xref: /netbsd-src/sys/dev/ic/rtl8169.c (revision 93bf6008f8b7982c1d1a9486e4a4a0e687fe36eb)
1 /*	$NetBSD: rtl8169.c,v 1.116 2009/04/13 12:38:06 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 1.116 2009/04/13 12:38:06 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 		case RTK_HWREV_8168CP:
606 		case RTK_HWREV_8168D:
607 			sc->sc_rev = 24;
608 			sc->sc_quirk |= RTKQ_DESCV2 | RTKQ_NOEECMD;
609 			/*
610 			 * From FreeBSD driver:
611 			 *
612 			 * These (8168/8111) controllers support jumbo frame
613 			 * but it seems that enabling it requires touching
614 			 * additional magic registers. Depending on MAC
615 			 * revisions some controllers need to disable
616 			 * checksum offload. So disable jumbo frame until
617 			 * I have better idea what it really requires to
618 			 * make it support.
619 			 * RTL8168C/CP : supports up to 6KB jumbo frame.
620 			 * RTL8111C/CP : supports up to 9KB jumbo frame.
621 			 */
622 			sc->sc_quirk |= RTKQ_NOJUMBO;
623 			break;
624 		case RTK_HWREV_8102E:
625 		case RTK_HWREV_8102EL:
626 		case RTK_HWREV_8102EL_SPIN2:
627 			sc->sc_rev = 25;
628 			sc->sc_quirk |=
629 			    RTKQ_DESCV2 | RTKQ_NOEECMD | RTKQ_NOJUMBO;
630 			break;
631 		case RTK_HWREV_8100E:
632 		case RTK_HWREV_8100E_SPIN2:
633 			/* XXX not in the Realtek driver */
634 			sc->sc_rev = 0;
635 			sc->sc_quirk |= RTKQ_NOJUMBO;
636 			break;
637 		default:
638 			aprint_normal_dev(sc->sc_dev,
639 			    "Unknown revision (0x%08x)\n", hwrev);
640 			sc->sc_rev = 0;
641 			/* assume the latest features */
642 			sc->sc_quirk |= RTKQ_DESCV2 | RTKQ_NOEECMD;
643 			sc->sc_quirk |= RTKQ_NOJUMBO;
644 		}
645 
646 		/* Set RX length mask */
647 		sc->re_rxlenmask = RE_RDESC_STAT_GFRAGLEN;
648 		sc->re_ldata.re_tx_desc_cnt = RE_TX_DESC_CNT_8169;
649 	} else {
650 		sc->sc_quirk |= RTKQ_NOJUMBO;
651 
652 		/* Set RX length mask */
653 		sc->re_rxlenmask = RE_RDESC_STAT_FRAGLEN;
654 		sc->re_ldata.re_tx_desc_cnt = RE_TX_DESC_CNT_8139;
655 	}
656 
657 	/* Reset the adapter. */
658 	re_reset(sc);
659 
660 	if ((sc->sc_quirk & RTKQ_NOEECMD) != 0) {
661 		/*
662 		 * Get station address from ID registers.
663 		 */
664 		for (i = 0; i < ETHER_ADDR_LEN; i++)
665 			eaddr[i] = CSR_READ_1(sc, RTK_IDR0 + i);
666 	} else {
667 		/*
668 		 * Get station address from the EEPROM.
669 		 */
670 		if (rtk_read_eeprom(sc, RTK_EE_ID, RTK_EEADDR_LEN1) == 0x8129)
671 			addr_len = RTK_EEADDR_LEN1;
672 		else
673 			addr_len = RTK_EEADDR_LEN0;
674 
675 		/*
676 		 * Get station address from the EEPROM.
677 		 */
678 		for (i = 0; i < ETHER_ADDR_LEN / 2; i++) {
679 			val = rtk_read_eeprom(sc, RTK_EE_EADDR0 + i, addr_len);
680 			eaddr[(i * 2) + 0] = val & 0xff;
681 			eaddr[(i * 2) + 1] = val >> 8;
682 		}
683 	}
684 
685 	aprint_normal_dev(sc->sc_dev, "Ethernet address %s\n",
686 	    ether_sprintf(eaddr));
687 
688 	if (sc->re_ldata.re_tx_desc_cnt >
689 	    PAGE_SIZE / sizeof(struct re_desc)) {
690 		sc->re_ldata.re_tx_desc_cnt =
691 		    PAGE_SIZE / sizeof(struct re_desc);
692 	}
693 
694 	aprint_verbose_dev(sc->sc_dev, "using %d tx descriptors\n",
695 	    sc->re_ldata.re_tx_desc_cnt);
696 	KASSERT(RE_NEXT_TX_DESC(sc, RE_TX_DESC_CNT(sc) - 1) == 0);
697 
698 	/* Allocate DMA'able memory for the TX ring */
699 	if ((error = bus_dmamem_alloc(sc->sc_dmat, RE_TX_LIST_SZ(sc),
700 	    RE_RING_ALIGN, 0, &sc->re_ldata.re_tx_listseg, 1,
701 	    &sc->re_ldata.re_tx_listnseg, BUS_DMA_NOWAIT)) != 0) {
702 		aprint_error_dev(sc->sc_dev,
703 		    "can't allocate tx listseg, error = %d\n", error);
704 		goto fail_0;
705 	}
706 
707 	/* Load the map for the TX ring. */
708 	if ((error = bus_dmamem_map(sc->sc_dmat, &sc->re_ldata.re_tx_listseg,
709 	    sc->re_ldata.re_tx_listnseg, RE_TX_LIST_SZ(sc),
710 	    (void **)&sc->re_ldata.re_tx_list,
711 	    BUS_DMA_COHERENT | BUS_DMA_NOWAIT)) != 0) {
712 		aprint_error_dev(sc->sc_dev,
713 		    "can't map tx list, error = %d\n", error);
714 	  	goto fail_1;
715 	}
716 	memset(sc->re_ldata.re_tx_list, 0, RE_TX_LIST_SZ(sc));
717 
718 	if ((error = bus_dmamap_create(sc->sc_dmat, RE_TX_LIST_SZ(sc), 1,
719 	    RE_TX_LIST_SZ(sc), 0, 0,
720 	    &sc->re_ldata.re_tx_list_map)) != 0) {
721 		aprint_error_dev(sc->sc_dev,
722 		    "can't create tx list map, error = %d\n", error);
723 		goto fail_2;
724 	}
725 
726 
727 	if ((error = bus_dmamap_load(sc->sc_dmat,
728 	    sc->re_ldata.re_tx_list_map, sc->re_ldata.re_tx_list,
729 	    RE_TX_LIST_SZ(sc), NULL, BUS_DMA_NOWAIT)) != 0) {
730 		aprint_error_dev(sc->sc_dev,
731 		    "can't load tx list, error = %d\n", error);
732 		goto fail_3;
733 	}
734 
735 	/* Create DMA maps for TX buffers */
736 	for (i = 0; i < RE_TX_QLEN; i++) {
737 		error = bus_dmamap_create(sc->sc_dmat,
738 		    round_page(IP_MAXPACKET),
739 		    RE_TX_DESC_CNT(sc), RE_TDESC_CMD_FRAGLEN,
740 		    0, 0, &sc->re_ldata.re_txq[i].txq_dmamap);
741 		if (error) {
742 			aprint_error_dev(sc->sc_dev,
743 			    "can't create DMA map for TX\n");
744 			goto fail_4;
745 		}
746 	}
747 
748 	/* Allocate DMA'able memory for the RX ring */
749 	/* XXX see also a comment about RE_RX_DMAMEM_SZ in rtl81x9var.h */
750 	if ((error = bus_dmamem_alloc(sc->sc_dmat,
751 	    RE_RX_DMAMEM_SZ, RE_RING_ALIGN, 0, &sc->re_ldata.re_rx_listseg, 1,
752 	    &sc->re_ldata.re_rx_listnseg, BUS_DMA_NOWAIT)) != 0) {
753 		aprint_error_dev(sc->sc_dev,
754 		    "can't allocate rx listseg, error = %d\n", error);
755 		goto fail_4;
756 	}
757 
758 	/* Load the map for the RX ring. */
759 	if ((error = bus_dmamem_map(sc->sc_dmat, &sc->re_ldata.re_rx_listseg,
760 	    sc->re_ldata.re_rx_listnseg, RE_RX_DMAMEM_SZ,
761 	    (void **)&sc->re_ldata.re_rx_list,
762 	    BUS_DMA_COHERENT | BUS_DMA_NOWAIT)) != 0) {
763 		aprint_error_dev(sc->sc_dev,
764 		    "can't map rx list, error = %d\n", error);
765 		goto fail_5;
766 	}
767 	memset(sc->re_ldata.re_rx_list, 0, RE_RX_DMAMEM_SZ);
768 
769 	if ((error = bus_dmamap_create(sc->sc_dmat,
770 	    RE_RX_DMAMEM_SZ, 1, RE_RX_DMAMEM_SZ, 0, 0,
771 	    &sc->re_ldata.re_rx_list_map)) != 0) {
772 		aprint_error_dev(sc->sc_dev,
773 		    "can't create rx list map, error = %d\n", error);
774 		goto fail_6;
775 	}
776 
777 	if ((error = bus_dmamap_load(sc->sc_dmat,
778 	    sc->re_ldata.re_rx_list_map, sc->re_ldata.re_rx_list,
779 	    RE_RX_DMAMEM_SZ, NULL, BUS_DMA_NOWAIT)) != 0) {
780 		aprint_error_dev(sc->sc_dev,
781 		    "can't load rx list, error = %d\n", error);
782 		goto fail_7;
783 	}
784 
785 	/* Create DMA maps for RX buffers */
786 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
787 		error = bus_dmamap_create(sc->sc_dmat, MCLBYTES, 1, MCLBYTES,
788 		    0, 0, &sc->re_ldata.re_rxsoft[i].rxs_dmamap);
789 		if (error) {
790 			aprint_error_dev(sc->sc_dev,
791 			    "can't create DMA map for RX\n");
792 			goto fail_8;
793 		}
794 	}
795 
796 	/*
797 	 * Record interface as attached. From here, we should not fail.
798 	 */
799 	sc->sc_flags |= RTK_ATTACHED;
800 
801 	ifp = &sc->ethercom.ec_if;
802 	ifp->if_softc = sc;
803 	strlcpy(ifp->if_xname, device_xname(sc->sc_dev), IFNAMSIZ);
804 	ifp->if_mtu = ETHERMTU;
805 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
806 	ifp->if_ioctl = re_ioctl;
807 	sc->ethercom.ec_capabilities |=
808 	    ETHERCAP_VLAN_MTU | ETHERCAP_VLAN_HWTAGGING;
809 	ifp->if_start = re_start;
810 	ifp->if_stop = re_stop;
811 
812 	/*
813 	 * IFCAP_CSUM_IPv4_Tx on re(4) is broken for small packets,
814 	 * so we have a workaround to handle the bug by padding
815 	 * such packets manually.
816 	 */
817 	ifp->if_capabilities |=
818 	    IFCAP_CSUM_IPv4_Tx | IFCAP_CSUM_IPv4_Rx |
819 	    IFCAP_CSUM_TCPv4_Tx | IFCAP_CSUM_TCPv4_Rx |
820 	    IFCAP_CSUM_UDPv4_Tx | IFCAP_CSUM_UDPv4_Rx |
821 	    IFCAP_TSOv4;
822 
823 	/*
824 	 * XXX
825 	 * Still have no idea how to make TSO work on 8168C, 8168CP,
826 	 * 8102E, 8111C and 8111CP.
827 	 */
828 	if ((sc->sc_quirk & RTKQ_DESCV2) != 0)
829 		ifp->if_capabilities &= ~IFCAP_TSOv4;
830 
831 	ifp->if_watchdog = re_watchdog;
832 	ifp->if_init = re_init;
833 	ifp->if_snd.ifq_maxlen = RE_IFQ_MAXLEN;
834 	ifp->if_capenable = ifp->if_capabilities;
835 	IFQ_SET_READY(&ifp->if_snd);
836 
837 	callout_init(&sc->rtk_tick_ch, 0);
838 
839 	/* Do MII setup */
840 	sc->mii.mii_ifp = ifp;
841 	sc->mii.mii_readreg = re_miibus_readreg;
842 	sc->mii.mii_writereg = re_miibus_writereg;
843 	sc->mii.mii_statchg = re_miibus_statchg;
844 	sc->ethercom.ec_mii = &sc->mii;
845 	ifmedia_init(&sc->mii.mii_media, IFM_IMASK, ether_mediachange,
846 	    ether_mediastatus);
847 	mii_attach(sc->sc_dev, &sc->mii, 0xffffffff, MII_PHY_ANY,
848 	    MII_OFFSET_ANY, 0);
849 	ifmedia_set(&sc->mii.mii_media, IFM_ETHER | IFM_AUTO);
850 
851 	/*
852 	 * Call MI attach routine.
853 	 */
854 	if_attach(ifp);
855 	ether_ifattach(ifp, eaddr);
856 
857 	return;
858 
859  fail_8:
860 	/* Destroy DMA maps for RX buffers. */
861 	for (i = 0; i < RE_RX_DESC_CNT; i++)
862 		if (sc->re_ldata.re_rxsoft[i].rxs_dmamap != NULL)
863 			bus_dmamap_destroy(sc->sc_dmat,
864 			    sc->re_ldata.re_rxsoft[i].rxs_dmamap);
865 
866 	/* Free DMA'able memory for the RX ring. */
867 	bus_dmamap_unload(sc->sc_dmat, sc->re_ldata.re_rx_list_map);
868  fail_7:
869 	bus_dmamap_destroy(sc->sc_dmat, sc->re_ldata.re_rx_list_map);
870  fail_6:
871 	bus_dmamem_unmap(sc->sc_dmat,
872 	    (void *)sc->re_ldata.re_rx_list, RE_RX_DMAMEM_SZ);
873  fail_5:
874 	bus_dmamem_free(sc->sc_dmat,
875 	    &sc->re_ldata.re_rx_listseg, sc->re_ldata.re_rx_listnseg);
876 
877  fail_4:
878 	/* Destroy DMA maps for TX buffers. */
879 	for (i = 0; i < RE_TX_QLEN; i++)
880 		if (sc->re_ldata.re_txq[i].txq_dmamap != NULL)
881 			bus_dmamap_destroy(sc->sc_dmat,
882 			    sc->re_ldata.re_txq[i].txq_dmamap);
883 
884 	/* Free DMA'able memory for the TX ring. */
885 	bus_dmamap_unload(sc->sc_dmat, sc->re_ldata.re_tx_list_map);
886  fail_3:
887 	bus_dmamap_destroy(sc->sc_dmat, sc->re_ldata.re_tx_list_map);
888  fail_2:
889 	bus_dmamem_unmap(sc->sc_dmat,
890 	    (void *)sc->re_ldata.re_tx_list, RE_TX_LIST_SZ(sc));
891  fail_1:
892 	bus_dmamem_free(sc->sc_dmat,
893 	    &sc->re_ldata.re_tx_listseg, sc->re_ldata.re_tx_listnseg);
894  fail_0:
895 	return;
896 }
897 
898 
899 /*
900  * re_activate:
901  *     Handle device activation/deactivation requests.
902  */
903 int
904 re_activate(device_t self, enum devact act)
905 {
906 	struct rtk_softc *sc = device_private(self);
907 	int s, error = 0;
908 
909 	s = splnet();
910 	switch (act) {
911 	case DVACT_ACTIVATE:
912 		error = EOPNOTSUPP;
913 		break;
914 	case DVACT_DEACTIVATE:
915 		mii_activate(&sc->mii, act, MII_PHY_ANY, MII_OFFSET_ANY);
916 		if_deactivate(&sc->ethercom.ec_if);
917 		break;
918 	}
919 	splx(s);
920 
921 	return error;
922 }
923 
924 /*
925  * re_detach:
926  *     Detach a rtk interface.
927  */
928 int
929 re_detach(struct rtk_softc *sc)
930 {
931 	struct ifnet *ifp = &sc->ethercom.ec_if;
932 	int i;
933 
934 	/*
935 	 * Succeed now if there isn't any work to do.
936 	 */
937 	if ((sc->sc_flags & RTK_ATTACHED) == 0)
938 		return 0;
939 
940 	/* Unhook our tick handler. */
941 	callout_stop(&sc->rtk_tick_ch);
942 
943 	/* Detach all PHYs. */
944 	mii_detach(&sc->mii, MII_PHY_ANY, MII_OFFSET_ANY);
945 
946 	/* Delete all remaining media. */
947 	ifmedia_delete_instance(&sc->mii.mii_media, IFM_INST_ANY);
948 
949 	ether_ifdetach(ifp);
950 	if_detach(ifp);
951 
952 	/* Destroy DMA maps for RX buffers. */
953 	for (i = 0; i < RE_RX_DESC_CNT; i++)
954 		if (sc->re_ldata.re_rxsoft[i].rxs_dmamap != NULL)
955 			bus_dmamap_destroy(sc->sc_dmat,
956 			    sc->re_ldata.re_rxsoft[i].rxs_dmamap);
957 
958 	/* Free DMA'able memory for the RX ring. */
959 	bus_dmamap_unload(sc->sc_dmat, sc->re_ldata.re_rx_list_map);
960 	bus_dmamap_destroy(sc->sc_dmat, sc->re_ldata.re_rx_list_map);
961 	bus_dmamem_unmap(sc->sc_dmat,
962 	    (void *)sc->re_ldata.re_rx_list, RE_RX_DMAMEM_SZ);
963 	bus_dmamem_free(sc->sc_dmat,
964 	    &sc->re_ldata.re_rx_listseg, sc->re_ldata.re_rx_listnseg);
965 
966 	/* Destroy DMA maps for TX buffers. */
967 	for (i = 0; i < RE_TX_QLEN; i++)
968 		if (sc->re_ldata.re_txq[i].txq_dmamap != NULL)
969 			bus_dmamap_destroy(sc->sc_dmat,
970 			    sc->re_ldata.re_txq[i].txq_dmamap);
971 
972 	/* Free DMA'able memory for the TX ring. */
973 	bus_dmamap_unload(sc->sc_dmat, sc->re_ldata.re_tx_list_map);
974 	bus_dmamap_destroy(sc->sc_dmat, sc->re_ldata.re_tx_list_map);
975 	bus_dmamem_unmap(sc->sc_dmat,
976 	    (void *)sc->re_ldata.re_tx_list, RE_TX_LIST_SZ(sc));
977 	bus_dmamem_free(sc->sc_dmat,
978 	    &sc->re_ldata.re_tx_listseg, sc->re_ldata.re_tx_listnseg);
979 
980 	return 0;
981 }
982 
983 /*
984  * re_enable:
985  *     Enable the RTL81X9 chip.
986  */
987 static int
988 re_enable(struct rtk_softc *sc)
989 {
990 
991 	if (RTK_IS_ENABLED(sc) == 0 && sc->sc_enable != NULL) {
992 		if ((*sc->sc_enable)(sc) != 0) {
993 			printf("%s: device enable failed\n",
994 			    device_xname(sc->sc_dev));
995 			return EIO;
996 		}
997 		sc->sc_flags |= RTK_ENABLED;
998 	}
999 	return 0;
1000 }
1001 
1002 /*
1003  * re_disable:
1004  *     Disable the RTL81X9 chip.
1005  */
1006 static void
1007 re_disable(struct rtk_softc *sc)
1008 {
1009 
1010 	if (RTK_IS_ENABLED(sc) && sc->sc_disable != NULL) {
1011 		(*sc->sc_disable)(sc);
1012 		sc->sc_flags &= ~RTK_ENABLED;
1013 	}
1014 }
1015 
1016 static int
1017 re_newbuf(struct rtk_softc *sc, int idx, struct mbuf *m)
1018 {
1019 	struct mbuf *n = NULL;
1020 	bus_dmamap_t map;
1021 	struct re_desc *d;
1022 	struct re_rxsoft *rxs;
1023 	uint32_t cmdstat;
1024 	int error;
1025 
1026 	if (m == NULL) {
1027 		MGETHDR(n, M_DONTWAIT, MT_DATA);
1028 		if (n == NULL)
1029 			return ENOBUFS;
1030 
1031 		MCLGET(n, M_DONTWAIT);
1032 		if ((n->m_flags & M_EXT) == 0) {
1033 			m_freem(n);
1034 			return ENOBUFS;
1035 		}
1036 		m = n;
1037 	} else
1038 		m->m_data = m->m_ext.ext_buf;
1039 
1040 	/*
1041 	 * Initialize mbuf length fields and fixup
1042 	 * alignment so that the frame payload is
1043 	 * longword aligned.
1044 	 */
1045 	m->m_len = m->m_pkthdr.len = MCLBYTES - RE_ETHER_ALIGN;
1046 	m->m_data += RE_ETHER_ALIGN;
1047 
1048 	rxs = &sc->re_ldata.re_rxsoft[idx];
1049 	map = rxs->rxs_dmamap;
1050 	error = bus_dmamap_load_mbuf(sc->sc_dmat, map, m,
1051 	    BUS_DMA_READ|BUS_DMA_NOWAIT);
1052 
1053 	if (error)
1054 		goto out;
1055 
1056 	bus_dmamap_sync(sc->sc_dmat, map, 0, map->dm_mapsize,
1057 	    BUS_DMASYNC_PREREAD);
1058 
1059 	d = &sc->re_ldata.re_rx_list[idx];
1060 #ifdef DIAGNOSTIC
1061 	RE_RXDESCSYNC(sc, idx, BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1062 	cmdstat = le32toh(d->re_cmdstat);
1063 	RE_RXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD);
1064 	if (cmdstat & RE_RDESC_STAT_OWN) {
1065 		panic("%s: tried to map busy RX descriptor",
1066 		    device_xname(sc->sc_dev));
1067 	}
1068 #endif
1069 
1070 	rxs->rxs_mbuf = m;
1071 
1072 	d->re_vlanctl = 0;
1073 	cmdstat = map->dm_segs[0].ds_len;
1074 	if (idx == (RE_RX_DESC_CNT - 1))
1075 		cmdstat |= RE_RDESC_CMD_EOR;
1076 	re_set_bufaddr(d, map->dm_segs[0].ds_addr);
1077 	d->re_cmdstat = htole32(cmdstat);
1078 	RE_RXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1079 	cmdstat |= RE_RDESC_CMD_OWN;
1080 	d->re_cmdstat = htole32(cmdstat);
1081 	RE_RXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1082 
1083 	return 0;
1084  out:
1085 	if (n != NULL)
1086 		m_freem(n);
1087 	return ENOMEM;
1088 }
1089 
1090 static int
1091 re_tx_list_init(struct rtk_softc *sc)
1092 {
1093 	int i;
1094 
1095 	memset(sc->re_ldata.re_tx_list, 0, RE_TX_LIST_SZ(sc));
1096 	for (i = 0; i < RE_TX_QLEN; i++) {
1097 		sc->re_ldata.re_txq[i].txq_mbuf = NULL;
1098 	}
1099 
1100 	bus_dmamap_sync(sc->sc_dmat,
1101 	    sc->re_ldata.re_tx_list_map, 0,
1102 	    sc->re_ldata.re_tx_list_map->dm_mapsize,
1103 	    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1104 	sc->re_ldata.re_txq_prodidx = 0;
1105 	sc->re_ldata.re_txq_considx = 0;
1106 	sc->re_ldata.re_txq_free = RE_TX_QLEN;
1107 	sc->re_ldata.re_tx_free = RE_TX_DESC_CNT(sc);
1108 	sc->re_ldata.re_tx_nextfree = 0;
1109 
1110 	return 0;
1111 }
1112 
1113 static int
1114 re_rx_list_init(struct rtk_softc *sc)
1115 {
1116 	int i;
1117 
1118 	memset(sc->re_ldata.re_rx_list, 0, RE_RX_LIST_SZ);
1119 
1120 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
1121 		if (re_newbuf(sc, i, NULL) == ENOBUFS)
1122 			return ENOBUFS;
1123 	}
1124 
1125 	sc->re_ldata.re_rx_prodidx = 0;
1126 	sc->re_head = sc->re_tail = NULL;
1127 
1128 	return 0;
1129 }
1130 
1131 /*
1132  * RX handler for C+ and 8169. For the gigE chips, we support
1133  * the reception of jumbo frames that have been fragmented
1134  * across multiple 2K mbuf cluster buffers.
1135  */
1136 static void
1137 re_rxeof(struct rtk_softc *sc)
1138 {
1139 	struct mbuf *m;
1140 	struct ifnet *ifp;
1141 	int i, total_len;
1142 	struct re_desc *cur_rx;
1143 	struct re_rxsoft *rxs;
1144 	uint32_t rxstat, rxvlan;
1145 
1146 	ifp = &sc->ethercom.ec_if;
1147 
1148 	for (i = sc->re_ldata.re_rx_prodidx;; i = RE_NEXT_RX_DESC(sc, i)) {
1149 		cur_rx = &sc->re_ldata.re_rx_list[i];
1150 		RE_RXDESCSYNC(sc, i,
1151 		    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1152 		rxstat = le32toh(cur_rx->re_cmdstat);
1153 		rxvlan = le32toh(cur_rx->re_vlanctl);
1154 		RE_RXDESCSYNC(sc, i, BUS_DMASYNC_PREREAD);
1155 		if ((rxstat & RE_RDESC_STAT_OWN) != 0) {
1156 			break;
1157 		}
1158 		total_len = rxstat & sc->re_rxlenmask;
1159 		rxs = &sc->re_ldata.re_rxsoft[i];
1160 		m = rxs->rxs_mbuf;
1161 
1162 		/* Invalidate the RX mbuf and unload its map */
1163 
1164 		bus_dmamap_sync(sc->sc_dmat,
1165 		    rxs->rxs_dmamap, 0, rxs->rxs_dmamap->dm_mapsize,
1166 		    BUS_DMASYNC_POSTREAD);
1167 		bus_dmamap_unload(sc->sc_dmat, rxs->rxs_dmamap);
1168 
1169 		if ((rxstat & RE_RDESC_STAT_EOF) == 0) {
1170 			m->m_len = MCLBYTES - RE_ETHER_ALIGN;
1171 			if (sc->re_head == NULL)
1172 				sc->re_head = sc->re_tail = m;
1173 			else {
1174 				m->m_flags &= ~M_PKTHDR;
1175 				sc->re_tail->m_next = m;
1176 				sc->re_tail = m;
1177 			}
1178 			re_newbuf(sc, i, NULL);
1179 			continue;
1180 		}
1181 
1182 		/*
1183 		 * NOTE: for the 8139C+, the frame length field
1184 		 * is always 12 bits in size, but for the gigE chips,
1185 		 * it is 13 bits (since the max RX frame length is 16K).
1186 		 * Unfortunately, all 32 bits in the status word
1187 		 * were already used, so to make room for the extra
1188 		 * length bit, RealTek took out the 'frame alignment
1189 		 * error' bit and shifted the other status bits
1190 		 * over one slot. The OWN, EOR, FS and LS bits are
1191 		 * still in the same places. We have already extracted
1192 		 * the frame length and checked the OWN bit, so rather
1193 		 * than using an alternate bit mapping, we shift the
1194 		 * status bits one space to the right so we can evaluate
1195 		 * them using the 8169 status as though it was in the
1196 		 * same format as that of the 8139C+.
1197 		 */
1198 		if ((sc->sc_quirk & RTKQ_8139CPLUS) == 0)
1199 			rxstat >>= 1;
1200 
1201 		if (__predict_false((rxstat & RE_RDESC_STAT_RXERRSUM) != 0)) {
1202 #ifdef RE_DEBUG
1203 			printf("%s: RX error (rxstat = 0x%08x)",
1204 			    device_xname(sc->sc_dev), rxstat);
1205 			if (rxstat & RE_RDESC_STAT_FRALIGN)
1206 				printf(", frame alignment error");
1207 			if (rxstat & RE_RDESC_STAT_BUFOFLOW)
1208 				printf(", out of buffer space");
1209 			if (rxstat & RE_RDESC_STAT_FIFOOFLOW)
1210 				printf(", FIFO overrun");
1211 			if (rxstat & RE_RDESC_STAT_GIANT)
1212 				printf(", giant packet");
1213 			if (rxstat & RE_RDESC_STAT_RUNT)
1214 				printf(", runt packet");
1215 			if (rxstat & RE_RDESC_STAT_CRCERR)
1216 				printf(", CRC error");
1217 			printf("\n");
1218 #endif
1219 			ifp->if_ierrors++;
1220 			/*
1221 			 * If this is part of a multi-fragment packet,
1222 			 * discard all the pieces.
1223 			 */
1224 			if (sc->re_head != NULL) {
1225 				m_freem(sc->re_head);
1226 				sc->re_head = sc->re_tail = NULL;
1227 			}
1228 			re_newbuf(sc, i, m);
1229 			continue;
1230 		}
1231 
1232 		/*
1233 		 * If allocating a replacement mbuf fails,
1234 		 * reload the current one.
1235 		 */
1236 
1237 		if (__predict_false(re_newbuf(sc, i, NULL) != 0)) {
1238 			ifp->if_ierrors++;
1239 			if (sc->re_head != NULL) {
1240 				m_freem(sc->re_head);
1241 				sc->re_head = sc->re_tail = NULL;
1242 			}
1243 			re_newbuf(sc, i, m);
1244 			continue;
1245 		}
1246 
1247 		if (sc->re_head != NULL) {
1248 			m->m_len = total_len % (MCLBYTES - RE_ETHER_ALIGN);
1249 			/*
1250 			 * Special case: if there's 4 bytes or less
1251 			 * in this buffer, the mbuf can be discarded:
1252 			 * the last 4 bytes is the CRC, which we don't
1253 			 * care about anyway.
1254 			 */
1255 			if (m->m_len <= ETHER_CRC_LEN) {
1256 				sc->re_tail->m_len -=
1257 				    (ETHER_CRC_LEN - m->m_len);
1258 				m_freem(m);
1259 			} else {
1260 				m->m_len -= ETHER_CRC_LEN;
1261 				m->m_flags &= ~M_PKTHDR;
1262 				sc->re_tail->m_next = m;
1263 			}
1264 			m = sc->re_head;
1265 			sc->re_head = sc->re_tail = NULL;
1266 			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
1267 		} else
1268 			m->m_pkthdr.len = m->m_len =
1269 			    (total_len - ETHER_CRC_LEN);
1270 
1271 		ifp->if_ipackets++;
1272 		m->m_pkthdr.rcvif = ifp;
1273 
1274 		/* Do RX checksumming */
1275 
1276 		/* Check IP header checksum */
1277 		if ((rxstat & RE_RDESC_STAT_PROTOID) != 0 &&
1278 		    ((sc->sc_quirk & RTKQ_DESCV2) == 0 ||
1279 		     (rxvlan & RE_RDESC_VLANCTL_IPV4) != 0)) {
1280 			m->m_pkthdr.csum_flags |= M_CSUM_IPv4;
1281 			if (rxstat & RE_RDESC_STAT_IPSUMBAD)
1282 				m->m_pkthdr.csum_flags |= M_CSUM_IPv4_BAD;
1283 		}
1284 
1285 		/* Check TCP/UDP checksum */
1286 		if (RE_TCPPKT(rxstat)) {
1287 			m->m_pkthdr.csum_flags |= M_CSUM_TCPv4;
1288 			if (rxstat & RE_RDESC_STAT_TCPSUMBAD)
1289 				m->m_pkthdr.csum_flags |= 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 |= M_CSUM_TCP_UDP_BAD;
1294 		}
1295 
1296 		if (rxvlan & RE_RDESC_VLANCTL_TAG) {
1297 			VLAN_INPUT_TAG(ifp, m,
1298 			     bswap16(rxvlan & RE_RDESC_VLANCTL_DATA),
1299 			     continue);
1300 		}
1301 #if NBPFILTER > 0
1302 		if (ifp->if_bpf)
1303 			bpf_mtap(ifp->if_bpf, m);
1304 #endif
1305 		(*ifp->if_input)(ifp, m);
1306 	}
1307 
1308 	sc->re_ldata.re_rx_prodidx = i;
1309 }
1310 
1311 static void
1312 re_txeof(struct rtk_softc *sc)
1313 {
1314 	struct ifnet *ifp;
1315 	struct re_txq *txq;
1316 	uint32_t txstat;
1317 	int idx, descidx;
1318 
1319 	ifp = &sc->ethercom.ec_if;
1320 
1321 	for (idx = sc->re_ldata.re_txq_considx;
1322 	    sc->re_ldata.re_txq_free < RE_TX_QLEN;
1323 	    idx = RE_NEXT_TXQ(sc, idx), sc->re_ldata.re_txq_free++) {
1324 		txq = &sc->re_ldata.re_txq[idx];
1325 		KASSERT(txq->txq_mbuf != NULL);
1326 
1327 		descidx = txq->txq_descidx;
1328 		RE_TXDESCSYNC(sc, descidx,
1329 		    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1330 		txstat =
1331 		    le32toh(sc->re_ldata.re_tx_list[descidx].re_cmdstat);
1332 		RE_TXDESCSYNC(sc, descidx, BUS_DMASYNC_PREREAD);
1333 		KASSERT((txstat & RE_TDESC_CMD_EOF) != 0);
1334 		if (txstat & RE_TDESC_CMD_OWN) {
1335 			break;
1336 		}
1337 
1338 		sc->re_ldata.re_tx_free += txq->txq_nsegs;
1339 		KASSERT(sc->re_ldata.re_tx_free <= RE_TX_DESC_CNT(sc));
1340 		bus_dmamap_sync(sc->sc_dmat, txq->txq_dmamap,
1341 		    0, txq->txq_dmamap->dm_mapsize, BUS_DMASYNC_POSTWRITE);
1342 		bus_dmamap_unload(sc->sc_dmat, txq->txq_dmamap);
1343 		m_freem(txq->txq_mbuf);
1344 		txq->txq_mbuf = NULL;
1345 
1346 		if (txstat & (RE_TDESC_STAT_EXCESSCOL | RE_TDESC_STAT_COLCNT))
1347 			ifp->if_collisions++;
1348 		if (txstat & RE_TDESC_STAT_TXERRSUM)
1349 			ifp->if_oerrors++;
1350 		else
1351 			ifp->if_opackets++;
1352 	}
1353 
1354 	sc->re_ldata.re_txq_considx = idx;
1355 
1356 	if (sc->re_ldata.re_txq_free > RE_NTXDESC_RSVD)
1357 		ifp->if_flags &= ~IFF_OACTIVE;
1358 
1359 	/*
1360 	 * If not all descriptors have been released reaped yet,
1361 	 * reload the timer so that we will eventually get another
1362 	 * interrupt that will cause us to re-enter this routine.
1363 	 * This is done in case the transmitter has gone idle.
1364 	 */
1365 	if (sc->re_ldata.re_txq_free < RE_TX_QLEN) {
1366 		CSR_WRITE_4(sc, RTK_TIMERCNT, 1);
1367 		if ((sc->sc_quirk & RTKQ_PCIE) != 0) {
1368 			/*
1369 			 * Some chips will ignore a second TX request
1370 			 * issued while an existing transmission is in
1371 			 * progress. If the transmitter goes idle but
1372 			 * there are still packets waiting to be sent,
1373 			 * we need to restart the channel here to flush
1374 			 * them out. This only seems to be required with
1375 			 * the PCIe devices.
1376 			 */
1377 			CSR_WRITE_1(sc, RTK_GTXSTART, RTK_TXSTART_START);
1378 		}
1379 	} else
1380 		ifp->if_timer = 0;
1381 }
1382 
1383 static void
1384 re_tick(void *arg)
1385 {
1386 	struct rtk_softc *sc = arg;
1387 	int s;
1388 
1389 	/*XXX: just return for 8169S/8110S with rev 2 or newer phy */
1390 	s = splnet();
1391 
1392 	mii_tick(&sc->mii);
1393 	splx(s);
1394 
1395 	callout_reset(&sc->rtk_tick_ch, hz, re_tick, sc);
1396 }
1397 
1398 int
1399 re_intr(void *arg)
1400 {
1401 	struct rtk_softc *sc = arg;
1402 	struct ifnet *ifp;
1403 	uint16_t status;
1404 	int handled = 0;
1405 
1406 	if (!device_has_power(sc->sc_dev))
1407 		return 0;
1408 
1409 	ifp = &sc->ethercom.ec_if;
1410 
1411 	if ((ifp->if_flags & IFF_UP) == 0)
1412 		return 0;
1413 
1414 	for (;;) {
1415 
1416 		status = CSR_READ_2(sc, RTK_ISR);
1417 		/* If the card has gone away the read returns 0xffff. */
1418 		if (status == 0xffff)
1419 			break;
1420 		if (status) {
1421 			handled = 1;
1422 			CSR_WRITE_2(sc, RTK_ISR, status);
1423 		}
1424 
1425 		if ((status & RTK_INTRS_CPLUS) == 0)
1426 			break;
1427 
1428 		if (status & (RTK_ISR_RX_OK | RTK_ISR_RX_ERR))
1429 			re_rxeof(sc);
1430 
1431 		if (status & (RTK_ISR_TIMEOUT_EXPIRED | RTK_ISR_TX_ERR |
1432 		    RTK_ISR_TX_DESC_UNAVAIL))
1433 			re_txeof(sc);
1434 
1435 		if (status & RTK_ISR_SYSTEM_ERR) {
1436 			re_init(ifp);
1437 		}
1438 
1439 		if (status & RTK_ISR_LINKCHG) {
1440 			callout_stop(&sc->rtk_tick_ch);
1441 			re_tick(sc);
1442 		}
1443 	}
1444 
1445 	if (handled && !IFQ_IS_EMPTY(&ifp->if_snd))
1446 		re_start(ifp);
1447 
1448 	return handled;
1449 }
1450 
1451 
1452 
1453 /*
1454  * Main transmit routine for C+ and gigE NICs.
1455  */
1456 
1457 static void
1458 re_start(struct ifnet *ifp)
1459 {
1460 	struct rtk_softc *sc;
1461 	struct mbuf *m;
1462 	bus_dmamap_t map;
1463 	struct re_txq *txq;
1464 	struct re_desc *d;
1465 	struct m_tag *mtag;
1466 	uint32_t cmdstat, re_flags, vlanctl;
1467 	int ofree, idx, error, nsegs, seg;
1468 	int startdesc, curdesc, lastdesc;
1469 	bool pad;
1470 
1471 	sc = ifp->if_softc;
1472 	ofree = sc->re_ldata.re_txq_free;
1473 
1474 	for (idx = sc->re_ldata.re_txq_prodidx;; idx = RE_NEXT_TXQ(sc, idx)) {
1475 
1476 		IFQ_POLL(&ifp->if_snd, m);
1477 		if (m == NULL)
1478 			break;
1479 
1480 		if (sc->re_ldata.re_txq_free == 0 ||
1481 		    sc->re_ldata.re_tx_free == 0) {
1482 			/* no more free slots left */
1483 			ifp->if_flags |= IFF_OACTIVE;
1484 			break;
1485 		}
1486 
1487 		/*
1488 		 * Set up checksum offload. Note: checksum offload bits must
1489 		 * appear in all descriptors of a multi-descriptor transmit
1490 		 * attempt. (This is according to testing done with an 8169
1491 		 * chip. I'm not sure if this is a requirement or a bug.)
1492 		 */
1493 
1494 		vlanctl = 0;
1495 		if ((m->m_pkthdr.csum_flags & M_CSUM_TSOv4) != 0) {
1496 			uint32_t segsz = m->m_pkthdr.segsz;
1497 
1498 			re_flags = RE_TDESC_CMD_LGSEND |
1499 			    (segsz << RE_TDESC_CMD_MSSVAL_SHIFT);
1500 		} else {
1501 			/*
1502 			 * set RE_TDESC_CMD_IPCSUM if any checksum offloading
1503 			 * is requested.  otherwise, RE_TDESC_CMD_TCPCSUM/
1504 			 * RE_TDESC_CMD_UDPCSUM doesn't make effects.
1505 			 */
1506 			re_flags = 0;
1507 			if ((m->m_pkthdr.csum_flags &
1508 			    (M_CSUM_IPv4 | M_CSUM_TCPv4 | M_CSUM_UDPv4))
1509 			    != 0) {
1510 				if ((sc->sc_quirk & RTKQ_DESCV2) == 0) {
1511 					re_flags |= RE_TDESC_CMD_IPCSUM;
1512 					if (m->m_pkthdr.csum_flags &
1513 					    M_CSUM_TCPv4) {
1514 						re_flags |=
1515 						    RE_TDESC_CMD_TCPCSUM;
1516 					} else if (m->m_pkthdr.csum_flags &
1517 					    M_CSUM_UDPv4) {
1518 						re_flags |=
1519 						    RE_TDESC_CMD_UDPCSUM;
1520 					}
1521 				} else {
1522 					vlanctl |= RE_TDESC_VLANCTL_IPCSUM;
1523 					if (m->m_pkthdr.csum_flags &
1524 					    M_CSUM_TCPv4) {
1525 						vlanctl |=
1526 						    RE_TDESC_VLANCTL_TCPCSUM;
1527 					} else if (m->m_pkthdr.csum_flags &
1528 					    M_CSUM_UDPv4) {
1529 						vlanctl |=
1530 						    RE_TDESC_VLANCTL_UDPCSUM;
1531 					}
1532 				}
1533 			}
1534 		}
1535 
1536 		txq = &sc->re_ldata.re_txq[idx];
1537 		map = txq->txq_dmamap;
1538 		error = bus_dmamap_load_mbuf(sc->sc_dmat, map, m,
1539 		    BUS_DMA_WRITE|BUS_DMA_NOWAIT);
1540 
1541 		if (__predict_false(error)) {
1542 			/* XXX try to defrag if EFBIG? */
1543 			printf("%s: can't map mbuf (error %d)\n",
1544 			    device_xname(sc->sc_dev), error);
1545 
1546 			IFQ_DEQUEUE(&ifp->if_snd, m);
1547 			m_freem(m);
1548 			ifp->if_oerrors++;
1549 			continue;
1550 		}
1551 
1552 		nsegs = map->dm_nsegs;
1553 		pad = false;
1554 		if (__predict_false(m->m_pkthdr.len <= RE_IP4CSUMTX_PADLEN &&
1555 		    (re_flags & RE_TDESC_CMD_IPCSUM) != 0 &&
1556 		    (sc->sc_quirk & RTKQ_DESCV2) == 0)) {
1557 			pad = true;
1558 			nsegs++;
1559 		}
1560 
1561 		if (nsegs > sc->re_ldata.re_tx_free) {
1562 			/*
1563 			 * Not enough free descriptors to transmit this packet.
1564 			 */
1565 			ifp->if_flags |= IFF_OACTIVE;
1566 			bus_dmamap_unload(sc->sc_dmat, map);
1567 			break;
1568 		}
1569 
1570 		IFQ_DEQUEUE(&ifp->if_snd, m);
1571 
1572 		/*
1573 		 * Make sure that the caches are synchronized before we
1574 		 * ask the chip to start DMA for the packet data.
1575 		 */
1576 		bus_dmamap_sync(sc->sc_dmat, map, 0, map->dm_mapsize,
1577 		    BUS_DMASYNC_PREWRITE);
1578 
1579 		/*
1580 		 * Set up hardware VLAN tagging. Note: vlan tag info must
1581 		 * appear in all descriptors of a multi-descriptor
1582 		 * transmission attempt.
1583 		 */
1584 		if ((mtag = VLAN_OUTPUT_TAG(&sc->ethercom, m)) != NULL)
1585 			vlanctl |= bswap16(VLAN_TAG_VALUE(mtag)) |
1586 			    RE_TDESC_VLANCTL_TAG;
1587 
1588 		/*
1589 		 * Map the segment array into descriptors.
1590 		 * Note that we set the start-of-frame and
1591 		 * end-of-frame markers for either TX or RX,
1592 		 * but they really only have meaning in the TX case.
1593 		 * (In the RX case, it's the chip that tells us
1594 		 *  where packets begin and end.)
1595 		 * We also keep track of the end of the ring
1596 		 * and set the end-of-ring bits as needed,
1597 		 * and we set the ownership bits in all except
1598 		 * the very first descriptor. (The caller will
1599 		 * set this descriptor later when it start
1600 		 * transmission or reception.)
1601 		 */
1602 		curdesc = startdesc = sc->re_ldata.re_tx_nextfree;
1603 		lastdesc = -1;
1604 		for (seg = 0; seg < map->dm_nsegs;
1605 		    seg++, curdesc = RE_NEXT_TX_DESC(sc, curdesc)) {
1606 			d = &sc->re_ldata.re_tx_list[curdesc];
1607 #ifdef DIAGNOSTIC
1608 			RE_TXDESCSYNC(sc, curdesc,
1609 			    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1610 			cmdstat = le32toh(d->re_cmdstat);
1611 			RE_TXDESCSYNC(sc, curdesc, BUS_DMASYNC_PREREAD);
1612 			if (cmdstat & RE_TDESC_STAT_OWN) {
1613 				panic("%s: tried to map busy TX descriptor",
1614 				    device_xname(sc->sc_dev));
1615 			}
1616 #endif
1617 
1618 			d->re_vlanctl = htole32(vlanctl);
1619 			re_set_bufaddr(d, map->dm_segs[seg].ds_addr);
1620 			cmdstat = re_flags | map->dm_segs[seg].ds_len;
1621 			if (seg == 0)
1622 				cmdstat |= RE_TDESC_CMD_SOF;
1623 			else
1624 				cmdstat |= RE_TDESC_CMD_OWN;
1625 			if (curdesc == (RE_TX_DESC_CNT(sc) - 1))
1626 				cmdstat |= RE_TDESC_CMD_EOR;
1627 			if (seg == nsegs - 1) {
1628 				cmdstat |= RE_TDESC_CMD_EOF;
1629 				lastdesc = curdesc;
1630 			}
1631 			d->re_cmdstat = htole32(cmdstat);
1632 			RE_TXDESCSYNC(sc, curdesc,
1633 			    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1634 		}
1635 		if (__predict_false(pad)) {
1636 			bus_addr_t paddaddr;
1637 
1638 			d = &sc->re_ldata.re_tx_list[curdesc];
1639 			d->re_vlanctl = htole32(vlanctl);
1640 			paddaddr = RE_TXPADDADDR(sc);
1641 			re_set_bufaddr(d, paddaddr);
1642 			cmdstat = re_flags |
1643 			    RE_TDESC_CMD_OWN | RE_TDESC_CMD_EOF |
1644 			    (RE_IP4CSUMTX_PADLEN + 1 - m->m_pkthdr.len);
1645 			if (curdesc == (RE_TX_DESC_CNT(sc) - 1))
1646 				cmdstat |= RE_TDESC_CMD_EOR;
1647 			d->re_cmdstat = htole32(cmdstat);
1648 			RE_TXDESCSYNC(sc, curdesc,
1649 			    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1650 			lastdesc = curdesc;
1651 			curdesc = RE_NEXT_TX_DESC(sc, curdesc);
1652 		}
1653 		KASSERT(lastdesc != -1);
1654 
1655 		/* Transfer ownership of packet to the chip. */
1656 
1657 		sc->re_ldata.re_tx_list[startdesc].re_cmdstat |=
1658 		    htole32(RE_TDESC_CMD_OWN);
1659 		RE_TXDESCSYNC(sc, startdesc,
1660 		    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1661 
1662 		/* update info of TX queue and descriptors */
1663 		txq->txq_mbuf = m;
1664 		txq->txq_descidx = lastdesc;
1665 		txq->txq_nsegs = nsegs;
1666 
1667 		sc->re_ldata.re_txq_free--;
1668 		sc->re_ldata.re_tx_free -= nsegs;
1669 		sc->re_ldata.re_tx_nextfree = curdesc;
1670 
1671 #if NBPFILTER > 0
1672 		/*
1673 		 * If there's a BPF listener, bounce a copy of this frame
1674 		 * to him.
1675 		 */
1676 		if (ifp->if_bpf)
1677 			bpf_mtap(ifp->if_bpf, m);
1678 #endif
1679 	}
1680 
1681 	if (sc->re_ldata.re_txq_free < ofree) {
1682 		/*
1683 		 * TX packets are enqueued.
1684 		 */
1685 		sc->re_ldata.re_txq_prodidx = idx;
1686 
1687 		/*
1688 		 * Start the transmitter to poll.
1689 		 *
1690 		 * RealTek put the TX poll request register in a different
1691 		 * location on the 8169 gigE chip. I don't know why.
1692 		 */
1693 		if ((sc->sc_quirk & RTKQ_8139CPLUS) != 0)
1694 			CSR_WRITE_1(sc, RTK_TXSTART, RTK_TXSTART_START);
1695 		else
1696 			CSR_WRITE_1(sc, RTK_GTXSTART, RTK_TXSTART_START);
1697 
1698 		/*
1699 		 * Use the countdown timer for interrupt moderation.
1700 		 * 'TX done' interrupts are disabled. Instead, we reset the
1701 		 * countdown timer, which will begin counting until it hits
1702 		 * the value in the TIMERINT register, and then trigger an
1703 		 * interrupt. Each time we write to the TIMERCNT register,
1704 		 * the timer count is reset to 0.
1705 		 */
1706 		CSR_WRITE_4(sc, RTK_TIMERCNT, 1);
1707 
1708 		/*
1709 		 * Set a timeout in case the chip goes out to lunch.
1710 		 */
1711 		ifp->if_timer = 5;
1712 	}
1713 }
1714 
1715 static int
1716 re_init(struct ifnet *ifp)
1717 {
1718 	struct rtk_softc *sc = ifp->if_softc;
1719 	const uint8_t *enaddr;
1720 	uint32_t rxcfg = 0;
1721 	uint32_t reg;
1722 	int error;
1723 
1724 	if ((error = re_enable(sc)) != 0)
1725 		goto out;
1726 
1727 	/*
1728 	 * Cancel pending I/O and free all RX/TX buffers.
1729 	 */
1730 	re_stop(ifp, 0);
1731 
1732 	re_reset(sc);
1733 
1734 	/*
1735 	 * Enable C+ RX and TX mode, as well as VLAN stripping and
1736 	 * RX checksum offload. We must configure the C+ register
1737 	 * before all others.
1738 	 */
1739 	reg = 0;
1740 
1741 	/*
1742 	 * XXX: Realtek docs say bits 0 and 1 are reserved, for 8169S/8110S.
1743 	 * FreeBSD  drivers set these bits anyway (for 8139C+?).
1744 	 * So far, it works.
1745 	 */
1746 
1747 	/*
1748 	 * XXX: For old 8169 set bit 14.
1749 	 *      For 8169S/8110S and above, do not set bit 14.
1750 	 */
1751 	if ((sc->sc_quirk & RTKQ_8169NONS) != 0)
1752 		reg |= (0x1 << 14) | RTK_CPLUSCMD_PCI_MRW;
1753 
1754 	if (1)  {/* not for 8169S ? */
1755 		reg |=
1756 		    RTK_CPLUSCMD_VLANSTRIP |
1757 		    (ifp->if_capenable &
1758 		    (IFCAP_CSUM_IPv4_Rx | IFCAP_CSUM_TCPv4_Rx |
1759 		     IFCAP_CSUM_UDPv4_Rx) ?
1760 		    RTK_CPLUSCMD_RXCSUM_ENB : 0);
1761 	}
1762 
1763 	CSR_WRITE_2(sc, RTK_CPLUS_CMD,
1764 	    reg | RTK_CPLUSCMD_RXENB | RTK_CPLUSCMD_TXENB);
1765 
1766 	/* XXX: from Realtek-supplied Linux driver. Wholly undocumented. */
1767 	if ((sc->sc_quirk & RTKQ_8139CPLUS) == 0)
1768 		CSR_WRITE_2(sc, RTK_IM, 0x0000);
1769 
1770 	DELAY(10000);
1771 
1772 	/*
1773 	 * Init our MAC address.  Even though the chipset
1774 	 * documentation doesn't mention it, we need to enter "Config
1775 	 * register write enable" mode to modify the ID registers.
1776 	 */
1777 	CSR_WRITE_1(sc, RTK_EECMD, RTK_EEMODE_WRITECFG);
1778 	enaddr = CLLADDR(ifp->if_sadl);
1779 	reg = enaddr[0] | (enaddr[1] << 8) |
1780 	    (enaddr[2] << 16) | (enaddr[3] << 24);
1781 	CSR_WRITE_4(sc, RTK_IDR0, reg);
1782 	reg = enaddr[4] | (enaddr[5] << 8);
1783 	CSR_WRITE_4(sc, RTK_IDR4, reg);
1784 	CSR_WRITE_1(sc, RTK_EECMD, RTK_EEMODE_OFF);
1785 
1786 	/*
1787 	 * For C+ mode, initialize the RX descriptors and mbufs.
1788 	 */
1789 	re_rx_list_init(sc);
1790 	re_tx_list_init(sc);
1791 
1792 	/*
1793 	 * Load the addresses of the RX and TX lists into the chip.
1794 	 */
1795 	CSR_WRITE_4(sc, RTK_RXLIST_ADDR_HI,
1796 	    RE_ADDR_HI(sc->re_ldata.re_rx_list_map->dm_segs[0].ds_addr));
1797 	CSR_WRITE_4(sc, RTK_RXLIST_ADDR_LO,
1798 	    RE_ADDR_LO(sc->re_ldata.re_rx_list_map->dm_segs[0].ds_addr));
1799 
1800 	CSR_WRITE_4(sc, RTK_TXLIST_ADDR_HI,
1801 	    RE_ADDR_HI(sc->re_ldata.re_tx_list_map->dm_segs[0].ds_addr));
1802 	CSR_WRITE_4(sc, RTK_TXLIST_ADDR_LO,
1803 	    RE_ADDR_LO(sc->re_ldata.re_tx_list_map->dm_segs[0].ds_addr));
1804 
1805 	/*
1806 	 * Enable transmit and receive.
1807 	 */
1808 	CSR_WRITE_1(sc, RTK_COMMAND, RTK_CMD_TX_ENB | RTK_CMD_RX_ENB);
1809 
1810 	/*
1811 	 * Set the initial TX and RX configuration.
1812 	 */
1813 	if (sc->re_testmode && (sc->sc_quirk & RTKQ_8169NONS) != 0) {
1814 		/* test mode is needed only for old 8169 */
1815 		CSR_WRITE_4(sc, RTK_TXCFG,
1816 		    RE_TXCFG_CONFIG | RTK_LOOPTEST_ON);
1817 	} else
1818 		CSR_WRITE_4(sc, RTK_TXCFG, RE_TXCFG_CONFIG);
1819 
1820 	CSR_WRITE_1(sc, RTK_EARLY_TX_THRESH, 16);
1821 
1822 	CSR_WRITE_4(sc, RTK_RXCFG, RE_RXCFG_CONFIG);
1823 
1824 	/* Set the individual bit to receive frames for this host only. */
1825 	rxcfg = CSR_READ_4(sc, RTK_RXCFG);
1826 	rxcfg |= RTK_RXCFG_RX_INDIV;
1827 
1828 	/* If we want promiscuous mode, set the allframes bit. */
1829 	if (ifp->if_flags & IFF_PROMISC)
1830 		rxcfg |= RTK_RXCFG_RX_ALLPHYS;
1831 	else
1832 		rxcfg &= ~RTK_RXCFG_RX_ALLPHYS;
1833 	CSR_WRITE_4(sc, RTK_RXCFG, rxcfg);
1834 
1835 	/*
1836 	 * Set capture broadcast bit to capture broadcast frames.
1837 	 */
1838 	if (ifp->if_flags & IFF_BROADCAST)
1839 		rxcfg |= RTK_RXCFG_RX_BROAD;
1840 	else
1841 		rxcfg &= ~RTK_RXCFG_RX_BROAD;
1842 	CSR_WRITE_4(sc, RTK_RXCFG, rxcfg);
1843 
1844 	/*
1845 	 * Program the multicast filter, if necessary.
1846 	 */
1847 	rtk_setmulti(sc);
1848 
1849 	/*
1850 	 * Enable interrupts.
1851 	 */
1852 	if (sc->re_testmode)
1853 		CSR_WRITE_2(sc, RTK_IMR, 0);
1854 	else
1855 		CSR_WRITE_2(sc, RTK_IMR, RTK_INTRS_CPLUS);
1856 
1857 	/* Start RX/TX process. */
1858 	CSR_WRITE_4(sc, RTK_MISSEDPKT, 0);
1859 #ifdef notdef
1860 	/* Enable receiver and transmitter. */
1861 	CSR_WRITE_1(sc, RTK_COMMAND, RTK_CMD_TX_ENB | RTK_CMD_RX_ENB);
1862 #endif
1863 
1864 	/*
1865 	 * Initialize the timer interrupt register so that
1866 	 * a timer interrupt will be generated once the timer
1867 	 * reaches a certain number of ticks. The timer is
1868 	 * reloaded on each transmit. This gives us TX interrupt
1869 	 * moderation, which dramatically improves TX frame rate.
1870 	 */
1871 
1872 	if ((sc->sc_quirk & RTKQ_8139CPLUS) != 0)
1873 		CSR_WRITE_4(sc, RTK_TIMERINT, 0x400);
1874 	else {
1875 		CSR_WRITE_4(sc, RTK_TIMERINT_8169, 0x800);
1876 
1877 		/*
1878 		 * For 8169 gigE NICs, set the max allowed RX packet
1879 		 * size so we can receive jumbo frames.
1880 		 */
1881 		CSR_WRITE_2(sc, RTK_MAXRXPKTLEN, 16383);
1882 	}
1883 
1884 	if (sc->re_testmode)
1885 		return 0;
1886 
1887 	CSR_WRITE_1(sc, RTK_CFG1, RTK_CFG1_DRVLOAD);
1888 
1889 	ifp->if_flags |= IFF_RUNNING;
1890 	ifp->if_flags &= ~IFF_OACTIVE;
1891 
1892 	callout_reset(&sc->rtk_tick_ch, hz, re_tick, sc);
1893 
1894  out:
1895 	if (error) {
1896 		ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
1897 		ifp->if_timer = 0;
1898 		printf("%s: interface not running\n",
1899 		    device_xname(sc->sc_dev));
1900 	}
1901 
1902 	return error;
1903 }
1904 
1905 static int
1906 re_ioctl(struct ifnet *ifp, u_long command, void *data)
1907 {
1908 	struct rtk_softc *sc = ifp->if_softc;
1909 	struct ifreq *ifr = data;
1910 	int s, error = 0;
1911 
1912 	s = splnet();
1913 
1914 	switch (command) {
1915 	case SIOCSIFMTU:
1916 		/*
1917 		 * Disable jumbo frames if it's not supported.
1918 		 */
1919 		if ((sc->sc_quirk & RTKQ_NOJUMBO) != 0 &&
1920 		    ifr->ifr_mtu > ETHERMTU) {
1921 			error = EINVAL;
1922 			break;
1923 		}
1924 
1925 		if (ifr->ifr_mtu < ETHERMIN || ifr->ifr_mtu > ETHERMTU_JUMBO)
1926 			error = EINVAL;
1927 		else if ((error = ifioctl_common(ifp, command, data)) ==
1928 		    ENETRESET)
1929 			error = 0;
1930 		break;
1931 	default:
1932 		if ((error = ether_ioctl(ifp, command, data)) != ENETRESET)
1933 			break;
1934 
1935 		error = 0;
1936 
1937 		if (command == SIOCSIFCAP)
1938 			error = (*ifp->if_init)(ifp);
1939 		else if (command != SIOCADDMULTI && command != SIOCDELMULTI)
1940 			;
1941 		else if (ifp->if_flags & IFF_RUNNING)
1942 			rtk_setmulti(sc);
1943 		break;
1944 	}
1945 
1946 	splx(s);
1947 
1948 	return error;
1949 }
1950 
1951 static void
1952 re_watchdog(struct ifnet *ifp)
1953 {
1954 	struct rtk_softc *sc;
1955 	int s;
1956 
1957 	sc = ifp->if_softc;
1958 	s = splnet();
1959 	printf("%s: watchdog timeout\n", device_xname(sc->sc_dev));
1960 	ifp->if_oerrors++;
1961 
1962 	re_txeof(sc);
1963 	re_rxeof(sc);
1964 
1965 	re_init(ifp);
1966 
1967 	splx(s);
1968 }
1969 
1970 /*
1971  * Stop the adapter and free any mbufs allocated to the
1972  * RX and TX lists.
1973  */
1974 static void
1975 re_stop(struct ifnet *ifp, int disable)
1976 {
1977 	int i;
1978 	struct rtk_softc *sc = ifp->if_softc;
1979 
1980 	callout_stop(&sc->rtk_tick_ch);
1981 
1982 	mii_down(&sc->mii);
1983 
1984 	CSR_WRITE_1(sc, RTK_COMMAND, 0x00);
1985 	CSR_WRITE_2(sc, RTK_IMR, 0x0000);
1986 
1987 	if (sc->re_head != NULL) {
1988 		m_freem(sc->re_head);
1989 		sc->re_head = sc->re_tail = NULL;
1990 	}
1991 
1992 	/* Free the TX list buffers. */
1993 	for (i = 0; i < RE_TX_QLEN; i++) {
1994 		if (sc->re_ldata.re_txq[i].txq_mbuf != NULL) {
1995 			bus_dmamap_unload(sc->sc_dmat,
1996 			    sc->re_ldata.re_txq[i].txq_dmamap);
1997 			m_freem(sc->re_ldata.re_txq[i].txq_mbuf);
1998 			sc->re_ldata.re_txq[i].txq_mbuf = NULL;
1999 		}
2000 	}
2001 
2002 	/* Free the RX list buffers. */
2003 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
2004 		if (sc->re_ldata.re_rxsoft[i].rxs_mbuf != NULL) {
2005 			bus_dmamap_unload(sc->sc_dmat,
2006 			    sc->re_ldata.re_rxsoft[i].rxs_dmamap);
2007 			m_freem(sc->re_ldata.re_rxsoft[i].rxs_mbuf);
2008 			sc->re_ldata.re_rxsoft[i].rxs_mbuf = NULL;
2009 		}
2010 	}
2011 
2012 	if (disable)
2013 		re_disable(sc);
2014 
2015 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
2016 	ifp->if_timer = 0;
2017 }
2018