xref: /freebsd/sys/dev/re/if_re.c (revision 13014ca04aad1931d41958b56f71a2c65b9a7a2c)
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 /* Tunables. */
159 static int msi_disable = 1;
160 TUNABLE_INT("hw.re.msi_disable", &msi_disable);
161 
162 #define RE_CSUM_FEATURES    (CSUM_IP | CSUM_TCP | CSUM_UDP)
163 
164 /*
165  * Various supported device vendors/types and their names.
166  */
167 static struct rl_type re_devs[] = {
168 	{ DLINK_VENDORID, DLINK_DEVICEID_528T, 0,
169 	    "D-Link DGE-528(T) Gigabit Ethernet Adapter" },
170 	{ RT_VENDORID, RT_DEVICEID_8139, 0,
171 	    "RealTek 8139C+ 10/100BaseTX" },
172 	{ RT_VENDORID, RT_DEVICEID_8101E, 0,
173 	    "RealTek 8101E/8102E/8102EL PCIe 10/100baseTX" },
174 	{ RT_VENDORID, RT_DEVICEID_8168, 0,
175 	    "RealTek 8168/8168B/8168C/8168CP/8111B/8111C/8111CP PCIe "
176 	    "Gigabit Ethernet" },
177 	{ RT_VENDORID, RT_DEVICEID_8169, 0,
178 	    "RealTek 8169/8169S/8169SB(L)/8110S/8110SB(L) Gigabit Ethernet" },
179 	{ RT_VENDORID, RT_DEVICEID_8169SC, 0,
180 	    "RealTek 8169SC/8110SC Single-chip Gigabit Ethernet" },
181 	{ COREGA_VENDORID, COREGA_DEVICEID_CGLAPCIGT, 0,
182 	    "Corega CG-LAPCIGT (RTL8169S) Gigabit Ethernet" },
183 	{ LINKSYS_VENDORID, LINKSYS_DEVICEID_EG1032, 0,
184 	    "Linksys EG1032 (RTL8169S) Gigabit Ethernet" },
185 	{ USR_VENDORID, USR_DEVICEID_997902, 0,
186 	    "US Robotics 997902 (RTL8169S) Gigabit Ethernet" }
187 };
188 
189 static struct rl_hwrev re_hwrevs[] = {
190 	{ RL_HWREV_8139, RL_8139,  "" },
191 	{ RL_HWREV_8139A, RL_8139, "A" },
192 	{ RL_HWREV_8139AG, RL_8139, "A-G" },
193 	{ RL_HWREV_8139B, RL_8139, "B" },
194 	{ RL_HWREV_8130, RL_8139, "8130" },
195 	{ RL_HWREV_8139C, RL_8139, "C" },
196 	{ RL_HWREV_8139D, RL_8139, "8139D/8100B/8100C" },
197 	{ RL_HWREV_8139CPLUS, RL_8139CPLUS, "C+"},
198 	{ RL_HWREV_8168_SPIN1, RL_8169, "8168"},
199 	{ RL_HWREV_8169, RL_8169, "8169"},
200 	{ RL_HWREV_8169S, RL_8169, "8169S"},
201 	{ RL_HWREV_8110S, RL_8169, "8110S"},
202 	{ RL_HWREV_8169_8110SB, RL_8169, "8169SB"},
203 	{ RL_HWREV_8169_8110SC, RL_8169, "8169SC"},
204 	{ RL_HWREV_8169_8110SBL, RL_8169, "8169SBL"},
205 	{ RL_HWREV_8100, RL_8139, "8100"},
206 	{ RL_HWREV_8101, RL_8139, "8101"},
207 	{ RL_HWREV_8100E, RL_8169, "8100E"},
208 	{ RL_HWREV_8101E, RL_8169, "8101E"},
209 	{ RL_HWREV_8102E, RL_8169, "8102E"},
210 	{ RL_HWREV_8102EL, RL_8169, "8102EL"},
211 	{ RL_HWREV_8168_SPIN2, RL_8169, "8168"},
212 	{ RL_HWREV_8168_SPIN3, RL_8169, "8168"},
213 	{ RL_HWREV_8168C, RL_8169, "8168C/8111C"},
214 	{ RL_HWREV_8168C_SPIN2, RL_8169, "8168C/8111C"},
215 	{ RL_HWREV_8168CP, RL_8169, "8168CP/8111CP"},
216 	{ 0, 0, NULL }
217 };
218 
219 static int re_probe		(device_t);
220 static int re_attach		(device_t);
221 static int re_detach		(device_t);
222 
223 static int re_encap		(struct rl_softc *, struct mbuf **);
224 
225 static void re_dma_map_addr	(void *, bus_dma_segment_t *, int, int);
226 static int re_allocmem		(device_t, struct rl_softc *);
227 static __inline void re_discard_rxbuf
228 				(struct rl_softc *, int);
229 static int re_newbuf		(struct rl_softc *, int);
230 static int re_rx_list_init	(struct rl_softc *);
231 static int re_tx_list_init	(struct rl_softc *);
232 #ifdef RE_FIXUP_RX
233 static __inline void re_fixup_rx
234 				(struct mbuf *);
235 #endif
236 static int re_rxeof		(struct rl_softc *);
237 static void re_txeof		(struct rl_softc *);
238 #ifdef DEVICE_POLLING
239 static void re_poll		(struct ifnet *, enum poll_cmd, int);
240 static void re_poll_locked	(struct ifnet *, enum poll_cmd, int);
241 #endif
242 static int re_intr		(void *);
243 static void re_tick		(void *);
244 static void re_tx_task		(void *, int);
245 static void re_int_task		(void *, int);
246 static void re_start		(struct ifnet *);
247 static int re_ioctl		(struct ifnet *, u_long, caddr_t);
248 static void re_init		(void *);
249 static void re_init_locked	(struct rl_softc *);
250 static void re_stop		(struct rl_softc *);
251 static void re_watchdog		(struct rl_softc *);
252 static int re_suspend		(device_t);
253 static int re_resume		(device_t);
254 static int re_shutdown		(device_t);
255 static int re_ifmedia_upd	(struct ifnet *);
256 static void re_ifmedia_sts	(struct ifnet *, struct ifmediareq *);
257 
258 static void re_eeprom_putbyte	(struct rl_softc *, int);
259 static void re_eeprom_getword	(struct rl_softc *, int, u_int16_t *);
260 static void re_read_eeprom	(struct rl_softc *, caddr_t, int, int);
261 static int re_gmii_readreg	(device_t, int, int);
262 static int re_gmii_writereg	(device_t, int, int, int);
263 
264 static int re_miibus_readreg	(device_t, int, int);
265 static int re_miibus_writereg	(device_t, int, int, int);
266 static void re_miibus_statchg	(device_t);
267 
268 static void re_setmulti		(struct rl_softc *);
269 static void re_reset		(struct rl_softc *);
270 static void re_setwol		(struct rl_softc *);
271 static void re_clrwol		(struct rl_softc *);
272 
273 #ifdef RE_DIAG
274 static int re_diag		(struct rl_softc *);
275 #endif
276 
277 static device_method_t re_methods[] = {
278 	/* Device interface */
279 	DEVMETHOD(device_probe,		re_probe),
280 	DEVMETHOD(device_attach,	re_attach),
281 	DEVMETHOD(device_detach,	re_detach),
282 	DEVMETHOD(device_suspend,	re_suspend),
283 	DEVMETHOD(device_resume,	re_resume),
284 	DEVMETHOD(device_shutdown,	re_shutdown),
285 
286 	/* bus interface */
287 	DEVMETHOD(bus_print_child,	bus_generic_print_child),
288 	DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
289 
290 	/* MII interface */
291 	DEVMETHOD(miibus_readreg,	re_miibus_readreg),
292 	DEVMETHOD(miibus_writereg,	re_miibus_writereg),
293 	DEVMETHOD(miibus_statchg,	re_miibus_statchg),
294 
295 	{ 0, 0 }
296 };
297 
298 static driver_t re_driver = {
299 	"re",
300 	re_methods,
301 	sizeof(struct rl_softc)
302 };
303 
304 static devclass_t re_devclass;
305 
306 DRIVER_MODULE(re, pci, re_driver, re_devclass, 0, 0);
307 DRIVER_MODULE(re, cardbus, re_driver, re_devclass, 0, 0);
308 DRIVER_MODULE(miibus, re, miibus_driver, miibus_devclass, 0, 0);
309 
310 #define EE_SET(x)					\
311 	CSR_WRITE_1(sc, RL_EECMD,			\
312 		CSR_READ_1(sc, RL_EECMD) | x)
313 
314 #define EE_CLR(x)					\
315 	CSR_WRITE_1(sc, RL_EECMD,			\
316 		CSR_READ_1(sc, RL_EECMD) & ~x)
317 
318 /*
319  * Send a read command and address to the EEPROM, check for ACK.
320  */
321 static void
322 re_eeprom_putbyte(struct rl_softc *sc, int addr)
323 {
324 	int			d, i;
325 
326 	d = addr | (RL_9346_READ << sc->rl_eewidth);
327 
328 	/*
329 	 * Feed in each bit and strobe the clock.
330 	 */
331 
332 	for (i = 1 << (sc->rl_eewidth + 3); i; i >>= 1) {
333 		if (d & i) {
334 			EE_SET(RL_EE_DATAIN);
335 		} else {
336 			EE_CLR(RL_EE_DATAIN);
337 		}
338 		DELAY(100);
339 		EE_SET(RL_EE_CLK);
340 		DELAY(150);
341 		EE_CLR(RL_EE_CLK);
342 		DELAY(100);
343 	}
344 }
345 
346 /*
347  * Read a word of data stored in the EEPROM at address 'addr.'
348  */
349 static void
350 re_eeprom_getword(struct rl_softc *sc, int addr, u_int16_t *dest)
351 {
352 	int			i;
353 	u_int16_t		word = 0;
354 
355 	/*
356 	 * Send address of word we want to read.
357 	 */
358 	re_eeprom_putbyte(sc, addr);
359 
360 	/*
361 	 * Start reading bits from EEPROM.
362 	 */
363 	for (i = 0x8000; i; i >>= 1) {
364 		EE_SET(RL_EE_CLK);
365 		DELAY(100);
366 		if (CSR_READ_1(sc, RL_EECMD) & RL_EE_DATAOUT)
367 			word |= i;
368 		EE_CLR(RL_EE_CLK);
369 		DELAY(100);
370 	}
371 
372 	*dest = word;
373 }
374 
375 /*
376  * Read a sequence of words from the EEPROM.
377  */
378 static void
379 re_read_eeprom(struct rl_softc *sc, caddr_t dest, int off, int cnt)
380 {
381 	int			i;
382 	u_int16_t		word = 0, *ptr;
383 
384 	CSR_SETBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM);
385 
386         DELAY(100);
387 
388 	for (i = 0; i < cnt; i++) {
389 		CSR_SETBIT_1(sc, RL_EECMD, RL_EE_SEL);
390 		re_eeprom_getword(sc, off + i, &word);
391 		CSR_CLRBIT_1(sc, RL_EECMD, RL_EE_SEL);
392 		ptr = (u_int16_t *)(dest + (i * 2));
393                 *ptr = word;
394 	}
395 
396 	CSR_CLRBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM);
397 }
398 
399 static int
400 re_gmii_readreg(device_t dev, int phy, int reg)
401 {
402 	struct rl_softc		*sc;
403 	u_int32_t		rval;
404 	int			i;
405 
406 	if (phy != 1)
407 		return (0);
408 
409 	sc = device_get_softc(dev);
410 
411 	/* Let the rgephy driver read the GMEDIASTAT register */
412 
413 	if (reg == RL_GMEDIASTAT) {
414 		rval = CSR_READ_1(sc, RL_GMEDIASTAT);
415 		return (rval);
416 	}
417 
418 	CSR_WRITE_4(sc, RL_PHYAR, reg << 16);
419 	DELAY(1000);
420 
421 	for (i = 0; i < RL_TIMEOUT; i++) {
422 		rval = CSR_READ_4(sc, RL_PHYAR);
423 		if (rval & RL_PHYAR_BUSY)
424 			break;
425 		DELAY(100);
426 	}
427 
428 	if (i == RL_TIMEOUT) {
429 		device_printf(sc->rl_dev, "PHY read failed\n");
430 		return (0);
431 	}
432 
433 	return (rval & RL_PHYAR_PHYDATA);
434 }
435 
436 static int
437 re_gmii_writereg(device_t dev, int phy, int reg, int data)
438 {
439 	struct rl_softc		*sc;
440 	u_int32_t		rval;
441 	int			i;
442 
443 	sc = device_get_softc(dev);
444 
445 	CSR_WRITE_4(sc, RL_PHYAR, (reg << 16) |
446 	    (data & RL_PHYAR_PHYDATA) | RL_PHYAR_BUSY);
447 	DELAY(1000);
448 
449 	for (i = 0; i < RL_TIMEOUT; i++) {
450 		rval = CSR_READ_4(sc, RL_PHYAR);
451 		if (!(rval & RL_PHYAR_BUSY))
452 			break;
453 		DELAY(100);
454 	}
455 
456 	if (i == RL_TIMEOUT) {
457 		device_printf(sc->rl_dev, "PHY write failed\n");
458 		return (0);
459 	}
460 
461 	return (0);
462 }
463 
464 static int
465 re_miibus_readreg(device_t dev, int phy, int reg)
466 {
467 	struct rl_softc		*sc;
468 	u_int16_t		rval = 0;
469 	u_int16_t		re8139_reg = 0;
470 
471 	sc = device_get_softc(dev);
472 
473 	if (sc->rl_type == RL_8169) {
474 		rval = re_gmii_readreg(dev, phy, reg);
475 		return (rval);
476 	}
477 
478 	/* Pretend the internal PHY is only at address 0 */
479 	if (phy) {
480 		return (0);
481 	}
482 	switch (reg) {
483 	case MII_BMCR:
484 		re8139_reg = RL_BMCR;
485 		break;
486 	case MII_BMSR:
487 		re8139_reg = RL_BMSR;
488 		break;
489 	case MII_ANAR:
490 		re8139_reg = RL_ANAR;
491 		break;
492 	case MII_ANER:
493 		re8139_reg = RL_ANER;
494 		break;
495 	case MII_ANLPAR:
496 		re8139_reg = RL_LPAR;
497 		break;
498 	case MII_PHYIDR1:
499 	case MII_PHYIDR2:
500 		return (0);
501 	/*
502 	 * Allow the rlphy driver to read the media status
503 	 * register. If we have a link partner which does not
504 	 * support NWAY, this is the register which will tell
505 	 * us the results of parallel detection.
506 	 */
507 	case RL_MEDIASTAT:
508 		rval = CSR_READ_1(sc, RL_MEDIASTAT);
509 		return (rval);
510 	default:
511 		device_printf(sc->rl_dev, "bad phy register\n");
512 		return (0);
513 	}
514 	rval = CSR_READ_2(sc, re8139_reg);
515 	if (sc->rl_type == RL_8139CPLUS && re8139_reg == RL_BMCR) {
516 		/* 8139C+ has different bit layout. */
517 		rval &= ~(BMCR_LOOP | BMCR_ISO);
518 	}
519 	return (rval);
520 }
521 
522 static int
523 re_miibus_writereg(device_t dev, int phy, int reg, int data)
524 {
525 	struct rl_softc		*sc;
526 	u_int16_t		re8139_reg = 0;
527 	int			rval = 0;
528 
529 	sc = device_get_softc(dev);
530 
531 	if (sc->rl_type == RL_8169) {
532 		rval = re_gmii_writereg(dev, phy, reg, data);
533 		return (rval);
534 	}
535 
536 	/* Pretend the internal PHY is only at address 0 */
537 	if (phy)
538 		return (0);
539 
540 	switch (reg) {
541 	case MII_BMCR:
542 		re8139_reg = RL_BMCR;
543 		if (sc->rl_type == RL_8139CPLUS) {
544 			/* 8139C+ has different bit layout. */
545 			data &= ~(BMCR_LOOP | BMCR_ISO);
546 		}
547 		break;
548 	case MII_BMSR:
549 		re8139_reg = RL_BMSR;
550 		break;
551 	case MII_ANAR:
552 		re8139_reg = RL_ANAR;
553 		break;
554 	case MII_ANER:
555 		re8139_reg = RL_ANER;
556 		break;
557 	case MII_ANLPAR:
558 		re8139_reg = RL_LPAR;
559 		break;
560 	case MII_PHYIDR1:
561 	case MII_PHYIDR2:
562 		return (0);
563 		break;
564 	default:
565 		device_printf(sc->rl_dev, "bad phy register\n");
566 		return (0);
567 	}
568 	CSR_WRITE_2(sc, re8139_reg, data);
569 	return (0);
570 }
571 
572 static void
573 re_miibus_statchg(device_t dev)
574 {
575 
576 }
577 
578 /*
579  * Program the 64-bit multicast hash filter.
580  */
581 static void
582 re_setmulti(struct rl_softc *sc)
583 {
584 	struct ifnet		*ifp;
585 	int			h = 0;
586 	u_int32_t		hashes[2] = { 0, 0 };
587 	struct ifmultiaddr	*ifma;
588 	u_int32_t		rxfilt;
589 	int			mcnt = 0;
590 
591 	RL_LOCK_ASSERT(sc);
592 
593 	ifp = sc->rl_ifp;
594 
595 
596 	rxfilt = CSR_READ_4(sc, RL_RXCFG);
597 	rxfilt &= ~(RL_RXCFG_RX_ALLPHYS | RL_RXCFG_RX_MULTI);
598 	if (ifp->if_flags & IFF_ALLMULTI || ifp->if_flags & IFF_PROMISC) {
599 		if (ifp->if_flags & IFF_PROMISC)
600 			rxfilt |= RL_RXCFG_RX_ALLPHYS;
601 		/*
602 		 * Unlike other hardwares, we have to explicitly set
603 		 * RL_RXCFG_RX_MULTI to receive multicast frames in
604 		 * promiscuous mode.
605 		 */
606 		rxfilt |= RL_RXCFG_RX_MULTI;
607 		CSR_WRITE_4(sc, RL_RXCFG, rxfilt);
608 		CSR_WRITE_4(sc, RL_MAR0, 0xFFFFFFFF);
609 		CSR_WRITE_4(sc, RL_MAR4, 0xFFFFFFFF);
610 		return;
611 	}
612 
613 	/* first, zot all the existing hash bits */
614 	CSR_WRITE_4(sc, RL_MAR0, 0);
615 	CSR_WRITE_4(sc, RL_MAR4, 0);
616 
617 	/* now program new ones */
618 	IF_ADDR_LOCK(ifp);
619 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
620 		if (ifma->ifma_addr->sa_family != AF_LINK)
621 			continue;
622 		h = ether_crc32_be(LLADDR((struct sockaddr_dl *)
623 		    ifma->ifma_addr), ETHER_ADDR_LEN) >> 26;
624 		if (h < 32)
625 			hashes[0] |= (1 << h);
626 		else
627 			hashes[1] |= (1 << (h - 32));
628 		mcnt++;
629 	}
630 	IF_ADDR_UNLOCK(ifp);
631 
632 	if (mcnt)
633 		rxfilt |= RL_RXCFG_RX_MULTI;
634 	else
635 		rxfilt &= ~RL_RXCFG_RX_MULTI;
636 
637 	CSR_WRITE_4(sc, RL_RXCFG, rxfilt);
638 
639 	/*
640 	 * For some unfathomable reason, RealTek decided to reverse
641 	 * the order of the multicast hash registers in the PCI Express
642 	 * parts. This means we have to write the hash pattern in reverse
643 	 * order for those devices.
644 	 */
645 
646 	if ((sc->rl_flags & RL_FLAG_INVMAR) != 0) {
647 		CSR_WRITE_4(sc, RL_MAR0, bswap32(hashes[1]));
648 		CSR_WRITE_4(sc, RL_MAR4, bswap32(hashes[0]));
649 	} else {
650 		CSR_WRITE_4(sc, RL_MAR0, hashes[0]);
651 		CSR_WRITE_4(sc, RL_MAR4, hashes[1]);
652 	}
653 }
654 
655 static void
656 re_reset(struct rl_softc *sc)
657 {
658 	int			i;
659 
660 	RL_LOCK_ASSERT(sc);
661 
662 	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RESET);
663 
664 	for (i = 0; i < RL_TIMEOUT; i++) {
665 		DELAY(10);
666 		if (!(CSR_READ_1(sc, RL_COMMAND) & RL_CMD_RESET))
667 			break;
668 	}
669 	if (i == RL_TIMEOUT)
670 		device_printf(sc->rl_dev, "reset never completed!\n");
671 
672 	CSR_WRITE_1(sc, 0x82, 1);
673 }
674 
675 #ifdef RE_DIAG
676 
677 /*
678  * The following routine is designed to test for a defect on some
679  * 32-bit 8169 cards. Some of these NICs have the REQ64# and ACK64#
680  * lines connected to the bus, however for a 32-bit only card, they
681  * should be pulled high. The result of this defect is that the
682  * NIC will not work right if you plug it into a 64-bit slot: DMA
683  * operations will be done with 64-bit transfers, which will fail
684  * because the 64-bit data lines aren't connected.
685  *
686  * There's no way to work around this (short of talking a soldering
687  * iron to the board), however we can detect it. The method we use
688  * here is to put the NIC into digital loopback mode, set the receiver
689  * to promiscuous mode, and then try to send a frame. We then compare
690  * the frame data we sent to what was received. If the data matches,
691  * then the NIC is working correctly, otherwise we know the user has
692  * a defective NIC which has been mistakenly plugged into a 64-bit PCI
693  * slot. In the latter case, there's no way the NIC can work correctly,
694  * so we print out a message on the console and abort the device attach.
695  */
696 
697 static int
698 re_diag(struct rl_softc *sc)
699 {
700 	struct ifnet		*ifp = sc->rl_ifp;
701 	struct mbuf		*m0;
702 	struct ether_header	*eh;
703 	struct rl_desc		*cur_rx;
704 	u_int16_t		status;
705 	u_int32_t		rxstat;
706 	int			total_len, i, error = 0, phyaddr;
707 	u_int8_t		dst[] = { 0x00, 'h', 'e', 'l', 'l', 'o' };
708 	u_int8_t		src[] = { 0x00, 'w', 'o', 'r', 'l', 'd' };
709 
710 	/* Allocate a single mbuf */
711 	MGETHDR(m0, M_DONTWAIT, MT_DATA);
712 	if (m0 == NULL)
713 		return (ENOBUFS);
714 
715 	RL_LOCK(sc);
716 
717 	/*
718 	 * Initialize the NIC in test mode. This sets the chip up
719 	 * so that it can send and receive frames, but performs the
720 	 * following special functions:
721 	 * - Puts receiver in promiscuous mode
722 	 * - Enables digital loopback mode
723 	 * - Leaves interrupts turned off
724 	 */
725 
726 	ifp->if_flags |= IFF_PROMISC;
727 	sc->rl_testmode = 1;
728 	re_reset(sc);
729 	re_init_locked(sc);
730 	sc->rl_flags |= RL_FLAG_LINK;
731 	if (sc->rl_type == RL_8169)
732 		phyaddr = 1;
733 	else
734 		phyaddr = 0;
735 
736 	re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_RESET);
737 	for (i = 0; i < RL_TIMEOUT; i++) {
738 		status = re_miibus_readreg(sc->rl_dev, phyaddr, MII_BMCR);
739 		if (!(status & BMCR_RESET))
740 			break;
741 	}
742 
743 	re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_LOOP);
744 	CSR_WRITE_2(sc, RL_ISR, RL_INTRS);
745 
746 	DELAY(100000);
747 
748 	/* Put some data in the mbuf */
749 
750 	eh = mtod(m0, struct ether_header *);
751 	bcopy ((char *)&dst, eh->ether_dhost, ETHER_ADDR_LEN);
752 	bcopy ((char *)&src, eh->ether_shost, ETHER_ADDR_LEN);
753 	eh->ether_type = htons(ETHERTYPE_IP);
754 	m0->m_pkthdr.len = m0->m_len = ETHER_MIN_LEN - ETHER_CRC_LEN;
755 
756 	/*
757 	 * Queue the packet, start transmission.
758 	 * Note: IF_HANDOFF() ultimately calls re_start() for us.
759 	 */
760 
761 	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
762 	RL_UNLOCK(sc);
763 	/* XXX: re_diag must not be called when in ALTQ mode */
764 	IF_HANDOFF(&ifp->if_snd, m0, ifp);
765 	RL_LOCK(sc);
766 	m0 = NULL;
767 
768 	/* Wait for it to propagate through the chip */
769 
770 	DELAY(100000);
771 	for (i = 0; i < RL_TIMEOUT; i++) {
772 		status = CSR_READ_2(sc, RL_ISR);
773 		CSR_WRITE_2(sc, RL_ISR, status);
774 		if ((status & (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK)) ==
775 		    (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK))
776 			break;
777 		DELAY(10);
778 	}
779 
780 	if (i == RL_TIMEOUT) {
781 		device_printf(sc->rl_dev,
782 		    "diagnostic failed, failed to receive packet in"
783 		    " loopback mode\n");
784 		error = EIO;
785 		goto done;
786 	}
787 
788 	/*
789 	 * The packet should have been dumped into the first
790 	 * entry in the RX DMA ring. Grab it from there.
791 	 */
792 
793 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
794 	    sc->rl_ldata.rl_rx_list_map,
795 	    BUS_DMASYNC_POSTREAD);
796 	bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag,
797 	    sc->rl_ldata.rl_rx_desc[0].rx_dmamap,
798 	    BUS_DMASYNC_POSTREAD);
799 	bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag,
800 	    sc->rl_ldata.rl_rx_desc[0].rx_dmamap);
801 
802 	m0 = sc->rl_ldata.rl_rx_desc[0].rx_m;
803 	sc->rl_ldata.rl_rx_desc[0].rx_m = NULL;
804 	eh = mtod(m0, struct ether_header *);
805 
806 	cur_rx = &sc->rl_ldata.rl_rx_list[0];
807 	total_len = RL_RXBYTES(cur_rx);
808 	rxstat = le32toh(cur_rx->rl_cmdstat);
809 
810 	if (total_len != ETHER_MIN_LEN) {
811 		device_printf(sc->rl_dev,
812 		    "diagnostic failed, received short packet\n");
813 		error = EIO;
814 		goto done;
815 	}
816 
817 	/* Test that the received packet data matches what we sent. */
818 
819 	if (bcmp((char *)&eh->ether_dhost, (char *)&dst, ETHER_ADDR_LEN) ||
820 	    bcmp((char *)&eh->ether_shost, (char *)&src, ETHER_ADDR_LEN) ||
821 	    ntohs(eh->ether_type) != ETHERTYPE_IP) {
822 		device_printf(sc->rl_dev, "WARNING, DMA FAILURE!\n");
823 		device_printf(sc->rl_dev, "expected TX data: %6D/%6D/0x%x\n",
824 		    dst, ":", src, ":", ETHERTYPE_IP);
825 		device_printf(sc->rl_dev, "received RX data: %6D/%6D/0x%x\n",
826 		    eh->ether_dhost, ":",  eh->ether_shost, ":",
827 		    ntohs(eh->ether_type));
828 		device_printf(sc->rl_dev, "You may have a defective 32-bit "
829 		    "NIC plugged into a 64-bit PCI slot.\n");
830 		device_printf(sc->rl_dev, "Please re-install the NIC in a "
831 		    "32-bit slot for proper operation.\n");
832 		device_printf(sc->rl_dev, "Read the re(4) man page for more "
833 		    "details.\n");
834 		error = EIO;
835 	}
836 
837 done:
838 	/* Turn interface off, release resources */
839 
840 	sc->rl_testmode = 0;
841 	sc->rl_flags &= ~RL_FLAG_LINK;
842 	ifp->if_flags &= ~IFF_PROMISC;
843 	re_stop(sc);
844 	if (m0 != NULL)
845 		m_freem(m0);
846 
847 	RL_UNLOCK(sc);
848 
849 	return (error);
850 }
851 
852 #endif
853 
854 /*
855  * Probe for a RealTek 8139C+/8169/8110 chip. Check the PCI vendor and device
856  * IDs against our list and return a device name if we find a match.
857  */
858 static int
859 re_probe(device_t dev)
860 {
861 	struct rl_type		*t;
862 	uint16_t		devid, vendor;
863 	uint16_t		revid, sdevid;
864 	int			i;
865 
866 	vendor = pci_get_vendor(dev);
867 	devid = pci_get_device(dev);
868 	revid = pci_get_revid(dev);
869 	sdevid = pci_get_subdevice(dev);
870 
871 	if (vendor == LINKSYS_VENDORID && devid == LINKSYS_DEVICEID_EG1032) {
872 		if (sdevid != LINKSYS_SUBDEVICE_EG1032_REV3) {
873 			/*
874 			 * Only attach to rev. 3 of the Linksys EG1032 adapter.
875 			 * Rev. 2 is supported by sk(4).
876 			 */
877 			return (ENXIO);
878 		}
879 	}
880 
881 	if (vendor == RT_VENDORID && devid == RT_DEVICEID_8139) {
882 		if (revid != 0x20) {
883 			/* 8139, let rl(4) take care of this device. */
884 			return (ENXIO);
885 		}
886 	}
887 
888 	t = re_devs;
889 	for (i = 0; i < sizeof(re_devs) / sizeof(re_devs[0]); i++, t++) {
890 		if (vendor == t->rl_vid && devid == t->rl_did) {
891 			device_set_desc(dev, t->rl_name);
892 			return (BUS_PROBE_DEFAULT);
893 		}
894 	}
895 
896 	return (ENXIO);
897 }
898 
899 /*
900  * Map a single buffer address.
901  */
902 
903 static void
904 re_dma_map_addr(void *arg, bus_dma_segment_t *segs, int nseg, int error)
905 {
906 	bus_addr_t		*addr;
907 
908 	if (error)
909 		return;
910 
911 	KASSERT(nseg == 1, ("too many DMA segments, %d should be 1", nseg));
912 	addr = arg;
913 	*addr = segs->ds_addr;
914 }
915 
916 static int
917 re_allocmem(device_t dev, struct rl_softc *sc)
918 {
919 	bus_size_t		rx_list_size, tx_list_size;
920 	int			error;
921 	int			i;
922 
923 	rx_list_size = sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc);
924 	tx_list_size = sc->rl_ldata.rl_tx_desc_cnt * sizeof(struct rl_desc);
925 
926 	/*
927 	 * Allocate the parent bus DMA tag appropriate for PCI.
928 	 * In order to use DAC, RL_CPLUSCMD_PCI_DAC bit of RL_CPLUS_CMD
929 	 * register should be set. However some RealTek chips are known
930 	 * to be buggy on DAC handling, therefore disable DAC by limiting
931 	 * DMA address space to 32bit. PCIe variants of RealTek chips
932 	 * may not have the limitation but I took safer path.
933 	 */
934 	error = bus_dma_tag_create(bus_get_dma_tag(dev), 1, 0,
935 	    BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL, NULL,
936 	    BUS_SPACE_MAXSIZE_32BIT, 0, BUS_SPACE_MAXSIZE_32BIT, 0,
937 	    NULL, NULL, &sc->rl_parent_tag);
938 	if (error) {
939 		device_printf(dev, "could not allocate parent DMA tag\n");
940 		return (error);
941 	}
942 
943 	/*
944 	 * Allocate map for TX mbufs.
945 	 */
946 	error = bus_dma_tag_create(sc->rl_parent_tag, 1, 0,
947 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL,
948 	    NULL, MCLBYTES * RL_NTXSEGS, RL_NTXSEGS, 4096, 0,
949 	    NULL, NULL, &sc->rl_ldata.rl_tx_mtag);
950 	if (error) {
951 		device_printf(dev, "could not allocate TX DMA tag\n");
952 		return (error);
953 	}
954 
955 	/*
956 	 * Allocate map for RX mbufs.
957 	 */
958 
959 	error = bus_dma_tag_create(sc->rl_parent_tag, sizeof(uint64_t), 0,
960 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
961 	    MCLBYTES, 1, MCLBYTES, 0, NULL, NULL, &sc->rl_ldata.rl_rx_mtag);
962 	if (error) {
963 		device_printf(dev, "could not allocate RX DMA tag\n");
964 		return (error);
965 	}
966 
967 	/*
968 	 * Allocate map for TX descriptor list.
969 	 */
970 	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
971 	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
972 	    NULL, tx_list_size, 1, tx_list_size, 0,
973 	    NULL, NULL, &sc->rl_ldata.rl_tx_list_tag);
974 	if (error) {
975 		device_printf(dev, "could not allocate TX DMA ring tag\n");
976 		return (error);
977 	}
978 
979 	/* Allocate DMA'able memory for the TX ring */
980 
981 	error = bus_dmamem_alloc(sc->rl_ldata.rl_tx_list_tag,
982 	    (void **)&sc->rl_ldata.rl_tx_list,
983 	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
984 	    &sc->rl_ldata.rl_tx_list_map);
985 	if (error) {
986 		device_printf(dev, "could not allocate TX DMA ring\n");
987 		return (error);
988 	}
989 
990 	/* Load the map for the TX ring. */
991 
992 	sc->rl_ldata.rl_tx_list_addr = 0;
993 	error = bus_dmamap_load(sc->rl_ldata.rl_tx_list_tag,
994 	     sc->rl_ldata.rl_tx_list_map, sc->rl_ldata.rl_tx_list,
995 	     tx_list_size, re_dma_map_addr,
996 	     &sc->rl_ldata.rl_tx_list_addr, BUS_DMA_NOWAIT);
997 	if (error != 0 || sc->rl_ldata.rl_tx_list_addr == 0) {
998 		device_printf(dev, "could not load TX DMA ring\n");
999 		return (ENOMEM);
1000 	}
1001 
1002 	/* Create DMA maps for TX buffers */
1003 
1004 	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
1005 		error = bus_dmamap_create(sc->rl_ldata.rl_tx_mtag, 0,
1006 		    &sc->rl_ldata.rl_tx_desc[i].tx_dmamap);
1007 		if (error) {
1008 			device_printf(dev, "could not create DMA map for TX\n");
1009 			return (error);
1010 		}
1011 	}
1012 
1013 	/*
1014 	 * Allocate map for RX descriptor list.
1015 	 */
1016 	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
1017 	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1018 	    NULL, rx_list_size, 1, rx_list_size, 0,
1019 	    NULL, NULL, &sc->rl_ldata.rl_rx_list_tag);
1020 	if (error) {
1021 		device_printf(dev, "could not create RX DMA ring tag\n");
1022 		return (error);
1023 	}
1024 
1025 	/* Allocate DMA'able memory for the RX ring */
1026 
1027 	error = bus_dmamem_alloc(sc->rl_ldata.rl_rx_list_tag,
1028 	    (void **)&sc->rl_ldata.rl_rx_list,
1029 	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1030 	    &sc->rl_ldata.rl_rx_list_map);
1031 	if (error) {
1032 		device_printf(dev, "could not allocate RX DMA ring\n");
1033 		return (error);
1034 	}
1035 
1036 	/* Load the map for the RX ring. */
1037 
1038 	sc->rl_ldata.rl_rx_list_addr = 0;
1039 	error = bus_dmamap_load(sc->rl_ldata.rl_rx_list_tag,
1040 	     sc->rl_ldata.rl_rx_list_map, sc->rl_ldata.rl_rx_list,
1041 	     rx_list_size, re_dma_map_addr,
1042 	     &sc->rl_ldata.rl_rx_list_addr, BUS_DMA_NOWAIT);
1043 	if (error != 0 || sc->rl_ldata.rl_rx_list_addr == 0) {
1044 		device_printf(dev, "could not load RX DMA ring\n");
1045 		return (ENOMEM);
1046 	}
1047 
1048 	/* Create DMA maps for RX buffers */
1049 
1050 	error = bus_dmamap_create(sc->rl_ldata.rl_rx_mtag, 0,
1051 	    &sc->rl_ldata.rl_rx_sparemap);
1052 	if (error) {
1053 		device_printf(dev, "could not create spare DMA map for RX\n");
1054 		return (error);
1055 	}
1056 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1057 		error = bus_dmamap_create(sc->rl_ldata.rl_rx_mtag, 0,
1058 		    &sc->rl_ldata.rl_rx_desc[i].rx_dmamap);
1059 		if (error) {
1060 			device_printf(dev, "could not create DMA map for RX\n");
1061 			return (error);
1062 		}
1063 	}
1064 
1065 	return (0);
1066 }
1067 
1068 /*
1069  * Attach the interface. Allocate softc structures, do ifmedia
1070  * setup and ethernet/BPF attach.
1071  */
1072 static int
1073 re_attach(device_t dev)
1074 {
1075 	u_char			eaddr[ETHER_ADDR_LEN];
1076 	u_int16_t		as[ETHER_ADDR_LEN / 2];
1077 	struct rl_softc		*sc;
1078 	struct ifnet		*ifp;
1079 	struct rl_hwrev		*hw_rev;
1080 	int			hwrev;
1081 	u_int16_t		devid, re_did = 0;
1082 	int			error = 0, rid, i;
1083 	int			msic, reg;
1084 	uint8_t			cfg;
1085 
1086 	sc = device_get_softc(dev);
1087 	sc->rl_dev = dev;
1088 
1089 	mtx_init(&sc->rl_mtx, device_get_nameunit(dev), MTX_NETWORK_LOCK,
1090 	    MTX_DEF);
1091 	callout_init_mtx(&sc->rl_stat_callout, &sc->rl_mtx, 0);
1092 
1093 	/*
1094 	 * Map control/status registers.
1095 	 */
1096 	pci_enable_busmaster(dev);
1097 
1098 	devid = pci_get_device(dev);
1099 	/* Prefer memory space register mapping over IO space. */
1100 	sc->rl_res_id = PCIR_BAR(1);
1101 	sc->rl_res_type = SYS_RES_MEMORY;
1102 	/* RTL8168/8101E seems to use different BARs. */
1103 	if (devid == RT_DEVICEID_8168 || devid == RT_DEVICEID_8101E)
1104 		sc->rl_res_id = PCIR_BAR(2);
1105 	sc->rl_res = bus_alloc_resource_any(dev, sc->rl_res_type,
1106 	    &sc->rl_res_id, RF_ACTIVE);
1107 
1108 	if (sc->rl_res == NULL) {
1109 		sc->rl_res_id = PCIR_BAR(0);
1110 		sc->rl_res_type = SYS_RES_IOPORT;
1111 		sc->rl_res = bus_alloc_resource_any(dev, sc->rl_res_type,
1112 		    &sc->rl_res_id, RF_ACTIVE);
1113 		if (sc->rl_res == NULL) {
1114 			device_printf(dev, "couldn't map ports/memory\n");
1115 			error = ENXIO;
1116 			goto fail;
1117 		}
1118 	}
1119 
1120 	sc->rl_btag = rman_get_bustag(sc->rl_res);
1121 	sc->rl_bhandle = rman_get_bushandle(sc->rl_res);
1122 
1123 	msic = 0;
1124 	if (pci_find_extcap(dev, PCIY_EXPRESS, &reg) == 0) {
1125 		msic = pci_msi_count(dev);
1126 		if (bootverbose)
1127 			device_printf(dev, "MSI count : %d\n", msic);
1128 	}
1129 	if (msic == RL_MSI_MESSAGES  && msi_disable == 0) {
1130 		if (pci_alloc_msi(dev, &msic) == 0) {
1131 			if (msic == RL_MSI_MESSAGES) {
1132 				device_printf(dev, "Using %d MSI messages\n",
1133 				    msic);
1134 				sc->rl_flags |= RL_FLAG_MSI;
1135 				/* Explicitly set MSI enable bit. */
1136 				CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1137 				cfg = CSR_READ_1(sc, RL_CFG2);
1138 				cfg |= RL_CFG2_MSI;
1139 				CSR_WRITE_1(sc, RL_CFG2, cfg);
1140 				CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1141 			} else
1142 				pci_release_msi(dev);
1143 		}
1144 	}
1145 
1146 	/* Allocate interrupt */
1147 	if ((sc->rl_flags & RL_FLAG_MSI) == 0) {
1148 		rid = 0;
1149 		sc->rl_irq[0] = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
1150 		    RF_SHAREABLE | RF_ACTIVE);
1151 		if (sc->rl_irq[0] == NULL) {
1152 			device_printf(dev, "couldn't allocate IRQ resources\n");
1153 			error = ENXIO;
1154 			goto fail;
1155 		}
1156 	} else {
1157 		for (i = 0, rid = 1; i < RL_MSI_MESSAGES; i++, rid++) {
1158 			sc->rl_irq[i] = bus_alloc_resource_any(dev,
1159 			    SYS_RES_IRQ, &rid, RF_ACTIVE);
1160 			if (sc->rl_irq[i] == NULL) {
1161 				device_printf(dev,
1162 				    "couldn't llocate IRQ resources for "
1163 				    "message %d\n", rid);
1164 				error = ENXIO;
1165 				goto fail;
1166 			}
1167 		}
1168 	}
1169 
1170 	if ((sc->rl_flags & RL_FLAG_MSI) == 0) {
1171 		CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1172 		cfg = CSR_READ_1(sc, RL_CFG2);
1173 		if ((cfg & RL_CFG2_MSI) != 0) {
1174 			device_printf(dev, "turning off MSI enable bit.\n");
1175 			cfg &= ~RL_CFG2_MSI;
1176 			CSR_WRITE_1(sc, RL_CFG2, cfg);
1177 		}
1178 		CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1179 	}
1180 
1181 	/* Reset the adapter. */
1182 	RL_LOCK(sc);
1183 	re_reset(sc);
1184 	RL_UNLOCK(sc);
1185 
1186 	hw_rev = re_hwrevs;
1187 	hwrev = CSR_READ_4(sc, RL_TXCFG);
1188 	device_printf(dev, "Chip rev. 0x%08x\n", hwrev & 0x7c800000);
1189 	device_printf(dev, "MAC rev. 0x%08x\n", hwrev & 0x00700000);
1190 	hwrev &= RL_TXCFG_HWREV;
1191 	while (hw_rev->rl_desc != NULL) {
1192 		if (hw_rev->rl_rev == hwrev) {
1193 			sc->rl_type = hw_rev->rl_type;
1194 			break;
1195 		}
1196 		hw_rev++;
1197 	}
1198 	if (hw_rev->rl_desc == NULL) {
1199 		device_printf(dev, "Unknown H/W revision: 0x%08x\n", hwrev);
1200 		error = ENXIO;
1201 		goto fail;
1202 	}
1203 
1204 	switch (hw_rev->rl_rev) {
1205 	case RL_HWREV_8139CPLUS:
1206 		sc->rl_flags |= RL_FLAG_NOJUMBO;
1207 		break;
1208 	case RL_HWREV_8100E:
1209 	case RL_HWREV_8101E:
1210 		sc->rl_flags |= RL_FLAG_NOJUMBO | RL_FLAG_INVMAR |
1211 		    RL_FLAG_PHYWAKE;
1212 		break;
1213 	case RL_HWREV_8102E:
1214 	case RL_HWREV_8102EL:
1215 		sc->rl_flags |= RL_FLAG_NOJUMBO | RL_FLAG_INVMAR |
1216 		    RL_FLAG_PHYWAKE | RL_FLAG_PAR | RL_FLAG_DESCV2 |
1217 		    RL_FLAG_MACSTAT;
1218 		break;
1219 	case RL_HWREV_8168_SPIN1:
1220 	case RL_HWREV_8168_SPIN2:
1221 	case RL_HWREV_8168_SPIN3:
1222 		sc->rl_flags |= RL_FLAG_INVMAR | RL_FLAG_PHYWAKE |
1223 		    RL_FLAG_MACSTAT;
1224 		break;
1225 	case RL_HWREV_8168C:
1226 	case RL_HWREV_8168C_SPIN2:
1227 	case RL_HWREV_8168CP:
1228 		sc->rl_flags |= RL_FLAG_INVMAR | RL_FLAG_PHYWAKE |
1229 		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT;
1230 		/*
1231 		 * These controllers support jumbo frame but it seems
1232 		 * that enabling it requires touching additional magic
1233 		 * registers. Depending on MAC revisions some
1234 		 * controllers need to disable checksum offload. So
1235 		 * disable jumbo frame until I have better idea what
1236 		 * it really requires to make it support.
1237 		 * RTL8168C/CP : supports up to 6KB jumbo frame.
1238 		 * RTL8111C/CP : supports up to 9KB jumbo frame.
1239 		 */
1240 		sc->rl_flags |= RL_FLAG_NOJUMBO;
1241 		break;
1242 	case RL_HWREV_8169_8110SB:
1243 	case RL_HWREV_8169_8110SC:
1244 	case RL_HWREV_8169_8110SBL:
1245 		sc->rl_flags |= RL_FLAG_PHYWAKE;
1246 		break;
1247 	default:
1248 		break;
1249 	}
1250 
1251 	/* Enable PME. */
1252 	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1253 	cfg = CSR_READ_1(sc, RL_CFG1);
1254 	cfg |= RL_CFG1_PME;
1255 	CSR_WRITE_1(sc, RL_CFG1, cfg);
1256 	cfg = CSR_READ_1(sc, RL_CFG5);
1257 	cfg &= RL_CFG5_PME_STS;
1258 	CSR_WRITE_1(sc, RL_CFG5, cfg);
1259 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1260 
1261 	if ((sc->rl_flags & RL_FLAG_PAR) != 0) {
1262 		/*
1263 		 * XXX Should have a better way to extract station
1264 		 * address from EEPROM.
1265 		 */
1266 		for (i = 0; i < ETHER_ADDR_LEN; i++)
1267 			eaddr[i] = CSR_READ_1(sc, RL_IDR0 + i);
1268 	} else {
1269 		sc->rl_eewidth = RL_9356_ADDR_LEN;
1270 		re_read_eeprom(sc, (caddr_t)&re_did, 0, 1);
1271 		if (re_did != 0x8129)
1272 			sc->rl_eewidth = RL_9346_ADDR_LEN;
1273 
1274 		/*
1275 		 * Get station address from the EEPROM.
1276 		 */
1277 		re_read_eeprom(sc, (caddr_t)as, RL_EE_EADDR, 3);
1278 		for (i = 0; i < ETHER_ADDR_LEN / 2; i++)
1279 			as[i] = le16toh(as[i]);
1280 		bcopy(as, eaddr, sizeof(eaddr));
1281 	}
1282 
1283 	if (sc->rl_type == RL_8169) {
1284 		/* Set RX length mask and number of descriptors. */
1285 		sc->rl_rxlenmask = RL_RDESC_STAT_GFRAGLEN;
1286 		sc->rl_txstart = RL_GTXSTART;
1287 		sc->rl_ldata.rl_tx_desc_cnt = RL_8169_TX_DESC_CNT;
1288 		sc->rl_ldata.rl_rx_desc_cnt = RL_8169_RX_DESC_CNT;
1289 	} else {
1290 		/* Set RX length mask and number of descriptors. */
1291 		sc->rl_rxlenmask = RL_RDESC_STAT_FRAGLEN;
1292 		sc->rl_txstart = RL_TXSTART;
1293 		sc->rl_ldata.rl_tx_desc_cnt = RL_8139_TX_DESC_CNT;
1294 		sc->rl_ldata.rl_rx_desc_cnt = RL_8139_RX_DESC_CNT;
1295 	}
1296 
1297 	error = re_allocmem(dev, sc);
1298 	if (error)
1299 		goto fail;
1300 
1301 	ifp = sc->rl_ifp = if_alloc(IFT_ETHER);
1302 	if (ifp == NULL) {
1303 		device_printf(dev, "can not if_alloc()\n");
1304 		error = ENOSPC;
1305 		goto fail;
1306 	}
1307 
1308 	/* Take PHY out of power down mode. */
1309 	if ((sc->rl_flags & RL_FLAG_PHYWAKE) != 0) {
1310 		re_gmii_writereg(dev, 1, 0x1f, 0);
1311 		re_gmii_writereg(dev, 1, 0x0e, 0);
1312 	}
1313 
1314 	/* Do MII setup */
1315 	if (mii_phy_probe(dev, &sc->rl_miibus,
1316 	    re_ifmedia_upd, re_ifmedia_sts)) {
1317 		device_printf(dev, "MII without any phy!\n");
1318 		error = ENXIO;
1319 		goto fail;
1320 	}
1321 
1322 	ifp->if_softc = sc;
1323 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1324 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1325 	ifp->if_ioctl = re_ioctl;
1326 	ifp->if_start = re_start;
1327 	ifp->if_hwassist = RE_CSUM_FEATURES;
1328 	ifp->if_capabilities = IFCAP_HWCSUM;
1329 	ifp->if_capenable = ifp->if_capabilities;
1330 	ifp->if_init = re_init;
1331 	IFQ_SET_MAXLEN(&ifp->if_snd, RL_IFQ_MAXLEN);
1332 	ifp->if_snd.ifq_drv_maxlen = RL_IFQ_MAXLEN;
1333 	IFQ_SET_READY(&ifp->if_snd);
1334 
1335 	TASK_INIT(&sc->rl_txtask, 1, re_tx_task, ifp);
1336 	TASK_INIT(&sc->rl_inttask, 0, re_int_task, sc);
1337 
1338 	/*
1339 	 * XXX
1340 	 * Still have no idea how to make TSO work on 8168C, 8168CP,
1341 	 * 8111C and 8111CP.
1342 	 */
1343 	if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
1344 		ifp->if_hwassist |= CSUM_TSO;
1345 		ifp->if_capabilities |= IFCAP_TSO4;
1346 	}
1347 
1348 	/*
1349 	 * Call MI attach routine.
1350 	 */
1351 	ether_ifattach(ifp, eaddr);
1352 
1353 	/* VLAN capability setup */
1354 	ifp->if_capabilities |= IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING;
1355 	if (ifp->if_capabilities & IFCAP_HWCSUM)
1356 		ifp->if_capabilities |= IFCAP_VLAN_HWCSUM;
1357 	/* Enable WOL if PM is supported. */
1358 	if (pci_find_extcap(sc->rl_dev, PCIY_PMG, &reg) == 0)
1359 		ifp->if_capabilities |= IFCAP_WOL;
1360 	ifp->if_capenable = ifp->if_capabilities;
1361 	/*
1362 	 * Don't enable TSO by default. Under certain
1363 	 * circumtances the controller generated corrupted
1364 	 * packets in TSO size.
1365 	 */
1366 	ifp->if_hwassist &= ~CSUM_TSO;
1367 	ifp->if_capenable &= ~IFCAP_TSO4;
1368 #ifdef DEVICE_POLLING
1369 	ifp->if_capabilities |= IFCAP_POLLING;
1370 #endif
1371 	/*
1372 	 * Tell the upper layer(s) we support long frames.
1373 	 * Must appear after the call to ether_ifattach() because
1374 	 * ether_ifattach() sets ifi_hdrlen to the default value.
1375 	 */
1376 	ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
1377 
1378 #ifdef RE_DIAG
1379 	/*
1380 	 * Perform hardware diagnostic on the original RTL8169.
1381 	 * Some 32-bit cards were incorrectly wired and would
1382 	 * malfunction if plugged into a 64-bit slot.
1383 	 */
1384 
1385 	if (hwrev == RL_HWREV_8169) {
1386 		error = re_diag(sc);
1387 		if (error) {
1388 			device_printf(dev,
1389 		    	"attach aborted due to hardware diag failure\n");
1390 			ether_ifdetach(ifp);
1391 			goto fail;
1392 		}
1393 	}
1394 #endif
1395 
1396 	/* Hook interrupt last to avoid having to lock softc */
1397 	if ((sc->rl_flags & RL_FLAG_MSI) == 0)
1398 		error = bus_setup_intr(dev, sc->rl_irq[0],
1399 		    INTR_TYPE_NET | INTR_MPSAFE, re_intr, NULL, sc,
1400 		    &sc->rl_intrhand[0]);
1401 	else {
1402 		for (i = 0; i < RL_MSI_MESSAGES; i++) {
1403 			error = bus_setup_intr(dev, sc->rl_irq[i],
1404 			    INTR_TYPE_NET | INTR_MPSAFE, re_intr, NULL, sc,
1405 		    	    &sc->rl_intrhand[i]);
1406 			if (error != 0)
1407 				break;
1408 		}
1409 	}
1410 	if (error) {
1411 		device_printf(dev, "couldn't set up irq\n");
1412 		ether_ifdetach(ifp);
1413 	}
1414 
1415 fail:
1416 
1417 	if (error)
1418 		re_detach(dev);
1419 
1420 	return (error);
1421 }
1422 
1423 /*
1424  * Shutdown hardware and free up resources. This can be called any
1425  * time after the mutex has been initialized. It is called in both
1426  * the error case in attach and the normal detach case so it needs
1427  * to be careful about only freeing resources that have actually been
1428  * allocated.
1429  */
1430 static int
1431 re_detach(device_t dev)
1432 {
1433 	struct rl_softc		*sc;
1434 	struct ifnet		*ifp;
1435 	int			i, rid;
1436 
1437 	sc = device_get_softc(dev);
1438 	ifp = sc->rl_ifp;
1439 	KASSERT(mtx_initialized(&sc->rl_mtx), ("re mutex not initialized"));
1440 
1441 	/* These should only be active if attach succeeded */
1442 	if (device_is_attached(dev)) {
1443 #ifdef DEVICE_POLLING
1444 		if (ifp->if_capenable & IFCAP_POLLING)
1445 			ether_poll_deregister(ifp);
1446 #endif
1447 		RL_LOCK(sc);
1448 #if 0
1449 		sc->suspended = 1;
1450 #endif
1451 		re_stop(sc);
1452 		RL_UNLOCK(sc);
1453 		callout_drain(&sc->rl_stat_callout);
1454 		taskqueue_drain(taskqueue_fast, &sc->rl_inttask);
1455 		taskqueue_drain(taskqueue_fast, &sc->rl_txtask);
1456 		/*
1457 		 * Force off the IFF_UP flag here, in case someone
1458 		 * still had a BPF descriptor attached to this
1459 		 * interface. If they do, ether_ifdetach() will cause
1460 		 * the BPF code to try and clear the promisc mode
1461 		 * flag, which will bubble down to re_ioctl(),
1462 		 * which will try to call re_init() again. This will
1463 		 * turn the NIC back on and restart the MII ticker,
1464 		 * which will panic the system when the kernel tries
1465 		 * to invoke the re_tick() function that isn't there
1466 		 * anymore.
1467 		 */
1468 		ifp->if_flags &= ~IFF_UP;
1469 		ether_ifdetach(ifp);
1470 	}
1471 	if (sc->rl_miibus)
1472 		device_delete_child(dev, sc->rl_miibus);
1473 	bus_generic_detach(dev);
1474 
1475 	/*
1476 	 * The rest is resource deallocation, so we should already be
1477 	 * stopped here.
1478 	 */
1479 
1480 	for (i = 0; i < RL_MSI_MESSAGES; i++) {
1481 		if (sc->rl_intrhand[i] != NULL) {
1482 			bus_teardown_intr(dev, sc->rl_irq[i],
1483 			    sc->rl_intrhand[i]);
1484 			sc->rl_intrhand[i] = NULL;
1485 		}
1486 	}
1487 	if (ifp != NULL)
1488 		if_free(ifp);
1489 	if ((sc->rl_flags & RL_FLAG_MSI) == 0) {
1490 		if (sc->rl_irq[0] != NULL) {
1491 			bus_release_resource(dev, SYS_RES_IRQ, 0,
1492 			    sc->rl_irq[0]);
1493 			sc->rl_irq[0] = NULL;
1494 		}
1495 	} else {
1496 		for (i = 0, rid = 1; i < RL_MSI_MESSAGES; i++, rid++) {
1497 			if (sc->rl_irq[i] != NULL) {
1498 				bus_release_resource(dev, SYS_RES_IRQ, rid,
1499 				    sc->rl_irq[i]);
1500 				sc->rl_irq[i] = NULL;
1501 			}
1502 		}
1503 		pci_release_msi(dev);
1504 	}
1505 	if (sc->rl_res)
1506 		bus_release_resource(dev, sc->rl_res_type, sc->rl_res_id,
1507 		    sc->rl_res);
1508 
1509 	/* Unload and free the RX DMA ring memory and map */
1510 
1511 	if (sc->rl_ldata.rl_rx_list_tag) {
1512 		bus_dmamap_unload(sc->rl_ldata.rl_rx_list_tag,
1513 		    sc->rl_ldata.rl_rx_list_map);
1514 		bus_dmamem_free(sc->rl_ldata.rl_rx_list_tag,
1515 		    sc->rl_ldata.rl_rx_list,
1516 		    sc->rl_ldata.rl_rx_list_map);
1517 		bus_dma_tag_destroy(sc->rl_ldata.rl_rx_list_tag);
1518 	}
1519 
1520 	/* Unload and free the TX DMA ring memory and map */
1521 
1522 	if (sc->rl_ldata.rl_tx_list_tag) {
1523 		bus_dmamap_unload(sc->rl_ldata.rl_tx_list_tag,
1524 		    sc->rl_ldata.rl_tx_list_map);
1525 		bus_dmamem_free(sc->rl_ldata.rl_tx_list_tag,
1526 		    sc->rl_ldata.rl_tx_list,
1527 		    sc->rl_ldata.rl_tx_list_map);
1528 		bus_dma_tag_destroy(sc->rl_ldata.rl_tx_list_tag);
1529 	}
1530 
1531 	/* Destroy all the RX and TX buffer maps */
1532 
1533 	if (sc->rl_ldata.rl_tx_mtag) {
1534 		for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++)
1535 			bus_dmamap_destroy(sc->rl_ldata.rl_tx_mtag,
1536 			    sc->rl_ldata.rl_tx_desc[i].tx_dmamap);
1537 		bus_dma_tag_destroy(sc->rl_ldata.rl_tx_mtag);
1538 	}
1539 	if (sc->rl_ldata.rl_rx_mtag) {
1540 		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++)
1541 			bus_dmamap_destroy(sc->rl_ldata.rl_rx_mtag,
1542 			    sc->rl_ldata.rl_rx_desc[i].rx_dmamap);
1543 		if (sc->rl_ldata.rl_rx_sparemap)
1544 			bus_dmamap_destroy(sc->rl_ldata.rl_rx_mtag,
1545 			    sc->rl_ldata.rl_rx_sparemap);
1546 		bus_dma_tag_destroy(sc->rl_ldata.rl_rx_mtag);
1547 	}
1548 
1549 	/* Unload and free the stats buffer and map */
1550 
1551 	if (sc->rl_ldata.rl_stag) {
1552 		bus_dmamap_unload(sc->rl_ldata.rl_stag,
1553 		    sc->rl_ldata.rl_rx_list_map);
1554 		bus_dmamem_free(sc->rl_ldata.rl_stag,
1555 		    sc->rl_ldata.rl_stats,
1556 		    sc->rl_ldata.rl_smap);
1557 		bus_dma_tag_destroy(sc->rl_ldata.rl_stag);
1558 	}
1559 
1560 	if (sc->rl_parent_tag)
1561 		bus_dma_tag_destroy(sc->rl_parent_tag);
1562 
1563 	mtx_destroy(&sc->rl_mtx);
1564 
1565 	return (0);
1566 }
1567 
1568 static __inline void
1569 re_discard_rxbuf(struct rl_softc *sc, int idx)
1570 {
1571 	struct rl_desc		*desc;
1572 	struct rl_rxdesc	*rxd;
1573 	uint32_t		cmdstat;
1574 
1575 	rxd = &sc->rl_ldata.rl_rx_desc[idx];
1576 	desc = &sc->rl_ldata.rl_rx_list[idx];
1577 	desc->rl_vlanctl = 0;
1578 	cmdstat = rxd->rx_size;
1579 	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
1580 		cmdstat |= RL_RDESC_CMD_EOR;
1581 	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
1582 }
1583 
1584 static int
1585 re_newbuf(struct rl_softc *sc, int idx)
1586 {
1587 	struct mbuf		*m;
1588 	struct rl_rxdesc	*rxd;
1589 	bus_dma_segment_t	segs[1];
1590 	bus_dmamap_t		map;
1591 	struct rl_desc		*desc;
1592 	uint32_t		cmdstat;
1593 	int			error, nsegs;
1594 
1595 	m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
1596 	if (m == NULL)
1597 		return (ENOBUFS);
1598 
1599 	m->m_len = m->m_pkthdr.len = MCLBYTES;
1600 #ifdef RE_FIXUP_RX
1601 	/*
1602 	 * This is part of an evil trick to deal with non-x86 platforms.
1603 	 * The RealTek chip requires RX buffers to be aligned on 64-bit
1604 	 * boundaries, but that will hose non-x86 machines. To get around
1605 	 * this, we leave some empty space at the start of each buffer
1606 	 * and for non-x86 hosts, we copy the buffer back six bytes
1607 	 * to achieve word alignment. This is slightly more efficient
1608 	 * than allocating a new buffer, copying the contents, and
1609 	 * discarding the old buffer.
1610 	 */
1611 	m_adj(m, RE_ETHER_ALIGN);
1612 #endif
1613 	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_rx_mtag,
1614 	    sc->rl_ldata.rl_rx_sparemap, m, segs, &nsegs, BUS_DMA_NOWAIT);
1615 	if (error != 0) {
1616 		m_freem(m);
1617 		return (ENOBUFS);
1618 	}
1619 	KASSERT(nsegs == 1, ("%s: %d segment returned!", __func__, nsegs));
1620 
1621 	rxd = &sc->rl_ldata.rl_rx_desc[idx];
1622 	if (rxd->rx_m != NULL) {
1623 		bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap,
1624 		    BUS_DMASYNC_POSTREAD);
1625 		bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap);
1626 	}
1627 
1628 	rxd->rx_m = m;
1629 	map = rxd->rx_dmamap;
1630 	rxd->rx_dmamap = sc->rl_ldata.rl_rx_sparemap;
1631 	rxd->rx_size = segs[0].ds_len;
1632 	sc->rl_ldata.rl_rx_sparemap = map;
1633 	bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap,
1634 	    BUS_DMASYNC_PREREAD);
1635 
1636 	desc = &sc->rl_ldata.rl_rx_list[idx];
1637 	desc->rl_vlanctl = 0;
1638 	desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[0].ds_addr));
1639 	desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[0].ds_addr));
1640 	cmdstat = segs[0].ds_len;
1641 	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
1642 		cmdstat |= RL_RDESC_CMD_EOR;
1643 	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
1644 
1645 	return (0);
1646 }
1647 
1648 #ifdef RE_FIXUP_RX
1649 static __inline void
1650 re_fixup_rx(struct mbuf *m)
1651 {
1652 	int                     i;
1653 	uint16_t                *src, *dst;
1654 
1655 	src = mtod(m, uint16_t *);
1656 	dst = src - (RE_ETHER_ALIGN - ETHER_ALIGN) / sizeof *src;
1657 
1658 	for (i = 0; i < (m->m_len / sizeof(uint16_t) + 1); i++)
1659 		*dst++ = *src++;
1660 
1661 	m->m_data -= RE_ETHER_ALIGN - ETHER_ALIGN;
1662 }
1663 #endif
1664 
1665 static int
1666 re_tx_list_init(struct rl_softc *sc)
1667 {
1668 	struct rl_desc		*desc;
1669 	int			i;
1670 
1671 	RL_LOCK_ASSERT(sc);
1672 
1673 	bzero(sc->rl_ldata.rl_tx_list,
1674 	    sc->rl_ldata.rl_tx_desc_cnt * sizeof(struct rl_desc));
1675 	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++)
1676 		sc->rl_ldata.rl_tx_desc[i].tx_m = NULL;
1677 	/* Set EOR. */
1678 	desc = &sc->rl_ldata.rl_tx_list[sc->rl_ldata.rl_tx_desc_cnt - 1];
1679 	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_EOR);
1680 
1681 	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
1682 	    sc->rl_ldata.rl_tx_list_map,
1683 	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1684 
1685 	sc->rl_ldata.rl_tx_prodidx = 0;
1686 	sc->rl_ldata.rl_tx_considx = 0;
1687 	sc->rl_ldata.rl_tx_free = sc->rl_ldata.rl_tx_desc_cnt;
1688 
1689 	return (0);
1690 }
1691 
1692 static int
1693 re_rx_list_init(struct rl_softc *sc)
1694 {
1695 	int			error, i;
1696 
1697 	bzero(sc->rl_ldata.rl_rx_list,
1698 	    sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc));
1699 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1700 		sc->rl_ldata.rl_rx_desc[i].rx_m = NULL;
1701 		if ((error = re_newbuf(sc, i)) != 0)
1702 			return (error);
1703 	}
1704 
1705 	/* Flush the RX descriptors */
1706 
1707 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1708 	    sc->rl_ldata.rl_rx_list_map,
1709 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
1710 
1711 	sc->rl_ldata.rl_rx_prodidx = 0;
1712 	sc->rl_head = sc->rl_tail = NULL;
1713 
1714 	return (0);
1715 }
1716 
1717 /*
1718  * RX handler for C+ and 8169. For the gigE chips, we support
1719  * the reception of jumbo frames that have been fragmented
1720  * across multiple 2K mbuf cluster buffers.
1721  */
1722 static int
1723 re_rxeof(struct rl_softc *sc)
1724 {
1725 	struct mbuf		*m;
1726 	struct ifnet		*ifp;
1727 	int			i, total_len;
1728 	struct rl_desc		*cur_rx;
1729 	u_int32_t		rxstat, rxvlan;
1730 	int			maxpkt = 16;
1731 
1732 	RL_LOCK_ASSERT(sc);
1733 
1734 	ifp = sc->rl_ifp;
1735 
1736 	/* Invalidate the descriptor memory */
1737 
1738 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1739 	    sc->rl_ldata.rl_rx_list_map,
1740 	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
1741 
1742 	for (i = sc->rl_ldata.rl_rx_prodidx; maxpkt > 0;
1743 	    i = RL_RX_DESC_NXT(sc, i)) {
1744 		cur_rx = &sc->rl_ldata.rl_rx_list[i];
1745 		rxstat = le32toh(cur_rx->rl_cmdstat);
1746 		if ((rxstat & RL_RDESC_STAT_OWN) != 0)
1747 			break;
1748 		total_len = rxstat & sc->rl_rxlenmask;
1749 		rxvlan = le32toh(cur_rx->rl_vlanctl);
1750 		m = sc->rl_ldata.rl_rx_desc[i].rx_m;
1751 
1752 		if (!(rxstat & RL_RDESC_STAT_EOF)) {
1753 			if (re_newbuf(sc, i) != 0) {
1754 				/*
1755 				 * If this is part of a multi-fragment packet,
1756 				 * discard all the pieces.
1757 				 */
1758 				if (sc->rl_head != NULL) {
1759 					m_freem(sc->rl_head);
1760 					sc->rl_head = sc->rl_tail = NULL;
1761 				}
1762 				re_discard_rxbuf(sc, i);
1763 				continue;
1764 			}
1765 			m->m_len = RE_RX_DESC_BUFLEN;
1766 			if (sc->rl_head == NULL)
1767 				sc->rl_head = sc->rl_tail = m;
1768 			else {
1769 				m->m_flags &= ~M_PKTHDR;
1770 				sc->rl_tail->m_next = m;
1771 				sc->rl_tail = m;
1772 			}
1773 			continue;
1774 		}
1775 
1776 		/*
1777 		 * NOTE: for the 8139C+, the frame length field
1778 		 * is always 12 bits in size, but for the gigE chips,
1779 		 * it is 13 bits (since the max RX frame length is 16K).
1780 		 * Unfortunately, all 32 bits in the status word
1781 		 * were already used, so to make room for the extra
1782 		 * length bit, RealTek took out the 'frame alignment
1783 		 * error' bit and shifted the other status bits
1784 		 * over one slot. The OWN, EOR, FS and LS bits are
1785 		 * still in the same places. We have already extracted
1786 		 * the frame length and checked the OWN bit, so rather
1787 		 * than using an alternate bit mapping, we shift the
1788 		 * status bits one space to the right so we can evaluate
1789 		 * them using the 8169 status as though it was in the
1790 		 * same format as that of the 8139C+.
1791 		 */
1792 		if (sc->rl_type == RL_8169)
1793 			rxstat >>= 1;
1794 
1795 		/*
1796 		 * if total_len > 2^13-1, both _RXERRSUM and _GIANT will be
1797 		 * set, but if CRC is clear, it will still be a valid frame.
1798 		 */
1799 		if (rxstat & RL_RDESC_STAT_RXERRSUM && !(total_len > 8191 &&
1800 		    (rxstat & RL_RDESC_STAT_ERRS) == RL_RDESC_STAT_GIANT)) {
1801 			ifp->if_ierrors++;
1802 			/*
1803 			 * If this is part of a multi-fragment packet,
1804 			 * discard all the pieces.
1805 			 */
1806 			if (sc->rl_head != NULL) {
1807 				m_freem(sc->rl_head);
1808 				sc->rl_head = sc->rl_tail = NULL;
1809 			}
1810 			re_discard_rxbuf(sc, i);
1811 			continue;
1812 		}
1813 
1814 		/*
1815 		 * If allocating a replacement mbuf fails,
1816 		 * reload the current one.
1817 		 */
1818 
1819 		if (re_newbuf(sc, i) != 0) {
1820 			ifp->if_iqdrops++;
1821 			if (sc->rl_head != NULL) {
1822 				m_freem(sc->rl_head);
1823 				sc->rl_head = sc->rl_tail = NULL;
1824 			}
1825 			re_discard_rxbuf(sc, i);
1826 			continue;
1827 		}
1828 
1829 		if (sc->rl_head != NULL) {
1830 			m->m_len = total_len % RE_RX_DESC_BUFLEN;
1831 			if (m->m_len == 0)
1832 				m->m_len = RE_RX_DESC_BUFLEN;
1833 			/*
1834 			 * Special case: if there's 4 bytes or less
1835 			 * in this buffer, the mbuf can be discarded:
1836 			 * the last 4 bytes is the CRC, which we don't
1837 			 * care about anyway.
1838 			 */
1839 			if (m->m_len <= ETHER_CRC_LEN) {
1840 				sc->rl_tail->m_len -=
1841 				    (ETHER_CRC_LEN - m->m_len);
1842 				m_freem(m);
1843 			} else {
1844 				m->m_len -= ETHER_CRC_LEN;
1845 				m->m_flags &= ~M_PKTHDR;
1846 				sc->rl_tail->m_next = m;
1847 			}
1848 			m = sc->rl_head;
1849 			sc->rl_head = sc->rl_tail = NULL;
1850 			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
1851 		} else
1852 			m->m_pkthdr.len = m->m_len =
1853 			    (total_len - ETHER_CRC_LEN);
1854 
1855 #ifdef RE_FIXUP_RX
1856 		re_fixup_rx(m);
1857 #endif
1858 		ifp->if_ipackets++;
1859 		m->m_pkthdr.rcvif = ifp;
1860 
1861 		/* Do RX checksumming if enabled */
1862 
1863 		if (ifp->if_capenable & IFCAP_RXCSUM) {
1864 			if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
1865 				/* Check IP header checksum */
1866 				if (rxstat & RL_RDESC_STAT_PROTOID)
1867 					m->m_pkthdr.csum_flags |=
1868 					    CSUM_IP_CHECKED;
1869 				if (!(rxstat & RL_RDESC_STAT_IPSUMBAD))
1870 					m->m_pkthdr.csum_flags |=
1871 					    CSUM_IP_VALID;
1872 
1873 				/* Check TCP/UDP checksum */
1874 				if ((RL_TCPPKT(rxstat) &&
1875 				    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
1876 				    (RL_UDPPKT(rxstat) &&
1877 				     !(rxstat & RL_RDESC_STAT_UDPSUMBAD))) {
1878 					m->m_pkthdr.csum_flags |=
1879 						CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
1880 					m->m_pkthdr.csum_data = 0xffff;
1881 				}
1882 			} else {
1883 				/*
1884 				 * RTL8168C/RTL816CP/RTL8111C/RTL8111CP
1885 				 */
1886 				if ((rxstat & RL_RDESC_STAT_PROTOID) &&
1887 				    (rxvlan & RL_RDESC_IPV4))
1888 					m->m_pkthdr.csum_flags |=
1889 					    CSUM_IP_CHECKED;
1890 				if (!(rxstat & RL_RDESC_STAT_IPSUMBAD) &&
1891 				    (rxvlan & RL_RDESC_IPV4))
1892 					m->m_pkthdr.csum_flags |=
1893 					    CSUM_IP_VALID;
1894 				if (((rxstat & RL_RDESC_STAT_TCP) &&
1895 				    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
1896 				    ((rxstat & RL_RDESC_STAT_UDP) &&
1897 				    !(rxstat & RL_RDESC_STAT_UDPSUMBAD))) {
1898 					m->m_pkthdr.csum_flags |=
1899 						CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
1900 					m->m_pkthdr.csum_data = 0xffff;
1901 				}
1902 			}
1903 		}
1904 		maxpkt--;
1905 		if (rxvlan & RL_RDESC_VLANCTL_TAG) {
1906 			m->m_pkthdr.ether_vtag =
1907 			    bswap16((rxvlan & RL_RDESC_VLANCTL_DATA));
1908 			m->m_flags |= M_VLANTAG;
1909 		}
1910 		RL_UNLOCK(sc);
1911 		(*ifp->if_input)(ifp, m);
1912 		RL_LOCK(sc);
1913 	}
1914 
1915 	/* Flush the RX DMA ring */
1916 
1917 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1918 	    sc->rl_ldata.rl_rx_list_map,
1919 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
1920 
1921 	sc->rl_ldata.rl_rx_prodidx = i;
1922 
1923 	if (maxpkt)
1924 		return(EAGAIN);
1925 
1926 	return(0);
1927 }
1928 
1929 static void
1930 re_txeof(struct rl_softc *sc)
1931 {
1932 	struct ifnet		*ifp;
1933 	struct rl_txdesc	*txd;
1934 	u_int32_t		txstat;
1935 	int			cons;
1936 
1937 	cons = sc->rl_ldata.rl_tx_considx;
1938 	if (cons == sc->rl_ldata.rl_tx_prodidx)
1939 		return;
1940 
1941 	ifp = sc->rl_ifp;
1942 	/* Invalidate the TX descriptor list */
1943 	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
1944 	    sc->rl_ldata.rl_tx_list_map,
1945 	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
1946 
1947 	for (; cons != sc->rl_ldata.rl_tx_prodidx;
1948 	    cons = RL_TX_DESC_NXT(sc, cons)) {
1949 		txstat = le32toh(sc->rl_ldata.rl_tx_list[cons].rl_cmdstat);
1950 		if (txstat & RL_TDESC_STAT_OWN)
1951 			break;
1952 		/*
1953 		 * We only stash mbufs in the last descriptor
1954 		 * in a fragment chain, which also happens to
1955 		 * be the only place where the TX status bits
1956 		 * are valid.
1957 		 */
1958 		if (txstat & RL_TDESC_CMD_EOF) {
1959 			txd = &sc->rl_ldata.rl_tx_desc[cons];
1960 			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
1961 			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
1962 			bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag,
1963 			    txd->tx_dmamap);
1964 			KASSERT(txd->tx_m != NULL,
1965 			    ("%s: freeing NULL mbufs!", __func__));
1966 			m_freem(txd->tx_m);
1967 			txd->tx_m = NULL;
1968 			if (txstat & (RL_TDESC_STAT_EXCESSCOL|
1969 			    RL_TDESC_STAT_COLCNT))
1970 				ifp->if_collisions++;
1971 			if (txstat & RL_TDESC_STAT_TXERRSUM)
1972 				ifp->if_oerrors++;
1973 			else
1974 				ifp->if_opackets++;
1975 		}
1976 		sc->rl_ldata.rl_tx_free++;
1977 		ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1978 	}
1979 	sc->rl_ldata.rl_tx_considx = cons;
1980 
1981 	/* No changes made to the TX ring, so no flush needed */
1982 
1983 	if (sc->rl_ldata.rl_tx_free != sc->rl_ldata.rl_tx_desc_cnt) {
1984 		/*
1985 		 * Some chips will ignore a second TX request issued
1986 		 * while an existing transmission is in progress. If
1987 		 * the transmitter goes idle but there are still
1988 		 * packets waiting to be sent, we need to restart the
1989 		 * channel here to flush them out. This only seems to
1990 		 * be required with the PCIe devices.
1991 		 */
1992 		CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
1993 
1994 #ifdef RE_TX_MODERATION
1995 		/*
1996 		 * If not all descriptors have been reaped yet, reload
1997 		 * the timer so that we will eventually get another
1998 		 * interrupt that will cause us to re-enter this routine.
1999 		 * This is done in case the transmitter has gone idle.
2000 		 */
2001 		CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2002 #endif
2003 	} else
2004 		sc->rl_watchdog_timer = 0;
2005 }
2006 
2007 static void
2008 re_tick(void *xsc)
2009 {
2010 	struct rl_softc		*sc;
2011 	struct mii_data		*mii;
2012 	struct ifnet		*ifp;
2013 
2014 	sc = xsc;
2015 	ifp = sc->rl_ifp;
2016 
2017 	RL_LOCK_ASSERT(sc);
2018 
2019 	re_watchdog(sc);
2020 
2021 	mii = device_get_softc(sc->rl_miibus);
2022 	mii_tick(mii);
2023 	if ((sc->rl_flags & RL_FLAG_LINK) != 0) {
2024 		if (!(mii->mii_media_status & IFM_ACTIVE))
2025 			sc->rl_flags &= ~RL_FLAG_LINK;
2026 	} else {
2027 		if (mii->mii_media_status & IFM_ACTIVE &&
2028 		    IFM_SUBTYPE(mii->mii_media_active) != IFM_NONE) {
2029 			sc->rl_flags |= RL_FLAG_LINK;
2030 			if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2031 				taskqueue_enqueue_fast(taskqueue_fast,
2032 				    &sc->rl_txtask);
2033 		}
2034 	}
2035 
2036 	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
2037 }
2038 
2039 #ifdef DEVICE_POLLING
2040 static void
2041 re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
2042 {
2043 	struct rl_softc *sc = ifp->if_softc;
2044 
2045 	RL_LOCK(sc);
2046 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
2047 		re_poll_locked(ifp, cmd, count);
2048 	RL_UNLOCK(sc);
2049 }
2050 
2051 static void
2052 re_poll_locked(struct ifnet *ifp, enum poll_cmd cmd, int count)
2053 {
2054 	struct rl_softc *sc = ifp->if_softc;
2055 
2056 	RL_LOCK_ASSERT(sc);
2057 
2058 	sc->rxcycles = count;
2059 	re_rxeof(sc);
2060 	re_txeof(sc);
2061 
2062 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2063 		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_txtask);
2064 
2065 	if (cmd == POLL_AND_CHECK_STATUS) { /* also check status register */
2066 		u_int16_t       status;
2067 
2068 		status = CSR_READ_2(sc, RL_ISR);
2069 		if (status == 0xffff)
2070 			return;
2071 		if (status)
2072 			CSR_WRITE_2(sc, RL_ISR, status);
2073 
2074 		/*
2075 		 * XXX check behaviour on receiver stalls.
2076 		 */
2077 
2078 		if (status & RL_ISR_SYSTEM_ERR) {
2079 			re_reset(sc);
2080 			re_init_locked(sc);
2081 		}
2082 	}
2083 }
2084 #endif /* DEVICE_POLLING */
2085 
2086 static int
2087 re_intr(void *arg)
2088 {
2089 	struct rl_softc		*sc;
2090 	uint16_t		status;
2091 
2092 	sc = arg;
2093 
2094 	status = CSR_READ_2(sc, RL_ISR);
2095 	if (status == 0xFFFF || (status & RL_INTRS_CPLUS) == 0)
2096                 return (FILTER_STRAY);
2097 	CSR_WRITE_2(sc, RL_IMR, 0);
2098 
2099 	taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_inttask);
2100 
2101 	return (FILTER_HANDLED);
2102 }
2103 
2104 static void
2105 re_int_task(void *arg, int npending)
2106 {
2107 	struct rl_softc		*sc;
2108 	struct ifnet		*ifp;
2109 	u_int16_t		status;
2110 	int			rval = 0;
2111 
2112 	sc = arg;
2113 	ifp = sc->rl_ifp;
2114 
2115 	RL_LOCK(sc);
2116 
2117 	status = CSR_READ_2(sc, RL_ISR);
2118         CSR_WRITE_2(sc, RL_ISR, status);
2119 
2120 	if (sc->suspended ||
2121 	    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2122 		RL_UNLOCK(sc);
2123 		return;
2124 	}
2125 
2126 #ifdef DEVICE_POLLING
2127 	if  (ifp->if_capenable & IFCAP_POLLING) {
2128 		RL_UNLOCK(sc);
2129 		return;
2130 	}
2131 #endif
2132 
2133 	if (status & (RL_ISR_RX_OK|RL_ISR_RX_ERR|RL_ISR_FIFO_OFLOW))
2134 		rval = re_rxeof(sc);
2135 
2136 	if (status & (
2137 #ifdef RE_TX_MODERATION
2138 	    RL_ISR_TIMEOUT_EXPIRED|
2139 #else
2140 	    RL_ISR_TX_OK|
2141 #endif
2142 	    RL_ISR_TX_ERR|RL_ISR_TX_DESC_UNAVAIL))
2143 		re_txeof(sc);
2144 
2145 	if (status & RL_ISR_SYSTEM_ERR) {
2146 		re_reset(sc);
2147 		re_init_locked(sc);
2148 	}
2149 
2150 	if (status & RL_ISR_LINKCHG) {
2151 		callout_stop(&sc->rl_stat_callout);
2152 		re_tick(sc);
2153 	}
2154 
2155 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2156 		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_txtask);
2157 
2158 	RL_UNLOCK(sc);
2159 
2160         if ((CSR_READ_2(sc, RL_ISR) & RL_INTRS_CPLUS) || rval) {
2161 		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_inttask);
2162 		return;
2163 	}
2164 
2165 	CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2166 }
2167 
2168 static int
2169 re_encap(struct rl_softc *sc, struct mbuf **m_head)
2170 {
2171 	struct rl_txdesc	*txd, *txd_last;
2172 	bus_dma_segment_t	segs[RL_NTXSEGS];
2173 	bus_dmamap_t		map;
2174 	struct mbuf		*m_new;
2175 	struct rl_desc		*desc;
2176 	int			nsegs, prod;
2177 	int			i, error, ei, si;
2178 	int			padlen;
2179 	uint32_t		cmdstat, csum_flags, vlanctl;
2180 
2181 	RL_LOCK_ASSERT(sc);
2182 	M_ASSERTPKTHDR((*m_head));
2183 
2184 	/*
2185 	 * With some of the RealTek chips, using the checksum offload
2186 	 * support in conjunction with the autopadding feature results
2187 	 * in the transmission of corrupt frames. For example, if we
2188 	 * need to send a really small IP fragment that's less than 60
2189 	 * bytes in size, and IP header checksumming is enabled, the
2190 	 * resulting ethernet frame that appears on the wire will
2191 	 * have garbled payload. To work around this, if TX IP checksum
2192 	 * offload is enabled, we always manually pad short frames out
2193 	 * to the minimum ethernet frame size.
2194 	 */
2195 	if ((sc->rl_flags & RL_FLAG_DESCV2) == 0 &&
2196 	    (*m_head)->m_pkthdr.len < RL_IP4CSUMTX_PADLEN &&
2197 	    ((*m_head)->m_pkthdr.csum_flags & CSUM_IP) != 0) {
2198 		padlen = RL_MIN_FRAMELEN - (*m_head)->m_pkthdr.len;
2199 		if (M_WRITABLE(*m_head) == 0) {
2200 			/* Get a writable copy. */
2201 			m_new = m_dup(*m_head, M_DONTWAIT);
2202 			m_freem(*m_head);
2203 			if (m_new == NULL) {
2204 				*m_head = NULL;
2205 				return (ENOBUFS);
2206 			}
2207 			*m_head = m_new;
2208 		}
2209 		if ((*m_head)->m_next != NULL ||
2210 		    M_TRAILINGSPACE(*m_head) < padlen) {
2211 			m_new = m_defrag(*m_head, M_DONTWAIT);
2212 			if (m_new == NULL) {
2213 				m_freem(*m_head);
2214 				*m_head = NULL;
2215 				return (ENOBUFS);
2216 			}
2217 		} else
2218 			m_new = *m_head;
2219 
2220 		/*
2221 		 * Manually pad short frames, and zero the pad space
2222 		 * to avoid leaking data.
2223 		 */
2224 		bzero(mtod(m_new, char *) + m_new->m_pkthdr.len, padlen);
2225 		m_new->m_pkthdr.len += padlen;
2226 		m_new->m_len = m_new->m_pkthdr.len;
2227 		*m_head = m_new;
2228 	}
2229 
2230 	prod = sc->rl_ldata.rl_tx_prodidx;
2231 	txd = &sc->rl_ldata.rl_tx_desc[prod];
2232 	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap,
2233 	    *m_head, segs, &nsegs, BUS_DMA_NOWAIT);
2234 	if (error == EFBIG) {
2235 		m_new = m_collapse(*m_head, M_DONTWAIT, RL_NTXSEGS);
2236 		if (m_new == NULL) {
2237 			m_freem(*m_head);
2238 			*m_head = NULL;
2239 			return (ENOBUFS);
2240 		}
2241 		*m_head = m_new;
2242 		error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_tx_mtag,
2243 		    txd->tx_dmamap, *m_head, segs, &nsegs, BUS_DMA_NOWAIT);
2244 		if (error != 0) {
2245 			m_freem(*m_head);
2246 			*m_head = NULL;
2247 			return (error);
2248 		}
2249 	} else if (error != 0)
2250 		return (error);
2251 	if (nsegs == 0) {
2252 		m_freem(*m_head);
2253 		*m_head = NULL;
2254 		return (EIO);
2255 	}
2256 
2257 	/* Check for number of available descriptors. */
2258 	if (sc->rl_ldata.rl_tx_free - nsegs <= 1) {
2259 		bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap);
2260 		return (ENOBUFS);
2261 	}
2262 
2263 	bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap,
2264 	    BUS_DMASYNC_PREWRITE);
2265 
2266 	/*
2267 	 * Set up checksum offload. Note: checksum offload bits must
2268 	 * appear in all descriptors of a multi-descriptor transmit
2269 	 * attempt. This is according to testing done with an 8169
2270 	 * chip. This is a requirement.
2271 	 */
2272 	vlanctl = 0;
2273 	csum_flags = 0;
2274 	if (((*m_head)->m_pkthdr.csum_flags & CSUM_TSO) != 0)
2275 		csum_flags = RL_TDESC_CMD_LGSEND |
2276 		    ((uint32_t)(*m_head)->m_pkthdr.tso_segsz <<
2277 		    RL_TDESC_CMD_MSSVAL_SHIFT);
2278 	else {
2279 		/*
2280 		 * Unconditionally enable IP checksum if TCP or UDP
2281 		 * checksum is required. Otherwise, TCP/UDP checksum
2282 		 * does't make effects.
2283 		 */
2284 		if (((*m_head)->m_pkthdr.csum_flags & RE_CSUM_FEATURES) != 0) {
2285 			if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
2286 				csum_flags |= RL_TDESC_CMD_IPCSUM;
2287 				if (((*m_head)->m_pkthdr.csum_flags &
2288 				    CSUM_TCP) != 0)
2289 					csum_flags |= RL_TDESC_CMD_TCPCSUM;
2290 				if (((*m_head)->m_pkthdr.csum_flags &
2291 				    CSUM_UDP) != 0)
2292 					csum_flags |= RL_TDESC_CMD_UDPCSUM;
2293 			} else {
2294 				vlanctl |= RL_TDESC_CMD_IPCSUMV2;
2295 				if (((*m_head)->m_pkthdr.csum_flags &
2296 				    CSUM_TCP) != 0)
2297 					vlanctl |= RL_TDESC_CMD_TCPCSUMV2;
2298 				if (((*m_head)->m_pkthdr.csum_flags &
2299 				    CSUM_UDP) != 0)
2300 					vlanctl |= RL_TDESC_CMD_UDPCSUMV2;
2301 			}
2302 		}
2303 	}
2304 
2305 	/*
2306 	 * Set up hardware VLAN tagging. Note: vlan tag info must
2307 	 * appear in all descriptors of a multi-descriptor
2308 	 * transmission attempt.
2309 	 */
2310 	if ((*m_head)->m_flags & M_VLANTAG)
2311 		vlanctl |= bswap16((*m_head)->m_pkthdr.ether_vtag) |
2312 		    RL_TDESC_VLANCTL_TAG;
2313 
2314 	si = prod;
2315 	for (i = 0; i < nsegs; i++, prod = RL_TX_DESC_NXT(sc, prod)) {
2316 		desc = &sc->rl_ldata.rl_tx_list[prod];
2317 		desc->rl_vlanctl = htole32(vlanctl);
2318 		desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[i].ds_addr));
2319 		desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[i].ds_addr));
2320 		cmdstat = segs[i].ds_len;
2321 		if (i != 0)
2322 			cmdstat |= RL_TDESC_CMD_OWN;
2323 		if (prod == sc->rl_ldata.rl_tx_desc_cnt - 1)
2324 			cmdstat |= RL_TDESC_CMD_EOR;
2325 		desc->rl_cmdstat = htole32(cmdstat | csum_flags);
2326 		sc->rl_ldata.rl_tx_free--;
2327 	}
2328 	/* Update producer index. */
2329 	sc->rl_ldata.rl_tx_prodidx = prod;
2330 
2331 	/* Set EOF on the last descriptor. */
2332 	ei = RL_TX_DESC_PRV(sc, prod);
2333 	desc = &sc->rl_ldata.rl_tx_list[ei];
2334 	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_EOF);
2335 
2336 	desc = &sc->rl_ldata.rl_tx_list[si];
2337 	/* Set SOF and transfer ownership of packet to the chip. */
2338 	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_OWN | RL_TDESC_CMD_SOF);
2339 
2340 	/*
2341 	 * Insure that the map for this transmission
2342 	 * is placed at the array index of the last descriptor
2343 	 * in this chain.  (Swap last and first dmamaps.)
2344 	 */
2345 	txd_last = &sc->rl_ldata.rl_tx_desc[ei];
2346 	map = txd->tx_dmamap;
2347 	txd->tx_dmamap = txd_last->tx_dmamap;
2348 	txd_last->tx_dmamap = map;
2349 	txd_last->tx_m = *m_head;
2350 
2351 	return (0);
2352 }
2353 
2354 static void
2355 re_tx_task(void *arg, int npending)
2356 {
2357 	struct ifnet		*ifp;
2358 
2359 	ifp = arg;
2360 	re_start(ifp);
2361 }
2362 
2363 /*
2364  * Main transmit routine for C+ and gigE NICs.
2365  */
2366 static void
2367 re_start(struct ifnet *ifp)
2368 {
2369 	struct rl_softc		*sc;
2370 	struct mbuf		*m_head;
2371 	int			queued;
2372 
2373 	sc = ifp->if_softc;
2374 
2375 	RL_LOCK(sc);
2376 
2377 	if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
2378 	    IFF_DRV_RUNNING || (sc->rl_flags & RL_FLAG_LINK) == 0) {
2379 		RL_UNLOCK(sc);
2380 		return;
2381 	}
2382 
2383 	for (queued = 0; !IFQ_DRV_IS_EMPTY(&ifp->if_snd) &&
2384 	    sc->rl_ldata.rl_tx_free > 1;) {
2385 		IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
2386 		if (m_head == NULL)
2387 			break;
2388 
2389 		if (re_encap(sc, &m_head) != 0) {
2390 			if (m_head == NULL)
2391 				break;
2392 			IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
2393 			ifp->if_drv_flags |= IFF_DRV_OACTIVE;
2394 			break;
2395 		}
2396 
2397 		/*
2398 		 * If there's a BPF listener, bounce a copy of this frame
2399 		 * to him.
2400 		 */
2401 		ETHER_BPF_MTAP(ifp, m_head);
2402 
2403 		queued++;
2404 	}
2405 
2406 	if (queued == 0) {
2407 #ifdef RE_TX_MODERATION
2408 		if (sc->rl_ldata.rl_tx_free != sc->rl_ldata.rl_tx_desc_cnt)
2409 			CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2410 #endif
2411 		RL_UNLOCK(sc);
2412 		return;
2413 	}
2414 
2415 	/* Flush the TX descriptors */
2416 
2417 	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2418 	    sc->rl_ldata.rl_tx_list_map,
2419 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2420 
2421 	CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2422 
2423 #ifdef RE_TX_MODERATION
2424 	/*
2425 	 * Use the countdown timer for interrupt moderation.
2426 	 * 'TX done' interrupts are disabled. Instead, we reset the
2427 	 * countdown timer, which will begin counting until it hits
2428 	 * the value in the TIMERINT register, and then trigger an
2429 	 * interrupt. Each time we write to the TIMERCNT register,
2430 	 * the timer count is reset to 0.
2431 	 */
2432 	CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2433 #endif
2434 
2435 	/*
2436 	 * Set a timeout in case the chip goes out to lunch.
2437 	 */
2438 	sc->rl_watchdog_timer = 5;
2439 
2440 	RL_UNLOCK(sc);
2441 }
2442 
2443 static void
2444 re_init(void *xsc)
2445 {
2446 	struct rl_softc		*sc = xsc;
2447 
2448 	RL_LOCK(sc);
2449 	re_init_locked(sc);
2450 	RL_UNLOCK(sc);
2451 }
2452 
2453 static void
2454 re_init_locked(struct rl_softc *sc)
2455 {
2456 	struct ifnet		*ifp = sc->rl_ifp;
2457 	struct mii_data		*mii;
2458 	u_int32_t		rxcfg = 0;
2459 	uint16_t		cfg;
2460 	union {
2461 		uint32_t align_dummy;
2462 		u_char eaddr[ETHER_ADDR_LEN];
2463         } eaddr;
2464 
2465 	RL_LOCK_ASSERT(sc);
2466 
2467 	mii = device_get_softc(sc->rl_miibus);
2468 
2469 	/*
2470 	 * Cancel pending I/O and free all RX/TX buffers.
2471 	 */
2472 	re_stop(sc);
2473 
2474 	/*
2475 	 * Enable C+ RX and TX mode, as well as VLAN stripping and
2476 	 * RX checksum offload. We must configure the C+ register
2477 	 * before all others.
2478 	 */
2479 	cfg = RL_CPLUSCMD_PCI_MRW;
2480 	if ((ifp->if_capenable & IFCAP_RXCSUM) != 0)
2481 		cfg |= RL_CPLUSCMD_RXCSUM_ENB;
2482 	if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0)
2483 		cfg |= RL_CPLUSCMD_VLANSTRIP;
2484 	if ((sc->rl_flags & RL_FLAG_MACSTAT) != 0) {
2485 		cfg |= RL_CPLUSCMD_MACSTAT_DIS;
2486 		/* XXX magic. */
2487 		cfg |= 0x0001;
2488 	} else
2489 		cfg |= RL_CPLUSCMD_RXENB | RL_CPLUSCMD_TXENB;
2490 	CSR_WRITE_2(sc, RL_CPLUS_CMD, cfg);
2491 	/*
2492 	 * Disable TSO if interface MTU size is greater than MSS
2493 	 * allowed in controller.
2494 	 */
2495 	if (ifp->if_mtu > RL_TSO_MTU && (ifp->if_capenable & IFCAP_TSO4) != 0) {
2496 		ifp->if_capenable &= ~IFCAP_TSO4;
2497 		ifp->if_hwassist &= ~CSUM_TSO;
2498 	}
2499 
2500 	/*
2501 	 * Init our MAC address.  Even though the chipset
2502 	 * documentation doesn't mention it, we need to enter "Config
2503 	 * register write enable" mode to modify the ID registers.
2504 	 */
2505 	/* Copy MAC address on stack to align. */
2506 	bcopy(IF_LLADDR(ifp), eaddr.eaddr, ETHER_ADDR_LEN);
2507 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG);
2508 	CSR_WRITE_4(sc, RL_IDR0,
2509 	    htole32(*(u_int32_t *)(&eaddr.eaddr[0])));
2510 	CSR_WRITE_4(sc, RL_IDR4,
2511 	    htole32(*(u_int32_t *)(&eaddr.eaddr[4])));
2512 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
2513 
2514 	/*
2515 	 * For C+ mode, initialize the RX descriptors and mbufs.
2516 	 */
2517 	re_rx_list_init(sc);
2518 	re_tx_list_init(sc);
2519 
2520 	/*
2521 	 * Load the addresses of the RX and TX lists into the chip.
2522 	 */
2523 
2524 	CSR_WRITE_4(sc, RL_RXLIST_ADDR_HI,
2525 	    RL_ADDR_HI(sc->rl_ldata.rl_rx_list_addr));
2526 	CSR_WRITE_4(sc, RL_RXLIST_ADDR_LO,
2527 	    RL_ADDR_LO(sc->rl_ldata.rl_rx_list_addr));
2528 
2529 	CSR_WRITE_4(sc, RL_TXLIST_ADDR_HI,
2530 	    RL_ADDR_HI(sc->rl_ldata.rl_tx_list_addr));
2531 	CSR_WRITE_4(sc, RL_TXLIST_ADDR_LO,
2532 	    RL_ADDR_LO(sc->rl_ldata.rl_tx_list_addr));
2533 
2534 	/*
2535 	 * Enable transmit and receive.
2536 	 */
2537 	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
2538 
2539 	/*
2540 	 * Set the initial TX and RX configuration.
2541 	 */
2542 	if (sc->rl_testmode) {
2543 		if (sc->rl_type == RL_8169)
2544 			CSR_WRITE_4(sc, RL_TXCFG,
2545 			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON);
2546 		else
2547 			CSR_WRITE_4(sc, RL_TXCFG,
2548 			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON_CPLUS);
2549 	} else
2550 		CSR_WRITE_4(sc, RL_TXCFG, RL_TXCFG_CONFIG);
2551 
2552 	CSR_WRITE_1(sc, RL_EARLY_TX_THRESH, 16);
2553 
2554 	CSR_WRITE_4(sc, RL_RXCFG, RL_RXCFG_CONFIG);
2555 
2556 	/* Set the individual bit to receive frames for this host only. */
2557 	rxcfg = CSR_READ_4(sc, RL_RXCFG);
2558 	rxcfg |= RL_RXCFG_RX_INDIV;
2559 
2560 	/* If we want promiscuous mode, set the allframes bit. */
2561 	if (ifp->if_flags & IFF_PROMISC)
2562 		rxcfg |= RL_RXCFG_RX_ALLPHYS;
2563 	else
2564 		rxcfg &= ~RL_RXCFG_RX_ALLPHYS;
2565 	CSR_WRITE_4(sc, RL_RXCFG, rxcfg);
2566 
2567 	/*
2568 	 * Set capture broadcast bit to capture broadcast frames.
2569 	 */
2570 	if (ifp->if_flags & IFF_BROADCAST)
2571 		rxcfg |= RL_RXCFG_RX_BROAD;
2572 	else
2573 		rxcfg &= ~RL_RXCFG_RX_BROAD;
2574 	CSR_WRITE_4(sc, RL_RXCFG, rxcfg);
2575 
2576 	/*
2577 	 * Program the multicast filter, if necessary.
2578 	 */
2579 	re_setmulti(sc);
2580 
2581 #ifdef DEVICE_POLLING
2582 	/*
2583 	 * Disable interrupts if we are polling.
2584 	 */
2585 	if (ifp->if_capenable & IFCAP_POLLING)
2586 		CSR_WRITE_2(sc, RL_IMR, 0);
2587 	else	/* otherwise ... */
2588 #endif
2589 
2590 	/*
2591 	 * Enable interrupts.
2592 	 */
2593 	if (sc->rl_testmode)
2594 		CSR_WRITE_2(sc, RL_IMR, 0);
2595 	else
2596 		CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2597 	CSR_WRITE_2(sc, RL_ISR, RL_INTRS_CPLUS);
2598 
2599 	/* Set initial TX threshold */
2600 	sc->rl_txthresh = RL_TX_THRESH_INIT;
2601 
2602 	/* Start RX/TX process. */
2603 	CSR_WRITE_4(sc, RL_MISSEDPKT, 0);
2604 #ifdef notdef
2605 	/* Enable receiver and transmitter. */
2606 	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
2607 #endif
2608 
2609 #ifdef RE_TX_MODERATION
2610 	/*
2611 	 * Initialize the timer interrupt register so that
2612 	 * a timer interrupt will be generated once the timer
2613 	 * reaches a certain number of ticks. The timer is
2614 	 * reloaded on each transmit. This gives us TX interrupt
2615 	 * moderation, which dramatically improves TX frame rate.
2616 	 */
2617 	if (sc->rl_type == RL_8169)
2618 		CSR_WRITE_4(sc, RL_TIMERINT_8169, 0x800);
2619 	else
2620 		CSR_WRITE_4(sc, RL_TIMERINT, 0x400);
2621 #endif
2622 
2623 	/*
2624 	 * For 8169 gigE NICs, set the max allowed RX packet
2625 	 * size so we can receive jumbo frames.
2626 	 */
2627 	if (sc->rl_type == RL_8169)
2628 		CSR_WRITE_2(sc, RL_MAXRXPKTLEN, 16383);
2629 
2630 	if (sc->rl_testmode)
2631 		return;
2632 
2633 	mii_mediachg(mii);
2634 
2635 	CSR_WRITE_1(sc, RL_CFG1, CSR_READ_1(sc, RL_CFG1) | RL_CFG1_DRVLOAD);
2636 
2637 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
2638 	ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
2639 
2640 	sc->rl_flags &= ~RL_FLAG_LINK;
2641 	sc->rl_watchdog_timer = 0;
2642 	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
2643 }
2644 
2645 /*
2646  * Set media options.
2647  */
2648 static int
2649 re_ifmedia_upd(struct ifnet *ifp)
2650 {
2651 	struct rl_softc		*sc;
2652 	struct mii_data		*mii;
2653 
2654 	sc = ifp->if_softc;
2655 	mii = device_get_softc(sc->rl_miibus);
2656 	RL_LOCK(sc);
2657 	mii_mediachg(mii);
2658 	RL_UNLOCK(sc);
2659 
2660 	return (0);
2661 }
2662 
2663 /*
2664  * Report current media status.
2665  */
2666 static void
2667 re_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
2668 {
2669 	struct rl_softc		*sc;
2670 	struct mii_data		*mii;
2671 
2672 	sc = ifp->if_softc;
2673 	mii = device_get_softc(sc->rl_miibus);
2674 
2675 	RL_LOCK(sc);
2676 	mii_pollstat(mii);
2677 	RL_UNLOCK(sc);
2678 	ifmr->ifm_active = mii->mii_media_active;
2679 	ifmr->ifm_status = mii->mii_media_status;
2680 }
2681 
2682 static int
2683 re_ioctl(struct ifnet *ifp, u_long command, caddr_t data)
2684 {
2685 	struct rl_softc		*sc = ifp->if_softc;
2686 	struct ifreq		*ifr = (struct ifreq *) data;
2687 	struct mii_data		*mii;
2688 	int			error = 0;
2689 
2690 	switch (command) {
2691 	case SIOCSIFMTU:
2692 		if (ifr->ifr_mtu < ETHERMIN || ifr->ifr_mtu > RL_JUMBO_MTU) {
2693 			error = EINVAL;
2694 			break;
2695 		}
2696 		if ((sc->rl_flags & RL_FLAG_NOJUMBO) != 0 &&
2697 		    ifr->ifr_mtu > RL_MAX_FRAMELEN) {
2698 			error = EINVAL;
2699 			break;
2700 		}
2701 		RL_LOCK(sc);
2702 		if (ifp->if_mtu != ifr->ifr_mtu)
2703 			ifp->if_mtu = ifr->ifr_mtu;
2704 		if (ifp->if_mtu > RL_TSO_MTU &&
2705 		    (ifp->if_capenable & IFCAP_TSO4) != 0) {
2706 			ifp->if_capenable &= ~IFCAP_TSO4;
2707 			ifp->if_hwassist &= ~CSUM_TSO;
2708 		}
2709 		RL_UNLOCK(sc);
2710 		break;
2711 	case SIOCSIFFLAGS:
2712 		RL_LOCK(sc);
2713 		if ((ifp->if_flags & IFF_UP) != 0) {
2714 			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
2715 				if (((ifp->if_flags ^ sc->rl_if_flags)
2716 				    & (IFF_PROMISC | IFF_ALLMULTI)) != 0)
2717 					re_setmulti(sc);
2718 			} else
2719 				re_init_locked(sc);
2720 		} else {
2721 			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
2722 				re_stop(sc);
2723 		}
2724 		sc->rl_if_flags = ifp->if_flags;
2725 		RL_UNLOCK(sc);
2726 		break;
2727 	case SIOCADDMULTI:
2728 	case SIOCDELMULTI:
2729 		RL_LOCK(sc);
2730 		re_setmulti(sc);
2731 		RL_UNLOCK(sc);
2732 		break;
2733 	case SIOCGIFMEDIA:
2734 	case SIOCSIFMEDIA:
2735 		mii = device_get_softc(sc->rl_miibus);
2736 		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, command);
2737 		break;
2738 	case SIOCSIFCAP:
2739 	    {
2740 		int mask, reinit;
2741 
2742 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
2743 		reinit = 0;
2744 #ifdef DEVICE_POLLING
2745 		if (mask & IFCAP_POLLING) {
2746 			if (ifr->ifr_reqcap & IFCAP_POLLING) {
2747 				error = ether_poll_register(re_poll, ifp);
2748 				if (error)
2749 					return(error);
2750 				RL_LOCK(sc);
2751 				/* Disable interrupts */
2752 				CSR_WRITE_2(sc, RL_IMR, 0x0000);
2753 				ifp->if_capenable |= IFCAP_POLLING;
2754 				RL_UNLOCK(sc);
2755 			} else {
2756 				error = ether_poll_deregister(ifp);
2757 				/* Enable interrupts. */
2758 				RL_LOCK(sc);
2759 				CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2760 				ifp->if_capenable &= ~IFCAP_POLLING;
2761 				RL_UNLOCK(sc);
2762 			}
2763 		}
2764 #endif /* DEVICE_POLLING */
2765 		if (mask & IFCAP_HWCSUM) {
2766 			ifp->if_capenable ^= IFCAP_HWCSUM;
2767 			if (ifp->if_capenable & IFCAP_TXCSUM)
2768 				ifp->if_hwassist |= RE_CSUM_FEATURES;
2769 			else
2770 				ifp->if_hwassist &= ~RE_CSUM_FEATURES;
2771 			reinit = 1;
2772 		}
2773 		if (mask & IFCAP_VLAN_HWTAGGING) {
2774 			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
2775 			reinit = 1;
2776 		}
2777 		if (mask & IFCAP_TSO4) {
2778 			ifp->if_capenable ^= IFCAP_TSO4;
2779 			if ((IFCAP_TSO4 & ifp->if_capenable) &&
2780 			    (IFCAP_TSO4 & ifp->if_capabilities))
2781 				ifp->if_hwassist |= CSUM_TSO;
2782 			else
2783 				ifp->if_hwassist &= ~CSUM_TSO;
2784 			if (ifp->if_mtu > RL_TSO_MTU &&
2785 			    (ifp->if_capenable & IFCAP_TSO4) != 0) {
2786 				ifp->if_capenable &= ~IFCAP_TSO4;
2787 				ifp->if_hwassist &= ~CSUM_TSO;
2788 			}
2789 		}
2790 		if ((mask & IFCAP_WOL) != 0 &&
2791 		    (ifp->if_capabilities & IFCAP_WOL) != 0) {
2792 			if ((mask & IFCAP_WOL_UCAST) != 0)
2793 				ifp->if_capenable ^= IFCAP_WOL_UCAST;
2794 			if ((mask & IFCAP_WOL_MCAST) != 0)
2795 				ifp->if_capenable ^= IFCAP_WOL_MCAST;
2796 			if ((mask & IFCAP_WOL_MAGIC) != 0)
2797 				ifp->if_capenable ^= IFCAP_WOL_MAGIC;
2798 		}
2799 		if (reinit && ifp->if_drv_flags & IFF_DRV_RUNNING)
2800 			re_init(sc);
2801 		VLAN_CAPABILITIES(ifp);
2802 	    }
2803 		break;
2804 	default:
2805 		error = ether_ioctl(ifp, command, data);
2806 		break;
2807 	}
2808 
2809 	return (error);
2810 }
2811 
2812 static void
2813 re_watchdog(struct rl_softc *sc)
2814 {
2815 
2816 	RL_LOCK_ASSERT(sc);
2817 
2818 	if (sc->rl_watchdog_timer == 0 || --sc->rl_watchdog_timer != 0)
2819 		return;
2820 
2821 	device_printf(sc->rl_dev, "watchdog timeout\n");
2822 	sc->rl_ifp->if_oerrors++;
2823 
2824 	re_txeof(sc);
2825 	re_rxeof(sc);
2826 	re_init_locked(sc);
2827 }
2828 
2829 /*
2830  * Stop the adapter and free any mbufs allocated to the
2831  * RX and TX lists.
2832  */
2833 static void
2834 re_stop(struct rl_softc *sc)
2835 {
2836 	int			i;
2837 	struct ifnet		*ifp;
2838 	struct rl_txdesc	*txd;
2839 	struct rl_rxdesc	*rxd;
2840 
2841 	RL_LOCK_ASSERT(sc);
2842 
2843 	ifp = sc->rl_ifp;
2844 
2845 	sc->rl_watchdog_timer = 0;
2846 	callout_stop(&sc->rl_stat_callout);
2847 	ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
2848 
2849 	CSR_WRITE_1(sc, RL_COMMAND, 0x00);
2850 	CSR_WRITE_2(sc, RL_IMR, 0x0000);
2851 	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
2852 
2853 	if (sc->rl_head != NULL) {
2854 		m_freem(sc->rl_head);
2855 		sc->rl_head = sc->rl_tail = NULL;
2856 	}
2857 
2858 	/* Free the TX list buffers. */
2859 
2860 	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
2861 		txd = &sc->rl_ldata.rl_tx_desc[i];
2862 		if (txd->tx_m != NULL) {
2863 			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
2864 			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
2865 			bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag,
2866 			    txd->tx_dmamap);
2867 			m_freem(txd->tx_m);
2868 			txd->tx_m = NULL;
2869 		}
2870 	}
2871 
2872 	/* Free the RX list buffers. */
2873 
2874 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
2875 		rxd = &sc->rl_ldata.rl_rx_desc[i];
2876 		if (rxd->rx_m != NULL) {
2877 			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
2878 			    rxd->rx_dmamap, BUS_DMASYNC_POSTREAD);
2879 			bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag,
2880 			    rxd->rx_dmamap);
2881 			m_freem(rxd->rx_m);
2882 			rxd->rx_m = NULL;
2883 		}
2884 	}
2885 }
2886 
2887 /*
2888  * Device suspend routine.  Stop the interface and save some PCI
2889  * settings in case the BIOS doesn't restore them properly on
2890  * resume.
2891  */
2892 static int
2893 re_suspend(device_t dev)
2894 {
2895 	struct rl_softc		*sc;
2896 
2897 	sc = device_get_softc(dev);
2898 
2899 	RL_LOCK(sc);
2900 	re_stop(sc);
2901 	re_setwol(sc);
2902 	sc->suspended = 1;
2903 	RL_UNLOCK(sc);
2904 
2905 	return (0);
2906 }
2907 
2908 /*
2909  * Device resume routine.  Restore some PCI settings in case the BIOS
2910  * doesn't, re-enable busmastering, and restart the interface if
2911  * appropriate.
2912  */
2913 static int
2914 re_resume(device_t dev)
2915 {
2916 	struct rl_softc		*sc;
2917 	struct ifnet		*ifp;
2918 
2919 	sc = device_get_softc(dev);
2920 
2921 	RL_LOCK(sc);
2922 
2923 	ifp = sc->rl_ifp;
2924 
2925 	/* reinitialize interface if necessary */
2926 	if (ifp->if_flags & IFF_UP)
2927 		re_init_locked(sc);
2928 
2929 	/*
2930 	 * Clear WOL matching such that normal Rx filtering
2931 	 * wouldn't interfere with WOL patterns.
2932 	 */
2933 	re_clrwol(sc);
2934 	sc->suspended = 0;
2935 	RL_UNLOCK(sc);
2936 
2937 	return (0);
2938 }
2939 
2940 /*
2941  * Stop all chip I/O so that the kernel's probe routines don't
2942  * get confused by errant DMAs when rebooting.
2943  */
2944 static int
2945 re_shutdown(device_t dev)
2946 {
2947 	struct rl_softc		*sc;
2948 
2949 	sc = device_get_softc(dev);
2950 
2951 	RL_LOCK(sc);
2952 	re_stop(sc);
2953 	/*
2954 	 * Mark interface as down since otherwise we will panic if
2955 	 * interrupt comes in later on, which can happen in some
2956 	 * cases.
2957 	 */
2958 	sc->rl_ifp->if_flags &= ~IFF_UP;
2959 	re_setwol(sc);
2960 	RL_UNLOCK(sc);
2961 
2962 	return (0);
2963 }
2964 
2965 static void
2966 re_setwol(struct rl_softc *sc)
2967 {
2968 	struct ifnet		*ifp;
2969 	int			pmc;
2970 	uint16_t		pmstat;
2971 	uint8_t			v;
2972 
2973 	RL_LOCK_ASSERT(sc);
2974 
2975 	if (pci_find_extcap(sc->rl_dev, PCIY_PMG, &pmc) != 0)
2976 		return;
2977 
2978 	ifp = sc->rl_ifp;
2979 	/* Enable config register write. */
2980 	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
2981 
2982 	/* Enable PME. */
2983 	v = CSR_READ_1(sc, RL_CFG1);
2984 	v &= ~RL_CFG1_PME;
2985 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
2986 		v |= RL_CFG1_PME;
2987 	CSR_WRITE_1(sc, RL_CFG1, v);
2988 
2989 	v = CSR_READ_1(sc, RL_CFG3);
2990 	v &= ~(RL_CFG3_WOL_LINK | RL_CFG3_WOL_MAGIC);
2991 	if ((ifp->if_capenable & IFCAP_WOL_MAGIC) != 0)
2992 		v |= RL_CFG3_WOL_MAGIC;
2993 	CSR_WRITE_1(sc, RL_CFG3, v);
2994 
2995 	/* Config register write done. */
2996 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
2997 
2998 	v = CSR_READ_1(sc, RL_CFG5);
2999 	v &= ~(RL_CFG5_WOL_BCAST | RL_CFG5_WOL_MCAST | RL_CFG5_WOL_UCAST);
3000 	v &= ~RL_CFG5_WOL_LANWAKE;
3001 	if ((ifp->if_capenable & IFCAP_WOL_UCAST) != 0)
3002 		v |= RL_CFG5_WOL_UCAST;
3003 	if ((ifp->if_capenable & IFCAP_WOL_MCAST) != 0)
3004 		v |= RL_CFG5_WOL_MCAST | RL_CFG5_WOL_BCAST;
3005 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3006 		v |= RL_CFG5_WOL_LANWAKE;
3007 	CSR_WRITE_1(sc, RL_CFG5, v);
3008 
3009 	/*
3010 	 * It seems that hardware resets its link speed to 100Mbps in
3011 	 * power down mode so switching to 100Mbps in driver is not
3012 	 * needed.
3013 	 */
3014 
3015 	/* Request PME if WOL is requested. */
3016 	pmstat = pci_read_config(sc->rl_dev, pmc + PCIR_POWER_STATUS, 2);
3017 	pmstat &= ~(PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE);
3018 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3019 		pmstat |= PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE;
3020 	pci_write_config(sc->rl_dev, pmc + PCIR_POWER_STATUS, pmstat, 2);
3021 }
3022 
3023 static void
3024 re_clrwol(struct rl_softc *sc)
3025 {
3026 	int			pmc;
3027 	uint8_t			v;
3028 
3029 	RL_LOCK_ASSERT(sc);
3030 
3031 	if (pci_find_extcap(sc->rl_dev, PCIY_PMG, &pmc) != 0)
3032 		return;
3033 
3034 	/* Enable config register write. */
3035 	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
3036 
3037 	v = CSR_READ_1(sc, RL_CFG3);
3038 	v &= ~(RL_CFG3_WOL_LINK | RL_CFG3_WOL_MAGIC);
3039 	CSR_WRITE_1(sc, RL_CFG3, v);
3040 
3041 	/* Config register write done. */
3042 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3043 
3044 	v = CSR_READ_1(sc, RL_CFG5);
3045 	v &= ~(RL_CFG5_WOL_BCAST | RL_CFG5_WOL_MCAST | RL_CFG5_WOL_UCAST);
3046 	v &= ~RL_CFG5_WOL_LANWAKE;
3047 	CSR_WRITE_1(sc, RL_CFG5, v);
3048 }
3049