xref: /dflybsd-src/sys/dev/netif/re/if_re.c (revision 6ea70f7669242fe9fd042834f455f06a4adfaadf)
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 1.28 2006/11/14 13:35:49 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 /*
153  * The hardware supports checksumming but, as usual, some chipsets screw it
154  * all up and produce bogus packets, so we disable it by default.
155  */
156 #define RE_CSUM_FEATURES    (CSUM_IP | CSUM_TCP | CSUM_UDP)
157 #if 0
158 #define RE_DISABLE_HWCSUM
159 #endif
160 
161 /*
162  * Various supported device vendors/types and their names.
163  */
164 static struct re_type re_devs[] = {
165 	{ PCI_VENDOR_DLINK, PCI_PRODUCT_DLINK_DGE528T, RE_HWREV_8169S,
166 		"D-Link DGE-528(T) Gigabit Ethernet Adapter" },
167 	{ PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8139, RE_HWREV_8139CPLUS,
168 		"RealTek 8139C+ 10/100BaseTX" },
169 	{ PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8101E, RE_HWREV_8101E,
170 		"RealTek 8101E PCIe 10/100baseTX" },
171 	{ PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8168, RE_HWREV_8168_SPIN1,
172 		"RealTek 8168/8111B PCIe Gigabit Ethernet" },
173 	{ PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8168, RE_HWREV_8168_SPIN2,
174 		"RealTek 8168/8111B PCIe Gigabit Ethernet" },
175 	{ PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8169, RE_HWREV_8169,
176 		"RealTek 8169 Gigabit Ethernet" },
177 	{ PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8169, RE_HWREV_8169S,
178 		"RealTek 8169S Single-chip Gigabit Ethernet" },
179 	{ PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8169, RE_HWREV_8169_8110SB,
180 		"RealTek 8169SB/8110SB Single-chip Gigabit Ethernet" },
181 	{ PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8169SC, RE_HWREV_8169_8110SC,
182 		"RealTek 8169SC/8110SC Single-chip Gigabit Ethernet" },
183 	{ PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8169, RE_HWREV_8110S,
184 		"RealTek 8110S Single-chip Gigabit Ethernet" },
185 	{ PCI_VENDOR_COREGA, PCI_PRODUCT_COREGA_CG_LAPCIGT, RE_HWREV_8169S,
186 		"Corega CG-LAPCIGT Gigabit Ethernet" },
187 	{ PCI_VENDOR_LINKSYS, PCI_PRODUCT_LINKSYS_EG1032, RE_HWREV_8169S,
188 		"Linksys EG1032 Gigabit Ethernet" },
189 	{ PCI_VENDOR_USR2, PCI_PRODUCT_USR2_997902, RE_HWREV_8169S,
190 		"US Robotics 997902 Gigabit Ethernet" },
191 	{ 0, 0, 0, NULL }
192 };
193 
194 static struct re_hwrev re_hwrevs[] = {
195 	{ RE_HWREV_8139CPLUS, RE_8139CPLUS, "C+"},
196 	{ RE_HWREV_8168_SPIN1, RE_8169, "8168"},
197 	{ RE_HWREV_8169, RE_8169, "8169"},
198 	{ RE_HWREV_8169S, RE_8169, "8169S"},
199 	{ RE_HWREV_8110S, RE_8169, "8110S"},
200 	{ RE_HWREV_8169_8110SB, RE_8169, "8169SB"},
201 	{ RE_HWREV_8169_8110SC, RE_8169, "8169SC"},
202 	{ RE_HWREV_8100E, RE_8169, "8100E"},
203 	{ RE_HWREV_8101E, RE_8169, "8101E"},
204 	{ RE_HWREV_8168_SPIN2, RE_8169, "8168"},
205 	{ 0, 0, NULL }
206 };
207 
208 static int	re_probe(device_t);
209 static int	re_attach(device_t);
210 static int	re_detach(device_t);
211 
212 static int	re_encap(struct re_softc *, struct mbuf **, int *, int *);
213 
214 static void	re_dma_map_addr(void *, bus_dma_segment_t *, int, int);
215 static void	re_dma_map_desc(void *, bus_dma_segment_t *, int,
216 				bus_size_t, int);
217 static int	re_allocmem(device_t, struct re_softc *);
218 static int	re_newbuf(struct re_softc *, int, struct mbuf *);
219 static int	re_rx_list_init(struct re_softc *);
220 static int	re_tx_list_init(struct re_softc *);
221 static void	re_rxeof(struct re_softc *);
222 static void	re_txeof(struct re_softc *);
223 static void	re_intr(void *);
224 static void	re_tick(void *);
225 static void	re_tick_serialized(void *);
226 static void	re_start(struct ifnet *);
227 static int	re_ioctl(struct ifnet *, u_long, caddr_t, struct ucred *);
228 static void	re_init(void *);
229 static void	re_stop(struct re_softc *);
230 static void	re_watchdog(struct ifnet *);
231 static int	re_suspend(device_t);
232 static int	re_resume(device_t);
233 static void	re_shutdown(device_t);
234 static int	re_ifmedia_upd(struct ifnet *);
235 static void	re_ifmedia_sts(struct ifnet *, struct ifmediareq *);
236 
237 static void	re_eeprom_putbyte(struct re_softc *, int);
238 static void	re_eeprom_getword(struct re_softc *, int, u_int16_t *);
239 static void	re_read_eeprom(struct re_softc *, caddr_t, int, int);
240 static int	re_gmii_readreg(device_t, int, int);
241 static int	re_gmii_writereg(device_t, int, int, int);
242 
243 static int	re_miibus_readreg(device_t, int, int);
244 static int	re_miibus_writereg(device_t, int, int, int);
245 static void	re_miibus_statchg(device_t);
246 
247 static void	re_setmulti(struct re_softc *);
248 static void	re_reset(struct re_softc *);
249 
250 #ifdef RE_DIAG
251 static int	re_diag(struct re_softc *);
252 #endif
253 
254 #ifdef DEVICE_POLLING
255 static void	re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count);
256 #endif
257 
258 static int	re_sysctl_tx_moderation(SYSCTL_HANDLER_ARGS);
259 
260 static device_method_t re_methods[] = {
261 	/* Device interface */
262 	DEVMETHOD(device_probe,		re_probe),
263 	DEVMETHOD(device_attach,	re_attach),
264 	DEVMETHOD(device_detach,	re_detach),
265 	DEVMETHOD(device_suspend,	re_suspend),
266 	DEVMETHOD(device_resume,	re_resume),
267 	DEVMETHOD(device_shutdown,	re_shutdown),
268 
269 	/* bus interface */
270 	DEVMETHOD(bus_print_child,	bus_generic_print_child),
271 	DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
272 
273 	/* MII interface */
274 	DEVMETHOD(miibus_readreg,	re_miibus_readreg),
275 	DEVMETHOD(miibus_writereg,	re_miibus_writereg),
276 	DEVMETHOD(miibus_statchg,	re_miibus_statchg),
277 
278 	{ 0, 0 }
279 };
280 
281 static driver_t re_driver = {
282 	"re",
283 	re_methods,
284 	sizeof(struct re_softc)
285 };
286 
287 static devclass_t re_devclass;
288 
289 DECLARE_DUMMY_MODULE(if_re);
290 DRIVER_MODULE(if_re, pci, re_driver, re_devclass, 0, 0);
291 DRIVER_MODULE(if_re, cardbus, re_driver, re_devclass, 0, 0);
292 DRIVER_MODULE(miibus, re, miibus_driver, miibus_devclass, 0, 0);
293 
294 #define EE_SET(x)	\
295 	CSR_WRITE_1(sc, RE_EECMD, CSR_READ_1(sc, RE_EECMD) | (x))
296 
297 #define EE_CLR(x)	\
298 	CSR_WRITE_1(sc, RE_EECMD, CSR_READ_1(sc, RE_EECMD) & ~(x))
299 
300 /*
301  * Send a read command and address to the EEPROM, check for ACK.
302  */
303 static void
304 re_eeprom_putbyte(struct re_softc *sc, int addr)
305 {
306 	int d, i;
307 
308 	d = addr | (RE_9346_READ << sc->re_eewidth);
309 
310 	/*
311 	 * Feed in each bit and strobe the clock.
312 	 */
313 	for (i = 1 << (sc->re_eewidth + 3); i; i >>= 1) {
314 		if (d & i)
315 			EE_SET(RE_EE_DATAIN);
316 		else
317 			EE_CLR(RE_EE_DATAIN);
318 		DELAY(100);
319 		EE_SET(RE_EE_CLK);
320 		DELAY(150);
321 		EE_CLR(RE_EE_CLK);
322 		DELAY(100);
323 	}
324 }
325 
326 /*
327  * Read a word of data stored in the EEPROM at address 'addr.'
328  */
329 static void
330 re_eeprom_getword(struct re_softc *sc, int addr, uint16_t *dest)
331 {
332 	int i;
333 	uint16_t word = 0;
334 
335 	/*
336 	 * Send address of word we want to read.
337 	 */
338 	re_eeprom_putbyte(sc, addr);
339 
340 	/*
341 	 * Start reading bits from EEPROM.
342 	 */
343 	for (i = 0x8000; i != 0; i >>= 1) {
344 		EE_SET(RE_EE_CLK);
345 		DELAY(100);
346 		if (CSR_READ_1(sc, RE_EECMD) & RE_EE_DATAOUT)
347 			word |= i;
348 		EE_CLR(RE_EE_CLK);
349 		DELAY(100);
350 	}
351 
352 	*dest = word;
353 }
354 
355 /*
356  * Read a sequence of words from the EEPROM.
357  */
358 static void
359 re_read_eeprom(struct re_softc *sc, caddr_t dest, int off, int cnt)
360 {
361 	int i;
362 	uint16_t word = 0, *ptr;
363 
364 	CSR_SETBIT_1(sc, RE_EECMD, RE_EEMODE_PROGRAM);
365 	DELAY(100);
366 
367 	for (i = 0; i < cnt; i++) {
368 		CSR_SETBIT_1(sc, RE_EECMD, RE_EE_SEL);
369 		re_eeprom_getword(sc, off + i, &word);
370 		CSR_CLRBIT_1(sc, RE_EECMD, RE_EE_SEL);
371 		ptr = (uint16_t *)(dest + (i * 2));
372 		*ptr = word;
373 	}
374 
375 	CSR_CLRBIT_1(sc, RE_EECMD, RE_EEMODE_PROGRAM);
376 }
377 
378 static int
379 re_gmii_readreg(device_t dev, int phy, int reg)
380 {
381 	struct re_softc *sc = device_get_softc(dev);
382 	u_int32_t rval;
383 	int i;
384 
385 	if (phy != 1)
386 		return(0);
387 
388 	/* Let the rgephy driver read the GMEDIASTAT register */
389 
390 	if (reg == RE_GMEDIASTAT)
391 		return(CSR_READ_1(sc, RE_GMEDIASTAT));
392 
393 	CSR_WRITE_4(sc, RE_PHYAR, reg << 16);
394 	DELAY(1000);
395 
396 	for (i = 0; i < RE_TIMEOUT; i++) {
397 		rval = CSR_READ_4(sc, RE_PHYAR);
398 		if (rval & RE_PHYAR_BUSY)
399 			break;
400 		DELAY(100);
401 	}
402 
403 	if (i == RE_TIMEOUT) {
404 		device_printf(dev, "PHY read failed\n");
405 		return(0);
406 	}
407 
408 	return(rval & RE_PHYAR_PHYDATA);
409 }
410 
411 static int
412 re_gmii_writereg(device_t dev, int phy, int reg, int data)
413 {
414 	struct re_softc *sc = device_get_softc(dev);
415 	uint32_t rval;
416 	int i;
417 
418 	CSR_WRITE_4(sc, RE_PHYAR,
419 		    (reg << 16) | (data & RE_PHYAR_PHYDATA) | RE_PHYAR_BUSY);
420 	DELAY(1000);
421 
422 	for (i = 0; i < RE_TIMEOUT; i++) {
423 		rval = CSR_READ_4(sc, RE_PHYAR);
424 		if ((rval & RE_PHYAR_BUSY) == 0)
425 			break;
426 		DELAY(100);
427 	}
428 
429 	if (i == RE_TIMEOUT)
430 		device_printf(dev, "PHY write failed\n");
431 
432 	return(0);
433 }
434 
435 static int
436 re_miibus_readreg(device_t dev, int phy, int reg)
437 {
438 	struct re_softc	*sc = device_get_softc(dev);
439 	uint16_t rval = 0;
440 	uint16_t re8139_reg = 0;
441 
442 	if (sc->re_type == RE_8169) {
443 		rval = re_gmii_readreg(dev, phy, reg);
444 		return(rval);
445 	}
446 
447 	/* Pretend the internal PHY is only at address 0 */
448 	if (phy)
449 		return(0);
450 
451 	switch(reg) {
452 	case MII_BMCR:
453 		re8139_reg = RE_BMCR;
454 		break;
455 	case MII_BMSR:
456 		re8139_reg = RE_BMSR;
457 		break;
458 	case MII_ANAR:
459 		re8139_reg = RE_ANAR;
460 		break;
461 	case MII_ANER:
462 		re8139_reg = RE_ANER;
463 		break;
464 	case MII_ANLPAR:
465 		re8139_reg = RE_LPAR;
466 		break;
467 	case MII_PHYIDR1:
468 	case MII_PHYIDR2:
469 		return(0);
470 	/*
471 	 * Allow the rlphy driver to read the media status
472 	 * register. If we have a link partner which does not
473 	 * support NWAY, this is the register which will tell
474 	 * us the results of parallel detection.
475 	 */
476 	case RE_MEDIASTAT:
477 		return(CSR_READ_1(sc, RE_MEDIASTAT));
478 	default:
479 		device_printf(dev, "bad phy register\n");
480 		return(0);
481 	}
482 	rval = CSR_READ_2(sc, re8139_reg);
483 	if (sc->re_type == RE_8139CPLUS && re8139_reg == RE_BMCR) {
484 		/* 8139C+ has different bit layout. */
485 		rval &= ~(BMCR_LOOP | BMCR_ISO);
486 	}
487 	return(rval);
488 }
489 
490 static int
491 re_miibus_writereg(device_t dev, int phy, int reg, int data)
492 {
493 	struct re_softc *sc= device_get_softc(dev);
494 	u_int16_t re8139_reg = 0;
495 
496 	if (sc->re_type == RE_8169)
497 		return(re_gmii_writereg(dev, phy, reg, data));
498 
499 	/* Pretend the internal PHY is only at address 0 */
500 	if (phy)
501 		return(0);
502 
503 	switch(reg) {
504 	case MII_BMCR:
505 		re8139_reg = RE_BMCR;
506 		if (sc->re_type == RE_8139CPLUS) {
507 			/* 8139C+ has different bit layout. */
508 			data &= ~(BMCR_LOOP | BMCR_ISO);
509 		}
510 		break;
511 	case MII_BMSR:
512 		re8139_reg = RE_BMSR;
513 		break;
514 	case MII_ANAR:
515 		re8139_reg = RE_ANAR;
516 		break;
517 	case MII_ANER:
518 		re8139_reg = RE_ANER;
519 		break;
520 	case MII_ANLPAR:
521 		re8139_reg = RE_LPAR;
522 		break;
523 	case MII_PHYIDR1:
524 	case MII_PHYIDR2:
525 		return(0);
526 	default:
527 		device_printf(dev, "bad phy register\n");
528 		return(0);
529 	}
530 	CSR_WRITE_2(sc, re8139_reg, data);
531 	return(0);
532 }
533 
534 static void
535 re_miibus_statchg(device_t dev)
536 {
537 }
538 
539 /*
540  * Program the 64-bit multicast hash filter.
541  */
542 static void
543 re_setmulti(struct re_softc *sc)
544 {
545 	struct ifnet *ifp = &sc->arpcom.ac_if;
546 	int h = 0;
547 	uint32_t hashes[2] = { 0, 0 };
548 	struct ifmultiaddr *ifma;
549 	uint32_t rxfilt;
550 	int mcnt = 0;
551 
552 	rxfilt = CSR_READ_4(sc, RE_RXCFG);
553 
554 	if (ifp->if_flags & IFF_ALLMULTI || ifp->if_flags & IFF_PROMISC) {
555 		rxfilt |= RE_RXCFG_RX_MULTI;
556 		CSR_WRITE_4(sc, RE_RXCFG, rxfilt);
557 		CSR_WRITE_4(sc, RE_MAR0, 0xFFFFFFFF);
558 		CSR_WRITE_4(sc, RE_MAR4, 0xFFFFFFFF);
559 		return;
560 	}
561 
562 	/* first, zot all the existing hash bits */
563 	CSR_WRITE_4(sc, RE_MAR0, 0);
564 	CSR_WRITE_4(sc, RE_MAR4, 0);
565 
566 	/* now program new ones */
567 	LIST_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
568 		if (ifma->ifma_addr->sa_family != AF_LINK)
569 			continue;
570 		h = ether_crc32_be(LLADDR((struct sockaddr_dl *)
571 		    ifma->ifma_addr), ETHER_ADDR_LEN) >> 26;
572 		if (h < 32)
573 			hashes[0] |= (1 << h);
574 		else
575 			hashes[1] |= (1 << (h - 32));
576 		mcnt++;
577 	}
578 
579 	if (mcnt)
580 		rxfilt |= RE_RXCFG_RX_MULTI;
581 	else
582 		rxfilt &= ~RE_RXCFG_RX_MULTI;
583 
584 	CSR_WRITE_4(sc, RE_RXCFG, rxfilt);
585 	CSR_WRITE_4(sc, RE_MAR0, hashes[0]);
586 	CSR_WRITE_4(sc, RE_MAR4, hashes[1]);
587 }
588 
589 static void
590 re_reset(struct re_softc *sc)
591 {
592 	int i;
593 
594 	CSR_WRITE_1(sc, RE_COMMAND, RE_CMD_RESET);
595 
596 	for (i = 0; i < RE_TIMEOUT; i++) {
597 		DELAY(10);
598 		if ((CSR_READ_1(sc, RE_COMMAND) & RE_CMD_RESET) == 0)
599 			break;
600 	}
601 	if (i == RE_TIMEOUT)
602 		if_printf(&sc->arpcom.ac_if, "reset never completed!\n");
603 
604 	CSR_WRITE_1(sc, 0x82, 1);
605 }
606 
607 #ifdef RE_DIAG
608 /*
609  * The following routine is designed to test for a defect on some
610  * 32-bit 8169 cards. Some of these NICs have the REQ64# and ACK64#
611  * lines connected to the bus, however for a 32-bit only card, they
612  * should be pulled high. The result of this defect is that the
613  * NIC will not work right if you plug it into a 64-bit slot: DMA
614  * operations will be done with 64-bit transfers, which will fail
615  * because the 64-bit data lines aren't connected.
616  *
617  * There's no way to work around this (short of talking a soldering
618  * iron to the board), however we can detect it. The method we use
619  * here is to put the NIC into digital loopback mode, set the receiver
620  * to promiscuous mode, and then try to send a frame. We then compare
621  * the frame data we sent to what was received. If the data matches,
622  * then the NIC is working correctly, otherwise we know the user has
623  * a defective NIC which has been mistakenly plugged into a 64-bit PCI
624  * slot. In the latter case, there's no way the NIC can work correctly,
625  * so we print out a message on the console and abort the device attach.
626  */
627 
628 static int
629 re_diag(struct re_softc *sc)
630 {
631 	struct ifnet *ifp = &sc->arpcom.ac_if;
632 	struct mbuf *m0;
633 	struct ether_header *eh;
634 	struct re_desc *cur_rx;
635 	uint16_t status;
636 	uint32_t rxstat;
637 	int total_len, i, error = 0, phyaddr;
638 	uint8_t dst[ETHER_ADDR_LEN] = { 0x00, 'h', 'e', 'l', 'l', 'o' };
639 	uint8_t src[ETHER_ADDR_LEN] = { 0x00, 'w', 'o', 'r', 'l', 'd' };
640 
641 	/* Allocate a single mbuf */
642 
643 	MGETHDR(m0, MB_DONTWAIT, MT_DATA);
644 	if (m0 == NULL)
645 		return(ENOBUFS);
646 
647 	/*
648 	 * Initialize the NIC in test mode. This sets the chip up
649 	 * so that it can send and receive frames, but performs the
650 	 * following special functions:
651 	 * - Puts receiver in promiscuous mode
652 	 * - Enables digital loopback mode
653 	 * - Leaves interrupts turned off
654 	 */
655 
656 	ifp->if_flags |= IFF_PROMISC;
657 	sc->re_testmode = 1;
658 	re_reset(sc);
659 	re_init(sc);
660 	sc->re_link = 1;
661 	if (sc->re_type == RE_8169)
662 		phyaddr = 1;
663 	else
664 		phyaddr = 0;
665 
666 	re_miibus_writereg(sc->re_dev, phyaddr, MII_BMCR, BMCR_RESET);
667 	for (i = 0; i < RE_TIMEOUT; i++) {
668 		status = re_miibus_readreg(sc->re_dev, phyaddr, MII_BMCR);
669 		if (!(status & BMCR_RESET))
670 			break;
671 	}
672 
673 	re_miibus_writereg(sc->re_dev, phyaddr, MII_BMCR, BMCR_LOOP);
674 	CSR_WRITE_2(sc, RE_ISR, RE_INTRS_DIAG);
675 
676 	DELAY(100000);
677 
678 	/* Put some data in the mbuf */
679 
680 	eh = mtod(m0, struct ether_header *);
681 	bcopy (dst, eh->ether_dhost, ETHER_ADDR_LEN);
682 	bcopy (src, eh->ether_shost, ETHER_ADDR_LEN);
683 	eh->ether_type = htons(ETHERTYPE_IP);
684 	m0->m_pkthdr.len = m0->m_len = ETHER_MIN_LEN - ETHER_CRC_LEN;
685 
686 	/*
687 	 * Queue the packet, start transmission.
688 	 * Note: ifq_handoff() ultimately calls re_start() for us.
689 	 */
690 
691 	CSR_WRITE_2(sc, RE_ISR, 0xFFFF);
692 	error = ifq_handoff(ifp, m0, NULL);
693 	if (error) {
694 		m0 = NULL;
695 		goto done;
696 	}
697 	m0 = NULL;
698 
699 	/* Wait for it to propagate through the chip */
700 
701 	DELAY(100000);
702 	for (i = 0; i < RE_TIMEOUT; i++) {
703 		status = CSR_READ_2(sc, RE_ISR);
704 		CSR_WRITE_2(sc, RE_ISR, status);
705 		if ((status & (RE_ISR_TIMEOUT_EXPIRED|RE_ISR_RX_OK)) ==
706 		    (RE_ISR_TIMEOUT_EXPIRED|RE_ISR_RX_OK))
707 			break;
708 		DELAY(10);
709 	}
710 
711 	if (i == RE_TIMEOUT) {
712 		if_printf(ifp, "diagnostic failed to receive packet "
713 			  "in loopback mode\n");
714 		error = EIO;
715 		goto done;
716 	}
717 
718 	/*
719 	 * The packet should have been dumped into the first
720 	 * entry in the RX DMA ring. Grab it from there.
721 	 */
722 
723 	bus_dmamap_sync(sc->re_ldata.re_rx_list_tag,
724 			sc->re_ldata.re_rx_list_map, BUS_DMASYNC_POSTREAD);
725 	bus_dmamap_sync(sc->re_ldata.re_mtag, sc->re_ldata.re_rx_dmamap[0],
726 			BUS_DMASYNC_POSTWRITE);
727 	bus_dmamap_unload(sc->re_ldata.re_mtag, sc->re_ldata.re_rx_dmamap[0]);
728 
729 	m0 = sc->re_ldata.re_rx_mbuf[0];
730 	sc->re_ldata.re_rx_mbuf[0] = NULL;
731 	eh = mtod(m0, struct ether_header *);
732 
733 	cur_rx = &sc->re_ldata.re_rx_list[0];
734 	total_len = RE_RXBYTES(cur_rx);
735 	rxstat = le32toh(cur_rx->re_cmdstat);
736 
737 	if (total_len != ETHER_MIN_LEN) {
738 		if_printf(ifp, "diagnostic failed, received short packet\n");
739 		error = EIO;
740 		goto done;
741 	}
742 
743 	/* Test that the received packet data matches what we sent. */
744 
745 	if (bcmp(eh->ether_dhost, dst, ETHER_ADDR_LEN) ||
746 	    bcmp(eh->ether_shost, &src, ETHER_ADDR_LEN) ||
747 	    be16toh(eh->ether_type) != ETHERTYPE_IP) {
748 		if_printf(ifp, "WARNING, DMA FAILURE!\n");
749 		if_printf(ifp, "expected TX data: %6D/%6D/0x%x\n",
750 		    dst, ":", src, ":", ETHERTYPE_IP);
751 		if_printf(ifp, "received RX data: %6D/%6D/0x%x\n",
752 		    eh->ether_dhost, ":",  eh->ether_shost, ":",
753 		    ntohs(eh->ether_type));
754 		if_printf(ifp, "You may have a defective 32-bit NIC plugged "
755 		    "into a 64-bit PCI slot.\n");
756 		if_printf(ifp, "Please re-install the NIC in a 32-bit slot "
757 		    "for proper operation.\n");
758 		if_printf(ifp, "Read the re(4) man page for more details.\n");
759 		error = EIO;
760 	}
761 
762 done:
763 	/* Turn interface off, release resources */
764 
765 	sc->re_testmode = 0;
766 	sc->re_link = 0;
767 	ifp->if_flags &= ~IFF_PROMISC;
768 	re_stop(sc);
769 	if (m0 != NULL)
770 		m_freem(m0);
771 
772 	return (error);
773 }
774 #endif	/* RE_DIAG */
775 
776 /*
777  * Probe for a RealTek 8139C+/8169/8110 chip. Check the PCI vendor and device
778  * IDs against our list and return a device name if we find a match.
779  */
780 static int
781 re_probe(device_t dev)
782 {
783 	struct re_type *t;
784 	struct re_softc *sc;
785 	int rid;
786 	uint32_t hwrev;
787 	uint16_t vendor, product;
788 
789 	t = re_devs;
790 
791 	vendor = pci_get_vendor(dev);
792 	product = pci_get_device(dev);
793 
794 	/*
795 	 * Only attach to rev.3 of the Linksys EG1032 adapter.
796 	 * Rev.2 is supported by sk(4).
797 	 */
798 	if (vendor == PCI_VENDOR_LINKSYS &&
799 	    product == PCI_PRODUCT_LINKSYS_EG1032 &&
800 	    pci_get_subdevice(dev) != PCI_SUBDEVICE_LINKSYS_EG1032_REV3)
801 			return ENXIO;
802 
803 	for (t = re_devs; t->re_name != NULL; t++) {
804 		if (product == t->re_did && vendor == t->re_vid)
805 			break;
806 	}
807 
808 	/*
809 	 * Check if we found a RealTek device.
810 	 */
811 	if (t->re_name == NULL)
812 		return(ENXIO);
813 
814 	/*
815 	 * Temporarily map the I/O space so we can read the chip ID register.
816 	 */
817 	sc = kmalloc(sizeof(*sc), M_TEMP, M_WAITOK | M_ZERO);
818 	rid = RE_PCI_LOIO;
819 	sc->re_res = bus_alloc_resource_any(dev, SYS_RES_IOPORT, &rid,
820 					    RF_ACTIVE);
821 	if (sc->re_res == NULL) {
822 		device_printf(dev, "couldn't map ports/memory\n");
823 		kfree(sc, M_TEMP);
824 		return(ENXIO);
825 	}
826 
827 	sc->re_btag = rman_get_bustag(sc->re_res);
828 	sc->re_bhandle = rman_get_bushandle(sc->re_res);
829 
830 	hwrev = CSR_READ_4(sc, RE_TXCFG) & RE_TXCFG_HWREV;
831 	bus_release_resource(dev, SYS_RES_IOPORT, RE_PCI_LOIO, sc->re_res);
832 	kfree(sc, M_TEMP);
833 
834 	/*
835 	 * and continue matching for the specific chip...
836 	 */
837 	for (; t->re_name != NULL; t++) {
838 		if (product == t->re_did && vendor == t->re_vid &&
839 		    t->re_basetype == hwrev) {
840 			device_set_desc(dev, t->re_name);
841 			return(0);
842 		}
843 	}
844 	return(ENXIO);
845 }
846 
847 /*
848  * This routine takes the segment list provided as the result of
849  * a bus_dma_map_load() operation and assigns the addresses/lengths
850  * to RealTek DMA descriptors. This can be called either by the RX
851  * code or the TX code. In the RX case, we'll probably wind up mapping
852  * at most one segment. For the TX case, there could be any number of
853  * segments since TX packets may span multiple mbufs. In either case,
854  * if the number of segments is larger than the re_maxsegs limit
855  * specified by the caller, we abort the mapping operation. Sadly,
856  * whoever designed the buffer mapping API did not provide a way to
857  * return an error from here, so we have to fake it a bit.
858  */
859 
860 static void
861 re_dma_map_desc(void *arg, bus_dma_segment_t *segs, int nseg,
862 		bus_size_t mapsize, int error)
863 {
864 	struct re_dmaload_arg *ctx;
865 	struct re_desc *d = NULL;
866 	int i = 0, idx;
867 	uint32_t cmdstat;
868 
869 	if (error)
870 		return;
871 
872 	ctx = arg;
873 
874 	/* Signal error to caller if there's too many segments */
875 	if (nseg > ctx->re_maxsegs) {
876 		ctx->re_maxsegs = 0;
877 		return;
878 	}
879 
880 	/*
881 	 * Map the segment array into descriptors. Note that we set the
882 	 * start-of-frame and end-of-frame markers for either TX or RX, but
883 	 * they really only have meaning in the TX case. (In the RX case,
884 	 * it's the chip that tells us where packets begin and end.)
885 	 * We also keep track of the end of the ring and set the
886 	 * end-of-ring bits as needed, and we set the ownership bits
887 	 * in all except the very first descriptor. (The caller will
888 	 * set this descriptor later when it start transmission or
889 	 * reception.)
890 	 */
891 	idx = ctx->re_idx;
892 	for (;;) {
893 		d = &ctx->re_ring[idx];
894 		if (le32toh(d->re_cmdstat) & RE_RDESC_STAT_OWN) {
895 			ctx->re_maxsegs = 0;
896 			return;
897 		}
898 		cmdstat = segs[i].ds_len;
899 		d->re_bufaddr_lo = htole32(RE_ADDR_LO(segs[i].ds_addr));
900 		d->re_bufaddr_hi = htole32(RE_ADDR_HI(segs[i].ds_addr));
901 		if (i == 0)
902 			cmdstat |= RE_TDESC_CMD_SOF;
903 		else
904 			cmdstat |= RE_TDESC_CMD_OWN;
905 		if (idx == (RE_RX_DESC_CNT - 1))
906 			cmdstat |= RE_TDESC_CMD_EOR;
907 		d->re_cmdstat = htole32(cmdstat | ctx->re_flags);
908 		i++;
909 		if (i == nseg)
910 			break;
911 		RE_DESC_INC(idx);
912 	}
913 
914 	d->re_cmdstat |= htole32(RE_TDESC_CMD_EOF);
915 	ctx->re_maxsegs = nseg;
916 	ctx->re_idx = idx;
917 }
918 
919 /*
920  * Map a single buffer address.
921  */
922 
923 static void
924 re_dma_map_addr(void *arg, bus_dma_segment_t *segs, int nseg, int error)
925 {
926 	uint32_t *addr;
927 
928 	if (error)
929 		return;
930 
931 	KASSERT(nseg == 1, ("too many DMA segments, %d should be 1", nseg));
932 	addr = arg;
933 	*addr = segs->ds_addr;
934 }
935 
936 static int
937 re_allocmem(device_t dev, struct re_softc *sc)
938 {
939 	int error, i, nseg;
940 
941 	/*
942 	 * Allocate map for RX mbufs.
943 	 */
944 	nseg = 32;
945 	error = bus_dma_tag_create(sc->re_parent_tag, ETHER_ALIGN, 0,
946 	    BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
947 	    NULL, MCLBYTES * nseg, nseg, MCLBYTES, BUS_DMA_ALLOCNOW,
948 	    &sc->re_ldata.re_mtag);
949 	if (error) {
950 		device_printf(dev, "could not allocate dma tag\n");
951 		return(error);
952 	}
953 
954 	/*
955 	 * Allocate map for TX descriptor list.
956 	 */
957 	error = bus_dma_tag_create(sc->re_parent_tag, RE_RING_ALIGN,
958 	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
959             NULL, RE_TX_LIST_SZ, 1, RE_TX_LIST_SZ, BUS_DMA_ALLOCNOW,
960 	    &sc->re_ldata.re_tx_list_tag);
961 	if (error) {
962 		device_printf(dev, "could not allocate dma tag\n");
963 		return(error);
964 	}
965 
966 	/* Allocate DMA'able memory for the TX ring */
967 
968         error = bus_dmamem_alloc(sc->re_ldata.re_tx_list_tag,
969 	    (void **)&sc->re_ldata.re_tx_list, BUS_DMA_WAITOK | BUS_DMA_ZERO,
970             &sc->re_ldata.re_tx_list_map);
971         if (error) {
972 		device_printf(dev, "could not allocate TX ring\n");
973                 return(error);
974 	}
975 
976 	/* Load the map for the TX ring. */
977 
978 	error = bus_dmamap_load(sc->re_ldata.re_tx_list_tag,
979 	     sc->re_ldata.re_tx_list_map, sc->re_ldata.re_tx_list,
980 	     RE_TX_LIST_SZ, re_dma_map_addr,
981 	     &sc->re_ldata.re_tx_list_addr, BUS_DMA_NOWAIT);
982 	if (error) {
983 		device_printf(dev, "could not get addres of TX ring\n");
984 		return(error);
985 	}
986 
987 	/* Create DMA maps for TX buffers */
988 
989 	for (i = 0; i < RE_TX_DESC_CNT; i++) {
990 		error = bus_dmamap_create(sc->re_ldata.re_mtag, 0,
991 			    &sc->re_ldata.re_tx_dmamap[i]);
992 		if (error) {
993 			device_printf(dev, "can't create DMA map for TX\n");
994 			return(error);
995 		}
996 	}
997 
998 	/*
999 	 * Allocate map for RX descriptor list.
1000 	 */
1001 	error = bus_dma_tag_create(sc->re_parent_tag, RE_RING_ALIGN,
1002 	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1003             NULL, RE_RX_LIST_SZ, 1, RE_RX_LIST_SZ, BUS_DMA_ALLOCNOW,
1004 	    &sc->re_ldata.re_rx_list_tag);
1005 	if (error) {
1006 		device_printf(dev, "could not allocate dma tag\n");
1007 		return(error);
1008 	}
1009 
1010 	/* Allocate DMA'able memory for the RX ring */
1011 
1012         error = bus_dmamem_alloc(sc->re_ldata.re_rx_list_tag,
1013 	    (void **)&sc->re_ldata.re_rx_list, BUS_DMA_WAITOK | BUS_DMA_ZERO,
1014             &sc->re_ldata.re_rx_list_map);
1015         if (error) {
1016 		device_printf(dev, "could not allocate RX ring\n");
1017                 return(error);
1018 	}
1019 
1020 	/* Load the map for the RX ring. */
1021 
1022 	error = bus_dmamap_load(sc->re_ldata.re_rx_list_tag,
1023 	     sc->re_ldata.re_rx_list_map, sc->re_ldata.re_rx_list,
1024 	     RE_RX_LIST_SZ, re_dma_map_addr,
1025 	     &sc->re_ldata.re_rx_list_addr, BUS_DMA_NOWAIT);
1026 	if (error) {
1027 		device_printf(dev, "could not get address of RX ring\n");
1028 		return(error);
1029 	}
1030 
1031 	/* Create DMA maps for RX buffers */
1032 
1033 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
1034 		error = bus_dmamap_create(sc->re_ldata.re_mtag, 0,
1035 			    &sc->re_ldata.re_rx_dmamap[i]);
1036 		if (error) {
1037 			device_printf(dev, "can't create DMA map for RX\n");
1038 			return(ENOMEM);
1039 		}
1040 	}
1041 
1042 	return(0);
1043 }
1044 
1045 /*
1046  * Attach the interface. Allocate softc structures, do ifmedia
1047  * setup and ethernet/BPF attach.
1048  */
1049 static int
1050 re_attach(device_t dev)
1051 {
1052 	struct re_softc	*sc = device_get_softc(dev);
1053 	struct ifnet *ifp;
1054 	struct re_hwrev *hw_rev;
1055 	uint8_t eaddr[ETHER_ADDR_LEN];
1056 	uint16_t as[ETHER_ADDR_LEN / 2];
1057 	int hwrev;
1058 	uint16_t re_did = 0;
1059 	int error = 0, rid, i;
1060 
1061 	callout_init(&sc->re_timer);
1062 #ifdef RE_DIAG
1063 	sc->re_dev = dev;
1064 #endif
1065 
1066 	RE_ENABLE_TX_MODERATION(sc);
1067 
1068 	sysctl_ctx_init(&sc->re_sysctl_ctx);
1069 	sc->re_sysctl_tree = SYSCTL_ADD_NODE(&sc->re_sysctl_ctx,
1070 					     SYSCTL_STATIC_CHILDREN(_hw),
1071 					     OID_AUTO,
1072 					     device_get_nameunit(dev),
1073 					     CTLFLAG_RD, 0, "");
1074 	if (sc->re_sysctl_tree == NULL) {
1075 		device_printf(dev, "can't add sysctl node\n");
1076 		error = ENXIO;
1077 		goto fail;
1078 	}
1079 	SYSCTL_ADD_PROC(&sc->re_sysctl_ctx,
1080 			SYSCTL_CHILDREN(sc->re_sysctl_tree),
1081 			OID_AUTO, "tx_moderation",
1082 			CTLTYPE_INT | CTLFLAG_RW,
1083 			sc, 0, re_sysctl_tx_moderation, "I",
1084 			"Enable/Disable TX moderation");
1085 
1086 #ifndef BURN_BRIDGES
1087 	/*
1088 	 * Handle power management nonsense.
1089 	 */
1090 
1091 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D0) {
1092 		uint32_t membase, irq;
1093 
1094 		/* Save important PCI config data. */
1095 		membase = pci_read_config(dev, RE_PCI_LOMEM, 4);
1096 		irq = pci_read_config(dev, PCIR_INTLINE, 4);
1097 
1098 		/* Reset the power state. */
1099 		device_printf(dev, "chip is is in D%d power mode "
1100 		    "-- setting to D0\n", pci_get_powerstate(dev));
1101 
1102 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
1103 
1104 		/* Restore PCI config data. */
1105 		pci_write_config(dev, RE_PCI_LOMEM, membase, 4);
1106 		pci_write_config(dev, PCIR_INTLINE, irq, 4);
1107 	}
1108 #endif
1109 	/*
1110 	 * Map control/status registers.
1111 	 */
1112 	pci_enable_busmaster(dev);
1113 
1114 	rid = RE_PCI_LOIO;
1115 	sc->re_res = bus_alloc_resource_any(dev, SYS_RES_IOPORT, &rid,
1116 					    RF_ACTIVE);
1117 
1118 	if (sc->re_res == NULL) {
1119 		device_printf(dev, "couldn't map ports\n");
1120 		error = ENXIO;
1121 		goto fail;
1122 	}
1123 
1124 	sc->re_btag = rman_get_bustag(sc->re_res);
1125 	sc->re_bhandle = rman_get_bushandle(sc->re_res);
1126 
1127 	/* Allocate interrupt */
1128 	rid = 0;
1129 	sc->re_irq = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
1130 					    RF_SHAREABLE | RF_ACTIVE);
1131 
1132 	if (sc->re_irq == NULL) {
1133 		device_printf(dev, "couldn't map interrupt\n");
1134 		error = ENXIO;
1135 		goto fail;
1136 	}
1137 
1138 	/* Reset the adapter. */
1139 	re_reset(sc);
1140 
1141 	hwrev = CSR_READ_4(sc, RE_TXCFG) & RE_TXCFG_HWREV;
1142 	for (hw_rev = re_hwrevs; hw_rev->re_desc != NULL; hw_rev++) {
1143 		if (hw_rev->re_rev == hwrev) {
1144 			sc->re_type = hw_rev->re_type;
1145 			break;
1146 		}
1147 	}
1148 
1149 	sc->re_eewidth = 6;
1150 	re_read_eeprom(sc, (caddr_t)&re_did, 0, 1);
1151 	if (re_did != 0x8129)
1152 	        sc->re_eewidth = 8;
1153 
1154 	/*
1155 	 * Get station address from the EEPROM.
1156 	 */
1157 	re_read_eeprom(sc, (caddr_t)as, RE_EE_EADDR, 3);
1158 	for (i = 0; i < ETHER_ADDR_LEN / 2; i++)
1159 		as[i] = le16toh(as[i]);
1160 	bcopy(as, eaddr, sizeof(eaddr));
1161 
1162 	if (sc->re_type == RE_8169) {
1163 		/* Set RX length mask */
1164 		sc->re_rxlenmask = RE_RDESC_STAT_GFRAGLEN;
1165 		sc->re_txstart = RE_GTXSTART;
1166 	} else {
1167 		/* Set RX length mask */
1168 		sc->re_rxlenmask = RE_RDESC_STAT_FRAGLEN;
1169 		sc->re_txstart = RE_TXSTART;
1170 	}
1171 
1172 	/*
1173 	 * Allocate the parent bus DMA tag appropriate for PCI.
1174 	 */
1175 #define RE_NSEG_NEW 32
1176 	error = bus_dma_tag_create(NULL,	/* parent */
1177 			1, 0,			/* alignment, boundary */
1178 			BUS_SPACE_MAXADDR_32BIT,/* lowaddr */
1179 			BUS_SPACE_MAXADDR,	/* highaddr */
1180 			NULL, NULL,		/* filter, filterarg */
1181 			MAXBSIZE, RE_NSEG_NEW,	/* maxsize, nsegments */
1182 			BUS_SPACE_MAXSIZE_32BIT,/* maxsegsize */
1183 			BUS_DMA_ALLOCNOW,	/* flags */
1184 			&sc->re_parent_tag);
1185 	if (error)
1186 		goto fail;
1187 
1188 	error = re_allocmem(dev, sc);
1189 
1190 	if (error)
1191 		goto fail;
1192 
1193 	/* Do MII setup */
1194 	if (mii_phy_probe(dev, &sc->re_miibus,
1195 	    re_ifmedia_upd, re_ifmedia_sts)) {
1196 		device_printf(dev, "MII without any phy!\n");
1197 		error = ENXIO;
1198 		goto fail;
1199 	}
1200 
1201 	ifp = &sc->arpcom.ac_if;
1202 	ifp->if_softc = sc;
1203 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1204 	ifp->if_mtu = ETHERMTU;
1205 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1206 	ifp->if_ioctl = re_ioctl;
1207 	ifp->if_capabilities = IFCAP_VLAN_MTU;
1208 	ifp->if_start = re_start;
1209 	ifp->if_capabilities |= IFCAP_HWCSUM|IFCAP_VLAN_HWTAGGING;
1210 #ifdef DEVICE_POLLING
1211 	ifp->if_poll = re_poll;
1212 #endif
1213 	ifp->if_watchdog = re_watchdog;
1214 	ifp->if_init = re_init;
1215 	if (sc->re_type == RE_8169)
1216 		ifp->if_baudrate = 1000000000;
1217 	else
1218 		ifp->if_baudrate = 100000000;
1219 	ifq_set_maxlen(&ifp->if_snd, RE_IFQ_MAXLEN);
1220 	ifq_set_ready(&ifp->if_snd);
1221 #ifdef RE_DISABLE_HWCSUM
1222 	ifp->if_capenable = ifp->if_capabilities & ~IFCAP_HWCSUM;
1223 	ifp->if_hwassist = 0;
1224 #else
1225 	ifp->if_capenable = ifp->if_capabilities;
1226 	ifp->if_hwassist = RE_CSUM_FEATURES;
1227 #endif
1228 
1229 	/*
1230 	 * Call MI attach routine.
1231 	 */
1232 	ether_ifattach(ifp, eaddr, NULL);
1233 
1234 #ifdef RE_DIAG
1235 	/*
1236 	 * Perform hardware diagnostic on the original RTL8169.
1237 	 * Some 32-bit cards were incorrectly wired and would
1238 	 * malfunction if plugged into a 64-bit slot.
1239 	 */
1240 	if (hwrev == RE_HWREV_8169) {
1241 		lwkt_serialize_enter(ifp->if_serializer);
1242 		error = re_diag(sc);
1243 		lwkt_serialize_exit(ifp->if_serializer);
1244 
1245 		if (error) {
1246 			device_printf(dev, "hardware diagnostic failure\n");
1247 			ether_ifdetach(ifp);
1248 			goto fail;
1249 		}
1250 	}
1251 #endif	/* RE_DIAG */
1252 
1253 	/* Hook interrupt last to avoid having to lock softc */
1254 	error = bus_setup_intr(dev, sc->re_irq, INTR_NETSAFE, re_intr, sc,
1255 			       &sc->re_intrhand, ifp->if_serializer);
1256 
1257 	if (error) {
1258 		device_printf(dev, "couldn't set up irq\n");
1259 		ether_ifdetach(ifp);
1260 		goto fail;
1261 	}
1262 
1263 fail:
1264 	if (error)
1265 		re_detach(dev);
1266 
1267 	return (error);
1268 }
1269 
1270 /*
1271  * Shutdown hardware and free up resources. This can be called any
1272  * time after the mutex has been initialized. It is called in both
1273  * the error case in attach and the normal detach case so it needs
1274  * to be careful about only freeing resources that have actually been
1275  * allocated.
1276  */
1277 static int
1278 re_detach(device_t dev)
1279 {
1280 	struct re_softc *sc = device_get_softc(dev);
1281 	struct ifnet *ifp = &sc->arpcom.ac_if;
1282 	int i;
1283 
1284 	/* These should only be active if attach succeeded */
1285 	if (device_is_attached(dev)) {
1286 		lwkt_serialize_enter(ifp->if_serializer);
1287 		re_stop(sc);
1288 		bus_teardown_intr(dev, sc->re_irq, sc->re_intrhand);
1289 		lwkt_serialize_exit(ifp->if_serializer);
1290 
1291 		ether_ifdetach(ifp);
1292 	}
1293 	if (sc->re_miibus)
1294 		device_delete_child(dev, sc->re_miibus);
1295 	bus_generic_detach(dev);
1296 
1297 	if (sc->re_irq)
1298 		bus_release_resource(dev, SYS_RES_IRQ, 0, sc->re_irq);
1299 	if (sc->re_res) {
1300 		bus_release_resource(dev, SYS_RES_IOPORT, RE_PCI_LOIO,
1301 				     sc->re_res);
1302 	}
1303 
1304 	/* Unload and free the RX DMA ring memory and map */
1305 
1306 	if (sc->re_ldata.re_rx_list_tag) {
1307 		bus_dmamap_unload(sc->re_ldata.re_rx_list_tag,
1308 		    sc->re_ldata.re_rx_list_map);
1309 		bus_dmamem_free(sc->re_ldata.re_rx_list_tag,
1310 		    sc->re_ldata.re_rx_list,
1311 		    sc->re_ldata.re_rx_list_map);
1312 		bus_dma_tag_destroy(sc->re_ldata.re_rx_list_tag);
1313 	}
1314 
1315 	/* Unload and free the TX DMA ring memory and map */
1316 
1317 	if (sc->re_ldata.re_tx_list_tag) {
1318 		bus_dmamap_unload(sc->re_ldata.re_tx_list_tag,
1319 		    sc->re_ldata.re_tx_list_map);
1320 		bus_dmamem_free(sc->re_ldata.re_tx_list_tag,
1321 		    sc->re_ldata.re_tx_list,
1322 		    sc->re_ldata.re_tx_list_map);
1323 		bus_dma_tag_destroy(sc->re_ldata.re_tx_list_tag);
1324 	}
1325 
1326 	/* Destroy all the RX and TX buffer maps */
1327 
1328 	if (sc->re_ldata.re_mtag) {
1329 		for (i = 0; i < RE_TX_DESC_CNT; i++)
1330 			bus_dmamap_destroy(sc->re_ldata.re_mtag,
1331 			    sc->re_ldata.re_tx_dmamap[i]);
1332 		for (i = 0; i < RE_RX_DESC_CNT; i++)
1333 			bus_dmamap_destroy(sc->re_ldata.re_mtag,
1334 			    sc->re_ldata.re_rx_dmamap[i]);
1335 		bus_dma_tag_destroy(sc->re_ldata.re_mtag);
1336 	}
1337 
1338 	/* Unload and free the stats buffer and map */
1339 
1340 	if (sc->re_ldata.re_stag) {
1341 		bus_dmamap_unload(sc->re_ldata.re_stag,
1342 		    sc->re_ldata.re_rx_list_map);
1343 		bus_dmamem_free(sc->re_ldata.re_stag,
1344 		    sc->re_ldata.re_stats,
1345 		    sc->re_ldata.re_smap);
1346 		bus_dma_tag_destroy(sc->re_ldata.re_stag);
1347 	}
1348 
1349 	if (sc->re_parent_tag)
1350 		bus_dma_tag_destroy(sc->re_parent_tag);
1351 
1352 	return(0);
1353 }
1354 
1355 static int
1356 re_newbuf(struct re_softc *sc, int idx, struct mbuf *m)
1357 {
1358 	struct re_dmaload_arg arg;
1359 	struct mbuf *n = NULL;
1360 	int error;
1361 
1362 	if (m == NULL) {
1363 		n = m_getcl(MB_DONTWAIT, MT_DATA, M_PKTHDR);
1364 		if (n == NULL)
1365 			return(ENOBUFS);
1366 		m = n;
1367 	} else
1368 		m->m_data = m->m_ext.ext_buf;
1369 
1370 	/*
1371 	 * Initialize mbuf length fields and fixup
1372 	 * alignment so that the frame payload is
1373 	 * longword aligned.
1374 	 */
1375 	m->m_len = m->m_pkthdr.len = MCLBYTES;
1376 	m_adj(m, ETHER_ALIGN);
1377 
1378 	arg.sc = sc;
1379 	arg.re_idx = idx;
1380 	arg.re_maxsegs = 1;
1381 	arg.re_flags = 0;
1382 	arg.re_ring = sc->re_ldata.re_rx_list;
1383 
1384         error = bus_dmamap_load_mbuf(sc->re_ldata.re_mtag,
1385 	    sc->re_ldata.re_rx_dmamap[idx], m, re_dma_map_desc,
1386 	    &arg, BUS_DMA_NOWAIT);
1387 	if (error || arg.re_maxsegs != 1) {
1388 		if (n != NULL)
1389 			m_freem(n);
1390 		return (ENOMEM);
1391 	}
1392 
1393 	sc->re_ldata.re_rx_list[idx].re_cmdstat |= htole32(RE_RDESC_CMD_OWN);
1394 	sc->re_ldata.re_rx_mbuf[idx] = m;
1395 
1396         bus_dmamap_sync(sc->re_ldata.re_mtag, sc->re_ldata.re_rx_dmamap[idx],
1397 		        BUS_DMASYNC_PREREAD);
1398 
1399 	return(0);
1400 }
1401 
1402 static int
1403 re_tx_list_init(struct re_softc *sc)
1404 {
1405 	bzero(sc->re_ldata.re_tx_list, RE_TX_LIST_SZ);
1406 	bzero(&sc->re_ldata.re_tx_mbuf, RE_TX_DESC_CNT * sizeof(struct mbuf *));
1407 
1408 	bus_dmamap_sync(sc->re_ldata.re_tx_list_tag,
1409 			sc->re_ldata.re_tx_list_map, BUS_DMASYNC_PREWRITE);
1410 	sc->re_ldata.re_tx_prodidx = 0;
1411 	sc->re_ldata.re_tx_considx = 0;
1412 	sc->re_ldata.re_tx_free = RE_TX_DESC_CNT;
1413 
1414 	return(0);
1415 }
1416 
1417 static int
1418 re_rx_list_init(struct re_softc *sc)
1419 {
1420 	int i, error;
1421 
1422 	bzero(sc->re_ldata.re_rx_list, RE_RX_LIST_SZ);
1423 	bzero(&sc->re_ldata.re_rx_mbuf, RE_RX_DESC_CNT * sizeof(struct mbuf *));
1424 
1425 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
1426 		error = re_newbuf(sc, i, NULL);
1427 		if (error)
1428 			return(error);
1429 	}
1430 
1431 	/* Flush the RX descriptors */
1432 
1433 	bus_dmamap_sync(sc->re_ldata.re_rx_list_tag,
1434 			sc->re_ldata.re_rx_list_map, BUS_DMASYNC_PREWRITE);
1435 
1436 	sc->re_ldata.re_rx_prodidx = 0;
1437 	sc->re_head = sc->re_tail = NULL;
1438 
1439 	return(0);
1440 }
1441 
1442 /*
1443  * RX handler for C+ and 8169. For the gigE chips, we support
1444  * the reception of jumbo frames that have been fragmented
1445  * across multiple 2K mbuf cluster buffers.
1446  */
1447 static void
1448 re_rxeof(struct re_softc *sc)
1449 {
1450 	struct ifnet *ifp = &sc->arpcom.ac_if;
1451 	struct mbuf *m;
1452 	struct re_desc 	*cur_rx;
1453 	uint32_t rxstat, rxvlan;
1454 	int i, total_len;
1455 
1456 	/* Invalidate the descriptor memory */
1457 
1458 	bus_dmamap_sync(sc->re_ldata.re_rx_list_tag,
1459 			sc->re_ldata.re_rx_list_map, BUS_DMASYNC_POSTREAD);
1460 
1461 	for (i = sc->re_ldata.re_rx_prodidx;
1462 	     RE_OWN(&sc->re_ldata.re_rx_list[i]) == 0 ; RE_DESC_INC(i)) {
1463 		cur_rx = &sc->re_ldata.re_rx_list[i];
1464 		m = sc->re_ldata.re_rx_mbuf[i];
1465 		total_len = RE_RXBYTES(cur_rx);
1466 		rxstat = le32toh(cur_rx->re_cmdstat);
1467 		rxvlan = le32toh(cur_rx->re_vlanctl);
1468 
1469 		/* Invalidate the RX mbuf and unload its map */
1470 
1471 		bus_dmamap_sync(sc->re_ldata.re_mtag,
1472 				sc->re_ldata.re_rx_dmamap[i],
1473 				BUS_DMASYNC_POSTWRITE);
1474 		bus_dmamap_unload(sc->re_ldata.re_mtag,
1475 				  sc->re_ldata.re_rx_dmamap[i]);
1476 
1477 		if ((rxstat & RE_RDESC_STAT_EOF) == 0) {
1478 			m->m_len = MCLBYTES - ETHER_ALIGN;
1479 			if (sc->re_head == NULL) {
1480 				sc->re_head = sc->re_tail = m;
1481 			} else {
1482 				sc->re_tail->m_next = m;
1483 				sc->re_tail = m;
1484 			}
1485 			re_newbuf(sc, i, NULL);
1486 			continue;
1487 		}
1488 
1489 		/*
1490 		 * NOTE: for the 8139C+, the frame length field
1491 		 * is always 12 bits in size, but for the gigE chips,
1492 		 * it is 13 bits (since the max RX frame length is 16K).
1493 		 * Unfortunately, all 32 bits in the status word
1494 		 * were already used, so to make room for the extra
1495 		 * length bit, RealTek took out the 'frame alignment
1496 		 * error' bit and shifted the other status bits
1497 		 * over one slot. The OWN, EOR, FS and LS bits are
1498 		 * still in the same places. We have already extracted
1499 		 * the frame length and checked the OWN bit, so rather
1500 		 * than using an alternate bit mapping, we shift the
1501 		 * status bits one space to the right so we can evaluate
1502 		 * them using the 8169 status as though it was in the
1503 		 * same format as that of the 8139C+.
1504 		 */
1505 		if (sc->re_type == RE_8169)
1506 			rxstat >>= 1;
1507 
1508 		if (rxstat & RE_RDESC_STAT_RXERRSUM) {
1509 			ifp->if_ierrors++;
1510 			/*
1511 			 * If this is part of a multi-fragment packet,
1512 			 * discard all the pieces.
1513 			 */
1514 			if (sc->re_head != NULL) {
1515 				m_freem(sc->re_head);
1516 				sc->re_head = sc->re_tail = NULL;
1517 			}
1518 			re_newbuf(sc, i, m);
1519 			continue;
1520 		}
1521 
1522 		/*
1523 		 * If allocating a replacement mbuf fails,
1524 		 * reload the current one.
1525 		 */
1526 
1527 		if (re_newbuf(sc, i, NULL)) {
1528 			ifp->if_ierrors++;
1529 			if (sc->re_head != NULL) {
1530 				m_freem(sc->re_head);
1531 				sc->re_head = sc->re_tail = NULL;
1532 			}
1533 			re_newbuf(sc, i, m);
1534 			continue;
1535 		}
1536 
1537 		if (sc->re_head != NULL) {
1538 			m->m_len = total_len % (MCLBYTES - ETHER_ALIGN);
1539 			/*
1540 			 * Special case: if there's 4 bytes or less
1541 			 * in this buffer, the mbuf can be discarded:
1542 			 * the last 4 bytes is the CRC, which we don't
1543 			 * care about anyway.
1544 			 */
1545 			if (m->m_len <= ETHER_CRC_LEN) {
1546 				sc->re_tail->m_len -=
1547 				    (ETHER_CRC_LEN - m->m_len);
1548 				m_freem(m);
1549 			} else {
1550 				m->m_len -= ETHER_CRC_LEN;
1551 				sc->re_tail->m_next = m;
1552 			}
1553 			m = sc->re_head;
1554 			sc->re_head = sc->re_tail = NULL;
1555 			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
1556 		} else
1557 			m->m_pkthdr.len = m->m_len =
1558 			    (total_len - ETHER_CRC_LEN);
1559 
1560 		ifp->if_ipackets++;
1561 		m->m_pkthdr.rcvif = ifp;
1562 
1563 		/* Do RX checksumming if enabled */
1564 
1565 		if (ifp->if_capenable & IFCAP_RXCSUM) {
1566 
1567 			/* Check IP header checksum */
1568 			if (rxstat & RE_RDESC_STAT_PROTOID)
1569 				m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED;
1570 			if ((rxstat & RE_RDESC_STAT_IPSUMBAD) == 0)
1571 				m->m_pkthdr.csum_flags |= CSUM_IP_VALID;
1572 
1573 			/* Check TCP/UDP checksum */
1574 			if ((RE_TCPPKT(rxstat) &&
1575 			    (rxstat & RE_RDESC_STAT_TCPSUMBAD) == 0) ||
1576 			    (RE_UDPPKT(rxstat) &&
1577 			    (rxstat & RE_RDESC_STAT_UDPSUMBAD)) == 0) {
1578 				m->m_pkthdr.csum_flags |=
1579 				    CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
1580 				m->m_pkthdr.csum_data = 0xffff;
1581 			}
1582 		}
1583 
1584 		if (rxvlan & RE_RDESC_VLANCTL_TAG) {
1585 			VLAN_INPUT_TAG(m,
1586 			   be16toh((rxvlan & RE_RDESC_VLANCTL_DATA)));
1587 		} else {
1588 			ifp->if_input(ifp, m);
1589 		}
1590 	}
1591 
1592 	/* Flush the RX DMA ring */
1593 
1594 	bus_dmamap_sync(sc->re_ldata.re_rx_list_tag,
1595 			sc->re_ldata.re_rx_list_map, BUS_DMASYNC_PREWRITE);
1596 
1597 	sc->re_ldata.re_rx_prodidx = i;
1598 }
1599 
1600 static void
1601 re_txeof(struct re_softc *sc)
1602 {
1603 	struct ifnet *ifp = &sc->arpcom.ac_if;
1604 	uint32_t txstat;
1605 	int idx;
1606 
1607 	/* Invalidate the TX descriptor list */
1608 
1609 	bus_dmamap_sync(sc->re_ldata.re_tx_list_tag,
1610 			sc->re_ldata.re_tx_list_map, BUS_DMASYNC_POSTREAD);
1611 
1612 	for (idx = sc->re_ldata.re_tx_considx;
1613 	     sc->re_ldata.re_tx_free < RE_TX_DESC_CNT; RE_DESC_INC(idx)) {
1614 		txstat = le32toh(sc->re_ldata.re_tx_list[idx].re_cmdstat);
1615 		if (txstat & RE_TDESC_CMD_OWN)
1616 			break;
1617 
1618 		sc->re_ldata.re_tx_list[idx].re_bufaddr_lo = 0;
1619 
1620 		/*
1621 		 * We only stash mbufs in the last descriptor
1622 		 * in a fragment chain, which also happens to
1623 		 * be the only place where the TX status bits
1624 		 * are valid.
1625 		 */
1626 		if (txstat & RE_TDESC_CMD_EOF) {
1627 			m_freem(sc->re_ldata.re_tx_mbuf[idx]);
1628 			sc->re_ldata.re_tx_mbuf[idx] = NULL;
1629 			bus_dmamap_unload(sc->re_ldata.re_mtag,
1630 			    sc->re_ldata.re_tx_dmamap[idx]);
1631 			if (txstat & (RE_TDESC_STAT_EXCESSCOL|
1632 			    RE_TDESC_STAT_COLCNT))
1633 				ifp->if_collisions++;
1634 			if (txstat & RE_TDESC_STAT_TXERRSUM)
1635 				ifp->if_oerrors++;
1636 			else
1637 				ifp->if_opackets++;
1638 		}
1639 		sc->re_ldata.re_tx_free++;
1640 	}
1641 
1642 	/* No changes made to the TX ring, so no flush needed */
1643 	if (sc->re_ldata.re_tx_free) {
1644 		sc->re_ldata.re_tx_considx = idx;
1645 		ifp->if_flags &= ~IFF_OACTIVE;
1646 		ifp->if_timer = 0;
1647 	}
1648 
1649 	/*
1650 	 * Some chips will ignore a second TX request issued while an
1651 	 * existing transmission is in progress. If the transmitter goes
1652 	 * idle but there are still packets waiting to be sent, we need
1653 	 * to restart the channel here to flush them out. This only seems
1654 	 * to be required with the PCIe devices.
1655 	 */
1656 	if (sc->re_ldata.re_tx_free < RE_TX_DESC_CNT)
1657 		CSR_WRITE_1(sc, sc->re_txstart, RE_TXSTART_START);
1658 
1659 	/*
1660 	 * If not all descriptors have been released reaped yet,
1661 	 * reload the timer so that we will eventually get another
1662 	 * interrupt that will cause us to re-enter this routine.
1663 	 * This is done in case the transmitter has gone idle.
1664 	 */
1665 	if (RE_TX_MODERATION_IS_ENABLED(sc) &&
1666 	    sc->re_ldata.re_tx_free < RE_TX_DESC_CNT)
1667                 CSR_WRITE_4(sc, RE_TIMERCNT, 1);
1668 }
1669 
1670 static void
1671 re_tick(void *xsc)
1672 {
1673 	struct re_softc *sc = xsc;
1674 
1675 	lwkt_serialize_enter(sc->arpcom.ac_if.if_serializer);
1676 	re_tick_serialized(xsc);
1677 	lwkt_serialize_exit(sc->arpcom.ac_if.if_serializer);
1678 }
1679 
1680 static void
1681 re_tick_serialized(void *xsc)
1682 {
1683 	struct re_softc *sc = xsc;
1684 	struct ifnet *ifp = &sc->arpcom.ac_if;
1685 	struct mii_data *mii;
1686 
1687 	mii = device_get_softc(sc->re_miibus);
1688 	mii_tick(mii);
1689 	if (sc->re_link) {
1690 		if (!(mii->mii_media_status & IFM_ACTIVE))
1691 			sc->re_link = 0;
1692 	} else {
1693 		if (mii->mii_media_status & IFM_ACTIVE &&
1694 		    IFM_SUBTYPE(mii->mii_media_active) != IFM_NONE) {
1695 			sc->re_link = 1;
1696 			if (!ifq_is_empty(&ifp->if_snd))
1697 				ifp->if_start(ifp);
1698 		}
1699 	}
1700 
1701 	callout_reset(&sc->re_timer, hz, re_tick, sc);
1702 }
1703 
1704 #ifdef DEVICE_POLLING
1705 
1706 static void
1707 re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
1708 {
1709 	struct re_softc *sc = ifp->if_softc;
1710 
1711 	switch(cmd) {
1712 	case POLL_REGISTER:
1713 		/* disable interrupts */
1714 		CSR_WRITE_2(sc, RE_IMR, 0x0000);
1715 		break;
1716 	case POLL_DEREGISTER:
1717 		/* enable interrupts */
1718 		CSR_WRITE_2(sc, RE_IMR, sc->re_intrs);
1719 		break;
1720 	default:
1721 		sc->rxcycles = count;
1722 		re_rxeof(sc);
1723 		re_txeof(sc);
1724 
1725 		if (!ifq_is_empty(&ifp->if_snd))
1726 			(*ifp->if_start)(ifp);
1727 
1728 		if (cmd == POLL_AND_CHECK_STATUS) { /* also check status register */
1729 			uint16_t       status;
1730 
1731 			status = CSR_READ_2(sc, RE_ISR);
1732 			if (status == 0xffff)
1733 				return;
1734 			if (status)
1735 				CSR_WRITE_2(sc, RE_ISR, status);
1736 
1737 			/*
1738 			 * XXX check behaviour on receiver stalls.
1739 			 */
1740 
1741 			if (status & RE_ISR_SYSTEM_ERR) {
1742 				re_reset(sc);
1743 				re_init(sc);
1744 			}
1745 		}
1746 		break;
1747 	}
1748 }
1749 #endif /* DEVICE_POLLING */
1750 
1751 static void
1752 re_intr(void *arg)
1753 {
1754 	struct re_softc	*sc = arg;
1755 	struct ifnet *ifp = &sc->arpcom.ac_if;
1756 	uint16_t status;
1757 
1758 	if (sc->suspended || (ifp->if_flags & IFF_UP) == 0)
1759 		return;
1760 
1761 	for (;;) {
1762 		status = CSR_READ_2(sc, RE_ISR);
1763 		/* If the card has gone away the read returns 0xffff. */
1764 		if (status == 0xffff)
1765 			break;
1766 		if (status)
1767 			CSR_WRITE_2(sc, RE_ISR, status);
1768 
1769 		if ((status & sc->re_intrs) == 0)
1770 			break;
1771 
1772 		if (status & (RE_ISR_RX_OK | RE_ISR_RX_ERR | RE_ISR_FIFO_OFLOW))
1773 			re_rxeof(sc);
1774 
1775 		if ((status & sc->re_tx_ack) ||
1776 		    (status & RE_ISR_TX_ERR) ||
1777 		    (status & RE_ISR_TX_DESC_UNAVAIL))
1778 			re_txeof(sc);
1779 
1780 		if (status & RE_ISR_SYSTEM_ERR) {
1781 			re_reset(sc);
1782 			re_init(sc);
1783 		}
1784 
1785 		if (status & RE_ISR_LINKCHG) {
1786 			callout_stop(&sc->re_timer);
1787 			re_tick_serialized(sc);
1788 		}
1789 	}
1790 
1791 	if (!ifq_is_empty(&ifp->if_snd))
1792 		(*ifp->if_start)(ifp);
1793 }
1794 
1795 static int
1796 re_encap(struct re_softc *sc, struct mbuf **m_head, int *idx, int *called_defrag)
1797 {
1798 	struct ifnet *ifp = &sc->arpcom.ac_if;
1799 	struct mbuf *m, *m_new = NULL;
1800 	struct re_dmaload_arg	arg;
1801 	bus_dmamap_t		map;
1802 	int			error;
1803 
1804 	*called_defrag = 0;
1805 	if (sc->re_ldata.re_tx_free <= 4)
1806 		return(EFBIG);
1807 
1808 	m = *m_head;
1809 
1810 	/*
1811 	 * Set up checksum offload. Note: checksum offload bits must
1812 	 * appear in all descriptors of a multi-descriptor transmit
1813 	 * attempt. (This is according to testing done with an 8169
1814 	 * chip. I'm not sure if this is a requirement or a bug.)
1815 	 */
1816 
1817 	arg.re_flags = 0;
1818 
1819 	if (m->m_pkthdr.csum_flags & CSUM_IP)
1820 		arg.re_flags |= RE_TDESC_CMD_IPCSUM;
1821 	if (m->m_pkthdr.csum_flags & CSUM_TCP)
1822 		arg.re_flags |= RE_TDESC_CMD_TCPCSUM;
1823 	if (m->m_pkthdr.csum_flags & CSUM_UDP)
1824 		arg.re_flags |= RE_TDESC_CMD_UDPCSUM;
1825 
1826 	arg.sc = sc;
1827 	arg.re_idx = *idx;
1828 	arg.re_maxsegs = sc->re_ldata.re_tx_free;
1829 	if (arg.re_maxsegs > 4)
1830 		arg.re_maxsegs -= 4;
1831 	arg.re_ring = sc->re_ldata.re_tx_list;
1832 
1833 	map = sc->re_ldata.re_tx_dmamap[*idx];
1834 
1835 	/*
1836 	 * With some of the RealTek chips, using the checksum offload
1837 	 * support in conjunction with the autopadding feature results
1838 	 * in the transmission of corrupt frames. For example, if we
1839 	 * need to send a really small IP fragment that's less than 60
1840 	 * bytes in size, and IP header checksumming is enabled, the
1841 	 * resulting ethernet frame that appears on the wire will
1842 	 * have garbled payload. To work around this, if TX checksum
1843 	 * offload is enabled, we always manually pad short frames out
1844 	 * to the minimum ethernet frame size. We do this by pretending
1845 	 * the mbuf chain has too many fragments so the coalescing code
1846 	 * below can assemble the packet into a single buffer that's
1847 	 * padded out to the mininum frame size.
1848 	 */
1849 	if (arg.re_flags && m->m_pkthdr.len < RE_MIN_FRAMELEN) {
1850 		error = EFBIG;
1851 	} else {
1852 		error = bus_dmamap_load_mbuf(sc->re_ldata.re_mtag, map,
1853 		    m, re_dma_map_desc, &arg, BUS_DMA_NOWAIT);
1854 	}
1855 
1856 	if (error && error != EFBIG) {
1857 		if_printf(ifp, "can't map mbuf (error %d)\n", error);
1858 		return(ENOBUFS);
1859 	}
1860 
1861 	/* Too many segments to map, coalesce into a single mbuf */
1862 
1863 	if (error || arg.re_maxsegs == 0) {
1864 		m_new = m_defrag_nofree(m, MB_DONTWAIT);
1865 		if (m_new == NULL)
1866 			return(1);
1867 		else {
1868 			m = m_new;
1869 			*m_head = m;
1870 		}
1871 
1872 		/*
1873 		 * Manually pad short frames, and zero the pad space
1874 		 * to avoid leaking data.
1875 		 */
1876 		if (m_new->m_pkthdr.len < RE_MIN_FRAMELEN) {
1877 			bzero(mtod(m_new, char *) + m_new->m_pkthdr.len,
1878 			    RE_MIN_FRAMELEN - m_new->m_pkthdr.len);
1879 			m_new->m_pkthdr.len += RE_MIN_FRAMELEN -
1880 			    m_new->m_pkthdr.len;
1881 			m_new->m_len = m_new->m_pkthdr.len;
1882 		}
1883 
1884 		*called_defrag = 1;
1885 		arg.sc = sc;
1886 		arg.re_idx = *idx;
1887 		arg.re_maxsegs = sc->re_ldata.re_tx_free;
1888 		arg.re_ring = sc->re_ldata.re_tx_list;
1889 
1890 		error = bus_dmamap_load_mbuf(sc->re_ldata.re_mtag, map,
1891 		    m, re_dma_map_desc, &arg, BUS_DMA_NOWAIT);
1892 		if (error) {
1893 			m_freem(m);
1894 			if_printf(ifp, "can't map mbuf (error %d)\n", error);
1895 			return(EFBIG);
1896 		}
1897 	}
1898 
1899 	/*
1900 	 * Insure that the map for this transmission
1901 	 * is placed at the array index of the last descriptor
1902 	 * in this chain.
1903 	 */
1904 	sc->re_ldata.re_tx_dmamap[*idx] =
1905 	    sc->re_ldata.re_tx_dmamap[arg.re_idx];
1906 	sc->re_ldata.re_tx_dmamap[arg.re_idx] = map;
1907 
1908 	sc->re_ldata.re_tx_mbuf[arg.re_idx] = m;
1909 	sc->re_ldata.re_tx_free -= arg.re_maxsegs;
1910 
1911 	/*
1912 	 * Set up hardware VLAN tagging. Note: vlan tag info must
1913 	 * appear in the first descriptor of a multi-descriptor
1914 	 * transmission attempt.
1915 	 */
1916 
1917 	if ((m->m_flags & (M_PROTO1|M_PKTHDR)) == (M_PROTO1|M_PKTHDR) &&
1918 	    m->m_pkthdr.rcvif != NULL &&
1919 	    m->m_pkthdr.rcvif->if_type == IFT_L2VLAN) {
1920 	    	struct ifvlan *ifv;
1921 		ifv = m->m_pkthdr.rcvif->if_softc;
1922 		if (ifv != NULL)
1923 			sc->re_ldata.re_tx_list[*idx].re_vlanctl =
1924 			    htole32(htobe16(ifv->ifv_tag) | RE_TDESC_VLANCTL_TAG);
1925 	}
1926 
1927 	/* Transfer ownership of packet to the chip. */
1928 
1929 	sc->re_ldata.re_tx_list[arg.re_idx].re_cmdstat |=
1930 	    htole32(RE_TDESC_CMD_OWN);
1931 	if (*idx != arg.re_idx)
1932 		sc->re_ldata.re_tx_list[*idx].re_cmdstat |=
1933 		    htole32(RE_TDESC_CMD_OWN);
1934 
1935 	RE_DESC_INC(arg.re_idx);
1936 	*idx = arg.re_idx;
1937 
1938 	return(0);
1939 }
1940 
1941 /*
1942  * Main transmit routine for C+ and gigE NICs.
1943  */
1944 
1945 static void
1946 re_start(struct ifnet *ifp)
1947 {
1948 	struct re_softc	*sc = ifp->if_softc;
1949 	struct mbuf *m_head;
1950 	struct mbuf *m_head2;
1951 	int called_defrag, idx, need_trans;
1952 
1953 	if (!sc->re_link || (ifp->if_flags & IFF_OACTIVE))
1954 		return;
1955 
1956 	idx = sc->re_ldata.re_tx_prodidx;
1957 
1958 	need_trans = 0;
1959 	while (sc->re_ldata.re_tx_mbuf[idx] == NULL) {
1960 		m_head = ifq_poll(&ifp->if_snd);
1961 		if (m_head == NULL)
1962 			break;
1963 		m_head2 = m_head;
1964 		if (re_encap(sc, &m_head2, &idx, &called_defrag)) {
1965 			/*
1966 			 * If we could not encapsulate the defragged packet,
1967 			 * the returned m_head2 is garbage and we must dequeue
1968 			 * and throw away the original packet.
1969 			 */
1970 			if (called_defrag) {
1971 				ifq_dequeue(&ifp->if_snd, m_head);
1972 				m_freem(m_head);
1973 			}
1974 			ifp->if_flags |= IFF_OACTIVE;
1975 			break;
1976 		}
1977 
1978 		/*
1979 		 * Clean out the packet we encapsulated.  If we defragged
1980 		 * the packet the m_head2 is the one that got encapsulated
1981 		 * and the original must be thrown away.  Otherwise m_head2
1982 		 * *IS* the original.
1983 		 */
1984 		ifq_dequeue(&ifp->if_snd, m_head);
1985 		if (called_defrag)
1986 			m_freem(m_head);
1987 		need_trans = 1;
1988 
1989 		/*
1990 		 * If there's a BPF listener, bounce a copy of this frame
1991 		 * to him.
1992 		 */
1993 		BPF_MTAP(ifp, m_head2);
1994 	}
1995 
1996 	if (!need_trans) {
1997 		if (RE_TX_MODERATION_IS_ENABLED(sc) &&
1998 		    sc->re_ldata.re_tx_free != RE_TX_DESC_CNT)
1999 			CSR_WRITE_4(sc, RE_TIMERCNT, 1);
2000 		return;
2001 	}
2002 
2003 	/* Flush the TX descriptors */
2004 	bus_dmamap_sync(sc->re_ldata.re_tx_list_tag,
2005 			sc->re_ldata.re_tx_list_map, BUS_DMASYNC_PREWRITE);
2006 
2007 	sc->re_ldata.re_tx_prodidx = idx;
2008 
2009 	/*
2010 	 * RealTek put the TX poll request register in a different
2011 	 * location on the 8169 gigE chip. I don't know why.
2012 	 */
2013 	CSR_WRITE_1(sc, sc->re_txstart, RE_TXSTART_START);
2014 
2015 	if (RE_TX_MODERATION_IS_ENABLED(sc)) {
2016 		/*
2017 		 * Use the countdown timer for interrupt moderation.
2018 		 * 'TX done' interrupts are disabled. Instead, we reset the
2019 		 * countdown timer, which will begin counting until it hits
2020 		 * the value in the TIMERINT register, and then trigger an
2021 		 * interrupt. Each time we write to the TIMERCNT register,
2022 		 * the timer count is reset to 0.
2023 		 */
2024 		CSR_WRITE_4(sc, RE_TIMERCNT, 1);
2025 	}
2026 
2027 	/*
2028 	 * Set a timeout in case the chip goes out to lunch.
2029 	 */
2030 	ifp->if_timer = 5;
2031 }
2032 
2033 static void
2034 re_init(void *xsc)
2035 {
2036 	struct re_softc *sc = xsc;
2037 	struct ifnet *ifp = &sc->arpcom.ac_if;
2038 	struct mii_data *mii;
2039 	uint32_t rxcfg = 0;
2040 
2041 	mii = device_get_softc(sc->re_miibus);
2042 
2043 	/*
2044 	 * Cancel pending I/O and free all RX/TX buffers.
2045 	 */
2046 	re_stop(sc);
2047 
2048 	/*
2049 	 * Enable C+ RX and TX mode, as well as VLAN stripping and
2050 	 * RX checksum offload. We must configure the C+ register
2051 	 * before all others.
2052 	 */
2053 	CSR_WRITE_2(sc, RE_CPLUS_CMD, RE_CPLUSCMD_RXENB | RE_CPLUSCMD_TXENB |
2054 		    RE_CPLUSCMD_PCI_MRW | RE_CPLUSCMD_VLANSTRIP |
2055 		    (ifp->if_capenable & IFCAP_RXCSUM ?
2056 		     RE_CPLUSCMD_RXCSUM_ENB : 0));
2057 
2058 	/*
2059 	 * Init our MAC address.  Even though the chipset
2060 	 * documentation doesn't mention it, we need to enter "Config
2061 	 * register write enable" mode to modify the ID registers.
2062 	 */
2063 	CSR_WRITE_1(sc, RE_EECMD, RE_EEMODE_WRITECFG);
2064 	CSR_WRITE_4(sc, RE_IDR0,
2065 	    htole32(*(uint32_t *)(&sc->arpcom.ac_enaddr[0])));
2066 	CSR_WRITE_4(sc, RE_IDR4,
2067 	    htole32(*(uint32_t *)(&sc->arpcom.ac_enaddr[4])));
2068 	CSR_WRITE_1(sc, RE_EECMD, RE_EEMODE_OFF);
2069 
2070 	/*
2071 	 * For C+ mode, initialize the RX descriptors and mbufs.
2072 	 */
2073 	re_rx_list_init(sc);
2074 	re_tx_list_init(sc);
2075 
2076 	/*
2077 	 * Enable transmit and receive.
2078 	 */
2079 	CSR_WRITE_1(sc, RE_COMMAND, RE_CMD_TX_ENB|RE_CMD_RX_ENB);
2080 
2081 	/*
2082 	 * Set the initial TX and RX configuration.
2083 	 */
2084 	if (sc->re_testmode) {
2085 		if (sc->re_type == RE_8169)
2086 			CSR_WRITE_4(sc, RE_TXCFG,
2087 				    RE_TXCFG_CONFIG | RE_LOOPTEST_ON);
2088 		else
2089 			CSR_WRITE_4(sc, RE_TXCFG,
2090 				    RE_TXCFG_CONFIG | RE_LOOPTEST_ON_CPLUS);
2091 	} else
2092 		CSR_WRITE_4(sc, RE_TXCFG, RE_TXCFG_CONFIG);
2093 	CSR_WRITE_4(sc, RE_RXCFG, RE_RXCFG_CONFIG);
2094 
2095 	/* Set the individual bit to receive frames for this host only. */
2096 	rxcfg = CSR_READ_4(sc, RE_RXCFG);
2097 	rxcfg |= RE_RXCFG_RX_INDIV;
2098 
2099 	/* If we want promiscuous mode, set the allframes bit. */
2100 	if (ifp->if_flags & IFF_PROMISC) {
2101 		rxcfg |= RE_RXCFG_RX_ALLPHYS;
2102 		CSR_WRITE_4(sc, RE_RXCFG, rxcfg);
2103 	} else {
2104 		rxcfg &= ~RE_RXCFG_RX_ALLPHYS;
2105 		CSR_WRITE_4(sc, RE_RXCFG, rxcfg);
2106 	}
2107 
2108 	/*
2109 	 * Set capture broadcast bit to capture broadcast frames.
2110 	 */
2111 	if (ifp->if_flags & IFF_BROADCAST) {
2112 		rxcfg |= RE_RXCFG_RX_BROAD;
2113 		CSR_WRITE_4(sc, RE_RXCFG, rxcfg);
2114 	} else {
2115 		rxcfg &= ~RE_RXCFG_RX_BROAD;
2116 		CSR_WRITE_4(sc, RE_RXCFG, rxcfg);
2117 	}
2118 
2119 	/*
2120 	 * Program the multicast filter, if necessary.
2121 	 */
2122 	re_setmulti(sc);
2123 
2124 #ifdef DEVICE_POLLING
2125 	/*
2126 	 * Disable interrupts if we are polling.
2127 	 */
2128 	if (ifp->if_flags & IFF_POLLING)
2129 		CSR_WRITE_2(sc, RE_IMR, 0);
2130 	else	/* otherwise ... */
2131 #endif /* DEVICE_POLLING */
2132 	/*
2133 	 * Enable interrupts.
2134 	 */
2135 	if (sc->re_testmode)
2136 		CSR_WRITE_2(sc, RE_IMR, 0);
2137 	else
2138 		CSR_WRITE_2(sc, RE_IMR, sc->re_intrs);
2139 	CSR_WRITE_2(sc, RE_ISR, sc->re_intrs);
2140 
2141 	/* Set initial TX threshold */
2142 	sc->re_txthresh = RE_TX_THRESH_INIT;
2143 
2144 	/* Start RX/TX process. */
2145 	CSR_WRITE_4(sc, RE_MISSEDPKT, 0);
2146 #ifdef notdef
2147 	/* Enable receiver and transmitter. */
2148 	CSR_WRITE_1(sc, RE_COMMAND, RE_CMD_TX_ENB|RE_CMD_RX_ENB);
2149 #endif
2150 	/*
2151 	 * Load the addresses of the RX and TX lists into the chip.
2152 	 */
2153 
2154 	CSR_WRITE_4(sc, RE_RXLIST_ADDR_HI,
2155 	    RE_ADDR_HI(sc->re_ldata.re_rx_list_addr));
2156 	CSR_WRITE_4(sc, RE_RXLIST_ADDR_LO,
2157 	    RE_ADDR_LO(sc->re_ldata.re_rx_list_addr));
2158 
2159 	CSR_WRITE_4(sc, RE_TXLIST_ADDR_HI,
2160 	    RE_ADDR_HI(sc->re_ldata.re_tx_list_addr));
2161 	CSR_WRITE_4(sc, RE_TXLIST_ADDR_LO,
2162 	    RE_ADDR_LO(sc->re_ldata.re_tx_list_addr));
2163 
2164 	CSR_WRITE_1(sc, RE_EARLY_TX_THRESH, 16);
2165 
2166 	if (RE_TX_MODERATION_IS_ENABLED(sc)) {
2167 		/*
2168 		 * Initialize the timer interrupt register so that
2169 		 * a timer interrupt will be generated once the timer
2170 		 * reaches a certain number of ticks. The timer is
2171 		 * reloaded on each transmit. This gives us TX interrupt
2172 		 * moderation, which dramatically improves TX frame rate.
2173 		 */
2174 		if (sc->re_type == RE_8169)
2175 			CSR_WRITE_4(sc, RE_TIMERINT_8169, 0x800);
2176 		else
2177 			CSR_WRITE_4(sc, RE_TIMERINT, 0x400);
2178 	}
2179 
2180 	/*
2181 	 * For 8169 gigE NICs, set the max allowed RX packet
2182 	 * size so we can receive jumbo frames.
2183 	 */
2184 	if (sc->re_type == RE_8169)
2185 		CSR_WRITE_2(sc, RE_MAXRXPKTLEN, 16383);
2186 
2187 	if (sc->re_testmode) {
2188 		return;
2189 	}
2190 
2191 	mii_mediachg(mii);
2192 
2193 	CSR_WRITE_1(sc, RE_CFG1, RE_CFG1_DRVLOAD|RE_CFG1_FULLDUPLEX);
2194 
2195 	ifp->if_flags |= IFF_RUNNING;
2196 	ifp->if_flags &= ~IFF_OACTIVE;
2197 
2198 	sc->re_link = 0;
2199 	callout_reset(&sc->re_timer, hz, re_tick, sc);
2200 }
2201 
2202 /*
2203  * Set media options.
2204  */
2205 static int
2206 re_ifmedia_upd(struct ifnet *ifp)
2207 {
2208 	struct re_softc *sc = ifp->if_softc;
2209 	struct mii_data *mii;
2210 
2211 	mii = device_get_softc(sc->re_miibus);
2212 	mii_mediachg(mii);
2213 
2214 	return(0);
2215 }
2216 
2217 /*
2218  * Report current media status.
2219  */
2220 static void
2221 re_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
2222 {
2223 	struct re_softc *sc = ifp->if_softc;
2224 	struct mii_data *mii;
2225 
2226 	mii = device_get_softc(sc->re_miibus);
2227 
2228 	mii_pollstat(mii);
2229 	ifmr->ifm_active = mii->mii_media_active;
2230 	ifmr->ifm_status = mii->mii_media_status;
2231 }
2232 
2233 static int
2234 re_ioctl(struct ifnet *ifp, u_long command, caddr_t data, struct ucred *cr)
2235 {
2236 	struct re_softc *sc = ifp->if_softc;
2237 	struct ifreq *ifr = (struct ifreq *) data;
2238 	struct mii_data *mii;
2239 	int error = 0;
2240 
2241 	switch(command) {
2242 	case SIOCSIFMTU:
2243 		if (ifr->ifr_mtu > RE_JUMBO_MTU)
2244 			error = EINVAL;
2245 		ifp->if_mtu = ifr->ifr_mtu;
2246 		break;
2247 	case SIOCSIFFLAGS:
2248 		if (ifp->if_flags & IFF_UP)
2249 			re_init(sc);
2250 		else if (ifp->if_flags & IFF_RUNNING)
2251 			re_stop(sc);
2252 		break;
2253 	case SIOCADDMULTI:
2254 	case SIOCDELMULTI:
2255 		re_setmulti(sc);
2256 		error = 0;
2257 		break;
2258 	case SIOCGIFMEDIA:
2259 	case SIOCSIFMEDIA:
2260 		mii = device_get_softc(sc->re_miibus);
2261 		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, command);
2262 		break;
2263 	case SIOCSIFCAP:
2264 		ifp->if_capenable &= ~(IFCAP_HWCSUM);
2265 		ifp->if_capenable |=
2266 		    ifr->ifr_reqcap & (IFCAP_HWCSUM);
2267 		if (ifp->if_capenable & IFCAP_TXCSUM)
2268 			ifp->if_hwassist = RE_CSUM_FEATURES;
2269 		else
2270 			ifp->if_hwassist = 0;
2271 		if (ifp->if_flags & IFF_RUNNING)
2272 			re_init(sc);
2273 		break;
2274 	default:
2275 		error = ether_ioctl(ifp, command, data);
2276 		break;
2277 	}
2278 	return(error);
2279 }
2280 
2281 static void
2282 re_watchdog(struct ifnet *ifp)
2283 {
2284 	struct re_softc *sc = ifp->if_softc;
2285 
2286 	if_printf(ifp, "watchdog timeout\n");
2287 
2288 	ifp->if_oerrors++;
2289 
2290 	re_txeof(sc);
2291 	re_rxeof(sc);
2292 
2293 	re_init(sc);
2294 
2295 	if (!ifq_is_empty(&ifp->if_snd))
2296 		ifp->if_start(ifp);
2297 }
2298 
2299 /*
2300  * Stop the adapter and free any mbufs allocated to the
2301  * RX and TX lists.
2302  */
2303 static void
2304 re_stop(struct re_softc *sc)
2305 {
2306 	struct ifnet *ifp = &sc->arpcom.ac_if;
2307 	int i;
2308 
2309 	ifp->if_timer = 0;
2310 	callout_stop(&sc->re_timer);
2311 
2312 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
2313 
2314 	CSR_WRITE_1(sc, RE_COMMAND, 0x00);
2315 	CSR_WRITE_2(sc, RE_IMR, 0x0000);
2316 	CSR_WRITE_2(sc, RE_ISR, 0xFFFF);
2317 
2318 	if (sc->re_head != NULL) {
2319 		m_freem(sc->re_head);
2320 		sc->re_head = sc->re_tail = NULL;
2321 	}
2322 
2323 	/* Free the TX list buffers. */
2324 	for (i = 0; i < RE_TX_DESC_CNT; i++) {
2325 		if (sc->re_ldata.re_tx_mbuf[i] != NULL) {
2326 			bus_dmamap_unload(sc->re_ldata.re_mtag,
2327 					  sc->re_ldata.re_tx_dmamap[i]);
2328 			m_freem(sc->re_ldata.re_tx_mbuf[i]);
2329 			sc->re_ldata.re_tx_mbuf[i] = NULL;
2330 		}
2331 	}
2332 
2333 	/* Free the RX list buffers. */
2334 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
2335 		if (sc->re_ldata.re_rx_mbuf[i] != NULL) {
2336 			bus_dmamap_unload(sc->re_ldata.re_mtag,
2337 					  sc->re_ldata.re_rx_dmamap[i]);
2338 			m_freem(sc->re_ldata.re_rx_mbuf[i]);
2339 			sc->re_ldata.re_rx_mbuf[i] = NULL;
2340 		}
2341 	}
2342 }
2343 
2344 /*
2345  * Device suspend routine.  Stop the interface and save some PCI
2346  * settings in case the BIOS doesn't restore them properly on
2347  * resume.
2348  */
2349 static int
2350 re_suspend(device_t dev)
2351 {
2352 #ifndef BURN_BRIDGES
2353 	int i;
2354 #endif
2355 	struct re_softc *sc = device_get_softc(dev);
2356 
2357 	re_stop(sc);
2358 
2359 #ifndef BURN_BRIDGES
2360 	for (i = 0; i < 5; i++)
2361 		sc->saved_maps[i] = pci_read_config(dev, PCIR_MAPS + i * 4, 4);
2362 	sc->saved_biosaddr = pci_read_config(dev, PCIR_BIOS, 4);
2363 	sc->saved_intline = pci_read_config(dev, PCIR_INTLINE, 1);
2364 	sc->saved_cachelnsz = pci_read_config(dev, PCIR_CACHELNSZ, 1);
2365 	sc->saved_lattimer = pci_read_config(dev, PCIR_LATTIMER, 1);
2366 #endif
2367 
2368 	sc->suspended = 1;
2369 
2370 	return (0);
2371 }
2372 
2373 /*
2374  * Device resume routine.  Restore some PCI settings in case the BIOS
2375  * doesn't, re-enable busmastering, and restart the interface if
2376  * appropriate.
2377  */
2378 static int
2379 re_resume(device_t dev)
2380 {
2381 	struct re_softc *sc = device_get_softc(dev);
2382 	struct ifnet *ifp = &sc->arpcom.ac_if;
2383 #ifndef BURN_BRIDGES
2384 	int i;
2385 #endif
2386 
2387 #ifndef BURN_BRIDGES
2388 	/* better way to do this? */
2389 	for (i = 0; i < 5; i++)
2390 		pci_write_config(dev, PCIR_MAPS + i * 4, sc->saved_maps[i], 4);
2391 	pci_write_config(dev, PCIR_BIOS, sc->saved_biosaddr, 4);
2392 	pci_write_config(dev, PCIR_INTLINE, sc->saved_intline, 1);
2393 	pci_write_config(dev, PCIR_CACHELNSZ, sc->saved_cachelnsz, 1);
2394 	pci_write_config(dev, PCIR_LATTIMER, sc->saved_lattimer, 1);
2395 
2396 	/* reenable busmastering */
2397 	pci_enable_busmaster(dev);
2398 	pci_enable_io(dev, SYS_RES_IOPORT);
2399 #endif
2400 
2401 	/* reinitialize interface if necessary */
2402 	if (ifp->if_flags & IFF_UP)
2403 		re_init(sc);
2404 
2405 	sc->suspended = 0;
2406 
2407 	return (0);
2408 }
2409 
2410 /*
2411  * Stop all chip I/O so that the kernel's probe routines don't
2412  * get confused by errant DMAs when rebooting.
2413  */
2414 static void
2415 re_shutdown(device_t dev)
2416 {
2417 	struct re_softc *sc = device_get_softc(dev);
2418 	struct ifnet *ifp = &sc->arpcom.ac_if;
2419 
2420 	lwkt_serialize_enter(ifp->if_serializer);
2421 	re_stop(sc);
2422 	lwkt_serialize_exit(ifp->if_serializer);
2423 }
2424 
2425 static int
2426 re_sysctl_tx_moderation(SYSCTL_HANDLER_ARGS)
2427 {
2428 	struct re_softc *sc = arg1;
2429 	struct ifnet *ifp = &sc->arpcom.ac_if;
2430 	int error = 0, mod, mod_old;
2431 
2432 	lwkt_serialize_enter(ifp->if_serializer);
2433 
2434 	mod_old = mod = RE_TX_MODERATION_IS_ENABLED(sc);
2435 
2436 	error = sysctl_handle_int(oidp, &mod, 0, req);
2437 	if (error || req->newptr == NULL || mod == mod_old)
2438 		goto back;
2439 	if (mod != 0 && mod != 1) {
2440 		error = EINVAL;
2441 		goto back;
2442 	}
2443 
2444 	if (mod)
2445 		RE_ENABLE_TX_MODERATION(sc);
2446 	else
2447 		RE_DISABLE_TX_MODERATION(sc);
2448 
2449 	re_init(sc);
2450 back:
2451 	lwkt_serialize_exit(ifp->if_serializer);
2452 	return error;
2453 }
2454