xref: /dflybsd-src/sys/dev/netif/re/if_re.c (revision b5b0912b1891e95ccc48cad83f09239ccb7ffc16)
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 1.12 2005/05/25 01:44:27 dillon 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 7.5K, so the max MTU possible with this
111  * driver is 7500 bytes.
112  */
113 
114 #include <sys/param.h>
115 #include <sys/endian.h>
116 #include <sys/systm.h>
117 #include <sys/sockio.h>
118 #include <sys/mbuf.h>
119 #include <sys/malloc.h>
120 #include <sys/module.h>
121 #include <sys/kernel.h>
122 #include <sys/socket.h>
123 
124 #include <net/if.h>
125 #include <net/ifq_var.h>
126 #include <net/if_arp.h>
127 #include <net/ethernet.h>
128 #include <net/if_dl.h>
129 #include <net/if_media.h>
130 #include <net/if_types.h>
131 #include <net/vlan/if_vlan_var.h>
132 
133 #include <net/bpf.h>
134 
135 #include <machine/bus_pio.h>
136 #include <machine/bus_memio.h>
137 #include <machine/bus.h>
138 #include <machine/resource.h>
139 #include <sys/bus.h>
140 #include <sys/rman.h>
141 
142 #include <dev/netif/mii_layer/mii.h>
143 #include <dev/netif/mii_layer/miivar.h>
144 
145 #include <bus/pci/pcireg.h>
146 #include <bus/pci/pcivar.h>
147 
148 /* "controller miibus0" required.  See GENERIC if you get errors here. */
149 #include "miibus_if.h"
150 
151 #include <dev/netif/re/if_rereg.h>
152 
153 /*
154  * The hardware supports checksumming but, as usual, some chipsets screw it
155  * all up and produce bogus packets, so we disable it by default.
156  */
157 #define RE_CSUM_FEATURES    (CSUM_IP | CSUM_TCP | CSUM_UDP)
158 #define RE_DISABLE_HWCSUM
159 
160 /*
161  * Various supported device vendors/types and their names.
162  */
163 static struct re_type re_devs[] = {
164 	{ RT_VENDORID, RT_DEVICEID_8139, RE_HWREV_8139CPLUS,
165 		"RealTek 8139C+ 10/100BaseTX" },
166 	{ RT_VENDORID, RT_DEVICEID_8169, RE_HWREV_8169,
167 		"RealTek 8169 Gigabit Ethernet" },
168 	{ RT_VENDORID, RT_DEVICEID_8169, RE_HWREV_8169S,
169 		"RealTek 8169S Single-chip Gigabit Ethernet" },
170 	{ RT_VENDORID, RT_DEVICEID_8169, RE_HWREV_8110S,
171 		"RealTek 8110S Single-chip Gigabit Ethernet" },
172 	{ 0, 0, 0, NULL }
173 };
174 
175 static struct re_hwrev re_hwrevs[] = {
176 	{ RE_HWREV_8139CPLUS, RE_8139CPLUS, "C+"},
177 	{ RE_HWREV_8169, RE_8169, "8169"},
178 	{ RE_HWREV_8169S, RE_8169, "8169S"},
179 	{ RE_HWREV_8110S, RE_8169, "8110S"},
180 	{ 0, 0, NULL }
181 };
182 
183 static int	re_probe(device_t);
184 static int	re_attach(device_t);
185 static int	re_detach(device_t);
186 
187 static int	re_encap(struct re_softc *, struct mbuf **, int *, int *);
188 
189 static void	re_dma_map_addr(void *, bus_dma_segment_t *, int, int);
190 static void	re_dma_map_desc(void *, bus_dma_segment_t *, int,
191 				bus_size_t, int);
192 static int	re_allocmem(device_t, struct re_softc *);
193 static int	re_newbuf(struct re_softc *, int, struct mbuf *);
194 static int	re_rx_list_init(struct re_softc *);
195 static int	re_tx_list_init(struct re_softc *);
196 static void	re_rxeof(struct re_softc *);
197 static void	re_txeof(struct re_softc *);
198 static void	re_intr(void *);
199 static void	re_tick(void *);
200 static void	re_start(struct ifnet *);
201 static int	re_ioctl(struct ifnet *, u_long, caddr_t, struct ucred *);
202 static void	re_init(void *);
203 static void	re_stop(struct re_softc *);
204 static void	re_watchdog(struct ifnet *);
205 static int	re_suspend(device_t);
206 static int	re_resume(device_t);
207 static void	re_shutdown(device_t);
208 static int	re_ifmedia_upd(struct ifnet *);
209 static void	re_ifmedia_sts(struct ifnet *, struct ifmediareq *);
210 
211 static void	re_eeprom_putbyte(struct re_softc *, int);
212 static void	re_eeprom_getword(struct re_softc *, int, u_int16_t *);
213 static void	re_read_eeprom(struct re_softc *, caddr_t, int, int, int);
214 static int	re_gmii_readreg(device_t, int, int);
215 static int	re_gmii_writereg(device_t, int, int, int);
216 
217 static int	re_miibus_readreg(device_t, int, int);
218 static int	re_miibus_writereg(device_t, int, int, int);
219 static void	re_miibus_statchg(device_t);
220 
221 static void	re_setmulti(struct re_softc *);
222 static void	re_reset(struct re_softc *);
223 
224 static int	re_diag(struct re_softc *);
225 #ifdef DEVICE_POLLING
226 static void	re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count);
227 #endif
228 
229 static device_method_t re_methods[] = {
230 	/* Device interface */
231 	DEVMETHOD(device_probe,		re_probe),
232 	DEVMETHOD(device_attach,	re_attach),
233 	DEVMETHOD(device_detach,	re_detach),
234 	DEVMETHOD(device_suspend,	re_suspend),
235 	DEVMETHOD(device_resume,	re_resume),
236 	DEVMETHOD(device_shutdown,	re_shutdown),
237 
238 	/* bus interface */
239 	DEVMETHOD(bus_print_child,	bus_generic_print_child),
240 	DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
241 
242 	/* MII interface */
243 	DEVMETHOD(miibus_readreg,	re_miibus_readreg),
244 	DEVMETHOD(miibus_writereg,	re_miibus_writereg),
245 	DEVMETHOD(miibus_statchg,	re_miibus_statchg),
246 
247 	{ 0, 0 }
248 };
249 
250 static driver_t re_driver = {
251 	"re",
252 	re_methods,
253 	sizeof(struct re_softc)
254 };
255 
256 static devclass_t re_devclass;
257 
258 DECLARE_DUMMY_MODULE(if_re);
259 DRIVER_MODULE(if_re, pci, re_driver, re_devclass, 0, 0);
260 DRIVER_MODULE(if_re, cardbus, re_driver, re_devclass, 0, 0);
261 DRIVER_MODULE(miibus, re, miibus_driver, miibus_devclass, 0, 0);
262 
263 #define EE_SET(x)	\
264 	CSR_WRITE_1(sc, RE_EECMD, CSR_READ_1(sc, RE_EECMD) | (x))
265 
266 #define EE_CLR(x)	\
267 	CSR_WRITE_1(sc, RE_EECMD, CSR_READ_1(sc, RE_EECMD) & ~(x))
268 
269 /*
270  * Send a read command and address to the EEPROM, check for ACK.
271  */
272 static void
273 re_eeprom_putbyte(struct re_softc *sc, int addr)
274 {
275 	int d, i;
276 
277 	d = addr | sc->re_eecmd_read;
278 
279 	/*
280 	 * Feed in each bit and strobe the clock.
281 	 */
282 	for (i = 0x400; i != 0; i >>= 1) {
283 		if (d & i)
284 			EE_SET(RE_EE_DATAIN);
285 		else
286 			EE_CLR(RE_EE_DATAIN);
287 		DELAY(100);
288 		EE_SET(RE_EE_CLK);
289 		DELAY(150);
290 		EE_CLR(RE_EE_CLK);
291 		DELAY(100);
292 	}
293 }
294 
295 /*
296  * Read a word of data stored in the EEPROM at address 'addr.'
297  */
298 static void
299 re_eeprom_getword(struct re_softc *sc, int addr, uint16_t *dest)
300 {
301 	int i;
302 	uint16_t word = 0;
303 
304 	/* Enter EEPROM access mode. */
305 	CSR_WRITE_1(sc, RE_EECMD, RE_EEMODE_PROGRAM|RE_EE_SEL);
306 
307 	/*
308 	 * Send address of word we want to read.
309 	 */
310 	re_eeprom_putbyte(sc, addr);
311 
312 	CSR_WRITE_1(sc, RE_EECMD, RE_EEMODE_PROGRAM|RE_EE_SEL);
313 
314 	/*
315 	 * Start reading bits from EEPROM.
316 	 */
317 	for (i = 0x8000; i != 0; i >>= 1) {
318 		EE_SET(RE_EE_CLK);
319 		DELAY(100);
320 		if (CSR_READ_1(sc, RE_EECMD) & RE_EE_DATAOUT)
321 			word |= i;
322 		EE_CLR(RE_EE_CLK);
323 		DELAY(100);
324 	}
325 
326 	/* Turn off EEPROM access mode. */
327 	CSR_WRITE_1(sc, RE_EECMD, RE_EEMODE_OFF);
328 
329 	*dest = word;
330 }
331 
332 /*
333  * Read a sequence of words from the EEPROM.
334  */
335 static void
336 re_read_eeprom(struct re_softc *sc, caddr_t dest, int off, int cnt, int swap)
337 {
338 	int i;
339 	uint16_t word = 0, *ptr;
340 
341 	for (i = 0; i < cnt; i++) {
342 		re_eeprom_getword(sc, off + i, &word);
343 		ptr = (u_int16_t *)(dest + (i * 2));
344 		if (swap)
345 			*ptr = be16toh(word);
346 		else
347 			*ptr = word;
348 	}
349 }
350 
351 static int
352 re_gmii_readreg(device_t dev, int phy, int reg)
353 {
354 	struct re_softc *sc = device_get_softc(dev);
355 	u_int32_t rval;
356 	int i;
357 
358 	if (phy != 1)
359 		return(0);
360 
361 	/* Let the rgephy driver read the GMEDIASTAT register */
362 
363 	if (reg == RE_GMEDIASTAT)
364 		return(CSR_READ_1(sc, RE_GMEDIASTAT));
365 
366 	CSR_WRITE_4(sc, RE_PHYAR, reg << 16);
367 	DELAY(1000);
368 
369 	for (i = 0; i < RE_TIMEOUT; i++) {
370 		rval = CSR_READ_4(sc, RE_PHYAR);
371 		if (rval & RE_PHYAR_BUSY)
372 			break;
373 		DELAY(100);
374 	}
375 
376 	if (i == RE_TIMEOUT) {
377 		device_printf(dev, "PHY read failed\n");
378 		return(0);
379 	}
380 
381 	return(rval & RE_PHYAR_PHYDATA);
382 }
383 
384 static int
385 re_gmii_writereg(device_t dev, int phy, int reg, int data)
386 {
387 	struct re_softc *sc = device_get_softc(dev);
388 	uint32_t rval;
389 	int i;
390 
391 	CSR_WRITE_4(sc, RE_PHYAR,
392 		    (reg << 16) | (data & RE_PHYAR_PHYDATA) | RE_PHYAR_BUSY);
393 	DELAY(1000);
394 
395 	for (i = 0; i < RE_TIMEOUT; i++) {
396 		rval = CSR_READ_4(sc, RE_PHYAR);
397 		if ((rval & RE_PHYAR_BUSY) == 0)
398 			break;
399 		DELAY(100);
400 	}
401 
402 	if (i == RE_TIMEOUT)
403 		device_printf(dev, "PHY write failed\n");
404 
405 	return(0);
406 }
407 
408 static int
409 re_miibus_readreg(device_t dev, int phy, int reg)
410 {
411 	struct re_softc	*sc = device_get_softc(dev);
412 	uint16_t rval = 0;
413 	uint16_t re8139_reg = 0;
414 
415 	if (sc->re_type == RE_8169) {
416 		rval = re_gmii_readreg(dev, phy, reg);
417 		return(rval);
418 	}
419 
420 	/* Pretend the internal PHY is only at address 0 */
421 	if (phy)
422 		return(0);
423 
424 	switch(reg) {
425 	case MII_BMCR:
426 		re8139_reg = RE_BMCR;
427 		break;
428 	case MII_BMSR:
429 		re8139_reg = RE_BMSR;
430 		break;
431 	case MII_ANAR:
432 		re8139_reg = RE_ANAR;
433 		break;
434 	case MII_ANER:
435 		re8139_reg = RE_ANER;
436 		break;
437 	case MII_ANLPAR:
438 		re8139_reg = RE_LPAR;
439 		break;
440 	case MII_PHYIDR1:
441 	case MII_PHYIDR2:
442 		return(0);
443 	/*
444 	 * Allow the rlphy driver to read the media status
445 	 * register. If we have a link partner which does not
446 	 * support NWAY, this is the register which will tell
447 	 * us the results of parallel detection.
448 	 */
449 	case RE_MEDIASTAT:
450 		return(CSR_READ_1(sc, RE_MEDIASTAT));
451 	default:
452 		device_printf(dev, "bad phy register\n");
453 		return(0);
454 	}
455 	rval = CSR_READ_2(sc, re8139_reg);
456 	return(rval);
457 }
458 
459 static int
460 re_miibus_writereg(device_t dev, int phy, int reg, int data)
461 {
462 	struct re_softc *sc= device_get_softc(dev);
463 	u_int16_t re8139_reg = 0;
464 
465 	if (sc->re_type == RE_8169)
466 		return(re_gmii_writereg(dev, phy, reg, data));
467 
468 	/* Pretend the internal PHY is only at address 0 */
469 	if (phy)
470 		return(0);
471 
472 	switch(reg) {
473 	case MII_BMCR:
474 		re8139_reg = RE_BMCR;
475 		break;
476 	case MII_BMSR:
477 		re8139_reg = RE_BMSR;
478 		break;
479 	case MII_ANAR:
480 		re8139_reg = RE_ANAR;
481 		break;
482 	case MII_ANER:
483 		re8139_reg = RE_ANER;
484 		break;
485 	case MII_ANLPAR:
486 		re8139_reg = RE_LPAR;
487 		break;
488 	case MII_PHYIDR1:
489 	case MII_PHYIDR2:
490 		return(0);
491 	default:
492 		device_printf(dev, "bad phy register\n");
493 		return(0);
494 	}
495 	CSR_WRITE_2(sc, re8139_reg, data);
496 	return(0);
497 }
498 
499 static void
500 re_miibus_statchg(device_t dev)
501 {
502 }
503 
504 /*
505  * Program the 64-bit multicast hash filter.
506  */
507 static void
508 re_setmulti(struct re_softc *sc)
509 {
510 	struct ifnet *ifp = &sc->arpcom.ac_if;
511 	int h = 0;
512 	uint32_t hashes[2] = { 0, 0 };
513 	struct ifmultiaddr *ifma;
514 	uint32_t rxfilt;
515 	int mcnt = 0;
516 
517 	rxfilt = CSR_READ_4(sc, RE_RXCFG);
518 
519 	if (ifp->if_flags & IFF_ALLMULTI || ifp->if_flags & IFF_PROMISC) {
520 		rxfilt |= RE_RXCFG_RX_MULTI;
521 		CSR_WRITE_4(sc, RE_RXCFG, rxfilt);
522 		CSR_WRITE_4(sc, RE_MAR0, 0xFFFFFFFF);
523 		CSR_WRITE_4(sc, RE_MAR4, 0xFFFFFFFF);
524 		return;
525 	}
526 
527 	/* first, zot all the existing hash bits */
528 	CSR_WRITE_4(sc, RE_MAR0, 0);
529 	CSR_WRITE_4(sc, RE_MAR4, 0);
530 
531 	/* now program new ones */
532 	LIST_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
533 		if (ifma->ifma_addr->sa_family != AF_LINK)
534 			continue;
535 		h = ether_crc32_be(LLADDR((struct sockaddr_dl *)
536 		    ifma->ifma_addr), ETHER_ADDR_LEN) >> 26;
537 		if (h < 32)
538 			hashes[0] |= (1 << h);
539 		else
540 			hashes[1] |= (1 << (h - 32));
541 		mcnt++;
542 	}
543 
544 	if (mcnt)
545 		rxfilt |= RE_RXCFG_RX_MULTI;
546 	else
547 		rxfilt &= ~RE_RXCFG_RX_MULTI;
548 
549 	CSR_WRITE_4(sc, RE_RXCFG, rxfilt);
550 	CSR_WRITE_4(sc, RE_MAR0, hashes[0]);
551 	CSR_WRITE_4(sc, RE_MAR4, hashes[1]);
552 }
553 
554 static void
555 re_reset(struct re_softc *sc)
556 {
557 	int i;
558 
559 	CSR_WRITE_1(sc, RE_COMMAND, RE_CMD_RESET);
560 
561 	for (i = 0; i < RE_TIMEOUT; i++) {
562 		DELAY(10);
563 		if ((CSR_READ_1(sc, RE_COMMAND) & RE_CMD_RESET) == 0)
564 			break;
565 	}
566 	if (i == RE_TIMEOUT)
567 		if_printf(&sc->arpcom.ac_if, "reset never completed!\n");
568 
569 	CSR_WRITE_1(sc, 0x82, 1);
570 }
571 
572 /*
573  * The following routine is designed to test for a defect on some
574  * 32-bit 8169 cards. Some of these NICs have the REQ64# and ACK64#
575  * lines connected to the bus, however for a 32-bit only card, they
576  * should be pulled high. The result of this defect is that the
577  * NIC will not work right if you plug it into a 64-bit slot: DMA
578  * operations will be done with 64-bit transfers, which will fail
579  * because the 64-bit data lines aren't connected.
580  *
581  * There's no way to work around this (short of talking a soldering
582  * iron to the board), however we can detect it. The method we use
583  * here is to put the NIC into digital loopback mode, set the receiver
584  * to promiscuous mode, and then try to send a frame. We then compare
585  * the frame data we sent to what was received. If the data matches,
586  * then the NIC is working correctly, otherwise we know the user has
587  * a defective NIC which has been mistakenly plugged into a 64-bit PCI
588  * slot. In the latter case, there's no way the NIC can work correctly,
589  * so we print out a message on the console and abort the device attach.
590  */
591 
592 static int
593 re_diag(struct re_softc *sc)
594 {
595 	struct ifnet *ifp = &sc->arpcom.ac_if;
596 	struct mbuf *m0;
597 	struct ether_header *eh;
598 	struct re_desc *cur_rx;
599 	uint16_t status;
600 	uint32_t rxstat;
601 	int total_len, i, error = 0;
602 	uint8_t dst[ETHER_ADDR_LEN] = { 0x00, 'h', 'e', 'l', 'l', 'o' };
603 	uint8_t src[ETHER_ADDR_LEN] = { 0x00, 'w', 'o', 'r', 'l', 'd' };
604 
605 	/* Allocate a single mbuf */
606 
607 	MGETHDR(m0, MB_DONTWAIT, MT_DATA);
608 	if (m0 == NULL)
609 		return(ENOBUFS);
610 
611 	/*
612 	 * Initialize the NIC in test mode. This sets the chip up
613 	 * so that it can send and receive frames, but performs the
614 	 * following special functions:
615 	 * - Puts receiver in promiscuous mode
616 	 * - Enables digital loopback mode
617 	 * - Leaves interrupts turned off
618 	 */
619 
620 	ifp->if_flags |= IFF_PROMISC;
621 	sc->re_testmode = 1;
622 	re_init(sc);
623 	re_stop(sc);
624 	DELAY(100000);
625 	re_init(sc);
626 
627 	/* Put some data in the mbuf */
628 
629 	eh = mtod(m0, struct ether_header *);
630 	bcopy (dst, eh->ether_dhost, ETHER_ADDR_LEN);
631 	bcopy (src, eh->ether_shost, ETHER_ADDR_LEN);
632 	eh->ether_type = htons(ETHERTYPE_IP);
633 	m0->m_pkthdr.len = m0->m_len = ETHER_MIN_LEN - ETHER_CRC_LEN;
634 
635 	/*
636 	 * Queue the packet, start transmission.
637 	 * Note: ifq_handoff() ultimately calls re_start() for us.
638 	 */
639 
640 	CSR_WRITE_2(sc, RE_ISR, 0xFFFF);
641 	error = ifq_handoff(ifp, m0, NULL);
642 	if (error) {
643 		m0 = NULL;
644 		goto done;
645 	}
646 	m0 = NULL;
647 
648 	/* Wait for it to propagate through the chip */
649 
650 	DELAY(100000);
651 	for (i = 0; i < RE_TIMEOUT; i++) {
652 		status = CSR_READ_2(sc, RE_ISR);
653 		if ((status & (RE_ISR_TIMEOUT_EXPIRED|RE_ISR_RX_OK)) ==
654 		    (RE_ISR_TIMEOUT_EXPIRED|RE_ISR_RX_OK))
655 			break;
656 		DELAY(10);
657 	}
658 
659 	if (i == RE_TIMEOUT) {
660 		if_printf(ifp, "diagnostic failed to receive packet "
661 			  "in loopback mode\n");
662 		error = EIO;
663 		goto done;
664 	}
665 
666 	/*
667 	 * The packet should have been dumped into the first
668 	 * entry in the RX DMA ring. Grab it from there.
669 	 */
670 
671 	bus_dmamap_sync(sc->re_ldata.re_rx_list_tag,
672 			sc->re_ldata.re_rx_list_map, BUS_DMASYNC_POSTREAD);
673 	bus_dmamap_sync(sc->re_ldata.re_mtag, sc->re_ldata.re_rx_dmamap[0],
674 			BUS_DMASYNC_POSTWRITE);
675 	bus_dmamap_unload(sc->re_ldata.re_mtag, sc->re_ldata.re_rx_dmamap[0]);
676 
677 	m0 = sc->re_ldata.re_rx_mbuf[0];
678 	sc->re_ldata.re_rx_mbuf[0] = NULL;
679 	eh = mtod(m0, struct ether_header *);
680 
681 	cur_rx = &sc->re_ldata.re_rx_list[0];
682 	total_len = RE_RXBYTES(cur_rx);
683 	rxstat = le32toh(cur_rx->re_cmdstat);
684 
685 	if (total_len != ETHER_MIN_LEN) {
686 		if_printf(ifp, "diagnostic failed, received short packet\n");
687 		error = EIO;
688 		goto done;
689 	}
690 
691 	/* Test that the received packet data matches what we sent. */
692 
693 	if (bcmp(eh->ether_dhost, dst, ETHER_ADDR_LEN) ||
694 	    bcmp(eh->ether_shost, &src, ETHER_ADDR_LEN) ||
695 	    be16toh(eh->ether_type) != ETHERTYPE_IP) {
696 		if_printf(ifp, "WARNING, DMA FAILURE!\n");
697 		if_printf(ifp, "expected TX data: %6D/%6D/0x%x\n",
698 		    dst, ":", src, ":", ETHERTYPE_IP);
699 		if_printf(ifp, "received RX data: %6D/%6D/0x%x\n",
700 		    eh->ether_dhost, ":",  eh->ether_shost, ":",
701 		    ntohs(eh->ether_type));
702 		if_printf(ifp, "You may have a defective 32-bit NIC plugged "
703 		    "into a 64-bit PCI slot.\n");
704 		if_printf(ifp, "Please re-install the NIC in a 32-bit slot "
705 		    "for proper operation.\n");
706 		if_printf(ifp, "Read the re(4) man page for more details.\n");
707 		error = EIO;
708 	}
709 
710 done:
711 	/* Turn interface off, release resources */
712 
713 	sc->re_testmode = 0;
714 	ifp->if_flags &= ~IFF_PROMISC;
715 	re_stop(sc);
716 	if (m0 != NULL)
717 		m_freem(m0);
718 
719 	return (error);
720 }
721 
722 /*
723  * Probe for a RealTek 8139C+/8169/8110 chip. Check the PCI vendor and device
724  * IDs against our list and return a device name if we find a match.
725  */
726 static int
727 re_probe(device_t dev)
728 {
729 	struct re_type *t;
730 	struct re_softc *sc;
731 	int rid;
732 	uint32_t hwrev;
733 	uint16_t vendor, product;
734 
735 	t = re_devs;
736 
737 	vendor = pci_get_vendor(dev);
738 	product = pci_get_device(dev);
739 
740 	for (t = re_devs; t->re_name != NULL; t++) {
741 		if (product == t->re_did && vendor == t->re_vid)
742 			break;
743 	}
744 
745 	/*
746 	 * Check if we found a RealTek device.
747 	 */
748 	if (t->re_name == NULL)
749 		return(ENXIO);
750 
751 	/*
752 	 * Temporarily map the I/O space so we can read the chip ID register.
753 	 */
754 	sc = malloc(sizeof(*sc), M_TEMP, M_WAITOK | M_ZERO);
755 	rid = RE_PCI_LOIO;
756 	sc->re_res = bus_alloc_resource_any(dev, SYS_RES_IOPORT, &rid,
757 					    RF_ACTIVE);
758 	if (sc->re_res == NULL) {
759 		device_printf(dev, "couldn't map ports/memory\n");
760 		free(sc, M_TEMP);
761 		return(ENXIO);
762 	}
763 
764 	sc->re_btag = rman_get_bustag(sc->re_res);
765 	sc->re_bhandle = rman_get_bushandle(sc->re_res);
766 
767 	hwrev = CSR_READ_4(sc, RE_TXCFG) & RE_TXCFG_HWREV;
768 	bus_release_resource(dev, SYS_RES_IOPORT, RE_PCI_LOIO, sc->re_res);
769 	free(sc, M_TEMP);
770 
771 	/*
772 	 * and continue matching for the specific chip...
773 	 */
774 	for (; t->re_name != NULL; t++) {
775 		if (product == t->re_did && vendor == t->re_vid &&
776 		    t->re_basetype == hwrev) {
777 			device_set_desc(dev, t->re_name);
778 			return(0);
779 		}
780 	}
781 	return(ENXIO);
782 }
783 
784 /*
785  * This routine takes the segment list provided as the result of
786  * a bus_dma_map_load() operation and assigns the addresses/lengths
787  * to RealTek DMA descriptors. This can be called either by the RX
788  * code or the TX code. In the RX case, we'll probably wind up mapping
789  * at most one segment. For the TX case, there could be any number of
790  * segments since TX packets may span multiple mbufs. In either case,
791  * if the number of segments is larger than the re_maxsegs limit
792  * specified by the caller, we abort the mapping operation. Sadly,
793  * whoever designed the buffer mapping API did not provide a way to
794  * return an error from here, so we have to fake it a bit.
795  */
796 
797 static void
798 re_dma_map_desc(void *arg, bus_dma_segment_t *segs, int nseg,
799 		bus_size_t mapsize, int error)
800 {
801 	struct re_dmaload_arg *ctx;
802 	struct re_desc *d = NULL;
803 	int i = 0, idx;
804 	uint32_t cmdstat;
805 
806 	if (error)
807 		return;
808 
809 	ctx = arg;
810 
811 	/* Signal error to caller if there's too many segments */
812 	if (nseg > ctx->re_maxsegs) {
813 		ctx->re_maxsegs = 0;
814 		return;
815 	}
816 
817 	/*
818 	 * Map the segment array into descriptors. Note that we set the
819 	 * start-of-frame and end-of-frame markers for either TX or RX, but
820 	 * they really only have meaning in the TX case. (In the RX case,
821 	 * it's the chip that tells us where packets begin and end.)
822 	 * We also keep track of the end of the ring and set the
823 	 * end-of-ring bits as needed, and we set the ownership bits
824 	 * in all except the very first descriptor. (The caller will
825 	 * set this descriptor later when it start transmission or
826 	 * reception.)
827 	 */
828 	idx = ctx->re_idx;
829 	for (;;) {
830 		d = &ctx->re_ring[idx];
831 		if (le32toh(d->re_cmdstat) & RE_RDESC_STAT_OWN) {
832 			ctx->re_maxsegs = 0;
833 			return;
834 		}
835 		cmdstat = segs[i].ds_len;
836 		d->re_bufaddr_lo = htole32(RE_ADDR_LO(segs[i].ds_addr));
837 		d->re_bufaddr_hi = htole32(RE_ADDR_HI(segs[i].ds_addr));
838 		if (i == 0)
839 			cmdstat |= RE_TDESC_CMD_SOF;
840 		else
841 			cmdstat |= RE_TDESC_CMD_OWN;
842 		if (idx == (RE_RX_DESC_CNT - 1))
843 			cmdstat |= RE_TDESC_CMD_EOR;
844 		d->re_cmdstat = htole32(cmdstat | ctx->re_flags);
845 		i++;
846 		if (i == nseg)
847 			break;
848 		RE_DESC_INC(idx);
849 	}
850 
851 	d->re_cmdstat |= htole32(RE_TDESC_CMD_EOF);
852 	ctx->re_maxsegs = nseg;
853 	ctx->re_idx = idx;
854 }
855 
856 /*
857  * Map a single buffer address.
858  */
859 
860 static void
861 re_dma_map_addr(void *arg, bus_dma_segment_t *segs, int nseg, int error)
862 {
863 	uint32_t *addr;
864 
865 	if (error)
866 		return;
867 
868 	KASSERT(nseg == 1, ("too many DMA segments, %d should be 1", nseg));
869 	addr = arg;
870 	*addr = segs->ds_addr;
871 }
872 
873 static int
874 re_allocmem(device_t dev, struct re_softc *sc)
875 {
876 	int error, i, nseg;
877 
878 	/*
879 	 * Allocate map for RX mbufs.
880 	 */
881 	nseg = 32;
882 	error = bus_dma_tag_create(sc->re_parent_tag, ETHER_ALIGN, 0,
883 	    BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
884 	    NULL, MCLBYTES * nseg, nseg, MCLBYTES, BUS_DMA_ALLOCNOW,
885 	    &sc->re_ldata.re_mtag);
886 	if (error) {
887 		device_printf(dev, "could not allocate dma tag\n");
888 		return(error);
889 	}
890 
891 	/*
892 	 * Allocate map for TX descriptor list.
893 	 */
894 	error = bus_dma_tag_create(sc->re_parent_tag, RE_RING_ALIGN,
895 	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
896             NULL, RE_TX_LIST_SZ, 1, RE_TX_LIST_SZ, BUS_DMA_ALLOCNOW,
897 	    &sc->re_ldata.re_tx_list_tag);
898 	if (error) {
899 		device_printf(dev, "could not allocate dma tag\n");
900 		return(error);
901 	}
902 
903 	/* Allocate DMA'able memory for the TX ring */
904 
905         error = bus_dmamem_alloc(sc->re_ldata.re_tx_list_tag,
906 	    (void **)&sc->re_ldata.re_tx_list, BUS_DMA_WAITOK | BUS_DMA_ZERO,
907             &sc->re_ldata.re_tx_list_map);
908         if (error) {
909 		device_printf(dev, "could not allocate TX ring\n");
910                 return(error);
911 	}
912 
913 	/* Load the map for the TX ring. */
914 
915 	error = bus_dmamap_load(sc->re_ldata.re_tx_list_tag,
916 	     sc->re_ldata.re_tx_list_map, sc->re_ldata.re_tx_list,
917 	     RE_TX_LIST_SZ, re_dma_map_addr,
918 	     &sc->re_ldata.re_tx_list_addr, BUS_DMA_NOWAIT);
919 	if (error) {
920 		device_printf(dev, "could not get addres of TX ring\n");
921 		return(error);
922 	}
923 
924 	/* Create DMA maps for TX buffers */
925 
926 	for (i = 0; i < RE_TX_DESC_CNT; i++) {
927 		error = bus_dmamap_create(sc->re_ldata.re_mtag, 0,
928 			    &sc->re_ldata.re_tx_dmamap[i]);
929 		if (error) {
930 			device_printf(dev, "can't create DMA map for TX\n");
931 			return(error);
932 		}
933 	}
934 
935 	/*
936 	 * Allocate map for RX descriptor list.
937 	 */
938 	error = bus_dma_tag_create(sc->re_parent_tag, RE_RING_ALIGN,
939 	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
940             NULL, RE_TX_LIST_SZ, 1, RE_TX_LIST_SZ, BUS_DMA_ALLOCNOW,
941 	    &sc->re_ldata.re_rx_list_tag);
942 	if (error) {
943 		device_printf(dev, "could not allocate dma tag\n");
944 		return(error);
945 	}
946 
947 	/* Allocate DMA'able memory for the RX ring */
948 
949         error = bus_dmamem_alloc(sc->re_ldata.re_rx_list_tag,
950 	    (void **)&sc->re_ldata.re_rx_list, BUS_DMA_WAITOK | BUS_DMA_ZERO,
951             &sc->re_ldata.re_rx_list_map);
952         if (error) {
953 		device_printf(dev, "could not allocate RX ring\n");
954                 return(error);
955 	}
956 
957 	/* Load the map for the RX ring. */
958 
959 	error = bus_dmamap_load(sc->re_ldata.re_rx_list_tag,
960 	     sc->re_ldata.re_rx_list_map, sc->re_ldata.re_rx_list,
961 	     RE_TX_LIST_SZ, re_dma_map_addr,
962 	     &sc->re_ldata.re_rx_list_addr, BUS_DMA_NOWAIT);
963 	if (error) {
964 		device_printf(dev, "could not get address of RX ring\n");
965 		return(error);
966 	}
967 
968 	/* Create DMA maps for RX buffers */
969 
970 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
971 		error = bus_dmamap_create(sc->re_ldata.re_mtag, 0,
972 			    &sc->re_ldata.re_rx_dmamap[i]);
973 		if (error) {
974 			device_printf(dev, "can't create DMA map for RX\n");
975 			return(ENOMEM);
976 		}
977 	}
978 
979 	return(0);
980 }
981 
982 /*
983  * Attach the interface. Allocate softc structures, do ifmedia
984  * setup and ethernet/BPF attach.
985  */
986 static int
987 re_attach(device_t dev)
988 {
989 	struct re_softc	*sc = device_get_softc(dev);
990 	struct ifnet *ifp;
991 	struct re_hwrev *hw_rev;
992 	uint8_t eaddr[ETHER_ADDR_LEN];
993 	int hwrev;
994 	u_int16_t re_did = 0;
995 	int error = 0, rid, i;
996 
997 	callout_init(&sc->re_timer);
998 
999 #ifndef BURN_BRIDGES
1000 	/*
1001 	 * Handle power management nonsense.
1002 	 */
1003 
1004 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D0) {
1005 		uint32_t membase, irq;
1006 
1007 		/* Save important PCI config data. */
1008 		membase = pci_read_config(dev, RE_PCI_LOMEM, 4);
1009 		irq = pci_read_config(dev, PCIR_INTLINE, 4);
1010 
1011 		/* Reset the power state. */
1012 		device_printf(dev, "chip is is in D%d power mode "
1013 		    "-- setting to D0\n", pci_get_powerstate(dev));
1014 
1015 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
1016 
1017 		/* Restore PCI config data. */
1018 		pci_write_config(dev, RE_PCI_LOMEM, membase, 4);
1019 		pci_write_config(dev, PCIR_INTLINE, irq, 4);
1020 	}
1021 #endif
1022 	/*
1023 	 * Map control/status registers.
1024 	 */
1025 	pci_enable_busmaster(dev);
1026 
1027 	rid = RE_PCI_LOIO;
1028 	sc->re_res = bus_alloc_resource_any(dev, SYS_RES_IOPORT, &rid,
1029 					    RF_ACTIVE);
1030 
1031 	if (sc->re_res == NULL) {
1032 		device_printf(dev, "couldn't map ports/memory\n");
1033 		error = ENXIO;
1034 		goto fail;
1035 	}
1036 
1037 	sc->re_btag = rman_get_bustag(sc->re_res);
1038 	sc->re_bhandle = rman_get_bushandle(sc->re_res);
1039 
1040 	/* Allocate interrupt */
1041 	rid = 0;
1042 	sc->re_irq = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
1043 					    RF_SHAREABLE | RF_ACTIVE);
1044 
1045 	if (sc->re_irq == NULL) {
1046 		device_printf(dev, "couldn't map interrupt\n");
1047 		error = ENXIO;
1048 		goto fail;
1049 	}
1050 
1051 	/* Reset the adapter. */
1052 	re_reset(sc);
1053 
1054 	hwrev = CSR_READ_4(sc, RE_TXCFG) & RE_TXCFG_HWREV;
1055 	for (hw_rev = re_hwrevs; hw_rev->re_desc != NULL; hw_rev++) {
1056 		if (hw_rev->re_rev == hwrev) {
1057 			sc->re_type = hw_rev->re_type;
1058 			break;
1059 		}
1060 	}
1061 
1062 	if (sc->re_type == RE_8169) {
1063 		/* Set RX length mask */
1064 		sc->re_rxlenmask = RE_RDESC_STAT_GFRAGLEN;
1065 
1066 		/* Force station address autoload from the EEPROM */
1067 		CSR_WRITE_1(sc, RE_EECMD, RE_EEMODE_AUTOLOAD);
1068 		for (i = 0; i < RE_TIMEOUT; i++) {
1069 			if ((CSR_READ_1(sc, RE_EECMD) & RE_EEMODE_AUTOLOAD) == 0)
1070 				break;
1071 			DELAY(100);
1072 		}
1073 		if (i == RE_TIMEOUT)
1074 			device_printf(dev, "eeprom autoload timed out\n");
1075 
1076 		for (i = 0; i < ETHER_ADDR_LEN; i++)
1077 			eaddr[i] = CSR_READ_1(sc, RE_IDR0 + i);
1078 	} else {
1079 		uint16_t as[3];
1080 
1081 		/* Set RX length mask */
1082 		sc->re_rxlenmask = RE_RDESC_STAT_FRAGLEN;
1083 
1084 		sc->re_eecmd_read = RE_EECMD_READ_6BIT;
1085 		re_read_eeprom(sc, (caddr_t)&re_did, 0, 1, 0);
1086 		if (re_did != 0x8129)
1087 			sc->re_eecmd_read = RE_EECMD_READ_8BIT;
1088 
1089 		/*
1090 		 * Get station address from the EEPROM.
1091 		 */
1092 		re_read_eeprom(sc, (caddr_t)as, RE_EE_EADDR, 3, 0);
1093 		for (i = 0; i < 3; i++) {
1094 			eaddr[(i * 2) + 0] = as[i] & 0xff;
1095 			eaddr[(i * 2) + 1] = as[i] >> 8;
1096 		}
1097 	}
1098 
1099 	/*
1100 	 * Allocate the parent bus DMA tag appropriate for PCI.
1101 	 */
1102 #define RE_NSEG_NEW 32
1103 	error = bus_dma_tag_create(NULL,	/* parent */
1104 			1, 0,			/* alignment, boundary */
1105 			BUS_SPACE_MAXADDR_32BIT,/* lowaddr */
1106 			BUS_SPACE_MAXADDR,	/* highaddr */
1107 			NULL, NULL,		/* filter, filterarg */
1108 			MAXBSIZE, RE_NSEG_NEW,	/* maxsize, nsegments */
1109 			BUS_SPACE_MAXSIZE_32BIT,/* maxsegsize */
1110 			BUS_DMA_ALLOCNOW,	/* flags */
1111 			&sc->re_parent_tag);
1112 	if (error)
1113 		goto fail;
1114 
1115 	error = re_allocmem(dev, sc);
1116 
1117 	if (error)
1118 		goto fail;
1119 
1120 	/* Do MII setup */
1121 	if (mii_phy_probe(dev, &sc->re_miibus,
1122 	    re_ifmedia_upd, re_ifmedia_sts)) {
1123 		device_printf(dev, "MII without any phy!\n");
1124 		error = ENXIO;
1125 		goto fail;
1126 	}
1127 
1128 	ifp = &sc->arpcom.ac_if;
1129 	ifp->if_softc = sc;
1130 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1131 	ifp->if_mtu = ETHERMTU;
1132 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1133 	ifp->if_ioctl = re_ioctl;
1134 	ifp->if_capabilities = IFCAP_VLAN_MTU;
1135 	ifp->if_start = re_start;
1136 	ifp->if_capabilities |= IFCAP_HWCSUM|IFCAP_VLAN_HWTAGGING;
1137 #ifdef DEVICE_POLLING
1138 	ifp->if_poll = re_poll;
1139 #endif
1140 	ifp->if_watchdog = re_watchdog;
1141 	ifp->if_init = re_init;
1142 	if (sc->re_type == RE_8169)
1143 		ifp->if_baudrate = 1000000000;
1144 	else
1145 		ifp->if_baudrate = 100000000;
1146 	ifq_set_maxlen(&ifp->if_snd, RE_IFQ_MAXLEN);
1147 	ifq_set_ready(&ifp->if_snd);
1148 #ifdef RE_DISABLE_HWCSUM
1149 	ifp->if_capenable = ifp->if_capabilities & ~IFCAP_HWCSUM;
1150 	ifp->if_hwassist = 0;
1151 #else
1152 	ifp->if_capenable = ifp->if_capabilities;
1153 	ifp->if_hwassist = RE_CSUM_FEATURES;
1154 #endif
1155 
1156 	/*
1157 	 * Call MI attach routine.
1158 	 */
1159 	ether_ifattach(ifp, eaddr);
1160 
1161 	/* Perform hardware diagnostic. */
1162 	error = re_diag(sc);
1163 
1164 	if (error) {
1165 		device_printf(dev, "hardware diagnostic failure\n");
1166 		ether_ifdetach(ifp);
1167 		goto fail;
1168 	}
1169 
1170 	/* Hook interrupt last to avoid having to lock softc */
1171 	error = bus_setup_intr(dev, sc->re_irq, INTR_TYPE_NET, re_intr, sc,
1172 			       &sc->re_intrhand, NULL);
1173 
1174 	if (error) {
1175 		device_printf(dev, "couldn't set up irq\n");
1176 		ether_ifdetach(ifp);
1177 		goto fail;
1178 	}
1179 
1180 fail:
1181 	if (error)
1182 		re_detach(dev);
1183 
1184 	return (error);
1185 }
1186 
1187 /*
1188  * Shutdown hardware and free up resources. This can be called any
1189  * time after the mutex has been initialized. It is called in both
1190  * the error case in attach and the normal detach case so it needs
1191  * to be careful about only freeing resources that have actually been
1192  * allocated.
1193  */
1194 static int
1195 re_detach(device_t dev)
1196 {
1197 	struct re_softc *sc = device_get_softc(dev);
1198 	struct ifnet *ifp = &sc->arpcom.ac_if;
1199 	int i, s;
1200 
1201 	s = splimp();
1202 
1203 	/* These should only be active if attach succeeded */
1204 	if (device_is_attached(dev)) {
1205 		re_stop(sc);
1206 		ether_ifdetach(ifp);
1207 	}
1208 	if (sc->re_miibus)
1209 		device_delete_child(dev, sc->re_miibus);
1210 	bus_generic_detach(dev);
1211 
1212 	if (sc->re_intrhand)
1213 		bus_teardown_intr(dev, sc->re_irq, sc->re_intrhand);
1214 	if (sc->re_irq)
1215 		bus_release_resource(dev, SYS_RES_IRQ, 0, sc->re_irq);
1216 	if (sc->re_res)
1217 		bus_release_resource(dev, SYS_RES_IOPORT, RE_PCI_LOIO,
1218 				     sc->re_res);
1219 
1220 	/* Unload and free the RX DMA ring memory and map */
1221 
1222 	if (sc->re_ldata.re_rx_list_tag) {
1223 		bus_dmamap_unload(sc->re_ldata.re_rx_list_tag,
1224 		    sc->re_ldata.re_rx_list_map);
1225 		bus_dmamem_free(sc->re_ldata.re_rx_list_tag,
1226 		    sc->re_ldata.re_rx_list,
1227 		    sc->re_ldata.re_rx_list_map);
1228 		bus_dma_tag_destroy(sc->re_ldata.re_rx_list_tag);
1229 	}
1230 
1231 	/* Unload and free the TX DMA ring memory and map */
1232 
1233 	if (sc->re_ldata.re_tx_list_tag) {
1234 		bus_dmamap_unload(sc->re_ldata.re_tx_list_tag,
1235 		    sc->re_ldata.re_tx_list_map);
1236 		bus_dmamem_free(sc->re_ldata.re_tx_list_tag,
1237 		    sc->re_ldata.re_tx_list,
1238 		    sc->re_ldata.re_tx_list_map);
1239 		bus_dma_tag_destroy(sc->re_ldata.re_tx_list_tag);
1240 	}
1241 
1242 	/* Destroy all the RX and TX buffer maps */
1243 
1244 	if (sc->re_ldata.re_mtag) {
1245 		for (i = 0; i < RE_TX_DESC_CNT; i++)
1246 			bus_dmamap_destroy(sc->re_ldata.re_mtag,
1247 			    sc->re_ldata.re_tx_dmamap[i]);
1248 		for (i = 0; i < RE_RX_DESC_CNT; i++)
1249 			bus_dmamap_destroy(sc->re_ldata.re_mtag,
1250 			    sc->re_ldata.re_rx_dmamap[i]);
1251 		bus_dma_tag_destroy(sc->re_ldata.re_mtag);
1252 	}
1253 
1254 	/* Unload and free the stats buffer and map */
1255 
1256 	if (sc->re_ldata.re_stag) {
1257 		bus_dmamap_unload(sc->re_ldata.re_stag,
1258 		    sc->re_ldata.re_rx_list_map);
1259 		bus_dmamem_free(sc->re_ldata.re_stag,
1260 		    sc->re_ldata.re_stats,
1261 		    sc->re_ldata.re_smap);
1262 		bus_dma_tag_destroy(sc->re_ldata.re_stag);
1263 	}
1264 
1265 	if (sc->re_parent_tag)
1266 		bus_dma_tag_destroy(sc->re_parent_tag);
1267 
1268 	splx(s);
1269 
1270 	return(0);
1271 }
1272 
1273 static int
1274 re_newbuf(struct re_softc *sc, int idx, struct mbuf *m)
1275 {
1276 	struct re_dmaload_arg arg;
1277 	struct mbuf *n = NULL;
1278 	int error;
1279 
1280 	if (m == NULL) {
1281 		n = m_getcl(MB_DONTWAIT, MT_DATA, M_PKTHDR);
1282 		if (n == NULL)
1283 			return(ENOBUFS);
1284 		m = n;
1285 	} else
1286 		m->m_data = m->m_ext.ext_buf;
1287 
1288 	/*
1289 	 * Initialize mbuf length fields and fixup
1290 	 * alignment so that the frame payload is
1291 	 * longword aligned.
1292 	 */
1293 	m->m_len = m->m_pkthdr.len = MCLBYTES;
1294 	m_adj(m, ETHER_ALIGN);
1295 
1296 	arg.sc = sc;
1297 	arg.re_idx = idx;
1298 	arg.re_maxsegs = 1;
1299 	arg.re_flags = 0;
1300 	arg.re_ring = sc->re_ldata.re_rx_list;
1301 
1302         error = bus_dmamap_load_mbuf(sc->re_ldata.re_mtag,
1303 	    sc->re_ldata.re_rx_dmamap[idx], m, re_dma_map_desc,
1304 	    &arg, BUS_DMA_NOWAIT);
1305 	if (error || arg.re_maxsegs != 1) {
1306 		if (n != NULL)
1307 			m_freem(n);
1308 		return (ENOMEM);
1309 	}
1310 
1311 	sc->re_ldata.re_rx_list[idx].re_cmdstat |= htole32(RE_RDESC_CMD_OWN);
1312 	sc->re_ldata.re_rx_mbuf[idx] = m;
1313 
1314         bus_dmamap_sync(sc->re_ldata.re_mtag, sc->re_ldata.re_rx_dmamap[idx],
1315 		        BUS_DMASYNC_PREREAD);
1316 
1317 	return(0);
1318 }
1319 
1320 static int
1321 re_tx_list_init(struct re_softc *sc)
1322 {
1323 	bzero(sc->re_ldata.re_tx_list, RE_TX_LIST_SZ);
1324 	bzero(&sc->re_ldata.re_tx_mbuf, RE_TX_DESC_CNT * sizeof(struct mbuf *));
1325 
1326 	bus_dmamap_sync(sc->re_ldata.re_tx_list_tag,
1327 			sc->re_ldata.re_tx_list_map, BUS_DMASYNC_PREWRITE);
1328 	sc->re_ldata.re_tx_prodidx = 0;
1329 	sc->re_ldata.re_tx_considx = 0;
1330 	sc->re_ldata.re_tx_free = RE_TX_DESC_CNT;
1331 
1332 	return(0);
1333 }
1334 
1335 static int
1336 re_rx_list_init(struct re_softc *sc)
1337 {
1338 	int i, error;
1339 
1340 	bzero(sc->re_ldata.re_rx_list, RE_RX_LIST_SZ);
1341 	bzero(&sc->re_ldata.re_rx_mbuf, RE_RX_DESC_CNT * sizeof(struct mbuf *));
1342 
1343 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
1344 		error = re_newbuf(sc, i, NULL);
1345 		if (error)
1346 			return(error);
1347 	}
1348 
1349 	/* Flush the RX descriptors */
1350 
1351 	bus_dmamap_sync(sc->re_ldata.re_rx_list_tag,
1352 	    sc->re_ldata.re_rx_list_map,
1353 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
1354 
1355 	sc->re_ldata.re_rx_prodidx = 0;
1356 	sc->re_head = sc->re_tail = NULL;
1357 
1358 	return(0);
1359 }
1360 
1361 /*
1362  * RX handler for C+ and 8169. For the gigE chips, we support
1363  * the reception of jumbo frames that have been fragmented
1364  * across multiple 2K mbuf cluster buffers.
1365  */
1366 static void
1367 re_rxeof(struct re_softc *sc)
1368 {
1369 	struct ifnet *ifp = &sc->arpcom.ac_if;
1370 	struct mbuf *m;
1371 	struct re_desc 	*cur_rx;
1372 	uint32_t rxstat, rxvlan;
1373 	int i, total_len;
1374 
1375 	/* Invalidate the descriptor memory */
1376 
1377 	bus_dmamap_sync(sc->re_ldata.re_rx_list_tag,
1378 			sc->re_ldata.re_rx_list_map, BUS_DMASYNC_POSTREAD);
1379 
1380 	for (i = sc->re_ldata.re_rx_prodidx;
1381 	     RE_OWN(&sc->re_ldata.re_rx_list[i]) == 0 ; RE_DESC_INC(i)) {
1382 		cur_rx = &sc->re_ldata.re_rx_list[i];
1383 		m = sc->re_ldata.re_rx_mbuf[i];
1384 		total_len = RE_RXBYTES(cur_rx);
1385 		rxstat = le32toh(cur_rx->re_cmdstat);
1386 		rxvlan = le32toh(cur_rx->re_vlanctl);
1387 
1388 		/* Invalidate the RX mbuf and unload its map */
1389 
1390 		bus_dmamap_sync(sc->re_ldata.re_mtag,
1391 				sc->re_ldata.re_rx_dmamap[i],
1392 				BUS_DMASYNC_POSTWRITE);
1393 		bus_dmamap_unload(sc->re_ldata.re_mtag,
1394 				  sc->re_ldata.re_rx_dmamap[i]);
1395 
1396 		if ((rxstat & RE_RDESC_STAT_EOF) == 0) {
1397 			m->m_len = MCLBYTES - ETHER_ALIGN;
1398 			if (sc->re_head == NULL) {
1399 				sc->re_head = sc->re_tail = m;
1400 			} else {
1401 				m->m_flags &= ~M_PKTHDR;
1402 				sc->re_tail->m_next = m;
1403 				sc->re_tail = m;
1404 			}
1405 			re_newbuf(sc, i, NULL);
1406 			continue;
1407 		}
1408 
1409 		/*
1410 		 * NOTE: for the 8139C+, the frame length field
1411 		 * is always 12 bits in size, but for the gigE chips,
1412 		 * it is 13 bits (since the max RX frame length is 16K).
1413 		 * Unfortunately, all 32 bits in the status word
1414 		 * were already used, so to make room for the extra
1415 		 * length bit, RealTek took out the 'frame alignment
1416 		 * error' bit and shifted the other status bits
1417 		 * over one slot. The OWN, EOR, FS and LS bits are
1418 		 * still in the same places. We have already extracted
1419 		 * the frame length and checked the OWN bit, so rather
1420 		 * than using an alternate bit mapping, we shift the
1421 		 * status bits one space to the right so we can evaluate
1422 		 * them using the 8169 status as though it was in the
1423 		 * same format as that of the 8139C+.
1424 		 */
1425 		if (sc->re_type == RE_8169)
1426 			rxstat >>= 1;
1427 
1428 		if (rxstat & RE_RDESC_STAT_RXERRSUM) {
1429 			ifp->if_ierrors++;
1430 			/*
1431 			 * If this is part of a multi-fragment packet,
1432 			 * discard all the pieces.
1433 			 */
1434 			if (sc->re_head != NULL) {
1435 				m_freem(sc->re_head);
1436 				sc->re_head = sc->re_tail = NULL;
1437 			}
1438 			re_newbuf(sc, i, m);
1439 			continue;
1440 		}
1441 
1442 		/*
1443 		 * If allocating a replacement mbuf fails,
1444 		 * reload the current one.
1445 		 */
1446 
1447 		if (re_newbuf(sc, i, NULL)) {
1448 			ifp->if_ierrors++;
1449 			if (sc->re_head != NULL) {
1450 				m_freem(sc->re_head);
1451 				sc->re_head = sc->re_tail = NULL;
1452 			}
1453 			re_newbuf(sc, i, m);
1454 			continue;
1455 		}
1456 
1457 		if (sc->re_head != NULL) {
1458 			m->m_len = total_len % (MCLBYTES - ETHER_ALIGN);
1459 			/*
1460 			 * Special case: if there's 4 bytes or less
1461 			 * in this buffer, the mbuf can be discarded:
1462 			 * the last 4 bytes is the CRC, which we don't
1463 			 * care about anyway.
1464 			 */
1465 			if (m->m_len <= ETHER_CRC_LEN) {
1466 				sc->re_tail->m_len -=
1467 				    (ETHER_CRC_LEN - m->m_len);
1468 				m_freem(m);
1469 			} else {
1470 				m->m_len -= ETHER_CRC_LEN;
1471 				m->m_flags &= ~M_PKTHDR;
1472 				sc->re_tail->m_next = m;
1473 			}
1474 			m = sc->re_head;
1475 			sc->re_head = sc->re_tail = NULL;
1476 			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
1477 		} else
1478 			m->m_pkthdr.len = m->m_len =
1479 			    (total_len - ETHER_CRC_LEN);
1480 
1481 		ifp->if_ipackets++;
1482 		m->m_pkthdr.rcvif = ifp;
1483 
1484 		/* Do RX checksumming if enabled */
1485 
1486 		if (ifp->if_capenable & IFCAP_RXCSUM) {
1487 
1488 			/* Check IP header checksum */
1489 			if (rxstat & RE_RDESC_STAT_PROTOID)
1490 				m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED;
1491 			if ((rxstat & RE_RDESC_STAT_IPSUMBAD) == 0)
1492 				m->m_pkthdr.csum_flags |= CSUM_IP_VALID;
1493 
1494 			/* Check TCP/UDP checksum */
1495 			if ((RE_TCPPKT(rxstat) &&
1496 			    (rxstat & RE_RDESC_STAT_TCPSUMBAD) == 0) ||
1497 			    (RE_UDPPKT(rxstat) &&
1498 			    (rxstat & RE_RDESC_STAT_UDPSUMBAD)) == 0) {
1499 				m->m_pkthdr.csum_flags |=
1500 				    CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
1501 				m->m_pkthdr.csum_data = 0xffff;
1502 			}
1503 		}
1504 
1505 		if (rxvlan & RE_RDESC_VLANCTL_TAG)
1506 			VLAN_INPUT_TAG(m,
1507 			   be16toh((rxvlan & RE_RDESC_VLANCTL_DATA)));
1508 		else
1509 			(*ifp->if_input)(ifp, m);
1510 	}
1511 
1512 	/* Flush the RX DMA ring */
1513 
1514 	bus_dmamap_sync(sc->re_ldata.re_rx_list_tag,
1515 			sc->re_ldata.re_rx_list_map,
1516 			BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
1517 
1518 	sc->re_ldata.re_rx_prodidx = i;
1519 }
1520 
1521 static void
1522 re_txeof(struct re_softc *sc)
1523 {
1524 	struct ifnet *ifp = &sc->arpcom.ac_if;
1525 	uint32_t txstat;
1526 	int idx;
1527 
1528 	/* Invalidate the TX descriptor list */
1529 
1530 	bus_dmamap_sync(sc->re_ldata.re_tx_list_tag,
1531 	    sc->re_ldata.re_tx_list_map,
1532 	    BUS_DMASYNC_POSTREAD);
1533 
1534 	for (idx = sc->re_ldata.re_tx_considx;
1535 	     idx != sc->re_ldata.re_tx_prodidx; RE_DESC_INC(idx)) {
1536 		txstat = le32toh(sc->re_ldata.re_tx_list[idx].re_cmdstat);
1537 		if (txstat & RE_TDESC_CMD_OWN)
1538 			break;
1539 
1540 		/*
1541 		 * We only stash mbufs in the last descriptor
1542 		 * in a fragment chain, which also happens to
1543 		 * be the only place where the TX status bits
1544 		 * are valid.
1545 		 */
1546 		if (txstat & RE_TDESC_CMD_EOF) {
1547 			m_freem(sc->re_ldata.re_tx_mbuf[idx]);
1548 			sc->re_ldata.re_tx_mbuf[idx] = NULL;
1549 			bus_dmamap_unload(sc->re_ldata.re_mtag,
1550 			    sc->re_ldata.re_tx_dmamap[idx]);
1551 			if (txstat & (RE_TDESC_STAT_EXCESSCOL|
1552 			    RE_TDESC_STAT_COLCNT))
1553 				ifp->if_collisions++;
1554 			if (txstat & RE_TDESC_STAT_TXERRSUM)
1555 				ifp->if_oerrors++;
1556 			else
1557 				ifp->if_opackets++;
1558 		}
1559 		sc->re_ldata.re_tx_free++;
1560 	}
1561 
1562 	/* No changes made to the TX ring, so no flush needed */
1563 	if (idx != sc->re_ldata.re_tx_considx) {
1564 		sc->re_ldata.re_tx_considx = idx;
1565 		ifp->if_flags &= ~IFF_OACTIVE;
1566 		ifp->if_timer = 0;
1567 	}
1568 
1569 	/*
1570 	 * If not all descriptors have been released reaped yet,
1571 	 * reload the timer so that we will eventually get another
1572 	 * interrupt that will cause us to re-enter this routine.
1573 	 * This is done in case the transmitter has gone idle.
1574 	 */
1575 	if (sc->re_ldata.re_tx_free != RE_TX_DESC_CNT)
1576                 CSR_WRITE_4(sc, RE_TIMERCNT, 1);
1577 }
1578 
1579 static void
1580 re_tick(void *xsc)
1581 {
1582 	struct re_softc *sc = xsc;
1583 	struct mii_data *mii;
1584 	int s;
1585 
1586 	s = splimp();
1587 
1588 	mii = device_get_softc(sc->re_miibus);
1589 	mii_tick(mii);
1590 
1591 	callout_reset(&sc->re_timer, hz, re_tick, sc);
1592 	splx(s);
1593 }
1594 
1595 #ifdef DEVICE_POLLING
1596 
1597 static void
1598 re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
1599 {
1600 	struct re_softc *sc = ifp->if_softc;
1601 
1602 	switch(cmd) {
1603 	case POLL_REGISTER:
1604 		/* disable interrupts */
1605 		CSR_WRITE_2(sc, RE_IMR, 0x0000);
1606 		break;
1607 	case POLL_DEREGISTER:
1608 		/* enable interrupts */
1609 		CSR_WRITE_2(sc, RE_IMR, RE_INTRS_CPLUS);
1610 		break;
1611 	default:
1612 		sc->rxcycles = count;
1613 		re_rxeof(sc);
1614 		re_txeof(sc);
1615 
1616 		if (!ifq_is_empty(&ifp->if_snd))
1617 			(*ifp->if_start)(ifp);
1618 
1619 		if (cmd == POLL_AND_CHECK_STATUS) { /* also check status register */
1620 			uint16_t       status;
1621 
1622 			status = CSR_READ_2(sc, RE_ISR);
1623 			if (status == 0xffff)
1624 				return;
1625 			if (status)
1626 				CSR_WRITE_2(sc, RE_ISR, status);
1627 
1628 			/*
1629 			 * XXX check behaviour on receiver stalls.
1630 			 */
1631 
1632 			if (status & RE_ISR_SYSTEM_ERR) {
1633 				re_reset(sc);
1634 				re_init(sc);
1635 			}
1636 		}
1637 		break;
1638 	}
1639 }
1640 #endif /* DEVICE_POLLING */
1641 
1642 static void
1643 re_intr(void *arg)
1644 {
1645 	struct re_softc	*sc = arg;
1646 	struct ifnet *ifp = &sc->arpcom.ac_if;
1647 	uint16_t status;
1648 	int s;
1649 
1650 	if (sc->suspended || (ifp->if_flags & IFF_UP) == 0)
1651 		return;
1652 
1653 	s = splimp();
1654 
1655 	for (;;) {
1656 		status = CSR_READ_2(sc, RE_ISR);
1657 		/* If the card has gone away the read returns 0xffff. */
1658 		if (status == 0xffff)
1659 			break;
1660 		if (status)
1661 			CSR_WRITE_2(sc, RE_ISR, status);
1662 
1663 		if ((status & RE_INTRS_CPLUS) == 0)
1664 			break;
1665 
1666 		if (status & RE_ISR_RX_OK)
1667 			re_rxeof(sc);
1668 
1669 		if (status & RE_ISR_RX_ERR)
1670 			re_rxeof(sc);
1671 
1672 		if ((status & RE_ISR_TIMEOUT_EXPIRED) ||
1673 		    (status & RE_ISR_TX_ERR) ||
1674 		    (status & RE_ISR_TX_DESC_UNAVAIL))
1675 			re_txeof(sc);
1676 
1677 		if (status & RE_ISR_SYSTEM_ERR) {
1678 			re_reset(sc);
1679 			re_init(sc);
1680 		}
1681 
1682 		if (status & RE_ISR_LINKCHG)
1683 			re_tick(sc);
1684 	}
1685 
1686 	if (!ifq_is_empty(&ifp->if_snd))
1687 		(*ifp->if_start)(ifp);
1688 
1689 	splx(s);
1690 }
1691 
1692 static int
1693 re_encap(struct re_softc *sc, struct mbuf **m_head, int *idx, int *called_defrag)
1694 {
1695 	struct ifnet *ifp = &sc->arpcom.ac_if;
1696 	struct mbuf *m, *m_new = NULL;
1697 	struct re_dmaload_arg	arg;
1698 	bus_dmamap_t		map;
1699 	int			error;
1700 
1701 	*called_defrag = 0;
1702 	if (sc->re_ldata.re_tx_free <= 4)
1703 		return(EFBIG);
1704 
1705 	m = *m_head;
1706 
1707 	/*
1708 	 * Set up checksum offload. Note: checksum offload bits must
1709 	 * appear in all descriptors of a multi-descriptor transmit
1710 	 * attempt. (This is according to testing done with an 8169
1711 	 * chip. I'm not sure if this is a requirement or a bug.)
1712 	 */
1713 
1714 	arg.re_flags = 0;
1715 
1716 	if (m->m_pkthdr.csum_flags & CSUM_IP)
1717 		arg.re_flags |= RE_TDESC_CMD_IPCSUM;
1718 	if (m->m_pkthdr.csum_flags & CSUM_TCP)
1719 		arg.re_flags |= RE_TDESC_CMD_TCPCSUM;
1720 	if (m->m_pkthdr.csum_flags & CSUM_UDP)
1721 		arg.re_flags |= RE_TDESC_CMD_UDPCSUM;
1722 
1723 	arg.sc = sc;
1724 	arg.re_idx = *idx;
1725 	arg.re_maxsegs = sc->re_ldata.re_tx_free;
1726 	if (arg.re_maxsegs > 4)
1727 		arg.re_maxsegs -= 4;
1728 	arg.re_ring = sc->re_ldata.re_tx_list;
1729 
1730 	map = sc->re_ldata.re_tx_dmamap[*idx];
1731 	error = bus_dmamap_load_mbuf(sc->re_ldata.re_mtag, map,
1732 	    m, re_dma_map_desc, &arg, BUS_DMA_NOWAIT);
1733 
1734 	if (error && error != EFBIG) {
1735 		if_printf(ifp, "can't map mbuf (error %d)\n", error);
1736 		return(ENOBUFS);
1737 	}
1738 
1739 	/* Too many segments to map, coalesce into a single mbuf */
1740 
1741 	if (error || arg.re_maxsegs == 0) {
1742 		m_new = m_defrag_nofree(m, MB_DONTWAIT);
1743 		if (m_new == NULL)
1744 			return(1);
1745 		else {
1746 			m = m_new;
1747 			*m_head = m;
1748 		}
1749 
1750 		*called_defrag = 1;
1751 		arg.sc = sc;
1752 		arg.re_idx = *idx;
1753 		arg.re_maxsegs = sc->re_ldata.re_tx_free;
1754 		arg.re_ring = sc->re_ldata.re_tx_list;
1755 
1756 		error = bus_dmamap_load_mbuf(sc->re_ldata.re_mtag, map,
1757 		    m, re_dma_map_desc, &arg, BUS_DMA_NOWAIT);
1758 		if (error) {
1759 			m_freem(m);
1760 			if_printf(ifp, "can't map mbuf (error %d)\n", error);
1761 			return(EFBIG);
1762 		}
1763 	}
1764 
1765 	/*
1766 	 * Insure that the map for this transmission
1767 	 * is placed at the array index of the last descriptor
1768 	 * in this chain.
1769 	 */
1770 	sc->re_ldata.re_tx_dmamap[*idx] =
1771 	    sc->re_ldata.re_tx_dmamap[arg.re_idx];
1772 	sc->re_ldata.re_tx_dmamap[arg.re_idx] = map;
1773 
1774 	sc->re_ldata.re_tx_mbuf[arg.re_idx] = m;
1775 	sc->re_ldata.re_tx_free -= arg.re_maxsegs;
1776 
1777 	/*
1778 	 * Set up hardware VLAN tagging. Note: vlan tag info must
1779 	 * appear in the first descriptor of a multi-descriptor
1780 	 * transmission attempt.
1781 	 */
1782 
1783 	if ((m->m_flags & (M_PROTO1|M_PKTHDR)) == (M_PROTO1|M_PKTHDR) &&
1784 	    m->m_pkthdr.rcvif != NULL &&
1785 	    m->m_pkthdr.rcvif->if_type == IFT_L2VLAN) {
1786 	    	struct ifvlan *ifv;
1787 		ifv = m->m_pkthdr.rcvif->if_softc;
1788 		if (ifv != NULL)
1789 			sc->re_ldata.re_tx_list[*idx].re_vlanctl =
1790 			    htole32(htobe16(ifv->ifv_tag) | RE_TDESC_VLANCTL_TAG);
1791 	}
1792 
1793 	/* Transfer ownership of packet to the chip. */
1794 
1795 	sc->re_ldata.re_tx_list[arg.re_idx].re_cmdstat |=
1796 	    htole32(RE_TDESC_CMD_OWN);
1797 	if (*idx != arg.re_idx)
1798 		sc->re_ldata.re_tx_list[*idx].re_cmdstat |=
1799 		    htole32(RE_TDESC_CMD_OWN);
1800 
1801 	RE_DESC_INC(arg.re_idx);
1802 	*idx = arg.re_idx;
1803 
1804 	return(0);
1805 }
1806 
1807 /*
1808  * Main transmit routine for C+ and gigE NICs.
1809  */
1810 
1811 static void
1812 re_start(struct ifnet *ifp)
1813 {
1814 	struct re_softc	*sc = ifp->if_softc;
1815 	struct mbuf *m_head = NULL, *m_head2;
1816 	int called_defrag, idx, s;
1817 
1818 	s = splimp();
1819 
1820 	idx = sc->re_ldata.re_tx_prodidx;
1821 
1822 	while (sc->re_ldata.re_tx_mbuf[idx] == NULL) {
1823 		m_head = ifq_poll(&ifp->if_snd);
1824 		if (m_head == NULL)
1825 			break;
1826 
1827 		if (re_encap(sc, &m_head, &idx, &called_defrag)) {
1828 			if (called_defrag) {
1829 				m_head2 = ifq_dequeue(&ifp->if_snd);
1830 				m_freem(m_head2);
1831 			}
1832 			ifp->if_flags |= IFF_OACTIVE;
1833 			break;
1834 		}
1835 
1836 		m_head2 = ifq_dequeue(&ifp->if_snd);
1837 		if (called_defrag)
1838 			m_freem(m_head2);
1839 
1840 		/*
1841 		 * If there's a BPF listener, bounce a copy of this frame
1842 		 * to him.
1843 		 */
1844 		BPF_MTAP(ifp, m_head);
1845 	}
1846 
1847 	/* Flush the TX descriptors */
1848 	bus_dmamap_sync(sc->re_ldata.re_tx_list_tag,
1849 			sc->re_ldata.re_tx_list_map,
1850 			BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
1851 
1852 	sc->re_ldata.re_tx_prodidx = idx;
1853 
1854 	/*
1855 	 * RealTek put the TX poll request register in a different
1856 	 * location on the 8169 gigE chip. I don't know why.
1857 	 */
1858 	if (sc->re_type == RE_8169)
1859 		CSR_WRITE_2(sc, RE_GTXSTART, RE_TXSTART_START);
1860 	else
1861 		CSR_WRITE_2(sc, RE_TXSTART, RE_TXSTART_START);
1862 
1863 	/*
1864 	 * Use the countdown timer for interrupt moderation.
1865 	 * 'TX done' interrupts are disabled. Instead, we reset the
1866 	 * countdown timer, which will begin counting until it hits
1867 	 * the value in the TIMERINT register, and then trigger an
1868 	 * interrupt. Each time we write to the TIMERCNT register,
1869 	 * the timer count is reset to 0.
1870 	 */
1871 	CSR_WRITE_4(sc, RE_TIMERCNT, 1);
1872 
1873 	splx(s);
1874 
1875 	/*
1876 	 * Set a timeout in case the chip goes out to lunch.
1877 	 */
1878 	ifp->if_timer = 5;
1879 }
1880 
1881 static void
1882 re_init(void *xsc)
1883 {
1884 	struct re_softc *sc = xsc;
1885 	struct ifnet *ifp = &sc->arpcom.ac_if;
1886 	struct mii_data *mii;
1887 	uint32_t rxcfg = 0;
1888 	int s;
1889 
1890 	s = splimp();
1891 	mii = device_get_softc(sc->re_miibus);
1892 
1893 	/*
1894 	 * Cancel pending I/O and free all RX/TX buffers.
1895 	 */
1896 	re_stop(sc);
1897 
1898 	/*
1899 	 * Enable C+ RX and TX mode, as well as VLAN stripping and
1900 	 * RX checksum offload. We must configure the C+ register
1901 	 * before all others.
1902 	 */
1903 	CSR_WRITE_2(sc, RE_CPLUS_CMD, RE_CPLUSCMD_RXENB | RE_CPLUSCMD_TXENB |
1904 		    RE_CPLUSCMD_PCI_MRW | RE_CPLUSCMD_VLANSTRIP |
1905 		    (ifp->if_capenable & IFCAP_RXCSUM ?
1906 		     RE_CPLUSCMD_RXCSUM_ENB : 0));
1907 
1908 	/*
1909 	 * Init our MAC address.  Even though the chipset
1910 	 * documentation doesn't mention it, we need to enter "Config
1911 	 * register write enable" mode to modify the ID registers.
1912 	 */
1913 	CSR_WRITE_1(sc, RE_EECMD, RE_EEMODE_WRITECFG);
1914 	CSR_WRITE_STREAM_4(sc, RE_IDR0,
1915 	    *(u_int32_t *)(&sc->arpcom.ac_enaddr[0]));
1916 	CSR_WRITE_STREAM_4(sc, RE_IDR4,
1917 	    *(u_int32_t *)(&sc->arpcom.ac_enaddr[4]));
1918 	CSR_WRITE_1(sc, RE_EECMD, RE_EEMODE_OFF);
1919 
1920 	/*
1921 	 * For C+ mode, initialize the RX descriptors and mbufs.
1922 	 */
1923 	re_rx_list_init(sc);
1924 	re_tx_list_init(sc);
1925 
1926 	/*
1927 	 * Enable transmit and receive.
1928 	 */
1929 	CSR_WRITE_1(sc, RE_COMMAND, RE_CMD_TX_ENB|RE_CMD_RX_ENB);
1930 
1931 	/*
1932 	 * Set the initial TX and RX configuration.
1933 	 */
1934 	if (sc->re_testmode) {
1935 		if (sc->re_type == RE_8169)
1936 			CSR_WRITE_4(sc, RE_TXCFG,
1937 				    RE_TXCFG_CONFIG | RE_LOOPTEST_ON);
1938 		else
1939 			CSR_WRITE_4(sc, RE_TXCFG,
1940 				    RE_TXCFG_CONFIG | RE_LOOPTEST_ON_CPLUS);
1941 	} else
1942 		CSR_WRITE_4(sc, RE_TXCFG, RE_TXCFG_CONFIG);
1943 	CSR_WRITE_4(sc, RE_RXCFG, RE_RXCFG_CONFIG);
1944 
1945 	/* Set the individual bit to receive frames for this host only. */
1946 	rxcfg = CSR_READ_4(sc, RE_RXCFG);
1947 	rxcfg |= RE_RXCFG_RX_INDIV;
1948 
1949 	/* If we want promiscuous mode, set the allframes bit. */
1950 	if (ifp->if_flags & IFF_PROMISC) {
1951 		rxcfg |= RE_RXCFG_RX_ALLPHYS;
1952 		CSR_WRITE_4(sc, RE_RXCFG, rxcfg);
1953 	} else {
1954 		rxcfg &= ~RE_RXCFG_RX_ALLPHYS;
1955 		CSR_WRITE_4(sc, RE_RXCFG, rxcfg);
1956 	}
1957 
1958 	/*
1959 	 * Set capture broadcast bit to capture broadcast frames.
1960 	 */
1961 	if (ifp->if_flags & IFF_BROADCAST) {
1962 		rxcfg |= RE_RXCFG_RX_BROAD;
1963 		CSR_WRITE_4(sc, RE_RXCFG, rxcfg);
1964 	} else {
1965 		rxcfg &= ~RE_RXCFG_RX_BROAD;
1966 		CSR_WRITE_4(sc, RE_RXCFG, rxcfg);
1967 	}
1968 
1969 	/*
1970 	 * Program the multicast filter, if necessary.
1971 	 */
1972 	re_setmulti(sc);
1973 
1974 #ifdef DEVICE_POLLING
1975 	/*
1976 	 * Disable interrupts if we are polling.
1977 	 */
1978 	if (ifp->if_flags & IFF_POLLING)
1979 		CSR_WRITE_2(sc, RE_IMR, 0);
1980 	else	/* otherwise ... */
1981 #endif /* DEVICE_POLLING */
1982 	/*
1983 	 * Enable interrupts.
1984 	 */
1985 	if (sc->re_testmode)
1986 		CSR_WRITE_2(sc, RE_IMR, 0);
1987 	else
1988 		CSR_WRITE_2(sc, RE_IMR, RE_INTRS_CPLUS);
1989 
1990 	/* Set initial TX threshold */
1991 	sc->re_txthresh = RE_TX_THRESH_INIT;
1992 
1993 	/* Start RX/TX process. */
1994 	CSR_WRITE_4(sc, RE_MISSEDPKT, 0);
1995 #ifdef notdef
1996 	/* Enable receiver and transmitter. */
1997 	CSR_WRITE_1(sc, RE_COMMAND, RE_CMD_TX_ENB|RE_CMD_RX_ENB);
1998 #endif
1999 	/*
2000 	 * Load the addresses of the RX and TX lists into the chip.
2001 	 */
2002 
2003 	CSR_WRITE_4(sc, RE_RXLIST_ADDR_HI,
2004 	    RE_ADDR_HI(sc->re_ldata.re_rx_list_addr));
2005 	CSR_WRITE_4(sc, RE_RXLIST_ADDR_LO,
2006 	    RE_ADDR_LO(sc->re_ldata.re_rx_list_addr));
2007 
2008 	CSR_WRITE_4(sc, RE_TXLIST_ADDR_HI,
2009 	    RE_ADDR_HI(sc->re_ldata.re_tx_list_addr));
2010 	CSR_WRITE_4(sc, RE_TXLIST_ADDR_LO,
2011 	    RE_ADDR_LO(sc->re_ldata.re_tx_list_addr));
2012 
2013 	CSR_WRITE_1(sc, RE_EARLY_TX_THRESH, 16);
2014 
2015 	/*
2016 	 * Initialize the timer interrupt register so that
2017 	 * a timer interrupt will be generated once the timer
2018 	 * reaches a certain number of ticks. The timer is
2019 	 * reloaded on each transmit. This gives us TX interrupt
2020 	 * moderation, which dramatically improves TX frame rate.
2021 	 */
2022 
2023 	if (sc->re_type == RE_8169)
2024 		CSR_WRITE_4(sc, RE_TIMERINT_8169, 0x800);
2025 	else
2026 		CSR_WRITE_4(sc, RE_TIMERINT, 0x400);
2027 
2028 	/*
2029 	 * For 8169 gigE NICs, set the max allowed RX packet
2030 	 * size so we can receive jumbo frames.
2031 	 */
2032 	if (sc->re_type == RE_8169)
2033 		CSR_WRITE_2(sc, RE_MAXRXPKTLEN, 16383);
2034 
2035 	if (sc->re_testmode) {
2036 		splx(s);
2037 		return;
2038 	}
2039 
2040 	mii_mediachg(mii);
2041 
2042 	CSR_WRITE_1(sc, RE_CFG1, RE_CFG1_DRVLOAD|RE_CFG1_FULLDUPLEX);
2043 
2044 	ifp->if_flags |= IFF_RUNNING;
2045 	ifp->if_flags &= ~IFF_OACTIVE;
2046 
2047 	callout_reset(&sc->re_timer, hz, re_tick, sc);
2048 	splx(s);
2049 }
2050 
2051 /*
2052  * Set media options.
2053  */
2054 static int
2055 re_ifmedia_upd(struct ifnet *ifp)
2056 {
2057 	struct re_softc *sc = ifp->if_softc;
2058 	struct mii_data *mii;
2059 
2060 	mii = device_get_softc(sc->re_miibus);
2061 	mii_mediachg(mii);
2062 
2063 	return(0);
2064 }
2065 
2066 /*
2067  * Report current media status.
2068  */
2069 static void
2070 re_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
2071 {
2072 	struct re_softc *sc = ifp->if_softc;
2073 	struct mii_data *mii;
2074 
2075 	mii = device_get_softc(sc->re_miibus);
2076 
2077 	mii_pollstat(mii);
2078 	ifmr->ifm_active = mii->mii_media_active;
2079 	ifmr->ifm_status = mii->mii_media_status;
2080 }
2081 
2082 static int
2083 re_ioctl(struct ifnet *ifp, u_long command, caddr_t data, struct ucred *cr)
2084 {
2085 	struct re_softc *sc = ifp->if_softc;
2086 	struct ifreq *ifr = (struct ifreq *) data;
2087 	struct mii_data *mii;
2088 	int error = 0, s;
2089 
2090 	s = splimp();
2091 
2092 	switch(command) {
2093 	case SIOCSIFMTU:
2094 		if (ifr->ifr_mtu > RE_JUMBO_MTU)
2095 			error = EINVAL;
2096 		ifp->if_mtu = ifr->ifr_mtu;
2097 		break;
2098 	case SIOCSIFFLAGS:
2099 		if (ifp->if_flags & IFF_UP)
2100 			re_init(sc);
2101 		else if (ifp->if_flags & IFF_RUNNING)
2102 				re_stop(sc);
2103 		error = 0;
2104 		break;
2105 	case SIOCADDMULTI:
2106 	case SIOCDELMULTI:
2107 		re_setmulti(sc);
2108 		error = 0;
2109 		break;
2110 	case SIOCGIFMEDIA:
2111 	case SIOCSIFMEDIA:
2112 		mii = device_get_softc(sc->re_miibus);
2113 		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, command);
2114 		break;
2115 	case SIOCSIFCAP:
2116 		ifp->if_capenable &= ~(IFCAP_HWCSUM);
2117 		ifp->if_capenable |=
2118 		    ifr->ifr_reqcap & (IFCAP_HWCSUM);
2119 		if (ifp->if_capenable & IFCAP_TXCSUM)
2120 			ifp->if_hwassist = RE_CSUM_FEATURES;
2121 		else
2122 			ifp->if_hwassist = 0;
2123 		if (ifp->if_flags & IFF_RUNNING)
2124 			re_init(sc);
2125 		break;
2126 	default:
2127 		error = ether_ioctl(ifp, command, data);
2128 		break;
2129 	}
2130 
2131 	splx(s);
2132 
2133 	return(error);
2134 }
2135 
2136 static void
2137 re_watchdog(struct ifnet *ifp)
2138 {
2139 	struct re_softc *sc = ifp->if_softc;
2140 	int s;
2141 
2142 	s = splimp();
2143 	if_printf(ifp, "watchdog timeout\n");
2144 	ifp->if_oerrors++;
2145 
2146 	re_txeof(sc);
2147 	re_rxeof(sc);
2148 
2149 	re_init(sc);
2150 
2151 	splx(s);
2152 }
2153 
2154 /*
2155  * Stop the adapter and free any mbufs allocated to the
2156  * RX and TX lists.
2157  */
2158 static void
2159 re_stop(struct re_softc *sc)
2160 {
2161 	struct ifnet *ifp = &sc->arpcom.ac_if;
2162 	int i, s;
2163 
2164 	s = splimp();
2165 	ifp->if_timer = 0;
2166 	callout_stop(&sc->re_timer);
2167 
2168 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
2169 
2170 	CSR_WRITE_1(sc, RE_COMMAND, 0x00);
2171 	CSR_WRITE_2(sc, RE_IMR, 0x0000);
2172 
2173 	if (sc->re_head != NULL) {
2174 		m_freem(sc->re_head);
2175 		sc->re_head = sc->re_tail = NULL;
2176 	}
2177 
2178 	/* Free the TX list buffers. */
2179 	for (i = 0; i < RE_TX_DESC_CNT; i++) {
2180 		if (sc->re_ldata.re_tx_mbuf[i] != NULL) {
2181 			bus_dmamap_unload(sc->re_ldata.re_mtag,
2182 					  sc->re_ldata.re_tx_dmamap[i]);
2183 			m_freem(sc->re_ldata.re_tx_mbuf[i]);
2184 			sc->re_ldata.re_tx_mbuf[i] = NULL;
2185 		}
2186 	}
2187 
2188 	/* Free the RX list buffers. */
2189 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
2190 		if (sc->re_ldata.re_rx_mbuf[i] != NULL) {
2191 			bus_dmamap_unload(sc->re_ldata.re_mtag,
2192 					  sc->re_ldata.re_rx_dmamap[i]);
2193 			m_freem(sc->re_ldata.re_rx_mbuf[i]);
2194 			sc->re_ldata.re_rx_mbuf[i] = NULL;
2195 		}
2196 	}
2197 
2198 	splx(s);
2199 }
2200 
2201 /*
2202  * Device suspend routine.  Stop the interface and save some PCI
2203  * settings in case the BIOS doesn't restore them properly on
2204  * resume.
2205  */
2206 static int
2207 re_suspend(device_t dev)
2208 {
2209 #ifndef BURN_BRIDGES
2210 	int i;
2211 #endif
2212 	struct re_softc *sc = device_get_softc(dev);
2213 
2214 	re_stop(sc);
2215 
2216 #ifndef BURN_BRIDGES
2217 	for (i = 0; i < 5; i++)
2218 		sc->saved_maps[i] = pci_read_config(dev, PCIR_MAPS + i * 4, 4);
2219 	sc->saved_biosaddr = pci_read_config(dev, PCIR_BIOS, 4);
2220 	sc->saved_intline = pci_read_config(dev, PCIR_INTLINE, 1);
2221 	sc->saved_cachelnsz = pci_read_config(dev, PCIR_CACHELNSZ, 1);
2222 	sc->saved_lattimer = pci_read_config(dev, PCIR_LATTIMER, 1);
2223 #endif
2224 
2225 	sc->suspended = 1;
2226 
2227 	return (0);
2228 }
2229 
2230 /*
2231  * Device resume routine.  Restore some PCI settings in case the BIOS
2232  * doesn't, re-enable busmastering, and restart the interface if
2233  * appropriate.
2234  */
2235 static int
2236 re_resume(device_t dev)
2237 {
2238 	struct re_softc *sc = device_get_softc(dev);
2239 	struct ifnet *ifp = &sc->arpcom.ac_if;
2240 #ifndef BURN_BRIDGES
2241 	int i;
2242 #endif
2243 
2244 #ifndef BURN_BRIDGES
2245 	/* better way to do this? */
2246 	for (i = 0; i < 5; i++)
2247 		pci_write_config(dev, PCIR_MAPS + i * 4, sc->saved_maps[i], 4);
2248 	pci_write_config(dev, PCIR_BIOS, sc->saved_biosaddr, 4);
2249 	pci_write_config(dev, PCIR_INTLINE, sc->saved_intline, 1);
2250 	pci_write_config(dev, PCIR_CACHELNSZ, sc->saved_cachelnsz, 1);
2251 	pci_write_config(dev, PCIR_LATTIMER, sc->saved_lattimer, 1);
2252 
2253 	/* reenable busmastering */
2254 	pci_enable_busmaster(dev);
2255 	pci_enable_io(dev, SYS_RES_IOPORT);
2256 #endif
2257 
2258 	/* reinitialize interface if necessary */
2259 	if (ifp->if_flags & IFF_UP)
2260 		re_init(sc);
2261 
2262 	sc->suspended = 0;
2263 
2264 	return (0);
2265 }
2266 
2267 /*
2268  * Stop all chip I/O so that the kernel's probe routines don't
2269  * get confused by errant DMAs when rebooting.
2270  */
2271 static void
2272 re_shutdown(device_t dev)
2273 {
2274 	struct re_softc *sc = device_get_softc(dev);
2275 
2276 	re_stop(sc);
2277 }
2278