xref: /freebsd/sys/dev/re/if_re.c (revision 30d239bc4c510432e65a84fa1c14ed67a3ab1c92)
1 /*-
2  * Copyright (c) 1997, 1998-2003
3  *	Bill Paul <wpaul@windriver.com>.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. All advertising materials mentioning features or use of this software
14  *    must display the following acknowledgement:
15  *	This product includes software developed by Bill Paul.
16  * 4. Neither the name of the author nor the names of any co-contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY Bill Paul AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL Bill Paul OR THE VOICES IN HIS HEAD
24  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
30  * THE POSSIBILITY OF SUCH DAMAGE.
31  */
32 
33 #include <sys/cdefs.h>
34 __FBSDID("$FreeBSD$");
35 
36 /*
37  * RealTek 8139C+/8169/8169S/8110S/8168/8111/8101E PCI NIC driver
38  *
39  * Written by Bill Paul <wpaul@windriver.com>
40  * Senior Networking Software Engineer
41  * Wind River Systems
42  */
43 
44 /*
45  * This driver is designed to support RealTek's next generation of
46  * 10/100 and 10/100/1000 PCI ethernet controllers. There are currently
47  * seven devices in this family: the RTL8139C+, the RTL8169, the RTL8169S,
48  * RTL8110S, the RTL8168, the RTL8111 and the RTL8101E.
49  *
50  * The 8139C+ is a 10/100 ethernet chip. It is backwards compatible
51  * with the older 8139 family, however it also supports a special
52  * C+ mode of operation that provides several new performance enhancing
53  * features. These include:
54  *
55  *	o Descriptor based DMA mechanism. Each descriptor represents
56  *	  a single packet fragment. Data buffers may be aligned on
57  *	  any byte boundary.
58  *
59  *	o 64-bit DMA
60  *
61  *	o TCP/IP checksum offload for both RX and TX
62  *
63  *	o High and normal priority transmit DMA rings
64  *
65  *	o VLAN tag insertion and extraction
66  *
67  *	o TCP large send (segmentation offload)
68  *
69  * Like the 8139, the 8139C+ also has a built-in 10/100 PHY. The C+
70  * programming API is fairly straightforward. The RX filtering, EEPROM
71  * access and PHY access is the same as it is on the older 8139 series
72  * chips.
73  *
74  * The 8169 is a 64-bit 10/100/1000 gigabit ethernet MAC. It has almost the
75  * same programming API and feature set as the 8139C+ with the following
76  * differences and additions:
77  *
78  *	o 1000Mbps mode
79  *
80  *	o Jumbo frames
81  *
82  *	o GMII and TBI ports/registers for interfacing with copper
83  *	  or fiber PHYs
84  *
85  *	o RX and TX DMA rings can have up to 1024 descriptors
86  *	  (the 8139C+ allows a maximum of 64)
87  *
88  *	o Slight differences in register layout from the 8139C+
89  *
90  * The TX start and timer interrupt registers are at different locations
91  * on the 8169 than they are on the 8139C+. Also, the status word in the
92  * RX descriptor has a slightly different bit layout. The 8169 does not
93  * have a built-in PHY. Most reference boards use a Marvell 88E1000 'Alaska'
94  * copper gigE PHY.
95  *
96  * The 8169S/8110S 10/100/1000 devices have built-in copper gigE PHYs
97  * (the 'S' stands for 'single-chip'). These devices have the same
98  * programming API as the older 8169, but also have some vendor-specific
99  * registers for the on-board PHY. The 8110S is a LAN-on-motherboard
100  * part designed to be pin-compatible with the RealTek 8100 10/100 chip.
101  *
102  * This driver takes advantage of the RX and TX checksum offload and
103  * VLAN tag insertion/extraction features. It also implements TX
104  * interrupt moderation using the timer interrupt registers, which
105  * significantly reduces TX interrupt load. There is also support
106  * for jumbo frames, however the 8169/8169S/8110S can not transmit
107  * jumbo frames larger than 7440, so the max MTU possible with this
108  * driver is 7422 bytes.
109  */
110 
111 #ifdef HAVE_KERNEL_OPTION_HEADERS
112 #include "opt_device_polling.h"
113 #endif
114 
115 #include <sys/param.h>
116 #include <sys/endian.h>
117 #include <sys/systm.h>
118 #include <sys/sockio.h>
119 #include <sys/mbuf.h>
120 #include <sys/malloc.h>
121 #include <sys/module.h>
122 #include <sys/kernel.h>
123 #include <sys/socket.h>
124 #include <sys/lock.h>
125 #include <sys/mutex.h>
126 #include <sys/taskqueue.h>
127 
128 #include <net/if.h>
129 #include <net/if_arp.h>
130 #include <net/ethernet.h>
131 #include <net/if_dl.h>
132 #include <net/if_media.h>
133 #include <net/if_types.h>
134 #include <net/if_vlan_var.h>
135 
136 #include <net/bpf.h>
137 
138 #include <machine/bus.h>
139 #include <machine/resource.h>
140 #include <sys/bus.h>
141 #include <sys/rman.h>
142 
143 #include <dev/mii/mii.h>
144 #include <dev/mii/miivar.h>
145 
146 #include <dev/pci/pcireg.h>
147 #include <dev/pci/pcivar.h>
148 
149 MODULE_DEPEND(re, pci, 1, 1, 1);
150 MODULE_DEPEND(re, ether, 1, 1, 1);
151 MODULE_DEPEND(re, miibus, 1, 1, 1);
152 
153 /* "device miibus" required.  See GENERIC if you get errors here. */
154 #include "miibus_if.h"
155 
156 /*
157  * Default to using PIO access for this driver.
158  */
159 #define RE_USEIOSPACE
160 
161 #include <pci/if_rlreg.h>
162 
163 /* Tunables. */
164 static int msi_disable = 0;
165 TUNABLE_INT("hw.re.msi_disable", &msi_disable);
166 
167 #define RE_CSUM_FEATURES    (CSUM_IP | CSUM_TCP | CSUM_UDP)
168 
169 /*
170  * Various supported device vendors/types and their names.
171  */
172 static struct rl_type re_devs[] = {
173 	{ DLINK_VENDORID, DLINK_DEVICEID_528T, RL_HWREV_8169S,
174 		"D-Link DGE-528(T) Gigabit Ethernet Adapter" },
175 	{ RT_VENDORID, RT_DEVICEID_8139, RL_HWREV_8139CPLUS,
176 		"RealTek 8139C+ 10/100BaseTX" },
177 	{ RT_VENDORID, RT_DEVICEID_8101E, RL_HWREV_8101E,
178 		"RealTek 8101E PCIe 10/100baseTX" },
179 	{ RT_VENDORID, RT_DEVICEID_8168, RL_HWREV_8168_SPIN1,
180 		"RealTek 8168/8111B PCIe Gigabit Ethernet" },
181 	{ RT_VENDORID, RT_DEVICEID_8168, RL_HWREV_8168_SPIN2,
182 		"RealTek 8168/8111B PCIe Gigabit Ethernet" },
183 	{ RT_VENDORID, RT_DEVICEID_8169, RL_HWREV_8169,
184 		"RealTek 8169 Gigabit Ethernet" },
185 	{ RT_VENDORID, RT_DEVICEID_8169, RL_HWREV_8169S,
186 		"RealTek 8169S Single-chip Gigabit Ethernet" },
187 	{ RT_VENDORID, RT_DEVICEID_8169, RL_HWREV_8169_8110SB,
188 		"RealTek 8169SB/8110SB Single-chip Gigabit Ethernet" },
189 	{ RT_VENDORID, RT_DEVICEID_8169, RL_HWREV_8169_8110SC,
190 		"RealTek 8169SC/8110SC Single-chip Gigabit Ethernet" },
191 	{ RT_VENDORID, RT_DEVICEID_8169SC, RL_HWREV_8169_8110SC,
192 		"RealTek 8169SC/8110SC Single-chip Gigabit Ethernet" },
193 	{ RT_VENDORID, RT_DEVICEID_8169, RL_HWREV_8110S,
194 		"RealTek 8110S Single-chip Gigabit Ethernet" },
195 	{ COREGA_VENDORID, COREGA_DEVICEID_CGLAPCIGT, RL_HWREV_8169S,
196 		"Corega CG-LAPCIGT (RTL8169S) Gigabit Ethernet" },
197 	{ LINKSYS_VENDORID, LINKSYS_DEVICEID_EG1032, RL_HWREV_8169S,
198 		"Linksys EG1032 (RTL8169S) Gigabit Ethernet" },
199 	{ USR_VENDORID, USR_DEVICEID_997902, RL_HWREV_8169S,
200 		"US Robotics 997902 (RTL8169S) Gigabit Ethernet" },
201 	{ 0, 0, 0, NULL }
202 };
203 
204 static struct rl_hwrev re_hwrevs[] = {
205 	{ RL_HWREV_8139, RL_8139,  "" },
206 	{ RL_HWREV_8139A, RL_8139, "A" },
207 	{ RL_HWREV_8139AG, RL_8139, "A-G" },
208 	{ RL_HWREV_8139B, RL_8139, "B" },
209 	{ RL_HWREV_8130, RL_8139, "8130" },
210 	{ RL_HWREV_8139C, RL_8139, "C" },
211 	{ RL_HWREV_8139D, RL_8139, "8139D/8100B/8100C" },
212 	{ RL_HWREV_8139CPLUS, RL_8139CPLUS, "C+"},
213 	{ RL_HWREV_8168_SPIN1, RL_8169, "8168"},
214 	{ RL_HWREV_8169, RL_8169, "8169"},
215 	{ RL_HWREV_8169S, RL_8169, "8169S"},
216 	{ RL_HWREV_8110S, RL_8169, "8110S"},
217 	{ RL_HWREV_8169_8110SB, RL_8169, "8169SB"},
218 	{ RL_HWREV_8169_8110SC, RL_8169, "8169SC"},
219 	{ RL_HWREV_8100, RL_8139, "8100"},
220 	{ RL_HWREV_8101, RL_8139, "8101"},
221 	{ RL_HWREV_8100E, RL_8169, "8100E"},
222 	{ RL_HWREV_8101E, RL_8169, "8101E"},
223 	{ RL_HWREV_8168_SPIN2, RL_8169, "8168"},
224 	{ 0, 0, NULL }
225 };
226 
227 static int re_probe		(device_t);
228 static int re_attach		(device_t);
229 static int re_detach		(device_t);
230 
231 static int re_encap		(struct rl_softc *, struct mbuf **, int *);
232 
233 static void re_dma_map_addr	(void *, bus_dma_segment_t *, int, int);
234 static void re_dma_map_desc	(void *, bus_dma_segment_t *, int,
235 				    bus_size_t, int);
236 static int re_allocmem		(device_t, struct rl_softc *);
237 static int re_newbuf		(struct rl_softc *, int, struct mbuf *);
238 static int re_rx_list_init	(struct rl_softc *);
239 static int re_tx_list_init	(struct rl_softc *);
240 #ifdef RE_FIXUP_RX
241 static __inline void re_fixup_rx
242 				(struct mbuf *);
243 #endif
244 static int re_rxeof		(struct rl_softc *);
245 static void re_txeof		(struct rl_softc *);
246 #ifdef DEVICE_POLLING
247 static void re_poll		(struct ifnet *, enum poll_cmd, int);
248 static void re_poll_locked	(struct ifnet *, enum poll_cmd, int);
249 #endif
250 static int re_intr		(void *);
251 static void re_tick		(void *);
252 static void re_tx_task		(void *, int);
253 static void re_int_task		(void *, int);
254 static void re_start		(struct ifnet *);
255 static int re_ioctl		(struct ifnet *, u_long, caddr_t);
256 static void re_init		(void *);
257 static void re_init_locked	(struct rl_softc *);
258 static void re_stop		(struct rl_softc *);
259 static void re_watchdog		(struct rl_softc *);
260 static int re_suspend		(device_t);
261 static int re_resume		(device_t);
262 static void re_shutdown		(device_t);
263 static int re_ifmedia_upd	(struct ifnet *);
264 static void re_ifmedia_sts	(struct ifnet *, struct ifmediareq *);
265 
266 static void re_eeprom_putbyte	(struct rl_softc *, int);
267 static void re_eeprom_getword	(struct rl_softc *, int, u_int16_t *);
268 static void re_read_eeprom	(struct rl_softc *, caddr_t, int, int);
269 static int re_gmii_readreg	(device_t, int, int);
270 static int re_gmii_writereg	(device_t, int, int, int);
271 
272 static int re_miibus_readreg	(device_t, int, int);
273 static int re_miibus_writereg	(device_t, int, int, int);
274 static void re_miibus_statchg	(device_t);
275 
276 static void re_setmulti		(struct rl_softc *);
277 static void re_reset		(struct rl_softc *);
278 
279 #ifdef RE_DIAG
280 static int re_diag		(struct rl_softc *);
281 #endif
282 
283 #ifdef RE_USEIOSPACE
284 #define RL_RES			SYS_RES_IOPORT
285 #define RL_RID			RL_PCI_LOIO
286 #else
287 #define RL_RES			SYS_RES_MEMORY
288 #define RL_RID			RL_PCI_LOMEM
289 #endif
290 
291 static device_method_t re_methods[] = {
292 	/* Device interface */
293 	DEVMETHOD(device_probe,		re_probe),
294 	DEVMETHOD(device_attach,	re_attach),
295 	DEVMETHOD(device_detach,	re_detach),
296 	DEVMETHOD(device_suspend,	re_suspend),
297 	DEVMETHOD(device_resume,	re_resume),
298 	DEVMETHOD(device_shutdown,	re_shutdown),
299 
300 	/* bus interface */
301 	DEVMETHOD(bus_print_child,	bus_generic_print_child),
302 	DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
303 
304 	/* MII interface */
305 	DEVMETHOD(miibus_readreg,	re_miibus_readreg),
306 	DEVMETHOD(miibus_writereg,	re_miibus_writereg),
307 	DEVMETHOD(miibus_statchg,	re_miibus_statchg),
308 
309 	{ 0, 0 }
310 };
311 
312 static driver_t re_driver = {
313 	"re",
314 	re_methods,
315 	sizeof(struct rl_softc)
316 };
317 
318 static devclass_t re_devclass;
319 
320 DRIVER_MODULE(re, pci, re_driver, re_devclass, 0, 0);
321 DRIVER_MODULE(re, cardbus, 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(sc, addr)
337 	struct rl_softc		*sc;
338 	int			addr;
339 {
340 	register int		d, i;
341 
342 	d = addr | (RL_9346_READ << sc->rl_eewidth);
343 
344 	/*
345 	 * Feed in each bit and strobe the clock.
346 	 */
347 
348 	for (i = 1 << (sc->rl_eewidth + 3); i; i >>= 1) {
349 		if (d & i) {
350 			EE_SET(RL_EE_DATAIN);
351 		} else {
352 			EE_CLR(RL_EE_DATAIN);
353 		}
354 		DELAY(100);
355 		EE_SET(RL_EE_CLK);
356 		DELAY(150);
357 		EE_CLR(RL_EE_CLK);
358 		DELAY(100);
359 	}
360 
361 	return;
362 }
363 
364 /*
365  * Read a word of data stored in the EEPROM at address 'addr.'
366  */
367 static void
368 re_eeprom_getword(sc, addr, dest)
369 	struct rl_softc		*sc;
370 	int			addr;
371 	u_int16_t		*dest;
372 {
373 	register int		i;
374 	u_int16_t		word = 0;
375 
376 	/*
377 	 * Send address of word we want to read.
378 	 */
379 	re_eeprom_putbyte(sc, addr);
380 
381 	/*
382 	 * Start reading bits from EEPROM.
383 	 */
384 	for (i = 0x8000; i; i >>= 1) {
385 		EE_SET(RL_EE_CLK);
386 		DELAY(100);
387 		if (CSR_READ_1(sc, RL_EECMD) & RL_EE_DATAOUT)
388 			word |= i;
389 		EE_CLR(RL_EE_CLK);
390 		DELAY(100);
391 	}
392 
393 	*dest = word;
394 
395 	return;
396 }
397 
398 /*
399  * Read a sequence of words from the EEPROM.
400  */
401 static void
402 re_read_eeprom(sc, dest, off, cnt)
403 	struct rl_softc		*sc;
404 	caddr_t			dest;
405 	int			off;
406 	int			cnt;
407 {
408 	int			i;
409 	u_int16_t		word = 0, *ptr;
410 
411 	CSR_SETBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM);
412 
413         DELAY(100);
414 
415 	for (i = 0; i < cnt; i++) {
416 		CSR_SETBIT_1(sc, RL_EECMD, RL_EE_SEL);
417 		re_eeprom_getword(sc, off + i, &word);
418 		CSR_CLRBIT_1(sc, RL_EECMD, RL_EE_SEL);
419 		ptr = (u_int16_t *)(dest + (i * 2));
420                 *ptr = word;
421 	}
422 
423 	CSR_CLRBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM);
424 
425 	return;
426 }
427 
428 static int
429 re_gmii_readreg(dev, phy, reg)
430 	device_t		dev;
431 	int			phy, reg;
432 {
433 	struct rl_softc		*sc;
434 	u_int32_t		rval;
435 	int			i;
436 
437 	if (phy != 1)
438 		return (0);
439 
440 	sc = device_get_softc(dev);
441 
442 	/* Let the rgephy driver read the GMEDIASTAT register */
443 
444 	if (reg == RL_GMEDIASTAT) {
445 		rval = CSR_READ_1(sc, RL_GMEDIASTAT);
446 		return (rval);
447 	}
448 
449 	CSR_WRITE_4(sc, RL_PHYAR, reg << 16);
450 	DELAY(1000);
451 
452 	for (i = 0; i < RL_TIMEOUT; i++) {
453 		rval = CSR_READ_4(sc, RL_PHYAR);
454 		if (rval & RL_PHYAR_BUSY)
455 			break;
456 		DELAY(100);
457 	}
458 
459 	if (i == RL_TIMEOUT) {
460 		device_printf(sc->rl_dev, "PHY read failed\n");
461 		return (0);
462 	}
463 
464 	return (rval & RL_PHYAR_PHYDATA);
465 }
466 
467 static int
468 re_gmii_writereg(dev, phy, reg, data)
469 	device_t		dev;
470 	int			phy, reg, data;
471 {
472 	struct rl_softc		*sc;
473 	u_int32_t		rval;
474 	int			i;
475 
476 	sc = device_get_softc(dev);
477 
478 	CSR_WRITE_4(sc, RL_PHYAR, (reg << 16) |
479 	    (data & RL_PHYAR_PHYDATA) | RL_PHYAR_BUSY);
480 	DELAY(1000);
481 
482 	for (i = 0; i < RL_TIMEOUT; i++) {
483 		rval = CSR_READ_4(sc, RL_PHYAR);
484 		if (!(rval & RL_PHYAR_BUSY))
485 			break;
486 		DELAY(100);
487 	}
488 
489 	if (i == RL_TIMEOUT) {
490 		device_printf(sc->rl_dev, "PHY write failed\n");
491 		return (0);
492 	}
493 
494 	return (0);
495 }
496 
497 static int
498 re_miibus_readreg(dev, phy, reg)
499 	device_t		dev;
500 	int			phy, reg;
501 {
502 	struct rl_softc		*sc;
503 	u_int16_t		rval = 0;
504 	u_int16_t		re8139_reg = 0;
505 
506 	sc = device_get_softc(dev);
507 
508 	if (sc->rl_type == RL_8169) {
509 		rval = re_gmii_readreg(dev, phy, reg);
510 		return (rval);
511 	}
512 
513 	/* Pretend the internal PHY is only at address 0 */
514 	if (phy) {
515 		return (0);
516 	}
517 	switch (reg) {
518 	case MII_BMCR:
519 		re8139_reg = RL_BMCR;
520 		break;
521 	case MII_BMSR:
522 		re8139_reg = RL_BMSR;
523 		break;
524 	case MII_ANAR:
525 		re8139_reg = RL_ANAR;
526 		break;
527 	case MII_ANER:
528 		re8139_reg = RL_ANER;
529 		break;
530 	case MII_ANLPAR:
531 		re8139_reg = RL_LPAR;
532 		break;
533 	case MII_PHYIDR1:
534 	case MII_PHYIDR2:
535 		return (0);
536 	/*
537 	 * Allow the rlphy driver to read the media status
538 	 * register. If we have a link partner which does not
539 	 * support NWAY, this is the register which will tell
540 	 * us the results of parallel detection.
541 	 */
542 	case RL_MEDIASTAT:
543 		rval = CSR_READ_1(sc, RL_MEDIASTAT);
544 		return (rval);
545 	default:
546 		device_printf(sc->rl_dev, "bad phy register\n");
547 		return (0);
548 	}
549 	rval = CSR_READ_2(sc, re8139_reg);
550 	if (sc->rl_type == RL_8139CPLUS && re8139_reg == RL_BMCR) {
551 		/* 8139C+ has different bit layout. */
552 		rval &= ~(BMCR_LOOP | BMCR_ISO);
553 	}
554 	return (rval);
555 }
556 
557 static int
558 re_miibus_writereg(dev, phy, reg, data)
559 	device_t		dev;
560 	int			phy, reg, data;
561 {
562 	struct rl_softc		*sc;
563 	u_int16_t		re8139_reg = 0;
564 	int			rval = 0;
565 
566 	sc = device_get_softc(dev);
567 
568 	if (sc->rl_type == RL_8169) {
569 		rval = re_gmii_writereg(dev, phy, reg, data);
570 		return (rval);
571 	}
572 
573 	/* Pretend the internal PHY is only at address 0 */
574 	if (phy)
575 		return (0);
576 
577 	switch (reg) {
578 	case MII_BMCR:
579 		re8139_reg = RL_BMCR;
580 		if (sc->rl_type == RL_8139CPLUS) {
581 			/* 8139C+ has different bit layout. */
582 			data &= ~(BMCR_LOOP | BMCR_ISO);
583 		}
584 		break;
585 	case MII_BMSR:
586 		re8139_reg = RL_BMSR;
587 		break;
588 	case MII_ANAR:
589 		re8139_reg = RL_ANAR;
590 		break;
591 	case MII_ANER:
592 		re8139_reg = RL_ANER;
593 		break;
594 	case MII_ANLPAR:
595 		re8139_reg = RL_LPAR;
596 		break;
597 	case MII_PHYIDR1:
598 	case MII_PHYIDR2:
599 		return (0);
600 		break;
601 	default:
602 		device_printf(sc->rl_dev, "bad phy register\n");
603 		return (0);
604 	}
605 	CSR_WRITE_2(sc, re8139_reg, data);
606 	return (0);
607 }
608 
609 static void
610 re_miibus_statchg(dev)
611 	device_t		dev;
612 {
613 
614 }
615 
616 /*
617  * Program the 64-bit multicast hash filter.
618  */
619 static void
620 re_setmulti(sc)
621 	struct rl_softc		*sc;
622 {
623 	struct ifnet		*ifp;
624 	int			h = 0;
625 	u_int32_t		hashes[2] = { 0, 0 };
626 	struct ifmultiaddr	*ifma;
627 	u_int32_t		rxfilt;
628 	int			mcnt = 0;
629 	u_int32_t		hwrev;
630 
631 	RL_LOCK_ASSERT(sc);
632 
633 	ifp = sc->rl_ifp;
634 
635 	rxfilt = CSR_READ_4(sc, RL_RXCFG);
636 
637 	if (ifp->if_flags & IFF_ALLMULTI || ifp->if_flags & IFF_PROMISC) {
638 		rxfilt |= RL_RXCFG_RX_MULTI;
639 		CSR_WRITE_4(sc, RL_RXCFG, rxfilt);
640 		CSR_WRITE_4(sc, RL_MAR0, 0xFFFFFFFF);
641 		CSR_WRITE_4(sc, RL_MAR4, 0xFFFFFFFF);
642 		return;
643 	}
644 
645 	/* first, zot all the existing hash bits */
646 	CSR_WRITE_4(sc, RL_MAR0, 0);
647 	CSR_WRITE_4(sc, RL_MAR4, 0);
648 
649 	/* now program new ones */
650 	IF_ADDR_LOCK(ifp);
651 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
652 		if (ifma->ifma_addr->sa_family != AF_LINK)
653 			continue;
654 		h = ether_crc32_be(LLADDR((struct sockaddr_dl *)
655 		    ifma->ifma_addr), ETHER_ADDR_LEN) >> 26;
656 		if (h < 32)
657 			hashes[0] |= (1 << h);
658 		else
659 			hashes[1] |= (1 << (h - 32));
660 		mcnt++;
661 	}
662 	IF_ADDR_UNLOCK(ifp);
663 
664 	if (mcnt)
665 		rxfilt |= RL_RXCFG_RX_MULTI;
666 	else
667 		rxfilt &= ~RL_RXCFG_RX_MULTI;
668 
669 	CSR_WRITE_4(sc, RL_RXCFG, rxfilt);
670 
671 	/*
672 	 * For some unfathomable reason, RealTek decided to reverse
673 	 * the order of the multicast hash registers in the PCI Express
674 	 * parts. This means we have to write the hash pattern in reverse
675 	 * order for those devices.
676 	 */
677 
678 	hwrev = CSR_READ_4(sc, RL_TXCFG) & RL_TXCFG_HWREV;
679 
680 	if (hwrev == RL_HWREV_8100E || hwrev == RL_HWREV_8101E ||
681 	    hwrev == RL_HWREV_8168_SPIN1 || hwrev == RL_HWREV_8168_SPIN2) {
682 		CSR_WRITE_4(sc, RL_MAR0, bswap32(hashes[1]));
683 		CSR_WRITE_4(sc, RL_MAR4, bswap32(hashes[0]));
684 	} else {
685 		CSR_WRITE_4(sc, RL_MAR0, hashes[0]);
686 		CSR_WRITE_4(sc, RL_MAR4, hashes[1]);
687 	}
688 }
689 
690 static void
691 re_reset(sc)
692 	struct rl_softc		*sc;
693 {
694 	register int		i;
695 
696 	RL_LOCK_ASSERT(sc);
697 
698 	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RESET);
699 
700 	for (i = 0; i < RL_TIMEOUT; i++) {
701 		DELAY(10);
702 		if (!(CSR_READ_1(sc, RL_COMMAND) & RL_CMD_RESET))
703 			break;
704 	}
705 	if (i == RL_TIMEOUT)
706 		device_printf(sc->rl_dev, "reset never completed!\n");
707 
708 	CSR_WRITE_1(sc, 0x82, 1);
709 }
710 
711 #ifdef RE_DIAG
712 
713 /*
714  * The following routine is designed to test for a defect on some
715  * 32-bit 8169 cards. Some of these NICs have the REQ64# and ACK64#
716  * lines connected to the bus, however for a 32-bit only card, they
717  * should be pulled high. The result of this defect is that the
718  * NIC will not work right if you plug it into a 64-bit slot: DMA
719  * operations will be done with 64-bit transfers, which will fail
720  * because the 64-bit data lines aren't connected.
721  *
722  * There's no way to work around this (short of talking a soldering
723  * iron to the board), however we can detect it. The method we use
724  * here is to put the NIC into digital loopback mode, set the receiver
725  * to promiscuous mode, and then try to send a frame. We then compare
726  * the frame data we sent to what was received. If the data matches,
727  * then the NIC is working correctly, otherwise we know the user has
728  * a defective NIC which has been mistakenly plugged into a 64-bit PCI
729  * slot. In the latter case, there's no way the NIC can work correctly,
730  * so we print out a message on the console and abort the device attach.
731  */
732 
733 static int
734 re_diag(sc)
735 	struct rl_softc		*sc;
736 {
737 	struct ifnet		*ifp = sc->rl_ifp;
738 	struct mbuf		*m0;
739 	struct ether_header	*eh;
740 	struct rl_desc		*cur_rx;
741 	u_int16_t		status;
742 	u_int32_t		rxstat;
743 	int			total_len, i, error = 0, phyaddr;
744 	u_int8_t		dst[] = { 0x00, 'h', 'e', 'l', 'l', 'o' };
745 	u_int8_t		src[] = { 0x00, 'w', 'o', 'r', 'l', 'd' };
746 
747 	/* Allocate a single mbuf */
748 	MGETHDR(m0, M_DONTWAIT, MT_DATA);
749 	if (m0 == NULL)
750 		return (ENOBUFS);
751 
752 	RL_LOCK(sc);
753 
754 	/*
755 	 * Initialize the NIC in test mode. This sets the chip up
756 	 * so that it can send and receive frames, but performs the
757 	 * following special functions:
758 	 * - Puts receiver in promiscuous mode
759 	 * - Enables digital loopback mode
760 	 * - Leaves interrupts turned off
761 	 */
762 
763 	ifp->if_flags |= IFF_PROMISC;
764 	sc->rl_testmode = 1;
765 	re_reset(sc);
766 	re_init_locked(sc);
767 	sc->rl_link = 1;
768 	if (sc->rl_type == RL_8169)
769 		phyaddr = 1;
770 	else
771 		phyaddr = 0;
772 
773 	re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_RESET);
774 	for (i = 0; i < RL_TIMEOUT; i++) {
775 		status = re_miibus_readreg(sc->rl_dev, phyaddr, MII_BMCR);
776 		if (!(status & BMCR_RESET))
777 			break;
778 	}
779 
780 	re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_LOOP);
781 	CSR_WRITE_2(sc, RL_ISR, RL_INTRS);
782 
783 	DELAY(100000);
784 
785 	/* Put some data in the mbuf */
786 
787 	eh = mtod(m0, struct ether_header *);
788 	bcopy ((char *)&dst, eh->ether_dhost, ETHER_ADDR_LEN);
789 	bcopy ((char *)&src, eh->ether_shost, ETHER_ADDR_LEN);
790 	eh->ether_type = htons(ETHERTYPE_IP);
791 	m0->m_pkthdr.len = m0->m_len = ETHER_MIN_LEN - ETHER_CRC_LEN;
792 
793 	/*
794 	 * Queue the packet, start transmission.
795 	 * Note: IF_HANDOFF() ultimately calls re_start() for us.
796 	 */
797 
798 	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
799 	RL_UNLOCK(sc);
800 	/* XXX: re_diag must not be called when in ALTQ mode */
801 	IF_HANDOFF(&ifp->if_snd, m0, ifp);
802 	RL_LOCK(sc);
803 	m0 = NULL;
804 
805 	/* Wait for it to propagate through the chip */
806 
807 	DELAY(100000);
808 	for (i = 0; i < RL_TIMEOUT; i++) {
809 		status = CSR_READ_2(sc, RL_ISR);
810 		CSR_WRITE_2(sc, RL_ISR, status);
811 		if ((status & (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK)) ==
812 		    (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK))
813 			break;
814 		DELAY(10);
815 	}
816 
817 	if (i == RL_TIMEOUT) {
818 		device_printf(sc->rl_dev,
819 		    "diagnostic failed, failed to receive packet in"
820 		    " loopback mode\n");
821 		error = EIO;
822 		goto done;
823 	}
824 
825 	/*
826 	 * The packet should have been dumped into the first
827 	 * entry in the RX DMA ring. Grab it from there.
828 	 */
829 
830 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
831 	    sc->rl_ldata.rl_rx_list_map,
832 	    BUS_DMASYNC_POSTREAD);
833 	bus_dmamap_sync(sc->rl_ldata.rl_mtag,
834 	    sc->rl_ldata.rl_rx_dmamap[0],
835 	    BUS_DMASYNC_POSTWRITE);
836 	bus_dmamap_unload(sc->rl_ldata.rl_mtag,
837 	    sc->rl_ldata.rl_rx_dmamap[0]);
838 
839 	m0 = sc->rl_ldata.rl_rx_mbuf[0];
840 	sc->rl_ldata.rl_rx_mbuf[0] = NULL;
841 	eh = mtod(m0, struct ether_header *);
842 
843 	cur_rx = &sc->rl_ldata.rl_rx_list[0];
844 	total_len = RL_RXBYTES(cur_rx);
845 	rxstat = le32toh(cur_rx->rl_cmdstat);
846 
847 	if (total_len != ETHER_MIN_LEN) {
848 		device_printf(sc->rl_dev,
849 		    "diagnostic failed, received short packet\n");
850 		error = EIO;
851 		goto done;
852 	}
853 
854 	/* Test that the received packet data matches what we sent. */
855 
856 	if (bcmp((char *)&eh->ether_dhost, (char *)&dst, ETHER_ADDR_LEN) ||
857 	    bcmp((char *)&eh->ether_shost, (char *)&src, ETHER_ADDR_LEN) ||
858 	    ntohs(eh->ether_type) != ETHERTYPE_IP) {
859 		device_printf(sc->rl_dev, "WARNING, DMA FAILURE!\n");
860 		device_printf(sc->rl_dev, "expected TX data: %6D/%6D/0x%x\n",
861 		    dst, ":", src, ":", ETHERTYPE_IP);
862 		device_printf(sc->rl_dev, "received RX data: %6D/%6D/0x%x\n",
863 		    eh->ether_dhost, ":",  eh->ether_shost, ":",
864 		    ntohs(eh->ether_type));
865 		device_printf(sc->rl_dev, "You may have a defective 32-bit "
866 		    "NIC plugged into a 64-bit PCI slot.\n");
867 		device_printf(sc->rl_dev, "Please re-install the NIC in a "
868 		    "32-bit slot for proper operation.\n");
869 		device_printf(sc->rl_dev, "Read the re(4) man page for more "
870 		    "details.\n");
871 		error = EIO;
872 	}
873 
874 done:
875 	/* Turn interface off, release resources */
876 
877 	sc->rl_testmode = 0;
878 	sc->rl_link = 0;
879 	ifp->if_flags &= ~IFF_PROMISC;
880 	re_stop(sc);
881 	if (m0 != NULL)
882 		m_freem(m0);
883 
884 	RL_UNLOCK(sc);
885 
886 	return (error);
887 }
888 
889 #endif
890 
891 /*
892  * Probe for a RealTek 8139C+/8169/8110 chip. Check the PCI vendor and device
893  * IDs against our list and return a device name if we find a match.
894  */
895 static int
896 re_probe(dev)
897 	device_t		dev;
898 {
899 	struct rl_type		*t;
900 	struct rl_softc		*sc;
901 	int			rid;
902 	u_int32_t		hwrev;
903 
904 	t = re_devs;
905 	sc = device_get_softc(dev);
906 
907 	while (t->rl_name != NULL) {
908 		if ((pci_get_vendor(dev) == t->rl_vid) &&
909 		    (pci_get_device(dev) == t->rl_did)) {
910 			/*
911 			 * Only attach to rev. 3 of the Linksys EG1032 adapter.
912 			 * Rev. 2 i supported by sk(4).
913 			 */
914 			if ((t->rl_vid == LINKSYS_VENDORID) &&
915 				(t->rl_did == LINKSYS_DEVICEID_EG1032) &&
916 				(pci_get_subdevice(dev) !=
917 				LINKSYS_SUBDEVICE_EG1032_REV3)) {
918 				t++;
919 				continue;
920 			}
921 
922 			/*
923 			 * Temporarily map the I/O space
924 			 * so we can read the chip ID register.
925 			 */
926 			rid = RL_RID;
927 			sc->rl_res = bus_alloc_resource_any(dev, RL_RES, &rid,
928 			    RF_ACTIVE);
929 			if (sc->rl_res == NULL) {
930 				device_printf(dev,
931 				    "couldn't map ports/memory\n");
932 				return (ENXIO);
933 			}
934 			sc->rl_btag = rman_get_bustag(sc->rl_res);
935 			sc->rl_bhandle = rman_get_bushandle(sc->rl_res);
936 			hwrev = CSR_READ_4(sc, RL_TXCFG) & RL_TXCFG_HWREV;
937 			bus_release_resource(dev, RL_RES,
938 			    RL_RID, sc->rl_res);
939 			if (t->rl_basetype == hwrev) {
940 				device_set_desc(dev, t->rl_name);
941 				return (BUS_PROBE_DEFAULT);
942 			}
943 		}
944 		t++;
945 	}
946 
947 	return (ENXIO);
948 }
949 
950 /*
951  * This routine takes the segment list provided as the result of
952  * a bus_dma_map_load() operation and assigns the addresses/lengths
953  * to RealTek DMA descriptors. This can be called either by the RX
954  * code or the TX code. In the RX case, we'll probably wind up mapping
955  * at most one segment. For the TX case, there could be any number of
956  * segments since TX packets may span multiple mbufs. In either case,
957  * if the number of segments is larger than the rl_maxsegs limit
958  * specified by the caller, we abort the mapping operation. Sadly,
959  * whoever designed the buffer mapping API did not provide a way to
960  * return an error from here, so we have to fake it a bit.
961  */
962 
963 static void
964 re_dma_map_desc(arg, segs, nseg, mapsize, error)
965 	void			*arg;
966 	bus_dma_segment_t	*segs;
967 	int			nseg;
968 	bus_size_t		mapsize;
969 	int			error;
970 {
971 	struct rl_dmaload_arg	*ctx;
972 	struct rl_desc		*d = NULL;
973 	int			i = 0, idx;
974 	u_int32_t		cmdstat;
975 	int			totlen = 0;
976 
977 	if (error)
978 		return;
979 
980 	ctx = arg;
981 
982 	/* Signal error to caller if there's too many segments */
983 	if (nseg > ctx->rl_maxsegs) {
984 		ctx->rl_maxsegs = 0;
985 		return;
986 	}
987 
988 	/*
989 	 * Map the segment array into descriptors. Note that we set the
990 	 * start-of-frame and end-of-frame markers for either TX or RX, but
991 	 * they really only have meaning in the TX case. (In the RX case,
992 	 * it's the chip that tells us where packets begin and end.)
993 	 * We also keep track of the end of the ring and set the
994 	 * end-of-ring bits as needed, and we set the ownership bits
995 	 * in all except the very first descriptor. (The caller will
996 	 * set this descriptor later when it start transmission or
997 	 * reception.)
998 	 */
999 	idx = ctx->rl_idx;
1000 	for (;;) {
1001 		d = &ctx->rl_ring[idx];
1002 		if (le32toh(d->rl_cmdstat) & RL_RDESC_STAT_OWN) {
1003 			ctx->rl_maxsegs = 0;
1004 			return;
1005 		}
1006 		cmdstat = segs[i].ds_len;
1007 		totlen += segs[i].ds_len;
1008 		d->rl_vlanctl = 0;
1009 		d->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[i].ds_addr));
1010 		d->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[i].ds_addr));
1011 		if (i == 0)
1012 			cmdstat |= RL_TDESC_CMD_SOF;
1013 		else
1014 			cmdstat |= RL_TDESC_CMD_OWN;
1015 		if (idx == (RL_RX_DESC_CNT - 1))
1016 			cmdstat |= RL_TDESC_CMD_EOR;
1017 		d->rl_cmdstat = htole32(cmdstat | ctx->rl_flags);
1018 		i++;
1019 		if (i == nseg)
1020 			break;
1021 		RL_DESC_INC(idx);
1022 	}
1023 
1024 	d->rl_cmdstat |= htole32(RL_TDESC_CMD_EOF);
1025 	ctx->rl_maxsegs = nseg;
1026 	ctx->rl_idx = idx;
1027 }
1028 
1029 /*
1030  * Map a single buffer address.
1031  */
1032 
1033 static void
1034 re_dma_map_addr(arg, segs, nseg, error)
1035 	void			*arg;
1036 	bus_dma_segment_t	*segs;
1037 	int			nseg;
1038 	int			error;
1039 {
1040 	bus_addr_t		*addr;
1041 
1042 	if (error)
1043 		return;
1044 
1045 	KASSERT(nseg == 1, ("too many DMA segments, %d should be 1", nseg));
1046 	addr = arg;
1047 	*addr = segs->ds_addr;
1048 }
1049 
1050 static int
1051 re_allocmem(dev, sc)
1052 	device_t		dev;
1053 	struct rl_softc		*sc;
1054 {
1055 	int			error;
1056 	int			nseg;
1057 	int			i;
1058 
1059 	/*
1060 	 * Allocate map for RX mbufs.
1061 	 */
1062 	nseg = 32;
1063 	error = bus_dma_tag_create(sc->rl_parent_tag, ETHER_ALIGN, 0,
1064 	    BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1065 	    NULL, MCLBYTES * nseg, nseg, MCLBYTES, BUS_DMA_ALLOCNOW,
1066 	    NULL, NULL, &sc->rl_ldata.rl_mtag);
1067 	if (error) {
1068 		device_printf(dev, "could not allocate dma tag\n");
1069 		return (ENOMEM);
1070 	}
1071 
1072 	/*
1073 	 * Allocate map for TX descriptor list.
1074 	 */
1075 	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
1076 	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1077 	    NULL, RL_TX_LIST_SZ, 1, RL_TX_LIST_SZ, 0,
1078 	    NULL, NULL, &sc->rl_ldata.rl_tx_list_tag);
1079 	if (error) {
1080 		device_printf(dev, "could not allocate dma tag\n");
1081 		return (ENOMEM);
1082 	}
1083 
1084 	/* Allocate DMA'able memory for the TX ring */
1085 
1086 	error = bus_dmamem_alloc(sc->rl_ldata.rl_tx_list_tag,
1087 	    (void **)&sc->rl_ldata.rl_tx_list, BUS_DMA_NOWAIT | BUS_DMA_ZERO,
1088 	    &sc->rl_ldata.rl_tx_list_map);
1089 	if (error)
1090 		return (ENOMEM);
1091 
1092 	/* Load the map for the TX ring. */
1093 
1094 	error = bus_dmamap_load(sc->rl_ldata.rl_tx_list_tag,
1095 	     sc->rl_ldata.rl_tx_list_map, sc->rl_ldata.rl_tx_list,
1096 	     RL_TX_LIST_SZ, re_dma_map_addr,
1097 	     &sc->rl_ldata.rl_tx_list_addr, BUS_DMA_NOWAIT);
1098 
1099 	/* Create DMA maps for TX buffers */
1100 
1101 	for (i = 0; i < RL_TX_DESC_CNT; i++) {
1102 		error = bus_dmamap_create(sc->rl_ldata.rl_mtag, 0,
1103 			    &sc->rl_ldata.rl_tx_dmamap[i]);
1104 		if (error) {
1105 			device_printf(dev, "can't create DMA map for TX\n");
1106 			return (ENOMEM);
1107 		}
1108 	}
1109 
1110 	/*
1111 	 * Allocate map for RX descriptor list.
1112 	 */
1113 	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
1114 	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1115 	    NULL, RL_RX_LIST_SZ, 1, RL_RX_LIST_SZ, 0,
1116 	    NULL, NULL, &sc->rl_ldata.rl_rx_list_tag);
1117 	if (error) {
1118 		device_printf(dev, "could not allocate dma tag\n");
1119 		return (ENOMEM);
1120 	}
1121 
1122 	/* Allocate DMA'able memory for the RX ring */
1123 
1124 	error = bus_dmamem_alloc(sc->rl_ldata.rl_rx_list_tag,
1125 	    (void **)&sc->rl_ldata.rl_rx_list, BUS_DMA_NOWAIT | BUS_DMA_ZERO,
1126 	    &sc->rl_ldata.rl_rx_list_map);
1127 	if (error)
1128 		return (ENOMEM);
1129 
1130 	/* Load the map for the RX ring. */
1131 
1132 	error = bus_dmamap_load(sc->rl_ldata.rl_rx_list_tag,
1133 	     sc->rl_ldata.rl_rx_list_map, sc->rl_ldata.rl_rx_list,
1134 	     RL_RX_LIST_SZ, re_dma_map_addr,
1135 	     &sc->rl_ldata.rl_rx_list_addr, BUS_DMA_NOWAIT);
1136 
1137 	/* Create DMA maps for RX buffers */
1138 
1139 	for (i = 0; i < RL_RX_DESC_CNT; i++) {
1140 		error = bus_dmamap_create(sc->rl_ldata.rl_mtag, 0,
1141 			    &sc->rl_ldata.rl_rx_dmamap[i]);
1142 		if (error) {
1143 			device_printf(dev, "can't create DMA map for RX\n");
1144 			return (ENOMEM);
1145 		}
1146 	}
1147 
1148 	return (0);
1149 }
1150 
1151 /*
1152  * Attach the interface. Allocate softc structures, do ifmedia
1153  * setup and ethernet/BPF attach.
1154  */
1155 static int
1156 re_attach(dev)
1157 	device_t		dev;
1158 {
1159 	u_char			eaddr[ETHER_ADDR_LEN];
1160 	u_int16_t		as[ETHER_ADDR_LEN / 2];
1161 	struct rl_softc		*sc;
1162 	struct ifnet		*ifp;
1163 	struct rl_hwrev		*hw_rev;
1164 	int			hwrev;
1165 	u_int16_t		re_did = 0;
1166 	int			error = 0, rid, i;
1167 	int			msic, reg;
1168 
1169 	sc = device_get_softc(dev);
1170 	sc->rl_dev = dev;
1171 
1172 	mtx_init(&sc->rl_mtx, device_get_nameunit(dev), MTX_NETWORK_LOCK,
1173 	    MTX_DEF);
1174 	callout_init_mtx(&sc->rl_stat_callout, &sc->rl_mtx, 0);
1175 
1176 	/*
1177 	 * Map control/status registers.
1178 	 */
1179 	pci_enable_busmaster(dev);
1180 
1181 	rid = RL_RID;
1182 	sc->rl_res = bus_alloc_resource_any(dev, RL_RES, &rid,
1183 	    RF_ACTIVE);
1184 
1185 	if (sc->rl_res == NULL) {
1186 		device_printf(dev, "couldn't map ports/memory\n");
1187 		error = ENXIO;
1188 		goto fail;
1189 	}
1190 
1191 	sc->rl_btag = rman_get_bustag(sc->rl_res);
1192 	sc->rl_bhandle = rman_get_bushandle(sc->rl_res);
1193 
1194 	msic = 0;
1195 	if (pci_find_extcap(dev, PCIY_EXPRESS, &reg) == 0) {
1196 		msic = pci_msi_count(dev);
1197 		if (bootverbose)
1198 			device_printf(dev, "MSI count : %d\n", msic);
1199 	}
1200 	if (msic == RL_MSI_MESSAGES  && msi_disable == 0) {
1201 		if (pci_alloc_msi(dev, &msic) == 0) {
1202 			if (msic == RL_MSI_MESSAGES) {
1203 				device_printf(dev, "Using %d MSI messages\n",
1204 				    msic);
1205 				sc->rl_msi = 1;
1206 			} else
1207 				pci_release_msi(dev);
1208 		}
1209 	}
1210 
1211 	/* Allocate interrupt */
1212 	if (sc->rl_msi == 0) {
1213 		rid = 0;
1214 		sc->rl_irq[0] = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
1215 		    RF_SHAREABLE | RF_ACTIVE);
1216 		if (sc->rl_irq[0] == NULL) {
1217 			device_printf(dev, "couldn't allocate IRQ resources\n");
1218 			error = ENXIO;
1219 			goto fail;
1220 		}
1221 	} else {
1222 		for (i = 0, rid = 1; i < RL_MSI_MESSAGES; i++, rid++) {
1223 			sc->rl_irq[i] = bus_alloc_resource_any(dev,
1224 			    SYS_RES_IRQ, &rid, RF_ACTIVE);
1225 			if (sc->rl_irq[i] == NULL) {
1226 				device_printf(dev,
1227 				    "couldn't llocate IRQ resources for "
1228 				    "message %d\n", rid);
1229 				error = ENXIO;
1230 				goto fail;
1231 			}
1232 		}
1233 	}
1234 
1235 	/* Reset the adapter. */
1236 	RL_LOCK(sc);
1237 	re_reset(sc);
1238 	RL_UNLOCK(sc);
1239 
1240 	hw_rev = re_hwrevs;
1241 	hwrev = CSR_READ_4(sc, RL_TXCFG) & RL_TXCFG_HWREV;
1242 	while (hw_rev->rl_desc != NULL) {
1243 		if (hw_rev->rl_rev == hwrev) {
1244 			sc->rl_type = hw_rev->rl_type;
1245 			break;
1246 		}
1247 		hw_rev++;
1248 	}
1249 
1250 	sc->rl_eewidth = RL_9356_ADDR_LEN;
1251 	re_read_eeprom(sc, (caddr_t)&re_did, 0, 1);
1252 	if (re_did != 0x8129)
1253 	        sc->rl_eewidth = RL_9346_ADDR_LEN;
1254 
1255 	/*
1256 	 * Get station address from the EEPROM.
1257 	 */
1258 	re_read_eeprom(sc, (caddr_t)as, RL_EE_EADDR, 3);
1259 	for (i = 0; i < ETHER_ADDR_LEN / 2; i++)
1260 		as[i] = le16toh(as[i]);
1261 	bcopy(as, eaddr, sizeof(eaddr));
1262 
1263 	if (sc->rl_type == RL_8169) {
1264 		/* Set RX length mask */
1265 		sc->rl_rxlenmask = RL_RDESC_STAT_GFRAGLEN;
1266 		sc->rl_txstart = RL_GTXSTART;
1267 	} else {
1268 		/* Set RX length mask */
1269 		sc->rl_rxlenmask = RL_RDESC_STAT_FRAGLEN;
1270 		sc->rl_txstart = RL_TXSTART;
1271 	}
1272 
1273 	/*
1274 	 * Allocate the parent bus DMA tag appropriate for PCI.
1275 	 */
1276 #define RL_NSEG_NEW 32
1277 	error = bus_dma_tag_create(bus_get_dma_tag(dev), 1, 0,
1278 	    BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL, NULL,
1279 	    MAXBSIZE, RL_NSEG_NEW, BUS_SPACE_MAXSIZE_32BIT, 0,
1280 	    NULL, NULL, &sc->rl_parent_tag);
1281 	if (error)
1282 		goto fail;
1283 
1284 	error = re_allocmem(dev, sc);
1285 
1286 	if (error)
1287 		goto fail;
1288 
1289 	ifp = sc->rl_ifp = if_alloc(IFT_ETHER);
1290 	if (ifp == NULL) {
1291 		device_printf(dev, "can not if_alloc()\n");
1292 		error = ENOSPC;
1293 		goto fail;
1294 	}
1295 
1296 	/* Do MII setup */
1297 	if (mii_phy_probe(dev, &sc->rl_miibus,
1298 	    re_ifmedia_upd, re_ifmedia_sts)) {
1299 		device_printf(dev, "MII without any phy!\n");
1300 		error = ENXIO;
1301 		goto fail;
1302 	}
1303 
1304 	/* Take PHY out of power down mode. */
1305 	if (sc->rl_type == RL_8169) {
1306 		uint32_t rev;
1307 
1308 		rev = CSR_READ_4(sc, RL_TXCFG);
1309 		/* HWVERID 0, 1 and 2 :  bit26-30, bit23 */
1310 		rev &= 0x7c800000;
1311 		if (rev != 0) {
1312 			/* RTL8169S single chip */
1313 			switch (rev) {
1314 			case RL_HWREV_8169_8110SB:
1315 			case RL_HWREV_8169_8110SC:
1316 			case RL_HWREV_8168_SPIN2:
1317 				re_gmii_writereg(dev, 1, 0x1f, 0);
1318 				re_gmii_writereg(dev, 1, 0x0e, 0);
1319 				break;
1320 			default:
1321 				break;
1322 			}
1323 		}
1324 	}
1325 
1326 	ifp->if_softc = sc;
1327 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1328 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1329 	ifp->if_ioctl = re_ioctl;
1330 	ifp->if_start = re_start;
1331 	ifp->if_hwassist = RE_CSUM_FEATURES;
1332 	ifp->if_capabilities = IFCAP_HWCSUM;
1333 	ifp->if_capenable = ifp->if_capabilities;
1334 	ifp->if_init = re_init;
1335 	IFQ_SET_MAXLEN(&ifp->if_snd, RL_IFQ_MAXLEN);
1336 	ifp->if_snd.ifq_drv_maxlen = RL_IFQ_MAXLEN;
1337 	IFQ_SET_READY(&ifp->if_snd);
1338 
1339 	TASK_INIT(&sc->rl_txtask, 1, re_tx_task, ifp);
1340 	TASK_INIT(&sc->rl_inttask, 0, re_int_task, sc);
1341 
1342 	/*
1343 	 * Call MI attach routine.
1344 	 */
1345 	ether_ifattach(ifp, eaddr);
1346 
1347 	/* VLAN capability setup */
1348 	ifp->if_capabilities |= IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING;
1349 	if (ifp->if_capabilities & IFCAP_HWCSUM)
1350 		ifp->if_capabilities |= IFCAP_VLAN_HWCSUM;
1351 	ifp->if_capenable = ifp->if_capabilities;
1352 #ifdef DEVICE_POLLING
1353 	ifp->if_capabilities |= IFCAP_POLLING;
1354 #endif
1355 	/*
1356 	 * Tell the upper layer(s) we support long frames.
1357 	 * Must appear after the call to ether_ifattach() because
1358 	 * ether_ifattach() sets ifi_hdrlen to the default value.
1359 	 */
1360 	ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
1361 
1362 #ifdef RE_DIAG
1363 	/*
1364 	 * Perform hardware diagnostic on the original RTL8169.
1365 	 * Some 32-bit cards were incorrectly wired and would
1366 	 * malfunction if plugged into a 64-bit slot.
1367 	 */
1368 
1369 	if (hwrev == RL_HWREV_8169) {
1370 		error = re_diag(sc);
1371 		if (error) {
1372 			device_printf(dev,
1373 		    	"attach aborted due to hardware diag failure\n");
1374 			ether_ifdetach(ifp);
1375 			goto fail;
1376 		}
1377 	}
1378 #endif
1379 
1380 	/* Hook interrupt last to avoid having to lock softc */
1381 	if (sc->rl_msi == 0)
1382 		error = bus_setup_intr(dev, sc->rl_irq[0],
1383 		    INTR_TYPE_NET | INTR_MPSAFE, re_intr, NULL, sc,
1384 		    &sc->rl_intrhand[0]);
1385 	else {
1386 		for (i = 0; i < RL_MSI_MESSAGES; i++) {
1387 			error = bus_setup_intr(dev, sc->rl_irq[i],
1388 			    INTR_TYPE_NET | INTR_MPSAFE, re_intr, NULL, sc,
1389 		    	    &sc->rl_intrhand[i]);
1390 			if (error != 0)
1391 				break;
1392 		}
1393 	}
1394 	if (error) {
1395 		device_printf(dev, "couldn't set up irq\n");
1396 		ether_ifdetach(ifp);
1397 	}
1398 
1399 fail:
1400 
1401 	if (error)
1402 		re_detach(dev);
1403 
1404 	return (error);
1405 }
1406 
1407 /*
1408  * Shutdown hardware and free up resources. This can be called any
1409  * time after the mutex has been initialized. It is called in both
1410  * the error case in attach and the normal detach case so it needs
1411  * to be careful about only freeing resources that have actually been
1412  * allocated.
1413  */
1414 static int
1415 re_detach(dev)
1416 	device_t		dev;
1417 {
1418 	struct rl_softc		*sc;
1419 	struct ifnet		*ifp;
1420 	int			i, rid;
1421 
1422 	sc = device_get_softc(dev);
1423 	ifp = sc->rl_ifp;
1424 	KASSERT(mtx_initialized(&sc->rl_mtx), ("re mutex not initialized"));
1425 
1426 #ifdef DEVICE_POLLING
1427 	if (ifp->if_capenable & IFCAP_POLLING)
1428 		ether_poll_deregister(ifp);
1429 #endif
1430 	/* These should only be active if attach succeeded */
1431 	if (device_is_attached(dev)) {
1432 		RL_LOCK(sc);
1433 #if 0
1434 		sc->suspended = 1;
1435 #endif
1436 		re_stop(sc);
1437 		RL_UNLOCK(sc);
1438 		callout_drain(&sc->rl_stat_callout);
1439 		taskqueue_drain(taskqueue_fast, &sc->rl_inttask);
1440 		taskqueue_drain(taskqueue_fast, &sc->rl_txtask);
1441 		/*
1442 		 * Force off the IFF_UP flag here, in case someone
1443 		 * still had a BPF descriptor attached to this
1444 		 * interface. If they do, ether_ifdetach() will cause
1445 		 * the BPF code to try and clear the promisc mode
1446 		 * flag, which will bubble down to re_ioctl(),
1447 		 * which will try to call re_init() again. This will
1448 		 * turn the NIC back on and restart the MII ticker,
1449 		 * which will panic the system when the kernel tries
1450 		 * to invoke the re_tick() function that isn't there
1451 		 * anymore.
1452 		 */
1453 		ifp->if_flags &= ~IFF_UP;
1454 		ether_ifdetach(ifp);
1455 	}
1456 	if (sc->rl_miibus)
1457 		device_delete_child(dev, sc->rl_miibus);
1458 	bus_generic_detach(dev);
1459 
1460 	/*
1461 	 * The rest is resource deallocation, so we should already be
1462 	 * stopped here.
1463 	 */
1464 
1465 	for (i = 0; i < RL_MSI_MESSAGES; i++) {
1466 		if (sc->rl_intrhand[i] != NULL) {
1467 			bus_teardown_intr(dev, sc->rl_irq[i],
1468 			    sc->rl_intrhand[i]);
1469 			sc->rl_intrhand[i] = NULL;
1470 		}
1471 	}
1472 	if (ifp != NULL)
1473 		if_free(ifp);
1474 	if (sc->rl_msi == 0) {
1475 		if (sc->rl_irq[0] != NULL) {
1476 			bus_release_resource(dev, SYS_RES_IRQ, 0,
1477 			    sc->rl_irq[0]);
1478 			sc->rl_irq[0] = NULL;
1479 		}
1480 	} else {
1481 		for (i = 0, rid = 1; i < RL_MSI_MESSAGES; i++, rid++) {
1482 			if (sc->rl_irq[i] != NULL) {
1483 				bus_release_resource(dev, SYS_RES_IRQ, rid,
1484 				    sc->rl_irq[i]);
1485 				sc->rl_irq[i] = NULL;
1486 			}
1487 		}
1488 		pci_release_msi(dev);
1489 	}
1490 	if (sc->rl_res)
1491 		bus_release_resource(dev, RL_RES, RL_RID, sc->rl_res);
1492 
1493 	/* Unload and free the RX DMA ring memory and map */
1494 
1495 	if (sc->rl_ldata.rl_rx_list_tag) {
1496 		bus_dmamap_unload(sc->rl_ldata.rl_rx_list_tag,
1497 		    sc->rl_ldata.rl_rx_list_map);
1498 		bus_dmamem_free(sc->rl_ldata.rl_rx_list_tag,
1499 		    sc->rl_ldata.rl_rx_list,
1500 		    sc->rl_ldata.rl_rx_list_map);
1501 		bus_dma_tag_destroy(sc->rl_ldata.rl_rx_list_tag);
1502 	}
1503 
1504 	/* Unload and free the TX DMA ring memory and map */
1505 
1506 	if (sc->rl_ldata.rl_tx_list_tag) {
1507 		bus_dmamap_unload(sc->rl_ldata.rl_tx_list_tag,
1508 		    sc->rl_ldata.rl_tx_list_map);
1509 		bus_dmamem_free(sc->rl_ldata.rl_tx_list_tag,
1510 		    sc->rl_ldata.rl_tx_list,
1511 		    sc->rl_ldata.rl_tx_list_map);
1512 		bus_dma_tag_destroy(sc->rl_ldata.rl_tx_list_tag);
1513 	}
1514 
1515 	/* Destroy all the RX and TX buffer maps */
1516 
1517 	if (sc->rl_ldata.rl_mtag) {
1518 		for (i = 0; i < RL_TX_DESC_CNT; i++)
1519 			bus_dmamap_destroy(sc->rl_ldata.rl_mtag,
1520 			    sc->rl_ldata.rl_tx_dmamap[i]);
1521 		for (i = 0; i < RL_RX_DESC_CNT; i++)
1522 			bus_dmamap_destroy(sc->rl_ldata.rl_mtag,
1523 			    sc->rl_ldata.rl_rx_dmamap[i]);
1524 		bus_dma_tag_destroy(sc->rl_ldata.rl_mtag);
1525 	}
1526 
1527 	/* Unload and free the stats buffer and map */
1528 
1529 	if (sc->rl_ldata.rl_stag) {
1530 		bus_dmamap_unload(sc->rl_ldata.rl_stag,
1531 		    sc->rl_ldata.rl_rx_list_map);
1532 		bus_dmamem_free(sc->rl_ldata.rl_stag,
1533 		    sc->rl_ldata.rl_stats,
1534 		    sc->rl_ldata.rl_smap);
1535 		bus_dma_tag_destroy(sc->rl_ldata.rl_stag);
1536 	}
1537 
1538 	if (sc->rl_parent_tag)
1539 		bus_dma_tag_destroy(sc->rl_parent_tag);
1540 
1541 	mtx_destroy(&sc->rl_mtx);
1542 
1543 	return (0);
1544 }
1545 
1546 static int
1547 re_newbuf(sc, idx, m)
1548 	struct rl_softc		*sc;
1549 	int			idx;
1550 	struct mbuf		*m;
1551 {
1552 	struct rl_dmaload_arg	arg;
1553 	struct mbuf		*n = NULL;
1554 	int			error;
1555 
1556 	if (m == NULL) {
1557 		n = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
1558 		if (n == NULL)
1559 			return (ENOBUFS);
1560 		m = n;
1561 	} else
1562 		m->m_data = m->m_ext.ext_buf;
1563 
1564 	m->m_len = m->m_pkthdr.len = MCLBYTES;
1565 #ifdef RE_FIXUP_RX
1566 	/*
1567 	 * This is part of an evil trick to deal with non-x86 platforms.
1568 	 * The RealTek chip requires RX buffers to be aligned on 64-bit
1569 	 * boundaries, but that will hose non-x86 machines. To get around
1570 	 * this, we leave some empty space at the start of each buffer
1571 	 * and for non-x86 hosts, we copy the buffer back six bytes
1572 	 * to achieve word alignment. This is slightly more efficient
1573 	 * than allocating a new buffer, copying the contents, and
1574 	 * discarding the old buffer.
1575 	 */
1576 	m_adj(m, RE_ETHER_ALIGN);
1577 #endif
1578 	arg.rl_idx = idx;
1579 	arg.rl_maxsegs = 1;
1580 	arg.rl_flags = 0;
1581 	arg.rl_ring = sc->rl_ldata.rl_rx_list;
1582 
1583 	error = bus_dmamap_load_mbuf(sc->rl_ldata.rl_mtag,
1584 	    sc->rl_ldata.rl_rx_dmamap[idx], m, re_dma_map_desc,
1585 	    &arg, BUS_DMA_NOWAIT);
1586 	if (error || arg.rl_maxsegs != 1) {
1587 		if (n != NULL)
1588 			m_freem(n);
1589 		if (arg.rl_maxsegs == 0)
1590 			bus_dmamap_unload(sc->rl_ldata.rl_mtag,
1591 			    sc->rl_ldata.rl_rx_dmamap[idx]);
1592 		return (ENOMEM);
1593 	}
1594 
1595 	sc->rl_ldata.rl_rx_list[idx].rl_cmdstat |= htole32(RL_RDESC_CMD_OWN);
1596 	sc->rl_ldata.rl_rx_mbuf[idx] = m;
1597 
1598 	bus_dmamap_sync(sc->rl_ldata.rl_mtag,
1599 	    sc->rl_ldata.rl_rx_dmamap[idx],
1600 	    BUS_DMASYNC_PREREAD);
1601 
1602 	return (0);
1603 }
1604 
1605 #ifdef RE_FIXUP_RX
1606 static __inline void
1607 re_fixup_rx(m)
1608 	struct mbuf		*m;
1609 {
1610 	int                     i;
1611 	uint16_t                *src, *dst;
1612 
1613 	src = mtod(m, uint16_t *);
1614 	dst = src - (RE_ETHER_ALIGN - ETHER_ALIGN) / sizeof *src;
1615 
1616 	for (i = 0; i < (m->m_len / sizeof(uint16_t) + 1); i++)
1617 		*dst++ = *src++;
1618 
1619 	m->m_data -= RE_ETHER_ALIGN - ETHER_ALIGN;
1620 
1621 	return;
1622 }
1623 #endif
1624 
1625 static int
1626 re_tx_list_init(sc)
1627 	struct rl_softc		*sc;
1628 {
1629 
1630 	RL_LOCK_ASSERT(sc);
1631 
1632 	bzero ((char *)sc->rl_ldata.rl_tx_list, RL_TX_LIST_SZ);
1633 	bzero ((char *)&sc->rl_ldata.rl_tx_mbuf,
1634 	    (RL_TX_DESC_CNT * sizeof(struct mbuf *)));
1635 
1636 	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
1637 	    sc->rl_ldata.rl_tx_list_map, BUS_DMASYNC_PREWRITE);
1638 	sc->rl_ldata.rl_tx_prodidx = 0;
1639 	sc->rl_ldata.rl_tx_considx = 0;
1640 	sc->rl_ldata.rl_tx_free = RL_TX_DESC_CNT;
1641 
1642 	return (0);
1643 }
1644 
1645 static int
1646 re_rx_list_init(sc)
1647 	struct rl_softc		*sc;
1648 {
1649 	int			i;
1650 
1651 	bzero ((char *)sc->rl_ldata.rl_rx_list, RL_RX_LIST_SZ);
1652 	bzero ((char *)&sc->rl_ldata.rl_rx_mbuf,
1653 	    (RL_RX_DESC_CNT * sizeof(struct mbuf *)));
1654 
1655 	for (i = 0; i < RL_RX_DESC_CNT; i++) {
1656 		if (re_newbuf(sc, i, NULL) == ENOBUFS)
1657 			return (ENOBUFS);
1658 	}
1659 
1660 	/* Flush the RX descriptors */
1661 
1662 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1663 	    sc->rl_ldata.rl_rx_list_map,
1664 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
1665 
1666 	sc->rl_ldata.rl_rx_prodidx = 0;
1667 	sc->rl_head = sc->rl_tail = NULL;
1668 
1669 	return (0);
1670 }
1671 
1672 /*
1673  * RX handler for C+ and 8169. For the gigE chips, we support
1674  * the reception of jumbo frames that have been fragmented
1675  * across multiple 2K mbuf cluster buffers.
1676  */
1677 static int
1678 re_rxeof(sc)
1679 	struct rl_softc		*sc;
1680 {
1681 	struct mbuf		*m;
1682 	struct ifnet		*ifp;
1683 	int			i, total_len;
1684 	struct rl_desc		*cur_rx;
1685 	u_int32_t		rxstat, rxvlan;
1686 	int			maxpkt = 16;
1687 
1688 	RL_LOCK_ASSERT(sc);
1689 
1690 	ifp = sc->rl_ifp;
1691 	i = sc->rl_ldata.rl_rx_prodidx;
1692 
1693 	/* Invalidate the descriptor memory */
1694 
1695 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1696 	    sc->rl_ldata.rl_rx_list_map,
1697 	    BUS_DMASYNC_POSTREAD);
1698 
1699 	while (!RL_OWN(&sc->rl_ldata.rl_rx_list[i]) && maxpkt) {
1700 		cur_rx = &sc->rl_ldata.rl_rx_list[i];
1701 		m = sc->rl_ldata.rl_rx_mbuf[i];
1702 		total_len = RL_RXBYTES(cur_rx);
1703 		rxstat = le32toh(cur_rx->rl_cmdstat);
1704 		rxvlan = le32toh(cur_rx->rl_vlanctl);
1705 
1706 		/* Invalidate the RX mbuf and unload its map */
1707 
1708 		bus_dmamap_sync(sc->rl_ldata.rl_mtag,
1709 		    sc->rl_ldata.rl_rx_dmamap[i],
1710 		    BUS_DMASYNC_POSTWRITE);
1711 		bus_dmamap_unload(sc->rl_ldata.rl_mtag,
1712 		    sc->rl_ldata.rl_rx_dmamap[i]);
1713 
1714 		if (!(rxstat & RL_RDESC_STAT_EOF)) {
1715 			m->m_len = RE_RX_DESC_BUFLEN;
1716 			if (sc->rl_head == NULL)
1717 				sc->rl_head = sc->rl_tail = m;
1718 			else {
1719 				m->m_flags &= ~M_PKTHDR;
1720 				sc->rl_tail->m_next = m;
1721 				sc->rl_tail = m;
1722 			}
1723 			re_newbuf(sc, i, NULL);
1724 			RL_DESC_INC(i);
1725 			continue;
1726 		}
1727 
1728 		/*
1729 		 * NOTE: for the 8139C+, the frame length field
1730 		 * is always 12 bits in size, but for the gigE chips,
1731 		 * it is 13 bits (since the max RX frame length is 16K).
1732 		 * Unfortunately, all 32 bits in the status word
1733 		 * were already used, so to make room for the extra
1734 		 * length bit, RealTek took out the 'frame alignment
1735 		 * error' bit and shifted the other status bits
1736 		 * over one slot. The OWN, EOR, FS and LS bits are
1737 		 * still in the same places. We have already extracted
1738 		 * the frame length and checked the OWN bit, so rather
1739 		 * than using an alternate bit mapping, we shift the
1740 		 * status bits one space to the right so we can evaluate
1741 		 * them using the 8169 status as though it was in the
1742 		 * same format as that of the 8139C+.
1743 		 */
1744 		if (sc->rl_type == RL_8169)
1745 			rxstat >>= 1;
1746 
1747 		/*
1748 		 * if total_len > 2^13-1, both _RXERRSUM and _GIANT will be
1749 		 * set, but if CRC is clear, it will still be a valid frame.
1750 		 */
1751 		if (rxstat & RL_RDESC_STAT_RXERRSUM && !(total_len > 8191 &&
1752 		    (rxstat & RL_RDESC_STAT_ERRS) == RL_RDESC_STAT_GIANT)) {
1753 			ifp->if_ierrors++;
1754 			/*
1755 			 * If this is part of a multi-fragment packet,
1756 			 * discard all the pieces.
1757 			 */
1758 			if (sc->rl_head != NULL) {
1759 				m_freem(sc->rl_head);
1760 				sc->rl_head = sc->rl_tail = NULL;
1761 			}
1762 			re_newbuf(sc, i, m);
1763 			RL_DESC_INC(i);
1764 			continue;
1765 		}
1766 
1767 		/*
1768 		 * If allocating a replacement mbuf fails,
1769 		 * reload the current one.
1770 		 */
1771 
1772 		if (re_newbuf(sc, i, NULL)) {
1773 			ifp->if_ierrors++;
1774 			if (sc->rl_head != NULL) {
1775 				m_freem(sc->rl_head);
1776 				sc->rl_head = sc->rl_tail = NULL;
1777 			}
1778 			re_newbuf(sc, i, m);
1779 			RL_DESC_INC(i);
1780 			continue;
1781 		}
1782 
1783 		RL_DESC_INC(i);
1784 
1785 		if (sc->rl_head != NULL) {
1786 			m->m_len = total_len % RE_RX_DESC_BUFLEN;
1787 			if (m->m_len == 0)
1788 				m->m_len = RE_RX_DESC_BUFLEN;
1789 			/*
1790 			 * Special case: if there's 4 bytes or less
1791 			 * in this buffer, the mbuf can be discarded:
1792 			 * the last 4 bytes is the CRC, which we don't
1793 			 * care about anyway.
1794 			 */
1795 			if (m->m_len <= ETHER_CRC_LEN) {
1796 				sc->rl_tail->m_len -=
1797 				    (ETHER_CRC_LEN - m->m_len);
1798 				m_freem(m);
1799 			} else {
1800 				m->m_len -= ETHER_CRC_LEN;
1801 				m->m_flags &= ~M_PKTHDR;
1802 				sc->rl_tail->m_next = m;
1803 			}
1804 			m = sc->rl_head;
1805 			sc->rl_head = sc->rl_tail = NULL;
1806 			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
1807 		} else
1808 			m->m_pkthdr.len = m->m_len =
1809 			    (total_len - ETHER_CRC_LEN);
1810 
1811 #ifdef RE_FIXUP_RX
1812 		re_fixup_rx(m);
1813 #endif
1814 		ifp->if_ipackets++;
1815 		m->m_pkthdr.rcvif = ifp;
1816 
1817 		/* Do RX checksumming if enabled */
1818 
1819 		if (ifp->if_capenable & IFCAP_RXCSUM) {
1820 
1821 			/* Check IP header checksum */
1822 			if (rxstat & RL_RDESC_STAT_PROTOID)
1823 				m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED;
1824 			if (!(rxstat & RL_RDESC_STAT_IPSUMBAD))
1825 				m->m_pkthdr.csum_flags |= CSUM_IP_VALID;
1826 
1827 			/* Check TCP/UDP checksum */
1828 			if ((RL_TCPPKT(rxstat) &&
1829 			    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
1830 			    (RL_UDPPKT(rxstat) &&
1831 			    !(rxstat & RL_RDESC_STAT_UDPSUMBAD))) {
1832 				m->m_pkthdr.csum_flags |=
1833 				    CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
1834 				m->m_pkthdr.csum_data = 0xffff;
1835 			}
1836 		}
1837 		maxpkt--;
1838 		if (rxvlan & RL_RDESC_VLANCTL_TAG) {
1839 			m->m_pkthdr.ether_vtag =
1840 			    ntohs((rxvlan & RL_RDESC_VLANCTL_DATA));
1841 			m->m_flags |= M_VLANTAG;
1842 		}
1843 		RL_UNLOCK(sc);
1844 		(*ifp->if_input)(ifp, m);
1845 		RL_LOCK(sc);
1846 	}
1847 
1848 	/* Flush the RX DMA ring */
1849 
1850 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1851 	    sc->rl_ldata.rl_rx_list_map,
1852 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
1853 
1854 	sc->rl_ldata.rl_rx_prodidx = i;
1855 
1856 	if (maxpkt)
1857 		return(EAGAIN);
1858 
1859 	return(0);
1860 }
1861 
1862 static void
1863 re_txeof(sc)
1864 	struct rl_softc		*sc;
1865 {
1866 	struct ifnet		*ifp;
1867 	u_int32_t		txstat;
1868 	int			idx;
1869 
1870 	ifp = sc->rl_ifp;
1871 	idx = sc->rl_ldata.rl_tx_considx;
1872 
1873 	/* Invalidate the TX descriptor list */
1874 	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
1875 	    sc->rl_ldata.rl_tx_list_map,
1876 	    BUS_DMASYNC_POSTREAD);
1877 
1878 	while (sc->rl_ldata.rl_tx_free < RL_TX_DESC_CNT) {
1879 		txstat = le32toh(sc->rl_ldata.rl_tx_list[idx].rl_cmdstat);
1880 		if (txstat & RL_TDESC_CMD_OWN)
1881 			break;
1882 
1883 		sc->rl_ldata.rl_tx_list[idx].rl_bufaddr_lo = 0;
1884 
1885 		/*
1886 		 * We only stash mbufs in the last descriptor
1887 		 * in a fragment chain, which also happens to
1888 		 * be the only place where the TX status bits
1889 		 * are valid.
1890 		 */
1891 		if (txstat & RL_TDESC_CMD_EOF) {
1892 			m_freem(sc->rl_ldata.rl_tx_mbuf[idx]);
1893 			sc->rl_ldata.rl_tx_mbuf[idx] = NULL;
1894 			bus_dmamap_unload(sc->rl_ldata.rl_mtag,
1895 			    sc->rl_ldata.rl_tx_dmamap[idx]);
1896 			if (txstat & (RL_TDESC_STAT_EXCESSCOL|
1897 			    RL_TDESC_STAT_COLCNT))
1898 				ifp->if_collisions++;
1899 			if (txstat & RL_TDESC_STAT_TXERRSUM)
1900 				ifp->if_oerrors++;
1901 			else
1902 				ifp->if_opackets++;
1903 		}
1904 		sc->rl_ldata.rl_tx_free++;
1905 		RL_DESC_INC(idx);
1906 	}
1907 	sc->rl_ldata.rl_tx_considx = idx;
1908 
1909 	/* No changes made to the TX ring, so no flush needed */
1910 
1911 	if (sc->rl_ldata.rl_tx_free > RL_TX_DESC_THLD)
1912 		ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1913 
1914 	if (sc->rl_ldata.rl_tx_free < RL_TX_DESC_CNT) {
1915 		/*
1916 		 * Some chips will ignore a second TX request issued
1917 		 * while an existing transmission is in progress. If
1918 		 * the transmitter goes idle but there are still
1919 		 * packets waiting to be sent, we need to restart the
1920 		 * channel here to flush them out. This only seems to
1921 		 * be required with the PCIe devices.
1922 		 */
1923 		CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
1924 
1925 #ifdef RE_TX_MODERATION
1926 		/*
1927 		 * If not all descriptors have been reaped yet, reload
1928 		 * the timer so that we will eventually get another
1929 		 * interrupt that will cause us to re-enter this routine.
1930 		 * This is done in case the transmitter has gone idle.
1931 		 */
1932 		CSR_WRITE_4(sc, RL_TIMERCNT, 1);
1933 #endif
1934 	} else
1935 		sc->rl_watchdog_timer = 0;
1936 }
1937 
1938 static void
1939 re_tick(xsc)
1940 	void			*xsc;
1941 {
1942 	struct rl_softc		*sc;
1943 	struct mii_data		*mii;
1944 	struct ifnet		*ifp;
1945 
1946 	sc = xsc;
1947 	ifp = sc->rl_ifp;
1948 
1949 	RL_LOCK_ASSERT(sc);
1950 
1951 	re_watchdog(sc);
1952 
1953 	mii = device_get_softc(sc->rl_miibus);
1954 	mii_tick(mii);
1955 	if (sc->rl_link) {
1956 		if (!(mii->mii_media_status & IFM_ACTIVE))
1957 			sc->rl_link = 0;
1958 	} else {
1959 		if (mii->mii_media_status & IFM_ACTIVE &&
1960 		    IFM_SUBTYPE(mii->mii_media_active) != IFM_NONE) {
1961 			sc->rl_link = 1;
1962 			if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
1963 				taskqueue_enqueue_fast(taskqueue_fast,
1964 				    &sc->rl_txtask);
1965 		}
1966 	}
1967 
1968 	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
1969 }
1970 
1971 #ifdef DEVICE_POLLING
1972 static void
1973 re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
1974 {
1975 	struct rl_softc *sc = ifp->if_softc;
1976 
1977 	RL_LOCK(sc);
1978 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1979 		re_poll_locked(ifp, cmd, count);
1980 	RL_UNLOCK(sc);
1981 }
1982 
1983 static void
1984 re_poll_locked(struct ifnet *ifp, enum poll_cmd cmd, int count)
1985 {
1986 	struct rl_softc *sc = ifp->if_softc;
1987 
1988 	RL_LOCK_ASSERT(sc);
1989 
1990 	sc->rxcycles = count;
1991 	re_rxeof(sc);
1992 	re_txeof(sc);
1993 
1994 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
1995 		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_txtask);
1996 
1997 	if (cmd == POLL_AND_CHECK_STATUS) { /* also check status register */
1998 		u_int16_t       status;
1999 
2000 		status = CSR_READ_2(sc, RL_ISR);
2001 		if (status == 0xffff)
2002 			return;
2003 		if (status)
2004 			CSR_WRITE_2(sc, RL_ISR, status);
2005 
2006 		/*
2007 		 * XXX check behaviour on receiver stalls.
2008 		 */
2009 
2010 		if (status & RL_ISR_SYSTEM_ERR) {
2011 			re_reset(sc);
2012 			re_init_locked(sc);
2013 		}
2014 	}
2015 }
2016 #endif /* DEVICE_POLLING */
2017 
2018 static int
2019 re_intr(arg)
2020 	void			*arg;
2021 {
2022 	struct rl_softc		*sc;
2023 	uint16_t		status;
2024 
2025 	sc = arg;
2026 
2027 	status = CSR_READ_2(sc, RL_ISR);
2028 	if (status == 0xFFFF || (status & RL_INTRS_CPLUS) == 0)
2029                 return (FILTER_STRAY);
2030 	CSR_WRITE_2(sc, RL_IMR, 0);
2031 
2032 	taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_inttask);
2033 
2034 	return (FILTER_HANDLED);
2035 }
2036 
2037 static void
2038 re_int_task(arg, npending)
2039 	void			*arg;
2040 	int			npending;
2041 {
2042 	struct rl_softc		*sc;
2043 	struct ifnet		*ifp;
2044 	u_int16_t		status;
2045 	int			rval = 0;
2046 
2047 	sc = arg;
2048 	ifp = sc->rl_ifp;
2049 
2050 	RL_LOCK(sc);
2051 
2052 	status = CSR_READ_2(sc, RL_ISR);
2053         CSR_WRITE_2(sc, RL_ISR, status);
2054 
2055 	if (sc->suspended || !(ifp->if_flags & IFF_UP)) {
2056 		RL_UNLOCK(sc);
2057 		return;
2058 	}
2059 
2060 #ifdef DEVICE_POLLING
2061 	if  (ifp->if_capenable & IFCAP_POLLING) {
2062 		RL_UNLOCK(sc);
2063 		return;
2064 	}
2065 #endif
2066 
2067 	if (status & (RL_ISR_RX_OK|RL_ISR_RX_ERR|RL_ISR_FIFO_OFLOW))
2068 		rval = re_rxeof(sc);
2069 
2070 #ifdef RE_TX_MODERATION
2071 	if (status & (RL_ISR_TIMEOUT_EXPIRED|
2072 #else
2073 	if (status & (RL_ISR_TX_OK|
2074 #endif
2075 	    RL_ISR_TX_ERR|RL_ISR_TX_DESC_UNAVAIL))
2076 		re_txeof(sc);
2077 
2078 	if (status & RL_ISR_SYSTEM_ERR) {
2079 		re_reset(sc);
2080 		re_init_locked(sc);
2081 	}
2082 
2083 	if (status & RL_ISR_LINKCHG) {
2084 		callout_stop(&sc->rl_stat_callout);
2085 		re_tick(sc);
2086 	}
2087 
2088 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2089 		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_txtask);
2090 
2091 	RL_UNLOCK(sc);
2092 
2093         if ((CSR_READ_2(sc, RL_ISR) & RL_INTRS_CPLUS) || rval) {
2094 		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_inttask);
2095 		return;
2096 	}
2097 
2098 	CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2099 
2100 	return;
2101 }
2102 
2103 static int
2104 re_encap(sc, m_head, idx)
2105 	struct rl_softc		*sc;
2106 	struct mbuf		**m_head;
2107 	int			*idx;
2108 {
2109 	struct mbuf		*m_new = NULL;
2110 	struct rl_dmaload_arg	arg;
2111 	bus_dmamap_t		map;
2112 	int			error;
2113 
2114 	RL_LOCK_ASSERT(sc);
2115 
2116 	if (sc->rl_ldata.rl_tx_free <= RL_TX_DESC_THLD)
2117 		return (EFBIG);
2118 
2119 	/*
2120 	 * Set up checksum offload. Note: checksum offload bits must
2121 	 * appear in all descriptors of a multi-descriptor transmit
2122 	 * attempt. This is according to testing done with an 8169
2123 	 * chip. This is a requirement.
2124 	 */
2125 
2126 	arg.rl_flags = 0;
2127 
2128 	if (((*m_head)->m_pkthdr.csum_flags & CSUM_TSO) != 0)
2129 		arg.rl_flags = RL_TDESC_CMD_LGSEND |
2130 		    ((uint32_t)(*m_head)->m_pkthdr.tso_segsz <<
2131 		    RL_TDESC_CMD_MSSVAL_SHIFT);
2132 	else {
2133 		if ((*m_head)->m_pkthdr.csum_flags & CSUM_IP)
2134 			arg.rl_flags |= RL_TDESC_CMD_IPCSUM;
2135 		if ((*m_head)->m_pkthdr.csum_flags & CSUM_TCP)
2136 			arg.rl_flags |= RL_TDESC_CMD_TCPCSUM;
2137 		if ((*m_head)->m_pkthdr.csum_flags & CSUM_UDP)
2138 			arg.rl_flags |= RL_TDESC_CMD_UDPCSUM;
2139 	}
2140 
2141 	arg.rl_idx = *idx;
2142 	arg.rl_maxsegs = sc->rl_ldata.rl_tx_free;
2143 	if (arg.rl_maxsegs > RL_TX_DESC_THLD)
2144 		arg.rl_maxsegs -= RL_TX_DESC_THLD;
2145 	arg.rl_ring = sc->rl_ldata.rl_tx_list;
2146 
2147 	map = sc->rl_ldata.rl_tx_dmamap[*idx];
2148 
2149 	/*
2150 	 * With some of the RealTek chips, using the checksum offload
2151 	 * support in conjunction with the autopadding feature results
2152 	 * in the transmission of corrupt frames. For example, if we
2153 	 * need to send a really small IP fragment that's less than 60
2154 	 * bytes in size, and IP header checksumming is enabled, the
2155 	 * resulting ethernet frame that appears on the wire will
2156 	 * have garbled payload. To work around this, if TX checksum
2157 	 * offload is enabled, we always manually pad short frames out
2158 	 * to the minimum ethernet frame size. We do this by pretending
2159 	 * the mbuf chain has too many fragments so the coalescing code
2160 	 * below can assemble the packet into a single buffer that's
2161 	 * padded out to the mininum frame size.
2162 	 *
2163 	 * Note: this appears unnecessary for TCP, and doing it for TCP
2164 	 * with PCIe adapters seems to result in bad checksums.
2165 	 */
2166 
2167 	if (arg.rl_flags && !(arg.rl_flags & RL_TDESC_CMD_TCPCSUM) &&
2168             (*m_head)->m_pkthdr.len < RL_MIN_FRAMELEN)
2169 		error = EFBIG;
2170 	else
2171 		error = bus_dmamap_load_mbuf(sc->rl_ldata.rl_mtag, map,
2172 		    *m_head, re_dma_map_desc, &arg, BUS_DMA_NOWAIT);
2173 
2174 	if (error && error != EFBIG) {
2175 		device_printf(sc->rl_dev, "can't map mbuf (error %d)\n", error);
2176 		return (ENOBUFS);
2177 	}
2178 
2179 	/* Too many segments to map, coalesce into a single mbuf */
2180 
2181 	if (error || arg.rl_maxsegs == 0) {
2182 		if (arg.rl_maxsegs == 0)
2183 			bus_dmamap_unload(sc->rl_ldata.rl_mtag, map);
2184 		m_new = m_defrag(*m_head, M_DONTWAIT);
2185 		if (m_new == NULL) {
2186 			m_freem(*m_head);
2187 			*m_head = NULL;
2188 			return (ENOBUFS);
2189 		}
2190 		*m_head = m_new;
2191 
2192 		/*
2193 		 * Manually pad short frames, and zero the pad space
2194 		 * to avoid leaking data.
2195 		 */
2196 		if (m_new->m_pkthdr.len < RL_MIN_FRAMELEN) {
2197 			bzero(mtod(m_new, char *) + m_new->m_pkthdr.len,
2198 			    RL_MIN_FRAMELEN - m_new->m_pkthdr.len);
2199 			m_new->m_pkthdr.len += RL_MIN_FRAMELEN -
2200 			    m_new->m_pkthdr.len;
2201 			m_new->m_len = m_new->m_pkthdr.len;
2202 		}
2203 
2204 		/* Note that we'll run over RL_TX_DESC_THLD here. */
2205 		arg.rl_maxsegs = sc->rl_ldata.rl_tx_free;
2206 		error = bus_dmamap_load_mbuf(sc->rl_ldata.rl_mtag, map,
2207 		    *m_head, re_dma_map_desc, &arg, BUS_DMA_NOWAIT);
2208 		if (error || arg.rl_maxsegs == 0) {
2209 			device_printf(sc->rl_dev,
2210 			    "can't map defragmented mbuf (error %d)\n", error);
2211 			m_freem(m_new);
2212 			*m_head = NULL;
2213 			if (arg.rl_maxsegs == 0)
2214 				bus_dmamap_unload(sc->rl_ldata.rl_mtag, map);
2215 			return (EFBIG);
2216 		}
2217 	}
2218 
2219 	/*
2220 	 * Insure that the map for this transmission
2221 	 * is placed at the array index of the last descriptor
2222 	 * in this chain.  (Swap last and first dmamaps.)
2223 	 */
2224 	sc->rl_ldata.rl_tx_dmamap[*idx] =
2225 	    sc->rl_ldata.rl_tx_dmamap[arg.rl_idx];
2226 	sc->rl_ldata.rl_tx_dmamap[arg.rl_idx] = map;
2227 
2228 	sc->rl_ldata.rl_tx_mbuf[arg.rl_idx] = *m_head;
2229 	sc->rl_ldata.rl_tx_free -= arg.rl_maxsegs;
2230 
2231 	/*
2232 	 * Set up hardware VLAN tagging. Note: vlan tag info must
2233 	 * appear in the first descriptor of a multi-descriptor
2234 	 * transmission attempt.
2235 	 */
2236 	if ((*m_head)->m_flags & M_VLANTAG)
2237 		sc->rl_ldata.rl_tx_list[*idx].rl_vlanctl =
2238 		    htole32(htons((*m_head)->m_pkthdr.ether_vtag) |
2239 		    RL_TDESC_VLANCTL_TAG);
2240 
2241 	/* Transfer ownership of packet to the chip. */
2242 
2243 	sc->rl_ldata.rl_tx_list[arg.rl_idx].rl_cmdstat |=
2244 	    htole32(RL_TDESC_CMD_OWN);
2245 	if (*idx != arg.rl_idx)
2246 		sc->rl_ldata.rl_tx_list[*idx].rl_cmdstat |=
2247 		    htole32(RL_TDESC_CMD_OWN);
2248 
2249         RL_DESC_INC(arg.rl_idx);
2250 	*idx = arg.rl_idx;
2251 
2252 	return (0);
2253 }
2254 
2255 static void
2256 re_tx_task(arg, npending)
2257 	void			*arg;
2258 	int			npending;
2259 {
2260 	struct ifnet		*ifp;
2261 
2262 	ifp = arg;
2263 	re_start(ifp);
2264 
2265 	return;
2266 }
2267 
2268 /*
2269  * Main transmit routine for C+ and gigE NICs.
2270  */
2271 static void
2272 re_start(ifp)
2273 	struct ifnet		*ifp;
2274 {
2275 	struct rl_softc		*sc;
2276 	struct mbuf		*m_head = NULL;
2277 	int			idx, queued = 0;
2278 
2279 	sc = ifp->if_softc;
2280 
2281 	RL_LOCK(sc);
2282 
2283 	if (!sc->rl_link || ifp->if_drv_flags & IFF_DRV_OACTIVE) {
2284 		RL_UNLOCK(sc);
2285 		return;
2286 	}
2287 
2288 	idx = sc->rl_ldata.rl_tx_prodidx;
2289 
2290 	while (sc->rl_ldata.rl_tx_mbuf[idx] == NULL) {
2291 		IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
2292 		if (m_head == NULL)
2293 			break;
2294 
2295 		if (re_encap(sc, &m_head, &idx)) {
2296 			if (m_head == NULL)
2297 				break;
2298 			IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
2299 			ifp->if_drv_flags |= IFF_DRV_OACTIVE;
2300 			break;
2301 		}
2302 
2303 		/*
2304 		 * If there's a BPF listener, bounce a copy of this frame
2305 		 * to him.
2306 		 */
2307 		ETHER_BPF_MTAP(ifp, m_head);
2308 
2309 		queued++;
2310 	}
2311 
2312 	if (queued == 0) {
2313 #ifdef RE_TX_MODERATION
2314 		if (sc->rl_ldata.rl_tx_free != RL_TX_DESC_CNT)
2315 			CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2316 #endif
2317 		RL_UNLOCK(sc);
2318 		return;
2319 	}
2320 
2321 	/* Flush the TX descriptors */
2322 
2323 	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2324 	    sc->rl_ldata.rl_tx_list_map,
2325 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2326 
2327 	sc->rl_ldata.rl_tx_prodidx = idx;
2328 
2329 	CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2330 
2331 #ifdef RE_TX_MODERATION
2332 	/*
2333 	 * Use the countdown timer for interrupt moderation.
2334 	 * 'TX done' interrupts are disabled. Instead, we reset the
2335 	 * countdown timer, which will begin counting until it hits
2336 	 * the value in the TIMERINT register, and then trigger an
2337 	 * interrupt. Each time we write to the TIMERCNT register,
2338 	 * the timer count is reset to 0.
2339 	 */
2340 	CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2341 #endif
2342 
2343 	/*
2344 	 * Set a timeout in case the chip goes out to lunch.
2345 	 */
2346 	sc->rl_watchdog_timer = 5;
2347 
2348 	RL_UNLOCK(sc);
2349 
2350 	return;
2351 }
2352 
2353 static void
2354 re_init(xsc)
2355 	void			*xsc;
2356 {
2357 	struct rl_softc		*sc = xsc;
2358 
2359 	RL_LOCK(sc);
2360 	re_init_locked(sc);
2361 	RL_UNLOCK(sc);
2362 }
2363 
2364 static void
2365 re_init_locked(sc)
2366 	struct rl_softc		*sc;
2367 {
2368 	struct ifnet		*ifp = sc->rl_ifp;
2369 	struct mii_data		*mii;
2370 	u_int32_t		rxcfg = 0;
2371 	union {
2372 		uint32_t align_dummy;
2373 		u_char eaddr[ETHER_ADDR_LEN];
2374         } eaddr;
2375 
2376 	RL_LOCK_ASSERT(sc);
2377 
2378 	mii = device_get_softc(sc->rl_miibus);
2379 
2380 	/*
2381 	 * Cancel pending I/O and free all RX/TX buffers.
2382 	 */
2383 	re_stop(sc);
2384 
2385 	/*
2386 	 * Enable C+ RX and TX mode, as well as VLAN stripping and
2387 	 * RX checksum offload. We must configure the C+ register
2388 	 * before all others.
2389 	 */
2390 	CSR_WRITE_2(sc, RL_CPLUS_CMD, RL_CPLUSCMD_RXENB|
2391 	    RL_CPLUSCMD_TXENB|RL_CPLUSCMD_PCI_MRW|
2392 	    RL_CPLUSCMD_VLANSTRIP|RL_CPLUSCMD_RXCSUM_ENB);
2393 
2394 	/*
2395 	 * Init our MAC address.  Even though the chipset
2396 	 * documentation doesn't mention it, we need to enter "Config
2397 	 * register write enable" mode to modify the ID registers.
2398 	 */
2399 	/* Copy MAC address on stack to align. */
2400 	bcopy(IF_LLADDR(ifp), eaddr.eaddr, ETHER_ADDR_LEN);
2401 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG);
2402 	CSR_WRITE_4(sc, RL_IDR0,
2403 	    htole32(*(u_int32_t *)(&eaddr.eaddr[0])));
2404 	CSR_WRITE_4(sc, RL_IDR4,
2405 	    htole32(*(u_int32_t *)(&eaddr.eaddr[4])));
2406 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
2407 
2408 	/*
2409 	 * For C+ mode, initialize the RX descriptors and mbufs.
2410 	 */
2411 	re_rx_list_init(sc);
2412 	re_tx_list_init(sc);
2413 
2414 	/*
2415 	 * Load the addresses of the RX and TX lists into the chip.
2416 	 */
2417 
2418 	CSR_WRITE_4(sc, RL_RXLIST_ADDR_HI,
2419 	    RL_ADDR_HI(sc->rl_ldata.rl_rx_list_addr));
2420 	CSR_WRITE_4(sc, RL_RXLIST_ADDR_LO,
2421 	    RL_ADDR_LO(sc->rl_ldata.rl_rx_list_addr));
2422 
2423 	CSR_WRITE_4(sc, RL_TXLIST_ADDR_HI,
2424 	    RL_ADDR_HI(sc->rl_ldata.rl_tx_list_addr));
2425 	CSR_WRITE_4(sc, RL_TXLIST_ADDR_LO,
2426 	    RL_ADDR_LO(sc->rl_ldata.rl_tx_list_addr));
2427 
2428 	/*
2429 	 * Enable transmit and receive.
2430 	 */
2431 	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
2432 
2433 	/*
2434 	 * Set the initial TX and RX configuration.
2435 	 */
2436 	if (sc->rl_testmode) {
2437 		if (sc->rl_type == RL_8169)
2438 			CSR_WRITE_4(sc, RL_TXCFG,
2439 			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON);
2440 		else
2441 			CSR_WRITE_4(sc, RL_TXCFG,
2442 			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON_CPLUS);
2443 	} else
2444 		CSR_WRITE_4(sc, RL_TXCFG, RL_TXCFG_CONFIG);
2445 
2446 	CSR_WRITE_1(sc, RL_EARLY_TX_THRESH, 16);
2447 
2448 	CSR_WRITE_4(sc, RL_RXCFG, RL_RXCFG_CONFIG);
2449 
2450 	/* Set the individual bit to receive frames for this host only. */
2451 	rxcfg = CSR_READ_4(sc, RL_RXCFG);
2452 	rxcfg |= RL_RXCFG_RX_INDIV;
2453 
2454 	/* If we want promiscuous mode, set the allframes bit. */
2455 	if (ifp->if_flags & IFF_PROMISC)
2456 		rxcfg |= RL_RXCFG_RX_ALLPHYS;
2457 	else
2458 		rxcfg &= ~RL_RXCFG_RX_ALLPHYS;
2459 	CSR_WRITE_4(sc, RL_RXCFG, rxcfg);
2460 
2461 	/*
2462 	 * Set capture broadcast bit to capture broadcast frames.
2463 	 */
2464 	if (ifp->if_flags & IFF_BROADCAST)
2465 		rxcfg |= RL_RXCFG_RX_BROAD;
2466 	else
2467 		rxcfg &= ~RL_RXCFG_RX_BROAD;
2468 	CSR_WRITE_4(sc, RL_RXCFG, rxcfg);
2469 
2470 	/*
2471 	 * Program the multicast filter, if necessary.
2472 	 */
2473 	re_setmulti(sc);
2474 
2475 #ifdef DEVICE_POLLING
2476 	/*
2477 	 * Disable interrupts if we are polling.
2478 	 */
2479 	if (ifp->if_capenable & IFCAP_POLLING)
2480 		CSR_WRITE_2(sc, RL_IMR, 0);
2481 	else	/* otherwise ... */
2482 #endif
2483 
2484 	/*
2485 	 * Enable interrupts.
2486 	 */
2487 	if (sc->rl_testmode)
2488 		CSR_WRITE_2(sc, RL_IMR, 0);
2489 	else
2490 		CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2491 	CSR_WRITE_2(sc, RL_ISR, RL_INTRS_CPLUS);
2492 
2493 	/* Set initial TX threshold */
2494 	sc->rl_txthresh = RL_TX_THRESH_INIT;
2495 
2496 	/* Start RX/TX process. */
2497 	CSR_WRITE_4(sc, RL_MISSEDPKT, 0);
2498 #ifdef notdef
2499 	/* Enable receiver and transmitter. */
2500 	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
2501 #endif
2502 
2503 #ifdef RE_TX_MODERATION
2504 	/*
2505 	 * Initialize the timer interrupt register so that
2506 	 * a timer interrupt will be generated once the timer
2507 	 * reaches a certain number of ticks. The timer is
2508 	 * reloaded on each transmit. This gives us TX interrupt
2509 	 * moderation, which dramatically improves TX frame rate.
2510 	 */
2511 	if (sc->rl_type == RL_8169)
2512 		CSR_WRITE_4(sc, RL_TIMERINT_8169, 0x800);
2513 	else
2514 		CSR_WRITE_4(sc, RL_TIMERINT, 0x400);
2515 #endif
2516 
2517 	/*
2518 	 * For 8169 gigE NICs, set the max allowed RX packet
2519 	 * size so we can receive jumbo frames.
2520 	 */
2521 	if (sc->rl_type == RL_8169)
2522 		CSR_WRITE_2(sc, RL_MAXRXPKTLEN, 16383);
2523 
2524 	if (sc->rl_testmode)
2525 		return;
2526 
2527 	mii_mediachg(mii);
2528 
2529 	CSR_WRITE_1(sc, RL_CFG1, CSR_READ_1(sc, RL_CFG1) | RL_CFG1_DRVLOAD);
2530 
2531 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
2532 	ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
2533 
2534 	sc->rl_link = 0;
2535 	sc->rl_watchdog_timer = 0;
2536 	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
2537 }
2538 
2539 /*
2540  * Set media options.
2541  */
2542 static int
2543 re_ifmedia_upd(ifp)
2544 	struct ifnet		*ifp;
2545 {
2546 	struct rl_softc		*sc;
2547 	struct mii_data		*mii;
2548 
2549 	sc = ifp->if_softc;
2550 	mii = device_get_softc(sc->rl_miibus);
2551 	RL_LOCK(sc);
2552 	mii_mediachg(mii);
2553 	RL_UNLOCK(sc);
2554 
2555 	return (0);
2556 }
2557 
2558 /*
2559  * Report current media status.
2560  */
2561 static void
2562 re_ifmedia_sts(ifp, ifmr)
2563 	struct ifnet		*ifp;
2564 	struct ifmediareq	*ifmr;
2565 {
2566 	struct rl_softc		*sc;
2567 	struct mii_data		*mii;
2568 
2569 	sc = ifp->if_softc;
2570 	mii = device_get_softc(sc->rl_miibus);
2571 
2572 	RL_LOCK(sc);
2573 	mii_pollstat(mii);
2574 	RL_UNLOCK(sc);
2575 	ifmr->ifm_active = mii->mii_media_active;
2576 	ifmr->ifm_status = mii->mii_media_status;
2577 }
2578 
2579 static int
2580 re_ioctl(ifp, command, data)
2581 	struct ifnet		*ifp;
2582 	u_long			command;
2583 	caddr_t			data;
2584 {
2585 	struct rl_softc		*sc = ifp->if_softc;
2586 	struct ifreq		*ifr = (struct ifreq *) data;
2587 	struct mii_data		*mii;
2588 	int			error = 0;
2589 
2590 	switch (command) {
2591 	case SIOCSIFMTU:
2592 		RL_LOCK(sc);
2593 		if (ifr->ifr_mtu > RL_JUMBO_MTU)
2594 			error = EINVAL;
2595 		ifp->if_mtu = ifr->ifr_mtu;
2596 		RL_UNLOCK(sc);
2597 		break;
2598 	case SIOCSIFFLAGS:
2599 		RL_LOCK(sc);
2600 		if ((ifp->if_flags & IFF_UP) != 0) {
2601 			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
2602 				if (((ifp->if_flags ^ sc->rl_if_flags)
2603 				    & IFF_PROMISC) != 0)
2604 					re_setmulti(sc);
2605 			} else
2606 				re_init_locked(sc);
2607 		} else {
2608 			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
2609 				re_stop(sc);
2610 		}
2611 		sc->rl_if_flags = ifp->if_flags;
2612 		RL_UNLOCK(sc);
2613 		break;
2614 	case SIOCADDMULTI:
2615 	case SIOCDELMULTI:
2616 		RL_LOCK(sc);
2617 		re_setmulti(sc);
2618 		RL_UNLOCK(sc);
2619 		break;
2620 	case SIOCGIFMEDIA:
2621 	case SIOCSIFMEDIA:
2622 		mii = device_get_softc(sc->rl_miibus);
2623 		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, command);
2624 		break;
2625 	case SIOCSIFCAP:
2626 	    {
2627 		int mask, reinit;
2628 
2629 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
2630 		reinit = 0;
2631 #ifdef DEVICE_POLLING
2632 		if (mask & IFCAP_POLLING) {
2633 			if (ifr->ifr_reqcap & IFCAP_POLLING) {
2634 				error = ether_poll_register(re_poll, ifp);
2635 				if (error)
2636 					return(error);
2637 				RL_LOCK(sc);
2638 				/* Disable interrupts */
2639 				CSR_WRITE_2(sc, RL_IMR, 0x0000);
2640 				ifp->if_capenable |= IFCAP_POLLING;
2641 				RL_UNLOCK(sc);
2642 			} else {
2643 				error = ether_poll_deregister(ifp);
2644 				/* Enable interrupts. */
2645 				RL_LOCK(sc);
2646 				CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2647 				ifp->if_capenable &= ~IFCAP_POLLING;
2648 				RL_UNLOCK(sc);
2649 			}
2650 		}
2651 #endif /* DEVICE_POLLING */
2652 		if (mask & IFCAP_HWCSUM) {
2653 			ifp->if_capenable ^= IFCAP_HWCSUM;
2654 			if (ifp->if_capenable & IFCAP_TXCSUM)
2655 				ifp->if_hwassist |= RE_CSUM_FEATURES;
2656 			else
2657 				ifp->if_hwassist &= ~RE_CSUM_FEATURES;
2658 			reinit = 1;
2659 		}
2660 		if (mask & IFCAP_VLAN_HWTAGGING) {
2661 			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
2662 			reinit = 1;
2663 		}
2664 		if (mask & IFCAP_TSO4) {
2665 			ifp->if_capenable ^= IFCAP_TSO4;
2666 			if ((IFCAP_TSO4 & ifp->if_capenable) &&
2667 			    (IFCAP_TSO4 & ifp->if_capabilities))
2668 				ifp->if_hwassist |= CSUM_TSO;
2669 			else
2670 				ifp->if_hwassist &= ~CSUM_TSO;
2671 		}
2672 		if (reinit && ifp->if_drv_flags & IFF_DRV_RUNNING)
2673 			re_init(sc);
2674 		VLAN_CAPABILITIES(ifp);
2675 	    }
2676 		break;
2677 	default:
2678 		error = ether_ioctl(ifp, command, data);
2679 		break;
2680 	}
2681 
2682 	return (error);
2683 }
2684 
2685 static void
2686 re_watchdog(sc)
2687 	struct rl_softc		*sc;
2688 {
2689 
2690 	RL_LOCK_ASSERT(sc);
2691 
2692 	if (sc->rl_watchdog_timer == 0 || --sc->rl_watchdog_timer != 0)
2693 		return;
2694 
2695 	device_printf(sc->rl_dev, "watchdog timeout\n");
2696 	sc->rl_ifp->if_oerrors++;
2697 
2698 	re_txeof(sc);
2699 	re_rxeof(sc);
2700 	re_init_locked(sc);
2701 }
2702 
2703 /*
2704  * Stop the adapter and free any mbufs allocated to the
2705  * RX and TX lists.
2706  */
2707 static void
2708 re_stop(sc)
2709 	struct rl_softc		*sc;
2710 {
2711 	register int		i;
2712 	struct ifnet		*ifp;
2713 
2714 	RL_LOCK_ASSERT(sc);
2715 
2716 	ifp = sc->rl_ifp;
2717 
2718 	sc->rl_watchdog_timer = 0;
2719 	callout_stop(&sc->rl_stat_callout);
2720 	ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
2721 
2722 	CSR_WRITE_1(sc, RL_COMMAND, 0x00);
2723 	CSR_WRITE_2(sc, RL_IMR, 0x0000);
2724 	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
2725 
2726 	if (sc->rl_head != NULL) {
2727 		m_freem(sc->rl_head);
2728 		sc->rl_head = sc->rl_tail = NULL;
2729 	}
2730 
2731 	/* Free the TX list buffers. */
2732 
2733 	for (i = 0; i < RL_TX_DESC_CNT; i++) {
2734 		if (sc->rl_ldata.rl_tx_mbuf[i] != NULL) {
2735 			bus_dmamap_unload(sc->rl_ldata.rl_mtag,
2736 			    sc->rl_ldata.rl_tx_dmamap[i]);
2737 			m_freem(sc->rl_ldata.rl_tx_mbuf[i]);
2738 			sc->rl_ldata.rl_tx_mbuf[i] = NULL;
2739 		}
2740 	}
2741 
2742 	/* Free the RX list buffers. */
2743 
2744 	for (i = 0; i < RL_RX_DESC_CNT; i++) {
2745 		if (sc->rl_ldata.rl_rx_mbuf[i] != NULL) {
2746 			bus_dmamap_unload(sc->rl_ldata.rl_mtag,
2747 			    sc->rl_ldata.rl_rx_dmamap[i]);
2748 			m_freem(sc->rl_ldata.rl_rx_mbuf[i]);
2749 			sc->rl_ldata.rl_rx_mbuf[i] = NULL;
2750 		}
2751 	}
2752 }
2753 
2754 /*
2755  * Device suspend routine.  Stop the interface and save some PCI
2756  * settings in case the BIOS doesn't restore them properly on
2757  * resume.
2758  */
2759 static int
2760 re_suspend(dev)
2761 	device_t		dev;
2762 {
2763 	struct rl_softc		*sc;
2764 
2765 	sc = device_get_softc(dev);
2766 
2767 	RL_LOCK(sc);
2768 	re_stop(sc);
2769 	sc->suspended = 1;
2770 	RL_UNLOCK(sc);
2771 
2772 	return (0);
2773 }
2774 
2775 /*
2776  * Device resume routine.  Restore some PCI settings in case the BIOS
2777  * doesn't, re-enable busmastering, and restart the interface if
2778  * appropriate.
2779  */
2780 static int
2781 re_resume(dev)
2782 	device_t		dev;
2783 {
2784 	struct rl_softc		*sc;
2785 	struct ifnet		*ifp;
2786 
2787 	sc = device_get_softc(dev);
2788 
2789 	RL_LOCK(sc);
2790 
2791 	ifp = sc->rl_ifp;
2792 
2793 	/* reinitialize interface if necessary */
2794 	if (ifp->if_flags & IFF_UP)
2795 		re_init_locked(sc);
2796 
2797 	sc->suspended = 0;
2798 	RL_UNLOCK(sc);
2799 
2800 	return (0);
2801 }
2802 
2803 /*
2804  * Stop all chip I/O so that the kernel's probe routines don't
2805  * get confused by errant DMAs when rebooting.
2806  */
2807 static void
2808 re_shutdown(dev)
2809 	device_t		dev;
2810 {
2811 	struct rl_softc		*sc;
2812 
2813 	sc = device_get_softc(dev);
2814 
2815 	RL_LOCK(sc);
2816 	re_stop(sc);
2817 	/*
2818 	 * Mark interface as down since otherwise we will panic if
2819 	 * interrupt comes in later on, which can happen in some
2820 	 * cases.
2821 	 */
2822 	sc->rl_ifp->if_flags &= ~IFF_UP;
2823 	RL_UNLOCK(sc);
2824 }
2825