xref: /dragonfly/sys/dev/netif/re/if_re.c (revision d600454b)
1 /*
2  * Copyright (c) 2004
3  *	Joerg Sonnenberger <joerg@bec.de>.  All rights reserved.
4  *
5  * Copyright (c) 1997, 1998-2003
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: src/sys/dev/re/if_re.c,v 1.25 2004/06/09 14:34:01 naddy Exp $
36  * $DragonFly: src/sys/dev/netif/re/if_re.c,v 1.21 2005/12/31 14:08:00 sephe Exp $
37  */
38 
39 /*
40  * RealTek 8139C+/8169/8169S/8110S PCI NIC driver
41  *
42  * Written by Bill Paul <wpaul@windriver.com>
43  * Senior Networking Software Engineer
44  * Wind River Systems
45  */
46 
47 /*
48  * This driver is designed to support RealTek's next generation of
49  * 10/100 and 10/100/1000 PCI ethernet controllers. There are currently
50  * four devices in this family: the RTL8139C+, the RTL8169, the RTL8169S
51  * and the RTL8110S.
52  *
53  * The 8139C+ is a 10/100 ethernet chip. It is backwards compatible
54  * with the older 8139 family, however it also supports a special
55  * C+ mode of operation that provides several new performance enhancing
56  * features. These include:
57  *
58  *	o Descriptor based DMA mechanism. Each descriptor represents
59  *	  a single packet fragment. Data buffers may be aligned on
60  *	  any byte boundary.
61  *
62  *	o 64-bit DMA
63  *
64  *	o TCP/IP checksum offload for both RX and TX
65  *
66  *	o High and normal priority transmit DMA rings
67  *
68  *	o VLAN tag insertion and extraction
69  *
70  *	o TCP large send (segmentation offload)
71  *
72  * Like the 8139, the 8139C+ also has a built-in 10/100 PHY. The C+
73  * programming API is fairly straightforward. The RX filtering, EEPROM
74  * access and PHY access is the same as it is on the older 8139 series
75  * chips.
76  *
77  * The 8169 is a 64-bit 10/100/1000 gigabit ethernet MAC. It has almost the
78  * same programming API and feature set as the 8139C+ with the following
79  * differences and additions:
80  *
81  *	o 1000Mbps mode
82  *
83  *	o Jumbo frames
84  *
85  * 	o GMII and TBI ports/registers for interfacing with copper
86  *	  or fiber PHYs
87  *
88  *      o RX and TX DMA rings can have up to 1024 descriptors
89  *        (the 8139C+ allows a maximum of 64)
90  *
91  *	o Slight differences in register layout from the 8139C+
92  *
93  * The TX start and timer interrupt registers are at different locations
94  * on the 8169 than they are on the 8139C+. Also, the status word in the
95  * RX descriptor has a slightly different bit layout. The 8169 does not
96  * have a built-in PHY. Most reference boards use a Marvell 88E1000 'Alaska'
97  * copper gigE PHY.
98  *
99  * The 8169S/8110S 10/100/1000 devices have built-in copper gigE PHYs
100  * (the 'S' stands for 'single-chip'). These devices have the same
101  * programming API as the older 8169, but also have some vendor-specific
102  * registers for the on-board PHY. The 8110S is a LAN-on-motherboard
103  * part designed to be pin-compatible with the RealTek 8100 10/100 chip.
104  *
105  * This driver takes advantage of the RX and TX checksum offload and
106  * VLAN tag insertion/extraction features. It also implements TX
107  * interrupt moderation using the timer interrupt registers, which
108  * significantly reduces TX interrupt load. There is also support
109  * for jumbo frames, however the 8169/8169S/8110S can not transmit
110  * jumbo frames larger than 7440, so the max MTU possible with this
111  * driver is 7422 bytes.
112  */
113 
114 #include "opt_polling.h"
115 
116 #include <sys/param.h>
117 #include <sys/endian.h>
118 #include <sys/systm.h>
119 #include <sys/sockio.h>
120 #include <sys/mbuf.h>
121 #include <sys/malloc.h>
122 #include <sys/module.h>
123 #include <sys/kernel.h>
124 #include <sys/socket.h>
125 #include <sys/serialize.h>
126 #include <sys/thread2.h>
127 
128 #include <net/if.h>
129 #include <net/ifq_var.h>
130 #include <net/if_arp.h>
131 #include <net/ethernet.h>
132 #include <net/if_dl.h>
133 #include <net/if_media.h>
134 #include <net/if_types.h>
135 #include <net/vlan/if_vlan_var.h>
136 
137 #include <net/bpf.h>
138 
139 #include <machine/bus_pio.h>
140 #include <machine/bus_memio.h>
141 #include <machine/bus.h>
142 #include <machine/resource.h>
143 #include <sys/bus.h>
144 #include <sys/rman.h>
145 
146 #include <dev/netif/mii_layer/mii.h>
147 #include <dev/netif/mii_layer/miivar.h>
148 
149 #include <bus/pci/pcireg.h>
150 #include <bus/pci/pcivar.h>
151 
152 /* "controller miibus0" required.  See GENERIC if you get errors here. */
153 #include "miibus_if.h"
154 
155 #include <dev/netif/re/if_rereg.h>
156 
157 /*
158  * The hardware supports checksumming but, as usual, some chipsets screw it
159  * all up and produce bogus packets, so we disable it by default.
160  */
161 #define RE_CSUM_FEATURES    (CSUM_IP | CSUM_TCP | CSUM_UDP)
162 #define RE_DISABLE_HWCSUM
163 
164 /*
165  * Various supported device vendors/types and their names.
166  */
167 static struct re_type re_devs[] = {
168 	{ DLINK_VENDORID, DLINK_DEVICEID_528T, RE_HWREV_8169S,
169 		"D-Link DGE-528(T) Gigabit Ethernet Adapter" },
170 	{ RT_VENDORID, RT_DEVICEID_8139, RE_HWREV_8139CPLUS,
171 		"RealTek 8139C+ 10/100BaseTX" },
172 	{ RT_VENDORID, RT_DEVICEID_8169, RE_HWREV_8169,
173 		"RealTek 8169 Gigabit Ethernet" },
174 	{ RT_VENDORID, RT_DEVICEID_8169, RE_HWREV_8169S,
175 		"RealTek 8169S Single-chip Gigabit Ethernet" },
176 	{ RT_VENDORID, RT_DEVICEID_8169, RE_HWREV_8110S,
177 		"RealTek 8110S Single-chip Gigabit Ethernet" },
178 	{ 0, 0, 0, NULL }
179 };
180 
181 static struct re_hwrev re_hwrevs[] = {
182 	{ RE_HWREV_8139CPLUS, RE_8139CPLUS, "C+"},
183 	{ RE_HWREV_8169, RE_8169, "8169"},
184 	{ RE_HWREV_8169S, RE_8169, "8169S"},
185 	{ RE_HWREV_8110S, RE_8169, "8110S"},
186 	{ 0, 0, NULL }
187 };
188 
189 static int	re_probe(device_t);
190 static int	re_attach(device_t);
191 static int	re_detach(device_t);
192 
193 static int	re_encap(struct re_softc *, struct mbuf **, int *, int *);
194 
195 static void	re_dma_map_addr(void *, bus_dma_segment_t *, int, int);
196 static void	re_dma_map_desc(void *, bus_dma_segment_t *, int,
197 				bus_size_t, int);
198 static int	re_allocmem(device_t, struct re_softc *);
199 static int	re_newbuf(struct re_softc *, int, struct mbuf *);
200 static int	re_rx_list_init(struct re_softc *);
201 static int	re_tx_list_init(struct re_softc *);
202 static void	re_rxeof(struct re_softc *);
203 static void	re_txeof(struct re_softc *);
204 static void	re_intr(void *);
205 static void	re_tick(void *);
206 static void	re_tick_serialized(void *);
207 static void	re_start(struct ifnet *);
208 static int	re_ioctl(struct ifnet *, u_long, caddr_t, struct ucred *);
209 static void	re_init(void *);
210 static void	re_stop(struct re_softc *);
211 static void	re_watchdog(struct ifnet *);
212 static int	re_suspend(device_t);
213 static int	re_resume(device_t);
214 static void	re_shutdown(device_t);
215 static int	re_ifmedia_upd(struct ifnet *);
216 static void	re_ifmedia_sts(struct ifnet *, struct ifmediareq *);
217 
218 static void	re_eeprom_putbyte(struct re_softc *, int);
219 static void	re_eeprom_getword(struct re_softc *, int, u_int16_t *);
220 static void	re_read_eeprom(struct re_softc *, caddr_t, int, int, int);
221 static int	re_gmii_readreg(device_t, int, int);
222 static int	re_gmii_writereg(device_t, int, int, int);
223 
224 static int	re_miibus_readreg(device_t, int, int);
225 static int	re_miibus_writereg(device_t, int, int, int);
226 static void	re_miibus_statchg(device_t);
227 
228 static void	re_setmulti(struct re_softc *);
229 static void	re_reset(struct re_softc *);
230 
231 static int	re_diag(struct re_softc *);
232 #ifdef DEVICE_POLLING
233 static void	re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count);
234 #endif
235 
236 static device_method_t re_methods[] = {
237 	/* Device interface */
238 	DEVMETHOD(device_probe,		re_probe),
239 	DEVMETHOD(device_attach,	re_attach),
240 	DEVMETHOD(device_detach,	re_detach),
241 	DEVMETHOD(device_suspend,	re_suspend),
242 	DEVMETHOD(device_resume,	re_resume),
243 	DEVMETHOD(device_shutdown,	re_shutdown),
244 
245 	/* bus interface */
246 	DEVMETHOD(bus_print_child,	bus_generic_print_child),
247 	DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
248 
249 	/* MII interface */
250 	DEVMETHOD(miibus_readreg,	re_miibus_readreg),
251 	DEVMETHOD(miibus_writereg,	re_miibus_writereg),
252 	DEVMETHOD(miibus_statchg,	re_miibus_statchg),
253 
254 	{ 0, 0 }
255 };
256 
257 static driver_t re_driver = {
258 	"re",
259 	re_methods,
260 	sizeof(struct re_softc)
261 };
262 
263 static devclass_t re_devclass;
264 
265 DECLARE_DUMMY_MODULE(if_re);
266 DRIVER_MODULE(if_re, pci, re_driver, re_devclass, 0, 0);
267 DRIVER_MODULE(if_re, cardbus, re_driver, re_devclass, 0, 0);
268 DRIVER_MODULE(miibus, re, miibus_driver, miibus_devclass, 0, 0);
269 
270 #define EE_SET(x)	\
271 	CSR_WRITE_1(sc, RE_EECMD, CSR_READ_1(sc, RE_EECMD) | (x))
272 
273 #define EE_CLR(x)	\
274 	CSR_WRITE_1(sc, RE_EECMD, CSR_READ_1(sc, RE_EECMD) & ~(x))
275 
276 /*
277  * Send a read command and address to the EEPROM, check for ACK.
278  */
279 static void
280 re_eeprom_putbyte(struct re_softc *sc, int addr)
281 {
282 	int d, i;
283 
284 	d = addr | sc->re_eecmd_read;
285 
286 	/*
287 	 * Feed in each bit and strobe the clock.
288 	 */
289 	for (i = 0x400; i != 0; i >>= 1) {
290 		if (d & i)
291 			EE_SET(RE_EE_DATAIN);
292 		else
293 			EE_CLR(RE_EE_DATAIN);
294 		DELAY(100);
295 		EE_SET(RE_EE_CLK);
296 		DELAY(150);
297 		EE_CLR(RE_EE_CLK);
298 		DELAY(100);
299 	}
300 }
301 
302 /*
303  * Read a word of data stored in the EEPROM at address 'addr.'
304  */
305 static void
306 re_eeprom_getword(struct re_softc *sc, int addr, uint16_t *dest)
307 {
308 	int i;
309 	uint16_t word = 0;
310 
311 	/* Enter EEPROM access mode. */
312 	CSR_WRITE_1(sc, RE_EECMD, RE_EEMODE_PROGRAM|RE_EE_SEL);
313 
314 	/*
315 	 * Send address of word we want to read.
316 	 */
317 	re_eeprom_putbyte(sc, addr);
318 
319 	CSR_WRITE_1(sc, RE_EECMD, RE_EEMODE_PROGRAM|RE_EE_SEL);
320 
321 	/*
322 	 * Start reading bits from EEPROM.
323 	 */
324 	for (i = 0x8000; i != 0; i >>= 1) {
325 		EE_SET(RE_EE_CLK);
326 		DELAY(100);
327 		if (CSR_READ_1(sc, RE_EECMD) & RE_EE_DATAOUT)
328 			word |= i;
329 		EE_CLR(RE_EE_CLK);
330 		DELAY(100);
331 	}
332 
333 	/* Turn off EEPROM access mode. */
334 	CSR_WRITE_1(sc, RE_EECMD, RE_EEMODE_OFF);
335 
336 	*dest = word;
337 }
338 
339 /*
340  * Read a sequence of words from the EEPROM.
341  */
342 static void
343 re_read_eeprom(struct re_softc *sc, caddr_t dest, int off, int cnt, int swap)
344 {
345 	int i;
346 	uint16_t word = 0, *ptr;
347 
348 	for (i = 0; i < cnt; i++) {
349 		re_eeprom_getword(sc, off + i, &word);
350 		ptr = (u_int16_t *)(dest + (i * 2));
351 		if (swap)
352 			*ptr = be16toh(word);
353 		else
354 			*ptr = word;
355 	}
356 }
357 
358 static int
359 re_gmii_readreg(device_t dev, int phy, int reg)
360 {
361 	struct re_softc *sc = device_get_softc(dev);
362 	u_int32_t rval;
363 	int i;
364 
365 	if (phy != 1)
366 		return(0);
367 
368 	/* Let the rgephy driver read the GMEDIASTAT register */
369 
370 	if (reg == RE_GMEDIASTAT)
371 		return(CSR_READ_1(sc, RE_GMEDIASTAT));
372 
373 	CSR_WRITE_4(sc, RE_PHYAR, reg << 16);
374 	DELAY(1000);
375 
376 	for (i = 0; i < RE_TIMEOUT; i++) {
377 		rval = CSR_READ_4(sc, RE_PHYAR);
378 		if (rval & RE_PHYAR_BUSY)
379 			break;
380 		DELAY(100);
381 	}
382 
383 	if (i == RE_TIMEOUT) {
384 		device_printf(dev, "PHY read failed\n");
385 		return(0);
386 	}
387 
388 	return(rval & RE_PHYAR_PHYDATA);
389 }
390 
391 static int
392 re_gmii_writereg(device_t dev, int phy, int reg, int data)
393 {
394 	struct re_softc *sc = device_get_softc(dev);
395 	uint32_t rval;
396 	int i;
397 
398 	CSR_WRITE_4(sc, RE_PHYAR,
399 		    (reg << 16) | (data & RE_PHYAR_PHYDATA) | RE_PHYAR_BUSY);
400 	DELAY(1000);
401 
402 	for (i = 0; i < RE_TIMEOUT; i++) {
403 		rval = CSR_READ_4(sc, RE_PHYAR);
404 		if ((rval & RE_PHYAR_BUSY) == 0)
405 			break;
406 		DELAY(100);
407 	}
408 
409 	if (i == RE_TIMEOUT)
410 		device_printf(dev, "PHY write failed\n");
411 
412 	return(0);
413 }
414 
415 static int
416 re_miibus_readreg(device_t dev, int phy, int reg)
417 {
418 	struct re_softc	*sc = device_get_softc(dev);
419 	uint16_t rval = 0;
420 	uint16_t re8139_reg = 0;
421 
422 	if (sc->re_type == RE_8169) {
423 		rval = re_gmii_readreg(dev, phy, reg);
424 		return(rval);
425 	}
426 
427 	/* Pretend the internal PHY is only at address 0 */
428 	if (phy)
429 		return(0);
430 
431 	switch(reg) {
432 	case MII_BMCR:
433 		re8139_reg = RE_BMCR;
434 		break;
435 	case MII_BMSR:
436 		re8139_reg = RE_BMSR;
437 		break;
438 	case MII_ANAR:
439 		re8139_reg = RE_ANAR;
440 		break;
441 	case MII_ANER:
442 		re8139_reg = RE_ANER;
443 		break;
444 	case MII_ANLPAR:
445 		re8139_reg = RE_LPAR;
446 		break;
447 	case MII_PHYIDR1:
448 	case MII_PHYIDR2:
449 		return(0);
450 	/*
451 	 * Allow the rlphy driver to read the media status
452 	 * register. If we have a link partner which does not
453 	 * support NWAY, this is the register which will tell
454 	 * us the results of parallel detection.
455 	 */
456 	case RE_MEDIASTAT:
457 		return(CSR_READ_1(sc, RE_MEDIASTAT));
458 	default:
459 		device_printf(dev, "bad phy register\n");
460 		return(0);
461 	}
462 	rval = CSR_READ_2(sc, re8139_reg);
463 	return(rval);
464 }
465 
466 static int
467 re_miibus_writereg(device_t dev, int phy, int reg, int data)
468 {
469 	struct re_softc *sc= device_get_softc(dev);
470 	u_int16_t re8139_reg = 0;
471 
472 	if (sc->re_type == RE_8169)
473 		return(re_gmii_writereg(dev, phy, reg, data));
474 
475 	/* Pretend the internal PHY is only at address 0 */
476 	if (phy)
477 		return(0);
478 
479 	switch(reg) {
480 	case MII_BMCR:
481 		re8139_reg = RE_BMCR;
482 		break;
483 	case MII_BMSR:
484 		re8139_reg = RE_BMSR;
485 		break;
486 	case MII_ANAR:
487 		re8139_reg = RE_ANAR;
488 		break;
489 	case MII_ANER:
490 		re8139_reg = RE_ANER;
491 		break;
492 	case MII_ANLPAR:
493 		re8139_reg = RE_LPAR;
494 		break;
495 	case MII_PHYIDR1:
496 	case MII_PHYIDR2:
497 		return(0);
498 	default:
499 		device_printf(dev, "bad phy register\n");
500 		return(0);
501 	}
502 	CSR_WRITE_2(sc, re8139_reg, data);
503 	return(0);
504 }
505 
506 static void
507 re_miibus_statchg(device_t dev)
508 {
509 }
510 
511 /*
512  * Program the 64-bit multicast hash filter.
513  */
514 static void
515 re_setmulti(struct re_softc *sc)
516 {
517 	struct ifnet *ifp = &sc->arpcom.ac_if;
518 	int h = 0;
519 	uint32_t hashes[2] = { 0, 0 };
520 	struct ifmultiaddr *ifma;
521 	uint32_t rxfilt;
522 	int mcnt = 0;
523 
524 	rxfilt = CSR_READ_4(sc, RE_RXCFG);
525 
526 	if (ifp->if_flags & IFF_ALLMULTI || ifp->if_flags & IFF_PROMISC) {
527 		rxfilt |= RE_RXCFG_RX_MULTI;
528 		CSR_WRITE_4(sc, RE_RXCFG, rxfilt);
529 		CSR_WRITE_4(sc, RE_MAR0, 0xFFFFFFFF);
530 		CSR_WRITE_4(sc, RE_MAR4, 0xFFFFFFFF);
531 		return;
532 	}
533 
534 	/* first, zot all the existing hash bits */
535 	CSR_WRITE_4(sc, RE_MAR0, 0);
536 	CSR_WRITE_4(sc, RE_MAR4, 0);
537 
538 	/* now program new ones */
539 	LIST_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
540 		if (ifma->ifma_addr->sa_family != AF_LINK)
541 			continue;
542 		h = ether_crc32_be(LLADDR((struct sockaddr_dl *)
543 		    ifma->ifma_addr), ETHER_ADDR_LEN) >> 26;
544 		if (h < 32)
545 			hashes[0] |= (1 << h);
546 		else
547 			hashes[1] |= (1 << (h - 32));
548 		mcnt++;
549 	}
550 
551 	if (mcnt)
552 		rxfilt |= RE_RXCFG_RX_MULTI;
553 	else
554 		rxfilt &= ~RE_RXCFG_RX_MULTI;
555 
556 	CSR_WRITE_4(sc, RE_RXCFG, rxfilt);
557 	CSR_WRITE_4(sc, RE_MAR0, hashes[0]);
558 	CSR_WRITE_4(sc, RE_MAR4, hashes[1]);
559 }
560 
561 static void
562 re_reset(struct re_softc *sc)
563 {
564 	int i;
565 
566 	CSR_WRITE_1(sc, RE_COMMAND, RE_CMD_RESET);
567 
568 	for (i = 0; i < RE_TIMEOUT; i++) {
569 		DELAY(10);
570 		if ((CSR_READ_1(sc, RE_COMMAND) & RE_CMD_RESET) == 0)
571 			break;
572 	}
573 	if (i == RE_TIMEOUT)
574 		if_printf(&sc->arpcom.ac_if, "reset never completed!\n");
575 
576 	CSR_WRITE_1(sc, 0x82, 1);
577 }
578 
579 /*
580  * The following routine is designed to test for a defect on some
581  * 32-bit 8169 cards. Some of these NICs have the REQ64# and ACK64#
582  * lines connected to the bus, however for a 32-bit only card, they
583  * should be pulled high. The result of this defect is that the
584  * NIC will not work right if you plug it into a 64-bit slot: DMA
585  * operations will be done with 64-bit transfers, which will fail
586  * because the 64-bit data lines aren't connected.
587  *
588  * There's no way to work around this (short of talking a soldering
589  * iron to the board), however we can detect it. The method we use
590  * here is to put the NIC into digital loopback mode, set the receiver
591  * to promiscuous mode, and then try to send a frame. We then compare
592  * the frame data we sent to what was received. If the data matches,
593  * then the NIC is working correctly, otherwise we know the user has
594  * a defective NIC which has been mistakenly plugged into a 64-bit PCI
595  * slot. In the latter case, there's no way the NIC can work correctly,
596  * so we print out a message on the console and abort the device attach.
597  */
598 
599 static int
600 re_diag(struct re_softc *sc)
601 {
602 	struct ifnet *ifp = &sc->arpcom.ac_if;
603 	struct mbuf *m0;
604 	struct ether_header *eh;
605 	struct re_desc *cur_rx;
606 	uint16_t status;
607 	uint32_t rxstat;
608 	int total_len, i, error = 0;
609 	uint8_t dst[ETHER_ADDR_LEN] = { 0x00, 'h', 'e', 'l', 'l', 'o' };
610 	uint8_t src[ETHER_ADDR_LEN] = { 0x00, 'w', 'o', 'r', 'l', 'd' };
611 
612 	/* Allocate a single mbuf */
613 
614 	MGETHDR(m0, MB_DONTWAIT, MT_DATA);
615 	if (m0 == NULL)
616 		return(ENOBUFS);
617 
618 	/*
619 	 * Initialize the NIC in test mode. This sets the chip up
620 	 * so that it can send and receive frames, but performs the
621 	 * following special functions:
622 	 * - Puts receiver in promiscuous mode
623 	 * - Enables digital loopback mode
624 	 * - Leaves interrupts turned off
625 	 */
626 
627 	ifp->if_flags |= IFF_PROMISC;
628 	sc->re_testmode = 1;
629 	re_init(sc);
630 	re_stop(sc);
631 	DELAY(100000);
632 	re_init(sc);
633 
634 	/* Put some data in the mbuf */
635 
636 	eh = mtod(m0, struct ether_header *);
637 	bcopy (dst, eh->ether_dhost, ETHER_ADDR_LEN);
638 	bcopy (src, eh->ether_shost, ETHER_ADDR_LEN);
639 	eh->ether_type = htons(ETHERTYPE_IP);
640 	m0->m_pkthdr.len = m0->m_len = ETHER_MIN_LEN - ETHER_CRC_LEN;
641 
642 	/*
643 	 * Queue the packet, start transmission.
644 	 * Note: ifq_handoff() ultimately calls re_start() for us.
645 	 */
646 
647 	CSR_WRITE_2(sc, RE_ISR, 0xFFFF);
648 	error = ifq_handoff(ifp, m0, NULL);
649 	if (error) {
650 		m0 = NULL;
651 		goto done;
652 	}
653 	m0 = NULL;
654 
655 	/* Wait for it to propagate through the chip */
656 
657 	DELAY(100000);
658 	for (i = 0; i < RE_TIMEOUT; i++) {
659 		status = CSR_READ_2(sc, RE_ISR);
660 		if ((status & (RE_ISR_TIMEOUT_EXPIRED|RE_ISR_RX_OK)) ==
661 		    (RE_ISR_TIMEOUT_EXPIRED|RE_ISR_RX_OK))
662 			break;
663 		DELAY(10);
664 	}
665 
666 	if (i == RE_TIMEOUT) {
667 		if_printf(ifp, "diagnostic failed to receive packet "
668 			  "in loopback mode\n");
669 		error = EIO;
670 		goto done;
671 	}
672 
673 	/*
674 	 * The packet should have been dumped into the first
675 	 * entry in the RX DMA ring. Grab it from there.
676 	 */
677 
678 	bus_dmamap_sync(sc->re_ldata.re_rx_list_tag,
679 			sc->re_ldata.re_rx_list_map, BUS_DMASYNC_POSTREAD);
680 	bus_dmamap_sync(sc->re_ldata.re_mtag, sc->re_ldata.re_rx_dmamap[0],
681 			BUS_DMASYNC_POSTWRITE);
682 	bus_dmamap_unload(sc->re_ldata.re_mtag, sc->re_ldata.re_rx_dmamap[0]);
683 
684 	m0 = sc->re_ldata.re_rx_mbuf[0];
685 	sc->re_ldata.re_rx_mbuf[0] = NULL;
686 	eh = mtod(m0, struct ether_header *);
687 
688 	cur_rx = &sc->re_ldata.re_rx_list[0];
689 	total_len = RE_RXBYTES(cur_rx);
690 	rxstat = le32toh(cur_rx->re_cmdstat);
691 
692 	if (total_len != ETHER_MIN_LEN) {
693 		if_printf(ifp, "diagnostic failed, received short packet\n");
694 		error = EIO;
695 		goto done;
696 	}
697 
698 	/* Test that the received packet data matches what we sent. */
699 
700 	if (bcmp(eh->ether_dhost, dst, ETHER_ADDR_LEN) ||
701 	    bcmp(eh->ether_shost, &src, ETHER_ADDR_LEN) ||
702 	    be16toh(eh->ether_type) != ETHERTYPE_IP) {
703 		if_printf(ifp, "WARNING, DMA FAILURE!\n");
704 		if_printf(ifp, "expected TX data: %6D/%6D/0x%x\n",
705 		    dst, ":", src, ":", ETHERTYPE_IP);
706 		if_printf(ifp, "received RX data: %6D/%6D/0x%x\n",
707 		    eh->ether_dhost, ":",  eh->ether_shost, ":",
708 		    ntohs(eh->ether_type));
709 		if_printf(ifp, "You may have a defective 32-bit NIC plugged "
710 		    "into a 64-bit PCI slot.\n");
711 		if_printf(ifp, "Please re-install the NIC in a 32-bit slot "
712 		    "for proper operation.\n");
713 		if_printf(ifp, "Read the re(4) man page for more details.\n");
714 		error = EIO;
715 	}
716 
717 done:
718 	/* Turn interface off, release resources */
719 
720 	sc->re_testmode = 0;
721 	ifp->if_flags &= ~IFF_PROMISC;
722 	re_stop(sc);
723 	if (m0 != NULL)
724 		m_freem(m0);
725 
726 	return (error);
727 }
728 
729 /*
730  * Probe for a RealTek 8139C+/8169/8110 chip. Check the PCI vendor and device
731  * IDs against our list and return a device name if we find a match.
732  */
733 static int
734 re_probe(device_t dev)
735 {
736 	struct re_type *t;
737 	struct re_softc *sc;
738 	int rid;
739 	uint32_t hwrev;
740 	uint16_t vendor, product;
741 
742 	t = re_devs;
743 
744 	vendor = pci_get_vendor(dev);
745 	product = pci_get_device(dev);
746 
747 	for (t = re_devs; t->re_name != NULL; t++) {
748 		if (product == t->re_did && vendor == t->re_vid)
749 			break;
750 	}
751 
752 	/*
753 	 * Check if we found a RealTek device.
754 	 */
755 	if (t->re_name == NULL)
756 		return(ENXIO);
757 
758 	/*
759 	 * Temporarily map the I/O space so we can read the chip ID register.
760 	 */
761 	sc = malloc(sizeof(*sc), M_TEMP, M_WAITOK | M_ZERO);
762 	rid = RE_PCI_LOIO;
763 	sc->re_res = bus_alloc_resource_any(dev, SYS_RES_IOPORT, &rid,
764 					    RF_ACTIVE);
765 	if (sc->re_res == NULL) {
766 		device_printf(dev, "couldn't map ports/memory\n");
767 		free(sc, M_TEMP);
768 		return(ENXIO);
769 	}
770 
771 	sc->re_btag = rman_get_bustag(sc->re_res);
772 	sc->re_bhandle = rman_get_bushandle(sc->re_res);
773 
774 	hwrev = CSR_READ_4(sc, RE_TXCFG) & RE_TXCFG_HWREV;
775 	bus_release_resource(dev, SYS_RES_IOPORT, RE_PCI_LOIO, sc->re_res);
776 	free(sc, M_TEMP);
777 
778 	/*
779 	 * and continue matching for the specific chip...
780 	 */
781 	for (; t->re_name != NULL; t++) {
782 		if (product == t->re_did && vendor == t->re_vid &&
783 		    t->re_basetype == hwrev) {
784 			device_set_desc(dev, t->re_name);
785 			return(0);
786 		}
787 	}
788 	return(ENXIO);
789 }
790 
791 /*
792  * This routine takes the segment list provided as the result of
793  * a bus_dma_map_load() operation and assigns the addresses/lengths
794  * to RealTek DMA descriptors. This can be called either by the RX
795  * code or the TX code. In the RX case, we'll probably wind up mapping
796  * at most one segment. For the TX case, there could be any number of
797  * segments since TX packets may span multiple mbufs. In either case,
798  * if the number of segments is larger than the re_maxsegs limit
799  * specified by the caller, we abort the mapping operation. Sadly,
800  * whoever designed the buffer mapping API did not provide a way to
801  * return an error from here, so we have to fake it a bit.
802  */
803 
804 static void
805 re_dma_map_desc(void *arg, bus_dma_segment_t *segs, int nseg,
806 		bus_size_t mapsize, int error)
807 {
808 	struct re_dmaload_arg *ctx;
809 	struct re_desc *d = NULL;
810 	int i = 0, idx;
811 	uint32_t cmdstat;
812 
813 	if (error)
814 		return;
815 
816 	ctx = arg;
817 
818 	/* Signal error to caller if there's too many segments */
819 	if (nseg > ctx->re_maxsegs) {
820 		ctx->re_maxsegs = 0;
821 		return;
822 	}
823 
824 	/*
825 	 * Map the segment array into descriptors. Note that we set the
826 	 * start-of-frame and end-of-frame markers for either TX or RX, but
827 	 * they really only have meaning in the TX case. (In the RX case,
828 	 * it's the chip that tells us where packets begin and end.)
829 	 * We also keep track of the end of the ring and set the
830 	 * end-of-ring bits as needed, and we set the ownership bits
831 	 * in all except the very first descriptor. (The caller will
832 	 * set this descriptor later when it start transmission or
833 	 * reception.)
834 	 */
835 	idx = ctx->re_idx;
836 	for (;;) {
837 		d = &ctx->re_ring[idx];
838 		if (le32toh(d->re_cmdstat) & RE_RDESC_STAT_OWN) {
839 			ctx->re_maxsegs = 0;
840 			return;
841 		}
842 		cmdstat = segs[i].ds_len;
843 		d->re_bufaddr_lo = htole32(RE_ADDR_LO(segs[i].ds_addr));
844 		d->re_bufaddr_hi = htole32(RE_ADDR_HI(segs[i].ds_addr));
845 		if (i == 0)
846 			cmdstat |= RE_TDESC_CMD_SOF;
847 		else
848 			cmdstat |= RE_TDESC_CMD_OWN;
849 		if (idx == (RE_RX_DESC_CNT - 1))
850 			cmdstat |= RE_TDESC_CMD_EOR;
851 		d->re_cmdstat = htole32(cmdstat | ctx->re_flags);
852 		i++;
853 		if (i == nseg)
854 			break;
855 		RE_DESC_INC(idx);
856 	}
857 
858 	d->re_cmdstat |= htole32(RE_TDESC_CMD_EOF);
859 	ctx->re_maxsegs = nseg;
860 	ctx->re_idx = idx;
861 }
862 
863 /*
864  * Map a single buffer address.
865  */
866 
867 static void
868 re_dma_map_addr(void *arg, bus_dma_segment_t *segs, int nseg, int error)
869 {
870 	uint32_t *addr;
871 
872 	if (error)
873 		return;
874 
875 	KASSERT(nseg == 1, ("too many DMA segments, %d should be 1", nseg));
876 	addr = arg;
877 	*addr = segs->ds_addr;
878 }
879 
880 static int
881 re_allocmem(device_t dev, struct re_softc *sc)
882 {
883 	int error, i, nseg;
884 
885 	/*
886 	 * Allocate map for RX mbufs.
887 	 */
888 	nseg = 32;
889 	error = bus_dma_tag_create(sc->re_parent_tag, ETHER_ALIGN, 0,
890 	    BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
891 	    NULL, MCLBYTES * nseg, nseg, MCLBYTES, BUS_DMA_ALLOCNOW,
892 	    &sc->re_ldata.re_mtag);
893 	if (error) {
894 		device_printf(dev, "could not allocate dma tag\n");
895 		return(error);
896 	}
897 
898 	/*
899 	 * Allocate map for TX descriptor list.
900 	 */
901 	error = bus_dma_tag_create(sc->re_parent_tag, RE_RING_ALIGN,
902 	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
903             NULL, RE_TX_LIST_SZ, 1, RE_TX_LIST_SZ, BUS_DMA_ALLOCNOW,
904 	    &sc->re_ldata.re_tx_list_tag);
905 	if (error) {
906 		device_printf(dev, "could not allocate dma tag\n");
907 		return(error);
908 	}
909 
910 	/* Allocate DMA'able memory for the TX ring */
911 
912         error = bus_dmamem_alloc(sc->re_ldata.re_tx_list_tag,
913 	    (void **)&sc->re_ldata.re_tx_list, BUS_DMA_WAITOK | BUS_DMA_ZERO,
914             &sc->re_ldata.re_tx_list_map);
915         if (error) {
916 		device_printf(dev, "could not allocate TX ring\n");
917                 return(error);
918 	}
919 
920 	/* Load the map for the TX ring. */
921 
922 	error = bus_dmamap_load(sc->re_ldata.re_tx_list_tag,
923 	     sc->re_ldata.re_tx_list_map, sc->re_ldata.re_tx_list,
924 	     RE_TX_LIST_SZ, re_dma_map_addr,
925 	     &sc->re_ldata.re_tx_list_addr, BUS_DMA_NOWAIT);
926 	if (error) {
927 		device_printf(dev, "could not get addres of TX ring\n");
928 		return(error);
929 	}
930 
931 	/* Create DMA maps for TX buffers */
932 
933 	for (i = 0; i < RE_TX_DESC_CNT; i++) {
934 		error = bus_dmamap_create(sc->re_ldata.re_mtag, 0,
935 			    &sc->re_ldata.re_tx_dmamap[i]);
936 		if (error) {
937 			device_printf(dev, "can't create DMA map for TX\n");
938 			return(error);
939 		}
940 	}
941 
942 	/*
943 	 * Allocate map for RX descriptor list.
944 	 */
945 	error = bus_dma_tag_create(sc->re_parent_tag, RE_RING_ALIGN,
946 	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
947             NULL, RE_TX_LIST_SZ, 1, RE_TX_LIST_SZ, BUS_DMA_ALLOCNOW,
948 	    &sc->re_ldata.re_rx_list_tag);
949 	if (error) {
950 		device_printf(dev, "could not allocate dma tag\n");
951 		return(error);
952 	}
953 
954 	/* Allocate DMA'able memory for the RX ring */
955 
956         error = bus_dmamem_alloc(sc->re_ldata.re_rx_list_tag,
957 	    (void **)&sc->re_ldata.re_rx_list, BUS_DMA_WAITOK | BUS_DMA_ZERO,
958             &sc->re_ldata.re_rx_list_map);
959         if (error) {
960 		device_printf(dev, "could not allocate RX ring\n");
961                 return(error);
962 	}
963 
964 	/* Load the map for the RX ring. */
965 
966 	error = bus_dmamap_load(sc->re_ldata.re_rx_list_tag,
967 	     sc->re_ldata.re_rx_list_map, sc->re_ldata.re_rx_list,
968 	     RE_TX_LIST_SZ, re_dma_map_addr,
969 	     &sc->re_ldata.re_rx_list_addr, BUS_DMA_NOWAIT);
970 	if (error) {
971 		device_printf(dev, "could not get address of RX ring\n");
972 		return(error);
973 	}
974 
975 	/* Create DMA maps for RX buffers */
976 
977 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
978 		error = bus_dmamap_create(sc->re_ldata.re_mtag, 0,
979 			    &sc->re_ldata.re_rx_dmamap[i]);
980 		if (error) {
981 			device_printf(dev, "can't create DMA map for RX\n");
982 			return(ENOMEM);
983 		}
984 	}
985 
986 	return(0);
987 }
988 
989 /*
990  * Attach the interface. Allocate softc structures, do ifmedia
991  * setup and ethernet/BPF attach.
992  */
993 static int
994 re_attach(device_t dev)
995 {
996 	struct re_softc	*sc = device_get_softc(dev);
997 	struct ifnet *ifp;
998 	struct re_hwrev *hw_rev;
999 	uint8_t eaddr[ETHER_ADDR_LEN];
1000 	int hwrev;
1001 	u_int16_t re_did = 0;
1002 	int error = 0, rid, i;
1003 
1004 	callout_init(&sc->re_timer);
1005 
1006 #ifndef BURN_BRIDGES
1007 	/*
1008 	 * Handle power management nonsense.
1009 	 */
1010 
1011 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D0) {
1012 		uint32_t membase, irq;
1013 
1014 		/* Save important PCI config data. */
1015 		membase = pci_read_config(dev, RE_PCI_LOMEM, 4);
1016 		irq = pci_read_config(dev, PCIR_INTLINE, 4);
1017 
1018 		/* Reset the power state. */
1019 		device_printf(dev, "chip is is in D%d power mode "
1020 		    "-- setting to D0\n", pci_get_powerstate(dev));
1021 
1022 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
1023 
1024 		/* Restore PCI config data. */
1025 		pci_write_config(dev, RE_PCI_LOMEM, membase, 4);
1026 		pci_write_config(dev, PCIR_INTLINE, irq, 4);
1027 	}
1028 #endif
1029 	/*
1030 	 * Map control/status registers.
1031 	 */
1032 	pci_enable_busmaster(dev);
1033 
1034 	rid = RE_PCI_LOIO;
1035 	sc->re_res = bus_alloc_resource_any(dev, SYS_RES_IOPORT, &rid,
1036 					    RF_ACTIVE);
1037 
1038 	if (sc->re_res == NULL) {
1039 		device_printf(dev, "couldn't map ports/memory\n");
1040 		error = ENXIO;
1041 		goto fail;
1042 	}
1043 
1044 	sc->re_btag = rman_get_bustag(sc->re_res);
1045 	sc->re_bhandle = rman_get_bushandle(sc->re_res);
1046 
1047 	/* Allocate interrupt */
1048 	rid = 0;
1049 	sc->re_irq = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
1050 					    RF_SHAREABLE | RF_ACTIVE);
1051 
1052 	if (sc->re_irq == NULL) {
1053 		device_printf(dev, "couldn't map interrupt\n");
1054 		error = ENXIO;
1055 		goto fail;
1056 	}
1057 
1058 	/* Reset the adapter. */
1059 	re_reset(sc);
1060 
1061 	hwrev = CSR_READ_4(sc, RE_TXCFG) & RE_TXCFG_HWREV;
1062 	for (hw_rev = re_hwrevs; hw_rev->re_desc != NULL; hw_rev++) {
1063 		if (hw_rev->re_rev == hwrev) {
1064 			sc->re_type = hw_rev->re_type;
1065 			break;
1066 		}
1067 	}
1068 
1069 	if (sc->re_type == RE_8169) {
1070 		/* Set RX length mask */
1071 		sc->re_rxlenmask = RE_RDESC_STAT_GFRAGLEN;
1072 
1073 		/* Force station address autoload from the EEPROM */
1074 		CSR_WRITE_1(sc, RE_EECMD, RE_EEMODE_AUTOLOAD);
1075 		for (i = 0; i < RE_TIMEOUT; i++) {
1076 			if ((CSR_READ_1(sc, RE_EECMD) & RE_EEMODE_AUTOLOAD) == 0)
1077 				break;
1078 			DELAY(100);
1079 		}
1080 		if (i == RE_TIMEOUT)
1081 			device_printf(dev, "eeprom autoload timed out\n");
1082 
1083 		for (i = 0; i < ETHER_ADDR_LEN; i++)
1084 			eaddr[i] = CSR_READ_1(sc, RE_IDR0 + i);
1085 	} else {
1086 		uint16_t as[3];
1087 
1088 		/* Set RX length mask */
1089 		sc->re_rxlenmask = RE_RDESC_STAT_FRAGLEN;
1090 
1091 		sc->re_eecmd_read = RE_EECMD_READ_6BIT;
1092 		re_read_eeprom(sc, (caddr_t)&re_did, 0, 1, 0);
1093 		if (re_did != 0x8129)
1094 			sc->re_eecmd_read = RE_EECMD_READ_8BIT;
1095 
1096 		/*
1097 		 * Get station address from the EEPROM.
1098 		 */
1099 		re_read_eeprom(sc, (caddr_t)as, RE_EE_EADDR, 3, 0);
1100 		for (i = 0; i < 3; i++) {
1101 			eaddr[(i * 2) + 0] = as[i] & 0xff;
1102 			eaddr[(i * 2) + 1] = as[i] >> 8;
1103 		}
1104 	}
1105 
1106 	/*
1107 	 * Allocate the parent bus DMA tag appropriate for PCI.
1108 	 */
1109 #define RE_NSEG_NEW 32
1110 	error = bus_dma_tag_create(NULL,	/* parent */
1111 			1, 0,			/* alignment, boundary */
1112 			BUS_SPACE_MAXADDR_32BIT,/* lowaddr */
1113 			BUS_SPACE_MAXADDR,	/* highaddr */
1114 			NULL, NULL,		/* filter, filterarg */
1115 			MAXBSIZE, RE_NSEG_NEW,	/* maxsize, nsegments */
1116 			BUS_SPACE_MAXSIZE_32BIT,/* maxsegsize */
1117 			BUS_DMA_ALLOCNOW,	/* flags */
1118 			&sc->re_parent_tag);
1119 	if (error)
1120 		goto fail;
1121 
1122 	error = re_allocmem(dev, sc);
1123 
1124 	if (error)
1125 		goto fail;
1126 
1127 	/* Do MII setup */
1128 	if (mii_phy_probe(dev, &sc->re_miibus,
1129 	    re_ifmedia_upd, re_ifmedia_sts)) {
1130 		device_printf(dev, "MII without any phy!\n");
1131 		error = ENXIO;
1132 		goto fail;
1133 	}
1134 
1135 	ifp = &sc->arpcom.ac_if;
1136 	ifp->if_softc = sc;
1137 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1138 	ifp->if_mtu = ETHERMTU;
1139 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1140 	ifp->if_ioctl = re_ioctl;
1141 	ifp->if_capabilities = IFCAP_VLAN_MTU;
1142 	ifp->if_start = re_start;
1143 	ifp->if_capabilities |= IFCAP_HWCSUM|IFCAP_VLAN_HWTAGGING;
1144 #ifdef DEVICE_POLLING
1145 	ifp->if_poll = re_poll;
1146 #endif
1147 	ifp->if_watchdog = re_watchdog;
1148 	ifp->if_init = re_init;
1149 	if (sc->re_type == RE_8169)
1150 		ifp->if_baudrate = 1000000000;
1151 	else
1152 		ifp->if_baudrate = 100000000;
1153 	ifq_set_maxlen(&ifp->if_snd, RE_IFQ_MAXLEN);
1154 	ifq_set_ready(&ifp->if_snd);
1155 #ifdef RE_DISABLE_HWCSUM
1156 	ifp->if_capenable = ifp->if_capabilities & ~IFCAP_HWCSUM;
1157 	ifp->if_hwassist = 0;
1158 #else
1159 	ifp->if_capenable = ifp->if_capabilities;
1160 	ifp->if_hwassist = RE_CSUM_FEATURES;
1161 #endif
1162 
1163 	/*
1164 	 * Call MI attach routine.
1165 	 */
1166 	ether_ifattach(ifp, eaddr, NULL);
1167 
1168 	lwkt_serialize_enter(ifp->if_serializer);
1169 	/* Perform hardware diagnostic. */
1170 	error = re_diag(sc);
1171 	lwkt_serialize_exit(ifp->if_serializer);
1172 
1173 	if (error) {
1174 		device_printf(dev, "hardware diagnostic failure\n");
1175 		ether_ifdetach(ifp);
1176 		goto fail;
1177 	}
1178 
1179 	/* Hook interrupt last to avoid having to lock softc */
1180 	error = bus_setup_intr(dev, sc->re_irq, INTR_NETSAFE, re_intr, sc,
1181 			       &sc->re_intrhand, ifp->if_serializer);
1182 
1183 	if (error) {
1184 		device_printf(dev, "couldn't set up irq\n");
1185 		ether_ifdetach(ifp);
1186 		goto fail;
1187 	}
1188 
1189 fail:
1190 	if (error)
1191 		re_detach(dev);
1192 
1193 	return (error);
1194 }
1195 
1196 /*
1197  * Shutdown hardware and free up resources. This can be called any
1198  * time after the mutex has been initialized. It is called in both
1199  * the error case in attach and the normal detach case so it needs
1200  * to be careful about only freeing resources that have actually been
1201  * allocated.
1202  */
1203 static int
1204 re_detach(device_t dev)
1205 {
1206 	struct re_softc *sc = device_get_softc(dev);
1207 	struct ifnet *ifp = &sc->arpcom.ac_if;
1208 	int i;
1209 
1210 	/* These should only be active if attach succeeded */
1211 	if (device_is_attached(dev)) {
1212 		lwkt_serialize_enter(ifp->if_serializer);
1213 		re_stop(sc);
1214 		bus_teardown_intr(dev, sc->re_irq, sc->re_intrhand);
1215 		lwkt_serialize_exit(ifp->if_serializer);
1216 
1217 		ether_ifdetach(ifp);
1218 	}
1219 	if (sc->re_miibus)
1220 		device_delete_child(dev, sc->re_miibus);
1221 	bus_generic_detach(dev);
1222 
1223 	if (sc->re_irq)
1224 		bus_release_resource(dev, SYS_RES_IRQ, 0, sc->re_irq);
1225 	if (sc->re_res) {
1226 		bus_release_resource(dev, SYS_RES_IOPORT, RE_PCI_LOIO,
1227 				     sc->re_res);
1228 	}
1229 
1230 	/* Unload and free the RX DMA ring memory and map */
1231 
1232 	if (sc->re_ldata.re_rx_list_tag) {
1233 		bus_dmamap_unload(sc->re_ldata.re_rx_list_tag,
1234 		    sc->re_ldata.re_rx_list_map);
1235 		bus_dmamem_free(sc->re_ldata.re_rx_list_tag,
1236 		    sc->re_ldata.re_rx_list,
1237 		    sc->re_ldata.re_rx_list_map);
1238 		bus_dma_tag_destroy(sc->re_ldata.re_rx_list_tag);
1239 	}
1240 
1241 	/* Unload and free the TX DMA ring memory and map */
1242 
1243 	if (sc->re_ldata.re_tx_list_tag) {
1244 		bus_dmamap_unload(sc->re_ldata.re_tx_list_tag,
1245 		    sc->re_ldata.re_tx_list_map);
1246 		bus_dmamem_free(sc->re_ldata.re_tx_list_tag,
1247 		    sc->re_ldata.re_tx_list,
1248 		    sc->re_ldata.re_tx_list_map);
1249 		bus_dma_tag_destroy(sc->re_ldata.re_tx_list_tag);
1250 	}
1251 
1252 	/* Destroy all the RX and TX buffer maps */
1253 
1254 	if (sc->re_ldata.re_mtag) {
1255 		for (i = 0; i < RE_TX_DESC_CNT; i++)
1256 			bus_dmamap_destroy(sc->re_ldata.re_mtag,
1257 			    sc->re_ldata.re_tx_dmamap[i]);
1258 		for (i = 0; i < RE_RX_DESC_CNT; i++)
1259 			bus_dmamap_destroy(sc->re_ldata.re_mtag,
1260 			    sc->re_ldata.re_rx_dmamap[i]);
1261 		bus_dma_tag_destroy(sc->re_ldata.re_mtag);
1262 	}
1263 
1264 	/* Unload and free the stats buffer and map */
1265 
1266 	if (sc->re_ldata.re_stag) {
1267 		bus_dmamap_unload(sc->re_ldata.re_stag,
1268 		    sc->re_ldata.re_rx_list_map);
1269 		bus_dmamem_free(sc->re_ldata.re_stag,
1270 		    sc->re_ldata.re_stats,
1271 		    sc->re_ldata.re_smap);
1272 		bus_dma_tag_destroy(sc->re_ldata.re_stag);
1273 	}
1274 
1275 	if (sc->re_parent_tag)
1276 		bus_dma_tag_destroy(sc->re_parent_tag);
1277 
1278 	return(0);
1279 }
1280 
1281 static int
1282 re_newbuf(struct re_softc *sc, int idx, struct mbuf *m)
1283 {
1284 	struct re_dmaload_arg arg;
1285 	struct mbuf *n = NULL;
1286 	int error;
1287 
1288 	if (m == NULL) {
1289 		n = m_getcl(MB_DONTWAIT, MT_DATA, M_PKTHDR);
1290 		if (n == NULL)
1291 			return(ENOBUFS);
1292 		m = n;
1293 	} else
1294 		m->m_data = m->m_ext.ext_buf;
1295 
1296 	/*
1297 	 * Initialize mbuf length fields and fixup
1298 	 * alignment so that the frame payload is
1299 	 * longword aligned.
1300 	 */
1301 	m->m_len = m->m_pkthdr.len = MCLBYTES;
1302 	m_adj(m, ETHER_ALIGN);
1303 
1304 	arg.sc = sc;
1305 	arg.re_idx = idx;
1306 	arg.re_maxsegs = 1;
1307 	arg.re_flags = 0;
1308 	arg.re_ring = sc->re_ldata.re_rx_list;
1309 
1310         error = bus_dmamap_load_mbuf(sc->re_ldata.re_mtag,
1311 	    sc->re_ldata.re_rx_dmamap[idx], m, re_dma_map_desc,
1312 	    &arg, BUS_DMA_NOWAIT);
1313 	if (error || arg.re_maxsegs != 1) {
1314 		if (n != NULL)
1315 			m_freem(n);
1316 		return (ENOMEM);
1317 	}
1318 
1319 	sc->re_ldata.re_rx_list[idx].re_cmdstat |= htole32(RE_RDESC_CMD_OWN);
1320 	sc->re_ldata.re_rx_mbuf[idx] = m;
1321 
1322         bus_dmamap_sync(sc->re_ldata.re_mtag, sc->re_ldata.re_rx_dmamap[idx],
1323 		        BUS_DMASYNC_PREREAD);
1324 
1325 	return(0);
1326 }
1327 
1328 static int
1329 re_tx_list_init(struct re_softc *sc)
1330 {
1331 	bzero(sc->re_ldata.re_tx_list, RE_TX_LIST_SZ);
1332 	bzero(&sc->re_ldata.re_tx_mbuf, RE_TX_DESC_CNT * sizeof(struct mbuf *));
1333 
1334 	bus_dmamap_sync(sc->re_ldata.re_tx_list_tag,
1335 			sc->re_ldata.re_tx_list_map, BUS_DMASYNC_PREWRITE);
1336 	sc->re_ldata.re_tx_prodidx = 0;
1337 	sc->re_ldata.re_tx_considx = 0;
1338 	sc->re_ldata.re_tx_free = RE_TX_DESC_CNT;
1339 
1340 	return(0);
1341 }
1342 
1343 static int
1344 re_rx_list_init(struct re_softc *sc)
1345 {
1346 	int i, error;
1347 
1348 	bzero(sc->re_ldata.re_rx_list, RE_RX_LIST_SZ);
1349 	bzero(&sc->re_ldata.re_rx_mbuf, RE_RX_DESC_CNT * sizeof(struct mbuf *));
1350 
1351 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
1352 		error = re_newbuf(sc, i, NULL);
1353 		if (error)
1354 			return(error);
1355 	}
1356 
1357 	/* Flush the RX descriptors */
1358 
1359 	bus_dmamap_sync(sc->re_ldata.re_rx_list_tag,
1360 	    sc->re_ldata.re_rx_list_map,
1361 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
1362 
1363 	sc->re_ldata.re_rx_prodidx = 0;
1364 	sc->re_head = sc->re_tail = NULL;
1365 
1366 	return(0);
1367 }
1368 
1369 /*
1370  * RX handler for C+ and 8169. For the gigE chips, we support
1371  * the reception of jumbo frames that have been fragmented
1372  * across multiple 2K mbuf cluster buffers.
1373  */
1374 static void
1375 re_rxeof(struct re_softc *sc)
1376 {
1377 	struct ifnet *ifp = &sc->arpcom.ac_if;
1378 	struct mbuf *m;
1379 	struct re_desc 	*cur_rx;
1380 	uint32_t rxstat, rxvlan;
1381 	int i, total_len;
1382 
1383 	/* Invalidate the descriptor memory */
1384 
1385 	bus_dmamap_sync(sc->re_ldata.re_rx_list_tag,
1386 			sc->re_ldata.re_rx_list_map, BUS_DMASYNC_POSTREAD);
1387 
1388 	for (i = sc->re_ldata.re_rx_prodidx;
1389 	     RE_OWN(&sc->re_ldata.re_rx_list[i]) == 0 ; RE_DESC_INC(i)) {
1390 		cur_rx = &sc->re_ldata.re_rx_list[i];
1391 		m = sc->re_ldata.re_rx_mbuf[i];
1392 		total_len = RE_RXBYTES(cur_rx);
1393 		rxstat = le32toh(cur_rx->re_cmdstat);
1394 		rxvlan = le32toh(cur_rx->re_vlanctl);
1395 
1396 		/* Invalidate the RX mbuf and unload its map */
1397 
1398 		bus_dmamap_sync(sc->re_ldata.re_mtag,
1399 				sc->re_ldata.re_rx_dmamap[i],
1400 				BUS_DMASYNC_POSTWRITE);
1401 		bus_dmamap_unload(sc->re_ldata.re_mtag,
1402 				  sc->re_ldata.re_rx_dmamap[i]);
1403 
1404 		if ((rxstat & RE_RDESC_STAT_EOF) == 0) {
1405 			m->m_len = MCLBYTES - ETHER_ALIGN;
1406 			if (sc->re_head == NULL) {
1407 				sc->re_head = sc->re_tail = m;
1408 			} else {
1409 				sc->re_tail->m_next = m;
1410 				sc->re_tail = m;
1411 			}
1412 			re_newbuf(sc, i, NULL);
1413 			continue;
1414 		}
1415 
1416 		/*
1417 		 * NOTE: for the 8139C+, the frame length field
1418 		 * is always 12 bits in size, but for the gigE chips,
1419 		 * it is 13 bits (since the max RX frame length is 16K).
1420 		 * Unfortunately, all 32 bits in the status word
1421 		 * were already used, so to make room for the extra
1422 		 * length bit, RealTek took out the 'frame alignment
1423 		 * error' bit and shifted the other status bits
1424 		 * over one slot. The OWN, EOR, FS and LS bits are
1425 		 * still in the same places. We have already extracted
1426 		 * the frame length and checked the OWN bit, so rather
1427 		 * than using an alternate bit mapping, we shift the
1428 		 * status bits one space to the right so we can evaluate
1429 		 * them using the 8169 status as though it was in the
1430 		 * same format as that of the 8139C+.
1431 		 */
1432 		if (sc->re_type == RE_8169)
1433 			rxstat >>= 1;
1434 
1435 		if (rxstat & RE_RDESC_STAT_RXERRSUM) {
1436 			ifp->if_ierrors++;
1437 			/*
1438 			 * If this is part of a multi-fragment packet,
1439 			 * discard all the pieces.
1440 			 */
1441 			if (sc->re_head != NULL) {
1442 				m_freem(sc->re_head);
1443 				sc->re_head = sc->re_tail = NULL;
1444 			}
1445 			re_newbuf(sc, i, m);
1446 			continue;
1447 		}
1448 
1449 		/*
1450 		 * If allocating a replacement mbuf fails,
1451 		 * reload the current one.
1452 		 */
1453 
1454 		if (re_newbuf(sc, i, NULL)) {
1455 			ifp->if_ierrors++;
1456 			if (sc->re_head != NULL) {
1457 				m_freem(sc->re_head);
1458 				sc->re_head = sc->re_tail = NULL;
1459 			}
1460 			re_newbuf(sc, i, m);
1461 			continue;
1462 		}
1463 
1464 		if (sc->re_head != NULL) {
1465 			m->m_len = total_len % (MCLBYTES - ETHER_ALIGN);
1466 			/*
1467 			 * Special case: if there's 4 bytes or less
1468 			 * in this buffer, the mbuf can be discarded:
1469 			 * the last 4 bytes is the CRC, which we don't
1470 			 * care about anyway.
1471 			 */
1472 			if (m->m_len <= ETHER_CRC_LEN) {
1473 				sc->re_tail->m_len -=
1474 				    (ETHER_CRC_LEN - m->m_len);
1475 				m_freem(m);
1476 			} else {
1477 				m->m_len -= ETHER_CRC_LEN;
1478 				sc->re_tail->m_next = m;
1479 			}
1480 			m = sc->re_head;
1481 			sc->re_head = sc->re_tail = NULL;
1482 			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
1483 		} else
1484 			m->m_pkthdr.len = m->m_len =
1485 			    (total_len - ETHER_CRC_LEN);
1486 
1487 		ifp->if_ipackets++;
1488 		m->m_pkthdr.rcvif = ifp;
1489 
1490 		/* Do RX checksumming if enabled */
1491 
1492 		if (ifp->if_capenable & IFCAP_RXCSUM) {
1493 
1494 			/* Check IP header checksum */
1495 			if (rxstat & RE_RDESC_STAT_PROTOID)
1496 				m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED;
1497 			if ((rxstat & RE_RDESC_STAT_IPSUMBAD) == 0)
1498 				m->m_pkthdr.csum_flags |= CSUM_IP_VALID;
1499 
1500 			/* Check TCP/UDP checksum */
1501 			if ((RE_TCPPKT(rxstat) &&
1502 			    (rxstat & RE_RDESC_STAT_TCPSUMBAD) == 0) ||
1503 			    (RE_UDPPKT(rxstat) &&
1504 			    (rxstat & RE_RDESC_STAT_UDPSUMBAD)) == 0) {
1505 				m->m_pkthdr.csum_flags |=
1506 				    CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
1507 				m->m_pkthdr.csum_data = 0xffff;
1508 			}
1509 		}
1510 
1511 		if (rxvlan & RE_RDESC_VLANCTL_TAG) {
1512 			VLAN_INPUT_TAG(m,
1513 			   be16toh((rxvlan & RE_RDESC_VLANCTL_DATA)));
1514 		} else {
1515 			ifp->if_input(ifp, m);
1516 		}
1517 	}
1518 
1519 	/* Flush the RX DMA ring */
1520 
1521 	bus_dmamap_sync(sc->re_ldata.re_rx_list_tag,
1522 			sc->re_ldata.re_rx_list_map,
1523 			BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
1524 
1525 	sc->re_ldata.re_rx_prodidx = i;
1526 }
1527 
1528 static void
1529 re_txeof(struct re_softc *sc)
1530 {
1531 	struct ifnet *ifp = &sc->arpcom.ac_if;
1532 	uint32_t txstat;
1533 	int idx;
1534 
1535 	/* Invalidate the TX descriptor list */
1536 
1537 	bus_dmamap_sync(sc->re_ldata.re_tx_list_tag,
1538 	    sc->re_ldata.re_tx_list_map,
1539 	    BUS_DMASYNC_POSTREAD);
1540 
1541 	for (idx = sc->re_ldata.re_tx_considx;
1542 	     idx != sc->re_ldata.re_tx_prodidx; RE_DESC_INC(idx)) {
1543 		txstat = le32toh(sc->re_ldata.re_tx_list[idx].re_cmdstat);
1544 		if (txstat & RE_TDESC_CMD_OWN)
1545 			break;
1546 
1547 		/*
1548 		 * We only stash mbufs in the last descriptor
1549 		 * in a fragment chain, which also happens to
1550 		 * be the only place where the TX status bits
1551 		 * are valid.
1552 		 */
1553 		if (txstat & RE_TDESC_CMD_EOF) {
1554 			m_freem(sc->re_ldata.re_tx_mbuf[idx]);
1555 			sc->re_ldata.re_tx_mbuf[idx] = NULL;
1556 			bus_dmamap_unload(sc->re_ldata.re_mtag,
1557 			    sc->re_ldata.re_tx_dmamap[idx]);
1558 			if (txstat & (RE_TDESC_STAT_EXCESSCOL|
1559 			    RE_TDESC_STAT_COLCNT))
1560 				ifp->if_collisions++;
1561 			if (txstat & RE_TDESC_STAT_TXERRSUM)
1562 				ifp->if_oerrors++;
1563 			else
1564 				ifp->if_opackets++;
1565 		}
1566 		sc->re_ldata.re_tx_free++;
1567 	}
1568 
1569 	/* No changes made to the TX ring, so no flush needed */
1570 	if (idx != sc->re_ldata.re_tx_considx) {
1571 		sc->re_ldata.re_tx_considx = idx;
1572 		ifp->if_flags &= ~IFF_OACTIVE;
1573 		ifp->if_timer = 0;
1574 	}
1575 
1576 	/*
1577 	 * If not all descriptors have been released reaped yet,
1578 	 * reload the timer so that we will eventually get another
1579 	 * interrupt that will cause us to re-enter this routine.
1580 	 * This is done in case the transmitter has gone idle.
1581 	 */
1582 	if (sc->re_ldata.re_tx_free != RE_TX_DESC_CNT)
1583                 CSR_WRITE_4(sc, RE_TIMERCNT, 1);
1584 }
1585 
1586 static void
1587 re_tick(void *xsc)
1588 {
1589 	struct re_softc *sc = xsc;
1590 
1591 	lwkt_serialize_enter(sc->arpcom.ac_if.if_serializer);
1592 	re_tick_serialized(xsc);
1593 	lwkt_serialize_exit(sc->arpcom.ac_if.if_serializer);
1594 }
1595 
1596 static void
1597 re_tick_serialized(void *xsc)
1598 {
1599 	struct re_softc *sc = xsc;
1600 	struct mii_data *mii;
1601 
1602 	mii = device_get_softc(sc->re_miibus);
1603 	mii_tick(mii);
1604 
1605 	callout_reset(&sc->re_timer, hz, re_tick, sc);
1606 }
1607 
1608 #ifdef DEVICE_POLLING
1609 
1610 static void
1611 re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
1612 {
1613 	struct re_softc *sc = ifp->if_softc;
1614 
1615 	switch(cmd) {
1616 	case POLL_REGISTER:
1617 		/* disable interrupts */
1618 		CSR_WRITE_2(sc, RE_IMR, 0x0000);
1619 		break;
1620 	case POLL_DEREGISTER:
1621 		/* enable interrupts */
1622 		CSR_WRITE_2(sc, RE_IMR, RE_INTRS_CPLUS);
1623 		break;
1624 	default:
1625 		sc->rxcycles = count;
1626 		re_rxeof(sc);
1627 		re_txeof(sc);
1628 
1629 		if (!ifq_is_empty(&ifp->if_snd))
1630 			(*ifp->if_start)(ifp);
1631 
1632 		if (cmd == POLL_AND_CHECK_STATUS) { /* also check status register */
1633 			uint16_t       status;
1634 
1635 			status = CSR_READ_2(sc, RE_ISR);
1636 			if (status == 0xffff)
1637 				return;
1638 			if (status)
1639 				CSR_WRITE_2(sc, RE_ISR, status);
1640 
1641 			/*
1642 			 * XXX check behaviour on receiver stalls.
1643 			 */
1644 
1645 			if (status & RE_ISR_SYSTEM_ERR) {
1646 				re_reset(sc);
1647 				re_init(sc);
1648 			}
1649 		}
1650 		break;
1651 	}
1652 }
1653 #endif /* DEVICE_POLLING */
1654 
1655 static void
1656 re_intr(void *arg)
1657 {
1658 	struct re_softc	*sc = arg;
1659 	struct ifnet *ifp = &sc->arpcom.ac_if;
1660 	uint16_t status;
1661 
1662 	if (sc->suspended || (ifp->if_flags & IFF_UP) == 0)
1663 		return;
1664 
1665 	for (;;) {
1666 		status = CSR_READ_2(sc, RE_ISR);
1667 		/* If the card has gone away the read returns 0xffff. */
1668 		if (status == 0xffff)
1669 			break;
1670 		if (status)
1671 			CSR_WRITE_2(sc, RE_ISR, status);
1672 
1673 		if ((status & RE_INTRS_CPLUS) == 0)
1674 			break;
1675 
1676 		if (status & RE_ISR_RX_OK)
1677 			re_rxeof(sc);
1678 
1679 		if (status & RE_ISR_RX_ERR)
1680 			re_rxeof(sc);
1681 
1682 		if ((status & RE_ISR_TIMEOUT_EXPIRED) ||
1683 		    (status & RE_ISR_TX_ERR) ||
1684 		    (status & RE_ISR_TX_DESC_UNAVAIL))
1685 			re_txeof(sc);
1686 
1687 		if (status & RE_ISR_SYSTEM_ERR) {
1688 			re_reset(sc);
1689 			re_init(sc);
1690 		}
1691 
1692 		if (status & RE_ISR_LINKCHG)
1693 			re_tick_serialized(sc);
1694 	}
1695 
1696 	if (!ifq_is_empty(&ifp->if_snd))
1697 		(*ifp->if_start)(ifp);
1698 }
1699 
1700 static int
1701 re_encap(struct re_softc *sc, struct mbuf **m_head, int *idx, int *called_defrag)
1702 {
1703 	struct ifnet *ifp = &sc->arpcom.ac_if;
1704 	struct mbuf *m, *m_new = NULL;
1705 	struct re_dmaload_arg	arg;
1706 	bus_dmamap_t		map;
1707 	int			error;
1708 
1709 	*called_defrag = 0;
1710 	if (sc->re_ldata.re_tx_free <= 4)
1711 		return(EFBIG);
1712 
1713 	m = *m_head;
1714 
1715 	/*
1716 	 * Set up checksum offload. Note: checksum offload bits must
1717 	 * appear in all descriptors of a multi-descriptor transmit
1718 	 * attempt. (This is according to testing done with an 8169
1719 	 * chip. I'm not sure if this is a requirement or a bug.)
1720 	 */
1721 
1722 	arg.re_flags = 0;
1723 
1724 	if (m->m_pkthdr.csum_flags & CSUM_IP)
1725 		arg.re_flags |= RE_TDESC_CMD_IPCSUM;
1726 	if (m->m_pkthdr.csum_flags & CSUM_TCP)
1727 		arg.re_flags |= RE_TDESC_CMD_TCPCSUM;
1728 	if (m->m_pkthdr.csum_flags & CSUM_UDP)
1729 		arg.re_flags |= RE_TDESC_CMD_UDPCSUM;
1730 
1731 	arg.sc = sc;
1732 	arg.re_idx = *idx;
1733 	arg.re_maxsegs = sc->re_ldata.re_tx_free;
1734 	if (arg.re_maxsegs > 4)
1735 		arg.re_maxsegs -= 4;
1736 	arg.re_ring = sc->re_ldata.re_tx_list;
1737 
1738 	map = sc->re_ldata.re_tx_dmamap[*idx];
1739 	error = bus_dmamap_load_mbuf(sc->re_ldata.re_mtag, map,
1740 	    m, re_dma_map_desc, &arg, BUS_DMA_NOWAIT);
1741 
1742 	if (error && error != EFBIG) {
1743 		if_printf(ifp, "can't map mbuf (error %d)\n", error);
1744 		return(ENOBUFS);
1745 	}
1746 
1747 	/* Too many segments to map, coalesce into a single mbuf */
1748 
1749 	if (error || arg.re_maxsegs == 0) {
1750 		m_new = m_defrag_nofree(m, MB_DONTWAIT);
1751 		if (m_new == NULL)
1752 			return(1);
1753 		else {
1754 			m = m_new;
1755 			*m_head = m;
1756 		}
1757 
1758 		*called_defrag = 1;
1759 		arg.sc = sc;
1760 		arg.re_idx = *idx;
1761 		arg.re_maxsegs = sc->re_ldata.re_tx_free;
1762 		arg.re_ring = sc->re_ldata.re_tx_list;
1763 
1764 		error = bus_dmamap_load_mbuf(sc->re_ldata.re_mtag, map,
1765 		    m, re_dma_map_desc, &arg, BUS_DMA_NOWAIT);
1766 		if (error) {
1767 			m_freem(m);
1768 			if_printf(ifp, "can't map mbuf (error %d)\n", error);
1769 			return(EFBIG);
1770 		}
1771 	}
1772 
1773 	/*
1774 	 * Insure that the map for this transmission
1775 	 * is placed at the array index of the last descriptor
1776 	 * in this chain.
1777 	 */
1778 	sc->re_ldata.re_tx_dmamap[*idx] =
1779 	    sc->re_ldata.re_tx_dmamap[arg.re_idx];
1780 	sc->re_ldata.re_tx_dmamap[arg.re_idx] = map;
1781 
1782 	sc->re_ldata.re_tx_mbuf[arg.re_idx] = m;
1783 	sc->re_ldata.re_tx_free -= arg.re_maxsegs;
1784 
1785 	/*
1786 	 * Set up hardware VLAN tagging. Note: vlan tag info must
1787 	 * appear in the first descriptor of a multi-descriptor
1788 	 * transmission attempt.
1789 	 */
1790 
1791 	if ((m->m_flags & (M_PROTO1|M_PKTHDR)) == (M_PROTO1|M_PKTHDR) &&
1792 	    m->m_pkthdr.rcvif != NULL &&
1793 	    m->m_pkthdr.rcvif->if_type == IFT_L2VLAN) {
1794 	    	struct ifvlan *ifv;
1795 		ifv = m->m_pkthdr.rcvif->if_softc;
1796 		if (ifv != NULL)
1797 			sc->re_ldata.re_tx_list[*idx].re_vlanctl =
1798 			    htole32(htobe16(ifv->ifv_tag) | RE_TDESC_VLANCTL_TAG);
1799 	}
1800 
1801 	/* Transfer ownership of packet to the chip. */
1802 
1803 	sc->re_ldata.re_tx_list[arg.re_idx].re_cmdstat |=
1804 	    htole32(RE_TDESC_CMD_OWN);
1805 	if (*idx != arg.re_idx)
1806 		sc->re_ldata.re_tx_list[*idx].re_cmdstat |=
1807 		    htole32(RE_TDESC_CMD_OWN);
1808 
1809 	RE_DESC_INC(arg.re_idx);
1810 	*idx = arg.re_idx;
1811 
1812 	return(0);
1813 }
1814 
1815 /*
1816  * Main transmit routine for C+ and gigE NICs.
1817  */
1818 
1819 static void
1820 re_start(struct ifnet *ifp)
1821 {
1822 	struct re_softc	*sc = ifp->if_softc;
1823 	struct mbuf *m_head;
1824 	struct mbuf *m_head2;
1825 	int called_defrag, idx, need_trans;
1826 
1827 	idx = sc->re_ldata.re_tx_prodidx;
1828 
1829 	need_trans = 0;
1830 	while (sc->re_ldata.re_tx_mbuf[idx] == NULL) {
1831 		m_head = ifq_poll(&ifp->if_snd);
1832 		if (m_head == NULL)
1833 			break;
1834 		m_head2 = m_head;
1835 		if (re_encap(sc, &m_head2, &idx, &called_defrag)) {
1836 			/*
1837 			 * If we could not encapsulate the defragged packet,
1838 			 * the returned m_head2 is garbage and we must dequeue
1839 			 * and throw away the original packet.
1840 			 */
1841 			if (called_defrag) {
1842 				ifq_dequeue(&ifp->if_snd, m_head);
1843 				m_freem(m_head);
1844 			}
1845 			ifp->if_flags |= IFF_OACTIVE;
1846 			break;
1847 		}
1848 
1849 		/*
1850 		 * Clean out the packet we encapsulated.  If we defragged
1851 		 * the packet the m_head2 is the one that got encapsulated
1852 		 * and the original must be thrown away.  Otherwise m_head2
1853 		 * *IS* the original.
1854 		 */
1855 		ifq_dequeue(&ifp->if_snd, m_head);
1856 		if (called_defrag)
1857 			m_freem(m_head);
1858 		need_trans = 1;
1859 
1860 		/*
1861 		 * If there's a BPF listener, bounce a copy of this frame
1862 		 * to him.
1863 		 */
1864 		BPF_MTAP(ifp, m_head2);
1865 	}
1866 
1867 	if (!need_trans) {
1868 		return;
1869 	}
1870 
1871 	/* Flush the TX descriptors */
1872 	bus_dmamap_sync(sc->re_ldata.re_tx_list_tag,
1873 			sc->re_ldata.re_tx_list_map,
1874 			BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
1875 
1876 	sc->re_ldata.re_tx_prodidx = idx;
1877 
1878 	/*
1879 	 * RealTek put the TX poll request register in a different
1880 	 * location on the 8169 gigE chip. I don't know why.
1881 	 */
1882 	if (sc->re_type == RE_8169)
1883 		CSR_WRITE_2(sc, RE_GTXSTART, RE_TXSTART_START);
1884 	else
1885 		CSR_WRITE_2(sc, RE_TXSTART, RE_TXSTART_START);
1886 
1887 	/*
1888 	 * Use the countdown timer for interrupt moderation.
1889 	 * 'TX done' interrupts are disabled. Instead, we reset the
1890 	 * countdown timer, which will begin counting until it hits
1891 	 * the value in the TIMERINT register, and then trigger an
1892 	 * interrupt. Each time we write to the TIMERCNT register,
1893 	 * the timer count is reset to 0.
1894 	 */
1895 	CSR_WRITE_4(sc, RE_TIMERCNT, 1);
1896 
1897 	/*
1898 	 * Set a timeout in case the chip goes out to lunch.
1899 	 */
1900 	ifp->if_timer = 5;
1901 }
1902 
1903 static void
1904 re_init(void *xsc)
1905 {
1906 	struct re_softc *sc = xsc;
1907 	struct ifnet *ifp = &sc->arpcom.ac_if;
1908 	struct mii_data *mii;
1909 	uint32_t rxcfg = 0;
1910 
1911 	mii = device_get_softc(sc->re_miibus);
1912 
1913 	/*
1914 	 * Cancel pending I/O and free all RX/TX buffers.
1915 	 */
1916 	re_stop(sc);
1917 
1918 	/*
1919 	 * Enable C+ RX and TX mode, as well as VLAN stripping and
1920 	 * RX checksum offload. We must configure the C+ register
1921 	 * before all others.
1922 	 */
1923 	CSR_WRITE_2(sc, RE_CPLUS_CMD, RE_CPLUSCMD_RXENB | RE_CPLUSCMD_TXENB |
1924 		    RE_CPLUSCMD_PCI_MRW | RE_CPLUSCMD_VLANSTRIP |
1925 		    (ifp->if_capenable & IFCAP_RXCSUM ?
1926 		     RE_CPLUSCMD_RXCSUM_ENB : 0));
1927 
1928 	/*
1929 	 * Init our MAC address.  Even though the chipset
1930 	 * documentation doesn't mention it, we need to enter "Config
1931 	 * register write enable" mode to modify the ID registers.
1932 	 */
1933 	CSR_WRITE_1(sc, RE_EECMD, RE_EEMODE_WRITECFG);
1934 	CSR_WRITE_STREAM_4(sc, RE_IDR0,
1935 	    *(u_int32_t *)(&sc->arpcom.ac_enaddr[0]));
1936 	CSR_WRITE_STREAM_4(sc, RE_IDR4,
1937 	    *(u_int32_t *)(&sc->arpcom.ac_enaddr[4]));
1938 	CSR_WRITE_1(sc, RE_EECMD, RE_EEMODE_OFF);
1939 
1940 	/*
1941 	 * For C+ mode, initialize the RX descriptors and mbufs.
1942 	 */
1943 	re_rx_list_init(sc);
1944 	re_tx_list_init(sc);
1945 
1946 	/*
1947 	 * Enable transmit and receive.
1948 	 */
1949 	CSR_WRITE_1(sc, RE_COMMAND, RE_CMD_TX_ENB|RE_CMD_RX_ENB);
1950 
1951 	/*
1952 	 * Set the initial TX and RX configuration.
1953 	 */
1954 	if (sc->re_testmode) {
1955 		if (sc->re_type == RE_8169)
1956 			CSR_WRITE_4(sc, RE_TXCFG,
1957 				    RE_TXCFG_CONFIG | RE_LOOPTEST_ON);
1958 		else
1959 			CSR_WRITE_4(sc, RE_TXCFG,
1960 				    RE_TXCFG_CONFIG | RE_LOOPTEST_ON_CPLUS);
1961 	} else
1962 		CSR_WRITE_4(sc, RE_TXCFG, RE_TXCFG_CONFIG);
1963 	CSR_WRITE_4(sc, RE_RXCFG, RE_RXCFG_CONFIG);
1964 
1965 	/* Set the individual bit to receive frames for this host only. */
1966 	rxcfg = CSR_READ_4(sc, RE_RXCFG);
1967 	rxcfg |= RE_RXCFG_RX_INDIV;
1968 
1969 	/* If we want promiscuous mode, set the allframes bit. */
1970 	if (ifp->if_flags & IFF_PROMISC) {
1971 		rxcfg |= RE_RXCFG_RX_ALLPHYS;
1972 		CSR_WRITE_4(sc, RE_RXCFG, rxcfg);
1973 	} else {
1974 		rxcfg &= ~RE_RXCFG_RX_ALLPHYS;
1975 		CSR_WRITE_4(sc, RE_RXCFG, rxcfg);
1976 	}
1977 
1978 	/*
1979 	 * Set capture broadcast bit to capture broadcast frames.
1980 	 */
1981 	if (ifp->if_flags & IFF_BROADCAST) {
1982 		rxcfg |= RE_RXCFG_RX_BROAD;
1983 		CSR_WRITE_4(sc, RE_RXCFG, rxcfg);
1984 	} else {
1985 		rxcfg &= ~RE_RXCFG_RX_BROAD;
1986 		CSR_WRITE_4(sc, RE_RXCFG, rxcfg);
1987 	}
1988 
1989 	/*
1990 	 * Program the multicast filter, if necessary.
1991 	 */
1992 	re_setmulti(sc);
1993 
1994 #ifdef DEVICE_POLLING
1995 	/*
1996 	 * Disable interrupts if we are polling.
1997 	 */
1998 	if (ifp->if_flags & IFF_POLLING)
1999 		CSR_WRITE_2(sc, RE_IMR, 0);
2000 	else	/* otherwise ... */
2001 #endif /* DEVICE_POLLING */
2002 	/*
2003 	 * Enable interrupts.
2004 	 */
2005 	if (sc->re_testmode)
2006 		CSR_WRITE_2(sc, RE_IMR, 0);
2007 	else
2008 		CSR_WRITE_2(sc, RE_IMR, RE_INTRS_CPLUS);
2009 
2010 	/* Set initial TX threshold */
2011 	sc->re_txthresh = RE_TX_THRESH_INIT;
2012 
2013 	/* Start RX/TX process. */
2014 	CSR_WRITE_4(sc, RE_MISSEDPKT, 0);
2015 #ifdef notdef
2016 	/* Enable receiver and transmitter. */
2017 	CSR_WRITE_1(sc, RE_COMMAND, RE_CMD_TX_ENB|RE_CMD_RX_ENB);
2018 #endif
2019 	/*
2020 	 * Load the addresses of the RX and TX lists into the chip.
2021 	 */
2022 
2023 	CSR_WRITE_4(sc, RE_RXLIST_ADDR_HI,
2024 	    RE_ADDR_HI(sc->re_ldata.re_rx_list_addr));
2025 	CSR_WRITE_4(sc, RE_RXLIST_ADDR_LO,
2026 	    RE_ADDR_LO(sc->re_ldata.re_rx_list_addr));
2027 
2028 	CSR_WRITE_4(sc, RE_TXLIST_ADDR_HI,
2029 	    RE_ADDR_HI(sc->re_ldata.re_tx_list_addr));
2030 	CSR_WRITE_4(sc, RE_TXLIST_ADDR_LO,
2031 	    RE_ADDR_LO(sc->re_ldata.re_tx_list_addr));
2032 
2033 	CSR_WRITE_1(sc, RE_EARLY_TX_THRESH, 16);
2034 
2035 	/*
2036 	 * Initialize the timer interrupt register so that
2037 	 * a timer interrupt will be generated once the timer
2038 	 * reaches a certain number of ticks. The timer is
2039 	 * reloaded on each transmit. This gives us TX interrupt
2040 	 * moderation, which dramatically improves TX frame rate.
2041 	 */
2042 
2043 	if (sc->re_type == RE_8169)
2044 		CSR_WRITE_4(sc, RE_TIMERINT_8169, 0x800);
2045 	else
2046 		CSR_WRITE_4(sc, RE_TIMERINT, 0x400);
2047 
2048 	/*
2049 	 * For 8169 gigE NICs, set the max allowed RX packet
2050 	 * size so we can receive jumbo frames.
2051 	 */
2052 	if (sc->re_type == RE_8169)
2053 		CSR_WRITE_2(sc, RE_MAXRXPKTLEN, 16383);
2054 
2055 	if (sc->re_testmode) {
2056 		return;
2057 	}
2058 
2059 	mii_mediachg(mii);
2060 
2061 	CSR_WRITE_1(sc, RE_CFG1, RE_CFG1_DRVLOAD|RE_CFG1_FULLDUPLEX);
2062 
2063 	ifp->if_flags |= IFF_RUNNING;
2064 	ifp->if_flags &= ~IFF_OACTIVE;
2065 
2066 	callout_reset(&sc->re_timer, hz, re_tick, sc);
2067 }
2068 
2069 /*
2070  * Set media options.
2071  */
2072 static int
2073 re_ifmedia_upd(struct ifnet *ifp)
2074 {
2075 	struct re_softc *sc = ifp->if_softc;
2076 	struct mii_data *mii;
2077 
2078 	mii = device_get_softc(sc->re_miibus);
2079 	mii_mediachg(mii);
2080 
2081 	return(0);
2082 }
2083 
2084 /*
2085  * Report current media status.
2086  */
2087 static void
2088 re_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
2089 {
2090 	struct re_softc *sc = ifp->if_softc;
2091 	struct mii_data *mii;
2092 
2093 	mii = device_get_softc(sc->re_miibus);
2094 
2095 	mii_pollstat(mii);
2096 	ifmr->ifm_active = mii->mii_media_active;
2097 	ifmr->ifm_status = mii->mii_media_status;
2098 }
2099 
2100 static int
2101 re_ioctl(struct ifnet *ifp, u_long command, caddr_t data, struct ucred *cr)
2102 {
2103 	struct re_softc *sc = ifp->if_softc;
2104 	struct ifreq *ifr = (struct ifreq *) data;
2105 	struct mii_data *mii;
2106 	int error = 0;
2107 
2108 	switch(command) {
2109 	case SIOCSIFMTU:
2110 		if (ifr->ifr_mtu > RE_JUMBO_MTU)
2111 			error = EINVAL;
2112 		ifp->if_mtu = ifr->ifr_mtu;
2113 		break;
2114 	case SIOCSIFFLAGS:
2115 		if (ifp->if_flags & IFF_UP)
2116 			re_init(sc);
2117 		else if (ifp->if_flags & IFF_RUNNING)
2118 				re_stop(sc);
2119 		error = 0;
2120 		break;
2121 	case SIOCADDMULTI:
2122 	case SIOCDELMULTI:
2123 		re_setmulti(sc);
2124 		error = 0;
2125 		break;
2126 	case SIOCGIFMEDIA:
2127 	case SIOCSIFMEDIA:
2128 		mii = device_get_softc(sc->re_miibus);
2129 		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, command);
2130 		break;
2131 	case SIOCSIFCAP:
2132 		ifp->if_capenable &= ~(IFCAP_HWCSUM);
2133 		ifp->if_capenable |=
2134 		    ifr->ifr_reqcap & (IFCAP_HWCSUM);
2135 		if (ifp->if_capenable & IFCAP_TXCSUM)
2136 			ifp->if_hwassist = RE_CSUM_FEATURES;
2137 		else
2138 			ifp->if_hwassist = 0;
2139 		if (ifp->if_flags & IFF_RUNNING)
2140 			re_init(sc);
2141 		break;
2142 	default:
2143 		error = ether_ioctl(ifp, command, data);
2144 		break;
2145 	}
2146 	return(error);
2147 }
2148 
2149 static void
2150 re_watchdog(struct ifnet *ifp)
2151 {
2152 	struct re_softc *sc = ifp->if_softc;
2153 
2154 	if_printf(ifp, "watchdog timeout\n");
2155 
2156 	ifp->if_oerrors++;
2157 
2158 	re_txeof(sc);
2159 	re_rxeof(sc);
2160 
2161 	re_init(sc);
2162 
2163 	if (!ifq_is_empty(&ifp->if_snd))
2164 		ifp->if_start(ifp);
2165 }
2166 
2167 /*
2168  * Stop the adapter and free any mbufs allocated to the
2169  * RX and TX lists.
2170  */
2171 static void
2172 re_stop(struct re_softc *sc)
2173 {
2174 	struct ifnet *ifp = &sc->arpcom.ac_if;
2175 	int i;
2176 
2177 	ifp->if_timer = 0;
2178 	callout_stop(&sc->re_timer);
2179 
2180 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
2181 
2182 	CSR_WRITE_1(sc, RE_COMMAND, 0x00);
2183 	CSR_WRITE_2(sc, RE_IMR, 0x0000);
2184 
2185 	if (sc->re_head != NULL) {
2186 		m_freem(sc->re_head);
2187 		sc->re_head = sc->re_tail = NULL;
2188 	}
2189 
2190 	/* Free the TX list buffers. */
2191 	for (i = 0; i < RE_TX_DESC_CNT; i++) {
2192 		if (sc->re_ldata.re_tx_mbuf[i] != NULL) {
2193 			bus_dmamap_unload(sc->re_ldata.re_mtag,
2194 					  sc->re_ldata.re_tx_dmamap[i]);
2195 			m_freem(sc->re_ldata.re_tx_mbuf[i]);
2196 			sc->re_ldata.re_tx_mbuf[i] = NULL;
2197 		}
2198 	}
2199 
2200 	/* Free the RX list buffers. */
2201 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
2202 		if (sc->re_ldata.re_rx_mbuf[i] != NULL) {
2203 			bus_dmamap_unload(sc->re_ldata.re_mtag,
2204 					  sc->re_ldata.re_rx_dmamap[i]);
2205 			m_freem(sc->re_ldata.re_rx_mbuf[i]);
2206 			sc->re_ldata.re_rx_mbuf[i] = NULL;
2207 		}
2208 	}
2209 }
2210 
2211 /*
2212  * Device suspend routine.  Stop the interface and save some PCI
2213  * settings in case the BIOS doesn't restore them properly on
2214  * resume.
2215  */
2216 static int
2217 re_suspend(device_t dev)
2218 {
2219 #ifndef BURN_BRIDGES
2220 	int i;
2221 #endif
2222 	struct re_softc *sc = device_get_softc(dev);
2223 
2224 	re_stop(sc);
2225 
2226 #ifndef BURN_BRIDGES
2227 	for (i = 0; i < 5; i++)
2228 		sc->saved_maps[i] = pci_read_config(dev, PCIR_MAPS + i * 4, 4);
2229 	sc->saved_biosaddr = pci_read_config(dev, PCIR_BIOS, 4);
2230 	sc->saved_intline = pci_read_config(dev, PCIR_INTLINE, 1);
2231 	sc->saved_cachelnsz = pci_read_config(dev, PCIR_CACHELNSZ, 1);
2232 	sc->saved_lattimer = pci_read_config(dev, PCIR_LATTIMER, 1);
2233 #endif
2234 
2235 	sc->suspended = 1;
2236 
2237 	return (0);
2238 }
2239 
2240 /*
2241  * Device resume routine.  Restore some PCI settings in case the BIOS
2242  * doesn't, re-enable busmastering, and restart the interface if
2243  * appropriate.
2244  */
2245 static int
2246 re_resume(device_t dev)
2247 {
2248 	struct re_softc *sc = device_get_softc(dev);
2249 	struct ifnet *ifp = &sc->arpcom.ac_if;
2250 #ifndef BURN_BRIDGES
2251 	int i;
2252 #endif
2253 
2254 #ifndef BURN_BRIDGES
2255 	/* better way to do this? */
2256 	for (i = 0; i < 5; i++)
2257 		pci_write_config(dev, PCIR_MAPS + i * 4, sc->saved_maps[i], 4);
2258 	pci_write_config(dev, PCIR_BIOS, sc->saved_biosaddr, 4);
2259 	pci_write_config(dev, PCIR_INTLINE, sc->saved_intline, 1);
2260 	pci_write_config(dev, PCIR_CACHELNSZ, sc->saved_cachelnsz, 1);
2261 	pci_write_config(dev, PCIR_LATTIMER, sc->saved_lattimer, 1);
2262 
2263 	/* reenable busmastering */
2264 	pci_enable_busmaster(dev);
2265 	pci_enable_io(dev, SYS_RES_IOPORT);
2266 #endif
2267 
2268 	/* reinitialize interface if necessary */
2269 	if (ifp->if_flags & IFF_UP)
2270 		re_init(sc);
2271 
2272 	sc->suspended = 0;
2273 
2274 	return (0);
2275 }
2276 
2277 /*
2278  * Stop all chip I/O so that the kernel's probe routines don't
2279  * get confused by errant DMAs when rebooting.
2280  */
2281 static void
2282 re_shutdown(device_t dev)
2283 {
2284 	struct re_softc *sc = device_get_softc(dev);
2285 
2286 	re_stop(sc);
2287 }
2288