xref: /netbsd-src/sys/dev/ic/rtl8169.c (revision 6a493d6bc668897c91594964a732d38505b70cbb)
1 /*	$NetBSD: rtl8169.c,v 1.139 2013/05/10 14:55:08 tsutsui Exp $	*/
2 
3 /*
4  * Copyright (c) 1997, 1998-2003
5  *	Bill Paul <wpaul@windriver.com>.  All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. All advertising materials mentioning features or use of this software
16  *    must display the following acknowledgement:
17  *	This product includes software developed by Bill Paul.
18  * 4. Neither the name of the author nor the names of any co-contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY Bill Paul AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL Bill Paul OR THE VOICES IN HIS HEAD
26  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
27  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
28  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
29  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
30  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
31  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
32  * THE POSSIBILITY OF SUCH DAMAGE.
33  */
34 
35 #include <sys/cdefs.h>
36 __KERNEL_RCSID(0, "$NetBSD: rtl8169.c,v 1.139 2013/05/10 14:55:08 tsutsui Exp $");
37 /* $FreeBSD: /repoman/r/ncvs/src/sys/dev/re/if_re.c,v 1.20 2004/04/11 20:34:08 ru Exp $ */
38 
39 /*
40  * RealTek 8139C+/8169/8169S/8168/8110S 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  * six devices in this family: the RTL8139C+, the RTL8169, the RTL8169S,
51  * RTL8110S, the RTL8168 and the RTL8111.
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 
115 #include <sys/param.h>
116 #include <sys/endian.h>
117 #include <sys/systm.h>
118 #include <sys/sockio.h>
119 #include <sys/mbuf.h>
120 #include <sys/malloc.h>
121 #include <sys/kernel.h>
122 #include <sys/socket.h>
123 #include <sys/device.h>
124 
125 #include <net/if.h>
126 #include <net/if_arp.h>
127 #include <net/if_dl.h>
128 #include <net/if_ether.h>
129 #include <net/if_media.h>
130 #include <net/if_vlanvar.h>
131 
132 #include <netinet/in_systm.h>	/* XXX for IP_MAXPACKET */
133 #include <netinet/in.h>		/* XXX for IP_MAXPACKET */
134 #include <netinet/ip.h>		/* XXX for IP_MAXPACKET */
135 
136 #include <net/bpf.h>
137 #include <sys/rnd.h>
138 
139 #include <sys/bus.h>
140 
141 #include <dev/mii/mii.h>
142 #include <dev/mii/miivar.h>
143 
144 #include <dev/ic/rtl81x9reg.h>
145 #include <dev/ic/rtl81x9var.h>
146 
147 #include <dev/ic/rtl8169var.h>
148 
149 static inline void re_set_bufaddr(struct re_desc *, bus_addr_t);
150 
151 static int re_newbuf(struct rtk_softc *, int, struct mbuf *);
152 static int re_rx_list_init(struct rtk_softc *);
153 static int re_tx_list_init(struct rtk_softc *);
154 static void re_rxeof(struct rtk_softc *);
155 static void re_txeof(struct rtk_softc *);
156 static void re_tick(void *);
157 static void re_start(struct ifnet *);
158 static int re_ioctl(struct ifnet *, u_long, void *);
159 static int re_init(struct ifnet *);
160 static void re_stop(struct ifnet *, int);
161 static void re_watchdog(struct ifnet *);
162 
163 static int re_enable(struct rtk_softc *);
164 static void re_disable(struct rtk_softc *);
165 
166 static int re_gmii_readreg(device_t, int, int);
167 static void re_gmii_writereg(device_t, int, int, int);
168 
169 static int re_miibus_readreg(device_t, int, int);
170 static void re_miibus_writereg(device_t, int, int, int);
171 static void re_miibus_statchg(struct ifnet *);
172 
173 static void re_reset(struct rtk_softc *);
174 
175 static inline void
176 re_set_bufaddr(struct re_desc *d, bus_addr_t addr)
177 {
178 
179 	d->re_bufaddr_lo = htole32((uint32_t)addr);
180 	if (sizeof(bus_addr_t) == sizeof(uint64_t))
181 		d->re_bufaddr_hi = htole32((uint64_t)addr >> 32);
182 	else
183 		d->re_bufaddr_hi = 0;
184 }
185 
186 static int
187 re_gmii_readreg(device_t dev, int phy, int reg)
188 {
189 	struct rtk_softc *sc = device_private(dev);
190 	uint32_t rval;
191 	int i;
192 
193 	if (phy != 7)
194 		return 0;
195 
196 	/* Let the rgephy driver read the GMEDIASTAT register */
197 
198 	if (reg == RTK_GMEDIASTAT) {
199 		rval = CSR_READ_1(sc, RTK_GMEDIASTAT);
200 		return rval;
201 	}
202 
203 	CSR_WRITE_4(sc, RTK_PHYAR, reg << 16);
204 	DELAY(1000);
205 
206 	for (i = 0; i < RTK_TIMEOUT; i++) {
207 		rval = CSR_READ_4(sc, RTK_PHYAR);
208 		if (rval & RTK_PHYAR_BUSY)
209 			break;
210 		DELAY(100);
211 	}
212 
213 	if (i == RTK_TIMEOUT) {
214 		printf("%s: PHY read failed\n", device_xname(sc->sc_dev));
215 		return 0;
216 	}
217 
218 	return rval & RTK_PHYAR_PHYDATA;
219 }
220 
221 static void
222 re_gmii_writereg(device_t dev, int phy, int reg, int data)
223 {
224 	struct rtk_softc *sc = device_private(dev);
225 	uint32_t rval;
226 	int i;
227 
228 	CSR_WRITE_4(sc, RTK_PHYAR, (reg << 16) |
229 	    (data & RTK_PHYAR_PHYDATA) | RTK_PHYAR_BUSY);
230 	DELAY(1000);
231 
232 	for (i = 0; i < RTK_TIMEOUT; i++) {
233 		rval = CSR_READ_4(sc, RTK_PHYAR);
234 		if (!(rval & RTK_PHYAR_BUSY))
235 			break;
236 		DELAY(100);
237 	}
238 
239 	if (i == RTK_TIMEOUT) {
240 		printf("%s: PHY write reg %x <- %x failed\n",
241 		    device_xname(sc->sc_dev), reg, data);
242 	}
243 }
244 
245 static int
246 re_miibus_readreg(device_t dev, int phy, int reg)
247 {
248 	struct rtk_softc *sc = device_private(dev);
249 	uint16_t rval = 0;
250 	uint16_t re8139_reg = 0;
251 	int s;
252 
253 	s = splnet();
254 
255 	if ((sc->sc_quirk & RTKQ_8139CPLUS) == 0) {
256 		rval = re_gmii_readreg(dev, phy, reg);
257 		splx(s);
258 		return rval;
259 	}
260 
261 	/* Pretend the internal PHY is only at address 0 */
262 	if (phy) {
263 		splx(s);
264 		return 0;
265 	}
266 	switch (reg) {
267 	case MII_BMCR:
268 		re8139_reg = RTK_BMCR;
269 		break;
270 	case MII_BMSR:
271 		re8139_reg = RTK_BMSR;
272 		break;
273 	case MII_ANAR:
274 		re8139_reg = RTK_ANAR;
275 		break;
276 	case MII_ANER:
277 		re8139_reg = RTK_ANER;
278 		break;
279 	case MII_ANLPAR:
280 		re8139_reg = RTK_LPAR;
281 		break;
282 	case MII_PHYIDR1:
283 	case MII_PHYIDR2:
284 		splx(s);
285 		return 0;
286 	/*
287 	 * Allow the rlphy driver to read the media status
288 	 * register. If we have a link partner which does not
289 	 * support NWAY, this is the register which will tell
290 	 * us the results of parallel detection.
291 	 */
292 	case RTK_MEDIASTAT:
293 		rval = CSR_READ_1(sc, RTK_MEDIASTAT);
294 		splx(s);
295 		return rval;
296 	default:
297 		printf("%s: bad phy register\n", device_xname(sc->sc_dev));
298 		splx(s);
299 		return 0;
300 	}
301 	rval = CSR_READ_2(sc, re8139_reg);
302 	if ((sc->sc_quirk & RTKQ_8139CPLUS) != 0 && re8139_reg == RTK_BMCR) {
303 		/* 8139C+ has different bit layout. */
304 		rval &= ~(BMCR_LOOP | BMCR_ISO);
305 	}
306 	splx(s);
307 	return rval;
308 }
309 
310 static void
311 re_miibus_writereg(device_t dev, int phy, int reg, int data)
312 {
313 	struct rtk_softc *sc = device_private(dev);
314 	uint16_t re8139_reg = 0;
315 	int s;
316 
317 	s = splnet();
318 
319 	if ((sc->sc_quirk & RTKQ_8139CPLUS) == 0) {
320 		re_gmii_writereg(dev, phy, reg, data);
321 		splx(s);
322 		return;
323 	}
324 
325 	/* Pretend the internal PHY is only at address 0 */
326 	if (phy) {
327 		splx(s);
328 		return;
329 	}
330 	switch (reg) {
331 	case MII_BMCR:
332 		re8139_reg = RTK_BMCR;
333 		if ((sc->sc_quirk & RTKQ_8139CPLUS) != 0) {
334 			/* 8139C+ has different bit layout. */
335 			data &= ~(BMCR_LOOP | BMCR_ISO);
336 		}
337 		break;
338 	case MII_BMSR:
339 		re8139_reg = RTK_BMSR;
340 		break;
341 	case MII_ANAR:
342 		re8139_reg = RTK_ANAR;
343 		break;
344 	case MII_ANER:
345 		re8139_reg = RTK_ANER;
346 		break;
347 	case MII_ANLPAR:
348 		re8139_reg = RTK_LPAR;
349 		break;
350 	case MII_PHYIDR1:
351 	case MII_PHYIDR2:
352 		splx(s);
353 		return;
354 		break;
355 	default:
356 		printf("%s: bad phy register\n", device_xname(sc->sc_dev));
357 		splx(s);
358 		return;
359 	}
360 	CSR_WRITE_2(sc, re8139_reg, data);
361 	splx(s);
362 	return;
363 }
364 
365 static void
366 re_miibus_statchg(struct ifnet *ifp)
367 {
368 
369 	return;
370 }
371 
372 static void
373 re_reset(struct rtk_softc *sc)
374 {
375 	int i;
376 
377 	CSR_WRITE_1(sc, RTK_COMMAND, RTK_CMD_RESET);
378 
379 	for (i = 0; i < RTK_TIMEOUT; i++) {
380 		DELAY(10);
381 		if ((CSR_READ_1(sc, RTK_COMMAND) & RTK_CMD_RESET) == 0)
382 			break;
383 	}
384 	if (i == RTK_TIMEOUT)
385 		printf("%s: reset never completed!\n",
386 		    device_xname(sc->sc_dev));
387 
388 	/*
389 	 * NB: Realtek-supplied FreeBSD driver does this only for MACFG_3,
390 	 *     but also says "Rtl8169s sigle chip detected".
391 	 */
392 	if ((sc->sc_quirk & RTKQ_MACLDPS) != 0)
393 		CSR_WRITE_1(sc, RTK_LDPS, 1);
394 
395 }
396 
397 /*
398  * The following routine is designed to test for a defect on some
399  * 32-bit 8169 cards. Some of these NICs have the REQ64# and ACK64#
400  * lines connected to the bus, however for a 32-bit only card, they
401  * should be pulled high. The result of this defect is that the
402  * NIC will not work right if you plug it into a 64-bit slot: DMA
403  * operations will be done with 64-bit transfers, which will fail
404  * because the 64-bit data lines aren't connected.
405  *
406  * There's no way to work around this (short of talking a soldering
407  * iron to the board), however we can detect it. The method we use
408  * here is to put the NIC into digital loopback mode, set the receiver
409  * to promiscuous mode, and then try to send a frame. We then compare
410  * the frame data we sent to what was received. If the data matches,
411  * then the NIC is working correctly, otherwise we know the user has
412  * a defective NIC which has been mistakenly plugged into a 64-bit PCI
413  * slot. In the latter case, there's no way the NIC can work correctly,
414  * so we print out a message on the console and abort the device attach.
415  */
416 
417 int
418 re_diag(struct rtk_softc *sc)
419 {
420 	struct ifnet *ifp = &sc->ethercom.ec_if;
421 	struct mbuf *m0;
422 	struct ether_header *eh;
423 	struct re_rxsoft *rxs;
424 	struct re_desc *cur_rx;
425 	bus_dmamap_t dmamap;
426 	uint16_t status;
427 	uint32_t rxstat;
428 	int total_len, i, s, error = 0;
429 	static const uint8_t dst[] = { 0x00, 'h', 'e', 'l', 'l', 'o' };
430 	static const uint8_t src[] = { 0x00, 'w', 'o', 'r', 'l', 'd' };
431 
432 	/* Allocate a single mbuf */
433 
434 	MGETHDR(m0, M_DONTWAIT, MT_DATA);
435 	if (m0 == NULL)
436 		return ENOBUFS;
437 
438 	/*
439 	 * Initialize the NIC in test mode. This sets the chip up
440 	 * so that it can send and receive frames, but performs the
441 	 * following special functions:
442 	 * - Puts receiver in promiscuous mode
443 	 * - Enables digital loopback mode
444 	 * - Leaves interrupts turned off
445 	 */
446 
447 	ifp->if_flags |= IFF_PROMISC;
448 	sc->re_testmode = 1;
449 	re_init(ifp);
450 	re_stop(ifp, 0);
451 	DELAY(100000);
452 	re_init(ifp);
453 
454 	/* Put some data in the mbuf */
455 
456 	eh = mtod(m0, struct ether_header *);
457 	memcpy(eh->ether_dhost, (char *)&dst, ETHER_ADDR_LEN);
458 	memcpy(eh->ether_shost, (char *)&src, ETHER_ADDR_LEN);
459 	eh->ether_type = htons(ETHERTYPE_IP);
460 	m0->m_pkthdr.len = m0->m_len = ETHER_MIN_LEN - ETHER_CRC_LEN;
461 
462 	/*
463 	 * Queue the packet, start transmission.
464 	 */
465 
466 	CSR_WRITE_2(sc, RTK_ISR, 0xFFFF);
467 	s = splnet();
468 	IF_ENQUEUE(&ifp->if_snd, m0);
469 	re_start(ifp);
470 	splx(s);
471 	m0 = NULL;
472 
473 	/* Wait for it to propagate through the chip */
474 
475 	DELAY(100000);
476 	for (i = 0; i < RTK_TIMEOUT; i++) {
477 		status = CSR_READ_2(sc, RTK_ISR);
478 		if ((status & (RTK_ISR_TIMEOUT_EXPIRED | RTK_ISR_RX_OK)) ==
479 		    (RTK_ISR_TIMEOUT_EXPIRED | RTK_ISR_RX_OK))
480 			break;
481 		DELAY(10);
482 	}
483 	if (i == RTK_TIMEOUT) {
484 		aprint_error_dev(sc->sc_dev,
485 		    "diagnostic failed, failed to receive packet "
486 		    "in loopback mode\n");
487 		error = EIO;
488 		goto done;
489 	}
490 
491 	/*
492 	 * The packet should have been dumped into the first
493 	 * entry in the RX DMA ring. Grab it from there.
494 	 */
495 
496 	rxs = &sc->re_ldata.re_rxsoft[0];
497 	dmamap = rxs->rxs_dmamap;
498 	bus_dmamap_sync(sc->sc_dmat, dmamap, 0, dmamap->dm_mapsize,
499 	    BUS_DMASYNC_POSTREAD);
500 	bus_dmamap_unload(sc->sc_dmat, dmamap);
501 
502 	m0 = rxs->rxs_mbuf;
503 	rxs->rxs_mbuf = NULL;
504 	eh = mtod(m0, struct ether_header *);
505 
506 	RE_RXDESCSYNC(sc, 0, BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
507 	cur_rx = &sc->re_ldata.re_rx_list[0];
508 	rxstat = le32toh(cur_rx->re_cmdstat);
509 	total_len = rxstat & sc->re_rxlenmask;
510 
511 	if (total_len != ETHER_MIN_LEN) {
512 		aprint_error_dev(sc->sc_dev,
513 		    "diagnostic failed, received short packet\n");
514 		error = EIO;
515 		goto done;
516 	}
517 
518 	/* Test that the received packet data matches what we sent. */
519 
520 	if (memcmp((char *)&eh->ether_dhost, (char *)&dst, ETHER_ADDR_LEN) ||
521 	    memcmp((char *)&eh->ether_shost, (char *)&src, ETHER_ADDR_LEN) ||
522 	    ntohs(eh->ether_type) != ETHERTYPE_IP) {
523 		aprint_error_dev(sc->sc_dev, "WARNING, DMA FAILURE!\n"
524 		    "expected TX data: %s/%s/0x%x\n"
525 		    "received RX data: %s/%s/0x%x\n"
526 		    "You may have a defective 32-bit NIC plugged "
527 		    "into a 64-bit PCI slot.\n"
528 		    "Please re-install the NIC in a 32-bit slot "
529 		    "for proper operation.\n"
530 		    "Read the re(4) man page for more details.\n" ,
531 		    ether_sprintf(dst),  ether_sprintf(src), ETHERTYPE_IP,
532 		    ether_sprintf(eh->ether_dhost),
533 		    ether_sprintf(eh->ether_shost), ntohs(eh->ether_type));
534 		error = EIO;
535 	}
536 
537  done:
538 	/* Turn interface off, release resources */
539 
540 	sc->re_testmode = 0;
541 	ifp->if_flags &= ~IFF_PROMISC;
542 	re_stop(ifp, 0);
543 	if (m0 != NULL)
544 		m_freem(m0);
545 
546 	return error;
547 }
548 
549 
550 /*
551  * Attach the interface. Allocate softc structures, do ifmedia
552  * setup and ethernet/BPF attach.
553  */
554 void
555 re_attach(struct rtk_softc *sc)
556 {
557 	uint8_t eaddr[ETHER_ADDR_LEN];
558 	struct ifnet *ifp;
559 	int error = 0, i;
560 
561 	if ((sc->sc_quirk & RTKQ_8139CPLUS) == 0) {
562 		uint32_t hwrev;
563 
564 		/* Revision of 8169/8169S/8110s in bits 30..26, 23 */
565 		hwrev = CSR_READ_4(sc, RTK_TXCFG) & RTK_TXCFG_HWREV;
566 		switch (hwrev) {
567 		case RTK_HWREV_8169:
568 			sc->sc_quirk |= RTKQ_8169NONS;
569 			break;
570 		case RTK_HWREV_8169S:
571 		case RTK_HWREV_8110S:
572 		case RTK_HWREV_8169_8110SB:
573 		case RTK_HWREV_8169_8110SBL:
574 		case RTK_HWREV_8169_8110SC:
575 			sc->sc_quirk |= RTKQ_MACLDPS;
576 			break;
577 		case RTK_HWREV_8168_SPIN1:
578 		case RTK_HWREV_8168_SPIN2:
579 		case RTK_HWREV_8168_SPIN3:
580 			sc->sc_quirk |= RTKQ_MACSTAT;
581 			break;
582 		case RTK_HWREV_8168C:
583 		case RTK_HWREV_8168C_SPIN2:
584 		case RTK_HWREV_8168CP:
585 		case RTK_HWREV_8168D:
586 		case RTK_HWREV_8168DP:
587 			sc->sc_quirk |= RTKQ_DESCV2 | RTKQ_NOEECMD |
588 			    RTKQ_MACSTAT | RTKQ_CMDSTOP;
589 			/*
590 			 * From FreeBSD driver:
591 			 *
592 			 * These (8168/8111) controllers support jumbo frame
593 			 * but it seems that enabling it requires touching
594 			 * additional magic registers. Depending on MAC
595 			 * revisions some controllers need to disable
596 			 * checksum offload. So disable jumbo frame until
597 			 * I have better idea what it really requires to
598 			 * make it support.
599 			 * RTL8168C/CP : supports up to 6KB jumbo frame.
600 			 * RTL8111C/CP : supports up to 9KB jumbo frame.
601 			 */
602 			sc->sc_quirk |= RTKQ_NOJUMBO;
603 			break;
604 		case RTK_HWREV_8168E:
605 			sc->sc_quirk |= RTKQ_DESCV2 | RTKQ_NOEECMD |
606 			    RTKQ_MACSTAT | RTKQ_CMDSTOP | RTKQ_PHYWAKE_PM |
607 			    RTKQ_NOJUMBO;
608 			break;
609 		case RTK_HWREV_8168E_VL:
610 		case RTK_HWREV_8168F:
611 			sc->sc_quirk |= RTKQ_DESCV2 | RTKQ_NOEECMD |
612 			    RTKQ_MACSTAT | RTKQ_CMDSTOP | RTKQ_NOJUMBO;
613 			break;
614 		case RTK_HWREV_8100E:
615 		case RTK_HWREV_8100E_SPIN2:
616 		case RTK_HWREV_8101E:
617 			sc->sc_quirk |= RTKQ_NOJUMBO;
618 			break;
619 		case RTK_HWREV_8102E:
620 		case RTK_HWREV_8102EL:
621 		case RTK_HWREV_8103E:
622 			sc->sc_quirk |= RTKQ_DESCV2 | RTKQ_NOEECMD |
623 			    RTKQ_MACSTAT | RTKQ_CMDSTOP | RTKQ_NOJUMBO;
624 			break;
625 		default:
626 			aprint_normal_dev(sc->sc_dev,
627 			    "Unknown revision (0x%08x)\n", hwrev);
628 			/* assume the latest features */
629 			sc->sc_quirk |= RTKQ_DESCV2 | RTKQ_NOEECMD;
630 			sc->sc_quirk |= RTKQ_NOJUMBO;
631 		}
632 
633 		/* Set RX length mask */
634 		sc->re_rxlenmask = RE_RDESC_STAT_GFRAGLEN;
635 		sc->re_ldata.re_tx_desc_cnt = RE_TX_DESC_CNT_8169;
636 	} else {
637 		sc->sc_quirk |= RTKQ_NOJUMBO;
638 
639 		/* Set RX length mask */
640 		sc->re_rxlenmask = RE_RDESC_STAT_FRAGLEN;
641 		sc->re_ldata.re_tx_desc_cnt = RE_TX_DESC_CNT_8139;
642 	}
643 
644 	/* Reset the adapter. */
645 	re_reset(sc);
646 
647 	/*
648 	 * RTL81x9 chips automatically read EEPROM to init MAC address,
649 	 * and some NAS override its MAC address per own configuration,
650 	 * so no need to explicitely read EEPROM and set ID registers.
651 	 */
652 #ifdef RE_USE_EECMD
653 	if ((sc->sc_quirk & RTKQ_NOEECMD) != 0) {
654 		/*
655 		 * Get station address from ID registers.
656 		 */
657 		for (i = 0; i < ETHER_ADDR_LEN; i++)
658 			eaddr[i] = CSR_READ_1(sc, RTK_IDR0 + i);
659 	} else {
660 		uint16_t val;
661 		int addr_len;
662 
663 		/*
664 		 * Get station address from the EEPROM.
665 		 */
666 		if (rtk_read_eeprom(sc, RTK_EE_ID, RTK_EEADDR_LEN1) == 0x8129)
667 			addr_len = RTK_EEADDR_LEN1;
668 		else
669 			addr_len = RTK_EEADDR_LEN0;
670 
671 		/*
672 		 * Get station address from the EEPROM.
673 		 */
674 		for (i = 0; i < ETHER_ADDR_LEN / 2; i++) {
675 			val = rtk_read_eeprom(sc, RTK_EE_EADDR0 + i, addr_len);
676 			eaddr[(i * 2) + 0] = val & 0xff;
677 			eaddr[(i * 2) + 1] = val >> 8;
678 		}
679 	}
680 #else
681 	/*
682 	 * Get station address from ID registers.
683 	 */
684 	for (i = 0; i < ETHER_ADDR_LEN; i++)
685 		eaddr[i] = CSR_READ_1(sc, RTK_IDR0 + i);
686 #endif
687 
688 	/* Take PHY out of power down mode. */
689 	if ((sc->sc_quirk & RTKQ_PHYWAKE_PM) != 0)
690 		CSR_WRITE_1(sc, RTK_PMCH, CSR_READ_1(sc, RTK_PMCH) | 0x80);
691 
692 	aprint_normal_dev(sc->sc_dev, "Ethernet address %s\n",
693 	    ether_sprintf(eaddr));
694 
695 	if (sc->re_ldata.re_tx_desc_cnt >
696 	    PAGE_SIZE / sizeof(struct re_desc)) {
697 		sc->re_ldata.re_tx_desc_cnt =
698 		    PAGE_SIZE / sizeof(struct re_desc);
699 	}
700 
701 	aprint_verbose_dev(sc->sc_dev, "using %d tx descriptors\n",
702 	    sc->re_ldata.re_tx_desc_cnt);
703 	KASSERT(RE_NEXT_TX_DESC(sc, RE_TX_DESC_CNT(sc) - 1) == 0);
704 
705 	/* Allocate DMA'able memory for the TX ring */
706 	if ((error = bus_dmamem_alloc(sc->sc_dmat, RE_TX_LIST_SZ(sc),
707 	    RE_RING_ALIGN, 0, &sc->re_ldata.re_tx_listseg, 1,
708 	    &sc->re_ldata.re_tx_listnseg, BUS_DMA_NOWAIT)) != 0) {
709 		aprint_error_dev(sc->sc_dev,
710 		    "can't allocate tx listseg, error = %d\n", error);
711 		goto fail_0;
712 	}
713 
714 	/* Load the map for the TX ring. */
715 	if ((error = bus_dmamem_map(sc->sc_dmat, &sc->re_ldata.re_tx_listseg,
716 	    sc->re_ldata.re_tx_listnseg, RE_TX_LIST_SZ(sc),
717 	    (void **)&sc->re_ldata.re_tx_list,
718 	    BUS_DMA_COHERENT | BUS_DMA_NOWAIT)) != 0) {
719 		aprint_error_dev(sc->sc_dev,
720 		    "can't map tx list, error = %d\n", error);
721 		goto fail_1;
722 	}
723 	memset(sc->re_ldata.re_tx_list, 0, RE_TX_LIST_SZ(sc));
724 
725 	if ((error = bus_dmamap_create(sc->sc_dmat, RE_TX_LIST_SZ(sc), 1,
726 	    RE_TX_LIST_SZ(sc), 0, 0,
727 	    &sc->re_ldata.re_tx_list_map)) != 0) {
728 		aprint_error_dev(sc->sc_dev,
729 		    "can't create tx list map, error = %d\n", error);
730 		goto fail_2;
731 	}
732 
733 
734 	if ((error = bus_dmamap_load(sc->sc_dmat,
735 	    sc->re_ldata.re_tx_list_map, sc->re_ldata.re_tx_list,
736 	    RE_TX_LIST_SZ(sc), NULL, BUS_DMA_NOWAIT)) != 0) {
737 		aprint_error_dev(sc->sc_dev,
738 		    "can't load tx list, error = %d\n", error);
739 		goto fail_3;
740 	}
741 
742 	/* Create DMA maps for TX buffers */
743 	for (i = 0; i < RE_TX_QLEN; i++) {
744 		error = bus_dmamap_create(sc->sc_dmat,
745 		    round_page(IP_MAXPACKET),
746 		    RE_TX_DESC_CNT(sc), RE_TDESC_CMD_FRAGLEN,
747 		    0, 0, &sc->re_ldata.re_txq[i].txq_dmamap);
748 		if (error) {
749 			aprint_error_dev(sc->sc_dev,
750 			    "can't create DMA map for TX\n");
751 			goto fail_4;
752 		}
753 	}
754 
755 	/* Allocate DMA'able memory for the RX ring */
756 	/* XXX see also a comment about RE_RX_DMAMEM_SZ in rtl81x9var.h */
757 	if ((error = bus_dmamem_alloc(sc->sc_dmat,
758 	    RE_RX_DMAMEM_SZ, RE_RING_ALIGN, 0, &sc->re_ldata.re_rx_listseg, 1,
759 	    &sc->re_ldata.re_rx_listnseg, BUS_DMA_NOWAIT)) != 0) {
760 		aprint_error_dev(sc->sc_dev,
761 		    "can't allocate rx listseg, error = %d\n", error);
762 		goto fail_4;
763 	}
764 
765 	/* Load the map for the RX ring. */
766 	if ((error = bus_dmamem_map(sc->sc_dmat, &sc->re_ldata.re_rx_listseg,
767 	    sc->re_ldata.re_rx_listnseg, RE_RX_DMAMEM_SZ,
768 	    (void **)&sc->re_ldata.re_rx_list,
769 	    BUS_DMA_COHERENT | BUS_DMA_NOWAIT)) != 0) {
770 		aprint_error_dev(sc->sc_dev,
771 		    "can't map rx list, error = %d\n", error);
772 		goto fail_5;
773 	}
774 	memset(sc->re_ldata.re_rx_list, 0, RE_RX_DMAMEM_SZ);
775 
776 	if ((error = bus_dmamap_create(sc->sc_dmat,
777 	    RE_RX_DMAMEM_SZ, 1, RE_RX_DMAMEM_SZ, 0, 0,
778 	    &sc->re_ldata.re_rx_list_map)) != 0) {
779 		aprint_error_dev(sc->sc_dev,
780 		    "can't create rx list map, error = %d\n", error);
781 		goto fail_6;
782 	}
783 
784 	if ((error = bus_dmamap_load(sc->sc_dmat,
785 	    sc->re_ldata.re_rx_list_map, sc->re_ldata.re_rx_list,
786 	    RE_RX_DMAMEM_SZ, NULL, BUS_DMA_NOWAIT)) != 0) {
787 		aprint_error_dev(sc->sc_dev,
788 		    "can't load rx list, error = %d\n", error);
789 		goto fail_7;
790 	}
791 
792 	/* Create DMA maps for RX buffers */
793 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
794 		error = bus_dmamap_create(sc->sc_dmat, MCLBYTES, 1, MCLBYTES,
795 		    0, 0, &sc->re_ldata.re_rxsoft[i].rxs_dmamap);
796 		if (error) {
797 			aprint_error_dev(sc->sc_dev,
798 			    "can't create DMA map for RX\n");
799 			goto fail_8;
800 		}
801 	}
802 
803 	/*
804 	 * Record interface as attached. From here, we should not fail.
805 	 */
806 	sc->sc_flags |= RTK_ATTACHED;
807 
808 	ifp = &sc->ethercom.ec_if;
809 	ifp->if_softc = sc;
810 	strlcpy(ifp->if_xname, device_xname(sc->sc_dev), IFNAMSIZ);
811 	ifp->if_mtu = ETHERMTU;
812 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
813 	ifp->if_ioctl = re_ioctl;
814 	sc->ethercom.ec_capabilities |=
815 	    ETHERCAP_VLAN_MTU | ETHERCAP_VLAN_HWTAGGING;
816 	ifp->if_start = re_start;
817 	ifp->if_stop = re_stop;
818 
819 	/*
820 	 * IFCAP_CSUM_IPv4_Tx on re(4) is broken for small packets,
821 	 * so we have a workaround to handle the bug by padding
822 	 * such packets manually.
823 	 */
824 	ifp->if_capabilities |=
825 	    IFCAP_CSUM_IPv4_Tx | IFCAP_CSUM_IPv4_Rx |
826 	    IFCAP_CSUM_TCPv4_Tx | IFCAP_CSUM_TCPv4_Rx |
827 	    IFCAP_CSUM_UDPv4_Tx | IFCAP_CSUM_UDPv4_Rx |
828 	    IFCAP_TSOv4;
829 
830 	/*
831 	 * XXX
832 	 * Still have no idea how to make TSO work on 8168C, 8168CP,
833 	 * 8102E, 8111C and 8111CP.
834 	 */
835 	if ((sc->sc_quirk & RTKQ_DESCV2) != 0)
836 		ifp->if_capabilities &= ~IFCAP_TSOv4;
837 
838 	ifp->if_watchdog = re_watchdog;
839 	ifp->if_init = re_init;
840 	ifp->if_snd.ifq_maxlen = RE_IFQ_MAXLEN;
841 	ifp->if_capenable = ifp->if_capabilities;
842 	IFQ_SET_READY(&ifp->if_snd);
843 
844 	callout_init(&sc->rtk_tick_ch, 0);
845 
846 	/* Do MII setup */
847 	sc->mii.mii_ifp = ifp;
848 	sc->mii.mii_readreg = re_miibus_readreg;
849 	sc->mii.mii_writereg = re_miibus_writereg;
850 	sc->mii.mii_statchg = re_miibus_statchg;
851 	sc->ethercom.ec_mii = &sc->mii;
852 	ifmedia_init(&sc->mii.mii_media, IFM_IMASK, ether_mediachange,
853 	    ether_mediastatus);
854 	mii_attach(sc->sc_dev, &sc->mii, 0xffffffff, MII_PHY_ANY,
855 	    MII_OFFSET_ANY, 0);
856 	ifmedia_set(&sc->mii.mii_media, IFM_ETHER | IFM_AUTO);
857 
858 	/*
859 	 * Call MI attach routine.
860 	 */
861 	if_attach(ifp);
862 	ether_ifattach(ifp, eaddr);
863 
864 	rnd_attach_source(&sc->rnd_source, device_xname(sc->sc_dev),
865 	    RND_TYPE_NET, 0);
866 
867 	if (pmf_device_register(sc->sc_dev, NULL, NULL))
868 		pmf_class_network_register(sc->sc_dev, ifp);
869 	else
870 		aprint_error_dev(sc->sc_dev,
871 		    "couldn't establish power handler\n");
872 
873 	return;
874 
875  fail_8:
876 	/* Destroy DMA maps for RX buffers. */
877 	for (i = 0; i < RE_RX_DESC_CNT; i++)
878 		if (sc->re_ldata.re_rxsoft[i].rxs_dmamap != NULL)
879 			bus_dmamap_destroy(sc->sc_dmat,
880 			    sc->re_ldata.re_rxsoft[i].rxs_dmamap);
881 
882 	/* Free DMA'able memory for the RX ring. */
883 	bus_dmamap_unload(sc->sc_dmat, sc->re_ldata.re_rx_list_map);
884  fail_7:
885 	bus_dmamap_destroy(sc->sc_dmat, sc->re_ldata.re_rx_list_map);
886  fail_6:
887 	bus_dmamem_unmap(sc->sc_dmat,
888 	    (void *)sc->re_ldata.re_rx_list, RE_RX_DMAMEM_SZ);
889  fail_5:
890 	bus_dmamem_free(sc->sc_dmat,
891 	    &sc->re_ldata.re_rx_listseg, sc->re_ldata.re_rx_listnseg);
892 
893  fail_4:
894 	/* Destroy DMA maps for TX buffers. */
895 	for (i = 0; i < RE_TX_QLEN; i++)
896 		if (sc->re_ldata.re_txq[i].txq_dmamap != NULL)
897 			bus_dmamap_destroy(sc->sc_dmat,
898 			    sc->re_ldata.re_txq[i].txq_dmamap);
899 
900 	/* Free DMA'able memory for the TX ring. */
901 	bus_dmamap_unload(sc->sc_dmat, sc->re_ldata.re_tx_list_map);
902  fail_3:
903 	bus_dmamap_destroy(sc->sc_dmat, sc->re_ldata.re_tx_list_map);
904  fail_2:
905 	bus_dmamem_unmap(sc->sc_dmat,
906 	    (void *)sc->re_ldata.re_tx_list, RE_TX_LIST_SZ(sc));
907  fail_1:
908 	bus_dmamem_free(sc->sc_dmat,
909 	    &sc->re_ldata.re_tx_listseg, sc->re_ldata.re_tx_listnseg);
910  fail_0:
911 	return;
912 }
913 
914 
915 /*
916  * re_activate:
917  *     Handle device activation/deactivation requests.
918  */
919 int
920 re_activate(device_t self, enum devact act)
921 {
922 	struct rtk_softc *sc = device_private(self);
923 
924 	switch (act) {
925 	case DVACT_DEACTIVATE:
926 		if_deactivate(&sc->ethercom.ec_if);
927 		return 0;
928 	default:
929 		return EOPNOTSUPP;
930 	}
931 }
932 
933 /*
934  * re_detach:
935  *     Detach a rtk interface.
936  */
937 int
938 re_detach(struct rtk_softc *sc)
939 {
940 	struct ifnet *ifp = &sc->ethercom.ec_if;
941 	int i;
942 
943 	/*
944 	 * Succeed now if there isn't any work to do.
945 	 */
946 	if ((sc->sc_flags & RTK_ATTACHED) == 0)
947 		return 0;
948 
949 	/* Unhook our tick handler. */
950 	callout_stop(&sc->rtk_tick_ch);
951 
952 	/* Detach all PHYs. */
953 	mii_detach(&sc->mii, MII_PHY_ANY, MII_OFFSET_ANY);
954 
955 	/* Delete all remaining media. */
956 	ifmedia_delete_instance(&sc->mii.mii_media, IFM_INST_ANY);
957 
958 	rnd_detach_source(&sc->rnd_source);
959 	ether_ifdetach(ifp);
960 	if_detach(ifp);
961 
962 	/* Destroy DMA maps for RX buffers. */
963 	for (i = 0; i < RE_RX_DESC_CNT; i++)
964 		if (sc->re_ldata.re_rxsoft[i].rxs_dmamap != NULL)
965 			bus_dmamap_destroy(sc->sc_dmat,
966 			    sc->re_ldata.re_rxsoft[i].rxs_dmamap);
967 
968 	/* Free DMA'able memory for the RX ring. */
969 	bus_dmamap_unload(sc->sc_dmat, sc->re_ldata.re_rx_list_map);
970 	bus_dmamap_destroy(sc->sc_dmat, sc->re_ldata.re_rx_list_map);
971 	bus_dmamem_unmap(sc->sc_dmat,
972 	    (void *)sc->re_ldata.re_rx_list, RE_RX_DMAMEM_SZ);
973 	bus_dmamem_free(sc->sc_dmat,
974 	    &sc->re_ldata.re_rx_listseg, sc->re_ldata.re_rx_listnseg);
975 
976 	/* Destroy DMA maps for TX buffers. */
977 	for (i = 0; i < RE_TX_QLEN; i++)
978 		if (sc->re_ldata.re_txq[i].txq_dmamap != NULL)
979 			bus_dmamap_destroy(sc->sc_dmat,
980 			    sc->re_ldata.re_txq[i].txq_dmamap);
981 
982 	/* Free DMA'able memory for the TX ring. */
983 	bus_dmamap_unload(sc->sc_dmat, sc->re_ldata.re_tx_list_map);
984 	bus_dmamap_destroy(sc->sc_dmat, sc->re_ldata.re_tx_list_map);
985 	bus_dmamem_unmap(sc->sc_dmat,
986 	    (void *)sc->re_ldata.re_tx_list, RE_TX_LIST_SZ(sc));
987 	bus_dmamem_free(sc->sc_dmat,
988 	    &sc->re_ldata.re_tx_listseg, sc->re_ldata.re_tx_listnseg);
989 
990 	pmf_device_deregister(sc->sc_dev);
991 
992 	/* we don't want to run again */
993 	sc->sc_flags &= ~RTK_ATTACHED;
994 
995 	return 0;
996 }
997 
998 /*
999  * re_enable:
1000  *     Enable the RTL81X9 chip.
1001  */
1002 static int
1003 re_enable(struct rtk_softc *sc)
1004 {
1005 
1006 	if (RTK_IS_ENABLED(sc) == 0 && sc->sc_enable != NULL) {
1007 		if ((*sc->sc_enable)(sc) != 0) {
1008 			printf("%s: device enable failed\n",
1009 			    device_xname(sc->sc_dev));
1010 			return EIO;
1011 		}
1012 		sc->sc_flags |= RTK_ENABLED;
1013 	}
1014 	return 0;
1015 }
1016 
1017 /*
1018  * re_disable:
1019  *     Disable the RTL81X9 chip.
1020  */
1021 static void
1022 re_disable(struct rtk_softc *sc)
1023 {
1024 
1025 	if (RTK_IS_ENABLED(sc) && sc->sc_disable != NULL) {
1026 		(*sc->sc_disable)(sc);
1027 		sc->sc_flags &= ~RTK_ENABLED;
1028 	}
1029 }
1030 
1031 static int
1032 re_newbuf(struct rtk_softc *sc, int idx, struct mbuf *m)
1033 {
1034 	struct mbuf *n = NULL;
1035 	bus_dmamap_t map;
1036 	struct re_desc *d;
1037 	struct re_rxsoft *rxs;
1038 	uint32_t cmdstat;
1039 	int error;
1040 
1041 	if (m == NULL) {
1042 		MGETHDR(n, M_DONTWAIT, MT_DATA);
1043 		if (n == NULL)
1044 			return ENOBUFS;
1045 
1046 		MCLGET(n, M_DONTWAIT);
1047 		if ((n->m_flags & M_EXT) == 0) {
1048 			m_freem(n);
1049 			return ENOBUFS;
1050 		}
1051 		m = n;
1052 	} else
1053 		m->m_data = m->m_ext.ext_buf;
1054 
1055 	/*
1056 	 * Initialize mbuf length fields and fixup
1057 	 * alignment so that the frame payload is
1058 	 * longword aligned.
1059 	 */
1060 	m->m_len = m->m_pkthdr.len = MCLBYTES - RE_ETHER_ALIGN;
1061 	m->m_data += RE_ETHER_ALIGN;
1062 
1063 	rxs = &sc->re_ldata.re_rxsoft[idx];
1064 	map = rxs->rxs_dmamap;
1065 	error = bus_dmamap_load_mbuf(sc->sc_dmat, map, m,
1066 	    BUS_DMA_READ|BUS_DMA_NOWAIT);
1067 
1068 	if (error)
1069 		goto out;
1070 
1071 	bus_dmamap_sync(sc->sc_dmat, map, 0, map->dm_mapsize,
1072 	    BUS_DMASYNC_PREREAD);
1073 
1074 	d = &sc->re_ldata.re_rx_list[idx];
1075 #ifdef DIAGNOSTIC
1076 	RE_RXDESCSYNC(sc, idx, BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1077 	cmdstat = le32toh(d->re_cmdstat);
1078 	RE_RXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD);
1079 	if (cmdstat & RE_RDESC_STAT_OWN) {
1080 		panic("%s: tried to map busy RX descriptor",
1081 		    device_xname(sc->sc_dev));
1082 	}
1083 #endif
1084 
1085 	rxs->rxs_mbuf = m;
1086 
1087 	d->re_vlanctl = 0;
1088 	cmdstat = map->dm_segs[0].ds_len;
1089 	if (idx == (RE_RX_DESC_CNT - 1))
1090 		cmdstat |= RE_RDESC_CMD_EOR;
1091 	re_set_bufaddr(d, map->dm_segs[0].ds_addr);
1092 	d->re_cmdstat = htole32(cmdstat);
1093 	RE_RXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1094 	cmdstat |= RE_RDESC_CMD_OWN;
1095 	d->re_cmdstat = htole32(cmdstat);
1096 	RE_RXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1097 
1098 	return 0;
1099  out:
1100 	if (n != NULL)
1101 		m_freem(n);
1102 	return ENOMEM;
1103 }
1104 
1105 static int
1106 re_tx_list_init(struct rtk_softc *sc)
1107 {
1108 	int i;
1109 
1110 	memset(sc->re_ldata.re_tx_list, 0, RE_TX_LIST_SZ(sc));
1111 	for (i = 0; i < RE_TX_QLEN; i++) {
1112 		sc->re_ldata.re_txq[i].txq_mbuf = NULL;
1113 	}
1114 
1115 	bus_dmamap_sync(sc->sc_dmat,
1116 	    sc->re_ldata.re_tx_list_map, 0,
1117 	    sc->re_ldata.re_tx_list_map->dm_mapsize,
1118 	    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1119 	sc->re_ldata.re_txq_prodidx = 0;
1120 	sc->re_ldata.re_txq_considx = 0;
1121 	sc->re_ldata.re_txq_free = RE_TX_QLEN;
1122 	sc->re_ldata.re_tx_free = RE_TX_DESC_CNT(sc);
1123 	sc->re_ldata.re_tx_nextfree = 0;
1124 
1125 	return 0;
1126 }
1127 
1128 static int
1129 re_rx_list_init(struct rtk_softc *sc)
1130 {
1131 	int i;
1132 
1133 	memset(sc->re_ldata.re_rx_list, 0, RE_RX_LIST_SZ);
1134 
1135 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
1136 		if (re_newbuf(sc, i, NULL) == ENOBUFS)
1137 			return ENOBUFS;
1138 	}
1139 
1140 	sc->re_ldata.re_rx_prodidx = 0;
1141 	sc->re_head = sc->re_tail = NULL;
1142 
1143 	return 0;
1144 }
1145 
1146 /*
1147  * RX handler for C+ and 8169. For the gigE chips, we support
1148  * the reception of jumbo frames that have been fragmented
1149  * across multiple 2K mbuf cluster buffers.
1150  */
1151 static void
1152 re_rxeof(struct rtk_softc *sc)
1153 {
1154 	struct mbuf *m;
1155 	struct ifnet *ifp;
1156 	int i, total_len;
1157 	struct re_desc *cur_rx;
1158 	struct re_rxsoft *rxs;
1159 	uint32_t rxstat, rxvlan;
1160 
1161 	ifp = &sc->ethercom.ec_if;
1162 
1163 	for (i = sc->re_ldata.re_rx_prodidx;; i = RE_NEXT_RX_DESC(sc, i)) {
1164 		cur_rx = &sc->re_ldata.re_rx_list[i];
1165 		RE_RXDESCSYNC(sc, i,
1166 		    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1167 		rxstat = le32toh(cur_rx->re_cmdstat);
1168 		rxvlan = le32toh(cur_rx->re_vlanctl);
1169 		RE_RXDESCSYNC(sc, i, BUS_DMASYNC_PREREAD);
1170 		if ((rxstat & RE_RDESC_STAT_OWN) != 0) {
1171 			break;
1172 		}
1173 		total_len = rxstat & sc->re_rxlenmask;
1174 		rxs = &sc->re_ldata.re_rxsoft[i];
1175 		m = rxs->rxs_mbuf;
1176 
1177 		/* Invalidate the RX mbuf and unload its map */
1178 
1179 		bus_dmamap_sync(sc->sc_dmat,
1180 		    rxs->rxs_dmamap, 0, rxs->rxs_dmamap->dm_mapsize,
1181 		    BUS_DMASYNC_POSTREAD);
1182 		bus_dmamap_unload(sc->sc_dmat, rxs->rxs_dmamap);
1183 
1184 		if ((rxstat & RE_RDESC_STAT_EOF) == 0) {
1185 			m->m_len = MCLBYTES - RE_ETHER_ALIGN;
1186 			if (sc->re_head == NULL)
1187 				sc->re_head = sc->re_tail = m;
1188 			else {
1189 				m->m_flags &= ~M_PKTHDR;
1190 				sc->re_tail->m_next = m;
1191 				sc->re_tail = m;
1192 			}
1193 			re_newbuf(sc, i, NULL);
1194 			continue;
1195 		}
1196 
1197 		/*
1198 		 * NOTE: for the 8139C+, the frame length field
1199 		 * is always 12 bits in size, but for the gigE chips,
1200 		 * it is 13 bits (since the max RX frame length is 16K).
1201 		 * Unfortunately, all 32 bits in the status word
1202 		 * were already used, so to make room for the extra
1203 		 * length bit, RealTek took out the 'frame alignment
1204 		 * error' bit and shifted the other status bits
1205 		 * over one slot. The OWN, EOR, FS and LS bits are
1206 		 * still in the same places. We have already extracted
1207 		 * the frame length and checked the OWN bit, so rather
1208 		 * than using an alternate bit mapping, we shift the
1209 		 * status bits one space to the right so we can evaluate
1210 		 * them using the 8169 status as though it was in the
1211 		 * same format as that of the 8139C+.
1212 		 */
1213 		if ((sc->sc_quirk & RTKQ_8139CPLUS) == 0)
1214 			rxstat >>= 1;
1215 
1216 		if (__predict_false((rxstat & RE_RDESC_STAT_RXERRSUM) != 0)) {
1217 #ifdef RE_DEBUG
1218 			printf("%s: RX error (rxstat = 0x%08x)",
1219 			    device_xname(sc->sc_dev), rxstat);
1220 			if (rxstat & RE_RDESC_STAT_FRALIGN)
1221 				printf(", frame alignment error");
1222 			if (rxstat & RE_RDESC_STAT_BUFOFLOW)
1223 				printf(", out of buffer space");
1224 			if (rxstat & RE_RDESC_STAT_FIFOOFLOW)
1225 				printf(", FIFO overrun");
1226 			if (rxstat & RE_RDESC_STAT_GIANT)
1227 				printf(", giant packet");
1228 			if (rxstat & RE_RDESC_STAT_RUNT)
1229 				printf(", runt packet");
1230 			if (rxstat & RE_RDESC_STAT_CRCERR)
1231 				printf(", CRC error");
1232 			printf("\n");
1233 #endif
1234 			ifp->if_ierrors++;
1235 			/*
1236 			 * If this is part of a multi-fragment packet,
1237 			 * discard all the pieces.
1238 			 */
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 		/*
1248 		 * If allocating a replacement mbuf fails,
1249 		 * reload the current one.
1250 		 */
1251 
1252 		if (__predict_false(re_newbuf(sc, i, NULL) != 0)) {
1253 			ifp->if_ierrors++;
1254 			if (sc->re_head != NULL) {
1255 				m_freem(sc->re_head);
1256 				sc->re_head = sc->re_tail = NULL;
1257 			}
1258 			re_newbuf(sc, i, m);
1259 			continue;
1260 		}
1261 
1262 		if (sc->re_head != NULL) {
1263 			m->m_len = total_len % (MCLBYTES - RE_ETHER_ALIGN);
1264 			/*
1265 			 * Special case: if there's 4 bytes or less
1266 			 * in this buffer, the mbuf can be discarded:
1267 			 * the last 4 bytes is the CRC, which we don't
1268 			 * care about anyway.
1269 			 */
1270 			if (m->m_len <= ETHER_CRC_LEN) {
1271 				sc->re_tail->m_len -=
1272 				    (ETHER_CRC_LEN - m->m_len);
1273 				m_freem(m);
1274 			} else {
1275 				m->m_len -= ETHER_CRC_LEN;
1276 				m->m_flags &= ~M_PKTHDR;
1277 				sc->re_tail->m_next = m;
1278 			}
1279 			m = sc->re_head;
1280 			sc->re_head = sc->re_tail = NULL;
1281 			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
1282 		} else
1283 			m->m_pkthdr.len = m->m_len =
1284 			    (total_len - ETHER_CRC_LEN);
1285 
1286 		ifp->if_ipackets++;
1287 		m->m_pkthdr.rcvif = ifp;
1288 
1289 		/* Do RX checksumming */
1290 		if ((sc->sc_quirk & RTKQ_DESCV2) == 0) {
1291 			/* Check IP header checksum */
1292 			if ((rxstat & RE_RDESC_STAT_PROTOID) != 0) {
1293 				m->m_pkthdr.csum_flags |= M_CSUM_IPv4;
1294 				if (rxstat & RE_RDESC_STAT_IPSUMBAD)
1295 					m->m_pkthdr.csum_flags |=
1296 					    M_CSUM_IPv4_BAD;
1297 
1298 				/* Check TCP/UDP checksum */
1299 				if (RE_TCPPKT(rxstat)) {
1300 					m->m_pkthdr.csum_flags |= M_CSUM_TCPv4;
1301 					if (rxstat & RE_RDESC_STAT_TCPSUMBAD)
1302 						m->m_pkthdr.csum_flags |=
1303 						    M_CSUM_TCP_UDP_BAD;
1304 				} else if (RE_UDPPKT(rxstat)) {
1305 					m->m_pkthdr.csum_flags |= M_CSUM_UDPv4;
1306 					if (rxstat & RE_RDESC_STAT_UDPSUMBAD)
1307 						m->m_pkthdr.csum_flags |=
1308 						    M_CSUM_TCP_UDP_BAD;
1309 				}
1310 			}
1311 		} else {
1312 			/* Check IPv4 header checksum */
1313 			if ((rxvlan & RE_RDESC_VLANCTL_IPV4) != 0) {
1314 				m->m_pkthdr.csum_flags |= M_CSUM_IPv4;
1315 				if (rxstat & RE_RDESC_STAT_IPSUMBAD)
1316 					m->m_pkthdr.csum_flags |=
1317 					    M_CSUM_IPv4_BAD;
1318 
1319 				/* Check TCPv4/UDPv4 checksum */
1320 				if (RE_TCPPKT(rxstat)) {
1321 					m->m_pkthdr.csum_flags |= M_CSUM_TCPv4;
1322 					if (rxstat & RE_RDESC_STAT_TCPSUMBAD)
1323 						m->m_pkthdr.csum_flags |=
1324 						    M_CSUM_TCP_UDP_BAD;
1325 				} else if (RE_UDPPKT(rxstat)) {
1326 					m->m_pkthdr.csum_flags |= M_CSUM_UDPv4;
1327 					if (rxstat & RE_RDESC_STAT_UDPSUMBAD)
1328 						m->m_pkthdr.csum_flags |=
1329 						    M_CSUM_TCP_UDP_BAD;
1330 				}
1331 			}
1332 			/* XXX Check TCPv6/UDPv6 checksum? */
1333 		}
1334 
1335 		if (rxvlan & RE_RDESC_VLANCTL_TAG) {
1336 			VLAN_INPUT_TAG(ifp, m,
1337 			     bswap16(rxvlan & RE_RDESC_VLANCTL_DATA),
1338 			     continue);
1339 		}
1340 		bpf_mtap(ifp, m);
1341 		(*ifp->if_input)(ifp, m);
1342 	}
1343 
1344 	sc->re_ldata.re_rx_prodidx = i;
1345 }
1346 
1347 static void
1348 re_txeof(struct rtk_softc *sc)
1349 {
1350 	struct ifnet *ifp;
1351 	struct re_txq *txq;
1352 	uint32_t txstat;
1353 	int idx, descidx;
1354 
1355 	ifp = &sc->ethercom.ec_if;
1356 
1357 	for (idx = sc->re_ldata.re_txq_considx;
1358 	    sc->re_ldata.re_txq_free < RE_TX_QLEN;
1359 	    idx = RE_NEXT_TXQ(sc, idx), sc->re_ldata.re_txq_free++) {
1360 		txq = &sc->re_ldata.re_txq[idx];
1361 		KASSERT(txq->txq_mbuf != NULL);
1362 
1363 		descidx = txq->txq_descidx;
1364 		RE_TXDESCSYNC(sc, descidx,
1365 		    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1366 		txstat =
1367 		    le32toh(sc->re_ldata.re_tx_list[descidx].re_cmdstat);
1368 		RE_TXDESCSYNC(sc, descidx, BUS_DMASYNC_PREREAD);
1369 		KASSERT((txstat & RE_TDESC_CMD_EOF) != 0);
1370 		if (txstat & RE_TDESC_CMD_OWN) {
1371 			break;
1372 		}
1373 
1374 		sc->re_ldata.re_tx_free += txq->txq_nsegs;
1375 		KASSERT(sc->re_ldata.re_tx_free <= RE_TX_DESC_CNT(sc));
1376 		bus_dmamap_sync(sc->sc_dmat, txq->txq_dmamap,
1377 		    0, txq->txq_dmamap->dm_mapsize, BUS_DMASYNC_POSTWRITE);
1378 		bus_dmamap_unload(sc->sc_dmat, txq->txq_dmamap);
1379 		m_freem(txq->txq_mbuf);
1380 		txq->txq_mbuf = NULL;
1381 
1382 		if (txstat & (RE_TDESC_STAT_EXCESSCOL | RE_TDESC_STAT_COLCNT))
1383 			ifp->if_collisions++;
1384 		if (txstat & RE_TDESC_STAT_TXERRSUM)
1385 			ifp->if_oerrors++;
1386 		else
1387 			ifp->if_opackets++;
1388 	}
1389 
1390 	sc->re_ldata.re_txq_considx = idx;
1391 
1392 	if (sc->re_ldata.re_txq_free > RE_NTXDESC_RSVD)
1393 		ifp->if_flags &= ~IFF_OACTIVE;
1394 
1395 	/*
1396 	 * If not all descriptors have been released reaped yet,
1397 	 * reload the timer so that we will eventually get another
1398 	 * interrupt that will cause us to re-enter this routine.
1399 	 * This is done in case the transmitter has gone idle.
1400 	 */
1401 	if (sc->re_ldata.re_txq_free < RE_TX_QLEN) {
1402 		CSR_WRITE_4(sc, RTK_TIMERCNT, 1);
1403 		if ((sc->sc_quirk & RTKQ_PCIE) != 0) {
1404 			/*
1405 			 * Some chips will ignore a second TX request
1406 			 * issued while an existing transmission is in
1407 			 * progress. If the transmitter goes idle but
1408 			 * there are still packets waiting to be sent,
1409 			 * we need to restart the channel here to flush
1410 			 * them out. This only seems to be required with
1411 			 * the PCIe devices.
1412 			 */
1413 			CSR_WRITE_1(sc, RTK_GTXSTART, RTK_TXSTART_START);
1414 		}
1415 	} else
1416 		ifp->if_timer = 0;
1417 }
1418 
1419 static void
1420 re_tick(void *arg)
1421 {
1422 	struct rtk_softc *sc = arg;
1423 	int s;
1424 
1425 	/* XXX: just return for 8169S/8110S with rev 2 or newer phy */
1426 	s = splnet();
1427 
1428 	mii_tick(&sc->mii);
1429 	splx(s);
1430 
1431 	callout_reset(&sc->rtk_tick_ch, hz, re_tick, sc);
1432 }
1433 
1434 int
1435 re_intr(void *arg)
1436 {
1437 	struct rtk_softc *sc = arg;
1438 	struct ifnet *ifp;
1439 	uint16_t status;
1440 	int handled = 0;
1441 
1442 	if (!device_has_power(sc->sc_dev))
1443 		return 0;
1444 
1445 	ifp = &sc->ethercom.ec_if;
1446 
1447 	if ((ifp->if_flags & IFF_UP) == 0)
1448 		return 0;
1449 
1450 	for (;;) {
1451 
1452 		status = CSR_READ_2(sc, RTK_ISR);
1453 		/* If the card has gone away the read returns 0xffff. */
1454 		if (status == 0xffff)
1455 			break;
1456 		if (status) {
1457 			handled = 1;
1458 			CSR_WRITE_2(sc, RTK_ISR, status);
1459 		}
1460 
1461 		if ((status & RTK_INTRS_CPLUS) == 0)
1462 			break;
1463 
1464 		if (status & (RTK_ISR_RX_OK | RTK_ISR_RX_ERR))
1465 			re_rxeof(sc);
1466 
1467 		if (status & (RTK_ISR_TIMEOUT_EXPIRED | RTK_ISR_TX_ERR |
1468 		    RTK_ISR_TX_DESC_UNAVAIL))
1469 			re_txeof(sc);
1470 
1471 		if (status & RTK_ISR_SYSTEM_ERR) {
1472 			re_init(ifp);
1473 		}
1474 
1475 		if (status & RTK_ISR_LINKCHG) {
1476 			callout_stop(&sc->rtk_tick_ch);
1477 			re_tick(sc);
1478 		}
1479 	}
1480 
1481 	if (handled && !IFQ_IS_EMPTY(&ifp->if_snd))
1482 		re_start(ifp);
1483 
1484 	rnd_add_uint32(&sc->rnd_source, status);
1485 
1486 	return handled;
1487 }
1488 
1489 
1490 
1491 /*
1492  * Main transmit routine for C+ and gigE NICs.
1493  */
1494 
1495 static void
1496 re_start(struct ifnet *ifp)
1497 {
1498 	struct rtk_softc *sc;
1499 	struct mbuf *m;
1500 	bus_dmamap_t map;
1501 	struct re_txq *txq;
1502 	struct re_desc *d;
1503 	struct m_tag *mtag;
1504 	uint32_t cmdstat, re_flags, vlanctl;
1505 	int ofree, idx, error, nsegs, seg;
1506 	int startdesc, curdesc, lastdesc;
1507 	bool pad;
1508 
1509 	sc = ifp->if_softc;
1510 	ofree = sc->re_ldata.re_txq_free;
1511 
1512 	for (idx = sc->re_ldata.re_txq_prodidx;; idx = RE_NEXT_TXQ(sc, idx)) {
1513 
1514 		IFQ_POLL(&ifp->if_snd, m);
1515 		if (m == NULL)
1516 			break;
1517 
1518 		if (sc->re_ldata.re_txq_free == 0 ||
1519 		    sc->re_ldata.re_tx_free == 0) {
1520 			/* no more free slots left */
1521 			ifp->if_flags |= IFF_OACTIVE;
1522 			break;
1523 		}
1524 
1525 		/*
1526 		 * Set up checksum offload. Note: checksum offload bits must
1527 		 * appear in all descriptors of a multi-descriptor transmit
1528 		 * attempt. (This is according to testing done with an 8169
1529 		 * chip. I'm not sure if this is a requirement or a bug.)
1530 		 */
1531 
1532 		vlanctl = 0;
1533 		if ((m->m_pkthdr.csum_flags & M_CSUM_TSOv4) != 0) {
1534 			uint32_t segsz = m->m_pkthdr.segsz;
1535 
1536 			re_flags = RE_TDESC_CMD_LGSEND |
1537 			    (segsz << RE_TDESC_CMD_MSSVAL_SHIFT);
1538 		} else {
1539 			/*
1540 			 * set RE_TDESC_CMD_IPCSUM if any checksum offloading
1541 			 * is requested.  otherwise, RE_TDESC_CMD_TCPCSUM/
1542 			 * RE_TDESC_CMD_UDPCSUM doesn't make effects.
1543 			 */
1544 			re_flags = 0;
1545 			if ((m->m_pkthdr.csum_flags &
1546 			    (M_CSUM_IPv4 | M_CSUM_TCPv4 | M_CSUM_UDPv4))
1547 			    != 0) {
1548 				if ((sc->sc_quirk & RTKQ_DESCV2) == 0) {
1549 					re_flags |= RE_TDESC_CMD_IPCSUM;
1550 					if (m->m_pkthdr.csum_flags &
1551 					    M_CSUM_TCPv4) {
1552 						re_flags |=
1553 						    RE_TDESC_CMD_TCPCSUM;
1554 					} else if (m->m_pkthdr.csum_flags &
1555 					    M_CSUM_UDPv4) {
1556 						re_flags |=
1557 						    RE_TDESC_CMD_UDPCSUM;
1558 					}
1559 				} else {
1560 					vlanctl |= RE_TDESC_VLANCTL_IPCSUM;
1561 					if (m->m_pkthdr.csum_flags &
1562 					    M_CSUM_TCPv4) {
1563 						vlanctl |=
1564 						    RE_TDESC_VLANCTL_TCPCSUM;
1565 					} else if (m->m_pkthdr.csum_flags &
1566 					    M_CSUM_UDPv4) {
1567 						vlanctl |=
1568 						    RE_TDESC_VLANCTL_UDPCSUM;
1569 					}
1570 				}
1571 			}
1572 		}
1573 
1574 		txq = &sc->re_ldata.re_txq[idx];
1575 		map = txq->txq_dmamap;
1576 		error = bus_dmamap_load_mbuf(sc->sc_dmat, map, m,
1577 		    BUS_DMA_WRITE|BUS_DMA_NOWAIT);
1578 
1579 		if (__predict_false(error)) {
1580 			/* XXX try to defrag if EFBIG? */
1581 			printf("%s: can't map mbuf (error %d)\n",
1582 			    device_xname(sc->sc_dev), error);
1583 
1584 			IFQ_DEQUEUE(&ifp->if_snd, m);
1585 			m_freem(m);
1586 			ifp->if_oerrors++;
1587 			continue;
1588 		}
1589 
1590 		nsegs = map->dm_nsegs;
1591 		pad = false;
1592 		if (__predict_false(m->m_pkthdr.len <= RE_IP4CSUMTX_PADLEN &&
1593 		    (re_flags & RE_TDESC_CMD_IPCSUM) != 0 &&
1594 		    (sc->sc_quirk & RTKQ_DESCV2) == 0)) {
1595 			pad = true;
1596 			nsegs++;
1597 		}
1598 
1599 		if (nsegs > sc->re_ldata.re_tx_free) {
1600 			/*
1601 			 * Not enough free descriptors to transmit this packet.
1602 			 */
1603 			ifp->if_flags |= IFF_OACTIVE;
1604 			bus_dmamap_unload(sc->sc_dmat, map);
1605 			break;
1606 		}
1607 
1608 		IFQ_DEQUEUE(&ifp->if_snd, m);
1609 
1610 		/*
1611 		 * Make sure that the caches are synchronized before we
1612 		 * ask the chip to start DMA for the packet data.
1613 		 */
1614 		bus_dmamap_sync(sc->sc_dmat, map, 0, map->dm_mapsize,
1615 		    BUS_DMASYNC_PREWRITE);
1616 
1617 		/*
1618 		 * Set up hardware VLAN tagging. Note: vlan tag info must
1619 		 * appear in all descriptors of a multi-descriptor
1620 		 * transmission attempt.
1621 		 */
1622 		if ((mtag = VLAN_OUTPUT_TAG(&sc->ethercom, m)) != NULL)
1623 			vlanctl |= bswap16(VLAN_TAG_VALUE(mtag)) |
1624 			    RE_TDESC_VLANCTL_TAG;
1625 
1626 		/*
1627 		 * Map the segment array into descriptors.
1628 		 * Note that we set the start-of-frame and
1629 		 * end-of-frame markers for either TX or RX,
1630 		 * but they really only have meaning in the TX case.
1631 		 * (In the RX case, it's the chip that tells us
1632 		 *  where packets begin and end.)
1633 		 * We also keep track of the end of the ring
1634 		 * and set the end-of-ring bits as needed,
1635 		 * and we set the ownership bits in all except
1636 		 * the very first descriptor. (The caller will
1637 		 * set this descriptor later when it start
1638 		 * transmission or reception.)
1639 		 */
1640 		curdesc = startdesc = sc->re_ldata.re_tx_nextfree;
1641 		lastdesc = -1;
1642 		for (seg = 0; seg < map->dm_nsegs;
1643 		    seg++, curdesc = RE_NEXT_TX_DESC(sc, curdesc)) {
1644 			d = &sc->re_ldata.re_tx_list[curdesc];
1645 #ifdef DIAGNOSTIC
1646 			RE_TXDESCSYNC(sc, curdesc,
1647 			    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1648 			cmdstat = le32toh(d->re_cmdstat);
1649 			RE_TXDESCSYNC(sc, curdesc, BUS_DMASYNC_PREREAD);
1650 			if (cmdstat & RE_TDESC_STAT_OWN) {
1651 				panic("%s: tried to map busy TX descriptor",
1652 				    device_xname(sc->sc_dev));
1653 			}
1654 #endif
1655 
1656 			d->re_vlanctl = htole32(vlanctl);
1657 			re_set_bufaddr(d, map->dm_segs[seg].ds_addr);
1658 			cmdstat = re_flags | map->dm_segs[seg].ds_len;
1659 			if (seg == 0)
1660 				cmdstat |= RE_TDESC_CMD_SOF;
1661 			else
1662 				cmdstat |= RE_TDESC_CMD_OWN;
1663 			if (curdesc == (RE_TX_DESC_CNT(sc) - 1))
1664 				cmdstat |= RE_TDESC_CMD_EOR;
1665 			if (seg == nsegs - 1) {
1666 				cmdstat |= RE_TDESC_CMD_EOF;
1667 				lastdesc = curdesc;
1668 			}
1669 			d->re_cmdstat = htole32(cmdstat);
1670 			RE_TXDESCSYNC(sc, curdesc,
1671 			    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1672 		}
1673 		if (__predict_false(pad)) {
1674 			d = &sc->re_ldata.re_tx_list[curdesc];
1675 			d->re_vlanctl = htole32(vlanctl);
1676 			re_set_bufaddr(d, RE_TXPADDADDR(sc));
1677 			cmdstat = re_flags |
1678 			    RE_TDESC_CMD_OWN | RE_TDESC_CMD_EOF |
1679 			    (RE_IP4CSUMTX_PADLEN + 1 - m->m_pkthdr.len);
1680 			if (curdesc == (RE_TX_DESC_CNT(sc) - 1))
1681 				cmdstat |= RE_TDESC_CMD_EOR;
1682 			d->re_cmdstat = htole32(cmdstat);
1683 			RE_TXDESCSYNC(sc, curdesc,
1684 			    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1685 			lastdesc = curdesc;
1686 			curdesc = RE_NEXT_TX_DESC(sc, curdesc);
1687 		}
1688 		KASSERT(lastdesc != -1);
1689 
1690 		/* Transfer ownership of packet to the chip. */
1691 
1692 		sc->re_ldata.re_tx_list[startdesc].re_cmdstat |=
1693 		    htole32(RE_TDESC_CMD_OWN);
1694 		RE_TXDESCSYNC(sc, startdesc,
1695 		    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1696 
1697 		/* update info of TX queue and descriptors */
1698 		txq->txq_mbuf = m;
1699 		txq->txq_descidx = lastdesc;
1700 		txq->txq_nsegs = nsegs;
1701 
1702 		sc->re_ldata.re_txq_free--;
1703 		sc->re_ldata.re_tx_free -= nsegs;
1704 		sc->re_ldata.re_tx_nextfree = curdesc;
1705 
1706 		/*
1707 		 * If there's a BPF listener, bounce a copy of this frame
1708 		 * to him.
1709 		 */
1710 		bpf_mtap(ifp, m);
1711 	}
1712 
1713 	if (sc->re_ldata.re_txq_free < ofree) {
1714 		/*
1715 		 * TX packets are enqueued.
1716 		 */
1717 		sc->re_ldata.re_txq_prodidx = idx;
1718 
1719 		/*
1720 		 * Start the transmitter to poll.
1721 		 *
1722 		 * RealTek put the TX poll request register in a different
1723 		 * location on the 8169 gigE chip. I don't know why.
1724 		 */
1725 		if ((sc->sc_quirk & RTKQ_8139CPLUS) != 0)
1726 			CSR_WRITE_1(sc, RTK_TXSTART, RTK_TXSTART_START);
1727 		else
1728 			CSR_WRITE_1(sc, RTK_GTXSTART, RTK_TXSTART_START);
1729 
1730 		/*
1731 		 * Use the countdown timer for interrupt moderation.
1732 		 * 'TX done' interrupts are disabled. Instead, we reset the
1733 		 * countdown timer, which will begin counting until it hits
1734 		 * the value in the TIMERINT register, and then trigger an
1735 		 * interrupt. Each time we write to the TIMERCNT register,
1736 		 * the timer count is reset to 0.
1737 		 */
1738 		CSR_WRITE_4(sc, RTK_TIMERCNT, 1);
1739 
1740 		/*
1741 		 * Set a timeout in case the chip goes out to lunch.
1742 		 */
1743 		ifp->if_timer = 5;
1744 	}
1745 }
1746 
1747 static int
1748 re_init(struct ifnet *ifp)
1749 {
1750 	struct rtk_softc *sc = ifp->if_softc;
1751 	uint32_t rxcfg = 0;
1752 	uint16_t cfg;
1753 	int error;
1754 #ifdef RE_USE_EECMD
1755 	const uint8_t *enaddr;
1756 	uint32_t reg;
1757 #endif
1758 
1759 	if ((error = re_enable(sc)) != 0)
1760 		goto out;
1761 
1762 	/*
1763 	 * Cancel pending I/O and free all RX/TX buffers.
1764 	 */
1765 	re_stop(ifp, 0);
1766 
1767 	re_reset(sc);
1768 
1769 	/*
1770 	 * Enable C+ RX and TX mode, as well as VLAN stripping and
1771 	 * RX checksum offload. We must configure the C+ register
1772 	 * before all others.
1773 	 */
1774 	cfg = RE_CPLUSCMD_PCI_MRW;
1775 
1776 	/*
1777 	 * XXX: For old 8169 set bit 14.
1778 	 *      For 8169S/8110S and above, do not set bit 14.
1779 	 */
1780 	if ((sc->sc_quirk & RTKQ_8169NONS) != 0)
1781 		cfg |= (0x1 << 14);
1782 
1783 	if ((sc->ethercom.ec_capenable & ETHERCAP_VLAN_HWTAGGING) != 0)
1784 		cfg |= RE_CPLUSCMD_VLANSTRIP;
1785 	if ((ifp->if_capenable & (IFCAP_CSUM_IPv4_Rx |
1786 	     IFCAP_CSUM_TCPv4_Rx | IFCAP_CSUM_UDPv4_Rx)) != 0)
1787 		cfg |= RE_CPLUSCMD_RXCSUM_ENB;
1788 	if ((sc->sc_quirk & RTKQ_MACSTAT) != 0) {
1789 		cfg |= RE_CPLUSCMD_MACSTAT_DIS;
1790 		cfg |= RE_CPLUSCMD_TXENB;
1791 	} else
1792 		cfg |= RE_CPLUSCMD_RXENB | RE_CPLUSCMD_TXENB;
1793 
1794 	CSR_WRITE_2(sc, RTK_CPLUS_CMD, cfg);
1795 
1796 	/* XXX: from Realtek-supplied Linux driver. Wholly undocumented. */
1797 	if ((sc->sc_quirk & RTKQ_8139CPLUS) == 0)
1798 		CSR_WRITE_2(sc, RTK_IM, 0x0000);
1799 
1800 	DELAY(10000);
1801 
1802 #ifdef RE_USE_EECMD
1803 	/*
1804 	 * Init our MAC address.  Even though the chipset
1805 	 * documentation doesn't mention it, we need to enter "Config
1806 	 * register write enable" mode to modify the ID registers.
1807 	 */
1808 	CSR_WRITE_1(sc, RTK_EECMD, RTK_EEMODE_WRITECFG);
1809 	enaddr = CLLADDR(ifp->if_sadl);
1810 	reg = enaddr[0] | (enaddr[1] << 8) |
1811 	    (enaddr[2] << 16) | (enaddr[3] << 24);
1812 	CSR_WRITE_4(sc, RTK_IDR0, reg);
1813 	reg = enaddr[4] | (enaddr[5] << 8);
1814 	CSR_WRITE_4(sc, RTK_IDR4, reg);
1815 	CSR_WRITE_1(sc, RTK_EECMD, RTK_EEMODE_OFF);
1816 #endif
1817 
1818 	/*
1819 	 * For C+ mode, initialize the RX descriptors and mbufs.
1820 	 */
1821 	re_rx_list_init(sc);
1822 	re_tx_list_init(sc);
1823 
1824 	/*
1825 	 * Load the addresses of the RX and TX lists into the chip.
1826 	 */
1827 	CSR_WRITE_4(sc, RTK_RXLIST_ADDR_HI,
1828 	    RE_ADDR_HI(sc->re_ldata.re_rx_list_map->dm_segs[0].ds_addr));
1829 	CSR_WRITE_4(sc, RTK_RXLIST_ADDR_LO,
1830 	    RE_ADDR_LO(sc->re_ldata.re_rx_list_map->dm_segs[0].ds_addr));
1831 
1832 	CSR_WRITE_4(sc, RTK_TXLIST_ADDR_HI,
1833 	    RE_ADDR_HI(sc->re_ldata.re_tx_list_map->dm_segs[0].ds_addr));
1834 	CSR_WRITE_4(sc, RTK_TXLIST_ADDR_LO,
1835 	    RE_ADDR_LO(sc->re_ldata.re_tx_list_map->dm_segs[0].ds_addr));
1836 
1837 	/*
1838 	 * Enable transmit and receive.
1839 	 */
1840 	CSR_WRITE_1(sc, RTK_COMMAND, RTK_CMD_TX_ENB | RTK_CMD_RX_ENB);
1841 
1842 	/*
1843 	 * Set the initial TX and RX configuration.
1844 	 */
1845 	if (sc->re_testmode && (sc->sc_quirk & RTKQ_8169NONS) != 0) {
1846 		/* test mode is needed only for old 8169 */
1847 		CSR_WRITE_4(sc, RTK_TXCFG,
1848 		    RE_TXCFG_CONFIG | RTK_LOOPTEST_ON);
1849 	} else
1850 		CSR_WRITE_4(sc, RTK_TXCFG, RE_TXCFG_CONFIG);
1851 
1852 	CSR_WRITE_1(sc, RTK_EARLY_TX_THRESH, 16);
1853 
1854 	CSR_WRITE_4(sc, RTK_RXCFG, RE_RXCFG_CONFIG);
1855 
1856 	/* Set the individual bit to receive frames for this host only. */
1857 	rxcfg = CSR_READ_4(sc, RTK_RXCFG);
1858 	rxcfg |= RTK_RXCFG_RX_INDIV;
1859 
1860 	/* If we want promiscuous mode, set the allframes bit. */
1861 	if (ifp->if_flags & IFF_PROMISC)
1862 		rxcfg |= RTK_RXCFG_RX_ALLPHYS;
1863 	else
1864 		rxcfg &= ~RTK_RXCFG_RX_ALLPHYS;
1865 	CSR_WRITE_4(sc, RTK_RXCFG, rxcfg);
1866 
1867 	/*
1868 	 * Set capture broadcast bit to capture broadcast frames.
1869 	 */
1870 	if (ifp->if_flags & IFF_BROADCAST)
1871 		rxcfg |= RTK_RXCFG_RX_BROAD;
1872 	else
1873 		rxcfg &= ~RTK_RXCFG_RX_BROAD;
1874 	CSR_WRITE_4(sc, RTK_RXCFG, rxcfg);
1875 
1876 	/*
1877 	 * Program the multicast filter, if necessary.
1878 	 */
1879 	rtk_setmulti(sc);
1880 
1881 	/*
1882 	 * Enable interrupts.
1883 	 */
1884 	if (sc->re_testmode)
1885 		CSR_WRITE_2(sc, RTK_IMR, 0);
1886 	else
1887 		CSR_WRITE_2(sc, RTK_IMR, RTK_INTRS_CPLUS);
1888 
1889 	/* Start RX/TX process. */
1890 	CSR_WRITE_4(sc, RTK_MISSEDPKT, 0);
1891 #ifdef notdef
1892 	/* Enable receiver and transmitter. */
1893 	CSR_WRITE_1(sc, RTK_COMMAND, RTK_CMD_TX_ENB | RTK_CMD_RX_ENB);
1894 #endif
1895 
1896 	/*
1897 	 * Initialize the timer interrupt register so that
1898 	 * a timer interrupt will be generated once the timer
1899 	 * reaches a certain number of ticks. The timer is
1900 	 * reloaded on each transmit. This gives us TX interrupt
1901 	 * moderation, which dramatically improves TX frame rate.
1902 	 */
1903 
1904 	if ((sc->sc_quirk & RTKQ_8139CPLUS) != 0)
1905 		CSR_WRITE_4(sc, RTK_TIMERINT, 0x400);
1906 	else {
1907 		CSR_WRITE_4(sc, RTK_TIMERINT_8169, 0x800);
1908 
1909 		/*
1910 		 * For 8169 gigE NICs, set the max allowed RX packet
1911 		 * size so we can receive jumbo frames.
1912 		 */
1913 		CSR_WRITE_2(sc, RTK_MAXRXPKTLEN, 16383);
1914 	}
1915 
1916 	if (sc->re_testmode)
1917 		return 0;
1918 
1919 	CSR_WRITE_1(sc, RTK_CFG1, RTK_CFG1_DRVLOAD);
1920 
1921 	ifp->if_flags |= IFF_RUNNING;
1922 	ifp->if_flags &= ~IFF_OACTIVE;
1923 
1924 	callout_reset(&sc->rtk_tick_ch, hz, re_tick, sc);
1925 
1926  out:
1927 	if (error) {
1928 		ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
1929 		ifp->if_timer = 0;
1930 		printf("%s: interface not running\n",
1931 		    device_xname(sc->sc_dev));
1932 	}
1933 
1934 	return error;
1935 }
1936 
1937 static int
1938 re_ioctl(struct ifnet *ifp, u_long command, void *data)
1939 {
1940 	struct rtk_softc *sc = ifp->if_softc;
1941 	struct ifreq *ifr = data;
1942 	int s, error = 0;
1943 
1944 	s = splnet();
1945 
1946 	switch (command) {
1947 	case SIOCSIFMTU:
1948 		/*
1949 		 * Disable jumbo frames if it's not supported.
1950 		 */
1951 		if ((sc->sc_quirk & RTKQ_NOJUMBO) != 0 &&
1952 		    ifr->ifr_mtu > ETHERMTU) {
1953 			error = EINVAL;
1954 			break;
1955 		}
1956 
1957 		if (ifr->ifr_mtu < ETHERMIN || ifr->ifr_mtu > ETHERMTU_JUMBO)
1958 			error = EINVAL;
1959 		else if ((error = ifioctl_common(ifp, command, data)) ==
1960 		    ENETRESET)
1961 			error = 0;
1962 		break;
1963 	default:
1964 		if ((error = ether_ioctl(ifp, command, data)) != ENETRESET)
1965 			break;
1966 
1967 		error = 0;
1968 
1969 		if (command == SIOCSIFCAP)
1970 			error = (*ifp->if_init)(ifp);
1971 		else if (command != SIOCADDMULTI && command != SIOCDELMULTI)
1972 			;
1973 		else if (ifp->if_flags & IFF_RUNNING)
1974 			rtk_setmulti(sc);
1975 		break;
1976 	}
1977 
1978 	splx(s);
1979 
1980 	return error;
1981 }
1982 
1983 static void
1984 re_watchdog(struct ifnet *ifp)
1985 {
1986 	struct rtk_softc *sc;
1987 	int s;
1988 
1989 	sc = ifp->if_softc;
1990 	s = splnet();
1991 	printf("%s: watchdog timeout\n", device_xname(sc->sc_dev));
1992 	ifp->if_oerrors++;
1993 
1994 	re_txeof(sc);
1995 	re_rxeof(sc);
1996 
1997 	re_init(ifp);
1998 
1999 	splx(s);
2000 }
2001 
2002 /*
2003  * Stop the adapter and free any mbufs allocated to the
2004  * RX and TX lists.
2005  */
2006 static void
2007 re_stop(struct ifnet *ifp, int disable)
2008 {
2009 	int i;
2010 	struct rtk_softc *sc = ifp->if_softc;
2011 
2012 	callout_stop(&sc->rtk_tick_ch);
2013 
2014 	mii_down(&sc->mii);
2015 
2016 	if ((sc->sc_quirk & RTKQ_CMDSTOP) != 0)
2017 		CSR_WRITE_1(sc, RTK_COMMAND, RTK_CMD_STOPREQ | RTK_CMD_TX_ENB |
2018 		    RTK_CMD_RX_ENB);
2019 	else
2020 		CSR_WRITE_1(sc, RTK_COMMAND, 0x00);
2021 	DELAY(1000);
2022 	CSR_WRITE_2(sc, RTK_IMR, 0x0000);
2023 	CSR_WRITE_2(sc, RTK_ISR, 0xFFFF);
2024 
2025 	if (sc->re_head != NULL) {
2026 		m_freem(sc->re_head);
2027 		sc->re_head = sc->re_tail = NULL;
2028 	}
2029 
2030 	/* Free the TX list buffers. */
2031 	for (i = 0; i < RE_TX_QLEN; i++) {
2032 		if (sc->re_ldata.re_txq[i].txq_mbuf != NULL) {
2033 			bus_dmamap_unload(sc->sc_dmat,
2034 			    sc->re_ldata.re_txq[i].txq_dmamap);
2035 			m_freem(sc->re_ldata.re_txq[i].txq_mbuf);
2036 			sc->re_ldata.re_txq[i].txq_mbuf = NULL;
2037 		}
2038 	}
2039 
2040 	/* Free the RX list buffers. */
2041 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
2042 		if (sc->re_ldata.re_rxsoft[i].rxs_mbuf != NULL) {
2043 			bus_dmamap_unload(sc->sc_dmat,
2044 			    sc->re_ldata.re_rxsoft[i].rxs_dmamap);
2045 			m_freem(sc->re_ldata.re_rxsoft[i].rxs_mbuf);
2046 			sc->re_ldata.re_rxsoft[i].rxs_mbuf = NULL;
2047 		}
2048 	}
2049 
2050 	if (disable)
2051 		re_disable(sc);
2052 
2053 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
2054 	ifp->if_timer = 0;
2055 }
2056