xref: /netbsd-src/sys/dev/ic/rtl8169.c (revision 3816d47b2c42fcd6e549e3407f842a5b1a1d23ad)
1 /*	$NetBSD: rtl8169.c,v 1.128 2009/09/15 19:29:17 dyoung 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 1.128 2009/09/15 19:29:17 dyoung 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 
116 #include <sys/param.h>
117 #include <sys/endian.h>
118 #include <sys/systm.h>
119 #include <sys/sockio.h>
120 #include <sys/mbuf.h>
121 #include <sys/malloc.h>
122 #include <sys/kernel.h>
123 #include <sys/socket.h>
124 #include <sys/device.h>
125 
126 #include <net/if.h>
127 #include <net/if_arp.h>
128 #include <net/if_dl.h>
129 #include <net/if_ether.h>
130 #include <net/if_media.h>
131 #include <net/if_vlanvar.h>
132 
133 #include <netinet/in_systm.h>	/* XXX for IP_MAXPACKET */
134 #include <netinet/in.h>		/* XXX for IP_MAXPACKET */
135 #include <netinet/ip.h>		/* XXX for IP_MAXPACKET */
136 
137 #if NBPFILTER > 0
138 #include <net/bpf.h>
139 #endif
140 
141 #include <sys/bus.h>
142 
143 #include <dev/mii/mii.h>
144 #include <dev/mii/miivar.h>
145 
146 #include <dev/ic/rtl81x9reg.h>
147 #include <dev/ic/rtl81x9var.h>
148 
149 #include <dev/ic/rtl8169var.h>
150 
151 static inline void re_set_bufaddr(struct re_desc *, bus_addr_t);
152 
153 static int re_newbuf(struct rtk_softc *, int, struct mbuf *);
154 static int re_rx_list_init(struct rtk_softc *);
155 static int re_tx_list_init(struct rtk_softc *);
156 static void re_rxeof(struct rtk_softc *);
157 static void re_txeof(struct rtk_softc *);
158 static void re_tick(void *);
159 static void re_start(struct ifnet *);
160 static int re_ioctl(struct ifnet *, u_long, void *);
161 static int re_init(struct ifnet *);
162 static void re_stop(struct ifnet *, int);
163 static void re_watchdog(struct ifnet *);
164 
165 static int re_enable(struct rtk_softc *);
166 static void re_disable(struct rtk_softc *);
167 
168 static int re_gmii_readreg(device_t, int, int);
169 static void re_gmii_writereg(device_t, int, int, int);
170 
171 static int re_miibus_readreg(device_t, int, int);
172 static void re_miibus_writereg(device_t, int, int, int);
173 static void re_miibus_statchg(device_t);
174 
175 static void re_reset(struct rtk_softc *);
176 
177 static inline void
178 re_set_bufaddr(struct re_desc *d, bus_addr_t addr)
179 {
180 
181 	d->re_bufaddr_lo = htole32((uint32_t)addr);
182 	if (sizeof(bus_addr_t) == sizeof(uint64_t))
183 		d->re_bufaddr_hi = htole32((uint64_t)addr >> 32);
184 	else
185 		d->re_bufaddr_hi = 0;
186 }
187 
188 static int
189 re_gmii_readreg(device_t dev, int phy, int reg)
190 {
191 	struct rtk_softc *sc = device_private(dev);
192 	uint32_t rval;
193 	int i;
194 
195 	if (phy != 7)
196 		return 0;
197 
198 	/* Let the rgephy driver read the GMEDIASTAT register */
199 
200 	if (reg == RTK_GMEDIASTAT) {
201 		rval = CSR_READ_1(sc, RTK_GMEDIASTAT);
202 		return rval;
203 	}
204 
205 	CSR_WRITE_4(sc, RTK_PHYAR, reg << 16);
206 	DELAY(1000);
207 
208 	for (i = 0; i < RTK_TIMEOUT; i++) {
209 		rval = CSR_READ_4(sc, RTK_PHYAR);
210 		if (rval & RTK_PHYAR_BUSY)
211 			break;
212 		DELAY(100);
213 	}
214 
215 	if (i == RTK_TIMEOUT) {
216 		printf("%s: PHY read failed\n", device_xname(sc->sc_dev));
217 		return 0;
218 	}
219 
220 	return rval & RTK_PHYAR_PHYDATA;
221 }
222 
223 static void
224 re_gmii_writereg(device_t dev, int phy, int reg, int data)
225 {
226 	struct rtk_softc *sc = device_private(dev);
227 	uint32_t rval;
228 	int i;
229 
230 	CSR_WRITE_4(sc, RTK_PHYAR, (reg << 16) |
231 	    (data & RTK_PHYAR_PHYDATA) | RTK_PHYAR_BUSY);
232 	DELAY(1000);
233 
234 	for (i = 0; i < RTK_TIMEOUT; i++) {
235 		rval = CSR_READ_4(sc, RTK_PHYAR);
236 		if (!(rval & RTK_PHYAR_BUSY))
237 			break;
238 		DELAY(100);
239 	}
240 
241 	if (i == RTK_TIMEOUT) {
242 		printf("%s: PHY write reg %x <- %x failed\n",
243 		    device_xname(sc->sc_dev), reg, data);
244 	}
245 }
246 
247 static int
248 re_miibus_readreg(device_t dev, int phy, int reg)
249 {
250 	struct rtk_softc *sc = device_private(dev);
251 	uint16_t rval = 0;
252 	uint16_t re8139_reg = 0;
253 	int s;
254 
255 	s = splnet();
256 
257 	if ((sc->sc_quirk & RTKQ_8139CPLUS) == 0) {
258 		rval = re_gmii_readreg(dev, phy, reg);
259 		splx(s);
260 		return rval;
261 	}
262 
263 	/* Pretend the internal PHY is only at address 0 */
264 	if (phy) {
265 		splx(s);
266 		return 0;
267 	}
268 	switch (reg) {
269 	case MII_BMCR:
270 		re8139_reg = RTK_BMCR;
271 		break;
272 	case MII_BMSR:
273 		re8139_reg = RTK_BMSR;
274 		break;
275 	case MII_ANAR:
276 		re8139_reg = RTK_ANAR;
277 		break;
278 	case MII_ANER:
279 		re8139_reg = RTK_ANER;
280 		break;
281 	case MII_ANLPAR:
282 		re8139_reg = RTK_LPAR;
283 		break;
284 	case MII_PHYIDR1:
285 	case MII_PHYIDR2:
286 		splx(s);
287 		return 0;
288 	/*
289 	 * Allow the rlphy driver to read the media status
290 	 * register. If we have a link partner which does not
291 	 * support NWAY, this is the register which will tell
292 	 * us the results of parallel detection.
293 	 */
294 	case RTK_MEDIASTAT:
295 		rval = CSR_READ_1(sc, RTK_MEDIASTAT);
296 		splx(s);
297 		return rval;
298 	default:
299 		printf("%s: bad phy register\n", device_xname(sc->sc_dev));
300 		splx(s);
301 		return 0;
302 	}
303 	rval = CSR_READ_2(sc, re8139_reg);
304 	if ((sc->sc_quirk & RTKQ_8139CPLUS) != 0 && re8139_reg == RTK_BMCR) {
305 		/* 8139C+ has different bit layout. */
306 		rval &= ~(BMCR_LOOP | BMCR_ISO);
307 	}
308 	splx(s);
309 	return rval;
310 }
311 
312 static void
313 re_miibus_writereg(device_t dev, int phy, int reg, int data)
314 {
315 	struct rtk_softc *sc = device_private(dev);
316 	uint16_t re8139_reg = 0;
317 	int s;
318 
319 	s = splnet();
320 
321 	if ((sc->sc_quirk & RTKQ_8139CPLUS) == 0) {
322 		re_gmii_writereg(dev, phy, reg, data);
323 		splx(s);
324 		return;
325 	}
326 
327 	/* Pretend the internal PHY is only at address 0 */
328 	if (phy) {
329 		splx(s);
330 		return;
331 	}
332 	switch (reg) {
333 	case MII_BMCR:
334 		re8139_reg = RTK_BMCR;
335 		if ((sc->sc_quirk & RTKQ_8139CPLUS) != 0) {
336 			/* 8139C+ has different bit layout. */
337 			data &= ~(BMCR_LOOP | BMCR_ISO);
338 		}
339 		break;
340 	case MII_BMSR:
341 		re8139_reg = RTK_BMSR;
342 		break;
343 	case MII_ANAR:
344 		re8139_reg = RTK_ANAR;
345 		break;
346 	case MII_ANER:
347 		re8139_reg = RTK_ANER;
348 		break;
349 	case MII_ANLPAR:
350 		re8139_reg = RTK_LPAR;
351 		break;
352 	case MII_PHYIDR1:
353 	case MII_PHYIDR2:
354 		splx(s);
355 		return;
356 		break;
357 	default:
358 		printf("%s: bad phy register\n", device_xname(sc->sc_dev));
359 		splx(s);
360 		return;
361 	}
362 	CSR_WRITE_2(sc, re8139_reg, data);
363 	splx(s);
364 	return;
365 }
366 
367 static void
368 re_miibus_statchg(device_t dev)
369 {
370 
371 	return;
372 }
373 
374 static void
375 re_reset(struct rtk_softc *sc)
376 {
377 	int i;
378 
379 	CSR_WRITE_1(sc, RTK_COMMAND, RTK_CMD_RESET);
380 
381 	for (i = 0; i < RTK_TIMEOUT; i++) {
382 		DELAY(10);
383 		if ((CSR_READ_1(sc, RTK_COMMAND) & RTK_CMD_RESET) == 0)
384 			break;
385 	}
386 	if (i == RTK_TIMEOUT)
387 		printf("%s: reset never completed!\n",
388 		    device_xname(sc->sc_dev));
389 
390 	/*
391 	 * NB: Realtek-supplied FreeBSD driver does this only for MACFG_3,
392 	 *     but also says "Rtl8169s sigle chip detected".
393 	 */
394 	if ((sc->sc_quirk & RTKQ_MACLDPS) != 0)
395 		CSR_WRITE_1(sc, RTK_LDPS, 1);
396 
397 }
398 
399 /*
400  * The following routine is designed to test for a defect on some
401  * 32-bit 8169 cards. Some of these NICs have the REQ64# and ACK64#
402  * lines connected to the bus, however for a 32-bit only card, they
403  * should be pulled high. The result of this defect is that the
404  * NIC will not work right if you plug it into a 64-bit slot: DMA
405  * operations will be done with 64-bit transfers, which will fail
406  * because the 64-bit data lines aren't connected.
407  *
408  * There's no way to work around this (short of talking a soldering
409  * iron to the board), however we can detect it. The method we use
410  * here is to put the NIC into digital loopback mode, set the receiver
411  * to promiscuous mode, and then try to send a frame. We then compare
412  * the frame data we sent to what was received. If the data matches,
413  * then the NIC is working correctly, otherwise we know the user has
414  * a defective NIC which has been mistakenly plugged into a 64-bit PCI
415  * slot. In the latter case, there's no way the NIC can work correctly,
416  * so we print out a message on the console and abort the device attach.
417  */
418 
419 int
420 re_diag(struct rtk_softc *sc)
421 {
422 	struct ifnet *ifp = &sc->ethercom.ec_if;
423 	struct mbuf *m0;
424 	struct ether_header *eh;
425 	struct re_rxsoft *rxs;
426 	struct re_desc *cur_rx;
427 	bus_dmamap_t dmamap;
428 	uint16_t status;
429 	uint32_t rxstat;
430 	int total_len, i, s, error = 0;
431 	static const uint8_t dst[] = { 0x00, 'h', 'e', 'l', 'l', 'o' };
432 	static const uint8_t src[] = { 0x00, 'w', 'o', 'r', 'l', 'd' };
433 
434 	/* Allocate a single mbuf */
435 
436 	MGETHDR(m0, M_DONTWAIT, MT_DATA);
437 	if (m0 == NULL)
438 		return ENOBUFS;
439 
440 	/*
441 	 * Initialize the NIC in test mode. This sets the chip up
442 	 * so that it can send and receive frames, but performs the
443 	 * following special functions:
444 	 * - Puts receiver in promiscuous mode
445 	 * - Enables digital loopback mode
446 	 * - Leaves interrupts turned off
447 	 */
448 
449 	ifp->if_flags |= IFF_PROMISC;
450 	sc->re_testmode = 1;
451 	re_init(ifp);
452 	re_stop(ifp, 0);
453 	DELAY(100000);
454 	re_init(ifp);
455 
456 	/* Put some data in the mbuf */
457 
458 	eh = mtod(m0, struct ether_header *);
459 	memcpy(eh->ether_dhost, (char *)&dst, ETHER_ADDR_LEN);
460 	memcpy(eh->ether_shost, (char *)&src, ETHER_ADDR_LEN);
461 	eh->ether_type = htons(ETHERTYPE_IP);
462 	m0->m_pkthdr.len = m0->m_len = ETHER_MIN_LEN - ETHER_CRC_LEN;
463 
464 	/*
465 	 * Queue the packet, start transmission.
466 	 */
467 
468 	CSR_WRITE_2(sc, RTK_ISR, 0xFFFF);
469 	s = splnet();
470 	IF_ENQUEUE(&ifp->if_snd, m0);
471 	re_start(ifp);
472 	splx(s);
473 	m0 = NULL;
474 
475 	/* Wait for it to propagate through the chip */
476 
477 	DELAY(100000);
478 	for (i = 0; i < RTK_TIMEOUT; i++) {
479 		status = CSR_READ_2(sc, RTK_ISR);
480 		if ((status & (RTK_ISR_TIMEOUT_EXPIRED | RTK_ISR_RX_OK)) ==
481 		    (RTK_ISR_TIMEOUT_EXPIRED | RTK_ISR_RX_OK))
482 			break;
483 		DELAY(10);
484 	}
485 	if (i == RTK_TIMEOUT) {
486 		aprint_error_dev(sc->sc_dev,
487 		    "diagnostic failed, failed to receive packet "
488 		    "in loopback mode\n");
489 		error = EIO;
490 		goto done;
491 	}
492 
493 	/*
494 	 * The packet should have been dumped into the first
495 	 * entry in the RX DMA ring. Grab it from there.
496 	 */
497 
498 	rxs = &sc->re_ldata.re_rxsoft[0];
499 	dmamap = rxs->rxs_dmamap;
500 	bus_dmamap_sync(sc->sc_dmat, dmamap, 0, dmamap->dm_mapsize,
501 	    BUS_DMASYNC_POSTREAD);
502 	bus_dmamap_unload(sc->sc_dmat, dmamap);
503 
504 	m0 = rxs->rxs_mbuf;
505 	rxs->rxs_mbuf = NULL;
506 	eh = mtod(m0, struct ether_header *);
507 
508 	RE_RXDESCSYNC(sc, 0, BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
509 	cur_rx = &sc->re_ldata.re_rx_list[0];
510 	rxstat = le32toh(cur_rx->re_cmdstat);
511 	total_len = rxstat & sc->re_rxlenmask;
512 
513 	if (total_len != ETHER_MIN_LEN) {
514 		aprint_error_dev(sc->sc_dev,
515 		    "diagnostic failed, received short packet\n");
516 		error = EIO;
517 		goto done;
518 	}
519 
520 	/* Test that the received packet data matches what we sent. */
521 
522 	if (memcmp((char *)&eh->ether_dhost, (char *)&dst, ETHER_ADDR_LEN) ||
523 	    memcmp((char *)&eh->ether_shost, (char *)&src, ETHER_ADDR_LEN) ||
524 	    ntohs(eh->ether_type) != ETHERTYPE_IP) {
525 		aprint_error_dev(sc->sc_dev, "WARNING, DMA FAILURE!\n"
526 		    "expected TX data: %s/%s/0x%x\n"
527 		    "received RX data: %s/%s/0x%x\n"
528 		    "You may have a defective 32-bit NIC plugged "
529 		    "into a 64-bit PCI slot.\n"
530 		    "Please re-install the NIC in a 32-bit slot "
531 		    "for proper operation.\n"
532 		    "Read the re(4) man page for more details.\n" ,
533 		    ether_sprintf(dst),  ether_sprintf(src), ETHERTYPE_IP,
534 		    ether_sprintf(eh->ether_dhost),
535 		    ether_sprintf(eh->ether_shost), ntohs(eh->ether_type));
536 		error = EIO;
537 	}
538 
539  done:
540 	/* Turn interface off, release resources */
541 
542 	sc->re_testmode = 0;
543 	ifp->if_flags &= ~IFF_PROMISC;
544 	re_stop(ifp, 0);
545 	if (m0 != NULL)
546 		m_freem(m0);
547 
548 	return error;
549 }
550 
551 
552 /*
553  * Attach the interface. Allocate softc structures, do ifmedia
554  * setup and ethernet/BPF attach.
555  */
556 void
557 re_attach(struct rtk_softc *sc)
558 {
559 	uint8_t eaddr[ETHER_ADDR_LEN];
560 	uint16_t val;
561 	struct ifnet *ifp;
562 	int error = 0, i, addr_len;
563 
564 	if ((sc->sc_quirk & RTKQ_8139CPLUS) == 0) {
565 		uint32_t hwrev;
566 
567 		/* Revision of 8169/8169S/8110s in bits 30..26, 23 */
568 		hwrev = CSR_READ_4(sc, RTK_TXCFG) & RTK_TXCFG_HWREV;
569 		switch (hwrev) {
570 		case RTK_HWREV_8169:
571 			sc->sc_quirk |= RTKQ_8169NONS;
572 			break;
573 		case RTK_HWREV_8169S:
574 		case RTK_HWREV_8110S:
575 		case RTK_HWREV_8169_8110SB:
576 		case RTK_HWREV_8169_8110SC:
577 			sc->sc_quirk |= RTKQ_MACLDPS;
578 			break;
579 		case RTK_HWREV_8168_SPIN1:
580 		case RTK_HWREV_8168_SPIN2:
581 		case RTK_HWREV_8168_SPIN3:
582 			sc->sc_quirk |= RTKQ_MACSTAT;
583 			break;
584 		case RTK_HWREV_8168C:
585 		case RTK_HWREV_8168C_SPIN2:
586 		case RTK_HWREV_8168CP:
587 		case RTK_HWREV_8168D:
588 		case RTK_HWREV_8168DP:
589 			sc->sc_quirk |= RTKQ_DESCV2 | RTKQ_NOEECMD |
590 			    RTKQ_MACSTAT | RTKQ_CMDSTOP;
591 			/*
592 			 * From FreeBSD driver:
593 			 *
594 			 * These (8168/8111) controllers support jumbo frame
595 			 * but it seems that enabling it requires touching
596 			 * additional magic registers. Depending on MAC
597 			 * revisions some controllers need to disable
598 			 * checksum offload. So disable jumbo frame until
599 			 * I have better idea what it really requires to
600 			 * make it support.
601 			 * RTL8168C/CP : supports up to 6KB jumbo frame.
602 			 * RTL8111C/CP : supports up to 9KB jumbo frame.
603 			 */
604 			sc->sc_quirk |= RTKQ_NOJUMBO;
605 			break;
606 		case RTK_HWREV_8100E:
607 		case RTK_HWREV_8100E_SPIN2:
608 		case RTK_HWREV_8101E:
609 			sc->sc_quirk |= RTKQ_NOJUMBO;
610 			break;
611 		case RTK_HWREV_8102E:
612 		case RTK_HWREV_8102EL:
613 		case RTK_HWREV_8103E:
614 			sc->sc_quirk |= RTKQ_DESCV2 | RTKQ_NOEECMD |
615 			    RTKQ_MACSTAT | RTKQ_CMDSTOP | RTKQ_NOJUMBO;
616 			break;
617 		default:
618 			aprint_normal_dev(sc->sc_dev,
619 			    "Unknown revision (0x%08x)\n", hwrev);
620 			/* assume the latest features */
621 			sc->sc_quirk |= RTKQ_DESCV2 | RTKQ_NOEECMD;
622 			sc->sc_quirk |= RTKQ_NOJUMBO;
623 		}
624 
625 		/* Set RX length mask */
626 		sc->re_rxlenmask = RE_RDESC_STAT_GFRAGLEN;
627 		sc->re_ldata.re_tx_desc_cnt = RE_TX_DESC_CNT_8169;
628 	} else {
629 		sc->sc_quirk |= RTKQ_NOJUMBO;
630 
631 		/* Set RX length mask */
632 		sc->re_rxlenmask = RE_RDESC_STAT_FRAGLEN;
633 		sc->re_ldata.re_tx_desc_cnt = RE_TX_DESC_CNT_8139;
634 	}
635 
636 	/* Reset the adapter. */
637 	re_reset(sc);
638 
639 	if ((sc->sc_quirk & RTKQ_NOEECMD) != 0) {
640 		/*
641 		 * Get station address from ID registers.
642 		 */
643 		for (i = 0; i < ETHER_ADDR_LEN; i++)
644 			eaddr[i] = CSR_READ_1(sc, RTK_IDR0 + i);
645 	} else {
646 		/*
647 		 * Get station address from the EEPROM.
648 		 */
649 		if (rtk_read_eeprom(sc, RTK_EE_ID, RTK_EEADDR_LEN1) == 0x8129)
650 			addr_len = RTK_EEADDR_LEN1;
651 		else
652 			addr_len = RTK_EEADDR_LEN0;
653 
654 		/*
655 		 * Get station address from the EEPROM.
656 		 */
657 		for (i = 0; i < ETHER_ADDR_LEN / 2; i++) {
658 			val = rtk_read_eeprom(sc, RTK_EE_EADDR0 + i, addr_len);
659 			eaddr[(i * 2) + 0] = val & 0xff;
660 			eaddr[(i * 2) + 1] = val >> 8;
661 		}
662 	}
663 
664 	aprint_normal_dev(sc->sc_dev, "Ethernet address %s\n",
665 	    ether_sprintf(eaddr));
666 
667 	if (sc->re_ldata.re_tx_desc_cnt >
668 	    PAGE_SIZE / sizeof(struct re_desc)) {
669 		sc->re_ldata.re_tx_desc_cnt =
670 		    PAGE_SIZE / sizeof(struct re_desc);
671 	}
672 
673 	aprint_verbose_dev(sc->sc_dev, "using %d tx descriptors\n",
674 	    sc->re_ldata.re_tx_desc_cnt);
675 	KASSERT(RE_NEXT_TX_DESC(sc, RE_TX_DESC_CNT(sc) - 1) == 0);
676 
677 	/* Allocate DMA'able memory for the TX ring */
678 	if ((error = bus_dmamem_alloc(sc->sc_dmat, RE_TX_LIST_SZ(sc),
679 	    RE_RING_ALIGN, 0, &sc->re_ldata.re_tx_listseg, 1,
680 	    &sc->re_ldata.re_tx_listnseg, BUS_DMA_NOWAIT)) != 0) {
681 		aprint_error_dev(sc->sc_dev,
682 		    "can't allocate tx listseg, error = %d\n", error);
683 		goto fail_0;
684 	}
685 
686 	/* Load the map for the TX ring. */
687 	if ((error = bus_dmamem_map(sc->sc_dmat, &sc->re_ldata.re_tx_listseg,
688 	    sc->re_ldata.re_tx_listnseg, RE_TX_LIST_SZ(sc),
689 	    (void **)&sc->re_ldata.re_tx_list,
690 	    BUS_DMA_COHERENT | BUS_DMA_NOWAIT)) != 0) {
691 		aprint_error_dev(sc->sc_dev,
692 		    "can't map tx list, error = %d\n", error);
693 		goto fail_1;
694 	}
695 	memset(sc->re_ldata.re_tx_list, 0, RE_TX_LIST_SZ(sc));
696 
697 	if ((error = bus_dmamap_create(sc->sc_dmat, RE_TX_LIST_SZ(sc), 1,
698 	    RE_TX_LIST_SZ(sc), 0, 0,
699 	    &sc->re_ldata.re_tx_list_map)) != 0) {
700 		aprint_error_dev(sc->sc_dev,
701 		    "can't create tx list map, error = %d\n", error);
702 		goto fail_2;
703 	}
704 
705 
706 	if ((error = bus_dmamap_load(sc->sc_dmat,
707 	    sc->re_ldata.re_tx_list_map, sc->re_ldata.re_tx_list,
708 	    RE_TX_LIST_SZ(sc), NULL, BUS_DMA_NOWAIT)) != 0) {
709 		aprint_error_dev(sc->sc_dev,
710 		    "can't load tx list, error = %d\n", error);
711 		goto fail_3;
712 	}
713 
714 	/* Create DMA maps for TX buffers */
715 	for (i = 0; i < RE_TX_QLEN; i++) {
716 		error = bus_dmamap_create(sc->sc_dmat,
717 		    round_page(IP_MAXPACKET),
718 		    RE_TX_DESC_CNT(sc), RE_TDESC_CMD_FRAGLEN,
719 		    0, 0, &sc->re_ldata.re_txq[i].txq_dmamap);
720 		if (error) {
721 			aprint_error_dev(sc->sc_dev,
722 			    "can't create DMA map for TX\n");
723 			goto fail_4;
724 		}
725 	}
726 
727 	/* Allocate DMA'able memory for the RX ring */
728 	/* XXX see also a comment about RE_RX_DMAMEM_SZ in rtl81x9var.h */
729 	if ((error = bus_dmamem_alloc(sc->sc_dmat,
730 	    RE_RX_DMAMEM_SZ, RE_RING_ALIGN, 0, &sc->re_ldata.re_rx_listseg, 1,
731 	    &sc->re_ldata.re_rx_listnseg, BUS_DMA_NOWAIT)) != 0) {
732 		aprint_error_dev(sc->sc_dev,
733 		    "can't allocate rx listseg, error = %d\n", error);
734 		goto fail_4;
735 	}
736 
737 	/* Load the map for the RX ring. */
738 	if ((error = bus_dmamem_map(sc->sc_dmat, &sc->re_ldata.re_rx_listseg,
739 	    sc->re_ldata.re_rx_listnseg, RE_RX_DMAMEM_SZ,
740 	    (void **)&sc->re_ldata.re_rx_list,
741 	    BUS_DMA_COHERENT | BUS_DMA_NOWAIT)) != 0) {
742 		aprint_error_dev(sc->sc_dev,
743 		    "can't map rx list, error = %d\n", error);
744 		goto fail_5;
745 	}
746 	memset(sc->re_ldata.re_rx_list, 0, RE_RX_DMAMEM_SZ);
747 
748 	if ((error = bus_dmamap_create(sc->sc_dmat,
749 	    RE_RX_DMAMEM_SZ, 1, RE_RX_DMAMEM_SZ, 0, 0,
750 	    &sc->re_ldata.re_rx_list_map)) != 0) {
751 		aprint_error_dev(sc->sc_dev,
752 		    "can't create rx list map, error = %d\n", error);
753 		goto fail_6;
754 	}
755 
756 	if ((error = bus_dmamap_load(sc->sc_dmat,
757 	    sc->re_ldata.re_rx_list_map, sc->re_ldata.re_rx_list,
758 	    RE_RX_DMAMEM_SZ, NULL, BUS_DMA_NOWAIT)) != 0) {
759 		aprint_error_dev(sc->sc_dev,
760 		    "can't load rx list, error = %d\n", error);
761 		goto fail_7;
762 	}
763 
764 	/* Create DMA maps for RX buffers */
765 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
766 		error = bus_dmamap_create(sc->sc_dmat, MCLBYTES, 1, MCLBYTES,
767 		    0, 0, &sc->re_ldata.re_rxsoft[i].rxs_dmamap);
768 		if (error) {
769 			aprint_error_dev(sc->sc_dev,
770 			    "can't create DMA map for RX\n");
771 			goto fail_8;
772 		}
773 	}
774 
775 	/*
776 	 * Record interface as attached. From here, we should not fail.
777 	 */
778 	sc->sc_flags |= RTK_ATTACHED;
779 
780 	ifp = &sc->ethercom.ec_if;
781 	ifp->if_softc = sc;
782 	strlcpy(ifp->if_xname, device_xname(sc->sc_dev), IFNAMSIZ);
783 	ifp->if_mtu = ETHERMTU;
784 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
785 	ifp->if_ioctl = re_ioctl;
786 	sc->ethercom.ec_capabilities |=
787 	    ETHERCAP_VLAN_MTU | ETHERCAP_VLAN_HWTAGGING;
788 	ifp->if_start = re_start;
789 	ifp->if_stop = re_stop;
790 
791 	/*
792 	 * IFCAP_CSUM_IPv4_Tx on re(4) is broken for small packets,
793 	 * so we have a workaround to handle the bug by padding
794 	 * such packets manually.
795 	 */
796 	ifp->if_capabilities |=
797 	    IFCAP_CSUM_IPv4_Tx | IFCAP_CSUM_IPv4_Rx |
798 	    IFCAP_CSUM_TCPv4_Tx | IFCAP_CSUM_TCPv4_Rx |
799 	    IFCAP_CSUM_UDPv4_Tx | IFCAP_CSUM_UDPv4_Rx |
800 	    IFCAP_TSOv4;
801 
802 	/*
803 	 * XXX
804 	 * Still have no idea how to make TSO work on 8168C, 8168CP,
805 	 * 8102E, 8111C and 8111CP.
806 	 */
807 	if ((sc->sc_quirk & RTKQ_DESCV2) != 0)
808 		ifp->if_capabilities &= ~IFCAP_TSOv4;
809 
810 	ifp->if_watchdog = re_watchdog;
811 	ifp->if_init = re_init;
812 	ifp->if_snd.ifq_maxlen = RE_IFQ_MAXLEN;
813 	ifp->if_capenable = ifp->if_capabilities;
814 	IFQ_SET_READY(&ifp->if_snd);
815 
816 	callout_init(&sc->rtk_tick_ch, 0);
817 
818 	/* Do MII setup */
819 	sc->mii.mii_ifp = ifp;
820 	sc->mii.mii_readreg = re_miibus_readreg;
821 	sc->mii.mii_writereg = re_miibus_writereg;
822 	sc->mii.mii_statchg = re_miibus_statchg;
823 	sc->ethercom.ec_mii = &sc->mii;
824 	ifmedia_init(&sc->mii.mii_media, IFM_IMASK, ether_mediachange,
825 	    ether_mediastatus);
826 	mii_attach(sc->sc_dev, &sc->mii, 0xffffffff, MII_PHY_ANY,
827 	    MII_OFFSET_ANY, 0);
828 	ifmedia_set(&sc->mii.mii_media, IFM_ETHER | IFM_AUTO);
829 
830 	/*
831 	 * Call MI attach routine.
832 	 */
833 	if_attach(ifp);
834 	ether_ifattach(ifp, eaddr);
835 
836 	if (pmf_device_register(sc->sc_dev, NULL, NULL))
837 		pmf_class_network_register(sc->sc_dev, ifp);
838 	else
839 		aprint_error_dev(sc->sc_dev,
840 		    "couldn't establish power handler\n");
841 
842 	return;
843 
844  fail_8:
845 	/* Destroy DMA maps for RX buffers. */
846 	for (i = 0; i < RE_RX_DESC_CNT; i++)
847 		if (sc->re_ldata.re_rxsoft[i].rxs_dmamap != NULL)
848 			bus_dmamap_destroy(sc->sc_dmat,
849 			    sc->re_ldata.re_rxsoft[i].rxs_dmamap);
850 
851 	/* Free DMA'able memory for the RX ring. */
852 	bus_dmamap_unload(sc->sc_dmat, sc->re_ldata.re_rx_list_map);
853  fail_7:
854 	bus_dmamap_destroy(sc->sc_dmat, sc->re_ldata.re_rx_list_map);
855  fail_6:
856 	bus_dmamem_unmap(sc->sc_dmat,
857 	    (void *)sc->re_ldata.re_rx_list, RE_RX_DMAMEM_SZ);
858  fail_5:
859 	bus_dmamem_free(sc->sc_dmat,
860 	    &sc->re_ldata.re_rx_listseg, sc->re_ldata.re_rx_listnseg);
861 
862  fail_4:
863 	/* Destroy DMA maps for TX buffers. */
864 	for (i = 0; i < RE_TX_QLEN; i++)
865 		if (sc->re_ldata.re_txq[i].txq_dmamap != NULL)
866 			bus_dmamap_destroy(sc->sc_dmat,
867 			    sc->re_ldata.re_txq[i].txq_dmamap);
868 
869 	/* Free DMA'able memory for the TX ring. */
870 	bus_dmamap_unload(sc->sc_dmat, sc->re_ldata.re_tx_list_map);
871  fail_3:
872 	bus_dmamap_destroy(sc->sc_dmat, sc->re_ldata.re_tx_list_map);
873  fail_2:
874 	bus_dmamem_unmap(sc->sc_dmat,
875 	    (void *)sc->re_ldata.re_tx_list, RE_TX_LIST_SZ(sc));
876  fail_1:
877 	bus_dmamem_free(sc->sc_dmat,
878 	    &sc->re_ldata.re_tx_listseg, sc->re_ldata.re_tx_listnseg);
879  fail_0:
880 	return;
881 }
882 
883 
884 /*
885  * re_activate:
886  *     Handle device activation/deactivation requests.
887  */
888 int
889 re_activate(device_t self, enum devact act)
890 {
891 	struct rtk_softc *sc = device_private(self);
892 
893 	switch (act) {
894 	case DVACT_DEACTIVATE:
895 		if_deactivate(&sc->ethercom.ec_if);
896 		return 0;
897 	default:
898 		return EOPNOTSUPP;
899 	}
900 }
901 
902 /*
903  * re_detach:
904  *     Detach a rtk interface.
905  */
906 int
907 re_detach(struct rtk_softc *sc)
908 {
909 	struct ifnet *ifp = &sc->ethercom.ec_if;
910 	int i;
911 
912 	/*
913 	 * Succeed now if there isn't any work to do.
914 	 */
915 	if ((sc->sc_flags & RTK_ATTACHED) == 0)
916 		return 0;
917 
918 	/* Unhook our tick handler. */
919 	callout_stop(&sc->rtk_tick_ch);
920 
921 	/* Detach all PHYs. */
922 	mii_detach(&sc->mii, MII_PHY_ANY, MII_OFFSET_ANY);
923 
924 	/* Delete all remaining media. */
925 	ifmedia_delete_instance(&sc->mii.mii_media, IFM_INST_ANY);
926 
927 	ether_ifdetach(ifp);
928 	if_detach(ifp);
929 
930 	/* Destroy DMA maps for RX buffers. */
931 	for (i = 0; i < RE_RX_DESC_CNT; i++)
932 		if (sc->re_ldata.re_rxsoft[i].rxs_dmamap != NULL)
933 			bus_dmamap_destroy(sc->sc_dmat,
934 			    sc->re_ldata.re_rxsoft[i].rxs_dmamap);
935 
936 	/* Free DMA'able memory for the RX ring. */
937 	bus_dmamap_unload(sc->sc_dmat, sc->re_ldata.re_rx_list_map);
938 	bus_dmamap_destroy(sc->sc_dmat, sc->re_ldata.re_rx_list_map);
939 	bus_dmamem_unmap(sc->sc_dmat,
940 	    (void *)sc->re_ldata.re_rx_list, RE_RX_DMAMEM_SZ);
941 	bus_dmamem_free(sc->sc_dmat,
942 	    &sc->re_ldata.re_rx_listseg, sc->re_ldata.re_rx_listnseg);
943 
944 	/* Destroy DMA maps for TX buffers. */
945 	for (i = 0; i < RE_TX_QLEN; i++)
946 		if (sc->re_ldata.re_txq[i].txq_dmamap != NULL)
947 			bus_dmamap_destroy(sc->sc_dmat,
948 			    sc->re_ldata.re_txq[i].txq_dmamap);
949 
950 	/* Free DMA'able memory for the TX ring. */
951 	bus_dmamap_unload(sc->sc_dmat, sc->re_ldata.re_tx_list_map);
952 	bus_dmamap_destroy(sc->sc_dmat, sc->re_ldata.re_tx_list_map);
953 	bus_dmamem_unmap(sc->sc_dmat,
954 	    (void *)sc->re_ldata.re_tx_list, RE_TX_LIST_SZ(sc));
955 	bus_dmamem_free(sc->sc_dmat,
956 	    &sc->re_ldata.re_tx_listseg, sc->re_ldata.re_tx_listnseg);
957 
958 	pmf_device_deregister(sc->sc_dev);
959 
960 	return 0;
961 }
962 
963 /*
964  * re_enable:
965  *     Enable the RTL81X9 chip.
966  */
967 static int
968 re_enable(struct rtk_softc *sc)
969 {
970 
971 	if (RTK_IS_ENABLED(sc) == 0 && sc->sc_enable != NULL) {
972 		if ((*sc->sc_enable)(sc) != 0) {
973 			printf("%s: device enable failed\n",
974 			    device_xname(sc->sc_dev));
975 			return EIO;
976 		}
977 		sc->sc_flags |= RTK_ENABLED;
978 	}
979 	return 0;
980 }
981 
982 /*
983  * re_disable:
984  *     Disable the RTL81X9 chip.
985  */
986 static void
987 re_disable(struct rtk_softc *sc)
988 {
989 
990 	if (RTK_IS_ENABLED(sc) && sc->sc_disable != NULL) {
991 		(*sc->sc_disable)(sc);
992 		sc->sc_flags &= ~RTK_ENABLED;
993 	}
994 }
995 
996 static int
997 re_newbuf(struct rtk_softc *sc, int idx, struct mbuf *m)
998 {
999 	struct mbuf *n = NULL;
1000 	bus_dmamap_t map;
1001 	struct re_desc *d;
1002 	struct re_rxsoft *rxs;
1003 	uint32_t cmdstat;
1004 	int error;
1005 
1006 	if (m == NULL) {
1007 		MGETHDR(n, M_DONTWAIT, MT_DATA);
1008 		if (n == NULL)
1009 			return ENOBUFS;
1010 
1011 		MCLGET(n, M_DONTWAIT);
1012 		if ((n->m_flags & M_EXT) == 0) {
1013 			m_freem(n);
1014 			return ENOBUFS;
1015 		}
1016 		m = n;
1017 	} else
1018 		m->m_data = m->m_ext.ext_buf;
1019 
1020 	/*
1021 	 * Initialize mbuf length fields and fixup
1022 	 * alignment so that the frame payload is
1023 	 * longword aligned.
1024 	 */
1025 	m->m_len = m->m_pkthdr.len = MCLBYTES - RE_ETHER_ALIGN;
1026 	m->m_data += RE_ETHER_ALIGN;
1027 
1028 	rxs = &sc->re_ldata.re_rxsoft[idx];
1029 	map = rxs->rxs_dmamap;
1030 	error = bus_dmamap_load_mbuf(sc->sc_dmat, map, m,
1031 	    BUS_DMA_READ|BUS_DMA_NOWAIT);
1032 
1033 	if (error)
1034 		goto out;
1035 
1036 	bus_dmamap_sync(sc->sc_dmat, map, 0, map->dm_mapsize,
1037 	    BUS_DMASYNC_PREREAD);
1038 
1039 	d = &sc->re_ldata.re_rx_list[idx];
1040 #ifdef DIAGNOSTIC
1041 	RE_RXDESCSYNC(sc, idx, BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1042 	cmdstat = le32toh(d->re_cmdstat);
1043 	RE_RXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD);
1044 	if (cmdstat & RE_RDESC_STAT_OWN) {
1045 		panic("%s: tried to map busy RX descriptor",
1046 		    device_xname(sc->sc_dev));
1047 	}
1048 #endif
1049 
1050 	rxs->rxs_mbuf = m;
1051 
1052 	d->re_vlanctl = 0;
1053 	cmdstat = map->dm_segs[0].ds_len;
1054 	if (idx == (RE_RX_DESC_CNT - 1))
1055 		cmdstat |= RE_RDESC_CMD_EOR;
1056 	re_set_bufaddr(d, map->dm_segs[0].ds_addr);
1057 	d->re_cmdstat = htole32(cmdstat);
1058 	RE_RXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1059 	cmdstat |= RE_RDESC_CMD_OWN;
1060 	d->re_cmdstat = htole32(cmdstat);
1061 	RE_RXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1062 
1063 	return 0;
1064  out:
1065 	if (n != NULL)
1066 		m_freem(n);
1067 	return ENOMEM;
1068 }
1069 
1070 static int
1071 re_tx_list_init(struct rtk_softc *sc)
1072 {
1073 	int i;
1074 
1075 	memset(sc->re_ldata.re_tx_list, 0, RE_TX_LIST_SZ(sc));
1076 	for (i = 0; i < RE_TX_QLEN; i++) {
1077 		sc->re_ldata.re_txq[i].txq_mbuf = NULL;
1078 	}
1079 
1080 	bus_dmamap_sync(sc->sc_dmat,
1081 	    sc->re_ldata.re_tx_list_map, 0,
1082 	    sc->re_ldata.re_tx_list_map->dm_mapsize,
1083 	    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1084 	sc->re_ldata.re_txq_prodidx = 0;
1085 	sc->re_ldata.re_txq_considx = 0;
1086 	sc->re_ldata.re_txq_free = RE_TX_QLEN;
1087 	sc->re_ldata.re_tx_free = RE_TX_DESC_CNT(sc);
1088 	sc->re_ldata.re_tx_nextfree = 0;
1089 
1090 	return 0;
1091 }
1092 
1093 static int
1094 re_rx_list_init(struct rtk_softc *sc)
1095 {
1096 	int i;
1097 
1098 	memset(sc->re_ldata.re_rx_list, 0, RE_RX_LIST_SZ);
1099 
1100 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
1101 		if (re_newbuf(sc, i, NULL) == ENOBUFS)
1102 			return ENOBUFS;
1103 	}
1104 
1105 	sc->re_ldata.re_rx_prodidx = 0;
1106 	sc->re_head = sc->re_tail = NULL;
1107 
1108 	return 0;
1109 }
1110 
1111 /*
1112  * RX handler for C+ and 8169. For the gigE chips, we support
1113  * the reception of jumbo frames that have been fragmented
1114  * across multiple 2K mbuf cluster buffers.
1115  */
1116 static void
1117 re_rxeof(struct rtk_softc *sc)
1118 {
1119 	struct mbuf *m;
1120 	struct ifnet *ifp;
1121 	int i, total_len;
1122 	struct re_desc *cur_rx;
1123 	struct re_rxsoft *rxs;
1124 	uint32_t rxstat, rxvlan;
1125 
1126 	ifp = &sc->ethercom.ec_if;
1127 
1128 	for (i = sc->re_ldata.re_rx_prodidx;; i = RE_NEXT_RX_DESC(sc, i)) {
1129 		cur_rx = &sc->re_ldata.re_rx_list[i];
1130 		RE_RXDESCSYNC(sc, i,
1131 		    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1132 		rxstat = le32toh(cur_rx->re_cmdstat);
1133 		rxvlan = le32toh(cur_rx->re_vlanctl);
1134 		RE_RXDESCSYNC(sc, i, BUS_DMASYNC_PREREAD);
1135 		if ((rxstat & RE_RDESC_STAT_OWN) != 0) {
1136 			break;
1137 		}
1138 		total_len = rxstat & sc->re_rxlenmask;
1139 		rxs = &sc->re_ldata.re_rxsoft[i];
1140 		m = rxs->rxs_mbuf;
1141 
1142 		/* Invalidate the RX mbuf and unload its map */
1143 
1144 		bus_dmamap_sync(sc->sc_dmat,
1145 		    rxs->rxs_dmamap, 0, rxs->rxs_dmamap->dm_mapsize,
1146 		    BUS_DMASYNC_POSTREAD);
1147 		bus_dmamap_unload(sc->sc_dmat, rxs->rxs_dmamap);
1148 
1149 		if ((rxstat & RE_RDESC_STAT_EOF) == 0) {
1150 			m->m_len = MCLBYTES - RE_ETHER_ALIGN;
1151 			if (sc->re_head == NULL)
1152 				sc->re_head = sc->re_tail = m;
1153 			else {
1154 				m->m_flags &= ~M_PKTHDR;
1155 				sc->re_tail->m_next = m;
1156 				sc->re_tail = m;
1157 			}
1158 			re_newbuf(sc, i, NULL);
1159 			continue;
1160 		}
1161 
1162 		/*
1163 		 * NOTE: for the 8139C+, the frame length field
1164 		 * is always 12 bits in size, but for the gigE chips,
1165 		 * it is 13 bits (since the max RX frame length is 16K).
1166 		 * Unfortunately, all 32 bits in the status word
1167 		 * were already used, so to make room for the extra
1168 		 * length bit, RealTek took out the 'frame alignment
1169 		 * error' bit and shifted the other status bits
1170 		 * over one slot. The OWN, EOR, FS and LS bits are
1171 		 * still in the same places. We have already extracted
1172 		 * the frame length and checked the OWN bit, so rather
1173 		 * than using an alternate bit mapping, we shift the
1174 		 * status bits one space to the right so we can evaluate
1175 		 * them using the 8169 status as though it was in the
1176 		 * same format as that of the 8139C+.
1177 		 */
1178 		if ((sc->sc_quirk & RTKQ_8139CPLUS) == 0)
1179 			rxstat >>= 1;
1180 
1181 		if (__predict_false((rxstat & RE_RDESC_STAT_RXERRSUM) != 0)) {
1182 #ifdef RE_DEBUG
1183 			printf("%s: RX error (rxstat = 0x%08x)",
1184 			    device_xname(sc->sc_dev), rxstat);
1185 			if (rxstat & RE_RDESC_STAT_FRALIGN)
1186 				printf(", frame alignment error");
1187 			if (rxstat & RE_RDESC_STAT_BUFOFLOW)
1188 				printf(", out of buffer space");
1189 			if (rxstat & RE_RDESC_STAT_FIFOOFLOW)
1190 				printf(", FIFO overrun");
1191 			if (rxstat & RE_RDESC_STAT_GIANT)
1192 				printf(", giant packet");
1193 			if (rxstat & RE_RDESC_STAT_RUNT)
1194 				printf(", runt packet");
1195 			if (rxstat & RE_RDESC_STAT_CRCERR)
1196 				printf(", CRC error");
1197 			printf("\n");
1198 #endif
1199 			ifp->if_ierrors++;
1200 			/*
1201 			 * If this is part of a multi-fragment packet,
1202 			 * discard all the pieces.
1203 			 */
1204 			if (sc->re_head != NULL) {
1205 				m_freem(sc->re_head);
1206 				sc->re_head = sc->re_tail = NULL;
1207 			}
1208 			re_newbuf(sc, i, m);
1209 			continue;
1210 		}
1211 
1212 		/*
1213 		 * If allocating a replacement mbuf fails,
1214 		 * reload the current one.
1215 		 */
1216 
1217 		if (__predict_false(re_newbuf(sc, i, NULL) != 0)) {
1218 			ifp->if_ierrors++;
1219 			if (sc->re_head != NULL) {
1220 				m_freem(sc->re_head);
1221 				sc->re_head = sc->re_tail = NULL;
1222 			}
1223 			re_newbuf(sc, i, m);
1224 			continue;
1225 		}
1226 
1227 		if (sc->re_head != NULL) {
1228 			m->m_len = total_len % (MCLBYTES - RE_ETHER_ALIGN);
1229 			/*
1230 			 * Special case: if there's 4 bytes or less
1231 			 * in this buffer, the mbuf can be discarded:
1232 			 * the last 4 bytes is the CRC, which we don't
1233 			 * care about anyway.
1234 			 */
1235 			if (m->m_len <= ETHER_CRC_LEN) {
1236 				sc->re_tail->m_len -=
1237 				    (ETHER_CRC_LEN - m->m_len);
1238 				m_freem(m);
1239 			} else {
1240 				m->m_len -= ETHER_CRC_LEN;
1241 				m->m_flags &= ~M_PKTHDR;
1242 				sc->re_tail->m_next = m;
1243 			}
1244 			m = sc->re_head;
1245 			sc->re_head = sc->re_tail = NULL;
1246 			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
1247 		} else
1248 			m->m_pkthdr.len = m->m_len =
1249 			    (total_len - ETHER_CRC_LEN);
1250 
1251 		ifp->if_ipackets++;
1252 		m->m_pkthdr.rcvif = ifp;
1253 
1254 		/* Do RX checksumming */
1255 		if ((sc->sc_quirk & RTKQ_DESCV2) == 0) {
1256 			/* Check IP header checksum */
1257 			if ((rxstat & RE_RDESC_STAT_PROTOID) != 0) {
1258 				m->m_pkthdr.csum_flags |= M_CSUM_IPv4;
1259 				if (rxstat & RE_RDESC_STAT_IPSUMBAD)
1260 					m->m_pkthdr.csum_flags |=
1261 					    M_CSUM_IPv4_BAD;
1262 
1263 				/* Check TCP/UDP checksum */
1264 				if (RE_TCPPKT(rxstat)) {
1265 					m->m_pkthdr.csum_flags |= M_CSUM_TCPv4;
1266 					if (rxstat & RE_RDESC_STAT_TCPSUMBAD)
1267 						m->m_pkthdr.csum_flags |=
1268 						    M_CSUM_TCP_UDP_BAD;
1269 				} else if (RE_UDPPKT(rxstat)) {
1270 					m->m_pkthdr.csum_flags |= M_CSUM_UDPv4;
1271 					if (rxstat & RE_RDESC_STAT_UDPSUMBAD)
1272 						m->m_pkthdr.csum_flags |=
1273 						    M_CSUM_TCP_UDP_BAD;
1274 				}
1275 			}
1276 		} else {
1277 			/* Check IPv4 header checksum */
1278 			if ((rxvlan & RE_RDESC_VLANCTL_IPV4) != 0) {
1279 				m->m_pkthdr.csum_flags |= M_CSUM_IPv4;
1280 				if (rxstat & RE_RDESC_STAT_IPSUMBAD)
1281 					m->m_pkthdr.csum_flags |=
1282 					    M_CSUM_IPv4_BAD;
1283 
1284 				/* Check TCPv4/UDPv4 checksum */
1285 				if (RE_TCPPKT(rxstat)) {
1286 					m->m_pkthdr.csum_flags |= M_CSUM_TCPv4;
1287 					if (rxstat & RE_RDESC_STAT_TCPSUMBAD)
1288 						m->m_pkthdr.csum_flags |=
1289 						    M_CSUM_TCP_UDP_BAD;
1290 				} else if (RE_UDPPKT(rxstat)) {
1291 					m->m_pkthdr.csum_flags |= M_CSUM_UDPv4;
1292 					if (rxstat & RE_RDESC_STAT_UDPSUMBAD)
1293 						m->m_pkthdr.csum_flags |=
1294 						    M_CSUM_TCP_UDP_BAD;
1295 				}
1296 			}
1297 			/* XXX Check TCPv6/UDPv6 checksum? */
1298 		}
1299 
1300 		if (rxvlan & RE_RDESC_VLANCTL_TAG) {
1301 			VLAN_INPUT_TAG(ifp, m,
1302 			     bswap16(rxvlan & RE_RDESC_VLANCTL_DATA),
1303 			     continue);
1304 		}
1305 #if NBPFILTER > 0
1306 		if (ifp->if_bpf)
1307 			bpf_mtap(ifp->if_bpf, m);
1308 #endif
1309 		(*ifp->if_input)(ifp, m);
1310 	}
1311 
1312 	sc->re_ldata.re_rx_prodidx = i;
1313 }
1314 
1315 static void
1316 re_txeof(struct rtk_softc *sc)
1317 {
1318 	struct ifnet *ifp;
1319 	struct re_txq *txq;
1320 	uint32_t txstat;
1321 	int idx, descidx;
1322 
1323 	ifp = &sc->ethercom.ec_if;
1324 
1325 	for (idx = sc->re_ldata.re_txq_considx;
1326 	    sc->re_ldata.re_txq_free < RE_TX_QLEN;
1327 	    idx = RE_NEXT_TXQ(sc, idx), sc->re_ldata.re_txq_free++) {
1328 		txq = &sc->re_ldata.re_txq[idx];
1329 		KASSERT(txq->txq_mbuf != NULL);
1330 
1331 		descidx = txq->txq_descidx;
1332 		RE_TXDESCSYNC(sc, descidx,
1333 		    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1334 		txstat =
1335 		    le32toh(sc->re_ldata.re_tx_list[descidx].re_cmdstat);
1336 		RE_TXDESCSYNC(sc, descidx, BUS_DMASYNC_PREREAD);
1337 		KASSERT((txstat & RE_TDESC_CMD_EOF) != 0);
1338 		if (txstat & RE_TDESC_CMD_OWN) {
1339 			break;
1340 		}
1341 
1342 		sc->re_ldata.re_tx_free += txq->txq_nsegs;
1343 		KASSERT(sc->re_ldata.re_tx_free <= RE_TX_DESC_CNT(sc));
1344 		bus_dmamap_sync(sc->sc_dmat, txq->txq_dmamap,
1345 		    0, txq->txq_dmamap->dm_mapsize, BUS_DMASYNC_POSTWRITE);
1346 		bus_dmamap_unload(sc->sc_dmat, txq->txq_dmamap);
1347 		m_freem(txq->txq_mbuf);
1348 		txq->txq_mbuf = NULL;
1349 
1350 		if (txstat & (RE_TDESC_STAT_EXCESSCOL | RE_TDESC_STAT_COLCNT))
1351 			ifp->if_collisions++;
1352 		if (txstat & RE_TDESC_STAT_TXERRSUM)
1353 			ifp->if_oerrors++;
1354 		else
1355 			ifp->if_opackets++;
1356 	}
1357 
1358 	sc->re_ldata.re_txq_considx = idx;
1359 
1360 	if (sc->re_ldata.re_txq_free > RE_NTXDESC_RSVD)
1361 		ifp->if_flags &= ~IFF_OACTIVE;
1362 
1363 	/*
1364 	 * If not all descriptors have been released reaped yet,
1365 	 * reload the timer so that we will eventually get another
1366 	 * interrupt that will cause us to re-enter this routine.
1367 	 * This is done in case the transmitter has gone idle.
1368 	 */
1369 	if (sc->re_ldata.re_txq_free < RE_TX_QLEN) {
1370 		CSR_WRITE_4(sc, RTK_TIMERCNT, 1);
1371 		if ((sc->sc_quirk & RTKQ_PCIE) != 0) {
1372 			/*
1373 			 * Some chips will ignore a second TX request
1374 			 * issued while an existing transmission is in
1375 			 * progress. If the transmitter goes idle but
1376 			 * there are still packets waiting to be sent,
1377 			 * we need to restart the channel here to flush
1378 			 * them out. This only seems to be required with
1379 			 * the PCIe devices.
1380 			 */
1381 			CSR_WRITE_1(sc, RTK_GTXSTART, RTK_TXSTART_START);
1382 		}
1383 	} else
1384 		ifp->if_timer = 0;
1385 }
1386 
1387 static void
1388 re_tick(void *arg)
1389 {
1390 	struct rtk_softc *sc = arg;
1391 	int s;
1392 
1393 	/* XXX: just return for 8169S/8110S with rev 2 or newer phy */
1394 	s = splnet();
1395 
1396 	mii_tick(&sc->mii);
1397 	splx(s);
1398 
1399 	callout_reset(&sc->rtk_tick_ch, hz, re_tick, sc);
1400 }
1401 
1402 int
1403 re_intr(void *arg)
1404 {
1405 	struct rtk_softc *sc = arg;
1406 	struct ifnet *ifp;
1407 	uint16_t status;
1408 	int handled = 0;
1409 
1410 	if (!device_has_power(sc->sc_dev))
1411 		return 0;
1412 
1413 	ifp = &sc->ethercom.ec_if;
1414 
1415 	if ((ifp->if_flags & IFF_UP) == 0)
1416 		return 0;
1417 
1418 	for (;;) {
1419 
1420 		status = CSR_READ_2(sc, RTK_ISR);
1421 		/* If the card has gone away the read returns 0xffff. */
1422 		if (status == 0xffff)
1423 			break;
1424 		if (status) {
1425 			handled = 1;
1426 			CSR_WRITE_2(sc, RTK_ISR, status);
1427 		}
1428 
1429 		if ((status & RTK_INTRS_CPLUS) == 0)
1430 			break;
1431 
1432 		if (status & (RTK_ISR_RX_OK | RTK_ISR_RX_ERR))
1433 			re_rxeof(sc);
1434 
1435 		if (status & (RTK_ISR_TIMEOUT_EXPIRED | RTK_ISR_TX_ERR |
1436 		    RTK_ISR_TX_DESC_UNAVAIL))
1437 			re_txeof(sc);
1438 
1439 		if (status & RTK_ISR_SYSTEM_ERR) {
1440 			re_init(ifp);
1441 		}
1442 
1443 		if (status & RTK_ISR_LINKCHG) {
1444 			callout_stop(&sc->rtk_tick_ch);
1445 			re_tick(sc);
1446 		}
1447 	}
1448 
1449 	if (handled && !IFQ_IS_EMPTY(&ifp->if_snd))
1450 		re_start(ifp);
1451 
1452 	return handled;
1453 }
1454 
1455 
1456 
1457 /*
1458  * Main transmit routine for C+ and gigE NICs.
1459  */
1460 
1461 static void
1462 re_start(struct ifnet *ifp)
1463 {
1464 	struct rtk_softc *sc;
1465 	struct mbuf *m;
1466 	bus_dmamap_t map;
1467 	struct re_txq *txq;
1468 	struct re_desc *d;
1469 	struct m_tag *mtag;
1470 	uint32_t cmdstat, re_flags, vlanctl;
1471 	int ofree, idx, error, nsegs, seg;
1472 	int startdesc, curdesc, lastdesc;
1473 	bool pad;
1474 
1475 	sc = ifp->if_softc;
1476 	ofree = sc->re_ldata.re_txq_free;
1477 
1478 	for (idx = sc->re_ldata.re_txq_prodidx;; idx = RE_NEXT_TXQ(sc, idx)) {
1479 
1480 		IFQ_POLL(&ifp->if_snd, m);
1481 		if (m == NULL)
1482 			break;
1483 
1484 		if (sc->re_ldata.re_txq_free == 0 ||
1485 		    sc->re_ldata.re_tx_free == 0) {
1486 			/* no more free slots left */
1487 			ifp->if_flags |= IFF_OACTIVE;
1488 			break;
1489 		}
1490 
1491 		/*
1492 		 * Set up checksum offload. Note: checksum offload bits must
1493 		 * appear in all descriptors of a multi-descriptor transmit
1494 		 * attempt. (This is according to testing done with an 8169
1495 		 * chip. I'm not sure if this is a requirement or a bug.)
1496 		 */
1497 
1498 		vlanctl = 0;
1499 		if ((m->m_pkthdr.csum_flags & M_CSUM_TSOv4) != 0) {
1500 			uint32_t segsz = m->m_pkthdr.segsz;
1501 
1502 			re_flags = RE_TDESC_CMD_LGSEND |
1503 			    (segsz << RE_TDESC_CMD_MSSVAL_SHIFT);
1504 		} else {
1505 			/*
1506 			 * set RE_TDESC_CMD_IPCSUM if any checksum offloading
1507 			 * is requested.  otherwise, RE_TDESC_CMD_TCPCSUM/
1508 			 * RE_TDESC_CMD_UDPCSUM doesn't make effects.
1509 			 */
1510 			re_flags = 0;
1511 			if ((m->m_pkthdr.csum_flags &
1512 			    (M_CSUM_IPv4 | M_CSUM_TCPv4 | M_CSUM_UDPv4))
1513 			    != 0) {
1514 				if ((sc->sc_quirk & RTKQ_DESCV2) == 0) {
1515 					re_flags |= RE_TDESC_CMD_IPCSUM;
1516 					if (m->m_pkthdr.csum_flags &
1517 					    M_CSUM_TCPv4) {
1518 						re_flags |=
1519 						    RE_TDESC_CMD_TCPCSUM;
1520 					} else if (m->m_pkthdr.csum_flags &
1521 					    M_CSUM_UDPv4) {
1522 						re_flags |=
1523 						    RE_TDESC_CMD_UDPCSUM;
1524 					}
1525 				} else {
1526 					vlanctl |= RE_TDESC_VLANCTL_IPCSUM;
1527 					if (m->m_pkthdr.csum_flags &
1528 					    M_CSUM_TCPv4) {
1529 						vlanctl |=
1530 						    RE_TDESC_VLANCTL_TCPCSUM;
1531 					} else if (m->m_pkthdr.csum_flags &
1532 					    M_CSUM_UDPv4) {
1533 						vlanctl |=
1534 						    RE_TDESC_VLANCTL_UDPCSUM;
1535 					}
1536 				}
1537 			}
1538 		}
1539 
1540 		txq = &sc->re_ldata.re_txq[idx];
1541 		map = txq->txq_dmamap;
1542 		error = bus_dmamap_load_mbuf(sc->sc_dmat, map, m,
1543 		    BUS_DMA_WRITE|BUS_DMA_NOWAIT);
1544 
1545 		if (__predict_false(error)) {
1546 			/* XXX try to defrag if EFBIG? */
1547 			printf("%s: can't map mbuf (error %d)\n",
1548 			    device_xname(sc->sc_dev), error);
1549 
1550 			IFQ_DEQUEUE(&ifp->if_snd, m);
1551 			m_freem(m);
1552 			ifp->if_oerrors++;
1553 			continue;
1554 		}
1555 
1556 		nsegs = map->dm_nsegs;
1557 		pad = false;
1558 		if (__predict_false(m->m_pkthdr.len <= RE_IP4CSUMTX_PADLEN &&
1559 		    (re_flags & RE_TDESC_CMD_IPCSUM) != 0 &&
1560 		    (sc->sc_quirk & RTKQ_DESCV2) == 0)) {
1561 			pad = true;
1562 			nsegs++;
1563 		}
1564 
1565 		if (nsegs > sc->re_ldata.re_tx_free) {
1566 			/*
1567 			 * Not enough free descriptors to transmit this packet.
1568 			 */
1569 			ifp->if_flags |= IFF_OACTIVE;
1570 			bus_dmamap_unload(sc->sc_dmat, map);
1571 			break;
1572 		}
1573 
1574 		IFQ_DEQUEUE(&ifp->if_snd, m);
1575 
1576 		/*
1577 		 * Make sure that the caches are synchronized before we
1578 		 * ask the chip to start DMA for the packet data.
1579 		 */
1580 		bus_dmamap_sync(sc->sc_dmat, map, 0, map->dm_mapsize,
1581 		    BUS_DMASYNC_PREWRITE);
1582 
1583 		/*
1584 		 * Set up hardware VLAN tagging. Note: vlan tag info must
1585 		 * appear in all descriptors of a multi-descriptor
1586 		 * transmission attempt.
1587 		 */
1588 		if ((mtag = VLAN_OUTPUT_TAG(&sc->ethercom, m)) != NULL)
1589 			vlanctl |= bswap16(VLAN_TAG_VALUE(mtag)) |
1590 			    RE_TDESC_VLANCTL_TAG;
1591 
1592 		/*
1593 		 * Map the segment array into descriptors.
1594 		 * Note that we set the start-of-frame and
1595 		 * end-of-frame markers for either TX or RX,
1596 		 * but they really only have meaning in the TX case.
1597 		 * (In the RX case, it's the chip that tells us
1598 		 *  where packets begin and end.)
1599 		 * We also keep track of the end of the ring
1600 		 * and set the end-of-ring bits as needed,
1601 		 * and we set the ownership bits in all except
1602 		 * the very first descriptor. (The caller will
1603 		 * set this descriptor later when it start
1604 		 * transmission or reception.)
1605 		 */
1606 		curdesc = startdesc = sc->re_ldata.re_tx_nextfree;
1607 		lastdesc = -1;
1608 		for (seg = 0; seg < map->dm_nsegs;
1609 		    seg++, curdesc = RE_NEXT_TX_DESC(sc, curdesc)) {
1610 			d = &sc->re_ldata.re_tx_list[curdesc];
1611 #ifdef DIAGNOSTIC
1612 			RE_TXDESCSYNC(sc, curdesc,
1613 			    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1614 			cmdstat = le32toh(d->re_cmdstat);
1615 			RE_TXDESCSYNC(sc, curdesc, BUS_DMASYNC_PREREAD);
1616 			if (cmdstat & RE_TDESC_STAT_OWN) {
1617 				panic("%s: tried to map busy TX descriptor",
1618 				    device_xname(sc->sc_dev));
1619 			}
1620 #endif
1621 
1622 			d->re_vlanctl = htole32(vlanctl);
1623 			re_set_bufaddr(d, map->dm_segs[seg].ds_addr);
1624 			cmdstat = re_flags | map->dm_segs[seg].ds_len;
1625 			if (seg == 0)
1626 				cmdstat |= RE_TDESC_CMD_SOF;
1627 			else
1628 				cmdstat |= RE_TDESC_CMD_OWN;
1629 			if (curdesc == (RE_TX_DESC_CNT(sc) - 1))
1630 				cmdstat |= RE_TDESC_CMD_EOR;
1631 			if (seg == nsegs - 1) {
1632 				cmdstat |= RE_TDESC_CMD_EOF;
1633 				lastdesc = curdesc;
1634 			}
1635 			d->re_cmdstat = htole32(cmdstat);
1636 			RE_TXDESCSYNC(sc, curdesc,
1637 			    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1638 		}
1639 		if (__predict_false(pad)) {
1640 			d = &sc->re_ldata.re_tx_list[curdesc];
1641 			d->re_vlanctl = htole32(vlanctl);
1642 			re_set_bufaddr(d, RE_TXPADDADDR(sc));
1643 			cmdstat = re_flags |
1644 			    RE_TDESC_CMD_OWN | RE_TDESC_CMD_EOF |
1645 			    (RE_IP4CSUMTX_PADLEN + 1 - m->m_pkthdr.len);
1646 			if (curdesc == (RE_TX_DESC_CNT(sc) - 1))
1647 				cmdstat |= RE_TDESC_CMD_EOR;
1648 			d->re_cmdstat = htole32(cmdstat);
1649 			RE_TXDESCSYNC(sc, curdesc,
1650 			    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1651 			lastdesc = curdesc;
1652 			curdesc = RE_NEXT_TX_DESC(sc, curdesc);
1653 		}
1654 		KASSERT(lastdesc != -1);
1655 
1656 		/* Transfer ownership of packet to the chip. */
1657 
1658 		sc->re_ldata.re_tx_list[startdesc].re_cmdstat |=
1659 		    htole32(RE_TDESC_CMD_OWN);
1660 		RE_TXDESCSYNC(sc, startdesc,
1661 		    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1662 
1663 		/* update info of TX queue and descriptors */
1664 		txq->txq_mbuf = m;
1665 		txq->txq_descidx = lastdesc;
1666 		txq->txq_nsegs = nsegs;
1667 
1668 		sc->re_ldata.re_txq_free--;
1669 		sc->re_ldata.re_tx_free -= nsegs;
1670 		sc->re_ldata.re_tx_nextfree = curdesc;
1671 
1672 #if NBPFILTER > 0
1673 		/*
1674 		 * If there's a BPF listener, bounce a copy of this frame
1675 		 * to him.
1676 		 */
1677 		if (ifp->if_bpf)
1678 			bpf_mtap(ifp->if_bpf, m);
1679 #endif
1680 	}
1681 
1682 	if (sc->re_ldata.re_txq_free < ofree) {
1683 		/*
1684 		 * TX packets are enqueued.
1685 		 */
1686 		sc->re_ldata.re_txq_prodidx = idx;
1687 
1688 		/*
1689 		 * Start the transmitter to poll.
1690 		 *
1691 		 * RealTek put the TX poll request register in a different
1692 		 * location on the 8169 gigE chip. I don't know why.
1693 		 */
1694 		if ((sc->sc_quirk & RTKQ_8139CPLUS) != 0)
1695 			CSR_WRITE_1(sc, RTK_TXSTART, RTK_TXSTART_START);
1696 		else
1697 			CSR_WRITE_1(sc, RTK_GTXSTART, RTK_TXSTART_START);
1698 
1699 		/*
1700 		 * Use the countdown timer for interrupt moderation.
1701 		 * 'TX done' interrupts are disabled. Instead, we reset the
1702 		 * countdown timer, which will begin counting until it hits
1703 		 * the value in the TIMERINT register, and then trigger an
1704 		 * interrupt. Each time we write to the TIMERCNT register,
1705 		 * the timer count is reset to 0.
1706 		 */
1707 		CSR_WRITE_4(sc, RTK_TIMERCNT, 1);
1708 
1709 		/*
1710 		 * Set a timeout in case the chip goes out to lunch.
1711 		 */
1712 		ifp->if_timer = 5;
1713 	}
1714 }
1715 
1716 static int
1717 re_init(struct ifnet *ifp)
1718 {
1719 	struct rtk_softc *sc = ifp->if_softc;
1720 	const uint8_t *enaddr;
1721 	uint32_t rxcfg = 0;
1722 	uint32_t reg;
1723 	uint16_t cfg;
1724 	int error;
1725 
1726 	if ((error = re_enable(sc)) != 0)
1727 		goto out;
1728 
1729 	/*
1730 	 * Cancel pending I/O and free all RX/TX buffers.
1731 	 */
1732 	re_stop(ifp, 0);
1733 
1734 	re_reset(sc);
1735 
1736 	/*
1737 	 * Enable C+ RX and TX mode, as well as VLAN stripping and
1738 	 * RX checksum offload. We must configure the C+ register
1739 	 * before all others.
1740 	 */
1741 	cfg = RE_CPLUSCMD_PCI_MRW;
1742 
1743 	/*
1744 	 * XXX: For old 8169 set bit 14.
1745 	 *      For 8169S/8110S and above, do not set bit 14.
1746 	 */
1747 	if ((sc->sc_quirk & RTKQ_8169NONS) != 0)
1748 		cfg |= (0x1 << 14);
1749 
1750 	if ((ifp->if_capenable & ETHERCAP_VLAN_HWTAGGING) != 0)
1751 		cfg |= RE_CPLUSCMD_VLANSTRIP;
1752 	if ((ifp->if_capenable & (IFCAP_CSUM_IPv4_Rx |
1753 	     IFCAP_CSUM_TCPv4_Rx | IFCAP_CSUM_UDPv4_Rx)) != 0)
1754 		cfg |= RE_CPLUSCMD_RXCSUM_ENB;
1755 	if ((sc->sc_quirk & RTKQ_MACSTAT) != 0) {
1756 		cfg |= RE_CPLUSCMD_MACSTAT_DIS;
1757 		cfg |= RE_CPLUSCMD_TXENB;
1758 	} else
1759 		cfg |= RE_CPLUSCMD_RXENB | RE_CPLUSCMD_TXENB;
1760 
1761 	CSR_WRITE_2(sc, RTK_CPLUS_CMD, cfg);
1762 
1763 	/* XXX: from Realtek-supplied Linux driver. Wholly undocumented. */
1764 	if ((sc->sc_quirk & RTKQ_8139CPLUS) == 0)
1765 		CSR_WRITE_2(sc, RTK_IM, 0x0000);
1766 
1767 	DELAY(10000);
1768 
1769 	/*
1770 	 * Init our MAC address.  Even though the chipset
1771 	 * documentation doesn't mention it, we need to enter "Config
1772 	 * register write enable" mode to modify the ID registers.
1773 	 */
1774 	CSR_WRITE_1(sc, RTK_EECMD, RTK_EEMODE_WRITECFG);
1775 	enaddr = CLLADDR(ifp->if_sadl);
1776 	reg = enaddr[0] | (enaddr[1] << 8) |
1777 	    (enaddr[2] << 16) | (enaddr[3] << 24);
1778 	CSR_WRITE_4(sc, RTK_IDR0, reg);
1779 	reg = enaddr[4] | (enaddr[5] << 8);
1780 	CSR_WRITE_4(sc, RTK_IDR4, reg);
1781 	CSR_WRITE_1(sc, RTK_EECMD, RTK_EEMODE_OFF);
1782 
1783 	/*
1784 	 * For C+ mode, initialize the RX descriptors and mbufs.
1785 	 */
1786 	re_rx_list_init(sc);
1787 	re_tx_list_init(sc);
1788 
1789 	/*
1790 	 * Load the addresses of the RX and TX lists into the chip.
1791 	 */
1792 	CSR_WRITE_4(sc, RTK_RXLIST_ADDR_HI,
1793 	    RE_ADDR_HI(sc->re_ldata.re_rx_list_map->dm_segs[0].ds_addr));
1794 	CSR_WRITE_4(sc, RTK_RXLIST_ADDR_LO,
1795 	    RE_ADDR_LO(sc->re_ldata.re_rx_list_map->dm_segs[0].ds_addr));
1796 
1797 	CSR_WRITE_4(sc, RTK_TXLIST_ADDR_HI,
1798 	    RE_ADDR_HI(sc->re_ldata.re_tx_list_map->dm_segs[0].ds_addr));
1799 	CSR_WRITE_4(sc, RTK_TXLIST_ADDR_LO,
1800 	    RE_ADDR_LO(sc->re_ldata.re_tx_list_map->dm_segs[0].ds_addr));
1801 
1802 	/*
1803 	 * Enable transmit and receive.
1804 	 */
1805 	CSR_WRITE_1(sc, RTK_COMMAND, RTK_CMD_TX_ENB | RTK_CMD_RX_ENB);
1806 
1807 	/*
1808 	 * Set the initial TX and RX configuration.
1809 	 */
1810 	if (sc->re_testmode && (sc->sc_quirk & RTKQ_8169NONS) != 0) {
1811 		/* test mode is needed only for old 8169 */
1812 		CSR_WRITE_4(sc, RTK_TXCFG,
1813 		    RE_TXCFG_CONFIG | RTK_LOOPTEST_ON);
1814 	} else
1815 		CSR_WRITE_4(sc, RTK_TXCFG, RE_TXCFG_CONFIG);
1816 
1817 	CSR_WRITE_1(sc, RTK_EARLY_TX_THRESH, 16);
1818 
1819 	CSR_WRITE_4(sc, RTK_RXCFG, RE_RXCFG_CONFIG);
1820 
1821 	/* Set the individual bit to receive frames for this host only. */
1822 	rxcfg = CSR_READ_4(sc, RTK_RXCFG);
1823 	rxcfg |= RTK_RXCFG_RX_INDIV;
1824 
1825 	/* If we want promiscuous mode, set the allframes bit. */
1826 	if (ifp->if_flags & IFF_PROMISC)
1827 		rxcfg |= RTK_RXCFG_RX_ALLPHYS;
1828 	else
1829 		rxcfg &= ~RTK_RXCFG_RX_ALLPHYS;
1830 	CSR_WRITE_4(sc, RTK_RXCFG, rxcfg);
1831 
1832 	/*
1833 	 * Set capture broadcast bit to capture broadcast frames.
1834 	 */
1835 	if (ifp->if_flags & IFF_BROADCAST)
1836 		rxcfg |= RTK_RXCFG_RX_BROAD;
1837 	else
1838 		rxcfg &= ~RTK_RXCFG_RX_BROAD;
1839 	CSR_WRITE_4(sc, RTK_RXCFG, rxcfg);
1840 
1841 	/*
1842 	 * Program the multicast filter, if necessary.
1843 	 */
1844 	rtk_setmulti(sc);
1845 
1846 	/*
1847 	 * Enable interrupts.
1848 	 */
1849 	if (sc->re_testmode)
1850 		CSR_WRITE_2(sc, RTK_IMR, 0);
1851 	else
1852 		CSR_WRITE_2(sc, RTK_IMR, RTK_INTRS_CPLUS);
1853 
1854 	/* Start RX/TX process. */
1855 	CSR_WRITE_4(sc, RTK_MISSEDPKT, 0);
1856 #ifdef notdef
1857 	/* Enable receiver and transmitter. */
1858 	CSR_WRITE_1(sc, RTK_COMMAND, RTK_CMD_TX_ENB | RTK_CMD_RX_ENB);
1859 #endif
1860 
1861 	/*
1862 	 * Initialize the timer interrupt register so that
1863 	 * a timer interrupt will be generated once the timer
1864 	 * reaches a certain number of ticks. The timer is
1865 	 * reloaded on each transmit. This gives us TX interrupt
1866 	 * moderation, which dramatically improves TX frame rate.
1867 	 */
1868 
1869 	if ((sc->sc_quirk & RTKQ_8139CPLUS) != 0)
1870 		CSR_WRITE_4(sc, RTK_TIMERINT, 0x400);
1871 	else {
1872 		CSR_WRITE_4(sc, RTK_TIMERINT_8169, 0x800);
1873 
1874 		/*
1875 		 * For 8169 gigE NICs, set the max allowed RX packet
1876 		 * size so we can receive jumbo frames.
1877 		 */
1878 		CSR_WRITE_2(sc, RTK_MAXRXPKTLEN, 16383);
1879 	}
1880 
1881 	if (sc->re_testmode)
1882 		return 0;
1883 
1884 	CSR_WRITE_1(sc, RTK_CFG1, RTK_CFG1_DRVLOAD);
1885 
1886 	ifp->if_flags |= IFF_RUNNING;
1887 	ifp->if_flags &= ~IFF_OACTIVE;
1888 
1889 	callout_reset(&sc->rtk_tick_ch, hz, re_tick, sc);
1890 
1891  out:
1892 	if (error) {
1893 		ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
1894 		ifp->if_timer = 0;
1895 		printf("%s: interface not running\n",
1896 		    device_xname(sc->sc_dev));
1897 	}
1898 
1899 	return error;
1900 }
1901 
1902 static int
1903 re_ioctl(struct ifnet *ifp, u_long command, void *data)
1904 {
1905 	struct rtk_softc *sc = ifp->if_softc;
1906 	struct ifreq *ifr = data;
1907 	int s, error = 0;
1908 
1909 	s = splnet();
1910 
1911 	switch (command) {
1912 	case SIOCSIFMTU:
1913 		/*
1914 		 * Disable jumbo frames if it's not supported.
1915 		 */
1916 		if ((sc->sc_quirk & RTKQ_NOJUMBO) != 0 &&
1917 		    ifr->ifr_mtu > ETHERMTU) {
1918 			error = EINVAL;
1919 			break;
1920 		}
1921 
1922 		if (ifr->ifr_mtu < ETHERMIN || ifr->ifr_mtu > ETHERMTU_JUMBO)
1923 			error = EINVAL;
1924 		else if ((error = ifioctl_common(ifp, command, data)) ==
1925 		    ENETRESET)
1926 			error = 0;
1927 		break;
1928 	default:
1929 		if ((error = ether_ioctl(ifp, command, data)) != ENETRESET)
1930 			break;
1931 
1932 		error = 0;
1933 
1934 		if (command == SIOCSIFCAP)
1935 			error = (*ifp->if_init)(ifp);
1936 		else if (command != SIOCADDMULTI && command != SIOCDELMULTI)
1937 			;
1938 		else if (ifp->if_flags & IFF_RUNNING)
1939 			rtk_setmulti(sc);
1940 		break;
1941 	}
1942 
1943 	splx(s);
1944 
1945 	return error;
1946 }
1947 
1948 static void
1949 re_watchdog(struct ifnet *ifp)
1950 {
1951 	struct rtk_softc *sc;
1952 	int s;
1953 
1954 	sc = ifp->if_softc;
1955 	s = splnet();
1956 	printf("%s: watchdog timeout\n", device_xname(sc->sc_dev));
1957 	ifp->if_oerrors++;
1958 
1959 	re_txeof(sc);
1960 	re_rxeof(sc);
1961 
1962 	re_init(ifp);
1963 
1964 	splx(s);
1965 }
1966 
1967 /*
1968  * Stop the adapter and free any mbufs allocated to the
1969  * RX and TX lists.
1970  */
1971 static void
1972 re_stop(struct ifnet *ifp, int disable)
1973 {
1974 	int i;
1975 	struct rtk_softc *sc = ifp->if_softc;
1976 
1977 	callout_stop(&sc->rtk_tick_ch);
1978 
1979 	mii_down(&sc->mii);
1980 
1981 	if ((sc->sc_quirk & RTKQ_CMDSTOP) != 0)
1982 		CSR_WRITE_1(sc, RTK_COMMAND, RTK_CMD_STOPREQ | RTK_CMD_TX_ENB |
1983 		    RTK_CMD_RX_ENB);
1984 	else
1985 		CSR_WRITE_1(sc, RTK_COMMAND, 0x00);
1986 	DELAY(1000);
1987 	CSR_WRITE_2(sc, RTK_IMR, 0x0000);
1988 	CSR_WRITE_2(sc, RTK_ISR, 0xFFFF);
1989 
1990 	if (sc->re_head != NULL) {
1991 		m_freem(sc->re_head);
1992 		sc->re_head = sc->re_tail = NULL;
1993 	}
1994 
1995 	/* Free the TX list buffers. */
1996 	for (i = 0; i < RE_TX_QLEN; i++) {
1997 		if (sc->re_ldata.re_txq[i].txq_mbuf != NULL) {
1998 			bus_dmamap_unload(sc->sc_dmat,
1999 			    sc->re_ldata.re_txq[i].txq_dmamap);
2000 			m_freem(sc->re_ldata.re_txq[i].txq_mbuf);
2001 			sc->re_ldata.re_txq[i].txq_mbuf = NULL;
2002 		}
2003 	}
2004 
2005 	/* Free the RX list buffers. */
2006 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
2007 		if (sc->re_ldata.re_rxsoft[i].rxs_mbuf != NULL) {
2008 			bus_dmamap_unload(sc->sc_dmat,
2009 			    sc->re_ldata.re_rxsoft[i].rxs_dmamap);
2010 			m_freem(sc->re_ldata.re_rxsoft[i].rxs_mbuf);
2011 			sc->re_ldata.re_rxsoft[i].rxs_mbuf = NULL;
2012 		}
2013 	}
2014 
2015 	if (disable)
2016 		re_disable(sc);
2017 
2018 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
2019 	ifp->if_timer = 0;
2020 }
2021