xref: /netbsd-src/sys/dev/ic/rtl8169.c (revision 274254cdae52594c1aa480a736aef78313d15c9c)
1 /*	$NetBSD: rtl8169.c,v 1.113 2009/03/28 22:16:08 tsutsui 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 1.113 2009/03/28 22:16:08 tsutsui 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 FreeBSD driver does this only for MACFG_3,
393 	 *     but also says "Rtl8169s sigle chip detected".
394 	 */
395 	if ((sc->sc_quirk & RTKQ_MACLDPS) != 0)
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 	if ((sc->sc_quirk & RTKQ_8139CPLUS) == 0) {
566 		uint32_t hwrev;
567 
568 		/* Revision of 8169/8169S/8110s in bits 30..26, 23 */
569 		hwrev = CSR_READ_4(sc, RTK_TXCFG) & RTK_TXCFG_HWREV;
570 		/* These rev numbers are taken from Realtek's driver */
571 		switch (hwrev) {
572 		case RTK_HWREV_8169:
573 			/* XXX not in the Realtek driver */
574 			sc->sc_rev = 1;
575 			sc->sc_quirk |= RTKQ_8169NONS;
576 			break;
577 		case RTK_HWREV_8169S:
578 		case RTK_HWREV_8110S:
579 			sc->sc_rev = 3;
580 			sc->sc_quirk |= RTKQ_MACLDPS;
581 			break;
582 		case RTK_HWREV_8169_8110SB:
583 			sc->sc_rev = 4;
584 			sc->sc_quirk |= RTKQ_MACLDPS;
585 			break;
586 		case RTK_HWREV_8169_8110SC:
587 			sc->sc_rev = 5;
588 			sc->sc_quirk |= RTKQ_MACLDPS;
589 			break;
590 		case RTK_HWREV_8101E:
591 			sc->sc_rev = 11;
592 			sc->sc_quirk |= RTKQ_NOJUMBO;
593 			break;
594 		case RTK_HWREV_8168_SPIN1:
595 			sc->sc_rev = 21;
596 			break;
597 		case RTK_HWREV_8168_SPIN2:
598 			sc->sc_rev = 22;
599 			break;
600 		case RTK_HWREV_8168_SPIN3:
601 			sc->sc_rev = 23;
602 			break;
603 		case RTK_HWREV_8168C:
604 		case RTK_HWREV_8168C_SPIN2:
605 			sc->sc_rev = 24;
606 			sc->sc_quirk |= RTKQ_DESCV2 | RTKQ_NOEECMD;
607 			/*
608 			 * From FreeBSD driver:
609 			 *
610 			 * These (8168/8111) controllers support jumbo frame
611 			 * but it seems that enabling it requires touching
612 			 * additional magic registers. Depending on MAC
613 			 * revisions some controllers need to disable
614 			 * checksum offload. So disable jumbo frame until
615 			 * I have better idea what it really requires to
616 			 * make it support.
617 			 * RTL8168C/CP : supports up to 6KB jumbo frame.
618 			 * RTL8111C/CP : supports up to 9KB jumbo frame.
619 			 */
620 			sc->sc_quirk |= RTKQ_NOJUMBO;
621 			break;
622 		case RTK_HWREV_8102E:
623 		case RTK_HWREV_8102EL:
624 			sc->sc_rev = 25;
625 			sc->sc_quirk |=
626 			    RTKQ_DESCV2 | RTKQ_NOEECMD | RTKQ_NOJUMBO;
627 			break;
628 		case RTK_HWREV_8100E:
629 		case RTK_HWREV_8100E_SPIN2:
630 			/* XXX not in the Realtek driver */
631 			sc->sc_rev = 0;
632 			sc->sc_quirk |= RTKQ_NOJUMBO;
633 			break;
634 		default:
635 			aprint_normal_dev(sc->sc_dev,
636 			    "Unknown revision (0x%08x)\n", hwrev);
637 			sc->sc_rev = 0;
638 		}
639 
640 		/* Set RX length mask */
641 		sc->re_rxlenmask = RE_RDESC_STAT_GFRAGLEN;
642 		sc->re_ldata.re_tx_desc_cnt = RE_TX_DESC_CNT_8169;
643 	} else {
644 		sc->sc_quirk |= RTKQ_NOJUMBO;
645 
646 		/* Set RX length mask */
647 		sc->re_rxlenmask = RE_RDESC_STAT_FRAGLEN;
648 		sc->re_ldata.re_tx_desc_cnt = RE_TX_DESC_CNT_8139;
649 	}
650 
651 	/* Reset the adapter. */
652 	re_reset(sc);
653 
654 	if ((sc->sc_quirk & RTKQ_NOEECMD) != 0) {
655 		/*
656 		 * Get station address from ID registers.
657 		 */
658 		for (i = 0; i < ETHER_ADDR_LEN; i++)
659 			eaddr[i] = CSR_READ_1(sc, RTK_IDR0 + i);
660 	} else {
661 		/*
662 		 * Get station address from the EEPROM.
663 		 */
664 		if (rtk_read_eeprom(sc, RTK_EE_ID, RTK_EEADDR_LEN1) == 0x8129)
665 			addr_len = RTK_EEADDR_LEN1;
666 		else
667 			addr_len = RTK_EEADDR_LEN0;
668 
669 		/*
670 		 * Get station address from the EEPROM.
671 		 */
672 		for (i = 0; i < ETHER_ADDR_LEN / 2; i++) {
673 			val = rtk_read_eeprom(sc, RTK_EE_EADDR0 + i, addr_len);
674 			eaddr[(i * 2) + 0] = val & 0xff;
675 			eaddr[(i * 2) + 1] = val >> 8;
676 		}
677 	}
678 
679 	aprint_normal_dev(sc->sc_dev, "Ethernet address %s\n",
680 	    ether_sprintf(eaddr));
681 
682 	if (sc->re_ldata.re_tx_desc_cnt >
683 	    PAGE_SIZE / sizeof(struct re_desc)) {
684 		sc->re_ldata.re_tx_desc_cnt =
685 		    PAGE_SIZE / sizeof(struct re_desc);
686 	}
687 
688 	aprint_verbose_dev(sc->sc_dev, "using %d tx descriptors\n",
689 	    sc->re_ldata.re_tx_desc_cnt);
690 	KASSERT(RE_NEXT_TX_DESC(sc, RE_TX_DESC_CNT(sc) - 1) == 0);
691 
692 	/* Allocate DMA'able memory for the TX ring */
693 	if ((error = bus_dmamem_alloc(sc->sc_dmat, RE_TX_LIST_SZ(sc),
694 	    RE_RING_ALIGN, 0, &sc->re_ldata.re_tx_listseg, 1,
695 	    &sc->re_ldata.re_tx_listnseg, BUS_DMA_NOWAIT)) != 0) {
696 		aprint_error_dev(sc->sc_dev,
697 		    "can't allocate tx listseg, error = %d\n", error);
698 		goto fail_0;
699 	}
700 
701 	/* Load the map for the TX ring. */
702 	if ((error = bus_dmamem_map(sc->sc_dmat, &sc->re_ldata.re_tx_listseg,
703 	    sc->re_ldata.re_tx_listnseg, RE_TX_LIST_SZ(sc),
704 	    (void **)&sc->re_ldata.re_tx_list,
705 	    BUS_DMA_COHERENT | BUS_DMA_NOWAIT)) != 0) {
706 		aprint_error_dev(sc->sc_dev,
707 		    "can't map tx list, error = %d\n", error);
708 	  	goto fail_1;
709 	}
710 	memset(sc->re_ldata.re_tx_list, 0, RE_TX_LIST_SZ(sc));
711 
712 	if ((error = bus_dmamap_create(sc->sc_dmat, RE_TX_LIST_SZ(sc), 1,
713 	    RE_TX_LIST_SZ(sc), 0, 0,
714 	    &sc->re_ldata.re_tx_list_map)) != 0) {
715 		aprint_error_dev(sc->sc_dev,
716 		    "can't create tx list map, error = %d\n", error);
717 		goto fail_2;
718 	}
719 
720 
721 	if ((error = bus_dmamap_load(sc->sc_dmat,
722 	    sc->re_ldata.re_tx_list_map, sc->re_ldata.re_tx_list,
723 	    RE_TX_LIST_SZ(sc), NULL, BUS_DMA_NOWAIT)) != 0) {
724 		aprint_error_dev(sc->sc_dev,
725 		    "can't load tx list, error = %d\n", error);
726 		goto fail_3;
727 	}
728 
729 	/* Create DMA maps for TX buffers */
730 	for (i = 0; i < RE_TX_QLEN; i++) {
731 		error = bus_dmamap_create(sc->sc_dmat,
732 		    round_page(IP_MAXPACKET),
733 		    RE_TX_DESC_CNT(sc), RE_TDESC_CMD_FRAGLEN,
734 		    0, 0, &sc->re_ldata.re_txq[i].txq_dmamap);
735 		if (error) {
736 			aprint_error_dev(sc->sc_dev,
737 			    "can't create DMA map for TX\n");
738 			goto fail_4;
739 		}
740 	}
741 
742 	/* Allocate DMA'able memory for the RX ring */
743 	/* XXX see also a comment about RE_RX_DMAMEM_SZ in rtl81x9var.h */
744 	if ((error = bus_dmamem_alloc(sc->sc_dmat,
745 	    RE_RX_DMAMEM_SZ, RE_RING_ALIGN, 0, &sc->re_ldata.re_rx_listseg, 1,
746 	    &sc->re_ldata.re_rx_listnseg, BUS_DMA_NOWAIT)) != 0) {
747 		aprint_error_dev(sc->sc_dev,
748 		    "can't allocate rx listseg, error = %d\n", error);
749 		goto fail_4;
750 	}
751 
752 	/* Load the map for the RX ring. */
753 	if ((error = bus_dmamem_map(sc->sc_dmat, &sc->re_ldata.re_rx_listseg,
754 	    sc->re_ldata.re_rx_listnseg, RE_RX_DMAMEM_SZ,
755 	    (void **)&sc->re_ldata.re_rx_list,
756 	    BUS_DMA_COHERENT | BUS_DMA_NOWAIT)) != 0) {
757 		aprint_error_dev(sc->sc_dev,
758 		    "can't map rx list, error = %d\n", error);
759 		goto fail_5;
760 	}
761 	memset(sc->re_ldata.re_rx_list, 0, RE_RX_DMAMEM_SZ);
762 
763 	if ((error = bus_dmamap_create(sc->sc_dmat,
764 	    RE_RX_DMAMEM_SZ, 1, RE_RX_DMAMEM_SZ, 0, 0,
765 	    &sc->re_ldata.re_rx_list_map)) != 0) {
766 		aprint_error_dev(sc->sc_dev,
767 		    "can't create rx list map, error = %d\n", error);
768 		goto fail_6;
769 	}
770 
771 	if ((error = bus_dmamap_load(sc->sc_dmat,
772 	    sc->re_ldata.re_rx_list_map, sc->re_ldata.re_rx_list,
773 	    RE_RX_DMAMEM_SZ, NULL, BUS_DMA_NOWAIT)) != 0) {
774 		aprint_error_dev(sc->sc_dev,
775 		    "can't load rx list, error = %d\n", error);
776 		goto fail_7;
777 	}
778 
779 	/* Create DMA maps for RX buffers */
780 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
781 		error = bus_dmamap_create(sc->sc_dmat, MCLBYTES, 1, MCLBYTES,
782 		    0, 0, &sc->re_ldata.re_rxsoft[i].rxs_dmamap);
783 		if (error) {
784 			aprint_error_dev(sc->sc_dev,
785 			    "can't create DMA map for RX\n");
786 			goto fail_8;
787 		}
788 	}
789 
790 	/*
791 	 * Record interface as attached. From here, we should not fail.
792 	 */
793 	sc->sc_flags |= RTK_ATTACHED;
794 
795 	ifp = &sc->ethercom.ec_if;
796 	ifp->if_softc = sc;
797 	strlcpy(ifp->if_xname, device_xname(sc->sc_dev), IFNAMSIZ);
798 	ifp->if_mtu = ETHERMTU;
799 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
800 	ifp->if_ioctl = re_ioctl;
801 	sc->ethercom.ec_capabilities |=
802 	    ETHERCAP_VLAN_MTU | ETHERCAP_VLAN_HWTAGGING;
803 	ifp->if_start = re_start;
804 	ifp->if_stop = re_stop;
805 
806 	/*
807 	 * IFCAP_CSUM_IPv4_Tx on re(4) is broken for small packets,
808 	 * so we have a workaround to handle the bug by padding
809 	 * such packets manually.
810 	 */
811 	ifp->if_capabilities |=
812 	    IFCAP_CSUM_IPv4_Tx | IFCAP_CSUM_IPv4_Rx |
813 	    IFCAP_CSUM_TCPv4_Tx | IFCAP_CSUM_TCPv4_Rx |
814 	    IFCAP_CSUM_UDPv4_Tx | IFCAP_CSUM_UDPv4_Rx |
815 	    IFCAP_TSOv4;
816 
817 	/*
818 	 * XXX
819 	 * Still have no idea how to make TSO work on 8168C, 8168CP,
820 	 * 8102E, 8111C and 8111CP.
821 	 */
822 	if ((sc->sc_quirk & RTKQ_DESCV2) != 0)
823 		ifp->if_capabilities &= ~IFCAP_TSOv4;
824 
825 	ifp->if_watchdog = re_watchdog;
826 	ifp->if_init = re_init;
827 	ifp->if_snd.ifq_maxlen = RE_IFQ_MAXLEN;
828 	ifp->if_capenable = ifp->if_capabilities;
829 	IFQ_SET_READY(&ifp->if_snd);
830 
831 	callout_init(&sc->rtk_tick_ch, 0);
832 
833 	/* Do MII setup */
834 	sc->mii.mii_ifp = ifp;
835 	sc->mii.mii_readreg = re_miibus_readreg;
836 	sc->mii.mii_writereg = re_miibus_writereg;
837 	sc->mii.mii_statchg = re_miibus_statchg;
838 	sc->ethercom.ec_mii = &sc->mii;
839 	ifmedia_init(&sc->mii.mii_media, IFM_IMASK, ether_mediachange,
840 	    ether_mediastatus);
841 	mii_attach(sc->sc_dev, &sc->mii, 0xffffffff, MII_PHY_ANY,
842 	    MII_OFFSET_ANY, 0);
843 	ifmedia_set(&sc->mii.mii_media, IFM_ETHER | IFM_AUTO);
844 
845 	/*
846 	 * Call MI attach routine.
847 	 */
848 	if_attach(ifp);
849 	ether_ifattach(ifp, eaddr);
850 
851 	return;
852 
853  fail_8:
854 	/* Destroy DMA maps for RX buffers. */
855 	for (i = 0; i < RE_RX_DESC_CNT; i++)
856 		if (sc->re_ldata.re_rxsoft[i].rxs_dmamap != NULL)
857 			bus_dmamap_destroy(sc->sc_dmat,
858 			    sc->re_ldata.re_rxsoft[i].rxs_dmamap);
859 
860 	/* Free DMA'able memory for the RX ring. */
861 	bus_dmamap_unload(sc->sc_dmat, sc->re_ldata.re_rx_list_map);
862  fail_7:
863 	bus_dmamap_destroy(sc->sc_dmat, sc->re_ldata.re_rx_list_map);
864  fail_6:
865 	bus_dmamem_unmap(sc->sc_dmat,
866 	    (void *)sc->re_ldata.re_rx_list, RE_RX_DMAMEM_SZ);
867  fail_5:
868 	bus_dmamem_free(sc->sc_dmat,
869 	    &sc->re_ldata.re_rx_listseg, sc->re_ldata.re_rx_listnseg);
870 
871  fail_4:
872 	/* Destroy DMA maps for TX buffers. */
873 	for (i = 0; i < RE_TX_QLEN; i++)
874 		if (sc->re_ldata.re_txq[i].txq_dmamap != NULL)
875 			bus_dmamap_destroy(sc->sc_dmat,
876 			    sc->re_ldata.re_txq[i].txq_dmamap);
877 
878 	/* Free DMA'able memory for the TX ring. */
879 	bus_dmamap_unload(sc->sc_dmat, sc->re_ldata.re_tx_list_map);
880  fail_3:
881 	bus_dmamap_destroy(sc->sc_dmat, sc->re_ldata.re_tx_list_map);
882  fail_2:
883 	bus_dmamem_unmap(sc->sc_dmat,
884 	    (void *)sc->re_ldata.re_tx_list, RE_TX_LIST_SZ(sc));
885  fail_1:
886 	bus_dmamem_free(sc->sc_dmat,
887 	    &sc->re_ldata.re_tx_listseg, sc->re_ldata.re_tx_listnseg);
888  fail_0:
889 	return;
890 }
891 
892 
893 /*
894  * re_activate:
895  *     Handle device activation/deactivation requests.
896  */
897 int
898 re_activate(device_t self, enum devact act)
899 {
900 	struct rtk_softc *sc = device_private(self);
901 	int s, error = 0;
902 
903 	s = splnet();
904 	switch (act) {
905 	case DVACT_ACTIVATE:
906 		error = EOPNOTSUPP;
907 		break;
908 	case DVACT_DEACTIVATE:
909 		mii_activate(&sc->mii, act, MII_PHY_ANY, MII_OFFSET_ANY);
910 		if_deactivate(&sc->ethercom.ec_if);
911 		break;
912 	}
913 	splx(s);
914 
915 	return error;
916 }
917 
918 /*
919  * re_detach:
920  *     Detach a rtk interface.
921  */
922 int
923 re_detach(struct rtk_softc *sc)
924 {
925 	struct ifnet *ifp = &sc->ethercom.ec_if;
926 	int i;
927 
928 	/*
929 	 * Succeed now if there isn't any work to do.
930 	 */
931 	if ((sc->sc_flags & RTK_ATTACHED) == 0)
932 		return 0;
933 
934 	/* Unhook our tick handler. */
935 	callout_stop(&sc->rtk_tick_ch);
936 
937 	/* Detach all PHYs. */
938 	mii_detach(&sc->mii, MII_PHY_ANY, MII_OFFSET_ANY);
939 
940 	/* Delete all remaining media. */
941 	ifmedia_delete_instance(&sc->mii.mii_media, IFM_INST_ANY);
942 
943 	ether_ifdetach(ifp);
944 	if_detach(ifp);
945 
946 	/* Destroy DMA maps for RX buffers. */
947 	for (i = 0; i < RE_RX_DESC_CNT; i++)
948 		if (sc->re_ldata.re_rxsoft[i].rxs_dmamap != NULL)
949 			bus_dmamap_destroy(sc->sc_dmat,
950 			    sc->re_ldata.re_rxsoft[i].rxs_dmamap);
951 
952 	/* Free DMA'able memory for the RX ring. */
953 	bus_dmamap_unload(sc->sc_dmat, sc->re_ldata.re_rx_list_map);
954 	bus_dmamap_destroy(sc->sc_dmat, sc->re_ldata.re_rx_list_map);
955 	bus_dmamem_unmap(sc->sc_dmat,
956 	    (void *)sc->re_ldata.re_rx_list, RE_RX_DMAMEM_SZ);
957 	bus_dmamem_free(sc->sc_dmat,
958 	    &sc->re_ldata.re_rx_listseg, sc->re_ldata.re_rx_listnseg);
959 
960 	/* Destroy DMA maps for TX buffers. */
961 	for (i = 0; i < RE_TX_QLEN; i++)
962 		if (sc->re_ldata.re_txq[i].txq_dmamap != NULL)
963 			bus_dmamap_destroy(sc->sc_dmat,
964 			    sc->re_ldata.re_txq[i].txq_dmamap);
965 
966 	/* Free DMA'able memory for the TX ring. */
967 	bus_dmamap_unload(sc->sc_dmat, sc->re_ldata.re_tx_list_map);
968 	bus_dmamap_destroy(sc->sc_dmat, sc->re_ldata.re_tx_list_map);
969 	bus_dmamem_unmap(sc->sc_dmat,
970 	    (void *)sc->re_ldata.re_tx_list, RE_TX_LIST_SZ(sc));
971 	bus_dmamem_free(sc->sc_dmat,
972 	    &sc->re_ldata.re_tx_listseg, sc->re_ldata.re_tx_listnseg);
973 
974 	return 0;
975 }
976 
977 /*
978  * re_enable:
979  *     Enable the RTL81X9 chip.
980  */
981 static int
982 re_enable(struct rtk_softc *sc)
983 {
984 
985 	if (RTK_IS_ENABLED(sc) == 0 && sc->sc_enable != NULL) {
986 		if ((*sc->sc_enable)(sc) != 0) {
987 			printf("%s: device enable failed\n",
988 			    device_xname(sc->sc_dev));
989 			return EIO;
990 		}
991 		sc->sc_flags |= RTK_ENABLED;
992 	}
993 	return 0;
994 }
995 
996 /*
997  * re_disable:
998  *     Disable the RTL81X9 chip.
999  */
1000 static void
1001 re_disable(struct rtk_softc *sc)
1002 {
1003 
1004 	if (RTK_IS_ENABLED(sc) && sc->sc_disable != NULL) {
1005 		(*sc->sc_disable)(sc);
1006 		sc->sc_flags &= ~RTK_ENABLED;
1007 	}
1008 }
1009 
1010 static int
1011 re_newbuf(struct rtk_softc *sc, int idx, struct mbuf *m)
1012 {
1013 	struct mbuf *n = NULL;
1014 	bus_dmamap_t map;
1015 	struct re_desc *d;
1016 	struct re_rxsoft *rxs;
1017 	uint32_t cmdstat;
1018 	int error;
1019 
1020 	if (m == NULL) {
1021 		MGETHDR(n, M_DONTWAIT, MT_DATA);
1022 		if (n == NULL)
1023 			return ENOBUFS;
1024 
1025 		MCLGET(n, M_DONTWAIT);
1026 		if ((n->m_flags & M_EXT) == 0) {
1027 			m_freem(n);
1028 			return ENOBUFS;
1029 		}
1030 		m = n;
1031 	} else
1032 		m->m_data = m->m_ext.ext_buf;
1033 
1034 	/*
1035 	 * Initialize mbuf length fields and fixup
1036 	 * alignment so that the frame payload is
1037 	 * longword aligned.
1038 	 */
1039 	m->m_len = m->m_pkthdr.len = MCLBYTES - RE_ETHER_ALIGN;
1040 	m->m_data += RE_ETHER_ALIGN;
1041 
1042 	rxs = &sc->re_ldata.re_rxsoft[idx];
1043 	map = rxs->rxs_dmamap;
1044 	error = bus_dmamap_load_mbuf(sc->sc_dmat, map, m,
1045 	    BUS_DMA_READ|BUS_DMA_NOWAIT);
1046 
1047 	if (error)
1048 		goto out;
1049 
1050 	bus_dmamap_sync(sc->sc_dmat, map, 0, map->dm_mapsize,
1051 	    BUS_DMASYNC_PREREAD);
1052 
1053 	d = &sc->re_ldata.re_rx_list[idx];
1054 #ifdef DIAGNOSTIC
1055 	RE_RXDESCSYNC(sc, idx, BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1056 	cmdstat = le32toh(d->re_cmdstat);
1057 	RE_RXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD);
1058 	if (cmdstat & RE_RDESC_STAT_OWN) {
1059 		panic("%s: tried to map busy RX descriptor",
1060 		    device_xname(sc->sc_dev));
1061 	}
1062 #endif
1063 
1064 	rxs->rxs_mbuf = m;
1065 
1066 	d->re_vlanctl = 0;
1067 	cmdstat = map->dm_segs[0].ds_len;
1068 	if (idx == (RE_RX_DESC_CNT - 1))
1069 		cmdstat |= RE_RDESC_CMD_EOR;
1070 	re_set_bufaddr(d, map->dm_segs[0].ds_addr);
1071 	d->re_cmdstat = htole32(cmdstat);
1072 	RE_RXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1073 	cmdstat |= RE_RDESC_CMD_OWN;
1074 	d->re_cmdstat = htole32(cmdstat);
1075 	RE_RXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1076 
1077 	return 0;
1078  out:
1079 	if (n != NULL)
1080 		m_freem(n);
1081 	return ENOMEM;
1082 }
1083 
1084 static int
1085 re_tx_list_init(struct rtk_softc *sc)
1086 {
1087 	int i;
1088 
1089 	memset(sc->re_ldata.re_tx_list, 0, RE_TX_LIST_SZ(sc));
1090 	for (i = 0; i < RE_TX_QLEN; i++) {
1091 		sc->re_ldata.re_txq[i].txq_mbuf = NULL;
1092 	}
1093 
1094 	bus_dmamap_sync(sc->sc_dmat,
1095 	    sc->re_ldata.re_tx_list_map, 0,
1096 	    sc->re_ldata.re_tx_list_map->dm_mapsize,
1097 	    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1098 	sc->re_ldata.re_txq_prodidx = 0;
1099 	sc->re_ldata.re_txq_considx = 0;
1100 	sc->re_ldata.re_txq_free = RE_TX_QLEN;
1101 	sc->re_ldata.re_tx_free = RE_TX_DESC_CNT(sc);
1102 	sc->re_ldata.re_tx_nextfree = 0;
1103 
1104 	return 0;
1105 }
1106 
1107 static int
1108 re_rx_list_init(struct rtk_softc *sc)
1109 {
1110 	int i;
1111 
1112 	memset(sc->re_ldata.re_rx_list, 0, RE_RX_LIST_SZ);
1113 
1114 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
1115 		if (re_newbuf(sc, i, NULL) == ENOBUFS)
1116 			return ENOBUFS;
1117 	}
1118 
1119 	sc->re_ldata.re_rx_prodidx = 0;
1120 	sc->re_head = sc->re_tail = NULL;
1121 
1122 	return 0;
1123 }
1124 
1125 /*
1126  * RX handler for C+ and 8169. For the gigE chips, we support
1127  * the reception of jumbo frames that have been fragmented
1128  * across multiple 2K mbuf cluster buffers.
1129  */
1130 static void
1131 re_rxeof(struct rtk_softc *sc)
1132 {
1133 	struct mbuf *m;
1134 	struct ifnet *ifp;
1135 	int i, total_len;
1136 	struct re_desc *cur_rx;
1137 	struct re_rxsoft *rxs;
1138 	uint32_t rxstat, rxvlan;
1139 
1140 	ifp = &sc->ethercom.ec_if;
1141 
1142 	for (i = sc->re_ldata.re_rx_prodidx;; i = RE_NEXT_RX_DESC(sc, i)) {
1143 		cur_rx = &sc->re_ldata.re_rx_list[i];
1144 		RE_RXDESCSYNC(sc, i,
1145 		    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1146 		rxstat = le32toh(cur_rx->re_cmdstat);
1147 		rxvlan = le32toh(cur_rx->re_vlanctl);
1148 		RE_RXDESCSYNC(sc, i, BUS_DMASYNC_PREREAD);
1149 		if ((rxstat & RE_RDESC_STAT_OWN) != 0) {
1150 			break;
1151 		}
1152 		total_len = rxstat & sc->re_rxlenmask;
1153 		rxs = &sc->re_ldata.re_rxsoft[i];
1154 		m = rxs->rxs_mbuf;
1155 
1156 		/* Invalidate the RX mbuf and unload its map */
1157 
1158 		bus_dmamap_sync(sc->sc_dmat,
1159 		    rxs->rxs_dmamap, 0, rxs->rxs_dmamap->dm_mapsize,
1160 		    BUS_DMASYNC_POSTREAD);
1161 		bus_dmamap_unload(sc->sc_dmat, rxs->rxs_dmamap);
1162 
1163 		if ((rxstat & RE_RDESC_STAT_EOF) == 0) {
1164 			m->m_len = MCLBYTES - RE_ETHER_ALIGN;
1165 			if (sc->re_head == NULL)
1166 				sc->re_head = sc->re_tail = m;
1167 			else {
1168 				m->m_flags &= ~M_PKTHDR;
1169 				sc->re_tail->m_next = m;
1170 				sc->re_tail = m;
1171 			}
1172 			re_newbuf(sc, i, NULL);
1173 			continue;
1174 		}
1175 
1176 		/*
1177 		 * NOTE: for the 8139C+, the frame length field
1178 		 * is always 12 bits in size, but for the gigE chips,
1179 		 * it is 13 bits (since the max RX frame length is 16K).
1180 		 * Unfortunately, all 32 bits in the status word
1181 		 * were already used, so to make room for the extra
1182 		 * length bit, RealTek took out the 'frame alignment
1183 		 * error' bit and shifted the other status bits
1184 		 * over one slot. The OWN, EOR, FS and LS bits are
1185 		 * still in the same places. We have already extracted
1186 		 * the frame length and checked the OWN bit, so rather
1187 		 * than using an alternate bit mapping, we shift the
1188 		 * status bits one space to the right so we can evaluate
1189 		 * them using the 8169 status as though it was in the
1190 		 * same format as that of the 8139C+.
1191 		 */
1192 		if ((sc->sc_quirk & RTKQ_8139CPLUS) == 0)
1193 			rxstat >>= 1;
1194 
1195 		if (__predict_false((rxstat & RE_RDESC_STAT_RXERRSUM) != 0)) {
1196 #ifdef RE_DEBUG
1197 			printf("%s: RX error (rxstat = 0x%08x)",
1198 			    device_xname(sc->sc_dev), rxstat);
1199 			if (rxstat & RE_RDESC_STAT_FRALIGN)
1200 				printf(", frame alignment error");
1201 			if (rxstat & RE_RDESC_STAT_BUFOFLOW)
1202 				printf(", out of buffer space");
1203 			if (rxstat & RE_RDESC_STAT_FIFOOFLOW)
1204 				printf(", FIFO overrun");
1205 			if (rxstat & RE_RDESC_STAT_GIANT)
1206 				printf(", giant packet");
1207 			if (rxstat & RE_RDESC_STAT_RUNT)
1208 				printf(", runt packet");
1209 			if (rxstat & RE_RDESC_STAT_CRCERR)
1210 				printf(", CRC error");
1211 			printf("\n");
1212 #endif
1213 			ifp->if_ierrors++;
1214 			/*
1215 			 * If this is part of a multi-fragment packet,
1216 			 * discard all the pieces.
1217 			 */
1218 			if (sc->re_head != NULL) {
1219 				m_freem(sc->re_head);
1220 				sc->re_head = sc->re_tail = NULL;
1221 			}
1222 			re_newbuf(sc, i, m);
1223 			continue;
1224 		}
1225 
1226 		/*
1227 		 * If allocating a replacement mbuf fails,
1228 		 * reload the current one.
1229 		 */
1230 
1231 		if (__predict_false(re_newbuf(sc, i, NULL) != 0)) {
1232 			ifp->if_ierrors++;
1233 			if (sc->re_head != NULL) {
1234 				m_freem(sc->re_head);
1235 				sc->re_head = sc->re_tail = NULL;
1236 			}
1237 			re_newbuf(sc, i, m);
1238 			continue;
1239 		}
1240 
1241 		if (sc->re_head != NULL) {
1242 			m->m_len = total_len % (MCLBYTES - RE_ETHER_ALIGN);
1243 			/*
1244 			 * Special case: if there's 4 bytes or less
1245 			 * in this buffer, the mbuf can be discarded:
1246 			 * the last 4 bytes is the CRC, which we don't
1247 			 * care about anyway.
1248 			 */
1249 			if (m->m_len <= ETHER_CRC_LEN) {
1250 				sc->re_tail->m_len -=
1251 				    (ETHER_CRC_LEN - m->m_len);
1252 				m_freem(m);
1253 			} else {
1254 				m->m_len -= ETHER_CRC_LEN;
1255 				m->m_flags &= ~M_PKTHDR;
1256 				sc->re_tail->m_next = m;
1257 			}
1258 			m = sc->re_head;
1259 			sc->re_head = sc->re_tail = NULL;
1260 			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
1261 		} else
1262 			m->m_pkthdr.len = m->m_len =
1263 			    (total_len - ETHER_CRC_LEN);
1264 
1265 		ifp->if_ipackets++;
1266 		m->m_pkthdr.rcvif = ifp;
1267 
1268 		/* Do RX checksumming */
1269 
1270 		/* Check IP header checksum */
1271 		if ((rxstat & RE_RDESC_STAT_PROTOID) != 0 &&
1272 		    ((sc->sc_quirk & RTKQ_DESCV2) == 0 ||
1273 		     (rxvlan & RE_RDESC_VLANCTL_IPV4) != 0)) {
1274 			m->m_pkthdr.csum_flags |= M_CSUM_IPv4;
1275 			if (rxstat & RE_RDESC_STAT_IPSUMBAD)
1276 				m->m_pkthdr.csum_flags |= M_CSUM_IPv4_BAD;
1277 		}
1278 
1279 		/* Check TCP/UDP checksum */
1280 		if (RE_TCPPKT(rxstat)) {
1281 			m->m_pkthdr.csum_flags |= M_CSUM_TCPv4;
1282 			if (rxstat & RE_RDESC_STAT_TCPSUMBAD)
1283 				m->m_pkthdr.csum_flags |= M_CSUM_TCP_UDP_BAD;
1284 		} else if (RE_UDPPKT(rxstat)) {
1285 			m->m_pkthdr.csum_flags |= M_CSUM_UDPv4;
1286 			if (rxstat & RE_RDESC_STAT_UDPSUMBAD)
1287 				m->m_pkthdr.csum_flags |= M_CSUM_TCP_UDP_BAD;
1288 		}
1289 
1290 		if (rxvlan & RE_RDESC_VLANCTL_TAG) {
1291 			VLAN_INPUT_TAG(ifp, m,
1292 			     bswap16(rxvlan & RE_RDESC_VLANCTL_DATA),
1293 			     continue);
1294 		}
1295 #if NBPFILTER > 0
1296 		if (ifp->if_bpf)
1297 			bpf_mtap(ifp->if_bpf, m);
1298 #endif
1299 		(*ifp->if_input)(ifp, m);
1300 	}
1301 
1302 	sc->re_ldata.re_rx_prodidx = i;
1303 }
1304 
1305 static void
1306 re_txeof(struct rtk_softc *sc)
1307 {
1308 	struct ifnet *ifp;
1309 	struct re_txq *txq;
1310 	uint32_t txstat;
1311 	int idx, descidx;
1312 
1313 	ifp = &sc->ethercom.ec_if;
1314 
1315 	for (idx = sc->re_ldata.re_txq_considx;
1316 	    sc->re_ldata.re_txq_free < RE_TX_QLEN;
1317 	    idx = RE_NEXT_TXQ(sc, idx), sc->re_ldata.re_txq_free++) {
1318 		txq = &sc->re_ldata.re_txq[idx];
1319 		KASSERT(txq->txq_mbuf != NULL);
1320 
1321 		descidx = txq->txq_descidx;
1322 		RE_TXDESCSYNC(sc, descidx,
1323 		    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1324 		txstat =
1325 		    le32toh(sc->re_ldata.re_tx_list[descidx].re_cmdstat);
1326 		RE_TXDESCSYNC(sc, descidx, BUS_DMASYNC_PREREAD);
1327 		KASSERT((txstat & RE_TDESC_CMD_EOF) != 0);
1328 		if (txstat & RE_TDESC_CMD_OWN) {
1329 			break;
1330 		}
1331 
1332 		sc->re_ldata.re_tx_free += txq->txq_nsegs;
1333 		KASSERT(sc->re_ldata.re_tx_free <= RE_TX_DESC_CNT(sc));
1334 		bus_dmamap_sync(sc->sc_dmat, txq->txq_dmamap,
1335 		    0, txq->txq_dmamap->dm_mapsize, BUS_DMASYNC_POSTWRITE);
1336 		bus_dmamap_unload(sc->sc_dmat, txq->txq_dmamap);
1337 		m_freem(txq->txq_mbuf);
1338 		txq->txq_mbuf = NULL;
1339 
1340 		if (txstat & (RE_TDESC_STAT_EXCESSCOL | RE_TDESC_STAT_COLCNT))
1341 			ifp->if_collisions++;
1342 		if (txstat & RE_TDESC_STAT_TXERRSUM)
1343 			ifp->if_oerrors++;
1344 		else
1345 			ifp->if_opackets++;
1346 	}
1347 
1348 	sc->re_ldata.re_txq_considx = idx;
1349 
1350 	if (sc->re_ldata.re_txq_free > RE_NTXDESC_RSVD)
1351 		ifp->if_flags &= ~IFF_OACTIVE;
1352 
1353 	/*
1354 	 * If not all descriptors have been released reaped yet,
1355 	 * reload the timer so that we will eventually get another
1356 	 * interrupt that will cause us to re-enter this routine.
1357 	 * This is done in case the transmitter has gone idle.
1358 	 */
1359 	if (sc->re_ldata.re_txq_free < RE_TX_QLEN) {
1360 		CSR_WRITE_4(sc, RTK_TIMERCNT, 1);
1361 		if ((sc->sc_quirk & RTKQ_PCIE) != 0) {
1362 			/*
1363 			 * Some chips will ignore a second TX request
1364 			 * issued while an existing transmission is in
1365 			 * progress. If the transmitter goes idle but
1366 			 * there are still packets waiting to be sent,
1367 			 * we need to restart the channel here to flush
1368 			 * them out. This only seems to be required with
1369 			 * the PCIe devices.
1370 			 */
1371 			CSR_WRITE_1(sc, RTK_GTXSTART, RTK_TXSTART_START);
1372 		}
1373 	} else
1374 		ifp->if_timer = 0;
1375 }
1376 
1377 static void
1378 re_tick(void *arg)
1379 {
1380 	struct rtk_softc *sc = arg;
1381 	int s;
1382 
1383 	/*XXX: just return for 8169S/8110S with rev 2 or newer phy */
1384 	s = splnet();
1385 
1386 	mii_tick(&sc->mii);
1387 	splx(s);
1388 
1389 	callout_reset(&sc->rtk_tick_ch, hz, re_tick, sc);
1390 }
1391 
1392 int
1393 re_intr(void *arg)
1394 {
1395 	struct rtk_softc *sc = arg;
1396 	struct ifnet *ifp;
1397 	uint16_t status;
1398 	int handled = 0;
1399 
1400 	if (!device_has_power(sc->sc_dev))
1401 		return 0;
1402 
1403 	ifp = &sc->ethercom.ec_if;
1404 
1405 	if ((ifp->if_flags & IFF_UP) == 0)
1406 		return 0;
1407 
1408 	for (;;) {
1409 
1410 		status = CSR_READ_2(sc, RTK_ISR);
1411 		/* If the card has gone away the read returns 0xffff. */
1412 		if (status == 0xffff)
1413 			break;
1414 		if (status) {
1415 			handled = 1;
1416 			CSR_WRITE_2(sc, RTK_ISR, status);
1417 		}
1418 
1419 		if ((status & RTK_INTRS_CPLUS) == 0)
1420 			break;
1421 
1422 		if (status & (RTK_ISR_RX_OK | RTK_ISR_RX_ERR))
1423 			re_rxeof(sc);
1424 
1425 		if (status & (RTK_ISR_TIMEOUT_EXPIRED | RTK_ISR_TX_ERR |
1426 		    RTK_ISR_TX_DESC_UNAVAIL))
1427 			re_txeof(sc);
1428 
1429 		if (status & RTK_ISR_SYSTEM_ERR) {
1430 			re_init(ifp);
1431 		}
1432 
1433 		if (status & RTK_ISR_LINKCHG) {
1434 			callout_stop(&sc->rtk_tick_ch);
1435 			re_tick(sc);
1436 		}
1437 	}
1438 
1439 	if (handled && !IFQ_IS_EMPTY(&ifp->if_snd))
1440 		re_start(ifp);
1441 
1442 	return handled;
1443 }
1444 
1445 
1446 
1447 /*
1448  * Main transmit routine for C+ and gigE NICs.
1449  */
1450 
1451 static void
1452 re_start(struct ifnet *ifp)
1453 {
1454 	struct rtk_softc *sc;
1455 	struct mbuf *m;
1456 	bus_dmamap_t map;
1457 	struct re_txq *txq;
1458 	struct re_desc *d;
1459 	struct m_tag *mtag;
1460 	uint32_t cmdstat, re_flags, vlanctl;
1461 	int ofree, idx, error, nsegs, seg;
1462 	int startdesc, curdesc, lastdesc;
1463 	bool pad;
1464 
1465 	sc = ifp->if_softc;
1466 	ofree = sc->re_ldata.re_txq_free;
1467 
1468 	for (idx = sc->re_ldata.re_txq_prodidx;; idx = RE_NEXT_TXQ(sc, idx)) {
1469 
1470 		IFQ_POLL(&ifp->if_snd, m);
1471 		if (m == NULL)
1472 			break;
1473 
1474 		if (sc->re_ldata.re_txq_free == 0 ||
1475 		    sc->re_ldata.re_tx_free == 0) {
1476 			/* no more free slots left */
1477 			ifp->if_flags |= IFF_OACTIVE;
1478 			break;
1479 		}
1480 
1481 		/*
1482 		 * Set up checksum offload. Note: checksum offload bits must
1483 		 * appear in all descriptors of a multi-descriptor transmit
1484 		 * attempt. (This is according to testing done with an 8169
1485 		 * chip. I'm not sure if this is a requirement or a bug.)
1486 		 */
1487 
1488 		vlanctl = 0;
1489 		if ((m->m_pkthdr.csum_flags & M_CSUM_TSOv4) != 0) {
1490 			uint32_t segsz = m->m_pkthdr.segsz;
1491 
1492 			re_flags = RE_TDESC_CMD_LGSEND |
1493 			    (segsz << RE_TDESC_CMD_MSSVAL_SHIFT);
1494 		} else {
1495 			/*
1496 			 * set RE_TDESC_CMD_IPCSUM if any checksum offloading
1497 			 * is requested.  otherwise, RE_TDESC_CMD_TCPCSUM/
1498 			 * RE_TDESC_CMD_UDPCSUM doesn't make effects.
1499 			 */
1500 			re_flags = 0;
1501 			if ((m->m_pkthdr.csum_flags &
1502 			    (M_CSUM_IPv4 | M_CSUM_TCPv4 | M_CSUM_UDPv4))
1503 			    != 0) {
1504 				if ((sc->sc_quirk & RTKQ_DESCV2) == 0) {
1505 					re_flags |= RE_TDESC_CMD_IPCSUM;
1506 					if (m->m_pkthdr.csum_flags &
1507 					    M_CSUM_TCPv4) {
1508 						re_flags |=
1509 						    RE_TDESC_CMD_TCPCSUM;
1510 					} else if (m->m_pkthdr.csum_flags &
1511 					    M_CSUM_UDPv4) {
1512 						re_flags |=
1513 						    RE_TDESC_CMD_UDPCSUM;
1514 					}
1515 				} else {
1516 					vlanctl |= RE_TDESC_VLANCTL_IPCSUM;
1517 					if (m->m_pkthdr.csum_flags &
1518 					    M_CSUM_TCPv4) {
1519 						vlanctl |=
1520 						    RE_TDESC_VLANCTL_TCPCSUM;
1521 					} else if (m->m_pkthdr.csum_flags &
1522 					    M_CSUM_UDPv4) {
1523 						vlanctl |=
1524 						    RE_TDESC_VLANCTL_UDPCSUM;
1525 					}
1526 				}
1527 			}
1528 		}
1529 
1530 		txq = &sc->re_ldata.re_txq[idx];
1531 		map = txq->txq_dmamap;
1532 		error = bus_dmamap_load_mbuf(sc->sc_dmat, map, m,
1533 		    BUS_DMA_WRITE|BUS_DMA_NOWAIT);
1534 
1535 		if (__predict_false(error)) {
1536 			/* XXX try to defrag if EFBIG? */
1537 			printf("%s: can't map mbuf (error %d)\n",
1538 			    device_xname(sc->sc_dev), error);
1539 
1540 			IFQ_DEQUEUE(&ifp->if_snd, m);
1541 			m_freem(m);
1542 			ifp->if_oerrors++;
1543 			continue;
1544 		}
1545 
1546 		nsegs = map->dm_nsegs;
1547 		pad = false;
1548 		if (__predict_false(m->m_pkthdr.len <= RE_IP4CSUMTX_PADLEN &&
1549 		    (re_flags & RE_TDESC_CMD_IPCSUM) != 0 &&
1550 		    (sc->sc_quirk & RTKQ_DESCV2) == 0)) {
1551 			pad = true;
1552 			nsegs++;
1553 		}
1554 
1555 		if (nsegs > sc->re_ldata.re_tx_free) {
1556 			/*
1557 			 * Not enough free descriptors to transmit this packet.
1558 			 */
1559 			ifp->if_flags |= IFF_OACTIVE;
1560 			bus_dmamap_unload(sc->sc_dmat, map);
1561 			break;
1562 		}
1563 
1564 		IFQ_DEQUEUE(&ifp->if_snd, m);
1565 
1566 		/*
1567 		 * Make sure that the caches are synchronized before we
1568 		 * ask the chip to start DMA for the packet data.
1569 		 */
1570 		bus_dmamap_sync(sc->sc_dmat, map, 0, map->dm_mapsize,
1571 		    BUS_DMASYNC_PREWRITE);
1572 
1573 		/*
1574 		 * Set up hardware VLAN tagging. Note: vlan tag info must
1575 		 * appear in all descriptors of a multi-descriptor
1576 		 * transmission attempt.
1577 		 */
1578 		if ((mtag = VLAN_OUTPUT_TAG(&sc->ethercom, m)) != NULL)
1579 			vlanctl |= bswap16(VLAN_TAG_VALUE(mtag)) |
1580 			    RE_TDESC_VLANCTL_TAG;
1581 
1582 		/*
1583 		 * Map the segment array into descriptors.
1584 		 * Note that we set the start-of-frame and
1585 		 * end-of-frame markers for either TX or RX,
1586 		 * but they really only have meaning in the TX case.
1587 		 * (In the RX case, it's the chip that tells us
1588 		 *  where packets begin and end.)
1589 		 * We also keep track of the end of the ring
1590 		 * and set the end-of-ring bits as needed,
1591 		 * and we set the ownership bits in all except
1592 		 * the very first descriptor. (The caller will
1593 		 * set this descriptor later when it start
1594 		 * transmission or reception.)
1595 		 */
1596 		curdesc = startdesc = sc->re_ldata.re_tx_nextfree;
1597 		lastdesc = -1;
1598 		for (seg = 0; seg < map->dm_nsegs;
1599 		    seg++, curdesc = RE_NEXT_TX_DESC(sc, curdesc)) {
1600 			d = &sc->re_ldata.re_tx_list[curdesc];
1601 #ifdef DIAGNOSTIC
1602 			RE_TXDESCSYNC(sc, curdesc,
1603 			    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1604 			cmdstat = le32toh(d->re_cmdstat);
1605 			RE_TXDESCSYNC(sc, curdesc, BUS_DMASYNC_PREREAD);
1606 			if (cmdstat & RE_TDESC_STAT_OWN) {
1607 				panic("%s: tried to map busy TX descriptor",
1608 				    device_xname(sc->sc_dev));
1609 			}
1610 #endif
1611 
1612 			d->re_vlanctl = htole32(vlanctl);
1613 			re_set_bufaddr(d, map->dm_segs[seg].ds_addr);
1614 			cmdstat = re_flags | map->dm_segs[seg].ds_len;
1615 			if (seg == 0)
1616 				cmdstat |= RE_TDESC_CMD_SOF;
1617 			else
1618 				cmdstat |= RE_TDESC_CMD_OWN;
1619 			if (curdesc == (RE_TX_DESC_CNT(sc) - 1))
1620 				cmdstat |= RE_TDESC_CMD_EOR;
1621 			if (seg == nsegs - 1) {
1622 				cmdstat |= RE_TDESC_CMD_EOF;
1623 				lastdesc = curdesc;
1624 			}
1625 			d->re_cmdstat = htole32(cmdstat);
1626 			RE_TXDESCSYNC(sc, curdesc,
1627 			    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1628 		}
1629 		if (__predict_false(pad)) {
1630 			bus_addr_t paddaddr;
1631 
1632 			d = &sc->re_ldata.re_tx_list[curdesc];
1633 			d->re_vlanctl = htole32(vlanctl);
1634 			paddaddr = RE_TXPADDADDR(sc);
1635 			re_set_bufaddr(d, paddaddr);
1636 			cmdstat = re_flags |
1637 			    RE_TDESC_CMD_OWN | RE_TDESC_CMD_EOF |
1638 			    (RE_IP4CSUMTX_PADLEN + 1 - m->m_pkthdr.len);
1639 			if (curdesc == (RE_TX_DESC_CNT(sc) - 1))
1640 				cmdstat |= RE_TDESC_CMD_EOR;
1641 			d->re_cmdstat = htole32(cmdstat);
1642 			RE_TXDESCSYNC(sc, curdesc,
1643 			    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1644 			lastdesc = curdesc;
1645 			curdesc = RE_NEXT_TX_DESC(sc, curdesc);
1646 		}
1647 		KASSERT(lastdesc != -1);
1648 
1649 		/* Transfer ownership of packet to the chip. */
1650 
1651 		sc->re_ldata.re_tx_list[startdesc].re_cmdstat |=
1652 		    htole32(RE_TDESC_CMD_OWN);
1653 		RE_TXDESCSYNC(sc, startdesc,
1654 		    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1655 
1656 		/* update info of TX queue and descriptors */
1657 		txq->txq_mbuf = m;
1658 		txq->txq_descidx = lastdesc;
1659 		txq->txq_nsegs = nsegs;
1660 
1661 		sc->re_ldata.re_txq_free--;
1662 		sc->re_ldata.re_tx_free -= nsegs;
1663 		sc->re_ldata.re_tx_nextfree = curdesc;
1664 
1665 #if NBPFILTER > 0
1666 		/*
1667 		 * If there's a BPF listener, bounce a copy of this frame
1668 		 * to him.
1669 		 */
1670 		if (ifp->if_bpf)
1671 			bpf_mtap(ifp->if_bpf, m);
1672 #endif
1673 	}
1674 
1675 	if (sc->re_ldata.re_txq_free < ofree) {
1676 		/*
1677 		 * TX packets are enqueued.
1678 		 */
1679 		sc->re_ldata.re_txq_prodidx = idx;
1680 
1681 		/*
1682 		 * Start the transmitter to poll.
1683 		 *
1684 		 * RealTek put the TX poll request register in a different
1685 		 * location on the 8169 gigE chip. I don't know why.
1686 		 */
1687 		if ((sc->sc_quirk & RTKQ_8139CPLUS) != 0)
1688 			CSR_WRITE_1(sc, RTK_TXSTART, RTK_TXSTART_START);
1689 		else
1690 			CSR_WRITE_1(sc, RTK_GTXSTART, RTK_TXSTART_START);
1691 
1692 		/*
1693 		 * Use the countdown timer for interrupt moderation.
1694 		 * 'TX done' interrupts are disabled. Instead, we reset the
1695 		 * countdown timer, which will begin counting until it hits
1696 		 * the value in the TIMERINT register, and then trigger an
1697 		 * interrupt. Each time we write to the TIMERCNT register,
1698 		 * the timer count is reset to 0.
1699 		 */
1700 		CSR_WRITE_4(sc, RTK_TIMERCNT, 1);
1701 
1702 		/*
1703 		 * Set a timeout in case the chip goes out to lunch.
1704 		 */
1705 		ifp->if_timer = 5;
1706 	}
1707 }
1708 
1709 static int
1710 re_init(struct ifnet *ifp)
1711 {
1712 	struct rtk_softc *sc = ifp->if_softc;
1713 	const uint8_t *enaddr;
1714 	uint32_t rxcfg = 0;
1715 	uint32_t reg;
1716 	int error;
1717 
1718 	if ((error = re_enable(sc)) != 0)
1719 		goto out;
1720 
1721 	/*
1722 	 * Cancel pending I/O and free all RX/TX buffers.
1723 	 */
1724 	re_stop(ifp, 0);
1725 
1726 	re_reset(sc);
1727 
1728 	/*
1729 	 * Enable C+ RX and TX mode, as well as VLAN stripping and
1730 	 * RX checksum offload. We must configure the C+ register
1731 	 * before all others.
1732 	 */
1733 	reg = 0;
1734 
1735 	/*
1736 	 * XXX: Realtek docs say bits 0 and 1 are reserved, for 8169S/8110S.
1737 	 * FreeBSD  drivers set these bits anyway (for 8139C+?).
1738 	 * So far, it works.
1739 	 */
1740 
1741 	/*
1742 	 * XXX: For old 8169 set bit 14.
1743 	 *      For 8169S/8110S and above, do not set bit 14.
1744 	 */
1745 	if ((sc->sc_quirk & RTKQ_8169NONS) != 0)
1746 		reg |= (0x1 << 14) | RTK_CPLUSCMD_PCI_MRW;
1747 
1748 	if (1)  {/* not for 8169S ? */
1749 		reg |=
1750 		    RTK_CPLUSCMD_VLANSTRIP |
1751 		    (ifp->if_capenable &
1752 		    (IFCAP_CSUM_IPv4_Rx | IFCAP_CSUM_TCPv4_Rx |
1753 		     IFCAP_CSUM_UDPv4_Rx) ?
1754 		    RTK_CPLUSCMD_RXCSUM_ENB : 0);
1755 	}
1756 
1757 	CSR_WRITE_2(sc, RTK_CPLUS_CMD,
1758 	    reg | RTK_CPLUSCMD_RXENB | RTK_CPLUSCMD_TXENB);
1759 
1760 	/* XXX: from Realtek-supplied Linux driver. Wholly undocumented. */
1761 	if ((sc->sc_quirk & RTKQ_8139CPLUS) == 0)
1762 		CSR_WRITE_2(sc, RTK_IM, 0x0000);
1763 
1764 	DELAY(10000);
1765 
1766 	/*
1767 	 * Init our MAC address.  Even though the chipset
1768 	 * documentation doesn't mention it, we need to enter "Config
1769 	 * register write enable" mode to modify the ID registers.
1770 	 */
1771 	CSR_WRITE_1(sc, RTK_EECMD, RTK_EEMODE_WRITECFG);
1772 	enaddr = CLLADDR(ifp->if_sadl);
1773 	reg = enaddr[0] | (enaddr[1] << 8) |
1774 	    (enaddr[2] << 16) | (enaddr[3] << 24);
1775 	CSR_WRITE_4(sc, RTK_IDR0, reg);
1776 	reg = enaddr[4] | (enaddr[5] << 8);
1777 	CSR_WRITE_4(sc, RTK_IDR4, reg);
1778 	CSR_WRITE_1(sc, RTK_EECMD, RTK_EEMODE_OFF);
1779 
1780 	/*
1781 	 * For C+ mode, initialize the RX descriptors and mbufs.
1782 	 */
1783 	re_rx_list_init(sc);
1784 	re_tx_list_init(sc);
1785 
1786 	/*
1787 	 * Load the addresses of the RX and TX lists into the chip.
1788 	 */
1789 	CSR_WRITE_4(sc, RTK_RXLIST_ADDR_HI,
1790 	    RE_ADDR_HI(sc->re_ldata.re_rx_list_map->dm_segs[0].ds_addr));
1791 	CSR_WRITE_4(sc, RTK_RXLIST_ADDR_LO,
1792 	    RE_ADDR_LO(sc->re_ldata.re_rx_list_map->dm_segs[0].ds_addr));
1793 
1794 	CSR_WRITE_4(sc, RTK_TXLIST_ADDR_HI,
1795 	    RE_ADDR_HI(sc->re_ldata.re_tx_list_map->dm_segs[0].ds_addr));
1796 	CSR_WRITE_4(sc, RTK_TXLIST_ADDR_LO,
1797 	    RE_ADDR_LO(sc->re_ldata.re_tx_list_map->dm_segs[0].ds_addr));
1798 
1799 	/*
1800 	 * Enable transmit and receive.
1801 	 */
1802 	CSR_WRITE_1(sc, RTK_COMMAND, RTK_CMD_TX_ENB | RTK_CMD_RX_ENB);
1803 
1804 	/*
1805 	 * Set the initial TX and RX configuration.
1806 	 */
1807 	if (sc->re_testmode && (sc->sc_quirk & RTKQ_8169NONS) != 0) {
1808 		/* test mode is needed only for old 8169 */
1809 		CSR_WRITE_4(sc, RTK_TXCFG,
1810 		    RE_TXCFG_CONFIG | RTK_LOOPTEST_ON);
1811 	} else
1812 		CSR_WRITE_4(sc, RTK_TXCFG, RE_TXCFG_CONFIG);
1813 
1814 	CSR_WRITE_1(sc, RTK_EARLY_TX_THRESH, 16);
1815 
1816 	CSR_WRITE_4(sc, RTK_RXCFG, RE_RXCFG_CONFIG);
1817 
1818 	/* Set the individual bit to receive frames for this host only. */
1819 	rxcfg = CSR_READ_4(sc, RTK_RXCFG);
1820 	rxcfg |= RTK_RXCFG_RX_INDIV;
1821 
1822 	/* If we want promiscuous mode, set the allframes bit. */
1823 	if (ifp->if_flags & IFF_PROMISC)
1824 		rxcfg |= RTK_RXCFG_RX_ALLPHYS;
1825 	else
1826 		rxcfg &= ~RTK_RXCFG_RX_ALLPHYS;
1827 	CSR_WRITE_4(sc, RTK_RXCFG, rxcfg);
1828 
1829 	/*
1830 	 * Set capture broadcast bit to capture broadcast frames.
1831 	 */
1832 	if (ifp->if_flags & IFF_BROADCAST)
1833 		rxcfg |= RTK_RXCFG_RX_BROAD;
1834 	else
1835 		rxcfg &= ~RTK_RXCFG_RX_BROAD;
1836 	CSR_WRITE_4(sc, RTK_RXCFG, rxcfg);
1837 
1838 	/*
1839 	 * Program the multicast filter, if necessary.
1840 	 */
1841 	rtk_setmulti(sc);
1842 
1843 	/*
1844 	 * Enable interrupts.
1845 	 */
1846 	if (sc->re_testmode)
1847 		CSR_WRITE_2(sc, RTK_IMR, 0);
1848 	else
1849 		CSR_WRITE_2(sc, RTK_IMR, RTK_INTRS_CPLUS);
1850 
1851 	/* Start RX/TX process. */
1852 	CSR_WRITE_4(sc, RTK_MISSEDPKT, 0);
1853 #ifdef notdef
1854 	/* Enable receiver and transmitter. */
1855 	CSR_WRITE_1(sc, RTK_COMMAND, RTK_CMD_TX_ENB | RTK_CMD_RX_ENB);
1856 #endif
1857 
1858 	/*
1859 	 * Initialize the timer interrupt register so that
1860 	 * a timer interrupt will be generated once the timer
1861 	 * reaches a certain number of ticks. The timer is
1862 	 * reloaded on each transmit. This gives us TX interrupt
1863 	 * moderation, which dramatically improves TX frame rate.
1864 	 */
1865 
1866 	if ((sc->sc_quirk & RTKQ_8139CPLUS) != 0)
1867 		CSR_WRITE_4(sc, RTK_TIMERINT, 0x400);
1868 	else {
1869 		CSR_WRITE_4(sc, RTK_TIMERINT_8169, 0x800);
1870 
1871 		/*
1872 		 * For 8169 gigE NICs, set the max allowed RX packet
1873 		 * size so we can receive jumbo frames.
1874 		 */
1875 		CSR_WRITE_2(sc, RTK_MAXRXPKTLEN, 16383);
1876 	}
1877 
1878 	if (sc->re_testmode)
1879 		return 0;
1880 
1881 	CSR_WRITE_1(sc, RTK_CFG1, RTK_CFG1_DRVLOAD);
1882 
1883 	ifp->if_flags |= IFF_RUNNING;
1884 	ifp->if_flags &= ~IFF_OACTIVE;
1885 
1886 	callout_reset(&sc->rtk_tick_ch, hz, re_tick, sc);
1887 
1888  out:
1889 	if (error) {
1890 		ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
1891 		ifp->if_timer = 0;
1892 		printf("%s: interface not running\n",
1893 		    device_xname(sc->sc_dev));
1894 	}
1895 
1896 	return error;
1897 }
1898 
1899 static int
1900 re_ioctl(struct ifnet *ifp, u_long command, void *data)
1901 {
1902 	struct rtk_softc *sc = ifp->if_softc;
1903 	struct ifreq *ifr = data;
1904 	int s, error = 0;
1905 
1906 	s = splnet();
1907 
1908 	switch (command) {
1909 	case SIOCSIFMTU:
1910 		/*
1911 		 * Disable jumbo frames if it's not supported.
1912 		 */
1913 		if ((sc->sc_quirk & RTKQ_NOJUMBO) != 0 &&
1914 		    ifr->ifr_mtu > ETHERMTU) {
1915 			error = EINVAL;
1916 			break;
1917 		}
1918 
1919 		if (ifr->ifr_mtu < ETHERMIN || ifr->ifr_mtu > ETHERMTU_JUMBO)
1920 			error = EINVAL;
1921 		else if ((error = ifioctl_common(ifp, command, data)) ==
1922 		    ENETRESET)
1923 			error = 0;
1924 		break;
1925 	default:
1926 		if ((error = ether_ioctl(ifp, command, data)) != ENETRESET)
1927 			break;
1928 
1929 		error = 0;
1930 
1931 		if (command == SIOCSIFCAP)
1932 			error = (*ifp->if_init)(ifp);
1933 		else if (command != SIOCADDMULTI && command != SIOCDELMULTI)
1934 			;
1935 		else if (ifp->if_flags & IFF_RUNNING)
1936 			rtk_setmulti(sc);
1937 		break;
1938 	}
1939 
1940 	splx(s);
1941 
1942 	return error;
1943 }
1944 
1945 static void
1946 re_watchdog(struct ifnet *ifp)
1947 {
1948 	struct rtk_softc *sc;
1949 	int s;
1950 
1951 	sc = ifp->if_softc;
1952 	s = splnet();
1953 	printf("%s: watchdog timeout\n", device_xname(sc->sc_dev));
1954 	ifp->if_oerrors++;
1955 
1956 	re_txeof(sc);
1957 	re_rxeof(sc);
1958 
1959 	re_init(ifp);
1960 
1961 	splx(s);
1962 }
1963 
1964 /*
1965  * Stop the adapter and free any mbufs allocated to the
1966  * RX and TX lists.
1967  */
1968 static void
1969 re_stop(struct ifnet *ifp, int disable)
1970 {
1971 	int i;
1972 	struct rtk_softc *sc = ifp->if_softc;
1973 
1974 	callout_stop(&sc->rtk_tick_ch);
1975 
1976 	mii_down(&sc->mii);
1977 
1978 	CSR_WRITE_1(sc, RTK_COMMAND, 0x00);
1979 	CSR_WRITE_2(sc, RTK_IMR, 0x0000);
1980 
1981 	if (sc->re_head != NULL) {
1982 		m_freem(sc->re_head);
1983 		sc->re_head = sc->re_tail = NULL;
1984 	}
1985 
1986 	/* Free the TX list buffers. */
1987 	for (i = 0; i < RE_TX_QLEN; i++) {
1988 		if (sc->re_ldata.re_txq[i].txq_mbuf != NULL) {
1989 			bus_dmamap_unload(sc->sc_dmat,
1990 			    sc->re_ldata.re_txq[i].txq_dmamap);
1991 			m_freem(sc->re_ldata.re_txq[i].txq_mbuf);
1992 			sc->re_ldata.re_txq[i].txq_mbuf = NULL;
1993 		}
1994 	}
1995 
1996 	/* Free the RX list buffers. */
1997 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
1998 		if (sc->re_ldata.re_rxsoft[i].rxs_mbuf != NULL) {
1999 			bus_dmamap_unload(sc->sc_dmat,
2000 			    sc->re_ldata.re_rxsoft[i].rxs_dmamap);
2001 			m_freem(sc->re_ldata.re_rxsoft[i].rxs_mbuf);
2002 			sc->re_ldata.re_rxsoft[i].rxs_mbuf = NULL;
2003 		}
2004 	}
2005 
2006 	if (disable)
2007 		re_disable(sc);
2008 
2009 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
2010 	ifp->if_timer = 0;
2011 }
2012