xref: /freebsd/sys/dev/re/if_re.c (revision dcc3a33188bceb5b6e819efdb9c5f72d059084b6)
1 /*-
2  * Copyright (c) 1997, 1998-2003
3  *	Bill Paul <wpaul@windriver.com>.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. All advertising materials mentioning features or use of this software
14  *    must display the following acknowledgement:
15  *	This product includes software developed by Bill Paul.
16  * 4. Neither the name of the author nor the names of any co-contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY Bill Paul AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL Bill Paul OR THE VOICES IN HIS HEAD
24  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
30  * THE POSSIBILITY OF SUCH DAMAGE.
31  */
32 
33 #include <sys/cdefs.h>
34 __FBSDID("$FreeBSD$");
35 
36 /*
37  * RealTek 8139C+/8169/8169S/8110S/8168/8111/8101E PCI NIC driver
38  *
39  * Written by Bill Paul <wpaul@windriver.com>
40  * Senior Networking Software Engineer
41  * Wind River Systems
42  */
43 
44 /*
45  * This driver is designed to support RealTek's next generation of
46  * 10/100 and 10/100/1000 PCI ethernet controllers. There are currently
47  * seven devices in this family: the RTL8139C+, the RTL8169, the RTL8169S,
48  * RTL8110S, the RTL8168, the RTL8111 and the RTL8101E.
49  *
50  * The 8139C+ is a 10/100 ethernet chip. It is backwards compatible
51  * with the older 8139 family, however it also supports a special
52  * C+ mode of operation that provides several new performance enhancing
53  * features. These include:
54  *
55  *	o Descriptor based DMA mechanism. Each descriptor represents
56  *	  a single packet fragment. Data buffers may be aligned on
57  *	  any byte boundary.
58  *
59  *	o 64-bit DMA
60  *
61  *	o TCP/IP checksum offload for both RX and TX
62  *
63  *	o High and normal priority transmit DMA rings
64  *
65  *	o VLAN tag insertion and extraction
66  *
67  *	o TCP large send (segmentation offload)
68  *
69  * Like the 8139, the 8139C+ also has a built-in 10/100 PHY. The C+
70  * programming API is fairly straightforward. The RX filtering, EEPROM
71  * access and PHY access is the same as it is on the older 8139 series
72  * chips.
73  *
74  * The 8169 is a 64-bit 10/100/1000 gigabit ethernet MAC. It has almost the
75  * same programming API and feature set as the 8139C+ with the following
76  * differences and additions:
77  *
78  *	o 1000Mbps mode
79  *
80  *	o Jumbo frames
81  *
82  *	o GMII and TBI ports/registers for interfacing with copper
83  *	  or fiber PHYs
84  *
85  *	o RX and TX DMA rings can have up to 1024 descriptors
86  *	  (the 8139C+ allows a maximum of 64)
87  *
88  *	o Slight differences in register layout from the 8139C+
89  *
90  * The TX start and timer interrupt registers are at different locations
91  * on the 8169 than they are on the 8139C+. Also, the status word in the
92  * RX descriptor has a slightly different bit layout. The 8169 does not
93  * have a built-in PHY. Most reference boards use a Marvell 88E1000 'Alaska'
94  * copper gigE PHY.
95  *
96  * The 8169S/8110S 10/100/1000 devices have built-in copper gigE PHYs
97  * (the 'S' stands for 'single-chip'). These devices have the same
98  * programming API as the older 8169, but also have some vendor-specific
99  * registers for the on-board PHY. The 8110S is a LAN-on-motherboard
100  * part designed to be pin-compatible with the RealTek 8100 10/100 chip.
101  *
102  * This driver takes advantage of the RX and TX checksum offload and
103  * VLAN tag insertion/extraction features. It also implements TX
104  * interrupt moderation using the timer interrupt registers, which
105  * significantly reduces TX interrupt load. There is also support
106  * for jumbo frames, however the 8169/8169S/8110S can not transmit
107  * jumbo frames larger than 7440, so the max MTU possible with this
108  * driver is 7422 bytes.
109  */
110 
111 #ifdef HAVE_KERNEL_OPTION_HEADERS
112 #include "opt_device_polling.h"
113 #endif
114 
115 #include <sys/param.h>
116 #include <sys/endian.h>
117 #include <sys/systm.h>
118 #include <sys/sockio.h>
119 #include <sys/mbuf.h>
120 #include <sys/malloc.h>
121 #include <sys/module.h>
122 #include <sys/kernel.h>
123 #include <sys/socket.h>
124 #include <sys/lock.h>
125 #include <sys/mutex.h>
126 #include <sys/taskqueue.h>
127 
128 #include <net/if.h>
129 #include <net/if_arp.h>
130 #include <net/ethernet.h>
131 #include <net/if_dl.h>
132 #include <net/if_media.h>
133 #include <net/if_types.h>
134 #include <net/if_vlan_var.h>
135 
136 #include <net/bpf.h>
137 
138 #include <machine/bus.h>
139 #include <machine/resource.h>
140 #include <sys/bus.h>
141 #include <sys/rman.h>
142 
143 #include <dev/mii/mii.h>
144 #include <dev/mii/miivar.h>
145 
146 #include <dev/pci/pcireg.h>
147 #include <dev/pci/pcivar.h>
148 
149 #include <pci/if_rlreg.h>
150 
151 MODULE_DEPEND(re, pci, 1, 1, 1);
152 MODULE_DEPEND(re, ether, 1, 1, 1);
153 MODULE_DEPEND(re, miibus, 1, 1, 1);
154 
155 /* "device miibus" required.  See GENERIC if you get errors here. */
156 #include "miibus_if.h"
157 
158 /* Tunables. */
159 static int msi_disable = 0;
160 TUNABLE_INT("hw.re.msi_disable", &msi_disable);
161 static int prefer_iomap = 0;
162 TUNABLE_INT("hw.re.prefer_iomap", &prefer_iomap);
163 
164 #define RE_CSUM_FEATURES    (CSUM_IP | CSUM_TCP | CSUM_UDP)
165 
166 /*
167  * Various supported device vendors/types and their names.
168  */
169 static struct rl_type re_devs[] = {
170 	{ DLINK_VENDORID, DLINK_DEVICEID_528T, 0,
171 	    "D-Link DGE-528(T) Gigabit Ethernet Adapter" },
172 	{ RT_VENDORID, RT_DEVICEID_8139, 0,
173 	    "RealTek 8139C+ 10/100BaseTX" },
174 	{ RT_VENDORID, RT_DEVICEID_8101E, 0,
175 	    "RealTek 8101E/8102E/8102EL PCIe 10/100baseTX" },
176 	{ RT_VENDORID, RT_DEVICEID_8168, 0,
177 	    "RealTek 8168/8168B/8168C/8168CP/8168D/8168DP/"
178 	    "8111B/8111C/8111CP/8111DP PCIe Gigabit Ethernet" },
179 	{ RT_VENDORID, RT_DEVICEID_8169, 0,
180 	    "RealTek 8169/8169S/8169SB(L)/8110S/8110SB(L) Gigabit Ethernet" },
181 	{ RT_VENDORID, RT_DEVICEID_8169SC, 0,
182 	    "RealTek 8169SC/8110SC Single-chip Gigabit Ethernet" },
183 	{ COREGA_VENDORID, COREGA_DEVICEID_CGLAPCIGT, 0,
184 	    "Corega CG-LAPCIGT (RTL8169S) Gigabit Ethernet" },
185 	{ LINKSYS_VENDORID, LINKSYS_DEVICEID_EG1032, 0,
186 	    "Linksys EG1032 (RTL8169S) Gigabit Ethernet" },
187 	{ USR_VENDORID, USR_DEVICEID_997902, 0,
188 	    "US Robotics 997902 (RTL8169S) Gigabit Ethernet" }
189 };
190 
191 static struct rl_hwrev re_hwrevs[] = {
192 	{ RL_HWREV_8139, RL_8139,  "" },
193 	{ RL_HWREV_8139A, RL_8139, "A" },
194 	{ RL_HWREV_8139AG, RL_8139, "A-G" },
195 	{ RL_HWREV_8139B, RL_8139, "B" },
196 	{ RL_HWREV_8130, RL_8139, "8130" },
197 	{ RL_HWREV_8139C, RL_8139, "C" },
198 	{ RL_HWREV_8139D, RL_8139, "8139D/8100B/8100C" },
199 	{ RL_HWREV_8139CPLUS, RL_8139CPLUS, "C+"},
200 	{ RL_HWREV_8168_SPIN1, RL_8169, "8168"},
201 	{ RL_HWREV_8169, RL_8169, "8169"},
202 	{ RL_HWREV_8169S, RL_8169, "8169S"},
203 	{ RL_HWREV_8110S, RL_8169, "8110S"},
204 	{ RL_HWREV_8169_8110SB, RL_8169, "8169SB/8110SB"},
205 	{ RL_HWREV_8169_8110SC, RL_8169, "8169SC/8110SC"},
206 	{ RL_HWREV_8169_8110SBL, RL_8169, "8169SBL/8110SBL"},
207 	{ RL_HWREV_8169_8110SCE, RL_8169, "8169SC/8110SC"},
208 	{ RL_HWREV_8100, RL_8139, "8100"},
209 	{ RL_HWREV_8101, RL_8139, "8101"},
210 	{ RL_HWREV_8100E, RL_8169, "8100E"},
211 	{ RL_HWREV_8101E, RL_8169, "8101E"},
212 	{ RL_HWREV_8102E, RL_8169, "8102E"},
213 	{ RL_HWREV_8102EL, RL_8169, "8102EL"},
214 	{ RL_HWREV_8102EL_SPIN1, RL_8169, "8102EL"},
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 	{ 0, 0, NULL }
223 };
224 
225 static int re_probe		(device_t);
226 static int re_attach		(device_t);
227 static int re_detach		(device_t);
228 
229 static int re_encap		(struct rl_softc *, struct mbuf **);
230 
231 static void re_dma_map_addr	(void *, bus_dma_segment_t *, int, int);
232 static int re_allocmem		(device_t, struct rl_softc *);
233 static __inline void re_discard_rxbuf
234 				(struct rl_softc *, int);
235 static int re_newbuf		(struct rl_softc *, int);
236 static int re_rx_list_init	(struct rl_softc *);
237 static int re_tx_list_init	(struct rl_softc *);
238 #ifdef RE_FIXUP_RX
239 static __inline void re_fixup_rx
240 				(struct mbuf *);
241 #endif
242 static int re_rxeof		(struct rl_softc *, int *);
243 static void re_txeof		(struct rl_softc *);
244 #ifdef DEVICE_POLLING
245 static int re_poll		(struct ifnet *, enum poll_cmd, int);
246 static int re_poll_locked	(struct ifnet *, enum poll_cmd, int);
247 #endif
248 static int re_intr		(void *);
249 static void re_tick		(void *);
250 static void re_tx_task		(void *, int);
251 static void re_int_task		(void *, int);
252 static void re_start		(struct ifnet *);
253 static int re_ioctl		(struct ifnet *, u_long, caddr_t);
254 static void re_init		(void *);
255 static void re_init_locked	(struct rl_softc *);
256 static void re_stop		(struct rl_softc *);
257 static void re_watchdog		(struct rl_softc *);
258 static int re_suspend		(device_t);
259 static int re_resume		(device_t);
260 static int re_shutdown		(device_t);
261 static int re_ifmedia_upd	(struct ifnet *);
262 static void re_ifmedia_sts	(struct ifnet *, struct ifmediareq *);
263 
264 static void re_eeprom_putbyte	(struct rl_softc *, int);
265 static void re_eeprom_getword	(struct rl_softc *, int, u_int16_t *);
266 static void re_read_eeprom	(struct rl_softc *, caddr_t, int, int);
267 static int re_gmii_readreg	(device_t, int, int);
268 static int re_gmii_writereg	(device_t, int, int, int);
269 
270 static int re_miibus_readreg	(device_t, int, int);
271 static int re_miibus_writereg	(device_t, int, int, int);
272 static void re_miibus_statchg	(device_t);
273 
274 static void re_set_rxmode		(struct rl_softc *);
275 static void re_reset		(struct rl_softc *);
276 static void re_setwol		(struct rl_softc *);
277 static void re_clrwol		(struct rl_softc *);
278 
279 #ifdef RE_DIAG
280 static int re_diag		(struct rl_softc *);
281 #endif
282 
283 static device_method_t re_methods[] = {
284 	/* Device interface */
285 	DEVMETHOD(device_probe,		re_probe),
286 	DEVMETHOD(device_attach,	re_attach),
287 	DEVMETHOD(device_detach,	re_detach),
288 	DEVMETHOD(device_suspend,	re_suspend),
289 	DEVMETHOD(device_resume,	re_resume),
290 	DEVMETHOD(device_shutdown,	re_shutdown),
291 
292 	/* bus interface */
293 	DEVMETHOD(bus_print_child,	bus_generic_print_child),
294 	DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
295 
296 	/* MII interface */
297 	DEVMETHOD(miibus_readreg,	re_miibus_readreg),
298 	DEVMETHOD(miibus_writereg,	re_miibus_writereg),
299 	DEVMETHOD(miibus_statchg,	re_miibus_statchg),
300 
301 	{ 0, 0 }
302 };
303 
304 static driver_t re_driver = {
305 	"re",
306 	re_methods,
307 	sizeof(struct rl_softc)
308 };
309 
310 static devclass_t re_devclass;
311 
312 DRIVER_MODULE(re, pci, re_driver, re_devclass, 0, 0);
313 DRIVER_MODULE(miibus, re, miibus_driver, miibus_devclass, 0, 0);
314 
315 #define EE_SET(x)					\
316 	CSR_WRITE_1(sc, RL_EECMD,			\
317 		CSR_READ_1(sc, RL_EECMD) | x)
318 
319 #define EE_CLR(x)					\
320 	CSR_WRITE_1(sc, RL_EECMD,			\
321 		CSR_READ_1(sc, RL_EECMD) & ~x)
322 
323 /*
324  * Send a read command and address to the EEPROM, check for ACK.
325  */
326 static void
327 re_eeprom_putbyte(struct rl_softc *sc, int addr)
328 {
329 	int			d, i;
330 
331 	d = addr | (RL_9346_READ << sc->rl_eewidth);
332 
333 	/*
334 	 * Feed in each bit and strobe the clock.
335 	 */
336 
337 	for (i = 1 << (sc->rl_eewidth + 3); i; i >>= 1) {
338 		if (d & i) {
339 			EE_SET(RL_EE_DATAIN);
340 		} else {
341 			EE_CLR(RL_EE_DATAIN);
342 		}
343 		DELAY(100);
344 		EE_SET(RL_EE_CLK);
345 		DELAY(150);
346 		EE_CLR(RL_EE_CLK);
347 		DELAY(100);
348 	}
349 }
350 
351 /*
352  * Read a word of data stored in the EEPROM at address 'addr.'
353  */
354 static void
355 re_eeprom_getword(struct rl_softc *sc, int addr, u_int16_t *dest)
356 {
357 	int			i;
358 	u_int16_t		word = 0;
359 
360 	/*
361 	 * Send address of word we want to read.
362 	 */
363 	re_eeprom_putbyte(sc, addr);
364 
365 	/*
366 	 * Start reading bits from EEPROM.
367 	 */
368 	for (i = 0x8000; i; i >>= 1) {
369 		EE_SET(RL_EE_CLK);
370 		DELAY(100);
371 		if (CSR_READ_1(sc, RL_EECMD) & RL_EE_DATAOUT)
372 			word |= i;
373 		EE_CLR(RL_EE_CLK);
374 		DELAY(100);
375 	}
376 
377 	*dest = word;
378 }
379 
380 /*
381  * Read a sequence of words from the EEPROM.
382  */
383 static void
384 re_read_eeprom(struct rl_softc *sc, caddr_t dest, int off, int cnt)
385 {
386 	int			i;
387 	u_int16_t		word = 0, *ptr;
388 
389 	CSR_SETBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM);
390 
391         DELAY(100);
392 
393 	for (i = 0; i < cnt; i++) {
394 		CSR_SETBIT_1(sc, RL_EECMD, RL_EE_SEL);
395 		re_eeprom_getword(sc, off + i, &word);
396 		CSR_CLRBIT_1(sc, RL_EECMD, RL_EE_SEL);
397 		ptr = (u_int16_t *)(dest + (i * 2));
398                 *ptr = word;
399 	}
400 
401 	CSR_CLRBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM);
402 }
403 
404 static int
405 re_gmii_readreg(device_t dev, int phy, int reg)
406 {
407 	struct rl_softc		*sc;
408 	u_int32_t		rval;
409 	int			i;
410 
411 	if (phy != 1)
412 		return (0);
413 
414 	sc = device_get_softc(dev);
415 
416 	/* Let the rgephy driver read the GMEDIASTAT register */
417 
418 	if (reg == RL_GMEDIASTAT) {
419 		rval = CSR_READ_1(sc, RL_GMEDIASTAT);
420 		return (rval);
421 	}
422 
423 	CSR_WRITE_4(sc, RL_PHYAR, reg << 16);
424 	DELAY(1000);
425 
426 	for (i = 0; i < RL_PHY_TIMEOUT; i++) {
427 		rval = CSR_READ_4(sc, RL_PHYAR);
428 		if (rval & RL_PHYAR_BUSY)
429 			break;
430 		DELAY(100);
431 	}
432 
433 	if (i == RL_PHY_TIMEOUT) {
434 		device_printf(sc->rl_dev, "PHY read failed\n");
435 		return (0);
436 	}
437 
438 	return (rval & RL_PHYAR_PHYDATA);
439 }
440 
441 static int
442 re_gmii_writereg(device_t dev, int phy, int reg, int data)
443 {
444 	struct rl_softc		*sc;
445 	u_int32_t		rval;
446 	int			i;
447 
448 	sc = device_get_softc(dev);
449 
450 	CSR_WRITE_4(sc, RL_PHYAR, (reg << 16) |
451 	    (data & RL_PHYAR_PHYDATA) | RL_PHYAR_BUSY);
452 	DELAY(1000);
453 
454 	for (i = 0; i < RL_PHY_TIMEOUT; i++) {
455 		rval = CSR_READ_4(sc, RL_PHYAR);
456 		if (!(rval & RL_PHYAR_BUSY))
457 			break;
458 		DELAY(100);
459 	}
460 
461 	if (i == RL_PHY_TIMEOUT) {
462 		device_printf(sc->rl_dev, "PHY write failed\n");
463 		return (0);
464 	}
465 
466 	return (0);
467 }
468 
469 static int
470 re_miibus_readreg(device_t dev, int phy, int reg)
471 {
472 	struct rl_softc		*sc;
473 	u_int16_t		rval = 0;
474 	u_int16_t		re8139_reg = 0;
475 
476 	sc = device_get_softc(dev);
477 
478 	if (sc->rl_type == RL_8169) {
479 		rval = re_gmii_readreg(dev, phy, reg);
480 		return (rval);
481 	}
482 
483 	/* Pretend the internal PHY is only at address 0 */
484 	if (phy) {
485 		return (0);
486 	}
487 	switch (reg) {
488 	case MII_BMCR:
489 		re8139_reg = RL_BMCR;
490 		break;
491 	case MII_BMSR:
492 		re8139_reg = RL_BMSR;
493 		break;
494 	case MII_ANAR:
495 		re8139_reg = RL_ANAR;
496 		break;
497 	case MII_ANER:
498 		re8139_reg = RL_ANER;
499 		break;
500 	case MII_ANLPAR:
501 		re8139_reg = RL_LPAR;
502 		break;
503 	case MII_PHYIDR1:
504 	case MII_PHYIDR2:
505 		return (0);
506 	/*
507 	 * Allow the rlphy driver to read the media status
508 	 * register. If we have a link partner which does not
509 	 * support NWAY, this is the register which will tell
510 	 * us the results of parallel detection.
511 	 */
512 	case RL_MEDIASTAT:
513 		rval = CSR_READ_1(sc, RL_MEDIASTAT);
514 		return (rval);
515 	default:
516 		device_printf(sc->rl_dev, "bad phy register\n");
517 		return (0);
518 	}
519 	rval = CSR_READ_2(sc, re8139_reg);
520 	if (sc->rl_type == RL_8139CPLUS && re8139_reg == RL_BMCR) {
521 		/* 8139C+ has different bit layout. */
522 		rval &= ~(BMCR_LOOP | BMCR_ISO);
523 	}
524 	return (rval);
525 }
526 
527 static int
528 re_miibus_writereg(device_t dev, int phy, int reg, int data)
529 {
530 	struct rl_softc		*sc;
531 	u_int16_t		re8139_reg = 0;
532 	int			rval = 0;
533 
534 	sc = device_get_softc(dev);
535 
536 	if (sc->rl_type == RL_8169) {
537 		rval = re_gmii_writereg(dev, phy, reg, data);
538 		return (rval);
539 	}
540 
541 	/* Pretend the internal PHY is only at address 0 */
542 	if (phy)
543 		return (0);
544 
545 	switch (reg) {
546 	case MII_BMCR:
547 		re8139_reg = RL_BMCR;
548 		if (sc->rl_type == RL_8139CPLUS) {
549 			/* 8139C+ has different bit layout. */
550 			data &= ~(BMCR_LOOP | BMCR_ISO);
551 		}
552 		break;
553 	case MII_BMSR:
554 		re8139_reg = RL_BMSR;
555 		break;
556 	case MII_ANAR:
557 		re8139_reg = RL_ANAR;
558 		break;
559 	case MII_ANER:
560 		re8139_reg = RL_ANER;
561 		break;
562 	case MII_ANLPAR:
563 		re8139_reg = RL_LPAR;
564 		break;
565 	case MII_PHYIDR1:
566 	case MII_PHYIDR2:
567 		return (0);
568 		break;
569 	default:
570 		device_printf(sc->rl_dev, "bad phy register\n");
571 		return (0);
572 	}
573 	CSR_WRITE_2(sc, re8139_reg, data);
574 	return (0);
575 }
576 
577 static void
578 re_miibus_statchg(device_t dev)
579 {
580 	struct rl_softc		*sc;
581 	struct ifnet		*ifp;
582 	struct mii_data		*mii;
583 
584 	sc = device_get_softc(dev);
585 	mii = device_get_softc(sc->rl_miibus);
586 	ifp = sc->rl_ifp;
587 	if (mii == NULL || ifp == NULL ||
588 	    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
589 		return;
590 
591 	sc->rl_flags &= ~RL_FLAG_LINK;
592 	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
593 	    (IFM_ACTIVE | IFM_AVALID)) {
594 		switch (IFM_SUBTYPE(mii->mii_media_active)) {
595 		case IFM_10_T:
596 		case IFM_100_TX:
597 			sc->rl_flags |= RL_FLAG_LINK;
598 			break;
599 		case IFM_1000_T:
600 			if ((sc->rl_flags & RL_FLAG_FASTETHER) != 0)
601 				break;
602 			sc->rl_flags |= RL_FLAG_LINK;
603 			break;
604 		default:
605 			break;
606 		}
607 	}
608 	/*
609 	 * RealTek controllers does not provide any interface to
610 	 * Tx/Rx MACs for resolved speed, duplex and flow-control
611 	 * parameters.
612 	 */
613 }
614 
615 /*
616  * Set the RX configuration and 64-bit multicast hash filter.
617  */
618 static void
619 re_set_rxmode(struct rl_softc *sc)
620 {
621 	struct ifnet		*ifp;
622 	struct ifmultiaddr	*ifma;
623 	uint32_t		hashes[2] = { 0, 0 };
624 	uint32_t		h, rxfilt;
625 
626 	RL_LOCK_ASSERT(sc);
627 
628 	ifp = sc->rl_ifp;
629 
630 	rxfilt = RL_RXCFG_CONFIG | RL_RXCFG_RX_INDIV | RL_RXCFG_RX_BROAD;
631 
632 	if (ifp->if_flags & (IFF_ALLMULTI | IFF_PROMISC)) {
633 		if (ifp->if_flags & IFF_PROMISC)
634 			rxfilt |= RL_RXCFG_RX_ALLPHYS;
635 		/*
636 		 * Unlike other hardwares, we have to explicitly set
637 		 * RL_RXCFG_RX_MULTI to receive multicast frames in
638 		 * promiscuous mode.
639 		 */
640 		rxfilt |= RL_RXCFG_RX_MULTI;
641 		hashes[0] = hashes[1] = 0xffffffff;
642 		goto done;
643 	}
644 
645 	if_maddr_rlock(ifp);
646 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
647 		if (ifma->ifma_addr->sa_family != AF_LINK)
648 			continue;
649 		h = ether_crc32_be(LLADDR((struct sockaddr_dl *)
650 		    ifma->ifma_addr), ETHER_ADDR_LEN) >> 26;
651 		if (h < 32)
652 			hashes[0] |= (1 << h);
653 		else
654 			hashes[1] |= (1 << (h - 32));
655 	}
656 	if_maddr_runlock(ifp);
657 
658 	if (hashes[0] != 0 || hashes[1] != 0) {
659 		/*
660 		 * For some unfathomable reason, RealTek decided to
661 		 * reverse the order of the multicast hash registers
662 		 * in the PCI Express parts.  This means we have to
663 		 * write the hash pattern in reverse order for those
664 		 * devices.
665 		 */
666 		if ((sc->rl_flags & RL_FLAG_PCIE) != 0) {
667 			h = bswap32(hashes[0]);
668 			hashes[0] = bswap32(hashes[1]);
669 			hashes[1] = h;
670 		}
671 		rxfilt |= RL_RXCFG_RX_MULTI;
672 	}
673 
674 done:
675 	CSR_WRITE_4(sc, RL_MAR0, hashes[0]);
676 	CSR_WRITE_4(sc, RL_MAR4, hashes[1]);
677 	CSR_WRITE_4(sc, RL_RXCFG, rxfilt);
678 }
679 
680 static void
681 re_reset(struct rl_softc *sc)
682 {
683 	int			i;
684 
685 	RL_LOCK_ASSERT(sc);
686 
687 	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RESET);
688 
689 	for (i = 0; i < RL_TIMEOUT; i++) {
690 		DELAY(10);
691 		if (!(CSR_READ_1(sc, RL_COMMAND) & RL_CMD_RESET))
692 			break;
693 	}
694 	if (i == RL_TIMEOUT)
695 		device_printf(sc->rl_dev, "reset never completed!\n");
696 
697 	if ((sc->rl_flags & RL_FLAG_MACRESET) != 0)
698 		CSR_WRITE_1(sc, 0x82, 1);
699 	if (sc->rl_hwrev == RL_HWREV_8169S)
700 		re_gmii_writereg(sc->rl_dev, 1, 0x0b, 0);
701 }
702 
703 #ifdef RE_DIAG
704 
705 /*
706  * The following routine is designed to test for a defect on some
707  * 32-bit 8169 cards. Some of these NICs have the REQ64# and ACK64#
708  * lines connected to the bus, however for a 32-bit only card, they
709  * should be pulled high. The result of this defect is that the
710  * NIC will not work right if you plug it into a 64-bit slot: DMA
711  * operations will be done with 64-bit transfers, which will fail
712  * because the 64-bit data lines aren't connected.
713  *
714  * There's no way to work around this (short of talking a soldering
715  * iron to the board), however we can detect it. The method we use
716  * here is to put the NIC into digital loopback mode, set the receiver
717  * to promiscuous mode, and then try to send a frame. We then compare
718  * the frame data we sent to what was received. If the data matches,
719  * then the NIC is working correctly, otherwise we know the user has
720  * a defective NIC which has been mistakenly plugged into a 64-bit PCI
721  * slot. In the latter case, there's no way the NIC can work correctly,
722  * so we print out a message on the console and abort the device attach.
723  */
724 
725 static int
726 re_diag(struct rl_softc *sc)
727 {
728 	struct ifnet		*ifp = sc->rl_ifp;
729 	struct mbuf		*m0;
730 	struct ether_header	*eh;
731 	struct rl_desc		*cur_rx;
732 	u_int16_t		status;
733 	u_int32_t		rxstat;
734 	int			total_len, i, error = 0, phyaddr;
735 	u_int8_t		dst[] = { 0x00, 'h', 'e', 'l', 'l', 'o' };
736 	u_int8_t		src[] = { 0x00, 'w', 'o', 'r', 'l', 'd' };
737 
738 	/* Allocate a single mbuf */
739 	MGETHDR(m0, M_DONTWAIT, MT_DATA);
740 	if (m0 == NULL)
741 		return (ENOBUFS);
742 
743 	RL_LOCK(sc);
744 
745 	/*
746 	 * Initialize the NIC in test mode. This sets the chip up
747 	 * so that it can send and receive frames, but performs the
748 	 * following special functions:
749 	 * - Puts receiver in promiscuous mode
750 	 * - Enables digital loopback mode
751 	 * - Leaves interrupts turned off
752 	 */
753 
754 	ifp->if_flags |= IFF_PROMISC;
755 	sc->rl_testmode = 1;
756 	re_init_locked(sc);
757 	sc->rl_flags |= RL_FLAG_LINK;
758 	if (sc->rl_type == RL_8169)
759 		phyaddr = 1;
760 	else
761 		phyaddr = 0;
762 
763 	re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_RESET);
764 	for (i = 0; i < RL_TIMEOUT; i++) {
765 		status = re_miibus_readreg(sc->rl_dev, phyaddr, MII_BMCR);
766 		if (!(status & BMCR_RESET))
767 			break;
768 	}
769 
770 	re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_LOOP);
771 	CSR_WRITE_2(sc, RL_ISR, RL_INTRS);
772 
773 	DELAY(100000);
774 
775 	/* Put some data in the mbuf */
776 
777 	eh = mtod(m0, struct ether_header *);
778 	bcopy ((char *)&dst, eh->ether_dhost, ETHER_ADDR_LEN);
779 	bcopy ((char *)&src, eh->ether_shost, ETHER_ADDR_LEN);
780 	eh->ether_type = htons(ETHERTYPE_IP);
781 	m0->m_pkthdr.len = m0->m_len = ETHER_MIN_LEN - ETHER_CRC_LEN;
782 
783 	/*
784 	 * Queue the packet, start transmission.
785 	 * Note: IF_HANDOFF() ultimately calls re_start() for us.
786 	 */
787 
788 	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
789 	RL_UNLOCK(sc);
790 	/* XXX: re_diag must not be called when in ALTQ mode */
791 	IF_HANDOFF(&ifp->if_snd, m0, ifp);
792 	RL_LOCK(sc);
793 	m0 = NULL;
794 
795 	/* Wait for it to propagate through the chip */
796 
797 	DELAY(100000);
798 	for (i = 0; i < RL_TIMEOUT; i++) {
799 		status = CSR_READ_2(sc, RL_ISR);
800 		CSR_WRITE_2(sc, RL_ISR, status);
801 		if ((status & (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK)) ==
802 		    (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK))
803 			break;
804 		DELAY(10);
805 	}
806 
807 	if (i == RL_TIMEOUT) {
808 		device_printf(sc->rl_dev,
809 		    "diagnostic failed, failed to receive packet in"
810 		    " loopback mode\n");
811 		error = EIO;
812 		goto done;
813 	}
814 
815 	/*
816 	 * The packet should have been dumped into the first
817 	 * entry in the RX DMA ring. Grab it from there.
818 	 */
819 
820 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
821 	    sc->rl_ldata.rl_rx_list_map,
822 	    BUS_DMASYNC_POSTREAD);
823 	bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag,
824 	    sc->rl_ldata.rl_rx_desc[0].rx_dmamap,
825 	    BUS_DMASYNC_POSTREAD);
826 	bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag,
827 	    sc->rl_ldata.rl_rx_desc[0].rx_dmamap);
828 
829 	m0 = sc->rl_ldata.rl_rx_desc[0].rx_m;
830 	sc->rl_ldata.rl_rx_desc[0].rx_m = NULL;
831 	eh = mtod(m0, struct ether_header *);
832 
833 	cur_rx = &sc->rl_ldata.rl_rx_list[0];
834 	total_len = RL_RXBYTES(cur_rx);
835 	rxstat = le32toh(cur_rx->rl_cmdstat);
836 
837 	if (total_len != ETHER_MIN_LEN) {
838 		device_printf(sc->rl_dev,
839 		    "diagnostic failed, received short packet\n");
840 		error = EIO;
841 		goto done;
842 	}
843 
844 	/* Test that the received packet data matches what we sent. */
845 
846 	if (bcmp((char *)&eh->ether_dhost, (char *)&dst, ETHER_ADDR_LEN) ||
847 	    bcmp((char *)&eh->ether_shost, (char *)&src, ETHER_ADDR_LEN) ||
848 	    ntohs(eh->ether_type) != ETHERTYPE_IP) {
849 		device_printf(sc->rl_dev, "WARNING, DMA FAILURE!\n");
850 		device_printf(sc->rl_dev, "expected TX data: %6D/%6D/0x%x\n",
851 		    dst, ":", src, ":", ETHERTYPE_IP);
852 		device_printf(sc->rl_dev, "received RX data: %6D/%6D/0x%x\n",
853 		    eh->ether_dhost, ":",  eh->ether_shost, ":",
854 		    ntohs(eh->ether_type));
855 		device_printf(sc->rl_dev, "You may have a defective 32-bit "
856 		    "NIC plugged into a 64-bit PCI slot.\n");
857 		device_printf(sc->rl_dev, "Please re-install the NIC in a "
858 		    "32-bit slot for proper operation.\n");
859 		device_printf(sc->rl_dev, "Read the re(4) man page for more "
860 		    "details.\n");
861 		error = EIO;
862 	}
863 
864 done:
865 	/* Turn interface off, release resources */
866 
867 	sc->rl_testmode = 0;
868 	sc->rl_flags &= ~RL_FLAG_LINK;
869 	ifp->if_flags &= ~IFF_PROMISC;
870 	re_stop(sc);
871 	if (m0 != NULL)
872 		m_freem(m0);
873 
874 	RL_UNLOCK(sc);
875 
876 	return (error);
877 }
878 
879 #endif
880 
881 /*
882  * Probe for a RealTek 8139C+/8169/8110 chip. Check the PCI vendor and device
883  * IDs against our list and return a device name if we find a match.
884  */
885 static int
886 re_probe(device_t dev)
887 {
888 	struct rl_type		*t;
889 	uint16_t		devid, vendor;
890 	uint16_t		revid, sdevid;
891 	int			i;
892 
893 	vendor = pci_get_vendor(dev);
894 	devid = pci_get_device(dev);
895 	revid = pci_get_revid(dev);
896 	sdevid = pci_get_subdevice(dev);
897 
898 	if (vendor == LINKSYS_VENDORID && devid == LINKSYS_DEVICEID_EG1032) {
899 		if (sdevid != LINKSYS_SUBDEVICE_EG1032_REV3) {
900 			/*
901 			 * Only attach to rev. 3 of the Linksys EG1032 adapter.
902 			 * Rev. 2 is supported by sk(4).
903 			 */
904 			return (ENXIO);
905 		}
906 	}
907 
908 	if (vendor == RT_VENDORID && devid == RT_DEVICEID_8139) {
909 		if (revid != 0x20) {
910 			/* 8139, let rl(4) take care of this device. */
911 			return (ENXIO);
912 		}
913 	}
914 
915 	t = re_devs;
916 	for (i = 0; i < sizeof(re_devs) / sizeof(re_devs[0]); i++, t++) {
917 		if (vendor == t->rl_vid && devid == t->rl_did) {
918 			device_set_desc(dev, t->rl_name);
919 			return (BUS_PROBE_DEFAULT);
920 		}
921 	}
922 
923 	return (ENXIO);
924 }
925 
926 /*
927  * Map a single buffer address.
928  */
929 
930 static void
931 re_dma_map_addr(void *arg, bus_dma_segment_t *segs, int nseg, int error)
932 {
933 	bus_addr_t		*addr;
934 
935 	if (error)
936 		return;
937 
938 	KASSERT(nseg == 1, ("too many DMA segments, %d should be 1", nseg));
939 	addr = arg;
940 	*addr = segs->ds_addr;
941 }
942 
943 static int
944 re_allocmem(device_t dev, struct rl_softc *sc)
945 {
946 	bus_size_t		rx_list_size, tx_list_size;
947 	int			error;
948 	int			i;
949 
950 	rx_list_size = sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc);
951 	tx_list_size = sc->rl_ldata.rl_tx_desc_cnt * sizeof(struct rl_desc);
952 
953 	/*
954 	 * Allocate the parent bus DMA tag appropriate for PCI.
955 	 * In order to use DAC, RL_CPLUSCMD_PCI_DAC bit of RL_CPLUS_CMD
956 	 * register should be set. However some RealTek chips are known
957 	 * to be buggy on DAC handling, therefore disable DAC by limiting
958 	 * DMA address space to 32bit. PCIe variants of RealTek chips
959 	 * may not have the limitation but I took safer path.
960 	 */
961 	error = bus_dma_tag_create(bus_get_dma_tag(dev), 1, 0,
962 	    BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL, NULL,
963 	    BUS_SPACE_MAXSIZE_32BIT, 0, BUS_SPACE_MAXSIZE_32BIT, 0,
964 	    NULL, NULL, &sc->rl_parent_tag);
965 	if (error) {
966 		device_printf(dev, "could not allocate parent DMA tag\n");
967 		return (error);
968 	}
969 
970 	/*
971 	 * Allocate map for TX mbufs.
972 	 */
973 	error = bus_dma_tag_create(sc->rl_parent_tag, 1, 0,
974 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL,
975 	    NULL, MCLBYTES * RL_NTXSEGS, RL_NTXSEGS, 4096, 0,
976 	    NULL, NULL, &sc->rl_ldata.rl_tx_mtag);
977 	if (error) {
978 		device_printf(dev, "could not allocate TX DMA tag\n");
979 		return (error);
980 	}
981 
982 	/*
983 	 * Allocate map for RX mbufs.
984 	 */
985 
986 	error = bus_dma_tag_create(sc->rl_parent_tag, sizeof(uint64_t), 0,
987 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
988 	    MCLBYTES, 1, MCLBYTES, 0, NULL, NULL, &sc->rl_ldata.rl_rx_mtag);
989 	if (error) {
990 		device_printf(dev, "could not allocate RX DMA tag\n");
991 		return (error);
992 	}
993 
994 	/*
995 	 * Allocate map for TX descriptor list.
996 	 */
997 	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
998 	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
999 	    NULL, tx_list_size, 1, tx_list_size, 0,
1000 	    NULL, NULL, &sc->rl_ldata.rl_tx_list_tag);
1001 	if (error) {
1002 		device_printf(dev, "could not allocate TX DMA ring tag\n");
1003 		return (error);
1004 	}
1005 
1006 	/* Allocate DMA'able memory for the TX ring */
1007 
1008 	error = bus_dmamem_alloc(sc->rl_ldata.rl_tx_list_tag,
1009 	    (void **)&sc->rl_ldata.rl_tx_list,
1010 	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1011 	    &sc->rl_ldata.rl_tx_list_map);
1012 	if (error) {
1013 		device_printf(dev, "could not allocate TX DMA ring\n");
1014 		return (error);
1015 	}
1016 
1017 	/* Load the map for the TX ring. */
1018 
1019 	sc->rl_ldata.rl_tx_list_addr = 0;
1020 	error = bus_dmamap_load(sc->rl_ldata.rl_tx_list_tag,
1021 	     sc->rl_ldata.rl_tx_list_map, sc->rl_ldata.rl_tx_list,
1022 	     tx_list_size, re_dma_map_addr,
1023 	     &sc->rl_ldata.rl_tx_list_addr, BUS_DMA_NOWAIT);
1024 	if (error != 0 || sc->rl_ldata.rl_tx_list_addr == 0) {
1025 		device_printf(dev, "could not load TX DMA ring\n");
1026 		return (ENOMEM);
1027 	}
1028 
1029 	/* Create DMA maps for TX buffers */
1030 
1031 	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
1032 		error = bus_dmamap_create(sc->rl_ldata.rl_tx_mtag, 0,
1033 		    &sc->rl_ldata.rl_tx_desc[i].tx_dmamap);
1034 		if (error) {
1035 			device_printf(dev, "could not create DMA map for TX\n");
1036 			return (error);
1037 		}
1038 	}
1039 
1040 	/*
1041 	 * Allocate map for RX descriptor list.
1042 	 */
1043 	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
1044 	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1045 	    NULL, rx_list_size, 1, rx_list_size, 0,
1046 	    NULL, NULL, &sc->rl_ldata.rl_rx_list_tag);
1047 	if (error) {
1048 		device_printf(dev, "could not create RX DMA ring tag\n");
1049 		return (error);
1050 	}
1051 
1052 	/* Allocate DMA'able memory for the RX ring */
1053 
1054 	error = bus_dmamem_alloc(sc->rl_ldata.rl_rx_list_tag,
1055 	    (void **)&sc->rl_ldata.rl_rx_list,
1056 	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1057 	    &sc->rl_ldata.rl_rx_list_map);
1058 	if (error) {
1059 		device_printf(dev, "could not allocate RX DMA ring\n");
1060 		return (error);
1061 	}
1062 
1063 	/* Load the map for the RX ring. */
1064 
1065 	sc->rl_ldata.rl_rx_list_addr = 0;
1066 	error = bus_dmamap_load(sc->rl_ldata.rl_rx_list_tag,
1067 	     sc->rl_ldata.rl_rx_list_map, sc->rl_ldata.rl_rx_list,
1068 	     rx_list_size, re_dma_map_addr,
1069 	     &sc->rl_ldata.rl_rx_list_addr, BUS_DMA_NOWAIT);
1070 	if (error != 0 || sc->rl_ldata.rl_rx_list_addr == 0) {
1071 		device_printf(dev, "could not load RX DMA ring\n");
1072 		return (ENOMEM);
1073 	}
1074 
1075 	/* Create DMA maps for RX buffers */
1076 
1077 	error = bus_dmamap_create(sc->rl_ldata.rl_rx_mtag, 0,
1078 	    &sc->rl_ldata.rl_rx_sparemap);
1079 	if (error) {
1080 		device_printf(dev, "could not create spare DMA map for RX\n");
1081 		return (error);
1082 	}
1083 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1084 		error = bus_dmamap_create(sc->rl_ldata.rl_rx_mtag, 0,
1085 		    &sc->rl_ldata.rl_rx_desc[i].rx_dmamap);
1086 		if (error) {
1087 			device_printf(dev, "could not create DMA map for RX\n");
1088 			return (error);
1089 		}
1090 	}
1091 
1092 	return (0);
1093 }
1094 
1095 /*
1096  * Attach the interface. Allocate softc structures, do ifmedia
1097  * setup and ethernet/BPF attach.
1098  */
1099 static int
1100 re_attach(device_t dev)
1101 {
1102 	u_char			eaddr[ETHER_ADDR_LEN];
1103 	u_int16_t		as[ETHER_ADDR_LEN / 2];
1104 	struct rl_softc		*sc;
1105 	struct ifnet		*ifp;
1106 	struct rl_hwrev		*hw_rev;
1107 	int			hwrev;
1108 	u_int16_t		devid, re_did = 0;
1109 	int			error = 0, rid, i;
1110 	int			msic, reg;
1111 	uint8_t			cfg;
1112 
1113 	sc = device_get_softc(dev);
1114 	sc->rl_dev = dev;
1115 
1116 	mtx_init(&sc->rl_mtx, device_get_nameunit(dev), MTX_NETWORK_LOCK,
1117 	    MTX_DEF);
1118 	callout_init_mtx(&sc->rl_stat_callout, &sc->rl_mtx, 0);
1119 
1120 	/*
1121 	 * Map control/status registers.
1122 	 */
1123 	pci_enable_busmaster(dev);
1124 
1125 	devid = pci_get_device(dev);
1126 	/*
1127 	 * Prefer memory space register mapping over IO space.
1128 	 * Because RTL8169SC does not seem to work when memory mapping
1129 	 * is used always activate io mapping.
1130 	 */
1131 	if (devid == RT_DEVICEID_8169SC)
1132 		prefer_iomap = 1;
1133 	if (prefer_iomap == 0) {
1134 		sc->rl_res_id = PCIR_BAR(1);
1135 		sc->rl_res_type = SYS_RES_MEMORY;
1136 		/* RTL8168/8101E seems to use different BARs. */
1137 		if (devid == RT_DEVICEID_8168 || devid == RT_DEVICEID_8101E)
1138 			sc->rl_res_id = PCIR_BAR(2);
1139 	} else {
1140 		sc->rl_res_id = PCIR_BAR(0);
1141 		sc->rl_res_type = SYS_RES_IOPORT;
1142 	}
1143 	sc->rl_res = bus_alloc_resource_any(dev, sc->rl_res_type,
1144 	    &sc->rl_res_id, RF_ACTIVE);
1145 	if (sc->rl_res == NULL && prefer_iomap == 0) {
1146 		sc->rl_res_id = PCIR_BAR(0);
1147 		sc->rl_res_type = SYS_RES_IOPORT;
1148 		sc->rl_res = bus_alloc_resource_any(dev, sc->rl_res_type,
1149 		    &sc->rl_res_id, RF_ACTIVE);
1150 	}
1151 	if (sc->rl_res == NULL) {
1152 		device_printf(dev, "couldn't map ports/memory\n");
1153 		error = ENXIO;
1154 		goto fail;
1155 	}
1156 
1157 	sc->rl_btag = rman_get_bustag(sc->rl_res);
1158 	sc->rl_bhandle = rman_get_bushandle(sc->rl_res);
1159 
1160 	msic = 0;
1161 	if (pci_find_extcap(dev, PCIY_EXPRESS, &reg) == 0) {
1162 		sc->rl_flags |= RL_FLAG_PCIE;
1163 		msic = pci_msi_count(dev);
1164 		if (bootverbose)
1165 			device_printf(dev, "MSI count : %d\n", msic);
1166 	}
1167 	if (msic > 0 && msi_disable == 0) {
1168 		msic = 1;
1169 		if (pci_alloc_msi(dev, &msic) == 0) {
1170 			if (msic == RL_MSI_MESSAGES) {
1171 				device_printf(dev, "Using %d MSI messages\n",
1172 				    msic);
1173 				sc->rl_flags |= RL_FLAG_MSI;
1174 				/* Explicitly set MSI enable bit. */
1175 				CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1176 				cfg = CSR_READ_1(sc, RL_CFG2);
1177 				cfg |= RL_CFG2_MSI;
1178 				CSR_WRITE_1(sc, RL_CFG2, cfg);
1179 				CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1180 			} else
1181 				pci_release_msi(dev);
1182 		}
1183 	}
1184 
1185 	/* Allocate interrupt */
1186 	if ((sc->rl_flags & RL_FLAG_MSI) == 0) {
1187 		rid = 0;
1188 		sc->rl_irq[0] = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
1189 		    RF_SHAREABLE | RF_ACTIVE);
1190 		if (sc->rl_irq[0] == NULL) {
1191 			device_printf(dev, "couldn't allocate IRQ resources\n");
1192 			error = ENXIO;
1193 			goto fail;
1194 		}
1195 	} else {
1196 		for (i = 0, rid = 1; i < RL_MSI_MESSAGES; i++, rid++) {
1197 			sc->rl_irq[i] = bus_alloc_resource_any(dev,
1198 			    SYS_RES_IRQ, &rid, RF_ACTIVE);
1199 			if (sc->rl_irq[i] == NULL) {
1200 				device_printf(dev,
1201 				    "couldn't llocate IRQ resources for "
1202 				    "message %d\n", rid);
1203 				error = ENXIO;
1204 				goto fail;
1205 			}
1206 		}
1207 	}
1208 
1209 	if ((sc->rl_flags & RL_FLAG_MSI) == 0) {
1210 		CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1211 		cfg = CSR_READ_1(sc, RL_CFG2);
1212 		if ((cfg & RL_CFG2_MSI) != 0) {
1213 			device_printf(dev, "turning off MSI enable bit.\n");
1214 			cfg &= ~RL_CFG2_MSI;
1215 			CSR_WRITE_1(sc, RL_CFG2, cfg);
1216 		}
1217 		CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1218 	}
1219 
1220 	/* Reset the adapter. */
1221 	RL_LOCK(sc);
1222 	re_reset(sc);
1223 	RL_UNLOCK(sc);
1224 
1225 	hw_rev = re_hwrevs;
1226 	hwrev = CSR_READ_4(sc, RL_TXCFG);
1227 	switch (hwrev & 0x70000000) {
1228 	case 0x00000000:
1229 	case 0x10000000:
1230 		device_printf(dev, "Chip rev. 0x%08x\n", hwrev & 0xfc800000);
1231 		hwrev &= (RL_TXCFG_HWREV | 0x80000000);
1232 		break;
1233 	default:
1234 		device_printf(dev, "Chip rev. 0x%08x\n", hwrev & 0x7c800000);
1235 		hwrev &= RL_TXCFG_HWREV;
1236 		break;
1237 	}
1238 	device_printf(dev, "MAC rev. 0x%08x\n", hwrev & 0x00700000);
1239 	while (hw_rev->rl_desc != NULL) {
1240 		if (hw_rev->rl_rev == hwrev) {
1241 			sc->rl_type = hw_rev->rl_type;
1242 			sc->rl_hwrev = hw_rev->rl_rev;
1243 			break;
1244 		}
1245 		hw_rev++;
1246 	}
1247 	if (hw_rev->rl_desc == NULL) {
1248 		device_printf(dev, "Unknown H/W revision: 0x%08x\n", hwrev);
1249 		error = ENXIO;
1250 		goto fail;
1251 	}
1252 
1253 	switch (hw_rev->rl_rev) {
1254 	case RL_HWREV_8139CPLUS:
1255 		sc->rl_flags |= RL_FLAG_NOJUMBO | RL_FLAG_FASTETHER |
1256 		    RL_FLAG_AUTOPAD;
1257 		break;
1258 	case RL_HWREV_8100E:
1259 	case RL_HWREV_8101E:
1260 		sc->rl_flags |= RL_FLAG_NOJUMBO | RL_FLAG_PHYWAKE |
1261 		    RL_FLAG_FASTETHER;
1262 		break;
1263 	case RL_HWREV_8102E:
1264 	case RL_HWREV_8102EL:
1265 	case RL_HWREV_8102EL_SPIN1:
1266 		sc->rl_flags |= RL_FLAG_NOJUMBO | RL_FLAG_PHYWAKE |
1267 		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
1268 		    RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD;
1269 		break;
1270 	case RL_HWREV_8168_SPIN1:
1271 	case RL_HWREV_8168_SPIN2:
1272 		sc->rl_flags |= RL_FLAG_WOLRXENB;
1273 		/* FALLTHROUGH */
1274 	case RL_HWREV_8168_SPIN3:
1275 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_MACSTAT;
1276 		break;
1277 	case RL_HWREV_8168C_SPIN2:
1278 		sc->rl_flags |= RL_FLAG_MACSLEEP;
1279 		/* FALLTHROUGH */
1280 	case RL_HWREV_8168C:
1281 		if ((hwrev & 0x00700000) == 0x00200000)
1282 			sc->rl_flags |= RL_FLAG_MACSLEEP;
1283 		/* FALLTHROUGH */
1284 	case RL_HWREV_8168CP:
1285 	case RL_HWREV_8168D:
1286 	case RL_HWREV_8168DP:
1287 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
1288 		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP |
1289 		    RL_FLAG_AUTOPAD;
1290 		/*
1291 		 * These controllers support jumbo frame but it seems
1292 		 * that enabling it requires touching additional magic
1293 		 * registers. Depending on MAC revisions some
1294 		 * controllers need to disable checksum offload. So
1295 		 * disable jumbo frame until I have better idea what
1296 		 * it really requires to make it support.
1297 		 * RTL8168C/CP : supports up to 6KB jumbo frame.
1298 		 * RTL8111C/CP : supports up to 9KB jumbo frame.
1299 		 */
1300 		sc->rl_flags |= RL_FLAG_NOJUMBO;
1301 		break;
1302 	case RL_HWREV_8169_8110SB:
1303 	case RL_HWREV_8169_8110SBL:
1304 	case RL_HWREV_8169_8110SC:
1305 	case RL_HWREV_8169_8110SCE:
1306 		sc->rl_flags |= RL_FLAG_PHYWAKE;
1307 		/* FALLTHROUGH */
1308 	case RL_HWREV_8169:
1309 	case RL_HWREV_8169S:
1310 	case RL_HWREV_8110S:
1311 		sc->rl_flags |= RL_FLAG_MACRESET;
1312 		break;
1313 	default:
1314 		break;
1315 	}
1316 
1317 	/* Enable PME. */
1318 	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1319 	cfg = CSR_READ_1(sc, RL_CFG1);
1320 	cfg |= RL_CFG1_PME;
1321 	CSR_WRITE_1(sc, RL_CFG1, cfg);
1322 	cfg = CSR_READ_1(sc, RL_CFG5);
1323 	cfg &= RL_CFG5_PME_STS;
1324 	CSR_WRITE_1(sc, RL_CFG5, cfg);
1325 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1326 
1327 	if ((sc->rl_flags & RL_FLAG_PAR) != 0) {
1328 		/*
1329 		 * XXX Should have a better way to extract station
1330 		 * address from EEPROM.
1331 		 */
1332 		for (i = 0; i < ETHER_ADDR_LEN; i++)
1333 			eaddr[i] = CSR_READ_1(sc, RL_IDR0 + i);
1334 	} else {
1335 		sc->rl_eewidth = RL_9356_ADDR_LEN;
1336 		re_read_eeprom(sc, (caddr_t)&re_did, 0, 1);
1337 		if (re_did != 0x8129)
1338 			sc->rl_eewidth = RL_9346_ADDR_LEN;
1339 
1340 		/*
1341 		 * Get station address from the EEPROM.
1342 		 */
1343 		re_read_eeprom(sc, (caddr_t)as, RL_EE_EADDR, 3);
1344 		for (i = 0; i < ETHER_ADDR_LEN / 2; i++)
1345 			as[i] = le16toh(as[i]);
1346 		bcopy(as, eaddr, sizeof(eaddr));
1347 	}
1348 
1349 	if (sc->rl_type == RL_8169) {
1350 		/* Set RX length mask and number of descriptors. */
1351 		sc->rl_rxlenmask = RL_RDESC_STAT_GFRAGLEN;
1352 		sc->rl_txstart = RL_GTXSTART;
1353 		sc->rl_ldata.rl_tx_desc_cnt = RL_8169_TX_DESC_CNT;
1354 		sc->rl_ldata.rl_rx_desc_cnt = RL_8169_RX_DESC_CNT;
1355 	} else {
1356 		/* Set RX length mask and number of descriptors. */
1357 		sc->rl_rxlenmask = RL_RDESC_STAT_FRAGLEN;
1358 		sc->rl_txstart = RL_TXSTART;
1359 		sc->rl_ldata.rl_tx_desc_cnt = RL_8139_TX_DESC_CNT;
1360 		sc->rl_ldata.rl_rx_desc_cnt = RL_8139_RX_DESC_CNT;
1361 	}
1362 
1363 	error = re_allocmem(dev, sc);
1364 	if (error)
1365 		goto fail;
1366 
1367 	ifp = sc->rl_ifp = if_alloc(IFT_ETHER);
1368 	if (ifp == NULL) {
1369 		device_printf(dev, "can not if_alloc()\n");
1370 		error = ENOSPC;
1371 		goto fail;
1372 	}
1373 
1374 	/* Take controller out of deep sleep mode. */
1375 	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
1376 		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
1377 			CSR_WRITE_1(sc, RL_GPIO,
1378 			    CSR_READ_1(sc, RL_GPIO) | 0x01);
1379 		else
1380 			CSR_WRITE_1(sc, RL_GPIO,
1381 			    CSR_READ_1(sc, RL_GPIO) & ~0x01);
1382 	}
1383 
1384 	/* Take PHY out of power down mode. */
1385 	if ((sc->rl_flags & RL_FLAG_PHYWAKE) != 0) {
1386 		re_gmii_writereg(dev, 1, 0x1f, 0);
1387 		re_gmii_writereg(dev, 1, 0x0e, 0);
1388 	}
1389 
1390 	/* Do MII setup */
1391 	if (mii_phy_probe(dev, &sc->rl_miibus,
1392 	    re_ifmedia_upd, re_ifmedia_sts)) {
1393 		device_printf(dev, "MII without any phy!\n");
1394 		error = ENXIO;
1395 		goto fail;
1396 	}
1397 
1398 	ifp->if_softc = sc;
1399 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1400 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1401 	ifp->if_ioctl = re_ioctl;
1402 	ifp->if_start = re_start;
1403 	ifp->if_hwassist = RE_CSUM_FEATURES;
1404 	ifp->if_capabilities = IFCAP_HWCSUM;
1405 	ifp->if_capenable = ifp->if_capabilities;
1406 	ifp->if_init = re_init;
1407 	IFQ_SET_MAXLEN(&ifp->if_snd, RL_IFQ_MAXLEN);
1408 	ifp->if_snd.ifq_drv_maxlen = RL_IFQ_MAXLEN;
1409 	IFQ_SET_READY(&ifp->if_snd);
1410 
1411 	TASK_INIT(&sc->rl_txtask, 1, re_tx_task, ifp);
1412 	TASK_INIT(&sc->rl_inttask, 0, re_int_task, sc);
1413 
1414 	/*
1415 	 * XXX
1416 	 * Still have no idea how to make TSO work on 8168C, 8168CP,
1417 	 * 8111C and 8111CP.
1418 	 */
1419 	if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
1420 		ifp->if_hwassist |= CSUM_TSO;
1421 		ifp->if_capabilities |= IFCAP_TSO4;
1422 	}
1423 
1424 	/*
1425 	 * Call MI attach routine.
1426 	 */
1427 	ether_ifattach(ifp, eaddr);
1428 
1429 	/* VLAN capability setup */
1430 	ifp->if_capabilities |= IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING;
1431 	if (ifp->if_capabilities & IFCAP_HWCSUM)
1432 		ifp->if_capabilities |= IFCAP_VLAN_HWCSUM;
1433 	/* Enable WOL if PM is supported. */
1434 	if (pci_find_extcap(sc->rl_dev, PCIY_PMG, &reg) == 0)
1435 		ifp->if_capabilities |= IFCAP_WOL;
1436 	ifp->if_capenable = ifp->if_capabilities;
1437 	/*
1438 	 * Don't enable TSO by default. Under certain
1439 	 * circumtances the controller generated corrupted
1440 	 * packets in TSO size.
1441 	 */
1442 	ifp->if_hwassist &= ~CSUM_TSO;
1443 	ifp->if_capenable &= ~IFCAP_TSO4;
1444 #ifdef DEVICE_POLLING
1445 	ifp->if_capabilities |= IFCAP_POLLING;
1446 #endif
1447 	/*
1448 	 * Tell the upper layer(s) we support long frames.
1449 	 * Must appear after the call to ether_ifattach() because
1450 	 * ether_ifattach() sets ifi_hdrlen to the default value.
1451 	 */
1452 	ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
1453 
1454 #ifdef RE_DIAG
1455 	/*
1456 	 * Perform hardware diagnostic on the original RTL8169.
1457 	 * Some 32-bit cards were incorrectly wired and would
1458 	 * malfunction if plugged into a 64-bit slot.
1459 	 */
1460 
1461 	if (hwrev == RL_HWREV_8169) {
1462 		error = re_diag(sc);
1463 		if (error) {
1464 			device_printf(dev,
1465 		    	"attach aborted due to hardware diag failure\n");
1466 			ether_ifdetach(ifp);
1467 			goto fail;
1468 		}
1469 	}
1470 #endif
1471 
1472 	/* Hook interrupt last to avoid having to lock softc */
1473 	if ((sc->rl_flags & RL_FLAG_MSI) == 0)
1474 		error = bus_setup_intr(dev, sc->rl_irq[0],
1475 		    INTR_TYPE_NET | INTR_MPSAFE, re_intr, NULL, sc,
1476 		    &sc->rl_intrhand[0]);
1477 	else {
1478 		for (i = 0; i < RL_MSI_MESSAGES; i++) {
1479 			error = bus_setup_intr(dev, sc->rl_irq[i],
1480 			    INTR_TYPE_NET | INTR_MPSAFE, re_intr, NULL, sc,
1481 		    	    &sc->rl_intrhand[i]);
1482 			if (error != 0)
1483 				break;
1484 		}
1485 	}
1486 	if (error) {
1487 		device_printf(dev, "couldn't set up irq\n");
1488 		ether_ifdetach(ifp);
1489 	}
1490 
1491 fail:
1492 
1493 	if (error)
1494 		re_detach(dev);
1495 
1496 	return (error);
1497 }
1498 
1499 /*
1500  * Shutdown hardware and free up resources. This can be called any
1501  * time after the mutex has been initialized. It is called in both
1502  * the error case in attach and the normal detach case so it needs
1503  * to be careful about only freeing resources that have actually been
1504  * allocated.
1505  */
1506 static int
1507 re_detach(device_t dev)
1508 {
1509 	struct rl_softc		*sc;
1510 	struct ifnet		*ifp;
1511 	int			i, rid;
1512 
1513 	sc = device_get_softc(dev);
1514 	ifp = sc->rl_ifp;
1515 	KASSERT(mtx_initialized(&sc->rl_mtx), ("re mutex not initialized"));
1516 
1517 	/* These should only be active if attach succeeded */
1518 	if (device_is_attached(dev)) {
1519 #ifdef DEVICE_POLLING
1520 		if (ifp->if_capenable & IFCAP_POLLING)
1521 			ether_poll_deregister(ifp);
1522 #endif
1523 		RL_LOCK(sc);
1524 #if 0
1525 		sc->suspended = 1;
1526 #endif
1527 		re_stop(sc);
1528 		RL_UNLOCK(sc);
1529 		callout_drain(&sc->rl_stat_callout);
1530 		taskqueue_drain(taskqueue_fast, &sc->rl_inttask);
1531 		taskqueue_drain(taskqueue_fast, &sc->rl_txtask);
1532 		/*
1533 		 * Force off the IFF_UP flag here, in case someone
1534 		 * still had a BPF descriptor attached to this
1535 		 * interface. If they do, ether_ifdetach() will cause
1536 		 * the BPF code to try and clear the promisc mode
1537 		 * flag, which will bubble down to re_ioctl(),
1538 		 * which will try to call re_init() again. This will
1539 		 * turn the NIC back on and restart the MII ticker,
1540 		 * which will panic the system when the kernel tries
1541 		 * to invoke the re_tick() function that isn't there
1542 		 * anymore.
1543 		 */
1544 		ifp->if_flags &= ~IFF_UP;
1545 		ether_ifdetach(ifp);
1546 	}
1547 	if (sc->rl_miibus)
1548 		device_delete_child(dev, sc->rl_miibus);
1549 	bus_generic_detach(dev);
1550 
1551 	/*
1552 	 * The rest is resource deallocation, so we should already be
1553 	 * stopped here.
1554 	 */
1555 
1556 	for (i = 0; i < RL_MSI_MESSAGES; i++) {
1557 		if (sc->rl_intrhand[i] != NULL) {
1558 			bus_teardown_intr(dev, sc->rl_irq[i],
1559 			    sc->rl_intrhand[i]);
1560 			sc->rl_intrhand[i] = NULL;
1561 		}
1562 	}
1563 	if (ifp != NULL)
1564 		if_free(ifp);
1565 	if ((sc->rl_flags & RL_FLAG_MSI) == 0) {
1566 		if (sc->rl_irq[0] != NULL) {
1567 			bus_release_resource(dev, SYS_RES_IRQ, 0,
1568 			    sc->rl_irq[0]);
1569 			sc->rl_irq[0] = NULL;
1570 		}
1571 	} else {
1572 		for (i = 0, rid = 1; i < RL_MSI_MESSAGES; i++, rid++) {
1573 			if (sc->rl_irq[i] != NULL) {
1574 				bus_release_resource(dev, SYS_RES_IRQ, rid,
1575 				    sc->rl_irq[i]);
1576 				sc->rl_irq[i] = NULL;
1577 			}
1578 		}
1579 		pci_release_msi(dev);
1580 	}
1581 	if (sc->rl_res)
1582 		bus_release_resource(dev, sc->rl_res_type, sc->rl_res_id,
1583 		    sc->rl_res);
1584 
1585 	/* Unload and free the RX DMA ring memory and map */
1586 
1587 	if (sc->rl_ldata.rl_rx_list_tag) {
1588 		bus_dmamap_unload(sc->rl_ldata.rl_rx_list_tag,
1589 		    sc->rl_ldata.rl_rx_list_map);
1590 		bus_dmamem_free(sc->rl_ldata.rl_rx_list_tag,
1591 		    sc->rl_ldata.rl_rx_list,
1592 		    sc->rl_ldata.rl_rx_list_map);
1593 		bus_dma_tag_destroy(sc->rl_ldata.rl_rx_list_tag);
1594 	}
1595 
1596 	/* Unload and free the TX DMA ring memory and map */
1597 
1598 	if (sc->rl_ldata.rl_tx_list_tag) {
1599 		bus_dmamap_unload(sc->rl_ldata.rl_tx_list_tag,
1600 		    sc->rl_ldata.rl_tx_list_map);
1601 		bus_dmamem_free(sc->rl_ldata.rl_tx_list_tag,
1602 		    sc->rl_ldata.rl_tx_list,
1603 		    sc->rl_ldata.rl_tx_list_map);
1604 		bus_dma_tag_destroy(sc->rl_ldata.rl_tx_list_tag);
1605 	}
1606 
1607 	/* Destroy all the RX and TX buffer maps */
1608 
1609 	if (sc->rl_ldata.rl_tx_mtag) {
1610 		for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++)
1611 			bus_dmamap_destroy(sc->rl_ldata.rl_tx_mtag,
1612 			    sc->rl_ldata.rl_tx_desc[i].tx_dmamap);
1613 		bus_dma_tag_destroy(sc->rl_ldata.rl_tx_mtag);
1614 	}
1615 	if (sc->rl_ldata.rl_rx_mtag) {
1616 		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++)
1617 			bus_dmamap_destroy(sc->rl_ldata.rl_rx_mtag,
1618 			    sc->rl_ldata.rl_rx_desc[i].rx_dmamap);
1619 		if (sc->rl_ldata.rl_rx_sparemap)
1620 			bus_dmamap_destroy(sc->rl_ldata.rl_rx_mtag,
1621 			    sc->rl_ldata.rl_rx_sparemap);
1622 		bus_dma_tag_destroy(sc->rl_ldata.rl_rx_mtag);
1623 	}
1624 
1625 	/* Unload and free the stats buffer and map */
1626 
1627 	if (sc->rl_ldata.rl_stag) {
1628 		bus_dmamap_unload(sc->rl_ldata.rl_stag,
1629 		    sc->rl_ldata.rl_rx_list_map);
1630 		bus_dmamem_free(sc->rl_ldata.rl_stag,
1631 		    sc->rl_ldata.rl_stats,
1632 		    sc->rl_ldata.rl_smap);
1633 		bus_dma_tag_destroy(sc->rl_ldata.rl_stag);
1634 	}
1635 
1636 	if (sc->rl_parent_tag)
1637 		bus_dma_tag_destroy(sc->rl_parent_tag);
1638 
1639 	mtx_destroy(&sc->rl_mtx);
1640 
1641 	return (0);
1642 }
1643 
1644 static __inline void
1645 re_discard_rxbuf(struct rl_softc *sc, int idx)
1646 {
1647 	struct rl_desc		*desc;
1648 	struct rl_rxdesc	*rxd;
1649 	uint32_t		cmdstat;
1650 
1651 	rxd = &sc->rl_ldata.rl_rx_desc[idx];
1652 	desc = &sc->rl_ldata.rl_rx_list[idx];
1653 	desc->rl_vlanctl = 0;
1654 	cmdstat = rxd->rx_size;
1655 	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
1656 		cmdstat |= RL_RDESC_CMD_EOR;
1657 	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
1658 }
1659 
1660 static int
1661 re_newbuf(struct rl_softc *sc, int idx)
1662 {
1663 	struct mbuf		*m;
1664 	struct rl_rxdesc	*rxd;
1665 	bus_dma_segment_t	segs[1];
1666 	bus_dmamap_t		map;
1667 	struct rl_desc		*desc;
1668 	uint32_t		cmdstat;
1669 	int			error, nsegs;
1670 
1671 	m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
1672 	if (m == NULL)
1673 		return (ENOBUFS);
1674 
1675 	m->m_len = m->m_pkthdr.len = MCLBYTES;
1676 #ifdef RE_FIXUP_RX
1677 	/*
1678 	 * This is part of an evil trick to deal with non-x86 platforms.
1679 	 * The RealTek chip requires RX buffers to be aligned on 64-bit
1680 	 * boundaries, but that will hose non-x86 machines. To get around
1681 	 * this, we leave some empty space at the start of each buffer
1682 	 * and for non-x86 hosts, we copy the buffer back six bytes
1683 	 * to achieve word alignment. This is slightly more efficient
1684 	 * than allocating a new buffer, copying the contents, and
1685 	 * discarding the old buffer.
1686 	 */
1687 	m_adj(m, RE_ETHER_ALIGN);
1688 #endif
1689 	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_rx_mtag,
1690 	    sc->rl_ldata.rl_rx_sparemap, m, segs, &nsegs, BUS_DMA_NOWAIT);
1691 	if (error != 0) {
1692 		m_freem(m);
1693 		return (ENOBUFS);
1694 	}
1695 	KASSERT(nsegs == 1, ("%s: %d segment returned!", __func__, nsegs));
1696 
1697 	rxd = &sc->rl_ldata.rl_rx_desc[idx];
1698 	if (rxd->rx_m != NULL) {
1699 		bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap,
1700 		    BUS_DMASYNC_POSTREAD);
1701 		bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap);
1702 	}
1703 
1704 	rxd->rx_m = m;
1705 	map = rxd->rx_dmamap;
1706 	rxd->rx_dmamap = sc->rl_ldata.rl_rx_sparemap;
1707 	rxd->rx_size = segs[0].ds_len;
1708 	sc->rl_ldata.rl_rx_sparemap = map;
1709 	bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap,
1710 	    BUS_DMASYNC_PREREAD);
1711 
1712 	desc = &sc->rl_ldata.rl_rx_list[idx];
1713 	desc->rl_vlanctl = 0;
1714 	desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[0].ds_addr));
1715 	desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[0].ds_addr));
1716 	cmdstat = segs[0].ds_len;
1717 	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
1718 		cmdstat |= RL_RDESC_CMD_EOR;
1719 	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
1720 
1721 	return (0);
1722 }
1723 
1724 #ifdef RE_FIXUP_RX
1725 static __inline void
1726 re_fixup_rx(struct mbuf *m)
1727 {
1728 	int                     i;
1729 	uint16_t                *src, *dst;
1730 
1731 	src = mtod(m, uint16_t *);
1732 	dst = src - (RE_ETHER_ALIGN - ETHER_ALIGN) / sizeof *src;
1733 
1734 	for (i = 0; i < (m->m_len / sizeof(uint16_t) + 1); i++)
1735 		*dst++ = *src++;
1736 
1737 	m->m_data -= RE_ETHER_ALIGN - ETHER_ALIGN;
1738 }
1739 #endif
1740 
1741 static int
1742 re_tx_list_init(struct rl_softc *sc)
1743 {
1744 	struct rl_desc		*desc;
1745 	int			i;
1746 
1747 	RL_LOCK_ASSERT(sc);
1748 
1749 	bzero(sc->rl_ldata.rl_tx_list,
1750 	    sc->rl_ldata.rl_tx_desc_cnt * sizeof(struct rl_desc));
1751 	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++)
1752 		sc->rl_ldata.rl_tx_desc[i].tx_m = NULL;
1753 	/* Set EOR. */
1754 	desc = &sc->rl_ldata.rl_tx_list[sc->rl_ldata.rl_tx_desc_cnt - 1];
1755 	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_EOR);
1756 
1757 	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
1758 	    sc->rl_ldata.rl_tx_list_map,
1759 	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1760 
1761 	sc->rl_ldata.rl_tx_prodidx = 0;
1762 	sc->rl_ldata.rl_tx_considx = 0;
1763 	sc->rl_ldata.rl_tx_free = sc->rl_ldata.rl_tx_desc_cnt;
1764 
1765 	return (0);
1766 }
1767 
1768 static int
1769 re_rx_list_init(struct rl_softc *sc)
1770 {
1771 	int			error, i;
1772 
1773 	bzero(sc->rl_ldata.rl_rx_list,
1774 	    sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc));
1775 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1776 		sc->rl_ldata.rl_rx_desc[i].rx_m = NULL;
1777 		if ((error = re_newbuf(sc, i)) != 0)
1778 			return (error);
1779 	}
1780 
1781 	/* Flush the RX descriptors */
1782 
1783 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1784 	    sc->rl_ldata.rl_rx_list_map,
1785 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
1786 
1787 	sc->rl_ldata.rl_rx_prodidx = 0;
1788 	sc->rl_head = sc->rl_tail = NULL;
1789 
1790 	return (0);
1791 }
1792 
1793 /*
1794  * RX handler for C+ and 8169. For the gigE chips, we support
1795  * the reception of jumbo frames that have been fragmented
1796  * across multiple 2K mbuf cluster buffers.
1797  */
1798 static int
1799 re_rxeof(struct rl_softc *sc, int *rx_npktsp)
1800 {
1801 	struct mbuf		*m;
1802 	struct ifnet		*ifp;
1803 	int			i, total_len;
1804 	struct rl_desc		*cur_rx;
1805 	u_int32_t		rxstat, rxvlan;
1806 	int			maxpkt = 16, rx_npkts = 0;
1807 
1808 	RL_LOCK_ASSERT(sc);
1809 
1810 	ifp = sc->rl_ifp;
1811 
1812 	/* Invalidate the descriptor memory */
1813 
1814 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1815 	    sc->rl_ldata.rl_rx_list_map,
1816 	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
1817 
1818 	for (i = sc->rl_ldata.rl_rx_prodidx; maxpkt > 0;
1819 	    i = RL_RX_DESC_NXT(sc, i)) {
1820 		cur_rx = &sc->rl_ldata.rl_rx_list[i];
1821 		rxstat = le32toh(cur_rx->rl_cmdstat);
1822 		if ((rxstat & RL_RDESC_STAT_OWN) != 0)
1823 			break;
1824 		total_len = rxstat & sc->rl_rxlenmask;
1825 		rxvlan = le32toh(cur_rx->rl_vlanctl);
1826 		m = sc->rl_ldata.rl_rx_desc[i].rx_m;
1827 
1828 		if (!(rxstat & RL_RDESC_STAT_EOF)) {
1829 			if (re_newbuf(sc, i) != 0) {
1830 				/*
1831 				 * If this is part of a multi-fragment packet,
1832 				 * discard all the pieces.
1833 				 */
1834 				if (sc->rl_head != NULL) {
1835 					m_freem(sc->rl_head);
1836 					sc->rl_head = sc->rl_tail = NULL;
1837 				}
1838 				re_discard_rxbuf(sc, i);
1839 				continue;
1840 			}
1841 			m->m_len = RE_RX_DESC_BUFLEN;
1842 			if (sc->rl_head == NULL)
1843 				sc->rl_head = sc->rl_tail = m;
1844 			else {
1845 				m->m_flags &= ~M_PKTHDR;
1846 				sc->rl_tail->m_next = m;
1847 				sc->rl_tail = m;
1848 			}
1849 			continue;
1850 		}
1851 
1852 		/*
1853 		 * NOTE: for the 8139C+, the frame length field
1854 		 * is always 12 bits in size, but for the gigE chips,
1855 		 * it is 13 bits (since the max RX frame length is 16K).
1856 		 * Unfortunately, all 32 bits in the status word
1857 		 * were already used, so to make room for the extra
1858 		 * length bit, RealTek took out the 'frame alignment
1859 		 * error' bit and shifted the other status bits
1860 		 * over one slot. The OWN, EOR, FS and LS bits are
1861 		 * still in the same places. We have already extracted
1862 		 * the frame length and checked the OWN bit, so rather
1863 		 * than using an alternate bit mapping, we shift the
1864 		 * status bits one space to the right so we can evaluate
1865 		 * them using the 8169 status as though it was in the
1866 		 * same format as that of the 8139C+.
1867 		 */
1868 		if (sc->rl_type == RL_8169)
1869 			rxstat >>= 1;
1870 
1871 		/*
1872 		 * if total_len > 2^13-1, both _RXERRSUM and _GIANT will be
1873 		 * set, but if CRC is clear, it will still be a valid frame.
1874 		 */
1875 		if (rxstat & RL_RDESC_STAT_RXERRSUM && !(total_len > 8191 &&
1876 		    (rxstat & RL_RDESC_STAT_ERRS) == RL_RDESC_STAT_GIANT)) {
1877 			ifp->if_ierrors++;
1878 			/*
1879 			 * If this is part of a multi-fragment packet,
1880 			 * discard all the pieces.
1881 			 */
1882 			if (sc->rl_head != NULL) {
1883 				m_freem(sc->rl_head);
1884 				sc->rl_head = sc->rl_tail = NULL;
1885 			}
1886 			re_discard_rxbuf(sc, i);
1887 			continue;
1888 		}
1889 
1890 		/*
1891 		 * If allocating a replacement mbuf fails,
1892 		 * reload the current one.
1893 		 */
1894 
1895 		if (re_newbuf(sc, i) != 0) {
1896 			ifp->if_iqdrops++;
1897 			if (sc->rl_head != NULL) {
1898 				m_freem(sc->rl_head);
1899 				sc->rl_head = sc->rl_tail = NULL;
1900 			}
1901 			re_discard_rxbuf(sc, i);
1902 			continue;
1903 		}
1904 
1905 		if (sc->rl_head != NULL) {
1906 			m->m_len = total_len % RE_RX_DESC_BUFLEN;
1907 			if (m->m_len == 0)
1908 				m->m_len = RE_RX_DESC_BUFLEN;
1909 			/*
1910 			 * Special case: if there's 4 bytes or less
1911 			 * in this buffer, the mbuf can be discarded:
1912 			 * the last 4 bytes is the CRC, which we don't
1913 			 * care about anyway.
1914 			 */
1915 			if (m->m_len <= ETHER_CRC_LEN) {
1916 				sc->rl_tail->m_len -=
1917 				    (ETHER_CRC_LEN - m->m_len);
1918 				m_freem(m);
1919 			} else {
1920 				m->m_len -= ETHER_CRC_LEN;
1921 				m->m_flags &= ~M_PKTHDR;
1922 				sc->rl_tail->m_next = m;
1923 			}
1924 			m = sc->rl_head;
1925 			sc->rl_head = sc->rl_tail = NULL;
1926 			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
1927 		} else
1928 			m->m_pkthdr.len = m->m_len =
1929 			    (total_len - ETHER_CRC_LEN);
1930 
1931 #ifdef RE_FIXUP_RX
1932 		re_fixup_rx(m);
1933 #endif
1934 		ifp->if_ipackets++;
1935 		m->m_pkthdr.rcvif = ifp;
1936 
1937 		/* Do RX checksumming if enabled */
1938 
1939 		if (ifp->if_capenable & IFCAP_RXCSUM) {
1940 			if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
1941 				/* Check IP header checksum */
1942 				if (rxstat & RL_RDESC_STAT_PROTOID)
1943 					m->m_pkthdr.csum_flags |=
1944 					    CSUM_IP_CHECKED;
1945 				if (!(rxstat & RL_RDESC_STAT_IPSUMBAD))
1946 					m->m_pkthdr.csum_flags |=
1947 					    CSUM_IP_VALID;
1948 
1949 				/* Check TCP/UDP checksum */
1950 				if ((RL_TCPPKT(rxstat) &&
1951 				    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
1952 				    (RL_UDPPKT(rxstat) &&
1953 				     !(rxstat & RL_RDESC_STAT_UDPSUMBAD))) {
1954 					m->m_pkthdr.csum_flags |=
1955 						CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
1956 					m->m_pkthdr.csum_data = 0xffff;
1957 				}
1958 			} else {
1959 				/*
1960 				 * RTL8168C/RTL816CP/RTL8111C/RTL8111CP
1961 				 */
1962 				if ((rxstat & RL_RDESC_STAT_PROTOID) &&
1963 				    (rxvlan & RL_RDESC_IPV4))
1964 					m->m_pkthdr.csum_flags |=
1965 					    CSUM_IP_CHECKED;
1966 				if (!(rxstat & RL_RDESC_STAT_IPSUMBAD) &&
1967 				    (rxvlan & RL_RDESC_IPV4))
1968 					m->m_pkthdr.csum_flags |=
1969 					    CSUM_IP_VALID;
1970 				if (((rxstat & RL_RDESC_STAT_TCP) &&
1971 				    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
1972 				    ((rxstat & RL_RDESC_STAT_UDP) &&
1973 				    !(rxstat & RL_RDESC_STAT_UDPSUMBAD))) {
1974 					m->m_pkthdr.csum_flags |=
1975 						CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
1976 					m->m_pkthdr.csum_data = 0xffff;
1977 				}
1978 			}
1979 		}
1980 		maxpkt--;
1981 		if (rxvlan & RL_RDESC_VLANCTL_TAG) {
1982 			m->m_pkthdr.ether_vtag =
1983 			    bswap16((rxvlan & RL_RDESC_VLANCTL_DATA));
1984 			m->m_flags |= M_VLANTAG;
1985 		}
1986 		RL_UNLOCK(sc);
1987 		(*ifp->if_input)(ifp, m);
1988 		RL_LOCK(sc);
1989 		rx_npkts++;
1990 	}
1991 
1992 	/* Flush the RX DMA ring */
1993 
1994 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1995 	    sc->rl_ldata.rl_rx_list_map,
1996 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
1997 
1998 	sc->rl_ldata.rl_rx_prodidx = i;
1999 
2000 	if (rx_npktsp != NULL)
2001 		*rx_npktsp = rx_npkts;
2002 	if (maxpkt)
2003 		return(EAGAIN);
2004 
2005 	return(0);
2006 }
2007 
2008 static void
2009 re_txeof(struct rl_softc *sc)
2010 {
2011 	struct ifnet		*ifp;
2012 	struct rl_txdesc	*txd;
2013 	u_int32_t		txstat;
2014 	int			cons;
2015 
2016 	cons = sc->rl_ldata.rl_tx_considx;
2017 	if (cons == sc->rl_ldata.rl_tx_prodidx)
2018 		return;
2019 
2020 	ifp = sc->rl_ifp;
2021 	/* Invalidate the TX descriptor list */
2022 	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2023 	    sc->rl_ldata.rl_tx_list_map,
2024 	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2025 
2026 	for (; cons != sc->rl_ldata.rl_tx_prodidx;
2027 	    cons = RL_TX_DESC_NXT(sc, cons)) {
2028 		txstat = le32toh(sc->rl_ldata.rl_tx_list[cons].rl_cmdstat);
2029 		if (txstat & RL_TDESC_STAT_OWN)
2030 			break;
2031 		/*
2032 		 * We only stash mbufs in the last descriptor
2033 		 * in a fragment chain, which also happens to
2034 		 * be the only place where the TX status bits
2035 		 * are valid.
2036 		 */
2037 		if (txstat & RL_TDESC_CMD_EOF) {
2038 			txd = &sc->rl_ldata.rl_tx_desc[cons];
2039 			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
2040 			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
2041 			bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag,
2042 			    txd->tx_dmamap);
2043 			KASSERT(txd->tx_m != NULL,
2044 			    ("%s: freeing NULL mbufs!", __func__));
2045 			m_freem(txd->tx_m);
2046 			txd->tx_m = NULL;
2047 			if (txstat & (RL_TDESC_STAT_EXCESSCOL|
2048 			    RL_TDESC_STAT_COLCNT))
2049 				ifp->if_collisions++;
2050 			if (txstat & RL_TDESC_STAT_TXERRSUM)
2051 				ifp->if_oerrors++;
2052 			else
2053 				ifp->if_opackets++;
2054 		}
2055 		sc->rl_ldata.rl_tx_free++;
2056 		ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
2057 	}
2058 	sc->rl_ldata.rl_tx_considx = cons;
2059 
2060 	/* No changes made to the TX ring, so no flush needed */
2061 
2062 	if (sc->rl_ldata.rl_tx_free != sc->rl_ldata.rl_tx_desc_cnt) {
2063 #ifdef RE_TX_MODERATION
2064 		/*
2065 		 * If not all descriptors have been reaped yet, reload
2066 		 * the timer so that we will eventually get another
2067 		 * interrupt that will cause us to re-enter this routine.
2068 		 * This is done in case the transmitter has gone idle.
2069 		 */
2070 		CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2071 #endif
2072 	} else
2073 		sc->rl_watchdog_timer = 0;
2074 }
2075 
2076 static void
2077 re_tick(void *xsc)
2078 {
2079 	struct rl_softc		*sc;
2080 	struct mii_data		*mii;
2081 
2082 	sc = xsc;
2083 
2084 	RL_LOCK_ASSERT(sc);
2085 
2086 	mii = device_get_softc(sc->rl_miibus);
2087 	mii_tick(mii);
2088 	if ((sc->rl_flags & RL_FLAG_LINK) == 0)
2089 		re_miibus_statchg(sc->rl_dev);
2090 	/*
2091 	 * Reclaim transmitted frames here. Technically it is not
2092 	 * necessary to do here but it ensures periodic reclamation
2093 	 * regardless of Tx completion interrupt which seems to be
2094 	 * lost on PCIe based controllers under certain situations.
2095 	 */
2096 	re_txeof(sc);
2097 	re_watchdog(sc);
2098 	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
2099 }
2100 
2101 #ifdef DEVICE_POLLING
2102 static int
2103 re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
2104 {
2105 	struct rl_softc *sc = ifp->if_softc;
2106 	int rx_npkts = 0;
2107 
2108 	RL_LOCK(sc);
2109 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
2110 		rx_npkts = re_poll_locked(ifp, cmd, count);
2111 	RL_UNLOCK(sc);
2112 	return (rx_npkts);
2113 }
2114 
2115 static int
2116 re_poll_locked(struct ifnet *ifp, enum poll_cmd cmd, int count)
2117 {
2118 	struct rl_softc *sc = ifp->if_softc;
2119 	int rx_npkts;
2120 
2121 	RL_LOCK_ASSERT(sc);
2122 
2123 	sc->rxcycles = count;
2124 	re_rxeof(sc, &rx_npkts);
2125 	re_txeof(sc);
2126 
2127 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2128 		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_txtask);
2129 
2130 	if (cmd == POLL_AND_CHECK_STATUS) { /* also check status register */
2131 		u_int16_t       status;
2132 
2133 		status = CSR_READ_2(sc, RL_ISR);
2134 		if (status == 0xffff)
2135 			return (rx_npkts);
2136 		if (status)
2137 			CSR_WRITE_2(sc, RL_ISR, status);
2138 		if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
2139 		    (sc->rl_flags & RL_FLAG_PCIE))
2140 			CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2141 
2142 		/*
2143 		 * XXX check behaviour on receiver stalls.
2144 		 */
2145 
2146 		if (status & RL_ISR_SYSTEM_ERR)
2147 			re_init_locked(sc);
2148 	}
2149 	return (rx_npkts);
2150 }
2151 #endif /* DEVICE_POLLING */
2152 
2153 static int
2154 re_intr(void *arg)
2155 {
2156 	struct rl_softc		*sc;
2157 	uint16_t		status;
2158 
2159 	sc = arg;
2160 
2161 	status = CSR_READ_2(sc, RL_ISR);
2162 	if (status == 0xFFFF || (status & RL_INTRS_CPLUS) == 0)
2163                 return (FILTER_STRAY);
2164 	CSR_WRITE_2(sc, RL_IMR, 0);
2165 
2166 	taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_inttask);
2167 
2168 	return (FILTER_HANDLED);
2169 }
2170 
2171 static void
2172 re_int_task(void *arg, int npending)
2173 {
2174 	struct rl_softc		*sc;
2175 	struct ifnet		*ifp;
2176 	u_int16_t		status;
2177 	int			rval = 0;
2178 
2179 	sc = arg;
2180 	ifp = sc->rl_ifp;
2181 
2182 	RL_LOCK(sc);
2183 
2184 	status = CSR_READ_2(sc, RL_ISR);
2185         CSR_WRITE_2(sc, RL_ISR, status);
2186 
2187 	if (sc->suspended ||
2188 	    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2189 		RL_UNLOCK(sc);
2190 		return;
2191 	}
2192 
2193 #ifdef DEVICE_POLLING
2194 	if  (ifp->if_capenable & IFCAP_POLLING) {
2195 		RL_UNLOCK(sc);
2196 		return;
2197 	}
2198 #endif
2199 
2200 	if (status & (RL_ISR_RX_OK|RL_ISR_RX_ERR|RL_ISR_FIFO_OFLOW))
2201 		rval = re_rxeof(sc, NULL);
2202 
2203 	/*
2204 	 * Some chips will ignore a second TX request issued
2205 	 * while an existing transmission is in progress. If
2206 	 * the transmitter goes idle but there are still
2207 	 * packets waiting to be sent, we need to restart the
2208 	 * channel here to flush them out. This only seems to
2209 	 * be required with the PCIe devices.
2210 	 */
2211 	if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
2212 	    (sc->rl_flags & RL_FLAG_PCIE))
2213 		CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2214 	if (status & (
2215 #ifdef RE_TX_MODERATION
2216 	    RL_ISR_TIMEOUT_EXPIRED|
2217 #else
2218 	    RL_ISR_TX_OK|
2219 #endif
2220 	    RL_ISR_TX_ERR|RL_ISR_TX_DESC_UNAVAIL))
2221 		re_txeof(sc);
2222 
2223 	if (status & RL_ISR_SYSTEM_ERR)
2224 		re_init_locked(sc);
2225 
2226 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2227 		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_txtask);
2228 
2229 	RL_UNLOCK(sc);
2230 
2231         if ((CSR_READ_2(sc, RL_ISR) & RL_INTRS_CPLUS) || rval) {
2232 		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_inttask);
2233 		return;
2234 	}
2235 
2236 	CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2237 }
2238 
2239 static int
2240 re_encap(struct rl_softc *sc, struct mbuf **m_head)
2241 {
2242 	struct rl_txdesc	*txd, *txd_last;
2243 	bus_dma_segment_t	segs[RL_NTXSEGS];
2244 	bus_dmamap_t		map;
2245 	struct mbuf		*m_new;
2246 	struct rl_desc		*desc;
2247 	int			nsegs, prod;
2248 	int			i, error, ei, si;
2249 	int			padlen;
2250 	uint32_t		cmdstat, csum_flags, vlanctl;
2251 
2252 	RL_LOCK_ASSERT(sc);
2253 	M_ASSERTPKTHDR((*m_head));
2254 
2255 	/*
2256 	 * With some of the RealTek chips, using the checksum offload
2257 	 * support in conjunction with the autopadding feature results
2258 	 * in the transmission of corrupt frames. For example, if we
2259 	 * need to send a really small IP fragment that's less than 60
2260 	 * bytes in size, and IP header checksumming is enabled, the
2261 	 * resulting ethernet frame that appears on the wire will
2262 	 * have garbled payload. To work around this, if TX IP checksum
2263 	 * offload is enabled, we always manually pad short frames out
2264 	 * to the minimum ethernet frame size.
2265 	 */
2266 	if ((sc->rl_flags & RL_FLAG_AUTOPAD) == 0 &&
2267 	    (*m_head)->m_pkthdr.len < RL_IP4CSUMTX_PADLEN &&
2268 	    ((*m_head)->m_pkthdr.csum_flags & CSUM_IP) != 0) {
2269 		padlen = RL_MIN_FRAMELEN - (*m_head)->m_pkthdr.len;
2270 		if (M_WRITABLE(*m_head) == 0) {
2271 			/* Get a writable copy. */
2272 			m_new = m_dup(*m_head, M_DONTWAIT);
2273 			m_freem(*m_head);
2274 			if (m_new == NULL) {
2275 				*m_head = NULL;
2276 				return (ENOBUFS);
2277 			}
2278 			*m_head = m_new;
2279 		}
2280 		if ((*m_head)->m_next != NULL ||
2281 		    M_TRAILINGSPACE(*m_head) < padlen) {
2282 			m_new = m_defrag(*m_head, M_DONTWAIT);
2283 			if (m_new == NULL) {
2284 				m_freem(*m_head);
2285 				*m_head = NULL;
2286 				return (ENOBUFS);
2287 			}
2288 		} else
2289 			m_new = *m_head;
2290 
2291 		/*
2292 		 * Manually pad short frames, and zero the pad space
2293 		 * to avoid leaking data.
2294 		 */
2295 		bzero(mtod(m_new, char *) + m_new->m_pkthdr.len, padlen);
2296 		m_new->m_pkthdr.len += padlen;
2297 		m_new->m_len = m_new->m_pkthdr.len;
2298 		*m_head = m_new;
2299 	}
2300 
2301 	prod = sc->rl_ldata.rl_tx_prodidx;
2302 	txd = &sc->rl_ldata.rl_tx_desc[prod];
2303 	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap,
2304 	    *m_head, segs, &nsegs, BUS_DMA_NOWAIT);
2305 	if (error == EFBIG) {
2306 		m_new = m_collapse(*m_head, M_DONTWAIT, RL_NTXSEGS);
2307 		if (m_new == NULL) {
2308 			m_freem(*m_head);
2309 			*m_head = NULL;
2310 			return (ENOBUFS);
2311 		}
2312 		*m_head = m_new;
2313 		error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_tx_mtag,
2314 		    txd->tx_dmamap, *m_head, segs, &nsegs, BUS_DMA_NOWAIT);
2315 		if (error != 0) {
2316 			m_freem(*m_head);
2317 			*m_head = NULL;
2318 			return (error);
2319 		}
2320 	} else if (error != 0)
2321 		return (error);
2322 	if (nsegs == 0) {
2323 		m_freem(*m_head);
2324 		*m_head = NULL;
2325 		return (EIO);
2326 	}
2327 
2328 	/* Check for number of available descriptors. */
2329 	if (sc->rl_ldata.rl_tx_free - nsegs <= 1) {
2330 		bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap);
2331 		return (ENOBUFS);
2332 	}
2333 
2334 	bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap,
2335 	    BUS_DMASYNC_PREWRITE);
2336 
2337 	/*
2338 	 * Set up checksum offload. Note: checksum offload bits must
2339 	 * appear in all descriptors of a multi-descriptor transmit
2340 	 * attempt. This is according to testing done with an 8169
2341 	 * chip. This is a requirement.
2342 	 */
2343 	vlanctl = 0;
2344 	csum_flags = 0;
2345 	if (((*m_head)->m_pkthdr.csum_flags & CSUM_TSO) != 0)
2346 		csum_flags = RL_TDESC_CMD_LGSEND |
2347 		    ((uint32_t)(*m_head)->m_pkthdr.tso_segsz <<
2348 		    RL_TDESC_CMD_MSSVAL_SHIFT);
2349 	else {
2350 		/*
2351 		 * Unconditionally enable IP checksum if TCP or UDP
2352 		 * checksum is required. Otherwise, TCP/UDP checksum
2353 		 * does't make effects.
2354 		 */
2355 		if (((*m_head)->m_pkthdr.csum_flags & RE_CSUM_FEATURES) != 0) {
2356 			if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
2357 				csum_flags |= RL_TDESC_CMD_IPCSUM;
2358 				if (((*m_head)->m_pkthdr.csum_flags &
2359 				    CSUM_TCP) != 0)
2360 					csum_flags |= RL_TDESC_CMD_TCPCSUM;
2361 				if (((*m_head)->m_pkthdr.csum_flags &
2362 				    CSUM_UDP) != 0)
2363 					csum_flags |= RL_TDESC_CMD_UDPCSUM;
2364 			} else {
2365 				vlanctl |= RL_TDESC_CMD_IPCSUMV2;
2366 				if (((*m_head)->m_pkthdr.csum_flags &
2367 				    CSUM_TCP) != 0)
2368 					vlanctl |= RL_TDESC_CMD_TCPCSUMV2;
2369 				if (((*m_head)->m_pkthdr.csum_flags &
2370 				    CSUM_UDP) != 0)
2371 					vlanctl |= RL_TDESC_CMD_UDPCSUMV2;
2372 			}
2373 		}
2374 	}
2375 
2376 	/*
2377 	 * Set up hardware VLAN tagging. Note: vlan tag info must
2378 	 * appear in all descriptors of a multi-descriptor
2379 	 * transmission attempt.
2380 	 */
2381 	if ((*m_head)->m_flags & M_VLANTAG)
2382 		vlanctl |= bswap16((*m_head)->m_pkthdr.ether_vtag) |
2383 		    RL_TDESC_VLANCTL_TAG;
2384 
2385 	si = prod;
2386 	for (i = 0; i < nsegs; i++, prod = RL_TX_DESC_NXT(sc, prod)) {
2387 		desc = &sc->rl_ldata.rl_tx_list[prod];
2388 		desc->rl_vlanctl = htole32(vlanctl);
2389 		desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[i].ds_addr));
2390 		desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[i].ds_addr));
2391 		cmdstat = segs[i].ds_len;
2392 		if (i != 0)
2393 			cmdstat |= RL_TDESC_CMD_OWN;
2394 		if (prod == sc->rl_ldata.rl_tx_desc_cnt - 1)
2395 			cmdstat |= RL_TDESC_CMD_EOR;
2396 		desc->rl_cmdstat = htole32(cmdstat | csum_flags);
2397 		sc->rl_ldata.rl_tx_free--;
2398 	}
2399 	/* Update producer index. */
2400 	sc->rl_ldata.rl_tx_prodidx = prod;
2401 
2402 	/* Set EOF on the last descriptor. */
2403 	ei = RL_TX_DESC_PRV(sc, prod);
2404 	desc = &sc->rl_ldata.rl_tx_list[ei];
2405 	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_EOF);
2406 
2407 	desc = &sc->rl_ldata.rl_tx_list[si];
2408 	/* Set SOF and transfer ownership of packet to the chip. */
2409 	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_OWN | RL_TDESC_CMD_SOF);
2410 
2411 	/*
2412 	 * Insure that the map for this transmission
2413 	 * is placed at the array index of the last descriptor
2414 	 * in this chain.  (Swap last and first dmamaps.)
2415 	 */
2416 	txd_last = &sc->rl_ldata.rl_tx_desc[ei];
2417 	map = txd->tx_dmamap;
2418 	txd->tx_dmamap = txd_last->tx_dmamap;
2419 	txd_last->tx_dmamap = map;
2420 	txd_last->tx_m = *m_head;
2421 
2422 	return (0);
2423 }
2424 
2425 static void
2426 re_tx_task(void *arg, int npending)
2427 {
2428 	struct ifnet		*ifp;
2429 
2430 	ifp = arg;
2431 	re_start(ifp);
2432 }
2433 
2434 /*
2435  * Main transmit routine for C+ and gigE NICs.
2436  */
2437 static void
2438 re_start(struct ifnet *ifp)
2439 {
2440 	struct rl_softc		*sc;
2441 	struct mbuf		*m_head;
2442 	int			queued;
2443 
2444 	sc = ifp->if_softc;
2445 
2446 	RL_LOCK(sc);
2447 
2448 	if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
2449 	    IFF_DRV_RUNNING || (sc->rl_flags & RL_FLAG_LINK) == 0) {
2450 		RL_UNLOCK(sc);
2451 		return;
2452 	}
2453 
2454 	for (queued = 0; !IFQ_DRV_IS_EMPTY(&ifp->if_snd) &&
2455 	    sc->rl_ldata.rl_tx_free > 1;) {
2456 		IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
2457 		if (m_head == NULL)
2458 			break;
2459 
2460 		if (re_encap(sc, &m_head) != 0) {
2461 			if (m_head == NULL)
2462 				break;
2463 			IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
2464 			ifp->if_drv_flags |= IFF_DRV_OACTIVE;
2465 			break;
2466 		}
2467 
2468 		/*
2469 		 * If there's a BPF listener, bounce a copy of this frame
2470 		 * to him.
2471 		 */
2472 		ETHER_BPF_MTAP(ifp, m_head);
2473 
2474 		queued++;
2475 	}
2476 
2477 	if (queued == 0) {
2478 #ifdef RE_TX_MODERATION
2479 		if (sc->rl_ldata.rl_tx_free != sc->rl_ldata.rl_tx_desc_cnt)
2480 			CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2481 #endif
2482 		RL_UNLOCK(sc);
2483 		return;
2484 	}
2485 
2486 	/* Flush the TX descriptors */
2487 
2488 	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2489 	    sc->rl_ldata.rl_tx_list_map,
2490 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2491 
2492 	CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2493 
2494 #ifdef RE_TX_MODERATION
2495 	/*
2496 	 * Use the countdown timer for interrupt moderation.
2497 	 * 'TX done' interrupts are disabled. Instead, we reset the
2498 	 * countdown timer, which will begin counting until it hits
2499 	 * the value in the TIMERINT register, and then trigger an
2500 	 * interrupt. Each time we write to the TIMERCNT register,
2501 	 * the timer count is reset to 0.
2502 	 */
2503 	CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2504 #endif
2505 
2506 	/*
2507 	 * Set a timeout in case the chip goes out to lunch.
2508 	 */
2509 	sc->rl_watchdog_timer = 5;
2510 
2511 	RL_UNLOCK(sc);
2512 }
2513 
2514 static void
2515 re_init(void *xsc)
2516 {
2517 	struct rl_softc		*sc = xsc;
2518 
2519 	RL_LOCK(sc);
2520 	re_init_locked(sc);
2521 	RL_UNLOCK(sc);
2522 }
2523 
2524 static void
2525 re_init_locked(struct rl_softc *sc)
2526 {
2527 	struct ifnet		*ifp = sc->rl_ifp;
2528 	struct mii_data		*mii;
2529 	uint32_t		reg;
2530 	uint16_t		cfg;
2531 	union {
2532 		uint32_t align_dummy;
2533 		u_char eaddr[ETHER_ADDR_LEN];
2534         } eaddr;
2535 
2536 	RL_LOCK_ASSERT(sc);
2537 
2538 	mii = device_get_softc(sc->rl_miibus);
2539 
2540 	/*
2541 	 * Cancel pending I/O and free all RX/TX buffers.
2542 	 */
2543 	re_stop(sc);
2544 
2545 	/* Put controller into known state. */
2546 	re_reset(sc);
2547 
2548 	/*
2549 	 * Enable C+ RX and TX mode, as well as VLAN stripping and
2550 	 * RX checksum offload. We must configure the C+ register
2551 	 * before all others.
2552 	 */
2553 	cfg = RL_CPLUSCMD_PCI_MRW;
2554 	if ((ifp->if_capenable & IFCAP_RXCSUM) != 0)
2555 		cfg |= RL_CPLUSCMD_RXCSUM_ENB;
2556 	if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0)
2557 		cfg |= RL_CPLUSCMD_VLANSTRIP;
2558 	if ((sc->rl_flags & RL_FLAG_MACSTAT) != 0) {
2559 		cfg |= RL_CPLUSCMD_MACSTAT_DIS;
2560 		/* XXX magic. */
2561 		cfg |= 0x0001;
2562 	} else
2563 		cfg |= RL_CPLUSCMD_RXENB | RL_CPLUSCMD_TXENB;
2564 	CSR_WRITE_2(sc, RL_CPLUS_CMD, cfg);
2565 	if (sc->rl_hwrev == RL_HWREV_8169_8110SC ||
2566 	    sc->rl_hwrev == RL_HWREV_8169_8110SCE) {
2567 		reg = 0x000fff00;
2568 		if ((CSR_READ_1(sc, RL_CFG2) & RL_CFG2_PCI66MHZ) != 0)
2569 			reg |= 0x000000ff;
2570 		if (sc->rl_hwrev == RL_HWREV_8169_8110SCE)
2571 			reg |= 0x00f00000;
2572 		CSR_WRITE_4(sc, 0x7c, reg);
2573 		/* Disable interrupt mitigation. */
2574 		CSR_WRITE_2(sc, 0xe2, 0);
2575 	}
2576 	/*
2577 	 * Disable TSO if interface MTU size is greater than MSS
2578 	 * allowed in controller.
2579 	 */
2580 	if (ifp->if_mtu > RL_TSO_MTU && (ifp->if_capenable & IFCAP_TSO4) != 0) {
2581 		ifp->if_capenable &= ~IFCAP_TSO4;
2582 		ifp->if_hwassist &= ~CSUM_TSO;
2583 	}
2584 
2585 	/*
2586 	 * Init our MAC address.  Even though the chipset
2587 	 * documentation doesn't mention it, we need to enter "Config
2588 	 * register write enable" mode to modify the ID registers.
2589 	 */
2590 	/* Copy MAC address on stack to align. */
2591 	bcopy(IF_LLADDR(ifp), eaddr.eaddr, ETHER_ADDR_LEN);
2592 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG);
2593 	CSR_WRITE_4(sc, RL_IDR0,
2594 	    htole32(*(u_int32_t *)(&eaddr.eaddr[0])));
2595 	CSR_WRITE_4(sc, RL_IDR4,
2596 	    htole32(*(u_int32_t *)(&eaddr.eaddr[4])));
2597 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
2598 
2599 	/*
2600 	 * For C+ mode, initialize the RX descriptors and mbufs.
2601 	 */
2602 	re_rx_list_init(sc);
2603 	re_tx_list_init(sc);
2604 
2605 	/*
2606 	 * Load the addresses of the RX and TX lists into the chip.
2607 	 */
2608 
2609 	CSR_WRITE_4(sc, RL_RXLIST_ADDR_HI,
2610 	    RL_ADDR_HI(sc->rl_ldata.rl_rx_list_addr));
2611 	CSR_WRITE_4(sc, RL_RXLIST_ADDR_LO,
2612 	    RL_ADDR_LO(sc->rl_ldata.rl_rx_list_addr));
2613 
2614 	CSR_WRITE_4(sc, RL_TXLIST_ADDR_HI,
2615 	    RL_ADDR_HI(sc->rl_ldata.rl_tx_list_addr));
2616 	CSR_WRITE_4(sc, RL_TXLIST_ADDR_LO,
2617 	    RL_ADDR_LO(sc->rl_ldata.rl_tx_list_addr));
2618 
2619 	/*
2620 	 * Enable transmit and receive.
2621 	 */
2622 	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
2623 
2624 	/*
2625 	 * Set the initial TX configuration.
2626 	 */
2627 	if (sc->rl_testmode) {
2628 		if (sc->rl_type == RL_8169)
2629 			CSR_WRITE_4(sc, RL_TXCFG,
2630 			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON);
2631 		else
2632 			CSR_WRITE_4(sc, RL_TXCFG,
2633 			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON_CPLUS);
2634 	} else
2635 		CSR_WRITE_4(sc, RL_TXCFG, RL_TXCFG_CONFIG);
2636 
2637 	CSR_WRITE_1(sc, RL_EARLY_TX_THRESH, 16);
2638 
2639 	/*
2640 	 * Set the initial RX configuration.
2641 	 */
2642 	re_set_rxmode(sc);
2643 
2644 #ifdef DEVICE_POLLING
2645 	/*
2646 	 * Disable interrupts if we are polling.
2647 	 */
2648 	if (ifp->if_capenable & IFCAP_POLLING)
2649 		CSR_WRITE_2(sc, RL_IMR, 0);
2650 	else	/* otherwise ... */
2651 #endif
2652 
2653 	/*
2654 	 * Enable interrupts.
2655 	 */
2656 	if (sc->rl_testmode)
2657 		CSR_WRITE_2(sc, RL_IMR, 0);
2658 	else
2659 		CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2660 	CSR_WRITE_2(sc, RL_ISR, RL_INTRS_CPLUS);
2661 
2662 	/* Set initial TX threshold */
2663 	sc->rl_txthresh = RL_TX_THRESH_INIT;
2664 
2665 	/* Start RX/TX process. */
2666 	CSR_WRITE_4(sc, RL_MISSEDPKT, 0);
2667 #ifdef notdef
2668 	/* Enable receiver and transmitter. */
2669 	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
2670 #endif
2671 
2672 #ifdef RE_TX_MODERATION
2673 	/*
2674 	 * Initialize the timer interrupt register so that
2675 	 * a timer interrupt will be generated once the timer
2676 	 * reaches a certain number of ticks. The timer is
2677 	 * reloaded on each transmit. This gives us TX interrupt
2678 	 * moderation, which dramatically improves TX frame rate.
2679 	 */
2680 	if (sc->rl_type == RL_8169)
2681 		CSR_WRITE_4(sc, RL_TIMERINT_8169, 0x800);
2682 	else
2683 		CSR_WRITE_4(sc, RL_TIMERINT, 0x400);
2684 #endif
2685 
2686 	/*
2687 	 * For 8169 gigE NICs, set the max allowed RX packet
2688 	 * size so we can receive jumbo frames.
2689 	 */
2690 	if (sc->rl_type == RL_8169)
2691 		CSR_WRITE_2(sc, RL_MAXRXPKTLEN, 16383);
2692 
2693 	if (sc->rl_testmode)
2694 		return;
2695 
2696 	mii_mediachg(mii);
2697 
2698 	CSR_WRITE_1(sc, RL_CFG1, CSR_READ_1(sc, RL_CFG1) | RL_CFG1_DRVLOAD);
2699 
2700 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
2701 	ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
2702 
2703 	sc->rl_flags &= ~RL_FLAG_LINK;
2704 	sc->rl_watchdog_timer = 0;
2705 	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
2706 }
2707 
2708 /*
2709  * Set media options.
2710  */
2711 static int
2712 re_ifmedia_upd(struct ifnet *ifp)
2713 {
2714 	struct rl_softc		*sc;
2715 	struct mii_data		*mii;
2716 	int			error;
2717 
2718 	sc = ifp->if_softc;
2719 	mii = device_get_softc(sc->rl_miibus);
2720 	RL_LOCK(sc);
2721 	error = mii_mediachg(mii);
2722 	RL_UNLOCK(sc);
2723 
2724 	return (error);
2725 }
2726 
2727 /*
2728  * Report current media status.
2729  */
2730 static void
2731 re_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
2732 {
2733 	struct rl_softc		*sc;
2734 	struct mii_data		*mii;
2735 
2736 	sc = ifp->if_softc;
2737 	mii = device_get_softc(sc->rl_miibus);
2738 
2739 	RL_LOCK(sc);
2740 	mii_pollstat(mii);
2741 	RL_UNLOCK(sc);
2742 	ifmr->ifm_active = mii->mii_media_active;
2743 	ifmr->ifm_status = mii->mii_media_status;
2744 }
2745 
2746 static int
2747 re_ioctl(struct ifnet *ifp, u_long command, caddr_t data)
2748 {
2749 	struct rl_softc		*sc = ifp->if_softc;
2750 	struct ifreq		*ifr = (struct ifreq *) data;
2751 	struct mii_data		*mii;
2752 	int			error = 0;
2753 
2754 	switch (command) {
2755 	case SIOCSIFMTU:
2756 		if (ifr->ifr_mtu < ETHERMIN || ifr->ifr_mtu > RL_JUMBO_MTU) {
2757 			error = EINVAL;
2758 			break;
2759 		}
2760 		if ((sc->rl_flags & RL_FLAG_NOJUMBO) != 0 &&
2761 		    ifr->ifr_mtu > RL_MAX_FRAMELEN) {
2762 			error = EINVAL;
2763 			break;
2764 		}
2765 		RL_LOCK(sc);
2766 		if (ifp->if_mtu != ifr->ifr_mtu)
2767 			ifp->if_mtu = ifr->ifr_mtu;
2768 		if (ifp->if_mtu > RL_TSO_MTU &&
2769 		    (ifp->if_capenable & IFCAP_TSO4) != 0) {
2770 			ifp->if_capenable &= ~IFCAP_TSO4;
2771 			ifp->if_hwassist &= ~CSUM_TSO;
2772 		}
2773 		RL_UNLOCK(sc);
2774 		break;
2775 	case SIOCSIFFLAGS:
2776 		RL_LOCK(sc);
2777 		if ((ifp->if_flags & IFF_UP) != 0) {
2778 			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
2779 				if (((ifp->if_flags ^ sc->rl_if_flags)
2780 				    & (IFF_PROMISC | IFF_ALLMULTI)) != 0)
2781 					re_set_rxmode(sc);
2782 			} else
2783 				re_init_locked(sc);
2784 		} else {
2785 			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
2786 				re_stop(sc);
2787 		}
2788 		sc->rl_if_flags = ifp->if_flags;
2789 		RL_UNLOCK(sc);
2790 		break;
2791 	case SIOCADDMULTI:
2792 	case SIOCDELMULTI:
2793 		RL_LOCK(sc);
2794 		re_set_rxmode(sc);
2795 		RL_UNLOCK(sc);
2796 		break;
2797 	case SIOCGIFMEDIA:
2798 	case SIOCSIFMEDIA:
2799 		mii = device_get_softc(sc->rl_miibus);
2800 		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, command);
2801 		break;
2802 	case SIOCSIFCAP:
2803 	    {
2804 		int mask, reinit;
2805 
2806 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
2807 		reinit = 0;
2808 #ifdef DEVICE_POLLING
2809 		if (mask & IFCAP_POLLING) {
2810 			if (ifr->ifr_reqcap & IFCAP_POLLING) {
2811 				error = ether_poll_register(re_poll, ifp);
2812 				if (error)
2813 					return(error);
2814 				RL_LOCK(sc);
2815 				/* Disable interrupts */
2816 				CSR_WRITE_2(sc, RL_IMR, 0x0000);
2817 				ifp->if_capenable |= IFCAP_POLLING;
2818 				RL_UNLOCK(sc);
2819 			} else {
2820 				error = ether_poll_deregister(ifp);
2821 				/* Enable interrupts. */
2822 				RL_LOCK(sc);
2823 				CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2824 				ifp->if_capenable &= ~IFCAP_POLLING;
2825 				RL_UNLOCK(sc);
2826 			}
2827 		}
2828 #endif /* DEVICE_POLLING */
2829 		if (mask & IFCAP_HWCSUM) {
2830 			ifp->if_capenable ^= IFCAP_HWCSUM;
2831 			if (ifp->if_capenable & IFCAP_TXCSUM)
2832 				ifp->if_hwassist |= RE_CSUM_FEATURES;
2833 			else
2834 				ifp->if_hwassist &= ~RE_CSUM_FEATURES;
2835 			reinit = 1;
2836 		}
2837 		if (mask & IFCAP_VLAN_HWTAGGING) {
2838 			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
2839 			reinit = 1;
2840 		}
2841 		if (mask & IFCAP_TSO4) {
2842 			ifp->if_capenable ^= IFCAP_TSO4;
2843 			if ((IFCAP_TSO4 & ifp->if_capenable) &&
2844 			    (IFCAP_TSO4 & ifp->if_capabilities))
2845 				ifp->if_hwassist |= CSUM_TSO;
2846 			else
2847 				ifp->if_hwassist &= ~CSUM_TSO;
2848 			if (ifp->if_mtu > RL_TSO_MTU &&
2849 			    (ifp->if_capenable & IFCAP_TSO4) != 0) {
2850 				ifp->if_capenable &= ~IFCAP_TSO4;
2851 				ifp->if_hwassist &= ~CSUM_TSO;
2852 			}
2853 		}
2854 		if ((mask & IFCAP_WOL) != 0 &&
2855 		    (ifp->if_capabilities & IFCAP_WOL) != 0) {
2856 			if ((mask & IFCAP_WOL_UCAST) != 0)
2857 				ifp->if_capenable ^= IFCAP_WOL_UCAST;
2858 			if ((mask & IFCAP_WOL_MCAST) != 0)
2859 				ifp->if_capenable ^= IFCAP_WOL_MCAST;
2860 			if ((mask & IFCAP_WOL_MAGIC) != 0)
2861 				ifp->if_capenable ^= IFCAP_WOL_MAGIC;
2862 		}
2863 		if (reinit && ifp->if_drv_flags & IFF_DRV_RUNNING)
2864 			re_init(sc);
2865 		VLAN_CAPABILITIES(ifp);
2866 	    }
2867 		break;
2868 	default:
2869 		error = ether_ioctl(ifp, command, data);
2870 		break;
2871 	}
2872 
2873 	return (error);
2874 }
2875 
2876 static void
2877 re_watchdog(struct rl_softc *sc)
2878 {
2879 	struct ifnet		*ifp;
2880 
2881 	RL_LOCK_ASSERT(sc);
2882 
2883 	if (sc->rl_watchdog_timer == 0 || --sc->rl_watchdog_timer != 0)
2884 		return;
2885 
2886 	ifp = sc->rl_ifp;
2887 	re_txeof(sc);
2888 	if (sc->rl_ldata.rl_tx_free == sc->rl_ldata.rl_tx_desc_cnt) {
2889 		if_printf(ifp, "watchdog timeout (missed Tx interrupts) "
2890 		    "-- recovering\n");
2891 		if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2892 			taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_txtask);
2893 		return;
2894 	}
2895 
2896 	if_printf(ifp, "watchdog timeout\n");
2897 	ifp->if_oerrors++;
2898 
2899 	re_rxeof(sc, NULL);
2900 	re_init_locked(sc);
2901 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2902 		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_txtask);
2903 }
2904 
2905 /*
2906  * Stop the adapter and free any mbufs allocated to the
2907  * RX and TX lists.
2908  */
2909 static void
2910 re_stop(struct rl_softc *sc)
2911 {
2912 	int			i;
2913 	struct ifnet		*ifp;
2914 	struct rl_txdesc	*txd;
2915 	struct rl_rxdesc	*rxd;
2916 
2917 	RL_LOCK_ASSERT(sc);
2918 
2919 	ifp = sc->rl_ifp;
2920 
2921 	sc->rl_watchdog_timer = 0;
2922 	callout_stop(&sc->rl_stat_callout);
2923 	ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
2924 
2925 	if ((sc->rl_flags & RL_FLAG_CMDSTOP) != 0)
2926 		CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_STOPREQ | RL_CMD_TX_ENB |
2927 		    RL_CMD_RX_ENB);
2928 	else
2929 		CSR_WRITE_1(sc, RL_COMMAND, 0x00);
2930 	DELAY(1000);
2931 	CSR_WRITE_2(sc, RL_IMR, 0x0000);
2932 	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
2933 
2934 	if (sc->rl_head != NULL) {
2935 		m_freem(sc->rl_head);
2936 		sc->rl_head = sc->rl_tail = NULL;
2937 	}
2938 
2939 	/* Free the TX list buffers. */
2940 
2941 	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
2942 		txd = &sc->rl_ldata.rl_tx_desc[i];
2943 		if (txd->tx_m != NULL) {
2944 			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
2945 			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
2946 			bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag,
2947 			    txd->tx_dmamap);
2948 			m_freem(txd->tx_m);
2949 			txd->tx_m = NULL;
2950 		}
2951 	}
2952 
2953 	/* Free the RX list buffers. */
2954 
2955 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
2956 		rxd = &sc->rl_ldata.rl_rx_desc[i];
2957 		if (rxd->rx_m != NULL) {
2958 			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
2959 			    rxd->rx_dmamap, BUS_DMASYNC_POSTREAD);
2960 			bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag,
2961 			    rxd->rx_dmamap);
2962 			m_freem(rxd->rx_m);
2963 			rxd->rx_m = NULL;
2964 		}
2965 	}
2966 }
2967 
2968 /*
2969  * Device suspend routine.  Stop the interface and save some PCI
2970  * settings in case the BIOS doesn't restore them properly on
2971  * resume.
2972  */
2973 static int
2974 re_suspend(device_t dev)
2975 {
2976 	struct rl_softc		*sc;
2977 
2978 	sc = device_get_softc(dev);
2979 
2980 	RL_LOCK(sc);
2981 	re_stop(sc);
2982 	re_setwol(sc);
2983 	sc->suspended = 1;
2984 	RL_UNLOCK(sc);
2985 
2986 	return (0);
2987 }
2988 
2989 /*
2990  * Device resume routine.  Restore some PCI settings in case the BIOS
2991  * doesn't, re-enable busmastering, and restart the interface if
2992  * appropriate.
2993  */
2994 static int
2995 re_resume(device_t dev)
2996 {
2997 	struct rl_softc		*sc;
2998 	struct ifnet		*ifp;
2999 
3000 	sc = device_get_softc(dev);
3001 
3002 	RL_LOCK(sc);
3003 
3004 	ifp = sc->rl_ifp;
3005 	/* Take controller out of sleep mode. */
3006 	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
3007 		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
3008 			CSR_WRITE_1(sc, RL_GPIO,
3009 			    CSR_READ_1(sc, RL_GPIO) | 0x01);
3010 	}
3011 
3012 	/* reinitialize interface if necessary */
3013 	if (ifp->if_flags & IFF_UP)
3014 		re_init_locked(sc);
3015 
3016 	/*
3017 	 * Clear WOL matching such that normal Rx filtering
3018 	 * wouldn't interfere with WOL patterns.
3019 	 */
3020 	re_clrwol(sc);
3021 	sc->suspended = 0;
3022 	RL_UNLOCK(sc);
3023 
3024 	return (0);
3025 }
3026 
3027 /*
3028  * Stop all chip I/O so that the kernel's probe routines don't
3029  * get confused by errant DMAs when rebooting.
3030  */
3031 static int
3032 re_shutdown(device_t dev)
3033 {
3034 	struct rl_softc		*sc;
3035 
3036 	sc = device_get_softc(dev);
3037 
3038 	RL_LOCK(sc);
3039 	re_stop(sc);
3040 	/*
3041 	 * Mark interface as down since otherwise we will panic if
3042 	 * interrupt comes in later on, which can happen in some
3043 	 * cases.
3044 	 */
3045 	sc->rl_ifp->if_flags &= ~IFF_UP;
3046 	re_setwol(sc);
3047 	RL_UNLOCK(sc);
3048 
3049 	return (0);
3050 }
3051 
3052 static void
3053 re_setwol(struct rl_softc *sc)
3054 {
3055 	struct ifnet		*ifp;
3056 	int			pmc;
3057 	uint16_t		pmstat;
3058 	uint8_t			v;
3059 
3060 	RL_LOCK_ASSERT(sc);
3061 
3062 	if (pci_find_extcap(sc->rl_dev, PCIY_PMG, &pmc) != 0)
3063 		return;
3064 
3065 	ifp = sc->rl_ifp;
3066 	/* Put controller into sleep mode. */
3067 	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
3068 		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
3069 			CSR_WRITE_1(sc, RL_GPIO,
3070 			    CSR_READ_1(sc, RL_GPIO) & ~0x01);
3071 	}
3072 	if ((ifp->if_capenable & IFCAP_WOL) != 0 &&
3073 	    (sc->rl_flags & RL_FLAG_WOLRXENB) != 0)
3074 		CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RX_ENB);
3075 	/* Enable config register write. */
3076 	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
3077 
3078 	/* Enable PME. */
3079 	v = CSR_READ_1(sc, RL_CFG1);
3080 	v &= ~RL_CFG1_PME;
3081 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3082 		v |= RL_CFG1_PME;
3083 	CSR_WRITE_1(sc, RL_CFG1, v);
3084 
3085 	v = CSR_READ_1(sc, RL_CFG3);
3086 	v &= ~(RL_CFG3_WOL_LINK | RL_CFG3_WOL_MAGIC);
3087 	if ((ifp->if_capenable & IFCAP_WOL_MAGIC) != 0)
3088 		v |= RL_CFG3_WOL_MAGIC;
3089 	CSR_WRITE_1(sc, RL_CFG3, v);
3090 
3091 	/* Config register write done. */
3092 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3093 
3094 	v = CSR_READ_1(sc, RL_CFG5);
3095 	v &= ~(RL_CFG5_WOL_BCAST | RL_CFG5_WOL_MCAST | RL_CFG5_WOL_UCAST);
3096 	v &= ~RL_CFG5_WOL_LANWAKE;
3097 	if ((ifp->if_capenable & IFCAP_WOL_UCAST) != 0)
3098 		v |= RL_CFG5_WOL_UCAST;
3099 	if ((ifp->if_capenable & IFCAP_WOL_MCAST) != 0)
3100 		v |= RL_CFG5_WOL_MCAST | RL_CFG5_WOL_BCAST;
3101 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3102 		v |= RL_CFG5_WOL_LANWAKE;
3103 	CSR_WRITE_1(sc, RL_CFG5, v);
3104 
3105 	/*
3106 	 * It seems that hardware resets its link speed to 100Mbps in
3107 	 * power down mode so switching to 100Mbps in driver is not
3108 	 * needed.
3109 	 */
3110 
3111 	/* Request PME if WOL is requested. */
3112 	pmstat = pci_read_config(sc->rl_dev, pmc + PCIR_POWER_STATUS, 2);
3113 	pmstat &= ~(PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE);
3114 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3115 		pmstat |= PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE;
3116 	pci_write_config(sc->rl_dev, pmc + PCIR_POWER_STATUS, pmstat, 2);
3117 }
3118 
3119 static void
3120 re_clrwol(struct rl_softc *sc)
3121 {
3122 	int			pmc;
3123 	uint8_t			v;
3124 
3125 	RL_LOCK_ASSERT(sc);
3126 
3127 	if (pci_find_extcap(sc->rl_dev, PCIY_PMG, &pmc) != 0)
3128 		return;
3129 
3130 	/* Enable config register write. */
3131 	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
3132 
3133 	v = CSR_READ_1(sc, RL_CFG3);
3134 	v &= ~(RL_CFG3_WOL_LINK | RL_CFG3_WOL_MAGIC);
3135 	CSR_WRITE_1(sc, RL_CFG3, v);
3136 
3137 	/* Config register write done. */
3138 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3139 
3140 	v = CSR_READ_1(sc, RL_CFG5);
3141 	v &= ~(RL_CFG5_WOL_BCAST | RL_CFG5_WOL_MCAST | RL_CFG5_WOL_UCAST);
3142 	v &= ~RL_CFG5_WOL_LANWAKE;
3143 	CSR_WRITE_1(sc, RL_CFG5, v);
3144 }
3145