xref: /freebsd/sys/dev/re/if_re.c (revision dadef94c7a762d05890e2891bc4a7d1dfe0cf758)
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 	sc = device_get_softc(dev);
413 
414 	/* Let the rgephy driver read the GMEDIASTAT register */
415 
416 	if (reg == RL_GMEDIASTAT) {
417 		rval = CSR_READ_1(sc, RL_GMEDIASTAT);
418 		return (rval);
419 	}
420 
421 	CSR_WRITE_4(sc, RL_PHYAR, reg << 16);
422 	DELAY(1000);
423 
424 	for (i = 0; i < RL_PHY_TIMEOUT; i++) {
425 		rval = CSR_READ_4(sc, RL_PHYAR);
426 		if (rval & RL_PHYAR_BUSY)
427 			break;
428 		DELAY(100);
429 	}
430 
431 	if (i == RL_PHY_TIMEOUT) {
432 		device_printf(sc->rl_dev, "PHY read failed\n");
433 		return (0);
434 	}
435 
436 	return (rval & RL_PHYAR_PHYDATA);
437 }
438 
439 static int
440 re_gmii_writereg(device_t dev, int phy, int reg, int data)
441 {
442 	struct rl_softc		*sc;
443 	u_int32_t		rval;
444 	int			i;
445 
446 	sc = device_get_softc(dev);
447 
448 	CSR_WRITE_4(sc, RL_PHYAR, (reg << 16) |
449 	    (data & RL_PHYAR_PHYDATA) | RL_PHYAR_BUSY);
450 	DELAY(1000);
451 
452 	for (i = 0; i < RL_PHY_TIMEOUT; i++) {
453 		rval = CSR_READ_4(sc, RL_PHYAR);
454 		if (!(rval & RL_PHYAR_BUSY))
455 			break;
456 		DELAY(100);
457 	}
458 
459 	if (i == RL_PHY_TIMEOUT) {
460 		device_printf(sc->rl_dev, "PHY write failed\n");
461 		return (0);
462 	}
463 
464 	return (0);
465 }
466 
467 static int
468 re_miibus_readreg(device_t dev, int phy, int reg)
469 {
470 	struct rl_softc		*sc;
471 	u_int16_t		rval = 0;
472 	u_int16_t		re8139_reg = 0;
473 
474 	sc = device_get_softc(dev);
475 
476 	if (sc->rl_type == RL_8169) {
477 		rval = re_gmii_readreg(dev, phy, reg);
478 		return (rval);
479 	}
480 
481 	switch (reg) {
482 	case MII_BMCR:
483 		re8139_reg = RL_BMCR;
484 		break;
485 	case MII_BMSR:
486 		re8139_reg = RL_BMSR;
487 		break;
488 	case MII_ANAR:
489 		re8139_reg = RL_ANAR;
490 		break;
491 	case MII_ANER:
492 		re8139_reg = RL_ANER;
493 		break;
494 	case MII_ANLPAR:
495 		re8139_reg = RL_LPAR;
496 		break;
497 	case MII_PHYIDR1:
498 	case MII_PHYIDR2:
499 		return (0);
500 	/*
501 	 * Allow the rlphy driver to read the media status
502 	 * register. If we have a link partner which does not
503 	 * support NWAY, this is the register which will tell
504 	 * us the results of parallel detection.
505 	 */
506 	case RL_MEDIASTAT:
507 		rval = CSR_READ_1(sc, RL_MEDIASTAT);
508 		return (rval);
509 	default:
510 		device_printf(sc->rl_dev, "bad phy register\n");
511 		return (0);
512 	}
513 	rval = CSR_READ_2(sc, re8139_reg);
514 	if (sc->rl_type == RL_8139CPLUS && re8139_reg == RL_BMCR) {
515 		/* 8139C+ has different bit layout. */
516 		rval &= ~(BMCR_LOOP | BMCR_ISO);
517 	}
518 	return (rval);
519 }
520 
521 static int
522 re_miibus_writereg(device_t dev, int phy, int reg, int data)
523 {
524 	struct rl_softc		*sc;
525 	u_int16_t		re8139_reg = 0;
526 	int			rval = 0;
527 
528 	sc = device_get_softc(dev);
529 
530 	if (sc->rl_type == RL_8169) {
531 		rval = re_gmii_writereg(dev, phy, reg, data);
532 		return (rval);
533 	}
534 
535 	switch (reg) {
536 	case MII_BMCR:
537 		re8139_reg = RL_BMCR;
538 		if (sc->rl_type == RL_8139CPLUS) {
539 			/* 8139C+ has different bit layout. */
540 			data &= ~(BMCR_LOOP | BMCR_ISO);
541 		}
542 		break;
543 	case MII_BMSR:
544 		re8139_reg = RL_BMSR;
545 		break;
546 	case MII_ANAR:
547 		re8139_reg = RL_ANAR;
548 		break;
549 	case MII_ANER:
550 		re8139_reg = RL_ANER;
551 		break;
552 	case MII_ANLPAR:
553 		re8139_reg = RL_LPAR;
554 		break;
555 	case MII_PHYIDR1:
556 	case MII_PHYIDR2:
557 		return (0);
558 		break;
559 	default:
560 		device_printf(sc->rl_dev, "bad phy register\n");
561 		return (0);
562 	}
563 	CSR_WRITE_2(sc, re8139_reg, data);
564 	return (0);
565 }
566 
567 static void
568 re_miibus_statchg(device_t dev)
569 {
570 	struct rl_softc		*sc;
571 	struct ifnet		*ifp;
572 	struct mii_data		*mii;
573 
574 	sc = device_get_softc(dev);
575 	mii = device_get_softc(sc->rl_miibus);
576 	ifp = sc->rl_ifp;
577 	if (mii == NULL || ifp == NULL ||
578 	    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
579 		return;
580 
581 	sc->rl_flags &= ~RL_FLAG_LINK;
582 	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
583 	    (IFM_ACTIVE | IFM_AVALID)) {
584 		switch (IFM_SUBTYPE(mii->mii_media_active)) {
585 		case IFM_10_T:
586 		case IFM_100_TX:
587 			sc->rl_flags |= RL_FLAG_LINK;
588 			break;
589 		case IFM_1000_T:
590 			if ((sc->rl_flags & RL_FLAG_FASTETHER) != 0)
591 				break;
592 			sc->rl_flags |= RL_FLAG_LINK;
593 			break;
594 		default:
595 			break;
596 		}
597 	}
598 	/*
599 	 * RealTek controllers does not provide any interface to
600 	 * Tx/Rx MACs for resolved speed, duplex and flow-control
601 	 * parameters.
602 	 */
603 }
604 
605 /*
606  * Set the RX configuration and 64-bit multicast hash filter.
607  */
608 static void
609 re_set_rxmode(struct rl_softc *sc)
610 {
611 	struct ifnet		*ifp;
612 	struct ifmultiaddr	*ifma;
613 	uint32_t		hashes[2] = { 0, 0 };
614 	uint32_t		h, rxfilt;
615 
616 	RL_LOCK_ASSERT(sc);
617 
618 	ifp = sc->rl_ifp;
619 
620 	rxfilt = RL_RXCFG_CONFIG | RL_RXCFG_RX_INDIV | RL_RXCFG_RX_BROAD;
621 
622 	if (ifp->if_flags & (IFF_ALLMULTI | IFF_PROMISC)) {
623 		if (ifp->if_flags & IFF_PROMISC)
624 			rxfilt |= RL_RXCFG_RX_ALLPHYS;
625 		/*
626 		 * Unlike other hardwares, we have to explicitly set
627 		 * RL_RXCFG_RX_MULTI to receive multicast frames in
628 		 * promiscuous mode.
629 		 */
630 		rxfilt |= RL_RXCFG_RX_MULTI;
631 		hashes[0] = hashes[1] = 0xffffffff;
632 		goto done;
633 	}
634 
635 	if_maddr_rlock(ifp);
636 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
637 		if (ifma->ifma_addr->sa_family != AF_LINK)
638 			continue;
639 		h = ether_crc32_be(LLADDR((struct sockaddr_dl *)
640 		    ifma->ifma_addr), ETHER_ADDR_LEN) >> 26;
641 		if (h < 32)
642 			hashes[0] |= (1 << h);
643 		else
644 			hashes[1] |= (1 << (h - 32));
645 	}
646 	if_maddr_runlock(ifp);
647 
648 	if (hashes[0] != 0 || hashes[1] != 0) {
649 		/*
650 		 * For some unfathomable reason, RealTek decided to
651 		 * reverse the order of the multicast hash registers
652 		 * in the PCI Express parts.  This means we have to
653 		 * write the hash pattern in reverse order for those
654 		 * devices.
655 		 */
656 		if ((sc->rl_flags & RL_FLAG_PCIE) != 0) {
657 			h = bswap32(hashes[0]);
658 			hashes[0] = bswap32(hashes[1]);
659 			hashes[1] = h;
660 		}
661 		rxfilt |= RL_RXCFG_RX_MULTI;
662 	}
663 
664 done:
665 	CSR_WRITE_4(sc, RL_MAR0, hashes[0]);
666 	CSR_WRITE_4(sc, RL_MAR4, hashes[1]);
667 	CSR_WRITE_4(sc, RL_RXCFG, rxfilt);
668 }
669 
670 static void
671 re_reset(struct rl_softc *sc)
672 {
673 	int			i;
674 
675 	RL_LOCK_ASSERT(sc);
676 
677 	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RESET);
678 
679 	for (i = 0; i < RL_TIMEOUT; i++) {
680 		DELAY(10);
681 		if (!(CSR_READ_1(sc, RL_COMMAND) & RL_CMD_RESET))
682 			break;
683 	}
684 	if (i == RL_TIMEOUT)
685 		device_printf(sc->rl_dev, "reset never completed!\n");
686 
687 	if ((sc->rl_flags & RL_FLAG_MACRESET) != 0)
688 		CSR_WRITE_1(sc, 0x82, 1);
689 	if (sc->rl_hwrev == RL_HWREV_8169S)
690 		re_gmii_writereg(sc->rl_dev, 1, 0x0b, 0);
691 }
692 
693 #ifdef RE_DIAG
694 
695 /*
696  * The following routine is designed to test for a defect on some
697  * 32-bit 8169 cards. Some of these NICs have the REQ64# and ACK64#
698  * lines connected to the bus, however for a 32-bit only card, they
699  * should be pulled high. The result of this defect is that the
700  * NIC will not work right if you plug it into a 64-bit slot: DMA
701  * operations will be done with 64-bit transfers, which will fail
702  * because the 64-bit data lines aren't connected.
703  *
704  * There's no way to work around this (short of talking a soldering
705  * iron to the board), however we can detect it. The method we use
706  * here is to put the NIC into digital loopback mode, set the receiver
707  * to promiscuous mode, and then try to send a frame. We then compare
708  * the frame data we sent to what was received. If the data matches,
709  * then the NIC is working correctly, otherwise we know the user has
710  * a defective NIC which has been mistakenly plugged into a 64-bit PCI
711  * slot. In the latter case, there's no way the NIC can work correctly,
712  * so we print out a message on the console and abort the device attach.
713  */
714 
715 static int
716 re_diag(struct rl_softc *sc)
717 {
718 	struct ifnet		*ifp = sc->rl_ifp;
719 	struct mbuf		*m0;
720 	struct ether_header	*eh;
721 	struct rl_desc		*cur_rx;
722 	u_int16_t		status;
723 	u_int32_t		rxstat;
724 	int			total_len, i, error = 0, phyaddr;
725 	u_int8_t		dst[] = { 0x00, 'h', 'e', 'l', 'l', 'o' };
726 	u_int8_t		src[] = { 0x00, 'w', 'o', 'r', 'l', 'd' };
727 
728 	/* Allocate a single mbuf */
729 	MGETHDR(m0, M_DONTWAIT, MT_DATA);
730 	if (m0 == NULL)
731 		return (ENOBUFS);
732 
733 	RL_LOCK(sc);
734 
735 	/*
736 	 * Initialize the NIC in test mode. This sets the chip up
737 	 * so that it can send and receive frames, but performs the
738 	 * following special functions:
739 	 * - Puts receiver in promiscuous mode
740 	 * - Enables digital loopback mode
741 	 * - Leaves interrupts turned off
742 	 */
743 
744 	ifp->if_flags |= IFF_PROMISC;
745 	sc->rl_testmode = 1;
746 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
747 	re_init_locked(sc);
748 	sc->rl_flags |= RL_FLAG_LINK;
749 	if (sc->rl_type == RL_8169)
750 		phyaddr = 1;
751 	else
752 		phyaddr = 0;
753 
754 	re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_RESET);
755 	for (i = 0; i < RL_TIMEOUT; i++) {
756 		status = re_miibus_readreg(sc->rl_dev, phyaddr, MII_BMCR);
757 		if (!(status & BMCR_RESET))
758 			break;
759 	}
760 
761 	re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_LOOP);
762 	CSR_WRITE_2(sc, RL_ISR, RL_INTRS);
763 
764 	DELAY(100000);
765 
766 	/* Put some data in the mbuf */
767 
768 	eh = mtod(m0, struct ether_header *);
769 	bcopy ((char *)&dst, eh->ether_dhost, ETHER_ADDR_LEN);
770 	bcopy ((char *)&src, eh->ether_shost, ETHER_ADDR_LEN);
771 	eh->ether_type = htons(ETHERTYPE_IP);
772 	m0->m_pkthdr.len = m0->m_len = ETHER_MIN_LEN - ETHER_CRC_LEN;
773 
774 	/*
775 	 * Queue the packet, start transmission.
776 	 * Note: IF_HANDOFF() ultimately calls re_start() for us.
777 	 */
778 
779 	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
780 	RL_UNLOCK(sc);
781 	/* XXX: re_diag must not be called when in ALTQ mode */
782 	IF_HANDOFF(&ifp->if_snd, m0, ifp);
783 	RL_LOCK(sc);
784 	m0 = NULL;
785 
786 	/* Wait for it to propagate through the chip */
787 
788 	DELAY(100000);
789 	for (i = 0; i < RL_TIMEOUT; i++) {
790 		status = CSR_READ_2(sc, RL_ISR);
791 		CSR_WRITE_2(sc, RL_ISR, status);
792 		if ((status & (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK)) ==
793 		    (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK))
794 			break;
795 		DELAY(10);
796 	}
797 
798 	if (i == RL_TIMEOUT) {
799 		device_printf(sc->rl_dev,
800 		    "diagnostic failed, failed to receive packet in"
801 		    " loopback mode\n");
802 		error = EIO;
803 		goto done;
804 	}
805 
806 	/*
807 	 * The packet should have been dumped into the first
808 	 * entry in the RX DMA ring. Grab it from there.
809 	 */
810 
811 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
812 	    sc->rl_ldata.rl_rx_list_map,
813 	    BUS_DMASYNC_POSTREAD);
814 	bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag,
815 	    sc->rl_ldata.rl_rx_desc[0].rx_dmamap,
816 	    BUS_DMASYNC_POSTREAD);
817 	bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag,
818 	    sc->rl_ldata.rl_rx_desc[0].rx_dmamap);
819 
820 	m0 = sc->rl_ldata.rl_rx_desc[0].rx_m;
821 	sc->rl_ldata.rl_rx_desc[0].rx_m = NULL;
822 	eh = mtod(m0, struct ether_header *);
823 
824 	cur_rx = &sc->rl_ldata.rl_rx_list[0];
825 	total_len = RL_RXBYTES(cur_rx);
826 	rxstat = le32toh(cur_rx->rl_cmdstat);
827 
828 	if (total_len != ETHER_MIN_LEN) {
829 		device_printf(sc->rl_dev,
830 		    "diagnostic failed, received short packet\n");
831 		error = EIO;
832 		goto done;
833 	}
834 
835 	/* Test that the received packet data matches what we sent. */
836 
837 	if (bcmp((char *)&eh->ether_dhost, (char *)&dst, ETHER_ADDR_LEN) ||
838 	    bcmp((char *)&eh->ether_shost, (char *)&src, ETHER_ADDR_LEN) ||
839 	    ntohs(eh->ether_type) != ETHERTYPE_IP) {
840 		device_printf(sc->rl_dev, "WARNING, DMA FAILURE!\n");
841 		device_printf(sc->rl_dev, "expected TX data: %6D/%6D/0x%x\n",
842 		    dst, ":", src, ":", ETHERTYPE_IP);
843 		device_printf(sc->rl_dev, "received RX data: %6D/%6D/0x%x\n",
844 		    eh->ether_dhost, ":",  eh->ether_shost, ":",
845 		    ntohs(eh->ether_type));
846 		device_printf(sc->rl_dev, "You may have a defective 32-bit "
847 		    "NIC plugged into a 64-bit PCI slot.\n");
848 		device_printf(sc->rl_dev, "Please re-install the NIC in a "
849 		    "32-bit slot for proper operation.\n");
850 		device_printf(sc->rl_dev, "Read the re(4) man page for more "
851 		    "details.\n");
852 		error = EIO;
853 	}
854 
855 done:
856 	/* Turn interface off, release resources */
857 
858 	sc->rl_testmode = 0;
859 	sc->rl_flags &= ~RL_FLAG_LINK;
860 	ifp->if_flags &= ~IFF_PROMISC;
861 	re_stop(sc);
862 	if (m0 != NULL)
863 		m_freem(m0);
864 
865 	RL_UNLOCK(sc);
866 
867 	return (error);
868 }
869 
870 #endif
871 
872 /*
873  * Probe for a RealTek 8139C+/8169/8110 chip. Check the PCI vendor and device
874  * IDs against our list and return a device name if we find a match.
875  */
876 static int
877 re_probe(device_t dev)
878 {
879 	struct rl_type		*t;
880 	uint16_t		devid, vendor;
881 	uint16_t		revid, sdevid;
882 	int			i;
883 
884 	vendor = pci_get_vendor(dev);
885 	devid = pci_get_device(dev);
886 	revid = pci_get_revid(dev);
887 	sdevid = pci_get_subdevice(dev);
888 
889 	if (vendor == LINKSYS_VENDORID && devid == LINKSYS_DEVICEID_EG1032) {
890 		if (sdevid != LINKSYS_SUBDEVICE_EG1032_REV3) {
891 			/*
892 			 * Only attach to rev. 3 of the Linksys EG1032 adapter.
893 			 * Rev. 2 is supported by sk(4).
894 			 */
895 			return (ENXIO);
896 		}
897 	}
898 
899 	if (vendor == RT_VENDORID && devid == RT_DEVICEID_8139) {
900 		if (revid != 0x20) {
901 			/* 8139, let rl(4) take care of this device. */
902 			return (ENXIO);
903 		}
904 	}
905 
906 	t = re_devs;
907 	for (i = 0; i < sizeof(re_devs) / sizeof(re_devs[0]); i++, t++) {
908 		if (vendor == t->rl_vid && devid == t->rl_did) {
909 			device_set_desc(dev, t->rl_name);
910 			return (BUS_PROBE_DEFAULT);
911 		}
912 	}
913 
914 	return (ENXIO);
915 }
916 
917 /*
918  * Map a single buffer address.
919  */
920 
921 static void
922 re_dma_map_addr(void *arg, bus_dma_segment_t *segs, int nseg, int error)
923 {
924 	bus_addr_t		*addr;
925 
926 	if (error)
927 		return;
928 
929 	KASSERT(nseg == 1, ("too many DMA segments, %d should be 1", nseg));
930 	addr = arg;
931 	*addr = segs->ds_addr;
932 }
933 
934 static int
935 re_allocmem(device_t dev, struct rl_softc *sc)
936 {
937 	bus_size_t		rx_list_size, tx_list_size;
938 	int			error;
939 	int			i;
940 
941 	rx_list_size = sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc);
942 	tx_list_size = sc->rl_ldata.rl_tx_desc_cnt * sizeof(struct rl_desc);
943 
944 	/*
945 	 * Allocate the parent bus DMA tag appropriate for PCI.
946 	 * In order to use DAC, RL_CPLUSCMD_PCI_DAC bit of RL_CPLUS_CMD
947 	 * register should be set. However some RealTek chips are known
948 	 * to be buggy on DAC handling, therefore disable DAC by limiting
949 	 * DMA address space to 32bit. PCIe variants of RealTek chips
950 	 * may not have the limitation but I took safer path.
951 	 */
952 	error = bus_dma_tag_create(bus_get_dma_tag(dev), 1, 0,
953 	    BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL, NULL,
954 	    BUS_SPACE_MAXSIZE_32BIT, 0, BUS_SPACE_MAXSIZE_32BIT, 0,
955 	    NULL, NULL, &sc->rl_parent_tag);
956 	if (error) {
957 		device_printf(dev, "could not allocate parent DMA tag\n");
958 		return (error);
959 	}
960 
961 	/*
962 	 * Allocate map for TX mbufs.
963 	 */
964 	error = bus_dma_tag_create(sc->rl_parent_tag, 1, 0,
965 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL,
966 	    NULL, MCLBYTES * RL_NTXSEGS, RL_NTXSEGS, 4096, 0,
967 	    NULL, NULL, &sc->rl_ldata.rl_tx_mtag);
968 	if (error) {
969 		device_printf(dev, "could not allocate TX DMA tag\n");
970 		return (error);
971 	}
972 
973 	/*
974 	 * Allocate map for RX mbufs.
975 	 */
976 
977 	error = bus_dma_tag_create(sc->rl_parent_tag, sizeof(uint64_t), 0,
978 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
979 	    MCLBYTES, 1, MCLBYTES, 0, NULL, NULL, &sc->rl_ldata.rl_rx_mtag);
980 	if (error) {
981 		device_printf(dev, "could not allocate RX DMA tag\n");
982 		return (error);
983 	}
984 
985 	/*
986 	 * Allocate map for TX descriptor list.
987 	 */
988 	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
989 	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
990 	    NULL, tx_list_size, 1, tx_list_size, 0,
991 	    NULL, NULL, &sc->rl_ldata.rl_tx_list_tag);
992 	if (error) {
993 		device_printf(dev, "could not allocate TX DMA ring tag\n");
994 		return (error);
995 	}
996 
997 	/* Allocate DMA'able memory for the TX ring */
998 
999 	error = bus_dmamem_alloc(sc->rl_ldata.rl_tx_list_tag,
1000 	    (void **)&sc->rl_ldata.rl_tx_list,
1001 	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1002 	    &sc->rl_ldata.rl_tx_list_map);
1003 	if (error) {
1004 		device_printf(dev, "could not allocate TX DMA ring\n");
1005 		return (error);
1006 	}
1007 
1008 	/* Load the map for the TX ring. */
1009 
1010 	sc->rl_ldata.rl_tx_list_addr = 0;
1011 	error = bus_dmamap_load(sc->rl_ldata.rl_tx_list_tag,
1012 	     sc->rl_ldata.rl_tx_list_map, sc->rl_ldata.rl_tx_list,
1013 	     tx_list_size, re_dma_map_addr,
1014 	     &sc->rl_ldata.rl_tx_list_addr, BUS_DMA_NOWAIT);
1015 	if (error != 0 || sc->rl_ldata.rl_tx_list_addr == 0) {
1016 		device_printf(dev, "could not load TX DMA ring\n");
1017 		return (ENOMEM);
1018 	}
1019 
1020 	/* Create DMA maps for TX buffers */
1021 
1022 	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
1023 		error = bus_dmamap_create(sc->rl_ldata.rl_tx_mtag, 0,
1024 		    &sc->rl_ldata.rl_tx_desc[i].tx_dmamap);
1025 		if (error) {
1026 			device_printf(dev, "could not create DMA map for TX\n");
1027 			return (error);
1028 		}
1029 	}
1030 
1031 	/*
1032 	 * Allocate map for RX descriptor list.
1033 	 */
1034 	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
1035 	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1036 	    NULL, rx_list_size, 1, rx_list_size, 0,
1037 	    NULL, NULL, &sc->rl_ldata.rl_rx_list_tag);
1038 	if (error) {
1039 		device_printf(dev, "could not create RX DMA ring tag\n");
1040 		return (error);
1041 	}
1042 
1043 	/* Allocate DMA'able memory for the RX ring */
1044 
1045 	error = bus_dmamem_alloc(sc->rl_ldata.rl_rx_list_tag,
1046 	    (void **)&sc->rl_ldata.rl_rx_list,
1047 	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1048 	    &sc->rl_ldata.rl_rx_list_map);
1049 	if (error) {
1050 		device_printf(dev, "could not allocate RX DMA ring\n");
1051 		return (error);
1052 	}
1053 
1054 	/* Load the map for the RX ring. */
1055 
1056 	sc->rl_ldata.rl_rx_list_addr = 0;
1057 	error = bus_dmamap_load(sc->rl_ldata.rl_rx_list_tag,
1058 	     sc->rl_ldata.rl_rx_list_map, sc->rl_ldata.rl_rx_list,
1059 	     rx_list_size, re_dma_map_addr,
1060 	     &sc->rl_ldata.rl_rx_list_addr, BUS_DMA_NOWAIT);
1061 	if (error != 0 || sc->rl_ldata.rl_rx_list_addr == 0) {
1062 		device_printf(dev, "could not load RX DMA ring\n");
1063 		return (ENOMEM);
1064 	}
1065 
1066 	/* Create DMA maps for RX buffers */
1067 
1068 	error = bus_dmamap_create(sc->rl_ldata.rl_rx_mtag, 0,
1069 	    &sc->rl_ldata.rl_rx_sparemap);
1070 	if (error) {
1071 		device_printf(dev, "could not create spare DMA map for RX\n");
1072 		return (error);
1073 	}
1074 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1075 		error = bus_dmamap_create(sc->rl_ldata.rl_rx_mtag, 0,
1076 		    &sc->rl_ldata.rl_rx_desc[i].rx_dmamap);
1077 		if (error) {
1078 			device_printf(dev, "could not create DMA map for RX\n");
1079 			return (error);
1080 		}
1081 	}
1082 
1083 	return (0);
1084 }
1085 
1086 /*
1087  * Attach the interface. Allocate softc structures, do ifmedia
1088  * setup and ethernet/BPF attach.
1089  */
1090 static int
1091 re_attach(device_t dev)
1092 {
1093 	u_char			eaddr[ETHER_ADDR_LEN];
1094 	u_int16_t		as[ETHER_ADDR_LEN / 2];
1095 	struct rl_softc		*sc;
1096 	struct ifnet		*ifp;
1097 	struct rl_hwrev		*hw_rev;
1098 	int			hwrev;
1099 	u_int16_t		devid, re_did = 0;
1100 	int			error = 0, i, phy, rid;
1101 	int			msic, reg;
1102 	uint8_t			cfg;
1103 
1104 	sc = device_get_softc(dev);
1105 	sc->rl_dev = dev;
1106 
1107 	mtx_init(&sc->rl_mtx, device_get_nameunit(dev), MTX_NETWORK_LOCK,
1108 	    MTX_DEF);
1109 	callout_init_mtx(&sc->rl_stat_callout, &sc->rl_mtx, 0);
1110 
1111 	/*
1112 	 * Map control/status registers.
1113 	 */
1114 	pci_enable_busmaster(dev);
1115 
1116 	devid = pci_get_device(dev);
1117 	/*
1118 	 * Prefer memory space register mapping over IO space.
1119 	 * Because RTL8169SC does not seem to work when memory mapping
1120 	 * is used always activate io mapping.
1121 	 */
1122 	if (devid == RT_DEVICEID_8169SC)
1123 		prefer_iomap = 1;
1124 	if (prefer_iomap == 0) {
1125 		sc->rl_res_id = PCIR_BAR(1);
1126 		sc->rl_res_type = SYS_RES_MEMORY;
1127 		/* RTL8168/8101E seems to use different BARs. */
1128 		if (devid == RT_DEVICEID_8168 || devid == RT_DEVICEID_8101E)
1129 			sc->rl_res_id = PCIR_BAR(2);
1130 	} else {
1131 		sc->rl_res_id = PCIR_BAR(0);
1132 		sc->rl_res_type = SYS_RES_IOPORT;
1133 	}
1134 	sc->rl_res = bus_alloc_resource_any(dev, sc->rl_res_type,
1135 	    &sc->rl_res_id, RF_ACTIVE);
1136 	if (sc->rl_res == NULL && prefer_iomap == 0) {
1137 		sc->rl_res_id = PCIR_BAR(0);
1138 		sc->rl_res_type = SYS_RES_IOPORT;
1139 		sc->rl_res = bus_alloc_resource_any(dev, sc->rl_res_type,
1140 		    &sc->rl_res_id, RF_ACTIVE);
1141 	}
1142 	if (sc->rl_res == NULL) {
1143 		device_printf(dev, "couldn't map ports/memory\n");
1144 		error = ENXIO;
1145 		goto fail;
1146 	}
1147 
1148 	sc->rl_btag = rman_get_bustag(sc->rl_res);
1149 	sc->rl_bhandle = rman_get_bushandle(sc->rl_res);
1150 
1151 	msic = 0;
1152 	if (pci_find_extcap(dev, PCIY_EXPRESS, &reg) == 0) {
1153 		sc->rl_flags |= RL_FLAG_PCIE;
1154 		if (devid != RT_DEVICEID_8101E) {
1155 			/* Set PCIe maximum read request size to 2048. */
1156 			if (pci_get_max_read_req(dev) < 2048)
1157 				pci_set_max_read_req(dev, 2048);
1158 		}
1159 		msic = pci_msi_count(dev);
1160 		if (bootverbose)
1161 			device_printf(dev, "MSI count : %d\n", msic);
1162 	}
1163 	if (msic > 0 && msi_disable == 0) {
1164 		msic = 1;
1165 		if (pci_alloc_msi(dev, &msic) == 0) {
1166 			if (msic == RL_MSI_MESSAGES) {
1167 				device_printf(dev, "Using %d MSI messages\n",
1168 				    msic);
1169 				sc->rl_flags |= RL_FLAG_MSI;
1170 				/* Explicitly set MSI enable bit. */
1171 				CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1172 				cfg = CSR_READ_1(sc, RL_CFG2);
1173 				cfg |= RL_CFG2_MSI;
1174 				CSR_WRITE_1(sc, RL_CFG2, cfg);
1175 				CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1176 			} else
1177 				pci_release_msi(dev);
1178 		}
1179 	}
1180 
1181 	/* Allocate interrupt */
1182 	if ((sc->rl_flags & RL_FLAG_MSI) == 0) {
1183 		rid = 0;
1184 		sc->rl_irq[0] = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
1185 		    RF_SHAREABLE | RF_ACTIVE);
1186 		if (sc->rl_irq[0] == NULL) {
1187 			device_printf(dev, "couldn't allocate IRQ resources\n");
1188 			error = ENXIO;
1189 			goto fail;
1190 		}
1191 	} else {
1192 		for (i = 0, rid = 1; i < RL_MSI_MESSAGES; i++, rid++) {
1193 			sc->rl_irq[i] = bus_alloc_resource_any(dev,
1194 			    SYS_RES_IRQ, &rid, RF_ACTIVE);
1195 			if (sc->rl_irq[i] == NULL) {
1196 				device_printf(dev,
1197 				    "couldn't llocate IRQ resources for "
1198 				    "message %d\n", rid);
1199 				error = ENXIO;
1200 				goto fail;
1201 			}
1202 		}
1203 	}
1204 
1205 	if ((sc->rl_flags & RL_FLAG_MSI) == 0) {
1206 		CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1207 		cfg = CSR_READ_1(sc, RL_CFG2);
1208 		if ((cfg & RL_CFG2_MSI) != 0) {
1209 			device_printf(dev, "turning off MSI enable bit.\n");
1210 			cfg &= ~RL_CFG2_MSI;
1211 			CSR_WRITE_1(sc, RL_CFG2, cfg);
1212 		}
1213 		CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1214 	}
1215 
1216 	/* Reset the adapter. */
1217 	RL_LOCK(sc);
1218 	re_reset(sc);
1219 	RL_UNLOCK(sc);
1220 
1221 	hw_rev = re_hwrevs;
1222 	hwrev = CSR_READ_4(sc, RL_TXCFG);
1223 	switch (hwrev & 0x70000000) {
1224 	case 0x00000000:
1225 	case 0x10000000:
1226 		device_printf(dev, "Chip rev. 0x%08x\n", hwrev & 0xfc800000);
1227 		hwrev &= (RL_TXCFG_HWREV | 0x80000000);
1228 		break;
1229 	default:
1230 		device_printf(dev, "Chip rev. 0x%08x\n", hwrev & 0x7c800000);
1231 		hwrev &= RL_TXCFG_HWREV;
1232 		break;
1233 	}
1234 	device_printf(dev, "MAC rev. 0x%08x\n", hwrev & 0x00700000);
1235 	while (hw_rev->rl_desc != NULL) {
1236 		if (hw_rev->rl_rev == hwrev) {
1237 			sc->rl_type = hw_rev->rl_type;
1238 			sc->rl_hwrev = hw_rev->rl_rev;
1239 			break;
1240 		}
1241 		hw_rev++;
1242 	}
1243 	if (hw_rev->rl_desc == NULL) {
1244 		device_printf(dev, "Unknown H/W revision: 0x%08x\n", hwrev);
1245 		error = ENXIO;
1246 		goto fail;
1247 	}
1248 
1249 	switch (hw_rev->rl_rev) {
1250 	case RL_HWREV_8139CPLUS:
1251 		sc->rl_flags |= RL_FLAG_NOJUMBO | RL_FLAG_FASTETHER |
1252 		    RL_FLAG_AUTOPAD;
1253 		break;
1254 	case RL_HWREV_8100E:
1255 	case RL_HWREV_8101E:
1256 		sc->rl_flags |= RL_FLAG_NOJUMBO | RL_FLAG_PHYWAKE |
1257 		    RL_FLAG_FASTETHER;
1258 		break;
1259 	case RL_HWREV_8102E:
1260 	case RL_HWREV_8102EL:
1261 	case RL_HWREV_8102EL_SPIN1:
1262 		sc->rl_flags |= RL_FLAG_NOJUMBO | RL_FLAG_PHYWAKE |
1263 		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
1264 		    RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD;
1265 		break;
1266 	case RL_HWREV_8103E:
1267 		sc->rl_flags |= RL_FLAG_NOJUMBO | RL_FLAG_PHYWAKE |
1268 		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
1269 		    RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD |
1270 		    RL_FLAG_MACSLEEP;
1271 		break;
1272 	case RL_HWREV_8168_SPIN1:
1273 	case RL_HWREV_8168_SPIN2:
1274 		sc->rl_flags |= RL_FLAG_WOLRXENB;
1275 		/* FALLTHROUGH */
1276 	case RL_HWREV_8168_SPIN3:
1277 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_MACSTAT;
1278 		break;
1279 	case RL_HWREV_8168C_SPIN2:
1280 		sc->rl_flags |= RL_FLAG_MACSLEEP;
1281 		/* FALLTHROUGH */
1282 	case RL_HWREV_8168C:
1283 		if ((hwrev & 0x00700000) == 0x00200000)
1284 			sc->rl_flags |= RL_FLAG_MACSLEEP;
1285 		/* FALLTHROUGH */
1286 	case RL_HWREV_8168CP:
1287 	case RL_HWREV_8168D:
1288 	case RL_HWREV_8168DP:
1289 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
1290 		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP |
1291 		    RL_FLAG_AUTOPAD;
1292 		/*
1293 		 * These controllers support jumbo frame but it seems
1294 		 * that enabling it requires touching additional magic
1295 		 * registers. Depending on MAC revisions some
1296 		 * controllers need to disable checksum offload. So
1297 		 * disable jumbo frame until I have better idea what
1298 		 * it really requires to make it support.
1299 		 * RTL8168C/CP : supports up to 6KB jumbo frame.
1300 		 * RTL8111C/CP : supports up to 9KB jumbo frame.
1301 		 */
1302 		sc->rl_flags |= RL_FLAG_NOJUMBO;
1303 		break;
1304 	case RL_HWREV_8168E:
1305 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
1306 		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
1307 		    RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD | RL_FLAG_NOJUMBO;
1308 		break;
1309 	case RL_HWREV_8169_8110SB:
1310 	case RL_HWREV_8169_8110SBL:
1311 	case RL_HWREV_8169_8110SC:
1312 	case RL_HWREV_8169_8110SCE:
1313 		sc->rl_flags |= RL_FLAG_PHYWAKE;
1314 		/* FALLTHROUGH */
1315 	case RL_HWREV_8169:
1316 	case RL_HWREV_8169S:
1317 	case RL_HWREV_8110S:
1318 		sc->rl_flags |= RL_FLAG_MACRESET;
1319 		break;
1320 	default:
1321 		break;
1322 	}
1323 
1324 	/* Enable PME. */
1325 	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1326 	cfg = CSR_READ_1(sc, RL_CFG1);
1327 	cfg |= RL_CFG1_PME;
1328 	CSR_WRITE_1(sc, RL_CFG1, cfg);
1329 	cfg = CSR_READ_1(sc, RL_CFG5);
1330 	cfg &= RL_CFG5_PME_STS;
1331 	CSR_WRITE_1(sc, RL_CFG5, cfg);
1332 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1333 
1334 	if ((sc->rl_flags & RL_FLAG_PAR) != 0) {
1335 		/*
1336 		 * XXX Should have a better way to extract station
1337 		 * address from EEPROM.
1338 		 */
1339 		for (i = 0; i < ETHER_ADDR_LEN; i++)
1340 			eaddr[i] = CSR_READ_1(sc, RL_IDR0 + i);
1341 	} else {
1342 		sc->rl_eewidth = RL_9356_ADDR_LEN;
1343 		re_read_eeprom(sc, (caddr_t)&re_did, 0, 1);
1344 		if (re_did != 0x8129)
1345 			sc->rl_eewidth = RL_9346_ADDR_LEN;
1346 
1347 		/*
1348 		 * Get station address from the EEPROM.
1349 		 */
1350 		re_read_eeprom(sc, (caddr_t)as, RL_EE_EADDR, 3);
1351 		for (i = 0; i < ETHER_ADDR_LEN / 2; i++)
1352 			as[i] = le16toh(as[i]);
1353 		bcopy(as, eaddr, sizeof(eaddr));
1354 	}
1355 
1356 	if (sc->rl_type == RL_8169) {
1357 		/* Set RX length mask and number of descriptors. */
1358 		sc->rl_rxlenmask = RL_RDESC_STAT_GFRAGLEN;
1359 		sc->rl_txstart = RL_GTXSTART;
1360 		sc->rl_ldata.rl_tx_desc_cnt = RL_8169_TX_DESC_CNT;
1361 		sc->rl_ldata.rl_rx_desc_cnt = RL_8169_RX_DESC_CNT;
1362 	} else {
1363 		/* Set RX length mask and number of descriptors. */
1364 		sc->rl_rxlenmask = RL_RDESC_STAT_FRAGLEN;
1365 		sc->rl_txstart = RL_TXSTART;
1366 		sc->rl_ldata.rl_tx_desc_cnt = RL_8139_TX_DESC_CNT;
1367 		sc->rl_ldata.rl_rx_desc_cnt = RL_8139_RX_DESC_CNT;
1368 	}
1369 
1370 	error = re_allocmem(dev, sc);
1371 	if (error)
1372 		goto fail;
1373 
1374 	ifp = sc->rl_ifp = if_alloc(IFT_ETHER);
1375 	if (ifp == NULL) {
1376 		device_printf(dev, "can not if_alloc()\n");
1377 		error = ENOSPC;
1378 		goto fail;
1379 	}
1380 
1381 	/* Take controller out of deep sleep mode. */
1382 	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
1383 		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
1384 			CSR_WRITE_1(sc, RL_GPIO,
1385 			    CSR_READ_1(sc, RL_GPIO) | 0x01);
1386 		else
1387 			CSR_WRITE_1(sc, RL_GPIO,
1388 			    CSR_READ_1(sc, RL_GPIO) & ~0x01);
1389 	}
1390 
1391 	/* Take PHY out of power down mode. */
1392 	if ((sc->rl_flags & RL_FLAG_PHYWAKE_PM) != 0)
1393 		CSR_WRITE_1(sc, RL_PMCH, CSR_READ_1(sc, RL_PMCH) | 0x80);
1394 	if ((sc->rl_flags & RL_FLAG_PHYWAKE) != 0) {
1395 		re_gmii_writereg(dev, 1, 0x1f, 0);
1396 		re_gmii_writereg(dev, 1, 0x0e, 0);
1397 	}
1398 
1399 #define	RE_PHYAD_INTERNAL	 0
1400 
1401 	/* Do MII setup. */
1402 	phy = RE_PHYAD_INTERNAL;
1403 	if (sc->rl_type == RL_8169)
1404 		phy = 1;
1405 	error = mii_attach(dev, &sc->rl_miibus, ifp, re_ifmedia_upd,
1406 	    re_ifmedia_sts, BMSR_DEFCAPMASK, phy, MII_OFFSET_ANY, 0);
1407 	if (error != 0) {
1408 		device_printf(dev, "attaching PHYs failed\n");
1409 		goto fail;
1410 	}
1411 
1412 	ifp->if_softc = sc;
1413 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1414 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1415 	ifp->if_ioctl = re_ioctl;
1416 	ifp->if_start = re_start;
1417 	ifp->if_hwassist = RE_CSUM_FEATURES;
1418 	ifp->if_capabilities = IFCAP_HWCSUM;
1419 	ifp->if_capenable = ifp->if_capabilities;
1420 	ifp->if_init = re_init;
1421 	IFQ_SET_MAXLEN(&ifp->if_snd, RL_IFQ_MAXLEN);
1422 	ifp->if_snd.ifq_drv_maxlen = RL_IFQ_MAXLEN;
1423 	IFQ_SET_READY(&ifp->if_snd);
1424 
1425 	TASK_INIT(&sc->rl_txtask, 1, re_tx_task, ifp);
1426 	TASK_INIT(&sc->rl_inttask, 0, re_int_task, sc);
1427 
1428 	/*
1429 	 * XXX
1430 	 * Still have no idea how to make TSO work on 8168C, 8168CP,
1431 	 * 8111C and 8111CP.
1432 	 */
1433 	if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
1434 		ifp->if_hwassist |= CSUM_TSO;
1435 		ifp->if_capabilities |= IFCAP_TSO4 | IFCAP_VLAN_HWTSO;
1436 	}
1437 
1438 	/*
1439 	 * Call MI attach routine.
1440 	 */
1441 	ether_ifattach(ifp, eaddr);
1442 
1443 	/* VLAN capability setup */
1444 	ifp->if_capabilities |= IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING;
1445 	if (ifp->if_capabilities & IFCAP_HWCSUM)
1446 		ifp->if_capabilities |= IFCAP_VLAN_HWCSUM;
1447 	/* Enable WOL if PM is supported. */
1448 	if (pci_find_extcap(sc->rl_dev, PCIY_PMG, &reg) == 0)
1449 		ifp->if_capabilities |= IFCAP_WOL;
1450 	ifp->if_capenable = ifp->if_capabilities;
1451 	/*
1452 	 * Don't enable TSO by default. Under certain
1453 	 * circumtances the controller generated corrupted
1454 	 * packets in TSO size.
1455 	 */
1456 	ifp->if_hwassist &= ~CSUM_TSO;
1457 	ifp->if_capenable &= ~(IFCAP_TSO4 | IFCAP_VLAN_HWTSO);
1458 #ifdef DEVICE_POLLING
1459 	ifp->if_capabilities |= IFCAP_POLLING;
1460 #endif
1461 	/*
1462 	 * Tell the upper layer(s) we support long frames.
1463 	 * Must appear after the call to ether_ifattach() because
1464 	 * ether_ifattach() sets ifi_hdrlen to the default value.
1465 	 */
1466 	ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
1467 
1468 #ifdef RE_DIAG
1469 	/*
1470 	 * Perform hardware diagnostic on the original RTL8169.
1471 	 * Some 32-bit cards were incorrectly wired and would
1472 	 * malfunction if plugged into a 64-bit slot.
1473 	 */
1474 
1475 	if (hwrev == RL_HWREV_8169) {
1476 		error = re_diag(sc);
1477 		if (error) {
1478 			device_printf(dev,
1479 		    	"attach aborted due to hardware diag failure\n");
1480 			ether_ifdetach(ifp);
1481 			goto fail;
1482 		}
1483 	}
1484 #endif
1485 
1486 	/* Hook interrupt last to avoid having to lock softc */
1487 	if ((sc->rl_flags & RL_FLAG_MSI) == 0)
1488 		error = bus_setup_intr(dev, sc->rl_irq[0],
1489 		    INTR_TYPE_NET | INTR_MPSAFE, re_intr, NULL, sc,
1490 		    &sc->rl_intrhand[0]);
1491 	else {
1492 		for (i = 0; i < RL_MSI_MESSAGES; i++) {
1493 			error = bus_setup_intr(dev, sc->rl_irq[i],
1494 			    INTR_TYPE_NET | INTR_MPSAFE, re_intr, NULL, sc,
1495 		    	    &sc->rl_intrhand[i]);
1496 			if (error != 0)
1497 				break;
1498 		}
1499 	}
1500 	if (error) {
1501 		device_printf(dev, "couldn't set up irq\n");
1502 		ether_ifdetach(ifp);
1503 	}
1504 
1505 fail:
1506 
1507 	if (error)
1508 		re_detach(dev);
1509 
1510 	return (error);
1511 }
1512 
1513 /*
1514  * Shutdown hardware and free up resources. This can be called any
1515  * time after the mutex has been initialized. It is called in both
1516  * the error case in attach and the normal detach case so it needs
1517  * to be careful about only freeing resources that have actually been
1518  * allocated.
1519  */
1520 static int
1521 re_detach(device_t dev)
1522 {
1523 	struct rl_softc		*sc;
1524 	struct ifnet		*ifp;
1525 	int			i, rid;
1526 
1527 	sc = device_get_softc(dev);
1528 	ifp = sc->rl_ifp;
1529 	KASSERT(mtx_initialized(&sc->rl_mtx), ("re mutex not initialized"));
1530 
1531 	/* These should only be active if attach succeeded */
1532 	if (device_is_attached(dev)) {
1533 #ifdef DEVICE_POLLING
1534 		if (ifp->if_capenable & IFCAP_POLLING)
1535 			ether_poll_deregister(ifp);
1536 #endif
1537 		RL_LOCK(sc);
1538 #if 0
1539 		sc->suspended = 1;
1540 #endif
1541 		re_stop(sc);
1542 		RL_UNLOCK(sc);
1543 		callout_drain(&sc->rl_stat_callout);
1544 		taskqueue_drain(taskqueue_fast, &sc->rl_inttask);
1545 		taskqueue_drain(taskqueue_fast, &sc->rl_txtask);
1546 		/*
1547 		 * Force off the IFF_UP flag here, in case someone
1548 		 * still had a BPF descriptor attached to this
1549 		 * interface. If they do, ether_ifdetach() will cause
1550 		 * the BPF code to try and clear the promisc mode
1551 		 * flag, which will bubble down to re_ioctl(),
1552 		 * which will try to call re_init() again. This will
1553 		 * turn the NIC back on and restart the MII ticker,
1554 		 * which will panic the system when the kernel tries
1555 		 * to invoke the re_tick() function that isn't there
1556 		 * anymore.
1557 		 */
1558 		ifp->if_flags &= ~IFF_UP;
1559 		ether_ifdetach(ifp);
1560 	}
1561 	if (sc->rl_miibus)
1562 		device_delete_child(dev, sc->rl_miibus);
1563 	bus_generic_detach(dev);
1564 
1565 	/*
1566 	 * The rest is resource deallocation, so we should already be
1567 	 * stopped here.
1568 	 */
1569 
1570 	for (i = 0; i < RL_MSI_MESSAGES; i++) {
1571 		if (sc->rl_intrhand[i] != NULL) {
1572 			bus_teardown_intr(dev, sc->rl_irq[i],
1573 			    sc->rl_intrhand[i]);
1574 			sc->rl_intrhand[i] = NULL;
1575 		}
1576 	}
1577 	if (ifp != NULL)
1578 		if_free(ifp);
1579 	if ((sc->rl_flags & RL_FLAG_MSI) == 0) {
1580 		if (sc->rl_irq[0] != NULL) {
1581 			bus_release_resource(dev, SYS_RES_IRQ, 0,
1582 			    sc->rl_irq[0]);
1583 			sc->rl_irq[0] = NULL;
1584 		}
1585 	} else {
1586 		for (i = 0, rid = 1; i < RL_MSI_MESSAGES; i++, rid++) {
1587 			if (sc->rl_irq[i] != NULL) {
1588 				bus_release_resource(dev, SYS_RES_IRQ, rid,
1589 				    sc->rl_irq[i]);
1590 				sc->rl_irq[i] = NULL;
1591 			}
1592 		}
1593 		pci_release_msi(dev);
1594 	}
1595 	if (sc->rl_res)
1596 		bus_release_resource(dev, sc->rl_res_type, sc->rl_res_id,
1597 		    sc->rl_res);
1598 
1599 	/* Unload and free the RX DMA ring memory and map */
1600 
1601 	if (sc->rl_ldata.rl_rx_list_tag) {
1602 		bus_dmamap_unload(sc->rl_ldata.rl_rx_list_tag,
1603 		    sc->rl_ldata.rl_rx_list_map);
1604 		bus_dmamem_free(sc->rl_ldata.rl_rx_list_tag,
1605 		    sc->rl_ldata.rl_rx_list,
1606 		    sc->rl_ldata.rl_rx_list_map);
1607 		bus_dma_tag_destroy(sc->rl_ldata.rl_rx_list_tag);
1608 	}
1609 
1610 	/* Unload and free the TX DMA ring memory and map */
1611 
1612 	if (sc->rl_ldata.rl_tx_list_tag) {
1613 		bus_dmamap_unload(sc->rl_ldata.rl_tx_list_tag,
1614 		    sc->rl_ldata.rl_tx_list_map);
1615 		bus_dmamem_free(sc->rl_ldata.rl_tx_list_tag,
1616 		    sc->rl_ldata.rl_tx_list,
1617 		    sc->rl_ldata.rl_tx_list_map);
1618 		bus_dma_tag_destroy(sc->rl_ldata.rl_tx_list_tag);
1619 	}
1620 
1621 	/* Destroy all the RX and TX buffer maps */
1622 
1623 	if (sc->rl_ldata.rl_tx_mtag) {
1624 		for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++)
1625 			bus_dmamap_destroy(sc->rl_ldata.rl_tx_mtag,
1626 			    sc->rl_ldata.rl_tx_desc[i].tx_dmamap);
1627 		bus_dma_tag_destroy(sc->rl_ldata.rl_tx_mtag);
1628 	}
1629 	if (sc->rl_ldata.rl_rx_mtag) {
1630 		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++)
1631 			bus_dmamap_destroy(sc->rl_ldata.rl_rx_mtag,
1632 			    sc->rl_ldata.rl_rx_desc[i].rx_dmamap);
1633 		if (sc->rl_ldata.rl_rx_sparemap)
1634 			bus_dmamap_destroy(sc->rl_ldata.rl_rx_mtag,
1635 			    sc->rl_ldata.rl_rx_sparemap);
1636 		bus_dma_tag_destroy(sc->rl_ldata.rl_rx_mtag);
1637 	}
1638 
1639 	/* Unload and free the stats buffer and map */
1640 
1641 	if (sc->rl_ldata.rl_stag) {
1642 		bus_dmamap_unload(sc->rl_ldata.rl_stag,
1643 		    sc->rl_ldata.rl_rx_list_map);
1644 		bus_dmamem_free(sc->rl_ldata.rl_stag,
1645 		    sc->rl_ldata.rl_stats,
1646 		    sc->rl_ldata.rl_smap);
1647 		bus_dma_tag_destroy(sc->rl_ldata.rl_stag);
1648 	}
1649 
1650 	if (sc->rl_parent_tag)
1651 		bus_dma_tag_destroy(sc->rl_parent_tag);
1652 
1653 	mtx_destroy(&sc->rl_mtx);
1654 
1655 	return (0);
1656 }
1657 
1658 static __inline void
1659 re_discard_rxbuf(struct rl_softc *sc, int idx)
1660 {
1661 	struct rl_desc		*desc;
1662 	struct rl_rxdesc	*rxd;
1663 	uint32_t		cmdstat;
1664 
1665 	rxd = &sc->rl_ldata.rl_rx_desc[idx];
1666 	desc = &sc->rl_ldata.rl_rx_list[idx];
1667 	desc->rl_vlanctl = 0;
1668 	cmdstat = rxd->rx_size;
1669 	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
1670 		cmdstat |= RL_RDESC_CMD_EOR;
1671 	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
1672 }
1673 
1674 static int
1675 re_newbuf(struct rl_softc *sc, int idx)
1676 {
1677 	struct mbuf		*m;
1678 	struct rl_rxdesc	*rxd;
1679 	bus_dma_segment_t	segs[1];
1680 	bus_dmamap_t		map;
1681 	struct rl_desc		*desc;
1682 	uint32_t		cmdstat;
1683 	int			error, nsegs;
1684 
1685 	m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
1686 	if (m == NULL)
1687 		return (ENOBUFS);
1688 
1689 	m->m_len = m->m_pkthdr.len = MCLBYTES;
1690 #ifdef RE_FIXUP_RX
1691 	/*
1692 	 * This is part of an evil trick to deal with non-x86 platforms.
1693 	 * The RealTek chip requires RX buffers to be aligned on 64-bit
1694 	 * boundaries, but that will hose non-x86 machines. To get around
1695 	 * this, we leave some empty space at the start of each buffer
1696 	 * and for non-x86 hosts, we copy the buffer back six bytes
1697 	 * to achieve word alignment. This is slightly more efficient
1698 	 * than allocating a new buffer, copying the contents, and
1699 	 * discarding the old buffer.
1700 	 */
1701 	m_adj(m, RE_ETHER_ALIGN);
1702 #endif
1703 	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_rx_mtag,
1704 	    sc->rl_ldata.rl_rx_sparemap, m, segs, &nsegs, BUS_DMA_NOWAIT);
1705 	if (error != 0) {
1706 		m_freem(m);
1707 		return (ENOBUFS);
1708 	}
1709 	KASSERT(nsegs == 1, ("%s: %d segment returned!", __func__, nsegs));
1710 
1711 	rxd = &sc->rl_ldata.rl_rx_desc[idx];
1712 	if (rxd->rx_m != NULL) {
1713 		bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap,
1714 		    BUS_DMASYNC_POSTREAD);
1715 		bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap);
1716 	}
1717 
1718 	rxd->rx_m = m;
1719 	map = rxd->rx_dmamap;
1720 	rxd->rx_dmamap = sc->rl_ldata.rl_rx_sparemap;
1721 	rxd->rx_size = segs[0].ds_len;
1722 	sc->rl_ldata.rl_rx_sparemap = map;
1723 	bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap,
1724 	    BUS_DMASYNC_PREREAD);
1725 
1726 	desc = &sc->rl_ldata.rl_rx_list[idx];
1727 	desc->rl_vlanctl = 0;
1728 	desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[0].ds_addr));
1729 	desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[0].ds_addr));
1730 	cmdstat = segs[0].ds_len;
1731 	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
1732 		cmdstat |= RL_RDESC_CMD_EOR;
1733 	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
1734 
1735 	return (0);
1736 }
1737 
1738 #ifdef RE_FIXUP_RX
1739 static __inline void
1740 re_fixup_rx(struct mbuf *m)
1741 {
1742 	int                     i;
1743 	uint16_t                *src, *dst;
1744 
1745 	src = mtod(m, uint16_t *);
1746 	dst = src - (RE_ETHER_ALIGN - ETHER_ALIGN) / sizeof *src;
1747 
1748 	for (i = 0; i < (m->m_len / sizeof(uint16_t) + 1); i++)
1749 		*dst++ = *src++;
1750 
1751 	m->m_data -= RE_ETHER_ALIGN - ETHER_ALIGN;
1752 }
1753 #endif
1754 
1755 static int
1756 re_tx_list_init(struct rl_softc *sc)
1757 {
1758 	struct rl_desc		*desc;
1759 	int			i;
1760 
1761 	RL_LOCK_ASSERT(sc);
1762 
1763 	bzero(sc->rl_ldata.rl_tx_list,
1764 	    sc->rl_ldata.rl_tx_desc_cnt * sizeof(struct rl_desc));
1765 	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++)
1766 		sc->rl_ldata.rl_tx_desc[i].tx_m = NULL;
1767 	/* Set EOR. */
1768 	desc = &sc->rl_ldata.rl_tx_list[sc->rl_ldata.rl_tx_desc_cnt - 1];
1769 	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_EOR);
1770 
1771 	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
1772 	    sc->rl_ldata.rl_tx_list_map,
1773 	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1774 
1775 	sc->rl_ldata.rl_tx_prodidx = 0;
1776 	sc->rl_ldata.rl_tx_considx = 0;
1777 	sc->rl_ldata.rl_tx_free = sc->rl_ldata.rl_tx_desc_cnt;
1778 
1779 	return (0);
1780 }
1781 
1782 static int
1783 re_rx_list_init(struct rl_softc *sc)
1784 {
1785 	int			error, i;
1786 
1787 	bzero(sc->rl_ldata.rl_rx_list,
1788 	    sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc));
1789 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1790 		sc->rl_ldata.rl_rx_desc[i].rx_m = NULL;
1791 		if ((error = re_newbuf(sc, i)) != 0)
1792 			return (error);
1793 	}
1794 
1795 	/* Flush the RX descriptors */
1796 
1797 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1798 	    sc->rl_ldata.rl_rx_list_map,
1799 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
1800 
1801 	sc->rl_ldata.rl_rx_prodidx = 0;
1802 	sc->rl_head = sc->rl_tail = NULL;
1803 
1804 	return (0);
1805 }
1806 
1807 /*
1808  * RX handler for C+ and 8169. For the gigE chips, we support
1809  * the reception of jumbo frames that have been fragmented
1810  * across multiple 2K mbuf cluster buffers.
1811  */
1812 static int
1813 re_rxeof(struct rl_softc *sc, int *rx_npktsp)
1814 {
1815 	struct mbuf		*m;
1816 	struct ifnet		*ifp;
1817 	int			i, total_len;
1818 	struct rl_desc		*cur_rx;
1819 	u_int32_t		rxstat, rxvlan;
1820 	int			maxpkt = 16, rx_npkts = 0;
1821 
1822 	RL_LOCK_ASSERT(sc);
1823 
1824 	ifp = sc->rl_ifp;
1825 
1826 	/* Invalidate the descriptor memory */
1827 
1828 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1829 	    sc->rl_ldata.rl_rx_list_map,
1830 	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
1831 
1832 	for (i = sc->rl_ldata.rl_rx_prodidx; maxpkt > 0;
1833 	    i = RL_RX_DESC_NXT(sc, i)) {
1834 		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1835 			break;
1836 		cur_rx = &sc->rl_ldata.rl_rx_list[i];
1837 		rxstat = le32toh(cur_rx->rl_cmdstat);
1838 		if ((rxstat & RL_RDESC_STAT_OWN) != 0)
1839 			break;
1840 		total_len = rxstat & sc->rl_rxlenmask;
1841 		rxvlan = le32toh(cur_rx->rl_vlanctl);
1842 		m = sc->rl_ldata.rl_rx_desc[i].rx_m;
1843 
1844 		if (!(rxstat & RL_RDESC_STAT_EOF)) {
1845 			if (re_newbuf(sc, i) != 0) {
1846 				/*
1847 				 * If this is part of a multi-fragment packet,
1848 				 * discard all the pieces.
1849 				 */
1850 				if (sc->rl_head != NULL) {
1851 					m_freem(sc->rl_head);
1852 					sc->rl_head = sc->rl_tail = NULL;
1853 				}
1854 				re_discard_rxbuf(sc, i);
1855 				continue;
1856 			}
1857 			m->m_len = RE_RX_DESC_BUFLEN;
1858 			if (sc->rl_head == NULL)
1859 				sc->rl_head = sc->rl_tail = m;
1860 			else {
1861 				m->m_flags &= ~M_PKTHDR;
1862 				sc->rl_tail->m_next = m;
1863 				sc->rl_tail = m;
1864 			}
1865 			continue;
1866 		}
1867 
1868 		/*
1869 		 * NOTE: for the 8139C+, the frame length field
1870 		 * is always 12 bits in size, but for the gigE chips,
1871 		 * it is 13 bits (since the max RX frame length is 16K).
1872 		 * Unfortunately, all 32 bits in the status word
1873 		 * were already used, so to make room for the extra
1874 		 * length bit, RealTek took out the 'frame alignment
1875 		 * error' bit and shifted the other status bits
1876 		 * over one slot. The OWN, EOR, FS and LS bits are
1877 		 * still in the same places. We have already extracted
1878 		 * the frame length and checked the OWN bit, so rather
1879 		 * than using an alternate bit mapping, we shift the
1880 		 * status bits one space to the right so we can evaluate
1881 		 * them using the 8169 status as though it was in the
1882 		 * same format as that of the 8139C+.
1883 		 */
1884 		if (sc->rl_type == RL_8169)
1885 			rxstat >>= 1;
1886 
1887 		/*
1888 		 * if total_len > 2^13-1, both _RXERRSUM and _GIANT will be
1889 		 * set, but if CRC is clear, it will still be a valid frame.
1890 		 */
1891 		if (rxstat & RL_RDESC_STAT_RXERRSUM && !(total_len > 8191 &&
1892 		    (rxstat & RL_RDESC_STAT_ERRS) == RL_RDESC_STAT_GIANT)) {
1893 			ifp->if_ierrors++;
1894 			/*
1895 			 * If this is part of a multi-fragment packet,
1896 			 * discard all the pieces.
1897 			 */
1898 			if (sc->rl_head != NULL) {
1899 				m_freem(sc->rl_head);
1900 				sc->rl_head = sc->rl_tail = NULL;
1901 			}
1902 			re_discard_rxbuf(sc, i);
1903 			continue;
1904 		}
1905 
1906 		/*
1907 		 * If allocating a replacement mbuf fails,
1908 		 * reload the current one.
1909 		 */
1910 
1911 		if (re_newbuf(sc, i) != 0) {
1912 			ifp->if_iqdrops++;
1913 			if (sc->rl_head != NULL) {
1914 				m_freem(sc->rl_head);
1915 				sc->rl_head = sc->rl_tail = NULL;
1916 			}
1917 			re_discard_rxbuf(sc, i);
1918 			continue;
1919 		}
1920 
1921 		if (sc->rl_head != NULL) {
1922 			m->m_len = total_len % RE_RX_DESC_BUFLEN;
1923 			if (m->m_len == 0)
1924 				m->m_len = RE_RX_DESC_BUFLEN;
1925 			/*
1926 			 * Special case: if there's 4 bytes or less
1927 			 * in this buffer, the mbuf can be discarded:
1928 			 * the last 4 bytes is the CRC, which we don't
1929 			 * care about anyway.
1930 			 */
1931 			if (m->m_len <= ETHER_CRC_LEN) {
1932 				sc->rl_tail->m_len -=
1933 				    (ETHER_CRC_LEN - m->m_len);
1934 				m_freem(m);
1935 			} else {
1936 				m->m_len -= ETHER_CRC_LEN;
1937 				m->m_flags &= ~M_PKTHDR;
1938 				sc->rl_tail->m_next = m;
1939 			}
1940 			m = sc->rl_head;
1941 			sc->rl_head = sc->rl_tail = NULL;
1942 			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
1943 		} else
1944 			m->m_pkthdr.len = m->m_len =
1945 			    (total_len - ETHER_CRC_LEN);
1946 
1947 #ifdef RE_FIXUP_RX
1948 		re_fixup_rx(m);
1949 #endif
1950 		ifp->if_ipackets++;
1951 		m->m_pkthdr.rcvif = ifp;
1952 
1953 		/* Do RX checksumming if enabled */
1954 
1955 		if (ifp->if_capenable & IFCAP_RXCSUM) {
1956 			if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
1957 				/* Check IP header checksum */
1958 				if (rxstat & RL_RDESC_STAT_PROTOID)
1959 					m->m_pkthdr.csum_flags |=
1960 					    CSUM_IP_CHECKED;
1961 				if (!(rxstat & RL_RDESC_STAT_IPSUMBAD))
1962 					m->m_pkthdr.csum_flags |=
1963 					    CSUM_IP_VALID;
1964 
1965 				/* Check TCP/UDP checksum */
1966 				if ((RL_TCPPKT(rxstat) &&
1967 				    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
1968 				    (RL_UDPPKT(rxstat) &&
1969 				     !(rxstat & RL_RDESC_STAT_UDPSUMBAD))) {
1970 					m->m_pkthdr.csum_flags |=
1971 						CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
1972 					m->m_pkthdr.csum_data = 0xffff;
1973 				}
1974 			} else {
1975 				/*
1976 				 * RTL8168C/RTL816CP/RTL8111C/RTL8111CP
1977 				 */
1978 				if ((rxstat & RL_RDESC_STAT_PROTOID) &&
1979 				    (rxvlan & RL_RDESC_IPV4))
1980 					m->m_pkthdr.csum_flags |=
1981 					    CSUM_IP_CHECKED;
1982 				if (!(rxstat & RL_RDESC_STAT_IPSUMBAD) &&
1983 				    (rxvlan & RL_RDESC_IPV4))
1984 					m->m_pkthdr.csum_flags |=
1985 					    CSUM_IP_VALID;
1986 				if (((rxstat & RL_RDESC_STAT_TCP) &&
1987 				    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
1988 				    ((rxstat & RL_RDESC_STAT_UDP) &&
1989 				    !(rxstat & RL_RDESC_STAT_UDPSUMBAD))) {
1990 					m->m_pkthdr.csum_flags |=
1991 						CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
1992 					m->m_pkthdr.csum_data = 0xffff;
1993 				}
1994 			}
1995 		}
1996 		maxpkt--;
1997 		if (rxvlan & RL_RDESC_VLANCTL_TAG) {
1998 			m->m_pkthdr.ether_vtag =
1999 			    bswap16((rxvlan & RL_RDESC_VLANCTL_DATA));
2000 			m->m_flags |= M_VLANTAG;
2001 		}
2002 		RL_UNLOCK(sc);
2003 		(*ifp->if_input)(ifp, m);
2004 		RL_LOCK(sc);
2005 		rx_npkts++;
2006 	}
2007 
2008 	/* Flush the RX DMA ring */
2009 
2010 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
2011 	    sc->rl_ldata.rl_rx_list_map,
2012 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2013 
2014 	sc->rl_ldata.rl_rx_prodidx = i;
2015 
2016 	if (rx_npktsp != NULL)
2017 		*rx_npktsp = rx_npkts;
2018 	if (maxpkt)
2019 		return(EAGAIN);
2020 
2021 	return(0);
2022 }
2023 
2024 static void
2025 re_txeof(struct rl_softc *sc)
2026 {
2027 	struct ifnet		*ifp;
2028 	struct rl_txdesc	*txd;
2029 	u_int32_t		txstat;
2030 	int			cons;
2031 
2032 	cons = sc->rl_ldata.rl_tx_considx;
2033 	if (cons == sc->rl_ldata.rl_tx_prodidx)
2034 		return;
2035 
2036 	ifp = sc->rl_ifp;
2037 	/* Invalidate the TX descriptor list */
2038 	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2039 	    sc->rl_ldata.rl_tx_list_map,
2040 	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2041 
2042 	for (; cons != sc->rl_ldata.rl_tx_prodidx;
2043 	    cons = RL_TX_DESC_NXT(sc, cons)) {
2044 		txstat = le32toh(sc->rl_ldata.rl_tx_list[cons].rl_cmdstat);
2045 		if (txstat & RL_TDESC_STAT_OWN)
2046 			break;
2047 		/*
2048 		 * We only stash mbufs in the last descriptor
2049 		 * in a fragment chain, which also happens to
2050 		 * be the only place where the TX status bits
2051 		 * are valid.
2052 		 */
2053 		if (txstat & RL_TDESC_CMD_EOF) {
2054 			txd = &sc->rl_ldata.rl_tx_desc[cons];
2055 			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
2056 			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
2057 			bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag,
2058 			    txd->tx_dmamap);
2059 			KASSERT(txd->tx_m != NULL,
2060 			    ("%s: freeing NULL mbufs!", __func__));
2061 			m_freem(txd->tx_m);
2062 			txd->tx_m = NULL;
2063 			if (txstat & (RL_TDESC_STAT_EXCESSCOL|
2064 			    RL_TDESC_STAT_COLCNT))
2065 				ifp->if_collisions++;
2066 			if (txstat & RL_TDESC_STAT_TXERRSUM)
2067 				ifp->if_oerrors++;
2068 			else
2069 				ifp->if_opackets++;
2070 		}
2071 		sc->rl_ldata.rl_tx_free++;
2072 		ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
2073 	}
2074 	sc->rl_ldata.rl_tx_considx = cons;
2075 
2076 	/* No changes made to the TX ring, so no flush needed */
2077 
2078 	if (sc->rl_ldata.rl_tx_free != sc->rl_ldata.rl_tx_desc_cnt) {
2079 #ifdef RE_TX_MODERATION
2080 		/*
2081 		 * If not all descriptors have been reaped yet, reload
2082 		 * the timer so that we will eventually get another
2083 		 * interrupt that will cause us to re-enter this routine.
2084 		 * This is done in case the transmitter has gone idle.
2085 		 */
2086 		CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2087 #endif
2088 	} else
2089 		sc->rl_watchdog_timer = 0;
2090 }
2091 
2092 static void
2093 re_tick(void *xsc)
2094 {
2095 	struct rl_softc		*sc;
2096 	struct mii_data		*mii;
2097 
2098 	sc = xsc;
2099 
2100 	RL_LOCK_ASSERT(sc);
2101 
2102 	mii = device_get_softc(sc->rl_miibus);
2103 	mii_tick(mii);
2104 	if ((sc->rl_flags & RL_FLAG_LINK) == 0)
2105 		re_miibus_statchg(sc->rl_dev);
2106 	/*
2107 	 * Reclaim transmitted frames here. Technically it is not
2108 	 * necessary to do here but it ensures periodic reclamation
2109 	 * regardless of Tx completion interrupt which seems to be
2110 	 * lost on PCIe based controllers under certain situations.
2111 	 */
2112 	re_txeof(sc);
2113 	re_watchdog(sc);
2114 	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
2115 }
2116 
2117 #ifdef DEVICE_POLLING
2118 static int
2119 re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
2120 {
2121 	struct rl_softc *sc = ifp->if_softc;
2122 	int rx_npkts = 0;
2123 
2124 	RL_LOCK(sc);
2125 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
2126 		rx_npkts = re_poll_locked(ifp, cmd, count);
2127 	RL_UNLOCK(sc);
2128 	return (rx_npkts);
2129 }
2130 
2131 static int
2132 re_poll_locked(struct ifnet *ifp, enum poll_cmd cmd, int count)
2133 {
2134 	struct rl_softc *sc = ifp->if_softc;
2135 	int rx_npkts;
2136 
2137 	RL_LOCK_ASSERT(sc);
2138 
2139 	sc->rxcycles = count;
2140 	re_rxeof(sc, &rx_npkts);
2141 	re_txeof(sc);
2142 
2143 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2144 		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_txtask);
2145 
2146 	if (cmd == POLL_AND_CHECK_STATUS) { /* also check status register */
2147 		u_int16_t       status;
2148 
2149 		status = CSR_READ_2(sc, RL_ISR);
2150 		if (status == 0xffff)
2151 			return (rx_npkts);
2152 		if (status)
2153 			CSR_WRITE_2(sc, RL_ISR, status);
2154 		if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
2155 		    (sc->rl_flags & RL_FLAG_PCIE))
2156 			CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2157 
2158 		/*
2159 		 * XXX check behaviour on receiver stalls.
2160 		 */
2161 
2162 		if (status & RL_ISR_SYSTEM_ERR) {
2163 			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2164 			re_init_locked(sc);
2165 		}
2166 	}
2167 	return (rx_npkts);
2168 }
2169 #endif /* DEVICE_POLLING */
2170 
2171 static int
2172 re_intr(void *arg)
2173 {
2174 	struct rl_softc		*sc;
2175 	uint16_t		status;
2176 
2177 	sc = arg;
2178 
2179 	status = CSR_READ_2(sc, RL_ISR);
2180 	if (status == 0xFFFF || (status & RL_INTRS_CPLUS) == 0)
2181                 return (FILTER_STRAY);
2182 	CSR_WRITE_2(sc, RL_IMR, 0);
2183 
2184 	taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_inttask);
2185 
2186 	return (FILTER_HANDLED);
2187 }
2188 
2189 static void
2190 re_int_task(void *arg, int npending)
2191 {
2192 	struct rl_softc		*sc;
2193 	struct ifnet		*ifp;
2194 	u_int16_t		status;
2195 	int			rval = 0;
2196 
2197 	sc = arg;
2198 	ifp = sc->rl_ifp;
2199 
2200 	RL_LOCK(sc);
2201 
2202 	status = CSR_READ_2(sc, RL_ISR);
2203         CSR_WRITE_2(sc, RL_ISR, status);
2204 
2205 	if (sc->suspended ||
2206 	    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2207 		RL_UNLOCK(sc);
2208 		return;
2209 	}
2210 
2211 #ifdef DEVICE_POLLING
2212 	if  (ifp->if_capenable & IFCAP_POLLING) {
2213 		RL_UNLOCK(sc);
2214 		return;
2215 	}
2216 #endif
2217 
2218 	if (status & (RL_ISR_RX_OK|RL_ISR_RX_ERR|RL_ISR_FIFO_OFLOW))
2219 		rval = re_rxeof(sc, NULL);
2220 
2221 	/*
2222 	 * Some chips will ignore a second TX request issued
2223 	 * while an existing transmission is in progress. If
2224 	 * the transmitter goes idle but there are still
2225 	 * packets waiting to be sent, we need to restart the
2226 	 * channel here to flush them out. This only seems to
2227 	 * be required with the PCIe devices.
2228 	 */
2229 	if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
2230 	    (sc->rl_flags & RL_FLAG_PCIE))
2231 		CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2232 	if (status & (
2233 #ifdef RE_TX_MODERATION
2234 	    RL_ISR_TIMEOUT_EXPIRED|
2235 #else
2236 	    RL_ISR_TX_OK|
2237 #endif
2238 	    RL_ISR_TX_ERR|RL_ISR_TX_DESC_UNAVAIL))
2239 		re_txeof(sc);
2240 
2241 	if (status & RL_ISR_SYSTEM_ERR) {
2242 		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2243 		re_init_locked(sc);
2244 	}
2245 
2246 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2247 		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_txtask);
2248 
2249 	RL_UNLOCK(sc);
2250 
2251         if ((CSR_READ_2(sc, RL_ISR) & RL_INTRS_CPLUS) || rval) {
2252 		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_inttask);
2253 		return;
2254 	}
2255 
2256 	CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2257 }
2258 
2259 static int
2260 re_encap(struct rl_softc *sc, struct mbuf **m_head)
2261 {
2262 	struct rl_txdesc	*txd, *txd_last;
2263 	bus_dma_segment_t	segs[RL_NTXSEGS];
2264 	bus_dmamap_t		map;
2265 	struct mbuf		*m_new;
2266 	struct rl_desc		*desc;
2267 	int			nsegs, prod;
2268 	int			i, error, ei, si;
2269 	int			padlen;
2270 	uint32_t		cmdstat, csum_flags, vlanctl;
2271 
2272 	RL_LOCK_ASSERT(sc);
2273 	M_ASSERTPKTHDR((*m_head));
2274 
2275 	/*
2276 	 * With some of the RealTek chips, using the checksum offload
2277 	 * support in conjunction with the autopadding feature results
2278 	 * in the transmission of corrupt frames. For example, if we
2279 	 * need to send a really small IP fragment that's less than 60
2280 	 * bytes in size, and IP header checksumming is enabled, the
2281 	 * resulting ethernet frame that appears on the wire will
2282 	 * have garbled payload. To work around this, if TX IP checksum
2283 	 * offload is enabled, we always manually pad short frames out
2284 	 * to the minimum ethernet frame size.
2285 	 */
2286 	if ((sc->rl_flags & RL_FLAG_AUTOPAD) == 0 &&
2287 	    (*m_head)->m_pkthdr.len < RL_IP4CSUMTX_PADLEN &&
2288 	    ((*m_head)->m_pkthdr.csum_flags & CSUM_IP) != 0) {
2289 		padlen = RL_MIN_FRAMELEN - (*m_head)->m_pkthdr.len;
2290 		if (M_WRITABLE(*m_head) == 0) {
2291 			/* Get a writable copy. */
2292 			m_new = m_dup(*m_head, M_DONTWAIT);
2293 			m_freem(*m_head);
2294 			if (m_new == NULL) {
2295 				*m_head = NULL;
2296 				return (ENOBUFS);
2297 			}
2298 			*m_head = m_new;
2299 		}
2300 		if ((*m_head)->m_next != NULL ||
2301 		    M_TRAILINGSPACE(*m_head) < padlen) {
2302 			m_new = m_defrag(*m_head, M_DONTWAIT);
2303 			if (m_new == NULL) {
2304 				m_freem(*m_head);
2305 				*m_head = NULL;
2306 				return (ENOBUFS);
2307 			}
2308 		} else
2309 			m_new = *m_head;
2310 
2311 		/*
2312 		 * Manually pad short frames, and zero the pad space
2313 		 * to avoid leaking data.
2314 		 */
2315 		bzero(mtod(m_new, char *) + m_new->m_pkthdr.len, padlen);
2316 		m_new->m_pkthdr.len += padlen;
2317 		m_new->m_len = m_new->m_pkthdr.len;
2318 		*m_head = m_new;
2319 	}
2320 
2321 	prod = sc->rl_ldata.rl_tx_prodidx;
2322 	txd = &sc->rl_ldata.rl_tx_desc[prod];
2323 	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap,
2324 	    *m_head, segs, &nsegs, BUS_DMA_NOWAIT);
2325 	if (error == EFBIG) {
2326 		m_new = m_collapse(*m_head, M_DONTWAIT, RL_NTXSEGS);
2327 		if (m_new == NULL) {
2328 			m_freem(*m_head);
2329 			*m_head = NULL;
2330 			return (ENOBUFS);
2331 		}
2332 		*m_head = m_new;
2333 		error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_tx_mtag,
2334 		    txd->tx_dmamap, *m_head, segs, &nsegs, BUS_DMA_NOWAIT);
2335 		if (error != 0) {
2336 			m_freem(*m_head);
2337 			*m_head = NULL;
2338 			return (error);
2339 		}
2340 	} else if (error != 0)
2341 		return (error);
2342 	if (nsegs == 0) {
2343 		m_freem(*m_head);
2344 		*m_head = NULL;
2345 		return (EIO);
2346 	}
2347 
2348 	/* Check for number of available descriptors. */
2349 	if (sc->rl_ldata.rl_tx_free - nsegs <= 1) {
2350 		bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap);
2351 		return (ENOBUFS);
2352 	}
2353 
2354 	bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap,
2355 	    BUS_DMASYNC_PREWRITE);
2356 
2357 	/*
2358 	 * Set up checksum offload. Note: checksum offload bits must
2359 	 * appear in all descriptors of a multi-descriptor transmit
2360 	 * attempt. This is according to testing done with an 8169
2361 	 * chip. This is a requirement.
2362 	 */
2363 	vlanctl = 0;
2364 	csum_flags = 0;
2365 	if (((*m_head)->m_pkthdr.csum_flags & CSUM_TSO) != 0)
2366 		csum_flags = RL_TDESC_CMD_LGSEND |
2367 		    ((uint32_t)(*m_head)->m_pkthdr.tso_segsz <<
2368 		    RL_TDESC_CMD_MSSVAL_SHIFT);
2369 	else {
2370 		/*
2371 		 * Unconditionally enable IP checksum if TCP or UDP
2372 		 * checksum is required. Otherwise, TCP/UDP checksum
2373 		 * does't make effects.
2374 		 */
2375 		if (((*m_head)->m_pkthdr.csum_flags & RE_CSUM_FEATURES) != 0) {
2376 			if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
2377 				csum_flags |= RL_TDESC_CMD_IPCSUM;
2378 				if (((*m_head)->m_pkthdr.csum_flags &
2379 				    CSUM_TCP) != 0)
2380 					csum_flags |= RL_TDESC_CMD_TCPCSUM;
2381 				if (((*m_head)->m_pkthdr.csum_flags &
2382 				    CSUM_UDP) != 0)
2383 					csum_flags |= RL_TDESC_CMD_UDPCSUM;
2384 			} else {
2385 				vlanctl |= RL_TDESC_CMD_IPCSUMV2;
2386 				if (((*m_head)->m_pkthdr.csum_flags &
2387 				    CSUM_TCP) != 0)
2388 					vlanctl |= RL_TDESC_CMD_TCPCSUMV2;
2389 				if (((*m_head)->m_pkthdr.csum_flags &
2390 				    CSUM_UDP) != 0)
2391 					vlanctl |= RL_TDESC_CMD_UDPCSUMV2;
2392 			}
2393 		}
2394 	}
2395 
2396 	/*
2397 	 * Set up hardware VLAN tagging. Note: vlan tag info must
2398 	 * appear in all descriptors of a multi-descriptor
2399 	 * transmission attempt.
2400 	 */
2401 	if ((*m_head)->m_flags & M_VLANTAG)
2402 		vlanctl |= bswap16((*m_head)->m_pkthdr.ether_vtag) |
2403 		    RL_TDESC_VLANCTL_TAG;
2404 
2405 	si = prod;
2406 	for (i = 0; i < nsegs; i++, prod = RL_TX_DESC_NXT(sc, prod)) {
2407 		desc = &sc->rl_ldata.rl_tx_list[prod];
2408 		desc->rl_vlanctl = htole32(vlanctl);
2409 		desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[i].ds_addr));
2410 		desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[i].ds_addr));
2411 		cmdstat = segs[i].ds_len;
2412 		if (i != 0)
2413 			cmdstat |= RL_TDESC_CMD_OWN;
2414 		if (prod == sc->rl_ldata.rl_tx_desc_cnt - 1)
2415 			cmdstat |= RL_TDESC_CMD_EOR;
2416 		desc->rl_cmdstat = htole32(cmdstat | csum_flags);
2417 		sc->rl_ldata.rl_tx_free--;
2418 	}
2419 	/* Update producer index. */
2420 	sc->rl_ldata.rl_tx_prodidx = prod;
2421 
2422 	/* Set EOF on the last descriptor. */
2423 	ei = RL_TX_DESC_PRV(sc, prod);
2424 	desc = &sc->rl_ldata.rl_tx_list[ei];
2425 	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_EOF);
2426 
2427 	desc = &sc->rl_ldata.rl_tx_list[si];
2428 	/* Set SOF and transfer ownership of packet to the chip. */
2429 	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_OWN | RL_TDESC_CMD_SOF);
2430 
2431 	/*
2432 	 * Insure that the map for this transmission
2433 	 * is placed at the array index of the last descriptor
2434 	 * in this chain.  (Swap last and first dmamaps.)
2435 	 */
2436 	txd_last = &sc->rl_ldata.rl_tx_desc[ei];
2437 	map = txd->tx_dmamap;
2438 	txd->tx_dmamap = txd_last->tx_dmamap;
2439 	txd_last->tx_dmamap = map;
2440 	txd_last->tx_m = *m_head;
2441 
2442 	return (0);
2443 }
2444 
2445 static void
2446 re_tx_task(void *arg, int npending)
2447 {
2448 	struct ifnet		*ifp;
2449 
2450 	ifp = arg;
2451 	re_start(ifp);
2452 }
2453 
2454 /*
2455  * Main transmit routine for C+ and gigE NICs.
2456  */
2457 static void
2458 re_start(struct ifnet *ifp)
2459 {
2460 	struct rl_softc		*sc;
2461 	struct mbuf		*m_head;
2462 	int			queued;
2463 
2464 	sc = ifp->if_softc;
2465 
2466 	RL_LOCK(sc);
2467 
2468 	if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
2469 	    IFF_DRV_RUNNING || (sc->rl_flags & RL_FLAG_LINK) == 0) {
2470 		RL_UNLOCK(sc);
2471 		return;
2472 	}
2473 
2474 	for (queued = 0; !IFQ_DRV_IS_EMPTY(&ifp->if_snd) &&
2475 	    sc->rl_ldata.rl_tx_free > 1;) {
2476 		IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
2477 		if (m_head == NULL)
2478 			break;
2479 
2480 		if (re_encap(sc, &m_head) != 0) {
2481 			if (m_head == NULL)
2482 				break;
2483 			IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
2484 			ifp->if_drv_flags |= IFF_DRV_OACTIVE;
2485 			break;
2486 		}
2487 
2488 		/*
2489 		 * If there's a BPF listener, bounce a copy of this frame
2490 		 * to him.
2491 		 */
2492 		ETHER_BPF_MTAP(ifp, m_head);
2493 
2494 		queued++;
2495 	}
2496 
2497 	if (queued == 0) {
2498 #ifdef RE_TX_MODERATION
2499 		if (sc->rl_ldata.rl_tx_free != sc->rl_ldata.rl_tx_desc_cnt)
2500 			CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2501 #endif
2502 		RL_UNLOCK(sc);
2503 		return;
2504 	}
2505 
2506 	/* Flush the TX descriptors */
2507 
2508 	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2509 	    sc->rl_ldata.rl_tx_list_map,
2510 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2511 
2512 	CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2513 
2514 #ifdef RE_TX_MODERATION
2515 	/*
2516 	 * Use the countdown timer for interrupt moderation.
2517 	 * 'TX done' interrupts are disabled. Instead, we reset the
2518 	 * countdown timer, which will begin counting until it hits
2519 	 * the value in the TIMERINT register, and then trigger an
2520 	 * interrupt. Each time we write to the TIMERCNT register,
2521 	 * the timer count is reset to 0.
2522 	 */
2523 	CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2524 #endif
2525 
2526 	/*
2527 	 * Set a timeout in case the chip goes out to lunch.
2528 	 */
2529 	sc->rl_watchdog_timer = 5;
2530 
2531 	RL_UNLOCK(sc);
2532 }
2533 
2534 static void
2535 re_init(void *xsc)
2536 {
2537 	struct rl_softc		*sc = xsc;
2538 
2539 	RL_LOCK(sc);
2540 	re_init_locked(sc);
2541 	RL_UNLOCK(sc);
2542 }
2543 
2544 static void
2545 re_init_locked(struct rl_softc *sc)
2546 {
2547 	struct ifnet		*ifp = sc->rl_ifp;
2548 	struct mii_data		*mii;
2549 	uint32_t		reg;
2550 	uint16_t		cfg;
2551 	union {
2552 		uint32_t align_dummy;
2553 		u_char eaddr[ETHER_ADDR_LEN];
2554         } eaddr;
2555 
2556 	RL_LOCK_ASSERT(sc);
2557 
2558 	mii = device_get_softc(sc->rl_miibus);
2559 
2560 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
2561 		return;
2562 
2563 	/*
2564 	 * Cancel pending I/O and free all RX/TX buffers.
2565 	 */
2566 	re_stop(sc);
2567 
2568 	/* Put controller into known state. */
2569 	re_reset(sc);
2570 
2571 	/*
2572 	 * Enable C+ RX and TX mode, as well as VLAN stripping and
2573 	 * RX checksum offload. We must configure the C+ register
2574 	 * before all others.
2575 	 */
2576 	cfg = RL_CPLUSCMD_PCI_MRW;
2577 	if ((ifp->if_capenable & IFCAP_RXCSUM) != 0)
2578 		cfg |= RL_CPLUSCMD_RXCSUM_ENB;
2579 	if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0)
2580 		cfg |= RL_CPLUSCMD_VLANSTRIP;
2581 	if ((sc->rl_flags & RL_FLAG_MACSTAT) != 0) {
2582 		cfg |= RL_CPLUSCMD_MACSTAT_DIS;
2583 		/* XXX magic. */
2584 		cfg |= 0x0001;
2585 	} else
2586 		cfg |= RL_CPLUSCMD_RXENB | RL_CPLUSCMD_TXENB;
2587 	CSR_WRITE_2(sc, RL_CPLUS_CMD, cfg);
2588 	if (sc->rl_hwrev == RL_HWREV_8169_8110SC ||
2589 	    sc->rl_hwrev == RL_HWREV_8169_8110SCE) {
2590 		reg = 0x000fff00;
2591 		if ((CSR_READ_1(sc, RL_CFG2) & RL_CFG2_PCI66MHZ) != 0)
2592 			reg |= 0x000000ff;
2593 		if (sc->rl_hwrev == RL_HWREV_8169_8110SCE)
2594 			reg |= 0x00f00000;
2595 		CSR_WRITE_4(sc, 0x7c, reg);
2596 		/* Disable interrupt mitigation. */
2597 		CSR_WRITE_2(sc, 0xe2, 0);
2598 	}
2599 	/*
2600 	 * Disable TSO if interface MTU size is greater than MSS
2601 	 * allowed in controller.
2602 	 */
2603 	if (ifp->if_mtu > RL_TSO_MTU && (ifp->if_capenable & IFCAP_TSO4) != 0) {
2604 		ifp->if_capenable &= ~IFCAP_TSO4;
2605 		ifp->if_hwassist &= ~CSUM_TSO;
2606 	}
2607 
2608 	/*
2609 	 * Init our MAC address.  Even though the chipset
2610 	 * documentation doesn't mention it, we need to enter "Config
2611 	 * register write enable" mode to modify the ID registers.
2612 	 */
2613 	/* Copy MAC address on stack to align. */
2614 	bcopy(IF_LLADDR(ifp), eaddr.eaddr, ETHER_ADDR_LEN);
2615 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG);
2616 	CSR_WRITE_4(sc, RL_IDR0,
2617 	    htole32(*(u_int32_t *)(&eaddr.eaddr[0])));
2618 	CSR_WRITE_4(sc, RL_IDR4,
2619 	    htole32(*(u_int32_t *)(&eaddr.eaddr[4])));
2620 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
2621 
2622 	/*
2623 	 * For C+ mode, initialize the RX descriptors and mbufs.
2624 	 */
2625 	re_rx_list_init(sc);
2626 	re_tx_list_init(sc);
2627 
2628 	/*
2629 	 * Load the addresses of the RX and TX lists into the chip.
2630 	 */
2631 
2632 	CSR_WRITE_4(sc, RL_RXLIST_ADDR_HI,
2633 	    RL_ADDR_HI(sc->rl_ldata.rl_rx_list_addr));
2634 	CSR_WRITE_4(sc, RL_RXLIST_ADDR_LO,
2635 	    RL_ADDR_LO(sc->rl_ldata.rl_rx_list_addr));
2636 
2637 	CSR_WRITE_4(sc, RL_TXLIST_ADDR_HI,
2638 	    RL_ADDR_HI(sc->rl_ldata.rl_tx_list_addr));
2639 	CSR_WRITE_4(sc, RL_TXLIST_ADDR_LO,
2640 	    RL_ADDR_LO(sc->rl_ldata.rl_tx_list_addr));
2641 
2642 	/*
2643 	 * Enable transmit and receive.
2644 	 */
2645 	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
2646 
2647 	/*
2648 	 * Set the initial TX configuration.
2649 	 */
2650 	if (sc->rl_testmode) {
2651 		if (sc->rl_type == RL_8169)
2652 			CSR_WRITE_4(sc, RL_TXCFG,
2653 			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON);
2654 		else
2655 			CSR_WRITE_4(sc, RL_TXCFG,
2656 			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON_CPLUS);
2657 	} else
2658 		CSR_WRITE_4(sc, RL_TXCFG, RL_TXCFG_CONFIG);
2659 
2660 	CSR_WRITE_1(sc, RL_EARLY_TX_THRESH, 16);
2661 
2662 	/*
2663 	 * Set the initial RX configuration.
2664 	 */
2665 	re_set_rxmode(sc);
2666 
2667 #ifdef DEVICE_POLLING
2668 	/*
2669 	 * Disable interrupts if we are polling.
2670 	 */
2671 	if (ifp->if_capenable & IFCAP_POLLING)
2672 		CSR_WRITE_2(sc, RL_IMR, 0);
2673 	else	/* otherwise ... */
2674 #endif
2675 
2676 	/*
2677 	 * Enable interrupts.
2678 	 */
2679 	if (sc->rl_testmode)
2680 		CSR_WRITE_2(sc, RL_IMR, 0);
2681 	else
2682 		CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2683 	CSR_WRITE_2(sc, RL_ISR, RL_INTRS_CPLUS);
2684 
2685 	/* Set initial TX threshold */
2686 	sc->rl_txthresh = RL_TX_THRESH_INIT;
2687 
2688 	/* Start RX/TX process. */
2689 	CSR_WRITE_4(sc, RL_MISSEDPKT, 0);
2690 #ifdef notdef
2691 	/* Enable receiver and transmitter. */
2692 	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
2693 #endif
2694 
2695 #ifdef RE_TX_MODERATION
2696 	/*
2697 	 * Initialize the timer interrupt register so that
2698 	 * a timer interrupt will be generated once the timer
2699 	 * reaches a certain number of ticks. The timer is
2700 	 * reloaded on each transmit. This gives us TX interrupt
2701 	 * moderation, which dramatically improves TX frame rate.
2702 	 */
2703 	if (sc->rl_type == RL_8169)
2704 		CSR_WRITE_4(sc, RL_TIMERINT_8169, 0x800);
2705 	else
2706 		CSR_WRITE_4(sc, RL_TIMERINT, 0x400);
2707 #endif
2708 
2709 	/*
2710 	 * For 8169 gigE NICs, set the max allowed RX packet
2711 	 * size so we can receive jumbo frames.
2712 	 */
2713 	if (sc->rl_type == RL_8169)
2714 		CSR_WRITE_2(sc, RL_MAXRXPKTLEN, 16383);
2715 
2716 	if (sc->rl_testmode)
2717 		return;
2718 
2719 	mii_mediachg(mii);
2720 
2721 	CSR_WRITE_1(sc, RL_CFG1, CSR_READ_1(sc, RL_CFG1) | RL_CFG1_DRVLOAD);
2722 
2723 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
2724 	ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
2725 
2726 	sc->rl_flags &= ~RL_FLAG_LINK;
2727 	sc->rl_watchdog_timer = 0;
2728 	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
2729 }
2730 
2731 /*
2732  * Set media options.
2733  */
2734 static int
2735 re_ifmedia_upd(struct ifnet *ifp)
2736 {
2737 	struct rl_softc		*sc;
2738 	struct mii_data		*mii;
2739 	int			error;
2740 
2741 	sc = ifp->if_softc;
2742 	mii = device_get_softc(sc->rl_miibus);
2743 	RL_LOCK(sc);
2744 	error = mii_mediachg(mii);
2745 	RL_UNLOCK(sc);
2746 
2747 	return (error);
2748 }
2749 
2750 /*
2751  * Report current media status.
2752  */
2753 static void
2754 re_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
2755 {
2756 	struct rl_softc		*sc;
2757 	struct mii_data		*mii;
2758 
2759 	sc = ifp->if_softc;
2760 	mii = device_get_softc(sc->rl_miibus);
2761 
2762 	RL_LOCK(sc);
2763 	mii_pollstat(mii);
2764 	RL_UNLOCK(sc);
2765 	ifmr->ifm_active = mii->mii_media_active;
2766 	ifmr->ifm_status = mii->mii_media_status;
2767 }
2768 
2769 static int
2770 re_ioctl(struct ifnet *ifp, u_long command, caddr_t data)
2771 {
2772 	struct rl_softc		*sc = ifp->if_softc;
2773 	struct ifreq		*ifr = (struct ifreq *) data;
2774 	struct mii_data		*mii;
2775 	int			error = 0;
2776 
2777 	switch (command) {
2778 	case SIOCSIFMTU:
2779 		if (ifr->ifr_mtu < ETHERMIN || ifr->ifr_mtu > RL_JUMBO_MTU) {
2780 			error = EINVAL;
2781 			break;
2782 		}
2783 		if ((sc->rl_flags & RL_FLAG_NOJUMBO) != 0 &&
2784 		    ifr->ifr_mtu > RL_MAX_FRAMELEN) {
2785 			error = EINVAL;
2786 			break;
2787 		}
2788 		RL_LOCK(sc);
2789 		if (ifp->if_mtu != ifr->ifr_mtu)
2790 			ifp->if_mtu = ifr->ifr_mtu;
2791 		if (ifp->if_mtu > RL_TSO_MTU &&
2792 		    (ifp->if_capenable & IFCAP_TSO4) != 0) {
2793 			ifp->if_capenable &= ~IFCAP_TSO4;
2794 			ifp->if_hwassist &= ~CSUM_TSO;
2795 			VLAN_CAPABILITIES(ifp);
2796 		}
2797 		RL_UNLOCK(sc);
2798 		break;
2799 	case SIOCSIFFLAGS:
2800 		RL_LOCK(sc);
2801 		if ((ifp->if_flags & IFF_UP) != 0) {
2802 			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
2803 				if (((ifp->if_flags ^ sc->rl_if_flags)
2804 				    & (IFF_PROMISC | IFF_ALLMULTI)) != 0)
2805 					re_set_rxmode(sc);
2806 			} else
2807 				re_init_locked(sc);
2808 		} else {
2809 			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
2810 				re_stop(sc);
2811 		}
2812 		sc->rl_if_flags = ifp->if_flags;
2813 		RL_UNLOCK(sc);
2814 		break;
2815 	case SIOCADDMULTI:
2816 	case SIOCDELMULTI:
2817 		RL_LOCK(sc);
2818 		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
2819 			re_set_rxmode(sc);
2820 		RL_UNLOCK(sc);
2821 		break;
2822 	case SIOCGIFMEDIA:
2823 	case SIOCSIFMEDIA:
2824 		mii = device_get_softc(sc->rl_miibus);
2825 		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, command);
2826 		break;
2827 	case SIOCSIFCAP:
2828 	    {
2829 		int mask, reinit;
2830 
2831 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
2832 		reinit = 0;
2833 #ifdef DEVICE_POLLING
2834 		if (mask & IFCAP_POLLING) {
2835 			if (ifr->ifr_reqcap & IFCAP_POLLING) {
2836 				error = ether_poll_register(re_poll, ifp);
2837 				if (error)
2838 					return(error);
2839 				RL_LOCK(sc);
2840 				/* Disable interrupts */
2841 				CSR_WRITE_2(sc, RL_IMR, 0x0000);
2842 				ifp->if_capenable |= IFCAP_POLLING;
2843 				RL_UNLOCK(sc);
2844 			} else {
2845 				error = ether_poll_deregister(ifp);
2846 				/* Enable interrupts. */
2847 				RL_LOCK(sc);
2848 				CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2849 				ifp->if_capenable &= ~IFCAP_POLLING;
2850 				RL_UNLOCK(sc);
2851 			}
2852 		}
2853 #endif /* DEVICE_POLLING */
2854 		if (mask & IFCAP_HWCSUM) {
2855 			ifp->if_capenable ^= IFCAP_HWCSUM;
2856 			if (ifp->if_capenable & IFCAP_TXCSUM)
2857 				ifp->if_hwassist |= RE_CSUM_FEATURES;
2858 			else
2859 				ifp->if_hwassist &= ~RE_CSUM_FEATURES;
2860 			reinit = 1;
2861 		}
2862 		if ((mask & IFCAP_TSO4) != 0 &&
2863 		    (ifp->if_capabilities & IFCAP_TSO) != 0) {
2864 			ifp->if_capenable ^= IFCAP_TSO4;
2865 			if ((IFCAP_TSO4 & ifp->if_capenable) != 0)
2866 				ifp->if_hwassist |= CSUM_TSO;
2867 			else
2868 				ifp->if_hwassist &= ~CSUM_TSO;
2869 			if (ifp->if_mtu > RL_TSO_MTU &&
2870 			    (ifp->if_capenable & IFCAP_TSO4) != 0) {
2871 				ifp->if_capenable &= ~IFCAP_TSO4;
2872 				ifp->if_hwassist &= ~CSUM_TSO;
2873 			}
2874 		}
2875 		if ((mask & IFCAP_VLAN_HWTSO) != 0 &&
2876 		    (ifp->if_capabilities & IFCAP_VLAN_HWTSO) != 0)
2877 			ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
2878 		if ((mask & IFCAP_VLAN_HWTAGGING) != 0 &&
2879 		    (ifp->if_capabilities & IFCAP_VLAN_HWTAGGING) != 0) {
2880 			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
2881 			/* TSO over VLAN requires VLAN hardware tagging. */
2882 			if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) == 0)
2883 				ifp->if_capenable &= ~IFCAP_VLAN_HWTSO;
2884 			reinit = 1;
2885 		}
2886 		if ((mask & IFCAP_WOL) != 0 &&
2887 		    (ifp->if_capabilities & IFCAP_WOL) != 0) {
2888 			if ((mask & IFCAP_WOL_UCAST) != 0)
2889 				ifp->if_capenable ^= IFCAP_WOL_UCAST;
2890 			if ((mask & IFCAP_WOL_MCAST) != 0)
2891 				ifp->if_capenable ^= IFCAP_WOL_MCAST;
2892 			if ((mask & IFCAP_WOL_MAGIC) != 0)
2893 				ifp->if_capenable ^= IFCAP_WOL_MAGIC;
2894 		}
2895 		if (reinit && ifp->if_drv_flags & IFF_DRV_RUNNING) {
2896 			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2897 			re_init(sc);
2898 		}
2899 		VLAN_CAPABILITIES(ifp);
2900 	    }
2901 		break;
2902 	default:
2903 		error = ether_ioctl(ifp, command, data);
2904 		break;
2905 	}
2906 
2907 	return (error);
2908 }
2909 
2910 static void
2911 re_watchdog(struct rl_softc *sc)
2912 {
2913 	struct ifnet		*ifp;
2914 
2915 	RL_LOCK_ASSERT(sc);
2916 
2917 	if (sc->rl_watchdog_timer == 0 || --sc->rl_watchdog_timer != 0)
2918 		return;
2919 
2920 	ifp = sc->rl_ifp;
2921 	re_txeof(sc);
2922 	if (sc->rl_ldata.rl_tx_free == sc->rl_ldata.rl_tx_desc_cnt) {
2923 		if_printf(ifp, "watchdog timeout (missed Tx interrupts) "
2924 		    "-- recovering\n");
2925 		if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2926 			taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_txtask);
2927 		return;
2928 	}
2929 
2930 	if_printf(ifp, "watchdog timeout\n");
2931 	ifp->if_oerrors++;
2932 
2933 	re_rxeof(sc, NULL);
2934 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2935 	re_init_locked(sc);
2936 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2937 		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_txtask);
2938 }
2939 
2940 /*
2941  * Stop the adapter and free any mbufs allocated to the
2942  * RX and TX lists.
2943  */
2944 static void
2945 re_stop(struct rl_softc *sc)
2946 {
2947 	int			i;
2948 	struct ifnet		*ifp;
2949 	struct rl_txdesc	*txd;
2950 	struct rl_rxdesc	*rxd;
2951 
2952 	RL_LOCK_ASSERT(sc);
2953 
2954 	ifp = sc->rl_ifp;
2955 
2956 	sc->rl_watchdog_timer = 0;
2957 	callout_stop(&sc->rl_stat_callout);
2958 	ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
2959 
2960 	if ((sc->rl_flags & RL_FLAG_CMDSTOP) != 0)
2961 		CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_STOPREQ | RL_CMD_TX_ENB |
2962 		    RL_CMD_RX_ENB);
2963 	else
2964 		CSR_WRITE_1(sc, RL_COMMAND, 0x00);
2965 	DELAY(1000);
2966 	CSR_WRITE_2(sc, RL_IMR, 0x0000);
2967 	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
2968 
2969 	if (sc->rl_head != NULL) {
2970 		m_freem(sc->rl_head);
2971 		sc->rl_head = sc->rl_tail = NULL;
2972 	}
2973 
2974 	/* Free the TX list buffers. */
2975 
2976 	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
2977 		txd = &sc->rl_ldata.rl_tx_desc[i];
2978 		if (txd->tx_m != NULL) {
2979 			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
2980 			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
2981 			bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag,
2982 			    txd->tx_dmamap);
2983 			m_freem(txd->tx_m);
2984 			txd->tx_m = NULL;
2985 		}
2986 	}
2987 
2988 	/* Free the RX list buffers. */
2989 
2990 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
2991 		rxd = &sc->rl_ldata.rl_rx_desc[i];
2992 		if (rxd->rx_m != NULL) {
2993 			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
2994 			    rxd->rx_dmamap, BUS_DMASYNC_POSTREAD);
2995 			bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag,
2996 			    rxd->rx_dmamap);
2997 			m_freem(rxd->rx_m);
2998 			rxd->rx_m = NULL;
2999 		}
3000 	}
3001 }
3002 
3003 /*
3004  * Device suspend routine.  Stop the interface and save some PCI
3005  * settings in case the BIOS doesn't restore them properly on
3006  * resume.
3007  */
3008 static int
3009 re_suspend(device_t dev)
3010 {
3011 	struct rl_softc		*sc;
3012 
3013 	sc = device_get_softc(dev);
3014 
3015 	RL_LOCK(sc);
3016 	re_stop(sc);
3017 	re_setwol(sc);
3018 	sc->suspended = 1;
3019 	RL_UNLOCK(sc);
3020 
3021 	return (0);
3022 }
3023 
3024 /*
3025  * Device resume routine.  Restore some PCI settings in case the BIOS
3026  * doesn't, re-enable busmastering, and restart the interface if
3027  * appropriate.
3028  */
3029 static int
3030 re_resume(device_t dev)
3031 {
3032 	struct rl_softc		*sc;
3033 	struct ifnet		*ifp;
3034 
3035 	sc = device_get_softc(dev);
3036 
3037 	RL_LOCK(sc);
3038 
3039 	ifp = sc->rl_ifp;
3040 	/* Take controller out of sleep mode. */
3041 	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
3042 		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
3043 			CSR_WRITE_1(sc, RL_GPIO,
3044 			    CSR_READ_1(sc, RL_GPIO) | 0x01);
3045 	}
3046 
3047 	/*
3048 	 * Clear WOL matching such that normal Rx filtering
3049 	 * wouldn't interfere with WOL patterns.
3050 	 */
3051 	re_clrwol(sc);
3052 
3053 	/* reinitialize interface if necessary */
3054 	if (ifp->if_flags & IFF_UP)
3055 		re_init_locked(sc);
3056 
3057 	sc->suspended = 0;
3058 	RL_UNLOCK(sc);
3059 
3060 	return (0);
3061 }
3062 
3063 /*
3064  * Stop all chip I/O so that the kernel's probe routines don't
3065  * get confused by errant DMAs when rebooting.
3066  */
3067 static int
3068 re_shutdown(device_t dev)
3069 {
3070 	struct rl_softc		*sc;
3071 
3072 	sc = device_get_softc(dev);
3073 
3074 	RL_LOCK(sc);
3075 	re_stop(sc);
3076 	/*
3077 	 * Mark interface as down since otherwise we will panic if
3078 	 * interrupt comes in later on, which can happen in some
3079 	 * cases.
3080 	 */
3081 	sc->rl_ifp->if_flags &= ~IFF_UP;
3082 	re_setwol(sc);
3083 	RL_UNLOCK(sc);
3084 
3085 	return (0);
3086 }
3087 
3088 static void
3089 re_setwol(struct rl_softc *sc)
3090 {
3091 	struct ifnet		*ifp;
3092 	int			pmc;
3093 	uint16_t		pmstat;
3094 	uint8_t			v;
3095 
3096 	RL_LOCK_ASSERT(sc);
3097 
3098 	if (pci_find_extcap(sc->rl_dev, PCIY_PMG, &pmc) != 0)
3099 		return;
3100 
3101 	ifp = sc->rl_ifp;
3102 	/* Put controller into sleep mode. */
3103 	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
3104 		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
3105 			CSR_WRITE_1(sc, RL_GPIO,
3106 			    CSR_READ_1(sc, RL_GPIO) & ~0x01);
3107 	}
3108 	if ((ifp->if_capenable & IFCAP_WOL) != 0 &&
3109 	    (sc->rl_flags & RL_FLAG_WOLRXENB) != 0)
3110 		CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RX_ENB);
3111 	/* Enable config register write. */
3112 	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
3113 
3114 	/* Enable PME. */
3115 	v = CSR_READ_1(sc, RL_CFG1);
3116 	v &= ~RL_CFG1_PME;
3117 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3118 		v |= RL_CFG1_PME;
3119 	CSR_WRITE_1(sc, RL_CFG1, v);
3120 
3121 	v = CSR_READ_1(sc, RL_CFG3);
3122 	v &= ~(RL_CFG3_WOL_LINK | RL_CFG3_WOL_MAGIC);
3123 	if ((ifp->if_capenable & IFCAP_WOL_MAGIC) != 0)
3124 		v |= RL_CFG3_WOL_MAGIC;
3125 	CSR_WRITE_1(sc, RL_CFG3, v);
3126 
3127 	/* Config register write done. */
3128 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3129 
3130 	v = CSR_READ_1(sc, RL_CFG5);
3131 	v &= ~(RL_CFG5_WOL_BCAST | RL_CFG5_WOL_MCAST | RL_CFG5_WOL_UCAST);
3132 	v &= ~RL_CFG5_WOL_LANWAKE;
3133 	if ((ifp->if_capenable & IFCAP_WOL_UCAST) != 0)
3134 		v |= RL_CFG5_WOL_UCAST;
3135 	if ((ifp->if_capenable & IFCAP_WOL_MCAST) != 0)
3136 		v |= RL_CFG5_WOL_MCAST | RL_CFG5_WOL_BCAST;
3137 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3138 		v |= RL_CFG5_WOL_LANWAKE;
3139 	CSR_WRITE_1(sc, RL_CFG5, v);
3140 
3141 	if ((ifp->if_capenable & IFCAP_WOL) != 0 &&
3142 	    (sc->rl_flags & RL_FLAG_PHYWAKE_PM) != 0)
3143 		CSR_WRITE_1(sc, RL_PMCH, CSR_READ_1(sc, RL_PMCH) & ~0x80);
3144 	/*
3145 	 * It seems that hardware resets its link speed to 100Mbps in
3146 	 * power down mode so switching to 100Mbps in driver is not
3147 	 * needed.
3148 	 */
3149 
3150 	/* Request PME if WOL is requested. */
3151 	pmstat = pci_read_config(sc->rl_dev, pmc + PCIR_POWER_STATUS, 2);
3152 	pmstat &= ~(PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE);
3153 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3154 		pmstat |= PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE;
3155 	pci_write_config(sc->rl_dev, pmc + PCIR_POWER_STATUS, pmstat, 2);
3156 }
3157 
3158 static void
3159 re_clrwol(struct rl_softc *sc)
3160 {
3161 	int			pmc;
3162 	uint8_t			v;
3163 
3164 	RL_LOCK_ASSERT(sc);
3165 
3166 	if (pci_find_extcap(sc->rl_dev, PCIY_PMG, &pmc) != 0)
3167 		return;
3168 
3169 	/* Enable config register write. */
3170 	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
3171 
3172 	v = CSR_READ_1(sc, RL_CFG3);
3173 	v &= ~(RL_CFG3_WOL_LINK | RL_CFG3_WOL_MAGIC);
3174 	CSR_WRITE_1(sc, RL_CFG3, v);
3175 
3176 	/* Config register write done. */
3177 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3178 
3179 	v = CSR_READ_1(sc, RL_CFG5);
3180 	v &= ~(RL_CFG5_WOL_BCAST | RL_CFG5_WOL_MCAST | RL_CFG5_WOL_UCAST);
3181 	v &= ~RL_CFG5_WOL_LANWAKE;
3182 	CSR_WRITE_1(sc, RL_CFG5, v);
3183 }
3184