xref: /freebsd/sys/dev/re/if_re.c (revision c0020399a650364d0134f79f3fa319f84064372d)
1 /*-
2  * Copyright (c) 1997, 1998-2003
3  *	Bill Paul <wpaul@windriver.com>.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. All advertising materials mentioning features or use of this software
14  *    must display the following acknowledgement:
15  *	This product includes software developed by Bill Paul.
16  * 4. Neither the name of the author nor the names of any co-contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY Bill Paul AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL Bill Paul OR THE VOICES IN HIS HEAD
24  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
30  * THE POSSIBILITY OF SUCH DAMAGE.
31  */
32 
33 #include <sys/cdefs.h>
34 __FBSDID("$FreeBSD$");
35 
36 /*
37  * RealTek 8139C+/8169/8169S/8110S/8168/8111/8101E PCI NIC driver
38  *
39  * Written by Bill Paul <wpaul@windriver.com>
40  * Senior Networking Software Engineer
41  * Wind River Systems
42  */
43 
44 /*
45  * This driver is designed to support RealTek's next generation of
46  * 10/100 and 10/100/1000 PCI ethernet controllers. There are currently
47  * seven devices in this family: the RTL8139C+, the RTL8169, the RTL8169S,
48  * RTL8110S, the RTL8168, the RTL8111 and the RTL8101E.
49  *
50  * The 8139C+ is a 10/100 ethernet chip. It is backwards compatible
51  * with the older 8139 family, however it also supports a special
52  * C+ mode of operation that provides several new performance enhancing
53  * features. These include:
54  *
55  *	o Descriptor based DMA mechanism. Each descriptor represents
56  *	  a single packet fragment. Data buffers may be aligned on
57  *	  any byte boundary.
58  *
59  *	o 64-bit DMA
60  *
61  *	o TCP/IP checksum offload for both RX and TX
62  *
63  *	o High and normal priority transmit DMA rings
64  *
65  *	o VLAN tag insertion and extraction
66  *
67  *	o TCP large send (segmentation offload)
68  *
69  * Like the 8139, the 8139C+ also has a built-in 10/100 PHY. The C+
70  * programming API is fairly straightforward. The RX filtering, EEPROM
71  * access and PHY access is the same as it is on the older 8139 series
72  * chips.
73  *
74  * The 8169 is a 64-bit 10/100/1000 gigabit ethernet MAC. It has almost the
75  * same programming API and feature set as the 8139C+ with the following
76  * differences and additions:
77  *
78  *	o 1000Mbps mode
79  *
80  *	o Jumbo frames
81  *
82  *	o GMII and TBI ports/registers for interfacing with copper
83  *	  or fiber PHYs
84  *
85  *	o RX and TX DMA rings can have up to 1024 descriptors
86  *	  (the 8139C+ allows a maximum of 64)
87  *
88  *	o Slight differences in register layout from the 8139C+
89  *
90  * The TX start and timer interrupt registers are at different locations
91  * on the 8169 than they are on the 8139C+. Also, the status word in the
92  * RX descriptor has a slightly different bit layout. The 8169 does not
93  * have a built-in PHY. Most reference boards use a Marvell 88E1000 'Alaska'
94  * copper gigE PHY.
95  *
96  * The 8169S/8110S 10/100/1000 devices have built-in copper gigE PHYs
97  * (the 'S' stands for 'single-chip'). These devices have the same
98  * programming API as the older 8169, but also have some vendor-specific
99  * registers for the on-board PHY. The 8110S is a LAN-on-motherboard
100  * part designed to be pin-compatible with the RealTek 8100 10/100 chip.
101  *
102  * This driver takes advantage of the RX and TX checksum offload and
103  * VLAN tag insertion/extraction features. It also implements TX
104  * interrupt moderation using the timer interrupt registers, which
105  * significantly reduces TX interrupt load. There is also support
106  * for jumbo frames, however the 8169/8169S/8110S can not transmit
107  * jumbo frames larger than 7440, so the max MTU possible with this
108  * driver is 7422 bytes.
109  */
110 
111 #ifdef HAVE_KERNEL_OPTION_HEADERS
112 #include "opt_device_polling.h"
113 #endif
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/module.h>
122 #include <sys/kernel.h>
123 #include <sys/socket.h>
124 #include <sys/lock.h>
125 #include <sys/mutex.h>
126 #include <sys/taskqueue.h>
127 
128 #include <net/if.h>
129 #include <net/if_arp.h>
130 #include <net/ethernet.h>
131 #include <net/if_dl.h>
132 #include <net/if_media.h>
133 #include <net/if_types.h>
134 #include <net/if_vlan_var.h>
135 
136 #include <net/bpf.h>
137 
138 #include <machine/bus.h>
139 #include <machine/resource.h>
140 #include <sys/bus.h>
141 #include <sys/rman.h>
142 
143 #include <dev/mii/mii.h>
144 #include <dev/mii/miivar.h>
145 
146 #include <dev/pci/pcireg.h>
147 #include <dev/pci/pcivar.h>
148 
149 #include <pci/if_rlreg.h>
150 
151 MODULE_DEPEND(re, pci, 1, 1, 1);
152 MODULE_DEPEND(re, ether, 1, 1, 1);
153 MODULE_DEPEND(re, miibus, 1, 1, 1);
154 
155 /* "device miibus" required.  See GENERIC if you get errors here. */
156 #include "miibus_if.h"
157 
158 /* Tunables. */
159 static int msi_disable = 0;
160 TUNABLE_INT("hw.re.msi_disable", &msi_disable);
161 static int prefer_iomap = 0;
162 TUNABLE_INT("hw.re.prefer_iomap", &prefer_iomap);
163 
164 #define RE_CSUM_FEATURES    (CSUM_IP | CSUM_TCP | CSUM_UDP)
165 
166 /*
167  * Various supported device vendors/types and their names.
168  */
169 static struct rl_type re_devs[] = {
170 	{ DLINK_VENDORID, DLINK_DEVICEID_528T, 0,
171 	    "D-Link DGE-528(T) Gigabit Ethernet Adapter" },
172 	{ RT_VENDORID, RT_DEVICEID_8139, 0,
173 	    "RealTek 8139C+ 10/100BaseTX" },
174 	{ RT_VENDORID, RT_DEVICEID_8101E, 0,
175 	    "RealTek 8101E/8102E/8102EL PCIe 10/100baseTX" },
176 	{ RT_VENDORID, RT_DEVICEID_8168, 0,
177 	    "RealTek 8168/8168B/8168C/8168CP/8168D/8111B/8111C/8111CP PCIe "
178 	    "Gigabit Ethernet" },
179 	{ RT_VENDORID, RT_DEVICEID_8169, 0,
180 	    "RealTek 8169/8169S/8169SB(L)/8110S/8110SB(L) Gigabit Ethernet" },
181 	{ RT_VENDORID, RT_DEVICEID_8169SC, 0,
182 	    "RealTek 8169SC/8110SC Single-chip Gigabit Ethernet" },
183 	{ COREGA_VENDORID, COREGA_DEVICEID_CGLAPCIGT, 0,
184 	    "Corega CG-LAPCIGT (RTL8169S) Gigabit Ethernet" },
185 	{ LINKSYS_VENDORID, LINKSYS_DEVICEID_EG1032, 0,
186 	    "Linksys EG1032 (RTL8169S) Gigabit Ethernet" },
187 	{ USR_VENDORID, USR_DEVICEID_997902, 0,
188 	    "US Robotics 997902 (RTL8169S) Gigabit Ethernet" }
189 };
190 
191 static struct rl_hwrev re_hwrevs[] = {
192 	{ RL_HWREV_8139, RL_8139,  "" },
193 	{ RL_HWREV_8139A, RL_8139, "A" },
194 	{ RL_HWREV_8139AG, RL_8139, "A-G" },
195 	{ RL_HWREV_8139B, RL_8139, "B" },
196 	{ RL_HWREV_8130, RL_8139, "8130" },
197 	{ RL_HWREV_8139C, RL_8139, "C" },
198 	{ RL_HWREV_8139D, RL_8139, "8139D/8100B/8100C" },
199 	{ RL_HWREV_8139CPLUS, RL_8139CPLUS, "C+"},
200 	{ RL_HWREV_8168_SPIN1, RL_8169, "8168"},
201 	{ RL_HWREV_8169, RL_8169, "8169"},
202 	{ RL_HWREV_8169S, RL_8169, "8169S"},
203 	{ RL_HWREV_8110S, RL_8169, "8110S"},
204 	{ RL_HWREV_8169_8110SB, RL_8169, "8169SB/8110SB"},
205 	{ RL_HWREV_8169_8110SC, RL_8169, "8169SC/8110SC"},
206 	{ RL_HWREV_8169_8110SBL, RL_8169, "8169SBL/8110SBL"},
207 	{ RL_HWREV_8169_8110SCE, RL_8169, "8169SC/8110SC"},
208 	{ RL_HWREV_8100, RL_8139, "8100"},
209 	{ RL_HWREV_8101, RL_8139, "8101"},
210 	{ RL_HWREV_8100E, RL_8169, "8100E"},
211 	{ RL_HWREV_8101E, RL_8169, "8101E"},
212 	{ RL_HWREV_8102E, RL_8169, "8102E"},
213 	{ RL_HWREV_8102EL, RL_8169, "8102EL"},
214 	{ RL_HWREV_8168_SPIN2, RL_8169, "8168"},
215 	{ RL_HWREV_8168_SPIN3, RL_8169, "8168"},
216 	{ RL_HWREV_8168C, RL_8169, "8168C/8111C"},
217 	{ RL_HWREV_8168C_SPIN2, RL_8169, "8168C/8111C"},
218 	{ RL_HWREV_8168CP, RL_8169, "8168CP/8111CP"},
219 	{ RL_HWREV_8168D, RL_8169, "8168D"},
220 	{ 0, 0, NULL }
221 };
222 
223 static int re_probe		(device_t);
224 static int re_attach		(device_t);
225 static int re_detach		(device_t);
226 
227 static int re_encap		(struct rl_softc *, struct mbuf **);
228 
229 static void re_dma_map_addr	(void *, bus_dma_segment_t *, int, int);
230 static int re_allocmem		(device_t, struct rl_softc *);
231 static __inline void re_discard_rxbuf
232 				(struct rl_softc *, int);
233 static int re_newbuf		(struct rl_softc *, int);
234 static int re_rx_list_init	(struct rl_softc *);
235 static int re_tx_list_init	(struct rl_softc *);
236 #ifdef RE_FIXUP_RX
237 static __inline void re_fixup_rx
238 				(struct mbuf *);
239 #endif
240 static int re_rxeof		(struct rl_softc *);
241 static void re_txeof		(struct rl_softc *);
242 #ifdef DEVICE_POLLING
243 static void re_poll		(struct ifnet *, enum poll_cmd, int);
244 static void re_poll_locked	(struct ifnet *, enum poll_cmd, int);
245 #endif
246 static int re_intr		(void *);
247 static void re_tick		(void *);
248 static void re_tx_task		(void *, int);
249 static void re_int_task		(void *, int);
250 static void re_start		(struct ifnet *);
251 static int re_ioctl		(struct ifnet *, u_long, caddr_t);
252 static void re_init		(void *);
253 static void re_init_locked	(struct rl_softc *);
254 static void re_stop		(struct rl_softc *);
255 static void re_watchdog		(struct rl_softc *);
256 static int re_suspend		(device_t);
257 static int re_resume		(device_t);
258 static int re_shutdown		(device_t);
259 static int re_ifmedia_upd	(struct ifnet *);
260 static void re_ifmedia_sts	(struct ifnet *, struct ifmediareq *);
261 
262 static void re_eeprom_putbyte	(struct rl_softc *, int);
263 static void re_eeprom_getword	(struct rl_softc *, int, u_int16_t *);
264 static void re_read_eeprom	(struct rl_softc *, caddr_t, int, int);
265 static int re_gmii_readreg	(device_t, int, int);
266 static int re_gmii_writereg	(device_t, int, int, int);
267 
268 static int re_miibus_readreg	(device_t, int, int);
269 static int re_miibus_writereg	(device_t, int, int, int);
270 static void re_miibus_statchg	(device_t);
271 
272 static void re_set_rxmode		(struct rl_softc *);
273 static void re_reset		(struct rl_softc *);
274 static void re_setwol		(struct rl_softc *);
275 static void re_clrwol		(struct rl_softc *);
276 
277 #ifdef RE_DIAG
278 static int re_diag		(struct rl_softc *);
279 #endif
280 
281 static device_method_t re_methods[] = {
282 	/* Device interface */
283 	DEVMETHOD(device_probe,		re_probe),
284 	DEVMETHOD(device_attach,	re_attach),
285 	DEVMETHOD(device_detach,	re_detach),
286 	DEVMETHOD(device_suspend,	re_suspend),
287 	DEVMETHOD(device_resume,	re_resume),
288 	DEVMETHOD(device_shutdown,	re_shutdown),
289 
290 	/* bus interface */
291 	DEVMETHOD(bus_print_child,	bus_generic_print_child),
292 	DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
293 
294 	/* MII interface */
295 	DEVMETHOD(miibus_readreg,	re_miibus_readreg),
296 	DEVMETHOD(miibus_writereg,	re_miibus_writereg),
297 	DEVMETHOD(miibus_statchg,	re_miibus_statchg),
298 
299 	{ 0, 0 }
300 };
301 
302 static driver_t re_driver = {
303 	"re",
304 	re_methods,
305 	sizeof(struct rl_softc)
306 };
307 
308 static devclass_t re_devclass;
309 
310 DRIVER_MODULE(re, pci, re_driver, re_devclass, 0, 0);
311 DRIVER_MODULE(miibus, re, miibus_driver, miibus_devclass, 0, 0);
312 
313 #define EE_SET(x)					\
314 	CSR_WRITE_1(sc, RL_EECMD,			\
315 		CSR_READ_1(sc, RL_EECMD) | x)
316 
317 #define EE_CLR(x)					\
318 	CSR_WRITE_1(sc, RL_EECMD,			\
319 		CSR_READ_1(sc, RL_EECMD) & ~x)
320 
321 /*
322  * Send a read command and address to the EEPROM, check for ACK.
323  */
324 static void
325 re_eeprom_putbyte(struct rl_softc *sc, int addr)
326 {
327 	int			d, i;
328 
329 	d = addr | (RL_9346_READ << sc->rl_eewidth);
330 
331 	/*
332 	 * Feed in each bit and strobe the clock.
333 	 */
334 
335 	for (i = 1 << (sc->rl_eewidth + 3); i; i >>= 1) {
336 		if (d & i) {
337 			EE_SET(RL_EE_DATAIN);
338 		} else {
339 			EE_CLR(RL_EE_DATAIN);
340 		}
341 		DELAY(100);
342 		EE_SET(RL_EE_CLK);
343 		DELAY(150);
344 		EE_CLR(RL_EE_CLK);
345 		DELAY(100);
346 	}
347 }
348 
349 /*
350  * Read a word of data stored in the EEPROM at address 'addr.'
351  */
352 static void
353 re_eeprom_getword(struct rl_softc *sc, int addr, u_int16_t *dest)
354 {
355 	int			i;
356 	u_int16_t		word = 0;
357 
358 	/*
359 	 * Send address of word we want to read.
360 	 */
361 	re_eeprom_putbyte(sc, addr);
362 
363 	/*
364 	 * Start reading bits from EEPROM.
365 	 */
366 	for (i = 0x8000; i; i >>= 1) {
367 		EE_SET(RL_EE_CLK);
368 		DELAY(100);
369 		if (CSR_READ_1(sc, RL_EECMD) & RL_EE_DATAOUT)
370 			word |= i;
371 		EE_CLR(RL_EE_CLK);
372 		DELAY(100);
373 	}
374 
375 	*dest = word;
376 }
377 
378 /*
379  * Read a sequence of words from the EEPROM.
380  */
381 static void
382 re_read_eeprom(struct rl_softc *sc, caddr_t dest, int off, int cnt)
383 {
384 	int			i;
385 	u_int16_t		word = 0, *ptr;
386 
387 	CSR_SETBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM);
388 
389         DELAY(100);
390 
391 	for (i = 0; i < cnt; i++) {
392 		CSR_SETBIT_1(sc, RL_EECMD, RL_EE_SEL);
393 		re_eeprom_getword(sc, off + i, &word);
394 		CSR_CLRBIT_1(sc, RL_EECMD, RL_EE_SEL);
395 		ptr = (u_int16_t *)(dest + (i * 2));
396                 *ptr = word;
397 	}
398 
399 	CSR_CLRBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM);
400 }
401 
402 static int
403 re_gmii_readreg(device_t dev, int phy, int reg)
404 {
405 	struct rl_softc		*sc;
406 	u_int32_t		rval;
407 	int			i;
408 
409 	if (phy != 1)
410 		return (0);
411 
412 	sc = device_get_softc(dev);
413 
414 	/* Let the rgephy driver read the GMEDIASTAT register */
415 
416 	if (reg == RL_GMEDIASTAT) {
417 		rval = CSR_READ_1(sc, RL_GMEDIASTAT);
418 		return (rval);
419 	}
420 
421 	CSR_WRITE_4(sc, RL_PHYAR, reg << 16);
422 	DELAY(1000);
423 
424 	for (i = 0; i < RL_PHY_TIMEOUT; i++) {
425 		rval = CSR_READ_4(sc, RL_PHYAR);
426 		if (rval & RL_PHYAR_BUSY)
427 			break;
428 		DELAY(100);
429 	}
430 
431 	if (i == RL_PHY_TIMEOUT) {
432 		device_printf(sc->rl_dev, "PHY read failed\n");
433 		return (0);
434 	}
435 
436 	return (rval & RL_PHYAR_PHYDATA);
437 }
438 
439 static int
440 re_gmii_writereg(device_t dev, int phy, int reg, int data)
441 {
442 	struct rl_softc		*sc;
443 	u_int32_t		rval;
444 	int			i;
445 
446 	sc = device_get_softc(dev);
447 
448 	CSR_WRITE_4(sc, RL_PHYAR, (reg << 16) |
449 	    (data & RL_PHYAR_PHYDATA) | RL_PHYAR_BUSY);
450 	DELAY(1000);
451 
452 	for (i = 0; i < RL_PHY_TIMEOUT; i++) {
453 		rval = CSR_READ_4(sc, RL_PHYAR);
454 		if (!(rval & RL_PHYAR_BUSY))
455 			break;
456 		DELAY(100);
457 	}
458 
459 	if (i == RL_PHY_TIMEOUT) {
460 		device_printf(sc->rl_dev, "PHY write failed\n");
461 		return (0);
462 	}
463 
464 	return (0);
465 }
466 
467 static int
468 re_miibus_readreg(device_t dev, int phy, int reg)
469 {
470 	struct rl_softc		*sc;
471 	u_int16_t		rval = 0;
472 	u_int16_t		re8139_reg = 0;
473 
474 	sc = device_get_softc(dev);
475 
476 	if (sc->rl_type == RL_8169) {
477 		rval = re_gmii_readreg(dev, phy, reg);
478 		return (rval);
479 	}
480 
481 	/* Pretend the internal PHY is only at address 0 */
482 	if (phy) {
483 		return (0);
484 	}
485 	switch (reg) {
486 	case MII_BMCR:
487 		re8139_reg = RL_BMCR;
488 		break;
489 	case MII_BMSR:
490 		re8139_reg = RL_BMSR;
491 		break;
492 	case MII_ANAR:
493 		re8139_reg = RL_ANAR;
494 		break;
495 	case MII_ANER:
496 		re8139_reg = RL_ANER;
497 		break;
498 	case MII_ANLPAR:
499 		re8139_reg = RL_LPAR;
500 		break;
501 	case MII_PHYIDR1:
502 	case MII_PHYIDR2:
503 		return (0);
504 	/*
505 	 * Allow the rlphy driver to read the media status
506 	 * register. If we have a link partner which does not
507 	 * support NWAY, this is the register which will tell
508 	 * us the results of parallel detection.
509 	 */
510 	case RL_MEDIASTAT:
511 		rval = CSR_READ_1(sc, RL_MEDIASTAT);
512 		return (rval);
513 	default:
514 		device_printf(sc->rl_dev, "bad phy register\n");
515 		return (0);
516 	}
517 	rval = CSR_READ_2(sc, re8139_reg);
518 	if (sc->rl_type == RL_8139CPLUS && re8139_reg == RL_BMCR) {
519 		/* 8139C+ has different bit layout. */
520 		rval &= ~(BMCR_LOOP | BMCR_ISO);
521 	}
522 	return (rval);
523 }
524 
525 static int
526 re_miibus_writereg(device_t dev, int phy, int reg, int data)
527 {
528 	struct rl_softc		*sc;
529 	u_int16_t		re8139_reg = 0;
530 	int			rval = 0;
531 
532 	sc = device_get_softc(dev);
533 
534 	if (sc->rl_type == RL_8169) {
535 		rval = re_gmii_writereg(dev, phy, reg, data);
536 		return (rval);
537 	}
538 
539 	/* Pretend the internal PHY is only at address 0 */
540 	if (phy)
541 		return (0);
542 
543 	switch (reg) {
544 	case MII_BMCR:
545 		re8139_reg = RL_BMCR;
546 		if (sc->rl_type == RL_8139CPLUS) {
547 			/* 8139C+ has different bit layout. */
548 			data &= ~(BMCR_LOOP | BMCR_ISO);
549 		}
550 		break;
551 	case MII_BMSR:
552 		re8139_reg = RL_BMSR;
553 		break;
554 	case MII_ANAR:
555 		re8139_reg = RL_ANAR;
556 		break;
557 	case MII_ANER:
558 		re8139_reg = RL_ANER;
559 		break;
560 	case MII_ANLPAR:
561 		re8139_reg = RL_LPAR;
562 		break;
563 	case MII_PHYIDR1:
564 	case MII_PHYIDR2:
565 		return (0);
566 		break;
567 	default:
568 		device_printf(sc->rl_dev, "bad phy register\n");
569 		return (0);
570 	}
571 	CSR_WRITE_2(sc, re8139_reg, data);
572 	return (0);
573 }
574 
575 static void
576 re_miibus_statchg(device_t dev)
577 {
578 	struct rl_softc		*sc;
579 	struct ifnet		*ifp;
580 	struct mii_data		*mii;
581 
582 	sc = device_get_softc(dev);
583 	mii = device_get_softc(sc->rl_miibus);
584 	ifp = sc->rl_ifp;
585 	if (mii == NULL || ifp == NULL ||
586 	    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
587 		return;
588 
589 	sc->rl_flags &= ~RL_FLAG_LINK;
590 	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
591 	    (IFM_ACTIVE | IFM_AVALID)) {
592 		switch (IFM_SUBTYPE(mii->mii_media_active)) {
593 		case IFM_10_T:
594 		case IFM_100_TX:
595 			sc->rl_flags |= RL_FLAG_LINK;
596 			break;
597 		case IFM_1000_T:
598 			if ((sc->rl_flags & RL_FLAG_FASTETHER) != 0)
599 				break;
600 			sc->rl_flags |= RL_FLAG_LINK;
601 			break;
602 		default:
603 			break;
604 		}
605 	}
606 	/*
607 	 * RealTek controllers does not provide any interface to
608 	 * Tx/Rx MACs for resolved speed, duplex and flow-control
609 	 * parameters.
610 	 */
611 }
612 
613 /*
614  * Set the RX configuration and 64-bit multicast hash filter.
615  */
616 static void
617 re_set_rxmode(struct rl_softc *sc)
618 {
619 	struct ifnet		*ifp;
620 	struct ifmultiaddr	*ifma;
621 	uint32_t		hashes[2] = { 0, 0 };
622 	uint32_t		h, rxfilt;
623 
624 	RL_LOCK_ASSERT(sc);
625 
626 	ifp = sc->rl_ifp;
627 
628 	rxfilt = RL_RXCFG_CONFIG | RL_RXCFG_RX_INDIV | RL_RXCFG_RX_BROAD;
629 
630 	if (ifp->if_flags & (IFF_ALLMULTI | IFF_PROMISC)) {
631 		if (ifp->if_flags & IFF_PROMISC)
632 			rxfilt |= RL_RXCFG_RX_ALLPHYS;
633 		/*
634 		 * Unlike other hardwares, we have to explicitly set
635 		 * RL_RXCFG_RX_MULTI to receive multicast frames in
636 		 * promiscuous mode.
637 		 */
638 		rxfilt |= RL_RXCFG_RX_MULTI;
639 		hashes[0] = hashes[1] = 0xffffffff;
640 		goto done;
641 	}
642 
643 	IF_ADDR_LOCK(ifp);
644 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
645 		if (ifma->ifma_addr->sa_family != AF_LINK)
646 			continue;
647 		h = ether_crc32_be(LLADDR((struct sockaddr_dl *)
648 		    ifma->ifma_addr), ETHER_ADDR_LEN) >> 26;
649 		if (h < 32)
650 			hashes[0] |= (1 << h);
651 		else
652 			hashes[1] |= (1 << (h - 32));
653 	}
654 	IF_ADDR_UNLOCK(ifp);
655 
656 	if (hashes[0] != 0 || hashes[1] != 0) {
657 		/*
658 		 * For some unfathomable reason, RealTek decided to
659 		 * reverse the order of the multicast hash registers
660 		 * in the PCI Express parts.  This means we have to
661 		 * write the hash pattern in reverse order for those
662 		 * devices.
663 		 */
664 		if ((sc->rl_flags & RL_FLAG_PCIE) != 0) {
665 			h = bswap32(hashes[0]);
666 			hashes[0] = bswap32(hashes[1]);
667 			hashes[1] = h;
668 		}
669 		rxfilt |= RL_RXCFG_RX_MULTI;
670 	}
671 
672 done:
673 	CSR_WRITE_4(sc, RL_MAR0, hashes[0]);
674 	CSR_WRITE_4(sc, RL_MAR4, hashes[1]);
675 	CSR_WRITE_4(sc, RL_RXCFG, rxfilt);
676 }
677 
678 static void
679 re_reset(struct rl_softc *sc)
680 {
681 	int			i;
682 
683 	RL_LOCK_ASSERT(sc);
684 
685 	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RESET);
686 
687 	for (i = 0; i < RL_TIMEOUT; i++) {
688 		DELAY(10);
689 		if (!(CSR_READ_1(sc, RL_COMMAND) & RL_CMD_RESET))
690 			break;
691 	}
692 	if (i == RL_TIMEOUT)
693 		device_printf(sc->rl_dev, "reset never completed!\n");
694 
695 	if ((sc->rl_flags & RL_FLAG_MACRESET) != 0)
696 		CSR_WRITE_1(sc, 0x82, 1);
697 	if (sc->rl_hwrev == RL_HWREV_8169S)
698 		re_gmii_writereg(sc->rl_dev, 1, 0x0b, 0);
699 }
700 
701 #ifdef RE_DIAG
702 
703 /*
704  * The following routine is designed to test for a defect on some
705  * 32-bit 8169 cards. Some of these NICs have the REQ64# and ACK64#
706  * lines connected to the bus, however for a 32-bit only card, they
707  * should be pulled high. The result of this defect is that the
708  * NIC will not work right if you plug it into a 64-bit slot: DMA
709  * operations will be done with 64-bit transfers, which will fail
710  * because the 64-bit data lines aren't connected.
711  *
712  * There's no way to work around this (short of talking a soldering
713  * iron to the board), however we can detect it. The method we use
714  * here is to put the NIC into digital loopback mode, set the receiver
715  * to promiscuous mode, and then try to send a frame. We then compare
716  * the frame data we sent to what was received. If the data matches,
717  * then the NIC is working correctly, otherwise we know the user has
718  * a defective NIC which has been mistakenly plugged into a 64-bit PCI
719  * slot. In the latter case, there's no way the NIC can work correctly,
720  * so we print out a message on the console and abort the device attach.
721  */
722 
723 static int
724 re_diag(struct rl_softc *sc)
725 {
726 	struct ifnet		*ifp = sc->rl_ifp;
727 	struct mbuf		*m0;
728 	struct ether_header	*eh;
729 	struct rl_desc		*cur_rx;
730 	u_int16_t		status;
731 	u_int32_t		rxstat;
732 	int			total_len, i, error = 0, phyaddr;
733 	u_int8_t		dst[] = { 0x00, 'h', 'e', 'l', 'l', 'o' };
734 	u_int8_t		src[] = { 0x00, 'w', 'o', 'r', 'l', 'd' };
735 
736 	/* Allocate a single mbuf */
737 	MGETHDR(m0, M_DONTWAIT, MT_DATA);
738 	if (m0 == NULL)
739 		return (ENOBUFS);
740 
741 	RL_LOCK(sc);
742 
743 	/*
744 	 * Initialize the NIC in test mode. This sets the chip up
745 	 * so that it can send and receive frames, but performs the
746 	 * following special functions:
747 	 * - Puts receiver in promiscuous mode
748 	 * - Enables digital loopback mode
749 	 * - Leaves interrupts turned off
750 	 */
751 
752 	ifp->if_flags |= IFF_PROMISC;
753 	sc->rl_testmode = 1;
754 	re_init_locked(sc);
755 	sc->rl_flags |= RL_FLAG_LINK;
756 	if (sc->rl_type == RL_8169)
757 		phyaddr = 1;
758 	else
759 		phyaddr = 0;
760 
761 	re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_RESET);
762 	for (i = 0; i < RL_TIMEOUT; i++) {
763 		status = re_miibus_readreg(sc->rl_dev, phyaddr, MII_BMCR);
764 		if (!(status & BMCR_RESET))
765 			break;
766 	}
767 
768 	re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_LOOP);
769 	CSR_WRITE_2(sc, RL_ISR, RL_INTRS);
770 
771 	DELAY(100000);
772 
773 	/* Put some data in the mbuf */
774 
775 	eh = mtod(m0, struct ether_header *);
776 	bcopy ((char *)&dst, eh->ether_dhost, ETHER_ADDR_LEN);
777 	bcopy ((char *)&src, eh->ether_shost, ETHER_ADDR_LEN);
778 	eh->ether_type = htons(ETHERTYPE_IP);
779 	m0->m_pkthdr.len = m0->m_len = ETHER_MIN_LEN - ETHER_CRC_LEN;
780 
781 	/*
782 	 * Queue the packet, start transmission.
783 	 * Note: IF_HANDOFF() ultimately calls re_start() for us.
784 	 */
785 
786 	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
787 	RL_UNLOCK(sc);
788 	/* XXX: re_diag must not be called when in ALTQ mode */
789 	IF_HANDOFF(&ifp->if_snd, m0, ifp);
790 	RL_LOCK(sc);
791 	m0 = NULL;
792 
793 	/* Wait for it to propagate through the chip */
794 
795 	DELAY(100000);
796 	for (i = 0; i < RL_TIMEOUT; i++) {
797 		status = CSR_READ_2(sc, RL_ISR);
798 		CSR_WRITE_2(sc, RL_ISR, status);
799 		if ((status & (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK)) ==
800 		    (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK))
801 			break;
802 		DELAY(10);
803 	}
804 
805 	if (i == RL_TIMEOUT) {
806 		device_printf(sc->rl_dev,
807 		    "diagnostic failed, failed to receive packet in"
808 		    " loopback mode\n");
809 		error = EIO;
810 		goto done;
811 	}
812 
813 	/*
814 	 * The packet should have been dumped into the first
815 	 * entry in the RX DMA ring. Grab it from there.
816 	 */
817 
818 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
819 	    sc->rl_ldata.rl_rx_list_map,
820 	    BUS_DMASYNC_POSTREAD);
821 	bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag,
822 	    sc->rl_ldata.rl_rx_desc[0].rx_dmamap,
823 	    BUS_DMASYNC_POSTREAD);
824 	bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag,
825 	    sc->rl_ldata.rl_rx_desc[0].rx_dmamap);
826 
827 	m0 = sc->rl_ldata.rl_rx_desc[0].rx_m;
828 	sc->rl_ldata.rl_rx_desc[0].rx_m = NULL;
829 	eh = mtod(m0, struct ether_header *);
830 
831 	cur_rx = &sc->rl_ldata.rl_rx_list[0];
832 	total_len = RL_RXBYTES(cur_rx);
833 	rxstat = le32toh(cur_rx->rl_cmdstat);
834 
835 	if (total_len != ETHER_MIN_LEN) {
836 		device_printf(sc->rl_dev,
837 		    "diagnostic failed, received short packet\n");
838 		error = EIO;
839 		goto done;
840 	}
841 
842 	/* Test that the received packet data matches what we sent. */
843 
844 	if (bcmp((char *)&eh->ether_dhost, (char *)&dst, ETHER_ADDR_LEN) ||
845 	    bcmp((char *)&eh->ether_shost, (char *)&src, ETHER_ADDR_LEN) ||
846 	    ntohs(eh->ether_type) != ETHERTYPE_IP) {
847 		device_printf(sc->rl_dev, "WARNING, DMA FAILURE!\n");
848 		device_printf(sc->rl_dev, "expected TX data: %6D/%6D/0x%x\n",
849 		    dst, ":", src, ":", ETHERTYPE_IP);
850 		device_printf(sc->rl_dev, "received RX data: %6D/%6D/0x%x\n",
851 		    eh->ether_dhost, ":",  eh->ether_shost, ":",
852 		    ntohs(eh->ether_type));
853 		device_printf(sc->rl_dev, "You may have a defective 32-bit "
854 		    "NIC plugged into a 64-bit PCI slot.\n");
855 		device_printf(sc->rl_dev, "Please re-install the NIC in a "
856 		    "32-bit slot for proper operation.\n");
857 		device_printf(sc->rl_dev, "Read the re(4) man page for more "
858 		    "details.\n");
859 		error = EIO;
860 	}
861 
862 done:
863 	/* Turn interface off, release resources */
864 
865 	sc->rl_testmode = 0;
866 	sc->rl_flags &= ~RL_FLAG_LINK;
867 	ifp->if_flags &= ~IFF_PROMISC;
868 	re_stop(sc);
869 	if (m0 != NULL)
870 		m_freem(m0);
871 
872 	RL_UNLOCK(sc);
873 
874 	return (error);
875 }
876 
877 #endif
878 
879 /*
880  * Probe for a RealTek 8139C+/8169/8110 chip. Check the PCI vendor and device
881  * IDs against our list and return a device name if we find a match.
882  */
883 static int
884 re_probe(device_t dev)
885 {
886 	struct rl_type		*t;
887 	uint16_t		devid, vendor;
888 	uint16_t		revid, sdevid;
889 	int			i;
890 
891 	vendor = pci_get_vendor(dev);
892 	devid = pci_get_device(dev);
893 	revid = pci_get_revid(dev);
894 	sdevid = pci_get_subdevice(dev);
895 
896 	if (vendor == LINKSYS_VENDORID && devid == LINKSYS_DEVICEID_EG1032) {
897 		if (sdevid != LINKSYS_SUBDEVICE_EG1032_REV3) {
898 			/*
899 			 * Only attach to rev. 3 of the Linksys EG1032 adapter.
900 			 * Rev. 2 is supported by sk(4).
901 			 */
902 			return (ENXIO);
903 		}
904 	}
905 
906 	if (vendor == RT_VENDORID && devid == RT_DEVICEID_8139) {
907 		if (revid != 0x20) {
908 			/* 8139, let rl(4) take care of this device. */
909 			return (ENXIO);
910 		}
911 	}
912 
913 	t = re_devs;
914 	for (i = 0; i < sizeof(re_devs) / sizeof(re_devs[0]); i++, t++) {
915 		if (vendor == t->rl_vid && devid == t->rl_did) {
916 			device_set_desc(dev, t->rl_name);
917 			return (BUS_PROBE_DEFAULT);
918 		}
919 	}
920 
921 	return (ENXIO);
922 }
923 
924 /*
925  * Map a single buffer address.
926  */
927 
928 static void
929 re_dma_map_addr(void *arg, bus_dma_segment_t *segs, int nseg, int error)
930 {
931 	bus_addr_t		*addr;
932 
933 	if (error)
934 		return;
935 
936 	KASSERT(nseg == 1, ("too many DMA segments, %d should be 1", nseg));
937 	addr = arg;
938 	*addr = segs->ds_addr;
939 }
940 
941 static int
942 re_allocmem(device_t dev, struct rl_softc *sc)
943 {
944 	bus_size_t		rx_list_size, tx_list_size;
945 	int			error;
946 	int			i;
947 
948 	rx_list_size = sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc);
949 	tx_list_size = sc->rl_ldata.rl_tx_desc_cnt * sizeof(struct rl_desc);
950 
951 	/*
952 	 * Allocate the parent bus DMA tag appropriate for PCI.
953 	 * In order to use DAC, RL_CPLUSCMD_PCI_DAC bit of RL_CPLUS_CMD
954 	 * register should be set. However some RealTek chips are known
955 	 * to be buggy on DAC handling, therefore disable DAC by limiting
956 	 * DMA address space to 32bit. PCIe variants of RealTek chips
957 	 * may not have the limitation but I took safer path.
958 	 */
959 	error = bus_dma_tag_create(bus_get_dma_tag(dev), 1, 0,
960 	    BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL, NULL,
961 	    BUS_SPACE_MAXSIZE_32BIT, 0, BUS_SPACE_MAXSIZE_32BIT, 0,
962 	    NULL, NULL, &sc->rl_parent_tag);
963 	if (error) {
964 		device_printf(dev, "could not allocate parent DMA tag\n");
965 		return (error);
966 	}
967 
968 	/*
969 	 * Allocate map for TX mbufs.
970 	 */
971 	error = bus_dma_tag_create(sc->rl_parent_tag, 1, 0,
972 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL,
973 	    NULL, MCLBYTES * RL_NTXSEGS, RL_NTXSEGS, 4096, 0,
974 	    NULL, NULL, &sc->rl_ldata.rl_tx_mtag);
975 	if (error) {
976 		device_printf(dev, "could not allocate TX DMA tag\n");
977 		return (error);
978 	}
979 
980 	/*
981 	 * Allocate map for RX mbufs.
982 	 */
983 
984 	error = bus_dma_tag_create(sc->rl_parent_tag, sizeof(uint64_t), 0,
985 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
986 	    MCLBYTES, 1, MCLBYTES, 0, NULL, NULL, &sc->rl_ldata.rl_rx_mtag);
987 	if (error) {
988 		device_printf(dev, "could not allocate RX DMA tag\n");
989 		return (error);
990 	}
991 
992 	/*
993 	 * Allocate map for TX descriptor list.
994 	 */
995 	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
996 	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
997 	    NULL, tx_list_size, 1, tx_list_size, 0,
998 	    NULL, NULL, &sc->rl_ldata.rl_tx_list_tag);
999 	if (error) {
1000 		device_printf(dev, "could not allocate TX DMA ring tag\n");
1001 		return (error);
1002 	}
1003 
1004 	/* Allocate DMA'able memory for the TX ring */
1005 
1006 	error = bus_dmamem_alloc(sc->rl_ldata.rl_tx_list_tag,
1007 	    (void **)&sc->rl_ldata.rl_tx_list,
1008 	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1009 	    &sc->rl_ldata.rl_tx_list_map);
1010 	if (error) {
1011 		device_printf(dev, "could not allocate TX DMA ring\n");
1012 		return (error);
1013 	}
1014 
1015 	/* Load the map for the TX ring. */
1016 
1017 	sc->rl_ldata.rl_tx_list_addr = 0;
1018 	error = bus_dmamap_load(sc->rl_ldata.rl_tx_list_tag,
1019 	     sc->rl_ldata.rl_tx_list_map, sc->rl_ldata.rl_tx_list,
1020 	     tx_list_size, re_dma_map_addr,
1021 	     &sc->rl_ldata.rl_tx_list_addr, BUS_DMA_NOWAIT);
1022 	if (error != 0 || sc->rl_ldata.rl_tx_list_addr == 0) {
1023 		device_printf(dev, "could not load TX DMA ring\n");
1024 		return (ENOMEM);
1025 	}
1026 
1027 	/* Create DMA maps for TX buffers */
1028 
1029 	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
1030 		error = bus_dmamap_create(sc->rl_ldata.rl_tx_mtag, 0,
1031 		    &sc->rl_ldata.rl_tx_desc[i].tx_dmamap);
1032 		if (error) {
1033 			device_printf(dev, "could not create DMA map for TX\n");
1034 			return (error);
1035 		}
1036 	}
1037 
1038 	/*
1039 	 * Allocate map for RX descriptor list.
1040 	 */
1041 	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
1042 	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1043 	    NULL, rx_list_size, 1, rx_list_size, 0,
1044 	    NULL, NULL, &sc->rl_ldata.rl_rx_list_tag);
1045 	if (error) {
1046 		device_printf(dev, "could not create RX DMA ring tag\n");
1047 		return (error);
1048 	}
1049 
1050 	/* Allocate DMA'able memory for the RX ring */
1051 
1052 	error = bus_dmamem_alloc(sc->rl_ldata.rl_rx_list_tag,
1053 	    (void **)&sc->rl_ldata.rl_rx_list,
1054 	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1055 	    &sc->rl_ldata.rl_rx_list_map);
1056 	if (error) {
1057 		device_printf(dev, "could not allocate RX DMA ring\n");
1058 		return (error);
1059 	}
1060 
1061 	/* Load the map for the RX ring. */
1062 
1063 	sc->rl_ldata.rl_rx_list_addr = 0;
1064 	error = bus_dmamap_load(sc->rl_ldata.rl_rx_list_tag,
1065 	     sc->rl_ldata.rl_rx_list_map, sc->rl_ldata.rl_rx_list,
1066 	     rx_list_size, re_dma_map_addr,
1067 	     &sc->rl_ldata.rl_rx_list_addr, BUS_DMA_NOWAIT);
1068 	if (error != 0 || sc->rl_ldata.rl_rx_list_addr == 0) {
1069 		device_printf(dev, "could not load RX DMA ring\n");
1070 		return (ENOMEM);
1071 	}
1072 
1073 	/* Create DMA maps for RX buffers */
1074 
1075 	error = bus_dmamap_create(sc->rl_ldata.rl_rx_mtag, 0,
1076 	    &sc->rl_ldata.rl_rx_sparemap);
1077 	if (error) {
1078 		device_printf(dev, "could not create spare DMA map for RX\n");
1079 		return (error);
1080 	}
1081 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1082 		error = bus_dmamap_create(sc->rl_ldata.rl_rx_mtag, 0,
1083 		    &sc->rl_ldata.rl_rx_desc[i].rx_dmamap);
1084 		if (error) {
1085 			device_printf(dev, "could not create DMA map for RX\n");
1086 			return (error);
1087 		}
1088 	}
1089 
1090 	return (0);
1091 }
1092 
1093 /*
1094  * Attach the interface. Allocate softc structures, do ifmedia
1095  * setup and ethernet/BPF attach.
1096  */
1097 static int
1098 re_attach(device_t dev)
1099 {
1100 	u_char			eaddr[ETHER_ADDR_LEN];
1101 	u_int16_t		as[ETHER_ADDR_LEN / 2];
1102 	struct rl_softc		*sc;
1103 	struct ifnet		*ifp;
1104 	struct rl_hwrev		*hw_rev;
1105 	int			hwrev;
1106 	u_int16_t		devid, re_did = 0;
1107 	int			error = 0, rid, i;
1108 	int			msic, reg;
1109 	uint8_t			cfg;
1110 
1111 	sc = device_get_softc(dev);
1112 	sc->rl_dev = dev;
1113 
1114 	mtx_init(&sc->rl_mtx, device_get_nameunit(dev), MTX_NETWORK_LOCK,
1115 	    MTX_DEF);
1116 	callout_init_mtx(&sc->rl_stat_callout, &sc->rl_mtx, 0);
1117 
1118 	/*
1119 	 * Map control/status registers.
1120 	 */
1121 	pci_enable_busmaster(dev);
1122 
1123 	devid = pci_get_device(dev);
1124 	/*
1125 	 * Prefer memory space register mapping over IO space.
1126 	 * Because RTL8169SC does not seem to work when memory mapping
1127 	 * is used always activate io mapping.
1128 	 */
1129 	if (devid == RT_DEVICEID_8169SC)
1130 		prefer_iomap = 1;
1131 	if (prefer_iomap == 0) {
1132 		sc->rl_res_id = PCIR_BAR(1);
1133 		sc->rl_res_type = SYS_RES_MEMORY;
1134 		/* RTL8168/8101E seems to use different BARs. */
1135 		if (devid == RT_DEVICEID_8168 || devid == RT_DEVICEID_8101E)
1136 			sc->rl_res_id = PCIR_BAR(2);
1137 	} else {
1138 		sc->rl_res_id = PCIR_BAR(0);
1139 		sc->rl_res_type = SYS_RES_IOPORT;
1140 	}
1141 	sc->rl_res = bus_alloc_resource_any(dev, sc->rl_res_type,
1142 	    &sc->rl_res_id, RF_ACTIVE);
1143 	if (sc->rl_res == NULL && prefer_iomap == 0) {
1144 		sc->rl_res_id = PCIR_BAR(0);
1145 		sc->rl_res_type = SYS_RES_IOPORT;
1146 		sc->rl_res = bus_alloc_resource_any(dev, sc->rl_res_type,
1147 		    &sc->rl_res_id, RF_ACTIVE);
1148 	}
1149 	if (sc->rl_res == NULL) {
1150 		device_printf(dev, "couldn't map ports/memory\n");
1151 		error = ENXIO;
1152 		goto fail;
1153 	}
1154 
1155 	sc->rl_btag = rman_get_bustag(sc->rl_res);
1156 	sc->rl_bhandle = rman_get_bushandle(sc->rl_res);
1157 
1158 	msic = 0;
1159 	if (pci_find_extcap(dev, PCIY_EXPRESS, &reg) == 0) {
1160 		sc->rl_flags |= RL_FLAG_PCIE;
1161 		msic = pci_msi_count(dev);
1162 		if (bootverbose)
1163 			device_printf(dev, "MSI count : %d\n", msic);
1164 	}
1165 	if (msic > 0 && msi_disable == 0) {
1166 		msic = 1;
1167 		if (pci_alloc_msi(dev, &msic) == 0) {
1168 			if (msic == RL_MSI_MESSAGES) {
1169 				device_printf(dev, "Using %d MSI messages\n",
1170 				    msic);
1171 				sc->rl_flags |= RL_FLAG_MSI;
1172 				/* Explicitly set MSI enable bit. */
1173 				CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1174 				cfg = CSR_READ_1(sc, RL_CFG2);
1175 				cfg |= RL_CFG2_MSI;
1176 				CSR_WRITE_1(sc, RL_CFG2, cfg);
1177 				CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1178 			} else
1179 				pci_release_msi(dev);
1180 		}
1181 	}
1182 
1183 	/* Allocate interrupt */
1184 	if ((sc->rl_flags & RL_FLAG_MSI) == 0) {
1185 		rid = 0;
1186 		sc->rl_irq[0] = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
1187 		    RF_SHAREABLE | RF_ACTIVE);
1188 		if (sc->rl_irq[0] == NULL) {
1189 			device_printf(dev, "couldn't allocate IRQ resources\n");
1190 			error = ENXIO;
1191 			goto fail;
1192 		}
1193 	} else {
1194 		for (i = 0, rid = 1; i < RL_MSI_MESSAGES; i++, rid++) {
1195 			sc->rl_irq[i] = bus_alloc_resource_any(dev,
1196 			    SYS_RES_IRQ, &rid, RF_ACTIVE);
1197 			if (sc->rl_irq[i] == NULL) {
1198 				device_printf(dev,
1199 				    "couldn't llocate IRQ resources for "
1200 				    "message %d\n", rid);
1201 				error = ENXIO;
1202 				goto fail;
1203 			}
1204 		}
1205 	}
1206 
1207 	if ((sc->rl_flags & RL_FLAG_MSI) == 0) {
1208 		CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1209 		cfg = CSR_READ_1(sc, RL_CFG2);
1210 		if ((cfg & RL_CFG2_MSI) != 0) {
1211 			device_printf(dev, "turning off MSI enable bit.\n");
1212 			cfg &= ~RL_CFG2_MSI;
1213 			CSR_WRITE_1(sc, RL_CFG2, cfg);
1214 		}
1215 		CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1216 	}
1217 
1218 	/* Reset the adapter. */
1219 	RL_LOCK(sc);
1220 	re_reset(sc);
1221 	RL_UNLOCK(sc);
1222 
1223 	hw_rev = re_hwrevs;
1224 	hwrev = CSR_READ_4(sc, RL_TXCFG);
1225 	switch (hwrev & 0x70000000) {
1226 	case 0x00000000:
1227 	case 0x10000000:
1228 		device_printf(dev, "Chip rev. 0x%08x\n", hwrev & 0xfc800000);
1229 		hwrev &= (RL_TXCFG_HWREV | 0x80000000);
1230 		break;
1231 	default:
1232 		device_printf(dev, "Chip rev. 0x%08x\n", hwrev & 0x7c800000);
1233 		hwrev &= RL_TXCFG_HWREV;
1234 		break;
1235 	}
1236 	device_printf(dev, "MAC rev. 0x%08x\n", hwrev & 0x00700000);
1237 	while (hw_rev->rl_desc != NULL) {
1238 		if (hw_rev->rl_rev == hwrev) {
1239 			sc->rl_type = hw_rev->rl_type;
1240 			sc->rl_hwrev = hw_rev->rl_rev;
1241 			break;
1242 		}
1243 		hw_rev++;
1244 	}
1245 	if (hw_rev->rl_desc == NULL) {
1246 		device_printf(dev, "Unknown H/W revision: 0x%08x\n", hwrev);
1247 		error = ENXIO;
1248 		goto fail;
1249 	}
1250 
1251 	switch (hw_rev->rl_rev) {
1252 	case RL_HWREV_8139CPLUS:
1253 		sc->rl_flags |= RL_FLAG_NOJUMBO | RL_FLAG_FASTETHER;
1254 		break;
1255 	case RL_HWREV_8100E:
1256 	case RL_HWREV_8101E:
1257 		sc->rl_flags |= RL_FLAG_NOJUMBO | RL_FLAG_PHYWAKE |
1258 		    RL_FLAG_FASTETHER;
1259 		break;
1260 	case RL_HWREV_8102E:
1261 	case RL_HWREV_8102EL:
1262 		sc->rl_flags |= RL_FLAG_NOJUMBO | RL_FLAG_PHYWAKE |
1263 		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
1264 		    RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP;
1265 		break;
1266 	case RL_HWREV_8168_SPIN1:
1267 	case RL_HWREV_8168_SPIN2:
1268 		sc->rl_flags |= RL_FLAG_WOLRXENB;
1269 		/* FALLTHROUGH */
1270 	case RL_HWREV_8168_SPIN3:
1271 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_MACSTAT;
1272 		break;
1273 	case RL_HWREV_8168C_SPIN2:
1274 		sc->rl_flags |= RL_FLAG_MACSLEEP;
1275 		/* FALLTHROUGH */
1276 	case RL_HWREV_8168C:
1277 		if ((hwrev & 0x00700000) == 0x00200000)
1278 			sc->rl_flags |= RL_FLAG_MACSLEEP;
1279 		/* FALLTHROUGH */
1280 	case RL_HWREV_8168CP:
1281 	case RL_HWREV_8168D:
1282 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
1283 		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP;
1284 		/*
1285 		 * These controllers support jumbo frame but it seems
1286 		 * that enabling it requires touching additional magic
1287 		 * registers. Depending on MAC revisions some
1288 		 * controllers need to disable checksum offload. So
1289 		 * disable jumbo frame until I have better idea what
1290 		 * it really requires to make it support.
1291 		 * RTL8168C/CP : supports up to 6KB jumbo frame.
1292 		 * RTL8111C/CP : supports up to 9KB jumbo frame.
1293 		 */
1294 		sc->rl_flags |= RL_FLAG_NOJUMBO;
1295 		break;
1296 	case RL_HWREV_8169_8110SB:
1297 	case RL_HWREV_8169_8110SBL:
1298 	case RL_HWREV_8169_8110SC:
1299 	case RL_HWREV_8169_8110SCE:
1300 		sc->rl_flags |= RL_FLAG_PHYWAKE;
1301 		/* FALLTHROUGH */
1302 	case RL_HWREV_8169:
1303 	case RL_HWREV_8169S:
1304 	case RL_HWREV_8110S:
1305 		sc->rl_flags |= RL_FLAG_MACRESET;
1306 		break;
1307 	default:
1308 		break;
1309 	}
1310 
1311 	/* Enable PME. */
1312 	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1313 	cfg = CSR_READ_1(sc, RL_CFG1);
1314 	cfg |= RL_CFG1_PME;
1315 	CSR_WRITE_1(sc, RL_CFG1, cfg);
1316 	cfg = CSR_READ_1(sc, RL_CFG5);
1317 	cfg &= RL_CFG5_PME_STS;
1318 	CSR_WRITE_1(sc, RL_CFG5, cfg);
1319 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1320 
1321 	if ((sc->rl_flags & RL_FLAG_PAR) != 0) {
1322 		/*
1323 		 * XXX Should have a better way to extract station
1324 		 * address from EEPROM.
1325 		 */
1326 		for (i = 0; i < ETHER_ADDR_LEN; i++)
1327 			eaddr[i] = CSR_READ_1(sc, RL_IDR0 + i);
1328 	} else {
1329 		sc->rl_eewidth = RL_9356_ADDR_LEN;
1330 		re_read_eeprom(sc, (caddr_t)&re_did, 0, 1);
1331 		if (re_did != 0x8129)
1332 			sc->rl_eewidth = RL_9346_ADDR_LEN;
1333 
1334 		/*
1335 		 * Get station address from the EEPROM.
1336 		 */
1337 		re_read_eeprom(sc, (caddr_t)as, RL_EE_EADDR, 3);
1338 		for (i = 0; i < ETHER_ADDR_LEN / 2; i++)
1339 			as[i] = le16toh(as[i]);
1340 		bcopy(as, eaddr, sizeof(eaddr));
1341 	}
1342 
1343 	if (sc->rl_type == RL_8169) {
1344 		/* Set RX length mask and number of descriptors. */
1345 		sc->rl_rxlenmask = RL_RDESC_STAT_GFRAGLEN;
1346 		sc->rl_txstart = RL_GTXSTART;
1347 		sc->rl_ldata.rl_tx_desc_cnt = RL_8169_TX_DESC_CNT;
1348 		sc->rl_ldata.rl_rx_desc_cnt = RL_8169_RX_DESC_CNT;
1349 	} else {
1350 		/* Set RX length mask and number of descriptors. */
1351 		sc->rl_rxlenmask = RL_RDESC_STAT_FRAGLEN;
1352 		sc->rl_txstart = RL_TXSTART;
1353 		sc->rl_ldata.rl_tx_desc_cnt = RL_8139_TX_DESC_CNT;
1354 		sc->rl_ldata.rl_rx_desc_cnt = RL_8139_RX_DESC_CNT;
1355 	}
1356 
1357 	error = re_allocmem(dev, sc);
1358 	if (error)
1359 		goto fail;
1360 
1361 	ifp = sc->rl_ifp = if_alloc(IFT_ETHER);
1362 	if (ifp == NULL) {
1363 		device_printf(dev, "can not if_alloc()\n");
1364 		error = ENOSPC;
1365 		goto fail;
1366 	}
1367 
1368 	/* Take controller out of deep sleep mode. */
1369 	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
1370 		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
1371 			CSR_WRITE_1(sc, RL_GPIO,
1372 			    CSR_READ_1(sc, RL_GPIO) | 0x01);
1373 		else
1374 			CSR_WRITE_1(sc, RL_GPIO,
1375 			    CSR_READ_1(sc, RL_GPIO) & ~0x01);
1376 	}
1377 
1378 	/* Take PHY out of power down mode. */
1379 	if ((sc->rl_flags & RL_FLAG_PHYWAKE) != 0) {
1380 		re_gmii_writereg(dev, 1, 0x1f, 0);
1381 		re_gmii_writereg(dev, 1, 0x0e, 0);
1382 	}
1383 
1384 	/* Do MII setup */
1385 	if (mii_phy_probe(dev, &sc->rl_miibus,
1386 	    re_ifmedia_upd, re_ifmedia_sts)) {
1387 		device_printf(dev, "MII without any phy!\n");
1388 		error = ENXIO;
1389 		goto fail;
1390 	}
1391 
1392 	ifp->if_softc = sc;
1393 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1394 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1395 	ifp->if_ioctl = re_ioctl;
1396 	ifp->if_start = re_start;
1397 	ifp->if_hwassist = RE_CSUM_FEATURES;
1398 	ifp->if_capabilities = IFCAP_HWCSUM;
1399 	ifp->if_capenable = ifp->if_capabilities;
1400 	ifp->if_init = re_init;
1401 	IFQ_SET_MAXLEN(&ifp->if_snd, RL_IFQ_MAXLEN);
1402 	ifp->if_snd.ifq_drv_maxlen = RL_IFQ_MAXLEN;
1403 	IFQ_SET_READY(&ifp->if_snd);
1404 
1405 	TASK_INIT(&sc->rl_txtask, 1, re_tx_task, ifp);
1406 	TASK_INIT(&sc->rl_inttask, 0, re_int_task, sc);
1407 
1408 	/*
1409 	 * XXX
1410 	 * Still have no idea how to make TSO work on 8168C, 8168CP,
1411 	 * 8111C and 8111CP.
1412 	 */
1413 	if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
1414 		ifp->if_hwassist |= CSUM_TSO;
1415 		ifp->if_capabilities |= IFCAP_TSO4;
1416 	}
1417 
1418 	/*
1419 	 * Call MI attach routine.
1420 	 */
1421 	ether_ifattach(ifp, eaddr);
1422 
1423 	/* VLAN capability setup */
1424 	ifp->if_capabilities |= IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING;
1425 	if (ifp->if_capabilities & IFCAP_HWCSUM)
1426 		ifp->if_capabilities |= IFCAP_VLAN_HWCSUM;
1427 	/* Enable WOL if PM is supported. */
1428 	if (pci_find_extcap(sc->rl_dev, PCIY_PMG, &reg) == 0)
1429 		ifp->if_capabilities |= IFCAP_WOL;
1430 	ifp->if_capenable = ifp->if_capabilities;
1431 	/*
1432 	 * Don't enable TSO by default. Under certain
1433 	 * circumtances the controller generated corrupted
1434 	 * packets in TSO size.
1435 	 */
1436 	ifp->if_hwassist &= ~CSUM_TSO;
1437 	ifp->if_capenable &= ~IFCAP_TSO4;
1438 #ifdef DEVICE_POLLING
1439 	ifp->if_capabilities |= IFCAP_POLLING;
1440 #endif
1441 	/*
1442 	 * Tell the upper layer(s) we support long frames.
1443 	 * Must appear after the call to ether_ifattach() because
1444 	 * ether_ifattach() sets ifi_hdrlen to the default value.
1445 	 */
1446 	ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
1447 
1448 #ifdef RE_DIAG
1449 	/*
1450 	 * Perform hardware diagnostic on the original RTL8169.
1451 	 * Some 32-bit cards were incorrectly wired and would
1452 	 * malfunction if plugged into a 64-bit slot.
1453 	 */
1454 
1455 	if (hwrev == RL_HWREV_8169) {
1456 		error = re_diag(sc);
1457 		if (error) {
1458 			device_printf(dev,
1459 		    	"attach aborted due to hardware diag failure\n");
1460 			ether_ifdetach(ifp);
1461 			goto fail;
1462 		}
1463 	}
1464 #endif
1465 
1466 	/* Hook interrupt last to avoid having to lock softc */
1467 	if ((sc->rl_flags & RL_FLAG_MSI) == 0)
1468 		error = bus_setup_intr(dev, sc->rl_irq[0],
1469 		    INTR_TYPE_NET | INTR_MPSAFE, re_intr, NULL, sc,
1470 		    &sc->rl_intrhand[0]);
1471 	else {
1472 		for (i = 0; i < RL_MSI_MESSAGES; i++) {
1473 			error = bus_setup_intr(dev, sc->rl_irq[i],
1474 			    INTR_TYPE_NET | INTR_MPSAFE, re_intr, NULL, sc,
1475 		    	    &sc->rl_intrhand[i]);
1476 			if (error != 0)
1477 				break;
1478 		}
1479 	}
1480 	if (error) {
1481 		device_printf(dev, "couldn't set up irq\n");
1482 		ether_ifdetach(ifp);
1483 	}
1484 
1485 fail:
1486 
1487 	if (error)
1488 		re_detach(dev);
1489 
1490 	return (error);
1491 }
1492 
1493 /*
1494  * Shutdown hardware and free up resources. This can be called any
1495  * time after the mutex has been initialized. It is called in both
1496  * the error case in attach and the normal detach case so it needs
1497  * to be careful about only freeing resources that have actually been
1498  * allocated.
1499  */
1500 static int
1501 re_detach(device_t dev)
1502 {
1503 	struct rl_softc		*sc;
1504 	struct ifnet		*ifp;
1505 	int			i, rid;
1506 
1507 	sc = device_get_softc(dev);
1508 	ifp = sc->rl_ifp;
1509 	KASSERT(mtx_initialized(&sc->rl_mtx), ("re mutex not initialized"));
1510 
1511 	/* These should only be active if attach succeeded */
1512 	if (device_is_attached(dev)) {
1513 #ifdef DEVICE_POLLING
1514 		if (ifp->if_capenable & IFCAP_POLLING)
1515 			ether_poll_deregister(ifp);
1516 #endif
1517 		RL_LOCK(sc);
1518 #if 0
1519 		sc->suspended = 1;
1520 #endif
1521 		re_stop(sc);
1522 		RL_UNLOCK(sc);
1523 		callout_drain(&sc->rl_stat_callout);
1524 		taskqueue_drain(taskqueue_fast, &sc->rl_inttask);
1525 		taskqueue_drain(taskqueue_fast, &sc->rl_txtask);
1526 		/*
1527 		 * Force off the IFF_UP flag here, in case someone
1528 		 * still had a BPF descriptor attached to this
1529 		 * interface. If they do, ether_ifdetach() will cause
1530 		 * the BPF code to try and clear the promisc mode
1531 		 * flag, which will bubble down to re_ioctl(),
1532 		 * which will try to call re_init() again. This will
1533 		 * turn the NIC back on and restart the MII ticker,
1534 		 * which will panic the system when the kernel tries
1535 		 * to invoke the re_tick() function that isn't there
1536 		 * anymore.
1537 		 */
1538 		ifp->if_flags &= ~IFF_UP;
1539 		ether_ifdetach(ifp);
1540 	}
1541 	if (sc->rl_miibus)
1542 		device_delete_child(dev, sc->rl_miibus);
1543 	bus_generic_detach(dev);
1544 
1545 	/*
1546 	 * The rest is resource deallocation, so we should already be
1547 	 * stopped here.
1548 	 */
1549 
1550 	for (i = 0; i < RL_MSI_MESSAGES; i++) {
1551 		if (sc->rl_intrhand[i] != NULL) {
1552 			bus_teardown_intr(dev, sc->rl_irq[i],
1553 			    sc->rl_intrhand[i]);
1554 			sc->rl_intrhand[i] = NULL;
1555 		}
1556 	}
1557 	if (ifp != NULL)
1558 		if_free(ifp);
1559 	if ((sc->rl_flags & RL_FLAG_MSI) == 0) {
1560 		if (sc->rl_irq[0] != NULL) {
1561 			bus_release_resource(dev, SYS_RES_IRQ, 0,
1562 			    sc->rl_irq[0]);
1563 			sc->rl_irq[0] = NULL;
1564 		}
1565 	} else {
1566 		for (i = 0, rid = 1; i < RL_MSI_MESSAGES; i++, rid++) {
1567 			if (sc->rl_irq[i] != NULL) {
1568 				bus_release_resource(dev, SYS_RES_IRQ, rid,
1569 				    sc->rl_irq[i]);
1570 				sc->rl_irq[i] = NULL;
1571 			}
1572 		}
1573 		pci_release_msi(dev);
1574 	}
1575 	if (sc->rl_res)
1576 		bus_release_resource(dev, sc->rl_res_type, sc->rl_res_id,
1577 		    sc->rl_res);
1578 
1579 	/* Unload and free the RX DMA ring memory and map */
1580 
1581 	if (sc->rl_ldata.rl_rx_list_tag) {
1582 		bus_dmamap_unload(sc->rl_ldata.rl_rx_list_tag,
1583 		    sc->rl_ldata.rl_rx_list_map);
1584 		bus_dmamem_free(sc->rl_ldata.rl_rx_list_tag,
1585 		    sc->rl_ldata.rl_rx_list,
1586 		    sc->rl_ldata.rl_rx_list_map);
1587 		bus_dma_tag_destroy(sc->rl_ldata.rl_rx_list_tag);
1588 	}
1589 
1590 	/* Unload and free the TX DMA ring memory and map */
1591 
1592 	if (sc->rl_ldata.rl_tx_list_tag) {
1593 		bus_dmamap_unload(sc->rl_ldata.rl_tx_list_tag,
1594 		    sc->rl_ldata.rl_tx_list_map);
1595 		bus_dmamem_free(sc->rl_ldata.rl_tx_list_tag,
1596 		    sc->rl_ldata.rl_tx_list,
1597 		    sc->rl_ldata.rl_tx_list_map);
1598 		bus_dma_tag_destroy(sc->rl_ldata.rl_tx_list_tag);
1599 	}
1600 
1601 	/* Destroy all the RX and TX buffer maps */
1602 
1603 	if (sc->rl_ldata.rl_tx_mtag) {
1604 		for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++)
1605 			bus_dmamap_destroy(sc->rl_ldata.rl_tx_mtag,
1606 			    sc->rl_ldata.rl_tx_desc[i].tx_dmamap);
1607 		bus_dma_tag_destroy(sc->rl_ldata.rl_tx_mtag);
1608 	}
1609 	if (sc->rl_ldata.rl_rx_mtag) {
1610 		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++)
1611 			bus_dmamap_destroy(sc->rl_ldata.rl_rx_mtag,
1612 			    sc->rl_ldata.rl_rx_desc[i].rx_dmamap);
1613 		if (sc->rl_ldata.rl_rx_sparemap)
1614 			bus_dmamap_destroy(sc->rl_ldata.rl_rx_mtag,
1615 			    sc->rl_ldata.rl_rx_sparemap);
1616 		bus_dma_tag_destroy(sc->rl_ldata.rl_rx_mtag);
1617 	}
1618 
1619 	/* Unload and free the stats buffer and map */
1620 
1621 	if (sc->rl_ldata.rl_stag) {
1622 		bus_dmamap_unload(sc->rl_ldata.rl_stag,
1623 		    sc->rl_ldata.rl_rx_list_map);
1624 		bus_dmamem_free(sc->rl_ldata.rl_stag,
1625 		    sc->rl_ldata.rl_stats,
1626 		    sc->rl_ldata.rl_smap);
1627 		bus_dma_tag_destroy(sc->rl_ldata.rl_stag);
1628 	}
1629 
1630 	if (sc->rl_parent_tag)
1631 		bus_dma_tag_destroy(sc->rl_parent_tag);
1632 
1633 	mtx_destroy(&sc->rl_mtx);
1634 
1635 	return (0);
1636 }
1637 
1638 static __inline void
1639 re_discard_rxbuf(struct rl_softc *sc, int idx)
1640 {
1641 	struct rl_desc		*desc;
1642 	struct rl_rxdesc	*rxd;
1643 	uint32_t		cmdstat;
1644 
1645 	rxd = &sc->rl_ldata.rl_rx_desc[idx];
1646 	desc = &sc->rl_ldata.rl_rx_list[idx];
1647 	desc->rl_vlanctl = 0;
1648 	cmdstat = rxd->rx_size;
1649 	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
1650 		cmdstat |= RL_RDESC_CMD_EOR;
1651 	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
1652 }
1653 
1654 static int
1655 re_newbuf(struct rl_softc *sc, int idx)
1656 {
1657 	struct mbuf		*m;
1658 	struct rl_rxdesc	*rxd;
1659 	bus_dma_segment_t	segs[1];
1660 	bus_dmamap_t		map;
1661 	struct rl_desc		*desc;
1662 	uint32_t		cmdstat;
1663 	int			error, nsegs;
1664 
1665 	m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
1666 	if (m == NULL)
1667 		return (ENOBUFS);
1668 
1669 	m->m_len = m->m_pkthdr.len = MCLBYTES;
1670 #ifdef RE_FIXUP_RX
1671 	/*
1672 	 * This is part of an evil trick to deal with non-x86 platforms.
1673 	 * The RealTek chip requires RX buffers to be aligned on 64-bit
1674 	 * boundaries, but that will hose non-x86 machines. To get around
1675 	 * this, we leave some empty space at the start of each buffer
1676 	 * and for non-x86 hosts, we copy the buffer back six bytes
1677 	 * to achieve word alignment. This is slightly more efficient
1678 	 * than allocating a new buffer, copying the contents, and
1679 	 * discarding the old buffer.
1680 	 */
1681 	m_adj(m, RE_ETHER_ALIGN);
1682 #endif
1683 	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_rx_mtag,
1684 	    sc->rl_ldata.rl_rx_sparemap, m, segs, &nsegs, BUS_DMA_NOWAIT);
1685 	if (error != 0) {
1686 		m_freem(m);
1687 		return (ENOBUFS);
1688 	}
1689 	KASSERT(nsegs == 1, ("%s: %d segment returned!", __func__, nsegs));
1690 
1691 	rxd = &sc->rl_ldata.rl_rx_desc[idx];
1692 	if (rxd->rx_m != NULL) {
1693 		bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap,
1694 		    BUS_DMASYNC_POSTREAD);
1695 		bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap);
1696 	}
1697 
1698 	rxd->rx_m = m;
1699 	map = rxd->rx_dmamap;
1700 	rxd->rx_dmamap = sc->rl_ldata.rl_rx_sparemap;
1701 	rxd->rx_size = segs[0].ds_len;
1702 	sc->rl_ldata.rl_rx_sparemap = map;
1703 	bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap,
1704 	    BUS_DMASYNC_PREREAD);
1705 
1706 	desc = &sc->rl_ldata.rl_rx_list[idx];
1707 	desc->rl_vlanctl = 0;
1708 	desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[0].ds_addr));
1709 	desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[0].ds_addr));
1710 	cmdstat = segs[0].ds_len;
1711 	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
1712 		cmdstat |= RL_RDESC_CMD_EOR;
1713 	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
1714 
1715 	return (0);
1716 }
1717 
1718 #ifdef RE_FIXUP_RX
1719 static __inline void
1720 re_fixup_rx(struct mbuf *m)
1721 {
1722 	int                     i;
1723 	uint16_t                *src, *dst;
1724 
1725 	src = mtod(m, uint16_t *);
1726 	dst = src - (RE_ETHER_ALIGN - ETHER_ALIGN) / sizeof *src;
1727 
1728 	for (i = 0; i < (m->m_len / sizeof(uint16_t) + 1); i++)
1729 		*dst++ = *src++;
1730 
1731 	m->m_data -= RE_ETHER_ALIGN - ETHER_ALIGN;
1732 }
1733 #endif
1734 
1735 static int
1736 re_tx_list_init(struct rl_softc *sc)
1737 {
1738 	struct rl_desc		*desc;
1739 	int			i;
1740 
1741 	RL_LOCK_ASSERT(sc);
1742 
1743 	bzero(sc->rl_ldata.rl_tx_list,
1744 	    sc->rl_ldata.rl_tx_desc_cnt * sizeof(struct rl_desc));
1745 	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++)
1746 		sc->rl_ldata.rl_tx_desc[i].tx_m = NULL;
1747 	/* Set EOR. */
1748 	desc = &sc->rl_ldata.rl_tx_list[sc->rl_ldata.rl_tx_desc_cnt - 1];
1749 	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_EOR);
1750 
1751 	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
1752 	    sc->rl_ldata.rl_tx_list_map,
1753 	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1754 
1755 	sc->rl_ldata.rl_tx_prodidx = 0;
1756 	sc->rl_ldata.rl_tx_considx = 0;
1757 	sc->rl_ldata.rl_tx_free = sc->rl_ldata.rl_tx_desc_cnt;
1758 
1759 	return (0);
1760 }
1761 
1762 static int
1763 re_rx_list_init(struct rl_softc *sc)
1764 {
1765 	int			error, i;
1766 
1767 	bzero(sc->rl_ldata.rl_rx_list,
1768 	    sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc));
1769 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1770 		sc->rl_ldata.rl_rx_desc[i].rx_m = NULL;
1771 		if ((error = re_newbuf(sc, i)) != 0)
1772 			return (error);
1773 	}
1774 
1775 	/* Flush the RX descriptors */
1776 
1777 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1778 	    sc->rl_ldata.rl_rx_list_map,
1779 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
1780 
1781 	sc->rl_ldata.rl_rx_prodidx = 0;
1782 	sc->rl_head = sc->rl_tail = NULL;
1783 
1784 	return (0);
1785 }
1786 
1787 /*
1788  * RX handler for C+ and 8169. For the gigE chips, we support
1789  * the reception of jumbo frames that have been fragmented
1790  * across multiple 2K mbuf cluster buffers.
1791  */
1792 static int
1793 re_rxeof(struct rl_softc *sc)
1794 {
1795 	struct mbuf		*m;
1796 	struct ifnet		*ifp;
1797 	int			i, total_len;
1798 	struct rl_desc		*cur_rx;
1799 	u_int32_t		rxstat, rxvlan;
1800 	int			maxpkt = 16;
1801 
1802 	RL_LOCK_ASSERT(sc);
1803 
1804 	ifp = sc->rl_ifp;
1805 
1806 	/* Invalidate the descriptor memory */
1807 
1808 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1809 	    sc->rl_ldata.rl_rx_list_map,
1810 	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
1811 
1812 	for (i = sc->rl_ldata.rl_rx_prodidx; maxpkt > 0;
1813 	    i = RL_RX_DESC_NXT(sc, i)) {
1814 		cur_rx = &sc->rl_ldata.rl_rx_list[i];
1815 		rxstat = le32toh(cur_rx->rl_cmdstat);
1816 		if ((rxstat & RL_RDESC_STAT_OWN) != 0)
1817 			break;
1818 		total_len = rxstat & sc->rl_rxlenmask;
1819 		rxvlan = le32toh(cur_rx->rl_vlanctl);
1820 		m = sc->rl_ldata.rl_rx_desc[i].rx_m;
1821 
1822 		if (!(rxstat & RL_RDESC_STAT_EOF)) {
1823 			if (re_newbuf(sc, i) != 0) {
1824 				/*
1825 				 * If this is part of a multi-fragment packet,
1826 				 * discard all the pieces.
1827 				 */
1828 				if (sc->rl_head != NULL) {
1829 					m_freem(sc->rl_head);
1830 					sc->rl_head = sc->rl_tail = NULL;
1831 				}
1832 				re_discard_rxbuf(sc, i);
1833 				continue;
1834 			}
1835 			m->m_len = RE_RX_DESC_BUFLEN;
1836 			if (sc->rl_head == NULL)
1837 				sc->rl_head = sc->rl_tail = m;
1838 			else {
1839 				m->m_flags &= ~M_PKTHDR;
1840 				sc->rl_tail->m_next = m;
1841 				sc->rl_tail = m;
1842 			}
1843 			continue;
1844 		}
1845 
1846 		/*
1847 		 * NOTE: for the 8139C+, the frame length field
1848 		 * is always 12 bits in size, but for the gigE chips,
1849 		 * it is 13 bits (since the max RX frame length is 16K).
1850 		 * Unfortunately, all 32 bits in the status word
1851 		 * were already used, so to make room for the extra
1852 		 * length bit, RealTek took out the 'frame alignment
1853 		 * error' bit and shifted the other status bits
1854 		 * over one slot. The OWN, EOR, FS and LS bits are
1855 		 * still in the same places. We have already extracted
1856 		 * the frame length and checked the OWN bit, so rather
1857 		 * than using an alternate bit mapping, we shift the
1858 		 * status bits one space to the right so we can evaluate
1859 		 * them using the 8169 status as though it was in the
1860 		 * same format as that of the 8139C+.
1861 		 */
1862 		if (sc->rl_type == RL_8169)
1863 			rxstat >>= 1;
1864 
1865 		/*
1866 		 * if total_len > 2^13-1, both _RXERRSUM and _GIANT will be
1867 		 * set, but if CRC is clear, it will still be a valid frame.
1868 		 */
1869 		if (rxstat & RL_RDESC_STAT_RXERRSUM && !(total_len > 8191 &&
1870 		    (rxstat & RL_RDESC_STAT_ERRS) == RL_RDESC_STAT_GIANT)) {
1871 			ifp->if_ierrors++;
1872 			/*
1873 			 * If this is part of a multi-fragment packet,
1874 			 * discard all the pieces.
1875 			 */
1876 			if (sc->rl_head != NULL) {
1877 				m_freem(sc->rl_head);
1878 				sc->rl_head = sc->rl_tail = NULL;
1879 			}
1880 			re_discard_rxbuf(sc, i);
1881 			continue;
1882 		}
1883 
1884 		/*
1885 		 * If allocating a replacement mbuf fails,
1886 		 * reload the current one.
1887 		 */
1888 
1889 		if (re_newbuf(sc, i) != 0) {
1890 			ifp->if_iqdrops++;
1891 			if (sc->rl_head != NULL) {
1892 				m_freem(sc->rl_head);
1893 				sc->rl_head = sc->rl_tail = NULL;
1894 			}
1895 			re_discard_rxbuf(sc, i);
1896 			continue;
1897 		}
1898 
1899 		if (sc->rl_head != NULL) {
1900 			m->m_len = total_len % RE_RX_DESC_BUFLEN;
1901 			if (m->m_len == 0)
1902 				m->m_len = RE_RX_DESC_BUFLEN;
1903 			/*
1904 			 * Special case: if there's 4 bytes or less
1905 			 * in this buffer, the mbuf can be discarded:
1906 			 * the last 4 bytes is the CRC, which we don't
1907 			 * care about anyway.
1908 			 */
1909 			if (m->m_len <= ETHER_CRC_LEN) {
1910 				sc->rl_tail->m_len -=
1911 				    (ETHER_CRC_LEN - m->m_len);
1912 				m_freem(m);
1913 			} else {
1914 				m->m_len -= ETHER_CRC_LEN;
1915 				m->m_flags &= ~M_PKTHDR;
1916 				sc->rl_tail->m_next = m;
1917 			}
1918 			m = sc->rl_head;
1919 			sc->rl_head = sc->rl_tail = NULL;
1920 			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
1921 		} else
1922 			m->m_pkthdr.len = m->m_len =
1923 			    (total_len - ETHER_CRC_LEN);
1924 
1925 #ifdef RE_FIXUP_RX
1926 		re_fixup_rx(m);
1927 #endif
1928 		ifp->if_ipackets++;
1929 		m->m_pkthdr.rcvif = ifp;
1930 
1931 		/* Do RX checksumming if enabled */
1932 
1933 		if (ifp->if_capenable & IFCAP_RXCSUM) {
1934 			if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
1935 				/* Check IP header checksum */
1936 				if (rxstat & RL_RDESC_STAT_PROTOID)
1937 					m->m_pkthdr.csum_flags |=
1938 					    CSUM_IP_CHECKED;
1939 				if (!(rxstat & RL_RDESC_STAT_IPSUMBAD))
1940 					m->m_pkthdr.csum_flags |=
1941 					    CSUM_IP_VALID;
1942 
1943 				/* Check TCP/UDP checksum */
1944 				if ((RL_TCPPKT(rxstat) &&
1945 				    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
1946 				    (RL_UDPPKT(rxstat) &&
1947 				     !(rxstat & RL_RDESC_STAT_UDPSUMBAD))) {
1948 					m->m_pkthdr.csum_flags |=
1949 						CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
1950 					m->m_pkthdr.csum_data = 0xffff;
1951 				}
1952 			} else {
1953 				/*
1954 				 * RTL8168C/RTL816CP/RTL8111C/RTL8111CP
1955 				 */
1956 				if ((rxstat & RL_RDESC_STAT_PROTOID) &&
1957 				    (rxvlan & RL_RDESC_IPV4))
1958 					m->m_pkthdr.csum_flags |=
1959 					    CSUM_IP_CHECKED;
1960 				if (!(rxstat & RL_RDESC_STAT_IPSUMBAD) &&
1961 				    (rxvlan & RL_RDESC_IPV4))
1962 					m->m_pkthdr.csum_flags |=
1963 					    CSUM_IP_VALID;
1964 				if (((rxstat & RL_RDESC_STAT_TCP) &&
1965 				    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
1966 				    ((rxstat & RL_RDESC_STAT_UDP) &&
1967 				    !(rxstat & RL_RDESC_STAT_UDPSUMBAD))) {
1968 					m->m_pkthdr.csum_flags |=
1969 						CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
1970 					m->m_pkthdr.csum_data = 0xffff;
1971 				}
1972 			}
1973 		}
1974 		maxpkt--;
1975 		if (rxvlan & RL_RDESC_VLANCTL_TAG) {
1976 			m->m_pkthdr.ether_vtag =
1977 			    bswap16((rxvlan & RL_RDESC_VLANCTL_DATA));
1978 			m->m_flags |= M_VLANTAG;
1979 		}
1980 		RL_UNLOCK(sc);
1981 		(*ifp->if_input)(ifp, m);
1982 		RL_LOCK(sc);
1983 	}
1984 
1985 	/* Flush the RX DMA ring */
1986 
1987 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1988 	    sc->rl_ldata.rl_rx_list_map,
1989 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
1990 
1991 	sc->rl_ldata.rl_rx_prodidx = i;
1992 
1993 	if (maxpkt)
1994 		return(EAGAIN);
1995 
1996 	return(0);
1997 }
1998 
1999 static void
2000 re_txeof(struct rl_softc *sc)
2001 {
2002 	struct ifnet		*ifp;
2003 	struct rl_txdesc	*txd;
2004 	u_int32_t		txstat;
2005 	int			cons;
2006 
2007 	cons = sc->rl_ldata.rl_tx_considx;
2008 	if (cons == sc->rl_ldata.rl_tx_prodidx)
2009 		return;
2010 
2011 	ifp = sc->rl_ifp;
2012 	/* Invalidate the TX descriptor list */
2013 	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2014 	    sc->rl_ldata.rl_tx_list_map,
2015 	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2016 
2017 	for (; cons != sc->rl_ldata.rl_tx_prodidx;
2018 	    cons = RL_TX_DESC_NXT(sc, cons)) {
2019 		txstat = le32toh(sc->rl_ldata.rl_tx_list[cons].rl_cmdstat);
2020 		if (txstat & RL_TDESC_STAT_OWN)
2021 			break;
2022 		/*
2023 		 * We only stash mbufs in the last descriptor
2024 		 * in a fragment chain, which also happens to
2025 		 * be the only place where the TX status bits
2026 		 * are valid.
2027 		 */
2028 		if (txstat & RL_TDESC_CMD_EOF) {
2029 			txd = &sc->rl_ldata.rl_tx_desc[cons];
2030 			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
2031 			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
2032 			bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag,
2033 			    txd->tx_dmamap);
2034 			KASSERT(txd->tx_m != NULL,
2035 			    ("%s: freeing NULL mbufs!", __func__));
2036 			m_freem(txd->tx_m);
2037 			txd->tx_m = NULL;
2038 			if (txstat & (RL_TDESC_STAT_EXCESSCOL|
2039 			    RL_TDESC_STAT_COLCNT))
2040 				ifp->if_collisions++;
2041 			if (txstat & RL_TDESC_STAT_TXERRSUM)
2042 				ifp->if_oerrors++;
2043 			else
2044 				ifp->if_opackets++;
2045 		}
2046 		sc->rl_ldata.rl_tx_free++;
2047 		ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
2048 	}
2049 	sc->rl_ldata.rl_tx_considx = cons;
2050 
2051 	/* No changes made to the TX ring, so no flush needed */
2052 
2053 	if (sc->rl_ldata.rl_tx_free != sc->rl_ldata.rl_tx_desc_cnt) {
2054 #ifdef RE_TX_MODERATION
2055 		/*
2056 		 * If not all descriptors have been reaped yet, reload
2057 		 * the timer so that we will eventually get another
2058 		 * interrupt that will cause us to re-enter this routine.
2059 		 * This is done in case the transmitter has gone idle.
2060 		 */
2061 		CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2062 #endif
2063 	} else
2064 		sc->rl_watchdog_timer = 0;
2065 }
2066 
2067 static void
2068 re_tick(void *xsc)
2069 {
2070 	struct rl_softc		*sc;
2071 	struct mii_data		*mii;
2072 
2073 	sc = xsc;
2074 
2075 	RL_LOCK_ASSERT(sc);
2076 
2077 	mii = device_get_softc(sc->rl_miibus);
2078 	mii_tick(mii);
2079 	if ((sc->rl_flags & RL_FLAG_LINK) == 0)
2080 		re_miibus_statchg(sc->rl_dev);
2081 	/*
2082 	 * Reclaim transmitted frames here. Technically it is not
2083 	 * necessary to do here but it ensures periodic reclamation
2084 	 * regardless of Tx completion interrupt which seems to be
2085 	 * lost on PCIe based controllers under certain situations.
2086 	 */
2087 	re_txeof(sc);
2088 	re_watchdog(sc);
2089 	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
2090 }
2091 
2092 #ifdef DEVICE_POLLING
2093 static void
2094 re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
2095 {
2096 	struct rl_softc *sc = ifp->if_softc;
2097 
2098 	RL_LOCK(sc);
2099 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
2100 		re_poll_locked(ifp, cmd, count);
2101 	RL_UNLOCK(sc);
2102 }
2103 
2104 static void
2105 re_poll_locked(struct ifnet *ifp, enum poll_cmd cmd, int count)
2106 {
2107 	struct rl_softc *sc = ifp->if_softc;
2108 
2109 	RL_LOCK_ASSERT(sc);
2110 
2111 	sc->rxcycles = count;
2112 	re_rxeof(sc);
2113 	re_txeof(sc);
2114 
2115 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2116 		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_txtask);
2117 
2118 	if (cmd == POLL_AND_CHECK_STATUS) { /* also check status register */
2119 		u_int16_t       status;
2120 
2121 		status = CSR_READ_2(sc, RL_ISR);
2122 		if (status == 0xffff)
2123 			return;
2124 		if (status)
2125 			CSR_WRITE_2(sc, RL_ISR, status);
2126 		if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
2127 		    (sc->rl_flags & RL_FLAG_PCIE))
2128 			CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2129 
2130 		/*
2131 		 * XXX check behaviour on receiver stalls.
2132 		 */
2133 
2134 		if (status & RL_ISR_SYSTEM_ERR)
2135 			re_init_locked(sc);
2136 	}
2137 }
2138 #endif /* DEVICE_POLLING */
2139 
2140 static int
2141 re_intr(void *arg)
2142 {
2143 	struct rl_softc		*sc;
2144 	uint16_t		status;
2145 
2146 	sc = arg;
2147 
2148 	status = CSR_READ_2(sc, RL_ISR);
2149 	if (status == 0xFFFF || (status & RL_INTRS_CPLUS) == 0)
2150                 return (FILTER_STRAY);
2151 	CSR_WRITE_2(sc, RL_IMR, 0);
2152 
2153 	taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_inttask);
2154 
2155 	return (FILTER_HANDLED);
2156 }
2157 
2158 static void
2159 re_int_task(void *arg, int npending)
2160 {
2161 	struct rl_softc		*sc;
2162 	struct ifnet		*ifp;
2163 	u_int16_t		status;
2164 	int			rval = 0;
2165 
2166 	sc = arg;
2167 	ifp = sc->rl_ifp;
2168 
2169 	RL_LOCK(sc);
2170 
2171 	status = CSR_READ_2(sc, RL_ISR);
2172         CSR_WRITE_2(sc, RL_ISR, status);
2173 
2174 	if (sc->suspended ||
2175 	    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2176 		RL_UNLOCK(sc);
2177 		return;
2178 	}
2179 
2180 #ifdef DEVICE_POLLING
2181 	if  (ifp->if_capenable & IFCAP_POLLING) {
2182 		RL_UNLOCK(sc);
2183 		return;
2184 	}
2185 #endif
2186 
2187 	if (status & (RL_ISR_RX_OK|RL_ISR_RX_ERR|RL_ISR_FIFO_OFLOW))
2188 		rval = re_rxeof(sc);
2189 
2190 	/*
2191 	 * Some chips will ignore a second TX request issued
2192 	 * while an existing transmission is in progress. If
2193 	 * the transmitter goes idle but there are still
2194 	 * packets waiting to be sent, we need to restart the
2195 	 * channel here to flush them out. This only seems to
2196 	 * be required with the PCIe devices.
2197 	 */
2198 	if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
2199 	    (sc->rl_flags & RL_FLAG_PCIE))
2200 		CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2201 	if (status & (
2202 #ifdef RE_TX_MODERATION
2203 	    RL_ISR_TIMEOUT_EXPIRED|
2204 #else
2205 	    RL_ISR_TX_OK|
2206 #endif
2207 	    RL_ISR_TX_ERR|RL_ISR_TX_DESC_UNAVAIL))
2208 		re_txeof(sc);
2209 
2210 	if (status & RL_ISR_SYSTEM_ERR)
2211 		re_init_locked(sc);
2212 
2213 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2214 		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_txtask);
2215 
2216 	RL_UNLOCK(sc);
2217 
2218         if ((CSR_READ_2(sc, RL_ISR) & RL_INTRS_CPLUS) || rval) {
2219 		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_inttask);
2220 		return;
2221 	}
2222 
2223 	CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2224 }
2225 
2226 static int
2227 re_encap(struct rl_softc *sc, struct mbuf **m_head)
2228 {
2229 	struct rl_txdesc	*txd, *txd_last;
2230 	bus_dma_segment_t	segs[RL_NTXSEGS];
2231 	bus_dmamap_t		map;
2232 	struct mbuf		*m_new;
2233 	struct rl_desc		*desc;
2234 	int			nsegs, prod;
2235 	int			i, error, ei, si;
2236 	int			padlen;
2237 	uint32_t		cmdstat, csum_flags, vlanctl;
2238 
2239 	RL_LOCK_ASSERT(sc);
2240 	M_ASSERTPKTHDR((*m_head));
2241 
2242 	/*
2243 	 * With some of the RealTek chips, using the checksum offload
2244 	 * support in conjunction with the autopadding feature results
2245 	 * in the transmission of corrupt frames. For example, if we
2246 	 * need to send a really small IP fragment that's less than 60
2247 	 * bytes in size, and IP header checksumming is enabled, the
2248 	 * resulting ethernet frame that appears on the wire will
2249 	 * have garbled payload. To work around this, if TX IP checksum
2250 	 * offload is enabled, we always manually pad short frames out
2251 	 * to the minimum ethernet frame size.
2252 	 */
2253 	if ((sc->rl_flags & RL_FLAG_DESCV2) == 0 &&
2254 	    (*m_head)->m_pkthdr.len < RL_IP4CSUMTX_PADLEN &&
2255 	    ((*m_head)->m_pkthdr.csum_flags & CSUM_IP) != 0) {
2256 		padlen = RL_MIN_FRAMELEN - (*m_head)->m_pkthdr.len;
2257 		if (M_WRITABLE(*m_head) == 0) {
2258 			/* Get a writable copy. */
2259 			m_new = m_dup(*m_head, M_DONTWAIT);
2260 			m_freem(*m_head);
2261 			if (m_new == NULL) {
2262 				*m_head = NULL;
2263 				return (ENOBUFS);
2264 			}
2265 			*m_head = m_new;
2266 		}
2267 		if ((*m_head)->m_next != NULL ||
2268 		    M_TRAILINGSPACE(*m_head) < padlen) {
2269 			m_new = m_defrag(*m_head, M_DONTWAIT);
2270 			if (m_new == NULL) {
2271 				m_freem(*m_head);
2272 				*m_head = NULL;
2273 				return (ENOBUFS);
2274 			}
2275 		} else
2276 			m_new = *m_head;
2277 
2278 		/*
2279 		 * Manually pad short frames, and zero the pad space
2280 		 * to avoid leaking data.
2281 		 */
2282 		bzero(mtod(m_new, char *) + m_new->m_pkthdr.len, padlen);
2283 		m_new->m_pkthdr.len += padlen;
2284 		m_new->m_len = m_new->m_pkthdr.len;
2285 		*m_head = m_new;
2286 	}
2287 
2288 	prod = sc->rl_ldata.rl_tx_prodidx;
2289 	txd = &sc->rl_ldata.rl_tx_desc[prod];
2290 	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap,
2291 	    *m_head, segs, &nsegs, BUS_DMA_NOWAIT);
2292 	if (error == EFBIG) {
2293 		m_new = m_collapse(*m_head, M_DONTWAIT, RL_NTXSEGS);
2294 		if (m_new == NULL) {
2295 			m_freem(*m_head);
2296 			*m_head = NULL;
2297 			return (ENOBUFS);
2298 		}
2299 		*m_head = m_new;
2300 		error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_tx_mtag,
2301 		    txd->tx_dmamap, *m_head, segs, &nsegs, BUS_DMA_NOWAIT);
2302 		if (error != 0) {
2303 			m_freem(*m_head);
2304 			*m_head = NULL;
2305 			return (error);
2306 		}
2307 	} else if (error != 0)
2308 		return (error);
2309 	if (nsegs == 0) {
2310 		m_freem(*m_head);
2311 		*m_head = NULL;
2312 		return (EIO);
2313 	}
2314 
2315 	/* Check for number of available descriptors. */
2316 	if (sc->rl_ldata.rl_tx_free - nsegs <= 1) {
2317 		bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap);
2318 		return (ENOBUFS);
2319 	}
2320 
2321 	bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap,
2322 	    BUS_DMASYNC_PREWRITE);
2323 
2324 	/*
2325 	 * Set up checksum offload. Note: checksum offload bits must
2326 	 * appear in all descriptors of a multi-descriptor transmit
2327 	 * attempt. This is according to testing done with an 8169
2328 	 * chip. This is a requirement.
2329 	 */
2330 	vlanctl = 0;
2331 	csum_flags = 0;
2332 	if (((*m_head)->m_pkthdr.csum_flags & CSUM_TSO) != 0)
2333 		csum_flags = RL_TDESC_CMD_LGSEND |
2334 		    ((uint32_t)(*m_head)->m_pkthdr.tso_segsz <<
2335 		    RL_TDESC_CMD_MSSVAL_SHIFT);
2336 	else {
2337 		/*
2338 		 * Unconditionally enable IP checksum if TCP or UDP
2339 		 * checksum is required. Otherwise, TCP/UDP checksum
2340 		 * does't make effects.
2341 		 */
2342 		if (((*m_head)->m_pkthdr.csum_flags & RE_CSUM_FEATURES) != 0) {
2343 			if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
2344 				csum_flags |= RL_TDESC_CMD_IPCSUM;
2345 				if (((*m_head)->m_pkthdr.csum_flags &
2346 				    CSUM_TCP) != 0)
2347 					csum_flags |= RL_TDESC_CMD_TCPCSUM;
2348 				if (((*m_head)->m_pkthdr.csum_flags &
2349 				    CSUM_UDP) != 0)
2350 					csum_flags |= RL_TDESC_CMD_UDPCSUM;
2351 			} else {
2352 				vlanctl |= RL_TDESC_CMD_IPCSUMV2;
2353 				if (((*m_head)->m_pkthdr.csum_flags &
2354 				    CSUM_TCP) != 0)
2355 					vlanctl |= RL_TDESC_CMD_TCPCSUMV2;
2356 				if (((*m_head)->m_pkthdr.csum_flags &
2357 				    CSUM_UDP) != 0)
2358 					vlanctl |= RL_TDESC_CMD_UDPCSUMV2;
2359 			}
2360 		}
2361 	}
2362 
2363 	/*
2364 	 * Set up hardware VLAN tagging. Note: vlan tag info must
2365 	 * appear in all descriptors of a multi-descriptor
2366 	 * transmission attempt.
2367 	 */
2368 	if ((*m_head)->m_flags & M_VLANTAG)
2369 		vlanctl |= bswap16((*m_head)->m_pkthdr.ether_vtag) |
2370 		    RL_TDESC_VLANCTL_TAG;
2371 
2372 	si = prod;
2373 	for (i = 0; i < nsegs; i++, prod = RL_TX_DESC_NXT(sc, prod)) {
2374 		desc = &sc->rl_ldata.rl_tx_list[prod];
2375 		desc->rl_vlanctl = htole32(vlanctl);
2376 		desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[i].ds_addr));
2377 		desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[i].ds_addr));
2378 		cmdstat = segs[i].ds_len;
2379 		if (i != 0)
2380 			cmdstat |= RL_TDESC_CMD_OWN;
2381 		if (prod == sc->rl_ldata.rl_tx_desc_cnt - 1)
2382 			cmdstat |= RL_TDESC_CMD_EOR;
2383 		desc->rl_cmdstat = htole32(cmdstat | csum_flags);
2384 		sc->rl_ldata.rl_tx_free--;
2385 	}
2386 	/* Update producer index. */
2387 	sc->rl_ldata.rl_tx_prodidx = prod;
2388 
2389 	/* Set EOF on the last descriptor. */
2390 	ei = RL_TX_DESC_PRV(sc, prod);
2391 	desc = &sc->rl_ldata.rl_tx_list[ei];
2392 	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_EOF);
2393 
2394 	desc = &sc->rl_ldata.rl_tx_list[si];
2395 	/* Set SOF and transfer ownership of packet to the chip. */
2396 	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_OWN | RL_TDESC_CMD_SOF);
2397 
2398 	/*
2399 	 * Insure that the map for this transmission
2400 	 * is placed at the array index of the last descriptor
2401 	 * in this chain.  (Swap last and first dmamaps.)
2402 	 */
2403 	txd_last = &sc->rl_ldata.rl_tx_desc[ei];
2404 	map = txd->tx_dmamap;
2405 	txd->tx_dmamap = txd_last->tx_dmamap;
2406 	txd_last->tx_dmamap = map;
2407 	txd_last->tx_m = *m_head;
2408 
2409 	return (0);
2410 }
2411 
2412 static void
2413 re_tx_task(void *arg, int npending)
2414 {
2415 	struct ifnet		*ifp;
2416 
2417 	ifp = arg;
2418 	re_start(ifp);
2419 }
2420 
2421 /*
2422  * Main transmit routine for C+ and gigE NICs.
2423  */
2424 static void
2425 re_start(struct ifnet *ifp)
2426 {
2427 	struct rl_softc		*sc;
2428 	struct mbuf		*m_head;
2429 	int			queued;
2430 
2431 	sc = ifp->if_softc;
2432 
2433 	RL_LOCK(sc);
2434 
2435 	if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
2436 	    IFF_DRV_RUNNING || (sc->rl_flags & RL_FLAG_LINK) == 0) {
2437 		RL_UNLOCK(sc);
2438 		return;
2439 	}
2440 
2441 	for (queued = 0; !IFQ_DRV_IS_EMPTY(&ifp->if_snd) &&
2442 	    sc->rl_ldata.rl_tx_free > 1;) {
2443 		IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
2444 		if (m_head == NULL)
2445 			break;
2446 
2447 		if (re_encap(sc, &m_head) != 0) {
2448 			if (m_head == NULL)
2449 				break;
2450 			IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
2451 			ifp->if_drv_flags |= IFF_DRV_OACTIVE;
2452 			break;
2453 		}
2454 
2455 		/*
2456 		 * If there's a BPF listener, bounce a copy of this frame
2457 		 * to him.
2458 		 */
2459 		ETHER_BPF_MTAP(ifp, m_head);
2460 
2461 		queued++;
2462 	}
2463 
2464 	if (queued == 0) {
2465 #ifdef RE_TX_MODERATION
2466 		if (sc->rl_ldata.rl_tx_free != sc->rl_ldata.rl_tx_desc_cnt)
2467 			CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2468 #endif
2469 		RL_UNLOCK(sc);
2470 		return;
2471 	}
2472 
2473 	/* Flush the TX descriptors */
2474 
2475 	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2476 	    sc->rl_ldata.rl_tx_list_map,
2477 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2478 
2479 	CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2480 
2481 #ifdef RE_TX_MODERATION
2482 	/*
2483 	 * Use the countdown timer for interrupt moderation.
2484 	 * 'TX done' interrupts are disabled. Instead, we reset the
2485 	 * countdown timer, which will begin counting until it hits
2486 	 * the value in the TIMERINT register, and then trigger an
2487 	 * interrupt. Each time we write to the TIMERCNT register,
2488 	 * the timer count is reset to 0.
2489 	 */
2490 	CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2491 #endif
2492 
2493 	/*
2494 	 * Set a timeout in case the chip goes out to lunch.
2495 	 */
2496 	sc->rl_watchdog_timer = 5;
2497 
2498 	RL_UNLOCK(sc);
2499 }
2500 
2501 static void
2502 re_init(void *xsc)
2503 {
2504 	struct rl_softc		*sc = xsc;
2505 
2506 	RL_LOCK(sc);
2507 	re_init_locked(sc);
2508 	RL_UNLOCK(sc);
2509 }
2510 
2511 static void
2512 re_init_locked(struct rl_softc *sc)
2513 {
2514 	struct ifnet		*ifp = sc->rl_ifp;
2515 	struct mii_data		*mii;
2516 	uint32_t		reg;
2517 	uint16_t		cfg;
2518 	union {
2519 		uint32_t align_dummy;
2520 		u_char eaddr[ETHER_ADDR_LEN];
2521         } eaddr;
2522 
2523 	RL_LOCK_ASSERT(sc);
2524 
2525 	mii = device_get_softc(sc->rl_miibus);
2526 
2527 	/*
2528 	 * Cancel pending I/O and free all RX/TX buffers.
2529 	 */
2530 	re_stop(sc);
2531 
2532 	/* Put controller into known state. */
2533 	re_reset(sc);
2534 
2535 	/*
2536 	 * Enable C+ RX and TX mode, as well as VLAN stripping and
2537 	 * RX checksum offload. We must configure the C+ register
2538 	 * before all others.
2539 	 */
2540 	cfg = RL_CPLUSCMD_PCI_MRW;
2541 	if ((ifp->if_capenable & IFCAP_RXCSUM) != 0)
2542 		cfg |= RL_CPLUSCMD_RXCSUM_ENB;
2543 	if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0)
2544 		cfg |= RL_CPLUSCMD_VLANSTRIP;
2545 	if ((sc->rl_flags & RL_FLAG_MACSTAT) != 0) {
2546 		cfg |= RL_CPLUSCMD_MACSTAT_DIS;
2547 		/* XXX magic. */
2548 		cfg |= 0x0001;
2549 	} else
2550 		cfg |= RL_CPLUSCMD_RXENB | RL_CPLUSCMD_TXENB;
2551 	CSR_WRITE_2(sc, RL_CPLUS_CMD, cfg);
2552 	if (sc->rl_hwrev == RL_HWREV_8169_8110SC ||
2553 	    sc->rl_hwrev == RL_HWREV_8169_8110SCE) {
2554 		reg = 0x000fff00;
2555 		if ((CSR_READ_1(sc, RL_CFG2) & RL_CFG2_PCI66MHZ) != 0)
2556 			reg |= 0x000000ff;
2557 		if (sc->rl_hwrev == RL_HWREV_8169_8110SCE)
2558 			reg |= 0x00f00000;
2559 		CSR_WRITE_4(sc, 0x7c, reg);
2560 		/* Disable interrupt mitigation. */
2561 		CSR_WRITE_2(sc, 0xe2, 0);
2562 	}
2563 	/*
2564 	 * Disable TSO if interface MTU size is greater than MSS
2565 	 * allowed in controller.
2566 	 */
2567 	if (ifp->if_mtu > RL_TSO_MTU && (ifp->if_capenable & IFCAP_TSO4) != 0) {
2568 		ifp->if_capenable &= ~IFCAP_TSO4;
2569 		ifp->if_hwassist &= ~CSUM_TSO;
2570 	}
2571 
2572 	/*
2573 	 * Init our MAC address.  Even though the chipset
2574 	 * documentation doesn't mention it, we need to enter "Config
2575 	 * register write enable" mode to modify the ID registers.
2576 	 */
2577 	/* Copy MAC address on stack to align. */
2578 	bcopy(IF_LLADDR(ifp), eaddr.eaddr, ETHER_ADDR_LEN);
2579 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG);
2580 	CSR_WRITE_4(sc, RL_IDR0,
2581 	    htole32(*(u_int32_t *)(&eaddr.eaddr[0])));
2582 	CSR_WRITE_4(sc, RL_IDR4,
2583 	    htole32(*(u_int32_t *)(&eaddr.eaddr[4])));
2584 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
2585 
2586 	/*
2587 	 * For C+ mode, initialize the RX descriptors and mbufs.
2588 	 */
2589 	re_rx_list_init(sc);
2590 	re_tx_list_init(sc);
2591 
2592 	/*
2593 	 * Load the addresses of the RX and TX lists into the chip.
2594 	 */
2595 
2596 	CSR_WRITE_4(sc, RL_RXLIST_ADDR_HI,
2597 	    RL_ADDR_HI(sc->rl_ldata.rl_rx_list_addr));
2598 	CSR_WRITE_4(sc, RL_RXLIST_ADDR_LO,
2599 	    RL_ADDR_LO(sc->rl_ldata.rl_rx_list_addr));
2600 
2601 	CSR_WRITE_4(sc, RL_TXLIST_ADDR_HI,
2602 	    RL_ADDR_HI(sc->rl_ldata.rl_tx_list_addr));
2603 	CSR_WRITE_4(sc, RL_TXLIST_ADDR_LO,
2604 	    RL_ADDR_LO(sc->rl_ldata.rl_tx_list_addr));
2605 
2606 	/*
2607 	 * Enable transmit and receive.
2608 	 */
2609 	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
2610 
2611 	/*
2612 	 * Set the initial TX configuration.
2613 	 */
2614 	if (sc->rl_testmode) {
2615 		if (sc->rl_type == RL_8169)
2616 			CSR_WRITE_4(sc, RL_TXCFG,
2617 			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON);
2618 		else
2619 			CSR_WRITE_4(sc, RL_TXCFG,
2620 			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON_CPLUS);
2621 	} else
2622 		CSR_WRITE_4(sc, RL_TXCFG, RL_TXCFG_CONFIG);
2623 
2624 	CSR_WRITE_1(sc, RL_EARLY_TX_THRESH, 16);
2625 
2626 	/*
2627 	 * Set the initial RX configuration.
2628 	 */
2629 	re_set_rxmode(sc);
2630 
2631 #ifdef DEVICE_POLLING
2632 	/*
2633 	 * Disable interrupts if we are polling.
2634 	 */
2635 	if (ifp->if_capenable & IFCAP_POLLING)
2636 		CSR_WRITE_2(sc, RL_IMR, 0);
2637 	else	/* otherwise ... */
2638 #endif
2639 
2640 	/*
2641 	 * Enable interrupts.
2642 	 */
2643 	if (sc->rl_testmode)
2644 		CSR_WRITE_2(sc, RL_IMR, 0);
2645 	else
2646 		CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2647 	CSR_WRITE_2(sc, RL_ISR, RL_INTRS_CPLUS);
2648 
2649 	/* Set initial TX threshold */
2650 	sc->rl_txthresh = RL_TX_THRESH_INIT;
2651 
2652 	/* Start RX/TX process. */
2653 	CSR_WRITE_4(sc, RL_MISSEDPKT, 0);
2654 #ifdef notdef
2655 	/* Enable receiver and transmitter. */
2656 	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
2657 #endif
2658 
2659 #ifdef RE_TX_MODERATION
2660 	/*
2661 	 * Initialize the timer interrupt register so that
2662 	 * a timer interrupt will be generated once the timer
2663 	 * reaches a certain number of ticks. The timer is
2664 	 * reloaded on each transmit. This gives us TX interrupt
2665 	 * moderation, which dramatically improves TX frame rate.
2666 	 */
2667 	if (sc->rl_type == RL_8169)
2668 		CSR_WRITE_4(sc, RL_TIMERINT_8169, 0x800);
2669 	else
2670 		CSR_WRITE_4(sc, RL_TIMERINT, 0x400);
2671 #endif
2672 
2673 	/*
2674 	 * For 8169 gigE NICs, set the max allowed RX packet
2675 	 * size so we can receive jumbo frames.
2676 	 */
2677 	if (sc->rl_type == RL_8169)
2678 		CSR_WRITE_2(sc, RL_MAXRXPKTLEN, 16383);
2679 
2680 	if (sc->rl_testmode)
2681 		return;
2682 
2683 	mii_mediachg(mii);
2684 
2685 	CSR_WRITE_1(sc, RL_CFG1, CSR_READ_1(sc, RL_CFG1) | RL_CFG1_DRVLOAD);
2686 
2687 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
2688 	ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
2689 
2690 	sc->rl_flags &= ~RL_FLAG_LINK;
2691 	sc->rl_watchdog_timer = 0;
2692 	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
2693 }
2694 
2695 /*
2696  * Set media options.
2697  */
2698 static int
2699 re_ifmedia_upd(struct ifnet *ifp)
2700 {
2701 	struct rl_softc		*sc;
2702 	struct mii_data		*mii;
2703 	int			error;
2704 
2705 	sc = ifp->if_softc;
2706 	mii = device_get_softc(sc->rl_miibus);
2707 	RL_LOCK(sc);
2708 	error = mii_mediachg(mii);
2709 	RL_UNLOCK(sc);
2710 
2711 	return (error);
2712 }
2713 
2714 /*
2715  * Report current media status.
2716  */
2717 static void
2718 re_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
2719 {
2720 	struct rl_softc		*sc;
2721 	struct mii_data		*mii;
2722 
2723 	sc = ifp->if_softc;
2724 	mii = device_get_softc(sc->rl_miibus);
2725 
2726 	RL_LOCK(sc);
2727 	mii_pollstat(mii);
2728 	RL_UNLOCK(sc);
2729 	ifmr->ifm_active = mii->mii_media_active;
2730 	ifmr->ifm_status = mii->mii_media_status;
2731 }
2732 
2733 static int
2734 re_ioctl(struct ifnet *ifp, u_long command, caddr_t data)
2735 {
2736 	struct rl_softc		*sc = ifp->if_softc;
2737 	struct ifreq		*ifr = (struct ifreq *) data;
2738 	struct mii_data		*mii;
2739 	int			error = 0;
2740 
2741 	switch (command) {
2742 	case SIOCSIFMTU:
2743 		if (ifr->ifr_mtu < ETHERMIN || ifr->ifr_mtu > RL_JUMBO_MTU) {
2744 			error = EINVAL;
2745 			break;
2746 		}
2747 		if ((sc->rl_flags & RL_FLAG_NOJUMBO) != 0 &&
2748 		    ifr->ifr_mtu > RL_MAX_FRAMELEN) {
2749 			error = EINVAL;
2750 			break;
2751 		}
2752 		RL_LOCK(sc);
2753 		if (ifp->if_mtu != ifr->ifr_mtu)
2754 			ifp->if_mtu = ifr->ifr_mtu;
2755 		if (ifp->if_mtu > RL_TSO_MTU &&
2756 		    (ifp->if_capenable & IFCAP_TSO4) != 0) {
2757 			ifp->if_capenable &= ~IFCAP_TSO4;
2758 			ifp->if_hwassist &= ~CSUM_TSO;
2759 		}
2760 		RL_UNLOCK(sc);
2761 		break;
2762 	case SIOCSIFFLAGS:
2763 		RL_LOCK(sc);
2764 		if ((ifp->if_flags & IFF_UP) != 0) {
2765 			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
2766 				if (((ifp->if_flags ^ sc->rl_if_flags)
2767 				    & (IFF_PROMISC | IFF_ALLMULTI)) != 0)
2768 					re_set_rxmode(sc);
2769 			} else
2770 				re_init_locked(sc);
2771 		} else {
2772 			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
2773 				re_stop(sc);
2774 		}
2775 		sc->rl_if_flags = ifp->if_flags;
2776 		RL_UNLOCK(sc);
2777 		break;
2778 	case SIOCADDMULTI:
2779 	case SIOCDELMULTI:
2780 		RL_LOCK(sc);
2781 		re_set_rxmode(sc);
2782 		RL_UNLOCK(sc);
2783 		break;
2784 	case SIOCGIFMEDIA:
2785 	case SIOCSIFMEDIA:
2786 		mii = device_get_softc(sc->rl_miibus);
2787 		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, command);
2788 		break;
2789 	case SIOCSIFCAP:
2790 	    {
2791 		int mask, reinit;
2792 
2793 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
2794 		reinit = 0;
2795 #ifdef DEVICE_POLLING
2796 		if (mask & IFCAP_POLLING) {
2797 			if (ifr->ifr_reqcap & IFCAP_POLLING) {
2798 				error = ether_poll_register(re_poll, ifp);
2799 				if (error)
2800 					return(error);
2801 				RL_LOCK(sc);
2802 				/* Disable interrupts */
2803 				CSR_WRITE_2(sc, RL_IMR, 0x0000);
2804 				ifp->if_capenable |= IFCAP_POLLING;
2805 				RL_UNLOCK(sc);
2806 			} else {
2807 				error = ether_poll_deregister(ifp);
2808 				/* Enable interrupts. */
2809 				RL_LOCK(sc);
2810 				CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2811 				ifp->if_capenable &= ~IFCAP_POLLING;
2812 				RL_UNLOCK(sc);
2813 			}
2814 		}
2815 #endif /* DEVICE_POLLING */
2816 		if (mask & IFCAP_HWCSUM) {
2817 			ifp->if_capenable ^= IFCAP_HWCSUM;
2818 			if (ifp->if_capenable & IFCAP_TXCSUM)
2819 				ifp->if_hwassist |= RE_CSUM_FEATURES;
2820 			else
2821 				ifp->if_hwassist &= ~RE_CSUM_FEATURES;
2822 			reinit = 1;
2823 		}
2824 		if (mask & IFCAP_VLAN_HWTAGGING) {
2825 			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
2826 			reinit = 1;
2827 		}
2828 		if (mask & IFCAP_TSO4) {
2829 			ifp->if_capenable ^= IFCAP_TSO4;
2830 			if ((IFCAP_TSO4 & ifp->if_capenable) &&
2831 			    (IFCAP_TSO4 & ifp->if_capabilities))
2832 				ifp->if_hwassist |= CSUM_TSO;
2833 			else
2834 				ifp->if_hwassist &= ~CSUM_TSO;
2835 			if (ifp->if_mtu > RL_TSO_MTU &&
2836 			    (ifp->if_capenable & IFCAP_TSO4) != 0) {
2837 				ifp->if_capenable &= ~IFCAP_TSO4;
2838 				ifp->if_hwassist &= ~CSUM_TSO;
2839 			}
2840 		}
2841 		if ((mask & IFCAP_WOL) != 0 &&
2842 		    (ifp->if_capabilities & IFCAP_WOL) != 0) {
2843 			if ((mask & IFCAP_WOL_UCAST) != 0)
2844 				ifp->if_capenable ^= IFCAP_WOL_UCAST;
2845 			if ((mask & IFCAP_WOL_MCAST) != 0)
2846 				ifp->if_capenable ^= IFCAP_WOL_MCAST;
2847 			if ((mask & IFCAP_WOL_MAGIC) != 0)
2848 				ifp->if_capenable ^= IFCAP_WOL_MAGIC;
2849 		}
2850 		if (reinit && ifp->if_drv_flags & IFF_DRV_RUNNING)
2851 			re_init(sc);
2852 		VLAN_CAPABILITIES(ifp);
2853 	    }
2854 		break;
2855 	default:
2856 		error = ether_ioctl(ifp, command, data);
2857 		break;
2858 	}
2859 
2860 	return (error);
2861 }
2862 
2863 static void
2864 re_watchdog(struct rl_softc *sc)
2865 {
2866 	struct ifnet		*ifp;
2867 
2868 	RL_LOCK_ASSERT(sc);
2869 
2870 	if (sc->rl_watchdog_timer == 0 || --sc->rl_watchdog_timer != 0)
2871 		return;
2872 
2873 	ifp = sc->rl_ifp;
2874 	re_txeof(sc);
2875 	if (sc->rl_ldata.rl_tx_free == sc->rl_ldata.rl_tx_desc_cnt) {
2876 		if_printf(ifp, "watchdog timeout (missed Tx interrupts) "
2877 		    "-- recovering\n");
2878 		if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2879 			taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_txtask);
2880 		return;
2881 	}
2882 
2883 	if_printf(ifp, "watchdog timeout\n");
2884 	ifp->if_oerrors++;
2885 
2886 	re_rxeof(sc);
2887 	re_init_locked(sc);
2888 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2889 		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_txtask);
2890 }
2891 
2892 /*
2893  * Stop the adapter and free any mbufs allocated to the
2894  * RX and TX lists.
2895  */
2896 static void
2897 re_stop(struct rl_softc *sc)
2898 {
2899 	int			i;
2900 	struct ifnet		*ifp;
2901 	struct rl_txdesc	*txd;
2902 	struct rl_rxdesc	*rxd;
2903 
2904 	RL_LOCK_ASSERT(sc);
2905 
2906 	ifp = sc->rl_ifp;
2907 
2908 	sc->rl_watchdog_timer = 0;
2909 	callout_stop(&sc->rl_stat_callout);
2910 	ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
2911 
2912 	if ((sc->rl_flags & RL_FLAG_CMDSTOP) != 0)
2913 		CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_STOPREQ | RL_CMD_TX_ENB |
2914 		    RL_CMD_RX_ENB);
2915 	else
2916 		CSR_WRITE_1(sc, RL_COMMAND, 0x00);
2917 	DELAY(1000);
2918 	CSR_WRITE_2(sc, RL_IMR, 0x0000);
2919 	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
2920 
2921 	if (sc->rl_head != NULL) {
2922 		m_freem(sc->rl_head);
2923 		sc->rl_head = sc->rl_tail = NULL;
2924 	}
2925 
2926 	/* Free the TX list buffers. */
2927 
2928 	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
2929 		txd = &sc->rl_ldata.rl_tx_desc[i];
2930 		if (txd->tx_m != NULL) {
2931 			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
2932 			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
2933 			bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag,
2934 			    txd->tx_dmamap);
2935 			m_freem(txd->tx_m);
2936 			txd->tx_m = NULL;
2937 		}
2938 	}
2939 
2940 	/* Free the RX list buffers. */
2941 
2942 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
2943 		rxd = &sc->rl_ldata.rl_rx_desc[i];
2944 		if (rxd->rx_m != NULL) {
2945 			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
2946 			    rxd->rx_dmamap, BUS_DMASYNC_POSTREAD);
2947 			bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag,
2948 			    rxd->rx_dmamap);
2949 			m_freem(rxd->rx_m);
2950 			rxd->rx_m = NULL;
2951 		}
2952 	}
2953 }
2954 
2955 /*
2956  * Device suspend routine.  Stop the interface and save some PCI
2957  * settings in case the BIOS doesn't restore them properly on
2958  * resume.
2959  */
2960 static int
2961 re_suspend(device_t dev)
2962 {
2963 	struct rl_softc		*sc;
2964 
2965 	sc = device_get_softc(dev);
2966 
2967 	RL_LOCK(sc);
2968 	re_stop(sc);
2969 	re_setwol(sc);
2970 	sc->suspended = 1;
2971 	RL_UNLOCK(sc);
2972 
2973 	return (0);
2974 }
2975 
2976 /*
2977  * Device resume routine.  Restore some PCI settings in case the BIOS
2978  * doesn't, re-enable busmastering, and restart the interface if
2979  * appropriate.
2980  */
2981 static int
2982 re_resume(device_t dev)
2983 {
2984 	struct rl_softc		*sc;
2985 	struct ifnet		*ifp;
2986 
2987 	sc = device_get_softc(dev);
2988 
2989 	RL_LOCK(sc);
2990 
2991 	ifp = sc->rl_ifp;
2992 	/* Take controller out of sleep mode. */
2993 	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
2994 		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
2995 			CSR_WRITE_1(sc, RL_GPIO,
2996 			    CSR_READ_1(sc, RL_GPIO) | 0x01);
2997 	}
2998 
2999 	/* reinitialize interface if necessary */
3000 	if (ifp->if_flags & IFF_UP)
3001 		re_init_locked(sc);
3002 
3003 	/*
3004 	 * Clear WOL matching such that normal Rx filtering
3005 	 * wouldn't interfere with WOL patterns.
3006 	 */
3007 	re_clrwol(sc);
3008 	sc->suspended = 0;
3009 	RL_UNLOCK(sc);
3010 
3011 	return (0);
3012 }
3013 
3014 /*
3015  * Stop all chip I/O so that the kernel's probe routines don't
3016  * get confused by errant DMAs when rebooting.
3017  */
3018 static int
3019 re_shutdown(device_t dev)
3020 {
3021 	struct rl_softc		*sc;
3022 
3023 	sc = device_get_softc(dev);
3024 
3025 	RL_LOCK(sc);
3026 	re_stop(sc);
3027 	/*
3028 	 * Mark interface as down since otherwise we will panic if
3029 	 * interrupt comes in later on, which can happen in some
3030 	 * cases.
3031 	 */
3032 	sc->rl_ifp->if_flags &= ~IFF_UP;
3033 	re_setwol(sc);
3034 	RL_UNLOCK(sc);
3035 
3036 	return (0);
3037 }
3038 
3039 static void
3040 re_setwol(struct rl_softc *sc)
3041 {
3042 	struct ifnet		*ifp;
3043 	int			pmc;
3044 	uint16_t		pmstat;
3045 	uint8_t			v;
3046 
3047 	RL_LOCK_ASSERT(sc);
3048 
3049 	if (pci_find_extcap(sc->rl_dev, PCIY_PMG, &pmc) != 0)
3050 		return;
3051 
3052 	ifp = sc->rl_ifp;
3053 	/* Put controller into sleep mode. */
3054 	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
3055 		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
3056 			CSR_WRITE_1(sc, RL_GPIO,
3057 			    CSR_READ_1(sc, RL_GPIO) & ~0x01);
3058 	}
3059 	if ((ifp->if_capenable & IFCAP_WOL) != 0 &&
3060 	    (sc->rl_flags & RL_FLAG_WOLRXENB) != 0)
3061 		CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RX_ENB);
3062 	/* Enable config register write. */
3063 	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
3064 
3065 	/* Enable PME. */
3066 	v = CSR_READ_1(sc, RL_CFG1);
3067 	v &= ~RL_CFG1_PME;
3068 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3069 		v |= RL_CFG1_PME;
3070 	CSR_WRITE_1(sc, RL_CFG1, v);
3071 
3072 	v = CSR_READ_1(sc, RL_CFG3);
3073 	v &= ~(RL_CFG3_WOL_LINK | RL_CFG3_WOL_MAGIC);
3074 	if ((ifp->if_capenable & IFCAP_WOL_MAGIC) != 0)
3075 		v |= RL_CFG3_WOL_MAGIC;
3076 	CSR_WRITE_1(sc, RL_CFG3, v);
3077 
3078 	/* Config register write done. */
3079 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3080 
3081 	v = CSR_READ_1(sc, RL_CFG5);
3082 	v &= ~(RL_CFG5_WOL_BCAST | RL_CFG5_WOL_MCAST | RL_CFG5_WOL_UCAST);
3083 	v &= ~RL_CFG5_WOL_LANWAKE;
3084 	if ((ifp->if_capenable & IFCAP_WOL_UCAST) != 0)
3085 		v |= RL_CFG5_WOL_UCAST;
3086 	if ((ifp->if_capenable & IFCAP_WOL_MCAST) != 0)
3087 		v |= RL_CFG5_WOL_MCAST | RL_CFG5_WOL_BCAST;
3088 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3089 		v |= RL_CFG5_WOL_LANWAKE;
3090 	CSR_WRITE_1(sc, RL_CFG5, v);
3091 
3092 	/*
3093 	 * It seems that hardware resets its link speed to 100Mbps in
3094 	 * power down mode so switching to 100Mbps in driver is not
3095 	 * needed.
3096 	 */
3097 
3098 	/* Request PME if WOL is requested. */
3099 	pmstat = pci_read_config(sc->rl_dev, pmc + PCIR_POWER_STATUS, 2);
3100 	pmstat &= ~(PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE);
3101 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3102 		pmstat |= PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE;
3103 	pci_write_config(sc->rl_dev, pmc + PCIR_POWER_STATUS, pmstat, 2);
3104 }
3105 
3106 static void
3107 re_clrwol(struct rl_softc *sc)
3108 {
3109 	int			pmc;
3110 	uint8_t			v;
3111 
3112 	RL_LOCK_ASSERT(sc);
3113 
3114 	if (pci_find_extcap(sc->rl_dev, PCIY_PMG, &pmc) != 0)
3115 		return;
3116 
3117 	/* Enable config register write. */
3118 	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
3119 
3120 	v = CSR_READ_1(sc, RL_CFG3);
3121 	v &= ~(RL_CFG3_WOL_LINK | RL_CFG3_WOL_MAGIC);
3122 	CSR_WRITE_1(sc, RL_CFG3, v);
3123 
3124 	/* Config register write done. */
3125 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3126 
3127 	v = CSR_READ_1(sc, RL_CFG5);
3128 	v &= ~(RL_CFG5_WOL_BCAST | RL_CFG5_WOL_MCAST | RL_CFG5_WOL_UCAST);
3129 	v &= ~RL_CFG5_WOL_LANWAKE;
3130 	CSR_WRITE_1(sc, RL_CFG5, v);
3131 }
3132