xref: /netbsd-src/sys/dev/ic/rtl8169.c (revision 10ad5ffa714ce1a679dcc9dd8159648df2d67b5a)
1 /*	$NetBSD: rtl8169.c,v 1.121 2009/05/31 05:10:47 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 1.121 2009/05/31 05:10:47 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(device_t, int, int);
170 static void re_gmii_writereg(device_t, int, int, int);
171 
172 static int re_miibus_readreg(device_t, int, int);
173 static void re_miibus_writereg(device_t, int, int, int);
174 static void re_miibus_statchg(device_t);
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 		switch (hwrev) {
571 		case RTK_HWREV_8169:
572 			sc->sc_quirk |= RTKQ_8169NONS;
573 			break;
574 		case RTK_HWREV_8169S:
575 		case RTK_HWREV_8110S:
576 		case RTK_HWREV_8169_8110SB:
577 		case RTK_HWREV_8169_8110SC:
578 			sc->sc_quirk |= RTKQ_MACLDPS;
579 			break;
580 		case RTK_HWREV_8168_SPIN1:
581 		case RTK_HWREV_8168_SPIN2:
582 		case RTK_HWREV_8168_SPIN3:
583 			sc->sc_quirk |= RTKQ_MACSTAT;
584 			break;
585 		case RTK_HWREV_8168C:
586 		case RTK_HWREV_8168C_SPIN2:
587 		case RTK_HWREV_8168CP:
588 		case RTK_HWREV_8168D:
589 			sc->sc_quirk |= RTKQ_DESCV2 | RTKQ_NOEECMD |
590 			    RTKQ_MACSTAT | RTKQ_CMDSTOP;
591 			/*
592 			 * From FreeBSD driver:
593 			 *
594 			 * These (8168/8111) controllers support jumbo frame
595 			 * but it seems that enabling it requires touching
596 			 * additional magic registers. Depending on MAC
597 			 * revisions some controllers need to disable
598 			 * checksum offload. So disable jumbo frame until
599 			 * I have better idea what it really requires to
600 			 * make it support.
601 			 * RTL8168C/CP : supports up to 6KB jumbo frame.
602 			 * RTL8111C/CP : supports up to 9KB jumbo frame.
603 			 */
604 			sc->sc_quirk |= RTKQ_NOJUMBO;
605 			break;
606 		case RTK_HWREV_8100E:
607 		case RTK_HWREV_8100E_SPIN2:
608 		case RTK_HWREV_8101E:
609 			sc->sc_quirk |= RTKQ_NOJUMBO;
610 			break;
611 		case RTK_HWREV_8102E:
612 		case RTK_HWREV_8102EL:
613 		case RTK_HWREV_8103E:
614 			sc->sc_quirk |= RTKQ_DESCV2 | RTKQ_NOEECMD |
615 			    RTKQ_MACSTAT | RTKQ_CMDSTOP | RTKQ_NOJUMBO;
616 			break;
617 		default:
618 			aprint_normal_dev(sc->sc_dev,
619 			    "Unknown revision (0x%08x)\n", hwrev);
620 			/* assume the latest features */
621 			sc->sc_quirk |= RTKQ_DESCV2 | RTKQ_NOEECMD;
622 			sc->sc_quirk |= RTKQ_NOJUMBO;
623 		}
624 
625 		/* Set RX length mask */
626 		sc->re_rxlenmask = RE_RDESC_STAT_GFRAGLEN;
627 		sc->re_ldata.re_tx_desc_cnt = RE_TX_DESC_CNT_8169;
628 	} else {
629 		sc->sc_quirk |= RTKQ_NOJUMBO;
630 
631 		/* Set RX length mask */
632 		sc->re_rxlenmask = RE_RDESC_STAT_FRAGLEN;
633 		sc->re_ldata.re_tx_desc_cnt = RE_TX_DESC_CNT_8139;
634 	}
635 
636 	/* Reset the adapter. */
637 	re_reset(sc);
638 
639 	if ((sc->sc_quirk & RTKQ_NOEECMD) != 0) {
640 		/*
641 		 * Get station address from ID registers.
642 		 */
643 		for (i = 0; i < ETHER_ADDR_LEN; i++)
644 			eaddr[i] = CSR_READ_1(sc, RTK_IDR0 + i);
645 	} else {
646 		/*
647 		 * Get station address from the EEPROM.
648 		 */
649 		if (rtk_read_eeprom(sc, RTK_EE_ID, RTK_EEADDR_LEN1) == 0x8129)
650 			addr_len = RTK_EEADDR_LEN1;
651 		else
652 			addr_len = RTK_EEADDR_LEN0;
653 
654 		/*
655 		 * Get station address from the EEPROM.
656 		 */
657 		for (i = 0; i < ETHER_ADDR_LEN / 2; i++) {
658 			val = rtk_read_eeprom(sc, RTK_EE_EADDR0 + i, addr_len);
659 			eaddr[(i * 2) + 0] = val & 0xff;
660 			eaddr[(i * 2) + 1] = val >> 8;
661 		}
662 	}
663 
664 	aprint_normal_dev(sc->sc_dev, "Ethernet address %s\n",
665 	    ether_sprintf(eaddr));
666 
667 	if (sc->re_ldata.re_tx_desc_cnt >
668 	    PAGE_SIZE / sizeof(struct re_desc)) {
669 		sc->re_ldata.re_tx_desc_cnt =
670 		    PAGE_SIZE / sizeof(struct re_desc);
671 	}
672 
673 	aprint_verbose_dev(sc->sc_dev, "using %d tx descriptors\n",
674 	    sc->re_ldata.re_tx_desc_cnt);
675 	KASSERT(RE_NEXT_TX_DESC(sc, RE_TX_DESC_CNT(sc) - 1) == 0);
676 
677 	/* Allocate DMA'able memory for the TX ring */
678 	if ((error = bus_dmamem_alloc(sc->sc_dmat, RE_TX_LIST_SZ(sc),
679 	    RE_RING_ALIGN, 0, &sc->re_ldata.re_tx_listseg, 1,
680 	    &sc->re_ldata.re_tx_listnseg, BUS_DMA_NOWAIT)) != 0) {
681 		aprint_error_dev(sc->sc_dev,
682 		    "can't allocate tx listseg, error = %d\n", error);
683 		goto fail_0;
684 	}
685 
686 	/* Load the map for the TX ring. */
687 	if ((error = bus_dmamem_map(sc->sc_dmat, &sc->re_ldata.re_tx_listseg,
688 	    sc->re_ldata.re_tx_listnseg, RE_TX_LIST_SZ(sc),
689 	    (void **)&sc->re_ldata.re_tx_list,
690 	    BUS_DMA_COHERENT | BUS_DMA_NOWAIT)) != 0) {
691 		aprint_error_dev(sc->sc_dev,
692 		    "can't map tx list, error = %d\n", error);
693 	  	goto fail_1;
694 	}
695 	memset(sc->re_ldata.re_tx_list, 0, RE_TX_LIST_SZ(sc));
696 
697 	if ((error = bus_dmamap_create(sc->sc_dmat, RE_TX_LIST_SZ(sc), 1,
698 	    RE_TX_LIST_SZ(sc), 0, 0,
699 	    &sc->re_ldata.re_tx_list_map)) != 0) {
700 		aprint_error_dev(sc->sc_dev,
701 		    "can't create tx list map, error = %d\n", error);
702 		goto fail_2;
703 	}
704 
705 
706 	if ((error = bus_dmamap_load(sc->sc_dmat,
707 	    sc->re_ldata.re_tx_list_map, sc->re_ldata.re_tx_list,
708 	    RE_TX_LIST_SZ(sc), NULL, BUS_DMA_NOWAIT)) != 0) {
709 		aprint_error_dev(sc->sc_dev,
710 		    "can't load tx list, error = %d\n", error);
711 		goto fail_3;
712 	}
713 
714 	/* Create DMA maps for TX buffers */
715 	for (i = 0; i < RE_TX_QLEN; i++) {
716 		error = bus_dmamap_create(sc->sc_dmat,
717 		    round_page(IP_MAXPACKET),
718 		    RE_TX_DESC_CNT(sc), RE_TDESC_CMD_FRAGLEN,
719 		    0, 0, &sc->re_ldata.re_txq[i].txq_dmamap);
720 		if (error) {
721 			aprint_error_dev(sc->sc_dev,
722 			    "can't create DMA map for TX\n");
723 			goto fail_4;
724 		}
725 	}
726 
727 	/* Allocate DMA'able memory for the RX ring */
728 	/* XXX see also a comment about RE_RX_DMAMEM_SZ in rtl81x9var.h */
729 	if ((error = bus_dmamem_alloc(sc->sc_dmat,
730 	    RE_RX_DMAMEM_SZ, RE_RING_ALIGN, 0, &sc->re_ldata.re_rx_listseg, 1,
731 	    &sc->re_ldata.re_rx_listnseg, BUS_DMA_NOWAIT)) != 0) {
732 		aprint_error_dev(sc->sc_dev,
733 		    "can't allocate rx listseg, error = %d\n", error);
734 		goto fail_4;
735 	}
736 
737 	/* Load the map for the RX ring. */
738 	if ((error = bus_dmamem_map(sc->sc_dmat, &sc->re_ldata.re_rx_listseg,
739 	    sc->re_ldata.re_rx_listnseg, RE_RX_DMAMEM_SZ,
740 	    (void **)&sc->re_ldata.re_rx_list,
741 	    BUS_DMA_COHERENT | BUS_DMA_NOWAIT)) != 0) {
742 		aprint_error_dev(sc->sc_dev,
743 		    "can't map rx list, error = %d\n", error);
744 		goto fail_5;
745 	}
746 	memset(sc->re_ldata.re_rx_list, 0, RE_RX_DMAMEM_SZ);
747 
748 	if ((error = bus_dmamap_create(sc->sc_dmat,
749 	    RE_RX_DMAMEM_SZ, 1, RE_RX_DMAMEM_SZ, 0, 0,
750 	    &sc->re_ldata.re_rx_list_map)) != 0) {
751 		aprint_error_dev(sc->sc_dev,
752 		    "can't create rx list map, error = %d\n", error);
753 		goto fail_6;
754 	}
755 
756 	if ((error = bus_dmamap_load(sc->sc_dmat,
757 	    sc->re_ldata.re_rx_list_map, sc->re_ldata.re_rx_list,
758 	    RE_RX_DMAMEM_SZ, NULL, BUS_DMA_NOWAIT)) != 0) {
759 		aprint_error_dev(sc->sc_dev,
760 		    "can't load rx list, error = %d\n", error);
761 		goto fail_7;
762 	}
763 
764 	/* Create DMA maps for RX buffers */
765 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
766 		error = bus_dmamap_create(sc->sc_dmat, MCLBYTES, 1, MCLBYTES,
767 		    0, 0, &sc->re_ldata.re_rxsoft[i].rxs_dmamap);
768 		if (error) {
769 			aprint_error_dev(sc->sc_dev,
770 			    "can't create DMA map for RX\n");
771 			goto fail_8;
772 		}
773 	}
774 
775 	/*
776 	 * Record interface as attached. From here, we should not fail.
777 	 */
778 	sc->sc_flags |= RTK_ATTACHED;
779 
780 	ifp = &sc->ethercom.ec_if;
781 	ifp->if_softc = sc;
782 	strlcpy(ifp->if_xname, device_xname(sc->sc_dev), IFNAMSIZ);
783 	ifp->if_mtu = ETHERMTU;
784 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
785 	ifp->if_ioctl = re_ioctl;
786 	sc->ethercom.ec_capabilities |=
787 	    ETHERCAP_VLAN_MTU | ETHERCAP_VLAN_HWTAGGING;
788 	ifp->if_start = re_start;
789 	ifp->if_stop = re_stop;
790 
791 	/*
792 	 * IFCAP_CSUM_IPv4_Tx on re(4) is broken for small packets,
793 	 * so we have a workaround to handle the bug by padding
794 	 * such packets manually.
795 	 */
796 	ifp->if_capabilities |=
797 	    IFCAP_CSUM_IPv4_Tx | IFCAP_CSUM_IPv4_Rx |
798 	    IFCAP_CSUM_TCPv4_Tx | IFCAP_CSUM_TCPv4_Rx |
799 	    IFCAP_CSUM_UDPv4_Tx | IFCAP_CSUM_UDPv4_Rx |
800 	    IFCAP_TSOv4;
801 
802 	/*
803 	 * XXX
804 	 * Still have no idea how to make TSO work on 8168C, 8168CP,
805 	 * 8102E, 8111C and 8111CP.
806 	 */
807 	if ((sc->sc_quirk & RTKQ_DESCV2) != 0)
808 		ifp->if_capabilities &= ~IFCAP_TSOv4;
809 
810 	ifp->if_watchdog = re_watchdog;
811 	ifp->if_init = re_init;
812 	ifp->if_snd.ifq_maxlen = RE_IFQ_MAXLEN;
813 	ifp->if_capenable = ifp->if_capabilities;
814 	IFQ_SET_READY(&ifp->if_snd);
815 
816 	callout_init(&sc->rtk_tick_ch, 0);
817 
818 	/* Do MII setup */
819 	sc->mii.mii_ifp = ifp;
820 	sc->mii.mii_readreg = re_miibus_readreg;
821 	sc->mii.mii_writereg = re_miibus_writereg;
822 	sc->mii.mii_statchg = re_miibus_statchg;
823 	sc->ethercom.ec_mii = &sc->mii;
824 	ifmedia_init(&sc->mii.mii_media, IFM_IMASK, ether_mediachange,
825 	    ether_mediastatus);
826 	mii_attach(sc->sc_dev, &sc->mii, 0xffffffff, MII_PHY_ANY,
827 	    MII_OFFSET_ANY, 0);
828 	ifmedia_set(&sc->mii.mii_media, IFM_ETHER | IFM_AUTO);
829 
830 	/*
831 	 * Call MI attach routine.
832 	 */
833 	if_attach(ifp);
834 	ether_ifattach(ifp, eaddr);
835 
836 	return;
837 
838  fail_8:
839 	/* Destroy DMA maps for RX buffers. */
840 	for (i = 0; i < RE_RX_DESC_CNT; i++)
841 		if (sc->re_ldata.re_rxsoft[i].rxs_dmamap != NULL)
842 			bus_dmamap_destroy(sc->sc_dmat,
843 			    sc->re_ldata.re_rxsoft[i].rxs_dmamap);
844 
845 	/* Free DMA'able memory for the RX ring. */
846 	bus_dmamap_unload(sc->sc_dmat, sc->re_ldata.re_rx_list_map);
847  fail_7:
848 	bus_dmamap_destroy(sc->sc_dmat, sc->re_ldata.re_rx_list_map);
849  fail_6:
850 	bus_dmamem_unmap(sc->sc_dmat,
851 	    (void *)sc->re_ldata.re_rx_list, RE_RX_DMAMEM_SZ);
852  fail_5:
853 	bus_dmamem_free(sc->sc_dmat,
854 	    &sc->re_ldata.re_rx_listseg, sc->re_ldata.re_rx_listnseg);
855 
856  fail_4:
857 	/* Destroy DMA maps for TX buffers. */
858 	for (i = 0; i < RE_TX_QLEN; i++)
859 		if (sc->re_ldata.re_txq[i].txq_dmamap != NULL)
860 			bus_dmamap_destroy(sc->sc_dmat,
861 			    sc->re_ldata.re_txq[i].txq_dmamap);
862 
863 	/* Free DMA'able memory for the TX ring. */
864 	bus_dmamap_unload(sc->sc_dmat, sc->re_ldata.re_tx_list_map);
865  fail_3:
866 	bus_dmamap_destroy(sc->sc_dmat, sc->re_ldata.re_tx_list_map);
867  fail_2:
868 	bus_dmamem_unmap(sc->sc_dmat,
869 	    (void *)sc->re_ldata.re_tx_list, RE_TX_LIST_SZ(sc));
870  fail_1:
871 	bus_dmamem_free(sc->sc_dmat,
872 	    &sc->re_ldata.re_tx_listseg, sc->re_ldata.re_tx_listnseg);
873  fail_0:
874 	return;
875 }
876 
877 
878 /*
879  * re_activate:
880  *     Handle device activation/deactivation requests.
881  */
882 int
883 re_activate(device_t self, enum devact act)
884 {
885 	struct rtk_softc *sc = device_private(self);
886 	int s, error = 0;
887 
888 	s = splnet();
889 	switch (act) {
890 	case DVACT_ACTIVATE:
891 		error = EOPNOTSUPP;
892 		break;
893 	case DVACT_DEACTIVATE:
894 		mii_activate(&sc->mii, act, MII_PHY_ANY, MII_OFFSET_ANY);
895 		if_deactivate(&sc->ethercom.ec_if);
896 		break;
897 	}
898 	splx(s);
899 
900 	return error;
901 }
902 
903 /*
904  * re_detach:
905  *     Detach a rtk interface.
906  */
907 int
908 re_detach(struct rtk_softc *sc)
909 {
910 	struct ifnet *ifp = &sc->ethercom.ec_if;
911 	int i;
912 
913 	/*
914 	 * Succeed now if there isn't any work to do.
915 	 */
916 	if ((sc->sc_flags & RTK_ATTACHED) == 0)
917 		return 0;
918 
919 	/* Unhook our tick handler. */
920 	callout_stop(&sc->rtk_tick_ch);
921 
922 	/* Detach all PHYs. */
923 	mii_detach(&sc->mii, MII_PHY_ANY, MII_OFFSET_ANY);
924 
925 	/* Delete all remaining media. */
926 	ifmedia_delete_instance(&sc->mii.mii_media, IFM_INST_ANY);
927 
928 	ether_ifdetach(ifp);
929 	if_detach(ifp);
930 
931 	/* Destroy DMA maps for RX buffers. */
932 	for (i = 0; i < RE_RX_DESC_CNT; i++)
933 		if (sc->re_ldata.re_rxsoft[i].rxs_dmamap != NULL)
934 			bus_dmamap_destroy(sc->sc_dmat,
935 			    sc->re_ldata.re_rxsoft[i].rxs_dmamap);
936 
937 	/* Free DMA'able memory for the RX ring. */
938 	bus_dmamap_unload(sc->sc_dmat, sc->re_ldata.re_rx_list_map);
939 	bus_dmamap_destroy(sc->sc_dmat, sc->re_ldata.re_rx_list_map);
940 	bus_dmamem_unmap(sc->sc_dmat,
941 	    (void *)sc->re_ldata.re_rx_list, RE_RX_DMAMEM_SZ);
942 	bus_dmamem_free(sc->sc_dmat,
943 	    &sc->re_ldata.re_rx_listseg, sc->re_ldata.re_rx_listnseg);
944 
945 	/* Destroy DMA maps for TX buffers. */
946 	for (i = 0; i < RE_TX_QLEN; i++)
947 		if (sc->re_ldata.re_txq[i].txq_dmamap != NULL)
948 			bus_dmamap_destroy(sc->sc_dmat,
949 			    sc->re_ldata.re_txq[i].txq_dmamap);
950 
951 	/* Free DMA'able memory for the TX ring. */
952 	bus_dmamap_unload(sc->sc_dmat, sc->re_ldata.re_tx_list_map);
953 	bus_dmamap_destroy(sc->sc_dmat, sc->re_ldata.re_tx_list_map);
954 	bus_dmamem_unmap(sc->sc_dmat,
955 	    (void *)sc->re_ldata.re_tx_list, RE_TX_LIST_SZ(sc));
956 	bus_dmamem_free(sc->sc_dmat,
957 	    &sc->re_ldata.re_tx_listseg, sc->re_ldata.re_tx_listnseg);
958 
959 	return 0;
960 }
961 
962 /*
963  * re_enable:
964  *     Enable the RTL81X9 chip.
965  */
966 static int
967 re_enable(struct rtk_softc *sc)
968 {
969 
970 	if (RTK_IS_ENABLED(sc) == 0 && sc->sc_enable != NULL) {
971 		if ((*sc->sc_enable)(sc) != 0) {
972 			printf("%s: device enable failed\n",
973 			    device_xname(sc->sc_dev));
974 			return EIO;
975 		}
976 		sc->sc_flags |= RTK_ENABLED;
977 	}
978 	return 0;
979 }
980 
981 /*
982  * re_disable:
983  *     Disable the RTL81X9 chip.
984  */
985 static void
986 re_disable(struct rtk_softc *sc)
987 {
988 
989 	if (RTK_IS_ENABLED(sc) && sc->sc_disable != NULL) {
990 		(*sc->sc_disable)(sc);
991 		sc->sc_flags &= ~RTK_ENABLED;
992 	}
993 }
994 
995 static int
996 re_newbuf(struct rtk_softc *sc, int idx, struct mbuf *m)
997 {
998 	struct mbuf *n = NULL;
999 	bus_dmamap_t map;
1000 	struct re_desc *d;
1001 	struct re_rxsoft *rxs;
1002 	uint32_t cmdstat;
1003 	int error;
1004 
1005 	if (m == NULL) {
1006 		MGETHDR(n, M_DONTWAIT, MT_DATA);
1007 		if (n == NULL)
1008 			return ENOBUFS;
1009 
1010 		MCLGET(n, M_DONTWAIT);
1011 		if ((n->m_flags & M_EXT) == 0) {
1012 			m_freem(n);
1013 			return ENOBUFS;
1014 		}
1015 		m = n;
1016 	} else
1017 		m->m_data = m->m_ext.ext_buf;
1018 
1019 	/*
1020 	 * Initialize mbuf length fields and fixup
1021 	 * alignment so that the frame payload is
1022 	 * longword aligned.
1023 	 */
1024 	m->m_len = m->m_pkthdr.len = MCLBYTES - RE_ETHER_ALIGN;
1025 	m->m_data += RE_ETHER_ALIGN;
1026 
1027 	rxs = &sc->re_ldata.re_rxsoft[idx];
1028 	map = rxs->rxs_dmamap;
1029 	error = bus_dmamap_load_mbuf(sc->sc_dmat, map, m,
1030 	    BUS_DMA_READ|BUS_DMA_NOWAIT);
1031 
1032 	if (error)
1033 		goto out;
1034 
1035 	bus_dmamap_sync(sc->sc_dmat, map, 0, map->dm_mapsize,
1036 	    BUS_DMASYNC_PREREAD);
1037 
1038 	d = &sc->re_ldata.re_rx_list[idx];
1039 #ifdef DIAGNOSTIC
1040 	RE_RXDESCSYNC(sc, idx, BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1041 	cmdstat = le32toh(d->re_cmdstat);
1042 	RE_RXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD);
1043 	if (cmdstat & RE_RDESC_STAT_OWN) {
1044 		panic("%s: tried to map busy RX descriptor",
1045 		    device_xname(sc->sc_dev));
1046 	}
1047 #endif
1048 
1049 	rxs->rxs_mbuf = m;
1050 
1051 	d->re_vlanctl = 0;
1052 	cmdstat = map->dm_segs[0].ds_len;
1053 	if (idx == (RE_RX_DESC_CNT - 1))
1054 		cmdstat |= RE_RDESC_CMD_EOR;
1055 	re_set_bufaddr(d, map->dm_segs[0].ds_addr);
1056 	d->re_cmdstat = htole32(cmdstat);
1057 	RE_RXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1058 	cmdstat |= RE_RDESC_CMD_OWN;
1059 	d->re_cmdstat = htole32(cmdstat);
1060 	RE_RXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1061 
1062 	return 0;
1063  out:
1064 	if (n != NULL)
1065 		m_freem(n);
1066 	return ENOMEM;
1067 }
1068 
1069 static int
1070 re_tx_list_init(struct rtk_softc *sc)
1071 {
1072 	int i;
1073 
1074 	memset(sc->re_ldata.re_tx_list, 0, RE_TX_LIST_SZ(sc));
1075 	for (i = 0; i < RE_TX_QLEN; i++) {
1076 		sc->re_ldata.re_txq[i].txq_mbuf = NULL;
1077 	}
1078 
1079 	bus_dmamap_sync(sc->sc_dmat,
1080 	    sc->re_ldata.re_tx_list_map, 0,
1081 	    sc->re_ldata.re_tx_list_map->dm_mapsize,
1082 	    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1083 	sc->re_ldata.re_txq_prodidx = 0;
1084 	sc->re_ldata.re_txq_considx = 0;
1085 	sc->re_ldata.re_txq_free = RE_TX_QLEN;
1086 	sc->re_ldata.re_tx_free = RE_TX_DESC_CNT(sc);
1087 	sc->re_ldata.re_tx_nextfree = 0;
1088 
1089 	return 0;
1090 }
1091 
1092 static int
1093 re_rx_list_init(struct rtk_softc *sc)
1094 {
1095 	int i;
1096 
1097 	memset(sc->re_ldata.re_rx_list, 0, RE_RX_LIST_SZ);
1098 
1099 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
1100 		if (re_newbuf(sc, i, NULL) == ENOBUFS)
1101 			return ENOBUFS;
1102 	}
1103 
1104 	sc->re_ldata.re_rx_prodidx = 0;
1105 	sc->re_head = sc->re_tail = NULL;
1106 
1107 	return 0;
1108 }
1109 
1110 /*
1111  * RX handler for C+ and 8169. For the gigE chips, we support
1112  * the reception of jumbo frames that have been fragmented
1113  * across multiple 2K mbuf cluster buffers.
1114  */
1115 static void
1116 re_rxeof(struct rtk_softc *sc)
1117 {
1118 	struct mbuf *m;
1119 	struct ifnet *ifp;
1120 	int i, total_len;
1121 	struct re_desc *cur_rx;
1122 	struct re_rxsoft *rxs;
1123 	uint32_t rxstat, rxvlan;
1124 
1125 	ifp = &sc->ethercom.ec_if;
1126 
1127 	for (i = sc->re_ldata.re_rx_prodidx;; i = RE_NEXT_RX_DESC(sc, i)) {
1128 		cur_rx = &sc->re_ldata.re_rx_list[i];
1129 		RE_RXDESCSYNC(sc, i,
1130 		    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1131 		rxstat = le32toh(cur_rx->re_cmdstat);
1132 		rxvlan = le32toh(cur_rx->re_vlanctl);
1133 		RE_RXDESCSYNC(sc, i, BUS_DMASYNC_PREREAD);
1134 		if ((rxstat & RE_RDESC_STAT_OWN) != 0) {
1135 			break;
1136 		}
1137 		total_len = rxstat & sc->re_rxlenmask;
1138 		rxs = &sc->re_ldata.re_rxsoft[i];
1139 		m = rxs->rxs_mbuf;
1140 
1141 		/* Invalidate the RX mbuf and unload its map */
1142 
1143 		bus_dmamap_sync(sc->sc_dmat,
1144 		    rxs->rxs_dmamap, 0, rxs->rxs_dmamap->dm_mapsize,
1145 		    BUS_DMASYNC_POSTREAD);
1146 		bus_dmamap_unload(sc->sc_dmat, rxs->rxs_dmamap);
1147 
1148 		if ((rxstat & RE_RDESC_STAT_EOF) == 0) {
1149 			m->m_len = MCLBYTES - RE_ETHER_ALIGN;
1150 			if (sc->re_head == NULL)
1151 				sc->re_head = sc->re_tail = m;
1152 			else {
1153 				m->m_flags &= ~M_PKTHDR;
1154 				sc->re_tail->m_next = m;
1155 				sc->re_tail = m;
1156 			}
1157 			re_newbuf(sc, i, NULL);
1158 			continue;
1159 		}
1160 
1161 		/*
1162 		 * NOTE: for the 8139C+, the frame length field
1163 		 * is always 12 bits in size, but for the gigE chips,
1164 		 * it is 13 bits (since the max RX frame length is 16K).
1165 		 * Unfortunately, all 32 bits in the status word
1166 		 * were already used, so to make room for the extra
1167 		 * length bit, RealTek took out the 'frame alignment
1168 		 * error' bit and shifted the other status bits
1169 		 * over one slot. The OWN, EOR, FS and LS bits are
1170 		 * still in the same places. We have already extracted
1171 		 * the frame length and checked the OWN bit, so rather
1172 		 * than using an alternate bit mapping, we shift the
1173 		 * status bits one space to the right so we can evaluate
1174 		 * them using the 8169 status as though it was in the
1175 		 * same format as that of the 8139C+.
1176 		 */
1177 		if ((sc->sc_quirk & RTKQ_8139CPLUS) == 0)
1178 			rxstat >>= 1;
1179 
1180 		if (__predict_false((rxstat & RE_RDESC_STAT_RXERRSUM) != 0)) {
1181 #ifdef RE_DEBUG
1182 			printf("%s: RX error (rxstat = 0x%08x)",
1183 			    device_xname(sc->sc_dev), rxstat);
1184 			if (rxstat & RE_RDESC_STAT_FRALIGN)
1185 				printf(", frame alignment error");
1186 			if (rxstat & RE_RDESC_STAT_BUFOFLOW)
1187 				printf(", out of buffer space");
1188 			if (rxstat & RE_RDESC_STAT_FIFOOFLOW)
1189 				printf(", FIFO overrun");
1190 			if (rxstat & RE_RDESC_STAT_GIANT)
1191 				printf(", giant packet");
1192 			if (rxstat & RE_RDESC_STAT_RUNT)
1193 				printf(", runt packet");
1194 			if (rxstat & RE_RDESC_STAT_CRCERR)
1195 				printf(", CRC error");
1196 			printf("\n");
1197 #endif
1198 			ifp->if_ierrors++;
1199 			/*
1200 			 * If this is part of a multi-fragment packet,
1201 			 * discard all the pieces.
1202 			 */
1203 			if (sc->re_head != NULL) {
1204 				m_freem(sc->re_head);
1205 				sc->re_head = sc->re_tail = NULL;
1206 			}
1207 			re_newbuf(sc, i, m);
1208 			continue;
1209 		}
1210 
1211 		/*
1212 		 * If allocating a replacement mbuf fails,
1213 		 * reload the current one.
1214 		 */
1215 
1216 		if (__predict_false(re_newbuf(sc, i, NULL) != 0)) {
1217 			ifp->if_ierrors++;
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 		if (sc->re_head != NULL) {
1227 			m->m_len = total_len % (MCLBYTES - RE_ETHER_ALIGN);
1228 			/*
1229 			 * Special case: if there's 4 bytes or less
1230 			 * in this buffer, the mbuf can be discarded:
1231 			 * the last 4 bytes is the CRC, which we don't
1232 			 * care about anyway.
1233 			 */
1234 			if (m->m_len <= ETHER_CRC_LEN) {
1235 				sc->re_tail->m_len -=
1236 				    (ETHER_CRC_LEN - m->m_len);
1237 				m_freem(m);
1238 			} else {
1239 				m->m_len -= ETHER_CRC_LEN;
1240 				m->m_flags &= ~M_PKTHDR;
1241 				sc->re_tail->m_next = m;
1242 			}
1243 			m = sc->re_head;
1244 			sc->re_head = sc->re_tail = NULL;
1245 			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
1246 		} else
1247 			m->m_pkthdr.len = m->m_len =
1248 			    (total_len - ETHER_CRC_LEN);
1249 
1250 		ifp->if_ipackets++;
1251 		m->m_pkthdr.rcvif = ifp;
1252 
1253 		/* Do RX checksumming */
1254 		if ((sc->sc_quirk & RTKQ_DESCV2) == 0) {
1255 			/* Check IP header checksum */
1256 			if ((rxstat & RE_RDESC_STAT_PROTOID) != 0) {
1257 				m->m_pkthdr.csum_flags |= M_CSUM_IPv4;
1258 				if (rxstat & RE_RDESC_STAT_IPSUMBAD)
1259 					m->m_pkthdr.csum_flags |=
1260 					    M_CSUM_IPv4_BAD;
1261 
1262 				/* Check TCP/UDP checksum */
1263 				if (RE_TCPPKT(rxstat)) {
1264 					m->m_pkthdr.csum_flags |= M_CSUM_TCPv4;
1265 					if (rxstat & RE_RDESC_STAT_TCPSUMBAD)
1266 						m->m_pkthdr.csum_flags |=
1267 						    M_CSUM_TCP_UDP_BAD;
1268 				} else if (RE_UDPPKT(rxstat)) {
1269 					m->m_pkthdr.csum_flags |= M_CSUM_UDPv4;
1270 					if (rxstat & RE_RDESC_STAT_UDPSUMBAD)
1271 						m->m_pkthdr.csum_flags |=
1272 						    M_CSUM_TCP_UDP_BAD;
1273 				}
1274 			}
1275 		} else {
1276 			/* Check IPv4 header checksum */
1277 			if ((rxvlan & RE_RDESC_VLANCTL_IPV4) != 0) {
1278 				m->m_pkthdr.csum_flags |= M_CSUM_IPv4;
1279 				if (rxstat & RE_RDESC_STAT_IPSUMBAD)
1280 					m->m_pkthdr.csum_flags |=
1281 					    M_CSUM_IPv4_BAD;
1282 
1283 				/* Check TCPv4/UDPv4 checksum */
1284 				if (RE_TCPPKT(rxstat)) {
1285 					m->m_pkthdr.csum_flags |= M_CSUM_TCPv4;
1286 					if (rxstat & RE_RDESC_STAT_TCPSUMBAD)
1287 						m->m_pkthdr.csum_flags |=
1288 						    M_CSUM_TCP_UDP_BAD;
1289 				} else if (RE_UDPPKT(rxstat)) {
1290 					m->m_pkthdr.csum_flags |= M_CSUM_UDPv4;
1291 					if (rxstat & RE_RDESC_STAT_UDPSUMBAD)
1292 						m->m_pkthdr.csum_flags |=
1293 						    M_CSUM_TCP_UDP_BAD;
1294 				}
1295 			}
1296 			/* XXX Check TCPv6/UDPv6 checksum? */
1297 		}
1298 
1299 		if (rxvlan & RE_RDESC_VLANCTL_TAG) {
1300 			VLAN_INPUT_TAG(ifp, m,
1301 			     bswap16(rxvlan & RE_RDESC_VLANCTL_DATA),
1302 			     continue);
1303 		}
1304 #if NBPFILTER > 0
1305 		if (ifp->if_bpf)
1306 			bpf_mtap(ifp->if_bpf, m);
1307 #endif
1308 		(*ifp->if_input)(ifp, m);
1309 	}
1310 
1311 	sc->re_ldata.re_rx_prodidx = i;
1312 }
1313 
1314 static void
1315 re_txeof(struct rtk_softc *sc)
1316 {
1317 	struct ifnet *ifp;
1318 	struct re_txq *txq;
1319 	uint32_t txstat;
1320 	int idx, descidx;
1321 
1322 	ifp = &sc->ethercom.ec_if;
1323 
1324 	for (idx = sc->re_ldata.re_txq_considx;
1325 	    sc->re_ldata.re_txq_free < RE_TX_QLEN;
1326 	    idx = RE_NEXT_TXQ(sc, idx), sc->re_ldata.re_txq_free++) {
1327 		txq = &sc->re_ldata.re_txq[idx];
1328 		KASSERT(txq->txq_mbuf != NULL);
1329 
1330 		descidx = txq->txq_descidx;
1331 		RE_TXDESCSYNC(sc, descidx,
1332 		    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1333 		txstat =
1334 		    le32toh(sc->re_ldata.re_tx_list[descidx].re_cmdstat);
1335 		RE_TXDESCSYNC(sc, descidx, BUS_DMASYNC_PREREAD);
1336 		KASSERT((txstat & RE_TDESC_CMD_EOF) != 0);
1337 		if (txstat & RE_TDESC_CMD_OWN) {
1338 			break;
1339 		}
1340 
1341 		sc->re_ldata.re_tx_free += txq->txq_nsegs;
1342 		KASSERT(sc->re_ldata.re_tx_free <= RE_TX_DESC_CNT(sc));
1343 		bus_dmamap_sync(sc->sc_dmat, txq->txq_dmamap,
1344 		    0, txq->txq_dmamap->dm_mapsize, BUS_DMASYNC_POSTWRITE);
1345 		bus_dmamap_unload(sc->sc_dmat, txq->txq_dmamap);
1346 		m_freem(txq->txq_mbuf);
1347 		txq->txq_mbuf = NULL;
1348 
1349 		if (txstat & (RE_TDESC_STAT_EXCESSCOL | RE_TDESC_STAT_COLCNT))
1350 			ifp->if_collisions++;
1351 		if (txstat & RE_TDESC_STAT_TXERRSUM)
1352 			ifp->if_oerrors++;
1353 		else
1354 			ifp->if_opackets++;
1355 	}
1356 
1357 	sc->re_ldata.re_txq_considx = idx;
1358 
1359 	if (sc->re_ldata.re_txq_free > RE_NTXDESC_RSVD)
1360 		ifp->if_flags &= ~IFF_OACTIVE;
1361 
1362 	/*
1363 	 * If not all descriptors have been released reaped yet,
1364 	 * reload the timer so that we will eventually get another
1365 	 * interrupt that will cause us to re-enter this routine.
1366 	 * This is done in case the transmitter has gone idle.
1367 	 */
1368 	if (sc->re_ldata.re_txq_free < RE_TX_QLEN) {
1369 		CSR_WRITE_4(sc, RTK_TIMERCNT, 1);
1370 		if ((sc->sc_quirk & RTKQ_PCIE) != 0) {
1371 			/*
1372 			 * Some chips will ignore a second TX request
1373 			 * issued while an existing transmission is in
1374 			 * progress. If the transmitter goes idle but
1375 			 * there are still packets waiting to be sent,
1376 			 * we need to restart the channel here to flush
1377 			 * them out. This only seems to be required with
1378 			 * the PCIe devices.
1379 			 */
1380 			CSR_WRITE_1(sc, RTK_GTXSTART, RTK_TXSTART_START);
1381 		}
1382 	} else
1383 		ifp->if_timer = 0;
1384 }
1385 
1386 static void
1387 re_tick(void *arg)
1388 {
1389 	struct rtk_softc *sc = arg;
1390 	int s;
1391 
1392 	/*XXX: just return for 8169S/8110S with rev 2 or newer phy */
1393 	s = splnet();
1394 
1395 	mii_tick(&sc->mii);
1396 	splx(s);
1397 
1398 	callout_reset(&sc->rtk_tick_ch, hz, re_tick, sc);
1399 }
1400 
1401 int
1402 re_intr(void *arg)
1403 {
1404 	struct rtk_softc *sc = arg;
1405 	struct ifnet *ifp;
1406 	uint16_t status;
1407 	int handled = 0;
1408 
1409 	if (!device_has_power(sc->sc_dev))
1410 		return 0;
1411 
1412 	ifp = &sc->ethercom.ec_if;
1413 
1414 	if ((ifp->if_flags & IFF_UP) == 0)
1415 		return 0;
1416 
1417 	for (;;) {
1418 
1419 		status = CSR_READ_2(sc, RTK_ISR);
1420 		/* If the card has gone away the read returns 0xffff. */
1421 		if (status == 0xffff)
1422 			break;
1423 		if (status) {
1424 			handled = 1;
1425 			CSR_WRITE_2(sc, RTK_ISR, status);
1426 		}
1427 
1428 		if ((status & RTK_INTRS_CPLUS) == 0)
1429 			break;
1430 
1431 		if (status & (RTK_ISR_RX_OK | RTK_ISR_RX_ERR))
1432 			re_rxeof(sc);
1433 
1434 		if (status & (RTK_ISR_TIMEOUT_EXPIRED | RTK_ISR_TX_ERR |
1435 		    RTK_ISR_TX_DESC_UNAVAIL))
1436 			re_txeof(sc);
1437 
1438 		if (status & RTK_ISR_SYSTEM_ERR) {
1439 			re_init(ifp);
1440 		}
1441 
1442 		if (status & RTK_ISR_LINKCHG) {
1443 			callout_stop(&sc->rtk_tick_ch);
1444 			re_tick(sc);
1445 		}
1446 	}
1447 
1448 	if (handled && !IFQ_IS_EMPTY(&ifp->if_snd))
1449 		re_start(ifp);
1450 
1451 	return handled;
1452 }
1453 
1454 
1455 
1456 /*
1457  * Main transmit routine for C+ and gigE NICs.
1458  */
1459 
1460 static void
1461 re_start(struct ifnet *ifp)
1462 {
1463 	struct rtk_softc *sc;
1464 	struct mbuf *m;
1465 	bus_dmamap_t map;
1466 	struct re_txq *txq;
1467 	struct re_desc *d;
1468 	struct m_tag *mtag;
1469 	uint32_t cmdstat, re_flags, vlanctl;
1470 	int ofree, idx, error, nsegs, seg;
1471 	int startdesc, curdesc, lastdesc;
1472 	bool pad;
1473 
1474 	sc = ifp->if_softc;
1475 	ofree = sc->re_ldata.re_txq_free;
1476 
1477 	for (idx = sc->re_ldata.re_txq_prodidx;; idx = RE_NEXT_TXQ(sc, idx)) {
1478 
1479 		IFQ_POLL(&ifp->if_snd, m);
1480 		if (m == NULL)
1481 			break;
1482 
1483 		if (sc->re_ldata.re_txq_free == 0 ||
1484 		    sc->re_ldata.re_tx_free == 0) {
1485 			/* no more free slots left */
1486 			ifp->if_flags |= IFF_OACTIVE;
1487 			break;
1488 		}
1489 
1490 		/*
1491 		 * Set up checksum offload. Note: checksum offload bits must
1492 		 * appear in all descriptors of a multi-descriptor transmit
1493 		 * attempt. (This is according to testing done with an 8169
1494 		 * chip. I'm not sure if this is a requirement or a bug.)
1495 		 */
1496 
1497 		vlanctl = 0;
1498 		if ((m->m_pkthdr.csum_flags & M_CSUM_TSOv4) != 0) {
1499 			uint32_t segsz = m->m_pkthdr.segsz;
1500 
1501 			re_flags = RE_TDESC_CMD_LGSEND |
1502 			    (segsz << RE_TDESC_CMD_MSSVAL_SHIFT);
1503 		} else {
1504 			/*
1505 			 * set RE_TDESC_CMD_IPCSUM if any checksum offloading
1506 			 * is requested.  otherwise, RE_TDESC_CMD_TCPCSUM/
1507 			 * RE_TDESC_CMD_UDPCSUM doesn't make effects.
1508 			 */
1509 			re_flags = 0;
1510 			if ((m->m_pkthdr.csum_flags &
1511 			    (M_CSUM_IPv4 | M_CSUM_TCPv4 | M_CSUM_UDPv4))
1512 			    != 0) {
1513 				if ((sc->sc_quirk & RTKQ_DESCV2) == 0) {
1514 					re_flags |= RE_TDESC_CMD_IPCSUM;
1515 					if (m->m_pkthdr.csum_flags &
1516 					    M_CSUM_TCPv4) {
1517 						re_flags |=
1518 						    RE_TDESC_CMD_TCPCSUM;
1519 					} else if (m->m_pkthdr.csum_flags &
1520 					    M_CSUM_UDPv4) {
1521 						re_flags |=
1522 						    RE_TDESC_CMD_UDPCSUM;
1523 					}
1524 				} else {
1525 					vlanctl |= RE_TDESC_VLANCTL_IPCSUM;
1526 					if (m->m_pkthdr.csum_flags &
1527 					    M_CSUM_TCPv4) {
1528 						vlanctl |=
1529 						    RE_TDESC_VLANCTL_TCPCSUM;
1530 					} else if (m->m_pkthdr.csum_flags &
1531 					    M_CSUM_UDPv4) {
1532 						vlanctl |=
1533 						    RE_TDESC_VLANCTL_UDPCSUM;
1534 					}
1535 				}
1536 			}
1537 		}
1538 
1539 		txq = &sc->re_ldata.re_txq[idx];
1540 		map = txq->txq_dmamap;
1541 		error = bus_dmamap_load_mbuf(sc->sc_dmat, map, m,
1542 		    BUS_DMA_WRITE|BUS_DMA_NOWAIT);
1543 
1544 		if (__predict_false(error)) {
1545 			/* XXX try to defrag if EFBIG? */
1546 			printf("%s: can't map mbuf (error %d)\n",
1547 			    device_xname(sc->sc_dev), error);
1548 
1549 			IFQ_DEQUEUE(&ifp->if_snd, m);
1550 			m_freem(m);
1551 			ifp->if_oerrors++;
1552 			continue;
1553 		}
1554 
1555 		nsegs = map->dm_nsegs;
1556 		pad = false;
1557 		if (__predict_false(m->m_pkthdr.len <= RE_IP4CSUMTX_PADLEN &&
1558 		    (re_flags & RE_TDESC_CMD_IPCSUM) != 0 &&
1559 		    (sc->sc_quirk & RTKQ_DESCV2) == 0)) {
1560 			pad = true;
1561 			nsegs++;
1562 		}
1563 
1564 		if (nsegs > sc->re_ldata.re_tx_free) {
1565 			/*
1566 			 * Not enough free descriptors to transmit this packet.
1567 			 */
1568 			ifp->if_flags |= IFF_OACTIVE;
1569 			bus_dmamap_unload(sc->sc_dmat, map);
1570 			break;
1571 		}
1572 
1573 		IFQ_DEQUEUE(&ifp->if_snd, m);
1574 
1575 		/*
1576 		 * Make sure that the caches are synchronized before we
1577 		 * ask the chip to start DMA for the packet data.
1578 		 */
1579 		bus_dmamap_sync(sc->sc_dmat, map, 0, map->dm_mapsize,
1580 		    BUS_DMASYNC_PREWRITE);
1581 
1582 		/*
1583 		 * Set up hardware VLAN tagging. Note: vlan tag info must
1584 		 * appear in all descriptors of a multi-descriptor
1585 		 * transmission attempt.
1586 		 */
1587 		if ((mtag = VLAN_OUTPUT_TAG(&sc->ethercom, m)) != NULL)
1588 			vlanctl |= bswap16(VLAN_TAG_VALUE(mtag)) |
1589 			    RE_TDESC_VLANCTL_TAG;
1590 
1591 		/*
1592 		 * Map the segment array into descriptors.
1593 		 * Note that we set the start-of-frame and
1594 		 * end-of-frame markers for either TX or RX,
1595 		 * but they really only have meaning in the TX case.
1596 		 * (In the RX case, it's the chip that tells us
1597 		 *  where packets begin and end.)
1598 		 * We also keep track of the end of the ring
1599 		 * and set the end-of-ring bits as needed,
1600 		 * and we set the ownership bits in all except
1601 		 * the very first descriptor. (The caller will
1602 		 * set this descriptor later when it start
1603 		 * transmission or reception.)
1604 		 */
1605 		curdesc = startdesc = sc->re_ldata.re_tx_nextfree;
1606 		lastdesc = -1;
1607 		for (seg = 0; seg < map->dm_nsegs;
1608 		    seg++, curdesc = RE_NEXT_TX_DESC(sc, curdesc)) {
1609 			d = &sc->re_ldata.re_tx_list[curdesc];
1610 #ifdef DIAGNOSTIC
1611 			RE_TXDESCSYNC(sc, curdesc,
1612 			    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1613 			cmdstat = le32toh(d->re_cmdstat);
1614 			RE_TXDESCSYNC(sc, curdesc, BUS_DMASYNC_PREREAD);
1615 			if (cmdstat & RE_TDESC_STAT_OWN) {
1616 				panic("%s: tried to map busy TX descriptor",
1617 				    device_xname(sc->sc_dev));
1618 			}
1619 #endif
1620 
1621 			d->re_vlanctl = htole32(vlanctl);
1622 			re_set_bufaddr(d, map->dm_segs[seg].ds_addr);
1623 			cmdstat = re_flags | map->dm_segs[seg].ds_len;
1624 			if (seg == 0)
1625 				cmdstat |= RE_TDESC_CMD_SOF;
1626 			else
1627 				cmdstat |= RE_TDESC_CMD_OWN;
1628 			if (curdesc == (RE_TX_DESC_CNT(sc) - 1))
1629 				cmdstat |= RE_TDESC_CMD_EOR;
1630 			if (seg == nsegs - 1) {
1631 				cmdstat |= RE_TDESC_CMD_EOF;
1632 				lastdesc = curdesc;
1633 			}
1634 			d->re_cmdstat = htole32(cmdstat);
1635 			RE_TXDESCSYNC(sc, curdesc,
1636 			    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1637 		}
1638 		if (__predict_false(pad)) {
1639 			bus_addr_t paddaddr;
1640 
1641 			d = &sc->re_ldata.re_tx_list[curdesc];
1642 			d->re_vlanctl = htole32(vlanctl);
1643 			paddaddr = RE_TXPADDADDR(sc);
1644 			re_set_bufaddr(d, paddaddr);
1645 			cmdstat = re_flags |
1646 			    RE_TDESC_CMD_OWN | RE_TDESC_CMD_EOF |
1647 			    (RE_IP4CSUMTX_PADLEN + 1 - m->m_pkthdr.len);
1648 			if (curdesc == (RE_TX_DESC_CNT(sc) - 1))
1649 				cmdstat |= RE_TDESC_CMD_EOR;
1650 			d->re_cmdstat = htole32(cmdstat);
1651 			RE_TXDESCSYNC(sc, curdesc,
1652 			    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1653 			lastdesc = curdesc;
1654 			curdesc = RE_NEXT_TX_DESC(sc, curdesc);
1655 		}
1656 		KASSERT(lastdesc != -1);
1657 
1658 		/* Transfer ownership of packet to the chip. */
1659 
1660 		sc->re_ldata.re_tx_list[startdesc].re_cmdstat |=
1661 		    htole32(RE_TDESC_CMD_OWN);
1662 		RE_TXDESCSYNC(sc, startdesc,
1663 		    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1664 
1665 		/* update info of TX queue and descriptors */
1666 		txq->txq_mbuf = m;
1667 		txq->txq_descidx = lastdesc;
1668 		txq->txq_nsegs = nsegs;
1669 
1670 		sc->re_ldata.re_txq_free--;
1671 		sc->re_ldata.re_tx_free -= nsegs;
1672 		sc->re_ldata.re_tx_nextfree = curdesc;
1673 
1674 #if NBPFILTER > 0
1675 		/*
1676 		 * If there's a BPF listener, bounce a copy of this frame
1677 		 * to him.
1678 		 */
1679 		if (ifp->if_bpf)
1680 			bpf_mtap(ifp->if_bpf, m);
1681 #endif
1682 	}
1683 
1684 	if (sc->re_ldata.re_txq_free < ofree) {
1685 		/*
1686 		 * TX packets are enqueued.
1687 		 */
1688 		sc->re_ldata.re_txq_prodidx = idx;
1689 
1690 		/*
1691 		 * Start the transmitter to poll.
1692 		 *
1693 		 * RealTek put the TX poll request register in a different
1694 		 * location on the 8169 gigE chip. I don't know why.
1695 		 */
1696 		if ((sc->sc_quirk & RTKQ_8139CPLUS) != 0)
1697 			CSR_WRITE_1(sc, RTK_TXSTART, RTK_TXSTART_START);
1698 		else
1699 			CSR_WRITE_1(sc, RTK_GTXSTART, RTK_TXSTART_START);
1700 
1701 		/*
1702 		 * Use the countdown timer for interrupt moderation.
1703 		 * 'TX done' interrupts are disabled. Instead, we reset the
1704 		 * countdown timer, which will begin counting until it hits
1705 		 * the value in the TIMERINT register, and then trigger an
1706 		 * interrupt. Each time we write to the TIMERCNT register,
1707 		 * the timer count is reset to 0.
1708 		 */
1709 		CSR_WRITE_4(sc, RTK_TIMERCNT, 1);
1710 
1711 		/*
1712 		 * Set a timeout in case the chip goes out to lunch.
1713 		 */
1714 		ifp->if_timer = 5;
1715 	}
1716 }
1717 
1718 static int
1719 re_init(struct ifnet *ifp)
1720 {
1721 	struct rtk_softc *sc = ifp->if_softc;
1722 	const uint8_t *enaddr;
1723 	uint32_t rxcfg = 0;
1724 	uint32_t reg;
1725 	uint16_t cfg;
1726 	int error;
1727 
1728 	if ((error = re_enable(sc)) != 0)
1729 		goto out;
1730 
1731 	/*
1732 	 * Cancel pending I/O and free all RX/TX buffers.
1733 	 */
1734 	re_stop(ifp, 0);
1735 
1736 	re_reset(sc);
1737 
1738 	/*
1739 	 * Enable C+ RX and TX mode, as well as VLAN stripping and
1740 	 * RX checksum offload. We must configure the C+ register
1741 	 * before all others.
1742 	 */
1743 	cfg = RE_CPLUSCMD_PCI_MRW;
1744 
1745 	/*
1746 	 * XXX: For old 8169 set bit 14.
1747 	 *      For 8169S/8110S and above, do not set bit 14.
1748 	 */
1749 	if ((sc->sc_quirk & RTKQ_8169NONS) != 0)
1750 		cfg |= (0x1 << 14);
1751 
1752 	if ((ifp->if_capenable & ETHERCAP_VLAN_HWTAGGING) != 0)
1753 		cfg |= RE_CPLUSCMD_VLANSTRIP;
1754 	if ((ifp->if_capenable & (IFCAP_CSUM_IPv4_Rx |
1755 	     IFCAP_CSUM_TCPv4_Rx | IFCAP_CSUM_UDPv4_Rx)) != 0)
1756 		cfg |= RE_CPLUSCMD_RXCSUM_ENB;
1757 	if ((sc->sc_quirk & RTKQ_MACSTAT) != 0) {
1758 		cfg |= RE_CPLUSCMD_MACSTAT_DIS;
1759 		cfg |= RE_CPLUSCMD_TXENB;
1760 	} else
1761 		cfg |= RE_CPLUSCMD_RXENB | RE_CPLUSCMD_TXENB;
1762 
1763 	CSR_WRITE_2(sc, RTK_CPLUS_CMD, cfg);
1764 
1765 	/* XXX: from Realtek-supplied Linux driver. Wholly undocumented. */
1766 	if ((sc->sc_quirk & RTKQ_8139CPLUS) == 0)
1767 		CSR_WRITE_2(sc, RTK_IM, 0x0000);
1768 
1769 	DELAY(10000);
1770 
1771 	/*
1772 	 * Init our MAC address.  Even though the chipset
1773 	 * documentation doesn't mention it, we need to enter "Config
1774 	 * register write enable" mode to modify the ID registers.
1775 	 */
1776 	CSR_WRITE_1(sc, RTK_EECMD, RTK_EEMODE_WRITECFG);
1777 	enaddr = CLLADDR(ifp->if_sadl);
1778 	reg = enaddr[0] | (enaddr[1] << 8) |
1779 	    (enaddr[2] << 16) | (enaddr[3] << 24);
1780 	CSR_WRITE_4(sc, RTK_IDR0, reg);
1781 	reg = enaddr[4] | (enaddr[5] << 8);
1782 	CSR_WRITE_4(sc, RTK_IDR4, reg);
1783 	CSR_WRITE_1(sc, RTK_EECMD, RTK_EEMODE_OFF);
1784 
1785 	/*
1786 	 * For C+ mode, initialize the RX descriptors and mbufs.
1787 	 */
1788 	re_rx_list_init(sc);
1789 	re_tx_list_init(sc);
1790 
1791 	/*
1792 	 * Load the addresses of the RX and TX lists into the chip.
1793 	 */
1794 	CSR_WRITE_4(sc, RTK_RXLIST_ADDR_HI,
1795 	    RE_ADDR_HI(sc->re_ldata.re_rx_list_map->dm_segs[0].ds_addr));
1796 	CSR_WRITE_4(sc, RTK_RXLIST_ADDR_LO,
1797 	    RE_ADDR_LO(sc->re_ldata.re_rx_list_map->dm_segs[0].ds_addr));
1798 
1799 	CSR_WRITE_4(sc, RTK_TXLIST_ADDR_HI,
1800 	    RE_ADDR_HI(sc->re_ldata.re_tx_list_map->dm_segs[0].ds_addr));
1801 	CSR_WRITE_4(sc, RTK_TXLIST_ADDR_LO,
1802 	    RE_ADDR_LO(sc->re_ldata.re_tx_list_map->dm_segs[0].ds_addr));
1803 
1804 	/*
1805 	 * Enable transmit and receive.
1806 	 */
1807 	CSR_WRITE_1(sc, RTK_COMMAND, RTK_CMD_TX_ENB | RTK_CMD_RX_ENB);
1808 
1809 	/*
1810 	 * Set the initial TX and RX configuration.
1811 	 */
1812 	if (sc->re_testmode && (sc->sc_quirk & RTKQ_8169NONS) != 0) {
1813 		/* test mode is needed only for old 8169 */
1814 		CSR_WRITE_4(sc, RTK_TXCFG,
1815 		    RE_TXCFG_CONFIG | RTK_LOOPTEST_ON);
1816 	} else
1817 		CSR_WRITE_4(sc, RTK_TXCFG, RE_TXCFG_CONFIG);
1818 
1819 	CSR_WRITE_1(sc, RTK_EARLY_TX_THRESH, 16);
1820 
1821 	CSR_WRITE_4(sc, RTK_RXCFG, RE_RXCFG_CONFIG);
1822 
1823 	/* Set the individual bit to receive frames for this host only. */
1824 	rxcfg = CSR_READ_4(sc, RTK_RXCFG);
1825 	rxcfg |= RTK_RXCFG_RX_INDIV;
1826 
1827 	/* If we want promiscuous mode, set the allframes bit. */
1828 	if (ifp->if_flags & IFF_PROMISC)
1829 		rxcfg |= RTK_RXCFG_RX_ALLPHYS;
1830 	else
1831 		rxcfg &= ~RTK_RXCFG_RX_ALLPHYS;
1832 	CSR_WRITE_4(sc, RTK_RXCFG, rxcfg);
1833 
1834 	/*
1835 	 * Set capture broadcast bit to capture broadcast frames.
1836 	 */
1837 	if (ifp->if_flags & IFF_BROADCAST)
1838 		rxcfg |= RTK_RXCFG_RX_BROAD;
1839 	else
1840 		rxcfg &= ~RTK_RXCFG_RX_BROAD;
1841 	CSR_WRITE_4(sc, RTK_RXCFG, rxcfg);
1842 
1843 	/*
1844 	 * Program the multicast filter, if necessary.
1845 	 */
1846 	rtk_setmulti(sc);
1847 
1848 	/*
1849 	 * Enable interrupts.
1850 	 */
1851 	if (sc->re_testmode)
1852 		CSR_WRITE_2(sc, RTK_IMR, 0);
1853 	else
1854 		CSR_WRITE_2(sc, RTK_IMR, RTK_INTRS_CPLUS);
1855 
1856 	/* Start RX/TX process. */
1857 	CSR_WRITE_4(sc, RTK_MISSEDPKT, 0);
1858 #ifdef notdef
1859 	/* Enable receiver and transmitter. */
1860 	CSR_WRITE_1(sc, RTK_COMMAND, RTK_CMD_TX_ENB | RTK_CMD_RX_ENB);
1861 #endif
1862 
1863 	/*
1864 	 * Initialize the timer interrupt register so that
1865 	 * a timer interrupt will be generated once the timer
1866 	 * reaches a certain number of ticks. The timer is
1867 	 * reloaded on each transmit. This gives us TX interrupt
1868 	 * moderation, which dramatically improves TX frame rate.
1869 	 */
1870 
1871 	if ((sc->sc_quirk & RTKQ_8139CPLUS) != 0)
1872 		CSR_WRITE_4(sc, RTK_TIMERINT, 0x400);
1873 	else {
1874 		CSR_WRITE_4(sc, RTK_TIMERINT_8169, 0x800);
1875 
1876 		/*
1877 		 * For 8169 gigE NICs, set the max allowed RX packet
1878 		 * size so we can receive jumbo frames.
1879 		 */
1880 		CSR_WRITE_2(sc, RTK_MAXRXPKTLEN, 16383);
1881 	}
1882 
1883 	if (sc->re_testmode)
1884 		return 0;
1885 
1886 	CSR_WRITE_1(sc, RTK_CFG1, RTK_CFG1_DRVLOAD);
1887 
1888 	ifp->if_flags |= IFF_RUNNING;
1889 	ifp->if_flags &= ~IFF_OACTIVE;
1890 
1891 	callout_reset(&sc->rtk_tick_ch, hz, re_tick, sc);
1892 
1893  out:
1894 	if (error) {
1895 		ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
1896 		ifp->if_timer = 0;
1897 		printf("%s: interface not running\n",
1898 		    device_xname(sc->sc_dev));
1899 	}
1900 
1901 	return error;
1902 }
1903 
1904 static int
1905 re_ioctl(struct ifnet *ifp, u_long command, void *data)
1906 {
1907 	struct rtk_softc *sc = ifp->if_softc;
1908 	struct ifreq *ifr = data;
1909 	int s, error = 0;
1910 
1911 	s = splnet();
1912 
1913 	switch (command) {
1914 	case SIOCSIFMTU:
1915 		/*
1916 		 * Disable jumbo frames if it's not supported.
1917 		 */
1918 		if ((sc->sc_quirk & RTKQ_NOJUMBO) != 0 &&
1919 		    ifr->ifr_mtu > ETHERMTU) {
1920 			error = EINVAL;
1921 			break;
1922 		}
1923 
1924 		if (ifr->ifr_mtu < ETHERMIN || ifr->ifr_mtu > ETHERMTU_JUMBO)
1925 			error = EINVAL;
1926 		else if ((error = ifioctl_common(ifp, command, data)) ==
1927 		    ENETRESET)
1928 			error = 0;
1929 		break;
1930 	default:
1931 		if ((error = ether_ioctl(ifp, command, data)) != ENETRESET)
1932 			break;
1933 
1934 		error = 0;
1935 
1936 		if (command == SIOCSIFCAP)
1937 			error = (*ifp->if_init)(ifp);
1938 		else if (command != SIOCADDMULTI && command != SIOCDELMULTI)
1939 			;
1940 		else if (ifp->if_flags & IFF_RUNNING)
1941 			rtk_setmulti(sc);
1942 		break;
1943 	}
1944 
1945 	splx(s);
1946 
1947 	return error;
1948 }
1949 
1950 static void
1951 re_watchdog(struct ifnet *ifp)
1952 {
1953 	struct rtk_softc *sc;
1954 	int s;
1955 
1956 	sc = ifp->if_softc;
1957 	s = splnet();
1958 	printf("%s: watchdog timeout\n", device_xname(sc->sc_dev));
1959 	ifp->if_oerrors++;
1960 
1961 	re_txeof(sc);
1962 	re_rxeof(sc);
1963 
1964 	re_init(ifp);
1965 
1966 	splx(s);
1967 }
1968 
1969 /*
1970  * Stop the adapter and free any mbufs allocated to the
1971  * RX and TX lists.
1972  */
1973 static void
1974 re_stop(struct ifnet *ifp, int disable)
1975 {
1976 	int i;
1977 	struct rtk_softc *sc = ifp->if_softc;
1978 
1979 	callout_stop(&sc->rtk_tick_ch);
1980 
1981 	mii_down(&sc->mii);
1982 
1983 	if ((sc->sc_quirk & RTKQ_CMDSTOP) != 0)
1984 		CSR_WRITE_1(sc, RTK_COMMAND, RTK_CMD_STOPREQ | RTK_CMD_TX_ENB |
1985 		    RTK_CMD_RX_ENB);
1986 	else
1987 		CSR_WRITE_1(sc, RTK_COMMAND, 0x00);
1988 	DELAY(1000);
1989 	CSR_WRITE_2(sc, RTK_IMR, 0x0000);
1990 	CSR_WRITE_2(sc, RTK_ISR, 0xFFFF);
1991 
1992 	if (sc->re_head != NULL) {
1993 		m_freem(sc->re_head);
1994 		sc->re_head = sc->re_tail = NULL;
1995 	}
1996 
1997 	/* Free the TX list buffers. */
1998 	for (i = 0; i < RE_TX_QLEN; i++) {
1999 		if (sc->re_ldata.re_txq[i].txq_mbuf != NULL) {
2000 			bus_dmamap_unload(sc->sc_dmat,
2001 			    sc->re_ldata.re_txq[i].txq_dmamap);
2002 			m_freem(sc->re_ldata.re_txq[i].txq_mbuf);
2003 			sc->re_ldata.re_txq[i].txq_mbuf = NULL;
2004 		}
2005 	}
2006 
2007 	/* Free the RX list buffers. */
2008 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
2009 		if (sc->re_ldata.re_rxsoft[i].rxs_mbuf != NULL) {
2010 			bus_dmamap_unload(sc->sc_dmat,
2011 			    sc->re_ldata.re_rxsoft[i].rxs_dmamap);
2012 			m_freem(sc->re_ldata.re_rxsoft[i].rxs_mbuf);
2013 			sc->re_ldata.re_rxsoft[i].rxs_mbuf = NULL;
2014 		}
2015 	}
2016 
2017 	if (disable)
2018 		re_disable(sc);
2019 
2020 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
2021 	ifp->if_timer = 0;
2022 }
2023