xref: /freebsd/sys/dev/re/if_re.c (revision d8b878873e7aa8df1972cc6a642804b17eb61087)
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/8103E PCIe 10/100baseTX" },
176 	{ RT_VENDORID, RT_DEVICEID_8168, 0,
177 	    "RealTek 8168/8111 B/C/CP/D/DP/E PCIe Gigabit Ethernet" },
178 	{ RT_VENDORID, RT_DEVICEID_8169, 0,
179 	    "RealTek 8169/8169S/8169SB(L)/8110S/8110SB(L) Gigabit Ethernet" },
180 	{ RT_VENDORID, RT_DEVICEID_8169SC, 0,
181 	    "RealTek 8169SC/8110SC Single-chip Gigabit Ethernet" },
182 	{ COREGA_VENDORID, COREGA_DEVICEID_CGLAPCIGT, 0,
183 	    "Corega CG-LAPCIGT (RTL8169S) Gigabit Ethernet" },
184 	{ LINKSYS_VENDORID, LINKSYS_DEVICEID_EG1032, 0,
185 	    "Linksys EG1032 (RTL8169S) Gigabit Ethernet" },
186 	{ USR_VENDORID, USR_DEVICEID_997902, 0,
187 	    "US Robotics 997902 (RTL8169S) Gigabit Ethernet" }
188 };
189 
190 static struct rl_hwrev re_hwrevs[] = {
191 	{ RL_HWREV_8139, RL_8139,  "" },
192 	{ RL_HWREV_8139A, RL_8139, "A" },
193 	{ RL_HWREV_8139AG, RL_8139, "A-G" },
194 	{ RL_HWREV_8139B, RL_8139, "B" },
195 	{ RL_HWREV_8130, RL_8139, "8130" },
196 	{ RL_HWREV_8139C, RL_8139, "C" },
197 	{ RL_HWREV_8139D, RL_8139, "8139D/8100B/8100C" },
198 	{ RL_HWREV_8139CPLUS, RL_8139CPLUS, "C+"},
199 	{ RL_HWREV_8168_SPIN1, RL_8169, "8168"},
200 	{ RL_HWREV_8169, RL_8169, "8169"},
201 	{ RL_HWREV_8169S, RL_8169, "8169S"},
202 	{ RL_HWREV_8110S, RL_8169, "8110S"},
203 	{ RL_HWREV_8169_8110SB, RL_8169, "8169SB/8110SB"},
204 	{ RL_HWREV_8169_8110SC, RL_8169, "8169SC/8110SC"},
205 	{ RL_HWREV_8169_8110SBL, RL_8169, "8169SBL/8110SBL"},
206 	{ RL_HWREV_8169_8110SCE, RL_8169, "8169SC/8110SC"},
207 	{ RL_HWREV_8100, RL_8139, "8100"},
208 	{ RL_HWREV_8101, RL_8139, "8101"},
209 	{ RL_HWREV_8100E, RL_8169, "8100E"},
210 	{ RL_HWREV_8101E, RL_8169, "8101E"},
211 	{ RL_HWREV_8102E, RL_8169, "8102E"},
212 	{ RL_HWREV_8102EL, RL_8169, "8102EL"},
213 	{ RL_HWREV_8102EL_SPIN1, RL_8169, "8102EL"},
214 	{ RL_HWREV_8103E, RL_8169, "8103E"},
215 	{ RL_HWREV_8168_SPIN2, RL_8169, "8168"},
216 	{ RL_HWREV_8168_SPIN3, RL_8169, "8168"},
217 	{ RL_HWREV_8168C, RL_8169, "8168C/8111C"},
218 	{ RL_HWREV_8168C_SPIN2, RL_8169, "8168C/8111C"},
219 	{ RL_HWREV_8168CP, RL_8169, "8168CP/8111CP"},
220 	{ RL_HWREV_8168D, RL_8169, "8168D/8111D"},
221 	{ RL_HWREV_8168DP, RL_8169, "8168DP/8111DP"},
222 	{ RL_HWREV_8168E, RL_8169, "8168E/8111E"},
223 	{ 0, 0, NULL }
224 };
225 
226 static int re_probe		(device_t);
227 static int re_attach		(device_t);
228 static int re_detach		(device_t);
229 
230 static int re_encap		(struct rl_softc *, struct mbuf **);
231 
232 static void re_dma_map_addr	(void *, bus_dma_segment_t *, int, int);
233 static int re_allocmem		(device_t, struct rl_softc *);
234 static __inline void re_discard_rxbuf
235 				(struct rl_softc *, int);
236 static int re_newbuf		(struct rl_softc *, int);
237 static int re_rx_list_init	(struct rl_softc *);
238 static int re_tx_list_init	(struct rl_softc *);
239 #ifdef RE_FIXUP_RX
240 static __inline void re_fixup_rx
241 				(struct mbuf *);
242 #endif
243 static int re_rxeof		(struct rl_softc *, int *);
244 static void re_txeof		(struct rl_softc *);
245 #ifdef DEVICE_POLLING
246 static int re_poll		(struct ifnet *, enum poll_cmd, int);
247 static int re_poll_locked	(struct ifnet *, enum poll_cmd, int);
248 #endif
249 static int re_intr		(void *);
250 static void re_tick		(void *);
251 static void re_tx_task		(void *, int);
252 static void re_int_task		(void *, int);
253 static void re_start		(struct ifnet *);
254 static int re_ioctl		(struct ifnet *, u_long, caddr_t);
255 static void re_init		(void *);
256 static void re_init_locked	(struct rl_softc *);
257 static void re_stop		(struct rl_softc *);
258 static void re_watchdog		(struct rl_softc *);
259 static int re_suspend		(device_t);
260 static int re_resume		(device_t);
261 static int re_shutdown		(device_t);
262 static int re_ifmedia_upd	(struct ifnet *);
263 static void re_ifmedia_sts	(struct ifnet *, struct ifmediareq *);
264 
265 static void re_eeprom_putbyte	(struct rl_softc *, int);
266 static void re_eeprom_getword	(struct rl_softc *, int, u_int16_t *);
267 static void re_read_eeprom	(struct rl_softc *, caddr_t, int, int);
268 static int re_gmii_readreg	(device_t, int, int);
269 static int re_gmii_writereg	(device_t, int, int, int);
270 
271 static int re_miibus_readreg	(device_t, int, int);
272 static int re_miibus_writereg	(device_t, int, int, int);
273 static void re_miibus_statchg	(device_t);
274 
275 static void re_set_rxmode		(struct rl_softc *);
276 static void re_reset		(struct rl_softc *);
277 static void re_setwol		(struct rl_softc *);
278 static void re_clrwol		(struct rl_softc *);
279 
280 #ifdef RE_DIAG
281 static int re_diag		(struct rl_softc *);
282 #endif
283 
284 static device_method_t re_methods[] = {
285 	/* Device interface */
286 	DEVMETHOD(device_probe,		re_probe),
287 	DEVMETHOD(device_attach,	re_attach),
288 	DEVMETHOD(device_detach,	re_detach),
289 	DEVMETHOD(device_suspend,	re_suspend),
290 	DEVMETHOD(device_resume,	re_resume),
291 	DEVMETHOD(device_shutdown,	re_shutdown),
292 
293 	/* bus interface */
294 	DEVMETHOD(bus_print_child,	bus_generic_print_child),
295 	DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
296 
297 	/* MII interface */
298 	DEVMETHOD(miibus_readreg,	re_miibus_readreg),
299 	DEVMETHOD(miibus_writereg,	re_miibus_writereg),
300 	DEVMETHOD(miibus_statchg,	re_miibus_statchg),
301 
302 	{ 0, 0 }
303 };
304 
305 static driver_t re_driver = {
306 	"re",
307 	re_methods,
308 	sizeof(struct rl_softc)
309 };
310 
311 static devclass_t re_devclass;
312 
313 DRIVER_MODULE(re, pci, re_driver, re_devclass, 0, 0);
314 DRIVER_MODULE(miibus, re, miibus_driver, miibus_devclass, 0, 0);
315 
316 #define EE_SET(x)					\
317 	CSR_WRITE_1(sc, RL_EECMD,			\
318 		CSR_READ_1(sc, RL_EECMD) | x)
319 
320 #define EE_CLR(x)					\
321 	CSR_WRITE_1(sc, RL_EECMD,			\
322 		CSR_READ_1(sc, RL_EECMD) & ~x)
323 
324 /*
325  * Send a read command and address to the EEPROM, check for ACK.
326  */
327 static void
328 re_eeprom_putbyte(struct rl_softc *sc, int addr)
329 {
330 	int			d, i;
331 
332 	d = addr | (RL_9346_READ << sc->rl_eewidth);
333 
334 	/*
335 	 * Feed in each bit and strobe the clock.
336 	 */
337 
338 	for (i = 1 << (sc->rl_eewidth + 3); i; i >>= 1) {
339 		if (d & i) {
340 			EE_SET(RL_EE_DATAIN);
341 		} else {
342 			EE_CLR(RL_EE_DATAIN);
343 		}
344 		DELAY(100);
345 		EE_SET(RL_EE_CLK);
346 		DELAY(150);
347 		EE_CLR(RL_EE_CLK);
348 		DELAY(100);
349 	}
350 }
351 
352 /*
353  * Read a word of data stored in the EEPROM at address 'addr.'
354  */
355 static void
356 re_eeprom_getword(struct rl_softc *sc, int addr, u_int16_t *dest)
357 {
358 	int			i;
359 	u_int16_t		word = 0;
360 
361 	/*
362 	 * Send address of word we want to read.
363 	 */
364 	re_eeprom_putbyte(sc, addr);
365 
366 	/*
367 	 * Start reading bits from EEPROM.
368 	 */
369 	for (i = 0x8000; i; i >>= 1) {
370 		EE_SET(RL_EE_CLK);
371 		DELAY(100);
372 		if (CSR_READ_1(sc, RL_EECMD) & RL_EE_DATAOUT)
373 			word |= i;
374 		EE_CLR(RL_EE_CLK);
375 		DELAY(100);
376 	}
377 
378 	*dest = word;
379 }
380 
381 /*
382  * Read a sequence of words from the EEPROM.
383  */
384 static void
385 re_read_eeprom(struct rl_softc *sc, caddr_t dest, int off, int cnt)
386 {
387 	int			i;
388 	u_int16_t		word = 0, *ptr;
389 
390 	CSR_SETBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM);
391 
392         DELAY(100);
393 
394 	for (i = 0; i < cnt; i++) {
395 		CSR_SETBIT_1(sc, RL_EECMD, RL_EE_SEL);
396 		re_eeprom_getword(sc, off + i, &word);
397 		CSR_CLRBIT_1(sc, RL_EECMD, RL_EE_SEL);
398 		ptr = (u_int16_t *)(dest + (i * 2));
399                 *ptr = word;
400 	}
401 
402 	CSR_CLRBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM);
403 }
404 
405 static int
406 re_gmii_readreg(device_t dev, int phy, int reg)
407 {
408 	struct rl_softc		*sc;
409 	u_int32_t		rval;
410 	int			i;
411 
412 	if (phy != 1)
413 		return (0);
414 
415 	sc = device_get_softc(dev);
416 
417 	/* Let the rgephy driver read the GMEDIASTAT register */
418 
419 	if (reg == RL_GMEDIASTAT) {
420 		rval = CSR_READ_1(sc, RL_GMEDIASTAT);
421 		return (rval);
422 	}
423 
424 	CSR_WRITE_4(sc, RL_PHYAR, reg << 16);
425 	DELAY(1000);
426 
427 	for (i = 0; i < RL_PHY_TIMEOUT; i++) {
428 		rval = CSR_READ_4(sc, RL_PHYAR);
429 		if (rval & RL_PHYAR_BUSY)
430 			break;
431 		DELAY(100);
432 	}
433 
434 	if (i == RL_PHY_TIMEOUT) {
435 		device_printf(sc->rl_dev, "PHY read failed\n");
436 		return (0);
437 	}
438 
439 	return (rval & RL_PHYAR_PHYDATA);
440 }
441 
442 static int
443 re_gmii_writereg(device_t dev, int phy, int reg, int data)
444 {
445 	struct rl_softc		*sc;
446 	u_int32_t		rval;
447 	int			i;
448 
449 	sc = device_get_softc(dev);
450 
451 	CSR_WRITE_4(sc, RL_PHYAR, (reg << 16) |
452 	    (data & RL_PHYAR_PHYDATA) | RL_PHYAR_BUSY);
453 	DELAY(1000);
454 
455 	for (i = 0; i < RL_PHY_TIMEOUT; i++) {
456 		rval = CSR_READ_4(sc, RL_PHYAR);
457 		if (!(rval & RL_PHYAR_BUSY))
458 			break;
459 		DELAY(100);
460 	}
461 
462 	if (i == RL_PHY_TIMEOUT) {
463 		device_printf(sc->rl_dev, "PHY write failed\n");
464 		return (0);
465 	}
466 
467 	return (0);
468 }
469 
470 static int
471 re_miibus_readreg(device_t dev, int phy, int reg)
472 {
473 	struct rl_softc		*sc;
474 	u_int16_t		rval = 0;
475 	u_int16_t		re8139_reg = 0;
476 
477 	sc = device_get_softc(dev);
478 
479 	if (sc->rl_type == RL_8169) {
480 		rval = re_gmii_readreg(dev, phy, reg);
481 		return (rval);
482 	}
483 
484 	/* Pretend the internal PHY is only at address 0 */
485 	if (phy) {
486 		return (0);
487 	}
488 	switch (reg) {
489 	case MII_BMCR:
490 		re8139_reg = RL_BMCR;
491 		break;
492 	case MII_BMSR:
493 		re8139_reg = RL_BMSR;
494 		break;
495 	case MII_ANAR:
496 		re8139_reg = RL_ANAR;
497 		break;
498 	case MII_ANER:
499 		re8139_reg = RL_ANER;
500 		break;
501 	case MII_ANLPAR:
502 		re8139_reg = RL_LPAR;
503 		break;
504 	case MII_PHYIDR1:
505 	case MII_PHYIDR2:
506 		return (0);
507 	/*
508 	 * Allow the rlphy driver to read the media status
509 	 * register. If we have a link partner which does not
510 	 * support NWAY, this is the register which will tell
511 	 * us the results of parallel detection.
512 	 */
513 	case RL_MEDIASTAT:
514 		rval = CSR_READ_1(sc, RL_MEDIASTAT);
515 		return (rval);
516 	default:
517 		device_printf(sc->rl_dev, "bad phy register\n");
518 		return (0);
519 	}
520 	rval = CSR_READ_2(sc, re8139_reg);
521 	if (sc->rl_type == RL_8139CPLUS && re8139_reg == RL_BMCR) {
522 		/* 8139C+ has different bit layout. */
523 		rval &= ~(BMCR_LOOP | BMCR_ISO);
524 	}
525 	return (rval);
526 }
527 
528 static int
529 re_miibus_writereg(device_t dev, int phy, int reg, int data)
530 {
531 	struct rl_softc		*sc;
532 	u_int16_t		re8139_reg = 0;
533 	int			rval = 0;
534 
535 	sc = device_get_softc(dev);
536 
537 	if (sc->rl_type == RL_8169) {
538 		rval = re_gmii_writereg(dev, phy, reg, data);
539 		return (rval);
540 	}
541 
542 	/* Pretend the internal PHY is only at address 0 */
543 	if (phy)
544 		return (0);
545 
546 	switch (reg) {
547 	case MII_BMCR:
548 		re8139_reg = RL_BMCR;
549 		if (sc->rl_type == RL_8139CPLUS) {
550 			/* 8139C+ has different bit layout. */
551 			data &= ~(BMCR_LOOP | BMCR_ISO);
552 		}
553 		break;
554 	case MII_BMSR:
555 		re8139_reg = RL_BMSR;
556 		break;
557 	case MII_ANAR:
558 		re8139_reg = RL_ANAR;
559 		break;
560 	case MII_ANER:
561 		re8139_reg = RL_ANER;
562 		break;
563 	case MII_ANLPAR:
564 		re8139_reg = RL_LPAR;
565 		break;
566 	case MII_PHYIDR1:
567 	case MII_PHYIDR2:
568 		return (0);
569 		break;
570 	default:
571 		device_printf(sc->rl_dev, "bad phy register\n");
572 		return (0);
573 	}
574 	CSR_WRITE_2(sc, re8139_reg, data);
575 	return (0);
576 }
577 
578 static void
579 re_miibus_statchg(device_t dev)
580 {
581 	struct rl_softc		*sc;
582 	struct ifnet		*ifp;
583 	struct mii_data		*mii;
584 
585 	sc = device_get_softc(dev);
586 	mii = device_get_softc(sc->rl_miibus);
587 	ifp = sc->rl_ifp;
588 	if (mii == NULL || ifp == NULL ||
589 	    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
590 		return;
591 
592 	sc->rl_flags &= ~RL_FLAG_LINK;
593 	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
594 	    (IFM_ACTIVE | IFM_AVALID)) {
595 		switch (IFM_SUBTYPE(mii->mii_media_active)) {
596 		case IFM_10_T:
597 		case IFM_100_TX:
598 			sc->rl_flags |= RL_FLAG_LINK;
599 			break;
600 		case IFM_1000_T:
601 			if ((sc->rl_flags & RL_FLAG_FASTETHER) != 0)
602 				break;
603 			sc->rl_flags |= RL_FLAG_LINK;
604 			break;
605 		default:
606 			break;
607 		}
608 	}
609 	/*
610 	 * RealTek controllers does not provide any interface to
611 	 * Tx/Rx MACs for resolved speed, duplex and flow-control
612 	 * parameters.
613 	 */
614 }
615 
616 /*
617  * Set the RX configuration and 64-bit multicast hash filter.
618  */
619 static void
620 re_set_rxmode(struct rl_softc *sc)
621 {
622 	struct ifnet		*ifp;
623 	struct ifmultiaddr	*ifma;
624 	uint32_t		hashes[2] = { 0, 0 };
625 	uint32_t		h, rxfilt;
626 
627 	RL_LOCK_ASSERT(sc);
628 
629 	ifp = sc->rl_ifp;
630 
631 	rxfilt = RL_RXCFG_CONFIG | RL_RXCFG_RX_INDIV | RL_RXCFG_RX_BROAD;
632 
633 	if (ifp->if_flags & (IFF_ALLMULTI | IFF_PROMISC)) {
634 		if (ifp->if_flags & IFF_PROMISC)
635 			rxfilt |= RL_RXCFG_RX_ALLPHYS;
636 		/*
637 		 * Unlike other hardwares, we have to explicitly set
638 		 * RL_RXCFG_RX_MULTI to receive multicast frames in
639 		 * promiscuous mode.
640 		 */
641 		rxfilt |= RL_RXCFG_RX_MULTI;
642 		hashes[0] = hashes[1] = 0xffffffff;
643 		goto done;
644 	}
645 
646 	if_maddr_rlock(ifp);
647 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
648 		if (ifma->ifma_addr->sa_family != AF_LINK)
649 			continue;
650 		h = ether_crc32_be(LLADDR((struct sockaddr_dl *)
651 		    ifma->ifma_addr), ETHER_ADDR_LEN) >> 26;
652 		if (h < 32)
653 			hashes[0] |= (1 << h);
654 		else
655 			hashes[1] |= (1 << (h - 32));
656 	}
657 	if_maddr_runlock(ifp);
658 
659 	if (hashes[0] != 0 || hashes[1] != 0) {
660 		/*
661 		 * For some unfathomable reason, RealTek decided to
662 		 * reverse the order of the multicast hash registers
663 		 * in the PCI Express parts.  This means we have to
664 		 * write the hash pattern in reverse order for those
665 		 * devices.
666 		 */
667 		if ((sc->rl_flags & RL_FLAG_PCIE) != 0) {
668 			h = bswap32(hashes[0]);
669 			hashes[0] = bswap32(hashes[1]);
670 			hashes[1] = h;
671 		}
672 		rxfilt |= RL_RXCFG_RX_MULTI;
673 	}
674 
675 done:
676 	CSR_WRITE_4(sc, RL_MAR0, hashes[0]);
677 	CSR_WRITE_4(sc, RL_MAR4, hashes[1]);
678 	CSR_WRITE_4(sc, RL_RXCFG, rxfilt);
679 }
680 
681 static void
682 re_reset(struct rl_softc *sc)
683 {
684 	int			i;
685 
686 	RL_LOCK_ASSERT(sc);
687 
688 	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RESET);
689 
690 	for (i = 0; i < RL_TIMEOUT; i++) {
691 		DELAY(10);
692 		if (!(CSR_READ_1(sc, RL_COMMAND) & RL_CMD_RESET))
693 			break;
694 	}
695 	if (i == RL_TIMEOUT)
696 		device_printf(sc->rl_dev, "reset never completed!\n");
697 
698 	if ((sc->rl_flags & RL_FLAG_MACRESET) != 0)
699 		CSR_WRITE_1(sc, 0x82, 1);
700 	if (sc->rl_hwrev == RL_HWREV_8169S)
701 		re_gmii_writereg(sc->rl_dev, 1, 0x0b, 0);
702 }
703 
704 #ifdef RE_DIAG
705 
706 /*
707  * The following routine is designed to test for a defect on some
708  * 32-bit 8169 cards. Some of these NICs have the REQ64# and ACK64#
709  * lines connected to the bus, however for a 32-bit only card, they
710  * should be pulled high. The result of this defect is that the
711  * NIC will not work right if you plug it into a 64-bit slot: DMA
712  * operations will be done with 64-bit transfers, which will fail
713  * because the 64-bit data lines aren't connected.
714  *
715  * There's no way to work around this (short of talking a soldering
716  * iron to the board), however we can detect it. The method we use
717  * here is to put the NIC into digital loopback mode, set the receiver
718  * to promiscuous mode, and then try to send a frame. We then compare
719  * the frame data we sent to what was received. If the data matches,
720  * then the NIC is working correctly, otherwise we know the user has
721  * a defective NIC which has been mistakenly plugged into a 64-bit PCI
722  * slot. In the latter case, there's no way the NIC can work correctly,
723  * so we print out a message on the console and abort the device attach.
724  */
725 
726 static int
727 re_diag(struct rl_softc *sc)
728 {
729 	struct ifnet		*ifp = sc->rl_ifp;
730 	struct mbuf		*m0;
731 	struct ether_header	*eh;
732 	struct rl_desc		*cur_rx;
733 	u_int16_t		status;
734 	u_int32_t		rxstat;
735 	int			total_len, i, error = 0, phyaddr;
736 	u_int8_t		dst[] = { 0x00, 'h', 'e', 'l', 'l', 'o' };
737 	u_int8_t		src[] = { 0x00, 'w', 'o', 'r', 'l', 'd' };
738 
739 	/* Allocate a single mbuf */
740 	MGETHDR(m0, M_DONTWAIT, MT_DATA);
741 	if (m0 == NULL)
742 		return (ENOBUFS);
743 
744 	RL_LOCK(sc);
745 
746 	/*
747 	 * Initialize the NIC in test mode. This sets the chip up
748 	 * so that it can send and receive frames, but performs the
749 	 * following special functions:
750 	 * - Puts receiver in promiscuous mode
751 	 * - Enables digital loopback mode
752 	 * - Leaves interrupts turned off
753 	 */
754 
755 	ifp->if_flags |= IFF_PROMISC;
756 	sc->rl_testmode = 1;
757 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
758 	re_init_locked(sc);
759 	sc->rl_flags |= RL_FLAG_LINK;
760 	if (sc->rl_type == RL_8169)
761 		phyaddr = 1;
762 	else
763 		phyaddr = 0;
764 
765 	re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_RESET);
766 	for (i = 0; i < RL_TIMEOUT; i++) {
767 		status = re_miibus_readreg(sc->rl_dev, phyaddr, MII_BMCR);
768 		if (!(status & BMCR_RESET))
769 			break;
770 	}
771 
772 	re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_LOOP);
773 	CSR_WRITE_2(sc, RL_ISR, RL_INTRS);
774 
775 	DELAY(100000);
776 
777 	/* Put some data in the mbuf */
778 
779 	eh = mtod(m0, struct ether_header *);
780 	bcopy ((char *)&dst, eh->ether_dhost, ETHER_ADDR_LEN);
781 	bcopy ((char *)&src, eh->ether_shost, ETHER_ADDR_LEN);
782 	eh->ether_type = htons(ETHERTYPE_IP);
783 	m0->m_pkthdr.len = m0->m_len = ETHER_MIN_LEN - ETHER_CRC_LEN;
784 
785 	/*
786 	 * Queue the packet, start transmission.
787 	 * Note: IF_HANDOFF() ultimately calls re_start() for us.
788 	 */
789 
790 	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
791 	RL_UNLOCK(sc);
792 	/* XXX: re_diag must not be called when in ALTQ mode */
793 	IF_HANDOFF(&ifp->if_snd, m0, ifp);
794 	RL_LOCK(sc);
795 	m0 = NULL;
796 
797 	/* Wait for it to propagate through the chip */
798 
799 	DELAY(100000);
800 	for (i = 0; i < RL_TIMEOUT; i++) {
801 		status = CSR_READ_2(sc, RL_ISR);
802 		CSR_WRITE_2(sc, RL_ISR, status);
803 		if ((status & (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK)) ==
804 		    (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK))
805 			break;
806 		DELAY(10);
807 	}
808 
809 	if (i == RL_TIMEOUT) {
810 		device_printf(sc->rl_dev,
811 		    "diagnostic failed, failed to receive packet in"
812 		    " loopback mode\n");
813 		error = EIO;
814 		goto done;
815 	}
816 
817 	/*
818 	 * The packet should have been dumped into the first
819 	 * entry in the RX DMA ring. Grab it from there.
820 	 */
821 
822 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
823 	    sc->rl_ldata.rl_rx_list_map,
824 	    BUS_DMASYNC_POSTREAD);
825 	bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag,
826 	    sc->rl_ldata.rl_rx_desc[0].rx_dmamap,
827 	    BUS_DMASYNC_POSTREAD);
828 	bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag,
829 	    sc->rl_ldata.rl_rx_desc[0].rx_dmamap);
830 
831 	m0 = sc->rl_ldata.rl_rx_desc[0].rx_m;
832 	sc->rl_ldata.rl_rx_desc[0].rx_m = NULL;
833 	eh = mtod(m0, struct ether_header *);
834 
835 	cur_rx = &sc->rl_ldata.rl_rx_list[0];
836 	total_len = RL_RXBYTES(cur_rx);
837 	rxstat = le32toh(cur_rx->rl_cmdstat);
838 
839 	if (total_len != ETHER_MIN_LEN) {
840 		device_printf(sc->rl_dev,
841 		    "diagnostic failed, received short packet\n");
842 		error = EIO;
843 		goto done;
844 	}
845 
846 	/* Test that the received packet data matches what we sent. */
847 
848 	if (bcmp((char *)&eh->ether_dhost, (char *)&dst, ETHER_ADDR_LEN) ||
849 	    bcmp((char *)&eh->ether_shost, (char *)&src, ETHER_ADDR_LEN) ||
850 	    ntohs(eh->ether_type) != ETHERTYPE_IP) {
851 		device_printf(sc->rl_dev, "WARNING, DMA FAILURE!\n");
852 		device_printf(sc->rl_dev, "expected TX data: %6D/%6D/0x%x\n",
853 		    dst, ":", src, ":", ETHERTYPE_IP);
854 		device_printf(sc->rl_dev, "received RX data: %6D/%6D/0x%x\n",
855 		    eh->ether_dhost, ":",  eh->ether_shost, ":",
856 		    ntohs(eh->ether_type));
857 		device_printf(sc->rl_dev, "You may have a defective 32-bit "
858 		    "NIC plugged into a 64-bit PCI slot.\n");
859 		device_printf(sc->rl_dev, "Please re-install the NIC in a "
860 		    "32-bit slot for proper operation.\n");
861 		device_printf(sc->rl_dev, "Read the re(4) man page for more "
862 		    "details.\n");
863 		error = EIO;
864 	}
865 
866 done:
867 	/* Turn interface off, release resources */
868 
869 	sc->rl_testmode = 0;
870 	sc->rl_flags &= ~RL_FLAG_LINK;
871 	ifp->if_flags &= ~IFF_PROMISC;
872 	re_stop(sc);
873 	if (m0 != NULL)
874 		m_freem(m0);
875 
876 	RL_UNLOCK(sc);
877 
878 	return (error);
879 }
880 
881 #endif
882 
883 /*
884  * Probe for a RealTek 8139C+/8169/8110 chip. Check the PCI vendor and device
885  * IDs against our list and return a device name if we find a match.
886  */
887 static int
888 re_probe(device_t dev)
889 {
890 	struct rl_type		*t;
891 	uint16_t		devid, vendor;
892 	uint16_t		revid, sdevid;
893 	int			i;
894 
895 	vendor = pci_get_vendor(dev);
896 	devid = pci_get_device(dev);
897 	revid = pci_get_revid(dev);
898 	sdevid = pci_get_subdevice(dev);
899 
900 	if (vendor == LINKSYS_VENDORID && devid == LINKSYS_DEVICEID_EG1032) {
901 		if (sdevid != LINKSYS_SUBDEVICE_EG1032_REV3) {
902 			/*
903 			 * Only attach to rev. 3 of the Linksys EG1032 adapter.
904 			 * Rev. 2 is supported by sk(4).
905 			 */
906 			return (ENXIO);
907 		}
908 	}
909 
910 	if (vendor == RT_VENDORID && devid == RT_DEVICEID_8139) {
911 		if (revid != 0x20) {
912 			/* 8139, let rl(4) take care of this device. */
913 			return (ENXIO);
914 		}
915 	}
916 
917 	t = re_devs;
918 	for (i = 0; i < sizeof(re_devs) / sizeof(re_devs[0]); i++, t++) {
919 		if (vendor == t->rl_vid && devid == t->rl_did) {
920 			device_set_desc(dev, t->rl_name);
921 			return (BUS_PROBE_DEFAULT);
922 		}
923 	}
924 
925 	return (ENXIO);
926 }
927 
928 /*
929  * Map a single buffer address.
930  */
931 
932 static void
933 re_dma_map_addr(void *arg, bus_dma_segment_t *segs, int nseg, int error)
934 {
935 	bus_addr_t		*addr;
936 
937 	if (error)
938 		return;
939 
940 	KASSERT(nseg == 1, ("too many DMA segments, %d should be 1", nseg));
941 	addr = arg;
942 	*addr = segs->ds_addr;
943 }
944 
945 static int
946 re_allocmem(device_t dev, struct rl_softc *sc)
947 {
948 	bus_size_t		rx_list_size, tx_list_size;
949 	int			error;
950 	int			i;
951 
952 	rx_list_size = sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc);
953 	tx_list_size = sc->rl_ldata.rl_tx_desc_cnt * sizeof(struct rl_desc);
954 
955 	/*
956 	 * Allocate the parent bus DMA tag appropriate for PCI.
957 	 * In order to use DAC, RL_CPLUSCMD_PCI_DAC bit of RL_CPLUS_CMD
958 	 * register should be set. However some RealTek chips are known
959 	 * to be buggy on DAC handling, therefore disable DAC by limiting
960 	 * DMA address space to 32bit. PCIe variants of RealTek chips
961 	 * may not have the limitation but I took safer path.
962 	 */
963 	error = bus_dma_tag_create(bus_get_dma_tag(dev), 1, 0,
964 	    BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL, NULL,
965 	    BUS_SPACE_MAXSIZE_32BIT, 0, BUS_SPACE_MAXSIZE_32BIT, 0,
966 	    NULL, NULL, &sc->rl_parent_tag);
967 	if (error) {
968 		device_printf(dev, "could not allocate parent DMA tag\n");
969 		return (error);
970 	}
971 
972 	/*
973 	 * Allocate map for TX mbufs.
974 	 */
975 	error = bus_dma_tag_create(sc->rl_parent_tag, 1, 0,
976 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL,
977 	    NULL, MCLBYTES * RL_NTXSEGS, RL_NTXSEGS, 4096, 0,
978 	    NULL, NULL, &sc->rl_ldata.rl_tx_mtag);
979 	if (error) {
980 		device_printf(dev, "could not allocate TX DMA tag\n");
981 		return (error);
982 	}
983 
984 	/*
985 	 * Allocate map for RX mbufs.
986 	 */
987 
988 	error = bus_dma_tag_create(sc->rl_parent_tag, sizeof(uint64_t), 0,
989 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
990 	    MCLBYTES, 1, MCLBYTES, 0, NULL, NULL, &sc->rl_ldata.rl_rx_mtag);
991 	if (error) {
992 		device_printf(dev, "could not allocate RX DMA tag\n");
993 		return (error);
994 	}
995 
996 	/*
997 	 * Allocate map for TX descriptor list.
998 	 */
999 	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
1000 	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1001 	    NULL, tx_list_size, 1, tx_list_size, 0,
1002 	    NULL, NULL, &sc->rl_ldata.rl_tx_list_tag);
1003 	if (error) {
1004 		device_printf(dev, "could not allocate TX DMA ring tag\n");
1005 		return (error);
1006 	}
1007 
1008 	/* Allocate DMA'able memory for the TX ring */
1009 
1010 	error = bus_dmamem_alloc(sc->rl_ldata.rl_tx_list_tag,
1011 	    (void **)&sc->rl_ldata.rl_tx_list,
1012 	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1013 	    &sc->rl_ldata.rl_tx_list_map);
1014 	if (error) {
1015 		device_printf(dev, "could not allocate TX DMA ring\n");
1016 		return (error);
1017 	}
1018 
1019 	/* Load the map for the TX ring. */
1020 
1021 	sc->rl_ldata.rl_tx_list_addr = 0;
1022 	error = bus_dmamap_load(sc->rl_ldata.rl_tx_list_tag,
1023 	     sc->rl_ldata.rl_tx_list_map, sc->rl_ldata.rl_tx_list,
1024 	     tx_list_size, re_dma_map_addr,
1025 	     &sc->rl_ldata.rl_tx_list_addr, BUS_DMA_NOWAIT);
1026 	if (error != 0 || sc->rl_ldata.rl_tx_list_addr == 0) {
1027 		device_printf(dev, "could not load TX DMA ring\n");
1028 		return (ENOMEM);
1029 	}
1030 
1031 	/* Create DMA maps for TX buffers */
1032 
1033 	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
1034 		error = bus_dmamap_create(sc->rl_ldata.rl_tx_mtag, 0,
1035 		    &sc->rl_ldata.rl_tx_desc[i].tx_dmamap);
1036 		if (error) {
1037 			device_printf(dev, "could not create DMA map for TX\n");
1038 			return (error);
1039 		}
1040 	}
1041 
1042 	/*
1043 	 * Allocate map for RX descriptor list.
1044 	 */
1045 	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
1046 	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1047 	    NULL, rx_list_size, 1, rx_list_size, 0,
1048 	    NULL, NULL, &sc->rl_ldata.rl_rx_list_tag);
1049 	if (error) {
1050 		device_printf(dev, "could not create RX DMA ring tag\n");
1051 		return (error);
1052 	}
1053 
1054 	/* Allocate DMA'able memory for the RX ring */
1055 
1056 	error = bus_dmamem_alloc(sc->rl_ldata.rl_rx_list_tag,
1057 	    (void **)&sc->rl_ldata.rl_rx_list,
1058 	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1059 	    &sc->rl_ldata.rl_rx_list_map);
1060 	if (error) {
1061 		device_printf(dev, "could not allocate RX DMA ring\n");
1062 		return (error);
1063 	}
1064 
1065 	/* Load the map for the RX ring. */
1066 
1067 	sc->rl_ldata.rl_rx_list_addr = 0;
1068 	error = bus_dmamap_load(sc->rl_ldata.rl_rx_list_tag,
1069 	     sc->rl_ldata.rl_rx_list_map, sc->rl_ldata.rl_rx_list,
1070 	     rx_list_size, re_dma_map_addr,
1071 	     &sc->rl_ldata.rl_rx_list_addr, BUS_DMA_NOWAIT);
1072 	if (error != 0 || sc->rl_ldata.rl_rx_list_addr == 0) {
1073 		device_printf(dev, "could not load RX DMA ring\n");
1074 		return (ENOMEM);
1075 	}
1076 
1077 	/* Create DMA maps for RX buffers */
1078 
1079 	error = bus_dmamap_create(sc->rl_ldata.rl_rx_mtag, 0,
1080 	    &sc->rl_ldata.rl_rx_sparemap);
1081 	if (error) {
1082 		device_printf(dev, "could not create spare DMA map for RX\n");
1083 		return (error);
1084 	}
1085 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1086 		error = bus_dmamap_create(sc->rl_ldata.rl_rx_mtag, 0,
1087 		    &sc->rl_ldata.rl_rx_desc[i].rx_dmamap);
1088 		if (error) {
1089 			device_printf(dev, "could not create DMA map for RX\n");
1090 			return (error);
1091 		}
1092 	}
1093 
1094 	return (0);
1095 }
1096 
1097 /*
1098  * Attach the interface. Allocate softc structures, do ifmedia
1099  * setup and ethernet/BPF attach.
1100  */
1101 static int
1102 re_attach(device_t dev)
1103 {
1104 	u_char			eaddr[ETHER_ADDR_LEN];
1105 	u_int16_t		as[ETHER_ADDR_LEN / 2];
1106 	struct rl_softc		*sc;
1107 	struct ifnet		*ifp;
1108 	struct rl_hwrev		*hw_rev;
1109 	int			hwrev;
1110 	u_int16_t		devid, re_did = 0;
1111 	int			error = 0, rid, i;
1112 	int			msic, reg;
1113 	uint8_t			cfg;
1114 
1115 	sc = device_get_softc(dev);
1116 	sc->rl_dev = dev;
1117 
1118 	mtx_init(&sc->rl_mtx, device_get_nameunit(dev), MTX_NETWORK_LOCK,
1119 	    MTX_DEF);
1120 	callout_init_mtx(&sc->rl_stat_callout, &sc->rl_mtx, 0);
1121 
1122 	/*
1123 	 * Map control/status registers.
1124 	 */
1125 	pci_enable_busmaster(dev);
1126 
1127 	devid = pci_get_device(dev);
1128 	/*
1129 	 * Prefer memory space register mapping over IO space.
1130 	 * Because RTL8169SC does not seem to work when memory mapping
1131 	 * is used always activate io mapping.
1132 	 */
1133 	if (devid == RT_DEVICEID_8169SC)
1134 		prefer_iomap = 1;
1135 	if (prefer_iomap == 0) {
1136 		sc->rl_res_id = PCIR_BAR(1);
1137 		sc->rl_res_type = SYS_RES_MEMORY;
1138 		/* RTL8168/8101E seems to use different BARs. */
1139 		if (devid == RT_DEVICEID_8168 || devid == RT_DEVICEID_8101E)
1140 			sc->rl_res_id = PCIR_BAR(2);
1141 	} else {
1142 		sc->rl_res_id = PCIR_BAR(0);
1143 		sc->rl_res_type = SYS_RES_IOPORT;
1144 	}
1145 	sc->rl_res = bus_alloc_resource_any(dev, sc->rl_res_type,
1146 	    &sc->rl_res_id, RF_ACTIVE);
1147 	if (sc->rl_res == NULL && prefer_iomap == 0) {
1148 		sc->rl_res_id = PCIR_BAR(0);
1149 		sc->rl_res_type = SYS_RES_IOPORT;
1150 		sc->rl_res = bus_alloc_resource_any(dev, sc->rl_res_type,
1151 		    &sc->rl_res_id, RF_ACTIVE);
1152 	}
1153 	if (sc->rl_res == NULL) {
1154 		device_printf(dev, "couldn't map ports/memory\n");
1155 		error = ENXIO;
1156 		goto fail;
1157 	}
1158 
1159 	sc->rl_btag = rman_get_bustag(sc->rl_res);
1160 	sc->rl_bhandle = rman_get_bushandle(sc->rl_res);
1161 
1162 	msic = 0;
1163 	if (pci_find_extcap(dev, PCIY_EXPRESS, &reg) == 0) {
1164 		sc->rl_flags |= RL_FLAG_PCIE;
1165 		/* Set PCIe maximum read request size to 2048. */
1166 		if (pci_get_max_read_req(dev) < 2048)
1167 			pci_set_max_read_req(dev, 2048);
1168 		msic = pci_msi_count(dev);
1169 		if (bootverbose)
1170 			device_printf(dev, "MSI count : %d\n", msic);
1171 	}
1172 	if (msic > 0 && msi_disable == 0) {
1173 		msic = 1;
1174 		if (pci_alloc_msi(dev, &msic) == 0) {
1175 			if (msic == RL_MSI_MESSAGES) {
1176 				device_printf(dev, "Using %d MSI messages\n",
1177 				    msic);
1178 				sc->rl_flags |= RL_FLAG_MSI;
1179 				/* Explicitly set MSI enable bit. */
1180 				CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1181 				cfg = CSR_READ_1(sc, RL_CFG2);
1182 				cfg |= RL_CFG2_MSI;
1183 				CSR_WRITE_1(sc, RL_CFG2, cfg);
1184 				CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1185 			} else
1186 				pci_release_msi(dev);
1187 		}
1188 	}
1189 
1190 	/* Allocate interrupt */
1191 	if ((sc->rl_flags & RL_FLAG_MSI) == 0) {
1192 		rid = 0;
1193 		sc->rl_irq[0] = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
1194 		    RF_SHAREABLE | RF_ACTIVE);
1195 		if (sc->rl_irq[0] == NULL) {
1196 			device_printf(dev, "couldn't allocate IRQ resources\n");
1197 			error = ENXIO;
1198 			goto fail;
1199 		}
1200 	} else {
1201 		for (i = 0, rid = 1; i < RL_MSI_MESSAGES; i++, rid++) {
1202 			sc->rl_irq[i] = bus_alloc_resource_any(dev,
1203 			    SYS_RES_IRQ, &rid, RF_ACTIVE);
1204 			if (sc->rl_irq[i] == NULL) {
1205 				device_printf(dev,
1206 				    "couldn't llocate IRQ resources for "
1207 				    "message %d\n", rid);
1208 				error = ENXIO;
1209 				goto fail;
1210 			}
1211 		}
1212 	}
1213 
1214 	if ((sc->rl_flags & RL_FLAG_MSI) == 0) {
1215 		CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1216 		cfg = CSR_READ_1(sc, RL_CFG2);
1217 		if ((cfg & RL_CFG2_MSI) != 0) {
1218 			device_printf(dev, "turning off MSI enable bit.\n");
1219 			cfg &= ~RL_CFG2_MSI;
1220 			CSR_WRITE_1(sc, RL_CFG2, cfg);
1221 		}
1222 		CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1223 	}
1224 
1225 	/* Reset the adapter. */
1226 	RL_LOCK(sc);
1227 	re_reset(sc);
1228 	RL_UNLOCK(sc);
1229 
1230 	hw_rev = re_hwrevs;
1231 	hwrev = CSR_READ_4(sc, RL_TXCFG);
1232 	switch (hwrev & 0x70000000) {
1233 	case 0x00000000:
1234 	case 0x10000000:
1235 		device_printf(dev, "Chip rev. 0x%08x\n", hwrev & 0xfc800000);
1236 		hwrev &= (RL_TXCFG_HWREV | 0x80000000);
1237 		break;
1238 	default:
1239 		device_printf(dev, "Chip rev. 0x%08x\n", hwrev & 0x7c800000);
1240 		hwrev &= RL_TXCFG_HWREV;
1241 		break;
1242 	}
1243 	device_printf(dev, "MAC rev. 0x%08x\n", hwrev & 0x00700000);
1244 	while (hw_rev->rl_desc != NULL) {
1245 		if (hw_rev->rl_rev == hwrev) {
1246 			sc->rl_type = hw_rev->rl_type;
1247 			sc->rl_hwrev = hw_rev->rl_rev;
1248 			break;
1249 		}
1250 		hw_rev++;
1251 	}
1252 	if (hw_rev->rl_desc == NULL) {
1253 		device_printf(dev, "Unknown H/W revision: 0x%08x\n", hwrev);
1254 		error = ENXIO;
1255 		goto fail;
1256 	}
1257 
1258 	switch (hw_rev->rl_rev) {
1259 	case RL_HWREV_8139CPLUS:
1260 		sc->rl_flags |= RL_FLAG_NOJUMBO | RL_FLAG_FASTETHER |
1261 		    RL_FLAG_AUTOPAD;
1262 		break;
1263 	case RL_HWREV_8100E:
1264 	case RL_HWREV_8101E:
1265 		sc->rl_flags |= RL_FLAG_NOJUMBO | RL_FLAG_PHYWAKE |
1266 		    RL_FLAG_FASTETHER;
1267 		break;
1268 	case RL_HWREV_8102E:
1269 	case RL_HWREV_8102EL:
1270 	case RL_HWREV_8102EL_SPIN1:
1271 		sc->rl_flags |= RL_FLAG_NOJUMBO | RL_FLAG_PHYWAKE |
1272 		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
1273 		    RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD;
1274 		break;
1275 	case RL_HWREV_8103E:
1276 		sc->rl_flags |= RL_FLAG_NOJUMBO | RL_FLAG_PHYWAKE |
1277 		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
1278 		    RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD |
1279 		    RL_FLAG_MACSLEEP;
1280 		break;
1281 	case RL_HWREV_8168_SPIN1:
1282 	case RL_HWREV_8168_SPIN2:
1283 		sc->rl_flags |= RL_FLAG_WOLRXENB;
1284 		/* FALLTHROUGH */
1285 	case RL_HWREV_8168_SPIN3:
1286 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_MACSTAT;
1287 		break;
1288 	case RL_HWREV_8168C_SPIN2:
1289 		sc->rl_flags |= RL_FLAG_MACSLEEP;
1290 		/* FALLTHROUGH */
1291 	case RL_HWREV_8168C:
1292 		if ((hwrev & 0x00700000) == 0x00200000)
1293 			sc->rl_flags |= RL_FLAG_MACSLEEP;
1294 		/* FALLTHROUGH */
1295 	case RL_HWREV_8168CP:
1296 	case RL_HWREV_8168D:
1297 	case RL_HWREV_8168DP:
1298 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
1299 		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP |
1300 		    RL_FLAG_AUTOPAD;
1301 		/*
1302 		 * These controllers support jumbo frame but it seems
1303 		 * that enabling it requires touching additional magic
1304 		 * registers. Depending on MAC revisions some
1305 		 * controllers need to disable checksum offload. So
1306 		 * disable jumbo frame until I have better idea what
1307 		 * it really requires to make it support.
1308 		 * RTL8168C/CP : supports up to 6KB jumbo frame.
1309 		 * RTL8111C/CP : supports up to 9KB jumbo frame.
1310 		 */
1311 		sc->rl_flags |= RL_FLAG_NOJUMBO;
1312 		break;
1313 	case RL_HWREV_8168E:
1314 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
1315 		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
1316 		    RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD | RL_FLAG_NOJUMBO;
1317 		break;
1318 	case RL_HWREV_8169_8110SB:
1319 	case RL_HWREV_8169_8110SBL:
1320 	case RL_HWREV_8169_8110SC:
1321 	case RL_HWREV_8169_8110SCE:
1322 		sc->rl_flags |= RL_FLAG_PHYWAKE;
1323 		/* FALLTHROUGH */
1324 	case RL_HWREV_8169:
1325 	case RL_HWREV_8169S:
1326 	case RL_HWREV_8110S:
1327 		sc->rl_flags |= RL_FLAG_MACRESET;
1328 		break;
1329 	default:
1330 		break;
1331 	}
1332 
1333 	/* Enable PME. */
1334 	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1335 	cfg = CSR_READ_1(sc, RL_CFG1);
1336 	cfg |= RL_CFG1_PME;
1337 	CSR_WRITE_1(sc, RL_CFG1, cfg);
1338 	cfg = CSR_READ_1(sc, RL_CFG5);
1339 	cfg &= RL_CFG5_PME_STS;
1340 	CSR_WRITE_1(sc, RL_CFG5, cfg);
1341 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1342 
1343 	if ((sc->rl_flags & RL_FLAG_PAR) != 0) {
1344 		/*
1345 		 * XXX Should have a better way to extract station
1346 		 * address from EEPROM.
1347 		 */
1348 		for (i = 0; i < ETHER_ADDR_LEN; i++)
1349 			eaddr[i] = CSR_READ_1(sc, RL_IDR0 + i);
1350 	} else {
1351 		sc->rl_eewidth = RL_9356_ADDR_LEN;
1352 		re_read_eeprom(sc, (caddr_t)&re_did, 0, 1);
1353 		if (re_did != 0x8129)
1354 			sc->rl_eewidth = RL_9346_ADDR_LEN;
1355 
1356 		/*
1357 		 * Get station address from the EEPROM.
1358 		 */
1359 		re_read_eeprom(sc, (caddr_t)as, RL_EE_EADDR, 3);
1360 		for (i = 0; i < ETHER_ADDR_LEN / 2; i++)
1361 			as[i] = le16toh(as[i]);
1362 		bcopy(as, eaddr, sizeof(eaddr));
1363 	}
1364 
1365 	if (sc->rl_type == RL_8169) {
1366 		/* Set RX length mask and number of descriptors. */
1367 		sc->rl_rxlenmask = RL_RDESC_STAT_GFRAGLEN;
1368 		sc->rl_txstart = RL_GTXSTART;
1369 		sc->rl_ldata.rl_tx_desc_cnt = RL_8169_TX_DESC_CNT;
1370 		sc->rl_ldata.rl_rx_desc_cnt = RL_8169_RX_DESC_CNT;
1371 	} else {
1372 		/* Set RX length mask and number of descriptors. */
1373 		sc->rl_rxlenmask = RL_RDESC_STAT_FRAGLEN;
1374 		sc->rl_txstart = RL_TXSTART;
1375 		sc->rl_ldata.rl_tx_desc_cnt = RL_8139_TX_DESC_CNT;
1376 		sc->rl_ldata.rl_rx_desc_cnt = RL_8139_RX_DESC_CNT;
1377 	}
1378 
1379 	error = re_allocmem(dev, sc);
1380 	if (error)
1381 		goto fail;
1382 
1383 	ifp = sc->rl_ifp = if_alloc(IFT_ETHER);
1384 	if (ifp == NULL) {
1385 		device_printf(dev, "can not if_alloc()\n");
1386 		error = ENOSPC;
1387 		goto fail;
1388 	}
1389 
1390 	/* Take controller out of deep sleep mode. */
1391 	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
1392 		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
1393 			CSR_WRITE_1(sc, RL_GPIO,
1394 			    CSR_READ_1(sc, RL_GPIO) | 0x01);
1395 		else
1396 			CSR_WRITE_1(sc, RL_GPIO,
1397 			    CSR_READ_1(sc, RL_GPIO) & ~0x01);
1398 	}
1399 
1400 	/* Take PHY out of power down mode. */
1401 	if ((sc->rl_flags & RL_FLAG_PHYWAKE_PM) != 0)
1402 		CSR_WRITE_1(sc, RL_PMCH, CSR_READ_1(sc, RL_PMCH) | 0x80);
1403 	if ((sc->rl_flags & RL_FLAG_PHYWAKE) != 0) {
1404 		re_gmii_writereg(dev, 1, 0x1f, 0);
1405 		re_gmii_writereg(dev, 1, 0x0e, 0);
1406 	}
1407 
1408 	/* Do MII setup */
1409 	if (mii_phy_probe(dev, &sc->rl_miibus,
1410 	    re_ifmedia_upd, re_ifmedia_sts)) {
1411 		device_printf(dev, "MII without any phy!\n");
1412 		error = ENXIO;
1413 		goto fail;
1414 	}
1415 
1416 	ifp->if_softc = sc;
1417 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1418 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1419 	ifp->if_ioctl = re_ioctl;
1420 	ifp->if_start = re_start;
1421 	ifp->if_hwassist = RE_CSUM_FEATURES;
1422 	ifp->if_capabilities = IFCAP_HWCSUM;
1423 	ifp->if_capenable = ifp->if_capabilities;
1424 	ifp->if_init = re_init;
1425 	IFQ_SET_MAXLEN(&ifp->if_snd, RL_IFQ_MAXLEN);
1426 	ifp->if_snd.ifq_drv_maxlen = RL_IFQ_MAXLEN;
1427 	IFQ_SET_READY(&ifp->if_snd);
1428 
1429 	TASK_INIT(&sc->rl_txtask, 1, re_tx_task, ifp);
1430 	TASK_INIT(&sc->rl_inttask, 0, re_int_task, sc);
1431 
1432 	/*
1433 	 * XXX
1434 	 * Still have no idea how to make TSO work on 8168C, 8168CP,
1435 	 * 8111C and 8111CP.
1436 	 */
1437 	if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
1438 		ifp->if_hwassist |= CSUM_TSO;
1439 		ifp->if_capabilities |= IFCAP_TSO4 | IFCAP_VLAN_HWTSO;
1440 	}
1441 
1442 	/*
1443 	 * Call MI attach routine.
1444 	 */
1445 	ether_ifattach(ifp, eaddr);
1446 
1447 	/* VLAN capability setup */
1448 	ifp->if_capabilities |= IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING;
1449 	if (ifp->if_capabilities & IFCAP_HWCSUM)
1450 		ifp->if_capabilities |= IFCAP_VLAN_HWCSUM;
1451 	/* Enable WOL if PM is supported. */
1452 	if (pci_find_extcap(sc->rl_dev, PCIY_PMG, &reg) == 0)
1453 		ifp->if_capabilities |= IFCAP_WOL;
1454 	ifp->if_capenable = ifp->if_capabilities;
1455 	/*
1456 	 * Don't enable TSO by default. Under certain
1457 	 * circumtances the controller generated corrupted
1458 	 * packets in TSO size.
1459 	 */
1460 	ifp->if_hwassist &= ~CSUM_TSO;
1461 	ifp->if_capenable &= ~(IFCAP_TSO4 | IFCAP_VLAN_HWTSO);
1462 #ifdef DEVICE_POLLING
1463 	ifp->if_capabilities |= IFCAP_POLLING;
1464 #endif
1465 	/*
1466 	 * Tell the upper layer(s) we support long frames.
1467 	 * Must appear after the call to ether_ifattach() because
1468 	 * ether_ifattach() sets ifi_hdrlen to the default value.
1469 	 */
1470 	ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
1471 
1472 #ifdef RE_DIAG
1473 	/*
1474 	 * Perform hardware diagnostic on the original RTL8169.
1475 	 * Some 32-bit cards were incorrectly wired and would
1476 	 * malfunction if plugged into a 64-bit slot.
1477 	 */
1478 
1479 	if (hwrev == RL_HWREV_8169) {
1480 		error = re_diag(sc);
1481 		if (error) {
1482 			device_printf(dev,
1483 		    	"attach aborted due to hardware diag failure\n");
1484 			ether_ifdetach(ifp);
1485 			goto fail;
1486 		}
1487 	}
1488 #endif
1489 
1490 	/* Hook interrupt last to avoid having to lock softc */
1491 	if ((sc->rl_flags & RL_FLAG_MSI) == 0)
1492 		error = bus_setup_intr(dev, sc->rl_irq[0],
1493 		    INTR_TYPE_NET | INTR_MPSAFE, re_intr, NULL, sc,
1494 		    &sc->rl_intrhand[0]);
1495 	else {
1496 		for (i = 0; i < RL_MSI_MESSAGES; i++) {
1497 			error = bus_setup_intr(dev, sc->rl_irq[i],
1498 			    INTR_TYPE_NET | INTR_MPSAFE, re_intr, NULL, sc,
1499 		    	    &sc->rl_intrhand[i]);
1500 			if (error != 0)
1501 				break;
1502 		}
1503 	}
1504 	if (error) {
1505 		device_printf(dev, "couldn't set up irq\n");
1506 		ether_ifdetach(ifp);
1507 	}
1508 
1509 fail:
1510 
1511 	if (error)
1512 		re_detach(dev);
1513 
1514 	return (error);
1515 }
1516 
1517 /*
1518  * Shutdown hardware and free up resources. This can be called any
1519  * time after the mutex has been initialized. It is called in both
1520  * the error case in attach and the normal detach case so it needs
1521  * to be careful about only freeing resources that have actually been
1522  * allocated.
1523  */
1524 static int
1525 re_detach(device_t dev)
1526 {
1527 	struct rl_softc		*sc;
1528 	struct ifnet		*ifp;
1529 	int			i, rid;
1530 
1531 	sc = device_get_softc(dev);
1532 	ifp = sc->rl_ifp;
1533 	KASSERT(mtx_initialized(&sc->rl_mtx), ("re mutex not initialized"));
1534 
1535 	/* These should only be active if attach succeeded */
1536 	if (device_is_attached(dev)) {
1537 #ifdef DEVICE_POLLING
1538 		if (ifp->if_capenable & IFCAP_POLLING)
1539 			ether_poll_deregister(ifp);
1540 #endif
1541 		RL_LOCK(sc);
1542 #if 0
1543 		sc->suspended = 1;
1544 #endif
1545 		re_stop(sc);
1546 		RL_UNLOCK(sc);
1547 		callout_drain(&sc->rl_stat_callout);
1548 		taskqueue_drain(taskqueue_fast, &sc->rl_inttask);
1549 		taskqueue_drain(taskqueue_fast, &sc->rl_txtask);
1550 		/*
1551 		 * Force off the IFF_UP flag here, in case someone
1552 		 * still had a BPF descriptor attached to this
1553 		 * interface. If they do, ether_ifdetach() will cause
1554 		 * the BPF code to try and clear the promisc mode
1555 		 * flag, which will bubble down to re_ioctl(),
1556 		 * which will try to call re_init() again. This will
1557 		 * turn the NIC back on and restart the MII ticker,
1558 		 * which will panic the system when the kernel tries
1559 		 * to invoke the re_tick() function that isn't there
1560 		 * anymore.
1561 		 */
1562 		ifp->if_flags &= ~IFF_UP;
1563 		ether_ifdetach(ifp);
1564 	}
1565 	if (sc->rl_miibus)
1566 		device_delete_child(dev, sc->rl_miibus);
1567 	bus_generic_detach(dev);
1568 
1569 	/*
1570 	 * The rest is resource deallocation, so we should already be
1571 	 * stopped here.
1572 	 */
1573 
1574 	for (i = 0; i < RL_MSI_MESSAGES; i++) {
1575 		if (sc->rl_intrhand[i] != NULL) {
1576 			bus_teardown_intr(dev, sc->rl_irq[i],
1577 			    sc->rl_intrhand[i]);
1578 			sc->rl_intrhand[i] = NULL;
1579 		}
1580 	}
1581 	if (ifp != NULL)
1582 		if_free(ifp);
1583 	if ((sc->rl_flags & RL_FLAG_MSI) == 0) {
1584 		if (sc->rl_irq[0] != NULL) {
1585 			bus_release_resource(dev, SYS_RES_IRQ, 0,
1586 			    sc->rl_irq[0]);
1587 			sc->rl_irq[0] = NULL;
1588 		}
1589 	} else {
1590 		for (i = 0, rid = 1; i < RL_MSI_MESSAGES; i++, rid++) {
1591 			if (sc->rl_irq[i] != NULL) {
1592 				bus_release_resource(dev, SYS_RES_IRQ, rid,
1593 				    sc->rl_irq[i]);
1594 				sc->rl_irq[i] = NULL;
1595 			}
1596 		}
1597 		pci_release_msi(dev);
1598 	}
1599 	if (sc->rl_res)
1600 		bus_release_resource(dev, sc->rl_res_type, sc->rl_res_id,
1601 		    sc->rl_res);
1602 
1603 	/* Unload and free the RX DMA ring memory and map */
1604 
1605 	if (sc->rl_ldata.rl_rx_list_tag) {
1606 		bus_dmamap_unload(sc->rl_ldata.rl_rx_list_tag,
1607 		    sc->rl_ldata.rl_rx_list_map);
1608 		bus_dmamem_free(sc->rl_ldata.rl_rx_list_tag,
1609 		    sc->rl_ldata.rl_rx_list,
1610 		    sc->rl_ldata.rl_rx_list_map);
1611 		bus_dma_tag_destroy(sc->rl_ldata.rl_rx_list_tag);
1612 	}
1613 
1614 	/* Unload and free the TX DMA ring memory and map */
1615 
1616 	if (sc->rl_ldata.rl_tx_list_tag) {
1617 		bus_dmamap_unload(sc->rl_ldata.rl_tx_list_tag,
1618 		    sc->rl_ldata.rl_tx_list_map);
1619 		bus_dmamem_free(sc->rl_ldata.rl_tx_list_tag,
1620 		    sc->rl_ldata.rl_tx_list,
1621 		    sc->rl_ldata.rl_tx_list_map);
1622 		bus_dma_tag_destroy(sc->rl_ldata.rl_tx_list_tag);
1623 	}
1624 
1625 	/* Destroy all the RX and TX buffer maps */
1626 
1627 	if (sc->rl_ldata.rl_tx_mtag) {
1628 		for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++)
1629 			bus_dmamap_destroy(sc->rl_ldata.rl_tx_mtag,
1630 			    sc->rl_ldata.rl_tx_desc[i].tx_dmamap);
1631 		bus_dma_tag_destroy(sc->rl_ldata.rl_tx_mtag);
1632 	}
1633 	if (sc->rl_ldata.rl_rx_mtag) {
1634 		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++)
1635 			bus_dmamap_destroy(sc->rl_ldata.rl_rx_mtag,
1636 			    sc->rl_ldata.rl_rx_desc[i].rx_dmamap);
1637 		if (sc->rl_ldata.rl_rx_sparemap)
1638 			bus_dmamap_destroy(sc->rl_ldata.rl_rx_mtag,
1639 			    sc->rl_ldata.rl_rx_sparemap);
1640 		bus_dma_tag_destroy(sc->rl_ldata.rl_rx_mtag);
1641 	}
1642 
1643 	/* Unload and free the stats buffer and map */
1644 
1645 	if (sc->rl_ldata.rl_stag) {
1646 		bus_dmamap_unload(sc->rl_ldata.rl_stag,
1647 		    sc->rl_ldata.rl_rx_list_map);
1648 		bus_dmamem_free(sc->rl_ldata.rl_stag,
1649 		    sc->rl_ldata.rl_stats,
1650 		    sc->rl_ldata.rl_smap);
1651 		bus_dma_tag_destroy(sc->rl_ldata.rl_stag);
1652 	}
1653 
1654 	if (sc->rl_parent_tag)
1655 		bus_dma_tag_destroy(sc->rl_parent_tag);
1656 
1657 	mtx_destroy(&sc->rl_mtx);
1658 
1659 	return (0);
1660 }
1661 
1662 static __inline void
1663 re_discard_rxbuf(struct rl_softc *sc, int idx)
1664 {
1665 	struct rl_desc		*desc;
1666 	struct rl_rxdesc	*rxd;
1667 	uint32_t		cmdstat;
1668 
1669 	rxd = &sc->rl_ldata.rl_rx_desc[idx];
1670 	desc = &sc->rl_ldata.rl_rx_list[idx];
1671 	desc->rl_vlanctl = 0;
1672 	cmdstat = rxd->rx_size;
1673 	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
1674 		cmdstat |= RL_RDESC_CMD_EOR;
1675 	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
1676 }
1677 
1678 static int
1679 re_newbuf(struct rl_softc *sc, int idx)
1680 {
1681 	struct mbuf		*m;
1682 	struct rl_rxdesc	*rxd;
1683 	bus_dma_segment_t	segs[1];
1684 	bus_dmamap_t		map;
1685 	struct rl_desc		*desc;
1686 	uint32_t		cmdstat;
1687 	int			error, nsegs;
1688 
1689 	m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
1690 	if (m == NULL)
1691 		return (ENOBUFS);
1692 
1693 	m->m_len = m->m_pkthdr.len = MCLBYTES;
1694 #ifdef RE_FIXUP_RX
1695 	/*
1696 	 * This is part of an evil trick to deal with non-x86 platforms.
1697 	 * The RealTek chip requires RX buffers to be aligned on 64-bit
1698 	 * boundaries, but that will hose non-x86 machines. To get around
1699 	 * this, we leave some empty space at the start of each buffer
1700 	 * and for non-x86 hosts, we copy the buffer back six bytes
1701 	 * to achieve word alignment. This is slightly more efficient
1702 	 * than allocating a new buffer, copying the contents, and
1703 	 * discarding the old buffer.
1704 	 */
1705 	m_adj(m, RE_ETHER_ALIGN);
1706 #endif
1707 	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_rx_mtag,
1708 	    sc->rl_ldata.rl_rx_sparemap, m, segs, &nsegs, BUS_DMA_NOWAIT);
1709 	if (error != 0) {
1710 		m_freem(m);
1711 		return (ENOBUFS);
1712 	}
1713 	KASSERT(nsegs == 1, ("%s: %d segment returned!", __func__, nsegs));
1714 
1715 	rxd = &sc->rl_ldata.rl_rx_desc[idx];
1716 	if (rxd->rx_m != NULL) {
1717 		bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap,
1718 		    BUS_DMASYNC_POSTREAD);
1719 		bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap);
1720 	}
1721 
1722 	rxd->rx_m = m;
1723 	map = rxd->rx_dmamap;
1724 	rxd->rx_dmamap = sc->rl_ldata.rl_rx_sparemap;
1725 	rxd->rx_size = segs[0].ds_len;
1726 	sc->rl_ldata.rl_rx_sparemap = map;
1727 	bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap,
1728 	    BUS_DMASYNC_PREREAD);
1729 
1730 	desc = &sc->rl_ldata.rl_rx_list[idx];
1731 	desc->rl_vlanctl = 0;
1732 	desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[0].ds_addr));
1733 	desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[0].ds_addr));
1734 	cmdstat = segs[0].ds_len;
1735 	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
1736 		cmdstat |= RL_RDESC_CMD_EOR;
1737 	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
1738 
1739 	return (0);
1740 }
1741 
1742 #ifdef RE_FIXUP_RX
1743 static __inline void
1744 re_fixup_rx(struct mbuf *m)
1745 {
1746 	int                     i;
1747 	uint16_t                *src, *dst;
1748 
1749 	src = mtod(m, uint16_t *);
1750 	dst = src - (RE_ETHER_ALIGN - ETHER_ALIGN) / sizeof *src;
1751 
1752 	for (i = 0; i < (m->m_len / sizeof(uint16_t) + 1); i++)
1753 		*dst++ = *src++;
1754 
1755 	m->m_data -= RE_ETHER_ALIGN - ETHER_ALIGN;
1756 }
1757 #endif
1758 
1759 static int
1760 re_tx_list_init(struct rl_softc *sc)
1761 {
1762 	struct rl_desc		*desc;
1763 	int			i;
1764 
1765 	RL_LOCK_ASSERT(sc);
1766 
1767 	bzero(sc->rl_ldata.rl_tx_list,
1768 	    sc->rl_ldata.rl_tx_desc_cnt * sizeof(struct rl_desc));
1769 	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++)
1770 		sc->rl_ldata.rl_tx_desc[i].tx_m = NULL;
1771 	/* Set EOR. */
1772 	desc = &sc->rl_ldata.rl_tx_list[sc->rl_ldata.rl_tx_desc_cnt - 1];
1773 	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_EOR);
1774 
1775 	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
1776 	    sc->rl_ldata.rl_tx_list_map,
1777 	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1778 
1779 	sc->rl_ldata.rl_tx_prodidx = 0;
1780 	sc->rl_ldata.rl_tx_considx = 0;
1781 	sc->rl_ldata.rl_tx_free = sc->rl_ldata.rl_tx_desc_cnt;
1782 
1783 	return (0);
1784 }
1785 
1786 static int
1787 re_rx_list_init(struct rl_softc *sc)
1788 {
1789 	int			error, i;
1790 
1791 	bzero(sc->rl_ldata.rl_rx_list,
1792 	    sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc));
1793 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1794 		sc->rl_ldata.rl_rx_desc[i].rx_m = NULL;
1795 		if ((error = re_newbuf(sc, i)) != 0)
1796 			return (error);
1797 	}
1798 
1799 	/* Flush the RX descriptors */
1800 
1801 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1802 	    sc->rl_ldata.rl_rx_list_map,
1803 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
1804 
1805 	sc->rl_ldata.rl_rx_prodidx = 0;
1806 	sc->rl_head = sc->rl_tail = NULL;
1807 
1808 	return (0);
1809 }
1810 
1811 /*
1812  * RX handler for C+ and 8169. For the gigE chips, we support
1813  * the reception of jumbo frames that have been fragmented
1814  * across multiple 2K mbuf cluster buffers.
1815  */
1816 static int
1817 re_rxeof(struct rl_softc *sc, int *rx_npktsp)
1818 {
1819 	struct mbuf		*m;
1820 	struct ifnet		*ifp;
1821 	int			i, total_len;
1822 	struct rl_desc		*cur_rx;
1823 	u_int32_t		rxstat, rxvlan;
1824 	int			maxpkt = 16, rx_npkts = 0;
1825 
1826 	RL_LOCK_ASSERT(sc);
1827 
1828 	ifp = sc->rl_ifp;
1829 
1830 	/* Invalidate the descriptor memory */
1831 
1832 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1833 	    sc->rl_ldata.rl_rx_list_map,
1834 	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
1835 
1836 	for (i = sc->rl_ldata.rl_rx_prodidx; maxpkt > 0;
1837 	    i = RL_RX_DESC_NXT(sc, i)) {
1838 		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1839 			break;
1840 		cur_rx = &sc->rl_ldata.rl_rx_list[i];
1841 		rxstat = le32toh(cur_rx->rl_cmdstat);
1842 		if ((rxstat & RL_RDESC_STAT_OWN) != 0)
1843 			break;
1844 		total_len = rxstat & sc->rl_rxlenmask;
1845 		rxvlan = le32toh(cur_rx->rl_vlanctl);
1846 		m = sc->rl_ldata.rl_rx_desc[i].rx_m;
1847 
1848 		if (!(rxstat & RL_RDESC_STAT_EOF)) {
1849 			if (re_newbuf(sc, i) != 0) {
1850 				/*
1851 				 * If this is part of a multi-fragment packet,
1852 				 * discard all the pieces.
1853 				 */
1854 				if (sc->rl_head != NULL) {
1855 					m_freem(sc->rl_head);
1856 					sc->rl_head = sc->rl_tail = NULL;
1857 				}
1858 				re_discard_rxbuf(sc, i);
1859 				continue;
1860 			}
1861 			m->m_len = RE_RX_DESC_BUFLEN;
1862 			if (sc->rl_head == NULL)
1863 				sc->rl_head = sc->rl_tail = m;
1864 			else {
1865 				m->m_flags &= ~M_PKTHDR;
1866 				sc->rl_tail->m_next = m;
1867 				sc->rl_tail = m;
1868 			}
1869 			continue;
1870 		}
1871 
1872 		/*
1873 		 * NOTE: for the 8139C+, the frame length field
1874 		 * is always 12 bits in size, but for the gigE chips,
1875 		 * it is 13 bits (since the max RX frame length is 16K).
1876 		 * Unfortunately, all 32 bits in the status word
1877 		 * were already used, so to make room for the extra
1878 		 * length bit, RealTek took out the 'frame alignment
1879 		 * error' bit and shifted the other status bits
1880 		 * over one slot. The OWN, EOR, FS and LS bits are
1881 		 * still in the same places. We have already extracted
1882 		 * the frame length and checked the OWN bit, so rather
1883 		 * than using an alternate bit mapping, we shift the
1884 		 * status bits one space to the right so we can evaluate
1885 		 * them using the 8169 status as though it was in the
1886 		 * same format as that of the 8139C+.
1887 		 */
1888 		if (sc->rl_type == RL_8169)
1889 			rxstat >>= 1;
1890 
1891 		/*
1892 		 * if total_len > 2^13-1, both _RXERRSUM and _GIANT will be
1893 		 * set, but if CRC is clear, it will still be a valid frame.
1894 		 */
1895 		if (rxstat & RL_RDESC_STAT_RXERRSUM && !(total_len > 8191 &&
1896 		    (rxstat & RL_RDESC_STAT_ERRS) == RL_RDESC_STAT_GIANT)) {
1897 			ifp->if_ierrors++;
1898 			/*
1899 			 * If this is part of a multi-fragment packet,
1900 			 * discard all the pieces.
1901 			 */
1902 			if (sc->rl_head != NULL) {
1903 				m_freem(sc->rl_head);
1904 				sc->rl_head = sc->rl_tail = NULL;
1905 			}
1906 			re_discard_rxbuf(sc, i);
1907 			continue;
1908 		}
1909 
1910 		/*
1911 		 * If allocating a replacement mbuf fails,
1912 		 * reload the current one.
1913 		 */
1914 
1915 		if (re_newbuf(sc, i) != 0) {
1916 			ifp->if_iqdrops++;
1917 			if (sc->rl_head != NULL) {
1918 				m_freem(sc->rl_head);
1919 				sc->rl_head = sc->rl_tail = NULL;
1920 			}
1921 			re_discard_rxbuf(sc, i);
1922 			continue;
1923 		}
1924 
1925 		if (sc->rl_head != NULL) {
1926 			m->m_len = total_len % RE_RX_DESC_BUFLEN;
1927 			if (m->m_len == 0)
1928 				m->m_len = RE_RX_DESC_BUFLEN;
1929 			/*
1930 			 * Special case: if there's 4 bytes or less
1931 			 * in this buffer, the mbuf can be discarded:
1932 			 * the last 4 bytes is the CRC, which we don't
1933 			 * care about anyway.
1934 			 */
1935 			if (m->m_len <= ETHER_CRC_LEN) {
1936 				sc->rl_tail->m_len -=
1937 				    (ETHER_CRC_LEN - m->m_len);
1938 				m_freem(m);
1939 			} else {
1940 				m->m_len -= ETHER_CRC_LEN;
1941 				m->m_flags &= ~M_PKTHDR;
1942 				sc->rl_tail->m_next = m;
1943 			}
1944 			m = sc->rl_head;
1945 			sc->rl_head = sc->rl_tail = NULL;
1946 			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
1947 		} else
1948 			m->m_pkthdr.len = m->m_len =
1949 			    (total_len - ETHER_CRC_LEN);
1950 
1951 #ifdef RE_FIXUP_RX
1952 		re_fixup_rx(m);
1953 #endif
1954 		ifp->if_ipackets++;
1955 		m->m_pkthdr.rcvif = ifp;
1956 
1957 		/* Do RX checksumming if enabled */
1958 
1959 		if (ifp->if_capenable & IFCAP_RXCSUM) {
1960 			if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
1961 				/* Check IP header checksum */
1962 				if (rxstat & RL_RDESC_STAT_PROTOID)
1963 					m->m_pkthdr.csum_flags |=
1964 					    CSUM_IP_CHECKED;
1965 				if (!(rxstat & RL_RDESC_STAT_IPSUMBAD))
1966 					m->m_pkthdr.csum_flags |=
1967 					    CSUM_IP_VALID;
1968 
1969 				/* Check TCP/UDP checksum */
1970 				if ((RL_TCPPKT(rxstat) &&
1971 				    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
1972 				    (RL_UDPPKT(rxstat) &&
1973 				     !(rxstat & RL_RDESC_STAT_UDPSUMBAD))) {
1974 					m->m_pkthdr.csum_flags |=
1975 						CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
1976 					m->m_pkthdr.csum_data = 0xffff;
1977 				}
1978 			} else {
1979 				/*
1980 				 * RTL8168C/RTL816CP/RTL8111C/RTL8111CP
1981 				 */
1982 				if ((rxstat & RL_RDESC_STAT_PROTOID) &&
1983 				    (rxvlan & RL_RDESC_IPV4))
1984 					m->m_pkthdr.csum_flags |=
1985 					    CSUM_IP_CHECKED;
1986 				if (!(rxstat & RL_RDESC_STAT_IPSUMBAD) &&
1987 				    (rxvlan & RL_RDESC_IPV4))
1988 					m->m_pkthdr.csum_flags |=
1989 					    CSUM_IP_VALID;
1990 				if (((rxstat & RL_RDESC_STAT_TCP) &&
1991 				    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
1992 				    ((rxstat & RL_RDESC_STAT_UDP) &&
1993 				    !(rxstat & RL_RDESC_STAT_UDPSUMBAD))) {
1994 					m->m_pkthdr.csum_flags |=
1995 						CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
1996 					m->m_pkthdr.csum_data = 0xffff;
1997 				}
1998 			}
1999 		}
2000 		maxpkt--;
2001 		if (rxvlan & RL_RDESC_VLANCTL_TAG) {
2002 			m->m_pkthdr.ether_vtag =
2003 			    bswap16((rxvlan & RL_RDESC_VLANCTL_DATA));
2004 			m->m_flags |= M_VLANTAG;
2005 		}
2006 		RL_UNLOCK(sc);
2007 		(*ifp->if_input)(ifp, m);
2008 		RL_LOCK(sc);
2009 		rx_npkts++;
2010 	}
2011 
2012 	/* Flush the RX DMA ring */
2013 
2014 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
2015 	    sc->rl_ldata.rl_rx_list_map,
2016 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2017 
2018 	sc->rl_ldata.rl_rx_prodidx = i;
2019 
2020 	if (rx_npktsp != NULL)
2021 		*rx_npktsp = rx_npkts;
2022 	if (maxpkt)
2023 		return(EAGAIN);
2024 
2025 	return(0);
2026 }
2027 
2028 static void
2029 re_txeof(struct rl_softc *sc)
2030 {
2031 	struct ifnet		*ifp;
2032 	struct rl_txdesc	*txd;
2033 	u_int32_t		txstat;
2034 	int			cons;
2035 
2036 	cons = sc->rl_ldata.rl_tx_considx;
2037 	if (cons == sc->rl_ldata.rl_tx_prodidx)
2038 		return;
2039 
2040 	ifp = sc->rl_ifp;
2041 	/* Invalidate the TX descriptor list */
2042 	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2043 	    sc->rl_ldata.rl_tx_list_map,
2044 	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2045 
2046 	for (; cons != sc->rl_ldata.rl_tx_prodidx;
2047 	    cons = RL_TX_DESC_NXT(sc, cons)) {
2048 		txstat = le32toh(sc->rl_ldata.rl_tx_list[cons].rl_cmdstat);
2049 		if (txstat & RL_TDESC_STAT_OWN)
2050 			break;
2051 		/*
2052 		 * We only stash mbufs in the last descriptor
2053 		 * in a fragment chain, which also happens to
2054 		 * be the only place where the TX status bits
2055 		 * are valid.
2056 		 */
2057 		if (txstat & RL_TDESC_CMD_EOF) {
2058 			txd = &sc->rl_ldata.rl_tx_desc[cons];
2059 			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
2060 			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
2061 			bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag,
2062 			    txd->tx_dmamap);
2063 			KASSERT(txd->tx_m != NULL,
2064 			    ("%s: freeing NULL mbufs!", __func__));
2065 			m_freem(txd->tx_m);
2066 			txd->tx_m = NULL;
2067 			if (txstat & (RL_TDESC_STAT_EXCESSCOL|
2068 			    RL_TDESC_STAT_COLCNT))
2069 				ifp->if_collisions++;
2070 			if (txstat & RL_TDESC_STAT_TXERRSUM)
2071 				ifp->if_oerrors++;
2072 			else
2073 				ifp->if_opackets++;
2074 		}
2075 		sc->rl_ldata.rl_tx_free++;
2076 		ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
2077 	}
2078 	sc->rl_ldata.rl_tx_considx = cons;
2079 
2080 	/* No changes made to the TX ring, so no flush needed */
2081 
2082 	if (sc->rl_ldata.rl_tx_free != sc->rl_ldata.rl_tx_desc_cnt) {
2083 #ifdef RE_TX_MODERATION
2084 		/*
2085 		 * If not all descriptors have been reaped yet, reload
2086 		 * the timer so that we will eventually get another
2087 		 * interrupt that will cause us to re-enter this routine.
2088 		 * This is done in case the transmitter has gone idle.
2089 		 */
2090 		CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2091 #endif
2092 	} else
2093 		sc->rl_watchdog_timer = 0;
2094 }
2095 
2096 static void
2097 re_tick(void *xsc)
2098 {
2099 	struct rl_softc		*sc;
2100 	struct mii_data		*mii;
2101 
2102 	sc = xsc;
2103 
2104 	RL_LOCK_ASSERT(sc);
2105 
2106 	mii = device_get_softc(sc->rl_miibus);
2107 	mii_tick(mii);
2108 	if ((sc->rl_flags & RL_FLAG_LINK) == 0)
2109 		re_miibus_statchg(sc->rl_dev);
2110 	/*
2111 	 * Reclaim transmitted frames here. Technically it is not
2112 	 * necessary to do here but it ensures periodic reclamation
2113 	 * regardless of Tx completion interrupt which seems to be
2114 	 * lost on PCIe based controllers under certain situations.
2115 	 */
2116 	re_txeof(sc);
2117 	re_watchdog(sc);
2118 	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
2119 }
2120 
2121 #ifdef DEVICE_POLLING
2122 static int
2123 re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
2124 {
2125 	struct rl_softc *sc = ifp->if_softc;
2126 	int rx_npkts = 0;
2127 
2128 	RL_LOCK(sc);
2129 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
2130 		rx_npkts = re_poll_locked(ifp, cmd, count);
2131 	RL_UNLOCK(sc);
2132 	return (rx_npkts);
2133 }
2134 
2135 static int
2136 re_poll_locked(struct ifnet *ifp, enum poll_cmd cmd, int count)
2137 {
2138 	struct rl_softc *sc = ifp->if_softc;
2139 	int rx_npkts;
2140 
2141 	RL_LOCK_ASSERT(sc);
2142 
2143 	sc->rxcycles = count;
2144 	re_rxeof(sc, &rx_npkts);
2145 	re_txeof(sc);
2146 
2147 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2148 		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_txtask);
2149 
2150 	if (cmd == POLL_AND_CHECK_STATUS) { /* also check status register */
2151 		u_int16_t       status;
2152 
2153 		status = CSR_READ_2(sc, RL_ISR);
2154 		if (status == 0xffff)
2155 			return (rx_npkts);
2156 		if (status)
2157 			CSR_WRITE_2(sc, RL_ISR, status);
2158 		if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
2159 		    (sc->rl_flags & RL_FLAG_PCIE))
2160 			CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2161 
2162 		/*
2163 		 * XXX check behaviour on receiver stalls.
2164 		 */
2165 
2166 		if (status & RL_ISR_SYSTEM_ERR) {
2167 			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2168 			re_init_locked(sc);
2169 		}
2170 	}
2171 	return (rx_npkts);
2172 }
2173 #endif /* DEVICE_POLLING */
2174 
2175 static int
2176 re_intr(void *arg)
2177 {
2178 	struct rl_softc		*sc;
2179 	uint16_t		status;
2180 
2181 	sc = arg;
2182 
2183 	status = CSR_READ_2(sc, RL_ISR);
2184 	if (status == 0xFFFF || (status & RL_INTRS_CPLUS) == 0)
2185                 return (FILTER_STRAY);
2186 	CSR_WRITE_2(sc, RL_IMR, 0);
2187 
2188 	taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_inttask);
2189 
2190 	return (FILTER_HANDLED);
2191 }
2192 
2193 static void
2194 re_int_task(void *arg, int npending)
2195 {
2196 	struct rl_softc		*sc;
2197 	struct ifnet		*ifp;
2198 	u_int16_t		status;
2199 	int			rval = 0;
2200 
2201 	sc = arg;
2202 	ifp = sc->rl_ifp;
2203 
2204 	RL_LOCK(sc);
2205 
2206 	status = CSR_READ_2(sc, RL_ISR);
2207         CSR_WRITE_2(sc, RL_ISR, status);
2208 
2209 	if (sc->suspended ||
2210 	    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2211 		RL_UNLOCK(sc);
2212 		return;
2213 	}
2214 
2215 #ifdef DEVICE_POLLING
2216 	if  (ifp->if_capenable & IFCAP_POLLING) {
2217 		RL_UNLOCK(sc);
2218 		return;
2219 	}
2220 #endif
2221 
2222 	if (status & (RL_ISR_RX_OK|RL_ISR_RX_ERR|RL_ISR_FIFO_OFLOW))
2223 		rval = re_rxeof(sc, NULL);
2224 
2225 	/*
2226 	 * Some chips will ignore a second TX request issued
2227 	 * while an existing transmission is in progress. If
2228 	 * the transmitter goes idle but there are still
2229 	 * packets waiting to be sent, we need to restart the
2230 	 * channel here to flush them out. This only seems to
2231 	 * be required with the PCIe devices.
2232 	 */
2233 	if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
2234 	    (sc->rl_flags & RL_FLAG_PCIE))
2235 		CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2236 	if (status & (
2237 #ifdef RE_TX_MODERATION
2238 	    RL_ISR_TIMEOUT_EXPIRED|
2239 #else
2240 	    RL_ISR_TX_OK|
2241 #endif
2242 	    RL_ISR_TX_ERR|RL_ISR_TX_DESC_UNAVAIL))
2243 		re_txeof(sc);
2244 
2245 	if (status & RL_ISR_SYSTEM_ERR) {
2246 		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2247 		re_init_locked(sc);
2248 	}
2249 
2250 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2251 		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_txtask);
2252 
2253 	RL_UNLOCK(sc);
2254 
2255         if ((CSR_READ_2(sc, RL_ISR) & RL_INTRS_CPLUS) || rval) {
2256 		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_inttask);
2257 		return;
2258 	}
2259 
2260 	CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2261 }
2262 
2263 static int
2264 re_encap(struct rl_softc *sc, struct mbuf **m_head)
2265 {
2266 	struct rl_txdesc	*txd, *txd_last;
2267 	bus_dma_segment_t	segs[RL_NTXSEGS];
2268 	bus_dmamap_t		map;
2269 	struct mbuf		*m_new;
2270 	struct rl_desc		*desc;
2271 	int			nsegs, prod;
2272 	int			i, error, ei, si;
2273 	int			padlen;
2274 	uint32_t		cmdstat, csum_flags, vlanctl;
2275 
2276 	RL_LOCK_ASSERT(sc);
2277 	M_ASSERTPKTHDR((*m_head));
2278 
2279 	/*
2280 	 * With some of the RealTek chips, using the checksum offload
2281 	 * support in conjunction with the autopadding feature results
2282 	 * in the transmission of corrupt frames. For example, if we
2283 	 * need to send a really small IP fragment that's less than 60
2284 	 * bytes in size, and IP header checksumming is enabled, the
2285 	 * resulting ethernet frame that appears on the wire will
2286 	 * have garbled payload. To work around this, if TX IP checksum
2287 	 * offload is enabled, we always manually pad short frames out
2288 	 * to the minimum ethernet frame size.
2289 	 */
2290 	if ((sc->rl_flags & RL_FLAG_AUTOPAD) == 0 &&
2291 	    (*m_head)->m_pkthdr.len < RL_IP4CSUMTX_PADLEN &&
2292 	    ((*m_head)->m_pkthdr.csum_flags & CSUM_IP) != 0) {
2293 		padlen = RL_MIN_FRAMELEN - (*m_head)->m_pkthdr.len;
2294 		if (M_WRITABLE(*m_head) == 0) {
2295 			/* Get a writable copy. */
2296 			m_new = m_dup(*m_head, M_DONTWAIT);
2297 			m_freem(*m_head);
2298 			if (m_new == NULL) {
2299 				*m_head = NULL;
2300 				return (ENOBUFS);
2301 			}
2302 			*m_head = m_new;
2303 		}
2304 		if ((*m_head)->m_next != NULL ||
2305 		    M_TRAILINGSPACE(*m_head) < padlen) {
2306 			m_new = m_defrag(*m_head, M_DONTWAIT);
2307 			if (m_new == NULL) {
2308 				m_freem(*m_head);
2309 				*m_head = NULL;
2310 				return (ENOBUFS);
2311 			}
2312 		} else
2313 			m_new = *m_head;
2314 
2315 		/*
2316 		 * Manually pad short frames, and zero the pad space
2317 		 * to avoid leaking data.
2318 		 */
2319 		bzero(mtod(m_new, char *) + m_new->m_pkthdr.len, padlen);
2320 		m_new->m_pkthdr.len += padlen;
2321 		m_new->m_len = m_new->m_pkthdr.len;
2322 		*m_head = m_new;
2323 	}
2324 
2325 	prod = sc->rl_ldata.rl_tx_prodidx;
2326 	txd = &sc->rl_ldata.rl_tx_desc[prod];
2327 	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap,
2328 	    *m_head, segs, &nsegs, BUS_DMA_NOWAIT);
2329 	if (error == EFBIG) {
2330 		m_new = m_collapse(*m_head, M_DONTWAIT, RL_NTXSEGS);
2331 		if (m_new == NULL) {
2332 			m_freem(*m_head);
2333 			*m_head = NULL;
2334 			return (ENOBUFS);
2335 		}
2336 		*m_head = m_new;
2337 		error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_tx_mtag,
2338 		    txd->tx_dmamap, *m_head, segs, &nsegs, BUS_DMA_NOWAIT);
2339 		if (error != 0) {
2340 			m_freem(*m_head);
2341 			*m_head = NULL;
2342 			return (error);
2343 		}
2344 	} else if (error != 0)
2345 		return (error);
2346 	if (nsegs == 0) {
2347 		m_freem(*m_head);
2348 		*m_head = NULL;
2349 		return (EIO);
2350 	}
2351 
2352 	/* Check for number of available descriptors. */
2353 	if (sc->rl_ldata.rl_tx_free - nsegs <= 1) {
2354 		bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap);
2355 		return (ENOBUFS);
2356 	}
2357 
2358 	bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap,
2359 	    BUS_DMASYNC_PREWRITE);
2360 
2361 	/*
2362 	 * Set up checksum offload. Note: checksum offload bits must
2363 	 * appear in all descriptors of a multi-descriptor transmit
2364 	 * attempt. This is according to testing done with an 8169
2365 	 * chip. This is a requirement.
2366 	 */
2367 	vlanctl = 0;
2368 	csum_flags = 0;
2369 	if (((*m_head)->m_pkthdr.csum_flags & CSUM_TSO) != 0)
2370 		csum_flags = RL_TDESC_CMD_LGSEND |
2371 		    ((uint32_t)(*m_head)->m_pkthdr.tso_segsz <<
2372 		    RL_TDESC_CMD_MSSVAL_SHIFT);
2373 	else {
2374 		/*
2375 		 * Unconditionally enable IP checksum if TCP or UDP
2376 		 * checksum is required. Otherwise, TCP/UDP checksum
2377 		 * does't make effects.
2378 		 */
2379 		if (((*m_head)->m_pkthdr.csum_flags & RE_CSUM_FEATURES) != 0) {
2380 			if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
2381 				csum_flags |= RL_TDESC_CMD_IPCSUM;
2382 				if (((*m_head)->m_pkthdr.csum_flags &
2383 				    CSUM_TCP) != 0)
2384 					csum_flags |= RL_TDESC_CMD_TCPCSUM;
2385 				if (((*m_head)->m_pkthdr.csum_flags &
2386 				    CSUM_UDP) != 0)
2387 					csum_flags |= RL_TDESC_CMD_UDPCSUM;
2388 			} else {
2389 				vlanctl |= RL_TDESC_CMD_IPCSUMV2;
2390 				if (((*m_head)->m_pkthdr.csum_flags &
2391 				    CSUM_TCP) != 0)
2392 					vlanctl |= RL_TDESC_CMD_TCPCSUMV2;
2393 				if (((*m_head)->m_pkthdr.csum_flags &
2394 				    CSUM_UDP) != 0)
2395 					vlanctl |= RL_TDESC_CMD_UDPCSUMV2;
2396 			}
2397 		}
2398 	}
2399 
2400 	/*
2401 	 * Set up hardware VLAN tagging. Note: vlan tag info must
2402 	 * appear in all descriptors of a multi-descriptor
2403 	 * transmission attempt.
2404 	 */
2405 	if ((*m_head)->m_flags & M_VLANTAG)
2406 		vlanctl |= bswap16((*m_head)->m_pkthdr.ether_vtag) |
2407 		    RL_TDESC_VLANCTL_TAG;
2408 
2409 	si = prod;
2410 	for (i = 0; i < nsegs; i++, prod = RL_TX_DESC_NXT(sc, prod)) {
2411 		desc = &sc->rl_ldata.rl_tx_list[prod];
2412 		desc->rl_vlanctl = htole32(vlanctl);
2413 		desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[i].ds_addr));
2414 		desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[i].ds_addr));
2415 		cmdstat = segs[i].ds_len;
2416 		if (i != 0)
2417 			cmdstat |= RL_TDESC_CMD_OWN;
2418 		if (prod == sc->rl_ldata.rl_tx_desc_cnt - 1)
2419 			cmdstat |= RL_TDESC_CMD_EOR;
2420 		desc->rl_cmdstat = htole32(cmdstat | csum_flags);
2421 		sc->rl_ldata.rl_tx_free--;
2422 	}
2423 	/* Update producer index. */
2424 	sc->rl_ldata.rl_tx_prodidx = prod;
2425 
2426 	/* Set EOF on the last descriptor. */
2427 	ei = RL_TX_DESC_PRV(sc, prod);
2428 	desc = &sc->rl_ldata.rl_tx_list[ei];
2429 	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_EOF);
2430 
2431 	desc = &sc->rl_ldata.rl_tx_list[si];
2432 	/* Set SOF and transfer ownership of packet to the chip. */
2433 	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_OWN | RL_TDESC_CMD_SOF);
2434 
2435 	/*
2436 	 * Insure that the map for this transmission
2437 	 * is placed at the array index of the last descriptor
2438 	 * in this chain.  (Swap last and first dmamaps.)
2439 	 */
2440 	txd_last = &sc->rl_ldata.rl_tx_desc[ei];
2441 	map = txd->tx_dmamap;
2442 	txd->tx_dmamap = txd_last->tx_dmamap;
2443 	txd_last->tx_dmamap = map;
2444 	txd_last->tx_m = *m_head;
2445 
2446 	return (0);
2447 }
2448 
2449 static void
2450 re_tx_task(void *arg, int npending)
2451 {
2452 	struct ifnet		*ifp;
2453 
2454 	ifp = arg;
2455 	re_start(ifp);
2456 }
2457 
2458 /*
2459  * Main transmit routine for C+ and gigE NICs.
2460  */
2461 static void
2462 re_start(struct ifnet *ifp)
2463 {
2464 	struct rl_softc		*sc;
2465 	struct mbuf		*m_head;
2466 	int			queued;
2467 
2468 	sc = ifp->if_softc;
2469 
2470 	RL_LOCK(sc);
2471 
2472 	if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
2473 	    IFF_DRV_RUNNING || (sc->rl_flags & RL_FLAG_LINK) == 0) {
2474 		RL_UNLOCK(sc);
2475 		return;
2476 	}
2477 
2478 	for (queued = 0; !IFQ_DRV_IS_EMPTY(&ifp->if_snd) &&
2479 	    sc->rl_ldata.rl_tx_free > 1;) {
2480 		IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
2481 		if (m_head == NULL)
2482 			break;
2483 
2484 		if (re_encap(sc, &m_head) != 0) {
2485 			if (m_head == NULL)
2486 				break;
2487 			IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
2488 			ifp->if_drv_flags |= IFF_DRV_OACTIVE;
2489 			break;
2490 		}
2491 
2492 		/*
2493 		 * If there's a BPF listener, bounce a copy of this frame
2494 		 * to him.
2495 		 */
2496 		ETHER_BPF_MTAP(ifp, m_head);
2497 
2498 		queued++;
2499 	}
2500 
2501 	if (queued == 0) {
2502 #ifdef RE_TX_MODERATION
2503 		if (sc->rl_ldata.rl_tx_free != sc->rl_ldata.rl_tx_desc_cnt)
2504 			CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2505 #endif
2506 		RL_UNLOCK(sc);
2507 		return;
2508 	}
2509 
2510 	/* Flush the TX descriptors */
2511 
2512 	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2513 	    sc->rl_ldata.rl_tx_list_map,
2514 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2515 
2516 	CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2517 
2518 #ifdef RE_TX_MODERATION
2519 	/*
2520 	 * Use the countdown timer for interrupt moderation.
2521 	 * 'TX done' interrupts are disabled. Instead, we reset the
2522 	 * countdown timer, which will begin counting until it hits
2523 	 * the value in the TIMERINT register, and then trigger an
2524 	 * interrupt. Each time we write to the TIMERCNT register,
2525 	 * the timer count is reset to 0.
2526 	 */
2527 	CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2528 #endif
2529 
2530 	/*
2531 	 * Set a timeout in case the chip goes out to lunch.
2532 	 */
2533 	sc->rl_watchdog_timer = 5;
2534 
2535 	RL_UNLOCK(sc);
2536 }
2537 
2538 static void
2539 re_init(void *xsc)
2540 {
2541 	struct rl_softc		*sc = xsc;
2542 
2543 	RL_LOCK(sc);
2544 	re_init_locked(sc);
2545 	RL_UNLOCK(sc);
2546 }
2547 
2548 static void
2549 re_init_locked(struct rl_softc *sc)
2550 {
2551 	struct ifnet		*ifp = sc->rl_ifp;
2552 	struct mii_data		*mii;
2553 	uint32_t		reg;
2554 	uint16_t		cfg;
2555 	union {
2556 		uint32_t align_dummy;
2557 		u_char eaddr[ETHER_ADDR_LEN];
2558         } eaddr;
2559 
2560 	RL_LOCK_ASSERT(sc);
2561 
2562 	mii = device_get_softc(sc->rl_miibus);
2563 
2564 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
2565 		return;
2566 
2567 	/*
2568 	 * Cancel pending I/O and free all RX/TX buffers.
2569 	 */
2570 	re_stop(sc);
2571 
2572 	/* Put controller into known state. */
2573 	re_reset(sc);
2574 
2575 	/*
2576 	 * Enable C+ RX and TX mode, as well as VLAN stripping and
2577 	 * RX checksum offload. We must configure the C+ register
2578 	 * before all others.
2579 	 */
2580 	cfg = RL_CPLUSCMD_PCI_MRW;
2581 	if ((ifp->if_capenable & IFCAP_RXCSUM) != 0)
2582 		cfg |= RL_CPLUSCMD_RXCSUM_ENB;
2583 	if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0)
2584 		cfg |= RL_CPLUSCMD_VLANSTRIP;
2585 	if ((sc->rl_flags & RL_FLAG_MACSTAT) != 0) {
2586 		cfg |= RL_CPLUSCMD_MACSTAT_DIS;
2587 		/* XXX magic. */
2588 		cfg |= 0x0001;
2589 	} else
2590 		cfg |= RL_CPLUSCMD_RXENB | RL_CPLUSCMD_TXENB;
2591 	CSR_WRITE_2(sc, RL_CPLUS_CMD, cfg);
2592 	if (sc->rl_hwrev == RL_HWREV_8169_8110SC ||
2593 	    sc->rl_hwrev == RL_HWREV_8169_8110SCE) {
2594 		reg = 0x000fff00;
2595 		if ((CSR_READ_1(sc, RL_CFG2) & RL_CFG2_PCI66MHZ) != 0)
2596 			reg |= 0x000000ff;
2597 		if (sc->rl_hwrev == RL_HWREV_8169_8110SCE)
2598 			reg |= 0x00f00000;
2599 		CSR_WRITE_4(sc, 0x7c, reg);
2600 		/* Disable interrupt mitigation. */
2601 		CSR_WRITE_2(sc, 0xe2, 0);
2602 	}
2603 	/*
2604 	 * Disable TSO if interface MTU size is greater than MSS
2605 	 * allowed in controller.
2606 	 */
2607 	if (ifp->if_mtu > RL_TSO_MTU && (ifp->if_capenable & IFCAP_TSO4) != 0) {
2608 		ifp->if_capenable &= ~IFCAP_TSO4;
2609 		ifp->if_hwassist &= ~CSUM_TSO;
2610 	}
2611 
2612 	/*
2613 	 * Init our MAC address.  Even though the chipset
2614 	 * documentation doesn't mention it, we need to enter "Config
2615 	 * register write enable" mode to modify the ID registers.
2616 	 */
2617 	/* Copy MAC address on stack to align. */
2618 	bcopy(IF_LLADDR(ifp), eaddr.eaddr, ETHER_ADDR_LEN);
2619 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG);
2620 	CSR_WRITE_4(sc, RL_IDR0,
2621 	    htole32(*(u_int32_t *)(&eaddr.eaddr[0])));
2622 	CSR_WRITE_4(sc, RL_IDR4,
2623 	    htole32(*(u_int32_t *)(&eaddr.eaddr[4])));
2624 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
2625 
2626 	/*
2627 	 * For C+ mode, initialize the RX descriptors and mbufs.
2628 	 */
2629 	re_rx_list_init(sc);
2630 	re_tx_list_init(sc);
2631 
2632 	/*
2633 	 * Load the addresses of the RX and TX lists into the chip.
2634 	 */
2635 
2636 	CSR_WRITE_4(sc, RL_RXLIST_ADDR_HI,
2637 	    RL_ADDR_HI(sc->rl_ldata.rl_rx_list_addr));
2638 	CSR_WRITE_4(sc, RL_RXLIST_ADDR_LO,
2639 	    RL_ADDR_LO(sc->rl_ldata.rl_rx_list_addr));
2640 
2641 	CSR_WRITE_4(sc, RL_TXLIST_ADDR_HI,
2642 	    RL_ADDR_HI(sc->rl_ldata.rl_tx_list_addr));
2643 	CSR_WRITE_4(sc, RL_TXLIST_ADDR_LO,
2644 	    RL_ADDR_LO(sc->rl_ldata.rl_tx_list_addr));
2645 
2646 	/*
2647 	 * Enable transmit and receive.
2648 	 */
2649 	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
2650 
2651 	/*
2652 	 * Set the initial TX configuration.
2653 	 */
2654 	if (sc->rl_testmode) {
2655 		if (sc->rl_type == RL_8169)
2656 			CSR_WRITE_4(sc, RL_TXCFG,
2657 			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON);
2658 		else
2659 			CSR_WRITE_4(sc, RL_TXCFG,
2660 			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON_CPLUS);
2661 	} else
2662 		CSR_WRITE_4(sc, RL_TXCFG, RL_TXCFG_CONFIG);
2663 
2664 	CSR_WRITE_1(sc, RL_EARLY_TX_THRESH, 16);
2665 
2666 	/*
2667 	 * Set the initial RX configuration.
2668 	 */
2669 	re_set_rxmode(sc);
2670 
2671 #ifdef DEVICE_POLLING
2672 	/*
2673 	 * Disable interrupts if we are polling.
2674 	 */
2675 	if (ifp->if_capenable & IFCAP_POLLING)
2676 		CSR_WRITE_2(sc, RL_IMR, 0);
2677 	else	/* otherwise ... */
2678 #endif
2679 
2680 	/*
2681 	 * Enable interrupts.
2682 	 */
2683 	if (sc->rl_testmode)
2684 		CSR_WRITE_2(sc, RL_IMR, 0);
2685 	else
2686 		CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2687 	CSR_WRITE_2(sc, RL_ISR, RL_INTRS_CPLUS);
2688 
2689 	/* Set initial TX threshold */
2690 	sc->rl_txthresh = RL_TX_THRESH_INIT;
2691 
2692 	/* Start RX/TX process. */
2693 	CSR_WRITE_4(sc, RL_MISSEDPKT, 0);
2694 #ifdef notdef
2695 	/* Enable receiver and transmitter. */
2696 	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
2697 #endif
2698 
2699 #ifdef RE_TX_MODERATION
2700 	/*
2701 	 * Initialize the timer interrupt register so that
2702 	 * a timer interrupt will be generated once the timer
2703 	 * reaches a certain number of ticks. The timer is
2704 	 * reloaded on each transmit. This gives us TX interrupt
2705 	 * moderation, which dramatically improves TX frame rate.
2706 	 */
2707 	if (sc->rl_type == RL_8169)
2708 		CSR_WRITE_4(sc, RL_TIMERINT_8169, 0x800);
2709 	else
2710 		CSR_WRITE_4(sc, RL_TIMERINT, 0x400);
2711 #endif
2712 
2713 	/*
2714 	 * For 8169 gigE NICs, set the max allowed RX packet
2715 	 * size so we can receive jumbo frames.
2716 	 */
2717 	if (sc->rl_type == RL_8169)
2718 		CSR_WRITE_2(sc, RL_MAXRXPKTLEN, 16383);
2719 
2720 	if (sc->rl_testmode)
2721 		return;
2722 
2723 	mii_mediachg(mii);
2724 
2725 	CSR_WRITE_1(sc, RL_CFG1, CSR_READ_1(sc, RL_CFG1) | RL_CFG1_DRVLOAD);
2726 
2727 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
2728 	ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
2729 
2730 	sc->rl_flags &= ~RL_FLAG_LINK;
2731 	sc->rl_watchdog_timer = 0;
2732 	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
2733 }
2734 
2735 /*
2736  * Set media options.
2737  */
2738 static int
2739 re_ifmedia_upd(struct ifnet *ifp)
2740 {
2741 	struct rl_softc		*sc;
2742 	struct mii_data		*mii;
2743 	int			error;
2744 
2745 	sc = ifp->if_softc;
2746 	mii = device_get_softc(sc->rl_miibus);
2747 	RL_LOCK(sc);
2748 	error = mii_mediachg(mii);
2749 	RL_UNLOCK(sc);
2750 
2751 	return (error);
2752 }
2753 
2754 /*
2755  * Report current media status.
2756  */
2757 static void
2758 re_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
2759 {
2760 	struct rl_softc		*sc;
2761 	struct mii_data		*mii;
2762 
2763 	sc = ifp->if_softc;
2764 	mii = device_get_softc(sc->rl_miibus);
2765 
2766 	RL_LOCK(sc);
2767 	mii_pollstat(mii);
2768 	RL_UNLOCK(sc);
2769 	ifmr->ifm_active = mii->mii_media_active;
2770 	ifmr->ifm_status = mii->mii_media_status;
2771 }
2772 
2773 static int
2774 re_ioctl(struct ifnet *ifp, u_long command, caddr_t data)
2775 {
2776 	struct rl_softc		*sc = ifp->if_softc;
2777 	struct ifreq		*ifr = (struct ifreq *) data;
2778 	struct mii_data		*mii;
2779 	int			error = 0;
2780 
2781 	switch (command) {
2782 	case SIOCSIFMTU:
2783 		if (ifr->ifr_mtu < ETHERMIN || ifr->ifr_mtu > RL_JUMBO_MTU) {
2784 			error = EINVAL;
2785 			break;
2786 		}
2787 		if ((sc->rl_flags & RL_FLAG_NOJUMBO) != 0 &&
2788 		    ifr->ifr_mtu > RL_MAX_FRAMELEN) {
2789 			error = EINVAL;
2790 			break;
2791 		}
2792 		RL_LOCK(sc);
2793 		if (ifp->if_mtu != ifr->ifr_mtu)
2794 			ifp->if_mtu = ifr->ifr_mtu;
2795 		if (ifp->if_mtu > RL_TSO_MTU &&
2796 		    (ifp->if_capenable & IFCAP_TSO4) != 0) {
2797 			ifp->if_capenable &= ~IFCAP_TSO4;
2798 			ifp->if_hwassist &= ~CSUM_TSO;
2799 			VLAN_CAPABILITIES(ifp);
2800 		}
2801 		RL_UNLOCK(sc);
2802 		break;
2803 	case SIOCSIFFLAGS:
2804 		RL_LOCK(sc);
2805 		if ((ifp->if_flags & IFF_UP) != 0) {
2806 			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
2807 				if (((ifp->if_flags ^ sc->rl_if_flags)
2808 				    & (IFF_PROMISC | IFF_ALLMULTI)) != 0)
2809 					re_set_rxmode(sc);
2810 			} else
2811 				re_init_locked(sc);
2812 		} else {
2813 			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
2814 				re_stop(sc);
2815 		}
2816 		sc->rl_if_flags = ifp->if_flags;
2817 		RL_UNLOCK(sc);
2818 		break;
2819 	case SIOCADDMULTI:
2820 	case SIOCDELMULTI:
2821 		RL_LOCK(sc);
2822 		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
2823 			re_set_rxmode(sc);
2824 		RL_UNLOCK(sc);
2825 		break;
2826 	case SIOCGIFMEDIA:
2827 	case SIOCSIFMEDIA:
2828 		mii = device_get_softc(sc->rl_miibus);
2829 		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, command);
2830 		break;
2831 	case SIOCSIFCAP:
2832 	    {
2833 		int mask, reinit;
2834 
2835 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
2836 		reinit = 0;
2837 #ifdef DEVICE_POLLING
2838 		if (mask & IFCAP_POLLING) {
2839 			if (ifr->ifr_reqcap & IFCAP_POLLING) {
2840 				error = ether_poll_register(re_poll, ifp);
2841 				if (error)
2842 					return(error);
2843 				RL_LOCK(sc);
2844 				/* Disable interrupts */
2845 				CSR_WRITE_2(sc, RL_IMR, 0x0000);
2846 				ifp->if_capenable |= IFCAP_POLLING;
2847 				RL_UNLOCK(sc);
2848 			} else {
2849 				error = ether_poll_deregister(ifp);
2850 				/* Enable interrupts. */
2851 				RL_LOCK(sc);
2852 				CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2853 				ifp->if_capenable &= ~IFCAP_POLLING;
2854 				RL_UNLOCK(sc);
2855 			}
2856 		}
2857 #endif /* DEVICE_POLLING */
2858 		if (mask & IFCAP_HWCSUM) {
2859 			ifp->if_capenable ^= IFCAP_HWCSUM;
2860 			if (ifp->if_capenable & IFCAP_TXCSUM)
2861 				ifp->if_hwassist |= RE_CSUM_FEATURES;
2862 			else
2863 				ifp->if_hwassist &= ~RE_CSUM_FEATURES;
2864 			reinit = 1;
2865 		}
2866 		if ((mask & IFCAP_TSO4) != 0 &&
2867 		    (ifp->if_capabilities & IFCAP_TSO) != 0) {
2868 			ifp->if_capenable ^= IFCAP_TSO4;
2869 			if ((IFCAP_TSO4 & ifp->if_capenable) != 0)
2870 				ifp->if_hwassist |= CSUM_TSO;
2871 			else
2872 				ifp->if_hwassist &= ~CSUM_TSO;
2873 			if (ifp->if_mtu > RL_TSO_MTU &&
2874 			    (ifp->if_capenable & IFCAP_TSO4) != 0) {
2875 				ifp->if_capenable &= ~IFCAP_TSO4;
2876 				ifp->if_hwassist &= ~CSUM_TSO;
2877 			}
2878 		}
2879 		if ((mask & IFCAP_VLAN_HWTSO) != 0 &&
2880 		    (ifp->if_capabilities & IFCAP_VLAN_HWTSO) != 0)
2881 			ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
2882 		if ((mask & IFCAP_VLAN_HWTAGGING) != 0 &&
2883 		    (ifp->if_capabilities & IFCAP_VLAN_HWTAGGING) != 0) {
2884 			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
2885 			/* TSO over VLAN requires VLAN hardware tagging. */
2886 			if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) == 0)
2887 				ifp->if_capenable &= ~IFCAP_VLAN_HWTSO;
2888 			reinit = 1;
2889 		}
2890 		if ((mask & IFCAP_WOL) != 0 &&
2891 		    (ifp->if_capabilities & IFCAP_WOL) != 0) {
2892 			if ((mask & IFCAP_WOL_UCAST) != 0)
2893 				ifp->if_capenable ^= IFCAP_WOL_UCAST;
2894 			if ((mask & IFCAP_WOL_MCAST) != 0)
2895 				ifp->if_capenable ^= IFCAP_WOL_MCAST;
2896 			if ((mask & IFCAP_WOL_MAGIC) != 0)
2897 				ifp->if_capenable ^= IFCAP_WOL_MAGIC;
2898 		}
2899 		if (reinit && ifp->if_drv_flags & IFF_DRV_RUNNING) {
2900 			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2901 			re_init(sc);
2902 		}
2903 		VLAN_CAPABILITIES(ifp);
2904 	    }
2905 		break;
2906 	default:
2907 		error = ether_ioctl(ifp, command, data);
2908 		break;
2909 	}
2910 
2911 	return (error);
2912 }
2913 
2914 static void
2915 re_watchdog(struct rl_softc *sc)
2916 {
2917 	struct ifnet		*ifp;
2918 
2919 	RL_LOCK_ASSERT(sc);
2920 
2921 	if (sc->rl_watchdog_timer == 0 || --sc->rl_watchdog_timer != 0)
2922 		return;
2923 
2924 	ifp = sc->rl_ifp;
2925 	re_txeof(sc);
2926 	if (sc->rl_ldata.rl_tx_free == sc->rl_ldata.rl_tx_desc_cnt) {
2927 		if_printf(ifp, "watchdog timeout (missed Tx interrupts) "
2928 		    "-- recovering\n");
2929 		if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2930 			taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_txtask);
2931 		return;
2932 	}
2933 
2934 	if_printf(ifp, "watchdog timeout\n");
2935 	ifp->if_oerrors++;
2936 
2937 	re_rxeof(sc, NULL);
2938 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2939 	re_init_locked(sc);
2940 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2941 		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_txtask);
2942 }
2943 
2944 /*
2945  * Stop the adapter and free any mbufs allocated to the
2946  * RX and TX lists.
2947  */
2948 static void
2949 re_stop(struct rl_softc *sc)
2950 {
2951 	int			i;
2952 	struct ifnet		*ifp;
2953 	struct rl_txdesc	*txd;
2954 	struct rl_rxdesc	*rxd;
2955 
2956 	RL_LOCK_ASSERT(sc);
2957 
2958 	ifp = sc->rl_ifp;
2959 
2960 	sc->rl_watchdog_timer = 0;
2961 	callout_stop(&sc->rl_stat_callout);
2962 	ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
2963 
2964 	if ((sc->rl_flags & RL_FLAG_CMDSTOP) != 0)
2965 		CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_STOPREQ | RL_CMD_TX_ENB |
2966 		    RL_CMD_RX_ENB);
2967 	else
2968 		CSR_WRITE_1(sc, RL_COMMAND, 0x00);
2969 	DELAY(1000);
2970 	CSR_WRITE_2(sc, RL_IMR, 0x0000);
2971 	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
2972 
2973 	if (sc->rl_head != NULL) {
2974 		m_freem(sc->rl_head);
2975 		sc->rl_head = sc->rl_tail = NULL;
2976 	}
2977 
2978 	/* Free the TX list buffers. */
2979 
2980 	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
2981 		txd = &sc->rl_ldata.rl_tx_desc[i];
2982 		if (txd->tx_m != NULL) {
2983 			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
2984 			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
2985 			bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag,
2986 			    txd->tx_dmamap);
2987 			m_freem(txd->tx_m);
2988 			txd->tx_m = NULL;
2989 		}
2990 	}
2991 
2992 	/* Free the RX list buffers. */
2993 
2994 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
2995 		rxd = &sc->rl_ldata.rl_rx_desc[i];
2996 		if (rxd->rx_m != NULL) {
2997 			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
2998 			    rxd->rx_dmamap, BUS_DMASYNC_POSTREAD);
2999 			bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag,
3000 			    rxd->rx_dmamap);
3001 			m_freem(rxd->rx_m);
3002 			rxd->rx_m = NULL;
3003 		}
3004 	}
3005 }
3006 
3007 /*
3008  * Device suspend routine.  Stop the interface and save some PCI
3009  * settings in case the BIOS doesn't restore them properly on
3010  * resume.
3011  */
3012 static int
3013 re_suspend(device_t dev)
3014 {
3015 	struct rl_softc		*sc;
3016 
3017 	sc = device_get_softc(dev);
3018 
3019 	RL_LOCK(sc);
3020 	re_stop(sc);
3021 	re_setwol(sc);
3022 	sc->suspended = 1;
3023 	RL_UNLOCK(sc);
3024 
3025 	return (0);
3026 }
3027 
3028 /*
3029  * Device resume routine.  Restore some PCI settings in case the BIOS
3030  * doesn't, re-enable busmastering, and restart the interface if
3031  * appropriate.
3032  */
3033 static int
3034 re_resume(device_t dev)
3035 {
3036 	struct rl_softc		*sc;
3037 	struct ifnet		*ifp;
3038 
3039 	sc = device_get_softc(dev);
3040 
3041 	RL_LOCK(sc);
3042 
3043 	ifp = sc->rl_ifp;
3044 	/* Take controller out of sleep mode. */
3045 	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
3046 		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
3047 			CSR_WRITE_1(sc, RL_GPIO,
3048 			    CSR_READ_1(sc, RL_GPIO) | 0x01);
3049 	}
3050 
3051 	/*
3052 	 * Clear WOL matching such that normal Rx filtering
3053 	 * wouldn't interfere with WOL patterns.
3054 	 */
3055 	re_clrwol(sc);
3056 
3057 	/* reinitialize interface if necessary */
3058 	if (ifp->if_flags & IFF_UP)
3059 		re_init_locked(sc);
3060 
3061 	sc->suspended = 0;
3062 	RL_UNLOCK(sc);
3063 
3064 	return (0);
3065 }
3066 
3067 /*
3068  * Stop all chip I/O so that the kernel's probe routines don't
3069  * get confused by errant DMAs when rebooting.
3070  */
3071 static int
3072 re_shutdown(device_t dev)
3073 {
3074 	struct rl_softc		*sc;
3075 
3076 	sc = device_get_softc(dev);
3077 
3078 	RL_LOCK(sc);
3079 	re_stop(sc);
3080 	/*
3081 	 * Mark interface as down since otherwise we will panic if
3082 	 * interrupt comes in later on, which can happen in some
3083 	 * cases.
3084 	 */
3085 	sc->rl_ifp->if_flags &= ~IFF_UP;
3086 	re_setwol(sc);
3087 	RL_UNLOCK(sc);
3088 
3089 	return (0);
3090 }
3091 
3092 static void
3093 re_setwol(struct rl_softc *sc)
3094 {
3095 	struct ifnet		*ifp;
3096 	int			pmc;
3097 	uint16_t		pmstat;
3098 	uint8_t			v;
3099 
3100 	RL_LOCK_ASSERT(sc);
3101 
3102 	if (pci_find_extcap(sc->rl_dev, PCIY_PMG, &pmc) != 0)
3103 		return;
3104 
3105 	ifp = sc->rl_ifp;
3106 	/* Put controller into sleep mode. */
3107 	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
3108 		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
3109 			CSR_WRITE_1(sc, RL_GPIO,
3110 			    CSR_READ_1(sc, RL_GPIO) & ~0x01);
3111 	}
3112 	if ((ifp->if_capenable & IFCAP_WOL) != 0 &&
3113 	    (sc->rl_flags & RL_FLAG_WOLRXENB) != 0)
3114 		CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RX_ENB);
3115 	/* Enable config register write. */
3116 	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
3117 
3118 	/* Enable PME. */
3119 	v = CSR_READ_1(sc, RL_CFG1);
3120 	v &= ~RL_CFG1_PME;
3121 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3122 		v |= RL_CFG1_PME;
3123 	CSR_WRITE_1(sc, RL_CFG1, v);
3124 
3125 	v = CSR_READ_1(sc, RL_CFG3);
3126 	v &= ~(RL_CFG3_WOL_LINK | RL_CFG3_WOL_MAGIC);
3127 	if ((ifp->if_capenable & IFCAP_WOL_MAGIC) != 0)
3128 		v |= RL_CFG3_WOL_MAGIC;
3129 	CSR_WRITE_1(sc, RL_CFG3, v);
3130 
3131 	/* Config register write done. */
3132 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3133 
3134 	v = CSR_READ_1(sc, RL_CFG5);
3135 	v &= ~(RL_CFG5_WOL_BCAST | RL_CFG5_WOL_MCAST | RL_CFG5_WOL_UCAST);
3136 	v &= ~RL_CFG5_WOL_LANWAKE;
3137 	if ((ifp->if_capenable & IFCAP_WOL_UCAST) != 0)
3138 		v |= RL_CFG5_WOL_UCAST;
3139 	if ((ifp->if_capenable & IFCAP_WOL_MCAST) != 0)
3140 		v |= RL_CFG5_WOL_MCAST | RL_CFG5_WOL_BCAST;
3141 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3142 		v |= RL_CFG5_WOL_LANWAKE;
3143 	CSR_WRITE_1(sc, RL_CFG5, v);
3144 
3145 	if ((ifp->if_capenable & IFCAP_WOL) != 0 &&
3146 	    (sc->rl_flags & RL_FLAG_PHYWAKE_PM) != 0)
3147 		CSR_WRITE_1(sc, RL_PMCH, CSR_READ_1(sc, RL_PMCH) & ~0x80);
3148 	/*
3149 	 * It seems that hardware resets its link speed to 100Mbps in
3150 	 * power down mode so switching to 100Mbps in driver is not
3151 	 * needed.
3152 	 */
3153 
3154 	/* Request PME if WOL is requested. */
3155 	pmstat = pci_read_config(sc->rl_dev, pmc + PCIR_POWER_STATUS, 2);
3156 	pmstat &= ~(PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE);
3157 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3158 		pmstat |= PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE;
3159 	pci_write_config(sc->rl_dev, pmc + PCIR_POWER_STATUS, pmstat, 2);
3160 }
3161 
3162 static void
3163 re_clrwol(struct rl_softc *sc)
3164 {
3165 	int			pmc;
3166 	uint8_t			v;
3167 
3168 	RL_LOCK_ASSERT(sc);
3169 
3170 	if (pci_find_extcap(sc->rl_dev, PCIY_PMG, &pmc) != 0)
3171 		return;
3172 
3173 	/* Enable config register write. */
3174 	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
3175 
3176 	v = CSR_READ_1(sc, RL_CFG3);
3177 	v &= ~(RL_CFG3_WOL_LINK | RL_CFG3_WOL_MAGIC);
3178 	CSR_WRITE_1(sc, RL_CFG3, v);
3179 
3180 	/* Config register write done. */
3181 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3182 
3183 	v = CSR_READ_1(sc, RL_CFG5);
3184 	v &= ~(RL_CFG5_WOL_BCAST | RL_CFG5_WOL_MCAST | RL_CFG5_WOL_UCAST);
3185 	v &= ~RL_CFG5_WOL_LANWAKE;
3186 	CSR_WRITE_1(sc, RL_CFG5, v);
3187 }
3188