xref: /dragonfly/sys/dev/netif/re/if_re.c (revision 509221ae)
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 1.14 2005/06/09 20:04:44 joerg 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 #include <sys/thread2.h>
124 
125 #include <net/if.h>
126 #include <net/ifq_var.h>
127 #include <net/if_arp.h>
128 #include <net/ethernet.h>
129 #include <net/if_dl.h>
130 #include <net/if_media.h>
131 #include <net/if_types.h>
132 #include <net/vlan/if_vlan_var.h>
133 
134 #include <net/bpf.h>
135 
136 #include <machine/bus_pio.h>
137 #include <machine/bus_memio.h>
138 #include <machine/bus.h>
139 #include <machine/resource.h>
140 #include <sys/bus.h>
141 #include <sys/rman.h>
142 
143 #include <dev/netif/mii_layer/mii.h>
144 #include <dev/netif/mii_layer/miivar.h>
145 
146 #include <bus/pci/pcireg.h>
147 #include <bus/pci/pcivar.h>
148 
149 /* "controller miibus0" required.  See GENERIC if you get errors here. */
150 #include "miibus_if.h"
151 
152 #include <dev/netif/re/if_rereg.h>
153 
154 /*
155  * The hardware supports checksumming but, as usual, some chipsets screw it
156  * all up and produce bogus packets, so we disable it by default.
157  */
158 #define RE_CSUM_FEATURES    (CSUM_IP | CSUM_TCP | CSUM_UDP)
159 #define RE_DISABLE_HWCSUM
160 
161 /*
162  * Various supported device vendors/types and their names.
163  */
164 static struct re_type re_devs[] = {
165 	{ RT_VENDORID, RT_DEVICEID_8139, RE_HWREV_8139CPLUS,
166 		"RealTek 8139C+ 10/100BaseTX" },
167 	{ RT_VENDORID, RT_DEVICEID_8169, RE_HWREV_8169,
168 		"RealTek 8169 Gigabit Ethernet" },
169 	{ RT_VENDORID, RT_DEVICEID_8169, RE_HWREV_8169S,
170 		"RealTek 8169S Single-chip Gigabit Ethernet" },
171 	{ RT_VENDORID, RT_DEVICEID_8169, RE_HWREV_8110S,
172 		"RealTek 8110S Single-chip Gigabit Ethernet" },
173 	{ 0, 0, 0, NULL }
174 };
175 
176 static struct re_hwrev re_hwrevs[] = {
177 	{ RE_HWREV_8139CPLUS, RE_8139CPLUS, "C+"},
178 	{ RE_HWREV_8169, RE_8169, "8169"},
179 	{ RE_HWREV_8169S, RE_8169, "8169S"},
180 	{ RE_HWREV_8110S, RE_8169, "8110S"},
181 	{ 0, 0, NULL }
182 };
183 
184 static int	re_probe(device_t);
185 static int	re_attach(device_t);
186 static int	re_detach(device_t);
187 
188 static int	re_encap(struct re_softc *, struct mbuf **, int *, int *);
189 
190 static void	re_dma_map_addr(void *, bus_dma_segment_t *, int, int);
191 static void	re_dma_map_desc(void *, bus_dma_segment_t *, int,
192 				bus_size_t, int);
193 static int	re_allocmem(device_t, struct re_softc *);
194 static int	re_newbuf(struct re_softc *, int, struct mbuf *);
195 static int	re_rx_list_init(struct re_softc *);
196 static int	re_tx_list_init(struct re_softc *);
197 static void	re_rxeof(struct re_softc *);
198 static void	re_txeof(struct re_softc *);
199 static void	re_intr(void *);
200 static void	re_tick(void *);
201 static void	re_start(struct ifnet *);
202 static int	re_ioctl(struct ifnet *, u_long, caddr_t, struct ucred *);
203 static void	re_init(void *);
204 static void	re_stop(struct re_softc *);
205 static void	re_watchdog(struct ifnet *);
206 static int	re_suspend(device_t);
207 static int	re_resume(device_t);
208 static void	re_shutdown(device_t);
209 static int	re_ifmedia_upd(struct ifnet *);
210 static void	re_ifmedia_sts(struct ifnet *, struct ifmediareq *);
211 
212 static void	re_eeprom_putbyte(struct re_softc *, int);
213 static void	re_eeprom_getword(struct re_softc *, int, u_int16_t *);
214 static void	re_read_eeprom(struct re_softc *, caddr_t, int, int, int);
215 static int	re_gmii_readreg(device_t, int, int);
216 static int	re_gmii_writereg(device_t, int, int, int);
217 
218 static int	re_miibus_readreg(device_t, int, int);
219 static int	re_miibus_writereg(device_t, int, int, int);
220 static void	re_miibus_statchg(device_t);
221 
222 static void	re_setmulti(struct re_softc *);
223 static void	re_reset(struct re_softc *);
224 
225 static int	re_diag(struct re_softc *);
226 #ifdef DEVICE_POLLING
227 static void	re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count);
228 #endif
229 
230 static device_method_t re_methods[] = {
231 	/* Device interface */
232 	DEVMETHOD(device_probe,		re_probe),
233 	DEVMETHOD(device_attach,	re_attach),
234 	DEVMETHOD(device_detach,	re_detach),
235 	DEVMETHOD(device_suspend,	re_suspend),
236 	DEVMETHOD(device_resume,	re_resume),
237 	DEVMETHOD(device_shutdown,	re_shutdown),
238 
239 	/* bus interface */
240 	DEVMETHOD(bus_print_child,	bus_generic_print_child),
241 	DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
242 
243 	/* MII interface */
244 	DEVMETHOD(miibus_readreg,	re_miibus_readreg),
245 	DEVMETHOD(miibus_writereg,	re_miibus_writereg),
246 	DEVMETHOD(miibus_statchg,	re_miibus_statchg),
247 
248 	{ 0, 0 }
249 };
250 
251 static driver_t re_driver = {
252 	"re",
253 	re_methods,
254 	sizeof(struct re_softc)
255 };
256 
257 static devclass_t re_devclass;
258 
259 DECLARE_DUMMY_MODULE(if_re);
260 DRIVER_MODULE(if_re, pci, re_driver, re_devclass, 0, 0);
261 DRIVER_MODULE(if_re, cardbus, re_driver, re_devclass, 0, 0);
262 DRIVER_MODULE(miibus, re, miibus_driver, miibus_devclass, 0, 0);
263 
264 #define EE_SET(x)	\
265 	CSR_WRITE_1(sc, RE_EECMD, CSR_READ_1(sc, RE_EECMD) | (x))
266 
267 #define EE_CLR(x)	\
268 	CSR_WRITE_1(sc, RE_EECMD, CSR_READ_1(sc, RE_EECMD) & ~(x))
269 
270 /*
271  * Send a read command and address to the EEPROM, check for ACK.
272  */
273 static void
274 re_eeprom_putbyte(struct re_softc *sc, int addr)
275 {
276 	int d, i;
277 
278 	d = addr | sc->re_eecmd_read;
279 
280 	/*
281 	 * Feed in each bit and strobe the clock.
282 	 */
283 	for (i = 0x400; i != 0; i >>= 1) {
284 		if (d & i)
285 			EE_SET(RE_EE_DATAIN);
286 		else
287 			EE_CLR(RE_EE_DATAIN);
288 		DELAY(100);
289 		EE_SET(RE_EE_CLK);
290 		DELAY(150);
291 		EE_CLR(RE_EE_CLK);
292 		DELAY(100);
293 	}
294 }
295 
296 /*
297  * Read a word of data stored in the EEPROM at address 'addr.'
298  */
299 static void
300 re_eeprom_getword(struct re_softc *sc, int addr, uint16_t *dest)
301 {
302 	int i;
303 	uint16_t word = 0;
304 
305 	/* Enter EEPROM access mode. */
306 	CSR_WRITE_1(sc, RE_EECMD, RE_EEMODE_PROGRAM|RE_EE_SEL);
307 
308 	/*
309 	 * Send address of word we want to read.
310 	 */
311 	re_eeprom_putbyte(sc, addr);
312 
313 	CSR_WRITE_1(sc, RE_EECMD, RE_EEMODE_PROGRAM|RE_EE_SEL);
314 
315 	/*
316 	 * Start reading bits from EEPROM.
317 	 */
318 	for (i = 0x8000; i != 0; i >>= 1) {
319 		EE_SET(RE_EE_CLK);
320 		DELAY(100);
321 		if (CSR_READ_1(sc, RE_EECMD) & RE_EE_DATAOUT)
322 			word |= i;
323 		EE_CLR(RE_EE_CLK);
324 		DELAY(100);
325 	}
326 
327 	/* Turn off EEPROM access mode. */
328 	CSR_WRITE_1(sc, RE_EECMD, RE_EEMODE_OFF);
329 
330 	*dest = word;
331 }
332 
333 /*
334  * Read a sequence of words from the EEPROM.
335  */
336 static void
337 re_read_eeprom(struct re_softc *sc, caddr_t dest, int off, int cnt, int swap)
338 {
339 	int i;
340 	uint16_t word = 0, *ptr;
341 
342 	for (i = 0; i < cnt; i++) {
343 		re_eeprom_getword(sc, off + i, &word);
344 		ptr = (u_int16_t *)(dest + (i * 2));
345 		if (swap)
346 			*ptr = be16toh(word);
347 		else
348 			*ptr = word;
349 	}
350 }
351 
352 static int
353 re_gmii_readreg(device_t dev, int phy, int reg)
354 {
355 	struct re_softc *sc = device_get_softc(dev);
356 	u_int32_t rval;
357 	int i;
358 
359 	if (phy != 1)
360 		return(0);
361 
362 	/* Let the rgephy driver read the GMEDIASTAT register */
363 
364 	if (reg == RE_GMEDIASTAT)
365 		return(CSR_READ_1(sc, RE_GMEDIASTAT));
366 
367 	CSR_WRITE_4(sc, RE_PHYAR, reg << 16);
368 	DELAY(1000);
369 
370 	for (i = 0; i < RE_TIMEOUT; i++) {
371 		rval = CSR_READ_4(sc, RE_PHYAR);
372 		if (rval & RE_PHYAR_BUSY)
373 			break;
374 		DELAY(100);
375 	}
376 
377 	if (i == RE_TIMEOUT) {
378 		device_printf(dev, "PHY read failed\n");
379 		return(0);
380 	}
381 
382 	return(rval & RE_PHYAR_PHYDATA);
383 }
384 
385 static int
386 re_gmii_writereg(device_t dev, int phy, int reg, int data)
387 {
388 	struct re_softc *sc = device_get_softc(dev);
389 	uint32_t rval;
390 	int i;
391 
392 	CSR_WRITE_4(sc, RE_PHYAR,
393 		    (reg << 16) | (data & RE_PHYAR_PHYDATA) | RE_PHYAR_BUSY);
394 	DELAY(1000);
395 
396 	for (i = 0; i < RE_TIMEOUT; i++) {
397 		rval = CSR_READ_4(sc, RE_PHYAR);
398 		if ((rval & RE_PHYAR_BUSY) == 0)
399 			break;
400 		DELAY(100);
401 	}
402 
403 	if (i == RE_TIMEOUT)
404 		device_printf(dev, "PHY write failed\n");
405 
406 	return(0);
407 }
408 
409 static int
410 re_miibus_readreg(device_t dev, int phy, int reg)
411 {
412 	struct re_softc	*sc = device_get_softc(dev);
413 	uint16_t rval = 0;
414 	uint16_t re8139_reg = 0;
415 
416 	if (sc->re_type == RE_8169) {
417 		rval = re_gmii_readreg(dev, phy, reg);
418 		return(rval);
419 	}
420 
421 	/* Pretend the internal PHY is only at address 0 */
422 	if (phy)
423 		return(0);
424 
425 	switch(reg) {
426 	case MII_BMCR:
427 		re8139_reg = RE_BMCR;
428 		break;
429 	case MII_BMSR:
430 		re8139_reg = RE_BMSR;
431 		break;
432 	case MII_ANAR:
433 		re8139_reg = RE_ANAR;
434 		break;
435 	case MII_ANER:
436 		re8139_reg = RE_ANER;
437 		break;
438 	case MII_ANLPAR:
439 		re8139_reg = RE_LPAR;
440 		break;
441 	case MII_PHYIDR1:
442 	case MII_PHYIDR2:
443 		return(0);
444 	/*
445 	 * Allow the rlphy driver to read the media status
446 	 * register. If we have a link partner which does not
447 	 * support NWAY, this is the register which will tell
448 	 * us the results of parallel detection.
449 	 */
450 	case RE_MEDIASTAT:
451 		return(CSR_READ_1(sc, RE_MEDIASTAT));
452 	default:
453 		device_printf(dev, "bad phy register\n");
454 		return(0);
455 	}
456 	rval = CSR_READ_2(sc, re8139_reg);
457 	return(rval);
458 }
459 
460 static int
461 re_miibus_writereg(device_t dev, int phy, int reg, int data)
462 {
463 	struct re_softc *sc= device_get_softc(dev);
464 	u_int16_t re8139_reg = 0;
465 
466 	if (sc->re_type == RE_8169)
467 		return(re_gmii_writereg(dev, phy, reg, data));
468 
469 	/* Pretend the internal PHY is only at address 0 */
470 	if (phy)
471 		return(0);
472 
473 	switch(reg) {
474 	case MII_BMCR:
475 		re8139_reg = RE_BMCR;
476 		break;
477 	case MII_BMSR:
478 		re8139_reg = RE_BMSR;
479 		break;
480 	case MII_ANAR:
481 		re8139_reg = RE_ANAR;
482 		break;
483 	case MII_ANER:
484 		re8139_reg = RE_ANER;
485 		break;
486 	case MII_ANLPAR:
487 		re8139_reg = RE_LPAR;
488 		break;
489 	case MII_PHYIDR1:
490 	case MII_PHYIDR2:
491 		return(0);
492 	default:
493 		device_printf(dev, "bad phy register\n");
494 		return(0);
495 	}
496 	CSR_WRITE_2(sc, re8139_reg, data);
497 	return(0);
498 }
499 
500 static void
501 re_miibus_statchg(device_t dev)
502 {
503 }
504 
505 /*
506  * Program the 64-bit multicast hash filter.
507  */
508 static void
509 re_setmulti(struct re_softc *sc)
510 {
511 	struct ifnet *ifp = &sc->arpcom.ac_if;
512 	int h = 0;
513 	uint32_t hashes[2] = { 0, 0 };
514 	struct ifmultiaddr *ifma;
515 	uint32_t rxfilt;
516 	int mcnt = 0;
517 
518 	rxfilt = CSR_READ_4(sc, RE_RXCFG);
519 
520 	if (ifp->if_flags & IFF_ALLMULTI || ifp->if_flags & IFF_PROMISC) {
521 		rxfilt |= RE_RXCFG_RX_MULTI;
522 		CSR_WRITE_4(sc, RE_RXCFG, rxfilt);
523 		CSR_WRITE_4(sc, RE_MAR0, 0xFFFFFFFF);
524 		CSR_WRITE_4(sc, RE_MAR4, 0xFFFFFFFF);
525 		return;
526 	}
527 
528 	/* first, zot all the existing hash bits */
529 	CSR_WRITE_4(sc, RE_MAR0, 0);
530 	CSR_WRITE_4(sc, RE_MAR4, 0);
531 
532 	/* now program new ones */
533 	LIST_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
534 		if (ifma->ifma_addr->sa_family != AF_LINK)
535 			continue;
536 		h = ether_crc32_be(LLADDR((struct sockaddr_dl *)
537 		    ifma->ifma_addr), ETHER_ADDR_LEN) >> 26;
538 		if (h < 32)
539 			hashes[0] |= (1 << h);
540 		else
541 			hashes[1] |= (1 << (h - 32));
542 		mcnt++;
543 	}
544 
545 	if (mcnt)
546 		rxfilt |= RE_RXCFG_RX_MULTI;
547 	else
548 		rxfilt &= ~RE_RXCFG_RX_MULTI;
549 
550 	CSR_WRITE_4(sc, RE_RXCFG, rxfilt);
551 	CSR_WRITE_4(sc, RE_MAR0, hashes[0]);
552 	CSR_WRITE_4(sc, RE_MAR4, hashes[1]);
553 }
554 
555 static void
556 re_reset(struct re_softc *sc)
557 {
558 	int i;
559 
560 	CSR_WRITE_1(sc, RE_COMMAND, RE_CMD_RESET);
561 
562 	for (i = 0; i < RE_TIMEOUT; i++) {
563 		DELAY(10);
564 		if ((CSR_READ_1(sc, RE_COMMAND) & RE_CMD_RESET) == 0)
565 			break;
566 	}
567 	if (i == RE_TIMEOUT)
568 		if_printf(&sc->arpcom.ac_if, "reset never completed!\n");
569 
570 	CSR_WRITE_1(sc, 0x82, 1);
571 }
572 
573 /*
574  * The following routine is designed to test for a defect on some
575  * 32-bit 8169 cards. Some of these NICs have the REQ64# and ACK64#
576  * lines connected to the bus, however for a 32-bit only card, they
577  * should be pulled high. The result of this defect is that the
578  * NIC will not work right if you plug it into a 64-bit slot: DMA
579  * operations will be done with 64-bit transfers, which will fail
580  * because the 64-bit data lines aren't connected.
581  *
582  * There's no way to work around this (short of talking a soldering
583  * iron to the board), however we can detect it. The method we use
584  * here is to put the NIC into digital loopback mode, set the receiver
585  * to promiscuous mode, and then try to send a frame. We then compare
586  * the frame data we sent to what was received. If the data matches,
587  * then the NIC is working correctly, otherwise we know the user has
588  * a defective NIC which has been mistakenly plugged into a 64-bit PCI
589  * slot. In the latter case, there's no way the NIC can work correctly,
590  * so we print out a message on the console and abort the device attach.
591  */
592 
593 static int
594 re_diag(struct re_softc *sc)
595 {
596 	struct ifnet *ifp = &sc->arpcom.ac_if;
597 	struct mbuf *m0;
598 	struct ether_header *eh;
599 	struct re_desc *cur_rx;
600 	uint16_t status;
601 	uint32_t rxstat;
602 	int total_len, i, error = 0;
603 	uint8_t dst[ETHER_ADDR_LEN] = { 0x00, 'h', 'e', 'l', 'l', 'o' };
604 	uint8_t src[ETHER_ADDR_LEN] = { 0x00, 'w', 'o', 'r', 'l', 'd' };
605 
606 	/* Allocate a single mbuf */
607 
608 	MGETHDR(m0, MB_DONTWAIT, MT_DATA);
609 	if (m0 == NULL)
610 		return(ENOBUFS);
611 
612 	/*
613 	 * Initialize the NIC in test mode. This sets the chip up
614 	 * so that it can send and receive frames, but performs the
615 	 * following special functions:
616 	 * - Puts receiver in promiscuous mode
617 	 * - Enables digital loopback mode
618 	 * - Leaves interrupts turned off
619 	 */
620 
621 	ifp->if_flags |= IFF_PROMISC;
622 	sc->re_testmode = 1;
623 	re_init(sc);
624 	re_stop(sc);
625 	DELAY(100000);
626 	re_init(sc);
627 
628 	/* Put some data in the mbuf */
629 
630 	eh = mtod(m0, struct ether_header *);
631 	bcopy (dst, eh->ether_dhost, ETHER_ADDR_LEN);
632 	bcopy (src, eh->ether_shost, ETHER_ADDR_LEN);
633 	eh->ether_type = htons(ETHERTYPE_IP);
634 	m0->m_pkthdr.len = m0->m_len = ETHER_MIN_LEN - ETHER_CRC_LEN;
635 
636 	/*
637 	 * Queue the packet, start transmission.
638 	 * Note: ifq_handoff() ultimately calls re_start() for us.
639 	 */
640 
641 	CSR_WRITE_2(sc, RE_ISR, 0xFFFF);
642 	error = ifq_handoff(ifp, m0, NULL);
643 	if (error) {
644 		m0 = NULL;
645 		goto done;
646 	}
647 	m0 = NULL;
648 
649 	/* Wait for it to propagate through the chip */
650 
651 	DELAY(100000);
652 	for (i = 0; i < RE_TIMEOUT; i++) {
653 		status = CSR_READ_2(sc, RE_ISR);
654 		if ((status & (RE_ISR_TIMEOUT_EXPIRED|RE_ISR_RX_OK)) ==
655 		    (RE_ISR_TIMEOUT_EXPIRED|RE_ISR_RX_OK))
656 			break;
657 		DELAY(10);
658 	}
659 
660 	if (i == RE_TIMEOUT) {
661 		if_printf(ifp, "diagnostic failed to receive packet "
662 			  "in loopback mode\n");
663 		error = EIO;
664 		goto done;
665 	}
666 
667 	/*
668 	 * The packet should have been dumped into the first
669 	 * entry in the RX DMA ring. Grab it from there.
670 	 */
671 
672 	bus_dmamap_sync(sc->re_ldata.re_rx_list_tag,
673 			sc->re_ldata.re_rx_list_map, BUS_DMASYNC_POSTREAD);
674 	bus_dmamap_sync(sc->re_ldata.re_mtag, sc->re_ldata.re_rx_dmamap[0],
675 			BUS_DMASYNC_POSTWRITE);
676 	bus_dmamap_unload(sc->re_ldata.re_mtag, sc->re_ldata.re_rx_dmamap[0]);
677 
678 	m0 = sc->re_ldata.re_rx_mbuf[0];
679 	sc->re_ldata.re_rx_mbuf[0] = NULL;
680 	eh = mtod(m0, struct ether_header *);
681 
682 	cur_rx = &sc->re_ldata.re_rx_list[0];
683 	total_len = RE_RXBYTES(cur_rx);
684 	rxstat = le32toh(cur_rx->re_cmdstat);
685 
686 	if (total_len != ETHER_MIN_LEN) {
687 		if_printf(ifp, "diagnostic failed, received short packet\n");
688 		error = EIO;
689 		goto done;
690 	}
691 
692 	/* Test that the received packet data matches what we sent. */
693 
694 	if (bcmp(eh->ether_dhost, dst, ETHER_ADDR_LEN) ||
695 	    bcmp(eh->ether_shost, &src, ETHER_ADDR_LEN) ||
696 	    be16toh(eh->ether_type) != ETHERTYPE_IP) {
697 		if_printf(ifp, "WARNING, DMA FAILURE!\n");
698 		if_printf(ifp, "expected TX data: %6D/%6D/0x%x\n",
699 		    dst, ":", src, ":", ETHERTYPE_IP);
700 		if_printf(ifp, "received RX data: %6D/%6D/0x%x\n",
701 		    eh->ether_dhost, ":",  eh->ether_shost, ":",
702 		    ntohs(eh->ether_type));
703 		if_printf(ifp, "You may have a defective 32-bit NIC plugged "
704 		    "into a 64-bit PCI slot.\n");
705 		if_printf(ifp, "Please re-install the NIC in a 32-bit slot "
706 		    "for proper operation.\n");
707 		if_printf(ifp, "Read the re(4) man page for more details.\n");
708 		error = EIO;
709 	}
710 
711 done:
712 	/* Turn interface off, release resources */
713 
714 	sc->re_testmode = 0;
715 	ifp->if_flags &= ~IFF_PROMISC;
716 	re_stop(sc);
717 	if (m0 != NULL)
718 		m_freem(m0);
719 
720 	return (error);
721 }
722 
723 /*
724  * Probe for a RealTek 8139C+/8169/8110 chip. Check the PCI vendor and device
725  * IDs against our list and return a device name if we find a match.
726  */
727 static int
728 re_probe(device_t dev)
729 {
730 	struct re_type *t;
731 	struct re_softc *sc;
732 	int rid;
733 	uint32_t hwrev;
734 	uint16_t vendor, product;
735 
736 	t = re_devs;
737 
738 	vendor = pci_get_vendor(dev);
739 	product = pci_get_device(dev);
740 
741 	for (t = re_devs; t->re_name != NULL; t++) {
742 		if (product == t->re_did && vendor == t->re_vid)
743 			break;
744 	}
745 
746 	/*
747 	 * Check if we found a RealTek device.
748 	 */
749 	if (t->re_name == NULL)
750 		return(ENXIO);
751 
752 	/*
753 	 * Temporarily map the I/O space so we can read the chip ID register.
754 	 */
755 	sc = malloc(sizeof(*sc), M_TEMP, M_WAITOK | M_ZERO);
756 	rid = RE_PCI_LOIO;
757 	sc->re_res = bus_alloc_resource_any(dev, SYS_RES_IOPORT, &rid,
758 					    RF_ACTIVE);
759 	if (sc->re_res == NULL) {
760 		device_printf(dev, "couldn't map ports/memory\n");
761 		free(sc, M_TEMP);
762 		return(ENXIO);
763 	}
764 
765 	sc->re_btag = rman_get_bustag(sc->re_res);
766 	sc->re_bhandle = rman_get_bushandle(sc->re_res);
767 
768 	hwrev = CSR_READ_4(sc, RE_TXCFG) & RE_TXCFG_HWREV;
769 	bus_release_resource(dev, SYS_RES_IOPORT, RE_PCI_LOIO, sc->re_res);
770 	free(sc, M_TEMP);
771 
772 	/*
773 	 * and continue matching for the specific chip...
774 	 */
775 	for (; t->re_name != NULL; t++) {
776 		if (product == t->re_did && vendor == t->re_vid &&
777 		    t->re_basetype == hwrev) {
778 			device_set_desc(dev, t->re_name);
779 			return(0);
780 		}
781 	}
782 	return(ENXIO);
783 }
784 
785 /*
786  * This routine takes the segment list provided as the result of
787  * a bus_dma_map_load() operation and assigns the addresses/lengths
788  * to RealTek DMA descriptors. This can be called either by the RX
789  * code or the TX code. In the RX case, we'll probably wind up mapping
790  * at most one segment. For the TX case, there could be any number of
791  * segments since TX packets may span multiple mbufs. In either case,
792  * if the number of segments is larger than the re_maxsegs limit
793  * specified by the caller, we abort the mapping operation. Sadly,
794  * whoever designed the buffer mapping API did not provide a way to
795  * return an error from here, so we have to fake it a bit.
796  */
797 
798 static void
799 re_dma_map_desc(void *arg, bus_dma_segment_t *segs, int nseg,
800 		bus_size_t mapsize, int error)
801 {
802 	struct re_dmaload_arg *ctx;
803 	struct re_desc *d = NULL;
804 	int i = 0, idx;
805 	uint32_t cmdstat;
806 
807 	if (error)
808 		return;
809 
810 	ctx = arg;
811 
812 	/* Signal error to caller if there's too many segments */
813 	if (nseg > ctx->re_maxsegs) {
814 		ctx->re_maxsegs = 0;
815 		return;
816 	}
817 
818 	/*
819 	 * Map the segment array into descriptors. Note that we set the
820 	 * start-of-frame and end-of-frame markers for either TX or RX, but
821 	 * they really only have meaning in the TX case. (In the RX case,
822 	 * it's the chip that tells us where packets begin and end.)
823 	 * We also keep track of the end of the ring and set the
824 	 * end-of-ring bits as needed, and we set the ownership bits
825 	 * in all except the very first descriptor. (The caller will
826 	 * set this descriptor later when it start transmission or
827 	 * reception.)
828 	 */
829 	idx = ctx->re_idx;
830 	for (;;) {
831 		d = &ctx->re_ring[idx];
832 		if (le32toh(d->re_cmdstat) & RE_RDESC_STAT_OWN) {
833 			ctx->re_maxsegs = 0;
834 			return;
835 		}
836 		cmdstat = segs[i].ds_len;
837 		d->re_bufaddr_lo = htole32(RE_ADDR_LO(segs[i].ds_addr));
838 		d->re_bufaddr_hi = htole32(RE_ADDR_HI(segs[i].ds_addr));
839 		if (i == 0)
840 			cmdstat |= RE_TDESC_CMD_SOF;
841 		else
842 			cmdstat |= RE_TDESC_CMD_OWN;
843 		if (idx == (RE_RX_DESC_CNT - 1))
844 			cmdstat |= RE_TDESC_CMD_EOR;
845 		d->re_cmdstat = htole32(cmdstat | ctx->re_flags);
846 		i++;
847 		if (i == nseg)
848 			break;
849 		RE_DESC_INC(idx);
850 	}
851 
852 	d->re_cmdstat |= htole32(RE_TDESC_CMD_EOF);
853 	ctx->re_maxsegs = nseg;
854 	ctx->re_idx = idx;
855 }
856 
857 /*
858  * Map a single buffer address.
859  */
860 
861 static void
862 re_dma_map_addr(void *arg, bus_dma_segment_t *segs, int nseg, int error)
863 {
864 	uint32_t *addr;
865 
866 	if (error)
867 		return;
868 
869 	KASSERT(nseg == 1, ("too many DMA segments, %d should be 1", nseg));
870 	addr = arg;
871 	*addr = segs->ds_addr;
872 }
873 
874 static int
875 re_allocmem(device_t dev, struct re_softc *sc)
876 {
877 	int error, i, nseg;
878 
879 	/*
880 	 * Allocate map for RX mbufs.
881 	 */
882 	nseg = 32;
883 	error = bus_dma_tag_create(sc->re_parent_tag, ETHER_ALIGN, 0,
884 	    BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
885 	    NULL, MCLBYTES * nseg, nseg, MCLBYTES, BUS_DMA_ALLOCNOW,
886 	    &sc->re_ldata.re_mtag);
887 	if (error) {
888 		device_printf(dev, "could not allocate dma tag\n");
889 		return(error);
890 	}
891 
892 	/*
893 	 * Allocate map for TX descriptor list.
894 	 */
895 	error = bus_dma_tag_create(sc->re_parent_tag, RE_RING_ALIGN,
896 	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
897             NULL, RE_TX_LIST_SZ, 1, RE_TX_LIST_SZ, BUS_DMA_ALLOCNOW,
898 	    &sc->re_ldata.re_tx_list_tag);
899 	if (error) {
900 		device_printf(dev, "could not allocate dma tag\n");
901 		return(error);
902 	}
903 
904 	/* Allocate DMA'able memory for the TX ring */
905 
906         error = bus_dmamem_alloc(sc->re_ldata.re_tx_list_tag,
907 	    (void **)&sc->re_ldata.re_tx_list, BUS_DMA_WAITOK | BUS_DMA_ZERO,
908             &sc->re_ldata.re_tx_list_map);
909         if (error) {
910 		device_printf(dev, "could not allocate TX ring\n");
911                 return(error);
912 	}
913 
914 	/* Load the map for the TX ring. */
915 
916 	error = bus_dmamap_load(sc->re_ldata.re_tx_list_tag,
917 	     sc->re_ldata.re_tx_list_map, sc->re_ldata.re_tx_list,
918 	     RE_TX_LIST_SZ, re_dma_map_addr,
919 	     &sc->re_ldata.re_tx_list_addr, BUS_DMA_NOWAIT);
920 	if (error) {
921 		device_printf(dev, "could not get addres of TX ring\n");
922 		return(error);
923 	}
924 
925 	/* Create DMA maps for TX buffers */
926 
927 	for (i = 0; i < RE_TX_DESC_CNT; i++) {
928 		error = bus_dmamap_create(sc->re_ldata.re_mtag, 0,
929 			    &sc->re_ldata.re_tx_dmamap[i]);
930 		if (error) {
931 			device_printf(dev, "can't create DMA map for TX\n");
932 			return(error);
933 		}
934 	}
935 
936 	/*
937 	 * Allocate map for RX descriptor list.
938 	 */
939 	error = bus_dma_tag_create(sc->re_parent_tag, RE_RING_ALIGN,
940 	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
941             NULL, RE_TX_LIST_SZ, 1, RE_TX_LIST_SZ, BUS_DMA_ALLOCNOW,
942 	    &sc->re_ldata.re_rx_list_tag);
943 	if (error) {
944 		device_printf(dev, "could not allocate dma tag\n");
945 		return(error);
946 	}
947 
948 	/* Allocate DMA'able memory for the RX ring */
949 
950         error = bus_dmamem_alloc(sc->re_ldata.re_rx_list_tag,
951 	    (void **)&sc->re_ldata.re_rx_list, BUS_DMA_WAITOK | BUS_DMA_ZERO,
952             &sc->re_ldata.re_rx_list_map);
953         if (error) {
954 		device_printf(dev, "could not allocate RX ring\n");
955                 return(error);
956 	}
957 
958 	/* Load the map for the RX ring. */
959 
960 	error = bus_dmamap_load(sc->re_ldata.re_rx_list_tag,
961 	     sc->re_ldata.re_rx_list_map, sc->re_ldata.re_rx_list,
962 	     RE_TX_LIST_SZ, re_dma_map_addr,
963 	     &sc->re_ldata.re_rx_list_addr, BUS_DMA_NOWAIT);
964 	if (error) {
965 		device_printf(dev, "could not get address of RX ring\n");
966 		return(error);
967 	}
968 
969 	/* Create DMA maps for RX buffers */
970 
971 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
972 		error = bus_dmamap_create(sc->re_ldata.re_mtag, 0,
973 			    &sc->re_ldata.re_rx_dmamap[i]);
974 		if (error) {
975 			device_printf(dev, "can't create DMA map for RX\n");
976 			return(ENOMEM);
977 		}
978 	}
979 
980 	return(0);
981 }
982 
983 /*
984  * Attach the interface. Allocate softc structures, do ifmedia
985  * setup and ethernet/BPF attach.
986  */
987 static int
988 re_attach(device_t dev)
989 {
990 	struct re_softc	*sc = device_get_softc(dev);
991 	struct ifnet *ifp;
992 	struct re_hwrev *hw_rev;
993 	uint8_t eaddr[ETHER_ADDR_LEN];
994 	int hwrev;
995 	u_int16_t re_did = 0;
996 	int error = 0, rid, i;
997 
998 	callout_init(&sc->re_timer);
999 
1000 #ifndef BURN_BRIDGES
1001 	/*
1002 	 * Handle power management nonsense.
1003 	 */
1004 
1005 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D0) {
1006 		uint32_t membase, irq;
1007 
1008 		/* Save important PCI config data. */
1009 		membase = pci_read_config(dev, RE_PCI_LOMEM, 4);
1010 		irq = pci_read_config(dev, PCIR_INTLINE, 4);
1011 
1012 		/* Reset the power state. */
1013 		device_printf(dev, "chip is is in D%d power mode "
1014 		    "-- setting to D0\n", pci_get_powerstate(dev));
1015 
1016 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
1017 
1018 		/* Restore PCI config data. */
1019 		pci_write_config(dev, RE_PCI_LOMEM, membase, 4);
1020 		pci_write_config(dev, PCIR_INTLINE, irq, 4);
1021 	}
1022 #endif
1023 	/*
1024 	 * Map control/status registers.
1025 	 */
1026 	pci_enable_busmaster(dev);
1027 
1028 	rid = RE_PCI_LOIO;
1029 	sc->re_res = bus_alloc_resource_any(dev, SYS_RES_IOPORT, &rid,
1030 					    RF_ACTIVE);
1031 
1032 	if (sc->re_res == NULL) {
1033 		device_printf(dev, "couldn't map ports/memory\n");
1034 		error = ENXIO;
1035 		goto fail;
1036 	}
1037 
1038 	sc->re_btag = rman_get_bustag(sc->re_res);
1039 	sc->re_bhandle = rman_get_bushandle(sc->re_res);
1040 
1041 	/* Allocate interrupt */
1042 	rid = 0;
1043 	sc->re_irq = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
1044 					    RF_SHAREABLE | RF_ACTIVE);
1045 
1046 	if (sc->re_irq == NULL) {
1047 		device_printf(dev, "couldn't map interrupt\n");
1048 		error = ENXIO;
1049 		goto fail;
1050 	}
1051 
1052 	/* Reset the adapter. */
1053 	re_reset(sc);
1054 
1055 	hwrev = CSR_READ_4(sc, RE_TXCFG) & RE_TXCFG_HWREV;
1056 	for (hw_rev = re_hwrevs; hw_rev->re_desc != NULL; hw_rev++) {
1057 		if (hw_rev->re_rev == hwrev) {
1058 			sc->re_type = hw_rev->re_type;
1059 			break;
1060 		}
1061 	}
1062 
1063 	if (sc->re_type == RE_8169) {
1064 		/* Set RX length mask */
1065 		sc->re_rxlenmask = RE_RDESC_STAT_GFRAGLEN;
1066 
1067 		/* Force station address autoload from the EEPROM */
1068 		CSR_WRITE_1(sc, RE_EECMD, RE_EEMODE_AUTOLOAD);
1069 		for (i = 0; i < RE_TIMEOUT; i++) {
1070 			if ((CSR_READ_1(sc, RE_EECMD) & RE_EEMODE_AUTOLOAD) == 0)
1071 				break;
1072 			DELAY(100);
1073 		}
1074 		if (i == RE_TIMEOUT)
1075 			device_printf(dev, "eeprom autoload timed out\n");
1076 
1077 		for (i = 0; i < ETHER_ADDR_LEN; i++)
1078 			eaddr[i] = CSR_READ_1(sc, RE_IDR0 + i);
1079 	} else {
1080 		uint16_t as[3];
1081 
1082 		/* Set RX length mask */
1083 		sc->re_rxlenmask = RE_RDESC_STAT_FRAGLEN;
1084 
1085 		sc->re_eecmd_read = RE_EECMD_READ_6BIT;
1086 		re_read_eeprom(sc, (caddr_t)&re_did, 0, 1, 0);
1087 		if (re_did != 0x8129)
1088 			sc->re_eecmd_read = RE_EECMD_READ_8BIT;
1089 
1090 		/*
1091 		 * Get station address from the EEPROM.
1092 		 */
1093 		re_read_eeprom(sc, (caddr_t)as, RE_EE_EADDR, 3, 0);
1094 		for (i = 0; i < 3; i++) {
1095 			eaddr[(i * 2) + 0] = as[i] & 0xff;
1096 			eaddr[(i * 2) + 1] = as[i] >> 8;
1097 		}
1098 	}
1099 
1100 	/*
1101 	 * Allocate the parent bus DMA tag appropriate for PCI.
1102 	 */
1103 #define RE_NSEG_NEW 32
1104 	error = bus_dma_tag_create(NULL,	/* parent */
1105 			1, 0,			/* alignment, boundary */
1106 			BUS_SPACE_MAXADDR_32BIT,/* lowaddr */
1107 			BUS_SPACE_MAXADDR,	/* highaddr */
1108 			NULL, NULL,		/* filter, filterarg */
1109 			MAXBSIZE, RE_NSEG_NEW,	/* maxsize, nsegments */
1110 			BUS_SPACE_MAXSIZE_32BIT,/* maxsegsize */
1111 			BUS_DMA_ALLOCNOW,	/* flags */
1112 			&sc->re_parent_tag);
1113 	if (error)
1114 		goto fail;
1115 
1116 	error = re_allocmem(dev, sc);
1117 
1118 	if (error)
1119 		goto fail;
1120 
1121 	/* Do MII setup */
1122 	if (mii_phy_probe(dev, &sc->re_miibus,
1123 	    re_ifmedia_upd, re_ifmedia_sts)) {
1124 		device_printf(dev, "MII without any phy!\n");
1125 		error = ENXIO;
1126 		goto fail;
1127 	}
1128 
1129 	ifp = &sc->arpcom.ac_if;
1130 	ifp->if_softc = sc;
1131 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1132 	ifp->if_mtu = ETHERMTU;
1133 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1134 	ifp->if_ioctl = re_ioctl;
1135 	ifp->if_capabilities = IFCAP_VLAN_MTU;
1136 	ifp->if_start = re_start;
1137 	ifp->if_capabilities |= IFCAP_HWCSUM|IFCAP_VLAN_HWTAGGING;
1138 #ifdef DEVICE_POLLING
1139 	ifp->if_poll = re_poll;
1140 #endif
1141 	ifp->if_watchdog = re_watchdog;
1142 	ifp->if_init = re_init;
1143 	if (sc->re_type == RE_8169)
1144 		ifp->if_baudrate = 1000000000;
1145 	else
1146 		ifp->if_baudrate = 100000000;
1147 	ifq_set_maxlen(&ifp->if_snd, RE_IFQ_MAXLEN);
1148 	ifq_set_ready(&ifp->if_snd);
1149 #ifdef RE_DISABLE_HWCSUM
1150 	ifp->if_capenable = ifp->if_capabilities & ~IFCAP_HWCSUM;
1151 	ifp->if_hwassist = 0;
1152 #else
1153 	ifp->if_capenable = ifp->if_capabilities;
1154 	ifp->if_hwassist = RE_CSUM_FEATURES;
1155 #endif
1156 
1157 	/*
1158 	 * Call MI attach routine.
1159 	 */
1160 	ether_ifattach(ifp, eaddr);
1161 
1162 	/* Perform hardware diagnostic. */
1163 	error = re_diag(sc);
1164 
1165 	if (error) {
1166 		device_printf(dev, "hardware diagnostic failure\n");
1167 		ether_ifdetach(ifp);
1168 		goto fail;
1169 	}
1170 
1171 	/* Hook interrupt last to avoid having to lock softc */
1172 	error = bus_setup_intr(dev, sc->re_irq, INTR_TYPE_NET, re_intr, sc,
1173 			       &sc->re_intrhand, NULL);
1174 
1175 	if (error) {
1176 		device_printf(dev, "couldn't set up irq\n");
1177 		ether_ifdetach(ifp);
1178 		goto fail;
1179 	}
1180 
1181 fail:
1182 	if (error)
1183 		re_detach(dev);
1184 
1185 	return (error);
1186 }
1187 
1188 /*
1189  * Shutdown hardware and free up resources. This can be called any
1190  * time after the mutex has been initialized. It is called in both
1191  * the error case in attach and the normal detach case so it needs
1192  * to be careful about only freeing resources that have actually been
1193  * allocated.
1194  */
1195 static int
1196 re_detach(device_t dev)
1197 {
1198 	struct re_softc *sc = device_get_softc(dev);
1199 	struct ifnet *ifp = &sc->arpcom.ac_if;
1200 	int i;
1201 
1202 	crit_enter();
1203 
1204 	/* These should only be active if attach succeeded */
1205 	if (device_is_attached(dev)) {
1206 		re_stop(sc);
1207 		ether_ifdetach(ifp);
1208 	}
1209 	if (sc->re_miibus)
1210 		device_delete_child(dev, sc->re_miibus);
1211 	bus_generic_detach(dev);
1212 
1213 	if (sc->re_intrhand)
1214 		bus_teardown_intr(dev, sc->re_irq, sc->re_intrhand);
1215 
1216 	crit_exit();
1217 
1218 	if (sc->re_irq)
1219 		bus_release_resource(dev, SYS_RES_IRQ, 0, sc->re_irq);
1220 	if (sc->re_res)
1221 		bus_release_resource(dev, SYS_RES_IOPORT, RE_PCI_LOIO,
1222 				     sc->re_res);
1223 
1224 	/* Unload and free the RX DMA ring memory and map */
1225 
1226 	if (sc->re_ldata.re_rx_list_tag) {
1227 		bus_dmamap_unload(sc->re_ldata.re_rx_list_tag,
1228 		    sc->re_ldata.re_rx_list_map);
1229 		bus_dmamem_free(sc->re_ldata.re_rx_list_tag,
1230 		    sc->re_ldata.re_rx_list,
1231 		    sc->re_ldata.re_rx_list_map);
1232 		bus_dma_tag_destroy(sc->re_ldata.re_rx_list_tag);
1233 	}
1234 
1235 	/* Unload and free the TX DMA ring memory and map */
1236 
1237 	if (sc->re_ldata.re_tx_list_tag) {
1238 		bus_dmamap_unload(sc->re_ldata.re_tx_list_tag,
1239 		    sc->re_ldata.re_tx_list_map);
1240 		bus_dmamem_free(sc->re_ldata.re_tx_list_tag,
1241 		    sc->re_ldata.re_tx_list,
1242 		    sc->re_ldata.re_tx_list_map);
1243 		bus_dma_tag_destroy(sc->re_ldata.re_tx_list_tag);
1244 	}
1245 
1246 	/* Destroy all the RX and TX buffer maps */
1247 
1248 	if (sc->re_ldata.re_mtag) {
1249 		for (i = 0; i < RE_TX_DESC_CNT; i++)
1250 			bus_dmamap_destroy(sc->re_ldata.re_mtag,
1251 			    sc->re_ldata.re_tx_dmamap[i]);
1252 		for (i = 0; i < RE_RX_DESC_CNT; i++)
1253 			bus_dmamap_destroy(sc->re_ldata.re_mtag,
1254 			    sc->re_ldata.re_rx_dmamap[i]);
1255 		bus_dma_tag_destroy(sc->re_ldata.re_mtag);
1256 	}
1257 
1258 	/* Unload and free the stats buffer and map */
1259 
1260 	if (sc->re_ldata.re_stag) {
1261 		bus_dmamap_unload(sc->re_ldata.re_stag,
1262 		    sc->re_ldata.re_rx_list_map);
1263 		bus_dmamem_free(sc->re_ldata.re_stag,
1264 		    sc->re_ldata.re_stats,
1265 		    sc->re_ldata.re_smap);
1266 		bus_dma_tag_destroy(sc->re_ldata.re_stag);
1267 	}
1268 
1269 	if (sc->re_parent_tag)
1270 		bus_dma_tag_destroy(sc->re_parent_tag);
1271 
1272 	return(0);
1273 }
1274 
1275 static int
1276 re_newbuf(struct re_softc *sc, int idx, struct mbuf *m)
1277 {
1278 	struct re_dmaload_arg arg;
1279 	struct mbuf *n = NULL;
1280 	int error;
1281 
1282 	if (m == NULL) {
1283 		n = m_getcl(MB_DONTWAIT, MT_DATA, M_PKTHDR);
1284 		if (n == NULL)
1285 			return(ENOBUFS);
1286 		m = n;
1287 	} else
1288 		m->m_data = m->m_ext.ext_buf;
1289 
1290 	/*
1291 	 * Initialize mbuf length fields and fixup
1292 	 * alignment so that the frame payload is
1293 	 * longword aligned.
1294 	 */
1295 	m->m_len = m->m_pkthdr.len = MCLBYTES;
1296 	m_adj(m, ETHER_ALIGN);
1297 
1298 	arg.sc = sc;
1299 	arg.re_idx = idx;
1300 	arg.re_maxsegs = 1;
1301 	arg.re_flags = 0;
1302 	arg.re_ring = sc->re_ldata.re_rx_list;
1303 
1304         error = bus_dmamap_load_mbuf(sc->re_ldata.re_mtag,
1305 	    sc->re_ldata.re_rx_dmamap[idx], m, re_dma_map_desc,
1306 	    &arg, BUS_DMA_NOWAIT);
1307 	if (error || arg.re_maxsegs != 1) {
1308 		if (n != NULL)
1309 			m_freem(n);
1310 		return (ENOMEM);
1311 	}
1312 
1313 	sc->re_ldata.re_rx_list[idx].re_cmdstat |= htole32(RE_RDESC_CMD_OWN);
1314 	sc->re_ldata.re_rx_mbuf[idx] = m;
1315 
1316         bus_dmamap_sync(sc->re_ldata.re_mtag, sc->re_ldata.re_rx_dmamap[idx],
1317 		        BUS_DMASYNC_PREREAD);
1318 
1319 	return(0);
1320 }
1321 
1322 static int
1323 re_tx_list_init(struct re_softc *sc)
1324 {
1325 	bzero(sc->re_ldata.re_tx_list, RE_TX_LIST_SZ);
1326 	bzero(&sc->re_ldata.re_tx_mbuf, RE_TX_DESC_CNT * sizeof(struct mbuf *));
1327 
1328 	bus_dmamap_sync(sc->re_ldata.re_tx_list_tag,
1329 			sc->re_ldata.re_tx_list_map, BUS_DMASYNC_PREWRITE);
1330 	sc->re_ldata.re_tx_prodidx = 0;
1331 	sc->re_ldata.re_tx_considx = 0;
1332 	sc->re_ldata.re_tx_free = RE_TX_DESC_CNT;
1333 
1334 	return(0);
1335 }
1336 
1337 static int
1338 re_rx_list_init(struct re_softc *sc)
1339 {
1340 	int i, error;
1341 
1342 	bzero(sc->re_ldata.re_rx_list, RE_RX_LIST_SZ);
1343 	bzero(&sc->re_ldata.re_rx_mbuf, RE_RX_DESC_CNT * sizeof(struct mbuf *));
1344 
1345 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
1346 		error = re_newbuf(sc, i, NULL);
1347 		if (error)
1348 			return(error);
1349 	}
1350 
1351 	/* Flush the RX descriptors */
1352 
1353 	bus_dmamap_sync(sc->re_ldata.re_rx_list_tag,
1354 	    sc->re_ldata.re_rx_list_map,
1355 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
1356 
1357 	sc->re_ldata.re_rx_prodidx = 0;
1358 	sc->re_head = sc->re_tail = NULL;
1359 
1360 	return(0);
1361 }
1362 
1363 /*
1364  * RX handler for C+ and 8169. For the gigE chips, we support
1365  * the reception of jumbo frames that have been fragmented
1366  * across multiple 2K mbuf cluster buffers.
1367  */
1368 static void
1369 re_rxeof(struct re_softc *sc)
1370 {
1371 	struct ifnet *ifp = &sc->arpcom.ac_if;
1372 	struct mbuf *m;
1373 	struct re_desc 	*cur_rx;
1374 	uint32_t rxstat, rxvlan;
1375 	int i, total_len;
1376 
1377 	/* Invalidate the descriptor memory */
1378 
1379 	bus_dmamap_sync(sc->re_ldata.re_rx_list_tag,
1380 			sc->re_ldata.re_rx_list_map, BUS_DMASYNC_POSTREAD);
1381 
1382 	for (i = sc->re_ldata.re_rx_prodidx;
1383 	     RE_OWN(&sc->re_ldata.re_rx_list[i]) == 0 ; RE_DESC_INC(i)) {
1384 		cur_rx = &sc->re_ldata.re_rx_list[i];
1385 		m = sc->re_ldata.re_rx_mbuf[i];
1386 		total_len = RE_RXBYTES(cur_rx);
1387 		rxstat = le32toh(cur_rx->re_cmdstat);
1388 		rxvlan = le32toh(cur_rx->re_vlanctl);
1389 
1390 		/* Invalidate the RX mbuf and unload its map */
1391 
1392 		bus_dmamap_sync(sc->re_ldata.re_mtag,
1393 				sc->re_ldata.re_rx_dmamap[i],
1394 				BUS_DMASYNC_POSTWRITE);
1395 		bus_dmamap_unload(sc->re_ldata.re_mtag,
1396 				  sc->re_ldata.re_rx_dmamap[i]);
1397 
1398 		if ((rxstat & RE_RDESC_STAT_EOF) == 0) {
1399 			m->m_len = MCLBYTES - ETHER_ALIGN;
1400 			if (sc->re_head == NULL) {
1401 				sc->re_head = sc->re_tail = m;
1402 			} else {
1403 				sc->re_tail->m_next = m;
1404 				sc->re_tail = m;
1405 			}
1406 			re_newbuf(sc, i, NULL);
1407 			continue;
1408 		}
1409 
1410 		/*
1411 		 * NOTE: for the 8139C+, the frame length field
1412 		 * is always 12 bits in size, but for the gigE chips,
1413 		 * it is 13 bits (since the max RX frame length is 16K).
1414 		 * Unfortunately, all 32 bits in the status word
1415 		 * were already used, so to make room for the extra
1416 		 * length bit, RealTek took out the 'frame alignment
1417 		 * error' bit and shifted the other status bits
1418 		 * over one slot. The OWN, EOR, FS and LS bits are
1419 		 * still in the same places. We have already extracted
1420 		 * the frame length and checked the OWN bit, so rather
1421 		 * than using an alternate bit mapping, we shift the
1422 		 * status bits one space to the right so we can evaluate
1423 		 * them using the 8169 status as though it was in the
1424 		 * same format as that of the 8139C+.
1425 		 */
1426 		if (sc->re_type == RE_8169)
1427 			rxstat >>= 1;
1428 
1429 		if (rxstat & RE_RDESC_STAT_RXERRSUM) {
1430 			ifp->if_ierrors++;
1431 			/*
1432 			 * If this is part of a multi-fragment packet,
1433 			 * discard all the pieces.
1434 			 */
1435 			if (sc->re_head != NULL) {
1436 				m_freem(sc->re_head);
1437 				sc->re_head = sc->re_tail = NULL;
1438 			}
1439 			re_newbuf(sc, i, m);
1440 			continue;
1441 		}
1442 
1443 		/*
1444 		 * If allocating a replacement mbuf fails,
1445 		 * reload the current one.
1446 		 */
1447 
1448 		if (re_newbuf(sc, i, NULL)) {
1449 			ifp->if_ierrors++;
1450 			if (sc->re_head != NULL) {
1451 				m_freem(sc->re_head);
1452 				sc->re_head = sc->re_tail = NULL;
1453 			}
1454 			re_newbuf(sc, i, m);
1455 			continue;
1456 		}
1457 
1458 		if (sc->re_head != NULL) {
1459 			m->m_len = total_len % (MCLBYTES - ETHER_ALIGN);
1460 			/*
1461 			 * Special case: if there's 4 bytes or less
1462 			 * in this buffer, the mbuf can be discarded:
1463 			 * the last 4 bytes is the CRC, which we don't
1464 			 * care about anyway.
1465 			 */
1466 			if (m->m_len <= ETHER_CRC_LEN) {
1467 				sc->re_tail->m_len -=
1468 				    (ETHER_CRC_LEN - m->m_len);
1469 				m_freem(m);
1470 			} else {
1471 				m->m_len -= ETHER_CRC_LEN;
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 
1585 	crit_enter();
1586 
1587 	mii = device_get_softc(sc->re_miibus);
1588 	mii_tick(mii);
1589 
1590 	callout_reset(&sc->re_timer, hz, re_tick, sc);
1591 
1592 	crit_exit();
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 
1649 	if (sc->suspended || (ifp->if_flags & IFF_UP) == 0)
1650 		return;
1651 
1652 	for (;;) {
1653 		status = CSR_READ_2(sc, RE_ISR);
1654 		/* If the card has gone away the read returns 0xffff. */
1655 		if (status == 0xffff)
1656 			break;
1657 		if (status)
1658 			CSR_WRITE_2(sc, RE_ISR, status);
1659 
1660 		if ((status & RE_INTRS_CPLUS) == 0)
1661 			break;
1662 
1663 		if (status & RE_ISR_RX_OK)
1664 			re_rxeof(sc);
1665 
1666 		if (status & RE_ISR_RX_ERR)
1667 			re_rxeof(sc);
1668 
1669 		if ((status & RE_ISR_TIMEOUT_EXPIRED) ||
1670 		    (status & RE_ISR_TX_ERR) ||
1671 		    (status & RE_ISR_TX_DESC_UNAVAIL))
1672 			re_txeof(sc);
1673 
1674 		if (status & RE_ISR_SYSTEM_ERR) {
1675 			re_reset(sc);
1676 			re_init(sc);
1677 		}
1678 
1679 		if (status & RE_ISR_LINKCHG)
1680 			re_tick(sc);
1681 	}
1682 
1683 	if (!ifq_is_empty(&ifp->if_snd))
1684 		(*ifp->if_start)(ifp);
1685 }
1686 
1687 static int
1688 re_encap(struct re_softc *sc, struct mbuf **m_head, int *idx, int *called_defrag)
1689 {
1690 	struct ifnet *ifp = &sc->arpcom.ac_if;
1691 	struct mbuf *m, *m_new = NULL;
1692 	struct re_dmaload_arg	arg;
1693 	bus_dmamap_t		map;
1694 	int			error;
1695 
1696 	*called_defrag = 0;
1697 	if (sc->re_ldata.re_tx_free <= 4)
1698 		return(EFBIG);
1699 
1700 	m = *m_head;
1701 
1702 	/*
1703 	 * Set up checksum offload. Note: checksum offload bits must
1704 	 * appear in all descriptors of a multi-descriptor transmit
1705 	 * attempt. (This is according to testing done with an 8169
1706 	 * chip. I'm not sure if this is a requirement or a bug.)
1707 	 */
1708 
1709 	arg.re_flags = 0;
1710 
1711 	if (m->m_pkthdr.csum_flags & CSUM_IP)
1712 		arg.re_flags |= RE_TDESC_CMD_IPCSUM;
1713 	if (m->m_pkthdr.csum_flags & CSUM_TCP)
1714 		arg.re_flags |= RE_TDESC_CMD_TCPCSUM;
1715 	if (m->m_pkthdr.csum_flags & CSUM_UDP)
1716 		arg.re_flags |= RE_TDESC_CMD_UDPCSUM;
1717 
1718 	arg.sc = sc;
1719 	arg.re_idx = *idx;
1720 	arg.re_maxsegs = sc->re_ldata.re_tx_free;
1721 	if (arg.re_maxsegs > 4)
1722 		arg.re_maxsegs -= 4;
1723 	arg.re_ring = sc->re_ldata.re_tx_list;
1724 
1725 	map = sc->re_ldata.re_tx_dmamap[*idx];
1726 	error = bus_dmamap_load_mbuf(sc->re_ldata.re_mtag, map,
1727 	    m, re_dma_map_desc, &arg, BUS_DMA_NOWAIT);
1728 
1729 	if (error && error != EFBIG) {
1730 		if_printf(ifp, "can't map mbuf (error %d)\n", error);
1731 		return(ENOBUFS);
1732 	}
1733 
1734 	/* Too many segments to map, coalesce into a single mbuf */
1735 
1736 	if (error || arg.re_maxsegs == 0) {
1737 		m_new = m_defrag_nofree(m, MB_DONTWAIT);
1738 		if (m_new == NULL)
1739 			return(1);
1740 		else {
1741 			m = m_new;
1742 			*m_head = m;
1743 		}
1744 
1745 		*called_defrag = 1;
1746 		arg.sc = sc;
1747 		arg.re_idx = *idx;
1748 		arg.re_maxsegs = sc->re_ldata.re_tx_free;
1749 		arg.re_ring = sc->re_ldata.re_tx_list;
1750 
1751 		error = bus_dmamap_load_mbuf(sc->re_ldata.re_mtag, map,
1752 		    m, re_dma_map_desc, &arg, BUS_DMA_NOWAIT);
1753 		if (error) {
1754 			m_freem(m);
1755 			if_printf(ifp, "can't map mbuf (error %d)\n", error);
1756 			return(EFBIG);
1757 		}
1758 	}
1759 
1760 	/*
1761 	 * Insure that the map for this transmission
1762 	 * is placed at the array index of the last descriptor
1763 	 * in this chain.
1764 	 */
1765 	sc->re_ldata.re_tx_dmamap[*idx] =
1766 	    sc->re_ldata.re_tx_dmamap[arg.re_idx];
1767 	sc->re_ldata.re_tx_dmamap[arg.re_idx] = map;
1768 
1769 	sc->re_ldata.re_tx_mbuf[arg.re_idx] = m;
1770 	sc->re_ldata.re_tx_free -= arg.re_maxsegs;
1771 
1772 	/*
1773 	 * Set up hardware VLAN tagging. Note: vlan tag info must
1774 	 * appear in the first descriptor of a multi-descriptor
1775 	 * transmission attempt.
1776 	 */
1777 
1778 	if ((m->m_flags & (M_PROTO1|M_PKTHDR)) == (M_PROTO1|M_PKTHDR) &&
1779 	    m->m_pkthdr.rcvif != NULL &&
1780 	    m->m_pkthdr.rcvif->if_type == IFT_L2VLAN) {
1781 	    	struct ifvlan *ifv;
1782 		ifv = m->m_pkthdr.rcvif->if_softc;
1783 		if (ifv != NULL)
1784 			sc->re_ldata.re_tx_list[*idx].re_vlanctl =
1785 			    htole32(htobe16(ifv->ifv_tag) | RE_TDESC_VLANCTL_TAG);
1786 	}
1787 
1788 	/* Transfer ownership of packet to the chip. */
1789 
1790 	sc->re_ldata.re_tx_list[arg.re_idx].re_cmdstat |=
1791 	    htole32(RE_TDESC_CMD_OWN);
1792 	if (*idx != arg.re_idx)
1793 		sc->re_ldata.re_tx_list[*idx].re_cmdstat |=
1794 		    htole32(RE_TDESC_CMD_OWN);
1795 
1796 	RE_DESC_INC(arg.re_idx);
1797 	*idx = arg.re_idx;
1798 
1799 	return(0);
1800 }
1801 
1802 /*
1803  * Main transmit routine for C+ and gigE NICs.
1804  */
1805 
1806 static void
1807 re_start(struct ifnet *ifp)
1808 {
1809 	struct re_softc	*sc = ifp->if_softc;
1810 	struct mbuf *m_head = NULL, *m_head2;
1811 	int called_defrag, idx;
1812 
1813 	crit_enter();
1814 
1815 	idx = sc->re_ldata.re_tx_prodidx;
1816 
1817 	while (sc->re_ldata.re_tx_mbuf[idx] == NULL) {
1818 		m_head = ifq_poll(&ifp->if_snd);
1819 		if (m_head == NULL)
1820 			break;
1821 
1822 		if (re_encap(sc, &m_head, &idx, &called_defrag)) {
1823 			if (called_defrag) {
1824 				m_head2 = ifq_dequeue(&ifp->if_snd);
1825 				m_freem(m_head2);
1826 			}
1827 			ifp->if_flags |= IFF_OACTIVE;
1828 			break;
1829 		}
1830 
1831 		m_head2 = ifq_dequeue(&ifp->if_snd);
1832 		if (called_defrag)
1833 			m_freem(m_head2);
1834 
1835 		/*
1836 		 * If there's a BPF listener, bounce a copy of this frame
1837 		 * to him.
1838 		 */
1839 		BPF_MTAP(ifp, m_head);
1840 	}
1841 
1842 	/* Flush the TX descriptors */
1843 	bus_dmamap_sync(sc->re_ldata.re_tx_list_tag,
1844 			sc->re_ldata.re_tx_list_map,
1845 			BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
1846 
1847 	sc->re_ldata.re_tx_prodidx = idx;
1848 
1849 	/*
1850 	 * RealTek put the TX poll request register in a different
1851 	 * location on the 8169 gigE chip. I don't know why.
1852 	 */
1853 	if (sc->re_type == RE_8169)
1854 		CSR_WRITE_2(sc, RE_GTXSTART, RE_TXSTART_START);
1855 	else
1856 		CSR_WRITE_2(sc, RE_TXSTART, RE_TXSTART_START);
1857 
1858 	/*
1859 	 * Use the countdown timer for interrupt moderation.
1860 	 * 'TX done' interrupts are disabled. Instead, we reset the
1861 	 * countdown timer, which will begin counting until it hits
1862 	 * the value in the TIMERINT register, and then trigger an
1863 	 * interrupt. Each time we write to the TIMERCNT register,
1864 	 * the timer count is reset to 0.
1865 	 */
1866 	CSR_WRITE_4(sc, RE_TIMERCNT, 1);
1867 
1868 	/*
1869 	 * Set a timeout in case the chip goes out to lunch.
1870 	 */
1871 	ifp->if_timer = 5;
1872 
1873 	crit_exit();
1874 }
1875 
1876 static void
1877 re_init(void *xsc)
1878 {
1879 	struct re_softc *sc = xsc;
1880 	struct ifnet *ifp = &sc->arpcom.ac_if;
1881 	struct mii_data *mii;
1882 	uint32_t rxcfg = 0;
1883 
1884 	crit_enter();
1885 
1886 	mii = device_get_softc(sc->re_miibus);
1887 
1888 	/*
1889 	 * Cancel pending I/O and free all RX/TX buffers.
1890 	 */
1891 	re_stop(sc);
1892 
1893 	/*
1894 	 * Enable C+ RX and TX mode, as well as VLAN stripping and
1895 	 * RX checksum offload. We must configure the C+ register
1896 	 * before all others.
1897 	 */
1898 	CSR_WRITE_2(sc, RE_CPLUS_CMD, RE_CPLUSCMD_RXENB | RE_CPLUSCMD_TXENB |
1899 		    RE_CPLUSCMD_PCI_MRW | RE_CPLUSCMD_VLANSTRIP |
1900 		    (ifp->if_capenable & IFCAP_RXCSUM ?
1901 		     RE_CPLUSCMD_RXCSUM_ENB : 0));
1902 
1903 	/*
1904 	 * Init our MAC address.  Even though the chipset
1905 	 * documentation doesn't mention it, we need to enter "Config
1906 	 * register write enable" mode to modify the ID registers.
1907 	 */
1908 	CSR_WRITE_1(sc, RE_EECMD, RE_EEMODE_WRITECFG);
1909 	CSR_WRITE_STREAM_4(sc, RE_IDR0,
1910 	    *(u_int32_t *)(&sc->arpcom.ac_enaddr[0]));
1911 	CSR_WRITE_STREAM_4(sc, RE_IDR4,
1912 	    *(u_int32_t *)(&sc->arpcom.ac_enaddr[4]));
1913 	CSR_WRITE_1(sc, RE_EECMD, RE_EEMODE_OFF);
1914 
1915 	/*
1916 	 * For C+ mode, initialize the RX descriptors and mbufs.
1917 	 */
1918 	re_rx_list_init(sc);
1919 	re_tx_list_init(sc);
1920 
1921 	/*
1922 	 * Enable transmit and receive.
1923 	 */
1924 	CSR_WRITE_1(sc, RE_COMMAND, RE_CMD_TX_ENB|RE_CMD_RX_ENB);
1925 
1926 	/*
1927 	 * Set the initial TX and RX configuration.
1928 	 */
1929 	if (sc->re_testmode) {
1930 		if (sc->re_type == RE_8169)
1931 			CSR_WRITE_4(sc, RE_TXCFG,
1932 				    RE_TXCFG_CONFIG | RE_LOOPTEST_ON);
1933 		else
1934 			CSR_WRITE_4(sc, RE_TXCFG,
1935 				    RE_TXCFG_CONFIG | RE_LOOPTEST_ON_CPLUS);
1936 	} else
1937 		CSR_WRITE_4(sc, RE_TXCFG, RE_TXCFG_CONFIG);
1938 	CSR_WRITE_4(sc, RE_RXCFG, RE_RXCFG_CONFIG);
1939 
1940 	/* Set the individual bit to receive frames for this host only. */
1941 	rxcfg = CSR_READ_4(sc, RE_RXCFG);
1942 	rxcfg |= RE_RXCFG_RX_INDIV;
1943 
1944 	/* If we want promiscuous mode, set the allframes bit. */
1945 	if (ifp->if_flags & IFF_PROMISC) {
1946 		rxcfg |= RE_RXCFG_RX_ALLPHYS;
1947 		CSR_WRITE_4(sc, RE_RXCFG, rxcfg);
1948 	} else {
1949 		rxcfg &= ~RE_RXCFG_RX_ALLPHYS;
1950 		CSR_WRITE_4(sc, RE_RXCFG, rxcfg);
1951 	}
1952 
1953 	/*
1954 	 * Set capture broadcast bit to capture broadcast frames.
1955 	 */
1956 	if (ifp->if_flags & IFF_BROADCAST) {
1957 		rxcfg |= RE_RXCFG_RX_BROAD;
1958 		CSR_WRITE_4(sc, RE_RXCFG, rxcfg);
1959 	} else {
1960 		rxcfg &= ~RE_RXCFG_RX_BROAD;
1961 		CSR_WRITE_4(sc, RE_RXCFG, rxcfg);
1962 	}
1963 
1964 	/*
1965 	 * Program the multicast filter, if necessary.
1966 	 */
1967 	re_setmulti(sc);
1968 
1969 #ifdef DEVICE_POLLING
1970 	/*
1971 	 * Disable interrupts if we are polling.
1972 	 */
1973 	if (ifp->if_flags & IFF_POLLING)
1974 		CSR_WRITE_2(sc, RE_IMR, 0);
1975 	else	/* otherwise ... */
1976 #endif /* DEVICE_POLLING */
1977 	/*
1978 	 * Enable interrupts.
1979 	 */
1980 	if (sc->re_testmode)
1981 		CSR_WRITE_2(sc, RE_IMR, 0);
1982 	else
1983 		CSR_WRITE_2(sc, RE_IMR, RE_INTRS_CPLUS);
1984 
1985 	/* Set initial TX threshold */
1986 	sc->re_txthresh = RE_TX_THRESH_INIT;
1987 
1988 	/* Start RX/TX process. */
1989 	CSR_WRITE_4(sc, RE_MISSEDPKT, 0);
1990 #ifdef notdef
1991 	/* Enable receiver and transmitter. */
1992 	CSR_WRITE_1(sc, RE_COMMAND, RE_CMD_TX_ENB|RE_CMD_RX_ENB);
1993 #endif
1994 	/*
1995 	 * Load the addresses of the RX and TX lists into the chip.
1996 	 */
1997 
1998 	CSR_WRITE_4(sc, RE_RXLIST_ADDR_HI,
1999 	    RE_ADDR_HI(sc->re_ldata.re_rx_list_addr));
2000 	CSR_WRITE_4(sc, RE_RXLIST_ADDR_LO,
2001 	    RE_ADDR_LO(sc->re_ldata.re_rx_list_addr));
2002 
2003 	CSR_WRITE_4(sc, RE_TXLIST_ADDR_HI,
2004 	    RE_ADDR_HI(sc->re_ldata.re_tx_list_addr));
2005 	CSR_WRITE_4(sc, RE_TXLIST_ADDR_LO,
2006 	    RE_ADDR_LO(sc->re_ldata.re_tx_list_addr));
2007 
2008 	CSR_WRITE_1(sc, RE_EARLY_TX_THRESH, 16);
2009 
2010 	/*
2011 	 * Initialize the timer interrupt register so that
2012 	 * a timer interrupt will be generated once the timer
2013 	 * reaches a certain number of ticks. The timer is
2014 	 * reloaded on each transmit. This gives us TX interrupt
2015 	 * moderation, which dramatically improves TX frame rate.
2016 	 */
2017 
2018 	if (sc->re_type == RE_8169)
2019 		CSR_WRITE_4(sc, RE_TIMERINT_8169, 0x800);
2020 	else
2021 		CSR_WRITE_4(sc, RE_TIMERINT, 0x400);
2022 
2023 	/*
2024 	 * For 8169 gigE NICs, set the max allowed RX packet
2025 	 * size so we can receive jumbo frames.
2026 	 */
2027 	if (sc->re_type == RE_8169)
2028 		CSR_WRITE_2(sc, RE_MAXRXPKTLEN, 16383);
2029 
2030 	if (sc->re_testmode) {
2031 		crit_exit();
2032 		return;
2033 	}
2034 
2035 	mii_mediachg(mii);
2036 
2037 	CSR_WRITE_1(sc, RE_CFG1, RE_CFG1_DRVLOAD|RE_CFG1_FULLDUPLEX);
2038 
2039 	ifp->if_flags |= IFF_RUNNING;
2040 	ifp->if_flags &= ~IFF_OACTIVE;
2041 
2042 	callout_reset(&sc->re_timer, hz, re_tick, sc);
2043 
2044 	crit_exit();
2045 }
2046 
2047 /*
2048  * Set media options.
2049  */
2050 static int
2051 re_ifmedia_upd(struct ifnet *ifp)
2052 {
2053 	struct re_softc *sc = ifp->if_softc;
2054 	struct mii_data *mii;
2055 
2056 	mii = device_get_softc(sc->re_miibus);
2057 	mii_mediachg(mii);
2058 
2059 	return(0);
2060 }
2061 
2062 /*
2063  * Report current media status.
2064  */
2065 static void
2066 re_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
2067 {
2068 	struct re_softc *sc = ifp->if_softc;
2069 	struct mii_data *mii;
2070 
2071 	mii = device_get_softc(sc->re_miibus);
2072 
2073 	mii_pollstat(mii);
2074 	ifmr->ifm_active = mii->mii_media_active;
2075 	ifmr->ifm_status = mii->mii_media_status;
2076 }
2077 
2078 static int
2079 re_ioctl(struct ifnet *ifp, u_long command, caddr_t data, struct ucred *cr)
2080 {
2081 	struct re_softc *sc = ifp->if_softc;
2082 	struct ifreq *ifr = (struct ifreq *) data;
2083 	struct mii_data *mii;
2084 	int error = 0;
2085 
2086 	crit_enter();
2087 
2088 	switch(command) {
2089 	case SIOCSIFMTU:
2090 		if (ifr->ifr_mtu > RE_JUMBO_MTU)
2091 			error = EINVAL;
2092 		ifp->if_mtu = ifr->ifr_mtu;
2093 		break;
2094 	case SIOCSIFFLAGS:
2095 		if (ifp->if_flags & IFF_UP)
2096 			re_init(sc);
2097 		else if (ifp->if_flags & IFF_RUNNING)
2098 				re_stop(sc);
2099 		error = 0;
2100 		break;
2101 	case SIOCADDMULTI:
2102 	case SIOCDELMULTI:
2103 		re_setmulti(sc);
2104 		error = 0;
2105 		break;
2106 	case SIOCGIFMEDIA:
2107 	case SIOCSIFMEDIA:
2108 		mii = device_get_softc(sc->re_miibus);
2109 		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, command);
2110 		break;
2111 	case SIOCSIFCAP:
2112 		ifp->if_capenable &= ~(IFCAP_HWCSUM);
2113 		ifp->if_capenable |=
2114 		    ifr->ifr_reqcap & (IFCAP_HWCSUM);
2115 		if (ifp->if_capenable & IFCAP_TXCSUM)
2116 			ifp->if_hwassist = RE_CSUM_FEATURES;
2117 		else
2118 			ifp->if_hwassist = 0;
2119 		if (ifp->if_flags & IFF_RUNNING)
2120 			re_init(sc);
2121 		break;
2122 	default:
2123 		error = ether_ioctl(ifp, command, data);
2124 		break;
2125 	}
2126 
2127 	crit_exit();
2128 
2129 	return(error);
2130 }
2131 
2132 static void
2133 re_watchdog(struct ifnet *ifp)
2134 {
2135 	struct re_softc *sc = ifp->if_softc;
2136 
2137 	if_printf(ifp, "watchdog timeout\n");
2138 
2139 	crit_enter();
2140 
2141 	ifp->if_oerrors++;
2142 
2143 	re_txeof(sc);
2144 	re_rxeof(sc);
2145 
2146 	re_init(sc);
2147 
2148 	crit_exit();
2149 }
2150 
2151 /*
2152  * Stop the adapter and free any mbufs allocated to the
2153  * RX and TX lists.
2154  */
2155 static void
2156 re_stop(struct re_softc *sc)
2157 {
2158 	struct ifnet *ifp = &sc->arpcom.ac_if;
2159 	int i;
2160 
2161 	crit_enter();
2162 
2163 	ifp->if_timer = 0;
2164 	callout_stop(&sc->re_timer);
2165 
2166 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
2167 
2168 	CSR_WRITE_1(sc, RE_COMMAND, 0x00);
2169 	CSR_WRITE_2(sc, RE_IMR, 0x0000);
2170 
2171 	if (sc->re_head != NULL) {
2172 		m_freem(sc->re_head);
2173 		sc->re_head = sc->re_tail = NULL;
2174 	}
2175 
2176 	/* Free the TX list buffers. */
2177 	for (i = 0; i < RE_TX_DESC_CNT; i++) {
2178 		if (sc->re_ldata.re_tx_mbuf[i] != NULL) {
2179 			bus_dmamap_unload(sc->re_ldata.re_mtag,
2180 					  sc->re_ldata.re_tx_dmamap[i]);
2181 			m_freem(sc->re_ldata.re_tx_mbuf[i]);
2182 			sc->re_ldata.re_tx_mbuf[i] = NULL;
2183 		}
2184 	}
2185 
2186 	/* Free the RX list buffers. */
2187 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
2188 		if (sc->re_ldata.re_rx_mbuf[i] != NULL) {
2189 			bus_dmamap_unload(sc->re_ldata.re_mtag,
2190 					  sc->re_ldata.re_rx_dmamap[i]);
2191 			m_freem(sc->re_ldata.re_rx_mbuf[i]);
2192 			sc->re_ldata.re_rx_mbuf[i] = NULL;
2193 		}
2194 	}
2195 
2196 	crit_exit();
2197 }
2198 
2199 /*
2200  * Device suspend routine.  Stop the interface and save some PCI
2201  * settings in case the BIOS doesn't restore them properly on
2202  * resume.
2203  */
2204 static int
2205 re_suspend(device_t dev)
2206 {
2207 #ifndef BURN_BRIDGES
2208 	int i;
2209 #endif
2210 	struct re_softc *sc = device_get_softc(dev);
2211 
2212 	re_stop(sc);
2213 
2214 #ifndef BURN_BRIDGES
2215 	for (i = 0; i < 5; i++)
2216 		sc->saved_maps[i] = pci_read_config(dev, PCIR_MAPS + i * 4, 4);
2217 	sc->saved_biosaddr = pci_read_config(dev, PCIR_BIOS, 4);
2218 	sc->saved_intline = pci_read_config(dev, PCIR_INTLINE, 1);
2219 	sc->saved_cachelnsz = pci_read_config(dev, PCIR_CACHELNSZ, 1);
2220 	sc->saved_lattimer = pci_read_config(dev, PCIR_LATTIMER, 1);
2221 #endif
2222 
2223 	sc->suspended = 1;
2224 
2225 	return (0);
2226 }
2227 
2228 /*
2229  * Device resume routine.  Restore some PCI settings in case the BIOS
2230  * doesn't, re-enable busmastering, and restart the interface if
2231  * appropriate.
2232  */
2233 static int
2234 re_resume(device_t dev)
2235 {
2236 	struct re_softc *sc = device_get_softc(dev);
2237 	struct ifnet *ifp = &sc->arpcom.ac_if;
2238 #ifndef BURN_BRIDGES
2239 	int i;
2240 #endif
2241 
2242 #ifndef BURN_BRIDGES
2243 	/* better way to do this? */
2244 	for (i = 0; i < 5; i++)
2245 		pci_write_config(dev, PCIR_MAPS + i * 4, sc->saved_maps[i], 4);
2246 	pci_write_config(dev, PCIR_BIOS, sc->saved_biosaddr, 4);
2247 	pci_write_config(dev, PCIR_INTLINE, sc->saved_intline, 1);
2248 	pci_write_config(dev, PCIR_CACHELNSZ, sc->saved_cachelnsz, 1);
2249 	pci_write_config(dev, PCIR_LATTIMER, sc->saved_lattimer, 1);
2250 
2251 	/* reenable busmastering */
2252 	pci_enable_busmaster(dev);
2253 	pci_enable_io(dev, SYS_RES_IOPORT);
2254 #endif
2255 
2256 	/* reinitialize interface if necessary */
2257 	if (ifp->if_flags & IFF_UP)
2258 		re_init(sc);
2259 
2260 	sc->suspended = 0;
2261 
2262 	return (0);
2263 }
2264 
2265 /*
2266  * Stop all chip I/O so that the kernel's probe routines don't
2267  * get confused by errant DMAs when rebooting.
2268  */
2269 static void
2270 re_shutdown(device_t dev)
2271 {
2272 	struct re_softc *sc = device_get_softc(dev);
2273 
2274 	re_stop(sc);
2275 }
2276