xref: /dflybsd-src/sys/dev/netif/re/if_re.c (revision 9b42cabedff59ea3b351e8eccb2da9bab5cefe7e)
1 /*
2  * Copyright (c) 2004
3  *	Joerg Sonnenberger <joerg@bec.de>.  All rights reserved.
4  *
5  * Copyright (c) 1997, 1998-2003
6  *	Bill Paul <wpaul@windriver.com>.  All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. All advertising materials mentioning features or use of this software
17  *    must display the following acknowledgement:
18  *	This product includes software developed by Bill Paul.
19  * 4. Neither the name of the author nor the names of any co-contributors
20  *    may be used to endorse or promote products derived from this software
21  *    without specific prior written permission.
22  *
23  * THIS SOFTWARE IS PROVIDED BY Bill Paul AND CONTRIBUTORS ``AS IS'' AND
24  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26  * ARE DISCLAIMED.  IN NO EVENT SHALL Bill Paul OR THE VOICES IN HIS HEAD
27  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
28  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
29  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
30  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
31  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
32  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
33  * THE POSSIBILITY OF SUCH DAMAGE.
34  *
35  * $FreeBSD: src/sys/dev/re/if_re.c,v 1.25 2004/06/09 14:34:01 naddy Exp $
36  * $DragonFly: src/sys/dev/netif/re/if_re.c,v 1.44 2008/06/25 11:02:33 sephe Exp $
37  */
38 
39 /*
40  * RealTek 8139C+/8169/8169S/8110S/8168/8111/8101E PCI NIC driver
41  *
42  * Written by Bill Paul <wpaul@windriver.com>
43  * Senior Networking Software Engineer
44  * Wind River Systems
45  */
46 
47 /*
48  * This driver is designed to support RealTek's next generation of
49  * 10/100 and 10/100/1000 PCI ethernet controllers. There are currently
50  * seven devices in this family: the RTL8139C+, the RTL8169, the RTL8169S,
51  * RTL8110S, the RTL8168, the RTL8111 and the RTL8101E.
52  *
53  * The 8139C+ is a 10/100 ethernet chip. It is backwards compatible
54  * with the older 8139 family, however it also supports a special
55  * C+ mode of operation that provides several new performance enhancing
56  * features. These include:
57  *
58  *	o Descriptor based DMA mechanism. Each descriptor represents
59  *	  a single packet fragment. Data buffers may be aligned on
60  *	  any byte boundary.
61  *
62  *	o 64-bit DMA
63  *
64  *	o TCP/IP checksum offload for both RX and TX
65  *
66  *	o High and normal priority transmit DMA rings
67  *
68  *	o VLAN tag insertion and extraction
69  *
70  *	o TCP large send (segmentation offload)
71  *
72  * Like the 8139, the 8139C+ also has a built-in 10/100 PHY. The C+
73  * programming API is fairly straightforward. The RX filtering, EEPROM
74  * access and PHY access is the same as it is on the older 8139 series
75  * chips.
76  *
77  * The 8169 is a 64-bit 10/100/1000 gigabit ethernet MAC. It has almost the
78  * same programming API and feature set as the 8139C+ with the following
79  * differences and additions:
80  *
81  *	o 1000Mbps mode
82  *
83  *	o Jumbo frames
84  *
85  * 	o GMII and TBI ports/registers for interfacing with copper
86  *	  or fiber PHYs
87  *
88  *      o RX and TX DMA rings can have up to 1024 descriptors
89  *        (the 8139C+ allows a maximum of 64)
90  *
91  *	o Slight differences in register layout from the 8139C+
92  *
93  * The TX start and timer interrupt registers are at different locations
94  * on the 8169 than they are on the 8139C+. Also, the status word in the
95  * RX descriptor has a slightly different bit layout. The 8169 does not
96  * have a built-in PHY. Most reference boards use a Marvell 88E1000 'Alaska'
97  * copper gigE PHY.
98  *
99  * The 8169S/8110S 10/100/1000 devices have built-in copper gigE PHYs
100  * (the 'S' stands for 'single-chip'). These devices have the same
101  * programming API as the older 8169, but also have some vendor-specific
102  * registers for the on-board PHY. The 8110S is a LAN-on-motherboard
103  * part designed to be pin-compatible with the RealTek 8100 10/100 chip.
104  *
105  * This driver takes advantage of the RX and TX checksum offload and
106  * VLAN tag insertion/extraction features. It also implements TX
107  * interrupt moderation using the timer interrupt registers, which
108  * significantly reduces TX interrupt load. There is also support
109  * for jumbo frames, however the 8169/8169S/8110S can not transmit
110  * jumbo frames larger than 7440, so the max MTU possible with this
111  * driver is 7422 bytes.
112  */
113 
114 #include "opt_polling.h"
115 #include "opt_ethernet.h"
116 
117 #include <sys/param.h>
118 #include <sys/bus.h>
119 #include <sys/endian.h>
120 #include <sys/kernel.h>
121 #include <sys/interrupt.h>
122 #include <sys/malloc.h>
123 #include <sys/mbuf.h>
124 #include <sys/rman.h>
125 #include <sys/serialize.h>
126 #include <sys/socket.h>
127 #include <sys/sockio.h>
128 #include <sys/sysctl.h>
129 
130 #include <net/bpf.h>
131 #include <net/ethernet.h>
132 #include <net/if.h>
133 #include <net/ifq_var.h>
134 #include <net/if_arp.h>
135 #include <net/if_dl.h>
136 #include <net/if_media.h>
137 #include <net/if_types.h>
138 #include <net/vlan/if_vlan_var.h>
139 #include <net/vlan/if_vlan_ether.h>
140 
141 #include <dev/netif/mii_layer/mii.h>
142 #include <dev/netif/mii_layer/miivar.h>
143 
144 #include <bus/pci/pcidevs.h>
145 #include <bus/pci/pcireg.h>
146 #include <bus/pci/pcivar.h>
147 
148 /* "device miibus" required.  See GENERIC if you get errors here. */
149 #include "miibus_if.h"
150 
151 #include <dev/netif/re/if_rereg.h>
152 #include <dev/netif/re/if_revar.h>
153 
154 #define RE_CSUM_FEATURES    (CSUM_IP | CSUM_TCP | CSUM_UDP)
155 #if 0
156 #define RE_DISABLE_HWCSUM
157 #endif
158 
159 /*
160  * Various supported device vendors/types and their names.
161  */
162 static const struct re_type re_devs[] = {
163 	{ PCI_VENDOR_DLINK, PCI_PRODUCT_DLINK_DGE528T, RE_HWREV_8169S,
164 		"D-Link DGE-528(T) Gigabit Ethernet Adapter" },
165 	{ PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8139, RE_HWREV_8139CPLUS,
166 		"RealTek 8139C+ 10/100BaseTX" },
167 	{ PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8101E, RE_HWREV_8101E,
168 		"RealTek 8101E PCIe 10/100baseTX" },
169 	{ PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8168, RE_HWREV_8168_SPIN1,
170 		"RealTek 8168/8111B PCIe Gigabit Ethernet" },
171 	{ PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8168, RE_HWREV_8168_SPIN2,
172 		"RealTek 8168/8111B PCIe Gigabit Ethernet" },
173 	{ PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8168, RE_HWREV_8168_SPIN3,
174 		"RealTek 8168B/8111B PCIe Gigabit Ethernet" },
175 	{ PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8168, RE_HWREV_8168C,
176 		"RealTek 8168C/8111C PCIe Gigabit Ethernet" },
177 	{ PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8169, RE_HWREV_8169,
178 		"RealTek 8169 Gigabit Ethernet" },
179 	{ PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8169, RE_HWREV_8169S,
180 		"RealTek 8169S Single-chip Gigabit Ethernet" },
181 	{ PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8169, RE_HWREV_8169_8110SB,
182 		"RealTek 8169SB/8110SB Single-chip Gigabit Ethernet" },
183 	{ PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8169, RE_HWREV_8169_8110SC,
184 		"RealTek 8169SC/8110SC Single-chip Gigabit Ethernet" },
185 	{ PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8169SC, RE_HWREV_8169_8110SC,
186 		"RealTek 8169SC/8110SC Single-chip Gigabit Ethernet" },
187 	{ PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8169, RE_HWREV_8110S,
188 		"RealTek 8110S Single-chip Gigabit Ethernet" },
189 	{ PCI_VENDOR_COREGA, PCI_PRODUCT_COREGA_CG_LAPCIGT, RE_HWREV_8169S,
190 		"Corega CG-LAPCIGT Gigabit Ethernet" },
191 	{ PCI_VENDOR_LINKSYS, PCI_PRODUCT_LINKSYS_EG1032, RE_HWREV_8169S,
192 		"Linksys EG1032 Gigabit Ethernet" },
193 	{ PCI_VENDOR_USR2, PCI_PRODUCT_USR2_997902, RE_HWREV_8169S,
194 		"US Robotics 997902 Gigabit Ethernet" },
195 	{ 0, 0, 0, NULL }
196 };
197 
198 static const struct re_hwrev re_hwrevs[] = {
199 	{ RE_HWREV_8139CPLUS,	RE_8139CPLUS,	RE_F_HASMPC,	"C+" },
200 	{ RE_HWREV_8168_SPIN1,	RE_8169,	RE_F_PCIE,	"8168" },
201 	{ RE_HWREV_8168_SPIN2,	RE_8169,	RE_F_PCIE,	"8168" },
202 	{ RE_HWREV_8168_SPIN3,	RE_8169,	RE_F_PCIE,	"8168" },
203 	{ RE_HWREV_8168C,	RE_8169,	RE_F_PCIE,	"8168C" },
204 	{ RE_HWREV_8169,	RE_8169,	RE_F_HASMPC,	"8169" },
205 	{ RE_HWREV_8169S,	RE_8169,	RE_F_HASMPC,	"8169S" },
206 	{ RE_HWREV_8110S,	RE_8169,	RE_F_HASMPC,	"8110S" },
207 	{ RE_HWREV_8169_8110SB,	RE_8169,	RE_F_HASMPC,	"8169SB" },
208 	{ RE_HWREV_8169_8110SC,	RE_8169,	0,		"8169SC" },
209 	{ RE_HWREV_8100E,	RE_8169,	RE_F_HASMPC,	"8100E" },
210 	{ RE_HWREV_8101E,	RE_8169,	RE_F_PCIE,	"8101E" },
211 	{ 0, 0, 0, NULL }
212 };
213 
214 static int	re_probe(device_t);
215 static int	re_attach(device_t);
216 static int	re_detach(device_t);
217 
218 static int	re_encap(struct re_softc *, struct mbuf **, int *, int *);
219 
220 static void	re_dma_map_addr(void *, bus_dma_segment_t *, int, int);
221 static void	re_dma_map_desc(void *, bus_dma_segment_t *, int,
222 				bus_size_t, int);
223 static int	re_allocmem(device_t, struct re_softc *);
224 static int	re_newbuf(struct re_softc *, int, struct mbuf *);
225 static int	re_rx_list_init(struct re_softc *);
226 static int	re_tx_list_init(struct re_softc *);
227 static void	re_rxeof(struct re_softc *);
228 static void	re_txeof(struct re_softc *);
229 static void	re_intr(void *);
230 static void	re_tick(void *);
231 static void	re_tick_serialized(void *);
232 static void	re_start(struct ifnet *);
233 static int	re_ioctl(struct ifnet *, u_long, caddr_t, struct ucred *);
234 static void	re_init(void *);
235 static void	re_stop(struct re_softc *);
236 static void	re_watchdog(struct ifnet *);
237 static int	re_suspend(device_t);
238 static int	re_resume(device_t);
239 static void	re_shutdown(device_t);
240 static int	re_ifmedia_upd(struct ifnet *);
241 static void	re_ifmedia_sts(struct ifnet *, struct ifmediareq *);
242 
243 static void	re_eeprom_putbyte(struct re_softc *, int);
244 static void	re_eeprom_getword(struct re_softc *, int, u_int16_t *);
245 static void	re_read_eeprom(struct re_softc *, caddr_t, int, int);
246 static int	re_gmii_readreg(device_t, int, int);
247 static int	re_gmii_writereg(device_t, int, int, int);
248 
249 static int	re_miibus_readreg(device_t, int, int);
250 static int	re_miibus_writereg(device_t, int, int, int);
251 static void	re_miibus_statchg(device_t);
252 
253 static void	re_setmulti(struct re_softc *);
254 static void	re_reset(struct re_softc *);
255 
256 #ifdef RE_DIAG
257 static int	re_diag(struct re_softc *);
258 #endif
259 
260 #ifdef DEVICE_POLLING
261 static void	re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count);
262 #endif
263 
264 static int	re_sysctl_tx_moderation(SYSCTL_HANDLER_ARGS);
265 
266 static device_method_t re_methods[] = {
267 	/* Device interface */
268 	DEVMETHOD(device_probe,		re_probe),
269 	DEVMETHOD(device_attach,	re_attach),
270 	DEVMETHOD(device_detach,	re_detach),
271 	DEVMETHOD(device_suspend,	re_suspend),
272 	DEVMETHOD(device_resume,	re_resume),
273 	DEVMETHOD(device_shutdown,	re_shutdown),
274 
275 	/* bus interface */
276 	DEVMETHOD(bus_print_child,	bus_generic_print_child),
277 	DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
278 
279 	/* MII interface */
280 	DEVMETHOD(miibus_readreg,	re_miibus_readreg),
281 	DEVMETHOD(miibus_writereg,	re_miibus_writereg),
282 	DEVMETHOD(miibus_statchg,	re_miibus_statchg),
283 
284 	{ 0, 0 }
285 };
286 
287 static driver_t re_driver = {
288 	"re",
289 	re_methods,
290 	sizeof(struct re_softc)
291 };
292 
293 static devclass_t re_devclass;
294 
295 DECLARE_DUMMY_MODULE(if_re);
296 DRIVER_MODULE(if_re, pci, re_driver, re_devclass, 0, 0);
297 DRIVER_MODULE(if_re, cardbus, re_driver, re_devclass, 0, 0);
298 DRIVER_MODULE(miibus, re, miibus_driver, miibus_devclass, 0, 0);
299 
300 #define EE_SET(x)	\
301 	CSR_WRITE_1(sc, RE_EECMD, CSR_READ_1(sc, RE_EECMD) | (x))
302 
303 #define EE_CLR(x)	\
304 	CSR_WRITE_1(sc, RE_EECMD, CSR_READ_1(sc, RE_EECMD) & ~(x))
305 
306 /*
307  * Send a read command and address to the EEPROM, check for ACK.
308  */
309 static void
310 re_eeprom_putbyte(struct re_softc *sc, int addr)
311 {
312 	int d, i;
313 
314 	d = addr | (RE_9346_READ << sc->re_eewidth);
315 
316 	/*
317 	 * Feed in each bit and strobe the clock.
318 	 */
319 	for (i = 1 << (sc->re_eewidth + 3); i; i >>= 1) {
320 		if (d & i)
321 			EE_SET(RE_EE_DATAIN);
322 		else
323 			EE_CLR(RE_EE_DATAIN);
324 		DELAY(100);
325 		EE_SET(RE_EE_CLK);
326 		DELAY(150);
327 		EE_CLR(RE_EE_CLK);
328 		DELAY(100);
329 	}
330 }
331 
332 /*
333  * Read a word of data stored in the EEPROM at address 'addr.'
334  */
335 static void
336 re_eeprom_getword(struct re_softc *sc, int addr, uint16_t *dest)
337 {
338 	int i;
339 	uint16_t word = 0;
340 
341 	/*
342 	 * Send address of word we want to read.
343 	 */
344 	re_eeprom_putbyte(sc, addr);
345 
346 	/*
347 	 * Start reading bits from EEPROM.
348 	 */
349 	for (i = 0x8000; i != 0; i >>= 1) {
350 		EE_SET(RE_EE_CLK);
351 		DELAY(100);
352 		if (CSR_READ_1(sc, RE_EECMD) & RE_EE_DATAOUT)
353 			word |= i;
354 		EE_CLR(RE_EE_CLK);
355 		DELAY(100);
356 	}
357 
358 	*dest = word;
359 }
360 
361 /*
362  * Read a sequence of words from the EEPROM.
363  */
364 static void
365 re_read_eeprom(struct re_softc *sc, caddr_t dest, int off, int cnt)
366 {
367 	int i;
368 	uint16_t word = 0, *ptr;
369 
370 	CSR_SETBIT_1(sc, RE_EECMD, RE_EEMODE_PROGRAM);
371 	DELAY(100);
372 
373 	for (i = 0; i < cnt; i++) {
374 		CSR_SETBIT_1(sc, RE_EECMD, RE_EE_SEL);
375 		re_eeprom_getword(sc, off + i, &word);
376 		CSR_CLRBIT_1(sc, RE_EECMD, RE_EE_SEL);
377 		ptr = (uint16_t *)(dest + (i * 2));
378 		*ptr = word;
379 	}
380 
381 	CSR_CLRBIT_1(sc, RE_EECMD, RE_EEMODE_PROGRAM);
382 }
383 
384 static int
385 re_gmii_readreg(device_t dev, int phy, int reg)
386 {
387 	struct re_softc *sc = device_get_softc(dev);
388 	u_int32_t rval;
389 	int i;
390 
391 	if (phy != 1)
392 		return(0);
393 
394 	/* Let the rgephy driver read the GMEDIASTAT register */
395 
396 	if (reg == RE_GMEDIASTAT)
397 		return(CSR_READ_1(sc, RE_GMEDIASTAT));
398 
399 	CSR_WRITE_4(sc, RE_PHYAR, reg << 16);
400 	DELAY(1000);
401 
402 	for (i = 0; i < RE_TIMEOUT; i++) {
403 		rval = CSR_READ_4(sc, RE_PHYAR);
404 		if (rval & RE_PHYAR_BUSY)
405 			break;
406 		DELAY(100);
407 	}
408 
409 	if (i == RE_TIMEOUT) {
410 		device_printf(dev, "PHY read failed\n");
411 		return(0);
412 	}
413 
414 	return(rval & RE_PHYAR_PHYDATA);
415 }
416 
417 static int
418 re_gmii_writereg(device_t dev, int phy, int reg, int data)
419 {
420 	struct re_softc *sc = device_get_softc(dev);
421 	uint32_t rval;
422 	int i;
423 
424 	CSR_WRITE_4(sc, RE_PHYAR,
425 		    (reg << 16) | (data & RE_PHYAR_PHYDATA) | RE_PHYAR_BUSY);
426 	DELAY(1000);
427 
428 	for (i = 0; i < RE_TIMEOUT; i++) {
429 		rval = CSR_READ_4(sc, RE_PHYAR);
430 		if ((rval & RE_PHYAR_BUSY) == 0)
431 			break;
432 		DELAY(100);
433 	}
434 
435 	if (i == RE_TIMEOUT)
436 		device_printf(dev, "PHY write failed\n");
437 
438 	return(0);
439 }
440 
441 static int
442 re_miibus_readreg(device_t dev, int phy, int reg)
443 {
444 	struct re_softc	*sc = device_get_softc(dev);
445 	uint16_t rval = 0;
446 	uint16_t re8139_reg = 0;
447 
448 	if (sc->re_type == RE_8169) {
449 		rval = re_gmii_readreg(dev, phy, reg);
450 		return(rval);
451 	}
452 
453 	/* Pretend the internal PHY is only at address 0 */
454 	if (phy)
455 		return(0);
456 
457 	switch(reg) {
458 	case MII_BMCR:
459 		re8139_reg = RE_BMCR;
460 		break;
461 	case MII_BMSR:
462 		re8139_reg = RE_BMSR;
463 		break;
464 	case MII_ANAR:
465 		re8139_reg = RE_ANAR;
466 		break;
467 	case MII_ANER:
468 		re8139_reg = RE_ANER;
469 		break;
470 	case MII_ANLPAR:
471 		re8139_reg = RE_LPAR;
472 		break;
473 	case MII_PHYIDR1:
474 	case MII_PHYIDR2:
475 		return(0);
476 	/*
477 	 * Allow the rlphy driver to read the media status
478 	 * register. If we have a link partner which does not
479 	 * support NWAY, this is the register which will tell
480 	 * us the results of parallel detection.
481 	 */
482 	case RE_MEDIASTAT:
483 		return(CSR_READ_1(sc, RE_MEDIASTAT));
484 	default:
485 		device_printf(dev, "bad phy register\n");
486 		return(0);
487 	}
488 	rval = CSR_READ_2(sc, re8139_reg);
489 	if (sc->re_type == RE_8139CPLUS && re8139_reg == RE_BMCR) {
490 		/* 8139C+ has different bit layout. */
491 		rval &= ~(BMCR_LOOP | BMCR_ISO);
492 	}
493 	return(rval);
494 }
495 
496 static int
497 re_miibus_writereg(device_t dev, int phy, int reg, int data)
498 {
499 	struct re_softc *sc= device_get_softc(dev);
500 	u_int16_t re8139_reg = 0;
501 
502 	if (sc->re_type == RE_8169)
503 		return(re_gmii_writereg(dev, phy, reg, data));
504 
505 	/* Pretend the internal PHY is only at address 0 */
506 	if (phy)
507 		return(0);
508 
509 	switch(reg) {
510 	case MII_BMCR:
511 		re8139_reg = RE_BMCR;
512 		if (sc->re_type == RE_8139CPLUS) {
513 			/* 8139C+ has different bit layout. */
514 			data &= ~(BMCR_LOOP | BMCR_ISO);
515 		}
516 		break;
517 	case MII_BMSR:
518 		re8139_reg = RE_BMSR;
519 		break;
520 	case MII_ANAR:
521 		re8139_reg = RE_ANAR;
522 		break;
523 	case MII_ANER:
524 		re8139_reg = RE_ANER;
525 		break;
526 	case MII_ANLPAR:
527 		re8139_reg = RE_LPAR;
528 		break;
529 	case MII_PHYIDR1:
530 	case MII_PHYIDR2:
531 		return(0);
532 	default:
533 		device_printf(dev, "bad phy register\n");
534 		return(0);
535 	}
536 	CSR_WRITE_2(sc, re8139_reg, data);
537 	return(0);
538 }
539 
540 static void
541 re_miibus_statchg(device_t dev)
542 {
543 }
544 
545 /*
546  * Program the 64-bit multicast hash filter.
547  */
548 static void
549 re_setmulti(struct re_softc *sc)
550 {
551 	struct ifnet *ifp = &sc->arpcom.ac_if;
552 	int h = 0;
553 	uint32_t hashes[2] = { 0, 0 };
554 	struct ifmultiaddr *ifma;
555 	uint32_t rxfilt;
556 	int mcnt = 0;
557 
558 	rxfilt = CSR_READ_4(sc, RE_RXCFG);
559 
560 	if (ifp->if_flags & IFF_ALLMULTI || ifp->if_flags & IFF_PROMISC) {
561 		rxfilt |= RE_RXCFG_RX_MULTI;
562 		CSR_WRITE_4(sc, RE_RXCFG, rxfilt);
563 		CSR_WRITE_4(sc, RE_MAR0, 0xFFFFFFFF);
564 		CSR_WRITE_4(sc, RE_MAR4, 0xFFFFFFFF);
565 		return;
566 	}
567 
568 	/* first, zot all the existing hash bits */
569 	CSR_WRITE_4(sc, RE_MAR0, 0);
570 	CSR_WRITE_4(sc, RE_MAR4, 0);
571 
572 	/* now program new ones */
573 	LIST_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
574 		if (ifma->ifma_addr->sa_family != AF_LINK)
575 			continue;
576 		h = ether_crc32_be(LLADDR((struct sockaddr_dl *)
577 		    ifma->ifma_addr), ETHER_ADDR_LEN) >> 26;
578 		if (h < 32)
579 			hashes[0] |= (1 << h);
580 		else
581 			hashes[1] |= (1 << (h - 32));
582 		mcnt++;
583 	}
584 
585 	if (mcnt)
586 		rxfilt |= RE_RXCFG_RX_MULTI;
587 	else
588 		rxfilt &= ~RE_RXCFG_RX_MULTI;
589 
590 	CSR_WRITE_4(sc, RE_RXCFG, rxfilt);
591 
592 	/*
593 	 * For some unfathomable reason, RealTek decided to reverse
594 	 * the order of the multicast hash registers in the PCI Express
595 	 * parts. This means we have to write the hash pattern in reverse
596 	 * order for those devices.
597 	 */
598 	if (sc->re_flags & RE_F_PCIE) {
599 		CSR_WRITE_4(sc, RE_MAR0, bswap32(hashes[0]));
600 		CSR_WRITE_4(sc, RE_MAR4, bswap32(hashes[1]));
601 	} else {
602 		CSR_WRITE_4(sc, RE_MAR0, hashes[0]);
603 		CSR_WRITE_4(sc, RE_MAR4, hashes[1]);
604 	}
605 }
606 
607 static void
608 re_reset(struct re_softc *sc)
609 {
610 	int i;
611 
612 	CSR_WRITE_1(sc, RE_COMMAND, RE_CMD_RESET);
613 
614 	for (i = 0; i < RE_TIMEOUT; i++) {
615 		DELAY(10);
616 		if ((CSR_READ_1(sc, RE_COMMAND) & RE_CMD_RESET) == 0)
617 			break;
618 	}
619 	if (i == RE_TIMEOUT)
620 		if_printf(&sc->arpcom.ac_if, "reset never completed!\n");
621 
622 	CSR_WRITE_1(sc, 0x82, 1);
623 }
624 
625 #ifdef RE_DIAG
626 /*
627  * The following routine is designed to test for a defect on some
628  * 32-bit 8169 cards. Some of these NICs have the REQ64# and ACK64#
629  * lines connected to the bus, however for a 32-bit only card, they
630  * should be pulled high. The result of this defect is that the
631  * NIC will not work right if you plug it into a 64-bit slot: DMA
632  * operations will be done with 64-bit transfers, which will fail
633  * because the 64-bit data lines aren't connected.
634  *
635  * There's no way to work around this (short of talking a soldering
636  * iron to the board), however we can detect it. The method we use
637  * here is to put the NIC into digital loopback mode, set the receiver
638  * to promiscuous mode, and then try to send a frame. We then compare
639  * the frame data we sent to what was received. If the data matches,
640  * then the NIC is working correctly, otherwise we know the user has
641  * a defective NIC which has been mistakenly plugged into a 64-bit PCI
642  * slot. In the latter case, there's no way the NIC can work correctly,
643  * so we print out a message on the console and abort the device attach.
644  */
645 
646 static int
647 re_diag(struct re_softc *sc)
648 {
649 	struct ifnet *ifp = &sc->arpcom.ac_if;
650 	struct mbuf *m0;
651 	struct ether_header *eh;
652 	struct re_desc *cur_rx;
653 	uint16_t status;
654 	uint32_t rxstat;
655 	int total_len, i, error = 0, phyaddr;
656 	uint8_t dst[ETHER_ADDR_LEN] = { 0x00, 'h', 'e', 'l', 'l', 'o' };
657 	uint8_t src[ETHER_ADDR_LEN] = { 0x00, 'w', 'o', 'r', 'l', 'd' };
658 
659 	/* Allocate a single mbuf */
660 
661 	MGETHDR(m0, MB_DONTWAIT, MT_DATA);
662 	if (m0 == NULL)
663 		return(ENOBUFS);
664 
665 	/*
666 	 * Initialize the NIC in test mode. This sets the chip up
667 	 * so that it can send and receive frames, but performs the
668 	 * following special functions:
669 	 * - Puts receiver in promiscuous mode
670 	 * - Enables digital loopback mode
671 	 * - Leaves interrupts turned off
672 	 */
673 
674 	ifp->if_flags |= IFF_PROMISC;
675 	sc->re_testmode = 1;
676 	re_reset(sc);
677 	re_init(sc);
678 	sc->re_link = 1;
679 	if (sc->re_type == RE_8169)
680 		phyaddr = 1;
681 	else
682 		phyaddr = 0;
683 
684 	re_miibus_writereg(sc->re_dev, phyaddr, MII_BMCR, BMCR_RESET);
685 	for (i = 0; i < RE_TIMEOUT; i++) {
686 		status = re_miibus_readreg(sc->re_dev, phyaddr, MII_BMCR);
687 		if (!(status & BMCR_RESET))
688 			break;
689 	}
690 
691 	re_miibus_writereg(sc->re_dev, phyaddr, MII_BMCR, BMCR_LOOP);
692 	CSR_WRITE_2(sc, RE_ISR, RE_INTRS_DIAG);
693 
694 	DELAY(100000);
695 
696 	/* Put some data in the mbuf */
697 
698 	eh = mtod(m0, struct ether_header *);
699 	bcopy (dst, eh->ether_dhost, ETHER_ADDR_LEN);
700 	bcopy (src, eh->ether_shost, ETHER_ADDR_LEN);
701 	eh->ether_type = htons(ETHERTYPE_IP);
702 	m0->m_pkthdr.len = m0->m_len = ETHER_MIN_LEN - ETHER_CRC_LEN;
703 
704 	/*
705 	 * Queue the packet, start transmission.
706 	 * Note: ifq_handoff() ultimately calls re_start() for us.
707 	 */
708 
709 	CSR_WRITE_2(sc, RE_ISR, 0xFFFF);
710 	error = ifq_handoff(ifp, m0, NULL);
711 	if (error) {
712 		m0 = NULL;
713 		goto done;
714 	}
715 	m0 = NULL;
716 
717 	/* Wait for it to propagate through the chip */
718 
719 	DELAY(100000);
720 	for (i = 0; i < RE_TIMEOUT; i++) {
721 		status = CSR_READ_2(sc, RE_ISR);
722 		CSR_WRITE_2(sc, RE_ISR, status);
723 		if ((status & (RE_ISR_TIMEOUT_EXPIRED|RE_ISR_RX_OK)) ==
724 		    (RE_ISR_TIMEOUT_EXPIRED|RE_ISR_RX_OK))
725 			break;
726 		DELAY(10);
727 	}
728 
729 	if (i == RE_TIMEOUT) {
730 		if_printf(ifp, "diagnostic failed to receive packet "
731 			  "in loopback mode\n");
732 		error = EIO;
733 		goto done;
734 	}
735 
736 	/*
737 	 * The packet should have been dumped into the first
738 	 * entry in the RX DMA ring. Grab it from there.
739 	 */
740 
741 	bus_dmamap_sync(sc->re_ldata.re_rx_list_tag,
742 			sc->re_ldata.re_rx_list_map, BUS_DMASYNC_POSTREAD);
743 	bus_dmamap_sync(sc->re_ldata.re_mtag, sc->re_ldata.re_rx_dmamap[0],
744 			BUS_DMASYNC_POSTWRITE);
745 	bus_dmamap_unload(sc->re_ldata.re_mtag, sc->re_ldata.re_rx_dmamap[0]);
746 
747 	m0 = sc->re_ldata.re_rx_mbuf[0];
748 	sc->re_ldata.re_rx_mbuf[0] = NULL;
749 	eh = mtod(m0, struct ether_header *);
750 
751 	cur_rx = &sc->re_ldata.re_rx_list[0];
752 	total_len = RE_RXBYTES(cur_rx);
753 	rxstat = le32toh(cur_rx->re_cmdstat);
754 
755 	if (total_len != ETHER_MIN_LEN) {
756 		if_printf(ifp, "diagnostic failed, received short packet\n");
757 		error = EIO;
758 		goto done;
759 	}
760 
761 	/* Test that the received packet data matches what we sent. */
762 
763 	if (bcmp(eh->ether_dhost, dst, ETHER_ADDR_LEN) ||
764 	    bcmp(eh->ether_shost, &src, ETHER_ADDR_LEN) ||
765 	    be16toh(eh->ether_type) != ETHERTYPE_IP) {
766 		if_printf(ifp, "WARNING, DMA FAILURE!\n");
767 		if_printf(ifp, "expected TX data: %6D/%6D/0x%x\n",
768 		    dst, ":", src, ":", ETHERTYPE_IP);
769 		if_printf(ifp, "received RX data: %6D/%6D/0x%x\n",
770 		    eh->ether_dhost, ":",  eh->ether_shost, ":",
771 		    ntohs(eh->ether_type));
772 		if_printf(ifp, "You may have a defective 32-bit NIC plugged "
773 		    "into a 64-bit PCI slot.\n");
774 		if_printf(ifp, "Please re-install the NIC in a 32-bit slot "
775 		    "for proper operation.\n");
776 		if_printf(ifp, "Read the re(4) man page for more details.\n");
777 		error = EIO;
778 	}
779 
780 done:
781 	/* Turn interface off, release resources */
782 
783 	sc->re_testmode = 0;
784 	sc->re_link = 0;
785 	ifp->if_flags &= ~IFF_PROMISC;
786 	re_stop(sc);
787 	if (m0 != NULL)
788 		m_freem(m0);
789 
790 	return (error);
791 }
792 #endif	/* RE_DIAG */
793 
794 /*
795  * Probe for a RealTek 8139C+/8169/8110 chip. Check the PCI vendor and device
796  * IDs against our list and return a device name if we find a match.
797  */
798 static int
799 re_probe(device_t dev)
800 {
801 	const struct re_type *t;
802 	struct re_softc *sc;
803 	int rid;
804 	uint32_t hwrev;
805 	uint16_t vendor, product;
806 
807 	t = re_devs;
808 
809 	vendor = pci_get_vendor(dev);
810 	product = pci_get_device(dev);
811 
812 	/*
813 	 * Only attach to rev.3 of the Linksys EG1032 adapter.
814 	 * Rev.2 is supported by sk(4).
815 	 */
816 	if (vendor == PCI_VENDOR_LINKSYS &&
817 	    product == PCI_PRODUCT_LINKSYS_EG1032 &&
818 	    pci_get_subdevice(dev) != PCI_SUBDEVICE_LINKSYS_EG1032_REV3)
819 			return ENXIO;
820 
821 	for (t = re_devs; t->re_name != NULL; t++) {
822 		if (product == t->re_did && vendor == t->re_vid)
823 			break;
824 	}
825 
826 	/*
827 	 * Check if we found a RealTek device.
828 	 */
829 	if (t->re_name == NULL)
830 		return(ENXIO);
831 
832 	/*
833 	 * Temporarily map the I/O space so we can read the chip ID register.
834 	 */
835 	sc = kmalloc(sizeof(*sc), M_TEMP, M_WAITOK | M_ZERO);
836 	rid = RE_PCI_LOIO;
837 	sc->re_res = bus_alloc_resource_any(dev, SYS_RES_IOPORT, &rid,
838 					    RF_ACTIVE);
839 	if (sc->re_res == NULL) {
840 		device_printf(dev, "couldn't map ports/memory\n");
841 		kfree(sc, M_TEMP);
842 		return(ENXIO);
843 	}
844 
845 	sc->re_btag = rman_get_bustag(sc->re_res);
846 	sc->re_bhandle = rman_get_bushandle(sc->re_res);
847 
848 	hwrev = CSR_READ_4(sc, RE_TXCFG) & RE_TXCFG_HWREV;
849 	bus_release_resource(dev, SYS_RES_IOPORT, RE_PCI_LOIO, sc->re_res);
850 	kfree(sc, M_TEMP);
851 
852 	/*
853 	 * and continue matching for the specific chip...
854 	 */
855 	for (; t->re_name != NULL; t++) {
856 		if (product == t->re_did && vendor == t->re_vid &&
857 		    t->re_basetype == hwrev) {
858 			device_set_desc(dev, t->re_name);
859 			return(0);
860 		}
861 	}
862 
863 	if (bootverbose)
864 		kprintf("re: unknown hwrev %#x\n", hwrev);
865 	return(ENXIO);
866 }
867 
868 /*
869  * This routine takes the segment list provided as the result of
870  * a bus_dma_map_load() operation and assigns the addresses/lengths
871  * to RealTek DMA descriptors. This can be called either by the RX
872  * code or the TX code. In the RX case, we'll probably wind up mapping
873  * at most one segment. For the TX case, there could be any number of
874  * segments since TX packets may span multiple mbufs. In either case,
875  * if the number of segments is larger than the re_maxsegs limit
876  * specified by the caller, we abort the mapping operation. Sadly,
877  * whoever designed the buffer mapping API did not provide a way to
878  * return an error from here, so we have to fake it a bit.
879  */
880 
881 static void
882 re_dma_map_desc(void *arg, bus_dma_segment_t *segs, int nseg,
883 		bus_size_t mapsize, int error)
884 {
885 	struct re_dmaload_arg *ctx;
886 	struct re_desc *d = NULL;
887 	int i = 0, idx;
888 	uint32_t cmdstat;
889 
890 	if (error)
891 		return;
892 
893 	ctx = arg;
894 
895 	/* Signal error to caller if there's too many segments */
896 	if (nseg > ctx->re_maxsegs) {
897 		ctx->re_maxsegs = 0;
898 		return;
899 	}
900 
901 	/*
902 	 * Map the segment array into descriptors. Note that we set the
903 	 * start-of-frame and end-of-frame markers for either TX or RX, but
904 	 * they really only have meaning in the TX case. (In the RX case,
905 	 * it's the chip that tells us where packets begin and end.)
906 	 * We also keep track of the end of the ring and set the
907 	 * end-of-ring bits as needed, and we set the ownership bits
908 	 * in all except the very first descriptor. (The caller will
909 	 * set this descriptor later when it start transmission or
910 	 * reception.)
911 	 */
912 	idx = ctx->re_idx;
913 	for (;;) {
914 		d = &ctx->re_ring[idx];
915 		if (le32toh(d->re_cmdstat) & RE_RDESC_STAT_OWN) {
916 			ctx->re_maxsegs = 0;
917 			return;
918 		}
919 		cmdstat = segs[i].ds_len;
920 		d->re_bufaddr_lo = htole32(RE_ADDR_LO(segs[i].ds_addr));
921 		d->re_bufaddr_hi = htole32(RE_ADDR_HI(segs[i].ds_addr));
922 		if (i == 0)
923 			cmdstat |= RE_TDESC_CMD_SOF;
924 		else
925 			cmdstat |= RE_TDESC_CMD_OWN;
926 		if (idx == (RE_RX_DESC_CNT - 1))
927 			cmdstat |= RE_TDESC_CMD_EOR;
928 		d->re_cmdstat = htole32(cmdstat | ctx->re_flags);
929 		i++;
930 		if (i == nseg)
931 			break;
932 		RE_DESC_INC(idx);
933 	}
934 
935 	d->re_cmdstat |= htole32(RE_TDESC_CMD_EOF);
936 	ctx->re_maxsegs = nseg;
937 	ctx->re_idx = idx;
938 }
939 
940 /*
941  * Map a single buffer address.
942  */
943 
944 static void
945 re_dma_map_addr(void *arg, bus_dma_segment_t *segs, int nseg, int error)
946 {
947 	uint32_t *addr;
948 
949 	if (error)
950 		return;
951 
952 	KASSERT(nseg == 1, ("too many DMA segments, %d should be 1", nseg));
953 	addr = arg;
954 	*addr = segs->ds_addr;
955 }
956 
957 static int
958 re_allocmem(device_t dev, struct re_softc *sc)
959 {
960 	int error, i, nseg;
961 
962 	/*
963 	 * Allocate map for RX mbufs.
964 	 */
965 	nseg = 32;
966 	error = bus_dma_tag_create(sc->re_parent_tag, ETHER_ALIGN, 0,
967 	    BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
968 	    NULL, MCLBYTES * nseg, nseg, MCLBYTES, BUS_DMA_ALLOCNOW,
969 	    &sc->re_ldata.re_mtag);
970 	if (error) {
971 		device_printf(dev, "could not allocate dma tag\n");
972 		return(error);
973 	}
974 
975 	/*
976 	 * Allocate map for TX descriptor list.
977 	 */
978 	error = bus_dma_tag_create(sc->re_parent_tag, RE_RING_ALIGN,
979 	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
980             NULL, RE_TX_LIST_SZ, 1, RE_TX_LIST_SZ, BUS_DMA_ALLOCNOW,
981 	    &sc->re_ldata.re_tx_list_tag);
982 	if (error) {
983 		device_printf(dev, "could not allocate dma tag\n");
984 		return(error);
985 	}
986 
987 	/* Allocate DMA'able memory for the TX ring */
988 
989         error = bus_dmamem_alloc(sc->re_ldata.re_tx_list_tag,
990 	    (void **)&sc->re_ldata.re_tx_list, BUS_DMA_WAITOK | BUS_DMA_ZERO,
991             &sc->re_ldata.re_tx_list_map);
992         if (error) {
993 		device_printf(dev, "could not allocate TX ring\n");
994                 return(error);
995 	}
996 
997 	/* Load the map for the TX ring. */
998 
999 	error = bus_dmamap_load(sc->re_ldata.re_tx_list_tag,
1000 	     sc->re_ldata.re_tx_list_map, sc->re_ldata.re_tx_list,
1001 	     RE_TX_LIST_SZ, re_dma_map_addr,
1002 	     &sc->re_ldata.re_tx_list_addr, BUS_DMA_NOWAIT);
1003 	if (error) {
1004 		device_printf(dev, "could not get address of TX ring\n");
1005 		return(error);
1006 	}
1007 
1008 	/* Create DMA maps for TX buffers */
1009 
1010 	for (i = 0; i < RE_TX_DESC_CNT; i++) {
1011 		error = bus_dmamap_create(sc->re_ldata.re_mtag, 0,
1012 			    &sc->re_ldata.re_tx_dmamap[i]);
1013 		if (error) {
1014 			device_printf(dev, "can't create DMA map for TX\n");
1015 			return(error);
1016 		}
1017 	}
1018 
1019 	/*
1020 	 * Allocate map for RX descriptor list.
1021 	 */
1022 	error = bus_dma_tag_create(sc->re_parent_tag, RE_RING_ALIGN,
1023 	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1024             NULL, RE_RX_LIST_SZ, 1, RE_RX_LIST_SZ, BUS_DMA_ALLOCNOW,
1025 	    &sc->re_ldata.re_rx_list_tag);
1026 	if (error) {
1027 		device_printf(dev, "could not allocate dma tag\n");
1028 		return(error);
1029 	}
1030 
1031 	/* Allocate DMA'able memory for the RX ring */
1032 
1033         error = bus_dmamem_alloc(sc->re_ldata.re_rx_list_tag,
1034 	    (void **)&sc->re_ldata.re_rx_list, BUS_DMA_WAITOK | BUS_DMA_ZERO,
1035             &sc->re_ldata.re_rx_list_map);
1036         if (error) {
1037 		device_printf(dev, "could not allocate RX ring\n");
1038                 return(error);
1039 	}
1040 
1041 	/* Load the map for the RX ring. */
1042 
1043 	error = bus_dmamap_load(sc->re_ldata.re_rx_list_tag,
1044 	     sc->re_ldata.re_rx_list_map, sc->re_ldata.re_rx_list,
1045 	     RE_RX_LIST_SZ, re_dma_map_addr,
1046 	     &sc->re_ldata.re_rx_list_addr, BUS_DMA_NOWAIT);
1047 	if (error) {
1048 		device_printf(dev, "could not get address of RX ring\n");
1049 		return(error);
1050 	}
1051 
1052 	/* Create DMA maps for RX buffers */
1053 
1054 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
1055 		error = bus_dmamap_create(sc->re_ldata.re_mtag, 0,
1056 			    &sc->re_ldata.re_rx_dmamap[i]);
1057 		if (error) {
1058 			device_printf(dev, "can't create DMA map for RX\n");
1059 			return(ENOMEM);
1060 		}
1061 	}
1062 
1063 	return(0);
1064 }
1065 
1066 /*
1067  * Attach the interface. Allocate softc structures, do ifmedia
1068  * setup and ethernet/BPF attach.
1069  */
1070 static int
1071 re_attach(device_t dev)
1072 {
1073 	struct re_softc	*sc = device_get_softc(dev);
1074 	struct ifnet *ifp;
1075 	const struct re_hwrev *hw_rev;
1076 	uint8_t eaddr[ETHER_ADDR_LEN];
1077 	uint16_t as[ETHER_ADDR_LEN / 2];
1078 	uint16_t re_did = 0;
1079 	uint32_t hwrev;
1080 	int error = 0, rid, i;
1081 
1082 	callout_init(&sc->re_timer);
1083 #ifdef RE_DIAG
1084 	sc->re_dev = dev;
1085 #endif
1086 
1087 	RE_ENABLE_TX_MODERATION(sc);
1088 
1089 	sysctl_ctx_init(&sc->re_sysctl_ctx);
1090 	sc->re_sysctl_tree = SYSCTL_ADD_NODE(&sc->re_sysctl_ctx,
1091 					     SYSCTL_STATIC_CHILDREN(_hw),
1092 					     OID_AUTO,
1093 					     device_get_nameunit(dev),
1094 					     CTLFLAG_RD, 0, "");
1095 	if (sc->re_sysctl_tree == NULL) {
1096 		device_printf(dev, "can't add sysctl node\n");
1097 		error = ENXIO;
1098 		goto fail;
1099 	}
1100 	SYSCTL_ADD_PROC(&sc->re_sysctl_ctx,
1101 			SYSCTL_CHILDREN(sc->re_sysctl_tree),
1102 			OID_AUTO, "tx_moderation",
1103 			CTLTYPE_INT | CTLFLAG_RW,
1104 			sc, 0, re_sysctl_tx_moderation, "I",
1105 			"Enable/Disable TX moderation");
1106 
1107 #ifndef BURN_BRIDGES
1108 	/*
1109 	 * Handle power management nonsense.
1110 	 */
1111 
1112 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D0) {
1113 		uint32_t membase, irq;
1114 
1115 		/* Save important PCI config data. */
1116 		membase = pci_read_config(dev, RE_PCI_LOMEM, 4);
1117 		irq = pci_read_config(dev, PCIR_INTLINE, 4);
1118 
1119 		/* Reset the power state. */
1120 		device_printf(dev, "chip is in D%d power mode "
1121 		    "-- setting to D0\n", pci_get_powerstate(dev));
1122 
1123 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
1124 
1125 		/* Restore PCI config data. */
1126 		pci_write_config(dev, RE_PCI_LOMEM, membase, 4);
1127 		pci_write_config(dev, PCIR_INTLINE, irq, 4);
1128 	}
1129 #endif
1130 	/*
1131 	 * Map control/status registers.
1132 	 */
1133 	pci_enable_busmaster(dev);
1134 
1135 	rid = RE_PCI_LOIO;
1136 	sc->re_res = bus_alloc_resource_any(dev, SYS_RES_IOPORT, &rid,
1137 					    RF_ACTIVE);
1138 
1139 	if (sc->re_res == NULL) {
1140 		device_printf(dev, "couldn't map ports\n");
1141 		error = ENXIO;
1142 		goto fail;
1143 	}
1144 
1145 	sc->re_btag = rman_get_bustag(sc->re_res);
1146 	sc->re_bhandle = rman_get_bushandle(sc->re_res);
1147 
1148 	/* Allocate interrupt */
1149 	rid = 0;
1150 	sc->re_irq = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
1151 					    RF_SHAREABLE | RF_ACTIVE);
1152 
1153 	if (sc->re_irq == NULL) {
1154 		device_printf(dev, "couldn't map interrupt\n");
1155 		error = ENXIO;
1156 		goto fail;
1157 	}
1158 
1159 	/* Reset the adapter. */
1160 	re_reset(sc);
1161 
1162 	hwrev = CSR_READ_4(sc, RE_TXCFG) & RE_TXCFG_HWREV;
1163 	for (hw_rev = re_hwrevs; hw_rev->re_desc != NULL; hw_rev++) {
1164 		if (hw_rev->re_rev == hwrev) {
1165 			sc->re_type = hw_rev->re_type;
1166 			sc->re_flags = hw_rev->re_flags;
1167 			break;
1168 		}
1169 	}
1170 
1171 	sc->re_eewidth = 6;
1172 	re_read_eeprom(sc, (caddr_t)&re_did, 0, 1);
1173 	if (re_did != 0x8129)
1174 	        sc->re_eewidth = 8;
1175 
1176 	/*
1177 	 * Get station address from the EEPROM.
1178 	 */
1179 	re_read_eeprom(sc, (caddr_t)as, RE_EE_EADDR, 3);
1180 	for (i = 0; i < ETHER_ADDR_LEN / 2; i++)
1181 		as[i] = le16toh(as[i]);
1182 	bcopy(as, eaddr, sizeof(eaddr));
1183 
1184 	if (sc->re_type == RE_8169) {
1185 		/* Set RX length mask */
1186 		sc->re_rxlenmask = RE_RDESC_STAT_GFRAGLEN;
1187 		sc->re_txstart = RE_GTXSTART;
1188 	} else {
1189 		/* Set RX length mask */
1190 		sc->re_rxlenmask = RE_RDESC_STAT_FRAGLEN;
1191 		sc->re_txstart = RE_TXSTART;
1192 	}
1193 
1194 	/*
1195 	 * Allocate the parent bus DMA tag appropriate for PCI.
1196 	 */
1197 #define RE_NSEG_NEW 32
1198 	error = bus_dma_tag_create(NULL,	/* parent */
1199 			1, 0,			/* alignment, boundary */
1200 			BUS_SPACE_MAXADDR_32BIT,/* lowaddr */
1201 			BUS_SPACE_MAXADDR,	/* highaddr */
1202 			NULL, NULL,		/* filter, filterarg */
1203 			MAXBSIZE, RE_NSEG_NEW,	/* maxsize, nsegments */
1204 			BUS_SPACE_MAXSIZE_32BIT,/* maxsegsize */
1205 			BUS_DMA_ALLOCNOW,	/* flags */
1206 			&sc->re_parent_tag);
1207 	if (error)
1208 		goto fail;
1209 
1210 	error = re_allocmem(dev, sc);
1211 
1212 	if (error)
1213 		goto fail;
1214 
1215 	/* Do MII setup */
1216 	if (mii_phy_probe(dev, &sc->re_miibus,
1217 	    re_ifmedia_upd, re_ifmedia_sts)) {
1218 		device_printf(dev, "MII without any phy!\n");
1219 		error = ENXIO;
1220 		goto fail;
1221 	}
1222 
1223 	ifp = &sc->arpcom.ac_if;
1224 	ifp->if_softc = sc;
1225 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1226 	ifp->if_mtu = ETHERMTU;
1227 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1228 	ifp->if_ioctl = re_ioctl;
1229 	ifp->if_start = re_start;
1230 	ifp->if_capabilities = IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING;
1231 	if (hwrev != RE_HWREV_8168C)	/* XXX does not work yet */
1232 		ifp->if_capabilities |= IFCAP_HWCSUM;
1233 #ifdef DEVICE_POLLING
1234 	ifp->if_poll = re_poll;
1235 #endif
1236 	ifp->if_watchdog = re_watchdog;
1237 	ifp->if_init = re_init;
1238 	if (sc->re_type == RE_8169)
1239 		ifp->if_baudrate = 1000000000;
1240 	else
1241 		ifp->if_baudrate = 100000000;
1242 	ifq_set_maxlen(&ifp->if_snd, RE_IFQ_MAXLEN);
1243 	ifq_set_ready(&ifp->if_snd);
1244 
1245 #ifdef RE_DISABLE_HWCSUM
1246 	ifp->if_capenable = ifp->if_capabilities & ~IFCAP_HWCSUM;
1247 	ifp->if_hwassist = 0;
1248 #else
1249 	ifp->if_capenable = ifp->if_capabilities;
1250 	if (ifp->if_capabilities & IFCAP_HWCSUM)
1251 		ifp->if_hwassist = RE_CSUM_FEATURES;
1252 	else
1253 		ifp->if_hwassist = 0;
1254 #endif	/* RE_DISABLE_HWCSUM */
1255 
1256 	/*
1257 	 * Call MI attach routine.
1258 	 */
1259 	ether_ifattach(ifp, eaddr, NULL);
1260 
1261 #ifdef RE_DIAG
1262 	/*
1263 	 * Perform hardware diagnostic on the original RTL8169.
1264 	 * Some 32-bit cards were incorrectly wired and would
1265 	 * malfunction if plugged into a 64-bit slot.
1266 	 */
1267 	if (hwrev == RE_HWREV_8169) {
1268 		lwkt_serialize_enter(ifp->if_serializer);
1269 		error = re_diag(sc);
1270 		lwkt_serialize_exit(ifp->if_serializer);
1271 
1272 		if (error) {
1273 			device_printf(dev, "hardware diagnostic failure\n");
1274 			ether_ifdetach(ifp);
1275 			goto fail;
1276 		}
1277 	}
1278 #endif	/* RE_DIAG */
1279 
1280 	/* Hook interrupt last to avoid having to lock softc */
1281 	error = bus_setup_intr(dev, sc->re_irq, INTR_NETSAFE, re_intr, sc,
1282 			       &sc->re_intrhand, ifp->if_serializer);
1283 
1284 	if (error) {
1285 		device_printf(dev, "couldn't set up irq\n");
1286 		ether_ifdetach(ifp);
1287 		goto fail;
1288 	}
1289 
1290 	ifp->if_cpuid = ithread_cpuid(rman_get_start(sc->re_irq));
1291 	KKASSERT(ifp->if_cpuid >= 0 && ifp->if_cpuid < ncpus);
1292 
1293 fail:
1294 	if (error)
1295 		re_detach(dev);
1296 
1297 	return (error);
1298 }
1299 
1300 /*
1301  * Shutdown hardware and free up resources. This can be called any
1302  * time after the mutex has been initialized. It is called in both
1303  * the error case in attach and the normal detach case so it needs
1304  * to be careful about only freeing resources that have actually been
1305  * allocated.
1306  */
1307 static int
1308 re_detach(device_t dev)
1309 {
1310 	struct re_softc *sc = device_get_softc(dev);
1311 	struct ifnet *ifp = &sc->arpcom.ac_if;
1312 	int i;
1313 
1314 	/* These should only be active if attach succeeded */
1315 	if (device_is_attached(dev)) {
1316 		lwkt_serialize_enter(ifp->if_serializer);
1317 		re_stop(sc);
1318 		bus_teardown_intr(dev, sc->re_irq, sc->re_intrhand);
1319 		lwkt_serialize_exit(ifp->if_serializer);
1320 
1321 		ether_ifdetach(ifp);
1322 	}
1323 	if (sc->re_miibus)
1324 		device_delete_child(dev, sc->re_miibus);
1325 	bus_generic_detach(dev);
1326 
1327 	if (sc->re_irq)
1328 		bus_release_resource(dev, SYS_RES_IRQ, 0, sc->re_irq);
1329 	if (sc->re_res) {
1330 		bus_release_resource(dev, SYS_RES_IOPORT, RE_PCI_LOIO,
1331 				     sc->re_res);
1332 	}
1333 
1334 	/* Unload and free the RX DMA ring memory and map */
1335 
1336 	if (sc->re_ldata.re_rx_list_tag) {
1337 		bus_dmamap_unload(sc->re_ldata.re_rx_list_tag,
1338 		    sc->re_ldata.re_rx_list_map);
1339 		bus_dmamem_free(sc->re_ldata.re_rx_list_tag,
1340 		    sc->re_ldata.re_rx_list,
1341 		    sc->re_ldata.re_rx_list_map);
1342 		bus_dma_tag_destroy(sc->re_ldata.re_rx_list_tag);
1343 	}
1344 
1345 	/* Unload and free the TX DMA ring memory and map */
1346 
1347 	if (sc->re_ldata.re_tx_list_tag) {
1348 		bus_dmamap_unload(sc->re_ldata.re_tx_list_tag,
1349 		    sc->re_ldata.re_tx_list_map);
1350 		bus_dmamem_free(sc->re_ldata.re_tx_list_tag,
1351 		    sc->re_ldata.re_tx_list,
1352 		    sc->re_ldata.re_tx_list_map);
1353 		bus_dma_tag_destroy(sc->re_ldata.re_tx_list_tag);
1354 	}
1355 
1356 	/* Destroy all the RX and TX buffer maps */
1357 
1358 	if (sc->re_ldata.re_mtag) {
1359 		for (i = 0; i < RE_TX_DESC_CNT; i++)
1360 			bus_dmamap_destroy(sc->re_ldata.re_mtag,
1361 			    sc->re_ldata.re_tx_dmamap[i]);
1362 		for (i = 0; i < RE_RX_DESC_CNT; i++)
1363 			bus_dmamap_destroy(sc->re_ldata.re_mtag,
1364 			    sc->re_ldata.re_rx_dmamap[i]);
1365 		bus_dma_tag_destroy(sc->re_ldata.re_mtag);
1366 	}
1367 
1368 	/* Unload and free the stats buffer and map */
1369 
1370 	if (sc->re_ldata.re_stag) {
1371 		bus_dmamap_unload(sc->re_ldata.re_stag,
1372 		    sc->re_ldata.re_rx_list_map);
1373 		bus_dmamem_free(sc->re_ldata.re_stag,
1374 		    sc->re_ldata.re_stats,
1375 		    sc->re_ldata.re_smap);
1376 		bus_dma_tag_destroy(sc->re_ldata.re_stag);
1377 	}
1378 
1379 	if (sc->re_parent_tag)
1380 		bus_dma_tag_destroy(sc->re_parent_tag);
1381 
1382 	return(0);
1383 }
1384 
1385 static int
1386 re_newbuf(struct re_softc *sc, int idx, struct mbuf *m)
1387 {
1388 	struct re_dmaload_arg arg;
1389 	struct mbuf *n = NULL;
1390 	int error;
1391 
1392 	if (m == NULL) {
1393 		n = m_getcl(MB_DONTWAIT, MT_DATA, M_PKTHDR);
1394 		if (n == NULL)
1395 			return(ENOBUFS);
1396 		m = n;
1397 	} else
1398 		m->m_data = m->m_ext.ext_buf;
1399 
1400 	m->m_len = m->m_pkthdr.len = MCLBYTES;
1401 
1402 	/*
1403 	 * NOTE:
1404 	 * Some re(4) chips(e.g. RTL8101E) need address of the receive buffer
1405 	 * to be 8-byte aligned, so don't call m_adj(m, ETHER_ALIGN) here.
1406 	 */
1407 
1408 	arg.sc = sc;
1409 	arg.re_idx = idx;
1410 	arg.re_maxsegs = 1;
1411 	arg.re_flags = 0;
1412 	arg.re_ring = sc->re_ldata.re_rx_list;
1413 
1414         error = bus_dmamap_load_mbuf(sc->re_ldata.re_mtag,
1415 	    sc->re_ldata.re_rx_dmamap[idx], m, re_dma_map_desc,
1416 	    &arg, BUS_DMA_NOWAIT);
1417 	if (error || arg.re_maxsegs != 1) {
1418 		if (n != NULL)
1419 			m_freem(n);
1420 		return (ENOMEM);
1421 	}
1422 
1423 	sc->re_ldata.re_rx_list[idx].re_cmdstat |= htole32(RE_RDESC_CMD_OWN);
1424 	sc->re_ldata.re_rx_mbuf[idx] = m;
1425 
1426         bus_dmamap_sync(sc->re_ldata.re_mtag, sc->re_ldata.re_rx_dmamap[idx],
1427 		        BUS_DMASYNC_PREREAD);
1428 
1429 	return(0);
1430 }
1431 
1432 static int
1433 re_tx_list_init(struct re_softc *sc)
1434 {
1435 	bzero(sc->re_ldata.re_tx_list, RE_TX_LIST_SZ);
1436 	bzero(&sc->re_ldata.re_tx_mbuf, RE_TX_DESC_CNT * sizeof(struct mbuf *));
1437 
1438 	bus_dmamap_sync(sc->re_ldata.re_tx_list_tag,
1439 			sc->re_ldata.re_tx_list_map, BUS_DMASYNC_PREWRITE);
1440 	sc->re_ldata.re_tx_prodidx = 0;
1441 	sc->re_ldata.re_tx_considx = 0;
1442 	sc->re_ldata.re_tx_free = RE_TX_DESC_CNT;
1443 
1444 	return(0);
1445 }
1446 
1447 static int
1448 re_rx_list_init(struct re_softc *sc)
1449 {
1450 	int i, error;
1451 
1452 	bzero(sc->re_ldata.re_rx_list, RE_RX_LIST_SZ);
1453 	bzero(&sc->re_ldata.re_rx_mbuf, RE_RX_DESC_CNT * sizeof(struct mbuf *));
1454 
1455 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
1456 		error = re_newbuf(sc, i, NULL);
1457 		if (error)
1458 			return(error);
1459 	}
1460 
1461 	/* Flush the RX descriptors */
1462 
1463 	bus_dmamap_sync(sc->re_ldata.re_rx_list_tag,
1464 			sc->re_ldata.re_rx_list_map, BUS_DMASYNC_PREWRITE);
1465 
1466 	sc->re_ldata.re_rx_prodidx = 0;
1467 	sc->re_head = sc->re_tail = NULL;
1468 
1469 	return(0);
1470 }
1471 
1472 /*
1473  * RX handler for C+ and 8169. For the gigE chips, we support
1474  * the reception of jumbo frames that have been fragmented
1475  * across multiple 2K mbuf cluster buffers.
1476  */
1477 static void
1478 re_rxeof(struct re_softc *sc)
1479 {
1480 	struct ifnet *ifp = &sc->arpcom.ac_if;
1481 	struct mbuf *m;
1482 	struct re_desc 	*cur_rx;
1483 	uint32_t rxstat, rxvlan;
1484 	int i, total_len;
1485 #ifdef ETHER_INPUT_CHAIN
1486 	struct mbuf_chain chain[MAXCPU];
1487 #endif
1488 
1489 	/* Invalidate the descriptor memory */
1490 
1491 	bus_dmamap_sync(sc->re_ldata.re_rx_list_tag,
1492 			sc->re_ldata.re_rx_list_map, BUS_DMASYNC_POSTREAD);
1493 
1494 #ifdef ETHER_INPUT_CHAIN
1495 	ether_input_chain_init(chain);
1496 #endif
1497 
1498 	for (i = sc->re_ldata.re_rx_prodidx;
1499 	     RE_OWN(&sc->re_ldata.re_rx_list[i]) == 0 ; RE_DESC_INC(i)) {
1500 		cur_rx = &sc->re_ldata.re_rx_list[i];
1501 		m = sc->re_ldata.re_rx_mbuf[i];
1502 		total_len = RE_RXBYTES(cur_rx);
1503 		rxstat = le32toh(cur_rx->re_cmdstat);
1504 		rxvlan = le32toh(cur_rx->re_vlanctl);
1505 
1506 		/* Invalidate the RX mbuf and unload its map */
1507 
1508 		bus_dmamap_sync(sc->re_ldata.re_mtag,
1509 				sc->re_ldata.re_rx_dmamap[i],
1510 				BUS_DMASYNC_POSTWRITE);
1511 		bus_dmamap_unload(sc->re_ldata.re_mtag,
1512 				  sc->re_ldata.re_rx_dmamap[i]);
1513 
1514 		if ((rxstat & RE_RDESC_STAT_EOF) == 0) {
1515 			m->m_len = MCLBYTES - ETHER_ALIGN;
1516 			if (sc->re_head == NULL) {
1517 				sc->re_head = sc->re_tail = m;
1518 			} else {
1519 				sc->re_tail->m_next = m;
1520 				sc->re_tail = m;
1521 			}
1522 			re_newbuf(sc, i, NULL);
1523 			continue;
1524 		}
1525 
1526 		/*
1527 		 * NOTE: for the 8139C+, the frame length field
1528 		 * is always 12 bits in size, but for the gigE chips,
1529 		 * it is 13 bits (since the max RX frame length is 16K).
1530 		 * Unfortunately, all 32 bits in the status word
1531 		 * were already used, so to make room for the extra
1532 		 * length bit, RealTek took out the 'frame alignment
1533 		 * error' bit and shifted the other status bits
1534 		 * over one slot. The OWN, EOR, FS and LS bits are
1535 		 * still in the same places. We have already extracted
1536 		 * the frame length and checked the OWN bit, so rather
1537 		 * than using an alternate bit mapping, we shift the
1538 		 * status bits one space to the right so we can evaluate
1539 		 * them using the 8169 status as though it was in the
1540 		 * same format as that of the 8139C+.
1541 		 */
1542 		if (sc->re_type == RE_8169)
1543 			rxstat >>= 1;
1544 
1545 		if (rxstat & RE_RDESC_STAT_RXERRSUM) {
1546 			ifp->if_ierrors++;
1547 			/*
1548 			 * If this is part of a multi-fragment packet,
1549 			 * discard all the pieces.
1550 			 */
1551 			if (sc->re_head != NULL) {
1552 				m_freem(sc->re_head);
1553 				sc->re_head = sc->re_tail = NULL;
1554 			}
1555 			re_newbuf(sc, i, m);
1556 			continue;
1557 		}
1558 
1559 		/*
1560 		 * If allocating a replacement mbuf fails,
1561 		 * reload the current one.
1562 		 */
1563 
1564 		if (re_newbuf(sc, i, NULL)) {
1565 			ifp->if_ierrors++;
1566 			if (sc->re_head != NULL) {
1567 				m_freem(sc->re_head);
1568 				sc->re_head = sc->re_tail = NULL;
1569 			}
1570 			re_newbuf(sc, i, m);
1571 			continue;
1572 		}
1573 
1574 		if (sc->re_head != NULL) {
1575 			m->m_len = total_len % (MCLBYTES - ETHER_ALIGN);
1576 			/*
1577 			 * Special case: if there's 4 bytes or less
1578 			 * in this buffer, the mbuf can be discarded:
1579 			 * the last 4 bytes is the CRC, which we don't
1580 			 * care about anyway.
1581 			 */
1582 			if (m->m_len <= ETHER_CRC_LEN) {
1583 				sc->re_tail->m_len -=
1584 				    (ETHER_CRC_LEN - m->m_len);
1585 				m_freem(m);
1586 			} else {
1587 				m->m_len -= ETHER_CRC_LEN;
1588 				sc->re_tail->m_next = m;
1589 			}
1590 			m = sc->re_head;
1591 			sc->re_head = sc->re_tail = NULL;
1592 			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
1593 		} else
1594 			m->m_pkthdr.len = m->m_len =
1595 			    (total_len - ETHER_CRC_LEN);
1596 
1597 		ifp->if_ipackets++;
1598 		m->m_pkthdr.rcvif = ifp;
1599 
1600 		/* Do RX checksumming if enabled */
1601 
1602 		if (ifp->if_capenable & IFCAP_RXCSUM) {
1603 
1604 			/* Check IP header checksum */
1605 			if (rxstat & RE_RDESC_STAT_PROTOID)
1606 				m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED;
1607 			if ((rxstat & RE_RDESC_STAT_IPSUMBAD) == 0)
1608 				m->m_pkthdr.csum_flags |= CSUM_IP_VALID;
1609 
1610 			/* Check TCP/UDP checksum */
1611 			if ((RE_TCPPKT(rxstat) &&
1612 			    (rxstat & RE_RDESC_STAT_TCPSUMBAD) == 0) ||
1613 			    (RE_UDPPKT(rxstat) &&
1614 			    (rxstat & RE_RDESC_STAT_UDPSUMBAD)) == 0) {
1615 				m->m_pkthdr.csum_flags |=
1616 				    CSUM_DATA_VALID|CSUM_PSEUDO_HDR|
1617 				    CSUM_FRAG_NOT_CHECKED;
1618 				m->m_pkthdr.csum_data = 0xffff;
1619 			}
1620 		}
1621 
1622 		if (rxvlan & RE_RDESC_VLANCTL_TAG) {
1623 			m->m_flags |= M_VLANTAG;
1624 			m->m_pkthdr.ether_vlantag =
1625 				be16toh((rxvlan & RE_RDESC_VLANCTL_DATA));
1626 		}
1627 #ifdef ETHER_INPUT_CHAIN
1628 #ifdef ETHER_INPUT2
1629 		ether_input_chain2(ifp, m, chain);
1630 #else
1631 		ether_input_chain(ifp, m, chain);
1632 #endif
1633 #else
1634 		ifp->if_input(ifp, m);
1635 #endif
1636 	}
1637 
1638 #ifdef ETHER_INPUT_CHAIN
1639 	ether_input_dispatch(chain);
1640 #endif
1641 
1642 	/* Flush the RX DMA ring */
1643 
1644 	bus_dmamap_sync(sc->re_ldata.re_rx_list_tag,
1645 			sc->re_ldata.re_rx_list_map, BUS_DMASYNC_PREWRITE);
1646 
1647 	sc->re_ldata.re_rx_prodidx = i;
1648 }
1649 
1650 static void
1651 re_txeof(struct re_softc *sc)
1652 {
1653 	struct ifnet *ifp = &sc->arpcom.ac_if;
1654 	uint32_t txstat;
1655 	int idx;
1656 
1657 	/* Invalidate the TX descriptor list */
1658 
1659 	bus_dmamap_sync(sc->re_ldata.re_tx_list_tag,
1660 			sc->re_ldata.re_tx_list_map, BUS_DMASYNC_POSTREAD);
1661 
1662 	for (idx = sc->re_ldata.re_tx_considx;
1663 	     sc->re_ldata.re_tx_free < RE_TX_DESC_CNT; RE_DESC_INC(idx)) {
1664 		txstat = le32toh(sc->re_ldata.re_tx_list[idx].re_cmdstat);
1665 		if (txstat & RE_TDESC_CMD_OWN)
1666 			break;
1667 
1668 		sc->re_ldata.re_tx_list[idx].re_bufaddr_lo = 0;
1669 
1670 		/*
1671 		 * We only stash mbufs in the last descriptor
1672 		 * in a fragment chain, which also happens to
1673 		 * be the only place where the TX status bits
1674 		 * are valid.
1675 		 */
1676 		if (txstat & RE_TDESC_CMD_EOF) {
1677 			m_freem(sc->re_ldata.re_tx_mbuf[idx]);
1678 			sc->re_ldata.re_tx_mbuf[idx] = NULL;
1679 			bus_dmamap_unload(sc->re_ldata.re_mtag,
1680 			    sc->re_ldata.re_tx_dmamap[idx]);
1681 			if (txstat & (RE_TDESC_STAT_EXCESSCOL|
1682 			    RE_TDESC_STAT_COLCNT))
1683 				ifp->if_collisions++;
1684 			if (txstat & RE_TDESC_STAT_TXERRSUM)
1685 				ifp->if_oerrors++;
1686 			else
1687 				ifp->if_opackets++;
1688 		}
1689 		sc->re_ldata.re_tx_free++;
1690 	}
1691 
1692 	/* No changes made to the TX ring, so no flush needed */
1693 	if (sc->re_ldata.re_tx_free) {
1694 		sc->re_ldata.re_tx_considx = idx;
1695 		ifp->if_flags &= ~IFF_OACTIVE;
1696 		ifp->if_timer = 0;
1697 	}
1698 
1699 	/*
1700 	 * Some chips will ignore a second TX request issued while an
1701 	 * existing transmission is in progress. If the transmitter goes
1702 	 * idle but there are still packets waiting to be sent, we need
1703 	 * to restart the channel here to flush them out. This only seems
1704 	 * to be required with the PCIe devices.
1705 	 */
1706 	if (sc->re_ldata.re_tx_free < RE_TX_DESC_CNT)
1707 		CSR_WRITE_1(sc, sc->re_txstart, RE_TXSTART_START);
1708 
1709 	/*
1710 	 * If not all descriptors have been released reaped yet,
1711 	 * reload the timer so that we will eventually get another
1712 	 * interrupt that will cause us to re-enter this routine.
1713 	 * This is done in case the transmitter has gone idle.
1714 	 */
1715 	if (RE_TX_MODERATION_IS_ENABLED(sc) &&
1716 	    sc->re_ldata.re_tx_free < RE_TX_DESC_CNT)
1717                 CSR_WRITE_4(sc, RE_TIMERCNT, 1);
1718 }
1719 
1720 static void
1721 re_tick(void *xsc)
1722 {
1723 	struct re_softc *sc = xsc;
1724 
1725 	lwkt_serialize_enter(sc->arpcom.ac_if.if_serializer);
1726 	re_tick_serialized(xsc);
1727 	lwkt_serialize_exit(sc->arpcom.ac_if.if_serializer);
1728 }
1729 
1730 static void
1731 re_tick_serialized(void *xsc)
1732 {
1733 	struct re_softc *sc = xsc;
1734 	struct ifnet *ifp = &sc->arpcom.ac_if;
1735 	struct mii_data *mii;
1736 
1737 	mii = device_get_softc(sc->re_miibus);
1738 	mii_tick(mii);
1739 	if (sc->re_link) {
1740 		if (!(mii->mii_media_status & IFM_ACTIVE))
1741 			sc->re_link = 0;
1742 	} else {
1743 		if (mii->mii_media_status & IFM_ACTIVE &&
1744 		    IFM_SUBTYPE(mii->mii_media_active) != IFM_NONE) {
1745 			sc->re_link = 1;
1746 			if (!ifq_is_empty(&ifp->if_snd))
1747 				if_devstart(ifp);
1748 		}
1749 	}
1750 
1751 	callout_reset(&sc->re_timer, hz, re_tick, sc);
1752 }
1753 
1754 #ifdef DEVICE_POLLING
1755 
1756 static void
1757 re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
1758 {
1759 	struct re_softc *sc = ifp->if_softc;
1760 
1761 	switch(cmd) {
1762 	case POLL_REGISTER:
1763 		/* disable interrupts */
1764 		CSR_WRITE_2(sc, RE_IMR, 0x0000);
1765 		break;
1766 	case POLL_DEREGISTER:
1767 		/* enable interrupts */
1768 		CSR_WRITE_2(sc, RE_IMR, sc->re_intrs);
1769 		break;
1770 	default:
1771 		sc->rxcycles = count;
1772 		re_rxeof(sc);
1773 		re_txeof(sc);
1774 
1775 		if (!ifq_is_empty(&ifp->if_snd))
1776 			if_devstart(ifp);
1777 
1778 		if (cmd == POLL_AND_CHECK_STATUS) { /* also check status register */
1779 			uint16_t       status;
1780 
1781 			status = CSR_READ_2(sc, RE_ISR);
1782 			if (status == 0xffff)
1783 				return;
1784 			if (status)
1785 				CSR_WRITE_2(sc, RE_ISR, status);
1786 
1787 			/*
1788 			 * XXX check behaviour on receiver stalls.
1789 			 */
1790 
1791 			if (status & RE_ISR_SYSTEM_ERR) {
1792 				re_reset(sc);
1793 				re_init(sc);
1794 			}
1795 		}
1796 		break;
1797 	}
1798 }
1799 #endif /* DEVICE_POLLING */
1800 
1801 static void
1802 re_intr(void *arg)
1803 {
1804 	struct re_softc	*sc = arg;
1805 	struct ifnet *ifp = &sc->arpcom.ac_if;
1806 	uint16_t status;
1807 
1808 	if (sc->suspended || (ifp->if_flags & IFF_UP) == 0)
1809 		return;
1810 
1811 	for (;;) {
1812 		status = CSR_READ_2(sc, RE_ISR);
1813 		/* If the card has gone away the read returns 0xffff. */
1814 		if (status == 0xffff)
1815 			break;
1816 		if (status)
1817 			CSR_WRITE_2(sc, RE_ISR, status);
1818 
1819 		if ((status & sc->re_intrs) == 0)
1820 			break;
1821 
1822 		if (status & (RE_ISR_RX_OK | RE_ISR_RX_ERR | RE_ISR_FIFO_OFLOW))
1823 			re_rxeof(sc);
1824 
1825 		if ((status & sc->re_tx_ack) ||
1826 		    (status & RE_ISR_TX_ERR) ||
1827 		    (status & RE_ISR_TX_DESC_UNAVAIL))
1828 			re_txeof(sc);
1829 
1830 		if (status & RE_ISR_SYSTEM_ERR) {
1831 			re_reset(sc);
1832 			re_init(sc);
1833 		}
1834 
1835 		if (status & RE_ISR_LINKCHG) {
1836 			callout_stop(&sc->re_timer);
1837 			re_tick_serialized(sc);
1838 		}
1839 	}
1840 
1841 	if (!ifq_is_empty(&ifp->if_snd))
1842 		if_devstart(ifp);
1843 }
1844 
1845 static int
1846 re_encap(struct re_softc *sc, struct mbuf **m_head, int *idx, int *called_defrag)
1847 {
1848 	struct ifnet *ifp = &sc->arpcom.ac_if;
1849 	struct mbuf *m, *m_new = NULL;
1850 	struct re_dmaload_arg	arg;
1851 	bus_dmamap_t		map;
1852 	int			error;
1853 
1854 	KASSERT(sc->re_ldata.re_tx_free > 4, ("not enough free TX desc\n"));
1855 
1856 	*called_defrag = 0;
1857 	m = *m_head;
1858 
1859 	/*
1860 	 * Set up checksum offload. Note: checksum offload bits must
1861 	 * appear in all descriptors of a multi-descriptor transmit
1862 	 * attempt. (This is according to testing done with an 8169
1863 	 * chip. I'm not sure if this is a requirement or a bug.)
1864 	 */
1865 
1866 	arg.re_flags = 0;
1867 
1868 	if (m->m_pkthdr.csum_flags & CSUM_IP)
1869 		arg.re_flags |= RE_TDESC_CMD_IPCSUM;
1870 	if (m->m_pkthdr.csum_flags & CSUM_TCP)
1871 		arg.re_flags |= RE_TDESC_CMD_TCPCSUM;
1872 	if (m->m_pkthdr.csum_flags & CSUM_UDP)
1873 		arg.re_flags |= RE_TDESC_CMD_UDPCSUM;
1874 
1875 	arg.sc = sc;
1876 	arg.re_idx = *idx;
1877 	arg.re_maxsegs = sc->re_ldata.re_tx_free;
1878 	if (arg.re_maxsegs > 4)
1879 		arg.re_maxsegs -= 4;
1880 	arg.re_ring = sc->re_ldata.re_tx_list;
1881 
1882 	map = sc->re_ldata.re_tx_dmamap[*idx];
1883 
1884 	/*
1885 	 * With some of the RealTek chips, using the checksum offload
1886 	 * support in conjunction with the autopadding feature results
1887 	 * in the transmission of corrupt frames. For example, if we
1888 	 * need to send a really small IP fragment that's less than 60
1889 	 * bytes in size, and IP header checksumming is enabled, the
1890 	 * resulting ethernet frame that appears on the wire will
1891 	 * have garbled payload. To work around this, if TX checksum
1892 	 * offload is enabled, we always manually pad short frames out
1893 	 * to the minimum ethernet frame size. We do this by pretending
1894 	 * the mbuf chain has too many fragments so the coalescing code
1895 	 * below can assemble the packet into a single buffer that's
1896 	 * padded out to the mininum frame size.
1897 	 *
1898 	 * Note: this appears unnecessary for TCP, and doing it for TCP
1899 	 * with PCIe adapters seems to result in bad checksums.
1900 	 */
1901 	if (arg.re_flags && !(arg.re_flags & RE_TDESC_CMD_TCPCSUM) &&
1902 	    m->m_pkthdr.len < RE_MIN_FRAMELEN) {
1903 		error = EFBIG;
1904 	} else {
1905 		error = bus_dmamap_load_mbuf(sc->re_ldata.re_mtag, map,
1906 		    m, re_dma_map_desc, &arg, BUS_DMA_NOWAIT);
1907 	}
1908 
1909 	if (error && error != EFBIG) {
1910 		if_printf(ifp, "can't map mbuf (error %d)\n", error);
1911 		return(ENOBUFS);
1912 	}
1913 
1914 	/* Too many segments to map, coalesce into a single mbuf */
1915 
1916 	if (error || arg.re_maxsegs == 0) {
1917 		m_new = m_defrag_nofree(m, MB_DONTWAIT);
1918 		if (m_new == NULL) {
1919 			return(1);
1920 		} else {
1921 			m = m_new;
1922 			*m_head = m;
1923 		}
1924 
1925 		/*
1926 		 * Manually pad short frames, and zero the pad space
1927 		 * to avoid leaking data.
1928 		 */
1929 		if (m_new->m_pkthdr.len < RE_MIN_FRAMELEN) {
1930 			bzero(mtod(m_new, char *) + m_new->m_pkthdr.len,
1931 			    RE_MIN_FRAMELEN - m_new->m_pkthdr.len);
1932 			m_new->m_pkthdr.len += RE_MIN_FRAMELEN -
1933 			    m_new->m_pkthdr.len;
1934 			m_new->m_len = m_new->m_pkthdr.len;
1935 		}
1936 
1937 		*called_defrag = 1;
1938 		arg.sc = sc;
1939 		arg.re_idx = *idx;
1940 		arg.re_maxsegs = sc->re_ldata.re_tx_free;
1941 		arg.re_ring = sc->re_ldata.re_tx_list;
1942 
1943 		error = bus_dmamap_load_mbuf(sc->re_ldata.re_mtag, map,
1944 		    m, re_dma_map_desc, &arg, BUS_DMA_NOWAIT);
1945 		if (error) {
1946 			m_freem(m);
1947 			if_printf(ifp, "can't map mbuf (error %d)\n", error);
1948 			return(EFBIG);
1949 		}
1950 	}
1951 
1952 	/*
1953 	 * Insure that the map for this transmission
1954 	 * is placed at the array index of the last descriptor
1955 	 * in this chain.
1956 	 */
1957 	sc->re_ldata.re_tx_dmamap[*idx] =
1958 	    sc->re_ldata.re_tx_dmamap[arg.re_idx];
1959 	sc->re_ldata.re_tx_dmamap[arg.re_idx] = map;
1960 
1961 	sc->re_ldata.re_tx_mbuf[arg.re_idx] = m;
1962 	sc->re_ldata.re_tx_free -= arg.re_maxsegs;
1963 
1964 	/*
1965 	 * Set up hardware VLAN tagging. Note: vlan tag info must
1966 	 * appear in the first descriptor of a multi-descriptor
1967 	 * transmission attempt.
1968 	 */
1969 
1970 	if (m->m_flags & M_VLANTAG) {
1971 		sc->re_ldata.re_tx_list[*idx].re_vlanctl =
1972 		    htole32(htobe16(m->m_pkthdr.ether_vlantag) |
1973 		    	    RE_TDESC_VLANCTL_TAG);
1974 	}
1975 
1976 	/* Transfer ownership of packet to the chip. */
1977 
1978 	sc->re_ldata.re_tx_list[arg.re_idx].re_cmdstat |=
1979 	    htole32(RE_TDESC_CMD_OWN);
1980 	if (*idx != arg.re_idx)
1981 		sc->re_ldata.re_tx_list[*idx].re_cmdstat |=
1982 		    htole32(RE_TDESC_CMD_OWN);
1983 
1984 	RE_DESC_INC(arg.re_idx);
1985 	*idx = arg.re_idx;
1986 
1987 	return(0);
1988 }
1989 
1990 /*
1991  * Main transmit routine for C+ and gigE NICs.
1992  */
1993 
1994 static void
1995 re_start(struct ifnet *ifp)
1996 {
1997 	struct re_softc	*sc = ifp->if_softc;
1998 	struct mbuf *m_head;
1999 	struct mbuf *m_head2;
2000 	int called_defrag, idx, need_trans;
2001 
2002 	if (!sc->re_link) {
2003 		ifq_purge(&ifp->if_snd);
2004 		return;
2005 	}
2006 
2007 	if ((ifp->if_flags & (IFF_OACTIVE | IFF_RUNNING)) != IFF_RUNNING)
2008 		return;
2009 
2010 	idx = sc->re_ldata.re_tx_prodidx;
2011 
2012 	need_trans = 0;
2013 	while (sc->re_ldata.re_tx_mbuf[idx] == NULL) {
2014 		if (sc->re_ldata.re_tx_free <= 4) {
2015 			ifp->if_flags |= IFF_OACTIVE;
2016 			break;
2017 		}
2018 
2019 		m_head = ifq_dequeue(&ifp->if_snd, NULL);
2020 		if (m_head == NULL)
2021 			break;
2022 
2023 		m_head2 = m_head;
2024 		if (re_encap(sc, &m_head2, &idx, &called_defrag)) {
2025 			/*
2026 			 * If we could not encapsulate the defragged packet,
2027 			 * the returned m_head2 is garbage and we must dequeue
2028 			 * and throw away the original packet.
2029 			 */
2030 			if (called_defrag)
2031 				m_freem(m_head);
2032 			ifp->if_flags |= IFF_OACTIVE;
2033 			break;
2034 		}
2035 
2036 		/*
2037 		 * Clean out the packet we encapsulated.  If we defragged
2038 		 * the packet the m_head2 is the one that got encapsulated
2039 		 * and the original must be thrown away.  Otherwise m_head2
2040 		 * *IS* the original.
2041 		 */
2042 		if (called_defrag)
2043 			m_freem(m_head);
2044 		need_trans = 1;
2045 
2046 		/*
2047 		 * If there's a BPF listener, bounce a copy of this frame
2048 		 * to him.
2049 		 */
2050 		ETHER_BPF_MTAP(ifp, m_head2);
2051 	}
2052 
2053 	if (!need_trans) {
2054 		if (RE_TX_MODERATION_IS_ENABLED(sc) &&
2055 		    sc->re_ldata.re_tx_free != RE_TX_DESC_CNT)
2056 			CSR_WRITE_4(sc, RE_TIMERCNT, 1);
2057 		return;
2058 	}
2059 
2060 	/* Flush the TX descriptors */
2061 	bus_dmamap_sync(sc->re_ldata.re_tx_list_tag,
2062 			sc->re_ldata.re_tx_list_map, BUS_DMASYNC_PREWRITE);
2063 
2064 	sc->re_ldata.re_tx_prodidx = idx;
2065 
2066 	/*
2067 	 * RealTek put the TX poll request register in a different
2068 	 * location on the 8169 gigE chip. I don't know why.
2069 	 */
2070 	CSR_WRITE_1(sc, sc->re_txstart, RE_TXSTART_START);
2071 
2072 	if (RE_TX_MODERATION_IS_ENABLED(sc)) {
2073 		/*
2074 		 * Use the countdown timer for interrupt moderation.
2075 		 * 'TX done' interrupts are disabled. Instead, we reset the
2076 		 * countdown timer, which will begin counting until it hits
2077 		 * the value in the TIMERINT register, and then trigger an
2078 		 * interrupt. Each time we write to the TIMERCNT register,
2079 		 * the timer count is reset to 0.
2080 		 */
2081 		CSR_WRITE_4(sc, RE_TIMERCNT, 1);
2082 	}
2083 
2084 	/*
2085 	 * Set a timeout in case the chip goes out to lunch.
2086 	 */
2087 	ifp->if_timer = 5;
2088 }
2089 
2090 static void
2091 re_init(void *xsc)
2092 {
2093 	struct re_softc *sc = xsc;
2094 	struct ifnet *ifp = &sc->arpcom.ac_if;
2095 	struct mii_data *mii;
2096 	uint32_t rxcfg = 0;
2097 
2098 	mii = device_get_softc(sc->re_miibus);
2099 
2100 	/*
2101 	 * Cancel pending I/O and free all RX/TX buffers.
2102 	 */
2103 	re_stop(sc);
2104 
2105 	/*
2106 	 * Enable C+ RX and TX mode, as well as VLAN stripping and
2107 	 * RX checksum offload. We must configure the C+ register
2108 	 * before all others.
2109 	 */
2110 	CSR_WRITE_2(sc, RE_CPLUS_CMD, RE_CPLUSCMD_RXENB | RE_CPLUSCMD_TXENB |
2111 		    RE_CPLUSCMD_PCI_MRW | RE_CPLUSCMD_VLANSTRIP |
2112 		    (ifp->if_capenable & IFCAP_RXCSUM ?
2113 		     RE_CPLUSCMD_RXCSUM_ENB : 0));
2114 
2115 	/*
2116 	 * Init our MAC address.  Even though the chipset
2117 	 * documentation doesn't mention it, we need to enter "Config
2118 	 * register write enable" mode to modify the ID registers.
2119 	 */
2120 	CSR_WRITE_1(sc, RE_EECMD, RE_EEMODE_WRITECFG);
2121 	CSR_WRITE_4(sc, RE_IDR0,
2122 	    htole32(*(uint32_t *)(&sc->arpcom.ac_enaddr[0])));
2123 	CSR_WRITE_2(sc, RE_IDR4,
2124 	    htole16(*(uint16_t *)(&sc->arpcom.ac_enaddr[4])));
2125 	CSR_WRITE_1(sc, RE_EECMD, RE_EEMODE_OFF);
2126 
2127 	/*
2128 	 * For C+ mode, initialize the RX descriptors and mbufs.
2129 	 */
2130 	re_rx_list_init(sc);
2131 	re_tx_list_init(sc);
2132 
2133 	/*
2134 	 * Load the addresses of the RX and TX lists into the chip.
2135 	 */
2136 	CSR_WRITE_4(sc, RE_RXLIST_ADDR_HI,
2137 	    RE_ADDR_HI(sc->re_ldata.re_rx_list_addr));
2138 	CSR_WRITE_4(sc, RE_RXLIST_ADDR_LO,
2139 	    RE_ADDR_LO(sc->re_ldata.re_rx_list_addr));
2140 
2141 	CSR_WRITE_4(sc, RE_TXLIST_ADDR_HI,
2142 	    RE_ADDR_HI(sc->re_ldata.re_tx_list_addr));
2143 	CSR_WRITE_4(sc, RE_TXLIST_ADDR_LO,
2144 	    RE_ADDR_LO(sc->re_ldata.re_tx_list_addr));
2145 
2146 	/*
2147 	 * Enable transmit and receive.
2148 	 */
2149 	CSR_WRITE_1(sc, RE_COMMAND, RE_CMD_TX_ENB|RE_CMD_RX_ENB);
2150 
2151 	/*
2152 	 * Set the initial TX and RX configuration.
2153 	 */
2154 	if (sc->re_testmode) {
2155 		if (sc->re_type == RE_8169)
2156 			CSR_WRITE_4(sc, RE_TXCFG,
2157 				    RE_TXCFG_CONFIG | RE_LOOPTEST_ON);
2158 		else
2159 			CSR_WRITE_4(sc, RE_TXCFG,
2160 				    RE_TXCFG_CONFIG | RE_LOOPTEST_ON_CPLUS);
2161 	} else
2162 		CSR_WRITE_4(sc, RE_TXCFG, RE_TXCFG_CONFIG);
2163 
2164 	CSR_WRITE_1(sc, RE_EARLY_TX_THRESH, 16);
2165 
2166 	CSR_WRITE_4(sc, RE_RXCFG, RE_RXCFG_CONFIG);
2167 
2168 	/* Set the individual bit to receive frames for this host only. */
2169 	rxcfg = CSR_READ_4(sc, RE_RXCFG);
2170 	rxcfg |= RE_RXCFG_RX_INDIV;
2171 
2172 	/* If we want promiscuous mode, set the allframes bit. */
2173 	if (ifp->if_flags & IFF_PROMISC) {
2174 		rxcfg |= RE_RXCFG_RX_ALLPHYS;
2175 		CSR_WRITE_4(sc, RE_RXCFG, rxcfg);
2176 	} else {
2177 		rxcfg &= ~RE_RXCFG_RX_ALLPHYS;
2178 		CSR_WRITE_4(sc, RE_RXCFG, rxcfg);
2179 	}
2180 
2181 	/*
2182 	 * Set capture broadcast bit to capture broadcast frames.
2183 	 */
2184 	if (ifp->if_flags & IFF_BROADCAST) {
2185 		rxcfg |= RE_RXCFG_RX_BROAD;
2186 		CSR_WRITE_4(sc, RE_RXCFG, rxcfg);
2187 	} else {
2188 		rxcfg &= ~RE_RXCFG_RX_BROAD;
2189 		CSR_WRITE_4(sc, RE_RXCFG, rxcfg);
2190 	}
2191 
2192 	/*
2193 	 * Program the multicast filter, if necessary.
2194 	 */
2195 	re_setmulti(sc);
2196 
2197 #ifdef DEVICE_POLLING
2198 	/*
2199 	 * Disable interrupts if we are polling.
2200 	 */
2201 	if (ifp->if_flags & IFF_POLLING)
2202 		CSR_WRITE_2(sc, RE_IMR, 0);
2203 	else	/* otherwise ... */
2204 #endif /* DEVICE_POLLING */
2205 	/*
2206 	 * Enable interrupts.
2207 	 */
2208 	if (sc->re_testmode)
2209 		CSR_WRITE_2(sc, RE_IMR, 0);
2210 	else
2211 		CSR_WRITE_2(sc, RE_IMR, sc->re_intrs);
2212 	CSR_WRITE_2(sc, RE_ISR, sc->re_intrs);
2213 
2214 	/* Set initial TX threshold */
2215 	sc->re_txthresh = RE_TX_THRESH_INIT;
2216 
2217 	/* Start RX/TX process. */
2218 	if (sc->re_flags & RE_F_HASMPC)
2219 		CSR_WRITE_4(sc, RE_MISSEDPKT, 0);
2220 #ifdef notdef
2221 	/* Enable receiver and transmitter. */
2222 	CSR_WRITE_1(sc, RE_COMMAND, RE_CMD_TX_ENB|RE_CMD_RX_ENB);
2223 #endif
2224 
2225 	if (RE_TX_MODERATION_IS_ENABLED(sc)) {
2226 		/*
2227 		 * Initialize the timer interrupt register so that
2228 		 * a timer interrupt will be generated once the timer
2229 		 * reaches a certain number of ticks. The timer is
2230 		 * reloaded on each transmit. This gives us TX interrupt
2231 		 * moderation, which dramatically improves TX frame rate.
2232 		 */
2233 		if (sc->re_type == RE_8169)
2234 			CSR_WRITE_4(sc, RE_TIMERINT_8169, 0x800);
2235 		else
2236 			CSR_WRITE_4(sc, RE_TIMERINT, 0x400);
2237 	}
2238 
2239 	/*
2240 	 * For 8169 gigE NICs, set the max allowed RX packet
2241 	 * size so we can receive jumbo frames.
2242 	 */
2243 	if (sc->re_type == RE_8169)
2244 		CSR_WRITE_2(sc, RE_MAXRXPKTLEN, 16383);
2245 
2246 	if (sc->re_testmode) {
2247 		return;
2248 	}
2249 
2250 	mii_mediachg(mii);
2251 
2252 	CSR_WRITE_1(sc, RE_CFG1, RE_CFG1_DRVLOAD|RE_CFG1_FULLDUPLEX);
2253 
2254 	ifp->if_flags |= IFF_RUNNING;
2255 	ifp->if_flags &= ~IFF_OACTIVE;
2256 
2257 	sc->re_link = 0;
2258 	callout_reset(&sc->re_timer, hz, re_tick, sc);
2259 }
2260 
2261 /*
2262  * Set media options.
2263  */
2264 static int
2265 re_ifmedia_upd(struct ifnet *ifp)
2266 {
2267 	struct re_softc *sc = ifp->if_softc;
2268 	struct mii_data *mii;
2269 
2270 	mii = device_get_softc(sc->re_miibus);
2271 	mii_mediachg(mii);
2272 
2273 	return(0);
2274 }
2275 
2276 /*
2277  * Report current media status.
2278  */
2279 static void
2280 re_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
2281 {
2282 	struct re_softc *sc = ifp->if_softc;
2283 	struct mii_data *mii;
2284 
2285 	mii = device_get_softc(sc->re_miibus);
2286 
2287 	mii_pollstat(mii);
2288 	ifmr->ifm_active = mii->mii_media_active;
2289 	ifmr->ifm_status = mii->mii_media_status;
2290 }
2291 
2292 static int
2293 re_ioctl(struct ifnet *ifp, u_long command, caddr_t data, struct ucred *cr)
2294 {
2295 	struct re_softc *sc = ifp->if_softc;
2296 	struct ifreq *ifr = (struct ifreq *) data;
2297 	struct mii_data *mii;
2298 	int error = 0;
2299 
2300 	switch(command) {
2301 	case SIOCSIFMTU:
2302 		if (ifr->ifr_mtu > RE_JUMBO_MTU)
2303 			error = EINVAL;
2304 		ifp->if_mtu = ifr->ifr_mtu;
2305 		break;
2306 	case SIOCSIFFLAGS:
2307 		if (ifp->if_flags & IFF_UP)
2308 			re_init(sc);
2309 		else if (ifp->if_flags & IFF_RUNNING)
2310 			re_stop(sc);
2311 		break;
2312 	case SIOCADDMULTI:
2313 	case SIOCDELMULTI:
2314 		re_setmulti(sc);
2315 		error = 0;
2316 		break;
2317 	case SIOCGIFMEDIA:
2318 	case SIOCSIFMEDIA:
2319 		mii = device_get_softc(sc->re_miibus);
2320 		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, command);
2321 		break;
2322 	case SIOCSIFCAP:
2323 		ifp->if_capenable &= ~(IFCAP_HWCSUM);
2324 		ifp->if_capenable |=
2325 		    ifr->ifr_reqcap & (IFCAP_HWCSUM);
2326 		if (ifp->if_capenable & IFCAP_TXCSUM)
2327 			ifp->if_hwassist = RE_CSUM_FEATURES;
2328 		else
2329 			ifp->if_hwassist = 0;
2330 		if (ifp->if_flags & IFF_RUNNING)
2331 			re_init(sc);
2332 		break;
2333 	default:
2334 		error = ether_ioctl(ifp, command, data);
2335 		break;
2336 	}
2337 	return(error);
2338 }
2339 
2340 static void
2341 re_watchdog(struct ifnet *ifp)
2342 {
2343 	struct re_softc *sc = ifp->if_softc;
2344 
2345 	if_printf(ifp, "watchdog timeout\n");
2346 
2347 	ifp->if_oerrors++;
2348 
2349 	re_txeof(sc);
2350 	re_rxeof(sc);
2351 
2352 	re_init(sc);
2353 
2354 	if (!ifq_is_empty(&ifp->if_snd))
2355 		if_devstart(ifp);
2356 }
2357 
2358 /*
2359  * Stop the adapter and free any mbufs allocated to the
2360  * RX and TX lists.
2361  */
2362 static void
2363 re_stop(struct re_softc *sc)
2364 {
2365 	struct ifnet *ifp = &sc->arpcom.ac_if;
2366 	int i;
2367 
2368 	ifp->if_timer = 0;
2369 	callout_stop(&sc->re_timer);
2370 
2371 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
2372 
2373 	CSR_WRITE_1(sc, RE_COMMAND, 0x00);
2374 	CSR_WRITE_2(sc, RE_IMR, 0x0000);
2375 	CSR_WRITE_2(sc, RE_ISR, 0xFFFF);
2376 
2377 	if (sc->re_head != NULL) {
2378 		m_freem(sc->re_head);
2379 		sc->re_head = sc->re_tail = NULL;
2380 	}
2381 
2382 	/* Free the TX list buffers. */
2383 	for (i = 0; i < RE_TX_DESC_CNT; i++) {
2384 		if (sc->re_ldata.re_tx_mbuf[i] != NULL) {
2385 			bus_dmamap_unload(sc->re_ldata.re_mtag,
2386 					  sc->re_ldata.re_tx_dmamap[i]);
2387 			m_freem(sc->re_ldata.re_tx_mbuf[i]);
2388 			sc->re_ldata.re_tx_mbuf[i] = NULL;
2389 		}
2390 	}
2391 
2392 	/* Free the RX list buffers. */
2393 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
2394 		if (sc->re_ldata.re_rx_mbuf[i] != NULL) {
2395 			bus_dmamap_unload(sc->re_ldata.re_mtag,
2396 					  sc->re_ldata.re_rx_dmamap[i]);
2397 			m_freem(sc->re_ldata.re_rx_mbuf[i]);
2398 			sc->re_ldata.re_rx_mbuf[i] = NULL;
2399 		}
2400 	}
2401 }
2402 
2403 /*
2404  * Device suspend routine.  Stop the interface and save some PCI
2405  * settings in case the BIOS doesn't restore them properly on
2406  * resume.
2407  */
2408 static int
2409 re_suspend(device_t dev)
2410 {
2411 #ifndef BURN_BRIDGES
2412 	int i;
2413 #endif
2414 	struct re_softc *sc = device_get_softc(dev);
2415 
2416 	re_stop(sc);
2417 
2418 #ifndef BURN_BRIDGES
2419 	for (i = 0; i < 5; i++)
2420 		sc->saved_maps[i] = pci_read_config(dev, PCIR_MAPS + i * 4, 4);
2421 	sc->saved_biosaddr = pci_read_config(dev, PCIR_BIOS, 4);
2422 	sc->saved_intline = pci_read_config(dev, PCIR_INTLINE, 1);
2423 	sc->saved_cachelnsz = pci_read_config(dev, PCIR_CACHELNSZ, 1);
2424 	sc->saved_lattimer = pci_read_config(dev, PCIR_LATTIMER, 1);
2425 #endif
2426 
2427 	sc->suspended = 1;
2428 
2429 	return (0);
2430 }
2431 
2432 /*
2433  * Device resume routine.  Restore some PCI settings in case the BIOS
2434  * doesn't, re-enable busmastering, and restart the interface if
2435  * appropriate.
2436  */
2437 static int
2438 re_resume(device_t dev)
2439 {
2440 	struct re_softc *sc = device_get_softc(dev);
2441 	struct ifnet *ifp = &sc->arpcom.ac_if;
2442 #ifndef BURN_BRIDGES
2443 	int i;
2444 #endif
2445 
2446 #ifndef BURN_BRIDGES
2447 	/* better way to do this? */
2448 	for (i = 0; i < 5; i++)
2449 		pci_write_config(dev, PCIR_MAPS + i * 4, sc->saved_maps[i], 4);
2450 	pci_write_config(dev, PCIR_BIOS, sc->saved_biosaddr, 4);
2451 	pci_write_config(dev, PCIR_INTLINE, sc->saved_intline, 1);
2452 	pci_write_config(dev, PCIR_CACHELNSZ, sc->saved_cachelnsz, 1);
2453 	pci_write_config(dev, PCIR_LATTIMER, sc->saved_lattimer, 1);
2454 
2455 	/* reenable busmastering */
2456 	pci_enable_busmaster(dev);
2457 	pci_enable_io(dev, SYS_RES_IOPORT);
2458 #endif
2459 
2460 	/* reinitialize interface if necessary */
2461 	if (ifp->if_flags & IFF_UP)
2462 		re_init(sc);
2463 
2464 	sc->suspended = 0;
2465 
2466 	return (0);
2467 }
2468 
2469 /*
2470  * Stop all chip I/O so that the kernel's probe routines don't
2471  * get confused by errant DMAs when rebooting.
2472  */
2473 static void
2474 re_shutdown(device_t dev)
2475 {
2476 	struct re_softc *sc = device_get_softc(dev);
2477 	struct ifnet *ifp = &sc->arpcom.ac_if;
2478 
2479 	lwkt_serialize_enter(ifp->if_serializer);
2480 	re_stop(sc);
2481 	lwkt_serialize_exit(ifp->if_serializer);
2482 }
2483 
2484 static int
2485 re_sysctl_tx_moderation(SYSCTL_HANDLER_ARGS)
2486 {
2487 	struct re_softc *sc = arg1;
2488 	struct ifnet *ifp = &sc->arpcom.ac_if;
2489 	int error = 0, mod, mod_old;
2490 
2491 	lwkt_serialize_enter(ifp->if_serializer);
2492 
2493 	mod_old = mod = RE_TX_MODERATION_IS_ENABLED(sc);
2494 
2495 	error = sysctl_handle_int(oidp, &mod, 0, req);
2496 	if (error || req->newptr == NULL || mod == mod_old)
2497 		goto back;
2498 	if (mod != 0 && mod != 1) {
2499 		error = EINVAL;
2500 		goto back;
2501 	}
2502 
2503 	if (mod)
2504 		RE_ENABLE_TX_MODERATION(sc);
2505 	else
2506 		RE_DISABLE_TX_MODERATION(sc);
2507 
2508 	if ((ifp->if_flags & (IFF_RUNNING | IFF_UP)) == (IFF_RUNNING | IFF_UP))
2509 		re_init(sc);
2510 back:
2511 	lwkt_serialize_exit(ifp->if_serializer);
2512 	return error;
2513 }
2514