xref: /netbsd-src/sys/dev/pci/if_bge.c (revision 3117ece4fc4a4ca4489ba793710b60b0d26bab6c)
1 /*	$NetBSD: if_bge.c,v 1.396 2024/09/14 07:01:33 skrll 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 1.396 2024/09/14 07:01:33 skrll 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 void bge_stop(struct ifnet *, int);
207 static bool bge_watchdog_tick(struct ifnet *);
208 static int bge_ifmedia_upd(struct ifnet *);
209 static void bge_ifmedia_sts(struct ifnet *, struct ifmediareq *);
210 static void bge_handle_reset_work(struct work *, void *);
211 
212 static uint8_t bge_nvram_getbyte(struct bge_softc *, int, uint8_t *);
213 static int bge_read_nvram(struct bge_softc *, uint8_t *, int, int);
214 
215 static uint8_t bge_eeprom_getbyte(struct bge_softc *, int, uint8_t *);
216 static int bge_read_eeprom(struct bge_softc *, void *, int, int);
217 static void bge_setmulti(struct bge_softc *);
218 
219 static void bge_handle_events(struct bge_softc *);
220 static int bge_alloc_jumbo_mem(struct bge_softc *);
221 static void bge_free_jumbo_mem(struct bge_softc *);
222 static void *bge_jalloc(struct bge_softc *);
223 static void bge_jfree(struct mbuf *, void *, size_t, void *);
224 static int bge_newbuf_jumbo(struct bge_softc *, int, struct mbuf *);
225 static int bge_init_rx_ring_jumbo(struct bge_softc *);
226 static void bge_free_rx_ring_jumbo(struct bge_softc *);
227 
228 static int bge_newbuf_std(struct bge_softc *, int);
229 static int bge_init_rx_ring_std(struct bge_softc *);
230 static void bge_fill_rx_ring_std(struct bge_softc *);
231 static void bge_free_rx_ring_std(struct bge_softc *m);
232 
233 static void bge_free_tx_ring(struct bge_softc *m, bool);
234 static int bge_init_tx_ring(struct bge_softc *);
235 
236 static int bge_chipinit(struct bge_softc *);
237 static int bge_blockinit(struct bge_softc *);
238 static int bge_phy_addr(struct bge_softc *);
239 static uint32_t bge_readmem_ind(struct bge_softc *, int);
240 static void bge_writemem_ind(struct bge_softc *, int, int);
241 static void bge_writembx(struct bge_softc *, int, int);
242 static void bge_writembx_flush(struct bge_softc *, int, int);
243 static void bge_writemem_direct(struct bge_softc *, int, int);
244 static void bge_writereg_ind(struct bge_softc *, int, int);
245 static void bge_set_max_readrq(struct bge_softc *);
246 
247 static int bge_miibus_readreg(device_t, int, int, uint16_t *);
248 static int bge_miibus_writereg(device_t, int, int, uint16_t);
249 static void bge_miibus_statchg(struct ifnet *);
250 
251 #define BGE_RESET_SHUTDOWN	0
252 #define	BGE_RESET_START		1
253 #define	BGE_RESET_SUSPEND	2
254 static void bge_sig_post_reset(struct bge_softc *, int);
255 static void bge_sig_legacy(struct bge_softc *, int);
256 static void bge_sig_pre_reset(struct bge_softc *, int);
257 static void bge_wait_for_event_ack(struct bge_softc *);
258 static void bge_stop_fw(struct bge_softc *);
259 static int bge_reset(struct bge_softc *);
260 static void bge_link_upd(struct bge_softc *);
261 static void bge_sysctl_init(struct bge_softc *);
262 static int bge_sysctl_verify(SYSCTLFN_PROTO);
263 
264 static void bge_ape_lock_init(struct bge_softc *);
265 static void bge_ape_read_fw_ver(struct bge_softc *);
266 static int bge_ape_lock(struct bge_softc *, int);
267 static void bge_ape_unlock(struct bge_softc *, int);
268 static void bge_ape_send_event(struct bge_softc *, uint32_t);
269 static void bge_ape_driver_state_change(struct bge_softc *, int);
270 
271 #ifdef BGE_DEBUG
272 #define DPRINTF(x)	if (bgedebug) printf x
273 #define DPRINTFN(n, x)	if (bgedebug >= (n)) printf x
274 #define BGE_TSO_PRINTF(x)  do { if (bge_tso_debug) printf x ;} while (0)
275 int	bgedebug = 0;
276 int	bge_tso_debug = 0;
277 void	bge_debug_info(struct bge_softc *);
278 #else
279 #define DPRINTF(x)
280 #define DPRINTFN(n, x)
281 #define BGE_TSO_PRINTF(x)
282 #endif
283 
284 #ifdef BGE_EVENT_COUNTERS
285 #define	BGE_EVCNT_INCR(ev)	(ev).ev_count++
286 #define	BGE_EVCNT_ADD(ev, val)	(ev).ev_count += (val)
287 #define	BGE_EVCNT_UPD(ev, val)	(ev).ev_count = (val)
288 #else
289 #define	BGE_EVCNT_INCR(ev)	/* nothing */
290 #define	BGE_EVCNT_ADD(ev, val)	/* nothing */
291 #define	BGE_EVCNT_UPD(ev, val)	/* nothing */
292 #endif
293 
294 #define VIDDID(a, b) PCI_VENDOR_ ## a, PCI_PRODUCT_ ## a ## _ ## b
295 /*
296  * The BCM5700 documentation seems to indicate that the hardware still has the
297  * Alteon vendor ID burned into it, though it should always be overridden by
298  * the value in the EEPROM.  We'll check for it anyway.
299  */
300 static const struct bge_product {
301 	pci_vendor_id_t		bp_vendor;
302 	pci_product_id_t	bp_product;
303 	const char		*bp_name;
304 } bge_products[] = {
305 	{ VIDDID(ALTEON,   BCM5700),	"Broadcom BCM5700 Gigabit" },
306 	{ VIDDID(ALTEON,   BCM5701),	"Broadcom BCM5701 Gigabit" },
307 	{ VIDDID(ALTIMA,   AC1000),	"Altima AC1000 Gigabit" },
308 	{ VIDDID(ALTIMA,   AC1001),	"Altima AC1001 Gigabit" },
309 	{ VIDDID(ALTIMA,   AC1003),	"Altima AC1003 Gigabit" },
310 	{ VIDDID(ALTIMA,   AC9100),	"Altima AC9100 Gigabit" },
311 	{ VIDDID(APPLE,	   BCM5701),	"APPLE BCM5701 Gigabit" },
312 	{ VIDDID(BROADCOM, BCM5700),	"Broadcom BCM5700 Gigabit" },
313 	{ VIDDID(BROADCOM, BCM5701),	"Broadcom BCM5701 Gigabit" },
314 	{ VIDDID(BROADCOM, BCM5702),	"Broadcom BCM5702 Gigabit" },
315 	{ VIDDID(BROADCOM, BCM5702FE),	"Broadcom BCM5702FE Fast" },
316 	{ VIDDID(BROADCOM, BCM5702X),	"Broadcom BCM5702X Gigabit" },
317 	{ VIDDID(BROADCOM, BCM5703),	"Broadcom BCM5703 Gigabit" },
318 	{ VIDDID(BROADCOM, BCM5703X),	"Broadcom BCM5703X Gigabit" },
319 	{ VIDDID(BROADCOM, BCM5703_ALT),"Broadcom BCM5703 Gigabit" },
320 	{ VIDDID(BROADCOM, BCM5704C),	"Broadcom BCM5704C Dual Gigabit" },
321 	{ VIDDID(BROADCOM, BCM5704S),	"Broadcom BCM5704S Dual Gigabit" },
322 	{ VIDDID(BROADCOM, BCM5704S_ALT),"Broadcom BCM5704S Dual Gigabit" },
323 	{ VIDDID(BROADCOM, BCM5705),	"Broadcom BCM5705 Gigabit" },
324 	{ VIDDID(BROADCOM, BCM5705F),	"Broadcom BCM5705F Gigabit" },
325 	{ VIDDID(BROADCOM, BCM5705K),	"Broadcom BCM5705K Gigabit" },
326 	{ VIDDID(BROADCOM, BCM5705M),	"Broadcom BCM5705M Gigabit" },
327 	{ VIDDID(BROADCOM, BCM5705M_ALT),"Broadcom BCM5705M Gigabit" },
328 	{ VIDDID(BROADCOM, BCM5714),	"Broadcom BCM5714 Gigabit" },
329 	{ VIDDID(BROADCOM, BCM5714S),	"Broadcom BCM5714S Gigabit" },
330 	{ VIDDID(BROADCOM, BCM5715),	"Broadcom BCM5715 Gigabit" },
331 	{ VIDDID(BROADCOM, BCM5715S),	"Broadcom BCM5715S Gigabit" },
332 	{ VIDDID(BROADCOM, BCM5717),	"Broadcom BCM5717 Gigabit" },
333 	{ VIDDID(BROADCOM, BCM5717C),	"Broadcom BCM5717 Gigabit" },
334 	{ VIDDID(BROADCOM, BCM5718),	"Broadcom BCM5718 Gigabit" },
335 	{ VIDDID(BROADCOM, BCM5719),	"Broadcom BCM5719 Gigabit" },
336 	{ VIDDID(BROADCOM, BCM5720),	"Broadcom BCM5720 Gigabit" },
337 	{ VIDDID(BROADCOM, BCM5721),	"Broadcom BCM5721 Gigabit" },
338 	{ VIDDID(BROADCOM, BCM5722),	"Broadcom BCM5722 Gigabit" },
339 	{ VIDDID(BROADCOM, BCM5723),	"Broadcom BCM5723 Gigabit" },
340 	{ VIDDID(BROADCOM, BCM5725),	"Broadcom BCM5725 Gigabit" },
341 	{ VIDDID(BROADCOM, BCM5727),	"Broadcom BCM5727 Gigabit" },
342 	{ VIDDID(BROADCOM, BCM5750),	"Broadcom BCM5750 Gigabit" },
343 	{ VIDDID(BROADCOM, BCM5751),	"Broadcom BCM5751 Gigabit" },
344 	{ VIDDID(BROADCOM, BCM5751F),	"Broadcom BCM5751F Gigabit" },
345 	{ VIDDID(BROADCOM, BCM5751M),	"Broadcom BCM5751M Gigabit" },
346 	{ VIDDID(BROADCOM, BCM5752),	"Broadcom BCM5752 Gigabit" },
347 	{ VIDDID(BROADCOM, BCM5752M),	"Broadcom BCM5752M Gigabit" },
348 	{ VIDDID(BROADCOM, BCM5753),	"Broadcom BCM5753 Gigabit" },
349 	{ VIDDID(BROADCOM, BCM5753F),	"Broadcom BCM5753F Gigabit" },
350 	{ VIDDID(BROADCOM, BCM5753M),	"Broadcom BCM5753M Gigabit" },
351 	{ VIDDID(BROADCOM, BCM5754),	"Broadcom BCM5754 Gigabit" },
352 	{ VIDDID(BROADCOM, BCM5754M),	"Broadcom BCM5754M Gigabit" },
353 	{ VIDDID(BROADCOM, BCM5755),	"Broadcom BCM5755 Gigabit" },
354 	{ VIDDID(BROADCOM, BCM5755M),	"Broadcom BCM5755M Gigabit" },
355 	{ VIDDID(BROADCOM, BCM5756),	"Broadcom BCM5756 Gigabit" },
356 	{ VIDDID(BROADCOM, BCM5761),	"Broadcom BCM5761 Gigabit" },
357 	{ VIDDID(BROADCOM, BCM5761E),	"Broadcom BCM5761E Gigabit" },
358 	{ VIDDID(BROADCOM, BCM5761S),	"Broadcom BCM5761S Gigabit" },
359 	{ VIDDID(BROADCOM, BCM5761SE),	"Broadcom BCM5761SE Gigabit" },
360 	{ VIDDID(BROADCOM, BCM5762),	"Broadcom BCM5762 Gigabit" },
361 	{ VIDDID(BROADCOM, BCM5764),	"Broadcom BCM5764 Gigabit" },
362 	{ VIDDID(BROADCOM, BCM5780),	"Broadcom BCM5780 Gigabit" },
363 	{ VIDDID(BROADCOM, BCM5780S),	"Broadcom BCM5780S Gigabit" },
364 	{ VIDDID(BROADCOM, BCM5781),	"Broadcom BCM5781 Gigabit" },
365 	{ VIDDID(BROADCOM, BCM5782),	"Broadcom BCM5782 Gigabit" },
366 	{ VIDDID(BROADCOM, BCM5784M),	"BCM5784M NetLink 1000baseT" },
367 	{ VIDDID(BROADCOM, BCM5785F),	"BCM5785F NetLink 10/100" },
368 	{ VIDDID(BROADCOM, BCM5785G),	"BCM5785G NetLink 1000baseT" },
369 	{ VIDDID(BROADCOM, BCM5786),	"Broadcom BCM5786 Gigabit" },
370 	{ VIDDID(BROADCOM, BCM5787),	"Broadcom BCM5787 Gigabit" },
371 	{ VIDDID(BROADCOM, BCM5787F),	"Broadcom BCM5787F 10/100" },
372 	{ VIDDID(BROADCOM, BCM5787M),	"Broadcom BCM5787M Gigabit" },
373 	{ VIDDID(BROADCOM, BCM5788),	"Broadcom BCM5788 Gigabit" },
374 	{ VIDDID(BROADCOM, BCM5789),	"Broadcom BCM5789 Gigabit" },
375 	{ VIDDID(BROADCOM, BCM5901),	"Broadcom BCM5901 Fast" },
376 	{ VIDDID(BROADCOM, BCM5901A2),	"Broadcom BCM5901A2 Fast" },
377 	{ VIDDID(BROADCOM, BCM5903M),	"Broadcom BCM5903M Fast" },
378 	{ VIDDID(BROADCOM, BCM5906),	"Broadcom BCM5906 Fast" },
379 	{ VIDDID(BROADCOM, BCM5906M),	"Broadcom BCM5906M Fast" },
380 	{ VIDDID(BROADCOM, BCM57760),	"Broadcom BCM57760 Gigabit" },
381 	{ VIDDID(BROADCOM, BCM57761),	"Broadcom BCM57761 Gigabit" },
382 	{ VIDDID(BROADCOM, BCM57762),	"Broadcom BCM57762 Gigabit" },
383 	{ VIDDID(BROADCOM, BCM57764),	"Broadcom BCM57764 Gigabit" },
384 	{ VIDDID(BROADCOM, BCM57765),	"Broadcom BCM57765 Gigabit" },
385 	{ VIDDID(BROADCOM, BCM57766),	"Broadcom BCM57766 Gigabit" },
386 	{ VIDDID(BROADCOM, BCM57767),	"Broadcom BCM57767 Gigabit" },
387 	{ VIDDID(BROADCOM, BCM57780),	"Broadcom BCM57780 Gigabit" },
388 	{ VIDDID(BROADCOM, BCM57781),	"Broadcom BCM57781 Gigabit" },
389 	{ VIDDID(BROADCOM, BCM57782),	"Broadcom BCM57782 Gigabit" },
390 	{ VIDDID(BROADCOM, BCM57785),	"Broadcom BCM57785 Gigabit" },
391 	{ VIDDID(BROADCOM, BCM57786),	"Broadcom BCM57786 Gigabit" },
392 	{ VIDDID(BROADCOM, BCM57787),	"Broadcom BCM57787 Gigabit" },
393 	{ VIDDID(BROADCOM, BCM57788),	"Broadcom BCM57788 Gigabit" },
394 	{ VIDDID(BROADCOM, BCM57790),	"Broadcom BCM57790 Gigabit" },
395 	{ VIDDID(BROADCOM, BCM57791),	"Broadcom BCM57791 Gigabit" },
396 	{ VIDDID(BROADCOM, BCM57795),	"Broadcom BCM57795 Gigabit" },
397 	{ VIDDID(SCHNEIDERKOCH, SK_9DX1),"SysKonnect SK-9Dx1 Gigabit" },
398 	{ VIDDID(SCHNEIDERKOCH, SK_9MXX),"SysKonnect SK-9Mxx Gigabit" },
399 	{ VIDDID(3COM, 3C996),		"3Com 3c996 Gigabit" },
400 	{ VIDDID(FUJITSU4, PW008GE4),	"Fujitsu PW008GE4 Gigabit" },
401 	{ VIDDID(FUJITSU4, PW008GE5),	"Fujitsu PW008GE5 Gigabit" },
402 	{ VIDDID(FUJITSU4, PP250_450_LAN),"Fujitsu Primepower 250/450 Gigabit" },
403 	{ 0, 0, NULL },
404 };
405 
406 #define BGE_IS_JUMBO_CAPABLE(sc)	((sc)->bge_flags & BGEF_JUMBO_CAPABLE)
407 #define BGE_IS_5700_FAMILY(sc)		((sc)->bge_flags & BGEF_5700_FAMILY)
408 #define BGE_IS_5705_PLUS(sc)		((sc)->bge_flags & BGEF_5705_PLUS)
409 #define BGE_IS_5714_FAMILY(sc)		((sc)->bge_flags & BGEF_5714_FAMILY)
410 #define BGE_IS_575X_PLUS(sc)		((sc)->bge_flags & BGEF_575X_PLUS)
411 #define BGE_IS_5755_PLUS(sc)		((sc)->bge_flags & BGEF_5755_PLUS)
412 #define BGE_IS_57765_FAMILY(sc)		((sc)->bge_flags & BGEF_57765_FAMILY)
413 #define BGE_IS_57765_PLUS(sc)		((sc)->bge_flags & BGEF_57765_PLUS)
414 #define BGE_IS_5717_PLUS(sc)		((sc)->bge_flags & BGEF_5717_PLUS)
415 
416 static const struct bge_revision {
417 	uint32_t		br_chipid;
418 	const char		*br_name;
419 } bge_revisions[] = {
420 	{ BGE_CHIPID_BCM5700_A0, "BCM5700 A0" },
421 	{ BGE_CHIPID_BCM5700_A1, "BCM5700 A1" },
422 	{ BGE_CHIPID_BCM5700_B0, "BCM5700 B0" },
423 	{ BGE_CHIPID_BCM5700_B1, "BCM5700 B1" },
424 	{ BGE_CHIPID_BCM5700_B2, "BCM5700 B2" },
425 	{ BGE_CHIPID_BCM5700_B3, "BCM5700 B3" },
426 	{ BGE_CHIPID_BCM5700_ALTIMA, "BCM5700 Altima" },
427 	{ BGE_CHIPID_BCM5700_C0, "BCM5700 C0" },
428 	{ BGE_CHIPID_BCM5701_A0, "BCM5701 A0" },
429 	{ BGE_CHIPID_BCM5701_B0, "BCM5701 B0" },
430 	{ BGE_CHIPID_BCM5701_B2, "BCM5701 B2" },
431 	{ BGE_CHIPID_BCM5701_B5, "BCM5701 B5" },
432 	{ BGE_CHIPID_BCM5703_A0, "BCM5702/5703 A0" },
433 	{ BGE_CHIPID_BCM5703_A1, "BCM5702/5703 A1" },
434 	{ BGE_CHIPID_BCM5703_A2, "BCM5702/5703 A2" },
435 	{ BGE_CHIPID_BCM5703_A3, "BCM5702/5703 A3" },
436 	{ BGE_CHIPID_BCM5703_B0, "BCM5702/5703 B0" },
437 	{ BGE_CHIPID_BCM5704_A0, "BCM5704 A0" },
438 	{ BGE_CHIPID_BCM5704_A1, "BCM5704 A1" },
439 	{ BGE_CHIPID_BCM5704_A2, "BCM5704 A2" },
440 	{ BGE_CHIPID_BCM5704_A3, "BCM5704 A3" },
441 	{ BGE_CHIPID_BCM5704_B0, "BCM5704 B0" },
442 	{ BGE_CHIPID_BCM5705_A0, "BCM5705 A0" },
443 	{ BGE_CHIPID_BCM5705_A1, "BCM5705 A1" },
444 	{ BGE_CHIPID_BCM5705_A2, "BCM5705 A2" },
445 	{ BGE_CHIPID_BCM5705_A3, "BCM5705 A3" },
446 	{ BGE_CHIPID_BCM5750_A0, "BCM5750 A0" },
447 	{ BGE_CHIPID_BCM5750_A1, "BCM5750 A1" },
448 	{ BGE_CHIPID_BCM5750_A3, "BCM5750 A3" },
449 	{ BGE_CHIPID_BCM5750_B0, "BCM5750 B0" },
450 	{ BGE_CHIPID_BCM5750_B1, "BCM5750 B1" },
451 	{ BGE_CHIPID_BCM5750_C0, "BCM5750 C0" },
452 	{ BGE_CHIPID_BCM5750_C1, "BCM5750 C1" },
453 	{ BGE_CHIPID_BCM5750_C2, "BCM5750 C2" },
454 	{ BGE_CHIPID_BCM5752_A0, "BCM5752 A0" },
455 	{ BGE_CHIPID_BCM5752_A1, "BCM5752 A1" },
456 	{ BGE_CHIPID_BCM5752_A2, "BCM5752 A2" },
457 	{ BGE_CHIPID_BCM5714_A0, "BCM5714 A0" },
458 	{ BGE_CHIPID_BCM5714_B0, "BCM5714 B0" },
459 	{ BGE_CHIPID_BCM5714_B3, "BCM5714 B3" },
460 	{ BGE_CHIPID_BCM5715_A0, "BCM5715 A0" },
461 	{ BGE_CHIPID_BCM5715_A1, "BCM5715 A1" },
462 	{ BGE_CHIPID_BCM5715_A3, "BCM5715 A3" },
463 	{ BGE_CHIPID_BCM5717_A0, "BCM5717 A0" },
464 	{ BGE_CHIPID_BCM5717_B0, "BCM5717 B0" },
465 	{ BGE_CHIPID_BCM5719_A0, "BCM5719 A0" },
466 	{ BGE_CHIPID_BCM5720_A0, "BCM5720 A0" },
467 	{ BGE_CHIPID_BCM5755_A0, "BCM5755 A0" },
468 	{ BGE_CHIPID_BCM5755_A1, "BCM5755 A1" },
469 	{ BGE_CHIPID_BCM5755_A2, "BCM5755 A2" },
470 	{ BGE_CHIPID_BCM5755_C0, "BCM5755 C0" },
471 	{ BGE_CHIPID_BCM5761_A0, "BCM5761 A0" },
472 	{ BGE_CHIPID_BCM5761_A1, "BCM5761 A1" },
473 	{ BGE_CHIPID_BCM5762_A0, "BCM5762 A0" },
474 	{ BGE_CHIPID_BCM5762_B0, "BCM5762 B0" },
475 	{ BGE_CHIPID_BCM5784_A0, "BCM5784 A0" },
476 	{ BGE_CHIPID_BCM5784_A1, "BCM5784 A1" },
477 	{ BGE_CHIPID_BCM5784_B0, "BCM5784 B0" },
478 	/* 5754 and 5787 share the same ASIC ID */
479 	{ BGE_CHIPID_BCM5787_A0, "BCM5754/5787 A0" },
480 	{ BGE_CHIPID_BCM5787_A1, "BCM5754/5787 A1" },
481 	{ BGE_CHIPID_BCM5787_A2, "BCM5754/5787 A2" },
482 	{ BGE_CHIPID_BCM5906_A0, "BCM5906 A0" },
483 	{ BGE_CHIPID_BCM5906_A1, "BCM5906 A1" },
484 	{ BGE_CHIPID_BCM5906_A2, "BCM5906 A2" },
485 	{ BGE_CHIPID_BCM57765_A0, "BCM57765 A0" },
486 	{ BGE_CHIPID_BCM57765_B0, "BCM57765 B0" },
487 	{ BGE_CHIPID_BCM57766_A0, "BCM57766 A0" },
488 	{ BGE_CHIPID_BCM57780_A0, "BCM57780 A0" },
489 	{ BGE_CHIPID_BCM57780_A1, "BCM57780 A1" },
490 
491 	{ 0, NULL }
492 };
493 
494 /*
495  * Some defaults for major revisions, so that newer steppings
496  * that we don't know about have a shot at working.
497  */
498 static const struct bge_revision bge_majorrevs[] = {
499 	{ BGE_ASICREV_BCM5700, "unknown BCM5700" },
500 	{ BGE_ASICREV_BCM5701, "unknown BCM5701" },
501 	{ BGE_ASICREV_BCM5703, "unknown BCM5703" },
502 	{ BGE_ASICREV_BCM5704, "unknown BCM5704" },
503 	{ BGE_ASICREV_BCM5705, "unknown BCM5705" },
504 	{ BGE_ASICREV_BCM5750, "unknown BCM5750" },
505 	{ BGE_ASICREV_BCM5714, "unknown BCM5714" },
506 	{ BGE_ASICREV_BCM5714_A0, "unknown BCM5714" },
507 	{ BGE_ASICREV_BCM5752, "unknown BCM5752" },
508 	{ BGE_ASICREV_BCM5780, "unknown BCM5780" },
509 	{ BGE_ASICREV_BCM5755, "unknown BCM5755" },
510 	{ BGE_ASICREV_BCM5761, "unknown BCM5761" },
511 	{ BGE_ASICREV_BCM5784, "unknown BCM5784" },
512 	{ BGE_ASICREV_BCM5785, "unknown BCM5785" },
513 	/* 5754 and 5787 share the same ASIC ID */
514 	{ BGE_ASICREV_BCM5787, "unknown BCM5754/5787" },
515 	{ BGE_ASICREV_BCM5906, "unknown BCM5906" },
516 	{ BGE_ASICREV_BCM57765, "unknown BCM57765" },
517 	{ BGE_ASICREV_BCM57766, "unknown BCM57766" },
518 	{ BGE_ASICREV_BCM57780, "unknown BCM57780" },
519 	{ BGE_ASICREV_BCM5717, "unknown BCM5717" },
520 	{ BGE_ASICREV_BCM5719, "unknown BCM5719" },
521 	{ BGE_ASICREV_BCM5720, "unknown BCM5720" },
522 	{ BGE_ASICREV_BCM5762, "unknown BCM5762" },
523 
524 	{ 0, NULL }
525 };
526 
527 static int bge_allow_asf = 1;
528 
529 #ifndef BGE_WATCHDOG_TIMEOUT
530 #define BGE_WATCHDOG_TIMEOUT 5
531 #endif
532 static int bge_watchdog_timeout = BGE_WATCHDOG_TIMEOUT;
533 
534 
535 CFATTACH_DECL3_NEW(bge, sizeof(struct bge_softc),
536     bge_probe, bge_attach, bge_detach, NULL, NULL, NULL, DVF_DETACH_SHUTDOWN);
537 
538 static uint32_t
539 bge_readmem_ind(struct bge_softc *sc, int off)
540 {
541 	pcireg_t val;
542 
543 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906 &&
544 	    off >= BGE_STATS_BLOCK && off < BGE_SEND_RING_1_TO_4)
545 		return 0;
546 
547 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MEMWIN_BASEADDR, off);
548 	val = pci_conf_read(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MEMWIN_DATA);
549 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MEMWIN_BASEADDR, 0);
550 	return val;
551 }
552 
553 static void
554 bge_writemem_ind(struct bge_softc *sc, int off, int val)
555 {
556 
557 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MEMWIN_BASEADDR, off);
558 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MEMWIN_DATA, val);
559 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MEMWIN_BASEADDR, 0);
560 }
561 
562 /*
563  * PCI Express only
564  */
565 static void
566 bge_set_max_readrq(struct bge_softc *sc)
567 {
568 	pcireg_t val;
569 
570 	val = pci_conf_read(sc->sc_pc, sc->sc_pcitag, sc->bge_pciecap
571 	    + PCIE_DCSR);
572 	val &= ~PCIE_DCSR_MAX_READ_REQ;
573 	switch (sc->bge_expmrq) {
574 	case 2048:
575 		val |= BGE_PCIE_DEVCTL_MAX_READRQ_2048;
576 		break;
577 	case 4096:
578 		val |= BGE_PCIE_DEVCTL_MAX_READRQ_4096;
579 		break;
580 	default:
581 		panic("incorrect expmrq value(%d)", sc->bge_expmrq);
582 		break;
583 	}
584 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, sc->bge_pciecap
585 	    + PCIE_DCSR, val);
586 }
587 
588 #ifdef notdef
589 static uint32_t
590 bge_readreg_ind(struct bge_softc *sc, int off)
591 {
592 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_REG_BASEADDR, off);
593 	return pci_conf_read(sc->sc_pc, sc->sc_pcitag, BGE_PCI_REG_DATA);
594 }
595 #endif
596 
597 static void
598 bge_writereg_ind(struct bge_softc *sc, int off, int val)
599 {
600 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_REG_BASEADDR, off);
601 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_REG_DATA, val);
602 }
603 
604 static void
605 bge_writemem_direct(struct bge_softc *sc, int off, int val)
606 {
607 	CSR_WRITE_4(sc, off, val);
608 }
609 
610 static void
611 bge_writembx(struct bge_softc *sc, int off, int val)
612 {
613 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906)
614 		off += BGE_LPMBX_IRQ0_HI - BGE_MBX_IRQ0_HI;
615 
616 	CSR_WRITE_4(sc, off, val);
617 }
618 
619 static void
620 bge_writembx_flush(struct bge_softc *sc, int off, int val)
621 {
622 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906)
623 		off += BGE_LPMBX_IRQ0_HI - BGE_MBX_IRQ0_HI;
624 
625 	CSR_WRITE_4_FLUSH(sc, off, val);
626 }
627 
628 /*
629  * Clear all stale locks and select the lock for this driver instance.
630  */
631 void
632 bge_ape_lock_init(struct bge_softc *sc)
633 {
634 	struct pci_attach_args *pa = &(sc->bge_pa);
635 	uint32_t bit, regbase;
636 	int i;
637 
638 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5761)
639 		regbase = BGE_APE_LOCK_GRANT;
640 	else
641 		regbase = BGE_APE_PER_LOCK_GRANT;
642 
643 	/* Clear any stale locks. */
644 	for (i = BGE_APE_LOCK_PHY0; i <= BGE_APE_LOCK_GPIO; i++) {
645 		switch (i) {
646 		case BGE_APE_LOCK_PHY0:
647 		case BGE_APE_LOCK_PHY1:
648 		case BGE_APE_LOCK_PHY2:
649 		case BGE_APE_LOCK_PHY3:
650 			bit = BGE_APE_LOCK_GRANT_DRIVER0;
651 			break;
652 		default:
653 			if (pa->pa_function == 0)
654 				bit = BGE_APE_LOCK_GRANT_DRIVER0;
655 			else
656 				bit = (1 << pa->pa_function);
657 		}
658 		APE_WRITE_4(sc, regbase + 4 * i, bit);
659 	}
660 
661 	/* Select the PHY lock based on the device's function number. */
662 	switch (pa->pa_function) {
663 	case 0:
664 		sc->bge_phy_ape_lock = BGE_APE_LOCK_PHY0;
665 		break;
666 	case 1:
667 		sc->bge_phy_ape_lock = BGE_APE_LOCK_PHY1;
668 		break;
669 	case 2:
670 		sc->bge_phy_ape_lock = BGE_APE_LOCK_PHY2;
671 		break;
672 	case 3:
673 		sc->bge_phy_ape_lock = BGE_APE_LOCK_PHY3;
674 		break;
675 	default:
676 		printf("%s: PHY lock not supported on function\n",
677 		    device_xname(sc->bge_dev));
678 		break;
679 	}
680 }
681 
682 /*
683  * Check for APE firmware, set flags, and print version info.
684  */
685 void
686 bge_ape_read_fw_ver(struct bge_softc *sc)
687 {
688 	const char *fwtype;
689 	uint32_t apedata, features;
690 
691 	/* Check for a valid APE signature in shared memory. */
692 	apedata = APE_READ_4(sc, BGE_APE_SEG_SIG);
693 	if (apedata != BGE_APE_SEG_SIG_MAGIC) {
694 		sc->bge_mfw_flags &= ~ BGE_MFW_ON_APE;
695 		return;
696 	}
697 
698 	/* Check if APE firmware is running. */
699 	apedata = APE_READ_4(sc, BGE_APE_FW_STATUS);
700 	if ((apedata & BGE_APE_FW_STATUS_READY) == 0) {
701 		printf("%s: APE signature found but FW status not ready! "
702 		    "0x%08x\n", device_xname(sc->bge_dev), apedata);
703 		return;
704 	}
705 
706 	sc->bge_mfw_flags |= BGE_MFW_ON_APE;
707 
708 	/* Fetch the APE firmware type and version. */
709 	apedata = APE_READ_4(sc, BGE_APE_FW_VERSION);
710 	features = APE_READ_4(sc, BGE_APE_FW_FEATURES);
711 	if ((features & BGE_APE_FW_FEATURE_NCSI) != 0) {
712 		sc->bge_mfw_flags |= BGE_MFW_TYPE_NCSI;
713 		fwtype = "NCSI";
714 	} else if ((features & BGE_APE_FW_FEATURE_DASH) != 0) {
715 		sc->bge_mfw_flags |= BGE_MFW_TYPE_DASH;
716 		fwtype = "DASH";
717 	} else
718 		fwtype = "UNKN";
719 
720 	/* Print the APE firmware version. */
721 	aprint_normal_dev(sc->bge_dev, "APE firmware %s %d.%d.%d.%d\n", fwtype,
722 	    (apedata & BGE_APE_FW_VERSION_MAJMSK) >> BGE_APE_FW_VERSION_MAJSFT,
723 	    (apedata & BGE_APE_FW_VERSION_MINMSK) >> BGE_APE_FW_VERSION_MINSFT,
724 	    (apedata & BGE_APE_FW_VERSION_REVMSK) >> BGE_APE_FW_VERSION_REVSFT,
725 	    (apedata & BGE_APE_FW_VERSION_BLDMSK));
726 }
727 
728 int
729 bge_ape_lock(struct bge_softc *sc, int locknum)
730 {
731 	struct pci_attach_args *pa = &(sc->bge_pa);
732 	uint32_t bit, gnt, req, status;
733 	int i, off;
734 
735 	if ((sc->bge_mfw_flags & BGE_MFW_ON_APE) == 0)
736 		return 0;
737 
738 	/* Lock request/grant registers have different bases. */
739 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5761) {
740 		req = BGE_APE_LOCK_REQ;
741 		gnt = BGE_APE_LOCK_GRANT;
742 	} else {
743 		req = BGE_APE_PER_LOCK_REQ;
744 		gnt = BGE_APE_PER_LOCK_GRANT;
745 	}
746 
747 	off = 4 * locknum;
748 
749 	switch (locknum) {
750 	case BGE_APE_LOCK_GPIO:
751 		/* Lock required when using GPIO. */
752 		if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5761)
753 			return 0;
754 		if (pa->pa_function == 0)
755 			bit = BGE_APE_LOCK_REQ_DRIVER0;
756 		else
757 			bit = (1 << pa->pa_function);
758 		break;
759 	case BGE_APE_LOCK_GRC:
760 		/* Lock required to reset the device. */
761 		if (pa->pa_function == 0)
762 			bit = BGE_APE_LOCK_REQ_DRIVER0;
763 		else
764 			bit = (1 << pa->pa_function);
765 		break;
766 	case BGE_APE_LOCK_MEM:
767 		/* Lock required when accessing certain APE memory. */
768 		if (pa->pa_function == 0)
769 			bit = BGE_APE_LOCK_REQ_DRIVER0;
770 		else
771 			bit = (1 << pa->pa_function);
772 		break;
773 	case BGE_APE_LOCK_PHY0:
774 	case BGE_APE_LOCK_PHY1:
775 	case BGE_APE_LOCK_PHY2:
776 	case BGE_APE_LOCK_PHY3:
777 		/* Lock required when accessing PHYs. */
778 		bit = BGE_APE_LOCK_REQ_DRIVER0;
779 		break;
780 	default:
781 		return EINVAL;
782 	}
783 
784 	/* Request a lock. */
785 	APE_WRITE_4_FLUSH(sc, req + off, bit);
786 
787 	/* Wait up to 1 second to acquire lock. */
788 	for (i = 0; i < 20000; i++) {
789 		status = APE_READ_4(sc, gnt + off);
790 		if (status == bit)
791 			break;
792 		DELAY(50);
793 	}
794 
795 	/* Handle any errors. */
796 	if (status != bit) {
797 		printf("%s: APE lock %d request failed! "
798 		    "request = 0x%04x[0x%04x], status = 0x%04x[0x%04x]\n",
799 		    device_xname(sc->bge_dev),
800 		    locknum, req + off, bit & 0xFFFF, gnt + off,
801 		    status & 0xFFFF);
802 		/* Revoke the lock request. */
803 		APE_WRITE_4(sc, gnt + off, bit);
804 		return EBUSY;
805 	}
806 
807 	return 0;
808 }
809 
810 void
811 bge_ape_unlock(struct bge_softc *sc, int locknum)
812 {
813 	struct pci_attach_args *pa = &(sc->bge_pa);
814 	uint32_t bit, gnt;
815 	int off;
816 
817 	if ((sc->bge_mfw_flags & BGE_MFW_ON_APE) == 0)
818 		return;
819 
820 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5761)
821 		gnt = BGE_APE_LOCK_GRANT;
822 	else
823 		gnt = BGE_APE_PER_LOCK_GRANT;
824 
825 	off = 4 * locknum;
826 
827 	switch (locknum) {
828 	case BGE_APE_LOCK_GPIO:
829 		if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5761)
830 			return;
831 		if (pa->pa_function == 0)
832 			bit = BGE_APE_LOCK_GRANT_DRIVER0;
833 		else
834 			bit = (1 << pa->pa_function);
835 		break;
836 	case BGE_APE_LOCK_GRC:
837 		if (pa->pa_function == 0)
838 			bit = BGE_APE_LOCK_GRANT_DRIVER0;
839 		else
840 			bit = (1 << pa->pa_function);
841 		break;
842 	case BGE_APE_LOCK_MEM:
843 		if (pa->pa_function == 0)
844 			bit = BGE_APE_LOCK_GRANT_DRIVER0;
845 		else
846 			bit = (1 << pa->pa_function);
847 		break;
848 	case BGE_APE_LOCK_PHY0:
849 	case BGE_APE_LOCK_PHY1:
850 	case BGE_APE_LOCK_PHY2:
851 	case BGE_APE_LOCK_PHY3:
852 		bit = BGE_APE_LOCK_GRANT_DRIVER0;
853 		break;
854 	default:
855 		return;
856 	}
857 
858 	/* Write and flush for consecutive bge_ape_lock() */
859 	APE_WRITE_4_FLUSH(sc, gnt + off, bit);
860 }
861 
862 /*
863  * Send an event to the APE firmware.
864  */
865 void
866 bge_ape_send_event(struct bge_softc *sc, uint32_t event)
867 {
868 	uint32_t apedata;
869 	int i;
870 
871 	/* NCSI does not support APE events. */
872 	if ((sc->bge_mfw_flags & BGE_MFW_ON_APE) == 0)
873 		return;
874 
875 	/* Wait up to 1ms for APE to service previous event. */
876 	for (i = 10; i > 0; i--) {
877 		if (bge_ape_lock(sc, BGE_APE_LOCK_MEM) != 0)
878 			break;
879 		apedata = APE_READ_4(sc, BGE_APE_EVENT_STATUS);
880 		if ((apedata & BGE_APE_EVENT_STATUS_EVENT_PENDING) == 0) {
881 			APE_WRITE_4(sc, BGE_APE_EVENT_STATUS, event |
882 			    BGE_APE_EVENT_STATUS_EVENT_PENDING);
883 			bge_ape_unlock(sc, BGE_APE_LOCK_MEM);
884 			APE_WRITE_4(sc, BGE_APE_EVENT, BGE_APE_EVENT_1);
885 			break;
886 		}
887 		bge_ape_unlock(sc, BGE_APE_LOCK_MEM);
888 		DELAY(100);
889 	}
890 	if (i == 0) {
891 		printf("%s: APE event 0x%08x send timed out\n",
892 		    device_xname(sc->bge_dev), event);
893 	}
894 }
895 
896 void
897 bge_ape_driver_state_change(struct bge_softc *sc, int kind)
898 {
899 	uint32_t apedata, event;
900 
901 	if ((sc->bge_mfw_flags & BGE_MFW_ON_APE) == 0)
902 		return;
903 
904 	switch (kind) {
905 	case BGE_RESET_START:
906 		/* If this is the first load, clear the load counter. */
907 		apedata = APE_READ_4(sc, BGE_APE_HOST_SEG_SIG);
908 		if (apedata != BGE_APE_HOST_SEG_SIG_MAGIC)
909 			APE_WRITE_4(sc, BGE_APE_HOST_INIT_COUNT, 0);
910 		else {
911 			apedata = APE_READ_4(sc, BGE_APE_HOST_INIT_COUNT);
912 			APE_WRITE_4(sc, BGE_APE_HOST_INIT_COUNT, ++apedata);
913 		}
914 		APE_WRITE_4(sc, BGE_APE_HOST_SEG_SIG,
915 		    BGE_APE_HOST_SEG_SIG_MAGIC);
916 		APE_WRITE_4(sc, BGE_APE_HOST_SEG_LEN,
917 		    BGE_APE_HOST_SEG_LEN_MAGIC);
918 
919 		/* Add some version info if bge(4) supports it. */
920 		APE_WRITE_4(sc, BGE_APE_HOST_DRIVER_ID,
921 		    BGE_APE_HOST_DRIVER_ID_MAGIC(1, 0));
922 		APE_WRITE_4(sc, BGE_APE_HOST_BEHAVIOR,
923 		    BGE_APE_HOST_BEHAV_NO_PHYLOCK);
924 		APE_WRITE_4(sc, BGE_APE_HOST_HEARTBEAT_INT_MS,
925 		    BGE_APE_HOST_HEARTBEAT_INT_DISABLE);
926 		APE_WRITE_4(sc, BGE_APE_HOST_DRVR_STATE,
927 		    BGE_APE_HOST_DRVR_STATE_START);
928 		event = BGE_APE_EVENT_STATUS_STATE_START;
929 		break;
930 	case BGE_RESET_SHUTDOWN:
931 		APE_WRITE_4(sc, BGE_APE_HOST_DRVR_STATE,
932 		    BGE_APE_HOST_DRVR_STATE_UNLOAD);
933 		event = BGE_APE_EVENT_STATUS_STATE_UNLOAD;
934 		break;
935 	case BGE_RESET_SUSPEND:
936 		event = BGE_APE_EVENT_STATUS_STATE_SUSPEND;
937 		break;
938 	default:
939 		return;
940 	}
941 
942 	bge_ape_send_event(sc, event | BGE_APE_EVENT_STATUS_DRIVER_EVNT |
943 	    BGE_APE_EVENT_STATUS_STATE_CHNGE);
944 }
945 
946 static uint8_t
947 bge_nvram_getbyte(struct bge_softc *sc, int addr, uint8_t *dest)
948 {
949 	uint32_t access, byte = 0;
950 	int i;
951 
952 	/* Lock. */
953 	CSR_WRITE_4(sc, BGE_NVRAM_SWARB, BGE_NVRAMSWARB_SET1);
954 	for (i = 0; i < 8000; i++) {
955 		if (CSR_READ_4(sc, BGE_NVRAM_SWARB) & BGE_NVRAMSWARB_GNT1)
956 			break;
957 		DELAY(20);
958 	}
959 	if (i == 8000)
960 		return 1;
961 
962 	/* Enable access. */
963 	access = CSR_READ_4(sc, BGE_NVRAM_ACCESS);
964 	CSR_WRITE_4(sc, BGE_NVRAM_ACCESS, access | BGE_NVRAMACC_ENABLE);
965 
966 	CSR_WRITE_4(sc, BGE_NVRAM_ADDR, addr & 0xfffffffc);
967 	CSR_WRITE_4(sc, BGE_NVRAM_CMD, BGE_NVRAM_READCMD);
968 	for (i = 0; i < BGE_TIMEOUT * 10; i++) {
969 		DELAY(10);
970 		if (CSR_READ_4(sc, BGE_NVRAM_CMD) & BGE_NVRAMCMD_DONE) {
971 			DELAY(10);
972 			break;
973 		}
974 	}
975 
976 	if (i == BGE_TIMEOUT * 10) {
977 		aprint_error_dev(sc->bge_dev, "nvram read timed out\n");
978 		return 1;
979 	}
980 
981 	/* Get result. */
982 	byte = CSR_READ_4(sc, BGE_NVRAM_RDDATA);
983 
984 	*dest = (bswap32(byte) >> ((addr % 4) * 8)) & 0xFF;
985 
986 	/* Disable access. */
987 	CSR_WRITE_4(sc, BGE_NVRAM_ACCESS, access);
988 
989 	/* Unlock. */
990 	CSR_WRITE_4_FLUSH(sc, BGE_NVRAM_SWARB, BGE_NVRAMSWARB_CLR1);
991 
992 	return 0;
993 }
994 
995 /*
996  * Read a sequence of bytes from NVRAM.
997  */
998 static int
999 bge_read_nvram(struct bge_softc *sc, uint8_t *dest, int off, int cnt)
1000 {
1001 	int error = 0, i;
1002 	uint8_t byte = 0;
1003 
1004 	if (BGE_ASICREV(sc->bge_chipid) != BGE_ASICREV_BCM5906)
1005 		return 1;
1006 
1007 	for (i = 0; i < cnt; i++) {
1008 		error = bge_nvram_getbyte(sc, off + i, &byte);
1009 		if (error)
1010 			break;
1011 		*(dest + i) = byte;
1012 	}
1013 
1014 	return error ? 1 : 0;
1015 }
1016 
1017 /*
1018  * Read a byte of data stored in the EEPROM at address 'addr.' The
1019  * BCM570x supports both the traditional bitbang interface and an
1020  * auto access interface for reading the EEPROM. We use the auto
1021  * access method.
1022  */
1023 static uint8_t
1024 bge_eeprom_getbyte(struct bge_softc *sc, int addr, uint8_t *dest)
1025 {
1026 	int i;
1027 	uint32_t byte = 0;
1028 
1029 	/*
1030 	 * Enable use of auto EEPROM access so we can avoid
1031 	 * having to use the bitbang method.
1032 	 */
1033 	BGE_SETBIT_FLUSH(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_AUTO_EEPROM);
1034 
1035 	/* Reset the EEPROM, load the clock period. */
1036 	CSR_WRITE_4_FLUSH(sc, BGE_EE_ADDR,
1037 	    BGE_EEADDR_RESET | BGE_EEHALFCLK(BGE_HALFCLK_384SCL));
1038 	DELAY(20);
1039 
1040 	/* Issue the read EEPROM command. */
1041 	CSR_WRITE_4(sc, BGE_EE_ADDR, BGE_EE_READCMD | addr);
1042 
1043 	/* Wait for completion */
1044 	for (i = 0; i < BGE_TIMEOUT * 10; i++) {
1045 		DELAY(10);
1046 		if (CSR_READ_4(sc, BGE_EE_ADDR) & BGE_EEADDR_DONE)
1047 			break;
1048 	}
1049 
1050 	if (i == BGE_TIMEOUT * 10) {
1051 		aprint_error_dev(sc->bge_dev, "eeprom read timed out\n");
1052 		return 1;
1053 	}
1054 
1055 	/* Get result. */
1056 	byte = CSR_READ_4(sc, BGE_EE_DATA);
1057 
1058 	*dest = (byte >> ((addr % 4) * 8)) & 0xFF;
1059 
1060 	return 0;
1061 }
1062 
1063 /*
1064  * Read a sequence of bytes from the EEPROM.
1065  */
1066 static int
1067 bge_read_eeprom(struct bge_softc *sc, void *destv, int off, int cnt)
1068 {
1069 	int error = 0, i;
1070 	uint8_t byte = 0;
1071 	char *dest = destv;
1072 
1073 	for (i = 0; i < cnt; i++) {
1074 		error = bge_eeprom_getbyte(sc, off + i, &byte);
1075 		if (error)
1076 			break;
1077 		*(dest + i) = byte;
1078 	}
1079 
1080 	return error ? 1 : 0;
1081 }
1082 
1083 static int
1084 bge_miibus_readreg(device_t dev, int phy, int reg, uint16_t *val)
1085 {
1086 	struct bge_softc * const sc = device_private(dev);
1087 	uint32_t data;
1088 	uint32_t autopoll;
1089 	int rv = 0;
1090 	int i;
1091 
1092 	KASSERT(mutex_owned(sc->sc_intr_lock));
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 	KASSERT(mutex_owned(sc->sc_intr_lock));
1148 
1149 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906 &&
1150 	    (reg == MII_GTCR || reg == BRGPHY_MII_AUXCTL))
1151 		return 0;
1152 
1153 	if (bge_ape_lock(sc, sc->bge_phy_ape_lock) != 0)
1154 		return -1;
1155 
1156 	/* Reading with autopolling on may trigger PCI errors */
1157 	autopoll = CSR_READ_4(sc, BGE_MI_MODE);
1158 	if (autopoll & BGE_MIMODE_AUTOPOLL) {
1159 		BGE_STS_CLRBIT(sc, BGE_STS_AUTOPOLL);
1160 		BGE_CLRBIT_FLUSH(sc, BGE_MI_MODE, BGE_MIMODE_AUTOPOLL);
1161 		DELAY(80);
1162 	}
1163 
1164 	CSR_WRITE_4_FLUSH(sc, BGE_MI_COMM, BGE_MICMD_WRITE | BGE_MICOMM_BUSY |
1165 	    BGE_MIPHY(phy) | BGE_MIREG(reg) | val);
1166 
1167 	for (i = 0; i < BGE_TIMEOUT; i++) {
1168 		delay(10);
1169 		data = CSR_READ_4(sc, BGE_MI_COMM);
1170 		if (!(data & BGE_MICOMM_BUSY)) {
1171 			delay(5);
1172 			data = CSR_READ_4(sc, BGE_MI_COMM);
1173 			break;
1174 		}
1175 	}
1176 
1177 	if (i == BGE_TIMEOUT) {
1178 		aprint_error_dev(sc->bge_dev, "PHY write timed out\n");
1179 		rv = ETIMEDOUT;
1180 	} else if ((data & BGE_MICOMM_READFAIL) != 0) {
1181 		aprint_error_dev(sc->bge_dev, "PHY write I/O error\n");
1182 		rv = EIO;
1183 	}
1184 
1185 	if (autopoll & BGE_MIMODE_AUTOPOLL) {
1186 		BGE_STS_SETBIT(sc, BGE_STS_AUTOPOLL);
1187 		BGE_SETBIT_FLUSH(sc, BGE_MI_MODE, BGE_MIMODE_AUTOPOLL);
1188 		delay(80);
1189 	}
1190 
1191 	bge_ape_unlock(sc, sc->bge_phy_ape_lock);
1192 
1193 	return rv;
1194 }
1195 
1196 static void
1197 bge_miibus_statchg(struct ifnet *ifp)
1198 {
1199 	struct bge_softc * const sc = ifp->if_softc;
1200 	struct mii_data *mii = &sc->bge_mii;
1201 	uint32_t mac_mode, rx_mode, tx_mode;
1202 
1203 	KASSERT(mutex_owned(sc->sc_intr_lock));
1204 
1205 	/*
1206 	 * Get flow control negotiation result.
1207 	 */
1208 	if (IFM_SUBTYPE(mii->mii_media.ifm_cur->ifm_media) == IFM_AUTO &&
1209 	    (mii->mii_media_active & IFM_ETH_FMASK) != sc->bge_flowflags)
1210 		sc->bge_flowflags = mii->mii_media_active & IFM_ETH_FMASK;
1211 
1212 	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_SETBIT(sc, BGE_STS_LINK);
1216 	else if (BGE_STS_BIT(sc, BGE_STS_LINK) &&
1217 	    (!(mii->mii_media_status & IFM_ACTIVE) ||
1218 	    IFM_SUBTYPE(mii->mii_media_active) == IFM_NONE))
1219 		BGE_STS_CLRBIT(sc, BGE_STS_LINK);
1220 
1221 	if (!BGE_STS_BIT(sc, BGE_STS_LINK))
1222 		return;
1223 
1224 	/* Set the port mode (MII/GMII) to match the link speed. */
1225 	mac_mode = CSR_READ_4(sc, BGE_MAC_MODE) &
1226 	    ~(BGE_MACMODE_PORTMODE | BGE_MACMODE_HALF_DUPLEX);
1227 	tx_mode = CSR_READ_4(sc, BGE_TX_MODE);
1228 	rx_mode = CSR_READ_4(sc, BGE_RX_MODE);
1229 	if (IFM_SUBTYPE(mii->mii_media_active) == IFM_1000_T ||
1230 	    IFM_SUBTYPE(mii->mii_media_active) == IFM_1000_SX)
1231 		mac_mode |= BGE_PORTMODE_GMII;
1232 	else
1233 		mac_mode |= BGE_PORTMODE_MII;
1234 
1235 	tx_mode &= ~BGE_TXMODE_FLOWCTL_ENABLE;
1236 	rx_mode &= ~BGE_RXMODE_FLOWCTL_ENABLE;
1237 	if ((mii->mii_media_active & IFM_FDX) != 0) {
1238 		if (sc->bge_flowflags & IFM_ETH_TXPAUSE)
1239 			tx_mode |= BGE_TXMODE_FLOWCTL_ENABLE;
1240 		if (sc->bge_flowflags & IFM_ETH_RXPAUSE)
1241 			rx_mode |= BGE_RXMODE_FLOWCTL_ENABLE;
1242 	} else
1243 		mac_mode |= BGE_MACMODE_HALF_DUPLEX;
1244 
1245 	CSR_WRITE_4_FLUSH(sc, BGE_MAC_MODE, mac_mode);
1246 	DELAY(40);
1247 	CSR_WRITE_4(sc, BGE_TX_MODE, tx_mode);
1248 	CSR_WRITE_4(sc, BGE_RX_MODE, rx_mode);
1249 }
1250 
1251 /*
1252  * Update rx threshold levels to values in a particular slot
1253  * of the interrupt-mitigation table bge_rx_threshes.
1254  */
1255 static void
1256 bge_set_thresh(struct ifnet *ifp, int lvl)
1257 {
1258 	struct bge_softc * const sc = ifp->if_softc;
1259 
1260 	/*
1261 	 * For now, just save the new Rx-intr thresholds and record
1262 	 * that a threshold update is pending.  Updating the hardware
1263 	 * registers here (even at splhigh()) is observed to
1264 	 * occasionally cause glitches where Rx-interrupts are not
1265 	 * honoured for up to 10 seconds. jonathan@NetBSD.org, 2003-04-05
1266 	 */
1267 	mutex_enter(sc->sc_intr_lock);
1268 	sc->bge_rx_coal_ticks = bge_rx_threshes[lvl].rx_ticks;
1269 	sc->bge_rx_max_coal_bds = bge_rx_threshes[lvl].rx_max_bds;
1270 	sc->bge_pending_rxintr_change = true;
1271 	mutex_exit(sc->sc_intr_lock);
1272 }
1273 
1274 
1275 /*
1276  * Update Rx thresholds of all bge devices
1277  */
1278 static void
1279 bge_update_all_threshes(int lvl)
1280 {
1281 	const char * const namebuf = "bge";
1282 	const size_t namelen = strlen(namebuf);
1283 	struct ifnet *ifp;
1284 
1285 	if (lvl < 0)
1286 		lvl = 0;
1287 	else if (lvl >= NBGE_RX_THRESH)
1288 		lvl = NBGE_RX_THRESH - 1;
1289 
1290 	/*
1291 	 * Now search all the interfaces for this name/number
1292 	 */
1293 	int s = pserialize_read_enter();
1294 	IFNET_READER_FOREACH(ifp) {
1295 		if (strncmp(ifp->if_xname, namebuf, namelen) != 0)
1296 			continue;
1297 		/* We got a match: update if doing auto-threshold-tuning */
1298 		if (bge_auto_thresh)
1299 			bge_set_thresh(ifp, lvl);
1300 	}
1301 	pserialize_read_exit(s);
1302 }
1303 
1304 /*
1305  * Handle events that have triggered interrupts.
1306  */
1307 static void
1308 bge_handle_events(struct bge_softc *sc)
1309 {
1310 
1311 	return;
1312 }
1313 
1314 /*
1315  * Memory management for jumbo frames.
1316  */
1317 
1318 static int
1319 bge_alloc_jumbo_mem(struct bge_softc *sc)
1320 {
1321 	char *ptr, *kva;
1322 	int i, rseg, state, error;
1323 	struct bge_jpool_entry *entry;
1324 
1325 	state = error = 0;
1326 
1327 	/* Grab a big chunk o' storage. */
1328 	if (bus_dmamem_alloc(sc->bge_dmatag, BGE_JMEM, PAGE_SIZE, 0,
1329 	    &sc->bge_cdata.bge_rx_jumbo_seg, 1, &rseg, BUS_DMA_WAITOK)) {
1330 		aprint_error_dev(sc->bge_dev, "can't alloc rx buffers\n");
1331 		return ENOBUFS;
1332 	}
1333 
1334 	state = 1;
1335 	if (bus_dmamem_map(sc->bge_dmatag, &sc->bge_cdata.bge_rx_jumbo_seg,
1336 	    rseg, BGE_JMEM, (void **)&kva, BUS_DMA_WAITOK)) {
1337 		aprint_error_dev(sc->bge_dev,
1338 		    "can't map DMA buffers (%d bytes)\n", (int)BGE_JMEM);
1339 		error = ENOBUFS;
1340 		goto out;
1341 	}
1342 
1343 	state = 2;
1344 	if (bus_dmamap_create(sc->bge_dmatag, BGE_JMEM, 1, BGE_JMEM, 0,
1345 	    BUS_DMA_WAITOK, &sc->bge_cdata.bge_rx_jumbo_map)) {
1346 		aprint_error_dev(sc->bge_dev, "can't create DMA map\n");
1347 		error = ENOBUFS;
1348 		goto out;
1349 	}
1350 
1351 	state = 3;
1352 	if (bus_dmamap_load(sc->bge_dmatag, sc->bge_cdata.bge_rx_jumbo_map,
1353 	    kva, BGE_JMEM, NULL, BUS_DMA_WAITOK)) {
1354 		aprint_error_dev(sc->bge_dev, "can't load DMA map\n");
1355 		error = ENOBUFS;
1356 		goto out;
1357 	}
1358 
1359 	state = 4;
1360 	sc->bge_cdata.bge_jumbo_buf = (void *)kva;
1361 	DPRINTFN(1,("bge_jumbo_buf = %p\n", sc->bge_cdata.bge_jumbo_buf));
1362 
1363 	SLIST_INIT(&sc->bge_jfree_listhead);
1364 	SLIST_INIT(&sc->bge_jinuse_listhead);
1365 
1366 	/*
1367 	 * Now divide it up into 9K pieces and save the addresses
1368 	 * in an array.
1369 	 */
1370 	ptr = sc->bge_cdata.bge_jumbo_buf;
1371 	for (i = 0; i < BGE_JSLOTS; i++) {
1372 		sc->bge_cdata.bge_jslots[i] = ptr;
1373 		ptr += BGE_JLEN;
1374 		entry = kmem_alloc(sizeof(*entry), KM_SLEEP);
1375 		entry->slot = i;
1376 		SLIST_INSERT_HEAD(&sc->bge_jfree_listhead,
1377 				 entry, jpool_entries);
1378 	}
1379 out:
1380 	if (error != 0) {
1381 		switch (state) {
1382 		case 4:
1383 			bus_dmamap_unload(sc->bge_dmatag,
1384 			    sc->bge_cdata.bge_rx_jumbo_map);
1385 			/* FALLTHROUGH */
1386 		case 3:
1387 			bus_dmamap_destroy(sc->bge_dmatag,
1388 			    sc->bge_cdata.bge_rx_jumbo_map);
1389 			/* FALLTHROUGH */
1390 		case 2:
1391 			bus_dmamem_unmap(sc->bge_dmatag, kva, BGE_JMEM);
1392 			/* FALLTHROUGH */
1393 		case 1:
1394 			bus_dmamem_free(sc->bge_dmatag,
1395 			    &sc->bge_cdata.bge_rx_jumbo_seg, rseg);
1396 			break;
1397 		default:
1398 			break;
1399 		}
1400 	}
1401 
1402 	return error;
1403 }
1404 
1405 static void
1406 bge_free_jumbo_mem(struct bge_softc *sc)
1407 {
1408 	struct bge_jpool_entry *entry, *tmp;
1409 
1410 	KASSERT(SLIST_EMPTY(&sc->bge_jinuse_listhead));
1411 
1412 	SLIST_FOREACH_SAFE(entry, &sc->bge_jfree_listhead, jpool_entries, tmp) {
1413 		kmem_free(entry, sizeof(*entry));
1414 	}
1415 
1416 	bus_dmamap_unload(sc->bge_dmatag, sc->bge_cdata.bge_rx_jumbo_map);
1417 
1418 	bus_dmamap_destroy(sc->bge_dmatag, sc->bge_cdata.bge_rx_jumbo_map);
1419 
1420 	bus_dmamem_unmap(sc->bge_dmatag, sc->bge_cdata.bge_jumbo_buf, BGE_JMEM);
1421 
1422 	bus_dmamem_free(sc->bge_dmatag, &sc->bge_cdata.bge_rx_jumbo_seg, 1);
1423 }
1424 
1425 /*
1426  * Allocate a jumbo buffer.
1427  */
1428 static void *
1429 bge_jalloc(struct bge_softc *sc)
1430 {
1431 	struct bge_jpool_entry	 *entry;
1432 
1433 	entry = SLIST_FIRST(&sc->bge_jfree_listhead);
1434 
1435 	if (entry == NULL) {
1436 		aprint_error_dev(sc->bge_dev, "no free jumbo buffers\n");
1437 		return NULL;
1438 	}
1439 
1440 	SLIST_REMOVE_HEAD(&sc->bge_jfree_listhead, jpool_entries);
1441 	SLIST_INSERT_HEAD(&sc->bge_jinuse_listhead, entry, jpool_entries);
1442 	return sc->bge_cdata.bge_jslots[entry->slot];
1443 }
1444 
1445 /*
1446  * Release a jumbo buffer.
1447  */
1448 static void
1449 bge_jfree(struct mbuf *m, void *buf, size_t size, void *arg)
1450 {
1451 	struct bge_jpool_entry *entry;
1452 	struct bge_softc * const sc = arg;
1453 
1454 	if (sc == NULL)
1455 		panic("bge_jfree: can't find softc pointer!");
1456 
1457 	/* calculate the slot this buffer belongs to */
1458 	int i = ((char *)buf - (char *)sc->bge_cdata.bge_jumbo_buf) / BGE_JLEN;
1459 
1460 	if (i < 0 || i >= BGE_JSLOTS)
1461 		panic("bge_jfree: asked to free buffer that we don't manage!");
1462 
1463 	mutex_enter(sc->sc_intr_lock);
1464 	entry = SLIST_FIRST(&sc->bge_jinuse_listhead);
1465 	if (entry == NULL)
1466 		panic("bge_jfree: buffer not in use!");
1467 	entry->slot = i;
1468 	SLIST_REMOVE_HEAD(&sc->bge_jinuse_listhead, jpool_entries);
1469 	SLIST_INSERT_HEAD(&sc->bge_jfree_listhead, entry, jpool_entries);
1470 	mutex_exit(sc->sc_intr_lock);
1471 
1472 	if (__predict_true(m != NULL))
1473 		pool_cache_put(mb_cache, m);
1474 }
1475 
1476 
1477 /*
1478  * Initialize a standard receive ring descriptor.
1479  */
1480 static int
1481 bge_newbuf_std(struct bge_softc *sc, int i)
1482 {
1483 	const bus_dmamap_t dmamap = sc->bge_cdata.bge_rx_std_map[i];
1484 	struct mbuf *m;
1485 
1486 	MGETHDR(m, M_DONTWAIT, MT_DATA);
1487 	if (m == NULL)
1488 		return ENOBUFS;
1489 
1490 	MCLGET(m, M_DONTWAIT);
1491 	if (!(m->m_flags & M_EXT)) {
1492 		m_freem(m);
1493 		return ENOBUFS;
1494 	}
1495 	m->m_len = m->m_pkthdr.len = MCLBYTES;
1496 
1497 	if (!(sc->bge_flags & BGEF_RX_ALIGNBUG))
1498 	    m_adj(m, ETHER_ALIGN);
1499 	if (bus_dmamap_load_mbuf(sc->bge_dmatag, dmamap, m,
1500 	    BUS_DMA_READ | BUS_DMA_NOWAIT)) {
1501 		m_freem(m);
1502 		return ENOBUFS;
1503 	}
1504 	bus_dmamap_sync(sc->bge_dmatag, dmamap, 0, dmamap->dm_mapsize,
1505 	    BUS_DMASYNC_PREREAD);
1506 	sc->bge_cdata.bge_rx_std_chain[i] = m;
1507 
1508 	bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
1509 	    offsetof(struct bge_ring_data, bge_rx_std_ring) +
1510 		i * sizeof(struct bge_rx_bd),
1511 	    sizeof(struct bge_rx_bd),
1512 	    BUS_DMASYNC_POSTWRITE);
1513 
1514 	struct bge_rx_bd * const r = &sc->bge_rdata->bge_rx_std_ring[i];
1515 	BGE_HOSTADDR(r->bge_addr, dmamap->dm_segs[0].ds_addr);
1516 	r->bge_flags = BGE_RXBDFLAG_END;
1517 	r->bge_len = m->m_len;
1518 	r->bge_idx = i;
1519 
1520 	bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
1521 	    offsetof(struct bge_ring_data, bge_rx_std_ring) +
1522 		i * sizeof(struct bge_rx_bd),
1523 	    sizeof(struct bge_rx_bd),
1524 	    BUS_DMASYNC_PREWRITE);
1525 
1526 	sc->bge_std_cnt++;
1527 
1528 	return 0;
1529 }
1530 
1531 /*
1532  * Initialize a jumbo receive ring descriptor. This allocates
1533  * a jumbo buffer from the pool managed internally by the driver.
1534  */
1535 static int
1536 bge_newbuf_jumbo(struct bge_softc *sc, int i, struct mbuf *m)
1537 {
1538 	struct mbuf *m_new = NULL;
1539 	struct bge_rx_bd *r;
1540 	void *buf = NULL;
1541 
1542 	if (m == NULL) {
1543 
1544 		/* Allocate the mbuf. */
1545 		MGETHDR(m_new, M_DONTWAIT, MT_DATA);
1546 		if (m_new == NULL)
1547 			return ENOBUFS;
1548 
1549 		/* Allocate the jumbo buffer */
1550 		buf = bge_jalloc(sc);
1551 		if (buf == NULL) {
1552 			m_freem(m_new);
1553 			aprint_error_dev(sc->bge_dev,
1554 			    "jumbo allocation failed -- packet dropped!\n");
1555 			return ENOBUFS;
1556 		}
1557 
1558 		/* Attach the buffer to the mbuf. */
1559 		m_new->m_len = m_new->m_pkthdr.len = BGE_JUMBO_FRAMELEN;
1560 		MEXTADD(m_new, buf, BGE_JUMBO_FRAMELEN, M_DEVBUF,
1561 		    bge_jfree, sc);
1562 		m_new->m_flags |= M_EXT_RW;
1563 	} else {
1564 		m_new = m;
1565 		buf = m_new->m_data = m_new->m_ext.ext_buf;
1566 		m_new->m_ext.ext_size = BGE_JUMBO_FRAMELEN;
1567 	}
1568 	if (!(sc->bge_flags & BGEF_RX_ALIGNBUG))
1569 	    m_adj(m_new, ETHER_ALIGN);
1570 	bus_dmamap_sync(sc->bge_dmatag, sc->bge_cdata.bge_rx_jumbo_map,
1571 	    mtod(m_new, char *) - (char *)sc->bge_cdata.bge_jumbo_buf,
1572 	    BGE_JLEN, BUS_DMASYNC_PREREAD);
1573 
1574 	/* Set up the descriptor. */
1575 	r = &sc->bge_rdata->bge_rx_jumbo_ring[i];
1576 	sc->bge_cdata.bge_rx_jumbo_chain[i] = m_new;
1577 	BGE_HOSTADDR(r->bge_addr, BGE_JUMBO_DMA_ADDR(sc, m_new));
1578 	r->bge_flags = BGE_RXBDFLAG_END | BGE_RXBDFLAG_JUMBO_RING;
1579 	r->bge_len = m_new->m_len;
1580 	r->bge_idx = i;
1581 
1582 	bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
1583 	    offsetof(struct bge_ring_data, bge_rx_jumbo_ring) +
1584 		i * sizeof(struct bge_rx_bd),
1585 	    sizeof(struct bge_rx_bd),
1586 	    BUS_DMASYNC_PREWRITE | BUS_DMASYNC_PREREAD);
1587 
1588 	return 0;
1589 }
1590 
1591 static int
1592 bge_init_rx_ring_std(struct bge_softc *sc)
1593 {
1594 	bus_dmamap_t dmamap;
1595 	int error = 0;
1596 	u_int i;
1597 
1598 	if (sc->bge_flags & BGEF_RXRING_VALID)
1599 		return 0;
1600 
1601 	for (i = 0; i < BGE_STD_RX_RING_CNT; i++) {
1602 		error = bus_dmamap_create(sc->bge_dmatag, MCLBYTES, 1,
1603 		    MCLBYTES, 0, BUS_DMA_WAITOK | BUS_DMA_ALLOCNOW, &dmamap);
1604 		if (error)
1605 			goto uncreate;
1606 
1607 		sc->bge_cdata.bge_rx_std_map[i] = dmamap;
1608 		memset(&sc->bge_rdata->bge_rx_std_ring[i], 0,
1609 		    sizeof(struct bge_rx_bd));
1610 	}
1611 
1612 	sc->bge_std = i - 1;
1613 	sc->bge_std_cnt = 0;
1614 	bge_fill_rx_ring_std(sc);
1615 
1616 	sc->bge_flags |= BGEF_RXRING_VALID;
1617 
1618 	return 0;
1619 
1620 uncreate:
1621 	while (--i) {
1622 		bus_dmamap_destroy(sc->bge_dmatag,
1623 		    sc->bge_cdata.bge_rx_std_map[i]);
1624 	}
1625 	return error;
1626 }
1627 
1628 static void
1629 bge_fill_rx_ring_std(struct bge_softc *sc)
1630 {
1631 	int i = sc->bge_std;
1632 	bool post = false;
1633 
1634 	while (sc->bge_std_cnt < BGE_STD_RX_RING_CNT) {
1635 		BGE_INC(i, BGE_STD_RX_RING_CNT);
1636 
1637 		if (bge_newbuf_std(sc, i) != 0)
1638 			break;
1639 
1640 		sc->bge_std = i;
1641 		post = true;
1642 	}
1643 
1644 	if (post)
1645 		bge_writembx(sc, BGE_MBX_RX_STD_PROD_LO, sc->bge_std);
1646 }
1647 
1648 
1649 static void
1650 bge_free_rx_ring_std(struct bge_softc *sc)
1651 {
1652 
1653 	if (!(sc->bge_flags & BGEF_RXRING_VALID))
1654 		return;
1655 
1656 	for (u_int i = 0; i < BGE_STD_RX_RING_CNT; i++) {
1657 		const bus_dmamap_t dmap = sc->bge_cdata.bge_rx_std_map[i];
1658 		struct mbuf * const m = sc->bge_cdata.bge_rx_std_chain[i];
1659 		if (m != NULL) {
1660 			bus_dmamap_sync(sc->bge_dmatag, dmap, 0,
1661 			    dmap->dm_mapsize, BUS_DMASYNC_POSTREAD);
1662 			bus_dmamap_unload(sc->bge_dmatag, dmap);
1663 			m_freem(m);
1664 			sc->bge_cdata.bge_rx_std_chain[i] = NULL;
1665 		}
1666 		bus_dmamap_destroy(sc->bge_dmatag,
1667 		    sc->bge_cdata.bge_rx_std_map[i]);
1668 		sc->bge_cdata.bge_rx_std_map[i] = NULL;
1669 		memset((char *)&sc->bge_rdata->bge_rx_std_ring[i], 0,
1670 		    sizeof(struct bge_rx_bd));
1671 	}
1672 
1673 	sc->bge_flags &= ~BGEF_RXRING_VALID;
1674 }
1675 
1676 static int
1677 bge_init_rx_ring_jumbo(struct bge_softc *sc)
1678 {
1679 	int i;
1680 	volatile struct bge_rcb *rcb;
1681 
1682 	if (sc->bge_flags & BGEF_JUMBO_RXRING_VALID)
1683 		return 0;
1684 
1685 	for (i = 0; i < BGE_JUMBO_RX_RING_CNT; i++) {
1686 		if (bge_newbuf_jumbo(sc, i, NULL) == ENOBUFS)
1687 			return ENOBUFS;
1688 	}
1689 
1690 	sc->bge_jumbo = i - 1;
1691 	sc->bge_flags |= BGEF_JUMBO_RXRING_VALID;
1692 
1693 	rcb = &sc->bge_rdata->bge_info.bge_jumbo_rx_rcb;
1694 	rcb->bge_maxlen_flags = 0;
1695 	CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_MAXLEN_FLAGS, rcb->bge_maxlen_flags);
1696 
1697 	bge_writembx(sc, BGE_MBX_RX_JUMBO_PROD_LO, sc->bge_jumbo);
1698 
1699 	return 0;
1700 }
1701 
1702 static void
1703 bge_free_rx_ring_jumbo(struct bge_softc *sc)
1704 {
1705 	int i;
1706 
1707 	if (!(sc->bge_flags & BGEF_JUMBO_RXRING_VALID))
1708 		return;
1709 
1710 	for (i = 0; i < BGE_JUMBO_RX_RING_CNT; i++) {
1711 		m_freem(sc->bge_cdata.bge_rx_jumbo_chain[i]);
1712 		sc->bge_cdata.bge_rx_jumbo_chain[i] = NULL;
1713 		memset((char *)&sc->bge_rdata->bge_rx_jumbo_ring[i], 0,
1714 		    sizeof(struct bge_rx_bd));
1715 	}
1716 
1717 	sc->bge_flags &= ~BGEF_JUMBO_RXRING_VALID;
1718 }
1719 
1720 static void
1721 bge_free_tx_ring(struct bge_softc *sc, bool disable)
1722 {
1723 	int i;
1724 	struct txdmamap_pool_entry *dma;
1725 
1726 	if (!(sc->bge_flags & BGEF_TXRING_VALID))
1727 		return;
1728 
1729 	for (i = 0; i < BGE_TX_RING_CNT; i++) {
1730 		if (sc->bge_cdata.bge_tx_chain[i] != NULL) {
1731 			m_freem(sc->bge_cdata.bge_tx_chain[i]);
1732 			sc->bge_cdata.bge_tx_chain[i] = NULL;
1733 			SLIST_INSERT_HEAD(&sc->txdma_list, sc->txdma[i],
1734 					    link);
1735 			sc->txdma[i] = 0;
1736 		}
1737 		memset((char *)&sc->bge_rdata->bge_tx_ring[i], 0,
1738 		    sizeof(struct bge_tx_bd));
1739 	}
1740 
1741 	if (disable) {
1742 		while ((dma = SLIST_FIRST(&sc->txdma_list))) {
1743 			SLIST_REMOVE_HEAD(&sc->txdma_list, link);
1744 			bus_dmamap_destroy(sc->bge_dmatag, dma->dmamap);
1745 			if (sc->bge_dma64) {
1746 				bus_dmamap_destroy(sc->bge_dmatag32,
1747 				    dma->dmamap32);
1748 			}
1749 			kmem_free(dma, sizeof(*dma));
1750 		}
1751 		SLIST_INIT(&sc->txdma_list);
1752 	}
1753 
1754 	sc->bge_flags &= ~BGEF_TXRING_VALID;
1755 }
1756 
1757 static int
1758 bge_init_tx_ring(struct bge_softc *sc)
1759 {
1760 	struct ifnet * const ifp = &sc->ethercom.ec_if;
1761 	int i;
1762 	bus_dmamap_t dmamap, dmamap32;
1763 	bus_size_t maxsegsz;
1764 	struct txdmamap_pool_entry *dma;
1765 
1766 	if (sc->bge_flags & BGEF_TXRING_VALID)
1767 		return 0;
1768 
1769 	sc->bge_txcnt = 0;
1770 	sc->bge_tx_saved_considx = 0;
1771 
1772 	/* Initialize transmit producer index for host-memory send ring. */
1773 	sc->bge_tx_prodidx = 0;
1774 	bge_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, sc->bge_tx_prodidx);
1775 	/* 5700 b2 errata */
1776 	if (BGE_CHIPREV(sc->bge_chipid) == BGE_CHIPREV_5700_BX)
1777 		bge_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, sc->bge_tx_prodidx);
1778 
1779 	/* NIC-memory send ring not used; initialize to zero. */
1780 	bge_writembx(sc, BGE_MBX_TX_NIC_PROD0_LO, 0);
1781 	/* 5700 b2 errata */
1782 	if (BGE_CHIPREV(sc->bge_chipid) == BGE_CHIPREV_5700_BX)
1783 		bge_writembx(sc, BGE_MBX_TX_NIC_PROD0_LO, 0);
1784 
1785 	/* Limit DMA segment size for some chips */
1786 	if ((BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM57766) &&
1787 	    (ifp->if_mtu <= ETHERMTU))
1788 		maxsegsz = 2048;
1789 	else if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5719)
1790 		maxsegsz = 4096;
1791 	else
1792 		maxsegsz = ETHER_MAX_LEN_JUMBO;
1793 
1794 	if (SLIST_FIRST(&sc->txdma_list) != NULL)
1795 		goto alloc_done;
1796 
1797 	for (i = 0; i < BGE_TX_RING_CNT; i++) {
1798 		if (bus_dmamap_create(sc->bge_dmatag, BGE_TXDMA_MAX,
1799 		    BGE_NTXSEG, maxsegsz, 0, BUS_DMA_WAITOK | BUS_DMA_ALLOCNOW,
1800 		    &dmamap))
1801 			return ENOBUFS;
1802 		if (dmamap == NULL)
1803 			panic("dmamap NULL in bge_init_tx_ring");
1804 		if (sc->bge_dma64) {
1805 			if (bus_dmamap_create(sc->bge_dmatag32, BGE_TXDMA_MAX,
1806 			    BGE_NTXSEG, maxsegsz, 0,
1807 			    BUS_DMA_WAITOK | BUS_DMA_ALLOCNOW,
1808 			    &dmamap32)) {
1809 				bus_dmamap_destroy(sc->bge_dmatag, dmamap);
1810 				return ENOBUFS;
1811 			}
1812 			if (dmamap32 == NULL)
1813 				panic("dmamap32 NULL in bge_init_tx_ring");
1814 		} else
1815 			dmamap32 = dmamap;
1816 		dma = kmem_alloc(sizeof(*dma), KM_NOSLEEP);
1817 		if (dma == NULL) {
1818 			aprint_error_dev(sc->bge_dev,
1819 			    "can't alloc txdmamap_pool_entry\n");
1820 			bus_dmamap_destroy(sc->bge_dmatag, dmamap);
1821 			if (sc->bge_dma64)
1822 				bus_dmamap_destroy(sc->bge_dmatag32, dmamap32);
1823 			return ENOMEM;
1824 		}
1825 		dma->dmamap = dmamap;
1826 		dma->dmamap32 = dmamap32;
1827 		SLIST_INSERT_HEAD(&sc->txdma_list, dma, link);
1828 	}
1829 alloc_done:
1830 	sc->bge_flags |= BGEF_TXRING_VALID;
1831 
1832 	return 0;
1833 }
1834 
1835 static void
1836 bge_setmulti(struct bge_softc *sc)
1837 {
1838 	struct ethercom * const ec = &sc->ethercom;
1839 	struct ether_multi	*enm;
1840 	struct ether_multistep	step;
1841 	uint32_t		hashes[4] = { 0, 0, 0, 0 };
1842 	uint32_t		h;
1843 	int			i;
1844 
1845 	KASSERT(mutex_owned(sc->sc_mcast_lock));
1846 	if (sc->bge_if_flags & IFF_PROMISC)
1847 		goto allmulti;
1848 
1849 	/* Now program new ones. */
1850 	ETHER_LOCK(ec);
1851 	ETHER_FIRST_MULTI(step, ec, enm);
1852 	while (enm != NULL) {
1853 		if (memcmp(enm->enm_addrlo, enm->enm_addrhi, ETHER_ADDR_LEN)) {
1854 			/*
1855 			 * We must listen to a range of multicast addresses.
1856 			 * For now, just accept all multicasts, rather than
1857 			 * trying to set only those filter bits needed to match
1858 			 * the range.  (At this time, the only use of address
1859 			 * ranges is for IP multicast routing, for which the
1860 			 * range is big enough to require all bits set.)
1861 			 */
1862 			ETHER_UNLOCK(ec);
1863 			goto allmulti;
1864 		}
1865 
1866 		h = ether_crc32_le(enm->enm_addrlo, ETHER_ADDR_LEN);
1867 
1868 		/* Just want the 7 least-significant bits. */
1869 		h &= 0x7f;
1870 
1871 		hashes[(h & 0x60) >> 5] |= 1U << (h & 0x1F);
1872 		ETHER_NEXT_MULTI(step, enm);
1873 	}
1874 	ec->ec_flags &= ~ETHER_F_ALLMULTI;
1875 	ETHER_UNLOCK(ec);
1876 
1877 	goto setit;
1878 
1879  allmulti:
1880 	ETHER_LOCK(ec);
1881 	ec->ec_flags |= ETHER_F_ALLMULTI;
1882 	ETHER_UNLOCK(ec);
1883 	hashes[0] = hashes[1] = hashes[2] = hashes[3] = 0xffffffff;
1884 
1885  setit:
1886 	for (i = 0; i < 4; i++)
1887 		CSR_WRITE_4(sc, BGE_MAR0 + (i * 4), hashes[i]);
1888 }
1889 
1890 static void
1891 bge_sig_pre_reset(struct bge_softc *sc, int type)
1892 {
1893 
1894 	/*
1895 	 * Some chips don't like this so only do this if ASF is enabled
1896 	 */
1897 	if (sc->bge_asf_mode)
1898 		bge_writemem_ind(sc, BGE_SRAM_FW_MB, BGE_SRAM_FW_MB_MAGIC);
1899 
1900 	if (sc->bge_asf_mode & ASF_NEW_HANDSHAKE) {
1901 		switch (type) {
1902 		case BGE_RESET_START:
1903 			bge_writemem_ind(sc, BGE_SRAM_FW_DRV_STATE_MB,
1904 			    BGE_FW_DRV_STATE_START);
1905 			break;
1906 		case BGE_RESET_SHUTDOWN:
1907 			bge_writemem_ind(sc, BGE_SRAM_FW_DRV_STATE_MB,
1908 			    BGE_FW_DRV_STATE_UNLOAD);
1909 			break;
1910 		case BGE_RESET_SUSPEND:
1911 			bge_writemem_ind(sc, BGE_SRAM_FW_DRV_STATE_MB,
1912 			    BGE_FW_DRV_STATE_SUSPEND);
1913 			break;
1914 		}
1915 	}
1916 
1917 	if (type == BGE_RESET_START || type == BGE_RESET_SUSPEND)
1918 		bge_ape_driver_state_change(sc, type);
1919 }
1920 
1921 static void
1922 bge_sig_post_reset(struct bge_softc *sc, int type)
1923 {
1924 
1925 	if (sc->bge_asf_mode & ASF_NEW_HANDSHAKE) {
1926 		switch (type) {
1927 		case BGE_RESET_START:
1928 			bge_writemem_ind(sc, BGE_SRAM_FW_DRV_STATE_MB,
1929 			    BGE_FW_DRV_STATE_START_DONE);
1930 			/* START DONE */
1931 			break;
1932 		case BGE_RESET_SHUTDOWN:
1933 			bge_writemem_ind(sc, BGE_SRAM_FW_DRV_STATE_MB,
1934 			    BGE_FW_DRV_STATE_UNLOAD_DONE);
1935 			break;
1936 		}
1937 	}
1938 
1939 	if (type == BGE_RESET_SHUTDOWN)
1940 		bge_ape_driver_state_change(sc, type);
1941 }
1942 
1943 static void
1944 bge_sig_legacy(struct bge_softc *sc, int type)
1945 {
1946 
1947 	if (sc->bge_asf_mode) {
1948 		switch (type) {
1949 		case BGE_RESET_START:
1950 			bge_writemem_ind(sc, BGE_SRAM_FW_DRV_STATE_MB,
1951 			    BGE_FW_DRV_STATE_START);
1952 			break;
1953 		case BGE_RESET_SHUTDOWN:
1954 			bge_writemem_ind(sc, BGE_SRAM_FW_DRV_STATE_MB,
1955 			    BGE_FW_DRV_STATE_UNLOAD);
1956 			break;
1957 		}
1958 	}
1959 }
1960 
1961 static void
1962 bge_wait_for_event_ack(struct bge_softc *sc)
1963 {
1964 	int i;
1965 
1966 	/* wait up to 2500usec */
1967 	for (i = 0; i < 250; i++) {
1968 		if (!(CSR_READ_4(sc, BGE_RX_CPU_EVENT) &
1969 			BGE_RX_CPU_DRV_EVENT))
1970 			break;
1971 		DELAY(10);
1972 	}
1973 }
1974 
1975 static void
1976 bge_stop_fw(struct bge_softc *sc)
1977 {
1978 
1979 	if (sc->bge_asf_mode) {
1980 		bge_wait_for_event_ack(sc);
1981 
1982 		bge_writemem_ind(sc, BGE_SRAM_FW_CMD_MB, BGE_FW_CMD_PAUSE);
1983 		CSR_WRITE_4_FLUSH(sc, BGE_RX_CPU_EVENT,
1984 		    CSR_READ_4(sc, BGE_RX_CPU_EVENT) | BGE_RX_CPU_DRV_EVENT);
1985 
1986 		bge_wait_for_event_ack(sc);
1987 	}
1988 }
1989 
1990 static int
1991 bge_poll_fw(struct bge_softc *sc)
1992 {
1993 	uint32_t val;
1994 	int i;
1995 
1996 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906) {
1997 		for (i = 0; i < BGE_TIMEOUT; i++) {
1998 			val = CSR_READ_4(sc, BGE_VCPU_STATUS);
1999 			if (val & BGE_VCPU_STATUS_INIT_DONE)
2000 				break;
2001 			DELAY(100);
2002 		}
2003 		if (i >= BGE_TIMEOUT) {
2004 			aprint_error_dev(sc->bge_dev, "reset timed out\n");
2005 			return -1;
2006 		}
2007 	} else {
2008 		/*
2009 		 * Poll the value location we just wrote until
2010 		 * we see the 1's complement of the magic number.
2011 		 * This indicates that the firmware initialization
2012 		 * is complete.
2013 		 * XXX 1000ms for Flash and 10000ms for SEEPROM.
2014 		 */
2015 		for (i = 0; i < BGE_TIMEOUT; i++) {
2016 			val = bge_readmem_ind(sc, BGE_SRAM_FW_MB);
2017 			if (val == ~BGE_SRAM_FW_MB_MAGIC)
2018 				break;
2019 			DELAY(10);
2020 		}
2021 
2022 		if ((i >= BGE_TIMEOUT)
2023 		    && ((sc->bge_flags & BGEF_NO_EEPROM) == 0)) {
2024 			aprint_error_dev(sc->bge_dev,
2025 			    "firmware handshake timed out, val = %x\n", val);
2026 			return -1;
2027 		}
2028 	}
2029 
2030 	if (sc->bge_chipid == BGE_CHIPID_BCM57765_A0) {
2031 		/* tg3 says we have to wait extra time */
2032 		delay(10 * 1000);
2033 	}
2034 
2035 	return 0;
2036 }
2037 
2038 int
2039 bge_phy_addr(struct bge_softc *sc)
2040 {
2041 	struct pci_attach_args *pa = &(sc->bge_pa);
2042 	int phy_addr = 1;
2043 
2044 	/*
2045 	 * PHY address mapping for various devices.
2046 	 *
2047 	 *	    | F0 Cu | F0 Sr | F1 Cu | F1 Sr |
2048 	 * ---------+-------+-------+-------+-------+
2049 	 * BCM57XX  |	1   |	X   |	X   |	X   |
2050 	 * BCM5704  |	1   |	X   |	1   |	X   |
2051 	 * BCM5717  |	1   |	8   |	2   |	9   |
2052 	 * BCM5719  |	1   |	8   |	2   |	9   |
2053 	 * BCM5720  |	1   |	8   |	2   |	9   |
2054 	 *
2055 	 *	    | F2 Cu | F2 Sr | F3 Cu | F3 Sr |
2056 	 * ---------+-------+-------+-------+-------+
2057 	 * BCM57XX  |	X   |	X   |	X   |	X   |
2058 	 * BCM5704  |	X   |	X   |	X   |	X   |
2059 	 * BCM5717  |	X   |	X   |	X   |	X   |
2060 	 * BCM5719  |	3   |	10  |	4   |	11  |
2061 	 * BCM5720  |	X   |	X   |	X   |	X   |
2062 	 *
2063 	 * Other addresses may respond but they are not
2064 	 * IEEE compliant PHYs and should be ignored.
2065 	 */
2066 	switch (BGE_ASICREV(sc->bge_chipid)) {
2067 	case BGE_ASICREV_BCM5717:
2068 	case BGE_ASICREV_BCM5719:
2069 	case BGE_ASICREV_BCM5720:
2070 		phy_addr = pa->pa_function;
2071 		if (sc->bge_chipid != BGE_CHIPID_BCM5717_A0) {
2072 			phy_addr += (CSR_READ_4(sc, BGE_SGDIG_STS) &
2073 			    BGE_SGDIGSTS_IS_SERDES) ? 8 : 1;
2074 		} else {
2075 			phy_addr += (CSR_READ_4(sc, BGE_CPMU_PHY_STRAP) &
2076 			    BGE_CPMU_PHY_STRAP_IS_SERDES) ? 8 : 1;
2077 		}
2078 	}
2079 
2080 	return phy_addr;
2081 }
2082 
2083 /*
2084  * Do endian, PCI and DMA initialization. Also check the on-board ROM
2085  * self-test results.
2086  */
2087 static int
2088 bge_chipinit(struct bge_softc *sc)
2089 {
2090 	uint32_t dma_rw_ctl, misc_ctl, mode_ctl, reg;
2091 	int i;
2092 
2093 	/* Set endianness before we access any non-PCI registers. */
2094 	misc_ctl = BGE_INIT;
2095 	if (sc->bge_flags & BGEF_TAGGED_STATUS)
2096 		misc_ctl |= BGE_PCIMISCCTL_TAGGED_STATUS;
2097 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MISC_CTL,
2098 	    misc_ctl);
2099 
2100 	/*
2101 	 * Clear the MAC statistics block in the NIC's
2102 	 * internal memory.
2103 	 */
2104 	for (i = BGE_STATS_BLOCK;
2105 	    i < BGE_STATS_BLOCK_END + 1; i += sizeof(uint32_t))
2106 		BGE_MEMWIN_WRITE(sc->sc_pc, sc->sc_pcitag, i, 0);
2107 
2108 	for (i = BGE_STATUS_BLOCK;
2109 	    i < BGE_STATUS_BLOCK_END + 1; i += sizeof(uint32_t))
2110 		BGE_MEMWIN_WRITE(sc->sc_pc, sc->sc_pcitag, i, 0);
2111 
2112 	/* 5717 workaround from tg3 */
2113 	if (sc->bge_chipid == BGE_CHIPID_BCM5717_A0) {
2114 		/* Save */
2115 		mode_ctl = CSR_READ_4(sc, BGE_MODE_CTL);
2116 
2117 		/* Temporary modify MODE_CTL to control TLP */
2118 		reg = mode_ctl & ~BGE_MODECTL_PCIE_TLPADDRMASK;
2119 		CSR_WRITE_4(sc, BGE_MODE_CTL, reg | BGE_MODECTL_PCIE_TLPADDR1);
2120 
2121 		/* Control TLP */
2122 		reg = CSR_READ_4(sc, BGE_TLP_CONTROL_REG +
2123 		    BGE_TLP_PHYCTL1);
2124 		CSR_WRITE_4(sc, BGE_TLP_CONTROL_REG + BGE_TLP_PHYCTL1,
2125 		    reg | BGE_TLP_PHYCTL1_EN_L1PLLPD);
2126 
2127 		/* Restore */
2128 		CSR_WRITE_4(sc, BGE_MODE_CTL, mode_ctl);
2129 	}
2130 
2131 	if (BGE_IS_57765_FAMILY(sc)) {
2132 		if (sc->bge_chipid == BGE_CHIPID_BCM57765_A0) {
2133 			/* Save */
2134 			mode_ctl = CSR_READ_4(sc, BGE_MODE_CTL);
2135 
2136 			/* Temporary modify MODE_CTL to control TLP */
2137 			reg = mode_ctl & ~BGE_MODECTL_PCIE_TLPADDRMASK;
2138 			CSR_WRITE_4(sc, BGE_MODE_CTL,
2139 			    reg | BGE_MODECTL_PCIE_TLPADDR1);
2140 
2141 			/* Control TLP */
2142 			reg = CSR_READ_4(sc, BGE_TLP_CONTROL_REG +
2143 			    BGE_TLP_PHYCTL5);
2144 			CSR_WRITE_4(sc, BGE_TLP_CONTROL_REG + BGE_TLP_PHYCTL5,
2145 			    reg | BGE_TLP_PHYCTL5_DIS_L2CLKREQ);
2146 
2147 			/* Restore */
2148 			CSR_WRITE_4(sc, BGE_MODE_CTL, mode_ctl);
2149 		}
2150 		if (BGE_CHIPREV(sc->bge_chipid) != BGE_CHIPREV_57765_AX) {
2151 			/*
2152 			 * For the 57766 and non Ax versions of 57765, bootcode
2153 			 * needs to setup the PCIE Fast Training Sequence (FTS)
2154 			 * value to prevent transmit hangs.
2155 			 */
2156 			reg = CSR_READ_4(sc, BGE_CPMU_PADRNG_CTL);
2157 			CSR_WRITE_4(sc, BGE_CPMU_PADRNG_CTL,
2158 			    reg | BGE_CPMU_PADRNG_CTL_RDIV2);
2159 
2160 			/* Save */
2161 			mode_ctl = CSR_READ_4(sc, BGE_MODE_CTL);
2162 
2163 			/* Temporary modify MODE_CTL to control TLP */
2164 			reg = mode_ctl & ~BGE_MODECTL_PCIE_TLPADDRMASK;
2165 			CSR_WRITE_4(sc, BGE_MODE_CTL,
2166 			    reg | BGE_MODECTL_PCIE_TLPADDR0);
2167 
2168 			/* Control TLP */
2169 			reg = CSR_READ_4(sc, BGE_TLP_CONTROL_REG +
2170 			    BGE_TLP_FTSMAX);
2171 			reg &= ~BGE_TLP_FTSMAX_MSK;
2172 			CSR_WRITE_4(sc, BGE_TLP_CONTROL_REG + BGE_TLP_FTSMAX,
2173 			    reg | BGE_TLP_FTSMAX_VAL);
2174 
2175 			/* Restore */
2176 			CSR_WRITE_4(sc, BGE_MODE_CTL, mode_ctl);
2177 		}
2178 
2179 		reg = CSR_READ_4(sc, BGE_CPMU_LSPD_10MB_CLK);
2180 		reg &= ~BGE_CPMU_LSPD_10MB_MACCLK_MASK;
2181 		reg |= BGE_CPMU_LSPD_10MB_MACCLK_6_25;
2182 		CSR_WRITE_4(sc, BGE_CPMU_LSPD_10MB_CLK, reg);
2183 	}
2184 
2185 	/* Set up the PCI DMA control register. */
2186 	dma_rw_ctl = BGE_PCI_READ_CMD | BGE_PCI_WRITE_CMD;
2187 	if (sc->bge_flags & BGEF_PCIE) {
2188 		/* Read watermark not used, 128 bytes for write. */
2189 		DPRINTFN(4, ("(%s: PCI-Express DMA setting)\n",
2190 		    device_xname(sc->bge_dev)));
2191 		if (sc->bge_mps >= 256)
2192 			dma_rw_ctl |= BGE_PCIDMARWCTL_WR_WAT_SHIFT(7);
2193 		else
2194 			dma_rw_ctl |= BGE_PCIDMARWCTL_WR_WAT_SHIFT(3);
2195 	} else if (sc->bge_flags & BGEF_PCIX) {
2196 		DPRINTFN(4, ("(:%s: PCI-X DMA setting)\n",
2197 		    device_xname(sc->bge_dev)));
2198 		/* PCI-X bus */
2199 		if (BGE_IS_5714_FAMILY(sc)) {
2200 			/* 256 bytes for read and write. */
2201 			dma_rw_ctl |= BGE_PCIDMARWCTL_RD_WAT_SHIFT(2) |
2202 			    BGE_PCIDMARWCTL_WR_WAT_SHIFT(2);
2203 
2204 			if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5780)
2205 				dma_rw_ctl |= BGE_PCIDMARWCTL_ONEDMA_ATONCE_GLOBAL;
2206 			else
2207 				dma_rw_ctl |= BGE_PCIDMARWCTL_ONEDMA_ATONCE_LOCAL;
2208 		} else if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5703) {
2209 			/*
2210 			 * In the BCM5703, the DMA read watermark should
2211 			 * be set to less than or equal to the maximum
2212 			 * memory read byte count of the PCI-X command
2213 			 * register.
2214 			 */
2215 			dma_rw_ctl |= BGE_PCIDMARWCTL_RD_WAT_SHIFT(4) |
2216 			    BGE_PCIDMARWCTL_WR_WAT_SHIFT(3);
2217 		} else if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5704) {
2218 			/* 1536 bytes for read, 384 bytes for write. */
2219 			dma_rw_ctl |= BGE_PCIDMARWCTL_RD_WAT_SHIFT(7) |
2220 			    BGE_PCIDMARWCTL_WR_WAT_SHIFT(3);
2221 		} else {
2222 			/* 384 bytes for read and write. */
2223 			dma_rw_ctl |= BGE_PCIDMARWCTL_RD_WAT_SHIFT(3) |
2224 			    BGE_PCIDMARWCTL_WR_WAT_SHIFT(3) |
2225 			    (0x0F);
2226 		}
2227 
2228 		if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5703 ||
2229 		    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5704) {
2230 			uint32_t tmp;
2231 
2232 			/* Set ONEDMA_ATONCE for hardware workaround. */
2233 			tmp = CSR_READ_4(sc, BGE_PCI_CLKCTL) & 0x1f;
2234 			if (tmp == 6 || tmp == 7)
2235 				dma_rw_ctl |=
2236 				    BGE_PCIDMARWCTL_ONEDMA_ATONCE_GLOBAL;
2237 
2238 			/* Set PCI-X DMA write workaround. */
2239 			dma_rw_ctl |= BGE_PCIDMARWCTL_ASRT_ALL_BE;
2240 		}
2241 	} else {
2242 		/* Conventional PCI bus: 256 bytes for read and write. */
2243 		DPRINTFN(4, ("(%s: PCI 2.2 DMA setting)\n",
2244 		    device_xname(sc->bge_dev)));
2245 		dma_rw_ctl |= BGE_PCIDMARWCTL_RD_WAT_SHIFT(7) |
2246 		    BGE_PCIDMARWCTL_WR_WAT_SHIFT(7);
2247 
2248 		if (BGE_ASICREV(sc->bge_chipid) != BGE_ASICREV_BCM5705 &&
2249 		    BGE_ASICREV(sc->bge_chipid) != BGE_ASICREV_BCM5750)
2250 			dma_rw_ctl |= 0x0F;
2251 	}
2252 
2253 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5700 ||
2254 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5701)
2255 		dma_rw_ctl |= BGE_PCIDMARWCTL_USE_MRM |
2256 		    BGE_PCIDMARWCTL_ASRT_ALL_BE;
2257 
2258 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5703 ||
2259 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5704)
2260 		dma_rw_ctl &= ~BGE_PCIDMARWCTL_MINDMA;
2261 
2262 	if (BGE_IS_57765_PLUS(sc)) {
2263 		dma_rw_ctl &= ~BGE_PCIDMARWCTL_DIS_CACHE_ALIGNMENT;
2264 		if (sc->bge_chipid == BGE_CHIPID_BCM57765_A0)
2265 			dma_rw_ctl &= ~BGE_PCIDMARWCTL_CRDRDR_RDMA_MRRS_MSK;
2266 
2267 		/*
2268 		 * Enable HW workaround for controllers that misinterpret
2269 		 * a status tag update and leave interrupts permanently
2270 		 * disabled.
2271 		 */
2272 		if (!BGE_IS_57765_FAMILY(sc) &&
2273 		    BGE_ASICREV(sc->bge_chipid) != BGE_ASICREV_BCM5717 &&
2274 		    BGE_ASICREV(sc->bge_chipid) != BGE_ASICREV_BCM5762)
2275 			dma_rw_ctl |= BGE_PCIDMARWCTL_TAGGED_STATUS_WA;
2276 	}
2277 
2278 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_DMA_RW_CTL,
2279 	    dma_rw_ctl);
2280 
2281 	/*
2282 	 * Set up general mode register.
2283 	 */
2284 	mode_ctl = BGE_DMA_SWAP_OPTIONS;
2285 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5720 ||
2286 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5762) {
2287 		/* Retain Host-2-BMC settings written by APE firmware. */
2288 		mode_ctl |= CSR_READ_4(sc, BGE_MODE_CTL) &
2289 		    (BGE_MODECTL_BYTESWAP_B2HRX_DATA |
2290 		    BGE_MODECTL_WORDSWAP_B2HRX_DATA |
2291 		    BGE_MODECTL_B2HRX_ENABLE | BGE_MODECTL_HTX2B_ENABLE);
2292 	}
2293 	mode_ctl |= BGE_MODECTL_MAC_ATTN_INTR | BGE_MODECTL_HOST_SEND_BDS |
2294 	    BGE_MODECTL_TX_NO_PHDR_CSUM;
2295 
2296 	/*
2297 	 * BCM5701 B5 have a bug causing data corruption when using
2298 	 * 64-bit DMA reads, which can be terminated early and then
2299 	 * completed later as 32-bit accesses, in combination with
2300 	 * certain bridges.
2301 	 */
2302 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5701 &&
2303 	    sc->bge_chipid == BGE_CHIPID_BCM5701_B5)
2304 		mode_ctl |= BGE_MODECTL_FORCE_PCI32;
2305 
2306 	/*
2307 	 * Tell the firmware the driver is running
2308 	 */
2309 	if (sc->bge_asf_mode & ASF_STACKUP)
2310 		mode_ctl |= BGE_MODECTL_STACKUP;
2311 
2312 	CSR_WRITE_4(sc, BGE_MODE_CTL, mode_ctl);
2313 
2314 	/*
2315 	 * Disable memory write invalidate.  Apparently it is not supported
2316 	 * properly by these devices.
2317 	 */
2318 	PCI_CLRBIT(sc->sc_pc, sc->sc_pcitag, PCI_COMMAND_STATUS_REG,
2319 		   PCI_COMMAND_INVALIDATE_ENABLE);
2320 
2321 #ifdef __brokenalpha__
2322 	/*
2323 	 * Must insure that we do not cross an 8K (bytes) boundary
2324 	 * for DMA reads.  Our highest limit is 1K bytes.  This is a
2325 	 * restriction on some ALPHA platforms with early revision
2326 	 * 21174 PCI chipsets, such as the AlphaPC 164lx
2327 	 */
2328 	PCI_SETBIT(sc, BGE_PCI_DMA_RW_CTL, BGE_PCI_READ_BNDRY_1024, 4);
2329 #endif
2330 
2331 	/* Set the timer prescaler (always 66MHz) */
2332 	CSR_WRITE_4_FLUSH(sc, BGE_MISC_CFG, BGE_32BITTIME_66MHZ);
2333 
2334 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906) {
2335 		DELAY(40);	/* XXX */
2336 
2337 		/* Put PHY into ready state */
2338 		BGE_CLRBIT_FLUSH(sc, BGE_MISC_CFG, BGE_MISCCFG_EPHY_IDDQ);
2339 		DELAY(40);
2340 	}
2341 
2342 	return 0;
2343 }
2344 
2345 static int
2346 bge_blockinit(struct bge_softc *sc)
2347 {
2348 	volatile struct bge_rcb	 *rcb;
2349 	bus_size_t rcb_addr;
2350 	struct ifnet * const ifp = &sc->ethercom.ec_if;
2351 	bge_hostaddr taddr;
2352 	uint32_t	dmactl, rdmareg, mimode, val;
2353 	int		i, limit;
2354 
2355 	/*
2356 	 * Initialize the memory window pointer register so that
2357 	 * we can access the first 32K of internal NIC RAM. This will
2358 	 * allow us to set up the TX send ring RCBs and the RX return
2359 	 * ring RCBs, plus other things which live in NIC memory.
2360 	 */
2361 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MEMWIN_BASEADDR, 0);
2362 
2363 	if (!BGE_IS_5705_PLUS(sc)) {
2364 		/* 57XX step 33 */
2365 		/* Configure mbuf memory pool */
2366 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_BASEADDR, BGE_BUFFPOOL_1);
2367 
2368 		if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5704)
2369 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_LEN, 0x10000);
2370 		else
2371 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_LEN, 0x18000);
2372 
2373 		/* 57XX step 34 */
2374 		/* Configure DMA resource pool */
2375 		CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_BASEADDR,
2376 		    BGE_DMA_DESCRIPTORS);
2377 		CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_LEN, 0x2000);
2378 	}
2379 
2380 	/* 5718 step 11, 57XX step 35 */
2381 	/*
2382 	 * Configure mbuf pool watermarks. New broadcom docs strongly
2383 	 * recommend these.
2384 	 */
2385 	if (BGE_IS_5717_PLUS(sc)) {
2386 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 0x0);
2387 		if (ifp->if_mtu > ETHERMTU) {
2388 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x7e);
2389 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0xea);
2390 		} else {
2391 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x2a);
2392 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0xa0);
2393 		}
2394 	} else if (BGE_IS_5705_PLUS(sc)) {
2395 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 0x0);
2396 
2397 		if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906) {
2398 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x04);
2399 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0x10);
2400 		} else {
2401 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x10);
2402 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0x60);
2403 		}
2404 	} else {
2405 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 0x50);
2406 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x20);
2407 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0x60);
2408 	}
2409 
2410 	/* 57XX step 36 */
2411 	/* Configure DMA resource watermarks */
2412 	CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_LOWAT, 5);
2413 	CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_HIWAT, 10);
2414 
2415 	/* 5718 step 13, 57XX step 38 */
2416 	/* Enable buffer manager */
2417 	val = BGE_BMANMODE_ENABLE | BGE_BMANMODE_ATTN;
2418 	/*
2419 	 * Change the arbitration algorithm of TXMBUF read request to
2420 	 * round-robin instead of priority based for BCM5719.  When
2421 	 * TXFIFO is almost empty, RDMA will hold its request until
2422 	 * TXFIFO is not almost empty.
2423 	 */
2424 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5719)
2425 		val |= BGE_BMANMODE_NO_TX_UNDERRUN;
2426 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5717 ||
2427 		sc->bge_chipid == BGE_CHIPID_BCM5719_A0 ||
2428 		sc->bge_chipid == BGE_CHIPID_BCM5720_A0)
2429 		val |= BGE_BMANMODE_LOMBUF_ATTN;
2430 	CSR_WRITE_4(sc, BGE_BMAN_MODE, val);
2431 
2432 	/* 57XX step 39 */
2433 	/* Poll for buffer manager start indication */
2434 	for (i = 0; i < BGE_TIMEOUT * 2; i++) {
2435 		DELAY(10);
2436 		if (CSR_READ_4(sc, BGE_BMAN_MODE) & BGE_BMANMODE_ENABLE)
2437 			break;
2438 	}
2439 
2440 	if (i == BGE_TIMEOUT * 2) {
2441 		aprint_error_dev(sc->bge_dev,
2442 		    "buffer manager failed to start\n");
2443 		return ENXIO;
2444 	}
2445 
2446 	/* 57XX step 40 */
2447 	/* Enable flow-through queues */
2448 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0xFFFFFFFF);
2449 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0);
2450 
2451 	/* Wait until queue initialization is complete */
2452 	for (i = 0; i < BGE_TIMEOUT * 2; i++) {
2453 		if (CSR_READ_4(sc, BGE_FTQ_RESET) == 0)
2454 			break;
2455 		DELAY(10);
2456 	}
2457 
2458 	if (i == BGE_TIMEOUT * 2) {
2459 		aprint_error_dev(sc->bge_dev,
2460 		    "flow-through queue init failed\n");
2461 		return ENXIO;
2462 	}
2463 
2464 	/*
2465 	 * Summary of rings supported by the controller:
2466 	 *
2467 	 * Standard Receive Producer Ring
2468 	 * - This ring is used to feed receive buffers for "standard"
2469 	 *   sized frames (typically 1536 bytes) to the controller.
2470 	 *
2471 	 * Jumbo Receive Producer Ring
2472 	 * - This ring is used to feed receive buffers for jumbo sized
2473 	 *   frames (i.e. anything bigger than the "standard" frames)
2474 	 *   to the controller.
2475 	 *
2476 	 * Mini Receive Producer Ring
2477 	 * - This ring is used to feed receive buffers for "mini"
2478 	 *   sized frames to the controller.
2479 	 * - This feature required external memory for the controller
2480 	 *   but was never used in a production system.  Should always
2481 	 *   be disabled.
2482 	 *
2483 	 * Receive Return Ring
2484 	 * - After the controller has placed an incoming frame into a
2485 	 *   receive buffer that buffer is moved into a receive return
2486 	 *   ring.  The driver is then responsible to passing the
2487 	 *   buffer up to the stack.  Many versions of the controller
2488 	 *   support multiple RR rings.
2489 	 *
2490 	 * Send Ring
2491 	 * - This ring is used for outgoing frames.  Many versions of
2492 	 *   the controller support multiple send rings.
2493 	 */
2494 
2495 	/* 5718 step 15, 57XX step 41 */
2496 	/* Initialize the standard RX ring control block */
2497 	rcb = &sc->bge_rdata->bge_info.bge_std_rx_rcb;
2498 	BGE_HOSTADDR(rcb->bge_hostaddr, BGE_RING_DMA_ADDR(sc, bge_rx_std_ring));
2499 	/* 5718 step 16 */
2500 	if (BGE_IS_57765_PLUS(sc)) {
2501 		/*
2502 		 * Bits 31-16: Programmable ring size (2048, 1024, 512, .., 32)
2503 		 * Bits 15-2 : Maximum RX frame size
2504 		 * Bit 1     : 1 = Ring Disabled, 0 = Ring Enabled
2505 		 * Bit 0     : Reserved
2506 		 */
2507 		rcb->bge_maxlen_flags =
2508 		    BGE_RCB_MAXLEN_FLAGS(512, BGE_MAX_FRAMELEN << 2);
2509 	} else if (BGE_IS_5705_PLUS(sc)) {
2510 		/*
2511 		 * Bits 31-16: Programmable ring size (512, 256, 128, 64, 32)
2512 		 * Bits 15-2 : Reserved (should be 0)
2513 		 * Bit 1     : 1 = Ring Disabled, 0 = Ring Enabled
2514 		 * Bit 0     : Reserved
2515 		 */
2516 		rcb->bge_maxlen_flags = BGE_RCB_MAXLEN_FLAGS(512, 0);
2517 	} else {
2518 		/*
2519 		 * Ring size is always XXX entries
2520 		 * Bits 31-16: Maximum RX frame size
2521 		 * Bits 15-2 : Reserved (should be 0)
2522 		 * Bit 1     : 1 = Ring Disabled, 0 = Ring Enabled
2523 		 * Bit 0     : Reserved
2524 		 */
2525 		rcb->bge_maxlen_flags =
2526 		    BGE_RCB_MAXLEN_FLAGS(BGE_MAX_FRAMELEN, 0);
2527 	}
2528 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5717 ||
2529 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5719 ||
2530 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5720)
2531 		rcb->bge_nicaddr = BGE_STD_RX_RINGS_5717;
2532 	else
2533 		rcb->bge_nicaddr = BGE_STD_RX_RINGS;
2534 	/* Write the standard receive producer ring control block. */
2535 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_HADDR_HI, rcb->bge_hostaddr.bge_addr_hi);
2536 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_HADDR_LO, rcb->bge_hostaddr.bge_addr_lo);
2537 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_MAXLEN_FLAGS, rcb->bge_maxlen_flags);
2538 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_NICADDR, rcb->bge_nicaddr);
2539 
2540 	/* Reset the standard receive producer ring producer index. */
2541 	bge_writembx(sc, BGE_MBX_RX_STD_PROD_LO, 0);
2542 
2543 	/* 57XX step 42 */
2544 	/*
2545 	 * Initialize the jumbo RX ring control block
2546 	 * We set the 'ring disabled' bit in the flags
2547 	 * field until we're actually ready to start
2548 	 * using this ring (i.e. once we set the MTU
2549 	 * high enough to require it).
2550 	 */
2551 	if (BGE_IS_JUMBO_CAPABLE(sc)) {
2552 		rcb = &sc->bge_rdata->bge_info.bge_jumbo_rx_rcb;
2553 		BGE_HOSTADDR(rcb->bge_hostaddr,
2554 		    BGE_RING_DMA_ADDR(sc, bge_rx_jumbo_ring));
2555 		rcb->bge_maxlen_flags = BGE_RCB_MAXLEN_FLAGS(0,
2556 		    BGE_RCB_FLAG_USE_EXT_RX_BD | BGE_RCB_FLAG_RING_DISABLED);
2557 		if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5717 ||
2558 		    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5719 ||
2559 		    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5720)
2560 			rcb->bge_nicaddr = BGE_JUMBO_RX_RINGS_5717;
2561 		else
2562 			rcb->bge_nicaddr = BGE_JUMBO_RX_RINGS;
2563 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_HADDR_HI,
2564 		    rcb->bge_hostaddr.bge_addr_hi);
2565 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_HADDR_LO,
2566 		    rcb->bge_hostaddr.bge_addr_lo);
2567 		/* Program the jumbo receive producer ring RCB parameters. */
2568 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_MAXLEN_FLAGS,
2569 		    rcb->bge_maxlen_flags);
2570 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_NICADDR, rcb->bge_nicaddr);
2571 		/* Reset the jumbo receive producer ring producer index. */
2572 		bge_writembx(sc, BGE_MBX_RX_JUMBO_PROD_LO, 0);
2573 	}
2574 
2575 	/* 57XX step 43 */
2576 	/* Disable the mini receive producer ring RCB. */
2577 	if (BGE_IS_5700_FAMILY(sc)) {
2578 		/* Set up dummy disabled mini ring RCB */
2579 		rcb = &sc->bge_rdata->bge_info.bge_mini_rx_rcb;
2580 		rcb->bge_maxlen_flags =
2581 		    BGE_RCB_MAXLEN_FLAGS(0, BGE_RCB_FLAG_RING_DISABLED);
2582 		CSR_WRITE_4(sc, BGE_RX_MINI_RCB_MAXLEN_FLAGS,
2583 		    rcb->bge_maxlen_flags);
2584 		/* Reset the mini receive producer ring producer index. */
2585 		bge_writembx(sc, BGE_MBX_RX_MINI_PROD_LO, 0);
2586 
2587 		bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
2588 		    offsetof(struct bge_ring_data, bge_info),
2589 		    sizeof(struct bge_gib),
2590 		    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
2591 	}
2592 
2593 	/* Choose de-pipeline mode for BCM5906 A0, A1 and A2. */
2594 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906) {
2595 		if (sc->bge_chipid == BGE_CHIPID_BCM5906_A0 ||
2596 		    sc->bge_chipid == BGE_CHIPID_BCM5906_A1 ||
2597 		    sc->bge_chipid == BGE_CHIPID_BCM5906_A2)
2598 			CSR_WRITE_4(sc, BGE_ISO_PKT_TX,
2599 			    (CSR_READ_4(sc, BGE_ISO_PKT_TX) & ~3) | 2);
2600 	}
2601 	/* 5718 step 14, 57XX step 44 */
2602 	/*
2603 	 * The BD ring replenish thresholds control how often the
2604 	 * hardware fetches new BD's from the producer rings in host
2605 	 * memory.  Setting the value too low on a busy system can
2606 	 * starve the hardware and reduce the throughput.
2607 	 *
2608 	 * Set the BD ring replenish thresholds. The recommended
2609 	 * values are 1/8th the number of descriptors allocated to
2610 	 * each ring, but since we try to avoid filling the entire
2611 	 * ring we set these to the minimal value of 8.  This needs to
2612 	 * be done on several of the supported chip revisions anyway,
2613 	 * to work around HW bugs.
2614 	 */
2615 	CSR_WRITE_4(sc, BGE_RBDI_STD_REPL_THRESH, 8);
2616 	if (BGE_IS_JUMBO_CAPABLE(sc))
2617 		CSR_WRITE_4(sc, BGE_RBDI_JUMBO_REPL_THRESH, 8);
2618 
2619 	/* 5718 step 18 */
2620 	if (BGE_IS_5717_PLUS(sc)) {
2621 		CSR_WRITE_4(sc, BGE_STD_REPL_LWM, 4);
2622 		CSR_WRITE_4(sc, BGE_JUMBO_REPL_LWM, 4);
2623 	}
2624 
2625 	/* 57XX step 45 */
2626 	/*
2627 	 * Disable all send rings by setting the 'ring disabled' bit
2628 	 * in the flags field of all the TX send ring control blocks,
2629 	 * located in NIC memory.
2630 	 */
2631 	if (BGE_IS_5700_FAMILY(sc)) {
2632 		/* 5700 to 5704 had 16 send rings. */
2633 		limit = BGE_TX_RINGS_EXTSSRAM_MAX;
2634 	} else if (BGE_IS_5717_PLUS(sc)) {
2635 		limit = BGE_TX_RINGS_5717_MAX;
2636 	} else if (BGE_IS_57765_FAMILY(sc) ||
2637 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5762) {
2638 		limit = BGE_TX_RINGS_57765_MAX;
2639 	} else
2640 		limit = 1;
2641 	rcb_addr = BGE_MEMWIN_START + BGE_SEND_RING_RCB;
2642 	for (i = 0; i < limit; i++) {
2643 		RCB_WRITE_4(sc, rcb_addr, bge_maxlen_flags,
2644 		    BGE_RCB_MAXLEN_FLAGS(0, BGE_RCB_FLAG_RING_DISABLED));
2645 		RCB_WRITE_4(sc, rcb_addr, bge_nicaddr, 0);
2646 		rcb_addr += sizeof(struct bge_rcb);
2647 	}
2648 
2649 	/* 57XX step 46 and 47 */
2650 	/* Configure send ring RCB 0 (we use only the first ring) */
2651 	rcb_addr = BGE_MEMWIN_START + BGE_SEND_RING_RCB;
2652 	BGE_HOSTADDR(taddr, BGE_RING_DMA_ADDR(sc, bge_tx_ring));
2653 	RCB_WRITE_4(sc, rcb_addr, bge_hostaddr.bge_addr_hi, taddr.bge_addr_hi);
2654 	RCB_WRITE_4(sc, rcb_addr, bge_hostaddr.bge_addr_lo, taddr.bge_addr_lo);
2655 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5717 ||
2656 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5719 ||
2657 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5720)
2658 		RCB_WRITE_4(sc, rcb_addr, bge_nicaddr, BGE_SEND_RING_5717);
2659 	else
2660 		RCB_WRITE_4(sc, rcb_addr, bge_nicaddr,
2661 		    BGE_NIC_TXRING_ADDR(0, BGE_TX_RING_CNT));
2662 	RCB_WRITE_4(sc, rcb_addr, bge_maxlen_flags,
2663 	    BGE_RCB_MAXLEN_FLAGS(BGE_TX_RING_CNT, 0));
2664 
2665 	/* 57XX step 48 */
2666 	/*
2667 	 * Disable all receive return rings by setting the
2668 	 * 'ring disabled' bit in the flags field of all the receive
2669 	 * return ring control blocks, located in NIC memory.
2670 	 */
2671 	if (BGE_IS_5717_PLUS(sc)) {
2672 		/* Should be 17, use 16 until we get an SRAM map. */
2673 		limit = 16;
2674 	} else if (BGE_IS_5700_FAMILY(sc))
2675 		limit = BGE_RX_RINGS_MAX;
2676 	else if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5755 ||
2677 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5762 ||
2678 	    BGE_IS_57765_FAMILY(sc))
2679 		limit = 4;
2680 	else
2681 		limit = 1;
2682 	/* Disable all receive return rings */
2683 	rcb_addr = BGE_MEMWIN_START + BGE_RX_RETURN_RING_RCB;
2684 	for (i = 0; i < limit; i++) {
2685 		RCB_WRITE_4(sc, rcb_addr, bge_hostaddr.bge_addr_hi, 0);
2686 		RCB_WRITE_4(sc, rcb_addr, bge_hostaddr.bge_addr_lo, 0);
2687 		RCB_WRITE_4(sc, rcb_addr, bge_maxlen_flags,
2688 		    BGE_RCB_MAXLEN_FLAGS(sc->bge_return_ring_cnt,
2689 			BGE_RCB_FLAG_RING_DISABLED));
2690 		RCB_WRITE_4(sc, rcb_addr, bge_nicaddr, 0);
2691 		bge_writembx(sc, BGE_MBX_RX_CONS0_LO +
2692 		    (i * (sizeof(uint64_t))), 0);
2693 		rcb_addr += sizeof(struct bge_rcb);
2694 	}
2695 
2696 	/* 57XX step 49 */
2697 	/*
2698 	 * Set up receive return ring 0.  Note that the NIC address
2699 	 * for RX return rings is 0x0.  The return rings live entirely
2700 	 * within the host, so the nicaddr field in the RCB isn't used.
2701 	 */
2702 	rcb_addr = BGE_MEMWIN_START + BGE_RX_RETURN_RING_RCB;
2703 	BGE_HOSTADDR(taddr, BGE_RING_DMA_ADDR(sc, bge_rx_return_ring));
2704 	RCB_WRITE_4(sc, rcb_addr, bge_hostaddr.bge_addr_hi, taddr.bge_addr_hi);
2705 	RCB_WRITE_4(sc, rcb_addr, bge_hostaddr.bge_addr_lo, taddr.bge_addr_lo);
2706 	RCB_WRITE_4(sc, rcb_addr, bge_nicaddr, 0x00000000);
2707 	RCB_WRITE_4(sc, rcb_addr, bge_maxlen_flags,
2708 	    BGE_RCB_MAXLEN_FLAGS(sc->bge_return_ring_cnt, 0));
2709 
2710 	/* 5718 step 24, 57XX step 53 */
2711 	/* Set random backoff seed for TX */
2712 	CSR_WRITE_4(sc, BGE_TX_RANDOM_BACKOFF,
2713 	    (CLLADDR(ifp->if_sadl)[0] + CLLADDR(ifp->if_sadl)[1] +
2714 		CLLADDR(ifp->if_sadl)[2] + CLLADDR(ifp->if_sadl)[3] +
2715 		CLLADDR(ifp->if_sadl)[4] + CLLADDR(ifp->if_sadl)[5]) &
2716 	    BGE_TX_BACKOFF_SEED_MASK);
2717 
2718 	/* 5718 step 26, 57XX step 55 */
2719 	/* Set inter-packet gap */
2720 	val = 0x2620;
2721 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5720 ||
2722 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5762)
2723 		val |= CSR_READ_4(sc, BGE_TX_LENGTHS) &
2724 		    (BGE_TXLEN_JMB_FRM_LEN_MSK | BGE_TXLEN_CNT_DN_VAL_MSK);
2725 	CSR_WRITE_4(sc, BGE_TX_LENGTHS, val);
2726 
2727 	/* 5718 step 27, 57XX step 56 */
2728 	/*
2729 	 * Specify which ring to use for packets that don't match
2730 	 * any RX rules.
2731 	 */
2732 	CSR_WRITE_4(sc, BGE_RX_RULES_CFG, 0x08);
2733 
2734 	/* 5718 step 28, 57XX step 57 */
2735 	/*
2736 	 * Configure number of RX lists. One interrupt distribution
2737 	 * list, sixteen active lists, one bad frames class.
2738 	 */
2739 	CSR_WRITE_4(sc, BGE_RXLP_CFG, 0x181);
2740 
2741 	/* 5718 step 29, 57XX step 58 */
2742 	/* Initialize RX list placement stats mask. */
2743 	if (BGE_IS_575X_PLUS(sc)) {
2744 		val = CSR_READ_4(sc, BGE_RXLP_STATS_ENABLE_MASK);
2745 		val &= ~BGE_RXLPSTATCONTROL_DACK_FIX;
2746 		CSR_WRITE_4(sc, BGE_RXLP_STATS_ENABLE_MASK, val);
2747 	} else
2748 		CSR_WRITE_4(sc, BGE_RXLP_STATS_ENABLE_MASK, 0x007FFFFF);
2749 
2750 	/* 5718 step 30, 57XX step 59 */
2751 	CSR_WRITE_4(sc, BGE_RXLP_STATS_CTL, 0x1);
2752 
2753 	/* 5718 step 33, 57XX step 62 */
2754 	/* Disable host coalescing until we get it set up */
2755 	CSR_WRITE_4(sc, BGE_HCC_MODE, 0x00000000);
2756 
2757 	/* 5718 step 34, 57XX step 63 */
2758 	/* Poll to make sure it's shut down. */
2759 	for (i = 0; i < BGE_TIMEOUT * 2; i++) {
2760 		DELAY(10);
2761 		if (!(CSR_READ_4(sc, BGE_HCC_MODE) & BGE_HCCMODE_ENABLE))
2762 			break;
2763 	}
2764 
2765 	if (i == BGE_TIMEOUT * 2) {
2766 		aprint_error_dev(sc->bge_dev,
2767 		    "host coalescing engine failed to idle\n");
2768 		return ENXIO;
2769 	}
2770 
2771 	/* 5718 step 35, 36, 37 */
2772 	/* Set up host coalescing defaults */
2773 	mutex_enter(sc->sc_intr_lock);
2774 	const uint32_t rx_coal_ticks = sc->bge_rx_coal_ticks;
2775 	const uint32_t tx_coal_ticks = sc->bge_tx_coal_ticks;
2776 	const uint32_t rx_max_coal_bds = sc->bge_rx_max_coal_bds;
2777 	const uint32_t tx_max_coal_bds = sc->bge_tx_max_coal_bds;
2778 	mutex_exit(sc->sc_intr_lock);
2779 	CSR_WRITE_4(sc, BGE_HCC_RX_COAL_TICKS, rx_coal_ticks);
2780 	CSR_WRITE_4(sc, BGE_HCC_TX_COAL_TICKS, tx_coal_ticks);
2781 	CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS, rx_max_coal_bds);
2782 	CSR_WRITE_4(sc, BGE_HCC_TX_MAX_COAL_BDS, tx_max_coal_bds);
2783 	if (!(BGE_IS_5705_PLUS(sc))) {
2784 		CSR_WRITE_4(sc, BGE_HCC_RX_COAL_TICKS_INT, 0);
2785 		CSR_WRITE_4(sc, BGE_HCC_TX_COAL_TICKS_INT, 0);
2786 	}
2787 	CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS_INT, 0);
2788 	CSR_WRITE_4(sc, BGE_HCC_TX_MAX_COAL_BDS_INT, 0);
2789 
2790 	/* Set up address of statistics block */
2791 	if (BGE_IS_5700_FAMILY(sc)) {
2792 		BGE_HOSTADDR(taddr, BGE_RING_DMA_ADDR(sc, bge_info.bge_stats));
2793 		CSR_WRITE_4(sc, BGE_HCC_STATS_TICKS, sc->bge_stat_ticks);
2794 		CSR_WRITE_4(sc, BGE_HCC_STATS_BASEADDR, BGE_STATS_BLOCK);
2795 		CSR_WRITE_4(sc, BGE_HCC_STATS_ADDR_HI, taddr.bge_addr_hi);
2796 		CSR_WRITE_4(sc, BGE_HCC_STATS_ADDR_LO, taddr.bge_addr_lo);
2797 	}
2798 
2799 	/* 5718 step 38 */
2800 	/* Set up address of status block */
2801 	BGE_HOSTADDR(taddr, BGE_RING_DMA_ADDR(sc, bge_status_block));
2802 	CSR_WRITE_4(sc, BGE_HCC_STATUSBLK_BASEADDR, BGE_STATUS_BLOCK);
2803 	CSR_WRITE_4(sc, BGE_HCC_STATUSBLK_ADDR_HI, taddr.bge_addr_hi);
2804 	CSR_WRITE_4(sc, BGE_HCC_STATUSBLK_ADDR_LO, taddr.bge_addr_lo);
2805 	sc->bge_rdata->bge_status_block.bge_idx[0].bge_rx_prod_idx = 0;
2806 	sc->bge_rdata->bge_status_block.bge_idx[0].bge_tx_cons_idx = 0;
2807 
2808 	/* Set up status block size. */
2809 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5700 &&
2810 	    sc->bge_chipid != BGE_CHIPID_BCM5700_C0) {
2811 		val = BGE_STATBLKSZ_FULL;
2812 		bzero(&sc->bge_rdata->bge_status_block, BGE_STATUS_BLK_SZ);
2813 	} else {
2814 		val = BGE_STATBLKSZ_32BYTE;
2815 		bzero(&sc->bge_rdata->bge_status_block, 32);
2816 	}
2817 
2818 	/* 5718 step 39, 57XX step 73 */
2819 	/* Turn on host coalescing state machine */
2820 	CSR_WRITE_4(sc, BGE_HCC_MODE, val | BGE_HCCMODE_ENABLE);
2821 
2822 	/* 5718 step 40, 57XX step 74 */
2823 	/* Turn on RX BD completion state machine and enable attentions */
2824 	CSR_WRITE_4(sc, BGE_RBDC_MODE,
2825 	    BGE_RBDCMODE_ENABLE | BGE_RBDCMODE_ATTN);
2826 
2827 	/* 5718 step 41, 57XX step 75 */
2828 	/* Turn on RX list placement state machine */
2829 	CSR_WRITE_4(sc, BGE_RXLP_MODE, BGE_RXLPMODE_ENABLE);
2830 
2831 	/* 57XX step 76 */
2832 	/* Turn on RX list selector state machine. */
2833 	if (!(BGE_IS_5705_PLUS(sc)))
2834 		CSR_WRITE_4(sc, BGE_RXLS_MODE, BGE_RXLSMODE_ENABLE);
2835 
2836 	val = BGE_MACMODE_TXDMA_ENB | BGE_MACMODE_RXDMA_ENB |
2837 	    BGE_MACMODE_RX_STATS_CLEAR | BGE_MACMODE_TX_STATS_CLEAR |
2838 	    BGE_MACMODE_RX_STATS_ENB | BGE_MACMODE_TX_STATS_ENB |
2839 	    BGE_MACMODE_FRMHDR_DMA_ENB;
2840 
2841 	if (sc->bge_flags & BGEF_FIBER_TBI)
2842 		val |= BGE_PORTMODE_TBI;
2843 	else if (sc->bge_flags & BGEF_FIBER_MII)
2844 		val |= BGE_PORTMODE_GMII;
2845 	else
2846 		val |= BGE_PORTMODE_MII;
2847 
2848 	/* 5718 step 42 and 43, 57XX step 77 and 78 */
2849 	/* Allow APE to send/receive frames. */
2850 	if ((sc->bge_mfw_flags & BGE_MFW_ON_APE) != 0)
2851 		val |= BGE_MACMODE_APE_RX_EN | BGE_MACMODE_APE_TX_EN;
2852 
2853 	/* Turn on DMA, clear stats */
2854 	CSR_WRITE_4_FLUSH(sc, BGE_MAC_MODE, val);
2855 	/* 5718 step 44 */
2856 	DELAY(40);
2857 
2858 	/* 5718 step 45, 57XX step 79 */
2859 	/* Set misc. local control, enable interrupts on attentions */
2860 	BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_INTR_ONATTN);
2861 	if (BGE_IS_5717_PLUS(sc)) {
2862 		CSR_READ_4(sc, BGE_MISC_LOCAL_CTL); /* Flush */
2863 		/* 5718 step 46 */
2864 		DELAY(100);
2865 	}
2866 
2867 	/* 57XX step 81 */
2868 	/* Turn on DMA completion state machine */
2869 	if (!(BGE_IS_5705_PLUS(sc)))
2870 		CSR_WRITE_4(sc, BGE_DMAC_MODE, BGE_DMACMODE_ENABLE);
2871 
2872 	/* 5718 step 47, 57XX step 82 */
2873 	val = BGE_WDMAMODE_ENABLE | BGE_WDMAMODE_ALL_ATTNS;
2874 
2875 	/* 5718 step 48 */
2876 	/* Enable host coalescing bug fix. */
2877 	if (BGE_IS_5755_PLUS(sc))
2878 		val |= BGE_WDMAMODE_STATUS_TAG_FIX;
2879 
2880 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5785)
2881 		val |= BGE_WDMAMODE_BURST_ALL_DATA;
2882 
2883 	/* Turn on write DMA state machine */
2884 	CSR_WRITE_4_FLUSH(sc, BGE_WDMA_MODE, val);
2885 	/* 5718 step 49 */
2886 	DELAY(40);
2887 
2888 	val = BGE_RDMAMODE_ENABLE | BGE_RDMAMODE_ALL_ATTNS;
2889 
2890 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5717)
2891 		val |= BGE_RDMAMODE_MULT_DMA_RD_DIS;
2892 
2893 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5784 ||
2894 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5785 ||
2895 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM57780)
2896 		val |= BGE_RDMAMODE_BD_SBD_CRPT_ATTN |
2897 		    BGE_RDMAMODE_MBUF_RBD_CRPT_ATTN |
2898 		    BGE_RDMAMODE_MBUF_SBD_CRPT_ATTN;
2899 
2900 	if (sc->bge_flags & BGEF_PCIE)
2901 		val |= BGE_RDMAMODE_FIFO_LONG_BURST;
2902 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM57766) {
2903 		if (ifp->if_mtu <= ETHERMTU)
2904 			val |= BGE_RDMAMODE_JMB_2K_MMRR;
2905 	}
2906 	if (sc->bge_flags & BGEF_TSO) {
2907 		val |= BGE_RDMAMODE_TSO4_ENABLE;
2908 		if (BGE_IS_5717_PLUS(sc))
2909 			val |= BGE_RDMAMODE_TSO6_ENABLE;
2910 	}
2911 
2912 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5720 ||
2913 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5762) {
2914 		val |= CSR_READ_4(sc, BGE_RDMA_MODE) &
2915 		    BGE_RDMAMODE_H2BNC_VLAN_DET;
2916 		/*
2917 		 * Allow multiple outstanding read requests from
2918 		 * non-LSO read DMA engine.
2919 		 */
2920 		val &= ~BGE_RDMAMODE_MULT_DMA_RD_DIS;
2921 	}
2922 
2923 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5761 ||
2924 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5784 ||
2925 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5785 ||
2926 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM57780 ||
2927 	    BGE_IS_57765_PLUS(sc)) {
2928 		if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5762)
2929 			rdmareg = BGE_RDMA_RSRVCTRL_REG2;
2930 		else
2931 			rdmareg = BGE_RDMA_RSRVCTRL;
2932 		dmactl = CSR_READ_4(sc, rdmareg);
2933 		/*
2934 		 * Adjust tx margin to prevent TX data corruption and
2935 		 * fix internal FIFO overflow.
2936 		 */
2937 		if (sc->bge_chipid == BGE_CHIPID_BCM5719_A0 ||
2938 		    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5762) {
2939 			dmactl &= ~(BGE_RDMA_RSRVCTRL_FIFO_LWM_MASK |
2940 			    BGE_RDMA_RSRVCTRL_FIFO_HWM_MASK |
2941 			    BGE_RDMA_RSRVCTRL_TXMRGN_MASK);
2942 			dmactl |= BGE_RDMA_RSRVCTRL_FIFO_LWM_1_5K |
2943 			    BGE_RDMA_RSRVCTRL_FIFO_HWM_1_5K |
2944 			    BGE_RDMA_RSRVCTRL_TXMRGN_320B;
2945 		}
2946 		/*
2947 		 * Enable fix for read DMA FIFO overruns.
2948 		 * The fix is to limit the number of RX BDs
2949 		 * the hardware would fetch at a time.
2950 		 */
2951 		CSR_WRITE_4(sc, rdmareg, dmactl |
2952 		    BGE_RDMA_RSRVCTRL_FIFO_OFLW_FIX);
2953 	}
2954 
2955 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5719) {
2956 		CSR_WRITE_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL,
2957 		    CSR_READ_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL) |
2958 		    BGE_RDMA_LSO_CRPTEN_CTRL_BLEN_BD_4K |
2959 		    BGE_RDMA_LSO_CRPTEN_CTRL_BLEN_LSO_4K);
2960 	} else if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5720) {
2961 		/*
2962 		 * Allow 4KB burst length reads for non-LSO frames.
2963 		 * Enable 512B burst length reads for buffer descriptors.
2964 		 */
2965 		CSR_WRITE_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL,
2966 		    CSR_READ_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL) |
2967 		    BGE_RDMA_LSO_CRPTEN_CTRL_BLEN_BD_512 |
2968 		    BGE_RDMA_LSO_CRPTEN_CTRL_BLEN_LSO_4K);
2969 	} else if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5762) {
2970 		CSR_WRITE_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL_REG2,
2971 		    CSR_READ_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL_REG2) |
2972 		    BGE_RDMA_LSO_CRPTEN_CTRL_BLEN_BD_4K |
2973 		    BGE_RDMA_LSO_CRPTEN_CTRL_BLEN_LSO_4K);
2974 	}
2975 	/* Turn on read DMA state machine */
2976 	CSR_WRITE_4_FLUSH(sc, BGE_RDMA_MODE, val);
2977 	/* 5718 step 52 */
2978 	delay(40);
2979 
2980 	if (sc->bge_flags & BGEF_RDMA_BUG) {
2981 		for (i = 0; i < BGE_NUM_RDMA_CHANNELS / 2; i++) {
2982 			val = CSR_READ_4(sc, BGE_RDMA_LENGTH + i * 4);
2983 			if ((val & 0xFFFF) > BGE_FRAMELEN)
2984 				break;
2985 			if (((val >> 16) & 0xFFFF) > BGE_FRAMELEN)
2986 				break;
2987 		}
2988 		if (i != BGE_NUM_RDMA_CHANNELS / 2) {
2989 			val = CSR_READ_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL);
2990 			if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5719)
2991 				val |= BGE_RDMA_TX_LENGTH_WA_5719;
2992 			else
2993 				val |= BGE_RDMA_TX_LENGTH_WA_5720;
2994 			CSR_WRITE_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL, val);
2995 		}
2996 	}
2997 
2998 	/* 5718 step 56, 57XX step 84 */
2999 	/* Turn on RX data completion state machine */
3000 	CSR_WRITE_4(sc, BGE_RDC_MODE, BGE_RDCMODE_ENABLE);
3001 
3002 	/* Turn on RX data and RX BD initiator state machine */
3003 	CSR_WRITE_4(sc, BGE_RDBDI_MODE, BGE_RDBDIMODE_ENABLE);
3004 
3005 	/* 57XX step 85 */
3006 	/* Turn on Mbuf cluster free state machine */
3007 	if (!BGE_IS_5705_PLUS(sc))
3008 		CSR_WRITE_4(sc, BGE_MBCF_MODE, BGE_MBCFMODE_ENABLE);
3009 
3010 	/* 5718 step 57, 57XX step 86 */
3011 	/* Turn on send data completion state machine */
3012 	val = BGE_SDCMODE_ENABLE;
3013 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5761)
3014 		val |= BGE_SDCMODE_CDELAY;
3015 	CSR_WRITE_4(sc, BGE_SDC_MODE, val);
3016 
3017 	/* 5718 step 58 */
3018 	/* Turn on send BD completion state machine */
3019 	CSR_WRITE_4(sc, BGE_SBDC_MODE, BGE_SBDCMODE_ENABLE);
3020 
3021 	/* 57XX step 88 */
3022 	/* Turn on RX BD initiator state machine */
3023 	CSR_WRITE_4(sc, BGE_RBDI_MODE, BGE_RBDIMODE_ENABLE);
3024 
3025 	/* 5718 step 60, 57XX step 90 */
3026 	/* Turn on send data initiator state machine */
3027 	if (sc->bge_flags & BGEF_TSO) {
3028 		/* XXX: magic value from Linux driver */
3029 		CSR_WRITE_4(sc, BGE_SDI_MODE, BGE_SDIMODE_ENABLE |
3030 		    BGE_SDIMODE_HW_LSO_PRE_DMA);
3031 	} else
3032 		CSR_WRITE_4(sc, BGE_SDI_MODE, BGE_SDIMODE_ENABLE);
3033 
3034 	/* 5718 step 61, 57XX step 91 */
3035 	/* Turn on send BD initiator state machine */
3036 	CSR_WRITE_4(sc, BGE_SBDI_MODE, BGE_SBDIMODE_ENABLE);
3037 
3038 	/* 5718 step 62, 57XX step 92 */
3039 	/* Turn on send BD selector state machine */
3040 	CSR_WRITE_4(sc, BGE_SRS_MODE, BGE_SRSMODE_ENABLE);
3041 
3042 	/* 5718 step 31, 57XX step 60 */
3043 	CSR_WRITE_4(sc, BGE_SDI_STATS_ENABLE_MASK, 0x007FFFFF);
3044 	/* 5718 step 32, 57XX step 61 */
3045 	CSR_WRITE_4(sc, BGE_SDI_STATS_CTL,
3046 	    BGE_SDISTATSCTL_ENABLE | BGE_SDISTATSCTL_FASTER);
3047 
3048 	/* ack/clear link change events */
3049 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED |
3050 	    BGE_MACSTAT_CFG_CHANGED | BGE_MACSTAT_MI_COMPLETE |
3051 	    BGE_MACSTAT_LINK_CHANGED);
3052 	CSR_WRITE_4(sc, BGE_MI_STS, 0);
3053 
3054 	/*
3055 	 * Enable attention when the link has changed state for
3056 	 * devices that use auto polling.
3057 	 */
3058 	if (sc->bge_flags & BGEF_FIBER_TBI) {
3059 		CSR_WRITE_4(sc, BGE_MI_STS, BGE_MISTS_LINK);
3060 	} else {
3061 		if ((sc->bge_flags & BGEF_CPMU_PRESENT) != 0)
3062 			mimode = BGE_MIMODE_500KHZ_CONST;
3063 		else
3064 			mimode = BGE_MIMODE_BASE;
3065 		/* 5718 step 68. 5718 step 69 (optionally). */
3066 		if (BGE_IS_5700_FAMILY(sc) ||
3067 		    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5705) {
3068 			mimode |= BGE_MIMODE_AUTOPOLL;
3069 			BGE_STS_SETBIT(sc, BGE_STS_AUTOPOLL);
3070 		}
3071 		mimode |= BGE_MIMODE_PHYADDR(sc->bge_phy_addr);
3072 		CSR_WRITE_4(sc, BGE_MI_MODE, mimode);
3073 		if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5700)
3074 			CSR_WRITE_4(sc, BGE_MAC_EVT_ENB,
3075 			    BGE_EVTENB_MI_INTERRUPT);
3076 	}
3077 
3078 	/*
3079 	 * Clear any pending link state attention.
3080 	 * Otherwise some link state change events may be lost until attention
3081 	 * is cleared by bge_intr() -> bge_link_upd() sequence.
3082 	 * It's not necessary on newer BCM chips - perhaps enabling link
3083 	 * state change attentions implies clearing pending attention.
3084 	 */
3085 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED |
3086 	    BGE_MACSTAT_CFG_CHANGED | BGE_MACSTAT_MI_COMPLETE |
3087 	    BGE_MACSTAT_LINK_CHANGED);
3088 
3089 	/* Enable link state change attentions. */
3090 	BGE_SETBIT(sc, BGE_MAC_EVT_ENB, BGE_EVTENB_LINK_CHANGED);
3091 
3092 	return 0;
3093 }
3094 
3095 static const struct bge_revision *
3096 bge_lookup_rev(uint32_t chipid)
3097 {
3098 	const struct bge_revision *br;
3099 
3100 	for (br = bge_revisions; br->br_name != NULL; br++) {
3101 		if (br->br_chipid == chipid)
3102 			return br;
3103 	}
3104 
3105 	for (br = bge_majorrevs; br->br_name != NULL; br++) {
3106 		if (br->br_chipid == BGE_ASICREV(chipid))
3107 			return br;
3108 	}
3109 
3110 	return NULL;
3111 }
3112 
3113 static const struct bge_product *
3114 bge_lookup(const struct pci_attach_args *pa)
3115 {
3116 	const struct bge_product *bp;
3117 
3118 	for (bp = bge_products; bp->bp_name != NULL; bp++) {
3119 		if (PCI_VENDOR(pa->pa_id) == bp->bp_vendor &&
3120 		    PCI_PRODUCT(pa->pa_id) == bp->bp_product)
3121 			return bp;
3122 	}
3123 
3124 	return NULL;
3125 }
3126 
3127 static uint32_t
3128 bge_chipid(const struct pci_attach_args *pa)
3129 {
3130 	uint32_t id;
3131 
3132 	id = pci_conf_read(pa->pa_pc, pa->pa_tag, BGE_PCI_MISC_CTL)
3133 		>> BGE_PCIMISCCTL_ASICREV_SHIFT;
3134 
3135 	if (BGE_ASICREV(id) == BGE_ASICREV_USE_PRODID_REG) {
3136 		switch (PCI_PRODUCT(pa->pa_id)) {
3137 		case PCI_PRODUCT_BROADCOM_BCM5717:
3138 		case PCI_PRODUCT_BROADCOM_BCM5718:
3139 		case PCI_PRODUCT_BROADCOM_BCM5719:
3140 		case PCI_PRODUCT_BROADCOM_BCM5720:
3141 		case PCI_PRODUCT_BROADCOM_BCM5725:
3142 		case PCI_PRODUCT_BROADCOM_BCM5727:
3143 		case PCI_PRODUCT_BROADCOM_BCM5762:
3144 		case PCI_PRODUCT_BROADCOM_BCM57764:
3145 		case PCI_PRODUCT_BROADCOM_BCM57767:
3146 		case PCI_PRODUCT_BROADCOM_BCM57787:
3147 			id = pci_conf_read(pa->pa_pc, pa->pa_tag,
3148 			    BGE_PCI_GEN2_PRODID_ASICREV);
3149 			break;
3150 		case PCI_PRODUCT_BROADCOM_BCM57761:
3151 		case PCI_PRODUCT_BROADCOM_BCM57762:
3152 		case PCI_PRODUCT_BROADCOM_BCM57765:
3153 		case PCI_PRODUCT_BROADCOM_BCM57766:
3154 		case PCI_PRODUCT_BROADCOM_BCM57781:
3155 		case PCI_PRODUCT_BROADCOM_BCM57782:
3156 		case PCI_PRODUCT_BROADCOM_BCM57785:
3157 		case PCI_PRODUCT_BROADCOM_BCM57786:
3158 		case PCI_PRODUCT_BROADCOM_BCM57791:
3159 		case PCI_PRODUCT_BROADCOM_BCM57795:
3160 			id = pci_conf_read(pa->pa_pc, pa->pa_tag,
3161 			    BGE_PCI_GEN15_PRODID_ASICREV);
3162 			break;
3163 		default:
3164 			id = pci_conf_read(pa->pa_pc, pa->pa_tag,
3165 			    BGE_PCI_PRODID_ASICREV);
3166 			break;
3167 		}
3168 	}
3169 
3170 	return id;
3171 }
3172 
3173 /*
3174  * Return true if MSI can be used with this device.
3175  */
3176 static int
3177 bge_can_use_msi(struct bge_softc *sc)
3178 {
3179 	int can_use_msi = 0;
3180 
3181 	switch (BGE_ASICREV(sc->bge_chipid)) {
3182 	case BGE_ASICREV_BCM5714_A0:
3183 	case BGE_ASICREV_BCM5714:
3184 		/*
3185 		 * Apparently, MSI doesn't work when these chips are
3186 		 * configured in single-port mode.
3187 		 */
3188 		break;
3189 	case BGE_ASICREV_BCM5750:
3190 		if (BGE_CHIPREV(sc->bge_chipid) != BGE_CHIPREV_5750_AX &&
3191 		    BGE_CHIPREV(sc->bge_chipid) != BGE_CHIPREV_5750_BX)
3192 			can_use_msi = 1;
3193 		break;
3194 	default:
3195 		if (BGE_IS_575X_PLUS(sc))
3196 			can_use_msi = 1;
3197 	}
3198 	return can_use_msi;
3199 }
3200 
3201 /*
3202  * Probe for a Broadcom chip. Check the PCI vendor and device IDs
3203  * against our list and return its name if we find a match. Note
3204  * that since the Broadcom controller contains VPD support, we
3205  * can get the device name string from the controller itself instead
3206  * of the compiled-in string. This is a little slow, but it guarantees
3207  * we'll always announce the right product name.
3208  */
3209 static int
3210 bge_probe(device_t parent, cfdata_t match, void *aux)
3211 {
3212 	struct pci_attach_args *pa = (struct pci_attach_args *)aux;
3213 
3214 	if (bge_lookup(pa) != NULL)
3215 		return 1;
3216 
3217 	return 0;
3218 }
3219 
3220 static void
3221 bge_attach(device_t parent, device_t self, void *aux)
3222 {
3223 	struct bge_softc * const sc = device_private(self);
3224 	struct pci_attach_args * const pa = aux;
3225 	prop_dictionary_t dict;
3226 	const struct bge_product *bp;
3227 	const struct bge_revision *br;
3228 	pci_chipset_tag_t	pc;
3229 	const char		*intrstr = NULL;
3230 	uint32_t		hwcfg, hwcfg2, hwcfg3, hwcfg4, hwcfg5;
3231 	uint32_t		command;
3232 	struct ifnet		*ifp;
3233 	struct mii_data * const mii = &sc->bge_mii;
3234 	uint32_t		misccfg, mimode, macmode;
3235 	void *			kva;
3236 	u_char			eaddr[ETHER_ADDR_LEN];
3237 	pcireg_t		memtype, subid, reg;
3238 	bus_addr_t		memaddr;
3239 	uint32_t		pm_ctl;
3240 	bool			no_seeprom;
3241 	int			capmask, trys;
3242 	int			mii_flags;
3243 	int			map_flags;
3244 	char intrbuf[PCI_INTRSTR_LEN];
3245 
3246 	bp = bge_lookup(pa);
3247 	KASSERT(bp != NULL);
3248 
3249 	sc->sc_pc = pa->pa_pc;
3250 	sc->sc_pcitag = pa->pa_tag;
3251 	sc->bge_dev = self;
3252 
3253 	sc->bge_pa = *pa;
3254 	pc = sc->sc_pc;
3255 	subid = pci_conf_read(pc, sc->sc_pcitag, PCI_SUBSYS_ID_REG);
3256 
3257 	aprint_naive(": Ethernet controller\n");
3258 	aprint_normal(": %s Ethernet\n", bp->bp_name);
3259 
3260 	/*
3261 	 * Map control/status registers.
3262 	 */
3263 	DPRINTFN(5, ("Map control/status regs\n"));
3264 	command = pci_conf_read(pc, sc->sc_pcitag, PCI_COMMAND_STATUS_REG);
3265 	command |= PCI_COMMAND_MEM_ENABLE | PCI_COMMAND_MASTER_ENABLE;
3266 	pci_conf_write(pc, sc->sc_pcitag, PCI_COMMAND_STATUS_REG, command);
3267 	command = pci_conf_read(pc, sc->sc_pcitag, PCI_COMMAND_STATUS_REG);
3268 
3269 	if (!(command & PCI_COMMAND_MEM_ENABLE)) {
3270 		aprint_error_dev(sc->bge_dev,
3271 		    "failed to enable memory mapping!\n");
3272 		return;
3273 	}
3274 
3275 	DPRINTFN(5, ("pci_mem_find\n"));
3276 	memtype = pci_mapreg_type(sc->sc_pc, sc->sc_pcitag, BGE_PCI_BAR0);
3277 	switch (memtype) {
3278 	case PCI_MAPREG_TYPE_MEM | PCI_MAPREG_MEM_TYPE_32BIT:
3279 	case PCI_MAPREG_TYPE_MEM | PCI_MAPREG_MEM_TYPE_64BIT:
3280 #if 0
3281 		if (pci_mapreg_map(pa, BGE_PCI_BAR0,
3282 		    memtype, 0, &sc->bge_btag, &sc->bge_bhandle,
3283 		    &memaddr, &sc->bge_bsize) == 0)
3284 			break;
3285 #else
3286 		/*
3287 		 * Workaround for PCI prefetchable bit. Some BCM5717-5720 based
3288 		 * system get NMI on boot (PR#48451). This problem might not be
3289 		 * the driver's bug but our PCI common part's bug. Until we
3290 		 * find a real reason, we ignore the prefetchable bit.
3291 		 */
3292 		if (pci_mapreg_info(pa->pa_pc, pa->pa_tag, BGE_PCI_BAR0,
3293 		    memtype, &memaddr, &sc->bge_bsize, &map_flags) == 0) {
3294 			map_flags &= ~BUS_SPACE_MAP_PREFETCHABLE;
3295 			if (bus_space_map(pa->pa_memt, memaddr, sc->bge_bsize,
3296 			    map_flags, &sc->bge_bhandle) == 0) {
3297 				sc->bge_btag = pa->pa_memt;
3298 				break;
3299 			}
3300 		}
3301 #endif
3302 		/* FALLTHROUGH */
3303 	default:
3304 		aprint_error_dev(sc->bge_dev, "can't find mem space\n");
3305 		return;
3306 	}
3307 
3308 	sc->bge_txrx_stopping = false;
3309 
3310 	/* Save various chip information. */
3311 	sc->bge_chipid = bge_chipid(pa);
3312 	sc->bge_phy_addr = bge_phy_addr(sc);
3313 
3314 	if (pci_get_capability(sc->sc_pc, sc->sc_pcitag, PCI_CAP_PCIEXPRESS,
3315 	    &sc->bge_pciecap, NULL) != 0) {
3316 		/* PCIe */
3317 		sc->bge_flags |= BGEF_PCIE;
3318 		/* Extract supported maximum payload size. */
3319 		reg = pci_conf_read(sc->sc_pc, sc->sc_pcitag,
3320 		    sc->bge_pciecap + PCIE_DCAP);
3321 		sc->bge_mps = 128 << (reg & PCIE_DCAP_MAX_PAYLOAD);
3322 		if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5719 ||
3323 		    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5720)
3324 			sc->bge_expmrq = 2048;
3325 		else
3326 			sc->bge_expmrq = 4096;
3327 		bge_set_max_readrq(sc);
3328 	} else if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5785) {
3329 		/* PCIe without PCIe cap */
3330 		sc->bge_flags |= BGEF_PCIE;
3331 	} else if ((pci_conf_read(sc->sc_pc, sc->sc_pcitag, BGE_PCI_PCISTATE) &
3332 		BGE_PCISTATE_PCI_BUSMODE) == 0) {
3333 		/* PCI-X */
3334 		sc->bge_flags |= BGEF_PCIX;
3335 		if (pci_get_capability(pa->pa_pc, pa->pa_tag, PCI_CAP_PCIX,
3336 			&sc->bge_pcixcap, NULL) == 0)
3337 			aprint_error_dev(sc->bge_dev,
3338 			    "unable to find PCIX capability\n");
3339 	}
3340 
3341 	if (BGE_CHIPREV(sc->bge_chipid) == BGE_CHIPREV_5700_BX) {
3342 		/*
3343 		 * Kludge for 5700 Bx bug: a hardware bug (PCIX byte enable?)
3344 		 * can clobber the chip's PCI config-space power control
3345 		 * registers, leaving the card in D3 powersave state. We do
3346 		 * not have memory-mapped registers in this state, so force
3347 		 * device into D0 state before starting initialization.
3348 		 */
3349 		pm_ctl = pci_conf_read(pc, sc->sc_pcitag, BGE_PCI_PWRMGMT_CMD);
3350 		pm_ctl &= ~(PCI_PWR_D0 | PCI_PWR_D1 | PCI_PWR_D2 | PCI_PWR_D3);
3351 		pm_ctl |= (1 << 8) | PCI_PWR_D0 ; /* D0 state */
3352 		pci_conf_write(pc, sc->sc_pcitag, BGE_PCI_PWRMGMT_CMD, pm_ctl);
3353 		DELAY(1000);	/* 27 usec is allegedly sufficient */
3354 	}
3355 
3356 	/* Save chipset family. */
3357 	switch (BGE_ASICREV(sc->bge_chipid)) {
3358 	case BGE_ASICREV_BCM5717:
3359 	case BGE_ASICREV_BCM5719:
3360 	case BGE_ASICREV_BCM5720:
3361 		sc->bge_flags |= BGEF_5717_PLUS;
3362 		/* FALLTHROUGH */
3363 	case BGE_ASICREV_BCM5762:
3364 	case BGE_ASICREV_BCM57765:
3365 	case BGE_ASICREV_BCM57766:
3366 		if (!BGE_IS_5717_PLUS(sc))
3367 			sc->bge_flags |= BGEF_57765_FAMILY;
3368 		sc->bge_flags |= BGEF_57765_PLUS | BGEF_5755_PLUS |
3369 		    BGEF_575X_PLUS | BGEF_5705_PLUS | BGEF_JUMBO_CAPABLE;
3370 		if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5719 ||
3371 		    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5720) {
3372 			/*
3373 			 * Enable work around for DMA engine miscalculation
3374 			 * of TXMBUF available space.
3375 			 */
3376 			sc->bge_flags |= BGEF_RDMA_BUG;
3377 
3378 			if ((BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5719) &&
3379 			    (sc->bge_chipid == BGE_CHIPID_BCM5719_A0)) {
3380 				/* Jumbo frame on BCM5719 A0 does not work. */
3381 				sc->bge_flags &= ~BGEF_JUMBO_CAPABLE;
3382 			}
3383 		}
3384 		break;
3385 	case BGE_ASICREV_BCM5755:
3386 	case BGE_ASICREV_BCM5761:
3387 	case BGE_ASICREV_BCM5784:
3388 	case BGE_ASICREV_BCM5785:
3389 	case BGE_ASICREV_BCM5787:
3390 	case BGE_ASICREV_BCM57780:
3391 		sc->bge_flags |= BGEF_5755_PLUS | BGEF_575X_PLUS | BGEF_5705_PLUS;
3392 		break;
3393 	case BGE_ASICREV_BCM5700:
3394 	case BGE_ASICREV_BCM5701:
3395 	case BGE_ASICREV_BCM5703:
3396 	case BGE_ASICREV_BCM5704:
3397 		sc->bge_flags |= BGEF_5700_FAMILY | BGEF_JUMBO_CAPABLE;
3398 		break;
3399 	case BGE_ASICREV_BCM5714_A0:
3400 	case BGE_ASICREV_BCM5780:
3401 	case BGE_ASICREV_BCM5714:
3402 		sc->bge_flags |= BGEF_5714_FAMILY | BGEF_JUMBO_CAPABLE;
3403 		/* FALLTHROUGH */
3404 	case BGE_ASICREV_BCM5750:
3405 	case BGE_ASICREV_BCM5752:
3406 	case BGE_ASICREV_BCM5906:
3407 		sc->bge_flags |= BGEF_575X_PLUS;
3408 		/* FALLTHROUGH */
3409 	case BGE_ASICREV_BCM5705:
3410 		sc->bge_flags |= BGEF_5705_PLUS;
3411 		break;
3412 	}
3413 
3414 	/* Identify chips with APE processor. */
3415 	switch (BGE_ASICREV(sc->bge_chipid)) {
3416 	case BGE_ASICREV_BCM5717:
3417 	case BGE_ASICREV_BCM5719:
3418 	case BGE_ASICREV_BCM5720:
3419 	case BGE_ASICREV_BCM5761:
3420 	case BGE_ASICREV_BCM5762:
3421 		sc->bge_flags |= BGEF_APE;
3422 		break;
3423 	}
3424 
3425 	/*
3426 	 * The 40bit DMA bug applies to the 5714/5715 controllers and is
3427 	 * not actually a MAC controller bug but an issue with the embedded
3428 	 * PCIe to PCI-X bridge in the device. Use 40bit DMA workaround.
3429 	 */
3430 	if (BGE_IS_5714_FAMILY(sc) && ((sc->bge_flags & BGEF_PCIX) != 0))
3431 		sc->bge_flags |= BGEF_40BIT_BUG;
3432 
3433 	/* Chips with APE need BAR2 access for APE registers/memory. */
3434 	if ((sc->bge_flags & BGEF_APE) != 0) {
3435 		memtype = pci_mapreg_type(pa->pa_pc, pa->pa_tag, BGE_PCI_BAR2);
3436 #if 0
3437 		if (pci_mapreg_map(pa, BGE_PCI_BAR2, memtype, 0,
3438 			&sc->bge_apetag, &sc->bge_apehandle, NULL,
3439 			&sc->bge_apesize)) {
3440 			aprint_error_dev(sc->bge_dev,
3441 			    "couldn't map BAR2 memory\n");
3442 			return;
3443 		}
3444 #else
3445 		/*
3446 		 * Workaround for PCI prefetchable bit. Some BCM5717-5720 based
3447 		 * system get NMI on boot (PR#48451). This problem might not be
3448 		 * the driver's bug but our PCI common part's bug. Until we
3449 		 * find a real reason, we ignore the prefetchable bit.
3450 		 */
3451 		if (pci_mapreg_info(pa->pa_pc, pa->pa_tag, BGE_PCI_BAR2,
3452 		    memtype, &memaddr, &sc->bge_apesize, &map_flags) != 0) {
3453 			aprint_error_dev(sc->bge_dev,
3454 			    "couldn't map BAR2 memory\n");
3455 			return;
3456 		}
3457 
3458 		map_flags &= ~BUS_SPACE_MAP_PREFETCHABLE;
3459 		if (bus_space_map(pa->pa_memt, memaddr,
3460 		    sc->bge_apesize, map_flags, &sc->bge_apehandle) != 0) {
3461 			aprint_error_dev(sc->bge_dev,
3462 			    "couldn't map BAR2 memory\n");
3463 			return;
3464 		}
3465 		sc->bge_apetag = pa->pa_memt;
3466 #endif
3467 
3468 		/* Enable APE register/memory access by host driver. */
3469 		reg = pci_conf_read(pa->pa_pc, pa->pa_tag, BGE_PCI_PCISTATE);
3470 		reg |= BGE_PCISTATE_ALLOW_APE_CTLSPC_WR |
3471 		    BGE_PCISTATE_ALLOW_APE_SHMEM_WR |
3472 		    BGE_PCISTATE_ALLOW_APE_PSPACE_WR;
3473 		pci_conf_write(pa->pa_pc, pa->pa_tag, BGE_PCI_PCISTATE, reg);
3474 
3475 		bge_ape_lock_init(sc);
3476 		bge_ape_read_fw_ver(sc);
3477 	}
3478 
3479 	/* Identify the chips that use an CPMU. */
3480 	if (BGE_IS_5717_PLUS(sc) ||
3481 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5784 ||
3482 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5761 ||
3483 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5785 ||
3484 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM57780)
3485 		sc->bge_flags |= BGEF_CPMU_PRESENT;
3486 
3487 	/*
3488 	 * When using the BCM5701 in PCI-X mode, data corruption has
3489 	 * been observed in the first few bytes of some received packets.
3490 	 * Aligning the packet buffer in memory eliminates the corruption.
3491 	 * Unfortunately, this misaligns the packet payloads.  On platforms
3492 	 * which do not support unaligned accesses, we will realign the
3493 	 * payloads by copying the received packets.
3494 	 */
3495 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5701 &&
3496 	    sc->bge_flags & BGEF_PCIX)
3497 		sc->bge_flags |= BGEF_RX_ALIGNBUG;
3498 
3499 	if (BGE_IS_5700_FAMILY(sc))
3500 		sc->bge_flags |= BGEF_JUMBO_CAPABLE;
3501 
3502 	misccfg = CSR_READ_4(sc, BGE_MISC_CFG);
3503 	misccfg &= BGE_MISCCFG_BOARD_ID_MASK;
3504 
3505 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5705 &&
3506 	    (misccfg == BGE_MISCCFG_BOARD_ID_5788 ||
3507 	     misccfg == BGE_MISCCFG_BOARD_ID_5788M))
3508 		sc->bge_flags |= BGEF_IS_5788;
3509 
3510 	/*
3511 	 * Some controllers seem to require a special firmware to use
3512 	 * TSO. But the firmware is not available to FreeBSD and Linux
3513 	 * claims that the TSO performed by the firmware is slower than
3514 	 * hardware based TSO. Moreover the firmware based TSO has one
3515 	 * known bug which can't handle TSO if ethernet header + IP/TCP
3516 	 * header is greater than 80 bytes. The workaround for the TSO
3517 	 * bug exist but it seems it's too expensive than not using
3518 	 * TSO at all. Some hardware also have the TSO bug so limit
3519 	 * the TSO to the controllers that are not affected TSO issues
3520 	 * (e.g. 5755 or higher).
3521 	 */
3522 	if (BGE_IS_5755_PLUS(sc)) {
3523 		/*
3524 		 * BCM5754 and BCM5787 shares the same ASIC id so
3525 		 * explicit device id check is required.
3526 		 */
3527 		if ((PCI_PRODUCT(pa->pa_id) != PCI_PRODUCT_BROADCOM_BCM5754) &&
3528 		    (PCI_PRODUCT(pa->pa_id) != PCI_PRODUCT_BROADCOM_BCM5754M))
3529 			sc->bge_flags |= BGEF_TSO;
3530 		/* TSO on BCM5719 A0 does not work. */
3531 		if ((BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5719) &&
3532 		    (sc->bge_chipid == BGE_CHIPID_BCM5719_A0))
3533 			sc->bge_flags &= ~BGEF_TSO;
3534 	}
3535 
3536 	capmask = 0xffffffff; /* XXX BMSR_DEFCAPMASK */
3537 	if ((BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5703 &&
3538 	     (misccfg == 0x4000 || misccfg == 0x8000)) ||
3539 	    (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5705 &&
3540 	     PCI_VENDOR(pa->pa_id) == PCI_VENDOR_BROADCOM &&
3541 	     (PCI_PRODUCT(pa->pa_id) == PCI_PRODUCT_BROADCOM_BCM5901 ||
3542 	      PCI_PRODUCT(pa->pa_id) == PCI_PRODUCT_BROADCOM_BCM5901A2 ||
3543 	      PCI_PRODUCT(pa->pa_id) == PCI_PRODUCT_BROADCOM_BCM5705F)) ||
3544 	    (PCI_VENDOR(pa->pa_id) == PCI_VENDOR_BROADCOM &&
3545 	     (PCI_PRODUCT(pa->pa_id) == PCI_PRODUCT_BROADCOM_BCM5751F ||
3546 	      PCI_PRODUCT(pa->pa_id) == PCI_PRODUCT_BROADCOM_BCM5753F ||
3547 	      PCI_PRODUCT(pa->pa_id) == PCI_PRODUCT_BROADCOM_BCM5787F)) ||
3548 	    PCI_PRODUCT(pa->pa_id) == PCI_PRODUCT_BROADCOM_BCM57790 ||
3549 	    PCI_PRODUCT(pa->pa_id) == PCI_PRODUCT_BROADCOM_BCM57791 ||
3550 	    PCI_PRODUCT(pa->pa_id) == PCI_PRODUCT_BROADCOM_BCM57795 ||
3551 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906) {
3552 		/* These chips are 10/100 only. */
3553 		capmask &= ~BMSR_EXTSTAT;
3554 		sc->bge_phy_flags |= BGEPHYF_NO_WIRESPEED;
3555 	}
3556 
3557 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5700 ||
3558 	    (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5705 &&
3559 	     (sc->bge_chipid != BGE_CHIPID_BCM5705_A0 &&
3560 		 sc->bge_chipid != BGE_CHIPID_BCM5705_A1)))
3561 		sc->bge_phy_flags |= BGEPHYF_NO_WIRESPEED;
3562 
3563 	/* Set various PHY bug flags. */
3564 	if (sc->bge_chipid == BGE_CHIPID_BCM5701_A0 ||
3565 	    sc->bge_chipid == BGE_CHIPID_BCM5701_B0)
3566 		sc->bge_phy_flags |= BGEPHYF_CRC_BUG;
3567 	if (BGE_CHIPREV(sc->bge_chipid) == BGE_CHIPREV_5703_AX ||
3568 	    BGE_CHIPREV(sc->bge_chipid) == BGE_CHIPREV_5704_AX)
3569 		sc->bge_phy_flags |= BGEPHYF_ADC_BUG;
3570 	if (sc->bge_chipid == BGE_CHIPID_BCM5704_A0)
3571 		sc->bge_phy_flags |= BGEPHYF_5704_A0_BUG;
3572 	if ((BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5700 ||
3573 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5701) &&
3574 	    PCI_VENDOR(subid) == PCI_VENDOR_DELL)
3575 		sc->bge_phy_flags |= BGEPHYF_NO_3LED;
3576 	if (BGE_IS_5705_PLUS(sc) &&
3577 	    BGE_ASICREV(sc->bge_chipid) != BGE_ASICREV_BCM5906 &&
3578 	    BGE_ASICREV(sc->bge_chipid) != BGE_ASICREV_BCM5785 &&
3579 	    BGE_ASICREV(sc->bge_chipid) != BGE_ASICREV_BCM57780 &&
3580 	    !BGE_IS_57765_PLUS(sc)) {
3581 		if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5755 ||
3582 		    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5761 ||
3583 		    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5784 ||
3584 		    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5787) {
3585 			if (PCI_PRODUCT(pa->pa_id) != PCI_PRODUCT_BROADCOM_BCM5722 &&
3586 			    PCI_PRODUCT(pa->pa_id) != PCI_PRODUCT_BROADCOM_BCM5756)
3587 				sc->bge_phy_flags |= BGEPHYF_JITTER_BUG;
3588 			if (PCI_PRODUCT(pa->pa_id) == PCI_PRODUCT_BROADCOM_BCM5755M)
3589 				sc->bge_phy_flags |= BGEPHYF_ADJUST_TRIM;
3590 		} else
3591 			sc->bge_phy_flags |= BGEPHYF_BER_BUG;
3592 	}
3593 
3594 	/*
3595 	 * SEEPROM check.
3596 	 * First check if firmware knows we do not have SEEPROM.
3597 	 */
3598 	if (prop_dictionary_get_bool(device_properties(self),
3599 	    "without-seeprom", &no_seeprom) && no_seeprom)
3600 		sc->bge_flags |= BGEF_NO_EEPROM;
3601 
3602 	else if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906)
3603 		sc->bge_flags |= BGEF_NO_EEPROM;
3604 
3605 	/* Now check the 'ROM failed' bit on the RX CPU */
3606 	else if (CSR_READ_4(sc, BGE_RXCPU_MODE) & BGE_RXCPUMODE_ROMFAIL)
3607 		sc->bge_flags |= BGEF_NO_EEPROM;
3608 
3609 	sc->bge_asf_mode = 0;
3610 	/* No ASF if APE present. */
3611 	if ((sc->bge_flags & BGEF_APE) == 0) {
3612 		if (bge_allow_asf && (bge_readmem_ind(sc, BGE_SRAM_DATA_SIG) ==
3613 			BGE_SRAM_DATA_SIG_MAGIC)) {
3614 			if (bge_readmem_ind(sc, BGE_SRAM_DATA_CFG) &
3615 			    BGE_HWCFG_ASF) {
3616 				sc->bge_asf_mode |= ASF_ENABLE;
3617 				sc->bge_asf_mode |= ASF_STACKUP;
3618 				if (BGE_IS_575X_PLUS(sc))
3619 					sc->bge_asf_mode |= ASF_NEW_HANDSHAKE;
3620 			}
3621 		}
3622 	}
3623 
3624 	int counts[PCI_INTR_TYPE_SIZE] = {
3625 		[PCI_INTR_TYPE_INTX] = 1,
3626 		[PCI_INTR_TYPE_MSI] = 1,
3627 		[PCI_INTR_TYPE_MSIX] = 1,
3628 	};
3629 	int max_type = PCI_INTR_TYPE_MSIX;
3630 
3631 	if (!bge_can_use_msi(sc)) {
3632 		/* MSI broken, allow only INTx */
3633 		max_type = PCI_INTR_TYPE_INTX;
3634 	}
3635 
3636 	if (pci_intr_alloc(pa, &sc->bge_pihp, counts, max_type) != 0) {
3637 		aprint_error_dev(sc->bge_dev, "couldn't alloc interrupt\n");
3638 		return;
3639 	}
3640 
3641 	DPRINTFN(5, ("pci_intr_string\n"));
3642 	intrstr = pci_intr_string(pc, sc->bge_pihp[0], intrbuf,
3643 	    sizeof(intrbuf));
3644 	DPRINTFN(5, ("pci_intr_establish\n"));
3645 	sc->bge_intrhand = pci_intr_establish_xname(pc, sc->bge_pihp[0],
3646 	    IPL_NET, bge_intr, sc, device_xname(sc->bge_dev));
3647 	if (sc->bge_intrhand == NULL) {
3648 		pci_intr_release(pc, sc->bge_pihp, 1);
3649 		sc->bge_pihp = NULL;
3650 
3651 		aprint_error_dev(self, "couldn't establish interrupt");
3652 		if (intrstr != NULL)
3653 			aprint_error(" at %s", intrstr);
3654 		aprint_error("\n");
3655 		return;
3656 	}
3657 	aprint_normal_dev(sc->bge_dev, "interrupting at %s\n", intrstr);
3658 
3659 	switch (pci_intr_type(pc, sc->bge_pihp[0])) {
3660 	case PCI_INTR_TYPE_MSIX:
3661 	case PCI_INTR_TYPE_MSI:
3662 		KASSERT(bge_can_use_msi(sc));
3663 		sc->bge_flags |= BGEF_MSI;
3664 		break;
3665 	default:
3666 		/* nothing to do */
3667 		break;
3668 	}
3669 
3670 	char wqname[MAXCOMLEN];
3671 	snprintf(wqname, sizeof(wqname), "%sReset", device_xname(sc->bge_dev));
3672 	int error = workqueue_create(&sc->sc_reset_wq, wqname,
3673 	    bge_handle_reset_work, sc, PRI_NONE, IPL_SOFTCLOCK,
3674 	    WQ_MPSAFE);
3675 	if (error) {
3676 		aprint_error_dev(sc->bge_dev,
3677 		    "unable to create reset workqueue\n");
3678 		return;
3679 	}
3680 
3681 
3682 	/*
3683 	 * All controllers except BCM5700 supports tagged status but
3684 	 * we use tagged status only for MSI case on BCM5717. Otherwise
3685 	 * MSI on BCM5717 does not work.
3686 	 */
3687 	if (BGE_IS_57765_PLUS(sc) && sc->bge_flags & BGEF_MSI)
3688 		sc->bge_flags |= BGEF_TAGGED_STATUS;
3689 
3690 	/*
3691 	 * Reset NVRAM before bge_reset(). It's required to acquire NVRAM
3692 	 * lock in bge_reset().
3693 	 */
3694 	CSR_WRITE_4_FLUSH(sc, BGE_EE_ADDR,
3695 	    BGE_EEADDR_RESET | BGE_EEHALFCLK(BGE_HALFCLK_384SCL));
3696 	delay(1000);
3697 	BGE_SETBIT_FLUSH(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_AUTO_EEPROM);
3698 
3699 	bge_stop_fw(sc);
3700 	bge_sig_pre_reset(sc, BGE_RESET_SHUTDOWN);
3701 	if (bge_reset(sc))
3702 		aprint_error_dev(sc->bge_dev, "chip reset failed\n");
3703 
3704 	/*
3705 	 * Read the hardware config word in the first 32k of NIC internal
3706 	 * memory, or fall back to the config word in the EEPROM.
3707 	 * Note: on some BCM5700 cards, this value appears to be unset.
3708 	 */
3709 	hwcfg = hwcfg2 = hwcfg3 = hwcfg4 = hwcfg5 = 0;
3710 	if (bge_readmem_ind(sc, BGE_SRAM_DATA_SIG) ==
3711 	    BGE_SRAM_DATA_SIG_MAGIC) {
3712 		uint32_t tmp;
3713 
3714 		hwcfg = bge_readmem_ind(sc, BGE_SRAM_DATA_CFG);
3715 		tmp = bge_readmem_ind(sc, BGE_SRAM_DATA_VER) >>
3716 		    BGE_SRAM_DATA_VER_SHIFT;
3717 		if ((0 < tmp) && (tmp < 0x100))
3718 			hwcfg2 = bge_readmem_ind(sc, BGE_SRAM_DATA_CFG_2);
3719 		if (sc->bge_flags & BGEF_PCIE)
3720 			hwcfg3 = bge_readmem_ind(sc, BGE_SRAM_DATA_CFG_3);
3721 		if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5785)
3722 			hwcfg4 = bge_readmem_ind(sc, BGE_SRAM_DATA_CFG_4);
3723 		if (BGE_IS_5717_PLUS(sc))
3724 			hwcfg5 = bge_readmem_ind(sc, BGE_SRAM_DATA_CFG_5);
3725 	} else if (!(sc->bge_flags & BGEF_NO_EEPROM)) {
3726 		bge_read_eeprom(sc, (void *)&hwcfg,
3727 		    BGE_EE_HWCFG_OFFSET, sizeof(hwcfg));
3728 		hwcfg = be32toh(hwcfg);
3729 	}
3730 	aprint_normal_dev(sc->bge_dev,
3731 	    "HW config %08x, %08x, %08x, %08x %08x\n",
3732 	    hwcfg, hwcfg2, hwcfg3, hwcfg4, hwcfg5);
3733 
3734 	bge_sig_legacy(sc, BGE_RESET_SHUTDOWN);
3735 	bge_sig_post_reset(sc, BGE_RESET_SHUTDOWN);
3736 
3737 	if (bge_chipinit(sc)) {
3738 		aprint_error_dev(sc->bge_dev, "chip initialization failed\n");
3739 		bge_release_resources(sc);
3740 		return;
3741 	}
3742 
3743 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5700) {
3744 		BGE_SETBIT_FLUSH(sc, BGE_MISC_LOCAL_CTL,
3745 		    BGE_MLC_MISCIO_OUT1 | BGE_MLC_MISCIO_OUTEN1);
3746 		DELAY(100);
3747 	}
3748 
3749 	/* Set MI_MODE */
3750 	mimode = BGE_MIMODE_PHYADDR(sc->bge_phy_addr);
3751 	if ((sc->bge_flags & BGEF_CPMU_PRESENT) != 0)
3752 		mimode |= BGE_MIMODE_500KHZ_CONST;
3753 	else
3754 		mimode |= BGE_MIMODE_BASE;
3755 	CSR_WRITE_4_FLUSH(sc, BGE_MI_MODE, mimode);
3756 	DELAY(80);
3757 
3758 	/*
3759 	 * Get station address from the EEPROM.
3760 	 */
3761 	if (bge_get_eaddr(sc, eaddr)) {
3762 		aprint_error_dev(sc->bge_dev,
3763 		    "failed to read station address\n");
3764 		bge_release_resources(sc);
3765 		return;
3766 	}
3767 
3768 	br = bge_lookup_rev(sc->bge_chipid);
3769 
3770 	if (br == NULL) {
3771 		aprint_normal_dev(sc->bge_dev, "unknown ASIC (0x%x)",
3772 		    sc->bge_chipid);
3773 	} else {
3774 		aprint_normal_dev(sc->bge_dev, "ASIC %s (0x%x)",
3775 		    br->br_name, sc->bge_chipid);
3776 	}
3777 	aprint_normal(", Ethernet address %s\n", ether_sprintf(eaddr));
3778 
3779 	/* Allocate the general information block and ring buffers. */
3780 	if (pci_dma64_available(pa)) {
3781 		sc->bge_dmatag = pa->pa_dmat64;
3782 		sc->bge_dmatag32 = pa->pa_dmat;
3783 		sc->bge_dma64 = true;
3784 	} else {
3785 		sc->bge_dmatag = pa->pa_dmat;
3786 		sc->bge_dmatag32 = pa->pa_dmat;
3787 		sc->bge_dma64 = false;
3788 	}
3789 
3790 	/* 40bit DMA workaround */
3791 	if (sizeof(bus_addr_t) > 4) {
3792 		if ((sc->bge_flags & BGEF_40BIT_BUG) != 0) {
3793 			bus_dma_tag_t olddmatag = sc->bge_dmatag; /* save */
3794 
3795 			if (bus_dmatag_subregion(olddmatag, 0,
3796 			    (bus_addr_t)__MASK(40),
3797 			    &(sc->bge_dmatag), BUS_DMA_WAITOK) != 0) {
3798 				aprint_error_dev(self,
3799 				    "WARNING: failed to restrict dma range,"
3800 				    " falling back to parent bus dma range\n");
3801 				sc->bge_dmatag = olddmatag;
3802 			}
3803 		}
3804 	}
3805 	SLIST_INIT(&sc->txdma_list);
3806 	DPRINTFN(5, ("bus_dmamem_alloc\n"));
3807 	if (bus_dmamem_alloc(sc->bge_dmatag, sizeof(struct bge_ring_data),
3808 			     PAGE_SIZE, 0, &sc->bge_ring_seg, 1,
3809 		&sc->bge_ring_rseg, BUS_DMA_WAITOK)) {
3810 		aprint_error_dev(sc->bge_dev, "can't alloc rx buffers\n");
3811 		return;
3812 	}
3813 	DPRINTFN(5, ("bus_dmamem_map\n"));
3814 	if (bus_dmamem_map(sc->bge_dmatag, &sc->bge_ring_seg,
3815 		sc->bge_ring_rseg, sizeof(struct bge_ring_data), &kva,
3816 			   BUS_DMA_WAITOK)) {
3817 		aprint_error_dev(sc->bge_dev,
3818 		    "can't map DMA buffers (%zu bytes)\n",
3819 		    sizeof(struct bge_ring_data));
3820 		bus_dmamem_free(sc->bge_dmatag, &sc->bge_ring_seg,
3821 		    sc->bge_ring_rseg);
3822 		return;
3823 	}
3824 	DPRINTFN(5, ("bus_dmamap_create\n"));
3825 	if (bus_dmamap_create(sc->bge_dmatag, sizeof(struct bge_ring_data), 1,
3826 	    sizeof(struct bge_ring_data), 0,
3827 	    BUS_DMA_WAITOK, &sc->bge_ring_map)) {
3828 		aprint_error_dev(sc->bge_dev, "can't create DMA map\n");
3829 		bus_dmamem_unmap(sc->bge_dmatag, kva,
3830 				 sizeof(struct bge_ring_data));
3831 		bus_dmamem_free(sc->bge_dmatag, &sc->bge_ring_seg,
3832 		    sc->bge_ring_rseg);
3833 		return;
3834 	}
3835 	DPRINTFN(5, ("bus_dmamap_load\n"));
3836 	if (bus_dmamap_load(sc->bge_dmatag, sc->bge_ring_map, kva,
3837 			    sizeof(struct bge_ring_data), NULL,
3838 			    BUS_DMA_WAITOK)) {
3839 		bus_dmamap_destroy(sc->bge_dmatag, sc->bge_ring_map);
3840 		bus_dmamem_unmap(sc->bge_dmatag, kva,
3841 				 sizeof(struct bge_ring_data));
3842 		bus_dmamem_free(sc->bge_dmatag, &sc->bge_ring_seg,
3843 		    sc->bge_ring_rseg);
3844 		return;
3845 	}
3846 
3847 	DPRINTFN(5, ("bzero\n"));
3848 	sc->bge_rdata = (struct bge_ring_data *)kva;
3849 
3850 	memset(sc->bge_rdata, 0, sizeof(struct bge_ring_data));
3851 
3852 	/* Try to allocate memory for jumbo buffers. */
3853 	if (BGE_IS_JUMBO_CAPABLE(sc)) {
3854 		if (bge_alloc_jumbo_mem(sc)) {
3855 			aprint_error_dev(sc->bge_dev,
3856 			    "jumbo buffer allocation failed\n");
3857 		} else
3858 			sc->ethercom.ec_capabilities |= ETHERCAP_JUMBO_MTU;
3859 	}
3860 
3861 	/* Set default tuneable values. */
3862 	sc->bge_stat_ticks = BGE_TICKS_PER_SEC;
3863 	sc->bge_rx_coal_ticks = 150;
3864 	sc->bge_rx_max_coal_bds = 64;
3865 	sc->bge_tx_coal_ticks = 300;
3866 	sc->bge_tx_max_coal_bds = 400;
3867 	if (BGE_IS_5705_PLUS(sc)) {
3868 		sc->bge_tx_coal_ticks = (12 * 5);
3869 		sc->bge_tx_max_coal_bds = (12 * 5);
3870 			aprint_verbose_dev(sc->bge_dev,
3871 			    "setting short Tx thresholds\n");
3872 	}
3873 
3874 	if (BGE_IS_5717_PLUS(sc))
3875 		sc->bge_return_ring_cnt = BGE_RETURN_RING_CNT;
3876 	else if (BGE_IS_5705_PLUS(sc))
3877 		sc->bge_return_ring_cnt = BGE_RETURN_RING_CNT_5705;
3878 	else
3879 		sc->bge_return_ring_cnt = BGE_RETURN_RING_CNT;
3880 
3881 	sc->sc_mcast_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_SOFTNET);
3882 	sc->sc_intr_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NET);
3883 
3884 	/* Set up ifnet structure */
3885 	ifp = &sc->ethercom.ec_if;
3886 	ifp->if_softc = sc;
3887 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
3888 	ifp->if_extflags = IFEF_MPSAFE;
3889 	ifp->if_ioctl = bge_ioctl;
3890 	ifp->if_stop = bge_stop;
3891 	ifp->if_start = bge_start;
3892 	ifp->if_init = bge_init;
3893 	IFQ_SET_MAXLEN(&ifp->if_snd, uimax(BGE_TX_RING_CNT - 1, IFQ_MAXLEN));
3894 	IFQ_SET_READY(&ifp->if_snd);
3895 	DPRINTFN(5, ("strcpy if_xname\n"));
3896 	strcpy(ifp->if_xname, device_xname(sc->bge_dev));
3897 
3898 	if (sc->bge_chipid != BGE_CHIPID_BCM5700_B0)
3899 		sc->ethercom.ec_if.if_capabilities |=
3900 		    IFCAP_CSUM_IPv4_Tx | IFCAP_CSUM_IPv4_Rx;
3901 #if 1	/* XXX TCP/UDP checksum offload breaks with pf(4) */
3902 		sc->ethercom.ec_if.if_capabilities |=
3903 		    IFCAP_CSUM_TCPv4_Tx | IFCAP_CSUM_TCPv4_Rx |
3904 		    IFCAP_CSUM_UDPv4_Tx | IFCAP_CSUM_UDPv4_Rx;
3905 #endif
3906 	sc->ethercom.ec_capabilities |=
3907 	    ETHERCAP_VLAN_HWTAGGING | ETHERCAP_VLAN_MTU;
3908 	sc->ethercom.ec_capenable |= ETHERCAP_VLAN_HWTAGGING;
3909 
3910 	if (sc->bge_flags & BGEF_TSO)
3911 		sc->ethercom.ec_if.if_capabilities |= IFCAP_TSOv4;
3912 
3913 	/*
3914 	 * Do MII setup.
3915 	 */
3916 	DPRINTFN(5, ("mii setup\n"));
3917 	mii->mii_ifp = ifp;
3918 	mii->mii_readreg = bge_miibus_readreg;
3919 	mii->mii_writereg = bge_miibus_writereg;
3920 	mii->mii_statchg = bge_miibus_statchg;
3921 
3922 	/*
3923 	 * Figure out what sort of media we have by checking the hardware
3924 	 * config word.  Note: on some BCM5700 cards, this value appears to be
3925 	 * unset. If that's the case, we have to rely on identifying the NIC
3926 	 * by its PCI subsystem ID, as we do below for the SysKonnect SK-9D41.
3927 	 * The SysKonnect SK-9D41 is a 1000baseSX card.
3928 	 */
3929 	if (PCI_PRODUCT(subid) == SK_SUBSYSID_9D41 ||
3930 	    (hwcfg & BGE_HWCFG_MEDIA) == BGE_MEDIA_FIBER) {
3931 		if (BGE_IS_5705_PLUS(sc)) {
3932 			sc->bge_flags |= BGEF_FIBER_MII;
3933 			sc->bge_phy_flags |= BGEPHYF_NO_WIRESPEED;
3934 		} else
3935 			sc->bge_flags |= BGEF_FIBER_TBI;
3936 	}
3937 
3938 	/* Set bge_phy_flags before prop_dictionary_set_uint32() */
3939 	if (BGE_IS_JUMBO_CAPABLE(sc))
3940 		sc->bge_phy_flags |= BGEPHYF_JUMBO_CAPABLE;
3941 
3942 	/* set phyflags and chipid before mii_attach() */
3943 	dict = device_properties(self);
3944 	prop_dictionary_set_uint32(dict, "phyflags", sc->bge_phy_flags);
3945 	prop_dictionary_set_uint32(dict, "chipid", sc->bge_chipid);
3946 
3947 	macmode = CSR_READ_4(sc, BGE_MAC_MODE);
3948 	macmode &= ~BGE_MACMODE_PORTMODE;
3949 	/* Initialize ifmedia structures. */
3950 	if (sc->bge_flags & BGEF_FIBER_TBI) {
3951 		CSR_WRITE_4_FLUSH(sc, BGE_MAC_MODE,
3952 		    macmode | BGE_PORTMODE_TBI);
3953 		DELAY(40);
3954 
3955 		struct ifmedia * const ifm = &sc->bge_ifmedia;
3956 		sc->ethercom.ec_ifmedia = ifm;
3957 
3958 		ifmedia_init_with_lock(ifm, IFM_IMASK,
3959 		    bge_ifmedia_upd, bge_ifmedia_sts, sc->sc_intr_lock);
3960 		ifmedia_add(ifm, IFM_ETHER | IFM_1000_SX, 0, NULL);
3961 		ifmedia_add(ifm, IFM_ETHER | IFM_1000_SX | IFM_FDX, 0, NULL);
3962 		ifmedia_add(ifm, IFM_ETHER | IFM_AUTO, 0, NULL);
3963 		ifmedia_set(ifm, IFM_ETHER | IFM_AUTO);
3964 		/* Pretend the user requested this setting */
3965 		sc->bge_ifmedia.ifm_media = sc->bge_ifmedia.ifm_cur->ifm_media;
3966 	} else {
3967 		uint16_t phyreg;
3968 		int rv;
3969 		/*
3970 		 * Do transceiver setup and tell the firmware the
3971 		 * driver is down so we can try to get access the
3972 		 * probe if ASF is running.  Retry a couple of times
3973 		 * if we get a conflict with the ASF firmware accessing
3974 		 * the PHY.
3975 		 */
3976 		if (sc->bge_flags & BGEF_FIBER_MII)
3977 			macmode |= BGE_PORTMODE_GMII;
3978 		else
3979 			macmode |= BGE_PORTMODE_MII;
3980 		CSR_WRITE_4_FLUSH(sc, BGE_MAC_MODE, macmode);
3981 		DELAY(40);
3982 
3983 		/*
3984 		 * Do transceiver setup and tell the firmware the
3985 		 * driver is down so we can try to get access the
3986 		 * probe if ASF is running.  Retry a couple of times
3987 		 * if we get a conflict with the ASF firmware accessing
3988 		 * the PHY.
3989 		 */
3990 		trys = 0;
3991 		BGE_CLRBIT(sc, BGE_MODE_CTL, BGE_MODECTL_STACKUP);
3992 		sc->ethercom.ec_mii = mii;
3993 		ifmedia_init_with_lock(&mii->mii_media, 0, bge_ifmedia_upd,
3994 		    bge_ifmedia_sts, sc->sc_intr_lock);
3995 		mii_flags = MIIF_DOPAUSE;
3996 		if (sc->bge_flags & BGEF_FIBER_MII)
3997 			mii_flags |= MIIF_HAVEFIBER;
3998 again:
3999 		bge_asf_driver_up(sc);
4000 		mutex_enter(sc->sc_intr_lock);
4001 		rv = bge_miibus_readreg(sc->bge_dev, sc->bge_phy_addr,
4002 		    MII_BMCR, &phyreg);
4003 		if ((rv != 0) || ((phyreg & BMCR_PDOWN) != 0)) {
4004 			int i;
4005 
4006 			bge_miibus_writereg(sc->bge_dev, sc->bge_phy_addr,
4007 			    MII_BMCR, BMCR_RESET);
4008 			/* Wait up to 500ms for it to complete. */
4009 			for (i = 0; i < 500; i++) {
4010 				bge_miibus_readreg(sc->bge_dev,
4011 				    sc->bge_phy_addr, MII_BMCR, &phyreg);
4012 				if ((phyreg & BMCR_RESET) == 0)
4013 					break;
4014 				DELAY(1000);
4015 			}
4016 		}
4017 		mutex_exit(sc->sc_intr_lock);
4018 
4019 		mii_attach(sc->bge_dev, mii, capmask, sc->bge_phy_addr,
4020 		    MII_OFFSET_ANY, mii_flags);
4021 
4022 		if (LIST_EMPTY(&mii->mii_phys) && (trys++ < 4))
4023 			goto again;
4024 
4025 		if (LIST_EMPTY(&mii->mii_phys)) {
4026 			aprint_error_dev(sc->bge_dev, "no PHY found!\n");
4027 			ifmedia_add(&mii->mii_media, IFM_ETHER | IFM_MANUAL,
4028 			    0, NULL);
4029 			ifmedia_set(&mii->mii_media, IFM_ETHER | IFM_MANUAL);
4030 		} else
4031 			ifmedia_set(&mii->mii_media, IFM_ETHER | IFM_AUTO);
4032 
4033 		/*
4034 		 * Now tell the firmware we are going up after probing the PHY
4035 		 */
4036 		if (sc->bge_asf_mode & ASF_STACKUP)
4037 			BGE_SETBIT(sc, BGE_MODE_CTL, BGE_MODECTL_STACKUP);
4038 	}
4039 
4040 	/*
4041 	 * Call MI attach routine.
4042 	 */
4043 	DPRINTFN(5, ("if_initialize\n"));
4044 	if_initialize(ifp);
4045 	ifp->if_percpuq = if_percpuq_create(ifp);
4046 	if_deferred_start_init(ifp, NULL);
4047 	if_register(ifp);
4048 
4049 	DPRINTFN(5, ("ether_ifattach\n"));
4050 	ether_ifattach(ifp, eaddr);
4051 	ether_set_ifflags_cb(&sc->ethercom, bge_ifflags_cb);
4052 
4053 	rnd_attach_source(&sc->rnd_source, device_xname(sc->bge_dev),
4054 		RND_TYPE_NET, RND_FLAG_DEFAULT);
4055 #ifdef BGE_EVENT_COUNTERS
4056 	/*
4057 	 * Attach event counters.
4058 	 */
4059 	evcnt_attach_dynamic(&sc->bge_ev_intr, EVCNT_TYPE_INTR,
4060 	    NULL, device_xname(sc->bge_dev), "intr");
4061 	evcnt_attach_dynamic(&sc->bge_ev_intr_spurious, EVCNT_TYPE_INTR,
4062 	    NULL, device_xname(sc->bge_dev), "intr_spurious");
4063 	evcnt_attach_dynamic(&sc->bge_ev_intr_spurious2, EVCNT_TYPE_INTR,
4064 	    NULL, device_xname(sc->bge_dev), "intr_spurious2");
4065 	evcnt_attach_dynamic(&sc->bge_ev_tx_xoff, EVCNT_TYPE_MISC,
4066 	    NULL, device_xname(sc->bge_dev), "tx_xoff");
4067 	evcnt_attach_dynamic(&sc->bge_ev_tx_xon, EVCNT_TYPE_MISC,
4068 	    NULL, device_xname(sc->bge_dev), "tx_xon");
4069 	evcnt_attach_dynamic(&sc->bge_ev_rx_xoff, EVCNT_TYPE_MISC,
4070 	    NULL, device_xname(sc->bge_dev), "rx_xoff");
4071 	evcnt_attach_dynamic(&sc->bge_ev_rx_xon, EVCNT_TYPE_MISC,
4072 	    NULL, device_xname(sc->bge_dev), "rx_xon");
4073 	evcnt_attach_dynamic(&sc->bge_ev_rx_macctl, EVCNT_TYPE_MISC,
4074 	    NULL, device_xname(sc->bge_dev), "rx_macctl");
4075 	evcnt_attach_dynamic(&sc->bge_ev_xoffentered, EVCNT_TYPE_MISC,
4076 	    NULL, device_xname(sc->bge_dev), "xoffentered");
4077 #endif /* BGE_EVENT_COUNTERS */
4078 	DPRINTFN(5, ("callout_init\n"));
4079 	callout_init(&sc->bge_timeout, CALLOUT_MPSAFE);
4080 	callout_setfunc(&sc->bge_timeout, bge_tick, sc);
4081 
4082 	if (pmf_device_register(self, NULL, NULL))
4083 		pmf_class_network_register(self, ifp);
4084 	else
4085 		aprint_error_dev(self, "couldn't establish power handler\n");
4086 
4087 	bge_sysctl_init(sc);
4088 
4089 #ifdef BGE_DEBUG
4090 	bge_debug_info(sc);
4091 #endif
4092 
4093 	sc->bge_attached = true;
4094 }
4095 
4096 /*
4097  * Stop all chip I/O so that the kernel's probe routines don't
4098  * get confused by errant DMAs when rebooting.
4099  */
4100 static int
4101 bge_detach(device_t self, int flags __unused)
4102 {
4103 	struct bge_softc * const sc = device_private(self);
4104 	struct ifnet * const ifp = &sc->ethercom.ec_if;
4105 
4106 	if (!sc->bge_attached)
4107 		return 0;
4108 
4109 	IFNET_LOCK(ifp);
4110 
4111 	/* Stop the interface. Callouts are stopped in it. */
4112 	bge_stop(ifp, 1);
4113 	sc->bge_detaching = true;
4114 
4115 	IFNET_UNLOCK(ifp);
4116 
4117 	mii_detach(&sc->bge_mii, MII_PHY_ANY, MII_OFFSET_ANY);
4118 
4119 	ether_ifdetach(ifp);
4120 	if_detach(ifp);
4121 
4122 	/* Delete all remaining media. */
4123 	ifmedia_fini(&sc->bge_mii.mii_media);
4124 
4125 	bge_release_resources(sc);
4126 
4127 	return 0;
4128 }
4129 
4130 static void
4131 bge_release_resources(struct bge_softc *sc)
4132 {
4133 
4134 	/* Detach sysctl */
4135 	if (sc->bge_log != NULL)
4136 		sysctl_teardown(&sc->bge_log);
4137 
4138 	callout_destroy(&sc->bge_timeout);
4139 
4140 #ifdef BGE_EVENT_COUNTERS
4141 	/* Detach event counters. */
4142 	evcnt_detach(&sc->bge_ev_intr);
4143 	evcnt_detach(&sc->bge_ev_intr_spurious);
4144 	evcnt_detach(&sc->bge_ev_intr_spurious2);
4145 	evcnt_detach(&sc->bge_ev_tx_xoff);
4146 	evcnt_detach(&sc->bge_ev_tx_xon);
4147 	evcnt_detach(&sc->bge_ev_rx_xoff);
4148 	evcnt_detach(&sc->bge_ev_rx_xon);
4149 	evcnt_detach(&sc->bge_ev_rx_macctl);
4150 	evcnt_detach(&sc->bge_ev_xoffentered);
4151 #endif /* BGE_EVENT_COUNTERS */
4152 
4153 	/* Disestablish the interrupt handler */
4154 	if (sc->bge_intrhand != NULL) {
4155 		pci_intr_disestablish(sc->sc_pc, sc->bge_intrhand);
4156 		pci_intr_release(sc->sc_pc, sc->bge_pihp, 1);
4157 		sc->bge_intrhand = NULL;
4158 	}
4159 
4160 	if (sc->bge_cdata.bge_jumbo_buf != NULL)
4161 		bge_free_jumbo_mem(sc);
4162 
4163 	if (sc->bge_dmatag != NULL) {
4164 		bus_dmamap_unload(sc->bge_dmatag, sc->bge_ring_map);
4165 		bus_dmamap_destroy(sc->bge_dmatag, sc->bge_ring_map);
4166 		bus_dmamem_unmap(sc->bge_dmatag, (void *)sc->bge_rdata,
4167 		    sizeof(struct bge_ring_data));
4168 		bus_dmamem_free(sc->bge_dmatag, &sc->bge_ring_seg,
4169 		    sc->bge_ring_rseg);
4170 	}
4171 
4172 	/* Unmap the device registers */
4173 	if (sc->bge_bsize != 0) {
4174 		bus_space_unmap(sc->bge_btag, sc->bge_bhandle, sc->bge_bsize);
4175 		sc->bge_bsize = 0;
4176 	}
4177 
4178 	/* Unmap the APE registers */
4179 	if (sc->bge_apesize != 0) {
4180 		bus_space_unmap(sc->bge_apetag, sc->bge_apehandle,
4181 		    sc->bge_apesize);
4182 		sc->bge_apesize = 0;
4183 	}
4184 	if (sc->sc_intr_lock) {
4185 		mutex_obj_free(sc->sc_intr_lock);
4186 		sc->sc_intr_lock = NULL;
4187 	}
4188 	if (sc->sc_mcast_lock) {
4189 		mutex_obj_free(sc->sc_mcast_lock);
4190 		sc->sc_mcast_lock = NULL;
4191 	}
4192 }
4193 
4194 static int
4195 bge_reset(struct bge_softc *sc)
4196 {
4197 	uint32_t cachesize, command;
4198 	uint32_t reset, mac_mode, mac_mode_mask;
4199 	pcireg_t devctl, reg;
4200 	int i, val;
4201 	void (*write_op)(struct bge_softc *, int, int);
4202 
4203 	/* Make mask for BGE_MAC_MODE register. */
4204 	mac_mode_mask = BGE_MACMODE_HALF_DUPLEX | BGE_MACMODE_PORTMODE;
4205 	if ((sc->bge_mfw_flags & BGE_MFW_ON_APE) != 0)
4206 		mac_mode_mask |= BGE_MACMODE_APE_RX_EN | BGE_MACMODE_APE_TX_EN;
4207 	/* Keep mac_mode_mask's bits of BGE_MAC_MODE register into mac_mode */
4208 	mac_mode = CSR_READ_4(sc, BGE_MAC_MODE) & mac_mode_mask;
4209 
4210 	if (BGE_IS_575X_PLUS(sc) && !BGE_IS_5714_FAMILY(sc) &&
4211 	    (BGE_ASICREV(sc->bge_chipid) != BGE_ASICREV_BCM5906)) {
4212 		if (sc->bge_flags & BGEF_PCIE)
4213 			write_op = bge_writemem_direct;
4214 		else
4215 			write_op = bge_writemem_ind;
4216 	} else
4217 		write_op = bge_writereg_ind;
4218 
4219 	/* 57XX step 4 */
4220 	/* Acquire the NVM lock */
4221 	if ((sc->bge_flags & BGEF_NO_EEPROM) == 0 &&
4222 	    BGE_ASICREV(sc->bge_chipid) != BGE_ASICREV_BCM5700 &&
4223 	    BGE_ASICREV(sc->bge_chipid) != BGE_ASICREV_BCM5701) {
4224 		CSR_WRITE_4(sc, BGE_NVRAM_SWARB, BGE_NVRAMSWARB_SET1);
4225 		for (i = 0; i < 8000; i++) {
4226 			if (CSR_READ_4(sc, BGE_NVRAM_SWARB) &
4227 			    BGE_NVRAMSWARB_GNT1)
4228 				break;
4229 			DELAY(20);
4230 		}
4231 		if (i == 8000) {
4232 			printf("%s: NVRAM lock timedout!\n",
4233 			    device_xname(sc->bge_dev));
4234 		}
4235 	}
4236 
4237 	/* Take APE lock when performing reset. */
4238 	bge_ape_lock(sc, BGE_APE_LOCK_GRC);
4239 
4240 	/* 57XX step 3 */
4241 	/* Save some important PCI state. */
4242 	cachesize = pci_conf_read(sc->sc_pc, sc->sc_pcitag, BGE_PCI_CACHESZ);
4243 	/* 5718 reset step 3 */
4244 	command = pci_conf_read(sc->sc_pc, sc->sc_pcitag, BGE_PCI_CMD);
4245 
4246 	/* 5718 reset step 5, 57XX step 5b-5d */
4247 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MISC_CTL,
4248 	    BGE_PCIMISCCTL_INDIRECT_ACCESS | BGE_PCIMISCCTL_MASK_PCI_INTR |
4249 	    BGE_HIF_SWAP_OPTIONS | BGE_PCIMISCCTL_PCISTATE_RW);
4250 
4251 	/* XXX ???: Disable fastboot on controllers that support it. */
4252 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5752 ||
4253 	    BGE_IS_5755_PLUS(sc))
4254 		CSR_WRITE_4(sc, BGE_FASTBOOT_PC, 0);
4255 
4256 	/* 5718 reset step 2, 57XX step 6 */
4257 	/*
4258 	 * Write the magic number to SRAM at offset 0xB50.
4259 	 * When firmware finishes its initialization it will
4260 	 * write ~BGE_MAGIC_NUMBER to the same location.
4261 	 */
4262 	bge_writemem_ind(sc, BGE_SRAM_FW_MB, BGE_SRAM_FW_MB_MAGIC);
4263 
4264 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM57780) {
4265 		val = CSR_READ_4(sc, BGE_PCIE_LINKCTL);
4266 		val = (val & ~BGE_PCIE_LINKCTL_L1_PLL_PDEN)
4267 		    | BGE_PCIE_LINKCTL_L1_PLL_PDDIS;
4268 		CSR_WRITE_4(sc, BGE_PCIE_LINKCTL, val);
4269 	}
4270 
4271 	/* 5718 reset step 6, 57XX step 7 */
4272 	reset = BGE_MISCCFG_RESET_CORE_CLOCKS | BGE_32BITTIME_66MHZ;
4273 	/*
4274 	 * XXX: from FreeBSD/Linux; no documentation
4275 	 */
4276 	if (sc->bge_flags & BGEF_PCIE) {
4277 		if ((BGE_ASICREV(sc->bge_chipid) != BGE_ASICREV_BCM5785) &&
4278 		    !BGE_IS_57765_PLUS(sc) &&
4279 		    (CSR_READ_4(sc, BGE_PHY_TEST_CTRL_REG) ==
4280 			(BGE_PHY_PCIE_LTASS_MODE | BGE_PHY_PCIE_SCRAM_MODE))) {
4281 			/* PCI Express 1.0 system */
4282 			CSR_WRITE_4(sc, BGE_PHY_TEST_CTRL_REG,
4283 			    BGE_PHY_PCIE_SCRAM_MODE);
4284 		}
4285 		if (sc->bge_chipid != BGE_CHIPID_BCM5750_A0) {
4286 			/*
4287 			 * Prevent PCI Express link training
4288 			 * during global reset.
4289 			 */
4290 			CSR_WRITE_4(sc, BGE_MISC_CFG, 1 << 29);
4291 			reset |= (1 << 29);
4292 		}
4293 	}
4294 
4295 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906) {
4296 		i = CSR_READ_4(sc, BGE_VCPU_STATUS);
4297 		CSR_WRITE_4(sc, BGE_VCPU_STATUS,
4298 		    i | BGE_VCPU_STATUS_DRV_RESET);
4299 		i = CSR_READ_4(sc, BGE_VCPU_EXT_CTRL);
4300 		CSR_WRITE_4(sc, BGE_VCPU_EXT_CTRL,
4301 		    i & ~BGE_VCPU_EXT_CTRL_HALT_CPU);
4302 	}
4303 
4304 	/*
4305 	 * Set GPHY Power Down Override to leave GPHY
4306 	 * powered up in D0 uninitialized.
4307 	 */
4308 	if (BGE_IS_5705_PLUS(sc) &&
4309 	    (sc->bge_flags & BGEF_CPMU_PRESENT) == 0)
4310 		reset |= BGE_MISCCFG_GPHY_PD_OVERRIDE;
4311 
4312 	/* Issue global reset */
4313 	write_op(sc, BGE_MISC_CFG, reset);
4314 
4315 	/* 5718 reset step 7, 57XX step 8 */
4316 	if (sc->bge_flags & BGEF_PCIE)
4317 		delay(100*1000); /* too big */
4318 	else
4319 		delay(1000);
4320 
4321 	if (sc->bge_flags & BGEF_PCIE) {
4322 		if (sc->bge_chipid == BGE_CHIPID_BCM5750_A0) {
4323 			DELAY(500000);
4324 			/* XXX: Magic Numbers */
4325 			reg = pci_conf_read(sc->sc_pc, sc->sc_pcitag,
4326 			    BGE_PCI_UNKNOWN0);
4327 			pci_conf_write(sc->sc_pc, sc->sc_pcitag,
4328 			    BGE_PCI_UNKNOWN0,
4329 			    reg | (1 << 15));
4330 		}
4331 		devctl = pci_conf_read(sc->sc_pc, sc->sc_pcitag,
4332 		    sc->bge_pciecap + PCIE_DCSR);
4333 		/* Clear enable no snoop and disable relaxed ordering. */
4334 		devctl &= ~(PCIE_DCSR_ENA_RELAX_ORD |
4335 		    PCIE_DCSR_ENA_NO_SNOOP);
4336 
4337 		/* Set PCIE max payload size to 128 for older PCIe devices */
4338 		if ((sc->bge_flags & BGEF_CPMU_PRESENT) == 0)
4339 			devctl &= ~(0x00e0);
4340 		/* Clear device status register. Write 1b to clear */
4341 		devctl |= PCIE_DCSR_URD | PCIE_DCSR_FED
4342 		    | PCIE_DCSR_NFED | PCIE_DCSR_CED;
4343 		pci_conf_write(sc->sc_pc, sc->sc_pcitag,
4344 		    sc->bge_pciecap + PCIE_DCSR, devctl);
4345 		bge_set_max_readrq(sc);
4346 	}
4347 
4348 	/* From Linux: dummy read to flush PCI posted writes */
4349 	reg = pci_conf_read(sc->sc_pc, sc->sc_pcitag, BGE_PCI_CMD);
4350 
4351 	/*
4352 	 * Reset some of the PCI state that got zapped by reset
4353 	 * To modify the PCISTATE register, BGE_PCIMISCCTL_PCISTATE_RW must be
4354 	 * set, too.
4355 	 */
4356 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MISC_CTL,
4357 	    BGE_PCIMISCCTL_INDIRECT_ACCESS | BGE_PCIMISCCTL_MASK_PCI_INTR |
4358 	    BGE_HIF_SWAP_OPTIONS | BGE_PCIMISCCTL_PCISTATE_RW);
4359 	val = BGE_PCISTATE_ROM_ENABLE | BGE_PCISTATE_ROM_RETRY_ENABLE;
4360 	if (sc->bge_chipid == BGE_CHIPID_BCM5704_A0 &&
4361 	    (sc->bge_flags & BGEF_PCIX) != 0)
4362 		val |= BGE_PCISTATE_RETRY_SAME_DMA;
4363 	if ((sc->bge_mfw_flags & BGE_MFW_ON_APE) != 0)
4364 		val |= BGE_PCISTATE_ALLOW_APE_CTLSPC_WR |
4365 		    BGE_PCISTATE_ALLOW_APE_SHMEM_WR |
4366 		    BGE_PCISTATE_ALLOW_APE_PSPACE_WR;
4367 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_PCISTATE, val);
4368 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_CACHESZ, cachesize);
4369 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_CMD, command);
4370 
4371 	/* 57xx step 11: disable PCI-X Relaxed Ordering. */
4372 	if (sc->bge_flags & BGEF_PCIX) {
4373 		reg = pci_conf_read(sc->sc_pc, sc->sc_pcitag, sc->bge_pcixcap
4374 		    + PCIX_CMD);
4375 		/* Set max memory read byte count to 2K */
4376 		if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5703) {
4377 			reg &= ~PCIX_CMD_BYTECNT_MASK;
4378 			reg |= PCIX_CMD_BCNT_2048;
4379 		} else if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5704){
4380 			/*
4381 			 * For 5704, set max outstanding split transaction
4382 			 * field to 0 (0 means it supports 1 request)
4383 			 */
4384 			reg &= ~(PCIX_CMD_SPLTRANS_MASK
4385 			    | PCIX_CMD_BYTECNT_MASK);
4386 			reg |= PCIX_CMD_BCNT_2048;
4387 		}
4388 		pci_conf_write(sc->sc_pc, sc->sc_pcitag, sc->bge_pcixcap
4389 		    + PCIX_CMD, reg & ~PCIX_CMD_RELAXED_ORDER);
4390 	}
4391 
4392 	/* 5718 reset step 10, 57XX step 12 */
4393 	/* Enable memory arbiter. */
4394 	if (BGE_IS_5714_FAMILY(sc)) {
4395 		val = CSR_READ_4(sc, BGE_MARB_MODE);
4396 		CSR_WRITE_4(sc, BGE_MARB_MODE, BGE_MARBMODE_ENABLE | val);
4397 	} else
4398 		CSR_WRITE_4(sc, BGE_MARB_MODE, BGE_MARBMODE_ENABLE);
4399 
4400 	/* XXX 5721, 5751 and 5752 */
4401 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5750) {
4402 		/* Step 19: */
4403 		BGE_SETBIT(sc, BGE_TLP_CONTROL_REG, 1 << 29 | 1 << 25);
4404 		/* Step 20: */
4405 		BGE_SETBIT(sc, BGE_TLP_CONTROL_REG, BGE_TLP_DATA_FIFO_PROTECT);
4406 	}
4407 
4408 	/* 5718 reset step 12, 57XX step 15 and 16 */
4409 	/* Fix up byte swapping */
4410 	CSR_WRITE_4(sc, BGE_MODE_CTL, BGE_DMA_SWAP_OPTIONS);
4411 
4412 	/* 5718 reset step 13, 57XX step 17 */
4413 	/* Poll until the firmware initialization is complete */
4414 	bge_poll_fw(sc);
4415 
4416 	/* 57XX step 21 */
4417 	if (BGE_CHIPREV(sc->bge_chipid) == BGE_CHIPREV_5704_BX) {
4418 		pcireg_t msidata;
4419 
4420 		msidata = pci_conf_read(sc->sc_pc, sc->sc_pcitag,
4421 		    BGE_PCI_MSI_DATA);
4422 		msidata |= ((1 << 13 | 1 << 12 | 1 << 10) << 16);
4423 		pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MSI_DATA,
4424 		    msidata);
4425 	}
4426 
4427 	/* 57XX step 18 */
4428 	/* Write mac mode. */
4429 	val = CSR_READ_4(sc, BGE_MAC_MODE);
4430 	/* Restore mac_mode_mask's bits using mac_mode */
4431 	val = (val & ~mac_mode_mask) | mac_mode;
4432 	CSR_WRITE_4_FLUSH(sc, BGE_MAC_MODE, val);
4433 	DELAY(40);
4434 
4435 	bge_ape_unlock(sc, BGE_APE_LOCK_GRC);
4436 
4437 	/*
4438 	 * The 5704 in TBI mode apparently needs some special
4439 	 * adjustment to insure the SERDES drive level is set
4440 	 * to 1.2V.
4441 	 */
4442 	if (sc->bge_flags & BGEF_FIBER_TBI &&
4443 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5704) {
4444 		uint32_t serdescfg;
4445 
4446 		serdescfg = CSR_READ_4(sc, BGE_SERDES_CFG);
4447 		serdescfg = (serdescfg & ~0xFFF) | 0x880;
4448 		CSR_WRITE_4(sc, BGE_SERDES_CFG, serdescfg);
4449 	}
4450 
4451 	if (sc->bge_flags & BGEF_PCIE &&
4452 	    !BGE_IS_57765_PLUS(sc) &&
4453 	    sc->bge_chipid != BGE_CHIPID_BCM5750_A0 &&
4454 	    BGE_ASICREV(sc->bge_chipid) != BGE_ASICREV_BCM5785) {
4455 		uint32_t v;
4456 
4457 		/* Enable PCI Express bug fix */
4458 		v = CSR_READ_4(sc, BGE_TLP_CONTROL_REG);
4459 		CSR_WRITE_4(sc, BGE_TLP_CONTROL_REG,
4460 		    v | BGE_TLP_DATA_FIFO_PROTECT);
4461 	}
4462 
4463 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5720)
4464 		BGE_CLRBIT(sc, BGE_CPMU_CLCK_ORIDE,
4465 		    CPMU_CLCK_ORIDE_MAC_ORIDE_EN);
4466 
4467 	return 0;
4468 }
4469 
4470 /*
4471  * Frame reception handling. This is called if there's a frame
4472  * on the receive return list.
4473  *
4474  * Note: we have to be able to handle two possibilities here:
4475  * 1) the frame is from the jumbo receive ring
4476  * 2) the frame is from the standard receive ring
4477  */
4478 
4479 static void
4480 bge_rxeof(struct bge_softc *sc)
4481 {
4482 	struct ifnet * const ifp = &sc->ethercom.ec_if;
4483 	uint16_t rx_prod, rx_cons;
4484 	int stdcnt = 0, jumbocnt = 0;
4485 	bus_dmamap_t dmamap;
4486 	bus_addr_t offset, toff;
4487 	bus_size_t tlen;
4488 	int tosync;
4489 
4490 	KASSERT(mutex_owned(sc->sc_intr_lock));
4491 
4492 	bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
4493 	    offsetof(struct bge_ring_data, bge_status_block),
4494 	    sizeof(struct bge_status_block),
4495 	    BUS_DMASYNC_POSTREAD);
4496 
4497 	rx_cons = sc->bge_rx_saved_considx;
4498 	rx_prod = sc->bge_rdata->bge_status_block.bge_idx[0].bge_rx_prod_idx;
4499 
4500 	/* Nothing to do */
4501 	if (rx_cons == rx_prod)
4502 		return;
4503 
4504 	offset = offsetof(struct bge_ring_data, bge_rx_return_ring);
4505 	tosync = rx_prod - rx_cons;
4506 
4507 	if (tosync != 0)
4508 		rnd_add_uint32(&sc->rnd_source, tosync);
4509 
4510 	toff = offset + (rx_cons * sizeof(struct bge_rx_bd));
4511 
4512 	if (tosync < 0) {
4513 		tlen = (sc->bge_return_ring_cnt - rx_cons) *
4514 		    sizeof(struct bge_rx_bd);
4515 		bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
4516 		    toff, tlen, BUS_DMASYNC_POSTREAD);
4517 		tosync = rx_prod;
4518 		toff = offset;
4519 	}
4520 
4521 	if (tosync != 0) {
4522 		bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
4523 		    toff, tosync * sizeof(struct bge_rx_bd),
4524 		    BUS_DMASYNC_POSTREAD);
4525 	}
4526 
4527 	while (rx_cons != rx_prod) {
4528 		struct bge_rx_bd	*cur_rx;
4529 		uint32_t		rxidx;
4530 		struct mbuf		*m = NULL;
4531 
4532 		cur_rx = &sc->bge_rdata->bge_rx_return_ring[rx_cons];
4533 
4534 		rxidx = cur_rx->bge_idx;
4535 		BGE_INC(rx_cons, sc->bge_return_ring_cnt);
4536 
4537 		if (cur_rx->bge_flags & BGE_RXBDFLAG_JUMBO_RING) {
4538 			BGE_INC(sc->bge_jumbo, BGE_JUMBO_RX_RING_CNT);
4539 			m = sc->bge_cdata.bge_rx_jumbo_chain[rxidx];
4540 			sc->bge_cdata.bge_rx_jumbo_chain[rxidx] = NULL;
4541 			jumbocnt++;
4542 			bus_dmamap_sync(sc->bge_dmatag,
4543 			    sc->bge_cdata.bge_rx_jumbo_map,
4544 			    mtod(m, char *) - (char *)sc->bge_cdata.bge_jumbo_buf,
4545 			    BGE_JLEN, BUS_DMASYNC_POSTREAD);
4546 			if (cur_rx->bge_flags & BGE_RXBDFLAG_ERROR) {
4547 				if_statinc(ifp, if_ierrors);
4548 				bge_newbuf_jumbo(sc, sc->bge_jumbo, m);
4549 				continue;
4550 			}
4551 			if (bge_newbuf_jumbo(sc, sc->bge_jumbo,
4552 					     NULL) == ENOBUFS) {
4553 				if_statinc(ifp, if_ierrors);
4554 				bge_newbuf_jumbo(sc, sc->bge_jumbo, m);
4555 				continue;
4556 			}
4557 		} else {
4558 			m = sc->bge_cdata.bge_rx_std_chain[rxidx];
4559 			sc->bge_cdata.bge_rx_std_chain[rxidx] = NULL;
4560 
4561 			stdcnt++;
4562 			sc->bge_std_cnt--;
4563 
4564 			dmamap = sc->bge_cdata.bge_rx_std_map[rxidx];
4565 			bus_dmamap_sync(sc->bge_dmatag, dmamap, 0,
4566 			    dmamap->dm_mapsize, BUS_DMASYNC_POSTREAD);
4567 			bus_dmamap_unload(sc->bge_dmatag, dmamap);
4568 
4569 			if (cur_rx->bge_flags & BGE_RXBDFLAG_ERROR) {
4570 				m_free(m);
4571 				if_statinc(ifp, if_ierrors);
4572 				continue;
4573 			}
4574 		}
4575 
4576 #ifndef __NO_STRICT_ALIGNMENT
4577 		/*
4578 		 * XXX: if the 5701 PCIX-Rx-DMA workaround is in effect,
4579 		 * the Rx buffer has the layer-2 header unaligned.
4580 		 * If our CPU requires alignment, re-align by copying.
4581 		 */
4582 		if (sc->bge_flags & BGEF_RX_ALIGNBUG) {
4583 			memmove(mtod(m, char *) + ETHER_ALIGN, m->m_data,
4584 				cur_rx->bge_len);
4585 			m->m_data += ETHER_ALIGN;
4586 		}
4587 #endif
4588 
4589 		m->m_pkthdr.len = m->m_len = cur_rx->bge_len - ETHER_CRC_LEN;
4590 		m_set_rcvif(m, ifp);
4591 
4592 		bge_rxcsum(sc, cur_rx, m);
4593 
4594 		/*
4595 		 * If we received a packet with a vlan tag, pass it
4596 		 * to vlan_input() instead of ether_input().
4597 		 */
4598 		if (cur_rx->bge_flags & BGE_RXBDFLAG_VLAN_TAG)
4599 			vlan_set_tag(m, cur_rx->bge_vlan_tag);
4600 
4601 		if_percpuq_enqueue(ifp->if_percpuq, m);
4602 	}
4603 
4604 	sc->bge_rx_saved_considx = rx_cons;
4605 	bge_writembx(sc, BGE_MBX_RX_CONS0_LO, sc->bge_rx_saved_considx);
4606 	if (stdcnt)
4607 		bge_fill_rx_ring_std(sc);
4608 	if (jumbocnt)
4609 		bge_writembx(sc, BGE_MBX_RX_JUMBO_PROD_LO, sc->bge_jumbo);
4610 }
4611 
4612 static void
4613 bge_rxcsum(struct bge_softc *sc, struct bge_rx_bd *cur_rx, struct mbuf *m)
4614 {
4615 
4616 	if (BGE_IS_57765_PLUS(sc)) {
4617 		if ((cur_rx->bge_flags & BGE_RXBDFLAG_IPV6) == 0) {
4618 			if ((cur_rx->bge_flags & BGE_RXBDFLAG_IP_CSUM) != 0)
4619 				m->m_pkthdr.csum_flags = M_CSUM_IPv4;
4620 			if ((cur_rx->bge_error_flag &
4621 				BGE_RXERRFLAG_IP_CSUM_NOK) != 0)
4622 				m->m_pkthdr.csum_flags |= M_CSUM_IPv4_BAD;
4623 			if (cur_rx->bge_flags & BGE_RXBDFLAG_TCP_UDP_CSUM) {
4624 				m->m_pkthdr.csum_data =
4625 				    cur_rx->bge_tcp_udp_csum;
4626 				m->m_pkthdr.csum_flags |=
4627 				    (M_CSUM_TCPv4 | M_CSUM_UDPv4 |M_CSUM_DATA);
4628 			}
4629 		}
4630 	} else {
4631 		if ((cur_rx->bge_flags & BGE_RXBDFLAG_IP_CSUM) != 0)
4632 			m->m_pkthdr.csum_flags = M_CSUM_IPv4;
4633 		if ((cur_rx->bge_ip_csum ^ 0xffff) != 0)
4634 			m->m_pkthdr.csum_flags |= M_CSUM_IPv4_BAD;
4635 		/*
4636 		 * Rx transport checksum-offload may also
4637 		 * have bugs with packets which, when transmitted,
4638 		 * were `runts' requiring padding.
4639 		 */
4640 		if (cur_rx->bge_flags & BGE_RXBDFLAG_TCP_UDP_CSUM &&
4641 		    (/* (sc->_bge_quirks & BGE_QUIRK_SHORT_CKSUM_BUG) == 0 ||*/
4642 			    m->m_pkthdr.len >= ETHER_MIN_NOPAD)) {
4643 			m->m_pkthdr.csum_data =
4644 			    cur_rx->bge_tcp_udp_csum;
4645 			m->m_pkthdr.csum_flags |=
4646 			    (M_CSUM_TCPv4 | M_CSUM_UDPv4 | M_CSUM_DATA);
4647 		}
4648 	}
4649 }
4650 
4651 static void
4652 bge_txeof(struct bge_softc *sc)
4653 {
4654 	struct ifnet * const ifp = &sc->ethercom.ec_if;
4655 	struct bge_tx_bd *cur_tx = NULL;
4656 	struct txdmamap_pool_entry *dma;
4657 	bus_addr_t offset, toff;
4658 	bus_size_t tlen;
4659 	int tosync;
4660 	struct mbuf *m;
4661 
4662 	KASSERT(mutex_owned(sc->sc_intr_lock));
4663 
4664 	bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
4665 	    offsetof(struct bge_ring_data, bge_status_block),
4666 	    sizeof(struct bge_status_block),
4667 	    BUS_DMASYNC_POSTREAD);
4668 
4669 	const uint16_t hw_cons_idx =
4670 	    sc->bge_rdata->bge_status_block.bge_idx[0].bge_tx_cons_idx;
4671 	offset = offsetof(struct bge_ring_data, bge_tx_ring);
4672 	tosync = hw_cons_idx - sc->bge_tx_saved_considx;
4673 
4674 	if (tosync != 0)
4675 		rnd_add_uint32(&sc->rnd_source, tosync);
4676 
4677 	toff = offset + (sc->bge_tx_saved_considx * sizeof(struct bge_tx_bd));
4678 
4679 	if (tosync < 0) {
4680 		tlen = (BGE_TX_RING_CNT - sc->bge_tx_saved_considx) *
4681 		    sizeof(struct bge_tx_bd);
4682 		bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
4683 		    toff, tlen, BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
4684 		tosync = hw_cons_idx;
4685 		toff = offset;
4686 	}
4687 
4688 	if (tosync != 0) {
4689 		bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
4690 		    toff, tosync * sizeof(struct bge_tx_bd),
4691 		    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
4692 	}
4693 
4694 	/*
4695 	 * Go through our tx ring and free mbufs for those
4696 	 * frames that have been sent.
4697 	 */
4698 	while (sc->bge_tx_saved_considx != hw_cons_idx) {
4699 		uint32_t idx = sc->bge_tx_saved_considx;
4700 		cur_tx = &sc->bge_rdata->bge_tx_ring[idx];
4701 		if (cur_tx->bge_flags & BGE_TXBDFLAG_END)
4702 			if_statinc(ifp, if_opackets);
4703 		m = sc->bge_cdata.bge_tx_chain[idx];
4704 		if (m != NULL) {
4705 			sc->bge_cdata.bge_tx_chain[idx] = NULL;
4706 			dma = sc->txdma[idx];
4707 			if (dma->is_dma32) {
4708 				bus_dmamap_sync(sc->bge_dmatag32, dma->dmamap32,
4709 				    0, dma->dmamap32->dm_mapsize,
4710 				    BUS_DMASYNC_POSTWRITE);
4711 				bus_dmamap_unload(
4712 				    sc->bge_dmatag32, dma->dmamap32);
4713 			} else {
4714 				bus_dmamap_sync(sc->bge_dmatag, dma->dmamap,
4715 				    0, dma->dmamap->dm_mapsize,
4716 				    BUS_DMASYNC_POSTWRITE);
4717 				bus_dmamap_unload(sc->bge_dmatag, dma->dmamap);
4718 			}
4719 			SLIST_INSERT_HEAD(&sc->txdma_list, dma, link);
4720 			sc->txdma[idx] = NULL;
4721 
4722 			m_freem(m);
4723 		}
4724 		sc->bge_txcnt--;
4725 		BGE_INC(sc->bge_tx_saved_considx, BGE_TX_RING_CNT);
4726 		sc->bge_tx_sending = false;
4727 	}
4728 }
4729 
4730 static int
4731 bge_intr(void *xsc)
4732 {
4733 	struct bge_softc * const sc = xsc;
4734 	struct ifnet * const ifp = &sc->ethercom.ec_if;
4735 	uint32_t pcistate, statusword, statustag;
4736 	uint32_t intrmask = BGE_PCISTATE_INTR_NOT_ACTIVE;
4737 
4738 	/* 5717 and newer chips have no BGE_PCISTATE_INTR_NOT_ACTIVE bit */
4739 	if (BGE_IS_5717_PLUS(sc))
4740 		intrmask = 0;
4741 
4742 	mutex_enter(sc->sc_intr_lock);
4743 	if (sc->bge_txrx_stopping) {
4744 		mutex_exit(sc->sc_intr_lock);
4745 		return 1;
4746 	}
4747 
4748 	/*
4749 	 * It is possible for the interrupt to arrive before
4750 	 * the status block is updated prior to the interrupt.
4751 	 * Reading the PCI State register will confirm whether the
4752 	 * interrupt is ours and will flush the status block.
4753 	 */
4754 	pcistate = CSR_READ_4(sc, BGE_PCI_PCISTATE);
4755 
4756 	/* read status word from status block */
4757 	bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
4758 	    offsetof(struct bge_ring_data, bge_status_block),
4759 	    sizeof(struct bge_status_block),
4760 	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
4761 	statusword = sc->bge_rdata->bge_status_block.bge_status;
4762 	statustag = sc->bge_rdata->bge_status_block.bge_status_tag << 24;
4763 
4764 	if (sc->bge_flags & BGEF_TAGGED_STATUS) {
4765 		if (sc->bge_lasttag == statustag &&
4766 		    (~pcistate & intrmask)) {
4767 			BGE_EVCNT_INCR(sc->bge_ev_intr_spurious);
4768 			mutex_exit(sc->sc_intr_lock);
4769 			return 0;
4770 		}
4771 		sc->bge_lasttag = statustag;
4772 	} else {
4773 		if (!(statusword & BGE_STATFLAG_UPDATED) &&
4774 		    !(~pcistate & intrmask)) {
4775 			BGE_EVCNT_INCR(sc->bge_ev_intr_spurious2);
4776 			mutex_exit(sc->sc_intr_lock);
4777 			return 0;
4778 		}
4779 		statustag = 0;
4780 	}
4781 	/* Ack interrupt and stop others from occurring. */
4782 	bge_writembx_flush(sc, BGE_MBX_IRQ0_LO, 1);
4783 	BGE_EVCNT_INCR(sc->bge_ev_intr);
4784 
4785 	/* clear status word */
4786 	sc->bge_rdata->bge_status_block.bge_status = 0;
4787 
4788 	bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
4789 	    offsetof(struct bge_ring_data, bge_status_block),
4790 	    sizeof(struct bge_status_block),
4791 	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
4792 
4793 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5700 ||
4794 	    statusword & BGE_STATFLAG_LINKSTATE_CHANGED ||
4795 	    BGE_STS_BIT(sc, BGE_STS_LINK_EVT))
4796 		bge_link_upd(sc);
4797 
4798 	/* Check RX return ring producer/consumer */
4799 	bge_rxeof(sc);
4800 
4801 	/* Check TX ring producer/consumer */
4802 	bge_txeof(sc);
4803 
4804 	if (sc->bge_pending_rxintr_change) {
4805 		uint32_t rx_ticks = sc->bge_rx_coal_ticks;
4806 		uint32_t rx_bds = sc->bge_rx_max_coal_bds;
4807 
4808 		CSR_WRITE_4(sc, BGE_HCC_RX_COAL_TICKS, rx_ticks);
4809 		DELAY(10);
4810 		(void)CSR_READ_4(sc, BGE_HCC_RX_COAL_TICKS);
4811 
4812 		CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS, rx_bds);
4813 		DELAY(10);
4814 		(void)CSR_READ_4(sc, BGE_HCC_RX_MAX_COAL_BDS);
4815 
4816 		sc->bge_pending_rxintr_change = false;
4817 	}
4818 	bge_handle_events(sc);
4819 
4820 	/* Re-enable interrupts. */
4821 	bge_writembx_flush(sc, BGE_MBX_IRQ0_LO, statustag);
4822 
4823 	if_schedule_deferred_start(ifp);
4824 
4825 	mutex_exit(sc->sc_intr_lock);
4826 
4827 	return 1;
4828 }
4829 
4830 static void
4831 bge_asf_driver_up(struct bge_softc *sc)
4832 {
4833 	if (sc->bge_asf_mode & ASF_STACKUP) {
4834 		/* Send ASF heartbeat approx. every 2s */
4835 		if (sc->bge_asf_count)
4836 			sc->bge_asf_count --;
4837 		else {
4838 			sc->bge_asf_count = 2;
4839 
4840 			bge_wait_for_event_ack(sc);
4841 
4842 			bge_writemem_ind(sc, BGE_SRAM_FW_CMD_MB,
4843 			    BGE_FW_CMD_DRV_ALIVE3);
4844 			bge_writemem_ind(sc, BGE_SRAM_FW_CMD_LEN_MB, 4);
4845 			bge_writemem_ind(sc, BGE_SRAM_FW_CMD_DATA_MB,
4846 			    BGE_FW_HB_TIMEOUT_SEC);
4847 			CSR_WRITE_4_FLUSH(sc, BGE_RX_CPU_EVENT,
4848 			    CSR_READ_4(sc, BGE_RX_CPU_EVENT) |
4849 			    BGE_RX_CPU_DRV_EVENT);
4850 		}
4851 	}
4852 }
4853 
4854 static void
4855 bge_tick(void *xsc)
4856 {
4857 	struct bge_softc * const sc = xsc;
4858 	struct ifnet * const ifp = &sc->ethercom.ec_if;
4859 	struct mii_data * const mii = &sc->bge_mii;
4860 
4861 	mutex_enter(sc->sc_intr_lock);
4862 
4863 	if (BGE_IS_5705_PLUS(sc))
4864 		bge_stats_update_regs(sc);
4865 	else
4866 		bge_stats_update(sc);
4867 
4868 	if (sc->bge_flags & BGEF_FIBER_TBI) {
4869 		/*
4870 		 * Since in TBI mode auto-polling can't be used we should poll
4871 		 * link status manually. Here we register pending link event
4872 		 * and trigger interrupt.
4873 		 */
4874 		BGE_STS_SETBIT(sc, BGE_STS_LINK_EVT);
4875 		BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_INTR_SET);
4876 	} else {
4877 		/*
4878 		 * Do not touch PHY if we have link up. This could break
4879 		 * IPMI/ASF mode or produce extra input errors.
4880 		 * (extra input errors was reported for bcm5701 & bcm5704).
4881 		 */
4882 		if (!BGE_STS_BIT(sc, BGE_STS_LINK)) {
4883 			mii_tick(mii);
4884 		}
4885 	}
4886 
4887 	bge_asf_driver_up(sc);
4888 
4889 	const bool ok = bge_watchdog_tick(ifp);
4890 	if (ok)
4891 		callout_schedule(&sc->bge_timeout, hz);
4892 	mutex_exit(sc->sc_intr_lock);
4893 }
4894 
4895 static void
4896 bge_stats_update_regs(struct bge_softc *sc)
4897 {
4898 	struct ifnet * const ifp = &sc->ethercom.ec_if;
4899 
4900 	net_stat_ref_t nsr = IF_STAT_GETREF(ifp);
4901 
4902 	if_statadd_ref(ifp, nsr, if_collisions,
4903 	    CSR_READ_4(sc, BGE_MAC_STATS +
4904 	    offsetof(struct bge_mac_stats_regs, etherStatsCollisions)));
4905 
4906 	/*
4907 	 * On BCM5717, BCM5718, BCM5719 A0 and BCM5720 A0,
4908 	 * RXLP_LOCSTAT_IFIN_DROPS includes unwanted multicast frames
4909 	 * (silicon bug). There's no reliable workaround so just
4910 	 * ignore the counter
4911 	 */
4912 	if (BGE_ASICREV(sc->bge_chipid) != BGE_ASICREV_BCM5717 &&
4913 	    sc->bge_chipid != BGE_CHIPID_BCM5719_A0 &&
4914 	    sc->bge_chipid != BGE_CHIPID_BCM5720_A0) {
4915 		if_statadd_ref(ifp, nsr, if_ierrors,
4916 		    CSR_READ_4(sc, BGE_RXLP_LOCSTAT_IFIN_DROPS));
4917 	}
4918 	if_statadd_ref(ifp, nsr, if_ierrors,
4919 	    CSR_READ_4(sc, BGE_RXLP_LOCSTAT_IFIN_ERRORS));
4920 	if_statadd_ref(ifp, nsr, if_ierrors,
4921 	    CSR_READ_4(sc, BGE_RXLP_LOCSTAT_OUT_OF_BDS));
4922 
4923 	IF_STAT_PUTREF(ifp);
4924 
4925 	if (sc->bge_flags & BGEF_RDMA_BUG) {
4926 		uint32_t val, ucast, mcast, bcast;
4927 
4928 		ucast = CSR_READ_4(sc, BGE_MAC_STATS +
4929 		    offsetof(struct bge_mac_stats_regs, ifHCOutUcastPkts));
4930 		mcast = CSR_READ_4(sc, BGE_MAC_STATS +
4931 		    offsetof(struct bge_mac_stats_regs, ifHCOutMulticastPkts));
4932 		bcast = CSR_READ_4(sc, BGE_MAC_STATS +
4933 		    offsetof(struct bge_mac_stats_regs, ifHCOutBroadcastPkts));
4934 
4935 		/*
4936 		 * If controller transmitted more than BGE_NUM_RDMA_CHANNELS
4937 		 * frames, it's safe to disable workaround for DMA engine's
4938 		 * miscalculation of TXMBUF space.
4939 		 */
4940 		if (ucast + mcast + bcast > BGE_NUM_RDMA_CHANNELS) {
4941 			val = CSR_READ_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL);
4942 			if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5719)
4943 				val &= ~BGE_RDMA_TX_LENGTH_WA_5719;
4944 			else
4945 				val &= ~BGE_RDMA_TX_LENGTH_WA_5720;
4946 			CSR_WRITE_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL, val);
4947 			sc->bge_flags &= ~BGEF_RDMA_BUG;
4948 		}
4949 	}
4950 }
4951 
4952 static void
4953 bge_stats_update(struct bge_softc *sc)
4954 {
4955 	struct ifnet * const ifp = &sc->ethercom.ec_if;
4956 	bus_size_t stats = BGE_MEMWIN_START + BGE_STATS_BLOCK;
4957 
4958 #define READ_STAT(sc, stats, stat) \
4959 	  CSR_READ_4(sc, stats + offsetof(struct bge_stats, stat))
4960 
4961 	uint64_t collisions =
4962 	  (READ_STAT(sc, stats, dot3StatsSingleCollisionFrames.bge_addr_lo) +
4963 	   READ_STAT(sc, stats, dot3StatsMultipleCollisionFrames.bge_addr_lo) +
4964 	   READ_STAT(sc, stats, dot3StatsExcessiveCollisions.bge_addr_lo) +
4965 	   READ_STAT(sc, stats, dot3StatsLateCollisions.bge_addr_lo));
4966 
4967 	if_statadd(ifp, if_collisions, collisions - sc->bge_if_collisions);
4968 	sc->bge_if_collisions = collisions;
4969 
4970 
4971 	BGE_EVCNT_UPD(sc->bge_ev_tx_xoff,
4972 		      READ_STAT(sc, stats, outXoffSent.bge_addr_lo));
4973 	BGE_EVCNT_UPD(sc->bge_ev_tx_xon,
4974 		      READ_STAT(sc, stats, outXonSent.bge_addr_lo));
4975 	BGE_EVCNT_UPD(sc->bge_ev_rx_xoff,
4976 		      READ_STAT(sc, stats,
4977 				xoffPauseFramesReceived.bge_addr_lo));
4978 	BGE_EVCNT_UPD(sc->bge_ev_rx_xon,
4979 		      READ_STAT(sc, stats, xonPauseFramesReceived.bge_addr_lo));
4980 	BGE_EVCNT_UPD(sc->bge_ev_rx_macctl,
4981 		      READ_STAT(sc, stats,
4982 				macControlFramesReceived.bge_addr_lo));
4983 	BGE_EVCNT_UPD(sc->bge_ev_xoffentered,
4984 		      READ_STAT(sc, stats, xoffStateEntered.bge_addr_lo));
4985 
4986 #undef READ_STAT
4987 }
4988 
4989 /*
4990  * Pad outbound frame to ETHER_MIN_NOPAD for an unusual reason.
4991  * The bge hardware will pad out Tx runts to ETHER_MIN_NOPAD,
4992  * but when such padded frames employ the  bge IP/TCP checksum offload,
4993  * the hardware checksum assist gives incorrect results (possibly
4994  * from incorporating its own padding into the UDP/TCP checksum; who knows).
4995  * If we pad such runts with zeros, the onboard checksum comes out correct.
4996  */
4997 static inline int
4998 bge_cksum_pad(struct mbuf *pkt)
4999 {
5000 	struct mbuf *last = NULL;
5001 	int padlen;
5002 
5003 	padlen = ETHER_MIN_NOPAD - pkt->m_pkthdr.len;
5004 
5005 	/* if there's only the packet-header and we can pad there, use it. */
5006 	if (pkt->m_pkthdr.len == pkt->m_len &&
5007 	    M_TRAILINGSPACE(pkt) >= padlen) {
5008 		last = pkt;
5009 	} else {
5010 		/*
5011 		 * Walk packet chain to find last mbuf. We will either
5012 		 * pad there, or append a new mbuf and pad it
5013 		 * (thus perhaps avoiding the bcm5700 dma-min bug).
5014 		 */
5015 		for (last = pkt; last->m_next != NULL; last = last->m_next) {
5016 			continue; /* do nothing */
5017 		}
5018 
5019 		/* `last' now points to last in chain. */
5020 		if (M_TRAILINGSPACE(last) < padlen) {
5021 			/* Allocate new empty mbuf, pad it. Compact later. */
5022 			struct mbuf *n;
5023 			MGET(n, M_DONTWAIT, MT_DATA);
5024 			if (n == NULL)
5025 				return ENOBUFS;
5026 			n->m_len = 0;
5027 			last->m_next = n;
5028 			last = n;
5029 		}
5030 	}
5031 
5032 	KDASSERT(!M_READONLY(last));
5033 	KDASSERT(M_TRAILINGSPACE(last) >= padlen);
5034 
5035 	/* Now zero the pad area, to avoid the bge cksum-assist bug */
5036 	memset(mtod(last, char *) + last->m_len, 0, padlen);
5037 	last->m_len += padlen;
5038 	pkt->m_pkthdr.len += padlen;
5039 	return 0;
5040 }
5041 
5042 /*
5043  * Compact outbound packets to avoid bug with DMA segments less than 8 bytes.
5044  */
5045 static inline int
5046 bge_compact_dma_runt(struct mbuf *pkt)
5047 {
5048 	struct mbuf	*m, *prev;
5049 	int		totlen;
5050 
5051 	prev = NULL;
5052 	totlen = 0;
5053 
5054 	for (m = pkt; m != NULL; prev = m, m = m->m_next) {
5055 		int mlen = m->m_len;
5056 		int shortfall = 8 - mlen ;
5057 
5058 		totlen += mlen;
5059 		if (mlen == 0)
5060 			continue;
5061 		if (mlen >= 8)
5062 			continue;
5063 
5064 		/*
5065 		 * If we get here, mbuf data is too small for DMA engine.
5066 		 * Try to fix by shuffling data to prev or next in chain.
5067 		 * If that fails, do a compacting deep-copy of the whole chain.
5068 		 */
5069 
5070 		/* Internal frag. If fits in prev, copy it there. */
5071 		if (prev && M_TRAILINGSPACE(prev) >= m->m_len) {
5072 			memcpy(prev->m_data + prev->m_len, m->m_data, mlen);
5073 			prev->m_len += mlen;
5074 			m->m_len = 0;
5075 			/* XXX stitch chain */
5076 			prev->m_next = m_free(m);
5077 			m = prev;
5078 			continue;
5079 		} else if (m->m_next != NULL &&
5080 			    M_TRAILINGSPACE(m) >= shortfall &&
5081 			    m->m_next->m_len >= (8 + shortfall)) {
5082 		    /* m is writable and have enough data in next, pull up. */
5083 
5084 			memcpy(m->m_data + m->m_len, m->m_next->m_data,
5085 			    shortfall);
5086 			m->m_len += shortfall;
5087 			m->m_next->m_len -= shortfall;
5088 			m->m_next->m_data += shortfall;
5089 		} else if (m->m_next == NULL || 1) {
5090 			/*
5091 			 * Got a runt at the very end of the packet.
5092 			 * borrow data from the tail of the preceding mbuf and
5093 			 * update its length in-place. (The original data is
5094 			 * still valid, so we can do this even if prev is not
5095 			 * writable.)
5096 			 */
5097 
5098 			/*
5099 			 * If we'd make prev a runt, just move all of its data.
5100 			 */
5101 			KASSERT(prev != NULL /*, ("runt but null PREV")*/);
5102 			KASSERT(prev->m_len >= 8 /*, ("runt prev")*/);
5103 
5104 			if ((prev->m_len - shortfall) < 8)
5105 				shortfall = prev->m_len;
5106 
5107 #ifdef notyet	/* just do the safe slow thing for now */
5108 			if (!M_READONLY(m)) {
5109 				if (M_LEADINGSPACE(m) < shorfall) {
5110 					void *m_dat;
5111 					m_dat = M_BUFADDR(m);
5112 					memmove(m_dat, mtod(m, void*),
5113 					    m->m_len);
5114 					m->m_data = m_dat;
5115 				}
5116 			} else
5117 #endif	/* just do the safe slow thing */
5118 			{
5119 				struct mbuf * n = NULL;
5120 				int newprevlen = prev->m_len - shortfall;
5121 
5122 				MGET(n, M_NOWAIT, MT_DATA);
5123 				if (n == NULL)
5124 				   return ENOBUFS;
5125 				KASSERT(m->m_len + shortfall < MLEN
5126 					/*,
5127 					  ("runt %d +prev %d too big\n", m->m_len, shortfall)*/);
5128 
5129 				/* first copy the data we're stealing from prev */
5130 				memcpy(n->m_data, prev->m_data + newprevlen,
5131 				    shortfall);
5132 
5133 				/* update prev->m_len accordingly */
5134 				prev->m_len -= shortfall;
5135 
5136 				/* copy data from runt m */
5137 				memcpy(n->m_data + shortfall, m->m_data,
5138 				    m->m_len);
5139 
5140 				/* n holds what we stole from prev, plus m */
5141 				n->m_len = shortfall + m->m_len;
5142 
5143 				/* stitch n into chain and free m */
5144 				n->m_next = m->m_next;
5145 				prev->m_next = n;
5146 				/* KASSERT(m->m_next == NULL); */
5147 				m->m_next = NULL;
5148 				m_free(m);
5149 				m = n;	/* for continuing loop */
5150 			}
5151 		}
5152 	}
5153 	return 0;
5154 }
5155 
5156 /*
5157  * Encapsulate an mbuf chain in the tx ring by coupling the mbuf data
5158  * pointers to descriptors.
5159  */
5160 static int
5161 bge_encap(struct bge_softc *sc, struct mbuf *m_head, uint32_t *txidx)
5162 {
5163 	struct bge_tx_bd	*f, *prev_f;
5164 	uint32_t		frag, cur;
5165 	uint16_t		csum_flags = 0;
5166 	uint16_t		txbd_tso_flags = 0;
5167 	struct txdmamap_pool_entry *dma;
5168 	bus_dmamap_t dmamap;
5169 	bus_dma_tag_t dmatag;
5170 	int			i = 0;
5171 	int			use_tso, maxsegsize, error;
5172 	bool			have_vtag;
5173 	uint16_t		vtag;
5174 	bool			remap;
5175 
5176 	KASSERT(mutex_owned(sc->sc_intr_lock));
5177 
5178 	if (m_head->m_pkthdr.csum_flags) {
5179 		if (m_head->m_pkthdr.csum_flags & M_CSUM_IPv4)
5180 			csum_flags |= BGE_TXBDFLAG_IP_CSUM;
5181 		if (m_head->m_pkthdr.csum_flags & (M_CSUM_TCPv4 |M_CSUM_UDPv4))
5182 			csum_flags |= BGE_TXBDFLAG_TCP_UDP_CSUM;
5183 	}
5184 
5185 	/*
5186 	 * If we were asked to do an outboard checksum, and the NIC
5187 	 * has the bug where it sometimes adds in the Ethernet padding,
5188 	 * explicitly pad with zeros so the cksum will be correct either way.
5189 	 * (For now, do this for all chip versions, until newer
5190 	 * are confirmed to not require the workaround.)
5191 	 */
5192 	if ((csum_flags & BGE_TXBDFLAG_TCP_UDP_CSUM) == 0 ||
5193 #ifdef notyet
5194 	    (sc->bge_quirks & BGE_QUIRK_SHORT_CKSUM_BUG) == 0 ||
5195 #endif
5196 	    m_head->m_pkthdr.len >= ETHER_MIN_NOPAD)
5197 		goto check_dma_bug;
5198 
5199 	if (bge_cksum_pad(m_head) != 0)
5200 		return ENOBUFS;
5201 
5202 check_dma_bug:
5203 	if (!(BGE_CHIPREV(sc->bge_chipid) == BGE_CHIPREV_5700_BX))
5204 		goto doit;
5205 
5206 	/*
5207 	 * bcm5700 Revision B silicon cannot handle DMA descriptors with
5208 	 * less than eight bytes.  If we encounter a teeny mbuf
5209 	 * at the end of a chain, we can pad.  Otherwise, copy.
5210 	 */
5211 	if (bge_compact_dma_runt(m_head) != 0)
5212 		return ENOBUFS;
5213 
5214 doit:
5215 	dma = SLIST_FIRST(&sc->txdma_list);
5216 	if (dma == NULL) {
5217 		return ENOBUFS;
5218 	}
5219 	dmamap = dma->dmamap;
5220 	dmatag = sc->bge_dmatag;
5221 	dma->is_dma32 = false;
5222 
5223 	/*
5224 	 * Set up any necessary TSO state before we start packing...
5225 	 */
5226 	use_tso = (m_head->m_pkthdr.csum_flags & M_CSUM_TSOv4) != 0;
5227 	if (!use_tso) {
5228 		maxsegsize = 0;
5229 	} else {	/* TSO setup */
5230 		unsigned  mss;
5231 		struct ether_header *eh;
5232 		unsigned ip_tcp_hlen, iptcp_opt_words, tcp_seg_flags, offset;
5233 		unsigned bge_hlen;
5234 		struct mbuf * m0 = m_head;
5235 		struct ip *ip;
5236 		struct tcphdr *th;
5237 		int iphl, hlen;
5238 
5239 		/*
5240 		 * XXX It would be nice if the mbuf pkthdr had offset
5241 		 * fields for the protocol headers.
5242 		 */
5243 
5244 		eh = mtod(m0, struct ether_header *);
5245 		switch (htons(eh->ether_type)) {
5246 		case ETHERTYPE_IP:
5247 			offset = ETHER_HDR_LEN;
5248 			break;
5249 
5250 		case ETHERTYPE_VLAN:
5251 			offset = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
5252 			break;
5253 
5254 		default:
5255 			/*
5256 			 * Don't support this protocol or encapsulation.
5257 			 */
5258 			return ENOBUFS;
5259 		}
5260 
5261 		/*
5262 		 * TCP/IP headers are in the first mbuf; we can do
5263 		 * this the easy way.
5264 		 */
5265 		iphl = M_CSUM_DATA_IPv4_IPHL(m0->m_pkthdr.csum_data);
5266 		hlen = iphl + offset;
5267 		if (__predict_false(m0->m_len <
5268 				    (hlen + sizeof(struct tcphdr)))) {
5269 
5270 			aprint_error_dev(sc->bge_dev,
5271 			    "TSO: hard case m0->m_len == %d < ip/tcp hlen %zd,"
5272 			    "not handled yet\n",
5273 			    m0->m_len, hlen+ sizeof(struct tcphdr));
5274 #ifdef NOTYET
5275 			/*
5276 			 * XXX jonathan@NetBSD.org: untested.
5277 			 * how to force this branch to be taken?
5278 			 */
5279 			BGE_EVCNT_INCR(sc->bge_ev_txtsopain);
5280 
5281 			m_copydata(m0, offset, sizeof(ip), &ip);
5282 			m_copydata(m0, hlen, sizeof(th), &th);
5283 
5284 			ip.ip_len = 0;
5285 
5286 			m_copyback(m0, hlen + offsetof(struct ip, ip_len),
5287 			    sizeof(ip.ip_len), &ip.ip_len);
5288 
5289 			th.th_sum = in_cksum_phdr(ip.ip_src.s_addr,
5290 			    ip.ip_dst.s_addr, htons(IPPROTO_TCP));
5291 
5292 			m_copyback(m0, hlen + offsetof(struct tcphdr, th_sum),
5293 			    sizeof(th.th_sum), &th.th_sum);
5294 
5295 			hlen += th.th_off << 2;
5296 			iptcp_opt_words	= hlen;
5297 #else
5298 			/*
5299 			 * if_wm "hard" case not yet supported, can we not
5300 			 * mandate it out of existence?
5301 			 */
5302 			(void) ip; (void)th; (void) ip_tcp_hlen;
5303 
5304 			return ENOBUFS;
5305 #endif
5306 		} else {
5307 			ip = (struct ip *) (mtod(m0, char *) + offset);
5308 			th = (struct tcphdr *) (mtod(m0, char *) + hlen);
5309 			ip_tcp_hlen = iphl +  (th->th_off << 2);
5310 
5311 			/* Total IP/TCP options, in 32-bit words */
5312 			iptcp_opt_words = (ip_tcp_hlen
5313 					   - sizeof(struct tcphdr)
5314 					   - sizeof(struct ip)) >> 2;
5315 		}
5316 		if (BGE_IS_575X_PLUS(sc)) {
5317 			th->th_sum = 0;
5318 			csum_flags = 0;
5319 		} else {
5320 			/*
5321 			 * XXX jonathan@NetBSD.org: 5705 untested.
5322 			 * Requires TSO firmware patch for 5701/5703/5704.
5323 			 */
5324 			th->th_sum = in_cksum_phdr(ip->ip_src.s_addr,
5325 			    ip->ip_dst.s_addr, htons(IPPROTO_TCP));
5326 		}
5327 
5328 		mss = m_head->m_pkthdr.segsz;
5329 		txbd_tso_flags |=
5330 		    BGE_TXBDFLAG_CPU_PRE_DMA |
5331 		    BGE_TXBDFLAG_CPU_POST_DMA;
5332 
5333 		/*
5334 		 * Our NIC TSO-assist assumes TSO has standard, optionless
5335 		 * IPv4 and TCP headers, which total 40 bytes. By default,
5336 		 * the NIC copies 40 bytes of IP/TCP header from the
5337 		 * supplied header into the IP/TCP header portion of
5338 		 * each post-TSO-segment. If the supplied packet has IP or
5339 		 * TCP options, we need to tell the NIC to copy those extra
5340 		 * bytes into each  post-TSO header, in addition to the normal
5341 		 * 40-byte IP/TCP header (and to leave space accordingly).
5342 		 * Unfortunately, the driver encoding of option length
5343 		 * varies across different ASIC families.
5344 		 */
5345 		tcp_seg_flags = 0;
5346 		bge_hlen = ip_tcp_hlen >> 2;
5347 		if (BGE_IS_5717_PLUS(sc)) {
5348 			tcp_seg_flags = (bge_hlen & 0x3) << 14;
5349 			txbd_tso_flags |=
5350 			    ((bge_hlen & 0xF8) << 7) | ((bge_hlen & 0x4) << 2);
5351 		} else if (BGE_IS_5705_PLUS(sc)) {
5352 			tcp_seg_flags = bge_hlen << 11;
5353 		} else {
5354 			/* XXX iptcp_opt_words or bge_hlen ? */
5355 			txbd_tso_flags |= iptcp_opt_words << 12;
5356 		}
5357 		maxsegsize = mss | tcp_seg_flags;
5358 		ip->ip_len = htons(mss + ip_tcp_hlen);
5359 		ip->ip_sum = 0;
5360 
5361 	}	/* TSO setup */
5362 
5363 	have_vtag = vlan_has_tag(m_head);
5364 	if (have_vtag)
5365 		vtag = vlan_get_tag(m_head);
5366 
5367 	/*
5368 	 * Start packing the mbufs in this chain into
5369 	 * the fragment pointers. Stop when we run out
5370 	 * of fragments or hit the end of the mbuf chain.
5371 	 */
5372 	remap = true;
5373 load_again:
5374 	error = bus_dmamap_load_mbuf(dmatag, dmamap, m_head, BUS_DMA_NOWAIT);
5375 	if (__predict_false(error)) {
5376 		if (error == EFBIG && remap) {
5377 			struct mbuf *m;
5378 			remap = false;
5379 			m = m_defrag(m_head, M_NOWAIT);
5380 			if (m != NULL) {
5381 				KASSERT(m == m_head);
5382 				goto load_again;
5383 			}
5384 		}
5385 		return error;
5386 	}
5387 	/*
5388 	 * Sanity check: avoid coming within 16 descriptors
5389 	 * of the end of the ring.
5390 	 */
5391 	if (dmamap->dm_nsegs > (BGE_TX_RING_CNT - sc->bge_txcnt - 16)) {
5392 		BGE_TSO_PRINTF(("%s: "
5393 		    " dmamap_load_mbuf too close to ring wrap\n",
5394 		    device_xname(sc->bge_dev)));
5395 		goto fail_unload;
5396 	}
5397 
5398 	/* Iterate over dmap-map fragments. */
5399 	f = prev_f = NULL;
5400 	cur = frag = *txidx;
5401 
5402 	for (i = 0; i < dmamap->dm_nsegs; i++) {
5403 		f = &sc->bge_rdata->bge_tx_ring[frag];
5404 		if (sc->bge_cdata.bge_tx_chain[frag] != NULL)
5405 			break;
5406 
5407 		BGE_HOSTADDR(f->bge_addr, dmamap->dm_segs[i].ds_addr);
5408 		f->bge_len = dmamap->dm_segs[i].ds_len;
5409 		if (sizeof(bus_addr_t) > 4 && dma->is_dma32 == false && use_tso && (
5410 		    (dmamap->dm_segs[i].ds_addr & 0xffffffff00000000) !=
5411 		    ((dmamap->dm_segs[i].ds_addr + f->bge_len) & 0xffffffff00000000) ||
5412 		    (prev_f != NULL &&
5413 		     prev_f->bge_addr.bge_addr_hi != f->bge_addr.bge_addr_hi))
5414 		   ) {
5415 			/*
5416 			 * watchdog timeout issue was observed with TSO,
5417 			 * limiting DMA address space to 32bits seems to
5418 			 * address the issue.
5419 			 */
5420 			bus_dmamap_unload(dmatag, dmamap);
5421 			dmatag = sc->bge_dmatag32;
5422 			dmamap = dma->dmamap32;
5423 			dma->is_dma32 = true;
5424 			remap = true;
5425 			goto load_again;
5426 		}
5427 
5428 		/*
5429 		 * For 5751 and follow-ons, for TSO we must turn
5430 		 * off checksum-assist flag in the tx-descr, and
5431 		 * supply the ASIC-revision-specific encoding
5432 		 * of TSO flags and segsize.
5433 		 */
5434 		if (use_tso) {
5435 			if (BGE_IS_575X_PLUS(sc) || i == 0) {
5436 				f->bge_rsvd = maxsegsize;
5437 				f->bge_flags = csum_flags | txbd_tso_flags;
5438 			} else {
5439 				f->bge_rsvd = 0;
5440 				f->bge_flags =
5441 				  (csum_flags | txbd_tso_flags) & 0x0fff;
5442 			}
5443 		} else {
5444 			f->bge_rsvd = 0;
5445 			f->bge_flags = csum_flags;
5446 		}
5447 
5448 		if (have_vtag) {
5449 			f->bge_flags |= BGE_TXBDFLAG_VLAN_TAG;
5450 			f->bge_vlan_tag = vtag;
5451 		} else {
5452 			f->bge_vlan_tag = 0;
5453 		}
5454 		prev_f = f;
5455 		cur = frag;
5456 		BGE_INC(frag, BGE_TX_RING_CNT);
5457 	}
5458 
5459 	if (i < dmamap->dm_nsegs) {
5460 		BGE_TSO_PRINTF(("%s: reached %d < dm_nsegs %d\n",
5461 		    device_xname(sc->bge_dev), i, dmamap->dm_nsegs));
5462 		goto fail_unload;
5463 	}
5464 
5465 	bus_dmamap_sync(dmatag, dmamap, 0, dmamap->dm_mapsize,
5466 	    BUS_DMASYNC_PREWRITE);
5467 
5468 	if (frag == sc->bge_tx_saved_considx) {
5469 		BGE_TSO_PRINTF(("%s: frag %d = wrapped id %d?\n",
5470 		    device_xname(sc->bge_dev), frag, sc->bge_tx_saved_considx));
5471 
5472 		goto fail_unload;
5473 	}
5474 
5475 	sc->bge_rdata->bge_tx_ring[cur].bge_flags |= BGE_TXBDFLAG_END;
5476 	sc->bge_cdata.bge_tx_chain[cur] = m_head;
5477 	SLIST_REMOVE_HEAD(&sc->txdma_list, link);
5478 	sc->txdma[cur] = dma;
5479 	sc->bge_txcnt += dmamap->dm_nsegs;
5480 
5481 	*txidx = frag;
5482 
5483 	return 0;
5484 
5485 fail_unload:
5486 	bus_dmamap_unload(dmatag, dmamap);
5487 
5488 	return ENOBUFS;
5489 }
5490 
5491 
5492 static void
5493 bge_start(struct ifnet *ifp)
5494 {
5495 	struct bge_softc * const sc = ifp->if_softc;
5496 
5497 	mutex_enter(sc->sc_intr_lock);
5498 	if (!sc->bge_txrx_stopping)
5499 		bge_start_locked(ifp);
5500 	mutex_exit(sc->sc_intr_lock);
5501 }
5502 
5503 /*
5504  * Main transmit routine. To avoid having to do mbuf copies, we put pointers
5505  * to the mbuf data regions directly in the transmit descriptors.
5506  */
5507 static void
5508 bge_start_locked(struct ifnet *ifp)
5509 {
5510 	struct bge_softc * const sc = ifp->if_softc;
5511 	struct mbuf *m_head = NULL;
5512 	struct mbuf *m;
5513 	uint32_t prodidx;
5514 	int pkts = 0;
5515 	int error;
5516 
5517 	KASSERT(mutex_owned(sc->sc_intr_lock));
5518 
5519 	prodidx = sc->bge_tx_prodidx;
5520 
5521 	while (sc->bge_cdata.bge_tx_chain[prodidx] == NULL) {
5522 		IFQ_POLL(&ifp->if_snd, m_head);
5523 		if (m_head == NULL)
5524 			break;
5525 
5526 #if 0
5527 		/*
5528 		 * XXX
5529 		 * safety overkill.  If this is a fragmented packet chain
5530 		 * with delayed TCP/UDP checksums, then only encapsulate
5531 		 * it if we have enough descriptors to handle the entire
5532 		 * chain at once.
5533 		 * (paranoia -- may not actually be needed)
5534 		 */
5535 		if (m_head->m_flags & M_FIRSTFRAG &&
5536 		    m_head->m_pkthdr.csum_flags & (CSUM_DELAY_DATA)) {
5537 			if ((BGE_TX_RING_CNT - sc->bge_txcnt) <
5538 			    M_CSUM_DATA_IPv4_OFFSET(m_head->m_pkthdr.csum_data) + 16) {
5539 				ifp->if_flags |= IFF_OACTIVE;
5540 				break;
5541 			}
5542 		}
5543 #endif
5544 
5545 		/*
5546 		 * Pack the data into the transmit ring. If we
5547 		 * don't have room, set the OACTIVE flag and wait
5548 		 * for the NIC to drain the ring.
5549 		 */
5550 		error = bge_encap(sc, m_head, &prodidx);
5551 		if (__predict_false(error)) {
5552 			if (SLIST_EMPTY(&sc->txdma_list)) {
5553 				/* just wait for the transmit ring to drain */
5554 				break;
5555 			}
5556 			IFQ_DEQUEUE(&ifp->if_snd, m);
5557 			KASSERT(m == m_head);
5558 			m_freem(m_head);
5559 			continue;
5560 		}
5561 
5562 		/* now we are committed to transmit the packet */
5563 		IFQ_DEQUEUE(&ifp->if_snd, m);
5564 		KASSERT(m == m_head);
5565 		pkts++;
5566 
5567 		/*
5568 		 * If there's a BPF listener, bounce a copy of this frame
5569 		 * to him.
5570 		 */
5571 		bpf_mtap(ifp, m_head, BPF_D_OUT);
5572 	}
5573 	if (pkts == 0)
5574 		return;
5575 
5576 	/* Transmit */
5577 	bge_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, prodidx);
5578 	/* 5700 b2 errata */
5579 	if (BGE_CHIPREV(sc->bge_chipid) == BGE_CHIPREV_5700_BX)
5580 		bge_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, prodidx);
5581 
5582 	sc->bge_tx_prodidx = prodidx;
5583 	sc->bge_tx_lastsent = time_uptime;
5584 	sc->bge_tx_sending = true;
5585 }
5586 
5587 static int
5588 bge_init(struct ifnet *ifp)
5589 {
5590 	struct bge_softc * const sc = ifp->if_softc;
5591 	const uint16_t *m;
5592 	uint32_t mode, reg;
5593 	int error = 0;
5594 
5595 	ASSERT_SLEEPABLE();
5596 	KASSERT(IFNET_LOCKED(ifp));
5597 	KASSERT(ifp == &sc->ethercom.ec_if);
5598 
5599 	if (sc->bge_detaching)
5600 		return ENXIO;
5601 
5602 	/* Cancel pending I/O and flush buffers. */
5603 	bge_stop(ifp, 0);
5604 
5605 	bge_stop_fw(sc);
5606 	bge_sig_pre_reset(sc, BGE_RESET_START);
5607 	bge_reset(sc);
5608 	bge_sig_legacy(sc, BGE_RESET_START);
5609 
5610 	if (BGE_CHIPREV(sc->bge_chipid) == BGE_CHIPREV_5784_AX) {
5611 		reg = CSR_READ_4(sc, BGE_CPMU_CTRL);
5612 		reg &= ~(BGE_CPMU_CTRL_LINK_AWARE_MODE |
5613 		    BGE_CPMU_CTRL_LINK_IDLE_MODE);
5614 		CSR_WRITE_4(sc, BGE_CPMU_CTRL, reg);
5615 
5616 		reg = CSR_READ_4(sc, BGE_CPMU_LSPD_10MB_CLK);
5617 		reg &= ~BGE_CPMU_LSPD_10MB_CLK;
5618 		reg |= BGE_CPMU_LSPD_10MB_MACCLK_6_25;
5619 		CSR_WRITE_4(sc, BGE_CPMU_LSPD_10MB_CLK, reg);
5620 
5621 		reg = CSR_READ_4(sc, BGE_CPMU_LNK_AWARE_PWRMD);
5622 		reg &= ~BGE_CPMU_LNK_AWARE_MACCLK_MASK;
5623 		reg |= BGE_CPMU_LNK_AWARE_MACCLK_6_25;
5624 		CSR_WRITE_4(sc, BGE_CPMU_LNK_AWARE_PWRMD, reg);
5625 
5626 		reg = CSR_READ_4(sc, BGE_CPMU_HST_ACC);
5627 		reg &= ~BGE_CPMU_HST_ACC_MACCLK_MASK;
5628 		reg |= BGE_CPMU_HST_ACC_MACCLK_6_25;
5629 		CSR_WRITE_4(sc, BGE_CPMU_HST_ACC, reg);
5630 	}
5631 
5632 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM57780) {
5633 		pcireg_t aercap;
5634 
5635 		reg = CSR_READ_4(sc, BGE_PCIE_PWRMNG_THRESH);
5636 		reg = (reg & ~BGE_PCIE_PWRMNG_L1THRESH_MASK)
5637 		    | BGE_PCIE_PWRMNG_L1THRESH_4MS
5638 		    | BGE_PCIE_PWRMNG_EXTASPMTMR_EN;
5639 		CSR_WRITE_4(sc, BGE_PCIE_PWRMNG_THRESH, reg);
5640 
5641 		reg = CSR_READ_4(sc, BGE_PCIE_EIDLE_DELAY);
5642 		reg = (reg & ~BGE_PCIE_EIDLE_DELAY_MASK)
5643 		    | BGE_PCIE_EIDLE_DELAY_13CLK;
5644 		CSR_WRITE_4(sc, BGE_PCIE_EIDLE_DELAY, reg);
5645 
5646 		/* Clear correctable error */
5647 		if (pci_get_ext_capability(sc->sc_pc, sc->sc_pcitag,
5648 		    PCI_EXTCAP_AER, &aercap, NULL) != 0)
5649 			pci_conf_write(sc->sc_pc, sc->sc_pcitag,
5650 			    aercap + PCI_AER_COR_STATUS, 0xffffffff);
5651 
5652 		reg = CSR_READ_4(sc, BGE_PCIE_LINKCTL);
5653 		reg = (reg & ~BGE_PCIE_LINKCTL_L1_PLL_PDEN)
5654 		    | BGE_PCIE_LINKCTL_L1_PLL_PDDIS;
5655 		CSR_WRITE_4(sc, BGE_PCIE_LINKCTL, reg);
5656 	}
5657 
5658 	bge_sig_post_reset(sc, BGE_RESET_START);
5659 
5660 	bge_chipinit(sc);
5661 
5662 	/*
5663 	 * Init the various state machines, ring
5664 	 * control blocks and firmware.
5665 	 */
5666 	error = bge_blockinit(sc);
5667 	if (error != 0) {
5668 		aprint_error_dev(sc->bge_dev, "initialization error %d\n",
5669 		    error);
5670 		return error;
5671 	}
5672 
5673 	/* 5718 step 25, 57XX step 54 */
5674 	/* Specify MTU. */
5675 	CSR_WRITE_4(sc, BGE_RX_MTU, ifp->if_mtu +
5676 	    ETHER_HDR_LEN + ETHER_CRC_LEN + ETHER_VLAN_ENCAP_LEN);
5677 
5678 	/* 5718 step 23 */
5679 	/* Load our MAC address. */
5680 	m = (const uint16_t *)&(CLLADDR(ifp->if_sadl)[0]);
5681 	CSR_WRITE_4(sc, BGE_MAC_ADDR1_LO, htons(m[0]));
5682 	CSR_WRITE_4(sc, BGE_MAC_ADDR1_HI,
5683 	    ((uint32_t)htons(m[1]) << 16) | htons(m[2]));
5684 
5685 	/* Enable or disable promiscuous mode as needed. */
5686 	if (ifp->if_flags & IFF_PROMISC)
5687 		BGE_SETBIT(sc, BGE_RX_MODE, BGE_RXMODE_RX_PROMISC);
5688 	else
5689 		BGE_CLRBIT(sc, BGE_RX_MODE, BGE_RXMODE_RX_PROMISC);
5690 
5691 	/* Program multicast filter. */
5692 	mutex_enter(sc->sc_mcast_lock);
5693 	bge_setmulti(sc);
5694 	mutex_exit(sc->sc_mcast_lock);
5695 
5696 	/* Init RX ring. */
5697 	bge_init_rx_ring_std(sc);
5698 
5699 	/*
5700 	 * Workaround for a bug in 5705 ASIC rev A0. Poll the NIC's
5701 	 * memory to insure that the chip has in fact read the first
5702 	 * entry of the ring.
5703 	 */
5704 	if (sc->bge_chipid == BGE_CHIPID_BCM5705_A0) {
5705 		u_int i;
5706 		for (i = 0; i < 10; i++) {
5707 			DELAY(20);
5708 			uint32_t v = bge_readmem_ind(sc, BGE_STD_RX_RINGS + 8);
5709 			if (v == (MCLBYTES - ETHER_ALIGN))
5710 				break;
5711 		}
5712 		if (i == 10)
5713 			aprint_error_dev(sc->bge_dev,
5714 			    "5705 A0 chip failed to load RX ring\n");
5715 	}
5716 
5717 	/* Init jumbo RX ring. */
5718 	if (ifp->if_mtu > (ETHERMTU + ETHER_HDR_LEN + ETHER_CRC_LEN))
5719 		bge_init_rx_ring_jumbo(sc);
5720 
5721 	/* Init our RX return ring index */
5722 	sc->bge_rx_saved_considx = 0;
5723 
5724 	/* Init TX ring. */
5725 	bge_init_tx_ring(sc);
5726 
5727 	/* 5718 step 63, 57XX step 94 */
5728 	/* Enable TX MAC state machine lockup fix. */
5729 	mode = CSR_READ_4(sc, BGE_TX_MODE);
5730 	if (BGE_IS_5755_PLUS(sc) ||
5731 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906)
5732 		mode |= BGE_TXMODE_MBUF_LOCKUP_FIX;
5733 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5720 ||
5734 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5762) {
5735 		mode &= ~(BGE_TXMODE_JMB_FRM_LEN | BGE_TXMODE_CNT_DN_MODE);
5736 		mode |= CSR_READ_4(sc, BGE_TX_MODE) &
5737 		    (BGE_TXMODE_JMB_FRM_LEN | BGE_TXMODE_CNT_DN_MODE);
5738 	}
5739 
5740 	/* Turn on transmitter */
5741 	CSR_WRITE_4_FLUSH(sc, BGE_TX_MODE, mode | BGE_TXMODE_ENABLE);
5742 	/* 5718 step 64 */
5743 	DELAY(100);
5744 
5745 	/* 5718 step 65, 57XX step 95 */
5746 	/* Turn on receiver */
5747 	mode = CSR_READ_4(sc, BGE_RX_MODE);
5748 	if (BGE_IS_5755_PLUS(sc))
5749 		mode |= BGE_RXMODE_IPV6_ENABLE;
5750 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5762)
5751 		mode |= BGE_RXMODE_IPV4_FRAG_FIX;
5752 	CSR_WRITE_4_FLUSH(sc, BGE_RX_MODE, mode | BGE_RXMODE_ENABLE);
5753 	/* 5718 step 66 */
5754 	DELAY(10);
5755 
5756 	/* 5718 step 12, 57XX step 37 */
5757 	/*
5758 	 * XXX Documents of 5718 series and 577xx say the recommended value
5759 	 * is 1, but tg3 set 1 only on 57765 series.
5760 	 */
5761 	if (BGE_IS_57765_PLUS(sc))
5762 		reg = 1;
5763 	else
5764 		reg = 2;
5765 	CSR_WRITE_4_FLUSH(sc, BGE_MAX_RX_FRAME_LOWAT, reg);
5766 
5767 	/* Tell firmware we're alive. */
5768 	BGE_SETBIT(sc, BGE_MODE_CTL, BGE_MODECTL_STACKUP);
5769 
5770 	/* Enable host interrupts. */
5771 	BGE_SETBIT(sc, BGE_PCI_MISC_CTL, BGE_PCIMISCCTL_CLEAR_INTA);
5772 	BGE_CLRBIT(sc, BGE_PCI_MISC_CTL, BGE_PCIMISCCTL_MASK_PCI_INTR);
5773 	bge_writembx_flush(sc, BGE_MBX_IRQ0_LO, 0);
5774 
5775 	mutex_enter(sc->sc_intr_lock);
5776 	if ((error = bge_ifmedia_upd(ifp)) == 0) {
5777 		sc->bge_txrx_stopping = false;
5778 
5779 		/* IFNET_LOCKED asserted above */
5780 		ifp->if_flags |= IFF_RUNNING;
5781 
5782 		callout_schedule(&sc->bge_timeout, hz);
5783 	}
5784 	mutex_exit(sc->sc_intr_lock);
5785 
5786 	mutex_enter(sc->sc_mcast_lock);
5787 	sc->bge_if_flags = ifp->if_flags;
5788 	mutex_exit(sc->sc_mcast_lock);
5789 
5790 	return error;
5791 }
5792 
5793 /*
5794  * Set media options.
5795  */
5796 static int
5797 bge_ifmedia_upd(struct ifnet *ifp)
5798 {
5799 	struct bge_softc * const sc = ifp->if_softc;
5800 	struct mii_data * const mii = &sc->bge_mii;
5801 	struct ifmedia * const ifm = &sc->bge_ifmedia;
5802 	int rc;
5803 
5804 	KASSERT(mutex_owned(sc->sc_intr_lock));
5805 
5806 	/* If this is a 1000baseX NIC, enable the TBI port. */
5807 	if (sc->bge_flags & BGEF_FIBER_TBI) {
5808 		if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER)
5809 			return EINVAL;
5810 		switch (IFM_SUBTYPE(ifm->ifm_media)) {
5811 		case IFM_AUTO:
5812 			/*
5813 			 * The BCM5704 ASIC appears to have a special
5814 			 * mechanism for programming the autoneg
5815 			 * advertisement registers in TBI mode.
5816 			 */
5817 			if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5704) {
5818 				uint32_t sgdig;
5819 				sgdig = CSR_READ_4(sc, BGE_SGDIG_STS);
5820 				if (sgdig & BGE_SGDIGSTS_DONE) {
5821 					CSR_WRITE_4(sc, BGE_TX_TBI_AUTONEG, 0);
5822 					sgdig = CSR_READ_4(sc, BGE_SGDIG_CFG);
5823 					sgdig |= BGE_SGDIGCFG_AUTO |
5824 					    BGE_SGDIGCFG_PAUSE_CAP |
5825 					    BGE_SGDIGCFG_ASYM_PAUSE;
5826 					CSR_WRITE_4_FLUSH(sc, BGE_SGDIG_CFG,
5827 					    sgdig | BGE_SGDIGCFG_SEND);
5828 					DELAY(5);
5829 					CSR_WRITE_4_FLUSH(sc, BGE_SGDIG_CFG,
5830 					    sgdig);
5831 				}
5832 			}
5833 			break;
5834 		case IFM_1000_SX:
5835 			if ((ifm->ifm_media & IFM_FDX) != 0) {
5836 				BGE_CLRBIT_FLUSH(sc, BGE_MAC_MODE,
5837 				    BGE_MACMODE_HALF_DUPLEX);
5838 			} else {
5839 				BGE_SETBIT_FLUSH(sc, BGE_MAC_MODE,
5840 				    BGE_MACMODE_HALF_DUPLEX);
5841 			}
5842 			DELAY(40);
5843 			break;
5844 		default:
5845 			return EINVAL;
5846 		}
5847 		/* XXX 802.3x flow control for 1000BASE-SX */
5848 		return 0;
5849 	}
5850 
5851 	if ((BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5784) &&
5852 	    (BGE_CHIPREV(sc->bge_chipid) != BGE_CHIPREV_5784_AX)) {
5853 		uint32_t reg;
5854 
5855 		reg = CSR_READ_4(sc, BGE_CPMU_CTRL);
5856 		if ((reg & BGE_CPMU_CTRL_GPHY_10MB_RXONLY) != 0) {
5857 			reg &= ~BGE_CPMU_CTRL_GPHY_10MB_RXONLY;
5858 			CSR_WRITE_4(sc, BGE_CPMU_CTRL, reg);
5859 		}
5860 	}
5861 
5862 	BGE_STS_SETBIT(sc, BGE_STS_LINK_EVT);
5863 	if ((rc = mii_mediachg(mii)) == ENXIO)
5864 		return 0;
5865 
5866 	if (BGE_CHIPREV(sc->bge_chipid) == BGE_CHIPREV_5784_AX) {
5867 		uint32_t reg;
5868 
5869 		reg = CSR_READ_4(sc, BGE_CPMU_LSPD_1000MB_CLK);
5870 		if ((reg & BGE_CPMU_LSPD_1000MB_MACCLK_MASK)
5871 		    == (BGE_CPMU_LSPD_1000MB_MACCLK_12_5)) {
5872 			reg &= ~BGE_CPMU_LSPD_1000MB_MACCLK_MASK;
5873 			delay(40);
5874 			CSR_WRITE_4(sc, BGE_CPMU_LSPD_1000MB_CLK, reg);
5875 		}
5876 	}
5877 
5878 	/*
5879 	 * Force an interrupt so that we will call bge_link_upd
5880 	 * if needed and clear any pending link state attention.
5881 	 * Without this we are not getting any further interrupts
5882 	 * for link state changes and thus will not UP the link and
5883 	 * not be able to send in bge_start. The only way to get
5884 	 * things working was to receive a packet and get a RX intr.
5885 	 */
5886 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5700 ||
5887 	    sc->bge_flags & BGEF_IS_5788)
5888 		BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_INTR_SET);
5889 	else
5890 		BGE_SETBIT(sc, BGE_HCC_MODE, BGE_HCCMODE_COAL_NOW);
5891 
5892 	return rc;
5893 }
5894 
5895 /*
5896  * Report current media status.
5897  */
5898 static void
5899 bge_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
5900 {
5901 	struct bge_softc * const sc = ifp->if_softc;
5902 	struct mii_data * const mii = &sc->bge_mii;
5903 
5904 	KASSERT(mutex_owned(sc->sc_intr_lock));
5905 
5906 	if (sc->bge_flags & BGEF_FIBER_TBI) {
5907 		ifmr->ifm_status = IFM_AVALID;
5908 		ifmr->ifm_active = IFM_ETHER;
5909 		if (CSR_READ_4(sc, BGE_MAC_STS) &
5910 		    BGE_MACSTAT_TBI_PCS_SYNCHED)
5911 			ifmr->ifm_status |= IFM_ACTIVE;
5912 		ifmr->ifm_active |= IFM_1000_SX;
5913 		if (CSR_READ_4(sc, BGE_MAC_MODE) & BGE_MACMODE_HALF_DUPLEX)
5914 			ifmr->ifm_active |= IFM_HDX;
5915 		else
5916 			ifmr->ifm_active |= IFM_FDX;
5917 		return;
5918 	}
5919 
5920 	mii_pollstat(mii);
5921 	ifmr->ifm_status = mii->mii_media_status;
5922 	ifmr->ifm_active = (mii->mii_media_active & ~IFM_ETH_FMASK) |
5923 	    sc->bge_flowflags;
5924 }
5925 
5926 static int
5927 bge_ifflags_cb(struct ethercom *ec)
5928 {
5929 	struct ifnet * const ifp = &ec->ec_if;
5930 	struct bge_softc * const sc = ifp->if_softc;
5931 	int ret = 0;
5932 
5933 	KASSERT(IFNET_LOCKED(ifp));
5934 	mutex_enter(sc->sc_mcast_lock);
5935 
5936 	u_short change = ifp->if_flags ^ sc->bge_if_flags;
5937 	sc->bge_if_flags = ifp->if_flags;
5938 
5939 	if ((change & ~(IFF_CANTCHANGE | IFF_DEBUG)) != 0) {
5940 		ret = ENETRESET;
5941 	} else if ((change & (IFF_PROMISC | IFF_ALLMULTI)) != 0) {
5942 		if ((ifp->if_flags & IFF_PROMISC) == 0)
5943 			BGE_CLRBIT(sc, BGE_RX_MODE, BGE_RXMODE_RX_PROMISC);
5944 		else
5945 			BGE_SETBIT(sc, BGE_RX_MODE, BGE_RXMODE_RX_PROMISC);
5946 
5947 		bge_setmulti(sc);
5948 	}
5949 
5950 	mutex_exit(sc->sc_mcast_lock);
5951 
5952 	return ret;
5953 }
5954 
5955 static int
5956 bge_ioctl(struct ifnet *ifp, u_long command, void *data)
5957 {
5958 	struct bge_softc * const sc = ifp->if_softc;
5959 	struct ifreq * const ifr = (struct ifreq *) data;
5960 	int error = 0;
5961 
5962 	switch (command) {
5963 	case SIOCADDMULTI:
5964 	case SIOCDELMULTI:
5965 		break;
5966 	default:
5967 		KASSERT(IFNET_LOCKED(ifp));
5968 	}
5969 
5970 	const int s = splnet();
5971 
5972 	switch (command) {
5973 	case SIOCSIFMEDIA:
5974 		mutex_enter(sc->sc_intr_lock);
5975 		/* XXX Flow control is not supported for 1000BASE-SX */
5976 		if (sc->bge_flags & BGEF_FIBER_TBI) {
5977 			ifr->ifr_media &= ~IFM_ETH_FMASK;
5978 			sc->bge_flowflags = 0;
5979 		}
5980 
5981 		/* Flow control requires full-duplex mode. */
5982 		if (IFM_SUBTYPE(ifr->ifr_media) == IFM_AUTO ||
5983 		    (ifr->ifr_media & IFM_FDX) == 0) {
5984 			ifr->ifr_media &= ~IFM_ETH_FMASK;
5985 		}
5986 		if (IFM_SUBTYPE(ifr->ifr_media) != IFM_AUTO) {
5987 			if ((ifr->ifr_media & IFM_ETH_FMASK) == IFM_FLOW) {
5988 				/* We can do both TXPAUSE and RXPAUSE. */
5989 				ifr->ifr_media |=
5990 				    IFM_ETH_TXPAUSE | IFM_ETH_RXPAUSE;
5991 			}
5992 			sc->bge_flowflags = ifr->ifr_media & IFM_ETH_FMASK;
5993 		}
5994 		mutex_exit(sc->sc_intr_lock);
5995 
5996 		if (sc->bge_flags & BGEF_FIBER_TBI) {
5997 			error = ifmedia_ioctl(ifp, ifr, &sc->bge_ifmedia,
5998 			    command);
5999 		} else {
6000 			struct mii_data * const mii = &sc->bge_mii;
6001 			error = ifmedia_ioctl(ifp, ifr, &mii->mii_media,
6002 			    command);
6003 		}
6004 		break;
6005 	default:
6006 		if ((error = ether_ioctl(ifp, command, data)) != ENETRESET)
6007 			break;
6008 
6009 		error = 0;
6010 
6011 		if (command == SIOCADDMULTI || command == SIOCDELMULTI) {
6012 			mutex_enter(sc->sc_mcast_lock);
6013 			if (sc->bge_if_flags & IFF_RUNNING) {
6014 				bge_setmulti(sc);
6015 			}
6016 			mutex_exit(sc->sc_mcast_lock);
6017 		}
6018 		break;
6019 	}
6020 
6021 	splx(s);
6022 
6023 	return error;
6024 }
6025 
6026 static bool
6027 bge_watchdog_check(struct bge_softc * const sc)
6028 {
6029 
6030 	KASSERT(mutex_owned(sc->sc_intr_lock));
6031 
6032 	if (!sc->bge_tx_sending)
6033 		return true;
6034 
6035 	if (time_uptime - sc->bge_tx_lastsent <= bge_watchdog_timeout)
6036 		return true;
6037 
6038 	/* If pause frames are active then don't reset the hardware. */
6039 	if ((CSR_READ_4(sc, BGE_RX_MODE) & BGE_RXMODE_FLOWCTL_ENABLE) != 0) {
6040 		const uint32_t status = CSR_READ_4(sc, BGE_RX_STS);
6041 		if ((status & BGE_RXSTAT_REMOTE_XOFFED) != 0) {
6042 			/*
6043 			 * If link partner has us in XOFF state then wait for
6044 			 * the condition to clear.
6045 			 */
6046 			CSR_WRITE_4(sc, BGE_RX_STS, status);
6047 			sc->bge_tx_lastsent = time_uptime;
6048 			return true;
6049 		} else if ((status & BGE_RXSTAT_RCVD_XOFF) != 0 &&
6050 		    (status & BGE_RXSTAT_RCVD_XON) != 0) {
6051 			/*
6052 			 * If link partner has us in XOFF state then wait for
6053 			 * the condition to clear.
6054 			 */
6055 			CSR_WRITE_4(sc, BGE_RX_STS, status);
6056 			sc->bge_tx_lastsent = time_uptime;
6057 			return true;
6058 		}
6059 		/*
6060 		 * Any other condition is unexpected and the controller
6061 		 * should be reset.
6062 		 */
6063 	}
6064 
6065 	return false;
6066 }
6067 
6068 static bool
6069 bge_watchdog_tick(struct ifnet *ifp)
6070 {
6071 	struct bge_softc * const sc = ifp->if_softc;
6072 
6073 	KASSERT(mutex_owned(sc->sc_intr_lock));
6074 
6075 	if (!sc->sc_trigger_reset && bge_watchdog_check(sc))
6076 		return true;
6077 
6078 	if (atomic_swap_uint(&sc->sc_reset_pending, 1) == 0)
6079 		workqueue_enqueue(sc->sc_reset_wq, &sc->sc_reset_work, NULL);
6080 
6081 	return false;
6082 }
6083 
6084 /*
6085  * Perform an interface watchdog reset.
6086  */
6087 static void
6088 bge_handle_reset_work(struct work *work, void *arg)
6089 {
6090 	struct bge_softc * const sc = arg;
6091 	struct ifnet * const ifp = &sc->ethercom.ec_if;
6092 
6093 	printf("%s: watchdog timeout -- resetting\n", ifp->if_xname);
6094 
6095 	/* Don't want ioctl operations to happen */
6096 	IFNET_LOCK(ifp);
6097 
6098 	/* reset the interface. */
6099 	bge_init(ifp);
6100 
6101 	IFNET_UNLOCK(ifp);
6102 
6103 	/*
6104 	 * There are still some upper layer processing which call
6105 	 * ifp->if_start(). e.g. ALTQ or one CPU system
6106 	 */
6107 	/* Try to get more packets going. */
6108 	ifp->if_start(ifp);
6109 
6110 	atomic_store_relaxed(&sc->sc_reset_pending, 0);
6111 }
6112 
6113 static void
6114 bge_stop_block(struct bge_softc *sc, bus_addr_t reg, uint32_t bit)
6115 {
6116 	int i;
6117 
6118 	BGE_CLRBIT_FLUSH(sc, reg, bit);
6119 
6120 	for (i = 0; i < 1000; i++) {
6121 		delay(100);
6122 		if ((CSR_READ_4(sc, reg) & bit) == 0)
6123 			return;
6124 	}
6125 
6126 	/*
6127 	 * Doesn't print only when the register is BGE_SRS_MODE. It occurs
6128 	 * on some environment (and once after boot?)
6129 	 */
6130 	if (reg != BGE_SRS_MODE)
6131 		aprint_error_dev(sc->bge_dev,
6132 		    "block failed to stop: reg 0x%lx, bit 0x%08x\n",
6133 		    (u_long)reg, bit);
6134 }
6135 
6136 /*
6137  * Stop the adapter and free any mbufs allocated to the
6138  * RX and TX lists.
6139  */
6140 static void
6141 bge_stop(struct ifnet *ifp, int disable)
6142 {
6143 	struct bge_softc * const sc = ifp->if_softc;
6144 
6145 	ASSERT_SLEEPABLE();
6146 	KASSERT(IFNET_LOCKED(ifp));
6147 
6148 	mutex_enter(sc->sc_intr_lock);
6149 	sc->bge_txrx_stopping = true;
6150 	mutex_exit(sc->sc_intr_lock);
6151 
6152 	callout_halt(&sc->bge_timeout, NULL);
6153 
6154 	/* Disable host interrupts. */
6155 	BGE_SETBIT(sc, BGE_PCI_MISC_CTL, BGE_PCIMISCCTL_MASK_PCI_INTR);
6156 	bge_writembx_flush(sc, BGE_MBX_IRQ0_LO, 1);
6157 
6158 	/*
6159 	 * Tell firmware we're shutting down.
6160 	 */
6161 	bge_stop_fw(sc);
6162 	bge_sig_pre_reset(sc, BGE_RESET_SHUTDOWN);
6163 
6164 	/*
6165 	 * Disable all of the receiver blocks.
6166 	 */
6167 	bge_stop_block(sc, BGE_RX_MODE, BGE_RXMODE_ENABLE);
6168 	bge_stop_block(sc, BGE_RBDI_MODE, BGE_RBDIMODE_ENABLE);
6169 	bge_stop_block(sc, BGE_RXLP_MODE, BGE_RXLPMODE_ENABLE);
6170 	if (BGE_IS_5700_FAMILY(sc))
6171 		bge_stop_block(sc, BGE_RXLS_MODE, BGE_RXLSMODE_ENABLE);
6172 	bge_stop_block(sc, BGE_RDBDI_MODE, BGE_RBDIMODE_ENABLE);
6173 	bge_stop_block(sc, BGE_RDC_MODE, BGE_RDCMODE_ENABLE);
6174 	bge_stop_block(sc, BGE_RBDC_MODE, BGE_RBDCMODE_ENABLE);
6175 
6176 	/*
6177 	 * Disable all of the transmit blocks.
6178 	 */
6179 	bge_stop_block(sc, BGE_SRS_MODE, BGE_SRSMODE_ENABLE);
6180 	bge_stop_block(sc, BGE_SBDI_MODE, BGE_SBDIMODE_ENABLE);
6181 	bge_stop_block(sc, BGE_SDI_MODE, BGE_SDIMODE_ENABLE);
6182 	bge_stop_block(sc, BGE_RDMA_MODE, BGE_RDMAMODE_ENABLE);
6183 	bge_stop_block(sc, BGE_SDC_MODE, BGE_SDCMODE_ENABLE);
6184 	if (BGE_IS_5700_FAMILY(sc))
6185 		bge_stop_block(sc, BGE_DMAC_MODE, BGE_DMACMODE_ENABLE);
6186 	bge_stop_block(sc, BGE_SBDC_MODE, BGE_SBDCMODE_ENABLE);
6187 
6188 	BGE_CLRBIT_FLUSH(sc, BGE_MAC_MODE, BGE_MACMODE_TXDMA_ENB);
6189 	delay(40);
6190 
6191 	bge_stop_block(sc, BGE_TX_MODE, BGE_TXMODE_ENABLE);
6192 
6193 	/*
6194 	 * Shut down all of the memory managers and related
6195 	 * state machines.
6196 	 */
6197 	/* 5718 step 5a,5b */
6198 	bge_stop_block(sc, BGE_HCC_MODE, BGE_HCCMODE_ENABLE);
6199 	bge_stop_block(sc, BGE_WDMA_MODE, BGE_WDMAMODE_ENABLE);
6200 	if (BGE_IS_5700_FAMILY(sc))
6201 		bge_stop_block(sc, BGE_MBCF_MODE, BGE_MBCFMODE_ENABLE);
6202 
6203 	/* 5718 step 5c,5d */
6204 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0xFFFFFFFF);
6205 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0);
6206 
6207 	if (BGE_IS_5700_FAMILY(sc)) {
6208 		bge_stop_block(sc, BGE_BMAN_MODE, BGE_BMANMODE_ENABLE);
6209 		bge_stop_block(sc, BGE_MARB_MODE, BGE_MARBMODE_ENABLE);
6210 	}
6211 
6212 	bge_reset(sc);
6213 	bge_sig_legacy(sc, BGE_RESET_SHUTDOWN);
6214 	bge_sig_post_reset(sc, BGE_RESET_SHUTDOWN);
6215 
6216 	/*
6217 	 * Keep the ASF firmware running if up.
6218 	 */
6219 	if (sc->bge_asf_mode & ASF_STACKUP)
6220 		BGE_SETBIT(sc, BGE_MODE_CTL, BGE_MODECTL_STACKUP);
6221 	else
6222 		BGE_CLRBIT(sc, BGE_MODE_CTL, BGE_MODECTL_STACKUP);
6223 
6224 	/* Free the RX lists. */
6225 	bge_free_rx_ring_std(sc);
6226 
6227 	/* Free jumbo RX list. */
6228 	if (BGE_IS_JUMBO_CAPABLE(sc))
6229 		bge_free_rx_ring_jumbo(sc);
6230 
6231 	/* Free TX buffers. */
6232 	bge_free_tx_ring(sc, disable);
6233 
6234 	/*
6235 	 * Isolate/power down the PHY.
6236 	 */
6237 	if (!(sc->bge_flags & BGEF_FIBER_TBI)) {
6238 		mutex_enter(sc->sc_intr_lock);
6239 		mii_down(&sc->bge_mii);
6240 		mutex_exit(sc->sc_intr_lock);
6241 	}
6242 
6243 	sc->bge_tx_saved_considx = BGE_TXCONS_UNSET;
6244 
6245 	/* Clear MAC's link state (PHY may still have link UP). */
6246 	BGE_STS_CLRBIT(sc, BGE_STS_LINK);
6247 
6248 	ifp->if_flags &= ~IFF_RUNNING;
6249 
6250 	mutex_enter(sc->sc_mcast_lock);
6251 	sc->bge_if_flags = ifp->if_flags;
6252 	mutex_exit(sc->sc_mcast_lock);
6253 }
6254 
6255 static void
6256 bge_link_upd(struct bge_softc *sc)
6257 {
6258 	struct ifnet * const ifp = &sc->ethercom.ec_if;
6259 	struct mii_data * const mii = &sc->bge_mii;
6260 	uint32_t status;
6261 	uint16_t phyval;
6262 	int link;
6263 
6264 	KASSERT(sc->sc_intr_lock);
6265 
6266 	/* Clear 'pending link event' flag */
6267 	BGE_STS_CLRBIT(sc, BGE_STS_LINK_EVT);
6268 
6269 	/*
6270 	 * Process link state changes.
6271 	 * Grrr. The link status word in the status block does
6272 	 * not work correctly on the BCM5700 rev AX and BX chips,
6273 	 * according to all available information. Hence, we have
6274 	 * to enable MII interrupts in order to properly obtain
6275 	 * async link changes. Unfortunately, this also means that
6276 	 * we have to read the MAC status register to detect link
6277 	 * changes, thereby adding an additional register access to
6278 	 * the interrupt handler.
6279 	 */
6280 
6281 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5700) {
6282 		status = CSR_READ_4(sc, BGE_MAC_STS);
6283 		if (status & BGE_MACSTAT_MI_INTERRUPT) {
6284 			mii_pollstat(mii);
6285 
6286 			if (!BGE_STS_BIT(sc, BGE_STS_LINK) &&
6287 			    mii->mii_media_status & IFM_ACTIVE &&
6288 			    IFM_SUBTYPE(mii->mii_media_active) != IFM_NONE)
6289 				BGE_STS_SETBIT(sc, BGE_STS_LINK);
6290 			else if (BGE_STS_BIT(sc, BGE_STS_LINK) &&
6291 			    (!(mii->mii_media_status & IFM_ACTIVE) ||
6292 			    IFM_SUBTYPE(mii->mii_media_active) == IFM_NONE))
6293 				BGE_STS_CLRBIT(sc, BGE_STS_LINK);
6294 
6295 			/* Clear the interrupt */
6296 			CSR_WRITE_4(sc, BGE_MAC_EVT_ENB,
6297 			    BGE_EVTENB_MI_INTERRUPT);
6298 			bge_miibus_readreg(sc->bge_dev, sc->bge_phy_addr,
6299 			    BRGPHY_MII_ISR, &phyval);
6300 			bge_miibus_writereg(sc->bge_dev, sc->bge_phy_addr,
6301 			    BRGPHY_MII_IMR, BRGPHY_INTRS);
6302 		}
6303 		return;
6304 	}
6305 
6306 	if (sc->bge_flags & BGEF_FIBER_TBI) {
6307 		status = CSR_READ_4(sc, BGE_MAC_STS);
6308 		if (status & BGE_MACSTAT_TBI_PCS_SYNCHED) {
6309 			if (!BGE_STS_BIT(sc, BGE_STS_LINK)) {
6310 				BGE_STS_SETBIT(sc, BGE_STS_LINK);
6311 				if (BGE_ASICREV(sc->bge_chipid)
6312 				    == BGE_ASICREV_BCM5704) {
6313 					BGE_CLRBIT_FLUSH(sc, BGE_MAC_MODE,
6314 					    BGE_MACMODE_TBI_SEND_CFGS);
6315 					DELAY(40);
6316 				}
6317 				CSR_WRITE_4(sc, BGE_MAC_STS, 0xFFFFFFFF);
6318 				if_link_state_change(ifp, LINK_STATE_UP);
6319 			}
6320 		} else if (BGE_STS_BIT(sc, BGE_STS_LINK)) {
6321 			BGE_STS_CLRBIT(sc, BGE_STS_LINK);
6322 			if_link_state_change(ifp, LINK_STATE_DOWN);
6323 		}
6324 	} else if (BGE_STS_BIT(sc, BGE_STS_AUTOPOLL)) {
6325 		/*
6326 		 * Some broken BCM chips have BGE_STATFLAG_LINKSTATE_CHANGED
6327 		 * bit in status word always set. Workaround this bug by
6328 		 * reading PHY link status directly.
6329 		 */
6330 		link = (CSR_READ_4(sc, BGE_MI_STS) & BGE_MISTS_LINK)?
6331 		    BGE_STS_LINK : 0;
6332 
6333 		if (BGE_STS_BIT(sc, BGE_STS_LINK) != link) {
6334 			mii_pollstat(mii);
6335 
6336 			if (!BGE_STS_BIT(sc, BGE_STS_LINK) &&
6337 			    mii->mii_media_status & IFM_ACTIVE &&
6338 			    IFM_SUBTYPE(mii->mii_media_active) != IFM_NONE)
6339 				BGE_STS_SETBIT(sc, BGE_STS_LINK);
6340 			else if (BGE_STS_BIT(sc, BGE_STS_LINK) &&
6341 			    (!(mii->mii_media_status & IFM_ACTIVE) ||
6342 			    IFM_SUBTYPE(mii->mii_media_active) == IFM_NONE))
6343 				BGE_STS_CLRBIT(sc, BGE_STS_LINK);
6344 		}
6345 	} else {
6346 		/*
6347 		 * For controllers that call mii_tick, we have to poll
6348 		 * link status.
6349 		 */
6350 		mii_pollstat(mii);
6351 	}
6352 
6353 	if (BGE_CHIPREV(sc->bge_chipid) == BGE_CHIPREV_5784_AX) {
6354 		uint32_t reg, scale;
6355 
6356 		reg = CSR_READ_4(sc, BGE_CPMU_CLCK_STAT) &
6357 		    BGE_CPMU_CLCK_STAT_MAC_CLCK_MASK;
6358 		if (reg == BGE_CPMU_CLCK_STAT_MAC_CLCK_62_5)
6359 			scale = 65;
6360 		else if (reg == BGE_CPMU_CLCK_STAT_MAC_CLCK_6_25)
6361 			scale = 6;
6362 		else
6363 			scale = 12;
6364 
6365 		reg = CSR_READ_4(sc, BGE_MISC_CFG) &
6366 		    ~BGE_MISCCFG_TIMER_PRESCALER;
6367 		reg |= scale << 1;
6368 		CSR_WRITE_4(sc, BGE_MISC_CFG, reg);
6369 	}
6370 	/* Clear the attention */
6371 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED |
6372 	    BGE_MACSTAT_CFG_CHANGED | BGE_MACSTAT_MI_COMPLETE |
6373 	    BGE_MACSTAT_LINK_CHANGED);
6374 }
6375 
6376 static int
6377 bge_sysctl_verify(SYSCTLFN_ARGS)
6378 {
6379 	int error, t;
6380 	struct sysctlnode node;
6381 
6382 	node = *rnode;
6383 	t = *(int*)rnode->sysctl_data;
6384 	node.sysctl_data = &t;
6385 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
6386 	if (error || newp == NULL)
6387 		return error;
6388 
6389 #if 0
6390 	DPRINTF2(("%s: t = %d, nodenum = %d, rnodenum = %d\n", __func__, t,
6391 	    node.sysctl_num, rnode->sysctl_num));
6392 #endif
6393 
6394 	if (node.sysctl_num == bge_rxthresh_nodenum) {
6395 		if (t < 0 || t >= NBGE_RX_THRESH)
6396 			return EINVAL;
6397 		bge_update_all_threshes(t);
6398 	} else
6399 		return EINVAL;
6400 
6401 	*(int*)rnode->sysctl_data = t;
6402 
6403 	return 0;
6404 }
6405 
6406 /*
6407  * Set up sysctl(3) MIB, hw.bge.*.
6408  */
6409 static void
6410 bge_sysctl_init(struct bge_softc *sc)
6411 {
6412 	int rc, bge_root_num;
6413 	const struct sysctlnode *node;
6414 
6415 	if ((rc = sysctl_createv(&sc->bge_log, 0, NULL, &node,
6416 	    0, CTLTYPE_NODE, "bge",
6417 	    SYSCTL_DESCR("BGE interface controls"),
6418 	    NULL, 0, NULL, 0, CTL_HW, CTL_CREATE, CTL_EOL)) != 0) {
6419 		goto out;
6420 	}
6421 
6422 	bge_root_num = node->sysctl_num;
6423 
6424 	/* BGE Rx interrupt mitigation level */
6425 	if ((rc = sysctl_createv(&sc->bge_log, 0, NULL, &node,
6426 	    CTLFLAG_READWRITE,
6427 	    CTLTYPE_INT, "rx_lvl",
6428 	    SYSCTL_DESCR("BGE receive interrupt mitigation level"),
6429 	    bge_sysctl_verify, 0,
6430 	    &bge_rx_thresh_lvl,
6431 	    0, CTL_HW, bge_root_num, CTL_CREATE,
6432 	    CTL_EOL)) != 0) {
6433 		goto out;
6434 	}
6435 
6436 	bge_rxthresh_nodenum = node->sysctl_num;
6437 
6438 #ifdef BGE_DEBUG
6439 	if ((rc = sysctl_createv(&sc->bge_log, 0, NULL, &node,
6440 	    CTLFLAG_READWRITE,
6441 	    CTLTYPE_BOOL, "trigger_reset",
6442 	    SYSCTL_DESCR("Trigger an interface reset"),
6443 	    NULL, 0, &sc->sc_trigger_reset, 0, CTL_CREATE,
6444 	    CTL_EOL)) != 0) {
6445 		goto out;
6446 	}
6447 #endif
6448 	return;
6449 
6450 out:
6451 	aprint_error("%s: sysctl_createv failed (rc = %d)\n", __func__, rc);
6452 }
6453 
6454 #ifdef BGE_DEBUG
6455 void
6456 bge_debug_info(struct bge_softc *sc)
6457 {
6458 
6459 	printf("Hardware Flags:\n");
6460 	if (BGE_IS_57765_PLUS(sc))
6461 		printf(" - 57765 Plus\n");
6462 	if (BGE_IS_5717_PLUS(sc))
6463 		printf(" - 5717 Plus\n");
6464 	if (BGE_IS_5755_PLUS(sc))
6465 		printf(" - 5755 Plus\n");
6466 	if (BGE_IS_575X_PLUS(sc))
6467 		printf(" - 575X Plus\n");
6468 	if (BGE_IS_5705_PLUS(sc))
6469 		printf(" - 5705 Plus\n");
6470 	if (BGE_IS_5714_FAMILY(sc))
6471 		printf(" - 5714 Family\n");
6472 	if (BGE_IS_5700_FAMILY(sc))
6473 		printf(" - 5700 Family\n");
6474 	if (sc->bge_flags & BGEF_IS_5788)
6475 		printf(" - 5788\n");
6476 	if (sc->bge_flags & BGEF_JUMBO_CAPABLE)
6477 		printf(" - Supports Jumbo Frames\n");
6478 	if (sc->bge_flags & BGEF_NO_EEPROM)
6479 		printf(" - No EEPROM\n");
6480 	if (sc->bge_flags & BGEF_PCIX)
6481 		printf(" - PCI-X Bus\n");
6482 	if (sc->bge_flags & BGEF_PCIE)
6483 		printf(" - PCI Express Bus\n");
6484 	if (sc->bge_flags & BGEF_RX_ALIGNBUG)
6485 		printf(" - RX Alignment Bug\n");
6486 	if (sc->bge_flags & BGEF_APE)
6487 		printf(" - APE\n");
6488 	if (sc->bge_flags & BGEF_CPMU_PRESENT)
6489 		printf(" - CPMU\n");
6490 	if (sc->bge_flags & BGEF_TSO)
6491 		printf(" - TSO\n");
6492 	if (sc->bge_flags & BGEF_TAGGED_STATUS)
6493 		printf(" - TAGGED_STATUS\n");
6494 
6495 	/* PHY related */
6496 	if (sc->bge_phy_flags & BGEPHYF_NO_3LED)
6497 		printf(" - No 3 LEDs\n");
6498 	if (sc->bge_phy_flags & BGEPHYF_CRC_BUG)
6499 		printf(" - CRC bug\n");
6500 	if (sc->bge_phy_flags & BGEPHYF_ADC_BUG)
6501 		printf(" - ADC bug\n");
6502 	if (sc->bge_phy_flags & BGEPHYF_5704_A0_BUG)
6503 		printf(" - 5704 A0 bug\n");
6504 	if (sc->bge_phy_flags & BGEPHYF_JITTER_BUG)
6505 		printf(" - jitter bug\n");
6506 	if (sc->bge_phy_flags & BGEPHYF_BER_BUG)
6507 		printf(" - BER bug\n");
6508 	if (sc->bge_phy_flags & BGEPHYF_ADJUST_TRIM)
6509 		printf(" - adjust trim\n");
6510 	if (sc->bge_phy_flags & BGEPHYF_NO_WIRESPEED)
6511 		printf(" - no wirespeed\n");
6512 
6513 	/* ASF related */
6514 	if (sc->bge_asf_mode & ASF_ENABLE)
6515 		printf(" - ASF enable\n");
6516 	if (sc->bge_asf_mode & ASF_NEW_HANDSHAKE)
6517 		printf(" - ASF new handshake\n");
6518 	if (sc->bge_asf_mode & ASF_STACKUP)
6519 		printf(" - ASF stackup\n");
6520 }
6521 #endif /* BGE_DEBUG */
6522 
6523 static int
6524 bge_get_eaddr_fw(struct bge_softc *sc, uint8_t ether_addr[])
6525 {
6526 	prop_dictionary_t dict;
6527 	prop_data_t ea;
6528 
6529 	if ((sc->bge_flags & BGEF_NO_EEPROM) == 0)
6530 		return 1;
6531 
6532 	dict = device_properties(sc->bge_dev);
6533 	ea = prop_dictionary_get(dict, "mac-address");
6534 	if (ea != NULL) {
6535 		KASSERT(prop_object_type(ea) == PROP_TYPE_DATA);
6536 		KASSERT(prop_data_size(ea) == ETHER_ADDR_LEN);
6537 		memcpy(ether_addr, prop_data_value(ea), ETHER_ADDR_LEN);
6538 		return 0;
6539 	}
6540 
6541 	return 1;
6542 }
6543 
6544 static int
6545 bge_get_eaddr_mem(struct bge_softc *sc, uint8_t ether_addr[])
6546 {
6547 	uint32_t mac_addr;
6548 
6549 	mac_addr = bge_readmem_ind(sc, BGE_SRAM_MAC_ADDR_HIGH_MB);
6550 	if ((mac_addr >> 16) == 0x484b) {
6551 		ether_addr[0] = (uint8_t)(mac_addr >> 8);
6552 		ether_addr[1] = (uint8_t)mac_addr;
6553 		mac_addr = bge_readmem_ind(sc, BGE_SRAM_MAC_ADDR_LOW_MB);
6554 		ether_addr[2] = (uint8_t)(mac_addr >> 24);
6555 		ether_addr[3] = (uint8_t)(mac_addr >> 16);
6556 		ether_addr[4] = (uint8_t)(mac_addr >> 8);
6557 		ether_addr[5] = (uint8_t)mac_addr;
6558 		return 0;
6559 	}
6560 	return 1;
6561 }
6562 
6563 static int
6564 bge_get_eaddr_nvram(struct bge_softc *sc, uint8_t ether_addr[])
6565 {
6566 	int mac_offset = BGE_EE_MAC_OFFSET;
6567 
6568 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906)
6569 		mac_offset = BGE_EE_MAC_OFFSET_5906;
6570 
6571 	return (bge_read_nvram(sc, ether_addr, mac_offset + 2,
6572 	    ETHER_ADDR_LEN));
6573 }
6574 
6575 static int
6576 bge_get_eaddr_eeprom(struct bge_softc *sc, uint8_t ether_addr[])
6577 {
6578 
6579 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906)
6580 		return 1;
6581 
6582 	return (bge_read_eeprom(sc, ether_addr, BGE_EE_MAC_OFFSET + 2,
6583 	   ETHER_ADDR_LEN));
6584 }
6585 
6586 static int
6587 bge_get_eaddr(struct bge_softc *sc, uint8_t eaddr[])
6588 {
6589 	static const bge_eaddr_fcn_t bge_eaddr_funcs[] = {
6590 		/* NOTE: Order is critical */
6591 		bge_get_eaddr_fw,
6592 		bge_get_eaddr_mem,
6593 		bge_get_eaddr_nvram,
6594 		bge_get_eaddr_eeprom,
6595 		NULL
6596 	};
6597 	const bge_eaddr_fcn_t *func;
6598 
6599 	for (func = bge_eaddr_funcs; *func != NULL; ++func) {
6600 		if ((*func)(sc, eaddr) == 0)
6601 			break;
6602 	}
6603 	return *func == NULL ? ENXIO : 0;
6604 }
6605