xref: /dflybsd-src/sys/dev/netif/re/if_re.c (revision 744c01d0dc2aa1481a40e5b0988d15691602f5c9)
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 1.30 2007/01/15 12:53:26 sephe Exp $
37  */
38 
39 /*
40  * RealTek 8139C+/8169/8169S/8110S/8168/8111/8101E 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  * seven devices in this family: the RTL8139C+, the RTL8169, the RTL8169S,
51  * RTL8110S, the RTL8168, the RTL8111 and the RTL8101E.
52  *
53  * The 8139C+ is a 10/100 ethernet chip. It is backwards compatible
54  * with the older 8139 family, however it also supports a special
55  * C+ mode of operation that provides several new performance enhancing
56  * features. These include:
57  *
58  *	o Descriptor based DMA mechanism. Each descriptor represents
59  *	  a single packet fragment. Data buffers may be aligned on
60  *	  any byte boundary.
61  *
62  *	o 64-bit DMA
63  *
64  *	o TCP/IP checksum offload for both RX and TX
65  *
66  *	o High and normal priority transmit DMA rings
67  *
68  *	o VLAN tag insertion and extraction
69  *
70  *	o TCP large send (segmentation offload)
71  *
72  * Like the 8139, the 8139C+ also has a built-in 10/100 PHY. The C+
73  * programming API is fairly straightforward. The RX filtering, EEPROM
74  * access and PHY access is the same as it is on the older 8139 series
75  * chips.
76  *
77  * The 8169 is a 64-bit 10/100/1000 gigabit ethernet MAC. It has almost the
78  * same programming API and feature set as the 8139C+ with the following
79  * differences and additions:
80  *
81  *	o 1000Mbps mode
82  *
83  *	o Jumbo frames
84  *
85  * 	o GMII and TBI ports/registers for interfacing with copper
86  *	  or fiber PHYs
87  *
88  *      o RX and TX DMA rings can have up to 1024 descriptors
89  *        (the 8139C+ allows a maximum of 64)
90  *
91  *	o Slight differences in register layout from the 8139C+
92  *
93  * The TX start and timer interrupt registers are at different locations
94  * on the 8169 than they are on the 8139C+. Also, the status word in the
95  * RX descriptor has a slightly different bit layout. The 8169 does not
96  * have a built-in PHY. Most reference boards use a Marvell 88E1000 'Alaska'
97  * copper gigE PHY.
98  *
99  * The 8169S/8110S 10/100/1000 devices have built-in copper gigE PHYs
100  * (the 'S' stands for 'single-chip'). These devices have the same
101  * programming API as the older 8169, but also have some vendor-specific
102  * registers for the on-board PHY. The 8110S is a LAN-on-motherboard
103  * part designed to be pin-compatible with the RealTek 8100 10/100 chip.
104  *
105  * This driver takes advantage of the RX and TX checksum offload and
106  * VLAN tag insertion/extraction features. It also implements TX
107  * interrupt moderation using the timer interrupt registers, which
108  * significantly reduces TX interrupt load. There is also support
109  * for jumbo frames, however the 8169/8169S/8110S can not transmit
110  * jumbo frames larger than 7440, so the max MTU possible with this
111  * driver is 7422 bytes.
112  */
113 
114 #include "opt_polling.h"
115 
116 #include <sys/param.h>
117 #include <sys/bus.h>
118 #include <sys/endian.h>
119 #include <sys/kernel.h>
120 #include <sys/malloc.h>
121 #include <sys/mbuf.h>
122 /* #include <sys/module.h> */
123 #include <sys/rman.h>
124 #include <sys/serialize.h>
125 #include <sys/socket.h>
126 #include <sys/sockio.h>
127 #include <sys/sysctl.h>
128 
129 #include <net/bpf.h>
130 #include <net/ethernet.h>
131 #include <net/if.h>
132 #include <net/ifq_var.h>
133 #include <net/if_arp.h>
134 #include <net/if_dl.h>
135 #include <net/if_media.h>
136 #include <net/if_types.h>
137 #include <net/vlan/if_vlan_var.h>
138 
139 #include <dev/netif/mii_layer/mii.h>
140 #include <dev/netif/mii_layer/miivar.h>
141 
142 #include <bus/pci/pcidevs.h>
143 #include <bus/pci/pcireg.h>
144 #include <bus/pci/pcivar.h>
145 
146 /* "device miibus" required.  See GENERIC if you get errors here. */
147 #include "miibus_if.h"
148 
149 #include <dev/netif/re/if_rereg.h>
150 #include <dev/netif/re/if_revar.h>
151 
152 #define RE_CSUM_FEATURES    (CSUM_IP | CSUM_TCP | CSUM_UDP)
153 #if 0
154 #define RE_DISABLE_HWCSUM
155 #endif
156 
157 /*
158  * Various supported device vendors/types and their names.
159  */
160 static const struct re_type re_devs[] = {
161 	{ PCI_VENDOR_DLINK, PCI_PRODUCT_DLINK_DGE528T, RE_HWREV_8169S,
162 		"D-Link DGE-528(T) Gigabit Ethernet Adapter" },
163 	{ PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8139, RE_HWREV_8139CPLUS,
164 		"RealTek 8139C+ 10/100BaseTX" },
165 	{ PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8101E, RE_HWREV_8101E,
166 		"RealTek 8101E PCIe 10/100baseTX" },
167 	{ PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8168, RE_HWREV_8168_SPIN1,
168 		"RealTek 8168/8111B PCIe Gigabit Ethernet" },
169 	{ PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8168, RE_HWREV_8168_SPIN2,
170 		"RealTek 8168/8111B PCIe Gigabit Ethernet" },
171 	{ PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8169, RE_HWREV_8169,
172 		"RealTek 8169 Gigabit Ethernet" },
173 	{ PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8169, RE_HWREV_8169S,
174 		"RealTek 8169S Single-chip Gigabit Ethernet" },
175 	{ PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8169, RE_HWREV_8169_8110SB,
176 		"RealTek 8169SB/8110SB Single-chip Gigabit Ethernet" },
177 	{ PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8169SC, RE_HWREV_8169_8110SC,
178 		"RealTek 8169SC/8110SC Single-chip Gigabit Ethernet" },
179 	{ PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8169, RE_HWREV_8110S,
180 		"RealTek 8110S Single-chip Gigabit Ethernet" },
181 	{ PCI_VENDOR_COREGA, PCI_PRODUCT_COREGA_CG_LAPCIGT, RE_HWREV_8169S,
182 		"Corega CG-LAPCIGT Gigabit Ethernet" },
183 	{ PCI_VENDOR_LINKSYS, PCI_PRODUCT_LINKSYS_EG1032, RE_HWREV_8169S,
184 		"Linksys EG1032 Gigabit Ethernet" },
185 	{ PCI_VENDOR_USR2, PCI_PRODUCT_USR2_997902, RE_HWREV_8169S,
186 		"US Robotics 997902 Gigabit Ethernet" },
187 	{ 0, 0, 0, NULL }
188 };
189 
190 static const struct re_hwrev re_hwrevs[] = {
191 	{ RE_HWREV_8139CPLUS,	RE_8139CPLUS,	RE_F_HASMPC,	"C+" },
192 	{ RE_HWREV_8168_SPIN1,	RE_8169,	RE_F_PCIE,	"8168" },
193 	{ RE_HWREV_8168_SPIN2,	RE_8169,	RE_F_PCIE,	"8168" },
194 	{ RE_HWREV_8169,	RE_8169,	RE_F_HASMPC,	"8169" },
195 	{ RE_HWREV_8169S,	RE_8169,	RE_F_HASMPC,	"8169S" },
196 	{ RE_HWREV_8110S,	RE_8169,	RE_F_HASMPC,	"8110S" },
197 	{ RE_HWREV_8169_8110SB,	RE_8169,	RE_F_HASMPC,	"8169SB" },
198 	{ RE_HWREV_8169_8110SC,	RE_8169,	0,		"8169SC" },
199 	{ RE_HWREV_8100E,	RE_8169,	RE_F_HASMPC,	"8100E" },
200 	{ RE_HWREV_8101E,	RE_8169,	RE_F_PCIE,	"8101E" },
201 	{ 0, 0, 0, NULL }
202 };
203 
204 static int	re_probe(device_t);
205 static int	re_attach(device_t);
206 static int	re_detach(device_t);
207 
208 static int	re_encap(struct re_softc *, struct mbuf **, int *, int *);
209 
210 static void	re_dma_map_addr(void *, bus_dma_segment_t *, int, int);
211 static void	re_dma_map_desc(void *, bus_dma_segment_t *, int,
212 				bus_size_t, int);
213 static int	re_allocmem(device_t, struct re_softc *);
214 static int	re_newbuf(struct re_softc *, int, struct mbuf *);
215 static int	re_rx_list_init(struct re_softc *);
216 static int	re_tx_list_init(struct re_softc *);
217 static void	re_rxeof(struct re_softc *);
218 static void	re_txeof(struct re_softc *);
219 static void	re_intr(void *);
220 static void	re_tick(void *);
221 static void	re_tick_serialized(void *);
222 static void	re_start(struct ifnet *);
223 static int	re_ioctl(struct ifnet *, u_long, caddr_t, struct ucred *);
224 static void	re_init(void *);
225 static void	re_stop(struct re_softc *);
226 static void	re_watchdog(struct ifnet *);
227 static int	re_suspend(device_t);
228 static int	re_resume(device_t);
229 static void	re_shutdown(device_t);
230 static int	re_ifmedia_upd(struct ifnet *);
231 static void	re_ifmedia_sts(struct ifnet *, struct ifmediareq *);
232 
233 static void	re_eeprom_putbyte(struct re_softc *, int);
234 static void	re_eeprom_getword(struct re_softc *, int, u_int16_t *);
235 static void	re_read_eeprom(struct re_softc *, caddr_t, int, int);
236 static int	re_gmii_readreg(device_t, int, int);
237 static int	re_gmii_writereg(device_t, int, int, int);
238 
239 static int	re_miibus_readreg(device_t, int, int);
240 static int	re_miibus_writereg(device_t, int, int, int);
241 static void	re_miibus_statchg(device_t);
242 
243 static void	re_setmulti(struct re_softc *);
244 static void	re_reset(struct re_softc *);
245 
246 #ifdef RE_DIAG
247 static int	re_diag(struct re_softc *);
248 #endif
249 
250 #ifdef DEVICE_POLLING
251 static void	re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count);
252 #endif
253 
254 static int	re_sysctl_tx_moderation(SYSCTL_HANDLER_ARGS);
255 
256 static device_method_t re_methods[] = {
257 	/* Device interface */
258 	DEVMETHOD(device_probe,		re_probe),
259 	DEVMETHOD(device_attach,	re_attach),
260 	DEVMETHOD(device_detach,	re_detach),
261 	DEVMETHOD(device_suspend,	re_suspend),
262 	DEVMETHOD(device_resume,	re_resume),
263 	DEVMETHOD(device_shutdown,	re_shutdown),
264 
265 	/* bus interface */
266 	DEVMETHOD(bus_print_child,	bus_generic_print_child),
267 	DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
268 
269 	/* MII interface */
270 	DEVMETHOD(miibus_readreg,	re_miibus_readreg),
271 	DEVMETHOD(miibus_writereg,	re_miibus_writereg),
272 	DEVMETHOD(miibus_statchg,	re_miibus_statchg),
273 
274 	{ 0, 0 }
275 };
276 
277 static driver_t re_driver = {
278 	"re",
279 	re_methods,
280 	sizeof(struct re_softc)
281 };
282 
283 static devclass_t re_devclass;
284 
285 DECLARE_DUMMY_MODULE(if_re);
286 DRIVER_MODULE(if_re, pci, re_driver, re_devclass, 0, 0);
287 DRIVER_MODULE(if_re, cardbus, re_driver, re_devclass, 0, 0);
288 DRIVER_MODULE(miibus, re, miibus_driver, miibus_devclass, 0, 0);
289 
290 #define EE_SET(x)	\
291 	CSR_WRITE_1(sc, RE_EECMD, CSR_READ_1(sc, RE_EECMD) | (x))
292 
293 #define EE_CLR(x)	\
294 	CSR_WRITE_1(sc, RE_EECMD, CSR_READ_1(sc, RE_EECMD) & ~(x))
295 
296 /*
297  * Send a read command and address to the EEPROM, check for ACK.
298  */
299 static void
300 re_eeprom_putbyte(struct re_softc *sc, int addr)
301 {
302 	int d, i;
303 
304 	d = addr | (RE_9346_READ << sc->re_eewidth);
305 
306 	/*
307 	 * Feed in each bit and strobe the clock.
308 	 */
309 	for (i = 1 << (sc->re_eewidth + 3); i; i >>= 1) {
310 		if (d & i)
311 			EE_SET(RE_EE_DATAIN);
312 		else
313 			EE_CLR(RE_EE_DATAIN);
314 		DELAY(100);
315 		EE_SET(RE_EE_CLK);
316 		DELAY(150);
317 		EE_CLR(RE_EE_CLK);
318 		DELAY(100);
319 	}
320 }
321 
322 /*
323  * Read a word of data stored in the EEPROM at address 'addr.'
324  */
325 static void
326 re_eeprom_getword(struct re_softc *sc, int addr, uint16_t *dest)
327 {
328 	int i;
329 	uint16_t word = 0;
330 
331 	/*
332 	 * Send address of word we want to read.
333 	 */
334 	re_eeprom_putbyte(sc, addr);
335 
336 	/*
337 	 * Start reading bits from EEPROM.
338 	 */
339 	for (i = 0x8000; i != 0; i >>= 1) {
340 		EE_SET(RE_EE_CLK);
341 		DELAY(100);
342 		if (CSR_READ_1(sc, RE_EECMD) & RE_EE_DATAOUT)
343 			word |= i;
344 		EE_CLR(RE_EE_CLK);
345 		DELAY(100);
346 	}
347 
348 	*dest = word;
349 }
350 
351 /*
352  * Read a sequence of words from the EEPROM.
353  */
354 static void
355 re_read_eeprom(struct re_softc *sc, caddr_t dest, int off, int cnt)
356 {
357 	int i;
358 	uint16_t word = 0, *ptr;
359 
360 	CSR_SETBIT_1(sc, RE_EECMD, RE_EEMODE_PROGRAM);
361 	DELAY(100);
362 
363 	for (i = 0; i < cnt; i++) {
364 		CSR_SETBIT_1(sc, RE_EECMD, RE_EE_SEL);
365 		re_eeprom_getword(sc, off + i, &word);
366 		CSR_CLRBIT_1(sc, RE_EECMD, RE_EE_SEL);
367 		ptr = (uint16_t *)(dest + (i * 2));
368 		*ptr = word;
369 	}
370 
371 	CSR_CLRBIT_1(sc, RE_EECMD, RE_EEMODE_PROGRAM);
372 }
373 
374 static int
375 re_gmii_readreg(device_t dev, int phy, int reg)
376 {
377 	struct re_softc *sc = device_get_softc(dev);
378 	u_int32_t rval;
379 	int i;
380 
381 	if (phy != 1)
382 		return(0);
383 
384 	/* Let the rgephy driver read the GMEDIASTAT register */
385 
386 	if (reg == RE_GMEDIASTAT)
387 		return(CSR_READ_1(sc, RE_GMEDIASTAT));
388 
389 	CSR_WRITE_4(sc, RE_PHYAR, reg << 16);
390 	DELAY(1000);
391 
392 	for (i = 0; i < RE_TIMEOUT; i++) {
393 		rval = CSR_READ_4(sc, RE_PHYAR);
394 		if (rval & RE_PHYAR_BUSY)
395 			break;
396 		DELAY(100);
397 	}
398 
399 	if (i == RE_TIMEOUT) {
400 		device_printf(dev, "PHY read failed\n");
401 		return(0);
402 	}
403 
404 	return(rval & RE_PHYAR_PHYDATA);
405 }
406 
407 static int
408 re_gmii_writereg(device_t dev, int phy, int reg, int data)
409 {
410 	struct re_softc *sc = device_get_softc(dev);
411 	uint32_t rval;
412 	int i;
413 
414 	CSR_WRITE_4(sc, RE_PHYAR,
415 		    (reg << 16) | (data & RE_PHYAR_PHYDATA) | RE_PHYAR_BUSY);
416 	DELAY(1000);
417 
418 	for (i = 0; i < RE_TIMEOUT; i++) {
419 		rval = CSR_READ_4(sc, RE_PHYAR);
420 		if ((rval & RE_PHYAR_BUSY) == 0)
421 			break;
422 		DELAY(100);
423 	}
424 
425 	if (i == RE_TIMEOUT)
426 		device_printf(dev, "PHY write failed\n");
427 
428 	return(0);
429 }
430 
431 static int
432 re_miibus_readreg(device_t dev, int phy, int reg)
433 {
434 	struct re_softc	*sc = device_get_softc(dev);
435 	uint16_t rval = 0;
436 	uint16_t re8139_reg = 0;
437 
438 	if (sc->re_type == RE_8169) {
439 		rval = re_gmii_readreg(dev, phy, reg);
440 		return(rval);
441 	}
442 
443 	/* Pretend the internal PHY is only at address 0 */
444 	if (phy)
445 		return(0);
446 
447 	switch(reg) {
448 	case MII_BMCR:
449 		re8139_reg = RE_BMCR;
450 		break;
451 	case MII_BMSR:
452 		re8139_reg = RE_BMSR;
453 		break;
454 	case MII_ANAR:
455 		re8139_reg = RE_ANAR;
456 		break;
457 	case MII_ANER:
458 		re8139_reg = RE_ANER;
459 		break;
460 	case MII_ANLPAR:
461 		re8139_reg = RE_LPAR;
462 		break;
463 	case MII_PHYIDR1:
464 	case MII_PHYIDR2:
465 		return(0);
466 	/*
467 	 * Allow the rlphy driver to read the media status
468 	 * register. If we have a link partner which does not
469 	 * support NWAY, this is the register which will tell
470 	 * us the results of parallel detection.
471 	 */
472 	case RE_MEDIASTAT:
473 		return(CSR_READ_1(sc, RE_MEDIASTAT));
474 	default:
475 		device_printf(dev, "bad phy register\n");
476 		return(0);
477 	}
478 	rval = CSR_READ_2(sc, re8139_reg);
479 	if (sc->re_type == RE_8139CPLUS && re8139_reg == RE_BMCR) {
480 		/* 8139C+ has different bit layout. */
481 		rval &= ~(BMCR_LOOP | BMCR_ISO);
482 	}
483 	return(rval);
484 }
485 
486 static int
487 re_miibus_writereg(device_t dev, int phy, int reg, int data)
488 {
489 	struct re_softc *sc= device_get_softc(dev);
490 	u_int16_t re8139_reg = 0;
491 
492 	if (sc->re_type == RE_8169)
493 		return(re_gmii_writereg(dev, phy, reg, data));
494 
495 	/* Pretend the internal PHY is only at address 0 */
496 	if (phy)
497 		return(0);
498 
499 	switch(reg) {
500 	case MII_BMCR:
501 		re8139_reg = RE_BMCR;
502 		if (sc->re_type == RE_8139CPLUS) {
503 			/* 8139C+ has different bit layout. */
504 			data &= ~(BMCR_LOOP | BMCR_ISO);
505 		}
506 		break;
507 	case MII_BMSR:
508 		re8139_reg = RE_BMSR;
509 		break;
510 	case MII_ANAR:
511 		re8139_reg = RE_ANAR;
512 		break;
513 	case MII_ANER:
514 		re8139_reg = RE_ANER;
515 		break;
516 	case MII_ANLPAR:
517 		re8139_reg = RE_LPAR;
518 		break;
519 	case MII_PHYIDR1:
520 	case MII_PHYIDR2:
521 		return(0);
522 	default:
523 		device_printf(dev, "bad phy register\n");
524 		return(0);
525 	}
526 	CSR_WRITE_2(sc, re8139_reg, data);
527 	return(0);
528 }
529 
530 static void
531 re_miibus_statchg(device_t dev)
532 {
533 }
534 
535 /*
536  * Program the 64-bit multicast hash filter.
537  */
538 static void
539 re_setmulti(struct re_softc *sc)
540 {
541 	struct ifnet *ifp = &sc->arpcom.ac_if;
542 	int h = 0;
543 	uint32_t hashes[2] = { 0, 0 };
544 	struct ifmultiaddr *ifma;
545 	uint32_t rxfilt;
546 	int mcnt = 0;
547 
548 	rxfilt = CSR_READ_4(sc, RE_RXCFG);
549 
550 	if (ifp->if_flags & IFF_ALLMULTI || ifp->if_flags & IFF_PROMISC) {
551 		rxfilt |= RE_RXCFG_RX_MULTI;
552 		CSR_WRITE_4(sc, RE_RXCFG, rxfilt);
553 		CSR_WRITE_4(sc, RE_MAR0, 0xFFFFFFFF);
554 		CSR_WRITE_4(sc, RE_MAR4, 0xFFFFFFFF);
555 		return;
556 	}
557 
558 	/* first, zot all the existing hash bits */
559 	CSR_WRITE_4(sc, RE_MAR0, 0);
560 	CSR_WRITE_4(sc, RE_MAR4, 0);
561 
562 	/* now program new ones */
563 	LIST_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
564 		if (ifma->ifma_addr->sa_family != AF_LINK)
565 			continue;
566 		h = ether_crc32_be(LLADDR((struct sockaddr_dl *)
567 		    ifma->ifma_addr), ETHER_ADDR_LEN) >> 26;
568 		if (h < 32)
569 			hashes[0] |= (1 << h);
570 		else
571 			hashes[1] |= (1 << (h - 32));
572 		mcnt++;
573 	}
574 
575 	if (mcnt)
576 		rxfilt |= RE_RXCFG_RX_MULTI;
577 	else
578 		rxfilt &= ~RE_RXCFG_RX_MULTI;
579 
580 	CSR_WRITE_4(sc, RE_RXCFG, rxfilt);
581 	CSR_WRITE_4(sc, RE_MAR0, hashes[0]);
582 	CSR_WRITE_4(sc, RE_MAR4, hashes[1]);
583 }
584 
585 static void
586 re_reset(struct re_softc *sc)
587 {
588 	int i;
589 
590 	CSR_WRITE_1(sc, RE_COMMAND, RE_CMD_RESET);
591 
592 	for (i = 0; i < RE_TIMEOUT; i++) {
593 		DELAY(10);
594 		if ((CSR_READ_1(sc, RE_COMMAND) & RE_CMD_RESET) == 0)
595 			break;
596 	}
597 	if (i == RE_TIMEOUT)
598 		if_printf(&sc->arpcom.ac_if, "reset never completed!\n");
599 
600 	CSR_WRITE_1(sc, 0x82, 1);
601 }
602 
603 #ifdef RE_DIAG
604 /*
605  * The following routine is designed to test for a defect on some
606  * 32-bit 8169 cards. Some of these NICs have the REQ64# and ACK64#
607  * lines connected to the bus, however for a 32-bit only card, they
608  * should be pulled high. The result of this defect is that the
609  * NIC will not work right if you plug it into a 64-bit slot: DMA
610  * operations will be done with 64-bit transfers, which will fail
611  * because the 64-bit data lines aren't connected.
612  *
613  * There's no way to work around this (short of talking a soldering
614  * iron to the board), however we can detect it. The method we use
615  * here is to put the NIC into digital loopback mode, set the receiver
616  * to promiscuous mode, and then try to send a frame. We then compare
617  * the frame data we sent to what was received. If the data matches,
618  * then the NIC is working correctly, otherwise we know the user has
619  * a defective NIC which has been mistakenly plugged into a 64-bit PCI
620  * slot. In the latter case, there's no way the NIC can work correctly,
621  * so we print out a message on the console and abort the device attach.
622  */
623 
624 static int
625 re_diag(struct re_softc *sc)
626 {
627 	struct ifnet *ifp = &sc->arpcom.ac_if;
628 	struct mbuf *m0;
629 	struct ether_header *eh;
630 	struct re_desc *cur_rx;
631 	uint16_t status;
632 	uint32_t rxstat;
633 	int total_len, i, error = 0, phyaddr;
634 	uint8_t dst[ETHER_ADDR_LEN] = { 0x00, 'h', 'e', 'l', 'l', 'o' };
635 	uint8_t src[ETHER_ADDR_LEN] = { 0x00, 'w', 'o', 'r', 'l', 'd' };
636 
637 	/* Allocate a single mbuf */
638 
639 	MGETHDR(m0, MB_DONTWAIT, MT_DATA);
640 	if (m0 == NULL)
641 		return(ENOBUFS);
642 
643 	/*
644 	 * Initialize the NIC in test mode. This sets the chip up
645 	 * so that it can send and receive frames, but performs the
646 	 * following special functions:
647 	 * - Puts receiver in promiscuous mode
648 	 * - Enables digital loopback mode
649 	 * - Leaves interrupts turned off
650 	 */
651 
652 	ifp->if_flags |= IFF_PROMISC;
653 	sc->re_testmode = 1;
654 	re_reset(sc);
655 	re_init(sc);
656 	sc->re_link = 1;
657 	if (sc->re_type == RE_8169)
658 		phyaddr = 1;
659 	else
660 		phyaddr = 0;
661 
662 	re_miibus_writereg(sc->re_dev, phyaddr, MII_BMCR, BMCR_RESET);
663 	for (i = 0; i < RE_TIMEOUT; i++) {
664 		status = re_miibus_readreg(sc->re_dev, phyaddr, MII_BMCR);
665 		if (!(status & BMCR_RESET))
666 			break;
667 	}
668 
669 	re_miibus_writereg(sc->re_dev, phyaddr, MII_BMCR, BMCR_LOOP);
670 	CSR_WRITE_2(sc, RE_ISR, RE_INTRS_DIAG);
671 
672 	DELAY(100000);
673 
674 	/* Put some data in the mbuf */
675 
676 	eh = mtod(m0, struct ether_header *);
677 	bcopy (dst, eh->ether_dhost, ETHER_ADDR_LEN);
678 	bcopy (src, eh->ether_shost, ETHER_ADDR_LEN);
679 	eh->ether_type = htons(ETHERTYPE_IP);
680 	m0->m_pkthdr.len = m0->m_len = ETHER_MIN_LEN - ETHER_CRC_LEN;
681 
682 	/*
683 	 * Queue the packet, start transmission.
684 	 * Note: ifq_handoff() ultimately calls re_start() for us.
685 	 */
686 
687 	CSR_WRITE_2(sc, RE_ISR, 0xFFFF);
688 	error = ifq_handoff(ifp, m0, NULL);
689 	if (error) {
690 		m0 = NULL;
691 		goto done;
692 	}
693 	m0 = NULL;
694 
695 	/* Wait for it to propagate through the chip */
696 
697 	DELAY(100000);
698 	for (i = 0; i < RE_TIMEOUT; i++) {
699 		status = CSR_READ_2(sc, RE_ISR);
700 		CSR_WRITE_2(sc, RE_ISR, status);
701 		if ((status & (RE_ISR_TIMEOUT_EXPIRED|RE_ISR_RX_OK)) ==
702 		    (RE_ISR_TIMEOUT_EXPIRED|RE_ISR_RX_OK))
703 			break;
704 		DELAY(10);
705 	}
706 
707 	if (i == RE_TIMEOUT) {
708 		if_printf(ifp, "diagnostic failed to receive packet "
709 			  "in loopback mode\n");
710 		error = EIO;
711 		goto done;
712 	}
713 
714 	/*
715 	 * The packet should have been dumped into the first
716 	 * entry in the RX DMA ring. Grab it from there.
717 	 */
718 
719 	bus_dmamap_sync(sc->re_ldata.re_rx_list_tag,
720 			sc->re_ldata.re_rx_list_map, BUS_DMASYNC_POSTREAD);
721 	bus_dmamap_sync(sc->re_ldata.re_mtag, sc->re_ldata.re_rx_dmamap[0],
722 			BUS_DMASYNC_POSTWRITE);
723 	bus_dmamap_unload(sc->re_ldata.re_mtag, sc->re_ldata.re_rx_dmamap[0]);
724 
725 	m0 = sc->re_ldata.re_rx_mbuf[0];
726 	sc->re_ldata.re_rx_mbuf[0] = NULL;
727 	eh = mtod(m0, struct ether_header *);
728 
729 	cur_rx = &sc->re_ldata.re_rx_list[0];
730 	total_len = RE_RXBYTES(cur_rx);
731 	rxstat = le32toh(cur_rx->re_cmdstat);
732 
733 	if (total_len != ETHER_MIN_LEN) {
734 		if_printf(ifp, "diagnostic failed, received short packet\n");
735 		error = EIO;
736 		goto done;
737 	}
738 
739 	/* Test that the received packet data matches what we sent. */
740 
741 	if (bcmp(eh->ether_dhost, dst, ETHER_ADDR_LEN) ||
742 	    bcmp(eh->ether_shost, &src, ETHER_ADDR_LEN) ||
743 	    be16toh(eh->ether_type) != ETHERTYPE_IP) {
744 		if_printf(ifp, "WARNING, DMA FAILURE!\n");
745 		if_printf(ifp, "expected TX data: %6D/%6D/0x%x\n",
746 		    dst, ":", src, ":", ETHERTYPE_IP);
747 		if_printf(ifp, "received RX data: %6D/%6D/0x%x\n",
748 		    eh->ether_dhost, ":",  eh->ether_shost, ":",
749 		    ntohs(eh->ether_type));
750 		if_printf(ifp, "You may have a defective 32-bit NIC plugged "
751 		    "into a 64-bit PCI slot.\n");
752 		if_printf(ifp, "Please re-install the NIC in a 32-bit slot "
753 		    "for proper operation.\n");
754 		if_printf(ifp, "Read the re(4) man page for more details.\n");
755 		error = EIO;
756 	}
757 
758 done:
759 	/* Turn interface off, release resources */
760 
761 	sc->re_testmode = 0;
762 	sc->re_link = 0;
763 	ifp->if_flags &= ~IFF_PROMISC;
764 	re_stop(sc);
765 	if (m0 != NULL)
766 		m_freem(m0);
767 
768 	return (error);
769 }
770 #endif	/* RE_DIAG */
771 
772 /*
773  * Probe for a RealTek 8139C+/8169/8110 chip. Check the PCI vendor and device
774  * IDs against our list and return a device name if we find a match.
775  */
776 static int
777 re_probe(device_t dev)
778 {
779 	const struct re_type *t;
780 	struct re_softc *sc;
781 	int rid;
782 	uint32_t hwrev;
783 	uint16_t vendor, product;
784 
785 	t = re_devs;
786 
787 	vendor = pci_get_vendor(dev);
788 	product = pci_get_device(dev);
789 
790 	/*
791 	 * Only attach to rev.3 of the Linksys EG1032 adapter.
792 	 * Rev.2 is supported by sk(4).
793 	 */
794 	if (vendor == PCI_VENDOR_LINKSYS &&
795 	    product == PCI_PRODUCT_LINKSYS_EG1032 &&
796 	    pci_get_subdevice(dev) != PCI_SUBDEVICE_LINKSYS_EG1032_REV3)
797 			return ENXIO;
798 
799 	for (t = re_devs; t->re_name != NULL; t++) {
800 		if (product == t->re_did && vendor == t->re_vid)
801 			break;
802 	}
803 
804 	/*
805 	 * Check if we found a RealTek device.
806 	 */
807 	if (t->re_name == NULL)
808 		return(ENXIO);
809 
810 	/*
811 	 * Temporarily map the I/O space so we can read the chip ID register.
812 	 */
813 	sc = kmalloc(sizeof(*sc), M_TEMP, M_WAITOK | M_ZERO);
814 	rid = RE_PCI_LOIO;
815 	sc->re_res = bus_alloc_resource_any(dev, SYS_RES_IOPORT, &rid,
816 					    RF_ACTIVE);
817 	if (sc->re_res == NULL) {
818 		device_printf(dev, "couldn't map ports/memory\n");
819 		kfree(sc, M_TEMP);
820 		return(ENXIO);
821 	}
822 
823 	sc->re_btag = rman_get_bustag(sc->re_res);
824 	sc->re_bhandle = rman_get_bushandle(sc->re_res);
825 
826 	hwrev = CSR_READ_4(sc, RE_TXCFG) & RE_TXCFG_HWREV;
827 	bus_release_resource(dev, SYS_RES_IOPORT, RE_PCI_LOIO, sc->re_res);
828 	kfree(sc, M_TEMP);
829 
830 	/*
831 	 * and continue matching for the specific chip...
832 	 */
833 	for (; t->re_name != NULL; t++) {
834 		if (product == t->re_did && vendor == t->re_vid &&
835 		    t->re_basetype == hwrev) {
836 			device_set_desc(dev, t->re_name);
837 			return(0);
838 		}
839 	}
840 	return(ENXIO);
841 }
842 
843 /*
844  * This routine takes the segment list provided as the result of
845  * a bus_dma_map_load() operation and assigns the addresses/lengths
846  * to RealTek DMA descriptors. This can be called either by the RX
847  * code or the TX code. In the RX case, we'll probably wind up mapping
848  * at most one segment. For the TX case, there could be any number of
849  * segments since TX packets may span multiple mbufs. In either case,
850  * if the number of segments is larger than the re_maxsegs limit
851  * specified by the caller, we abort the mapping operation. Sadly,
852  * whoever designed the buffer mapping API did not provide a way to
853  * return an error from here, so we have to fake it a bit.
854  */
855 
856 static void
857 re_dma_map_desc(void *arg, bus_dma_segment_t *segs, int nseg,
858 		bus_size_t mapsize, int error)
859 {
860 	struct re_dmaload_arg *ctx;
861 	struct re_desc *d = NULL;
862 	int i = 0, idx;
863 	uint32_t cmdstat;
864 
865 	if (error)
866 		return;
867 
868 	ctx = arg;
869 
870 	/* Signal error to caller if there's too many segments */
871 	if (nseg > ctx->re_maxsegs) {
872 		ctx->re_maxsegs = 0;
873 		return;
874 	}
875 
876 	/*
877 	 * Map the segment array into descriptors. Note that we set the
878 	 * start-of-frame and end-of-frame markers for either TX or RX, but
879 	 * they really only have meaning in the TX case. (In the RX case,
880 	 * it's the chip that tells us where packets begin and end.)
881 	 * We also keep track of the end of the ring and set the
882 	 * end-of-ring bits as needed, and we set the ownership bits
883 	 * in all except the very first descriptor. (The caller will
884 	 * set this descriptor later when it start transmission or
885 	 * reception.)
886 	 */
887 	idx = ctx->re_idx;
888 	for (;;) {
889 		d = &ctx->re_ring[idx];
890 		if (le32toh(d->re_cmdstat) & RE_RDESC_STAT_OWN) {
891 			ctx->re_maxsegs = 0;
892 			return;
893 		}
894 		cmdstat = segs[i].ds_len;
895 		d->re_bufaddr_lo = htole32(RE_ADDR_LO(segs[i].ds_addr));
896 		d->re_bufaddr_hi = htole32(RE_ADDR_HI(segs[i].ds_addr));
897 		if (i == 0)
898 			cmdstat |= RE_TDESC_CMD_SOF;
899 		else
900 			cmdstat |= RE_TDESC_CMD_OWN;
901 		if (idx == (RE_RX_DESC_CNT - 1))
902 			cmdstat |= RE_TDESC_CMD_EOR;
903 		d->re_cmdstat = htole32(cmdstat | ctx->re_flags);
904 		i++;
905 		if (i == nseg)
906 			break;
907 		RE_DESC_INC(idx);
908 	}
909 
910 	d->re_cmdstat |= htole32(RE_TDESC_CMD_EOF);
911 	ctx->re_maxsegs = nseg;
912 	ctx->re_idx = idx;
913 }
914 
915 /*
916  * Map a single buffer address.
917  */
918 
919 static void
920 re_dma_map_addr(void *arg, bus_dma_segment_t *segs, int nseg, int error)
921 {
922 	uint32_t *addr;
923 
924 	if (error)
925 		return;
926 
927 	KASSERT(nseg == 1, ("too many DMA segments, %d should be 1", nseg));
928 	addr = arg;
929 	*addr = segs->ds_addr;
930 }
931 
932 static int
933 re_allocmem(device_t dev, struct re_softc *sc)
934 {
935 	int error, i, nseg;
936 
937 	/*
938 	 * Allocate map for RX mbufs.
939 	 */
940 	nseg = 32;
941 	error = bus_dma_tag_create(sc->re_parent_tag, ETHER_ALIGN, 0,
942 	    BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
943 	    NULL, MCLBYTES * nseg, nseg, MCLBYTES, BUS_DMA_ALLOCNOW,
944 	    &sc->re_ldata.re_mtag);
945 	if (error) {
946 		device_printf(dev, "could not allocate dma tag\n");
947 		return(error);
948 	}
949 
950 	/*
951 	 * Allocate map for TX descriptor list.
952 	 */
953 	error = bus_dma_tag_create(sc->re_parent_tag, RE_RING_ALIGN,
954 	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
955             NULL, RE_TX_LIST_SZ, 1, RE_TX_LIST_SZ, BUS_DMA_ALLOCNOW,
956 	    &sc->re_ldata.re_tx_list_tag);
957 	if (error) {
958 		device_printf(dev, "could not allocate dma tag\n");
959 		return(error);
960 	}
961 
962 	/* Allocate DMA'able memory for the TX ring */
963 
964         error = bus_dmamem_alloc(sc->re_ldata.re_tx_list_tag,
965 	    (void **)&sc->re_ldata.re_tx_list, BUS_DMA_WAITOK | BUS_DMA_ZERO,
966             &sc->re_ldata.re_tx_list_map);
967         if (error) {
968 		device_printf(dev, "could not allocate TX ring\n");
969                 return(error);
970 	}
971 
972 	/* Load the map for the TX ring. */
973 
974 	error = bus_dmamap_load(sc->re_ldata.re_tx_list_tag,
975 	     sc->re_ldata.re_tx_list_map, sc->re_ldata.re_tx_list,
976 	     RE_TX_LIST_SZ, re_dma_map_addr,
977 	     &sc->re_ldata.re_tx_list_addr, BUS_DMA_NOWAIT);
978 	if (error) {
979 		device_printf(dev, "could not get addres of TX ring\n");
980 		return(error);
981 	}
982 
983 	/* Create DMA maps for TX buffers */
984 
985 	for (i = 0; i < RE_TX_DESC_CNT; i++) {
986 		error = bus_dmamap_create(sc->re_ldata.re_mtag, 0,
987 			    &sc->re_ldata.re_tx_dmamap[i]);
988 		if (error) {
989 			device_printf(dev, "can't create DMA map for TX\n");
990 			return(error);
991 		}
992 	}
993 
994 	/*
995 	 * Allocate map for RX descriptor list.
996 	 */
997 	error = bus_dma_tag_create(sc->re_parent_tag, RE_RING_ALIGN,
998 	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
999             NULL, RE_RX_LIST_SZ, 1, RE_RX_LIST_SZ, BUS_DMA_ALLOCNOW,
1000 	    &sc->re_ldata.re_rx_list_tag);
1001 	if (error) {
1002 		device_printf(dev, "could not allocate dma tag\n");
1003 		return(error);
1004 	}
1005 
1006 	/* Allocate DMA'able memory for the RX ring */
1007 
1008         error = bus_dmamem_alloc(sc->re_ldata.re_rx_list_tag,
1009 	    (void **)&sc->re_ldata.re_rx_list, BUS_DMA_WAITOK | BUS_DMA_ZERO,
1010             &sc->re_ldata.re_rx_list_map);
1011         if (error) {
1012 		device_printf(dev, "could not allocate RX ring\n");
1013                 return(error);
1014 	}
1015 
1016 	/* Load the map for the RX ring. */
1017 
1018 	error = bus_dmamap_load(sc->re_ldata.re_rx_list_tag,
1019 	     sc->re_ldata.re_rx_list_map, sc->re_ldata.re_rx_list,
1020 	     RE_RX_LIST_SZ, re_dma_map_addr,
1021 	     &sc->re_ldata.re_rx_list_addr, BUS_DMA_NOWAIT);
1022 	if (error) {
1023 		device_printf(dev, "could not get address of RX ring\n");
1024 		return(error);
1025 	}
1026 
1027 	/* Create DMA maps for RX buffers */
1028 
1029 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
1030 		error = bus_dmamap_create(sc->re_ldata.re_mtag, 0,
1031 			    &sc->re_ldata.re_rx_dmamap[i]);
1032 		if (error) {
1033 			device_printf(dev, "can't create DMA map for RX\n");
1034 			return(ENOMEM);
1035 		}
1036 	}
1037 
1038 	return(0);
1039 }
1040 
1041 /*
1042  * Attach the interface. Allocate softc structures, do ifmedia
1043  * setup and ethernet/BPF attach.
1044  */
1045 static int
1046 re_attach(device_t dev)
1047 {
1048 	struct re_softc	*sc = device_get_softc(dev);
1049 	struct ifnet *ifp;
1050 	const struct re_hwrev *hw_rev;
1051 	uint8_t eaddr[ETHER_ADDR_LEN];
1052 	uint16_t as[ETHER_ADDR_LEN / 2];
1053 	uint16_t re_did = 0;
1054 	uint32_t hwrev;
1055 	int error = 0, rid, i;
1056 
1057 	callout_init(&sc->re_timer);
1058 #ifdef RE_DIAG
1059 	sc->re_dev = dev;
1060 #endif
1061 
1062 	RE_ENABLE_TX_MODERATION(sc);
1063 
1064 	sysctl_ctx_init(&sc->re_sysctl_ctx);
1065 	sc->re_sysctl_tree = SYSCTL_ADD_NODE(&sc->re_sysctl_ctx,
1066 					     SYSCTL_STATIC_CHILDREN(_hw),
1067 					     OID_AUTO,
1068 					     device_get_nameunit(dev),
1069 					     CTLFLAG_RD, 0, "");
1070 	if (sc->re_sysctl_tree == NULL) {
1071 		device_printf(dev, "can't add sysctl node\n");
1072 		error = ENXIO;
1073 		goto fail;
1074 	}
1075 	SYSCTL_ADD_PROC(&sc->re_sysctl_ctx,
1076 			SYSCTL_CHILDREN(sc->re_sysctl_tree),
1077 			OID_AUTO, "tx_moderation",
1078 			CTLTYPE_INT | CTLFLAG_RW,
1079 			sc, 0, re_sysctl_tx_moderation, "I",
1080 			"Enable/Disable TX moderation");
1081 
1082 #ifndef BURN_BRIDGES
1083 	/*
1084 	 * Handle power management nonsense.
1085 	 */
1086 
1087 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D0) {
1088 		uint32_t membase, irq;
1089 
1090 		/* Save important PCI config data. */
1091 		membase = pci_read_config(dev, RE_PCI_LOMEM, 4);
1092 		irq = pci_read_config(dev, PCIR_INTLINE, 4);
1093 
1094 		/* Reset the power state. */
1095 		device_printf(dev, "chip is is in D%d power mode "
1096 		    "-- setting to D0\n", pci_get_powerstate(dev));
1097 
1098 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
1099 
1100 		/* Restore PCI config data. */
1101 		pci_write_config(dev, RE_PCI_LOMEM, membase, 4);
1102 		pci_write_config(dev, PCIR_INTLINE, irq, 4);
1103 	}
1104 #endif
1105 	/*
1106 	 * Map control/status registers.
1107 	 */
1108 	pci_enable_busmaster(dev);
1109 
1110 	rid = RE_PCI_LOIO;
1111 	sc->re_res = bus_alloc_resource_any(dev, SYS_RES_IOPORT, &rid,
1112 					    RF_ACTIVE);
1113 
1114 	if (sc->re_res == NULL) {
1115 		device_printf(dev, "couldn't map ports\n");
1116 		error = ENXIO;
1117 		goto fail;
1118 	}
1119 
1120 	sc->re_btag = rman_get_bustag(sc->re_res);
1121 	sc->re_bhandle = rman_get_bushandle(sc->re_res);
1122 
1123 	/* Allocate interrupt */
1124 	rid = 0;
1125 	sc->re_irq = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
1126 					    RF_SHAREABLE | RF_ACTIVE);
1127 
1128 	if (sc->re_irq == NULL) {
1129 		device_printf(dev, "couldn't map interrupt\n");
1130 		error = ENXIO;
1131 		goto fail;
1132 	}
1133 
1134 	/* Reset the adapter. */
1135 	re_reset(sc);
1136 
1137 	hwrev = CSR_READ_4(sc, RE_TXCFG) & RE_TXCFG_HWREV;
1138 	for (hw_rev = re_hwrevs; hw_rev->re_desc != NULL; hw_rev++) {
1139 		if (hw_rev->re_rev == hwrev) {
1140 			sc->re_type = hw_rev->re_type;
1141 			sc->re_flags = hw_rev->re_flags;
1142 			break;
1143 		}
1144 	}
1145 
1146 	sc->re_eewidth = 6;
1147 	re_read_eeprom(sc, (caddr_t)&re_did, 0, 1);
1148 	if (re_did != 0x8129)
1149 	        sc->re_eewidth = 8;
1150 
1151 	/*
1152 	 * Get station address from the EEPROM.
1153 	 */
1154 	re_read_eeprom(sc, (caddr_t)as, RE_EE_EADDR, 3);
1155 	for (i = 0; i < ETHER_ADDR_LEN / 2; i++)
1156 		as[i] = le16toh(as[i]);
1157 	bcopy(as, eaddr, sizeof(eaddr));
1158 
1159 	if (sc->re_type == RE_8169) {
1160 		/* Set RX length mask */
1161 		sc->re_rxlenmask = RE_RDESC_STAT_GFRAGLEN;
1162 		sc->re_txstart = RE_GTXSTART;
1163 	} else {
1164 		/* Set RX length mask */
1165 		sc->re_rxlenmask = RE_RDESC_STAT_FRAGLEN;
1166 		sc->re_txstart = RE_TXSTART;
1167 	}
1168 
1169 	/*
1170 	 * Allocate the parent bus DMA tag appropriate for PCI.
1171 	 */
1172 #define RE_NSEG_NEW 32
1173 	error = bus_dma_tag_create(NULL,	/* parent */
1174 			1, 0,			/* alignment, boundary */
1175 			BUS_SPACE_MAXADDR_32BIT,/* lowaddr */
1176 			BUS_SPACE_MAXADDR,	/* highaddr */
1177 			NULL, NULL,		/* filter, filterarg */
1178 			MAXBSIZE, RE_NSEG_NEW,	/* maxsize, nsegments */
1179 			BUS_SPACE_MAXSIZE_32BIT,/* maxsegsize */
1180 			BUS_DMA_ALLOCNOW,	/* flags */
1181 			&sc->re_parent_tag);
1182 	if (error)
1183 		goto fail;
1184 
1185 	error = re_allocmem(dev, sc);
1186 
1187 	if (error)
1188 		goto fail;
1189 
1190 	/* Do MII setup */
1191 	if (mii_phy_probe(dev, &sc->re_miibus,
1192 	    re_ifmedia_upd, re_ifmedia_sts)) {
1193 		device_printf(dev, "MII without any phy!\n");
1194 		error = ENXIO;
1195 		goto fail;
1196 	}
1197 
1198 	ifp = &sc->arpcom.ac_if;
1199 	ifp->if_softc = sc;
1200 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1201 	ifp->if_mtu = ETHERMTU;
1202 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1203 	ifp->if_ioctl = re_ioctl;
1204 	ifp->if_capabilities = IFCAP_VLAN_MTU;
1205 	ifp->if_start = re_start;
1206 	ifp->if_capabilities |= IFCAP_HWCSUM|IFCAP_VLAN_HWTAGGING;
1207 #ifdef DEVICE_POLLING
1208 	ifp->if_poll = re_poll;
1209 #endif
1210 	ifp->if_watchdog = re_watchdog;
1211 	ifp->if_init = re_init;
1212 	if (sc->re_type == RE_8169)
1213 		ifp->if_baudrate = 1000000000;
1214 	else
1215 		ifp->if_baudrate = 100000000;
1216 	ifq_set_maxlen(&ifp->if_snd, RE_IFQ_MAXLEN);
1217 	ifq_set_ready(&ifp->if_snd);
1218 
1219 #ifdef RE_DISABLE_HWCSUM
1220 	ifp->if_capenable = ifp->if_capabilities & ~IFCAP_HWCSUM;
1221 	ifp->if_hwassist = 0;
1222 #else
1223 	if (sc->re_flags & RE_F_PCIE) {
1224 		/*
1225 		 * The hardware supports checksumming but, as usual, PCIe
1226 		 * chipsets screw it all up and produce bogus packets, so
1227 		 * we don't enable it by default.
1228 		 */
1229 		ifp->if_capenable = ifp->if_capabilities & ~IFCAP_HWCSUM;
1230 		ifp->if_hwassist = 0;
1231 	} else {
1232 		ifp->if_capenable = ifp->if_capabilities;
1233 		ifp->if_hwassist = RE_CSUM_FEATURES;
1234 	}
1235 #endif	/* RE_DISABLE_HWCSUM */
1236 
1237 	/*
1238 	 * Call MI attach routine.
1239 	 */
1240 	ether_ifattach(ifp, eaddr, NULL);
1241 
1242 #ifdef RE_DIAG
1243 	/*
1244 	 * Perform hardware diagnostic on the original RTL8169.
1245 	 * Some 32-bit cards were incorrectly wired and would
1246 	 * malfunction if plugged into a 64-bit slot.
1247 	 */
1248 	if (hwrev == RE_HWREV_8169) {
1249 		lwkt_serialize_enter(ifp->if_serializer);
1250 		error = re_diag(sc);
1251 		lwkt_serialize_exit(ifp->if_serializer);
1252 
1253 		if (error) {
1254 			device_printf(dev, "hardware diagnostic failure\n");
1255 			ether_ifdetach(ifp);
1256 			goto fail;
1257 		}
1258 	}
1259 #endif	/* RE_DIAG */
1260 
1261 	/* Hook interrupt last to avoid having to lock softc */
1262 	error = bus_setup_intr(dev, sc->re_irq, INTR_NETSAFE, re_intr, sc,
1263 			       &sc->re_intrhand, ifp->if_serializer);
1264 
1265 	if (error) {
1266 		device_printf(dev, "couldn't set up irq\n");
1267 		ether_ifdetach(ifp);
1268 		goto fail;
1269 	}
1270 
1271 fail:
1272 	if (error)
1273 		re_detach(dev);
1274 
1275 	return (error);
1276 }
1277 
1278 /*
1279  * Shutdown hardware and free up resources. This can be called any
1280  * time after the mutex has been initialized. It is called in both
1281  * the error case in attach and the normal detach case so it needs
1282  * to be careful about only freeing resources that have actually been
1283  * allocated.
1284  */
1285 static int
1286 re_detach(device_t dev)
1287 {
1288 	struct re_softc *sc = device_get_softc(dev);
1289 	struct ifnet *ifp = &sc->arpcom.ac_if;
1290 	int i;
1291 
1292 	/* These should only be active if attach succeeded */
1293 	if (device_is_attached(dev)) {
1294 		lwkt_serialize_enter(ifp->if_serializer);
1295 		re_stop(sc);
1296 		bus_teardown_intr(dev, sc->re_irq, sc->re_intrhand);
1297 		lwkt_serialize_exit(ifp->if_serializer);
1298 
1299 		ether_ifdetach(ifp);
1300 	}
1301 	if (sc->re_miibus)
1302 		device_delete_child(dev, sc->re_miibus);
1303 	bus_generic_detach(dev);
1304 
1305 	if (sc->re_irq)
1306 		bus_release_resource(dev, SYS_RES_IRQ, 0, sc->re_irq);
1307 	if (sc->re_res) {
1308 		bus_release_resource(dev, SYS_RES_IOPORT, RE_PCI_LOIO,
1309 				     sc->re_res);
1310 	}
1311 
1312 	/* Unload and free the RX DMA ring memory and map */
1313 
1314 	if (sc->re_ldata.re_rx_list_tag) {
1315 		bus_dmamap_unload(sc->re_ldata.re_rx_list_tag,
1316 		    sc->re_ldata.re_rx_list_map);
1317 		bus_dmamem_free(sc->re_ldata.re_rx_list_tag,
1318 		    sc->re_ldata.re_rx_list,
1319 		    sc->re_ldata.re_rx_list_map);
1320 		bus_dma_tag_destroy(sc->re_ldata.re_rx_list_tag);
1321 	}
1322 
1323 	/* Unload and free the TX DMA ring memory and map */
1324 
1325 	if (sc->re_ldata.re_tx_list_tag) {
1326 		bus_dmamap_unload(sc->re_ldata.re_tx_list_tag,
1327 		    sc->re_ldata.re_tx_list_map);
1328 		bus_dmamem_free(sc->re_ldata.re_tx_list_tag,
1329 		    sc->re_ldata.re_tx_list,
1330 		    sc->re_ldata.re_tx_list_map);
1331 		bus_dma_tag_destroy(sc->re_ldata.re_tx_list_tag);
1332 	}
1333 
1334 	/* Destroy all the RX and TX buffer maps */
1335 
1336 	if (sc->re_ldata.re_mtag) {
1337 		for (i = 0; i < RE_TX_DESC_CNT; i++)
1338 			bus_dmamap_destroy(sc->re_ldata.re_mtag,
1339 			    sc->re_ldata.re_tx_dmamap[i]);
1340 		for (i = 0; i < RE_RX_DESC_CNT; i++)
1341 			bus_dmamap_destroy(sc->re_ldata.re_mtag,
1342 			    sc->re_ldata.re_rx_dmamap[i]);
1343 		bus_dma_tag_destroy(sc->re_ldata.re_mtag);
1344 	}
1345 
1346 	/* Unload and free the stats buffer and map */
1347 
1348 	if (sc->re_ldata.re_stag) {
1349 		bus_dmamap_unload(sc->re_ldata.re_stag,
1350 		    sc->re_ldata.re_rx_list_map);
1351 		bus_dmamem_free(sc->re_ldata.re_stag,
1352 		    sc->re_ldata.re_stats,
1353 		    sc->re_ldata.re_smap);
1354 		bus_dma_tag_destroy(sc->re_ldata.re_stag);
1355 	}
1356 
1357 	if (sc->re_parent_tag)
1358 		bus_dma_tag_destroy(sc->re_parent_tag);
1359 
1360 	return(0);
1361 }
1362 
1363 static int
1364 re_newbuf(struct re_softc *sc, int idx, struct mbuf *m)
1365 {
1366 	struct re_dmaload_arg arg;
1367 	struct mbuf *n = NULL;
1368 	int error;
1369 
1370 	if (m == NULL) {
1371 		n = m_getcl(MB_DONTWAIT, MT_DATA, M_PKTHDR);
1372 		if (n == NULL)
1373 			return(ENOBUFS);
1374 		m = n;
1375 	} else
1376 		m->m_data = m->m_ext.ext_buf;
1377 
1378 	m->m_len = m->m_pkthdr.len = MCLBYTES;
1379 
1380 	/*
1381 	 * NOTE:
1382 	 * Some re(4) chips(e.g. RTL8101E) need address of the receive buffer
1383 	 * to be 8-byte aligned, so don't call m_adj(m, ETHER_ALIGN) here.
1384 	 */
1385 
1386 	arg.sc = sc;
1387 	arg.re_idx = idx;
1388 	arg.re_maxsegs = 1;
1389 	arg.re_flags = 0;
1390 	arg.re_ring = sc->re_ldata.re_rx_list;
1391 
1392         error = bus_dmamap_load_mbuf(sc->re_ldata.re_mtag,
1393 	    sc->re_ldata.re_rx_dmamap[idx], m, re_dma_map_desc,
1394 	    &arg, BUS_DMA_NOWAIT);
1395 	if (error || arg.re_maxsegs != 1) {
1396 		if (n != NULL)
1397 			m_freem(n);
1398 		return (ENOMEM);
1399 	}
1400 
1401 	sc->re_ldata.re_rx_list[idx].re_cmdstat |= htole32(RE_RDESC_CMD_OWN);
1402 	sc->re_ldata.re_rx_mbuf[idx] = m;
1403 
1404         bus_dmamap_sync(sc->re_ldata.re_mtag, sc->re_ldata.re_rx_dmamap[idx],
1405 		        BUS_DMASYNC_PREREAD);
1406 
1407 	return(0);
1408 }
1409 
1410 static int
1411 re_tx_list_init(struct re_softc *sc)
1412 {
1413 	bzero(sc->re_ldata.re_tx_list, RE_TX_LIST_SZ);
1414 	bzero(&sc->re_ldata.re_tx_mbuf, RE_TX_DESC_CNT * sizeof(struct mbuf *));
1415 
1416 	bus_dmamap_sync(sc->re_ldata.re_tx_list_tag,
1417 			sc->re_ldata.re_tx_list_map, BUS_DMASYNC_PREWRITE);
1418 	sc->re_ldata.re_tx_prodidx = 0;
1419 	sc->re_ldata.re_tx_considx = 0;
1420 	sc->re_ldata.re_tx_free = RE_TX_DESC_CNT;
1421 
1422 	return(0);
1423 }
1424 
1425 static int
1426 re_rx_list_init(struct re_softc *sc)
1427 {
1428 	int i, error;
1429 
1430 	bzero(sc->re_ldata.re_rx_list, RE_RX_LIST_SZ);
1431 	bzero(&sc->re_ldata.re_rx_mbuf, RE_RX_DESC_CNT * sizeof(struct mbuf *));
1432 
1433 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
1434 		error = re_newbuf(sc, i, NULL);
1435 		if (error)
1436 			return(error);
1437 	}
1438 
1439 	/* Flush the RX descriptors */
1440 
1441 	bus_dmamap_sync(sc->re_ldata.re_rx_list_tag,
1442 			sc->re_ldata.re_rx_list_map, BUS_DMASYNC_PREWRITE);
1443 
1444 	sc->re_ldata.re_rx_prodidx = 0;
1445 	sc->re_head = sc->re_tail = NULL;
1446 
1447 	return(0);
1448 }
1449 
1450 /*
1451  * RX handler for C+ and 8169. For the gigE chips, we support
1452  * the reception of jumbo frames that have been fragmented
1453  * across multiple 2K mbuf cluster buffers.
1454  */
1455 static void
1456 re_rxeof(struct re_softc *sc)
1457 {
1458 	struct ifnet *ifp = &sc->arpcom.ac_if;
1459 	struct mbuf *m;
1460 	struct re_desc 	*cur_rx;
1461 	uint32_t rxstat, rxvlan;
1462 	int i, total_len;
1463 
1464 	/* Invalidate the descriptor memory */
1465 
1466 	bus_dmamap_sync(sc->re_ldata.re_rx_list_tag,
1467 			sc->re_ldata.re_rx_list_map, BUS_DMASYNC_POSTREAD);
1468 
1469 	for (i = sc->re_ldata.re_rx_prodidx;
1470 	     RE_OWN(&sc->re_ldata.re_rx_list[i]) == 0 ; RE_DESC_INC(i)) {
1471 		cur_rx = &sc->re_ldata.re_rx_list[i];
1472 		m = sc->re_ldata.re_rx_mbuf[i];
1473 		total_len = RE_RXBYTES(cur_rx);
1474 		rxstat = le32toh(cur_rx->re_cmdstat);
1475 		rxvlan = le32toh(cur_rx->re_vlanctl);
1476 
1477 		/* Invalidate the RX mbuf and unload its map */
1478 
1479 		bus_dmamap_sync(sc->re_ldata.re_mtag,
1480 				sc->re_ldata.re_rx_dmamap[i],
1481 				BUS_DMASYNC_POSTWRITE);
1482 		bus_dmamap_unload(sc->re_ldata.re_mtag,
1483 				  sc->re_ldata.re_rx_dmamap[i]);
1484 
1485 		if ((rxstat & RE_RDESC_STAT_EOF) == 0) {
1486 			m->m_len = MCLBYTES - ETHER_ALIGN;
1487 			if (sc->re_head == NULL) {
1488 				sc->re_head = sc->re_tail = m;
1489 			} else {
1490 				sc->re_tail->m_next = m;
1491 				sc->re_tail = m;
1492 			}
1493 			re_newbuf(sc, i, NULL);
1494 			continue;
1495 		}
1496 
1497 		/*
1498 		 * NOTE: for the 8139C+, the frame length field
1499 		 * is always 12 bits in size, but for the gigE chips,
1500 		 * it is 13 bits (since the max RX frame length is 16K).
1501 		 * Unfortunately, all 32 bits in the status word
1502 		 * were already used, so to make room for the extra
1503 		 * length bit, RealTek took out the 'frame alignment
1504 		 * error' bit and shifted the other status bits
1505 		 * over one slot. The OWN, EOR, FS and LS bits are
1506 		 * still in the same places. We have already extracted
1507 		 * the frame length and checked the OWN bit, so rather
1508 		 * than using an alternate bit mapping, we shift the
1509 		 * status bits one space to the right so we can evaluate
1510 		 * them using the 8169 status as though it was in the
1511 		 * same format as that of the 8139C+.
1512 		 */
1513 		if (sc->re_type == RE_8169)
1514 			rxstat >>= 1;
1515 
1516 		if (rxstat & RE_RDESC_STAT_RXERRSUM) {
1517 			ifp->if_ierrors++;
1518 			/*
1519 			 * If this is part of a multi-fragment packet,
1520 			 * discard all the pieces.
1521 			 */
1522 			if (sc->re_head != NULL) {
1523 				m_freem(sc->re_head);
1524 				sc->re_head = sc->re_tail = NULL;
1525 			}
1526 			re_newbuf(sc, i, m);
1527 			continue;
1528 		}
1529 
1530 		/*
1531 		 * If allocating a replacement mbuf fails,
1532 		 * reload the current one.
1533 		 */
1534 
1535 		if (re_newbuf(sc, i, NULL)) {
1536 			ifp->if_ierrors++;
1537 			if (sc->re_head != NULL) {
1538 				m_freem(sc->re_head);
1539 				sc->re_head = sc->re_tail = NULL;
1540 			}
1541 			re_newbuf(sc, i, m);
1542 			continue;
1543 		}
1544 
1545 		if (sc->re_head != NULL) {
1546 			m->m_len = total_len % (MCLBYTES - ETHER_ALIGN);
1547 			/*
1548 			 * Special case: if there's 4 bytes or less
1549 			 * in this buffer, the mbuf can be discarded:
1550 			 * the last 4 bytes is the CRC, which we don't
1551 			 * care about anyway.
1552 			 */
1553 			if (m->m_len <= ETHER_CRC_LEN) {
1554 				sc->re_tail->m_len -=
1555 				    (ETHER_CRC_LEN - m->m_len);
1556 				m_freem(m);
1557 			} else {
1558 				m->m_len -= ETHER_CRC_LEN;
1559 				sc->re_tail->m_next = m;
1560 			}
1561 			m = sc->re_head;
1562 			sc->re_head = sc->re_tail = NULL;
1563 			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
1564 		} else
1565 			m->m_pkthdr.len = m->m_len =
1566 			    (total_len - ETHER_CRC_LEN);
1567 
1568 		ifp->if_ipackets++;
1569 		m->m_pkthdr.rcvif = ifp;
1570 
1571 		/* Do RX checksumming if enabled */
1572 
1573 		if (ifp->if_capenable & IFCAP_RXCSUM) {
1574 
1575 			/* Check IP header checksum */
1576 			if (rxstat & RE_RDESC_STAT_PROTOID)
1577 				m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED;
1578 			if ((rxstat & RE_RDESC_STAT_IPSUMBAD) == 0)
1579 				m->m_pkthdr.csum_flags |= CSUM_IP_VALID;
1580 
1581 			/* Check TCP/UDP checksum */
1582 			if ((RE_TCPPKT(rxstat) &&
1583 			    (rxstat & RE_RDESC_STAT_TCPSUMBAD) == 0) ||
1584 			    (RE_UDPPKT(rxstat) &&
1585 			    (rxstat & RE_RDESC_STAT_UDPSUMBAD)) == 0) {
1586 				m->m_pkthdr.csum_flags |=
1587 				    CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
1588 				m->m_pkthdr.csum_data = 0xffff;
1589 			}
1590 		}
1591 
1592 		if (rxvlan & RE_RDESC_VLANCTL_TAG) {
1593 			VLAN_INPUT_TAG(m,
1594 			   be16toh((rxvlan & RE_RDESC_VLANCTL_DATA)));
1595 		} else {
1596 			ifp->if_input(ifp, m);
1597 		}
1598 	}
1599 
1600 	/* Flush the RX DMA ring */
1601 
1602 	bus_dmamap_sync(sc->re_ldata.re_rx_list_tag,
1603 			sc->re_ldata.re_rx_list_map, BUS_DMASYNC_PREWRITE);
1604 
1605 	sc->re_ldata.re_rx_prodidx = i;
1606 }
1607 
1608 static void
1609 re_txeof(struct re_softc *sc)
1610 {
1611 	struct ifnet *ifp = &sc->arpcom.ac_if;
1612 	uint32_t txstat;
1613 	int idx;
1614 
1615 	/* Invalidate the TX descriptor list */
1616 
1617 	bus_dmamap_sync(sc->re_ldata.re_tx_list_tag,
1618 			sc->re_ldata.re_tx_list_map, BUS_DMASYNC_POSTREAD);
1619 
1620 	for (idx = sc->re_ldata.re_tx_considx;
1621 	     sc->re_ldata.re_tx_free < RE_TX_DESC_CNT; RE_DESC_INC(idx)) {
1622 		txstat = le32toh(sc->re_ldata.re_tx_list[idx].re_cmdstat);
1623 		if (txstat & RE_TDESC_CMD_OWN)
1624 			break;
1625 
1626 		sc->re_ldata.re_tx_list[idx].re_bufaddr_lo = 0;
1627 
1628 		/*
1629 		 * We only stash mbufs in the last descriptor
1630 		 * in a fragment chain, which also happens to
1631 		 * be the only place where the TX status bits
1632 		 * are valid.
1633 		 */
1634 		if (txstat & RE_TDESC_CMD_EOF) {
1635 			m_freem(sc->re_ldata.re_tx_mbuf[idx]);
1636 			sc->re_ldata.re_tx_mbuf[idx] = NULL;
1637 			bus_dmamap_unload(sc->re_ldata.re_mtag,
1638 			    sc->re_ldata.re_tx_dmamap[idx]);
1639 			if (txstat & (RE_TDESC_STAT_EXCESSCOL|
1640 			    RE_TDESC_STAT_COLCNT))
1641 				ifp->if_collisions++;
1642 			if (txstat & RE_TDESC_STAT_TXERRSUM)
1643 				ifp->if_oerrors++;
1644 			else
1645 				ifp->if_opackets++;
1646 		}
1647 		sc->re_ldata.re_tx_free++;
1648 	}
1649 
1650 	/* No changes made to the TX ring, so no flush needed */
1651 	if (sc->re_ldata.re_tx_free) {
1652 		sc->re_ldata.re_tx_considx = idx;
1653 		ifp->if_flags &= ~IFF_OACTIVE;
1654 		ifp->if_timer = 0;
1655 	}
1656 
1657 	/*
1658 	 * Some chips will ignore a second TX request issued while an
1659 	 * existing transmission is in progress. If the transmitter goes
1660 	 * idle but there are still packets waiting to be sent, we need
1661 	 * to restart the channel here to flush them out. This only seems
1662 	 * to be required with the PCIe devices.
1663 	 */
1664 	if (sc->re_ldata.re_tx_free < RE_TX_DESC_CNT)
1665 		CSR_WRITE_1(sc, sc->re_txstart, RE_TXSTART_START);
1666 
1667 	/*
1668 	 * If not all descriptors have been released reaped yet,
1669 	 * reload the timer so that we will eventually get another
1670 	 * interrupt that will cause us to re-enter this routine.
1671 	 * This is done in case the transmitter has gone idle.
1672 	 */
1673 	if (RE_TX_MODERATION_IS_ENABLED(sc) &&
1674 	    sc->re_ldata.re_tx_free < RE_TX_DESC_CNT)
1675                 CSR_WRITE_4(sc, RE_TIMERCNT, 1);
1676 }
1677 
1678 static void
1679 re_tick(void *xsc)
1680 {
1681 	struct re_softc *sc = xsc;
1682 
1683 	lwkt_serialize_enter(sc->arpcom.ac_if.if_serializer);
1684 	re_tick_serialized(xsc);
1685 	lwkt_serialize_exit(sc->arpcom.ac_if.if_serializer);
1686 }
1687 
1688 static void
1689 re_tick_serialized(void *xsc)
1690 {
1691 	struct re_softc *sc = xsc;
1692 	struct ifnet *ifp = &sc->arpcom.ac_if;
1693 	struct mii_data *mii;
1694 
1695 	mii = device_get_softc(sc->re_miibus);
1696 	mii_tick(mii);
1697 	if (sc->re_link) {
1698 		if (!(mii->mii_media_status & IFM_ACTIVE))
1699 			sc->re_link = 0;
1700 	} else {
1701 		if (mii->mii_media_status & IFM_ACTIVE &&
1702 		    IFM_SUBTYPE(mii->mii_media_active) != IFM_NONE) {
1703 			sc->re_link = 1;
1704 			if (!ifq_is_empty(&ifp->if_snd))
1705 				ifp->if_start(ifp);
1706 		}
1707 	}
1708 
1709 	callout_reset(&sc->re_timer, hz, re_tick, sc);
1710 }
1711 
1712 #ifdef DEVICE_POLLING
1713 
1714 static void
1715 re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
1716 {
1717 	struct re_softc *sc = ifp->if_softc;
1718 
1719 	switch(cmd) {
1720 	case POLL_REGISTER:
1721 		/* disable interrupts */
1722 		CSR_WRITE_2(sc, RE_IMR, 0x0000);
1723 		break;
1724 	case POLL_DEREGISTER:
1725 		/* enable interrupts */
1726 		CSR_WRITE_2(sc, RE_IMR, sc->re_intrs);
1727 		break;
1728 	default:
1729 		sc->rxcycles = count;
1730 		re_rxeof(sc);
1731 		re_txeof(sc);
1732 
1733 		if (!ifq_is_empty(&ifp->if_snd))
1734 			(*ifp->if_start)(ifp);
1735 
1736 		if (cmd == POLL_AND_CHECK_STATUS) { /* also check status register */
1737 			uint16_t       status;
1738 
1739 			status = CSR_READ_2(sc, RE_ISR);
1740 			if (status == 0xffff)
1741 				return;
1742 			if (status)
1743 				CSR_WRITE_2(sc, RE_ISR, status);
1744 
1745 			/*
1746 			 * XXX check behaviour on receiver stalls.
1747 			 */
1748 
1749 			if (status & RE_ISR_SYSTEM_ERR) {
1750 				re_reset(sc);
1751 				re_init(sc);
1752 			}
1753 		}
1754 		break;
1755 	}
1756 }
1757 #endif /* DEVICE_POLLING */
1758 
1759 static void
1760 re_intr(void *arg)
1761 {
1762 	struct re_softc	*sc = arg;
1763 	struct ifnet *ifp = &sc->arpcom.ac_if;
1764 	uint16_t status;
1765 
1766 	if (sc->suspended || (ifp->if_flags & IFF_UP) == 0)
1767 		return;
1768 
1769 	for (;;) {
1770 		status = CSR_READ_2(sc, RE_ISR);
1771 		/* If the card has gone away the read returns 0xffff. */
1772 		if (status == 0xffff)
1773 			break;
1774 		if (status)
1775 			CSR_WRITE_2(sc, RE_ISR, status);
1776 
1777 		if ((status & sc->re_intrs) == 0)
1778 			break;
1779 
1780 		if (status & (RE_ISR_RX_OK | RE_ISR_RX_ERR | RE_ISR_FIFO_OFLOW))
1781 			re_rxeof(sc);
1782 
1783 		if ((status & sc->re_tx_ack) ||
1784 		    (status & RE_ISR_TX_ERR) ||
1785 		    (status & RE_ISR_TX_DESC_UNAVAIL))
1786 			re_txeof(sc);
1787 
1788 		if (status & RE_ISR_SYSTEM_ERR) {
1789 			re_reset(sc);
1790 			re_init(sc);
1791 		}
1792 
1793 		if (status & RE_ISR_LINKCHG) {
1794 			callout_stop(&sc->re_timer);
1795 			re_tick_serialized(sc);
1796 		}
1797 	}
1798 
1799 	if (!ifq_is_empty(&ifp->if_snd))
1800 		(*ifp->if_start)(ifp);
1801 }
1802 
1803 static int
1804 re_encap(struct re_softc *sc, struct mbuf **m_head, int *idx, int *called_defrag)
1805 {
1806 	struct ifnet *ifp = &sc->arpcom.ac_if;
1807 	struct mbuf *m, *m_new = NULL;
1808 	struct re_dmaload_arg	arg;
1809 	bus_dmamap_t		map;
1810 	int			error;
1811 
1812 	*called_defrag = 0;
1813 	if (sc->re_ldata.re_tx_free <= 4)
1814 		return(EFBIG);
1815 
1816 	m = *m_head;
1817 
1818 	/*
1819 	 * Set up checksum offload. Note: checksum offload bits must
1820 	 * appear in all descriptors of a multi-descriptor transmit
1821 	 * attempt. (This is according to testing done with an 8169
1822 	 * chip. I'm not sure if this is a requirement or a bug.)
1823 	 */
1824 
1825 	arg.re_flags = 0;
1826 
1827 	if (m->m_pkthdr.csum_flags & CSUM_IP)
1828 		arg.re_flags |= RE_TDESC_CMD_IPCSUM;
1829 	if (m->m_pkthdr.csum_flags & CSUM_TCP)
1830 		arg.re_flags |= RE_TDESC_CMD_TCPCSUM;
1831 	if (m->m_pkthdr.csum_flags & CSUM_UDP)
1832 		arg.re_flags |= RE_TDESC_CMD_UDPCSUM;
1833 
1834 	arg.sc = sc;
1835 	arg.re_idx = *idx;
1836 	arg.re_maxsegs = sc->re_ldata.re_tx_free;
1837 	if (arg.re_maxsegs > 4)
1838 		arg.re_maxsegs -= 4;
1839 	arg.re_ring = sc->re_ldata.re_tx_list;
1840 
1841 	map = sc->re_ldata.re_tx_dmamap[*idx];
1842 
1843 	/*
1844 	 * With some of the RealTek chips, using the checksum offload
1845 	 * support in conjunction with the autopadding feature results
1846 	 * in the transmission of corrupt frames. For example, if we
1847 	 * need to send a really small IP fragment that's less than 60
1848 	 * bytes in size, and IP header checksumming is enabled, the
1849 	 * resulting ethernet frame that appears on the wire will
1850 	 * have garbled payload. To work around this, if TX checksum
1851 	 * offload is enabled, we always manually pad short frames out
1852 	 * to the minimum ethernet frame size. We do this by pretending
1853 	 * the mbuf chain has too many fragments so the coalescing code
1854 	 * below can assemble the packet into a single buffer that's
1855 	 * padded out to the mininum frame size.
1856 	 */
1857 	if (arg.re_flags && m->m_pkthdr.len < RE_MIN_FRAMELEN) {
1858 		error = EFBIG;
1859 	} else {
1860 		error = bus_dmamap_load_mbuf(sc->re_ldata.re_mtag, map,
1861 		    m, re_dma_map_desc, &arg, BUS_DMA_NOWAIT);
1862 	}
1863 
1864 	if (error && error != EFBIG) {
1865 		if_printf(ifp, "can't map mbuf (error %d)\n", error);
1866 		return(ENOBUFS);
1867 	}
1868 
1869 	/* Too many segments to map, coalesce into a single mbuf */
1870 
1871 	if (error || arg.re_maxsegs == 0) {
1872 		m_new = m_defrag_nofree(m, MB_DONTWAIT);
1873 		if (m_new == NULL)
1874 			return(1);
1875 		else {
1876 			m = m_new;
1877 			*m_head = m;
1878 		}
1879 
1880 		/*
1881 		 * Manually pad short frames, and zero the pad space
1882 		 * to avoid leaking data.
1883 		 */
1884 		if (m_new->m_pkthdr.len < RE_MIN_FRAMELEN) {
1885 			bzero(mtod(m_new, char *) + m_new->m_pkthdr.len,
1886 			    RE_MIN_FRAMELEN - m_new->m_pkthdr.len);
1887 			m_new->m_pkthdr.len += RE_MIN_FRAMELEN -
1888 			    m_new->m_pkthdr.len;
1889 			m_new->m_len = m_new->m_pkthdr.len;
1890 		}
1891 
1892 		*called_defrag = 1;
1893 		arg.sc = sc;
1894 		arg.re_idx = *idx;
1895 		arg.re_maxsegs = sc->re_ldata.re_tx_free;
1896 		arg.re_ring = sc->re_ldata.re_tx_list;
1897 
1898 		error = bus_dmamap_load_mbuf(sc->re_ldata.re_mtag, map,
1899 		    m, re_dma_map_desc, &arg, BUS_DMA_NOWAIT);
1900 		if (error) {
1901 			m_freem(m);
1902 			if_printf(ifp, "can't map mbuf (error %d)\n", error);
1903 			return(EFBIG);
1904 		}
1905 	}
1906 
1907 	/*
1908 	 * Insure that the map for this transmission
1909 	 * is placed at the array index of the last descriptor
1910 	 * in this chain.
1911 	 */
1912 	sc->re_ldata.re_tx_dmamap[*idx] =
1913 	    sc->re_ldata.re_tx_dmamap[arg.re_idx];
1914 	sc->re_ldata.re_tx_dmamap[arg.re_idx] = map;
1915 
1916 	sc->re_ldata.re_tx_mbuf[arg.re_idx] = m;
1917 	sc->re_ldata.re_tx_free -= arg.re_maxsegs;
1918 
1919 	/*
1920 	 * Set up hardware VLAN tagging. Note: vlan tag info must
1921 	 * appear in the first descriptor of a multi-descriptor
1922 	 * transmission attempt.
1923 	 */
1924 
1925 	if ((m->m_flags & (M_PROTO1|M_PKTHDR)) == (M_PROTO1|M_PKTHDR) &&
1926 	    m->m_pkthdr.rcvif != NULL &&
1927 	    m->m_pkthdr.rcvif->if_type == IFT_L2VLAN) {
1928 	    	struct ifvlan *ifv;
1929 		ifv = m->m_pkthdr.rcvif->if_softc;
1930 		if (ifv != NULL)
1931 			sc->re_ldata.re_tx_list[*idx].re_vlanctl =
1932 			    htole32(htobe16(ifv->ifv_tag) | RE_TDESC_VLANCTL_TAG);
1933 	}
1934 
1935 	/* Transfer ownership of packet to the chip. */
1936 
1937 	sc->re_ldata.re_tx_list[arg.re_idx].re_cmdstat |=
1938 	    htole32(RE_TDESC_CMD_OWN);
1939 	if (*idx != arg.re_idx)
1940 		sc->re_ldata.re_tx_list[*idx].re_cmdstat |=
1941 		    htole32(RE_TDESC_CMD_OWN);
1942 
1943 	RE_DESC_INC(arg.re_idx);
1944 	*idx = arg.re_idx;
1945 
1946 	return(0);
1947 }
1948 
1949 /*
1950  * Main transmit routine for C+ and gigE NICs.
1951  */
1952 
1953 static void
1954 re_start(struct ifnet *ifp)
1955 {
1956 	struct re_softc	*sc = ifp->if_softc;
1957 	struct mbuf *m_head;
1958 	struct mbuf *m_head2;
1959 	int called_defrag, idx, need_trans;
1960 
1961 	if (!sc->re_link || (ifp->if_flags & IFF_OACTIVE))
1962 		return;
1963 
1964 	idx = sc->re_ldata.re_tx_prodidx;
1965 
1966 	need_trans = 0;
1967 	while (sc->re_ldata.re_tx_mbuf[idx] == NULL) {
1968 		m_head = ifq_poll(&ifp->if_snd);
1969 		if (m_head == NULL)
1970 			break;
1971 		m_head2 = m_head;
1972 		if (re_encap(sc, &m_head2, &idx, &called_defrag)) {
1973 			/*
1974 			 * If we could not encapsulate the defragged packet,
1975 			 * the returned m_head2 is garbage and we must dequeue
1976 			 * and throw away the original packet.
1977 			 */
1978 			if (called_defrag) {
1979 				ifq_dequeue(&ifp->if_snd, m_head);
1980 				m_freem(m_head);
1981 			}
1982 			ifp->if_flags |= IFF_OACTIVE;
1983 			break;
1984 		}
1985 
1986 		/*
1987 		 * Clean out the packet we encapsulated.  If we defragged
1988 		 * the packet the m_head2 is the one that got encapsulated
1989 		 * and the original must be thrown away.  Otherwise m_head2
1990 		 * *IS* the original.
1991 		 */
1992 		ifq_dequeue(&ifp->if_snd, m_head);
1993 		if (called_defrag)
1994 			m_freem(m_head);
1995 		need_trans = 1;
1996 
1997 		/*
1998 		 * If there's a BPF listener, bounce a copy of this frame
1999 		 * to him.
2000 		 */
2001 		BPF_MTAP(ifp, m_head2);
2002 	}
2003 
2004 	if (!need_trans) {
2005 		if (RE_TX_MODERATION_IS_ENABLED(sc) &&
2006 		    sc->re_ldata.re_tx_free != RE_TX_DESC_CNT)
2007 			CSR_WRITE_4(sc, RE_TIMERCNT, 1);
2008 		return;
2009 	}
2010 
2011 	/* Flush the TX descriptors */
2012 	bus_dmamap_sync(sc->re_ldata.re_tx_list_tag,
2013 			sc->re_ldata.re_tx_list_map, BUS_DMASYNC_PREWRITE);
2014 
2015 	sc->re_ldata.re_tx_prodidx = idx;
2016 
2017 	/*
2018 	 * RealTek put the TX poll request register in a different
2019 	 * location on the 8169 gigE chip. I don't know why.
2020 	 */
2021 	CSR_WRITE_1(sc, sc->re_txstart, RE_TXSTART_START);
2022 
2023 	if (RE_TX_MODERATION_IS_ENABLED(sc)) {
2024 		/*
2025 		 * Use the countdown timer for interrupt moderation.
2026 		 * 'TX done' interrupts are disabled. Instead, we reset the
2027 		 * countdown timer, which will begin counting until it hits
2028 		 * the value in the TIMERINT register, and then trigger an
2029 		 * interrupt. Each time we write to the TIMERCNT register,
2030 		 * the timer count is reset to 0.
2031 		 */
2032 		CSR_WRITE_4(sc, RE_TIMERCNT, 1);
2033 	}
2034 
2035 	/*
2036 	 * Set a timeout in case the chip goes out to lunch.
2037 	 */
2038 	ifp->if_timer = 5;
2039 }
2040 
2041 static void
2042 re_init(void *xsc)
2043 {
2044 	struct re_softc *sc = xsc;
2045 	struct ifnet *ifp = &sc->arpcom.ac_if;
2046 	struct mii_data *mii;
2047 	uint32_t rxcfg = 0;
2048 
2049 	mii = device_get_softc(sc->re_miibus);
2050 
2051 	/*
2052 	 * Cancel pending I/O and free all RX/TX buffers.
2053 	 */
2054 	re_stop(sc);
2055 
2056 	/*
2057 	 * Enable C+ RX and TX mode, as well as VLAN stripping and
2058 	 * RX checksum offload. We must configure the C+ register
2059 	 * before all others.
2060 	 */
2061 	CSR_WRITE_2(sc, RE_CPLUS_CMD, RE_CPLUSCMD_RXENB | RE_CPLUSCMD_TXENB |
2062 		    RE_CPLUSCMD_PCI_MRW | RE_CPLUSCMD_VLANSTRIP |
2063 		    (ifp->if_capenable & IFCAP_RXCSUM ?
2064 		     RE_CPLUSCMD_RXCSUM_ENB : 0));
2065 
2066 	/*
2067 	 * Init our MAC address.  Even though the chipset
2068 	 * documentation doesn't mention it, we need to enter "Config
2069 	 * register write enable" mode to modify the ID registers.
2070 	 */
2071 	CSR_WRITE_1(sc, RE_EECMD, RE_EEMODE_WRITECFG);
2072 	CSR_WRITE_4(sc, RE_IDR0,
2073 	    htole32(*(uint32_t *)(&sc->arpcom.ac_enaddr[0])));
2074 	CSR_WRITE_2(sc, RE_IDR4,
2075 	    htole16(*(uint16_t *)(&sc->arpcom.ac_enaddr[4])));
2076 	CSR_WRITE_1(sc, RE_EECMD, RE_EEMODE_OFF);
2077 
2078 	/*
2079 	 * For C+ mode, initialize the RX descriptors and mbufs.
2080 	 */
2081 	re_rx_list_init(sc);
2082 	re_tx_list_init(sc);
2083 
2084 	/*
2085 	 * Load the addresses of the RX and TX lists into the chip.
2086 	 */
2087 	CSR_WRITE_4(sc, RE_RXLIST_ADDR_HI,
2088 	    RE_ADDR_HI(sc->re_ldata.re_rx_list_addr));
2089 	CSR_WRITE_4(sc, RE_RXLIST_ADDR_LO,
2090 	    RE_ADDR_LO(sc->re_ldata.re_rx_list_addr));
2091 
2092 	CSR_WRITE_4(sc, RE_TXLIST_ADDR_HI,
2093 	    RE_ADDR_HI(sc->re_ldata.re_tx_list_addr));
2094 	CSR_WRITE_4(sc, RE_TXLIST_ADDR_LO,
2095 	    RE_ADDR_LO(sc->re_ldata.re_tx_list_addr));
2096 
2097 	/*
2098 	 * Enable transmit and receive.
2099 	 */
2100 	CSR_WRITE_1(sc, RE_COMMAND, RE_CMD_TX_ENB|RE_CMD_RX_ENB);
2101 
2102 	/*
2103 	 * Set the initial TX and RX configuration.
2104 	 */
2105 	if (sc->re_testmode) {
2106 		if (sc->re_type == RE_8169)
2107 			CSR_WRITE_4(sc, RE_TXCFG,
2108 				    RE_TXCFG_CONFIG | RE_LOOPTEST_ON);
2109 		else
2110 			CSR_WRITE_4(sc, RE_TXCFG,
2111 				    RE_TXCFG_CONFIG | RE_LOOPTEST_ON_CPLUS);
2112 	} else
2113 		CSR_WRITE_4(sc, RE_TXCFG, RE_TXCFG_CONFIG);
2114 
2115 	CSR_WRITE_1(sc, RE_EARLY_TX_THRESH, 16);
2116 
2117 	CSR_WRITE_4(sc, RE_RXCFG, RE_RXCFG_CONFIG);
2118 
2119 	/* Set the individual bit to receive frames for this host only. */
2120 	rxcfg = CSR_READ_4(sc, RE_RXCFG);
2121 	rxcfg |= RE_RXCFG_RX_INDIV;
2122 
2123 	/* If we want promiscuous mode, set the allframes bit. */
2124 	if (ifp->if_flags & IFF_PROMISC) {
2125 		rxcfg |= RE_RXCFG_RX_ALLPHYS;
2126 		CSR_WRITE_4(sc, RE_RXCFG, rxcfg);
2127 	} else {
2128 		rxcfg &= ~RE_RXCFG_RX_ALLPHYS;
2129 		CSR_WRITE_4(sc, RE_RXCFG, rxcfg);
2130 	}
2131 
2132 	/*
2133 	 * Set capture broadcast bit to capture broadcast frames.
2134 	 */
2135 	if (ifp->if_flags & IFF_BROADCAST) {
2136 		rxcfg |= RE_RXCFG_RX_BROAD;
2137 		CSR_WRITE_4(sc, RE_RXCFG, rxcfg);
2138 	} else {
2139 		rxcfg &= ~RE_RXCFG_RX_BROAD;
2140 		CSR_WRITE_4(sc, RE_RXCFG, rxcfg);
2141 	}
2142 
2143 	/*
2144 	 * Program the multicast filter, if necessary.
2145 	 */
2146 	re_setmulti(sc);
2147 
2148 #ifdef DEVICE_POLLING
2149 	/*
2150 	 * Disable interrupts if we are polling.
2151 	 */
2152 	if (ifp->if_flags & IFF_POLLING)
2153 		CSR_WRITE_2(sc, RE_IMR, 0);
2154 	else	/* otherwise ... */
2155 #endif /* DEVICE_POLLING */
2156 	/*
2157 	 * Enable interrupts.
2158 	 */
2159 	if (sc->re_testmode)
2160 		CSR_WRITE_2(sc, RE_IMR, 0);
2161 	else
2162 		CSR_WRITE_2(sc, RE_IMR, sc->re_intrs);
2163 	CSR_WRITE_2(sc, RE_ISR, sc->re_intrs);
2164 
2165 	/* Set initial TX threshold */
2166 	sc->re_txthresh = RE_TX_THRESH_INIT;
2167 
2168 	/* Start RX/TX process. */
2169 	if (sc->re_flags & RE_F_HASMPC)
2170 		CSR_WRITE_4(sc, RE_MISSEDPKT, 0);
2171 #ifdef notdef
2172 	/* Enable receiver and transmitter. */
2173 	CSR_WRITE_1(sc, RE_COMMAND, RE_CMD_TX_ENB|RE_CMD_RX_ENB);
2174 #endif
2175 
2176 	if (RE_TX_MODERATION_IS_ENABLED(sc)) {
2177 		/*
2178 		 * Initialize the timer interrupt register so that
2179 		 * a timer interrupt will be generated once the timer
2180 		 * reaches a certain number of ticks. The timer is
2181 		 * reloaded on each transmit. This gives us TX interrupt
2182 		 * moderation, which dramatically improves TX frame rate.
2183 		 */
2184 		if (sc->re_type == RE_8169)
2185 			CSR_WRITE_4(sc, RE_TIMERINT_8169, 0x800);
2186 		else
2187 			CSR_WRITE_4(sc, RE_TIMERINT, 0x400);
2188 	}
2189 
2190 	/*
2191 	 * For 8169 gigE NICs, set the max allowed RX packet
2192 	 * size so we can receive jumbo frames.
2193 	 */
2194 	if (sc->re_type == RE_8169)
2195 		CSR_WRITE_2(sc, RE_MAXRXPKTLEN, 16383);
2196 
2197 	if (sc->re_testmode) {
2198 		return;
2199 	}
2200 
2201 	mii_mediachg(mii);
2202 
2203 	CSR_WRITE_1(sc, RE_CFG1, RE_CFG1_DRVLOAD|RE_CFG1_FULLDUPLEX);
2204 
2205 	ifp->if_flags |= IFF_RUNNING;
2206 	ifp->if_flags &= ~IFF_OACTIVE;
2207 
2208 	sc->re_link = 0;
2209 	callout_reset(&sc->re_timer, hz, re_tick, sc);
2210 }
2211 
2212 /*
2213  * Set media options.
2214  */
2215 static int
2216 re_ifmedia_upd(struct ifnet *ifp)
2217 {
2218 	struct re_softc *sc = ifp->if_softc;
2219 	struct mii_data *mii;
2220 
2221 	mii = device_get_softc(sc->re_miibus);
2222 	mii_mediachg(mii);
2223 
2224 	return(0);
2225 }
2226 
2227 /*
2228  * Report current media status.
2229  */
2230 static void
2231 re_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
2232 {
2233 	struct re_softc *sc = ifp->if_softc;
2234 	struct mii_data *mii;
2235 
2236 	mii = device_get_softc(sc->re_miibus);
2237 
2238 	mii_pollstat(mii);
2239 	ifmr->ifm_active = mii->mii_media_active;
2240 	ifmr->ifm_status = mii->mii_media_status;
2241 }
2242 
2243 static int
2244 re_ioctl(struct ifnet *ifp, u_long command, caddr_t data, struct ucred *cr)
2245 {
2246 	struct re_softc *sc = ifp->if_softc;
2247 	struct ifreq *ifr = (struct ifreq *) data;
2248 	struct mii_data *mii;
2249 	int error = 0;
2250 
2251 	switch(command) {
2252 	case SIOCSIFMTU:
2253 		if (ifr->ifr_mtu > RE_JUMBO_MTU)
2254 			error = EINVAL;
2255 		ifp->if_mtu = ifr->ifr_mtu;
2256 		break;
2257 	case SIOCSIFFLAGS:
2258 		if (ifp->if_flags & IFF_UP)
2259 			re_init(sc);
2260 		else if (ifp->if_flags & IFF_RUNNING)
2261 			re_stop(sc);
2262 		break;
2263 	case SIOCADDMULTI:
2264 	case SIOCDELMULTI:
2265 		re_setmulti(sc);
2266 		error = 0;
2267 		break;
2268 	case SIOCGIFMEDIA:
2269 	case SIOCSIFMEDIA:
2270 		mii = device_get_softc(sc->re_miibus);
2271 		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, command);
2272 		break;
2273 	case SIOCSIFCAP:
2274 		ifp->if_capenable &= ~(IFCAP_HWCSUM);
2275 		ifp->if_capenable |=
2276 		    ifr->ifr_reqcap & (IFCAP_HWCSUM);
2277 		if (ifp->if_capenable & IFCAP_TXCSUM)
2278 			ifp->if_hwassist = RE_CSUM_FEATURES;
2279 		else
2280 			ifp->if_hwassist = 0;
2281 		if (ifp->if_flags & IFF_RUNNING)
2282 			re_init(sc);
2283 		break;
2284 	default:
2285 		error = ether_ioctl(ifp, command, data);
2286 		break;
2287 	}
2288 	return(error);
2289 }
2290 
2291 static void
2292 re_watchdog(struct ifnet *ifp)
2293 {
2294 	struct re_softc *sc = ifp->if_softc;
2295 
2296 	if_printf(ifp, "watchdog timeout\n");
2297 
2298 	ifp->if_oerrors++;
2299 
2300 	re_txeof(sc);
2301 	re_rxeof(sc);
2302 
2303 	re_init(sc);
2304 
2305 	if (!ifq_is_empty(&ifp->if_snd))
2306 		ifp->if_start(ifp);
2307 }
2308 
2309 /*
2310  * Stop the adapter and free any mbufs allocated to the
2311  * RX and TX lists.
2312  */
2313 static void
2314 re_stop(struct re_softc *sc)
2315 {
2316 	struct ifnet *ifp = &sc->arpcom.ac_if;
2317 	int i;
2318 
2319 	ifp->if_timer = 0;
2320 	callout_stop(&sc->re_timer);
2321 
2322 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
2323 
2324 	CSR_WRITE_1(sc, RE_COMMAND, 0x00);
2325 	CSR_WRITE_2(sc, RE_IMR, 0x0000);
2326 	CSR_WRITE_2(sc, RE_ISR, 0xFFFF);
2327 
2328 	if (sc->re_head != NULL) {
2329 		m_freem(sc->re_head);
2330 		sc->re_head = sc->re_tail = NULL;
2331 	}
2332 
2333 	/* Free the TX list buffers. */
2334 	for (i = 0; i < RE_TX_DESC_CNT; i++) {
2335 		if (sc->re_ldata.re_tx_mbuf[i] != NULL) {
2336 			bus_dmamap_unload(sc->re_ldata.re_mtag,
2337 					  sc->re_ldata.re_tx_dmamap[i]);
2338 			m_freem(sc->re_ldata.re_tx_mbuf[i]);
2339 			sc->re_ldata.re_tx_mbuf[i] = NULL;
2340 		}
2341 	}
2342 
2343 	/* Free the RX list buffers. */
2344 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
2345 		if (sc->re_ldata.re_rx_mbuf[i] != NULL) {
2346 			bus_dmamap_unload(sc->re_ldata.re_mtag,
2347 					  sc->re_ldata.re_rx_dmamap[i]);
2348 			m_freem(sc->re_ldata.re_rx_mbuf[i]);
2349 			sc->re_ldata.re_rx_mbuf[i] = NULL;
2350 		}
2351 	}
2352 }
2353 
2354 /*
2355  * Device suspend routine.  Stop the interface and save some PCI
2356  * settings in case the BIOS doesn't restore them properly on
2357  * resume.
2358  */
2359 static int
2360 re_suspend(device_t dev)
2361 {
2362 #ifndef BURN_BRIDGES
2363 	int i;
2364 #endif
2365 	struct re_softc *sc = device_get_softc(dev);
2366 
2367 	re_stop(sc);
2368 
2369 #ifndef BURN_BRIDGES
2370 	for (i = 0; i < 5; i++)
2371 		sc->saved_maps[i] = pci_read_config(dev, PCIR_MAPS + i * 4, 4);
2372 	sc->saved_biosaddr = pci_read_config(dev, PCIR_BIOS, 4);
2373 	sc->saved_intline = pci_read_config(dev, PCIR_INTLINE, 1);
2374 	sc->saved_cachelnsz = pci_read_config(dev, PCIR_CACHELNSZ, 1);
2375 	sc->saved_lattimer = pci_read_config(dev, PCIR_LATTIMER, 1);
2376 #endif
2377 
2378 	sc->suspended = 1;
2379 
2380 	return (0);
2381 }
2382 
2383 /*
2384  * Device resume routine.  Restore some PCI settings in case the BIOS
2385  * doesn't, re-enable busmastering, and restart the interface if
2386  * appropriate.
2387  */
2388 static int
2389 re_resume(device_t dev)
2390 {
2391 	struct re_softc *sc = device_get_softc(dev);
2392 	struct ifnet *ifp = &sc->arpcom.ac_if;
2393 #ifndef BURN_BRIDGES
2394 	int i;
2395 #endif
2396 
2397 #ifndef BURN_BRIDGES
2398 	/* better way to do this? */
2399 	for (i = 0; i < 5; i++)
2400 		pci_write_config(dev, PCIR_MAPS + i * 4, sc->saved_maps[i], 4);
2401 	pci_write_config(dev, PCIR_BIOS, sc->saved_biosaddr, 4);
2402 	pci_write_config(dev, PCIR_INTLINE, sc->saved_intline, 1);
2403 	pci_write_config(dev, PCIR_CACHELNSZ, sc->saved_cachelnsz, 1);
2404 	pci_write_config(dev, PCIR_LATTIMER, sc->saved_lattimer, 1);
2405 
2406 	/* reenable busmastering */
2407 	pci_enable_busmaster(dev);
2408 	pci_enable_io(dev, SYS_RES_IOPORT);
2409 #endif
2410 
2411 	/* reinitialize interface if necessary */
2412 	if (ifp->if_flags & IFF_UP)
2413 		re_init(sc);
2414 
2415 	sc->suspended = 0;
2416 
2417 	return (0);
2418 }
2419 
2420 /*
2421  * Stop all chip I/O so that the kernel's probe routines don't
2422  * get confused by errant DMAs when rebooting.
2423  */
2424 static void
2425 re_shutdown(device_t dev)
2426 {
2427 	struct re_softc *sc = device_get_softc(dev);
2428 	struct ifnet *ifp = &sc->arpcom.ac_if;
2429 
2430 	lwkt_serialize_enter(ifp->if_serializer);
2431 	re_stop(sc);
2432 	lwkt_serialize_exit(ifp->if_serializer);
2433 }
2434 
2435 static int
2436 re_sysctl_tx_moderation(SYSCTL_HANDLER_ARGS)
2437 {
2438 	struct re_softc *sc = arg1;
2439 	struct ifnet *ifp = &sc->arpcom.ac_if;
2440 	int error = 0, mod, mod_old;
2441 
2442 	lwkt_serialize_enter(ifp->if_serializer);
2443 
2444 	mod_old = mod = RE_TX_MODERATION_IS_ENABLED(sc);
2445 
2446 	error = sysctl_handle_int(oidp, &mod, 0, req);
2447 	if (error || req->newptr == NULL || mod == mod_old)
2448 		goto back;
2449 	if (mod != 0 && mod != 1) {
2450 		error = EINVAL;
2451 		goto back;
2452 	}
2453 
2454 	if (mod)
2455 		RE_ENABLE_TX_MODERATION(sc);
2456 	else
2457 		RE_DISABLE_TX_MODERATION(sc);
2458 
2459 	if ((ifp->if_flags & (IFF_RUNNING | IFF_UP)) == (IFF_RUNNING | IFF_UP))
2460 		re_init(sc);
2461 back:
2462 	lwkt_serialize_exit(ifp->if_serializer);
2463 	return error;
2464 }
2465