xref: /netbsd-src/sys/dev/pci/if_bge.c (revision 965ff70d6cc168e208e3ec6b725c8ce156e95fd0)
1 /*	$NetBSD: if_bge.c,v 1.397 2024/11/10 11:44:08 mlelstv 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 1.397 2024/11/10 11:44:08 mlelstv 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 	MCLAIM(m, &sc->ethercom.ec_rx_mowner);
1490 
1491 	MCLGET(m, M_DONTWAIT);
1492 	if (!(m->m_flags & M_EXT)) {
1493 		m_freem(m);
1494 		return ENOBUFS;
1495 	}
1496 	m->m_len = m->m_pkthdr.len = MCLBYTES;
1497 
1498 	if (!(sc->bge_flags & BGEF_RX_ALIGNBUG))
1499 	    m_adj(m, ETHER_ALIGN);
1500 	if (bus_dmamap_load_mbuf(sc->bge_dmatag, dmamap, m,
1501 	    BUS_DMA_READ | BUS_DMA_NOWAIT)) {
1502 		m_freem(m);
1503 		return ENOBUFS;
1504 	}
1505 	bus_dmamap_sync(sc->bge_dmatag, dmamap, 0, dmamap->dm_mapsize,
1506 	    BUS_DMASYNC_PREREAD);
1507 	sc->bge_cdata.bge_rx_std_chain[i] = m;
1508 
1509 	bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
1510 	    offsetof(struct bge_ring_data, bge_rx_std_ring) +
1511 		i * sizeof(struct bge_rx_bd),
1512 	    sizeof(struct bge_rx_bd),
1513 	    BUS_DMASYNC_POSTWRITE);
1514 
1515 	struct bge_rx_bd * const r = &sc->bge_rdata->bge_rx_std_ring[i];
1516 	BGE_HOSTADDR(r->bge_addr, dmamap->dm_segs[0].ds_addr);
1517 	r->bge_flags = BGE_RXBDFLAG_END;
1518 	r->bge_len = m->m_len;
1519 	r->bge_idx = i;
1520 
1521 	bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
1522 	    offsetof(struct bge_ring_data, bge_rx_std_ring) +
1523 		i * sizeof(struct bge_rx_bd),
1524 	    sizeof(struct bge_rx_bd),
1525 	    BUS_DMASYNC_PREWRITE);
1526 
1527 	sc->bge_std_cnt++;
1528 
1529 	return 0;
1530 }
1531 
1532 /*
1533  * Initialize a jumbo receive ring descriptor. This allocates
1534  * a jumbo buffer from the pool managed internally by the driver.
1535  */
1536 static int
1537 bge_newbuf_jumbo(struct bge_softc *sc, int i, struct mbuf *m)
1538 {
1539 	struct mbuf *m_new = NULL;
1540 	struct bge_rx_bd *r;
1541 	void *buf = NULL;
1542 
1543 	if (m == NULL) {
1544 
1545 		/* Allocate the mbuf. */
1546 		MGETHDR(m_new, M_DONTWAIT, MT_DATA);
1547 		if (m_new == NULL)
1548 			return ENOBUFS;
1549 		MCLAIM(m, &sc->ethercom.ec_rx_mowner);
1550 
1551 		/* Allocate the jumbo buffer */
1552 		buf = bge_jalloc(sc);
1553 		if (buf == NULL) {
1554 			m_freem(m_new);
1555 			aprint_error_dev(sc->bge_dev,
1556 			    "jumbo allocation failed -- packet dropped!\n");
1557 			return ENOBUFS;
1558 		}
1559 
1560 		/* Attach the buffer to the mbuf. */
1561 		m_new->m_len = m_new->m_pkthdr.len = BGE_JUMBO_FRAMELEN;
1562 		MEXTADD(m_new, buf, BGE_JUMBO_FRAMELEN, M_DEVBUF,
1563 		    bge_jfree, sc);
1564 		m_new->m_flags |= M_EXT_RW;
1565 	} else {
1566 		m_new = m;
1567 		buf = m_new->m_data = m_new->m_ext.ext_buf;
1568 		m_new->m_ext.ext_size = BGE_JUMBO_FRAMELEN;
1569 	}
1570 	if (!(sc->bge_flags & BGEF_RX_ALIGNBUG))
1571 	    m_adj(m_new, ETHER_ALIGN);
1572 	bus_dmamap_sync(sc->bge_dmatag, sc->bge_cdata.bge_rx_jumbo_map,
1573 	    mtod(m_new, char *) - (char *)sc->bge_cdata.bge_jumbo_buf,
1574 	    BGE_JLEN, BUS_DMASYNC_PREREAD);
1575 
1576 	/* Set up the descriptor. */
1577 	r = &sc->bge_rdata->bge_rx_jumbo_ring[i];
1578 	sc->bge_cdata.bge_rx_jumbo_chain[i] = m_new;
1579 	BGE_HOSTADDR(r->bge_addr, BGE_JUMBO_DMA_ADDR(sc, m_new));
1580 	r->bge_flags = BGE_RXBDFLAG_END | BGE_RXBDFLAG_JUMBO_RING;
1581 	r->bge_len = m_new->m_len;
1582 	r->bge_idx = i;
1583 
1584 	bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
1585 	    offsetof(struct bge_ring_data, bge_rx_jumbo_ring) +
1586 		i * sizeof(struct bge_rx_bd),
1587 	    sizeof(struct bge_rx_bd),
1588 	    BUS_DMASYNC_PREWRITE | BUS_DMASYNC_PREREAD);
1589 
1590 	return 0;
1591 }
1592 
1593 static int
1594 bge_init_rx_ring_std(struct bge_softc *sc)
1595 {
1596 	bus_dmamap_t dmamap;
1597 	int error = 0;
1598 	u_int i;
1599 
1600 	if (sc->bge_flags & BGEF_RXRING_VALID)
1601 		return 0;
1602 
1603 	for (i = 0; i < BGE_STD_RX_RING_CNT; i++) {
1604 		error = bus_dmamap_create(sc->bge_dmatag, MCLBYTES, 1,
1605 		    MCLBYTES, 0, BUS_DMA_WAITOK | BUS_DMA_ALLOCNOW, &dmamap);
1606 		if (error)
1607 			goto uncreate;
1608 
1609 		sc->bge_cdata.bge_rx_std_map[i] = dmamap;
1610 		memset(&sc->bge_rdata->bge_rx_std_ring[i], 0,
1611 		    sizeof(struct bge_rx_bd));
1612 	}
1613 
1614 	sc->bge_std = i - 1;
1615 	sc->bge_std_cnt = 0;
1616 	bge_fill_rx_ring_std(sc);
1617 
1618 	sc->bge_flags |= BGEF_RXRING_VALID;
1619 
1620 	return 0;
1621 
1622 uncreate:
1623 	while (--i) {
1624 		bus_dmamap_destroy(sc->bge_dmatag,
1625 		    sc->bge_cdata.bge_rx_std_map[i]);
1626 	}
1627 	return error;
1628 }
1629 
1630 static void
1631 bge_fill_rx_ring_std(struct bge_softc *sc)
1632 {
1633 	int i = sc->bge_std;
1634 	bool post = false;
1635 
1636 	while (sc->bge_std_cnt < BGE_STD_RX_RING_CNT) {
1637 		BGE_INC(i, BGE_STD_RX_RING_CNT);
1638 
1639 		if (bge_newbuf_std(sc, i) != 0)
1640 			break;
1641 
1642 		sc->bge_std = i;
1643 		post = true;
1644 	}
1645 
1646 	if (post)
1647 		bge_writembx(sc, BGE_MBX_RX_STD_PROD_LO, sc->bge_std);
1648 }
1649 
1650 
1651 static void
1652 bge_free_rx_ring_std(struct bge_softc *sc)
1653 {
1654 
1655 	if (!(sc->bge_flags & BGEF_RXRING_VALID))
1656 		return;
1657 
1658 	for (u_int i = 0; i < BGE_STD_RX_RING_CNT; i++) {
1659 		const bus_dmamap_t dmap = sc->bge_cdata.bge_rx_std_map[i];
1660 		struct mbuf * const m = sc->bge_cdata.bge_rx_std_chain[i];
1661 		if (m != NULL) {
1662 			bus_dmamap_sync(sc->bge_dmatag, dmap, 0,
1663 			    dmap->dm_mapsize, BUS_DMASYNC_POSTREAD);
1664 			bus_dmamap_unload(sc->bge_dmatag, dmap);
1665 			m_freem(m);
1666 			sc->bge_cdata.bge_rx_std_chain[i] = NULL;
1667 		}
1668 		bus_dmamap_destroy(sc->bge_dmatag,
1669 		    sc->bge_cdata.bge_rx_std_map[i]);
1670 		sc->bge_cdata.bge_rx_std_map[i] = NULL;
1671 		memset((char *)&sc->bge_rdata->bge_rx_std_ring[i], 0,
1672 		    sizeof(struct bge_rx_bd));
1673 	}
1674 
1675 	sc->bge_flags &= ~BGEF_RXRING_VALID;
1676 }
1677 
1678 static int
1679 bge_init_rx_ring_jumbo(struct bge_softc *sc)
1680 {
1681 	int i;
1682 	volatile struct bge_rcb *rcb;
1683 
1684 	if (sc->bge_flags & BGEF_JUMBO_RXRING_VALID)
1685 		return 0;
1686 
1687 	for (i = 0; i < BGE_JUMBO_RX_RING_CNT; i++) {
1688 		if (bge_newbuf_jumbo(sc, i, NULL) == ENOBUFS)
1689 			return ENOBUFS;
1690 	}
1691 
1692 	sc->bge_jumbo = i - 1;
1693 	sc->bge_flags |= BGEF_JUMBO_RXRING_VALID;
1694 
1695 	rcb = &sc->bge_rdata->bge_info.bge_jumbo_rx_rcb;
1696 	rcb->bge_maxlen_flags = 0;
1697 	CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_MAXLEN_FLAGS, rcb->bge_maxlen_flags);
1698 
1699 	bge_writembx(sc, BGE_MBX_RX_JUMBO_PROD_LO, sc->bge_jumbo);
1700 
1701 	return 0;
1702 }
1703 
1704 static void
1705 bge_free_rx_ring_jumbo(struct bge_softc *sc)
1706 {
1707 	int i;
1708 
1709 	if (!(sc->bge_flags & BGEF_JUMBO_RXRING_VALID))
1710 		return;
1711 
1712 	for (i = 0; i < BGE_JUMBO_RX_RING_CNT; i++) {
1713 		m_freem(sc->bge_cdata.bge_rx_jumbo_chain[i]);
1714 		sc->bge_cdata.bge_rx_jumbo_chain[i] = NULL;
1715 		memset((char *)&sc->bge_rdata->bge_rx_jumbo_ring[i], 0,
1716 		    sizeof(struct bge_rx_bd));
1717 	}
1718 
1719 	sc->bge_flags &= ~BGEF_JUMBO_RXRING_VALID;
1720 }
1721 
1722 static void
1723 bge_free_tx_ring(struct bge_softc *sc, bool disable)
1724 {
1725 	int i;
1726 	struct txdmamap_pool_entry *dma;
1727 
1728 	if (!(sc->bge_flags & BGEF_TXRING_VALID))
1729 		return;
1730 
1731 	for (i = 0; i < BGE_TX_RING_CNT; i++) {
1732 		if (sc->bge_cdata.bge_tx_chain[i] != NULL) {
1733 			m_freem(sc->bge_cdata.bge_tx_chain[i]);
1734 			sc->bge_cdata.bge_tx_chain[i] = NULL;
1735 			SLIST_INSERT_HEAD(&sc->txdma_list, sc->txdma[i],
1736 					    link);
1737 			sc->txdma[i] = 0;
1738 		}
1739 		memset((char *)&sc->bge_rdata->bge_tx_ring[i], 0,
1740 		    sizeof(struct bge_tx_bd));
1741 	}
1742 
1743 	if (disable) {
1744 		while ((dma = SLIST_FIRST(&sc->txdma_list))) {
1745 			SLIST_REMOVE_HEAD(&sc->txdma_list, link);
1746 			bus_dmamap_destroy(sc->bge_dmatag, dma->dmamap);
1747 			if (sc->bge_dma64) {
1748 				bus_dmamap_destroy(sc->bge_dmatag32,
1749 				    dma->dmamap32);
1750 			}
1751 			kmem_free(dma, sizeof(*dma));
1752 		}
1753 		SLIST_INIT(&sc->txdma_list);
1754 	}
1755 
1756 	sc->bge_flags &= ~BGEF_TXRING_VALID;
1757 }
1758 
1759 static int
1760 bge_init_tx_ring(struct bge_softc *sc)
1761 {
1762 	struct ifnet * const ifp = &sc->ethercom.ec_if;
1763 	int i;
1764 	bus_dmamap_t dmamap, dmamap32;
1765 	bus_size_t maxsegsz;
1766 	struct txdmamap_pool_entry *dma;
1767 
1768 	if (sc->bge_flags & BGEF_TXRING_VALID)
1769 		return 0;
1770 
1771 	sc->bge_txcnt = 0;
1772 	sc->bge_tx_saved_considx = 0;
1773 
1774 	/* Initialize transmit producer index for host-memory send ring. */
1775 	sc->bge_tx_prodidx = 0;
1776 	bge_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, sc->bge_tx_prodidx);
1777 	/* 5700 b2 errata */
1778 	if (BGE_CHIPREV(sc->bge_chipid) == BGE_CHIPREV_5700_BX)
1779 		bge_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, sc->bge_tx_prodidx);
1780 
1781 	/* NIC-memory send ring not used; initialize to zero. */
1782 	bge_writembx(sc, BGE_MBX_TX_NIC_PROD0_LO, 0);
1783 	/* 5700 b2 errata */
1784 	if (BGE_CHIPREV(sc->bge_chipid) == BGE_CHIPREV_5700_BX)
1785 		bge_writembx(sc, BGE_MBX_TX_NIC_PROD0_LO, 0);
1786 
1787 	/* Limit DMA segment size for some chips */
1788 	if ((BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM57766) &&
1789 	    (ifp->if_mtu <= ETHERMTU))
1790 		maxsegsz = 2048;
1791 	else if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5719)
1792 		maxsegsz = 4096;
1793 	else
1794 		maxsegsz = ETHER_MAX_LEN_JUMBO;
1795 
1796 	if (SLIST_FIRST(&sc->txdma_list) != NULL)
1797 		goto alloc_done;
1798 
1799 	for (i = 0; i < BGE_TX_RING_CNT; i++) {
1800 		if (bus_dmamap_create(sc->bge_dmatag, BGE_TXDMA_MAX,
1801 		    BGE_NTXSEG, maxsegsz, 0, BUS_DMA_WAITOK | BUS_DMA_ALLOCNOW,
1802 		    &dmamap))
1803 			return ENOBUFS;
1804 		if (dmamap == NULL)
1805 			panic("dmamap NULL in bge_init_tx_ring");
1806 		if (sc->bge_dma64) {
1807 			if (bus_dmamap_create(sc->bge_dmatag32, BGE_TXDMA_MAX,
1808 			    BGE_NTXSEG, maxsegsz, 0,
1809 			    BUS_DMA_WAITOK | BUS_DMA_ALLOCNOW,
1810 			    &dmamap32)) {
1811 				bus_dmamap_destroy(sc->bge_dmatag, dmamap);
1812 				return ENOBUFS;
1813 			}
1814 			if (dmamap32 == NULL)
1815 				panic("dmamap32 NULL in bge_init_tx_ring");
1816 		} else
1817 			dmamap32 = dmamap;
1818 		dma = kmem_alloc(sizeof(*dma), KM_NOSLEEP);
1819 		if (dma == NULL) {
1820 			aprint_error_dev(sc->bge_dev,
1821 			    "can't alloc txdmamap_pool_entry\n");
1822 			bus_dmamap_destroy(sc->bge_dmatag, dmamap);
1823 			if (sc->bge_dma64)
1824 				bus_dmamap_destroy(sc->bge_dmatag32, dmamap32);
1825 			return ENOMEM;
1826 		}
1827 		dma->dmamap = dmamap;
1828 		dma->dmamap32 = dmamap32;
1829 		SLIST_INSERT_HEAD(&sc->txdma_list, dma, link);
1830 	}
1831 alloc_done:
1832 	sc->bge_flags |= BGEF_TXRING_VALID;
1833 
1834 	return 0;
1835 }
1836 
1837 static void
1838 bge_setmulti(struct bge_softc *sc)
1839 {
1840 	struct ethercom * const ec = &sc->ethercom;
1841 	struct ether_multi	*enm;
1842 	struct ether_multistep	step;
1843 	uint32_t		hashes[4] = { 0, 0, 0, 0 };
1844 	uint32_t		h;
1845 	int			i;
1846 
1847 	KASSERT(mutex_owned(sc->sc_mcast_lock));
1848 	if (sc->bge_if_flags & IFF_PROMISC)
1849 		goto allmulti;
1850 
1851 	/* Now program new ones. */
1852 	ETHER_LOCK(ec);
1853 	ETHER_FIRST_MULTI(step, ec, enm);
1854 	while (enm != NULL) {
1855 		if (memcmp(enm->enm_addrlo, enm->enm_addrhi, ETHER_ADDR_LEN)) {
1856 			/*
1857 			 * We must listen to a range of multicast addresses.
1858 			 * For now, just accept all multicasts, rather than
1859 			 * trying to set only those filter bits needed to match
1860 			 * the range.  (At this time, the only use of address
1861 			 * ranges is for IP multicast routing, for which the
1862 			 * range is big enough to require all bits set.)
1863 			 */
1864 			ETHER_UNLOCK(ec);
1865 			goto allmulti;
1866 		}
1867 
1868 		h = ether_crc32_le(enm->enm_addrlo, ETHER_ADDR_LEN);
1869 
1870 		/* Just want the 7 least-significant bits. */
1871 		h &= 0x7f;
1872 
1873 		hashes[(h & 0x60) >> 5] |= 1U << (h & 0x1F);
1874 		ETHER_NEXT_MULTI(step, enm);
1875 	}
1876 	ec->ec_flags &= ~ETHER_F_ALLMULTI;
1877 	ETHER_UNLOCK(ec);
1878 
1879 	goto setit;
1880 
1881  allmulti:
1882 	ETHER_LOCK(ec);
1883 	ec->ec_flags |= ETHER_F_ALLMULTI;
1884 	ETHER_UNLOCK(ec);
1885 	hashes[0] = hashes[1] = hashes[2] = hashes[3] = 0xffffffff;
1886 
1887  setit:
1888 	for (i = 0; i < 4; i++)
1889 		CSR_WRITE_4(sc, BGE_MAR0 + (i * 4), hashes[i]);
1890 }
1891 
1892 static void
1893 bge_sig_pre_reset(struct bge_softc *sc, int type)
1894 {
1895 
1896 	/*
1897 	 * Some chips don't like this so only do this if ASF is enabled
1898 	 */
1899 	if (sc->bge_asf_mode)
1900 		bge_writemem_ind(sc, BGE_SRAM_FW_MB, BGE_SRAM_FW_MB_MAGIC);
1901 
1902 	if (sc->bge_asf_mode & ASF_NEW_HANDSHAKE) {
1903 		switch (type) {
1904 		case BGE_RESET_START:
1905 			bge_writemem_ind(sc, BGE_SRAM_FW_DRV_STATE_MB,
1906 			    BGE_FW_DRV_STATE_START);
1907 			break;
1908 		case BGE_RESET_SHUTDOWN:
1909 			bge_writemem_ind(sc, BGE_SRAM_FW_DRV_STATE_MB,
1910 			    BGE_FW_DRV_STATE_UNLOAD);
1911 			break;
1912 		case BGE_RESET_SUSPEND:
1913 			bge_writemem_ind(sc, BGE_SRAM_FW_DRV_STATE_MB,
1914 			    BGE_FW_DRV_STATE_SUSPEND);
1915 			break;
1916 		}
1917 	}
1918 
1919 	if (type == BGE_RESET_START || type == BGE_RESET_SUSPEND)
1920 		bge_ape_driver_state_change(sc, type);
1921 }
1922 
1923 static void
1924 bge_sig_post_reset(struct bge_softc *sc, int type)
1925 {
1926 
1927 	if (sc->bge_asf_mode & ASF_NEW_HANDSHAKE) {
1928 		switch (type) {
1929 		case BGE_RESET_START:
1930 			bge_writemem_ind(sc, BGE_SRAM_FW_DRV_STATE_MB,
1931 			    BGE_FW_DRV_STATE_START_DONE);
1932 			/* START DONE */
1933 			break;
1934 		case BGE_RESET_SHUTDOWN:
1935 			bge_writemem_ind(sc, BGE_SRAM_FW_DRV_STATE_MB,
1936 			    BGE_FW_DRV_STATE_UNLOAD_DONE);
1937 			break;
1938 		}
1939 	}
1940 
1941 	if (type == BGE_RESET_SHUTDOWN)
1942 		bge_ape_driver_state_change(sc, type);
1943 }
1944 
1945 static void
1946 bge_sig_legacy(struct bge_softc *sc, int type)
1947 {
1948 
1949 	if (sc->bge_asf_mode) {
1950 		switch (type) {
1951 		case BGE_RESET_START:
1952 			bge_writemem_ind(sc, BGE_SRAM_FW_DRV_STATE_MB,
1953 			    BGE_FW_DRV_STATE_START);
1954 			break;
1955 		case BGE_RESET_SHUTDOWN:
1956 			bge_writemem_ind(sc, BGE_SRAM_FW_DRV_STATE_MB,
1957 			    BGE_FW_DRV_STATE_UNLOAD);
1958 			break;
1959 		}
1960 	}
1961 }
1962 
1963 static void
1964 bge_wait_for_event_ack(struct bge_softc *sc)
1965 {
1966 	int i;
1967 
1968 	/* wait up to 2500usec */
1969 	for (i = 0; i < 250; i++) {
1970 		if (!(CSR_READ_4(sc, BGE_RX_CPU_EVENT) &
1971 			BGE_RX_CPU_DRV_EVENT))
1972 			break;
1973 		DELAY(10);
1974 	}
1975 }
1976 
1977 static void
1978 bge_stop_fw(struct bge_softc *sc)
1979 {
1980 
1981 	if (sc->bge_asf_mode) {
1982 		bge_wait_for_event_ack(sc);
1983 
1984 		bge_writemem_ind(sc, BGE_SRAM_FW_CMD_MB, BGE_FW_CMD_PAUSE);
1985 		CSR_WRITE_4_FLUSH(sc, BGE_RX_CPU_EVENT,
1986 		    CSR_READ_4(sc, BGE_RX_CPU_EVENT) | BGE_RX_CPU_DRV_EVENT);
1987 
1988 		bge_wait_for_event_ack(sc);
1989 	}
1990 }
1991 
1992 static int
1993 bge_poll_fw(struct bge_softc *sc)
1994 {
1995 	uint32_t val;
1996 	int i;
1997 
1998 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906) {
1999 		for (i = 0; i < BGE_TIMEOUT; i++) {
2000 			val = CSR_READ_4(sc, BGE_VCPU_STATUS);
2001 			if (val & BGE_VCPU_STATUS_INIT_DONE)
2002 				break;
2003 			DELAY(100);
2004 		}
2005 		if (i >= BGE_TIMEOUT) {
2006 			aprint_error_dev(sc->bge_dev, "reset timed out\n");
2007 			return -1;
2008 		}
2009 	} else {
2010 		/*
2011 		 * Poll the value location we just wrote until
2012 		 * we see the 1's complement of the magic number.
2013 		 * This indicates that the firmware initialization
2014 		 * is complete.
2015 		 * XXX 1000ms for Flash and 10000ms for SEEPROM.
2016 		 */
2017 		for (i = 0; i < BGE_TIMEOUT; i++) {
2018 			val = bge_readmem_ind(sc, BGE_SRAM_FW_MB);
2019 			if (val == ~BGE_SRAM_FW_MB_MAGIC)
2020 				break;
2021 			DELAY(10);
2022 		}
2023 
2024 		if ((i >= BGE_TIMEOUT)
2025 		    && ((sc->bge_flags & BGEF_NO_EEPROM) == 0)) {
2026 			aprint_error_dev(sc->bge_dev,
2027 			    "firmware handshake timed out, val = %x\n", val);
2028 			return -1;
2029 		}
2030 	}
2031 
2032 	if (sc->bge_chipid == BGE_CHIPID_BCM57765_A0) {
2033 		/* tg3 says we have to wait extra time */
2034 		delay(10 * 1000);
2035 	}
2036 
2037 	return 0;
2038 }
2039 
2040 int
2041 bge_phy_addr(struct bge_softc *sc)
2042 {
2043 	struct pci_attach_args *pa = &(sc->bge_pa);
2044 	int phy_addr = 1;
2045 
2046 	/*
2047 	 * PHY address mapping for various devices.
2048 	 *
2049 	 *	    | F0 Cu | F0 Sr | F1 Cu | F1 Sr |
2050 	 * ---------+-------+-------+-------+-------+
2051 	 * BCM57XX  |	1   |	X   |	X   |	X   |
2052 	 * BCM5704  |	1   |	X   |	1   |	X   |
2053 	 * BCM5717  |	1   |	8   |	2   |	9   |
2054 	 * BCM5719  |	1   |	8   |	2   |	9   |
2055 	 * BCM5720  |	1   |	8   |	2   |	9   |
2056 	 *
2057 	 *	    | F2 Cu | F2 Sr | F3 Cu | F3 Sr |
2058 	 * ---------+-------+-------+-------+-------+
2059 	 * BCM57XX  |	X   |	X   |	X   |	X   |
2060 	 * BCM5704  |	X   |	X   |	X   |	X   |
2061 	 * BCM5717  |	X   |	X   |	X   |	X   |
2062 	 * BCM5719  |	3   |	10  |	4   |	11  |
2063 	 * BCM5720  |	X   |	X   |	X   |	X   |
2064 	 *
2065 	 * Other addresses may respond but they are not
2066 	 * IEEE compliant PHYs and should be ignored.
2067 	 */
2068 	switch (BGE_ASICREV(sc->bge_chipid)) {
2069 	case BGE_ASICREV_BCM5717:
2070 	case BGE_ASICREV_BCM5719:
2071 	case BGE_ASICREV_BCM5720:
2072 		phy_addr = pa->pa_function;
2073 		if (sc->bge_chipid != BGE_CHIPID_BCM5717_A0) {
2074 			phy_addr += (CSR_READ_4(sc, BGE_SGDIG_STS) &
2075 			    BGE_SGDIGSTS_IS_SERDES) ? 8 : 1;
2076 		} else {
2077 			phy_addr += (CSR_READ_4(sc, BGE_CPMU_PHY_STRAP) &
2078 			    BGE_CPMU_PHY_STRAP_IS_SERDES) ? 8 : 1;
2079 		}
2080 	}
2081 
2082 	return phy_addr;
2083 }
2084 
2085 /*
2086  * Do endian, PCI and DMA initialization. Also check the on-board ROM
2087  * self-test results.
2088  */
2089 static int
2090 bge_chipinit(struct bge_softc *sc)
2091 {
2092 	uint32_t dma_rw_ctl, misc_ctl, mode_ctl, reg;
2093 	int i;
2094 
2095 	/* Set endianness before we access any non-PCI registers. */
2096 	misc_ctl = BGE_INIT;
2097 	if (sc->bge_flags & BGEF_TAGGED_STATUS)
2098 		misc_ctl |= BGE_PCIMISCCTL_TAGGED_STATUS;
2099 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MISC_CTL,
2100 	    misc_ctl);
2101 
2102 	/*
2103 	 * Clear the MAC statistics block in the NIC's
2104 	 * internal memory.
2105 	 */
2106 	for (i = BGE_STATS_BLOCK;
2107 	    i < BGE_STATS_BLOCK_END + 1; i += sizeof(uint32_t))
2108 		BGE_MEMWIN_WRITE(sc->sc_pc, sc->sc_pcitag, i, 0);
2109 
2110 	for (i = BGE_STATUS_BLOCK;
2111 	    i < BGE_STATUS_BLOCK_END + 1; i += sizeof(uint32_t))
2112 		BGE_MEMWIN_WRITE(sc->sc_pc, sc->sc_pcitag, i, 0);
2113 
2114 	/* 5717 workaround from tg3 */
2115 	if (sc->bge_chipid == BGE_CHIPID_BCM5717_A0) {
2116 		/* Save */
2117 		mode_ctl = CSR_READ_4(sc, BGE_MODE_CTL);
2118 
2119 		/* Temporary modify MODE_CTL to control TLP */
2120 		reg = mode_ctl & ~BGE_MODECTL_PCIE_TLPADDRMASK;
2121 		CSR_WRITE_4(sc, BGE_MODE_CTL, reg | BGE_MODECTL_PCIE_TLPADDR1);
2122 
2123 		/* Control TLP */
2124 		reg = CSR_READ_4(sc, BGE_TLP_CONTROL_REG +
2125 		    BGE_TLP_PHYCTL1);
2126 		CSR_WRITE_4(sc, BGE_TLP_CONTROL_REG + BGE_TLP_PHYCTL1,
2127 		    reg | BGE_TLP_PHYCTL1_EN_L1PLLPD);
2128 
2129 		/* Restore */
2130 		CSR_WRITE_4(sc, BGE_MODE_CTL, mode_ctl);
2131 	}
2132 
2133 	if (BGE_IS_57765_FAMILY(sc)) {
2134 		if (sc->bge_chipid == BGE_CHIPID_BCM57765_A0) {
2135 			/* Save */
2136 			mode_ctl = CSR_READ_4(sc, BGE_MODE_CTL);
2137 
2138 			/* Temporary modify MODE_CTL to control TLP */
2139 			reg = mode_ctl & ~BGE_MODECTL_PCIE_TLPADDRMASK;
2140 			CSR_WRITE_4(sc, BGE_MODE_CTL,
2141 			    reg | BGE_MODECTL_PCIE_TLPADDR1);
2142 
2143 			/* Control TLP */
2144 			reg = CSR_READ_4(sc, BGE_TLP_CONTROL_REG +
2145 			    BGE_TLP_PHYCTL5);
2146 			CSR_WRITE_4(sc, BGE_TLP_CONTROL_REG + BGE_TLP_PHYCTL5,
2147 			    reg | BGE_TLP_PHYCTL5_DIS_L2CLKREQ);
2148 
2149 			/* Restore */
2150 			CSR_WRITE_4(sc, BGE_MODE_CTL, mode_ctl);
2151 		}
2152 		if (BGE_CHIPREV(sc->bge_chipid) != BGE_CHIPREV_57765_AX) {
2153 			/*
2154 			 * For the 57766 and non Ax versions of 57765, bootcode
2155 			 * needs to setup the PCIE Fast Training Sequence (FTS)
2156 			 * value to prevent transmit hangs.
2157 			 */
2158 			reg = CSR_READ_4(sc, BGE_CPMU_PADRNG_CTL);
2159 			CSR_WRITE_4(sc, BGE_CPMU_PADRNG_CTL,
2160 			    reg | BGE_CPMU_PADRNG_CTL_RDIV2);
2161 
2162 			/* Save */
2163 			mode_ctl = CSR_READ_4(sc, BGE_MODE_CTL);
2164 
2165 			/* Temporary modify MODE_CTL to control TLP */
2166 			reg = mode_ctl & ~BGE_MODECTL_PCIE_TLPADDRMASK;
2167 			CSR_WRITE_4(sc, BGE_MODE_CTL,
2168 			    reg | BGE_MODECTL_PCIE_TLPADDR0);
2169 
2170 			/* Control TLP */
2171 			reg = CSR_READ_4(sc, BGE_TLP_CONTROL_REG +
2172 			    BGE_TLP_FTSMAX);
2173 			reg &= ~BGE_TLP_FTSMAX_MSK;
2174 			CSR_WRITE_4(sc, BGE_TLP_CONTROL_REG + BGE_TLP_FTSMAX,
2175 			    reg | BGE_TLP_FTSMAX_VAL);
2176 
2177 			/* Restore */
2178 			CSR_WRITE_4(sc, BGE_MODE_CTL, mode_ctl);
2179 		}
2180 
2181 		reg = CSR_READ_4(sc, BGE_CPMU_LSPD_10MB_CLK);
2182 		reg &= ~BGE_CPMU_LSPD_10MB_MACCLK_MASK;
2183 		reg |= BGE_CPMU_LSPD_10MB_MACCLK_6_25;
2184 		CSR_WRITE_4(sc, BGE_CPMU_LSPD_10MB_CLK, reg);
2185 	}
2186 
2187 	/* Set up the PCI DMA control register. */
2188 	dma_rw_ctl = BGE_PCI_READ_CMD | BGE_PCI_WRITE_CMD;
2189 	if (sc->bge_flags & BGEF_PCIE) {
2190 		/* Read watermark not used, 128 bytes for write. */
2191 		DPRINTFN(4, ("(%s: PCI-Express DMA setting)\n",
2192 		    device_xname(sc->bge_dev)));
2193 		if (sc->bge_mps >= 256)
2194 			dma_rw_ctl |= BGE_PCIDMARWCTL_WR_WAT_SHIFT(7);
2195 		else
2196 			dma_rw_ctl |= BGE_PCIDMARWCTL_WR_WAT_SHIFT(3);
2197 	} else if (sc->bge_flags & BGEF_PCIX) {
2198 		DPRINTFN(4, ("(:%s: PCI-X DMA setting)\n",
2199 		    device_xname(sc->bge_dev)));
2200 		/* PCI-X bus */
2201 		if (BGE_IS_5714_FAMILY(sc)) {
2202 			/* 256 bytes for read and write. */
2203 			dma_rw_ctl |= BGE_PCIDMARWCTL_RD_WAT_SHIFT(2) |
2204 			    BGE_PCIDMARWCTL_WR_WAT_SHIFT(2);
2205 
2206 			if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5780)
2207 				dma_rw_ctl |= BGE_PCIDMARWCTL_ONEDMA_ATONCE_GLOBAL;
2208 			else
2209 				dma_rw_ctl |= BGE_PCIDMARWCTL_ONEDMA_ATONCE_LOCAL;
2210 		} else if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5703) {
2211 			/*
2212 			 * In the BCM5703, the DMA read watermark should
2213 			 * be set to less than or equal to the maximum
2214 			 * memory read byte count of the PCI-X command
2215 			 * register.
2216 			 */
2217 			dma_rw_ctl |= BGE_PCIDMARWCTL_RD_WAT_SHIFT(4) |
2218 			    BGE_PCIDMARWCTL_WR_WAT_SHIFT(3);
2219 		} else if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5704) {
2220 			/* 1536 bytes for read, 384 bytes for write. */
2221 			dma_rw_ctl |= BGE_PCIDMARWCTL_RD_WAT_SHIFT(7) |
2222 			    BGE_PCIDMARWCTL_WR_WAT_SHIFT(3);
2223 		} else {
2224 			/* 384 bytes for read and write. */
2225 			dma_rw_ctl |= BGE_PCIDMARWCTL_RD_WAT_SHIFT(3) |
2226 			    BGE_PCIDMARWCTL_WR_WAT_SHIFT(3) |
2227 			    (0x0F);
2228 		}
2229 
2230 		if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5703 ||
2231 		    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5704) {
2232 			uint32_t tmp;
2233 
2234 			/* Set ONEDMA_ATONCE for hardware workaround. */
2235 			tmp = CSR_READ_4(sc, BGE_PCI_CLKCTL) & 0x1f;
2236 			if (tmp == 6 || tmp == 7)
2237 				dma_rw_ctl |=
2238 				    BGE_PCIDMARWCTL_ONEDMA_ATONCE_GLOBAL;
2239 
2240 			/* Set PCI-X DMA write workaround. */
2241 			dma_rw_ctl |= BGE_PCIDMARWCTL_ASRT_ALL_BE;
2242 		}
2243 	} else {
2244 		/* Conventional PCI bus: 256 bytes for read and write. */
2245 		DPRINTFN(4, ("(%s: PCI 2.2 DMA setting)\n",
2246 		    device_xname(sc->bge_dev)));
2247 		dma_rw_ctl |= BGE_PCIDMARWCTL_RD_WAT_SHIFT(7) |
2248 		    BGE_PCIDMARWCTL_WR_WAT_SHIFT(7);
2249 
2250 		if (BGE_ASICREV(sc->bge_chipid) != BGE_ASICREV_BCM5705 &&
2251 		    BGE_ASICREV(sc->bge_chipid) != BGE_ASICREV_BCM5750)
2252 			dma_rw_ctl |= 0x0F;
2253 	}
2254 
2255 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5700 ||
2256 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5701)
2257 		dma_rw_ctl |= BGE_PCIDMARWCTL_USE_MRM |
2258 		    BGE_PCIDMARWCTL_ASRT_ALL_BE;
2259 
2260 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5703 ||
2261 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5704)
2262 		dma_rw_ctl &= ~BGE_PCIDMARWCTL_MINDMA;
2263 
2264 	if (BGE_IS_57765_PLUS(sc)) {
2265 		dma_rw_ctl &= ~BGE_PCIDMARWCTL_DIS_CACHE_ALIGNMENT;
2266 		if (sc->bge_chipid == BGE_CHIPID_BCM57765_A0)
2267 			dma_rw_ctl &= ~BGE_PCIDMARWCTL_CRDRDR_RDMA_MRRS_MSK;
2268 
2269 		/*
2270 		 * Enable HW workaround for controllers that misinterpret
2271 		 * a status tag update and leave interrupts permanently
2272 		 * disabled.
2273 		 */
2274 		if (!BGE_IS_57765_FAMILY(sc) &&
2275 		    BGE_ASICREV(sc->bge_chipid) != BGE_ASICREV_BCM5717 &&
2276 		    BGE_ASICREV(sc->bge_chipid) != BGE_ASICREV_BCM5762)
2277 			dma_rw_ctl |= BGE_PCIDMARWCTL_TAGGED_STATUS_WA;
2278 	}
2279 
2280 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_DMA_RW_CTL,
2281 	    dma_rw_ctl);
2282 
2283 	/*
2284 	 * Set up general mode register.
2285 	 */
2286 	mode_ctl = BGE_DMA_SWAP_OPTIONS;
2287 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5720 ||
2288 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5762) {
2289 		/* Retain Host-2-BMC settings written by APE firmware. */
2290 		mode_ctl |= CSR_READ_4(sc, BGE_MODE_CTL) &
2291 		    (BGE_MODECTL_BYTESWAP_B2HRX_DATA |
2292 		    BGE_MODECTL_WORDSWAP_B2HRX_DATA |
2293 		    BGE_MODECTL_B2HRX_ENABLE | BGE_MODECTL_HTX2B_ENABLE);
2294 	}
2295 	mode_ctl |= BGE_MODECTL_MAC_ATTN_INTR | BGE_MODECTL_HOST_SEND_BDS |
2296 	    BGE_MODECTL_TX_NO_PHDR_CSUM;
2297 
2298 	/*
2299 	 * BCM5701 B5 have a bug causing data corruption when using
2300 	 * 64-bit DMA reads, which can be terminated early and then
2301 	 * completed later as 32-bit accesses, in combination with
2302 	 * certain bridges.
2303 	 */
2304 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5701 &&
2305 	    sc->bge_chipid == BGE_CHIPID_BCM5701_B5)
2306 		mode_ctl |= BGE_MODECTL_FORCE_PCI32;
2307 
2308 	/*
2309 	 * Tell the firmware the driver is running
2310 	 */
2311 	if (sc->bge_asf_mode & ASF_STACKUP)
2312 		mode_ctl |= BGE_MODECTL_STACKUP;
2313 
2314 	CSR_WRITE_4(sc, BGE_MODE_CTL, mode_ctl);
2315 
2316 	/*
2317 	 * Disable memory write invalidate.  Apparently it is not supported
2318 	 * properly by these devices.
2319 	 */
2320 	PCI_CLRBIT(sc->sc_pc, sc->sc_pcitag, PCI_COMMAND_STATUS_REG,
2321 		   PCI_COMMAND_INVALIDATE_ENABLE);
2322 
2323 #ifdef __brokenalpha__
2324 	/*
2325 	 * Must insure that we do not cross an 8K (bytes) boundary
2326 	 * for DMA reads.  Our highest limit is 1K bytes.  This is a
2327 	 * restriction on some ALPHA platforms with early revision
2328 	 * 21174 PCI chipsets, such as the AlphaPC 164lx
2329 	 */
2330 	PCI_SETBIT(sc, BGE_PCI_DMA_RW_CTL, BGE_PCI_READ_BNDRY_1024, 4);
2331 #endif
2332 
2333 	/* Set the timer prescaler (always 66MHz) */
2334 	CSR_WRITE_4_FLUSH(sc, BGE_MISC_CFG, BGE_32BITTIME_66MHZ);
2335 
2336 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906) {
2337 		DELAY(40);	/* XXX */
2338 
2339 		/* Put PHY into ready state */
2340 		BGE_CLRBIT_FLUSH(sc, BGE_MISC_CFG, BGE_MISCCFG_EPHY_IDDQ);
2341 		DELAY(40);
2342 	}
2343 
2344 	return 0;
2345 }
2346 
2347 static int
2348 bge_blockinit(struct bge_softc *sc)
2349 {
2350 	volatile struct bge_rcb	 *rcb;
2351 	bus_size_t rcb_addr;
2352 	struct ifnet * const ifp = &sc->ethercom.ec_if;
2353 	bge_hostaddr taddr;
2354 	uint32_t	dmactl, rdmareg, mimode, val;
2355 	int		i, limit;
2356 
2357 	/*
2358 	 * Initialize the memory window pointer register so that
2359 	 * we can access the first 32K of internal NIC RAM. This will
2360 	 * allow us to set up the TX send ring RCBs and the RX return
2361 	 * ring RCBs, plus other things which live in NIC memory.
2362 	 */
2363 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MEMWIN_BASEADDR, 0);
2364 
2365 	if (!BGE_IS_5705_PLUS(sc)) {
2366 		/* 57XX step 33 */
2367 		/* Configure mbuf memory pool */
2368 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_BASEADDR, BGE_BUFFPOOL_1);
2369 
2370 		if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5704)
2371 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_LEN, 0x10000);
2372 		else
2373 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_LEN, 0x18000);
2374 
2375 		/* 57XX step 34 */
2376 		/* Configure DMA resource pool */
2377 		CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_BASEADDR,
2378 		    BGE_DMA_DESCRIPTORS);
2379 		CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_LEN, 0x2000);
2380 	}
2381 
2382 	/* 5718 step 11, 57XX step 35 */
2383 	/*
2384 	 * Configure mbuf pool watermarks. New broadcom docs strongly
2385 	 * recommend these.
2386 	 */
2387 	if (BGE_IS_5717_PLUS(sc)) {
2388 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 0x0);
2389 		if (ifp->if_mtu > ETHERMTU) {
2390 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x7e);
2391 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0xea);
2392 		} else {
2393 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x2a);
2394 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0xa0);
2395 		}
2396 	} else if (BGE_IS_5705_PLUS(sc)) {
2397 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 0x0);
2398 
2399 		if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906) {
2400 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x04);
2401 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0x10);
2402 		} else {
2403 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x10);
2404 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0x60);
2405 		}
2406 	} else {
2407 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 0x50);
2408 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x20);
2409 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0x60);
2410 	}
2411 
2412 	/* 57XX step 36 */
2413 	/* Configure DMA resource watermarks */
2414 	CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_LOWAT, 5);
2415 	CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_HIWAT, 10);
2416 
2417 	/* 5718 step 13, 57XX step 38 */
2418 	/* Enable buffer manager */
2419 	val = BGE_BMANMODE_ENABLE | BGE_BMANMODE_ATTN;
2420 	/*
2421 	 * Change the arbitration algorithm of TXMBUF read request to
2422 	 * round-robin instead of priority based for BCM5719.  When
2423 	 * TXFIFO is almost empty, RDMA will hold its request until
2424 	 * TXFIFO is not almost empty.
2425 	 */
2426 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5719)
2427 		val |= BGE_BMANMODE_NO_TX_UNDERRUN;
2428 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5717 ||
2429 		sc->bge_chipid == BGE_CHIPID_BCM5719_A0 ||
2430 		sc->bge_chipid == BGE_CHIPID_BCM5720_A0)
2431 		val |= BGE_BMANMODE_LOMBUF_ATTN;
2432 	CSR_WRITE_4(sc, BGE_BMAN_MODE, val);
2433 
2434 	/* 57XX step 39 */
2435 	/* Poll for buffer manager start indication */
2436 	for (i = 0; i < BGE_TIMEOUT * 2; i++) {
2437 		DELAY(10);
2438 		if (CSR_READ_4(sc, BGE_BMAN_MODE) & BGE_BMANMODE_ENABLE)
2439 			break;
2440 	}
2441 
2442 	if (i == BGE_TIMEOUT * 2) {
2443 		aprint_error_dev(sc->bge_dev,
2444 		    "buffer manager failed to start\n");
2445 		return ENXIO;
2446 	}
2447 
2448 	/* 57XX step 40 */
2449 	/* Enable flow-through queues */
2450 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0xFFFFFFFF);
2451 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0);
2452 
2453 	/* Wait until queue initialization is complete */
2454 	for (i = 0; i < BGE_TIMEOUT * 2; i++) {
2455 		if (CSR_READ_4(sc, BGE_FTQ_RESET) == 0)
2456 			break;
2457 		DELAY(10);
2458 	}
2459 
2460 	if (i == BGE_TIMEOUT * 2) {
2461 		aprint_error_dev(sc->bge_dev,
2462 		    "flow-through queue init failed\n");
2463 		return ENXIO;
2464 	}
2465 
2466 	/*
2467 	 * Summary of rings supported by the controller:
2468 	 *
2469 	 * Standard Receive Producer Ring
2470 	 * - This ring is used to feed receive buffers for "standard"
2471 	 *   sized frames (typically 1536 bytes) to the controller.
2472 	 *
2473 	 * Jumbo Receive Producer Ring
2474 	 * - This ring is used to feed receive buffers for jumbo sized
2475 	 *   frames (i.e. anything bigger than the "standard" frames)
2476 	 *   to the controller.
2477 	 *
2478 	 * Mini Receive Producer Ring
2479 	 * - This ring is used to feed receive buffers for "mini"
2480 	 *   sized frames to the controller.
2481 	 * - This feature required external memory for the controller
2482 	 *   but was never used in a production system.  Should always
2483 	 *   be disabled.
2484 	 *
2485 	 * Receive Return Ring
2486 	 * - After the controller has placed an incoming frame into a
2487 	 *   receive buffer that buffer is moved into a receive return
2488 	 *   ring.  The driver is then responsible to passing the
2489 	 *   buffer up to the stack.  Many versions of the controller
2490 	 *   support multiple RR rings.
2491 	 *
2492 	 * Send Ring
2493 	 * - This ring is used for outgoing frames.  Many versions of
2494 	 *   the controller support multiple send rings.
2495 	 */
2496 
2497 	/* 5718 step 15, 57XX step 41 */
2498 	/* Initialize the standard RX ring control block */
2499 	rcb = &sc->bge_rdata->bge_info.bge_std_rx_rcb;
2500 	BGE_HOSTADDR(rcb->bge_hostaddr, BGE_RING_DMA_ADDR(sc, bge_rx_std_ring));
2501 	/* 5718 step 16 */
2502 	if (BGE_IS_57765_PLUS(sc)) {
2503 		/*
2504 		 * Bits 31-16: Programmable ring size (2048, 1024, 512, .., 32)
2505 		 * Bits 15-2 : Maximum RX frame size
2506 		 * Bit 1     : 1 = Ring Disabled, 0 = Ring Enabled
2507 		 * Bit 0     : Reserved
2508 		 */
2509 		rcb->bge_maxlen_flags =
2510 		    BGE_RCB_MAXLEN_FLAGS(512, BGE_MAX_FRAMELEN << 2);
2511 	} else if (BGE_IS_5705_PLUS(sc)) {
2512 		/*
2513 		 * Bits 31-16: Programmable ring size (512, 256, 128, 64, 32)
2514 		 * Bits 15-2 : Reserved (should be 0)
2515 		 * Bit 1     : 1 = Ring Disabled, 0 = Ring Enabled
2516 		 * Bit 0     : Reserved
2517 		 */
2518 		rcb->bge_maxlen_flags = BGE_RCB_MAXLEN_FLAGS(512, 0);
2519 	} else {
2520 		/*
2521 		 * Ring size is always XXX entries
2522 		 * Bits 31-16: Maximum RX frame size
2523 		 * Bits 15-2 : Reserved (should be 0)
2524 		 * Bit 1     : 1 = Ring Disabled, 0 = Ring Enabled
2525 		 * Bit 0     : Reserved
2526 		 */
2527 		rcb->bge_maxlen_flags =
2528 		    BGE_RCB_MAXLEN_FLAGS(BGE_MAX_FRAMELEN, 0);
2529 	}
2530 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5717 ||
2531 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5719 ||
2532 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5720)
2533 		rcb->bge_nicaddr = BGE_STD_RX_RINGS_5717;
2534 	else
2535 		rcb->bge_nicaddr = BGE_STD_RX_RINGS;
2536 	/* Write the standard receive producer ring control block. */
2537 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_HADDR_HI, rcb->bge_hostaddr.bge_addr_hi);
2538 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_HADDR_LO, rcb->bge_hostaddr.bge_addr_lo);
2539 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_MAXLEN_FLAGS, rcb->bge_maxlen_flags);
2540 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_NICADDR, rcb->bge_nicaddr);
2541 
2542 	/* Reset the standard receive producer ring producer index. */
2543 	bge_writembx(sc, BGE_MBX_RX_STD_PROD_LO, 0);
2544 
2545 	/* 57XX step 42 */
2546 	/*
2547 	 * Initialize the jumbo RX ring control block
2548 	 * We set the 'ring disabled' bit in the flags
2549 	 * field until we're actually ready to start
2550 	 * using this ring (i.e. once we set the MTU
2551 	 * high enough to require it).
2552 	 */
2553 	if (BGE_IS_JUMBO_CAPABLE(sc)) {
2554 		rcb = &sc->bge_rdata->bge_info.bge_jumbo_rx_rcb;
2555 		BGE_HOSTADDR(rcb->bge_hostaddr,
2556 		    BGE_RING_DMA_ADDR(sc, bge_rx_jumbo_ring));
2557 		rcb->bge_maxlen_flags = BGE_RCB_MAXLEN_FLAGS(0,
2558 		    BGE_RCB_FLAG_USE_EXT_RX_BD | BGE_RCB_FLAG_RING_DISABLED);
2559 		if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5717 ||
2560 		    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5719 ||
2561 		    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5720)
2562 			rcb->bge_nicaddr = BGE_JUMBO_RX_RINGS_5717;
2563 		else
2564 			rcb->bge_nicaddr = BGE_JUMBO_RX_RINGS;
2565 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_HADDR_HI,
2566 		    rcb->bge_hostaddr.bge_addr_hi);
2567 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_HADDR_LO,
2568 		    rcb->bge_hostaddr.bge_addr_lo);
2569 		/* Program the jumbo receive producer ring RCB parameters. */
2570 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_MAXLEN_FLAGS,
2571 		    rcb->bge_maxlen_flags);
2572 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_NICADDR, rcb->bge_nicaddr);
2573 		/* Reset the jumbo receive producer ring producer index. */
2574 		bge_writembx(sc, BGE_MBX_RX_JUMBO_PROD_LO, 0);
2575 	}
2576 
2577 	/* 57XX step 43 */
2578 	/* Disable the mini receive producer ring RCB. */
2579 	if (BGE_IS_5700_FAMILY(sc)) {
2580 		/* Set up dummy disabled mini ring RCB */
2581 		rcb = &sc->bge_rdata->bge_info.bge_mini_rx_rcb;
2582 		rcb->bge_maxlen_flags =
2583 		    BGE_RCB_MAXLEN_FLAGS(0, BGE_RCB_FLAG_RING_DISABLED);
2584 		CSR_WRITE_4(sc, BGE_RX_MINI_RCB_MAXLEN_FLAGS,
2585 		    rcb->bge_maxlen_flags);
2586 		/* Reset the mini receive producer ring producer index. */
2587 		bge_writembx(sc, BGE_MBX_RX_MINI_PROD_LO, 0);
2588 
2589 		bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
2590 		    offsetof(struct bge_ring_data, bge_info),
2591 		    sizeof(struct bge_gib),
2592 		    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
2593 	}
2594 
2595 	/* Choose de-pipeline mode for BCM5906 A0, A1 and A2. */
2596 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906) {
2597 		if (sc->bge_chipid == BGE_CHIPID_BCM5906_A0 ||
2598 		    sc->bge_chipid == BGE_CHIPID_BCM5906_A1 ||
2599 		    sc->bge_chipid == BGE_CHIPID_BCM5906_A2)
2600 			CSR_WRITE_4(sc, BGE_ISO_PKT_TX,
2601 			    (CSR_READ_4(sc, BGE_ISO_PKT_TX) & ~3) | 2);
2602 	}
2603 	/* 5718 step 14, 57XX step 44 */
2604 	/*
2605 	 * The BD ring replenish thresholds control how often the
2606 	 * hardware fetches new BD's from the producer rings in host
2607 	 * memory.  Setting the value too low on a busy system can
2608 	 * starve the hardware and reduce the throughput.
2609 	 *
2610 	 * Set the BD ring replenish thresholds. The recommended
2611 	 * values are 1/8th the number of descriptors allocated to
2612 	 * each ring, but since we try to avoid filling the entire
2613 	 * ring we set these to the minimal value of 8.  This needs to
2614 	 * be done on several of the supported chip revisions anyway,
2615 	 * to work around HW bugs.
2616 	 */
2617 	CSR_WRITE_4(sc, BGE_RBDI_STD_REPL_THRESH, 8);
2618 	if (BGE_IS_JUMBO_CAPABLE(sc))
2619 		CSR_WRITE_4(sc, BGE_RBDI_JUMBO_REPL_THRESH, 8);
2620 
2621 	/* 5718 step 18 */
2622 	if (BGE_IS_5717_PLUS(sc)) {
2623 		CSR_WRITE_4(sc, BGE_STD_REPL_LWM, 4);
2624 		CSR_WRITE_4(sc, BGE_JUMBO_REPL_LWM, 4);
2625 	}
2626 
2627 	/* 57XX step 45 */
2628 	/*
2629 	 * Disable all send rings by setting the 'ring disabled' bit
2630 	 * in the flags field of all the TX send ring control blocks,
2631 	 * located in NIC memory.
2632 	 */
2633 	if (BGE_IS_5700_FAMILY(sc)) {
2634 		/* 5700 to 5704 had 16 send rings. */
2635 		limit = BGE_TX_RINGS_EXTSSRAM_MAX;
2636 	} else if (BGE_IS_5717_PLUS(sc)) {
2637 		limit = BGE_TX_RINGS_5717_MAX;
2638 	} else if (BGE_IS_57765_FAMILY(sc) ||
2639 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5762) {
2640 		limit = BGE_TX_RINGS_57765_MAX;
2641 	} else
2642 		limit = 1;
2643 	rcb_addr = BGE_MEMWIN_START + BGE_SEND_RING_RCB;
2644 	for (i = 0; i < limit; i++) {
2645 		RCB_WRITE_4(sc, rcb_addr, bge_maxlen_flags,
2646 		    BGE_RCB_MAXLEN_FLAGS(0, BGE_RCB_FLAG_RING_DISABLED));
2647 		RCB_WRITE_4(sc, rcb_addr, bge_nicaddr, 0);
2648 		rcb_addr += sizeof(struct bge_rcb);
2649 	}
2650 
2651 	/* 57XX step 46 and 47 */
2652 	/* Configure send ring RCB 0 (we use only the first ring) */
2653 	rcb_addr = BGE_MEMWIN_START + BGE_SEND_RING_RCB;
2654 	BGE_HOSTADDR(taddr, BGE_RING_DMA_ADDR(sc, bge_tx_ring));
2655 	RCB_WRITE_4(sc, rcb_addr, bge_hostaddr.bge_addr_hi, taddr.bge_addr_hi);
2656 	RCB_WRITE_4(sc, rcb_addr, bge_hostaddr.bge_addr_lo, taddr.bge_addr_lo);
2657 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5717 ||
2658 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5719 ||
2659 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5720)
2660 		RCB_WRITE_4(sc, rcb_addr, bge_nicaddr, BGE_SEND_RING_5717);
2661 	else
2662 		RCB_WRITE_4(sc, rcb_addr, bge_nicaddr,
2663 		    BGE_NIC_TXRING_ADDR(0, BGE_TX_RING_CNT));
2664 	RCB_WRITE_4(sc, rcb_addr, bge_maxlen_flags,
2665 	    BGE_RCB_MAXLEN_FLAGS(BGE_TX_RING_CNT, 0));
2666 
2667 	/* 57XX step 48 */
2668 	/*
2669 	 * Disable all receive return rings by setting the
2670 	 * 'ring disabled' bit in the flags field of all the receive
2671 	 * return ring control blocks, located in NIC memory.
2672 	 */
2673 	if (BGE_IS_5717_PLUS(sc)) {
2674 		/* Should be 17, use 16 until we get an SRAM map. */
2675 		limit = 16;
2676 	} else if (BGE_IS_5700_FAMILY(sc))
2677 		limit = BGE_RX_RINGS_MAX;
2678 	else if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5755 ||
2679 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5762 ||
2680 	    BGE_IS_57765_FAMILY(sc))
2681 		limit = 4;
2682 	else
2683 		limit = 1;
2684 	/* Disable all receive return rings */
2685 	rcb_addr = BGE_MEMWIN_START + BGE_RX_RETURN_RING_RCB;
2686 	for (i = 0; i < limit; i++) {
2687 		RCB_WRITE_4(sc, rcb_addr, bge_hostaddr.bge_addr_hi, 0);
2688 		RCB_WRITE_4(sc, rcb_addr, bge_hostaddr.bge_addr_lo, 0);
2689 		RCB_WRITE_4(sc, rcb_addr, bge_maxlen_flags,
2690 		    BGE_RCB_MAXLEN_FLAGS(sc->bge_return_ring_cnt,
2691 			BGE_RCB_FLAG_RING_DISABLED));
2692 		RCB_WRITE_4(sc, rcb_addr, bge_nicaddr, 0);
2693 		bge_writembx(sc, BGE_MBX_RX_CONS0_LO +
2694 		    (i * (sizeof(uint64_t))), 0);
2695 		rcb_addr += sizeof(struct bge_rcb);
2696 	}
2697 
2698 	/* 57XX step 49 */
2699 	/*
2700 	 * Set up receive return ring 0.  Note that the NIC address
2701 	 * for RX return rings is 0x0.  The return rings live entirely
2702 	 * within the host, so the nicaddr field in the RCB isn't used.
2703 	 */
2704 	rcb_addr = BGE_MEMWIN_START + BGE_RX_RETURN_RING_RCB;
2705 	BGE_HOSTADDR(taddr, BGE_RING_DMA_ADDR(sc, bge_rx_return_ring));
2706 	RCB_WRITE_4(sc, rcb_addr, bge_hostaddr.bge_addr_hi, taddr.bge_addr_hi);
2707 	RCB_WRITE_4(sc, rcb_addr, bge_hostaddr.bge_addr_lo, taddr.bge_addr_lo);
2708 	RCB_WRITE_4(sc, rcb_addr, bge_nicaddr, 0x00000000);
2709 	RCB_WRITE_4(sc, rcb_addr, bge_maxlen_flags,
2710 	    BGE_RCB_MAXLEN_FLAGS(sc->bge_return_ring_cnt, 0));
2711 
2712 	/* 5718 step 24, 57XX step 53 */
2713 	/* Set random backoff seed for TX */
2714 	CSR_WRITE_4(sc, BGE_TX_RANDOM_BACKOFF,
2715 	    (CLLADDR(ifp->if_sadl)[0] + CLLADDR(ifp->if_sadl)[1] +
2716 		CLLADDR(ifp->if_sadl)[2] + CLLADDR(ifp->if_sadl)[3] +
2717 		CLLADDR(ifp->if_sadl)[4] + CLLADDR(ifp->if_sadl)[5]) &
2718 	    BGE_TX_BACKOFF_SEED_MASK);
2719 
2720 	/* 5718 step 26, 57XX step 55 */
2721 	/* Set inter-packet gap */
2722 	val = 0x2620;
2723 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5720 ||
2724 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5762)
2725 		val |= CSR_READ_4(sc, BGE_TX_LENGTHS) &
2726 		    (BGE_TXLEN_JMB_FRM_LEN_MSK | BGE_TXLEN_CNT_DN_VAL_MSK);
2727 	CSR_WRITE_4(sc, BGE_TX_LENGTHS, val);
2728 
2729 	/* 5718 step 27, 57XX step 56 */
2730 	/*
2731 	 * Specify which ring to use for packets that don't match
2732 	 * any RX rules.
2733 	 */
2734 	CSR_WRITE_4(sc, BGE_RX_RULES_CFG, 0x08);
2735 
2736 	/* 5718 step 28, 57XX step 57 */
2737 	/*
2738 	 * Configure number of RX lists. One interrupt distribution
2739 	 * list, sixteen active lists, one bad frames class.
2740 	 */
2741 	CSR_WRITE_4(sc, BGE_RXLP_CFG, 0x181);
2742 
2743 	/* 5718 step 29, 57XX step 58 */
2744 	/* Initialize RX list placement stats mask. */
2745 	if (BGE_IS_575X_PLUS(sc)) {
2746 		val = CSR_READ_4(sc, BGE_RXLP_STATS_ENABLE_MASK);
2747 		val &= ~BGE_RXLPSTATCONTROL_DACK_FIX;
2748 		CSR_WRITE_4(sc, BGE_RXLP_STATS_ENABLE_MASK, val);
2749 	} else
2750 		CSR_WRITE_4(sc, BGE_RXLP_STATS_ENABLE_MASK, 0x007FFFFF);
2751 
2752 	/* 5718 step 30, 57XX step 59 */
2753 	CSR_WRITE_4(sc, BGE_RXLP_STATS_CTL, 0x1);
2754 
2755 	/* 5718 step 33, 57XX step 62 */
2756 	/* Disable host coalescing until we get it set up */
2757 	CSR_WRITE_4(sc, BGE_HCC_MODE, 0x00000000);
2758 
2759 	/* 5718 step 34, 57XX step 63 */
2760 	/* Poll to make sure it's shut down. */
2761 	for (i = 0; i < BGE_TIMEOUT * 2; i++) {
2762 		DELAY(10);
2763 		if (!(CSR_READ_4(sc, BGE_HCC_MODE) & BGE_HCCMODE_ENABLE))
2764 			break;
2765 	}
2766 
2767 	if (i == BGE_TIMEOUT * 2) {
2768 		aprint_error_dev(sc->bge_dev,
2769 		    "host coalescing engine failed to idle\n");
2770 		return ENXIO;
2771 	}
2772 
2773 	/* 5718 step 35, 36, 37 */
2774 	/* Set up host coalescing defaults */
2775 	mutex_enter(sc->sc_intr_lock);
2776 	const uint32_t rx_coal_ticks = sc->bge_rx_coal_ticks;
2777 	const uint32_t tx_coal_ticks = sc->bge_tx_coal_ticks;
2778 	const uint32_t rx_max_coal_bds = sc->bge_rx_max_coal_bds;
2779 	const uint32_t tx_max_coal_bds = sc->bge_tx_max_coal_bds;
2780 	mutex_exit(sc->sc_intr_lock);
2781 	CSR_WRITE_4(sc, BGE_HCC_RX_COAL_TICKS, rx_coal_ticks);
2782 	CSR_WRITE_4(sc, BGE_HCC_TX_COAL_TICKS, tx_coal_ticks);
2783 	CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS, rx_max_coal_bds);
2784 	CSR_WRITE_4(sc, BGE_HCC_TX_MAX_COAL_BDS, tx_max_coal_bds);
2785 	if (!(BGE_IS_5705_PLUS(sc))) {
2786 		CSR_WRITE_4(sc, BGE_HCC_RX_COAL_TICKS_INT, 0);
2787 		CSR_WRITE_4(sc, BGE_HCC_TX_COAL_TICKS_INT, 0);
2788 	}
2789 	CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS_INT, 0);
2790 	CSR_WRITE_4(sc, BGE_HCC_TX_MAX_COAL_BDS_INT, 0);
2791 
2792 	/* Set up address of statistics block */
2793 	if (BGE_IS_5700_FAMILY(sc)) {
2794 		BGE_HOSTADDR(taddr, BGE_RING_DMA_ADDR(sc, bge_info.bge_stats));
2795 		CSR_WRITE_4(sc, BGE_HCC_STATS_TICKS, sc->bge_stat_ticks);
2796 		CSR_WRITE_4(sc, BGE_HCC_STATS_BASEADDR, BGE_STATS_BLOCK);
2797 		CSR_WRITE_4(sc, BGE_HCC_STATS_ADDR_HI, taddr.bge_addr_hi);
2798 		CSR_WRITE_4(sc, BGE_HCC_STATS_ADDR_LO, taddr.bge_addr_lo);
2799 	}
2800 
2801 	/* 5718 step 38 */
2802 	/* Set up address of status block */
2803 	BGE_HOSTADDR(taddr, BGE_RING_DMA_ADDR(sc, bge_status_block));
2804 	CSR_WRITE_4(sc, BGE_HCC_STATUSBLK_BASEADDR, BGE_STATUS_BLOCK);
2805 	CSR_WRITE_4(sc, BGE_HCC_STATUSBLK_ADDR_HI, taddr.bge_addr_hi);
2806 	CSR_WRITE_4(sc, BGE_HCC_STATUSBLK_ADDR_LO, taddr.bge_addr_lo);
2807 	sc->bge_rdata->bge_status_block.bge_idx[0].bge_rx_prod_idx = 0;
2808 	sc->bge_rdata->bge_status_block.bge_idx[0].bge_tx_cons_idx = 0;
2809 
2810 	/* Set up status block size. */
2811 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5700 &&
2812 	    sc->bge_chipid != BGE_CHIPID_BCM5700_C0) {
2813 		val = BGE_STATBLKSZ_FULL;
2814 		bzero(&sc->bge_rdata->bge_status_block, BGE_STATUS_BLK_SZ);
2815 	} else {
2816 		val = BGE_STATBLKSZ_32BYTE;
2817 		bzero(&sc->bge_rdata->bge_status_block, 32);
2818 	}
2819 
2820 	/* 5718 step 39, 57XX step 73 */
2821 	/* Turn on host coalescing state machine */
2822 	CSR_WRITE_4(sc, BGE_HCC_MODE, val | BGE_HCCMODE_ENABLE);
2823 
2824 	/* 5718 step 40, 57XX step 74 */
2825 	/* Turn on RX BD completion state machine and enable attentions */
2826 	CSR_WRITE_4(sc, BGE_RBDC_MODE,
2827 	    BGE_RBDCMODE_ENABLE | BGE_RBDCMODE_ATTN);
2828 
2829 	/* 5718 step 41, 57XX step 75 */
2830 	/* Turn on RX list placement state machine */
2831 	CSR_WRITE_4(sc, BGE_RXLP_MODE, BGE_RXLPMODE_ENABLE);
2832 
2833 	/* 57XX step 76 */
2834 	/* Turn on RX list selector state machine. */
2835 	if (!(BGE_IS_5705_PLUS(sc)))
2836 		CSR_WRITE_4(sc, BGE_RXLS_MODE, BGE_RXLSMODE_ENABLE);
2837 
2838 	val = BGE_MACMODE_TXDMA_ENB | BGE_MACMODE_RXDMA_ENB |
2839 	    BGE_MACMODE_RX_STATS_CLEAR | BGE_MACMODE_TX_STATS_CLEAR |
2840 	    BGE_MACMODE_RX_STATS_ENB | BGE_MACMODE_TX_STATS_ENB |
2841 	    BGE_MACMODE_FRMHDR_DMA_ENB;
2842 
2843 	if (sc->bge_flags & BGEF_FIBER_TBI)
2844 		val |= BGE_PORTMODE_TBI;
2845 	else if (sc->bge_flags & BGEF_FIBER_MII)
2846 		val |= BGE_PORTMODE_GMII;
2847 	else
2848 		val |= BGE_PORTMODE_MII;
2849 
2850 	/* 5718 step 42 and 43, 57XX step 77 and 78 */
2851 	/* Allow APE to send/receive frames. */
2852 	if ((sc->bge_mfw_flags & BGE_MFW_ON_APE) != 0)
2853 		val |= BGE_MACMODE_APE_RX_EN | BGE_MACMODE_APE_TX_EN;
2854 
2855 	/* Turn on DMA, clear stats */
2856 	CSR_WRITE_4_FLUSH(sc, BGE_MAC_MODE, val);
2857 	/* 5718 step 44 */
2858 	DELAY(40);
2859 
2860 	/* 5718 step 45, 57XX step 79 */
2861 	/* Set misc. local control, enable interrupts on attentions */
2862 	BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_INTR_ONATTN);
2863 	if (BGE_IS_5717_PLUS(sc)) {
2864 		CSR_READ_4(sc, BGE_MISC_LOCAL_CTL); /* Flush */
2865 		/* 5718 step 46 */
2866 		DELAY(100);
2867 	}
2868 
2869 	/* 57XX step 81 */
2870 	/* Turn on DMA completion state machine */
2871 	if (!(BGE_IS_5705_PLUS(sc)))
2872 		CSR_WRITE_4(sc, BGE_DMAC_MODE, BGE_DMACMODE_ENABLE);
2873 
2874 	/* 5718 step 47, 57XX step 82 */
2875 	val = BGE_WDMAMODE_ENABLE | BGE_WDMAMODE_ALL_ATTNS;
2876 
2877 	/* 5718 step 48 */
2878 	/* Enable host coalescing bug fix. */
2879 	if (BGE_IS_5755_PLUS(sc))
2880 		val |= BGE_WDMAMODE_STATUS_TAG_FIX;
2881 
2882 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5785)
2883 		val |= BGE_WDMAMODE_BURST_ALL_DATA;
2884 
2885 	/* Turn on write DMA state machine */
2886 	CSR_WRITE_4_FLUSH(sc, BGE_WDMA_MODE, val);
2887 	/* 5718 step 49 */
2888 	DELAY(40);
2889 
2890 	val = BGE_RDMAMODE_ENABLE | BGE_RDMAMODE_ALL_ATTNS;
2891 
2892 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5717)
2893 		val |= BGE_RDMAMODE_MULT_DMA_RD_DIS;
2894 
2895 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5784 ||
2896 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5785 ||
2897 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM57780)
2898 		val |= BGE_RDMAMODE_BD_SBD_CRPT_ATTN |
2899 		    BGE_RDMAMODE_MBUF_RBD_CRPT_ATTN |
2900 		    BGE_RDMAMODE_MBUF_SBD_CRPT_ATTN;
2901 
2902 	if (sc->bge_flags & BGEF_PCIE)
2903 		val |= BGE_RDMAMODE_FIFO_LONG_BURST;
2904 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM57766) {
2905 		if (ifp->if_mtu <= ETHERMTU)
2906 			val |= BGE_RDMAMODE_JMB_2K_MMRR;
2907 	}
2908 	if (sc->bge_flags & BGEF_TSO) {
2909 		val |= BGE_RDMAMODE_TSO4_ENABLE;
2910 		if (BGE_IS_5717_PLUS(sc))
2911 			val |= BGE_RDMAMODE_TSO6_ENABLE;
2912 	}
2913 
2914 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5720 ||
2915 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5762) {
2916 		val |= CSR_READ_4(sc, BGE_RDMA_MODE) &
2917 		    BGE_RDMAMODE_H2BNC_VLAN_DET;
2918 		/*
2919 		 * Allow multiple outstanding read requests from
2920 		 * non-LSO read DMA engine.
2921 		 */
2922 		val &= ~BGE_RDMAMODE_MULT_DMA_RD_DIS;
2923 	}
2924 
2925 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5761 ||
2926 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5784 ||
2927 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5785 ||
2928 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM57780 ||
2929 	    BGE_IS_57765_PLUS(sc)) {
2930 		if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5762)
2931 			rdmareg = BGE_RDMA_RSRVCTRL_REG2;
2932 		else
2933 			rdmareg = BGE_RDMA_RSRVCTRL;
2934 		dmactl = CSR_READ_4(sc, rdmareg);
2935 		/*
2936 		 * Adjust tx margin to prevent TX data corruption and
2937 		 * fix internal FIFO overflow.
2938 		 */
2939 		if (sc->bge_chipid == BGE_CHIPID_BCM5719_A0 ||
2940 		    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5762) {
2941 			dmactl &= ~(BGE_RDMA_RSRVCTRL_FIFO_LWM_MASK |
2942 			    BGE_RDMA_RSRVCTRL_FIFO_HWM_MASK |
2943 			    BGE_RDMA_RSRVCTRL_TXMRGN_MASK);
2944 			dmactl |= BGE_RDMA_RSRVCTRL_FIFO_LWM_1_5K |
2945 			    BGE_RDMA_RSRVCTRL_FIFO_HWM_1_5K |
2946 			    BGE_RDMA_RSRVCTRL_TXMRGN_320B;
2947 		}
2948 		/*
2949 		 * Enable fix for read DMA FIFO overruns.
2950 		 * The fix is to limit the number of RX BDs
2951 		 * the hardware would fetch at a time.
2952 		 */
2953 		CSR_WRITE_4(sc, rdmareg, dmactl |
2954 		    BGE_RDMA_RSRVCTRL_FIFO_OFLW_FIX);
2955 	}
2956 
2957 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5719) {
2958 		CSR_WRITE_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL,
2959 		    CSR_READ_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL) |
2960 		    BGE_RDMA_LSO_CRPTEN_CTRL_BLEN_BD_4K |
2961 		    BGE_RDMA_LSO_CRPTEN_CTRL_BLEN_LSO_4K);
2962 	} else if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5720) {
2963 		/*
2964 		 * Allow 4KB burst length reads for non-LSO frames.
2965 		 * Enable 512B burst length reads for buffer descriptors.
2966 		 */
2967 		CSR_WRITE_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL,
2968 		    CSR_READ_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL) |
2969 		    BGE_RDMA_LSO_CRPTEN_CTRL_BLEN_BD_512 |
2970 		    BGE_RDMA_LSO_CRPTEN_CTRL_BLEN_LSO_4K);
2971 	} else if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5762) {
2972 		CSR_WRITE_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL_REG2,
2973 		    CSR_READ_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL_REG2) |
2974 		    BGE_RDMA_LSO_CRPTEN_CTRL_BLEN_BD_4K |
2975 		    BGE_RDMA_LSO_CRPTEN_CTRL_BLEN_LSO_4K);
2976 	}
2977 	/* Turn on read DMA state machine */
2978 	CSR_WRITE_4_FLUSH(sc, BGE_RDMA_MODE, val);
2979 	/* 5718 step 52 */
2980 	delay(40);
2981 
2982 	if (sc->bge_flags & BGEF_RDMA_BUG) {
2983 		for (i = 0; i < BGE_NUM_RDMA_CHANNELS / 2; i++) {
2984 			val = CSR_READ_4(sc, BGE_RDMA_LENGTH + i * 4);
2985 			if ((val & 0xFFFF) > BGE_FRAMELEN)
2986 				break;
2987 			if (((val >> 16) & 0xFFFF) > BGE_FRAMELEN)
2988 				break;
2989 		}
2990 		if (i != BGE_NUM_RDMA_CHANNELS / 2) {
2991 			val = CSR_READ_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL);
2992 			if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5719)
2993 				val |= BGE_RDMA_TX_LENGTH_WA_5719;
2994 			else
2995 				val |= BGE_RDMA_TX_LENGTH_WA_5720;
2996 			CSR_WRITE_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL, val);
2997 		}
2998 	}
2999 
3000 	/* 5718 step 56, 57XX step 84 */
3001 	/* Turn on RX data completion state machine */
3002 	CSR_WRITE_4(sc, BGE_RDC_MODE, BGE_RDCMODE_ENABLE);
3003 
3004 	/* Turn on RX data and RX BD initiator state machine */
3005 	CSR_WRITE_4(sc, BGE_RDBDI_MODE, BGE_RDBDIMODE_ENABLE);
3006 
3007 	/* 57XX step 85 */
3008 	/* Turn on Mbuf cluster free state machine */
3009 	if (!BGE_IS_5705_PLUS(sc))
3010 		CSR_WRITE_4(sc, BGE_MBCF_MODE, BGE_MBCFMODE_ENABLE);
3011 
3012 	/* 5718 step 57, 57XX step 86 */
3013 	/* Turn on send data completion state machine */
3014 	val = BGE_SDCMODE_ENABLE;
3015 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5761)
3016 		val |= BGE_SDCMODE_CDELAY;
3017 	CSR_WRITE_4(sc, BGE_SDC_MODE, val);
3018 
3019 	/* 5718 step 58 */
3020 	/* Turn on send BD completion state machine */
3021 	CSR_WRITE_4(sc, BGE_SBDC_MODE, BGE_SBDCMODE_ENABLE);
3022 
3023 	/* 57XX step 88 */
3024 	/* Turn on RX BD initiator state machine */
3025 	CSR_WRITE_4(sc, BGE_RBDI_MODE, BGE_RBDIMODE_ENABLE);
3026 
3027 	/* 5718 step 60, 57XX step 90 */
3028 	/* Turn on send data initiator state machine */
3029 	if (sc->bge_flags & BGEF_TSO) {
3030 		/* XXX: magic value from Linux driver */
3031 		CSR_WRITE_4(sc, BGE_SDI_MODE, BGE_SDIMODE_ENABLE |
3032 		    BGE_SDIMODE_HW_LSO_PRE_DMA);
3033 	} else
3034 		CSR_WRITE_4(sc, BGE_SDI_MODE, BGE_SDIMODE_ENABLE);
3035 
3036 	/* 5718 step 61, 57XX step 91 */
3037 	/* Turn on send BD initiator state machine */
3038 	CSR_WRITE_4(sc, BGE_SBDI_MODE, BGE_SBDIMODE_ENABLE);
3039 
3040 	/* 5718 step 62, 57XX step 92 */
3041 	/* Turn on send BD selector state machine */
3042 	CSR_WRITE_4(sc, BGE_SRS_MODE, BGE_SRSMODE_ENABLE);
3043 
3044 	/* 5718 step 31, 57XX step 60 */
3045 	CSR_WRITE_4(sc, BGE_SDI_STATS_ENABLE_MASK, 0x007FFFFF);
3046 	/* 5718 step 32, 57XX step 61 */
3047 	CSR_WRITE_4(sc, BGE_SDI_STATS_CTL,
3048 	    BGE_SDISTATSCTL_ENABLE | BGE_SDISTATSCTL_FASTER);
3049 
3050 	/* ack/clear link change events */
3051 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED |
3052 	    BGE_MACSTAT_CFG_CHANGED | BGE_MACSTAT_MI_COMPLETE |
3053 	    BGE_MACSTAT_LINK_CHANGED);
3054 	CSR_WRITE_4(sc, BGE_MI_STS, 0);
3055 
3056 	/*
3057 	 * Enable attention when the link has changed state for
3058 	 * devices that use auto polling.
3059 	 */
3060 	if (sc->bge_flags & BGEF_FIBER_TBI) {
3061 		CSR_WRITE_4(sc, BGE_MI_STS, BGE_MISTS_LINK);
3062 	} else {
3063 		if ((sc->bge_flags & BGEF_CPMU_PRESENT) != 0)
3064 			mimode = BGE_MIMODE_500KHZ_CONST;
3065 		else
3066 			mimode = BGE_MIMODE_BASE;
3067 		/* 5718 step 68. 5718 step 69 (optionally). */
3068 		if (BGE_IS_5700_FAMILY(sc) ||
3069 		    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5705) {
3070 			mimode |= BGE_MIMODE_AUTOPOLL;
3071 			BGE_STS_SETBIT(sc, BGE_STS_AUTOPOLL);
3072 		}
3073 		mimode |= BGE_MIMODE_PHYADDR(sc->bge_phy_addr);
3074 		CSR_WRITE_4(sc, BGE_MI_MODE, mimode);
3075 		if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5700)
3076 			CSR_WRITE_4(sc, BGE_MAC_EVT_ENB,
3077 			    BGE_EVTENB_MI_INTERRUPT);
3078 	}
3079 
3080 	/*
3081 	 * Clear any pending link state attention.
3082 	 * Otherwise some link state change events may be lost until attention
3083 	 * is cleared by bge_intr() -> bge_link_upd() sequence.
3084 	 * It's not necessary on newer BCM chips - perhaps enabling link
3085 	 * state change attentions implies clearing pending attention.
3086 	 */
3087 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED |
3088 	    BGE_MACSTAT_CFG_CHANGED | BGE_MACSTAT_MI_COMPLETE |
3089 	    BGE_MACSTAT_LINK_CHANGED);
3090 
3091 	/* Enable link state change attentions. */
3092 	BGE_SETBIT(sc, BGE_MAC_EVT_ENB, BGE_EVTENB_LINK_CHANGED);
3093 
3094 	return 0;
3095 }
3096 
3097 static const struct bge_revision *
3098 bge_lookup_rev(uint32_t chipid)
3099 {
3100 	const struct bge_revision *br;
3101 
3102 	for (br = bge_revisions; br->br_name != NULL; br++) {
3103 		if (br->br_chipid == chipid)
3104 			return br;
3105 	}
3106 
3107 	for (br = bge_majorrevs; br->br_name != NULL; br++) {
3108 		if (br->br_chipid == BGE_ASICREV(chipid))
3109 			return br;
3110 	}
3111 
3112 	return NULL;
3113 }
3114 
3115 static const struct bge_product *
3116 bge_lookup(const struct pci_attach_args *pa)
3117 {
3118 	const struct bge_product *bp;
3119 
3120 	for (bp = bge_products; bp->bp_name != NULL; bp++) {
3121 		if (PCI_VENDOR(pa->pa_id) == bp->bp_vendor &&
3122 		    PCI_PRODUCT(pa->pa_id) == bp->bp_product)
3123 			return bp;
3124 	}
3125 
3126 	return NULL;
3127 }
3128 
3129 static uint32_t
3130 bge_chipid(const struct pci_attach_args *pa)
3131 {
3132 	uint32_t id;
3133 
3134 	id = pci_conf_read(pa->pa_pc, pa->pa_tag, BGE_PCI_MISC_CTL)
3135 		>> BGE_PCIMISCCTL_ASICREV_SHIFT;
3136 
3137 	if (BGE_ASICREV(id) == BGE_ASICREV_USE_PRODID_REG) {
3138 		switch (PCI_PRODUCT(pa->pa_id)) {
3139 		case PCI_PRODUCT_BROADCOM_BCM5717:
3140 		case PCI_PRODUCT_BROADCOM_BCM5718:
3141 		case PCI_PRODUCT_BROADCOM_BCM5719:
3142 		case PCI_PRODUCT_BROADCOM_BCM5720:
3143 		case PCI_PRODUCT_BROADCOM_BCM5725:
3144 		case PCI_PRODUCT_BROADCOM_BCM5727:
3145 		case PCI_PRODUCT_BROADCOM_BCM5762:
3146 		case PCI_PRODUCT_BROADCOM_BCM57764:
3147 		case PCI_PRODUCT_BROADCOM_BCM57767:
3148 		case PCI_PRODUCT_BROADCOM_BCM57787:
3149 			id = pci_conf_read(pa->pa_pc, pa->pa_tag,
3150 			    BGE_PCI_GEN2_PRODID_ASICREV);
3151 			break;
3152 		case PCI_PRODUCT_BROADCOM_BCM57761:
3153 		case PCI_PRODUCT_BROADCOM_BCM57762:
3154 		case PCI_PRODUCT_BROADCOM_BCM57765:
3155 		case PCI_PRODUCT_BROADCOM_BCM57766:
3156 		case PCI_PRODUCT_BROADCOM_BCM57781:
3157 		case PCI_PRODUCT_BROADCOM_BCM57782:
3158 		case PCI_PRODUCT_BROADCOM_BCM57785:
3159 		case PCI_PRODUCT_BROADCOM_BCM57786:
3160 		case PCI_PRODUCT_BROADCOM_BCM57791:
3161 		case PCI_PRODUCT_BROADCOM_BCM57795:
3162 			id = pci_conf_read(pa->pa_pc, pa->pa_tag,
3163 			    BGE_PCI_GEN15_PRODID_ASICREV);
3164 			break;
3165 		default:
3166 			id = pci_conf_read(pa->pa_pc, pa->pa_tag,
3167 			    BGE_PCI_PRODID_ASICREV);
3168 			break;
3169 		}
3170 	}
3171 
3172 	return id;
3173 }
3174 
3175 /*
3176  * Return true if MSI can be used with this device.
3177  */
3178 static int
3179 bge_can_use_msi(struct bge_softc *sc)
3180 {
3181 	int can_use_msi = 0;
3182 
3183 	switch (BGE_ASICREV(sc->bge_chipid)) {
3184 	case BGE_ASICREV_BCM5714_A0:
3185 	case BGE_ASICREV_BCM5714:
3186 		/*
3187 		 * Apparently, MSI doesn't work when these chips are
3188 		 * configured in single-port mode.
3189 		 */
3190 		break;
3191 	case BGE_ASICREV_BCM5750:
3192 		if (BGE_CHIPREV(sc->bge_chipid) != BGE_CHIPREV_5750_AX &&
3193 		    BGE_CHIPREV(sc->bge_chipid) != BGE_CHIPREV_5750_BX)
3194 			can_use_msi = 1;
3195 		break;
3196 	default:
3197 		if (BGE_IS_575X_PLUS(sc))
3198 			can_use_msi = 1;
3199 	}
3200 	return can_use_msi;
3201 }
3202 
3203 /*
3204  * Probe for a Broadcom chip. Check the PCI vendor and device IDs
3205  * against our list and return its name if we find a match. Note
3206  * that since the Broadcom controller contains VPD support, we
3207  * can get the device name string from the controller itself instead
3208  * of the compiled-in string. This is a little slow, but it guarantees
3209  * we'll always announce the right product name.
3210  */
3211 static int
3212 bge_probe(device_t parent, cfdata_t match, void *aux)
3213 {
3214 	struct pci_attach_args *pa = (struct pci_attach_args *)aux;
3215 
3216 	if (bge_lookup(pa) != NULL)
3217 		return 1;
3218 
3219 	return 0;
3220 }
3221 
3222 static void
3223 bge_attach(device_t parent, device_t self, void *aux)
3224 {
3225 	struct bge_softc * const sc = device_private(self);
3226 	struct pci_attach_args * const pa = aux;
3227 	prop_dictionary_t dict;
3228 	const struct bge_product *bp;
3229 	const struct bge_revision *br;
3230 	pci_chipset_tag_t	pc;
3231 	const char		*intrstr = NULL;
3232 	uint32_t		hwcfg, hwcfg2, hwcfg3, hwcfg4, hwcfg5;
3233 	uint32_t		command;
3234 	struct ifnet		*ifp;
3235 	struct mii_data * const mii = &sc->bge_mii;
3236 	uint32_t		misccfg, mimode, macmode;
3237 	void *			kva;
3238 	u_char			eaddr[ETHER_ADDR_LEN];
3239 	pcireg_t		memtype, subid, reg;
3240 	bus_addr_t		memaddr;
3241 	uint32_t		pm_ctl;
3242 	bool			no_seeprom;
3243 	int			capmask, trys;
3244 	int			mii_flags;
3245 	int			map_flags;
3246 	char intrbuf[PCI_INTRSTR_LEN];
3247 
3248 	bp = bge_lookup(pa);
3249 	KASSERT(bp != NULL);
3250 
3251 	sc->sc_pc = pa->pa_pc;
3252 	sc->sc_pcitag = pa->pa_tag;
3253 	sc->bge_dev = self;
3254 
3255 	sc->bge_pa = *pa;
3256 	pc = sc->sc_pc;
3257 	subid = pci_conf_read(pc, sc->sc_pcitag, PCI_SUBSYS_ID_REG);
3258 
3259 	aprint_naive(": Ethernet controller\n");
3260 	aprint_normal(": %s Ethernet\n", bp->bp_name);
3261 
3262 	/*
3263 	 * Map control/status registers.
3264 	 */
3265 	DPRINTFN(5, ("Map control/status regs\n"));
3266 	command = pci_conf_read(pc, sc->sc_pcitag, PCI_COMMAND_STATUS_REG);
3267 	command |= PCI_COMMAND_MEM_ENABLE | PCI_COMMAND_MASTER_ENABLE;
3268 	pci_conf_write(pc, sc->sc_pcitag, PCI_COMMAND_STATUS_REG, command);
3269 	command = pci_conf_read(pc, sc->sc_pcitag, PCI_COMMAND_STATUS_REG);
3270 
3271 	if (!(command & PCI_COMMAND_MEM_ENABLE)) {
3272 		aprint_error_dev(sc->bge_dev,
3273 		    "failed to enable memory mapping!\n");
3274 		return;
3275 	}
3276 
3277 	DPRINTFN(5, ("pci_mem_find\n"));
3278 	memtype = pci_mapreg_type(sc->sc_pc, sc->sc_pcitag, BGE_PCI_BAR0);
3279 	switch (memtype) {
3280 	case PCI_MAPREG_TYPE_MEM | PCI_MAPREG_MEM_TYPE_32BIT:
3281 	case PCI_MAPREG_TYPE_MEM | PCI_MAPREG_MEM_TYPE_64BIT:
3282 #if 0
3283 		if (pci_mapreg_map(pa, BGE_PCI_BAR0,
3284 		    memtype, 0, &sc->bge_btag, &sc->bge_bhandle,
3285 		    &memaddr, &sc->bge_bsize) == 0)
3286 			break;
3287 #else
3288 		/*
3289 		 * Workaround for PCI prefetchable bit. Some BCM5717-5720 based
3290 		 * system get NMI on boot (PR#48451). This problem might not be
3291 		 * the driver's bug but our PCI common part's bug. Until we
3292 		 * find a real reason, we ignore the prefetchable bit.
3293 		 */
3294 		if (pci_mapreg_info(pa->pa_pc, pa->pa_tag, BGE_PCI_BAR0,
3295 		    memtype, &memaddr, &sc->bge_bsize, &map_flags) == 0) {
3296 			map_flags &= ~BUS_SPACE_MAP_PREFETCHABLE;
3297 			if (bus_space_map(pa->pa_memt, memaddr, sc->bge_bsize,
3298 			    map_flags, &sc->bge_bhandle) == 0) {
3299 				sc->bge_btag = pa->pa_memt;
3300 				break;
3301 			}
3302 		}
3303 #endif
3304 		/* FALLTHROUGH */
3305 	default:
3306 		aprint_error_dev(sc->bge_dev, "can't find mem space\n");
3307 		return;
3308 	}
3309 
3310 	sc->bge_txrx_stopping = false;
3311 
3312 	/* Save various chip information. */
3313 	sc->bge_chipid = bge_chipid(pa);
3314 	sc->bge_phy_addr = bge_phy_addr(sc);
3315 
3316 	if (pci_get_capability(sc->sc_pc, sc->sc_pcitag, PCI_CAP_PCIEXPRESS,
3317 	    &sc->bge_pciecap, NULL) != 0) {
3318 		/* PCIe */
3319 		sc->bge_flags |= BGEF_PCIE;
3320 		/* Extract supported maximum payload size. */
3321 		reg = pci_conf_read(sc->sc_pc, sc->sc_pcitag,
3322 		    sc->bge_pciecap + PCIE_DCAP);
3323 		sc->bge_mps = 128 << (reg & PCIE_DCAP_MAX_PAYLOAD);
3324 		if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5719 ||
3325 		    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5720)
3326 			sc->bge_expmrq = 2048;
3327 		else
3328 			sc->bge_expmrq = 4096;
3329 		bge_set_max_readrq(sc);
3330 	} else if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5785) {
3331 		/* PCIe without PCIe cap */
3332 		sc->bge_flags |= BGEF_PCIE;
3333 	} else if ((pci_conf_read(sc->sc_pc, sc->sc_pcitag, BGE_PCI_PCISTATE) &
3334 		BGE_PCISTATE_PCI_BUSMODE) == 0) {
3335 		/* PCI-X */
3336 		sc->bge_flags |= BGEF_PCIX;
3337 		if (pci_get_capability(pa->pa_pc, pa->pa_tag, PCI_CAP_PCIX,
3338 			&sc->bge_pcixcap, NULL) == 0)
3339 			aprint_error_dev(sc->bge_dev,
3340 			    "unable to find PCIX capability\n");
3341 	}
3342 
3343 	if (BGE_CHIPREV(sc->bge_chipid) == BGE_CHIPREV_5700_BX) {
3344 		/*
3345 		 * Kludge for 5700 Bx bug: a hardware bug (PCIX byte enable?)
3346 		 * can clobber the chip's PCI config-space power control
3347 		 * registers, leaving the card in D3 powersave state. We do
3348 		 * not have memory-mapped registers in this state, so force
3349 		 * device into D0 state before starting initialization.
3350 		 */
3351 		pm_ctl = pci_conf_read(pc, sc->sc_pcitag, BGE_PCI_PWRMGMT_CMD);
3352 		pm_ctl &= ~(PCI_PWR_D0 | PCI_PWR_D1 | PCI_PWR_D2 | PCI_PWR_D3);
3353 		pm_ctl |= (1 << 8) | PCI_PWR_D0 ; /* D0 state */
3354 		pci_conf_write(pc, sc->sc_pcitag, BGE_PCI_PWRMGMT_CMD, pm_ctl);
3355 		DELAY(1000);	/* 27 usec is allegedly sufficient */
3356 	}
3357 
3358 	/* Save chipset family. */
3359 	switch (BGE_ASICREV(sc->bge_chipid)) {
3360 	case BGE_ASICREV_BCM5717:
3361 	case BGE_ASICREV_BCM5719:
3362 	case BGE_ASICREV_BCM5720:
3363 		sc->bge_flags |= BGEF_5717_PLUS;
3364 		/* FALLTHROUGH */
3365 	case BGE_ASICREV_BCM5762:
3366 	case BGE_ASICREV_BCM57765:
3367 	case BGE_ASICREV_BCM57766:
3368 		if (!BGE_IS_5717_PLUS(sc))
3369 			sc->bge_flags |= BGEF_57765_FAMILY;
3370 		sc->bge_flags |= BGEF_57765_PLUS | BGEF_5755_PLUS |
3371 		    BGEF_575X_PLUS | BGEF_5705_PLUS | BGEF_JUMBO_CAPABLE;
3372 		if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5719 ||
3373 		    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5720) {
3374 			/*
3375 			 * Enable work around for DMA engine miscalculation
3376 			 * of TXMBUF available space.
3377 			 */
3378 			sc->bge_flags |= BGEF_RDMA_BUG;
3379 
3380 			if ((BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5719) &&
3381 			    (sc->bge_chipid == BGE_CHIPID_BCM5719_A0)) {
3382 				/* Jumbo frame on BCM5719 A0 does not work. */
3383 				sc->bge_flags &= ~BGEF_JUMBO_CAPABLE;
3384 			}
3385 		}
3386 		break;
3387 	case BGE_ASICREV_BCM5755:
3388 	case BGE_ASICREV_BCM5761:
3389 	case BGE_ASICREV_BCM5784:
3390 	case BGE_ASICREV_BCM5785:
3391 	case BGE_ASICREV_BCM5787:
3392 	case BGE_ASICREV_BCM57780:
3393 		sc->bge_flags |= BGEF_5755_PLUS | BGEF_575X_PLUS | BGEF_5705_PLUS;
3394 		break;
3395 	case BGE_ASICREV_BCM5700:
3396 	case BGE_ASICREV_BCM5701:
3397 	case BGE_ASICREV_BCM5703:
3398 	case BGE_ASICREV_BCM5704:
3399 		sc->bge_flags |= BGEF_5700_FAMILY | BGEF_JUMBO_CAPABLE;
3400 		break;
3401 	case BGE_ASICREV_BCM5714_A0:
3402 	case BGE_ASICREV_BCM5780:
3403 	case BGE_ASICREV_BCM5714:
3404 		sc->bge_flags |= BGEF_5714_FAMILY | BGEF_JUMBO_CAPABLE;
3405 		/* FALLTHROUGH */
3406 	case BGE_ASICREV_BCM5750:
3407 	case BGE_ASICREV_BCM5752:
3408 	case BGE_ASICREV_BCM5906:
3409 		sc->bge_flags |= BGEF_575X_PLUS;
3410 		/* FALLTHROUGH */
3411 	case BGE_ASICREV_BCM5705:
3412 		sc->bge_flags |= BGEF_5705_PLUS;
3413 		break;
3414 	}
3415 
3416 	/* Identify chips with APE processor. */
3417 	switch (BGE_ASICREV(sc->bge_chipid)) {
3418 	case BGE_ASICREV_BCM5717:
3419 	case BGE_ASICREV_BCM5719:
3420 	case BGE_ASICREV_BCM5720:
3421 	case BGE_ASICREV_BCM5761:
3422 	case BGE_ASICREV_BCM5762:
3423 		sc->bge_flags |= BGEF_APE;
3424 		break;
3425 	}
3426 
3427 	/*
3428 	 * The 40bit DMA bug applies to the 5714/5715 controllers and is
3429 	 * not actually a MAC controller bug but an issue with the embedded
3430 	 * PCIe to PCI-X bridge in the device. Use 40bit DMA workaround.
3431 	 */
3432 	if (BGE_IS_5714_FAMILY(sc) && ((sc->bge_flags & BGEF_PCIX) != 0))
3433 		sc->bge_flags |= BGEF_40BIT_BUG;
3434 
3435 	/* Chips with APE need BAR2 access for APE registers/memory. */
3436 	if ((sc->bge_flags & BGEF_APE) != 0) {
3437 		memtype = pci_mapreg_type(pa->pa_pc, pa->pa_tag, BGE_PCI_BAR2);
3438 #if 0
3439 		if (pci_mapreg_map(pa, BGE_PCI_BAR2, memtype, 0,
3440 			&sc->bge_apetag, &sc->bge_apehandle, NULL,
3441 			&sc->bge_apesize)) {
3442 			aprint_error_dev(sc->bge_dev,
3443 			    "couldn't map BAR2 memory\n");
3444 			return;
3445 		}
3446 #else
3447 		/*
3448 		 * Workaround for PCI prefetchable bit. Some BCM5717-5720 based
3449 		 * system get NMI on boot (PR#48451). This problem might not be
3450 		 * the driver's bug but our PCI common part's bug. Until we
3451 		 * find a real reason, we ignore the prefetchable bit.
3452 		 */
3453 		if (pci_mapreg_info(pa->pa_pc, pa->pa_tag, BGE_PCI_BAR2,
3454 		    memtype, &memaddr, &sc->bge_apesize, &map_flags) != 0) {
3455 			aprint_error_dev(sc->bge_dev,
3456 			    "couldn't map BAR2 memory\n");
3457 			return;
3458 		}
3459 
3460 		map_flags &= ~BUS_SPACE_MAP_PREFETCHABLE;
3461 		if (bus_space_map(pa->pa_memt, memaddr,
3462 		    sc->bge_apesize, map_flags, &sc->bge_apehandle) != 0) {
3463 			aprint_error_dev(sc->bge_dev,
3464 			    "couldn't map BAR2 memory\n");
3465 			return;
3466 		}
3467 		sc->bge_apetag = pa->pa_memt;
3468 #endif
3469 
3470 		/* Enable APE register/memory access by host driver. */
3471 		reg = pci_conf_read(pa->pa_pc, pa->pa_tag, BGE_PCI_PCISTATE);
3472 		reg |= BGE_PCISTATE_ALLOW_APE_CTLSPC_WR |
3473 		    BGE_PCISTATE_ALLOW_APE_SHMEM_WR |
3474 		    BGE_PCISTATE_ALLOW_APE_PSPACE_WR;
3475 		pci_conf_write(pa->pa_pc, pa->pa_tag, BGE_PCI_PCISTATE, reg);
3476 
3477 		bge_ape_lock_init(sc);
3478 		bge_ape_read_fw_ver(sc);
3479 	}
3480 
3481 	/* Identify the chips that use an CPMU. */
3482 	if (BGE_IS_5717_PLUS(sc) ||
3483 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5784 ||
3484 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5761 ||
3485 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5785 ||
3486 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM57780)
3487 		sc->bge_flags |= BGEF_CPMU_PRESENT;
3488 
3489 	/*
3490 	 * When using the BCM5701 in PCI-X mode, data corruption has
3491 	 * been observed in the first few bytes of some received packets.
3492 	 * Aligning the packet buffer in memory eliminates the corruption.
3493 	 * Unfortunately, this misaligns the packet payloads.  On platforms
3494 	 * which do not support unaligned accesses, we will realign the
3495 	 * payloads by copying the received packets.
3496 	 */
3497 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5701 &&
3498 	    sc->bge_flags & BGEF_PCIX)
3499 		sc->bge_flags |= BGEF_RX_ALIGNBUG;
3500 
3501 	if (BGE_IS_5700_FAMILY(sc))
3502 		sc->bge_flags |= BGEF_JUMBO_CAPABLE;
3503 
3504 	misccfg = CSR_READ_4(sc, BGE_MISC_CFG);
3505 	misccfg &= BGE_MISCCFG_BOARD_ID_MASK;
3506 
3507 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5705 &&
3508 	    (misccfg == BGE_MISCCFG_BOARD_ID_5788 ||
3509 	     misccfg == BGE_MISCCFG_BOARD_ID_5788M))
3510 		sc->bge_flags |= BGEF_IS_5788;
3511 
3512 	/*
3513 	 * Some controllers seem to require a special firmware to use
3514 	 * TSO. But the firmware is not available to FreeBSD and Linux
3515 	 * claims that the TSO performed by the firmware is slower than
3516 	 * hardware based TSO. Moreover the firmware based TSO has one
3517 	 * known bug which can't handle TSO if ethernet header + IP/TCP
3518 	 * header is greater than 80 bytes. The workaround for the TSO
3519 	 * bug exist but it seems it's too expensive than not using
3520 	 * TSO at all. Some hardware also have the TSO bug so limit
3521 	 * the TSO to the controllers that are not affected TSO issues
3522 	 * (e.g. 5755 or higher).
3523 	 */
3524 	if (BGE_IS_5755_PLUS(sc)) {
3525 		/*
3526 		 * BCM5754 and BCM5787 shares the same ASIC id so
3527 		 * explicit device id check is required.
3528 		 */
3529 		if ((PCI_PRODUCT(pa->pa_id) != PCI_PRODUCT_BROADCOM_BCM5754) &&
3530 		    (PCI_PRODUCT(pa->pa_id) != PCI_PRODUCT_BROADCOM_BCM5754M))
3531 			sc->bge_flags |= BGEF_TSO;
3532 		/* TSO on BCM5719 A0 does not work. */
3533 		if ((BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5719) &&
3534 		    (sc->bge_chipid == BGE_CHIPID_BCM5719_A0))
3535 			sc->bge_flags &= ~BGEF_TSO;
3536 	}
3537 
3538 	capmask = 0xffffffff; /* XXX BMSR_DEFCAPMASK */
3539 	if ((BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5703 &&
3540 	     (misccfg == 0x4000 || misccfg == 0x8000)) ||
3541 	    (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5705 &&
3542 	     PCI_VENDOR(pa->pa_id) == PCI_VENDOR_BROADCOM &&
3543 	     (PCI_PRODUCT(pa->pa_id) == PCI_PRODUCT_BROADCOM_BCM5901 ||
3544 	      PCI_PRODUCT(pa->pa_id) == PCI_PRODUCT_BROADCOM_BCM5901A2 ||
3545 	      PCI_PRODUCT(pa->pa_id) == PCI_PRODUCT_BROADCOM_BCM5705F)) ||
3546 	    (PCI_VENDOR(pa->pa_id) == PCI_VENDOR_BROADCOM &&
3547 	     (PCI_PRODUCT(pa->pa_id) == PCI_PRODUCT_BROADCOM_BCM5751F ||
3548 	      PCI_PRODUCT(pa->pa_id) == PCI_PRODUCT_BROADCOM_BCM5753F ||
3549 	      PCI_PRODUCT(pa->pa_id) == PCI_PRODUCT_BROADCOM_BCM5787F)) ||
3550 	    PCI_PRODUCT(pa->pa_id) == PCI_PRODUCT_BROADCOM_BCM57790 ||
3551 	    PCI_PRODUCT(pa->pa_id) == PCI_PRODUCT_BROADCOM_BCM57791 ||
3552 	    PCI_PRODUCT(pa->pa_id) == PCI_PRODUCT_BROADCOM_BCM57795 ||
3553 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906) {
3554 		/* These chips are 10/100 only. */
3555 		capmask &= ~BMSR_EXTSTAT;
3556 		sc->bge_phy_flags |= BGEPHYF_NO_WIRESPEED;
3557 	}
3558 
3559 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5700 ||
3560 	    (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5705 &&
3561 	     (sc->bge_chipid != BGE_CHIPID_BCM5705_A0 &&
3562 		 sc->bge_chipid != BGE_CHIPID_BCM5705_A1)))
3563 		sc->bge_phy_flags |= BGEPHYF_NO_WIRESPEED;
3564 
3565 	/* Set various PHY bug flags. */
3566 	if (sc->bge_chipid == BGE_CHIPID_BCM5701_A0 ||
3567 	    sc->bge_chipid == BGE_CHIPID_BCM5701_B0)
3568 		sc->bge_phy_flags |= BGEPHYF_CRC_BUG;
3569 	if (BGE_CHIPREV(sc->bge_chipid) == BGE_CHIPREV_5703_AX ||
3570 	    BGE_CHIPREV(sc->bge_chipid) == BGE_CHIPREV_5704_AX)
3571 		sc->bge_phy_flags |= BGEPHYF_ADC_BUG;
3572 	if (sc->bge_chipid == BGE_CHIPID_BCM5704_A0)
3573 		sc->bge_phy_flags |= BGEPHYF_5704_A0_BUG;
3574 	if ((BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5700 ||
3575 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5701) &&
3576 	    PCI_VENDOR(subid) == PCI_VENDOR_DELL)
3577 		sc->bge_phy_flags |= BGEPHYF_NO_3LED;
3578 	if (BGE_IS_5705_PLUS(sc) &&
3579 	    BGE_ASICREV(sc->bge_chipid) != BGE_ASICREV_BCM5906 &&
3580 	    BGE_ASICREV(sc->bge_chipid) != BGE_ASICREV_BCM5785 &&
3581 	    BGE_ASICREV(sc->bge_chipid) != BGE_ASICREV_BCM57780 &&
3582 	    !BGE_IS_57765_PLUS(sc)) {
3583 		if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5755 ||
3584 		    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5761 ||
3585 		    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5784 ||
3586 		    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5787) {
3587 			if (PCI_PRODUCT(pa->pa_id) != PCI_PRODUCT_BROADCOM_BCM5722 &&
3588 			    PCI_PRODUCT(pa->pa_id) != PCI_PRODUCT_BROADCOM_BCM5756)
3589 				sc->bge_phy_flags |= BGEPHYF_JITTER_BUG;
3590 			if (PCI_PRODUCT(pa->pa_id) == PCI_PRODUCT_BROADCOM_BCM5755M)
3591 				sc->bge_phy_flags |= BGEPHYF_ADJUST_TRIM;
3592 		} else
3593 			sc->bge_phy_flags |= BGEPHYF_BER_BUG;
3594 	}
3595 
3596 	/*
3597 	 * SEEPROM check.
3598 	 * First check if firmware knows we do not have SEEPROM.
3599 	 */
3600 	if (prop_dictionary_get_bool(device_properties(self),
3601 	    "without-seeprom", &no_seeprom) && no_seeprom)
3602 		sc->bge_flags |= BGEF_NO_EEPROM;
3603 
3604 	else if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906)
3605 		sc->bge_flags |= BGEF_NO_EEPROM;
3606 
3607 	/* Now check the 'ROM failed' bit on the RX CPU */
3608 	else if (CSR_READ_4(sc, BGE_RXCPU_MODE) & BGE_RXCPUMODE_ROMFAIL)
3609 		sc->bge_flags |= BGEF_NO_EEPROM;
3610 
3611 	sc->bge_asf_mode = 0;
3612 	/* No ASF if APE present. */
3613 	if ((sc->bge_flags & BGEF_APE) == 0) {
3614 		if (bge_allow_asf && (bge_readmem_ind(sc, BGE_SRAM_DATA_SIG) ==
3615 			BGE_SRAM_DATA_SIG_MAGIC)) {
3616 			if (bge_readmem_ind(sc, BGE_SRAM_DATA_CFG) &
3617 			    BGE_HWCFG_ASF) {
3618 				sc->bge_asf_mode |= ASF_ENABLE;
3619 				sc->bge_asf_mode |= ASF_STACKUP;
3620 				if (BGE_IS_575X_PLUS(sc))
3621 					sc->bge_asf_mode |= ASF_NEW_HANDSHAKE;
3622 			}
3623 		}
3624 	}
3625 
3626 	int counts[PCI_INTR_TYPE_SIZE] = {
3627 		[PCI_INTR_TYPE_INTX] = 1,
3628 		[PCI_INTR_TYPE_MSI] = 1,
3629 		[PCI_INTR_TYPE_MSIX] = 1,
3630 	};
3631 	int max_type = PCI_INTR_TYPE_MSIX;
3632 
3633 	if (!bge_can_use_msi(sc)) {
3634 		/* MSI broken, allow only INTx */
3635 		max_type = PCI_INTR_TYPE_INTX;
3636 	}
3637 
3638 	if (pci_intr_alloc(pa, &sc->bge_pihp, counts, max_type) != 0) {
3639 		aprint_error_dev(sc->bge_dev, "couldn't alloc interrupt\n");
3640 		return;
3641 	}
3642 
3643 	DPRINTFN(5, ("pci_intr_string\n"));
3644 	intrstr = pci_intr_string(pc, sc->bge_pihp[0], intrbuf,
3645 	    sizeof(intrbuf));
3646 	DPRINTFN(5, ("pci_intr_establish\n"));
3647 	sc->bge_intrhand = pci_intr_establish_xname(pc, sc->bge_pihp[0],
3648 	    IPL_NET, bge_intr, sc, device_xname(sc->bge_dev));
3649 	if (sc->bge_intrhand == NULL) {
3650 		pci_intr_release(pc, sc->bge_pihp, 1);
3651 		sc->bge_pihp = NULL;
3652 
3653 		aprint_error_dev(self, "couldn't establish interrupt");
3654 		if (intrstr != NULL)
3655 			aprint_error(" at %s", intrstr);
3656 		aprint_error("\n");
3657 		return;
3658 	}
3659 	aprint_normal_dev(sc->bge_dev, "interrupting at %s\n", intrstr);
3660 
3661 	switch (pci_intr_type(pc, sc->bge_pihp[0])) {
3662 	case PCI_INTR_TYPE_MSIX:
3663 	case PCI_INTR_TYPE_MSI:
3664 		KASSERT(bge_can_use_msi(sc));
3665 		sc->bge_flags |= BGEF_MSI;
3666 		break;
3667 	default:
3668 		/* nothing to do */
3669 		break;
3670 	}
3671 
3672 	char wqname[MAXCOMLEN];
3673 	snprintf(wqname, sizeof(wqname), "%sReset", device_xname(sc->bge_dev));
3674 	int error = workqueue_create(&sc->sc_reset_wq, wqname,
3675 	    bge_handle_reset_work, sc, PRI_NONE, IPL_SOFTCLOCK,
3676 	    WQ_MPSAFE);
3677 	if (error) {
3678 		aprint_error_dev(sc->bge_dev,
3679 		    "unable to create reset workqueue\n");
3680 		return;
3681 	}
3682 
3683 
3684 	/*
3685 	 * All controllers except BCM5700 supports tagged status but
3686 	 * we use tagged status only for MSI case on BCM5717. Otherwise
3687 	 * MSI on BCM5717 does not work.
3688 	 */
3689 	if (BGE_IS_57765_PLUS(sc) && sc->bge_flags & BGEF_MSI)
3690 		sc->bge_flags |= BGEF_TAGGED_STATUS;
3691 
3692 	/*
3693 	 * Reset NVRAM before bge_reset(). It's required to acquire NVRAM
3694 	 * lock in bge_reset().
3695 	 */
3696 	CSR_WRITE_4_FLUSH(sc, BGE_EE_ADDR,
3697 	    BGE_EEADDR_RESET | BGE_EEHALFCLK(BGE_HALFCLK_384SCL));
3698 	delay(1000);
3699 	BGE_SETBIT_FLUSH(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_AUTO_EEPROM);
3700 
3701 	bge_stop_fw(sc);
3702 	bge_sig_pre_reset(sc, BGE_RESET_SHUTDOWN);
3703 	if (bge_reset(sc))
3704 		aprint_error_dev(sc->bge_dev, "chip reset failed\n");
3705 
3706 	/*
3707 	 * Read the hardware config word in the first 32k of NIC internal
3708 	 * memory, or fall back to the config word in the EEPROM.
3709 	 * Note: on some BCM5700 cards, this value appears to be unset.
3710 	 */
3711 	hwcfg = hwcfg2 = hwcfg3 = hwcfg4 = hwcfg5 = 0;
3712 	if (bge_readmem_ind(sc, BGE_SRAM_DATA_SIG) ==
3713 	    BGE_SRAM_DATA_SIG_MAGIC) {
3714 		uint32_t tmp;
3715 
3716 		hwcfg = bge_readmem_ind(sc, BGE_SRAM_DATA_CFG);
3717 		tmp = bge_readmem_ind(sc, BGE_SRAM_DATA_VER) >>
3718 		    BGE_SRAM_DATA_VER_SHIFT;
3719 		if ((0 < tmp) && (tmp < 0x100))
3720 			hwcfg2 = bge_readmem_ind(sc, BGE_SRAM_DATA_CFG_2);
3721 		if (sc->bge_flags & BGEF_PCIE)
3722 			hwcfg3 = bge_readmem_ind(sc, BGE_SRAM_DATA_CFG_3);
3723 		if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5785)
3724 			hwcfg4 = bge_readmem_ind(sc, BGE_SRAM_DATA_CFG_4);
3725 		if (BGE_IS_5717_PLUS(sc))
3726 			hwcfg5 = bge_readmem_ind(sc, BGE_SRAM_DATA_CFG_5);
3727 	} else if (!(sc->bge_flags & BGEF_NO_EEPROM)) {
3728 		bge_read_eeprom(sc, (void *)&hwcfg,
3729 		    BGE_EE_HWCFG_OFFSET, sizeof(hwcfg));
3730 		hwcfg = be32toh(hwcfg);
3731 	}
3732 	aprint_normal_dev(sc->bge_dev,
3733 	    "HW config %08x, %08x, %08x, %08x %08x\n",
3734 	    hwcfg, hwcfg2, hwcfg3, hwcfg4, hwcfg5);
3735 
3736 	bge_sig_legacy(sc, BGE_RESET_SHUTDOWN);
3737 	bge_sig_post_reset(sc, BGE_RESET_SHUTDOWN);
3738 
3739 	if (bge_chipinit(sc)) {
3740 		aprint_error_dev(sc->bge_dev, "chip initialization failed\n");
3741 		bge_release_resources(sc);
3742 		return;
3743 	}
3744 
3745 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5700) {
3746 		BGE_SETBIT_FLUSH(sc, BGE_MISC_LOCAL_CTL,
3747 		    BGE_MLC_MISCIO_OUT1 | BGE_MLC_MISCIO_OUTEN1);
3748 		DELAY(100);
3749 	}
3750 
3751 	/* Set MI_MODE */
3752 	mimode = BGE_MIMODE_PHYADDR(sc->bge_phy_addr);
3753 	if ((sc->bge_flags & BGEF_CPMU_PRESENT) != 0)
3754 		mimode |= BGE_MIMODE_500KHZ_CONST;
3755 	else
3756 		mimode |= BGE_MIMODE_BASE;
3757 	CSR_WRITE_4_FLUSH(sc, BGE_MI_MODE, mimode);
3758 	DELAY(80);
3759 
3760 	/*
3761 	 * Get station address from the EEPROM.
3762 	 */
3763 	if (bge_get_eaddr(sc, eaddr)) {
3764 		aprint_error_dev(sc->bge_dev,
3765 		    "failed to read station address\n");
3766 		bge_release_resources(sc);
3767 		return;
3768 	}
3769 
3770 	br = bge_lookup_rev(sc->bge_chipid);
3771 
3772 	if (br == NULL) {
3773 		aprint_normal_dev(sc->bge_dev, "unknown ASIC (0x%x)",
3774 		    sc->bge_chipid);
3775 	} else {
3776 		aprint_normal_dev(sc->bge_dev, "ASIC %s (0x%x)",
3777 		    br->br_name, sc->bge_chipid);
3778 	}
3779 	aprint_normal(", Ethernet address %s\n", ether_sprintf(eaddr));
3780 
3781 	/* Allocate the general information block and ring buffers. */
3782 	if (pci_dma64_available(pa)) {
3783 		sc->bge_dmatag = pa->pa_dmat64;
3784 		sc->bge_dmatag32 = pa->pa_dmat;
3785 		sc->bge_dma64 = true;
3786 	} else {
3787 		sc->bge_dmatag = pa->pa_dmat;
3788 		sc->bge_dmatag32 = pa->pa_dmat;
3789 		sc->bge_dma64 = false;
3790 	}
3791 
3792 	/* 40bit DMA workaround */
3793 	if (sizeof(bus_addr_t) > 4) {
3794 		if ((sc->bge_flags & BGEF_40BIT_BUG) != 0) {
3795 			bus_dma_tag_t olddmatag = sc->bge_dmatag; /* save */
3796 
3797 			if (bus_dmatag_subregion(olddmatag, 0,
3798 			    (bus_addr_t)__MASK(40),
3799 			    &(sc->bge_dmatag), BUS_DMA_WAITOK) != 0) {
3800 				aprint_error_dev(self,
3801 				    "WARNING: failed to restrict dma range,"
3802 				    " falling back to parent bus dma range\n");
3803 				sc->bge_dmatag = olddmatag;
3804 			}
3805 		}
3806 	}
3807 	SLIST_INIT(&sc->txdma_list);
3808 	DPRINTFN(5, ("bus_dmamem_alloc\n"));
3809 	if (bus_dmamem_alloc(sc->bge_dmatag, sizeof(struct bge_ring_data),
3810 			     PAGE_SIZE, 0, &sc->bge_ring_seg, 1,
3811 		&sc->bge_ring_rseg, BUS_DMA_WAITOK)) {
3812 		aprint_error_dev(sc->bge_dev, "can't alloc rx buffers\n");
3813 		return;
3814 	}
3815 	DPRINTFN(5, ("bus_dmamem_map\n"));
3816 	if (bus_dmamem_map(sc->bge_dmatag, &sc->bge_ring_seg,
3817 		sc->bge_ring_rseg, sizeof(struct bge_ring_data), &kva,
3818 			   BUS_DMA_WAITOK)) {
3819 		aprint_error_dev(sc->bge_dev,
3820 		    "can't map DMA buffers (%zu bytes)\n",
3821 		    sizeof(struct bge_ring_data));
3822 		bus_dmamem_free(sc->bge_dmatag, &sc->bge_ring_seg,
3823 		    sc->bge_ring_rseg);
3824 		return;
3825 	}
3826 	DPRINTFN(5, ("bus_dmamap_create\n"));
3827 	if (bus_dmamap_create(sc->bge_dmatag, sizeof(struct bge_ring_data), 1,
3828 	    sizeof(struct bge_ring_data), 0,
3829 	    BUS_DMA_WAITOK, &sc->bge_ring_map)) {
3830 		aprint_error_dev(sc->bge_dev, "can't create DMA map\n");
3831 		bus_dmamem_unmap(sc->bge_dmatag, kva,
3832 				 sizeof(struct bge_ring_data));
3833 		bus_dmamem_free(sc->bge_dmatag, &sc->bge_ring_seg,
3834 		    sc->bge_ring_rseg);
3835 		return;
3836 	}
3837 	DPRINTFN(5, ("bus_dmamap_load\n"));
3838 	if (bus_dmamap_load(sc->bge_dmatag, sc->bge_ring_map, kva,
3839 			    sizeof(struct bge_ring_data), NULL,
3840 			    BUS_DMA_WAITOK)) {
3841 		bus_dmamap_destroy(sc->bge_dmatag, sc->bge_ring_map);
3842 		bus_dmamem_unmap(sc->bge_dmatag, kva,
3843 				 sizeof(struct bge_ring_data));
3844 		bus_dmamem_free(sc->bge_dmatag, &sc->bge_ring_seg,
3845 		    sc->bge_ring_rseg);
3846 		return;
3847 	}
3848 
3849 	DPRINTFN(5, ("bzero\n"));
3850 	sc->bge_rdata = (struct bge_ring_data *)kva;
3851 
3852 	memset(sc->bge_rdata, 0, sizeof(struct bge_ring_data));
3853 
3854 	/* Try to allocate memory for jumbo buffers. */
3855 	if (BGE_IS_JUMBO_CAPABLE(sc)) {
3856 		if (bge_alloc_jumbo_mem(sc)) {
3857 			aprint_error_dev(sc->bge_dev,
3858 			    "jumbo buffer allocation failed\n");
3859 		} else
3860 			sc->ethercom.ec_capabilities |= ETHERCAP_JUMBO_MTU;
3861 	}
3862 
3863 	/* Set default tuneable values. */
3864 	sc->bge_stat_ticks = BGE_TICKS_PER_SEC;
3865 	sc->bge_rx_coal_ticks = 150;
3866 	sc->bge_rx_max_coal_bds = 64;
3867 	sc->bge_tx_coal_ticks = 300;
3868 	sc->bge_tx_max_coal_bds = 400;
3869 	if (BGE_IS_5705_PLUS(sc)) {
3870 		sc->bge_tx_coal_ticks = (12 * 5);
3871 		sc->bge_tx_max_coal_bds = (12 * 5);
3872 			aprint_verbose_dev(sc->bge_dev,
3873 			    "setting short Tx thresholds\n");
3874 	}
3875 
3876 	if (BGE_IS_5717_PLUS(sc))
3877 		sc->bge_return_ring_cnt = BGE_RETURN_RING_CNT;
3878 	else if (BGE_IS_5705_PLUS(sc))
3879 		sc->bge_return_ring_cnt = BGE_RETURN_RING_CNT_5705;
3880 	else
3881 		sc->bge_return_ring_cnt = BGE_RETURN_RING_CNT;
3882 
3883 	sc->sc_mcast_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_SOFTNET);
3884 	sc->sc_intr_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NET);
3885 
3886 	/* Set up ifnet structure */
3887 	ifp = &sc->ethercom.ec_if;
3888 	ifp->if_softc = sc;
3889 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
3890 	ifp->if_extflags = IFEF_MPSAFE;
3891 	ifp->if_ioctl = bge_ioctl;
3892 	ifp->if_stop = bge_stop;
3893 	ifp->if_start = bge_start;
3894 	ifp->if_init = bge_init;
3895 	IFQ_SET_MAXLEN(&ifp->if_snd, uimax(BGE_TX_RING_CNT - 1, IFQ_MAXLEN));
3896 	IFQ_SET_READY(&ifp->if_snd);
3897 	DPRINTFN(5, ("strcpy if_xname\n"));
3898 	strcpy(ifp->if_xname, device_xname(sc->bge_dev));
3899 
3900 	if (sc->bge_chipid != BGE_CHIPID_BCM5700_B0)
3901 		sc->ethercom.ec_if.if_capabilities |=
3902 		    IFCAP_CSUM_IPv4_Tx | IFCAP_CSUM_IPv4_Rx;
3903 #if 1	/* XXX TCP/UDP checksum offload breaks with pf(4) */
3904 		sc->ethercom.ec_if.if_capabilities |=
3905 		    IFCAP_CSUM_TCPv4_Tx | IFCAP_CSUM_TCPv4_Rx |
3906 		    IFCAP_CSUM_UDPv4_Tx | IFCAP_CSUM_UDPv4_Rx;
3907 #endif
3908 	sc->ethercom.ec_capabilities |=
3909 	    ETHERCAP_VLAN_HWTAGGING | ETHERCAP_VLAN_MTU;
3910 	sc->ethercom.ec_capenable |= ETHERCAP_VLAN_HWTAGGING;
3911 
3912 	if (sc->bge_flags & BGEF_TSO)
3913 		sc->ethercom.ec_if.if_capabilities |= IFCAP_TSOv4;
3914 
3915 	/*
3916 	 * Do MII setup.
3917 	 */
3918 	DPRINTFN(5, ("mii setup\n"));
3919 	mii->mii_ifp = ifp;
3920 	mii->mii_readreg = bge_miibus_readreg;
3921 	mii->mii_writereg = bge_miibus_writereg;
3922 	mii->mii_statchg = bge_miibus_statchg;
3923 
3924 	/*
3925 	 * Figure out what sort of media we have by checking the hardware
3926 	 * config word.  Note: on some BCM5700 cards, this value appears to be
3927 	 * unset. If that's the case, we have to rely on identifying the NIC
3928 	 * by its PCI subsystem ID, as we do below for the SysKonnect SK-9D41.
3929 	 * The SysKonnect SK-9D41 is a 1000baseSX card.
3930 	 */
3931 	if (PCI_PRODUCT(subid) == SK_SUBSYSID_9D41 ||
3932 	    (hwcfg & BGE_HWCFG_MEDIA) == BGE_MEDIA_FIBER) {
3933 		if (BGE_IS_5705_PLUS(sc)) {
3934 			sc->bge_flags |= BGEF_FIBER_MII;
3935 			sc->bge_phy_flags |= BGEPHYF_NO_WIRESPEED;
3936 		} else
3937 			sc->bge_flags |= BGEF_FIBER_TBI;
3938 	}
3939 
3940 	/* Set bge_phy_flags before prop_dictionary_set_uint32() */
3941 	if (BGE_IS_JUMBO_CAPABLE(sc))
3942 		sc->bge_phy_flags |= BGEPHYF_JUMBO_CAPABLE;
3943 
3944 	/* set phyflags and chipid before mii_attach() */
3945 	dict = device_properties(self);
3946 	prop_dictionary_set_uint32(dict, "phyflags", sc->bge_phy_flags);
3947 	prop_dictionary_set_uint32(dict, "chipid", sc->bge_chipid);
3948 
3949 	macmode = CSR_READ_4(sc, BGE_MAC_MODE);
3950 	macmode &= ~BGE_MACMODE_PORTMODE;
3951 	/* Initialize ifmedia structures. */
3952 	if (sc->bge_flags & BGEF_FIBER_TBI) {
3953 		CSR_WRITE_4_FLUSH(sc, BGE_MAC_MODE,
3954 		    macmode | BGE_PORTMODE_TBI);
3955 		DELAY(40);
3956 
3957 		struct ifmedia * const ifm = &sc->bge_ifmedia;
3958 		sc->ethercom.ec_ifmedia = ifm;
3959 
3960 		ifmedia_init_with_lock(ifm, IFM_IMASK,
3961 		    bge_ifmedia_upd, bge_ifmedia_sts, sc->sc_intr_lock);
3962 		ifmedia_add(ifm, IFM_ETHER | IFM_1000_SX, 0, NULL);
3963 		ifmedia_add(ifm, IFM_ETHER | IFM_1000_SX | IFM_FDX, 0, NULL);
3964 		ifmedia_add(ifm, IFM_ETHER | IFM_AUTO, 0, NULL);
3965 		ifmedia_set(ifm, IFM_ETHER | IFM_AUTO);
3966 		/* Pretend the user requested this setting */
3967 		sc->bge_ifmedia.ifm_media = sc->bge_ifmedia.ifm_cur->ifm_media;
3968 	} else {
3969 		uint16_t phyreg;
3970 		int rv;
3971 		/*
3972 		 * Do transceiver setup and tell the firmware the
3973 		 * driver is down so we can try to get access the
3974 		 * probe if ASF is running.  Retry a couple of times
3975 		 * if we get a conflict with the ASF firmware accessing
3976 		 * the PHY.
3977 		 */
3978 		if (sc->bge_flags & BGEF_FIBER_MII)
3979 			macmode |= BGE_PORTMODE_GMII;
3980 		else
3981 			macmode |= BGE_PORTMODE_MII;
3982 		CSR_WRITE_4_FLUSH(sc, BGE_MAC_MODE, macmode);
3983 		DELAY(40);
3984 
3985 		/*
3986 		 * Do transceiver setup and tell the firmware the
3987 		 * driver is down so we can try to get access the
3988 		 * probe if ASF is running.  Retry a couple of times
3989 		 * if we get a conflict with the ASF firmware accessing
3990 		 * the PHY.
3991 		 */
3992 		trys = 0;
3993 		BGE_CLRBIT(sc, BGE_MODE_CTL, BGE_MODECTL_STACKUP);
3994 		sc->ethercom.ec_mii = mii;
3995 		ifmedia_init_with_lock(&mii->mii_media, 0, bge_ifmedia_upd,
3996 		    bge_ifmedia_sts, sc->sc_intr_lock);
3997 		mii_flags = MIIF_DOPAUSE;
3998 		if (sc->bge_flags & BGEF_FIBER_MII)
3999 			mii_flags |= MIIF_HAVEFIBER;
4000 again:
4001 		bge_asf_driver_up(sc);
4002 		mutex_enter(sc->sc_intr_lock);
4003 		rv = bge_miibus_readreg(sc->bge_dev, sc->bge_phy_addr,
4004 		    MII_BMCR, &phyreg);
4005 		if ((rv != 0) || ((phyreg & BMCR_PDOWN) != 0)) {
4006 			int i;
4007 
4008 			bge_miibus_writereg(sc->bge_dev, sc->bge_phy_addr,
4009 			    MII_BMCR, BMCR_RESET);
4010 			/* Wait up to 500ms for it to complete. */
4011 			for (i = 0; i < 500; i++) {
4012 				bge_miibus_readreg(sc->bge_dev,
4013 				    sc->bge_phy_addr, MII_BMCR, &phyreg);
4014 				if ((phyreg & BMCR_RESET) == 0)
4015 					break;
4016 				DELAY(1000);
4017 			}
4018 		}
4019 		mutex_exit(sc->sc_intr_lock);
4020 
4021 		mii_attach(sc->bge_dev, mii, capmask, sc->bge_phy_addr,
4022 		    MII_OFFSET_ANY, mii_flags);
4023 
4024 		if (LIST_EMPTY(&mii->mii_phys) && (trys++ < 4))
4025 			goto again;
4026 
4027 		if (LIST_EMPTY(&mii->mii_phys)) {
4028 			aprint_error_dev(sc->bge_dev, "no PHY found!\n");
4029 			ifmedia_add(&mii->mii_media, IFM_ETHER | IFM_MANUAL,
4030 			    0, NULL);
4031 			ifmedia_set(&mii->mii_media, IFM_ETHER | IFM_MANUAL);
4032 		} else
4033 			ifmedia_set(&mii->mii_media, IFM_ETHER | IFM_AUTO);
4034 
4035 		/*
4036 		 * Now tell the firmware we are going up after probing the PHY
4037 		 */
4038 		if (sc->bge_asf_mode & ASF_STACKUP)
4039 			BGE_SETBIT(sc, BGE_MODE_CTL, BGE_MODECTL_STACKUP);
4040 	}
4041 
4042 	/*
4043 	 * Call MI attach routine.
4044 	 */
4045 	DPRINTFN(5, ("if_initialize\n"));
4046 	if_initialize(ifp);
4047 	ifp->if_percpuq = if_percpuq_create(ifp);
4048 	if_deferred_start_init(ifp, NULL);
4049 	if_register(ifp);
4050 
4051 	DPRINTFN(5, ("ether_ifattach\n"));
4052 	ether_ifattach(ifp, eaddr);
4053 	ether_set_ifflags_cb(&sc->ethercom, bge_ifflags_cb);
4054 
4055 	rnd_attach_source(&sc->rnd_source, device_xname(sc->bge_dev),
4056 		RND_TYPE_NET, RND_FLAG_DEFAULT);
4057 #ifdef BGE_EVENT_COUNTERS
4058 	/*
4059 	 * Attach event counters.
4060 	 */
4061 	evcnt_attach_dynamic(&sc->bge_ev_intr, EVCNT_TYPE_INTR,
4062 	    NULL, device_xname(sc->bge_dev), "intr");
4063 	evcnt_attach_dynamic(&sc->bge_ev_intr_spurious, EVCNT_TYPE_INTR,
4064 	    NULL, device_xname(sc->bge_dev), "intr_spurious");
4065 	evcnt_attach_dynamic(&sc->bge_ev_intr_spurious2, EVCNT_TYPE_INTR,
4066 	    NULL, device_xname(sc->bge_dev), "intr_spurious2");
4067 	evcnt_attach_dynamic(&sc->bge_ev_tx_xoff, EVCNT_TYPE_MISC,
4068 	    NULL, device_xname(sc->bge_dev), "tx_xoff");
4069 	evcnt_attach_dynamic(&sc->bge_ev_tx_xon, EVCNT_TYPE_MISC,
4070 	    NULL, device_xname(sc->bge_dev), "tx_xon");
4071 	evcnt_attach_dynamic(&sc->bge_ev_rx_xoff, EVCNT_TYPE_MISC,
4072 	    NULL, device_xname(sc->bge_dev), "rx_xoff");
4073 	evcnt_attach_dynamic(&sc->bge_ev_rx_xon, EVCNT_TYPE_MISC,
4074 	    NULL, device_xname(sc->bge_dev), "rx_xon");
4075 	evcnt_attach_dynamic(&sc->bge_ev_rx_macctl, EVCNT_TYPE_MISC,
4076 	    NULL, device_xname(sc->bge_dev), "rx_macctl");
4077 	evcnt_attach_dynamic(&sc->bge_ev_xoffentered, EVCNT_TYPE_MISC,
4078 	    NULL, device_xname(sc->bge_dev), "xoffentered");
4079 #endif /* BGE_EVENT_COUNTERS */
4080 	DPRINTFN(5, ("callout_init\n"));
4081 	callout_init(&sc->bge_timeout, CALLOUT_MPSAFE);
4082 	callout_setfunc(&sc->bge_timeout, bge_tick, sc);
4083 
4084 	if (pmf_device_register(self, NULL, NULL))
4085 		pmf_class_network_register(self, ifp);
4086 	else
4087 		aprint_error_dev(self, "couldn't establish power handler\n");
4088 
4089 	bge_sysctl_init(sc);
4090 
4091 #ifdef BGE_DEBUG
4092 	bge_debug_info(sc);
4093 #endif
4094 
4095 	sc->bge_attached = true;
4096 }
4097 
4098 /*
4099  * Stop all chip I/O so that the kernel's probe routines don't
4100  * get confused by errant DMAs when rebooting.
4101  */
4102 static int
4103 bge_detach(device_t self, int flags __unused)
4104 {
4105 	struct bge_softc * const sc = device_private(self);
4106 	struct ifnet * const ifp = &sc->ethercom.ec_if;
4107 
4108 	if (!sc->bge_attached)
4109 		return 0;
4110 
4111 	IFNET_LOCK(ifp);
4112 
4113 	/* Stop the interface. Callouts are stopped in it. */
4114 	bge_stop(ifp, 1);
4115 	sc->bge_detaching = true;
4116 
4117 	IFNET_UNLOCK(ifp);
4118 
4119 	mii_detach(&sc->bge_mii, MII_PHY_ANY, MII_OFFSET_ANY);
4120 
4121 	ether_ifdetach(ifp);
4122 	if_detach(ifp);
4123 
4124 	/* Delete all remaining media. */
4125 	ifmedia_fini(&sc->bge_mii.mii_media);
4126 
4127 	bge_release_resources(sc);
4128 
4129 	return 0;
4130 }
4131 
4132 static void
4133 bge_release_resources(struct bge_softc *sc)
4134 {
4135 
4136 	/* Detach sysctl */
4137 	if (sc->bge_log != NULL)
4138 		sysctl_teardown(&sc->bge_log);
4139 
4140 	callout_destroy(&sc->bge_timeout);
4141 
4142 #ifdef BGE_EVENT_COUNTERS
4143 	/* Detach event counters. */
4144 	evcnt_detach(&sc->bge_ev_intr);
4145 	evcnt_detach(&sc->bge_ev_intr_spurious);
4146 	evcnt_detach(&sc->bge_ev_intr_spurious2);
4147 	evcnt_detach(&sc->bge_ev_tx_xoff);
4148 	evcnt_detach(&sc->bge_ev_tx_xon);
4149 	evcnt_detach(&sc->bge_ev_rx_xoff);
4150 	evcnt_detach(&sc->bge_ev_rx_xon);
4151 	evcnt_detach(&sc->bge_ev_rx_macctl);
4152 	evcnt_detach(&sc->bge_ev_xoffentered);
4153 #endif /* BGE_EVENT_COUNTERS */
4154 
4155 	/* Disestablish the interrupt handler */
4156 	if (sc->bge_intrhand != NULL) {
4157 		pci_intr_disestablish(sc->sc_pc, sc->bge_intrhand);
4158 		pci_intr_release(sc->sc_pc, sc->bge_pihp, 1);
4159 		sc->bge_intrhand = NULL;
4160 	}
4161 
4162 	if (sc->bge_cdata.bge_jumbo_buf != NULL)
4163 		bge_free_jumbo_mem(sc);
4164 
4165 	if (sc->bge_dmatag != NULL) {
4166 		bus_dmamap_unload(sc->bge_dmatag, sc->bge_ring_map);
4167 		bus_dmamap_destroy(sc->bge_dmatag, sc->bge_ring_map);
4168 		bus_dmamem_unmap(sc->bge_dmatag, (void *)sc->bge_rdata,
4169 		    sizeof(struct bge_ring_data));
4170 		bus_dmamem_free(sc->bge_dmatag, &sc->bge_ring_seg,
4171 		    sc->bge_ring_rseg);
4172 	}
4173 
4174 	/* Unmap the device registers */
4175 	if (sc->bge_bsize != 0) {
4176 		bus_space_unmap(sc->bge_btag, sc->bge_bhandle, sc->bge_bsize);
4177 		sc->bge_bsize = 0;
4178 	}
4179 
4180 	/* Unmap the APE registers */
4181 	if (sc->bge_apesize != 0) {
4182 		bus_space_unmap(sc->bge_apetag, sc->bge_apehandle,
4183 		    sc->bge_apesize);
4184 		sc->bge_apesize = 0;
4185 	}
4186 	if (sc->sc_intr_lock) {
4187 		mutex_obj_free(sc->sc_intr_lock);
4188 		sc->sc_intr_lock = NULL;
4189 	}
4190 	if (sc->sc_mcast_lock) {
4191 		mutex_obj_free(sc->sc_mcast_lock);
4192 		sc->sc_mcast_lock = NULL;
4193 	}
4194 }
4195 
4196 static int
4197 bge_reset(struct bge_softc *sc)
4198 {
4199 	uint32_t cachesize, command;
4200 	uint32_t reset, mac_mode, mac_mode_mask;
4201 	pcireg_t devctl, reg;
4202 	int i, val;
4203 	void (*write_op)(struct bge_softc *, int, int);
4204 
4205 	/* Make mask for BGE_MAC_MODE register. */
4206 	mac_mode_mask = BGE_MACMODE_HALF_DUPLEX | BGE_MACMODE_PORTMODE;
4207 	if ((sc->bge_mfw_flags & BGE_MFW_ON_APE) != 0)
4208 		mac_mode_mask |= BGE_MACMODE_APE_RX_EN | BGE_MACMODE_APE_TX_EN;
4209 	/* Keep mac_mode_mask's bits of BGE_MAC_MODE register into mac_mode */
4210 	mac_mode = CSR_READ_4(sc, BGE_MAC_MODE) & mac_mode_mask;
4211 
4212 	if (BGE_IS_575X_PLUS(sc) && !BGE_IS_5714_FAMILY(sc) &&
4213 	    (BGE_ASICREV(sc->bge_chipid) != BGE_ASICREV_BCM5906)) {
4214 		if (sc->bge_flags & BGEF_PCIE)
4215 			write_op = bge_writemem_direct;
4216 		else
4217 			write_op = bge_writemem_ind;
4218 	} else
4219 		write_op = bge_writereg_ind;
4220 
4221 	/* 57XX step 4 */
4222 	/* Acquire the NVM lock */
4223 	if ((sc->bge_flags & BGEF_NO_EEPROM) == 0 &&
4224 	    BGE_ASICREV(sc->bge_chipid) != BGE_ASICREV_BCM5700 &&
4225 	    BGE_ASICREV(sc->bge_chipid) != BGE_ASICREV_BCM5701) {
4226 		CSR_WRITE_4(sc, BGE_NVRAM_SWARB, BGE_NVRAMSWARB_SET1);
4227 		for (i = 0; i < 8000; i++) {
4228 			if (CSR_READ_4(sc, BGE_NVRAM_SWARB) &
4229 			    BGE_NVRAMSWARB_GNT1)
4230 				break;
4231 			DELAY(20);
4232 		}
4233 		if (i == 8000) {
4234 			printf("%s: NVRAM lock timedout!\n",
4235 			    device_xname(sc->bge_dev));
4236 		}
4237 	}
4238 
4239 	/* Take APE lock when performing reset. */
4240 	bge_ape_lock(sc, BGE_APE_LOCK_GRC);
4241 
4242 	/* 57XX step 3 */
4243 	/* Save some important PCI state. */
4244 	cachesize = pci_conf_read(sc->sc_pc, sc->sc_pcitag, BGE_PCI_CACHESZ);
4245 	/* 5718 reset step 3 */
4246 	command = pci_conf_read(sc->sc_pc, sc->sc_pcitag, BGE_PCI_CMD);
4247 
4248 	/* 5718 reset step 5, 57XX step 5b-5d */
4249 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MISC_CTL,
4250 	    BGE_PCIMISCCTL_INDIRECT_ACCESS | BGE_PCIMISCCTL_MASK_PCI_INTR |
4251 	    BGE_HIF_SWAP_OPTIONS | BGE_PCIMISCCTL_PCISTATE_RW);
4252 
4253 	/* XXX ???: Disable fastboot on controllers that support it. */
4254 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5752 ||
4255 	    BGE_IS_5755_PLUS(sc))
4256 		CSR_WRITE_4(sc, BGE_FASTBOOT_PC, 0);
4257 
4258 	/* 5718 reset step 2, 57XX step 6 */
4259 	/*
4260 	 * Write the magic number to SRAM at offset 0xB50.
4261 	 * When firmware finishes its initialization it will
4262 	 * write ~BGE_MAGIC_NUMBER to the same location.
4263 	 */
4264 	bge_writemem_ind(sc, BGE_SRAM_FW_MB, BGE_SRAM_FW_MB_MAGIC);
4265 
4266 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM57780) {
4267 		val = CSR_READ_4(sc, BGE_PCIE_LINKCTL);
4268 		val = (val & ~BGE_PCIE_LINKCTL_L1_PLL_PDEN)
4269 		    | BGE_PCIE_LINKCTL_L1_PLL_PDDIS;
4270 		CSR_WRITE_4(sc, BGE_PCIE_LINKCTL, val);
4271 	}
4272 
4273 	/* 5718 reset step 6, 57XX step 7 */
4274 	reset = BGE_MISCCFG_RESET_CORE_CLOCKS | BGE_32BITTIME_66MHZ;
4275 	/*
4276 	 * XXX: from FreeBSD/Linux; no documentation
4277 	 */
4278 	if (sc->bge_flags & BGEF_PCIE) {
4279 		if ((BGE_ASICREV(sc->bge_chipid) != BGE_ASICREV_BCM5785) &&
4280 		    !BGE_IS_57765_PLUS(sc) &&
4281 		    (CSR_READ_4(sc, BGE_PHY_TEST_CTRL_REG) ==
4282 			(BGE_PHY_PCIE_LTASS_MODE | BGE_PHY_PCIE_SCRAM_MODE))) {
4283 			/* PCI Express 1.0 system */
4284 			CSR_WRITE_4(sc, BGE_PHY_TEST_CTRL_REG,
4285 			    BGE_PHY_PCIE_SCRAM_MODE);
4286 		}
4287 		if (sc->bge_chipid != BGE_CHIPID_BCM5750_A0) {
4288 			/*
4289 			 * Prevent PCI Express link training
4290 			 * during global reset.
4291 			 */
4292 			CSR_WRITE_4(sc, BGE_MISC_CFG, 1 << 29);
4293 			reset |= (1 << 29);
4294 		}
4295 	}
4296 
4297 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906) {
4298 		i = CSR_READ_4(sc, BGE_VCPU_STATUS);
4299 		CSR_WRITE_4(sc, BGE_VCPU_STATUS,
4300 		    i | BGE_VCPU_STATUS_DRV_RESET);
4301 		i = CSR_READ_4(sc, BGE_VCPU_EXT_CTRL);
4302 		CSR_WRITE_4(sc, BGE_VCPU_EXT_CTRL,
4303 		    i & ~BGE_VCPU_EXT_CTRL_HALT_CPU);
4304 	}
4305 
4306 	/*
4307 	 * Set GPHY Power Down Override to leave GPHY
4308 	 * powered up in D0 uninitialized.
4309 	 */
4310 	if (BGE_IS_5705_PLUS(sc) &&
4311 	    (sc->bge_flags & BGEF_CPMU_PRESENT) == 0)
4312 		reset |= BGE_MISCCFG_GPHY_PD_OVERRIDE;
4313 
4314 	/* Issue global reset */
4315 	write_op(sc, BGE_MISC_CFG, reset);
4316 
4317 	/* 5718 reset step 7, 57XX step 8 */
4318 	if (sc->bge_flags & BGEF_PCIE)
4319 		delay(100*1000); /* too big */
4320 	else
4321 		delay(1000);
4322 
4323 	if (sc->bge_flags & BGEF_PCIE) {
4324 		if (sc->bge_chipid == BGE_CHIPID_BCM5750_A0) {
4325 			DELAY(500000);
4326 			/* XXX: Magic Numbers */
4327 			reg = pci_conf_read(sc->sc_pc, sc->sc_pcitag,
4328 			    BGE_PCI_UNKNOWN0);
4329 			pci_conf_write(sc->sc_pc, sc->sc_pcitag,
4330 			    BGE_PCI_UNKNOWN0,
4331 			    reg | (1 << 15));
4332 		}
4333 		devctl = pci_conf_read(sc->sc_pc, sc->sc_pcitag,
4334 		    sc->bge_pciecap + PCIE_DCSR);
4335 		/* Clear enable no snoop and disable relaxed ordering. */
4336 		devctl &= ~(PCIE_DCSR_ENA_RELAX_ORD |
4337 		    PCIE_DCSR_ENA_NO_SNOOP);
4338 
4339 		/* Set PCIE max payload size to 128 for older PCIe devices */
4340 		if ((sc->bge_flags & BGEF_CPMU_PRESENT) == 0)
4341 			devctl &= ~(0x00e0);
4342 		/* Clear device status register. Write 1b to clear */
4343 		devctl |= PCIE_DCSR_URD | PCIE_DCSR_FED
4344 		    | PCIE_DCSR_NFED | PCIE_DCSR_CED;
4345 		pci_conf_write(sc->sc_pc, sc->sc_pcitag,
4346 		    sc->bge_pciecap + PCIE_DCSR, devctl);
4347 		bge_set_max_readrq(sc);
4348 	}
4349 
4350 	/* From Linux: dummy read to flush PCI posted writes */
4351 	reg = pci_conf_read(sc->sc_pc, sc->sc_pcitag, BGE_PCI_CMD);
4352 
4353 	/*
4354 	 * Reset some of the PCI state that got zapped by reset
4355 	 * To modify the PCISTATE register, BGE_PCIMISCCTL_PCISTATE_RW must be
4356 	 * set, too.
4357 	 */
4358 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MISC_CTL,
4359 	    BGE_PCIMISCCTL_INDIRECT_ACCESS | BGE_PCIMISCCTL_MASK_PCI_INTR |
4360 	    BGE_HIF_SWAP_OPTIONS | BGE_PCIMISCCTL_PCISTATE_RW);
4361 	val = BGE_PCISTATE_ROM_ENABLE | BGE_PCISTATE_ROM_RETRY_ENABLE;
4362 	if (sc->bge_chipid == BGE_CHIPID_BCM5704_A0 &&
4363 	    (sc->bge_flags & BGEF_PCIX) != 0)
4364 		val |= BGE_PCISTATE_RETRY_SAME_DMA;
4365 	if ((sc->bge_mfw_flags & BGE_MFW_ON_APE) != 0)
4366 		val |= BGE_PCISTATE_ALLOW_APE_CTLSPC_WR |
4367 		    BGE_PCISTATE_ALLOW_APE_SHMEM_WR |
4368 		    BGE_PCISTATE_ALLOW_APE_PSPACE_WR;
4369 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_PCISTATE, val);
4370 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_CACHESZ, cachesize);
4371 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_CMD, command);
4372 
4373 	/* 57xx step 11: disable PCI-X Relaxed Ordering. */
4374 	if (sc->bge_flags & BGEF_PCIX) {
4375 		reg = pci_conf_read(sc->sc_pc, sc->sc_pcitag, sc->bge_pcixcap
4376 		    + PCIX_CMD);
4377 		/* Set max memory read byte count to 2K */
4378 		if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5703) {
4379 			reg &= ~PCIX_CMD_BYTECNT_MASK;
4380 			reg |= PCIX_CMD_BCNT_2048;
4381 		} else if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5704){
4382 			/*
4383 			 * For 5704, set max outstanding split transaction
4384 			 * field to 0 (0 means it supports 1 request)
4385 			 */
4386 			reg &= ~(PCIX_CMD_SPLTRANS_MASK
4387 			    | PCIX_CMD_BYTECNT_MASK);
4388 			reg |= PCIX_CMD_BCNT_2048;
4389 		}
4390 		pci_conf_write(sc->sc_pc, sc->sc_pcitag, sc->bge_pcixcap
4391 		    + PCIX_CMD, reg & ~PCIX_CMD_RELAXED_ORDER);
4392 	}
4393 
4394 	/* 5718 reset step 10, 57XX step 12 */
4395 	/* Enable memory arbiter. */
4396 	if (BGE_IS_5714_FAMILY(sc)) {
4397 		val = CSR_READ_4(sc, BGE_MARB_MODE);
4398 		CSR_WRITE_4(sc, BGE_MARB_MODE, BGE_MARBMODE_ENABLE | val);
4399 	} else
4400 		CSR_WRITE_4(sc, BGE_MARB_MODE, BGE_MARBMODE_ENABLE);
4401 
4402 	/* XXX 5721, 5751 and 5752 */
4403 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5750) {
4404 		/* Step 19: */
4405 		BGE_SETBIT(sc, BGE_TLP_CONTROL_REG, 1 << 29 | 1 << 25);
4406 		/* Step 20: */
4407 		BGE_SETBIT(sc, BGE_TLP_CONTROL_REG, BGE_TLP_DATA_FIFO_PROTECT);
4408 	}
4409 
4410 	/* 5718 reset step 12, 57XX step 15 and 16 */
4411 	/* Fix up byte swapping */
4412 	CSR_WRITE_4(sc, BGE_MODE_CTL, BGE_DMA_SWAP_OPTIONS);
4413 
4414 	/* 5718 reset step 13, 57XX step 17 */
4415 	/* Poll until the firmware initialization is complete */
4416 	bge_poll_fw(sc);
4417 
4418 	/* 57XX step 21 */
4419 	if (BGE_CHIPREV(sc->bge_chipid) == BGE_CHIPREV_5704_BX) {
4420 		pcireg_t msidata;
4421 
4422 		msidata = pci_conf_read(sc->sc_pc, sc->sc_pcitag,
4423 		    BGE_PCI_MSI_DATA);
4424 		msidata |= ((1 << 13 | 1 << 12 | 1 << 10) << 16);
4425 		pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MSI_DATA,
4426 		    msidata);
4427 	}
4428 
4429 	/* 57XX step 18 */
4430 	/* Write mac mode. */
4431 	val = CSR_READ_4(sc, BGE_MAC_MODE);
4432 	/* Restore mac_mode_mask's bits using mac_mode */
4433 	val = (val & ~mac_mode_mask) | mac_mode;
4434 	CSR_WRITE_4_FLUSH(sc, BGE_MAC_MODE, val);
4435 	DELAY(40);
4436 
4437 	bge_ape_unlock(sc, BGE_APE_LOCK_GRC);
4438 
4439 	/*
4440 	 * The 5704 in TBI mode apparently needs some special
4441 	 * adjustment to insure the SERDES drive level is set
4442 	 * to 1.2V.
4443 	 */
4444 	if (sc->bge_flags & BGEF_FIBER_TBI &&
4445 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5704) {
4446 		uint32_t serdescfg;
4447 
4448 		serdescfg = CSR_READ_4(sc, BGE_SERDES_CFG);
4449 		serdescfg = (serdescfg & ~0xFFF) | 0x880;
4450 		CSR_WRITE_4(sc, BGE_SERDES_CFG, serdescfg);
4451 	}
4452 
4453 	if (sc->bge_flags & BGEF_PCIE &&
4454 	    !BGE_IS_57765_PLUS(sc) &&
4455 	    sc->bge_chipid != BGE_CHIPID_BCM5750_A0 &&
4456 	    BGE_ASICREV(sc->bge_chipid) != BGE_ASICREV_BCM5785) {
4457 		uint32_t v;
4458 
4459 		/* Enable PCI Express bug fix */
4460 		v = CSR_READ_4(sc, BGE_TLP_CONTROL_REG);
4461 		CSR_WRITE_4(sc, BGE_TLP_CONTROL_REG,
4462 		    v | BGE_TLP_DATA_FIFO_PROTECT);
4463 	}
4464 
4465 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5720)
4466 		BGE_CLRBIT(sc, BGE_CPMU_CLCK_ORIDE,
4467 		    CPMU_CLCK_ORIDE_MAC_ORIDE_EN);
4468 
4469 	return 0;
4470 }
4471 
4472 /*
4473  * Frame reception handling. This is called if there's a frame
4474  * on the receive return list.
4475  *
4476  * Note: we have to be able to handle two possibilities here:
4477  * 1) the frame is from the jumbo receive ring
4478  * 2) the frame is from the standard receive ring
4479  */
4480 
4481 static void
4482 bge_rxeof(struct bge_softc *sc)
4483 {
4484 	struct ifnet * const ifp = &sc->ethercom.ec_if;
4485 	uint16_t rx_prod, rx_cons;
4486 	int stdcnt = 0, jumbocnt = 0;
4487 	bus_dmamap_t dmamap;
4488 	bus_addr_t offset, toff;
4489 	bus_size_t tlen;
4490 	int tosync;
4491 
4492 	KASSERT(mutex_owned(sc->sc_intr_lock));
4493 
4494 	bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
4495 	    offsetof(struct bge_ring_data, bge_status_block),
4496 	    sizeof(struct bge_status_block),
4497 	    BUS_DMASYNC_POSTREAD);
4498 
4499 	rx_cons = sc->bge_rx_saved_considx;
4500 	rx_prod = sc->bge_rdata->bge_status_block.bge_idx[0].bge_rx_prod_idx;
4501 
4502 	/* Nothing to do */
4503 	if (rx_cons == rx_prod)
4504 		return;
4505 
4506 	offset = offsetof(struct bge_ring_data, bge_rx_return_ring);
4507 	tosync = rx_prod - rx_cons;
4508 
4509 	if (tosync != 0)
4510 		rnd_add_uint32(&sc->rnd_source, tosync);
4511 
4512 	toff = offset + (rx_cons * sizeof(struct bge_rx_bd));
4513 
4514 	if (tosync < 0) {
4515 		tlen = (sc->bge_return_ring_cnt - rx_cons) *
4516 		    sizeof(struct bge_rx_bd);
4517 		bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
4518 		    toff, tlen, BUS_DMASYNC_POSTREAD);
4519 		tosync = rx_prod;
4520 		toff = offset;
4521 	}
4522 
4523 	if (tosync != 0) {
4524 		bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
4525 		    toff, tosync * sizeof(struct bge_rx_bd),
4526 		    BUS_DMASYNC_POSTREAD);
4527 	}
4528 
4529 	while (rx_cons != rx_prod) {
4530 		struct bge_rx_bd	*cur_rx;
4531 		uint32_t		rxidx;
4532 		struct mbuf		*m = NULL;
4533 
4534 		cur_rx = &sc->bge_rdata->bge_rx_return_ring[rx_cons];
4535 
4536 		rxidx = cur_rx->bge_idx;
4537 		BGE_INC(rx_cons, sc->bge_return_ring_cnt);
4538 
4539 		if (cur_rx->bge_flags & BGE_RXBDFLAG_JUMBO_RING) {
4540 			BGE_INC(sc->bge_jumbo, BGE_JUMBO_RX_RING_CNT);
4541 			m = sc->bge_cdata.bge_rx_jumbo_chain[rxidx];
4542 			sc->bge_cdata.bge_rx_jumbo_chain[rxidx] = NULL;
4543 			jumbocnt++;
4544 			bus_dmamap_sync(sc->bge_dmatag,
4545 			    sc->bge_cdata.bge_rx_jumbo_map,
4546 			    mtod(m, char *) - (char *)sc->bge_cdata.bge_jumbo_buf,
4547 			    BGE_JLEN, BUS_DMASYNC_POSTREAD);
4548 			if (cur_rx->bge_flags & BGE_RXBDFLAG_ERROR) {
4549 				if_statinc(ifp, if_ierrors);
4550 				bge_newbuf_jumbo(sc, sc->bge_jumbo, m);
4551 				continue;
4552 			}
4553 			if (bge_newbuf_jumbo(sc, sc->bge_jumbo,
4554 					     NULL) == ENOBUFS) {
4555 				if_statinc(ifp, if_ierrors);
4556 				bge_newbuf_jumbo(sc, sc->bge_jumbo, m);
4557 				continue;
4558 			}
4559 		} else {
4560 			m = sc->bge_cdata.bge_rx_std_chain[rxidx];
4561 			sc->bge_cdata.bge_rx_std_chain[rxidx] = NULL;
4562 
4563 			stdcnt++;
4564 			sc->bge_std_cnt--;
4565 
4566 			dmamap = sc->bge_cdata.bge_rx_std_map[rxidx];
4567 			bus_dmamap_sync(sc->bge_dmatag, dmamap, 0,
4568 			    dmamap->dm_mapsize, BUS_DMASYNC_POSTREAD);
4569 			bus_dmamap_unload(sc->bge_dmatag, dmamap);
4570 
4571 			if (cur_rx->bge_flags & BGE_RXBDFLAG_ERROR) {
4572 				m_free(m);
4573 				if_statinc(ifp, if_ierrors);
4574 				continue;
4575 			}
4576 		}
4577 
4578 #ifndef __NO_STRICT_ALIGNMENT
4579 		/*
4580 		 * XXX: if the 5701 PCIX-Rx-DMA workaround is in effect,
4581 		 * the Rx buffer has the layer-2 header unaligned.
4582 		 * If our CPU requires alignment, re-align by copying.
4583 		 */
4584 		if (sc->bge_flags & BGEF_RX_ALIGNBUG) {
4585 			memmove(mtod(m, char *) + ETHER_ALIGN, m->m_data,
4586 				cur_rx->bge_len);
4587 			m->m_data += ETHER_ALIGN;
4588 		}
4589 #endif
4590 
4591 		m->m_pkthdr.len = m->m_len = cur_rx->bge_len - ETHER_CRC_LEN;
4592 		m_set_rcvif(m, ifp);
4593 
4594 		bge_rxcsum(sc, cur_rx, m);
4595 
4596 		/*
4597 		 * If we received a packet with a vlan tag, pass it
4598 		 * to vlan_input() instead of ether_input().
4599 		 */
4600 		if (cur_rx->bge_flags & BGE_RXBDFLAG_VLAN_TAG)
4601 			vlan_set_tag(m, cur_rx->bge_vlan_tag);
4602 
4603 		if_percpuq_enqueue(ifp->if_percpuq, m);
4604 	}
4605 
4606 	sc->bge_rx_saved_considx = rx_cons;
4607 	bge_writembx(sc, BGE_MBX_RX_CONS0_LO, sc->bge_rx_saved_considx);
4608 	if (stdcnt)
4609 		bge_fill_rx_ring_std(sc);
4610 	if (jumbocnt)
4611 		bge_writembx(sc, BGE_MBX_RX_JUMBO_PROD_LO, sc->bge_jumbo);
4612 }
4613 
4614 static void
4615 bge_rxcsum(struct bge_softc *sc, struct bge_rx_bd *cur_rx, struct mbuf *m)
4616 {
4617 
4618 	if (BGE_IS_57765_PLUS(sc)) {
4619 		if ((cur_rx->bge_flags & BGE_RXBDFLAG_IPV6) == 0) {
4620 			if ((cur_rx->bge_flags & BGE_RXBDFLAG_IP_CSUM) != 0)
4621 				m->m_pkthdr.csum_flags = M_CSUM_IPv4;
4622 			if ((cur_rx->bge_error_flag &
4623 				BGE_RXERRFLAG_IP_CSUM_NOK) != 0)
4624 				m->m_pkthdr.csum_flags |= M_CSUM_IPv4_BAD;
4625 			if (cur_rx->bge_flags & BGE_RXBDFLAG_TCP_UDP_CSUM) {
4626 				m->m_pkthdr.csum_data =
4627 				    cur_rx->bge_tcp_udp_csum;
4628 				m->m_pkthdr.csum_flags |=
4629 				    (M_CSUM_TCPv4 | M_CSUM_UDPv4 |M_CSUM_DATA);
4630 			}
4631 		}
4632 	} else {
4633 		if ((cur_rx->bge_flags & BGE_RXBDFLAG_IP_CSUM) != 0)
4634 			m->m_pkthdr.csum_flags = M_CSUM_IPv4;
4635 		if ((cur_rx->bge_ip_csum ^ 0xffff) != 0)
4636 			m->m_pkthdr.csum_flags |= M_CSUM_IPv4_BAD;
4637 		/*
4638 		 * Rx transport checksum-offload may also
4639 		 * have bugs with packets which, when transmitted,
4640 		 * were `runts' requiring padding.
4641 		 */
4642 		if (cur_rx->bge_flags & BGE_RXBDFLAG_TCP_UDP_CSUM &&
4643 		    (/* (sc->_bge_quirks & BGE_QUIRK_SHORT_CKSUM_BUG) == 0 ||*/
4644 			    m->m_pkthdr.len >= ETHER_MIN_NOPAD)) {
4645 			m->m_pkthdr.csum_data =
4646 			    cur_rx->bge_tcp_udp_csum;
4647 			m->m_pkthdr.csum_flags |=
4648 			    (M_CSUM_TCPv4 | M_CSUM_UDPv4 | M_CSUM_DATA);
4649 		}
4650 	}
4651 }
4652 
4653 static void
4654 bge_txeof(struct bge_softc *sc)
4655 {
4656 	struct ifnet * const ifp = &sc->ethercom.ec_if;
4657 	struct bge_tx_bd *cur_tx = NULL;
4658 	struct txdmamap_pool_entry *dma;
4659 	bus_addr_t offset, toff;
4660 	bus_size_t tlen;
4661 	int tosync;
4662 	struct mbuf *m;
4663 
4664 	KASSERT(mutex_owned(sc->sc_intr_lock));
4665 
4666 	bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
4667 	    offsetof(struct bge_ring_data, bge_status_block),
4668 	    sizeof(struct bge_status_block),
4669 	    BUS_DMASYNC_POSTREAD);
4670 
4671 	const uint16_t hw_cons_idx =
4672 	    sc->bge_rdata->bge_status_block.bge_idx[0].bge_tx_cons_idx;
4673 	offset = offsetof(struct bge_ring_data, bge_tx_ring);
4674 	tosync = hw_cons_idx - sc->bge_tx_saved_considx;
4675 
4676 	if (tosync != 0)
4677 		rnd_add_uint32(&sc->rnd_source, tosync);
4678 
4679 	toff = offset + (sc->bge_tx_saved_considx * sizeof(struct bge_tx_bd));
4680 
4681 	if (tosync < 0) {
4682 		tlen = (BGE_TX_RING_CNT - sc->bge_tx_saved_considx) *
4683 		    sizeof(struct bge_tx_bd);
4684 		bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
4685 		    toff, tlen, BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
4686 		tosync = hw_cons_idx;
4687 		toff = offset;
4688 	}
4689 
4690 	if (tosync != 0) {
4691 		bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
4692 		    toff, tosync * sizeof(struct bge_tx_bd),
4693 		    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
4694 	}
4695 
4696 	/*
4697 	 * Go through our tx ring and free mbufs for those
4698 	 * frames that have been sent.
4699 	 */
4700 	while (sc->bge_tx_saved_considx != hw_cons_idx) {
4701 		uint32_t idx = sc->bge_tx_saved_considx;
4702 		cur_tx = &sc->bge_rdata->bge_tx_ring[idx];
4703 		if (cur_tx->bge_flags & BGE_TXBDFLAG_END)
4704 			if_statinc(ifp, if_opackets);
4705 		m = sc->bge_cdata.bge_tx_chain[idx];
4706 		if (m != NULL) {
4707 			sc->bge_cdata.bge_tx_chain[idx] = NULL;
4708 			dma = sc->txdma[idx];
4709 			if (dma->is_dma32) {
4710 				bus_dmamap_sync(sc->bge_dmatag32, dma->dmamap32,
4711 				    0, dma->dmamap32->dm_mapsize,
4712 				    BUS_DMASYNC_POSTWRITE);
4713 				bus_dmamap_unload(
4714 				    sc->bge_dmatag32, dma->dmamap32);
4715 			} else {
4716 				bus_dmamap_sync(sc->bge_dmatag, dma->dmamap,
4717 				    0, dma->dmamap->dm_mapsize,
4718 				    BUS_DMASYNC_POSTWRITE);
4719 				bus_dmamap_unload(sc->bge_dmatag, dma->dmamap);
4720 			}
4721 			SLIST_INSERT_HEAD(&sc->txdma_list, dma, link);
4722 			sc->txdma[idx] = NULL;
4723 
4724 			m_freem(m);
4725 		}
4726 		sc->bge_txcnt--;
4727 		BGE_INC(sc->bge_tx_saved_considx, BGE_TX_RING_CNT);
4728 		sc->bge_tx_sending = false;
4729 	}
4730 }
4731 
4732 static int
4733 bge_intr(void *xsc)
4734 {
4735 	struct bge_softc * const sc = xsc;
4736 	struct ifnet * const ifp = &sc->ethercom.ec_if;
4737 	uint32_t pcistate, statusword, statustag;
4738 	uint32_t intrmask = BGE_PCISTATE_INTR_NOT_ACTIVE;
4739 
4740 	/* 5717 and newer chips have no BGE_PCISTATE_INTR_NOT_ACTIVE bit */
4741 	if (BGE_IS_5717_PLUS(sc))
4742 		intrmask = 0;
4743 
4744 	mutex_enter(sc->sc_intr_lock);
4745 	if (sc->bge_txrx_stopping) {
4746 		mutex_exit(sc->sc_intr_lock);
4747 		return 1;
4748 	}
4749 
4750 	/*
4751 	 * It is possible for the interrupt to arrive before
4752 	 * the status block is updated prior to the interrupt.
4753 	 * Reading the PCI State register will confirm whether the
4754 	 * interrupt is ours and will flush the status block.
4755 	 */
4756 	pcistate = CSR_READ_4(sc, BGE_PCI_PCISTATE);
4757 
4758 	/* read status word from status block */
4759 	bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
4760 	    offsetof(struct bge_ring_data, bge_status_block),
4761 	    sizeof(struct bge_status_block),
4762 	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
4763 	statusword = sc->bge_rdata->bge_status_block.bge_status;
4764 	statustag = sc->bge_rdata->bge_status_block.bge_status_tag << 24;
4765 
4766 	if (sc->bge_flags & BGEF_TAGGED_STATUS) {
4767 		if (sc->bge_lasttag == statustag &&
4768 		    (~pcistate & intrmask)) {
4769 			BGE_EVCNT_INCR(sc->bge_ev_intr_spurious);
4770 			mutex_exit(sc->sc_intr_lock);
4771 			return 0;
4772 		}
4773 		sc->bge_lasttag = statustag;
4774 	} else {
4775 		if (!(statusword & BGE_STATFLAG_UPDATED) &&
4776 		    !(~pcistate & intrmask)) {
4777 			BGE_EVCNT_INCR(sc->bge_ev_intr_spurious2);
4778 			mutex_exit(sc->sc_intr_lock);
4779 			return 0;
4780 		}
4781 		statustag = 0;
4782 	}
4783 	/* Ack interrupt and stop others from occurring. */
4784 	bge_writembx_flush(sc, BGE_MBX_IRQ0_LO, 1);
4785 	BGE_EVCNT_INCR(sc->bge_ev_intr);
4786 
4787 	/* clear status word */
4788 	sc->bge_rdata->bge_status_block.bge_status = 0;
4789 
4790 	bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
4791 	    offsetof(struct bge_ring_data, bge_status_block),
4792 	    sizeof(struct bge_status_block),
4793 	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
4794 
4795 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5700 ||
4796 	    statusword & BGE_STATFLAG_LINKSTATE_CHANGED ||
4797 	    BGE_STS_BIT(sc, BGE_STS_LINK_EVT))
4798 		bge_link_upd(sc);
4799 
4800 	/* Check RX return ring producer/consumer */
4801 	bge_rxeof(sc);
4802 
4803 	/* Check TX ring producer/consumer */
4804 	bge_txeof(sc);
4805 
4806 	if (sc->bge_pending_rxintr_change) {
4807 		uint32_t rx_ticks = sc->bge_rx_coal_ticks;
4808 		uint32_t rx_bds = sc->bge_rx_max_coal_bds;
4809 
4810 		CSR_WRITE_4(sc, BGE_HCC_RX_COAL_TICKS, rx_ticks);
4811 		DELAY(10);
4812 		(void)CSR_READ_4(sc, BGE_HCC_RX_COAL_TICKS);
4813 
4814 		CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS, rx_bds);
4815 		DELAY(10);
4816 		(void)CSR_READ_4(sc, BGE_HCC_RX_MAX_COAL_BDS);
4817 
4818 		sc->bge_pending_rxintr_change = false;
4819 	}
4820 	bge_handle_events(sc);
4821 
4822 	/* Re-enable interrupts. */
4823 	bge_writembx_flush(sc, BGE_MBX_IRQ0_LO, statustag);
4824 
4825 	if_schedule_deferred_start(ifp);
4826 
4827 	mutex_exit(sc->sc_intr_lock);
4828 
4829 	return 1;
4830 }
4831 
4832 static void
4833 bge_asf_driver_up(struct bge_softc *sc)
4834 {
4835 	if (sc->bge_asf_mode & ASF_STACKUP) {
4836 		/* Send ASF heartbeat approx. every 2s */
4837 		if (sc->bge_asf_count)
4838 			sc->bge_asf_count --;
4839 		else {
4840 			sc->bge_asf_count = 2;
4841 
4842 			bge_wait_for_event_ack(sc);
4843 
4844 			bge_writemem_ind(sc, BGE_SRAM_FW_CMD_MB,
4845 			    BGE_FW_CMD_DRV_ALIVE3);
4846 			bge_writemem_ind(sc, BGE_SRAM_FW_CMD_LEN_MB, 4);
4847 			bge_writemem_ind(sc, BGE_SRAM_FW_CMD_DATA_MB,
4848 			    BGE_FW_HB_TIMEOUT_SEC);
4849 			CSR_WRITE_4_FLUSH(sc, BGE_RX_CPU_EVENT,
4850 			    CSR_READ_4(sc, BGE_RX_CPU_EVENT) |
4851 			    BGE_RX_CPU_DRV_EVENT);
4852 		}
4853 	}
4854 }
4855 
4856 static void
4857 bge_tick(void *xsc)
4858 {
4859 	struct bge_softc * const sc = xsc;
4860 	struct ifnet * const ifp = &sc->ethercom.ec_if;
4861 	struct mii_data * const mii = &sc->bge_mii;
4862 
4863 	mutex_enter(sc->sc_intr_lock);
4864 
4865 	if (BGE_IS_5705_PLUS(sc))
4866 		bge_stats_update_regs(sc);
4867 	else
4868 		bge_stats_update(sc);
4869 
4870 	if (sc->bge_flags & BGEF_FIBER_TBI) {
4871 		/*
4872 		 * Since in TBI mode auto-polling can't be used we should poll
4873 		 * link status manually. Here we register pending link event
4874 		 * and trigger interrupt.
4875 		 */
4876 		BGE_STS_SETBIT(sc, BGE_STS_LINK_EVT);
4877 		BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_INTR_SET);
4878 	} else {
4879 		/*
4880 		 * Do not touch PHY if we have link up. This could break
4881 		 * IPMI/ASF mode or produce extra input errors.
4882 		 * (extra input errors was reported for bcm5701 & bcm5704).
4883 		 */
4884 		if (!BGE_STS_BIT(sc, BGE_STS_LINK)) {
4885 			mii_tick(mii);
4886 		}
4887 	}
4888 
4889 	bge_asf_driver_up(sc);
4890 
4891 	const bool ok = bge_watchdog_tick(ifp);
4892 	if (ok)
4893 		callout_schedule(&sc->bge_timeout, hz);
4894 	mutex_exit(sc->sc_intr_lock);
4895 }
4896 
4897 static void
4898 bge_stats_update_regs(struct bge_softc *sc)
4899 {
4900 	struct ifnet * const ifp = &sc->ethercom.ec_if;
4901 
4902 	net_stat_ref_t nsr = IF_STAT_GETREF(ifp);
4903 
4904 	if_statadd_ref(ifp, nsr, if_collisions,
4905 	    CSR_READ_4(sc, BGE_MAC_STATS +
4906 	    offsetof(struct bge_mac_stats_regs, etherStatsCollisions)));
4907 
4908 	/*
4909 	 * On BCM5717, BCM5718, BCM5719 A0 and BCM5720 A0,
4910 	 * RXLP_LOCSTAT_IFIN_DROPS includes unwanted multicast frames
4911 	 * (silicon bug). There's no reliable workaround so just
4912 	 * ignore the counter
4913 	 */
4914 	if (BGE_ASICREV(sc->bge_chipid) != BGE_ASICREV_BCM5717 &&
4915 	    sc->bge_chipid != BGE_CHIPID_BCM5719_A0 &&
4916 	    sc->bge_chipid != BGE_CHIPID_BCM5720_A0) {
4917 		if_statadd_ref(ifp, nsr, if_ierrors,
4918 		    CSR_READ_4(sc, BGE_RXLP_LOCSTAT_IFIN_DROPS));
4919 	}
4920 	if_statadd_ref(ifp, nsr, if_ierrors,
4921 	    CSR_READ_4(sc, BGE_RXLP_LOCSTAT_IFIN_ERRORS));
4922 	if_statadd_ref(ifp, nsr, if_ierrors,
4923 	    CSR_READ_4(sc, BGE_RXLP_LOCSTAT_OUT_OF_BDS));
4924 
4925 	IF_STAT_PUTREF(ifp);
4926 
4927 	if (sc->bge_flags & BGEF_RDMA_BUG) {
4928 		uint32_t val, ucast, mcast, bcast;
4929 
4930 		ucast = CSR_READ_4(sc, BGE_MAC_STATS +
4931 		    offsetof(struct bge_mac_stats_regs, ifHCOutUcastPkts));
4932 		mcast = CSR_READ_4(sc, BGE_MAC_STATS +
4933 		    offsetof(struct bge_mac_stats_regs, ifHCOutMulticastPkts));
4934 		bcast = CSR_READ_4(sc, BGE_MAC_STATS +
4935 		    offsetof(struct bge_mac_stats_regs, ifHCOutBroadcastPkts));
4936 
4937 		/*
4938 		 * If controller transmitted more than BGE_NUM_RDMA_CHANNELS
4939 		 * frames, it's safe to disable workaround for DMA engine's
4940 		 * miscalculation of TXMBUF space.
4941 		 */
4942 		if (ucast + mcast + bcast > BGE_NUM_RDMA_CHANNELS) {
4943 			val = CSR_READ_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL);
4944 			if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5719)
4945 				val &= ~BGE_RDMA_TX_LENGTH_WA_5719;
4946 			else
4947 				val &= ~BGE_RDMA_TX_LENGTH_WA_5720;
4948 			CSR_WRITE_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL, val);
4949 			sc->bge_flags &= ~BGEF_RDMA_BUG;
4950 		}
4951 	}
4952 }
4953 
4954 static void
4955 bge_stats_update(struct bge_softc *sc)
4956 {
4957 	struct ifnet * const ifp = &sc->ethercom.ec_if;
4958 	bus_size_t stats = BGE_MEMWIN_START + BGE_STATS_BLOCK;
4959 
4960 #define READ_STAT(sc, stats, stat) \
4961 	  CSR_READ_4(sc, stats + offsetof(struct bge_stats, stat))
4962 
4963 	uint64_t collisions =
4964 	  (READ_STAT(sc, stats, dot3StatsSingleCollisionFrames.bge_addr_lo) +
4965 	   READ_STAT(sc, stats, dot3StatsMultipleCollisionFrames.bge_addr_lo) +
4966 	   READ_STAT(sc, stats, dot3StatsExcessiveCollisions.bge_addr_lo) +
4967 	   READ_STAT(sc, stats, dot3StatsLateCollisions.bge_addr_lo));
4968 
4969 	if_statadd(ifp, if_collisions, collisions - sc->bge_if_collisions);
4970 	sc->bge_if_collisions = collisions;
4971 
4972 
4973 	BGE_EVCNT_UPD(sc->bge_ev_tx_xoff,
4974 		      READ_STAT(sc, stats, outXoffSent.bge_addr_lo));
4975 	BGE_EVCNT_UPD(sc->bge_ev_tx_xon,
4976 		      READ_STAT(sc, stats, outXonSent.bge_addr_lo));
4977 	BGE_EVCNT_UPD(sc->bge_ev_rx_xoff,
4978 		      READ_STAT(sc, stats,
4979 				xoffPauseFramesReceived.bge_addr_lo));
4980 	BGE_EVCNT_UPD(sc->bge_ev_rx_xon,
4981 		      READ_STAT(sc, stats, xonPauseFramesReceived.bge_addr_lo));
4982 	BGE_EVCNT_UPD(sc->bge_ev_rx_macctl,
4983 		      READ_STAT(sc, stats,
4984 				macControlFramesReceived.bge_addr_lo));
4985 	BGE_EVCNT_UPD(sc->bge_ev_xoffentered,
4986 		      READ_STAT(sc, stats, xoffStateEntered.bge_addr_lo));
4987 
4988 #undef READ_STAT
4989 }
4990 
4991 /*
4992  * Pad outbound frame to ETHER_MIN_NOPAD for an unusual reason.
4993  * The bge hardware will pad out Tx runts to ETHER_MIN_NOPAD,
4994  * but when such padded frames employ the  bge IP/TCP checksum offload,
4995  * the hardware checksum assist gives incorrect results (possibly
4996  * from incorporating its own padding into the UDP/TCP checksum; who knows).
4997  * If we pad such runts with zeros, the onboard checksum comes out correct.
4998  */
4999 static inline int
5000 bge_cksum_pad(struct mbuf *pkt)
5001 {
5002 	struct mbuf *last = NULL;
5003 	int padlen;
5004 
5005 	padlen = ETHER_MIN_NOPAD - pkt->m_pkthdr.len;
5006 
5007 	/* if there's only the packet-header and we can pad there, use it. */
5008 	if (pkt->m_pkthdr.len == pkt->m_len &&
5009 	    M_TRAILINGSPACE(pkt) >= padlen) {
5010 		last = pkt;
5011 	} else {
5012 		/*
5013 		 * Walk packet chain to find last mbuf. We will either
5014 		 * pad there, or append a new mbuf and pad it
5015 		 * (thus perhaps avoiding the bcm5700 dma-min bug).
5016 		 */
5017 		for (last = pkt; last->m_next != NULL; last = last->m_next) {
5018 			continue; /* do nothing */
5019 		}
5020 
5021 		/* `last' now points to last in chain. */
5022 		if (M_TRAILINGSPACE(last) < padlen) {
5023 			/* Allocate new empty mbuf, pad it. Compact later. */
5024 			struct mbuf *n;
5025 			MGET(n, M_DONTWAIT, MT_DATA);
5026 			if (n == NULL)
5027 				return ENOBUFS;
5028 			MCLAIM(n, last->m_owner);
5029 			n->m_len = 0;
5030 			last->m_next = n;
5031 			last = n;
5032 		}
5033 	}
5034 
5035 	KDASSERT(!M_READONLY(last));
5036 	KDASSERT(M_TRAILINGSPACE(last) >= padlen);
5037 
5038 	/* Now zero the pad area, to avoid the bge cksum-assist bug */
5039 	memset(mtod(last, char *) + last->m_len, 0, padlen);
5040 	last->m_len += padlen;
5041 	pkt->m_pkthdr.len += padlen;
5042 	return 0;
5043 }
5044 
5045 /*
5046  * Compact outbound packets to avoid bug with DMA segments less than 8 bytes.
5047  */
5048 static inline int
5049 bge_compact_dma_runt(struct mbuf *pkt)
5050 {
5051 	struct mbuf	*m, *prev;
5052 	int		totlen;
5053 
5054 	prev = NULL;
5055 	totlen = 0;
5056 
5057 	for (m = pkt; m != NULL; prev = m, m = m->m_next) {
5058 		int mlen = m->m_len;
5059 		int shortfall = 8 - mlen ;
5060 
5061 		totlen += mlen;
5062 		if (mlen == 0)
5063 			continue;
5064 		if (mlen >= 8)
5065 			continue;
5066 
5067 		/*
5068 		 * If we get here, mbuf data is too small for DMA engine.
5069 		 * Try to fix by shuffling data to prev or next in chain.
5070 		 * If that fails, do a compacting deep-copy of the whole chain.
5071 		 */
5072 
5073 		/* Internal frag. If fits in prev, copy it there. */
5074 		if (prev && M_TRAILINGSPACE(prev) >= m->m_len) {
5075 			memcpy(prev->m_data + prev->m_len, m->m_data, mlen);
5076 			prev->m_len += mlen;
5077 			m->m_len = 0;
5078 			/* XXX stitch chain */
5079 			prev->m_next = m_free(m);
5080 			m = prev;
5081 			continue;
5082 		} else if (m->m_next != NULL &&
5083 			    M_TRAILINGSPACE(m) >= shortfall &&
5084 			    m->m_next->m_len >= (8 + shortfall)) {
5085 		    /* m is writable and have enough data in next, pull up. */
5086 
5087 			memcpy(m->m_data + m->m_len, m->m_next->m_data,
5088 			    shortfall);
5089 			m->m_len += shortfall;
5090 			m->m_next->m_len -= shortfall;
5091 			m->m_next->m_data += shortfall;
5092 		} else if (m->m_next == NULL || 1) {
5093 			/*
5094 			 * Got a runt at the very end of the packet.
5095 			 * borrow data from the tail of the preceding mbuf and
5096 			 * update its length in-place. (The original data is
5097 			 * still valid, so we can do this even if prev is not
5098 			 * writable.)
5099 			 */
5100 
5101 			/*
5102 			 * If we'd make prev a runt, just move all of its data.
5103 			 */
5104 			KASSERT(prev != NULL /*, ("runt but null PREV")*/);
5105 			KASSERT(prev->m_len >= 8 /*, ("runt prev")*/);
5106 
5107 			if ((prev->m_len - shortfall) < 8)
5108 				shortfall = prev->m_len;
5109 
5110 #ifdef notyet	/* just do the safe slow thing for now */
5111 			if (!M_READONLY(m)) {
5112 				if (M_LEADINGSPACE(m) < shorfall) {
5113 					void *m_dat;
5114 					m_dat = M_BUFADDR(m);
5115 					memmove(m_dat, mtod(m, void*),
5116 					    m->m_len);
5117 					m->m_data = m_dat;
5118 				}
5119 			} else
5120 #endif	/* just do the safe slow thing */
5121 			{
5122 				struct mbuf * n = NULL;
5123 				int newprevlen = prev->m_len - shortfall;
5124 
5125 				MGET(n, M_NOWAIT, MT_DATA);
5126 				if (n == NULL)
5127 				   return ENOBUFS;
5128 				MCLAIM(n, prev->m_owner);
5129 				KASSERT(m->m_len + shortfall < MLEN
5130 					/*,
5131 					  ("runt %d +prev %d too big\n", m->m_len, shortfall)*/);
5132 
5133 				/* first copy the data we're stealing from prev */
5134 				memcpy(n->m_data, prev->m_data + newprevlen,
5135 				    shortfall);
5136 
5137 				/* update prev->m_len accordingly */
5138 				prev->m_len -= shortfall;
5139 
5140 				/* copy data from runt m */
5141 				memcpy(n->m_data + shortfall, m->m_data,
5142 				    m->m_len);
5143 
5144 				/* n holds what we stole from prev, plus m */
5145 				n->m_len = shortfall + m->m_len;
5146 
5147 				/* stitch n into chain and free m */
5148 				n->m_next = m->m_next;
5149 				prev->m_next = n;
5150 				/* KASSERT(m->m_next == NULL); */
5151 				m->m_next = NULL;
5152 				m_free(m);
5153 				m = n;	/* for continuing loop */
5154 			}
5155 		}
5156 	}
5157 	return 0;
5158 }
5159 
5160 /*
5161  * Encapsulate an mbuf chain in the tx ring by coupling the mbuf data
5162  * pointers to descriptors.
5163  */
5164 static int
5165 bge_encap(struct bge_softc *sc, struct mbuf *m_head, uint32_t *txidx)
5166 {
5167 	struct bge_tx_bd	*f, *prev_f;
5168 	uint32_t		frag, cur;
5169 	uint16_t		csum_flags = 0;
5170 	uint16_t		txbd_tso_flags = 0;
5171 	struct txdmamap_pool_entry *dma;
5172 	bus_dmamap_t dmamap;
5173 	bus_dma_tag_t dmatag;
5174 	int			i = 0;
5175 	int			use_tso, maxsegsize, error;
5176 	bool			have_vtag;
5177 	uint16_t		vtag;
5178 	bool			remap;
5179 
5180 	KASSERT(mutex_owned(sc->sc_intr_lock));
5181 
5182 	if (m_head->m_pkthdr.csum_flags) {
5183 		if (m_head->m_pkthdr.csum_flags & M_CSUM_IPv4)
5184 			csum_flags |= BGE_TXBDFLAG_IP_CSUM;
5185 		if (m_head->m_pkthdr.csum_flags & (M_CSUM_TCPv4 |M_CSUM_UDPv4))
5186 			csum_flags |= BGE_TXBDFLAG_TCP_UDP_CSUM;
5187 	}
5188 
5189 	/*
5190 	 * If we were asked to do an outboard checksum, and the NIC
5191 	 * has the bug where it sometimes adds in the Ethernet padding,
5192 	 * explicitly pad with zeros so the cksum will be correct either way.
5193 	 * (For now, do this for all chip versions, until newer
5194 	 * are confirmed to not require the workaround.)
5195 	 */
5196 	if ((csum_flags & BGE_TXBDFLAG_TCP_UDP_CSUM) == 0 ||
5197 #ifdef notyet
5198 	    (sc->bge_quirks & BGE_QUIRK_SHORT_CKSUM_BUG) == 0 ||
5199 #endif
5200 	    m_head->m_pkthdr.len >= ETHER_MIN_NOPAD)
5201 		goto check_dma_bug;
5202 
5203 	if (bge_cksum_pad(m_head) != 0)
5204 		return ENOBUFS;
5205 
5206 check_dma_bug:
5207 	if (!(BGE_CHIPREV(sc->bge_chipid) == BGE_CHIPREV_5700_BX))
5208 		goto doit;
5209 
5210 	/*
5211 	 * bcm5700 Revision B silicon cannot handle DMA descriptors with
5212 	 * less than eight bytes.  If we encounter a teeny mbuf
5213 	 * at the end of a chain, we can pad.  Otherwise, copy.
5214 	 */
5215 	if (bge_compact_dma_runt(m_head) != 0)
5216 		return ENOBUFS;
5217 
5218 doit:
5219 	dma = SLIST_FIRST(&sc->txdma_list);
5220 	if (dma == NULL) {
5221 		return ENOBUFS;
5222 	}
5223 	dmamap = dma->dmamap;
5224 	dmatag = sc->bge_dmatag;
5225 	dma->is_dma32 = false;
5226 
5227 	/*
5228 	 * Set up any necessary TSO state before we start packing...
5229 	 */
5230 	use_tso = (m_head->m_pkthdr.csum_flags & M_CSUM_TSOv4) != 0;
5231 	if (!use_tso) {
5232 		maxsegsize = 0;
5233 	} else {	/* TSO setup */
5234 		unsigned  mss;
5235 		struct ether_header *eh;
5236 		unsigned ip_tcp_hlen, iptcp_opt_words, tcp_seg_flags, offset;
5237 		unsigned bge_hlen;
5238 		struct mbuf * m0 = m_head;
5239 		struct ip *ip;
5240 		struct tcphdr *th;
5241 		int iphl, hlen;
5242 
5243 		/*
5244 		 * XXX It would be nice if the mbuf pkthdr had offset
5245 		 * fields for the protocol headers.
5246 		 */
5247 
5248 		eh = mtod(m0, struct ether_header *);
5249 		switch (htons(eh->ether_type)) {
5250 		case ETHERTYPE_IP:
5251 			offset = ETHER_HDR_LEN;
5252 			break;
5253 
5254 		case ETHERTYPE_VLAN:
5255 			offset = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
5256 			break;
5257 
5258 		default:
5259 			/*
5260 			 * Don't support this protocol or encapsulation.
5261 			 */
5262 			return ENOBUFS;
5263 		}
5264 
5265 		/*
5266 		 * TCP/IP headers are in the first mbuf; we can do
5267 		 * this the easy way.
5268 		 */
5269 		iphl = M_CSUM_DATA_IPv4_IPHL(m0->m_pkthdr.csum_data);
5270 		hlen = iphl + offset;
5271 		if (__predict_false(m0->m_len <
5272 				    (hlen + sizeof(struct tcphdr)))) {
5273 
5274 			aprint_error_dev(sc->bge_dev,
5275 			    "TSO: hard case m0->m_len == %d < ip/tcp hlen %zd,"
5276 			    "not handled yet\n",
5277 			    m0->m_len, hlen+ sizeof(struct tcphdr));
5278 #ifdef NOTYET
5279 			/*
5280 			 * XXX jonathan@NetBSD.org: untested.
5281 			 * how to force this branch to be taken?
5282 			 */
5283 			BGE_EVCNT_INCR(sc->bge_ev_txtsopain);
5284 
5285 			m_copydata(m0, offset, sizeof(ip), &ip);
5286 			m_copydata(m0, hlen, sizeof(th), &th);
5287 
5288 			ip.ip_len = 0;
5289 
5290 			m_copyback(m0, hlen + offsetof(struct ip, ip_len),
5291 			    sizeof(ip.ip_len), &ip.ip_len);
5292 
5293 			th.th_sum = in_cksum_phdr(ip.ip_src.s_addr,
5294 			    ip.ip_dst.s_addr, htons(IPPROTO_TCP));
5295 
5296 			m_copyback(m0, hlen + offsetof(struct tcphdr, th_sum),
5297 			    sizeof(th.th_sum), &th.th_sum);
5298 
5299 			hlen += th.th_off << 2;
5300 			iptcp_opt_words	= hlen;
5301 #else
5302 			/*
5303 			 * if_wm "hard" case not yet supported, can we not
5304 			 * mandate it out of existence?
5305 			 */
5306 			(void) ip; (void)th; (void) ip_tcp_hlen;
5307 
5308 			return ENOBUFS;
5309 #endif
5310 		} else {
5311 			ip = (struct ip *) (mtod(m0, char *) + offset);
5312 			th = (struct tcphdr *) (mtod(m0, char *) + hlen);
5313 			ip_tcp_hlen = iphl +  (th->th_off << 2);
5314 
5315 			/* Total IP/TCP options, in 32-bit words */
5316 			iptcp_opt_words = (ip_tcp_hlen
5317 					   - sizeof(struct tcphdr)
5318 					   - sizeof(struct ip)) >> 2;
5319 		}
5320 		if (BGE_IS_575X_PLUS(sc)) {
5321 			th->th_sum = 0;
5322 			csum_flags = 0;
5323 		} else {
5324 			/*
5325 			 * XXX jonathan@NetBSD.org: 5705 untested.
5326 			 * Requires TSO firmware patch for 5701/5703/5704.
5327 			 */
5328 			th->th_sum = in_cksum_phdr(ip->ip_src.s_addr,
5329 			    ip->ip_dst.s_addr, htons(IPPROTO_TCP));
5330 		}
5331 
5332 		mss = m_head->m_pkthdr.segsz;
5333 		txbd_tso_flags |=
5334 		    BGE_TXBDFLAG_CPU_PRE_DMA |
5335 		    BGE_TXBDFLAG_CPU_POST_DMA;
5336 
5337 		/*
5338 		 * Our NIC TSO-assist assumes TSO has standard, optionless
5339 		 * IPv4 and TCP headers, which total 40 bytes. By default,
5340 		 * the NIC copies 40 bytes of IP/TCP header from the
5341 		 * supplied header into the IP/TCP header portion of
5342 		 * each post-TSO-segment. If the supplied packet has IP or
5343 		 * TCP options, we need to tell the NIC to copy those extra
5344 		 * bytes into each  post-TSO header, in addition to the normal
5345 		 * 40-byte IP/TCP header (and to leave space accordingly).
5346 		 * Unfortunately, the driver encoding of option length
5347 		 * varies across different ASIC families.
5348 		 */
5349 		tcp_seg_flags = 0;
5350 		bge_hlen = ip_tcp_hlen >> 2;
5351 		if (BGE_IS_5717_PLUS(sc)) {
5352 			tcp_seg_flags = (bge_hlen & 0x3) << 14;
5353 			txbd_tso_flags |=
5354 			    ((bge_hlen & 0xF8) << 7) | ((bge_hlen & 0x4) << 2);
5355 		} else if (BGE_IS_5705_PLUS(sc)) {
5356 			tcp_seg_flags = bge_hlen << 11;
5357 		} else {
5358 			/* XXX iptcp_opt_words or bge_hlen ? */
5359 			txbd_tso_flags |= iptcp_opt_words << 12;
5360 		}
5361 		maxsegsize = mss | tcp_seg_flags;
5362 		ip->ip_len = htons(mss + ip_tcp_hlen);
5363 		ip->ip_sum = 0;
5364 
5365 	}	/* TSO setup */
5366 
5367 	have_vtag = vlan_has_tag(m_head);
5368 	if (have_vtag)
5369 		vtag = vlan_get_tag(m_head);
5370 
5371 	/*
5372 	 * Start packing the mbufs in this chain into
5373 	 * the fragment pointers. Stop when we run out
5374 	 * of fragments or hit the end of the mbuf chain.
5375 	 */
5376 	remap = true;
5377 load_again:
5378 	error = bus_dmamap_load_mbuf(dmatag, dmamap, m_head, BUS_DMA_NOWAIT);
5379 	if (__predict_false(error)) {
5380 		if (error == EFBIG && remap) {
5381 			struct mbuf *m;
5382 			remap = false;
5383 			m = m_defrag(m_head, M_NOWAIT);
5384 			if (m != NULL) {
5385 				KASSERT(m == m_head);
5386 				goto load_again;
5387 			}
5388 		}
5389 		return error;
5390 	}
5391 	/*
5392 	 * Sanity check: avoid coming within 16 descriptors
5393 	 * of the end of the ring.
5394 	 */
5395 	if (dmamap->dm_nsegs > (BGE_TX_RING_CNT - sc->bge_txcnt - 16)) {
5396 		BGE_TSO_PRINTF(("%s: "
5397 		    " dmamap_load_mbuf too close to ring wrap\n",
5398 		    device_xname(sc->bge_dev)));
5399 		goto fail_unload;
5400 	}
5401 
5402 	/* Iterate over dmap-map fragments. */
5403 	f = prev_f = NULL;
5404 	cur = frag = *txidx;
5405 
5406 	for (i = 0; i < dmamap->dm_nsegs; i++) {
5407 		f = &sc->bge_rdata->bge_tx_ring[frag];
5408 		if (sc->bge_cdata.bge_tx_chain[frag] != NULL)
5409 			break;
5410 
5411 		BGE_HOSTADDR(f->bge_addr, dmamap->dm_segs[i].ds_addr);
5412 		f->bge_len = dmamap->dm_segs[i].ds_len;
5413 		if (sizeof(bus_addr_t) > 4 && dma->is_dma32 == false && use_tso && (
5414 		    (dmamap->dm_segs[i].ds_addr & 0xffffffff00000000) !=
5415 		    ((dmamap->dm_segs[i].ds_addr + f->bge_len) & 0xffffffff00000000) ||
5416 		    (prev_f != NULL &&
5417 		     prev_f->bge_addr.bge_addr_hi != f->bge_addr.bge_addr_hi))
5418 		   ) {
5419 			/*
5420 			 * watchdog timeout issue was observed with TSO,
5421 			 * limiting DMA address space to 32bits seems to
5422 			 * address the issue.
5423 			 */
5424 			bus_dmamap_unload(dmatag, dmamap);
5425 			dmatag = sc->bge_dmatag32;
5426 			dmamap = dma->dmamap32;
5427 			dma->is_dma32 = true;
5428 			remap = true;
5429 			goto load_again;
5430 		}
5431 
5432 		/*
5433 		 * For 5751 and follow-ons, for TSO we must turn
5434 		 * off checksum-assist flag in the tx-descr, and
5435 		 * supply the ASIC-revision-specific encoding
5436 		 * of TSO flags and segsize.
5437 		 */
5438 		if (use_tso) {
5439 			if (BGE_IS_575X_PLUS(sc) || i == 0) {
5440 				f->bge_rsvd = maxsegsize;
5441 				f->bge_flags = csum_flags | txbd_tso_flags;
5442 			} else {
5443 				f->bge_rsvd = 0;
5444 				f->bge_flags =
5445 				  (csum_flags | txbd_tso_flags) & 0x0fff;
5446 			}
5447 		} else {
5448 			f->bge_rsvd = 0;
5449 			f->bge_flags = csum_flags;
5450 		}
5451 
5452 		if (have_vtag) {
5453 			f->bge_flags |= BGE_TXBDFLAG_VLAN_TAG;
5454 			f->bge_vlan_tag = vtag;
5455 		} else {
5456 			f->bge_vlan_tag = 0;
5457 		}
5458 		prev_f = f;
5459 		cur = frag;
5460 		BGE_INC(frag, BGE_TX_RING_CNT);
5461 	}
5462 
5463 	if (i < dmamap->dm_nsegs) {
5464 		BGE_TSO_PRINTF(("%s: reached %d < dm_nsegs %d\n",
5465 		    device_xname(sc->bge_dev), i, dmamap->dm_nsegs));
5466 		goto fail_unload;
5467 	}
5468 
5469 	bus_dmamap_sync(dmatag, dmamap, 0, dmamap->dm_mapsize,
5470 	    BUS_DMASYNC_PREWRITE);
5471 
5472 	if (frag == sc->bge_tx_saved_considx) {
5473 		BGE_TSO_PRINTF(("%s: frag %d = wrapped id %d?\n",
5474 		    device_xname(sc->bge_dev), frag, sc->bge_tx_saved_considx));
5475 
5476 		goto fail_unload;
5477 	}
5478 
5479 	sc->bge_rdata->bge_tx_ring[cur].bge_flags |= BGE_TXBDFLAG_END;
5480 	sc->bge_cdata.bge_tx_chain[cur] = m_head;
5481 	SLIST_REMOVE_HEAD(&sc->txdma_list, link);
5482 	sc->txdma[cur] = dma;
5483 	sc->bge_txcnt += dmamap->dm_nsegs;
5484 
5485 	*txidx = frag;
5486 
5487 	return 0;
5488 
5489 fail_unload:
5490 	bus_dmamap_unload(dmatag, dmamap);
5491 
5492 	return ENOBUFS;
5493 }
5494 
5495 
5496 static void
5497 bge_start(struct ifnet *ifp)
5498 {
5499 	struct bge_softc * const sc = ifp->if_softc;
5500 
5501 	mutex_enter(sc->sc_intr_lock);
5502 	if (!sc->bge_txrx_stopping)
5503 		bge_start_locked(ifp);
5504 	mutex_exit(sc->sc_intr_lock);
5505 }
5506 
5507 /*
5508  * Main transmit routine. To avoid having to do mbuf copies, we put pointers
5509  * to the mbuf data regions directly in the transmit descriptors.
5510  */
5511 static void
5512 bge_start_locked(struct ifnet *ifp)
5513 {
5514 	struct bge_softc * const sc = ifp->if_softc;
5515 	struct mbuf *m_head = NULL;
5516 	struct mbuf *m;
5517 	uint32_t prodidx;
5518 	int pkts = 0;
5519 	int error;
5520 
5521 	KASSERT(mutex_owned(sc->sc_intr_lock));
5522 
5523 	prodidx = sc->bge_tx_prodidx;
5524 
5525 	while (sc->bge_cdata.bge_tx_chain[prodidx] == NULL) {
5526 		IFQ_POLL(&ifp->if_snd, m_head);
5527 		if (m_head == NULL)
5528 			break;
5529 
5530 #if 0
5531 		/*
5532 		 * XXX
5533 		 * safety overkill.  If this is a fragmented packet chain
5534 		 * with delayed TCP/UDP checksums, then only encapsulate
5535 		 * it if we have enough descriptors to handle the entire
5536 		 * chain at once.
5537 		 * (paranoia -- may not actually be needed)
5538 		 */
5539 		if (m_head->m_flags & M_FIRSTFRAG &&
5540 		    m_head->m_pkthdr.csum_flags & (CSUM_DELAY_DATA)) {
5541 			if ((BGE_TX_RING_CNT - sc->bge_txcnt) <
5542 			    M_CSUM_DATA_IPv4_OFFSET(m_head->m_pkthdr.csum_data) + 16) {
5543 				ifp->if_flags |= IFF_OACTIVE;
5544 				break;
5545 			}
5546 		}
5547 #endif
5548 
5549 		/*
5550 		 * Pack the data into the transmit ring. If we
5551 		 * don't have room, set the OACTIVE flag and wait
5552 		 * for the NIC to drain the ring.
5553 		 */
5554 		error = bge_encap(sc, m_head, &prodidx);
5555 		if (__predict_false(error)) {
5556 			if (SLIST_EMPTY(&sc->txdma_list)) {
5557 				/* just wait for the transmit ring to drain */
5558 				break;
5559 			}
5560 			IFQ_DEQUEUE(&ifp->if_snd, m);
5561 			KASSERT(m == m_head);
5562 			m_freem(m_head);
5563 			continue;
5564 		}
5565 
5566 		/* now we are committed to transmit the packet */
5567 		IFQ_DEQUEUE(&ifp->if_snd, m);
5568 		KASSERT(m == m_head);
5569 		pkts++;
5570 
5571 		/*
5572 		 * If there's a BPF listener, bounce a copy of this frame
5573 		 * to him.
5574 		 */
5575 		bpf_mtap(ifp, m_head, BPF_D_OUT);
5576 	}
5577 	if (pkts == 0)
5578 		return;
5579 
5580 	/* Transmit */
5581 	bge_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, prodidx);
5582 	/* 5700 b2 errata */
5583 	if (BGE_CHIPREV(sc->bge_chipid) == BGE_CHIPREV_5700_BX)
5584 		bge_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, prodidx);
5585 
5586 	sc->bge_tx_prodidx = prodidx;
5587 	sc->bge_tx_lastsent = time_uptime;
5588 	sc->bge_tx_sending = true;
5589 }
5590 
5591 static int
5592 bge_init(struct ifnet *ifp)
5593 {
5594 	struct bge_softc * const sc = ifp->if_softc;
5595 	const uint16_t *m;
5596 	uint32_t mode, reg;
5597 	int error = 0;
5598 
5599 	ASSERT_SLEEPABLE();
5600 	KASSERT(IFNET_LOCKED(ifp));
5601 	KASSERT(ifp == &sc->ethercom.ec_if);
5602 
5603 	if (sc->bge_detaching)
5604 		return ENXIO;
5605 
5606 	/* Cancel pending I/O and flush buffers. */
5607 	bge_stop(ifp, 0);
5608 
5609 	bge_stop_fw(sc);
5610 	bge_sig_pre_reset(sc, BGE_RESET_START);
5611 	bge_reset(sc);
5612 	bge_sig_legacy(sc, BGE_RESET_START);
5613 
5614 	if (BGE_CHIPREV(sc->bge_chipid) == BGE_CHIPREV_5784_AX) {
5615 		reg = CSR_READ_4(sc, BGE_CPMU_CTRL);
5616 		reg &= ~(BGE_CPMU_CTRL_LINK_AWARE_MODE |
5617 		    BGE_CPMU_CTRL_LINK_IDLE_MODE);
5618 		CSR_WRITE_4(sc, BGE_CPMU_CTRL, reg);
5619 
5620 		reg = CSR_READ_4(sc, BGE_CPMU_LSPD_10MB_CLK);
5621 		reg &= ~BGE_CPMU_LSPD_10MB_CLK;
5622 		reg |= BGE_CPMU_LSPD_10MB_MACCLK_6_25;
5623 		CSR_WRITE_4(sc, BGE_CPMU_LSPD_10MB_CLK, reg);
5624 
5625 		reg = CSR_READ_4(sc, BGE_CPMU_LNK_AWARE_PWRMD);
5626 		reg &= ~BGE_CPMU_LNK_AWARE_MACCLK_MASK;
5627 		reg |= BGE_CPMU_LNK_AWARE_MACCLK_6_25;
5628 		CSR_WRITE_4(sc, BGE_CPMU_LNK_AWARE_PWRMD, reg);
5629 
5630 		reg = CSR_READ_4(sc, BGE_CPMU_HST_ACC);
5631 		reg &= ~BGE_CPMU_HST_ACC_MACCLK_MASK;
5632 		reg |= BGE_CPMU_HST_ACC_MACCLK_6_25;
5633 		CSR_WRITE_4(sc, BGE_CPMU_HST_ACC, reg);
5634 	}
5635 
5636 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM57780) {
5637 		pcireg_t aercap;
5638 
5639 		reg = CSR_READ_4(sc, BGE_PCIE_PWRMNG_THRESH);
5640 		reg = (reg & ~BGE_PCIE_PWRMNG_L1THRESH_MASK)
5641 		    | BGE_PCIE_PWRMNG_L1THRESH_4MS
5642 		    | BGE_PCIE_PWRMNG_EXTASPMTMR_EN;
5643 		CSR_WRITE_4(sc, BGE_PCIE_PWRMNG_THRESH, reg);
5644 
5645 		reg = CSR_READ_4(sc, BGE_PCIE_EIDLE_DELAY);
5646 		reg = (reg & ~BGE_PCIE_EIDLE_DELAY_MASK)
5647 		    | BGE_PCIE_EIDLE_DELAY_13CLK;
5648 		CSR_WRITE_4(sc, BGE_PCIE_EIDLE_DELAY, reg);
5649 
5650 		/* Clear correctable error */
5651 		if (pci_get_ext_capability(sc->sc_pc, sc->sc_pcitag,
5652 		    PCI_EXTCAP_AER, &aercap, NULL) != 0)
5653 			pci_conf_write(sc->sc_pc, sc->sc_pcitag,
5654 			    aercap + PCI_AER_COR_STATUS, 0xffffffff);
5655 
5656 		reg = CSR_READ_4(sc, BGE_PCIE_LINKCTL);
5657 		reg = (reg & ~BGE_PCIE_LINKCTL_L1_PLL_PDEN)
5658 		    | BGE_PCIE_LINKCTL_L1_PLL_PDDIS;
5659 		CSR_WRITE_4(sc, BGE_PCIE_LINKCTL, reg);
5660 	}
5661 
5662 	bge_sig_post_reset(sc, BGE_RESET_START);
5663 
5664 	bge_chipinit(sc);
5665 
5666 	/*
5667 	 * Init the various state machines, ring
5668 	 * control blocks and firmware.
5669 	 */
5670 	error = bge_blockinit(sc);
5671 	if (error != 0) {
5672 		aprint_error_dev(sc->bge_dev, "initialization error %d\n",
5673 		    error);
5674 		return error;
5675 	}
5676 
5677 	/* 5718 step 25, 57XX step 54 */
5678 	/* Specify MTU. */
5679 	CSR_WRITE_4(sc, BGE_RX_MTU, ifp->if_mtu +
5680 	    ETHER_HDR_LEN + ETHER_CRC_LEN + ETHER_VLAN_ENCAP_LEN);
5681 
5682 	/* 5718 step 23 */
5683 	/* Load our MAC address. */
5684 	m = (const uint16_t *)&(CLLADDR(ifp->if_sadl)[0]);
5685 	CSR_WRITE_4(sc, BGE_MAC_ADDR1_LO, htons(m[0]));
5686 	CSR_WRITE_4(sc, BGE_MAC_ADDR1_HI,
5687 	    ((uint32_t)htons(m[1]) << 16) | htons(m[2]));
5688 
5689 	/* Enable or disable promiscuous mode as needed. */
5690 	if (ifp->if_flags & IFF_PROMISC)
5691 		BGE_SETBIT(sc, BGE_RX_MODE, BGE_RXMODE_RX_PROMISC);
5692 	else
5693 		BGE_CLRBIT(sc, BGE_RX_MODE, BGE_RXMODE_RX_PROMISC);
5694 
5695 	/* Program multicast filter. */
5696 	mutex_enter(sc->sc_mcast_lock);
5697 	bge_setmulti(sc);
5698 	mutex_exit(sc->sc_mcast_lock);
5699 
5700 	/* Init RX ring. */
5701 	bge_init_rx_ring_std(sc);
5702 
5703 	/*
5704 	 * Workaround for a bug in 5705 ASIC rev A0. Poll the NIC's
5705 	 * memory to insure that the chip has in fact read the first
5706 	 * entry of the ring.
5707 	 */
5708 	if (sc->bge_chipid == BGE_CHIPID_BCM5705_A0) {
5709 		u_int i;
5710 		for (i = 0; i < 10; i++) {
5711 			DELAY(20);
5712 			uint32_t v = bge_readmem_ind(sc, BGE_STD_RX_RINGS + 8);
5713 			if (v == (MCLBYTES - ETHER_ALIGN))
5714 				break;
5715 		}
5716 		if (i == 10)
5717 			aprint_error_dev(sc->bge_dev,
5718 			    "5705 A0 chip failed to load RX ring\n");
5719 	}
5720 
5721 	/* Init jumbo RX ring. */
5722 	if (ifp->if_mtu > (ETHERMTU + ETHER_HDR_LEN + ETHER_CRC_LEN))
5723 		bge_init_rx_ring_jumbo(sc);
5724 
5725 	/* Init our RX return ring index */
5726 	sc->bge_rx_saved_considx = 0;
5727 
5728 	/* Init TX ring. */
5729 	bge_init_tx_ring(sc);
5730 
5731 	/* 5718 step 63, 57XX step 94 */
5732 	/* Enable TX MAC state machine lockup fix. */
5733 	mode = CSR_READ_4(sc, BGE_TX_MODE);
5734 	if (BGE_IS_5755_PLUS(sc) ||
5735 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906)
5736 		mode |= BGE_TXMODE_MBUF_LOCKUP_FIX;
5737 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5720 ||
5738 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5762) {
5739 		mode &= ~(BGE_TXMODE_JMB_FRM_LEN | BGE_TXMODE_CNT_DN_MODE);
5740 		mode |= CSR_READ_4(sc, BGE_TX_MODE) &
5741 		    (BGE_TXMODE_JMB_FRM_LEN | BGE_TXMODE_CNT_DN_MODE);
5742 	}
5743 
5744 	/* Turn on transmitter */
5745 	CSR_WRITE_4_FLUSH(sc, BGE_TX_MODE, mode | BGE_TXMODE_ENABLE);
5746 	/* 5718 step 64 */
5747 	DELAY(100);
5748 
5749 	/* 5718 step 65, 57XX step 95 */
5750 	/* Turn on receiver */
5751 	mode = CSR_READ_4(sc, BGE_RX_MODE);
5752 	if (BGE_IS_5755_PLUS(sc))
5753 		mode |= BGE_RXMODE_IPV6_ENABLE;
5754 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5762)
5755 		mode |= BGE_RXMODE_IPV4_FRAG_FIX;
5756 	CSR_WRITE_4_FLUSH(sc, BGE_RX_MODE, mode | BGE_RXMODE_ENABLE);
5757 	/* 5718 step 66 */
5758 	DELAY(10);
5759 
5760 	/* 5718 step 12, 57XX step 37 */
5761 	/*
5762 	 * XXX Documents of 5718 series and 577xx say the recommended value
5763 	 * is 1, but tg3 set 1 only on 57765 series.
5764 	 */
5765 	if (BGE_IS_57765_PLUS(sc))
5766 		reg = 1;
5767 	else
5768 		reg = 2;
5769 	CSR_WRITE_4_FLUSH(sc, BGE_MAX_RX_FRAME_LOWAT, reg);
5770 
5771 	/* Tell firmware we're alive. */
5772 	BGE_SETBIT(sc, BGE_MODE_CTL, BGE_MODECTL_STACKUP);
5773 
5774 	/* Enable host interrupts. */
5775 	BGE_SETBIT(sc, BGE_PCI_MISC_CTL, BGE_PCIMISCCTL_CLEAR_INTA);
5776 	BGE_CLRBIT(sc, BGE_PCI_MISC_CTL, BGE_PCIMISCCTL_MASK_PCI_INTR);
5777 	bge_writembx_flush(sc, BGE_MBX_IRQ0_LO, 0);
5778 
5779 	mutex_enter(sc->sc_intr_lock);
5780 	if ((error = bge_ifmedia_upd(ifp)) == 0) {
5781 		sc->bge_txrx_stopping = false;
5782 
5783 		/* IFNET_LOCKED asserted above */
5784 		ifp->if_flags |= IFF_RUNNING;
5785 
5786 		callout_schedule(&sc->bge_timeout, hz);
5787 	}
5788 	mutex_exit(sc->sc_intr_lock);
5789 
5790 	mutex_enter(sc->sc_mcast_lock);
5791 	sc->bge_if_flags = ifp->if_flags;
5792 	mutex_exit(sc->sc_mcast_lock);
5793 
5794 	return error;
5795 }
5796 
5797 /*
5798  * Set media options.
5799  */
5800 static int
5801 bge_ifmedia_upd(struct ifnet *ifp)
5802 {
5803 	struct bge_softc * const sc = ifp->if_softc;
5804 	struct mii_data * const mii = &sc->bge_mii;
5805 	struct ifmedia * const ifm = &sc->bge_ifmedia;
5806 	int rc;
5807 
5808 	KASSERT(mutex_owned(sc->sc_intr_lock));
5809 
5810 	/* If this is a 1000baseX NIC, enable the TBI port. */
5811 	if (sc->bge_flags & BGEF_FIBER_TBI) {
5812 		if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER)
5813 			return EINVAL;
5814 		switch (IFM_SUBTYPE(ifm->ifm_media)) {
5815 		case IFM_AUTO:
5816 			/*
5817 			 * The BCM5704 ASIC appears to have a special
5818 			 * mechanism for programming the autoneg
5819 			 * advertisement registers in TBI mode.
5820 			 */
5821 			if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5704) {
5822 				uint32_t sgdig;
5823 				sgdig = CSR_READ_4(sc, BGE_SGDIG_STS);
5824 				if (sgdig & BGE_SGDIGSTS_DONE) {
5825 					CSR_WRITE_4(sc, BGE_TX_TBI_AUTONEG, 0);
5826 					sgdig = CSR_READ_4(sc, BGE_SGDIG_CFG);
5827 					sgdig |= BGE_SGDIGCFG_AUTO |
5828 					    BGE_SGDIGCFG_PAUSE_CAP |
5829 					    BGE_SGDIGCFG_ASYM_PAUSE;
5830 					CSR_WRITE_4_FLUSH(sc, BGE_SGDIG_CFG,
5831 					    sgdig | BGE_SGDIGCFG_SEND);
5832 					DELAY(5);
5833 					CSR_WRITE_4_FLUSH(sc, BGE_SGDIG_CFG,
5834 					    sgdig);
5835 				}
5836 			}
5837 			break;
5838 		case IFM_1000_SX:
5839 			if ((ifm->ifm_media & IFM_FDX) != 0) {
5840 				BGE_CLRBIT_FLUSH(sc, BGE_MAC_MODE,
5841 				    BGE_MACMODE_HALF_DUPLEX);
5842 			} else {
5843 				BGE_SETBIT_FLUSH(sc, BGE_MAC_MODE,
5844 				    BGE_MACMODE_HALF_DUPLEX);
5845 			}
5846 			DELAY(40);
5847 			break;
5848 		default:
5849 			return EINVAL;
5850 		}
5851 		/* XXX 802.3x flow control for 1000BASE-SX */
5852 		return 0;
5853 	}
5854 
5855 	if ((BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5784) &&
5856 	    (BGE_CHIPREV(sc->bge_chipid) != BGE_CHIPREV_5784_AX)) {
5857 		uint32_t reg;
5858 
5859 		reg = CSR_READ_4(sc, BGE_CPMU_CTRL);
5860 		if ((reg & BGE_CPMU_CTRL_GPHY_10MB_RXONLY) != 0) {
5861 			reg &= ~BGE_CPMU_CTRL_GPHY_10MB_RXONLY;
5862 			CSR_WRITE_4(sc, BGE_CPMU_CTRL, reg);
5863 		}
5864 	}
5865 
5866 	BGE_STS_SETBIT(sc, BGE_STS_LINK_EVT);
5867 	if ((rc = mii_mediachg(mii)) == ENXIO)
5868 		return 0;
5869 
5870 	if (BGE_CHIPREV(sc->bge_chipid) == BGE_CHIPREV_5784_AX) {
5871 		uint32_t reg;
5872 
5873 		reg = CSR_READ_4(sc, BGE_CPMU_LSPD_1000MB_CLK);
5874 		if ((reg & BGE_CPMU_LSPD_1000MB_MACCLK_MASK)
5875 		    == (BGE_CPMU_LSPD_1000MB_MACCLK_12_5)) {
5876 			reg &= ~BGE_CPMU_LSPD_1000MB_MACCLK_MASK;
5877 			delay(40);
5878 			CSR_WRITE_4(sc, BGE_CPMU_LSPD_1000MB_CLK, reg);
5879 		}
5880 	}
5881 
5882 	/*
5883 	 * Force an interrupt so that we will call bge_link_upd
5884 	 * if needed and clear any pending link state attention.
5885 	 * Without this we are not getting any further interrupts
5886 	 * for link state changes and thus will not UP the link and
5887 	 * not be able to send in bge_start. The only way to get
5888 	 * things working was to receive a packet and get a RX intr.
5889 	 */
5890 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5700 ||
5891 	    sc->bge_flags & BGEF_IS_5788)
5892 		BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_INTR_SET);
5893 	else
5894 		BGE_SETBIT(sc, BGE_HCC_MODE, BGE_HCCMODE_COAL_NOW);
5895 
5896 	return rc;
5897 }
5898 
5899 /*
5900  * Report current media status.
5901  */
5902 static void
5903 bge_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
5904 {
5905 	struct bge_softc * const sc = ifp->if_softc;
5906 	struct mii_data * const mii = &sc->bge_mii;
5907 
5908 	KASSERT(mutex_owned(sc->sc_intr_lock));
5909 
5910 	if (sc->bge_flags & BGEF_FIBER_TBI) {
5911 		ifmr->ifm_status = IFM_AVALID;
5912 		ifmr->ifm_active = IFM_ETHER;
5913 		if (CSR_READ_4(sc, BGE_MAC_STS) &
5914 		    BGE_MACSTAT_TBI_PCS_SYNCHED)
5915 			ifmr->ifm_status |= IFM_ACTIVE;
5916 		ifmr->ifm_active |= IFM_1000_SX;
5917 		if (CSR_READ_4(sc, BGE_MAC_MODE) & BGE_MACMODE_HALF_DUPLEX)
5918 			ifmr->ifm_active |= IFM_HDX;
5919 		else
5920 			ifmr->ifm_active |= IFM_FDX;
5921 		return;
5922 	}
5923 
5924 	mii_pollstat(mii);
5925 	ifmr->ifm_status = mii->mii_media_status;
5926 	ifmr->ifm_active = (mii->mii_media_active & ~IFM_ETH_FMASK) |
5927 	    sc->bge_flowflags;
5928 }
5929 
5930 static int
5931 bge_ifflags_cb(struct ethercom *ec)
5932 {
5933 	struct ifnet * const ifp = &ec->ec_if;
5934 	struct bge_softc * const sc = ifp->if_softc;
5935 	int ret = 0;
5936 
5937 	KASSERT(IFNET_LOCKED(ifp));
5938 	mutex_enter(sc->sc_mcast_lock);
5939 
5940 	u_short change = ifp->if_flags ^ sc->bge_if_flags;
5941 	sc->bge_if_flags = ifp->if_flags;
5942 
5943 	if ((change & ~(IFF_CANTCHANGE | IFF_DEBUG)) != 0) {
5944 		ret = ENETRESET;
5945 	} else if ((change & (IFF_PROMISC | IFF_ALLMULTI)) != 0) {
5946 		if ((ifp->if_flags & IFF_PROMISC) == 0)
5947 			BGE_CLRBIT(sc, BGE_RX_MODE, BGE_RXMODE_RX_PROMISC);
5948 		else
5949 			BGE_SETBIT(sc, BGE_RX_MODE, BGE_RXMODE_RX_PROMISC);
5950 
5951 		bge_setmulti(sc);
5952 	}
5953 
5954 	mutex_exit(sc->sc_mcast_lock);
5955 
5956 	return ret;
5957 }
5958 
5959 static int
5960 bge_ioctl(struct ifnet *ifp, u_long command, void *data)
5961 {
5962 	struct bge_softc * const sc = ifp->if_softc;
5963 	struct ifreq * const ifr = (struct ifreq *) data;
5964 	int error = 0;
5965 
5966 	switch (command) {
5967 	case SIOCADDMULTI:
5968 	case SIOCDELMULTI:
5969 		break;
5970 	default:
5971 		KASSERT(IFNET_LOCKED(ifp));
5972 	}
5973 
5974 	const int s = splnet();
5975 
5976 	switch (command) {
5977 	case SIOCSIFMEDIA:
5978 		mutex_enter(sc->sc_intr_lock);
5979 		/* XXX Flow control is not supported for 1000BASE-SX */
5980 		if (sc->bge_flags & BGEF_FIBER_TBI) {
5981 			ifr->ifr_media &= ~IFM_ETH_FMASK;
5982 			sc->bge_flowflags = 0;
5983 		}
5984 
5985 		/* Flow control requires full-duplex mode. */
5986 		if (IFM_SUBTYPE(ifr->ifr_media) == IFM_AUTO ||
5987 		    (ifr->ifr_media & IFM_FDX) == 0) {
5988 			ifr->ifr_media &= ~IFM_ETH_FMASK;
5989 		}
5990 		if (IFM_SUBTYPE(ifr->ifr_media) != IFM_AUTO) {
5991 			if ((ifr->ifr_media & IFM_ETH_FMASK) == IFM_FLOW) {
5992 				/* We can do both TXPAUSE and RXPAUSE. */
5993 				ifr->ifr_media |=
5994 				    IFM_ETH_TXPAUSE | IFM_ETH_RXPAUSE;
5995 			}
5996 			sc->bge_flowflags = ifr->ifr_media & IFM_ETH_FMASK;
5997 		}
5998 		mutex_exit(sc->sc_intr_lock);
5999 
6000 		if (sc->bge_flags & BGEF_FIBER_TBI) {
6001 			error = ifmedia_ioctl(ifp, ifr, &sc->bge_ifmedia,
6002 			    command);
6003 		} else {
6004 			struct mii_data * const mii = &sc->bge_mii;
6005 			error = ifmedia_ioctl(ifp, ifr, &mii->mii_media,
6006 			    command);
6007 		}
6008 		break;
6009 	default:
6010 		if ((error = ether_ioctl(ifp, command, data)) != ENETRESET)
6011 			break;
6012 
6013 		error = 0;
6014 
6015 		if (command == SIOCADDMULTI || command == SIOCDELMULTI) {
6016 			mutex_enter(sc->sc_mcast_lock);
6017 			if (sc->bge_if_flags & IFF_RUNNING) {
6018 				bge_setmulti(sc);
6019 			}
6020 			mutex_exit(sc->sc_mcast_lock);
6021 		}
6022 		break;
6023 	}
6024 
6025 	splx(s);
6026 
6027 	return error;
6028 }
6029 
6030 static bool
6031 bge_watchdog_check(struct bge_softc * const sc)
6032 {
6033 
6034 	KASSERT(mutex_owned(sc->sc_intr_lock));
6035 
6036 	if (!sc->bge_tx_sending)
6037 		return true;
6038 
6039 	if (time_uptime - sc->bge_tx_lastsent <= bge_watchdog_timeout)
6040 		return true;
6041 
6042 	/* If pause frames are active then don't reset the hardware. */
6043 	if ((CSR_READ_4(sc, BGE_RX_MODE) & BGE_RXMODE_FLOWCTL_ENABLE) != 0) {
6044 		const uint32_t status = CSR_READ_4(sc, BGE_RX_STS);
6045 		if ((status & BGE_RXSTAT_REMOTE_XOFFED) != 0) {
6046 			/*
6047 			 * If link partner has us in XOFF state then wait for
6048 			 * the condition to clear.
6049 			 */
6050 			CSR_WRITE_4(sc, BGE_RX_STS, status);
6051 			sc->bge_tx_lastsent = time_uptime;
6052 			return true;
6053 		} else if ((status & BGE_RXSTAT_RCVD_XOFF) != 0 &&
6054 		    (status & BGE_RXSTAT_RCVD_XON) != 0) {
6055 			/*
6056 			 * If link partner has us in XOFF state then wait for
6057 			 * the condition to clear.
6058 			 */
6059 			CSR_WRITE_4(sc, BGE_RX_STS, status);
6060 			sc->bge_tx_lastsent = time_uptime;
6061 			return true;
6062 		}
6063 		/*
6064 		 * Any other condition is unexpected and the controller
6065 		 * should be reset.
6066 		 */
6067 	}
6068 
6069 	return false;
6070 }
6071 
6072 static bool
6073 bge_watchdog_tick(struct ifnet *ifp)
6074 {
6075 	struct bge_softc * const sc = ifp->if_softc;
6076 
6077 	KASSERT(mutex_owned(sc->sc_intr_lock));
6078 
6079 	if (!sc->sc_trigger_reset && bge_watchdog_check(sc))
6080 		return true;
6081 
6082 	if (atomic_swap_uint(&sc->sc_reset_pending, 1) == 0)
6083 		workqueue_enqueue(sc->sc_reset_wq, &sc->sc_reset_work, NULL);
6084 
6085 	return false;
6086 }
6087 
6088 /*
6089  * Perform an interface watchdog reset.
6090  */
6091 static void
6092 bge_handle_reset_work(struct work *work, void *arg)
6093 {
6094 	struct bge_softc * const sc = arg;
6095 	struct ifnet * const ifp = &sc->ethercom.ec_if;
6096 
6097 	printf("%s: watchdog timeout -- resetting\n", ifp->if_xname);
6098 
6099 	/* Don't want ioctl operations to happen */
6100 	IFNET_LOCK(ifp);
6101 
6102 	/* reset the interface. */
6103 	bge_init(ifp);
6104 
6105 	IFNET_UNLOCK(ifp);
6106 
6107 	/*
6108 	 * There are still some upper layer processing which call
6109 	 * ifp->if_start(). e.g. ALTQ or one CPU system
6110 	 */
6111 	/* Try to get more packets going. */
6112 	ifp->if_start(ifp);
6113 
6114 	atomic_store_relaxed(&sc->sc_reset_pending, 0);
6115 }
6116 
6117 static void
6118 bge_stop_block(struct bge_softc *sc, bus_addr_t reg, uint32_t bit)
6119 {
6120 	int i;
6121 
6122 	BGE_CLRBIT_FLUSH(sc, reg, bit);
6123 
6124 	for (i = 0; i < 1000; i++) {
6125 		delay(100);
6126 		if ((CSR_READ_4(sc, reg) & bit) == 0)
6127 			return;
6128 	}
6129 
6130 	/*
6131 	 * Doesn't print only when the register is BGE_SRS_MODE. It occurs
6132 	 * on some environment (and once after boot?)
6133 	 */
6134 	if (reg != BGE_SRS_MODE)
6135 		aprint_error_dev(sc->bge_dev,
6136 		    "block failed to stop: reg 0x%lx, bit 0x%08x\n",
6137 		    (u_long)reg, bit);
6138 }
6139 
6140 /*
6141  * Stop the adapter and free any mbufs allocated to the
6142  * RX and TX lists.
6143  */
6144 static void
6145 bge_stop(struct ifnet *ifp, int disable)
6146 {
6147 	struct bge_softc * const sc = ifp->if_softc;
6148 
6149 	ASSERT_SLEEPABLE();
6150 	KASSERT(IFNET_LOCKED(ifp));
6151 
6152 	mutex_enter(sc->sc_intr_lock);
6153 	sc->bge_txrx_stopping = true;
6154 	mutex_exit(sc->sc_intr_lock);
6155 
6156 	callout_halt(&sc->bge_timeout, NULL);
6157 
6158 	/* Disable host interrupts. */
6159 	BGE_SETBIT(sc, BGE_PCI_MISC_CTL, BGE_PCIMISCCTL_MASK_PCI_INTR);
6160 	bge_writembx_flush(sc, BGE_MBX_IRQ0_LO, 1);
6161 
6162 	/*
6163 	 * Tell firmware we're shutting down.
6164 	 */
6165 	bge_stop_fw(sc);
6166 	bge_sig_pre_reset(sc, BGE_RESET_SHUTDOWN);
6167 
6168 	/*
6169 	 * Disable all of the receiver blocks.
6170 	 */
6171 	bge_stop_block(sc, BGE_RX_MODE, BGE_RXMODE_ENABLE);
6172 	bge_stop_block(sc, BGE_RBDI_MODE, BGE_RBDIMODE_ENABLE);
6173 	bge_stop_block(sc, BGE_RXLP_MODE, BGE_RXLPMODE_ENABLE);
6174 	if (BGE_IS_5700_FAMILY(sc))
6175 		bge_stop_block(sc, BGE_RXLS_MODE, BGE_RXLSMODE_ENABLE);
6176 	bge_stop_block(sc, BGE_RDBDI_MODE, BGE_RBDIMODE_ENABLE);
6177 	bge_stop_block(sc, BGE_RDC_MODE, BGE_RDCMODE_ENABLE);
6178 	bge_stop_block(sc, BGE_RBDC_MODE, BGE_RBDCMODE_ENABLE);
6179 
6180 	/*
6181 	 * Disable all of the transmit blocks.
6182 	 */
6183 	bge_stop_block(sc, BGE_SRS_MODE, BGE_SRSMODE_ENABLE);
6184 	bge_stop_block(sc, BGE_SBDI_MODE, BGE_SBDIMODE_ENABLE);
6185 	bge_stop_block(sc, BGE_SDI_MODE, BGE_SDIMODE_ENABLE);
6186 	bge_stop_block(sc, BGE_RDMA_MODE, BGE_RDMAMODE_ENABLE);
6187 	bge_stop_block(sc, BGE_SDC_MODE, BGE_SDCMODE_ENABLE);
6188 	if (BGE_IS_5700_FAMILY(sc))
6189 		bge_stop_block(sc, BGE_DMAC_MODE, BGE_DMACMODE_ENABLE);
6190 	bge_stop_block(sc, BGE_SBDC_MODE, BGE_SBDCMODE_ENABLE);
6191 
6192 	BGE_CLRBIT_FLUSH(sc, BGE_MAC_MODE, BGE_MACMODE_TXDMA_ENB);
6193 	delay(40);
6194 
6195 	bge_stop_block(sc, BGE_TX_MODE, BGE_TXMODE_ENABLE);
6196 
6197 	/*
6198 	 * Shut down all of the memory managers and related
6199 	 * state machines.
6200 	 */
6201 	/* 5718 step 5a,5b */
6202 	bge_stop_block(sc, BGE_HCC_MODE, BGE_HCCMODE_ENABLE);
6203 	bge_stop_block(sc, BGE_WDMA_MODE, BGE_WDMAMODE_ENABLE);
6204 	if (BGE_IS_5700_FAMILY(sc))
6205 		bge_stop_block(sc, BGE_MBCF_MODE, BGE_MBCFMODE_ENABLE);
6206 
6207 	/* 5718 step 5c,5d */
6208 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0xFFFFFFFF);
6209 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0);
6210 
6211 	if (BGE_IS_5700_FAMILY(sc)) {
6212 		bge_stop_block(sc, BGE_BMAN_MODE, BGE_BMANMODE_ENABLE);
6213 		bge_stop_block(sc, BGE_MARB_MODE, BGE_MARBMODE_ENABLE);
6214 	}
6215 
6216 	bge_reset(sc);
6217 	bge_sig_legacy(sc, BGE_RESET_SHUTDOWN);
6218 	bge_sig_post_reset(sc, BGE_RESET_SHUTDOWN);
6219 
6220 	/*
6221 	 * Keep the ASF firmware running if up.
6222 	 */
6223 	if (sc->bge_asf_mode & ASF_STACKUP)
6224 		BGE_SETBIT(sc, BGE_MODE_CTL, BGE_MODECTL_STACKUP);
6225 	else
6226 		BGE_CLRBIT(sc, BGE_MODE_CTL, BGE_MODECTL_STACKUP);
6227 
6228 	/* Free the RX lists. */
6229 	bge_free_rx_ring_std(sc);
6230 
6231 	/* Free jumbo RX list. */
6232 	if (BGE_IS_JUMBO_CAPABLE(sc))
6233 		bge_free_rx_ring_jumbo(sc);
6234 
6235 	/* Free TX buffers. */
6236 	bge_free_tx_ring(sc, disable);
6237 
6238 	/*
6239 	 * Isolate/power down the PHY.
6240 	 */
6241 	if (!(sc->bge_flags & BGEF_FIBER_TBI)) {
6242 		mutex_enter(sc->sc_intr_lock);
6243 		mii_down(&sc->bge_mii);
6244 		mutex_exit(sc->sc_intr_lock);
6245 	}
6246 
6247 	sc->bge_tx_saved_considx = BGE_TXCONS_UNSET;
6248 
6249 	/* Clear MAC's link state (PHY may still have link UP). */
6250 	BGE_STS_CLRBIT(sc, BGE_STS_LINK);
6251 
6252 	ifp->if_flags &= ~IFF_RUNNING;
6253 
6254 	mutex_enter(sc->sc_mcast_lock);
6255 	sc->bge_if_flags = ifp->if_flags;
6256 	mutex_exit(sc->sc_mcast_lock);
6257 }
6258 
6259 static void
6260 bge_link_upd(struct bge_softc *sc)
6261 {
6262 	struct ifnet * const ifp = &sc->ethercom.ec_if;
6263 	struct mii_data * const mii = &sc->bge_mii;
6264 	uint32_t status;
6265 	uint16_t phyval;
6266 	int link;
6267 
6268 	KASSERT(sc->sc_intr_lock);
6269 
6270 	/* Clear 'pending link event' flag */
6271 	BGE_STS_CLRBIT(sc, BGE_STS_LINK_EVT);
6272 
6273 	/*
6274 	 * Process link state changes.
6275 	 * Grrr. The link status word in the status block does
6276 	 * not work correctly on the BCM5700 rev AX and BX chips,
6277 	 * according to all available information. Hence, we have
6278 	 * to enable MII interrupts in order to properly obtain
6279 	 * async link changes. Unfortunately, this also means that
6280 	 * we have to read the MAC status register to detect link
6281 	 * changes, thereby adding an additional register access to
6282 	 * the interrupt handler.
6283 	 */
6284 
6285 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5700) {
6286 		status = CSR_READ_4(sc, BGE_MAC_STS);
6287 		if (status & BGE_MACSTAT_MI_INTERRUPT) {
6288 			mii_pollstat(mii);
6289 
6290 			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_SETBIT(sc, BGE_STS_LINK);
6294 			else if (BGE_STS_BIT(sc, BGE_STS_LINK) &&
6295 			    (!(mii->mii_media_status & IFM_ACTIVE) ||
6296 			    IFM_SUBTYPE(mii->mii_media_active) == IFM_NONE))
6297 				BGE_STS_CLRBIT(sc, BGE_STS_LINK);
6298 
6299 			/* Clear the interrupt */
6300 			CSR_WRITE_4(sc, BGE_MAC_EVT_ENB,
6301 			    BGE_EVTENB_MI_INTERRUPT);
6302 			bge_miibus_readreg(sc->bge_dev, sc->bge_phy_addr,
6303 			    BRGPHY_MII_ISR, &phyval);
6304 			bge_miibus_writereg(sc->bge_dev, sc->bge_phy_addr,
6305 			    BRGPHY_MII_IMR, BRGPHY_INTRS);
6306 		}
6307 		return;
6308 	}
6309 
6310 	if (sc->bge_flags & BGEF_FIBER_TBI) {
6311 		status = CSR_READ_4(sc, BGE_MAC_STS);
6312 		if (status & BGE_MACSTAT_TBI_PCS_SYNCHED) {
6313 			if (!BGE_STS_BIT(sc, BGE_STS_LINK)) {
6314 				BGE_STS_SETBIT(sc, BGE_STS_LINK);
6315 				if (BGE_ASICREV(sc->bge_chipid)
6316 				    == BGE_ASICREV_BCM5704) {
6317 					BGE_CLRBIT_FLUSH(sc, BGE_MAC_MODE,
6318 					    BGE_MACMODE_TBI_SEND_CFGS);
6319 					DELAY(40);
6320 				}
6321 				CSR_WRITE_4(sc, BGE_MAC_STS, 0xFFFFFFFF);
6322 				if_link_state_change(ifp, LINK_STATE_UP);
6323 			}
6324 		} else if (BGE_STS_BIT(sc, BGE_STS_LINK)) {
6325 			BGE_STS_CLRBIT(sc, BGE_STS_LINK);
6326 			if_link_state_change(ifp, LINK_STATE_DOWN);
6327 		}
6328 	} else if (BGE_STS_BIT(sc, BGE_STS_AUTOPOLL)) {
6329 		/*
6330 		 * Some broken BCM chips have BGE_STATFLAG_LINKSTATE_CHANGED
6331 		 * bit in status word always set. Workaround this bug by
6332 		 * reading PHY link status directly.
6333 		 */
6334 		link = (CSR_READ_4(sc, BGE_MI_STS) & BGE_MISTS_LINK)?
6335 		    BGE_STS_LINK : 0;
6336 
6337 		if (BGE_STS_BIT(sc, BGE_STS_LINK) != link) {
6338 			mii_pollstat(mii);
6339 
6340 			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_SETBIT(sc, BGE_STS_LINK);
6344 			else if (BGE_STS_BIT(sc, BGE_STS_LINK) &&
6345 			    (!(mii->mii_media_status & IFM_ACTIVE) ||
6346 			    IFM_SUBTYPE(mii->mii_media_active) == IFM_NONE))
6347 				BGE_STS_CLRBIT(sc, BGE_STS_LINK);
6348 		}
6349 	} else {
6350 		/*
6351 		 * For controllers that call mii_tick, we have to poll
6352 		 * link status.
6353 		 */
6354 		mii_pollstat(mii);
6355 	}
6356 
6357 	if (BGE_CHIPREV(sc->bge_chipid) == BGE_CHIPREV_5784_AX) {
6358 		uint32_t reg, scale;
6359 
6360 		reg = CSR_READ_4(sc, BGE_CPMU_CLCK_STAT) &
6361 		    BGE_CPMU_CLCK_STAT_MAC_CLCK_MASK;
6362 		if (reg == BGE_CPMU_CLCK_STAT_MAC_CLCK_62_5)
6363 			scale = 65;
6364 		else if (reg == BGE_CPMU_CLCK_STAT_MAC_CLCK_6_25)
6365 			scale = 6;
6366 		else
6367 			scale = 12;
6368 
6369 		reg = CSR_READ_4(sc, BGE_MISC_CFG) &
6370 		    ~BGE_MISCCFG_TIMER_PRESCALER;
6371 		reg |= scale << 1;
6372 		CSR_WRITE_4(sc, BGE_MISC_CFG, reg);
6373 	}
6374 	/* Clear the attention */
6375 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED |
6376 	    BGE_MACSTAT_CFG_CHANGED | BGE_MACSTAT_MI_COMPLETE |
6377 	    BGE_MACSTAT_LINK_CHANGED);
6378 }
6379 
6380 static int
6381 bge_sysctl_verify(SYSCTLFN_ARGS)
6382 {
6383 	int error, t;
6384 	struct sysctlnode node;
6385 
6386 	node = *rnode;
6387 	t = *(int*)rnode->sysctl_data;
6388 	node.sysctl_data = &t;
6389 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
6390 	if (error || newp == NULL)
6391 		return error;
6392 
6393 #if 0
6394 	DPRINTF2(("%s: t = %d, nodenum = %d, rnodenum = %d\n", __func__, t,
6395 	    node.sysctl_num, rnode->sysctl_num));
6396 #endif
6397 
6398 	if (node.sysctl_num == bge_rxthresh_nodenum) {
6399 		if (t < 0 || t >= NBGE_RX_THRESH)
6400 			return EINVAL;
6401 		bge_update_all_threshes(t);
6402 	} else
6403 		return EINVAL;
6404 
6405 	*(int*)rnode->sysctl_data = t;
6406 
6407 	return 0;
6408 }
6409 
6410 /*
6411  * Set up sysctl(3) MIB, hw.bge.*.
6412  */
6413 static void
6414 bge_sysctl_init(struct bge_softc *sc)
6415 {
6416 	int rc, bge_root_num;
6417 	const struct sysctlnode *node;
6418 
6419 	if ((rc = sysctl_createv(&sc->bge_log, 0, NULL, &node,
6420 	    0, CTLTYPE_NODE, "bge",
6421 	    SYSCTL_DESCR("BGE interface controls"),
6422 	    NULL, 0, NULL, 0, CTL_HW, CTL_CREATE, CTL_EOL)) != 0) {
6423 		goto out;
6424 	}
6425 
6426 	bge_root_num = node->sysctl_num;
6427 
6428 	/* BGE Rx interrupt mitigation level */
6429 	if ((rc = sysctl_createv(&sc->bge_log, 0, NULL, &node,
6430 	    CTLFLAG_READWRITE,
6431 	    CTLTYPE_INT, "rx_lvl",
6432 	    SYSCTL_DESCR("BGE receive interrupt mitigation level"),
6433 	    bge_sysctl_verify, 0,
6434 	    &bge_rx_thresh_lvl,
6435 	    0, CTL_HW, bge_root_num, CTL_CREATE,
6436 	    CTL_EOL)) != 0) {
6437 		goto out;
6438 	}
6439 
6440 	bge_rxthresh_nodenum = node->sysctl_num;
6441 
6442 #ifdef BGE_DEBUG
6443 	if ((rc = sysctl_createv(&sc->bge_log, 0, NULL, &node,
6444 	    CTLFLAG_READWRITE,
6445 	    CTLTYPE_BOOL, "trigger_reset",
6446 	    SYSCTL_DESCR("Trigger an interface reset"),
6447 	    NULL, 0, &sc->sc_trigger_reset, 0, CTL_CREATE,
6448 	    CTL_EOL)) != 0) {
6449 		goto out;
6450 	}
6451 #endif
6452 	return;
6453 
6454 out:
6455 	aprint_error("%s: sysctl_createv failed (rc = %d)\n", __func__, rc);
6456 }
6457 
6458 #ifdef BGE_DEBUG
6459 void
6460 bge_debug_info(struct bge_softc *sc)
6461 {
6462 
6463 	printf("Hardware Flags:\n");
6464 	if (BGE_IS_57765_PLUS(sc))
6465 		printf(" - 57765 Plus\n");
6466 	if (BGE_IS_5717_PLUS(sc))
6467 		printf(" - 5717 Plus\n");
6468 	if (BGE_IS_5755_PLUS(sc))
6469 		printf(" - 5755 Plus\n");
6470 	if (BGE_IS_575X_PLUS(sc))
6471 		printf(" - 575X Plus\n");
6472 	if (BGE_IS_5705_PLUS(sc))
6473 		printf(" - 5705 Plus\n");
6474 	if (BGE_IS_5714_FAMILY(sc))
6475 		printf(" - 5714 Family\n");
6476 	if (BGE_IS_5700_FAMILY(sc))
6477 		printf(" - 5700 Family\n");
6478 	if (sc->bge_flags & BGEF_IS_5788)
6479 		printf(" - 5788\n");
6480 	if (sc->bge_flags & BGEF_JUMBO_CAPABLE)
6481 		printf(" - Supports Jumbo Frames\n");
6482 	if (sc->bge_flags & BGEF_NO_EEPROM)
6483 		printf(" - No EEPROM\n");
6484 	if (sc->bge_flags & BGEF_PCIX)
6485 		printf(" - PCI-X Bus\n");
6486 	if (sc->bge_flags & BGEF_PCIE)
6487 		printf(" - PCI Express Bus\n");
6488 	if (sc->bge_flags & BGEF_RX_ALIGNBUG)
6489 		printf(" - RX Alignment Bug\n");
6490 	if (sc->bge_flags & BGEF_APE)
6491 		printf(" - APE\n");
6492 	if (sc->bge_flags & BGEF_CPMU_PRESENT)
6493 		printf(" - CPMU\n");
6494 	if (sc->bge_flags & BGEF_TSO)
6495 		printf(" - TSO\n");
6496 	if (sc->bge_flags & BGEF_TAGGED_STATUS)
6497 		printf(" - TAGGED_STATUS\n");
6498 
6499 	/* PHY related */
6500 	if (sc->bge_phy_flags & BGEPHYF_NO_3LED)
6501 		printf(" - No 3 LEDs\n");
6502 	if (sc->bge_phy_flags & BGEPHYF_CRC_BUG)
6503 		printf(" - CRC bug\n");
6504 	if (sc->bge_phy_flags & BGEPHYF_ADC_BUG)
6505 		printf(" - ADC bug\n");
6506 	if (sc->bge_phy_flags & BGEPHYF_5704_A0_BUG)
6507 		printf(" - 5704 A0 bug\n");
6508 	if (sc->bge_phy_flags & BGEPHYF_JITTER_BUG)
6509 		printf(" - jitter bug\n");
6510 	if (sc->bge_phy_flags & BGEPHYF_BER_BUG)
6511 		printf(" - BER bug\n");
6512 	if (sc->bge_phy_flags & BGEPHYF_ADJUST_TRIM)
6513 		printf(" - adjust trim\n");
6514 	if (sc->bge_phy_flags & BGEPHYF_NO_WIRESPEED)
6515 		printf(" - no wirespeed\n");
6516 
6517 	/* ASF related */
6518 	if (sc->bge_asf_mode & ASF_ENABLE)
6519 		printf(" - ASF enable\n");
6520 	if (sc->bge_asf_mode & ASF_NEW_HANDSHAKE)
6521 		printf(" - ASF new handshake\n");
6522 	if (sc->bge_asf_mode & ASF_STACKUP)
6523 		printf(" - ASF stackup\n");
6524 }
6525 #endif /* BGE_DEBUG */
6526 
6527 static int
6528 bge_get_eaddr_fw(struct bge_softc *sc, uint8_t ether_addr[])
6529 {
6530 	prop_dictionary_t dict;
6531 	prop_data_t ea;
6532 
6533 	if ((sc->bge_flags & BGEF_NO_EEPROM) == 0)
6534 		return 1;
6535 
6536 	dict = device_properties(sc->bge_dev);
6537 	ea = prop_dictionary_get(dict, "mac-address");
6538 	if (ea != NULL) {
6539 		KASSERT(prop_object_type(ea) == PROP_TYPE_DATA);
6540 		KASSERT(prop_data_size(ea) == ETHER_ADDR_LEN);
6541 		memcpy(ether_addr, prop_data_value(ea), ETHER_ADDR_LEN);
6542 		return 0;
6543 	}
6544 
6545 	return 1;
6546 }
6547 
6548 static int
6549 bge_get_eaddr_mem(struct bge_softc *sc, uint8_t ether_addr[])
6550 {
6551 	uint32_t mac_addr;
6552 
6553 	mac_addr = bge_readmem_ind(sc, BGE_SRAM_MAC_ADDR_HIGH_MB);
6554 	if ((mac_addr >> 16) == 0x484b) {
6555 		ether_addr[0] = (uint8_t)(mac_addr >> 8);
6556 		ether_addr[1] = (uint8_t)mac_addr;
6557 		mac_addr = bge_readmem_ind(sc, BGE_SRAM_MAC_ADDR_LOW_MB);
6558 		ether_addr[2] = (uint8_t)(mac_addr >> 24);
6559 		ether_addr[3] = (uint8_t)(mac_addr >> 16);
6560 		ether_addr[4] = (uint8_t)(mac_addr >> 8);
6561 		ether_addr[5] = (uint8_t)mac_addr;
6562 		return 0;
6563 	}
6564 	return 1;
6565 }
6566 
6567 static int
6568 bge_get_eaddr_nvram(struct bge_softc *sc, uint8_t ether_addr[])
6569 {
6570 	int mac_offset = BGE_EE_MAC_OFFSET;
6571 
6572 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906)
6573 		mac_offset = BGE_EE_MAC_OFFSET_5906;
6574 
6575 	return (bge_read_nvram(sc, ether_addr, mac_offset + 2,
6576 	    ETHER_ADDR_LEN));
6577 }
6578 
6579 static int
6580 bge_get_eaddr_eeprom(struct bge_softc *sc, uint8_t ether_addr[])
6581 {
6582 
6583 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906)
6584 		return 1;
6585 
6586 	return (bge_read_eeprom(sc, ether_addr, BGE_EE_MAC_OFFSET + 2,
6587 	   ETHER_ADDR_LEN));
6588 }
6589 
6590 static int
6591 bge_get_eaddr(struct bge_softc *sc, uint8_t eaddr[])
6592 {
6593 	static const bge_eaddr_fcn_t bge_eaddr_funcs[] = {
6594 		/* NOTE: Order is critical */
6595 		bge_get_eaddr_fw,
6596 		bge_get_eaddr_mem,
6597 		bge_get_eaddr_nvram,
6598 		bge_get_eaddr_eeprom,
6599 		NULL
6600 	};
6601 	const bge_eaddr_fcn_t *func;
6602 
6603 	for (func = bge_eaddr_funcs; *func != NULL; ++func) {
6604 		if ((*func)(sc, eaddr) == 0)
6605 			break;
6606 	}
6607 	return *func == NULL ? ENXIO : 0;
6608 }
6609