xref: /freebsd/sys/dev/re/if_re.c (revision d139ce67c0b39ab6532275f7baff67d220fe8001)
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 	u_int32_t		hwrev;
624 
625 	RL_LOCK_ASSERT(sc);
626 
627 	ifp = sc->rl_ifp;
628 
629 	rxfilt = CSR_READ_4(sc, RL_RXCFG);
630 
631 	if (ifp->if_flags & IFF_ALLMULTI || ifp->if_flags & IFF_PROMISC) {
632 		rxfilt |= RL_RXCFG_RX_MULTI;
633 		CSR_WRITE_4(sc, RL_RXCFG, rxfilt);
634 		CSR_WRITE_4(sc, RL_MAR0, 0xFFFFFFFF);
635 		CSR_WRITE_4(sc, RL_MAR4, 0xFFFFFFFF);
636 		return;
637 	}
638 
639 	/* first, zot all the existing hash bits */
640 	CSR_WRITE_4(sc, RL_MAR0, 0);
641 	CSR_WRITE_4(sc, RL_MAR4, 0);
642 
643 	/* now program new ones */
644 	IF_ADDR_LOCK(ifp);
645 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
646 		if (ifma->ifma_addr->sa_family != AF_LINK)
647 			continue;
648 		h = ether_crc32_be(LLADDR((struct sockaddr_dl *)
649 		    ifma->ifma_addr), ETHER_ADDR_LEN) >> 26;
650 		if (h < 32)
651 			hashes[0] |= (1 << h);
652 		else
653 			hashes[1] |= (1 << (h - 32));
654 		mcnt++;
655 	}
656 	IF_ADDR_UNLOCK(ifp);
657 
658 	if (mcnt)
659 		rxfilt |= RL_RXCFG_RX_MULTI;
660 	else
661 		rxfilt &= ~RL_RXCFG_RX_MULTI;
662 
663 	CSR_WRITE_4(sc, RL_RXCFG, rxfilt);
664 
665 	/*
666 	 * For some unfathomable reason, RealTek decided to reverse
667 	 * the order of the multicast hash registers in the PCI Express
668 	 * parts. This means we have to write the hash pattern in reverse
669 	 * order for those devices.
670 	 */
671 
672 	hwrev = CSR_READ_4(sc, RL_TXCFG) & RL_TXCFG_HWREV;
673 
674 	if (hwrev == RL_HWREV_8100E || hwrev == RL_HWREV_8101E ||
675 	    hwrev == RL_HWREV_8168_SPIN1 || hwrev == RL_HWREV_8168_SPIN2) {
676 		CSR_WRITE_4(sc, RL_MAR0, bswap32(hashes[1]));
677 		CSR_WRITE_4(sc, RL_MAR4, bswap32(hashes[0]));
678 	} else {
679 		CSR_WRITE_4(sc, RL_MAR0, hashes[0]);
680 		CSR_WRITE_4(sc, RL_MAR4, hashes[1]);
681 	}
682 }
683 
684 static void
685 re_reset(sc)
686 	struct rl_softc		*sc;
687 {
688 	register int		i;
689 
690 	RL_LOCK_ASSERT(sc);
691 
692 	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RESET);
693 
694 	for (i = 0; i < RL_TIMEOUT; i++) {
695 		DELAY(10);
696 		if (!(CSR_READ_1(sc, RL_COMMAND) & RL_CMD_RESET))
697 			break;
698 	}
699 	if (i == RL_TIMEOUT)
700 		device_printf(sc->rl_dev, "reset never completed!\n");
701 
702 	CSR_WRITE_1(sc, 0x82, 1);
703 }
704 
705 #ifdef RE_DIAG
706 
707 /*
708  * The following routine is designed to test for a defect on some
709  * 32-bit 8169 cards. Some of these NICs have the REQ64# and ACK64#
710  * lines connected to the bus, however for a 32-bit only card, they
711  * should be pulled high. The result of this defect is that the
712  * NIC will not work right if you plug it into a 64-bit slot: DMA
713  * operations will be done with 64-bit transfers, which will fail
714  * because the 64-bit data lines aren't connected.
715  *
716  * There's no way to work around this (short of talking a soldering
717  * iron to the board), however we can detect it. The method we use
718  * here is to put the NIC into digital loopback mode, set the receiver
719  * to promiscuous mode, and then try to send a frame. We then compare
720  * the frame data we sent to what was received. If the data matches,
721  * then the NIC is working correctly, otherwise we know the user has
722  * a defective NIC which has been mistakenly plugged into a 64-bit PCI
723  * slot. In the latter case, there's no way the NIC can work correctly,
724  * so we print out a message on the console and abort the device attach.
725  */
726 
727 static int
728 re_diag(sc)
729 	struct rl_softc		*sc;
730 {
731 	struct ifnet		*ifp = sc->rl_ifp;
732 	struct mbuf		*m0;
733 	struct ether_header	*eh;
734 	struct rl_desc		*cur_rx;
735 	u_int16_t		status;
736 	u_int32_t		rxstat;
737 	int			total_len, i, error = 0, phyaddr;
738 	u_int8_t		dst[] = { 0x00, 'h', 'e', 'l', 'l', 'o' };
739 	u_int8_t		src[] = { 0x00, 'w', 'o', 'r', 'l', 'd' };
740 
741 	/* Allocate a single mbuf */
742 	MGETHDR(m0, M_DONTWAIT, MT_DATA);
743 	if (m0 == NULL)
744 		return (ENOBUFS);
745 
746 	RL_LOCK(sc);
747 
748 	/*
749 	 * Initialize the NIC in test mode. This sets the chip up
750 	 * so that it can send and receive frames, but performs the
751 	 * following special functions:
752 	 * - Puts receiver in promiscuous mode
753 	 * - Enables digital loopback mode
754 	 * - Leaves interrupts turned off
755 	 */
756 
757 	ifp->if_flags |= IFF_PROMISC;
758 	sc->rl_testmode = 1;
759 	re_reset(sc);
760 	re_init_locked(sc);
761 	sc->rl_link = 1;
762 	if (sc->rl_type == RL_8169)
763 		phyaddr = 1;
764 	else
765 		phyaddr = 0;
766 
767 	re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_RESET);
768 	for (i = 0; i < RL_TIMEOUT; i++) {
769 		status = re_miibus_readreg(sc->rl_dev, phyaddr, MII_BMCR);
770 		if (!(status & BMCR_RESET))
771 			break;
772 	}
773 
774 	re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_LOOP);
775 	CSR_WRITE_2(sc, RL_ISR, RL_INTRS);
776 
777 	DELAY(100000);
778 
779 	/* Put some data in the mbuf */
780 
781 	eh = mtod(m0, struct ether_header *);
782 	bcopy ((char *)&dst, eh->ether_dhost, ETHER_ADDR_LEN);
783 	bcopy ((char *)&src, eh->ether_shost, ETHER_ADDR_LEN);
784 	eh->ether_type = htons(ETHERTYPE_IP);
785 	m0->m_pkthdr.len = m0->m_len = ETHER_MIN_LEN - ETHER_CRC_LEN;
786 
787 	/*
788 	 * Queue the packet, start transmission.
789 	 * Note: IF_HANDOFF() ultimately calls re_start() for us.
790 	 */
791 
792 	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
793 	RL_UNLOCK(sc);
794 	/* XXX: re_diag must not be called when in ALTQ mode */
795 	IF_HANDOFF(&ifp->if_snd, m0, ifp);
796 	RL_LOCK(sc);
797 	m0 = NULL;
798 
799 	/* Wait for it to propagate through the chip */
800 
801 	DELAY(100000);
802 	for (i = 0; i < RL_TIMEOUT; i++) {
803 		status = CSR_READ_2(sc, RL_ISR);
804 		CSR_WRITE_2(sc, RL_ISR, status);
805 		if ((status & (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK)) ==
806 		    (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK))
807 			break;
808 		DELAY(10);
809 	}
810 
811 	if (i == RL_TIMEOUT) {
812 		device_printf(sc->rl_dev,
813 		    "diagnostic failed, failed to receive packet in"
814 		    " loopback mode\n");
815 		error = EIO;
816 		goto done;
817 	}
818 
819 	/*
820 	 * The packet should have been dumped into the first
821 	 * entry in the RX DMA ring. Grab it from there.
822 	 */
823 
824 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
825 	    sc->rl_ldata.rl_rx_list_map,
826 	    BUS_DMASYNC_POSTREAD);
827 	bus_dmamap_sync(sc->rl_ldata.rl_mtag,
828 	    sc->rl_ldata.rl_rx_dmamap[0],
829 	    BUS_DMASYNC_POSTWRITE);
830 	bus_dmamap_unload(sc->rl_ldata.rl_mtag,
831 	    sc->rl_ldata.rl_rx_dmamap[0]);
832 
833 	m0 = sc->rl_ldata.rl_rx_mbuf[0];
834 	sc->rl_ldata.rl_rx_mbuf[0] = NULL;
835 	eh = mtod(m0, struct ether_header *);
836 
837 	cur_rx = &sc->rl_ldata.rl_rx_list[0];
838 	total_len = RL_RXBYTES(cur_rx);
839 	rxstat = le32toh(cur_rx->rl_cmdstat);
840 
841 	if (total_len != ETHER_MIN_LEN) {
842 		device_printf(sc->rl_dev,
843 		    "diagnostic failed, received short packet\n");
844 		error = EIO;
845 		goto done;
846 	}
847 
848 	/* Test that the received packet data matches what we sent. */
849 
850 	if (bcmp((char *)&eh->ether_dhost, (char *)&dst, ETHER_ADDR_LEN) ||
851 	    bcmp((char *)&eh->ether_shost, (char *)&src, ETHER_ADDR_LEN) ||
852 	    ntohs(eh->ether_type) != ETHERTYPE_IP) {
853 		device_printf(sc->rl_dev, "WARNING, DMA FAILURE!\n");
854 		device_printf(sc->rl_dev, "expected TX data: %6D/%6D/0x%x\n",
855 		    dst, ":", src, ":", ETHERTYPE_IP);
856 		device_printf(sc->rl_dev, "received RX data: %6D/%6D/0x%x\n",
857 		    eh->ether_dhost, ":",  eh->ether_shost, ":",
858 		    ntohs(eh->ether_type));
859 		device_printf(sc->rl_dev, "You may have a defective 32-bit "
860 		    "NIC plugged into a 64-bit PCI slot.\n");
861 		device_printf(sc->rl_dev, "Please re-install the NIC in a "
862 		    "32-bit slot for proper operation.\n");
863 		device_printf(sc->rl_dev, "Read the re(4) man page for more "
864 		    "details.\n");
865 		error = EIO;
866 	}
867 
868 done:
869 	/* Turn interface off, release resources */
870 
871 	sc->rl_testmode = 0;
872 	sc->rl_link = 0;
873 	ifp->if_flags &= ~IFF_PROMISC;
874 	re_stop(sc);
875 	if (m0 != NULL)
876 		m_freem(m0);
877 
878 	RL_UNLOCK(sc);
879 
880 	return (error);
881 }
882 
883 #endif
884 
885 /*
886  * Probe for a RealTek 8139C+/8169/8110 chip. Check the PCI vendor and device
887  * IDs against our list and return a device name if we find a match.
888  */
889 static int
890 re_probe(dev)
891 	device_t		dev;
892 {
893 	struct rl_type		*t;
894 	struct rl_softc		*sc;
895 	int			rid;
896 	u_int32_t		hwrev;
897 
898 	t = re_devs;
899 	sc = device_get_softc(dev);
900 
901 	while (t->rl_name != NULL) {
902 		if ((pci_get_vendor(dev) == t->rl_vid) &&
903 		    (pci_get_device(dev) == t->rl_did)) {
904 			/*
905 			 * Only attach to rev. 3 of the Linksys EG1032 adapter.
906 			 * Rev. 2 i supported by sk(4).
907 			 */
908 			if ((t->rl_vid == LINKSYS_VENDORID) &&
909 				(t->rl_did == LINKSYS_DEVICEID_EG1032) &&
910 				(pci_get_subdevice(dev) !=
911 				LINKSYS_SUBDEVICE_EG1032_REV3)) {
912 				t++;
913 				continue;
914 			}
915 
916 			/*
917 			 * Temporarily map the I/O space
918 			 * so we can read the chip ID register.
919 			 */
920 			rid = RL_RID;
921 			sc->rl_res = bus_alloc_resource_any(dev, RL_RES, &rid,
922 			    RF_ACTIVE);
923 			if (sc->rl_res == NULL) {
924 				device_printf(dev,
925 				    "couldn't map ports/memory\n");
926 				return (ENXIO);
927 			}
928 			sc->rl_btag = rman_get_bustag(sc->rl_res);
929 			sc->rl_bhandle = rman_get_bushandle(sc->rl_res);
930 			hwrev = CSR_READ_4(sc, RL_TXCFG) & RL_TXCFG_HWREV;
931 			bus_release_resource(dev, RL_RES,
932 			    RL_RID, sc->rl_res);
933 			if (t->rl_basetype == hwrev) {
934 				device_set_desc(dev, t->rl_name);
935 				return (BUS_PROBE_DEFAULT);
936 			}
937 		}
938 		t++;
939 	}
940 
941 	return (ENXIO);
942 }
943 
944 /*
945  * This routine takes the segment list provided as the result of
946  * a bus_dma_map_load() operation and assigns the addresses/lengths
947  * to RealTek DMA descriptors. This can be called either by the RX
948  * code or the TX code. In the RX case, we'll probably wind up mapping
949  * at most one segment. For the TX case, there could be any number of
950  * segments since TX packets may span multiple mbufs. In either case,
951  * if the number of segments is larger than the rl_maxsegs limit
952  * specified by the caller, we abort the mapping operation. Sadly,
953  * whoever designed the buffer mapping API did not provide a way to
954  * return an error from here, so we have to fake it a bit.
955  */
956 
957 static void
958 re_dma_map_desc(arg, segs, nseg, mapsize, error)
959 	void			*arg;
960 	bus_dma_segment_t	*segs;
961 	int			nseg;
962 	bus_size_t		mapsize;
963 	int			error;
964 {
965 	struct rl_dmaload_arg	*ctx;
966 	struct rl_desc		*d = NULL;
967 	int			i = 0, idx;
968 	u_int32_t		cmdstat;
969 	int			totlen = 0;
970 
971 	if (error)
972 		return;
973 
974 	ctx = arg;
975 
976 	/* Signal error to caller if there's too many segments */
977 	if (nseg > ctx->rl_maxsegs) {
978 		ctx->rl_maxsegs = 0;
979 		return;
980 	}
981 
982 	/*
983 	 * Map the segment array into descriptors. Note that we set the
984 	 * start-of-frame and end-of-frame markers for either TX or RX, but
985 	 * they really only have meaning in the TX case. (In the RX case,
986 	 * it's the chip that tells us where packets begin and end.)
987 	 * We also keep track of the end of the ring and set the
988 	 * end-of-ring bits as needed, and we set the ownership bits
989 	 * in all except the very first descriptor. (The caller will
990 	 * set this descriptor later when it start transmission or
991 	 * reception.)
992 	 */
993 	idx = ctx->rl_idx;
994 	for (;;) {
995 		d = &ctx->rl_ring[idx];
996 		if (le32toh(d->rl_cmdstat) & RL_RDESC_STAT_OWN) {
997 			ctx->rl_maxsegs = 0;
998 			return;
999 		}
1000 		cmdstat = segs[i].ds_len;
1001 		totlen += segs[i].ds_len;
1002 		d->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[i].ds_addr));
1003 		d->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[i].ds_addr));
1004 		if (i == 0)
1005 			cmdstat |= RL_TDESC_CMD_SOF;
1006 		else
1007 			cmdstat |= RL_TDESC_CMD_OWN;
1008 		if (idx == (RL_RX_DESC_CNT - 1))
1009 			cmdstat |= RL_TDESC_CMD_EOR;
1010 		d->rl_cmdstat = htole32(cmdstat | ctx->rl_flags);
1011 		i++;
1012 		if (i == nseg)
1013 			break;
1014 		RL_DESC_INC(idx);
1015 	}
1016 
1017 	d->rl_cmdstat |= htole32(RL_TDESC_CMD_EOF);
1018 	ctx->rl_maxsegs = nseg;
1019 	ctx->rl_idx = idx;
1020 }
1021 
1022 /*
1023  * Map a single buffer address.
1024  */
1025 
1026 static void
1027 re_dma_map_addr(arg, segs, nseg, error)
1028 	void			*arg;
1029 	bus_dma_segment_t	*segs;
1030 	int			nseg;
1031 	int			error;
1032 {
1033 	bus_addr_t		*addr;
1034 
1035 	if (error)
1036 		return;
1037 
1038 	KASSERT(nseg == 1, ("too many DMA segments, %d should be 1", nseg));
1039 	addr = arg;
1040 	*addr = segs->ds_addr;
1041 }
1042 
1043 static int
1044 re_allocmem(dev, sc)
1045 	device_t		dev;
1046 	struct rl_softc		*sc;
1047 {
1048 	int			error;
1049 	int			nseg;
1050 	int			i;
1051 
1052 	/*
1053 	 * Allocate map for RX mbufs.
1054 	 */
1055 	nseg = 32;
1056 	error = bus_dma_tag_create(sc->rl_parent_tag, ETHER_ALIGN, 0,
1057 	    BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1058 	    NULL, MCLBYTES * nseg, nseg, MCLBYTES, BUS_DMA_ALLOCNOW,
1059 	    NULL, NULL, &sc->rl_ldata.rl_mtag);
1060 	if (error) {
1061 		device_printf(dev, "could not allocate dma tag\n");
1062 		return (ENOMEM);
1063 	}
1064 
1065 	/*
1066 	 * Allocate map for TX descriptor list.
1067 	 */
1068 	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
1069 	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1070 	    NULL, RL_TX_LIST_SZ, 1, RL_TX_LIST_SZ, 0,
1071 	    NULL, NULL, &sc->rl_ldata.rl_tx_list_tag);
1072 	if (error) {
1073 		device_printf(dev, "could not allocate dma tag\n");
1074 		return (ENOMEM);
1075 	}
1076 
1077 	/* Allocate DMA'able memory for the TX ring */
1078 
1079 	error = bus_dmamem_alloc(sc->rl_ldata.rl_tx_list_tag,
1080 	    (void **)&sc->rl_ldata.rl_tx_list, BUS_DMA_NOWAIT | BUS_DMA_ZERO,
1081 	    &sc->rl_ldata.rl_tx_list_map);
1082 	if (error)
1083 		return (ENOMEM);
1084 
1085 	/* Load the map for the TX ring. */
1086 
1087 	error = bus_dmamap_load(sc->rl_ldata.rl_tx_list_tag,
1088 	     sc->rl_ldata.rl_tx_list_map, sc->rl_ldata.rl_tx_list,
1089 	     RL_TX_LIST_SZ, re_dma_map_addr,
1090 	     &sc->rl_ldata.rl_tx_list_addr, BUS_DMA_NOWAIT);
1091 
1092 	/* Create DMA maps for TX buffers */
1093 
1094 	for (i = 0; i < RL_TX_DESC_CNT; i++) {
1095 		error = bus_dmamap_create(sc->rl_ldata.rl_mtag, 0,
1096 			    &sc->rl_ldata.rl_tx_dmamap[i]);
1097 		if (error) {
1098 			device_printf(dev, "can't create DMA map for TX\n");
1099 			return (ENOMEM);
1100 		}
1101 	}
1102 
1103 	/*
1104 	 * Allocate map for RX descriptor list.
1105 	 */
1106 	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
1107 	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1108 	    NULL, RL_RX_LIST_SZ, 1, RL_RX_LIST_SZ, 0,
1109 	    NULL, NULL, &sc->rl_ldata.rl_rx_list_tag);
1110 	if (error) {
1111 		device_printf(dev, "could not allocate dma tag\n");
1112 		return (ENOMEM);
1113 	}
1114 
1115 	/* Allocate DMA'able memory for the RX ring */
1116 
1117 	error = bus_dmamem_alloc(sc->rl_ldata.rl_rx_list_tag,
1118 	    (void **)&sc->rl_ldata.rl_rx_list, BUS_DMA_NOWAIT | BUS_DMA_ZERO,
1119 	    &sc->rl_ldata.rl_rx_list_map);
1120 	if (error)
1121 		return (ENOMEM);
1122 
1123 	/* Load the map for the RX ring. */
1124 
1125 	error = bus_dmamap_load(sc->rl_ldata.rl_rx_list_tag,
1126 	     sc->rl_ldata.rl_rx_list_map, sc->rl_ldata.rl_rx_list,
1127 	     RL_RX_LIST_SZ, re_dma_map_addr,
1128 	     &sc->rl_ldata.rl_rx_list_addr, BUS_DMA_NOWAIT);
1129 
1130 	/* Create DMA maps for RX buffers */
1131 
1132 	for (i = 0; i < RL_RX_DESC_CNT; i++) {
1133 		error = bus_dmamap_create(sc->rl_ldata.rl_mtag, 0,
1134 			    &sc->rl_ldata.rl_rx_dmamap[i]);
1135 		if (error) {
1136 			device_printf(dev, "can't create DMA map for RX\n");
1137 			return (ENOMEM);
1138 		}
1139 	}
1140 
1141 	return (0);
1142 }
1143 
1144 /*
1145  * Attach the interface. Allocate softc structures, do ifmedia
1146  * setup and ethernet/BPF attach.
1147  */
1148 static int
1149 re_attach(dev)
1150 	device_t		dev;
1151 {
1152 	u_char			eaddr[ETHER_ADDR_LEN];
1153 	u_int16_t		as[ETHER_ADDR_LEN / 2];
1154 	struct rl_softc		*sc;
1155 	struct ifnet		*ifp;
1156 	struct rl_hwrev		*hw_rev;
1157 	int			hwrev;
1158 	u_int16_t		re_did = 0;
1159 	int			error = 0, rid, i;
1160 
1161 	sc = device_get_softc(dev);
1162 	sc->rl_dev = dev;
1163 
1164 	mtx_init(&sc->rl_mtx, device_get_nameunit(dev), MTX_NETWORK_LOCK,
1165 	    MTX_DEF);
1166 	callout_init_mtx(&sc->rl_stat_callout, &sc->rl_mtx, 0);
1167 
1168 	/*
1169 	 * Map control/status registers.
1170 	 */
1171 	pci_enable_busmaster(dev);
1172 
1173 	rid = RL_RID;
1174 	sc->rl_res = bus_alloc_resource_any(dev, RL_RES, &rid,
1175 	    RF_ACTIVE);
1176 
1177 	if (sc->rl_res == NULL) {
1178 		device_printf(dev, "couldn't map ports/memory\n");
1179 		error = ENXIO;
1180 		goto fail;
1181 	}
1182 
1183 	sc->rl_btag = rman_get_bustag(sc->rl_res);
1184 	sc->rl_bhandle = rman_get_bushandle(sc->rl_res);
1185 
1186 	/* Allocate interrupt */
1187 	rid = 0;
1188 	sc->rl_irq = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
1189 	    RF_SHAREABLE | RF_ACTIVE);
1190 
1191 	if (sc->rl_irq == NULL) {
1192 		device_printf(dev, "couldn't map interrupt\n");
1193 		error = ENXIO;
1194 		goto fail;
1195 	}
1196 
1197 	/* Reset the adapter. */
1198 	RL_LOCK(sc);
1199 	re_reset(sc);
1200 	RL_UNLOCK(sc);
1201 
1202 	hw_rev = re_hwrevs;
1203 	hwrev = CSR_READ_4(sc, RL_TXCFG) & RL_TXCFG_HWREV;
1204 	while (hw_rev->rl_desc != NULL) {
1205 		if (hw_rev->rl_rev == hwrev) {
1206 			sc->rl_type = hw_rev->rl_type;
1207 			break;
1208 		}
1209 		hw_rev++;
1210 	}
1211 
1212 	sc->rl_eewidth = 6;
1213 	re_read_eeprom(sc, (caddr_t)&re_did, 0, 1);
1214 	if (re_did != 0x8129)
1215 	        sc->rl_eewidth = 8;
1216 
1217 	/*
1218 	 * Get station address from the EEPROM.
1219 	 */
1220 	re_read_eeprom(sc, (caddr_t)as, RL_EE_EADDR, 3);
1221 	for (i = 0; i < ETHER_ADDR_LEN / 2; i++)
1222 		as[i] = le16toh(as[i]);
1223 	bcopy(as, eaddr, sizeof(eaddr));
1224 
1225 	if (sc->rl_type == RL_8169) {
1226 		/* Set RX length mask */
1227 		sc->rl_rxlenmask = RL_RDESC_STAT_GFRAGLEN;
1228 		sc->rl_txstart = RL_GTXSTART;
1229 	} else {
1230 		/* Set RX length mask */
1231 		sc->rl_rxlenmask = RL_RDESC_STAT_FRAGLEN;
1232 		sc->rl_txstart = RL_TXSTART;
1233 	}
1234 
1235 	/*
1236 	 * Allocate the parent bus DMA tag appropriate for PCI.
1237 	 */
1238 #define RL_NSEG_NEW 32
1239 	error = bus_dma_tag_create(bus_get_dma_tag(dev), 1, 0,
1240 	    BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL, NULL,
1241 	    MAXBSIZE, RL_NSEG_NEW, BUS_SPACE_MAXSIZE_32BIT, 0,
1242 	    NULL, NULL, &sc->rl_parent_tag);
1243 	if (error)
1244 		goto fail;
1245 
1246 	error = re_allocmem(dev, sc);
1247 
1248 	if (error)
1249 		goto fail;
1250 
1251 	ifp = sc->rl_ifp = if_alloc(IFT_ETHER);
1252 	if (ifp == NULL) {
1253 		device_printf(dev, "can not if_alloc()\n");
1254 		error = ENOSPC;
1255 		goto fail;
1256 	}
1257 
1258 	/* Do MII setup */
1259 	if (mii_phy_probe(dev, &sc->rl_miibus,
1260 	    re_ifmedia_upd, re_ifmedia_sts)) {
1261 		device_printf(dev, "MII without any phy!\n");
1262 		error = ENXIO;
1263 		goto fail;
1264 	}
1265 
1266 	ifp->if_softc = sc;
1267 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1268 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1269 	ifp->if_ioctl = re_ioctl;
1270 	ifp->if_start = re_start;
1271 	ifp->if_hwassist = RE_CSUM_FEATURES | CSUM_TSO;
1272 	ifp->if_capabilities = IFCAP_HWCSUM | IFCAP_TSO4;
1273 	ifp->if_capenable = ifp->if_capabilities;
1274 	ifp->if_init = re_init;
1275 	IFQ_SET_MAXLEN(&ifp->if_snd, RL_IFQ_MAXLEN);
1276 	ifp->if_snd.ifq_drv_maxlen = RL_IFQ_MAXLEN;
1277 	IFQ_SET_READY(&ifp->if_snd);
1278 
1279 	TASK_INIT(&sc->rl_txtask, 1, re_tx_task, ifp);
1280 	TASK_INIT(&sc->rl_inttask, 0, re_int_task, sc);
1281 
1282 	/*
1283 	 * Call MI attach routine.
1284 	 */
1285 	ether_ifattach(ifp, eaddr);
1286 
1287 	/* VLAN capability setup */
1288 	ifp->if_capabilities |= IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING;
1289 	if (ifp->if_capabilities & IFCAP_HWCSUM)
1290 		ifp->if_capabilities |= IFCAP_VLAN_HWCSUM;
1291 	ifp->if_capenable = ifp->if_capabilities;
1292 #ifdef DEVICE_POLLING
1293 	ifp->if_capabilities |= IFCAP_POLLING;
1294 #endif
1295 	/*
1296 	 * Tell the upper layer(s) we support long frames.
1297 	 * Must appear after the call to ether_ifattach() because
1298 	 * ether_ifattach() sets ifi_hdrlen to the default value.
1299 	 */
1300 	ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
1301 
1302 #ifdef RE_DIAG
1303 	/*
1304 	 * Perform hardware diagnostic on the original RTL8169.
1305 	 * Some 32-bit cards were incorrectly wired and would
1306 	 * malfunction if plugged into a 64-bit slot.
1307 	 */
1308 
1309 	if (hwrev == RL_HWREV_8169) {
1310 		error = re_diag(sc);
1311 		if (error) {
1312 			device_printf(dev,
1313 		    	"attach aborted due to hardware diag failure\n");
1314 			ether_ifdetach(ifp);
1315 			goto fail;
1316 		}
1317 	}
1318 #endif
1319 
1320 	/* Hook interrupt last to avoid having to lock softc */
1321 	error = bus_setup_intr(dev, sc->rl_irq, INTR_TYPE_NET | INTR_MPSAFE |
1322 	    INTR_FAST, re_intr, sc, &sc->rl_intrhand);
1323 	if (error) {
1324 		device_printf(dev, "couldn't set up irq\n");
1325 		ether_ifdetach(ifp);
1326 	}
1327 
1328 fail:
1329 
1330 	if (error)
1331 		re_detach(dev);
1332 
1333 	return (error);
1334 }
1335 
1336 /*
1337  * Shutdown hardware and free up resources. This can be called any
1338  * time after the mutex has been initialized. It is called in both
1339  * the error case in attach and the normal detach case so it needs
1340  * to be careful about only freeing resources that have actually been
1341  * allocated.
1342  */
1343 static int
1344 re_detach(dev)
1345 	device_t		dev;
1346 {
1347 	struct rl_softc		*sc;
1348 	struct ifnet		*ifp;
1349 	int			i;
1350 
1351 	sc = device_get_softc(dev);
1352 	ifp = sc->rl_ifp;
1353 	KASSERT(mtx_initialized(&sc->rl_mtx), ("re mutex not initialized"));
1354 
1355 #ifdef DEVICE_POLLING
1356 	if (ifp->if_capenable & IFCAP_POLLING)
1357 		ether_poll_deregister(ifp);
1358 #endif
1359 	/* These should only be active if attach succeeded */
1360 	if (device_is_attached(dev)) {
1361 		RL_LOCK(sc);
1362 #if 0
1363 		sc->suspended = 1;
1364 #endif
1365 		re_stop(sc);
1366 		RL_UNLOCK(sc);
1367 		callout_drain(&sc->rl_stat_callout);
1368 		/*
1369 		 * Force off the IFF_UP flag here, in case someone
1370 		 * still had a BPF descriptor attached to this
1371 		 * interface. If they do, ether_ifdetach() will cause
1372 		 * the BPF code to try and clear the promisc mode
1373 		 * flag, which will bubble down to re_ioctl(),
1374 		 * which will try to call re_init() again. This will
1375 		 * turn the NIC back on and restart the MII ticker,
1376 		 * which will panic the system when the kernel tries
1377 		 * to invoke the re_tick() function that isn't there
1378 		 * anymore.
1379 		 */
1380 		ifp->if_flags &= ~IFF_UP;
1381 		ether_ifdetach(ifp);
1382 	}
1383 	if (sc->rl_miibus)
1384 		device_delete_child(dev, sc->rl_miibus);
1385 	bus_generic_detach(dev);
1386 
1387 	/*
1388 	 * The rest is resource deallocation, so we should already be
1389 	 * stopped here.
1390 	 */
1391 
1392 	if (sc->rl_intrhand)
1393 		bus_teardown_intr(dev, sc->rl_irq, sc->rl_intrhand);
1394 	if (ifp != NULL)
1395 		if_free(ifp);
1396 	if (sc->rl_irq)
1397 		bus_release_resource(dev, SYS_RES_IRQ, 0, sc->rl_irq);
1398 	if (sc->rl_res)
1399 		bus_release_resource(dev, RL_RES, RL_RID, sc->rl_res);
1400 
1401 	/* Yield the CPU long enough for any tasks to drain */
1402 
1403         tsleep(sc, PPAUSE, "rewait", hz);
1404 
1405 	/* Unload and free the RX DMA ring memory and map */
1406 
1407 	if (sc->rl_ldata.rl_rx_list_tag) {
1408 		bus_dmamap_unload(sc->rl_ldata.rl_rx_list_tag,
1409 		    sc->rl_ldata.rl_rx_list_map);
1410 		bus_dmamem_free(sc->rl_ldata.rl_rx_list_tag,
1411 		    sc->rl_ldata.rl_rx_list,
1412 		    sc->rl_ldata.rl_rx_list_map);
1413 		bus_dma_tag_destroy(sc->rl_ldata.rl_rx_list_tag);
1414 	}
1415 
1416 	/* Unload and free the TX DMA ring memory and map */
1417 
1418 	if (sc->rl_ldata.rl_tx_list_tag) {
1419 		bus_dmamap_unload(sc->rl_ldata.rl_tx_list_tag,
1420 		    sc->rl_ldata.rl_tx_list_map);
1421 		bus_dmamem_free(sc->rl_ldata.rl_tx_list_tag,
1422 		    sc->rl_ldata.rl_tx_list,
1423 		    sc->rl_ldata.rl_tx_list_map);
1424 		bus_dma_tag_destroy(sc->rl_ldata.rl_tx_list_tag);
1425 	}
1426 
1427 	/* Destroy all the RX and TX buffer maps */
1428 
1429 	if (sc->rl_ldata.rl_mtag) {
1430 		for (i = 0; i < RL_TX_DESC_CNT; i++)
1431 			bus_dmamap_destroy(sc->rl_ldata.rl_mtag,
1432 			    sc->rl_ldata.rl_tx_dmamap[i]);
1433 		for (i = 0; i < RL_RX_DESC_CNT; i++)
1434 			bus_dmamap_destroy(sc->rl_ldata.rl_mtag,
1435 			    sc->rl_ldata.rl_rx_dmamap[i]);
1436 		bus_dma_tag_destroy(sc->rl_ldata.rl_mtag);
1437 	}
1438 
1439 	/* Unload and free the stats buffer and map */
1440 
1441 	if (sc->rl_ldata.rl_stag) {
1442 		bus_dmamap_unload(sc->rl_ldata.rl_stag,
1443 		    sc->rl_ldata.rl_rx_list_map);
1444 		bus_dmamem_free(sc->rl_ldata.rl_stag,
1445 		    sc->rl_ldata.rl_stats,
1446 		    sc->rl_ldata.rl_smap);
1447 		bus_dma_tag_destroy(sc->rl_ldata.rl_stag);
1448 	}
1449 
1450 	if (sc->rl_parent_tag)
1451 		bus_dma_tag_destroy(sc->rl_parent_tag);
1452 
1453 	mtx_destroy(&sc->rl_mtx);
1454 
1455 	return (0);
1456 }
1457 
1458 static int
1459 re_newbuf(sc, idx, m)
1460 	struct rl_softc		*sc;
1461 	int			idx;
1462 	struct mbuf		*m;
1463 {
1464 	struct rl_dmaload_arg	arg;
1465 	struct mbuf		*n = NULL;
1466 	int			error;
1467 
1468 	if (m == NULL) {
1469 		n = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
1470 		if (n == NULL)
1471 			return (ENOBUFS);
1472 		m = n;
1473 	} else
1474 		m->m_data = m->m_ext.ext_buf;
1475 
1476 	m->m_len = m->m_pkthdr.len = MCLBYTES;
1477 #ifdef RE_FIXUP_RX
1478 	/*
1479 	 * This is part of an evil trick to deal with non-x86 platforms.
1480 	 * The RealTek chip requires RX buffers to be aligned on 64-bit
1481 	 * boundaries, but that will hose non-x86 machines. To get around
1482 	 * this, we leave some empty space at the start of each buffer
1483 	 * and for non-x86 hosts, we copy the buffer back six bytes
1484 	 * to achieve word alignment. This is slightly more efficient
1485 	 * than allocating a new buffer, copying the contents, and
1486 	 * discarding the old buffer.
1487 	 */
1488 	m_adj(m, RE_ETHER_ALIGN);
1489 #endif
1490 	arg.rl_idx = idx;
1491 	arg.rl_maxsegs = 1;
1492 	arg.rl_flags = 0;
1493 	arg.rl_ring = sc->rl_ldata.rl_rx_list;
1494 
1495 	error = bus_dmamap_load_mbuf(sc->rl_ldata.rl_mtag,
1496 	    sc->rl_ldata.rl_rx_dmamap[idx], m, re_dma_map_desc,
1497 	    &arg, BUS_DMA_NOWAIT);
1498 	if (error || arg.rl_maxsegs != 1) {
1499 		if (n != NULL)
1500 			m_freem(n);
1501 		if (arg.rl_maxsegs == 0)
1502 			bus_dmamap_unload(sc->rl_ldata.rl_mtag,
1503 			    sc->rl_ldata.rl_rx_dmamap[idx]);
1504 		return (ENOMEM);
1505 	}
1506 
1507 	sc->rl_ldata.rl_rx_list[idx].rl_cmdstat |= htole32(RL_RDESC_CMD_OWN);
1508 	sc->rl_ldata.rl_rx_mbuf[idx] = m;
1509 
1510 	bus_dmamap_sync(sc->rl_ldata.rl_mtag,
1511 	    sc->rl_ldata.rl_rx_dmamap[idx],
1512 	    BUS_DMASYNC_PREREAD);
1513 
1514 	return (0);
1515 }
1516 
1517 #ifdef RE_FIXUP_RX
1518 static __inline void
1519 re_fixup_rx(m)
1520 	struct mbuf		*m;
1521 {
1522 	int                     i;
1523 	uint16_t                *src, *dst;
1524 
1525 	src = mtod(m, uint16_t *);
1526 	dst = src - (RE_ETHER_ALIGN - ETHER_ALIGN) / sizeof *src;
1527 
1528 	for (i = 0; i < (m->m_len / sizeof(uint16_t) + 1); i++)
1529 		*dst++ = *src++;
1530 
1531 	m->m_data -= RE_ETHER_ALIGN - ETHER_ALIGN;
1532 
1533 	return;
1534 }
1535 #endif
1536 
1537 static int
1538 re_tx_list_init(sc)
1539 	struct rl_softc		*sc;
1540 {
1541 
1542 	RL_LOCK_ASSERT(sc);
1543 
1544 	bzero ((char *)sc->rl_ldata.rl_tx_list, RL_TX_LIST_SZ);
1545 	bzero ((char *)&sc->rl_ldata.rl_tx_mbuf,
1546 	    (RL_TX_DESC_CNT * sizeof(struct mbuf *)));
1547 
1548 	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
1549 	    sc->rl_ldata.rl_tx_list_map, BUS_DMASYNC_PREWRITE);
1550 	sc->rl_ldata.rl_tx_prodidx = 0;
1551 	sc->rl_ldata.rl_tx_considx = 0;
1552 	sc->rl_ldata.rl_tx_free = RL_TX_DESC_CNT;
1553 
1554 	return (0);
1555 }
1556 
1557 static int
1558 re_rx_list_init(sc)
1559 	struct rl_softc		*sc;
1560 {
1561 	int			i;
1562 
1563 	bzero ((char *)sc->rl_ldata.rl_rx_list, RL_RX_LIST_SZ);
1564 	bzero ((char *)&sc->rl_ldata.rl_rx_mbuf,
1565 	    (RL_RX_DESC_CNT * sizeof(struct mbuf *)));
1566 
1567 	for (i = 0; i < RL_RX_DESC_CNT; i++) {
1568 		if (re_newbuf(sc, i, NULL) == ENOBUFS)
1569 			return (ENOBUFS);
1570 	}
1571 
1572 	/* Flush the RX descriptors */
1573 
1574 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1575 	    sc->rl_ldata.rl_rx_list_map,
1576 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
1577 
1578 	sc->rl_ldata.rl_rx_prodidx = 0;
1579 	sc->rl_head = sc->rl_tail = NULL;
1580 
1581 	return (0);
1582 }
1583 
1584 /*
1585  * RX handler for C+ and 8169. For the gigE chips, we support
1586  * the reception of jumbo frames that have been fragmented
1587  * across multiple 2K mbuf cluster buffers.
1588  */
1589 static int
1590 re_rxeof(sc)
1591 	struct rl_softc		*sc;
1592 {
1593 	struct mbuf		*m;
1594 	struct ifnet		*ifp;
1595 	int			i, total_len;
1596 	struct rl_desc		*cur_rx;
1597 	u_int32_t		rxstat, rxvlan;
1598 	int			maxpkt = 16;
1599 
1600 	RL_LOCK_ASSERT(sc);
1601 
1602 	ifp = sc->rl_ifp;
1603 	i = sc->rl_ldata.rl_rx_prodidx;
1604 
1605 	/* Invalidate the descriptor memory */
1606 
1607 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1608 	    sc->rl_ldata.rl_rx_list_map,
1609 	    BUS_DMASYNC_POSTREAD);
1610 
1611 	while (!RL_OWN(&sc->rl_ldata.rl_rx_list[i]) && maxpkt) {
1612 		cur_rx = &sc->rl_ldata.rl_rx_list[i];
1613 		m = sc->rl_ldata.rl_rx_mbuf[i];
1614 		total_len = RL_RXBYTES(cur_rx);
1615 		rxstat = le32toh(cur_rx->rl_cmdstat);
1616 		rxvlan = le32toh(cur_rx->rl_vlanctl);
1617 
1618 		/* Invalidate the RX mbuf and unload its map */
1619 
1620 		bus_dmamap_sync(sc->rl_ldata.rl_mtag,
1621 		    sc->rl_ldata.rl_rx_dmamap[i],
1622 		    BUS_DMASYNC_POSTWRITE);
1623 		bus_dmamap_unload(sc->rl_ldata.rl_mtag,
1624 		    sc->rl_ldata.rl_rx_dmamap[i]);
1625 
1626 		if (!(rxstat & RL_RDESC_STAT_EOF)) {
1627 			m->m_len = RE_RX_DESC_BUFLEN;
1628 			if (sc->rl_head == NULL)
1629 				sc->rl_head = sc->rl_tail = m;
1630 			else {
1631 				m->m_flags &= ~M_PKTHDR;
1632 				sc->rl_tail->m_next = m;
1633 				sc->rl_tail = m;
1634 			}
1635 			re_newbuf(sc, i, NULL);
1636 			RL_DESC_INC(i);
1637 			continue;
1638 		}
1639 
1640 		/*
1641 		 * NOTE: for the 8139C+, the frame length field
1642 		 * is always 12 bits in size, but for the gigE chips,
1643 		 * it is 13 bits (since the max RX frame length is 16K).
1644 		 * Unfortunately, all 32 bits in the status word
1645 		 * were already used, so to make room for the extra
1646 		 * length bit, RealTek took out the 'frame alignment
1647 		 * error' bit and shifted the other status bits
1648 		 * over one slot. The OWN, EOR, FS and LS bits are
1649 		 * still in the same places. We have already extracted
1650 		 * the frame length and checked the OWN bit, so rather
1651 		 * than using an alternate bit mapping, we shift the
1652 		 * status bits one space to the right so we can evaluate
1653 		 * them using the 8169 status as though it was in the
1654 		 * same format as that of the 8139C+.
1655 		 */
1656 		if (sc->rl_type == RL_8169)
1657 			rxstat >>= 1;
1658 
1659 		/*
1660 		 * if total_len > 2^13-1, both _RXERRSUM and _GIANT will be
1661 		 * set, but if CRC is clear, it will still be a valid frame.
1662 		 */
1663 		if (rxstat & RL_RDESC_STAT_RXERRSUM && !(total_len > 8191 &&
1664 		    (rxstat & RL_RDESC_STAT_ERRS) == RL_RDESC_STAT_GIANT)) {
1665 			ifp->if_ierrors++;
1666 			/*
1667 			 * If this is part of a multi-fragment packet,
1668 			 * discard all the pieces.
1669 			 */
1670 			if (sc->rl_head != NULL) {
1671 				m_freem(sc->rl_head);
1672 				sc->rl_head = sc->rl_tail = NULL;
1673 			}
1674 			re_newbuf(sc, i, m);
1675 			RL_DESC_INC(i);
1676 			continue;
1677 		}
1678 
1679 		/*
1680 		 * If allocating a replacement mbuf fails,
1681 		 * reload the current one.
1682 		 */
1683 
1684 		if (re_newbuf(sc, i, NULL)) {
1685 			ifp->if_ierrors++;
1686 			if (sc->rl_head != NULL) {
1687 				m_freem(sc->rl_head);
1688 				sc->rl_head = sc->rl_tail = NULL;
1689 			}
1690 			re_newbuf(sc, i, m);
1691 			RL_DESC_INC(i);
1692 			continue;
1693 		}
1694 
1695 		RL_DESC_INC(i);
1696 
1697 		if (sc->rl_head != NULL) {
1698 			m->m_len = total_len % RE_RX_DESC_BUFLEN;
1699 			if (m->m_len == 0)
1700 				m->m_len = RE_RX_DESC_BUFLEN;
1701 			/*
1702 			 * Special case: if there's 4 bytes or less
1703 			 * in this buffer, the mbuf can be discarded:
1704 			 * the last 4 bytes is the CRC, which we don't
1705 			 * care about anyway.
1706 			 */
1707 			if (m->m_len <= ETHER_CRC_LEN) {
1708 				sc->rl_tail->m_len -=
1709 				    (ETHER_CRC_LEN - m->m_len);
1710 				m_freem(m);
1711 			} else {
1712 				m->m_len -= ETHER_CRC_LEN;
1713 				m->m_flags &= ~M_PKTHDR;
1714 				sc->rl_tail->m_next = m;
1715 			}
1716 			m = sc->rl_head;
1717 			sc->rl_head = sc->rl_tail = NULL;
1718 			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
1719 		} else
1720 			m->m_pkthdr.len = m->m_len =
1721 			    (total_len - ETHER_CRC_LEN);
1722 
1723 #ifdef RE_FIXUP_RX
1724 		re_fixup_rx(m);
1725 #endif
1726 		ifp->if_ipackets++;
1727 		m->m_pkthdr.rcvif = ifp;
1728 
1729 		/* Do RX checksumming if enabled */
1730 
1731 		if (ifp->if_capenable & IFCAP_RXCSUM) {
1732 
1733 			/* Check IP header checksum */
1734 			if (rxstat & RL_RDESC_STAT_PROTOID)
1735 				m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED;
1736 			if (!(rxstat & RL_RDESC_STAT_IPSUMBAD))
1737 				m->m_pkthdr.csum_flags |= CSUM_IP_VALID;
1738 
1739 			/* Check TCP/UDP checksum */
1740 			if ((RL_TCPPKT(rxstat) &&
1741 			    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
1742 			    (RL_UDPPKT(rxstat) &&
1743 			    !(rxstat & RL_RDESC_STAT_UDPSUMBAD))) {
1744 				m->m_pkthdr.csum_flags |=
1745 				    CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
1746 				m->m_pkthdr.csum_data = 0xffff;
1747 			}
1748 		}
1749 		maxpkt--;
1750 		if (rxvlan & RL_RDESC_VLANCTL_TAG) {
1751 			m->m_pkthdr.ether_vtag =
1752 			    ntohs((rxvlan & RL_RDESC_VLANCTL_DATA));
1753 			m->m_flags |= M_VLANTAG;
1754 		}
1755 		RL_UNLOCK(sc);
1756 		(*ifp->if_input)(ifp, m);
1757 		RL_LOCK(sc);
1758 	}
1759 
1760 	/* Flush the RX DMA ring */
1761 
1762 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1763 	    sc->rl_ldata.rl_rx_list_map,
1764 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
1765 
1766 	sc->rl_ldata.rl_rx_prodidx = i;
1767 
1768 	if (maxpkt)
1769 		return(EAGAIN);
1770 
1771 	return(0);
1772 }
1773 
1774 static void
1775 re_txeof(sc)
1776 	struct rl_softc		*sc;
1777 {
1778 	struct ifnet		*ifp;
1779 	u_int32_t		txstat;
1780 	int			idx;
1781 
1782 	ifp = sc->rl_ifp;
1783 	idx = sc->rl_ldata.rl_tx_considx;
1784 
1785 	/* Invalidate the TX descriptor list */
1786 	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
1787 	    sc->rl_ldata.rl_tx_list_map,
1788 	    BUS_DMASYNC_POSTREAD);
1789 
1790 	while (sc->rl_ldata.rl_tx_free < RL_TX_DESC_CNT) {
1791 		txstat = le32toh(sc->rl_ldata.rl_tx_list[idx].rl_cmdstat);
1792 		if (txstat & RL_TDESC_CMD_OWN)
1793 			break;
1794 
1795 		sc->rl_ldata.rl_tx_list[idx].rl_bufaddr_lo = 0;
1796 
1797 		/*
1798 		 * We only stash mbufs in the last descriptor
1799 		 * in a fragment chain, which also happens to
1800 		 * be the only place where the TX status bits
1801 		 * are valid.
1802 		 */
1803 		if (txstat & RL_TDESC_CMD_EOF) {
1804 			m_freem(sc->rl_ldata.rl_tx_mbuf[idx]);
1805 			sc->rl_ldata.rl_tx_mbuf[idx] = NULL;
1806 			bus_dmamap_unload(sc->rl_ldata.rl_mtag,
1807 			    sc->rl_ldata.rl_tx_dmamap[idx]);
1808 			if (txstat & (RL_TDESC_STAT_EXCESSCOL|
1809 			    RL_TDESC_STAT_COLCNT))
1810 				ifp->if_collisions++;
1811 			if (txstat & RL_TDESC_STAT_TXERRSUM)
1812 				ifp->if_oerrors++;
1813 			else
1814 				ifp->if_opackets++;
1815 		}
1816 		sc->rl_ldata.rl_tx_free++;
1817 		RL_DESC_INC(idx);
1818 	}
1819 	sc->rl_ldata.rl_tx_considx = idx;
1820 
1821 	/* No changes made to the TX ring, so no flush needed */
1822 
1823 	if (sc->rl_ldata.rl_tx_free > RL_TX_DESC_THLD)
1824 		ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1825 
1826 	if (sc->rl_ldata.rl_tx_free < RL_TX_DESC_CNT) {
1827 		/*
1828 		 * Some chips will ignore a second TX request issued
1829 		 * while an existing transmission is in progress. If
1830 		 * the transmitter goes idle but there are still
1831 		 * packets waiting to be sent, we need to restart the
1832 		 * channel here to flush them out. This only seems to
1833 		 * be required with the PCIe devices.
1834 		 */
1835 		CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
1836 
1837 #ifdef RE_TX_MODERATION
1838 		/*
1839 		 * If not all descriptors have been reaped yet, reload
1840 		 * the timer so that we will eventually get another
1841 		 * interrupt that will cause us to re-enter this routine.
1842 		 * This is done in case the transmitter has gone idle.
1843 		 */
1844 		CSR_WRITE_4(sc, RL_TIMERCNT, 1);
1845 #endif
1846 	} else
1847 		sc->rl_watchdog_timer = 0;
1848 }
1849 
1850 static void
1851 re_tick(xsc)
1852 	void			*xsc;
1853 {
1854 	struct rl_softc		*sc;
1855 	struct mii_data		*mii;
1856 	struct ifnet		*ifp;
1857 
1858 	sc = xsc;
1859 	ifp = sc->rl_ifp;
1860 
1861 	RL_LOCK_ASSERT(sc);
1862 
1863 	re_watchdog(sc);
1864 
1865 	mii = device_get_softc(sc->rl_miibus);
1866 	mii_tick(mii);
1867 	if (sc->rl_link) {
1868 		if (!(mii->mii_media_status & IFM_ACTIVE))
1869 			sc->rl_link = 0;
1870 	} else {
1871 		if (mii->mii_media_status & IFM_ACTIVE &&
1872 		    IFM_SUBTYPE(mii->mii_media_active) != IFM_NONE) {
1873 			sc->rl_link = 1;
1874 			if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
1875 				taskqueue_enqueue_fast(taskqueue_fast,
1876 				    &sc->rl_txtask);
1877 		}
1878 	}
1879 
1880 	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
1881 }
1882 
1883 #ifdef DEVICE_POLLING
1884 static void
1885 re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
1886 {
1887 	struct rl_softc *sc = ifp->if_softc;
1888 
1889 	RL_LOCK(sc);
1890 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1891 		re_poll_locked(ifp, cmd, count);
1892 	RL_UNLOCK(sc);
1893 }
1894 
1895 static void
1896 re_poll_locked(struct ifnet *ifp, enum poll_cmd cmd, int count)
1897 {
1898 	struct rl_softc *sc = ifp->if_softc;
1899 
1900 	RL_LOCK_ASSERT(sc);
1901 
1902 	sc->rxcycles = count;
1903 	re_rxeof(sc);
1904 	re_txeof(sc);
1905 
1906 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
1907 		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_txtask);
1908 
1909 	if (cmd == POLL_AND_CHECK_STATUS) { /* also check status register */
1910 		u_int16_t       status;
1911 
1912 		status = CSR_READ_2(sc, RL_ISR);
1913 		if (status == 0xffff)
1914 			return;
1915 		if (status)
1916 			CSR_WRITE_2(sc, RL_ISR, status);
1917 
1918 		/*
1919 		 * XXX check behaviour on receiver stalls.
1920 		 */
1921 
1922 		if (status & RL_ISR_SYSTEM_ERR) {
1923 			re_reset(sc);
1924 			re_init_locked(sc);
1925 		}
1926 	}
1927 }
1928 #endif /* DEVICE_POLLING */
1929 
1930 static void
1931 re_intr(arg)
1932 	void			*arg;
1933 {
1934 	struct rl_softc		*sc;
1935 	uint16_t		status;
1936 
1937 	sc = arg;
1938 
1939 	status = CSR_READ_2(sc, RL_ISR);
1940 	if (status == 0xFFFF || (status & RL_INTRS_CPLUS) == 0)
1941                 return;
1942 	CSR_WRITE_2(sc, RL_IMR, 0);
1943 
1944 	taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_inttask);
1945 
1946 	return;
1947 }
1948 
1949 static void
1950 re_int_task(arg, npending)
1951 	void			*arg;
1952 	int			npending;
1953 {
1954 	struct rl_softc		*sc;
1955 	struct ifnet		*ifp;
1956 	u_int16_t		status;
1957 	int			rval = 0;
1958 
1959 	sc = arg;
1960 	ifp = sc->rl_ifp;
1961 
1962 	RL_LOCK(sc);
1963 
1964 	status = CSR_READ_2(sc, RL_ISR);
1965         CSR_WRITE_2(sc, RL_ISR, status);
1966 
1967 	if (sc->suspended || !(ifp->if_flags & IFF_UP)) {
1968 		RL_UNLOCK(sc);
1969 		return;
1970 	}
1971 
1972 #ifdef DEVICE_POLLING
1973 	if  (ifp->if_capenable & IFCAP_POLLING) {
1974 		RL_UNLOCK(sc);
1975 		return;
1976 	}
1977 #endif
1978 
1979 	if (status & (RL_ISR_RX_OK|RL_ISR_RX_ERR|RL_ISR_FIFO_OFLOW))
1980 		rval = re_rxeof(sc);
1981 
1982 #ifdef RE_TX_MODERATION
1983 	if (status & (RL_ISR_TIMEOUT_EXPIRED|
1984 #else
1985 	if (status & (RL_ISR_TX_OK|
1986 #endif
1987 	    RL_ISR_TX_ERR|RL_ISR_TX_DESC_UNAVAIL))
1988 		re_txeof(sc);
1989 
1990 	if (status & RL_ISR_SYSTEM_ERR) {
1991 		re_reset(sc);
1992 		re_init_locked(sc);
1993 	}
1994 
1995 	if (status & RL_ISR_LINKCHG) {
1996 		callout_stop(&sc->rl_stat_callout);
1997 		re_tick(sc);
1998 	}
1999 
2000 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2001 		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_txtask);
2002 
2003 	RL_UNLOCK(sc);
2004 
2005         if ((CSR_READ_2(sc, RL_ISR) & RL_INTRS_CPLUS) || rval) {
2006 		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_inttask);
2007 		return;
2008 	}
2009 
2010 	CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2011 
2012 	return;
2013 }
2014 
2015 static int
2016 re_encap(sc, m_head, idx)
2017 	struct rl_softc		*sc;
2018 	struct mbuf		**m_head;
2019 	int			*idx;
2020 {
2021 	struct mbuf		*m_new = NULL;
2022 	struct rl_dmaload_arg	arg;
2023 	bus_dmamap_t		map;
2024 	int			error;
2025 
2026 	RL_LOCK_ASSERT(sc);
2027 
2028 	if (sc->rl_ldata.rl_tx_free <= RL_TX_DESC_THLD)
2029 		return (EFBIG);
2030 
2031 	/*
2032 	 * Set up checksum offload. Note: checksum offload bits must
2033 	 * appear in all descriptors of a multi-descriptor transmit
2034 	 * attempt. This is according to testing done with an 8169
2035 	 * chip. This is a requirement.
2036 	 */
2037 
2038 	arg.rl_flags = 0;
2039 
2040 	if (((*m_head)->m_pkthdr.csum_flags & CSUM_TSO) != 0)
2041 		arg.rl_flags = RL_TDESC_CMD_LGSEND |
2042 		    ((uint32_t)(*m_head)->m_pkthdr.tso_segsz <<
2043 		    RL_TDESC_CMD_MSSVAL_SHIFT);
2044 	else {
2045 		if ((*m_head)->m_pkthdr.csum_flags & CSUM_IP)
2046 			arg.rl_flags |= RL_TDESC_CMD_IPCSUM;
2047 		if ((*m_head)->m_pkthdr.csum_flags & CSUM_TCP)
2048 			arg.rl_flags |= RL_TDESC_CMD_TCPCSUM;
2049 		if ((*m_head)->m_pkthdr.csum_flags & CSUM_UDP)
2050 			arg.rl_flags |= RL_TDESC_CMD_UDPCSUM;
2051 	}
2052 
2053 	arg.rl_idx = *idx;
2054 	arg.rl_maxsegs = sc->rl_ldata.rl_tx_free;
2055 	if (arg.rl_maxsegs > RL_TX_DESC_THLD)
2056 		arg.rl_maxsegs -= RL_TX_DESC_THLD;
2057 	arg.rl_ring = sc->rl_ldata.rl_tx_list;
2058 
2059 	map = sc->rl_ldata.rl_tx_dmamap[*idx];
2060 
2061 	/*
2062 	 * With some of the RealTek chips, using the checksum offload
2063 	 * support in conjunction with the autopadding feature results
2064 	 * in the transmission of corrupt frames. For example, if we
2065 	 * need to send a really small IP fragment that's less than 60
2066 	 * bytes in size, and IP header checksumming is enabled, the
2067 	 * resulting ethernet frame that appears on the wire will
2068 	 * have garbled payload. To work around this, if TX checksum
2069 	 * offload is enabled, we always manually pad short frames out
2070 	 * to the minimum ethernet frame size. We do this by pretending
2071 	 * the mbuf chain has too many fragments so the coalescing code
2072 	 * below can assemble the packet into a single buffer that's
2073 	 * padded out to the mininum frame size.
2074 	 *
2075 	 * Note: this appears unnecessary for TCP, and doing it for TCP
2076 	 * with PCIe adapters seems to result in bad checksums.
2077 	 */
2078 
2079 	if (arg.rl_flags && !(arg.rl_flags & RL_TDESC_CMD_TCPCSUM) &&
2080             (*m_head)->m_pkthdr.len < RL_MIN_FRAMELEN)
2081 		error = EFBIG;
2082 	else
2083 		error = bus_dmamap_load_mbuf(sc->rl_ldata.rl_mtag, map,
2084 		    *m_head, re_dma_map_desc, &arg, BUS_DMA_NOWAIT);
2085 
2086 	if (error && error != EFBIG) {
2087 		device_printf(sc->rl_dev, "can't map mbuf (error %d)\n", error);
2088 		return (ENOBUFS);
2089 	}
2090 
2091 	/* Too many segments to map, coalesce into a single mbuf */
2092 
2093 	if (error || arg.rl_maxsegs == 0) {
2094 		if (arg.rl_maxsegs == 0)
2095 			bus_dmamap_unload(sc->rl_ldata.rl_mtag, map);
2096 		m_new = m_defrag(*m_head, M_DONTWAIT);
2097 		if (m_new == NULL) {
2098 			m_freem(*m_head);
2099 			*m_head = NULL;
2100 			return (ENOBUFS);
2101 		}
2102 		*m_head = m_new;
2103 
2104 		/*
2105 		 * Manually pad short frames, and zero the pad space
2106 		 * to avoid leaking data.
2107 		 */
2108 		if (m_new->m_pkthdr.len < RL_MIN_FRAMELEN) {
2109 			bzero(mtod(m_new, char *) + m_new->m_pkthdr.len,
2110 			    RL_MIN_FRAMELEN - m_new->m_pkthdr.len);
2111 			m_new->m_pkthdr.len += RL_MIN_FRAMELEN -
2112 			    m_new->m_pkthdr.len;
2113 			m_new->m_len = m_new->m_pkthdr.len;
2114 		}
2115 
2116 		/* Note that we'll run over RL_TX_DESC_THLD here. */
2117 		arg.rl_maxsegs = sc->rl_ldata.rl_tx_free;
2118 		error = bus_dmamap_load_mbuf(sc->rl_ldata.rl_mtag, map,
2119 		    *m_head, re_dma_map_desc, &arg, BUS_DMA_NOWAIT);
2120 		if (error || arg.rl_maxsegs == 0) {
2121 			device_printf(sc->rl_dev,
2122 			    "can't map defragmented mbuf (error %d)\n", error);
2123 			m_freem(m_new);
2124 			*m_head = NULL;
2125 			if (arg.rl_maxsegs == 0)
2126 				bus_dmamap_unload(sc->rl_ldata.rl_mtag, map);
2127 			return (EFBIG);
2128 		}
2129 	}
2130 
2131 	/*
2132 	 * Insure that the map for this transmission
2133 	 * is placed at the array index of the last descriptor
2134 	 * in this chain.  (Swap last and first dmamaps.)
2135 	 */
2136 	sc->rl_ldata.rl_tx_dmamap[*idx] =
2137 	    sc->rl_ldata.rl_tx_dmamap[arg.rl_idx];
2138 	sc->rl_ldata.rl_tx_dmamap[arg.rl_idx] = map;
2139 
2140 	sc->rl_ldata.rl_tx_mbuf[arg.rl_idx] = *m_head;
2141 	sc->rl_ldata.rl_tx_free -= arg.rl_maxsegs;
2142 
2143 	/*
2144 	 * Set up hardware VLAN tagging. Note: vlan tag info must
2145 	 * appear in the first descriptor of a multi-descriptor
2146 	 * transmission attempt.
2147 	 */
2148 	if ((*m_head)->m_flags & M_VLANTAG)
2149 		sc->rl_ldata.rl_tx_list[*idx].rl_vlanctl =
2150 		    htole32(htons((*m_head)->m_pkthdr.ether_vtag) |
2151 		    RL_TDESC_VLANCTL_TAG);
2152 
2153 	/* Transfer ownership of packet to the chip. */
2154 
2155 	sc->rl_ldata.rl_tx_list[arg.rl_idx].rl_cmdstat |=
2156 	    htole32(RL_TDESC_CMD_OWN);
2157 	if (*idx != arg.rl_idx)
2158 		sc->rl_ldata.rl_tx_list[*idx].rl_cmdstat |=
2159 		    htole32(RL_TDESC_CMD_OWN);
2160 
2161         RL_DESC_INC(arg.rl_idx);
2162 	*idx = arg.rl_idx;
2163 
2164 	return (0);
2165 }
2166 
2167 static void
2168 re_tx_task(arg, npending)
2169 	void			*arg;
2170 	int			npending;
2171 {
2172 	struct ifnet		*ifp;
2173 
2174 	ifp = arg;
2175 	re_start(ifp);
2176 
2177 	return;
2178 }
2179 
2180 /*
2181  * Main transmit routine for C+ and gigE NICs.
2182  */
2183 static void
2184 re_start(ifp)
2185 	struct ifnet		*ifp;
2186 {
2187 	struct rl_softc		*sc;
2188 	struct mbuf		*m_head = NULL;
2189 	int			idx, queued = 0;
2190 
2191 	sc = ifp->if_softc;
2192 
2193 	RL_LOCK(sc);
2194 
2195 	if (!sc->rl_link || ifp->if_drv_flags & IFF_DRV_OACTIVE) {
2196 		RL_UNLOCK(sc);
2197 		return;
2198 	}
2199 
2200 	idx = sc->rl_ldata.rl_tx_prodidx;
2201 
2202 	while (sc->rl_ldata.rl_tx_mbuf[idx] == NULL) {
2203 		IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
2204 		if (m_head == NULL)
2205 			break;
2206 
2207 		if (re_encap(sc, &m_head, &idx)) {
2208 			if (m_head == NULL)
2209 				break;
2210 			IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
2211 			ifp->if_drv_flags |= IFF_DRV_OACTIVE;
2212 			break;
2213 		}
2214 
2215 		/*
2216 		 * If there's a BPF listener, bounce a copy of this frame
2217 		 * to him.
2218 		 */
2219 		BPF_MTAP(ifp, m_head);
2220 
2221 		queued++;
2222 	}
2223 
2224 	if (queued == 0) {
2225 #ifdef RE_TX_MODERATION
2226 		if (sc->rl_ldata.rl_tx_free != RL_TX_DESC_CNT)
2227 			CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2228 #endif
2229 		RL_UNLOCK(sc);
2230 		return;
2231 	}
2232 
2233 	/* Flush the TX descriptors */
2234 
2235 	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2236 	    sc->rl_ldata.rl_tx_list_map,
2237 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2238 
2239 	sc->rl_ldata.rl_tx_prodidx = idx;
2240 
2241 	CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2242 
2243 #ifdef RE_TX_MODERATION
2244 	/*
2245 	 * Use the countdown timer for interrupt moderation.
2246 	 * 'TX done' interrupts are disabled. Instead, we reset the
2247 	 * countdown timer, which will begin counting until it hits
2248 	 * the value in the TIMERINT register, and then trigger an
2249 	 * interrupt. Each time we write to the TIMERCNT register,
2250 	 * the timer count is reset to 0.
2251 	 */
2252 	CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2253 #endif
2254 
2255 	/*
2256 	 * Set a timeout in case the chip goes out to lunch.
2257 	 */
2258 	sc->rl_watchdog_timer = 5;
2259 
2260 	RL_UNLOCK(sc);
2261 
2262 	return;
2263 }
2264 
2265 static void
2266 re_init(xsc)
2267 	void			*xsc;
2268 {
2269 	struct rl_softc		*sc = xsc;
2270 
2271 	RL_LOCK(sc);
2272 	re_init_locked(sc);
2273 	RL_UNLOCK(sc);
2274 }
2275 
2276 static void
2277 re_init_locked(sc)
2278 	struct rl_softc		*sc;
2279 {
2280 	struct ifnet		*ifp = sc->rl_ifp;
2281 	struct mii_data		*mii;
2282 	u_int32_t		rxcfg = 0;
2283 	union {
2284 		uint32_t align_dummy;
2285 		u_char eaddr[ETHER_ADDR_LEN];
2286         } eaddr;
2287 
2288 	RL_LOCK_ASSERT(sc);
2289 
2290 	mii = device_get_softc(sc->rl_miibus);
2291 
2292 	/*
2293 	 * Cancel pending I/O and free all RX/TX buffers.
2294 	 */
2295 	re_stop(sc);
2296 
2297 	/*
2298 	 * Enable C+ RX and TX mode, as well as VLAN stripping and
2299 	 * RX checksum offload. We must configure the C+ register
2300 	 * before all others.
2301 	 */
2302 	CSR_WRITE_2(sc, RL_CPLUS_CMD, RL_CPLUSCMD_RXENB|
2303 	    RL_CPLUSCMD_TXENB|RL_CPLUSCMD_PCI_MRW|
2304 	    RL_CPLUSCMD_VLANSTRIP|RL_CPLUSCMD_RXCSUM_ENB);
2305 
2306 	/*
2307 	 * Init our MAC address.  Even though the chipset
2308 	 * documentation doesn't mention it, we need to enter "Config
2309 	 * register write enable" mode to modify the ID registers.
2310 	 */
2311 	/* Copy MAC address on stack to align. */
2312 	bcopy(IF_LLADDR(ifp), eaddr.eaddr, ETHER_ADDR_LEN);
2313 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG);
2314 	CSR_WRITE_4(sc, RL_IDR0,
2315 	    htole32(*(u_int32_t *)(&eaddr.eaddr[0])));
2316 	CSR_WRITE_4(sc, RL_IDR4,
2317 	    htole32(*(u_int32_t *)(&eaddr.eaddr[4])));
2318 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
2319 
2320 	/*
2321 	 * For C+ mode, initialize the RX descriptors and mbufs.
2322 	 */
2323 	re_rx_list_init(sc);
2324 	re_tx_list_init(sc);
2325 
2326 	/*
2327 	 * Load the addresses of the RX and TX lists into the chip.
2328 	 */
2329 
2330 	CSR_WRITE_4(sc, RL_RXLIST_ADDR_HI,
2331 	    RL_ADDR_HI(sc->rl_ldata.rl_rx_list_addr));
2332 	CSR_WRITE_4(sc, RL_RXLIST_ADDR_LO,
2333 	    RL_ADDR_LO(sc->rl_ldata.rl_rx_list_addr));
2334 
2335 	CSR_WRITE_4(sc, RL_TXLIST_ADDR_HI,
2336 	    RL_ADDR_HI(sc->rl_ldata.rl_tx_list_addr));
2337 	CSR_WRITE_4(sc, RL_TXLIST_ADDR_LO,
2338 	    RL_ADDR_LO(sc->rl_ldata.rl_tx_list_addr));
2339 
2340 	/*
2341 	 * Enable transmit and receive.
2342 	 */
2343 	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
2344 
2345 	/*
2346 	 * Set the initial TX and RX configuration.
2347 	 */
2348 	if (sc->rl_testmode) {
2349 		if (sc->rl_type == RL_8169)
2350 			CSR_WRITE_4(sc, RL_TXCFG,
2351 			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON);
2352 		else
2353 			CSR_WRITE_4(sc, RL_TXCFG,
2354 			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON_CPLUS);
2355 	} else
2356 		CSR_WRITE_4(sc, RL_TXCFG, RL_TXCFG_CONFIG);
2357 
2358 	CSR_WRITE_1(sc, RL_EARLY_TX_THRESH, 16);
2359 
2360 	CSR_WRITE_4(sc, RL_RXCFG, RL_RXCFG_CONFIG);
2361 
2362 	/* Set the individual bit to receive frames for this host only. */
2363 	rxcfg = CSR_READ_4(sc, RL_RXCFG);
2364 	rxcfg |= RL_RXCFG_RX_INDIV;
2365 
2366 	/* If we want promiscuous mode, set the allframes bit. */
2367 	if (ifp->if_flags & IFF_PROMISC)
2368 		rxcfg |= RL_RXCFG_RX_ALLPHYS;
2369 	else
2370 		rxcfg &= ~RL_RXCFG_RX_ALLPHYS;
2371 	CSR_WRITE_4(sc, RL_RXCFG, rxcfg);
2372 
2373 	/*
2374 	 * Set capture broadcast bit to capture broadcast frames.
2375 	 */
2376 	if (ifp->if_flags & IFF_BROADCAST)
2377 		rxcfg |= RL_RXCFG_RX_BROAD;
2378 	else
2379 		rxcfg &= ~RL_RXCFG_RX_BROAD;
2380 	CSR_WRITE_4(sc, RL_RXCFG, rxcfg);
2381 
2382 	/*
2383 	 * Program the multicast filter, if necessary.
2384 	 */
2385 	re_setmulti(sc);
2386 
2387 #ifdef DEVICE_POLLING
2388 	/*
2389 	 * Disable interrupts if we are polling.
2390 	 */
2391 	if (ifp->if_capenable & IFCAP_POLLING)
2392 		CSR_WRITE_2(sc, RL_IMR, 0);
2393 	else	/* otherwise ... */
2394 #endif
2395 
2396 	/*
2397 	 * Enable interrupts.
2398 	 */
2399 	if (sc->rl_testmode)
2400 		CSR_WRITE_2(sc, RL_IMR, 0);
2401 	else
2402 		CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2403 	CSR_WRITE_2(sc, RL_ISR, RL_INTRS_CPLUS);
2404 
2405 	/* Set initial TX threshold */
2406 	sc->rl_txthresh = RL_TX_THRESH_INIT;
2407 
2408 	/* Start RX/TX process. */
2409 	CSR_WRITE_4(sc, RL_MISSEDPKT, 0);
2410 #ifdef notdef
2411 	/* Enable receiver and transmitter. */
2412 	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
2413 #endif
2414 
2415 #ifdef RE_TX_MODERATION
2416 	/*
2417 	 * Initialize the timer interrupt register so that
2418 	 * a timer interrupt will be generated once the timer
2419 	 * reaches a certain number of ticks. The timer is
2420 	 * reloaded on each transmit. This gives us TX interrupt
2421 	 * moderation, which dramatically improves TX frame rate.
2422 	 */
2423 	if (sc->rl_type == RL_8169)
2424 		CSR_WRITE_4(sc, RL_TIMERINT_8169, 0x800);
2425 	else
2426 		CSR_WRITE_4(sc, RL_TIMERINT, 0x400);
2427 #endif
2428 
2429 	/*
2430 	 * For 8169 gigE NICs, set the max allowed RX packet
2431 	 * size so we can receive jumbo frames.
2432 	 */
2433 	if (sc->rl_type == RL_8169)
2434 		CSR_WRITE_2(sc, RL_MAXRXPKTLEN, 16383);
2435 
2436 	if (sc->rl_testmode)
2437 		return;
2438 
2439 	mii_mediachg(mii);
2440 
2441 	CSR_WRITE_1(sc, RL_CFG1, CSR_READ_1(sc, RL_CFG1) | RL_CFG1_DRVLOAD);
2442 
2443 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
2444 	ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
2445 
2446 	sc->rl_link = 0;
2447 	sc->rl_watchdog_timer = 0;
2448 	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
2449 }
2450 
2451 /*
2452  * Set media options.
2453  */
2454 static int
2455 re_ifmedia_upd(ifp)
2456 	struct ifnet		*ifp;
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 	RL_LOCK(sc);
2464 	mii_mediachg(mii);
2465 	RL_UNLOCK(sc);
2466 
2467 	return (0);
2468 }
2469 
2470 /*
2471  * Report current media status.
2472  */
2473 static void
2474 re_ifmedia_sts(ifp, ifmr)
2475 	struct ifnet		*ifp;
2476 	struct ifmediareq	*ifmr;
2477 {
2478 	struct rl_softc		*sc;
2479 	struct mii_data		*mii;
2480 
2481 	sc = ifp->if_softc;
2482 	mii = device_get_softc(sc->rl_miibus);
2483 
2484 	RL_LOCK(sc);
2485 	mii_pollstat(mii);
2486 	RL_UNLOCK(sc);
2487 	ifmr->ifm_active = mii->mii_media_active;
2488 	ifmr->ifm_status = mii->mii_media_status;
2489 }
2490 
2491 static int
2492 re_ioctl(ifp, command, data)
2493 	struct ifnet		*ifp;
2494 	u_long			command;
2495 	caddr_t			data;
2496 {
2497 	struct rl_softc		*sc = ifp->if_softc;
2498 	struct ifreq		*ifr = (struct ifreq *) data;
2499 	struct mii_data		*mii;
2500 	int			error = 0;
2501 
2502 	switch (command) {
2503 	case SIOCSIFMTU:
2504 		RL_LOCK(sc);
2505 		if (ifr->ifr_mtu > RL_JUMBO_MTU)
2506 			error = EINVAL;
2507 		ifp->if_mtu = ifr->ifr_mtu;
2508 		RL_UNLOCK(sc);
2509 		break;
2510 	case SIOCSIFFLAGS:
2511 		RL_LOCK(sc);
2512 		if (ifp->if_flags & IFF_UP)
2513 			re_init_locked(sc);
2514 		else if (ifp->if_drv_flags & IFF_DRV_RUNNING)
2515 			re_stop(sc);
2516 		RL_UNLOCK(sc);
2517 		break;
2518 	case SIOCADDMULTI:
2519 	case SIOCDELMULTI:
2520 		RL_LOCK(sc);
2521 		re_setmulti(sc);
2522 		RL_UNLOCK(sc);
2523 		break;
2524 	case SIOCGIFMEDIA:
2525 	case SIOCSIFMEDIA:
2526 		mii = device_get_softc(sc->rl_miibus);
2527 		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, command);
2528 		break;
2529 	case SIOCSIFCAP:
2530 	    {
2531 		int mask, reinit;
2532 
2533 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
2534 		reinit = 0;
2535 #ifdef DEVICE_POLLING
2536 		if (mask & IFCAP_POLLING) {
2537 			if (ifr->ifr_reqcap & IFCAP_POLLING) {
2538 				error = ether_poll_register(re_poll, ifp);
2539 				if (error)
2540 					return(error);
2541 				RL_LOCK(sc);
2542 				/* Disable interrupts */
2543 				CSR_WRITE_2(sc, RL_IMR, 0x0000);
2544 				ifp->if_capenable |= IFCAP_POLLING;
2545 				RL_UNLOCK(sc);
2546 			} else {
2547 				error = ether_poll_deregister(ifp);
2548 				/* Enable interrupts. */
2549 				RL_LOCK(sc);
2550 				CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2551 				ifp->if_capenable &= ~IFCAP_POLLING;
2552 				RL_UNLOCK(sc);
2553 			}
2554 		}
2555 #endif /* DEVICE_POLLING */
2556 		if (mask & IFCAP_HWCSUM) {
2557 			ifp->if_capenable ^= IFCAP_HWCSUM;
2558 			if (ifp->if_capenable & IFCAP_TXCSUM)
2559 				ifp->if_hwassist |= RE_CSUM_FEATURES;
2560 			else
2561 				ifp->if_hwassist &= ~RE_CSUM_FEATURES;
2562 			reinit = 1;
2563 		}
2564 		if (mask & IFCAP_VLAN_HWTAGGING) {
2565 			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
2566 			reinit = 1;
2567 		}
2568 		if (mask & IFCAP_TSO4) {
2569 			ifp->if_capenable ^= IFCAP_TSO4;
2570 			if ((IFCAP_TSO4 & ifp->if_capenable) &&
2571 			    (IFCAP_TSO4 & ifp->if_capabilities))
2572 				ifp->if_hwassist |= CSUM_TSO;
2573 			else
2574 				ifp->if_hwassist &= ~CSUM_TSO;
2575 		}
2576 		if (reinit && ifp->if_drv_flags & IFF_DRV_RUNNING)
2577 			re_init(sc);
2578 		VLAN_CAPABILITIES(ifp);
2579 	    }
2580 		break;
2581 	default:
2582 		error = ether_ioctl(ifp, command, data);
2583 		break;
2584 	}
2585 
2586 	return (error);
2587 }
2588 
2589 static void
2590 re_watchdog(sc)
2591 	struct rl_softc		*sc;
2592 {
2593 
2594 	RL_LOCK_ASSERT(sc);
2595 
2596 	if (sc->rl_watchdog_timer == 0 || --sc->rl_watchdog_timer != 0)
2597 		return;
2598 
2599 	device_printf(sc->rl_dev, "watchdog timeout\n");
2600 	sc->rl_ifp->if_oerrors++;
2601 
2602 	re_txeof(sc);
2603 	re_rxeof(sc);
2604 	re_init_locked(sc);
2605 }
2606 
2607 /*
2608  * Stop the adapter and free any mbufs allocated to the
2609  * RX and TX lists.
2610  */
2611 static void
2612 re_stop(sc)
2613 	struct rl_softc		*sc;
2614 {
2615 	register int		i;
2616 	struct ifnet		*ifp;
2617 
2618 	RL_LOCK_ASSERT(sc);
2619 
2620 	ifp = sc->rl_ifp;
2621 
2622 	sc->rl_watchdog_timer = 0;
2623 	callout_stop(&sc->rl_stat_callout);
2624 	ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
2625 
2626 	CSR_WRITE_1(sc, RL_COMMAND, 0x00);
2627 	CSR_WRITE_2(sc, RL_IMR, 0x0000);
2628 	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
2629 
2630 	if (sc->rl_head != NULL) {
2631 		m_freem(sc->rl_head);
2632 		sc->rl_head = sc->rl_tail = NULL;
2633 	}
2634 
2635 	/* Free the TX list buffers. */
2636 
2637 	for (i = 0; i < RL_TX_DESC_CNT; i++) {
2638 		if (sc->rl_ldata.rl_tx_mbuf[i] != NULL) {
2639 			bus_dmamap_unload(sc->rl_ldata.rl_mtag,
2640 			    sc->rl_ldata.rl_tx_dmamap[i]);
2641 			m_freem(sc->rl_ldata.rl_tx_mbuf[i]);
2642 			sc->rl_ldata.rl_tx_mbuf[i] = NULL;
2643 		}
2644 	}
2645 
2646 	/* Free the RX list buffers. */
2647 
2648 	for (i = 0; i < RL_RX_DESC_CNT; i++) {
2649 		if (sc->rl_ldata.rl_rx_mbuf[i] != NULL) {
2650 			bus_dmamap_unload(sc->rl_ldata.rl_mtag,
2651 			    sc->rl_ldata.rl_rx_dmamap[i]);
2652 			m_freem(sc->rl_ldata.rl_rx_mbuf[i]);
2653 			sc->rl_ldata.rl_rx_mbuf[i] = NULL;
2654 		}
2655 	}
2656 }
2657 
2658 /*
2659  * Device suspend routine.  Stop the interface and save some PCI
2660  * settings in case the BIOS doesn't restore them properly on
2661  * resume.
2662  */
2663 static int
2664 re_suspend(dev)
2665 	device_t		dev;
2666 {
2667 	struct rl_softc		*sc;
2668 
2669 	sc = device_get_softc(dev);
2670 
2671 	RL_LOCK(sc);
2672 	re_stop(sc);
2673 	sc->suspended = 1;
2674 	RL_UNLOCK(sc);
2675 
2676 	return (0);
2677 }
2678 
2679 /*
2680  * Device resume routine.  Restore some PCI settings in case the BIOS
2681  * doesn't, re-enable busmastering, and restart the interface if
2682  * appropriate.
2683  */
2684 static int
2685 re_resume(dev)
2686 	device_t		dev;
2687 {
2688 	struct rl_softc		*sc;
2689 	struct ifnet		*ifp;
2690 
2691 	sc = device_get_softc(dev);
2692 
2693 	RL_LOCK(sc);
2694 
2695 	ifp = sc->rl_ifp;
2696 
2697 	/* reinitialize interface if necessary */
2698 	if (ifp->if_flags & IFF_UP)
2699 		re_init_locked(sc);
2700 
2701 	sc->suspended = 0;
2702 	RL_UNLOCK(sc);
2703 
2704 	return (0);
2705 }
2706 
2707 /*
2708  * Stop all chip I/O so that the kernel's probe routines don't
2709  * get confused by errant DMAs when rebooting.
2710  */
2711 static void
2712 re_shutdown(dev)
2713 	device_t		dev;
2714 {
2715 	struct rl_softc		*sc;
2716 
2717 	sc = device_get_softc(dev);
2718 
2719 	RL_LOCK(sc);
2720 	re_stop(sc);
2721 	/*
2722 	 * Mark interface as down since otherwise we will panic if
2723 	 * interrupt comes in later on, which can happen in some
2724 	 * cases.
2725 	 */
2726 	sc->rl_ifp->if_flags &= ~IFF_UP;
2727 	RL_UNLOCK(sc);
2728 }
2729