xref: /dflybsd-src/sys/dev/netif/re/if_re.c (revision 52347f71f500fe4113e37cca4ea6edc6150698e7)
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 1.38 2008/03/10 12:59:51 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 #include <net/vlan/if_vlan_ether.h>
139 
140 #include <dev/netif/mii_layer/mii.h>
141 #include <dev/netif/mii_layer/miivar.h>
142 
143 #include <bus/pci/pcidevs.h>
144 #include <bus/pci/pcireg.h>
145 #include <bus/pci/pcivar.h>
146 
147 /* "device miibus" required.  See GENERIC if you get errors here. */
148 #include "miibus_if.h"
149 
150 #include <dev/netif/re/if_rereg.h>
151 #include <dev/netif/re/if_revar.h>
152 
153 #define RE_CSUM_FEATURES    (CSUM_IP | CSUM_TCP | CSUM_UDP)
154 #if 0
155 #define RE_DISABLE_HWCSUM
156 #endif
157 
158 /*
159  * Various supported device vendors/types and their names.
160  */
161 static const struct re_type re_devs[] = {
162 	{ PCI_VENDOR_DLINK, PCI_PRODUCT_DLINK_DGE528T, RE_HWREV_8169S,
163 		"D-Link DGE-528(T) Gigabit Ethernet Adapter" },
164 	{ PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8139, RE_HWREV_8139CPLUS,
165 		"RealTek 8139C+ 10/100BaseTX" },
166 	{ PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8101E, RE_HWREV_8101E,
167 		"RealTek 8101E PCIe 10/100baseTX" },
168 	{ PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8168, RE_HWREV_8168_SPIN1,
169 		"RealTek 8168/8111B PCIe Gigabit Ethernet" },
170 	{ PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8168, RE_HWREV_8168_SPIN2,
171 		"RealTek 8168/8111B PCIe Gigabit Ethernet" },
172 	{ PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8168, RE_HWREV_8168_SPIN3,
173 		"RealTek 8168B/8111B PCIe Gigabit Ethernet" },
174 	{ PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8169, RE_HWREV_8169,
175 		"RealTek 8169 Gigabit Ethernet" },
176 	{ PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8169, RE_HWREV_8169S,
177 		"RealTek 8169S Single-chip Gigabit Ethernet" },
178 	{ PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8169, RE_HWREV_8169_8110SB,
179 		"RealTek 8169SB/8110SB Single-chip Gigabit Ethernet" },
180 	{ PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8169, RE_HWREV_8169_8110SC,
181 		"RealTek 8169SC/8110SC Single-chip Gigabit Ethernet" },
182 	{ PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8169SC, RE_HWREV_8169_8110SC,
183 		"RealTek 8169SC/8110SC Single-chip Gigabit Ethernet" },
184 	{ PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8169, RE_HWREV_8110S,
185 		"RealTek 8110S Single-chip Gigabit Ethernet" },
186 	{ PCI_VENDOR_COREGA, PCI_PRODUCT_COREGA_CG_LAPCIGT, RE_HWREV_8169S,
187 		"Corega CG-LAPCIGT Gigabit Ethernet" },
188 	{ PCI_VENDOR_LINKSYS, PCI_PRODUCT_LINKSYS_EG1032, RE_HWREV_8169S,
189 		"Linksys EG1032 Gigabit Ethernet" },
190 	{ PCI_VENDOR_USR2, PCI_PRODUCT_USR2_997902, RE_HWREV_8169S,
191 		"US Robotics 997902 Gigabit Ethernet" },
192 	{ 0, 0, 0, NULL }
193 };
194 
195 static const struct re_hwrev re_hwrevs[] = {
196 	{ RE_HWREV_8139CPLUS,	RE_8139CPLUS,	RE_F_HASMPC,	"C+" },
197 	{ RE_HWREV_8168_SPIN1,	RE_8169,	RE_F_PCIE,	"8168" },
198 	{ RE_HWREV_8168_SPIN2,	RE_8169,	RE_F_PCIE,	"8168" },
199 	{ RE_HWREV_8168_SPIN3,	RE_8169,	RE_F_PCIE,	"8168" },
200 	{ RE_HWREV_8169,	RE_8169,	RE_F_HASMPC,	"8169" },
201 	{ RE_HWREV_8169S,	RE_8169,	RE_F_HASMPC,	"8169S" },
202 	{ RE_HWREV_8110S,	RE_8169,	RE_F_HASMPC,	"8110S" },
203 	{ RE_HWREV_8169_8110SB,	RE_8169,	RE_F_HASMPC,	"8169SB" },
204 	{ RE_HWREV_8169_8110SC,	RE_8169,	0,		"8169SC" },
205 	{ RE_HWREV_8100E,	RE_8169,	RE_F_HASMPC,	"8100E" },
206 	{ RE_HWREV_8101E,	RE_8169,	RE_F_PCIE,	"8101E" },
207 	{ 0, 0, 0, NULL }
208 };
209 
210 static int	re_probe(device_t);
211 static int	re_attach(device_t);
212 static int	re_detach(device_t);
213 
214 static int	re_encap(struct re_softc *, struct mbuf **, int *, int *);
215 
216 static void	re_dma_map_addr(void *, bus_dma_segment_t *, int, int);
217 static void	re_dma_map_desc(void *, bus_dma_segment_t *, int,
218 				bus_size_t, int);
219 static int	re_allocmem(device_t, struct re_softc *);
220 static int	re_newbuf(struct re_softc *, int, struct mbuf *);
221 static int	re_rx_list_init(struct re_softc *);
222 static int	re_tx_list_init(struct re_softc *);
223 static void	re_rxeof(struct re_softc *);
224 static void	re_txeof(struct re_softc *);
225 static void	re_intr(void *);
226 static void	re_tick(void *);
227 static void	re_tick_serialized(void *);
228 static void	re_start(struct ifnet *);
229 static int	re_ioctl(struct ifnet *, u_long, caddr_t, struct ucred *);
230 static void	re_init(void *);
231 static void	re_stop(struct re_softc *);
232 static void	re_watchdog(struct ifnet *);
233 static int	re_suspend(device_t);
234 static int	re_resume(device_t);
235 static void	re_shutdown(device_t);
236 static int	re_ifmedia_upd(struct ifnet *);
237 static void	re_ifmedia_sts(struct ifnet *, struct ifmediareq *);
238 
239 static void	re_eeprom_putbyte(struct re_softc *, int);
240 static void	re_eeprom_getword(struct re_softc *, int, u_int16_t *);
241 static void	re_read_eeprom(struct re_softc *, caddr_t, int, int);
242 static int	re_gmii_readreg(device_t, int, int);
243 static int	re_gmii_writereg(device_t, int, int, int);
244 
245 static int	re_miibus_readreg(device_t, int, int);
246 static int	re_miibus_writereg(device_t, int, int, int);
247 static void	re_miibus_statchg(device_t);
248 
249 static void	re_setmulti(struct re_softc *);
250 static void	re_reset(struct re_softc *);
251 
252 #ifdef RE_DIAG
253 static int	re_diag(struct re_softc *);
254 #endif
255 
256 #ifdef DEVICE_POLLING
257 static void	re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count);
258 #endif
259 
260 static int	re_sysctl_tx_moderation(SYSCTL_HANDLER_ARGS);
261 
262 static device_method_t re_methods[] = {
263 	/* Device interface */
264 	DEVMETHOD(device_probe,		re_probe),
265 	DEVMETHOD(device_attach,	re_attach),
266 	DEVMETHOD(device_detach,	re_detach),
267 	DEVMETHOD(device_suspend,	re_suspend),
268 	DEVMETHOD(device_resume,	re_resume),
269 	DEVMETHOD(device_shutdown,	re_shutdown),
270 
271 	/* bus interface */
272 	DEVMETHOD(bus_print_child,	bus_generic_print_child),
273 	DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
274 
275 	/* MII interface */
276 	DEVMETHOD(miibus_readreg,	re_miibus_readreg),
277 	DEVMETHOD(miibus_writereg,	re_miibus_writereg),
278 	DEVMETHOD(miibus_statchg,	re_miibus_statchg),
279 
280 	{ 0, 0 }
281 };
282 
283 static driver_t re_driver = {
284 	"re",
285 	re_methods,
286 	sizeof(struct re_softc)
287 };
288 
289 static devclass_t re_devclass;
290 
291 DECLARE_DUMMY_MODULE(if_re);
292 DRIVER_MODULE(if_re, pci, re_driver, re_devclass, 0, 0);
293 DRIVER_MODULE(if_re, cardbus, re_driver, re_devclass, 0, 0);
294 DRIVER_MODULE(miibus, re, miibus_driver, miibus_devclass, 0, 0);
295 
296 #define EE_SET(x)	\
297 	CSR_WRITE_1(sc, RE_EECMD, CSR_READ_1(sc, RE_EECMD) | (x))
298 
299 #define EE_CLR(x)	\
300 	CSR_WRITE_1(sc, RE_EECMD, CSR_READ_1(sc, RE_EECMD) & ~(x))
301 
302 /*
303  * Send a read command and address to the EEPROM, check for ACK.
304  */
305 static void
306 re_eeprom_putbyte(struct re_softc *sc, int addr)
307 {
308 	int d, i;
309 
310 	d = addr | (RE_9346_READ << sc->re_eewidth);
311 
312 	/*
313 	 * Feed in each bit and strobe the clock.
314 	 */
315 	for (i = 1 << (sc->re_eewidth + 3); i; i >>= 1) {
316 		if (d & i)
317 			EE_SET(RE_EE_DATAIN);
318 		else
319 			EE_CLR(RE_EE_DATAIN);
320 		DELAY(100);
321 		EE_SET(RE_EE_CLK);
322 		DELAY(150);
323 		EE_CLR(RE_EE_CLK);
324 		DELAY(100);
325 	}
326 }
327 
328 /*
329  * Read a word of data stored in the EEPROM at address 'addr.'
330  */
331 static void
332 re_eeprom_getword(struct re_softc *sc, int addr, uint16_t *dest)
333 {
334 	int i;
335 	uint16_t word = 0;
336 
337 	/*
338 	 * Send address of word we want to read.
339 	 */
340 	re_eeprom_putbyte(sc, addr);
341 
342 	/*
343 	 * Start reading bits from EEPROM.
344 	 */
345 	for (i = 0x8000; i != 0; i >>= 1) {
346 		EE_SET(RE_EE_CLK);
347 		DELAY(100);
348 		if (CSR_READ_1(sc, RE_EECMD) & RE_EE_DATAOUT)
349 			word |= i;
350 		EE_CLR(RE_EE_CLK);
351 		DELAY(100);
352 	}
353 
354 	*dest = word;
355 }
356 
357 /*
358  * Read a sequence of words from the EEPROM.
359  */
360 static void
361 re_read_eeprom(struct re_softc *sc, caddr_t dest, int off, int cnt)
362 {
363 	int i;
364 	uint16_t word = 0, *ptr;
365 
366 	CSR_SETBIT_1(sc, RE_EECMD, RE_EEMODE_PROGRAM);
367 	DELAY(100);
368 
369 	for (i = 0; i < cnt; i++) {
370 		CSR_SETBIT_1(sc, RE_EECMD, RE_EE_SEL);
371 		re_eeprom_getword(sc, off + i, &word);
372 		CSR_CLRBIT_1(sc, RE_EECMD, RE_EE_SEL);
373 		ptr = (uint16_t *)(dest + (i * 2));
374 		*ptr = word;
375 	}
376 
377 	CSR_CLRBIT_1(sc, RE_EECMD, RE_EEMODE_PROGRAM);
378 }
379 
380 static int
381 re_gmii_readreg(device_t dev, int phy, int reg)
382 {
383 	struct re_softc *sc = device_get_softc(dev);
384 	u_int32_t rval;
385 	int i;
386 
387 	if (phy != 1)
388 		return(0);
389 
390 	/* Let the rgephy driver read the GMEDIASTAT register */
391 
392 	if (reg == RE_GMEDIASTAT)
393 		return(CSR_READ_1(sc, RE_GMEDIASTAT));
394 
395 	CSR_WRITE_4(sc, RE_PHYAR, reg << 16);
396 	DELAY(1000);
397 
398 	for (i = 0; i < RE_TIMEOUT; i++) {
399 		rval = CSR_READ_4(sc, RE_PHYAR);
400 		if (rval & RE_PHYAR_BUSY)
401 			break;
402 		DELAY(100);
403 	}
404 
405 	if (i == RE_TIMEOUT) {
406 		device_printf(dev, "PHY read failed\n");
407 		return(0);
408 	}
409 
410 	return(rval & RE_PHYAR_PHYDATA);
411 }
412 
413 static int
414 re_gmii_writereg(device_t dev, int phy, int reg, int data)
415 {
416 	struct re_softc *sc = device_get_softc(dev);
417 	uint32_t rval;
418 	int i;
419 
420 	CSR_WRITE_4(sc, RE_PHYAR,
421 		    (reg << 16) | (data & RE_PHYAR_PHYDATA) | RE_PHYAR_BUSY);
422 	DELAY(1000);
423 
424 	for (i = 0; i < RE_TIMEOUT; i++) {
425 		rval = CSR_READ_4(sc, RE_PHYAR);
426 		if ((rval & RE_PHYAR_BUSY) == 0)
427 			break;
428 		DELAY(100);
429 	}
430 
431 	if (i == RE_TIMEOUT)
432 		device_printf(dev, "PHY write failed\n");
433 
434 	return(0);
435 }
436 
437 static int
438 re_miibus_readreg(device_t dev, int phy, int reg)
439 {
440 	struct re_softc	*sc = device_get_softc(dev);
441 	uint16_t rval = 0;
442 	uint16_t re8139_reg = 0;
443 
444 	if (sc->re_type == RE_8169) {
445 		rval = re_gmii_readreg(dev, phy, reg);
446 		return(rval);
447 	}
448 
449 	/* Pretend the internal PHY is only at address 0 */
450 	if (phy)
451 		return(0);
452 
453 	switch(reg) {
454 	case MII_BMCR:
455 		re8139_reg = RE_BMCR;
456 		break;
457 	case MII_BMSR:
458 		re8139_reg = RE_BMSR;
459 		break;
460 	case MII_ANAR:
461 		re8139_reg = RE_ANAR;
462 		break;
463 	case MII_ANER:
464 		re8139_reg = RE_ANER;
465 		break;
466 	case MII_ANLPAR:
467 		re8139_reg = RE_LPAR;
468 		break;
469 	case MII_PHYIDR1:
470 	case MII_PHYIDR2:
471 		return(0);
472 	/*
473 	 * Allow the rlphy driver to read the media status
474 	 * register. If we have a link partner which does not
475 	 * support NWAY, this is the register which will tell
476 	 * us the results of parallel detection.
477 	 */
478 	case RE_MEDIASTAT:
479 		return(CSR_READ_1(sc, RE_MEDIASTAT));
480 	default:
481 		device_printf(dev, "bad phy register\n");
482 		return(0);
483 	}
484 	rval = CSR_READ_2(sc, re8139_reg);
485 	if (sc->re_type == RE_8139CPLUS && re8139_reg == RE_BMCR) {
486 		/* 8139C+ has different bit layout. */
487 		rval &= ~(BMCR_LOOP | BMCR_ISO);
488 	}
489 	return(rval);
490 }
491 
492 static int
493 re_miibus_writereg(device_t dev, int phy, int reg, int data)
494 {
495 	struct re_softc *sc= device_get_softc(dev);
496 	u_int16_t re8139_reg = 0;
497 
498 	if (sc->re_type == RE_8169)
499 		return(re_gmii_writereg(dev, phy, reg, data));
500 
501 	/* Pretend the internal PHY is only at address 0 */
502 	if (phy)
503 		return(0);
504 
505 	switch(reg) {
506 	case MII_BMCR:
507 		re8139_reg = RE_BMCR;
508 		if (sc->re_type == RE_8139CPLUS) {
509 			/* 8139C+ has different bit layout. */
510 			data &= ~(BMCR_LOOP | BMCR_ISO);
511 		}
512 		break;
513 	case MII_BMSR:
514 		re8139_reg = RE_BMSR;
515 		break;
516 	case MII_ANAR:
517 		re8139_reg = RE_ANAR;
518 		break;
519 	case MII_ANER:
520 		re8139_reg = RE_ANER;
521 		break;
522 	case MII_ANLPAR:
523 		re8139_reg = RE_LPAR;
524 		break;
525 	case MII_PHYIDR1:
526 	case MII_PHYIDR2:
527 		return(0);
528 	default:
529 		device_printf(dev, "bad phy register\n");
530 		return(0);
531 	}
532 	CSR_WRITE_2(sc, re8139_reg, data);
533 	return(0);
534 }
535 
536 static void
537 re_miibus_statchg(device_t dev)
538 {
539 }
540 
541 /*
542  * Program the 64-bit multicast hash filter.
543  */
544 static void
545 re_setmulti(struct re_softc *sc)
546 {
547 	struct ifnet *ifp = &sc->arpcom.ac_if;
548 	int h = 0;
549 	uint32_t hashes[2] = { 0, 0 };
550 	struct ifmultiaddr *ifma;
551 	uint32_t rxfilt;
552 	int mcnt = 0;
553 
554 	rxfilt = CSR_READ_4(sc, RE_RXCFG);
555 
556 	if (ifp->if_flags & IFF_ALLMULTI || ifp->if_flags & IFF_PROMISC) {
557 		rxfilt |= RE_RXCFG_RX_MULTI;
558 		CSR_WRITE_4(sc, RE_RXCFG, rxfilt);
559 		CSR_WRITE_4(sc, RE_MAR0, 0xFFFFFFFF);
560 		CSR_WRITE_4(sc, RE_MAR4, 0xFFFFFFFF);
561 		return;
562 	}
563 
564 	/* first, zot all the existing hash bits */
565 	CSR_WRITE_4(sc, RE_MAR0, 0);
566 	CSR_WRITE_4(sc, RE_MAR4, 0);
567 
568 	/* now program new ones */
569 	LIST_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
570 		if (ifma->ifma_addr->sa_family != AF_LINK)
571 			continue;
572 		h = ether_crc32_be(LLADDR((struct sockaddr_dl *)
573 		    ifma->ifma_addr), ETHER_ADDR_LEN) >> 26;
574 		if (h < 32)
575 			hashes[0] |= (1 << h);
576 		else
577 			hashes[1] |= (1 << (h - 32));
578 		mcnt++;
579 	}
580 
581 	if (mcnt)
582 		rxfilt |= RE_RXCFG_RX_MULTI;
583 	else
584 		rxfilt &= ~RE_RXCFG_RX_MULTI;
585 
586 	CSR_WRITE_4(sc, RE_RXCFG, rxfilt);
587 
588 	/*
589 	 * For some unfathomable reason, RealTek decided to reverse
590 	 * the order of the multicast hash registers in the PCI Express
591 	 * parts. This means we have to write the hash pattern in reverse
592 	 * order for those devices.
593 	 */
594 	if (sc->re_flags & RE_F_PCIE) {
595 		CSR_WRITE_4(sc, RE_MAR0, bswap32(hashes[0]));
596 		CSR_WRITE_4(sc, RE_MAR4, bswap32(hashes[1]));
597 	} else {
598 		CSR_WRITE_4(sc, RE_MAR0, hashes[0]);
599 		CSR_WRITE_4(sc, RE_MAR4, hashes[1]);
600 	}
601 }
602 
603 static void
604 re_reset(struct re_softc *sc)
605 {
606 	int i;
607 
608 	CSR_WRITE_1(sc, RE_COMMAND, RE_CMD_RESET);
609 
610 	for (i = 0; i < RE_TIMEOUT; i++) {
611 		DELAY(10);
612 		if ((CSR_READ_1(sc, RE_COMMAND) & RE_CMD_RESET) == 0)
613 			break;
614 	}
615 	if (i == RE_TIMEOUT)
616 		if_printf(&sc->arpcom.ac_if, "reset never completed!\n");
617 
618 	CSR_WRITE_1(sc, 0x82, 1);
619 }
620 
621 #ifdef RE_DIAG
622 /*
623  * The following routine is designed to test for a defect on some
624  * 32-bit 8169 cards. Some of these NICs have the REQ64# and ACK64#
625  * lines connected to the bus, however for a 32-bit only card, they
626  * should be pulled high. The result of this defect is that the
627  * NIC will not work right if you plug it into a 64-bit slot: DMA
628  * operations will be done with 64-bit transfers, which will fail
629  * because the 64-bit data lines aren't connected.
630  *
631  * There's no way to work around this (short of talking a soldering
632  * iron to the board), however we can detect it. The method we use
633  * here is to put the NIC into digital loopback mode, set the receiver
634  * to promiscuous mode, and then try to send a frame. We then compare
635  * the frame data we sent to what was received. If the data matches,
636  * then the NIC is working correctly, otherwise we know the user has
637  * a defective NIC which has been mistakenly plugged into a 64-bit PCI
638  * slot. In the latter case, there's no way the NIC can work correctly,
639  * so we print out a message on the console and abort the device attach.
640  */
641 
642 static int
643 re_diag(struct re_softc *sc)
644 {
645 	struct ifnet *ifp = &sc->arpcom.ac_if;
646 	struct mbuf *m0;
647 	struct ether_header *eh;
648 	struct re_desc *cur_rx;
649 	uint16_t status;
650 	uint32_t rxstat;
651 	int total_len, i, error = 0, phyaddr;
652 	uint8_t dst[ETHER_ADDR_LEN] = { 0x00, 'h', 'e', 'l', 'l', 'o' };
653 	uint8_t src[ETHER_ADDR_LEN] = { 0x00, 'w', 'o', 'r', 'l', 'd' };
654 
655 	/* Allocate a single mbuf */
656 
657 	MGETHDR(m0, MB_DONTWAIT, MT_DATA);
658 	if (m0 == NULL)
659 		return(ENOBUFS);
660 
661 	/*
662 	 * Initialize the NIC in test mode. This sets the chip up
663 	 * so that it can send and receive frames, but performs the
664 	 * following special functions:
665 	 * - Puts receiver in promiscuous mode
666 	 * - Enables digital loopback mode
667 	 * - Leaves interrupts turned off
668 	 */
669 
670 	ifp->if_flags |= IFF_PROMISC;
671 	sc->re_testmode = 1;
672 	re_reset(sc);
673 	re_init(sc);
674 	sc->re_link = 1;
675 	if (sc->re_type == RE_8169)
676 		phyaddr = 1;
677 	else
678 		phyaddr = 0;
679 
680 	re_miibus_writereg(sc->re_dev, phyaddr, MII_BMCR, BMCR_RESET);
681 	for (i = 0; i < RE_TIMEOUT; i++) {
682 		status = re_miibus_readreg(sc->re_dev, phyaddr, MII_BMCR);
683 		if (!(status & BMCR_RESET))
684 			break;
685 	}
686 
687 	re_miibus_writereg(sc->re_dev, phyaddr, MII_BMCR, BMCR_LOOP);
688 	CSR_WRITE_2(sc, RE_ISR, RE_INTRS_DIAG);
689 
690 	DELAY(100000);
691 
692 	/* Put some data in the mbuf */
693 
694 	eh = mtod(m0, struct ether_header *);
695 	bcopy (dst, eh->ether_dhost, ETHER_ADDR_LEN);
696 	bcopy (src, eh->ether_shost, ETHER_ADDR_LEN);
697 	eh->ether_type = htons(ETHERTYPE_IP);
698 	m0->m_pkthdr.len = m0->m_len = ETHER_MIN_LEN - ETHER_CRC_LEN;
699 
700 	/*
701 	 * Queue the packet, start transmission.
702 	 * Note: ifq_handoff() ultimately calls re_start() for us.
703 	 */
704 
705 	CSR_WRITE_2(sc, RE_ISR, 0xFFFF);
706 	error = ifq_handoff(ifp, m0, NULL);
707 	if (error) {
708 		m0 = NULL;
709 		goto done;
710 	}
711 	m0 = NULL;
712 
713 	/* Wait for it to propagate through the chip */
714 
715 	DELAY(100000);
716 	for (i = 0; i < RE_TIMEOUT; i++) {
717 		status = CSR_READ_2(sc, RE_ISR);
718 		CSR_WRITE_2(sc, RE_ISR, status);
719 		if ((status & (RE_ISR_TIMEOUT_EXPIRED|RE_ISR_RX_OK)) ==
720 		    (RE_ISR_TIMEOUT_EXPIRED|RE_ISR_RX_OK))
721 			break;
722 		DELAY(10);
723 	}
724 
725 	if (i == RE_TIMEOUT) {
726 		if_printf(ifp, "diagnostic failed to receive packet "
727 			  "in loopback mode\n");
728 		error = EIO;
729 		goto done;
730 	}
731 
732 	/*
733 	 * The packet should have been dumped into the first
734 	 * entry in the RX DMA ring. Grab it from there.
735 	 */
736 
737 	bus_dmamap_sync(sc->re_ldata.re_rx_list_tag,
738 			sc->re_ldata.re_rx_list_map, BUS_DMASYNC_POSTREAD);
739 	bus_dmamap_sync(sc->re_ldata.re_mtag, sc->re_ldata.re_rx_dmamap[0],
740 			BUS_DMASYNC_POSTWRITE);
741 	bus_dmamap_unload(sc->re_ldata.re_mtag, sc->re_ldata.re_rx_dmamap[0]);
742 
743 	m0 = sc->re_ldata.re_rx_mbuf[0];
744 	sc->re_ldata.re_rx_mbuf[0] = NULL;
745 	eh = mtod(m0, struct ether_header *);
746 
747 	cur_rx = &sc->re_ldata.re_rx_list[0];
748 	total_len = RE_RXBYTES(cur_rx);
749 	rxstat = le32toh(cur_rx->re_cmdstat);
750 
751 	if (total_len != ETHER_MIN_LEN) {
752 		if_printf(ifp, "diagnostic failed, received short packet\n");
753 		error = EIO;
754 		goto done;
755 	}
756 
757 	/* Test that the received packet data matches what we sent. */
758 
759 	if (bcmp(eh->ether_dhost, dst, ETHER_ADDR_LEN) ||
760 	    bcmp(eh->ether_shost, &src, ETHER_ADDR_LEN) ||
761 	    be16toh(eh->ether_type) != ETHERTYPE_IP) {
762 		if_printf(ifp, "WARNING, DMA FAILURE!\n");
763 		if_printf(ifp, "expected TX data: %6D/%6D/0x%x\n",
764 		    dst, ":", src, ":", ETHERTYPE_IP);
765 		if_printf(ifp, "received RX data: %6D/%6D/0x%x\n",
766 		    eh->ether_dhost, ":",  eh->ether_shost, ":",
767 		    ntohs(eh->ether_type));
768 		if_printf(ifp, "You may have a defective 32-bit NIC plugged "
769 		    "into a 64-bit PCI slot.\n");
770 		if_printf(ifp, "Please re-install the NIC in a 32-bit slot "
771 		    "for proper operation.\n");
772 		if_printf(ifp, "Read the re(4) man page for more details.\n");
773 		error = EIO;
774 	}
775 
776 done:
777 	/* Turn interface off, release resources */
778 
779 	sc->re_testmode = 0;
780 	sc->re_link = 0;
781 	ifp->if_flags &= ~IFF_PROMISC;
782 	re_stop(sc);
783 	if (m0 != NULL)
784 		m_freem(m0);
785 
786 	return (error);
787 }
788 #endif	/* RE_DIAG */
789 
790 /*
791  * Probe for a RealTek 8139C+/8169/8110 chip. Check the PCI vendor and device
792  * IDs against our list and return a device name if we find a match.
793  */
794 static int
795 re_probe(device_t dev)
796 {
797 	const struct re_type *t;
798 	struct re_softc *sc;
799 	int rid;
800 	uint32_t hwrev;
801 	uint16_t vendor, product;
802 
803 	t = re_devs;
804 
805 	vendor = pci_get_vendor(dev);
806 	product = pci_get_device(dev);
807 
808 	/*
809 	 * Only attach to rev.3 of the Linksys EG1032 adapter.
810 	 * Rev.2 is supported by sk(4).
811 	 */
812 	if (vendor == PCI_VENDOR_LINKSYS &&
813 	    product == PCI_PRODUCT_LINKSYS_EG1032 &&
814 	    pci_get_subdevice(dev) != PCI_SUBDEVICE_LINKSYS_EG1032_REV3)
815 			return ENXIO;
816 
817 	for (t = re_devs; t->re_name != NULL; t++) {
818 		if (product == t->re_did && vendor == t->re_vid)
819 			break;
820 	}
821 
822 	/*
823 	 * Check if we found a RealTek device.
824 	 */
825 	if (t->re_name == NULL)
826 		return(ENXIO);
827 
828 	/*
829 	 * Temporarily map the I/O space so we can read the chip ID register.
830 	 */
831 	sc = kmalloc(sizeof(*sc), M_TEMP, M_WAITOK | M_ZERO);
832 	rid = RE_PCI_LOIO;
833 	sc->re_res = bus_alloc_resource_any(dev, SYS_RES_IOPORT, &rid,
834 					    RF_ACTIVE);
835 	if (sc->re_res == NULL) {
836 		device_printf(dev, "couldn't map ports/memory\n");
837 		kfree(sc, M_TEMP);
838 		return(ENXIO);
839 	}
840 
841 	sc->re_btag = rman_get_bustag(sc->re_res);
842 	sc->re_bhandle = rman_get_bushandle(sc->re_res);
843 
844 	hwrev = CSR_READ_4(sc, RE_TXCFG) & RE_TXCFG_HWREV;
845 	bus_release_resource(dev, SYS_RES_IOPORT, RE_PCI_LOIO, sc->re_res);
846 	kfree(sc, M_TEMP);
847 
848 	/*
849 	 * and continue matching for the specific chip...
850 	 */
851 	for (; t->re_name != NULL; t++) {
852 		if (product == t->re_did && vendor == t->re_vid &&
853 		    t->re_basetype == hwrev) {
854 			device_set_desc(dev, t->re_name);
855 			return(0);
856 		}
857 	}
858 	return(ENXIO);
859 }
860 
861 /*
862  * This routine takes the segment list provided as the result of
863  * a bus_dma_map_load() operation and assigns the addresses/lengths
864  * to RealTek DMA descriptors. This can be called either by the RX
865  * code or the TX code. In the RX case, we'll probably wind up mapping
866  * at most one segment. For the TX case, there could be any number of
867  * segments since TX packets may span multiple mbufs. In either case,
868  * if the number of segments is larger than the re_maxsegs limit
869  * specified by the caller, we abort the mapping operation. Sadly,
870  * whoever designed the buffer mapping API did not provide a way to
871  * return an error from here, so we have to fake it a bit.
872  */
873 
874 static void
875 re_dma_map_desc(void *arg, bus_dma_segment_t *segs, int nseg,
876 		bus_size_t mapsize, int error)
877 {
878 	struct re_dmaload_arg *ctx;
879 	struct re_desc *d = NULL;
880 	int i = 0, idx;
881 	uint32_t cmdstat;
882 
883 	if (error)
884 		return;
885 
886 	ctx = arg;
887 
888 	/* Signal error to caller if there's too many segments */
889 	if (nseg > ctx->re_maxsegs) {
890 		ctx->re_maxsegs = 0;
891 		return;
892 	}
893 
894 	/*
895 	 * Map the segment array into descriptors. Note that we set the
896 	 * start-of-frame and end-of-frame markers for either TX or RX, but
897 	 * they really only have meaning in the TX case. (In the RX case,
898 	 * it's the chip that tells us where packets begin and end.)
899 	 * We also keep track of the end of the ring and set the
900 	 * end-of-ring bits as needed, and we set the ownership bits
901 	 * in all except the very first descriptor. (The caller will
902 	 * set this descriptor later when it start transmission or
903 	 * reception.)
904 	 */
905 	idx = ctx->re_idx;
906 	for (;;) {
907 		d = &ctx->re_ring[idx];
908 		if (le32toh(d->re_cmdstat) & RE_RDESC_STAT_OWN) {
909 			ctx->re_maxsegs = 0;
910 			return;
911 		}
912 		cmdstat = segs[i].ds_len;
913 		d->re_bufaddr_lo = htole32(RE_ADDR_LO(segs[i].ds_addr));
914 		d->re_bufaddr_hi = htole32(RE_ADDR_HI(segs[i].ds_addr));
915 		if (i == 0)
916 			cmdstat |= RE_TDESC_CMD_SOF;
917 		else
918 			cmdstat |= RE_TDESC_CMD_OWN;
919 		if (idx == (RE_RX_DESC_CNT - 1))
920 			cmdstat |= RE_TDESC_CMD_EOR;
921 		d->re_cmdstat = htole32(cmdstat | ctx->re_flags);
922 		i++;
923 		if (i == nseg)
924 			break;
925 		RE_DESC_INC(idx);
926 	}
927 
928 	d->re_cmdstat |= htole32(RE_TDESC_CMD_EOF);
929 	ctx->re_maxsegs = nseg;
930 	ctx->re_idx = idx;
931 }
932 
933 /*
934  * Map a single buffer address.
935  */
936 
937 static void
938 re_dma_map_addr(void *arg, bus_dma_segment_t *segs, int nseg, int error)
939 {
940 	uint32_t *addr;
941 
942 	if (error)
943 		return;
944 
945 	KASSERT(nseg == 1, ("too many DMA segments, %d should be 1", nseg));
946 	addr = arg;
947 	*addr = segs->ds_addr;
948 }
949 
950 static int
951 re_allocmem(device_t dev, struct re_softc *sc)
952 {
953 	int error, i, nseg;
954 
955 	/*
956 	 * Allocate map for RX mbufs.
957 	 */
958 	nseg = 32;
959 	error = bus_dma_tag_create(sc->re_parent_tag, ETHER_ALIGN, 0,
960 	    BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
961 	    NULL, MCLBYTES * nseg, nseg, MCLBYTES, BUS_DMA_ALLOCNOW,
962 	    &sc->re_ldata.re_mtag);
963 	if (error) {
964 		device_printf(dev, "could not allocate dma tag\n");
965 		return(error);
966 	}
967 
968 	/*
969 	 * Allocate map for TX descriptor list.
970 	 */
971 	error = bus_dma_tag_create(sc->re_parent_tag, RE_RING_ALIGN,
972 	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
973             NULL, RE_TX_LIST_SZ, 1, RE_TX_LIST_SZ, BUS_DMA_ALLOCNOW,
974 	    &sc->re_ldata.re_tx_list_tag);
975 	if (error) {
976 		device_printf(dev, "could not allocate dma tag\n");
977 		return(error);
978 	}
979 
980 	/* Allocate DMA'able memory for the TX ring */
981 
982         error = bus_dmamem_alloc(sc->re_ldata.re_tx_list_tag,
983 	    (void **)&sc->re_ldata.re_tx_list, BUS_DMA_WAITOK | BUS_DMA_ZERO,
984             &sc->re_ldata.re_tx_list_map);
985         if (error) {
986 		device_printf(dev, "could not allocate TX ring\n");
987                 return(error);
988 	}
989 
990 	/* Load the map for the TX ring. */
991 
992 	error = bus_dmamap_load(sc->re_ldata.re_tx_list_tag,
993 	     sc->re_ldata.re_tx_list_map, sc->re_ldata.re_tx_list,
994 	     RE_TX_LIST_SZ, re_dma_map_addr,
995 	     &sc->re_ldata.re_tx_list_addr, BUS_DMA_NOWAIT);
996 	if (error) {
997 		device_printf(dev, "could not get addres of TX ring\n");
998 		return(error);
999 	}
1000 
1001 	/* Create DMA maps for TX buffers */
1002 
1003 	for (i = 0; i < RE_TX_DESC_CNT; i++) {
1004 		error = bus_dmamap_create(sc->re_ldata.re_mtag, 0,
1005 			    &sc->re_ldata.re_tx_dmamap[i]);
1006 		if (error) {
1007 			device_printf(dev, "can't create DMA map for TX\n");
1008 			return(error);
1009 		}
1010 	}
1011 
1012 	/*
1013 	 * Allocate map for RX descriptor list.
1014 	 */
1015 	error = bus_dma_tag_create(sc->re_parent_tag, RE_RING_ALIGN,
1016 	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1017             NULL, RE_RX_LIST_SZ, 1, RE_RX_LIST_SZ, BUS_DMA_ALLOCNOW,
1018 	    &sc->re_ldata.re_rx_list_tag);
1019 	if (error) {
1020 		device_printf(dev, "could not allocate dma tag\n");
1021 		return(error);
1022 	}
1023 
1024 	/* Allocate DMA'able memory for the RX ring */
1025 
1026         error = bus_dmamem_alloc(sc->re_ldata.re_rx_list_tag,
1027 	    (void **)&sc->re_ldata.re_rx_list, BUS_DMA_WAITOK | BUS_DMA_ZERO,
1028             &sc->re_ldata.re_rx_list_map);
1029         if (error) {
1030 		device_printf(dev, "could not allocate RX ring\n");
1031                 return(error);
1032 	}
1033 
1034 	/* Load the map for the RX ring. */
1035 
1036 	error = bus_dmamap_load(sc->re_ldata.re_rx_list_tag,
1037 	     sc->re_ldata.re_rx_list_map, sc->re_ldata.re_rx_list,
1038 	     RE_RX_LIST_SZ, re_dma_map_addr,
1039 	     &sc->re_ldata.re_rx_list_addr, BUS_DMA_NOWAIT);
1040 	if (error) {
1041 		device_printf(dev, "could not get address of RX ring\n");
1042 		return(error);
1043 	}
1044 
1045 	/* Create DMA maps for RX buffers */
1046 
1047 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
1048 		error = bus_dmamap_create(sc->re_ldata.re_mtag, 0,
1049 			    &sc->re_ldata.re_rx_dmamap[i]);
1050 		if (error) {
1051 			device_printf(dev, "can't create DMA map for RX\n");
1052 			return(ENOMEM);
1053 		}
1054 	}
1055 
1056 	return(0);
1057 }
1058 
1059 /*
1060  * Attach the interface. Allocate softc structures, do ifmedia
1061  * setup and ethernet/BPF attach.
1062  */
1063 static int
1064 re_attach(device_t dev)
1065 {
1066 	struct re_softc	*sc = device_get_softc(dev);
1067 	struct ifnet *ifp;
1068 	const struct re_hwrev *hw_rev;
1069 	uint8_t eaddr[ETHER_ADDR_LEN];
1070 	uint16_t as[ETHER_ADDR_LEN / 2];
1071 	uint16_t re_did = 0;
1072 	uint32_t hwrev;
1073 	int error = 0, rid, i;
1074 
1075 	callout_init(&sc->re_timer);
1076 #ifdef RE_DIAG
1077 	sc->re_dev = dev;
1078 #endif
1079 
1080 	RE_ENABLE_TX_MODERATION(sc);
1081 
1082 	sysctl_ctx_init(&sc->re_sysctl_ctx);
1083 	sc->re_sysctl_tree = SYSCTL_ADD_NODE(&sc->re_sysctl_ctx,
1084 					     SYSCTL_STATIC_CHILDREN(_hw),
1085 					     OID_AUTO,
1086 					     device_get_nameunit(dev),
1087 					     CTLFLAG_RD, 0, "");
1088 	if (sc->re_sysctl_tree == NULL) {
1089 		device_printf(dev, "can't add sysctl node\n");
1090 		error = ENXIO;
1091 		goto fail;
1092 	}
1093 	SYSCTL_ADD_PROC(&sc->re_sysctl_ctx,
1094 			SYSCTL_CHILDREN(sc->re_sysctl_tree),
1095 			OID_AUTO, "tx_moderation",
1096 			CTLTYPE_INT | CTLFLAG_RW,
1097 			sc, 0, re_sysctl_tx_moderation, "I",
1098 			"Enable/Disable TX moderation");
1099 
1100 #ifndef BURN_BRIDGES
1101 	/*
1102 	 * Handle power management nonsense.
1103 	 */
1104 
1105 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D0) {
1106 		uint32_t membase, irq;
1107 
1108 		/* Save important PCI config data. */
1109 		membase = pci_read_config(dev, RE_PCI_LOMEM, 4);
1110 		irq = pci_read_config(dev, PCIR_INTLINE, 4);
1111 
1112 		/* Reset the power state. */
1113 		device_printf(dev, "chip is in D%d power mode "
1114 		    "-- setting to D0\n", pci_get_powerstate(dev));
1115 
1116 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
1117 
1118 		/* Restore PCI config data. */
1119 		pci_write_config(dev, RE_PCI_LOMEM, membase, 4);
1120 		pci_write_config(dev, PCIR_INTLINE, irq, 4);
1121 	}
1122 #endif
1123 	/*
1124 	 * Map control/status registers.
1125 	 */
1126 	pci_enable_busmaster(dev);
1127 
1128 	rid = RE_PCI_LOIO;
1129 	sc->re_res = bus_alloc_resource_any(dev, SYS_RES_IOPORT, &rid,
1130 					    RF_ACTIVE);
1131 
1132 	if (sc->re_res == NULL) {
1133 		device_printf(dev, "couldn't map ports\n");
1134 		error = ENXIO;
1135 		goto fail;
1136 	}
1137 
1138 	sc->re_btag = rman_get_bustag(sc->re_res);
1139 	sc->re_bhandle = rman_get_bushandle(sc->re_res);
1140 
1141 	/* Allocate interrupt */
1142 	rid = 0;
1143 	sc->re_irq = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
1144 					    RF_SHAREABLE | RF_ACTIVE);
1145 
1146 	if (sc->re_irq == NULL) {
1147 		device_printf(dev, "couldn't map interrupt\n");
1148 		error = ENXIO;
1149 		goto fail;
1150 	}
1151 
1152 	/* Reset the adapter. */
1153 	re_reset(sc);
1154 
1155 	hwrev = CSR_READ_4(sc, RE_TXCFG) & RE_TXCFG_HWREV;
1156 	for (hw_rev = re_hwrevs; hw_rev->re_desc != NULL; hw_rev++) {
1157 		if (hw_rev->re_rev == hwrev) {
1158 			sc->re_type = hw_rev->re_type;
1159 			sc->re_flags = hw_rev->re_flags;
1160 			break;
1161 		}
1162 	}
1163 
1164 	sc->re_eewidth = 6;
1165 	re_read_eeprom(sc, (caddr_t)&re_did, 0, 1);
1166 	if (re_did != 0x8129)
1167 	        sc->re_eewidth = 8;
1168 
1169 	/*
1170 	 * Get station address from the EEPROM.
1171 	 */
1172 	re_read_eeprom(sc, (caddr_t)as, RE_EE_EADDR, 3);
1173 	for (i = 0; i < ETHER_ADDR_LEN / 2; i++)
1174 		as[i] = le16toh(as[i]);
1175 	bcopy(as, eaddr, sizeof(eaddr));
1176 
1177 	if (sc->re_type == RE_8169) {
1178 		/* Set RX length mask */
1179 		sc->re_rxlenmask = RE_RDESC_STAT_GFRAGLEN;
1180 		sc->re_txstart = RE_GTXSTART;
1181 	} else {
1182 		/* Set RX length mask */
1183 		sc->re_rxlenmask = RE_RDESC_STAT_FRAGLEN;
1184 		sc->re_txstart = RE_TXSTART;
1185 	}
1186 
1187 	/*
1188 	 * Allocate the parent bus DMA tag appropriate for PCI.
1189 	 */
1190 #define RE_NSEG_NEW 32
1191 	error = bus_dma_tag_create(NULL,	/* parent */
1192 			1, 0,			/* alignment, boundary */
1193 			BUS_SPACE_MAXADDR_32BIT,/* lowaddr */
1194 			BUS_SPACE_MAXADDR,	/* highaddr */
1195 			NULL, NULL,		/* filter, filterarg */
1196 			MAXBSIZE, RE_NSEG_NEW,	/* maxsize, nsegments */
1197 			BUS_SPACE_MAXSIZE_32BIT,/* maxsegsize */
1198 			BUS_DMA_ALLOCNOW,	/* flags */
1199 			&sc->re_parent_tag);
1200 	if (error)
1201 		goto fail;
1202 
1203 	error = re_allocmem(dev, sc);
1204 
1205 	if (error)
1206 		goto fail;
1207 
1208 	/* Do MII setup */
1209 	if (mii_phy_probe(dev, &sc->re_miibus,
1210 	    re_ifmedia_upd, re_ifmedia_sts)) {
1211 		device_printf(dev, "MII without any phy!\n");
1212 		error = ENXIO;
1213 		goto fail;
1214 	}
1215 
1216 	ifp = &sc->arpcom.ac_if;
1217 	ifp->if_softc = sc;
1218 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1219 	ifp->if_mtu = ETHERMTU;
1220 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1221 	ifp->if_ioctl = re_ioctl;
1222 	ifp->if_capabilities = IFCAP_VLAN_MTU;
1223 	ifp->if_start = re_start;
1224 	ifp->if_capabilities |= IFCAP_HWCSUM|IFCAP_VLAN_HWTAGGING;
1225 #ifdef DEVICE_POLLING
1226 	ifp->if_poll = re_poll;
1227 #endif
1228 	ifp->if_watchdog = re_watchdog;
1229 	ifp->if_init = re_init;
1230 	if (sc->re_type == RE_8169)
1231 		ifp->if_baudrate = 1000000000;
1232 	else
1233 		ifp->if_baudrate = 100000000;
1234 	ifq_set_maxlen(&ifp->if_snd, RE_IFQ_MAXLEN);
1235 	ifq_set_ready(&ifp->if_snd);
1236 
1237 #ifdef RE_DISABLE_HWCSUM
1238 	ifp->if_capenable = ifp->if_capabilities & ~IFCAP_HWCSUM;
1239 	ifp->if_hwassist = 0;
1240 #else
1241 	ifp->if_capenable = ifp->if_capabilities;
1242 	ifp->if_hwassist = RE_CSUM_FEATURES;
1243 #endif	/* RE_DISABLE_HWCSUM */
1244 
1245 	/*
1246 	 * Call MI attach routine.
1247 	 */
1248 	ether_ifattach(ifp, eaddr, NULL);
1249 
1250 #ifdef RE_DIAG
1251 	/*
1252 	 * Perform hardware diagnostic on the original RTL8169.
1253 	 * Some 32-bit cards were incorrectly wired and would
1254 	 * malfunction if plugged into a 64-bit slot.
1255 	 */
1256 	if (hwrev == RE_HWREV_8169) {
1257 		lwkt_serialize_enter(ifp->if_serializer);
1258 		error = re_diag(sc);
1259 		lwkt_serialize_exit(ifp->if_serializer);
1260 
1261 		if (error) {
1262 			device_printf(dev, "hardware diagnostic failure\n");
1263 			ether_ifdetach(ifp);
1264 			goto fail;
1265 		}
1266 	}
1267 #endif	/* RE_DIAG */
1268 
1269 	/* Hook interrupt last to avoid having to lock softc */
1270 	error = bus_setup_intr(dev, sc->re_irq, INTR_NETSAFE, re_intr, sc,
1271 			       &sc->re_intrhand, ifp->if_serializer);
1272 
1273 	if (error) {
1274 		device_printf(dev, "couldn't set up irq\n");
1275 		ether_ifdetach(ifp);
1276 		goto fail;
1277 	}
1278 
1279 fail:
1280 	if (error)
1281 		re_detach(dev);
1282 
1283 	return (error);
1284 }
1285 
1286 /*
1287  * Shutdown hardware and free up resources. This can be called any
1288  * time after the mutex has been initialized. It is called in both
1289  * the error case in attach and the normal detach case so it needs
1290  * to be careful about only freeing resources that have actually been
1291  * allocated.
1292  */
1293 static int
1294 re_detach(device_t dev)
1295 {
1296 	struct re_softc *sc = device_get_softc(dev);
1297 	struct ifnet *ifp = &sc->arpcom.ac_if;
1298 	int i;
1299 
1300 	/* These should only be active if attach succeeded */
1301 	if (device_is_attached(dev)) {
1302 		lwkt_serialize_enter(ifp->if_serializer);
1303 		re_stop(sc);
1304 		bus_teardown_intr(dev, sc->re_irq, sc->re_intrhand);
1305 		lwkt_serialize_exit(ifp->if_serializer);
1306 
1307 		ether_ifdetach(ifp);
1308 	}
1309 	if (sc->re_miibus)
1310 		device_delete_child(dev, sc->re_miibus);
1311 	bus_generic_detach(dev);
1312 
1313 	if (sc->re_irq)
1314 		bus_release_resource(dev, SYS_RES_IRQ, 0, sc->re_irq);
1315 	if (sc->re_res) {
1316 		bus_release_resource(dev, SYS_RES_IOPORT, RE_PCI_LOIO,
1317 				     sc->re_res);
1318 	}
1319 
1320 	/* Unload and free the RX DMA ring memory and map */
1321 
1322 	if (sc->re_ldata.re_rx_list_tag) {
1323 		bus_dmamap_unload(sc->re_ldata.re_rx_list_tag,
1324 		    sc->re_ldata.re_rx_list_map);
1325 		bus_dmamem_free(sc->re_ldata.re_rx_list_tag,
1326 		    sc->re_ldata.re_rx_list,
1327 		    sc->re_ldata.re_rx_list_map);
1328 		bus_dma_tag_destroy(sc->re_ldata.re_rx_list_tag);
1329 	}
1330 
1331 	/* Unload and free the TX DMA ring memory and map */
1332 
1333 	if (sc->re_ldata.re_tx_list_tag) {
1334 		bus_dmamap_unload(sc->re_ldata.re_tx_list_tag,
1335 		    sc->re_ldata.re_tx_list_map);
1336 		bus_dmamem_free(sc->re_ldata.re_tx_list_tag,
1337 		    sc->re_ldata.re_tx_list,
1338 		    sc->re_ldata.re_tx_list_map);
1339 		bus_dma_tag_destroy(sc->re_ldata.re_tx_list_tag);
1340 	}
1341 
1342 	/* Destroy all the RX and TX buffer maps */
1343 
1344 	if (sc->re_ldata.re_mtag) {
1345 		for (i = 0; i < RE_TX_DESC_CNT; i++)
1346 			bus_dmamap_destroy(sc->re_ldata.re_mtag,
1347 			    sc->re_ldata.re_tx_dmamap[i]);
1348 		for (i = 0; i < RE_RX_DESC_CNT; i++)
1349 			bus_dmamap_destroy(sc->re_ldata.re_mtag,
1350 			    sc->re_ldata.re_rx_dmamap[i]);
1351 		bus_dma_tag_destroy(sc->re_ldata.re_mtag);
1352 	}
1353 
1354 	/* Unload and free the stats buffer and map */
1355 
1356 	if (sc->re_ldata.re_stag) {
1357 		bus_dmamap_unload(sc->re_ldata.re_stag,
1358 		    sc->re_ldata.re_rx_list_map);
1359 		bus_dmamem_free(sc->re_ldata.re_stag,
1360 		    sc->re_ldata.re_stats,
1361 		    sc->re_ldata.re_smap);
1362 		bus_dma_tag_destroy(sc->re_ldata.re_stag);
1363 	}
1364 
1365 	if (sc->re_parent_tag)
1366 		bus_dma_tag_destroy(sc->re_parent_tag);
1367 
1368 	return(0);
1369 }
1370 
1371 static int
1372 re_newbuf(struct re_softc *sc, int idx, struct mbuf *m)
1373 {
1374 	struct re_dmaload_arg arg;
1375 	struct mbuf *n = NULL;
1376 	int error;
1377 
1378 	if (m == NULL) {
1379 		n = m_getcl(MB_DONTWAIT, MT_DATA, M_PKTHDR);
1380 		if (n == NULL)
1381 			return(ENOBUFS);
1382 		m = n;
1383 	} else
1384 		m->m_data = m->m_ext.ext_buf;
1385 
1386 	m->m_len = m->m_pkthdr.len = MCLBYTES;
1387 
1388 	/*
1389 	 * NOTE:
1390 	 * Some re(4) chips(e.g. RTL8101E) need address of the receive buffer
1391 	 * to be 8-byte aligned, so don't call m_adj(m, ETHER_ALIGN) here.
1392 	 */
1393 
1394 	arg.sc = sc;
1395 	arg.re_idx = idx;
1396 	arg.re_maxsegs = 1;
1397 	arg.re_flags = 0;
1398 	arg.re_ring = sc->re_ldata.re_rx_list;
1399 
1400         error = bus_dmamap_load_mbuf(sc->re_ldata.re_mtag,
1401 	    sc->re_ldata.re_rx_dmamap[idx], m, re_dma_map_desc,
1402 	    &arg, BUS_DMA_NOWAIT);
1403 	if (error || arg.re_maxsegs != 1) {
1404 		if (n != NULL)
1405 			m_freem(n);
1406 		return (ENOMEM);
1407 	}
1408 
1409 	sc->re_ldata.re_rx_list[idx].re_cmdstat |= htole32(RE_RDESC_CMD_OWN);
1410 	sc->re_ldata.re_rx_mbuf[idx] = m;
1411 
1412         bus_dmamap_sync(sc->re_ldata.re_mtag, sc->re_ldata.re_rx_dmamap[idx],
1413 		        BUS_DMASYNC_PREREAD);
1414 
1415 	return(0);
1416 }
1417 
1418 static int
1419 re_tx_list_init(struct re_softc *sc)
1420 {
1421 	bzero(sc->re_ldata.re_tx_list, RE_TX_LIST_SZ);
1422 	bzero(&sc->re_ldata.re_tx_mbuf, RE_TX_DESC_CNT * sizeof(struct mbuf *));
1423 
1424 	bus_dmamap_sync(sc->re_ldata.re_tx_list_tag,
1425 			sc->re_ldata.re_tx_list_map, BUS_DMASYNC_PREWRITE);
1426 	sc->re_ldata.re_tx_prodidx = 0;
1427 	sc->re_ldata.re_tx_considx = 0;
1428 	sc->re_ldata.re_tx_free = RE_TX_DESC_CNT;
1429 
1430 	return(0);
1431 }
1432 
1433 static int
1434 re_rx_list_init(struct re_softc *sc)
1435 {
1436 	int i, error;
1437 
1438 	bzero(sc->re_ldata.re_rx_list, RE_RX_LIST_SZ);
1439 	bzero(&sc->re_ldata.re_rx_mbuf, RE_RX_DESC_CNT * sizeof(struct mbuf *));
1440 
1441 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
1442 		error = re_newbuf(sc, i, NULL);
1443 		if (error)
1444 			return(error);
1445 	}
1446 
1447 	/* Flush the RX descriptors */
1448 
1449 	bus_dmamap_sync(sc->re_ldata.re_rx_list_tag,
1450 			sc->re_ldata.re_rx_list_map, BUS_DMASYNC_PREWRITE);
1451 
1452 	sc->re_ldata.re_rx_prodidx = 0;
1453 	sc->re_head = sc->re_tail = NULL;
1454 
1455 	return(0);
1456 }
1457 
1458 /*
1459  * RX handler for C+ and 8169. For the gigE chips, we support
1460  * the reception of jumbo frames that have been fragmented
1461  * across multiple 2K mbuf cluster buffers.
1462  */
1463 static void
1464 re_rxeof(struct re_softc *sc)
1465 {
1466 	struct ifnet *ifp = &sc->arpcom.ac_if;
1467 	struct mbuf *m;
1468 	struct re_desc 	*cur_rx;
1469 	uint32_t rxstat, rxvlan;
1470 	int i, total_len;
1471 
1472 	/* Invalidate the descriptor memory */
1473 
1474 	bus_dmamap_sync(sc->re_ldata.re_rx_list_tag,
1475 			sc->re_ldata.re_rx_list_map, BUS_DMASYNC_POSTREAD);
1476 
1477 	for (i = sc->re_ldata.re_rx_prodidx;
1478 	     RE_OWN(&sc->re_ldata.re_rx_list[i]) == 0 ; RE_DESC_INC(i)) {
1479 		cur_rx = &sc->re_ldata.re_rx_list[i];
1480 		m = sc->re_ldata.re_rx_mbuf[i];
1481 		total_len = RE_RXBYTES(cur_rx);
1482 		rxstat = le32toh(cur_rx->re_cmdstat);
1483 		rxvlan = le32toh(cur_rx->re_vlanctl);
1484 
1485 		/* Invalidate the RX mbuf and unload its map */
1486 
1487 		bus_dmamap_sync(sc->re_ldata.re_mtag,
1488 				sc->re_ldata.re_rx_dmamap[i],
1489 				BUS_DMASYNC_POSTWRITE);
1490 		bus_dmamap_unload(sc->re_ldata.re_mtag,
1491 				  sc->re_ldata.re_rx_dmamap[i]);
1492 
1493 		if ((rxstat & RE_RDESC_STAT_EOF) == 0) {
1494 			m->m_len = MCLBYTES - ETHER_ALIGN;
1495 			if (sc->re_head == NULL) {
1496 				sc->re_head = sc->re_tail = m;
1497 			} else {
1498 				sc->re_tail->m_next = m;
1499 				sc->re_tail = m;
1500 			}
1501 			re_newbuf(sc, i, NULL);
1502 			continue;
1503 		}
1504 
1505 		/*
1506 		 * NOTE: for the 8139C+, the frame length field
1507 		 * is always 12 bits in size, but for the gigE chips,
1508 		 * it is 13 bits (since the max RX frame length is 16K).
1509 		 * Unfortunately, all 32 bits in the status word
1510 		 * were already used, so to make room for the extra
1511 		 * length bit, RealTek took out the 'frame alignment
1512 		 * error' bit and shifted the other status bits
1513 		 * over one slot. The OWN, EOR, FS and LS bits are
1514 		 * still in the same places. We have already extracted
1515 		 * the frame length and checked the OWN bit, so rather
1516 		 * than using an alternate bit mapping, we shift the
1517 		 * status bits one space to the right so we can evaluate
1518 		 * them using the 8169 status as though it was in the
1519 		 * same format as that of the 8139C+.
1520 		 */
1521 		if (sc->re_type == RE_8169)
1522 			rxstat >>= 1;
1523 
1524 		if (rxstat & RE_RDESC_STAT_RXERRSUM) {
1525 			ifp->if_ierrors++;
1526 			/*
1527 			 * If this is part of a multi-fragment packet,
1528 			 * discard all the pieces.
1529 			 */
1530 			if (sc->re_head != NULL) {
1531 				m_freem(sc->re_head);
1532 				sc->re_head = sc->re_tail = NULL;
1533 			}
1534 			re_newbuf(sc, i, m);
1535 			continue;
1536 		}
1537 
1538 		/*
1539 		 * If allocating a replacement mbuf fails,
1540 		 * reload the current one.
1541 		 */
1542 
1543 		if (re_newbuf(sc, i, NULL)) {
1544 			ifp->if_ierrors++;
1545 			if (sc->re_head != NULL) {
1546 				m_freem(sc->re_head);
1547 				sc->re_head = sc->re_tail = NULL;
1548 			}
1549 			re_newbuf(sc, i, m);
1550 			continue;
1551 		}
1552 
1553 		if (sc->re_head != NULL) {
1554 			m->m_len = total_len % (MCLBYTES - ETHER_ALIGN);
1555 			/*
1556 			 * Special case: if there's 4 bytes or less
1557 			 * in this buffer, the mbuf can be discarded:
1558 			 * the last 4 bytes is the CRC, which we don't
1559 			 * care about anyway.
1560 			 */
1561 			if (m->m_len <= ETHER_CRC_LEN) {
1562 				sc->re_tail->m_len -=
1563 				    (ETHER_CRC_LEN - m->m_len);
1564 				m_freem(m);
1565 			} else {
1566 				m->m_len -= ETHER_CRC_LEN;
1567 				sc->re_tail->m_next = m;
1568 			}
1569 			m = sc->re_head;
1570 			sc->re_head = sc->re_tail = NULL;
1571 			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
1572 		} else
1573 			m->m_pkthdr.len = m->m_len =
1574 			    (total_len - ETHER_CRC_LEN);
1575 
1576 		ifp->if_ipackets++;
1577 		m->m_pkthdr.rcvif = ifp;
1578 
1579 		/* Do RX checksumming if enabled */
1580 
1581 		if (ifp->if_capenable & IFCAP_RXCSUM) {
1582 
1583 			/* Check IP header checksum */
1584 			if (rxstat & RE_RDESC_STAT_PROTOID)
1585 				m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED;
1586 			if ((rxstat & RE_RDESC_STAT_IPSUMBAD) == 0)
1587 				m->m_pkthdr.csum_flags |= CSUM_IP_VALID;
1588 
1589 			/* Check TCP/UDP checksum */
1590 			if ((RE_TCPPKT(rxstat) &&
1591 			    (rxstat & RE_RDESC_STAT_TCPSUMBAD) == 0) ||
1592 			    (RE_UDPPKT(rxstat) &&
1593 			    (rxstat & RE_RDESC_STAT_UDPSUMBAD)) == 0) {
1594 				m->m_pkthdr.csum_flags |=
1595 				    CSUM_DATA_VALID|CSUM_PSEUDO_HDR|
1596 				    CSUM_FRAG_NOT_CHECKED;
1597 				m->m_pkthdr.csum_data = 0xffff;
1598 			}
1599 		}
1600 
1601 		if (rxvlan & RE_RDESC_VLANCTL_TAG) {
1602 			VLAN_INPUT_TAG(m,
1603 			   be16toh((rxvlan & RE_RDESC_VLANCTL_DATA)));
1604 		} else {
1605 			ifp->if_input(ifp, m);
1606 		}
1607 	}
1608 
1609 	/* Flush the RX DMA ring */
1610 
1611 	bus_dmamap_sync(sc->re_ldata.re_rx_list_tag,
1612 			sc->re_ldata.re_rx_list_map, BUS_DMASYNC_PREWRITE);
1613 
1614 	sc->re_ldata.re_rx_prodidx = i;
1615 }
1616 
1617 static void
1618 re_txeof(struct re_softc *sc)
1619 {
1620 	struct ifnet *ifp = &sc->arpcom.ac_if;
1621 	uint32_t txstat;
1622 	int idx;
1623 
1624 	/* Invalidate the TX descriptor list */
1625 
1626 	bus_dmamap_sync(sc->re_ldata.re_tx_list_tag,
1627 			sc->re_ldata.re_tx_list_map, BUS_DMASYNC_POSTREAD);
1628 
1629 	for (idx = sc->re_ldata.re_tx_considx;
1630 	     sc->re_ldata.re_tx_free < RE_TX_DESC_CNT; RE_DESC_INC(idx)) {
1631 		txstat = le32toh(sc->re_ldata.re_tx_list[idx].re_cmdstat);
1632 		if (txstat & RE_TDESC_CMD_OWN)
1633 			break;
1634 
1635 		sc->re_ldata.re_tx_list[idx].re_bufaddr_lo = 0;
1636 
1637 		/*
1638 		 * We only stash mbufs in the last descriptor
1639 		 * in a fragment chain, which also happens to
1640 		 * be the only place where the TX status bits
1641 		 * are valid.
1642 		 */
1643 		if (txstat & RE_TDESC_CMD_EOF) {
1644 			m_freem(sc->re_ldata.re_tx_mbuf[idx]);
1645 			sc->re_ldata.re_tx_mbuf[idx] = NULL;
1646 			bus_dmamap_unload(sc->re_ldata.re_mtag,
1647 			    sc->re_ldata.re_tx_dmamap[idx]);
1648 			if (txstat & (RE_TDESC_STAT_EXCESSCOL|
1649 			    RE_TDESC_STAT_COLCNT))
1650 				ifp->if_collisions++;
1651 			if (txstat & RE_TDESC_STAT_TXERRSUM)
1652 				ifp->if_oerrors++;
1653 			else
1654 				ifp->if_opackets++;
1655 		}
1656 		sc->re_ldata.re_tx_free++;
1657 	}
1658 
1659 	/* No changes made to the TX ring, so no flush needed */
1660 	if (sc->re_ldata.re_tx_free) {
1661 		sc->re_ldata.re_tx_considx = idx;
1662 		ifp->if_flags &= ~IFF_OACTIVE;
1663 		ifp->if_timer = 0;
1664 	}
1665 
1666 	/*
1667 	 * Some chips will ignore a second TX request issued while an
1668 	 * existing transmission is in progress. If the transmitter goes
1669 	 * idle but there are still packets waiting to be sent, we need
1670 	 * to restart the channel here to flush them out. This only seems
1671 	 * to be required with the PCIe devices.
1672 	 */
1673 	if (sc->re_ldata.re_tx_free < RE_TX_DESC_CNT)
1674 		CSR_WRITE_1(sc, sc->re_txstart, RE_TXSTART_START);
1675 
1676 	/*
1677 	 * If not all descriptors have been released reaped yet,
1678 	 * reload the timer so that we will eventually get another
1679 	 * interrupt that will cause us to re-enter this routine.
1680 	 * This is done in case the transmitter has gone idle.
1681 	 */
1682 	if (RE_TX_MODERATION_IS_ENABLED(sc) &&
1683 	    sc->re_ldata.re_tx_free < RE_TX_DESC_CNT)
1684                 CSR_WRITE_4(sc, RE_TIMERCNT, 1);
1685 }
1686 
1687 static void
1688 re_tick(void *xsc)
1689 {
1690 	struct re_softc *sc = xsc;
1691 
1692 	lwkt_serialize_enter(sc->arpcom.ac_if.if_serializer);
1693 	re_tick_serialized(xsc);
1694 	lwkt_serialize_exit(sc->arpcom.ac_if.if_serializer);
1695 }
1696 
1697 static void
1698 re_tick_serialized(void *xsc)
1699 {
1700 	struct re_softc *sc = xsc;
1701 	struct ifnet *ifp = &sc->arpcom.ac_if;
1702 	struct mii_data *mii;
1703 
1704 	mii = device_get_softc(sc->re_miibus);
1705 	mii_tick(mii);
1706 	if (sc->re_link) {
1707 		if (!(mii->mii_media_status & IFM_ACTIVE))
1708 			sc->re_link = 0;
1709 	} else {
1710 		if (mii->mii_media_status & IFM_ACTIVE &&
1711 		    IFM_SUBTYPE(mii->mii_media_active) != IFM_NONE) {
1712 			sc->re_link = 1;
1713 			if (!ifq_is_empty(&ifp->if_snd))
1714 				ifp->if_start(ifp);
1715 		}
1716 	}
1717 
1718 	callout_reset(&sc->re_timer, hz, re_tick, sc);
1719 }
1720 
1721 #ifdef DEVICE_POLLING
1722 
1723 static void
1724 re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
1725 {
1726 	struct re_softc *sc = ifp->if_softc;
1727 
1728 	switch(cmd) {
1729 	case POLL_REGISTER:
1730 		/* disable interrupts */
1731 		CSR_WRITE_2(sc, RE_IMR, 0x0000);
1732 		break;
1733 	case POLL_DEREGISTER:
1734 		/* enable interrupts */
1735 		CSR_WRITE_2(sc, RE_IMR, sc->re_intrs);
1736 		break;
1737 	default:
1738 		sc->rxcycles = count;
1739 		re_rxeof(sc);
1740 		re_txeof(sc);
1741 
1742 		if (!ifq_is_empty(&ifp->if_snd))
1743 			(*ifp->if_start)(ifp);
1744 
1745 		if (cmd == POLL_AND_CHECK_STATUS) { /* also check status register */
1746 			uint16_t       status;
1747 
1748 			status = CSR_READ_2(sc, RE_ISR);
1749 			if (status == 0xffff)
1750 				return;
1751 			if (status)
1752 				CSR_WRITE_2(sc, RE_ISR, status);
1753 
1754 			/*
1755 			 * XXX check behaviour on receiver stalls.
1756 			 */
1757 
1758 			if (status & RE_ISR_SYSTEM_ERR) {
1759 				re_reset(sc);
1760 				re_init(sc);
1761 			}
1762 		}
1763 		break;
1764 	}
1765 }
1766 #endif /* DEVICE_POLLING */
1767 
1768 static void
1769 re_intr(void *arg)
1770 {
1771 	struct re_softc	*sc = arg;
1772 	struct ifnet *ifp = &sc->arpcom.ac_if;
1773 	uint16_t status;
1774 
1775 	if (sc->suspended || (ifp->if_flags & IFF_UP) == 0)
1776 		return;
1777 
1778 	for (;;) {
1779 		status = CSR_READ_2(sc, RE_ISR);
1780 		/* If the card has gone away the read returns 0xffff. */
1781 		if (status == 0xffff)
1782 			break;
1783 		if (status)
1784 			CSR_WRITE_2(sc, RE_ISR, status);
1785 
1786 		if ((status & sc->re_intrs) == 0)
1787 			break;
1788 
1789 		if (status & (RE_ISR_RX_OK | RE_ISR_RX_ERR | RE_ISR_FIFO_OFLOW))
1790 			re_rxeof(sc);
1791 
1792 		if ((status & sc->re_tx_ack) ||
1793 		    (status & RE_ISR_TX_ERR) ||
1794 		    (status & RE_ISR_TX_DESC_UNAVAIL))
1795 			re_txeof(sc);
1796 
1797 		if (status & RE_ISR_SYSTEM_ERR) {
1798 			re_reset(sc);
1799 			re_init(sc);
1800 		}
1801 
1802 		if (status & RE_ISR_LINKCHG) {
1803 			callout_stop(&sc->re_timer);
1804 			re_tick_serialized(sc);
1805 		}
1806 	}
1807 
1808 	if (!ifq_is_empty(&ifp->if_snd))
1809 		(*ifp->if_start)(ifp);
1810 }
1811 
1812 static int
1813 re_encap(struct re_softc *sc, struct mbuf **m_head, int *idx, int *called_defrag)
1814 {
1815 	struct ifnet *ifp = &sc->arpcom.ac_if;
1816 	struct mbuf *m, *m_new = NULL;
1817 	struct re_dmaload_arg	arg;
1818 	bus_dmamap_t		map;
1819 	int			error;
1820 
1821 	*called_defrag = 0;
1822 	if (sc->re_ldata.re_tx_free <= 4)
1823 		return(EFBIG);
1824 
1825 	m = *m_head;
1826 
1827 	/*
1828 	 * Set up checksum offload. Note: checksum offload bits must
1829 	 * appear in all descriptors of a multi-descriptor transmit
1830 	 * attempt. (This is according to testing done with an 8169
1831 	 * chip. I'm not sure if this is a requirement or a bug.)
1832 	 */
1833 
1834 	arg.re_flags = 0;
1835 
1836 	if (m->m_pkthdr.csum_flags & CSUM_IP)
1837 		arg.re_flags |= RE_TDESC_CMD_IPCSUM;
1838 	if (m->m_pkthdr.csum_flags & CSUM_TCP)
1839 		arg.re_flags |= RE_TDESC_CMD_TCPCSUM;
1840 	if (m->m_pkthdr.csum_flags & CSUM_UDP)
1841 		arg.re_flags |= RE_TDESC_CMD_UDPCSUM;
1842 
1843 	arg.sc = sc;
1844 	arg.re_idx = *idx;
1845 	arg.re_maxsegs = sc->re_ldata.re_tx_free;
1846 	if (arg.re_maxsegs > 4)
1847 		arg.re_maxsegs -= 4;
1848 	arg.re_ring = sc->re_ldata.re_tx_list;
1849 
1850 	map = sc->re_ldata.re_tx_dmamap[*idx];
1851 
1852 	/*
1853 	 * With some of the RealTek chips, using the checksum offload
1854 	 * support in conjunction with the autopadding feature results
1855 	 * in the transmission of corrupt frames. For example, if we
1856 	 * need to send a really small IP fragment that's less than 60
1857 	 * bytes in size, and IP header checksumming is enabled, the
1858 	 * resulting ethernet frame that appears on the wire will
1859 	 * have garbled payload. To work around this, if TX checksum
1860 	 * offload is enabled, we always manually pad short frames out
1861 	 * to the minimum ethernet frame size. We do this by pretending
1862 	 * the mbuf chain has too many fragments so the coalescing code
1863 	 * below can assemble the packet into a single buffer that's
1864 	 * padded out to the mininum frame size.
1865 	 *
1866 	 * Note: this appears unnecessary for TCP, and doing it for TCP
1867 	 * with PCIe adapters seems to result in bad checksums.
1868 	 */
1869 	if (arg.re_flags && !(arg.re_flags & RE_TDESC_CMD_TCPCSUM) &&
1870 	    m->m_pkthdr.len < RE_MIN_FRAMELEN) {
1871 		error = EFBIG;
1872 	} else {
1873 		error = bus_dmamap_load_mbuf(sc->re_ldata.re_mtag, map,
1874 		    m, re_dma_map_desc, &arg, BUS_DMA_NOWAIT);
1875 	}
1876 
1877 	if (error && error != EFBIG) {
1878 		if_printf(ifp, "can't map mbuf (error %d)\n", error);
1879 		return(ENOBUFS);
1880 	}
1881 
1882 	/* Too many segments to map, coalesce into a single mbuf */
1883 
1884 	if (error || arg.re_maxsegs == 0) {
1885 		m_new = m_defrag_nofree(m, MB_DONTWAIT);
1886 		if (m_new == NULL) {
1887 			return(1);
1888 		} else {
1889 			m = m_new;
1890 			*m_head = m;
1891 		}
1892 
1893 		/*
1894 		 * Manually pad short frames, and zero the pad space
1895 		 * to avoid leaking data.
1896 		 */
1897 		if (m_new->m_pkthdr.len < RE_MIN_FRAMELEN) {
1898 			bzero(mtod(m_new, char *) + m_new->m_pkthdr.len,
1899 			    RE_MIN_FRAMELEN - m_new->m_pkthdr.len);
1900 			m_new->m_pkthdr.len += RE_MIN_FRAMELEN -
1901 			    m_new->m_pkthdr.len;
1902 			m_new->m_len = m_new->m_pkthdr.len;
1903 		}
1904 
1905 		*called_defrag = 1;
1906 		arg.sc = sc;
1907 		arg.re_idx = *idx;
1908 		arg.re_maxsegs = sc->re_ldata.re_tx_free;
1909 		arg.re_ring = sc->re_ldata.re_tx_list;
1910 
1911 		error = bus_dmamap_load_mbuf(sc->re_ldata.re_mtag, map,
1912 		    m, re_dma_map_desc, &arg, BUS_DMA_NOWAIT);
1913 		if (error) {
1914 			m_freem(m);
1915 			if_printf(ifp, "can't map mbuf (error %d)\n", error);
1916 			return(EFBIG);
1917 		}
1918 	}
1919 
1920 	/*
1921 	 * Insure that the map for this transmission
1922 	 * is placed at the array index of the last descriptor
1923 	 * in this chain.
1924 	 */
1925 	sc->re_ldata.re_tx_dmamap[*idx] =
1926 	    sc->re_ldata.re_tx_dmamap[arg.re_idx];
1927 	sc->re_ldata.re_tx_dmamap[arg.re_idx] = map;
1928 
1929 	sc->re_ldata.re_tx_mbuf[arg.re_idx] = m;
1930 	sc->re_ldata.re_tx_free -= arg.re_maxsegs;
1931 
1932 	/*
1933 	 * Set up hardware VLAN tagging. Note: vlan tag info must
1934 	 * appear in the first descriptor of a multi-descriptor
1935 	 * transmission attempt.
1936 	 */
1937 
1938 	if (m->m_flags & M_VLANTAG) {
1939 		sc->re_ldata.re_tx_list[*idx].re_vlanctl =
1940 		    htole32(htobe16(m->m_pkthdr.ether_vlantag) |
1941 		    	    RE_TDESC_VLANCTL_TAG);
1942 	}
1943 
1944 	/* Transfer ownership of packet to the chip. */
1945 
1946 	sc->re_ldata.re_tx_list[arg.re_idx].re_cmdstat |=
1947 	    htole32(RE_TDESC_CMD_OWN);
1948 	if (*idx != arg.re_idx)
1949 		sc->re_ldata.re_tx_list[*idx].re_cmdstat |=
1950 		    htole32(RE_TDESC_CMD_OWN);
1951 
1952 	RE_DESC_INC(arg.re_idx);
1953 	*idx = arg.re_idx;
1954 
1955 	return(0);
1956 }
1957 
1958 /*
1959  * Main transmit routine for C+ and gigE NICs.
1960  */
1961 
1962 static void
1963 re_start(struct ifnet *ifp)
1964 {
1965 	struct re_softc	*sc = ifp->if_softc;
1966 	struct mbuf *m_head;
1967 	struct mbuf *m_head2;
1968 	int called_defrag, idx, need_trans;
1969 
1970 	if (!sc->re_link || (ifp->if_flags & IFF_OACTIVE))
1971 		return;
1972 
1973 	idx = sc->re_ldata.re_tx_prodidx;
1974 
1975 	need_trans = 0;
1976 	while (sc->re_ldata.re_tx_mbuf[idx] == NULL) {
1977 		m_head = ifq_poll(&ifp->if_snd);
1978 		if (m_head == NULL)
1979 			break;
1980 		m_head2 = m_head;
1981 		if (re_encap(sc, &m_head2, &idx, &called_defrag)) {
1982 			/*
1983 			 * If we could not encapsulate the defragged packet,
1984 			 * the returned m_head2 is garbage and we must dequeue
1985 			 * and throw away the original packet.
1986 			 */
1987 			if (called_defrag) {
1988 				ifq_dequeue(&ifp->if_snd, m_head);
1989 				m_freem(m_head);
1990 			}
1991 			ifp->if_flags |= IFF_OACTIVE;
1992 			break;
1993 		}
1994 
1995 		/*
1996 		 * Clean out the packet we encapsulated.  If we defragged
1997 		 * the packet the m_head2 is the one that got encapsulated
1998 		 * and the original must be thrown away.  Otherwise m_head2
1999 		 * *IS* the original.
2000 		 */
2001 		ifq_dequeue(&ifp->if_snd, m_head);
2002 		if (called_defrag)
2003 			m_freem(m_head);
2004 		need_trans = 1;
2005 
2006 		/*
2007 		 * If there's a BPF listener, bounce a copy of this frame
2008 		 * to him.
2009 		 */
2010 		ETHER_BPF_MTAP(ifp, m_head2);
2011 	}
2012 
2013 	if (!need_trans) {
2014 		if (RE_TX_MODERATION_IS_ENABLED(sc) &&
2015 		    sc->re_ldata.re_tx_free != RE_TX_DESC_CNT)
2016 			CSR_WRITE_4(sc, RE_TIMERCNT, 1);
2017 		return;
2018 	}
2019 
2020 	/* Flush the TX descriptors */
2021 	bus_dmamap_sync(sc->re_ldata.re_tx_list_tag,
2022 			sc->re_ldata.re_tx_list_map, BUS_DMASYNC_PREWRITE);
2023 
2024 	sc->re_ldata.re_tx_prodidx = idx;
2025 
2026 	/*
2027 	 * RealTek put the TX poll request register in a different
2028 	 * location on the 8169 gigE chip. I don't know why.
2029 	 */
2030 	CSR_WRITE_1(sc, sc->re_txstart, RE_TXSTART_START);
2031 
2032 	if (RE_TX_MODERATION_IS_ENABLED(sc)) {
2033 		/*
2034 		 * Use the countdown timer for interrupt moderation.
2035 		 * 'TX done' interrupts are disabled. Instead, we reset the
2036 		 * countdown timer, which will begin counting until it hits
2037 		 * the value in the TIMERINT register, and then trigger an
2038 		 * interrupt. Each time we write to the TIMERCNT register,
2039 		 * the timer count is reset to 0.
2040 		 */
2041 		CSR_WRITE_4(sc, RE_TIMERCNT, 1);
2042 	}
2043 
2044 	/*
2045 	 * Set a timeout in case the chip goes out to lunch.
2046 	 */
2047 	ifp->if_timer = 5;
2048 }
2049 
2050 static void
2051 re_init(void *xsc)
2052 {
2053 	struct re_softc *sc = xsc;
2054 	struct ifnet *ifp = &sc->arpcom.ac_if;
2055 	struct mii_data *mii;
2056 	uint32_t rxcfg = 0;
2057 
2058 	mii = device_get_softc(sc->re_miibus);
2059 
2060 	/*
2061 	 * Cancel pending I/O and free all RX/TX buffers.
2062 	 */
2063 	re_stop(sc);
2064 
2065 	/*
2066 	 * Enable C+ RX and TX mode, as well as VLAN stripping and
2067 	 * RX checksum offload. We must configure the C+ register
2068 	 * before all others.
2069 	 */
2070 	CSR_WRITE_2(sc, RE_CPLUS_CMD, RE_CPLUSCMD_RXENB | RE_CPLUSCMD_TXENB |
2071 		    RE_CPLUSCMD_PCI_MRW | RE_CPLUSCMD_VLANSTRIP |
2072 		    (ifp->if_capenable & IFCAP_RXCSUM ?
2073 		     RE_CPLUSCMD_RXCSUM_ENB : 0));
2074 
2075 	/*
2076 	 * Init our MAC address.  Even though the chipset
2077 	 * documentation doesn't mention it, we need to enter "Config
2078 	 * register write enable" mode to modify the ID registers.
2079 	 */
2080 	CSR_WRITE_1(sc, RE_EECMD, RE_EEMODE_WRITECFG);
2081 	CSR_WRITE_4(sc, RE_IDR0,
2082 	    htole32(*(uint32_t *)(&sc->arpcom.ac_enaddr[0])));
2083 	CSR_WRITE_2(sc, RE_IDR4,
2084 	    htole16(*(uint16_t *)(&sc->arpcom.ac_enaddr[4])));
2085 	CSR_WRITE_1(sc, RE_EECMD, RE_EEMODE_OFF);
2086 
2087 	/*
2088 	 * For C+ mode, initialize the RX descriptors and mbufs.
2089 	 */
2090 	re_rx_list_init(sc);
2091 	re_tx_list_init(sc);
2092 
2093 	/*
2094 	 * Load the addresses of the RX and TX lists into the chip.
2095 	 */
2096 	CSR_WRITE_4(sc, RE_RXLIST_ADDR_HI,
2097 	    RE_ADDR_HI(sc->re_ldata.re_rx_list_addr));
2098 	CSR_WRITE_4(sc, RE_RXLIST_ADDR_LO,
2099 	    RE_ADDR_LO(sc->re_ldata.re_rx_list_addr));
2100 
2101 	CSR_WRITE_4(sc, RE_TXLIST_ADDR_HI,
2102 	    RE_ADDR_HI(sc->re_ldata.re_tx_list_addr));
2103 	CSR_WRITE_4(sc, RE_TXLIST_ADDR_LO,
2104 	    RE_ADDR_LO(sc->re_ldata.re_tx_list_addr));
2105 
2106 	/*
2107 	 * Enable transmit and receive.
2108 	 */
2109 	CSR_WRITE_1(sc, RE_COMMAND, RE_CMD_TX_ENB|RE_CMD_RX_ENB);
2110 
2111 	/*
2112 	 * Set the initial TX and RX configuration.
2113 	 */
2114 	if (sc->re_testmode) {
2115 		if (sc->re_type == RE_8169)
2116 			CSR_WRITE_4(sc, RE_TXCFG,
2117 				    RE_TXCFG_CONFIG | RE_LOOPTEST_ON);
2118 		else
2119 			CSR_WRITE_4(sc, RE_TXCFG,
2120 				    RE_TXCFG_CONFIG | RE_LOOPTEST_ON_CPLUS);
2121 	} else
2122 		CSR_WRITE_4(sc, RE_TXCFG, RE_TXCFG_CONFIG);
2123 
2124 	CSR_WRITE_1(sc, RE_EARLY_TX_THRESH, 16);
2125 
2126 	CSR_WRITE_4(sc, RE_RXCFG, RE_RXCFG_CONFIG);
2127 
2128 	/* Set the individual bit to receive frames for this host only. */
2129 	rxcfg = CSR_READ_4(sc, RE_RXCFG);
2130 	rxcfg |= RE_RXCFG_RX_INDIV;
2131 
2132 	/* If we want promiscuous mode, set the allframes bit. */
2133 	if (ifp->if_flags & IFF_PROMISC) {
2134 		rxcfg |= RE_RXCFG_RX_ALLPHYS;
2135 		CSR_WRITE_4(sc, RE_RXCFG, rxcfg);
2136 	} else {
2137 		rxcfg &= ~RE_RXCFG_RX_ALLPHYS;
2138 		CSR_WRITE_4(sc, RE_RXCFG, rxcfg);
2139 	}
2140 
2141 	/*
2142 	 * Set capture broadcast bit to capture broadcast frames.
2143 	 */
2144 	if (ifp->if_flags & IFF_BROADCAST) {
2145 		rxcfg |= RE_RXCFG_RX_BROAD;
2146 		CSR_WRITE_4(sc, RE_RXCFG, rxcfg);
2147 	} else {
2148 		rxcfg &= ~RE_RXCFG_RX_BROAD;
2149 		CSR_WRITE_4(sc, RE_RXCFG, rxcfg);
2150 	}
2151 
2152 	/*
2153 	 * Program the multicast filter, if necessary.
2154 	 */
2155 	re_setmulti(sc);
2156 
2157 #ifdef DEVICE_POLLING
2158 	/*
2159 	 * Disable interrupts if we are polling.
2160 	 */
2161 	if (ifp->if_flags & IFF_POLLING)
2162 		CSR_WRITE_2(sc, RE_IMR, 0);
2163 	else	/* otherwise ... */
2164 #endif /* DEVICE_POLLING */
2165 	/*
2166 	 * Enable interrupts.
2167 	 */
2168 	if (sc->re_testmode)
2169 		CSR_WRITE_2(sc, RE_IMR, 0);
2170 	else
2171 		CSR_WRITE_2(sc, RE_IMR, sc->re_intrs);
2172 	CSR_WRITE_2(sc, RE_ISR, sc->re_intrs);
2173 
2174 	/* Set initial TX threshold */
2175 	sc->re_txthresh = RE_TX_THRESH_INIT;
2176 
2177 	/* Start RX/TX process. */
2178 	if (sc->re_flags & RE_F_HASMPC)
2179 		CSR_WRITE_4(sc, RE_MISSEDPKT, 0);
2180 #ifdef notdef
2181 	/* Enable receiver and transmitter. */
2182 	CSR_WRITE_1(sc, RE_COMMAND, RE_CMD_TX_ENB|RE_CMD_RX_ENB);
2183 #endif
2184 
2185 	if (RE_TX_MODERATION_IS_ENABLED(sc)) {
2186 		/*
2187 		 * Initialize the timer interrupt register so that
2188 		 * a timer interrupt will be generated once the timer
2189 		 * reaches a certain number of ticks. The timer is
2190 		 * reloaded on each transmit. This gives us TX interrupt
2191 		 * moderation, which dramatically improves TX frame rate.
2192 		 */
2193 		if (sc->re_type == RE_8169)
2194 			CSR_WRITE_4(sc, RE_TIMERINT_8169, 0x800);
2195 		else
2196 			CSR_WRITE_4(sc, RE_TIMERINT, 0x400);
2197 	}
2198 
2199 	/*
2200 	 * For 8169 gigE NICs, set the max allowed RX packet
2201 	 * size so we can receive jumbo frames.
2202 	 */
2203 	if (sc->re_type == RE_8169)
2204 		CSR_WRITE_2(sc, RE_MAXRXPKTLEN, 16383);
2205 
2206 	if (sc->re_testmode) {
2207 		return;
2208 	}
2209 
2210 	mii_mediachg(mii);
2211 
2212 	CSR_WRITE_1(sc, RE_CFG1, RE_CFG1_DRVLOAD|RE_CFG1_FULLDUPLEX);
2213 
2214 	ifp->if_flags |= IFF_RUNNING;
2215 	ifp->if_flags &= ~IFF_OACTIVE;
2216 
2217 	sc->re_link = 0;
2218 	callout_reset(&sc->re_timer, hz, re_tick, sc);
2219 }
2220 
2221 /*
2222  * Set media options.
2223  */
2224 static int
2225 re_ifmedia_upd(struct ifnet *ifp)
2226 {
2227 	struct re_softc *sc = ifp->if_softc;
2228 	struct mii_data *mii;
2229 
2230 	mii = device_get_softc(sc->re_miibus);
2231 	mii_mediachg(mii);
2232 
2233 	return(0);
2234 }
2235 
2236 /*
2237  * Report current media status.
2238  */
2239 static void
2240 re_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
2241 {
2242 	struct re_softc *sc = ifp->if_softc;
2243 	struct mii_data *mii;
2244 
2245 	mii = device_get_softc(sc->re_miibus);
2246 
2247 	mii_pollstat(mii);
2248 	ifmr->ifm_active = mii->mii_media_active;
2249 	ifmr->ifm_status = mii->mii_media_status;
2250 }
2251 
2252 static int
2253 re_ioctl(struct ifnet *ifp, u_long command, caddr_t data, struct ucred *cr)
2254 {
2255 	struct re_softc *sc = ifp->if_softc;
2256 	struct ifreq *ifr = (struct ifreq *) data;
2257 	struct mii_data *mii;
2258 	int error = 0;
2259 
2260 	switch(command) {
2261 	case SIOCSIFMTU:
2262 		if (ifr->ifr_mtu > RE_JUMBO_MTU)
2263 			error = EINVAL;
2264 		ifp->if_mtu = ifr->ifr_mtu;
2265 		break;
2266 	case SIOCSIFFLAGS:
2267 		if (ifp->if_flags & IFF_UP)
2268 			re_init(sc);
2269 		else if (ifp->if_flags & IFF_RUNNING)
2270 			re_stop(sc);
2271 		break;
2272 	case SIOCADDMULTI:
2273 	case SIOCDELMULTI:
2274 		re_setmulti(sc);
2275 		error = 0;
2276 		break;
2277 	case SIOCGIFMEDIA:
2278 	case SIOCSIFMEDIA:
2279 		mii = device_get_softc(sc->re_miibus);
2280 		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, command);
2281 		break;
2282 	case SIOCSIFCAP:
2283 		ifp->if_capenable &= ~(IFCAP_HWCSUM);
2284 		ifp->if_capenable |=
2285 		    ifr->ifr_reqcap & (IFCAP_HWCSUM);
2286 		if (ifp->if_capenable & IFCAP_TXCSUM)
2287 			ifp->if_hwassist = RE_CSUM_FEATURES;
2288 		else
2289 			ifp->if_hwassist = 0;
2290 		if (ifp->if_flags & IFF_RUNNING)
2291 			re_init(sc);
2292 		break;
2293 	default:
2294 		error = ether_ioctl(ifp, command, data);
2295 		break;
2296 	}
2297 	return(error);
2298 }
2299 
2300 static void
2301 re_watchdog(struct ifnet *ifp)
2302 {
2303 	struct re_softc *sc = ifp->if_softc;
2304 
2305 	if_printf(ifp, "watchdog timeout\n");
2306 
2307 	ifp->if_oerrors++;
2308 
2309 	re_txeof(sc);
2310 	re_rxeof(sc);
2311 
2312 	re_init(sc);
2313 
2314 	if (!ifq_is_empty(&ifp->if_snd))
2315 		ifp->if_start(ifp);
2316 }
2317 
2318 /*
2319  * Stop the adapter and free any mbufs allocated to the
2320  * RX and TX lists.
2321  */
2322 static void
2323 re_stop(struct re_softc *sc)
2324 {
2325 	struct ifnet *ifp = &sc->arpcom.ac_if;
2326 	int i;
2327 
2328 	ifp->if_timer = 0;
2329 	callout_stop(&sc->re_timer);
2330 
2331 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
2332 
2333 	CSR_WRITE_1(sc, RE_COMMAND, 0x00);
2334 	CSR_WRITE_2(sc, RE_IMR, 0x0000);
2335 	CSR_WRITE_2(sc, RE_ISR, 0xFFFF);
2336 
2337 	if (sc->re_head != NULL) {
2338 		m_freem(sc->re_head);
2339 		sc->re_head = sc->re_tail = NULL;
2340 	}
2341 
2342 	/* Free the TX list buffers. */
2343 	for (i = 0; i < RE_TX_DESC_CNT; i++) {
2344 		if (sc->re_ldata.re_tx_mbuf[i] != NULL) {
2345 			bus_dmamap_unload(sc->re_ldata.re_mtag,
2346 					  sc->re_ldata.re_tx_dmamap[i]);
2347 			m_freem(sc->re_ldata.re_tx_mbuf[i]);
2348 			sc->re_ldata.re_tx_mbuf[i] = NULL;
2349 		}
2350 	}
2351 
2352 	/* Free the RX list buffers. */
2353 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
2354 		if (sc->re_ldata.re_rx_mbuf[i] != NULL) {
2355 			bus_dmamap_unload(sc->re_ldata.re_mtag,
2356 					  sc->re_ldata.re_rx_dmamap[i]);
2357 			m_freem(sc->re_ldata.re_rx_mbuf[i]);
2358 			sc->re_ldata.re_rx_mbuf[i] = NULL;
2359 		}
2360 	}
2361 }
2362 
2363 /*
2364  * Device suspend routine.  Stop the interface and save some PCI
2365  * settings in case the BIOS doesn't restore them properly on
2366  * resume.
2367  */
2368 static int
2369 re_suspend(device_t dev)
2370 {
2371 #ifndef BURN_BRIDGES
2372 	int i;
2373 #endif
2374 	struct re_softc *sc = device_get_softc(dev);
2375 
2376 	re_stop(sc);
2377 
2378 #ifndef BURN_BRIDGES
2379 	for (i = 0; i < 5; i++)
2380 		sc->saved_maps[i] = pci_read_config(dev, PCIR_MAPS + i * 4, 4);
2381 	sc->saved_biosaddr = pci_read_config(dev, PCIR_BIOS, 4);
2382 	sc->saved_intline = pci_read_config(dev, PCIR_INTLINE, 1);
2383 	sc->saved_cachelnsz = pci_read_config(dev, PCIR_CACHELNSZ, 1);
2384 	sc->saved_lattimer = pci_read_config(dev, PCIR_LATTIMER, 1);
2385 #endif
2386 
2387 	sc->suspended = 1;
2388 
2389 	return (0);
2390 }
2391 
2392 /*
2393  * Device resume routine.  Restore some PCI settings in case the BIOS
2394  * doesn't, re-enable busmastering, and restart the interface if
2395  * appropriate.
2396  */
2397 static int
2398 re_resume(device_t dev)
2399 {
2400 	struct re_softc *sc = device_get_softc(dev);
2401 	struct ifnet *ifp = &sc->arpcom.ac_if;
2402 #ifndef BURN_BRIDGES
2403 	int i;
2404 #endif
2405 
2406 #ifndef BURN_BRIDGES
2407 	/* better way to do this? */
2408 	for (i = 0; i < 5; i++)
2409 		pci_write_config(dev, PCIR_MAPS + i * 4, sc->saved_maps[i], 4);
2410 	pci_write_config(dev, PCIR_BIOS, sc->saved_biosaddr, 4);
2411 	pci_write_config(dev, PCIR_INTLINE, sc->saved_intline, 1);
2412 	pci_write_config(dev, PCIR_CACHELNSZ, sc->saved_cachelnsz, 1);
2413 	pci_write_config(dev, PCIR_LATTIMER, sc->saved_lattimer, 1);
2414 
2415 	/* reenable busmastering */
2416 	pci_enable_busmaster(dev);
2417 	pci_enable_io(dev, SYS_RES_IOPORT);
2418 #endif
2419 
2420 	/* reinitialize interface if necessary */
2421 	if (ifp->if_flags & IFF_UP)
2422 		re_init(sc);
2423 
2424 	sc->suspended = 0;
2425 
2426 	return (0);
2427 }
2428 
2429 /*
2430  * Stop all chip I/O so that the kernel's probe routines don't
2431  * get confused by errant DMAs when rebooting.
2432  */
2433 static void
2434 re_shutdown(device_t dev)
2435 {
2436 	struct re_softc *sc = device_get_softc(dev);
2437 	struct ifnet *ifp = &sc->arpcom.ac_if;
2438 
2439 	lwkt_serialize_enter(ifp->if_serializer);
2440 	re_stop(sc);
2441 	lwkt_serialize_exit(ifp->if_serializer);
2442 }
2443 
2444 static int
2445 re_sysctl_tx_moderation(SYSCTL_HANDLER_ARGS)
2446 {
2447 	struct re_softc *sc = arg1;
2448 	struct ifnet *ifp = &sc->arpcom.ac_if;
2449 	int error = 0, mod, mod_old;
2450 
2451 	lwkt_serialize_enter(ifp->if_serializer);
2452 
2453 	mod_old = mod = RE_TX_MODERATION_IS_ENABLED(sc);
2454 
2455 	error = sysctl_handle_int(oidp, &mod, 0, req);
2456 	if (error || req->newptr == NULL || mod == mod_old)
2457 		goto back;
2458 	if (mod != 0 && mod != 1) {
2459 		error = EINVAL;
2460 		goto back;
2461 	}
2462 
2463 	if (mod)
2464 		RE_ENABLE_TX_MODERATION(sc);
2465 	else
2466 		RE_DISABLE_TX_MODERATION(sc);
2467 
2468 	if ((ifp->if_flags & (IFF_RUNNING | IFF_UP)) == (IFF_RUNNING | IFF_UP))
2469 		re_init(sc);
2470 back:
2471 	lwkt_serialize_exit(ifp->if_serializer);
2472 	return error;
2473 }
2474