xref: /openbsd-src/sys/dev/ic/re.c (revision 0b7734b3d77bb9b21afec6f4621cae6c805dbd45)
1 /*	$OpenBSD: re.c,v 1.192 2016/04/20 12:15:24 sthen Exp $	*/
2 /*	$FreeBSD: if_re.c,v 1.31 2004/09/04 07:54:05 ru Exp $	*/
3 /*
4  * Copyright (c) 1997, 1998-2003
5  *	Bill Paul <wpaul@windriver.com>.  All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. All advertising materials mentioning features or use of this software
16  *    must display the following acknowledgement:
17  *	This product includes software developed by Bill Paul.
18  * 4. Neither the name of the author nor the names of any co-contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY Bill Paul AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL Bill Paul OR THE VOICES IN HIS HEAD
26  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
27  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
28  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
29  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
30  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
31  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
32  * THE POSSIBILITY OF SUCH DAMAGE.
33  */
34 
35 /*
36  * Realtek 8139C+/8169/8169S/8110S PCI NIC driver
37  *
38  * Written by Bill Paul <wpaul@windriver.com>
39  * Senior Networking Software Engineer
40  * Wind River Systems
41  */
42 
43 /*
44  * This driver is designed to support Realtek's next generation of
45  * 10/100 and 10/100/1000 PCI ethernet controllers. There are currently
46  * seven devices in this family: the RTL8139C+, the RTL8169, the RTL8169S,
47  * RTL8110S, the RTL8168, the RTL8111 and the RTL8101E.
48  *
49  * The 8139C+ is a 10/100 ethernet chip. It is backwards compatible
50  * with the older 8139 family, however it also supports a special
51  * C+ mode of operation that provides several new performance enhancing
52  * features. These include:
53  *
54  *	o Descriptor based DMA mechanism. Each descriptor represents
55  *	  a single packet fragment. Data buffers may be aligned on
56  *	  any byte boundary.
57  *
58  *	o 64-bit DMA
59  *
60  *	o TCP/IP checksum offload for both RX and TX
61  *
62  *	o High and normal priority transmit DMA rings
63  *
64  *	o VLAN tag insertion and extraction
65  *
66  *	o TCP large send (segmentation offload)
67  *
68  * Like the 8139, the 8139C+ also has a built-in 10/100 PHY. The C+
69  * programming API is fairly straightforward. The RX filtering, EEPROM
70  * access and PHY access is the same as it is on the older 8139 series
71  * chips.
72  *
73  * The 8169 is a 64-bit 10/100/1000 gigabit ethernet MAC. It has almost the
74  * same programming API and feature set as the 8139C+ with the following
75  * differences and additions:
76  *
77  *	o 1000Mbps mode
78  *
79  *	o Jumbo frames
80  *
81  * 	o GMII and TBI ports/registers for interfacing with copper
82  *	  or fiber PHYs
83  *
84  *      o RX and TX DMA rings can have up to 1024 descriptors
85  *        (the 8139C+ allows a maximum of 64)
86  *
87  *	o Slight differences in register layout from the 8139C+
88  *
89  * The TX start and timer interrupt registers are at different locations
90  * on the 8169 than they are on the 8139C+. Also, the status word in the
91  * RX descriptor has a slightly different bit layout. The 8169 does not
92  * have a built-in PHY. Most reference boards use a Marvell 88E1000 'Alaska'
93  * copper gigE PHY.
94  *
95  * The 8169S/8110S 10/100/1000 devices have built-in copper gigE PHYs
96  * (the 'S' stands for 'single-chip'). These devices have the same
97  * programming API as the older 8169, but also have some vendor-specific
98  * registers for the on-board PHY. The 8110S is a LAN-on-motherboard
99  * part designed to be pin-compatible with the Realtek 8100 10/100 chip.
100  *
101  * This driver takes advantage of the RX and TX checksum offload and
102  * VLAN tag insertion/extraction features. It also implements TX
103  * interrupt moderation using the timer interrupt registers, which
104  * significantly reduces TX interrupt load. There is also support
105  * for jumbo frames, however the 8169/8169S/8110S can not transmit
106  * jumbo frames larger than 7440, so the max MTU possible with this
107  * driver is 7422 bytes.
108  */
109 
110 #include "bpfilter.h"
111 #include "vlan.h"
112 
113 #include <sys/param.h>
114 #include <sys/endian.h>
115 #include <sys/systm.h>
116 #include <sys/sockio.h>
117 #include <sys/mbuf.h>
118 #include <sys/malloc.h>
119 #include <sys/kernel.h>
120 #include <sys/device.h>
121 #include <sys/timeout.h>
122 #include <sys/socket.h>
123 #include <sys/atomic.h>
124 
125 #include <machine/bus.h>
126 
127 #include <net/if.h>
128 #include <net/if_media.h>
129 
130 #include <netinet/in.h>
131 #include <netinet/ip.h>
132 #include <netinet/if_ether.h>
133 
134 #if NBPFILTER > 0
135 #include <net/bpf.h>
136 #endif
137 
138 #include <dev/mii/mii.h>
139 #include <dev/mii/miivar.h>
140 
141 #include <dev/pci/pcidevs.h>
142 
143 #include <dev/ic/rtl81x9reg.h>
144 #include <dev/ic/revar.h>
145 
146 #ifdef RE_DEBUG
147 int redebug = 0;
148 #define DPRINTF(x)	do { if (redebug) printf x; } while (0)
149 #else
150 #define DPRINTF(x)
151 #endif
152 
153 static inline void re_set_bufaddr(struct rl_desc *, bus_addr_t);
154 
155 int	re_encap(struct rl_softc *, struct mbuf *, struct rl_txq *, int *);
156 
157 int	re_newbuf(struct rl_softc *);
158 int	re_rx_list_init(struct rl_softc *);
159 void	re_rx_list_fill(struct rl_softc *);
160 int	re_tx_list_init(struct rl_softc *);
161 int	re_rxeof(struct rl_softc *);
162 int	re_txeof(struct rl_softc *);
163 void	re_tick(void *);
164 void	re_start(struct ifnet *);
165 int	re_ioctl(struct ifnet *, u_long, caddr_t);
166 void	re_watchdog(struct ifnet *);
167 int	re_ifmedia_upd(struct ifnet *);
168 void	re_ifmedia_sts(struct ifnet *, struct ifmediareq *);
169 
170 void	re_set_jumbo(struct rl_softc *);
171 
172 void	re_eeprom_putbyte(struct rl_softc *, int);
173 void	re_eeprom_getword(struct rl_softc *, int, u_int16_t *);
174 void	re_read_eeprom(struct rl_softc *, caddr_t, int, int);
175 
176 int	re_gmii_readreg(struct device *, int, int);
177 void	re_gmii_writereg(struct device *, int, int, int);
178 
179 int	re_miibus_readreg(struct device *, int, int);
180 void	re_miibus_writereg(struct device *, int, int, int);
181 void	re_miibus_statchg(struct device *);
182 
183 void	re_iff(struct rl_softc *);
184 
185 void	re_setup_hw_im(struct rl_softc *);
186 void	re_setup_sim_im(struct rl_softc *);
187 void	re_disable_hw_im(struct rl_softc *);
188 void	re_disable_sim_im(struct rl_softc *);
189 void	re_config_imtype(struct rl_softc *, int);
190 void	re_setup_intr(struct rl_softc *, int, int);
191 #ifndef SMALL_KERNEL
192 int	re_wol(struct ifnet*, int);
193 #endif
194 
195 void	in_delayed_cksum(struct mbuf *);
196 
197 struct cfdriver re_cd = {
198 	0, "re", DV_IFNET
199 };
200 
201 extern char *hw_vendor, *hw_prod;
202 
203 #define EE_SET(x)					\
204 	CSR_WRITE_1(sc, RL_EECMD,			\
205 		CSR_READ_1(sc, RL_EECMD) | x)
206 
207 #define EE_CLR(x)					\
208 	CSR_WRITE_1(sc, RL_EECMD,			\
209 		CSR_READ_1(sc, RL_EECMD) & ~x)
210 
211 #define RL_FRAMELEN(mtu)				\
212 	(mtu + ETHER_HDR_LEN + ETHER_CRC_LEN +		\
213 		ETHER_VLAN_ENCAP_LEN)
214 
215 static const struct re_revision {
216 	u_int32_t		re_chipid;
217 	const char		*re_name;
218 } re_revisions[] = {
219 	{ RL_HWREV_8100,	"RTL8100" },
220 	{ RL_HWREV_8100E,	"RTL8100E" },
221 	{ RL_HWREV_8100E_SPIN2, "RTL8100E 2" },
222 	{ RL_HWREV_8101,	"RTL8101" },
223 	{ RL_HWREV_8101E,	"RTL8101E" },
224 	{ RL_HWREV_8102E,	"RTL8102E" },
225 	{ RL_HWREV_8106E,	"RTL8106E" },
226 	{ RL_HWREV_8401E,	"RTL8401E" },
227 	{ RL_HWREV_8402,	"RTL8402" },
228 	{ RL_HWREV_8411,	"RTL8411" },
229 	{ RL_HWREV_8411B,	"RTL8411B" },
230 	{ RL_HWREV_8102EL,	"RTL8102EL" },
231 	{ RL_HWREV_8102EL_SPIN1, "RTL8102EL 1" },
232 	{ RL_HWREV_8103E,       "RTL8103E" },
233 	{ RL_HWREV_8110S,	"RTL8110S" },
234 	{ RL_HWREV_8139CPLUS,	"RTL8139C+" },
235 	{ RL_HWREV_8168B_SPIN1,	"RTL8168 1" },
236 	{ RL_HWREV_8168B_SPIN2,	"RTL8168 2" },
237 	{ RL_HWREV_8168B_SPIN3,	"RTL8168 3" },
238 	{ RL_HWREV_8168C,	"RTL8168C/8111C" },
239 	{ RL_HWREV_8168C_SPIN2,	"RTL8168C/8111C" },
240 	{ RL_HWREV_8168CP,	"RTL8168CP/8111CP" },
241 	{ RL_HWREV_8168F,	"RTL8168F/8111F" },
242 	{ RL_HWREV_8168G,	"RTL8168G/8111G" },
243 	{ RL_HWREV_8168GU,	"RTL8168GU/8111GU" },
244 	{ RL_HWREV_8168H,	"RTL8168H/8111H" },
245 	{ RL_HWREV_8105E,	"RTL8105E" },
246 	{ RL_HWREV_8105E_SPIN1,	"RTL8105E" },
247 	{ RL_HWREV_8168D,	"RTL8168D/8111D" },
248 	{ RL_HWREV_8168DP,      "RTL8168DP/8111DP" },
249 	{ RL_HWREV_8168E,       "RTL8168E/8111E" },
250 	{ RL_HWREV_8168E_VL,	"RTL8168E/8111E-VL" },
251 	{ RL_HWREV_8168EP,	"RTL8168EP/8111EP" },
252 	{ RL_HWREV_8169,	"RTL8169" },
253 	{ RL_HWREV_8169_8110SB,	"RTL8169/8110SB" },
254 	{ RL_HWREV_8169_8110SBL, "RTL8169SBL" },
255 	{ RL_HWREV_8169_8110SCd, "RTL8169/8110SCd" },
256 	{ RL_HWREV_8169_8110SCe, "RTL8169/8110SCe" },
257 	{ RL_HWREV_8169S,	"RTL8169S" },
258 
259 	{ 0, NULL }
260 };
261 
262 
263 static inline void
264 re_set_bufaddr(struct rl_desc *d, bus_addr_t addr)
265 {
266 	d->rl_bufaddr_lo = htole32((uint32_t)addr);
267 	if (sizeof(bus_addr_t) == sizeof(uint64_t))
268 		d->rl_bufaddr_hi = htole32((uint64_t)addr >> 32);
269 	else
270 		d->rl_bufaddr_hi = 0;
271 }
272 
273 /*
274  * Send a read command and address to the EEPROM, check for ACK.
275  */
276 void
277 re_eeprom_putbyte(struct rl_softc *sc, int addr)
278 {
279 	int	d, i;
280 
281 	d = addr | (RL_9346_READ << sc->rl_eewidth);
282 
283 	/*
284 	 * Feed in each bit and strobe the clock.
285 	 */
286 
287 	for (i = 1 << (sc->rl_eewidth + 3); i; i >>= 1) {
288 		if (d & i)
289 			EE_SET(RL_EE_DATAIN);
290 		else
291 			EE_CLR(RL_EE_DATAIN);
292 		DELAY(100);
293 		EE_SET(RL_EE_CLK);
294 		DELAY(150);
295 		EE_CLR(RL_EE_CLK);
296 		DELAY(100);
297 	}
298 }
299 
300 /*
301  * Read a word of data stored in the EEPROM at address 'addr.'
302  */
303 void
304 re_eeprom_getword(struct rl_softc *sc, int addr, u_int16_t *dest)
305 {
306 	int		i;
307 	u_int16_t	word = 0;
308 
309 	/*
310 	 * Send address of word we want to read.
311 	 */
312 	re_eeprom_putbyte(sc, addr);
313 
314 	/*
315 	 * Start reading bits from EEPROM.
316 	 */
317 	for (i = 0x8000; i; i >>= 1) {
318 		EE_SET(RL_EE_CLK);
319 		DELAY(100);
320 		if (CSR_READ_1(sc, RL_EECMD) & RL_EE_DATAOUT)
321 			word |= i;
322 		EE_CLR(RL_EE_CLK);
323 		DELAY(100);
324 	}
325 
326 	*dest = word;
327 }
328 
329 /*
330  * Read a sequence of words from the EEPROM.
331  */
332 void
333 re_read_eeprom(struct rl_softc *sc, caddr_t dest, int off, int cnt)
334 {
335 	int		i;
336 	u_int16_t	word = 0, *ptr;
337 
338 	CSR_SETBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM);
339 
340 	DELAY(100);
341 
342 	for (i = 0; i < cnt; i++) {
343 		CSR_SETBIT_1(sc, RL_EECMD, RL_EE_SEL);
344 		re_eeprom_getword(sc, off + i, &word);
345 		CSR_CLRBIT_1(sc, RL_EECMD, RL_EE_SEL);
346 		ptr = (u_int16_t *)(dest + (i * 2));
347 		*ptr = word;
348 	}
349 
350 	CSR_CLRBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM);
351 }
352 
353 int
354 re_gmii_readreg(struct device *self, int phy, int reg)
355 {
356 	struct rl_softc	*sc = (struct rl_softc *)self;
357 	u_int32_t	rval;
358 	int		i;
359 
360 	if (phy != 7)
361 		return (0);
362 
363 	/* Let the rgephy driver read the GMEDIASTAT register */
364 
365 	if (reg == RL_GMEDIASTAT) {
366 		rval = CSR_READ_1(sc, RL_GMEDIASTAT);
367 		return (rval);
368 	}
369 
370 	CSR_WRITE_4(sc, RL_PHYAR, reg << 16);
371 
372 	for (i = 0; i < RL_PHY_TIMEOUT; i++) {
373 		rval = CSR_READ_4(sc, RL_PHYAR);
374 		if (rval & RL_PHYAR_BUSY)
375 			break;
376 		DELAY(25);
377 	}
378 
379 	if (i == RL_PHY_TIMEOUT) {
380 		printf ("%s: PHY read failed\n", sc->sc_dev.dv_xname);
381 		return (0);
382 	}
383 
384 	DELAY(20);
385 
386 	return (rval & RL_PHYAR_PHYDATA);
387 }
388 
389 void
390 re_gmii_writereg(struct device *dev, int phy, int reg, int data)
391 {
392 	struct rl_softc	*sc = (struct rl_softc *)dev;
393 	u_int32_t	rval;
394 	int		i;
395 
396 	CSR_WRITE_4(sc, RL_PHYAR, (reg << 16) |
397 	    (data & RL_PHYAR_PHYDATA) | RL_PHYAR_BUSY);
398 
399 	for (i = 0; i < RL_PHY_TIMEOUT; i++) {
400 		rval = CSR_READ_4(sc, RL_PHYAR);
401 		if (!(rval & RL_PHYAR_BUSY))
402 			break;
403 		DELAY(25);
404 	}
405 
406 	if (i == RL_PHY_TIMEOUT)
407 		printf ("%s: PHY write failed\n", sc->sc_dev.dv_xname);
408 
409 	DELAY(20);
410 }
411 
412 int
413 re_miibus_readreg(struct device *dev, int phy, int reg)
414 {
415 	struct rl_softc	*sc = (struct rl_softc *)dev;
416 	u_int16_t	rval = 0;
417 	u_int16_t	re8139_reg = 0;
418 	int		s;
419 
420 	s = splnet();
421 
422 	if (sc->sc_hwrev != RL_HWREV_8139CPLUS) {
423 		rval = re_gmii_readreg(dev, phy, reg);
424 		splx(s);
425 		return (rval);
426 	}
427 
428 	/* Pretend the internal PHY is only at address 0 */
429 	if (phy) {
430 		splx(s);
431 		return (0);
432 	}
433 	switch(reg) {
434 	case MII_BMCR:
435 		re8139_reg = RL_BMCR;
436 		break;
437 	case MII_BMSR:
438 		re8139_reg = RL_BMSR;
439 		break;
440 	case MII_ANAR:
441 		re8139_reg = RL_ANAR;
442 		break;
443 	case MII_ANER:
444 		re8139_reg = RL_ANER;
445 		break;
446 	case MII_ANLPAR:
447 		re8139_reg = RL_LPAR;
448 		break;
449 	case MII_PHYIDR1:
450 	case MII_PHYIDR2:
451 		splx(s);
452 		return (0);
453 	/*
454 	 * Allow the rlphy driver to read the media status
455 	 * register. If we have a link partner which does not
456 	 * support NWAY, this is the register which will tell
457 	 * us the results of parallel detection.
458 	 */
459 	case RL_MEDIASTAT:
460 		rval = CSR_READ_1(sc, RL_MEDIASTAT);
461 		splx(s);
462 		return (rval);
463 	default:
464 		printf("%s: bad phy register %x\n", sc->sc_dev.dv_xname, reg);
465 		splx(s);
466 		return (0);
467 	}
468 	rval = CSR_READ_2(sc, re8139_reg);
469 	if (re8139_reg == RL_BMCR) {
470 		/* 8139C+ has different bit layout. */
471 		rval &= ~(BMCR_LOOP | BMCR_ISO);
472 	}
473 	splx(s);
474 	return (rval);
475 }
476 
477 void
478 re_miibus_writereg(struct device *dev, int phy, int reg, int data)
479 {
480 	struct rl_softc	*sc = (struct rl_softc *)dev;
481 	u_int16_t	re8139_reg = 0;
482 	int		s;
483 
484 	s = splnet();
485 
486 	if (sc->sc_hwrev != RL_HWREV_8139CPLUS) {
487 		re_gmii_writereg(dev, phy, reg, data);
488 		splx(s);
489 		return;
490 	}
491 
492 	/* Pretend the internal PHY is only at address 0 */
493 	if (phy) {
494 		splx(s);
495 		return;
496 	}
497 	switch(reg) {
498 	case MII_BMCR:
499 		re8139_reg = RL_BMCR;
500 		/* 8139C+ has different bit layout. */
501 		data &= ~(BMCR_LOOP | BMCR_ISO);
502 		break;
503 	case MII_BMSR:
504 		re8139_reg = RL_BMSR;
505 		break;
506 	case MII_ANAR:
507 		re8139_reg = RL_ANAR;
508 		break;
509 	case MII_ANER:
510 		re8139_reg = RL_ANER;
511 		break;
512 	case MII_ANLPAR:
513 		re8139_reg = RL_LPAR;
514 		break;
515 	case MII_PHYIDR1:
516 	case MII_PHYIDR2:
517 		splx(s);
518 		return;
519 		break;
520 	default:
521 		printf("%s: bad phy register %x\n", sc->sc_dev.dv_xname, reg);
522 		splx(s);
523 		return;
524 	}
525 	CSR_WRITE_2(sc, re8139_reg, data);
526 	splx(s);
527 }
528 
529 void
530 re_miibus_statchg(struct device *dev)
531 {
532 	struct rl_softc		*sc = (struct rl_softc *)dev;
533 	struct ifnet		*ifp = &sc->sc_arpcom.ac_if;
534 	struct mii_data		*mii = &sc->sc_mii;
535 
536 	if ((ifp->if_flags & IFF_RUNNING) == 0)
537 		return;
538 
539 	sc->rl_flags &= ~RL_FLAG_LINK;
540 	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
541 	    (IFM_ACTIVE | IFM_AVALID)) {
542 		switch (IFM_SUBTYPE(mii->mii_media_active)) {
543 		case IFM_10_T:
544 		case IFM_100_TX:
545 			sc->rl_flags |= RL_FLAG_LINK;
546 			break;
547 		case IFM_1000_T:
548 			if ((sc->rl_flags & RL_FLAG_FASTETHER) != 0)
549 				break;
550 			sc->rl_flags |= RL_FLAG_LINK;
551 			break;
552 		default:
553 			break;
554 		}
555 	}
556 
557 	/*
558 	 * Realtek controllers do not provide an interface to
559 	 * Tx/Rx MACs for resolved speed, duplex and flow-control
560 	 * parameters.
561 	 */
562 }
563 
564 void
565 re_iff(struct rl_softc *sc)
566 {
567 	struct ifnet		*ifp = &sc->sc_arpcom.ac_if;
568 	int			h = 0;
569 	u_int32_t		hashes[2];
570 	u_int32_t		rxfilt;
571 	struct arpcom		*ac = &sc->sc_arpcom;
572 	struct ether_multi	*enm;
573 	struct ether_multistep	step;
574 
575 	rxfilt = CSR_READ_4(sc, RL_RXCFG);
576 	rxfilt &= ~(RL_RXCFG_RX_ALLPHYS | RL_RXCFG_RX_BROAD |
577 	    RL_RXCFG_RX_INDIV | RL_RXCFG_RX_MULTI);
578 	ifp->if_flags &= ~IFF_ALLMULTI;
579 
580 	/*
581 	 * Always accept frames destined to our station address.
582 	 * Always accept broadcast frames.
583 	 */
584 	rxfilt |= RL_RXCFG_RX_INDIV | RL_RXCFG_RX_BROAD;
585 
586 	if (ifp->if_flags & IFF_PROMISC || ac->ac_multirangecnt > 0) {
587 		ifp->if_flags |= IFF_ALLMULTI;
588 		rxfilt |= RL_RXCFG_RX_MULTI;
589 		if (ifp->if_flags & IFF_PROMISC)
590 			rxfilt |= RL_RXCFG_RX_ALLPHYS;
591 		hashes[0] = hashes[1] = 0xFFFFFFFF;
592 	} else {
593 		rxfilt |= RL_RXCFG_RX_MULTI;
594 		/* Program new filter. */
595 		bzero(hashes, sizeof(hashes));
596 
597 		ETHER_FIRST_MULTI(step, ac, enm);
598 		while (enm != NULL) {
599 			h = ether_crc32_be(enm->enm_addrlo,
600 			    ETHER_ADDR_LEN) >> 26;
601 
602 			if (h < 32)
603 				hashes[0] |= (1 << h);
604 			else
605 				hashes[1] |= (1 << (h - 32));
606 
607 			ETHER_NEXT_MULTI(step, enm);
608 		}
609 	}
610 
611 	/*
612 	 * For some unfathomable reason, Realtek decided to reverse
613 	 * the order of the multicast hash registers in the PCI Express
614 	 * parts. This means we have to write the hash pattern in reverse
615 	 * order for those devices.
616 	 */
617 	if (sc->rl_flags & RL_FLAG_PCIE) {
618 		CSR_WRITE_4(sc, RL_MAR0, swap32(hashes[1]));
619 		CSR_WRITE_4(sc, RL_MAR4, swap32(hashes[0]));
620 	} else {
621 		CSR_WRITE_4(sc, RL_MAR0, hashes[0]);
622 		CSR_WRITE_4(sc, RL_MAR4, hashes[1]);
623 	}
624 
625 	CSR_WRITE_4(sc, RL_RXCFG, rxfilt);
626 }
627 
628 void
629 re_reset(struct rl_softc *sc)
630 {
631 	int	i;
632 
633 	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RESET);
634 
635 	for (i = 0; i < RL_TIMEOUT; i++) {
636 		DELAY(10);
637 		if (!(CSR_READ_1(sc, RL_COMMAND) & RL_CMD_RESET))
638 			break;
639 	}
640 	if (i == RL_TIMEOUT)
641 		printf("%s: reset never completed!\n", sc->sc_dev.dv_xname);
642 
643 	if (sc->rl_flags & RL_FLAG_MACRESET)
644 		CSR_WRITE_1(sc, RL_LDPS, 1);
645 }
646 
647 #ifdef __armish__
648 /*
649  * Thecus N2100 doesn't store the full mac address in eeprom
650  * so we read the old mac address from the device before the reset
651  * in hopes that the proper mac address is already there.
652  */
653 union {
654 	u_int32_t eaddr_word[2];
655 	u_char eaddr[ETHER_ADDR_LEN];
656 } boot_eaddr;
657 int boot_eaddr_valid;
658 #endif /* __armish__ */
659 /*
660  * Attach the interface. Allocate softc structures, do ifmedia
661  * setup and ethernet/BPF attach.
662  */
663 int
664 re_attach(struct rl_softc *sc, const char *intrstr)
665 {
666 	u_char		eaddr[ETHER_ADDR_LEN];
667 	u_int16_t	as[ETHER_ADDR_LEN / 2];
668 	struct ifnet	*ifp;
669 	u_int16_t	re_did = 0;
670 	int		error = 0, i;
671 	const struct re_revision *rr;
672 	const char	*re_name = NULL;
673 	int		ntxsegs;
674 
675 	sc->sc_hwrev = CSR_READ_4(sc, RL_TXCFG) & RL_TXCFG_HWREV;
676 
677 	switch (sc->sc_hwrev) {
678 	case RL_HWREV_8139CPLUS:
679 		sc->rl_flags |= RL_FLAG_FASTETHER | RL_FLAG_AUTOPAD;
680 		sc->rl_max_mtu = RL_MTU;
681 		break;
682 	case RL_HWREV_8100E:
683 	case RL_HWREV_8100E_SPIN2:
684 	case RL_HWREV_8101E:
685 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_FASTETHER;
686 		sc->rl_max_mtu = RL_MTU;
687 		break;
688 	case RL_HWREV_8103E:
689 		sc->rl_flags |= RL_FLAG_MACSLEEP;
690 		/* FALLTHROUGH */
691 	case RL_HWREV_8102E:
692 	case RL_HWREV_8102EL:
693 	case RL_HWREV_8102EL_SPIN1:
694 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
695 		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_FASTETHER |
696 		    RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD;
697 		sc->rl_max_mtu = RL_MTU;
698 		break;
699 	case RL_HWREV_8401E:
700 	case RL_HWREV_8105E:
701 	case RL_HWREV_8105E_SPIN1:
702 	case RL_HWREV_8106E:
703 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
704 		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
705 		    RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD;
706 		sc->rl_max_mtu = RL_MTU;
707 		break;
708 	case RL_HWREV_8402:
709 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
710 		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
711 		    RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD |
712 		    RL_FLAG_CMDSTOP_WAIT_TXQ;
713 		sc->rl_max_mtu = RL_MTU;
714 		break;
715 	case RL_HWREV_8168B_SPIN1:
716 	case RL_HWREV_8168B_SPIN2:
717 		sc->rl_flags |= RL_FLAG_WOLRXENB;
718 		/* FALLTHROUGH */
719 	case RL_HWREV_8168B_SPIN3:
720 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_MACSTAT;
721 		sc->rl_max_mtu = RL_MTU;
722 		break;
723 	case RL_HWREV_8168C_SPIN2:
724 		sc->rl_flags |= RL_FLAG_MACSLEEP;
725 		/* FALLTHROUGH */
726 	case RL_HWREV_8168C:
727 	case RL_HWREV_8168CP:
728 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
729 		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP |
730 		    RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 | RL_FLAG_WOL_MANLINK;
731 		sc->rl_max_mtu = RL_JUMBO_MTU_6K;
732 		break;
733 	case RL_HWREV_8168D:
734 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
735 		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
736 		    RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 |
737 		    RL_FLAG_WOL_MANLINK;
738 		sc->rl_max_mtu = RL_JUMBO_MTU_9K;
739 		break;
740 	case RL_HWREV_8168DP:
741 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
742 		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_AUTOPAD |
743 		    RL_FLAG_JUMBOV2 | RL_FLAG_WAIT_TXPOLL | RL_FLAG_WOL_MANLINK;
744 		sc->rl_max_mtu = RL_JUMBO_MTU_9K;
745 		break;
746 	case RL_HWREV_8168E:
747 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
748 		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
749 		    RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 |
750 		    RL_FLAG_WOL_MANLINK;
751 		sc->rl_max_mtu = RL_JUMBO_MTU_9K;
752 		break;
753 	case RL_HWREV_8168E_VL:
754 		sc->rl_flags |= RL_FLAG_EARLYOFF | RL_FLAG_PHYWAKE | RL_FLAG_PAR |
755 		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP |
756 		    RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 | RL_FLAG_CMDSTOP_WAIT_TXQ |
757 		    RL_FLAG_WOL_MANLINK;
758 		sc->rl_max_mtu = RL_JUMBO_MTU_6K;
759 		break;
760 	case RL_HWREV_8168F:
761 		sc->rl_flags |= RL_FLAG_EARLYOFF;
762 		/* FALLTHROUGH */
763 	case RL_HWREV_8411:
764 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
765 		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP |
766 		    RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 | RL_FLAG_CMDSTOP_WAIT_TXQ |
767 		    RL_FLAG_WOL_MANLINK;
768 		sc->rl_max_mtu = RL_JUMBO_MTU_9K;
769 		break;
770 	case RL_HWREV_8168EP:
771 	case RL_HWREV_8168G:
772 	case RL_HWREV_8168GU:
773 	case RL_HWREV_8168H:
774 	case RL_HWREV_8411B:
775 		if (sc->sc_product == PCI_PRODUCT_REALTEK_RT8101E) {
776 			/* RTL8106EUS */
777 			sc->rl_flags |= RL_FLAG_FASTETHER;
778 			sc->rl_max_mtu = RL_MTU;
779 		} else {
780 			sc->rl_flags |= RL_FLAG_JUMBOV2 | RL_FLAG_WOL_MANLINK;
781 			sc->rl_max_mtu = RL_JUMBO_MTU_9K;
782 		}
783 
784 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
785 		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP |
786 		    RL_FLAG_AUTOPAD | RL_FLAG_CMDSTOP_WAIT_TXQ |
787 		    RL_FLAG_EARLYOFFV2 | RL_FLAG_RXDV_GATED;
788 		break;
789 	case RL_HWREV_8169_8110SB:
790 	case RL_HWREV_8169_8110SBL:
791 	case RL_HWREV_8169_8110SCd:
792 	case RL_HWREV_8169_8110SCe:
793 		sc->rl_flags |= RL_FLAG_PHYWAKE;
794 		/* FALLTHROUGH */
795 	case RL_HWREV_8169:
796 	case RL_HWREV_8169S:
797 	case RL_HWREV_8110S:
798 		sc->rl_flags |= RL_FLAG_MACRESET;
799 		sc->rl_max_mtu = RL_JUMBO_MTU_7K;
800 		break;
801 	default:
802 		break;
803 	}
804 
805 	if (sc->sc_hwrev == RL_HWREV_8139CPLUS) {
806 		sc->rl_cfg0 = RL_8139_CFG0;
807 		sc->rl_cfg1 = RL_8139_CFG1;
808 		sc->rl_cfg2 = 0;
809 		sc->rl_cfg3 = RL_8139_CFG3;
810 		sc->rl_cfg4 = RL_8139_CFG4;
811 		sc->rl_cfg5 = RL_8139_CFG5;
812 	} else {
813 		sc->rl_cfg0 = RL_CFG0;
814 		sc->rl_cfg1 = RL_CFG1;
815 		sc->rl_cfg2 = RL_CFG2;
816 		sc->rl_cfg3 = RL_CFG3;
817 		sc->rl_cfg4 = RL_CFG4;
818 		sc->rl_cfg5 = RL_CFG5;
819 	}
820 
821 	/* Reset the adapter. */
822 	re_reset(sc);
823 
824 	sc->rl_tx_time = 5;		/* 125us */
825 	sc->rl_rx_time = 2;		/* 50us */
826 	if (sc->rl_flags & RL_FLAG_PCIE)
827 		sc->rl_sim_time = 75;	/* 75us */
828 	else
829 		sc->rl_sim_time = 125;	/* 125us */
830 	sc->rl_imtype = RL_IMTYPE_SIM;	/* simulated interrupt moderation */
831 
832 	if (sc->sc_hwrev == RL_HWREV_8139CPLUS)
833 		sc->rl_bus_speed = 33; /* XXX */
834 	else if (sc->rl_flags & RL_FLAG_PCIE)
835 		sc->rl_bus_speed = 125;
836 	else {
837 		u_int8_t cfg2;
838 
839 		cfg2 = CSR_READ_1(sc, sc->rl_cfg2);
840 		switch (cfg2 & RL_CFG2_PCI_MASK) {
841 		case RL_CFG2_PCI_33MHZ:
842  			sc->rl_bus_speed = 33;
843 			break;
844 		case RL_CFG2_PCI_66MHZ:
845 			sc->rl_bus_speed = 66;
846 			break;
847 		default:
848 			printf("%s: unknown bus speed, assume 33MHz\n",
849 			    sc->sc_dev.dv_xname);
850 			sc->rl_bus_speed = 33;
851 			break;
852 		}
853 
854 		if (cfg2 & RL_CFG2_PCI_64BIT)
855 			sc->rl_flags |= RL_FLAG_PCI64;
856 	}
857 
858 	re_config_imtype(sc, sc->rl_imtype);
859 
860 	if (sc->rl_flags & RL_FLAG_PAR) {
861 		/*
862 		 * XXX Should have a better way to extract station
863 		 * address from EEPROM.
864 		 */
865 		for (i = 0; i < ETHER_ADDR_LEN; i++)
866 			eaddr[i] = CSR_READ_1(sc, RL_IDR0 + i);
867 	} else {
868 		sc->rl_eewidth = RL_9356_ADDR_LEN;
869 		re_read_eeprom(sc, (caddr_t)&re_did, 0, 1);
870 		if (re_did != 0x8129)
871 			sc->rl_eewidth = RL_9346_ADDR_LEN;
872 
873 		/*
874 		 * Get station address from the EEPROM.
875 		 */
876 		re_read_eeprom(sc, (caddr_t)as, RL_EE_EADDR, 3);
877 		for (i = 0; i < ETHER_ADDR_LEN / 2; i++)
878 			as[i] = letoh16(as[i]);
879 		bcopy(as, eaddr, ETHER_ADDR_LEN);
880 
881 #ifdef __armish__
882 		/*
883 		 * On the Thecus N2100, the MAC address in the EEPROM is
884 		 * always 00:14:fd:10:00:00.  The proper MAC address is
885 		 * stored in flash.  Fortunately RedBoot configures the
886 		 * proper MAC address (for the first onboard interface)
887 		 * which we can read from the IDR.
888 		 */
889 		if (eaddr[0] == 0x00 && eaddr[1] == 0x14 &&
890 		    eaddr[2] == 0xfd && eaddr[3] == 0x10 &&
891 		    eaddr[4] == 0x00 && eaddr[5] == 0x00) {
892 			if (boot_eaddr_valid == 0) {
893 				boot_eaddr.eaddr_word[1] =
894 				    letoh32(CSR_READ_4(sc, RL_IDR4));
895 				boot_eaddr.eaddr_word[0] =
896 				    letoh32(CSR_READ_4(sc, RL_IDR0));
897 				boot_eaddr_valid = 1;
898 			}
899 
900 			bcopy(boot_eaddr.eaddr, eaddr, sizeof(eaddr));
901 			eaddr[5] += sc->sc_dev.dv_unit;
902 		}
903 #endif
904 	}
905 
906 	/*
907 	 * Set RX length mask, TX poll request register
908 	 * and descriptor count.
909 	 */
910 	if (sc->sc_hwrev == RL_HWREV_8139CPLUS) {
911 		sc->rl_rxlenmask = RL_RDESC_STAT_FRAGLEN;
912 		sc->rl_txstart = RL_TXSTART;
913 		sc->rl_ldata.rl_tx_desc_cnt = RL_8139_TX_DESC_CNT;
914 		sc->rl_ldata.rl_rx_desc_cnt = RL_8139_RX_DESC_CNT;
915 		ntxsegs = RL_8139_NTXSEGS;
916 	} else {
917 		sc->rl_rxlenmask = RL_RDESC_STAT_GFRAGLEN;
918 		sc->rl_txstart = RL_GTXSTART;
919 		sc->rl_ldata.rl_tx_desc_cnt = RL_8169_TX_DESC_CNT;
920 		sc->rl_ldata.rl_rx_desc_cnt = RL_8169_RX_DESC_CNT;
921 		ntxsegs = RL_8169_NTXSEGS;
922 	}
923 
924 	bcopy(eaddr, (char *)&sc->sc_arpcom.ac_enaddr, ETHER_ADDR_LEN);
925 
926 	for (rr = re_revisions; rr->re_name != NULL; rr++) {
927 		if (rr->re_chipid == sc->sc_hwrev)
928 			re_name = rr->re_name;
929 	}
930 
931 	if (re_name == NULL)
932 		printf(": unknown ASIC (0x%04x)", sc->sc_hwrev >> 16);
933 	else
934 		printf(": %s (0x%04x)", re_name, sc->sc_hwrev >> 16);
935 
936 	printf(", %s, address %s\n", intrstr,
937 	    ether_sprintf(sc->sc_arpcom.ac_enaddr));
938 
939 	if (sc->rl_ldata.rl_tx_desc_cnt >
940 	    PAGE_SIZE / sizeof(struct rl_desc)) {
941 		sc->rl_ldata.rl_tx_desc_cnt =
942 		    PAGE_SIZE / sizeof(struct rl_desc);
943 	}
944 
945 	/* Allocate DMA'able memory for the TX ring */
946 	if ((error = bus_dmamem_alloc(sc->sc_dmat, RL_TX_LIST_SZ(sc),
947 		    RL_RING_ALIGN, 0, &sc->rl_ldata.rl_tx_listseg, 1,
948 		    &sc->rl_ldata.rl_tx_listnseg, BUS_DMA_NOWAIT |
949 		    BUS_DMA_ZERO)) != 0) {
950 		printf("%s: can't allocate tx listseg, error = %d\n",
951 		    sc->sc_dev.dv_xname, error);
952 		goto fail_0;
953 	}
954 
955 	/* Load the map for the TX ring. */
956 	if ((error = bus_dmamem_map(sc->sc_dmat, &sc->rl_ldata.rl_tx_listseg,
957 		    sc->rl_ldata.rl_tx_listnseg, RL_TX_LIST_SZ(sc),
958 		    (caddr_t *)&sc->rl_ldata.rl_tx_list,
959 		    BUS_DMA_COHERENT | BUS_DMA_NOWAIT)) != 0) {
960 		printf("%s: can't map tx list, error = %d\n",
961 		    sc->sc_dev.dv_xname, error);
962 		goto fail_1;
963 	}
964 
965 	if ((error = bus_dmamap_create(sc->sc_dmat, RL_TX_LIST_SZ(sc), 1,
966 		    RL_TX_LIST_SZ(sc), 0, 0,
967 		    &sc->rl_ldata.rl_tx_list_map)) != 0) {
968 		printf("%s: can't create tx list map, error = %d\n",
969 		    sc->sc_dev.dv_xname, error);
970 		goto fail_2;
971 	}
972 
973 	if ((error = bus_dmamap_load(sc->sc_dmat,
974 		    sc->rl_ldata.rl_tx_list_map, sc->rl_ldata.rl_tx_list,
975 		    RL_TX_LIST_SZ(sc), NULL, BUS_DMA_NOWAIT)) != 0) {
976 		printf("%s: can't load tx list, error = %d\n",
977 		    sc->sc_dev.dv_xname, error);
978 		goto fail_3;
979 	}
980 
981 	/* Create DMA maps for TX buffers */
982 	for (i = 0; i < RL_TX_QLEN; i++) {
983 		error = bus_dmamap_create(sc->sc_dmat,
984 		    RL_JUMBO_FRAMELEN, ntxsegs, RL_JUMBO_FRAMELEN,
985 		    0, 0, &sc->rl_ldata.rl_txq[i].txq_dmamap);
986 		if (error) {
987 			printf("%s: can't create DMA map for TX\n",
988 			    sc->sc_dev.dv_xname);
989 			goto fail_4;
990 		}
991 	}
992 
993         /* Allocate DMA'able memory for the RX ring */
994 	if ((error = bus_dmamem_alloc(sc->sc_dmat, RL_RX_DMAMEM_SZ(sc),
995 		    RL_RING_ALIGN, 0, &sc->rl_ldata.rl_rx_listseg, 1,
996 		    &sc->rl_ldata.rl_rx_listnseg, BUS_DMA_NOWAIT |
997 		    BUS_DMA_ZERO)) != 0) {
998 		printf("%s: can't allocate rx listnseg, error = %d\n",
999 		    sc->sc_dev.dv_xname, error);
1000 		goto fail_4;
1001 	}
1002 
1003         /* Load the map for the RX ring. */
1004 	if ((error = bus_dmamem_map(sc->sc_dmat, &sc->rl_ldata.rl_rx_listseg,
1005 		    sc->rl_ldata.rl_rx_listnseg, RL_RX_DMAMEM_SZ(sc),
1006 		    (caddr_t *)&sc->rl_ldata.rl_rx_list,
1007 		    BUS_DMA_COHERENT | BUS_DMA_NOWAIT)) != 0) {
1008 		printf("%s: can't map rx list, error = %d\n",
1009 		    sc->sc_dev.dv_xname, error);
1010 		goto fail_5;
1011 
1012 	}
1013 
1014 	if ((error = bus_dmamap_create(sc->sc_dmat, RL_RX_DMAMEM_SZ(sc), 1,
1015 		    RL_RX_DMAMEM_SZ(sc), 0, 0,
1016 		    &sc->rl_ldata.rl_rx_list_map)) != 0) {
1017 		printf("%s: can't create rx list map, error = %d\n",
1018 		    sc->sc_dev.dv_xname, error);
1019 		goto fail_6;
1020 	}
1021 
1022 	if ((error = bus_dmamap_load(sc->sc_dmat,
1023 		    sc->rl_ldata.rl_rx_list_map, sc->rl_ldata.rl_rx_list,
1024 		    RL_RX_DMAMEM_SZ(sc), NULL, BUS_DMA_NOWAIT)) != 0) {
1025 		printf("%s: can't load rx list, error = %d\n",
1026 		    sc->sc_dev.dv_xname, error);
1027 		goto fail_7;
1028 	}
1029 
1030 	/* Create DMA maps for RX buffers */
1031 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1032 		error = bus_dmamap_create(sc->sc_dmat,
1033 		    RL_FRAMELEN(sc->rl_max_mtu), 1,
1034 		    RL_FRAMELEN(sc->rl_max_mtu), 0, 0,
1035 		    &sc->rl_ldata.rl_rxsoft[i].rxs_dmamap);
1036 		if (error) {
1037 			printf("%s: can't create DMA map for RX\n",
1038 			    sc->sc_dev.dv_xname);
1039 			goto fail_8;
1040 		}
1041 	}
1042 
1043 	ifp = &sc->sc_arpcom.ac_if;
1044 	ifp->if_softc = sc;
1045 	strlcpy(ifp->if_xname, sc->sc_dev.dv_xname, IFNAMSIZ);
1046 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1047 	ifp->if_xflags = IFXF_MPSAFE;
1048 	ifp->if_ioctl = re_ioctl;
1049 	ifp->if_start = re_start;
1050 	ifp->if_watchdog = re_watchdog;
1051 	ifp->if_hardmtu = sc->rl_max_mtu;
1052 	IFQ_SET_MAXLEN(&ifp->if_snd, RL_TX_QLEN);
1053 
1054 	ifp->if_capabilities = IFCAP_VLAN_MTU | IFCAP_CSUM_TCPv4 |
1055 	    IFCAP_CSUM_UDPv4;
1056 
1057 	/*
1058 	 * RTL8168/8111C generates wrong IP checksummed frame if the
1059 	 * packet has IP options so disable TX IP checksum offloading.
1060 	 */
1061 	switch (sc->sc_hwrev) {
1062 	case RL_HWREV_8168C:
1063 	case RL_HWREV_8168C_SPIN2:
1064 	case RL_HWREV_8168CP:
1065 		break;
1066 	default:
1067 		ifp->if_capabilities |= IFCAP_CSUM_IPv4;
1068 	}
1069 
1070 #if NVLAN > 0
1071 	ifp->if_capabilities |= IFCAP_VLAN_HWTAGGING;
1072 #endif
1073 
1074 #ifndef SMALL_KERNEL
1075 	ifp->if_capabilities |= IFCAP_WOL;
1076 	ifp->if_wol = re_wol;
1077 	re_wol(ifp, 0);
1078 #endif
1079 	timeout_set(&sc->timer_handle, re_tick, sc);
1080 
1081 	/* Take PHY out of power down mode. */
1082 	if (sc->rl_flags & RL_FLAG_PHYWAKE_PM) {
1083 		CSR_WRITE_1(sc, RL_PMCH, CSR_READ_1(sc, RL_PMCH) | 0x80);
1084 		if (sc->sc_hwrev == RL_HWREV_8401E)
1085 			CSR_WRITE_1(sc, 0xD1, CSR_READ_1(sc, 0xD1) & ~0x08);
1086 	}
1087 	if (sc->rl_flags & RL_FLAG_PHYWAKE) {
1088 		re_gmii_writereg((struct device *)sc, 1, 0x1f, 0);
1089 		re_gmii_writereg((struct device *)sc, 1, 0x0e, 0);
1090 	}
1091 
1092 	/* Do MII setup */
1093 	sc->sc_mii.mii_ifp = ifp;
1094 	sc->sc_mii.mii_readreg = re_miibus_readreg;
1095 	sc->sc_mii.mii_writereg = re_miibus_writereg;
1096 	sc->sc_mii.mii_statchg = re_miibus_statchg;
1097 	ifmedia_init(&sc->sc_mii.mii_media, IFM_IMASK, re_ifmedia_upd,
1098 	    re_ifmedia_sts);
1099 	mii_attach(&sc->sc_dev, &sc->sc_mii, 0xffffffff, MII_PHY_ANY,
1100 	    MII_OFFSET_ANY, MIIF_DOPAUSE);
1101 	if (LIST_FIRST(&sc->sc_mii.mii_phys) == NULL) {
1102 		printf("%s: no PHY found!\n", sc->sc_dev.dv_xname);
1103 		ifmedia_add(&sc->sc_mii.mii_media,
1104 		    IFM_ETHER|IFM_NONE, 0, NULL);
1105 		ifmedia_set(&sc->sc_mii.mii_media,
1106 		    IFM_ETHER|IFM_NONE);
1107 	} else
1108 		ifmedia_set(&sc->sc_mii.mii_media, IFM_ETHER|IFM_AUTO);
1109 
1110 	/*
1111 	 * Call MI attach routine.
1112 	 */
1113 	if_attach(ifp);
1114 	ether_ifattach(ifp);
1115 
1116 	return (0);
1117 
1118 fail_8:
1119 	/* Destroy DMA maps for RX buffers. */
1120 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1121 		if (sc->rl_ldata.rl_rxsoft[i].rxs_dmamap != NULL)
1122 			bus_dmamap_destroy(sc->sc_dmat,
1123 			    sc->rl_ldata.rl_rxsoft[i].rxs_dmamap);
1124 	}
1125 
1126 	/* Free DMA'able memory for the RX ring. */
1127 	bus_dmamap_unload(sc->sc_dmat, sc->rl_ldata.rl_rx_list_map);
1128 fail_7:
1129 	bus_dmamap_destroy(sc->sc_dmat, sc->rl_ldata.rl_rx_list_map);
1130 fail_6:
1131 	bus_dmamem_unmap(sc->sc_dmat,
1132 	    (caddr_t)sc->rl_ldata.rl_rx_list, RL_RX_DMAMEM_SZ(sc));
1133 fail_5:
1134 	bus_dmamem_free(sc->sc_dmat,
1135 	    &sc->rl_ldata.rl_rx_listseg, sc->rl_ldata.rl_rx_listnseg);
1136 
1137 fail_4:
1138 	/* Destroy DMA maps for TX buffers. */
1139 	for (i = 0; i < RL_TX_QLEN; i++) {
1140 		if (sc->rl_ldata.rl_txq[i].txq_dmamap != NULL)
1141 			bus_dmamap_destroy(sc->sc_dmat,
1142 			    sc->rl_ldata.rl_txq[i].txq_dmamap);
1143 	}
1144 
1145 	/* Free DMA'able memory for the TX ring. */
1146 	bus_dmamap_unload(sc->sc_dmat, sc->rl_ldata.rl_tx_list_map);
1147 fail_3:
1148 	bus_dmamap_destroy(sc->sc_dmat, sc->rl_ldata.rl_tx_list_map);
1149 fail_2:
1150 	bus_dmamem_unmap(sc->sc_dmat,
1151 	    (caddr_t)sc->rl_ldata.rl_tx_list, RL_TX_LIST_SZ(sc));
1152 fail_1:
1153 	bus_dmamem_free(sc->sc_dmat,
1154 	    &sc->rl_ldata.rl_tx_listseg, sc->rl_ldata.rl_tx_listnseg);
1155 fail_0:
1156  	return (1);
1157 }
1158 
1159 
1160 int
1161 re_newbuf(struct rl_softc *sc)
1162 {
1163 	struct mbuf	*m;
1164 	bus_dmamap_t	map;
1165 	struct rl_desc	*d;
1166 	struct rl_rxsoft *rxs;
1167 	u_int32_t	cmdstat;
1168 	int		error, idx;
1169 
1170 	m = MCLGETI(NULL, M_DONTWAIT, NULL, RL_FRAMELEN(sc->rl_max_mtu));
1171 	if (!m)
1172 		return (ENOBUFS);
1173 
1174 	/*
1175 	 * Initialize mbuf length fields and fixup
1176 	 * alignment so that the frame payload is
1177 	 * longword aligned on strict alignment archs.
1178 	 */
1179 	m->m_len = m->m_pkthdr.len = RL_FRAMELEN(sc->rl_max_mtu);
1180 	m->m_data += RE_ETHER_ALIGN;
1181 
1182 	idx = sc->rl_ldata.rl_rx_prodidx;
1183 	rxs = &sc->rl_ldata.rl_rxsoft[idx];
1184 	map = rxs->rxs_dmamap;
1185 	error = bus_dmamap_load_mbuf(sc->sc_dmat, map, m,
1186 	    BUS_DMA_READ|BUS_DMA_NOWAIT);
1187 	if (error) {
1188 		m_freem(m);
1189 		return (ENOBUFS);
1190 	}
1191 
1192 	bus_dmamap_sync(sc->sc_dmat, map, 0, map->dm_mapsize,
1193 	    BUS_DMASYNC_PREREAD);
1194 
1195 	d = &sc->rl_ldata.rl_rx_list[idx];
1196 	RL_RXDESCSYNC(sc, idx, BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1197 	cmdstat = letoh32(d->rl_cmdstat);
1198 	RL_RXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD);
1199 	if (cmdstat & RL_RDESC_STAT_OWN) {
1200 		printf("%s: tried to map busy RX descriptor\n",
1201 		    sc->sc_dev.dv_xname);
1202 		m_freem(m);
1203 		return (ENOBUFS);
1204 	}
1205 
1206 	rxs->rxs_mbuf = m;
1207 
1208 	d->rl_vlanctl = 0;
1209 	cmdstat = map->dm_segs[0].ds_len;
1210 	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
1211 		cmdstat |= RL_RDESC_CMD_EOR;
1212 	re_set_bufaddr(d, map->dm_segs[0].ds_addr);
1213 	d->rl_cmdstat = htole32(cmdstat);
1214 	RL_RXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1215 	cmdstat |= RL_RDESC_CMD_OWN;
1216 	d->rl_cmdstat = htole32(cmdstat);
1217 	RL_RXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1218 
1219 	sc->rl_ldata.rl_rx_prodidx = RL_NEXT_RX_DESC(sc, idx);
1220 
1221 	return (0);
1222 }
1223 
1224 
1225 int
1226 re_tx_list_init(struct rl_softc *sc)
1227 {
1228 	int i;
1229 
1230 	memset(sc->rl_ldata.rl_tx_list, 0, RL_TX_LIST_SZ(sc));
1231 	for (i = 0; i < RL_TX_QLEN; i++) {
1232 		sc->rl_ldata.rl_txq[i].txq_mbuf = NULL;
1233 	}
1234 
1235 	bus_dmamap_sync(sc->sc_dmat,
1236 	    sc->rl_ldata.rl_tx_list_map, 0,
1237 	    sc->rl_ldata.rl_tx_list_map->dm_mapsize,
1238 	    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1239 	sc->rl_ldata.rl_txq_prodidx = 0;
1240 	sc->rl_ldata.rl_txq_considx = 0;
1241 	sc->rl_ldata.rl_tx_free = sc->rl_ldata.rl_tx_desc_cnt;
1242 	sc->rl_ldata.rl_tx_nextfree = 0;
1243 
1244 	return (0);
1245 }
1246 
1247 int
1248 re_rx_list_init(struct rl_softc *sc)
1249 {
1250 	bzero(sc->rl_ldata.rl_rx_list, RL_RX_LIST_SZ(sc));
1251 
1252 	sc->rl_ldata.rl_rx_prodidx = 0;
1253 	sc->rl_ldata.rl_rx_considx = 0;
1254 	sc->rl_head = sc->rl_tail = NULL;
1255 
1256 	if_rxr_init(&sc->rl_ldata.rl_rx_ring, 2, sc->rl_ldata.rl_rx_desc_cnt);
1257 	re_rx_list_fill(sc);
1258 
1259 	return (0);
1260 }
1261 
1262 void
1263 re_rx_list_fill(struct rl_softc *sc)
1264 {
1265 	u_int slots;
1266 
1267 	for (slots = if_rxr_get(&sc->rl_ldata.rl_rx_ring,
1268 	    sc->rl_ldata.rl_rx_desc_cnt);
1269 	    slots > 0; slots--) {
1270 		if (re_newbuf(sc) == ENOBUFS)
1271 			break;
1272 	}
1273 	if_rxr_put(&sc->rl_ldata.rl_rx_ring, slots);
1274 }
1275 
1276 /*
1277  * RX handler for C+ and 8169. For the gigE chips, we support
1278  * the reception of jumbo frames that have been fragmented
1279  * across multiple 2K mbuf cluster buffers.
1280  */
1281 int
1282 re_rxeof(struct rl_softc *sc)
1283 {
1284 	struct mbuf_list ml = MBUF_LIST_INITIALIZER();
1285 	struct mbuf	*m;
1286 	struct ifnet	*ifp;
1287 	int		i, total_len, rx = 0;
1288 	struct rl_desc	*cur_rx;
1289 	struct rl_rxsoft *rxs;
1290 	u_int32_t	rxstat, rxvlan;
1291 
1292 	ifp = &sc->sc_arpcom.ac_if;
1293 
1294 	for (i = sc->rl_ldata.rl_rx_considx;
1295 	    if_rxr_inuse(&sc->rl_ldata.rl_rx_ring) > 0;
1296 	     i = RL_NEXT_RX_DESC(sc, i)) {
1297 		cur_rx = &sc->rl_ldata.rl_rx_list[i];
1298 		RL_RXDESCSYNC(sc, i,
1299 		    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1300 		rxstat = letoh32(cur_rx->rl_cmdstat);
1301 		rxvlan = letoh32(cur_rx->rl_vlanctl);
1302 		RL_RXDESCSYNC(sc, i, BUS_DMASYNC_PREREAD);
1303 		if ((rxstat & RL_RDESC_STAT_OWN) != 0)
1304 			break;
1305 		total_len = rxstat & sc->rl_rxlenmask;
1306 		rxs = &sc->rl_ldata.rl_rxsoft[i];
1307 		m = rxs->rxs_mbuf;
1308 		rxs->rxs_mbuf = NULL;
1309 		if_rxr_put(&sc->rl_ldata.rl_rx_ring, 1);
1310 		rx = 1;
1311 
1312 		/* Invalidate the RX mbuf and unload its map */
1313 
1314 		bus_dmamap_sync(sc->sc_dmat,
1315 		    rxs->rxs_dmamap, 0, rxs->rxs_dmamap->dm_mapsize,
1316 		    BUS_DMASYNC_POSTREAD);
1317 		bus_dmamap_unload(sc->sc_dmat, rxs->rxs_dmamap);
1318 
1319 		if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0 &&
1320 		    (rxstat & (RL_RDESC_STAT_SOF | RL_RDESC_STAT_EOF)) !=
1321 		    (RL_RDESC_STAT_SOF | RL_RDESC_STAT_EOF)) {
1322 			continue;
1323 		} else if (!(rxstat & RL_RDESC_STAT_EOF)) {
1324 			m->m_len = RL_FRAMELEN(sc->rl_max_mtu);
1325 			if (sc->rl_head == NULL)
1326 				sc->rl_head = sc->rl_tail = m;
1327 			else {
1328 				m->m_flags &= ~M_PKTHDR;
1329 				sc->rl_tail->m_next = m;
1330 				sc->rl_tail = m;
1331 			}
1332 			continue;
1333 		}
1334 
1335 		/*
1336 		 * NOTE: for the 8139C+, the frame length field
1337 		 * is always 12 bits in size, but for the gigE chips,
1338 		 * it is 13 bits (since the max RX frame length is 16K).
1339 		 * Unfortunately, all 32 bits in the status word
1340 		 * were already used, so to make room for the extra
1341 		 * length bit, Realtek took out the 'frame alignment
1342 		 * error' bit and shifted the other status bits
1343 		 * over one slot. The OWN, EOR, FS and LS bits are
1344 		 * still in the same places. We have already extracted
1345 		 * the frame length and checked the OWN bit, so rather
1346 		 * than using an alternate bit mapping, we shift the
1347 		 * status bits one space to the right so we can evaluate
1348 		 * them using the 8169 status as though it was in the
1349 		 * same format as that of the 8139C+.
1350 		 */
1351 		if (sc->sc_hwrev != RL_HWREV_8139CPLUS)
1352 			rxstat >>= 1;
1353 
1354 		/*
1355 		 * if total_len > 2^13-1, both _RXERRSUM and _GIANT will be
1356 		 * set, but if CRC is clear, it will still be a valid frame.
1357 		 */
1358 		if ((rxstat & RL_RDESC_STAT_RXERRSUM) != 0 &&
1359 	 	    !(rxstat & RL_RDESC_STAT_RXERRSUM && !(total_len > 8191 &&
1360 		    (rxstat & RL_RDESC_STAT_ERRS) == RL_RDESC_STAT_GIANT))) {
1361 			ifp->if_ierrors++;
1362 			/*
1363 			 * If this is part of a multi-fragment packet,
1364 			 * discard all the pieces.
1365 			 */
1366 			if (sc->rl_head != NULL) {
1367 				m_freem(sc->rl_head);
1368 				sc->rl_head = sc->rl_tail = NULL;
1369 			}
1370 			continue;
1371 		}
1372 
1373 		if (sc->rl_head != NULL) {
1374 			m->m_len = total_len % RL_FRAMELEN(sc->rl_max_mtu);
1375 			if (m->m_len == 0)
1376 				m->m_len = RL_FRAMELEN(sc->rl_max_mtu);
1377 			/*
1378 			 * Special case: if there's 4 bytes or less
1379 			 * in this buffer, the mbuf can be discarded:
1380 			 * the last 4 bytes is the CRC, which we don't
1381 			 * care about anyway.
1382 			 */
1383 			if (m->m_len <= ETHER_CRC_LEN) {
1384 				sc->rl_tail->m_len -=
1385 				    (ETHER_CRC_LEN - m->m_len);
1386 				m_freem(m);
1387 			} else {
1388 				m->m_len -= ETHER_CRC_LEN;
1389 				m->m_flags &= ~M_PKTHDR;
1390 				sc->rl_tail->m_next = m;
1391 			}
1392 			m = sc->rl_head;
1393 			sc->rl_head = sc->rl_tail = NULL;
1394 			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
1395 		} else
1396 			m->m_pkthdr.len = m->m_len =
1397 			    (total_len - ETHER_CRC_LEN);
1398 
1399 		/* Do RX checksumming */
1400 
1401 		if (sc->rl_flags & RL_FLAG_DESCV2) {
1402 			/* Check IP header checksum */
1403 			if ((rxvlan & RL_RDESC_IPV4) &&
1404 			    !(rxstat & RL_RDESC_STAT_IPSUMBAD))
1405 				m->m_pkthdr.csum_flags |= M_IPV4_CSUM_IN_OK;
1406 
1407 			/* Check TCP/UDP checksum */
1408 			if ((rxvlan & (RL_RDESC_IPV4|RL_RDESC_IPV6)) &&
1409 			    (((rxstat & RL_RDESC_STAT_TCP) &&
1410 			    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
1411 			    ((rxstat & RL_RDESC_STAT_UDP) &&
1412 			    !(rxstat & RL_RDESC_STAT_UDPSUMBAD))))
1413 				m->m_pkthdr.csum_flags |= M_TCP_CSUM_IN_OK |
1414 				    M_UDP_CSUM_IN_OK;
1415 		} else {
1416 			/* Check IP header checksum */
1417 			if ((rxstat & RL_RDESC_STAT_PROTOID) &&
1418 			    !(rxstat & RL_RDESC_STAT_IPSUMBAD))
1419 				m->m_pkthdr.csum_flags |= M_IPV4_CSUM_IN_OK;
1420 
1421 			/* Check TCP/UDP checksum */
1422 			if ((RL_TCPPKT(rxstat) &&
1423 			    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
1424 			    (RL_UDPPKT(rxstat) &&
1425 			    !(rxstat & RL_RDESC_STAT_UDPSUMBAD)))
1426 				m->m_pkthdr.csum_flags |= M_TCP_CSUM_IN_OK |
1427 				    M_UDP_CSUM_IN_OK;
1428 		}
1429 #if NVLAN > 0
1430 		if (rxvlan & RL_RDESC_VLANCTL_TAG) {
1431 			m->m_pkthdr.ether_vtag =
1432 			    ntohs((rxvlan & RL_RDESC_VLANCTL_DATA));
1433 			m->m_flags |= M_VLANTAG;
1434 		}
1435 #endif
1436 
1437 		ml_enqueue(&ml, m);
1438 	}
1439 
1440 	sc->rl_ldata.rl_rx_considx = i;
1441 	re_rx_list_fill(sc);
1442 
1443 	if_input(ifp, &ml);
1444 
1445 	return (rx);
1446 }
1447 
1448 int
1449 re_txeof(struct rl_softc *sc)
1450 {
1451 	struct ifnet	*ifp;
1452 	struct rl_txq	*txq;
1453 	uint32_t	txstat;
1454 	int		idx, descidx, tx_free, freed = 0;
1455 
1456 	ifp = &sc->sc_arpcom.ac_if;
1457 
1458 	for (idx = sc->rl_ldata.rl_txq_considx;
1459 	    idx != sc->rl_ldata.rl_txq_prodidx; idx = RL_NEXT_TXQ(sc, idx)) {
1460 		txq = &sc->rl_ldata.rl_txq[idx];
1461 
1462 		descidx = txq->txq_descidx;
1463 		RL_TXDESCSYNC(sc, descidx,
1464 		    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1465 		txstat =
1466 		    letoh32(sc->rl_ldata.rl_tx_list[descidx].rl_cmdstat);
1467 		RL_TXDESCSYNC(sc, descidx, BUS_DMASYNC_PREREAD);
1468 		KASSERT((txstat & RL_TDESC_CMD_EOF) != 0);
1469 		if (txstat & RL_TDESC_CMD_OWN)
1470 			break;
1471 
1472 		freed += txq->txq_nsegs;
1473 		bus_dmamap_sync(sc->sc_dmat, txq->txq_dmamap,
1474 		    0, txq->txq_dmamap->dm_mapsize, BUS_DMASYNC_POSTWRITE);
1475 		bus_dmamap_unload(sc->sc_dmat, txq->txq_dmamap);
1476 		m_freem(txq->txq_mbuf);
1477 		txq->txq_mbuf = NULL;
1478 
1479 		if (txstat & (RL_TDESC_STAT_EXCESSCOL | RL_TDESC_STAT_COLCNT))
1480 			ifp->if_collisions++;
1481 		if (txstat & RL_TDESC_STAT_TXERRSUM)
1482 			ifp->if_oerrors++;
1483 		else
1484 			ifp->if_opackets++;
1485 	}
1486 
1487 	if (freed == 0)
1488 		return (0);
1489 
1490 	tx_free = atomic_add_int_nv(&sc->rl_ldata.rl_tx_free, freed);
1491 	KASSERT(tx_free <= sc->rl_ldata.rl_tx_desc_cnt);
1492 
1493 	sc->rl_ldata.rl_txq_considx = idx;
1494 
1495 	/*
1496 	 * Some chips will ignore a second TX request issued while an
1497 	 * existing transmission is in progress. If the transmitter goes
1498 	 * idle but there are still packets waiting to be sent, we need
1499 	 * to restart the channel here to flush them out. This only
1500 	 * seems to be required with the PCIe devices.
1501 	 */
1502 	if (ifq_is_oactive(&ifp->if_snd))
1503 		ifq_restart(&ifp->if_snd);
1504 	else if (tx_free < sc->rl_ldata.rl_tx_desc_cnt)
1505 		CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
1506 	else
1507 		ifp->if_timer = 0;
1508 
1509 	return (1);
1510 }
1511 
1512 void
1513 re_tick(void *xsc)
1514 {
1515 	struct rl_softc	*sc = xsc;
1516 	struct mii_data	*mii;
1517 	int s;
1518 
1519 	mii = &sc->sc_mii;
1520 
1521 	s = splnet();
1522 
1523 	mii_tick(mii);
1524 
1525 	if ((sc->rl_flags & RL_FLAG_LINK) == 0)
1526 		re_miibus_statchg(&sc->sc_dev);
1527 
1528 	splx(s);
1529 
1530 	timeout_add_sec(&sc->timer_handle, 1);
1531 }
1532 
1533 int
1534 re_intr(void *arg)
1535 {
1536 	struct rl_softc	*sc = arg;
1537 	struct ifnet	*ifp;
1538 	u_int16_t	status;
1539 	int		claimed = 0, rx, tx;
1540 
1541 	ifp = &sc->sc_arpcom.ac_if;
1542 
1543 	if (!(ifp->if_flags & IFF_RUNNING))
1544 		return (0);
1545 
1546 	/* Disable interrupts. */
1547 	CSR_WRITE_2(sc, RL_IMR, 0);
1548 
1549 	rx = tx = 0;
1550 	status = CSR_READ_2(sc, RL_ISR);
1551 	/* If the card has gone away the read returns 0xffff. */
1552 	if (status == 0xffff)
1553 		return (0);
1554 	if (status)
1555 		CSR_WRITE_2(sc, RL_ISR, status);
1556 
1557 	if (status & RL_ISR_TIMEOUT_EXPIRED)
1558 		claimed = 1;
1559 
1560 	if (status & RL_INTRS_CPLUS) {
1561 		if (status & (sc->rl_rx_ack | RL_ISR_RX_ERR)) {
1562 			rx |= re_rxeof(sc);
1563 			claimed = 1;
1564 		}
1565 
1566 		if (status & (sc->rl_tx_ack | RL_ISR_TX_ERR)) {
1567 			tx |= re_txeof(sc);
1568 			claimed = 1;
1569 		}
1570 
1571 		if (status & RL_ISR_SYSTEM_ERR) {
1572 			KERNEL_LOCK();
1573 			re_init(ifp);
1574 			KERNEL_UNLOCK();
1575 			claimed = 1;
1576 		}
1577 	}
1578 
1579 	if (sc->rl_imtype == RL_IMTYPE_SIM) {
1580 		if (sc->rl_timerintr) {
1581 			if ((tx | rx) == 0) {
1582 				/*
1583 				 * Nothing needs to be processed, fallback
1584 				 * to use TX/RX interrupts.
1585 				 */
1586 				re_setup_intr(sc, 1, RL_IMTYPE_NONE);
1587 
1588 				/*
1589 				 * Recollect, mainly to avoid the possible
1590 				 * race introduced by changing interrupt
1591 				 * masks.
1592 				 */
1593 				re_rxeof(sc);
1594 				re_txeof(sc);
1595 			} else
1596 				CSR_WRITE_4(sc, RL_TIMERCNT, 1); /* reload */
1597 		} else if (tx | rx) {
1598 			/*
1599 			 * Assume that using simulated interrupt moderation
1600 			 * (hardware timer based) could reduce the interrupt
1601 			 * rate.
1602 			 */
1603 			re_setup_intr(sc, 1, RL_IMTYPE_SIM);
1604 		}
1605 	}
1606 
1607 	CSR_WRITE_2(sc, RL_IMR, sc->rl_intrs);
1608 
1609 	return (claimed);
1610 }
1611 
1612 int
1613 re_encap(struct rl_softc *sc, struct mbuf *m, struct rl_txq *txq, int *used)
1614 {
1615 	bus_dmamap_t	map;
1616 	struct mbuf	*mp, mh;
1617 	int		error, seg, nsegs, uidx, startidx, curidx, lastidx, pad;
1618 	int		off;
1619 	struct ip	*ip;
1620 	struct rl_desc	*d;
1621 	u_int32_t	cmdstat, vlanctl = 0, csum_flags = 0;
1622 
1623 	/*
1624 	 * Set up checksum offload. Note: checksum offload bits must
1625 	 * appear in all descriptors of a multi-descriptor transmit
1626 	 * attempt. This is according to testing done with an 8169
1627 	 * chip. This is a requirement.
1628 	 */
1629 
1630 	/*
1631 	 * Set RL_TDESC_CMD_IPCSUM if any checksum offloading
1632 	 * is requested.  Otherwise, RL_TDESC_CMD_TCPCSUM/
1633 	 * RL_TDESC_CMD_UDPCSUM does not take affect.
1634 	 */
1635 
1636 	if ((sc->rl_flags & RL_FLAG_JUMBOV2) &&
1637 	    m->m_pkthdr.len > RL_MTU &&
1638 	    (m->m_pkthdr.csum_flags &
1639 	    (M_IPV4_CSUM_OUT|M_TCP_CSUM_OUT|M_UDP_CSUM_OUT)) != 0) {
1640 		mp = m_getptr(m, ETHER_HDR_LEN, &off);
1641 		mh.m_flags = 0;
1642 		mh.m_data = mtod(mp, caddr_t) + off;
1643 		mh.m_next = mp->m_next;
1644 		mh.m_pkthdr.len = mp->m_pkthdr.len - ETHER_HDR_LEN;
1645 		mh.m_len = mp->m_len - off;
1646 		ip = (struct ip *)mh.m_data;
1647 
1648 		if (m->m_pkthdr.csum_flags & M_IPV4_CSUM_OUT)
1649 			ip->ip_sum = in_cksum(&mh, sizeof(struct ip));
1650 		if (m->m_pkthdr.csum_flags & (M_TCP_CSUM_OUT|M_UDP_CSUM_OUT))
1651 			in_delayed_cksum(&mh);
1652 
1653 		m->m_pkthdr.csum_flags &=
1654 		    ~(M_IPV4_CSUM_OUT|M_TCP_CSUM_OUT|M_UDP_CSUM_OUT);
1655 	}
1656 
1657 	if ((m->m_pkthdr.csum_flags &
1658 	    (M_IPV4_CSUM_OUT|M_TCP_CSUM_OUT|M_UDP_CSUM_OUT)) != 0) {
1659 		if (sc->rl_flags & RL_FLAG_DESCV2) {
1660 			vlanctl |= RL_TDESC_CMD_IPCSUMV2;
1661 			if (m->m_pkthdr.csum_flags & M_TCP_CSUM_OUT)
1662 				vlanctl |= RL_TDESC_CMD_TCPCSUMV2;
1663 			if (m->m_pkthdr.csum_flags & M_UDP_CSUM_OUT)
1664 				vlanctl |= RL_TDESC_CMD_UDPCSUMV2;
1665 		} else {
1666 			csum_flags |= RL_TDESC_CMD_IPCSUM;
1667 			if (m->m_pkthdr.csum_flags & M_TCP_CSUM_OUT)
1668 				csum_flags |= RL_TDESC_CMD_TCPCSUM;
1669 			if (m->m_pkthdr.csum_flags & M_UDP_CSUM_OUT)
1670 				csum_flags |= RL_TDESC_CMD_UDPCSUM;
1671 		}
1672 	}
1673 
1674 	map = txq->txq_dmamap;
1675 
1676 	error = bus_dmamap_load_mbuf(sc->sc_dmat, map, m,
1677 	    BUS_DMA_WRITE|BUS_DMA_NOWAIT);
1678 	switch (error) {
1679 	case 0:
1680 		break;
1681 
1682 	case EFBIG:
1683 		if (m_defrag(m, M_DONTWAIT) == 0 &&
1684 		    bus_dmamap_load_mbuf(sc->sc_dmat, map, m,
1685 		    BUS_DMA_WRITE|BUS_DMA_NOWAIT) == 0)
1686 			break;
1687 
1688 		/* FALLTHROUGH */
1689 	default:
1690 		return (ENOMEM);
1691 	}
1692 
1693 	nsegs = map->dm_nsegs;
1694 	pad = 0;
1695 
1696 	/*
1697 	 * With some of the RealTek chips, using the checksum offload
1698 	 * support in conjunction with the autopadding feature results
1699 	 * in the transmission of corrupt frames. For example, if we
1700 	 * need to send a really small IP fragment that's less than 60
1701 	 * bytes in size, and IP header checksumming is enabled, the
1702 	 * resulting ethernet frame that appears on the wire will
1703 	 * have garbled payload. To work around this, if TX IP checksum
1704 	 * offload is enabled, we always manually pad short frames out
1705 	 * to the minimum ethernet frame size.
1706 	 */
1707 	if ((sc->rl_flags & RL_FLAG_AUTOPAD) == 0 &&
1708 	    m->m_pkthdr.len < RL_IP4CSUMTX_PADLEN &&
1709 	    (m->m_pkthdr.csum_flags & M_IPV4_CSUM_OUT) != 0) {
1710 		pad = 1;
1711 		nsegs++;
1712 	}
1713 
1714 	if (*used + nsegs + 1 >= sc->rl_ldata.rl_tx_free) {
1715 		error = ENOBUFS;
1716 		goto fail_unload;
1717 	}
1718 
1719 	/*
1720 	 * Make sure that the caches are synchronized before we
1721 	 * ask the chip to start DMA for the packet data.
1722 	 */
1723 	bus_dmamap_sync(sc->sc_dmat, map, 0, map->dm_mapsize,
1724 		BUS_DMASYNC_PREWRITE);
1725 
1726 	/*
1727 	 * Set up hardware VLAN tagging. Note: vlan tag info must
1728 	 * appear in all descriptors of a multi-descriptor
1729 	 * transmission attempt.
1730 	 */
1731 #if NVLAN > 0
1732 	if (m->m_flags & M_VLANTAG)
1733 		vlanctl |= swap16(m->m_pkthdr.ether_vtag) |
1734 		    RL_TDESC_VLANCTL_TAG;
1735 #endif
1736 
1737 	/*
1738 	 * Map the segment array into descriptors. Note that we set the
1739 	 * start-of-frame and end-of-frame markers for either TX or RX, but
1740 	 * they really only have meaning in the TX case. (In the RX case,
1741 	 * it's the chip that tells us where packets begin and end.)
1742 	 * We also keep track of the end of the ring and set the
1743 	 * end-of-ring bits as needed, and we set the ownership bits
1744 	 * in all except the very first descriptor. (The caller will
1745 	 * set this descriptor later when it start transmission or
1746 	 * reception.)
1747 	 */
1748 	curidx = startidx = sc->rl_ldata.rl_tx_nextfree;
1749 	lastidx = -1;
1750 	for (seg = 0; seg < map->dm_nsegs;
1751 	    seg++, curidx = RL_NEXT_TX_DESC(sc, curidx)) {
1752 		d = &sc->rl_ldata.rl_tx_list[curidx];
1753 		RL_TXDESCSYNC(sc, curidx,
1754 		    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1755 		cmdstat = letoh32(d->rl_cmdstat);
1756 		RL_TXDESCSYNC(sc, curidx, BUS_DMASYNC_PREREAD);
1757 		if (cmdstat & RL_TDESC_STAT_OWN) {
1758 			printf("%s: tried to map busy TX descriptor\n",
1759 			    sc->sc_dev.dv_xname);
1760 			for (; seg > 0; seg --) {
1761 				uidx = (curidx + sc->rl_ldata.rl_tx_desc_cnt -
1762 				    seg) % sc->rl_ldata.rl_tx_desc_cnt;
1763 				sc->rl_ldata.rl_tx_list[uidx].rl_cmdstat = 0;
1764 				RL_TXDESCSYNC(sc, uidx,
1765 				    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1766 			}
1767 			error = EBUSY;
1768 			goto fail_unload;
1769 		}
1770 
1771 		d->rl_vlanctl = htole32(vlanctl);
1772 		re_set_bufaddr(d, map->dm_segs[seg].ds_addr);
1773 		cmdstat = csum_flags | map->dm_segs[seg].ds_len;
1774 		if (seg == 0)
1775 			cmdstat |= RL_TDESC_CMD_SOF;
1776 		else
1777 			cmdstat |= RL_TDESC_CMD_OWN;
1778 		if (curidx == sc->rl_ldata.rl_tx_desc_cnt - 1)
1779 			cmdstat |= RL_TDESC_CMD_EOR;
1780 		if (seg == nsegs - 1) {
1781 			cmdstat |= RL_TDESC_CMD_EOF;
1782 			lastidx = curidx;
1783 		}
1784 		d->rl_cmdstat = htole32(cmdstat);
1785 		RL_TXDESCSYNC(sc, curidx,
1786 		    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1787 	}
1788 	if (pad) {
1789 		d = &sc->rl_ldata.rl_tx_list[curidx];
1790 		d->rl_vlanctl = htole32(vlanctl);
1791 		re_set_bufaddr(d, RL_TXPADDADDR(sc));
1792 		cmdstat = csum_flags |
1793 		    RL_TDESC_CMD_OWN | RL_TDESC_CMD_EOF |
1794 		    (RL_IP4CSUMTX_PADLEN + 1 - m->m_pkthdr.len);
1795 		if (curidx == sc->rl_ldata.rl_tx_desc_cnt - 1)
1796 			cmdstat |= RL_TDESC_CMD_EOR;
1797 		d->rl_cmdstat = htole32(cmdstat);
1798 		RL_TXDESCSYNC(sc, curidx,
1799 		    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1800 		lastidx = curidx;
1801 		curidx = RL_NEXT_TX_DESC(sc, curidx);
1802 	}
1803 	KASSERT(lastidx != -1);
1804 
1805 	/* Transfer ownership of packet to the chip. */
1806 
1807 	sc->rl_ldata.rl_tx_list[startidx].rl_cmdstat |=
1808 	    htole32(RL_TDESC_CMD_OWN);
1809 	RL_TXDESCSYNC(sc, startidx, BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1810 
1811 	/* update info of TX queue and descriptors */
1812 	txq->txq_mbuf = m;
1813 	txq->txq_descidx = lastidx;
1814 	txq->txq_nsegs = nsegs;
1815 
1816 	sc->rl_ldata.rl_tx_nextfree = curidx;
1817 
1818 	*used += nsegs;
1819 
1820 	return (0);
1821 
1822 fail_unload:
1823 	bus_dmamap_unload(sc->sc_dmat, map);
1824 
1825 	return (error);
1826 }
1827 
1828 /*
1829  * Main transmit routine for C+ and gigE NICs.
1830  */
1831 
1832 void
1833 re_start(struct ifnet *ifp)
1834 {
1835 	struct rl_softc	*sc = ifp->if_softc;
1836 	struct mbuf	*m;
1837 	int		idx, used = 0, txq_free, error;
1838 
1839 	if (!ISSET(sc->rl_flags, RL_FLAG_LINK)) {
1840 		IFQ_PURGE(&ifp->if_snd);
1841 		return;
1842 	}
1843 
1844 	txq_free = sc->rl_ldata.rl_txq_considx;
1845 	idx = sc->rl_ldata.rl_txq_prodidx;
1846 	if (txq_free <= idx)
1847 		txq_free += RL_TX_QLEN;
1848 	txq_free -= idx;
1849 
1850 	for (;;) {
1851 		if (txq_free <= 1) {
1852 			ifq_set_oactive(&ifp->if_snd);
1853 			break;
1854 		}
1855 
1856 		m = ifq_deq_begin(&ifp->if_snd);
1857 		if (m == NULL)
1858 			break;
1859 
1860 		error = re_encap(sc, m, &sc->rl_ldata.rl_txq[idx], &used);
1861 		if (error == 0)
1862 			ifq_deq_commit(&ifp->if_snd, m);
1863 		else if (error == ENOBUFS) {
1864 			ifq_deq_rollback(&ifp->if_snd, m);
1865 			ifq_set_oactive(&ifp->if_snd);
1866 			break;
1867 		} else {
1868 			ifq_deq_commit(&ifp->if_snd, m);
1869 			m_freem(m);
1870 			ifp->if_oerrors++;
1871 			continue;
1872 		}
1873 
1874 #if NBPFILTER > 0
1875 		if (ifp->if_bpf)
1876 			bpf_mtap_ether(ifp->if_bpf, m, BPF_DIRECTION_OUT);
1877 #endif
1878 		idx = RL_NEXT_TXQ(sc, idx);
1879 		txq_free--;
1880 	}
1881 
1882 	if (used == 0)
1883 		return;
1884 
1885 	ifp->if_timer = 5;
1886 	atomic_sub_int(&sc->rl_ldata.rl_tx_free, used);
1887 	KASSERT(sc->rl_ldata.rl_tx_free >= 0);
1888 
1889 	sc->rl_ldata.rl_txq_prodidx = idx;
1890 
1891 	CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
1892 }
1893 
1894 int
1895 re_init(struct ifnet *ifp)
1896 {
1897 	struct rl_softc *sc = ifp->if_softc;
1898 	u_int16_t	cfg;
1899 	uint32_t	rxcfg;
1900 	int		s;
1901 	union {
1902 		u_int32_t align_dummy;
1903 		u_char eaddr[ETHER_ADDR_LEN];
1904 	} eaddr;
1905 
1906 	s = splnet();
1907 
1908 	/*
1909 	 * Cancel pending I/O and free all RX/TX buffers.
1910 	 */
1911 	re_stop(ifp);
1912 
1913 	/* Put controller into known state. */
1914 	re_reset(sc);
1915 
1916 	/*
1917 	 * Enable C+ RX and TX mode, as well as VLAN stripping and
1918 	 * RX checksum offload. We must configure the C+ register
1919 	 * before all others.
1920 	 */
1921 	cfg = RL_CPLUSCMD_TXENB | RL_CPLUSCMD_PCI_MRW |
1922 	    RL_CPLUSCMD_RXCSUM_ENB;
1923 
1924 	if (ifp->if_capabilities & IFCAP_VLAN_HWTAGGING)
1925 		cfg |= RL_CPLUSCMD_VLANSTRIP;
1926 
1927 	if (sc->rl_flags & RL_FLAG_MACSTAT)
1928 		cfg |= RL_CPLUSCMD_MACSTAT_DIS;
1929 	else
1930 		cfg |= RL_CPLUSCMD_RXENB;
1931 
1932 	CSR_WRITE_2(sc, RL_CPLUS_CMD, cfg);
1933 
1934 	/*
1935 	 * Init our MAC address.  Even though the chipset
1936 	 * documentation doesn't mention it, we need to enter "Config
1937 	 * register write enable" mode to modify the ID registers.
1938 	 */
1939 	bcopy(sc->sc_arpcom.ac_enaddr, eaddr.eaddr, ETHER_ADDR_LEN);
1940 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG);
1941 	CSR_WRITE_4(sc, RL_IDR4,
1942 	    htole32(*(u_int32_t *)(&eaddr.eaddr[4])));
1943 	CSR_WRITE_4(sc, RL_IDR0,
1944 	    htole32(*(u_int32_t *)(&eaddr.eaddr[0])));
1945 	/*
1946 	 * Default on PC Engines APU1 is to have all LEDs off unless
1947 	 * there is network activity. Override to provide a link status
1948 	 * LED.
1949 	 */
1950 	if (sc->sc_hwrev == RL_HWREV_8168E &&
1951 	    hw_vendor != NULL && hw_prod != NULL &&
1952 	    strcmp(hw_vendor, "PC Engines") == 0 &&
1953 	    strcmp(hw_prod, "APU") == 0) {
1954 		CSR_SETBIT_1(sc, RL_CFG4, RL_CFG4_CUSTOM_LED);
1955 		CSR_WRITE_1(sc, RL_LEDSEL, RL_LED_LINK | RL_LED_ACT << 4);
1956 	}
1957 	/*
1958 	 * Protect config register again
1959 	 */
1960 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1961 
1962 	if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0)
1963 		re_set_jumbo(sc);
1964 
1965 	/*
1966 	 * For C+ mode, initialize the RX descriptors and mbufs.
1967 	 */
1968 	re_rx_list_init(sc);
1969 	re_tx_list_init(sc);
1970 
1971 	/*
1972 	 * Load the addresses of the RX and TX lists into the chip.
1973 	 */
1974 	CSR_WRITE_4(sc, RL_RXLIST_ADDR_HI,
1975 	    RL_ADDR_HI(sc->rl_ldata.rl_rx_list_map->dm_segs[0].ds_addr));
1976 	CSR_WRITE_4(sc, RL_RXLIST_ADDR_LO,
1977 	    RL_ADDR_LO(sc->rl_ldata.rl_rx_list_map->dm_segs[0].ds_addr));
1978 
1979 	CSR_WRITE_4(sc, RL_TXLIST_ADDR_HI,
1980 	    RL_ADDR_HI(sc->rl_ldata.rl_tx_list_map->dm_segs[0].ds_addr));
1981 	CSR_WRITE_4(sc, RL_TXLIST_ADDR_LO,
1982 	    RL_ADDR_LO(sc->rl_ldata.rl_tx_list_map->dm_segs[0].ds_addr));
1983 
1984 	if (sc->rl_flags & RL_FLAG_RXDV_GATED)
1985 		CSR_WRITE_4(sc, RL_MISC, CSR_READ_4(sc, RL_MISC) &
1986 		    ~0x00080000);
1987 
1988 	/*
1989 	 * Set the initial TX and RX configuration.
1990 	 */
1991 	CSR_WRITE_4(sc, RL_TXCFG, RL_TXCFG_CONFIG);
1992 
1993 	CSR_WRITE_1(sc, RL_EARLY_TX_THRESH, 16);
1994 
1995 	rxcfg = RL_RXCFG_CONFIG;
1996 	if (sc->rl_flags & RL_FLAG_EARLYOFF)
1997 		rxcfg |= RL_RXCFG_EARLYOFF;
1998 	else if (sc->rl_flags & RL_FLAG_EARLYOFFV2)
1999 		rxcfg |= RL_RXCFG_EARLYOFFV2;
2000 	CSR_WRITE_4(sc, RL_RXCFG, rxcfg);
2001 
2002 	/*
2003 	 * Enable transmit and receive.
2004 	 */
2005 	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB | RL_CMD_RX_ENB);
2006 
2007 	/* Program promiscuous mode and multicast filters. */
2008 	re_iff(sc);
2009 
2010 	/*
2011 	 * Enable interrupts.
2012 	 */
2013 	re_setup_intr(sc, 1, sc->rl_imtype);
2014 	CSR_WRITE_2(sc, RL_ISR, sc->rl_imtype);
2015 
2016 	/* Start RX/TX process. */
2017 	CSR_WRITE_4(sc, RL_MISSEDPKT, 0);
2018 
2019 	/*
2020 	 * For 8169 gigE NICs, set the max allowed RX packet
2021 	 * size so we can receive jumbo frames.
2022 	 */
2023 	if (sc->sc_hwrev != RL_HWREV_8139CPLUS) {
2024 		if (sc->rl_flags & RL_FLAG_PCIE &&
2025 		    (sc->rl_flags & RL_FLAG_JUMBOV2) == 0)
2026 			CSR_WRITE_2(sc, RL_MAXRXPKTLEN, RE_RX_DESC_BUFLEN);
2027 		else
2028 			CSR_WRITE_2(sc, RL_MAXRXPKTLEN, 16383);
2029 	}
2030 
2031 	CSR_WRITE_1(sc, sc->rl_cfg1, CSR_READ_1(sc, sc->rl_cfg1) |
2032 	    RL_CFG1_DRVLOAD);
2033 
2034 	ifp->if_flags |= IFF_RUNNING;
2035 	ifq_clr_oactive(&ifp->if_snd);
2036 
2037 	splx(s);
2038 
2039 	sc->rl_flags &= ~RL_FLAG_LINK;
2040 	mii_mediachg(&sc->sc_mii);
2041 
2042 	timeout_add_sec(&sc->timer_handle, 1);
2043 
2044 	return (0);
2045 }
2046 
2047 /*
2048  * Set media options.
2049  */
2050 int
2051 re_ifmedia_upd(struct ifnet *ifp)
2052 {
2053 	struct rl_softc	*sc;
2054 
2055 	sc = ifp->if_softc;
2056 
2057 	return (mii_mediachg(&sc->sc_mii));
2058 }
2059 
2060 /*
2061  * Report current media status.
2062  */
2063 void
2064 re_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
2065 {
2066 	struct rl_softc	*sc;
2067 
2068 	sc = ifp->if_softc;
2069 
2070 	mii_pollstat(&sc->sc_mii);
2071 	ifmr->ifm_active = sc->sc_mii.mii_media_active;
2072 	ifmr->ifm_status = sc->sc_mii.mii_media_status;
2073 }
2074 
2075 int
2076 re_ioctl(struct ifnet *ifp, u_long command, caddr_t data)
2077 {
2078 	struct rl_softc	*sc = ifp->if_softc;
2079 	struct ifreq	*ifr = (struct ifreq *) data;
2080 	int		s, error = 0;
2081 
2082 	s = splnet();
2083 
2084 	switch(command) {
2085 	case SIOCSIFADDR:
2086 		ifp->if_flags |= IFF_UP;
2087 		if (!(ifp->if_flags & IFF_RUNNING))
2088 			re_init(ifp);
2089 		break;
2090 	case SIOCSIFFLAGS:
2091 		if (ifp->if_flags & IFF_UP) {
2092 			if (ifp->if_flags & IFF_RUNNING)
2093 				error = ENETRESET;
2094 			else
2095 				re_init(ifp);
2096 		} else {
2097 			if (ifp->if_flags & IFF_RUNNING)
2098 				re_stop(ifp);
2099 		}
2100 		break;
2101 	case SIOCGIFMEDIA:
2102 	case SIOCSIFMEDIA:
2103 		error = ifmedia_ioctl(ifp, ifr, &sc->sc_mii.mii_media, command);
2104 		break;
2105 	case SIOCGIFRXR:
2106 		error = if_rxr_ioctl((struct if_rxrinfo *)ifr->ifr_data,
2107 		    NULL, RL_FRAMELEN(sc->rl_max_mtu), &sc->rl_ldata.rl_rx_ring);
2108  		break;
2109 	default:
2110 		error = ether_ioctl(ifp, &sc->sc_arpcom, command, data);
2111 	}
2112 
2113 	if (error == ENETRESET) {
2114 		if (ifp->if_flags & IFF_RUNNING)
2115 			re_iff(sc);
2116 		error = 0;
2117 	}
2118 
2119 	splx(s);
2120 	return (error);
2121 }
2122 
2123 void
2124 re_watchdog(struct ifnet *ifp)
2125 {
2126 	struct rl_softc	*sc;
2127 	int	s;
2128 
2129 	sc = ifp->if_softc;
2130 	s = splnet();
2131 	printf("%s: watchdog timeout\n", sc->sc_dev.dv_xname);
2132 
2133 	re_txeof(sc);
2134 	re_rxeof(sc);
2135 
2136 	re_init(ifp);
2137 
2138 	splx(s);
2139 }
2140 
2141 /*
2142  * Stop the adapter and free any mbufs allocated to the
2143  * RX and TX lists.
2144  */
2145 void
2146 re_stop(struct ifnet *ifp)
2147 {
2148 	struct rl_softc *sc;
2149 	int	i;
2150 
2151 	sc = ifp->if_softc;
2152 
2153 	ifp->if_timer = 0;
2154 	sc->rl_flags &= ~RL_FLAG_LINK;
2155 	sc->rl_timerintr = 0;
2156 
2157 	timeout_del(&sc->timer_handle);
2158 	ifp->if_flags &= ~IFF_RUNNING;
2159 
2160 	/*
2161 	 * Disable accepting frames to put RX MAC into idle state.
2162 	 * Otherwise it's possible to get frames while stop command
2163 	 * execution is in progress and controller can DMA the frame
2164 	 * to already freed RX buffer during that period.
2165 	 */
2166 	CSR_WRITE_4(sc, RL_RXCFG, CSR_READ_4(sc, RL_RXCFG) &
2167 	    ~(RL_RXCFG_RX_ALLPHYS | RL_RXCFG_RX_BROAD | RL_RXCFG_RX_INDIV |
2168 	    RL_RXCFG_RX_MULTI));
2169 
2170 	if (sc->rl_flags & RL_FLAG_WAIT_TXPOLL) {
2171 		for (i = RL_TIMEOUT; i > 0; i--) {
2172 			if ((CSR_READ_1(sc, sc->rl_txstart) &
2173 			    RL_TXSTART_START) == 0)
2174 				break;
2175 			DELAY(20);
2176 		}
2177 		if (i == 0)
2178 			printf("%s: stopping TX poll timed out!\n",
2179 			    sc->sc_dev.dv_xname);
2180 		CSR_WRITE_1(sc, RL_COMMAND, 0x00);
2181 	} else if (sc->rl_flags & RL_FLAG_CMDSTOP) {
2182 		CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_STOPREQ | RL_CMD_TX_ENB |
2183 		    RL_CMD_RX_ENB);
2184 		if (sc->rl_flags & RL_FLAG_CMDSTOP_WAIT_TXQ) {
2185 			for (i = RL_TIMEOUT; i > 0; i--) {
2186 				if ((CSR_READ_4(sc, RL_TXCFG) &
2187 				    RL_TXCFG_QUEUE_EMPTY) != 0)
2188 					break;
2189 				DELAY(100);
2190 			}
2191 			if (i == 0)
2192 				printf("%s: stopping TXQ timed out!\n",
2193 				    sc->sc_dev.dv_xname);
2194 		}
2195 	} else
2196 		CSR_WRITE_1(sc, RL_COMMAND, 0x00);
2197 	DELAY(1000);
2198 	CSR_WRITE_2(sc, RL_IMR, 0x0000);
2199 	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
2200 
2201 	intr_barrier(sc->sc_ih);
2202 	ifq_barrier(&ifp->if_snd);
2203 
2204 	ifq_clr_oactive(&ifp->if_snd);
2205 	mii_down(&sc->sc_mii);
2206 
2207 	if (sc->rl_head != NULL) {
2208 		m_freem(sc->rl_head);
2209 		sc->rl_head = sc->rl_tail = NULL;
2210 	}
2211 
2212 	/* Free the TX list buffers. */
2213 	for (i = 0; i < RL_TX_QLEN; i++) {
2214 		if (sc->rl_ldata.rl_txq[i].txq_mbuf != NULL) {
2215 			bus_dmamap_unload(sc->sc_dmat,
2216 			    sc->rl_ldata.rl_txq[i].txq_dmamap);
2217 			m_freem(sc->rl_ldata.rl_txq[i].txq_mbuf);
2218 			sc->rl_ldata.rl_txq[i].txq_mbuf = NULL;
2219 		}
2220 	}
2221 
2222 	/* Free the RX list buffers. */
2223 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
2224 		if (sc->rl_ldata.rl_rxsoft[i].rxs_mbuf != NULL) {
2225 			bus_dmamap_unload(sc->sc_dmat,
2226 			    sc->rl_ldata.rl_rxsoft[i].rxs_dmamap);
2227 			m_freem(sc->rl_ldata.rl_rxsoft[i].rxs_mbuf);
2228 			sc->rl_ldata.rl_rxsoft[i].rxs_mbuf = NULL;
2229 		}
2230 	}
2231 }
2232 
2233 void
2234 re_setup_hw_im(struct rl_softc *sc)
2235 {
2236 	KASSERT(sc->rl_flags & RL_FLAG_HWIM);
2237 
2238 	/*
2239 	 * Interrupt moderation
2240 	 *
2241 	 * 0xABCD
2242 	 * A - unknown (maybe TX related)
2243 	 * B - TX timer (unit: 25us)
2244 	 * C - unknown (maybe RX related)
2245 	 * D - RX timer (unit: 25us)
2246 	 *
2247 	 *
2248 	 * re(4)'s interrupt moderation is actually controlled by
2249 	 * two variables, like most other NICs (bge, bnx etc.)
2250 	 * o  timer
2251 	 * o  number of packets [P]
2252 	 *
2253 	 * The logic relationship between these two variables is
2254 	 * similar to other NICs too:
2255 	 * if (timer expire || packets > [P])
2256 	 *     Interrupt is delivered
2257 	 *
2258 	 * Currently we only know how to set 'timer', but not
2259 	 * 'number of packets', which should be ~30, as far as I
2260 	 * tested (sink ~900Kpps, interrupt rate is 30KHz)
2261 	 */
2262 	CSR_WRITE_2(sc, RL_IM,
2263 		    RL_IM_RXTIME(sc->rl_rx_time) |
2264 		    RL_IM_TXTIME(sc->rl_tx_time) |
2265 		    RL_IM_MAGIC);
2266 }
2267 
2268 void
2269 re_disable_hw_im(struct rl_softc *sc)
2270 {
2271 	if (sc->rl_flags & RL_FLAG_HWIM)
2272 		CSR_WRITE_2(sc, RL_IM, 0);
2273 }
2274 
2275 void
2276 re_setup_sim_im(struct rl_softc *sc)
2277 {
2278 	if (sc->sc_hwrev == RL_HWREV_8139CPLUS)
2279 		CSR_WRITE_4(sc, RL_TIMERINT, 0x400); /* XXX */
2280 	else {
2281 		u_int32_t nticks;
2282 
2283 		/*
2284 		 * Datasheet says tick decreases at bus speed,
2285 		 * but it seems the clock runs a little bit
2286 		 * faster, so we do some compensation here.
2287 		 */
2288 		nticks = (sc->rl_sim_time * sc->rl_bus_speed * 8) / 5;
2289 		CSR_WRITE_4(sc, RL_TIMERINT_8169, nticks);
2290 	}
2291 	CSR_WRITE_4(sc, RL_TIMERCNT, 1); /* reload */
2292 	sc->rl_timerintr = 1;
2293 }
2294 
2295 void
2296 re_disable_sim_im(struct rl_softc *sc)
2297 {
2298 	if (sc->sc_hwrev == RL_HWREV_8139CPLUS)
2299 		CSR_WRITE_4(sc, RL_TIMERINT, 0);
2300 	else
2301 		CSR_WRITE_4(sc, RL_TIMERINT_8169, 0);
2302 	sc->rl_timerintr = 0;
2303 }
2304 
2305 void
2306 re_config_imtype(struct rl_softc *sc, int imtype)
2307 {
2308 	switch (imtype) {
2309 	case RL_IMTYPE_HW:
2310 		KASSERT(sc->rl_flags & RL_FLAG_HWIM);
2311 		/* FALLTHROUGH */
2312 	case RL_IMTYPE_NONE:
2313 		sc->rl_intrs = RL_INTRS_CPLUS;
2314 		sc->rl_rx_ack = RL_ISR_RX_OK | RL_ISR_FIFO_OFLOW |
2315 				RL_ISR_RX_OVERRUN;
2316 		sc->rl_tx_ack = RL_ISR_TX_OK;
2317 		break;
2318 
2319 	case RL_IMTYPE_SIM:
2320 		sc->rl_intrs = RL_INTRS_TIMER;
2321 		sc->rl_rx_ack = RL_ISR_TIMEOUT_EXPIRED;
2322 		sc->rl_tx_ack = RL_ISR_TIMEOUT_EXPIRED;
2323 		break;
2324 
2325 	default:
2326 		panic("%s: unknown imtype %d",
2327 		      sc->sc_dev.dv_xname, imtype);
2328 	}
2329 }
2330 
2331 void
2332 re_set_jumbo(struct rl_softc *sc)
2333 {
2334 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG);
2335 	CSR_WRITE_1(sc, RL_CFG3, CSR_READ_1(sc, RL_CFG3) |
2336 	    RL_CFG3_JUMBO_EN0);
2337 
2338 	switch (sc->sc_hwrev) {
2339 	case RL_HWREV_8168DP:
2340 		break;
2341 	case RL_HWREV_8168E:
2342 		CSR_WRITE_1(sc, RL_CFG4, CSR_READ_1(sc, RL_CFG4) |
2343 		    RL_CFG4_8168E_JUMBO_EN1);
2344 		break;
2345 	default:
2346 		CSR_WRITE_1(sc, RL_CFG4, CSR_READ_1(sc, RL_CFG4) |
2347 		    RL_CFG4_JUMBO_EN1);
2348 		break;
2349 	}
2350 
2351 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
2352 }
2353 
2354 void
2355 re_setup_intr(struct rl_softc *sc, int enable_intrs, int imtype)
2356 {
2357 	re_config_imtype(sc, imtype);
2358 
2359 	if (enable_intrs)
2360 		CSR_WRITE_2(sc, RL_IMR, sc->rl_intrs);
2361 	else
2362 		CSR_WRITE_2(sc, RL_IMR, 0);
2363 
2364 	switch (imtype) {
2365 	case RL_IMTYPE_NONE:
2366 		re_disable_sim_im(sc);
2367 		re_disable_hw_im(sc);
2368 		break;
2369 
2370 	case RL_IMTYPE_HW:
2371 		KASSERT(sc->rl_flags & RL_FLAG_HWIM);
2372 		re_disable_sim_im(sc);
2373 		re_setup_hw_im(sc);
2374 		break;
2375 
2376 	case RL_IMTYPE_SIM:
2377 		re_disable_hw_im(sc);
2378 		re_setup_sim_im(sc);
2379 		break;
2380 
2381 	default:
2382 		panic("%s: unknown imtype %d",
2383 		      sc->sc_dev.dv_xname, imtype);
2384 	}
2385 }
2386 
2387 #ifndef SMALL_KERNEL
2388 int
2389 re_wol(struct ifnet *ifp, int enable)
2390 {
2391 	struct rl_softc *sc = ifp->if_softc;
2392 	u_int8_t val;
2393 
2394 	if (enable) {
2395 		if ((CSR_READ_1(sc, sc->rl_cfg1) & RL_CFG1_PME) == 0) {
2396 			printf("%s: power management is disabled, "
2397 			    "cannot do WOL\n", sc->sc_dev.dv_xname);
2398 			return (ENOTSUP);
2399 		}
2400 		if ((CSR_READ_1(sc, sc->rl_cfg2) & RL_CFG2_AUXPWR) == 0)
2401 			printf("%s: no auxiliary power, cannot do WOL from D3 "
2402 			    "(power-off) state\n", sc->sc_dev.dv_xname);
2403 	}
2404 
2405 	re_iff(sc);
2406 
2407 	/* Temporarily enable write to configuration registers. */
2408 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG);
2409 
2410 	/* Always disable all wake events except magic packet. */
2411 	if (enable) {
2412 		val = CSR_READ_1(sc, sc->rl_cfg5);
2413 		val &= ~(RL_CFG5_WOL_UCAST | RL_CFG5_WOL_MCAST |
2414 		    RL_CFG5_WOL_BCAST);
2415 		CSR_WRITE_1(sc, sc->rl_cfg5, val);
2416 
2417 		val = CSR_READ_1(sc, sc->rl_cfg3);
2418 		val |= RL_CFG3_WOL_MAGIC;
2419 		val &= ~RL_CFG3_WOL_LINK;
2420 		CSR_WRITE_1(sc, sc->rl_cfg3, val);
2421 	} else {
2422 		val = CSR_READ_1(sc, sc->rl_cfg5);
2423 		val &= ~(RL_CFG5_WOL_UCAST | RL_CFG5_WOL_MCAST |
2424 		    RL_CFG5_WOL_BCAST);
2425 		CSR_WRITE_1(sc, sc->rl_cfg5, val);
2426 
2427 		val = CSR_READ_1(sc, sc->rl_cfg3);
2428 		val &= ~(RL_CFG3_WOL_MAGIC | RL_CFG3_WOL_LINK);
2429 		CSR_WRITE_1(sc, sc->rl_cfg3, val);
2430 	}
2431 
2432 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
2433 
2434 	return (0);
2435 }
2436 #endif
2437