xref: /freebsd/sys/dev/re/if_re.c (revision 050570efa79efcc9cf5adeb545f1a679c8dc377b)
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/sysctl.h>
127 #include <sys/taskqueue.h>
128 
129 #include <net/if.h>
130 #include <net/if_arp.h>
131 #include <net/ethernet.h>
132 #include <net/if_dl.h>
133 #include <net/if_media.h>
134 #include <net/if_types.h>
135 #include <net/if_vlan_var.h>
136 
137 #include <net/bpf.h>
138 
139 #include <machine/bus.h>
140 #include <machine/resource.h>
141 #include <sys/bus.h>
142 #include <sys/rman.h>
143 
144 #include <dev/mii/mii.h>
145 #include <dev/mii/miivar.h>
146 
147 #include <dev/pci/pcireg.h>
148 #include <dev/pci/pcivar.h>
149 
150 #include <pci/if_rlreg.h>
151 
152 MODULE_DEPEND(re, pci, 1, 1, 1);
153 MODULE_DEPEND(re, ether, 1, 1, 1);
154 MODULE_DEPEND(re, miibus, 1, 1, 1);
155 
156 /* "device miibus" required.  See GENERIC if you get errors here. */
157 #include "miibus_if.h"
158 
159 /* Tunables. */
160 static int msi_disable = 0;
161 TUNABLE_INT("hw.re.msi_disable", &msi_disable);
162 static int prefer_iomap = 0;
163 TUNABLE_INT("hw.re.prefer_iomap", &prefer_iomap);
164 
165 #define RE_CSUM_FEATURES    (CSUM_IP | CSUM_TCP | CSUM_UDP)
166 
167 /*
168  * Various supported device vendors/types and their names.
169  */
170 static struct rl_type re_devs[] = {
171 	{ DLINK_VENDORID, DLINK_DEVICEID_528T, 0,
172 	    "D-Link DGE-528(T) Gigabit Ethernet Adapter" },
173 	{ RT_VENDORID, RT_DEVICEID_8139, 0,
174 	    "RealTek 8139C+ 10/100BaseTX" },
175 	{ RT_VENDORID, RT_DEVICEID_8101E, 0,
176 	    "RealTek 8101E/8102E/8102EL/8103E PCIe 10/100baseTX" },
177 	{ RT_VENDORID, RT_DEVICEID_8168, 0,
178 	    "RealTek 8168/8111 B/C/CP/D/DP/E 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,  "", RL_MTU },
193 	{ RL_HWREV_8139A, RL_8139, "A", RL_MTU },
194 	{ RL_HWREV_8139AG, RL_8139, "A-G", RL_MTU },
195 	{ RL_HWREV_8139B, RL_8139, "B", RL_MTU },
196 	{ RL_HWREV_8130, RL_8139, "8130", RL_MTU },
197 	{ RL_HWREV_8139C, RL_8139, "C", RL_MTU },
198 	{ RL_HWREV_8139D, RL_8139, "8139D/8100B/8100C", RL_MTU },
199 	{ RL_HWREV_8139CPLUS, RL_8139CPLUS, "C+", RL_MTU },
200 	{ RL_HWREV_8168B_SPIN1, RL_8169, "8168", RL_JUMBO_MTU },
201 	{ RL_HWREV_8169, RL_8169, "8169", RL_JUMBO_MTU },
202 	{ RL_HWREV_8169S, RL_8169, "8169S", RL_JUMBO_MTU },
203 	{ RL_HWREV_8110S, RL_8169, "8110S", RL_JUMBO_MTU },
204 	{ RL_HWREV_8169_8110SB, RL_8169, "8169SB/8110SB", RL_JUMBO_MTU },
205 	{ RL_HWREV_8169_8110SC, RL_8169, "8169SC/8110SC", RL_JUMBO_MTU },
206 	{ RL_HWREV_8169_8110SBL, RL_8169, "8169SBL/8110SBL", RL_JUMBO_MTU },
207 	{ RL_HWREV_8169_8110SCE, RL_8169, "8169SC/8110SC", RL_JUMBO_MTU },
208 	{ RL_HWREV_8100, RL_8139, "8100", RL_MTU },
209 	{ RL_HWREV_8101, RL_8139, "8101", RL_MTU },
210 	{ RL_HWREV_8100E, RL_8169, "8100E", RL_MTU },
211 	{ RL_HWREV_8101E, RL_8169, "8101E", RL_MTU },
212 	{ RL_HWREV_8102E, RL_8169, "8102E", RL_MTU },
213 	{ RL_HWREV_8102EL, RL_8169, "8102EL", RL_MTU },
214 	{ RL_HWREV_8102EL_SPIN1, RL_8169, "8102EL", RL_MTU },
215 	{ RL_HWREV_8103E, RL_8169, "8103E", RL_MTU },
216 	{ RL_HWREV_8168B_SPIN2, RL_8169, "8168", RL_JUMBO_MTU },
217 	{ RL_HWREV_8168B_SPIN3, RL_8169, "8168", RL_JUMBO_MTU },
218 	{ RL_HWREV_8168C, RL_8169, "8168C/8111C", RL_JUMBO_MTU_6K },
219 	{ RL_HWREV_8168C_SPIN2, RL_8169, "8168C/8111C", RL_JUMBO_MTU_6K },
220 	{ RL_HWREV_8168CP, RL_8169, "8168CP/8111CP", RL_JUMBO_MTU_6K },
221 	{ RL_HWREV_8168D, RL_8169, "8168D/8111D", RL_JUMBO_MTU_9K },
222 	{ RL_HWREV_8168DP, RL_8169, "8168DP/8111DP", RL_JUMBO_MTU_9K },
223 	{ RL_HWREV_8168E, RL_8169, "8168E/8111E", RL_JUMBO_MTU_9K},
224 	{ RL_HWREV_8168E_VL, RL_8169, "8168E/8111E-VL", RL_JUMBO_MTU_6K},
225 	{ 0, 0, NULL, 0 }
226 };
227 
228 static int re_probe		(device_t);
229 static int re_attach		(device_t);
230 static int re_detach		(device_t);
231 
232 static int re_encap		(struct rl_softc *, struct mbuf **);
233 
234 static void re_dma_map_addr	(void *, bus_dma_segment_t *, int, int);
235 static int re_allocmem		(device_t, struct rl_softc *);
236 static __inline void re_discard_rxbuf
237 				(struct rl_softc *, int);
238 static int re_newbuf		(struct rl_softc *, int);
239 static int re_jumbo_newbuf	(struct rl_softc *, int);
240 static int re_rx_list_init	(struct rl_softc *);
241 static int re_jrx_list_init	(struct rl_softc *);
242 static int re_tx_list_init	(struct rl_softc *);
243 #ifdef RE_FIXUP_RX
244 static __inline void re_fixup_rx
245 				(struct mbuf *);
246 #endif
247 static int re_rxeof		(struct rl_softc *, int *);
248 static void re_txeof		(struct rl_softc *);
249 #ifdef DEVICE_POLLING
250 static int re_poll		(struct ifnet *, enum poll_cmd, int);
251 static int re_poll_locked	(struct ifnet *, enum poll_cmd, int);
252 #endif
253 static int re_intr		(void *);
254 static void re_tick		(void *);
255 static void re_tx_task		(void *, int);
256 static void re_int_task		(void *, int);
257 static void re_start		(struct ifnet *);
258 static int re_ioctl		(struct ifnet *, u_long, caddr_t);
259 static void re_init		(void *);
260 static void re_init_locked	(struct rl_softc *);
261 static void re_stop		(struct rl_softc *);
262 static void re_watchdog		(struct rl_softc *);
263 static int re_suspend		(device_t);
264 static int re_resume		(device_t);
265 static int re_shutdown		(device_t);
266 static int re_ifmedia_upd	(struct ifnet *);
267 static void re_ifmedia_sts	(struct ifnet *, struct ifmediareq *);
268 
269 static void re_eeprom_putbyte	(struct rl_softc *, int);
270 static void re_eeprom_getword	(struct rl_softc *, int, u_int16_t *);
271 static void re_read_eeprom	(struct rl_softc *, caddr_t, int, int);
272 static int re_gmii_readreg	(device_t, int, int);
273 static int re_gmii_writereg	(device_t, int, int, int);
274 
275 static int re_miibus_readreg	(device_t, int, int);
276 static int re_miibus_writereg	(device_t, int, int, int);
277 static void re_miibus_statchg	(device_t);
278 
279 static void re_set_jumbo	(struct rl_softc *, int);
280 static void re_set_rxmode		(struct rl_softc *);
281 static void re_reset		(struct rl_softc *);
282 static void re_setwol		(struct rl_softc *);
283 static void re_clrwol		(struct rl_softc *);
284 
285 #ifdef RE_DIAG
286 static int re_diag		(struct rl_softc *);
287 #endif
288 
289 static void re_add_sysctls	(struct rl_softc *);
290 static int re_sysctl_stats	(SYSCTL_HANDLER_ARGS);
291 
292 static device_method_t re_methods[] = {
293 	/* Device interface */
294 	DEVMETHOD(device_probe,		re_probe),
295 	DEVMETHOD(device_attach,	re_attach),
296 	DEVMETHOD(device_detach,	re_detach),
297 	DEVMETHOD(device_suspend,	re_suspend),
298 	DEVMETHOD(device_resume,	re_resume),
299 	DEVMETHOD(device_shutdown,	re_shutdown),
300 
301 	/* bus interface */
302 	DEVMETHOD(bus_print_child,	bus_generic_print_child),
303 	DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
304 
305 	/* MII interface */
306 	DEVMETHOD(miibus_readreg,	re_miibus_readreg),
307 	DEVMETHOD(miibus_writereg,	re_miibus_writereg),
308 	DEVMETHOD(miibus_statchg,	re_miibus_statchg),
309 
310 	{ 0, 0 }
311 };
312 
313 static driver_t re_driver = {
314 	"re",
315 	re_methods,
316 	sizeof(struct rl_softc)
317 };
318 
319 static devclass_t re_devclass;
320 
321 DRIVER_MODULE(re, pci, re_driver, re_devclass, 0, 0);
322 DRIVER_MODULE(miibus, re, miibus_driver, miibus_devclass, 0, 0);
323 
324 #define EE_SET(x)					\
325 	CSR_WRITE_1(sc, RL_EECMD,			\
326 		CSR_READ_1(sc, RL_EECMD) | x)
327 
328 #define EE_CLR(x)					\
329 	CSR_WRITE_1(sc, RL_EECMD,			\
330 		CSR_READ_1(sc, RL_EECMD) & ~x)
331 
332 /*
333  * Send a read command and address to the EEPROM, check for ACK.
334  */
335 static void
336 re_eeprom_putbyte(struct rl_softc *sc, int addr)
337 {
338 	int			d, i;
339 
340 	d = addr | (RL_9346_READ << sc->rl_eewidth);
341 
342 	/*
343 	 * Feed in each bit and strobe the clock.
344 	 */
345 
346 	for (i = 1 << (sc->rl_eewidth + 3); i; i >>= 1) {
347 		if (d & i) {
348 			EE_SET(RL_EE_DATAIN);
349 		} else {
350 			EE_CLR(RL_EE_DATAIN);
351 		}
352 		DELAY(100);
353 		EE_SET(RL_EE_CLK);
354 		DELAY(150);
355 		EE_CLR(RL_EE_CLK);
356 		DELAY(100);
357 	}
358 }
359 
360 /*
361  * Read a word of data stored in the EEPROM at address 'addr.'
362  */
363 static void
364 re_eeprom_getword(struct rl_softc *sc, int addr, u_int16_t *dest)
365 {
366 	int			i;
367 	u_int16_t		word = 0;
368 
369 	/*
370 	 * Send address of word we want to read.
371 	 */
372 	re_eeprom_putbyte(sc, addr);
373 
374 	/*
375 	 * Start reading bits from EEPROM.
376 	 */
377 	for (i = 0x8000; i; i >>= 1) {
378 		EE_SET(RL_EE_CLK);
379 		DELAY(100);
380 		if (CSR_READ_1(sc, RL_EECMD) & RL_EE_DATAOUT)
381 			word |= i;
382 		EE_CLR(RL_EE_CLK);
383 		DELAY(100);
384 	}
385 
386 	*dest = word;
387 }
388 
389 /*
390  * Read a sequence of words from the EEPROM.
391  */
392 static void
393 re_read_eeprom(struct rl_softc *sc, caddr_t dest, int off, int cnt)
394 {
395 	int			i;
396 	u_int16_t		word = 0, *ptr;
397 
398 	CSR_SETBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM);
399 
400         DELAY(100);
401 
402 	for (i = 0; i < cnt; i++) {
403 		CSR_SETBIT_1(sc, RL_EECMD, RL_EE_SEL);
404 		re_eeprom_getword(sc, off + i, &word);
405 		CSR_CLRBIT_1(sc, RL_EECMD, RL_EE_SEL);
406 		ptr = (u_int16_t *)(dest + (i * 2));
407                 *ptr = word;
408 	}
409 
410 	CSR_CLRBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM);
411 }
412 
413 static int
414 re_gmii_readreg(device_t dev, int phy, int reg)
415 {
416 	struct rl_softc		*sc;
417 	u_int32_t		rval;
418 	int			i;
419 
420 	sc = device_get_softc(dev);
421 
422 	/* Let the rgephy driver read the GMEDIASTAT register */
423 
424 	if (reg == RL_GMEDIASTAT) {
425 		rval = CSR_READ_1(sc, RL_GMEDIASTAT);
426 		return (rval);
427 	}
428 
429 	CSR_WRITE_4(sc, RL_PHYAR, reg << 16);
430 
431 	for (i = 0; i < RL_PHY_TIMEOUT; i++) {
432 		rval = CSR_READ_4(sc, RL_PHYAR);
433 		if (rval & RL_PHYAR_BUSY)
434 			break;
435 		DELAY(25);
436 	}
437 
438 	if (i == RL_PHY_TIMEOUT) {
439 		device_printf(sc->rl_dev, "PHY read failed\n");
440 		return (0);
441 	}
442 
443 	/*
444 	 * Controller requires a 20us delay to process next MDIO request.
445 	 */
446 	DELAY(20);
447 
448 	return (rval & RL_PHYAR_PHYDATA);
449 }
450 
451 static int
452 re_gmii_writereg(device_t dev, int phy, int reg, int data)
453 {
454 	struct rl_softc		*sc;
455 	u_int32_t		rval;
456 	int			i;
457 
458 	sc = device_get_softc(dev);
459 
460 	CSR_WRITE_4(sc, RL_PHYAR, (reg << 16) |
461 	    (data & RL_PHYAR_PHYDATA) | RL_PHYAR_BUSY);
462 
463 	for (i = 0; i < RL_PHY_TIMEOUT; i++) {
464 		rval = CSR_READ_4(sc, RL_PHYAR);
465 		if (!(rval & RL_PHYAR_BUSY))
466 			break;
467 		DELAY(25);
468 	}
469 
470 	if (i == RL_PHY_TIMEOUT) {
471 		device_printf(sc->rl_dev, "PHY write failed\n");
472 		return (0);
473 	}
474 
475 	/*
476 	 * Controller requires a 20us delay to process next MDIO request.
477 	 */
478 	DELAY(20);
479 
480 	return (0);
481 }
482 
483 static int
484 re_miibus_readreg(device_t dev, int phy, int reg)
485 {
486 	struct rl_softc		*sc;
487 	u_int16_t		rval = 0;
488 	u_int16_t		re8139_reg = 0;
489 
490 	sc = device_get_softc(dev);
491 
492 	if (sc->rl_type == RL_8169) {
493 		rval = re_gmii_readreg(dev, phy, reg);
494 		return (rval);
495 	}
496 
497 	switch (reg) {
498 	case MII_BMCR:
499 		re8139_reg = RL_BMCR;
500 		break;
501 	case MII_BMSR:
502 		re8139_reg = RL_BMSR;
503 		break;
504 	case MII_ANAR:
505 		re8139_reg = RL_ANAR;
506 		break;
507 	case MII_ANER:
508 		re8139_reg = RL_ANER;
509 		break;
510 	case MII_ANLPAR:
511 		re8139_reg = RL_LPAR;
512 		break;
513 	case MII_PHYIDR1:
514 	case MII_PHYIDR2:
515 		return (0);
516 	/*
517 	 * Allow the rlphy driver to read the media status
518 	 * register. If we have a link partner which does not
519 	 * support NWAY, this is the register which will tell
520 	 * us the results of parallel detection.
521 	 */
522 	case RL_MEDIASTAT:
523 		rval = CSR_READ_1(sc, RL_MEDIASTAT);
524 		return (rval);
525 	default:
526 		device_printf(sc->rl_dev, "bad phy register\n");
527 		return (0);
528 	}
529 	rval = CSR_READ_2(sc, re8139_reg);
530 	if (sc->rl_type == RL_8139CPLUS && re8139_reg == RL_BMCR) {
531 		/* 8139C+ has different bit layout. */
532 		rval &= ~(BMCR_LOOP | BMCR_ISO);
533 	}
534 	return (rval);
535 }
536 
537 static int
538 re_miibus_writereg(device_t dev, int phy, int reg, int data)
539 {
540 	struct rl_softc		*sc;
541 	u_int16_t		re8139_reg = 0;
542 	int			rval = 0;
543 
544 	sc = device_get_softc(dev);
545 
546 	if (sc->rl_type == RL_8169) {
547 		rval = re_gmii_writereg(dev, phy, reg, data);
548 		return (rval);
549 	}
550 
551 	switch (reg) {
552 	case MII_BMCR:
553 		re8139_reg = RL_BMCR;
554 		if (sc->rl_type == RL_8139CPLUS) {
555 			/* 8139C+ has different bit layout. */
556 			data &= ~(BMCR_LOOP | BMCR_ISO);
557 		}
558 		break;
559 	case MII_BMSR:
560 		re8139_reg = RL_BMSR;
561 		break;
562 	case MII_ANAR:
563 		re8139_reg = RL_ANAR;
564 		break;
565 	case MII_ANER:
566 		re8139_reg = RL_ANER;
567 		break;
568 	case MII_ANLPAR:
569 		re8139_reg = RL_LPAR;
570 		break;
571 	case MII_PHYIDR1:
572 	case MII_PHYIDR2:
573 		return (0);
574 		break;
575 	default:
576 		device_printf(sc->rl_dev, "bad phy register\n");
577 		return (0);
578 	}
579 	CSR_WRITE_2(sc, re8139_reg, data);
580 	return (0);
581 }
582 
583 static void
584 re_miibus_statchg(device_t dev)
585 {
586 	struct rl_softc		*sc;
587 	struct ifnet		*ifp;
588 	struct mii_data		*mii;
589 
590 	sc = device_get_softc(dev);
591 	mii = device_get_softc(sc->rl_miibus);
592 	ifp = sc->rl_ifp;
593 	if (mii == NULL || ifp == NULL ||
594 	    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
595 		return;
596 
597 	sc->rl_flags &= ~RL_FLAG_LINK;
598 	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
599 	    (IFM_ACTIVE | IFM_AVALID)) {
600 		switch (IFM_SUBTYPE(mii->mii_media_active)) {
601 		case IFM_10_T:
602 		case IFM_100_TX:
603 			sc->rl_flags |= RL_FLAG_LINK;
604 			break;
605 		case IFM_1000_T:
606 			if ((sc->rl_flags & RL_FLAG_FASTETHER) != 0)
607 				break;
608 			sc->rl_flags |= RL_FLAG_LINK;
609 			break;
610 		default:
611 			break;
612 		}
613 	}
614 	/*
615 	 * RealTek controllers does not provide any interface to
616 	 * Tx/Rx MACs for resolved speed, duplex and flow-control
617 	 * parameters.
618 	 */
619 }
620 
621 /*
622  * Set the RX configuration and 64-bit multicast hash filter.
623  */
624 static void
625 re_set_rxmode(struct rl_softc *sc)
626 {
627 	struct ifnet		*ifp;
628 	struct ifmultiaddr	*ifma;
629 	uint32_t		hashes[2] = { 0, 0 };
630 	uint32_t		h, rxfilt;
631 
632 	RL_LOCK_ASSERT(sc);
633 
634 	ifp = sc->rl_ifp;
635 
636 	rxfilt = RL_RXCFG_CONFIG | RL_RXCFG_RX_INDIV | RL_RXCFG_RX_BROAD;
637 
638 	if (ifp->if_flags & (IFF_ALLMULTI | IFF_PROMISC)) {
639 		if (ifp->if_flags & IFF_PROMISC)
640 			rxfilt |= RL_RXCFG_RX_ALLPHYS;
641 		/*
642 		 * Unlike other hardwares, we have to explicitly set
643 		 * RL_RXCFG_RX_MULTI to receive multicast frames in
644 		 * promiscuous mode.
645 		 */
646 		rxfilt |= RL_RXCFG_RX_MULTI;
647 		hashes[0] = hashes[1] = 0xffffffff;
648 		goto done;
649 	}
650 
651 	if_maddr_rlock(ifp);
652 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
653 		if (ifma->ifma_addr->sa_family != AF_LINK)
654 			continue;
655 		h = ether_crc32_be(LLADDR((struct sockaddr_dl *)
656 		    ifma->ifma_addr), ETHER_ADDR_LEN) >> 26;
657 		if (h < 32)
658 			hashes[0] |= (1 << h);
659 		else
660 			hashes[1] |= (1 << (h - 32));
661 	}
662 	if_maddr_runlock(ifp);
663 
664 	if (hashes[0] != 0 || hashes[1] != 0) {
665 		/*
666 		 * For some unfathomable reason, RealTek decided to
667 		 * reverse the order of the multicast hash registers
668 		 * in the PCI Express parts.  This means we have to
669 		 * write the hash pattern in reverse order for those
670 		 * devices.
671 		 */
672 		if ((sc->rl_flags & RL_FLAG_PCIE) != 0) {
673 			h = bswap32(hashes[0]);
674 			hashes[0] = bswap32(hashes[1]);
675 			hashes[1] = h;
676 		}
677 		rxfilt |= RL_RXCFG_RX_MULTI;
678 	}
679 
680 done:
681 	CSR_WRITE_4(sc, RL_MAR0, hashes[0]);
682 	CSR_WRITE_4(sc, RL_MAR4, hashes[1]);
683 	CSR_WRITE_4(sc, RL_RXCFG, rxfilt);
684 }
685 
686 static void
687 re_reset(struct rl_softc *sc)
688 {
689 	int			i;
690 
691 	RL_LOCK_ASSERT(sc);
692 
693 	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RESET);
694 
695 	for (i = 0; i < RL_TIMEOUT; i++) {
696 		DELAY(10);
697 		if (!(CSR_READ_1(sc, RL_COMMAND) & RL_CMD_RESET))
698 			break;
699 	}
700 	if (i == RL_TIMEOUT)
701 		device_printf(sc->rl_dev, "reset never completed!\n");
702 
703 	if ((sc->rl_flags & RL_FLAG_MACRESET) != 0)
704 		CSR_WRITE_1(sc, 0x82, 1);
705 	if (sc->rl_hwrev->rl_rev == RL_HWREV_8169S)
706 		re_gmii_writereg(sc->rl_dev, 1, 0x0b, 0);
707 }
708 
709 #ifdef RE_DIAG
710 
711 /*
712  * The following routine is designed to test for a defect on some
713  * 32-bit 8169 cards. Some of these NICs have the REQ64# and ACK64#
714  * lines connected to the bus, however for a 32-bit only card, they
715  * should be pulled high. The result of this defect is that the
716  * NIC will not work right if you plug it into a 64-bit slot: DMA
717  * operations will be done with 64-bit transfers, which will fail
718  * because the 64-bit data lines aren't connected.
719  *
720  * There's no way to work around this (short of talking a soldering
721  * iron to the board), however we can detect it. The method we use
722  * here is to put the NIC into digital loopback mode, set the receiver
723  * to promiscuous mode, and then try to send a frame. We then compare
724  * the frame data we sent to what was received. If the data matches,
725  * then the NIC is working correctly, otherwise we know the user has
726  * a defective NIC which has been mistakenly plugged into a 64-bit PCI
727  * slot. In the latter case, there's no way the NIC can work correctly,
728  * so we print out a message on the console and abort the device attach.
729  */
730 
731 static int
732 re_diag(struct rl_softc *sc)
733 {
734 	struct ifnet		*ifp = sc->rl_ifp;
735 	struct mbuf		*m0;
736 	struct ether_header	*eh;
737 	struct rl_desc		*cur_rx;
738 	u_int16_t		status;
739 	u_int32_t		rxstat;
740 	int			total_len, i, error = 0, phyaddr;
741 	u_int8_t		dst[] = { 0x00, 'h', 'e', 'l', 'l', 'o' };
742 	u_int8_t		src[] = { 0x00, 'w', 'o', 'r', 'l', 'd' };
743 
744 	/* Allocate a single mbuf */
745 	MGETHDR(m0, M_DONTWAIT, MT_DATA);
746 	if (m0 == NULL)
747 		return (ENOBUFS);
748 
749 	RL_LOCK(sc);
750 
751 	/*
752 	 * Initialize the NIC in test mode. This sets the chip up
753 	 * so that it can send and receive frames, but performs the
754 	 * following special functions:
755 	 * - Puts receiver in promiscuous mode
756 	 * - Enables digital loopback mode
757 	 * - Leaves interrupts turned off
758 	 */
759 
760 	ifp->if_flags |= IFF_PROMISC;
761 	sc->rl_testmode = 1;
762 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
763 	re_init_locked(sc);
764 	sc->rl_flags |= RL_FLAG_LINK;
765 	if (sc->rl_type == RL_8169)
766 		phyaddr = 1;
767 	else
768 		phyaddr = 0;
769 
770 	re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_RESET);
771 	for (i = 0; i < RL_TIMEOUT; i++) {
772 		status = re_miibus_readreg(sc->rl_dev, phyaddr, MII_BMCR);
773 		if (!(status & BMCR_RESET))
774 			break;
775 	}
776 
777 	re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_LOOP);
778 	CSR_WRITE_2(sc, RL_ISR, RL_INTRS);
779 
780 	DELAY(100000);
781 
782 	/* Put some data in the mbuf */
783 
784 	eh = mtod(m0, struct ether_header *);
785 	bcopy ((char *)&dst, eh->ether_dhost, ETHER_ADDR_LEN);
786 	bcopy ((char *)&src, eh->ether_shost, ETHER_ADDR_LEN);
787 	eh->ether_type = htons(ETHERTYPE_IP);
788 	m0->m_pkthdr.len = m0->m_len = ETHER_MIN_LEN - ETHER_CRC_LEN;
789 
790 	/*
791 	 * Queue the packet, start transmission.
792 	 * Note: IF_HANDOFF() ultimately calls re_start() for us.
793 	 */
794 
795 	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
796 	RL_UNLOCK(sc);
797 	/* XXX: re_diag must not be called when in ALTQ mode */
798 	IF_HANDOFF(&ifp->if_snd, m0, ifp);
799 	RL_LOCK(sc);
800 	m0 = NULL;
801 
802 	/* Wait for it to propagate through the chip */
803 
804 	DELAY(100000);
805 	for (i = 0; i < RL_TIMEOUT; i++) {
806 		status = CSR_READ_2(sc, RL_ISR);
807 		CSR_WRITE_2(sc, RL_ISR, status);
808 		if ((status & (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK)) ==
809 		    (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK))
810 			break;
811 		DELAY(10);
812 	}
813 
814 	if (i == RL_TIMEOUT) {
815 		device_printf(sc->rl_dev,
816 		    "diagnostic failed, failed to receive packet in"
817 		    " loopback mode\n");
818 		error = EIO;
819 		goto done;
820 	}
821 
822 	/*
823 	 * The packet should have been dumped into the first
824 	 * entry in the RX DMA ring. Grab it from there.
825 	 */
826 
827 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
828 	    sc->rl_ldata.rl_rx_list_map,
829 	    BUS_DMASYNC_POSTREAD);
830 	bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag,
831 	    sc->rl_ldata.rl_rx_desc[0].rx_dmamap,
832 	    BUS_DMASYNC_POSTREAD);
833 	bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag,
834 	    sc->rl_ldata.rl_rx_desc[0].rx_dmamap);
835 
836 	m0 = sc->rl_ldata.rl_rx_desc[0].rx_m;
837 	sc->rl_ldata.rl_rx_desc[0].rx_m = NULL;
838 	eh = mtod(m0, struct ether_header *);
839 
840 	cur_rx = &sc->rl_ldata.rl_rx_list[0];
841 	total_len = RL_RXBYTES(cur_rx);
842 	rxstat = le32toh(cur_rx->rl_cmdstat);
843 
844 	if (total_len != ETHER_MIN_LEN) {
845 		device_printf(sc->rl_dev,
846 		    "diagnostic failed, received short packet\n");
847 		error = EIO;
848 		goto done;
849 	}
850 
851 	/* Test that the received packet data matches what we sent. */
852 
853 	if (bcmp((char *)&eh->ether_dhost, (char *)&dst, ETHER_ADDR_LEN) ||
854 	    bcmp((char *)&eh->ether_shost, (char *)&src, ETHER_ADDR_LEN) ||
855 	    ntohs(eh->ether_type) != ETHERTYPE_IP) {
856 		device_printf(sc->rl_dev, "WARNING, DMA FAILURE!\n");
857 		device_printf(sc->rl_dev, "expected TX data: %6D/%6D/0x%x\n",
858 		    dst, ":", src, ":", ETHERTYPE_IP);
859 		device_printf(sc->rl_dev, "received RX data: %6D/%6D/0x%x\n",
860 		    eh->ether_dhost, ":",  eh->ether_shost, ":",
861 		    ntohs(eh->ether_type));
862 		device_printf(sc->rl_dev, "You may have a defective 32-bit "
863 		    "NIC plugged into a 64-bit PCI slot.\n");
864 		device_printf(sc->rl_dev, "Please re-install the NIC in a "
865 		    "32-bit slot for proper operation.\n");
866 		device_printf(sc->rl_dev, "Read the re(4) man page for more "
867 		    "details.\n");
868 		error = EIO;
869 	}
870 
871 done:
872 	/* Turn interface off, release resources */
873 
874 	sc->rl_testmode = 0;
875 	sc->rl_flags &= ~RL_FLAG_LINK;
876 	ifp->if_flags &= ~IFF_PROMISC;
877 	re_stop(sc);
878 	if (m0 != NULL)
879 		m_freem(m0);
880 
881 	RL_UNLOCK(sc);
882 
883 	return (error);
884 }
885 
886 #endif
887 
888 /*
889  * Probe for a RealTek 8139C+/8169/8110 chip. Check the PCI vendor and device
890  * IDs against our list and return a device name if we find a match.
891  */
892 static int
893 re_probe(device_t dev)
894 {
895 	struct rl_type		*t;
896 	uint16_t		devid, vendor;
897 	uint16_t		revid, sdevid;
898 	int			i;
899 
900 	vendor = pci_get_vendor(dev);
901 	devid = pci_get_device(dev);
902 	revid = pci_get_revid(dev);
903 	sdevid = pci_get_subdevice(dev);
904 
905 	if (vendor == LINKSYS_VENDORID && devid == LINKSYS_DEVICEID_EG1032) {
906 		if (sdevid != LINKSYS_SUBDEVICE_EG1032_REV3) {
907 			/*
908 			 * Only attach to rev. 3 of the Linksys EG1032 adapter.
909 			 * Rev. 2 is supported by sk(4).
910 			 */
911 			return (ENXIO);
912 		}
913 	}
914 
915 	if (vendor == RT_VENDORID && devid == RT_DEVICEID_8139) {
916 		if (revid != 0x20) {
917 			/* 8139, let rl(4) take care of this device. */
918 			return (ENXIO);
919 		}
920 	}
921 
922 	t = re_devs;
923 	for (i = 0; i < sizeof(re_devs) / sizeof(re_devs[0]); i++, t++) {
924 		if (vendor == t->rl_vid && devid == t->rl_did) {
925 			device_set_desc(dev, t->rl_name);
926 			return (BUS_PROBE_DEFAULT);
927 		}
928 	}
929 
930 	return (ENXIO);
931 }
932 
933 /*
934  * Map a single buffer address.
935  */
936 
937 static void
938 re_dma_map_addr(void *arg, bus_dma_segment_t *segs, int nseg, int error)
939 {
940 	bus_addr_t		*addr;
941 
942 	if (error)
943 		return;
944 
945 	KASSERT(nseg == 1, ("too many DMA segments, %d should be 1", nseg));
946 	addr = arg;
947 	*addr = segs->ds_addr;
948 }
949 
950 static int
951 re_allocmem(device_t dev, struct rl_softc *sc)
952 {
953 	bus_addr_t		lowaddr;
954 	bus_size_t		rx_list_size, tx_list_size;
955 	int			error;
956 	int			i;
957 
958 	rx_list_size = sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc);
959 	tx_list_size = sc->rl_ldata.rl_tx_desc_cnt * sizeof(struct rl_desc);
960 
961 	/*
962 	 * Allocate the parent bus DMA tag appropriate for PCI.
963 	 * In order to use DAC, RL_CPLUSCMD_PCI_DAC bit of RL_CPLUS_CMD
964 	 * register should be set. However some RealTek chips are known
965 	 * to be buggy on DAC handling, therefore disable DAC by limiting
966 	 * DMA address space to 32bit. PCIe variants of RealTek chips
967 	 * may not have the limitation.
968 	 */
969 	lowaddr = BUS_SPACE_MAXADDR;
970 	if ((sc->rl_flags & RL_FLAG_PCIE) == 0)
971 		lowaddr = BUS_SPACE_MAXADDR_32BIT;
972 	error = bus_dma_tag_create(bus_get_dma_tag(dev), 1, 0,
973 	    lowaddr, BUS_SPACE_MAXADDR, NULL, NULL,
974 	    BUS_SPACE_MAXSIZE_32BIT, 0, BUS_SPACE_MAXSIZE_32BIT, 0,
975 	    NULL, NULL, &sc->rl_parent_tag);
976 	if (error) {
977 		device_printf(dev, "could not allocate parent DMA tag\n");
978 		return (error);
979 	}
980 
981 	/*
982 	 * Allocate map for TX mbufs.
983 	 */
984 	error = bus_dma_tag_create(sc->rl_parent_tag, 1, 0,
985 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL,
986 	    NULL, MCLBYTES * RL_NTXSEGS, RL_NTXSEGS, 4096, 0,
987 	    NULL, NULL, &sc->rl_ldata.rl_tx_mtag);
988 	if (error) {
989 		device_printf(dev, "could not allocate TX DMA tag\n");
990 		return (error);
991 	}
992 
993 	/*
994 	 * Allocate map for RX mbufs.
995 	 */
996 
997 	if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
998 		error = bus_dma_tag_create(sc->rl_parent_tag, sizeof(uint64_t),
999 		    0, BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
1000 		    MJUM9BYTES, 1, MJUM9BYTES, 0, NULL, NULL,
1001 		    &sc->rl_ldata.rl_jrx_mtag);
1002 		if (error) {
1003 			device_printf(dev,
1004 			    "could not allocate jumbo RX DMA tag\n");
1005 			return (error);
1006 		}
1007 	}
1008 	error = bus_dma_tag_create(sc->rl_parent_tag, sizeof(uint64_t), 0,
1009 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
1010 	    MCLBYTES, 1, MCLBYTES, 0, NULL, NULL, &sc->rl_ldata.rl_rx_mtag);
1011 	if (error) {
1012 		device_printf(dev, "could not allocate RX DMA tag\n");
1013 		return (error);
1014 	}
1015 
1016 	/*
1017 	 * Allocate map for TX descriptor list.
1018 	 */
1019 	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
1020 	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1021 	    NULL, tx_list_size, 1, tx_list_size, 0,
1022 	    NULL, NULL, &sc->rl_ldata.rl_tx_list_tag);
1023 	if (error) {
1024 		device_printf(dev, "could not allocate TX DMA ring tag\n");
1025 		return (error);
1026 	}
1027 
1028 	/* Allocate DMA'able memory for the TX ring */
1029 
1030 	error = bus_dmamem_alloc(sc->rl_ldata.rl_tx_list_tag,
1031 	    (void **)&sc->rl_ldata.rl_tx_list,
1032 	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1033 	    &sc->rl_ldata.rl_tx_list_map);
1034 	if (error) {
1035 		device_printf(dev, "could not allocate TX DMA ring\n");
1036 		return (error);
1037 	}
1038 
1039 	/* Load the map for the TX ring. */
1040 
1041 	sc->rl_ldata.rl_tx_list_addr = 0;
1042 	error = bus_dmamap_load(sc->rl_ldata.rl_tx_list_tag,
1043 	     sc->rl_ldata.rl_tx_list_map, sc->rl_ldata.rl_tx_list,
1044 	     tx_list_size, re_dma_map_addr,
1045 	     &sc->rl_ldata.rl_tx_list_addr, BUS_DMA_NOWAIT);
1046 	if (error != 0 || sc->rl_ldata.rl_tx_list_addr == 0) {
1047 		device_printf(dev, "could not load TX DMA ring\n");
1048 		return (ENOMEM);
1049 	}
1050 
1051 	/* Create DMA maps for TX buffers */
1052 
1053 	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
1054 		error = bus_dmamap_create(sc->rl_ldata.rl_tx_mtag, 0,
1055 		    &sc->rl_ldata.rl_tx_desc[i].tx_dmamap);
1056 		if (error) {
1057 			device_printf(dev, "could not create DMA map for TX\n");
1058 			return (error);
1059 		}
1060 	}
1061 
1062 	/*
1063 	 * Allocate map for RX descriptor list.
1064 	 */
1065 	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
1066 	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1067 	    NULL, rx_list_size, 1, rx_list_size, 0,
1068 	    NULL, NULL, &sc->rl_ldata.rl_rx_list_tag);
1069 	if (error) {
1070 		device_printf(dev, "could not create RX DMA ring tag\n");
1071 		return (error);
1072 	}
1073 
1074 	/* Allocate DMA'able memory for the RX ring */
1075 
1076 	error = bus_dmamem_alloc(sc->rl_ldata.rl_rx_list_tag,
1077 	    (void **)&sc->rl_ldata.rl_rx_list,
1078 	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1079 	    &sc->rl_ldata.rl_rx_list_map);
1080 	if (error) {
1081 		device_printf(dev, "could not allocate RX DMA ring\n");
1082 		return (error);
1083 	}
1084 
1085 	/* Load the map for the RX ring. */
1086 
1087 	sc->rl_ldata.rl_rx_list_addr = 0;
1088 	error = bus_dmamap_load(sc->rl_ldata.rl_rx_list_tag,
1089 	     sc->rl_ldata.rl_rx_list_map, sc->rl_ldata.rl_rx_list,
1090 	     rx_list_size, re_dma_map_addr,
1091 	     &sc->rl_ldata.rl_rx_list_addr, BUS_DMA_NOWAIT);
1092 	if (error != 0 || sc->rl_ldata.rl_rx_list_addr == 0) {
1093 		device_printf(dev, "could not load RX DMA ring\n");
1094 		return (ENOMEM);
1095 	}
1096 
1097 	/* Create DMA maps for RX buffers */
1098 
1099 	if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
1100 		error = bus_dmamap_create(sc->rl_ldata.rl_jrx_mtag, 0,
1101 		    &sc->rl_ldata.rl_jrx_sparemap);
1102 		if (error) {
1103 			device_printf(dev,
1104 			    "could not create spare DMA map for jumbo RX\n");
1105 			return (error);
1106 		}
1107 		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1108 			error = bus_dmamap_create(sc->rl_ldata.rl_jrx_mtag, 0,
1109 			    &sc->rl_ldata.rl_jrx_desc[i].rx_dmamap);
1110 			if (error) {
1111 				device_printf(dev,
1112 				    "could not create DMA map for jumbo RX\n");
1113 				return (error);
1114 			}
1115 		}
1116 	}
1117 	error = bus_dmamap_create(sc->rl_ldata.rl_rx_mtag, 0,
1118 	    &sc->rl_ldata.rl_rx_sparemap);
1119 	if (error) {
1120 		device_printf(dev, "could not create spare DMA map for RX\n");
1121 		return (error);
1122 	}
1123 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1124 		error = bus_dmamap_create(sc->rl_ldata.rl_rx_mtag, 0,
1125 		    &sc->rl_ldata.rl_rx_desc[i].rx_dmamap);
1126 		if (error) {
1127 			device_printf(dev, "could not create DMA map for RX\n");
1128 			return (error);
1129 		}
1130 	}
1131 
1132 	/* Create DMA map for statistics. */
1133 	error = bus_dma_tag_create(sc->rl_parent_tag, RL_DUMP_ALIGN, 0,
1134 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
1135 	    sizeof(struct rl_stats), 1, sizeof(struct rl_stats), 0, NULL, NULL,
1136 	    &sc->rl_ldata.rl_stag);
1137 	if (error) {
1138 		device_printf(dev, "could not create statistics DMA tag\n");
1139 		return (error);
1140 	}
1141 	/* Allocate DMA'able memory for statistics. */
1142 	error = bus_dmamem_alloc(sc->rl_ldata.rl_stag,
1143 	    (void **)&sc->rl_ldata.rl_stats,
1144 	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1145 	    &sc->rl_ldata.rl_smap);
1146 	if (error) {
1147 		device_printf(dev,
1148 		    "could not allocate statistics DMA memory\n");
1149 		return (error);
1150 	}
1151 	/* Load the map for statistics. */
1152 	sc->rl_ldata.rl_stats_addr = 0;
1153 	error = bus_dmamap_load(sc->rl_ldata.rl_stag, sc->rl_ldata.rl_smap,
1154 	    sc->rl_ldata.rl_stats, sizeof(struct rl_stats), re_dma_map_addr,
1155 	     &sc->rl_ldata.rl_stats_addr, BUS_DMA_NOWAIT);
1156 	if (error != 0 || sc->rl_ldata.rl_stats_addr == 0) {
1157 		device_printf(dev, "could not load statistics DMA memory\n");
1158 		return (ENOMEM);
1159 	}
1160 
1161 	return (0);
1162 }
1163 
1164 /*
1165  * Attach the interface. Allocate softc structures, do ifmedia
1166  * setup and ethernet/BPF attach.
1167  */
1168 static int
1169 re_attach(device_t dev)
1170 {
1171 	u_char			eaddr[ETHER_ADDR_LEN];
1172 	u_int16_t		as[ETHER_ADDR_LEN / 2];
1173 	struct rl_softc		*sc;
1174 	struct ifnet		*ifp;
1175 	struct rl_hwrev		*hw_rev;
1176 	int			hwrev;
1177 	u_int16_t		devid, re_did = 0;
1178 	int			error = 0, i, phy, rid;
1179 	int			msic, reg;
1180 	uint8_t			cfg;
1181 
1182 	sc = device_get_softc(dev);
1183 	sc->rl_dev = dev;
1184 
1185 	mtx_init(&sc->rl_mtx, device_get_nameunit(dev), MTX_NETWORK_LOCK,
1186 	    MTX_DEF);
1187 	callout_init_mtx(&sc->rl_stat_callout, &sc->rl_mtx, 0);
1188 
1189 	/*
1190 	 * Map control/status registers.
1191 	 */
1192 	pci_enable_busmaster(dev);
1193 
1194 	devid = pci_get_device(dev);
1195 	/*
1196 	 * Prefer memory space register mapping over IO space.
1197 	 * Because RTL8169SC does not seem to work when memory mapping
1198 	 * is used always activate io mapping.
1199 	 */
1200 	if (devid == RT_DEVICEID_8169SC)
1201 		prefer_iomap = 1;
1202 	if (prefer_iomap == 0) {
1203 		sc->rl_res_id = PCIR_BAR(1);
1204 		sc->rl_res_type = SYS_RES_MEMORY;
1205 		/* RTL8168/8101E seems to use different BARs. */
1206 		if (devid == RT_DEVICEID_8168 || devid == RT_DEVICEID_8101E)
1207 			sc->rl_res_id = PCIR_BAR(2);
1208 	} else {
1209 		sc->rl_res_id = PCIR_BAR(0);
1210 		sc->rl_res_type = SYS_RES_IOPORT;
1211 	}
1212 	sc->rl_res = bus_alloc_resource_any(dev, sc->rl_res_type,
1213 	    &sc->rl_res_id, RF_ACTIVE);
1214 	if (sc->rl_res == NULL && prefer_iomap == 0) {
1215 		sc->rl_res_id = PCIR_BAR(0);
1216 		sc->rl_res_type = SYS_RES_IOPORT;
1217 		sc->rl_res = bus_alloc_resource_any(dev, sc->rl_res_type,
1218 		    &sc->rl_res_id, RF_ACTIVE);
1219 	}
1220 	if (sc->rl_res == NULL) {
1221 		device_printf(dev, "couldn't map ports/memory\n");
1222 		error = ENXIO;
1223 		goto fail;
1224 	}
1225 
1226 	sc->rl_btag = rman_get_bustag(sc->rl_res);
1227 	sc->rl_bhandle = rman_get_bushandle(sc->rl_res);
1228 
1229 	msic = 0;
1230 	if (pci_find_extcap(dev, PCIY_EXPRESS, &reg) == 0) {
1231 		sc->rl_flags |= RL_FLAG_PCIE;
1232 		msic = pci_msi_count(dev);
1233 		if (bootverbose)
1234 			device_printf(dev, "MSI count : %d\n", msic);
1235 	}
1236 	if (msic > 0 && msi_disable == 0) {
1237 		msic = 1;
1238 		if (pci_alloc_msi(dev, &msic) == 0) {
1239 			if (msic == RL_MSI_MESSAGES) {
1240 				device_printf(dev, "Using %d MSI messages\n",
1241 				    msic);
1242 				sc->rl_flags |= RL_FLAG_MSI;
1243 				/* Explicitly set MSI enable bit. */
1244 				CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1245 				cfg = CSR_READ_1(sc, RL_CFG2);
1246 				cfg |= RL_CFG2_MSI;
1247 				CSR_WRITE_1(sc, RL_CFG2, cfg);
1248 				CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1249 			} else
1250 				pci_release_msi(dev);
1251 		}
1252 	}
1253 
1254 	/* Allocate interrupt */
1255 	if ((sc->rl_flags & RL_FLAG_MSI) == 0) {
1256 		rid = 0;
1257 		sc->rl_irq[0] = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
1258 		    RF_SHAREABLE | RF_ACTIVE);
1259 		if (sc->rl_irq[0] == NULL) {
1260 			device_printf(dev, "couldn't allocate IRQ resources\n");
1261 			error = ENXIO;
1262 			goto fail;
1263 		}
1264 	} else {
1265 		for (i = 0, rid = 1; i < RL_MSI_MESSAGES; i++, rid++) {
1266 			sc->rl_irq[i] = bus_alloc_resource_any(dev,
1267 			    SYS_RES_IRQ, &rid, RF_ACTIVE);
1268 			if (sc->rl_irq[i] == NULL) {
1269 				device_printf(dev,
1270 				    "couldn't llocate IRQ resources for "
1271 				    "message %d\n", rid);
1272 				error = ENXIO;
1273 				goto fail;
1274 			}
1275 		}
1276 	}
1277 
1278 	if ((sc->rl_flags & RL_FLAG_MSI) == 0) {
1279 		CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1280 		cfg = CSR_READ_1(sc, RL_CFG2);
1281 		if ((cfg & RL_CFG2_MSI) != 0) {
1282 			device_printf(dev, "turning off MSI enable bit.\n");
1283 			cfg &= ~RL_CFG2_MSI;
1284 			CSR_WRITE_1(sc, RL_CFG2, cfg);
1285 		}
1286 		CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1287 	}
1288 
1289 	hw_rev = re_hwrevs;
1290 	hwrev = CSR_READ_4(sc, RL_TXCFG);
1291 	switch (hwrev & 0x70000000) {
1292 	case 0x00000000:
1293 	case 0x10000000:
1294 		device_printf(dev, "Chip rev. 0x%08x\n", hwrev & 0xfc800000);
1295 		hwrev &= (RL_TXCFG_HWREV | 0x80000000);
1296 		break;
1297 	default:
1298 		device_printf(dev, "Chip rev. 0x%08x\n", hwrev & 0x7c800000);
1299 		hwrev &= RL_TXCFG_HWREV;
1300 		break;
1301 	}
1302 	device_printf(dev, "MAC rev. 0x%08x\n", hwrev & 0x00700000);
1303 	while (hw_rev->rl_desc != NULL) {
1304 		if (hw_rev->rl_rev == hwrev) {
1305 			sc->rl_type = hw_rev->rl_type;
1306 			sc->rl_hwrev = hw_rev;
1307 			break;
1308 		}
1309 		hw_rev++;
1310 	}
1311 	if (hw_rev->rl_desc == NULL) {
1312 		device_printf(dev, "Unknown H/W revision: 0x%08x\n", hwrev);
1313 		error = ENXIO;
1314 		goto fail;
1315 	}
1316 
1317 	switch (hw_rev->rl_rev) {
1318 	case RL_HWREV_8139CPLUS:
1319 		sc->rl_flags |= RL_FLAG_FASTETHER | RL_FLAG_AUTOPAD;
1320 		break;
1321 	case RL_HWREV_8100E:
1322 	case RL_HWREV_8101E:
1323 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_FASTETHER;
1324 		break;
1325 	case RL_HWREV_8102E:
1326 	case RL_HWREV_8102EL:
1327 	case RL_HWREV_8102EL_SPIN1:
1328 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR | RL_FLAG_DESCV2 |
1329 		    RL_FLAG_MACSTAT | RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP |
1330 		    RL_FLAG_AUTOPAD;
1331 		break;
1332 	case RL_HWREV_8103E:
1333 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR | RL_FLAG_DESCV2 |
1334 		    RL_FLAG_MACSTAT | RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP |
1335 		    RL_FLAG_AUTOPAD | RL_FLAG_MACSLEEP;
1336 		break;
1337 	case RL_HWREV_8168B_SPIN1:
1338 	case RL_HWREV_8168B_SPIN2:
1339 		sc->rl_flags |= RL_FLAG_WOLRXENB;
1340 		/* FALLTHROUGH */
1341 	case RL_HWREV_8168B_SPIN3:
1342 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_MACSTAT;
1343 		break;
1344 	case RL_HWREV_8168C_SPIN2:
1345 		sc->rl_flags |= RL_FLAG_MACSLEEP;
1346 		/* FALLTHROUGH */
1347 	case RL_HWREV_8168C:
1348 		if ((hwrev & 0x00700000) == 0x00200000)
1349 			sc->rl_flags |= RL_FLAG_MACSLEEP;
1350 		/* FALLTHROUGH */
1351 	case RL_HWREV_8168CP:
1352 	case RL_HWREV_8168D:
1353 	case RL_HWREV_8168DP:
1354 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
1355 		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP |
1356 		    RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2;
1357 		break;
1358 	case RL_HWREV_8168E:
1359 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
1360 		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
1361 		    RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2;
1362 		break;
1363 	case RL_HWREV_8168E_VL:
1364 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
1365 		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP |
1366 		    RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2;
1367 		break;
1368 	case RL_HWREV_8169_8110SB:
1369 	case RL_HWREV_8169_8110SBL:
1370 	case RL_HWREV_8169_8110SC:
1371 	case RL_HWREV_8169_8110SCE:
1372 		sc->rl_flags |= RL_FLAG_PHYWAKE;
1373 		/* FALLTHROUGH */
1374 	case RL_HWREV_8169:
1375 	case RL_HWREV_8169S:
1376 	case RL_HWREV_8110S:
1377 		sc->rl_flags |= RL_FLAG_MACRESET;
1378 		break;
1379 	default:
1380 		break;
1381 	}
1382 
1383 	/* Reset the adapter. */
1384 	RL_LOCK(sc);
1385 	re_reset(sc);
1386 	RL_UNLOCK(sc);
1387 
1388 	/* Enable PME. */
1389 	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1390 	cfg = CSR_READ_1(sc, RL_CFG1);
1391 	cfg |= RL_CFG1_PME;
1392 	CSR_WRITE_1(sc, RL_CFG1, cfg);
1393 	cfg = CSR_READ_1(sc, RL_CFG5);
1394 	cfg &= RL_CFG5_PME_STS;
1395 	CSR_WRITE_1(sc, RL_CFG5, cfg);
1396 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1397 
1398 	if ((sc->rl_flags & RL_FLAG_PAR) != 0) {
1399 		/*
1400 		 * XXX Should have a better way to extract station
1401 		 * address from EEPROM.
1402 		 */
1403 		for (i = 0; i < ETHER_ADDR_LEN; i++)
1404 			eaddr[i] = CSR_READ_1(sc, RL_IDR0 + i);
1405 	} else {
1406 		sc->rl_eewidth = RL_9356_ADDR_LEN;
1407 		re_read_eeprom(sc, (caddr_t)&re_did, 0, 1);
1408 		if (re_did != 0x8129)
1409 			sc->rl_eewidth = RL_9346_ADDR_LEN;
1410 
1411 		/*
1412 		 * Get station address from the EEPROM.
1413 		 */
1414 		re_read_eeprom(sc, (caddr_t)as, RL_EE_EADDR, 3);
1415 		for (i = 0; i < ETHER_ADDR_LEN / 2; i++)
1416 			as[i] = le16toh(as[i]);
1417 		bcopy(as, eaddr, sizeof(eaddr));
1418 	}
1419 
1420 	if (sc->rl_type == RL_8169) {
1421 		/* Set RX length mask and number of descriptors. */
1422 		sc->rl_rxlenmask = RL_RDESC_STAT_GFRAGLEN;
1423 		sc->rl_txstart = RL_GTXSTART;
1424 		sc->rl_ldata.rl_tx_desc_cnt = RL_8169_TX_DESC_CNT;
1425 		sc->rl_ldata.rl_rx_desc_cnt = RL_8169_RX_DESC_CNT;
1426 	} else {
1427 		/* Set RX length mask and number of descriptors. */
1428 		sc->rl_rxlenmask = RL_RDESC_STAT_FRAGLEN;
1429 		sc->rl_txstart = RL_TXSTART;
1430 		sc->rl_ldata.rl_tx_desc_cnt = RL_8139_TX_DESC_CNT;
1431 		sc->rl_ldata.rl_rx_desc_cnt = RL_8139_RX_DESC_CNT;
1432 	}
1433 
1434 	error = re_allocmem(dev, sc);
1435 	if (error)
1436 		goto fail;
1437 	re_add_sysctls(sc);
1438 
1439 	ifp = sc->rl_ifp = if_alloc(IFT_ETHER);
1440 	if (ifp == NULL) {
1441 		device_printf(dev, "can not if_alloc()\n");
1442 		error = ENOSPC;
1443 		goto fail;
1444 	}
1445 
1446 	/* Take controller out of deep sleep mode. */
1447 	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
1448 		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
1449 			CSR_WRITE_1(sc, RL_GPIO,
1450 			    CSR_READ_1(sc, RL_GPIO) | 0x01);
1451 		else
1452 			CSR_WRITE_1(sc, RL_GPIO,
1453 			    CSR_READ_1(sc, RL_GPIO) & ~0x01);
1454 	}
1455 
1456 	/* Take PHY out of power down mode. */
1457 	if ((sc->rl_flags & RL_FLAG_PHYWAKE_PM) != 0)
1458 		CSR_WRITE_1(sc, RL_PMCH, CSR_READ_1(sc, RL_PMCH) | 0x80);
1459 	if ((sc->rl_flags & RL_FLAG_PHYWAKE) != 0) {
1460 		re_gmii_writereg(dev, 1, 0x1f, 0);
1461 		re_gmii_writereg(dev, 1, 0x0e, 0);
1462 	}
1463 
1464 #define	RE_PHYAD_INTERNAL	 0
1465 
1466 	/* Do MII setup. */
1467 	phy = RE_PHYAD_INTERNAL;
1468 	if (sc->rl_type == RL_8169)
1469 		phy = 1;
1470 	error = mii_attach(dev, &sc->rl_miibus, ifp, re_ifmedia_upd,
1471 	    re_ifmedia_sts, BMSR_DEFCAPMASK, phy, MII_OFFSET_ANY, MIIF_DOPAUSE);
1472 	if (error != 0) {
1473 		device_printf(dev, "attaching PHYs failed\n");
1474 		goto fail;
1475 	}
1476 
1477 	ifp->if_softc = sc;
1478 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1479 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1480 	ifp->if_ioctl = re_ioctl;
1481 	ifp->if_start = re_start;
1482 	ifp->if_hwassist = RE_CSUM_FEATURES | CSUM_TSO;
1483 	ifp->if_capabilities = IFCAP_HWCSUM | IFCAP_TSO4;
1484 	ifp->if_capenable = ifp->if_capabilities;
1485 	ifp->if_init = re_init;
1486 	IFQ_SET_MAXLEN(&ifp->if_snd, RL_IFQ_MAXLEN);
1487 	ifp->if_snd.ifq_drv_maxlen = RL_IFQ_MAXLEN;
1488 	IFQ_SET_READY(&ifp->if_snd);
1489 
1490 	TASK_INIT(&sc->rl_txtask, 1, re_tx_task, ifp);
1491 	TASK_INIT(&sc->rl_inttask, 0, re_int_task, sc);
1492 
1493 	/*
1494 	 * Call MI attach routine.
1495 	 */
1496 	ether_ifattach(ifp, eaddr);
1497 
1498 	/* VLAN capability setup */
1499 	ifp->if_capabilities |= IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING;
1500 	if (ifp->if_capabilities & IFCAP_HWCSUM)
1501 		ifp->if_capabilities |= IFCAP_VLAN_HWCSUM;
1502 	/* Enable WOL if PM is supported. */
1503 	if (pci_find_extcap(sc->rl_dev, PCIY_PMG, &reg) == 0)
1504 		ifp->if_capabilities |= IFCAP_WOL;
1505 	ifp->if_capenable = ifp->if_capabilities;
1506 	/*
1507 	 * Don't enable TSO by default for old controllers. Under
1508 	 * certain circumtances the controller generated corrupted
1509 	 * packets in TSO size.
1510 	 */
1511 	if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
1512 		ifp->if_hwassist &= ~CSUM_TSO;
1513 		ifp->if_capenable &= ~(IFCAP_TSO4 | IFCAP_VLAN_HWTSO);
1514 	}
1515 #ifdef DEVICE_POLLING
1516 	ifp->if_capabilities |= IFCAP_POLLING;
1517 #endif
1518 	/*
1519 	 * Tell the upper layer(s) we support long frames.
1520 	 * Must appear after the call to ether_ifattach() because
1521 	 * ether_ifattach() sets ifi_hdrlen to the default value.
1522 	 */
1523 	ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
1524 
1525 #ifdef RE_DIAG
1526 	/*
1527 	 * Perform hardware diagnostic on the original RTL8169.
1528 	 * Some 32-bit cards were incorrectly wired and would
1529 	 * malfunction if plugged into a 64-bit slot.
1530 	 */
1531 
1532 	if (hwrev == RL_HWREV_8169) {
1533 		error = re_diag(sc);
1534 		if (error) {
1535 			device_printf(dev,
1536 		    	"attach aborted due to hardware diag failure\n");
1537 			ether_ifdetach(ifp);
1538 			goto fail;
1539 		}
1540 	}
1541 #endif
1542 
1543 	/* Hook interrupt last to avoid having to lock softc */
1544 	if ((sc->rl_flags & RL_FLAG_MSI) == 0)
1545 		error = bus_setup_intr(dev, sc->rl_irq[0],
1546 		    INTR_TYPE_NET | INTR_MPSAFE, re_intr, NULL, sc,
1547 		    &sc->rl_intrhand[0]);
1548 	else {
1549 		for (i = 0; i < RL_MSI_MESSAGES; i++) {
1550 			error = bus_setup_intr(dev, sc->rl_irq[i],
1551 			    INTR_TYPE_NET | INTR_MPSAFE, re_intr, NULL, sc,
1552 		    	    &sc->rl_intrhand[i]);
1553 			if (error != 0)
1554 				break;
1555 		}
1556 	}
1557 	if (error) {
1558 		device_printf(dev, "couldn't set up irq\n");
1559 		ether_ifdetach(ifp);
1560 	}
1561 
1562 fail:
1563 
1564 	if (error)
1565 		re_detach(dev);
1566 
1567 	return (error);
1568 }
1569 
1570 /*
1571  * Shutdown hardware and free up resources. This can be called any
1572  * time after the mutex has been initialized. It is called in both
1573  * the error case in attach and the normal detach case so it needs
1574  * to be careful about only freeing resources that have actually been
1575  * allocated.
1576  */
1577 static int
1578 re_detach(device_t dev)
1579 {
1580 	struct rl_softc		*sc;
1581 	struct ifnet		*ifp;
1582 	int			i, rid;
1583 
1584 	sc = device_get_softc(dev);
1585 	ifp = sc->rl_ifp;
1586 	KASSERT(mtx_initialized(&sc->rl_mtx), ("re mutex not initialized"));
1587 
1588 	/* These should only be active if attach succeeded */
1589 	if (device_is_attached(dev)) {
1590 #ifdef DEVICE_POLLING
1591 		if (ifp->if_capenable & IFCAP_POLLING)
1592 			ether_poll_deregister(ifp);
1593 #endif
1594 		RL_LOCK(sc);
1595 #if 0
1596 		sc->suspended = 1;
1597 #endif
1598 		re_stop(sc);
1599 		RL_UNLOCK(sc);
1600 		callout_drain(&sc->rl_stat_callout);
1601 		taskqueue_drain(taskqueue_fast, &sc->rl_inttask);
1602 		taskqueue_drain(taskqueue_fast, &sc->rl_txtask);
1603 		/*
1604 		 * Force off the IFF_UP flag here, in case someone
1605 		 * still had a BPF descriptor attached to this
1606 		 * interface. If they do, ether_ifdetach() will cause
1607 		 * the BPF code to try and clear the promisc mode
1608 		 * flag, which will bubble down to re_ioctl(),
1609 		 * which will try to call re_init() again. This will
1610 		 * turn the NIC back on and restart the MII ticker,
1611 		 * which will panic the system when the kernel tries
1612 		 * to invoke the re_tick() function that isn't there
1613 		 * anymore.
1614 		 */
1615 		ifp->if_flags &= ~IFF_UP;
1616 		ether_ifdetach(ifp);
1617 	}
1618 	if (sc->rl_miibus)
1619 		device_delete_child(dev, sc->rl_miibus);
1620 	bus_generic_detach(dev);
1621 
1622 	/*
1623 	 * The rest is resource deallocation, so we should already be
1624 	 * stopped here.
1625 	 */
1626 
1627 	for (i = 0; i < RL_MSI_MESSAGES; i++) {
1628 		if (sc->rl_intrhand[i] != NULL) {
1629 			bus_teardown_intr(dev, sc->rl_irq[i],
1630 			    sc->rl_intrhand[i]);
1631 			sc->rl_intrhand[i] = NULL;
1632 		}
1633 	}
1634 	if (ifp != NULL)
1635 		if_free(ifp);
1636 	if ((sc->rl_flags & RL_FLAG_MSI) == 0) {
1637 		if (sc->rl_irq[0] != NULL) {
1638 			bus_release_resource(dev, SYS_RES_IRQ, 0,
1639 			    sc->rl_irq[0]);
1640 			sc->rl_irq[0] = NULL;
1641 		}
1642 	} else {
1643 		for (i = 0, rid = 1; i < RL_MSI_MESSAGES; i++, rid++) {
1644 			if (sc->rl_irq[i] != NULL) {
1645 				bus_release_resource(dev, SYS_RES_IRQ, rid,
1646 				    sc->rl_irq[i]);
1647 				sc->rl_irq[i] = NULL;
1648 			}
1649 		}
1650 		pci_release_msi(dev);
1651 	}
1652 	if (sc->rl_res)
1653 		bus_release_resource(dev, sc->rl_res_type, sc->rl_res_id,
1654 		    sc->rl_res);
1655 
1656 	/* Unload and free the RX DMA ring memory and map */
1657 
1658 	if (sc->rl_ldata.rl_rx_list_tag) {
1659 		if (sc->rl_ldata.rl_rx_list_map)
1660 			bus_dmamap_unload(sc->rl_ldata.rl_rx_list_tag,
1661 			    sc->rl_ldata.rl_rx_list_map);
1662 		if (sc->rl_ldata.rl_rx_list_map && sc->rl_ldata.rl_rx_list)
1663 			bus_dmamem_free(sc->rl_ldata.rl_rx_list_tag,
1664 			    sc->rl_ldata.rl_rx_list,
1665 			    sc->rl_ldata.rl_rx_list_map);
1666 		bus_dma_tag_destroy(sc->rl_ldata.rl_rx_list_tag);
1667 	}
1668 
1669 	/* Unload and free the TX DMA ring memory and map */
1670 
1671 	if (sc->rl_ldata.rl_tx_list_tag) {
1672 		if (sc->rl_ldata.rl_tx_list_map)
1673 			bus_dmamap_unload(sc->rl_ldata.rl_tx_list_tag,
1674 			    sc->rl_ldata.rl_tx_list_map);
1675 		if (sc->rl_ldata.rl_tx_list_map && sc->rl_ldata.rl_tx_list)
1676 			bus_dmamem_free(sc->rl_ldata.rl_tx_list_tag,
1677 			    sc->rl_ldata.rl_tx_list,
1678 			    sc->rl_ldata.rl_tx_list_map);
1679 		bus_dma_tag_destroy(sc->rl_ldata.rl_tx_list_tag);
1680 	}
1681 
1682 	/* Destroy all the RX and TX buffer maps */
1683 
1684 	if (sc->rl_ldata.rl_tx_mtag) {
1685 		for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
1686 			if (sc->rl_ldata.rl_tx_desc[i].tx_dmamap)
1687 				bus_dmamap_destroy(sc->rl_ldata.rl_tx_mtag,
1688 				    sc->rl_ldata.rl_tx_desc[i].tx_dmamap);
1689 		}
1690 		bus_dma_tag_destroy(sc->rl_ldata.rl_tx_mtag);
1691 	}
1692 	if (sc->rl_ldata.rl_rx_mtag) {
1693 		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1694 			if (sc->rl_ldata.rl_rx_desc[i].rx_dmamap)
1695 				bus_dmamap_destroy(sc->rl_ldata.rl_rx_mtag,
1696 				    sc->rl_ldata.rl_rx_desc[i].rx_dmamap);
1697 		}
1698 		if (sc->rl_ldata.rl_rx_sparemap)
1699 			bus_dmamap_destroy(sc->rl_ldata.rl_rx_mtag,
1700 			    sc->rl_ldata.rl_rx_sparemap);
1701 		bus_dma_tag_destroy(sc->rl_ldata.rl_rx_mtag);
1702 	}
1703 	if (sc->rl_ldata.rl_jrx_mtag) {
1704 		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1705 			if (sc->rl_ldata.rl_jrx_desc[i].rx_dmamap)
1706 				bus_dmamap_destroy(sc->rl_ldata.rl_jrx_mtag,
1707 				    sc->rl_ldata.rl_jrx_desc[i].rx_dmamap);
1708 		}
1709 		if (sc->rl_ldata.rl_jrx_sparemap)
1710 			bus_dmamap_destroy(sc->rl_ldata.rl_jrx_mtag,
1711 			    sc->rl_ldata.rl_jrx_sparemap);
1712 		bus_dma_tag_destroy(sc->rl_ldata.rl_jrx_mtag);
1713 	}
1714 	/* Unload and free the stats buffer and map */
1715 
1716 	if (sc->rl_ldata.rl_stag) {
1717 		if (sc->rl_ldata.rl_smap)
1718 			bus_dmamap_unload(sc->rl_ldata.rl_stag,
1719 			    sc->rl_ldata.rl_smap);
1720 		if (sc->rl_ldata.rl_smap && sc->rl_ldata.rl_stats)
1721 			bus_dmamem_free(sc->rl_ldata.rl_stag,
1722 			    sc->rl_ldata.rl_stats, sc->rl_ldata.rl_smap);
1723 		bus_dma_tag_destroy(sc->rl_ldata.rl_stag);
1724 	}
1725 
1726 	if (sc->rl_parent_tag)
1727 		bus_dma_tag_destroy(sc->rl_parent_tag);
1728 
1729 	mtx_destroy(&sc->rl_mtx);
1730 
1731 	return (0);
1732 }
1733 
1734 static __inline void
1735 re_discard_rxbuf(struct rl_softc *sc, int idx)
1736 {
1737 	struct rl_desc		*desc;
1738 	struct rl_rxdesc	*rxd;
1739 	uint32_t		cmdstat;
1740 
1741 	if (sc->rl_ifp->if_mtu > RL_MTU &&
1742 	    (sc->rl_flags & RL_FLAG_JUMBOV2) != 0)
1743 		rxd = &sc->rl_ldata.rl_jrx_desc[idx];
1744 	else
1745 		rxd = &sc->rl_ldata.rl_rx_desc[idx];
1746 	desc = &sc->rl_ldata.rl_rx_list[idx];
1747 	desc->rl_vlanctl = 0;
1748 	cmdstat = rxd->rx_size;
1749 	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
1750 		cmdstat |= RL_RDESC_CMD_EOR;
1751 	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
1752 }
1753 
1754 static int
1755 re_newbuf(struct rl_softc *sc, int idx)
1756 {
1757 	struct mbuf		*m;
1758 	struct rl_rxdesc	*rxd;
1759 	bus_dma_segment_t	segs[1];
1760 	bus_dmamap_t		map;
1761 	struct rl_desc		*desc;
1762 	uint32_t		cmdstat;
1763 	int			error, nsegs;
1764 
1765 	m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
1766 	if (m == NULL)
1767 		return (ENOBUFS);
1768 
1769 	m->m_len = m->m_pkthdr.len = MCLBYTES;
1770 #ifdef RE_FIXUP_RX
1771 	/*
1772 	 * This is part of an evil trick to deal with non-x86 platforms.
1773 	 * The RealTek chip requires RX buffers to be aligned on 64-bit
1774 	 * boundaries, but that will hose non-x86 machines. To get around
1775 	 * this, we leave some empty space at the start of each buffer
1776 	 * and for non-x86 hosts, we copy the buffer back six bytes
1777 	 * to achieve word alignment. This is slightly more efficient
1778 	 * than allocating a new buffer, copying the contents, and
1779 	 * discarding the old buffer.
1780 	 */
1781 	m_adj(m, RE_ETHER_ALIGN);
1782 #endif
1783 	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_rx_mtag,
1784 	    sc->rl_ldata.rl_rx_sparemap, m, segs, &nsegs, BUS_DMA_NOWAIT);
1785 	if (error != 0) {
1786 		m_freem(m);
1787 		return (ENOBUFS);
1788 	}
1789 	KASSERT(nsegs == 1, ("%s: %d segment returned!", __func__, nsegs));
1790 
1791 	rxd = &sc->rl_ldata.rl_rx_desc[idx];
1792 	if (rxd->rx_m != NULL) {
1793 		bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap,
1794 		    BUS_DMASYNC_POSTREAD);
1795 		bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap);
1796 	}
1797 
1798 	rxd->rx_m = m;
1799 	map = rxd->rx_dmamap;
1800 	rxd->rx_dmamap = sc->rl_ldata.rl_rx_sparemap;
1801 	rxd->rx_size = segs[0].ds_len;
1802 	sc->rl_ldata.rl_rx_sparemap = map;
1803 	bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap,
1804 	    BUS_DMASYNC_PREREAD);
1805 
1806 	desc = &sc->rl_ldata.rl_rx_list[idx];
1807 	desc->rl_vlanctl = 0;
1808 	desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[0].ds_addr));
1809 	desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[0].ds_addr));
1810 	cmdstat = segs[0].ds_len;
1811 	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
1812 		cmdstat |= RL_RDESC_CMD_EOR;
1813 	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
1814 
1815 	return (0);
1816 }
1817 
1818 static int
1819 re_jumbo_newbuf(struct rl_softc *sc, int idx)
1820 {
1821 	struct mbuf		*m;
1822 	struct rl_rxdesc	*rxd;
1823 	bus_dma_segment_t	segs[1];
1824 	bus_dmamap_t		map;
1825 	struct rl_desc		*desc;
1826 	uint32_t		cmdstat;
1827 	int			error, nsegs;
1828 
1829 	m = m_getjcl(M_DONTWAIT, MT_DATA, M_PKTHDR, MJUM9BYTES);
1830 	if (m == NULL)
1831 		return (ENOBUFS);
1832 	m->m_len = m->m_pkthdr.len = MJUM9BYTES;
1833 #ifdef RE_FIXUP_RX
1834 	m_adj(m, RE_ETHER_ALIGN);
1835 #endif
1836 	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_jrx_mtag,
1837 	    sc->rl_ldata.rl_jrx_sparemap, m, segs, &nsegs, BUS_DMA_NOWAIT);
1838 	if (error != 0) {
1839 		m_freem(m);
1840 		return (ENOBUFS);
1841 	}
1842 	KASSERT(nsegs == 1, ("%s: %d segment returned!", __func__, nsegs));
1843 
1844 	rxd = &sc->rl_ldata.rl_jrx_desc[idx];
1845 	if (rxd->rx_m != NULL) {
1846 		bus_dmamap_sync(sc->rl_ldata.rl_jrx_mtag, rxd->rx_dmamap,
1847 		    BUS_DMASYNC_POSTREAD);
1848 		bus_dmamap_unload(sc->rl_ldata.rl_jrx_mtag, rxd->rx_dmamap);
1849 	}
1850 
1851 	rxd->rx_m = m;
1852 	map = rxd->rx_dmamap;
1853 	rxd->rx_dmamap = sc->rl_ldata.rl_jrx_sparemap;
1854 	rxd->rx_size = segs[0].ds_len;
1855 	sc->rl_ldata.rl_jrx_sparemap = map;
1856 	bus_dmamap_sync(sc->rl_ldata.rl_jrx_mtag, rxd->rx_dmamap,
1857 	    BUS_DMASYNC_PREREAD);
1858 
1859 	desc = &sc->rl_ldata.rl_rx_list[idx];
1860 	desc->rl_vlanctl = 0;
1861 	desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[0].ds_addr));
1862 	desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[0].ds_addr));
1863 	cmdstat = segs[0].ds_len;
1864 	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
1865 		cmdstat |= RL_RDESC_CMD_EOR;
1866 	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
1867 
1868 	return (0);
1869 }
1870 
1871 #ifdef RE_FIXUP_RX
1872 static __inline void
1873 re_fixup_rx(struct mbuf *m)
1874 {
1875 	int                     i;
1876 	uint16_t                *src, *dst;
1877 
1878 	src = mtod(m, uint16_t *);
1879 	dst = src - (RE_ETHER_ALIGN - ETHER_ALIGN) / sizeof *src;
1880 
1881 	for (i = 0; i < (m->m_len / sizeof(uint16_t) + 1); i++)
1882 		*dst++ = *src++;
1883 
1884 	m->m_data -= RE_ETHER_ALIGN - ETHER_ALIGN;
1885 }
1886 #endif
1887 
1888 static int
1889 re_tx_list_init(struct rl_softc *sc)
1890 {
1891 	struct rl_desc		*desc;
1892 	int			i;
1893 
1894 	RL_LOCK_ASSERT(sc);
1895 
1896 	bzero(sc->rl_ldata.rl_tx_list,
1897 	    sc->rl_ldata.rl_tx_desc_cnt * sizeof(struct rl_desc));
1898 	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++)
1899 		sc->rl_ldata.rl_tx_desc[i].tx_m = NULL;
1900 	/* Set EOR. */
1901 	desc = &sc->rl_ldata.rl_tx_list[sc->rl_ldata.rl_tx_desc_cnt - 1];
1902 	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_EOR);
1903 
1904 	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
1905 	    sc->rl_ldata.rl_tx_list_map,
1906 	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1907 
1908 	sc->rl_ldata.rl_tx_prodidx = 0;
1909 	sc->rl_ldata.rl_tx_considx = 0;
1910 	sc->rl_ldata.rl_tx_free = sc->rl_ldata.rl_tx_desc_cnt;
1911 
1912 	return (0);
1913 }
1914 
1915 static int
1916 re_rx_list_init(struct rl_softc *sc)
1917 {
1918 	int			error, i;
1919 
1920 	bzero(sc->rl_ldata.rl_rx_list,
1921 	    sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc));
1922 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1923 		sc->rl_ldata.rl_rx_desc[i].rx_m = NULL;
1924 		if ((error = re_newbuf(sc, i)) != 0)
1925 			return (error);
1926 	}
1927 
1928 	/* Flush the RX descriptors */
1929 
1930 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1931 	    sc->rl_ldata.rl_rx_list_map,
1932 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
1933 
1934 	sc->rl_ldata.rl_rx_prodidx = 0;
1935 	sc->rl_head = sc->rl_tail = NULL;
1936 
1937 	return (0);
1938 }
1939 
1940 static int
1941 re_jrx_list_init(struct rl_softc *sc)
1942 {
1943 	int			error, i;
1944 
1945 	bzero(sc->rl_ldata.rl_rx_list,
1946 	    sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc));
1947 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1948 		sc->rl_ldata.rl_jrx_desc[i].rx_m = NULL;
1949 		if ((error = re_jumbo_newbuf(sc, i)) != 0)
1950 			return (error);
1951 	}
1952 
1953 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1954 	    sc->rl_ldata.rl_rx_list_map,
1955 	    BUS_DMASYNC_PREWRITE | BUS_DMASYNC_PREREAD);
1956 
1957 	sc->rl_ldata.rl_rx_prodidx = 0;
1958 	sc->rl_head = sc->rl_tail = NULL;
1959 
1960 	return (0);
1961 }
1962 
1963 /*
1964  * RX handler for C+ and 8169. For the gigE chips, we support
1965  * the reception of jumbo frames that have been fragmented
1966  * across multiple 2K mbuf cluster buffers.
1967  */
1968 static int
1969 re_rxeof(struct rl_softc *sc, int *rx_npktsp)
1970 {
1971 	struct mbuf		*m;
1972 	struct ifnet		*ifp;
1973 	int			i, rxerr, total_len;
1974 	struct rl_desc		*cur_rx;
1975 	u_int32_t		rxstat, rxvlan;
1976 	int			jumbo, maxpkt = 16, rx_npkts = 0;
1977 
1978 	RL_LOCK_ASSERT(sc);
1979 
1980 	ifp = sc->rl_ifp;
1981 	if (ifp->if_mtu > RL_MTU && (sc->rl_flags & RL_FLAG_JUMBOV2) != 0)
1982 		jumbo = 1;
1983 	else
1984 		jumbo = 0;
1985 
1986 	/* Invalidate the descriptor memory */
1987 
1988 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1989 	    sc->rl_ldata.rl_rx_list_map,
1990 	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
1991 
1992 	for (i = sc->rl_ldata.rl_rx_prodidx; maxpkt > 0;
1993 	    i = RL_RX_DESC_NXT(sc, i)) {
1994 		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1995 			break;
1996 		cur_rx = &sc->rl_ldata.rl_rx_list[i];
1997 		rxstat = le32toh(cur_rx->rl_cmdstat);
1998 		if ((rxstat & RL_RDESC_STAT_OWN) != 0)
1999 			break;
2000 		total_len = rxstat & sc->rl_rxlenmask;
2001 		rxvlan = le32toh(cur_rx->rl_vlanctl);
2002 		if (jumbo != 0)
2003 			m = sc->rl_ldata.rl_jrx_desc[i].rx_m;
2004 		else
2005 			m = sc->rl_ldata.rl_rx_desc[i].rx_m;
2006 
2007 		if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0 &&
2008 		    (rxstat & (RL_RDESC_STAT_SOF | RL_RDESC_STAT_EOF)) !=
2009 		    (RL_RDESC_STAT_SOF | RL_RDESC_STAT_EOF)) {
2010 			/*
2011 			 * RTL8168C or later controllers do not
2012 			 * support multi-fragment packet.
2013 			 */
2014 			re_discard_rxbuf(sc, i);
2015 			continue;
2016 		} else if ((rxstat & RL_RDESC_STAT_EOF) == 0) {
2017 			if (re_newbuf(sc, i) != 0) {
2018 				/*
2019 				 * If this is part of a multi-fragment packet,
2020 				 * discard all the pieces.
2021 				 */
2022 				if (sc->rl_head != NULL) {
2023 					m_freem(sc->rl_head);
2024 					sc->rl_head = sc->rl_tail = NULL;
2025 				}
2026 				re_discard_rxbuf(sc, i);
2027 				continue;
2028 			}
2029 			m->m_len = RE_RX_DESC_BUFLEN;
2030 			if (sc->rl_head == NULL)
2031 				sc->rl_head = sc->rl_tail = m;
2032 			else {
2033 				m->m_flags &= ~M_PKTHDR;
2034 				sc->rl_tail->m_next = m;
2035 				sc->rl_tail = m;
2036 			}
2037 			continue;
2038 		}
2039 
2040 		/*
2041 		 * NOTE: for the 8139C+, the frame length field
2042 		 * is always 12 bits in size, but for the gigE chips,
2043 		 * it is 13 bits (since the max RX frame length is 16K).
2044 		 * Unfortunately, all 32 bits in the status word
2045 		 * were already used, so to make room for the extra
2046 		 * length bit, RealTek took out the 'frame alignment
2047 		 * error' bit and shifted the other status bits
2048 		 * over one slot. The OWN, EOR, FS and LS bits are
2049 		 * still in the same places. We have already extracted
2050 		 * the frame length and checked the OWN bit, so rather
2051 		 * than using an alternate bit mapping, we shift the
2052 		 * status bits one space to the right so we can evaluate
2053 		 * them using the 8169 status as though it was in the
2054 		 * same format as that of the 8139C+.
2055 		 */
2056 		if (sc->rl_type == RL_8169)
2057 			rxstat >>= 1;
2058 
2059 		/*
2060 		 * if total_len > 2^13-1, both _RXERRSUM and _GIANT will be
2061 		 * set, but if CRC is clear, it will still be a valid frame.
2062 		 */
2063 		if ((rxstat & RL_RDESC_STAT_RXERRSUM) != 0) {
2064 			rxerr = 1;
2065 			if ((sc->rl_flags & RL_FLAG_JUMBOV2) == 0 &&
2066 			    total_len > 8191 &&
2067 			    (rxstat & RL_RDESC_STAT_ERRS) == RL_RDESC_STAT_GIANT)
2068 				rxerr = 0;
2069 			if (rxerr != 0) {
2070 				ifp->if_ierrors++;
2071 				/*
2072 				 * If this is part of a multi-fragment packet,
2073 				 * discard all the pieces.
2074 				 */
2075 				if (sc->rl_head != NULL) {
2076 					m_freem(sc->rl_head);
2077 					sc->rl_head = sc->rl_tail = NULL;
2078 				}
2079 				re_discard_rxbuf(sc, i);
2080 				continue;
2081 			}
2082 		}
2083 
2084 		/*
2085 		 * If allocating a replacement mbuf fails,
2086 		 * reload the current one.
2087 		 */
2088 		if (jumbo != 0)
2089 			rxerr = re_jumbo_newbuf(sc, i);
2090 		else
2091 			rxerr = re_newbuf(sc, i);
2092 		if (rxerr != 0) {
2093 			ifp->if_iqdrops++;
2094 			if (sc->rl_head != NULL) {
2095 				m_freem(sc->rl_head);
2096 				sc->rl_head = sc->rl_tail = NULL;
2097 			}
2098 			re_discard_rxbuf(sc, i);
2099 			continue;
2100 		}
2101 
2102 		if (sc->rl_head != NULL) {
2103 			if (jumbo != 0)
2104 				m->m_len = total_len;
2105 			else {
2106 				m->m_len = total_len % RE_RX_DESC_BUFLEN;
2107 				if (m->m_len == 0)
2108 					m->m_len = RE_RX_DESC_BUFLEN;
2109 			}
2110 			/*
2111 			 * Special case: if there's 4 bytes or less
2112 			 * in this buffer, the mbuf can be discarded:
2113 			 * the last 4 bytes is the CRC, which we don't
2114 			 * care about anyway.
2115 			 */
2116 			if (m->m_len <= ETHER_CRC_LEN) {
2117 				sc->rl_tail->m_len -=
2118 				    (ETHER_CRC_LEN - m->m_len);
2119 				m_freem(m);
2120 			} else {
2121 				m->m_len -= ETHER_CRC_LEN;
2122 				m->m_flags &= ~M_PKTHDR;
2123 				sc->rl_tail->m_next = m;
2124 			}
2125 			m = sc->rl_head;
2126 			sc->rl_head = sc->rl_tail = NULL;
2127 			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
2128 		} else
2129 			m->m_pkthdr.len = m->m_len =
2130 			    (total_len - ETHER_CRC_LEN);
2131 
2132 #ifdef RE_FIXUP_RX
2133 		re_fixup_rx(m);
2134 #endif
2135 		ifp->if_ipackets++;
2136 		m->m_pkthdr.rcvif = ifp;
2137 
2138 		/* Do RX checksumming if enabled */
2139 
2140 		if (ifp->if_capenable & IFCAP_RXCSUM) {
2141 			if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
2142 				/* Check IP header checksum */
2143 				if (rxstat & RL_RDESC_STAT_PROTOID)
2144 					m->m_pkthdr.csum_flags |=
2145 					    CSUM_IP_CHECKED;
2146 				if (!(rxstat & RL_RDESC_STAT_IPSUMBAD))
2147 					m->m_pkthdr.csum_flags |=
2148 					    CSUM_IP_VALID;
2149 
2150 				/* Check TCP/UDP checksum */
2151 				if ((RL_TCPPKT(rxstat) &&
2152 				    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
2153 				    (RL_UDPPKT(rxstat) &&
2154 				     !(rxstat & RL_RDESC_STAT_UDPSUMBAD))) {
2155 					m->m_pkthdr.csum_flags |=
2156 						CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
2157 					m->m_pkthdr.csum_data = 0xffff;
2158 				}
2159 			} else {
2160 				/*
2161 				 * RTL8168C/RTL816CP/RTL8111C/RTL8111CP
2162 				 */
2163 				if ((rxstat & RL_RDESC_STAT_PROTOID) &&
2164 				    (rxvlan & RL_RDESC_IPV4))
2165 					m->m_pkthdr.csum_flags |=
2166 					    CSUM_IP_CHECKED;
2167 				if (!(rxstat & RL_RDESC_STAT_IPSUMBAD) &&
2168 				    (rxvlan & RL_RDESC_IPV4))
2169 					m->m_pkthdr.csum_flags |=
2170 					    CSUM_IP_VALID;
2171 				if (((rxstat & RL_RDESC_STAT_TCP) &&
2172 				    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
2173 				    ((rxstat & RL_RDESC_STAT_UDP) &&
2174 				    !(rxstat & RL_RDESC_STAT_UDPSUMBAD))) {
2175 					m->m_pkthdr.csum_flags |=
2176 						CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
2177 					m->m_pkthdr.csum_data = 0xffff;
2178 				}
2179 			}
2180 		}
2181 		maxpkt--;
2182 		if (rxvlan & RL_RDESC_VLANCTL_TAG) {
2183 			m->m_pkthdr.ether_vtag =
2184 			    bswap16((rxvlan & RL_RDESC_VLANCTL_DATA));
2185 			m->m_flags |= M_VLANTAG;
2186 		}
2187 		RL_UNLOCK(sc);
2188 		(*ifp->if_input)(ifp, m);
2189 		RL_LOCK(sc);
2190 		rx_npkts++;
2191 	}
2192 
2193 	/* Flush the RX DMA ring */
2194 
2195 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
2196 	    sc->rl_ldata.rl_rx_list_map,
2197 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2198 
2199 	sc->rl_ldata.rl_rx_prodidx = i;
2200 
2201 	if (rx_npktsp != NULL)
2202 		*rx_npktsp = rx_npkts;
2203 	if (maxpkt)
2204 		return (EAGAIN);
2205 
2206 	return (0);
2207 }
2208 
2209 static void
2210 re_txeof(struct rl_softc *sc)
2211 {
2212 	struct ifnet		*ifp;
2213 	struct rl_txdesc	*txd;
2214 	u_int32_t		txstat;
2215 	int			cons;
2216 
2217 	cons = sc->rl_ldata.rl_tx_considx;
2218 	if (cons == sc->rl_ldata.rl_tx_prodidx)
2219 		return;
2220 
2221 	ifp = sc->rl_ifp;
2222 	/* Invalidate the TX descriptor list */
2223 	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2224 	    sc->rl_ldata.rl_tx_list_map,
2225 	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2226 
2227 	for (; cons != sc->rl_ldata.rl_tx_prodidx;
2228 	    cons = RL_TX_DESC_NXT(sc, cons)) {
2229 		txstat = le32toh(sc->rl_ldata.rl_tx_list[cons].rl_cmdstat);
2230 		if (txstat & RL_TDESC_STAT_OWN)
2231 			break;
2232 		/*
2233 		 * We only stash mbufs in the last descriptor
2234 		 * in a fragment chain, which also happens to
2235 		 * be the only place where the TX status bits
2236 		 * are valid.
2237 		 */
2238 		if (txstat & RL_TDESC_CMD_EOF) {
2239 			txd = &sc->rl_ldata.rl_tx_desc[cons];
2240 			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
2241 			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
2242 			bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag,
2243 			    txd->tx_dmamap);
2244 			KASSERT(txd->tx_m != NULL,
2245 			    ("%s: freeing NULL mbufs!", __func__));
2246 			m_freem(txd->tx_m);
2247 			txd->tx_m = NULL;
2248 			if (txstat & (RL_TDESC_STAT_EXCESSCOL|
2249 			    RL_TDESC_STAT_COLCNT))
2250 				ifp->if_collisions++;
2251 			if (txstat & RL_TDESC_STAT_TXERRSUM)
2252 				ifp->if_oerrors++;
2253 			else
2254 				ifp->if_opackets++;
2255 		}
2256 		sc->rl_ldata.rl_tx_free++;
2257 		ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
2258 	}
2259 	sc->rl_ldata.rl_tx_considx = cons;
2260 
2261 	/* No changes made to the TX ring, so no flush needed */
2262 
2263 	if (sc->rl_ldata.rl_tx_free != sc->rl_ldata.rl_tx_desc_cnt) {
2264 #ifdef RE_TX_MODERATION
2265 		/*
2266 		 * If not all descriptors have been reaped yet, reload
2267 		 * the timer so that we will eventually get another
2268 		 * interrupt that will cause us to re-enter this routine.
2269 		 * This is done in case the transmitter has gone idle.
2270 		 */
2271 		CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2272 #endif
2273 	} else
2274 		sc->rl_watchdog_timer = 0;
2275 }
2276 
2277 static void
2278 re_tick(void *xsc)
2279 {
2280 	struct rl_softc		*sc;
2281 	struct mii_data		*mii;
2282 
2283 	sc = xsc;
2284 
2285 	RL_LOCK_ASSERT(sc);
2286 
2287 	mii = device_get_softc(sc->rl_miibus);
2288 	mii_tick(mii);
2289 	if ((sc->rl_flags & RL_FLAG_LINK) == 0)
2290 		re_miibus_statchg(sc->rl_dev);
2291 	/*
2292 	 * Reclaim transmitted frames here. Technically it is not
2293 	 * necessary to do here but it ensures periodic reclamation
2294 	 * regardless of Tx completion interrupt which seems to be
2295 	 * lost on PCIe based controllers under certain situations.
2296 	 */
2297 	re_txeof(sc);
2298 	re_watchdog(sc);
2299 	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
2300 }
2301 
2302 #ifdef DEVICE_POLLING
2303 static int
2304 re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
2305 {
2306 	struct rl_softc *sc = ifp->if_softc;
2307 	int rx_npkts = 0;
2308 
2309 	RL_LOCK(sc);
2310 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
2311 		rx_npkts = re_poll_locked(ifp, cmd, count);
2312 	RL_UNLOCK(sc);
2313 	return (rx_npkts);
2314 }
2315 
2316 static int
2317 re_poll_locked(struct ifnet *ifp, enum poll_cmd cmd, int count)
2318 {
2319 	struct rl_softc *sc = ifp->if_softc;
2320 	int rx_npkts;
2321 
2322 	RL_LOCK_ASSERT(sc);
2323 
2324 	sc->rxcycles = count;
2325 	re_rxeof(sc, &rx_npkts);
2326 	re_txeof(sc);
2327 
2328 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2329 		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_txtask);
2330 
2331 	if (cmd == POLL_AND_CHECK_STATUS) { /* also check status register */
2332 		u_int16_t       status;
2333 
2334 		status = CSR_READ_2(sc, RL_ISR);
2335 		if (status == 0xffff)
2336 			return (rx_npkts);
2337 		if (status)
2338 			CSR_WRITE_2(sc, RL_ISR, status);
2339 		if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
2340 		    (sc->rl_flags & RL_FLAG_PCIE))
2341 			CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2342 
2343 		/*
2344 		 * XXX check behaviour on receiver stalls.
2345 		 */
2346 
2347 		if (status & RL_ISR_SYSTEM_ERR) {
2348 			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2349 			re_init_locked(sc);
2350 		}
2351 	}
2352 	return (rx_npkts);
2353 }
2354 #endif /* DEVICE_POLLING */
2355 
2356 static int
2357 re_intr(void *arg)
2358 {
2359 	struct rl_softc		*sc;
2360 	uint16_t		status;
2361 
2362 	sc = arg;
2363 
2364 	status = CSR_READ_2(sc, RL_ISR);
2365 	if (status == 0xFFFF || (status & RL_INTRS_CPLUS) == 0)
2366                 return (FILTER_STRAY);
2367 	CSR_WRITE_2(sc, RL_IMR, 0);
2368 
2369 	taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_inttask);
2370 
2371 	return (FILTER_HANDLED);
2372 }
2373 
2374 static void
2375 re_int_task(void *arg, int npending)
2376 {
2377 	struct rl_softc		*sc;
2378 	struct ifnet		*ifp;
2379 	u_int16_t		status;
2380 	int			rval = 0;
2381 
2382 	sc = arg;
2383 	ifp = sc->rl_ifp;
2384 
2385 	RL_LOCK(sc);
2386 
2387 	status = CSR_READ_2(sc, RL_ISR);
2388         CSR_WRITE_2(sc, RL_ISR, status);
2389 
2390 	if (sc->suspended ||
2391 	    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2392 		RL_UNLOCK(sc);
2393 		return;
2394 	}
2395 
2396 #ifdef DEVICE_POLLING
2397 	if  (ifp->if_capenable & IFCAP_POLLING) {
2398 		RL_UNLOCK(sc);
2399 		return;
2400 	}
2401 #endif
2402 
2403 	if (status & (RL_ISR_RX_OK|RL_ISR_RX_ERR|RL_ISR_FIFO_OFLOW))
2404 		rval = re_rxeof(sc, NULL);
2405 
2406 	/*
2407 	 * Some chips will ignore a second TX request issued
2408 	 * while an existing transmission is in progress. If
2409 	 * the transmitter goes idle but there are still
2410 	 * packets waiting to be sent, we need to restart the
2411 	 * channel here to flush them out. This only seems to
2412 	 * be required with the PCIe devices.
2413 	 */
2414 	if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
2415 	    (sc->rl_flags & RL_FLAG_PCIE))
2416 		CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2417 	if (status & (
2418 #ifdef RE_TX_MODERATION
2419 	    RL_ISR_TIMEOUT_EXPIRED|
2420 #else
2421 	    RL_ISR_TX_OK|
2422 #endif
2423 	    RL_ISR_TX_ERR|RL_ISR_TX_DESC_UNAVAIL))
2424 		re_txeof(sc);
2425 
2426 	if (status & RL_ISR_SYSTEM_ERR) {
2427 		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2428 		re_init_locked(sc);
2429 	}
2430 
2431 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2432 		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_txtask);
2433 
2434 	RL_UNLOCK(sc);
2435 
2436         if ((CSR_READ_2(sc, RL_ISR) & RL_INTRS_CPLUS) || rval) {
2437 		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_inttask);
2438 		return;
2439 	}
2440 
2441 	CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2442 }
2443 
2444 static int
2445 re_encap(struct rl_softc *sc, struct mbuf **m_head)
2446 {
2447 	struct rl_txdesc	*txd, *txd_last;
2448 	bus_dma_segment_t	segs[RL_NTXSEGS];
2449 	bus_dmamap_t		map;
2450 	struct mbuf		*m_new;
2451 	struct rl_desc		*desc;
2452 	int			nsegs, prod;
2453 	int			i, error, ei, si;
2454 	int			padlen;
2455 	uint32_t		cmdstat, csum_flags, vlanctl;
2456 
2457 	RL_LOCK_ASSERT(sc);
2458 	M_ASSERTPKTHDR((*m_head));
2459 
2460 	/*
2461 	 * With some of the RealTek chips, using the checksum offload
2462 	 * support in conjunction with the autopadding feature results
2463 	 * in the transmission of corrupt frames. For example, if we
2464 	 * need to send a really small IP fragment that's less than 60
2465 	 * bytes in size, and IP header checksumming is enabled, the
2466 	 * resulting ethernet frame that appears on the wire will
2467 	 * have garbled payload. To work around this, if TX IP checksum
2468 	 * offload is enabled, we always manually pad short frames out
2469 	 * to the minimum ethernet frame size.
2470 	 */
2471 	if ((sc->rl_flags & RL_FLAG_AUTOPAD) == 0 &&
2472 	    (*m_head)->m_pkthdr.len < RL_IP4CSUMTX_PADLEN &&
2473 	    ((*m_head)->m_pkthdr.csum_flags & CSUM_IP) != 0) {
2474 		padlen = RL_MIN_FRAMELEN - (*m_head)->m_pkthdr.len;
2475 		if (M_WRITABLE(*m_head) == 0) {
2476 			/* Get a writable copy. */
2477 			m_new = m_dup(*m_head, M_DONTWAIT);
2478 			m_freem(*m_head);
2479 			if (m_new == NULL) {
2480 				*m_head = NULL;
2481 				return (ENOBUFS);
2482 			}
2483 			*m_head = m_new;
2484 		}
2485 		if ((*m_head)->m_next != NULL ||
2486 		    M_TRAILINGSPACE(*m_head) < padlen) {
2487 			m_new = m_defrag(*m_head, M_DONTWAIT);
2488 			if (m_new == NULL) {
2489 				m_freem(*m_head);
2490 				*m_head = NULL;
2491 				return (ENOBUFS);
2492 			}
2493 		} else
2494 			m_new = *m_head;
2495 
2496 		/*
2497 		 * Manually pad short frames, and zero the pad space
2498 		 * to avoid leaking data.
2499 		 */
2500 		bzero(mtod(m_new, char *) + m_new->m_pkthdr.len, padlen);
2501 		m_new->m_pkthdr.len += padlen;
2502 		m_new->m_len = m_new->m_pkthdr.len;
2503 		*m_head = m_new;
2504 	}
2505 
2506 	prod = sc->rl_ldata.rl_tx_prodidx;
2507 	txd = &sc->rl_ldata.rl_tx_desc[prod];
2508 	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap,
2509 	    *m_head, segs, &nsegs, BUS_DMA_NOWAIT);
2510 	if (error == EFBIG) {
2511 		m_new = m_collapse(*m_head, M_DONTWAIT, RL_NTXSEGS);
2512 		if (m_new == NULL) {
2513 			m_freem(*m_head);
2514 			*m_head = NULL;
2515 			return (ENOBUFS);
2516 		}
2517 		*m_head = m_new;
2518 		error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_tx_mtag,
2519 		    txd->tx_dmamap, *m_head, segs, &nsegs, BUS_DMA_NOWAIT);
2520 		if (error != 0) {
2521 			m_freem(*m_head);
2522 			*m_head = NULL;
2523 			return (error);
2524 		}
2525 	} else if (error != 0)
2526 		return (error);
2527 	if (nsegs == 0) {
2528 		m_freem(*m_head);
2529 		*m_head = NULL;
2530 		return (EIO);
2531 	}
2532 
2533 	/* Check for number of available descriptors. */
2534 	if (sc->rl_ldata.rl_tx_free - nsegs <= 1) {
2535 		bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap);
2536 		return (ENOBUFS);
2537 	}
2538 
2539 	bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap,
2540 	    BUS_DMASYNC_PREWRITE);
2541 
2542 	/*
2543 	 * Set up checksum offload. Note: checksum offload bits must
2544 	 * appear in all descriptors of a multi-descriptor transmit
2545 	 * attempt. This is according to testing done with an 8169
2546 	 * chip. This is a requirement.
2547 	 */
2548 	vlanctl = 0;
2549 	csum_flags = 0;
2550 	if (((*m_head)->m_pkthdr.csum_flags & CSUM_TSO) != 0) {
2551 		if ((sc->rl_flags & RL_FLAG_DESCV2) != 0) {
2552 			csum_flags |= RL_TDESC_CMD_LGSEND;
2553 			vlanctl |= ((uint32_t)(*m_head)->m_pkthdr.tso_segsz <<
2554 			    RL_TDESC_CMD_MSSVALV2_SHIFT);
2555 		} else {
2556 			csum_flags |= RL_TDESC_CMD_LGSEND |
2557 			    ((uint32_t)(*m_head)->m_pkthdr.tso_segsz <<
2558 			    RL_TDESC_CMD_MSSVAL_SHIFT);
2559 		}
2560 	} else {
2561 		/*
2562 		 * Unconditionally enable IP checksum if TCP or UDP
2563 		 * checksum is required. Otherwise, TCP/UDP checksum
2564 		 * does't make effects.
2565 		 */
2566 		if (((*m_head)->m_pkthdr.csum_flags & RE_CSUM_FEATURES) != 0) {
2567 			if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
2568 				csum_flags |= RL_TDESC_CMD_IPCSUM;
2569 				if (((*m_head)->m_pkthdr.csum_flags &
2570 				    CSUM_TCP) != 0)
2571 					csum_flags |= RL_TDESC_CMD_TCPCSUM;
2572 				if (((*m_head)->m_pkthdr.csum_flags &
2573 				    CSUM_UDP) != 0)
2574 					csum_flags |= RL_TDESC_CMD_UDPCSUM;
2575 			} else {
2576 				vlanctl |= RL_TDESC_CMD_IPCSUMV2;
2577 				if (((*m_head)->m_pkthdr.csum_flags &
2578 				    CSUM_TCP) != 0)
2579 					vlanctl |= RL_TDESC_CMD_TCPCSUMV2;
2580 				if (((*m_head)->m_pkthdr.csum_flags &
2581 				    CSUM_UDP) != 0)
2582 					vlanctl |= RL_TDESC_CMD_UDPCSUMV2;
2583 			}
2584 		}
2585 	}
2586 
2587 	/*
2588 	 * Set up hardware VLAN tagging. Note: vlan tag info must
2589 	 * appear in all descriptors of a multi-descriptor
2590 	 * transmission attempt.
2591 	 */
2592 	if ((*m_head)->m_flags & M_VLANTAG)
2593 		vlanctl |= bswap16((*m_head)->m_pkthdr.ether_vtag) |
2594 		    RL_TDESC_VLANCTL_TAG;
2595 
2596 	si = prod;
2597 	for (i = 0; i < nsegs; i++, prod = RL_TX_DESC_NXT(sc, prod)) {
2598 		desc = &sc->rl_ldata.rl_tx_list[prod];
2599 		desc->rl_vlanctl = htole32(vlanctl);
2600 		desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[i].ds_addr));
2601 		desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[i].ds_addr));
2602 		cmdstat = segs[i].ds_len;
2603 		if (i != 0)
2604 			cmdstat |= RL_TDESC_CMD_OWN;
2605 		if (prod == sc->rl_ldata.rl_tx_desc_cnt - 1)
2606 			cmdstat |= RL_TDESC_CMD_EOR;
2607 		desc->rl_cmdstat = htole32(cmdstat | csum_flags);
2608 		sc->rl_ldata.rl_tx_free--;
2609 	}
2610 	/* Update producer index. */
2611 	sc->rl_ldata.rl_tx_prodidx = prod;
2612 
2613 	/* Set EOF on the last descriptor. */
2614 	ei = RL_TX_DESC_PRV(sc, prod);
2615 	desc = &sc->rl_ldata.rl_tx_list[ei];
2616 	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_EOF);
2617 
2618 	desc = &sc->rl_ldata.rl_tx_list[si];
2619 	/* Set SOF and transfer ownership of packet to the chip. */
2620 	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_OWN | RL_TDESC_CMD_SOF);
2621 
2622 	/*
2623 	 * Insure that the map for this transmission
2624 	 * is placed at the array index of the last descriptor
2625 	 * in this chain.  (Swap last and first dmamaps.)
2626 	 */
2627 	txd_last = &sc->rl_ldata.rl_tx_desc[ei];
2628 	map = txd->tx_dmamap;
2629 	txd->tx_dmamap = txd_last->tx_dmamap;
2630 	txd_last->tx_dmamap = map;
2631 	txd_last->tx_m = *m_head;
2632 
2633 	return (0);
2634 }
2635 
2636 static void
2637 re_tx_task(void *arg, int npending)
2638 {
2639 	struct ifnet		*ifp;
2640 
2641 	ifp = arg;
2642 	re_start(ifp);
2643 }
2644 
2645 /*
2646  * Main transmit routine for C+ and gigE NICs.
2647  */
2648 static void
2649 re_start(struct ifnet *ifp)
2650 {
2651 	struct rl_softc		*sc;
2652 	struct mbuf		*m_head;
2653 	int			queued;
2654 
2655 	sc = ifp->if_softc;
2656 
2657 	RL_LOCK(sc);
2658 
2659 	if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
2660 	    IFF_DRV_RUNNING || (sc->rl_flags & RL_FLAG_LINK) == 0) {
2661 		RL_UNLOCK(sc);
2662 		return;
2663 	}
2664 
2665 	for (queued = 0; !IFQ_DRV_IS_EMPTY(&ifp->if_snd) &&
2666 	    sc->rl_ldata.rl_tx_free > 1;) {
2667 		IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
2668 		if (m_head == NULL)
2669 			break;
2670 
2671 		if (re_encap(sc, &m_head) != 0) {
2672 			if (m_head == NULL)
2673 				break;
2674 			IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
2675 			ifp->if_drv_flags |= IFF_DRV_OACTIVE;
2676 			break;
2677 		}
2678 
2679 		/*
2680 		 * If there's a BPF listener, bounce a copy of this frame
2681 		 * to him.
2682 		 */
2683 		ETHER_BPF_MTAP(ifp, m_head);
2684 
2685 		queued++;
2686 	}
2687 
2688 	if (queued == 0) {
2689 #ifdef RE_TX_MODERATION
2690 		if (sc->rl_ldata.rl_tx_free != sc->rl_ldata.rl_tx_desc_cnt)
2691 			CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2692 #endif
2693 		RL_UNLOCK(sc);
2694 		return;
2695 	}
2696 
2697 	/* Flush the TX descriptors */
2698 
2699 	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2700 	    sc->rl_ldata.rl_tx_list_map,
2701 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2702 
2703 	CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2704 
2705 #ifdef RE_TX_MODERATION
2706 	/*
2707 	 * Use the countdown timer for interrupt moderation.
2708 	 * 'TX done' interrupts are disabled. Instead, we reset the
2709 	 * countdown timer, which will begin counting until it hits
2710 	 * the value in the TIMERINT register, and then trigger an
2711 	 * interrupt. Each time we write to the TIMERCNT register,
2712 	 * the timer count is reset to 0.
2713 	 */
2714 	CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2715 #endif
2716 
2717 	/*
2718 	 * Set a timeout in case the chip goes out to lunch.
2719 	 */
2720 	sc->rl_watchdog_timer = 5;
2721 
2722 	RL_UNLOCK(sc);
2723 }
2724 
2725 static void
2726 re_set_jumbo(struct rl_softc *sc, int jumbo)
2727 {
2728 
2729 	if (sc->rl_hwrev->rl_rev == RL_HWREV_8168E_VL) {
2730 		pci_set_max_read_req(sc->rl_dev, 4096);
2731 		return;
2732 	}
2733 
2734 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG);
2735 	if (jumbo != 0) {
2736 		CSR_WRITE_1(sc, RL_CFG3, CSR_READ_1(sc, RL_CFG3) |
2737 		    RL_CFG3_JUMBO_EN0);
2738 		switch (sc->rl_hwrev->rl_rev) {
2739 		case RL_HWREV_8168DP:
2740 			break;
2741 		case RL_HWREV_8168E:
2742 			CSR_WRITE_1(sc, RL_CFG4, CSR_READ_1(sc, RL_CFG4) |
2743 			    0x01);
2744 			break;
2745 		default:
2746 			CSR_WRITE_1(sc, RL_CFG4, CSR_READ_1(sc, RL_CFG4) |
2747 			    RL_CFG4_JUMBO_EN1);
2748 		}
2749 	} else {
2750 		CSR_WRITE_1(sc, RL_CFG3, CSR_READ_1(sc, RL_CFG3) &
2751 		    ~RL_CFG3_JUMBO_EN0);
2752 		switch (sc->rl_hwrev->rl_rev) {
2753 		case RL_HWREV_8168DP:
2754 			break;
2755 		case RL_HWREV_8168E:
2756 			CSR_WRITE_1(sc, RL_CFG4, CSR_READ_1(sc, RL_CFG4) &
2757 			    ~0x01);
2758 			break;
2759 		default:
2760 			CSR_WRITE_1(sc, RL_CFG4, CSR_READ_1(sc, RL_CFG4) &
2761 			    ~RL_CFG4_JUMBO_EN1);
2762 		}
2763 	}
2764 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
2765 
2766 	switch (sc->rl_hwrev->rl_rev) {
2767 	case RL_HWREV_8168DP:
2768 		pci_set_max_read_req(sc->rl_dev, 4096);
2769 		break;
2770 	default:
2771 		if (jumbo != 0)
2772 			pci_set_max_read_req(sc->rl_dev, 512);
2773 		else
2774 			pci_set_max_read_req(sc->rl_dev, 4096);
2775 	}
2776 }
2777 
2778 static void
2779 re_init(void *xsc)
2780 {
2781 	struct rl_softc		*sc = xsc;
2782 
2783 	RL_LOCK(sc);
2784 	re_init_locked(sc);
2785 	RL_UNLOCK(sc);
2786 }
2787 
2788 static void
2789 re_init_locked(struct rl_softc *sc)
2790 {
2791 	struct ifnet		*ifp = sc->rl_ifp;
2792 	struct mii_data		*mii;
2793 	uint32_t		reg;
2794 	uint16_t		cfg;
2795 	union {
2796 		uint32_t align_dummy;
2797 		u_char eaddr[ETHER_ADDR_LEN];
2798         } eaddr;
2799 
2800 	RL_LOCK_ASSERT(sc);
2801 
2802 	mii = device_get_softc(sc->rl_miibus);
2803 
2804 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
2805 		return;
2806 
2807 	/*
2808 	 * Cancel pending I/O and free all RX/TX buffers.
2809 	 */
2810 	re_stop(sc);
2811 
2812 	/* Put controller into known state. */
2813 	re_reset(sc);
2814 
2815 	/*
2816 	 * For C+ mode, initialize the RX descriptors and mbufs.
2817 	 */
2818 	if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
2819 		if (ifp->if_mtu > RL_MTU) {
2820 			if (re_jrx_list_init(sc) != 0) {
2821 				device_printf(sc->rl_dev,
2822 				    "no memory for jumbo RX buffers\n");
2823 				re_stop(sc);
2824 				return;
2825 			}
2826 			/* Disable checksum offloading for jumbo frames. */
2827 			ifp->if_capenable &= ~(IFCAP_HWCSUM | IFCAP_TSO4);
2828 			ifp->if_hwassist &= ~(RE_CSUM_FEATURES | CSUM_TSO);
2829 		} else {
2830 			if (re_rx_list_init(sc) != 0) {
2831 				device_printf(sc->rl_dev,
2832 				    "no memory for RX buffers\n");
2833 				re_stop(sc);
2834 				return;
2835 			}
2836 		}
2837 		re_set_jumbo(sc, ifp->if_mtu > RL_MTU);
2838 	} else {
2839 		if (re_rx_list_init(sc) != 0) {
2840 			device_printf(sc->rl_dev, "no memory for RX buffers\n");
2841 			re_stop(sc);
2842 			return;
2843 		}
2844 		if ((sc->rl_flags & RL_FLAG_PCIE) != 0 &&
2845 		    pci_get_device(sc->rl_dev) != RT_DEVICEID_8101E) {
2846 			if (ifp->if_mtu > RL_MTU)
2847 				pci_set_max_read_req(sc->rl_dev, 512);
2848 			else
2849 				pci_set_max_read_req(sc->rl_dev, 4096);
2850 		}
2851 	}
2852 	re_tx_list_init(sc);
2853 
2854 	/*
2855 	 * Enable C+ RX and TX mode, as well as VLAN stripping and
2856 	 * RX checksum offload. We must configure the C+ register
2857 	 * before all others.
2858 	 */
2859 	cfg = RL_CPLUSCMD_PCI_MRW;
2860 	if ((ifp->if_capenable & IFCAP_RXCSUM) != 0)
2861 		cfg |= RL_CPLUSCMD_RXCSUM_ENB;
2862 	if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0)
2863 		cfg |= RL_CPLUSCMD_VLANSTRIP;
2864 	if ((sc->rl_flags & RL_FLAG_MACSTAT) != 0) {
2865 		cfg |= RL_CPLUSCMD_MACSTAT_DIS;
2866 		/* XXX magic. */
2867 		cfg |= 0x0001;
2868 	} else
2869 		cfg |= RL_CPLUSCMD_RXENB | RL_CPLUSCMD_TXENB;
2870 	CSR_WRITE_2(sc, RL_CPLUS_CMD, cfg);
2871 	if (sc->rl_hwrev->rl_rev == RL_HWREV_8169_8110SC ||
2872 	    sc->rl_hwrev->rl_rev == RL_HWREV_8169_8110SCE) {
2873 		reg = 0x000fff00;
2874 		if ((CSR_READ_1(sc, RL_CFG2) & RL_CFG2_PCI66MHZ) != 0)
2875 			reg |= 0x000000ff;
2876 		if (sc->rl_hwrev->rl_rev == RL_HWREV_8169_8110SCE)
2877 			reg |= 0x00f00000;
2878 		CSR_WRITE_4(sc, 0x7c, reg);
2879 		/* Disable interrupt mitigation. */
2880 		CSR_WRITE_2(sc, 0xe2, 0);
2881 	}
2882 	/*
2883 	 * Disable TSO if interface MTU size is greater than MSS
2884 	 * allowed in controller.
2885 	 */
2886 	if (ifp->if_mtu > RL_TSO_MTU && (ifp->if_capenable & IFCAP_TSO4) != 0) {
2887 		ifp->if_capenable &= ~IFCAP_TSO4;
2888 		ifp->if_hwassist &= ~CSUM_TSO;
2889 	}
2890 
2891 	/*
2892 	 * Init our MAC address.  Even though the chipset
2893 	 * documentation doesn't mention it, we need to enter "Config
2894 	 * register write enable" mode to modify the ID registers.
2895 	 */
2896 	/* Copy MAC address on stack to align. */
2897 	bcopy(IF_LLADDR(ifp), eaddr.eaddr, ETHER_ADDR_LEN);
2898 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG);
2899 	CSR_WRITE_4(sc, RL_IDR0,
2900 	    htole32(*(u_int32_t *)(&eaddr.eaddr[0])));
2901 	CSR_WRITE_4(sc, RL_IDR4,
2902 	    htole32(*(u_int32_t *)(&eaddr.eaddr[4])));
2903 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
2904 
2905 	/*
2906 	 * Load the addresses of the RX and TX lists into the chip.
2907 	 */
2908 
2909 	CSR_WRITE_4(sc, RL_RXLIST_ADDR_HI,
2910 	    RL_ADDR_HI(sc->rl_ldata.rl_rx_list_addr));
2911 	CSR_WRITE_4(sc, RL_RXLIST_ADDR_LO,
2912 	    RL_ADDR_LO(sc->rl_ldata.rl_rx_list_addr));
2913 
2914 	CSR_WRITE_4(sc, RL_TXLIST_ADDR_HI,
2915 	    RL_ADDR_HI(sc->rl_ldata.rl_tx_list_addr));
2916 	CSR_WRITE_4(sc, RL_TXLIST_ADDR_LO,
2917 	    RL_ADDR_LO(sc->rl_ldata.rl_tx_list_addr));
2918 
2919 	/*
2920 	 * Enable transmit and receive.
2921 	 */
2922 	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
2923 
2924 	/*
2925 	 * Set the initial TX configuration.
2926 	 */
2927 	if (sc->rl_testmode) {
2928 		if (sc->rl_type == RL_8169)
2929 			CSR_WRITE_4(sc, RL_TXCFG,
2930 			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON);
2931 		else
2932 			CSR_WRITE_4(sc, RL_TXCFG,
2933 			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON_CPLUS);
2934 	} else
2935 		CSR_WRITE_4(sc, RL_TXCFG, RL_TXCFG_CONFIG);
2936 
2937 	CSR_WRITE_1(sc, RL_EARLY_TX_THRESH, 16);
2938 
2939 	/*
2940 	 * Set the initial RX configuration.
2941 	 */
2942 	re_set_rxmode(sc);
2943 
2944 	/* Configure interrupt moderation. */
2945 	if (sc->rl_type == RL_8169) {
2946 		/* Magic from vendor. */
2947 		CSR_WRITE_2(sc, RL_INTRMOD, 0x5100);
2948 	}
2949 
2950 #ifdef DEVICE_POLLING
2951 	/*
2952 	 * Disable interrupts if we are polling.
2953 	 */
2954 	if (ifp->if_capenable & IFCAP_POLLING)
2955 		CSR_WRITE_2(sc, RL_IMR, 0);
2956 	else	/* otherwise ... */
2957 #endif
2958 
2959 	/*
2960 	 * Enable interrupts.
2961 	 */
2962 	if (sc->rl_testmode)
2963 		CSR_WRITE_2(sc, RL_IMR, 0);
2964 	else
2965 		CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2966 	CSR_WRITE_2(sc, RL_ISR, RL_INTRS_CPLUS);
2967 
2968 	/* Set initial TX threshold */
2969 	sc->rl_txthresh = RL_TX_THRESH_INIT;
2970 
2971 	/* Start RX/TX process. */
2972 	CSR_WRITE_4(sc, RL_MISSEDPKT, 0);
2973 #ifdef notdef
2974 	/* Enable receiver and transmitter. */
2975 	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
2976 #endif
2977 
2978 #ifdef RE_TX_MODERATION
2979 	/*
2980 	 * Initialize the timer interrupt register so that
2981 	 * a timer interrupt will be generated once the timer
2982 	 * reaches a certain number of ticks. The timer is
2983 	 * reloaded on each transmit. This gives us TX interrupt
2984 	 * moderation, which dramatically improves TX frame rate.
2985 	 */
2986 	if (sc->rl_type == RL_8169)
2987 		CSR_WRITE_4(sc, RL_TIMERINT_8169, 0x800);
2988 	else
2989 		CSR_WRITE_4(sc, RL_TIMERINT, 0x400);
2990 #endif
2991 
2992 	/*
2993 	 * For 8169 gigE NICs, set the max allowed RX packet
2994 	 * size so we can receive jumbo frames.
2995 	 */
2996 	if (sc->rl_type == RL_8169) {
2997 		if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
2998 			/*
2999 			 * For controllers that use new jumbo frame scheme,
3000 			 * set maximum size of jumbo frame depedning on
3001 			 * controller revisions.
3002 			 */
3003 			if (ifp->if_mtu > RL_MTU)
3004 				CSR_WRITE_2(sc, RL_MAXRXPKTLEN,
3005 				    sc->rl_hwrev->rl_max_mtu +
3006 				    ETHER_VLAN_ENCAP_LEN + ETHER_HDR_LEN +
3007 				    ETHER_CRC_LEN);
3008 			else
3009 				CSR_WRITE_2(sc, RL_MAXRXPKTLEN,
3010 				    RE_RX_DESC_BUFLEN);
3011 		} else if ((sc->rl_flags & RL_FLAG_PCIE) != 0 &&
3012 		    sc->rl_hwrev->rl_max_mtu == RL_MTU) {
3013 			/* RTL810x has no jumbo frame support. */
3014 			CSR_WRITE_2(sc, RL_MAXRXPKTLEN, RE_RX_DESC_BUFLEN);
3015 		} else
3016 			CSR_WRITE_2(sc, RL_MAXRXPKTLEN, 16383);
3017 	}
3018 
3019 	if (sc->rl_testmode)
3020 		return;
3021 
3022 	mii_mediachg(mii);
3023 
3024 	CSR_WRITE_1(sc, RL_CFG1, CSR_READ_1(sc, RL_CFG1) | RL_CFG1_DRVLOAD);
3025 
3026 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
3027 	ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
3028 
3029 	sc->rl_flags &= ~RL_FLAG_LINK;
3030 	sc->rl_watchdog_timer = 0;
3031 	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
3032 }
3033 
3034 /*
3035  * Set media options.
3036  */
3037 static int
3038 re_ifmedia_upd(struct ifnet *ifp)
3039 {
3040 	struct rl_softc		*sc;
3041 	struct mii_data		*mii;
3042 	int			error;
3043 
3044 	sc = ifp->if_softc;
3045 	mii = device_get_softc(sc->rl_miibus);
3046 	RL_LOCK(sc);
3047 	error = mii_mediachg(mii);
3048 	RL_UNLOCK(sc);
3049 
3050 	return (error);
3051 }
3052 
3053 /*
3054  * Report current media status.
3055  */
3056 static void
3057 re_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
3058 {
3059 	struct rl_softc		*sc;
3060 	struct mii_data		*mii;
3061 
3062 	sc = ifp->if_softc;
3063 	mii = device_get_softc(sc->rl_miibus);
3064 
3065 	RL_LOCK(sc);
3066 	mii_pollstat(mii);
3067 	RL_UNLOCK(sc);
3068 	ifmr->ifm_active = mii->mii_media_active;
3069 	ifmr->ifm_status = mii->mii_media_status;
3070 }
3071 
3072 static int
3073 re_ioctl(struct ifnet *ifp, u_long command, caddr_t data)
3074 {
3075 	struct rl_softc		*sc = ifp->if_softc;
3076 	struct ifreq		*ifr = (struct ifreq *) data;
3077 	struct mii_data		*mii;
3078 	int			error = 0;
3079 
3080 	switch (command) {
3081 	case SIOCSIFMTU:
3082 		if (ifr->ifr_mtu < ETHERMIN ||
3083 		    ifr->ifr_mtu > sc->rl_hwrev->rl_max_mtu) {
3084 			error = EINVAL;
3085 			break;
3086 		}
3087 		RL_LOCK(sc);
3088 		if (ifp->if_mtu != ifr->ifr_mtu) {
3089 			ifp->if_mtu = ifr->ifr_mtu;
3090 			if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0 &&
3091 			    (ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
3092 				ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3093 				re_init_locked(sc);
3094 			}
3095 			if (ifp->if_mtu > RL_TSO_MTU &&
3096 			    (ifp->if_capenable & IFCAP_TSO4) != 0) {
3097 				ifp->if_capenable &= ~(IFCAP_TSO4 |
3098 				    IFCAP_VLAN_HWTSO);
3099 				ifp->if_hwassist &= ~CSUM_TSO;
3100 			}
3101 			VLAN_CAPABILITIES(ifp);
3102 		}
3103 		RL_UNLOCK(sc);
3104 		break;
3105 	case SIOCSIFFLAGS:
3106 		RL_LOCK(sc);
3107 		if ((ifp->if_flags & IFF_UP) != 0) {
3108 			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
3109 				if (((ifp->if_flags ^ sc->rl_if_flags)
3110 				    & (IFF_PROMISC | IFF_ALLMULTI)) != 0)
3111 					re_set_rxmode(sc);
3112 			} else
3113 				re_init_locked(sc);
3114 		} else {
3115 			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
3116 				re_stop(sc);
3117 		}
3118 		sc->rl_if_flags = ifp->if_flags;
3119 		RL_UNLOCK(sc);
3120 		break;
3121 	case SIOCADDMULTI:
3122 	case SIOCDELMULTI:
3123 		RL_LOCK(sc);
3124 		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
3125 			re_set_rxmode(sc);
3126 		RL_UNLOCK(sc);
3127 		break;
3128 	case SIOCGIFMEDIA:
3129 	case SIOCSIFMEDIA:
3130 		mii = device_get_softc(sc->rl_miibus);
3131 		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, command);
3132 		break;
3133 	case SIOCSIFCAP:
3134 	    {
3135 		int mask, reinit;
3136 
3137 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
3138 		reinit = 0;
3139 #ifdef DEVICE_POLLING
3140 		if (mask & IFCAP_POLLING) {
3141 			if (ifr->ifr_reqcap & IFCAP_POLLING) {
3142 				error = ether_poll_register(re_poll, ifp);
3143 				if (error)
3144 					return (error);
3145 				RL_LOCK(sc);
3146 				/* Disable interrupts */
3147 				CSR_WRITE_2(sc, RL_IMR, 0x0000);
3148 				ifp->if_capenable |= IFCAP_POLLING;
3149 				RL_UNLOCK(sc);
3150 			} else {
3151 				error = ether_poll_deregister(ifp);
3152 				/* Enable interrupts. */
3153 				RL_LOCK(sc);
3154 				CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
3155 				ifp->if_capenable &= ~IFCAP_POLLING;
3156 				RL_UNLOCK(sc);
3157 			}
3158 		}
3159 #endif /* DEVICE_POLLING */
3160 		if ((mask & IFCAP_TXCSUM) != 0 &&
3161 		    (ifp->if_capabilities & IFCAP_TXCSUM) != 0) {
3162 			ifp->if_capenable ^= IFCAP_TXCSUM;
3163 			if ((ifp->if_capenable & IFCAP_TXCSUM) != 0)
3164 				ifp->if_hwassist |= RE_CSUM_FEATURES;
3165 			else
3166 				ifp->if_hwassist &= ~RE_CSUM_FEATURES;
3167 			reinit = 1;
3168 		}
3169 		if ((mask & IFCAP_RXCSUM) != 0 &&
3170 		    (ifp->if_capabilities & IFCAP_RXCSUM) != 0) {
3171 			ifp->if_capenable ^= IFCAP_RXCSUM;
3172 			reinit = 1;
3173 		}
3174 		if ((mask & IFCAP_TSO4) != 0 &&
3175 		    (ifp->if_capabilities & IFCAP_TSO) != 0) {
3176 			ifp->if_capenable ^= IFCAP_TSO4;
3177 			if ((IFCAP_TSO4 & ifp->if_capenable) != 0)
3178 				ifp->if_hwassist |= CSUM_TSO;
3179 			else
3180 				ifp->if_hwassist &= ~CSUM_TSO;
3181 			if (ifp->if_mtu > RL_TSO_MTU &&
3182 			    (ifp->if_capenable & IFCAP_TSO4) != 0) {
3183 				ifp->if_capenable &= ~IFCAP_TSO4;
3184 				ifp->if_hwassist &= ~CSUM_TSO;
3185 			}
3186 		}
3187 		if ((mask & IFCAP_VLAN_HWTSO) != 0 &&
3188 		    (ifp->if_capabilities & IFCAP_VLAN_HWTSO) != 0)
3189 			ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
3190 		if ((mask & IFCAP_VLAN_HWTAGGING) != 0 &&
3191 		    (ifp->if_capabilities & IFCAP_VLAN_HWTAGGING) != 0) {
3192 			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
3193 			/* TSO over VLAN requires VLAN hardware tagging. */
3194 			if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) == 0)
3195 				ifp->if_capenable &= ~IFCAP_VLAN_HWTSO;
3196 			reinit = 1;
3197 		}
3198 		if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0 &&
3199 		    (mask & (IFCAP_HWCSUM | IFCAP_TSO4 |
3200 		    IFCAP_VLAN_HWTSO)) != 0)
3201 				reinit = 1;
3202 		if ((mask & IFCAP_WOL) != 0 &&
3203 		    (ifp->if_capabilities & IFCAP_WOL) != 0) {
3204 			if ((mask & IFCAP_WOL_UCAST) != 0)
3205 				ifp->if_capenable ^= IFCAP_WOL_UCAST;
3206 			if ((mask & IFCAP_WOL_MCAST) != 0)
3207 				ifp->if_capenable ^= IFCAP_WOL_MCAST;
3208 			if ((mask & IFCAP_WOL_MAGIC) != 0)
3209 				ifp->if_capenable ^= IFCAP_WOL_MAGIC;
3210 		}
3211 		if (reinit && ifp->if_drv_flags & IFF_DRV_RUNNING) {
3212 			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3213 			re_init(sc);
3214 		}
3215 		VLAN_CAPABILITIES(ifp);
3216 	    }
3217 		break;
3218 	default:
3219 		error = ether_ioctl(ifp, command, data);
3220 		break;
3221 	}
3222 
3223 	return (error);
3224 }
3225 
3226 static void
3227 re_watchdog(struct rl_softc *sc)
3228 {
3229 	struct ifnet		*ifp;
3230 
3231 	RL_LOCK_ASSERT(sc);
3232 
3233 	if (sc->rl_watchdog_timer == 0 || --sc->rl_watchdog_timer != 0)
3234 		return;
3235 
3236 	ifp = sc->rl_ifp;
3237 	re_txeof(sc);
3238 	if (sc->rl_ldata.rl_tx_free == sc->rl_ldata.rl_tx_desc_cnt) {
3239 		if_printf(ifp, "watchdog timeout (missed Tx interrupts) "
3240 		    "-- recovering\n");
3241 		if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
3242 			taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_txtask);
3243 		return;
3244 	}
3245 
3246 	if_printf(ifp, "watchdog timeout\n");
3247 	ifp->if_oerrors++;
3248 
3249 	re_rxeof(sc, NULL);
3250 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3251 	re_init_locked(sc);
3252 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
3253 		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_txtask);
3254 }
3255 
3256 /*
3257  * Stop the adapter and free any mbufs allocated to the
3258  * RX and TX lists.
3259  */
3260 static void
3261 re_stop(struct rl_softc *sc)
3262 {
3263 	int			i;
3264 	struct ifnet		*ifp;
3265 	struct rl_txdesc	*txd;
3266 	struct rl_rxdesc	*rxd;
3267 
3268 	RL_LOCK_ASSERT(sc);
3269 
3270 	ifp = sc->rl_ifp;
3271 
3272 	sc->rl_watchdog_timer = 0;
3273 	callout_stop(&sc->rl_stat_callout);
3274 	ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
3275 
3276 	if ((sc->rl_flags & RL_FLAG_CMDSTOP) != 0)
3277 		CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_STOPREQ | RL_CMD_TX_ENB |
3278 		    RL_CMD_RX_ENB);
3279 	else
3280 		CSR_WRITE_1(sc, RL_COMMAND, 0x00);
3281 	DELAY(1000);
3282 	CSR_WRITE_2(sc, RL_IMR, 0x0000);
3283 	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
3284 
3285 	if (sc->rl_head != NULL) {
3286 		m_freem(sc->rl_head);
3287 		sc->rl_head = sc->rl_tail = NULL;
3288 	}
3289 
3290 	/* Free the TX list buffers. */
3291 
3292 	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
3293 		txd = &sc->rl_ldata.rl_tx_desc[i];
3294 		if (txd->tx_m != NULL) {
3295 			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
3296 			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
3297 			bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag,
3298 			    txd->tx_dmamap);
3299 			m_freem(txd->tx_m);
3300 			txd->tx_m = NULL;
3301 		}
3302 	}
3303 
3304 	/* Free the RX list buffers. */
3305 
3306 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
3307 		rxd = &sc->rl_ldata.rl_rx_desc[i];
3308 		if (rxd->rx_m != NULL) {
3309 			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
3310 			    rxd->rx_dmamap, BUS_DMASYNC_POSTREAD);
3311 			bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag,
3312 			    rxd->rx_dmamap);
3313 			m_freem(rxd->rx_m);
3314 			rxd->rx_m = NULL;
3315 		}
3316 	}
3317 }
3318 
3319 /*
3320  * Device suspend routine.  Stop the interface and save some PCI
3321  * settings in case the BIOS doesn't restore them properly on
3322  * resume.
3323  */
3324 static int
3325 re_suspend(device_t dev)
3326 {
3327 	struct rl_softc		*sc;
3328 
3329 	sc = device_get_softc(dev);
3330 
3331 	RL_LOCK(sc);
3332 	re_stop(sc);
3333 	re_setwol(sc);
3334 	sc->suspended = 1;
3335 	RL_UNLOCK(sc);
3336 
3337 	return (0);
3338 }
3339 
3340 /*
3341  * Device resume routine.  Restore some PCI settings in case the BIOS
3342  * doesn't, re-enable busmastering, and restart the interface if
3343  * appropriate.
3344  */
3345 static int
3346 re_resume(device_t dev)
3347 {
3348 	struct rl_softc		*sc;
3349 	struct ifnet		*ifp;
3350 
3351 	sc = device_get_softc(dev);
3352 
3353 	RL_LOCK(sc);
3354 
3355 	ifp = sc->rl_ifp;
3356 	/* Take controller out of sleep mode. */
3357 	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
3358 		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
3359 			CSR_WRITE_1(sc, RL_GPIO,
3360 			    CSR_READ_1(sc, RL_GPIO) | 0x01);
3361 	}
3362 
3363 	/*
3364 	 * Clear WOL matching such that normal Rx filtering
3365 	 * wouldn't interfere with WOL patterns.
3366 	 */
3367 	re_clrwol(sc);
3368 
3369 	/* reinitialize interface if necessary */
3370 	if (ifp->if_flags & IFF_UP)
3371 		re_init_locked(sc);
3372 
3373 	sc->suspended = 0;
3374 	RL_UNLOCK(sc);
3375 
3376 	return (0);
3377 }
3378 
3379 /*
3380  * Stop all chip I/O so that the kernel's probe routines don't
3381  * get confused by errant DMAs when rebooting.
3382  */
3383 static int
3384 re_shutdown(device_t dev)
3385 {
3386 	struct rl_softc		*sc;
3387 
3388 	sc = device_get_softc(dev);
3389 
3390 	RL_LOCK(sc);
3391 	re_stop(sc);
3392 	/*
3393 	 * Mark interface as down since otherwise we will panic if
3394 	 * interrupt comes in later on, which can happen in some
3395 	 * cases.
3396 	 */
3397 	sc->rl_ifp->if_flags &= ~IFF_UP;
3398 	re_setwol(sc);
3399 	RL_UNLOCK(sc);
3400 
3401 	return (0);
3402 }
3403 
3404 static void
3405 re_setwol(struct rl_softc *sc)
3406 {
3407 	struct ifnet		*ifp;
3408 	int			pmc;
3409 	uint16_t		pmstat;
3410 	uint8_t			v;
3411 
3412 	RL_LOCK_ASSERT(sc);
3413 
3414 	if (pci_find_extcap(sc->rl_dev, PCIY_PMG, &pmc) != 0)
3415 		return;
3416 
3417 	ifp = sc->rl_ifp;
3418 	/* Put controller into sleep mode. */
3419 	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
3420 		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
3421 			CSR_WRITE_1(sc, RL_GPIO,
3422 			    CSR_READ_1(sc, RL_GPIO) & ~0x01);
3423 	}
3424 	if ((ifp->if_capenable & IFCAP_WOL) != 0 &&
3425 	    (sc->rl_flags & RL_FLAG_WOLRXENB) != 0)
3426 		CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RX_ENB);
3427 	/* Enable config register write. */
3428 	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
3429 
3430 	/* Enable PME. */
3431 	v = CSR_READ_1(sc, RL_CFG1);
3432 	v &= ~RL_CFG1_PME;
3433 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3434 		v |= RL_CFG1_PME;
3435 	CSR_WRITE_1(sc, RL_CFG1, v);
3436 
3437 	v = CSR_READ_1(sc, RL_CFG3);
3438 	v &= ~(RL_CFG3_WOL_LINK | RL_CFG3_WOL_MAGIC);
3439 	if ((ifp->if_capenable & IFCAP_WOL_MAGIC) != 0)
3440 		v |= RL_CFG3_WOL_MAGIC;
3441 	CSR_WRITE_1(sc, RL_CFG3, v);
3442 
3443 	/* Config register write done. */
3444 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3445 
3446 	v = CSR_READ_1(sc, RL_CFG5);
3447 	v &= ~(RL_CFG5_WOL_BCAST | RL_CFG5_WOL_MCAST | RL_CFG5_WOL_UCAST);
3448 	v &= ~RL_CFG5_WOL_LANWAKE;
3449 	if ((ifp->if_capenable & IFCAP_WOL_UCAST) != 0)
3450 		v |= RL_CFG5_WOL_UCAST;
3451 	if ((ifp->if_capenable & IFCAP_WOL_MCAST) != 0)
3452 		v |= RL_CFG5_WOL_MCAST | RL_CFG5_WOL_BCAST;
3453 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3454 		v |= RL_CFG5_WOL_LANWAKE;
3455 	CSR_WRITE_1(sc, RL_CFG5, v);
3456 
3457 	if ((ifp->if_capenable & IFCAP_WOL) != 0 &&
3458 	    (sc->rl_flags & RL_FLAG_PHYWAKE_PM) != 0)
3459 		CSR_WRITE_1(sc, RL_PMCH, CSR_READ_1(sc, RL_PMCH) & ~0x80);
3460 	/*
3461 	 * It seems that hardware resets its link speed to 100Mbps in
3462 	 * power down mode so switching to 100Mbps in driver is not
3463 	 * needed.
3464 	 */
3465 
3466 	/* Request PME if WOL is requested. */
3467 	pmstat = pci_read_config(sc->rl_dev, pmc + PCIR_POWER_STATUS, 2);
3468 	pmstat &= ~(PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE);
3469 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3470 		pmstat |= PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE;
3471 	pci_write_config(sc->rl_dev, pmc + PCIR_POWER_STATUS, pmstat, 2);
3472 }
3473 
3474 static void
3475 re_clrwol(struct rl_softc *sc)
3476 {
3477 	int			pmc;
3478 	uint8_t			v;
3479 
3480 	RL_LOCK_ASSERT(sc);
3481 
3482 	if (pci_find_extcap(sc->rl_dev, PCIY_PMG, &pmc) != 0)
3483 		return;
3484 
3485 	/* Enable config register write. */
3486 	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
3487 
3488 	v = CSR_READ_1(sc, RL_CFG3);
3489 	v &= ~(RL_CFG3_WOL_LINK | RL_CFG3_WOL_MAGIC);
3490 	CSR_WRITE_1(sc, RL_CFG3, v);
3491 
3492 	/* Config register write done. */
3493 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3494 
3495 	v = CSR_READ_1(sc, RL_CFG5);
3496 	v &= ~(RL_CFG5_WOL_BCAST | RL_CFG5_WOL_MCAST | RL_CFG5_WOL_UCAST);
3497 	v &= ~RL_CFG5_WOL_LANWAKE;
3498 	CSR_WRITE_1(sc, RL_CFG5, v);
3499 }
3500 
3501 static void
3502 re_add_sysctls(struct rl_softc *sc)
3503 {
3504 	struct sysctl_ctx_list	*ctx;
3505 	struct sysctl_oid_list	*children;
3506 
3507 	ctx = device_get_sysctl_ctx(sc->rl_dev);
3508 	children = SYSCTL_CHILDREN(device_get_sysctl_tree(sc->rl_dev));
3509 
3510 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "stats",
3511 	    CTLTYPE_INT | CTLFLAG_RW, sc, 0, re_sysctl_stats, "I",
3512 	    "Statistics Information");
3513 }
3514 
3515 static int
3516 re_sysctl_stats(SYSCTL_HANDLER_ARGS)
3517 {
3518 	struct rl_softc		*sc;
3519 	struct rl_stats		*stats;
3520 	int			error, i, result;
3521 
3522 	result = -1;
3523 	error = sysctl_handle_int(oidp, &result, 0, req);
3524 	if (error || req->newptr == NULL)
3525 		return (error);
3526 
3527 	if (result == 1) {
3528 		sc = (struct rl_softc *)arg1;
3529 		RL_LOCK(sc);
3530 		if ((sc->rl_ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
3531 			RL_UNLOCK(sc);
3532 			goto done;
3533 		}
3534 		bus_dmamap_sync(sc->rl_ldata.rl_stag,
3535 		    sc->rl_ldata.rl_smap, BUS_DMASYNC_PREREAD);
3536 		CSR_WRITE_4(sc, RL_DUMPSTATS_HI,
3537 		    RL_ADDR_HI(sc->rl_ldata.rl_stats_addr));
3538 		CSR_WRITE_4(sc, RL_DUMPSTATS_LO,
3539 		    RL_ADDR_LO(sc->rl_ldata.rl_stats_addr));
3540 		CSR_WRITE_4(sc, RL_DUMPSTATS_LO,
3541 		    RL_ADDR_LO(sc->rl_ldata.rl_stats_addr |
3542 		    RL_DUMPSTATS_START));
3543 		for (i = RL_TIMEOUT; i > 0; i--) {
3544 			if ((CSR_READ_4(sc, RL_DUMPSTATS_LO) &
3545 			    RL_DUMPSTATS_START) == 0)
3546 				break;
3547 			DELAY(1000);
3548 		}
3549 		bus_dmamap_sync(sc->rl_ldata.rl_stag,
3550 		    sc->rl_ldata.rl_smap, BUS_DMASYNC_POSTREAD);
3551 		RL_UNLOCK(sc);
3552 		if (i == 0) {
3553 			device_printf(sc->rl_dev,
3554 			    "DUMP statistics request timedout\n");
3555 			return (ETIMEDOUT);
3556 		}
3557 done:
3558 		stats = sc->rl_ldata.rl_stats;
3559 		printf("%s statistics:\n", device_get_nameunit(sc->rl_dev));
3560 		printf("Tx frames : %ju\n",
3561 		    (uintmax_t)le64toh(stats->rl_tx_pkts));
3562 		printf("Rx frames : %ju\n",
3563 		    (uintmax_t)le64toh(stats->rl_rx_pkts));
3564 		printf("Tx errors : %ju\n",
3565 		    (uintmax_t)le64toh(stats->rl_tx_errs));
3566 		printf("Rx errors : %u\n",
3567 		    le32toh(stats->rl_rx_errs));
3568 		printf("Rx missed frames : %u\n",
3569 		    (uint32_t)le16toh(stats->rl_missed_pkts));
3570 		printf("Rx frame alignment errs : %u\n",
3571 		    (uint32_t)le16toh(stats->rl_rx_framealign_errs));
3572 		printf("Tx single collisions : %u\n",
3573 		    le32toh(stats->rl_tx_onecoll));
3574 		printf("Tx multiple collisions : %u\n",
3575 		    le32toh(stats->rl_tx_multicolls));
3576 		printf("Rx unicast frames : %ju\n",
3577 		    (uintmax_t)le64toh(stats->rl_rx_ucasts));
3578 		printf("Rx broadcast frames : %ju\n",
3579 		    (uintmax_t)le64toh(stats->rl_rx_bcasts));
3580 		printf("Rx multicast frames : %u\n",
3581 		    le32toh(stats->rl_rx_mcasts));
3582 		printf("Tx aborts : %u\n",
3583 		    (uint32_t)le16toh(stats->rl_tx_aborts));
3584 		printf("Tx underruns : %u\n",
3585 		    (uint32_t)le16toh(stats->rl_rx_underruns));
3586 	}
3587 
3588 	return (error);
3589 }
3590