488 		sc->bge_flags |= BGEF_RX_ALIGNBUG;
3489 
3490 	if (BGE_IS_5700_FAMILY(sc))
3491 		sc->bge_flags |= BGEF_JUMBO_CAPABLE;
3492 
3493 	misccfg = CSR_READ_4(sc, BGE_MISC_CFG);
3494 	misccfg &= BGE_MISCCFG_BOARD_ID_MASK;
3495 
3496 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5705 &&
3497 	    (misccfg == BGE_MISCCFG_BOARD_ID_5788 ||
3498 	     misccfg == BGE_MISCCFG_BOARD_ID_5788M))
3499 		sc->bge_flags |= BGEF_IS_5788;
3500 
3501 	/*
3502 	 * Some controllers seem to require a special firmware to use
3503 	 * TSO. But the firmware is not available to FreeBSD and Linux
3504 	 * claims that the TSO performed by the firmware is slower than
3505 	 * hardware based TSO. Moreover the firmware based TSO has one
3506 	 * known bug which can't handle TSO if ethernet header + IP/TCP
3507 	 * header is greater than 80 bytes. The workaround for the TSO
3508 	 * bug exist but it seems it's too expensive than not using
3509 	 * TSO at all. Some hardware also have the TSO bug so limit
3510 	 * the TSO to the controllers that are not affected TSO issues
3511 	 * (e.g. 5755 or higher).
3512 	 */
3513 	if (BGE_IS_5755_PLUS(sc)) {
3514 		/*
3515 		 * BCM5754 and BCM5787 shares the same ASIC id so
3516 		 * explicit device id check is required.
3517 		 */
3518 		if ((PCI_PRODUCT(pa->pa_id) != PCI_PRODUCT_BROADCOM_BCM5754) &&
3519 		    (PCI_PRODUCT(pa->pa_id) != PCI_PRODUCT_BROADCOM_BCM5754M))
3520 			sc->bge_flags |= BGEF_TSO;
3521 		/* TSO on BCM5719 A0 does not work. */
3522 		if ((BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5719) &&
3523 		    (sc->bge_chipid == BGE_CHIPID_BCM5719_A0))
3524 			sc->bge_flags &= ~BGEF_TSO;
3525 	}
3526 
3527 	capmask = 0xffffffff; /* XXX BMSR_DEFCAPMASK */
3528 	if ((BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5703 &&
3529 	     (misccfg == 0x4000 || misccfg == 0x8000)) ||
3530 	    (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5705 &&
3531 	     PCI_VENDOR(pa->pa_id) == PCI_VENDOR_BROADCOM &&
3532 	     (PCI_PRODUCT(pa->pa_id) == PCI_PRODUCT_BROADCOM_BCM5901 ||
3533 	      PCI_PRODUCT(pa->pa_id) == PCI_PRODUCT_BROADCOM_BCM5901A2 ||
3534 	      PCI_PRODUCT(pa->pa_id) == PCI_PRODUCT_BROADCOM_BCM5705F)) ||
3535 	    (PCI_VENDOR(pa->pa_id) == PCI_VENDOR_BROADCOM &&
3536 	     (PCI_PRODUCT(pa->pa_id) == PCI_PRODUCT_BROADCOM_BCM5751F ||
3537 	      PCI_PRODUCT(pa->pa_id) == PCI_PRODUCT_BROADCOM_BCM5753F ||
3538 	      PCI_PRODUCT(pa->pa_id) == PCI_PRODUCT_BROADCOM_BCM5787F)) ||
3539 	    PCI_PRODUCT(pa->pa_id) == PCI_PRODUCT_BROADCOM_BCM57790 ||
3540 	    PCI_PRODUCT(pa->pa_id) == PCI_PRODUCT_BROADCOM_BCM57791 ||
3541 	    PCI_PRODUCT(pa->pa_id) == PCI_PRODUCT_BROADCOM_BCM57795 ||
3542 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906) {
3543 		/* These chips are 10/100 only. */
3544 		capmask &= ~BMSR_EXTSTAT;
3545 		sc->bge_phy_flags |= BGEPHYF_NO_WIRESPEED;
3546 	}
3547 
3548 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5700 ||
3549 	    (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5705 &&
3550 	     (sc->bge_chipid != BGE_CHIPID_BCM5705_A0 &&
3551 		 sc->bge_chipid != BGE_CHIPID_BCM5705_A1)))
3552 		sc->bge_phy_flags |= BGEPHYF_NO_WIRESPEED;
3553 
3554 	/* Set various PHY bug flags. */
3555 	if (sc->bge_chipid == BGE_CHIPID_BCM5701_A0 ||
3556 	    sc->bge_chipid == BGE_CHIPID_BCM5701_B0)
3557 		sc->bge_phy_flags |= BGEPHYF_CRC_BUG;
3558 	if (BGE_CHIPREV(sc->bge_chipid) == BGE_CHIPREV_5703_AX ||
3559 	    BGE_CHIPREV(sc->bge_chipid) == BGE_CHIPREV_5704_AX)
3560 		sc->bge_phy_flags |= BGEPHYF_ADC_BUG;
3561 	if (sc->bge_chipid == BGE_CHIPID_BCM5704_A0)
3562 		sc->bge_phy_flags |= BGEPHYF_5704_A0_BUG;
3563 	if ((BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5700 ||
3564 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5701) &&
3565 	    PCI_VENDOR(subid) == PCI_VENDOR_DELL)
3566 		sc->bge_phy_flags |= BGEPHYF_NO_3LED;
3567 	if (BGE_IS_5705_PLUS(sc) &&
3568 	    BGE_ASICREV(sc->bge_chipid) != BGE_ASICREV_BCM5906 &&
3569 	    BGE_ASICREV(sc->bge_chipid) != BGE_ASICREV_BCM5785 &&
3570 	    BGE_ASICREV(sc->bge_chipid) != BGE_ASICREV_BCM57780 &&
3571 	    !BGE_IS_57765_PLUS(sc)) {
3572 		if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5755 ||
3573 		    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5761 ||
3574 		    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5784 ||
3575 		    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5787) {
3576 			if (PCI_PRODUCT(pa->pa_id) != PCI_PRODUCT_BROADCOM_BCM5722 &&
3577 			    PCI_PRODUCT(pa->pa_id) != PCI_PRODUCT_BROADCOM_BCM5756)
3578 				sc->bge_phy_flags |= BGEPHYF_JITTER_BUG;
3579 			if (PCI_PRODUCT(pa->pa_id) == PCI_PRODUCT_BROADCOM_BCM5755M)
3580 				sc->bge_phy_flags |= BGEPHYF_ADJUST_TRIM;
3581 		} else
3582 			sc->bge_phy_flags |= BGEPHYF_BER_BUG;
3583 	}
3584 
3585 	/*
3586 	 * SEEPROM check.
3587 	 * First check if firmware knows we do not have SEEPROM.
3588 	 */
3589 	if (prop_dictionary_get_bool(device_properties(self),
3590 	    "without-seeprom", &no_seeprom) && no_seeprom)
3591 		sc->bge_flags |= BGEF_NO_EEPROM;
3592 
3593 	else if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906)
3594 		sc->bge_flags |= BGEF_NO_EEPROM;
3595 
3596 	/* Now check the 'ROM failed' bit on the RX CPU */
3597 	else if (CSR_READ_4(sc, BGE_RXCPU_MODE) & BGE_RXCPUMODE_ROMFAIL)
3598 		sc->bge_flags |= BGEF_NO_EEPROM;
3599 
3600 	sc->bge_asf_mode = 0;
3601 	/* No ASF if APE present. */
3602 	if ((sc->bge_flags & BGEF_APE) == 0) {
3603 		if (bge_allow_asf && (bge_readmem_ind(sc, BGE_SRAM_DATA_SIG) ==
3604 			BGE_SRAM_DATA_SIG_MAGIC)) {
3605 			if (bge_readmem_ind(sc, BGE_SRAM_DATA_CFG) &
3606 			    BGE_HWCFG_ASF) {
3607 				sc->bge_asf_mode |= ASF_ENABLE;
3608 				sc->bge_asf_mode |= ASF_STACKUP;
3609 				if (BGE_IS_575X_PLUS(sc))
3610 					sc->bge_asf_mode |= ASF_NEW_HANDSHAKE;
3611 			}
3612 		}
3613 	}
3614 
3615 	int counts[PCI_INTR_TYPE_SIZE] = {
3616 		[PCI_INTR_TYPE_INTX] = 1,
3617 		[PCI_INTR_TYPE_MSI] = 1,
3618 		[PCI_INTR_TYPE_MSIX] = 1,
3619 	};
3620 	int max_type = PCI_INTR_TYPE_MSIX;
3621 
3622 	if (!bge_can_use_msi(sc)) {
3623 		/* MSI broken, allow only INTx */
3624 		max_type = PCI_INTR_TYPE_INTX;
3625 	}
3626 
3627 	if (pci_intr_alloc(pa, &sc->bge_pihp, counts, max_type) != 0) {
3628 		aprint_error_dev(sc->bge_dev, "couldn't alloc interrupt\n");
3629 		return;
3630 	}
3631 
3632 	DPRINTFN(5, ("pci_intr_string\n"));
3633 	intrstr = pci_intr_string(pc, sc->bge_pihp[0], intrbuf,
3634 	    sizeof(intrbuf));
3635 	DPRINTFN(5, ("pci_intr_establish\n"));
3636 	sc->bge_intrhand = pci_intr_establish_xname(pc, sc->bge_pihp[0],
3637 	    IPL_NET, bge_intr, sc, device_xname(sc->bge_dev));
3638 	if (sc->bge_intrhand == NULL) {
3639 		pci_intr_release(pc, sc->bge_pihp, 1);
3640 		sc->bge_pihp = NULL;
3641 
3642 		aprint_error_dev(self, "couldn't establish interrupt");
3643 		if (intrstr != NULL)
3644 			aprint_error(" at %s", intrstr);
3645 		aprint_error("\n");
3646 		return;
3647 	}
3648 	aprint_normal_dev(sc->bge_dev, "interrupting at %s\n", intrstr);
3649 
3650 	switch (pci_intr_type(pc, sc->bge_pihp[0])) {
3651 	case PCI_INTR_TYPE_MSIX:
3652 	case PCI_INTR_TYPE_MSI:
3653 		KASSERT(bge_can_use_msi(sc));
3654 		sc->bge_flags |= BGEF_MSI;
3655 		break;
3656 	default:
3657 		/* nothing to do */
3658 		break;
3659 	}
3660 
3661 	char wqname[MAXCOMLEN];
3662 	snprintf(wqname, sizeof(wqname), "%sReset", device_xname(sc->bge_dev));
3663 	int error = workqueue_create(&sc->sc_reset_wq, wqname,
3664 	    bge_handle_reset_work, sc, PRI_NONE, IPL_SOFTCLOCK,
3665 	    WQ_MPSAFE);
3666 	if (error) {
3667 		aprint_error_dev(sc->bge_dev,
3668 		    "unable to create reset workqueue\n");
3669 		return;
3670 	}
3671 
3672 
3673 	/*
3674 	 * All controllers except BCM5700 supports tagged status but
3675 	 * we use tagged status only for MSI case on BCM5717. Otherwise
3676 	 * MSI on BCM5717 does not work.
3677 	 */
3678 	if (BGE_IS_57765_PLUS(sc) && sc->bge_flags & BGEF_MSI)
3679 		sc->bge_flags |= BGEF_TAGGED_STATUS;
3680 
3681 	/*
3682 	 * Reset NVRAM before bge_reset(). It's required to acquire NVRAM
3683 	 * lock in bge_reset().
3684 	 */
3685 	CSR_WRITE_4_FLUSH(sc, BGE_EE_ADDR,
3686 	    BGE_EEADDR_RESET | BGE_EEHALFCLK(BGE_HALFCLK_384SCL));
3687 	delay(1000);
3688 	BGE_SETBIT_FLUSH(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_AUTO_EEPROM);
3689 
3690 	bge_stop_fw(sc);
3691 	bge_sig_pre_reset(sc, BGE_RESET_SHUTDOWN);
3692 	if (bge_reset(sc))
3693 		aprint_error_dev(sc->bge_dev, "chip reset failed\n");
3694 
3695 	/*
3696 	 * Read the hardware config word in the first 32k of NIC internal
3697 	 * memory, or fall back to the config word in the EEPROM.
3698 	 * Note: on some BCM5700 cards, this value appears to be unset.
3699 	 */
3700 	hwcfg = hwcfg2 = hwcfg3 = hwcfg4 = hwcfg5 = 0;
3701 	if (bge_readmem_ind(sc, BGE_SRAM_DATA_SIG) ==
3702 	    BGE_SRAM_DATA_SIG_MAGIC) {
3703 		uint32_t tmp;
3704 
3705 		hwcfg = bge_readmem_ind(sc, BGE_SRAM_DATA_CFG);
3706 		tmp = bge_readmem_ind(sc, BGE_SRAM_DATA_VER) >>
3707 		    BGE_SRAM_DATA_VER_SHIFT;
3708 		if ((0 < tmp) && (tmp < 0x100))
3709 			hwcfg2 = bge_readmem_ind(sc, BGE_SRAM_DATA_CFG_2);
3710 		if (sc->bge_flags & BGEF_PCIE)
3711 			hwcfg3 = bge_readmem_ind(sc, BGE_SRAM_DATA_CFG_3);
3712 		if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5785)
3713 			hwcfg4 = bge_readmem_ind(sc, BGE_SRAM_DATA_CFG_4);
3714 		if (BGE_IS_5717_PLUS(sc))
3715 			hwcfg5 = bge_readmem_ind(sc, BGE_SRAM_DATA_CFG_5);
3716 	} else if (!(sc->bge_flags & BGEF_NO_EEPROM)) {
3717 		bge_read_eeprom(sc, (void *)&hwcfg,
3718 		    BGE_EE_HWCFG_OFFSET, sizeof(hwcfg));
3719 		hwcfg = be32toh(hwcfg);
3720 	}
3721 	aprint_normal_dev(sc->bge_dev,
3722 	    "HW config %08x, %08x, %08x, %08x %08x\n",
3723 	    hwcfg, hwcfg2, hwcfg3, hwcfg4, hwcfg5);
3724 
3725 	bge_sig_legacy(sc, BGE_RESET_SHUTDOWN);
3726 	bge_sig_post_reset(sc, BGE_RESET_SHUTDOWN);
3727 
3728 	if (bge_chipinit(sc)) {
3729 		aprint_error_dev(sc->bge_dev, "chip initialization failed\n");
3730 		bge_release_resources(sc);
3731 		return;
3732 	}
3733 
3734 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5700) {
3735 		BGE_SETBIT_FLUSH(sc, BGE_MISC_LOCAL_CTL,
3736 		    BGE_MLC_MISCIO_OUT1 | BGE_MLC_MISCIO_OUTEN1);
3737 		DELAY(100);
3738 	}
3739 
3740 	/* Set MI_MODE */
3741 	mimode = BGE_MIMODE_PHYADDR(sc->bge_phy_addr);
3742 	if ((sc->bge_flags & BGEF_CPMU_PRESENT) != 0)
3743 		mimode |= BGE_MIMODE_500KHZ_CONST;
3744 	else
3745 		mimode |= BGE_MIMODE_BASE;
3746 	CSR_WRITE_4_FLUSH(sc, BGE_MI_MODE, mimode);
3747 	DELAY(80);
3748 
3749 	/*
3750 	 * Get station address from the EEPROM.
3751 	 */
3752 	if (bge_get_eaddr(sc, eaddr)) {
3753 		aprint_error_dev(sc->bge_dev,
3754 		    "failed to read station address\n");
3755 		bge_release_resources(sc);
3756 		return;
3757 	}
3758 
3759 	br = bge_lookup_rev(sc->bge_chipid);
3760 
3761 	if (br == NULL) {
3762 		aprint_normal_dev(sc->bge_dev, "unknown ASIC (0x%x)",
3763 		    sc->bge_chipid);
3764 	} else {
3765 		aprint_normal_dev(sc->bge_dev, "ASIC %s (0x%x)",
3766 		    br->br_name, sc->bge_chipid);
3767 	}
3768 	aprint_normal(", Ethernet address %s\n", ether_sprintf(eaddr));
3769 
3770 	/* Allocate the general information block and ring buffers. */
3771 	if (pci_dma64_available(pa)) {
3772 		sc->bge_dmatag = pa->pa_dmat64;
3773 		sc->bge_dmatag32 = pa->pa_dmat;
3774 		sc->bge_dma64 = true;
3775 	} else {
3776 		sc->bge_dmatag = pa->pa_dmat;
3777 		sc->bge_dmatag32 = pa->pa_dmat;
3778 		sc->bge_dma64 = false;
3779 	}
3780 
3781 	/* 40bit DMA workaround */
3782 	if (sizeof(bus_addr_t) > 4) {
3783 		if ((sc->bge_flags & BGEF_40BIT_BUG) != 0) {
3784 			bus_dma_tag_t olddmatag = sc->bge_dmatag; /* save */
3785 
3786 			if (bus_dmatag_subregion(olddmatag, 0,
3787 			    (bus_addr_t)__MASK(40),
3788 			    &(sc->bge_dmatag), BUS_DMA_WAITOK) != 0) {
3789 				aprint_error_dev(self,
3790 				    "WARNING: failed to restrict dma range,"
3791 				    " falling back to parent bus dma range\n");
3792 				sc->bge_dmatag = olddmatag;
3793 			}
3794 		}
3795 	}
3796 	SLIST_INIT(&sc->txdma_list);
3797 	DPRINTFN(5, ("bus_dmamem_alloc\n"));
3798 	if (bus_dmamem_alloc(sc->bge_dmatag, sizeof(struct bge_ring_data),
3799 			     PAGE_SIZE, 0, &sc->bge_ring_seg, 1,
3800 		&sc->bge_ring_rseg, BUS_DMA_WAITOK)) {
3801 		aprint_error_dev(sc->bge_dev, "can't alloc rx buffers\n");
3802 		return;
3803 	}
3804 	DPRINTFN(5, ("bus_dmamem_map\n"));
3805 	if (bus_dmamem_map(sc->bge_dmatag, &sc->bge_ring_seg,
3806 		sc->bge_ring_rseg, sizeof(struct bge_ring_data), &kva,
3807 			   BUS_DMA_WAITOK)) {
3808 		aprint_error_dev(sc->bge_dev,
3809 		    "can't map DMA buffers (%zu bytes)\n",
3810 		    sizeof(struct bge_ring_data));
3811 		bus_dmamem_free(sc->bge_dmatag, &sc->bge_ring_seg,
3812 		    sc->bge_ring_rseg);
3813 		return;
3814 	}
3815 	DPRINTFN(5, ("bus_dmamap_create\n"));
3816 	if (bus_dmamap_create(sc->bge_dmatag, sizeof(struct bge_ring_data), 1,
3817 	    sizeof(struct bge_ring_data), 0,
3818 	    BUS_DMA_WAITOK, &sc->bge_ring_map)) {
3819 		aprint_error_dev(sc->bge_dev, "can't create DMA map\n");
3820 		bus_dmamem_unmap(sc->bge_dmatag, kva,
3821 				 sizeof(struct bge_ring_data));
3822 		bus_dmamem_free(sc->bge_dmatag, &sc->bge_ring_seg,
3823 		    sc->bge_ring_rseg);
3824 		return;
3825 	}
3826 	DPRINTFN(5, ("bus_dmamap_load\n"));
3827 	if (bus_dmamap_load(sc->bge_dmatag, sc->bge_ring_map, kva,
3828 			    sizeof(struct bge_ring_data), NULL,
3829 			    BUS_DMA_WAITOK)) {
3830 		bus_dmamap_destroy(sc->bge_dmatag, sc->bge_ring_map);
3831 		bus_dmamem_unmap(sc->bge_dmatag, kva,
3832 				 sizeof(struct bge_ring_data));
3833 		bus_dmamem_free(sc->bge_dmatag, &sc->bge_ring_seg,
3834 		    sc->bge_ring_rseg);
3835 		return;
3836 	}
3837 
3838 	DPRINTFN(5, ("bzero\n"));
3839 	sc->bge_rdata = (struct bge_ring_data *)kva;
3840 
3841 	memset(sc->bge_rdata, 0, sizeof(struct bge_ring_data));
3842 
3843 	/* Try to allocate memory for jumbo buffers. */
3844 	if (BGE_IS_JUMBO_CAPABLE(sc)) {
3845 		if (bge_alloc_jumbo_mem(sc)) {
3846 			aprint_error_dev(sc->bge_dev,
3847 			    "jumbo buffer allocation failed\n");
3848 		} else
3849 			sc->ethercom.ec_capabilities |= ETHERCAP_JUMBO_MTU;
3850 	}
3851 
3852 	/* Set default tuneable values. */
3853 	sc->bge_stat_ticks = BGE_TICKS_PER_SEC;
3854 	sc->bge_rx_coal_ticks = 150;
3855 	sc->bge_rx_max_coal_bds = 64;
3856 	sc->bge_tx_coal_ticks = 300;
3857 	sc->bge_tx_max_coal_bds = 400;
3858 	if (BGE_IS_5705_PLUS(sc)) {
3859 		sc->bge_tx_coal_ticks = (12 * 5);
3860 		sc->bge_tx_max_coal_bds = (12 * 5);
3861 			aprint_verbose_dev(sc->bge_dev,
3862 			    "setting short Tx thresholds\n");
3863 	}
3864 
3865 	if (BGE_IS_5717_PLUS(sc))
3866 		sc->bge_return_ring_cnt = BGE_RETURN_RING_CNT;
3867 	else if (BGE_IS_5705_PLUS(sc))
3868 		sc->bge_return_ring_cnt = BGE_RETURN_RING_CNT_5705;
3869 	else
3870 		sc->bge_return_ring_cnt = BGE_RETURN_RING_CNT;
3871 
3872 	sc->sc_core_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
3873 	sc->sc_intr_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NET);
3874 
3875 	/* Set up ifnet structure */
3876 	ifp = &sc->ethercom.ec_if;
3877 	ifp->if_softc = sc;
3878 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
3879 	ifp->if_extflags = IFEF_MPSAFE;
3880 	ifp->if_ioctl = bge_ioctl;
3881 	ifp->if_stop = bge_stop;
3882 	ifp->if_start = bge_start;
3883 	ifp->if_init = bge_init;
3884 	IFQ_SET_MAXLEN(&ifp->if_snd, uimax(BGE_TX_RING_CNT - 1, IFQ_MAXLEN));
3885 	IFQ_SET_READY(&ifp->if_snd);
3886 	DPRINTFN(5, ("strcpy if_xname\n"));
3887 	strcpy(ifp->if_xname, device_xname(sc->bge_dev));
3888 
3889 	if (sc->bge_chipid != BGE_CHIPID_BCM5700_B0)
3890 		sc->ethercom.ec_if.if_capabilities |=
3891 		    IFCAP_CSUM_IPv4_Tx | IFCAP_CSUM_IPv4_Rx;
3892 #if 1	/* XXX TCP/UDP checksum offload breaks with pf(4) */
3893 		sc->ethercom.ec_if.if_capabilities |=
3894 		    IFCAP_CSUM_TCPv4_Tx | IFCAP_CSUM_TCPv4_Rx |
3895 		    IFCAP_CSUM_UDPv4_Tx | IFCAP_CSUM_UDPv4_Rx;
3896 #endif
3897 	sc->ethercom.ec_capabilities |=
3898 	    ETHERCAP_VLAN_HWTAGGING | ETHERCAP_VLAN_MTU;
3899 	sc->ethercom.ec_capenable |= ETHERCAP_VLAN_HWTAGGING;
3900 
3901 	if (sc->bge_flags & BGEF_TSO)
3902 		sc->ethercom.ec_if.if_capabilities |= IFCAP_TSOv4;
3903 
3904 	/*
3905 	 * Do MII setup.
3906 	 */
3907 	DPRINTFN(5, ("mii setup\n"));
3908 	mii->mii_ifp = ifp;
3909 	mii->mii_readreg = bge_miibus_readreg;
3910 	mii->mii_writereg = bge_miibus_writereg;
3911 	mii->mii_statchg = bge_miibus_statchg;
3912 
3913 	/*
3914 	 * Figure out what sort of media we have by checking the hardware
3915 	 * config word.  Note: on some BCM5700 cards, this value appears to be
3916 	 * unset. If that's the case, we have to rely on identifying the NIC
3917 	 * by its PCI subsystem ID, as we do below for the SysKonnect SK-9D41.
3918 	 * The SysKonnect SK-9D41 is a 1000baseSX card.
3919 	 */
3920 	if (PCI_PRODUCT(subid) == SK_SUBSYSID_9D41 ||
3921 	    (hwcfg & BGE_HWCFG_MEDIA) == BGE_MEDIA_FIBER) {
3922 		if (BGE_IS_5705_PLUS(sc)) {
3923 			sc->bge_flags |= BGEF_FIBER_MII;
3924 			sc->bge_phy_flags |= BGEPHYF_NO_WIRESPEED;
3925 		} else
3926 			sc->bge_flags |= BGEF_FIBER_TBI;
3927 	}
3928 
3929 	/* Set bge_phy_flags before prop_dictionary_set_uint32() */
3930 	if (BGE_IS_JUMBO_CAPABLE(sc))
3931 		sc->bge_phy_flags |= BGEPHYF_JUMBO_CAPABLE;
3932 
3933 	/* set phyflags and chipid before mii_attach() */
3934 	dict = device_properties(self);
3935 	prop_dictionary_set_uint32(dict, "phyflags", sc->bge_phy_flags);
3936 	prop_dictionary_set_uint32(dict, "chipid", sc->bge_chipid);
3937 
3938 	macmode = CSR_READ_4(sc, BGE_MAC_MODE);
3939 	macmode &= ~BGE_MACMODE_PORTMODE;
3940 	/* Initialize ifmedia structures. */
3941 	if (sc->bge_flags & BGEF_FIBER_TBI) {
3942 		CSR_WRITE_4_FLUSH(sc, BGE_MAC_MODE,
3943 		    macmode | BGE_PORTMODE_TBI);
3944 		DELAY(40);
3945 
3946 		struct ifmedia * const ifm = &sc->bge_ifmedia;
3947 		sc->ethercom.ec_ifmedia = ifm;
3948 
3949 		ifmedia_init_with_lock(ifm, IFM_IMASK,
3950 		    bge_ifmedia_upd, bge_ifmedia_sts, sc->sc_intr_lock);
3951 		ifmedia_add(ifm, IFM_ETHER | IFM_1000_SX, 0, NULL);
3952 		ifmedia_add(ifm, IFM_ETHER | IFM_1000_SX | IFM_FDX, 0, NULL);
3953 		ifmedia_add(ifm, IFM_ETHER | IFM_AUTO, 0, NULL);
3954 		ifmedia_set(ifm, IFM_ETHER | IFM_AUTO);
3955 		/* Pretend the user requested this setting */
3956 		sc->bge_ifmedia.ifm_media = sc->bge_ifmedia.ifm_cur->ifm_media;
3957 	} else {
3958 		uint16_t phyreg;
3959 		int rv;
3960 		/*
3961 		 * Do transceiver setup and tell the firmware the
3962 		 * driver is down so we can try to get access the
3963 		 * probe if ASF is running.  Retry a couple of times
3964 		 * if we get a conflict with the ASF firmware accessing
3965 		 * the PHY.
3966 		 */
3967 		if (sc->bge_flags & BGEF_FIBER_MII)
3968 			macmode |= BGE_PORTMODE_GMII;
3969 		else
3970 			macmode |= BGE_PORTMODE_MII;
3971 		CSR_WRITE_4_FLUSH(sc, BGE_MAC_MODE, macmode);
3972 		DELAY(40);
3973 
3974 		/*
3975 		 * Do transceiver setup and tell the firmware the
3976 		 * driver is down so we can try to get access the
3977 		 * probe if ASF is running.  Retry a couple of times
3978 		 * if we get a conflict with the ASF firmware accessing
3979 		 * the PHY.
3980 		 */
3981 		trys = 0;
3982 		BGE_CLRBIT(sc, BGE_MODE_CTL, BGE_MODECTL_STACKUP);
3983 		sc->ethercom.ec_mii = mii;
3984 		ifmedia_init_with_lock(&mii->mii_media, 0, bge_ifmedia_upd,
3985 		    bge_ifmedia_sts, sc->sc_intr_lock);
3986 		mii_flags = MIIF_DOPAUSE;
3987 		if (sc->bge_flags & BGEF_FIBER_MII)
3988 			mii_flags |= MIIF_HAVEFIBER;
3989 again:
3990 		bge_asf_driver_up(sc);
3991 		rv = bge_miibus_readreg(sc->bge_dev, sc->bge_phy_addr,
3992 		    MII_BMCR, &phyreg);
3993 		if ((rv != 0) || ((phyreg & BMCR_PDOWN) != 0)) {
3994 			int i;
3995 
3996 			bge_miibus_writereg(sc->bge_dev, sc->bge_phy_addr,
3997 			    MII_BMCR, BMCR_RESET);
3998 			/* Wait up to 500ms for it to complete. */
3999 			for (i = 0; i < 500; i++) {
4000 				bge_miibus_readreg(sc->bge_dev,
4001 				    sc->bge_phy_addr, MII_BMCR, &phyreg);
4002 				if ((phyreg & BMCR_RESET) == 0)
4003 					break;
4004 				DELAY(1000);
4005 			}
4006 		}
4007 
4008 		mii_attach(sc->bge_dev, mii, capmask, sc->bge_phy_addr,
4009 		    MII_OFFSET_ANY, mii_flags);
4010 
4011 		if (LIST_EMPTY(&mii->mii_phys) && (trys++ < 4))
4012 			goto again;
4013 
4014 		if (LIST_EMPTY(&mii->mii_phys)) {
4015 			aprint_error_dev(sc->bge_dev, "no PHY found!\n");
4016 			ifmedia_add(&mii->mii_media, IFM_ETHER | IFM_MANUAL,
4017 			    0, NULL);
4018 			ifmedia_set(&mii->mii_media, IFM_ETHER | IFM_MANUAL);
4019 		} else
4020 			ifmedia_set(&mii->mii_media, IFM_ETHER | IFM_AUTO);
4021 
4022 		/*
4023 		 * Now tell the firmware we are going up after probing the PHY
4024 		 */
4025 		if (sc->bge_asf_mode & ASF_STACKUP)
4026 			BGE_SETBIT(sc, BGE_MODE_CTL, BGE_MODECTL_STACKUP);
4027 	}
4028 
4029 	/*
4030 	 * Call MI attach routine.
4031 	 */
4032 	DPRINTFN(5, ("if_initialize\n"));
4033 	if_initialize(ifp);
4034 	ifp->if_percpuq = if_percpuq_create(ifp);
4035 	if_deferred_start_init(ifp, NULL);
4036 	if_register(ifp);
4037 
4038 	DPRINTFN(5, ("ether_ifattach\n"));
4039 	ether_ifattach(ifp, eaddr);
4040 	ether_set_ifflags_cb(&sc->ethercom, bge_ifflags_cb);
4041 
4042 	rnd_attach_source(&sc->rnd_source, device_xname(sc->bge_dev),
4043 		RND_TYPE_NET, RND_FLAG_DEFAULT);
4044 #ifdef BGE_EVENT_COUNTERS
4045 	/*
4046 	 * Attach event counters.
4047 	 */
4048 	evcnt_attach_dynamic(&sc->bge_ev_intr, EVCNT_TYPE_INTR,
4049 	    NULL, device_xname(sc->bge_dev), "intr");
4050 	evcnt_attach_dynamic(&sc->bge_ev_intr_spurious, EVCNT_TYPE_INTR,
4051 	    NULL, device_xname(sc->bge_dev), "intr_spurious");
4052 	evcnt_attach_dynamic(&sc->bge_ev_intr_spurious2, EVCNT_TYPE_INTR,
4053 	    NULL, device_xname(sc->bge_dev), "intr_spurious2");
4054 	evcnt_attach_dynamic(&sc->bge_ev_tx_xoff, EVCNT_TYPE_MISC,
4055 	    NULL, device_xname(sc->bge_dev), "tx_xoff");
4056 	evcnt_attach_dynamic(&sc->bge_ev_tx_xon, EVCNT_TYPE_MISC,
4057 	    NULL, device_xname(sc->bge_dev), "tx_xon");
4058 	evcnt_attach_dynamic(&sc->bge_ev_rx_xoff, EVCNT_TYPE_MISC,
4059 	    NULL, device_xname(sc->bge_dev), "rx_xoff");
4060 	evcnt_attach_dynamic(&sc->bge_ev_rx_xon, EVCNT_TYPE_MISC,
4061 	    NULL, device_xname(sc->bge_dev), "rx_xon");
4062 	evcnt_attach_dynamic(&sc->bge_ev_rx_macctl, EVCNT_TYPE_MISC,
4063 	    NULL, device_xname(sc->bge_dev), "rx_macctl");
4064 	evcnt_attach_dynamic(&sc->bge_ev_xoffentered, EVCNT_TYPE_MISC,
4065 	    NULL, device_xname(sc->bge_dev), "xoffentered");
4066 #endif /* BGE_EVENT_COUNTERS */
4067 	DPRINTFN(5, ("callout_init\n"));
4068 	callout_init(&sc->bge_timeout, CALLOUT_MPSAFE);
4069 	callout_setfunc(&sc->bge_timeout, bge_tick, sc);
4070 
4071 	if (pmf_device_register(self, NULL, NULL))
4072 		pmf_class_network_register(self, ifp);
4073 	else
4074 		aprint_error_dev(self, "couldn't establish power handler\n");
4075 
4076 	bge_sysctl_init(sc);
4077 
4078 #ifdef BGE_DEBUG
4079 	bge_debug_info(sc);
4080 #endif
4081 }
4082 
4083 /*
4084  * Stop all chip I/O so that the kernel's probe routines don't
4085  * get confused by errant DMAs when rebooting.
4086  */
4087 static int
4088 bge_detach(device_t self, int flags __unused)
4089 {
4090 	struct bge_softc * const sc = device_private(self);
4091 	struct ifnet * const ifp = &sc->ethercom.ec_if;
4092 
4093 	IFNET_LOCK(ifp);
4094 
4095 	/* Stop the interface. Callouts are stopped in it. */
4096 	bge_stop(ifp, 1);
4097 	sc->bge_detaching = true;
4098 
4099 	IFNET_UNLOCK(ifp);
4100 
4101 	mii_detach(&sc->bge_mii, MII_PHY_ANY, MII_OFFSET_ANY);
4102 
4103 	ether_ifdetach(ifp);
4104 	if_detach(ifp);
4105 
4106 	/* Delete all remaining media. */
4107 	ifmedia_fini(&sc->bge_mii.mii_media);
4108 
4109 	bge_release_resources(sc);
4110 
4111 	return 0;
4112 }
4113 
4114 static void
4115 bge_release_resources(struct bge_softc *sc)
4116 {
4117 
4118 	/* Detach sysctl */
4119 	if (sc->bge_log != NULL)
4120 		sysctl_teardown(&sc->bge_log);
4121 
4122 #ifdef BGE_EVENT_COUNTERS
4123 	/* Detach event counters. */
4124 	evcnt_detach(&sc->bge_ev_intr);
4125 	evcnt_detach(&sc->bge_ev_intr_spurious);
4126 	evcnt_detach(&sc->bge_ev_intr_spurious2);
4127 	evcnt_detach(&sc->bge_ev_tx_xoff);
4128 	evcnt_detach(&sc->bge_ev_tx_xon);
4129 	evcnt_detach(&sc->bge_ev_rx_xoff);
4130 	evcnt_detach(&sc->bge_ev_rx_xon);
4131 	evcnt_detach(&sc->bge_ev_rx_macctl);
4132 	evcnt_detach(&sc->bge_ev_xoffentered);
4133 #endif /* BGE_EVENT_COUNTERS */
4134 
4135 	/* Disestablish the interrupt handler */
4136 	if (sc->bge_intrhand != NULL) {
4137 		pci_intr_disestablish(sc->sc_pc, sc->bge_intrhand);
4138 		pci_intr_release(sc->sc_pc, sc->bge_pihp, 1);
4139 		sc->bge_intrhand = NULL;
4140 	}
4141 
4142 	if (sc->bge_cdata.bge_jumbo_buf != NULL)
4143 		bge_free_jumbo_mem(sc);
4144 
4145 	if (sc->bge_dmatag != NULL) {
4146 		bus_dmamap_unload(sc->bge_dmatag, sc->bge_ring_map);
4147 		bus_dmamap_destroy(sc->bge_dmatag, sc->bge_ring_map);
4148 		bus_dmamem_unmap(sc->bge_dmatag, (void *)sc->bge_rdata,
4149 		    sizeof(struct bge_ring_data));
4150 		bus_dmamem_free(sc->bge_dmatag, &sc->bge_ring_seg,
4151 		    sc->bge_ring_rseg);
4152 	}
4153 
4154 	/* Unmap the device registers */
4155 	if (sc->bge_bsize != 0) {
4156 		bus_space_unmap(sc->bge_btag, sc->bge_bhandle, sc->bge_bsize);
4157 		sc->bge_bsize = 0;
4158 	}
4159 
4160 	/* Unmap the APE registers */
4161 	if (sc->bge_apesize != 0) {
4162 		bus_space_unmap(sc->bge_apetag, sc->bge_apehandle,
4163 		    sc->bge_apesize);
4164 		sc->bge_apesize = 0;
4165 	}
4166 }
4167 
4168 static int
4169 bge_reset(struct bge_softc *sc)
4170 {
4171 	uint32_t cachesize, command;
4172 	uint32_t reset, mac_mode, mac_mode_mask;
4173 	pcireg_t devctl, reg;
4174 	int i, val;
4175 	void (*write_op)(struct bge_softc *, int, int);
4176 
4177 	/* Make mask for BGE_MAC_MODE register. */
4178 	mac_mode_mask = BGE_MACMODE_HALF_DUPLEX | BGE_MACMODE_PORTMODE;
4179 	if ((sc->bge_mfw_flags & BGE_MFW_ON_APE) != 0)
4180 		mac_mode_mask |= BGE_MACMODE_APE_RX_EN | BGE_MACMODE_APE_TX_EN;
4181 	/* Keep mac_mode_mask's bits of BGE_MAC_MODE register into mac_mode */
4182 	mac_mode = CSR_READ_4(sc, BGE_MAC_MODE) & mac_mode_mask;
4183 
4184 	if (BGE_IS_575X_PLUS(sc) && !BGE_IS_5714_FAMILY(sc) &&
4185 	    (BGE_ASICREV(sc->bge_chipid) != BGE_ASICREV_BCM5906)) {
4186 		if (sc->bge_flags & BGEF_PCIE)
4187 			write_op = bge_writemem_direct;
4188 		else
4189 			write_op = bge_writemem_ind;
4190 	} else
4191 		write_op = bge_writereg_ind;
4192 
4193 	/* 57XX step 4 */
4194 	/* Acquire the NVM lock */
4195 	if ((sc->bge_flags & BGEF_NO_EEPROM) == 0 &&
4196 	    BGE_ASICREV(sc->bge_chipid) != BGE_ASICREV_BCM5700 &&
4197 	    BGE_ASICREV(sc->bge_chipid) != BGE_ASICREV_BCM5701) {
4198 		CSR_WRITE_4(sc, BGE_NVRAM_SWARB, BGE_NVRAMSWARB_SET1);
4199 		for (i = 0; i < 8000; i++) {
4200 			if (CSR_READ_4(sc, BGE_NVRAM_SWARB) &
4201 			    BGE_NVRAMSWARB_GNT1)
4202 				break;
4203 			DELAY(20);
4204 		}
4205 		if (i == 8000) {
4206 			printf("%s: NVRAM lock timedout!\n",
4207 			    device_xname(sc->bge_dev));
4208 		}
4209 	}
4210 
4211 	/* Take APE lock when performing reset. */
4212 	bge_ape_lock(sc, BGE_APE_LOCK_GRC);
4213 
4214 	/* 57XX step 3 */
4215 	/* Save some important PCI state. */
4216 	cachesize = pci_conf_read(sc->sc_pc, sc->sc_pcitag, BGE_PCI_CACHESZ);
4217 	/* 5718 reset step 3 */
4218 	command = pci_conf_read(sc->sc_pc, sc->sc_pcitag, BGE_PCI_CMD);
4219 
4220 	/* 5718 reset step 5, 57XX step 5b-5d */
4221 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MISC_CTL,
4222 	    BGE_PCIMISCCTL_INDIRECT_ACCESS | BGE_PCIMISCCTL_MASK_PCI_INTR |
4223 	    BGE_HIF_SWAP_OPTIONS | BGE_PCIMISCCTL_PCISTATE_RW);
4224 
4225 	/* XXX ???: Disable fastboot on controllers that support it. */
4226 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5752 ||
4227 	    BGE_IS_5755_PLUS(sc))
4228 		CSR_WRITE_4(sc, BGE_FASTBOOT_PC, 0);
4229 
4230 	/* 5718 reset step 2, 57XX step 6 */
4231 	/*
4232 	 * Write the magic number to SRAM at offset 0xB50.
4233 	 * When firmware finishes its initialization it will
4234 	 * write ~BGE_MAGIC_NUMBER to the same location.
4235 	 */
4236 	bge_writemem_ind(sc, BGE_SRAM_FW_MB, BGE_SRAM_FW_MB_MAGIC);
4237 
4238 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM57780) {
4239 		val = CSR_READ_4(sc, BGE_PCIE_LINKCTL);
4240 		val = (val & ~BGE_PCIE_LINKCTL_L1_PLL_PDEN)
4241 		    | BGE_PCIE_LINKCTL_L1_PLL_PDDIS;
4242 		CSR_WRITE_4(sc, BGE_PCIE_LINKCTL, val);
4243 	}
4244 
4245 	/* 5718 reset step 6, 57XX step 7 */
4246 	reset = BGE_MISCCFG_RESET_CORE_CLOCKS | BGE_32BITTIME_66MHZ;
4247 	/*
4248 	 * XXX: from FreeBSD/Linux; no documentation
4249 	 */
4250 	if (sc->bge_flags & BGEF_PCIE) {
4251 		if ((BGE_ASICREV(sc->bge_chipid) != BGE_ASICREV_BCM5785) &&
4252 		    !BGE_IS_57765_PLUS(sc) &&
4253 		    (CSR_READ_4(sc, BGE_PHY_TEST_CTRL_REG) ==
4254 			(BGE_PHY_PCIE_LTASS_MODE | BGE_PHY_PCIE_SCRAM_MODE))) {
4255 			/* PCI Express 1.0 system */
4256 			CSR_WRITE_4(sc, BGE_PHY_TEST_CTRL_REG,
4257 			    BGE_PHY_PCIE_SCRAM_MODE);
4258 		}
4259 		if (sc->bge_chipid != BGE_CHIPID_BCM5750_A0) {
4260 			/*
4261 			 * Prevent PCI Express link training
4262 			 * during global reset.
4263 			 */
4264 			CSR_WRITE_4(sc, BGE_MISC_CFG, 1 << 29);
4265 			reset |= (1 << 29);
4266 		}
4267 	}
4268 
4269 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906) {
4270 		i = CSR_READ_4(sc, BGE_VCPU_STATUS);
4271 		CSR_WRITE_4(sc, BGE_VCPU_STATUS,
4272 		    i | BGE_VCPU_STATUS_DRV_RESET);
4273 		i = CSR_READ_4(sc, BGE_VCPU_EXT_CTRL);
4274 		CSR_WRITE_4(sc, BGE_VCPU_EXT_CTRL,
4275 		    i & ~BGE_VCPU_EXT_CTRL_HALT_CPU);
4276 	}
4277 
4278 	/*
4279 	 * Set GPHY Power Down Override to leave GPHY
4280 	 * powered up in D0 uninitialized.
4281 	 */
4282 	if (BGE_IS_5705_PLUS(sc) &&
4283 	    (sc->bge_flags & BGEF_CPMU_PRESENT) == 0)
4284 		reset |= BGE_MISCCFG_GPHY_PD_OVERRIDE;
4285 
4286 	/* Issue global reset */
4287 	write_op(sc, BGE_MISC_CFG, reset);
4288 
4289 	/* 5718 reset step 7, 57XX step 8 */
4290 	if (sc->bge_flags & BGEF_PCIE)
4291 		delay(100*1000); /* too big */
4292 	else
4293 		delay(1000);
4294 
4295 	if (sc->bge_flags & BGEF_PCIE) {
4296 		if (sc->bge_chipid == BGE_CHIPID_BCM5750_A0) {
4297 			DELAY(500000);
4298 			/* XXX: Magic Numbers */
4299 			reg = pci_conf_read(sc->sc_pc, sc->sc_pcitag,
4300 			    BGE_PCI_UNKNOWN0);
4301 			pci_conf_write(sc->sc_pc, sc->sc_pcitag,
4302 			    BGE_PCI_UNKNOWN0,
4303 			    reg | (1 << 15));
4304 		}
4305 		devctl = pci_conf_read(sc->sc_pc, sc->sc_pcitag,
4306 		    sc->bge_pciecap + PCIE_DCSR);
4307 		/* Clear enable no snoop and disable relaxed ordering. */
4308 		devctl &= ~(PCIE_DCSR_ENA_RELAX_ORD |
4309 		    PCIE_DCSR_ENA_NO_SNOOP);
4310 
4311 		/* Set PCIE max payload size to 128 for older PCIe devices */
4312 		if ((sc->bge_flags & BGEF_CPMU_PRESENT) == 0)
4313 			devctl &= ~(0x00e0);
4314 		/* Clear device status register. Write 1b to clear */
4315 		devctl |= PCIE_DCSR_URD | PCIE_DCSR_FED
4316 		    | PCIE_DCSR_NFED | PCIE_DCSR_CED;
4317 		pci_conf_write(sc->sc_pc, sc->sc_pcitag,
4318 		    sc->bge_pciecap + PCIE_DCSR, devctl);
4319 		bge_set_max_readrq(sc);
4320 	}
4321 
4322 	/* From Linux: dummy read to flush PCI posted writes */
4323 	reg = pci_conf_read(sc->sc_pc, sc->sc_pcitag, BGE_PCI_CMD);
4324 
4325 	/*
4326 	 * Reset some of the PCI state that got zapped by reset
4327 	 * To modify the PCISTATE register, BGE_PCIMISCCTL_PCISTATE_RW must be
4328 	 * set, too.
4329 	 */
4330 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MISC_CTL,
4331 	    BGE_PCIMISCCTL_INDIRECT_ACCESS | BGE_PCIMISCCTL_MASK_PCI_INTR |
4332 	    BGE_HIF_SWAP_OPTIONS | BGE_PCIMISCCTL_PCISTATE_RW);
4333 	val = BGE_PCISTATE_ROM_ENABLE | BGE_PCISTATE_ROM_RETRY_ENABLE;
4334 	if (sc->bge_chipid == BGE_CHIPID_BCM5704_A0 &&
4335 	    (sc->bge_flags & BGEF_PCIX) != 0)
4336 		val |= BGE_PCISTATE_RETRY_SAME_DMA;
4337 	if ((sc->bge_mfw_flags & BGE_MFW_ON_APE) != 0)
4338 		val |= BGE_PCISTATE_ALLOW_APE_CTLSPC_WR |
4339 		    BGE_PCISTATE_ALLOW_APE_SHMEM_WR |
4340 		    BGE_PCISTATE_ALLOW_APE_PSPACE_WR;
4341 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_PCISTATE, val);
4342 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_CACHESZ, cachesize);
4343 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_CMD, command);
4344 
4345 	/* 57xx step 11: disable PCI-X Relaxed Ordering. */
4346 	if (sc->bge_flags & BGEF_PCIX) {
4347 		reg = pci_conf_read(sc->sc_pc, sc->sc_pcitag, sc->bge_pcixcap
4348 		    + PCIX_CMD);
4349 		/* Set max memory read byte count to 2K */
4350 		if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5703) {
4351 			reg &= ~PCIX_CMD_BYTECNT_MASK;
4352 			reg |= PCIX_CMD_BCNT_2048;
4353 		} else if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5704){
4354 			/*
4355 			 * For 5704, set max outstanding split transaction
4356 			 * field to 0 (0 means it supports 1 request)
4357 			 */
4358 			reg &= ~(PCIX_CMD_SPLTRANS_MASK
4359 			    | PCIX_CMD_BYTECNT_MASK);
4360 			reg |= PCIX_CMD_BCNT_2048;
4361 		}
4362 		pci_conf_write(sc->sc_pc, sc->sc_pcitag, sc->bge_pcixcap
4363 		    + PCIX_CMD, reg & ~PCIX_CMD_RELAXED_ORDER);
4364 	}
4365 
4366 	/* 5718 reset step 10, 57XX step 12 */
4367 	/* Enable memory arbiter. */
4368 	if (BGE_IS_5714_FAMILY(sc)) {
4369 		val = CSR_READ_4(sc, BGE_MARB_MODE);
4370 		CSR_WRITE_4(sc, BGE_MARB_MODE, BGE_MARBMODE_ENABLE | val);
4371 	} else
4372 		CSR_WRITE_4(sc, BGE_MARB_MODE, BGE_MARBMODE_ENABLE);
4373 
4374 	/* XXX 5721, 5751 and 5752 */
4375 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5750) {
4376 		/* Step 19: */
4377 		BGE_SETBIT(sc, BGE_TLP_CONTROL_REG, 1 << 29 | 1 << 25);
4378 		/* Step 20: */
4379 		BGE_SETBIT(sc, BGE_TLP_CONTROL_REG, BGE_TLP_DATA_FIFO_PROTECT);
4380 	}
4381 
4382 	/* 5718 reset step 12, 57XX step 15 and 16 */
4383 	/* Fix up byte swapping */
4384 	CSR_WRITE_4(sc, BGE_MODE_CTL, BGE_DMA_SWAP_OPTIONS);
4385 
4386 	/* 5718 reset step 13, 57XX step 17 */
4387 	/* Poll until the firmware initialization is complete */
4388 	bge_poll_fw(sc);
4389 
4390 	/* 57XX step 21 */
4391 	if (BGE_CHIPREV(sc->bge_chipid) == BGE_CHIPREV_5704_BX) {
4392 		pcireg_t msidata;
4393 
4394 		msidata = pci_conf_read(sc->sc_pc, sc->sc_pcitag,
4395 		    BGE_PCI_MSI_DATA);
4396 		msidata |= ((1 << 13 | 1 << 12 | 1 << 10) << 16);
4397 		pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MSI_DATA,
4398 		    msidata);
4399 	}
4400 
4401 	/* 57XX step 18 */
4402 	/* Write mac mode. */
4403 	val = CSR_READ_4(sc, BGE_MAC_MODE);
4404 	/* Restore mac_mode_mask's bits using mac_mode */
4405 	val = (val & ~mac_mode_mask) | mac_mode;
4406 	CSR_WRITE_4_FLUSH(sc, BGE_MAC_MODE, val);
4407 	DELAY(40);
4408 
4409 	bge_ape_unlock(sc, BGE_APE_LOCK_GRC);
4410 
4411 	/*
4412 	 * The 5704 in TBI mode apparently needs some special
4413 	 * adjustment to insure the SERDES drive level is set
4414 	 * to 1.2V.
4415 	 */
4416 	if (sc->bge_flags & BGEF_FIBER_TBI &&
4417 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5704) {
4418 		uint32_t serdescfg;
4419 
4420 		serdescfg = CSR_READ_4(sc, BGE_SERDES_CFG);
4421 		serdescfg = (serdescfg & ~0xFFF) | 0x880;
4422 		CSR_WRITE_4(sc, BGE_SERDES_CFG, serdescfg);
4423 	}
4424 
4425 	if (sc->bge_flags & BGEF_PCIE &&
4426 	    !BGE_IS_57765_PLUS(sc) &&
4427 	    sc->bge_chipid != BGE_CHIPID_BCM5750_A0 &&
4428 	    BGE_ASICREV(sc->bge_chipid) != BGE_ASICREV_BCM5785) {
4429 		uint32_t v;
4430 
4431 		/* Enable PCI Express bug fix */
4432 		v = CSR_READ_4(sc, BGE_TLP_CONTROL_REG);
4433 		CSR_WRITE_4(sc, BGE_TLP_CONTROL_REG,
4434 		    v | BGE_TLP_DATA_FIFO_PROTECT);
4435 	}
4436 
4437 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5720)
4438 		BGE_CLRBIT(sc, BGE_CPMU_CLCK_ORIDE,
4439 		    CPMU_CLCK_ORIDE_MAC_ORIDE_EN);
4440 
4441 	return 0;
4442 }
4443 
4444 /*
4445  * Frame reception handling. This is called if there's a frame
4446  * on the receive return list.
4447  *
4448  * Note: we have to be able to handle two possibilities here:
4449  * 1) the frame is from the jumbo receive ring
4450  * 2) the frame is from the standard receive ring
4451  */
4452 
4453 static void
4454 bge_rxeof(struct bge_softc *sc)
4455 {
4456 	struct ifnet * const ifp = &sc->ethercom.ec_if;
4457 	uint16_t rx_prod, rx_cons;
4458 	int stdcnt = 0, jumbocnt = 0;
4459 	bus_dmamap_t dmamap;
4460 	bus_addr_t offset, toff;
4461 	bus_size_t tlen;
4462 	int tosync;
4463 
4464 	bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
4465 	    offsetof(struct bge_ring_data, bge_status_block),
4466 	    sizeof(struct bge_status_block),
4467 	    BUS_DMASYNC_POSTREAD);
4468 
4469 	rx_cons = sc->bge_rx_saved_considx;
4470 	rx_prod = sc->bge_rdata->bge_status_block.bge_idx[0].bge_rx_prod_idx;
4471 
4472 	/* Nothing to do */
4473 	if (rx_cons == rx_prod)
4474 		return;
4475 
4476 	offset = offsetof(struct bge_ring_data, bge_rx_return_ring);
4477 	tosync = rx_prod - rx_cons;
4478 
4479 	if (tosync != 0)
4480 		rnd_add_uint32(&sc->rnd_source, tosync);
4481 
4482 	toff = offset + (rx_cons * sizeof(struct bge_rx_bd));
4483 
4484 	if (tosync < 0) {
4485 		tlen = (sc->bge_return_ring_cnt - rx_cons) *
4486 		    sizeof(struct bge_rx_bd);
4487 		bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
4488 		    toff, tlen, BUS_DMASYNC_POSTREAD);
4489 		tosync = rx_prod;
4490 		toff = offset;
4491 	}
4492 
4493 	if (tosync != 0) {
4494 		bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
4495 		    toff, tosync * sizeof(struct bge_rx_bd),
4496 		    BUS_DMASYNC_POSTREAD);
4497 	}
4498 
4499 	while (rx_cons != rx_prod) {
4500 		struct bge_rx_bd	*cur_rx;
4501 		uint32_t		rxidx;
4502 		struct mbuf		*m = NULL;
4503 
4504 		cur_rx = &sc->bge_rdata->bge_rx_return_ring[rx_cons];
4505 
4506 		rxidx = cur_rx->bge_idx;
4507 		BGE_INC(rx_cons, sc->bge_return_ring_cnt);
4508 
4509 		if (cur_rx->bge_flags & BGE_RXBDFLAG_JUMBO_RING) {
4510 			BGE_INC(sc->bge_jumbo, BGE_JUMBO_RX_RING_CNT);
4511 			m = sc->bge_cdata.bge_rx_jumbo_chain[rxidx];
4512 			sc->bge_cdata.bge_rx_jumbo_chain[rxidx] = NULL;
4513 			jumbocnt++;
4514 			bus_dmamap_sync(sc->bge_dmatag,
4515 			    sc->bge_cdata.bge_rx_jumbo_map,
4516 			    mtod(m, char *) - (char *)sc->bge_cdata.bge_jumbo_buf,
4517 			    BGE_JLEN, BUS_DMASYNC_POSTREAD);
4518 			if (cur_rx->bge_flags & BGE_RXBDFLAG_ERROR) {
4519 				if_statinc(ifp, if_ierrors);
4520 				bge_newbuf_jumbo(sc, sc->bge_jumbo, m);
4521 				continue;
4522 			}
4523 			if (bge_newbuf_jumbo(sc, sc->bge_jumbo,
4524 					     NULL) == ENOBUFS) {
4525 				if_statinc(ifp, if_ierrors);
4526 				bge_newbuf_jumbo(sc, sc->bge_jumbo, m);
4527 				continue;
4528 			}
4529 		} else {
4530 			m = sc->bge_cdata.bge_rx_std_chain[rxidx];
4531 			sc->bge_cdata.bge_rx_std_chain[rxidx] = NULL;
4532 
4533 			stdcnt++;
4534 			sc->bge_std_cnt--;
4535 
4536 			dmamap = sc->bge_cdata.bge_rx_std_map[rxidx];
4537 			bus_dmamap_sync(sc->bge_dmatag, dmamap, 0,
4538 			    dmamap->dm_mapsize, BUS_DMASYNC_POSTREAD);
4539 			bus_dmamap_unload(sc->bge_dmatag, dmamap);
4540 
4541 			if (cur_rx->bge_flags & BGE_RXBDFLAG_ERROR) {
4542 				m_free(m);
4543 				if_statinc(ifp, if_ierrors);
4544 				continue;
4545 			}
4546 		}
4547 
4548 #ifndef __NO_STRICT_ALIGNMENT
4549 		/*
4550 		 * XXX: if the 5701 PCIX-Rx-DMA workaround is in effect,
4551 		 * the Rx buffer has the layer-2 header unaligned.
4552 		 * If our CPU requires alignment, re-align by copying.
4553 		 */
4554 		if (sc->bge_flags & BGEF_RX_ALIGNBUG) {
4555 			memmove(mtod(m, char *) + ETHER_ALIGN, m->m_data,
4556 				cur_rx->bge_len);
4557 			m->m_data += ETHER_ALIGN;
4558 		}
4559 #endif
4560 
4561 		m->m_pkthdr.len = m->m_len = cur_rx->bge_len - ETHER_CRC_LEN;
4562 		m_set_rcvif(m, ifp);
4563 
4564 		bge_rxcsum(sc, cur_rx, m);
4565 
4566 		/*
4567 		 * If we received a packet with a vlan tag, pass it
4568 		 * to vlan_input() instead of ether_input().
4569 		 */
4570 		if (cur_rx->bge_flags & BGE_RXBDFLAG_VLAN_TAG)
4571 			vlan_set_tag(m, cur_rx->bge_vlan_tag);
4572 
4573 		if_percpuq_enqueue(ifp->if_percpuq, m);
4574 	}
4575 
4576 	sc->bge_rx_saved_considx = rx_cons;
4577 	bge_writembx(sc, BGE_MBX_RX_CONS0_LO, sc->bge_rx_saved_considx);
4578 	if (stdcnt)
4579 		bge_fill_rx_ring_std(sc);
4580 	if (jumbocnt)
4581 		bge_writembx(sc, BGE_MBX_RX_JUMBO_PROD_LO, sc->bge_jumbo);
4582 }
4583 
4584 static void
4585 bge_rxcsum(struct bge_softc *sc, struct bge_rx_bd *cur_rx, struct mbuf *m)
4586 {
4587 
4588 	if (BGE_IS_57765_PLUS(sc)) {
4589 		if ((cur_rx->bge_flags & BGE_RXBDFLAG_IPV6) == 0) {
4590 			if ((cur_rx->bge_flags & BGE_RXBDFLAG_IP_CSUM) != 0)
4591 				m->m_pkthdr.csum_flags = M_CSUM_IPv4;
4592 			if ((cur_rx->bge_error_flag &
4593 				BGE_RXERRFLAG_IP_CSUM_NOK) != 0)
4594 				m->m_pkthdr.csum_flags |= M_CSUM_IPv4_BAD;
4595 			if (cur_rx->bge_flags & BGE_RXBDFLAG_TCP_UDP_CSUM) {
4596 				m->m_pkthdr.csum_data =
4597 				    cur_rx->bge_tcp_udp_csum;
4598 				m->m_pkthdr.csum_flags |=
4599 				    (M_CSUM_TCPv4 | M_CSUM_UDPv4 |M_CSUM_DATA);
4600 			}
4601 		}
4602 	} else {
4603 		if ((cur_rx->bge_flags & BGE_RXBDFLAG_IP_CSUM) != 0)
4604 			m->m_pkthdr.csum_flags = M_CSUM_IPv4;
4605 		if ((cur_rx->bge_ip_csum ^ 0xffff) != 0)
4606 			m->m_pkthdr.csum_flags |= M_CSUM_IPv4_BAD;
4607 		/*
4608 		 * Rx transport checksum-offload may also
4609 		 * have bugs with packets which, when transmitted,
4610 		 * were `runts' requiring padding.
4611 		 */
4612 		if (cur_rx->bge_flags & BGE_RXBDFLAG_TCP_UDP_CSUM &&
4613 		    (/* (sc->_bge_quirks & BGE_QUIRK_SHORT_CKSUM_BUG) == 0 ||*/
4614 			    m->m_pkthdr.len >= ETHER_MIN_NOPAD)) {
4615 			m->m_pkthdr.csum_data =
4616 			    cur_rx->bge_tcp_udp_csum;
4617 			m->m_pkthdr.csum_flags |=
4618 			    (M_CSUM_TCPv4 | M_CSUM_UDPv4 | M_CSUM_DATA);
4619 		}
4620 	}
4621 }
4622 
4623 static void
4624 bge_txeof(struct bge_softc *sc)
4625 {
4626 	struct ifnet * const ifp = &sc->ethercom.ec_if;
4627 	struct bge_tx_bd *cur_tx = NULL;
4628 	struct txdmamap_pool_entry *dma;
4629 	bus_addr_t offset, toff;
4630 	bus_size_t tlen;
4631 	int tosync;
4632 	struct mbuf *m;
4633 
4634 	bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
4635 	    offsetof(struct bge_ring_data, bge_status_block),
4636 	    sizeof(struct bge_status_block),
4637 	    BUS_DMASYNC_POSTREAD);
4638 
4639 	const uint16_t hw_cons_idx =
4640 	    sc->bge_rdata->bge_status_block.bge_idx[0].bge_tx_cons_idx;
4641 	offset = offsetof(struct bge_ring_data, bge_tx_ring);
4642 	tosync = hw_cons_idx - sc->bge_tx_saved_considx;
4643 
4644 	if (tosync != 0)
4645 		rnd_add_uint32(&sc->rnd_source, tosync);
4646 
4647 	toff = offset + (sc->bge_tx_saved_considx * sizeof(struct bge_tx_bd));
4648 
4649 	if (tosync < 0) {
4650 		tlen = (BGE_TX_RING_CNT - sc->bge_tx_saved_considx) *
4651 		    sizeof(struct bge_tx_bd);
4652 		bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
4653 		    toff, tlen, BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
4654 		tosync = hw_cons_idx;
4655 		toff = offset;
4656 	}
4657 
4658 	if (tosync != 0) {
4659 		bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
4660 		    toff, tosync * sizeof(struct bge_tx_bd),
4661 		    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
4662 	}
4663 
4664 	/*
4665 	 * Go through our tx ring and free mbufs for those
4666 	 * frames that have been sent.
4667 	 */
4668 	while (sc->bge_tx_saved_considx != hw_cons_idx) {
4669 		uint32_t idx = sc->bge_tx_saved_considx;
4670 		cur_tx = &sc->bge_rdata->bge_tx_ring[idx];
4671 		if (cur_tx->bge_flags & BGE_TXBDFLAG_END)
4672 			if_statinc(ifp, if_opackets);
4673 		m = sc->bge_cdata.bge_tx_chain[idx];
4674 		if (m != NULL) {
4675 			sc->bge_cdata.bge_tx_chain[idx] = NULL;
4676 			dma = sc->txdma[idx];
4677 			if (dma->is_dma32) {
4678 				bus_dmamap_sync(sc->bge_dmatag32, dma->dmamap32,
4679 				    0, dma->dmamap32->dm_mapsize,
4680 				    BUS_DMASYNC_POSTWRITE);
4681 				bus_dmamap_unload(
4682 				    sc->bge_dmatag32, dma->dmamap32);
4683 			} else {
4684 				bus_dmamap_sync(sc->bge_dmatag, dma->dmamap,
4685 				    0, dma->dmamap->dm_mapsize,
4686 				    BUS_DMASYNC_POSTWRITE);
4687 				bus_dmamap_unload(sc->bge_dmatag, dma->dmamap);
4688 			}
4689 			SLIST_INSERT_HEAD(&sc->txdma_list, dma, link);
4690 			sc->txdma[idx] = NULL;
4691 
4692 			m_freem(m);
4693 		}
4694 		sc->bge_txcnt--;
4695 		BGE_INC(sc->bge_tx_saved_considx, BGE_TX_RING_CNT);
4696 		sc->bge_tx_sending = false;
4697 	}
4698 }
4699 
4700 static int
4701 bge_intr(void *xsc)
4702 {
4703 	struct bge_softc * const sc = xsc;
4704 	struct ifnet * const ifp = &sc->ethercom.ec_if;
4705 	uint32_t pcistate, statusword, statustag;
4706 	uint32_t intrmask = BGE_PCISTATE_INTR_NOT_ACTIVE;
4707 
4708 	/* 5717 and newer chips have no BGE_PCISTATE_INTR_NOT_ACTIVE bit */
4709 	if (BGE_IS_5717_PLUS(sc))
4710 		intrmask = 0;
4711 
4712 	mutex_enter(sc->sc_intr_lock);
4713 	if (sc->bge_txrx_stopping) {
4714 		mutex_exit(sc->sc_intr_lock);
4715 		return 1;
4716 	}
4717 
4718 	/*
4719 	 * It is possible for the interrupt to arrive before
4720 	 * the status block is updated prior to the interrupt.
4721 	 * Reading the PCI State register will confirm whether the
4722 	 * interrupt is ours and will flush the status block.
4723 	 */
4724 	pcistate = CSR_READ_4(sc, BGE_PCI_PCISTATE);
4725 
4726 	/* read status word from status block */
4727 	bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
4728 	    offsetof(struct bge_ring_data, bge_status_block),
4729 	    sizeof(struct bge_status_block),
4730 	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
4731 	statusword = sc->bge_rdata->bge_status_block.bge_status;
4732 	statustag = sc->bge_rdata->bge_status_block.bge_status_tag << 24;
4733 
4734 	if (sc->bge_flags & BGEF_TAGGED_STATUS) {
4735 		if (sc->bge_lasttag == statustag &&
4736 		    (~pcistate & intrmask)) {
4737 			BGE_EVCNT_INCR(sc->bge_ev_intr_spurious);
4738 			mutex_exit(sc->sc_intr_lock);
4739 			return 0;
4740 		}
4741 		sc->bge_lasttag = statustag;
4742 	} else {
4743 		if (!(statusword & BGE_STATFLAG_UPDATED) &&
4744 		    !(~pcistate & intrmask)) {
4745 			BGE_EVCNT_INCR(sc->bge_ev_intr_spurious2);
4746 			mutex_exit(sc->sc_intr_lock);
4747 			return 0;
4748 		}
4749 		statustag = 0;
4750 	}
4751 	/* Ack interrupt and stop others from occurring. */
4752 	bge_writembx_flush(sc, BGE_MBX_IRQ0_LO, 1);
4753 	BGE_EVCNT_INCR(sc->bge_ev_intr);
4754 
4755 	/* clear status word */
4756 	sc->bge_rdata->bge_status_block.bge_status = 0;
4757 
4758 	bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
4759 	    offsetof(struct bge_ring_data, bge_status_block),
4760 	    sizeof(struct bge_status_block),
4761 	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
4762 
4763 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5700 ||
4764 	    statusword & BGE_STATFLAG_LINKSTATE_CHANGED ||
4765 	    BGE_STS_BIT(sc, BGE_STS_LINK_EVT))
4766 		bge_link_upd(sc);
4767 
4768 	/* Check RX return ring producer/consumer */
4769 	bge_rxeof(sc);
4770 
4771 	/* Check TX ring producer/consumer */
4772 	bge_txeof(sc);
4773 
4774 	if (sc->bge_pending_rxintr_change) {
4775 		uint32_t rx_ticks = sc->bge_rx_coal_ticks;
4776 		uint32_t rx_bds = sc->bge_rx_max_coal_bds;
4777 
4778 		CSR_WRITE_4(sc, BGE_HCC_RX_COAL_TICKS, rx_ticks);
4779 		DELAY(10);
4780 		(void)CSR_READ_4(sc, BGE_HCC_RX_COAL_TICKS);
4781 
4782 		CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS, rx_bds);
4783 		DELAY(10);
4784 		(void)CSR_READ_4(sc, BGE_HCC_RX_MAX_COAL_BDS);
4785 
4786 		sc->bge_pending_rxintr_change = false;
4787 	}
4788 	bge_handle_events(sc);
4789 
4790 	/* Re-enable interrupts. */
4791 	bge_writembx_flush(sc, BGE_MBX_IRQ0_LO, statustag);
4792 
4793 	if_schedule_deferred_start(ifp);
4794 
4795 	mutex_exit(sc->sc_intr_lock);
4796 
4797 	return 1;
4798 }
4799 
4800 static void
4801 bge_asf_driver_up(struct bge_softc *sc)
4802 {
4803 	if (sc->bge_asf_mode & ASF_STACKUP) {
4804 		/* Send ASF heartbeat approx. every 2s */
4805 		if (sc->bge_asf_count)
4806 			sc->bge_asf_count --;
4807 		else {
4808 			sc->bge_asf_count = 2;
4809 
4810 			bge_wait_for_event_ack(sc);
4811 
4812 			bge_writemem_ind(sc, BGE_SRAM_FW_CMD_MB,
4813 			    BGE_FW_CMD_DRV_ALIVE3);
4814 			bge_writemem_ind(sc, BGE_SRAM_FW_CMD_LEN_MB, 4);
4815 			bge_writemem_ind(sc, BGE_SRAM_FW_CMD_DATA_MB,
4816 			    BGE_FW_HB_TIMEOUT_SEC);
4817 			CSR_WRITE_4_FLUSH(sc, BGE_RX_CPU_EVENT,
4818 			    CSR_READ_4(sc, BGE_RX_CPU_EVENT) |
4819 			    BGE_RX_CPU_DRV_EVENT);
4820 		}
4821 	}
4822 }
4823 
4824 static void
4825 bge_tick(void *xsc)
4826 {
4827 	struct bge_softc * const sc = xsc;
4828 	struct ifnet * const ifp = &sc->ethercom.ec_if;
4829 	struct mii_data * const mii = &sc->bge_mii;
4830 
4831 	mutex_enter(sc->sc_core_lock);
4832 	if (sc->bge_stopping) {
4833 		mutex_exit(sc->sc_core_lock);
4834 		return;
4835 	}
4836 
4837 	if (BGE_IS_5705_PLUS(sc))
4838 		bge_stats_update_regs(sc);
4839 	else
4840 		bge_stats_update(sc);
4841 
4842 	if (sc->bge_flags & BGEF_FIBER_TBI) {
4843 		/*
4844 		 * Since in TBI mode auto-polling can't be used we should poll
4845 		 * link status manually. Here we register pending link event
4846 		 * and trigger interrupt.
4847 		 */
4848 		BGE_STS_SETBIT(sc, BGE_STS_LINK_EVT);
4849 		BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_INTR_SET);
4850 	} else {
4851 		/*
4852 		 * Do not touch PHY if we have link up. This could break
4853 		 * IPMI/ASF mode or produce extra input errors.
4854 		 * (extra input errors was reported for bcm5701 & bcm5704).
4855 		 */
4856 		if (!BGE_STS_BIT(sc, BGE_STS_LINK)) {
4857 			mutex_enter(sc->sc_intr_lock);
4858 			mii_tick(mii);
4859 			mutex_exit(sc->sc_intr_lock);
4860 		}
4861 	}
4862 
4863 	bge_asf_driver_up(sc);
4864 
4865 	const bool ok = bge_watchdog_tick(ifp);
4866 	if (ok)
4867 		callout_schedule(&sc->bge_timeout, hz);
4868 
4869 	mutex_exit(sc->sc_core_lock);
4870 }
4871 
4872 static void
4873 bge_stats_update_regs(struct bge_softc *sc)
4874 {
4875 	struct ifnet * const ifp = &sc->ethercom.ec_if;
4876 
4877 	net_stat_ref_t nsr = IF_STAT_GETREF(ifp);
4878 
4879 	if_statadd_ref(ifp, nsr, if_collisions,
4880 	    CSR_READ_4(sc, BGE_MAC_STATS +
4881 	    offsetof(struct bge_mac_stats_regs, etherStatsCollisions)));
4882 
4883 	/*
4884 	 * On BCM5717, BCM5718, BCM5719 A0 and BCM5720 A0,
4885 	 * RXLP_LOCSTAT_IFIN_DROPS includes unwanted multicast frames
4886 	 * (silicon bug). There's no reliable workaround so just
4887 	 * ignore the counter
4888 	 */
4889 	if (BGE_ASICREV(sc->bge_chipid) != BGE_ASICREV_BCM5717 &&
4890 	    sc->bge_chipid != BGE_CHIPID_BCM5719_A0 &&
4891 	    sc->bge_chipid != BGE_CHIPID_BCM5720_A0) {
4892 		if_statadd_ref(ifp, nsr, if_ierrors,
4893 		    CSR_READ_4(sc, BGE_RXLP_LOCSTAT_IFIN_DROPS));
4894 	}
4895 	if_statadd_ref(ifp, nsr, if_ierrors,
4896 	    CSR_READ_4(sc, BGE_RXLP_LOCSTAT_IFIN_ERRORS));
4897 	if_statadd_ref(ifp, nsr, if_ierrors,
4898 	    CSR_READ_4(sc, BGE_RXLP_LOCSTAT_OUT_OF_BDS));
4899 
4900 	IF_STAT_PUTREF(ifp);
4901 
4902 	if (sc->bge_flags & BGEF_RDMA_BUG) {
4903 		uint32_t val, ucast, mcast, bcast;
4904 
4905 		ucast = CSR_READ_4(sc, BGE_MAC_STATS +
4906 		    offsetof(struct bge_mac_stats_regs, ifHCOutUcastPkts));
4907 		mcast = CSR_READ_4(sc, BGE_MAC_STATS +
4908 		    offsetof(struct bge_mac_stats_regs, ifHCOutMulticastPkts));
4909 		bcast = CSR_READ_4(sc, BGE_MAC_STATS +
4910 		    offsetof(struct bge_mac_stats_regs, ifHCOutBroadcastPkts));
4911 
4912 		/*
4913 		 * If controller transmitted more than BGE_NUM_RDMA_CHANNELS
4914 		 * frames, it's safe to disable workaround for DMA engine's
4915 		 * miscalculation of TXMBUF space.
4916 		 */
4917 		if (ucast + mcast + bcast > BGE_NUM_RDMA_CHANNELS) {
4918 			val = CSR_READ_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL);
4919 			if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5719)
4920 				val &= ~BGE_RDMA_TX_LENGTH_WA_5719;
4921 			else
4922 				val &= ~BGE_RDMA_TX_LENGTH_WA_5720;
4923 			CSR_WRITE_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL, val);
4924 			sc->bge_flags &= ~BGEF_RDMA_BUG;
4925 		}
4926 	}
4927 }
4928 
4929 static void
4930 bge_stats_update(struct bge_softc *sc)
4931 {
4932 	struct ifnet * const ifp = &sc->ethercom.ec_if;
4933 	bus_size_t stats = BGE_MEMWIN_START + BGE_STATS_BLOCK;
4934 
4935 #define READ_STAT(sc, stats, stat) \
4936 	  CSR_READ_4(sc, stats + offsetof(struct bge_stats, stat))
4937 
4938 	uint64_t collisions =
4939 	  (READ_STAT(sc, stats, dot3StatsSingleCollisionFrames.bge_addr_lo) +
4940 	   READ_STAT(sc, stats, dot3StatsMultipleCollisionFrames.bge_addr_lo) +
4941 	   READ_STAT(sc, stats, dot3StatsExcessiveCollisions.bge_addr_lo) +
4942 	   READ_STAT(sc, stats, dot3StatsLateCollisions.bge_addr_lo));
4943 
4944 	if_statadd(ifp, if_collisions, collisions - sc->bge_if_collisions);
4945 	sc->bge_if_collisions = collisions;
4946 
4947 
4948 	BGE_EVCNT_UPD(sc->bge_ev_tx_xoff,
4949 		      READ_STAT(sc, stats, outXoffSent.bge_addr_lo));
4950 	BGE_EVCNT_UPD(sc->bge_ev_tx_xon,
4951 		      READ_STAT(sc, stats, outXonSent.bge_addr_lo));
4952 	BGE_EVCNT_UPD(sc->bge_ev_rx_xoff,
4953 		      READ_STAT(sc, stats,
4954 				xoffPauseFramesReceived.bge_addr_lo));
4955 	BGE_EVCNT_UPD(sc->bge_ev_rx_xon,
4956 		      READ_STAT(sc, stats, xonPauseFramesReceived.bge_addr_lo));
4957 	BGE_EVCNT_UPD(sc->bge_ev_rx_macctl,
4958 		      READ_STAT(sc, stats,
4959 				macControlFramesReceived.bge_addr_lo));
4960 	BGE_EVCNT_UPD(sc->bge_ev_xoffentered,
4961 		      READ_STAT(sc, stats, xoffStateEntered.bge_addr_lo));
4962 
4963 #undef READ_STAT
4964 }
4965 
4966 /*
4967  * Pad outbound frame to ETHER_MIN_NOPAD for an unusual reason.
4968  * The bge hardware will pad out Tx runts to ETHER_MIN_NOPAD,
4969  * but when such padded frames employ the  bge IP/TCP checksum offload,
4970  * the hardware checksum assist gives incorrect results (possibly
4971  * from incorporating its own padding into the UDP/TCP checksum; who knows).
4972  * If we pad such runts with zeros, the onboard checksum comes out correct.
4973  */
4974 static inline int
4975 bge_cksum_pad(struct mbuf *pkt)
4976 {
4977 	struct mbuf *last = NULL;
4978 	int padlen;
4979 
4980 	padlen = ETHER_MIN_NOPAD - pkt->m_pkthdr.len;
4981 
4982 	/* if there's only the packet-header and we can pad there, use it. */
4983 	if (pkt->m_pkthdr.len == pkt->m_len &&
4984 	    M_TRAILINGSPACE(pkt) >= padlen) {
4985 		last = pkt;
4986 	} else {
4987 		/*
4988 		 * Walk packet chain to find last mbuf. We will either
4989 		 * pad there, or append a new mbuf and pad it
4990 		 * (thus perhaps avoiding the bcm5700 dma-min bug).
4991 		 */
4992 		for (last = pkt; last->m_next != NULL; last = last->m_next) {
4993 			continue; /* do nothing */
4994 		}
4995 
4996 		/* `last' now points to last in chain. */
4997 		if (M_TRAILINGSPACE(last) < padlen) {
4998 			/* Allocate new empty mbuf, pad it. Compact later. */
4999 			struct mbuf *n;
5000 			MGET(n, M_DONTWAIT, MT_DATA);
5001 			if (n == NULL)
5002 				return ENOBUFS;
5003 			n->m_len = 0;
5004 			last->m_next = n;
5005 			last = n;
5006 		}
5007 	}
5008 
5009 	KDASSERT(!M_READONLY(last));
5010 	KDASSERT(M_TRAILINGSPACE(last) >= padlen);
5011 
5012 	/* Now zero the pad area, to avoid the bge cksum-assist bug */
5013 	memset(mtod(last, char *) + last->m_len, 0, padlen);
5014 	last->m_len += padlen;
5015 	pkt->m_pkthdr.len += padlen;
5016 	return 0;
5017 }
5018 
5019 /*
5020  * Compact outbound packets to avoid bug with DMA segments less than 8 bytes.
5021  */
5022 static inline int
5023 bge_compact_dma_runt(struct mbuf *pkt)
5024 {
5025 	struct mbuf	*m, *prev;
5026 	int		totlen;
5027 
5028 	prev = NULL;
5029 	totlen = 0;
5030 
5031 	for (m = pkt; m != NULL; prev = m, m = m->m_next) {
5032 		int mlen = m->m_len;
5033 		int shortfall = 8 - mlen ;
5034 
5035 		totlen += mlen;
5036 		if (mlen == 0)
5037 			continue;
5038 		if (mlen >= 8)
5039 			continue;
5040 
5041 		/*
5042 		 * If we get here, mbuf data is too small for DMA engine.
5043 		 * Try to fix by shuffling data to prev or next in chain.
5044 		 * If that fails, do a compacting deep-copy of the whole chain.
5045 		 */
5046 
5047 		/* Internal frag. If fits in prev, copy it there. */
5048 		if (prev && M_TRAILINGSPACE(prev) >= m->m_len) {
5049 			memcpy(prev->m_data + prev->m_len, m->m_data, mlen);
5050 			prev->m_len += mlen;
5051 			m->m_len = 0;
5052 			/* XXX stitch chain */
5053 			prev->m_next = m_free(m);
5054 			m = prev;
5055 			continue;
5056 		} else if (m->m_next != NULL &&
5057 			    M_TRAILINGSPACE(m) >= shortfall &&
5058 			    m->m_next->m_len >= (8 + shortfall)) {
5059 		    /* m is writable and have enough data in next, pull up. */
5060 
5061 			memcpy(m->m_data + m->m_len, m->m_next->m_data,
5062 			    shortfall);
5063 			m->m_len += shortfall;
5064 			m->m_next->m_len -= shortfall;
5065 			m->m_next->m_data += shortfall;
5066 		} else if (m->m_next == NULL || 1) {
5067 			/*
5068 			 * Got a runt at the very end of the packet.
5069 			 * borrow data from the tail of the preceding mbuf and
5070 			 * update its length in-place. (The original data is
5071 			 * still valid, so we can do this even if prev is not
5072 			 * writable.)
5073 			 */
5074 
5075 			/*
5076 			 * If we'd make prev a runt, just move all of its data.
5077 			 */
5078 			KASSERT(prev != NULL /*, ("runt but null PREV")*/);
5079 			KASSERT(prev->m_len >= 8 /*, ("runt prev")*/);
5080 
5081 			if ((prev->m_len - shortfall) < 8)
5082 				shortfall = prev->m_len;
5083 
5084 #ifdef notyet	/* just do the safe slow thing for now */
5085 			if (!M_READONLY(m)) {
5086 				if (M_LEADINGSPACE(m) < shorfall) {
5087 					void *m_dat;
5088 					m_dat = M_BUFADDR(m);
5089 					memmove(m_dat, mtod(m, void*),
5090 					    m->m_len);
5091 					m->m_data = m_dat;
5092 				}
5093 			} else
5094 #endif	/* just do the safe slow thing */
5095 			{
5096 				struct mbuf * n = NULL;
5097 				int newprevlen = prev->m_len - shortfall;
5098 
5099 				MGET(n, M_NOWAIT, MT_DATA);
5100 				if (n == NULL)
5101 				   return ENOBUFS;
5102 				KASSERT(m->m_len + shortfall < MLEN
5103 					/*,
5104 					  ("runt %d +prev %d too big\n", m->m_len, shortfall)*/);
5105 
5106 				/* first copy the data we're stealing from prev */
5107 				memcpy(n->m_data, prev->m_data + newprevlen,
5108 				    shortfall);
5109 
5110 				/* update prev->m_len accordingly */
5111 				prev->m_len -= shortfall;
5112 
5113 				/* copy data from runt m */
5114 				memcpy(n->m_data + shortfall, m->m_data,
5115 				    m->m_len);
5116 
5117 				/* n holds what we stole from prev, plus m */
5118 				n->m_len = shortfall + m->m_len;
5119 
5120 				/* stitch n into chain and free m */
5121 				n->m_next = m->m_next;
5122 				prev->m_next = n;
5123 				/* KASSERT(m->m_next == NULL); */
5124 				m->m_next = NULL;
5125 				m_free(m);
5126 				m = n;	/* for continuing loop */
5127 			}
5128 		}
5129 	}
5130 	return 0;
5131 }
5132 
5133 /*
5134  * Encapsulate an mbuf chain in the tx ring by coupling the mbuf data
5135  * pointers to descriptors.
5136  */
5137 static int
5138 bge_encap(struct bge_softc *sc, struct mbuf *m_head, uint32_t *txidx)
5139 {
5140 	struct bge_tx_bd	*f, *prev_f;
5141 	uint32_t		frag, cur;
5142 	uint16_t		csum_flags = 0;
5143 	uint16_t		txbd_tso_flags = 0;
5144 	struct txdmamap_pool_entry *dma;
5145 	bus_dmamap_t dmamap;
5146 	bus_dma_tag_t dmatag;
5147 	int			i = 0;
5148 	int			use_tso, maxsegsize, error;
5149 	bool			have_vtag;
5150 	uint16_t		vtag;
5151 	bool			remap;
5152 
5153 	if (m_head->m_pkthdr.csum_flags) {
5154 		if (m_head->m_pkthdr.csum_flags & M_CSUM_IPv4)
5155 			csum_flags |= BGE_TXBDFLAG_IP_CSUM;
5156 		if (m_head->m_pkthdr.csum_flags & (M_CSUM_TCPv4 |M_CSUM_UDPv4))
5157 			csum_flags |= BGE_TXBDFLAG_TCP_UDP_CSUM;
5158 	}
5159 
5160 	/*
5161 	 * If we were asked to do an outboard checksum, and the NIC
5162 	 * has the bug where it sometimes adds in the Ethernet padding,
5163 	 * explicitly pad with zeros so the cksum will be correct either way.
5164 	 * (For now, do this for all chip versions, until newer
5165 	 * are confirmed to not require the workaround.)
5166 	 */
5167 	if ((csum_flags & BGE_TXBDFLAG_TCP_UDP_CSUM) == 0 ||
5168 #ifdef notyet
5169 	    (sc->bge_quirks & BGE_QUIRK_SHORT_CKSUM_BUG) == 0 ||
5170 #endif
5171 	    m_head->m_pkthdr.len >= ETHER_MIN_NOPAD)
5172 		goto check_dma_bug;
5173 
5174 	if (bge_cksum_pad(m_head) != 0)
5175 		return ENOBUFS;
5176 
5177 check_dma_bug:
5178 	if (!(BGE_CHIPREV(sc->bge_chipid) == BGE_CHIPREV_5700_BX))
5179 		goto doit;
5180 
5181 	/*
5182 	 * bcm5700 Revision B silicon cannot handle DMA descriptors with
5183 	 * less than eight bytes.  If we encounter a teeny mbuf
5184 	 * at the end of a chain, we can pad.  Otherwise, copy.
5185 	 */
5186 	if (bge_compact_dma_runt(m_head) != 0)
5187 		return ENOBUFS;
5188 
5189 doit:
5190 	dma = SLIST_FIRST(&sc->txdma_list);
5191 	if (dma == NULL) {
5192 		return ENOBUFS;
5193 	}
5194 	dmamap = dma->dmamap;
5195 	dmatag = sc->bge_dmatag;
5196 	dma->is_dma32 = false;
5197 
5198 	/*
5199 	 * Set up any necessary TSO state before we start packing...
5200 	 */
5201 	use_tso = (m_head->m_pkthdr.csum_flags & M_CSUM_TSOv4) != 0;
5202 	if (!use_tso) {
5203 		maxsegsize = 0;
5204 	} else {	/* TSO setup */
5205 		unsigned  mss;
5206 		struct ether_header *eh;
5207 		unsigned ip_tcp_hlen, iptcp_opt_words, tcp_seg_flags, offset;
5208 		unsigned bge_hlen;
5209 		struct mbuf * m0 = m_head;
5210 		struct ip *ip;
5211 		struct tcphdr *th;
5212 		int iphl, hlen;
5213 
5214 		/*
5215 		 * XXX It would be nice if the mbuf pkthdr had offset
5216 		 * fields for the protocol headers.
5217 		 */
5218 
5219 		eh = mtod(m0, struct ether_header *);
5220 		switch (htons(eh->ether_type)) {
5221 		case ETHERTYPE_IP:
5222 			offset = ETHER_HDR_LEN;
5223 			break;
5224 
5225 		case ETHERTYPE_VLAN:
5226 			offset = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
5227 			break;
5228 
5229 		default:
5230 			/*
5231 			 * Don't support this protocol or encapsulation.
5232 			 */
5233 			return ENOBUFS;
5234 		}
5235 
5236 		/*
5237 		 * TCP/IP headers are in the first mbuf; we can do
5238 		 * this the easy way.
5239 		 */
5240 		iphl = M_CSUM_DATA_IPv4_IPHL(m0->m_pkthdr.csum_data);
5241 		hlen = iphl + offset;
5242 		if (__predict_false(m0->m_len <
5243 				    (hlen + sizeof(struct tcphdr)))) {
5244 
5245 			aprint_error_dev(sc->bge_dev,
5246 			    "TSO: hard case m0->m_len == %d < ip/tcp hlen %zd,"
5247 			    "not handled yet\n",
5248 			    m0->m_len, hlen+ sizeof(struct tcphdr));
5249 #ifdef NOTYET
5250 			/*
5251 			 * XXX jonathan@NetBSD.org: untested.
5252 			 * how to force this branch to be taken?
5253 			 */
5254 			BGE_EVCNT_INCR(sc->bge_ev_txtsopain);
5255 
5256 			m_copydata(m0, offset, sizeof(ip), &ip);
5257 			m_copydata(m0, hlen, sizeof(th), &th);
5258 
5259 			ip.ip_len = 0;
5260 
5261 			m_copyback(m0, hlen + offsetof(struct ip, ip_len),
5262 			    sizeof(ip.ip_len), &ip.ip_len);
5263 
5264 			th.th_sum = in_cksum_phdr(ip.ip_src.s_addr,
5265 			    ip.ip_dst.s_addr, htons(IPPROTO_TCP));
5266 
5267 			m_copyback(m0, hlen + offsetof(struct tcphdr, th_sum),
5268 			    sizeof(th.th_sum), &th.th_sum);
5269 
5270 			hlen += th.th_off << 2;
5271 			iptcp_opt_words	= hlen;
5272 #else
5273 			/*
5274 			 * if_wm "hard" case not yet supported, can we not
5275 			 * mandate it out of existence?
5276 			 */
5277 			(void) ip; (void)th; (void) ip_tcp_hlen;
5278 
5279 			return ENOBUFS;
5280 #endif
5281 		} else {
5282 			ip = (struct ip *) (mtod(m0, char *) + offset);
5283 			th = (struct tcphdr *) (mtod(m0, char *) + hlen);
5284 			ip_tcp_hlen = iphl +  (th->th_off << 2);
5285 
5286 			/* Total IP/TCP options, in 32-bit words */
5287 			iptcp_opt_words = (ip_tcp_hlen
5288 					   - sizeof(struct tcphdr)
5289 					   - sizeof(struct ip)) >> 2;
5290 		}
5291 		if (BGE_IS_575X_PLUS(sc)) {
5292 			th->th_sum = 0;
5293 			csum_flags = 0;
5294 		} else {
5295 			/*
5296 			 * XXX jonathan@NetBSD.org: 5705 untested.
5297 			 * Requires TSO firmware patch for 5701/5703/5704.
5298 			 */
5299 			th->th_sum = in_cksum_phdr(ip->ip_src.s_addr,
5300 			    ip->ip_dst.s_addr, htons(IPPROTO_TCP));
5301 		}
5302 
5303 		mss = m_head->m_pkthdr.segsz;
5304 		txbd_tso_flags |=
5305 		    BGE_TXBDFLAG_CPU_PRE_DMA |
5306 		    BGE_TXBDFLAG_CPU_POST_DMA;
5307 
5308 		/*
5309 		 * Our NIC TSO-assist assumes TSO has standard, optionless
5310 		 * IPv4 and TCP headers, which total 40 bytes. By default,
5311 		 * the NIC copies 40 bytes of IP/TCP header from the
5312 		 * supplied header into the IP/TCP header portion of
5313 		 * each post-TSO-segment. If the supplied packet has IP or
5314 		 * TCP options, we need to tell the NIC to copy those extra
5315 		 * bytes into each  post-TSO header, in addition to the normal
5316 		 * 40-byte IP/TCP header (and to leave space accordingly).
5317 		 * Unfortunately, the driver encoding of option length
5318 		 * varies across different ASIC families.
5319 		 */
5320 		tcp_seg_flags = 0;
5321 		bge_hlen = ip_tcp_hlen >> 2;
5322 		if (BGE_IS_5717_PLUS(sc)) {
5323 			tcp_seg_flags = (bge_hlen & 0x3) << 14;
5324 			txbd_tso_flags |=
5325 			    ((bge_hlen & 0xF8) << 7) | ((bge_hlen & 0x4) << 2);
5326 		} else if (BGE_IS_5705_PLUS(sc)) {
5327 			tcp_seg_flags = bge_hlen << 11;
5328 		} else {
5329 			/* XXX iptcp_opt_words or bge_hlen ? */
5330 			txbd_tso_flags |= iptcp_opt_words << 12;
5331 		}
5332 		maxsegsize = mss | tcp_seg_flags;
5333 		ip->ip_len = htons(mss + ip_tcp_hlen);
5334 		ip->ip_sum = 0;
5335 
5336 	}	/* TSO setup */
5337 
5338 	have_vtag = vlan_has_tag(m_head);
5339 	if (have_vtag)
5340 		vtag = vlan_get_tag(m_head);
5341 
5342 	/*
5343 	 * Start packing the mbufs in this chain into
5344 	 * the fragment pointers. Stop when we run out
5345 	 * of fragments or hit the end of the mbuf chain.
5346 	 */
5347 	remap = true;
5348 load_again:
5349 	error = bus_dmamap_load_mbuf(dmatag, dmamap, m_head, BUS_DMA_NOWAIT);
5350 	if (__predict_false(error)) {
5351 		if (error == EFBIG && remap) {
5352 			struct mbuf *m;
5353 			remap = false;
5354 			m = m_defrag(m_head, M_NOWAIT);
5355 			if (m != NULL) {
5356 				KASSERT(m == m_head);
5357 				goto load_again;
5358 			}
5359 		}
5360 		return error;
5361 	}
5362 	/*
5363 	 * Sanity check: avoid coming within 16 descriptors
5364 	 * of the end of the ring.
5365 	 */
5366 	if (dmamap->dm_nsegs > (BGE_TX_RING_CNT - sc->bge_txcnt - 16)) {
5367 		BGE_TSO_PRINTF(("%s: "
5368 		    " dmamap_load_mbuf too close to ring wrap\n",
5369 		    device_xname(sc->bge_dev)));
5370 		goto fail_unload;
5371 	}
5372 
5373 	/* Iterate over dmap-map fragments. */
5374 	f = prev_f = NULL;
5375 	cur = frag = *txidx;
5376 
5377 	for (i = 0; i < dmamap->dm_nsegs; i++) {
5378 		f = &sc->bge_rdata->bge_tx_ring[frag];
5379 		if (sc->bge_cdata.bge_tx_chain[frag] != NULL)
5380 			break;
5381 
5382 		BGE_HOSTADDR(f->bge_addr, dmamap->dm_segs[i].ds_addr);
5383 		f->bge_len = dmamap->dm_segs[i].ds_len;
5384 		if (sizeof(bus_addr_t) > 4 && dma->is_dma32 == false && use_tso && (
5385 		    (dmamap->dm_segs[i].ds_addr & 0xffffffff00000000) !=
5386 		    ((dmamap->dm_segs[i].ds_addr + f->bge_len) & 0xffffffff00000000) ||
5387 		    (prev_f != NULL &&
5388 		     prev_f->bge_addr.bge_addr_hi != f->bge_addr.bge_addr_hi))
5389 		   ) {
5390 			/*
5391 			 * watchdog timeout issue was observed with TSO,
5392 			 * limiting DMA address space to 32bits seems to
5393 			 * address the issue.
5394 			 */
5395 			bus_dmamap_unload(dmatag, dmamap);
5396 			dmatag = sc->bge_dmatag32;
5397 			dmamap = dma->dmamap32;
5398 			dma->is_dma32 = true;
5399 			remap = true;
5400 			goto load_again;
5401 		}
5402 
5403 		/*
5404 		 * For 5751 and follow-ons, for TSO we must turn
5405 		 * off checksum-assist flag in the tx-descr, and
5406 		 * supply the ASIC-revision-specific encoding
5407 		 * of TSO flags and segsize.
5408 		 */
5409 		if (use_tso) {
5410 			if (BGE_IS_575X_PLUS(sc) || i == 0) {
5411 				f->bge_rsvd = maxsegsize;
5412 				f->bge_flags = csum_flags | txbd_tso_flags;
5413 			} else {
5414 				f->bge_rsvd = 0;
5415 				f->bge_flags =
5416 				  (csum_flags | txbd_tso_flags) & 0x0fff;
5417 			}
5418 		} else {
5419 			f->bge_rsvd = 0;
5420 			f->bge_flags = csum_flags;
5421 		}
5422 
5423 		if (have_vtag) {
5424 			f->bge_flags |= BGE_TXBDFLAG_VLAN_TAG;
5425 			f->bge_vlan_tag = vtag;
5426 		} else {
5427 			f->bge_vlan_tag = 0;
5428 		}
5429 		prev_f = f;
5430 		cur = frag;
5431 		BGE_INC(frag, BGE_TX_RING_CNT);
5432 	}
5433 
5434 	if (i < dmamap->dm_nsegs) {
5435 		BGE_TSO_PRINTF(("%s: reached %d < dm_nsegs %d\n",
5436 		    device_xname(sc->bge_dev), i, dmamap->dm_nsegs));
5437 		goto fail_unload;
5438 	}
5439 
5440 	bus_dmamap_sync(dmatag, dmamap, 0, dmamap->dm_mapsize,
5441 	    BUS_DMASYNC_PREWRITE);
5442 
5443 	if (frag == sc->bge_tx_saved_considx) {
5444 		BGE_TSO_PRINTF(("%s: frag %d = wrapped id %d?\n",
5445 		    device_xname(sc->bge_dev), frag, sc->bge_tx_saved_considx));
5446 
5447 		goto fail_unload;
5448 	}
5449 
5450 	sc->bge_rdata->bge_tx_ring[cur].bge_flags |= BGE_TXBDFLAG_END;
5451 	sc->bge_cdata.bge_tx_chain[cur] = m_head;
5452 	SLIST_REMOVE_HEAD(&sc->txdma_list, link);
5453 	sc->txdma[cur] = dma;
5454 	sc->bge_txcnt += dmamap->dm_nsegs;
5455 
5456 	*txidx = frag;
5457 
5458 	return 0;
5459 
5460 fail_unload:
5461 	bus_dmamap_unload(dmatag, dmamap);
5462 
5463 	return ENOBUFS;
5464 }
5465 
5466 
5467 static void
5468 bge_start(struct ifnet *ifp)
5469 {
5470 	struct bge_softc * const sc = ifp->if_softc;
5471 
5472 	mutex_enter(sc->sc_intr_lock);
5473 	if (!sc->bge_txrx_stopping)
5474 		bge_start_locked(ifp);
5475 	mutex_exit(sc->sc_intr_lock);
5476 }
5477 
5478 /*
5479  * Main transmit routine. To avoid having to do mbuf copies, we put pointers
5480  * to the mbuf data regions directly in the transmit descriptors.
5481  */
5482 static void
5483 bge_start_locked(struct ifnet *ifp)
5484 {
5485 	struct bge_softc * const sc = ifp->if_softc;
5486 	struct mbuf *m_head = NULL;
5487 	struct mbuf *m;
5488 	uint32_t prodidx;
5489 	int pkts = 0;
5490 	int error;
5491 
5492 	KASSERT(mutex_owned(sc->sc_intr_lock));
5493 
5494 	prodidx = sc->bge_tx_prodidx;
5495 
5496 	while (sc->bge_cdata.bge_tx_chain[prodidx] == NULL) {
5497 		IFQ_POLL(&ifp->if_snd, m_head);
5498 		if (m_head == NULL)
5499 			break;
5500 
5501 #if 0
5502 		/*
5503 		 * XXX
5504 		 * safety overkill.  If this is a fragmented packet chain
5505 		 * with delayed TCP/UDP checksums, then only encapsulate
5506 		 * it if we have enough descriptors to handle the entire
5507 		 * chain at once.
5508 		 * (paranoia -- may not actually be needed)
5509 		 */
5510 		if (m_head->m_flags & M_FIRSTFRAG &&
5511 		    m_head->m_pkthdr.csum_flags & (CSUM_DELAY_DATA)) {
5512 			if ((BGE_TX_RING_CNT - sc->bge_txcnt) <
5513 			    M_CSUM_DATA_IPv4_OFFSET(m_head->m_pkthdr.csum_data) + 16) {
5514 				ifp->if_flags |= IFF_OACTIVE;
5515 				break;
5516 			}
5517 		}
5518 #endif
5519 
5520 		/*
5521 		 * Pack the data into the transmit ring. If we
5522 		 * don't have room, set the OACTIVE flag and wait
5523 		 * for the NIC to drain the ring.
5524 		 */
5525 		error = bge_encap(sc, m_head, &prodidx);
5526 		if (__predict_false(error)) {
5527 			if (SLIST_EMPTY(&sc->txdma_list)) {
5528 				/* just wait for the transmit ring to drain */
5529 				break;
5530 			}
5531 			IFQ_DEQUEUE(&ifp->if_snd, m);
5532 			KASSERT(m == m_head);
5533 			m_freem(m_head);
5534 			continue;
5535 		}
5536 
5537 		/* now we are committed to transmit the packet */
5538 		IFQ_DEQUEUE(&ifp->if_snd, m);
5539 		KASSERT(m == m_head);
5540 		pkts++;
5541 
5542 		/*
5543 		 * If there's a BPF listener, bounce a copy of this frame
5544 		 * to him.
5545 		 */
5546 		bpf_mtap(ifp, m_head, BPF_D_OUT);
5547 	}
5548 	if (pkts == 0)
5549 		return;
5550 
5551 	/* Transmit */
5552 	bge_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, prodidx);
5553 	/* 5700 b2 errata */
5554 	if (BGE_CHIPREV(sc->bge_chipid) == BGE_CHIPREV_5700_BX)
5555 		bge_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, prodidx);
5556 
5557 	sc->bge_tx_prodidx = prodidx;
5558 	sc->bge_tx_lastsent = time_uptime;
5559 	sc->bge_tx_sending = true;
5560 }
5561 
5562 static int
5563 bge_init(struct ifnet *ifp)
5564 {
5565 	struct bge_softc * const sc = ifp->if_softc;
5566 
5567 	KASSERT(IFNET_LOCKED(ifp));
5568 
5569 	if (sc->bge_detaching)
5570 		return ENXIO;
5571 
5572 	mutex_enter(sc->sc_core_lock);
5573 	int ret = bge_init_locked(ifp);
5574 	mutex_exit(sc->sc_core_lock);
5575 
5576 	return ret;
5577 }
5578 
5579 
5580 static int
5581 bge_init_locked(struct ifnet *ifp)
5582 {
5583 	struct bge_softc * const sc = ifp->if_softc;
5584 	const uint16_t *m;
5585 	uint32_t mode, reg;
5586 	int error = 0;
5587 
5588 	ASSERT_SLEEPABLE();
5589 	KASSERT(IFNET_LOCKED(ifp));
5590 	KASSERT(mutex_owned(sc->sc_core_lock));
5591 	KASSERT(ifp == &sc->ethercom.ec_if);
5592 
5593 	/* Cancel pending I/O and flush buffers. */
5594 	bge_stop_locked(ifp, false);
5595 
5596 	bge_stop_fw(sc);
5597 	bge_sig_pre_reset(sc, BGE_RESET_START);
5598 	bge_reset(sc);
5599 	bge_sig_legacy(sc, BGE_RESET_START);
5600 
5601 	if (BGE_CHIPREV(sc->bge_chipid) == BGE_CHIPREV_5784_AX) {
5602 		reg = CSR_READ_4(sc, BGE_CPMU_CTRL);
5603 		reg &= ~(BGE_CPMU_CTRL_LINK_AWARE_MODE |
5604 		    BGE_CPMU_CTRL_LINK_IDLE_MODE);
5605 		CSR_WRITE_4(sc, BGE_CPMU_CTRL, reg);
5606 
5607 		reg = CSR_READ_4(sc, BGE_CPMU_LSPD_10MB_CLK);
5608 		reg &= ~BGE_CPMU_LSPD_10MB_CLK;
5609 		reg |= BGE_CPMU_LSPD_10MB_MACCLK_6_25;
5610 		CSR_WRITE_4(sc, BGE_CPMU_LSPD_10MB_CLK, reg);
5611 
5612 		reg = CSR_READ_4(sc, BGE_CPMU_LNK_AWARE_PWRMD);
5613 		reg &= ~BGE_CPMU_LNK_AWARE_MACCLK_MASK;
5614 		reg |= BGE_CPMU_LNK_AWARE_MACCLK_6_25;
5615 		CSR_WRITE_4(sc, BGE_CPMU_LNK_AWARE_PWRMD, reg);
5616 
5617 		reg = CSR_READ_4(sc, BGE_CPMU_HST_ACC);
5618 		reg &= ~BGE_CPMU_HST_ACC_MACCLK_MASK;
5619 		reg |= BGE_CPMU_HST_ACC_MACCLK_6_25;
5620 		CSR_WRITE_4(sc, BGE_CPMU_HST_ACC, reg);
5621 	}
5622 
5623 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM57780) {
5624 		pcireg_t aercap;
5625 
5626 		reg = CSR_READ_4(sc, BGE_PCIE_PWRMNG_THRESH);
5627 		reg = (reg & ~BGE_PCIE_PWRMNG_L1THRESH_MASK)
5628 		    | BGE_PCIE_PWRMNG_L1THRESH_4MS
5629 		    | BGE_PCIE_PWRMNG_EXTASPMTMR_EN;
5630 		CSR_WRITE_4(sc, BGE_PCIE_PWRMNG_THRESH, reg);
5631 
5632 		reg = CSR_READ_4(sc, BGE_PCIE_EIDLE_DELAY);
5633 		reg = (reg & ~BGE_PCIE_EIDLE_DELAY_MASK)
5634 		    | BGE_PCIE_EIDLE_DELAY_13CLK;
5635 		CSR_WRITE_4(sc, BGE_PCIE_EIDLE_DELAY, reg);
5636 
5637 		/* Clear correctable error */
5638 		if (pci_get_ext_capability(sc->sc_pc, sc->sc_pcitag,
5639 		    PCI_EXTCAP_AER, &aercap, NULL) != 0)
5640 			pci_conf_write(sc->sc_pc, sc->sc_pcitag,
5641 			    aercap + PCI_AER_COR_STATUS, 0xffffffff);
5642 
5643 		reg = CSR_READ_4(sc, BGE_PCIE_LINKCTL);
5644 		reg = (reg & ~BGE_PCIE_LINKCTL_L1_PLL_PDEN)
5645 		    | BGE_PCIE_LINKCTL_L1_PLL_PDDIS;
5646 		CSR_WRITE_4(sc, BGE_PCIE_LINKCTL, reg);
5647 	}
5648 
5649 	bge_sig_post_reset(sc, BGE_RESET_START);
5650 
5651 	bge_chipinit(sc);
5652 
5653 	/*
5654 	 * Init the various state machines, ring
5655 	 * control blocks and firmware.
5656 	 */
5657 	error = bge_blockinit(sc);
5658 	if (error != 0) {
5659 		aprint_error_dev(sc->bge_dev, "initialization error %d\n",
5660 		    error);
5661 		return error;
5662 	}
5663 
5664 	/* 5718 step 25, 57XX step 54 */
5665 	/* Specify MTU. */
5666 	CSR_WRITE_4(sc, BGE_RX_MTU, ifp->if_mtu +
5667 	    ETHER_HDR_LEN + ETHER_CRC_LEN + ETHER_VLAN_ENCAP_LEN);
5668 
5669 	/* 5718 step 23 */
5670 	/* Load our MAC address. */
5671 	m = (const uint16_t *)&(CLLADDR(ifp->if_sadl)[0]);
5672 	CSR_WRITE_4(sc, BGE_MAC_ADDR1_LO, htons(m[0]));
5673 	CSR_WRITE_4(sc, BGE_MAC_ADDR1_HI,
5674 	    ((uint32_t)htons(m[1]) << 16) | htons(m[2]));
5675 
5676 	/* Enable or disable promiscuous mode as needed. */
5677 	if (ifp->if_flags & IFF_PROMISC)
5678 		BGE_SETBIT(sc, BGE_RX_MODE, BGE_RXMODE_RX_PROMISC);
5679 	else
5680 		BGE_CLRBIT(sc, BGE_RX_MODE, BGE_RXMODE_RX_PROMISC);
5681 
5682 	/* Program multicast filter. */
5683 	bge_setmulti(sc);
5684 
5685 	/* Init RX ring. */
5686 	bge_init_rx_ring_std(sc);
5687 
5688 	/*
5689 	 * Workaround for a bug in 5705 ASIC rev A0. Poll the NIC's
5690 	 * memory to insure that the chip has in fact read the first
5691 	 * entry of the ring.
5692 	 */
5693 	if (sc->bge_chipid == BGE_CHIPID_BCM5705_A0) {
5694 		u_int i;
5695 		for (i = 0; i < 10; i++) {
5696 			DELAY(20);
5697 			uint32_t v = bge_readmem_ind(sc, BGE_STD_RX_RINGS + 8);
5698 			if (v == (MCLBYTES - ETHER_ALIGN))
5699 				break;
5700 		}
5701 		if (i == 10)
5702 			aprint_error_dev(sc->bge_dev,
5703 			    "5705 A0 chip failed to load RX ring\n");
5704 	}
5705 
5706 	/* Init jumbo RX ring. */
5707 	if (ifp->if_mtu > (ETHERMTU + ETHER_HDR_LEN + ETHER_CRC_LEN))
5708 		bge_init_rx_ring_jumbo(sc);
5709 
5710 	/* Init our RX return ring index */
5711 	sc->bge_rx_saved_considx = 0;
5712 
5713 	/* Init TX ring. */
5714 	bge_init_tx_ring(sc);
5715 
5716 	/* 5718 step 63, 57XX step 94 */
5717 	/* Enable TX MAC state machine lockup fix. */
5718 	mode = CSR_READ_4(sc, BGE_TX_MODE);
5719 	if (BGE_IS_5755_PLUS(sc) ||
5720 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906)
5721 		mode |= BGE_TXMODE_MBUF_LOCKUP_FIX;
5722 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5720 ||
5723 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5762) {
5724 		mode &= ~(BGE_TXMODE_JMB_FRM_LEN | BGE_TXMODE_CNT_DN_MODE);
5725 		mode |= CSR_READ_4(sc, BGE_TX_MODE) &
5726 		    (BGE_TXMODE_JMB_FRM_LEN | BGE_TXMODE_CNT_DN_MODE);
5727 	}
5728 
5729 	/* Turn on transmitter */
5730 	CSR_WRITE_4_FLUSH(sc, BGE_TX_MODE, mode | BGE_TXMODE_ENABLE);
5731 	/* 5718 step 64 */
5732 	DELAY(100);
5733 
5734 	/* 5718 step 65, 57XX step 95 */
5735 	/* Turn on receiver */
5736 	mode = CSR_READ_4(sc, BGE_RX_MODE);
5737 	if (BGE_IS_5755_PLUS(sc))
5738 		mode |= BGE_RXMODE_IPV6_ENABLE;
5739 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5762)
5740 		mode |= BGE_RXMODE_IPV4_FRAG_FIX;
5741 	CSR_WRITE_4_FLUSH(sc, BGE_RX_MODE, mode | BGE_RXMODE_ENABLE);
5742 	/* 5718 step 66 */
5743 	DELAY(10);
5744 
5745 	/* 5718 step 12, 57XX step 37 */
5746 	/*
5747 	 * XXX Documents of 5718 series and 577xx say the recommended value
5748 	 * is 1, but tg3 set 1 only on 57765 series.
5749 	 */
5750 	if (BGE_IS_57765_PLUS(sc))
5751 		reg = 1;
5752 	else
5753 		reg = 2;
5754 	CSR_WRITE_4_FLUSH(sc, BGE_MAX_RX_FRAME_LOWAT, reg);
5755 
5756 	/* Tell firmware we're alive. */
5757 	BGE_SETBIT(sc, BGE_MODE_CTL, BGE_MODECTL_STACKUP);
5758 
5759 	/* Enable host interrupts. */
5760 	BGE_SETBIT(sc, BGE_PCI_MISC_CTL, BGE_PCIMISCCTL_CLEAR_INTA);
5761 	BGE_CLRBIT(sc, BGE_PCI_MISC_CTL, BGE_PCIMISCCTL_MASK_PCI_INTR);
5762 	bge_writembx_flush(sc, BGE_MBX_IRQ0_LO, 0);
5763 
5764 	mutex_enter(sc->sc_intr_lock);
5765 	if ((error = bge_ifmedia_upd(ifp)) == 0) {
5766 		sc->bge_stopping = false;
5767 		sc->bge_txrx_stopping = false;
5768 
5769 		/* IFNET_LOCKED asserted above */
5770 		ifp->if_flags |= IFF_RUNNING;
5771 
5772 		callout_schedule(&sc->bge_timeout, hz);
5773 	}
5774 	mutex_exit(sc->sc_intr_lock);
5775 
5776 	sc->bge_if_flags = ifp->if_flags;
5777 
5778 	return error;
5779 }
5780 
5781 /*
5782  * Set media options.
5783  */
5784 static int
5785 bge_ifmedia_upd(struct ifnet *ifp)
5786 {
5787 	struct bge_softc * const sc = ifp->if_softc;
5788 	struct mii_data * const mii = &sc->bge_mii;
5789 	struct ifmedia * const ifm = &sc->bge_ifmedia;
5790 	int rc;
5791 
5792 	KASSERT(mutex_owned(sc->sc_intr_lock));
5793 
5794 	/* If this is a 1000baseX NIC, enable the TBI port. */
5795 	if (sc->bge_flags & BGEF_FIBER_TBI) {
5796 		if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER)
5797 			return EINVAL;
5798 		switch (IFM_SUBTYPE(ifm->ifm_media)) {
5799 		case IFM_AUTO:
5800 			/*
5801 			 * The BCM5704 ASIC appears to have a special
5802 			 * mechanism for programming the autoneg
5803 			 * advertisement registers in TBI mode.
5804 			 */
5805 			if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5704) {
5806 				uint32_t sgdig;
5807 				sgdig = CSR_READ_4(sc, BGE_SGDIG_STS);
5808 				if (sgdig & BGE_SGDIGSTS_DONE) {
5809 					CSR_WRITE_4(sc, BGE_TX_TBI_AUTONEG, 0);
5810 					sgdig = CSR_READ_4(sc, BGE_SGDIG_CFG);
5811 					sgdig |= BGE_SGDIGCFG_AUTO |
5812 					    BGE_SGDIGCFG_PAUSE_CAP |
5813 					    BGE_SGDIGCFG_ASYM_PAUSE;
5814 					CSR_WRITE_4_FLUSH(sc, BGE_SGDIG_CFG,
5815 					    sgdig | BGE_SGDIGCFG_SEND);
5816 					DELAY(5);
5817 					CSR_WRITE_4_FLUSH(sc, BGE_SGDIG_CFG,
5818 					    sgdig);
5819 				}
5820 			}
5821 			break;
5822 		case IFM_1000_SX:
5823 			if ((ifm->ifm_media & IFM_FDX) != 0) {
5824 				BGE_CLRBIT_FLUSH(sc, BGE_MAC_MODE,
5825 				    BGE_MACMODE_HALF_DUPLEX);
5826 			} else {
5827 				BGE_SETBIT_FLUSH(sc, BGE_MAC_MODE,
5828 				    BGE_MACMODE_HALF_DUPLEX);
5829 			}
5830 			DELAY(40);
5831 			break;
5832 		default:
5833 			return EINVAL;
5834 		}
5835 		/* XXX 802.3x flow control for 1000BASE-SX */
5836 		return 0;
5837 	}
5838 
5839 	if ((BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5784) &&
5840 	    (BGE_CHIPREV(sc->bge_chipid) != BGE_CHIPREV_5784_AX)) {
5841 		uint32_t reg;
5842 
5843 		reg = CSR_READ_4(sc, BGE_CPMU_CTRL);
5844 		if ((reg & BGE_CPMU_CTRL_GPHY_10MB_RXONLY) != 0) {
5845 			reg &= ~BGE_CPMU_CTRL_GPHY_10MB_RXONLY;
5846 			CSR_WRITE_4(sc, BGE_CPMU_CTRL, reg);
5847 		}
5848 	}
5849 
5850 	BGE_STS_SETBIT(sc, BGE_STS_LINK_EVT);
5851 	if ((rc = mii_mediachg(mii)) == ENXIO)
5852 		return 0;
5853 
5854 	if (BGE_CHIPREV(sc->bge_chipid) == BGE_CHIPREV_5784_AX) {
5855 		uint32_t reg;
5856 
5857 		reg = CSR_READ_4(sc, BGE_CPMU_LSPD_1000MB_CLK);
5858 		if ((reg & BGE_CPMU_LSPD_1000MB_MACCLK_MASK)
5859 		    == (BGE_CPMU_LSPD_1000MB_MACCLK_12_5)) {
5860 			reg &= ~BGE_CPMU_LSPD_1000MB_MACCLK_MASK;
5861 			delay(40);
5862 			CSR_WRITE_4(sc, BGE_CPMU_LSPD_1000MB_CLK, reg);
5863 		}
5864 	}
5865 
5866 	/*
5867 	 * Force an interrupt so that we will call bge_link_upd
5868 	 * if needed and clear any pending link state attention.
5869 	 * Without this we are not getting any further interrupts
5870 	 * for link state changes and thus will not UP the link and
5871 	 * not be able to send in bge_start. The only way to get
5872 	 * things working was to receive a packet and get a RX intr.
5873 	 */
5874 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5700 ||
5875 	    sc->bge_flags & BGEF_IS_5788)
5876 		BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_INTR_SET);
5877 	else
5878 		BGE_SETBIT(sc, BGE_HCC_MODE, BGE_HCCMODE_COAL_NOW);
5879 
5880 	return rc;
5881 }
5882 
5883 /*
5884  * Report current media status.
5885  */
5886 static void
5887 bge_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
5888 {
5889 	struct bge_softc * const sc = ifp->if_softc;
5890 	struct mii_data * const mii = &sc->bge_mii;
5891 
5892 	KASSERT(mutex_owned(sc->sc_intr_lock));
5893 
5894 	if (sc->bge_flags & BGEF_FIBER_TBI) {
5895 		ifmr->ifm_status = IFM_AVALID;
5896 		ifmr->ifm_active = IFM_ETHER;
5897 		if (CSR_READ_4(sc, BGE_MAC_STS) &
5898 		    BGE_MACSTAT_TBI_PCS_SYNCHED)
5899 			ifmr->ifm_status |= IFM_ACTIVE;
5900 		ifmr->ifm_active |= IFM_1000_SX;
5901 		if (CSR_READ_4(sc, BGE_MAC_MODE) & BGE_MACMODE_HALF_DUPLEX)
5902 			ifmr->ifm_active |= IFM_HDX;
5903 		else
5904 			ifmr->ifm_active |= IFM_FDX;
5905 		return;
5906 	}
5907 
5908 	mii_pollstat(mii);
5909 	ifmr->ifm_status = mii->mii_media_status;
5910 	ifmr->ifm_active = (mii->mii_media_active & ~IFM_ETH_FMASK) |
5911 	    sc->bge_flowflags;
5912 }
5913 
5914 static int
5915 bge_ifflags_cb(struct ethercom *ec)
5916 {
5917 	struct ifnet * const ifp = &ec->ec_if;
5918 	struct bge_softc * const sc = ifp->if_softc;
5919 	int ret = 0;
5920 
5921 	KASSERT(IFNET_LOCKED(ifp));
5922 	mutex_enter(sc->sc_core_lock);
5923 
5924 	u_short change = ifp->if_flags ^ sc->bge_if_flags;
5925 
5926 	if ((change & ~(IFF_CANTCHANGE | IFF_DEBUG)) != 0) {
5927 		ret = ENETRESET;
5928 	} else if ((change & (IFF_PROMISC | IFF_ALLMULTI)) != 0) {
5929 		if ((ifp->if_flags & IFF_PROMISC) == 0)
5930 			BGE_CLRBIT(sc, BGE_RX_MODE, BGE_RXMODE_RX_PROMISC);
5931 		else
5932 			BGE_SETBIT(sc, BGE_RX_MODE, BGE_RXMODE_RX_PROMISC);
5933 
5934 		bge_setmulti(sc);
5935 	}
5936 
5937 	sc->bge_if_flags = ifp->if_flags;
5938 	mutex_exit(sc->sc_core_lock);
5939 
5940 	return ret;
5941 }
5942 
5943 static int
5944 bge_ioctl(struct ifnet *ifp, u_long command, void *data)
5945 {
5946 	struct bge_softc * const sc = ifp->if_softc;
5947 	struct ifreq * const ifr = (struct ifreq *) data;
5948 	int error = 0;
5949 
5950 	switch (command) {
5951 	case SIOCADDMULTI:
5952 	case SIOCDELMULTI:
5953 		break;
5954 	default:
5955 		KASSERT(IFNET_LOCKED(ifp));
5956 	}
5957 
5958 	const int s = splnet();
5959 
5960 	switch (command) {
5961 	case SIOCSIFMEDIA:
5962 		mutex_enter(sc->sc_core_lock);
5963 		/* XXX Flow control is not supported for 1000BASE-SX */
5964 		if (sc->bge_flags & BGEF_FIBER_TBI) {
5965 			ifr->ifr_media &= ~IFM_ETH_FMASK;
5966 			sc->bge_flowflags = 0;
5967 		}
5968 
5969 		/* Flow control requires full-duplex mode. */
5970 		if (IFM_SUBTYPE(ifr->ifr_media) == IFM_AUTO ||
5971 		    (ifr->ifr_media & IFM_FDX) == 0) {
5972 			ifr->ifr_media &= ~IFM_ETH_FMASK;
5973 		}
5974 		if (IFM_SUBTYPE(ifr->ifr_media) != IFM_AUTO) {
5975 			if ((ifr->ifr_media & IFM_ETH_FMASK) == IFM_FLOW) {
5976 				/* We can do both TXPAUSE and RXPAUSE. */
5977 				ifr->ifr_media |=
5978 				    IFM_ETH_TXPAUSE | IFM_ETH_RXPAUSE;
5979 			}
5980 			sc->bge_flowflags = ifr->ifr_media & IFM_ETH_FMASK;
5981 		}
5982 		mutex_exit(sc->sc_core_lock);
5983 
5984 		if (sc->bge_flags & BGEF_FIBER_TBI) {
5985 			error = ifmedia_ioctl(ifp, ifr, &sc->bge_ifmedia,
5986 			    command);
5987 		} else {
5988 			struct mii_data * const mii = &sc->bge_mii;
5989 			error = ifmedia_ioctl(ifp, ifr, &mii->mii_media,
5990 			    command);
5991 		}
5992 		break;
5993 	default:
5994 		if ((error = ether_ioctl(ifp, command, data)) != ENETRESET)
5995 			break;
5996 
5997 		error = 0;
5998 
5999 		if (command == SIOCADDMULTI || command == SIOCDELMULTI) {
6000 			mutex_enter(sc->sc_core_lock);
6001 			if (sc->bge_if_flags & IFF_RUNNING) {
6002 				bge_setmulti(sc);
6003 			}
6004 			mutex_exit(sc->sc_core_lock);
6005 		}
6006 		break;
6007 	}
6008 
6009 	splx(s);
6010 
6011 	return error;
6012 }
6013 
6014 static bool
6015 bge_watchdog_check(struct bge_softc * const sc)
6016 {
6017 
6018 	KASSERT(mutex_owned(sc->sc_core_lock));
6019 
6020 	if (!sc->bge_tx_sending)
6021 		return true;
6022 
6023 	if (time_uptime - sc->bge_tx_lastsent <= bge_watchdog_timeout)
6024 		return true;
6025 
6026 	/* If pause frames are active then don't reset the hardware. */
6027 	if ((CSR_READ_4(sc, BGE_RX_MODE) & BGE_RXMODE_FLOWCTL_ENABLE) != 0) {
6028 		const uint32_t status = CSR_READ_4(sc, BGE_RX_STS);
6029 		if ((status & BGE_RXSTAT_REMOTE_XOFFED) != 0) {
6030 			/*
6031 			 * If link partner has us in XOFF state then wait for
6032 			 * the condition to clear.
6033 			 */
6034 			CSR_WRITE_4(sc, BGE_RX_STS, status);
6035 			sc->bge_tx_lastsent = time_uptime;
6036 			return true;
6037 		} else if ((status & BGE_RXSTAT_RCVD_XOFF) != 0 &&
6038 		    (status & BGE_RXSTAT_RCVD_XON) != 0) {
6039 			/*
6040 			 * If link partner has us in XOFF state then wait for
6041 			 * the condition to clear.
6042 			 */
6043 			CSR_WRITE_4(sc, BGE_RX_STS, status);
6044 			sc->bge_tx_lastsent = time_uptime;
6045 			return true;
6046 		}
6047 		/*
6048 		 * Any other condition is unexpected and the controller
6049 		 * should be reset.
6050 		 */
6051 	}
6052 
6053 	return false;
6054 }
6055 
6056 static bool
6057 bge_watchdog_tick(struct ifnet *ifp)
6058 {
6059 	struct bge_softc * const sc = ifp->if_softc;
6060 
6061 	KASSERT(mutex_owned(sc->sc_core_lock));
6062 
6063 	if (!sc->sc_trigger_reset && bge_watchdog_check(sc))
6064 		return true;
6065 
6066 	if (atomic_swap_uint(&sc->sc_reset_pending, 1) == 0)
6067 		workqueue_enqueue(sc->sc_reset_wq, &sc->sc_reset_work, NULL);
6068 
6069 	return false;
6070 }
6071 
6072 /*
6073  * Perform an interface watchdog reset.
6074  */
6075 static void
6076 bge_handle_reset_work(struct work *work, void *arg)
6077 {
6078 	struct bge_softc * const sc = arg;
6079 	struct ifnet * const ifp = &sc->ethercom.ec_if;
6080 
6081 	printf("%s: watchdog timeout -- resetting\n", ifp->if_xname);
6082 
6083 	/* Don't want ioctl operations to happen */
6084 	IFNET_LOCK(ifp);
6085 
6086 	/* reset the interface. */
6087 	bge_init(ifp);
6088 
6089 	IFNET_UNLOCK(ifp);
6090 
6091 	/*
6092 	 * There are still some upper layer processing which call
6093 	 * ifp->if_start(). e.g. ALTQ or one CPU system
6094 	 */
6095 	/* Try to get more packets going. */
6096 	ifp->if_start(ifp);
6097 
6098 	atomic_store_relaxed(&sc->sc_reset_pending, 0);
6099 }
6100 
6101 static void
6102 bge_stop_block(struct bge_softc *sc, bus_addr_t reg, uint32_t bit)
6103 {
6104 	int i;
6105 
6106 	BGE_CLRBIT_FLUSH(sc, reg, bit);
6107 
6108 	for (i = 0; i < 1000; i++) {
6109 		delay(100);
6110 		if ((CSR_READ_4(sc, reg) & bit) == 0)
6111 			return;
6112 	}
6113 
6114 	/*
6115 	 * Doesn't print only when the register is BGE_SRS_MODE. It occurs
6116 	 * on some environment (and once after boot?)
6117 	 */
6118 	if (reg != BGE_SRS_MODE)
6119 		aprint_error_dev(sc->bge_dev,
6120 		    "block failed to stop: reg 0x%lx, bit 0x%08x\n",
6121 		    (u_long)reg, bit);
6122 }
6123 
6124 
6125 static void
6126 bge_stop(struct ifnet *ifp, int disable)
6127 {
6128 	struct bge_softc * const sc = ifp->if_softc;
6129 
6130 	ASSERT_SLEEPABLE();
6131 	KASSERT(IFNET_LOCKED(ifp));
6132 
6133 	mutex_enter(sc->sc_core_lock);
6134 	bge_stop_locked(ifp, disable ? true : false);
6135 	mutex_exit(sc->sc_core_lock);
6136 }
6137 
6138 /*
6139  * Stop the adapter and free any mbufs allocated to the
6140  * RX and TX lists.
6141  */
6142 static void
6143 bge_stop_locked(struct ifnet *ifp, bool disable)
6144 {
6145 	struct bge_softc * const sc = ifp->if_softc;
6146 
6147 	ASSERT_SLEEPABLE();
6148 	KASSERT(IFNET_LOCKED(ifp));
6149 	KASSERT(mutex_owned(sc->sc_core_lock));
6150 
6151 	sc->bge_stopping = true;
6152 
6153 	mutex_enter(sc->sc_intr_lock);
6154 	sc->bge_txrx_stopping = true;
6155 	mutex_exit(sc->sc_intr_lock);
6156 
6157 	callout_halt(&sc->bge_timeout, sc->sc_core_lock);
6158 
6159 	/* Disable host interrupts. */
6160 	BGE_SETBIT(sc, BGE_PCI_MISC_CTL, BGE_PCIMISCCTL_MASK_PCI_INTR);
6161 	bge_writembx_flush(sc, BGE_MBX_IRQ0_LO, 1);
6162 
6163 	/*
6164 	 * Tell firmware we're shutting down.
6165 	 */
6166 	bge_stop_fw(sc);
6167 	bge_sig_pre_reset(sc, BGE_RESET_SHUTDOWN);
6168 
6169 	/*
6170 	 * Disable all of the receiver blocks.
6171 	 */
6172 	bge_stop_block(sc, BGE_RX_MODE, BGE_RXMODE_ENABLE);
6173 	bge_stop_block(sc, BGE_RBDI_MODE, BGE_RBDIMODE_ENABLE);
6174 	bge_stop_block(sc, BGE_RXLP_MODE, BGE_RXLPMODE_ENABLE);
6175 	if (BGE_IS_5700_FAMILY(sc))
6176 		bge_stop_block(sc, BGE_RXLS_MODE, BGE_RXLSMODE_ENABLE);
6177 	bge_stop_block(sc, BGE_RDBDI_MODE, BGE_RBDIMODE_ENABLE);
6178 	bge_stop_block(sc, BGE_RDC_MODE, BGE_RDCMODE_ENABLE);
6179 	bge_stop_block(sc, BGE_RBDC_MODE, BGE_RBDCMODE_ENABLE);
6180 
6181 	/*
6182 	 * Disable all of the transmit blocks.
6183 	 */
6184 	bge_stop_block(sc, BGE_SRS_MODE, BGE_SRSMODE_ENABLE);
6185 	bge_stop_block(sc, BGE_SBDI_MODE, BGE_SBDIMODE_ENABLE);
6186 	bge_stop_block(sc, BGE_SDI_MODE, BGE_SDIMODE_ENABLE);
6187 	bge_stop_block(sc, BGE_RDMA_MODE, BGE_RDMAMODE_ENABLE);
6188 	bge_stop_block(sc, BGE_SDC_MODE, BGE_SDCMODE_ENABLE);
6189 	if (BGE_IS_5700_FAMILY(sc))
6190 		bge_stop_block(sc, BGE_DMAC_MODE, BGE_DMACMODE_ENABLE);
6191 	bge_stop_block(sc, BGE_SBDC_MODE, BGE_SBDCMODE_ENABLE);
6192 
6193 	BGE_CLRBIT_FLUSH(sc, BGE_MAC_MODE, BGE_MACMODE_TXDMA_ENB);
6194 	delay(40);
6195 
6196 	bge_stop_block(sc, BGE_TX_MODE, BGE_TXMODE_ENABLE);
6197 
6198 	/*
6199 	 * Shut down all of the memory managers and related
6200 	 * state machines.
6201 	 */
6202 	/* 5718 step 5a,5b */
6203 	bge_stop_block(sc, BGE_HCC_MODE, BGE_HCCMODE_ENABLE);
6204 	bge_stop_block(sc, BGE_WDMA_MODE, BGE_WDMAMODE_ENABLE);
6205 	if (BGE_IS_5700_FAMILY(sc))
6206 		bge_stop_block(sc, BGE_MBCF_MODE, BGE_MBCFMODE_ENABLE);
6207 
6208 	/* 5718 step 5c,5d */
6209 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0xFFFFFFFF);
6210 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0);
6211 
6212 	if (BGE_IS_5700_FAMILY(sc)) {
6213 		bge_stop_block(sc, BGE_BMAN_MODE, BGE_BMANMODE_ENABLE);
6214 		bge_stop_block(sc, BGE_MARB_MODE, BGE_MARBMODE_ENABLE);
6215 	}
6216 
6217 	bge_reset(sc);
6218 	bge_sig_legacy(sc, BGE_RESET_SHUTDOWN);
6219 	bge_sig_post_reset(sc, BGE_RESET_SHUTDOWN);
6220 
6221 	/*
6222 	 * Keep the ASF firmware running if up.
6223 	 */
6224 	if (sc->bge_asf_mode & ASF_STACKUP)
6225 		BGE_SETBIT(sc, BGE_MODE_CTL, BGE_MODECTL_STACKUP);
6226 	else
6227 		BGE_CLRBIT(sc, BGE_MODE_CTL, BGE_MODECTL_STACKUP);
6228 
6229 	/* Free the RX lists. */
6230 	bge_free_rx_ring_std(sc);
6231 
6232 	/* Free jumbo RX list. */
6233 	if (BGE_IS_JUMBO_CAPABLE(sc))
6234 		bge_free_rx_ring_jumbo(sc);
6235 
6236 	/* Free TX buffers. */
6237 	bge_free_tx_ring(sc, disable);
6238 
6239 	/*
6240 	 * Isolate/power down the PHY.
6241 	 */
6242 	if (!(sc->bge_flags & BGEF_FIBER_TBI)) {
6243 		mutex_enter(sc->sc_intr_lock);
6244 		mii_down(&sc->bge_mii);
6245 		mutex_exit(sc->sc_intr_lock);
6246 	}
6247 
6248 	sc->bge_tx_saved_considx = BGE_TXCONS_UNSET;
6249 
6250 	/* Clear MAC's link state (PHY may still have link UP). */
6251 	BGE_STS_CLRBIT(sc, BGE_STS_LINK);
6252 
6253 	ifp->if_flags &= ~IFF_RUNNING;
6254 
6255 	sc->bge_if_flags = ifp->if_flags;
6256 }
6257 
6258 static void
6259 bge_link_upd(struct bge_softc *sc)
6260 {
6261 	struct ifnet * const ifp = &sc->ethercom.ec_if;
6262 	struct mii_data * const mii = &sc->bge_mii;
6263 	uint32_t status;
6264 	uint16_t phyval;
6265 	int link;
6266 
6267 	KASSERT(sc->sc_intr_lock);
6268 
6269 	/* Clear 'pending link event' flag */
6270 	BGE_STS_CLRBIT(sc, BGE_STS_LINK_EVT);
6271 
6272 	/*
6273 	 * Process link state changes.
6274 	 * Grrr. The link status word in the status block does
6275 	 * not work correctly on the BCM5700 rev AX and BX chips,
6276 	 * according to all available information. Hence, we have
6277 	 * to enable MII interrupts in order to properly obtain
6278 	 * async link changes. Unfortunately, this also means that
6279 	 * we have to read the MAC status register to detect link
6280 	 * changes, thereby adding an additional register access to
6281 	 * the interrupt handler.
6282 	 */
6283 
6284 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5700) {
6285 		status = CSR_READ_4(sc, BGE_MAC_STS);
6286 		if (status & BGE_MACSTAT_MI_INTERRUPT) {
6287 			mii_pollstat(mii);
6288 
6289 			if (!BGE_STS_BIT(sc, BGE_STS_LINK) &&
6290 			    mii->mii_media_status & IFM_ACTIVE &&
6291 			    IFM_SUBTYPE(mii->mii_media_active) != IFM_NONE)
6292 				BGE_STS_SETBIT(sc, BGE_STS_LINK);
6293 			else if (BGE_STS_BIT(sc, BGE_STS_LINK) &&
6294 			    (!(mii->mii_media_status & IFM_ACTIVE) ||
6295 			    IFM_SUBTYPE(mii->mii_media_active) == IFM_NONE))
6296 				BGE_STS_CLRBIT(sc, BGE_STS_LINK);
6297 
6298 			/* Clear the interrupt */
6299 			CSR_WRITE_4(sc, BGE_MAC_EVT_ENB,
6300 			    BGE_EVTENB_MI_INTERRUPT);
6301 			bge_miibus_readreg(sc->bge_dev, sc->bge_phy_addr,
6302 			    BRGPHY_MII_ISR, &phyval);
6303 			bge_miibus_writereg(sc->bge_dev, sc->bge_phy_addr,
6304 			    BRGPHY_MII_IMR, BRGPHY_INTRS);
6305 		}
6306 		return;
6307 	}
6308 
6309 	if (sc->bge_flags & BGEF_FIBER_TBI) {
6310 		status = CSR_READ_4(sc, BGE_MAC_STS);
6311 		if (status & BGE_MACSTAT_TBI_PCS_SYNCHED) {
6312 			if (!BGE_STS_BIT(sc, BGE_STS_LINK)) {
6313 				BGE_STS_SETBIT(sc, BGE_STS_LINK);
6314 				if (BGE_ASICREV(sc->bge_chipid)
6315 				    == BGE_ASICREV_BCM5704) {
6316 					BGE_CLRBIT_FLUSH(sc, BGE_MAC_MODE,
6317 					    BGE_MACMODE_TBI_SEND_CFGS);
6318 					DELAY(40);
6319 				}
6320 				CSR_WRITE_4(sc, BGE_MAC_STS, 0xFFFFFFFF);
6321 				if_link_state_change(ifp, LINK_STATE_UP);
6322 			}
6323 		} else if (BGE_STS_BIT(sc, BGE_STS_LINK)) {
6324 			BGE_STS_CLRBIT(sc, BGE_STS_LINK);
6325 			if_link_state_change(ifp, LINK_STATE_DOWN);
6326 		}
6327 	} else if (BGE_STS_BIT(sc, BGE_STS_AUTOPOLL)) {
6328 		/*
6329 		 * Some broken BCM chips have BGE_STATFLAG_LINKSTATE_CHANGED
6330 		 * bit in status word always set. Workaround this bug by
6331 		 * reading PHY link status directly.
6332 		 */
6333 		link = (CSR_READ_4(sc, BGE_MI_STS) & BGE_MISTS_LINK)?
6334 		    BGE_STS_LINK : 0;
6335 
6336 		if (BGE_STS_BIT(sc, BGE_STS_LINK) != link) {
6337 			mii_pollstat(mii);
6338 
6339 			if (!BGE_STS_BIT(sc, BGE_STS_LINK) &&
6340 			    mii->mii_media_status & IFM_ACTIVE &&
6341 			    IFM_SUBTYPE(mii->mii_media_active) != IFM_NONE)
6342 				BGE_STS_SETBIT(sc, BGE_STS_LINK);
6343 			else if (BGE_STS_BIT(sc, BGE_STS_LINK) &&
6344 			    (!(mii->mii_media_status & IFM_ACTIVE) ||
6345 			    IFM_SUBTYPE(mii->mii_media_active) == IFM_NONE))
6346 				BGE_STS_CLRBIT(sc, BGE_STS_LINK);
6347 		}
6348 	} else {
6349 		/*
6350 		 * For controllers that call mii_tick, we have to poll
6351 		 * link status.
6352 		 */
6353 		mii_pollstat(mii);
6354 	}
6355 
6356 	if (BGE_CHIPREV(sc->bge_chipid) == BGE_CHIPREV_5784_AX) {
6357 		uint32_t reg, scale;
6358 
6359 		reg = CSR_READ_4(sc, BGE_CPMU_CLCK_STAT) &
6360 		    BGE_CPMU_CLCK_STAT_MAC_CLCK_MASK;
6361 		if (reg == BGE_CPMU_CLCK_STAT_MAC_CLCK_62_5)
6362 			scale = 65;
6363 		else if (reg == BGE_CPMU_CLCK_STAT_MAC_CLCK_6_25)
6364 			scale = 6;
6365 		else
6366 			scale = 12;
6367 
6368 		reg = CSR_READ_4(sc, BGE_MISC_CFG) &
6369 		    ~BGE_MISCCFG_TIMER_PRESCALER;
6370 		reg |= scale << 1;
6371 		CSR_WRITE_4(sc, BGE_MISC_CFG, reg);
6372 	}
6373 	/* Clear the attention */
6374 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED |
6375 	    BGE_MACSTAT_CFG_CHANGED | BGE_MACSTAT_MI_COMPLETE |
6376 	    BGE_MACSTAT_LINK_CHANGED);
6377 }
6378 
6379 static int
6380 bge_sysctl_verify(SYSCTLFN_ARGS)
6381 {
6382 	int error, t;
6383 	struct sysctlnode node;
6384 
6385 	node = *rnode;
6386 	t = *(int*)rnode->sysctl_data;
6387 	node.sysctl_data = &t;
6388 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
6389 	if (error || newp == NULL)
6390 		return error;
6391 
6392 #if 0
6393 	DPRINTF2(("%s: t = %d, nodenum = %d, rnodenum = %d\n", __func__, t,
6394 	    node.sysctl_num, rnode->sysctl_num));
6395 #endif
6396 
6397 	if (node.sysctl_num == bge_rxthresh_nodenum) {
6398 		if (t < 0 || t >= NBGE_RX_THRESH)
6399 			return EINVAL;
6400 		bge_update_all_threshes(t);
6401 	} else
6402 		return EINVAL;
6403 
6404 	*(int*)rnode->sysctl_data = t;
6405 
6406 	return 0;
6407 }
6408 
6409 /*
6410  * Set up sysctl(3) MIB, hw.bge.*.
6411  */
6412 static void
6413 bge_sysctl_init(struct bge_softc *sc)
6414 {
6415 	int rc, bge_root_num;
6416 	const struct sysctlnode *node;
6417 
6418 	if ((rc = sysctl_createv(&sc->bge_log, 0, NULL, &node,
6419 	    0, CTLTYPE_NODE, "bge",
6420 	    SYSCTL_DESCR("BGE interface controls"),
6421 	    NULL, 0, NULL, 0, CTL_HW, CTL_CREATE, CTL_EOL)) != 0) {
6422 		goto out;
6423 	}
6424 
6425 	bge_root_num = node->sysctl_num;
6426 
6427 	/* BGE Rx interrupt mitigation level */
6428 	if ((rc = sysctl_createv(&sc->bge_log, 0, NULL, &node,
6429 	    CTLFLAG_READWRITE,
6430 	    CTLTYPE_INT, "rx_lvl",
6431 	    SYSCTL_DESCR("BGE receive interrupt mitigation level"),
6432 	    bge_sysctl_verify, 0,
6433 	    &bge_rx_thresh_lvl,
6434 	    0, CTL_HW, bge_root_num, CTL_CREATE,
6435 	    CTL_EOL)) != 0) {
6436 		goto out;
6437 	}
6438 
6439 	bge_rxthresh_nodenum = node->sysctl_num;
6440 
6441 #ifdef BGE_DEBUG
6442 	if ((rc = sysctl_createv(&sc->bge_log, 0, NULL, &node,
6443 	    CTLFLAG_READWRITE,
6444 	    CTLTYPE_BOOL, "trigger_reset",
6445 	    SYSCTL_DESCR("Trigger an interface reset"),
6446 	    NULL, 0, &sc->sc_trigger_reset, 0, CTL_CREATE,
6447 	    CTL_EOL)) != 0) {
6448 		goto out;
6449 	}
6450 #endif
6451 	return;
6452 
6453 out:
6454 	aprint_error("%s: sysctl_createv failed (rc = %d)\n", __func__, rc);
6455 }
6456 
6457 #ifdef BGE_DEBUG
6458 void
6459 bge_debug_info(struct bge_softc *sc)
6460 {
6461 
6462 	printf("Hardware Flags:\n");
6463 	if (BGE_IS_57765_PLUS(sc))
6464 		printf(" - 57765 Plus\n");
6465 	if (BGE_IS_5717_PLUS(sc))
6466 		printf(" - 5717 Plus\n");
6467 	if (BGE_IS_5755_PLUS(sc))
6468 		printf(" - 5755 Plus\n");
6469 	if (BGE_IS_575X_PLUS(sc))
6470 		printf(" - 575X Plus\n");
6471 	if (BGE_IS_5705_PLUS(sc))
6472 		printf(" - 5705 Plus\n");
6473 	if (BGE_IS_5714_FAMILY(sc))
6474 		printf(" - 5714 Family\n");
6475 	if (BGE_IS_5700_FAMILY(sc))
6476 		printf(" - 5700 Family\n");
6477 	if (sc->bge_flags & BGEF_IS_5788)
6478 		printf(" - 5788\n");
6479 	if (sc->bge_flags & BGEF_JUMBO_CAPABLE)
6480 		printf(" - Supports Jumbo Frames\n");
6481 	if (sc->bge_flags & BGEF_NO_EEPROM)
6482 		printf(" - No EEPROM\n");
6483 	if (sc->bge_flags & BGEF_PCIX)
6484 		printf(" - PCI-X Bus\n");
6485 	if (sc->bge_flags & BGEF_PCIE)
6486 		printf(" - PCI Express Bus\n");
6487 	if (sc->bge_flags & BGEF_RX_ALIGNBUG)
6488 		printf(" - RX Alignment Bug\n");
6489 	if (sc->bge_flags & BGEF_APE)
6490 		printf(" - APE\n");
6491 	if (sc->bge_flags & BGEF_CPMU_PRESENT)
6492 		printf(" - CPMU\n");
6493 	if (sc->bge_flags & BGEF_TSO)
6494 		printf(" - TSO\n");
6495 	if (sc->bge_flags & BGEF_TAGGED_STATUS)
6496 		printf(" - TAGGED_STATUS\n");
6497 
6498 	/* PHY related */
6499 	if (sc->bge_phy_flags & BGEPHYF_NO_3LED)
6500 		printf(" - No 3 LEDs\n");
6501 	if (sc->bge_phy_flags & BGEPHYF_CRC_BUG)
6502 		printf(" - CRC bug\n");
6503 	if (sc->bge_phy_flags & BGEPHYF_ADC_BUG)
6504 		printf(" - ADC bug\n");
6505 	if (sc->bge_phy_flags & BGEPHYF_5704_A0_BUG)
6506 		printf(" - 5704 A0 bug\n");
6507 	if (sc->bge_phy_flags & BGEPHYF_JITTER_BUG)
6508 		printf(" - jitter bug\n");
6509 	if (sc->bge_phy_flags & BGEPHYF_BER_BUG)
6510 		printf(" - BER bug\n");
6511 	if (sc->bge_phy_flags & BGEPHYF_ADJUST_TRIM)
6512 		printf(" - adjust trim\n");
6513 	if (sc->bge_phy_flags & BGEPHYF_NO_WIRESPEED)
6514 		printf(" - no wirespeed\n");
6515 
6516 	/* ASF related */
6517 	if (sc->bge_asf_mode & ASF_ENABLE)
6518 		printf(" - ASF enable\n");
6519 	if (sc->bge_asf_mode & ASF_NEW_HANDSHAKE)
6520 		printf(" - ASF new handshake\n");
6521 	if (sc->bge_asf_mode & ASF_STACKUP)
6522 		printf(" - ASF stackup\n");
6523 }
6524 #endif /* BGE_DEBUG */
6525 
6526 static int
6527 bge_get_eaddr_fw(struct bge_softc *sc, uint8_t ether_addr[])
6528 {
6529 	prop_dictionary_t dict;
6530 	prop_data_t ea;
6531 
6532 	if ((sc->bge_flags & BGEF_NO_EEPROM) == 0)
6533 		return 1;
6534 
6535 	dict = device_properties(sc->bge_dev);
6536 	ea = prop_dictionary_get(dict, "mac-address");
6537 	if (ea != NULL) {
6538 		KASSERT(prop_object_type(ea) == PROP_TYPE_DATA);
6539 		KASSERT(prop_data_size(ea) == ETHER_ADDR_LEN);
6540 		memcpy(ether_addr, prop_data_value(ea), ETHER_ADDR_LEN);
6541 		return 0;
6542 	}
6543 
6544 	return 1;
6545 }
6546 
6547 static int
6548 bge_get_eaddr_mem(struct bge_softc *sc, uint8_t ether_addr[])
6549 {
6550 	uint32_t mac_addr;
6551 
6552 	mac_addr = bge_readmem_ind(sc, BGE_SRAM_MAC_ADDR_HIGH_MB);
6553 	if ((mac_addr >> 16) == 0x484b) {
6554 		ether_addr[0] = (uint8_t)(mac_addr >> 8);
6555 		ether_addr[1] = (uint8_t)mac_addr;
6556 		mac_addr = bge_readmem_ind(sc, BGE_SRAM_MAC_ADDR_LOW_MB);
6557 		ether_addr[2] = (uint8_t)(mac_addr >> 24);
6558 		ether_addr[3] = (uint8_t)(mac_addr >> 16);
6559 		ether_addr[4] = (uint8_t)(mac_addr >> 8);
6560 		ether_addr[5] = (uint8_t)mac_addr;
6561 		return 0;
6562 	}
6563 	return 1;
6564 }
6565 
6566 static int
6567 bge_get_eaddr_nvram(struct bge_softc *sc, uint8_t ether_addr[])
6568 {
6569 	int mac_offset = BGE_EE_MAC_OFFSET;
6570 
6571 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906)
6572 		mac_offset = BGE_EE_MAC_OFFSET_5906;
6573 
6574 	return (bge_read_nvram(sc, ether_addr, mac_offset + 2,
6575 	    ETHER_ADDR_LEN));
6576 }
6577 
6578 static int
6579 bge_get_eaddr_eeprom(struct bge_softc *sc, uint8_t ether_addr[])
6580 {
6581 
6582 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906)
6583 		return 1;
6584 
6585 	return (bge_read_eeprom(sc, ether_addr, BGE_EE_MAC_OFFSET + 2,
6586 	   ETHER_ADDR_LEN));
6587 }
6588 
6589 static int
6590 bge_get_eaddr(struct bge_softc *sc, uint8_t eaddr[])
6591 {
6592 	static const bge_eaddr_fcn_t bge_eaddr_funcs[] = {
6593 		/* NOTE: Order is critical */
6594 		bge_get_eaddr_fw,
6595 		bge_get_eaddr_mem,
6596 		bge_get_eaddr_nvram,
6597 		bge_get_eaddr_eeprom,
6598 		NULL
6599 	};
6600 	const bge_eaddr_fcn_t *func;
6601 
6602 	for (func = bge_eaddr_funcs; *func != NULL; ++func) {
6603 		if ((*func)(sc, eaddr) == 0)
6604 			break;
6605 	}
6606 	return *func == NULL ? ENXIO : 0;
6607 }
6608