xref: /openbsd-src/sys/dev/ic/re.c (revision f2da64fbbbf1b03f09f390ab01267c93dfd77c4c)
1 /*	$OpenBSD: re.c,v 1.193 2016/08/10 14:27:17 deraadt 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 /*
648  * Attach the interface. Allocate softc structures, do ifmedia
649  * setup and ethernet/BPF attach.
650  */
651 int
652 re_attach(struct rl_softc *sc, const char *intrstr)
653 {
654 	u_char		eaddr[ETHER_ADDR_LEN];
655 	u_int16_t	as[ETHER_ADDR_LEN / 2];
656 	struct ifnet	*ifp;
657 	u_int16_t	re_did = 0;
658 	int		error = 0, i;
659 	const struct re_revision *rr;
660 	const char	*re_name = NULL;
661 	int		ntxsegs;
662 
663 	sc->sc_hwrev = CSR_READ_4(sc, RL_TXCFG) & RL_TXCFG_HWREV;
664 
665 	switch (sc->sc_hwrev) {
666 	case RL_HWREV_8139CPLUS:
667 		sc->rl_flags |= RL_FLAG_FASTETHER | RL_FLAG_AUTOPAD;
668 		sc->rl_max_mtu = RL_MTU;
669 		break;
670 	case RL_HWREV_8100E:
671 	case RL_HWREV_8100E_SPIN2:
672 	case RL_HWREV_8101E:
673 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_FASTETHER;
674 		sc->rl_max_mtu = RL_MTU;
675 		break;
676 	case RL_HWREV_8103E:
677 		sc->rl_flags |= RL_FLAG_MACSLEEP;
678 		/* FALLTHROUGH */
679 	case RL_HWREV_8102E:
680 	case RL_HWREV_8102EL:
681 	case RL_HWREV_8102EL_SPIN1:
682 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
683 		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_FASTETHER |
684 		    RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD;
685 		sc->rl_max_mtu = RL_MTU;
686 		break;
687 	case RL_HWREV_8401E:
688 	case RL_HWREV_8105E:
689 	case RL_HWREV_8105E_SPIN1:
690 	case RL_HWREV_8106E:
691 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
692 		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
693 		    RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD;
694 		sc->rl_max_mtu = RL_MTU;
695 		break;
696 	case RL_HWREV_8402:
697 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
698 		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
699 		    RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD |
700 		    RL_FLAG_CMDSTOP_WAIT_TXQ;
701 		sc->rl_max_mtu = RL_MTU;
702 		break;
703 	case RL_HWREV_8168B_SPIN1:
704 	case RL_HWREV_8168B_SPIN2:
705 		sc->rl_flags |= RL_FLAG_WOLRXENB;
706 		/* FALLTHROUGH */
707 	case RL_HWREV_8168B_SPIN3:
708 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_MACSTAT;
709 		sc->rl_max_mtu = RL_MTU;
710 		break;
711 	case RL_HWREV_8168C_SPIN2:
712 		sc->rl_flags |= RL_FLAG_MACSLEEP;
713 		/* FALLTHROUGH */
714 	case RL_HWREV_8168C:
715 	case RL_HWREV_8168CP:
716 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
717 		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP |
718 		    RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 | RL_FLAG_WOL_MANLINK;
719 		sc->rl_max_mtu = RL_JUMBO_MTU_6K;
720 		break;
721 	case RL_HWREV_8168D:
722 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
723 		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
724 		    RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 |
725 		    RL_FLAG_WOL_MANLINK;
726 		sc->rl_max_mtu = RL_JUMBO_MTU_9K;
727 		break;
728 	case RL_HWREV_8168DP:
729 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
730 		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_AUTOPAD |
731 		    RL_FLAG_JUMBOV2 | RL_FLAG_WAIT_TXPOLL | RL_FLAG_WOL_MANLINK;
732 		sc->rl_max_mtu = RL_JUMBO_MTU_9K;
733 		break;
734 	case RL_HWREV_8168E:
735 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
736 		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
737 		    RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 |
738 		    RL_FLAG_WOL_MANLINK;
739 		sc->rl_max_mtu = RL_JUMBO_MTU_9K;
740 		break;
741 	case RL_HWREV_8168E_VL:
742 		sc->rl_flags |= RL_FLAG_EARLYOFF | RL_FLAG_PHYWAKE | RL_FLAG_PAR |
743 		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP |
744 		    RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 | RL_FLAG_CMDSTOP_WAIT_TXQ |
745 		    RL_FLAG_WOL_MANLINK;
746 		sc->rl_max_mtu = RL_JUMBO_MTU_6K;
747 		break;
748 	case RL_HWREV_8168F:
749 		sc->rl_flags |= RL_FLAG_EARLYOFF;
750 		/* FALLTHROUGH */
751 	case RL_HWREV_8411:
752 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
753 		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP |
754 		    RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 | RL_FLAG_CMDSTOP_WAIT_TXQ |
755 		    RL_FLAG_WOL_MANLINK;
756 		sc->rl_max_mtu = RL_JUMBO_MTU_9K;
757 		break;
758 	case RL_HWREV_8168EP:
759 	case RL_HWREV_8168G:
760 	case RL_HWREV_8168GU:
761 	case RL_HWREV_8168H:
762 	case RL_HWREV_8411B:
763 		if (sc->sc_product == PCI_PRODUCT_REALTEK_RT8101E) {
764 			/* RTL8106EUS */
765 			sc->rl_flags |= RL_FLAG_FASTETHER;
766 			sc->rl_max_mtu = RL_MTU;
767 		} else {
768 			sc->rl_flags |= RL_FLAG_JUMBOV2 | RL_FLAG_WOL_MANLINK;
769 			sc->rl_max_mtu = RL_JUMBO_MTU_9K;
770 		}
771 
772 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
773 		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP |
774 		    RL_FLAG_AUTOPAD | RL_FLAG_CMDSTOP_WAIT_TXQ |
775 		    RL_FLAG_EARLYOFFV2 | RL_FLAG_RXDV_GATED;
776 		break;
777 	case RL_HWREV_8169_8110SB:
778 	case RL_HWREV_8169_8110SBL:
779 	case RL_HWREV_8169_8110SCd:
780 	case RL_HWREV_8169_8110SCe:
781 		sc->rl_flags |= RL_FLAG_PHYWAKE;
782 		/* FALLTHROUGH */
783 	case RL_HWREV_8169:
784 	case RL_HWREV_8169S:
785 	case RL_HWREV_8110S:
786 		sc->rl_flags |= RL_FLAG_MACRESET;
787 		sc->rl_max_mtu = RL_JUMBO_MTU_7K;
788 		break;
789 	default:
790 		break;
791 	}
792 
793 	if (sc->sc_hwrev == RL_HWREV_8139CPLUS) {
794 		sc->rl_cfg0 = RL_8139_CFG0;
795 		sc->rl_cfg1 = RL_8139_CFG1;
796 		sc->rl_cfg2 = 0;
797 		sc->rl_cfg3 = RL_8139_CFG3;
798 		sc->rl_cfg4 = RL_8139_CFG4;
799 		sc->rl_cfg5 = RL_8139_CFG5;
800 	} else {
801 		sc->rl_cfg0 = RL_CFG0;
802 		sc->rl_cfg1 = RL_CFG1;
803 		sc->rl_cfg2 = RL_CFG2;
804 		sc->rl_cfg3 = RL_CFG3;
805 		sc->rl_cfg4 = RL_CFG4;
806 		sc->rl_cfg5 = RL_CFG5;
807 	}
808 
809 	/* Reset the adapter. */
810 	re_reset(sc);
811 
812 	sc->rl_tx_time = 5;		/* 125us */
813 	sc->rl_rx_time = 2;		/* 50us */
814 	if (sc->rl_flags & RL_FLAG_PCIE)
815 		sc->rl_sim_time = 75;	/* 75us */
816 	else
817 		sc->rl_sim_time = 125;	/* 125us */
818 	sc->rl_imtype = RL_IMTYPE_SIM;	/* simulated interrupt moderation */
819 
820 	if (sc->sc_hwrev == RL_HWREV_8139CPLUS)
821 		sc->rl_bus_speed = 33; /* XXX */
822 	else if (sc->rl_flags & RL_FLAG_PCIE)
823 		sc->rl_bus_speed = 125;
824 	else {
825 		u_int8_t cfg2;
826 
827 		cfg2 = CSR_READ_1(sc, sc->rl_cfg2);
828 		switch (cfg2 & RL_CFG2_PCI_MASK) {
829 		case RL_CFG2_PCI_33MHZ:
830  			sc->rl_bus_speed = 33;
831 			break;
832 		case RL_CFG2_PCI_66MHZ:
833 			sc->rl_bus_speed = 66;
834 			break;
835 		default:
836 			printf("%s: unknown bus speed, assume 33MHz\n",
837 			    sc->sc_dev.dv_xname);
838 			sc->rl_bus_speed = 33;
839 			break;
840 		}
841 
842 		if (cfg2 & RL_CFG2_PCI_64BIT)
843 			sc->rl_flags |= RL_FLAG_PCI64;
844 	}
845 
846 	re_config_imtype(sc, sc->rl_imtype);
847 
848 	if (sc->rl_flags & RL_FLAG_PAR) {
849 		/*
850 		 * XXX Should have a better way to extract station
851 		 * address from EEPROM.
852 		 */
853 		for (i = 0; i < ETHER_ADDR_LEN; i++)
854 			eaddr[i] = CSR_READ_1(sc, RL_IDR0 + i);
855 	} else {
856 		sc->rl_eewidth = RL_9356_ADDR_LEN;
857 		re_read_eeprom(sc, (caddr_t)&re_did, 0, 1);
858 		if (re_did != 0x8129)
859 			sc->rl_eewidth = RL_9346_ADDR_LEN;
860 
861 		/*
862 		 * Get station address from the EEPROM.
863 		 */
864 		re_read_eeprom(sc, (caddr_t)as, RL_EE_EADDR, 3);
865 		for (i = 0; i < ETHER_ADDR_LEN / 2; i++)
866 			as[i] = letoh16(as[i]);
867 		bcopy(as, eaddr, ETHER_ADDR_LEN);
868 	}
869 
870 	/*
871 	 * Set RX length mask, TX poll request register
872 	 * and descriptor count.
873 	 */
874 	if (sc->sc_hwrev == RL_HWREV_8139CPLUS) {
875 		sc->rl_rxlenmask = RL_RDESC_STAT_FRAGLEN;
876 		sc->rl_txstart = RL_TXSTART;
877 		sc->rl_ldata.rl_tx_desc_cnt = RL_8139_TX_DESC_CNT;
878 		sc->rl_ldata.rl_rx_desc_cnt = RL_8139_RX_DESC_CNT;
879 		ntxsegs = RL_8139_NTXSEGS;
880 	} else {
881 		sc->rl_rxlenmask = RL_RDESC_STAT_GFRAGLEN;
882 		sc->rl_txstart = RL_GTXSTART;
883 		sc->rl_ldata.rl_tx_desc_cnt = RL_8169_TX_DESC_CNT;
884 		sc->rl_ldata.rl_rx_desc_cnt = RL_8169_RX_DESC_CNT;
885 		ntxsegs = RL_8169_NTXSEGS;
886 	}
887 
888 	bcopy(eaddr, (char *)&sc->sc_arpcom.ac_enaddr, ETHER_ADDR_LEN);
889 
890 	for (rr = re_revisions; rr->re_name != NULL; rr++) {
891 		if (rr->re_chipid == sc->sc_hwrev)
892 			re_name = rr->re_name;
893 	}
894 
895 	if (re_name == NULL)
896 		printf(": unknown ASIC (0x%04x)", sc->sc_hwrev >> 16);
897 	else
898 		printf(": %s (0x%04x)", re_name, sc->sc_hwrev >> 16);
899 
900 	printf(", %s, address %s\n", intrstr,
901 	    ether_sprintf(sc->sc_arpcom.ac_enaddr));
902 
903 	if (sc->rl_ldata.rl_tx_desc_cnt >
904 	    PAGE_SIZE / sizeof(struct rl_desc)) {
905 		sc->rl_ldata.rl_tx_desc_cnt =
906 		    PAGE_SIZE / sizeof(struct rl_desc);
907 	}
908 
909 	/* Allocate DMA'able memory for the TX ring */
910 	if ((error = bus_dmamem_alloc(sc->sc_dmat, RL_TX_LIST_SZ(sc),
911 		    RL_RING_ALIGN, 0, &sc->rl_ldata.rl_tx_listseg, 1,
912 		    &sc->rl_ldata.rl_tx_listnseg, BUS_DMA_NOWAIT |
913 		    BUS_DMA_ZERO)) != 0) {
914 		printf("%s: can't allocate tx listseg, error = %d\n",
915 		    sc->sc_dev.dv_xname, error);
916 		goto fail_0;
917 	}
918 
919 	/* Load the map for the TX ring. */
920 	if ((error = bus_dmamem_map(sc->sc_dmat, &sc->rl_ldata.rl_tx_listseg,
921 		    sc->rl_ldata.rl_tx_listnseg, RL_TX_LIST_SZ(sc),
922 		    (caddr_t *)&sc->rl_ldata.rl_tx_list,
923 		    BUS_DMA_COHERENT | BUS_DMA_NOWAIT)) != 0) {
924 		printf("%s: can't map tx list, error = %d\n",
925 		    sc->sc_dev.dv_xname, error);
926 		goto fail_1;
927 	}
928 
929 	if ((error = bus_dmamap_create(sc->sc_dmat, RL_TX_LIST_SZ(sc), 1,
930 		    RL_TX_LIST_SZ(sc), 0, 0,
931 		    &sc->rl_ldata.rl_tx_list_map)) != 0) {
932 		printf("%s: can't create tx list map, error = %d\n",
933 		    sc->sc_dev.dv_xname, error);
934 		goto fail_2;
935 	}
936 
937 	if ((error = bus_dmamap_load(sc->sc_dmat,
938 		    sc->rl_ldata.rl_tx_list_map, sc->rl_ldata.rl_tx_list,
939 		    RL_TX_LIST_SZ(sc), NULL, BUS_DMA_NOWAIT)) != 0) {
940 		printf("%s: can't load tx list, error = %d\n",
941 		    sc->sc_dev.dv_xname, error);
942 		goto fail_3;
943 	}
944 
945 	/* Create DMA maps for TX buffers */
946 	for (i = 0; i < RL_TX_QLEN; i++) {
947 		error = bus_dmamap_create(sc->sc_dmat,
948 		    RL_JUMBO_FRAMELEN, ntxsegs, RL_JUMBO_FRAMELEN,
949 		    0, 0, &sc->rl_ldata.rl_txq[i].txq_dmamap);
950 		if (error) {
951 			printf("%s: can't create DMA map for TX\n",
952 			    sc->sc_dev.dv_xname);
953 			goto fail_4;
954 		}
955 	}
956 
957         /* Allocate DMA'able memory for the RX ring */
958 	if ((error = bus_dmamem_alloc(sc->sc_dmat, RL_RX_DMAMEM_SZ(sc),
959 		    RL_RING_ALIGN, 0, &sc->rl_ldata.rl_rx_listseg, 1,
960 		    &sc->rl_ldata.rl_rx_listnseg, BUS_DMA_NOWAIT |
961 		    BUS_DMA_ZERO)) != 0) {
962 		printf("%s: can't allocate rx listnseg, error = %d\n",
963 		    sc->sc_dev.dv_xname, error);
964 		goto fail_4;
965 	}
966 
967         /* Load the map for the RX ring. */
968 	if ((error = bus_dmamem_map(sc->sc_dmat, &sc->rl_ldata.rl_rx_listseg,
969 		    sc->rl_ldata.rl_rx_listnseg, RL_RX_DMAMEM_SZ(sc),
970 		    (caddr_t *)&sc->rl_ldata.rl_rx_list,
971 		    BUS_DMA_COHERENT | BUS_DMA_NOWAIT)) != 0) {
972 		printf("%s: can't map rx list, error = %d\n",
973 		    sc->sc_dev.dv_xname, error);
974 		goto fail_5;
975 
976 	}
977 
978 	if ((error = bus_dmamap_create(sc->sc_dmat, RL_RX_DMAMEM_SZ(sc), 1,
979 		    RL_RX_DMAMEM_SZ(sc), 0, 0,
980 		    &sc->rl_ldata.rl_rx_list_map)) != 0) {
981 		printf("%s: can't create rx list map, error = %d\n",
982 		    sc->sc_dev.dv_xname, error);
983 		goto fail_6;
984 	}
985 
986 	if ((error = bus_dmamap_load(sc->sc_dmat,
987 		    sc->rl_ldata.rl_rx_list_map, sc->rl_ldata.rl_rx_list,
988 		    RL_RX_DMAMEM_SZ(sc), NULL, BUS_DMA_NOWAIT)) != 0) {
989 		printf("%s: can't load rx list, error = %d\n",
990 		    sc->sc_dev.dv_xname, error);
991 		goto fail_7;
992 	}
993 
994 	/* Create DMA maps for RX buffers */
995 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
996 		error = bus_dmamap_create(sc->sc_dmat,
997 		    RL_FRAMELEN(sc->rl_max_mtu), 1,
998 		    RL_FRAMELEN(sc->rl_max_mtu), 0, 0,
999 		    &sc->rl_ldata.rl_rxsoft[i].rxs_dmamap);
1000 		if (error) {
1001 			printf("%s: can't create DMA map for RX\n",
1002 			    sc->sc_dev.dv_xname);
1003 			goto fail_8;
1004 		}
1005 	}
1006 
1007 	ifp = &sc->sc_arpcom.ac_if;
1008 	ifp->if_softc = sc;
1009 	strlcpy(ifp->if_xname, sc->sc_dev.dv_xname, IFNAMSIZ);
1010 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1011 	ifp->if_xflags = IFXF_MPSAFE;
1012 	ifp->if_ioctl = re_ioctl;
1013 	ifp->if_start = re_start;
1014 	ifp->if_watchdog = re_watchdog;
1015 	ifp->if_hardmtu = sc->rl_max_mtu;
1016 	IFQ_SET_MAXLEN(&ifp->if_snd, RL_TX_QLEN);
1017 
1018 	ifp->if_capabilities = IFCAP_VLAN_MTU | IFCAP_CSUM_TCPv4 |
1019 	    IFCAP_CSUM_UDPv4;
1020 
1021 	/*
1022 	 * RTL8168/8111C generates wrong IP checksummed frame if the
1023 	 * packet has IP options so disable TX IP checksum offloading.
1024 	 */
1025 	switch (sc->sc_hwrev) {
1026 	case RL_HWREV_8168C:
1027 	case RL_HWREV_8168C_SPIN2:
1028 	case RL_HWREV_8168CP:
1029 		break;
1030 	default:
1031 		ifp->if_capabilities |= IFCAP_CSUM_IPv4;
1032 	}
1033 
1034 #if NVLAN > 0
1035 	ifp->if_capabilities |= IFCAP_VLAN_HWTAGGING;
1036 #endif
1037 
1038 #ifndef SMALL_KERNEL
1039 	ifp->if_capabilities |= IFCAP_WOL;
1040 	ifp->if_wol = re_wol;
1041 	re_wol(ifp, 0);
1042 #endif
1043 	timeout_set(&sc->timer_handle, re_tick, sc);
1044 
1045 	/* Take PHY out of power down mode. */
1046 	if (sc->rl_flags & RL_FLAG_PHYWAKE_PM) {
1047 		CSR_WRITE_1(sc, RL_PMCH, CSR_READ_1(sc, RL_PMCH) | 0x80);
1048 		if (sc->sc_hwrev == RL_HWREV_8401E)
1049 			CSR_WRITE_1(sc, 0xD1, CSR_READ_1(sc, 0xD1) & ~0x08);
1050 	}
1051 	if (sc->rl_flags & RL_FLAG_PHYWAKE) {
1052 		re_gmii_writereg((struct device *)sc, 1, 0x1f, 0);
1053 		re_gmii_writereg((struct device *)sc, 1, 0x0e, 0);
1054 	}
1055 
1056 	/* Do MII setup */
1057 	sc->sc_mii.mii_ifp = ifp;
1058 	sc->sc_mii.mii_readreg = re_miibus_readreg;
1059 	sc->sc_mii.mii_writereg = re_miibus_writereg;
1060 	sc->sc_mii.mii_statchg = re_miibus_statchg;
1061 	ifmedia_init(&sc->sc_mii.mii_media, IFM_IMASK, re_ifmedia_upd,
1062 	    re_ifmedia_sts);
1063 	mii_attach(&sc->sc_dev, &sc->sc_mii, 0xffffffff, MII_PHY_ANY,
1064 	    MII_OFFSET_ANY, MIIF_DOPAUSE);
1065 	if (LIST_FIRST(&sc->sc_mii.mii_phys) == NULL) {
1066 		printf("%s: no PHY found!\n", sc->sc_dev.dv_xname);
1067 		ifmedia_add(&sc->sc_mii.mii_media,
1068 		    IFM_ETHER|IFM_NONE, 0, NULL);
1069 		ifmedia_set(&sc->sc_mii.mii_media,
1070 		    IFM_ETHER|IFM_NONE);
1071 	} else
1072 		ifmedia_set(&sc->sc_mii.mii_media, IFM_ETHER|IFM_AUTO);
1073 
1074 	/*
1075 	 * Call MI attach routine.
1076 	 */
1077 	if_attach(ifp);
1078 	ether_ifattach(ifp);
1079 
1080 	return (0);
1081 
1082 fail_8:
1083 	/* Destroy DMA maps for RX buffers. */
1084 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1085 		if (sc->rl_ldata.rl_rxsoft[i].rxs_dmamap != NULL)
1086 			bus_dmamap_destroy(sc->sc_dmat,
1087 			    sc->rl_ldata.rl_rxsoft[i].rxs_dmamap);
1088 	}
1089 
1090 	/* Free DMA'able memory for the RX ring. */
1091 	bus_dmamap_unload(sc->sc_dmat, sc->rl_ldata.rl_rx_list_map);
1092 fail_7:
1093 	bus_dmamap_destroy(sc->sc_dmat, sc->rl_ldata.rl_rx_list_map);
1094 fail_6:
1095 	bus_dmamem_unmap(sc->sc_dmat,
1096 	    (caddr_t)sc->rl_ldata.rl_rx_list, RL_RX_DMAMEM_SZ(sc));
1097 fail_5:
1098 	bus_dmamem_free(sc->sc_dmat,
1099 	    &sc->rl_ldata.rl_rx_listseg, sc->rl_ldata.rl_rx_listnseg);
1100 
1101 fail_4:
1102 	/* Destroy DMA maps for TX buffers. */
1103 	for (i = 0; i < RL_TX_QLEN; i++) {
1104 		if (sc->rl_ldata.rl_txq[i].txq_dmamap != NULL)
1105 			bus_dmamap_destroy(sc->sc_dmat,
1106 			    sc->rl_ldata.rl_txq[i].txq_dmamap);
1107 	}
1108 
1109 	/* Free DMA'able memory for the TX ring. */
1110 	bus_dmamap_unload(sc->sc_dmat, sc->rl_ldata.rl_tx_list_map);
1111 fail_3:
1112 	bus_dmamap_destroy(sc->sc_dmat, sc->rl_ldata.rl_tx_list_map);
1113 fail_2:
1114 	bus_dmamem_unmap(sc->sc_dmat,
1115 	    (caddr_t)sc->rl_ldata.rl_tx_list, RL_TX_LIST_SZ(sc));
1116 fail_1:
1117 	bus_dmamem_free(sc->sc_dmat,
1118 	    &sc->rl_ldata.rl_tx_listseg, sc->rl_ldata.rl_tx_listnseg);
1119 fail_0:
1120  	return (1);
1121 }
1122 
1123 
1124 int
1125 re_newbuf(struct rl_softc *sc)
1126 {
1127 	struct mbuf	*m;
1128 	bus_dmamap_t	map;
1129 	struct rl_desc	*d;
1130 	struct rl_rxsoft *rxs;
1131 	u_int32_t	cmdstat;
1132 	int		error, idx;
1133 
1134 	m = MCLGETI(NULL, M_DONTWAIT, NULL, RL_FRAMELEN(sc->rl_max_mtu));
1135 	if (!m)
1136 		return (ENOBUFS);
1137 
1138 	/*
1139 	 * Initialize mbuf length fields and fixup
1140 	 * alignment so that the frame payload is
1141 	 * longword aligned on strict alignment archs.
1142 	 */
1143 	m->m_len = m->m_pkthdr.len = RL_FRAMELEN(sc->rl_max_mtu);
1144 	m->m_data += RE_ETHER_ALIGN;
1145 
1146 	idx = sc->rl_ldata.rl_rx_prodidx;
1147 	rxs = &sc->rl_ldata.rl_rxsoft[idx];
1148 	map = rxs->rxs_dmamap;
1149 	error = bus_dmamap_load_mbuf(sc->sc_dmat, map, m,
1150 	    BUS_DMA_READ|BUS_DMA_NOWAIT);
1151 	if (error) {
1152 		m_freem(m);
1153 		return (ENOBUFS);
1154 	}
1155 
1156 	bus_dmamap_sync(sc->sc_dmat, map, 0, map->dm_mapsize,
1157 	    BUS_DMASYNC_PREREAD);
1158 
1159 	d = &sc->rl_ldata.rl_rx_list[idx];
1160 	RL_RXDESCSYNC(sc, idx, BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1161 	cmdstat = letoh32(d->rl_cmdstat);
1162 	RL_RXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD);
1163 	if (cmdstat & RL_RDESC_STAT_OWN) {
1164 		printf("%s: tried to map busy RX descriptor\n",
1165 		    sc->sc_dev.dv_xname);
1166 		m_freem(m);
1167 		return (ENOBUFS);
1168 	}
1169 
1170 	rxs->rxs_mbuf = m;
1171 
1172 	d->rl_vlanctl = 0;
1173 	cmdstat = map->dm_segs[0].ds_len;
1174 	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
1175 		cmdstat |= RL_RDESC_CMD_EOR;
1176 	re_set_bufaddr(d, map->dm_segs[0].ds_addr);
1177 	d->rl_cmdstat = htole32(cmdstat);
1178 	RL_RXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1179 	cmdstat |= RL_RDESC_CMD_OWN;
1180 	d->rl_cmdstat = htole32(cmdstat);
1181 	RL_RXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1182 
1183 	sc->rl_ldata.rl_rx_prodidx = RL_NEXT_RX_DESC(sc, idx);
1184 
1185 	return (0);
1186 }
1187 
1188 
1189 int
1190 re_tx_list_init(struct rl_softc *sc)
1191 {
1192 	int i;
1193 
1194 	memset(sc->rl_ldata.rl_tx_list, 0, RL_TX_LIST_SZ(sc));
1195 	for (i = 0; i < RL_TX_QLEN; i++) {
1196 		sc->rl_ldata.rl_txq[i].txq_mbuf = NULL;
1197 	}
1198 
1199 	bus_dmamap_sync(sc->sc_dmat,
1200 	    sc->rl_ldata.rl_tx_list_map, 0,
1201 	    sc->rl_ldata.rl_tx_list_map->dm_mapsize,
1202 	    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1203 	sc->rl_ldata.rl_txq_prodidx = 0;
1204 	sc->rl_ldata.rl_txq_considx = 0;
1205 	sc->rl_ldata.rl_tx_free = sc->rl_ldata.rl_tx_desc_cnt;
1206 	sc->rl_ldata.rl_tx_nextfree = 0;
1207 
1208 	return (0);
1209 }
1210 
1211 int
1212 re_rx_list_init(struct rl_softc *sc)
1213 {
1214 	bzero(sc->rl_ldata.rl_rx_list, RL_RX_LIST_SZ(sc));
1215 
1216 	sc->rl_ldata.rl_rx_prodidx = 0;
1217 	sc->rl_ldata.rl_rx_considx = 0;
1218 	sc->rl_head = sc->rl_tail = NULL;
1219 
1220 	if_rxr_init(&sc->rl_ldata.rl_rx_ring, 2, sc->rl_ldata.rl_rx_desc_cnt);
1221 	re_rx_list_fill(sc);
1222 
1223 	return (0);
1224 }
1225 
1226 void
1227 re_rx_list_fill(struct rl_softc *sc)
1228 {
1229 	u_int slots;
1230 
1231 	for (slots = if_rxr_get(&sc->rl_ldata.rl_rx_ring,
1232 	    sc->rl_ldata.rl_rx_desc_cnt);
1233 	    slots > 0; slots--) {
1234 		if (re_newbuf(sc) == ENOBUFS)
1235 			break;
1236 	}
1237 	if_rxr_put(&sc->rl_ldata.rl_rx_ring, slots);
1238 }
1239 
1240 /*
1241  * RX handler for C+ and 8169. For the gigE chips, we support
1242  * the reception of jumbo frames that have been fragmented
1243  * across multiple 2K mbuf cluster buffers.
1244  */
1245 int
1246 re_rxeof(struct rl_softc *sc)
1247 {
1248 	struct mbuf_list ml = MBUF_LIST_INITIALIZER();
1249 	struct mbuf	*m;
1250 	struct ifnet	*ifp;
1251 	int		i, total_len, rx = 0;
1252 	struct rl_desc	*cur_rx;
1253 	struct rl_rxsoft *rxs;
1254 	u_int32_t	rxstat, rxvlan;
1255 
1256 	ifp = &sc->sc_arpcom.ac_if;
1257 
1258 	for (i = sc->rl_ldata.rl_rx_considx;
1259 	    if_rxr_inuse(&sc->rl_ldata.rl_rx_ring) > 0;
1260 	     i = RL_NEXT_RX_DESC(sc, i)) {
1261 		cur_rx = &sc->rl_ldata.rl_rx_list[i];
1262 		RL_RXDESCSYNC(sc, i,
1263 		    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1264 		rxstat = letoh32(cur_rx->rl_cmdstat);
1265 		rxvlan = letoh32(cur_rx->rl_vlanctl);
1266 		RL_RXDESCSYNC(sc, i, BUS_DMASYNC_PREREAD);
1267 		if ((rxstat & RL_RDESC_STAT_OWN) != 0)
1268 			break;
1269 		total_len = rxstat & sc->rl_rxlenmask;
1270 		rxs = &sc->rl_ldata.rl_rxsoft[i];
1271 		m = rxs->rxs_mbuf;
1272 		rxs->rxs_mbuf = NULL;
1273 		if_rxr_put(&sc->rl_ldata.rl_rx_ring, 1);
1274 		rx = 1;
1275 
1276 		/* Invalidate the RX mbuf and unload its map */
1277 
1278 		bus_dmamap_sync(sc->sc_dmat,
1279 		    rxs->rxs_dmamap, 0, rxs->rxs_dmamap->dm_mapsize,
1280 		    BUS_DMASYNC_POSTREAD);
1281 		bus_dmamap_unload(sc->sc_dmat, rxs->rxs_dmamap);
1282 
1283 		if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0 &&
1284 		    (rxstat & (RL_RDESC_STAT_SOF | RL_RDESC_STAT_EOF)) !=
1285 		    (RL_RDESC_STAT_SOF | RL_RDESC_STAT_EOF)) {
1286 			continue;
1287 		} else if (!(rxstat & RL_RDESC_STAT_EOF)) {
1288 			m->m_len = RL_FRAMELEN(sc->rl_max_mtu);
1289 			if (sc->rl_head == NULL)
1290 				sc->rl_head = sc->rl_tail = m;
1291 			else {
1292 				m->m_flags &= ~M_PKTHDR;
1293 				sc->rl_tail->m_next = m;
1294 				sc->rl_tail = m;
1295 			}
1296 			continue;
1297 		}
1298 
1299 		/*
1300 		 * NOTE: for the 8139C+, the frame length field
1301 		 * is always 12 bits in size, but for the gigE chips,
1302 		 * it is 13 bits (since the max RX frame length is 16K).
1303 		 * Unfortunately, all 32 bits in the status word
1304 		 * were already used, so to make room for the extra
1305 		 * length bit, Realtek took out the 'frame alignment
1306 		 * error' bit and shifted the other status bits
1307 		 * over one slot. The OWN, EOR, FS and LS bits are
1308 		 * still in the same places. We have already extracted
1309 		 * the frame length and checked the OWN bit, so rather
1310 		 * than using an alternate bit mapping, we shift the
1311 		 * status bits one space to the right so we can evaluate
1312 		 * them using the 8169 status as though it was in the
1313 		 * same format as that of the 8139C+.
1314 		 */
1315 		if (sc->sc_hwrev != RL_HWREV_8139CPLUS)
1316 			rxstat >>= 1;
1317 
1318 		/*
1319 		 * if total_len > 2^13-1, both _RXERRSUM and _GIANT will be
1320 		 * set, but if CRC is clear, it will still be a valid frame.
1321 		 */
1322 		if ((rxstat & RL_RDESC_STAT_RXERRSUM) != 0 &&
1323 	 	    !(rxstat & RL_RDESC_STAT_RXERRSUM && !(total_len > 8191 &&
1324 		    (rxstat & RL_RDESC_STAT_ERRS) == RL_RDESC_STAT_GIANT))) {
1325 			ifp->if_ierrors++;
1326 			/*
1327 			 * If this is part of a multi-fragment packet,
1328 			 * discard all the pieces.
1329 			 */
1330 			if (sc->rl_head != NULL) {
1331 				m_freem(sc->rl_head);
1332 				sc->rl_head = sc->rl_tail = NULL;
1333 			}
1334 			continue;
1335 		}
1336 
1337 		if (sc->rl_head != NULL) {
1338 			m->m_len = total_len % RL_FRAMELEN(sc->rl_max_mtu);
1339 			if (m->m_len == 0)
1340 				m->m_len = RL_FRAMELEN(sc->rl_max_mtu);
1341 			/*
1342 			 * Special case: if there's 4 bytes or less
1343 			 * in this buffer, the mbuf can be discarded:
1344 			 * the last 4 bytes is the CRC, which we don't
1345 			 * care about anyway.
1346 			 */
1347 			if (m->m_len <= ETHER_CRC_LEN) {
1348 				sc->rl_tail->m_len -=
1349 				    (ETHER_CRC_LEN - m->m_len);
1350 				m_freem(m);
1351 			} else {
1352 				m->m_len -= ETHER_CRC_LEN;
1353 				m->m_flags &= ~M_PKTHDR;
1354 				sc->rl_tail->m_next = m;
1355 			}
1356 			m = sc->rl_head;
1357 			sc->rl_head = sc->rl_tail = NULL;
1358 			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
1359 		} else
1360 			m->m_pkthdr.len = m->m_len =
1361 			    (total_len - ETHER_CRC_LEN);
1362 
1363 		/* Do RX checksumming */
1364 
1365 		if (sc->rl_flags & RL_FLAG_DESCV2) {
1366 			/* Check IP header checksum */
1367 			if ((rxvlan & RL_RDESC_IPV4) &&
1368 			    !(rxstat & RL_RDESC_STAT_IPSUMBAD))
1369 				m->m_pkthdr.csum_flags |= M_IPV4_CSUM_IN_OK;
1370 
1371 			/* Check TCP/UDP checksum */
1372 			if ((rxvlan & (RL_RDESC_IPV4|RL_RDESC_IPV6)) &&
1373 			    (((rxstat & RL_RDESC_STAT_TCP) &&
1374 			    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
1375 			    ((rxstat & RL_RDESC_STAT_UDP) &&
1376 			    !(rxstat & RL_RDESC_STAT_UDPSUMBAD))))
1377 				m->m_pkthdr.csum_flags |= M_TCP_CSUM_IN_OK |
1378 				    M_UDP_CSUM_IN_OK;
1379 		} else {
1380 			/* Check IP header checksum */
1381 			if ((rxstat & RL_RDESC_STAT_PROTOID) &&
1382 			    !(rxstat & RL_RDESC_STAT_IPSUMBAD))
1383 				m->m_pkthdr.csum_flags |= M_IPV4_CSUM_IN_OK;
1384 
1385 			/* Check TCP/UDP checksum */
1386 			if ((RL_TCPPKT(rxstat) &&
1387 			    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
1388 			    (RL_UDPPKT(rxstat) &&
1389 			    !(rxstat & RL_RDESC_STAT_UDPSUMBAD)))
1390 				m->m_pkthdr.csum_flags |= M_TCP_CSUM_IN_OK |
1391 				    M_UDP_CSUM_IN_OK;
1392 		}
1393 #if NVLAN > 0
1394 		if (rxvlan & RL_RDESC_VLANCTL_TAG) {
1395 			m->m_pkthdr.ether_vtag =
1396 			    ntohs((rxvlan & RL_RDESC_VLANCTL_DATA));
1397 			m->m_flags |= M_VLANTAG;
1398 		}
1399 #endif
1400 
1401 		ml_enqueue(&ml, m);
1402 	}
1403 
1404 	sc->rl_ldata.rl_rx_considx = i;
1405 	re_rx_list_fill(sc);
1406 
1407 	if_input(ifp, &ml);
1408 
1409 	return (rx);
1410 }
1411 
1412 int
1413 re_txeof(struct rl_softc *sc)
1414 {
1415 	struct ifnet	*ifp;
1416 	struct rl_txq	*txq;
1417 	uint32_t	txstat;
1418 	int		idx, descidx, tx_free, freed = 0;
1419 
1420 	ifp = &sc->sc_arpcom.ac_if;
1421 
1422 	for (idx = sc->rl_ldata.rl_txq_considx;
1423 	    idx != sc->rl_ldata.rl_txq_prodidx; idx = RL_NEXT_TXQ(sc, idx)) {
1424 		txq = &sc->rl_ldata.rl_txq[idx];
1425 
1426 		descidx = txq->txq_descidx;
1427 		RL_TXDESCSYNC(sc, descidx,
1428 		    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1429 		txstat =
1430 		    letoh32(sc->rl_ldata.rl_tx_list[descidx].rl_cmdstat);
1431 		RL_TXDESCSYNC(sc, descidx, BUS_DMASYNC_PREREAD);
1432 		KASSERT((txstat & RL_TDESC_CMD_EOF) != 0);
1433 		if (txstat & RL_TDESC_CMD_OWN)
1434 			break;
1435 
1436 		freed += txq->txq_nsegs;
1437 		bus_dmamap_sync(sc->sc_dmat, txq->txq_dmamap,
1438 		    0, txq->txq_dmamap->dm_mapsize, BUS_DMASYNC_POSTWRITE);
1439 		bus_dmamap_unload(sc->sc_dmat, txq->txq_dmamap);
1440 		m_freem(txq->txq_mbuf);
1441 		txq->txq_mbuf = NULL;
1442 
1443 		if (txstat & (RL_TDESC_STAT_EXCESSCOL | RL_TDESC_STAT_COLCNT))
1444 			ifp->if_collisions++;
1445 		if (txstat & RL_TDESC_STAT_TXERRSUM)
1446 			ifp->if_oerrors++;
1447 		else
1448 			ifp->if_opackets++;
1449 	}
1450 
1451 	if (freed == 0)
1452 		return (0);
1453 
1454 	tx_free = atomic_add_int_nv(&sc->rl_ldata.rl_tx_free, freed);
1455 	KASSERT(tx_free <= sc->rl_ldata.rl_tx_desc_cnt);
1456 
1457 	sc->rl_ldata.rl_txq_considx = idx;
1458 
1459 	/*
1460 	 * Some chips will ignore a second TX request issued while an
1461 	 * existing transmission is in progress. If the transmitter goes
1462 	 * idle but there are still packets waiting to be sent, we need
1463 	 * to restart the channel here to flush them out. This only
1464 	 * seems to be required with the PCIe devices.
1465 	 */
1466 	if (ifq_is_oactive(&ifp->if_snd))
1467 		ifq_restart(&ifp->if_snd);
1468 	else if (tx_free < sc->rl_ldata.rl_tx_desc_cnt)
1469 		CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
1470 	else
1471 		ifp->if_timer = 0;
1472 
1473 	return (1);
1474 }
1475 
1476 void
1477 re_tick(void *xsc)
1478 {
1479 	struct rl_softc	*sc = xsc;
1480 	struct mii_data	*mii;
1481 	int s;
1482 
1483 	mii = &sc->sc_mii;
1484 
1485 	s = splnet();
1486 
1487 	mii_tick(mii);
1488 
1489 	if ((sc->rl_flags & RL_FLAG_LINK) == 0)
1490 		re_miibus_statchg(&sc->sc_dev);
1491 
1492 	splx(s);
1493 
1494 	timeout_add_sec(&sc->timer_handle, 1);
1495 }
1496 
1497 int
1498 re_intr(void *arg)
1499 {
1500 	struct rl_softc	*sc = arg;
1501 	struct ifnet	*ifp;
1502 	u_int16_t	status;
1503 	int		claimed = 0, rx, tx;
1504 
1505 	ifp = &sc->sc_arpcom.ac_if;
1506 
1507 	if (!(ifp->if_flags & IFF_RUNNING))
1508 		return (0);
1509 
1510 	/* Disable interrupts. */
1511 	CSR_WRITE_2(sc, RL_IMR, 0);
1512 
1513 	rx = tx = 0;
1514 	status = CSR_READ_2(sc, RL_ISR);
1515 	/* If the card has gone away the read returns 0xffff. */
1516 	if (status == 0xffff)
1517 		return (0);
1518 	if (status)
1519 		CSR_WRITE_2(sc, RL_ISR, status);
1520 
1521 	if (status & RL_ISR_TIMEOUT_EXPIRED)
1522 		claimed = 1;
1523 
1524 	if (status & RL_INTRS_CPLUS) {
1525 		if (status & (sc->rl_rx_ack | RL_ISR_RX_ERR)) {
1526 			rx |= re_rxeof(sc);
1527 			claimed = 1;
1528 		}
1529 
1530 		if (status & (sc->rl_tx_ack | RL_ISR_TX_ERR)) {
1531 			tx |= re_txeof(sc);
1532 			claimed = 1;
1533 		}
1534 
1535 		if (status & RL_ISR_SYSTEM_ERR) {
1536 			KERNEL_LOCK();
1537 			re_init(ifp);
1538 			KERNEL_UNLOCK();
1539 			claimed = 1;
1540 		}
1541 	}
1542 
1543 	if (sc->rl_imtype == RL_IMTYPE_SIM) {
1544 		if (sc->rl_timerintr) {
1545 			if ((tx | rx) == 0) {
1546 				/*
1547 				 * Nothing needs to be processed, fallback
1548 				 * to use TX/RX interrupts.
1549 				 */
1550 				re_setup_intr(sc, 1, RL_IMTYPE_NONE);
1551 
1552 				/*
1553 				 * Recollect, mainly to avoid the possible
1554 				 * race introduced by changing interrupt
1555 				 * masks.
1556 				 */
1557 				re_rxeof(sc);
1558 				re_txeof(sc);
1559 			} else
1560 				CSR_WRITE_4(sc, RL_TIMERCNT, 1); /* reload */
1561 		} else if (tx | rx) {
1562 			/*
1563 			 * Assume that using simulated interrupt moderation
1564 			 * (hardware timer based) could reduce the interrupt
1565 			 * rate.
1566 			 */
1567 			re_setup_intr(sc, 1, RL_IMTYPE_SIM);
1568 		}
1569 	}
1570 
1571 	CSR_WRITE_2(sc, RL_IMR, sc->rl_intrs);
1572 
1573 	return (claimed);
1574 }
1575 
1576 int
1577 re_encap(struct rl_softc *sc, struct mbuf *m, struct rl_txq *txq, int *used)
1578 {
1579 	bus_dmamap_t	map;
1580 	struct mbuf	*mp, mh;
1581 	int		error, seg, nsegs, uidx, startidx, curidx, lastidx, pad;
1582 	int		off;
1583 	struct ip	*ip;
1584 	struct rl_desc	*d;
1585 	u_int32_t	cmdstat, vlanctl = 0, csum_flags = 0;
1586 
1587 	/*
1588 	 * Set up checksum offload. Note: checksum offload bits must
1589 	 * appear in all descriptors of a multi-descriptor transmit
1590 	 * attempt. This is according to testing done with an 8169
1591 	 * chip. This is a requirement.
1592 	 */
1593 
1594 	/*
1595 	 * Set RL_TDESC_CMD_IPCSUM if any checksum offloading
1596 	 * is requested.  Otherwise, RL_TDESC_CMD_TCPCSUM/
1597 	 * RL_TDESC_CMD_UDPCSUM does not take affect.
1598 	 */
1599 
1600 	if ((sc->rl_flags & RL_FLAG_JUMBOV2) &&
1601 	    m->m_pkthdr.len > RL_MTU &&
1602 	    (m->m_pkthdr.csum_flags &
1603 	    (M_IPV4_CSUM_OUT|M_TCP_CSUM_OUT|M_UDP_CSUM_OUT)) != 0) {
1604 		mp = m_getptr(m, ETHER_HDR_LEN, &off);
1605 		mh.m_flags = 0;
1606 		mh.m_data = mtod(mp, caddr_t) + off;
1607 		mh.m_next = mp->m_next;
1608 		mh.m_pkthdr.len = mp->m_pkthdr.len - ETHER_HDR_LEN;
1609 		mh.m_len = mp->m_len - off;
1610 		ip = (struct ip *)mh.m_data;
1611 
1612 		if (m->m_pkthdr.csum_flags & M_IPV4_CSUM_OUT)
1613 			ip->ip_sum = in_cksum(&mh, sizeof(struct ip));
1614 		if (m->m_pkthdr.csum_flags & (M_TCP_CSUM_OUT|M_UDP_CSUM_OUT))
1615 			in_delayed_cksum(&mh);
1616 
1617 		m->m_pkthdr.csum_flags &=
1618 		    ~(M_IPV4_CSUM_OUT|M_TCP_CSUM_OUT|M_UDP_CSUM_OUT);
1619 	}
1620 
1621 	if ((m->m_pkthdr.csum_flags &
1622 	    (M_IPV4_CSUM_OUT|M_TCP_CSUM_OUT|M_UDP_CSUM_OUT)) != 0) {
1623 		if (sc->rl_flags & RL_FLAG_DESCV2) {
1624 			vlanctl |= RL_TDESC_CMD_IPCSUMV2;
1625 			if (m->m_pkthdr.csum_flags & M_TCP_CSUM_OUT)
1626 				vlanctl |= RL_TDESC_CMD_TCPCSUMV2;
1627 			if (m->m_pkthdr.csum_flags & M_UDP_CSUM_OUT)
1628 				vlanctl |= RL_TDESC_CMD_UDPCSUMV2;
1629 		} else {
1630 			csum_flags |= RL_TDESC_CMD_IPCSUM;
1631 			if (m->m_pkthdr.csum_flags & M_TCP_CSUM_OUT)
1632 				csum_flags |= RL_TDESC_CMD_TCPCSUM;
1633 			if (m->m_pkthdr.csum_flags & M_UDP_CSUM_OUT)
1634 				csum_flags |= RL_TDESC_CMD_UDPCSUM;
1635 		}
1636 	}
1637 
1638 	map = txq->txq_dmamap;
1639 
1640 	error = bus_dmamap_load_mbuf(sc->sc_dmat, map, m,
1641 	    BUS_DMA_WRITE|BUS_DMA_NOWAIT);
1642 	switch (error) {
1643 	case 0:
1644 		break;
1645 
1646 	case EFBIG:
1647 		if (m_defrag(m, M_DONTWAIT) == 0 &&
1648 		    bus_dmamap_load_mbuf(sc->sc_dmat, map, m,
1649 		    BUS_DMA_WRITE|BUS_DMA_NOWAIT) == 0)
1650 			break;
1651 
1652 		/* FALLTHROUGH */
1653 	default:
1654 		return (ENOMEM);
1655 	}
1656 
1657 	nsegs = map->dm_nsegs;
1658 	pad = 0;
1659 
1660 	/*
1661 	 * With some of the RealTek chips, using the checksum offload
1662 	 * support in conjunction with the autopadding feature results
1663 	 * in the transmission of corrupt frames. For example, if we
1664 	 * need to send a really small IP fragment that's less than 60
1665 	 * bytes in size, and IP header checksumming is enabled, the
1666 	 * resulting ethernet frame that appears on the wire will
1667 	 * have garbled payload. To work around this, if TX IP checksum
1668 	 * offload is enabled, we always manually pad short frames out
1669 	 * to the minimum ethernet frame size.
1670 	 */
1671 	if ((sc->rl_flags & RL_FLAG_AUTOPAD) == 0 &&
1672 	    m->m_pkthdr.len < RL_IP4CSUMTX_PADLEN &&
1673 	    (m->m_pkthdr.csum_flags & M_IPV4_CSUM_OUT) != 0) {
1674 		pad = 1;
1675 		nsegs++;
1676 	}
1677 
1678 	if (*used + nsegs + 1 >= sc->rl_ldata.rl_tx_free) {
1679 		error = ENOBUFS;
1680 		goto fail_unload;
1681 	}
1682 
1683 	/*
1684 	 * Make sure that the caches are synchronized before we
1685 	 * ask the chip to start DMA for the packet data.
1686 	 */
1687 	bus_dmamap_sync(sc->sc_dmat, map, 0, map->dm_mapsize,
1688 		BUS_DMASYNC_PREWRITE);
1689 
1690 	/*
1691 	 * Set up hardware VLAN tagging. Note: vlan tag info must
1692 	 * appear in all descriptors of a multi-descriptor
1693 	 * transmission attempt.
1694 	 */
1695 #if NVLAN > 0
1696 	if (m->m_flags & M_VLANTAG)
1697 		vlanctl |= swap16(m->m_pkthdr.ether_vtag) |
1698 		    RL_TDESC_VLANCTL_TAG;
1699 #endif
1700 
1701 	/*
1702 	 * Map the segment array into descriptors. Note that we set the
1703 	 * start-of-frame and end-of-frame markers for either TX or RX, but
1704 	 * they really only have meaning in the TX case. (In the RX case,
1705 	 * it's the chip that tells us where packets begin and end.)
1706 	 * We also keep track of the end of the ring and set the
1707 	 * end-of-ring bits as needed, and we set the ownership bits
1708 	 * in all except the very first descriptor. (The caller will
1709 	 * set this descriptor later when it start transmission or
1710 	 * reception.)
1711 	 */
1712 	curidx = startidx = sc->rl_ldata.rl_tx_nextfree;
1713 	lastidx = -1;
1714 	for (seg = 0; seg < map->dm_nsegs;
1715 	    seg++, curidx = RL_NEXT_TX_DESC(sc, curidx)) {
1716 		d = &sc->rl_ldata.rl_tx_list[curidx];
1717 		RL_TXDESCSYNC(sc, curidx,
1718 		    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1719 		cmdstat = letoh32(d->rl_cmdstat);
1720 		RL_TXDESCSYNC(sc, curidx, BUS_DMASYNC_PREREAD);
1721 		if (cmdstat & RL_TDESC_STAT_OWN) {
1722 			printf("%s: tried to map busy TX descriptor\n",
1723 			    sc->sc_dev.dv_xname);
1724 			for (; seg > 0; seg --) {
1725 				uidx = (curidx + sc->rl_ldata.rl_tx_desc_cnt -
1726 				    seg) % sc->rl_ldata.rl_tx_desc_cnt;
1727 				sc->rl_ldata.rl_tx_list[uidx].rl_cmdstat = 0;
1728 				RL_TXDESCSYNC(sc, uidx,
1729 				    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1730 			}
1731 			error = EBUSY;
1732 			goto fail_unload;
1733 		}
1734 
1735 		d->rl_vlanctl = htole32(vlanctl);
1736 		re_set_bufaddr(d, map->dm_segs[seg].ds_addr);
1737 		cmdstat = csum_flags | map->dm_segs[seg].ds_len;
1738 		if (seg == 0)
1739 			cmdstat |= RL_TDESC_CMD_SOF;
1740 		else
1741 			cmdstat |= RL_TDESC_CMD_OWN;
1742 		if (curidx == sc->rl_ldata.rl_tx_desc_cnt - 1)
1743 			cmdstat |= RL_TDESC_CMD_EOR;
1744 		if (seg == nsegs - 1) {
1745 			cmdstat |= RL_TDESC_CMD_EOF;
1746 			lastidx = curidx;
1747 		}
1748 		d->rl_cmdstat = htole32(cmdstat);
1749 		RL_TXDESCSYNC(sc, curidx,
1750 		    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1751 	}
1752 	if (pad) {
1753 		d = &sc->rl_ldata.rl_tx_list[curidx];
1754 		d->rl_vlanctl = htole32(vlanctl);
1755 		re_set_bufaddr(d, RL_TXPADDADDR(sc));
1756 		cmdstat = csum_flags |
1757 		    RL_TDESC_CMD_OWN | RL_TDESC_CMD_EOF |
1758 		    (RL_IP4CSUMTX_PADLEN + 1 - m->m_pkthdr.len);
1759 		if (curidx == sc->rl_ldata.rl_tx_desc_cnt - 1)
1760 			cmdstat |= RL_TDESC_CMD_EOR;
1761 		d->rl_cmdstat = htole32(cmdstat);
1762 		RL_TXDESCSYNC(sc, curidx,
1763 		    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1764 		lastidx = curidx;
1765 		curidx = RL_NEXT_TX_DESC(sc, curidx);
1766 	}
1767 	KASSERT(lastidx != -1);
1768 
1769 	/* Transfer ownership of packet to the chip. */
1770 
1771 	sc->rl_ldata.rl_tx_list[startidx].rl_cmdstat |=
1772 	    htole32(RL_TDESC_CMD_OWN);
1773 	RL_TXDESCSYNC(sc, startidx, BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1774 
1775 	/* update info of TX queue and descriptors */
1776 	txq->txq_mbuf = m;
1777 	txq->txq_descidx = lastidx;
1778 	txq->txq_nsegs = nsegs;
1779 
1780 	sc->rl_ldata.rl_tx_nextfree = curidx;
1781 
1782 	*used += nsegs;
1783 
1784 	return (0);
1785 
1786 fail_unload:
1787 	bus_dmamap_unload(sc->sc_dmat, map);
1788 
1789 	return (error);
1790 }
1791 
1792 /*
1793  * Main transmit routine for C+ and gigE NICs.
1794  */
1795 
1796 void
1797 re_start(struct ifnet *ifp)
1798 {
1799 	struct rl_softc	*sc = ifp->if_softc;
1800 	struct mbuf	*m;
1801 	int		idx, used = 0, txq_free, error;
1802 
1803 	if (!ISSET(sc->rl_flags, RL_FLAG_LINK)) {
1804 		IFQ_PURGE(&ifp->if_snd);
1805 		return;
1806 	}
1807 
1808 	txq_free = sc->rl_ldata.rl_txq_considx;
1809 	idx = sc->rl_ldata.rl_txq_prodidx;
1810 	if (txq_free <= idx)
1811 		txq_free += RL_TX_QLEN;
1812 	txq_free -= idx;
1813 
1814 	for (;;) {
1815 		if (txq_free <= 1) {
1816 			ifq_set_oactive(&ifp->if_snd);
1817 			break;
1818 		}
1819 
1820 		m = ifq_deq_begin(&ifp->if_snd);
1821 		if (m == NULL)
1822 			break;
1823 
1824 		error = re_encap(sc, m, &sc->rl_ldata.rl_txq[idx], &used);
1825 		if (error == 0)
1826 			ifq_deq_commit(&ifp->if_snd, m);
1827 		else if (error == ENOBUFS) {
1828 			ifq_deq_rollback(&ifp->if_snd, m);
1829 			ifq_set_oactive(&ifp->if_snd);
1830 			break;
1831 		} else {
1832 			ifq_deq_commit(&ifp->if_snd, m);
1833 			m_freem(m);
1834 			ifp->if_oerrors++;
1835 			continue;
1836 		}
1837 
1838 #if NBPFILTER > 0
1839 		if (ifp->if_bpf)
1840 			bpf_mtap_ether(ifp->if_bpf, m, BPF_DIRECTION_OUT);
1841 #endif
1842 		idx = RL_NEXT_TXQ(sc, idx);
1843 		txq_free--;
1844 	}
1845 
1846 	if (used == 0)
1847 		return;
1848 
1849 	ifp->if_timer = 5;
1850 	atomic_sub_int(&sc->rl_ldata.rl_tx_free, used);
1851 	KASSERT(sc->rl_ldata.rl_tx_free >= 0);
1852 
1853 	sc->rl_ldata.rl_txq_prodidx = idx;
1854 
1855 	CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
1856 }
1857 
1858 int
1859 re_init(struct ifnet *ifp)
1860 {
1861 	struct rl_softc *sc = ifp->if_softc;
1862 	u_int16_t	cfg;
1863 	uint32_t	rxcfg;
1864 	int		s;
1865 	union {
1866 		u_int32_t align_dummy;
1867 		u_char eaddr[ETHER_ADDR_LEN];
1868 	} eaddr;
1869 
1870 	s = splnet();
1871 
1872 	/*
1873 	 * Cancel pending I/O and free all RX/TX buffers.
1874 	 */
1875 	re_stop(ifp);
1876 
1877 	/* Put controller into known state. */
1878 	re_reset(sc);
1879 
1880 	/*
1881 	 * Enable C+ RX and TX mode, as well as VLAN stripping and
1882 	 * RX checksum offload. We must configure the C+ register
1883 	 * before all others.
1884 	 */
1885 	cfg = RL_CPLUSCMD_TXENB | RL_CPLUSCMD_PCI_MRW |
1886 	    RL_CPLUSCMD_RXCSUM_ENB;
1887 
1888 	if (ifp->if_capabilities & IFCAP_VLAN_HWTAGGING)
1889 		cfg |= RL_CPLUSCMD_VLANSTRIP;
1890 
1891 	if (sc->rl_flags & RL_FLAG_MACSTAT)
1892 		cfg |= RL_CPLUSCMD_MACSTAT_DIS;
1893 	else
1894 		cfg |= RL_CPLUSCMD_RXENB;
1895 
1896 	CSR_WRITE_2(sc, RL_CPLUS_CMD, cfg);
1897 
1898 	/*
1899 	 * Init our MAC address.  Even though the chipset
1900 	 * documentation doesn't mention it, we need to enter "Config
1901 	 * register write enable" mode to modify the ID registers.
1902 	 */
1903 	bcopy(sc->sc_arpcom.ac_enaddr, eaddr.eaddr, ETHER_ADDR_LEN);
1904 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG);
1905 	CSR_WRITE_4(sc, RL_IDR4,
1906 	    htole32(*(u_int32_t *)(&eaddr.eaddr[4])));
1907 	CSR_WRITE_4(sc, RL_IDR0,
1908 	    htole32(*(u_int32_t *)(&eaddr.eaddr[0])));
1909 	/*
1910 	 * Default on PC Engines APU1 is to have all LEDs off unless
1911 	 * there is network activity. Override to provide a link status
1912 	 * LED.
1913 	 */
1914 	if (sc->sc_hwrev == RL_HWREV_8168E &&
1915 	    hw_vendor != NULL && hw_prod != NULL &&
1916 	    strcmp(hw_vendor, "PC Engines") == 0 &&
1917 	    strcmp(hw_prod, "APU") == 0) {
1918 		CSR_SETBIT_1(sc, RL_CFG4, RL_CFG4_CUSTOM_LED);
1919 		CSR_WRITE_1(sc, RL_LEDSEL, RL_LED_LINK | RL_LED_ACT << 4);
1920 	}
1921 	/*
1922 	 * Protect config register again
1923 	 */
1924 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1925 
1926 	if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0)
1927 		re_set_jumbo(sc);
1928 
1929 	/*
1930 	 * For C+ mode, initialize the RX descriptors and mbufs.
1931 	 */
1932 	re_rx_list_init(sc);
1933 	re_tx_list_init(sc);
1934 
1935 	/*
1936 	 * Load the addresses of the RX and TX lists into the chip.
1937 	 */
1938 	CSR_WRITE_4(sc, RL_RXLIST_ADDR_HI,
1939 	    RL_ADDR_HI(sc->rl_ldata.rl_rx_list_map->dm_segs[0].ds_addr));
1940 	CSR_WRITE_4(sc, RL_RXLIST_ADDR_LO,
1941 	    RL_ADDR_LO(sc->rl_ldata.rl_rx_list_map->dm_segs[0].ds_addr));
1942 
1943 	CSR_WRITE_4(sc, RL_TXLIST_ADDR_HI,
1944 	    RL_ADDR_HI(sc->rl_ldata.rl_tx_list_map->dm_segs[0].ds_addr));
1945 	CSR_WRITE_4(sc, RL_TXLIST_ADDR_LO,
1946 	    RL_ADDR_LO(sc->rl_ldata.rl_tx_list_map->dm_segs[0].ds_addr));
1947 
1948 	if (sc->rl_flags & RL_FLAG_RXDV_GATED)
1949 		CSR_WRITE_4(sc, RL_MISC, CSR_READ_4(sc, RL_MISC) &
1950 		    ~0x00080000);
1951 
1952 	/*
1953 	 * Set the initial TX and RX configuration.
1954 	 */
1955 	CSR_WRITE_4(sc, RL_TXCFG, RL_TXCFG_CONFIG);
1956 
1957 	CSR_WRITE_1(sc, RL_EARLY_TX_THRESH, 16);
1958 
1959 	rxcfg = RL_RXCFG_CONFIG;
1960 	if (sc->rl_flags & RL_FLAG_EARLYOFF)
1961 		rxcfg |= RL_RXCFG_EARLYOFF;
1962 	else if (sc->rl_flags & RL_FLAG_EARLYOFFV2)
1963 		rxcfg |= RL_RXCFG_EARLYOFFV2;
1964 	CSR_WRITE_4(sc, RL_RXCFG, rxcfg);
1965 
1966 	/*
1967 	 * Enable transmit and receive.
1968 	 */
1969 	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB | RL_CMD_RX_ENB);
1970 
1971 	/* Program promiscuous mode and multicast filters. */
1972 	re_iff(sc);
1973 
1974 	/*
1975 	 * Enable interrupts.
1976 	 */
1977 	re_setup_intr(sc, 1, sc->rl_imtype);
1978 	CSR_WRITE_2(sc, RL_ISR, sc->rl_imtype);
1979 
1980 	/* Start RX/TX process. */
1981 	CSR_WRITE_4(sc, RL_MISSEDPKT, 0);
1982 
1983 	/*
1984 	 * For 8169 gigE NICs, set the max allowed RX packet
1985 	 * size so we can receive jumbo frames.
1986 	 */
1987 	if (sc->sc_hwrev != RL_HWREV_8139CPLUS) {
1988 		if (sc->rl_flags & RL_FLAG_PCIE &&
1989 		    (sc->rl_flags & RL_FLAG_JUMBOV2) == 0)
1990 			CSR_WRITE_2(sc, RL_MAXRXPKTLEN, RE_RX_DESC_BUFLEN);
1991 		else
1992 			CSR_WRITE_2(sc, RL_MAXRXPKTLEN, 16383);
1993 	}
1994 
1995 	CSR_WRITE_1(sc, sc->rl_cfg1, CSR_READ_1(sc, sc->rl_cfg1) |
1996 	    RL_CFG1_DRVLOAD);
1997 
1998 	ifp->if_flags |= IFF_RUNNING;
1999 	ifq_clr_oactive(&ifp->if_snd);
2000 
2001 	splx(s);
2002 
2003 	sc->rl_flags &= ~RL_FLAG_LINK;
2004 	mii_mediachg(&sc->sc_mii);
2005 
2006 	timeout_add_sec(&sc->timer_handle, 1);
2007 
2008 	return (0);
2009 }
2010 
2011 /*
2012  * Set media options.
2013  */
2014 int
2015 re_ifmedia_upd(struct ifnet *ifp)
2016 {
2017 	struct rl_softc	*sc;
2018 
2019 	sc = ifp->if_softc;
2020 
2021 	return (mii_mediachg(&sc->sc_mii));
2022 }
2023 
2024 /*
2025  * Report current media status.
2026  */
2027 void
2028 re_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
2029 {
2030 	struct rl_softc	*sc;
2031 
2032 	sc = ifp->if_softc;
2033 
2034 	mii_pollstat(&sc->sc_mii);
2035 	ifmr->ifm_active = sc->sc_mii.mii_media_active;
2036 	ifmr->ifm_status = sc->sc_mii.mii_media_status;
2037 }
2038 
2039 int
2040 re_ioctl(struct ifnet *ifp, u_long command, caddr_t data)
2041 {
2042 	struct rl_softc	*sc = ifp->if_softc;
2043 	struct ifreq	*ifr = (struct ifreq *) data;
2044 	int		s, error = 0;
2045 
2046 	s = splnet();
2047 
2048 	switch(command) {
2049 	case SIOCSIFADDR:
2050 		ifp->if_flags |= IFF_UP;
2051 		if (!(ifp->if_flags & IFF_RUNNING))
2052 			re_init(ifp);
2053 		break;
2054 	case SIOCSIFFLAGS:
2055 		if (ifp->if_flags & IFF_UP) {
2056 			if (ifp->if_flags & IFF_RUNNING)
2057 				error = ENETRESET;
2058 			else
2059 				re_init(ifp);
2060 		} else {
2061 			if (ifp->if_flags & IFF_RUNNING)
2062 				re_stop(ifp);
2063 		}
2064 		break;
2065 	case SIOCGIFMEDIA:
2066 	case SIOCSIFMEDIA:
2067 		error = ifmedia_ioctl(ifp, ifr, &sc->sc_mii.mii_media, command);
2068 		break;
2069 	case SIOCGIFRXR:
2070 		error = if_rxr_ioctl((struct if_rxrinfo *)ifr->ifr_data,
2071 		    NULL, RL_FRAMELEN(sc->rl_max_mtu), &sc->rl_ldata.rl_rx_ring);
2072  		break;
2073 	default:
2074 		error = ether_ioctl(ifp, &sc->sc_arpcom, command, data);
2075 	}
2076 
2077 	if (error == ENETRESET) {
2078 		if (ifp->if_flags & IFF_RUNNING)
2079 			re_iff(sc);
2080 		error = 0;
2081 	}
2082 
2083 	splx(s);
2084 	return (error);
2085 }
2086 
2087 void
2088 re_watchdog(struct ifnet *ifp)
2089 {
2090 	struct rl_softc	*sc;
2091 	int	s;
2092 
2093 	sc = ifp->if_softc;
2094 	s = splnet();
2095 	printf("%s: watchdog timeout\n", sc->sc_dev.dv_xname);
2096 
2097 	re_txeof(sc);
2098 	re_rxeof(sc);
2099 
2100 	re_init(ifp);
2101 
2102 	splx(s);
2103 }
2104 
2105 /*
2106  * Stop the adapter and free any mbufs allocated to the
2107  * RX and TX lists.
2108  */
2109 void
2110 re_stop(struct ifnet *ifp)
2111 {
2112 	struct rl_softc *sc;
2113 	int	i;
2114 
2115 	sc = ifp->if_softc;
2116 
2117 	ifp->if_timer = 0;
2118 	sc->rl_flags &= ~RL_FLAG_LINK;
2119 	sc->rl_timerintr = 0;
2120 
2121 	timeout_del(&sc->timer_handle);
2122 	ifp->if_flags &= ~IFF_RUNNING;
2123 
2124 	/*
2125 	 * Disable accepting frames to put RX MAC into idle state.
2126 	 * Otherwise it's possible to get frames while stop command
2127 	 * execution is in progress and controller can DMA the frame
2128 	 * to already freed RX buffer during that period.
2129 	 */
2130 	CSR_WRITE_4(sc, RL_RXCFG, CSR_READ_4(sc, RL_RXCFG) &
2131 	    ~(RL_RXCFG_RX_ALLPHYS | RL_RXCFG_RX_BROAD | RL_RXCFG_RX_INDIV |
2132 	    RL_RXCFG_RX_MULTI));
2133 
2134 	if (sc->rl_flags & RL_FLAG_WAIT_TXPOLL) {
2135 		for (i = RL_TIMEOUT; i > 0; i--) {
2136 			if ((CSR_READ_1(sc, sc->rl_txstart) &
2137 			    RL_TXSTART_START) == 0)
2138 				break;
2139 			DELAY(20);
2140 		}
2141 		if (i == 0)
2142 			printf("%s: stopping TX poll timed out!\n",
2143 			    sc->sc_dev.dv_xname);
2144 		CSR_WRITE_1(sc, RL_COMMAND, 0x00);
2145 	} else if (sc->rl_flags & RL_FLAG_CMDSTOP) {
2146 		CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_STOPREQ | RL_CMD_TX_ENB |
2147 		    RL_CMD_RX_ENB);
2148 		if (sc->rl_flags & RL_FLAG_CMDSTOP_WAIT_TXQ) {
2149 			for (i = RL_TIMEOUT; i > 0; i--) {
2150 				if ((CSR_READ_4(sc, RL_TXCFG) &
2151 				    RL_TXCFG_QUEUE_EMPTY) != 0)
2152 					break;
2153 				DELAY(100);
2154 			}
2155 			if (i == 0)
2156 				printf("%s: stopping TXQ timed out!\n",
2157 				    sc->sc_dev.dv_xname);
2158 		}
2159 	} else
2160 		CSR_WRITE_1(sc, RL_COMMAND, 0x00);
2161 	DELAY(1000);
2162 	CSR_WRITE_2(sc, RL_IMR, 0x0000);
2163 	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
2164 
2165 	intr_barrier(sc->sc_ih);
2166 	ifq_barrier(&ifp->if_snd);
2167 
2168 	ifq_clr_oactive(&ifp->if_snd);
2169 	mii_down(&sc->sc_mii);
2170 
2171 	if (sc->rl_head != NULL) {
2172 		m_freem(sc->rl_head);
2173 		sc->rl_head = sc->rl_tail = NULL;
2174 	}
2175 
2176 	/* Free the TX list buffers. */
2177 	for (i = 0; i < RL_TX_QLEN; i++) {
2178 		if (sc->rl_ldata.rl_txq[i].txq_mbuf != NULL) {
2179 			bus_dmamap_unload(sc->sc_dmat,
2180 			    sc->rl_ldata.rl_txq[i].txq_dmamap);
2181 			m_freem(sc->rl_ldata.rl_txq[i].txq_mbuf);
2182 			sc->rl_ldata.rl_txq[i].txq_mbuf = NULL;
2183 		}
2184 	}
2185 
2186 	/* Free the RX list buffers. */
2187 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
2188 		if (sc->rl_ldata.rl_rxsoft[i].rxs_mbuf != NULL) {
2189 			bus_dmamap_unload(sc->sc_dmat,
2190 			    sc->rl_ldata.rl_rxsoft[i].rxs_dmamap);
2191 			m_freem(sc->rl_ldata.rl_rxsoft[i].rxs_mbuf);
2192 			sc->rl_ldata.rl_rxsoft[i].rxs_mbuf = NULL;
2193 		}
2194 	}
2195 }
2196 
2197 void
2198 re_setup_hw_im(struct rl_softc *sc)
2199 {
2200 	KASSERT(sc->rl_flags & RL_FLAG_HWIM);
2201 
2202 	/*
2203 	 * Interrupt moderation
2204 	 *
2205 	 * 0xABCD
2206 	 * A - unknown (maybe TX related)
2207 	 * B - TX timer (unit: 25us)
2208 	 * C - unknown (maybe RX related)
2209 	 * D - RX timer (unit: 25us)
2210 	 *
2211 	 *
2212 	 * re(4)'s interrupt moderation is actually controlled by
2213 	 * two variables, like most other NICs (bge, bnx etc.)
2214 	 * o  timer
2215 	 * o  number of packets [P]
2216 	 *
2217 	 * The logic relationship between these two variables is
2218 	 * similar to other NICs too:
2219 	 * if (timer expire || packets > [P])
2220 	 *     Interrupt is delivered
2221 	 *
2222 	 * Currently we only know how to set 'timer', but not
2223 	 * 'number of packets', which should be ~30, as far as I
2224 	 * tested (sink ~900Kpps, interrupt rate is 30KHz)
2225 	 */
2226 	CSR_WRITE_2(sc, RL_IM,
2227 		    RL_IM_RXTIME(sc->rl_rx_time) |
2228 		    RL_IM_TXTIME(sc->rl_tx_time) |
2229 		    RL_IM_MAGIC);
2230 }
2231 
2232 void
2233 re_disable_hw_im(struct rl_softc *sc)
2234 {
2235 	if (sc->rl_flags & RL_FLAG_HWIM)
2236 		CSR_WRITE_2(sc, RL_IM, 0);
2237 }
2238 
2239 void
2240 re_setup_sim_im(struct rl_softc *sc)
2241 {
2242 	if (sc->sc_hwrev == RL_HWREV_8139CPLUS)
2243 		CSR_WRITE_4(sc, RL_TIMERINT, 0x400); /* XXX */
2244 	else {
2245 		u_int32_t nticks;
2246 
2247 		/*
2248 		 * Datasheet says tick decreases at bus speed,
2249 		 * but it seems the clock runs a little bit
2250 		 * faster, so we do some compensation here.
2251 		 */
2252 		nticks = (sc->rl_sim_time * sc->rl_bus_speed * 8) / 5;
2253 		CSR_WRITE_4(sc, RL_TIMERINT_8169, nticks);
2254 	}
2255 	CSR_WRITE_4(sc, RL_TIMERCNT, 1); /* reload */
2256 	sc->rl_timerintr = 1;
2257 }
2258 
2259 void
2260 re_disable_sim_im(struct rl_softc *sc)
2261 {
2262 	if (sc->sc_hwrev == RL_HWREV_8139CPLUS)
2263 		CSR_WRITE_4(sc, RL_TIMERINT, 0);
2264 	else
2265 		CSR_WRITE_4(sc, RL_TIMERINT_8169, 0);
2266 	sc->rl_timerintr = 0;
2267 }
2268 
2269 void
2270 re_config_imtype(struct rl_softc *sc, int imtype)
2271 {
2272 	switch (imtype) {
2273 	case RL_IMTYPE_HW:
2274 		KASSERT(sc->rl_flags & RL_FLAG_HWIM);
2275 		/* FALLTHROUGH */
2276 	case RL_IMTYPE_NONE:
2277 		sc->rl_intrs = RL_INTRS_CPLUS;
2278 		sc->rl_rx_ack = RL_ISR_RX_OK | RL_ISR_FIFO_OFLOW |
2279 				RL_ISR_RX_OVERRUN;
2280 		sc->rl_tx_ack = RL_ISR_TX_OK;
2281 		break;
2282 
2283 	case RL_IMTYPE_SIM:
2284 		sc->rl_intrs = RL_INTRS_TIMER;
2285 		sc->rl_rx_ack = RL_ISR_TIMEOUT_EXPIRED;
2286 		sc->rl_tx_ack = RL_ISR_TIMEOUT_EXPIRED;
2287 		break;
2288 
2289 	default:
2290 		panic("%s: unknown imtype %d",
2291 		      sc->sc_dev.dv_xname, imtype);
2292 	}
2293 }
2294 
2295 void
2296 re_set_jumbo(struct rl_softc *sc)
2297 {
2298 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG);
2299 	CSR_WRITE_1(sc, RL_CFG3, CSR_READ_1(sc, RL_CFG3) |
2300 	    RL_CFG3_JUMBO_EN0);
2301 
2302 	switch (sc->sc_hwrev) {
2303 	case RL_HWREV_8168DP:
2304 		break;
2305 	case RL_HWREV_8168E:
2306 		CSR_WRITE_1(sc, RL_CFG4, CSR_READ_1(sc, RL_CFG4) |
2307 		    RL_CFG4_8168E_JUMBO_EN1);
2308 		break;
2309 	default:
2310 		CSR_WRITE_1(sc, RL_CFG4, CSR_READ_1(sc, RL_CFG4) |
2311 		    RL_CFG4_JUMBO_EN1);
2312 		break;
2313 	}
2314 
2315 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
2316 }
2317 
2318 void
2319 re_setup_intr(struct rl_softc *sc, int enable_intrs, int imtype)
2320 {
2321 	re_config_imtype(sc, imtype);
2322 
2323 	if (enable_intrs)
2324 		CSR_WRITE_2(sc, RL_IMR, sc->rl_intrs);
2325 	else
2326 		CSR_WRITE_2(sc, RL_IMR, 0);
2327 
2328 	switch (imtype) {
2329 	case RL_IMTYPE_NONE:
2330 		re_disable_sim_im(sc);
2331 		re_disable_hw_im(sc);
2332 		break;
2333 
2334 	case RL_IMTYPE_HW:
2335 		KASSERT(sc->rl_flags & RL_FLAG_HWIM);
2336 		re_disable_sim_im(sc);
2337 		re_setup_hw_im(sc);
2338 		break;
2339 
2340 	case RL_IMTYPE_SIM:
2341 		re_disable_hw_im(sc);
2342 		re_setup_sim_im(sc);
2343 		break;
2344 
2345 	default:
2346 		panic("%s: unknown imtype %d",
2347 		      sc->sc_dev.dv_xname, imtype);
2348 	}
2349 }
2350 
2351 #ifndef SMALL_KERNEL
2352 int
2353 re_wol(struct ifnet *ifp, int enable)
2354 {
2355 	struct rl_softc *sc = ifp->if_softc;
2356 	u_int8_t val;
2357 
2358 	if (enable) {
2359 		if ((CSR_READ_1(sc, sc->rl_cfg1) & RL_CFG1_PME) == 0) {
2360 			printf("%s: power management is disabled, "
2361 			    "cannot do WOL\n", sc->sc_dev.dv_xname);
2362 			return (ENOTSUP);
2363 		}
2364 		if ((CSR_READ_1(sc, sc->rl_cfg2) & RL_CFG2_AUXPWR) == 0)
2365 			printf("%s: no auxiliary power, cannot do WOL from D3 "
2366 			    "(power-off) state\n", sc->sc_dev.dv_xname);
2367 	}
2368 
2369 	re_iff(sc);
2370 
2371 	/* Temporarily enable write to configuration registers. */
2372 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG);
2373 
2374 	/* Always disable all wake events except magic packet. */
2375 	if (enable) {
2376 		val = CSR_READ_1(sc, sc->rl_cfg5);
2377 		val &= ~(RL_CFG5_WOL_UCAST | RL_CFG5_WOL_MCAST |
2378 		    RL_CFG5_WOL_BCAST);
2379 		CSR_WRITE_1(sc, sc->rl_cfg5, val);
2380 
2381 		val = CSR_READ_1(sc, sc->rl_cfg3);
2382 		val |= RL_CFG3_WOL_MAGIC;
2383 		val &= ~RL_CFG3_WOL_LINK;
2384 		CSR_WRITE_1(sc, sc->rl_cfg3, val);
2385 	} else {
2386 		val = CSR_READ_1(sc, sc->rl_cfg5);
2387 		val &= ~(RL_CFG5_WOL_UCAST | RL_CFG5_WOL_MCAST |
2388 		    RL_CFG5_WOL_BCAST);
2389 		CSR_WRITE_1(sc, sc->rl_cfg5, val);
2390 
2391 		val = CSR_READ_1(sc, sc->rl_cfg3);
2392 		val &= ~(RL_CFG3_WOL_MAGIC | RL_CFG3_WOL_LINK);
2393 		CSR_WRITE_1(sc, sc->rl_cfg3, val);
2394 	}
2395 
2396 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
2397 
2398 	return (0);
2399 }
2400 #endif
2401