xref: /freebsd/sys/dev/re/if_re.c (revision f856af0466c076beef4ea9b15d088e1119a945b8)
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 #define RE_CSUM_FEATURES    (CSUM_IP | CSUM_TCP | CSUM_UDP)
164 
165 /*
166  * Various supported device vendors/types and their names.
167  */
168 static struct rl_type re_devs[] = {
169 	{ DLINK_VENDORID, DLINK_DEVICEID_528T, RL_HWREV_8169S,
170 		"D-Link DGE-528(T) Gigabit Ethernet Adapter" },
171 	{ RT_VENDORID, RT_DEVICEID_8139, RL_HWREV_8139CPLUS,
172 		"RealTek 8139C+ 10/100BaseTX" },
173 	{ RT_VENDORID, RT_DEVICEID_8101E, RL_HWREV_8101E,
174 		"RealTek 8101E PCIe 10/100baseTX" },
175 	{ RT_VENDORID, RT_DEVICEID_8168, RL_HWREV_8168_SPIN1,
176 		"RealTek 8168/8111B PCIe Gigabit Ethernet" },
177 	{ RT_VENDORID, RT_DEVICEID_8168, RL_HWREV_8168_SPIN2,
178 		"RealTek 8168/8111B PCIe Gigabit Ethernet" },
179 	{ RT_VENDORID, RT_DEVICEID_8169, RL_HWREV_8169,
180 		"RealTek 8169 Gigabit Ethernet" },
181 	{ RT_VENDORID, RT_DEVICEID_8169, RL_HWREV_8169S,
182 		"RealTek 8169S Single-chip Gigabit Ethernet" },
183 	{ RT_VENDORID, RT_DEVICEID_8169, RL_HWREV_8169_8110SB,
184 		"RealTek 8169SB/8110SB Single-chip Gigabit Ethernet" },
185 	{ RT_VENDORID, RT_DEVICEID_8169SC, RL_HWREV_8169_8110SC,
186 		"RealTek 8169SC/8110SC Single-chip Gigabit Ethernet" },
187 	{ RT_VENDORID, RT_DEVICEID_8169, RL_HWREV_8110S,
188 		"RealTek 8110S Single-chip Gigabit Ethernet" },
189 	{ COREGA_VENDORID, COREGA_DEVICEID_CGLAPCIGT, RL_HWREV_8169S,
190 		"Corega CG-LAPCIGT (RTL8169S) Gigabit Ethernet" },
191 	{ LINKSYS_VENDORID, LINKSYS_DEVICEID_EG1032, RL_HWREV_8169S,
192 		"Linksys EG1032 (RTL8169S) Gigabit Ethernet" },
193 	{ USR_VENDORID, USR_DEVICEID_997902, RL_HWREV_8169S,
194 		"US Robotics 997902 (RTL8169S) Gigabit Ethernet" },
195 	{ 0, 0, 0, NULL }
196 };
197 
198 static struct rl_hwrev re_hwrevs[] = {
199 	{ RL_HWREV_8139, RL_8139,  "" },
200 	{ RL_HWREV_8139A, RL_8139, "A" },
201 	{ RL_HWREV_8139AG, RL_8139, "A-G" },
202 	{ RL_HWREV_8139B, RL_8139, "B" },
203 	{ RL_HWREV_8130, RL_8139, "8130" },
204 	{ RL_HWREV_8139C, RL_8139, "C" },
205 	{ RL_HWREV_8139D, RL_8139, "8139D/8100B/8100C" },
206 	{ RL_HWREV_8139CPLUS, RL_8139CPLUS, "C+"},
207 	{ RL_HWREV_8168_SPIN1, RL_8169, "8168"},
208 	{ RL_HWREV_8169, RL_8169, "8169"},
209 	{ RL_HWREV_8169S, RL_8169, "8169S"},
210 	{ RL_HWREV_8110S, RL_8169, "8110S"},
211 	{ RL_HWREV_8169_8110SB, RL_8169, "8169SB"},
212 	{ RL_HWREV_8169_8110SC, RL_8169, "8169SC"},
213 	{ RL_HWREV_8100, RL_8139, "8100"},
214 	{ RL_HWREV_8101, RL_8139, "8101"},
215 	{ RL_HWREV_8100E, RL_8169, "8100E"},
216 	{ RL_HWREV_8101E, RL_8169, "8101E"},
217 	{ RL_HWREV_8168_SPIN2, RL_8169, "8168"},
218 	{ 0, 0, NULL }
219 };
220 
221 static int re_probe		(device_t);
222 static int re_attach		(device_t);
223 static int re_detach		(device_t);
224 
225 static int re_encap		(struct rl_softc *, struct mbuf **, int *);
226 
227 static void re_dma_map_addr	(void *, bus_dma_segment_t *, int, int);
228 static void re_dma_map_desc	(void *, bus_dma_segment_t *, int,
229 				    bus_size_t, int);
230 static int re_allocmem		(device_t, struct rl_softc *);
231 static int re_newbuf		(struct rl_softc *, int, struct mbuf *);
232 static int re_rx_list_init	(struct rl_softc *);
233 static int re_tx_list_init	(struct rl_softc *);
234 #ifdef RE_FIXUP_RX
235 static __inline void re_fixup_rx
236 				(struct mbuf *);
237 #endif
238 static int re_rxeof		(struct rl_softc *);
239 static void re_txeof		(struct rl_softc *);
240 #ifdef DEVICE_POLLING
241 static void re_poll		(struct ifnet *, enum poll_cmd, int);
242 static void re_poll_locked	(struct ifnet *, enum poll_cmd, int);
243 #endif
244 static void re_intr		(void *);
245 static void re_tick		(void *);
246 static void re_tx_task		(void *, int);
247 static void re_int_task		(void *, int);
248 static void re_start		(struct ifnet *);
249 static int re_ioctl		(struct ifnet *, u_long, caddr_t);
250 static void re_init		(void *);
251 static void re_init_locked	(struct rl_softc *);
252 static void re_stop		(struct rl_softc *);
253 static void re_watchdog		(struct rl_softc *);
254 static int re_suspend		(device_t);
255 static int re_resume		(device_t);
256 static void re_shutdown		(device_t);
257 static int re_ifmedia_upd	(struct ifnet *);
258 static void re_ifmedia_sts	(struct ifnet *, struct ifmediareq *);
259 
260 static void re_eeprom_putbyte	(struct rl_softc *, int);
261 static void re_eeprom_getword	(struct rl_softc *, int, u_int16_t *);
262 static void re_read_eeprom	(struct rl_softc *, caddr_t, int, int);
263 static int re_gmii_readreg	(device_t, int, int);
264 static int re_gmii_writereg	(device_t, int, int, int);
265 
266 static int re_miibus_readreg	(device_t, int, int);
267 static int re_miibus_writereg	(device_t, int, int, int);
268 static void re_miibus_statchg	(device_t);
269 
270 static void re_setmulti		(struct rl_softc *);
271 static void re_reset		(struct rl_softc *);
272 
273 #ifdef RE_DIAG
274 static int re_diag		(struct rl_softc *);
275 #endif
276 
277 #ifdef RE_USEIOSPACE
278 #define RL_RES			SYS_RES_IOPORT
279 #define RL_RID			RL_PCI_LOIO
280 #else
281 #define RL_RES			SYS_RES_MEMORY
282 #define RL_RID			RL_PCI_LOMEM
283 #endif
284 
285 static device_method_t re_methods[] = {
286 	/* Device interface */
287 	DEVMETHOD(device_probe,		re_probe),
288 	DEVMETHOD(device_attach,	re_attach),
289 	DEVMETHOD(device_detach,	re_detach),
290 	DEVMETHOD(device_suspend,	re_suspend),
291 	DEVMETHOD(device_resume,	re_resume),
292 	DEVMETHOD(device_shutdown,	re_shutdown),
293 
294 	/* bus interface */
295 	DEVMETHOD(bus_print_child,	bus_generic_print_child),
296 	DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
297 
298 	/* MII interface */
299 	DEVMETHOD(miibus_readreg,	re_miibus_readreg),
300 	DEVMETHOD(miibus_writereg,	re_miibus_writereg),
301 	DEVMETHOD(miibus_statchg,	re_miibus_statchg),
302 
303 	{ 0, 0 }
304 };
305 
306 static driver_t re_driver = {
307 	"re",
308 	re_methods,
309 	sizeof(struct rl_softc)
310 };
311 
312 static devclass_t re_devclass;
313 
314 DRIVER_MODULE(re, pci, re_driver, re_devclass, 0, 0);
315 DRIVER_MODULE(re, cardbus, re_driver, re_devclass, 0, 0);
316 DRIVER_MODULE(miibus, re, miibus_driver, miibus_devclass, 0, 0);
317 
318 #define EE_SET(x)					\
319 	CSR_WRITE_1(sc, RL_EECMD,			\
320 		CSR_READ_1(sc, RL_EECMD) | x)
321 
322 #define EE_CLR(x)					\
323 	CSR_WRITE_1(sc, RL_EECMD,			\
324 		CSR_READ_1(sc, RL_EECMD) & ~x)
325 
326 /*
327  * Send a read command and address to the EEPROM, check for ACK.
328  */
329 static void
330 re_eeprom_putbyte(sc, addr)
331 	struct rl_softc		*sc;
332 	int			addr;
333 {
334 	register int		d, i;
335 
336 	d = addr | (RL_9346_READ << sc->rl_eewidth);
337 
338 	/*
339 	 * Feed in each bit and strobe the clock.
340 	 */
341 
342 	for (i = 1 << (sc->rl_eewidth + 3); i; i >>= 1) {
343 		if (d & i) {
344 			EE_SET(RL_EE_DATAIN);
345 		} else {
346 			EE_CLR(RL_EE_DATAIN);
347 		}
348 		DELAY(100);
349 		EE_SET(RL_EE_CLK);
350 		DELAY(150);
351 		EE_CLR(RL_EE_CLK);
352 		DELAY(100);
353 	}
354 
355 	return;
356 }
357 
358 /*
359  * Read a word of data stored in the EEPROM at address 'addr.'
360  */
361 static void
362 re_eeprom_getword(sc, addr, dest)
363 	struct rl_softc		*sc;
364 	int			addr;
365 	u_int16_t		*dest;
366 {
367 	register int		i;
368 	u_int16_t		word = 0;
369 
370 	/*
371 	 * Send address of word we want to read.
372 	 */
373 	re_eeprom_putbyte(sc, addr);
374 
375 	/*
376 	 * Start reading bits from EEPROM.
377 	 */
378 	for (i = 0x8000; i; i >>= 1) {
379 		EE_SET(RL_EE_CLK);
380 		DELAY(100);
381 		if (CSR_READ_1(sc, RL_EECMD) & RL_EE_DATAOUT)
382 			word |= i;
383 		EE_CLR(RL_EE_CLK);
384 		DELAY(100);
385 	}
386 
387 	*dest = word;
388 
389 	return;
390 }
391 
392 /*
393  * Read a sequence of words from the EEPROM.
394  */
395 static void
396 re_read_eeprom(sc, dest, off, cnt)
397 	struct rl_softc		*sc;
398 	caddr_t			dest;
399 	int			off;
400 	int			cnt;
401 {
402 	int			i;
403 	u_int16_t		word = 0, *ptr;
404 
405 	CSR_SETBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM);
406 
407         DELAY(100);
408 
409 	for (i = 0; i < cnt; i++) {
410 		CSR_SETBIT_1(sc, RL_EECMD, RL_EE_SEL);
411 		re_eeprom_getword(sc, off + i, &word);
412 		CSR_CLRBIT_1(sc, RL_EECMD, RL_EE_SEL);
413 		ptr = (u_int16_t *)(dest + (i * 2));
414                 *ptr = word;
415 	}
416 
417 	CSR_CLRBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM);
418 
419 	return;
420 }
421 
422 static int
423 re_gmii_readreg(dev, phy, reg)
424 	device_t		dev;
425 	int			phy, reg;
426 {
427 	struct rl_softc		*sc;
428 	u_int32_t		rval;
429 	int			i;
430 
431 	if (phy != 1)
432 		return (0);
433 
434 	sc = device_get_softc(dev);
435 
436 	/* Let the rgephy driver read the GMEDIASTAT register */
437 
438 	if (reg == RL_GMEDIASTAT) {
439 		rval = CSR_READ_1(sc, RL_GMEDIASTAT);
440 		return (rval);
441 	}
442 
443 	CSR_WRITE_4(sc, RL_PHYAR, reg << 16);
444 	DELAY(1000);
445 
446 	for (i = 0; i < RL_TIMEOUT; i++) {
447 		rval = CSR_READ_4(sc, RL_PHYAR);
448 		if (rval & RL_PHYAR_BUSY)
449 			break;
450 		DELAY(100);
451 	}
452 
453 	if (i == RL_TIMEOUT) {
454 		device_printf(sc->rl_dev, "PHY read failed\n");
455 		return (0);
456 	}
457 
458 	return (rval & RL_PHYAR_PHYDATA);
459 }
460 
461 static int
462 re_gmii_writereg(dev, phy, reg, data)
463 	device_t		dev;
464 	int			phy, reg, data;
465 {
466 	struct rl_softc		*sc;
467 	u_int32_t		rval;
468 	int			i;
469 
470 	sc = device_get_softc(dev);
471 
472 	CSR_WRITE_4(sc, RL_PHYAR, (reg << 16) |
473 	    (data & RL_PHYAR_PHYDATA) | RL_PHYAR_BUSY);
474 	DELAY(1000);
475 
476 	for (i = 0; i < RL_TIMEOUT; i++) {
477 		rval = CSR_READ_4(sc, RL_PHYAR);
478 		if (!(rval & RL_PHYAR_BUSY))
479 			break;
480 		DELAY(100);
481 	}
482 
483 	if (i == RL_TIMEOUT) {
484 		device_printf(sc->rl_dev, "PHY write failed\n");
485 		return (0);
486 	}
487 
488 	return (0);
489 }
490 
491 static int
492 re_miibus_readreg(dev, phy, reg)
493 	device_t		dev;
494 	int			phy, reg;
495 {
496 	struct rl_softc		*sc;
497 	u_int16_t		rval = 0;
498 	u_int16_t		re8139_reg = 0;
499 
500 	sc = device_get_softc(dev);
501 
502 	if (sc->rl_type == RL_8169) {
503 		rval = re_gmii_readreg(dev, phy, reg);
504 		return (rval);
505 	}
506 
507 	/* Pretend the internal PHY is only at address 0 */
508 	if (phy) {
509 		return (0);
510 	}
511 	switch (reg) {
512 	case MII_BMCR:
513 		re8139_reg = RL_BMCR;
514 		break;
515 	case MII_BMSR:
516 		re8139_reg = RL_BMSR;
517 		break;
518 	case MII_ANAR:
519 		re8139_reg = RL_ANAR;
520 		break;
521 	case MII_ANER:
522 		re8139_reg = RL_ANER;
523 		break;
524 	case MII_ANLPAR:
525 		re8139_reg = RL_LPAR;
526 		break;
527 	case MII_PHYIDR1:
528 	case MII_PHYIDR2:
529 		return (0);
530 	/*
531 	 * Allow the rlphy driver to read the media status
532 	 * register. If we have a link partner which does not
533 	 * support NWAY, this is the register which will tell
534 	 * us the results of parallel detection.
535 	 */
536 	case RL_MEDIASTAT:
537 		rval = CSR_READ_1(sc, RL_MEDIASTAT);
538 		return (rval);
539 	default:
540 		device_printf(sc->rl_dev, "bad phy register\n");
541 		return (0);
542 	}
543 	rval = CSR_READ_2(sc, re8139_reg);
544 	if (sc->rl_type == RL_8139CPLUS && re8139_reg == RL_BMCR) {
545 		/* 8139C+ has different bit layout. */
546 		rval &= ~(BMCR_LOOP | BMCR_ISO);
547 	}
548 	return (rval);
549 }
550 
551 static int
552 re_miibus_writereg(dev, phy, reg, data)
553 	device_t		dev;
554 	int			phy, reg, data;
555 {
556 	struct rl_softc		*sc;
557 	u_int16_t		re8139_reg = 0;
558 	int			rval = 0;
559 
560 	sc = device_get_softc(dev);
561 
562 	if (sc->rl_type == RL_8169) {
563 		rval = re_gmii_writereg(dev, phy, reg, data);
564 		return (rval);
565 	}
566 
567 	/* Pretend the internal PHY is only at address 0 */
568 	if (phy)
569 		return (0);
570 
571 	switch (reg) {
572 	case MII_BMCR:
573 		re8139_reg = RL_BMCR;
574 		if (sc->rl_type == RL_8139CPLUS) {
575 			/* 8139C+ has different bit layout. */
576 			data &= ~(BMCR_LOOP | BMCR_ISO);
577 		}
578 		break;
579 	case MII_BMSR:
580 		re8139_reg = RL_BMSR;
581 		break;
582 	case MII_ANAR:
583 		re8139_reg = RL_ANAR;
584 		break;
585 	case MII_ANER:
586 		re8139_reg = RL_ANER;
587 		break;
588 	case MII_ANLPAR:
589 		re8139_reg = RL_LPAR;
590 		break;
591 	case MII_PHYIDR1:
592 	case MII_PHYIDR2:
593 		return (0);
594 		break;
595 	default:
596 		device_printf(sc->rl_dev, "bad phy register\n");
597 		return (0);
598 	}
599 	CSR_WRITE_2(sc, re8139_reg, data);
600 	return (0);
601 }
602 
603 static void
604 re_miibus_statchg(dev)
605 	device_t		dev;
606 {
607 
608 }
609 
610 /*
611  * Program the 64-bit multicast hash filter.
612  */
613 static void
614 re_setmulti(sc)
615 	struct rl_softc		*sc;
616 {
617 	struct ifnet		*ifp;
618 	int			h = 0;
619 	u_int32_t		hashes[2] = { 0, 0 };
620 	struct ifmultiaddr	*ifma;
621 	u_int32_t		rxfilt;
622 	int			mcnt = 0;
623 
624 	RL_LOCK_ASSERT(sc);
625 
626 	ifp = sc->rl_ifp;
627 
628 	rxfilt = CSR_READ_4(sc, RL_RXCFG);
629 
630 	if (ifp->if_flags & IFF_ALLMULTI || ifp->if_flags & IFF_PROMISC) {
631 		rxfilt |= RL_RXCFG_RX_MULTI;
632 		CSR_WRITE_4(sc, RL_RXCFG, rxfilt);
633 		CSR_WRITE_4(sc, RL_MAR0, 0xFFFFFFFF);
634 		CSR_WRITE_4(sc, RL_MAR4, 0xFFFFFFFF);
635 		return;
636 	}
637 
638 	/* first, zot all the existing hash bits */
639 	CSR_WRITE_4(sc, RL_MAR0, 0);
640 	CSR_WRITE_4(sc, RL_MAR4, 0);
641 
642 	/* now program new ones */
643 	IF_ADDR_LOCK(ifp);
644 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
645 		if (ifma->ifma_addr->sa_family != AF_LINK)
646 			continue;
647 		h = ether_crc32_be(LLADDR((struct sockaddr_dl *)
648 		    ifma->ifma_addr), ETHER_ADDR_LEN) >> 26;
649 		if (h < 32)
650 			hashes[0] |= (1 << h);
651 		else
652 			hashes[1] |= (1 << (h - 32));
653 		mcnt++;
654 	}
655 	IF_ADDR_UNLOCK(ifp);
656 
657 	if (mcnt)
658 		rxfilt |= RL_RXCFG_RX_MULTI;
659 	else
660 		rxfilt &= ~RL_RXCFG_RX_MULTI;
661 
662 	CSR_WRITE_4(sc, RL_RXCFG, rxfilt);
663 	CSR_WRITE_4(sc, RL_MAR0, hashes[0]);
664 	CSR_WRITE_4(sc, RL_MAR4, hashes[1]);
665 }
666 
667 static void
668 re_reset(sc)
669 	struct rl_softc		*sc;
670 {
671 	register int		i;
672 
673 	RL_LOCK_ASSERT(sc);
674 
675 	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RESET);
676 
677 	for (i = 0; i < RL_TIMEOUT; i++) {
678 		DELAY(10);
679 		if (!(CSR_READ_1(sc, RL_COMMAND) & RL_CMD_RESET))
680 			break;
681 	}
682 	if (i == RL_TIMEOUT)
683 		device_printf(sc->rl_dev, "reset never completed!\n");
684 
685 	CSR_WRITE_1(sc, 0x82, 1);
686 }
687 
688 #ifdef RE_DIAG
689 
690 /*
691  * The following routine is designed to test for a defect on some
692  * 32-bit 8169 cards. Some of these NICs have the REQ64# and ACK64#
693  * lines connected to the bus, however for a 32-bit only card, they
694  * should be pulled high. The result of this defect is that the
695  * NIC will not work right if you plug it into a 64-bit slot: DMA
696  * operations will be done with 64-bit transfers, which will fail
697  * because the 64-bit data lines aren't connected.
698  *
699  * There's no way to work around this (short of talking a soldering
700  * iron to the board), however we can detect it. The method we use
701  * here is to put the NIC into digital loopback mode, set the receiver
702  * to promiscuous mode, and then try to send a frame. We then compare
703  * the frame data we sent to what was received. If the data matches,
704  * then the NIC is working correctly, otherwise we know the user has
705  * a defective NIC which has been mistakenly plugged into a 64-bit PCI
706  * slot. In the latter case, there's no way the NIC can work correctly,
707  * so we print out a message on the console and abort the device attach.
708  */
709 
710 static int
711 re_diag(sc)
712 	struct rl_softc		*sc;
713 {
714 	struct ifnet		*ifp = sc->rl_ifp;
715 	struct mbuf		*m0;
716 	struct ether_header	*eh;
717 	struct rl_desc		*cur_rx;
718 	u_int16_t		status;
719 	u_int32_t		rxstat;
720 	int			total_len, i, error = 0, phyaddr;
721 	u_int8_t		dst[] = { 0x00, 'h', 'e', 'l', 'l', 'o' };
722 	u_int8_t		src[] = { 0x00, 'w', 'o', 'r', 'l', 'd' };
723 
724 	/* Allocate a single mbuf */
725 	MGETHDR(m0, M_DONTWAIT, MT_DATA);
726 	if (m0 == NULL)
727 		return (ENOBUFS);
728 
729 	RL_LOCK(sc);
730 
731 	/*
732 	 * Initialize the NIC in test mode. This sets the chip up
733 	 * so that it can send and receive frames, but performs the
734 	 * following special functions:
735 	 * - Puts receiver in promiscuous mode
736 	 * - Enables digital loopback mode
737 	 * - Leaves interrupts turned off
738 	 */
739 
740 	ifp->if_flags |= IFF_PROMISC;
741 	sc->rl_testmode = 1;
742 	re_reset(sc);
743 	re_init_locked(sc);
744 	sc->rl_link = 1;
745 	if (sc->rl_type == RL_8169)
746 		phyaddr = 1;
747 	else
748 		phyaddr = 0;
749 
750 	re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_RESET);
751 	for (i = 0; i < RL_TIMEOUT; i++) {
752 		status = re_miibus_readreg(sc->rl_dev, phyaddr, MII_BMCR);
753 		if (!(status & BMCR_RESET))
754 			break;
755 	}
756 
757 	re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_LOOP);
758 	CSR_WRITE_2(sc, RL_ISR, RL_INTRS);
759 
760 	DELAY(100000);
761 
762 	/* Put some data in the mbuf */
763 
764 	eh = mtod(m0, struct ether_header *);
765 	bcopy ((char *)&dst, eh->ether_dhost, ETHER_ADDR_LEN);
766 	bcopy ((char *)&src, eh->ether_shost, ETHER_ADDR_LEN);
767 	eh->ether_type = htons(ETHERTYPE_IP);
768 	m0->m_pkthdr.len = m0->m_len = ETHER_MIN_LEN - ETHER_CRC_LEN;
769 
770 	/*
771 	 * Queue the packet, start transmission.
772 	 * Note: IF_HANDOFF() ultimately calls re_start() for us.
773 	 */
774 
775 	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
776 	RL_UNLOCK(sc);
777 	/* XXX: re_diag must not be called when in ALTQ mode */
778 	IF_HANDOFF(&ifp->if_snd, m0, ifp);
779 	RL_LOCK(sc);
780 	m0 = NULL;
781 
782 	/* Wait for it to propagate through the chip */
783 
784 	DELAY(100000);
785 	for (i = 0; i < RL_TIMEOUT; i++) {
786 		status = CSR_READ_2(sc, RL_ISR);
787 		CSR_WRITE_2(sc, RL_ISR, status);
788 		if ((status & (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK)) ==
789 		    (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK))
790 			break;
791 		DELAY(10);
792 	}
793 
794 	if (i == RL_TIMEOUT) {
795 		device_printf(sc->rl_dev,
796 		    "diagnostic failed, failed to receive packet in"
797 		    " loopback mode\n");
798 		error = EIO;
799 		goto done;
800 	}
801 
802 	/*
803 	 * The packet should have been dumped into the first
804 	 * entry in the RX DMA ring. Grab it from there.
805 	 */
806 
807 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
808 	    sc->rl_ldata.rl_rx_list_map,
809 	    BUS_DMASYNC_POSTREAD);
810 	bus_dmamap_sync(sc->rl_ldata.rl_mtag,
811 	    sc->rl_ldata.rl_rx_dmamap[0],
812 	    BUS_DMASYNC_POSTWRITE);
813 	bus_dmamap_unload(sc->rl_ldata.rl_mtag,
814 	    sc->rl_ldata.rl_rx_dmamap[0]);
815 
816 	m0 = sc->rl_ldata.rl_rx_mbuf[0];
817 	sc->rl_ldata.rl_rx_mbuf[0] = NULL;
818 	eh = mtod(m0, struct ether_header *);
819 
820 	cur_rx = &sc->rl_ldata.rl_rx_list[0];
821 	total_len = RL_RXBYTES(cur_rx);
822 	rxstat = le32toh(cur_rx->rl_cmdstat);
823 
824 	if (total_len != ETHER_MIN_LEN) {
825 		device_printf(sc->rl_dev,
826 		    "diagnostic failed, received short packet\n");
827 		error = EIO;
828 		goto done;
829 	}
830 
831 	/* Test that the received packet data matches what we sent. */
832 
833 	if (bcmp((char *)&eh->ether_dhost, (char *)&dst, ETHER_ADDR_LEN) ||
834 	    bcmp((char *)&eh->ether_shost, (char *)&src, ETHER_ADDR_LEN) ||
835 	    ntohs(eh->ether_type) != ETHERTYPE_IP) {
836 		device_printf(sc->rl_dev, "WARNING, DMA FAILURE!\n");
837 		device_printf(sc->rl_dev, "expected TX data: %6D/%6D/0x%x\n",
838 		    dst, ":", src, ":", ETHERTYPE_IP);
839 		device_printf(sc->rl_dev, "received RX data: %6D/%6D/0x%x\n",
840 		    eh->ether_dhost, ":",  eh->ether_shost, ":",
841 		    ntohs(eh->ether_type));
842 		device_printf(sc->rl_dev, "You may have a defective 32-bit "
843 		    "NIC plugged into a 64-bit PCI slot.\n");
844 		device_printf(sc->rl_dev, "Please re-install the NIC in a "
845 		    "32-bit slot for proper operation.\n");
846 		device_printf(sc->rl_dev, "Read the re(4) man page for more "
847 		    "details.\n");
848 		error = EIO;
849 	}
850 
851 done:
852 	/* Turn interface off, release resources */
853 
854 	sc->rl_testmode = 0;
855 	sc->rl_link = 0;
856 	ifp->if_flags &= ~IFF_PROMISC;
857 	re_stop(sc);
858 	if (m0 != NULL)
859 		m_freem(m0);
860 
861 	RL_UNLOCK(sc);
862 
863 	return (error);
864 }
865 
866 #endif
867 
868 /*
869  * Probe for a RealTek 8139C+/8169/8110 chip. Check the PCI vendor and device
870  * IDs against our list and return a device name if we find a match.
871  */
872 static int
873 re_probe(dev)
874 	device_t		dev;
875 {
876 	struct rl_type		*t;
877 	struct rl_softc		*sc;
878 	int			rid;
879 	u_int32_t		hwrev;
880 
881 	t = re_devs;
882 	sc = device_get_softc(dev);
883 
884 	while (t->rl_name != NULL) {
885 		if ((pci_get_vendor(dev) == t->rl_vid) &&
886 		    (pci_get_device(dev) == t->rl_did)) {
887 			/*
888 			 * Only attach to rev. 3 of the Linksys EG1032 adapter.
889 			 * Rev. 2 i supported by sk(4).
890 			 */
891 			if ((t->rl_vid == LINKSYS_VENDORID) &&
892 				(t->rl_did == LINKSYS_DEVICEID_EG1032) &&
893 				(pci_get_subdevice(dev) !=
894 				LINKSYS_SUBDEVICE_EG1032_REV3)) {
895 				t++;
896 				continue;
897 			}
898 
899 			/*
900 			 * Temporarily map the I/O space
901 			 * so we can read the chip ID register.
902 			 */
903 			rid = RL_RID;
904 			sc->rl_res = bus_alloc_resource_any(dev, RL_RES, &rid,
905 			    RF_ACTIVE);
906 			if (sc->rl_res == NULL) {
907 				device_printf(dev,
908 				    "couldn't map ports/memory\n");
909 				return (ENXIO);
910 			}
911 			sc->rl_btag = rman_get_bustag(sc->rl_res);
912 			sc->rl_bhandle = rman_get_bushandle(sc->rl_res);
913 			hwrev = CSR_READ_4(sc, RL_TXCFG) & RL_TXCFG_HWREV;
914 			bus_release_resource(dev, RL_RES,
915 			    RL_RID, sc->rl_res);
916 			if (t->rl_basetype == hwrev) {
917 				device_set_desc(dev, t->rl_name);
918 				return (BUS_PROBE_DEFAULT);
919 			}
920 		}
921 		t++;
922 	}
923 
924 	return (ENXIO);
925 }
926 
927 /*
928  * This routine takes the segment list provided as the result of
929  * a bus_dma_map_load() operation and assigns the addresses/lengths
930  * to RealTek DMA descriptors. This can be called either by the RX
931  * code or the TX code. In the RX case, we'll probably wind up mapping
932  * at most one segment. For the TX case, there could be any number of
933  * segments since TX packets may span multiple mbufs. In either case,
934  * if the number of segments is larger than the rl_maxsegs limit
935  * specified by the caller, we abort the mapping operation. Sadly,
936  * whoever designed the buffer mapping API did not provide a way to
937  * return an error from here, so we have to fake it a bit.
938  */
939 
940 static void
941 re_dma_map_desc(arg, segs, nseg, mapsize, error)
942 	void			*arg;
943 	bus_dma_segment_t	*segs;
944 	int			nseg;
945 	bus_size_t		mapsize;
946 	int			error;
947 {
948 	struct rl_dmaload_arg	*ctx;
949 	struct rl_desc		*d = NULL;
950 	int			i = 0, idx;
951 	u_int32_t		cmdstat;
952 	int			totlen = 0;
953 
954 	if (error)
955 		return;
956 
957 	ctx = arg;
958 
959 	/* Signal error to caller if there's too many segments */
960 	if (nseg > ctx->rl_maxsegs) {
961 		ctx->rl_maxsegs = 0;
962 		return;
963 	}
964 
965 	/*
966 	 * Map the segment array into descriptors. Note that we set the
967 	 * start-of-frame and end-of-frame markers for either TX or RX, but
968 	 * they really only have meaning in the TX case. (In the RX case,
969 	 * it's the chip that tells us where packets begin and end.)
970 	 * We also keep track of the end of the ring and set the
971 	 * end-of-ring bits as needed, and we set the ownership bits
972 	 * in all except the very first descriptor. (The caller will
973 	 * set this descriptor later when it start transmission or
974 	 * reception.)
975 	 */
976 	idx = ctx->rl_idx;
977 	for (;;) {
978 		d = &ctx->rl_ring[idx];
979 		if (le32toh(d->rl_cmdstat) & RL_RDESC_STAT_OWN) {
980 			ctx->rl_maxsegs = 0;
981 			return;
982 		}
983 		cmdstat = segs[i].ds_len;
984 		totlen += segs[i].ds_len;
985 		d->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[i].ds_addr));
986 		d->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[i].ds_addr));
987 		if (i == 0)
988 			cmdstat |= RL_TDESC_CMD_SOF;
989 		else
990 			cmdstat |= RL_TDESC_CMD_OWN;
991 		if (idx == (RL_RX_DESC_CNT - 1))
992 			cmdstat |= RL_TDESC_CMD_EOR;
993 		d->rl_cmdstat = htole32(cmdstat | ctx->rl_flags);
994 		i++;
995 		if (i == nseg)
996 			break;
997 		RL_DESC_INC(idx);
998 	}
999 
1000 	d->rl_cmdstat |= htole32(RL_TDESC_CMD_EOF);
1001 	ctx->rl_maxsegs = nseg;
1002 	ctx->rl_idx = idx;
1003 }
1004 
1005 /*
1006  * Map a single buffer address.
1007  */
1008 
1009 static void
1010 re_dma_map_addr(arg, segs, nseg, error)
1011 	void			*arg;
1012 	bus_dma_segment_t	*segs;
1013 	int			nseg;
1014 	int			error;
1015 {
1016 	bus_addr_t		*addr;
1017 
1018 	if (error)
1019 		return;
1020 
1021 	KASSERT(nseg == 1, ("too many DMA segments, %d should be 1", nseg));
1022 	addr = arg;
1023 	*addr = segs->ds_addr;
1024 }
1025 
1026 static int
1027 re_allocmem(dev, sc)
1028 	device_t		dev;
1029 	struct rl_softc		*sc;
1030 {
1031 	int			error;
1032 	int			nseg;
1033 	int			i;
1034 
1035 	/*
1036 	 * Allocate map for RX mbufs.
1037 	 */
1038 	nseg = 32;
1039 	error = bus_dma_tag_create(sc->rl_parent_tag, ETHER_ALIGN, 0,
1040 	    BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1041 	    NULL, MCLBYTES * nseg, nseg, MCLBYTES, BUS_DMA_ALLOCNOW,
1042 	    NULL, NULL, &sc->rl_ldata.rl_mtag);
1043 	if (error) {
1044 		device_printf(dev, "could not allocate dma tag\n");
1045 		return (ENOMEM);
1046 	}
1047 
1048 	/*
1049 	 * Allocate map for TX descriptor list.
1050 	 */
1051 	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
1052 	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1053 	    NULL, RL_TX_LIST_SZ, 1, RL_TX_LIST_SZ, 0,
1054 	    NULL, NULL, &sc->rl_ldata.rl_tx_list_tag);
1055 	if (error) {
1056 		device_printf(dev, "could not allocate dma tag\n");
1057 		return (ENOMEM);
1058 	}
1059 
1060 	/* Allocate DMA'able memory for the TX ring */
1061 
1062 	error = bus_dmamem_alloc(sc->rl_ldata.rl_tx_list_tag,
1063 	    (void **)&sc->rl_ldata.rl_tx_list, BUS_DMA_NOWAIT | BUS_DMA_ZERO,
1064 	    &sc->rl_ldata.rl_tx_list_map);
1065 	if (error)
1066 		return (ENOMEM);
1067 
1068 	/* Load the map for the TX ring. */
1069 
1070 	error = bus_dmamap_load(sc->rl_ldata.rl_tx_list_tag,
1071 	     sc->rl_ldata.rl_tx_list_map, sc->rl_ldata.rl_tx_list,
1072 	     RL_TX_LIST_SZ, re_dma_map_addr,
1073 	     &sc->rl_ldata.rl_tx_list_addr, BUS_DMA_NOWAIT);
1074 
1075 	/* Create DMA maps for TX buffers */
1076 
1077 	for (i = 0; i < RL_TX_DESC_CNT; i++) {
1078 		error = bus_dmamap_create(sc->rl_ldata.rl_mtag, 0,
1079 			    &sc->rl_ldata.rl_tx_dmamap[i]);
1080 		if (error) {
1081 			device_printf(dev, "can't create DMA map for TX\n");
1082 			return (ENOMEM);
1083 		}
1084 	}
1085 
1086 	/*
1087 	 * Allocate map for RX descriptor list.
1088 	 */
1089 	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
1090 	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1091 	    NULL, RL_RX_LIST_SZ, 1, RL_RX_LIST_SZ, 0,
1092 	    NULL, NULL, &sc->rl_ldata.rl_rx_list_tag);
1093 	if (error) {
1094 		device_printf(dev, "could not allocate dma tag\n");
1095 		return (ENOMEM);
1096 	}
1097 
1098 	/* Allocate DMA'able memory for the RX ring */
1099 
1100 	error = bus_dmamem_alloc(sc->rl_ldata.rl_rx_list_tag,
1101 	    (void **)&sc->rl_ldata.rl_rx_list, BUS_DMA_NOWAIT | BUS_DMA_ZERO,
1102 	    &sc->rl_ldata.rl_rx_list_map);
1103 	if (error)
1104 		return (ENOMEM);
1105 
1106 	/* Load the map for the RX ring. */
1107 
1108 	error = bus_dmamap_load(sc->rl_ldata.rl_rx_list_tag,
1109 	     sc->rl_ldata.rl_rx_list_map, sc->rl_ldata.rl_rx_list,
1110 	     RL_RX_LIST_SZ, re_dma_map_addr,
1111 	     &sc->rl_ldata.rl_rx_list_addr, BUS_DMA_NOWAIT);
1112 
1113 	/* Create DMA maps for RX buffers */
1114 
1115 	for (i = 0; i < RL_RX_DESC_CNT; i++) {
1116 		error = bus_dmamap_create(sc->rl_ldata.rl_mtag, 0,
1117 			    &sc->rl_ldata.rl_rx_dmamap[i]);
1118 		if (error) {
1119 			device_printf(dev, "can't create DMA map for RX\n");
1120 			return (ENOMEM);
1121 		}
1122 	}
1123 
1124 	return (0);
1125 }
1126 
1127 /*
1128  * Attach the interface. Allocate softc structures, do ifmedia
1129  * setup and ethernet/BPF attach.
1130  */
1131 static int
1132 re_attach(dev)
1133 	device_t		dev;
1134 {
1135 	u_char			eaddr[ETHER_ADDR_LEN];
1136 	u_int16_t		as[ETHER_ADDR_LEN / 2];
1137 	struct rl_softc		*sc;
1138 	struct ifnet		*ifp;
1139 	struct rl_hwrev		*hw_rev;
1140 	int			hwrev;
1141 	u_int16_t		re_did = 0;
1142 	int			error = 0, rid, i;
1143 
1144 	sc = device_get_softc(dev);
1145 	sc->rl_dev = dev;
1146 
1147 	mtx_init(&sc->rl_mtx, device_get_nameunit(dev), MTX_NETWORK_LOCK,
1148 	    MTX_DEF);
1149 	callout_init_mtx(&sc->rl_stat_callout, &sc->rl_mtx, 0);
1150 
1151 	/*
1152 	 * Map control/status registers.
1153 	 */
1154 	pci_enable_busmaster(dev);
1155 
1156 	rid = RL_RID;
1157 	sc->rl_res = bus_alloc_resource_any(dev, RL_RES, &rid,
1158 	    RF_ACTIVE);
1159 
1160 	if (sc->rl_res == NULL) {
1161 		device_printf(dev, "couldn't map ports/memory\n");
1162 		error = ENXIO;
1163 		goto fail;
1164 	}
1165 
1166 	sc->rl_btag = rman_get_bustag(sc->rl_res);
1167 	sc->rl_bhandle = rman_get_bushandle(sc->rl_res);
1168 
1169 	/* Allocate interrupt */
1170 	rid = 0;
1171 	sc->rl_irq = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
1172 	    RF_SHAREABLE | RF_ACTIVE);
1173 
1174 	if (sc->rl_irq == NULL) {
1175 		device_printf(dev, "couldn't map interrupt\n");
1176 		error = ENXIO;
1177 		goto fail;
1178 	}
1179 
1180 	/* Reset the adapter. */
1181 	RL_LOCK(sc);
1182 	re_reset(sc);
1183 	RL_UNLOCK(sc);
1184 
1185 	hw_rev = re_hwrevs;
1186 	hwrev = CSR_READ_4(sc, RL_TXCFG) & RL_TXCFG_HWREV;
1187 	while (hw_rev->rl_desc != NULL) {
1188 		if (hw_rev->rl_rev == hwrev) {
1189 			sc->rl_type = hw_rev->rl_type;
1190 			break;
1191 		}
1192 		hw_rev++;
1193 	}
1194 
1195 	sc->rl_eewidth = 6;
1196 	re_read_eeprom(sc, (caddr_t)&re_did, 0, 1);
1197 	if (re_did != 0x8129)
1198 	        sc->rl_eewidth = 8;
1199 
1200 	/*
1201 	 * Get station address from the EEPROM.
1202 	 */
1203 	re_read_eeprom(sc, (caddr_t)as, RL_EE_EADDR, 3);
1204 	for (i = 0; i < ETHER_ADDR_LEN / 2; i++)
1205 		as[i] = le16toh(as[i]);
1206 	bcopy(as, eaddr, sizeof(eaddr));
1207 
1208 	if (sc->rl_type == RL_8169) {
1209 		/* Set RX length mask */
1210 		sc->rl_rxlenmask = RL_RDESC_STAT_GFRAGLEN;
1211 		sc->rl_txstart = RL_GTXSTART;
1212 	} else {
1213 		/* Set RX length mask */
1214 		sc->rl_rxlenmask = RL_RDESC_STAT_FRAGLEN;
1215 		sc->rl_txstart = RL_TXSTART;
1216 	}
1217 
1218 	/*
1219 	 * Allocate the parent bus DMA tag appropriate for PCI.
1220 	 */
1221 #define RL_NSEG_NEW 32
1222 	error = bus_dma_tag_create(bus_get_dma_tag(dev), 1, 0,
1223 	    BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL, NULL,
1224 	    MAXBSIZE, RL_NSEG_NEW, BUS_SPACE_MAXSIZE_32BIT, 0,
1225 	    NULL, NULL, &sc->rl_parent_tag);
1226 	if (error)
1227 		goto fail;
1228 
1229 	error = re_allocmem(dev, sc);
1230 
1231 	if (error)
1232 		goto fail;
1233 
1234 	ifp = sc->rl_ifp = if_alloc(IFT_ETHER);
1235 	if (ifp == NULL) {
1236 		device_printf(dev, "can not if_alloc()\n");
1237 		error = ENOSPC;
1238 		goto fail;
1239 	}
1240 
1241 	/* Do MII setup */
1242 	if (mii_phy_probe(dev, &sc->rl_miibus,
1243 	    re_ifmedia_upd, re_ifmedia_sts)) {
1244 		device_printf(dev, "MII without any phy!\n");
1245 		error = ENXIO;
1246 		goto fail;
1247 	}
1248 
1249 	ifp->if_softc = sc;
1250 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1251 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1252 	ifp->if_ioctl = re_ioctl;
1253 	ifp->if_start = re_start;
1254 	ifp->if_hwassist = RE_CSUM_FEATURES | CSUM_TSO;
1255 	ifp->if_capabilities = IFCAP_HWCSUM | IFCAP_TSO4;
1256 	ifp->if_capenable = ifp->if_capabilities;
1257 	ifp->if_init = re_init;
1258 	IFQ_SET_MAXLEN(&ifp->if_snd, RL_IFQ_MAXLEN);
1259 	ifp->if_snd.ifq_drv_maxlen = RL_IFQ_MAXLEN;
1260 	IFQ_SET_READY(&ifp->if_snd);
1261 
1262 	TASK_INIT(&sc->rl_txtask, 1, re_tx_task, ifp);
1263 	TASK_INIT(&sc->rl_inttask, 0, re_int_task, sc);
1264 
1265 	/*
1266 	 * Call MI attach routine.
1267 	 */
1268 	ether_ifattach(ifp, eaddr);
1269 
1270 	/* VLAN capability setup */
1271 	ifp->if_capabilities |= IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING;
1272 	if (ifp->if_capabilities & IFCAP_HWCSUM)
1273 		ifp->if_capabilities |= IFCAP_VLAN_HWCSUM;
1274 	ifp->if_capenable = ifp->if_capabilities;
1275 #ifdef DEVICE_POLLING
1276 	ifp->if_capabilities |= IFCAP_POLLING;
1277 #endif
1278 	/*
1279 	 * Tell the upper layer(s) we support long frames.
1280 	 * Must appear after the call to ether_ifattach() because
1281 	 * ether_ifattach() sets ifi_hdrlen to the default value.
1282 	 */
1283 	ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
1284 
1285 #ifdef RE_DIAG
1286 	/*
1287 	 * Perform hardware diagnostic on the original RTL8169.
1288 	 * Some 32-bit cards were incorrectly wired and would
1289 	 * malfunction if plugged into a 64-bit slot.
1290 	 */
1291 
1292 	if (hwrev == RL_HWREV_8169) {
1293 		error = re_diag(sc);
1294 		if (error) {
1295 			device_printf(dev,
1296 		    	"attach aborted due to hardware diag failure\n");
1297 			ether_ifdetach(ifp);
1298 			goto fail;
1299 		}
1300 	}
1301 #endif
1302 
1303 	/* Hook interrupt last to avoid having to lock softc */
1304 	error = bus_setup_intr(dev, sc->rl_irq, INTR_TYPE_NET | INTR_MPSAFE |
1305 	    INTR_FAST, re_intr, sc, &sc->rl_intrhand);
1306 	if (error) {
1307 		device_printf(dev, "couldn't set up irq\n");
1308 		ether_ifdetach(ifp);
1309 	}
1310 
1311 fail:
1312 
1313 	if (error)
1314 		re_detach(dev);
1315 
1316 	return (error);
1317 }
1318 
1319 /*
1320  * Shutdown hardware and free up resources. This can be called any
1321  * time after the mutex has been initialized. It is called in both
1322  * the error case in attach and the normal detach case so it needs
1323  * to be careful about only freeing resources that have actually been
1324  * allocated.
1325  */
1326 static int
1327 re_detach(dev)
1328 	device_t		dev;
1329 {
1330 	struct rl_softc		*sc;
1331 	struct ifnet		*ifp;
1332 	int			i;
1333 
1334 	sc = device_get_softc(dev);
1335 	ifp = sc->rl_ifp;
1336 	KASSERT(mtx_initialized(&sc->rl_mtx), ("re mutex not initialized"));
1337 
1338 #ifdef DEVICE_POLLING
1339 	if (ifp->if_capenable & IFCAP_POLLING)
1340 		ether_poll_deregister(ifp);
1341 #endif
1342 	/* These should only be active if attach succeeded */
1343 	if (device_is_attached(dev)) {
1344 		RL_LOCK(sc);
1345 #if 0
1346 		sc->suspended = 1;
1347 #endif
1348 		re_stop(sc);
1349 		RL_UNLOCK(sc);
1350 		callout_drain(&sc->rl_stat_callout);
1351 		/*
1352 		 * Force off the IFF_UP flag here, in case someone
1353 		 * still had a BPF descriptor attached to this
1354 		 * interface. If they do, ether_ifdetach() will cause
1355 		 * the BPF code to try and clear the promisc mode
1356 		 * flag, which will bubble down to re_ioctl(),
1357 		 * which will try to call re_init() again. This will
1358 		 * turn the NIC back on and restart the MII ticker,
1359 		 * which will panic the system when the kernel tries
1360 		 * to invoke the re_tick() function that isn't there
1361 		 * anymore.
1362 		 */
1363 		ifp->if_flags &= ~IFF_UP;
1364 		ether_ifdetach(ifp);
1365 	}
1366 	if (sc->rl_miibus)
1367 		device_delete_child(dev, sc->rl_miibus);
1368 	bus_generic_detach(dev);
1369 
1370 	/*
1371 	 * The rest is resource deallocation, so we should already be
1372 	 * stopped here.
1373 	 */
1374 
1375 	if (sc->rl_intrhand)
1376 		bus_teardown_intr(dev, sc->rl_irq, sc->rl_intrhand);
1377 	if (ifp != NULL)
1378 		if_free(ifp);
1379 	if (sc->rl_irq)
1380 		bus_release_resource(dev, SYS_RES_IRQ, 0, sc->rl_irq);
1381 	if (sc->rl_res)
1382 		bus_release_resource(dev, RL_RES, RL_RID, sc->rl_res);
1383 
1384 	/* Yield the CPU long enough for any tasks to drain */
1385 
1386         tsleep(sc, PPAUSE, "rewait", hz);
1387 
1388 	/* Unload and free the RX DMA ring memory and map */
1389 
1390 	if (sc->rl_ldata.rl_rx_list_tag) {
1391 		bus_dmamap_unload(sc->rl_ldata.rl_rx_list_tag,
1392 		    sc->rl_ldata.rl_rx_list_map);
1393 		bus_dmamem_free(sc->rl_ldata.rl_rx_list_tag,
1394 		    sc->rl_ldata.rl_rx_list,
1395 		    sc->rl_ldata.rl_rx_list_map);
1396 		bus_dma_tag_destroy(sc->rl_ldata.rl_rx_list_tag);
1397 	}
1398 
1399 	/* Unload and free the TX DMA ring memory and map */
1400 
1401 	if (sc->rl_ldata.rl_tx_list_tag) {
1402 		bus_dmamap_unload(sc->rl_ldata.rl_tx_list_tag,
1403 		    sc->rl_ldata.rl_tx_list_map);
1404 		bus_dmamem_free(sc->rl_ldata.rl_tx_list_tag,
1405 		    sc->rl_ldata.rl_tx_list,
1406 		    sc->rl_ldata.rl_tx_list_map);
1407 		bus_dma_tag_destroy(sc->rl_ldata.rl_tx_list_tag);
1408 	}
1409 
1410 	/* Destroy all the RX and TX buffer maps */
1411 
1412 	if (sc->rl_ldata.rl_mtag) {
1413 		for (i = 0; i < RL_TX_DESC_CNT; i++)
1414 			bus_dmamap_destroy(sc->rl_ldata.rl_mtag,
1415 			    sc->rl_ldata.rl_tx_dmamap[i]);
1416 		for (i = 0; i < RL_RX_DESC_CNT; i++)
1417 			bus_dmamap_destroy(sc->rl_ldata.rl_mtag,
1418 			    sc->rl_ldata.rl_rx_dmamap[i]);
1419 		bus_dma_tag_destroy(sc->rl_ldata.rl_mtag);
1420 	}
1421 
1422 	/* Unload and free the stats buffer and map */
1423 
1424 	if (sc->rl_ldata.rl_stag) {
1425 		bus_dmamap_unload(sc->rl_ldata.rl_stag,
1426 		    sc->rl_ldata.rl_rx_list_map);
1427 		bus_dmamem_free(sc->rl_ldata.rl_stag,
1428 		    sc->rl_ldata.rl_stats,
1429 		    sc->rl_ldata.rl_smap);
1430 		bus_dma_tag_destroy(sc->rl_ldata.rl_stag);
1431 	}
1432 
1433 	if (sc->rl_parent_tag)
1434 		bus_dma_tag_destroy(sc->rl_parent_tag);
1435 
1436 	mtx_destroy(&sc->rl_mtx);
1437 
1438 	return (0);
1439 }
1440 
1441 static int
1442 re_newbuf(sc, idx, m)
1443 	struct rl_softc		*sc;
1444 	int			idx;
1445 	struct mbuf		*m;
1446 {
1447 	struct rl_dmaload_arg	arg;
1448 	struct mbuf		*n = NULL;
1449 	int			error;
1450 
1451 	if (m == NULL) {
1452 		n = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
1453 		if (n == NULL)
1454 			return (ENOBUFS);
1455 		m = n;
1456 	} else
1457 		m->m_data = m->m_ext.ext_buf;
1458 
1459 	m->m_len = m->m_pkthdr.len = MCLBYTES;
1460 #ifdef RE_FIXUP_RX
1461 	/*
1462 	 * This is part of an evil trick to deal with non-x86 platforms.
1463 	 * The RealTek chip requires RX buffers to be aligned on 64-bit
1464 	 * boundaries, but that will hose non-x86 machines. To get around
1465 	 * this, we leave some empty space at the start of each buffer
1466 	 * and for non-x86 hosts, we copy the buffer back six bytes
1467 	 * to achieve word alignment. This is slightly more efficient
1468 	 * than allocating a new buffer, copying the contents, and
1469 	 * discarding the old buffer.
1470 	 */
1471 	m_adj(m, RE_ETHER_ALIGN);
1472 #endif
1473 	arg.sc = sc;
1474 	arg.rl_idx = idx;
1475 	arg.rl_maxsegs = 1;
1476 	arg.rl_flags = 0;
1477 	arg.rl_ring = sc->rl_ldata.rl_rx_list;
1478 
1479 	error = bus_dmamap_load_mbuf(sc->rl_ldata.rl_mtag,
1480 	    sc->rl_ldata.rl_rx_dmamap[idx], m, re_dma_map_desc,
1481 	    &arg, BUS_DMA_NOWAIT);
1482 	if (error || arg.rl_maxsegs != 1) {
1483 		if (n != NULL)
1484 			m_freem(n);
1485 		return (ENOMEM);
1486 	}
1487 
1488 	sc->rl_ldata.rl_rx_list[idx].rl_cmdstat |= htole32(RL_RDESC_CMD_OWN);
1489 	sc->rl_ldata.rl_rx_mbuf[idx] = m;
1490 
1491 	bus_dmamap_sync(sc->rl_ldata.rl_mtag,
1492 	    sc->rl_ldata.rl_rx_dmamap[idx],
1493 	    BUS_DMASYNC_PREREAD);
1494 
1495 	return (0);
1496 }
1497 
1498 #ifdef RE_FIXUP_RX
1499 static __inline void
1500 re_fixup_rx(m)
1501 	struct mbuf		*m;
1502 {
1503 	int                     i;
1504 	uint16_t                *src, *dst;
1505 
1506 	src = mtod(m, uint16_t *);
1507 	dst = src - (RE_ETHER_ALIGN - ETHER_ALIGN) / sizeof *src;
1508 
1509 	for (i = 0; i < (m->m_len / sizeof(uint16_t) + 1); i++)
1510 		*dst++ = *src++;
1511 
1512 	m->m_data -= RE_ETHER_ALIGN - ETHER_ALIGN;
1513 
1514 	return;
1515 }
1516 #endif
1517 
1518 static int
1519 re_tx_list_init(sc)
1520 	struct rl_softc		*sc;
1521 {
1522 
1523 	RL_LOCK_ASSERT(sc);
1524 
1525 	bzero ((char *)sc->rl_ldata.rl_tx_list, RL_TX_LIST_SZ);
1526 	bzero ((char *)&sc->rl_ldata.rl_tx_mbuf,
1527 	    (RL_TX_DESC_CNT * sizeof(struct mbuf *)));
1528 
1529 	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
1530 	    sc->rl_ldata.rl_tx_list_map, BUS_DMASYNC_PREWRITE);
1531 	sc->rl_ldata.rl_tx_prodidx = 0;
1532 	sc->rl_ldata.rl_tx_considx = 0;
1533 	sc->rl_ldata.rl_tx_free = RL_TX_DESC_CNT;
1534 
1535 	return (0);
1536 }
1537 
1538 static int
1539 re_rx_list_init(sc)
1540 	struct rl_softc		*sc;
1541 {
1542 	int			i;
1543 
1544 	bzero ((char *)sc->rl_ldata.rl_rx_list, RL_RX_LIST_SZ);
1545 	bzero ((char *)&sc->rl_ldata.rl_rx_mbuf,
1546 	    (RL_RX_DESC_CNT * sizeof(struct mbuf *)));
1547 
1548 	for (i = 0; i < RL_RX_DESC_CNT; i++) {
1549 		if (re_newbuf(sc, i, NULL) == ENOBUFS)
1550 			return (ENOBUFS);
1551 	}
1552 
1553 	/* Flush the RX descriptors */
1554 
1555 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1556 	    sc->rl_ldata.rl_rx_list_map,
1557 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
1558 
1559 	sc->rl_ldata.rl_rx_prodidx = 0;
1560 	sc->rl_head = sc->rl_tail = NULL;
1561 
1562 	return (0);
1563 }
1564 
1565 /*
1566  * RX handler for C+ and 8169. For the gigE chips, we support
1567  * the reception of jumbo frames that have been fragmented
1568  * across multiple 2K mbuf cluster buffers.
1569  */
1570 static int
1571 re_rxeof(sc)
1572 	struct rl_softc		*sc;
1573 {
1574 	struct mbuf		*m;
1575 	struct ifnet		*ifp;
1576 	int			i, total_len;
1577 	struct rl_desc		*cur_rx;
1578 	u_int32_t		rxstat, rxvlan;
1579 	int			maxpkt = 16;
1580 
1581 	RL_LOCK_ASSERT(sc);
1582 
1583 	ifp = sc->rl_ifp;
1584 	i = sc->rl_ldata.rl_rx_prodidx;
1585 
1586 	/* Invalidate the descriptor memory */
1587 
1588 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1589 	    sc->rl_ldata.rl_rx_list_map,
1590 	    BUS_DMASYNC_POSTREAD);
1591 
1592 	while (!RL_OWN(&sc->rl_ldata.rl_rx_list[i]) && maxpkt) {
1593 		cur_rx = &sc->rl_ldata.rl_rx_list[i];
1594 		m = sc->rl_ldata.rl_rx_mbuf[i];
1595 		total_len = RL_RXBYTES(cur_rx);
1596 		rxstat = le32toh(cur_rx->rl_cmdstat);
1597 		rxvlan = le32toh(cur_rx->rl_vlanctl);
1598 
1599 		/* Invalidate the RX mbuf and unload its map */
1600 
1601 		bus_dmamap_sync(sc->rl_ldata.rl_mtag,
1602 		    sc->rl_ldata.rl_rx_dmamap[i],
1603 		    BUS_DMASYNC_POSTWRITE);
1604 		bus_dmamap_unload(sc->rl_ldata.rl_mtag,
1605 		    sc->rl_ldata.rl_rx_dmamap[i]);
1606 
1607 		if (!(rxstat & RL_RDESC_STAT_EOF)) {
1608 			m->m_len = RE_RX_DESC_BUFLEN;
1609 			if (sc->rl_head == NULL)
1610 				sc->rl_head = sc->rl_tail = m;
1611 			else {
1612 				m->m_flags &= ~M_PKTHDR;
1613 				sc->rl_tail->m_next = m;
1614 				sc->rl_tail = m;
1615 			}
1616 			re_newbuf(sc, i, NULL);
1617 			RL_DESC_INC(i);
1618 			continue;
1619 		}
1620 
1621 		/*
1622 		 * NOTE: for the 8139C+, the frame length field
1623 		 * is always 12 bits in size, but for the gigE chips,
1624 		 * it is 13 bits (since the max RX frame length is 16K).
1625 		 * Unfortunately, all 32 bits in the status word
1626 		 * were already used, so to make room for the extra
1627 		 * length bit, RealTek took out the 'frame alignment
1628 		 * error' bit and shifted the other status bits
1629 		 * over one slot. The OWN, EOR, FS and LS bits are
1630 		 * still in the same places. We have already extracted
1631 		 * the frame length and checked the OWN bit, so rather
1632 		 * than using an alternate bit mapping, we shift the
1633 		 * status bits one space to the right so we can evaluate
1634 		 * them using the 8169 status as though it was in the
1635 		 * same format as that of the 8139C+.
1636 		 */
1637 		if (sc->rl_type == RL_8169)
1638 			rxstat >>= 1;
1639 
1640 		/*
1641 		 * if total_len > 2^13-1, both _RXERRSUM and _GIANT will be
1642 		 * set, but if CRC is clear, it will still be a valid frame.
1643 		 */
1644 		if (rxstat & RL_RDESC_STAT_RXERRSUM && !(total_len > 8191 &&
1645 		    (rxstat & RL_RDESC_STAT_ERRS) == RL_RDESC_STAT_GIANT)) {
1646 			ifp->if_ierrors++;
1647 			/*
1648 			 * If this is part of a multi-fragment packet,
1649 			 * discard all the pieces.
1650 			 */
1651 			if (sc->rl_head != NULL) {
1652 				m_freem(sc->rl_head);
1653 				sc->rl_head = sc->rl_tail = NULL;
1654 			}
1655 			re_newbuf(sc, i, m);
1656 			RL_DESC_INC(i);
1657 			continue;
1658 		}
1659 
1660 		/*
1661 		 * If allocating a replacement mbuf fails,
1662 		 * reload the current one.
1663 		 */
1664 
1665 		if (re_newbuf(sc, i, NULL)) {
1666 			ifp->if_ierrors++;
1667 			if (sc->rl_head != NULL) {
1668 				m_freem(sc->rl_head);
1669 				sc->rl_head = sc->rl_tail = NULL;
1670 			}
1671 			re_newbuf(sc, i, m);
1672 			RL_DESC_INC(i);
1673 			continue;
1674 		}
1675 
1676 		RL_DESC_INC(i);
1677 
1678 		if (sc->rl_head != NULL) {
1679 			m->m_len = total_len % RE_RX_DESC_BUFLEN;
1680 			if (m->m_len == 0)
1681 				m->m_len = RE_RX_DESC_BUFLEN;
1682 			/*
1683 			 * Special case: if there's 4 bytes or less
1684 			 * in this buffer, the mbuf can be discarded:
1685 			 * the last 4 bytes is the CRC, which we don't
1686 			 * care about anyway.
1687 			 */
1688 			if (m->m_len <= ETHER_CRC_LEN) {
1689 				sc->rl_tail->m_len -=
1690 				    (ETHER_CRC_LEN - m->m_len);
1691 				m_freem(m);
1692 			} else {
1693 				m->m_len -= ETHER_CRC_LEN;
1694 				m->m_flags &= ~M_PKTHDR;
1695 				sc->rl_tail->m_next = m;
1696 			}
1697 			m = sc->rl_head;
1698 			sc->rl_head = sc->rl_tail = NULL;
1699 			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
1700 		} else
1701 			m->m_pkthdr.len = m->m_len =
1702 			    (total_len - ETHER_CRC_LEN);
1703 
1704 #ifdef RE_FIXUP_RX
1705 		re_fixup_rx(m);
1706 #endif
1707 		ifp->if_ipackets++;
1708 		m->m_pkthdr.rcvif = ifp;
1709 
1710 		/* Do RX checksumming if enabled */
1711 
1712 		if (ifp->if_capenable & IFCAP_RXCSUM) {
1713 
1714 			/* Check IP header checksum */
1715 			if (rxstat & RL_RDESC_STAT_PROTOID)
1716 				m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED;
1717 			if (!(rxstat & RL_RDESC_STAT_IPSUMBAD))
1718 				m->m_pkthdr.csum_flags |= CSUM_IP_VALID;
1719 
1720 			/* Check TCP/UDP checksum */
1721 			if ((RL_TCPPKT(rxstat) &&
1722 			    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
1723 			    (RL_UDPPKT(rxstat) &&
1724 			    !(rxstat & RL_RDESC_STAT_UDPSUMBAD))) {
1725 				m->m_pkthdr.csum_flags |=
1726 				    CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
1727 				m->m_pkthdr.csum_data = 0xffff;
1728 			}
1729 		}
1730 		maxpkt--;
1731 		if (rxvlan & RL_RDESC_VLANCTL_TAG) {
1732 			m->m_pkthdr.ether_vtag =
1733 			    ntohs((rxvlan & RL_RDESC_VLANCTL_DATA));
1734 			m->m_flags |= M_VLANTAG;
1735 		}
1736 		RL_UNLOCK(sc);
1737 		(*ifp->if_input)(ifp, m);
1738 		RL_LOCK(sc);
1739 	}
1740 
1741 	/* Flush the RX DMA ring */
1742 
1743 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1744 	    sc->rl_ldata.rl_rx_list_map,
1745 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
1746 
1747 	sc->rl_ldata.rl_rx_prodidx = i;
1748 
1749 	if (maxpkt)
1750 		return(EAGAIN);
1751 
1752 	return(0);
1753 }
1754 
1755 static void
1756 re_txeof(sc)
1757 	struct rl_softc		*sc;
1758 {
1759 	struct ifnet		*ifp;
1760 	u_int32_t		txstat;
1761 	int			idx;
1762 
1763 	ifp = sc->rl_ifp;
1764 	idx = sc->rl_ldata.rl_tx_considx;
1765 
1766 	/* Invalidate the TX descriptor list */
1767 
1768 	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
1769 	    sc->rl_ldata.rl_tx_list_map,
1770 	    BUS_DMASYNC_POSTREAD);
1771 
1772 	while (sc->rl_ldata.rl_tx_free < RL_TX_DESC_CNT) {
1773 
1774 		txstat = le32toh(sc->rl_ldata.rl_tx_list[idx].rl_cmdstat);
1775 		if (txstat & RL_TDESC_CMD_OWN)
1776 			break;
1777 
1778 		sc->rl_ldata.rl_tx_list[idx].rl_bufaddr_lo = 0;
1779 
1780 		/*
1781 		 * We only stash mbufs in the last descriptor
1782 		 * in a fragment chain, which also happens to
1783 		 * be the only place where the TX status bits
1784 		 * are valid.
1785 		 */
1786 
1787 		if (txstat & RL_TDESC_CMD_EOF) {
1788 			m_freem(sc->rl_ldata.rl_tx_mbuf[idx]);
1789 			sc->rl_ldata.rl_tx_mbuf[idx] = NULL;
1790 			bus_dmamap_unload(sc->rl_ldata.rl_mtag,
1791 			    sc->rl_ldata.rl_tx_dmamap[idx]);
1792 			if (txstat & (RL_TDESC_STAT_EXCESSCOL|
1793 			    RL_TDESC_STAT_COLCNT))
1794 				ifp->if_collisions++;
1795 			if (txstat & RL_TDESC_STAT_TXERRSUM)
1796 				ifp->if_oerrors++;
1797 			else
1798 				ifp->if_opackets++;
1799 		}
1800 		sc->rl_ldata.rl_tx_free++;
1801 		RL_DESC_INC(idx);
1802 	}
1803 
1804 	/* No changes made to the TX ring, so no flush needed */
1805 
1806 	if (sc->rl_ldata.rl_tx_free) {
1807 		sc->rl_ldata.rl_tx_considx = idx;
1808 		ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1809 		sc->rl_watchdog_timer = 0;
1810 	}
1811 
1812 	/*
1813 	 * Some chips will ignore a second TX request issued while an
1814 	 * existing transmission is in progress. If the transmitter goes
1815 	 * idle but there are still packets waiting to be sent, we need
1816 	 * to restart the channel here to flush them out. This only seems
1817 	 * to be required with the PCIe devices.
1818 	 */
1819 
1820 	if (sc->rl_ldata.rl_tx_free < RL_TX_DESC_CNT)
1821 	    CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
1822 
1823 #ifdef RE_TX_MODERATION
1824 	/*
1825 	 * If not all descriptors have been released reaped yet,
1826 	 * reload the timer so that we will eventually get another
1827 	 * interrupt that will cause us to re-enter this routine.
1828 	 * This is done in case the transmitter has gone idle.
1829 	 */
1830 	if (sc->rl_ldata.rl_tx_free != RL_TX_DESC_CNT)
1831 		CSR_WRITE_4(sc, RL_TIMERCNT, 1);
1832 #endif
1833 
1834 }
1835 
1836 static void
1837 re_tick(xsc)
1838 	void			*xsc;
1839 {
1840 	struct rl_softc		*sc;
1841 	struct mii_data		*mii;
1842 	struct ifnet		*ifp;
1843 
1844 	sc = xsc;
1845 	ifp = sc->rl_ifp;
1846 
1847 	RL_LOCK_ASSERT(sc);
1848 
1849 	re_watchdog(sc);
1850 
1851 	mii = device_get_softc(sc->rl_miibus);
1852 	mii_tick(mii);
1853 	if (sc->rl_link) {
1854 		if (!(mii->mii_media_status & IFM_ACTIVE))
1855 			sc->rl_link = 0;
1856 	} else {
1857 		if (mii->mii_media_status & IFM_ACTIVE &&
1858 		    IFM_SUBTYPE(mii->mii_media_active) != IFM_NONE) {
1859 			sc->rl_link = 1;
1860 			if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
1861 				taskqueue_enqueue_fast(taskqueue_fast,
1862 				    &sc->rl_txtask);
1863 		}
1864 	}
1865 
1866 	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
1867 }
1868 
1869 #ifdef DEVICE_POLLING
1870 static void
1871 re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
1872 {
1873 	struct rl_softc *sc = ifp->if_softc;
1874 
1875 	RL_LOCK(sc);
1876 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1877 		re_poll_locked(ifp, cmd, count);
1878 	RL_UNLOCK(sc);
1879 }
1880 
1881 static void
1882 re_poll_locked(struct ifnet *ifp, enum poll_cmd cmd, int count)
1883 {
1884 	struct rl_softc *sc = ifp->if_softc;
1885 
1886 	RL_LOCK_ASSERT(sc);
1887 
1888 	sc->rxcycles = count;
1889 	re_rxeof(sc);
1890 	re_txeof(sc);
1891 
1892 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
1893 		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_txtask);
1894 
1895 	if (cmd == POLL_AND_CHECK_STATUS) { /* also check status register */
1896 		u_int16_t       status;
1897 
1898 		status = CSR_READ_2(sc, RL_ISR);
1899 		if (status == 0xffff)
1900 			return;
1901 		if (status)
1902 			CSR_WRITE_2(sc, RL_ISR, status);
1903 
1904 		/*
1905 		 * XXX check behaviour on receiver stalls.
1906 		 */
1907 
1908 		if (status & RL_ISR_SYSTEM_ERR) {
1909 			re_reset(sc);
1910 			re_init_locked(sc);
1911 		}
1912 	}
1913 }
1914 #endif /* DEVICE_POLLING */
1915 
1916 static void
1917 re_intr(arg)
1918 	void			*arg;
1919 {
1920 	struct rl_softc		*sc;
1921 	uint16_t		status;
1922 
1923 	sc = arg;
1924 
1925 	status = CSR_READ_2(sc, RL_ISR);
1926 	if (status == 0xFFFF || (status & RL_INTRS_CPLUS) == 0)
1927                 return;
1928 	CSR_WRITE_2(sc, RL_IMR, 0);
1929 
1930 	taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_inttask);
1931 
1932 	return;
1933 }
1934 
1935 static void
1936 re_int_task(arg, npending)
1937 	void			*arg;
1938 	int			npending;
1939 {
1940 	struct rl_softc		*sc;
1941 	struct ifnet		*ifp;
1942 	u_int16_t		status;
1943 	int			rval = 0;
1944 
1945 	sc = arg;
1946 	ifp = sc->rl_ifp;
1947 
1948 	RL_LOCK(sc);
1949 
1950 	status = CSR_READ_2(sc, RL_ISR);
1951         CSR_WRITE_2(sc, RL_ISR, status);
1952 
1953 	if (sc->suspended || !(ifp->if_flags & IFF_UP)) {
1954 		RL_UNLOCK(sc);
1955 		return;
1956 	}
1957 
1958 #ifdef DEVICE_POLLING
1959 	if  (ifp->if_capenable & IFCAP_POLLING) {
1960 		RL_UNLOCK(sc);
1961 		return;
1962 	}
1963 #endif
1964 
1965 	if (status & (RL_ISR_RX_OK|RL_ISR_RX_ERR|RL_ISR_FIFO_OFLOW))
1966 		rval = re_rxeof(sc);
1967 
1968 #ifdef RE_TX_MODERATION
1969 	if (status & (RL_ISR_TIMEOUT_EXPIRED|
1970 #else
1971 	if (status & (RL_ISR_TX_OK|
1972 #endif
1973 	    RL_ISR_TX_ERR|RL_ISR_TX_DESC_UNAVAIL))
1974 		re_txeof(sc);
1975 
1976 	if (status & RL_ISR_SYSTEM_ERR) {
1977 		re_reset(sc);
1978 		re_init_locked(sc);
1979 	}
1980 
1981 	if (status & RL_ISR_LINKCHG) {
1982 		callout_stop(&sc->rl_stat_callout);
1983 		re_tick(sc);
1984 	}
1985 
1986 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
1987 		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_txtask);
1988 
1989 	RL_UNLOCK(sc);
1990 
1991         if ((CSR_READ_2(sc, RL_ISR) & RL_INTRS_CPLUS) || rval) {
1992 		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_inttask);
1993 		return;
1994 	}
1995 
1996 	CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
1997 
1998 	return;
1999 }
2000 
2001 static int
2002 re_encap(sc, m_head, idx)
2003 	struct rl_softc		*sc;
2004 	struct mbuf		**m_head;
2005 	int			*idx;
2006 {
2007 	struct mbuf		*m_new = NULL;
2008 	struct rl_dmaload_arg	arg;
2009 	bus_dmamap_t		map;
2010 	int			error;
2011 
2012 	RL_LOCK_ASSERT(sc);
2013 
2014 	if (sc->rl_ldata.rl_tx_free <= 4)
2015 		return (EFBIG);
2016 
2017 	/*
2018 	 * Set up checksum offload. Note: checksum offload bits must
2019 	 * appear in all descriptors of a multi-descriptor transmit
2020 	 * attempt. This is according to testing done with an 8169
2021 	 * chip. This is a requirement.
2022 	 */
2023 
2024 	arg.rl_flags = 0;
2025 
2026 	if (((*m_head)->m_pkthdr.csum_flags & CSUM_TSO) != 0)
2027 		arg.rl_flags = RL_TDESC_CMD_LGSEND |
2028 		    ((uint32_t)(*m_head)->m_pkthdr.tso_segsz <<
2029 		    RL_TDESC_CMD_MSSVAL_SHIFT);
2030 	else {
2031 		if ((*m_head)->m_pkthdr.csum_flags & CSUM_IP)
2032 			arg.rl_flags |= RL_TDESC_CMD_IPCSUM;
2033 		if ((*m_head)->m_pkthdr.csum_flags & CSUM_TCP)
2034 			arg.rl_flags |= RL_TDESC_CMD_TCPCSUM;
2035 		if ((*m_head)->m_pkthdr.csum_flags & CSUM_UDP)
2036 			arg.rl_flags |= RL_TDESC_CMD_UDPCSUM;
2037 	}
2038 
2039 	arg.sc = sc;
2040 	arg.rl_idx = *idx;
2041 	arg.rl_maxsegs = sc->rl_ldata.rl_tx_free;
2042 	if (arg.rl_maxsegs > 4)
2043 		arg.rl_maxsegs -= 4;
2044 	arg.rl_ring = sc->rl_ldata.rl_tx_list;
2045 
2046 	map = sc->rl_ldata.rl_tx_dmamap[*idx];
2047 
2048 	/*
2049 	 * With some of the RealTek chips, using the checksum offload
2050 	 * support in conjunction with the autopadding feature results
2051 	 * in the transmission of corrupt frames. For example, if we
2052 	 * need to send a really small IP fragment that's less than 60
2053 	 * bytes in size, and IP header checksumming is enabled, the
2054 	 * resulting ethernet frame that appears on the wire will
2055 	 * have garbled payload. To work around this, if TX checksum
2056 	 * offload is enabled, we always manually pad short frames out
2057 	 * to the minimum ethernet frame size. We do this by pretending
2058 	 * the mbuf chain has too many fragments so the coalescing code
2059 	 * below can assemble the packet into a single buffer that's
2060 	 * padded out to the mininum frame size.
2061 	 */
2062 
2063 	if (arg.rl_flags && (*m_head)->m_pkthdr.len < RL_MIN_FRAMELEN)
2064 		error = EFBIG;
2065 	else
2066 		error = bus_dmamap_load_mbuf(sc->rl_ldata.rl_mtag, map,
2067 		    *m_head, re_dma_map_desc, &arg, BUS_DMA_NOWAIT);
2068 
2069 	if (error && error != EFBIG) {
2070 		device_printf(sc->rl_dev, "can't map mbuf (error %d)\n", error);
2071 		return (ENOBUFS);
2072 	}
2073 
2074 	/* Too many segments to map, coalesce into a single mbuf */
2075 
2076 	if (error || arg.rl_maxsegs == 0) {
2077 		m_new = m_defrag(*m_head, M_DONTWAIT);
2078 		if (m_new == NULL)
2079 			return (ENOBUFS);
2080 		else
2081 			*m_head = m_new;
2082 
2083 		/*
2084 		 * Manually pad short frames, and zero the pad space
2085 		 * to avoid leaking data.
2086 		 */
2087 
2088 		if (m_new->m_pkthdr.len < RL_MIN_FRAMELEN) {
2089 			bzero(mtod(m_new, char *) + m_new->m_pkthdr.len,
2090 			    RL_MIN_FRAMELEN - m_new->m_pkthdr.len);
2091 			m_new->m_pkthdr.len += RL_MIN_FRAMELEN -
2092 			    m_new->m_pkthdr.len;
2093 			m_new->m_len = m_new->m_pkthdr.len;
2094 		}
2095 
2096 		arg.sc = sc;
2097 		arg.rl_idx = *idx;
2098 		arg.rl_maxsegs = sc->rl_ldata.rl_tx_free;
2099 		arg.rl_ring = sc->rl_ldata.rl_tx_list;
2100 
2101 		error = bus_dmamap_load_mbuf(sc->rl_ldata.rl_mtag, map,
2102 		    *m_head, re_dma_map_desc, &arg, BUS_DMA_NOWAIT);
2103 		if (error) {
2104 			device_printf(sc->rl_dev, "can't map mbuf (error %d)\n",
2105 			    error);
2106 			return (EFBIG);
2107 		}
2108 	}
2109 
2110 	/*
2111 	 * Insure that the map for this transmission
2112 	 * is placed at the array index of the last descriptor
2113 	 * in this chain.  (Swap last and first dmamaps.)
2114 	 */
2115 	sc->rl_ldata.rl_tx_dmamap[*idx] =
2116 	    sc->rl_ldata.rl_tx_dmamap[arg.rl_idx];
2117 	sc->rl_ldata.rl_tx_dmamap[arg.rl_idx] = map;
2118 
2119 	sc->rl_ldata.rl_tx_mbuf[arg.rl_idx] = *m_head;
2120 	sc->rl_ldata.rl_tx_free -= arg.rl_maxsegs;
2121 
2122 	/*
2123 	 * Set up hardware VLAN tagging. Note: vlan tag info must
2124 	 * appear in the first descriptor of a multi-descriptor
2125 	 * transmission attempt.
2126 	 */
2127 	if ((*m_head)->m_flags & M_VLANTAG)
2128 		sc->rl_ldata.rl_tx_list[*idx].rl_vlanctl =
2129 		    htole32(htons((*m_head)->m_pkthdr.ether_vtag) |
2130 		    RL_TDESC_VLANCTL_TAG);
2131 
2132 	/* Transfer ownership of packet to the chip. */
2133 
2134 	sc->rl_ldata.rl_tx_list[arg.rl_idx].rl_cmdstat |=
2135 	    htole32(RL_TDESC_CMD_OWN);
2136 	if (*idx != arg.rl_idx)
2137 		sc->rl_ldata.rl_tx_list[*idx].rl_cmdstat |=
2138 		    htole32(RL_TDESC_CMD_OWN);
2139 
2140         RL_DESC_INC(arg.rl_idx);
2141 	*idx = arg.rl_idx;
2142 
2143 	return (0);
2144 }
2145 
2146 static void
2147 re_tx_task(arg, npending)
2148 	void			*arg;
2149 	int			npending;
2150 {
2151 	struct ifnet		*ifp;
2152 
2153 	ifp = arg;
2154 	re_start(ifp);
2155 
2156 	return;
2157 }
2158 
2159 /*
2160  * Main transmit routine for C+ and gigE NICs.
2161  */
2162 static void
2163 re_start(ifp)
2164 	struct ifnet		*ifp;
2165 {
2166 	struct rl_softc		*sc;
2167 	struct mbuf		*m_head = NULL;
2168 	int			idx, queued = 0;
2169 
2170 	sc = ifp->if_softc;
2171 
2172 	RL_LOCK(sc);
2173 
2174 	if (!sc->rl_link || ifp->if_drv_flags & IFF_DRV_OACTIVE) {
2175 		RL_UNLOCK(sc);
2176 		return;
2177 	}
2178 
2179 	idx = sc->rl_ldata.rl_tx_prodidx;
2180 
2181 	while (sc->rl_ldata.rl_tx_mbuf[idx] == NULL) {
2182 		IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
2183 		if (m_head == NULL)
2184 			break;
2185 
2186 		if (re_encap(sc, &m_head, &idx)) {
2187 			IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
2188 			ifp->if_drv_flags |= IFF_DRV_OACTIVE;
2189 			break;
2190 		}
2191 
2192 		/*
2193 		 * If there's a BPF listener, bounce a copy of this frame
2194 		 * to him.
2195 		 */
2196 		BPF_MTAP(ifp, m_head);
2197 
2198 		queued++;
2199 	}
2200 
2201 	if (queued == 0) {
2202 #ifdef RE_TX_MODERATION
2203 		if (sc->rl_ldata.rl_tx_free != RL_TX_DESC_CNT)
2204 			CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2205 #endif
2206 		RL_UNLOCK(sc);
2207 		return;
2208 	}
2209 
2210 	/* Flush the TX descriptors */
2211 
2212 	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2213 	    sc->rl_ldata.rl_tx_list_map,
2214 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2215 
2216 	sc->rl_ldata.rl_tx_prodidx = idx;
2217 
2218 	/*
2219 	 * RealTek put the TX poll request register in a different
2220 	 * location on the 8169 gigE chip. I don't know why.
2221 	 */
2222 
2223 	CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2224 
2225 #ifdef RE_TX_MODERATION
2226 	/*
2227 	 * Use the countdown timer for interrupt moderation.
2228 	 * 'TX done' interrupts are disabled. Instead, we reset the
2229 	 * countdown timer, which will begin counting until it hits
2230 	 * the value in the TIMERINT register, and then trigger an
2231 	 * interrupt. Each time we write to the TIMERCNT register,
2232 	 * the timer count is reset to 0.
2233 	 */
2234 	CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2235 #endif
2236 
2237 	/*
2238 	 * Set a timeout in case the chip goes out to lunch.
2239 	 */
2240 	sc->rl_watchdog_timer = 5;
2241 
2242 	RL_UNLOCK(sc);
2243 
2244 	return;
2245 }
2246 
2247 static void
2248 re_init(xsc)
2249 	void			*xsc;
2250 {
2251 	struct rl_softc		*sc = xsc;
2252 
2253 	RL_LOCK(sc);
2254 	re_init_locked(sc);
2255 	RL_UNLOCK(sc);
2256 }
2257 
2258 static void
2259 re_init_locked(sc)
2260 	struct rl_softc		*sc;
2261 {
2262 	struct ifnet		*ifp = sc->rl_ifp;
2263 	struct mii_data		*mii;
2264 	u_int32_t		rxcfg = 0;
2265 	union {
2266 		uint32_t align_dummy;
2267 		u_char eaddr[ETHER_ADDR_LEN];
2268         } eaddr;
2269 
2270 	RL_LOCK_ASSERT(sc);
2271 
2272 	mii = device_get_softc(sc->rl_miibus);
2273 
2274 	/*
2275 	 * Cancel pending I/O and free all RX/TX buffers.
2276 	 */
2277 	re_stop(sc);
2278 
2279 	/*
2280 	 * Enable C+ RX and TX mode, as well as VLAN stripping and
2281 	 * RX checksum offload. We must configure the C+ register
2282 	 * before all others.
2283 	 */
2284 	CSR_WRITE_2(sc, RL_CPLUS_CMD, RL_CPLUSCMD_RXENB|
2285 	    RL_CPLUSCMD_TXENB|RL_CPLUSCMD_PCI_MRW|
2286 	    RL_CPLUSCMD_VLANSTRIP|RL_CPLUSCMD_RXCSUM_ENB);
2287 
2288 	/*
2289 	 * Init our MAC address.  Even though the chipset
2290 	 * documentation doesn't mention it, we need to enter "Config
2291 	 * register write enable" mode to modify the ID registers.
2292 	 */
2293 	/* Copy MAC address on stack to align. */
2294 	bcopy(IF_LLADDR(ifp), eaddr.eaddr, ETHER_ADDR_LEN);
2295 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG);
2296 	CSR_WRITE_4(sc, RL_IDR0,
2297 	    htole32(*(u_int32_t *)(&eaddr.eaddr[0])));
2298 	CSR_WRITE_4(sc, RL_IDR4,
2299 	    htole32(*(u_int32_t *)(&eaddr.eaddr[4])));
2300 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
2301 
2302 	/*
2303 	 * For C+ mode, initialize the RX descriptors and mbufs.
2304 	 */
2305 	re_rx_list_init(sc);
2306 	re_tx_list_init(sc);
2307 
2308 	/*
2309 	 * Enable transmit and receive.
2310 	 */
2311 	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
2312 
2313 	/*
2314 	 * Set the initial TX and RX configuration.
2315 	 */
2316 	if (sc->rl_testmode) {
2317 		if (sc->rl_type == RL_8169)
2318 			CSR_WRITE_4(sc, RL_TXCFG,
2319 			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON);
2320 		else
2321 			CSR_WRITE_4(sc, RL_TXCFG,
2322 			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON_CPLUS);
2323 	} else
2324 		CSR_WRITE_4(sc, RL_TXCFG, RL_TXCFG_CONFIG);
2325 	CSR_WRITE_4(sc, RL_RXCFG, RL_RXCFG_CONFIG);
2326 
2327 	/* Set the individual bit to receive frames for this host only. */
2328 	rxcfg = CSR_READ_4(sc, RL_RXCFG);
2329 	rxcfg |= RL_RXCFG_RX_INDIV;
2330 
2331 	/* If we want promiscuous mode, set the allframes bit. */
2332 	if (ifp->if_flags & IFF_PROMISC)
2333 		rxcfg |= RL_RXCFG_RX_ALLPHYS;
2334 	else
2335 		rxcfg &= ~RL_RXCFG_RX_ALLPHYS;
2336 	CSR_WRITE_4(sc, RL_RXCFG, rxcfg);
2337 
2338 	/*
2339 	 * Set capture broadcast bit to capture broadcast frames.
2340 	 */
2341 	if (ifp->if_flags & IFF_BROADCAST)
2342 		rxcfg |= RL_RXCFG_RX_BROAD;
2343 	else
2344 		rxcfg &= ~RL_RXCFG_RX_BROAD;
2345 	CSR_WRITE_4(sc, RL_RXCFG, rxcfg);
2346 
2347 	/*
2348 	 * Program the multicast filter, if necessary.
2349 	 */
2350 	re_setmulti(sc);
2351 
2352 #ifdef DEVICE_POLLING
2353 	/*
2354 	 * Disable interrupts if we are polling.
2355 	 */
2356 	if (ifp->if_capenable & IFCAP_POLLING)
2357 		CSR_WRITE_2(sc, RL_IMR, 0);
2358 	else	/* otherwise ... */
2359 #endif
2360 
2361 	/*
2362 	 * Enable interrupts.
2363 	 */
2364 	if (sc->rl_testmode)
2365 		CSR_WRITE_2(sc, RL_IMR, 0);
2366 	else
2367 		CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2368 	CSR_WRITE_2(sc, RL_ISR, RL_INTRS_CPLUS);
2369 
2370 	/* Set initial TX threshold */
2371 	sc->rl_txthresh = RL_TX_THRESH_INIT;
2372 
2373 	/* Start RX/TX process. */
2374 	CSR_WRITE_4(sc, RL_MISSEDPKT, 0);
2375 #ifdef notdef
2376 	/* Enable receiver and transmitter. */
2377 	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
2378 #endif
2379 	/*
2380 	 * Load the addresses of the RX and TX lists into the chip.
2381 	 */
2382 
2383 	CSR_WRITE_4(sc, RL_RXLIST_ADDR_HI,
2384 	    RL_ADDR_HI(sc->rl_ldata.rl_rx_list_addr));
2385 	CSR_WRITE_4(sc, RL_RXLIST_ADDR_LO,
2386 	    RL_ADDR_LO(sc->rl_ldata.rl_rx_list_addr));
2387 
2388 	CSR_WRITE_4(sc, RL_TXLIST_ADDR_HI,
2389 	    RL_ADDR_HI(sc->rl_ldata.rl_tx_list_addr));
2390 	CSR_WRITE_4(sc, RL_TXLIST_ADDR_LO,
2391 	    RL_ADDR_LO(sc->rl_ldata.rl_tx_list_addr));
2392 
2393 	CSR_WRITE_1(sc, RL_EARLY_TX_THRESH, 16);
2394 
2395 #ifdef RE_TX_MODERATION
2396 	/*
2397 	 * Initialize the timer interrupt register so that
2398 	 * a timer interrupt will be generated once the timer
2399 	 * reaches a certain number of ticks. The timer is
2400 	 * reloaded on each transmit. This gives us TX interrupt
2401 	 * moderation, which dramatically improves TX frame rate.
2402 	 */
2403 	if (sc->rl_type == RL_8169)
2404 		CSR_WRITE_4(sc, RL_TIMERINT_8169, 0x800);
2405 	else
2406 		CSR_WRITE_4(sc, RL_TIMERINT, 0x400);
2407 #endif
2408 
2409 	/*
2410 	 * For 8169 gigE NICs, set the max allowed RX packet
2411 	 * size so we can receive jumbo frames.
2412 	 */
2413 	if (sc->rl_type == RL_8169)
2414 		CSR_WRITE_2(sc, RL_MAXRXPKTLEN, 16383);
2415 
2416 	if (sc->rl_testmode)
2417 		return;
2418 
2419 	mii_mediachg(mii);
2420 
2421 	CSR_WRITE_1(sc, RL_CFG1, CSR_READ_1(sc, RL_CFG1) | RL_CFG1_DRVLOAD);
2422 
2423 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
2424 	ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
2425 
2426 	sc->rl_link = 0;
2427 	sc->rl_watchdog_timer = 0;
2428 	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
2429 }
2430 
2431 /*
2432  * Set media options.
2433  */
2434 static int
2435 re_ifmedia_upd(ifp)
2436 	struct ifnet		*ifp;
2437 {
2438 	struct rl_softc		*sc;
2439 	struct mii_data		*mii;
2440 
2441 	sc = ifp->if_softc;
2442 	mii = device_get_softc(sc->rl_miibus);
2443 	RL_LOCK(sc);
2444 	mii_mediachg(mii);
2445 	RL_UNLOCK(sc);
2446 
2447 	return (0);
2448 }
2449 
2450 /*
2451  * Report current media status.
2452  */
2453 static void
2454 re_ifmedia_sts(ifp, ifmr)
2455 	struct ifnet		*ifp;
2456 	struct ifmediareq	*ifmr;
2457 {
2458 	struct rl_softc		*sc;
2459 	struct mii_data		*mii;
2460 
2461 	sc = ifp->if_softc;
2462 	mii = device_get_softc(sc->rl_miibus);
2463 
2464 	RL_LOCK(sc);
2465 	mii_pollstat(mii);
2466 	RL_UNLOCK(sc);
2467 	ifmr->ifm_active = mii->mii_media_active;
2468 	ifmr->ifm_status = mii->mii_media_status;
2469 }
2470 
2471 static int
2472 re_ioctl(ifp, command, data)
2473 	struct ifnet		*ifp;
2474 	u_long			command;
2475 	caddr_t			data;
2476 {
2477 	struct rl_softc		*sc = ifp->if_softc;
2478 	struct ifreq		*ifr = (struct ifreq *) data;
2479 	struct mii_data		*mii;
2480 	int			error = 0;
2481 
2482 	switch (command) {
2483 	case SIOCSIFMTU:
2484 		RL_LOCK(sc);
2485 		if (ifr->ifr_mtu > RL_JUMBO_MTU)
2486 			error = EINVAL;
2487 		ifp->if_mtu = ifr->ifr_mtu;
2488 		RL_UNLOCK(sc);
2489 		break;
2490 	case SIOCSIFFLAGS:
2491 		RL_LOCK(sc);
2492 		if (ifp->if_flags & IFF_UP)
2493 			re_init_locked(sc);
2494 		else if (ifp->if_drv_flags & IFF_DRV_RUNNING)
2495 			re_stop(sc);
2496 		RL_UNLOCK(sc);
2497 		break;
2498 	case SIOCADDMULTI:
2499 	case SIOCDELMULTI:
2500 		RL_LOCK(sc);
2501 		re_setmulti(sc);
2502 		RL_UNLOCK(sc);
2503 		break;
2504 	case SIOCGIFMEDIA:
2505 	case SIOCSIFMEDIA:
2506 		mii = device_get_softc(sc->rl_miibus);
2507 		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, command);
2508 		break;
2509 	case SIOCSIFCAP:
2510 	    {
2511 		int mask, reinit;
2512 
2513 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
2514 		reinit = 0;
2515 #ifdef DEVICE_POLLING
2516 		if (mask & IFCAP_POLLING) {
2517 			if (ifr->ifr_reqcap & IFCAP_POLLING) {
2518 				error = ether_poll_register(re_poll, ifp);
2519 				if (error)
2520 					return(error);
2521 				RL_LOCK(sc);
2522 				/* Disable interrupts */
2523 				CSR_WRITE_2(sc, RL_IMR, 0x0000);
2524 				ifp->if_capenable |= IFCAP_POLLING;
2525 				RL_UNLOCK(sc);
2526 			} else {
2527 				error = ether_poll_deregister(ifp);
2528 				/* Enable interrupts. */
2529 				RL_LOCK(sc);
2530 				CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2531 				ifp->if_capenable &= ~IFCAP_POLLING;
2532 				RL_UNLOCK(sc);
2533 			}
2534 		}
2535 #endif /* DEVICE_POLLING */
2536 		if (mask & IFCAP_HWCSUM) {
2537 			ifp->if_capenable ^= IFCAP_HWCSUM;
2538 			if (ifp->if_capenable & IFCAP_TXCSUM)
2539 				ifp->if_hwassist |= RE_CSUM_FEATURES;
2540 			else
2541 				ifp->if_hwassist &= ~RE_CSUM_FEATURES;
2542 			reinit = 1;
2543 		}
2544 		if (mask & IFCAP_VLAN_HWTAGGING) {
2545 			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
2546 			reinit = 1;
2547 		}
2548 		if (mask & IFCAP_TSO4) {
2549 			ifp->if_capenable ^= IFCAP_TSO4;
2550 			if ((IFCAP_TSO4 & ifp->if_capenable) &&
2551 			    (IFCAP_TSO4 & ifp->if_capabilities))
2552 				ifp->if_hwassist |= CSUM_TSO;
2553 			else
2554 				ifp->if_hwassist &= ~CSUM_TSO;
2555 		}
2556 		if (reinit && ifp->if_drv_flags & IFF_DRV_RUNNING)
2557 			re_init(sc);
2558 		VLAN_CAPABILITIES(ifp);
2559 	    }
2560 		break;
2561 	default:
2562 		error = ether_ioctl(ifp, command, data);
2563 		break;
2564 	}
2565 
2566 	return (error);
2567 }
2568 
2569 static void
2570 re_watchdog(sc)
2571 	struct rl_softc		*sc;
2572 {
2573 
2574 	RL_LOCK_ASSERT(sc);
2575 
2576 	if (sc->rl_watchdog_timer == 0 || --sc->rl_watchdog_timer != 0)
2577 		return;
2578 
2579 	device_printf(sc->rl_dev, "watchdog timeout\n");
2580 	sc->rl_ifp->if_oerrors++;
2581 
2582 	re_txeof(sc);
2583 	re_rxeof(sc);
2584 	re_init_locked(sc);
2585 }
2586 
2587 /*
2588  * Stop the adapter and free any mbufs allocated to the
2589  * RX and TX lists.
2590  */
2591 static void
2592 re_stop(sc)
2593 	struct rl_softc		*sc;
2594 {
2595 	register int		i;
2596 	struct ifnet		*ifp;
2597 
2598 	RL_LOCK_ASSERT(sc);
2599 
2600 	ifp = sc->rl_ifp;
2601 
2602 	sc->rl_watchdog_timer = 0;
2603 	callout_stop(&sc->rl_stat_callout);
2604 	ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
2605 
2606 	CSR_WRITE_1(sc, RL_COMMAND, 0x00);
2607 	CSR_WRITE_2(sc, RL_IMR, 0x0000);
2608 	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
2609 
2610 	if (sc->rl_head != NULL) {
2611 		m_freem(sc->rl_head);
2612 		sc->rl_head = sc->rl_tail = NULL;
2613 	}
2614 
2615 	/* Free the TX list buffers. */
2616 
2617 	for (i = 0; i < RL_TX_DESC_CNT; i++) {
2618 		if (sc->rl_ldata.rl_tx_mbuf[i] != NULL) {
2619 			bus_dmamap_unload(sc->rl_ldata.rl_mtag,
2620 			    sc->rl_ldata.rl_tx_dmamap[i]);
2621 			m_freem(sc->rl_ldata.rl_tx_mbuf[i]);
2622 			sc->rl_ldata.rl_tx_mbuf[i] = NULL;
2623 		}
2624 	}
2625 
2626 	/* Free the RX list buffers. */
2627 
2628 	for (i = 0; i < RL_RX_DESC_CNT; i++) {
2629 		if (sc->rl_ldata.rl_rx_mbuf[i] != NULL) {
2630 			bus_dmamap_unload(sc->rl_ldata.rl_mtag,
2631 			    sc->rl_ldata.rl_rx_dmamap[i]);
2632 			m_freem(sc->rl_ldata.rl_rx_mbuf[i]);
2633 			sc->rl_ldata.rl_rx_mbuf[i] = NULL;
2634 		}
2635 	}
2636 }
2637 
2638 /*
2639  * Device suspend routine.  Stop the interface and save some PCI
2640  * settings in case the BIOS doesn't restore them properly on
2641  * resume.
2642  */
2643 static int
2644 re_suspend(dev)
2645 	device_t		dev;
2646 {
2647 	struct rl_softc		*sc;
2648 
2649 	sc = device_get_softc(dev);
2650 
2651 	RL_LOCK(sc);
2652 	re_stop(sc);
2653 	sc->suspended = 1;
2654 	RL_UNLOCK(sc);
2655 
2656 	return (0);
2657 }
2658 
2659 /*
2660  * Device resume routine.  Restore some PCI settings in case the BIOS
2661  * doesn't, re-enable busmastering, and restart the interface if
2662  * appropriate.
2663  */
2664 static int
2665 re_resume(dev)
2666 	device_t		dev;
2667 {
2668 	struct rl_softc		*sc;
2669 	struct ifnet		*ifp;
2670 
2671 	sc = device_get_softc(dev);
2672 
2673 	RL_LOCK(sc);
2674 
2675 	ifp = sc->rl_ifp;
2676 
2677 	/* reinitialize interface if necessary */
2678 	if (ifp->if_flags & IFF_UP)
2679 		re_init_locked(sc);
2680 
2681 	sc->suspended = 0;
2682 	RL_UNLOCK(sc);
2683 
2684 	return (0);
2685 }
2686 
2687 /*
2688  * Stop all chip I/O so that the kernel's probe routines don't
2689  * get confused by errant DMAs when rebooting.
2690  */
2691 static void
2692 re_shutdown(dev)
2693 	device_t		dev;
2694 {
2695 	struct rl_softc		*sc;
2696 
2697 	sc = device_get_softc(dev);
2698 
2699 	RL_LOCK(sc);
2700 	re_stop(sc);
2701 	/*
2702 	 * Mark interface as down since otherwise we will panic if
2703 	 * interrupt comes in later on, which can happen in some
2704 	 * cases.
2705 	 */
2706 	sc->rl_ifp->if_flags &= ~IFF_UP;
2707 	RL_UNLOCK(sc);
2708 }
2709