xref: /netbsd-src/sys/dev/ic/rtl8169.c (revision 404fbe5fb94ca1e054339640cabb2801ce52dd30)
1 /*	$NetBSD: rtl8169.c,v 1.107 2009/01/03 03:43:22 yamt Exp $	*/
2 
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 #include <sys/cdefs.h>
36 __KERNEL_RCSID(0, "$NetBSD: rtl8169.c,v 1.107 2009/01/03 03:43:22 yamt Exp $");
37 /* $FreeBSD: /repoman/r/ncvs/src/sys/dev/re/if_re.c,v 1.20 2004/04/11 20:34:08 ru Exp $ */
38 
39 /*
40  * RealTek 8139C+/8169/8169S/8110S PCI NIC driver
41  *
42  * Written by Bill Paul <wpaul@windriver.com>
43  * Senior Networking Software Engineer
44  * Wind River Systems
45  */
46 
47 /*
48  * This driver is designed to support RealTek's next generation of
49  * 10/100 and 10/100/1000 PCI ethernet controllers. There are currently
50  * four devices in this family: the RTL8139C+, the RTL8169, the RTL8169S
51  * and the RTL8110S.
52  *
53  * The 8139C+ is a 10/100 ethernet chip. It is backwards compatible
54  * with the older 8139 family, however it also supports a special
55  * C+ mode of operation that provides several new performance enhancing
56  * features. These include:
57  *
58  *	o Descriptor based DMA mechanism. Each descriptor represents
59  *	  a single packet fragment. Data buffers may be aligned on
60  *	  any byte boundary.
61  *
62  *	o 64-bit DMA
63  *
64  *	o TCP/IP checksum offload for both RX and TX
65  *
66  *	o High and normal priority transmit DMA rings
67  *
68  *	o VLAN tag insertion and extraction
69  *
70  *	o TCP large send (segmentation offload)
71  *
72  * Like the 8139, the 8139C+ also has a built-in 10/100 PHY. The C+
73  * programming API is fairly straightforward. The RX filtering, EEPROM
74  * access and PHY access is the same as it is on the older 8139 series
75  * chips.
76  *
77  * The 8169 is a 64-bit 10/100/1000 gigabit ethernet MAC. It has almost the
78  * same programming API and feature set as the 8139C+ with the following
79  * differences and additions:
80  *
81  *	o 1000Mbps mode
82  *
83  *	o Jumbo frames
84  *
85  * 	o GMII and TBI ports/registers for interfacing with copper
86  *	  or fiber PHYs
87  *
88  *      o RX and TX DMA rings can have up to 1024 descriptors
89  *        (the 8139C+ allows a maximum of 64)
90  *
91  *	o Slight differences in register layout from the 8139C+
92  *
93  * The TX start and timer interrupt registers are at different locations
94  * on the 8169 than they are on the 8139C+. Also, the status word in the
95  * RX descriptor has a slightly different bit layout. The 8169 does not
96  * have a built-in PHY. Most reference boards use a Marvell 88E1000 'Alaska'
97  * copper gigE PHY.
98  *
99  * The 8169S/8110S 10/100/1000 devices have built-in copper gigE PHYs
100  * (the 'S' stands for 'single-chip'). These devices have the same
101  * programming API as the older 8169, but also have some vendor-specific
102  * registers for the on-board PHY. The 8110S is a LAN-on-motherboard
103  * part designed to be pin-compatible with the RealTek 8100 10/100 chip.
104  *
105  * This driver takes advantage of the RX and TX checksum offload and
106  * VLAN tag insertion/extraction features. It also implements TX
107  * interrupt moderation using the timer interrupt registers, which
108  * significantly reduces TX interrupt load. There is also support
109  * for jumbo frames, however the 8169/8169S/8110S can not transmit
110  * jumbo frames larger than 7.5K, so the max MTU possible with this
111  * driver is 7500 bytes.
112  */
113 
114 #include "bpfilter.h"
115 #include "vlan.h"
116 
117 #include <sys/param.h>
118 #include <sys/endian.h>
119 #include <sys/systm.h>
120 #include <sys/sockio.h>
121 #include <sys/mbuf.h>
122 #include <sys/malloc.h>
123 #include <sys/kernel.h>
124 #include <sys/socket.h>
125 #include <sys/device.h>
126 
127 #include <net/if.h>
128 #include <net/if_arp.h>
129 #include <net/if_dl.h>
130 #include <net/if_ether.h>
131 #include <net/if_media.h>
132 #include <net/if_vlanvar.h>
133 
134 #include <netinet/in_systm.h>	/* XXX for IP_MAXPACKET */
135 #include <netinet/in.h>		/* XXX for IP_MAXPACKET */
136 #include <netinet/ip.h>		/* XXX for IP_MAXPACKET */
137 
138 #if NBPFILTER > 0
139 #include <net/bpf.h>
140 #endif
141 
142 #include <sys/bus.h>
143 
144 #include <dev/mii/mii.h>
145 #include <dev/mii/miivar.h>
146 
147 #include <dev/ic/rtl81x9reg.h>
148 #include <dev/ic/rtl81x9var.h>
149 
150 #include <dev/ic/rtl8169var.h>
151 
152 static inline void re_set_bufaddr(struct re_desc *, bus_addr_t);
153 
154 static int re_newbuf(struct rtk_softc *, int, struct mbuf *);
155 static int re_rx_list_init(struct rtk_softc *);
156 static int re_tx_list_init(struct rtk_softc *);
157 static void re_rxeof(struct rtk_softc *);
158 static void re_txeof(struct rtk_softc *);
159 static void re_tick(void *);
160 static void re_start(struct ifnet *);
161 static int re_ioctl(struct ifnet *, u_long, void *);
162 static int re_init(struct ifnet *);
163 static void re_stop(struct ifnet *, int);
164 static void re_watchdog(struct ifnet *);
165 
166 static int re_enable(struct rtk_softc *);
167 static void re_disable(struct rtk_softc *);
168 
169 static int re_gmii_readreg(struct device *, int, int);
170 static void re_gmii_writereg(struct device *, int, int, int);
171 
172 static int re_miibus_readreg(struct device *, int, int);
173 static void re_miibus_writereg(struct device *, int, int, int);
174 static void re_miibus_statchg(struct device *);
175 
176 static void re_reset(struct rtk_softc *);
177 
178 static inline void
179 re_set_bufaddr(struct re_desc *d, bus_addr_t addr)
180 {
181 
182 	d->re_bufaddr_lo = htole32((uint32_t)addr);
183 	if (sizeof(bus_addr_t) == sizeof(uint64_t))
184 		d->re_bufaddr_hi = htole32((uint64_t)addr >> 32);
185 	else
186 		d->re_bufaddr_hi = 0;
187 }
188 
189 static int
190 re_gmii_readreg(device_t dev, int phy, int reg)
191 {
192 	struct rtk_softc *sc = device_private(dev);
193 	uint32_t rval;
194 	int i;
195 
196 	if (phy != 7)
197 		return 0;
198 
199 	/* Let the rgephy driver read the GMEDIASTAT register */
200 
201 	if (reg == RTK_GMEDIASTAT) {
202 		rval = CSR_READ_1(sc, RTK_GMEDIASTAT);
203 		return rval;
204 	}
205 
206 	CSR_WRITE_4(sc, RTK_PHYAR, reg << 16);
207 	DELAY(1000);
208 
209 	for (i = 0; i < RTK_TIMEOUT; i++) {
210 		rval = CSR_READ_4(sc, RTK_PHYAR);
211 		if (rval & RTK_PHYAR_BUSY)
212 			break;
213 		DELAY(100);
214 	}
215 
216 	if (i == RTK_TIMEOUT) {
217 		printf("%s: PHY read failed\n", device_xname(sc->sc_dev));
218 		return 0;
219 	}
220 
221 	return rval & RTK_PHYAR_PHYDATA;
222 }
223 
224 static void
225 re_gmii_writereg(device_t dev, int phy, int reg, int data)
226 {
227 	struct rtk_softc *sc = device_private(dev);
228 	uint32_t rval;
229 	int i;
230 
231 	CSR_WRITE_4(sc, RTK_PHYAR, (reg << 16) |
232 	    (data & RTK_PHYAR_PHYDATA) | RTK_PHYAR_BUSY);
233 	DELAY(1000);
234 
235 	for (i = 0; i < RTK_TIMEOUT; i++) {
236 		rval = CSR_READ_4(sc, RTK_PHYAR);
237 		if (!(rval & RTK_PHYAR_BUSY))
238 			break;
239 		DELAY(100);
240 	}
241 
242 	if (i == RTK_TIMEOUT) {
243 		printf("%s: PHY write reg %x <- %x failed\n",
244 		    device_xname(sc->sc_dev), reg, data);
245 	}
246 }
247 
248 static int
249 re_miibus_readreg(device_t dev, int phy, int reg)
250 {
251 	struct rtk_softc *sc = device_private(dev);
252 	uint16_t rval = 0;
253 	uint16_t re8139_reg = 0;
254 	int s;
255 
256 	s = splnet();
257 
258 	if ((sc->sc_quirk & RTKQ_8139CPLUS) == 0) {
259 		rval = re_gmii_readreg(dev, phy, reg);
260 		splx(s);
261 		return rval;
262 	}
263 
264 	/* Pretend the internal PHY is only at address 0 */
265 	if (phy) {
266 		splx(s);
267 		return 0;
268 	}
269 	switch (reg) {
270 	case MII_BMCR:
271 		re8139_reg = RTK_BMCR;
272 		break;
273 	case MII_BMSR:
274 		re8139_reg = RTK_BMSR;
275 		break;
276 	case MII_ANAR:
277 		re8139_reg = RTK_ANAR;
278 		break;
279 	case MII_ANER:
280 		re8139_reg = RTK_ANER;
281 		break;
282 	case MII_ANLPAR:
283 		re8139_reg = RTK_LPAR;
284 		break;
285 	case MII_PHYIDR1:
286 	case MII_PHYIDR2:
287 		splx(s);
288 		return 0;
289 	/*
290 	 * Allow the rlphy driver to read the media status
291 	 * register. If we have a link partner which does not
292 	 * support NWAY, this is the register which will tell
293 	 * us the results of parallel detection.
294 	 */
295 	case RTK_MEDIASTAT:
296 		rval = CSR_READ_1(sc, RTK_MEDIASTAT);
297 		splx(s);
298 		return rval;
299 	default:
300 		printf("%s: bad phy register\n", device_xname(sc->sc_dev));
301 		splx(s);
302 		return 0;
303 	}
304 	rval = CSR_READ_2(sc, re8139_reg);
305 	if ((sc->sc_quirk & RTKQ_8139CPLUS) != 0 && re8139_reg == RTK_BMCR) {
306 		/* 8139C+ has different bit layout. */
307 		rval &= ~(BMCR_LOOP | BMCR_ISO);
308 	}
309 	splx(s);
310 	return rval;
311 }
312 
313 static void
314 re_miibus_writereg(device_t dev, int phy, int reg, int data)
315 {
316 	struct rtk_softc *sc = device_private(dev);
317 	uint16_t re8139_reg = 0;
318 	int s;
319 
320 	s = splnet();
321 
322 	if ((sc->sc_quirk & RTKQ_8139CPLUS) == 0) {
323 		re_gmii_writereg(dev, phy, reg, data);
324 		splx(s);
325 		return;
326 	}
327 
328 	/* Pretend the internal PHY is only at address 0 */
329 	if (phy) {
330 		splx(s);
331 		return;
332 	}
333 	switch (reg) {
334 	case MII_BMCR:
335 		re8139_reg = RTK_BMCR;
336 		if ((sc->sc_quirk & RTKQ_8139CPLUS) != 0) {
337 			/* 8139C+ has different bit layout. */
338 			data &= ~(BMCR_LOOP | BMCR_ISO);
339 		}
340 		break;
341 	case MII_BMSR:
342 		re8139_reg = RTK_BMSR;
343 		break;
344 	case MII_ANAR:
345 		re8139_reg = RTK_ANAR;
346 		break;
347 	case MII_ANER:
348 		re8139_reg = RTK_ANER;
349 		break;
350 	case MII_ANLPAR:
351 		re8139_reg = RTK_LPAR;
352 		break;
353 	case MII_PHYIDR1:
354 	case MII_PHYIDR2:
355 		splx(s);
356 		return;
357 		break;
358 	default:
359 		printf("%s: bad phy register\n", device_xname(sc->sc_dev));
360 		splx(s);
361 		return;
362 	}
363 	CSR_WRITE_2(sc, re8139_reg, data);
364 	splx(s);
365 	return;
366 }
367 
368 static void
369 re_miibus_statchg(device_t dev)
370 {
371 
372 	return;
373 }
374 
375 static void
376 re_reset(struct rtk_softc *sc)
377 {
378 	int i;
379 
380 	CSR_WRITE_1(sc, RTK_COMMAND, RTK_CMD_RESET);
381 
382 	for (i = 0; i < RTK_TIMEOUT; i++) {
383 		DELAY(10);
384 		if ((CSR_READ_1(sc, RTK_COMMAND) & RTK_CMD_RESET) == 0)
385 			break;
386 	}
387 	if (i == RTK_TIMEOUT)
388 		printf("%s: reset never completed!\n",
389 		    device_xname(sc->sc_dev));
390 
391 	/*
392 	 * NB: Realtek-supplied Linux driver does this only for
393 	 * MCFG_METHOD_2, which corresponds to sc->sc_rev == 3.
394 	 */
395 	if (1) /* XXX check softc flag for 8169s version */
396 		CSR_WRITE_1(sc, RTK_LDPS, 1);
397 
398 }
399 
400 /*
401  * The following routine is designed to test for a defect on some
402  * 32-bit 8169 cards. Some of these NICs have the REQ64# and ACK64#
403  * lines connected to the bus, however for a 32-bit only card, they
404  * should be pulled high. The result of this defect is that the
405  * NIC will not work right if you plug it into a 64-bit slot: DMA
406  * operations will be done with 64-bit transfers, which will fail
407  * because the 64-bit data lines aren't connected.
408  *
409  * There's no way to work around this (short of talking a soldering
410  * iron to the board), however we can detect it. The method we use
411  * here is to put the NIC into digital loopback mode, set the receiver
412  * to promiscuous mode, and then try to send a frame. We then compare
413  * the frame data we sent to what was received. If the data matches,
414  * then the NIC is working correctly, otherwise we know the user has
415  * a defective NIC which has been mistakenly plugged into a 64-bit PCI
416  * slot. In the latter case, there's no way the NIC can work correctly,
417  * so we print out a message on the console and abort the device attach.
418  */
419 
420 int
421 re_diag(struct rtk_softc *sc)
422 {
423 	struct ifnet *ifp = &sc->ethercom.ec_if;
424 	struct mbuf *m0;
425 	struct ether_header *eh;
426 	struct re_rxsoft *rxs;
427 	struct re_desc *cur_rx;
428 	bus_dmamap_t dmamap;
429 	uint16_t status;
430 	uint32_t rxstat;
431 	int total_len, i, s, error = 0;
432 	static const uint8_t dst[] = { 0x00, 'h', 'e', 'l', 'l', 'o' };
433 	static const uint8_t src[] = { 0x00, 'w', 'o', 'r', 'l', 'd' };
434 
435 	/* Allocate a single mbuf */
436 
437 	MGETHDR(m0, M_DONTWAIT, MT_DATA);
438 	if (m0 == NULL)
439 		return ENOBUFS;
440 
441 	/*
442 	 * Initialize the NIC in test mode. This sets the chip up
443 	 * so that it can send and receive frames, but performs the
444 	 * following special functions:
445 	 * - Puts receiver in promiscuous mode
446 	 * - Enables digital loopback mode
447 	 * - Leaves interrupts turned off
448 	 */
449 
450 	ifp->if_flags |= IFF_PROMISC;
451 	sc->re_testmode = 1;
452 	re_init(ifp);
453 	re_stop(ifp, 0);
454 	DELAY(100000);
455 	re_init(ifp);
456 
457 	/* Put some data in the mbuf */
458 
459 	eh = mtod(m0, struct ether_header *);
460 	memcpy(eh->ether_dhost, (char *)&dst, ETHER_ADDR_LEN);
461 	memcpy(eh->ether_shost, (char *)&src, ETHER_ADDR_LEN);
462 	eh->ether_type = htons(ETHERTYPE_IP);
463 	m0->m_pkthdr.len = m0->m_len = ETHER_MIN_LEN - ETHER_CRC_LEN;
464 
465 	/*
466 	 * Queue the packet, start transmission.
467 	 */
468 
469 	CSR_WRITE_2(sc, RTK_ISR, 0xFFFF);
470 	s = splnet();
471 	IF_ENQUEUE(&ifp->if_snd, m0);
472 	re_start(ifp);
473 	splx(s);
474 	m0 = NULL;
475 
476 	/* Wait for it to propagate through the chip */
477 
478 	DELAY(100000);
479 	for (i = 0; i < RTK_TIMEOUT; i++) {
480 		status = CSR_READ_2(sc, RTK_ISR);
481 		if ((status & (RTK_ISR_TIMEOUT_EXPIRED | RTK_ISR_RX_OK)) ==
482 		    (RTK_ISR_TIMEOUT_EXPIRED | RTK_ISR_RX_OK))
483 			break;
484 		DELAY(10);
485 	}
486 	if (i == RTK_TIMEOUT) {
487 		aprint_error_dev(sc->sc_dev,
488 		    "diagnostic failed, failed to receive packet "
489 		    "in loopback mode\n");
490 		error = EIO;
491 		goto done;
492 	}
493 
494 	/*
495 	 * The packet should have been dumped into the first
496 	 * entry in the RX DMA ring. Grab it from there.
497 	 */
498 
499 	rxs = &sc->re_ldata.re_rxsoft[0];
500 	dmamap = rxs->rxs_dmamap;
501 	bus_dmamap_sync(sc->sc_dmat, dmamap, 0, dmamap->dm_mapsize,
502 	    BUS_DMASYNC_POSTREAD);
503 	bus_dmamap_unload(sc->sc_dmat, dmamap);
504 
505 	m0 = rxs->rxs_mbuf;
506 	rxs->rxs_mbuf = NULL;
507 	eh = mtod(m0, struct ether_header *);
508 
509 	RE_RXDESCSYNC(sc, 0, BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
510 	cur_rx = &sc->re_ldata.re_rx_list[0];
511 	rxstat = le32toh(cur_rx->re_cmdstat);
512 	total_len = rxstat & sc->re_rxlenmask;
513 
514 	if (total_len != ETHER_MIN_LEN) {
515 		aprint_error_dev(sc->sc_dev,
516 		    "diagnostic failed, received short packet\n");
517 		error = EIO;
518 		goto done;
519 	}
520 
521 	/* Test that the received packet data matches what we sent. */
522 
523 	if (memcmp((char *)&eh->ether_dhost, (char *)&dst, ETHER_ADDR_LEN) ||
524 	    memcmp((char *)&eh->ether_shost, (char *)&src, ETHER_ADDR_LEN) ||
525 	    ntohs(eh->ether_type) != ETHERTYPE_IP) {
526 		aprint_error_dev(sc->sc_dev, "WARNING, DMA FAILURE!\n"
527 		    "expected TX data: %s/%s/0x%x\n"
528 		    "received RX data: %s/%s/0x%x\n"
529 		    "You may have a defective 32-bit NIC plugged "
530 		    "into a 64-bit PCI slot.\n"
531 		    "Please re-install the NIC in a 32-bit slot "
532 		    "for proper operation.\n"
533 		    "Read the re(4) man page for more details.\n" ,
534 		    ether_sprintf(dst),  ether_sprintf(src), ETHERTYPE_IP,
535 		    ether_sprintf(eh->ether_dhost),
536 		    ether_sprintf(eh->ether_shost), ntohs(eh->ether_type));
537 		error = EIO;
538 	}
539 
540  done:
541 	/* Turn interface off, release resources */
542 
543 	sc->re_testmode = 0;
544 	ifp->if_flags &= ~IFF_PROMISC;
545 	re_stop(ifp, 0);
546 	if (m0 != NULL)
547 		m_freem(m0);
548 
549 	return error;
550 }
551 
552 
553 /*
554  * Attach the interface. Allocate softc structures, do ifmedia
555  * setup and ethernet/BPF attach.
556  */
557 void
558 re_attach(struct rtk_softc *sc)
559 {
560 	uint8_t eaddr[ETHER_ADDR_LEN];
561 	uint16_t val;
562 	struct ifnet *ifp;
563 	int error = 0, i, addr_len;
564 
565 	/* Reset the adapter. */
566 	re_reset(sc);
567 
568 	if ((sc->sc_quirk & RTKQ_8139CPLUS) == 0) {
569 		uint32_t hwrev;
570 
571 		/* Revision of 8169/8169S/8110s in bits 30..26, 23 */
572 		hwrev = CSR_READ_4(sc, RTK_TXCFG) & RTK_TXCFG_HWREV;
573 		/* These rev numbers are taken from Realtek's driver */
574 		switch (hwrev) {
575 		case RTK_HWREV_8169:
576 			/* XXX not in the Realtek driver */
577 			sc->sc_rev = 1;
578 			sc->sc_quirk |= RTKQ_8169NONS;
579 			break;
580 		case RTK_HWREV_8169S:
581 		case RTK_HWREV_8110S:
582 			sc->sc_rev = 3;
583 			break;
584 		case RTK_HWREV_8169_8110SB:
585 			sc->sc_rev = 4;
586 			break;
587 		case RTK_HWREV_8169_8110SC:
588 			sc->sc_rev = 5;
589 			break;
590 		case RTK_HWREV_8101E:
591 			sc->sc_rev = 11;
592 			break;
593 		case RTK_HWREV_8168_SPIN1:
594 			sc->sc_rev = 21;
595 			break;
596 		case RTK_HWREV_8168_SPIN2:
597 			sc->sc_rev = 22;
598 			break;
599 		case RTK_HWREV_8168_SPIN3:
600 			sc->sc_rev = 23;
601 			break;
602 		case RTK_HWREV_8168C:
603 		case RTK_HWREV_8168C_SPIN2:
604 			sc->sc_rev = 24;
605 			break;
606 		case RTK_HWREV_8102E:
607 		case RTK_HWREV_8102EL:
608 			sc->sc_rev = 25;
609 			break;
610 		case RTK_HWREV_8100E:
611 		case RTK_HWREV_8100E_SPIN2:
612 			/* XXX not in the Realtek driver */
613 			sc->sc_rev = 0;
614 			break;
615 		default:
616 			aprint_normal_dev(sc->sc_dev,
617 			    "Unknown revision (0x%08x)\n", hwrev);
618 			sc->sc_rev = 0;
619 		}
620 
621 		/* Set RX length mask */
622 		sc->re_rxlenmask = RE_RDESC_STAT_GFRAGLEN;
623 		sc->re_ldata.re_tx_desc_cnt = RE_TX_DESC_CNT_8169;
624 	} else {
625 		/* Set RX length mask */
626 		sc->re_rxlenmask = RE_RDESC_STAT_FRAGLEN;
627 		sc->re_ldata.re_tx_desc_cnt = RE_TX_DESC_CNT_8139;
628 	}
629 
630 	if (sc->sc_rev == 24 || sc->sc_rev == 25) {
631 		/*
632 		 * Get station address from ID registers.
633 		 */
634 		for (i = 0; i < ETHER_ADDR_LEN; i++)
635 			eaddr[i] = CSR_READ_1(sc, RTK_IDR0 + i);
636 	} else {
637 		/*
638 		 * Get station address from the EEPROM.
639 		 */
640 		if (rtk_read_eeprom(sc, RTK_EE_ID, RTK_EEADDR_LEN1) == 0x8129)
641 			addr_len = RTK_EEADDR_LEN1;
642 		else
643 			addr_len = RTK_EEADDR_LEN0;
644 
645 		/*
646 		 * Get station address from the EEPROM.
647 		 */
648 		for (i = 0; i < ETHER_ADDR_LEN / 2; i++) {
649 			val = rtk_read_eeprom(sc, RTK_EE_EADDR0 + i, addr_len);
650 			eaddr[(i * 2) + 0] = val & 0xff;
651 			eaddr[(i * 2) + 1] = val >> 8;
652 		}
653 	}
654 
655 	aprint_normal_dev(sc->sc_dev, "Ethernet address %s\n",
656 	    ether_sprintf(eaddr));
657 
658 	if (sc->re_ldata.re_tx_desc_cnt >
659 	    PAGE_SIZE / sizeof(struct re_desc)) {
660 		sc->re_ldata.re_tx_desc_cnt =
661 		    PAGE_SIZE / sizeof(struct re_desc);
662 	}
663 
664 	aprint_verbose_dev(sc->sc_dev, "using %d tx descriptors\n",
665 	    sc->re_ldata.re_tx_desc_cnt);
666 	KASSERT(RE_NEXT_TX_DESC(sc, RE_TX_DESC_CNT(sc) - 1) == 0);
667 
668 	/* Allocate DMA'able memory for the TX ring */
669 	if ((error = bus_dmamem_alloc(sc->sc_dmat, RE_TX_LIST_SZ(sc),
670 	    RE_RING_ALIGN, 0, &sc->re_ldata.re_tx_listseg, 1,
671 	    &sc->re_ldata.re_tx_listnseg, BUS_DMA_NOWAIT)) != 0) {
672 		aprint_error_dev(sc->sc_dev,
673 		    "can't allocate tx listseg, error = %d\n", error);
674 		goto fail_0;
675 	}
676 
677 	/* Load the map for the TX ring. */
678 	if ((error = bus_dmamem_map(sc->sc_dmat, &sc->re_ldata.re_tx_listseg,
679 	    sc->re_ldata.re_tx_listnseg, RE_TX_LIST_SZ(sc),
680 	    (void **)&sc->re_ldata.re_tx_list,
681 	    BUS_DMA_COHERENT | BUS_DMA_NOWAIT)) != 0) {
682 		aprint_error_dev(sc->sc_dev,
683 		    "can't map tx list, error = %d\n", error);
684 	  	goto fail_1;
685 	}
686 	memset(sc->re_ldata.re_tx_list, 0, RE_TX_LIST_SZ(sc));
687 
688 	if ((error = bus_dmamap_create(sc->sc_dmat, RE_TX_LIST_SZ(sc), 1,
689 	    RE_TX_LIST_SZ(sc), 0, 0,
690 	    &sc->re_ldata.re_tx_list_map)) != 0) {
691 		aprint_error_dev(sc->sc_dev,
692 		    "can't create tx list map, error = %d\n", error);
693 		goto fail_2;
694 	}
695 
696 
697 	if ((error = bus_dmamap_load(sc->sc_dmat,
698 	    sc->re_ldata.re_tx_list_map, sc->re_ldata.re_tx_list,
699 	    RE_TX_LIST_SZ(sc), NULL, BUS_DMA_NOWAIT)) != 0) {
700 		aprint_error_dev(sc->sc_dev,
701 		    "can't load tx list, error = %d\n", error);
702 		goto fail_3;
703 	}
704 
705 	/* Create DMA maps for TX buffers */
706 	for (i = 0; i < RE_TX_QLEN; i++) {
707 		error = bus_dmamap_create(sc->sc_dmat,
708 		    round_page(IP_MAXPACKET),
709 		    RE_TX_DESC_CNT(sc), RE_TDESC_CMD_FRAGLEN,
710 		    0, 0, &sc->re_ldata.re_txq[i].txq_dmamap);
711 		if (error) {
712 			aprint_error_dev(sc->sc_dev,
713 			    "can't create DMA map for TX\n");
714 			goto fail_4;
715 		}
716 	}
717 
718 	/* Allocate DMA'able memory for the RX ring */
719 	/* XXX see also a comment about RE_RX_DMAMEM_SZ in rtl81x9var.h */
720 	if ((error = bus_dmamem_alloc(sc->sc_dmat,
721 	    RE_RX_DMAMEM_SZ, RE_RING_ALIGN, 0, &sc->re_ldata.re_rx_listseg, 1,
722 	    &sc->re_ldata.re_rx_listnseg, BUS_DMA_NOWAIT)) != 0) {
723 		aprint_error_dev(sc->sc_dev,
724 		    "can't allocate rx listseg, error = %d\n", error);
725 		goto fail_4;
726 	}
727 
728 	/* Load the map for the RX ring. */
729 	if ((error = bus_dmamem_map(sc->sc_dmat, &sc->re_ldata.re_rx_listseg,
730 	    sc->re_ldata.re_rx_listnseg, RE_RX_DMAMEM_SZ,
731 	    (void **)&sc->re_ldata.re_rx_list,
732 	    BUS_DMA_COHERENT | BUS_DMA_NOWAIT)) != 0) {
733 		aprint_error_dev(sc->sc_dev,
734 		    "can't map rx list, error = %d\n", error);
735 		goto fail_5;
736 	}
737 	memset(sc->re_ldata.re_rx_list, 0, RE_RX_DMAMEM_SZ);
738 
739 	if ((error = bus_dmamap_create(sc->sc_dmat,
740 	    RE_RX_DMAMEM_SZ, 1, RE_RX_DMAMEM_SZ, 0, 0,
741 	    &sc->re_ldata.re_rx_list_map)) != 0) {
742 		aprint_error_dev(sc->sc_dev,
743 		    "can't create rx list map, error = %d\n", error);
744 		goto fail_6;
745 	}
746 
747 	if ((error = bus_dmamap_load(sc->sc_dmat,
748 	    sc->re_ldata.re_rx_list_map, sc->re_ldata.re_rx_list,
749 	    RE_RX_DMAMEM_SZ, NULL, BUS_DMA_NOWAIT)) != 0) {
750 		aprint_error_dev(sc->sc_dev,
751 		    "can't load rx list, error = %d\n", error);
752 		goto fail_7;
753 	}
754 
755 	/* Create DMA maps for RX buffers */
756 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
757 		error = bus_dmamap_create(sc->sc_dmat, MCLBYTES, 1, MCLBYTES,
758 		    0, 0, &sc->re_ldata.re_rxsoft[i].rxs_dmamap);
759 		if (error) {
760 			aprint_error_dev(sc->sc_dev,
761 			    "can't create DMA map for RX\n");
762 			goto fail_8;
763 		}
764 	}
765 
766 	/*
767 	 * Record interface as attached. From here, we should not fail.
768 	 */
769 	sc->sc_flags |= RTK_ATTACHED;
770 
771 	ifp = &sc->ethercom.ec_if;
772 	ifp->if_softc = sc;
773 	strlcpy(ifp->if_xname, device_xname(sc->sc_dev), IFNAMSIZ);
774 	ifp->if_mtu = ETHERMTU;
775 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
776 	ifp->if_ioctl = re_ioctl;
777 	sc->ethercom.ec_capabilities |=
778 	    ETHERCAP_VLAN_MTU | ETHERCAP_VLAN_HWTAGGING;
779 	ifp->if_start = re_start;
780 	ifp->if_stop = re_stop;
781 
782 	/*
783 	 * IFCAP_CSUM_IPv4_Tx on re(4) is broken for small packets,
784 	 * so we have a workaround to handle the bug by padding
785 	 * such packets manually.
786 	 */
787 	ifp->if_capabilities |=
788 	    IFCAP_CSUM_IPv4_Tx | IFCAP_CSUM_IPv4_Rx |
789 	    IFCAP_CSUM_TCPv4_Tx | IFCAP_CSUM_TCPv4_Rx |
790 	    IFCAP_CSUM_UDPv4_Tx | IFCAP_CSUM_UDPv4_Rx |
791 	    IFCAP_TSOv4;
792 	ifp->if_watchdog = re_watchdog;
793 	ifp->if_init = re_init;
794 	ifp->if_snd.ifq_maxlen = RE_IFQ_MAXLEN;
795 	ifp->if_capenable = ifp->if_capabilities;
796 	IFQ_SET_READY(&ifp->if_snd);
797 
798 	callout_init(&sc->rtk_tick_ch, 0);
799 
800 	/* Do MII setup */
801 	sc->mii.mii_ifp = ifp;
802 	sc->mii.mii_readreg = re_miibus_readreg;
803 	sc->mii.mii_writereg = re_miibus_writereg;
804 	sc->mii.mii_statchg = re_miibus_statchg;
805 	sc->ethercom.ec_mii = &sc->mii;
806 	ifmedia_init(&sc->mii.mii_media, IFM_IMASK, ether_mediachange,
807 	    ether_mediastatus);
808 	mii_attach(sc->sc_dev, &sc->mii, 0xffffffff, MII_PHY_ANY,
809 	    MII_OFFSET_ANY, 0);
810 	ifmedia_set(&sc->mii.mii_media, IFM_ETHER | IFM_AUTO);
811 
812 	/*
813 	 * Call MI attach routine.
814 	 */
815 	if_attach(ifp);
816 	ether_ifattach(ifp, eaddr);
817 
818 	return;
819 
820  fail_8:
821 	/* Destroy DMA maps for RX buffers. */
822 	for (i = 0; i < RE_RX_DESC_CNT; i++)
823 		if (sc->re_ldata.re_rxsoft[i].rxs_dmamap != NULL)
824 			bus_dmamap_destroy(sc->sc_dmat,
825 			    sc->re_ldata.re_rxsoft[i].rxs_dmamap);
826 
827 	/* Free DMA'able memory for the RX ring. */
828 	bus_dmamap_unload(sc->sc_dmat, sc->re_ldata.re_rx_list_map);
829  fail_7:
830 	bus_dmamap_destroy(sc->sc_dmat, sc->re_ldata.re_rx_list_map);
831  fail_6:
832 	bus_dmamem_unmap(sc->sc_dmat,
833 	    (void *)sc->re_ldata.re_rx_list, RE_RX_DMAMEM_SZ);
834  fail_5:
835 	bus_dmamem_free(sc->sc_dmat,
836 	    &sc->re_ldata.re_rx_listseg, sc->re_ldata.re_rx_listnseg);
837 
838  fail_4:
839 	/* Destroy DMA maps for TX buffers. */
840 	for (i = 0; i < RE_TX_QLEN; i++)
841 		if (sc->re_ldata.re_txq[i].txq_dmamap != NULL)
842 			bus_dmamap_destroy(sc->sc_dmat,
843 			    sc->re_ldata.re_txq[i].txq_dmamap);
844 
845 	/* Free DMA'able memory for the TX ring. */
846 	bus_dmamap_unload(sc->sc_dmat, sc->re_ldata.re_tx_list_map);
847  fail_3:
848 	bus_dmamap_destroy(sc->sc_dmat, sc->re_ldata.re_tx_list_map);
849  fail_2:
850 	bus_dmamem_unmap(sc->sc_dmat,
851 	    (void *)sc->re_ldata.re_tx_list, RE_TX_LIST_SZ(sc));
852  fail_1:
853 	bus_dmamem_free(sc->sc_dmat,
854 	    &sc->re_ldata.re_tx_listseg, sc->re_ldata.re_tx_listnseg);
855  fail_0:
856 	return;
857 }
858 
859 
860 /*
861  * re_activate:
862  *     Handle device activation/deactivation requests.
863  */
864 int
865 re_activate(device_t self, enum devact act)
866 {
867 	struct rtk_softc *sc = device_private(self);
868 	int s, error = 0;
869 
870 	s = splnet();
871 	switch (act) {
872 	case DVACT_ACTIVATE:
873 		error = EOPNOTSUPP;
874 		break;
875 	case DVACT_DEACTIVATE:
876 		mii_activate(&sc->mii, act, MII_PHY_ANY, MII_OFFSET_ANY);
877 		if_deactivate(&sc->ethercom.ec_if);
878 		break;
879 	}
880 	splx(s);
881 
882 	return error;
883 }
884 
885 /*
886  * re_detach:
887  *     Detach a rtk interface.
888  */
889 int
890 re_detach(struct rtk_softc *sc)
891 {
892 	struct ifnet *ifp = &sc->ethercom.ec_if;
893 	int i;
894 
895 	/*
896 	 * Succeed now if there isn't any work to do.
897 	 */
898 	if ((sc->sc_flags & RTK_ATTACHED) == 0)
899 		return 0;
900 
901 	/* Unhook our tick handler. */
902 	callout_stop(&sc->rtk_tick_ch);
903 
904 	/* Detach all PHYs. */
905 	mii_detach(&sc->mii, MII_PHY_ANY, MII_OFFSET_ANY);
906 
907 	/* Delete all remaining media. */
908 	ifmedia_delete_instance(&sc->mii.mii_media, IFM_INST_ANY);
909 
910 	ether_ifdetach(ifp);
911 	if_detach(ifp);
912 
913 	/* Destroy DMA maps for RX buffers. */
914 	for (i = 0; i < RE_RX_DESC_CNT; i++)
915 		if (sc->re_ldata.re_rxsoft[i].rxs_dmamap != NULL)
916 			bus_dmamap_destroy(sc->sc_dmat,
917 			    sc->re_ldata.re_rxsoft[i].rxs_dmamap);
918 
919 	/* Free DMA'able memory for the RX ring. */
920 	bus_dmamap_unload(sc->sc_dmat, sc->re_ldata.re_rx_list_map);
921 	bus_dmamap_destroy(sc->sc_dmat, sc->re_ldata.re_rx_list_map);
922 	bus_dmamem_unmap(sc->sc_dmat,
923 	    (void *)sc->re_ldata.re_rx_list, RE_RX_DMAMEM_SZ);
924 	bus_dmamem_free(sc->sc_dmat,
925 	    &sc->re_ldata.re_rx_listseg, sc->re_ldata.re_rx_listnseg);
926 
927 	/* Destroy DMA maps for TX buffers. */
928 	for (i = 0; i < RE_TX_QLEN; i++)
929 		if (sc->re_ldata.re_txq[i].txq_dmamap != NULL)
930 			bus_dmamap_destroy(sc->sc_dmat,
931 			    sc->re_ldata.re_txq[i].txq_dmamap);
932 
933 	/* Free DMA'able memory for the TX ring. */
934 	bus_dmamap_unload(sc->sc_dmat, sc->re_ldata.re_tx_list_map);
935 	bus_dmamap_destroy(sc->sc_dmat, sc->re_ldata.re_tx_list_map);
936 	bus_dmamem_unmap(sc->sc_dmat,
937 	    (void *)sc->re_ldata.re_tx_list, RE_TX_LIST_SZ(sc));
938 	bus_dmamem_free(sc->sc_dmat,
939 	    &sc->re_ldata.re_tx_listseg, sc->re_ldata.re_tx_listnseg);
940 
941 	return 0;
942 }
943 
944 /*
945  * re_enable:
946  *     Enable the RTL81X9 chip.
947  */
948 static int
949 re_enable(struct rtk_softc *sc)
950 {
951 
952 	if (RTK_IS_ENABLED(sc) == 0 && sc->sc_enable != NULL) {
953 		if ((*sc->sc_enable)(sc) != 0) {
954 			printf("%s: device enable failed\n",
955 			    device_xname(sc->sc_dev));
956 			return EIO;
957 		}
958 		sc->sc_flags |= RTK_ENABLED;
959 	}
960 	return 0;
961 }
962 
963 /*
964  * re_disable:
965  *     Disable the RTL81X9 chip.
966  */
967 static void
968 re_disable(struct rtk_softc *sc)
969 {
970 
971 	if (RTK_IS_ENABLED(sc) && sc->sc_disable != NULL) {
972 		(*sc->sc_disable)(sc);
973 		sc->sc_flags &= ~RTK_ENABLED;
974 	}
975 }
976 
977 static int
978 re_newbuf(struct rtk_softc *sc, int idx, struct mbuf *m)
979 {
980 	struct mbuf *n = NULL;
981 	bus_dmamap_t map;
982 	struct re_desc *d;
983 	struct re_rxsoft *rxs;
984 	uint32_t cmdstat;
985 	int error;
986 
987 	if (m == NULL) {
988 		MGETHDR(n, M_DONTWAIT, MT_DATA);
989 		if (n == NULL)
990 			return ENOBUFS;
991 
992 		MCLGET(n, M_DONTWAIT);
993 		if ((n->m_flags & M_EXT) == 0) {
994 			m_freem(n);
995 			return ENOBUFS;
996 		}
997 		m = n;
998 	} else
999 		m->m_data = m->m_ext.ext_buf;
1000 
1001 	/*
1002 	 * Initialize mbuf length fields and fixup
1003 	 * alignment so that the frame payload is
1004 	 * longword aligned.
1005 	 */
1006 	m->m_len = m->m_pkthdr.len = MCLBYTES - RE_ETHER_ALIGN;
1007 	m->m_data += RE_ETHER_ALIGN;
1008 
1009 	rxs = &sc->re_ldata.re_rxsoft[idx];
1010 	map = rxs->rxs_dmamap;
1011 	error = bus_dmamap_load_mbuf(sc->sc_dmat, map, m,
1012 	    BUS_DMA_READ|BUS_DMA_NOWAIT);
1013 
1014 	if (error)
1015 		goto out;
1016 
1017 	bus_dmamap_sync(sc->sc_dmat, map, 0, map->dm_mapsize,
1018 	    BUS_DMASYNC_PREREAD);
1019 
1020 	d = &sc->re_ldata.re_rx_list[idx];
1021 #ifdef DIAGNOSTIC
1022 	RE_RXDESCSYNC(sc, idx, BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1023 	cmdstat = le32toh(d->re_cmdstat);
1024 	RE_RXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD);
1025 	if (cmdstat & RE_RDESC_STAT_OWN) {
1026 		panic("%s: tried to map busy RX descriptor",
1027 		    device_xname(sc->sc_dev));
1028 	}
1029 #endif
1030 
1031 	rxs->rxs_mbuf = m;
1032 
1033 	d->re_vlanctl = 0;
1034 	cmdstat = map->dm_segs[0].ds_len;
1035 	if (idx == (RE_RX_DESC_CNT - 1))
1036 		cmdstat |= RE_RDESC_CMD_EOR;
1037 	re_set_bufaddr(d, map->dm_segs[0].ds_addr);
1038 	d->re_cmdstat = htole32(cmdstat);
1039 	RE_RXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1040 	cmdstat |= RE_RDESC_CMD_OWN;
1041 	d->re_cmdstat = htole32(cmdstat);
1042 	RE_RXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1043 
1044 	return 0;
1045  out:
1046 	if (n != NULL)
1047 		m_freem(n);
1048 	return ENOMEM;
1049 }
1050 
1051 static int
1052 re_tx_list_init(struct rtk_softc *sc)
1053 {
1054 	int i;
1055 
1056 	memset(sc->re_ldata.re_tx_list, 0, RE_TX_LIST_SZ(sc));
1057 	for (i = 0; i < RE_TX_QLEN; i++) {
1058 		sc->re_ldata.re_txq[i].txq_mbuf = NULL;
1059 	}
1060 
1061 	bus_dmamap_sync(sc->sc_dmat,
1062 	    sc->re_ldata.re_tx_list_map, 0,
1063 	    sc->re_ldata.re_tx_list_map->dm_mapsize,
1064 	    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1065 	sc->re_ldata.re_txq_prodidx = 0;
1066 	sc->re_ldata.re_txq_considx = 0;
1067 	sc->re_ldata.re_txq_free = RE_TX_QLEN;
1068 	sc->re_ldata.re_tx_free = RE_TX_DESC_CNT(sc);
1069 	sc->re_ldata.re_tx_nextfree = 0;
1070 
1071 	return 0;
1072 }
1073 
1074 static int
1075 re_rx_list_init(struct rtk_softc *sc)
1076 {
1077 	int i;
1078 
1079 	memset(sc->re_ldata.re_rx_list, 0, RE_RX_LIST_SZ);
1080 
1081 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
1082 		if (re_newbuf(sc, i, NULL) == ENOBUFS)
1083 			return ENOBUFS;
1084 	}
1085 
1086 	sc->re_ldata.re_rx_prodidx = 0;
1087 	sc->re_head = sc->re_tail = NULL;
1088 
1089 	return 0;
1090 }
1091 
1092 /*
1093  * RX handler for C+ and 8169. For the gigE chips, we support
1094  * the reception of jumbo frames that have been fragmented
1095  * across multiple 2K mbuf cluster buffers.
1096  */
1097 static void
1098 re_rxeof(struct rtk_softc *sc)
1099 {
1100 	struct mbuf *m;
1101 	struct ifnet *ifp;
1102 	int i, total_len;
1103 	struct re_desc *cur_rx;
1104 	struct re_rxsoft *rxs;
1105 	uint32_t rxstat, rxvlan;
1106 
1107 	ifp = &sc->ethercom.ec_if;
1108 
1109 	for (i = sc->re_ldata.re_rx_prodidx;; i = RE_NEXT_RX_DESC(sc, i)) {
1110 		cur_rx = &sc->re_ldata.re_rx_list[i];
1111 		RE_RXDESCSYNC(sc, i,
1112 		    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1113 		rxstat = le32toh(cur_rx->re_cmdstat);
1114 		rxvlan = le32toh(cur_rx->re_vlanctl);
1115 		RE_RXDESCSYNC(sc, i, BUS_DMASYNC_PREREAD);
1116 		if ((rxstat & RE_RDESC_STAT_OWN) != 0) {
1117 			break;
1118 		}
1119 		total_len = rxstat & sc->re_rxlenmask;
1120 		rxs = &sc->re_ldata.re_rxsoft[i];
1121 		m = rxs->rxs_mbuf;
1122 
1123 		/* Invalidate the RX mbuf and unload its map */
1124 
1125 		bus_dmamap_sync(sc->sc_dmat,
1126 		    rxs->rxs_dmamap, 0, rxs->rxs_dmamap->dm_mapsize,
1127 		    BUS_DMASYNC_POSTREAD);
1128 		bus_dmamap_unload(sc->sc_dmat, rxs->rxs_dmamap);
1129 
1130 		if ((rxstat & RE_RDESC_STAT_EOF) == 0) {
1131 			m->m_len = MCLBYTES - RE_ETHER_ALIGN;
1132 			if (sc->re_head == NULL)
1133 				sc->re_head = sc->re_tail = m;
1134 			else {
1135 				m->m_flags &= ~M_PKTHDR;
1136 				sc->re_tail->m_next = m;
1137 				sc->re_tail = m;
1138 			}
1139 			re_newbuf(sc, i, NULL);
1140 			continue;
1141 		}
1142 
1143 		/*
1144 		 * NOTE: for the 8139C+, the frame length field
1145 		 * is always 12 bits in size, but for the gigE chips,
1146 		 * it is 13 bits (since the max RX frame length is 16K).
1147 		 * Unfortunately, all 32 bits in the status word
1148 		 * were already used, so to make room for the extra
1149 		 * length bit, RealTek took out the 'frame alignment
1150 		 * error' bit and shifted the other status bits
1151 		 * over one slot. The OWN, EOR, FS and LS bits are
1152 		 * still in the same places. We have already extracted
1153 		 * the frame length and checked the OWN bit, so rather
1154 		 * than using an alternate bit mapping, we shift the
1155 		 * status bits one space to the right so we can evaluate
1156 		 * them using the 8169 status as though it was in the
1157 		 * same format as that of the 8139C+.
1158 		 */
1159 		if ((sc->sc_quirk & RTKQ_8139CPLUS) == 0)
1160 			rxstat >>= 1;
1161 
1162 		if (__predict_false((rxstat & RE_RDESC_STAT_RXERRSUM) != 0)) {
1163 #ifdef RE_DEBUG
1164 			printf("%s: RX error (rxstat = 0x%08x)",
1165 			    device_xname(sc->sc_dev), rxstat);
1166 			if (rxstat & RE_RDESC_STAT_FRALIGN)
1167 				printf(", frame alignment error");
1168 			if (rxstat & RE_RDESC_STAT_BUFOFLOW)
1169 				printf(", out of buffer space");
1170 			if (rxstat & RE_RDESC_STAT_FIFOOFLOW)
1171 				printf(", FIFO overrun");
1172 			if (rxstat & RE_RDESC_STAT_GIANT)
1173 				printf(", giant packet");
1174 			if (rxstat & RE_RDESC_STAT_RUNT)
1175 				printf(", runt packet");
1176 			if (rxstat & RE_RDESC_STAT_CRCERR)
1177 				printf(", CRC error");
1178 			printf("\n");
1179 #endif
1180 			ifp->if_ierrors++;
1181 			/*
1182 			 * If this is part of a multi-fragment packet,
1183 			 * discard all the pieces.
1184 			 */
1185 			if (sc->re_head != NULL) {
1186 				m_freem(sc->re_head);
1187 				sc->re_head = sc->re_tail = NULL;
1188 			}
1189 			re_newbuf(sc, i, m);
1190 			continue;
1191 		}
1192 
1193 		/*
1194 		 * If allocating a replacement mbuf fails,
1195 		 * reload the current one.
1196 		 */
1197 
1198 		if (__predict_false(re_newbuf(sc, i, NULL) != 0)) {
1199 			ifp->if_ierrors++;
1200 			if (sc->re_head != NULL) {
1201 				m_freem(sc->re_head);
1202 				sc->re_head = sc->re_tail = NULL;
1203 			}
1204 			re_newbuf(sc, i, m);
1205 			continue;
1206 		}
1207 
1208 		if (sc->re_head != NULL) {
1209 			m->m_len = total_len % (MCLBYTES - RE_ETHER_ALIGN);
1210 			/*
1211 			 * Special case: if there's 4 bytes or less
1212 			 * in this buffer, the mbuf can be discarded:
1213 			 * the last 4 bytes is the CRC, which we don't
1214 			 * care about anyway.
1215 			 */
1216 			if (m->m_len <= ETHER_CRC_LEN) {
1217 				sc->re_tail->m_len -=
1218 				    (ETHER_CRC_LEN - m->m_len);
1219 				m_freem(m);
1220 			} else {
1221 				m->m_len -= ETHER_CRC_LEN;
1222 				m->m_flags &= ~M_PKTHDR;
1223 				sc->re_tail->m_next = m;
1224 			}
1225 			m = sc->re_head;
1226 			sc->re_head = sc->re_tail = NULL;
1227 			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
1228 		} else
1229 			m->m_pkthdr.len = m->m_len =
1230 			    (total_len - ETHER_CRC_LEN);
1231 
1232 		ifp->if_ipackets++;
1233 		m->m_pkthdr.rcvif = ifp;
1234 
1235 		/* Do RX checksumming */
1236 
1237 		/* Check IP header checksum */
1238 		if (rxstat & RE_RDESC_STAT_PROTOID) {
1239 			m->m_pkthdr.csum_flags |= M_CSUM_IPv4;
1240 			if (rxstat & RE_RDESC_STAT_IPSUMBAD)
1241 				m->m_pkthdr.csum_flags |= M_CSUM_IPv4_BAD;
1242 		}
1243 
1244 		/* Check TCP/UDP checksum */
1245 		if (RE_TCPPKT(rxstat)) {
1246 			m->m_pkthdr.csum_flags |= M_CSUM_TCPv4;
1247 			if (rxstat & RE_RDESC_STAT_TCPSUMBAD)
1248 				m->m_pkthdr.csum_flags |= M_CSUM_TCP_UDP_BAD;
1249 		} else if (RE_UDPPKT(rxstat)) {
1250 			m->m_pkthdr.csum_flags |= M_CSUM_UDPv4;
1251 			if (rxstat & RE_RDESC_STAT_UDPSUMBAD)
1252 				m->m_pkthdr.csum_flags |= M_CSUM_TCP_UDP_BAD;
1253 		}
1254 
1255 		if (rxvlan & RE_RDESC_VLANCTL_TAG) {
1256 			VLAN_INPUT_TAG(ifp, m,
1257 			     bswap16(rxvlan & RE_RDESC_VLANCTL_DATA),
1258 			     continue);
1259 		}
1260 #if NBPFILTER > 0
1261 		if (ifp->if_bpf)
1262 			bpf_mtap(ifp->if_bpf, m);
1263 #endif
1264 		(*ifp->if_input)(ifp, m);
1265 	}
1266 
1267 	sc->re_ldata.re_rx_prodidx = i;
1268 }
1269 
1270 static void
1271 re_txeof(struct rtk_softc *sc)
1272 {
1273 	struct ifnet *ifp;
1274 	struct re_txq *txq;
1275 	uint32_t txstat;
1276 	int idx, descidx;
1277 
1278 	ifp = &sc->ethercom.ec_if;
1279 
1280 	for (idx = sc->re_ldata.re_txq_considx;
1281 	    sc->re_ldata.re_txq_free < RE_TX_QLEN;
1282 	    idx = RE_NEXT_TXQ(sc, idx), sc->re_ldata.re_txq_free++) {
1283 		txq = &sc->re_ldata.re_txq[idx];
1284 		KASSERT(txq->txq_mbuf != NULL);
1285 
1286 		descidx = txq->txq_descidx;
1287 		RE_TXDESCSYNC(sc, descidx,
1288 		    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1289 		txstat =
1290 		    le32toh(sc->re_ldata.re_tx_list[descidx].re_cmdstat);
1291 		RE_TXDESCSYNC(sc, descidx, BUS_DMASYNC_PREREAD);
1292 		KASSERT((txstat & RE_TDESC_CMD_EOF) != 0);
1293 		if (txstat & RE_TDESC_CMD_OWN) {
1294 			break;
1295 		}
1296 
1297 		sc->re_ldata.re_tx_free += txq->txq_nsegs;
1298 		KASSERT(sc->re_ldata.re_tx_free <= RE_TX_DESC_CNT(sc));
1299 		bus_dmamap_sync(sc->sc_dmat, txq->txq_dmamap,
1300 		    0, txq->txq_dmamap->dm_mapsize, BUS_DMASYNC_POSTWRITE);
1301 		bus_dmamap_unload(sc->sc_dmat, txq->txq_dmamap);
1302 		m_freem(txq->txq_mbuf);
1303 		txq->txq_mbuf = NULL;
1304 
1305 		if (txstat & (RE_TDESC_STAT_EXCESSCOL | RE_TDESC_STAT_COLCNT))
1306 			ifp->if_collisions++;
1307 		if (txstat & RE_TDESC_STAT_TXERRSUM)
1308 			ifp->if_oerrors++;
1309 		else
1310 			ifp->if_opackets++;
1311 	}
1312 
1313 	sc->re_ldata.re_txq_considx = idx;
1314 
1315 	if (sc->re_ldata.re_txq_free > RE_NTXDESC_RSVD)
1316 		ifp->if_flags &= ~IFF_OACTIVE;
1317 
1318 	/*
1319 	 * If not all descriptors have been released reaped yet,
1320 	 * reload the timer so that we will eventually get another
1321 	 * interrupt that will cause us to re-enter this routine.
1322 	 * This is done in case the transmitter has gone idle.
1323 	 */
1324 	if (sc->re_ldata.re_txq_free < RE_TX_QLEN) {
1325 		CSR_WRITE_4(sc, RTK_TIMERCNT, 1);
1326 		if ((sc->sc_quirk & RTKQ_PCIE) != 0) {
1327 			/*
1328 			 * Some chips will ignore a second TX request
1329 			 * issued while an existing transmission is in
1330 			 * progress. If the transmitter goes idle but
1331 			 * there are still packets waiting to be sent,
1332 			 * we need to restart the channel here to flush
1333 			 * them out. This only seems to be required with
1334 			 * the PCIe devices.
1335 			 */
1336 			CSR_WRITE_1(sc, RTK_GTXSTART, RTK_TXSTART_START);
1337 		}
1338 	} else
1339 		ifp->if_timer = 0;
1340 }
1341 
1342 static void
1343 re_tick(void *arg)
1344 {
1345 	struct rtk_softc *sc = arg;
1346 	int s;
1347 
1348 	/*XXX: just return for 8169S/8110S with rev 2 or newer phy */
1349 	s = splnet();
1350 
1351 	mii_tick(&sc->mii);
1352 	splx(s);
1353 
1354 	callout_reset(&sc->rtk_tick_ch, hz, re_tick, sc);
1355 }
1356 
1357 int
1358 re_intr(void *arg)
1359 {
1360 	struct rtk_softc *sc = arg;
1361 	struct ifnet *ifp;
1362 	uint16_t status;
1363 	int handled = 0;
1364 
1365 	if (!device_has_power(sc->sc_dev))
1366 		return 0;
1367 
1368 	ifp = &sc->ethercom.ec_if;
1369 
1370 	if ((ifp->if_flags & IFF_UP) == 0)
1371 		return 0;
1372 
1373 	for (;;) {
1374 
1375 		status = CSR_READ_2(sc, RTK_ISR);
1376 		/* If the card has gone away the read returns 0xffff. */
1377 		if (status == 0xffff)
1378 			break;
1379 		if (status) {
1380 			handled = 1;
1381 			CSR_WRITE_2(sc, RTK_ISR, status);
1382 		}
1383 
1384 		if ((status & RTK_INTRS_CPLUS) == 0)
1385 			break;
1386 
1387 		if (status & (RTK_ISR_RX_OK | RTK_ISR_RX_ERR))
1388 			re_rxeof(sc);
1389 
1390 		if (status & (RTK_ISR_TIMEOUT_EXPIRED | RTK_ISR_TX_ERR |
1391 		    RTK_ISR_TX_DESC_UNAVAIL))
1392 			re_txeof(sc);
1393 
1394 		if (status & RTK_ISR_SYSTEM_ERR) {
1395 			re_init(ifp);
1396 		}
1397 
1398 		if (status & RTK_ISR_LINKCHG) {
1399 			callout_stop(&sc->rtk_tick_ch);
1400 			re_tick(sc);
1401 		}
1402 	}
1403 
1404 	if (handled && !IFQ_IS_EMPTY(&ifp->if_snd))
1405 		re_start(ifp);
1406 
1407 	return handled;
1408 }
1409 
1410 
1411 
1412 /*
1413  * Main transmit routine for C+ and gigE NICs.
1414  */
1415 
1416 static void
1417 re_start(struct ifnet *ifp)
1418 {
1419 	struct rtk_softc *sc;
1420 	struct mbuf *m;
1421 	bus_dmamap_t map;
1422 	struct re_txq *txq;
1423 	struct re_desc *d;
1424 	struct m_tag *mtag;
1425 	uint32_t cmdstat, re_flags, vlanctl;
1426 	int ofree, idx, error, nsegs, seg;
1427 	int startdesc, curdesc, lastdesc;
1428 	bool pad;
1429 
1430 	sc = ifp->if_softc;
1431 	ofree = sc->re_ldata.re_txq_free;
1432 
1433 	for (idx = sc->re_ldata.re_txq_prodidx;; idx = RE_NEXT_TXQ(sc, idx)) {
1434 
1435 		IFQ_POLL(&ifp->if_snd, m);
1436 		if (m == NULL)
1437 			break;
1438 
1439 		if (sc->re_ldata.re_txq_free == 0 ||
1440 		    sc->re_ldata.re_tx_free == 0) {
1441 			/* no more free slots left */
1442 			ifp->if_flags |= IFF_OACTIVE;
1443 			break;
1444 		}
1445 
1446 		/*
1447 		 * Set up checksum offload. Note: checksum offload bits must
1448 		 * appear in all descriptors of a multi-descriptor transmit
1449 		 * attempt. (This is according to testing done with an 8169
1450 		 * chip. I'm not sure if this is a requirement or a bug.)
1451 		 */
1452 
1453 		if ((m->m_pkthdr.csum_flags & M_CSUM_TSOv4) != 0) {
1454 			uint32_t segsz = m->m_pkthdr.segsz;
1455 
1456 			re_flags = RE_TDESC_CMD_LGSEND |
1457 			    (segsz << RE_TDESC_CMD_MSSVAL_SHIFT);
1458 		} else {
1459 			/*
1460 			 * set RE_TDESC_CMD_IPCSUM if any checksum offloading
1461 			 * is requested.  otherwise, RE_TDESC_CMD_TCPCSUM/
1462 			 * RE_TDESC_CMD_UDPCSUM doesn't make effects.
1463 			 */
1464 			re_flags = 0;
1465 			if ((m->m_pkthdr.csum_flags &
1466 			    (M_CSUM_IPv4 | M_CSUM_TCPv4 | M_CSUM_UDPv4))
1467 			    != 0) {
1468 				re_flags |= RE_TDESC_CMD_IPCSUM;
1469 				if (m->m_pkthdr.csum_flags & M_CSUM_TCPv4) {
1470 					re_flags |= RE_TDESC_CMD_TCPCSUM;
1471 				} else if (m->m_pkthdr.csum_flags &
1472 				    M_CSUM_UDPv4) {
1473 					re_flags |= RE_TDESC_CMD_UDPCSUM;
1474 				}
1475 			}
1476 		}
1477 
1478 		txq = &sc->re_ldata.re_txq[idx];
1479 		map = txq->txq_dmamap;
1480 		error = bus_dmamap_load_mbuf(sc->sc_dmat, map, m,
1481 		    BUS_DMA_WRITE|BUS_DMA_NOWAIT);
1482 
1483 		if (__predict_false(error)) {
1484 			/* XXX try to defrag if EFBIG? */
1485 			printf("%s: can't map mbuf (error %d)\n",
1486 			    device_xname(sc->sc_dev), error);
1487 
1488 			IFQ_DEQUEUE(&ifp->if_snd, m);
1489 			m_freem(m);
1490 			ifp->if_oerrors++;
1491 			continue;
1492 		}
1493 
1494 		nsegs = map->dm_nsegs;
1495 		pad = false;
1496 		if (__predict_false(m->m_pkthdr.len <= RE_IP4CSUMTX_PADLEN &&
1497 		    (re_flags & RE_TDESC_CMD_IPCSUM) != 0)) {
1498 			pad = true;
1499 			nsegs++;
1500 		}
1501 
1502 		if (nsegs > sc->re_ldata.re_tx_free) {
1503 			/*
1504 			 * Not enough free descriptors to transmit this packet.
1505 			 */
1506 			ifp->if_flags |= IFF_OACTIVE;
1507 			bus_dmamap_unload(sc->sc_dmat, map);
1508 			break;
1509 		}
1510 
1511 		IFQ_DEQUEUE(&ifp->if_snd, m);
1512 
1513 		/*
1514 		 * Make sure that the caches are synchronized before we
1515 		 * ask the chip to start DMA for the packet data.
1516 		 */
1517 		bus_dmamap_sync(sc->sc_dmat, map, 0, map->dm_mapsize,
1518 		    BUS_DMASYNC_PREWRITE);
1519 
1520 		/*
1521 		 * Set up hardware VLAN tagging. Note: vlan tag info must
1522 		 * appear in all descriptors of a multi-descriptor
1523 		 * transmission attempt.
1524 		 */
1525 		vlanctl = 0;
1526 		if ((mtag = VLAN_OUTPUT_TAG(&sc->ethercom, m)) != NULL)
1527 			vlanctl = bswap16(VLAN_TAG_VALUE(mtag)) |
1528 			    RE_TDESC_VLANCTL_TAG;
1529 
1530 		/*
1531 		 * Map the segment array into descriptors.
1532 		 * Note that we set the start-of-frame and
1533 		 * end-of-frame markers for either TX or RX,
1534 		 * but they really only have meaning in the TX case.
1535 		 * (In the RX case, it's the chip that tells us
1536 		 *  where packets begin and end.)
1537 		 * We also keep track of the end of the ring
1538 		 * and set the end-of-ring bits as needed,
1539 		 * and we set the ownership bits in all except
1540 		 * the very first descriptor. (The caller will
1541 		 * set this descriptor later when it start
1542 		 * transmission or reception.)
1543 		 */
1544 		curdesc = startdesc = sc->re_ldata.re_tx_nextfree;
1545 		lastdesc = -1;
1546 		for (seg = 0; seg < map->dm_nsegs;
1547 		    seg++, curdesc = RE_NEXT_TX_DESC(sc, curdesc)) {
1548 			d = &sc->re_ldata.re_tx_list[curdesc];
1549 #ifdef DIAGNOSTIC
1550 			RE_TXDESCSYNC(sc, curdesc,
1551 			    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1552 			cmdstat = le32toh(d->re_cmdstat);
1553 			RE_TXDESCSYNC(sc, curdesc, BUS_DMASYNC_PREREAD);
1554 			if (cmdstat & RE_TDESC_STAT_OWN) {
1555 				panic("%s: tried to map busy TX descriptor",
1556 				    device_xname(sc->sc_dev));
1557 			}
1558 #endif
1559 
1560 			d->re_vlanctl = htole32(vlanctl);
1561 			re_set_bufaddr(d, map->dm_segs[seg].ds_addr);
1562 			cmdstat = re_flags | map->dm_segs[seg].ds_len;
1563 			if (seg == 0)
1564 				cmdstat |= RE_TDESC_CMD_SOF;
1565 			else
1566 				cmdstat |= RE_TDESC_CMD_OWN;
1567 			if (curdesc == (RE_TX_DESC_CNT(sc) - 1))
1568 				cmdstat |= RE_TDESC_CMD_EOR;
1569 			if (seg == nsegs - 1) {
1570 				cmdstat |= RE_TDESC_CMD_EOF;
1571 				lastdesc = curdesc;
1572 			}
1573 			d->re_cmdstat = htole32(cmdstat);
1574 			RE_TXDESCSYNC(sc, curdesc,
1575 			    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1576 		}
1577 		if (__predict_false(pad)) {
1578 			bus_addr_t paddaddr;
1579 
1580 			d = &sc->re_ldata.re_tx_list[curdesc];
1581 			d->re_vlanctl = htole32(vlanctl);
1582 			paddaddr = RE_TXPADDADDR(sc);
1583 			re_set_bufaddr(d, paddaddr);
1584 			cmdstat = re_flags |
1585 			    RE_TDESC_CMD_OWN | RE_TDESC_CMD_EOF |
1586 			    (RE_IP4CSUMTX_PADLEN + 1 - m->m_pkthdr.len);
1587 			if (curdesc == (RE_TX_DESC_CNT(sc) - 1))
1588 				cmdstat |= RE_TDESC_CMD_EOR;
1589 			d->re_cmdstat = htole32(cmdstat);
1590 			RE_TXDESCSYNC(sc, curdesc,
1591 			    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1592 			lastdesc = curdesc;
1593 			curdesc = RE_NEXT_TX_DESC(sc, curdesc);
1594 		}
1595 		KASSERT(lastdesc != -1);
1596 
1597 		/* Transfer ownership of packet to the chip. */
1598 
1599 		sc->re_ldata.re_tx_list[startdesc].re_cmdstat |=
1600 		    htole32(RE_TDESC_CMD_OWN);
1601 		RE_TXDESCSYNC(sc, startdesc,
1602 		    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1603 
1604 		/* update info of TX queue and descriptors */
1605 		txq->txq_mbuf = m;
1606 		txq->txq_descidx = lastdesc;
1607 		txq->txq_nsegs = nsegs;
1608 
1609 		sc->re_ldata.re_txq_free--;
1610 		sc->re_ldata.re_tx_free -= nsegs;
1611 		sc->re_ldata.re_tx_nextfree = curdesc;
1612 
1613 #if NBPFILTER > 0
1614 		/*
1615 		 * If there's a BPF listener, bounce a copy of this frame
1616 		 * to him.
1617 		 */
1618 		if (ifp->if_bpf)
1619 			bpf_mtap(ifp->if_bpf, m);
1620 #endif
1621 	}
1622 
1623 	if (sc->re_ldata.re_txq_free < ofree) {
1624 		/*
1625 		 * TX packets are enqueued.
1626 		 */
1627 		sc->re_ldata.re_txq_prodidx = idx;
1628 
1629 		/*
1630 		 * Start the transmitter to poll.
1631 		 *
1632 		 * RealTek put the TX poll request register in a different
1633 		 * location on the 8169 gigE chip. I don't know why.
1634 		 */
1635 		if ((sc->sc_quirk & RTKQ_8139CPLUS) != 0)
1636 			CSR_WRITE_1(sc, RTK_TXSTART, RTK_TXSTART_START);
1637 		else
1638 			CSR_WRITE_1(sc, RTK_GTXSTART, RTK_TXSTART_START);
1639 
1640 		/*
1641 		 * Use the countdown timer for interrupt moderation.
1642 		 * 'TX done' interrupts are disabled. Instead, we reset the
1643 		 * countdown timer, which will begin counting until it hits
1644 		 * the value in the TIMERINT register, and then trigger an
1645 		 * interrupt. Each time we write to the TIMERCNT register,
1646 		 * the timer count is reset to 0.
1647 		 */
1648 		CSR_WRITE_4(sc, RTK_TIMERCNT, 1);
1649 
1650 		/*
1651 		 * Set a timeout in case the chip goes out to lunch.
1652 		 */
1653 		ifp->if_timer = 5;
1654 	}
1655 }
1656 
1657 static int
1658 re_init(struct ifnet *ifp)
1659 {
1660 	struct rtk_softc *sc = ifp->if_softc;
1661 	const uint8_t *enaddr;
1662 	uint32_t rxcfg = 0;
1663 	uint32_t reg;
1664 	int error;
1665 
1666 	if ((error = re_enable(sc)) != 0)
1667 		goto out;
1668 
1669 	/*
1670 	 * Cancel pending I/O and free all RX/TX buffers.
1671 	 */
1672 	re_stop(ifp, 0);
1673 
1674 	re_reset(sc);
1675 
1676 	/*
1677 	 * Enable C+ RX and TX mode, as well as VLAN stripping and
1678 	 * RX checksum offload. We must configure the C+ register
1679 	 * before all others.
1680 	 */
1681 	reg = 0;
1682 
1683 	/*
1684 	 * XXX: Realtek docs say bits 0 and 1 are reserved, for 8169S/8110S.
1685 	 * FreeBSD  drivers set these bits anyway (for 8139C+?).
1686 	 * So far, it works.
1687 	 */
1688 
1689 	/*
1690 	 * XXX: For old 8169 set bit 14.
1691 	 *      For 8169S/8110S and above, do not set bit 14.
1692 	 */
1693 	if ((sc->sc_quirk & RTKQ_8169NONS) != 0)
1694 		reg |= (0x1 << 14) | RTK_CPLUSCMD_PCI_MRW;
1695 
1696 	if (1)  {/* not for 8169S ? */
1697 		reg |=
1698 		    RTK_CPLUSCMD_VLANSTRIP |
1699 		    (ifp->if_capenable &
1700 		    (IFCAP_CSUM_IPv4_Rx | IFCAP_CSUM_TCPv4_Rx |
1701 		     IFCAP_CSUM_UDPv4_Rx) ?
1702 		    RTK_CPLUSCMD_RXCSUM_ENB : 0);
1703 	}
1704 
1705 	CSR_WRITE_2(sc, RTK_CPLUS_CMD,
1706 	    reg | RTK_CPLUSCMD_RXENB | RTK_CPLUSCMD_TXENB);
1707 
1708 	/* XXX: from Realtek-supplied Linux driver. Wholly undocumented. */
1709 	if ((sc->sc_quirk & RTKQ_8139CPLUS) == 0)
1710 		CSR_WRITE_2(sc, RTK_IM, 0x0000);
1711 
1712 	DELAY(10000);
1713 
1714 	/*
1715 	 * Init our MAC address.  Even though the chipset
1716 	 * documentation doesn't mention it, we need to enter "Config
1717 	 * register write enable" mode to modify the ID registers.
1718 	 */
1719 	CSR_WRITE_1(sc, RTK_EECMD, RTK_EEMODE_WRITECFG);
1720 	enaddr = CLLADDR(ifp->if_sadl);
1721 	reg = enaddr[0] | (enaddr[1] << 8) |
1722 	    (enaddr[2] << 16) | (enaddr[3] << 24);
1723 	CSR_WRITE_4(sc, RTK_IDR0, reg);
1724 	reg = enaddr[4] | (enaddr[5] << 8);
1725 	CSR_WRITE_4(sc, RTK_IDR4, reg);
1726 	CSR_WRITE_1(sc, RTK_EECMD, RTK_EEMODE_OFF);
1727 
1728 	/*
1729 	 * For C+ mode, initialize the RX descriptors and mbufs.
1730 	 */
1731 	re_rx_list_init(sc);
1732 	re_tx_list_init(sc);
1733 
1734 	/*
1735 	 * Load the addresses of the RX and TX lists into the chip.
1736 	 */
1737 	CSR_WRITE_4(sc, RTK_RXLIST_ADDR_HI,
1738 	    RE_ADDR_HI(sc->re_ldata.re_rx_list_map->dm_segs[0].ds_addr));
1739 	CSR_WRITE_4(sc, RTK_RXLIST_ADDR_LO,
1740 	    RE_ADDR_LO(sc->re_ldata.re_rx_list_map->dm_segs[0].ds_addr));
1741 
1742 	CSR_WRITE_4(sc, RTK_TXLIST_ADDR_HI,
1743 	    RE_ADDR_HI(sc->re_ldata.re_tx_list_map->dm_segs[0].ds_addr));
1744 	CSR_WRITE_4(sc, RTK_TXLIST_ADDR_LO,
1745 	    RE_ADDR_LO(sc->re_ldata.re_tx_list_map->dm_segs[0].ds_addr));
1746 
1747 	/*
1748 	 * Enable transmit and receive.
1749 	 */
1750 	CSR_WRITE_1(sc, RTK_COMMAND, RTK_CMD_TX_ENB | RTK_CMD_RX_ENB);
1751 
1752 	/*
1753 	 * Set the initial TX and RX configuration.
1754 	 */
1755 	if (sc->re_testmode && (sc->sc_quirk & RTKQ_8169NONS) != 0) {
1756 		/* test mode is needed only for old 8169 */
1757 		CSR_WRITE_4(sc, RTK_TXCFG,
1758 		    RE_TXCFG_CONFIG | RTK_LOOPTEST_ON);
1759 	} else
1760 		CSR_WRITE_4(sc, RTK_TXCFG, RE_TXCFG_CONFIG);
1761 
1762 	CSR_WRITE_1(sc, RTK_EARLY_TX_THRESH, 16);
1763 
1764 	CSR_WRITE_4(sc, RTK_RXCFG, RE_RXCFG_CONFIG);
1765 
1766 	/* Set the individual bit to receive frames for this host only. */
1767 	rxcfg = CSR_READ_4(sc, RTK_RXCFG);
1768 	rxcfg |= RTK_RXCFG_RX_INDIV;
1769 
1770 	/* If we want promiscuous mode, set the allframes bit. */
1771 	if (ifp->if_flags & IFF_PROMISC)
1772 		rxcfg |= RTK_RXCFG_RX_ALLPHYS;
1773 	else
1774 		rxcfg &= ~RTK_RXCFG_RX_ALLPHYS;
1775 	CSR_WRITE_4(sc, RTK_RXCFG, rxcfg);
1776 
1777 	/*
1778 	 * Set capture broadcast bit to capture broadcast frames.
1779 	 */
1780 	if (ifp->if_flags & IFF_BROADCAST)
1781 		rxcfg |= RTK_RXCFG_RX_BROAD;
1782 	else
1783 		rxcfg &= ~RTK_RXCFG_RX_BROAD;
1784 	CSR_WRITE_4(sc, RTK_RXCFG, rxcfg);
1785 
1786 	/*
1787 	 * Program the multicast filter, if necessary.
1788 	 */
1789 	rtk_setmulti(sc);
1790 
1791 	/*
1792 	 * Enable interrupts.
1793 	 */
1794 	if (sc->re_testmode)
1795 		CSR_WRITE_2(sc, RTK_IMR, 0);
1796 	else
1797 		CSR_WRITE_2(sc, RTK_IMR, RTK_INTRS_CPLUS);
1798 
1799 	/* Start RX/TX process. */
1800 	CSR_WRITE_4(sc, RTK_MISSEDPKT, 0);
1801 #ifdef notdef
1802 	/* Enable receiver and transmitter. */
1803 	CSR_WRITE_1(sc, RTK_COMMAND, RTK_CMD_TX_ENB | RTK_CMD_RX_ENB);
1804 #endif
1805 
1806 	/*
1807 	 * Initialize the timer interrupt register so that
1808 	 * a timer interrupt will be generated once the timer
1809 	 * reaches a certain number of ticks. The timer is
1810 	 * reloaded on each transmit. This gives us TX interrupt
1811 	 * moderation, which dramatically improves TX frame rate.
1812 	 */
1813 
1814 	if ((sc->sc_quirk & RTKQ_8139CPLUS) != 0)
1815 		CSR_WRITE_4(sc, RTK_TIMERINT, 0x400);
1816 	else {
1817 		CSR_WRITE_4(sc, RTK_TIMERINT_8169, 0x800);
1818 
1819 		/*
1820 		 * For 8169 gigE NICs, set the max allowed RX packet
1821 		 * size so we can receive jumbo frames.
1822 		 */
1823 		CSR_WRITE_2(sc, RTK_MAXRXPKTLEN, 16383);
1824 	}
1825 
1826 	if (sc->re_testmode)
1827 		return 0;
1828 
1829 	CSR_WRITE_1(sc, RTK_CFG1, RTK_CFG1_DRVLOAD);
1830 
1831 	ifp->if_flags |= IFF_RUNNING;
1832 	ifp->if_flags &= ~IFF_OACTIVE;
1833 
1834 	callout_reset(&sc->rtk_tick_ch, hz, re_tick, sc);
1835 
1836  out:
1837 	if (error) {
1838 		ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
1839 		ifp->if_timer = 0;
1840 		printf("%s: interface not running\n",
1841 		    device_xname(sc->sc_dev));
1842 	}
1843 
1844 	return error;
1845 }
1846 
1847 static int
1848 re_ioctl(struct ifnet *ifp, u_long command, void *data)
1849 {
1850 	struct rtk_softc *sc = ifp->if_softc;
1851 	struct ifreq *ifr = data;
1852 	int s, error = 0;
1853 
1854 	s = splnet();
1855 
1856 	switch (command) {
1857 	case SIOCSIFMTU:
1858 		/*
1859 		 * According to FreeBSD, 8102E/8102EL use a different DMA
1860 		 * descriptor format. 8168C/8111C requires touching additional
1861 		 * magic registers. Depending on MAC revisions some controllers
1862 		 * need to disable checksum offload.
1863 		 *
1864 		 * Disable jumbo frames for those parts.
1865 		 */
1866 		if ((sc->sc_rev == 24 || sc->sc_rev == 25) &&
1867 		    ifr->ifr_mtu > ETHERMTU) {
1868 			error = EINVAL;
1869 			break;
1870 		}
1871 
1872 		if (ifr->ifr_mtu < ETHERMIN || ifr->ifr_mtu > ETHERMTU_JUMBO)
1873 			error = EINVAL;
1874 		else if ((error = ifioctl_common(ifp, command, data)) ==
1875 		    ENETRESET)
1876 			error = 0;
1877 		break;
1878 	default:
1879 		if ((error = ether_ioctl(ifp, command, data)) != ENETRESET)
1880 			break;
1881 
1882 		error = 0;
1883 
1884 		if (command == SIOCSIFCAP)
1885 			error = (*ifp->if_init)(ifp);
1886 		else if (command != SIOCADDMULTI && command != SIOCDELMULTI)
1887 			;
1888 		else if (ifp->if_flags & IFF_RUNNING)
1889 			rtk_setmulti(sc);
1890 		break;
1891 	}
1892 
1893 	splx(s);
1894 
1895 	return error;
1896 }
1897 
1898 static void
1899 re_watchdog(struct ifnet *ifp)
1900 {
1901 	struct rtk_softc *sc;
1902 	int s;
1903 
1904 	sc = ifp->if_softc;
1905 	s = splnet();
1906 	printf("%s: watchdog timeout\n", device_xname(sc->sc_dev));
1907 	ifp->if_oerrors++;
1908 
1909 	re_txeof(sc);
1910 	re_rxeof(sc);
1911 
1912 	re_init(ifp);
1913 
1914 	splx(s);
1915 }
1916 
1917 /*
1918  * Stop the adapter and free any mbufs allocated to the
1919  * RX and TX lists.
1920  */
1921 static void
1922 re_stop(struct ifnet *ifp, int disable)
1923 {
1924 	int i;
1925 	struct rtk_softc *sc = ifp->if_softc;
1926 
1927 	callout_stop(&sc->rtk_tick_ch);
1928 
1929 	mii_down(&sc->mii);
1930 
1931 	CSR_WRITE_1(sc, RTK_COMMAND, 0x00);
1932 	CSR_WRITE_2(sc, RTK_IMR, 0x0000);
1933 
1934 	if (sc->re_head != NULL) {
1935 		m_freem(sc->re_head);
1936 		sc->re_head = sc->re_tail = NULL;
1937 	}
1938 
1939 	/* Free the TX list buffers. */
1940 	for (i = 0; i < RE_TX_QLEN; i++) {
1941 		if (sc->re_ldata.re_txq[i].txq_mbuf != NULL) {
1942 			bus_dmamap_unload(sc->sc_dmat,
1943 			    sc->re_ldata.re_txq[i].txq_dmamap);
1944 			m_freem(sc->re_ldata.re_txq[i].txq_mbuf);
1945 			sc->re_ldata.re_txq[i].txq_mbuf = NULL;
1946 		}
1947 	}
1948 
1949 	/* Free the RX list buffers. */
1950 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
1951 		if (sc->re_ldata.re_rxsoft[i].rxs_mbuf != NULL) {
1952 			bus_dmamap_unload(sc->sc_dmat,
1953 			    sc->re_ldata.re_rxsoft[i].rxs_dmamap);
1954 			m_freem(sc->re_ldata.re_rxsoft[i].rxs_mbuf);
1955 			sc->re_ldata.re_rxsoft[i].rxs_mbuf = NULL;
1956 		}
1957 	}
1958 
1959 	if (disable)
1960 		re_disable(sc);
1961 
1962 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
1963 	ifp->if_timer = 0;
1964 }
1965