xref: /netbsd-src/sys/dev/pci/if_bge.c (revision 8b0f9554ff8762542c4defc4f70e1eb76fb508fa)
1 /*	$NetBSD: if_bge.c,v 1.141 2007/12/09 20:28:08 jmcneill 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 refered 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 1.141 2007/12/09 20:28:08 jmcneill Exp $");
83 
84 #include "bpfilter.h"
85 #include "vlan.h"
86 
87 #include <sys/param.h>
88 #include <sys/systm.h>
89 #include <sys/callout.h>
90 #include <sys/sockio.h>
91 #include <sys/mbuf.h>
92 #include <sys/malloc.h>
93 #include <sys/kernel.h>
94 #include <sys/device.h>
95 #include <sys/socket.h>
96 #include <sys/sysctl.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 
103 #ifdef INET
104 #include <netinet/in.h>
105 #include <netinet/in_systm.h>
106 #include <netinet/in_var.h>
107 #include <netinet/ip.h>
108 #endif
109 
110 /* Headers for TCP  Segmentation Offload (TSO) */
111 #include <netinet/in_systm.h>		/* n_time for <netinet/ip.h>... */
112 #include <netinet/in.h>			/* ip_{src,dst}, for <netinet/ip.h> */
113 #include <netinet/ip.h>			/* for struct ip */
114 #include <netinet/tcp.h>		/* for struct tcphdr */
115 
116 
117 #if NBPFILTER > 0
118 #include <net/bpf.h>
119 #endif
120 
121 #include <dev/pci/pcireg.h>
122 #include <dev/pci/pcivar.h>
123 #include <dev/pci/pcidevs.h>
124 
125 #include <dev/mii/mii.h>
126 #include <dev/mii/miivar.h>
127 #include <dev/mii/miidevs.h>
128 #include <dev/mii/brgphyreg.h>
129 
130 #include <dev/pci/if_bgereg.h>
131 
132 #include <uvm/uvm_extern.h>
133 
134 #define ETHER_MIN_NOPAD (ETHER_MIN_LEN - ETHER_CRC_LEN) /* i.e., 60 */
135 
136 
137 /*
138  * Tunable thresholds for rx-side bge interrupt mitigation.
139  */
140 
141 /*
142  * The pairs of values below were obtained from empirical measurement
143  * on bcm5700 rev B2; they ar designed to give roughly 1 receive
144  * interrupt for every N packets received, where N is, approximately,
145  * the second value (rx_max_bds) in each pair.  The values are chosen
146  * such that moving from one pair to the succeeding pair was observed
147  * to roughly halve interrupt rate under sustained input packet load.
148  * The values were empirically chosen to avoid overflowing internal
149  * limits on the  bcm5700: inreasing rx_ticks much beyond 600
150  * results in internal wrapping and higher interrupt rates.
151  * The limit of 46 frames was chosen to match NFS workloads.
152  *
153  * These values also work well on bcm5701, bcm5704C, and (less
154  * tested) bcm5703.  On other chipsets, (including the Altima chip
155  * family), the larger values may overflow internal chip limits,
156  * leading to increasing interrupt rates rather than lower interrupt
157  * rates.
158  *
159  * Applications using heavy interrupt mitigation (interrupting every
160  * 32 or 46 frames) in both directions may need to increase the TCP
161  * windowsize to above 131072 bytes (e.g., to 199608 bytes) to sustain
162  * full link bandwidth, due to ACKs and window updates lingering
163  * in the RX queue during the 30-to-40-frame interrupt-mitigation window.
164  */
165 static const struct bge_load_rx_thresh {
166 	int rx_ticks;
167 	int rx_max_bds; }
168 bge_rx_threshes[] = {
169 	{ 32,   2 },
170 	{ 50,   4 },
171 	{ 100,  8 },
172 	{ 192, 16 },
173 	{ 416, 32 },
174 	{ 598, 46 }
175 };
176 #define NBGE_RX_THRESH (sizeof(bge_rx_threshes) / sizeof(bge_rx_threshes[0]))
177 
178 /* XXX patchable; should be sysctl'able */
179 static int	bge_auto_thresh = 1;
180 static int	bge_rx_thresh_lvl;
181 
182 static int	bge_rxthresh_nodenum;
183 
184 static int	bge_probe(device_t, cfdata_t, void *);
185 static void	bge_attach(device_t, device_t, void *);
186 static void	bge_release_resources(struct bge_softc *);
187 static void	bge_txeof(struct bge_softc *);
188 static void	bge_rxeof(struct bge_softc *);
189 
190 static void	bge_tick(void *);
191 static void	bge_stats_update(struct bge_softc *);
192 static int	bge_encap(struct bge_softc *, struct mbuf *, u_int32_t *);
193 
194 static int	bge_intr(void *);
195 static void	bge_start(struct ifnet *);
196 static int	bge_ioctl(struct ifnet *, u_long, void *);
197 static int	bge_init(struct ifnet *);
198 static void	bge_stop(struct ifnet *, int);
199 static void	bge_watchdog(struct ifnet *);
200 static int	bge_ifmedia_upd(struct ifnet *);
201 static void	bge_ifmedia_sts(struct ifnet *, struct ifmediareq *);
202 
203 static void	bge_setmulti(struct bge_softc *);
204 
205 static void	bge_handle_events(struct bge_softc *);
206 static int	bge_alloc_jumbo_mem(struct bge_softc *);
207 #if 0 /* XXX */
208 static void	bge_free_jumbo_mem(struct bge_softc *);
209 #endif
210 static void	*bge_jalloc(struct bge_softc *);
211 static void	bge_jfree(struct mbuf *, void *, size_t, void *);
212 static int	bge_newbuf_std(struct bge_softc *, int, struct mbuf *,
213 			       bus_dmamap_t);
214 static int	bge_newbuf_jumbo(struct bge_softc *, int, struct mbuf *);
215 static int	bge_init_rx_ring_std(struct bge_softc *);
216 static void	bge_free_rx_ring_std(struct bge_softc *);
217 static int	bge_init_rx_ring_jumbo(struct bge_softc *);
218 static void	bge_free_rx_ring_jumbo(struct bge_softc *);
219 static void	bge_free_tx_ring(struct bge_softc *);
220 static int	bge_init_tx_ring(struct bge_softc *);
221 
222 static int	bge_chipinit(struct bge_softc *);
223 static int	bge_blockinit(struct bge_softc *);
224 static int	bge_setpowerstate(struct bge_softc *, int);
225 
226 static void	bge_reset(struct bge_softc *);
227 
228 #define BGE_DEBUG
229 #ifdef BGE_DEBUG
230 #define DPRINTF(x)	if (bgedebug) printf x
231 #define DPRINTFN(n,x)	if (bgedebug >= (n)) printf x
232 #define BGE_TSO_PRINTF(x)  do { if (bge_tso_debug) printf x ;} while (0)
233 int	bgedebug = 0;
234 int	bge_tso_debug = 0;
235 #else
236 #define DPRINTF(x)
237 #define DPRINTFN(n,x)
238 #define BGE_TSO_PRINTF(x)
239 #endif
240 
241 #ifdef BGE_EVENT_COUNTERS
242 #define	BGE_EVCNT_INCR(ev)	(ev).ev_count++
243 #define	BGE_EVCNT_ADD(ev, val)	(ev).ev_count += (val)
244 #define	BGE_EVCNT_UPD(ev, val)	(ev).ev_count = (val)
245 #else
246 #define	BGE_EVCNT_INCR(ev)	/* nothing */
247 #define	BGE_EVCNT_ADD(ev, val)	/* nothing */
248 #define	BGE_EVCNT_UPD(ev, val)	/* nothing */
249 #endif
250 
251 /* Various chip quirks. */
252 #define	BGE_QUIRK_LINK_STATE_BROKEN	0x00000001
253 #define	BGE_QUIRK_CSUM_BROKEN		0x00000002
254 #define	BGE_QUIRK_ONLY_PHY_1		0x00000004
255 #define	BGE_QUIRK_5700_SMALLDMA		0x00000008
256 #define	BGE_QUIRK_5700_PCIX_REG_BUG	0x00000010
257 #define	BGE_QUIRK_PRODUCER_BUG		0x00000020
258 #define	BGE_QUIRK_PCIX_DMA_ALIGN_BUG	0x00000040
259 #define	BGE_QUIRK_5705_CORE		0x00000080
260 #define	BGE_QUIRK_FEWER_MBUFS		0x00000100
261 
262 /*
263  * XXX: how to handle variants based on 5750 and derivatives:
264  * 5750 5751, 5721, possibly 5714, 5752, and 5708?, which
265  * in general behave like a 5705, except with additional quirks.
266  * This driver's current handling of the 5721 is wrong;
267  * how we map ASIC revision to "quirks" needs more thought.
268  * (defined here until the thought is done).
269  */
270 #define BGE_IS_5714_FAMILY(sc) \
271 	(BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5714_A0 || \
272 	 BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5780 ||	\
273 	 BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5714 )
274 
275 #define BGE_IS_5750_OR_BEYOND(sc)  \
276 	(BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5750 || \
277 	 BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5752 || \
278 	 BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5755 || \
279 	 BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5787 || \
280 	 BGE_IS_5714_FAMILY(sc) )
281 
282 #define BGE_IS_5705_OR_BEYOND(sc)  \
283 	( ((sc)->bge_quirks & BGE_QUIRK_5705_CORE) || \
284 	  BGE_IS_5750_OR_BEYOND(sc) )
285 
286 
287 /* following bugs are common to bcm5700 rev B, all flavours */
288 #define BGE_QUIRK_5700_COMMON \
289 	(BGE_QUIRK_5700_SMALLDMA|BGE_QUIRK_PRODUCER_BUG)
290 
291 CFATTACH_DECL_NEW(bge, sizeof(struct bge_softc),
292     bge_probe, bge_attach, NULL, NULL);
293 
294 static u_int32_t
295 bge_readmem_ind(struct bge_softc *sc, int off)
296 {
297 	pcireg_t val;
298 
299 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MEMWIN_BASEADDR, off);
300 	val = pci_conf_read(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MEMWIN_DATA);
301 	return val;
302 }
303 
304 static void
305 bge_writemem_ind(struct bge_softc *sc, int off, int val)
306 {
307 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MEMWIN_BASEADDR, off);
308 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MEMWIN_DATA, val);
309 }
310 
311 #ifdef notdef
312 static u_int32_t
313 bge_readreg_ind(struct bge_softc *sc, int off)
314 {
315 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_REG_BASEADDR, off);
316 	return(pci_conf_read(sc->sc_pc, sc->sc_pcitag, BGE_PCI_REG_DATA));
317 }
318 #endif
319 
320 static void
321 bge_writereg_ind(struct bge_softc *sc, int off, int val)
322 {
323 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_REG_BASEADDR, off);
324 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_REG_DATA, val);
325 }
326 
327 #ifdef notdef
328 static u_int8_t
329 bge_vpd_readbyte(struct bge_softc *sc, int addr)
330 {
331 	int i;
332 	u_int32_t val;
333 
334 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_VPD_ADDR, addr);
335 	for (i = 0; i < BGE_TIMEOUT * 10; i++) {
336 		DELAY(10);
337 		if (pci_conf_read(sc->sc_pc, sc->sc_pcitag, BGE_PCI_VPD_ADDR) &
338 		    BGE_VPD_FLAG)
339 			break;
340 	}
341 
342 	if (i == BGE_TIMEOUT) {
343 		aprint_error_dev(sc->bge_dev, "VPD read timed out\n");
344 		return(0);
345 	}
346 
347 	val = pci_conf_read(sc->sc_pc, sc->sca_pcitag, BGE_PCI_VPD_DATA);
348 
349 	return((val >> ((addr % 4) * 8)) & 0xFF);
350 }
351 
352 static void
353 bge_vpd_read_res(struct bge_softc *sc, struct vpd_res *res, int addr)
354 {
355 	int i;
356 	u_int8_t *ptr;
357 
358 	ptr = (u_int8_t *)res;
359 	for (i = 0; i < sizeof(struct vpd_res); i++)
360 		ptr[i] = bge_vpd_readbyte(sc, i + addr);
361 }
362 
363 static void
364 bge_vpd_read(struct bge_softc *sc)
365 {
366 	int pos = 0, i;
367 	struct vpd_res res;
368 
369 	if (sc->bge_vpd_prodname != NULL)
370 		free(sc->bge_vpd_prodname, M_DEVBUF);
371 	if (sc->bge_vpd_readonly != NULL)
372 		free(sc->bge_vpd_readonly, M_DEVBUF);
373 	sc->bge_vpd_prodname = NULL;
374 	sc->bge_vpd_readonly = NULL;
375 
376 	bge_vpd_read_res(sc, &res, pos);
377 
378 	if (res.vr_id != VPD_RES_ID) {
379 		aprint_error_dev("bad VPD resource id: expected %x got %x\n",
380 		    VPD_RES_ID, res.vr_id);
381 		return;
382 	}
383 
384 	pos += sizeof(res);
385 	sc->bge_vpd_prodname = malloc(res.vr_len + 1, M_DEVBUF, M_NOWAIT);
386 	if (sc->bge_vpd_prodname == NULL)
387 		panic("bge_vpd_read");
388 	for (i = 0; i < res.vr_len; i++)
389 		sc->bge_vpd_prodname[i] = bge_vpd_readbyte(sc, i + pos);
390 	sc->bge_vpd_prodname[i] = '\0';
391 	pos += i;
392 
393 	bge_vpd_read_res(sc, &res, pos);
394 
395 	if (res.vr_id != VPD_RES_READ) {
396 		aprint_error_dev(sc->bge_dev,
397 		    "bad VPD resource id: expected %x got %x\n",
398 		    VPD_RES_READ, res.vr_id);
399 		return;
400 	}
401 
402 	pos += sizeof(res);
403 	sc->bge_vpd_readonly = malloc(res.vr_len, M_DEVBUF, M_NOWAIT);
404 	if (sc->bge_vpd_readonly == NULL)
405 		panic("bge_vpd_read");
406 	for (i = 0; i < res.vr_len + 1; i++)
407 		sc->bge_vpd_readonly[i] = bge_vpd_readbyte(sc, i + pos);
408 }
409 #endif
410 
411 /*
412  * Read a byte of data stored in the EEPROM at address 'addr.' The
413  * BCM570x supports both the traditional bitbang interface and an
414  * auto access interface for reading the EEPROM. We use the auto
415  * access method.
416  */
417 static u_int8_t
418 bge_eeprom_getbyte(struct bge_softc *sc, int addr, u_int8_t *dest)
419 {
420 	int i;
421 	u_int32_t byte = 0;
422 
423 	/*
424 	 * Enable use of auto EEPROM access so we can avoid
425 	 * having to use the bitbang method.
426 	 */
427 	BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_AUTO_EEPROM);
428 
429 	/* Reset the EEPROM, load the clock period. */
430 	CSR_WRITE_4(sc, BGE_EE_ADDR,
431 	    BGE_EEADDR_RESET|BGE_EEHALFCLK(BGE_HALFCLK_384SCL));
432 	DELAY(20);
433 
434 	/* Issue the read EEPROM command. */
435 	CSR_WRITE_4(sc, BGE_EE_ADDR, BGE_EE_READCMD | addr);
436 
437 	/* Wait for completion */
438 	for(i = 0; i < BGE_TIMEOUT * 10; i++) {
439 		DELAY(10);
440 		if (CSR_READ_4(sc, BGE_EE_ADDR) & BGE_EEADDR_DONE)
441 			break;
442 	}
443 
444 	if (i == BGE_TIMEOUT) {
445 		aprint_error_dev(sc->bge_dev, "eeprom read timed out\n");
446 		return(0);
447 	}
448 
449 	/* Get result. */
450 	byte = CSR_READ_4(sc, BGE_EE_DATA);
451 
452 	*dest = (byte >> ((addr % 4) * 8)) & 0xFF;
453 
454 	return(0);
455 }
456 
457 /*
458  * Read a sequence of bytes from the EEPROM.
459  */
460 static int
461 bge_read_eeprom(struct bge_softc *sc, void *destv, int off, int cnt)
462 {
463 	int err = 0, i;
464 	u_int8_t byte = 0;
465 	char *dest = destv;
466 
467 	for (i = 0; i < cnt; i++) {
468 		err = bge_eeprom_getbyte(sc, off + i, &byte);
469 		if (err)
470 			break;
471 		*(dest + i) = byte;
472 	}
473 
474 	return(err ? 1 : 0);
475 }
476 
477 static int
478 bge_miibus_readreg(device_t dev, int phy, int reg)
479 {
480 	struct bge_softc *sc = device_private(dev);
481 	u_int32_t val;
482 	u_int32_t saved_autopoll;
483 	int i;
484 
485 	/*
486 	 * Several chips with builtin PHYs will incorrectly answer to
487 	 * other PHY instances than the builtin PHY at id 1.
488 	 */
489 	if (phy != 1 && (sc->bge_quirks & BGE_QUIRK_ONLY_PHY_1))
490 		return(0);
491 
492 	/* Reading with autopolling on may trigger PCI errors */
493 	saved_autopoll = CSR_READ_4(sc, BGE_MI_MODE);
494 	if (saved_autopoll & BGE_MIMODE_AUTOPOLL) {
495 		CSR_WRITE_4(sc, BGE_MI_MODE,
496 		    saved_autopoll &~ BGE_MIMODE_AUTOPOLL);
497 		DELAY(40);
498 	}
499 
500 	CSR_WRITE_4(sc, BGE_MI_COMM, BGE_MICMD_READ|BGE_MICOMM_BUSY|
501 	    BGE_MIPHY(phy)|BGE_MIREG(reg));
502 
503 	for (i = 0; i < BGE_TIMEOUT; i++) {
504 		val = CSR_READ_4(sc, BGE_MI_COMM);
505 		if (!(val & BGE_MICOMM_BUSY))
506 			break;
507 		delay(10);
508 	}
509 
510 	if (i == BGE_TIMEOUT) {
511 		aprint_error_dev(sc->bge_dev, "PHY read timed out\n");
512 		val = 0;
513 		goto done;
514 	}
515 
516 	val = CSR_READ_4(sc, BGE_MI_COMM);
517 
518 done:
519 	if (saved_autopoll & BGE_MIMODE_AUTOPOLL) {
520 		CSR_WRITE_4(sc, BGE_MI_MODE, saved_autopoll);
521 		DELAY(40);
522 	}
523 
524 	if (val & BGE_MICOMM_READFAIL)
525 		return(0);
526 
527 	return(val & 0xFFFF);
528 }
529 
530 static void
531 bge_miibus_writereg(device_t dev, int phy, int reg, int val)
532 {
533 	struct bge_softc *sc = device_private(dev);
534 	u_int32_t saved_autopoll;
535 	int i;
536 
537 	/* Touching the PHY while autopolling is on may trigger PCI errors */
538 	saved_autopoll = CSR_READ_4(sc, BGE_MI_MODE);
539 	if (saved_autopoll & BGE_MIMODE_AUTOPOLL) {
540 		delay(40);
541 		CSR_WRITE_4(sc, BGE_MI_MODE,
542 		    saved_autopoll & (~BGE_MIMODE_AUTOPOLL));
543 		delay(10); /* 40 usec is supposed to be adequate */
544 	}
545 
546 	CSR_WRITE_4(sc, BGE_MI_COMM, BGE_MICMD_WRITE|BGE_MICOMM_BUSY|
547 	    BGE_MIPHY(phy)|BGE_MIREG(reg)|val);
548 
549 	for (i = 0; i < BGE_TIMEOUT; i++) {
550 		if (!(CSR_READ_4(sc, BGE_MI_COMM) & BGE_MICOMM_BUSY))
551 			break;
552 		delay(10);
553 	}
554 
555 	if (saved_autopoll & BGE_MIMODE_AUTOPOLL) {
556 		CSR_WRITE_4(sc, BGE_MI_MODE, saved_autopoll);
557 		delay(40);
558 	}
559 
560 	if (i == BGE_TIMEOUT)
561 		aprint_error_dev(sc->bge_dev, "PHY read timed out\n");
562 }
563 
564 static void
565 bge_miibus_statchg(device_t dev)
566 {
567 	struct bge_softc *sc = device_private(dev);
568 	struct mii_data *mii = &sc->bge_mii;
569 
570 	/*
571 	 * Get flow control negotiation result.
572 	 */
573 	if (IFM_SUBTYPE(mii->mii_media.ifm_cur->ifm_media) == IFM_AUTO &&
574 	    (mii->mii_media_active & IFM_ETH_FMASK) != sc->bge_flowflags) {
575 		sc->bge_flowflags = mii->mii_media_active & IFM_ETH_FMASK;
576 		mii->mii_media_active &= ~IFM_ETH_FMASK;
577 	}
578 
579 	BGE_CLRBIT(sc, BGE_MAC_MODE, BGE_MACMODE_PORTMODE);
580 	if (IFM_SUBTYPE(mii->mii_media_active) == IFM_1000_T) {
581 		BGE_SETBIT(sc, BGE_MAC_MODE, BGE_PORTMODE_GMII);
582 	} else {
583 		BGE_SETBIT(sc, BGE_MAC_MODE, BGE_PORTMODE_MII);
584 	}
585 
586 	if ((mii->mii_media_active & IFM_GMASK) == IFM_FDX) {
587 		BGE_CLRBIT(sc, BGE_MAC_MODE, BGE_MACMODE_HALF_DUPLEX);
588 	} else {
589 		BGE_SETBIT(sc, BGE_MAC_MODE, BGE_MACMODE_HALF_DUPLEX);
590 	}
591 
592 	/*
593 	 * 802.3x flow control
594 	 */
595 	if (sc->bge_flowflags & IFM_ETH_RXPAUSE) {
596 		BGE_SETBIT(sc, BGE_RX_MODE, BGE_RXMODE_FLOWCTL_ENABLE);
597 	} else {
598 		BGE_CLRBIT(sc, BGE_RX_MODE, BGE_RXMODE_FLOWCTL_ENABLE);
599 	}
600 	if (sc->bge_flowflags & IFM_ETH_TXPAUSE) {
601 		BGE_SETBIT(sc, BGE_TX_MODE, BGE_TXMODE_FLOWCTL_ENABLE);
602 	} else {
603 		BGE_CLRBIT(sc, BGE_TX_MODE, BGE_TXMODE_FLOWCTL_ENABLE);
604 	}
605 }
606 
607 /*
608  * Update rx threshold levels to values in a particular slot
609  * of the interrupt-mitigation table bge_rx_threshes.
610  */
611 static void
612 bge_set_thresh(struct ifnet *ifp, int lvl)
613 {
614 	struct bge_softc *sc = ifp->if_softc;
615 	int s;
616 
617 	/* For now, just save the new Rx-intr thresholds and record
618 	 * that a threshold update is pending.  Updating the hardware
619 	 * registers here (even at splhigh()) is observed to
620 	 * occasionaly cause glitches where Rx-interrupts are not
621 	 * honoured for up to 10 seconds. jonathan@NetBSD.org, 2003-04-05
622 	 */
623 	s = splnet();
624 	sc->bge_rx_coal_ticks = bge_rx_threshes[lvl].rx_ticks;
625 	sc->bge_rx_max_coal_bds = bge_rx_threshes[lvl].rx_max_bds;
626 	sc->bge_pending_rxintr_change = 1;
627 	splx(s);
628 
629 	 return;
630 }
631 
632 
633 /*
634  * Update Rx thresholds of all bge devices
635  */
636 static void
637 bge_update_all_threshes(int lvl)
638 {
639 	struct ifnet *ifp;
640 	const char * const namebuf = "bge";
641 	int namelen;
642 
643 	if (lvl < 0)
644 		lvl = 0;
645 	else if( lvl >= NBGE_RX_THRESH)
646 		lvl = NBGE_RX_THRESH - 1;
647 
648 	namelen = strlen(namebuf);
649 	/*
650 	 * Now search all the interfaces for this name/number
651 	 */
652 	IFNET_FOREACH(ifp) {
653 		if (strncmp(ifp->if_xname, namebuf, namelen) != 0)
654 		      continue;
655 		/* We got a match: update if doing auto-threshold-tuning */
656 		if (bge_auto_thresh)
657 			bge_set_thresh(ifp, lvl);
658 	}
659 }
660 
661 /*
662  * Handle events that have triggered interrupts.
663  */
664 static void
665 bge_handle_events(struct bge_softc *sc)
666 {
667 
668 	return;
669 }
670 
671 /*
672  * Memory management for jumbo frames.
673  */
674 
675 static int
676 bge_alloc_jumbo_mem(struct bge_softc *sc)
677 {
678 	char *ptr, *kva;
679 	bus_dma_segment_t	seg;
680 	int		i, rseg, state, error;
681 	struct bge_jpool_entry   *entry;
682 
683 	state = error = 0;
684 
685 	/* Grab a big chunk o' storage. */
686 	if (bus_dmamem_alloc(sc->bge_dmatag, BGE_JMEM, PAGE_SIZE, 0,
687 	     &seg, 1, &rseg, BUS_DMA_NOWAIT)) {
688 		aprint_error_dev(sc->bge_dev, "can't alloc rx buffers\n");
689 		return ENOBUFS;
690 	}
691 
692 	state = 1;
693 	if (bus_dmamem_map(sc->bge_dmatag, &seg, rseg, BGE_JMEM, (void **)&kva,
694 	    BUS_DMA_NOWAIT)) {
695 		aprint_error_dev(sc->bge_dev,
696 		    "can't map DMA buffers (%d bytes)\n", (int)BGE_JMEM);
697 		error = ENOBUFS;
698 		goto out;
699 	}
700 
701 	state = 2;
702 	if (bus_dmamap_create(sc->bge_dmatag, BGE_JMEM, 1, BGE_JMEM, 0,
703 	    BUS_DMA_NOWAIT, &sc->bge_cdata.bge_rx_jumbo_map)) {
704 		aprint_error_dev(sc->bge_dev, "can't create DMA map\n");
705 		error = ENOBUFS;
706 		goto out;
707 	}
708 
709 	state = 3;
710 	if (bus_dmamap_load(sc->bge_dmatag, sc->bge_cdata.bge_rx_jumbo_map,
711 	    kva, BGE_JMEM, NULL, BUS_DMA_NOWAIT)) {
712 		aprint_error_dev(sc->bge_dev, "can't load DMA map\n");
713 		error = ENOBUFS;
714 		goto out;
715 	}
716 
717 	state = 4;
718 	sc->bge_cdata.bge_jumbo_buf = (void *)kva;
719 	DPRINTFN(1,("bge_jumbo_buf = %p\n", sc->bge_cdata.bge_jumbo_buf));
720 
721 	SLIST_INIT(&sc->bge_jfree_listhead);
722 	SLIST_INIT(&sc->bge_jinuse_listhead);
723 
724 	/*
725 	 * Now divide it up into 9K pieces and save the addresses
726 	 * in an array.
727 	 */
728 	ptr = sc->bge_cdata.bge_jumbo_buf;
729 	for (i = 0; i < BGE_JSLOTS; i++) {
730 		sc->bge_cdata.bge_jslots[i] = ptr;
731 		ptr += BGE_JLEN;
732 		entry = malloc(sizeof(struct bge_jpool_entry),
733 		    M_DEVBUF, M_NOWAIT);
734 		if (entry == NULL) {
735 			aprint_error_dev(sc->bge_dev,
736 			    "no memory for jumbo buffer queue!\n");
737 			error = ENOBUFS;
738 			goto out;
739 		}
740 		entry->slot = i;
741 		SLIST_INSERT_HEAD(&sc->bge_jfree_listhead,
742 				 entry, jpool_entries);
743 	}
744 out:
745 	if (error != 0) {
746 		switch (state) {
747 		case 4:
748 			bus_dmamap_unload(sc->bge_dmatag,
749 			    sc->bge_cdata.bge_rx_jumbo_map);
750 		case 3:
751 			bus_dmamap_destroy(sc->bge_dmatag,
752 			    sc->bge_cdata.bge_rx_jumbo_map);
753 		case 2:
754 			bus_dmamem_unmap(sc->bge_dmatag, kva, BGE_JMEM);
755 		case 1:
756 			bus_dmamem_free(sc->bge_dmatag, &seg, rseg);
757 			break;
758 		default:
759 			break;
760 		}
761 	}
762 
763 	return error;
764 }
765 
766 /*
767  * Allocate a jumbo buffer.
768  */
769 static void *
770 bge_jalloc(struct bge_softc *sc)
771 {
772 	struct bge_jpool_entry   *entry;
773 
774 	entry = SLIST_FIRST(&sc->bge_jfree_listhead);
775 
776 	if (entry == NULL) {
777 		aprint_error_dev(sc->bge_dev, "no free jumbo buffers\n");
778 		return(NULL);
779 	}
780 
781 	SLIST_REMOVE_HEAD(&sc->bge_jfree_listhead, jpool_entries);
782 	SLIST_INSERT_HEAD(&sc->bge_jinuse_listhead, entry, jpool_entries);
783 	return(sc->bge_cdata.bge_jslots[entry->slot]);
784 }
785 
786 /*
787  * Release a jumbo buffer.
788  */
789 static void
790 bge_jfree(struct mbuf *m, void *buf, size_t size, void *arg)
791 {
792 	struct bge_jpool_entry *entry;
793 	struct bge_softc *sc;
794 	int i, s;
795 
796 	/* Extract the softc struct pointer. */
797 	sc = (struct bge_softc *)arg;
798 
799 	if (sc == NULL)
800 		panic("bge_jfree: can't find softc pointer!");
801 
802 	/* calculate the slot this buffer belongs to */
803 
804 	i = ((char *)buf
805 	     - (char *)sc->bge_cdata.bge_jumbo_buf) / BGE_JLEN;
806 
807 	if ((i < 0) || (i >= BGE_JSLOTS))
808 		panic("bge_jfree: asked to free buffer that we don't manage!");
809 
810 	s = splvm();
811 	entry = SLIST_FIRST(&sc->bge_jinuse_listhead);
812 	if (entry == NULL)
813 		panic("bge_jfree: buffer not in use!");
814 	entry->slot = i;
815 	SLIST_REMOVE_HEAD(&sc->bge_jinuse_listhead, jpool_entries);
816 	SLIST_INSERT_HEAD(&sc->bge_jfree_listhead, entry, jpool_entries);
817 
818 	if (__predict_true(m != NULL))
819   		pool_cache_put(mb_cache, m);
820 	splx(s);
821 }
822 
823 
824 /*
825  * Intialize a standard receive ring descriptor.
826  */
827 static int
828 bge_newbuf_std(struct bge_softc *sc, int i, struct mbuf *m, bus_dmamap_t dmamap)
829 {
830 	struct mbuf		*m_new = NULL;
831 	struct bge_rx_bd	*r;
832 	int			error;
833 
834 	if (dmamap == NULL) {
835 		error = bus_dmamap_create(sc->bge_dmatag, MCLBYTES, 1,
836 		    MCLBYTES, 0, BUS_DMA_NOWAIT, &dmamap);
837 		if (error != 0)
838 			return error;
839 	}
840 
841 	sc->bge_cdata.bge_rx_std_map[i] = dmamap;
842 
843 	if (m == NULL) {
844 		MGETHDR(m_new, M_DONTWAIT, MT_DATA);
845 		if (m_new == NULL) {
846 			return(ENOBUFS);
847 		}
848 
849 		MCLGET(m_new, M_DONTWAIT);
850 		if (!(m_new->m_flags & M_EXT)) {
851 			m_freem(m_new);
852 			return(ENOBUFS);
853 		}
854 		m_new->m_len = m_new->m_pkthdr.len = MCLBYTES;
855 
856 	} else {
857 		m_new = m;
858 		m_new->m_len = m_new->m_pkthdr.len = MCLBYTES;
859 		m_new->m_data = m_new->m_ext.ext_buf;
860 	}
861 	if (!sc->bge_rx_alignment_bug)
862 	    m_adj(m_new, ETHER_ALIGN);
863 	if (bus_dmamap_load_mbuf(sc->bge_dmatag, dmamap, m_new,
864 	    BUS_DMA_READ|BUS_DMA_NOWAIT))
865 		return(ENOBUFS);
866 	bus_dmamap_sync(sc->bge_dmatag, dmamap, 0, dmamap->dm_mapsize,
867 	    BUS_DMASYNC_PREREAD);
868 
869 	sc->bge_cdata.bge_rx_std_chain[i] = m_new;
870 	r = &sc->bge_rdata->bge_rx_std_ring[i];
871 	bge_set_hostaddr(&r->bge_addr,
872 	    dmamap->dm_segs[0].ds_addr);
873 	r->bge_flags = BGE_RXBDFLAG_END;
874 	r->bge_len = m_new->m_len;
875 	r->bge_idx = i;
876 
877 	bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
878 	    offsetof(struct bge_ring_data, bge_rx_std_ring) +
879 		i * sizeof (struct bge_rx_bd),
880 	    sizeof (struct bge_rx_bd),
881 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
882 
883 	return(0);
884 }
885 
886 /*
887  * Initialize a jumbo receive ring descriptor. This allocates
888  * a jumbo buffer from the pool managed internally by the driver.
889  */
890 static int
891 bge_newbuf_jumbo(struct bge_softc *sc, int i, struct mbuf *m)
892 {
893 	struct mbuf *m_new = NULL;
894 	struct bge_rx_bd *r;
895 	void *buf = NULL;
896 
897 	if (m == NULL) {
898 
899 		/* Allocate the mbuf. */
900 		MGETHDR(m_new, M_DONTWAIT, MT_DATA);
901 		if (m_new == NULL) {
902 			return(ENOBUFS);
903 		}
904 
905 		/* Allocate the jumbo buffer */
906 		buf = bge_jalloc(sc);
907 		if (buf == NULL) {
908 			m_freem(m_new);
909 			aprint_error_dev(sc->bge_dev,
910 			    "jumbo allocation failed -- packet dropped!\n");
911 			return(ENOBUFS);
912 		}
913 
914 		/* Attach the buffer to the mbuf. */
915 		m_new->m_len = m_new->m_pkthdr.len = BGE_JUMBO_FRAMELEN;
916 		MEXTADD(m_new, buf, BGE_JUMBO_FRAMELEN, M_DEVBUF,
917 		    bge_jfree, sc);
918 		m_new->m_flags |= M_EXT_RW;
919 	} else {
920 		m_new = m;
921 		buf = m_new->m_data = m_new->m_ext.ext_buf;
922 		m_new->m_ext.ext_size = BGE_JUMBO_FRAMELEN;
923 	}
924 	if (!sc->bge_rx_alignment_bug)
925 	    m_adj(m_new, ETHER_ALIGN);
926 	bus_dmamap_sync(sc->bge_dmatag, sc->bge_cdata.bge_rx_jumbo_map,
927 	    mtod(m_new, char *) - (char *)sc->bge_cdata.bge_jumbo_buf, BGE_JLEN,
928 	    BUS_DMASYNC_PREREAD);
929 	/* Set up the descriptor. */
930 	r = &sc->bge_rdata->bge_rx_jumbo_ring[i];
931 	sc->bge_cdata.bge_rx_jumbo_chain[i] = m_new;
932 	bge_set_hostaddr(&r->bge_addr, BGE_JUMBO_DMA_ADDR(sc, m_new));
933 	r->bge_flags = BGE_RXBDFLAG_END|BGE_RXBDFLAG_JUMBO_RING;
934 	r->bge_len = m_new->m_len;
935 	r->bge_idx = i;
936 
937 	bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
938 	    offsetof(struct bge_ring_data, bge_rx_jumbo_ring) +
939 		i * sizeof (struct bge_rx_bd),
940 	    sizeof (struct bge_rx_bd),
941 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
942 
943 	return(0);
944 }
945 
946 /*
947  * The standard receive ring has 512 entries in it. At 2K per mbuf cluster,
948  * that's 1MB or memory, which is a lot. For now, we fill only the first
949  * 256 ring entries and hope that our CPU is fast enough to keep up with
950  * the NIC.
951  */
952 static int
953 bge_init_rx_ring_std(struct bge_softc *sc)
954 {
955 	int i;
956 
957 	if (sc->bge_flags & BGE_RXRING_VALID)
958 		return 0;
959 
960 	for (i = 0; i < BGE_SSLOTS; i++) {
961 		if (bge_newbuf_std(sc, i, NULL, 0) == ENOBUFS)
962 			return(ENOBUFS);
963 	}
964 
965 	sc->bge_std = i - 1;
966 	CSR_WRITE_4(sc, BGE_MBX_RX_STD_PROD_LO, sc->bge_std);
967 
968 	sc->bge_flags |= BGE_RXRING_VALID;
969 
970 	return(0);
971 }
972 
973 static void
974 bge_free_rx_ring_std(struct bge_softc *sc)
975 {
976 	int i;
977 
978 	if (!(sc->bge_flags & BGE_RXRING_VALID))
979 		return;
980 
981 	for (i = 0; i < BGE_STD_RX_RING_CNT; i++) {
982 		if (sc->bge_cdata.bge_rx_std_chain[i] != NULL) {
983 			m_freem(sc->bge_cdata.bge_rx_std_chain[i]);
984 			sc->bge_cdata.bge_rx_std_chain[i] = NULL;
985 			bus_dmamap_destroy(sc->bge_dmatag,
986 			    sc->bge_cdata.bge_rx_std_map[i]);
987 		}
988 		memset((char *)&sc->bge_rdata->bge_rx_std_ring[i], 0,
989 		    sizeof(struct bge_rx_bd));
990 	}
991 
992 	sc->bge_flags &= ~BGE_RXRING_VALID;
993 }
994 
995 static int
996 bge_init_rx_ring_jumbo(struct bge_softc *sc)
997 {
998 	int i;
999 	volatile struct bge_rcb *rcb;
1000 
1001 	if (sc->bge_flags & BGE_JUMBO_RXRING_VALID)
1002 		return 0;
1003 
1004 	for (i = 0; i < BGE_JUMBO_RX_RING_CNT; i++) {
1005 		if (bge_newbuf_jumbo(sc, i, NULL) == ENOBUFS)
1006 			return(ENOBUFS);
1007 	};
1008 
1009 	sc->bge_jumbo = i - 1;
1010 	sc->bge_flags |= BGE_JUMBO_RXRING_VALID;
1011 
1012 	rcb = &sc->bge_rdata->bge_info.bge_jumbo_rx_rcb;
1013 	rcb->bge_maxlen_flags = 0;
1014 	CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_MAXLEN_FLAGS, rcb->bge_maxlen_flags);
1015 
1016 	CSR_WRITE_4(sc, BGE_MBX_RX_JUMBO_PROD_LO, sc->bge_jumbo);
1017 
1018 	return(0);
1019 }
1020 
1021 static void
1022 bge_free_rx_ring_jumbo(struct bge_softc *sc)
1023 {
1024 	int i;
1025 
1026 	if (!(sc->bge_flags & BGE_JUMBO_RXRING_VALID))
1027 		return;
1028 
1029 	for (i = 0; i < BGE_JUMBO_RX_RING_CNT; i++) {
1030 		if (sc->bge_cdata.bge_rx_jumbo_chain[i] != NULL) {
1031 			m_freem(sc->bge_cdata.bge_rx_jumbo_chain[i]);
1032 			sc->bge_cdata.bge_rx_jumbo_chain[i] = NULL;
1033 		}
1034 		memset((char *)&sc->bge_rdata->bge_rx_jumbo_ring[i], 0,
1035 		    sizeof(struct bge_rx_bd));
1036 	}
1037 
1038 	sc->bge_flags &= ~BGE_JUMBO_RXRING_VALID;
1039 }
1040 
1041 static void
1042 bge_free_tx_ring(struct bge_softc *sc)
1043 {
1044 	int i, freed;
1045 	struct txdmamap_pool_entry *dma;
1046 
1047 	if (!(sc->bge_flags & BGE_TXRING_VALID))
1048 		return;
1049 
1050 	freed = 0;
1051 
1052 	for (i = 0; i < BGE_TX_RING_CNT; i++) {
1053 		if (sc->bge_cdata.bge_tx_chain[i] != NULL) {
1054 			freed++;
1055 			m_freem(sc->bge_cdata.bge_tx_chain[i]);
1056 			sc->bge_cdata.bge_tx_chain[i] = NULL;
1057 			SLIST_INSERT_HEAD(&sc->txdma_list, sc->txdma[i],
1058 					    link);
1059 			sc->txdma[i] = 0;
1060 		}
1061 		memset((char *)&sc->bge_rdata->bge_tx_ring[i], 0,
1062 		    sizeof(struct bge_tx_bd));
1063 	}
1064 
1065 	while ((dma = SLIST_FIRST(&sc->txdma_list))) {
1066 		SLIST_REMOVE_HEAD(&sc->txdma_list, link);
1067 		bus_dmamap_destroy(sc->bge_dmatag, dma->dmamap);
1068 		free(dma, M_DEVBUF);
1069 	}
1070 
1071 	sc->bge_flags &= ~BGE_TXRING_VALID;
1072 }
1073 
1074 static int
1075 bge_init_tx_ring(struct bge_softc *sc)
1076 {
1077 	int i;
1078 	bus_dmamap_t dmamap;
1079 	struct txdmamap_pool_entry *dma;
1080 
1081 	if (sc->bge_flags & BGE_TXRING_VALID)
1082 		return 0;
1083 
1084 	sc->bge_txcnt = 0;
1085 	sc->bge_tx_saved_considx = 0;
1086 
1087 	/* Initialize transmit producer index for host-memory send ring. */
1088 	sc->bge_tx_prodidx = 0;
1089 	CSR_WRITE_4(sc, BGE_MBX_TX_HOST_PROD0_LO, sc->bge_tx_prodidx);
1090 	if (sc->bge_quirks & BGE_QUIRK_PRODUCER_BUG)	/* 5700 b2 errata */
1091 		CSR_WRITE_4(sc, BGE_MBX_TX_HOST_PROD0_LO, sc->bge_tx_prodidx);
1092 
1093 	/* NIC-memory send ring  not used; initialize to zero. */
1094 	CSR_WRITE_4(sc, BGE_MBX_TX_NIC_PROD0_LO, 0);
1095 	if (sc->bge_quirks & BGE_QUIRK_PRODUCER_BUG)	/* 5700 b2 errata */
1096 		CSR_WRITE_4(sc, BGE_MBX_TX_NIC_PROD0_LO, 0);
1097 
1098 	SLIST_INIT(&sc->txdma_list);
1099 	for (i = 0; i < BGE_RSLOTS; i++) {
1100 		if (bus_dmamap_create(sc->bge_dmatag, BGE_TXDMA_MAX,
1101 		    BGE_NTXSEG, ETHER_MAX_LEN_JUMBO, 0, BUS_DMA_NOWAIT,
1102 		    &dmamap))
1103 			return(ENOBUFS);
1104 		if (dmamap == NULL)
1105 			panic("dmamap NULL in bge_init_tx_ring");
1106 		dma = malloc(sizeof(*dma), M_DEVBUF, M_NOWAIT);
1107 		if (dma == NULL) {
1108 			aprint_error_dev(sc->bge_dev,
1109 			    "can't alloc txdmamap_pool_entry\n");
1110 			bus_dmamap_destroy(sc->bge_dmatag, dmamap);
1111 			return (ENOMEM);
1112 		}
1113 		dma->dmamap = dmamap;
1114 		SLIST_INSERT_HEAD(&sc->txdma_list, dma, link);
1115 	}
1116 
1117 	sc->bge_flags |= BGE_TXRING_VALID;
1118 
1119 	return(0);
1120 }
1121 
1122 static void
1123 bge_setmulti(struct bge_softc *sc)
1124 {
1125 	struct ethercom		*ac = &sc->ethercom;
1126 	struct ifnet		*ifp = &ac->ec_if;
1127 	struct ether_multi	*enm;
1128 	struct ether_multistep  step;
1129 	u_int32_t		hashes[4] = { 0, 0, 0, 0 };
1130 	u_int32_t		h;
1131 	int			i;
1132 
1133 	if (ifp->if_flags & IFF_PROMISC)
1134 		goto allmulti;
1135 
1136 	/* Now program new ones. */
1137 	ETHER_FIRST_MULTI(step, ac, enm);
1138 	while (enm != NULL) {
1139 		if (memcmp(enm->enm_addrlo, enm->enm_addrhi, ETHER_ADDR_LEN)) {
1140 			/*
1141 			 * We must listen to a range of multicast addresses.
1142 			 * For now, just accept all multicasts, rather than
1143 			 * trying to set only those filter bits needed to match
1144 			 * the range.  (At this time, the only use of address
1145 			 * ranges is for IP multicast routing, for which the
1146 			 * range is big enough to require all bits set.)
1147 			 */
1148 			goto allmulti;
1149 		}
1150 
1151 		h = ether_crc32_le(enm->enm_addrlo, ETHER_ADDR_LEN);
1152 
1153 		/* Just want the 7 least-significant bits. */
1154 		h &= 0x7f;
1155 
1156 		hashes[(h & 0x60) >> 5] |= 1 << (h & 0x1F);
1157 		ETHER_NEXT_MULTI(step, enm);
1158 	}
1159 
1160 	ifp->if_flags &= ~IFF_ALLMULTI;
1161 	goto setit;
1162 
1163  allmulti:
1164 	ifp->if_flags |= IFF_ALLMULTI;
1165 	hashes[0] = hashes[1] = hashes[2] = hashes[3] = 0xffffffff;
1166 
1167  setit:
1168 	for (i = 0; i < 4; i++)
1169 		CSR_WRITE_4(sc, BGE_MAR0 + (i * 4), hashes[i]);
1170 }
1171 
1172 const int bge_swapbits[] = {
1173 	0,
1174 	BGE_MODECTL_BYTESWAP_DATA,
1175 	BGE_MODECTL_WORDSWAP_DATA,
1176 	BGE_MODECTL_BYTESWAP_NONFRAME,
1177 	BGE_MODECTL_WORDSWAP_NONFRAME,
1178 
1179 	BGE_MODECTL_BYTESWAP_DATA|BGE_MODECTL_WORDSWAP_DATA,
1180 	BGE_MODECTL_BYTESWAP_DATA|BGE_MODECTL_BYTESWAP_NONFRAME,
1181 	BGE_MODECTL_BYTESWAP_DATA|BGE_MODECTL_WORDSWAP_NONFRAME,
1182 
1183 	BGE_MODECTL_WORDSWAP_DATA|BGE_MODECTL_BYTESWAP_NONFRAME,
1184 	BGE_MODECTL_WORDSWAP_DATA|BGE_MODECTL_WORDSWAP_NONFRAME,
1185 
1186 	BGE_MODECTL_BYTESWAP_NONFRAME|BGE_MODECTL_WORDSWAP_NONFRAME,
1187 
1188 	BGE_MODECTL_BYTESWAP_DATA|BGE_MODECTL_WORDSWAP_DATA|
1189 	    BGE_MODECTL_BYTESWAP_NONFRAME,
1190 	BGE_MODECTL_BYTESWAP_DATA|BGE_MODECTL_WORDSWAP_DATA|
1191 	    BGE_MODECTL_WORDSWAP_NONFRAME,
1192 	BGE_MODECTL_BYTESWAP_DATA|BGE_MODECTL_BYTESWAP_NONFRAME|
1193 	    BGE_MODECTL_WORDSWAP_NONFRAME,
1194 	BGE_MODECTL_WORDSWAP_DATA|BGE_MODECTL_BYTESWAP_NONFRAME|
1195 	    BGE_MODECTL_WORDSWAP_NONFRAME,
1196 
1197 	BGE_MODECTL_BYTESWAP_DATA|BGE_MODECTL_WORDSWAP_DATA|
1198 	    BGE_MODECTL_BYTESWAP_NONFRAME|BGE_MODECTL_WORDSWAP_NONFRAME,
1199 };
1200 
1201 int bge_swapindex = 0;
1202 
1203 /*
1204  * Do endian, PCI and DMA initialization. Also check the on-board ROM
1205  * self-test results.
1206  */
1207 static int
1208 bge_chipinit(struct bge_softc *sc)
1209 {
1210 	u_int32_t		cachesize;
1211 	int			i;
1212 	u_int32_t		dma_rw_ctl;
1213 
1214 
1215 	/* Set endianness before we access any non-PCI registers. */
1216 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MISC_CTL,
1217 	    BGE_INIT);
1218 
1219 	/* Set power state to D0. */
1220 	bge_setpowerstate(sc, 0);
1221 
1222 	/*
1223 	 * Check the 'ROM failed' bit on the RX CPU to see if
1224 	 * self-tests passed.
1225 	 */
1226 	if (CSR_READ_4(sc, BGE_RXCPU_MODE) & BGE_RXCPUMODE_ROMFAIL) {
1227 		aprint_error_dev(sc->bge_dev,
1228 		    "RX CPU self-diagnostics failed!\n");
1229 		return(ENODEV);
1230 	}
1231 
1232 	/* Clear the MAC control register */
1233 	CSR_WRITE_4(sc, BGE_MAC_MODE, 0);
1234 
1235 	/*
1236 	 * Clear the MAC statistics block in the NIC's
1237 	 * internal memory.
1238 	 */
1239 	for (i = BGE_STATS_BLOCK;
1240 	    i < BGE_STATS_BLOCK_END + 1; i += sizeof(u_int32_t))
1241 		BGE_MEMWIN_WRITE(sc->sc_pc, sc->sc_pcitag, i, 0);
1242 
1243 	for (i = BGE_STATUS_BLOCK;
1244 	    i < BGE_STATUS_BLOCK_END + 1; i += sizeof(u_int32_t))
1245 		BGE_MEMWIN_WRITE(sc->sc_pc, sc->sc_pcitag, i, 0);
1246 
1247 	/* Set up the PCI DMA control register. */
1248 	if (sc->bge_pcie) {
1249 	  u_int32_t device_ctl;
1250 
1251 		/* From FreeBSD */
1252 		DPRINTFN(4, ("(%s: PCI-Express DMA setting)\n",
1253 		    device_xname(sc->bge_dev)));
1254 		dma_rw_ctl = (BGE_PCI_READ_CMD | BGE_PCI_WRITE_CMD |
1255 		    (0xf << BGE_PCIDMARWCTL_RD_WAT_SHIFT) |
1256 		    (0x2 << BGE_PCIDMARWCTL_WR_WAT_SHIFT));
1257 
1258 		/* jonathan: alternative from Linux driver */
1259 #define DMA_CTRL_WRITE_PCIE_H20MARK_128         0x00180000
1260 #define DMA_CTRL_WRITE_PCIE_H20MARK_256         0x00380000
1261 
1262 		dma_rw_ctl =   0x76000000; /* XXX XXX XXX */;
1263 		device_ctl = pci_conf_read(sc->sc_pc, sc->sc_pcitag,
1264 					   BGE_PCI_CONF_DEV_CTRL);
1265 		aprint_debug_dev(sc->bge_dev, "pcie mode=0x%x\n", device_ctl);
1266 
1267 		if ((device_ctl & 0x00e0) && 0) {
1268 			/*
1269 			 * XXX jonathan@NetBSD.org:
1270 			 * This clause is exactly what the Broadcom-supplied
1271 			 * Linux does; but given overall register programming
1272 			 * by if_bge(4), this larger DMA-write watermark
1273 			 * value causes bcm5721 chips to totally wedge.
1274 			 */
1275 			dma_rw_ctl |= BGE_PCIDMA_RWCTL_PCIE_WRITE_WATRMARK_256;
1276 		} else {
1277 			dma_rw_ctl |= BGE_PCIDMA_RWCTL_PCIE_WRITE_WATRMARK_128;
1278 		}
1279 	} else if (pci_conf_read(sc->sc_pc, sc->sc_pcitag,BGE_PCI_PCISTATE) &
1280 	    BGE_PCISTATE_PCI_BUSMODE) {
1281 		/* Conventional PCI bus */
1282 	  	DPRINTFN(4, ("(%s: PCI 2.2 DMA setting)\n",
1283 		    device_xname(sc->bge_dev)));
1284 		dma_rw_ctl = (BGE_PCI_READ_CMD | BGE_PCI_WRITE_CMD |
1285 		   (0x7 << BGE_PCIDMARWCTL_RD_WAT_SHIFT) |
1286 		   (0x7 << BGE_PCIDMARWCTL_WR_WAT_SHIFT));
1287 		if ((sc->bge_quirks & BGE_QUIRK_5705_CORE) == 0) {
1288 			dma_rw_ctl |= 0x0F;
1289 		}
1290 	} else {
1291 	  	DPRINTFN(4, ("(:%s: PCI-X DMA setting)\n",
1292 		    device_xname(sc->bge_dev)));
1293 		/* PCI-X bus */
1294 		dma_rw_ctl = BGE_PCI_READ_CMD|BGE_PCI_WRITE_CMD |
1295 		    (0x3 << BGE_PCIDMARWCTL_RD_WAT_SHIFT) |
1296 		    (0x3 << BGE_PCIDMARWCTL_WR_WAT_SHIFT) |
1297 		    (0x0F);
1298 		/*
1299 		 * 5703 and 5704 need ONEDMA_AT_ONCE as a workaround
1300 		 * for hardware bugs, which means we should also clear
1301 		 * the low-order MINDMA bits.  In addition, the 5704
1302 		 * uses a different encoding of read/write watermarks.
1303 		 */
1304 		if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5704) {
1305 			dma_rw_ctl = BGE_PCI_READ_CMD|BGE_PCI_WRITE_CMD |
1306 			  /* should be 0x1f0000 */
1307 			  (0x7 << BGE_PCIDMARWCTL_RD_WAT_SHIFT) |
1308 			  (0x3 << BGE_PCIDMARWCTL_WR_WAT_SHIFT);
1309 			dma_rw_ctl |= BGE_PCIDMARWCTL_ONEDMA_ATONCE;
1310 		}
1311 		else if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5703) {
1312 			dma_rw_ctl &=  0xfffffff0;
1313 			dma_rw_ctl |= BGE_PCIDMARWCTL_ONEDMA_ATONCE;
1314 		}
1315 		else if (BGE_IS_5714_FAMILY(sc)) {
1316 			dma_rw_ctl = BGE_PCI_READ_CMD|BGE_PCI_WRITE_CMD;
1317 			dma_rw_ctl &= ~BGE_PCIDMARWCTL_ONEDMA_ATONCE; /* XXX */
1318 			/* XXX magic values, Broadcom-supplied Linux driver */
1319 			if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5780)
1320 				dma_rw_ctl |= (1 << 20) | (1 << 18) |
1321 				  BGE_PCIDMARWCTL_ONEDMA_ATONCE;
1322 			else
1323 				dma_rw_ctl |= (1<<20) | (1<<18) | (1 << 15);
1324 		}
1325 	}
1326 
1327 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_DMA_RW_CTL, dma_rw_ctl);
1328 
1329 	/*
1330 	 * Set up general mode register.
1331 	 */
1332 	CSR_WRITE_4(sc, BGE_MODE_CTL, BGE_DMA_SWAP_OPTIONS|
1333 		    BGE_MODECTL_MAC_ATTN_INTR|BGE_MODECTL_HOST_SEND_BDS|
1334 		    BGE_MODECTL_TX_NO_PHDR_CSUM|BGE_MODECTL_RX_NO_PHDR_CSUM);
1335 
1336 	/* Get cache line size. */
1337 	cachesize = pci_conf_read(sc->sc_pc, sc->sc_pcitag, BGE_PCI_CACHESZ);
1338 
1339 	/*
1340 	 * Avoid violating PCI spec on certain chip revs.
1341 	 */
1342 	if (pci_conf_read(sc->sc_pc, sc->sc_pcitag, BGE_PCI_CMD) &
1343 	    PCIM_CMD_MWIEN) {
1344 		switch(cachesize) {
1345 		case 1:
1346 			PCI_SETBIT(sc->sc_pc, sc->sc_pcitag, BGE_PCI_DMA_RW_CTL,
1347 				   BGE_PCI_WRITE_BNDRY_16BYTES);
1348 			break;
1349 		case 2:
1350 			PCI_SETBIT(sc->sc_pc, sc->sc_pcitag, BGE_PCI_DMA_RW_CTL,
1351 				   BGE_PCI_WRITE_BNDRY_32BYTES);
1352 			break;
1353 		case 4:
1354 			PCI_SETBIT(sc->sc_pc, sc->sc_pcitag, BGE_PCI_DMA_RW_CTL,
1355 				   BGE_PCI_WRITE_BNDRY_64BYTES);
1356 			break;
1357 		case 8:
1358 			PCI_SETBIT(sc->sc_pc, sc->sc_pcitag, BGE_PCI_DMA_RW_CTL,
1359 				   BGE_PCI_WRITE_BNDRY_128BYTES);
1360 			break;
1361 		case 16:
1362 			PCI_SETBIT(sc->sc_pc, sc->sc_pcitag, BGE_PCI_DMA_RW_CTL,
1363 				   BGE_PCI_WRITE_BNDRY_256BYTES);
1364 			break;
1365 		case 32:
1366 			PCI_SETBIT(sc->sc_pc, sc->sc_pcitag, BGE_PCI_DMA_RW_CTL,
1367 				   BGE_PCI_WRITE_BNDRY_512BYTES);
1368 			break;
1369 		case 64:
1370 			PCI_SETBIT(sc->sc_pc, sc->sc_pcitag, BGE_PCI_DMA_RW_CTL,
1371 				   BGE_PCI_WRITE_BNDRY_1024BYTES);
1372 			break;
1373 		default:
1374 		/* Disable PCI memory write and invalidate. */
1375 #if 0
1376 			if (bootverbose)
1377 				aprint_error_dev(sc->bge_dev,
1378 				    "cache line size %d not supported "
1379 				    "disabling PCI MWI\n",
1380 #endif
1381 			PCI_CLRBIT(sc->sc_pc, sc->sc_pcitag, BGE_PCI_CMD,
1382 			    PCIM_CMD_MWIEN);
1383 			break;
1384 		}
1385 	}
1386 
1387 	/*
1388 	 * Disable memory write invalidate.  Apparently it is not supported
1389 	 * properly by these devices.
1390 	 */
1391 	PCI_CLRBIT(sc->sc_pc, sc->sc_pcitag, BGE_PCI_CMD, PCIM_CMD_MWIEN);
1392 
1393 
1394 #ifdef __brokenalpha__
1395 	/*
1396 	 * Must insure that we do not cross an 8K (bytes) boundary
1397 	 * for DMA reads.  Our highest limit is 1K bytes.  This is a
1398 	 * restriction on some ALPHA platforms with early revision
1399 	 * 21174 PCI chipsets, such as the AlphaPC 164lx
1400 	 */
1401 	PCI_SETBIT(sc, BGE_PCI_DMA_RW_CTL, BGE_PCI_READ_BNDRY_1024, 4);
1402 #endif
1403 
1404 	/* Set the timer prescaler (always 66MHz) */
1405 	CSR_WRITE_4(sc, BGE_MISC_CFG, 65 << 1/*BGE_32BITTIME_66MHZ*/);
1406 
1407 	return(0);
1408 }
1409 
1410 static int
1411 bge_blockinit(struct bge_softc *sc)
1412 {
1413 	volatile struct bge_rcb		*rcb;
1414 	bus_size_t		rcb_addr;
1415 	int			i;
1416 	struct ifnet		*ifp = &sc->ethercom.ec_if;
1417 	bge_hostaddr		taddr;
1418 
1419 	/*
1420 	 * Initialize the memory window pointer register so that
1421 	 * we can access the first 32K of internal NIC RAM. This will
1422 	 * allow us to set up the TX send ring RCBs and the RX return
1423 	 * ring RCBs, plus other things which live in NIC memory.
1424 	 */
1425 
1426 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MEMWIN_BASEADDR, 0);
1427 
1428 	/* Configure mbuf memory pool */
1429 	if ((sc->bge_quirks & BGE_QUIRK_5705_CORE) == 0) {
1430 		if (sc->bge_extram) {
1431 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_BASEADDR,
1432 			    BGE_EXT_SSRAM);
1433 			if ((sc->bge_quirks & BGE_QUIRK_FEWER_MBUFS) != 0)
1434 				CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_LEN, 0x10000);
1435 			else
1436 				CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_LEN, 0x18000);
1437 		} else {
1438 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_BASEADDR,
1439 			    BGE_BUFFPOOL_1);
1440 			if ((sc->bge_quirks & BGE_QUIRK_FEWER_MBUFS) != 0)
1441 				CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_LEN, 0x10000);
1442 			else
1443 				CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_LEN, 0x18000);
1444 		}
1445 
1446 		/* Configure DMA resource pool */
1447 		CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_BASEADDR,
1448 		    BGE_DMA_DESCRIPTORS);
1449 		CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_LEN, 0x2000);
1450 	}
1451 
1452 	/* Configure mbuf pool watermarks */
1453 #ifdef ORIG_WPAUL_VALUES
1454 	CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 24);
1455 	CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 24);
1456 	CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 48);
1457 #else
1458 	/* new broadcom docs strongly recommend these: */
1459 	if ((sc->bge_quirks & BGE_QUIRK_5705_CORE) == 0) {
1460 		if (ifp->if_mtu > ETHER_MAX_LEN) {
1461 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 0x50);
1462 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x20);
1463 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0x60);
1464 		} else {
1465 			/* Values from Linux driver... */
1466 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 304);
1467 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 152);
1468 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 380);
1469 		}
1470 	} else {
1471 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 0x0);
1472 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x10);
1473 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0x60);
1474 	}
1475 #endif
1476 
1477 	/* Configure DMA resource watermarks */
1478 	CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_LOWAT, 5);
1479 	CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_HIWAT, 10);
1480 
1481 	/* Enable buffer manager */
1482 	CSR_WRITE_4(sc, BGE_BMAN_MODE,
1483 	    BGE_BMANMODE_ENABLE|BGE_BMANMODE_LOMBUF_ATTN);
1484 
1485 	/* Poll for buffer manager start indication */
1486 	for (i = 0; i < BGE_TIMEOUT; i++) {
1487 		if (CSR_READ_4(sc, BGE_BMAN_MODE) & BGE_BMANMODE_ENABLE)
1488 			break;
1489 		DELAY(10);
1490 	}
1491 
1492 	if (i == BGE_TIMEOUT) {
1493 		aprint_error_dev(sc->bge_dev,
1494 		    "buffer manager failed to start\n");
1495 		return(ENXIO);
1496 	}
1497 
1498 	/* Enable flow-through queues */
1499 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0xFFFFFFFF);
1500 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0);
1501 
1502 	/* Wait until queue initialization is complete */
1503 	for (i = 0; i < BGE_TIMEOUT; i++) {
1504 		if (CSR_READ_4(sc, BGE_FTQ_RESET) == 0)
1505 			break;
1506 		DELAY(10);
1507 	}
1508 
1509 	if (i == BGE_TIMEOUT) {
1510 		aprint_error_dev(sc->bge_dev,
1511 		    "flow-through queue init failed\n");
1512 		return(ENXIO);
1513 	}
1514 
1515 	/* Initialize the standard RX ring control block */
1516 	rcb = &sc->bge_rdata->bge_info.bge_std_rx_rcb;
1517 	bge_set_hostaddr(&rcb->bge_hostaddr,
1518 	    BGE_RING_DMA_ADDR(sc, bge_rx_std_ring));
1519 	if ((sc->bge_quirks & BGE_QUIRK_5705_CORE) == 0) {
1520 		rcb->bge_maxlen_flags =
1521 		    BGE_RCB_MAXLEN_FLAGS(BGE_MAX_FRAMELEN, 0);
1522 	} else {
1523 		rcb->bge_maxlen_flags = BGE_RCB_MAXLEN_FLAGS(512, 0);
1524 	}
1525 	if (sc->bge_extram)
1526 		rcb->bge_nicaddr = BGE_EXT_STD_RX_RINGS;
1527 	else
1528 		rcb->bge_nicaddr = BGE_STD_RX_RINGS;
1529 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_HADDR_HI, rcb->bge_hostaddr.bge_addr_hi);
1530 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_HADDR_LO, rcb->bge_hostaddr.bge_addr_lo);
1531 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_MAXLEN_FLAGS, rcb->bge_maxlen_flags);
1532 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_NICADDR, rcb->bge_nicaddr);
1533 
1534 	if ((sc->bge_quirks & BGE_QUIRK_5705_CORE) == 0) {
1535 		sc->bge_return_ring_cnt = BGE_RETURN_RING_CNT;
1536 	} else {
1537 		sc->bge_return_ring_cnt = BGE_RETURN_RING_CNT_5705;
1538 	}
1539 
1540 	/*
1541 	 * Initialize the jumbo RX ring control block
1542 	 * We set the 'ring disabled' bit in the flags
1543 	 * field until we're actually ready to start
1544 	 * using this ring (i.e. once we set the MTU
1545 	 * high enough to require it).
1546 	 */
1547 	if ((sc->bge_quirks & BGE_QUIRK_5705_CORE) == 0) {
1548 		rcb = &sc->bge_rdata->bge_info.bge_jumbo_rx_rcb;
1549 		bge_set_hostaddr(&rcb->bge_hostaddr,
1550 		    BGE_RING_DMA_ADDR(sc, bge_rx_jumbo_ring));
1551 		rcb->bge_maxlen_flags =
1552 		    BGE_RCB_MAXLEN_FLAGS(BGE_MAX_FRAMELEN,
1553 			BGE_RCB_FLAG_RING_DISABLED);
1554 		if (sc->bge_extram)
1555 			rcb->bge_nicaddr = BGE_EXT_JUMBO_RX_RINGS;
1556 		else
1557 			rcb->bge_nicaddr = BGE_JUMBO_RX_RINGS;
1558 
1559 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_HADDR_HI,
1560 		    rcb->bge_hostaddr.bge_addr_hi);
1561 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_HADDR_LO,
1562 		    rcb->bge_hostaddr.bge_addr_lo);
1563 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_MAXLEN_FLAGS,
1564 		    rcb->bge_maxlen_flags);
1565 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_NICADDR, rcb->bge_nicaddr);
1566 
1567 		/* Set up dummy disabled mini ring RCB */
1568 		rcb = &sc->bge_rdata->bge_info.bge_mini_rx_rcb;
1569 		rcb->bge_maxlen_flags = BGE_RCB_MAXLEN_FLAGS(0,
1570 		    BGE_RCB_FLAG_RING_DISABLED);
1571 		CSR_WRITE_4(sc, BGE_RX_MINI_RCB_MAXLEN_FLAGS,
1572 		    rcb->bge_maxlen_flags);
1573 
1574 		bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
1575 		    offsetof(struct bge_ring_data, bge_info),
1576 		    sizeof (struct bge_gib),
1577 		    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1578 	}
1579 
1580 	/*
1581 	 * Set the BD ring replenish thresholds. The recommended
1582 	 * values are 1/8th the number of descriptors allocated to
1583 	 * each ring.
1584 	 */
1585 	i = BGE_STD_RX_RING_CNT / 8;
1586 
1587 	/*
1588  	 * Use a value of 8 for the following chips to workaround HW errata.
1589 	 * Some of these chips have been added based on empirical
1590 	 * evidence (they don't work unless this is done).
1591 	 */
1592 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5750 ||
1593 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5752 ||
1594 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5755 ||
1595 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5787)
1596 		i = 8;
1597 
1598 	CSR_WRITE_4(sc, BGE_RBDI_STD_REPL_THRESH, i);
1599 	CSR_WRITE_4(sc, BGE_RBDI_JUMBO_REPL_THRESH, BGE_JUMBO_RX_RING_CNT/8);
1600 
1601 	/*
1602 	 * Disable all unused send rings by setting the 'ring disabled'
1603 	 * bit in the flags field of all the TX send ring control blocks.
1604 	 * These are located in NIC memory.
1605 	 */
1606 	rcb_addr = BGE_MEMWIN_START + BGE_SEND_RING_RCB;
1607 	for (i = 0; i < BGE_TX_RINGS_EXTSSRAM_MAX; i++) {
1608 		RCB_WRITE_4(sc, rcb_addr, bge_maxlen_flags,
1609 		    BGE_RCB_MAXLEN_FLAGS(0,BGE_RCB_FLAG_RING_DISABLED));
1610 		RCB_WRITE_4(sc, rcb_addr, bge_nicaddr, 0);
1611 		rcb_addr += sizeof(struct bge_rcb);
1612 	}
1613 
1614 	/* Configure TX RCB 0 (we use only the first ring) */
1615 	rcb_addr = BGE_MEMWIN_START + BGE_SEND_RING_RCB;
1616 	bge_set_hostaddr(&taddr, BGE_RING_DMA_ADDR(sc, bge_tx_ring));
1617 	RCB_WRITE_4(sc, rcb_addr, bge_hostaddr.bge_addr_hi, taddr.bge_addr_hi);
1618 	RCB_WRITE_4(sc, rcb_addr, bge_hostaddr.bge_addr_lo, taddr.bge_addr_lo);
1619 	RCB_WRITE_4(sc, rcb_addr, bge_nicaddr,
1620 		    BGE_NIC_TXRING_ADDR(0, BGE_TX_RING_CNT));
1621 	if ((sc->bge_quirks & BGE_QUIRK_5705_CORE) == 0) {
1622 		RCB_WRITE_4(sc, rcb_addr, bge_maxlen_flags,
1623 		    BGE_RCB_MAXLEN_FLAGS(BGE_TX_RING_CNT, 0));
1624 	}
1625 
1626 	/* Disable all unused RX return rings */
1627 	rcb_addr = BGE_MEMWIN_START + BGE_RX_RETURN_RING_RCB;
1628 	for (i = 0; i < BGE_RX_RINGS_MAX; i++) {
1629 		RCB_WRITE_4(sc, rcb_addr, bge_hostaddr.bge_addr_hi, 0);
1630 		RCB_WRITE_4(sc, rcb_addr, bge_hostaddr.bge_addr_lo, 0);
1631 		RCB_WRITE_4(sc, rcb_addr, bge_maxlen_flags,
1632 			    BGE_RCB_MAXLEN_FLAGS(sc->bge_return_ring_cnt,
1633                                      BGE_RCB_FLAG_RING_DISABLED));
1634 		RCB_WRITE_4(sc, rcb_addr, bge_nicaddr, 0);
1635 		CSR_WRITE_4(sc, BGE_MBX_RX_CONS0_LO +
1636 		    (i * (sizeof(u_int64_t))), 0);
1637 		rcb_addr += sizeof(struct bge_rcb);
1638 	}
1639 
1640 	/* Initialize RX ring indexes */
1641 	CSR_WRITE_4(sc, BGE_MBX_RX_STD_PROD_LO, 0);
1642 	CSR_WRITE_4(sc, BGE_MBX_RX_JUMBO_PROD_LO, 0);
1643 	CSR_WRITE_4(sc, BGE_MBX_RX_MINI_PROD_LO, 0);
1644 
1645 	/*
1646 	 * Set up RX return ring 0
1647 	 * Note that the NIC address for RX return rings is 0x00000000.
1648 	 * The return rings live entirely within the host, so the
1649 	 * nicaddr field in the RCB isn't used.
1650 	 */
1651 	rcb_addr = BGE_MEMWIN_START + BGE_RX_RETURN_RING_RCB;
1652 	bge_set_hostaddr(&taddr, BGE_RING_DMA_ADDR(sc, bge_rx_return_ring));
1653 	RCB_WRITE_4(sc, rcb_addr, bge_hostaddr.bge_addr_hi, taddr.bge_addr_hi);
1654 	RCB_WRITE_4(sc, rcb_addr, bge_hostaddr.bge_addr_lo, taddr.bge_addr_lo);
1655 	RCB_WRITE_4(sc, rcb_addr, bge_nicaddr, 0x00000000);
1656 	RCB_WRITE_4(sc, rcb_addr, bge_maxlen_flags,
1657 	    BGE_RCB_MAXLEN_FLAGS(sc->bge_return_ring_cnt, 0));
1658 
1659 	/* Set random backoff seed for TX */
1660 	CSR_WRITE_4(sc, BGE_TX_RANDOM_BACKOFF,
1661 	    CLLADDR(ifp->if_sadl)[0] + CLLADDR(ifp->if_sadl)[1] +
1662 	    CLLADDR(ifp->if_sadl)[2] + CLLADDR(ifp->if_sadl)[3] +
1663 	    CLLADDR(ifp->if_sadl)[4] + CLLADDR(ifp->if_sadl)[5] +
1664 	    BGE_TX_BACKOFF_SEED_MASK);
1665 
1666 	/* Set inter-packet gap */
1667 	CSR_WRITE_4(sc, BGE_TX_LENGTHS, 0x2620);
1668 
1669 	/*
1670 	 * Specify which ring to use for packets that don't match
1671 	 * any RX rules.
1672 	 */
1673 	CSR_WRITE_4(sc, BGE_RX_RULES_CFG, 0x08);
1674 
1675 	/*
1676 	 * Configure number of RX lists. One interrupt distribution
1677 	 * list, sixteen active lists, one bad frames class.
1678 	 */
1679 	CSR_WRITE_4(sc, BGE_RXLP_CFG, 0x181);
1680 
1681 	/* Inialize RX list placement stats mask. */
1682 	CSR_WRITE_4(sc, BGE_RXLP_STATS_ENABLE_MASK, 0x007FFFFF);
1683 	CSR_WRITE_4(sc, BGE_RXLP_STATS_CTL, 0x1);
1684 
1685 	/* Disable host coalescing until we get it set up */
1686 	CSR_WRITE_4(sc, BGE_HCC_MODE, 0x00000000);
1687 
1688 	/* Poll to make sure it's shut down. */
1689 	for (i = 0; i < BGE_TIMEOUT; i++) {
1690 		if (!(CSR_READ_4(sc, BGE_HCC_MODE) & BGE_HCCMODE_ENABLE))
1691 			break;
1692 		DELAY(10);
1693 	}
1694 
1695 	if (i == BGE_TIMEOUT) {
1696 		aprint_error_dev(sc->bge_dev,
1697 		    "host coalescing engine failed to idle\n");
1698 		return(ENXIO);
1699 	}
1700 
1701 	/* Set up host coalescing defaults */
1702 	CSR_WRITE_4(sc, BGE_HCC_RX_COAL_TICKS, sc->bge_rx_coal_ticks);
1703 	CSR_WRITE_4(sc, BGE_HCC_TX_COAL_TICKS, sc->bge_tx_coal_ticks);
1704 	CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS, sc->bge_rx_max_coal_bds);
1705 	CSR_WRITE_4(sc, BGE_HCC_TX_MAX_COAL_BDS, sc->bge_tx_max_coal_bds);
1706 	if ((sc->bge_quirks & BGE_QUIRK_5705_CORE) == 0) {
1707 		CSR_WRITE_4(sc, BGE_HCC_RX_COAL_TICKS_INT, 0);
1708 		CSR_WRITE_4(sc, BGE_HCC_TX_COAL_TICKS_INT, 0);
1709 	}
1710 	CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS_INT, 0);
1711 	CSR_WRITE_4(sc, BGE_HCC_TX_MAX_COAL_BDS_INT, 0);
1712 
1713 	/* Set up address of statistics block */
1714 	if ((sc->bge_quirks & BGE_QUIRK_5705_CORE) == 0) {
1715 		bge_set_hostaddr(&taddr,
1716 		    BGE_RING_DMA_ADDR(sc, bge_info.bge_stats));
1717 		CSR_WRITE_4(sc, BGE_HCC_STATS_TICKS, sc->bge_stat_ticks);
1718 		CSR_WRITE_4(sc, BGE_HCC_STATS_BASEADDR, BGE_STATS_BLOCK);
1719 		CSR_WRITE_4(sc, BGE_HCC_STATS_ADDR_HI, taddr.bge_addr_hi);
1720 		CSR_WRITE_4(sc, BGE_HCC_STATS_ADDR_LO, taddr.bge_addr_lo);
1721 	}
1722 
1723 	/* Set up address of status block */
1724 	bge_set_hostaddr(&taddr, BGE_RING_DMA_ADDR(sc, bge_status_block));
1725 	CSR_WRITE_4(sc, BGE_HCC_STATUSBLK_BASEADDR, BGE_STATUS_BLOCK);
1726 	CSR_WRITE_4(sc, BGE_HCC_STATUSBLK_ADDR_HI, taddr.bge_addr_hi);
1727 	CSR_WRITE_4(sc, BGE_HCC_STATUSBLK_ADDR_LO, taddr.bge_addr_lo);
1728 	sc->bge_rdata->bge_status_block.bge_idx[0].bge_rx_prod_idx = 0;
1729 	sc->bge_rdata->bge_status_block.bge_idx[0].bge_tx_cons_idx = 0;
1730 
1731 	/* Turn on host coalescing state machine */
1732 	CSR_WRITE_4(sc, BGE_HCC_MODE, BGE_HCCMODE_ENABLE);
1733 
1734 	/* Turn on RX BD completion state machine and enable attentions */
1735 	CSR_WRITE_4(sc, BGE_RBDC_MODE,
1736 	    BGE_RBDCMODE_ENABLE|BGE_RBDCMODE_ATTN);
1737 
1738 	/* Turn on RX list placement state machine */
1739 	CSR_WRITE_4(sc, BGE_RXLP_MODE, BGE_RXLPMODE_ENABLE);
1740 
1741 	/* Turn on RX list selector state machine. */
1742 	if ((sc->bge_quirks & BGE_QUIRK_5705_CORE) == 0) {
1743 		CSR_WRITE_4(sc, BGE_RXLS_MODE, BGE_RXLSMODE_ENABLE);
1744 	}
1745 
1746 	/* Turn on DMA, clear stats */
1747 	CSR_WRITE_4(sc, BGE_MAC_MODE, BGE_MACMODE_TXDMA_ENB|
1748 	    BGE_MACMODE_RXDMA_ENB|BGE_MACMODE_RX_STATS_CLEAR|
1749 	    BGE_MACMODE_TX_STATS_CLEAR|BGE_MACMODE_RX_STATS_ENB|
1750 	    BGE_MACMODE_TX_STATS_ENB|BGE_MACMODE_FRMHDR_DMA_ENB|
1751 	    (sc->bge_tbi ? BGE_PORTMODE_TBI : BGE_PORTMODE_MII));
1752 
1753 	/* Set misc. local control, enable interrupts on attentions */
1754 	sc->bge_local_ctrl_reg = BGE_MLC_INTR_ONATTN | BGE_MLC_AUTO_EEPROM;
1755 
1756 #ifdef notdef
1757 	/* Assert GPIO pins for PHY reset */
1758 	BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_MISCIO_OUT0|
1759 	    BGE_MLC_MISCIO_OUT1|BGE_MLC_MISCIO_OUT2);
1760 	BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_MISCIO_OUTEN0|
1761 	    BGE_MLC_MISCIO_OUTEN1|BGE_MLC_MISCIO_OUTEN2);
1762 #endif
1763 
1764 #if defined(not_quite_yet)
1765 	/* Linux driver enables enable gpio pin #1 on 5700s */
1766 	if (sc->bge_chipid == BGE_CHIPID_BCM5700) {
1767 		sc->bge_local_ctrl_reg |=
1768 		  (BGE_MLC_MISCIO_OUT1|BGE_MLC_MISCIO_OUTEN1);
1769 	}
1770 #endif
1771 	CSR_WRITE_4(sc, BGE_MISC_LOCAL_CTL, sc->bge_local_ctrl_reg);
1772 
1773 	/* Turn on DMA completion state machine */
1774 	if ((sc->bge_quirks & BGE_QUIRK_5705_CORE) == 0) {
1775 		CSR_WRITE_4(sc, BGE_DMAC_MODE, BGE_DMACMODE_ENABLE);
1776 	}
1777 
1778 	/* Turn on write DMA state machine */
1779 	{
1780 		uint32_t bge_wdma_mode =
1781 			BGE_WDMAMODE_ENABLE|BGE_WDMAMODE_ALL_ATTNS;
1782 
1783 		if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5755 ||
1784 		    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5787)
1785 		  /* Enable host coalescing bug fix; see Linux tg3.c */
1786 		  bge_wdma_mode |= (1 << 29);
1787 
1788 		CSR_WRITE_4(sc, BGE_WDMA_MODE, bge_wdma_mode);
1789         }
1790 
1791 	/* Turn on read DMA state machine */
1792 	{
1793 		uint32_t dma_read_modebits;
1794 
1795 		dma_read_modebits =
1796 		  BGE_RDMAMODE_ENABLE | BGE_RDMAMODE_ALL_ATTNS;
1797 
1798 		if (sc->bge_pcie && 0) {
1799 			dma_read_modebits |= BGE_RDMA_MODE_FIFO_LONG_BURST;
1800 		} else if ((sc->bge_quirks & BGE_QUIRK_5705_CORE)) {
1801 			dma_read_modebits |= BGE_RDMA_MODE_FIFO_SIZE_128;
1802 		}
1803 
1804 		/* XXX broadcom-supplied linux driver; undocumented */
1805 		if (BGE_IS_5750_OR_BEYOND(sc)) {
1806  			/*
1807 			 * XXX: magic values.
1808 			 * From Broadcom-supplied Linux driver;  apparently
1809 			 * required to workaround a DMA bug affecting TSO
1810 			 * on bcm575x/bcm5721?
1811 			 */
1812 			dma_read_modebits |= (1 << 27);
1813 		}
1814 		CSR_WRITE_4(sc, BGE_RDMA_MODE, dma_read_modebits);
1815 	}
1816 
1817 	/* Turn on RX data completion state machine */
1818 	CSR_WRITE_4(sc, BGE_RDC_MODE, BGE_RDCMODE_ENABLE);
1819 
1820 	/* Turn on RX BD initiator state machine */
1821 	CSR_WRITE_4(sc, BGE_RBDI_MODE, BGE_RBDIMODE_ENABLE);
1822 
1823 	/* Turn on RX data and RX BD initiator state machine */
1824 	CSR_WRITE_4(sc, BGE_RDBDI_MODE, BGE_RDBDIMODE_ENABLE);
1825 
1826 	/* Turn on Mbuf cluster free state machine */
1827 	if ((sc->bge_quirks & BGE_QUIRK_5705_CORE) == 0) {
1828 		CSR_WRITE_4(sc, BGE_MBCF_MODE, BGE_MBCFMODE_ENABLE);
1829 	}
1830 
1831 	/* Turn on send BD completion state machine */
1832 	CSR_WRITE_4(sc, BGE_SBDC_MODE, BGE_SBDCMODE_ENABLE);
1833 
1834 	/* Turn on send data completion state machine */
1835 	CSR_WRITE_4(sc, BGE_SDC_MODE, BGE_SDCMODE_ENABLE);
1836 
1837 	/* Turn on send data initiator state machine */
1838 	if (BGE_IS_5750_OR_BEYOND(sc)) {
1839 		/* XXX: magic value from Linux driver */
1840 		CSR_WRITE_4(sc, BGE_SDI_MODE, BGE_SDIMODE_ENABLE | 0x08);
1841 	} else {
1842 		CSR_WRITE_4(sc, BGE_SDI_MODE, BGE_SDIMODE_ENABLE);
1843 	}
1844 
1845 	/* Turn on send BD initiator state machine */
1846 	CSR_WRITE_4(sc, BGE_SBDI_MODE, BGE_SBDIMODE_ENABLE);
1847 
1848 	/* Turn on send BD selector state machine */
1849 	CSR_WRITE_4(sc, BGE_SRS_MODE, BGE_SRSMODE_ENABLE);
1850 
1851 	CSR_WRITE_4(sc, BGE_SDI_STATS_ENABLE_MASK, 0x007FFFFF);
1852 	CSR_WRITE_4(sc, BGE_SDI_STATS_CTL,
1853 	    BGE_SDISTATSCTL_ENABLE|BGE_SDISTATSCTL_FASTER);
1854 
1855 	/* ack/clear link change events */
1856 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED|
1857 	    BGE_MACSTAT_CFG_CHANGED);
1858 	CSR_WRITE_4(sc, BGE_MI_STS, 0);
1859 
1860 	/* Enable PHY auto polling (for MII/GMII only) */
1861 	if (sc->bge_tbi) {
1862 		CSR_WRITE_4(sc, BGE_MI_STS, BGE_MISTS_LINK);
1863  	} else {
1864 		BGE_SETBIT(sc, BGE_MI_MODE, BGE_MIMODE_AUTOPOLL|10<<16);
1865 		if (sc->bge_quirks & BGE_QUIRK_LINK_STATE_BROKEN)
1866 			CSR_WRITE_4(sc, BGE_MAC_EVT_ENB,
1867 			    BGE_EVTENB_MI_INTERRUPT);
1868 	}
1869 
1870 	/* Enable link state change attentions. */
1871 	BGE_SETBIT(sc, BGE_MAC_EVT_ENB, BGE_EVTENB_LINK_CHANGED);
1872 
1873 	return(0);
1874 }
1875 
1876 static const struct bge_revision {
1877 	uint32_t		br_chipid;
1878 	uint32_t		br_quirks;
1879 	const char		*br_name;
1880 } bge_revisions[] = {
1881 	{ BGE_CHIPID_BCM5700_A0,
1882 	  BGE_QUIRK_LINK_STATE_BROKEN,
1883 	  "BCM5700 A0" },
1884 
1885 	{ BGE_CHIPID_BCM5700_A1,
1886 	  BGE_QUIRK_LINK_STATE_BROKEN,
1887 	  "BCM5700 A1" },
1888 
1889 	{ BGE_CHIPID_BCM5700_B0,
1890 	  BGE_QUIRK_LINK_STATE_BROKEN|BGE_QUIRK_CSUM_BROKEN|BGE_QUIRK_5700_COMMON,
1891 	  "BCM5700 B0" },
1892 
1893 	{ BGE_CHIPID_BCM5700_B1,
1894 	  BGE_QUIRK_LINK_STATE_BROKEN|BGE_QUIRK_5700_COMMON,
1895 	  "BCM5700 B1" },
1896 
1897 	{ BGE_CHIPID_BCM5700_B2,
1898 	  BGE_QUIRK_LINK_STATE_BROKEN|BGE_QUIRK_5700_COMMON,
1899 	  "BCM5700 B2" },
1900 
1901 	{ BGE_CHIPID_BCM5700_B3,
1902 	  BGE_QUIRK_LINK_STATE_BROKEN|BGE_QUIRK_5700_COMMON,
1903 	  "BCM5700 B3" },
1904 
1905 	/* This is treated like a BCM5700 Bx */
1906 	{ BGE_CHIPID_BCM5700_ALTIMA,
1907 	  BGE_QUIRK_LINK_STATE_BROKEN|BGE_QUIRK_5700_COMMON,
1908 	  "BCM5700 Altima" },
1909 
1910 	{ BGE_CHIPID_BCM5700_C0,
1911 	  0,
1912 	  "BCM5700 C0" },
1913 
1914 	{ BGE_CHIPID_BCM5701_A0,
1915 	  0, /*XXX really, just not known */
1916 	  "BCM5701 A0" },
1917 
1918 	{ BGE_CHIPID_BCM5701_B0,
1919 	  BGE_QUIRK_PCIX_DMA_ALIGN_BUG,
1920 	  "BCM5701 B0" },
1921 
1922 	{ BGE_CHIPID_BCM5701_B2,
1923 	  BGE_QUIRK_ONLY_PHY_1|BGE_QUIRK_PCIX_DMA_ALIGN_BUG,
1924 	  "BCM5701 B2" },
1925 
1926 	{ BGE_CHIPID_BCM5701_B5,
1927 	  BGE_QUIRK_ONLY_PHY_1|BGE_QUIRK_PCIX_DMA_ALIGN_BUG,
1928 	  "BCM5701 B5" },
1929 
1930 	{ BGE_CHIPID_BCM5703_A0,
1931 	  0,
1932 	  "BCM5703 A0" },
1933 
1934 	{ BGE_CHIPID_BCM5703_A1,
1935 	  0,
1936 	  "BCM5703 A1" },
1937 
1938 	{ BGE_CHIPID_BCM5703_A2,
1939 	  BGE_QUIRK_ONLY_PHY_1,
1940 	  "BCM5703 A2" },
1941 
1942 	{ BGE_CHIPID_BCM5703_A3,
1943 	  BGE_QUIRK_ONLY_PHY_1,
1944 	  "BCM5703 A3" },
1945 
1946 	{ BGE_CHIPID_BCM5703_B0,
1947 	  BGE_QUIRK_ONLY_PHY_1,
1948 	  "BCM5703 B0" },
1949 
1950 	{ BGE_CHIPID_BCM5704_A0,
1951   	  BGE_QUIRK_ONLY_PHY_1|BGE_QUIRK_FEWER_MBUFS,
1952 	  "BCM5704 A0" },
1953 
1954 	{ BGE_CHIPID_BCM5704_A1,
1955   	  BGE_QUIRK_ONLY_PHY_1|BGE_QUIRK_FEWER_MBUFS,
1956 	  "BCM5704 A1" },
1957 
1958 	{ BGE_CHIPID_BCM5704_A2,
1959   	  BGE_QUIRK_ONLY_PHY_1|BGE_QUIRK_FEWER_MBUFS,
1960 	  "BCM5704 A2" },
1961 
1962 	{ BGE_CHIPID_BCM5704_A3,
1963   	  BGE_QUIRK_ONLY_PHY_1|BGE_QUIRK_FEWER_MBUFS,
1964 	  "BCM5704 A3" },
1965 
1966 	{ BGE_CHIPID_BCM5705_A0,
1967 	  BGE_QUIRK_ONLY_PHY_1|BGE_QUIRK_5705_CORE,
1968 	  "BCM5705 A0" },
1969 
1970 	{ BGE_CHIPID_BCM5705_A1,
1971 	  BGE_QUIRK_ONLY_PHY_1|BGE_QUIRK_5705_CORE,
1972 	  "BCM5705 A1" },
1973 
1974 	{ BGE_CHIPID_BCM5705_A2,
1975 	  BGE_QUIRK_ONLY_PHY_1|BGE_QUIRK_5705_CORE,
1976 	  "BCM5705 A2" },
1977 
1978 	{ BGE_CHIPID_BCM5705_A3,
1979 	  BGE_QUIRK_ONLY_PHY_1|BGE_QUIRK_5705_CORE,
1980 	  "BCM5705 A3" },
1981 
1982 	{ BGE_CHIPID_BCM5750_A0,
1983 	  BGE_QUIRK_ONLY_PHY_1|BGE_QUIRK_5705_CORE,
1984 	  "BCM5750 A0" },
1985 
1986 	{ BGE_CHIPID_BCM5750_A1,
1987 	  BGE_QUIRK_ONLY_PHY_1|BGE_QUIRK_5705_CORE,
1988 	  "BCM5750 A1" },
1989 
1990 	{ BGE_CHIPID_BCM5751_A1,
1991 	  BGE_QUIRK_ONLY_PHY_1|BGE_QUIRK_5705_CORE,
1992 	  "BCM5751 A1" },
1993 
1994 	{ BGE_CHIPID_BCM5752_A0,
1995 	  BGE_QUIRK_ONLY_PHY_1|BGE_QUIRK_5705_CORE,
1996 	  "BCM5752 A0" },
1997 
1998 	{ BGE_CHIPID_BCM5752_A1,
1999 	  BGE_QUIRK_ONLY_PHY_1|BGE_QUIRK_5705_CORE,
2000 	  "BCM5752 A1" },
2001 
2002 	{ BGE_CHIPID_BCM5752_A2,
2003 	  BGE_QUIRK_ONLY_PHY_1|BGE_QUIRK_5705_CORE,
2004 	  "BCM5752 A2" },
2005 
2006 	{ BGE_CHIPID_BCM5787_A0,
2007 	  BGE_QUIRK_ONLY_PHY_1|BGE_QUIRK_5705_CORE,
2008 	  "BCM5754/5787 A0" },
2009 
2010 	{ BGE_CHIPID_BCM5787_A1,
2011 	  BGE_QUIRK_ONLY_PHY_1|BGE_QUIRK_5705_CORE,
2012 	  "BCM5754/5787 A1" },
2013 
2014 	{ BGE_CHIPID_BCM5787_A2,
2015 	  BGE_QUIRK_ONLY_PHY_1|BGE_QUIRK_5705_CORE,
2016 	  "BCM5754/5787 A2" },
2017 
2018 	{ 0, 0, NULL }
2019 };
2020 
2021 /*
2022  * Some defaults for major revisions, so that newer steppings
2023  * that we don't know about have a shot at working.
2024  */
2025 static const struct bge_revision bge_majorrevs[] = {
2026 	{ BGE_ASICREV_BCM5700,
2027 	  BGE_QUIRK_LINK_STATE_BROKEN,
2028 	  "unknown BCM5700" },
2029 
2030 	{ BGE_ASICREV_BCM5701,
2031 	  BGE_QUIRK_PCIX_DMA_ALIGN_BUG,
2032 	  "unknown BCM5701" },
2033 
2034 	{ BGE_ASICREV_BCM5703,
2035 	  0,
2036 	  "unknown BCM5703" },
2037 
2038 	{ BGE_ASICREV_BCM5704,
2039 	  BGE_QUIRK_ONLY_PHY_1,
2040 	  "unknown BCM5704" },
2041 
2042 	{ BGE_ASICREV_BCM5705,
2043 	  BGE_QUIRK_ONLY_PHY_1|BGE_QUIRK_5705_CORE,
2044 	  "unknown BCM5705" },
2045 
2046 	{ BGE_ASICREV_BCM5750,
2047 	  BGE_QUIRK_ONLY_PHY_1|BGE_QUIRK_5705_CORE,
2048 	  "unknown BCM575x family" },
2049 
2050 	{ BGE_ASICREV_BCM5714_A0,
2051 	  BGE_QUIRK_ONLY_PHY_1|BGE_QUIRK_5705_CORE,
2052 	  "unknown BCM5714" },
2053 
2054 	{ BGE_ASICREV_BCM5714,
2055 	  BGE_QUIRK_ONLY_PHY_1|BGE_QUIRK_5705_CORE,
2056 	  "unknown BCM5714" },
2057 
2058 	{ BGE_ASICREV_BCM5752,
2059 	  BGE_QUIRK_ONLY_PHY_1|BGE_QUIRK_5705_CORE,
2060 	  "unknown BCM5752 family" },
2061 
2062 	{ BGE_ASICREV_BCM5755,
2063 	  BGE_QUIRK_ONLY_PHY_1|BGE_QUIRK_5705_CORE,
2064 	  "unknown BCM5755" },
2065 
2066 	{ BGE_ASICREV_BCM5780,
2067 	  BGE_QUIRK_ONLY_PHY_1|BGE_QUIRK_5705_CORE,
2068 	  "unknown BCM5780" },
2069 
2070 	{ BGE_ASICREV_BCM5787,
2071 	  BGE_QUIRK_ONLY_PHY_1|BGE_QUIRK_5705_CORE,
2072 	  "unknown BCM5787" },
2073 
2074 	{ 0,
2075 	  0,
2076 	  NULL }
2077 };
2078 
2079 
2080 static const struct bge_revision *
2081 bge_lookup_rev(uint32_t chipid)
2082 {
2083 	const struct bge_revision *br;
2084 
2085 	for (br = bge_revisions; br->br_name != NULL; br++) {
2086 		if (br->br_chipid == chipid)
2087 			return (br);
2088 	}
2089 
2090 	for (br = bge_majorrevs; br->br_name != NULL; br++) {
2091 		if (br->br_chipid == BGE_ASICREV(chipid))
2092 			return (br);
2093 	}
2094 
2095 	return (NULL);
2096 }
2097 
2098 static const struct bge_product {
2099 	pci_vendor_id_t		bp_vendor;
2100 	pci_product_id_t	bp_product;
2101 	const char		*bp_name;
2102 } bge_products[] = {
2103 	/*
2104 	 * The BCM5700 documentation seems to indicate that the hardware
2105 	 * still has the Alteon vendor ID burned into it, though it
2106 	 * should always be overridden by the value in the EEPROM.  We'll
2107 	 * check for it anyway.
2108 	 */
2109 	{ PCI_VENDOR_ALTEON,
2110 	  PCI_PRODUCT_ALTEON_BCM5700,
2111 	  "Broadcom BCM5700 Gigabit Ethernet",
2112 	  },
2113 	{ PCI_VENDOR_ALTEON,
2114 	  PCI_PRODUCT_ALTEON_BCM5701,
2115 	  "Broadcom BCM5701 Gigabit Ethernet",
2116 	  },
2117 
2118 	{ PCI_VENDOR_ALTIMA,
2119 	  PCI_PRODUCT_ALTIMA_AC1000,
2120 	  "Altima AC1000 Gigabit Ethernet",
2121 	  },
2122 	{ PCI_VENDOR_ALTIMA,
2123 	  PCI_PRODUCT_ALTIMA_AC1001,
2124 	  "Altima AC1001 Gigabit Ethernet",
2125 	   },
2126 	{ PCI_VENDOR_ALTIMA,
2127 	  PCI_PRODUCT_ALTIMA_AC9100,
2128 	  "Altima AC9100 Gigabit Ethernet",
2129 	  },
2130 
2131 	{ PCI_VENDOR_BROADCOM,
2132 	  PCI_PRODUCT_BROADCOM_BCM5700,
2133 	  "Broadcom BCM5700 Gigabit Ethernet",
2134 	  },
2135 	{ PCI_VENDOR_BROADCOM,
2136 	  PCI_PRODUCT_BROADCOM_BCM5701,
2137 	  "Broadcom BCM5701 Gigabit Ethernet",
2138 	  },
2139 	{ PCI_VENDOR_BROADCOM,
2140 	  PCI_PRODUCT_BROADCOM_BCM5702,
2141 	  "Broadcom BCM5702 Gigabit Ethernet",
2142 	  },
2143 	{ PCI_VENDOR_BROADCOM,
2144 	  PCI_PRODUCT_BROADCOM_BCM5702X,
2145 	  "Broadcom BCM5702X Gigabit Ethernet" },
2146 
2147 	{ PCI_VENDOR_BROADCOM,
2148 	  PCI_PRODUCT_BROADCOM_BCM5703,
2149 	  "Broadcom BCM5703 Gigabit Ethernet",
2150 	  },
2151 	{ PCI_VENDOR_BROADCOM,
2152 	  PCI_PRODUCT_BROADCOM_BCM5703X,
2153 	  "Broadcom BCM5703X Gigabit Ethernet",
2154 	  },
2155 	{ PCI_VENDOR_BROADCOM,
2156 	  PCI_PRODUCT_BROADCOM_BCM5703_ALT,
2157 	  "Broadcom BCM5703 Gigabit Ethernet",
2158 	  },
2159 
2160    	{ PCI_VENDOR_BROADCOM,
2161 	  PCI_PRODUCT_BROADCOM_BCM5704C,
2162 	  "Broadcom BCM5704C Dual Gigabit Ethernet",
2163 	  },
2164    	{ PCI_VENDOR_BROADCOM,
2165 	  PCI_PRODUCT_BROADCOM_BCM5704S,
2166 	  "Broadcom BCM5704S Dual Gigabit Ethernet",
2167 	  },
2168 
2169    	{ PCI_VENDOR_BROADCOM,
2170 	  PCI_PRODUCT_BROADCOM_BCM5705,
2171 	  "Broadcom BCM5705 Gigabit Ethernet",
2172 	  },
2173    	{ PCI_VENDOR_BROADCOM,
2174 	  PCI_PRODUCT_BROADCOM_BCM5705K,
2175 	  "Broadcom BCM5705K Gigabit Ethernet",
2176 	  },
2177    	{ PCI_VENDOR_BROADCOM,
2178 	  PCI_PRODUCT_BROADCOM_BCM5705M,
2179 	  "Broadcom BCM5705M Gigabit Ethernet",
2180 	  },
2181    	{ PCI_VENDOR_BROADCOM,
2182 	  PCI_PRODUCT_BROADCOM_BCM5705M_ALT,
2183 	  "Broadcom BCM5705M Gigabit Ethernet",
2184 	  },
2185 
2186 	{ PCI_VENDOR_BROADCOM,
2187 	  PCI_PRODUCT_BROADCOM_BCM5714,
2188 	  "Broadcom BCM5714/5715 Gigabit Ethernet",
2189 	  },
2190 	{ PCI_VENDOR_BROADCOM,
2191 	  PCI_PRODUCT_BROADCOM_BCM5715,
2192 	  "Broadcom BCM5714/5715 Gigabit Ethernet",
2193 	  },
2194 	{ PCI_VENDOR_BROADCOM,
2195 	  PCI_PRODUCT_BROADCOM_BCM5789,
2196 	  "Broadcom BCM5789 Gigabit Ethernet",
2197 	  },
2198 
2199 	{ PCI_VENDOR_BROADCOM,
2200 	  PCI_PRODUCT_BROADCOM_BCM5721,
2201 	  "Broadcom BCM5721 Gigabit Ethernet",
2202 	  },
2203 
2204 	{ PCI_VENDOR_BROADCOM,
2205 	  PCI_PRODUCT_BROADCOM_BCM5750,
2206 	  "Broadcom BCM5750 Gigabit Ethernet",
2207 	  },
2208 
2209 	{ PCI_VENDOR_BROADCOM,
2210 	  PCI_PRODUCT_BROADCOM_BCM5750M,
2211 	  "Broadcom BCM5750M Gigabit Ethernet",
2212 	  },
2213 
2214 	{ PCI_VENDOR_BROADCOM,
2215 	  PCI_PRODUCT_BROADCOM_BCM5751,
2216 	  "Broadcom BCM5751 Gigabit Ethernet",
2217 	  },
2218 
2219 	{ PCI_VENDOR_BROADCOM,
2220 	  PCI_PRODUCT_BROADCOM_BCM5751M,
2221 	  "Broadcom BCM5751M Gigabit Ethernet",
2222 	  },
2223 
2224 	{ PCI_VENDOR_BROADCOM,
2225 	  PCI_PRODUCT_BROADCOM_BCM5752,
2226 	  "Broadcom BCM5752 Gigabit Ethernet",
2227 	  },
2228 
2229 	{ PCI_VENDOR_BROADCOM,
2230 	  PCI_PRODUCT_BROADCOM_BCM5752M,
2231 	  "Broadcom BCM5752M Gigabit Ethernet",
2232 	  },
2233 
2234 	{ PCI_VENDOR_BROADCOM,
2235 	  PCI_PRODUCT_BROADCOM_BCM5753,
2236 	  "Broadcom BCM5753 Gigabit Ethernet",
2237 	  },
2238 
2239 	{ PCI_VENDOR_BROADCOM,
2240 	  PCI_PRODUCT_BROADCOM_BCM5753M,
2241 	  "Broadcom BCM5753M Gigabit Ethernet",
2242 	  },
2243 
2244 	{ PCI_VENDOR_BROADCOM,
2245 	  PCI_PRODUCT_BROADCOM_BCM5754,
2246 	  "Broadcom BCM5754 Gigabit Ethernet",
2247 	},
2248 
2249 	{ PCI_VENDOR_BROADCOM,
2250 	  PCI_PRODUCT_BROADCOM_BCM5754M,
2251 	  "Broadcom BCM5754M Gigabit Ethernet",
2252 	},
2253 
2254 	{ PCI_VENDOR_BROADCOM,
2255 	  PCI_PRODUCT_BROADCOM_BCM5755,
2256 	  "Broadcom BCM5755 Gigabit Ethernet",
2257 	},
2258 
2259 	{ PCI_VENDOR_BROADCOM,
2260 	  PCI_PRODUCT_BROADCOM_BCM5755M,
2261 	  "Broadcom BCM5755M Gigabit Ethernet",
2262 	},
2263 
2264    	{ PCI_VENDOR_BROADCOM,
2265 	  PCI_PRODUCT_BROADCOM_BCM5780,
2266 	  "Broadcom BCM5780 Gigabit Ethernet",
2267 	  },
2268 
2269    	{ PCI_VENDOR_BROADCOM,
2270 	  PCI_PRODUCT_BROADCOM_BCM5780S,
2271 	  "Broadcom BCM5780S Gigabit Ethernet",
2272 	  },
2273 
2274    	{ PCI_VENDOR_BROADCOM,
2275 	  PCI_PRODUCT_BROADCOM_BCM5782,
2276 	  "Broadcom BCM5782 Gigabit Ethernet",
2277 	},
2278 
2279 	{ PCI_VENDOR_BROADCOM,
2280 	  PCI_PRODUCT_BROADCOM_BCM5786,
2281 	  "Broadcom BCM5786 Gigabit Ethernet",
2282 	},
2283 
2284 	{ PCI_VENDOR_BROADCOM,
2285 	  PCI_PRODUCT_BROADCOM_BCM5787,
2286 	  "Broadcom BCM5787 Gigabit Ethernet",
2287 	},
2288 
2289 	{ PCI_VENDOR_BROADCOM,
2290 	  PCI_PRODUCT_BROADCOM_BCM5787M,
2291 	  "Broadcom BCM5787M Gigabit Ethernet",
2292 	},
2293 
2294    	{ PCI_VENDOR_BROADCOM,
2295 	  PCI_PRODUCT_BROADCOM_BCM5788,
2296 	  "Broadcom BCM5788 Gigabit Ethernet",
2297 	  },
2298    	{ PCI_VENDOR_BROADCOM,
2299 	  PCI_PRODUCT_BROADCOM_BCM5789,
2300 	  "Broadcom BCM5789 Gigabit Ethernet",
2301 	  },
2302 
2303    	{ PCI_VENDOR_BROADCOM,
2304 	  PCI_PRODUCT_BROADCOM_BCM5901,
2305 	  "Broadcom BCM5901 Fast Ethernet",
2306 	  },
2307    	{ PCI_VENDOR_BROADCOM,
2308 	  PCI_PRODUCT_BROADCOM_BCM5901A2,
2309 	  "Broadcom BCM5901A2 Fast Ethernet",
2310 	  },
2311 
2312 	{ PCI_VENDOR_SCHNEIDERKOCH,
2313 	  PCI_PRODUCT_SCHNEIDERKOCH_SK_9DX1,
2314 	  "SysKonnect SK-9Dx1 Gigabit Ethernet",
2315 	  },
2316 
2317 	{ PCI_VENDOR_3COM,
2318 	  PCI_PRODUCT_3COM_3C996,
2319 	  "3Com 3c996 Gigabit Ethernet",
2320 	  },
2321 
2322 	{ 0,
2323 	  0,
2324 	  NULL },
2325 };
2326 
2327 static const struct bge_product *
2328 bge_lookup(const struct pci_attach_args *pa)
2329 {
2330 	const struct bge_product *bp;
2331 
2332 	for (bp = bge_products; bp->bp_name != NULL; bp++) {
2333 		if (PCI_VENDOR(pa->pa_id) == bp->bp_vendor &&
2334 		    PCI_PRODUCT(pa->pa_id) == bp->bp_product)
2335 			return (bp);
2336 	}
2337 
2338 	return (NULL);
2339 }
2340 
2341 static int
2342 bge_setpowerstate(struct bge_softc *sc, int powerlevel)
2343 {
2344 #ifdef NOTYET
2345 	u_int32_t pm_ctl = 0;
2346 
2347 	/* XXX FIXME: make sure indirect accesses enabled? */
2348 	pm_ctl = pci_conf_read(sc->bge_dev, BGE_PCI_MISC_CTL, 4);
2349 	pm_ctl |= BGE_PCIMISCCTL_INDIRECT_ACCESS;
2350 	pci_write_config(sc->bge_dev, BGE_PCI_MISC_CTL, pm_ctl, 4);
2351 
2352 	/* clear the PME_assert bit and power state bits, enable PME */
2353 	pm_ctl = pci_conf_read(sc->bge_dev, BGE_PCI_PWRMGMT_CMD, 2);
2354 	pm_ctl &= ~PCIM_PSTAT_DMASK;
2355 	pm_ctl |= (1 << 8);
2356 
2357 	if (powerlevel == 0) {
2358 		pm_ctl |= PCIM_PSTAT_D0;
2359 		pci_write_config(sc->bge_dev, BGE_PCI_PWRMGMT_CMD,
2360 		    pm_ctl, 2);
2361 		DELAY(10000);
2362 		CSR_WRITE_4(sc, BGE_MISC_LOCAL_CTL, sc->bge_local_ctrl_reg);
2363 		DELAY(10000);
2364 
2365 #ifdef NOTYET
2366 		/* XXX FIXME: write 0x02 to phy aux_Ctrl reg */
2367 		bge_miibus_writereg(sc->bge_dev, 1, 0x18, 0x02);
2368 #endif
2369 		DELAY(40); DELAY(40); DELAY(40);
2370 		DELAY(10000);	/* above not quite adequate on 5700 */
2371 		return 0;
2372 	}
2373 
2374 
2375 	/*
2376 	 * Entering ACPI power states D1-D3 is achieved by wiggling
2377 	 * GMII gpio pins. Example code assumes all hardware vendors
2378 	 * followed Broadom's sample pcb layout. Until we verify that
2379 	 * for all supported OEM cards, states D1-D3 are  unsupported.
2380 	 */
2381 	aprint_error_dev(sc->bge_dev,
2382 	    "power state %d unimplemented; check GPIO pins\n",
2383 	    powerlevel);
2384 #endif
2385 	return EOPNOTSUPP;
2386 }
2387 
2388 
2389 /*
2390  * Probe for a Broadcom chip. Check the PCI vendor and device IDs
2391  * against our list and return its name if we find a match. Note
2392  * that since the Broadcom controller contains VPD support, we
2393  * can get the device name string from the controller itself instead
2394  * of the compiled-in string. This is a little slow, but it guarantees
2395  * we'll always announce the right product name.
2396  */
2397 static int
2398 bge_probe(device_t parent, cfdata_t match, void *aux)
2399 {
2400 	struct pci_attach_args *pa = (struct pci_attach_args *)aux;
2401 
2402 	if (bge_lookup(pa) != NULL)
2403 		return (1);
2404 
2405 	return (0);
2406 }
2407 
2408 static void
2409 bge_attach(device_t parent, device_t self, void *aux)
2410 {
2411 	struct bge_softc	*sc = device_private(self);
2412 	struct pci_attach_args	*pa = aux;
2413 	const struct bge_product *bp;
2414 	const struct bge_revision *br;
2415 	pci_chipset_tag_t	pc = sc->sc_pc;
2416 	pci_intr_handle_t	ih;
2417 	const char		*intrstr = NULL;
2418 	bus_dma_segment_t	seg;
2419 	int			rseg;
2420 	u_int32_t		hwcfg = 0;
2421 	u_int32_t		mac_addr = 0;
2422 	u_int32_t		command;
2423 	struct ifnet		*ifp;
2424 	void *			kva;
2425 	u_char			eaddr[ETHER_ADDR_LEN];
2426 	pcireg_t		memtype;
2427 	bus_addr_t		memaddr;
2428 	bus_size_t		memsize;
2429 	u_int32_t		pm_ctl;
2430 
2431 	bp = bge_lookup(pa);
2432 	KASSERT(bp != NULL);
2433 
2434 	sc->sc_pc = pa->pa_pc;
2435 	sc->sc_pcitag = pa->pa_tag;
2436 	sc->bge_dev = self;
2437 
2438 	aprint_naive(": Ethernet controller\n");
2439 	aprint_normal(": %s\n", bp->bp_name);
2440 
2441 	/*
2442 	 * Map control/status registers.
2443 	 */
2444 	DPRINTFN(5, ("Map control/status regs\n"));
2445 	command = pci_conf_read(pc, sc->sc_pcitag, PCI_COMMAND_STATUS_REG);
2446 	command |= PCI_COMMAND_MEM_ENABLE | PCI_COMMAND_MASTER_ENABLE;
2447 	pci_conf_write(pc, sc->sc_pcitag, PCI_COMMAND_STATUS_REG, command);
2448 	command = pci_conf_read(pc, sc->sc_pcitag, PCI_COMMAND_STATUS_REG);
2449 
2450 	if (!(command & PCI_COMMAND_MEM_ENABLE)) {
2451 		aprint_error_dev(sc->bge_dev,
2452 		    "failed to enable memory mapping!\n");
2453 		return;
2454 	}
2455 
2456 	DPRINTFN(5, ("pci_mem_find\n"));
2457 	memtype = pci_mapreg_type(sc->sc_pc, sc->sc_pcitag, BGE_PCI_BAR0);
2458  	switch (memtype) {
2459 	case PCI_MAPREG_TYPE_MEM | PCI_MAPREG_MEM_TYPE_32BIT:
2460 	case PCI_MAPREG_TYPE_MEM | PCI_MAPREG_MEM_TYPE_64BIT:
2461 		if (pci_mapreg_map(pa, BGE_PCI_BAR0,
2462 		    memtype, 0, &sc->bge_btag, &sc->bge_bhandle,
2463 		    &memaddr, &memsize) == 0)
2464 			break;
2465 	default:
2466 		aprint_error_dev(sc->bge_dev, "can't find mem space\n");
2467 		return;
2468 	}
2469 
2470 	DPRINTFN(5, ("pci_intr_map\n"));
2471 	if (pci_intr_map(pa, &ih)) {
2472 		aprint_error_dev(sc->bge_dev, "couldn't map interrupt\n");
2473 		return;
2474 	}
2475 
2476 	DPRINTFN(5, ("pci_intr_string\n"));
2477 	intrstr = pci_intr_string(pc, ih);
2478 
2479 	DPRINTFN(5, ("pci_intr_establish\n"));
2480 	sc->bge_intrhand = pci_intr_establish(pc, ih, IPL_NET, bge_intr, sc);
2481 
2482 	if (sc->bge_intrhand == NULL) {
2483 		aprint_error_dev(sc->bge_dev,
2484 		    "couldn't establish interrupt%s%s\n",
2485 		    intrstr ? " at " : "", intrstr ? intrstr : "");
2486 		return;
2487 	}
2488 	aprint_normal_dev(sc->bge_dev, "interrupting at %s\n", intrstr);
2489 
2490 	/*
2491 	 * Kludge for 5700 Bx bug: a hardware bug (PCIX byte enable?)
2492 	 * can clobber the chip's PCI config-space power control registers,
2493 	 * leaving the card in D3 powersave state.
2494 	 * We do not have memory-mapped registers in this state,
2495 	 * so force device into D0 state before starting initialization.
2496 	 */
2497 	pm_ctl = pci_conf_read(pc, sc->sc_pcitag, BGE_PCI_PWRMGMT_CMD);
2498 	pm_ctl &= ~(PCI_PWR_D0|PCI_PWR_D1|PCI_PWR_D2|PCI_PWR_D3);
2499 	pm_ctl |= (1 << 8) | PCI_PWR_D0 ; /* D0 state */
2500 	pci_conf_write(pc, sc->sc_pcitag, BGE_PCI_PWRMGMT_CMD, pm_ctl);
2501 	DELAY(1000);	/* 27 usec is allegedly sufficent */
2502 
2503 	/*
2504 	 * Save ASIC rev.  Look up any quirks associated with this
2505 	 * ASIC.
2506 	 */
2507 	sc->bge_chipid =
2508 	    pci_conf_read(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MISC_CTL) &
2509 	    BGE_PCIMISCCTL_ASICREV;
2510 
2511 	/*
2512 	 * Detect PCI-Express devices
2513 	 * XXX: guessed from Linux/FreeBSD; no documentation
2514 	 */
2515 	if (pci_get_capability(sc->sc_pc, sc->sc_pcitag, PCI_CAP_PCIEXPRESS,
2516 	        NULL, NULL) != 0)
2517 		sc->bge_pcie = 1;
2518 	else
2519 		sc->bge_pcie = 0;
2520 
2521 	/* Try to reset the chip. */
2522 	DPRINTFN(5, ("bge_reset\n"));
2523 	bge_reset(sc);
2524 
2525 	if (bge_chipinit(sc)) {
2526 		aprint_error_dev(sc->bge_dev, "chip initialization failed\n");
2527 		bge_release_resources(sc);
2528 		return;
2529 	}
2530 
2531 	/*
2532 	 * Get station address from the EEPROM.
2533 	 */
2534 	mac_addr = bge_readmem_ind(sc, 0x0c14);
2535 	if ((mac_addr >> 16) == 0x484b) {
2536 		eaddr[0] = (u_char)(mac_addr >> 8);
2537 		eaddr[1] = (u_char)(mac_addr >> 0);
2538 		mac_addr = bge_readmem_ind(sc, 0x0c18);
2539 		eaddr[2] = (u_char)(mac_addr >> 24);
2540 		eaddr[3] = (u_char)(mac_addr >> 16);
2541 		eaddr[4] = (u_char)(mac_addr >> 8);
2542 		eaddr[5] = (u_char)(mac_addr >> 0);
2543 	} else if (bge_read_eeprom(sc, (void *)eaddr,
2544 	    BGE_EE_MAC_OFFSET + 2, ETHER_ADDR_LEN)) {
2545 		aprint_error_dev(sc->bge_dev,
2546 		    "failed to read station address\n");
2547 		bge_release_resources(sc);
2548 		return;
2549 	}
2550 
2551 	br = bge_lookup_rev(sc->bge_chipid);
2552 
2553 	if (br == NULL) {
2554 		aprint_normal_dev(sc->bge_dev, "unknown ASIC (0x%04x)",
2555 		    sc->bge_chipid >> 16);
2556 		sc->bge_quirks = 0;
2557 	} else {
2558 		aprint_normal_dev(sc->bge_dev, "ASIC %s (0x%04x)",
2559 		    br->br_name, sc->bge_chipid >> 16);
2560 		sc->bge_quirks |= br->br_quirks;
2561 	}
2562 	aprint_normal(", Ethernet address %s\n", ether_sprintf(eaddr));
2563 
2564 	/* Allocate the general information block and ring buffers. */
2565 	if (pci_dma64_available(pa))
2566 		sc->bge_dmatag = pa->pa_dmat64;
2567 	else
2568 		sc->bge_dmatag = pa->pa_dmat;
2569 	DPRINTFN(5, ("bus_dmamem_alloc\n"));
2570 	if (bus_dmamem_alloc(sc->bge_dmatag, sizeof(struct bge_ring_data),
2571 			     PAGE_SIZE, 0, &seg, 1, &rseg, BUS_DMA_NOWAIT)) {
2572 		aprint_error_dev(sc->bge_dev, "can't alloc rx buffers\n");
2573 		return;
2574 	}
2575 	DPRINTFN(5, ("bus_dmamem_map\n"));
2576 	if (bus_dmamem_map(sc->bge_dmatag, &seg, rseg,
2577 			   sizeof(struct bge_ring_data), &kva,
2578 			   BUS_DMA_NOWAIT)) {
2579 		aprint_error_dev(sc->bge_dev,
2580 		    "can't map DMA buffers (%zu bytes)\n",
2581 		    sizeof(struct bge_ring_data));
2582 		bus_dmamem_free(sc->bge_dmatag, &seg, rseg);
2583 		return;
2584 	}
2585 	DPRINTFN(5, ("bus_dmamem_create\n"));
2586 	if (bus_dmamap_create(sc->bge_dmatag, sizeof(struct bge_ring_data), 1,
2587 	    sizeof(struct bge_ring_data), 0,
2588 	    BUS_DMA_NOWAIT, &sc->bge_ring_map)) {
2589 		aprint_error_dev(sc->bge_dev, "can't create DMA map\n");
2590 		bus_dmamem_unmap(sc->bge_dmatag, kva,
2591 				 sizeof(struct bge_ring_data));
2592 		bus_dmamem_free(sc->bge_dmatag, &seg, rseg);
2593 		return;
2594 	}
2595 	DPRINTFN(5, ("bus_dmamem_load\n"));
2596 	if (bus_dmamap_load(sc->bge_dmatag, sc->bge_ring_map, kva,
2597 			    sizeof(struct bge_ring_data), NULL,
2598 			    BUS_DMA_NOWAIT)) {
2599 		bus_dmamap_destroy(sc->bge_dmatag, sc->bge_ring_map);
2600 		bus_dmamem_unmap(sc->bge_dmatag, kva,
2601 				 sizeof(struct bge_ring_data));
2602 		bus_dmamem_free(sc->bge_dmatag, &seg, rseg);
2603 		return;
2604 	}
2605 
2606 	DPRINTFN(5, ("bzero\n"));
2607 	sc->bge_rdata = (struct bge_ring_data *)kva;
2608 
2609 	memset(sc->bge_rdata, 0, sizeof(struct bge_ring_data));
2610 
2611 	/* Try to allocate memory for jumbo buffers. */
2612 	if ((sc->bge_quirks & BGE_QUIRK_5705_CORE) == 0) {
2613 		if (bge_alloc_jumbo_mem(sc)) {
2614 			aprint_error_dev(sc->bge_dev,
2615 			    "jumbo buffer allocation failed\n");
2616 		} else
2617 			sc->ethercom.ec_capabilities |= ETHERCAP_JUMBO_MTU;
2618 	}
2619 
2620 	/* Set default tuneable values. */
2621 	sc->bge_stat_ticks = BGE_TICKS_PER_SEC;
2622 	sc->bge_rx_coal_ticks = 150;
2623 	sc->bge_rx_max_coal_bds = 64;
2624 #ifdef ORIG_WPAUL_VALUES
2625 	sc->bge_tx_coal_ticks = 150;
2626 	sc->bge_tx_max_coal_bds = 128;
2627 #else
2628 	sc->bge_tx_coal_ticks = 300;
2629 	sc->bge_tx_max_coal_bds = 400;
2630 #endif
2631 	if (sc->bge_quirks & BGE_QUIRK_5705_CORE) {
2632 		sc->bge_tx_coal_ticks = (12 * 5);
2633 		sc->bge_rx_max_coal_bds = (12 * 5);
2634 			aprint_verbose_dev(sc->bge_dev,
2635 			    "setting short Tx thresholds\n");
2636 	}
2637 
2638 	/* Set up ifnet structure */
2639 	ifp = &sc->ethercom.ec_if;
2640 	ifp->if_softc = sc;
2641 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
2642 	ifp->if_ioctl = bge_ioctl;
2643 	ifp->if_stop = bge_stop;
2644 	ifp->if_start = bge_start;
2645 	ifp->if_init = bge_init;
2646 	ifp->if_watchdog = bge_watchdog;
2647 	IFQ_SET_MAXLEN(&ifp->if_snd, max(BGE_TX_RING_CNT - 1, IFQ_MAXLEN));
2648 	IFQ_SET_READY(&ifp->if_snd);
2649 	DPRINTFN(5, ("strcpy if_xname\n"));
2650 	strcpy(ifp->if_xname, device_xname(sc->bge_dev));
2651 
2652 	if ((sc->bge_quirks & BGE_QUIRK_CSUM_BROKEN) == 0)
2653 		sc->ethercom.ec_if.if_capabilities |=
2654 		    IFCAP_CSUM_IPv4_Tx | IFCAP_CSUM_IPv4_Rx |
2655 		    IFCAP_CSUM_TCPv4_Tx | IFCAP_CSUM_TCPv4_Rx |
2656 		    IFCAP_CSUM_UDPv4_Tx | IFCAP_CSUM_UDPv4_Rx;
2657 	sc->ethercom.ec_capabilities |=
2658 	    ETHERCAP_VLAN_HWTAGGING | ETHERCAP_VLAN_MTU;
2659 
2660 	if (sc->bge_pcie)
2661 		sc->ethercom.ec_if.if_capabilities |= IFCAP_TSOv4;
2662 
2663 	/*
2664 	 * Do MII setup.
2665 	 */
2666 	DPRINTFN(5, ("mii setup\n"));
2667 	sc->bge_mii.mii_ifp = ifp;
2668 	sc->bge_mii.mii_readreg = bge_miibus_readreg;
2669 	sc->bge_mii.mii_writereg = bge_miibus_writereg;
2670 	sc->bge_mii.mii_statchg = bge_miibus_statchg;
2671 
2672 	/*
2673 	 * Figure out what sort of media we have by checking the
2674 	 * hardware config word in the first 32k of NIC internal memory,
2675 	 * or fall back to the config word in the EEPROM. Note: on some BCM5700
2676 	 * cards, this value appears to be unset. If that's the
2677 	 * case, we have to rely on identifying the NIC by its PCI
2678 	 * subsystem ID, as we do below for the SysKonnect SK-9D41.
2679 	 */
2680 	if (bge_readmem_ind(sc, BGE_SOFTWARE_GENCOMM_SIG) == BGE_MAGIC_NUMBER) {
2681 		hwcfg = bge_readmem_ind(sc, BGE_SOFTWARE_GENCOMM_NICCFG);
2682 	} else {
2683 		bge_read_eeprom(sc, (void *)&hwcfg,
2684 		    BGE_EE_HWCFG_OFFSET, sizeof(hwcfg));
2685 		hwcfg = be32toh(hwcfg);
2686 	}
2687 	if ((hwcfg & BGE_HWCFG_MEDIA) == BGE_MEDIA_FIBER)
2688 		sc->bge_tbi = 1;
2689 
2690 	/* The SysKonnect SK-9D41 is a 1000baseSX card. */
2691 	if ((pci_conf_read(sc->sc_pc, sc->sc_pcitag, BGE_PCI_SUBSYS) >> 16) ==
2692 	    SK_SUBSYSID_9D41)
2693 		sc->bge_tbi = 1;
2694 
2695 	if (sc->bge_tbi) {
2696 		ifmedia_init(&sc->bge_ifmedia, IFM_IMASK, bge_ifmedia_upd,
2697 		    bge_ifmedia_sts);
2698 		ifmedia_add(&sc->bge_ifmedia, IFM_ETHER|IFM_1000_SX, 0, NULL);
2699 		ifmedia_add(&sc->bge_ifmedia, IFM_ETHER|IFM_1000_SX|IFM_FDX,
2700 			    0, NULL);
2701 		ifmedia_add(&sc->bge_ifmedia, IFM_ETHER|IFM_AUTO, 0, NULL);
2702 		ifmedia_set(&sc->bge_ifmedia, IFM_ETHER|IFM_AUTO);
2703 	} else {
2704 		/*
2705 		 * Do transceiver setup.
2706 		 */
2707 		ifmedia_init(&sc->bge_mii.mii_media, 0, bge_ifmedia_upd,
2708 			     bge_ifmedia_sts);
2709 		mii_attach(sc->bge_dev, &sc->bge_mii, 0xffffffff,
2710 			   MII_PHY_ANY, MII_OFFSET_ANY,
2711 			   MIIF_FORCEANEG|MIIF_DOPAUSE);
2712 
2713 		if (LIST_FIRST(&sc->bge_mii.mii_phys) == NULL) {
2714 			aprint_error_dev(sc->bge_dev, "no PHY found!\n");
2715 			ifmedia_add(&sc->bge_mii.mii_media,
2716 				    IFM_ETHER|IFM_MANUAL, 0, NULL);
2717 			ifmedia_set(&sc->bge_mii.mii_media,
2718 				    IFM_ETHER|IFM_MANUAL);
2719 		} else
2720 			ifmedia_set(&sc->bge_mii.mii_media,
2721 				    IFM_ETHER|IFM_AUTO);
2722 	}
2723 
2724 	/*
2725 	 * When using the BCM5701 in PCI-X mode, data corruption has
2726 	 * been observed in the first few bytes of some received packets.
2727 	 * Aligning the packet buffer in memory eliminates the corruption.
2728 	 * Unfortunately, this misaligns the packet payloads.  On platforms
2729 	 * which do not support unaligned accesses, we will realign the
2730 	 * payloads by copying the received packets.
2731 	 */
2732 	if (sc->bge_quirks & BGE_QUIRK_PCIX_DMA_ALIGN_BUG) {
2733 		/* If in PCI-X mode, work around the alignment bug. */
2734 		if ((pci_conf_read(sc->sc_pc, sc->sc_pcitag, BGE_PCI_PCISTATE) &
2735                     (BGE_PCISTATE_PCI_BUSMODE | BGE_PCISTATE_PCI_BUSSPEED)) ==
2736                          BGE_PCISTATE_PCI_BUSSPEED)
2737 		sc->bge_rx_alignment_bug = 1;
2738         }
2739 
2740 	/*
2741 	 * Call MI attach routine.
2742 	 */
2743 	DPRINTFN(5, ("if_attach\n"));
2744 	if_attach(ifp);
2745 	DPRINTFN(5, ("ether_ifattach\n"));
2746 	ether_ifattach(ifp, eaddr);
2747 #ifdef BGE_EVENT_COUNTERS
2748 	/*
2749 	 * Attach event counters.
2750 	 */
2751 	evcnt_attach_dynamic(&sc->bge_ev_intr, EVCNT_TYPE_INTR,
2752 	    NULL, device_xname(sc->bge_dev), "intr");
2753 	evcnt_attach_dynamic(&sc->bge_ev_tx_xoff, EVCNT_TYPE_MISC,
2754 	    NULL, device_xname(sc->bge_dev), "tx_xoff");
2755 	evcnt_attach_dynamic(&sc->bge_ev_tx_xon, EVCNT_TYPE_MISC,
2756 	    NULL, device_xname(sc->bge_dev), "tx_xon");
2757 	evcnt_attach_dynamic(&sc->bge_ev_rx_xoff, EVCNT_TYPE_MISC,
2758 	    NULL, device_xname(sc->bge_dev), "rx_xoff");
2759 	evcnt_attach_dynamic(&sc->bge_ev_rx_xon, EVCNT_TYPE_MISC,
2760 	    NULL, device_xname(sc->bge_dev), "rx_xon");
2761 	evcnt_attach_dynamic(&sc->bge_ev_rx_macctl, EVCNT_TYPE_MISC,
2762 	    NULL, device_xname(sc->bge_dev), "rx_macctl");
2763 	evcnt_attach_dynamic(&sc->bge_ev_xoffentered, EVCNT_TYPE_MISC,
2764 	    NULL, device_xname(sc->bge_dev), "xoffentered");
2765 #endif /* BGE_EVENT_COUNTERS */
2766 	DPRINTFN(5, ("callout_init\n"));
2767 	callout_init(&sc->bge_timeout, 0);
2768 
2769 	if (!pmf_device_register(self, NULL, NULL))
2770 		aprint_error_dev(self, "couldn't establish power handler\n");
2771 	else
2772 		pmf_class_network_register(self, ifp);
2773 }
2774 
2775 static void
2776 bge_release_resources(struct bge_softc *sc)
2777 {
2778 	if (sc->bge_vpd_prodname != NULL)
2779 		free(sc->bge_vpd_prodname, M_DEVBUF);
2780 
2781 	if (sc->bge_vpd_readonly != NULL)
2782 		free(sc->bge_vpd_readonly, M_DEVBUF);
2783 }
2784 
2785 static void
2786 bge_reset(struct bge_softc *sc)
2787 {
2788 	u_int32_t cachesize, command, pcistate, new_pcistate;
2789 	int i, val;
2790 
2791 	/* Save some important PCI state. */
2792 	cachesize = pci_conf_read(sc->sc_pc, sc->sc_pcitag, BGE_PCI_CACHESZ);
2793 	command = pci_conf_read(sc->sc_pc, sc->sc_pcitag, BGE_PCI_CMD);
2794 	pcistate = pci_conf_read(sc->sc_pc, sc->sc_pcitag, BGE_PCI_PCISTATE);
2795 
2796 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MISC_CTL,
2797 	    BGE_PCIMISCCTL_INDIRECT_ACCESS|BGE_PCIMISCCTL_MASK_PCI_INTR|
2798 	    BGE_HIF_SWAP_OPTIONS|BGE_PCIMISCCTL_PCISTATE_RW);
2799 
2800 	/*
2801 	 * Disable the firmware fastboot feature on 5752 ASIC
2802 	 * to avoid firmware timeout.
2803 	 */
2804 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5752 ||
2805 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5755 ||
2806 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5787)
2807 		CSR_WRITE_4(sc, BGE_FASTBOOT_PC, 0);
2808 
2809 	val = BGE_MISCCFG_RESET_CORE_CLOCKS | (65<<1);
2810 	/*
2811 	 * XXX: from FreeBSD/Linux; no documentation
2812 	 */
2813 	if (sc->bge_pcie) {
2814 		if (CSR_READ_4(sc, BGE_PCIE_CTL1) == 0x60)
2815 			CSR_WRITE_4(sc, BGE_PCIE_CTL1, 0x20);
2816 		if (sc->bge_chipid != BGE_CHIPID_BCM5750_A0) {
2817 			/* No idea what that actually means */
2818 			CSR_WRITE_4(sc, BGE_MISC_CFG, 1 << 29);
2819 			val |= (1<<29);
2820 		}
2821 	}
2822 
2823 	/* Issue global reset */
2824 	bge_writereg_ind(sc, BGE_MISC_CFG, val);
2825 
2826 	DELAY(1000);
2827 
2828 	/*
2829 	 * XXX: from FreeBSD/Linux; no documentation
2830 	 */
2831 	if (sc->bge_pcie) {
2832 		if (sc->bge_chipid == BGE_CHIPID_BCM5750_A0) {
2833 			pcireg_t reg;
2834 
2835 			DELAY(500000);
2836 			/* XXX: Magic Numbers */
2837 			reg = pci_conf_read(sc->sc_pc, sc->sc_pcitag, BGE_PCI_UNKNOWN0);
2838 			pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_UNKNOWN0,
2839 			    reg | (1 << 15));
2840 		}
2841 		/*
2842 		 * XXX: Magic Numbers.
2843 		 * Sets maximal PCI-e payload and clears any PCI-e errors.
2844 		 * Should be replaced with references to PCI config-space
2845 		 * capability block for PCI-Express.
2846 		 */
2847 		pci_conf_write(sc->sc_pc, sc->sc_pcitag,
2848 		    BGE_PCI_CONF_DEV_CTRL, 0xf5000);
2849 
2850 	}
2851 
2852 	/* Reset some of the PCI state that got zapped by reset */
2853 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MISC_CTL,
2854 	    BGE_PCIMISCCTL_INDIRECT_ACCESS|BGE_PCIMISCCTL_MASK_PCI_INTR|
2855 	    BGE_HIF_SWAP_OPTIONS|BGE_PCIMISCCTL_PCISTATE_RW);
2856 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_CMD, command);
2857 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_CACHESZ, cachesize);
2858 	bge_writereg_ind(sc, BGE_MISC_CFG, (65 << 1));
2859 
2860 	/* Enable memory arbiter. */
2861 	{
2862 		uint32_t marbmode = 0;
2863 		if (BGE_IS_5714_FAMILY(sc)) {
2864 			marbmode = CSR_READ_4(sc, BGE_MARB_MODE);
2865 		}
2866  		CSR_WRITE_4(sc, BGE_MARB_MODE, BGE_MARBMODE_ENABLE | marbmode);
2867 	}
2868 
2869 	/*
2870 	 * Write the magic number to the firmware mailbox at 0xb50
2871 	 * so that the driver can synchronize with the firmware.
2872 	 */
2873 	bge_writemem_ind(sc, BGE_SOFTWARE_GENCOMM, BGE_MAGIC_NUMBER);
2874 
2875 	/*
2876 	 * Poll the value location we just wrote until
2877 	 * we see the 1's complement of the magic number.
2878 	 * This indicates that the firmware initialization
2879 	 * is complete.
2880 	 */
2881 	for (i = 0; i < BGE_TIMEOUT; i++) {
2882 		val = bge_readmem_ind(sc, BGE_SOFTWARE_GENCOMM);
2883 		if (val == ~BGE_MAGIC_NUMBER)
2884 			break;
2885 		DELAY(1000);
2886 	}
2887 
2888 	if (i >= BGE_TIMEOUT) {
2889 		aprint_error_dev(sc->bge_dev,
2890 		    "firmware handshake timed out, val = %x\n", val);
2891 		/*
2892 		 * XXX: occasionally fired on bcm5721, but without
2893 		 * apparent harm.  For now, keep going if we timeout
2894 		 * against PCI-E devices.
2895 		 */
2896 		 if (!sc->bge_pcie)
2897 		  return;
2898 	}
2899 
2900 	/*
2901 	 * XXX Wait for the value of the PCISTATE register to
2902 	 * return to its original pre-reset state. This is a
2903 	 * fairly good indicator of reset completion. If we don't
2904 	 * wait for the reset to fully complete, trying to read
2905 	 * from the device's non-PCI registers may yield garbage
2906 	 * results.
2907 	 */
2908 	for (i = 0; i < 10000; i++) {
2909 		new_pcistate = pci_conf_read(sc->sc_pc, sc->sc_pcitag,
2910 		    BGE_PCI_PCISTATE);
2911 		if ((new_pcistate & ~BGE_PCISTATE_RESERVED) ==
2912 		    (pcistate & ~BGE_PCISTATE_RESERVED))
2913 			break;
2914 		DELAY(10);
2915 	}
2916 	if ((new_pcistate & ~BGE_PCISTATE_RESERVED) !=
2917 	    (pcistate & ~BGE_PCISTATE_RESERVED)) {
2918 		aprint_error_dev(sc->bge_dev, "pcistate failed to revert\n");
2919 	}
2920 
2921 	/* XXX: from FreeBSD/Linux; no documentation */
2922 	if (sc->bge_pcie && sc->bge_chipid != BGE_CHIPID_BCM5750_A0)
2923 		CSR_WRITE_4(sc, BGE_PCIE_CTL0, CSR_READ_4(sc, BGE_PCIE_CTL0) | (1<<25));
2924 
2925 	/* Enable memory arbiter. */
2926 	/* XXX why do this twice? */
2927 	{
2928 		uint32_t marbmode = 0;
2929 		if (BGE_IS_5714_FAMILY(sc)) {
2930 			marbmode = CSR_READ_4(sc, BGE_MARB_MODE);
2931 		}
2932  		CSR_WRITE_4(sc, BGE_MARB_MODE, BGE_MARBMODE_ENABLE | marbmode);
2933 	}
2934 
2935 	/* Fix up byte swapping */
2936 	CSR_WRITE_4(sc, BGE_MODE_CTL, BGE_DMA_SWAP_OPTIONS);
2937 
2938 	CSR_WRITE_4(sc, BGE_MAC_MODE, 0);
2939 
2940 	DELAY(10000);
2941 }
2942 
2943 /*
2944  * Frame reception handling. This is called if there's a frame
2945  * on the receive return list.
2946  *
2947  * Note: we have to be able to handle two possibilities here:
2948  * 1) the frame is from the jumbo recieve ring
2949  * 2) the frame is from the standard receive ring
2950  */
2951 
2952 static void
2953 bge_rxeof(struct bge_softc *sc)
2954 {
2955 	struct ifnet *ifp;
2956 	int stdcnt = 0, jumbocnt = 0;
2957 	bus_dmamap_t dmamap;
2958 	bus_addr_t offset, toff;
2959 	bus_size_t tlen;
2960 	int tosync;
2961 
2962 	ifp = &sc->ethercom.ec_if;
2963 
2964 	bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
2965 	    offsetof(struct bge_ring_data, bge_status_block),
2966 	    sizeof (struct bge_status_block),
2967 	    BUS_DMASYNC_POSTREAD);
2968 
2969 	offset = offsetof(struct bge_ring_data, bge_rx_return_ring);
2970 	tosync = sc->bge_rdata->bge_status_block.bge_idx[0].bge_rx_prod_idx -
2971 	    sc->bge_rx_saved_considx;
2972 
2973 	toff = offset + (sc->bge_rx_saved_considx * sizeof (struct bge_rx_bd));
2974 
2975 	if (tosync < 0) {
2976 		tlen = (sc->bge_return_ring_cnt - sc->bge_rx_saved_considx) *
2977 		    sizeof (struct bge_rx_bd);
2978 		bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
2979 		    toff, tlen, BUS_DMASYNC_POSTREAD);
2980 		tosync = -tosync;
2981 	}
2982 
2983 	bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
2984 	    offset, tosync * sizeof (struct bge_rx_bd),
2985 	    BUS_DMASYNC_POSTREAD);
2986 
2987 	while(sc->bge_rx_saved_considx !=
2988 	    sc->bge_rdata->bge_status_block.bge_idx[0].bge_rx_prod_idx) {
2989 		struct bge_rx_bd	*cur_rx;
2990 		u_int32_t		rxidx;
2991 		struct mbuf		*m = NULL;
2992 
2993 		cur_rx = &sc->bge_rdata->
2994 			bge_rx_return_ring[sc->bge_rx_saved_considx];
2995 
2996 		rxidx = cur_rx->bge_idx;
2997 		BGE_INC(sc->bge_rx_saved_considx, sc->bge_return_ring_cnt);
2998 
2999 		if (cur_rx->bge_flags & BGE_RXBDFLAG_JUMBO_RING) {
3000 			BGE_INC(sc->bge_jumbo, BGE_JUMBO_RX_RING_CNT);
3001 			m = sc->bge_cdata.bge_rx_jumbo_chain[rxidx];
3002 			sc->bge_cdata.bge_rx_jumbo_chain[rxidx] = NULL;
3003 			jumbocnt++;
3004 			bus_dmamap_sync(sc->bge_dmatag,
3005 			    sc->bge_cdata.bge_rx_jumbo_map,
3006 			    mtod(m, char *) - (char *)sc->bge_cdata.bge_jumbo_buf,
3007 			    BGE_JLEN, BUS_DMASYNC_POSTREAD);
3008 			if (cur_rx->bge_flags & BGE_RXBDFLAG_ERROR) {
3009 				ifp->if_ierrors++;
3010 				bge_newbuf_jumbo(sc, sc->bge_jumbo, m);
3011 				continue;
3012 			}
3013 			if (bge_newbuf_jumbo(sc, sc->bge_jumbo,
3014 					     NULL)== ENOBUFS) {
3015 				ifp->if_ierrors++;
3016 				bge_newbuf_jumbo(sc, sc->bge_jumbo, m);
3017 				continue;
3018 			}
3019 		} else {
3020 			BGE_INC(sc->bge_std, BGE_STD_RX_RING_CNT);
3021 			m = sc->bge_cdata.bge_rx_std_chain[rxidx];
3022 
3023 			sc->bge_cdata.bge_rx_std_chain[rxidx] = NULL;
3024 			stdcnt++;
3025 			dmamap = sc->bge_cdata.bge_rx_std_map[rxidx];
3026 			sc->bge_cdata.bge_rx_std_map[rxidx] = 0;
3027 			bus_dmamap_sync(sc->bge_dmatag, dmamap, 0,
3028 			    dmamap->dm_mapsize, BUS_DMASYNC_POSTREAD);
3029 			bus_dmamap_unload(sc->bge_dmatag, dmamap);
3030 			if (cur_rx->bge_flags & BGE_RXBDFLAG_ERROR) {
3031 				ifp->if_ierrors++;
3032 				bge_newbuf_std(sc, sc->bge_std, m, dmamap);
3033 				continue;
3034 			}
3035 			if (bge_newbuf_std(sc, sc->bge_std,
3036 			    NULL, dmamap) == ENOBUFS) {
3037 				ifp->if_ierrors++;
3038 				bge_newbuf_std(sc, sc->bge_std, m, dmamap);
3039 				continue;
3040 			}
3041 		}
3042 
3043 		ifp->if_ipackets++;
3044 #ifndef __NO_STRICT_ALIGNMENT
3045                 /*
3046                  * XXX: if the 5701 PCIX-Rx-DMA workaround is in effect,
3047                  * the Rx buffer has the layer-2 header unaligned.
3048                  * If our CPU requires alignment, re-align by copying.
3049                  */
3050 		if (sc->bge_rx_alignment_bug) {
3051 			memmove(mtod(m, char *) + ETHER_ALIGN, m->m_data,
3052                                 cur_rx->bge_len);
3053 			m->m_data += ETHER_ALIGN;
3054 		}
3055 #endif
3056 
3057 		m->m_pkthdr.len = m->m_len = cur_rx->bge_len - ETHER_CRC_LEN;
3058 		m->m_pkthdr.rcvif = ifp;
3059 
3060 #if NBPFILTER > 0
3061 		/*
3062 		 * Handle BPF listeners. Let the BPF user see the packet.
3063 		 */
3064 		if (ifp->if_bpf)
3065 			bpf_mtap(ifp->if_bpf, m);
3066 #endif
3067 
3068 		m->m_pkthdr.csum_flags = M_CSUM_IPv4;
3069 
3070 		if ((cur_rx->bge_ip_csum ^ 0xffff) != 0)
3071 			m->m_pkthdr.csum_flags |= M_CSUM_IPv4_BAD;
3072 		/*
3073 		 * Rx transport checksum-offload may also
3074 		 * have bugs with packets which, when transmitted,
3075 		 * were `runts' requiring padding.
3076 		 */
3077 		if (cur_rx->bge_flags & BGE_RXBDFLAG_TCP_UDP_CSUM &&
3078 		    (/* (sc->_bge_quirks & BGE_QUIRK_SHORT_CKSUM_BUG) == 0 ||*/
3079 		     m->m_pkthdr.len >= ETHER_MIN_NOPAD)) {
3080 			m->m_pkthdr.csum_data =
3081 			    cur_rx->bge_tcp_udp_csum;
3082 			m->m_pkthdr.csum_flags |=
3083 			    (M_CSUM_TCPv4|M_CSUM_UDPv4|
3084 			     M_CSUM_DATA|M_CSUM_NO_PSEUDOHDR);
3085 		}
3086 
3087 		/*
3088 		 * If we received a packet with a vlan tag, pass it
3089 		 * to vlan_input() instead of ether_input().
3090 		 */
3091 		if (cur_rx->bge_flags & BGE_RXBDFLAG_VLAN_TAG)
3092 			VLAN_INPUT_TAG(ifp, m, cur_rx->bge_vlan_tag, continue);
3093 
3094 		(*ifp->if_input)(ifp, m);
3095 	}
3096 
3097 	CSR_WRITE_4(sc, BGE_MBX_RX_CONS0_LO, sc->bge_rx_saved_considx);
3098 	if (stdcnt)
3099 		CSR_WRITE_4(sc, BGE_MBX_RX_STD_PROD_LO, sc->bge_std);
3100 	if (jumbocnt)
3101 		CSR_WRITE_4(sc, BGE_MBX_RX_JUMBO_PROD_LO, sc->bge_jumbo);
3102 }
3103 
3104 static void
3105 bge_txeof(struct bge_softc *sc)
3106 {
3107 	struct bge_tx_bd *cur_tx = NULL;
3108 	struct ifnet *ifp;
3109 	struct txdmamap_pool_entry *dma;
3110 	bus_addr_t offset, toff;
3111 	bus_size_t tlen;
3112 	int tosync;
3113 	struct mbuf *m;
3114 
3115 	ifp = &sc->ethercom.ec_if;
3116 
3117 	bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
3118 	    offsetof(struct bge_ring_data, bge_status_block),
3119 	    sizeof (struct bge_status_block),
3120 	    BUS_DMASYNC_POSTREAD);
3121 
3122 	offset = offsetof(struct bge_ring_data, bge_tx_ring);
3123 	tosync = sc->bge_rdata->bge_status_block.bge_idx[0].bge_tx_cons_idx -
3124 	    sc->bge_tx_saved_considx;
3125 
3126 	toff = offset + (sc->bge_tx_saved_considx * sizeof (struct bge_tx_bd));
3127 
3128 	if (tosync < 0) {
3129 		tlen = (BGE_TX_RING_CNT - sc->bge_tx_saved_considx) *
3130 		    sizeof (struct bge_tx_bd);
3131 		bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
3132 		    toff, tlen, BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
3133 		tosync = -tosync;
3134 	}
3135 
3136 	bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
3137 	    offset, tosync * sizeof (struct bge_tx_bd),
3138 	    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
3139 
3140 	/*
3141 	 * Go through our tx ring and free mbufs for those
3142 	 * frames that have been sent.
3143 	 */
3144 	while (sc->bge_tx_saved_considx !=
3145 	    sc->bge_rdata->bge_status_block.bge_idx[0].bge_tx_cons_idx) {
3146 		u_int32_t		idx = 0;
3147 
3148 		idx = sc->bge_tx_saved_considx;
3149 		cur_tx = &sc->bge_rdata->bge_tx_ring[idx];
3150 		if (cur_tx->bge_flags & BGE_TXBDFLAG_END)
3151 			ifp->if_opackets++;
3152 		m = sc->bge_cdata.bge_tx_chain[idx];
3153 		if (m != NULL) {
3154 			sc->bge_cdata.bge_tx_chain[idx] = NULL;
3155 			dma = sc->txdma[idx];
3156 			bus_dmamap_sync(sc->bge_dmatag, dma->dmamap, 0,
3157 			    dma->dmamap->dm_mapsize, BUS_DMASYNC_POSTWRITE);
3158 			bus_dmamap_unload(sc->bge_dmatag, dma->dmamap);
3159 			SLIST_INSERT_HEAD(&sc->txdma_list, dma, link);
3160 			sc->txdma[idx] = NULL;
3161 
3162 			m_freem(m);
3163 		}
3164 		sc->bge_txcnt--;
3165 		BGE_INC(sc->bge_tx_saved_considx, BGE_TX_RING_CNT);
3166 		ifp->if_timer = 0;
3167 	}
3168 
3169 	if (cur_tx != NULL)
3170 		ifp->if_flags &= ~IFF_OACTIVE;
3171 }
3172 
3173 static int
3174 bge_intr(void *xsc)
3175 {
3176 	struct bge_softc *sc;
3177 	struct ifnet *ifp;
3178 
3179 	sc = xsc;
3180 	ifp = &sc->ethercom.ec_if;
3181 
3182 #ifdef notdef
3183 	/* Avoid this for now -- checking this register is expensive. */
3184 	/* Make sure this is really our interrupt. */
3185 	if (!(CSR_READ_4(sc, BGE_MISC_LOCAL_CTL) & BGE_MLC_INTR_STATE))
3186 		return (0);
3187 #endif
3188 	/* Ack interrupt and stop others from occuring. */
3189 	CSR_WRITE_4(sc, BGE_MBX_IRQ0_LO, 1);
3190 
3191 	BGE_EVCNT_INCR(sc->bge_ev_intr);
3192 
3193 	/*
3194 	 * Process link state changes.
3195 	 * Grrr. The link status word in the status block does
3196 	 * not work correctly on the BCM5700 rev AX and BX chips,
3197 	 * according to all available information. Hence, we have
3198 	 * to enable MII interrupts in order to properly obtain
3199 	 * async link changes. Unfortunately, this also means that
3200 	 * we have to read the MAC status register to detect link
3201 	 * changes, thereby adding an additional register access to
3202 	 * the interrupt handler.
3203 	 */
3204 
3205 	if (sc->bge_quirks & BGE_QUIRK_LINK_STATE_BROKEN) {
3206 		u_int32_t		status;
3207 
3208 		status = CSR_READ_4(sc, BGE_MAC_STS);
3209 		if (status & BGE_MACSTAT_MI_INTERRUPT) {
3210 			sc->bge_link = 0;
3211 			callout_stop(&sc->bge_timeout);
3212 			bge_tick(sc);
3213 			/* Clear the interrupt */
3214 			CSR_WRITE_4(sc, BGE_MAC_EVT_ENB,
3215 			    BGE_EVTENB_MI_INTERRUPT);
3216 			bge_miibus_readreg(sc->bge_dev, 1, BRGPHY_MII_ISR);
3217 			bge_miibus_writereg(sc->bge_dev, 1, BRGPHY_MII_IMR,
3218 			    BRGPHY_INTRS);
3219 		}
3220 	} else {
3221 		if (sc->bge_rdata->bge_status_block.bge_status &
3222 		    BGE_STATFLAG_LINKSTATE_CHANGED) {
3223 			sc->bge_link = 0;
3224 			callout_stop(&sc->bge_timeout);
3225 			bge_tick(sc);
3226 			/* Clear the interrupt */
3227 			CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED|
3228 			    BGE_MACSTAT_CFG_CHANGED|BGE_MACSTAT_MI_COMPLETE|
3229 			    BGE_MACSTAT_LINK_CHANGED);
3230 		}
3231 	}
3232 
3233 	if (ifp->if_flags & IFF_RUNNING) {
3234 		/* Check RX return ring producer/consumer */
3235 		bge_rxeof(sc);
3236 
3237 		/* Check TX ring producer/consumer */
3238 		bge_txeof(sc);
3239 	}
3240 
3241 	if (sc->bge_pending_rxintr_change) {
3242 		uint32_t rx_ticks = sc->bge_rx_coal_ticks;
3243 		uint32_t rx_bds = sc->bge_rx_max_coal_bds;
3244 		uint32_t junk;
3245 
3246 		CSR_WRITE_4(sc, BGE_HCC_RX_COAL_TICKS, rx_ticks);
3247 		DELAY(10);
3248 		junk = CSR_READ_4(sc, BGE_HCC_RX_COAL_TICKS);
3249 
3250 		CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS, rx_bds);
3251 		DELAY(10);
3252 		junk = CSR_READ_4(sc, BGE_HCC_RX_MAX_COAL_BDS);
3253 
3254 		sc->bge_pending_rxintr_change = 0;
3255 	}
3256 	bge_handle_events(sc);
3257 
3258 	/* Re-enable interrupts. */
3259 	CSR_WRITE_4(sc, BGE_MBX_IRQ0_LO, 0);
3260 
3261 	if (ifp->if_flags & IFF_RUNNING && !IFQ_IS_EMPTY(&ifp->if_snd))
3262 		bge_start(ifp);
3263 
3264 	return (1);
3265 }
3266 
3267 static void
3268 bge_tick(void *xsc)
3269 {
3270 	struct bge_softc *sc = xsc;
3271 	struct mii_data *mii = &sc->bge_mii;
3272 	struct ifmedia *ifm = NULL;
3273 	struct ifnet *ifp = &sc->ethercom.ec_if;
3274 	int s;
3275 
3276 	s = splnet();
3277 
3278 	bge_stats_update(sc);
3279 	callout_reset(&sc->bge_timeout, hz, bge_tick, sc);
3280 	if (sc->bge_link) {
3281 		splx(s);
3282 		return;
3283 	}
3284 
3285 	if (sc->bge_tbi) {
3286 		ifm = &sc->bge_ifmedia;
3287 		if (CSR_READ_4(sc, BGE_MAC_STS) &
3288 		    BGE_MACSTAT_TBI_PCS_SYNCHED) {
3289 			sc->bge_link++;
3290 			CSR_WRITE_4(sc, BGE_MAC_STS, 0xFFFFFFFF);
3291 			if (!IFQ_IS_EMPTY(&ifp->if_snd))
3292 				bge_start(ifp);
3293 		}
3294 		splx(s);
3295 		return;
3296 	}
3297 
3298 	mii_tick(mii);
3299 
3300 	if (!sc->bge_link && mii->mii_media_status & IFM_ACTIVE &&
3301 	    IFM_SUBTYPE(mii->mii_media_active) != IFM_NONE) {
3302 		sc->bge_link++;
3303 		if (!IFQ_IS_EMPTY(&ifp->if_snd))
3304 			bge_start(ifp);
3305 	}
3306 
3307 	splx(s);
3308 }
3309 
3310 static void
3311 bge_stats_update(struct bge_softc *sc)
3312 {
3313 	struct ifnet *ifp = &sc->ethercom.ec_if;
3314 	bus_size_t stats = BGE_MEMWIN_START + BGE_STATS_BLOCK;
3315 	bus_size_t rstats = BGE_RX_STATS;
3316 
3317 #define READ_RSTAT(sc, stats, stat) \
3318 	  CSR_READ_4(sc, stats + offsetof(struct bge_mac_stats_regs, stat))
3319 
3320 	if (sc->bge_quirks & BGE_QUIRK_5705_CORE) {
3321 		ifp->if_collisions +=
3322 		    READ_RSTAT(sc, rstats, dot3StatsSingleCollisionFrames) +
3323 		    READ_RSTAT(sc, rstats, dot3StatsMultipleCollisionFrames) +
3324 		    READ_RSTAT(sc, rstats, dot3StatsExcessiveCollisions) +
3325 		    READ_RSTAT(sc, rstats, dot3StatsLateCollisions);
3326 
3327 		BGE_EVCNT_ADD(sc->bge_ev_tx_xoff,
3328 			      READ_RSTAT(sc, rstats, outXoffSent));
3329 		BGE_EVCNT_ADD(sc->bge_ev_tx_xon,
3330 			      READ_RSTAT(sc, rstats, outXonSent));
3331 		BGE_EVCNT_ADD(sc->bge_ev_rx_xoff,
3332 			      READ_RSTAT(sc, rstats, xoffPauseFramesReceived));
3333 		BGE_EVCNT_ADD(sc->bge_ev_rx_xon,
3334 			      READ_RSTAT(sc, rstats, xonPauseFramesReceived));
3335 		BGE_EVCNT_ADD(sc->bge_ev_rx_macctl,
3336 			      READ_RSTAT(sc, rstats, macControlFramesReceived));
3337 		BGE_EVCNT_ADD(sc->bge_ev_xoffentered,
3338 			      READ_RSTAT(sc, rstats, xoffStateEntered));
3339 		return;
3340 	}
3341 
3342 #undef READ_RSTAT
3343 #define READ_STAT(sc, stats, stat) \
3344 	  CSR_READ_4(sc, stats + offsetof(struct bge_stats, stat))
3345 
3346 	ifp->if_collisions +=
3347 	  (READ_STAT(sc, stats, dot3StatsSingleCollisionFrames.bge_addr_lo) +
3348 	   READ_STAT(sc, stats, dot3StatsMultipleCollisionFrames.bge_addr_lo) +
3349 	   READ_STAT(sc, stats, dot3StatsExcessiveCollisions.bge_addr_lo) +
3350 	   READ_STAT(sc, stats, dot3StatsLateCollisions.bge_addr_lo)) -
3351 	  ifp->if_collisions;
3352 
3353 	BGE_EVCNT_UPD(sc->bge_ev_tx_xoff,
3354 		      READ_STAT(sc, stats, outXoffSent.bge_addr_lo));
3355 	BGE_EVCNT_UPD(sc->bge_ev_tx_xon,
3356 		      READ_STAT(sc, stats, outXonSent.bge_addr_lo));
3357 	BGE_EVCNT_UPD(sc->bge_ev_rx_xoff,
3358 		      READ_STAT(sc, stats,
3359 		      		xoffPauseFramesReceived.bge_addr_lo));
3360 	BGE_EVCNT_UPD(sc->bge_ev_rx_xon,
3361 		      READ_STAT(sc, stats, xonPauseFramesReceived.bge_addr_lo));
3362 	BGE_EVCNT_UPD(sc->bge_ev_rx_macctl,
3363 		      READ_STAT(sc, stats,
3364 		      		macControlFramesReceived.bge_addr_lo));
3365 	BGE_EVCNT_UPD(sc->bge_ev_xoffentered,
3366 		      READ_STAT(sc, stats, xoffStateEntered.bge_addr_lo));
3367 
3368 #undef READ_STAT
3369 
3370 #ifdef notdef
3371 	ifp->if_collisions +=
3372 	   (sc->bge_rdata->bge_info.bge_stats.dot3StatsSingleCollisionFrames +
3373 	   sc->bge_rdata->bge_info.bge_stats.dot3StatsMultipleCollisionFrames +
3374 	   sc->bge_rdata->bge_info.bge_stats.dot3StatsExcessiveCollisions +
3375 	   sc->bge_rdata->bge_info.bge_stats.dot3StatsLateCollisions) -
3376 	   ifp->if_collisions;
3377 #endif
3378 }
3379 
3380 /*
3381  * Pad outbound frame to ETHER_MIN_NOPAD for an unusual reason.
3382  * The bge hardware will pad out Tx runts to ETHER_MIN_NOPAD,
3383  * but when such padded frames employ the  bge IP/TCP checksum offload,
3384  * the hardware checksum assist gives incorrect results (possibly
3385  * from incorporating its own padding into the UDP/TCP checksum; who knows).
3386  * If we pad such runts with zeros, the onboard checksum comes out correct.
3387  */
3388 static inline int
3389 bge_cksum_pad(struct mbuf *pkt)
3390 {
3391 	struct mbuf *last = NULL;
3392 	int padlen;
3393 
3394 	padlen = ETHER_MIN_NOPAD - pkt->m_pkthdr.len;
3395 
3396 	/* if there's only the packet-header and we can pad there, use it. */
3397 	if (pkt->m_pkthdr.len == pkt->m_len &&
3398 	    M_TRAILINGSPACE(pkt) >= padlen) {
3399 		last = pkt;
3400 	} else {
3401 		/*
3402 		 * Walk packet chain to find last mbuf. We will either
3403 		 * pad there, or append a new mbuf and pad it
3404 		 * (thus perhaps avoiding the bcm5700 dma-min bug).
3405 		 */
3406 		for (last = pkt; last->m_next != NULL; last = last->m_next) {
3407 	      	       continue; /* do nothing */
3408 		}
3409 
3410 		/* `last' now points to last in chain. */
3411 		if (M_TRAILINGSPACE(last) < padlen) {
3412 			/* Allocate new empty mbuf, pad it. Compact later. */
3413 			struct mbuf *n;
3414 			MGET(n, M_DONTWAIT, MT_DATA);
3415 			if (n == NULL)
3416 				return ENOBUFS;
3417 			n->m_len = 0;
3418 			last->m_next = n;
3419 			last = n;
3420 		}
3421 	}
3422 
3423 	KDASSERT(!M_READONLY(last));
3424 	KDASSERT(M_TRAILINGSPACE(last) >= padlen);
3425 
3426 	/* Now zero the pad area, to avoid the bge cksum-assist bug */
3427 	memset(mtod(last, char *) + last->m_len, 0, padlen);
3428 	last->m_len += padlen;
3429 	pkt->m_pkthdr.len += padlen;
3430 	return 0;
3431 }
3432 
3433 /*
3434  * Compact outbound packets to avoid bug with DMA segments less than 8 bytes.
3435  */
3436 static inline int
3437 bge_compact_dma_runt(struct mbuf *pkt)
3438 {
3439 	struct mbuf	*m, *prev;
3440 	int 		totlen, prevlen;
3441 
3442 	prev = NULL;
3443 	totlen = 0;
3444 	prevlen = -1;
3445 
3446 	for (m = pkt; m != NULL; prev = m,m = m->m_next) {
3447 		int mlen = m->m_len;
3448 		int shortfall = 8 - mlen ;
3449 
3450 		totlen += mlen;
3451 		if (mlen == 0) {
3452 			continue;
3453 		}
3454 		if (mlen >= 8)
3455 			continue;
3456 
3457 		/* If we get here, mbuf data is too small for DMA engine.
3458 		 * Try to fix by shuffling data to prev or next in chain.
3459 		 * If that fails, do a compacting deep-copy of the whole chain.
3460 		 */
3461 
3462 		/* Internal frag. If fits in prev, copy it there. */
3463 		if (prev && M_TRAILINGSPACE(prev) >= m->m_len) {
3464 		  	memcpy(prev->m_data + prev->m_len, m->m_data, mlen);
3465 			prev->m_len += mlen;
3466 			m->m_len = 0;
3467 			/* XXX stitch chain */
3468 			prev->m_next = m_free(m);
3469 			m = prev;
3470 			continue;
3471 		}
3472 		else if (m->m_next != NULL &&
3473 			     M_TRAILINGSPACE(m) >= shortfall &&
3474 			     m->m_next->m_len >= (8 + shortfall)) {
3475 		    /* m is writable and have enough data in next, pull up. */
3476 
3477 		  	memcpy(m->m_data + m->m_len, m->m_next->m_data,
3478 			    shortfall);
3479 			m->m_len += shortfall;
3480 			m->m_next->m_len -= shortfall;
3481 			m->m_next->m_data += shortfall;
3482 		}
3483 		else if (m->m_next == NULL || 1) {
3484 		  	/* Got a runt at the very end of the packet.
3485 			 * borrow data from the tail of the preceding mbuf and
3486 			 * update its length in-place. (The original data is still
3487 			 * valid, so we can do this even if prev is not writable.)
3488 			 */
3489 
3490 			/* if we'd make prev a runt, just move all of its data. */
3491 			KASSERT(prev != NULL /*, ("runt but null PREV")*/);
3492 			KASSERT(prev->m_len >= 8 /*, ("runt prev")*/);
3493 
3494 			if ((prev->m_len - shortfall) < 8)
3495 				shortfall = prev->m_len;
3496 
3497 #ifdef notyet	/* just do the safe slow thing for now */
3498 			if (!M_READONLY(m)) {
3499 				if (M_LEADINGSPACE(m) < shorfall) {
3500 					void *m_dat;
3501 					m_dat = (m->m_flags & M_PKTHDR) ?
3502 					  m->m_pktdat : m->dat;
3503 					memmove(m_dat, mtod(m, void*), m->m_len);
3504 					m->m_data = m_dat;
3505 				    }
3506 			} else
3507 #endif	/* just do the safe slow thing */
3508 			{
3509 				struct mbuf * n = NULL;
3510 				int newprevlen = prev->m_len - shortfall;
3511 
3512 				MGET(n, M_NOWAIT, MT_DATA);
3513 				if (n == NULL)
3514 				   return ENOBUFS;
3515 				KASSERT(m->m_len + shortfall < MLEN
3516 					/*,
3517 					  ("runt %d +prev %d too big\n", m->m_len, shortfall)*/);
3518 
3519 				/* first copy the data we're stealing from prev */
3520 				memcpy(n->m_data, prev->m_data + newprevlen,
3521 				    shortfall);
3522 
3523 				/* update prev->m_len accordingly */
3524 				prev->m_len -= shortfall;
3525 
3526 				/* copy data from runt m */
3527 				memcpy(n->m_data + shortfall, m->m_data,
3528 				    m->m_len);
3529 
3530 				/* n holds what we stole from prev, plus m */
3531 				n->m_len = shortfall + m->m_len;
3532 
3533 				/* stitch n into chain and free m */
3534 				n->m_next = m->m_next;
3535 				prev->m_next = n;
3536 				/* KASSERT(m->m_next == NULL); */
3537 				m->m_next = NULL;
3538 				m_free(m);
3539 				m = n;	/* for continuing loop */
3540 			}
3541 		}
3542 		prevlen = m->m_len;
3543 	}
3544 	return 0;
3545 }
3546 
3547 /*
3548  * Encapsulate an mbuf chain in the tx ring  by coupling the mbuf data
3549  * pointers to descriptors.
3550  */
3551 static int
3552 bge_encap(struct bge_softc *sc, struct mbuf *m_head, u_int32_t *txidx)
3553 {
3554 	struct bge_tx_bd	*f = NULL;
3555 	u_int32_t		frag, cur;
3556 	u_int16_t		csum_flags = 0;
3557 	u_int16_t		txbd_tso_flags = 0;
3558 	struct txdmamap_pool_entry *dma;
3559 	bus_dmamap_t dmamap;
3560 	int			i = 0;
3561 	struct m_tag		*mtag;
3562 	int			use_tso, maxsegsize, error;
3563 
3564 	cur = frag = *txidx;
3565 
3566 	if (m_head->m_pkthdr.csum_flags) {
3567 		if (m_head->m_pkthdr.csum_flags & M_CSUM_IPv4)
3568 			csum_flags |= BGE_TXBDFLAG_IP_CSUM;
3569 		if (m_head->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4))
3570 			csum_flags |= BGE_TXBDFLAG_TCP_UDP_CSUM;
3571 	}
3572 
3573 	/*
3574 	 * If we were asked to do an outboard checksum, and the NIC
3575 	 * has the bug where it sometimes adds in the Ethernet padding,
3576 	 * explicitly pad with zeros so the cksum will be correct either way.
3577 	 * (For now, do this for all chip versions, until newer
3578 	 * are confirmed to not require the workaround.)
3579 	 */
3580 	if ((csum_flags & BGE_TXBDFLAG_TCP_UDP_CSUM) == 0 ||
3581 #ifdef notyet
3582 	    (sc->bge_quirks & BGE_QUIRK_SHORT_CKSUM_BUG) == 0 ||
3583 #endif
3584 	    m_head->m_pkthdr.len >= ETHER_MIN_NOPAD)
3585 		goto check_dma_bug;
3586 
3587 	if (bge_cksum_pad(m_head) != 0) {
3588 	    return ENOBUFS;
3589 	}
3590 
3591 check_dma_bug:
3592 	if (!(sc->bge_quirks & BGE_QUIRK_5700_SMALLDMA))
3593 		goto doit;
3594 	/*
3595 	 * bcm5700 Revision B silicon cannot handle DMA descriptors with
3596 	 * less than eight bytes.  If we encounter a teeny mbuf
3597 	 * at the end of a chain, we can pad.  Otherwise, copy.
3598 	 */
3599 	if (bge_compact_dma_runt(m_head) != 0)
3600 		return ENOBUFS;
3601 
3602 doit:
3603 	dma = SLIST_FIRST(&sc->txdma_list);
3604 	if (dma == NULL)
3605 		return ENOBUFS;
3606 	dmamap = dma->dmamap;
3607 
3608 	/*
3609 	 * Set up any necessary TSO state before we start packing...
3610 	 */
3611 	use_tso = (m_head->m_pkthdr.csum_flags & M_CSUM_TSOv4) != 0;
3612 	if (!use_tso) {
3613 		maxsegsize = 0;
3614 	} else {	/* TSO setup */
3615 		unsigned  mss;
3616 		struct ether_header *eh;
3617 		unsigned ip_tcp_hlen, iptcp_opt_words, tcp_seg_flags, offset;
3618 		struct mbuf * m0 = m_head;
3619 		struct ip *ip;
3620 		struct tcphdr *th;
3621 		int iphl, hlen;
3622 
3623 		/*
3624 		 * XXX It would be nice if the mbuf pkthdr had offset
3625 		 * fields for the protocol headers.
3626 		 */
3627 
3628 		eh = mtod(m0, struct ether_header *);
3629 		switch (htons(eh->ether_type)) {
3630 		case ETHERTYPE_IP:
3631 			offset = ETHER_HDR_LEN;
3632 			break;
3633 
3634 		case ETHERTYPE_VLAN:
3635 			offset = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
3636 			break;
3637 
3638 		default:
3639 			/*
3640 			 * Don't support this protocol or encapsulation.
3641 			 */
3642 			return (ENOBUFS);
3643 		}
3644 
3645 		/*
3646 		 * TCP/IP headers are in the first mbuf; we can do
3647 		 * this the easy way.
3648 		 */
3649 		iphl = M_CSUM_DATA_IPv4_IPHL(m0->m_pkthdr.csum_data);
3650 		hlen = iphl + offset;
3651 		if (__predict_false(m0->m_len <
3652 				    (hlen + sizeof(struct tcphdr)))) {
3653 
3654 			aprint_debug_dev(sc->bge_dev,
3655 			    "TSO: hard case m0->m_len == %d < ip/tcp hlen %zd,"
3656 			    "not handled yet\n",
3657 			     m0->m_len, hlen+ sizeof(struct tcphdr));
3658 #ifdef NOTYET
3659 			/*
3660 			 * XXX jonathan@NetBSD.org: untested.
3661 			 * how to force  this branch to be taken?
3662 			 */
3663 			BGE_EVCNT_INCR(&sc->sc_ev_txtsopain);
3664 
3665 			m_copydata(m0, offset, sizeof(ip), &ip);
3666 			m_copydata(m0, hlen, sizeof(th), &th);
3667 
3668 			ip.ip_len = 0;
3669 
3670 			m_copyback(m0, hlen + offsetof(struct ip, ip_len),
3671 			    sizeof(ip.ip_len), &ip.ip_len);
3672 
3673 			th.th_sum = in_cksum_phdr(ip.ip_src.s_addr,
3674 			    ip.ip_dst.s_addr, htons(IPPROTO_TCP));
3675 
3676 			m_copyback(m0, hlen + offsetof(struct tcphdr, th_sum),
3677 			    sizeof(th.th_sum), &th.th_sum);
3678 
3679 			hlen += th.th_off << 2;
3680 			iptcp_opt_words	= hlen;
3681 #else
3682 			/*
3683 			 * if_wm "hard" case not yet supported, can we not
3684 			 * mandate it out of existence?
3685 			 */
3686 			(void) ip; (void)th; (void) ip_tcp_hlen;
3687 
3688 			return ENOBUFS;
3689 #endif
3690 		} else {
3691 			ip = (struct ip *) (mtod(m0, char *) + offset);
3692 			th = (struct tcphdr *) (mtod(m0, char *) + hlen);
3693 			ip_tcp_hlen = iphl +  (th->th_off << 2);
3694 
3695 			/* Total IP/TCP options, in 32-bit words */
3696 			iptcp_opt_words = (ip_tcp_hlen
3697 					   - sizeof(struct tcphdr)
3698 					   - sizeof(struct ip)) >> 2;
3699 		}
3700 		if (BGE_IS_5750_OR_BEYOND(sc)) {
3701 			th->th_sum = 0;
3702 			csum_flags &= ~(BGE_TXBDFLAG_TCP_UDP_CSUM);
3703 		} else {
3704 			/*
3705 			 * XXX jonathan@NetBSD.org: 5705 untested.
3706 			 * Requires TSO firmware patch for 5701/5703/5704.
3707 			 */
3708 			th->th_sum = in_cksum_phdr(ip->ip_src.s_addr,
3709 			    ip->ip_dst.s_addr, htons(IPPROTO_TCP));
3710 		}
3711 
3712 		mss = m_head->m_pkthdr.segsz;
3713 		txbd_tso_flags |=
3714 		    BGE_TXBDFLAG_CPU_PRE_DMA |
3715 		    BGE_TXBDFLAG_CPU_POST_DMA;
3716 
3717 		/*
3718 		 * Our NIC TSO-assist assumes TSO has standard, optionless
3719 		 * IPv4 and TCP headers, which total 40 bytes. By default,
3720 		 * the NIC copies 40 bytes of IP/TCP header from the
3721 		 * supplied header into the IP/TCP header portion of
3722 		 * each post-TSO-segment. If the supplied packet has IP or
3723 		 * TCP options, we need to tell the NIC to copy those extra
3724 		 * bytes into each  post-TSO header, in addition to the normal
3725 		 * 40-byte IP/TCP header (and to leave space accordingly).
3726 		 * Unfortunately, the driver encoding of option length
3727 		 * varies across different ASIC families.
3728 		 */
3729 		tcp_seg_flags = 0;
3730 		if (iptcp_opt_words) {
3731 			if ( BGE_IS_5705_OR_BEYOND(sc)) {
3732 				tcp_seg_flags =
3733 					iptcp_opt_words << 11;
3734 			} else {
3735 				txbd_tso_flags |=
3736 					iptcp_opt_words << 12;
3737 			}
3738 		}
3739 		maxsegsize = mss | tcp_seg_flags;
3740 		ip->ip_len = htons(mss + ip_tcp_hlen);
3741 
3742 	}	/* TSO setup */
3743 
3744 	/*
3745 	 * Start packing the mbufs in this chain into
3746 	 * the fragment pointers. Stop when we run out
3747 	 * of fragments or hit the end of the mbuf chain.
3748 	 */
3749 	error = bus_dmamap_load_mbuf(sc->bge_dmatag, dmamap, m_head,
3750 	    BUS_DMA_NOWAIT);
3751 	if (error) {
3752 		return(ENOBUFS);
3753 	}
3754 	/*
3755 	 * Sanity check: avoid coming within 16 descriptors
3756 	 * of the end of the ring.
3757 	 */
3758 	if (dmamap->dm_nsegs > (BGE_TX_RING_CNT - sc->bge_txcnt - 16)) {
3759 		BGE_TSO_PRINTF(("%s: "
3760 		    " dmamap_load_mbuf too close to ring wrap\n",
3761 		    device_xname(sc->bge_dev)));
3762 		goto fail_unload;
3763 	}
3764 
3765 	mtag = sc->ethercom.ec_nvlans ?
3766 	    m_tag_find(m_head, PACKET_TAG_VLAN, NULL) : NULL;
3767 
3768 
3769 	/* Iterate over dmap-map fragments. */
3770 	for (i = 0; i < dmamap->dm_nsegs; i++) {
3771 		f = &sc->bge_rdata->bge_tx_ring[frag];
3772 		if (sc->bge_cdata.bge_tx_chain[frag] != NULL)
3773 			break;
3774 
3775 		bge_set_hostaddr(&f->bge_addr, dmamap->dm_segs[i].ds_addr);
3776 		f->bge_len = dmamap->dm_segs[i].ds_len;
3777 
3778 		/*
3779 		 * For 5751 and follow-ons, for TSO we must turn
3780 		 * off checksum-assist flag in the tx-descr, and
3781 		 * supply the ASIC-revision-specific encoding
3782 		 * of TSO flags and segsize.
3783 		 */
3784 		if (use_tso) {
3785 			if (BGE_IS_5750_OR_BEYOND(sc) || i == 0) {
3786 				f->bge_rsvd = maxsegsize;
3787 				f->bge_flags = csum_flags | txbd_tso_flags;
3788 			} else {
3789 				f->bge_rsvd = 0;
3790 				f->bge_flags =
3791 				  (csum_flags | txbd_tso_flags) & 0x0fff;
3792 			}
3793 		} else {
3794 			f->bge_rsvd = 0;
3795 			f->bge_flags = csum_flags;
3796 		}
3797 
3798 		if (mtag != NULL) {
3799 			f->bge_flags |= BGE_TXBDFLAG_VLAN_TAG;
3800 			f->bge_vlan_tag = VLAN_TAG_VALUE(mtag);
3801 		} else {
3802 			f->bge_vlan_tag = 0;
3803 		}
3804 		cur = frag;
3805 		BGE_INC(frag, BGE_TX_RING_CNT);
3806 	}
3807 
3808 	if (i < dmamap->dm_nsegs) {
3809 		BGE_TSO_PRINTF(("%s: reached %d < dm_nsegs %d\n",
3810 		    device_xname(sc->bge_dev), i, dmamap->dm_nsegs));
3811 		goto fail_unload;
3812 	}
3813 
3814 	bus_dmamap_sync(sc->bge_dmatag, dmamap, 0, dmamap->dm_mapsize,
3815 	    BUS_DMASYNC_PREWRITE);
3816 
3817 	if (frag == sc->bge_tx_saved_considx) {
3818 		BGE_TSO_PRINTF(("%s: frag %d = wrapped id %d?\n",
3819 		    device_xname(sc->bge_dev), frag, sc->bge_tx_saved_considx));
3820 
3821 		goto fail_unload;
3822 	}
3823 
3824 	sc->bge_rdata->bge_tx_ring[cur].bge_flags |= BGE_TXBDFLAG_END;
3825 	sc->bge_cdata.bge_tx_chain[cur] = m_head;
3826 	SLIST_REMOVE_HEAD(&sc->txdma_list, link);
3827 	sc->txdma[cur] = dma;
3828 	sc->bge_txcnt += dmamap->dm_nsegs;
3829 
3830 	*txidx = frag;
3831 
3832 	return(0);
3833 
3834  fail_unload:
3835 	bus_dmamap_unload(sc->bge_dmatag, dmamap);
3836 
3837 	return ENOBUFS;
3838 }
3839 
3840 /*
3841  * Main transmit routine. To avoid having to do mbuf copies, we put pointers
3842  * to the mbuf data regions directly in the transmit descriptors.
3843  */
3844 static void
3845 bge_start(struct ifnet *ifp)
3846 {
3847 	struct bge_softc *sc;
3848 	struct mbuf *m_head = NULL;
3849 	u_int32_t prodidx;
3850 	int pkts = 0;
3851 
3852 	sc = ifp->if_softc;
3853 
3854 	if ((ifp->if_flags & (IFF_RUNNING|IFF_OACTIVE)) != IFF_RUNNING)
3855 		return;
3856 
3857 	prodidx = sc->bge_tx_prodidx;
3858 
3859 	while(sc->bge_cdata.bge_tx_chain[prodidx] == NULL) {
3860 		IFQ_POLL(&ifp->if_snd, m_head);
3861 		if (m_head == NULL)
3862 			break;
3863 
3864 #if 0
3865 		/*
3866 		 * XXX
3867 		 * safety overkill.  If this is a fragmented packet chain
3868 		 * with delayed TCP/UDP checksums, then only encapsulate
3869 		 * it if we have enough descriptors to handle the entire
3870 		 * chain at once.
3871 		 * (paranoia -- may not actually be needed)
3872 		 */
3873 		if (m_head->m_flags & M_FIRSTFRAG &&
3874 		    m_head->m_pkthdr.csum_flags & (CSUM_DELAY_DATA)) {
3875 			if ((BGE_TX_RING_CNT - sc->bge_txcnt) <
3876 			    M_CSUM_DATA_IPv4_OFFSET(m_head->m_pkthdr.csum_data) + 16) {
3877 				ifp->if_flags |= IFF_OACTIVE;
3878 				break;
3879 			}
3880 		}
3881 #endif
3882 
3883 		/*
3884 		 * Pack the data into the transmit ring. If we
3885 		 * don't have room, set the OACTIVE flag and wait
3886 		 * for the NIC to drain the ring.
3887 		 */
3888 		if (bge_encap(sc, m_head, &prodidx)) {
3889 			ifp->if_flags |= IFF_OACTIVE;
3890 			break;
3891 		}
3892 
3893 		/* now we are committed to transmit the packet */
3894 		IFQ_DEQUEUE(&ifp->if_snd, m_head);
3895 		pkts++;
3896 
3897 #if NBPFILTER > 0
3898 		/*
3899 		 * If there's a BPF listener, bounce a copy of this frame
3900 		 * to him.
3901 		 */
3902 		if (ifp->if_bpf)
3903 			bpf_mtap(ifp->if_bpf, m_head);
3904 #endif
3905 	}
3906 	if (pkts == 0)
3907 		return;
3908 
3909 	/* Transmit */
3910 	CSR_WRITE_4(sc, BGE_MBX_TX_HOST_PROD0_LO, prodidx);
3911 	if (sc->bge_quirks & BGE_QUIRK_PRODUCER_BUG)	/* 5700 b2 errata */
3912 		CSR_WRITE_4(sc, BGE_MBX_TX_HOST_PROD0_LO, prodidx);
3913 
3914 	sc->bge_tx_prodidx = prodidx;
3915 
3916 	/*
3917 	 * Set a timeout in case the chip goes out to lunch.
3918 	 */
3919 	ifp->if_timer = 5;
3920 }
3921 
3922 static int
3923 bge_init(struct ifnet *ifp)
3924 {
3925 	struct bge_softc *sc = ifp->if_softc;
3926 	const u_int16_t *m;
3927 	int s, error;
3928 
3929 	s = splnet();
3930 
3931 	ifp = &sc->ethercom.ec_if;
3932 
3933 	/* Cancel pending I/O and flush buffers. */
3934 	bge_stop(ifp, 0);
3935 	bge_reset(sc);
3936 	bge_chipinit(sc);
3937 
3938 	/*
3939 	 * Init the various state machines, ring
3940 	 * control blocks and firmware.
3941 	 */
3942 	error = bge_blockinit(sc);
3943 	if (error != 0) {
3944 		aprint_error_dev(sc->bge_dev, "initialization error %d\n",
3945 		    error);
3946 		splx(s);
3947 		return error;
3948 	}
3949 
3950 	ifp = &sc->ethercom.ec_if;
3951 
3952 	/* Specify MTU. */
3953 	CSR_WRITE_4(sc, BGE_RX_MTU, ifp->if_mtu +
3954 	    ETHER_HDR_LEN + ETHER_CRC_LEN + ETHER_VLAN_ENCAP_LEN);
3955 
3956 	/* Load our MAC address. */
3957 	m = (const u_int16_t *)&(CLLADDR(ifp->if_sadl)[0]);
3958 	CSR_WRITE_4(sc, BGE_MAC_ADDR1_LO, htons(m[0]));
3959 	CSR_WRITE_4(sc, BGE_MAC_ADDR1_HI, (htons(m[1]) << 16) | htons(m[2]));
3960 
3961 	/* Enable or disable promiscuous mode as needed. */
3962 	if (ifp->if_flags & IFF_PROMISC) {
3963 		BGE_SETBIT(sc, BGE_RX_MODE, BGE_RXMODE_RX_PROMISC);
3964 	} else {
3965 		BGE_CLRBIT(sc, BGE_RX_MODE, BGE_RXMODE_RX_PROMISC);
3966 	}
3967 
3968 	/* Program multicast filter. */
3969 	bge_setmulti(sc);
3970 
3971 	/* Init RX ring. */
3972 	bge_init_rx_ring_std(sc);
3973 
3974 	/* Init jumbo RX ring. */
3975 	if (ifp->if_mtu > (ETHERMTU + ETHER_HDR_LEN + ETHER_CRC_LEN))
3976 		bge_init_rx_ring_jumbo(sc);
3977 
3978 	/* Init our RX return ring index */
3979 	sc->bge_rx_saved_considx = 0;
3980 
3981 	/* Init TX ring. */
3982 	bge_init_tx_ring(sc);
3983 
3984 	/* Turn on transmitter */
3985 	BGE_SETBIT(sc, BGE_TX_MODE, BGE_TXMODE_ENABLE);
3986 
3987 	/* Turn on receiver */
3988 	BGE_SETBIT(sc, BGE_RX_MODE, BGE_RXMODE_ENABLE);
3989 
3990 	CSR_WRITE_4(sc, BGE_MAX_RX_FRAME_LOWAT, 2);
3991 
3992 	/* Tell firmware we're alive. */
3993 	BGE_SETBIT(sc, BGE_MODE_CTL, BGE_MODECTL_STACKUP);
3994 
3995 	/* Enable host interrupts. */
3996 	BGE_SETBIT(sc, BGE_PCI_MISC_CTL, BGE_PCIMISCCTL_CLEAR_INTA);
3997 	BGE_CLRBIT(sc, BGE_PCI_MISC_CTL, BGE_PCIMISCCTL_MASK_PCI_INTR);
3998 	CSR_WRITE_4(sc, BGE_MBX_IRQ0_LO, 0);
3999 
4000 	bge_ifmedia_upd(ifp);
4001 
4002 	ifp->if_flags |= IFF_RUNNING;
4003 	ifp->if_flags &= ~IFF_OACTIVE;
4004 
4005 	splx(s);
4006 
4007 	callout_reset(&sc->bge_timeout, hz, bge_tick, sc);
4008 
4009 	return 0;
4010 }
4011 
4012 /*
4013  * Set media options.
4014  */
4015 static int
4016 bge_ifmedia_upd(struct ifnet *ifp)
4017 {
4018 	struct bge_softc *sc = ifp->if_softc;
4019 	struct mii_data *mii = &sc->bge_mii;
4020 	struct ifmedia *ifm = &sc->bge_ifmedia;
4021 
4022 	/* If this is a 1000baseX NIC, enable the TBI port. */
4023 	if (sc->bge_tbi) {
4024 		if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER)
4025 			return(EINVAL);
4026 		switch(IFM_SUBTYPE(ifm->ifm_media)) {
4027 		case IFM_AUTO:
4028 			break;
4029 		case IFM_1000_SX:
4030 			if ((ifm->ifm_media & IFM_GMASK) == IFM_FDX) {
4031 				BGE_CLRBIT(sc, BGE_MAC_MODE,
4032 				    BGE_MACMODE_HALF_DUPLEX);
4033 			} else {
4034 				BGE_SETBIT(sc, BGE_MAC_MODE,
4035 				    BGE_MACMODE_HALF_DUPLEX);
4036 			}
4037 			break;
4038 		default:
4039 			return(EINVAL);
4040 		}
4041 		/* XXX 802.3x flow control for 1000BASE-SX */
4042 		return(0);
4043 	}
4044 
4045 	sc->bge_link = 0;
4046 	mii_mediachg(mii);
4047 
4048 	return(0);
4049 }
4050 
4051 /*
4052  * Report current media status.
4053  */
4054 static void
4055 bge_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
4056 {
4057 	struct bge_softc *sc = ifp->if_softc;
4058 	struct mii_data *mii = &sc->bge_mii;
4059 
4060 	if (sc->bge_tbi) {
4061 		ifmr->ifm_status = IFM_AVALID;
4062 		ifmr->ifm_active = IFM_ETHER;
4063 		if (CSR_READ_4(sc, BGE_MAC_STS) &
4064 		    BGE_MACSTAT_TBI_PCS_SYNCHED)
4065 			ifmr->ifm_status |= IFM_ACTIVE;
4066 		ifmr->ifm_active |= IFM_1000_SX;
4067 		if (CSR_READ_4(sc, BGE_MAC_MODE) & BGE_MACMODE_HALF_DUPLEX)
4068 			ifmr->ifm_active |= IFM_HDX;
4069 		else
4070 			ifmr->ifm_active |= IFM_FDX;
4071 		return;
4072 	}
4073 
4074 	mii_pollstat(mii);
4075 	ifmr->ifm_status = mii->mii_media_status;
4076 	ifmr->ifm_active = (mii->mii_media_active & ~IFM_ETH_FMASK) |
4077 	    sc->bge_flowflags;
4078 }
4079 
4080 static int
4081 bge_ioctl(struct ifnet *ifp, u_long command, void *data)
4082 {
4083 	struct bge_softc *sc = ifp->if_softc;
4084 	struct ifreq *ifr = (struct ifreq *) data;
4085 	int s, error = 0;
4086 	struct mii_data *mii;
4087 
4088 	s = splnet();
4089 
4090 	switch(command) {
4091 	case SIOCSIFFLAGS:
4092 		if (ifp->if_flags & IFF_UP) {
4093 			/*
4094 			 * If only the state of the PROMISC flag changed,
4095 			 * then just use the 'set promisc mode' command
4096 			 * instead of reinitializing the entire NIC. Doing
4097 			 * a full re-init means reloading the firmware and
4098 			 * waiting for it to start up, which may take a
4099 			 * second or two.
4100 			 */
4101 			if (ifp->if_flags & IFF_RUNNING &&
4102 			    ifp->if_flags & IFF_PROMISC &&
4103 			    !(sc->bge_if_flags & IFF_PROMISC)) {
4104 				BGE_SETBIT(sc, BGE_RX_MODE,
4105 				    BGE_RXMODE_RX_PROMISC);
4106 			} else if (ifp->if_flags & IFF_RUNNING &&
4107 			    !(ifp->if_flags & IFF_PROMISC) &&
4108 			    sc->bge_if_flags & IFF_PROMISC) {
4109 				BGE_CLRBIT(sc, BGE_RX_MODE,
4110 				    BGE_RXMODE_RX_PROMISC);
4111 			} else if (!(sc->bge_if_flags & IFF_UP))
4112 				bge_init(ifp);
4113 		} else {
4114 			if (ifp->if_flags & IFF_RUNNING)
4115 				bge_stop(ifp, 1);
4116 		}
4117 		sc->bge_if_flags = ifp->if_flags;
4118 		error = 0;
4119 		break;
4120 	case SIOCSIFMEDIA:
4121 		/* XXX Flow control is not supported for 1000BASE-SX */
4122 		if (sc->bge_tbi) {
4123 			ifr->ifr_media &= ~IFM_ETH_FMASK;
4124 			sc->bge_flowflags = 0;
4125 		}
4126 
4127 		/* Flow control requires full-duplex mode. */
4128 		if (IFM_SUBTYPE(ifr->ifr_media) == IFM_AUTO ||
4129 		    (ifr->ifr_media & IFM_FDX) == 0) {
4130 		    	ifr->ifr_media &= ~IFM_ETH_FMASK;
4131 		}
4132 		if (IFM_SUBTYPE(ifr->ifr_media) != IFM_AUTO) {
4133 			if ((ifr->ifr_media & IFM_ETH_FMASK) == IFM_FLOW) {
4134 				/* We an do both TXPAUSE and RXPAUSE. */
4135 				ifr->ifr_media |=
4136 				    IFM_ETH_TXPAUSE | IFM_ETH_RXPAUSE;
4137 			}
4138 			sc->bge_flowflags = ifr->ifr_media & IFM_ETH_FMASK;
4139 		}
4140 		/* FALLTHROUGH */
4141 	case SIOCGIFMEDIA:
4142 		if (sc->bge_tbi) {
4143 			error = ifmedia_ioctl(ifp, ifr, &sc->bge_ifmedia,
4144 			    command);
4145 		} else {
4146 			mii = &sc->bge_mii;
4147 			error = ifmedia_ioctl(ifp, ifr, &mii->mii_media,
4148 			    command);
4149 		}
4150 		break;
4151 	default:
4152 		error = ether_ioctl(ifp, command, data);
4153 		if (error == ENETRESET) {
4154 			if (ifp->if_flags & IFF_RUNNING)
4155 				bge_setmulti(sc);
4156 			error = 0;
4157 		}
4158 		break;
4159 	}
4160 
4161 	splx(s);
4162 
4163 	return(error);
4164 }
4165 
4166 static void
4167 bge_watchdog(struct ifnet *ifp)
4168 {
4169 	struct bge_softc *sc;
4170 
4171 	sc = ifp->if_softc;
4172 
4173 	aprint_error_dev(sc->bge_dev, "watchdog timeout -- resetting\n");
4174 
4175 	ifp->if_flags &= ~IFF_RUNNING;
4176 	bge_init(ifp);
4177 
4178 	ifp->if_oerrors++;
4179 }
4180 
4181 static void
4182 bge_stop_block(struct bge_softc *sc, bus_addr_t reg, uint32_t bit)
4183 {
4184 	int i;
4185 
4186 	BGE_CLRBIT(sc, reg, bit);
4187 
4188 	for (i = 0; i < BGE_TIMEOUT; i++) {
4189 		if ((CSR_READ_4(sc, reg) & bit) == 0)
4190 			return;
4191 		delay(100);
4192 		if (sc->bge_pcie)
4193 		  DELAY(1000);
4194 	}
4195 
4196 	aprint_error_dev(sc->bge_dev,
4197 	    "block failed to stop: reg 0x%lx, bit 0x%08x\n", (u_long)reg, bit);
4198 }
4199 
4200 /*
4201  * Stop the adapter and free any mbufs allocated to the
4202  * RX and TX lists.
4203  */
4204 static void
4205 bge_stop(struct ifnet *ifp, int disable)
4206 {
4207 	struct bge_softc *sc = ifp->if_softc;
4208 
4209 	callout_stop(&sc->bge_timeout);
4210 
4211 	/*
4212 	 * Disable all of the receiver blocks
4213 	 */
4214 	bge_stop_block(sc, BGE_RX_MODE, BGE_RXMODE_ENABLE);
4215 	bge_stop_block(sc, BGE_RBDI_MODE, BGE_RBDIMODE_ENABLE);
4216 	bge_stop_block(sc, BGE_RXLP_MODE, BGE_RXLPMODE_ENABLE);
4217 	if ((sc->bge_quirks & BGE_QUIRK_5705_CORE) == 0) {
4218 		bge_stop_block(sc, BGE_RXLS_MODE, BGE_RXLSMODE_ENABLE);
4219 	}
4220 	bge_stop_block(sc, BGE_RDBDI_MODE, BGE_RBDIMODE_ENABLE);
4221 	bge_stop_block(sc, BGE_RDC_MODE, BGE_RDCMODE_ENABLE);
4222 	bge_stop_block(sc, BGE_RBDC_MODE, BGE_RBDCMODE_ENABLE);
4223 
4224 	/*
4225 	 * Disable all of the transmit blocks
4226 	 */
4227 	bge_stop_block(sc, BGE_SRS_MODE, BGE_SRSMODE_ENABLE);
4228 	bge_stop_block(sc, BGE_SBDI_MODE, BGE_SBDIMODE_ENABLE);
4229 	bge_stop_block(sc, BGE_SDI_MODE, BGE_SDIMODE_ENABLE);
4230 	bge_stop_block(sc, BGE_RDMA_MODE, BGE_RDMAMODE_ENABLE);
4231 	bge_stop_block(sc, BGE_SDC_MODE, BGE_SDCMODE_ENABLE);
4232 	if ((sc->bge_quirks & BGE_QUIRK_5705_CORE) == 0) {
4233 		bge_stop_block(sc, BGE_DMAC_MODE, BGE_DMACMODE_ENABLE);
4234 	}
4235 	bge_stop_block(sc, BGE_SBDC_MODE, BGE_SBDCMODE_ENABLE);
4236 
4237 	/*
4238 	 * Shut down all of the memory managers and related
4239 	 * state machines.
4240 	 */
4241 	bge_stop_block(sc, BGE_HCC_MODE, BGE_HCCMODE_ENABLE);
4242 	bge_stop_block(sc, BGE_WDMA_MODE, BGE_WDMAMODE_ENABLE);
4243 	if ((sc->bge_quirks & BGE_QUIRK_5705_CORE) == 0) {
4244 		bge_stop_block(sc, BGE_MBCF_MODE, BGE_MBCFMODE_ENABLE);
4245 	}
4246 
4247 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0xFFFFFFFF);
4248 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0);
4249 
4250 	if ((sc->bge_quirks & BGE_QUIRK_5705_CORE) == 0) {
4251 		bge_stop_block(sc, BGE_BMAN_MODE, BGE_BMANMODE_ENABLE);
4252 		bge_stop_block(sc, BGE_MARB_MODE, BGE_MARBMODE_ENABLE);
4253 	}
4254 
4255 	/* Disable host interrupts. */
4256 	BGE_SETBIT(sc, BGE_PCI_MISC_CTL, BGE_PCIMISCCTL_MASK_PCI_INTR);
4257 	CSR_WRITE_4(sc, BGE_MBX_IRQ0_LO, 1);
4258 
4259 	/*
4260 	 * Tell firmware we're shutting down.
4261 	 */
4262 	BGE_CLRBIT(sc, BGE_MODE_CTL, BGE_MODECTL_STACKUP);
4263 
4264 	/* Free the RX lists. */
4265 	bge_free_rx_ring_std(sc);
4266 
4267 	/* Free jumbo RX list. */
4268 	bge_free_rx_ring_jumbo(sc);
4269 
4270 	/* Free TX buffers. */
4271 	bge_free_tx_ring(sc);
4272 
4273 	/*
4274 	 * Isolate/power down the PHY.
4275 	 */
4276 	if (!sc->bge_tbi)
4277 		mii_down(&sc->bge_mii);
4278 
4279 	sc->bge_link = 0;
4280 
4281 	sc->bge_tx_saved_considx = BGE_TXCONS_UNSET;
4282 
4283 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
4284 }
4285 
4286 static int
4287 sysctl_bge_verify(SYSCTLFN_ARGS)
4288 {
4289 	int error, t;
4290 	struct sysctlnode node;
4291 
4292 	node = *rnode;
4293 	t = *(int*)rnode->sysctl_data;
4294 	node.sysctl_data = &t;
4295 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
4296 	if (error || newp == NULL)
4297 		return (error);
4298 
4299 #if 0
4300 	DPRINTF2(("%s: t = %d, nodenum = %d, rnodenum = %d\n", __func__, t,
4301 	    node.sysctl_num, rnode->sysctl_num));
4302 #endif
4303 
4304 	if (node.sysctl_num == bge_rxthresh_nodenum) {
4305 		if (t < 0 || t >= NBGE_RX_THRESH)
4306 			return (EINVAL);
4307 		bge_update_all_threshes(t);
4308 	} else
4309 		return (EINVAL);
4310 
4311 	*(int*)rnode->sysctl_data = t;
4312 
4313 	return (0);
4314 }
4315 
4316 /*
4317  * Set up sysctl(3) MIB, hw.bge.*.
4318  *
4319  * TBD condition SYSCTL_PERMANENT on being an LKM or not
4320  */
4321 SYSCTL_SETUP(sysctl_bge, "sysctl bge subtree setup")
4322 {
4323 	int rc, bge_root_num;
4324 	const struct sysctlnode *node;
4325 
4326 	if ((rc = sysctl_createv(clog, 0, NULL, NULL,
4327 	    CTLFLAG_PERMANENT, CTLTYPE_NODE, "hw", NULL,
4328 	    NULL, 0, NULL, 0, CTL_HW, CTL_EOL)) != 0) {
4329 		goto err;
4330 	}
4331 
4332 	if ((rc = sysctl_createv(clog, 0, NULL, &node,
4333 	    CTLFLAG_PERMANENT, CTLTYPE_NODE, "bge",
4334 	    SYSCTL_DESCR("BGE interface controls"),
4335 	    NULL, 0, NULL, 0, CTL_HW, CTL_CREATE, CTL_EOL)) != 0) {
4336 		goto err;
4337 	}
4338 
4339 	bge_root_num = node->sysctl_num;
4340 
4341 	/* BGE Rx interrupt mitigation level */
4342 	if ((rc = sysctl_createv(clog, 0, NULL, &node,
4343 	    CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
4344 	    CTLTYPE_INT, "rx_lvl",
4345 	    SYSCTL_DESCR("BGE receive interrupt mitigation level"),
4346 	    sysctl_bge_verify, 0,
4347 	    &bge_rx_thresh_lvl,
4348 	    0, CTL_HW, bge_root_num, CTL_CREATE,
4349 	    CTL_EOL)) != 0) {
4350 		goto err;
4351 	}
4352 
4353 	bge_rxthresh_nodenum = node->sysctl_num;
4354 
4355 	return;
4356 
4357 err:
4358 	aprint_error("%s: sysctl_createv failed (rc = %d)\n", __func__, rc);
4359 }
4360