xref: /dragonfly/sys/dev/netif/re/if_re.c (revision 23265324)
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 1.31 2007/02/14 13:00:34 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 	ifp->if_capenable = ifp->if_capabilities;
1224 	ifp->if_hwassist = RE_CSUM_FEATURES;
1225 #endif	/* RE_DISABLE_HWCSUM */
1226 
1227 	/*
1228 	 * Call MI attach routine.
1229 	 */
1230 	ether_ifattach(ifp, eaddr, NULL);
1231 
1232 #ifdef RE_DIAG
1233 	/*
1234 	 * Perform hardware diagnostic on the original RTL8169.
1235 	 * Some 32-bit cards were incorrectly wired and would
1236 	 * malfunction if plugged into a 64-bit slot.
1237 	 */
1238 	if (hwrev == RE_HWREV_8169) {
1239 		lwkt_serialize_enter(ifp->if_serializer);
1240 		error = re_diag(sc);
1241 		lwkt_serialize_exit(ifp->if_serializer);
1242 
1243 		if (error) {
1244 			device_printf(dev, "hardware diagnostic failure\n");
1245 			ether_ifdetach(ifp);
1246 			goto fail;
1247 		}
1248 	}
1249 #endif	/* RE_DIAG */
1250 
1251 	/* Hook interrupt last to avoid having to lock softc */
1252 	error = bus_setup_intr(dev, sc->re_irq, INTR_NETSAFE, re_intr, sc,
1253 			       &sc->re_intrhand, ifp->if_serializer);
1254 
1255 	if (error) {
1256 		device_printf(dev, "couldn't set up irq\n");
1257 		ether_ifdetach(ifp);
1258 		goto fail;
1259 	}
1260 
1261 fail:
1262 	if (error)
1263 		re_detach(dev);
1264 
1265 	return (error);
1266 }
1267 
1268 /*
1269  * Shutdown hardware and free up resources. This can be called any
1270  * time after the mutex has been initialized. It is called in both
1271  * the error case in attach and the normal detach case so it needs
1272  * to be careful about only freeing resources that have actually been
1273  * allocated.
1274  */
1275 static int
1276 re_detach(device_t dev)
1277 {
1278 	struct re_softc *sc = device_get_softc(dev);
1279 	struct ifnet *ifp = &sc->arpcom.ac_if;
1280 	int i;
1281 
1282 	/* These should only be active if attach succeeded */
1283 	if (device_is_attached(dev)) {
1284 		lwkt_serialize_enter(ifp->if_serializer);
1285 		re_stop(sc);
1286 		bus_teardown_intr(dev, sc->re_irq, sc->re_intrhand);
1287 		lwkt_serialize_exit(ifp->if_serializer);
1288 
1289 		ether_ifdetach(ifp);
1290 	}
1291 	if (sc->re_miibus)
1292 		device_delete_child(dev, sc->re_miibus);
1293 	bus_generic_detach(dev);
1294 
1295 	if (sc->re_irq)
1296 		bus_release_resource(dev, SYS_RES_IRQ, 0, sc->re_irq);
1297 	if (sc->re_res) {
1298 		bus_release_resource(dev, SYS_RES_IOPORT, RE_PCI_LOIO,
1299 				     sc->re_res);
1300 	}
1301 
1302 	/* Unload and free the RX DMA ring memory and map */
1303 
1304 	if (sc->re_ldata.re_rx_list_tag) {
1305 		bus_dmamap_unload(sc->re_ldata.re_rx_list_tag,
1306 		    sc->re_ldata.re_rx_list_map);
1307 		bus_dmamem_free(sc->re_ldata.re_rx_list_tag,
1308 		    sc->re_ldata.re_rx_list,
1309 		    sc->re_ldata.re_rx_list_map);
1310 		bus_dma_tag_destroy(sc->re_ldata.re_rx_list_tag);
1311 	}
1312 
1313 	/* Unload and free the TX DMA ring memory and map */
1314 
1315 	if (sc->re_ldata.re_tx_list_tag) {
1316 		bus_dmamap_unload(sc->re_ldata.re_tx_list_tag,
1317 		    sc->re_ldata.re_tx_list_map);
1318 		bus_dmamem_free(sc->re_ldata.re_tx_list_tag,
1319 		    sc->re_ldata.re_tx_list,
1320 		    sc->re_ldata.re_tx_list_map);
1321 		bus_dma_tag_destroy(sc->re_ldata.re_tx_list_tag);
1322 	}
1323 
1324 	/* Destroy all the RX and TX buffer maps */
1325 
1326 	if (sc->re_ldata.re_mtag) {
1327 		for (i = 0; i < RE_TX_DESC_CNT; i++)
1328 			bus_dmamap_destroy(sc->re_ldata.re_mtag,
1329 			    sc->re_ldata.re_tx_dmamap[i]);
1330 		for (i = 0; i < RE_RX_DESC_CNT; i++)
1331 			bus_dmamap_destroy(sc->re_ldata.re_mtag,
1332 			    sc->re_ldata.re_rx_dmamap[i]);
1333 		bus_dma_tag_destroy(sc->re_ldata.re_mtag);
1334 	}
1335 
1336 	/* Unload and free the stats buffer and map */
1337 
1338 	if (sc->re_ldata.re_stag) {
1339 		bus_dmamap_unload(sc->re_ldata.re_stag,
1340 		    sc->re_ldata.re_rx_list_map);
1341 		bus_dmamem_free(sc->re_ldata.re_stag,
1342 		    sc->re_ldata.re_stats,
1343 		    sc->re_ldata.re_smap);
1344 		bus_dma_tag_destroy(sc->re_ldata.re_stag);
1345 	}
1346 
1347 	if (sc->re_parent_tag)
1348 		bus_dma_tag_destroy(sc->re_parent_tag);
1349 
1350 	return(0);
1351 }
1352 
1353 static int
1354 re_newbuf(struct re_softc *sc, int idx, struct mbuf *m)
1355 {
1356 	struct re_dmaload_arg arg;
1357 	struct mbuf *n = NULL;
1358 	int error;
1359 
1360 	if (m == NULL) {
1361 		n = m_getcl(MB_DONTWAIT, MT_DATA, M_PKTHDR);
1362 		if (n == NULL)
1363 			return(ENOBUFS);
1364 		m = n;
1365 	} else
1366 		m->m_data = m->m_ext.ext_buf;
1367 
1368 	m->m_len = m->m_pkthdr.len = MCLBYTES;
1369 
1370 	/*
1371 	 * NOTE:
1372 	 * Some re(4) chips(e.g. RTL8101E) need address of the receive buffer
1373 	 * to be 8-byte aligned, so don't call m_adj(m, ETHER_ALIGN) here.
1374 	 */
1375 
1376 	arg.sc = sc;
1377 	arg.re_idx = idx;
1378 	arg.re_maxsegs = 1;
1379 	arg.re_flags = 0;
1380 	arg.re_ring = sc->re_ldata.re_rx_list;
1381 
1382         error = bus_dmamap_load_mbuf(sc->re_ldata.re_mtag,
1383 	    sc->re_ldata.re_rx_dmamap[idx], m, re_dma_map_desc,
1384 	    &arg, BUS_DMA_NOWAIT);
1385 	if (error || arg.re_maxsegs != 1) {
1386 		if (n != NULL)
1387 			m_freem(n);
1388 		return (ENOMEM);
1389 	}
1390 
1391 	sc->re_ldata.re_rx_list[idx].re_cmdstat |= htole32(RE_RDESC_CMD_OWN);
1392 	sc->re_ldata.re_rx_mbuf[idx] = m;
1393 
1394         bus_dmamap_sync(sc->re_ldata.re_mtag, sc->re_ldata.re_rx_dmamap[idx],
1395 		        BUS_DMASYNC_PREREAD);
1396 
1397 	return(0);
1398 }
1399 
1400 static int
1401 re_tx_list_init(struct re_softc *sc)
1402 {
1403 	bzero(sc->re_ldata.re_tx_list, RE_TX_LIST_SZ);
1404 	bzero(&sc->re_ldata.re_tx_mbuf, RE_TX_DESC_CNT * sizeof(struct mbuf *));
1405 
1406 	bus_dmamap_sync(sc->re_ldata.re_tx_list_tag,
1407 			sc->re_ldata.re_tx_list_map, BUS_DMASYNC_PREWRITE);
1408 	sc->re_ldata.re_tx_prodidx = 0;
1409 	sc->re_ldata.re_tx_considx = 0;
1410 	sc->re_ldata.re_tx_free = RE_TX_DESC_CNT;
1411 
1412 	return(0);
1413 }
1414 
1415 static int
1416 re_rx_list_init(struct re_softc *sc)
1417 {
1418 	int i, error;
1419 
1420 	bzero(sc->re_ldata.re_rx_list, RE_RX_LIST_SZ);
1421 	bzero(&sc->re_ldata.re_rx_mbuf, RE_RX_DESC_CNT * sizeof(struct mbuf *));
1422 
1423 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
1424 		error = re_newbuf(sc, i, NULL);
1425 		if (error)
1426 			return(error);
1427 	}
1428 
1429 	/* Flush the RX descriptors */
1430 
1431 	bus_dmamap_sync(sc->re_ldata.re_rx_list_tag,
1432 			sc->re_ldata.re_rx_list_map, BUS_DMASYNC_PREWRITE);
1433 
1434 	sc->re_ldata.re_rx_prodidx = 0;
1435 	sc->re_head = sc->re_tail = NULL;
1436 
1437 	return(0);
1438 }
1439 
1440 /*
1441  * RX handler for C+ and 8169. For the gigE chips, we support
1442  * the reception of jumbo frames that have been fragmented
1443  * across multiple 2K mbuf cluster buffers.
1444  */
1445 static void
1446 re_rxeof(struct re_softc *sc)
1447 {
1448 	struct ifnet *ifp = &sc->arpcom.ac_if;
1449 	struct mbuf *m;
1450 	struct re_desc 	*cur_rx;
1451 	uint32_t rxstat, rxvlan;
1452 	int i, total_len;
1453 
1454 	/* Invalidate the descriptor memory */
1455 
1456 	bus_dmamap_sync(sc->re_ldata.re_rx_list_tag,
1457 			sc->re_ldata.re_rx_list_map, BUS_DMASYNC_POSTREAD);
1458 
1459 	for (i = sc->re_ldata.re_rx_prodidx;
1460 	     RE_OWN(&sc->re_ldata.re_rx_list[i]) == 0 ; RE_DESC_INC(i)) {
1461 		cur_rx = &sc->re_ldata.re_rx_list[i];
1462 		m = sc->re_ldata.re_rx_mbuf[i];
1463 		total_len = RE_RXBYTES(cur_rx);
1464 		rxstat = le32toh(cur_rx->re_cmdstat);
1465 		rxvlan = le32toh(cur_rx->re_vlanctl);
1466 
1467 		/* Invalidate the RX mbuf and unload its map */
1468 
1469 		bus_dmamap_sync(sc->re_ldata.re_mtag,
1470 				sc->re_ldata.re_rx_dmamap[i],
1471 				BUS_DMASYNC_POSTWRITE);
1472 		bus_dmamap_unload(sc->re_ldata.re_mtag,
1473 				  sc->re_ldata.re_rx_dmamap[i]);
1474 
1475 		if ((rxstat & RE_RDESC_STAT_EOF) == 0) {
1476 			m->m_len = MCLBYTES - ETHER_ALIGN;
1477 			if (sc->re_head == NULL) {
1478 				sc->re_head = sc->re_tail = m;
1479 			} else {
1480 				sc->re_tail->m_next = m;
1481 				sc->re_tail = m;
1482 			}
1483 			re_newbuf(sc, i, NULL);
1484 			continue;
1485 		}
1486 
1487 		/*
1488 		 * NOTE: for the 8139C+, the frame length field
1489 		 * is always 12 bits in size, but for the gigE chips,
1490 		 * it is 13 bits (since the max RX frame length is 16K).
1491 		 * Unfortunately, all 32 bits in the status word
1492 		 * were already used, so to make room for the extra
1493 		 * length bit, RealTek took out the 'frame alignment
1494 		 * error' bit and shifted the other status bits
1495 		 * over one slot. The OWN, EOR, FS and LS bits are
1496 		 * still in the same places. We have already extracted
1497 		 * the frame length and checked the OWN bit, so rather
1498 		 * than using an alternate bit mapping, we shift the
1499 		 * status bits one space to the right so we can evaluate
1500 		 * them using the 8169 status as though it was in the
1501 		 * same format as that of the 8139C+.
1502 		 */
1503 		if (sc->re_type == RE_8169)
1504 			rxstat >>= 1;
1505 
1506 		if (rxstat & RE_RDESC_STAT_RXERRSUM) {
1507 			ifp->if_ierrors++;
1508 			/*
1509 			 * If this is part of a multi-fragment packet,
1510 			 * discard all the pieces.
1511 			 */
1512 			if (sc->re_head != NULL) {
1513 				m_freem(sc->re_head);
1514 				sc->re_head = sc->re_tail = NULL;
1515 			}
1516 			re_newbuf(sc, i, m);
1517 			continue;
1518 		}
1519 
1520 		/*
1521 		 * If allocating a replacement mbuf fails,
1522 		 * reload the current one.
1523 		 */
1524 
1525 		if (re_newbuf(sc, i, NULL)) {
1526 			ifp->if_ierrors++;
1527 			if (sc->re_head != NULL) {
1528 				m_freem(sc->re_head);
1529 				sc->re_head = sc->re_tail = NULL;
1530 			}
1531 			re_newbuf(sc, i, m);
1532 			continue;
1533 		}
1534 
1535 		if (sc->re_head != NULL) {
1536 			m->m_len = total_len % (MCLBYTES - ETHER_ALIGN);
1537 			/*
1538 			 * Special case: if there's 4 bytes or less
1539 			 * in this buffer, the mbuf can be discarded:
1540 			 * the last 4 bytes is the CRC, which we don't
1541 			 * care about anyway.
1542 			 */
1543 			if (m->m_len <= ETHER_CRC_LEN) {
1544 				sc->re_tail->m_len -=
1545 				    (ETHER_CRC_LEN - m->m_len);
1546 				m_freem(m);
1547 			} else {
1548 				m->m_len -= ETHER_CRC_LEN;
1549 				sc->re_tail->m_next = m;
1550 			}
1551 			m = sc->re_head;
1552 			sc->re_head = sc->re_tail = NULL;
1553 			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
1554 		} else
1555 			m->m_pkthdr.len = m->m_len =
1556 			    (total_len - ETHER_CRC_LEN);
1557 
1558 		ifp->if_ipackets++;
1559 		m->m_pkthdr.rcvif = ifp;
1560 
1561 		/* Do RX checksumming if enabled */
1562 
1563 		if (ifp->if_capenable & IFCAP_RXCSUM) {
1564 
1565 			/* Check IP header checksum */
1566 			if (rxstat & RE_RDESC_STAT_PROTOID)
1567 				m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED;
1568 			if ((rxstat & RE_RDESC_STAT_IPSUMBAD) == 0)
1569 				m->m_pkthdr.csum_flags |= CSUM_IP_VALID;
1570 
1571 			/* Check TCP/UDP checksum */
1572 			if ((RE_TCPPKT(rxstat) &&
1573 			    (rxstat & RE_RDESC_STAT_TCPSUMBAD) == 0) ||
1574 			    (RE_UDPPKT(rxstat) &&
1575 			    (rxstat & RE_RDESC_STAT_UDPSUMBAD)) == 0) {
1576 				m->m_pkthdr.csum_flags |=
1577 				    CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
1578 				m->m_pkthdr.csum_data = 0xffff;
1579 			}
1580 		}
1581 
1582 		if (rxvlan & RE_RDESC_VLANCTL_TAG) {
1583 			VLAN_INPUT_TAG(m,
1584 			   be16toh((rxvlan & RE_RDESC_VLANCTL_DATA)));
1585 		} else {
1586 			ifp->if_input(ifp, m);
1587 		}
1588 	}
1589 
1590 	/* Flush the RX DMA ring */
1591 
1592 	bus_dmamap_sync(sc->re_ldata.re_rx_list_tag,
1593 			sc->re_ldata.re_rx_list_map, BUS_DMASYNC_PREWRITE);
1594 
1595 	sc->re_ldata.re_rx_prodidx = i;
1596 }
1597 
1598 static void
1599 re_txeof(struct re_softc *sc)
1600 {
1601 	struct ifnet *ifp = &sc->arpcom.ac_if;
1602 	uint32_t txstat;
1603 	int idx;
1604 
1605 	/* Invalidate the TX descriptor list */
1606 
1607 	bus_dmamap_sync(sc->re_ldata.re_tx_list_tag,
1608 			sc->re_ldata.re_tx_list_map, BUS_DMASYNC_POSTREAD);
1609 
1610 	for (idx = sc->re_ldata.re_tx_considx;
1611 	     sc->re_ldata.re_tx_free < RE_TX_DESC_CNT; RE_DESC_INC(idx)) {
1612 		txstat = le32toh(sc->re_ldata.re_tx_list[idx].re_cmdstat);
1613 		if (txstat & RE_TDESC_CMD_OWN)
1614 			break;
1615 
1616 		sc->re_ldata.re_tx_list[idx].re_bufaddr_lo = 0;
1617 
1618 		/*
1619 		 * We only stash mbufs in the last descriptor
1620 		 * in a fragment chain, which also happens to
1621 		 * be the only place where the TX status bits
1622 		 * are valid.
1623 		 */
1624 		if (txstat & RE_TDESC_CMD_EOF) {
1625 			m_freem(sc->re_ldata.re_tx_mbuf[idx]);
1626 			sc->re_ldata.re_tx_mbuf[idx] = NULL;
1627 			bus_dmamap_unload(sc->re_ldata.re_mtag,
1628 			    sc->re_ldata.re_tx_dmamap[idx]);
1629 			if (txstat & (RE_TDESC_STAT_EXCESSCOL|
1630 			    RE_TDESC_STAT_COLCNT))
1631 				ifp->if_collisions++;
1632 			if (txstat & RE_TDESC_STAT_TXERRSUM)
1633 				ifp->if_oerrors++;
1634 			else
1635 				ifp->if_opackets++;
1636 		}
1637 		sc->re_ldata.re_tx_free++;
1638 	}
1639 
1640 	/* No changes made to the TX ring, so no flush needed */
1641 	if (sc->re_ldata.re_tx_free) {
1642 		sc->re_ldata.re_tx_considx = idx;
1643 		ifp->if_flags &= ~IFF_OACTIVE;
1644 		ifp->if_timer = 0;
1645 	}
1646 
1647 	/*
1648 	 * Some chips will ignore a second TX request issued while an
1649 	 * existing transmission is in progress. If the transmitter goes
1650 	 * idle but there are still packets waiting to be sent, we need
1651 	 * to restart the channel here to flush them out. This only seems
1652 	 * to be required with the PCIe devices.
1653 	 */
1654 	if (sc->re_ldata.re_tx_free < RE_TX_DESC_CNT)
1655 		CSR_WRITE_1(sc, sc->re_txstart, RE_TXSTART_START);
1656 
1657 	/*
1658 	 * If not all descriptors have been released reaped yet,
1659 	 * reload the timer so that we will eventually get another
1660 	 * interrupt that will cause us to re-enter this routine.
1661 	 * This is done in case the transmitter has gone idle.
1662 	 */
1663 	if (RE_TX_MODERATION_IS_ENABLED(sc) &&
1664 	    sc->re_ldata.re_tx_free < RE_TX_DESC_CNT)
1665                 CSR_WRITE_4(sc, RE_TIMERCNT, 1);
1666 }
1667 
1668 static void
1669 re_tick(void *xsc)
1670 {
1671 	struct re_softc *sc = xsc;
1672 
1673 	lwkt_serialize_enter(sc->arpcom.ac_if.if_serializer);
1674 	re_tick_serialized(xsc);
1675 	lwkt_serialize_exit(sc->arpcom.ac_if.if_serializer);
1676 }
1677 
1678 static void
1679 re_tick_serialized(void *xsc)
1680 {
1681 	struct re_softc *sc = xsc;
1682 	struct ifnet *ifp = &sc->arpcom.ac_if;
1683 	struct mii_data *mii;
1684 
1685 	mii = device_get_softc(sc->re_miibus);
1686 	mii_tick(mii);
1687 	if (sc->re_link) {
1688 		if (!(mii->mii_media_status & IFM_ACTIVE))
1689 			sc->re_link = 0;
1690 	} else {
1691 		if (mii->mii_media_status & IFM_ACTIVE &&
1692 		    IFM_SUBTYPE(mii->mii_media_active) != IFM_NONE) {
1693 			sc->re_link = 1;
1694 			if (!ifq_is_empty(&ifp->if_snd))
1695 				ifp->if_start(ifp);
1696 		}
1697 	}
1698 
1699 	callout_reset(&sc->re_timer, hz, re_tick, sc);
1700 }
1701 
1702 #ifdef DEVICE_POLLING
1703 
1704 static void
1705 re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
1706 {
1707 	struct re_softc *sc = ifp->if_softc;
1708 
1709 	switch(cmd) {
1710 	case POLL_REGISTER:
1711 		/* disable interrupts */
1712 		CSR_WRITE_2(sc, RE_IMR, 0x0000);
1713 		break;
1714 	case POLL_DEREGISTER:
1715 		/* enable interrupts */
1716 		CSR_WRITE_2(sc, RE_IMR, sc->re_intrs);
1717 		break;
1718 	default:
1719 		sc->rxcycles = count;
1720 		re_rxeof(sc);
1721 		re_txeof(sc);
1722 
1723 		if (!ifq_is_empty(&ifp->if_snd))
1724 			(*ifp->if_start)(ifp);
1725 
1726 		if (cmd == POLL_AND_CHECK_STATUS) { /* also check status register */
1727 			uint16_t       status;
1728 
1729 			status = CSR_READ_2(sc, RE_ISR);
1730 			if (status == 0xffff)
1731 				return;
1732 			if (status)
1733 				CSR_WRITE_2(sc, RE_ISR, status);
1734 
1735 			/*
1736 			 * XXX check behaviour on receiver stalls.
1737 			 */
1738 
1739 			if (status & RE_ISR_SYSTEM_ERR) {
1740 				re_reset(sc);
1741 				re_init(sc);
1742 			}
1743 		}
1744 		break;
1745 	}
1746 }
1747 #endif /* DEVICE_POLLING */
1748 
1749 static void
1750 re_intr(void *arg)
1751 {
1752 	struct re_softc	*sc = arg;
1753 	struct ifnet *ifp = &sc->arpcom.ac_if;
1754 	uint16_t status;
1755 
1756 	if (sc->suspended || (ifp->if_flags & IFF_UP) == 0)
1757 		return;
1758 
1759 	for (;;) {
1760 		status = CSR_READ_2(sc, RE_ISR);
1761 		/* If the card has gone away the read returns 0xffff. */
1762 		if (status == 0xffff)
1763 			break;
1764 		if (status)
1765 			CSR_WRITE_2(sc, RE_ISR, status);
1766 
1767 		if ((status & sc->re_intrs) == 0)
1768 			break;
1769 
1770 		if (status & (RE_ISR_RX_OK | RE_ISR_RX_ERR | RE_ISR_FIFO_OFLOW))
1771 			re_rxeof(sc);
1772 
1773 		if ((status & sc->re_tx_ack) ||
1774 		    (status & RE_ISR_TX_ERR) ||
1775 		    (status & RE_ISR_TX_DESC_UNAVAIL))
1776 			re_txeof(sc);
1777 
1778 		if (status & RE_ISR_SYSTEM_ERR) {
1779 			re_reset(sc);
1780 			re_init(sc);
1781 		}
1782 
1783 		if (status & RE_ISR_LINKCHG) {
1784 			callout_stop(&sc->re_timer);
1785 			re_tick_serialized(sc);
1786 		}
1787 	}
1788 
1789 	if (!ifq_is_empty(&ifp->if_snd))
1790 		(*ifp->if_start)(ifp);
1791 }
1792 
1793 static int
1794 re_encap(struct re_softc *sc, struct mbuf **m_head, int *idx, int *called_defrag)
1795 {
1796 	struct ifnet *ifp = &sc->arpcom.ac_if;
1797 	struct mbuf *m, *m_new = NULL;
1798 	struct re_dmaload_arg	arg;
1799 	bus_dmamap_t		map;
1800 	int			error;
1801 
1802 	*called_defrag = 0;
1803 	if (sc->re_ldata.re_tx_free <= 4)
1804 		return(EFBIG);
1805 
1806 	m = *m_head;
1807 
1808 	/*
1809 	 * Set up checksum offload. Note: checksum offload bits must
1810 	 * appear in all descriptors of a multi-descriptor transmit
1811 	 * attempt. (This is according to testing done with an 8169
1812 	 * chip. I'm not sure if this is a requirement or a bug.)
1813 	 */
1814 
1815 	arg.re_flags = 0;
1816 
1817 	if (m->m_pkthdr.csum_flags & CSUM_IP)
1818 		arg.re_flags |= RE_TDESC_CMD_IPCSUM;
1819 	if (m->m_pkthdr.csum_flags & CSUM_TCP)
1820 		arg.re_flags |= RE_TDESC_CMD_TCPCSUM;
1821 	if (m->m_pkthdr.csum_flags & CSUM_UDP)
1822 		arg.re_flags |= RE_TDESC_CMD_UDPCSUM;
1823 
1824 	arg.sc = sc;
1825 	arg.re_idx = *idx;
1826 	arg.re_maxsegs = sc->re_ldata.re_tx_free;
1827 	if (arg.re_maxsegs > 4)
1828 		arg.re_maxsegs -= 4;
1829 	arg.re_ring = sc->re_ldata.re_tx_list;
1830 
1831 	map = sc->re_ldata.re_tx_dmamap[*idx];
1832 
1833 	/*
1834 	 * With some of the RealTek chips, using the checksum offload
1835 	 * support in conjunction with the autopadding feature results
1836 	 * in the transmission of corrupt frames. For example, if we
1837 	 * need to send a really small IP fragment that's less than 60
1838 	 * bytes in size, and IP header checksumming is enabled, the
1839 	 * resulting ethernet frame that appears on the wire will
1840 	 * have garbled payload. To work around this, if TX checksum
1841 	 * offload is enabled, we always manually pad short frames out
1842 	 * to the minimum ethernet frame size. We do this by pretending
1843 	 * the mbuf chain has too many fragments so the coalescing code
1844 	 * below can assemble the packet into a single buffer that's
1845 	 * padded out to the mininum frame size.
1846 	 *
1847 	 * Note: this appears unnecessary for TCP, and doing it for TCP
1848 	 * with PCIe adapters seems to result in bad checksums.
1849 	 */
1850 	if (arg.re_flags && !(arg.re_flags & RE_TDESC_CMD_TCPCSUM) &&
1851 	    m->m_pkthdr.len < RE_MIN_FRAMELEN) {
1852 		error = EFBIG;
1853 	} else {
1854 		error = bus_dmamap_load_mbuf(sc->re_ldata.re_mtag, map,
1855 		    m, re_dma_map_desc, &arg, BUS_DMA_NOWAIT);
1856 	}
1857 
1858 	if (error && error != EFBIG) {
1859 		if_printf(ifp, "can't map mbuf (error %d)\n", error);
1860 		return(ENOBUFS);
1861 	}
1862 
1863 	/* Too many segments to map, coalesce into a single mbuf */
1864 
1865 	if (error || arg.re_maxsegs == 0) {
1866 		m_new = m_defrag_nofree(m, MB_DONTWAIT);
1867 		if (m_new == NULL) {
1868 			return(1);
1869 		} else {
1870 			m = m_new;
1871 			*m_head = m;
1872 		}
1873 
1874 		/*
1875 		 * Manually pad short frames, and zero the pad space
1876 		 * to avoid leaking data.
1877 		 */
1878 		if (m_new->m_pkthdr.len < RE_MIN_FRAMELEN) {
1879 			bzero(mtod(m_new, char *) + m_new->m_pkthdr.len,
1880 			    RE_MIN_FRAMELEN - m_new->m_pkthdr.len);
1881 			m_new->m_pkthdr.len += RE_MIN_FRAMELEN -
1882 			    m_new->m_pkthdr.len;
1883 			m_new->m_len = m_new->m_pkthdr.len;
1884 		}
1885 
1886 		*called_defrag = 1;
1887 		arg.sc = sc;
1888 		arg.re_idx = *idx;
1889 		arg.re_maxsegs = sc->re_ldata.re_tx_free;
1890 		arg.re_ring = sc->re_ldata.re_tx_list;
1891 
1892 		error = bus_dmamap_load_mbuf(sc->re_ldata.re_mtag, map,
1893 		    m, re_dma_map_desc, &arg, BUS_DMA_NOWAIT);
1894 		if (error) {
1895 			m_freem(m);
1896 			if_printf(ifp, "can't map mbuf (error %d)\n", error);
1897 			return(EFBIG);
1898 		}
1899 	}
1900 
1901 	/*
1902 	 * Insure that the map for this transmission
1903 	 * is placed at the array index of the last descriptor
1904 	 * in this chain.
1905 	 */
1906 	sc->re_ldata.re_tx_dmamap[*idx] =
1907 	    sc->re_ldata.re_tx_dmamap[arg.re_idx];
1908 	sc->re_ldata.re_tx_dmamap[arg.re_idx] = map;
1909 
1910 	sc->re_ldata.re_tx_mbuf[arg.re_idx] = m;
1911 	sc->re_ldata.re_tx_free -= arg.re_maxsegs;
1912 
1913 	/*
1914 	 * Set up hardware VLAN tagging. Note: vlan tag info must
1915 	 * appear in the first descriptor of a multi-descriptor
1916 	 * transmission attempt.
1917 	 */
1918 
1919 	if ((m->m_flags & (M_PROTO1|M_PKTHDR)) == (M_PROTO1|M_PKTHDR) &&
1920 	    m->m_pkthdr.rcvif != NULL &&
1921 	    m->m_pkthdr.rcvif->if_type == IFT_L2VLAN) {
1922 	    	struct ifvlan *ifv;
1923 		ifv = m->m_pkthdr.rcvif->if_softc;
1924 		if (ifv != NULL)
1925 			sc->re_ldata.re_tx_list[*idx].re_vlanctl =
1926 			    htole32(htobe16(ifv->ifv_tag) | RE_TDESC_VLANCTL_TAG);
1927 	}
1928 
1929 	/* Transfer ownership of packet to the chip. */
1930 
1931 	sc->re_ldata.re_tx_list[arg.re_idx].re_cmdstat |=
1932 	    htole32(RE_TDESC_CMD_OWN);
1933 	if (*idx != arg.re_idx)
1934 		sc->re_ldata.re_tx_list[*idx].re_cmdstat |=
1935 		    htole32(RE_TDESC_CMD_OWN);
1936 
1937 	RE_DESC_INC(arg.re_idx);
1938 	*idx = arg.re_idx;
1939 
1940 	return(0);
1941 }
1942 
1943 /*
1944  * Main transmit routine for C+ and gigE NICs.
1945  */
1946 
1947 static void
1948 re_start(struct ifnet *ifp)
1949 {
1950 	struct re_softc	*sc = ifp->if_softc;
1951 	struct mbuf *m_head;
1952 	struct mbuf *m_head2;
1953 	int called_defrag, idx, need_trans;
1954 
1955 	if (!sc->re_link || (ifp->if_flags & IFF_OACTIVE))
1956 		return;
1957 
1958 	idx = sc->re_ldata.re_tx_prodidx;
1959 
1960 	need_trans = 0;
1961 	while (sc->re_ldata.re_tx_mbuf[idx] == NULL) {
1962 		m_head = ifq_poll(&ifp->if_snd);
1963 		if (m_head == NULL)
1964 			break;
1965 		m_head2 = m_head;
1966 		if (re_encap(sc, &m_head2, &idx, &called_defrag)) {
1967 			/*
1968 			 * If we could not encapsulate the defragged packet,
1969 			 * the returned m_head2 is garbage and we must dequeue
1970 			 * and throw away the original packet.
1971 			 */
1972 			if (called_defrag) {
1973 				ifq_dequeue(&ifp->if_snd, m_head);
1974 				m_freem(m_head);
1975 			}
1976 			ifp->if_flags |= IFF_OACTIVE;
1977 			break;
1978 		}
1979 
1980 		/*
1981 		 * Clean out the packet we encapsulated.  If we defragged
1982 		 * the packet the m_head2 is the one that got encapsulated
1983 		 * and the original must be thrown away.  Otherwise m_head2
1984 		 * *IS* the original.
1985 		 */
1986 		ifq_dequeue(&ifp->if_snd, m_head);
1987 		if (called_defrag)
1988 			m_freem(m_head);
1989 		need_trans = 1;
1990 
1991 		/*
1992 		 * If there's a BPF listener, bounce a copy of this frame
1993 		 * to him.
1994 		 */
1995 		BPF_MTAP(ifp, m_head2);
1996 	}
1997 
1998 	if (!need_trans) {
1999 		if (RE_TX_MODERATION_IS_ENABLED(sc) &&
2000 		    sc->re_ldata.re_tx_free != RE_TX_DESC_CNT)
2001 			CSR_WRITE_4(sc, RE_TIMERCNT, 1);
2002 		return;
2003 	}
2004 
2005 	/* Flush the TX descriptors */
2006 	bus_dmamap_sync(sc->re_ldata.re_tx_list_tag,
2007 			sc->re_ldata.re_tx_list_map, BUS_DMASYNC_PREWRITE);
2008 
2009 	sc->re_ldata.re_tx_prodidx = idx;
2010 
2011 	/*
2012 	 * RealTek put the TX poll request register in a different
2013 	 * location on the 8169 gigE chip. I don't know why.
2014 	 */
2015 	CSR_WRITE_1(sc, sc->re_txstart, RE_TXSTART_START);
2016 
2017 	if (RE_TX_MODERATION_IS_ENABLED(sc)) {
2018 		/*
2019 		 * Use the countdown timer for interrupt moderation.
2020 		 * 'TX done' interrupts are disabled. Instead, we reset the
2021 		 * countdown timer, which will begin counting until it hits
2022 		 * the value in the TIMERINT register, and then trigger an
2023 		 * interrupt. Each time we write to the TIMERCNT register,
2024 		 * the timer count is reset to 0.
2025 		 */
2026 		CSR_WRITE_4(sc, RE_TIMERCNT, 1);
2027 	}
2028 
2029 	/*
2030 	 * Set a timeout in case the chip goes out to lunch.
2031 	 */
2032 	ifp->if_timer = 5;
2033 }
2034 
2035 static void
2036 re_init(void *xsc)
2037 {
2038 	struct re_softc *sc = xsc;
2039 	struct ifnet *ifp = &sc->arpcom.ac_if;
2040 	struct mii_data *mii;
2041 	uint32_t rxcfg = 0;
2042 
2043 	mii = device_get_softc(sc->re_miibus);
2044 
2045 	/*
2046 	 * Cancel pending I/O and free all RX/TX buffers.
2047 	 */
2048 	re_stop(sc);
2049 
2050 	/*
2051 	 * Enable C+ RX and TX mode, as well as VLAN stripping and
2052 	 * RX checksum offload. We must configure the C+ register
2053 	 * before all others.
2054 	 */
2055 	CSR_WRITE_2(sc, RE_CPLUS_CMD, RE_CPLUSCMD_RXENB | RE_CPLUSCMD_TXENB |
2056 		    RE_CPLUSCMD_PCI_MRW | RE_CPLUSCMD_VLANSTRIP |
2057 		    (ifp->if_capenable & IFCAP_RXCSUM ?
2058 		     RE_CPLUSCMD_RXCSUM_ENB : 0));
2059 
2060 	/*
2061 	 * Init our MAC address.  Even though the chipset
2062 	 * documentation doesn't mention it, we need to enter "Config
2063 	 * register write enable" mode to modify the ID registers.
2064 	 */
2065 	CSR_WRITE_1(sc, RE_EECMD, RE_EEMODE_WRITECFG);
2066 	CSR_WRITE_4(sc, RE_IDR0,
2067 	    htole32(*(uint32_t *)(&sc->arpcom.ac_enaddr[0])));
2068 	CSR_WRITE_2(sc, RE_IDR4,
2069 	    htole16(*(uint16_t *)(&sc->arpcom.ac_enaddr[4])));
2070 	CSR_WRITE_1(sc, RE_EECMD, RE_EEMODE_OFF);
2071 
2072 	/*
2073 	 * For C+ mode, initialize the RX descriptors and mbufs.
2074 	 */
2075 	re_rx_list_init(sc);
2076 	re_tx_list_init(sc);
2077 
2078 	/*
2079 	 * Load the addresses of the RX and TX lists into the chip.
2080 	 */
2081 	CSR_WRITE_4(sc, RE_RXLIST_ADDR_HI,
2082 	    RE_ADDR_HI(sc->re_ldata.re_rx_list_addr));
2083 	CSR_WRITE_4(sc, RE_RXLIST_ADDR_LO,
2084 	    RE_ADDR_LO(sc->re_ldata.re_rx_list_addr));
2085 
2086 	CSR_WRITE_4(sc, RE_TXLIST_ADDR_HI,
2087 	    RE_ADDR_HI(sc->re_ldata.re_tx_list_addr));
2088 	CSR_WRITE_4(sc, RE_TXLIST_ADDR_LO,
2089 	    RE_ADDR_LO(sc->re_ldata.re_tx_list_addr));
2090 
2091 	/*
2092 	 * Enable transmit and receive.
2093 	 */
2094 	CSR_WRITE_1(sc, RE_COMMAND, RE_CMD_TX_ENB|RE_CMD_RX_ENB);
2095 
2096 	/*
2097 	 * Set the initial TX and RX configuration.
2098 	 */
2099 	if (sc->re_testmode) {
2100 		if (sc->re_type == RE_8169)
2101 			CSR_WRITE_4(sc, RE_TXCFG,
2102 				    RE_TXCFG_CONFIG | RE_LOOPTEST_ON);
2103 		else
2104 			CSR_WRITE_4(sc, RE_TXCFG,
2105 				    RE_TXCFG_CONFIG | RE_LOOPTEST_ON_CPLUS);
2106 	} else
2107 		CSR_WRITE_4(sc, RE_TXCFG, RE_TXCFG_CONFIG);
2108 
2109 	CSR_WRITE_1(sc, RE_EARLY_TX_THRESH, 16);
2110 
2111 	CSR_WRITE_4(sc, RE_RXCFG, RE_RXCFG_CONFIG);
2112 
2113 	/* Set the individual bit to receive frames for this host only. */
2114 	rxcfg = CSR_READ_4(sc, RE_RXCFG);
2115 	rxcfg |= RE_RXCFG_RX_INDIV;
2116 
2117 	/* If we want promiscuous mode, set the allframes bit. */
2118 	if (ifp->if_flags & IFF_PROMISC) {
2119 		rxcfg |= RE_RXCFG_RX_ALLPHYS;
2120 		CSR_WRITE_4(sc, RE_RXCFG, rxcfg);
2121 	} else {
2122 		rxcfg &= ~RE_RXCFG_RX_ALLPHYS;
2123 		CSR_WRITE_4(sc, RE_RXCFG, rxcfg);
2124 	}
2125 
2126 	/*
2127 	 * Set capture broadcast bit to capture broadcast frames.
2128 	 */
2129 	if (ifp->if_flags & IFF_BROADCAST) {
2130 		rxcfg |= RE_RXCFG_RX_BROAD;
2131 		CSR_WRITE_4(sc, RE_RXCFG, rxcfg);
2132 	} else {
2133 		rxcfg &= ~RE_RXCFG_RX_BROAD;
2134 		CSR_WRITE_4(sc, RE_RXCFG, rxcfg);
2135 	}
2136 
2137 	/*
2138 	 * Program the multicast filter, if necessary.
2139 	 */
2140 	re_setmulti(sc);
2141 
2142 #ifdef DEVICE_POLLING
2143 	/*
2144 	 * Disable interrupts if we are polling.
2145 	 */
2146 	if (ifp->if_flags & IFF_POLLING)
2147 		CSR_WRITE_2(sc, RE_IMR, 0);
2148 	else	/* otherwise ... */
2149 #endif /* DEVICE_POLLING */
2150 	/*
2151 	 * Enable interrupts.
2152 	 */
2153 	if (sc->re_testmode)
2154 		CSR_WRITE_2(sc, RE_IMR, 0);
2155 	else
2156 		CSR_WRITE_2(sc, RE_IMR, sc->re_intrs);
2157 	CSR_WRITE_2(sc, RE_ISR, sc->re_intrs);
2158 
2159 	/* Set initial TX threshold */
2160 	sc->re_txthresh = RE_TX_THRESH_INIT;
2161 
2162 	/* Start RX/TX process. */
2163 	if (sc->re_flags & RE_F_HASMPC)
2164 		CSR_WRITE_4(sc, RE_MISSEDPKT, 0);
2165 #ifdef notdef
2166 	/* Enable receiver and transmitter. */
2167 	CSR_WRITE_1(sc, RE_COMMAND, RE_CMD_TX_ENB|RE_CMD_RX_ENB);
2168 #endif
2169 
2170 	if (RE_TX_MODERATION_IS_ENABLED(sc)) {
2171 		/*
2172 		 * Initialize the timer interrupt register so that
2173 		 * a timer interrupt will be generated once the timer
2174 		 * reaches a certain number of ticks. The timer is
2175 		 * reloaded on each transmit. This gives us TX interrupt
2176 		 * moderation, which dramatically improves TX frame rate.
2177 		 */
2178 		if (sc->re_type == RE_8169)
2179 			CSR_WRITE_4(sc, RE_TIMERINT_8169, 0x800);
2180 		else
2181 			CSR_WRITE_4(sc, RE_TIMERINT, 0x400);
2182 	}
2183 
2184 	/*
2185 	 * For 8169 gigE NICs, set the max allowed RX packet
2186 	 * size so we can receive jumbo frames.
2187 	 */
2188 	if (sc->re_type == RE_8169)
2189 		CSR_WRITE_2(sc, RE_MAXRXPKTLEN, 16383);
2190 
2191 	if (sc->re_testmode) {
2192 		return;
2193 	}
2194 
2195 	mii_mediachg(mii);
2196 
2197 	CSR_WRITE_1(sc, RE_CFG1, RE_CFG1_DRVLOAD|RE_CFG1_FULLDUPLEX);
2198 
2199 	ifp->if_flags |= IFF_RUNNING;
2200 	ifp->if_flags &= ~IFF_OACTIVE;
2201 
2202 	sc->re_link = 0;
2203 	callout_reset(&sc->re_timer, hz, re_tick, sc);
2204 }
2205 
2206 /*
2207  * Set media options.
2208  */
2209 static int
2210 re_ifmedia_upd(struct ifnet *ifp)
2211 {
2212 	struct re_softc *sc = ifp->if_softc;
2213 	struct mii_data *mii;
2214 
2215 	mii = device_get_softc(sc->re_miibus);
2216 	mii_mediachg(mii);
2217 
2218 	return(0);
2219 }
2220 
2221 /*
2222  * Report current media status.
2223  */
2224 static void
2225 re_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
2226 {
2227 	struct re_softc *sc = ifp->if_softc;
2228 	struct mii_data *mii;
2229 
2230 	mii = device_get_softc(sc->re_miibus);
2231 
2232 	mii_pollstat(mii);
2233 	ifmr->ifm_active = mii->mii_media_active;
2234 	ifmr->ifm_status = mii->mii_media_status;
2235 }
2236 
2237 static int
2238 re_ioctl(struct ifnet *ifp, u_long command, caddr_t data, struct ucred *cr)
2239 {
2240 	struct re_softc *sc = ifp->if_softc;
2241 	struct ifreq *ifr = (struct ifreq *) data;
2242 	struct mii_data *mii;
2243 	int error = 0;
2244 
2245 	switch(command) {
2246 	case SIOCSIFMTU:
2247 		if (ifr->ifr_mtu > RE_JUMBO_MTU)
2248 			error = EINVAL;
2249 		ifp->if_mtu = ifr->ifr_mtu;
2250 		break;
2251 	case SIOCSIFFLAGS:
2252 		if (ifp->if_flags & IFF_UP)
2253 			re_init(sc);
2254 		else if (ifp->if_flags & IFF_RUNNING)
2255 			re_stop(sc);
2256 		break;
2257 	case SIOCADDMULTI:
2258 	case SIOCDELMULTI:
2259 		re_setmulti(sc);
2260 		error = 0;
2261 		break;
2262 	case SIOCGIFMEDIA:
2263 	case SIOCSIFMEDIA:
2264 		mii = device_get_softc(sc->re_miibus);
2265 		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, command);
2266 		break;
2267 	case SIOCSIFCAP:
2268 		ifp->if_capenable &= ~(IFCAP_HWCSUM);
2269 		ifp->if_capenable |=
2270 		    ifr->ifr_reqcap & (IFCAP_HWCSUM);
2271 		if (ifp->if_capenable & IFCAP_TXCSUM)
2272 			ifp->if_hwassist = RE_CSUM_FEATURES;
2273 		else
2274 			ifp->if_hwassist = 0;
2275 		if (ifp->if_flags & IFF_RUNNING)
2276 			re_init(sc);
2277 		break;
2278 	default:
2279 		error = ether_ioctl(ifp, command, data);
2280 		break;
2281 	}
2282 	return(error);
2283 }
2284 
2285 static void
2286 re_watchdog(struct ifnet *ifp)
2287 {
2288 	struct re_softc *sc = ifp->if_softc;
2289 
2290 	if_printf(ifp, "watchdog timeout\n");
2291 
2292 	ifp->if_oerrors++;
2293 
2294 	re_txeof(sc);
2295 	re_rxeof(sc);
2296 
2297 	re_init(sc);
2298 
2299 	if (!ifq_is_empty(&ifp->if_snd))
2300 		ifp->if_start(ifp);
2301 }
2302 
2303 /*
2304  * Stop the adapter and free any mbufs allocated to the
2305  * RX and TX lists.
2306  */
2307 static void
2308 re_stop(struct re_softc *sc)
2309 {
2310 	struct ifnet *ifp = &sc->arpcom.ac_if;
2311 	int i;
2312 
2313 	ifp->if_timer = 0;
2314 	callout_stop(&sc->re_timer);
2315 
2316 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
2317 
2318 	CSR_WRITE_1(sc, RE_COMMAND, 0x00);
2319 	CSR_WRITE_2(sc, RE_IMR, 0x0000);
2320 	CSR_WRITE_2(sc, RE_ISR, 0xFFFF);
2321 
2322 	if (sc->re_head != NULL) {
2323 		m_freem(sc->re_head);
2324 		sc->re_head = sc->re_tail = NULL;
2325 	}
2326 
2327 	/* Free the TX list buffers. */
2328 	for (i = 0; i < RE_TX_DESC_CNT; i++) {
2329 		if (sc->re_ldata.re_tx_mbuf[i] != NULL) {
2330 			bus_dmamap_unload(sc->re_ldata.re_mtag,
2331 					  sc->re_ldata.re_tx_dmamap[i]);
2332 			m_freem(sc->re_ldata.re_tx_mbuf[i]);
2333 			sc->re_ldata.re_tx_mbuf[i] = NULL;
2334 		}
2335 	}
2336 
2337 	/* Free the RX list buffers. */
2338 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
2339 		if (sc->re_ldata.re_rx_mbuf[i] != NULL) {
2340 			bus_dmamap_unload(sc->re_ldata.re_mtag,
2341 					  sc->re_ldata.re_rx_dmamap[i]);
2342 			m_freem(sc->re_ldata.re_rx_mbuf[i]);
2343 			sc->re_ldata.re_rx_mbuf[i] = NULL;
2344 		}
2345 	}
2346 }
2347 
2348 /*
2349  * Device suspend routine.  Stop the interface and save some PCI
2350  * settings in case the BIOS doesn't restore them properly on
2351  * resume.
2352  */
2353 static int
2354 re_suspend(device_t dev)
2355 {
2356 #ifndef BURN_BRIDGES
2357 	int i;
2358 #endif
2359 	struct re_softc *sc = device_get_softc(dev);
2360 
2361 	re_stop(sc);
2362 
2363 #ifndef BURN_BRIDGES
2364 	for (i = 0; i < 5; i++)
2365 		sc->saved_maps[i] = pci_read_config(dev, PCIR_MAPS + i * 4, 4);
2366 	sc->saved_biosaddr = pci_read_config(dev, PCIR_BIOS, 4);
2367 	sc->saved_intline = pci_read_config(dev, PCIR_INTLINE, 1);
2368 	sc->saved_cachelnsz = pci_read_config(dev, PCIR_CACHELNSZ, 1);
2369 	sc->saved_lattimer = pci_read_config(dev, PCIR_LATTIMER, 1);
2370 #endif
2371 
2372 	sc->suspended = 1;
2373 
2374 	return (0);
2375 }
2376 
2377 /*
2378  * Device resume routine.  Restore some PCI settings in case the BIOS
2379  * doesn't, re-enable busmastering, and restart the interface if
2380  * appropriate.
2381  */
2382 static int
2383 re_resume(device_t dev)
2384 {
2385 	struct re_softc *sc = device_get_softc(dev);
2386 	struct ifnet *ifp = &sc->arpcom.ac_if;
2387 #ifndef BURN_BRIDGES
2388 	int i;
2389 #endif
2390 
2391 #ifndef BURN_BRIDGES
2392 	/* better way to do this? */
2393 	for (i = 0; i < 5; i++)
2394 		pci_write_config(dev, PCIR_MAPS + i * 4, sc->saved_maps[i], 4);
2395 	pci_write_config(dev, PCIR_BIOS, sc->saved_biosaddr, 4);
2396 	pci_write_config(dev, PCIR_INTLINE, sc->saved_intline, 1);
2397 	pci_write_config(dev, PCIR_CACHELNSZ, sc->saved_cachelnsz, 1);
2398 	pci_write_config(dev, PCIR_LATTIMER, sc->saved_lattimer, 1);
2399 
2400 	/* reenable busmastering */
2401 	pci_enable_busmaster(dev);
2402 	pci_enable_io(dev, SYS_RES_IOPORT);
2403 #endif
2404 
2405 	/* reinitialize interface if necessary */
2406 	if (ifp->if_flags & IFF_UP)
2407 		re_init(sc);
2408 
2409 	sc->suspended = 0;
2410 
2411 	return (0);
2412 }
2413 
2414 /*
2415  * Stop all chip I/O so that the kernel's probe routines don't
2416  * get confused by errant DMAs when rebooting.
2417  */
2418 static void
2419 re_shutdown(device_t dev)
2420 {
2421 	struct re_softc *sc = device_get_softc(dev);
2422 	struct ifnet *ifp = &sc->arpcom.ac_if;
2423 
2424 	lwkt_serialize_enter(ifp->if_serializer);
2425 	re_stop(sc);
2426 	lwkt_serialize_exit(ifp->if_serializer);
2427 }
2428 
2429 static int
2430 re_sysctl_tx_moderation(SYSCTL_HANDLER_ARGS)
2431 {
2432 	struct re_softc *sc = arg1;
2433 	struct ifnet *ifp = &sc->arpcom.ac_if;
2434 	int error = 0, mod, mod_old;
2435 
2436 	lwkt_serialize_enter(ifp->if_serializer);
2437 
2438 	mod_old = mod = RE_TX_MODERATION_IS_ENABLED(sc);
2439 
2440 	error = sysctl_handle_int(oidp, &mod, 0, req);
2441 	if (error || req->newptr == NULL || mod == mod_old)
2442 		goto back;
2443 	if (mod != 0 && mod != 1) {
2444 		error = EINVAL;
2445 		goto back;
2446 	}
2447 
2448 	if (mod)
2449 		RE_ENABLE_TX_MODERATION(sc);
2450 	else
2451 		RE_DISABLE_TX_MODERATION(sc);
2452 
2453 	if ((ifp->if_flags & (IFF_RUNNING | IFF_UP)) == (IFF_RUNNING | IFF_UP))
2454 		re_init(sc);
2455 back:
2456 	lwkt_serialize_exit(ifp->if_serializer);
2457 	return error;
2458 }
2459