xref: /freebsd/sys/dev/re/if_re.c (revision 10f0bcab61ef441cb5af32fb706688d8cbd55dc0)
1 /*-
2  * Copyright (c) 1997, 1998-2003
3  *	Bill Paul <wpaul@windriver.com>.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. All advertising materials mentioning features or use of this software
14  *    must display the following acknowledgement:
15  *	This product includes software developed by Bill Paul.
16  * 4. Neither the name of the author nor the names of any co-contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY Bill Paul AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL Bill Paul OR THE VOICES IN HIS HEAD
24  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
30  * THE POSSIBILITY OF SUCH DAMAGE.
31  */
32 
33 #include <sys/cdefs.h>
34 __FBSDID("$FreeBSD$");
35 
36 /*
37  * RealTek 8139C+/8169/8169S/8110S/8168/8111/8101E PCI NIC driver
38  *
39  * Written by Bill Paul <wpaul@windriver.com>
40  * Senior Networking Software Engineer
41  * Wind River Systems
42  */
43 
44 /*
45  * This driver is designed to support RealTek's next generation of
46  * 10/100 and 10/100/1000 PCI ethernet controllers. There are currently
47  * seven devices in this family: the RTL8139C+, the RTL8169, the RTL8169S,
48  * RTL8110S, the RTL8168, the RTL8111 and the RTL8101E.
49  *
50  * The 8139C+ is a 10/100 ethernet chip. It is backwards compatible
51  * with the older 8139 family, however it also supports a special
52  * C+ mode of operation that provides several new performance enhancing
53  * features. These include:
54  *
55  *	o Descriptor based DMA mechanism. Each descriptor represents
56  *	  a single packet fragment. Data buffers may be aligned on
57  *	  any byte boundary.
58  *
59  *	o 64-bit DMA
60  *
61  *	o TCP/IP checksum offload for both RX and TX
62  *
63  *	o High and normal priority transmit DMA rings
64  *
65  *	o VLAN tag insertion and extraction
66  *
67  *	o TCP large send (segmentation offload)
68  *
69  * Like the 8139, the 8139C+ also has a built-in 10/100 PHY. The C+
70  * programming API is fairly straightforward. The RX filtering, EEPROM
71  * access and PHY access is the same as it is on the older 8139 series
72  * chips.
73  *
74  * The 8169 is a 64-bit 10/100/1000 gigabit ethernet MAC. It has almost the
75  * same programming API and feature set as the 8139C+ with the following
76  * differences and additions:
77  *
78  *	o 1000Mbps mode
79  *
80  *	o Jumbo frames
81  *
82  *	o GMII and TBI ports/registers for interfacing with copper
83  *	  or fiber PHYs
84  *
85  *	o RX and TX DMA rings can have up to 1024 descriptors
86  *	  (the 8139C+ allows a maximum of 64)
87  *
88  *	o Slight differences in register layout from the 8139C+
89  *
90  * The TX start and timer interrupt registers are at different locations
91  * on the 8169 than they are on the 8139C+. Also, the status word in the
92  * RX descriptor has a slightly different bit layout. The 8169 does not
93  * have a built-in PHY. Most reference boards use a Marvell 88E1000 'Alaska'
94  * copper gigE PHY.
95  *
96  * The 8169S/8110S 10/100/1000 devices have built-in copper gigE PHYs
97  * (the 'S' stands for 'single-chip'). These devices have the same
98  * programming API as the older 8169, but also have some vendor-specific
99  * registers for the on-board PHY. The 8110S is a LAN-on-motherboard
100  * part designed to be pin-compatible with the RealTek 8100 10/100 chip.
101  *
102  * This driver takes advantage of the RX and TX checksum offload and
103  * VLAN tag insertion/extraction features. It also implements TX
104  * interrupt moderation using the timer interrupt registers, which
105  * significantly reduces TX interrupt load. There is also support
106  * for jumbo frames, however the 8169/8169S/8110S can not transmit
107  * jumbo frames larger than 7440, so the max MTU possible with this
108  * driver is 7422 bytes.
109  */
110 
111 #ifdef HAVE_KERNEL_OPTION_HEADERS
112 #include "opt_device_polling.h"
113 #endif
114 
115 #include <sys/param.h>
116 #include <sys/endian.h>
117 #include <sys/systm.h>
118 #include <sys/sockio.h>
119 #include <sys/mbuf.h>
120 #include <sys/malloc.h>
121 #include <sys/module.h>
122 #include <sys/kernel.h>
123 #include <sys/socket.h>
124 #include <sys/lock.h>
125 #include <sys/mutex.h>
126 #include <sys/taskqueue.h>
127 
128 #include <net/if.h>
129 #include <net/if_arp.h>
130 #include <net/ethernet.h>
131 #include <net/if_dl.h>
132 #include <net/if_media.h>
133 #include <net/if_types.h>
134 #include <net/if_vlan_var.h>
135 
136 #include <net/bpf.h>
137 
138 #include <machine/bus.h>
139 #include <machine/resource.h>
140 #include <sys/bus.h>
141 #include <sys/rman.h>
142 
143 #include <dev/mii/mii.h>
144 #include <dev/mii/miivar.h>
145 
146 #include <dev/pci/pcireg.h>
147 #include <dev/pci/pcivar.h>
148 
149 #include <pci/if_rlreg.h>
150 
151 MODULE_DEPEND(re, pci, 1, 1, 1);
152 MODULE_DEPEND(re, ether, 1, 1, 1);
153 MODULE_DEPEND(re, miibus, 1, 1, 1);
154 
155 /* "device miibus" required.  See GENERIC if you get errors here. */
156 #include "miibus_if.h"
157 
158 /*
159  * Default to using PIO access for this driver.
160  */
161 #define RE_USEIOSPACE
162 
163 /* Tunables. */
164 static int msi_disable = 1;
165 TUNABLE_INT("hw.re.msi_disable", &msi_disable);
166 
167 #define RE_CSUM_FEATURES    (CSUM_IP | CSUM_TCP | CSUM_UDP)
168 
169 /*
170  * Various supported device vendors/types and their names.
171  */
172 static struct rl_type re_devs[] = {
173 	{ DLINK_VENDORID, DLINK_DEVICEID_528T, RL_HWREV_8169S,
174 		"D-Link DGE-528(T) Gigabit Ethernet Adapter" },
175 	{ DLINK_VENDORID, DLINK_DEVICEID_528T, RL_HWREV_8169_8110SB,
176 		"D-Link DGE-528(T) Rev.B1 Gigabit Ethernet Adapter" },
177 	{ RT_VENDORID, RT_DEVICEID_8139, RL_HWREV_8139CPLUS,
178 		"RealTek 8139C+ 10/100BaseTX" },
179 	{ RT_VENDORID, RT_DEVICEID_8101E, RL_HWREV_8101E,
180 		"RealTek 8101E PCIe 10/100baseTX" },
181 	{ RT_VENDORID, RT_DEVICEID_8168, RL_HWREV_8168_SPIN1,
182 		"RealTek 8168/8111B PCIe Gigabit Ethernet" },
183 	{ RT_VENDORID, RT_DEVICEID_8168, RL_HWREV_8168_SPIN2,
184 		"RealTek 8168/8111B PCIe Gigabit Ethernet" },
185 	{ RT_VENDORID, RT_DEVICEID_8168, RL_HWREV_8168_SPIN3,
186 		"RealTek 8168/8111B PCIe Gigabit Ethernet" },
187 	{ RT_VENDORID, RT_DEVICEID_8169, RL_HWREV_8169,
188 		"RealTek 8169 Gigabit Ethernet" },
189 	{ RT_VENDORID, RT_DEVICEID_8169, RL_HWREV_8169S,
190 		"RealTek 8169S Single-chip Gigabit Ethernet" },
191 	{ RT_VENDORID, RT_DEVICEID_8169, RL_HWREV_8169_8110SB,
192 		"RealTek 8169SB/8110SB Single-chip Gigabit Ethernet" },
193 	{ RT_VENDORID, RT_DEVICEID_8169, RL_HWREV_8169_8110SC,
194 		"RealTek 8169SC/8110SC Single-chip Gigabit Ethernet" },
195 	{ RT_VENDORID, RT_DEVICEID_8169SC, RL_HWREV_8169_8110SC,
196 		"RealTek 8169SC/8110SC Single-chip Gigabit Ethernet" },
197 	{ RT_VENDORID, RT_DEVICEID_8169, RL_HWREV_8110S,
198 		"RealTek 8110S Single-chip Gigabit Ethernet" },
199 	{ COREGA_VENDORID, COREGA_DEVICEID_CGLAPCIGT, RL_HWREV_8169S,
200 		"Corega CG-LAPCIGT (RTL8169S) Gigabit Ethernet" },
201 	{ LINKSYS_VENDORID, LINKSYS_DEVICEID_EG1032, RL_HWREV_8169S,
202 		"Linksys EG1032 (RTL8169S) Gigabit Ethernet" },
203 	{ USR_VENDORID, USR_DEVICEID_997902, RL_HWREV_8169S,
204 		"US Robotics 997902 (RTL8169S) Gigabit Ethernet" }
205 };
206 
207 static struct rl_hwrev re_hwrevs[] = {
208 	{ RL_HWREV_8139, RL_8139,  "" },
209 	{ RL_HWREV_8139A, RL_8139, "A" },
210 	{ RL_HWREV_8139AG, RL_8139, "A-G" },
211 	{ RL_HWREV_8139B, RL_8139, "B" },
212 	{ RL_HWREV_8130, RL_8139, "8130" },
213 	{ RL_HWREV_8139C, RL_8139, "C" },
214 	{ RL_HWREV_8139D, RL_8139, "8139D/8100B/8100C" },
215 	{ RL_HWREV_8139CPLUS, RL_8139CPLUS, "C+"},
216 	{ RL_HWREV_8168_SPIN1, RL_8169, "8168"},
217 	{ RL_HWREV_8169, RL_8169, "8169"},
218 	{ RL_HWREV_8169S, RL_8169, "8169S"},
219 	{ RL_HWREV_8110S, RL_8169, "8110S"},
220 	{ RL_HWREV_8169_8110SB, RL_8169, "8169SB"},
221 	{ RL_HWREV_8169_8110SC, RL_8169, "8169SC"},
222 	{ RL_HWREV_8100, RL_8139, "8100"},
223 	{ RL_HWREV_8101, RL_8139, "8101"},
224 	{ RL_HWREV_8100E, RL_8169, "8100E"},
225 	{ RL_HWREV_8101E, RL_8169, "8101E"},
226 	{ RL_HWREV_8168_SPIN2, RL_8169, "8168"},
227 	{ RL_HWREV_8168_SPIN3, RL_8169, "8168"},
228 	{ 0, 0, NULL }
229 };
230 
231 static int re_probe		(device_t);
232 static int re_attach		(device_t);
233 static int re_detach		(device_t);
234 
235 static int re_encap		(struct rl_softc *, struct mbuf **);
236 
237 static void re_dma_map_addr	(void *, bus_dma_segment_t *, int, int);
238 static int re_allocmem		(device_t, struct rl_softc *);
239 static __inline void re_discard_rxbuf
240 				(struct rl_softc *, int);
241 static int re_newbuf		(struct rl_softc *, int);
242 static int re_rx_list_init	(struct rl_softc *);
243 static int re_tx_list_init	(struct rl_softc *);
244 #ifdef RE_FIXUP_RX
245 static __inline void re_fixup_rx
246 				(struct mbuf *);
247 #endif
248 static int re_rxeof		(struct rl_softc *);
249 static void re_txeof		(struct rl_softc *);
250 #ifdef DEVICE_POLLING
251 static void re_poll		(struct ifnet *, enum poll_cmd, int);
252 static void re_poll_locked	(struct ifnet *, enum poll_cmd, int);
253 #endif
254 static int re_intr		(void *);
255 static void re_tick		(void *);
256 static void re_tx_task		(void *, int);
257 static void re_int_task		(void *, int);
258 static void re_start		(struct ifnet *);
259 static int re_ioctl		(struct ifnet *, u_long, caddr_t);
260 static void re_init		(void *);
261 static void re_init_locked	(struct rl_softc *);
262 static void re_stop		(struct rl_softc *);
263 static void re_watchdog		(struct rl_softc *);
264 static int re_suspend		(device_t);
265 static int re_resume		(device_t);
266 static int re_shutdown		(device_t);
267 static int re_ifmedia_upd	(struct ifnet *);
268 static void re_ifmedia_sts	(struct ifnet *, struct ifmediareq *);
269 
270 static void re_eeprom_putbyte	(struct rl_softc *, int);
271 static void re_eeprom_getword	(struct rl_softc *, int, u_int16_t *);
272 static void re_read_eeprom	(struct rl_softc *, caddr_t, int, int);
273 static int re_gmii_readreg	(device_t, int, int);
274 static int re_gmii_writereg	(device_t, int, int, int);
275 
276 static int re_miibus_readreg	(device_t, int, int);
277 static int re_miibus_writereg	(device_t, int, int, int);
278 static void re_miibus_statchg	(device_t);
279 
280 static void re_setmulti		(struct rl_softc *);
281 static void re_reset		(struct rl_softc *);
282 static void re_setwol		(struct rl_softc *);
283 static void re_clrwol		(struct rl_softc *);
284 
285 #ifdef RE_DIAG
286 static int re_diag		(struct rl_softc *);
287 #endif
288 
289 #ifdef RE_USEIOSPACE
290 #define RL_RES			SYS_RES_IOPORT
291 #define RL_RID			RL_PCI_LOIO
292 #else
293 #define RL_RES			SYS_RES_MEMORY
294 #define RL_RID			RL_PCI_LOMEM
295 #endif
296 
297 static device_method_t re_methods[] = {
298 	/* Device interface */
299 	DEVMETHOD(device_probe,		re_probe),
300 	DEVMETHOD(device_attach,	re_attach),
301 	DEVMETHOD(device_detach,	re_detach),
302 	DEVMETHOD(device_suspend,	re_suspend),
303 	DEVMETHOD(device_resume,	re_resume),
304 	DEVMETHOD(device_shutdown,	re_shutdown),
305 
306 	/* bus interface */
307 	DEVMETHOD(bus_print_child,	bus_generic_print_child),
308 	DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
309 
310 	/* MII interface */
311 	DEVMETHOD(miibus_readreg,	re_miibus_readreg),
312 	DEVMETHOD(miibus_writereg,	re_miibus_writereg),
313 	DEVMETHOD(miibus_statchg,	re_miibus_statchg),
314 
315 	{ 0, 0 }
316 };
317 
318 static driver_t re_driver = {
319 	"re",
320 	re_methods,
321 	sizeof(struct rl_softc)
322 };
323 
324 static devclass_t re_devclass;
325 
326 DRIVER_MODULE(re, pci, re_driver, re_devclass, 0, 0);
327 DRIVER_MODULE(re, cardbus, re_driver, re_devclass, 0, 0);
328 DRIVER_MODULE(miibus, re, miibus_driver, miibus_devclass, 0, 0);
329 
330 #define EE_SET(x)					\
331 	CSR_WRITE_1(sc, RL_EECMD,			\
332 		CSR_READ_1(sc, RL_EECMD) | x)
333 
334 #define EE_CLR(x)					\
335 	CSR_WRITE_1(sc, RL_EECMD,			\
336 		CSR_READ_1(sc, RL_EECMD) & ~x)
337 
338 /*
339  * Send a read command and address to the EEPROM, check for ACK.
340  */
341 static void
342 re_eeprom_putbyte(sc, addr)
343 	struct rl_softc		*sc;
344 	int			addr;
345 {
346 	register int		d, i;
347 
348 	d = addr | (RL_9346_READ << sc->rl_eewidth);
349 
350 	/*
351 	 * Feed in each bit and strobe the clock.
352 	 */
353 
354 	for (i = 1 << (sc->rl_eewidth + 3); i; i >>= 1) {
355 		if (d & i) {
356 			EE_SET(RL_EE_DATAIN);
357 		} else {
358 			EE_CLR(RL_EE_DATAIN);
359 		}
360 		DELAY(100);
361 		EE_SET(RL_EE_CLK);
362 		DELAY(150);
363 		EE_CLR(RL_EE_CLK);
364 		DELAY(100);
365 	}
366 
367 	return;
368 }
369 
370 /*
371  * Read a word of data stored in the EEPROM at address 'addr.'
372  */
373 static void
374 re_eeprom_getword(sc, addr, dest)
375 	struct rl_softc		*sc;
376 	int			addr;
377 	u_int16_t		*dest;
378 {
379 	register int		i;
380 	u_int16_t		word = 0;
381 
382 	/*
383 	 * Send address of word we want to read.
384 	 */
385 	re_eeprom_putbyte(sc, addr);
386 
387 	/*
388 	 * Start reading bits from EEPROM.
389 	 */
390 	for (i = 0x8000; i; i >>= 1) {
391 		EE_SET(RL_EE_CLK);
392 		DELAY(100);
393 		if (CSR_READ_1(sc, RL_EECMD) & RL_EE_DATAOUT)
394 			word |= i;
395 		EE_CLR(RL_EE_CLK);
396 		DELAY(100);
397 	}
398 
399 	*dest = word;
400 
401 	return;
402 }
403 
404 /*
405  * Read a sequence of words from the EEPROM.
406  */
407 static void
408 re_read_eeprom(sc, dest, off, cnt)
409 	struct rl_softc		*sc;
410 	caddr_t			dest;
411 	int			off;
412 	int			cnt;
413 {
414 	int			i;
415 	u_int16_t		word = 0, *ptr;
416 
417 	CSR_SETBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM);
418 
419         DELAY(100);
420 
421 	for (i = 0; i < cnt; i++) {
422 		CSR_SETBIT_1(sc, RL_EECMD, RL_EE_SEL);
423 		re_eeprom_getword(sc, off + i, &word);
424 		CSR_CLRBIT_1(sc, RL_EECMD, RL_EE_SEL);
425 		ptr = (u_int16_t *)(dest + (i * 2));
426                 *ptr = word;
427 	}
428 
429 	CSR_CLRBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM);
430 
431 	return;
432 }
433 
434 static int
435 re_gmii_readreg(dev, phy, reg)
436 	device_t		dev;
437 	int			phy, reg;
438 {
439 	struct rl_softc		*sc;
440 	u_int32_t		rval;
441 	int			i;
442 
443 	if (phy != 1)
444 		return (0);
445 
446 	sc = device_get_softc(dev);
447 
448 	/* Let the rgephy driver read the GMEDIASTAT register */
449 
450 	if (reg == RL_GMEDIASTAT) {
451 		rval = CSR_READ_1(sc, RL_GMEDIASTAT);
452 		return (rval);
453 	}
454 
455 	CSR_WRITE_4(sc, RL_PHYAR, reg << 16);
456 	DELAY(1000);
457 
458 	for (i = 0; i < RL_TIMEOUT; i++) {
459 		rval = CSR_READ_4(sc, RL_PHYAR);
460 		if (rval & RL_PHYAR_BUSY)
461 			break;
462 		DELAY(100);
463 	}
464 
465 	if (i == RL_TIMEOUT) {
466 		device_printf(sc->rl_dev, "PHY read failed\n");
467 		return (0);
468 	}
469 
470 	return (rval & RL_PHYAR_PHYDATA);
471 }
472 
473 static int
474 re_gmii_writereg(dev, phy, reg, data)
475 	device_t		dev;
476 	int			phy, reg, data;
477 {
478 	struct rl_softc		*sc;
479 	u_int32_t		rval;
480 	int			i;
481 
482 	sc = device_get_softc(dev);
483 
484 	CSR_WRITE_4(sc, RL_PHYAR, (reg << 16) |
485 	    (data & RL_PHYAR_PHYDATA) | RL_PHYAR_BUSY);
486 	DELAY(1000);
487 
488 	for (i = 0; i < RL_TIMEOUT; i++) {
489 		rval = CSR_READ_4(sc, RL_PHYAR);
490 		if (!(rval & RL_PHYAR_BUSY))
491 			break;
492 		DELAY(100);
493 	}
494 
495 	if (i == RL_TIMEOUT) {
496 		device_printf(sc->rl_dev, "PHY write failed\n");
497 		return (0);
498 	}
499 
500 	return (0);
501 }
502 
503 static int
504 re_miibus_readreg(dev, phy, reg)
505 	device_t		dev;
506 	int			phy, reg;
507 {
508 	struct rl_softc		*sc;
509 	u_int16_t		rval = 0;
510 	u_int16_t		re8139_reg = 0;
511 
512 	sc = device_get_softc(dev);
513 
514 	if (sc->rl_type == RL_8169) {
515 		rval = re_gmii_readreg(dev, phy, reg);
516 		return (rval);
517 	}
518 
519 	/* Pretend the internal PHY is only at address 0 */
520 	if (phy) {
521 		return (0);
522 	}
523 	switch (reg) {
524 	case MII_BMCR:
525 		re8139_reg = RL_BMCR;
526 		break;
527 	case MII_BMSR:
528 		re8139_reg = RL_BMSR;
529 		break;
530 	case MII_ANAR:
531 		re8139_reg = RL_ANAR;
532 		break;
533 	case MII_ANER:
534 		re8139_reg = RL_ANER;
535 		break;
536 	case MII_ANLPAR:
537 		re8139_reg = RL_LPAR;
538 		break;
539 	case MII_PHYIDR1:
540 	case MII_PHYIDR2:
541 		return (0);
542 	/*
543 	 * Allow the rlphy driver to read the media status
544 	 * register. If we have a link partner which does not
545 	 * support NWAY, this is the register which will tell
546 	 * us the results of parallel detection.
547 	 */
548 	case RL_MEDIASTAT:
549 		rval = CSR_READ_1(sc, RL_MEDIASTAT);
550 		return (rval);
551 	default:
552 		device_printf(sc->rl_dev, "bad phy register\n");
553 		return (0);
554 	}
555 	rval = CSR_READ_2(sc, re8139_reg);
556 	if (sc->rl_type == RL_8139CPLUS && re8139_reg == RL_BMCR) {
557 		/* 8139C+ has different bit layout. */
558 		rval &= ~(BMCR_LOOP | BMCR_ISO);
559 	}
560 	return (rval);
561 }
562 
563 static int
564 re_miibus_writereg(dev, phy, reg, data)
565 	device_t		dev;
566 	int			phy, reg, data;
567 {
568 	struct rl_softc		*sc;
569 	u_int16_t		re8139_reg = 0;
570 	int			rval = 0;
571 
572 	sc = device_get_softc(dev);
573 
574 	if (sc->rl_type == RL_8169) {
575 		rval = re_gmii_writereg(dev, phy, reg, data);
576 		return (rval);
577 	}
578 
579 	/* Pretend the internal PHY is only at address 0 */
580 	if (phy)
581 		return (0);
582 
583 	switch (reg) {
584 	case MII_BMCR:
585 		re8139_reg = RL_BMCR;
586 		if (sc->rl_type == RL_8139CPLUS) {
587 			/* 8139C+ has different bit layout. */
588 			data &= ~(BMCR_LOOP | BMCR_ISO);
589 		}
590 		break;
591 	case MII_BMSR:
592 		re8139_reg = RL_BMSR;
593 		break;
594 	case MII_ANAR:
595 		re8139_reg = RL_ANAR;
596 		break;
597 	case MII_ANER:
598 		re8139_reg = RL_ANER;
599 		break;
600 	case MII_ANLPAR:
601 		re8139_reg = RL_LPAR;
602 		break;
603 	case MII_PHYIDR1:
604 	case MII_PHYIDR2:
605 		return (0);
606 		break;
607 	default:
608 		device_printf(sc->rl_dev, "bad phy register\n");
609 		return (0);
610 	}
611 	CSR_WRITE_2(sc, re8139_reg, data);
612 	return (0);
613 }
614 
615 static void
616 re_miibus_statchg(dev)
617 	device_t		dev;
618 {
619 
620 }
621 
622 /*
623  * Program the 64-bit multicast hash filter.
624  */
625 static void
626 re_setmulti(sc)
627 	struct rl_softc		*sc;
628 {
629 	struct ifnet		*ifp;
630 	int			h = 0;
631 	u_int32_t		hashes[2] = { 0, 0 };
632 	struct ifmultiaddr	*ifma;
633 	u_int32_t		rxfilt;
634 	int			mcnt = 0;
635 	u_int32_t		hwrev;
636 
637 	RL_LOCK_ASSERT(sc);
638 
639 	ifp = sc->rl_ifp;
640 
641 
642 	rxfilt = CSR_READ_4(sc, RL_RXCFG);
643 	rxfilt &= ~(RL_RXCFG_RX_ALLPHYS | RL_RXCFG_RX_MULTI);
644 	if (ifp->if_flags & IFF_ALLMULTI || ifp->if_flags & IFF_PROMISC) {
645 		if (ifp->if_flags & IFF_PROMISC)
646 			rxfilt |= RL_RXCFG_RX_ALLPHYS;
647 		/*
648 		 * Unlike other hardwares, we have to explicitly set
649 		 * RL_RXCFG_RX_MULTI to receive multicast frames in
650 		 * promiscuous mode.
651 		 */
652 		rxfilt |= RL_RXCFG_RX_MULTI;
653 		CSR_WRITE_4(sc, RL_RXCFG, rxfilt);
654 		CSR_WRITE_4(sc, RL_MAR0, 0xFFFFFFFF);
655 		CSR_WRITE_4(sc, RL_MAR4, 0xFFFFFFFF);
656 		return;
657 	}
658 
659 	/* first, zot all the existing hash bits */
660 	CSR_WRITE_4(sc, RL_MAR0, 0);
661 	CSR_WRITE_4(sc, RL_MAR4, 0);
662 
663 	/* now program new ones */
664 	IF_ADDR_LOCK(ifp);
665 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
666 		if (ifma->ifma_addr->sa_family != AF_LINK)
667 			continue;
668 		h = ether_crc32_be(LLADDR((struct sockaddr_dl *)
669 		    ifma->ifma_addr), ETHER_ADDR_LEN) >> 26;
670 		if (h < 32)
671 			hashes[0] |= (1 << h);
672 		else
673 			hashes[1] |= (1 << (h - 32));
674 		mcnt++;
675 	}
676 	IF_ADDR_UNLOCK(ifp);
677 
678 	if (mcnt)
679 		rxfilt |= RL_RXCFG_RX_MULTI;
680 	else
681 		rxfilt &= ~RL_RXCFG_RX_MULTI;
682 
683 	CSR_WRITE_4(sc, RL_RXCFG, rxfilt);
684 
685 	/*
686 	 * For some unfathomable reason, RealTek decided to reverse
687 	 * the order of the multicast hash registers in the PCI Express
688 	 * parts. This means we have to write the hash pattern in reverse
689 	 * order for those devices.
690 	 */
691 
692 	hwrev = CSR_READ_4(sc, RL_TXCFG) & RL_TXCFG_HWREV;
693 
694 	switch (hwrev) {
695 	case RL_HWREV_8100E:
696 	case RL_HWREV_8101E:
697 	case RL_HWREV_8168_SPIN1:
698 	case RL_HWREV_8168_SPIN2:
699 	case RL_HWREV_8168_SPIN3:
700 		CSR_WRITE_4(sc, RL_MAR0, bswap32(hashes[1]));
701 		CSR_WRITE_4(sc, RL_MAR4, bswap32(hashes[0]));
702 		break;
703 	default:
704 		CSR_WRITE_4(sc, RL_MAR0, hashes[0]);
705 		CSR_WRITE_4(sc, RL_MAR4, hashes[1]);
706 		break;
707 	}
708 }
709 
710 static void
711 re_reset(sc)
712 	struct rl_softc		*sc;
713 {
714 	register int		i;
715 
716 	RL_LOCK_ASSERT(sc);
717 
718 	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RESET);
719 
720 	for (i = 0; i < RL_TIMEOUT; i++) {
721 		DELAY(10);
722 		if (!(CSR_READ_1(sc, RL_COMMAND) & RL_CMD_RESET))
723 			break;
724 	}
725 	if (i == RL_TIMEOUT)
726 		device_printf(sc->rl_dev, "reset never completed!\n");
727 
728 	CSR_WRITE_1(sc, 0x82, 1);
729 }
730 
731 #ifdef RE_DIAG
732 
733 /*
734  * The following routine is designed to test for a defect on some
735  * 32-bit 8169 cards. Some of these NICs have the REQ64# and ACK64#
736  * lines connected to the bus, however for a 32-bit only card, they
737  * should be pulled high. The result of this defect is that the
738  * NIC will not work right if you plug it into a 64-bit slot: DMA
739  * operations will be done with 64-bit transfers, which will fail
740  * because the 64-bit data lines aren't connected.
741  *
742  * There's no way to work around this (short of talking a soldering
743  * iron to the board), however we can detect it. The method we use
744  * here is to put the NIC into digital loopback mode, set the receiver
745  * to promiscuous mode, and then try to send a frame. We then compare
746  * the frame data we sent to what was received. If the data matches,
747  * then the NIC is working correctly, otherwise we know the user has
748  * a defective NIC which has been mistakenly plugged into a 64-bit PCI
749  * slot. In the latter case, there's no way the NIC can work correctly,
750  * so we print out a message on the console and abort the device attach.
751  */
752 
753 static int
754 re_diag(sc)
755 	struct rl_softc		*sc;
756 {
757 	struct ifnet		*ifp = sc->rl_ifp;
758 	struct mbuf		*m0;
759 	struct ether_header	*eh;
760 	struct rl_desc		*cur_rx;
761 	u_int16_t		status;
762 	u_int32_t		rxstat;
763 	int			total_len, i, error = 0, phyaddr;
764 	u_int8_t		dst[] = { 0x00, 'h', 'e', 'l', 'l', 'o' };
765 	u_int8_t		src[] = { 0x00, 'w', 'o', 'r', 'l', 'd' };
766 
767 	/* Allocate a single mbuf */
768 	MGETHDR(m0, M_DONTWAIT, MT_DATA);
769 	if (m0 == NULL)
770 		return (ENOBUFS);
771 
772 	RL_LOCK(sc);
773 
774 	/*
775 	 * Initialize the NIC in test mode. This sets the chip up
776 	 * so that it can send and receive frames, but performs the
777 	 * following special functions:
778 	 * - Puts receiver in promiscuous mode
779 	 * - Enables digital loopback mode
780 	 * - Leaves interrupts turned off
781 	 */
782 
783 	ifp->if_flags |= IFF_PROMISC;
784 	sc->rl_testmode = 1;
785 	re_reset(sc);
786 	re_init_locked(sc);
787 	sc->rl_link = 1;
788 	if (sc->rl_type == RL_8169)
789 		phyaddr = 1;
790 	else
791 		phyaddr = 0;
792 
793 	re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_RESET);
794 	for (i = 0; i < RL_TIMEOUT; i++) {
795 		status = re_miibus_readreg(sc->rl_dev, phyaddr, MII_BMCR);
796 		if (!(status & BMCR_RESET))
797 			break;
798 	}
799 
800 	re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_LOOP);
801 	CSR_WRITE_2(sc, RL_ISR, RL_INTRS);
802 
803 	DELAY(100000);
804 
805 	/* Put some data in the mbuf */
806 
807 	eh = mtod(m0, struct ether_header *);
808 	bcopy ((char *)&dst, eh->ether_dhost, ETHER_ADDR_LEN);
809 	bcopy ((char *)&src, eh->ether_shost, ETHER_ADDR_LEN);
810 	eh->ether_type = htons(ETHERTYPE_IP);
811 	m0->m_pkthdr.len = m0->m_len = ETHER_MIN_LEN - ETHER_CRC_LEN;
812 
813 	/*
814 	 * Queue the packet, start transmission.
815 	 * Note: IF_HANDOFF() ultimately calls re_start() for us.
816 	 */
817 
818 	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
819 	RL_UNLOCK(sc);
820 	/* XXX: re_diag must not be called when in ALTQ mode */
821 	IF_HANDOFF(&ifp->if_snd, m0, ifp);
822 	RL_LOCK(sc);
823 	m0 = NULL;
824 
825 	/* Wait for it to propagate through the chip */
826 
827 	DELAY(100000);
828 	for (i = 0; i < RL_TIMEOUT; i++) {
829 		status = CSR_READ_2(sc, RL_ISR);
830 		CSR_WRITE_2(sc, RL_ISR, status);
831 		if ((status & (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK)) ==
832 		    (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK))
833 			break;
834 		DELAY(10);
835 	}
836 
837 	if (i == RL_TIMEOUT) {
838 		device_printf(sc->rl_dev,
839 		    "diagnostic failed, failed to receive packet in"
840 		    " loopback mode\n");
841 		error = EIO;
842 		goto done;
843 	}
844 
845 	/*
846 	 * The packet should have been dumped into the first
847 	 * entry in the RX DMA ring. Grab it from there.
848 	 */
849 
850 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
851 	    sc->rl_ldata.rl_rx_list_map,
852 	    BUS_DMASYNC_POSTREAD);
853 	bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag,
854 	    sc->rl_ldata.rl_rx_desc[0].rx_dmamap,
855 	    BUS_DMASYNC_POSTREAD);
856 	bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag,
857 	    sc->rl_ldata.rl_rx_desc[0].rx_dmamap);
858 
859 	m0 = sc->rl_ldata.rl_rx_desc[0].rx_m;
860 	sc->rl_ldata.rl_rx_desc[0].rx_m = NULL;
861 	eh = mtod(m0, struct ether_header *);
862 
863 	cur_rx = &sc->rl_ldata.rl_rx_list[0];
864 	total_len = RL_RXBYTES(cur_rx);
865 	rxstat = le32toh(cur_rx->rl_cmdstat);
866 
867 	if (total_len != ETHER_MIN_LEN) {
868 		device_printf(sc->rl_dev,
869 		    "diagnostic failed, received short packet\n");
870 		error = EIO;
871 		goto done;
872 	}
873 
874 	/* Test that the received packet data matches what we sent. */
875 
876 	if (bcmp((char *)&eh->ether_dhost, (char *)&dst, ETHER_ADDR_LEN) ||
877 	    bcmp((char *)&eh->ether_shost, (char *)&src, ETHER_ADDR_LEN) ||
878 	    ntohs(eh->ether_type) != ETHERTYPE_IP) {
879 		device_printf(sc->rl_dev, "WARNING, DMA FAILURE!\n");
880 		device_printf(sc->rl_dev, "expected TX data: %6D/%6D/0x%x\n",
881 		    dst, ":", src, ":", ETHERTYPE_IP);
882 		device_printf(sc->rl_dev, "received RX data: %6D/%6D/0x%x\n",
883 		    eh->ether_dhost, ":",  eh->ether_shost, ":",
884 		    ntohs(eh->ether_type));
885 		device_printf(sc->rl_dev, "You may have a defective 32-bit "
886 		    "NIC plugged into a 64-bit PCI slot.\n");
887 		device_printf(sc->rl_dev, "Please re-install the NIC in a "
888 		    "32-bit slot for proper operation.\n");
889 		device_printf(sc->rl_dev, "Read the re(4) man page for more "
890 		    "details.\n");
891 		error = EIO;
892 	}
893 
894 done:
895 	/* Turn interface off, release resources */
896 
897 	sc->rl_testmode = 0;
898 	sc->rl_link = 0;
899 	ifp->if_flags &= ~IFF_PROMISC;
900 	re_stop(sc);
901 	if (m0 != NULL)
902 		m_freem(m0);
903 
904 	RL_UNLOCK(sc);
905 
906 	return (error);
907 }
908 
909 #endif
910 
911 /*
912  * Probe for a RealTek 8139C+/8169/8110 chip. Check the PCI vendor and device
913  * IDs against our list and return a device name if we find a match.
914  */
915 static int
916 re_probe(dev)
917 	device_t		dev;
918 {
919 	struct rl_type		*t;
920 	uint16_t		devid, vendor;
921 	uint16_t		revid, sdevid;
922 	int			i;
923 
924 	vendor = pci_get_vendor(dev);
925 	devid = pci_get_device(dev);
926 	revid = pci_get_revid(dev);
927 	sdevid = pci_get_subdevice(dev);
928 
929 	if (vendor == LINKSYS_VENDORID && devid == LINKSYS_DEVICEID_EG1032) {
930 		if (sdevid != LINKSYS_SUBDEVICE_EG1032_REV3) {
931 			/*
932 			 * Only attach to rev. 3 of the Linksys EG1032 adapter.
933 			 * Rev. 2 is supported by sk(4).
934 			 */
935 			return (ENXIO);
936 		}
937 	}
938 
939 	if (vendor == RT_VENDORID && devid == RT_DEVICEID_8139) {
940 		if (revid != 0x20) {
941 			/* 8139, let rl(4) take care of this device. */
942 			return (ENXIO);
943 		}
944 	}
945 
946 	t = re_devs;
947 	for (i = 0; i < sizeof(re_devs) / sizeof(re_devs[0]); i++, t++) {
948 		if (vendor == t->rl_vid && devid == t->rl_did) {
949 			device_set_desc(dev, t->rl_name);
950 			return (BUS_PROBE_DEFAULT);
951 		}
952 	}
953 
954 	return (ENXIO);
955 }
956 
957 /*
958  * Map a single buffer address.
959  */
960 
961 static void
962 re_dma_map_addr(arg, segs, nseg, error)
963 	void			*arg;
964 	bus_dma_segment_t	*segs;
965 	int			nseg;
966 	int			error;
967 {
968 	bus_addr_t		*addr;
969 
970 	if (error)
971 		return;
972 
973 	KASSERT(nseg == 1, ("too many DMA segments, %d should be 1", nseg));
974 	addr = arg;
975 	*addr = segs->ds_addr;
976 }
977 
978 static int
979 re_allocmem(dev, sc)
980 	device_t		dev;
981 	struct rl_softc		*sc;
982 {
983 	bus_size_t		rx_list_size, tx_list_size;
984 	int			error;
985 	int			i;
986 
987 	rx_list_size = sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc);
988 	tx_list_size = sc->rl_ldata.rl_tx_desc_cnt * sizeof(struct rl_desc);
989 
990 	/*
991 	 * Allocate the parent bus DMA tag appropriate for PCI.
992 	 * In order to use DAC, RL_CPLUSCMD_PCI_DAC bit of RL_CPLUS_CMD
993 	 * register should be set. However some RealTek chips are known
994 	 * to be buggy on DAC handling, therefore disable DAC by limiting
995 	 * DMA address space to 32bit. PCIe variants of RealTek chips
996 	 * may not have the limitation but I took safer path.
997 	 */
998 	error = bus_dma_tag_create(bus_get_dma_tag(dev), 1, 0,
999 	    BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL, NULL,
1000 	    BUS_SPACE_MAXSIZE_32BIT, 0, BUS_SPACE_MAXSIZE_32BIT, 0,
1001 	    NULL, NULL, &sc->rl_parent_tag);
1002 	if (error) {
1003 		device_printf(dev, "could not allocate parent DMA tag\n");
1004 		return (error);
1005 	}
1006 
1007 	/*
1008 	 * Allocate map for TX mbufs.
1009 	 */
1010 	error = bus_dma_tag_create(sc->rl_parent_tag, 1, 0,
1011 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL,
1012 	    NULL, MCLBYTES * RL_NTXSEGS, RL_NTXSEGS, 4096, 0,
1013 	    NULL, NULL, &sc->rl_ldata.rl_tx_mtag);
1014 	if (error) {
1015 		device_printf(dev, "could not allocate TX DMA tag\n");
1016 		return (error);
1017 	}
1018 
1019 	/*
1020 	 * Allocate map for RX mbufs.
1021 	 */
1022 
1023 	error = bus_dma_tag_create(sc->rl_parent_tag, sizeof(uint64_t), 0,
1024 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
1025 	    MCLBYTES, 1, MCLBYTES, 0, NULL, NULL, &sc->rl_ldata.rl_rx_mtag);
1026 	if (error) {
1027 		device_printf(dev, "could not allocate RX DMA tag\n");
1028 		return (error);
1029 	}
1030 
1031 	/*
1032 	 * Allocate map for TX descriptor list.
1033 	 */
1034 	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
1035 	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1036 	    NULL, tx_list_size, 1, tx_list_size, 0,
1037 	    NULL, NULL, &sc->rl_ldata.rl_tx_list_tag);
1038 	if (error) {
1039 		device_printf(dev, "could not allocate TX DMA ring tag\n");
1040 		return (error);
1041 	}
1042 
1043 	/* Allocate DMA'able memory for the TX ring */
1044 
1045 	error = bus_dmamem_alloc(sc->rl_ldata.rl_tx_list_tag,
1046 	    (void **)&sc->rl_ldata.rl_tx_list,
1047 	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1048 	    &sc->rl_ldata.rl_tx_list_map);
1049 	if (error) {
1050 		device_printf(dev, "could not allocate TX DMA ring\n");
1051 		return (error);
1052 	}
1053 
1054 	/* Load the map for the TX ring. */
1055 
1056 	sc->rl_ldata.rl_tx_list_addr = 0;
1057 	error = bus_dmamap_load(sc->rl_ldata.rl_tx_list_tag,
1058 	     sc->rl_ldata.rl_tx_list_map, sc->rl_ldata.rl_tx_list,
1059 	     tx_list_size, re_dma_map_addr,
1060 	     &sc->rl_ldata.rl_tx_list_addr, BUS_DMA_NOWAIT);
1061 	if (error != 0 || sc->rl_ldata.rl_tx_list_addr == 0) {
1062 		device_printf(dev, "could not load TX DMA ring\n");
1063 		return (ENOMEM);
1064 	}
1065 
1066 	/* Create DMA maps for TX buffers */
1067 
1068 	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
1069 		error = bus_dmamap_create(sc->rl_ldata.rl_tx_mtag, 0,
1070 		    &sc->rl_ldata.rl_tx_desc[i].tx_dmamap);
1071 		if (error) {
1072 			device_printf(dev, "could not create DMA map for TX\n");
1073 			return (error);
1074 		}
1075 	}
1076 
1077 	/*
1078 	 * Allocate map for RX descriptor list.
1079 	 */
1080 	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
1081 	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1082 	    NULL, rx_list_size, 1, rx_list_size, 0,
1083 	    NULL, NULL, &sc->rl_ldata.rl_rx_list_tag);
1084 	if (error) {
1085 		device_printf(dev, "could not create RX DMA ring tag\n");
1086 		return (error);
1087 	}
1088 
1089 	/* Allocate DMA'able memory for the RX ring */
1090 
1091 	error = bus_dmamem_alloc(sc->rl_ldata.rl_rx_list_tag,
1092 	    (void **)&sc->rl_ldata.rl_rx_list,
1093 	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1094 	    &sc->rl_ldata.rl_rx_list_map);
1095 	if (error) {
1096 		device_printf(dev, "could not allocate RX DMA ring\n");
1097 		return (error);
1098 	}
1099 
1100 	/* Load the map for the RX ring. */
1101 
1102 	sc->rl_ldata.rl_rx_list_addr = 0;
1103 	error = bus_dmamap_load(sc->rl_ldata.rl_rx_list_tag,
1104 	     sc->rl_ldata.rl_rx_list_map, sc->rl_ldata.rl_rx_list,
1105 	     rx_list_size, re_dma_map_addr,
1106 	     &sc->rl_ldata.rl_rx_list_addr, BUS_DMA_NOWAIT);
1107 	if (error != 0 || sc->rl_ldata.rl_rx_list_addr == 0) {
1108 		device_printf(dev, "could not load RX DMA ring\n");
1109 		return (ENOMEM);
1110 	}
1111 
1112 	/* Create DMA maps for RX buffers */
1113 
1114 	error = bus_dmamap_create(sc->rl_ldata.rl_rx_mtag, 0,
1115 	    &sc->rl_ldata.rl_rx_sparemap);
1116 	if (error) {
1117 		device_printf(dev, "could not create spare DMA map for RX\n");
1118 		return (error);
1119 	}
1120 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1121 		error = bus_dmamap_create(sc->rl_ldata.rl_rx_mtag, 0,
1122 		    &sc->rl_ldata.rl_rx_desc[i].rx_dmamap);
1123 		if (error) {
1124 			device_printf(dev, "could not create DMA map for RX\n");
1125 			return (error);
1126 		}
1127 	}
1128 
1129 	return (0);
1130 }
1131 
1132 /*
1133  * Attach the interface. Allocate softc structures, do ifmedia
1134  * setup and ethernet/BPF attach.
1135  */
1136 static int
1137 re_attach(dev)
1138 	device_t		dev;
1139 {
1140 	u_char			eaddr[ETHER_ADDR_LEN];
1141 	u_int16_t		as[ETHER_ADDR_LEN / 2];
1142 	struct rl_softc		*sc;
1143 	struct ifnet		*ifp;
1144 	struct rl_hwrev		*hw_rev;
1145 	int			hwrev;
1146 	u_int16_t		re_did = 0;
1147 	int			error = 0, rid, i;
1148 	int			msic, reg;
1149 	uint8_t			cfg;
1150 
1151 	sc = device_get_softc(dev);
1152 	sc->rl_dev = dev;
1153 
1154 	mtx_init(&sc->rl_mtx, device_get_nameunit(dev), MTX_NETWORK_LOCK,
1155 	    MTX_DEF);
1156 	callout_init_mtx(&sc->rl_stat_callout, &sc->rl_mtx, 0);
1157 
1158 	/*
1159 	 * Map control/status registers.
1160 	 */
1161 	pci_enable_busmaster(dev);
1162 
1163 	rid = RL_RID;
1164 	sc->rl_res = bus_alloc_resource_any(dev, RL_RES, &rid,
1165 	    RF_ACTIVE);
1166 
1167 	if (sc->rl_res == NULL) {
1168 		device_printf(dev, "couldn't map ports/memory\n");
1169 		error = ENXIO;
1170 		goto fail;
1171 	}
1172 
1173 	sc->rl_btag = rman_get_bustag(sc->rl_res);
1174 	sc->rl_bhandle = rman_get_bushandle(sc->rl_res);
1175 
1176 	msic = 0;
1177 	if (pci_find_extcap(dev, PCIY_EXPRESS, &reg) == 0) {
1178 		msic = pci_msi_count(dev);
1179 		if (bootverbose)
1180 			device_printf(dev, "MSI count : %d\n", msic);
1181 	}
1182 	if (msic == RL_MSI_MESSAGES  && msi_disable == 0) {
1183 		if (pci_alloc_msi(dev, &msic) == 0) {
1184 			if (msic == RL_MSI_MESSAGES) {
1185 				device_printf(dev, "Using %d MSI messages\n",
1186 				    msic);
1187 				sc->rl_msi = 1;
1188 			} else
1189 				pci_release_msi(dev);
1190 		}
1191 	}
1192 
1193 	/* For MSI capable hardwares, explicitily set/clear MSI enable bit. */
1194 	if (msic != 0) {
1195 		CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1196 		cfg = CSR_READ_1(sc, RL_CFG2);
1197 		if (sc->rl_msi != 0)
1198 			cfg |= RL_CFG2_MSI;
1199 		else
1200 			cfg &= ~RL_CFG2_MSI;
1201 		CSR_WRITE_1(sc, RL_CFG2, cfg);
1202 		CSR_WRITE_1(sc, RL_EECMD, 0);
1203 	}
1204 
1205 	/* Allocate interrupt */
1206 	if (sc->rl_msi == 0) {
1207 		rid = 0;
1208 		sc->rl_irq[0] = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
1209 		    RF_SHAREABLE | RF_ACTIVE);
1210 		if (sc->rl_irq[0] == NULL) {
1211 			device_printf(dev, "couldn't allocate IRQ resources\n");
1212 			error = ENXIO;
1213 			goto fail;
1214 		}
1215 	} else {
1216 		for (i = 0, rid = 1; i < RL_MSI_MESSAGES; i++, rid++) {
1217 			sc->rl_irq[i] = bus_alloc_resource_any(dev,
1218 			    SYS_RES_IRQ, &rid, RF_ACTIVE);
1219 			if (sc->rl_irq[i] == NULL) {
1220 				device_printf(dev,
1221 				    "couldn't llocate IRQ resources for "
1222 				    "message %d\n", rid);
1223 				error = ENXIO;
1224 				goto fail;
1225 			}
1226 		}
1227 	}
1228 
1229 	/* Reset the adapter. */
1230 	RL_LOCK(sc);
1231 	re_reset(sc);
1232 	RL_UNLOCK(sc);
1233 
1234 	hw_rev = re_hwrevs;
1235 	hwrev = CSR_READ_4(sc, RL_TXCFG) & RL_TXCFG_HWREV;
1236 	while (hw_rev->rl_desc != NULL) {
1237 		if (hw_rev->rl_rev == hwrev) {
1238 			sc->rl_type = hw_rev->rl_type;
1239 			break;
1240 		}
1241 		hw_rev++;
1242 	}
1243 	if (hw_rev->rl_desc == NULL) {
1244 		device_printf(dev, "Unknown H/W revision: %08x\n", hwrev);
1245 		error = ENXIO;
1246 		goto fail;
1247 	}
1248 
1249 	sc->rl_eewidth = RL_9356_ADDR_LEN;
1250 	re_read_eeprom(sc, (caddr_t)&re_did, 0, 1);
1251 	if (re_did != 0x8129)
1252 	        sc->rl_eewidth = RL_9346_ADDR_LEN;
1253 
1254 	/*
1255 	 * Get station address from the EEPROM.
1256 	 */
1257 	re_read_eeprom(sc, (caddr_t)as, RL_EE_EADDR, 3);
1258 	for (i = 0; i < ETHER_ADDR_LEN / 2; i++)
1259 		as[i] = le16toh(as[i]);
1260 	bcopy(as, eaddr, sizeof(eaddr));
1261 
1262 	if (sc->rl_type == RL_8169) {
1263 		/* Set RX length mask and number of descriptors. */
1264 		sc->rl_rxlenmask = RL_RDESC_STAT_GFRAGLEN;
1265 		sc->rl_txstart = RL_GTXSTART;
1266 		sc->rl_ldata.rl_tx_desc_cnt = RL_8169_TX_DESC_CNT;
1267 		sc->rl_ldata.rl_rx_desc_cnt = RL_8169_RX_DESC_CNT;
1268 	} else {
1269 		/* Set RX length mask and number of descriptors. */
1270 		sc->rl_rxlenmask = RL_RDESC_STAT_FRAGLEN;
1271 		sc->rl_txstart = RL_TXSTART;
1272 		sc->rl_ldata.rl_tx_desc_cnt = RL_8139_TX_DESC_CNT;
1273 		sc->rl_ldata.rl_rx_desc_cnt = RL_8139_RX_DESC_CNT;
1274 	}
1275 	if (hw_rev->rl_desc == NULL) {
1276 		device_printf(dev, "Unsupported revision : 0x%08x\n", hwrev);
1277 		error = ENXIO;
1278 		goto fail;
1279 	}
1280 
1281 	error = re_allocmem(dev, sc);
1282 	if (error)
1283 		goto fail;
1284 
1285 	ifp = sc->rl_ifp = if_alloc(IFT_ETHER);
1286 	if (ifp == NULL) {
1287 		device_printf(dev, "can not if_alloc()\n");
1288 		error = ENOSPC;
1289 		goto fail;
1290 	}
1291 
1292 	/* Do MII setup */
1293 	if (mii_phy_probe(dev, &sc->rl_miibus,
1294 	    re_ifmedia_upd, re_ifmedia_sts)) {
1295 		device_printf(dev, "MII without any phy!\n");
1296 		error = ENXIO;
1297 		goto fail;
1298 	}
1299 
1300 	/* Take PHY out of power down mode. */
1301 	if (sc->rl_type == RL_8169) {
1302 		uint32_t rev;
1303 
1304 		rev = CSR_READ_4(sc, RL_TXCFG);
1305 		/* HWVERID 0, 1 and 2 :  bit26-30, bit23 */
1306 		rev &= 0x7c800000;
1307 		if (rev != 0) {
1308 			/* RTL8169S single chip */
1309 			switch (rev) {
1310 			case RL_HWREV_8169_8110SB:
1311 			case RL_HWREV_8169_8110SC:
1312 			case RL_HWREV_8168_SPIN2:
1313 			case RL_HWREV_8168_SPIN3:
1314 				re_gmii_writereg(dev, 1, 0x1f, 0);
1315 				re_gmii_writereg(dev, 1, 0x0e, 0);
1316 				break;
1317 			default:
1318 				break;
1319 			}
1320 		}
1321 	}
1322 
1323 	ifp->if_softc = sc;
1324 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1325 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1326 	ifp->if_ioctl = re_ioctl;
1327 	ifp->if_start = re_start;
1328 	ifp->if_hwassist = RE_CSUM_FEATURES | CSUM_TSO;
1329 	ifp->if_capabilities = IFCAP_HWCSUM | IFCAP_TSO4;
1330 	ifp->if_capenable = ifp->if_capabilities;
1331 	ifp->if_init = re_init;
1332 	IFQ_SET_MAXLEN(&ifp->if_snd, RL_IFQ_MAXLEN);
1333 	ifp->if_snd.ifq_drv_maxlen = RL_IFQ_MAXLEN;
1334 	IFQ_SET_READY(&ifp->if_snd);
1335 
1336 	TASK_INIT(&sc->rl_txtask, 1, re_tx_task, ifp);
1337 	TASK_INIT(&sc->rl_inttask, 0, re_int_task, sc);
1338 
1339 	/*
1340 	 * Call MI attach routine.
1341 	 */
1342 	ether_ifattach(ifp, eaddr);
1343 
1344 	/* VLAN capability setup */
1345 	ifp->if_capabilities |= IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING;
1346 	if (ifp->if_capabilities & IFCAP_HWCSUM)
1347 		ifp->if_capabilities |= IFCAP_VLAN_HWCSUM;
1348 	/* Enable WOL if PM is supported. */
1349 	if (pci_find_extcap(sc->rl_dev, PCIY_PMG, &reg) == 0)
1350 		ifp->if_capabilities |= IFCAP_WOL;
1351 	ifp->if_capenable = ifp->if_capabilities;
1352 #ifdef DEVICE_POLLING
1353 	ifp->if_capabilities |= IFCAP_POLLING;
1354 #endif
1355 	/*
1356 	 * Tell the upper layer(s) we support long frames.
1357 	 * Must appear after the call to ether_ifattach() because
1358 	 * ether_ifattach() sets ifi_hdrlen to the default value.
1359 	 */
1360 	ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
1361 
1362 #ifdef RE_DIAG
1363 	/*
1364 	 * Perform hardware diagnostic on the original RTL8169.
1365 	 * Some 32-bit cards were incorrectly wired and would
1366 	 * malfunction if plugged into a 64-bit slot.
1367 	 */
1368 
1369 	if (hwrev == RL_HWREV_8169) {
1370 		error = re_diag(sc);
1371 		if (error) {
1372 			device_printf(dev,
1373 		    	"attach aborted due to hardware diag failure\n");
1374 			ether_ifdetach(ifp);
1375 			goto fail;
1376 		}
1377 	}
1378 #endif
1379 
1380 	/* Hook interrupt last to avoid having to lock softc */
1381 	if (sc->rl_msi == 0)
1382 		error = bus_setup_intr(dev, sc->rl_irq[0],
1383 		    INTR_TYPE_NET | INTR_MPSAFE, re_intr, NULL, sc,
1384 		    &sc->rl_intrhand[0]);
1385 	else {
1386 		for (i = 0; i < RL_MSI_MESSAGES; i++) {
1387 			error = bus_setup_intr(dev, sc->rl_irq[i],
1388 			    INTR_TYPE_NET | INTR_MPSAFE, re_intr, NULL, sc,
1389 		    	    &sc->rl_intrhand[i]);
1390 			if (error != 0)
1391 				break;
1392 		}
1393 	}
1394 	if (error) {
1395 		device_printf(dev, "couldn't set up irq\n");
1396 		ether_ifdetach(ifp);
1397 	}
1398 
1399 fail:
1400 
1401 	if (error)
1402 		re_detach(dev);
1403 
1404 	return (error);
1405 }
1406 
1407 /*
1408  * Shutdown hardware and free up resources. This can be called any
1409  * time after the mutex has been initialized. It is called in both
1410  * the error case in attach and the normal detach case so it needs
1411  * to be careful about only freeing resources that have actually been
1412  * allocated.
1413  */
1414 static int
1415 re_detach(dev)
1416 	device_t		dev;
1417 {
1418 	struct rl_softc		*sc;
1419 	struct ifnet		*ifp;
1420 	int			i, rid;
1421 
1422 	sc = device_get_softc(dev);
1423 	ifp = sc->rl_ifp;
1424 	KASSERT(mtx_initialized(&sc->rl_mtx), ("re mutex not initialized"));
1425 
1426 #ifdef DEVICE_POLLING
1427 	if (ifp->if_capenable & IFCAP_POLLING)
1428 		ether_poll_deregister(ifp);
1429 #endif
1430 	/* These should only be active if attach succeeded */
1431 	if (device_is_attached(dev)) {
1432 		RL_LOCK(sc);
1433 #if 0
1434 		sc->suspended = 1;
1435 #endif
1436 		re_stop(sc);
1437 		RL_UNLOCK(sc);
1438 		callout_drain(&sc->rl_stat_callout);
1439 		taskqueue_drain(taskqueue_fast, &sc->rl_inttask);
1440 		taskqueue_drain(taskqueue_fast, &sc->rl_txtask);
1441 		/*
1442 		 * Force off the IFF_UP flag here, in case someone
1443 		 * still had a BPF descriptor attached to this
1444 		 * interface. If they do, ether_ifdetach() will cause
1445 		 * the BPF code to try and clear the promisc mode
1446 		 * flag, which will bubble down to re_ioctl(),
1447 		 * which will try to call re_init() again. This will
1448 		 * turn the NIC back on and restart the MII ticker,
1449 		 * which will panic the system when the kernel tries
1450 		 * to invoke the re_tick() function that isn't there
1451 		 * anymore.
1452 		 */
1453 		ifp->if_flags &= ~IFF_UP;
1454 		ether_ifdetach(ifp);
1455 	}
1456 	if (sc->rl_miibus)
1457 		device_delete_child(dev, sc->rl_miibus);
1458 	bus_generic_detach(dev);
1459 
1460 	/*
1461 	 * The rest is resource deallocation, so we should already be
1462 	 * stopped here.
1463 	 */
1464 
1465 	for (i = 0; i < RL_MSI_MESSAGES; i++) {
1466 		if (sc->rl_intrhand[i] != NULL) {
1467 			bus_teardown_intr(dev, sc->rl_irq[i],
1468 			    sc->rl_intrhand[i]);
1469 			sc->rl_intrhand[i] = NULL;
1470 		}
1471 	}
1472 	if (ifp != NULL)
1473 		if_free(ifp);
1474 	if (sc->rl_msi == 0) {
1475 		if (sc->rl_irq[0] != NULL) {
1476 			bus_release_resource(dev, SYS_RES_IRQ, 0,
1477 			    sc->rl_irq[0]);
1478 			sc->rl_irq[0] = NULL;
1479 		}
1480 	} else {
1481 		for (i = 0, rid = 1; i < RL_MSI_MESSAGES; i++, rid++) {
1482 			if (sc->rl_irq[i] != NULL) {
1483 				bus_release_resource(dev, SYS_RES_IRQ, rid,
1484 				    sc->rl_irq[i]);
1485 				sc->rl_irq[i] = NULL;
1486 			}
1487 		}
1488 		pci_release_msi(dev);
1489 	}
1490 	if (sc->rl_res)
1491 		bus_release_resource(dev, RL_RES, RL_RID, sc->rl_res);
1492 
1493 	/* Unload and free the RX DMA ring memory and map */
1494 
1495 	if (sc->rl_ldata.rl_rx_list_tag) {
1496 		bus_dmamap_unload(sc->rl_ldata.rl_rx_list_tag,
1497 		    sc->rl_ldata.rl_rx_list_map);
1498 		bus_dmamem_free(sc->rl_ldata.rl_rx_list_tag,
1499 		    sc->rl_ldata.rl_rx_list,
1500 		    sc->rl_ldata.rl_rx_list_map);
1501 		bus_dma_tag_destroy(sc->rl_ldata.rl_rx_list_tag);
1502 	}
1503 
1504 	/* Unload and free the TX DMA ring memory and map */
1505 
1506 	if (sc->rl_ldata.rl_tx_list_tag) {
1507 		bus_dmamap_unload(sc->rl_ldata.rl_tx_list_tag,
1508 		    sc->rl_ldata.rl_tx_list_map);
1509 		bus_dmamem_free(sc->rl_ldata.rl_tx_list_tag,
1510 		    sc->rl_ldata.rl_tx_list,
1511 		    sc->rl_ldata.rl_tx_list_map);
1512 		bus_dma_tag_destroy(sc->rl_ldata.rl_tx_list_tag);
1513 	}
1514 
1515 	/* Destroy all the RX and TX buffer maps */
1516 
1517 	if (sc->rl_ldata.rl_tx_mtag) {
1518 		for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++)
1519 			bus_dmamap_destroy(sc->rl_ldata.rl_tx_mtag,
1520 			    sc->rl_ldata.rl_tx_desc[i].tx_dmamap);
1521 		bus_dma_tag_destroy(sc->rl_ldata.rl_tx_mtag);
1522 	}
1523 	if (sc->rl_ldata.rl_rx_mtag) {
1524 		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++)
1525 			bus_dmamap_destroy(sc->rl_ldata.rl_rx_mtag,
1526 			    sc->rl_ldata.rl_rx_desc[i].rx_dmamap);
1527 		if (sc->rl_ldata.rl_rx_sparemap)
1528 			bus_dmamap_destroy(sc->rl_ldata.rl_rx_mtag,
1529 			    sc->rl_ldata.rl_rx_sparemap);
1530 		bus_dma_tag_destroy(sc->rl_ldata.rl_rx_mtag);
1531 	}
1532 
1533 	/* Unload and free the stats buffer and map */
1534 
1535 	if (sc->rl_ldata.rl_stag) {
1536 		bus_dmamap_unload(sc->rl_ldata.rl_stag,
1537 		    sc->rl_ldata.rl_rx_list_map);
1538 		bus_dmamem_free(sc->rl_ldata.rl_stag,
1539 		    sc->rl_ldata.rl_stats,
1540 		    sc->rl_ldata.rl_smap);
1541 		bus_dma_tag_destroy(sc->rl_ldata.rl_stag);
1542 	}
1543 
1544 	if (sc->rl_parent_tag)
1545 		bus_dma_tag_destroy(sc->rl_parent_tag);
1546 
1547 	mtx_destroy(&sc->rl_mtx);
1548 
1549 	return (0);
1550 }
1551 
1552 static __inline void
1553 re_discard_rxbuf(sc, idx)
1554 	struct rl_softc		*sc;
1555 	int			idx;
1556 {
1557 	struct rl_desc		*desc;
1558 	struct rl_rxdesc	*rxd;
1559 	uint32_t		cmdstat;
1560 
1561 	rxd = &sc->rl_ldata.rl_rx_desc[idx];
1562 	desc = &sc->rl_ldata.rl_rx_list[idx];
1563 	desc->rl_vlanctl = 0;
1564 	cmdstat = rxd->rx_size;
1565 	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
1566 		cmdstat |= RL_RDESC_CMD_EOR;
1567 	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
1568 }
1569 
1570 static int
1571 re_newbuf(sc, idx)
1572 	struct rl_softc		*sc;
1573 	int			idx;
1574 {
1575 	struct mbuf		*m;
1576 	struct rl_rxdesc	*rxd;
1577 	bus_dma_segment_t	segs[1];
1578 	bus_dmamap_t		map;
1579 	struct rl_desc		*desc;
1580 	uint32_t		cmdstat;
1581 	int			error, nsegs;
1582 
1583 	m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
1584 	if (m == NULL)
1585 		return (ENOBUFS);
1586 
1587 	m->m_len = m->m_pkthdr.len = MCLBYTES;
1588 #ifdef RE_FIXUP_RX
1589 	/*
1590 	 * This is part of an evil trick to deal with non-x86 platforms.
1591 	 * The RealTek chip requires RX buffers to be aligned on 64-bit
1592 	 * boundaries, but that will hose non-x86 machines. To get around
1593 	 * this, we leave some empty space at the start of each buffer
1594 	 * and for non-x86 hosts, we copy the buffer back six bytes
1595 	 * to achieve word alignment. This is slightly more efficient
1596 	 * than allocating a new buffer, copying the contents, and
1597 	 * discarding the old buffer.
1598 	 */
1599 	m_adj(m, RE_ETHER_ALIGN);
1600 #endif
1601 	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_rx_mtag,
1602 	    sc->rl_ldata.rl_rx_sparemap, m, segs, &nsegs, BUS_DMA_NOWAIT);
1603 	if (error != 0) {
1604 		m_freem(m);
1605 		return (ENOBUFS);
1606 	}
1607 	KASSERT(nsegs == 1, ("%s: %d segment returned!", __func__, nsegs));
1608 
1609 	rxd = &sc->rl_ldata.rl_rx_desc[idx];
1610 	if (rxd->rx_m != NULL) {
1611 		bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap,
1612 		    BUS_DMASYNC_POSTREAD);
1613 		bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap);
1614 	}
1615 
1616 	rxd->rx_m = m;
1617 	map = rxd->rx_dmamap;
1618 	rxd->rx_dmamap = sc->rl_ldata.rl_rx_sparemap;
1619 	rxd->rx_size = segs[0].ds_len;
1620 	sc->rl_ldata.rl_rx_sparemap = map;
1621 	bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap,
1622 	    BUS_DMASYNC_PREREAD);
1623 
1624 	desc = &sc->rl_ldata.rl_rx_list[idx];
1625 	desc->rl_vlanctl = 0;
1626 	desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[0].ds_addr));
1627 	desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[0].ds_addr));
1628 	cmdstat = segs[0].ds_len;
1629 	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
1630 		cmdstat |= RL_RDESC_CMD_EOR;
1631 	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
1632 
1633 	return (0);
1634 }
1635 
1636 #ifdef RE_FIXUP_RX
1637 static __inline void
1638 re_fixup_rx(m)
1639 	struct mbuf		*m;
1640 {
1641 	int                     i;
1642 	uint16_t                *src, *dst;
1643 
1644 	src = mtod(m, uint16_t *);
1645 	dst = src - (RE_ETHER_ALIGN - ETHER_ALIGN) / sizeof *src;
1646 
1647 	for (i = 0; i < (m->m_len / sizeof(uint16_t) + 1); i++)
1648 		*dst++ = *src++;
1649 
1650 	m->m_data -= RE_ETHER_ALIGN - ETHER_ALIGN;
1651 
1652 	return;
1653 }
1654 #endif
1655 
1656 static int
1657 re_tx_list_init(sc)
1658 	struct rl_softc		*sc;
1659 {
1660 	struct rl_desc		*desc;
1661 	int			i;
1662 
1663 	RL_LOCK_ASSERT(sc);
1664 
1665 	bzero(sc->rl_ldata.rl_tx_list,
1666 	    sc->rl_ldata.rl_tx_desc_cnt * sizeof(struct rl_desc));
1667 	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++)
1668 		sc->rl_ldata.rl_tx_desc[i].tx_m = NULL;
1669 	/* Set EOR. */
1670 	desc = &sc->rl_ldata.rl_tx_list[sc->rl_ldata.rl_tx_desc_cnt - 1];
1671 	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_EOR);
1672 
1673 	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
1674 	    sc->rl_ldata.rl_tx_list_map,
1675 	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1676 
1677 	sc->rl_ldata.rl_tx_prodidx = 0;
1678 	sc->rl_ldata.rl_tx_considx = 0;
1679 	sc->rl_ldata.rl_tx_free = sc->rl_ldata.rl_tx_desc_cnt;
1680 
1681 	return (0);
1682 }
1683 
1684 static int
1685 re_rx_list_init(sc)
1686 	struct rl_softc		*sc;
1687 {
1688 	int			error, i;
1689 
1690 	bzero(sc->rl_ldata.rl_rx_list,
1691 	    sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc));
1692 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1693 		sc->rl_ldata.rl_rx_desc[i].rx_m = NULL;
1694 		if ((error = re_newbuf(sc, i)) != 0)
1695 			return (error);
1696 	}
1697 
1698 	/* Flush the RX descriptors */
1699 
1700 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1701 	    sc->rl_ldata.rl_rx_list_map,
1702 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
1703 
1704 	sc->rl_ldata.rl_rx_prodidx = 0;
1705 	sc->rl_head = sc->rl_tail = NULL;
1706 
1707 	return (0);
1708 }
1709 
1710 /*
1711  * RX handler for C+ and 8169. For the gigE chips, we support
1712  * the reception of jumbo frames that have been fragmented
1713  * across multiple 2K mbuf cluster buffers.
1714  */
1715 static int
1716 re_rxeof(sc)
1717 	struct rl_softc		*sc;
1718 {
1719 	struct mbuf		*m;
1720 	struct ifnet		*ifp;
1721 	int			i, total_len;
1722 	struct rl_desc		*cur_rx;
1723 	u_int32_t		rxstat, rxvlan;
1724 	int			maxpkt = 16;
1725 
1726 	RL_LOCK_ASSERT(sc);
1727 
1728 	ifp = sc->rl_ifp;
1729 
1730 	/* Invalidate the descriptor memory */
1731 
1732 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1733 	    sc->rl_ldata.rl_rx_list_map,
1734 	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
1735 
1736 	for (i = sc->rl_ldata.rl_rx_prodidx; maxpkt > 0;
1737 	    i = RL_RX_DESC_NXT(sc, i)) {
1738 		cur_rx = &sc->rl_ldata.rl_rx_list[i];
1739 		rxstat = le32toh(cur_rx->rl_cmdstat);
1740 		if ((rxstat & RL_RDESC_STAT_OWN) != 0)
1741 			break;
1742 		total_len = rxstat & sc->rl_rxlenmask;
1743 		rxvlan = le32toh(cur_rx->rl_vlanctl);
1744 		m = sc->rl_ldata.rl_rx_desc[i].rx_m;
1745 
1746 		if (!(rxstat & RL_RDESC_STAT_EOF)) {
1747 			if (re_newbuf(sc, i) != 0) {
1748 				/*
1749 				 * If this is part of a multi-fragment packet,
1750 				 * discard all the pieces.
1751 				 */
1752 				if (sc->rl_head != NULL) {
1753 					m_freem(sc->rl_head);
1754 					sc->rl_head = sc->rl_tail = NULL;
1755 				}
1756 				re_discard_rxbuf(sc, i);
1757 				continue;
1758 			}
1759 			m->m_len = RE_RX_DESC_BUFLEN;
1760 			if (sc->rl_head == NULL)
1761 				sc->rl_head = sc->rl_tail = m;
1762 			else {
1763 				m->m_flags &= ~M_PKTHDR;
1764 				sc->rl_tail->m_next = m;
1765 				sc->rl_tail = m;
1766 			}
1767 			continue;
1768 		}
1769 
1770 		/*
1771 		 * NOTE: for the 8139C+, the frame length field
1772 		 * is always 12 bits in size, but for the gigE chips,
1773 		 * it is 13 bits (since the max RX frame length is 16K).
1774 		 * Unfortunately, all 32 bits in the status word
1775 		 * were already used, so to make room for the extra
1776 		 * length bit, RealTek took out the 'frame alignment
1777 		 * error' bit and shifted the other status bits
1778 		 * over one slot. The OWN, EOR, FS and LS bits are
1779 		 * still in the same places. We have already extracted
1780 		 * the frame length and checked the OWN bit, so rather
1781 		 * than using an alternate bit mapping, we shift the
1782 		 * status bits one space to the right so we can evaluate
1783 		 * them using the 8169 status as though it was in the
1784 		 * same format as that of the 8139C+.
1785 		 */
1786 		if (sc->rl_type == RL_8169)
1787 			rxstat >>= 1;
1788 
1789 		/*
1790 		 * if total_len > 2^13-1, both _RXERRSUM and _GIANT will be
1791 		 * set, but if CRC is clear, it will still be a valid frame.
1792 		 */
1793 		if (rxstat & RL_RDESC_STAT_RXERRSUM && !(total_len > 8191 &&
1794 		    (rxstat & RL_RDESC_STAT_ERRS) == RL_RDESC_STAT_GIANT)) {
1795 			ifp->if_ierrors++;
1796 			/*
1797 			 * If this is part of a multi-fragment packet,
1798 			 * discard all the pieces.
1799 			 */
1800 			if (sc->rl_head != NULL) {
1801 				m_freem(sc->rl_head);
1802 				sc->rl_head = sc->rl_tail = NULL;
1803 			}
1804 			re_discard_rxbuf(sc, i);
1805 			continue;
1806 		}
1807 
1808 		/*
1809 		 * If allocating a replacement mbuf fails,
1810 		 * reload the current one.
1811 		 */
1812 
1813 		if (re_newbuf(sc, i) != 0) {
1814 			ifp->if_iqdrops++;
1815 			if (sc->rl_head != NULL) {
1816 				m_freem(sc->rl_head);
1817 				sc->rl_head = sc->rl_tail = NULL;
1818 			}
1819 			re_discard_rxbuf(sc, i);
1820 			continue;
1821 		}
1822 
1823 		if (sc->rl_head != NULL) {
1824 			m->m_len = total_len % RE_RX_DESC_BUFLEN;
1825 			if (m->m_len == 0)
1826 				m->m_len = RE_RX_DESC_BUFLEN;
1827 			/*
1828 			 * Special case: if there's 4 bytes or less
1829 			 * in this buffer, the mbuf can be discarded:
1830 			 * the last 4 bytes is the CRC, which we don't
1831 			 * care about anyway.
1832 			 */
1833 			if (m->m_len <= ETHER_CRC_LEN) {
1834 				sc->rl_tail->m_len -=
1835 				    (ETHER_CRC_LEN - m->m_len);
1836 				m_freem(m);
1837 			} else {
1838 				m->m_len -= ETHER_CRC_LEN;
1839 				m->m_flags &= ~M_PKTHDR;
1840 				sc->rl_tail->m_next = m;
1841 			}
1842 			m = sc->rl_head;
1843 			sc->rl_head = sc->rl_tail = NULL;
1844 			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
1845 		} else
1846 			m->m_pkthdr.len = m->m_len =
1847 			    (total_len - ETHER_CRC_LEN);
1848 
1849 #ifdef RE_FIXUP_RX
1850 		re_fixup_rx(m);
1851 #endif
1852 		ifp->if_ipackets++;
1853 		m->m_pkthdr.rcvif = ifp;
1854 
1855 		/* Do RX checksumming if enabled */
1856 
1857 		if (ifp->if_capenable & IFCAP_RXCSUM) {
1858 
1859 			/* Check IP header checksum */
1860 			if (rxstat & RL_RDESC_STAT_PROTOID)
1861 				m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED;
1862 			if (!(rxstat & RL_RDESC_STAT_IPSUMBAD))
1863 				m->m_pkthdr.csum_flags |= CSUM_IP_VALID;
1864 
1865 			/* Check TCP/UDP checksum */
1866 			if ((RL_TCPPKT(rxstat) &&
1867 			    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
1868 			    (RL_UDPPKT(rxstat) &&
1869 			    !(rxstat & RL_RDESC_STAT_UDPSUMBAD))) {
1870 				m->m_pkthdr.csum_flags |=
1871 				    CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
1872 				m->m_pkthdr.csum_data = 0xffff;
1873 			}
1874 		}
1875 		maxpkt--;
1876 		if (rxvlan & RL_RDESC_VLANCTL_TAG) {
1877 			m->m_pkthdr.ether_vtag =
1878 			    ntohs((rxvlan & RL_RDESC_VLANCTL_DATA));
1879 			m->m_flags |= M_VLANTAG;
1880 		}
1881 		RL_UNLOCK(sc);
1882 		(*ifp->if_input)(ifp, m);
1883 		RL_LOCK(sc);
1884 	}
1885 
1886 	/* Flush the RX DMA ring */
1887 
1888 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1889 	    sc->rl_ldata.rl_rx_list_map,
1890 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
1891 
1892 	sc->rl_ldata.rl_rx_prodidx = i;
1893 
1894 	if (maxpkt)
1895 		return(EAGAIN);
1896 
1897 	return(0);
1898 }
1899 
1900 static void
1901 re_txeof(sc)
1902 	struct rl_softc		*sc;
1903 {
1904 	struct ifnet		*ifp;
1905 	struct rl_txdesc	*txd;
1906 	u_int32_t		txstat;
1907 	int			cons;
1908 
1909 	cons = sc->rl_ldata.rl_tx_considx;
1910 	if (cons == sc->rl_ldata.rl_tx_prodidx)
1911 		return;
1912 
1913 	ifp = sc->rl_ifp;
1914 	/* Invalidate the TX descriptor list */
1915 	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
1916 	    sc->rl_ldata.rl_tx_list_map,
1917 	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
1918 
1919 	for (; cons != sc->rl_ldata.rl_tx_prodidx;
1920 	    cons = RL_TX_DESC_NXT(sc, cons)) {
1921 		txstat = le32toh(sc->rl_ldata.rl_tx_list[cons].rl_cmdstat);
1922 		if (txstat & RL_TDESC_STAT_OWN)
1923 			break;
1924 		/*
1925 		 * We only stash mbufs in the last descriptor
1926 		 * in a fragment chain, which also happens to
1927 		 * be the only place where the TX status bits
1928 		 * are valid.
1929 		 */
1930 		if (txstat & RL_TDESC_CMD_EOF) {
1931 			txd = &sc->rl_ldata.rl_tx_desc[cons];
1932 			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
1933 			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
1934 			bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag,
1935 			    txd->tx_dmamap);
1936 			KASSERT(txd->tx_m != NULL,
1937 			    ("%s: freeing NULL mbufs!", __func__));
1938 			m_freem(txd->tx_m);
1939 			txd->tx_m = NULL;
1940 			if (txstat & (RL_TDESC_STAT_EXCESSCOL|
1941 			    RL_TDESC_STAT_COLCNT))
1942 				ifp->if_collisions++;
1943 			if (txstat & RL_TDESC_STAT_TXERRSUM)
1944 				ifp->if_oerrors++;
1945 			else
1946 				ifp->if_opackets++;
1947 		}
1948 		sc->rl_ldata.rl_tx_free++;
1949 		ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1950 	}
1951 	sc->rl_ldata.rl_tx_considx = cons;
1952 
1953 	/* No changes made to the TX ring, so no flush needed */
1954 
1955 	if (sc->rl_ldata.rl_tx_free != sc->rl_ldata.rl_tx_desc_cnt) {
1956 		/*
1957 		 * Some chips will ignore a second TX request issued
1958 		 * while an existing transmission is in progress. If
1959 		 * the transmitter goes idle but there are still
1960 		 * packets waiting to be sent, we need to restart the
1961 		 * channel here to flush them out. This only seems to
1962 		 * be required with the PCIe devices.
1963 		 */
1964 		CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
1965 
1966 #ifdef RE_TX_MODERATION
1967 		/*
1968 		 * If not all descriptors have been reaped yet, reload
1969 		 * the timer so that we will eventually get another
1970 		 * interrupt that will cause us to re-enter this routine.
1971 		 * This is done in case the transmitter has gone idle.
1972 		 */
1973 		CSR_WRITE_4(sc, RL_TIMERCNT, 1);
1974 #endif
1975 	} else
1976 		sc->rl_watchdog_timer = 0;
1977 }
1978 
1979 static void
1980 re_tick(xsc)
1981 	void			*xsc;
1982 {
1983 	struct rl_softc		*sc;
1984 	struct mii_data		*mii;
1985 	struct ifnet		*ifp;
1986 
1987 	sc = xsc;
1988 	ifp = sc->rl_ifp;
1989 
1990 	RL_LOCK_ASSERT(sc);
1991 
1992 	re_watchdog(sc);
1993 
1994 	mii = device_get_softc(sc->rl_miibus);
1995 	mii_tick(mii);
1996 	if (sc->rl_link) {
1997 		if (!(mii->mii_media_status & IFM_ACTIVE))
1998 			sc->rl_link = 0;
1999 	} else {
2000 		if (mii->mii_media_status & IFM_ACTIVE &&
2001 		    IFM_SUBTYPE(mii->mii_media_active) != IFM_NONE) {
2002 			sc->rl_link = 1;
2003 			if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2004 				taskqueue_enqueue_fast(taskqueue_fast,
2005 				    &sc->rl_txtask);
2006 		}
2007 	}
2008 
2009 	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
2010 }
2011 
2012 #ifdef DEVICE_POLLING
2013 static void
2014 re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
2015 {
2016 	struct rl_softc *sc = ifp->if_softc;
2017 
2018 	RL_LOCK(sc);
2019 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
2020 		re_poll_locked(ifp, cmd, count);
2021 	RL_UNLOCK(sc);
2022 }
2023 
2024 static void
2025 re_poll_locked(struct ifnet *ifp, enum poll_cmd cmd, int count)
2026 {
2027 	struct rl_softc *sc = ifp->if_softc;
2028 
2029 	RL_LOCK_ASSERT(sc);
2030 
2031 	sc->rxcycles = count;
2032 	re_rxeof(sc);
2033 	re_txeof(sc);
2034 
2035 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2036 		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_txtask);
2037 
2038 	if (cmd == POLL_AND_CHECK_STATUS) { /* also check status register */
2039 		u_int16_t       status;
2040 
2041 		status = CSR_READ_2(sc, RL_ISR);
2042 		if (status == 0xffff)
2043 			return;
2044 		if (status)
2045 			CSR_WRITE_2(sc, RL_ISR, status);
2046 
2047 		/*
2048 		 * XXX check behaviour on receiver stalls.
2049 		 */
2050 
2051 		if (status & RL_ISR_SYSTEM_ERR) {
2052 			re_reset(sc);
2053 			re_init_locked(sc);
2054 		}
2055 	}
2056 }
2057 #endif /* DEVICE_POLLING */
2058 
2059 static int
2060 re_intr(arg)
2061 	void			*arg;
2062 {
2063 	struct rl_softc		*sc;
2064 	uint16_t		status;
2065 
2066 	sc = arg;
2067 
2068 	status = CSR_READ_2(sc, RL_ISR);
2069 	if (status == 0xFFFF || (status & RL_INTRS_CPLUS) == 0)
2070                 return (FILTER_STRAY);
2071 	CSR_WRITE_2(sc, RL_IMR, 0);
2072 
2073 	taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_inttask);
2074 
2075 	return (FILTER_HANDLED);
2076 }
2077 
2078 static void
2079 re_int_task(arg, npending)
2080 	void			*arg;
2081 	int			npending;
2082 {
2083 	struct rl_softc		*sc;
2084 	struct ifnet		*ifp;
2085 	u_int16_t		status;
2086 	int			rval = 0;
2087 
2088 	sc = arg;
2089 	ifp = sc->rl_ifp;
2090 
2091 	RL_LOCK(sc);
2092 
2093 	status = CSR_READ_2(sc, RL_ISR);
2094         CSR_WRITE_2(sc, RL_ISR, status);
2095 
2096 	if (sc->suspended ||
2097 	    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2098 		RL_UNLOCK(sc);
2099 		return;
2100 	}
2101 
2102 #ifdef DEVICE_POLLING
2103 	if  (ifp->if_capenable & IFCAP_POLLING) {
2104 		RL_UNLOCK(sc);
2105 		return;
2106 	}
2107 #endif
2108 
2109 	if (status & (RL_ISR_RX_OK|RL_ISR_RX_ERR|RL_ISR_FIFO_OFLOW))
2110 		rval = re_rxeof(sc);
2111 
2112 #ifdef RE_TX_MODERATION
2113 	if (status & (RL_ISR_TIMEOUT_EXPIRED|
2114 #else
2115 	if (status & (RL_ISR_TX_OK|
2116 #endif
2117 	    RL_ISR_TX_ERR|RL_ISR_TX_DESC_UNAVAIL))
2118 		re_txeof(sc);
2119 
2120 	if (status & RL_ISR_SYSTEM_ERR) {
2121 		re_reset(sc);
2122 		re_init_locked(sc);
2123 	}
2124 
2125 	if (status & RL_ISR_LINKCHG) {
2126 		callout_stop(&sc->rl_stat_callout);
2127 		re_tick(sc);
2128 	}
2129 
2130 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2131 		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_txtask);
2132 
2133 	RL_UNLOCK(sc);
2134 
2135         if ((CSR_READ_2(sc, RL_ISR) & RL_INTRS_CPLUS) || rval) {
2136 		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_inttask);
2137 		return;
2138 	}
2139 
2140 	CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2141 
2142 	return;
2143 }
2144 
2145 static int
2146 re_encap(sc, m_head)
2147 	struct rl_softc		*sc;
2148 	struct mbuf		**m_head;
2149 {
2150 	struct rl_txdesc	*txd, *txd_last;
2151 	bus_dma_segment_t	segs[RL_NTXSEGS];
2152 	bus_dmamap_t		map;
2153 	struct mbuf		*m_new;
2154 	struct rl_desc		*desc;
2155 	int			nsegs, prod;
2156 	int			i, error, ei, si;
2157 	int			padlen;
2158 	uint32_t		cmdstat, csum_flags, vlanctl;
2159 
2160 	RL_LOCK_ASSERT(sc);
2161 	M_ASSERTPKTHDR((*m_head));
2162 
2163 	/*
2164 	 * With some of the RealTek chips, using the checksum offload
2165 	 * support in conjunction with the autopadding feature results
2166 	 * in the transmission of corrupt frames. For example, if we
2167 	 * need to send a really small IP fragment that's less than 60
2168 	 * bytes in size, and IP header checksumming is enabled, the
2169 	 * resulting ethernet frame that appears on the wire will
2170 	 * have garbled payload. To work around this, if TX checksum
2171 	 * offload is enabled, we always manually pad short frames out
2172 	 * to the minimum ethernet frame size.
2173 	 *
2174 	 * Note: this appears unnecessary for TCP, and doing it for TCP
2175 	 * with PCIe adapters seems to result in bad checksums.
2176 	 */
2177 	if ((*m_head)->m_pkthdr.csum_flags & (CSUM_IP | CSUM_UDP) &&
2178 	    ((*m_head)->m_pkthdr.csum_flags & CSUM_TCP) == 0 &&
2179             (*m_head)->m_pkthdr.len < RL_MIN_FRAMELEN) {
2180 		padlen = RL_MIN_FRAMELEN - (*m_head)->m_pkthdr.len;
2181 		if (M_WRITABLE(*m_head) == 0) {
2182 			/* Get a writable copy. */
2183 			m_new = m_dup(*m_head, M_DONTWAIT);
2184 			m_freem(*m_head);
2185 			if (m_new == NULL) {
2186 				*m_head = NULL;
2187 				return (ENOBUFS);
2188 			}
2189 			*m_head = m_new;
2190 		}
2191 		if ((*m_head)->m_next != NULL ||
2192 		    M_TRAILINGSPACE(*m_head) < padlen) {
2193 			m_new = m_defrag(*m_head, M_DONTWAIT);
2194 			if (m_new == NULL) {
2195 				m_freem(*m_head);
2196 				*m_head = NULL;
2197 				return (ENOBUFS);
2198 			}
2199 		} else
2200 			m_new = *m_head;
2201 
2202 		/*
2203 		 * Manually pad short frames, and zero the pad space
2204 		 * to avoid leaking data.
2205 		 */
2206 		bzero(mtod(m_new, char *) + m_new->m_pkthdr.len, padlen);
2207 		m_new->m_pkthdr.len += padlen;
2208 		m_new->m_len = m_new->m_pkthdr.len;
2209 		*m_head = m_new;
2210 	}
2211 
2212 	prod = sc->rl_ldata.rl_tx_prodidx;
2213 	txd = &sc->rl_ldata.rl_tx_desc[prod];
2214 	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap,
2215 	    *m_head, segs, &nsegs, BUS_DMA_NOWAIT);
2216 	if (error == EFBIG) {
2217 		m_new = m_collapse(*m_head, M_DONTWAIT, RL_NTXSEGS);
2218 		if (m_new == NULL) {
2219 			m_freem(*m_head);
2220 			*m_head = NULL;
2221 			return (ENOBUFS);
2222 		}
2223 		*m_head = m_new;
2224 		error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_tx_mtag,
2225 		    txd->tx_dmamap, *m_head, segs, &nsegs, BUS_DMA_NOWAIT);
2226 		if (error != 0) {
2227 			m_freem(*m_head);
2228 			*m_head = NULL;
2229 			return (error);
2230 		}
2231 	} else if (error != 0)
2232 		return (error);
2233 	if (nsegs == 0) {
2234 		m_freem(*m_head);
2235 		*m_head = NULL;
2236 		return (EIO);
2237 	}
2238 
2239 	/* Check for number of available descriptors. */
2240 	if (sc->rl_ldata.rl_tx_free - nsegs <= 1) {
2241 		bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap);
2242 		return (ENOBUFS);
2243 	}
2244 
2245 	bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap,
2246 	    BUS_DMASYNC_PREWRITE);
2247 
2248 	/*
2249 	 * Set up checksum offload. Note: checksum offload bits must
2250 	 * appear in all descriptors of a multi-descriptor transmit
2251 	 * attempt. This is according to testing done with an 8169
2252 	 * chip. This is a requirement.
2253 	 */
2254 	csum_flags = 0;
2255 	if (((*m_head)->m_pkthdr.csum_flags & CSUM_TSO) != 0)
2256 		csum_flags = RL_TDESC_CMD_LGSEND |
2257 		    ((uint32_t)(*m_head)->m_pkthdr.tso_segsz <<
2258 		    RL_TDESC_CMD_MSSVAL_SHIFT);
2259 	else {
2260 		if ((*m_head)->m_pkthdr.csum_flags & CSUM_IP)
2261 			csum_flags |= RL_TDESC_CMD_IPCSUM;
2262 		if ((*m_head)->m_pkthdr.csum_flags & CSUM_TCP)
2263 			csum_flags |= RL_TDESC_CMD_TCPCSUM;
2264 		if ((*m_head)->m_pkthdr.csum_flags & CSUM_UDP)
2265 			csum_flags |= RL_TDESC_CMD_UDPCSUM;
2266 	}
2267 
2268 	/*
2269 	 * Set up hardware VLAN tagging. Note: vlan tag info must
2270 	 * appear in all descriptors of a multi-descriptor
2271 	 * transmission attempt.
2272 	 */
2273 	vlanctl = 0;
2274 	if ((*m_head)->m_flags & M_VLANTAG)
2275 		vlanctl =
2276 		    htole32(htons((*m_head)->m_pkthdr.ether_vtag) |
2277 		    RL_TDESC_VLANCTL_TAG);
2278 
2279 	si = prod;
2280 	for (i = 0; i < nsegs; i++, prod = RL_TX_DESC_NXT(sc, prod)) {
2281 		desc = &sc->rl_ldata.rl_tx_list[prod];
2282 		desc->rl_vlanctl = vlanctl;
2283 		desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[i].ds_addr));
2284 		desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[i].ds_addr));
2285 		cmdstat = segs[i].ds_len;
2286 		if (i != 0)
2287 			cmdstat |= RL_TDESC_CMD_OWN;
2288 		if (prod == sc->rl_ldata.rl_tx_desc_cnt - 1)
2289 			cmdstat |= RL_TDESC_CMD_EOR;
2290 		desc->rl_cmdstat = htole32(cmdstat | csum_flags);
2291 		sc->rl_ldata.rl_tx_free--;
2292 	}
2293 	/* Update producer index. */
2294 	sc->rl_ldata.rl_tx_prodidx = prod;
2295 
2296 	/* Set EOF on the last descriptor. */
2297 	ei = RL_TX_DESC_PRV(sc, prod);
2298 	desc = &sc->rl_ldata.rl_tx_list[ei];
2299 	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_EOF);
2300 
2301 	desc = &sc->rl_ldata.rl_tx_list[si];
2302 	/* Set SOF and transfer ownership of packet to the chip. */
2303 	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_OWN | RL_TDESC_CMD_SOF);
2304 
2305 	/*
2306 	 * Insure that the map for this transmission
2307 	 * is placed at the array index of the last descriptor
2308 	 * in this chain.  (Swap last and first dmamaps.)
2309 	 */
2310 	txd_last = &sc->rl_ldata.rl_tx_desc[ei];
2311 	map = txd->tx_dmamap;
2312 	txd->tx_dmamap = txd_last->tx_dmamap;
2313 	txd_last->tx_dmamap = map;
2314 	txd_last->tx_m = *m_head;
2315 
2316 	return (0);
2317 }
2318 
2319 static void
2320 re_tx_task(arg, npending)
2321 	void			*arg;
2322 	int			npending;
2323 {
2324 	struct ifnet		*ifp;
2325 
2326 	ifp = arg;
2327 	re_start(ifp);
2328 
2329 	return;
2330 }
2331 
2332 /*
2333  * Main transmit routine for C+ and gigE NICs.
2334  */
2335 static void
2336 re_start(ifp)
2337 	struct ifnet		*ifp;
2338 {
2339 	struct rl_softc		*sc;
2340 	struct mbuf		*m_head;
2341 	int			queued;
2342 
2343 	sc = ifp->if_softc;
2344 
2345 	RL_LOCK(sc);
2346 
2347 	if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
2348 	    IFF_DRV_RUNNING || sc->rl_link == 0) {
2349 		RL_UNLOCK(sc);
2350 		return;
2351 	}
2352 
2353 	for (queued = 0; !IFQ_DRV_IS_EMPTY(&ifp->if_snd) &&
2354 	    sc->rl_ldata.rl_tx_free > 1;) {
2355 		IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
2356 		if (m_head == NULL)
2357 			break;
2358 
2359 		if (re_encap(sc, &m_head) != 0) {
2360 			if (m_head == NULL)
2361 				break;
2362 			IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
2363 			ifp->if_drv_flags |= IFF_DRV_OACTIVE;
2364 			break;
2365 		}
2366 
2367 		/*
2368 		 * If there's a BPF listener, bounce a copy of this frame
2369 		 * to him.
2370 		 */
2371 		ETHER_BPF_MTAP(ifp, m_head);
2372 
2373 		queued++;
2374 	}
2375 
2376 	if (queued == 0) {
2377 #ifdef RE_TX_MODERATION
2378 		if (sc->rl_ldata.rl_tx_free != sc->rl_ldata.rl_tx_desc_cnt)
2379 			CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2380 #endif
2381 		RL_UNLOCK(sc);
2382 		return;
2383 	}
2384 
2385 	/* Flush the TX descriptors */
2386 
2387 	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2388 	    sc->rl_ldata.rl_tx_list_map,
2389 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2390 
2391 	CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2392 
2393 #ifdef RE_TX_MODERATION
2394 	/*
2395 	 * Use the countdown timer for interrupt moderation.
2396 	 * 'TX done' interrupts are disabled. Instead, we reset the
2397 	 * countdown timer, which will begin counting until it hits
2398 	 * the value in the TIMERINT register, and then trigger an
2399 	 * interrupt. Each time we write to the TIMERCNT register,
2400 	 * the timer count is reset to 0.
2401 	 */
2402 	CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2403 #endif
2404 
2405 	/*
2406 	 * Set a timeout in case the chip goes out to lunch.
2407 	 */
2408 	sc->rl_watchdog_timer = 5;
2409 
2410 	RL_UNLOCK(sc);
2411 
2412 	return;
2413 }
2414 
2415 static void
2416 re_init(xsc)
2417 	void			*xsc;
2418 {
2419 	struct rl_softc		*sc = xsc;
2420 
2421 	RL_LOCK(sc);
2422 	re_init_locked(sc);
2423 	RL_UNLOCK(sc);
2424 }
2425 
2426 static void
2427 re_init_locked(sc)
2428 	struct rl_softc		*sc;
2429 {
2430 	struct ifnet		*ifp = sc->rl_ifp;
2431 	struct mii_data		*mii;
2432 	u_int32_t		rxcfg = 0;
2433 	uint16_t		cfg;
2434 	union {
2435 		uint32_t align_dummy;
2436 		u_char eaddr[ETHER_ADDR_LEN];
2437         } eaddr;
2438 
2439 	RL_LOCK_ASSERT(sc);
2440 
2441 	mii = device_get_softc(sc->rl_miibus);
2442 
2443 	/*
2444 	 * Cancel pending I/O and free all RX/TX buffers.
2445 	 */
2446 	re_stop(sc);
2447 
2448 	/*
2449 	 * Enable C+ RX and TX mode, as well as VLAN stripping and
2450 	 * RX checksum offload. We must configure the C+ register
2451 	 * before all others.
2452 	 */
2453 	cfg = RL_CPLUSCMD_PCI_MRW;
2454 	if ((ifp->if_capenable & IFCAP_RXCSUM) != 0)
2455 		cfg |= RL_CPLUSCMD_RXCSUM_ENB;
2456 	if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0)
2457 		cfg |= RL_CPLUSCMD_VLANSTRIP;
2458 	CSR_WRITE_2(sc, RL_CPLUS_CMD,
2459 	    cfg | RL_CPLUSCMD_RXENB | RL_CPLUSCMD_TXENB);
2460 
2461 	/*
2462 	 * Init our MAC address.  Even though the chipset
2463 	 * documentation doesn't mention it, we need to enter "Config
2464 	 * register write enable" mode to modify the ID registers.
2465 	 */
2466 	/* Copy MAC address on stack to align. */
2467 	bcopy(IF_LLADDR(ifp), eaddr.eaddr, ETHER_ADDR_LEN);
2468 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG);
2469 	CSR_WRITE_4(sc, RL_IDR0,
2470 	    htole32(*(u_int32_t *)(&eaddr.eaddr[0])));
2471 	CSR_WRITE_4(sc, RL_IDR4,
2472 	    htole32(*(u_int32_t *)(&eaddr.eaddr[4])));
2473 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
2474 
2475 	/*
2476 	 * For C+ mode, initialize the RX descriptors and mbufs.
2477 	 */
2478 	re_rx_list_init(sc);
2479 	re_tx_list_init(sc);
2480 
2481 	/*
2482 	 * Load the addresses of the RX and TX lists into the chip.
2483 	 */
2484 
2485 	CSR_WRITE_4(sc, RL_RXLIST_ADDR_HI,
2486 	    RL_ADDR_HI(sc->rl_ldata.rl_rx_list_addr));
2487 	CSR_WRITE_4(sc, RL_RXLIST_ADDR_LO,
2488 	    RL_ADDR_LO(sc->rl_ldata.rl_rx_list_addr));
2489 
2490 	CSR_WRITE_4(sc, RL_TXLIST_ADDR_HI,
2491 	    RL_ADDR_HI(sc->rl_ldata.rl_tx_list_addr));
2492 	CSR_WRITE_4(sc, RL_TXLIST_ADDR_LO,
2493 	    RL_ADDR_LO(sc->rl_ldata.rl_tx_list_addr));
2494 
2495 	/*
2496 	 * Enable transmit and receive.
2497 	 */
2498 	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
2499 
2500 	/*
2501 	 * Set the initial TX and RX configuration.
2502 	 */
2503 	if (sc->rl_testmode) {
2504 		if (sc->rl_type == RL_8169)
2505 			CSR_WRITE_4(sc, RL_TXCFG,
2506 			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON);
2507 		else
2508 			CSR_WRITE_4(sc, RL_TXCFG,
2509 			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON_CPLUS);
2510 	} else
2511 		CSR_WRITE_4(sc, RL_TXCFG, RL_TXCFG_CONFIG);
2512 
2513 	CSR_WRITE_1(sc, RL_EARLY_TX_THRESH, 16);
2514 
2515 	CSR_WRITE_4(sc, RL_RXCFG, RL_RXCFG_CONFIG);
2516 
2517 	/* Set the individual bit to receive frames for this host only. */
2518 	rxcfg = CSR_READ_4(sc, RL_RXCFG);
2519 	rxcfg |= RL_RXCFG_RX_INDIV;
2520 
2521 	/* If we want promiscuous mode, set the allframes bit. */
2522 	if (ifp->if_flags & IFF_PROMISC)
2523 		rxcfg |= RL_RXCFG_RX_ALLPHYS;
2524 	else
2525 		rxcfg &= ~RL_RXCFG_RX_ALLPHYS;
2526 	CSR_WRITE_4(sc, RL_RXCFG, rxcfg);
2527 
2528 	/*
2529 	 * Set capture broadcast bit to capture broadcast frames.
2530 	 */
2531 	if (ifp->if_flags & IFF_BROADCAST)
2532 		rxcfg |= RL_RXCFG_RX_BROAD;
2533 	else
2534 		rxcfg &= ~RL_RXCFG_RX_BROAD;
2535 	CSR_WRITE_4(sc, RL_RXCFG, rxcfg);
2536 
2537 	/*
2538 	 * Program the multicast filter, if necessary.
2539 	 */
2540 	re_setmulti(sc);
2541 
2542 #ifdef DEVICE_POLLING
2543 	/*
2544 	 * Disable interrupts if we are polling.
2545 	 */
2546 	if (ifp->if_capenable & IFCAP_POLLING)
2547 		CSR_WRITE_2(sc, RL_IMR, 0);
2548 	else	/* otherwise ... */
2549 #endif
2550 
2551 	/*
2552 	 * Enable interrupts.
2553 	 */
2554 	if (sc->rl_testmode)
2555 		CSR_WRITE_2(sc, RL_IMR, 0);
2556 	else
2557 		CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2558 	CSR_WRITE_2(sc, RL_ISR, RL_INTRS_CPLUS);
2559 
2560 	/* Set initial TX threshold */
2561 	sc->rl_txthresh = RL_TX_THRESH_INIT;
2562 
2563 	/* Start RX/TX process. */
2564 	CSR_WRITE_4(sc, RL_MISSEDPKT, 0);
2565 #ifdef notdef
2566 	/* Enable receiver and transmitter. */
2567 	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
2568 #endif
2569 
2570 #ifdef RE_TX_MODERATION
2571 	/*
2572 	 * Initialize the timer interrupt register so that
2573 	 * a timer interrupt will be generated once the timer
2574 	 * reaches a certain number of ticks. The timer is
2575 	 * reloaded on each transmit. This gives us TX interrupt
2576 	 * moderation, which dramatically improves TX frame rate.
2577 	 */
2578 	if (sc->rl_type == RL_8169)
2579 		CSR_WRITE_4(sc, RL_TIMERINT_8169, 0x800);
2580 	else
2581 		CSR_WRITE_4(sc, RL_TIMERINT, 0x400);
2582 #endif
2583 
2584 	/*
2585 	 * For 8169 gigE NICs, set the max allowed RX packet
2586 	 * size so we can receive jumbo frames.
2587 	 */
2588 	if (sc->rl_type == RL_8169)
2589 		CSR_WRITE_2(sc, RL_MAXRXPKTLEN, 16383);
2590 
2591 	if (sc->rl_testmode)
2592 		return;
2593 
2594 	mii_mediachg(mii);
2595 
2596 	CSR_WRITE_1(sc, RL_CFG1, CSR_READ_1(sc, RL_CFG1) | RL_CFG1_DRVLOAD);
2597 
2598 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
2599 	ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
2600 
2601 	sc->rl_link = 0;
2602 	sc->rl_watchdog_timer = 0;
2603 	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
2604 }
2605 
2606 /*
2607  * Set media options.
2608  */
2609 static int
2610 re_ifmedia_upd(ifp)
2611 	struct ifnet		*ifp;
2612 {
2613 	struct rl_softc		*sc;
2614 	struct mii_data		*mii;
2615 
2616 	sc = ifp->if_softc;
2617 	mii = device_get_softc(sc->rl_miibus);
2618 	RL_LOCK(sc);
2619 	mii_mediachg(mii);
2620 	RL_UNLOCK(sc);
2621 
2622 	return (0);
2623 }
2624 
2625 /*
2626  * Report current media status.
2627  */
2628 static void
2629 re_ifmedia_sts(ifp, ifmr)
2630 	struct ifnet		*ifp;
2631 	struct ifmediareq	*ifmr;
2632 {
2633 	struct rl_softc		*sc;
2634 	struct mii_data		*mii;
2635 
2636 	sc = ifp->if_softc;
2637 	mii = device_get_softc(sc->rl_miibus);
2638 
2639 	RL_LOCK(sc);
2640 	mii_pollstat(mii);
2641 	RL_UNLOCK(sc);
2642 	ifmr->ifm_active = mii->mii_media_active;
2643 	ifmr->ifm_status = mii->mii_media_status;
2644 }
2645 
2646 static int
2647 re_ioctl(ifp, command, data)
2648 	struct ifnet		*ifp;
2649 	u_long			command;
2650 	caddr_t			data;
2651 {
2652 	struct rl_softc		*sc = ifp->if_softc;
2653 	struct ifreq		*ifr = (struct ifreq *) data;
2654 	struct mii_data		*mii;
2655 	int			error = 0;
2656 
2657 	switch (command) {
2658 	case SIOCSIFMTU:
2659 		if (ifr->ifr_mtu < ETHERMIN || ifr->ifr_mtu > RL_JUMBO_MTU) {
2660 			error = EINVAL;
2661 			break;
2662 		}
2663 		if (sc->rl_type == RL_8139CPLUS &&
2664 		    ifr->ifr_mtu > RL_MAX_FRAMELEN) {
2665 			error = EINVAL;
2666 			break;
2667 		}
2668 		RL_LOCK(sc);
2669 		if (ifp->if_mtu != ifr->ifr_mtu)
2670 			ifp->if_mtu = ifr->ifr_mtu;
2671 		RL_UNLOCK(sc);
2672 		break;
2673 	case SIOCSIFFLAGS:
2674 		RL_LOCK(sc);
2675 		if ((ifp->if_flags & IFF_UP) != 0) {
2676 			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
2677 				if (((ifp->if_flags ^ sc->rl_if_flags)
2678 				    & IFF_PROMISC) != 0)
2679 					re_setmulti(sc);
2680 			} else
2681 				re_init_locked(sc);
2682 		} else {
2683 			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
2684 				re_stop(sc);
2685 		}
2686 		sc->rl_if_flags = ifp->if_flags;
2687 		RL_UNLOCK(sc);
2688 		break;
2689 	case SIOCADDMULTI:
2690 	case SIOCDELMULTI:
2691 		RL_LOCK(sc);
2692 		re_setmulti(sc);
2693 		RL_UNLOCK(sc);
2694 		break;
2695 	case SIOCGIFMEDIA:
2696 	case SIOCSIFMEDIA:
2697 		mii = device_get_softc(sc->rl_miibus);
2698 		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, command);
2699 		break;
2700 	case SIOCSIFCAP:
2701 	    {
2702 		int mask, reinit;
2703 
2704 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
2705 		reinit = 0;
2706 #ifdef DEVICE_POLLING
2707 		if (mask & IFCAP_POLLING) {
2708 			if (ifr->ifr_reqcap & IFCAP_POLLING) {
2709 				error = ether_poll_register(re_poll, ifp);
2710 				if (error)
2711 					return(error);
2712 				RL_LOCK(sc);
2713 				/* Disable interrupts */
2714 				CSR_WRITE_2(sc, RL_IMR, 0x0000);
2715 				ifp->if_capenable |= IFCAP_POLLING;
2716 				RL_UNLOCK(sc);
2717 			} else {
2718 				error = ether_poll_deregister(ifp);
2719 				/* Enable interrupts. */
2720 				RL_LOCK(sc);
2721 				CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2722 				ifp->if_capenable &= ~IFCAP_POLLING;
2723 				RL_UNLOCK(sc);
2724 			}
2725 		}
2726 #endif /* DEVICE_POLLING */
2727 		if (mask & IFCAP_HWCSUM) {
2728 			ifp->if_capenable ^= IFCAP_HWCSUM;
2729 			if (ifp->if_capenable & IFCAP_TXCSUM)
2730 				ifp->if_hwassist |= RE_CSUM_FEATURES;
2731 			else
2732 				ifp->if_hwassist &= ~RE_CSUM_FEATURES;
2733 			reinit = 1;
2734 		}
2735 		if (mask & IFCAP_VLAN_HWTAGGING) {
2736 			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
2737 			reinit = 1;
2738 		}
2739 		if (mask & IFCAP_TSO4) {
2740 			ifp->if_capenable ^= IFCAP_TSO4;
2741 			if ((IFCAP_TSO4 & ifp->if_capenable) &&
2742 			    (IFCAP_TSO4 & ifp->if_capabilities))
2743 				ifp->if_hwassist |= CSUM_TSO;
2744 			else
2745 				ifp->if_hwassist &= ~CSUM_TSO;
2746 		}
2747 		if ((mask & IFCAP_WOL) != 0 &&
2748 		    (ifp->if_capabilities & IFCAP_WOL) != 0) {
2749 			if ((mask & IFCAP_WOL_UCAST) != 0)
2750 				ifp->if_capenable ^= IFCAP_WOL_UCAST;
2751 			if ((mask & IFCAP_WOL_MCAST) != 0)
2752 				ifp->if_capenable ^= IFCAP_WOL_MCAST;
2753 			if ((mask & IFCAP_WOL_MAGIC) != 0)
2754 				ifp->if_capenable ^= IFCAP_WOL_MAGIC;
2755 		}
2756 		if (reinit && ifp->if_drv_flags & IFF_DRV_RUNNING)
2757 			re_init(sc);
2758 		VLAN_CAPABILITIES(ifp);
2759 	    }
2760 		break;
2761 	default:
2762 		error = ether_ioctl(ifp, command, data);
2763 		break;
2764 	}
2765 
2766 	return (error);
2767 }
2768 
2769 static void
2770 re_watchdog(sc)
2771 	struct rl_softc		*sc;
2772 {
2773 
2774 	RL_LOCK_ASSERT(sc);
2775 
2776 	if (sc->rl_watchdog_timer == 0 || --sc->rl_watchdog_timer != 0)
2777 		return;
2778 
2779 	device_printf(sc->rl_dev, "watchdog timeout\n");
2780 	sc->rl_ifp->if_oerrors++;
2781 
2782 	re_txeof(sc);
2783 	re_rxeof(sc);
2784 	re_init_locked(sc);
2785 }
2786 
2787 /*
2788  * Stop the adapter and free any mbufs allocated to the
2789  * RX and TX lists.
2790  */
2791 static void
2792 re_stop(sc)
2793 	struct rl_softc		*sc;
2794 {
2795 	register int		i;
2796 	struct ifnet		*ifp;
2797 	struct rl_txdesc	*txd;
2798 	struct rl_rxdesc	*rxd;
2799 
2800 	RL_LOCK_ASSERT(sc);
2801 
2802 	ifp = sc->rl_ifp;
2803 
2804 	sc->rl_watchdog_timer = 0;
2805 	callout_stop(&sc->rl_stat_callout);
2806 	ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
2807 
2808 	CSR_WRITE_1(sc, RL_COMMAND, 0x00);
2809 	CSR_WRITE_2(sc, RL_IMR, 0x0000);
2810 	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
2811 
2812 	if (sc->rl_head != NULL) {
2813 		m_freem(sc->rl_head);
2814 		sc->rl_head = sc->rl_tail = NULL;
2815 	}
2816 
2817 	/* Free the TX list buffers. */
2818 
2819 	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
2820 		txd = &sc->rl_ldata.rl_tx_desc[i];
2821 		if (txd->tx_m != NULL) {
2822 			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
2823 			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
2824 			bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag,
2825 			    txd->tx_dmamap);
2826 			m_freem(txd->tx_m);
2827 			txd->tx_m = NULL;
2828 		}
2829 	}
2830 
2831 	/* Free the RX list buffers. */
2832 
2833 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
2834 		rxd = &sc->rl_ldata.rl_rx_desc[i];
2835 		if (rxd->rx_m != NULL) {
2836 			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
2837 			    rxd->rx_dmamap, BUS_DMASYNC_POSTREAD);
2838 			bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag,
2839 			    rxd->rx_dmamap);
2840 			m_freem(rxd->rx_m);
2841 			rxd->rx_m = NULL;
2842 		}
2843 	}
2844 }
2845 
2846 /*
2847  * Device suspend routine.  Stop the interface and save some PCI
2848  * settings in case the BIOS doesn't restore them properly on
2849  * resume.
2850  */
2851 static int
2852 re_suspend(dev)
2853 	device_t		dev;
2854 {
2855 	struct rl_softc		*sc;
2856 
2857 	sc = device_get_softc(dev);
2858 
2859 	RL_LOCK(sc);
2860 	re_stop(sc);
2861 	re_setwol(sc);
2862 	sc->suspended = 1;
2863 	RL_UNLOCK(sc);
2864 
2865 	return (0);
2866 }
2867 
2868 /*
2869  * Device resume routine.  Restore some PCI settings in case the BIOS
2870  * doesn't, re-enable busmastering, and restart the interface if
2871  * appropriate.
2872  */
2873 static int
2874 re_resume(dev)
2875 	device_t		dev;
2876 {
2877 	struct rl_softc		*sc;
2878 	struct ifnet		*ifp;
2879 
2880 	sc = device_get_softc(dev);
2881 
2882 	RL_LOCK(sc);
2883 
2884 	ifp = sc->rl_ifp;
2885 
2886 	/* reinitialize interface if necessary */
2887 	if (ifp->if_flags & IFF_UP)
2888 		re_init_locked(sc);
2889 
2890 	/*
2891 	 * Clear WOL matching such that normal Rx filtering
2892 	 * wouldn't interfere with WOL patterns.
2893 	 */
2894 	re_clrwol(sc);
2895 	sc->suspended = 0;
2896 	RL_UNLOCK(sc);
2897 
2898 	return (0);
2899 }
2900 
2901 /*
2902  * Stop all chip I/O so that the kernel's probe routines don't
2903  * get confused by errant DMAs when rebooting.
2904  */
2905 static int
2906 re_shutdown(dev)
2907 	device_t		dev;
2908 {
2909 	struct rl_softc		*sc;
2910 
2911 	sc = device_get_softc(dev);
2912 
2913 	RL_LOCK(sc);
2914 	re_stop(sc);
2915 	/*
2916 	 * Mark interface as down since otherwise we will panic if
2917 	 * interrupt comes in later on, which can happen in some
2918 	 * cases.
2919 	 */
2920 	sc->rl_ifp->if_flags &= ~IFF_UP;
2921 	re_setwol(sc);
2922 	RL_UNLOCK(sc);
2923 
2924 	return (0);
2925 }
2926 
2927 static void
2928 re_setwol(sc)
2929 	struct rl_softc		*sc;
2930 {
2931 	struct ifnet		*ifp;
2932 	int			pmc;
2933 	uint16_t		pmstat;
2934 	uint8_t			v;
2935 
2936 	RL_LOCK_ASSERT(sc);
2937 
2938 	if (pci_find_extcap(sc->rl_dev, PCIY_PMG, &pmc) != 0)
2939 		return;
2940 
2941 	ifp = sc->rl_ifp;
2942 	/* Enable config register write. */
2943 	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
2944 
2945 	/* Enable PME. */
2946 	v = CSR_READ_1(sc, RL_CFG1);
2947 	v &= ~RL_CFG1_PME;
2948 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
2949 		v |= RL_CFG1_PME;
2950 	CSR_WRITE_1(sc, RL_CFG1, v);
2951 
2952 	v = CSR_READ_1(sc, RL_CFG3);
2953 	v &= ~(RL_CFG3_WOL_LINK | RL_CFG3_WOL_MAGIC);
2954 	if ((ifp->if_capenable & IFCAP_WOL_MAGIC) != 0)
2955 		v |= RL_CFG3_WOL_MAGIC;
2956 	CSR_WRITE_1(sc, RL_CFG3, v);
2957 
2958 	/* Config register write done. */
2959 	CSR_WRITE_1(sc, RL_EECMD, 0);
2960 
2961 	v = CSR_READ_1(sc, RL_CFG5);
2962 	v &= ~(RL_CFG5_WOL_BCAST | RL_CFG5_WOL_MCAST | RL_CFG5_WOL_UCAST);
2963 	v &= ~RL_CFG5_WOL_LANWAKE;
2964 	if ((ifp->if_capenable & IFCAP_WOL_UCAST) != 0)
2965 		v |= RL_CFG5_WOL_UCAST;
2966 	if ((ifp->if_capenable & IFCAP_WOL_MCAST) != 0)
2967 		v |= RL_CFG5_WOL_MCAST | RL_CFG5_WOL_BCAST;
2968 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
2969 		v |= RL_CFG5_WOL_LANWAKE;
2970 	CSR_WRITE_1(sc, RL_CFG5, v);
2971 
2972 	/*
2973 	 * It seems that hardware resets its link speed to 100Mbps in
2974 	 * power down mode so switching to 100Mbps in driver is not
2975 	 * needed.
2976 	 */
2977 
2978 	/* Request PME if WOL is requested. */
2979 	pmstat = pci_read_config(sc->rl_dev, pmc + PCIR_POWER_STATUS, 2);
2980 	pmstat &= ~(PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE);
2981 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
2982 		pmstat |= PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE;
2983 	pci_write_config(sc->rl_dev, pmc + PCIR_POWER_STATUS, pmstat, 2);
2984 }
2985 
2986 static void
2987 re_clrwol(sc)
2988 	struct rl_softc		*sc;
2989 {
2990 	int			pmc;
2991 	uint8_t			v;
2992 
2993 	RL_LOCK_ASSERT(sc);
2994 
2995 	if (pci_find_extcap(sc->rl_dev, PCIY_PMG, &pmc) != 0)
2996 		return;
2997 
2998 	/* Enable config register write. */
2999 	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
3000 
3001 	v = CSR_READ_1(sc, RL_CFG3);
3002 	v &= ~(RL_CFG3_WOL_LINK | RL_CFG3_WOL_MAGIC);
3003 	CSR_WRITE_1(sc, RL_CFG3, v);
3004 
3005 	/* Config register write done. */
3006 	CSR_WRITE_1(sc, RL_EECMD, 0);
3007 
3008 	v = CSR_READ_1(sc, RL_CFG5);
3009 	v &= ~(RL_CFG5_WOL_BCAST | RL_CFG5_WOL_MCAST | RL_CFG5_WOL_UCAST);
3010 	v &= ~RL_CFG5_WOL_LANWAKE;
3011 	CSR_WRITE_1(sc, RL_CFG5, v);
3012 }
3013