xref: /dflybsd-src/sys/dev/netif/re/if_re.c (revision 99d569773b0a077e32a1ae305ce2ea743e686fa5)
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 1.48 2008/08/17 04:32:34 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 static int	re_suspend(device_t);
218 static int	re_resume(device_t);
219 static void	re_shutdown(device_t);
220 
221 static int	re_encap(struct re_softc *, struct mbuf **, int *, int *);
222 
223 static void	re_dma_map_addr(void *, bus_dma_segment_t *, int, int);
224 static void	re_dma_map_desc(void *, bus_dma_segment_t *, int,
225 				bus_size_t, int);
226 static int	re_allocmem(device_t, struct re_softc *);
227 static int	re_newbuf(struct re_softc *, int, struct mbuf *);
228 static int	re_rx_list_init(struct re_softc *);
229 static int	re_tx_list_init(struct re_softc *);
230 static void	re_rxeof(struct re_softc *);
231 static void	re_txeof(struct re_softc *);
232 static void	re_intr(void *);
233 static void	re_tick(void *);
234 static void	re_tick_serialized(void *);
235 
236 static void	re_start(struct ifnet *);
237 static int	re_ioctl(struct ifnet *, u_long, caddr_t, struct ucred *);
238 static void	re_init(void *);
239 static void	re_stop(struct re_softc *);
240 static void	re_watchdog(struct ifnet *);
241 static int	re_ifmedia_upd(struct ifnet *);
242 static void	re_ifmedia_sts(struct ifnet *, struct ifmediareq *);
243 
244 static void	re_eeprom_putbyte(struct re_softc *, int);
245 static void	re_eeprom_getword(struct re_softc *, int, u_int16_t *);
246 static void	re_read_eeprom(struct re_softc *, caddr_t, int, int);
247 static int	re_gmii_readreg(device_t, int, int);
248 static int	re_gmii_writereg(device_t, int, int, int);
249 
250 static int	re_miibus_readreg(device_t, int, int);
251 static int	re_miibus_writereg(device_t, int, int, int);
252 static void	re_miibus_statchg(device_t);
253 
254 static void	re_setmulti(struct re_softc *);
255 static void	re_reset(struct re_softc *);
256 
257 #ifdef RE_DIAG
258 static int	re_diag(struct re_softc *);
259 #endif
260 
261 #ifdef DEVICE_POLLING
262 static void	re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count);
263 #endif
264 
265 static int	re_sysctl_tx_moderation(SYSCTL_HANDLER_ARGS);
266 
267 static device_method_t re_methods[] = {
268 	/* Device interface */
269 	DEVMETHOD(device_probe,		re_probe),
270 	DEVMETHOD(device_attach,	re_attach),
271 	DEVMETHOD(device_detach,	re_detach),
272 	DEVMETHOD(device_suspend,	re_suspend),
273 	DEVMETHOD(device_resume,	re_resume),
274 	DEVMETHOD(device_shutdown,	re_shutdown),
275 
276 	/* bus interface */
277 	DEVMETHOD(bus_print_child,	bus_generic_print_child),
278 	DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
279 
280 	/* MII interface */
281 	DEVMETHOD(miibus_readreg,	re_miibus_readreg),
282 	DEVMETHOD(miibus_writereg,	re_miibus_writereg),
283 	DEVMETHOD(miibus_statchg,	re_miibus_statchg),
284 
285 	{ 0, 0 }
286 };
287 
288 static driver_t re_driver = {
289 	"re",
290 	re_methods,
291 	sizeof(struct re_softc)
292 };
293 
294 static devclass_t re_devclass;
295 
296 DECLARE_DUMMY_MODULE(if_re);
297 DRIVER_MODULE(if_re, pci, re_driver, re_devclass, 0, 0);
298 DRIVER_MODULE(if_re, cardbus, re_driver, re_devclass, 0, 0);
299 DRIVER_MODULE(miibus, re, miibus_driver, miibus_devclass, 0, 0);
300 
301 #define EE_SET(x)	\
302 	CSR_WRITE_1(sc, RE_EECMD, CSR_READ_1(sc, RE_EECMD) | (x))
303 
304 #define EE_CLR(x)	\
305 	CSR_WRITE_1(sc, RE_EECMD, CSR_READ_1(sc, RE_EECMD) & ~(x))
306 
307 /*
308  * Send a read command and address to the EEPROM, check for ACK.
309  */
310 static void
311 re_eeprom_putbyte(struct re_softc *sc, int addr)
312 {
313 	int d, i;
314 
315 	d = addr | (RE_9346_READ << sc->re_eewidth);
316 
317 	/*
318 	 * Feed in each bit and strobe the clock.
319 	 */
320 	for (i = 1 << (sc->re_eewidth + 3); i; i >>= 1) {
321 		if (d & i)
322 			EE_SET(RE_EE_DATAIN);
323 		else
324 			EE_CLR(RE_EE_DATAIN);
325 		DELAY(100);
326 		EE_SET(RE_EE_CLK);
327 		DELAY(150);
328 		EE_CLR(RE_EE_CLK);
329 		DELAY(100);
330 	}
331 }
332 
333 /*
334  * Read a word of data stored in the EEPROM at address 'addr.'
335  */
336 static void
337 re_eeprom_getword(struct re_softc *sc, int addr, uint16_t *dest)
338 {
339 	int i;
340 	uint16_t word = 0;
341 
342 	/*
343 	 * Send address of word we want to read.
344 	 */
345 	re_eeprom_putbyte(sc, addr);
346 
347 	/*
348 	 * Start reading bits from EEPROM.
349 	 */
350 	for (i = 0x8000; i != 0; i >>= 1) {
351 		EE_SET(RE_EE_CLK);
352 		DELAY(100);
353 		if (CSR_READ_1(sc, RE_EECMD) & RE_EE_DATAOUT)
354 			word |= i;
355 		EE_CLR(RE_EE_CLK);
356 		DELAY(100);
357 	}
358 
359 	*dest = word;
360 }
361 
362 /*
363  * Read a sequence of words from the EEPROM.
364  */
365 static void
366 re_read_eeprom(struct re_softc *sc, caddr_t dest, int off, int cnt)
367 {
368 	int i;
369 	uint16_t word = 0, *ptr;
370 
371 	CSR_SETBIT_1(sc, RE_EECMD, RE_EEMODE_PROGRAM);
372 	DELAY(100);
373 
374 	for (i = 0; i < cnt; i++) {
375 		CSR_SETBIT_1(sc, RE_EECMD, RE_EE_SEL);
376 		re_eeprom_getword(sc, off + i, &word);
377 		CSR_CLRBIT_1(sc, RE_EECMD, RE_EE_SEL);
378 		ptr = (uint16_t *)(dest + (i * 2));
379 		*ptr = word;
380 	}
381 
382 	CSR_CLRBIT_1(sc, RE_EECMD, RE_EEMODE_PROGRAM);
383 }
384 
385 static int
386 re_gmii_readreg(device_t dev, int phy, int reg)
387 {
388 	struct re_softc *sc = device_get_softc(dev);
389 	u_int32_t rval;
390 	int i;
391 
392 	if (phy != 1)
393 		return(0);
394 
395 	/* Let the rgephy driver read the GMEDIASTAT register */
396 
397 	if (reg == RE_GMEDIASTAT)
398 		return(CSR_READ_1(sc, RE_GMEDIASTAT));
399 
400 	CSR_WRITE_4(sc, RE_PHYAR, reg << 16);
401 	DELAY(1000);
402 
403 	for (i = 0; i < RE_TIMEOUT; i++) {
404 		rval = CSR_READ_4(sc, RE_PHYAR);
405 		if (rval & RE_PHYAR_BUSY)
406 			break;
407 		DELAY(100);
408 	}
409 
410 	if (i == RE_TIMEOUT) {
411 		device_printf(dev, "PHY read failed\n");
412 		return(0);
413 	}
414 
415 	return(rval & RE_PHYAR_PHYDATA);
416 }
417 
418 static int
419 re_gmii_writereg(device_t dev, int phy, int reg, int data)
420 {
421 	struct re_softc *sc = device_get_softc(dev);
422 	uint32_t rval;
423 	int i;
424 
425 	CSR_WRITE_4(sc, RE_PHYAR,
426 		    (reg << 16) | (data & RE_PHYAR_PHYDATA) | RE_PHYAR_BUSY);
427 	DELAY(1000);
428 
429 	for (i = 0; i < RE_TIMEOUT; i++) {
430 		rval = CSR_READ_4(sc, RE_PHYAR);
431 		if ((rval & RE_PHYAR_BUSY) == 0)
432 			break;
433 		DELAY(100);
434 	}
435 
436 	if (i == RE_TIMEOUT)
437 		device_printf(dev, "PHY write failed\n");
438 
439 	return(0);
440 }
441 
442 static int
443 re_miibus_readreg(device_t dev, int phy, int reg)
444 {
445 	struct re_softc	*sc = device_get_softc(dev);
446 	uint16_t rval = 0;
447 	uint16_t re8139_reg = 0;
448 
449 	if (sc->re_type == RE_8169) {
450 		rval = re_gmii_readreg(dev, phy, reg);
451 		return(rval);
452 	}
453 
454 	/* Pretend the internal PHY is only at address 0 */
455 	if (phy)
456 		return(0);
457 
458 	switch(reg) {
459 	case MII_BMCR:
460 		re8139_reg = RE_BMCR;
461 		break;
462 	case MII_BMSR:
463 		re8139_reg = RE_BMSR;
464 		break;
465 	case MII_ANAR:
466 		re8139_reg = RE_ANAR;
467 		break;
468 	case MII_ANER:
469 		re8139_reg = RE_ANER;
470 		break;
471 	case MII_ANLPAR:
472 		re8139_reg = RE_LPAR;
473 		break;
474 	case MII_PHYIDR1:
475 	case MII_PHYIDR2:
476 		return(0);
477 	/*
478 	 * Allow the rlphy driver to read the media status
479 	 * register. If we have a link partner which does not
480 	 * support NWAY, this is the register which will tell
481 	 * us the results of parallel detection.
482 	 */
483 	case RE_MEDIASTAT:
484 		return(CSR_READ_1(sc, RE_MEDIASTAT));
485 	default:
486 		device_printf(dev, "bad phy register\n");
487 		return(0);
488 	}
489 	rval = CSR_READ_2(sc, re8139_reg);
490 	if (sc->re_type == RE_8139CPLUS && re8139_reg == RE_BMCR) {
491 		/* 8139C+ has different bit layout. */
492 		rval &= ~(BMCR_LOOP | BMCR_ISO);
493 	}
494 	return(rval);
495 }
496 
497 static int
498 re_miibus_writereg(device_t dev, int phy, int reg, int data)
499 {
500 	struct re_softc *sc= device_get_softc(dev);
501 	u_int16_t re8139_reg = 0;
502 
503 	if (sc->re_type == RE_8169)
504 		return(re_gmii_writereg(dev, phy, reg, data));
505 
506 	/* Pretend the internal PHY is only at address 0 */
507 	if (phy)
508 		return(0);
509 
510 	switch(reg) {
511 	case MII_BMCR:
512 		re8139_reg = RE_BMCR;
513 		if (sc->re_type == RE_8139CPLUS) {
514 			/* 8139C+ has different bit layout. */
515 			data &= ~(BMCR_LOOP | BMCR_ISO);
516 		}
517 		break;
518 	case MII_BMSR:
519 		re8139_reg = RE_BMSR;
520 		break;
521 	case MII_ANAR:
522 		re8139_reg = RE_ANAR;
523 		break;
524 	case MII_ANER:
525 		re8139_reg = RE_ANER;
526 		break;
527 	case MII_ANLPAR:
528 		re8139_reg = RE_LPAR;
529 		break;
530 	case MII_PHYIDR1:
531 	case MII_PHYIDR2:
532 		return(0);
533 	default:
534 		device_printf(dev, "bad phy register\n");
535 		return(0);
536 	}
537 	CSR_WRITE_2(sc, re8139_reg, data);
538 	return(0);
539 }
540 
541 static void
542 re_miibus_statchg(device_t dev)
543 {
544 }
545 
546 /*
547  * Program the 64-bit multicast hash filter.
548  */
549 static void
550 re_setmulti(struct re_softc *sc)
551 {
552 	struct ifnet *ifp = &sc->arpcom.ac_if;
553 	int h = 0;
554 	uint32_t hashes[2] = { 0, 0 };
555 	struct ifmultiaddr *ifma;
556 	uint32_t rxfilt;
557 	int mcnt = 0;
558 
559 	rxfilt = CSR_READ_4(sc, RE_RXCFG);
560 
561 	if (ifp->if_flags & IFF_ALLMULTI || ifp->if_flags & IFF_PROMISC) {
562 		rxfilt |= RE_RXCFG_RX_MULTI;
563 		CSR_WRITE_4(sc, RE_RXCFG, rxfilt);
564 		CSR_WRITE_4(sc, RE_MAR0, 0xFFFFFFFF);
565 		CSR_WRITE_4(sc, RE_MAR4, 0xFFFFFFFF);
566 		return;
567 	}
568 
569 	/* first, zot all the existing hash bits */
570 	CSR_WRITE_4(sc, RE_MAR0, 0);
571 	CSR_WRITE_4(sc, RE_MAR4, 0);
572 
573 	/* now program new ones */
574 	LIST_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
575 		if (ifma->ifma_addr->sa_family != AF_LINK)
576 			continue;
577 		h = ether_crc32_be(LLADDR((struct sockaddr_dl *)
578 		    ifma->ifma_addr), ETHER_ADDR_LEN) >> 26;
579 		if (h < 32)
580 			hashes[0] |= (1 << h);
581 		else
582 			hashes[1] |= (1 << (h - 32));
583 		mcnt++;
584 	}
585 
586 	if (mcnt)
587 		rxfilt |= RE_RXCFG_RX_MULTI;
588 	else
589 		rxfilt &= ~RE_RXCFG_RX_MULTI;
590 
591 	CSR_WRITE_4(sc, RE_RXCFG, rxfilt);
592 
593 	/*
594 	 * For some unfathomable reason, RealTek decided to reverse
595 	 * the order of the multicast hash registers in the PCI Express
596 	 * parts. This means we have to write the hash pattern in reverse
597 	 * order for those devices.
598 	 */
599 	if (sc->re_flags & RE_F_PCIE) {
600 		CSR_WRITE_4(sc, RE_MAR0, bswap32(hashes[0]));
601 		CSR_WRITE_4(sc, RE_MAR4, bswap32(hashes[1]));
602 	} else {
603 		CSR_WRITE_4(sc, RE_MAR0, hashes[0]);
604 		CSR_WRITE_4(sc, RE_MAR4, hashes[1]);
605 	}
606 }
607 
608 static void
609 re_reset(struct re_softc *sc)
610 {
611 	int i;
612 
613 	CSR_WRITE_1(sc, RE_COMMAND, RE_CMD_RESET);
614 
615 	for (i = 0; i < RE_TIMEOUT; i++) {
616 		DELAY(10);
617 		if ((CSR_READ_1(sc, RE_COMMAND) & RE_CMD_RESET) == 0)
618 			break;
619 	}
620 	if (i == RE_TIMEOUT)
621 		if_printf(&sc->arpcom.ac_if, "reset never completed!\n");
622 
623 	CSR_WRITE_1(sc, 0x82, 1);
624 }
625 
626 #ifdef RE_DIAG
627 /*
628  * The following routine is designed to test for a defect on some
629  * 32-bit 8169 cards. Some of these NICs have the REQ64# and ACK64#
630  * lines connected to the bus, however for a 32-bit only card, they
631  * should be pulled high. The result of this defect is that the
632  * NIC will not work right if you plug it into a 64-bit slot: DMA
633  * operations will be done with 64-bit transfers, which will fail
634  * because the 64-bit data lines aren't connected.
635  *
636  * There's no way to work around this (short of talking a soldering
637  * iron to the board), however we can detect it. The method we use
638  * here is to put the NIC into digital loopback mode, set the receiver
639  * to promiscuous mode, and then try to send a frame. We then compare
640  * the frame data we sent to what was received. If the data matches,
641  * then the NIC is working correctly, otherwise we know the user has
642  * a defective NIC which has been mistakenly plugged into a 64-bit PCI
643  * slot. In the latter case, there's no way the NIC can work correctly,
644  * so we print out a message on the console and abort the device attach.
645  */
646 
647 static int
648 re_diag(struct re_softc *sc)
649 {
650 	struct ifnet *ifp = &sc->arpcom.ac_if;
651 	struct mbuf *m0;
652 	struct ether_header *eh;
653 	struct re_desc *cur_rx;
654 	uint16_t status;
655 	uint32_t rxstat;
656 	int total_len, i, error = 0, phyaddr;
657 	uint8_t dst[ETHER_ADDR_LEN] = { 0x00, 'h', 'e', 'l', 'l', 'o' };
658 	uint8_t src[ETHER_ADDR_LEN] = { 0x00, 'w', 'o', 'r', 'l', 'd' };
659 
660 	/* Allocate a single mbuf */
661 
662 	MGETHDR(m0, MB_DONTWAIT, MT_DATA);
663 	if (m0 == NULL)
664 		return(ENOBUFS);
665 
666 	/*
667 	 * Initialize the NIC in test mode. This sets the chip up
668 	 * so that it can send and receive frames, but performs the
669 	 * following special functions:
670 	 * - Puts receiver in promiscuous mode
671 	 * - Enables digital loopback mode
672 	 * - Leaves interrupts turned off
673 	 */
674 
675 	ifp->if_flags |= IFF_PROMISC;
676 	sc->re_testmode = 1;
677 	re_reset(sc);
678 	re_init(sc);
679 	sc->re_link = 1;
680 	if (sc->re_type == RE_8169)
681 		phyaddr = 1;
682 	else
683 		phyaddr = 0;
684 
685 	re_miibus_writereg(sc->re_dev, phyaddr, MII_BMCR, BMCR_RESET);
686 	for (i = 0; i < RE_TIMEOUT; i++) {
687 		status = re_miibus_readreg(sc->re_dev, phyaddr, MII_BMCR);
688 		if (!(status & BMCR_RESET))
689 			break;
690 	}
691 
692 	re_miibus_writereg(sc->re_dev, phyaddr, MII_BMCR, BMCR_LOOP);
693 	CSR_WRITE_2(sc, RE_ISR, RE_INTRS_DIAG);
694 
695 	DELAY(100000);
696 
697 	/* Put some data in the mbuf */
698 
699 	eh = mtod(m0, struct ether_header *);
700 	bcopy (dst, eh->ether_dhost, ETHER_ADDR_LEN);
701 	bcopy (src, eh->ether_shost, ETHER_ADDR_LEN);
702 	eh->ether_type = htons(ETHERTYPE_IP);
703 	m0->m_pkthdr.len = m0->m_len = ETHER_MIN_LEN - ETHER_CRC_LEN;
704 
705 	/*
706 	 * Queue the packet, start transmission.
707 	 * Note: ifq_handoff() ultimately calls re_start() for us.
708 	 */
709 
710 	CSR_WRITE_2(sc, RE_ISR, 0xFFFF);
711 	error = ifq_handoff(ifp, m0, NULL);
712 	if (error) {
713 		m0 = NULL;
714 		goto done;
715 	}
716 	m0 = NULL;
717 
718 	/* Wait for it to propagate through the chip */
719 
720 	DELAY(100000);
721 	for (i = 0; i < RE_TIMEOUT; i++) {
722 		status = CSR_READ_2(sc, RE_ISR);
723 		CSR_WRITE_2(sc, RE_ISR, status);
724 		if ((status & (RE_ISR_TIMEOUT_EXPIRED|RE_ISR_RX_OK)) ==
725 		    (RE_ISR_TIMEOUT_EXPIRED|RE_ISR_RX_OK))
726 			break;
727 		DELAY(10);
728 	}
729 
730 	if (i == RE_TIMEOUT) {
731 		if_printf(ifp, "diagnostic failed to receive packet "
732 			  "in loopback mode\n");
733 		error = EIO;
734 		goto done;
735 	}
736 
737 	/*
738 	 * The packet should have been dumped into the first
739 	 * entry in the RX DMA ring. Grab it from there.
740 	 */
741 
742 	bus_dmamap_sync(sc->re_ldata.re_rx_list_tag,
743 			sc->re_ldata.re_rx_list_map, BUS_DMASYNC_POSTREAD);
744 	bus_dmamap_sync(sc->re_ldata.re_mtag, sc->re_ldata.re_rx_dmamap[0],
745 			BUS_DMASYNC_POSTWRITE);
746 	bus_dmamap_unload(sc->re_ldata.re_mtag, sc->re_ldata.re_rx_dmamap[0]);
747 
748 	m0 = sc->re_ldata.re_rx_mbuf[0];
749 	sc->re_ldata.re_rx_mbuf[0] = NULL;
750 	eh = mtod(m0, struct ether_header *);
751 
752 	cur_rx = &sc->re_ldata.re_rx_list[0];
753 	total_len = RE_RXBYTES(cur_rx);
754 	rxstat = le32toh(cur_rx->re_cmdstat);
755 
756 	if (total_len != ETHER_MIN_LEN) {
757 		if_printf(ifp, "diagnostic failed, received short packet\n");
758 		error = EIO;
759 		goto done;
760 	}
761 
762 	/* Test that the received packet data matches what we sent. */
763 
764 	if (bcmp(eh->ether_dhost, dst, ETHER_ADDR_LEN) ||
765 	    bcmp(eh->ether_shost, &src, ETHER_ADDR_LEN) ||
766 	    be16toh(eh->ether_type) != ETHERTYPE_IP) {
767 		if_printf(ifp, "WARNING, DMA FAILURE!\n");
768 		if_printf(ifp, "expected TX data: %6D/%6D/0x%x\n",
769 		    dst, ":", src, ":", ETHERTYPE_IP);
770 		if_printf(ifp, "received RX data: %6D/%6D/0x%x\n",
771 		    eh->ether_dhost, ":",  eh->ether_shost, ":",
772 		    ntohs(eh->ether_type));
773 		if_printf(ifp, "You may have a defective 32-bit NIC plugged "
774 		    "into a 64-bit PCI slot.\n");
775 		if_printf(ifp, "Please re-install the NIC in a 32-bit slot "
776 		    "for proper operation.\n");
777 		if_printf(ifp, "Read the re(4) man page for more details.\n");
778 		error = EIO;
779 	}
780 
781 done:
782 	/* Turn interface off, release resources */
783 
784 	sc->re_testmode = 0;
785 	sc->re_link = 0;
786 	ifp->if_flags &= ~IFF_PROMISC;
787 	re_stop(sc);
788 	if (m0 != NULL)
789 		m_freem(m0);
790 
791 	return (error);
792 }
793 #endif	/* RE_DIAG */
794 
795 /*
796  * Probe for a RealTek 8139C+/8169/8110 chip. Check the PCI vendor and device
797  * IDs against our list and return a device name if we find a match.
798  */
799 static int
800 re_probe(device_t dev)
801 {
802 	const struct re_type *t;
803 	struct re_softc *sc;
804 	int rid;
805 	uint32_t hwrev;
806 	uint16_t vendor, product;
807 
808 	vendor = pci_get_vendor(dev);
809 	product = pci_get_device(dev);
810 
811 	/*
812 	 * Only attach to rev.3 of the Linksys EG1032 adapter.
813 	 * Rev.2 is supported by sk(4).
814 	 */
815 	if (vendor == PCI_VENDOR_LINKSYS &&
816 	    product == PCI_PRODUCT_LINKSYS_EG1032 &&
817 	    pci_get_subdevice(dev) != PCI_SUBDEVICE_LINKSYS_EG1032_REV3)
818 			return ENXIO;
819 
820 	for (t = re_devs; t->re_name != NULL; t++) {
821 		if (product == t->re_did && vendor == t->re_vid)
822 			break;
823 	}
824 
825 	/*
826 	 * Check if we found a RealTek device.
827 	 */
828 	if (t->re_name == NULL)
829 		return(ENXIO);
830 
831 	/*
832 	 * Temporarily map the I/O space so we can read the chip ID register.
833 	 */
834 	sc = kmalloc(sizeof(*sc), M_TEMP, M_WAITOK | M_ZERO);
835 	rid = RE_PCI_LOIO;
836 	sc->re_res = bus_alloc_resource_any(dev, SYS_RES_IOPORT, &rid,
837 					    RF_ACTIVE);
838 	if (sc->re_res == NULL) {
839 		device_printf(dev, "couldn't map ports/memory\n");
840 		kfree(sc, M_TEMP);
841 		return(ENXIO);
842 	}
843 
844 	sc->re_btag = rman_get_bustag(sc->re_res);
845 	sc->re_bhandle = rman_get_bushandle(sc->re_res);
846 
847 	hwrev = CSR_READ_4(sc, RE_TXCFG) & RE_TXCFG_HWREV;
848 	bus_release_resource(dev, SYS_RES_IOPORT, RE_PCI_LOIO, sc->re_res);
849 	kfree(sc, M_TEMP);
850 
851 	/*
852 	 * and continue matching for the specific chip...
853 	 */
854 	for (; t->re_name != NULL; t++) {
855 		if (product == t->re_did && vendor == t->re_vid &&
856 		    t->re_basetype == hwrev) {
857 			device_set_desc(dev, t->re_name);
858 			return(0);
859 		}
860 	}
861 
862 	if (bootverbose)
863 		kprintf("re: unknown hwrev %#x\n", hwrev);
864 	return(ENXIO);
865 }
866 
867 /*
868  * This routine takes the segment list provided as the result of
869  * a bus_dma_map_load() operation and assigns the addresses/lengths
870  * to RealTek DMA descriptors. This can be called either by the RX
871  * code or the TX code. In the RX case, we'll probably wind up mapping
872  * at most one segment. For the TX case, there could be any number of
873  * segments since TX packets may span multiple mbufs. In either case,
874  * if the number of segments is larger than the re_maxsegs limit
875  * specified by the caller, we abort the mapping operation. Sadly,
876  * whoever designed the buffer mapping API did not provide a way to
877  * return an error from here, so we have to fake it a bit.
878  */
879 
880 static void
881 re_dma_map_desc(void *arg, bus_dma_segment_t *segs, int nseg,
882 		bus_size_t mapsize, int error)
883 {
884 	struct re_dmaload_arg *ctx;
885 	struct re_desc *d = NULL;
886 	int i = 0, idx;
887 	uint32_t cmdstat;
888 
889 	if (error)
890 		return;
891 
892 	ctx = arg;
893 
894 	/* Signal error to caller if there's too many segments */
895 	if (nseg > ctx->re_maxsegs) {
896 		ctx->re_maxsegs = 0;
897 		return;
898 	}
899 
900 	/*
901 	 * Map the segment array into descriptors. Note that we set the
902 	 * start-of-frame and end-of-frame markers for either TX or RX, but
903 	 * they really only have meaning in the TX case. (In the RX case,
904 	 * it's the chip that tells us where packets begin and end.)
905 	 * We also keep track of the end of the ring and set the
906 	 * end-of-ring bits as needed, and we set the ownership bits
907 	 * in all except the very first descriptor. (The caller will
908 	 * set this descriptor later when it start transmission or
909 	 * reception.)
910 	 */
911 	idx = ctx->re_idx;
912 	for (;;) {
913 		d = &ctx->re_ring[idx];
914 		if (le32toh(d->re_cmdstat) & RE_RDESC_STAT_OWN) {
915 			ctx->re_maxsegs = 0;
916 			return;
917 		}
918 		cmdstat = segs[i].ds_len;
919 		d->re_bufaddr_lo = htole32(RE_ADDR_LO(segs[i].ds_addr));
920 		d->re_bufaddr_hi = htole32(RE_ADDR_HI(segs[i].ds_addr));
921 		if (i == 0)
922 			cmdstat |= RE_TDESC_CMD_SOF;
923 		else
924 			cmdstat |= RE_TDESC_CMD_OWN;
925 		if (idx == (RE_RX_DESC_CNT - 1))
926 			cmdstat |= RE_TDESC_CMD_EOR;
927 		d->re_cmdstat = htole32(cmdstat | ctx->re_flags);
928 		i++;
929 		if (i == nseg)
930 			break;
931 		RE_DESC_INC(idx);
932 	}
933 
934 	d->re_cmdstat |= htole32(RE_TDESC_CMD_EOF);
935 	ctx->re_maxsegs = nseg;
936 	ctx->re_idx = idx;
937 }
938 
939 /*
940  * Map a single buffer address.
941  */
942 
943 static void
944 re_dma_map_addr(void *arg, bus_dma_segment_t *segs, int nseg, int error)
945 {
946 	uint32_t *addr;
947 
948 	if (error)
949 		return;
950 
951 	KASSERT(nseg == 1, ("too many DMA segments, %d should be 1", nseg));
952 	addr = arg;
953 	*addr = segs->ds_addr;
954 }
955 
956 static int
957 re_allocmem(device_t dev, struct re_softc *sc)
958 {
959 	int error, i, nseg;
960 
961 	/*
962 	 * Allocate map for RX mbufs.
963 	 */
964 	nseg = 32;
965 	error = bus_dma_tag_create(sc->re_parent_tag, ETHER_ALIGN, 0,
966 	    BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
967 	    NULL, MCLBYTES * nseg, nseg, MCLBYTES, BUS_DMA_ALLOCNOW,
968 	    &sc->re_ldata.re_mtag);
969 	if (error) {
970 		device_printf(dev, "could not allocate dma tag\n");
971 		return(error);
972 	}
973 
974 	/*
975 	 * Allocate map for TX descriptor list.
976 	 */
977 	error = bus_dma_tag_create(sc->re_parent_tag, RE_RING_ALIGN,
978 	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
979             NULL, RE_TX_LIST_SZ, 1, RE_TX_LIST_SZ, BUS_DMA_ALLOCNOW,
980 	    &sc->re_ldata.re_tx_list_tag);
981 	if (error) {
982 		device_printf(dev, "could not allocate dma tag\n");
983 		return(error);
984 	}
985 
986 	/* Allocate DMA'able memory for the TX ring */
987 
988         error = bus_dmamem_alloc(sc->re_ldata.re_tx_list_tag,
989 	    (void **)&sc->re_ldata.re_tx_list, BUS_DMA_WAITOK | BUS_DMA_ZERO,
990             &sc->re_ldata.re_tx_list_map);
991         if (error) {
992 		device_printf(dev, "could not allocate TX ring\n");
993                 return(error);
994 	}
995 
996 	/* Load the map for the TX ring. */
997 
998 	error = bus_dmamap_load(sc->re_ldata.re_tx_list_tag,
999 	     sc->re_ldata.re_tx_list_map, sc->re_ldata.re_tx_list,
1000 	     RE_TX_LIST_SZ, re_dma_map_addr,
1001 	     &sc->re_ldata.re_tx_list_addr, BUS_DMA_NOWAIT);
1002 	if (error) {
1003 		device_printf(dev, "could not get address of TX ring\n");
1004 		return(error);
1005 	}
1006 
1007 	/* Create DMA maps for TX buffers */
1008 
1009 	for (i = 0; i < RE_TX_DESC_CNT; i++) {
1010 		error = bus_dmamap_create(sc->re_ldata.re_mtag, 0,
1011 			    &sc->re_ldata.re_tx_dmamap[i]);
1012 		if (error) {
1013 			device_printf(dev, "can't create DMA map for TX\n");
1014 			return(error);
1015 		}
1016 	}
1017 
1018 	/*
1019 	 * Allocate map for RX descriptor list.
1020 	 */
1021 	error = bus_dma_tag_create(sc->re_parent_tag, RE_RING_ALIGN,
1022 	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1023             NULL, RE_RX_LIST_SZ, 1, RE_RX_LIST_SZ, BUS_DMA_ALLOCNOW,
1024 	    &sc->re_ldata.re_rx_list_tag);
1025 	if (error) {
1026 		device_printf(dev, "could not allocate dma tag\n");
1027 		return(error);
1028 	}
1029 
1030 	/* Allocate DMA'able memory for the RX ring */
1031 
1032         error = bus_dmamem_alloc(sc->re_ldata.re_rx_list_tag,
1033 	    (void **)&sc->re_ldata.re_rx_list, BUS_DMA_WAITOK | BUS_DMA_ZERO,
1034             &sc->re_ldata.re_rx_list_map);
1035         if (error) {
1036 		device_printf(dev, "could not allocate RX ring\n");
1037                 return(error);
1038 	}
1039 
1040 	/* Load the map for the RX ring. */
1041 
1042 	error = bus_dmamap_load(sc->re_ldata.re_rx_list_tag,
1043 	     sc->re_ldata.re_rx_list_map, sc->re_ldata.re_rx_list,
1044 	     RE_RX_LIST_SZ, re_dma_map_addr,
1045 	     &sc->re_ldata.re_rx_list_addr, BUS_DMA_NOWAIT);
1046 	if (error) {
1047 		device_printf(dev, "could not get address of RX ring\n");
1048 		return(error);
1049 	}
1050 
1051 	/* Create DMA maps for RX buffers */
1052 
1053 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
1054 		error = bus_dmamap_create(sc->re_ldata.re_mtag, 0,
1055 			    &sc->re_ldata.re_rx_dmamap[i]);
1056 		if (error) {
1057 			device_printf(dev, "can't create DMA map for RX\n");
1058 			return(ENOMEM);
1059 		}
1060 	}
1061 
1062 	return(0);
1063 }
1064 
1065 /*
1066  * Attach the interface. Allocate softc structures, do ifmedia
1067  * setup and ethernet/BPF attach.
1068  */
1069 static int
1070 re_attach(device_t dev)
1071 {
1072 	struct re_softc	*sc = device_get_softc(dev);
1073 	struct ifnet *ifp;
1074 	const struct re_hwrev *hw_rev;
1075 	uint8_t eaddr[ETHER_ADDR_LEN];
1076 	uint16_t as[ETHER_ADDR_LEN / 2];
1077 	uint16_t re_did = 0;
1078 	uint32_t hwrev;
1079 	int error = 0, rid, i;
1080 
1081 	callout_init(&sc->re_timer);
1082 #ifdef RE_DIAG
1083 	sc->re_dev = dev;
1084 #endif
1085 
1086 	RE_ENABLE_TX_MODERATION(sc);
1087 
1088 	sysctl_ctx_init(&sc->re_sysctl_ctx);
1089 	sc->re_sysctl_tree = SYSCTL_ADD_NODE(&sc->re_sysctl_ctx,
1090 					     SYSCTL_STATIC_CHILDREN(_hw),
1091 					     OID_AUTO,
1092 					     device_get_nameunit(dev),
1093 					     CTLFLAG_RD, 0, "");
1094 	if (sc->re_sysctl_tree == NULL) {
1095 		device_printf(dev, "can't add sysctl node\n");
1096 		error = ENXIO;
1097 		goto fail;
1098 	}
1099 	SYSCTL_ADD_PROC(&sc->re_sysctl_ctx,
1100 			SYSCTL_CHILDREN(sc->re_sysctl_tree),
1101 			OID_AUTO, "tx_moderation",
1102 			CTLTYPE_INT | CTLFLAG_RW,
1103 			sc, 0, re_sysctl_tx_moderation, "I",
1104 			"Enable/Disable TX moderation");
1105 
1106 #ifndef BURN_BRIDGES
1107 	/*
1108 	 * Handle power management nonsense.
1109 	 */
1110 
1111 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D0) {
1112 		uint32_t membase, irq;
1113 
1114 		/* Save important PCI config data. */
1115 		membase = pci_read_config(dev, RE_PCI_LOMEM, 4);
1116 		irq = pci_read_config(dev, PCIR_INTLINE, 4);
1117 
1118 		/* Reset the power state. */
1119 		device_printf(dev, "chip is in D%d power mode "
1120 		    "-- setting to D0\n", pci_get_powerstate(dev));
1121 
1122 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
1123 
1124 		/* Restore PCI config data. */
1125 		pci_write_config(dev, RE_PCI_LOMEM, membase, 4);
1126 		pci_write_config(dev, PCIR_INTLINE, irq, 4);
1127 	}
1128 #endif
1129 	/*
1130 	 * Map control/status registers.
1131 	 */
1132 	pci_enable_busmaster(dev);
1133 
1134 	rid = RE_PCI_LOIO;
1135 	sc->re_res = bus_alloc_resource_any(dev, SYS_RES_IOPORT, &rid,
1136 					    RF_ACTIVE);
1137 
1138 	if (sc->re_res == NULL) {
1139 		device_printf(dev, "couldn't map ports\n");
1140 		error = ENXIO;
1141 		goto fail;
1142 	}
1143 
1144 	sc->re_btag = rman_get_bustag(sc->re_res);
1145 	sc->re_bhandle = rman_get_bushandle(sc->re_res);
1146 
1147 	/* Allocate interrupt */
1148 	rid = 0;
1149 	sc->re_irq = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
1150 					    RF_SHAREABLE | RF_ACTIVE);
1151 
1152 	if (sc->re_irq == NULL) {
1153 		device_printf(dev, "couldn't map interrupt\n");
1154 		error = ENXIO;
1155 		goto fail;
1156 	}
1157 
1158 	/* Reset the adapter. */
1159 	re_reset(sc);
1160 
1161 	hwrev = CSR_READ_4(sc, RE_TXCFG) & RE_TXCFG_HWREV;
1162 	for (hw_rev = re_hwrevs; hw_rev->re_desc != NULL; hw_rev++) {
1163 		if (hw_rev->re_rev == hwrev) {
1164 			sc->re_type = hw_rev->re_type;
1165 			sc->re_flags = hw_rev->re_flags;
1166 			break;
1167 		}
1168 	}
1169 
1170 	sc->re_eewidth = 6;
1171 	re_read_eeprom(sc, (caddr_t)&re_did, 0, 1);
1172 	if (re_did != 0x8129)
1173 	        sc->re_eewidth = 8;
1174 
1175 	/*
1176 	 * Get station address from the EEPROM.
1177 	 */
1178 	re_read_eeprom(sc, (caddr_t)as, RE_EE_EADDR, 3);
1179 	for (i = 0; i < ETHER_ADDR_LEN / 2; i++)
1180 		as[i] = le16toh(as[i]);
1181 	bcopy(as, eaddr, sizeof(eaddr));
1182 
1183 	if (sc->re_type == RE_8169) {
1184 		/* Set RX length mask */
1185 		sc->re_rxlenmask = RE_RDESC_STAT_GFRAGLEN;
1186 		sc->re_txstart = RE_GTXSTART;
1187 	} else {
1188 		/* Set RX length mask */
1189 		sc->re_rxlenmask = RE_RDESC_STAT_FRAGLEN;
1190 		sc->re_txstart = RE_TXSTART;
1191 	}
1192 
1193 	/*
1194 	 * Allocate the parent bus DMA tag appropriate for PCI.
1195 	 */
1196 #define RE_NSEG_NEW 32
1197 	error = bus_dma_tag_create(NULL,	/* parent */
1198 			1, 0,			/* alignment, boundary */
1199 			BUS_SPACE_MAXADDR_32BIT,/* lowaddr */
1200 			BUS_SPACE_MAXADDR,	/* highaddr */
1201 			NULL, NULL,		/* filter, filterarg */
1202 			MAXBSIZE, RE_NSEG_NEW,	/* maxsize, nsegments */
1203 			BUS_SPACE_MAXSIZE_32BIT,/* maxsegsize */
1204 			BUS_DMA_ALLOCNOW,	/* flags */
1205 			&sc->re_parent_tag);
1206 	if (error)
1207 		goto fail;
1208 
1209 	error = re_allocmem(dev, sc);
1210 
1211 	if (error)
1212 		goto fail;
1213 
1214 	/* Do MII setup */
1215 	if (mii_phy_probe(dev, &sc->re_miibus,
1216 	    re_ifmedia_upd, re_ifmedia_sts)) {
1217 		device_printf(dev, "MII without any phy!\n");
1218 		error = ENXIO;
1219 		goto fail;
1220 	}
1221 
1222 	ifp = &sc->arpcom.ac_if;
1223 	ifp->if_softc = sc;
1224 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1225 	ifp->if_mtu = ETHERMTU;
1226 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1227 	ifp->if_ioctl = re_ioctl;
1228 	ifp->if_start = re_start;
1229 	ifp->if_capabilities = IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING;
1230 	if (hwrev != RE_HWREV_8168C)	/* XXX does not work yet */
1231 		ifp->if_capabilities |= IFCAP_HWCSUM;
1232 #ifdef DEVICE_POLLING
1233 	ifp->if_poll = re_poll;
1234 #endif
1235 	ifp->if_watchdog = re_watchdog;
1236 	ifp->if_init = re_init;
1237 	if (sc->re_type == RE_8169)
1238 		ifp->if_baudrate = 1000000000;
1239 	else
1240 		ifp->if_baudrate = 100000000;
1241 	ifq_set_maxlen(&ifp->if_snd, RE_IFQ_MAXLEN);
1242 	ifq_set_ready(&ifp->if_snd);
1243 
1244 #ifdef RE_DISABLE_HWCSUM
1245 	ifp->if_capenable = ifp->if_capabilities & ~IFCAP_HWCSUM;
1246 	ifp->if_hwassist = 0;
1247 #else
1248 	ifp->if_capenable = ifp->if_capabilities;
1249 	if (ifp->if_capabilities & IFCAP_HWCSUM)
1250 		ifp->if_hwassist = RE_CSUM_FEATURES;
1251 	else
1252 		ifp->if_hwassist = 0;
1253 #endif	/* RE_DISABLE_HWCSUM */
1254 
1255 	/*
1256 	 * Call MI attach routine.
1257 	 */
1258 	ether_ifattach(ifp, eaddr, NULL);
1259 
1260 #ifdef RE_DIAG
1261 	/*
1262 	 * Perform hardware diagnostic on the original RTL8169.
1263 	 * Some 32-bit cards were incorrectly wired and would
1264 	 * malfunction if plugged into a 64-bit slot.
1265 	 */
1266 	if (hwrev == RE_HWREV_8169) {
1267 		lwkt_serialize_enter(ifp->if_serializer);
1268 		error = re_diag(sc);
1269 		lwkt_serialize_exit(ifp->if_serializer);
1270 
1271 		if (error) {
1272 			device_printf(dev, "hardware diagnostic failure\n");
1273 			ether_ifdetach(ifp);
1274 			goto fail;
1275 		}
1276 	}
1277 #endif	/* RE_DIAG */
1278 
1279 	/* Hook interrupt last to avoid having to lock softc */
1280 	error = bus_setup_intr(dev, sc->re_irq, INTR_MPSAFE, re_intr, sc,
1281 			       &sc->re_intrhand, ifp->if_serializer);
1282 
1283 	if (error) {
1284 		device_printf(dev, "couldn't set up irq\n");
1285 		ether_ifdetach(ifp);
1286 		goto fail;
1287 	}
1288 
1289 	ifp->if_cpuid = ithread_cpuid(rman_get_start(sc->re_irq));
1290 	KKASSERT(ifp->if_cpuid >= 0 && ifp->if_cpuid < ncpus);
1291 
1292 fail:
1293 	if (error)
1294 		re_detach(dev);
1295 
1296 	return (error);
1297 }
1298 
1299 /*
1300  * Shutdown hardware and free up resources. This can be called any
1301  * time after the mutex has been initialized. It is called in both
1302  * the error case in attach and the normal detach case so it needs
1303  * to be careful about only freeing resources that have actually been
1304  * allocated.
1305  */
1306 static int
1307 re_detach(device_t dev)
1308 {
1309 	struct re_softc *sc = device_get_softc(dev);
1310 	struct ifnet *ifp = &sc->arpcom.ac_if;
1311 	int i;
1312 
1313 	/* These should only be active if attach succeeded */
1314 	if (device_is_attached(dev)) {
1315 		lwkt_serialize_enter(ifp->if_serializer);
1316 		re_stop(sc);
1317 		bus_teardown_intr(dev, sc->re_irq, sc->re_intrhand);
1318 		lwkt_serialize_exit(ifp->if_serializer);
1319 
1320 		ether_ifdetach(ifp);
1321 	}
1322 	if (sc->re_miibus)
1323 		device_delete_child(dev, sc->re_miibus);
1324 	bus_generic_detach(dev);
1325 
1326 	if (sc->re_irq)
1327 		bus_release_resource(dev, SYS_RES_IRQ, 0, sc->re_irq);
1328 	if (sc->re_res) {
1329 		bus_release_resource(dev, SYS_RES_IOPORT, RE_PCI_LOIO,
1330 				     sc->re_res);
1331 	}
1332 
1333 	/* Unload and free the RX DMA ring memory and map */
1334 
1335 	if (sc->re_ldata.re_rx_list_tag) {
1336 		bus_dmamap_unload(sc->re_ldata.re_rx_list_tag,
1337 		    sc->re_ldata.re_rx_list_map);
1338 		bus_dmamem_free(sc->re_ldata.re_rx_list_tag,
1339 		    sc->re_ldata.re_rx_list,
1340 		    sc->re_ldata.re_rx_list_map);
1341 		bus_dma_tag_destroy(sc->re_ldata.re_rx_list_tag);
1342 	}
1343 
1344 	/* Unload and free the TX DMA ring memory and map */
1345 
1346 	if (sc->re_ldata.re_tx_list_tag) {
1347 		bus_dmamap_unload(sc->re_ldata.re_tx_list_tag,
1348 		    sc->re_ldata.re_tx_list_map);
1349 		bus_dmamem_free(sc->re_ldata.re_tx_list_tag,
1350 		    sc->re_ldata.re_tx_list,
1351 		    sc->re_ldata.re_tx_list_map);
1352 		bus_dma_tag_destroy(sc->re_ldata.re_tx_list_tag);
1353 	}
1354 
1355 	/* Destroy all the RX and TX buffer maps */
1356 
1357 	if (sc->re_ldata.re_mtag) {
1358 		for (i = 0; i < RE_TX_DESC_CNT; i++)
1359 			bus_dmamap_destroy(sc->re_ldata.re_mtag,
1360 			    sc->re_ldata.re_tx_dmamap[i]);
1361 		for (i = 0; i < RE_RX_DESC_CNT; i++)
1362 			bus_dmamap_destroy(sc->re_ldata.re_mtag,
1363 			    sc->re_ldata.re_rx_dmamap[i]);
1364 		bus_dma_tag_destroy(sc->re_ldata.re_mtag);
1365 	}
1366 
1367 	/* Unload and free the stats buffer and map */
1368 
1369 	if (sc->re_ldata.re_stag) {
1370 		bus_dmamap_unload(sc->re_ldata.re_stag,
1371 		    sc->re_ldata.re_rx_list_map);
1372 		bus_dmamem_free(sc->re_ldata.re_stag,
1373 		    sc->re_ldata.re_stats,
1374 		    sc->re_ldata.re_smap);
1375 		bus_dma_tag_destroy(sc->re_ldata.re_stag);
1376 	}
1377 
1378 	if (sc->re_parent_tag)
1379 		bus_dma_tag_destroy(sc->re_parent_tag);
1380 
1381 	return(0);
1382 }
1383 
1384 static int
1385 re_newbuf(struct re_softc *sc, int idx, struct mbuf *m)
1386 {
1387 	struct re_dmaload_arg arg;
1388 	struct mbuf *n = NULL;
1389 	int error;
1390 
1391 	if (m == NULL) {
1392 		n = m_getcl(MB_DONTWAIT, MT_DATA, M_PKTHDR);
1393 		if (n == NULL)
1394 			return(ENOBUFS);
1395 		m = n;
1396 	} else
1397 		m->m_data = m->m_ext.ext_buf;
1398 
1399 	m->m_len = m->m_pkthdr.len = MCLBYTES;
1400 
1401 	/*
1402 	 * NOTE:
1403 	 * Some re(4) chips(e.g. RTL8101E) need address of the receive buffer
1404 	 * to be 8-byte aligned, so don't call m_adj(m, ETHER_ALIGN) here.
1405 	 */
1406 
1407 	arg.sc = sc;
1408 	arg.re_idx = idx;
1409 	arg.re_maxsegs = 1;
1410 	arg.re_flags = 0;
1411 	arg.re_ring = sc->re_ldata.re_rx_list;
1412 
1413         error = bus_dmamap_load_mbuf(sc->re_ldata.re_mtag,
1414 	    sc->re_ldata.re_rx_dmamap[idx], m, re_dma_map_desc,
1415 	    &arg, BUS_DMA_NOWAIT);
1416 	if (error || arg.re_maxsegs != 1) {
1417 		if (n != NULL)
1418 			m_freem(n);
1419 		return (ENOMEM);
1420 	}
1421 
1422 	sc->re_ldata.re_rx_list[idx].re_cmdstat |= htole32(RE_RDESC_CMD_OWN);
1423 	sc->re_ldata.re_rx_mbuf[idx] = m;
1424 
1425         bus_dmamap_sync(sc->re_ldata.re_mtag, sc->re_ldata.re_rx_dmamap[idx],
1426 		        BUS_DMASYNC_PREREAD);
1427 
1428 	return(0);
1429 }
1430 
1431 static int
1432 re_tx_list_init(struct re_softc *sc)
1433 {
1434 	bzero(sc->re_ldata.re_tx_list, RE_TX_LIST_SZ);
1435 	bzero(&sc->re_ldata.re_tx_mbuf, RE_TX_DESC_CNT * sizeof(struct mbuf *));
1436 
1437 	bus_dmamap_sync(sc->re_ldata.re_tx_list_tag,
1438 			sc->re_ldata.re_tx_list_map, BUS_DMASYNC_PREWRITE);
1439 	sc->re_ldata.re_tx_prodidx = 0;
1440 	sc->re_ldata.re_tx_considx = 0;
1441 	sc->re_ldata.re_tx_free = RE_TX_DESC_CNT;
1442 
1443 	return(0);
1444 }
1445 
1446 static int
1447 re_rx_list_init(struct re_softc *sc)
1448 {
1449 	int i, error;
1450 
1451 	bzero(sc->re_ldata.re_rx_list, RE_RX_LIST_SZ);
1452 	bzero(&sc->re_ldata.re_rx_mbuf, RE_RX_DESC_CNT * sizeof(struct mbuf *));
1453 
1454 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
1455 		error = re_newbuf(sc, i, NULL);
1456 		if (error)
1457 			return(error);
1458 	}
1459 
1460 	/* Flush the RX descriptors */
1461 
1462 	bus_dmamap_sync(sc->re_ldata.re_rx_list_tag,
1463 			sc->re_ldata.re_rx_list_map, BUS_DMASYNC_PREWRITE);
1464 
1465 	sc->re_ldata.re_rx_prodidx = 0;
1466 	sc->re_head = sc->re_tail = NULL;
1467 
1468 	return(0);
1469 }
1470 
1471 /*
1472  * RX handler for C+ and 8169. For the gigE chips, we support
1473  * the reception of jumbo frames that have been fragmented
1474  * across multiple 2K mbuf cluster buffers.
1475  */
1476 static void
1477 re_rxeof(struct re_softc *sc)
1478 {
1479 	struct ifnet *ifp = &sc->arpcom.ac_if;
1480 	struct mbuf *m;
1481 	struct re_desc 	*cur_rx;
1482 	uint32_t rxstat, rxvlan;
1483 	int i, total_len;
1484 #ifdef ETHER_INPUT_CHAIN
1485 	struct mbuf_chain chain[MAXCPU];
1486 #endif
1487 
1488 	/* Invalidate the descriptor memory */
1489 
1490 	bus_dmamap_sync(sc->re_ldata.re_rx_list_tag,
1491 			sc->re_ldata.re_rx_list_map, BUS_DMASYNC_POSTREAD);
1492 
1493 #ifdef ETHER_INPUT_CHAIN
1494 	ether_input_chain_init(chain);
1495 #endif
1496 
1497 	for (i = sc->re_ldata.re_rx_prodidx;
1498 	     RE_OWN(&sc->re_ldata.re_rx_list[i]) == 0 ; RE_DESC_INC(i)) {
1499 		cur_rx = &sc->re_ldata.re_rx_list[i];
1500 		m = sc->re_ldata.re_rx_mbuf[i];
1501 		total_len = RE_RXBYTES(cur_rx);
1502 		rxstat = le32toh(cur_rx->re_cmdstat);
1503 		rxvlan = le32toh(cur_rx->re_vlanctl);
1504 
1505 		/* Invalidate the RX mbuf and unload its map */
1506 
1507 		bus_dmamap_sync(sc->re_ldata.re_mtag,
1508 				sc->re_ldata.re_rx_dmamap[i],
1509 				BUS_DMASYNC_POSTWRITE);
1510 		bus_dmamap_unload(sc->re_ldata.re_mtag,
1511 				  sc->re_ldata.re_rx_dmamap[i]);
1512 
1513 		if ((rxstat & RE_RDESC_STAT_EOF) == 0) {
1514 			m->m_len = MCLBYTES - ETHER_ALIGN;
1515 			if (sc->re_head == NULL) {
1516 				sc->re_head = sc->re_tail = m;
1517 			} else {
1518 				sc->re_tail->m_next = m;
1519 				sc->re_tail = m;
1520 			}
1521 			re_newbuf(sc, i, NULL);
1522 			continue;
1523 		}
1524 
1525 		/*
1526 		 * NOTE: for the 8139C+, the frame length field
1527 		 * is always 12 bits in size, but for the gigE chips,
1528 		 * it is 13 bits (since the max RX frame length is 16K).
1529 		 * Unfortunately, all 32 bits in the status word
1530 		 * were already used, so to make room for the extra
1531 		 * length bit, RealTek took out the 'frame alignment
1532 		 * error' bit and shifted the other status bits
1533 		 * over one slot. The OWN, EOR, FS and LS bits are
1534 		 * still in the same places. We have already extracted
1535 		 * the frame length and checked the OWN bit, so rather
1536 		 * than using an alternate bit mapping, we shift the
1537 		 * status bits one space to the right so we can evaluate
1538 		 * them using the 8169 status as though it was in the
1539 		 * same format as that of the 8139C+.
1540 		 */
1541 		if (sc->re_type == RE_8169)
1542 			rxstat >>= 1;
1543 
1544 		if (rxstat & RE_RDESC_STAT_RXERRSUM) {
1545 			ifp->if_ierrors++;
1546 			/*
1547 			 * If this is part of a multi-fragment packet,
1548 			 * discard all the pieces.
1549 			 */
1550 			if (sc->re_head != NULL) {
1551 				m_freem(sc->re_head);
1552 				sc->re_head = sc->re_tail = NULL;
1553 			}
1554 			re_newbuf(sc, i, m);
1555 			continue;
1556 		}
1557 
1558 		/*
1559 		 * If allocating a replacement mbuf fails,
1560 		 * reload the current one.
1561 		 */
1562 
1563 		if (re_newbuf(sc, i, NULL)) {
1564 			ifp->if_ierrors++;
1565 			if (sc->re_head != NULL) {
1566 				m_freem(sc->re_head);
1567 				sc->re_head = sc->re_tail = NULL;
1568 			}
1569 			re_newbuf(sc, i, m);
1570 			continue;
1571 		}
1572 
1573 		if (sc->re_head != NULL) {
1574 			m->m_len = total_len % (MCLBYTES - ETHER_ALIGN);
1575 			/*
1576 			 * Special case: if there's 4 bytes or less
1577 			 * in this buffer, the mbuf can be discarded:
1578 			 * the last 4 bytes is the CRC, which we don't
1579 			 * care about anyway.
1580 			 */
1581 			if (m->m_len <= ETHER_CRC_LEN) {
1582 				sc->re_tail->m_len -=
1583 				    (ETHER_CRC_LEN - m->m_len);
1584 				m_freem(m);
1585 			} else {
1586 				m->m_len -= ETHER_CRC_LEN;
1587 				sc->re_tail->m_next = m;
1588 			}
1589 			m = sc->re_head;
1590 			sc->re_head = sc->re_tail = NULL;
1591 			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
1592 		} else
1593 			m->m_pkthdr.len = m->m_len =
1594 			    (total_len - ETHER_CRC_LEN);
1595 
1596 		ifp->if_ipackets++;
1597 		m->m_pkthdr.rcvif = ifp;
1598 
1599 		/* Do RX checksumming if enabled */
1600 
1601 		if (ifp->if_capenable & IFCAP_RXCSUM) {
1602 
1603 			/* Check IP header checksum */
1604 			if (rxstat & RE_RDESC_STAT_PROTOID)
1605 				m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED;
1606 			if ((rxstat & RE_RDESC_STAT_IPSUMBAD) == 0)
1607 				m->m_pkthdr.csum_flags |= CSUM_IP_VALID;
1608 
1609 			/* Check TCP/UDP checksum */
1610 			if ((RE_TCPPKT(rxstat) &&
1611 			    (rxstat & RE_RDESC_STAT_TCPSUMBAD) == 0) ||
1612 			    (RE_UDPPKT(rxstat) &&
1613 			    (rxstat & RE_RDESC_STAT_UDPSUMBAD)) == 0) {
1614 				m->m_pkthdr.csum_flags |=
1615 				    CSUM_DATA_VALID|CSUM_PSEUDO_HDR|
1616 				    CSUM_FRAG_NOT_CHECKED;
1617 				m->m_pkthdr.csum_data = 0xffff;
1618 			}
1619 		}
1620 
1621 		if (rxvlan & RE_RDESC_VLANCTL_TAG) {
1622 			m->m_flags |= M_VLANTAG;
1623 			m->m_pkthdr.ether_vlantag =
1624 				be16toh((rxvlan & RE_RDESC_VLANCTL_DATA));
1625 		}
1626 #ifdef ETHER_INPUT_CHAIN
1627 		ether_input_chain2(ifp, m, chain);
1628 #else
1629 		ifp->if_input(ifp, m);
1630 #endif
1631 	}
1632 
1633 #ifdef ETHER_INPUT_CHAIN
1634 	ether_input_dispatch(chain);
1635 #endif
1636 
1637 	/* Flush the RX DMA ring */
1638 
1639 	bus_dmamap_sync(sc->re_ldata.re_rx_list_tag,
1640 			sc->re_ldata.re_rx_list_map, BUS_DMASYNC_PREWRITE);
1641 
1642 	sc->re_ldata.re_rx_prodidx = i;
1643 }
1644 
1645 static void
1646 re_txeof(struct re_softc *sc)
1647 {
1648 	struct ifnet *ifp = &sc->arpcom.ac_if;
1649 	uint32_t txstat;
1650 	int idx;
1651 
1652 	/* Invalidate the TX descriptor list */
1653 
1654 	bus_dmamap_sync(sc->re_ldata.re_tx_list_tag,
1655 			sc->re_ldata.re_tx_list_map, BUS_DMASYNC_POSTREAD);
1656 
1657 	for (idx = sc->re_ldata.re_tx_considx;
1658 	     sc->re_ldata.re_tx_free < RE_TX_DESC_CNT; RE_DESC_INC(idx)) {
1659 		txstat = le32toh(sc->re_ldata.re_tx_list[idx].re_cmdstat);
1660 		if (txstat & RE_TDESC_CMD_OWN)
1661 			break;
1662 
1663 		sc->re_ldata.re_tx_list[idx].re_bufaddr_lo = 0;
1664 
1665 		/*
1666 		 * We only stash mbufs in the last descriptor
1667 		 * in a fragment chain, which also happens to
1668 		 * be the only place where the TX status bits
1669 		 * are valid.
1670 		 */
1671 		if (txstat & RE_TDESC_CMD_EOF) {
1672 			m_freem(sc->re_ldata.re_tx_mbuf[idx]);
1673 			sc->re_ldata.re_tx_mbuf[idx] = NULL;
1674 			bus_dmamap_unload(sc->re_ldata.re_mtag,
1675 			    sc->re_ldata.re_tx_dmamap[idx]);
1676 			if (txstat & (RE_TDESC_STAT_EXCESSCOL|
1677 			    RE_TDESC_STAT_COLCNT))
1678 				ifp->if_collisions++;
1679 			if (txstat & RE_TDESC_STAT_TXERRSUM)
1680 				ifp->if_oerrors++;
1681 			else
1682 				ifp->if_opackets++;
1683 		}
1684 		sc->re_ldata.re_tx_free++;
1685 	}
1686 
1687 	/* No changes made to the TX ring, so no flush needed */
1688 	if (sc->re_ldata.re_tx_free) {
1689 		sc->re_ldata.re_tx_considx = idx;
1690 		ifp->if_flags &= ~IFF_OACTIVE;
1691 		ifp->if_timer = 0;
1692 	}
1693 
1694 	/*
1695 	 * Some chips will ignore a second TX request issued while an
1696 	 * existing transmission is in progress. If the transmitter goes
1697 	 * idle but there are still packets waiting to be sent, we need
1698 	 * to restart the channel here to flush them out. This only seems
1699 	 * to be required with the PCIe devices.
1700 	 */
1701 	if (sc->re_ldata.re_tx_free < RE_TX_DESC_CNT)
1702 		CSR_WRITE_1(sc, sc->re_txstart, RE_TXSTART_START);
1703 
1704 	/*
1705 	 * If not all descriptors have been released reaped yet,
1706 	 * reload the timer so that we will eventually get another
1707 	 * interrupt that will cause us to re-enter this routine.
1708 	 * This is done in case the transmitter has gone idle.
1709 	 */
1710 	if (RE_TX_MODERATION_IS_ENABLED(sc) &&
1711 	    sc->re_ldata.re_tx_free < RE_TX_DESC_CNT)
1712                 CSR_WRITE_4(sc, RE_TIMERCNT, 1);
1713 }
1714 
1715 static void
1716 re_tick(void *xsc)
1717 {
1718 	struct re_softc *sc = xsc;
1719 
1720 	lwkt_serialize_enter(sc->arpcom.ac_if.if_serializer);
1721 	re_tick_serialized(xsc);
1722 	lwkt_serialize_exit(sc->arpcom.ac_if.if_serializer);
1723 }
1724 
1725 static void
1726 re_tick_serialized(void *xsc)
1727 {
1728 	struct re_softc *sc = xsc;
1729 	struct ifnet *ifp = &sc->arpcom.ac_if;
1730 	struct mii_data *mii;
1731 
1732 	ASSERT_SERIALIZED(ifp->if_serializer);
1733 
1734 	mii = device_get_softc(sc->re_miibus);
1735 	mii_tick(mii);
1736 	if (sc->re_link) {
1737 		if (!(mii->mii_media_status & IFM_ACTIVE))
1738 			sc->re_link = 0;
1739 	} else {
1740 		if (mii->mii_media_status & IFM_ACTIVE &&
1741 		    IFM_SUBTYPE(mii->mii_media_active) != IFM_NONE) {
1742 			sc->re_link = 1;
1743 			if (!ifq_is_empty(&ifp->if_snd))
1744 				if_devstart(ifp);
1745 		}
1746 	}
1747 
1748 	callout_reset(&sc->re_timer, hz, re_tick, sc);
1749 }
1750 
1751 #ifdef DEVICE_POLLING
1752 
1753 static void
1754 re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
1755 {
1756 	struct re_softc *sc = ifp->if_softc;
1757 
1758 	ASSERT_SERIALIZED(ifp->if_serializer);
1759 
1760 	switch(cmd) {
1761 	case POLL_REGISTER:
1762 		/* disable interrupts */
1763 		CSR_WRITE_2(sc, RE_IMR, 0x0000);
1764 		break;
1765 	case POLL_DEREGISTER:
1766 		/* enable interrupts */
1767 		CSR_WRITE_2(sc, RE_IMR, sc->re_intrs);
1768 		break;
1769 	default:
1770 		sc->rxcycles = count;
1771 		re_rxeof(sc);
1772 		re_txeof(sc);
1773 
1774 		if (!ifq_is_empty(&ifp->if_snd))
1775 			if_devstart(ifp);
1776 
1777 		if (cmd == POLL_AND_CHECK_STATUS) { /* also check status register */
1778 			uint16_t       status;
1779 
1780 			status = CSR_READ_2(sc, RE_ISR);
1781 			if (status == 0xffff)
1782 				return;
1783 			if (status)
1784 				CSR_WRITE_2(sc, RE_ISR, status);
1785 
1786 			/*
1787 			 * XXX check behaviour on receiver stalls.
1788 			 */
1789 
1790 			if (status & RE_ISR_SYSTEM_ERR) {
1791 				re_reset(sc);
1792 				re_init(sc);
1793 			}
1794 		}
1795 		break;
1796 	}
1797 }
1798 #endif /* DEVICE_POLLING */
1799 
1800 static void
1801 re_intr(void *arg)
1802 {
1803 	struct re_softc	*sc = arg;
1804 	struct ifnet *ifp = &sc->arpcom.ac_if;
1805 	uint16_t status;
1806 
1807 	ASSERT_SERIALIZED(ifp->if_serializer);
1808 
1809 	if (sc->suspended || (ifp->if_flags & IFF_UP) == 0)
1810 		return;
1811 
1812 	for (;;) {
1813 		status = CSR_READ_2(sc, RE_ISR);
1814 		/* If the card has gone away the read returns 0xffff. */
1815 		if (status == 0xffff)
1816 			break;
1817 		if (status)
1818 			CSR_WRITE_2(sc, RE_ISR, status);
1819 
1820 		if ((status & sc->re_intrs) == 0)
1821 			break;
1822 
1823 		if (status & (RE_ISR_RX_OK | RE_ISR_RX_ERR | RE_ISR_FIFO_OFLOW))
1824 			re_rxeof(sc);
1825 
1826 		if ((status & sc->re_tx_ack) ||
1827 		    (status & RE_ISR_TX_ERR) ||
1828 		    (status & RE_ISR_TX_DESC_UNAVAIL))
1829 			re_txeof(sc);
1830 
1831 		if (status & RE_ISR_SYSTEM_ERR) {
1832 			re_reset(sc);
1833 			re_init(sc);
1834 		}
1835 
1836 		if (status & RE_ISR_LINKCHG) {
1837 			callout_stop(&sc->re_timer);
1838 			re_tick_serialized(sc);
1839 		}
1840 	}
1841 
1842 	if (!ifq_is_empty(&ifp->if_snd))
1843 		if_devstart(ifp);
1844 }
1845 
1846 static int
1847 re_encap(struct re_softc *sc, struct mbuf **m_head, int *idx, int *called_defrag)
1848 {
1849 	struct ifnet *ifp = &sc->arpcom.ac_if;
1850 	struct mbuf *m, *m_new = NULL;
1851 	struct re_dmaload_arg	arg;
1852 	bus_dmamap_t		map;
1853 	int			error;
1854 
1855 	KASSERT(sc->re_ldata.re_tx_free > 4, ("not enough free TX desc\n"));
1856 
1857 	*called_defrag = 0;
1858 	m = *m_head;
1859 
1860 	/*
1861 	 * Set up checksum offload. Note: checksum offload bits must
1862 	 * appear in all descriptors of a multi-descriptor transmit
1863 	 * attempt. (This is according to testing done with an 8169
1864 	 * chip. I'm not sure if this is a requirement or a bug.)
1865 	 */
1866 
1867 	arg.re_flags = 0;
1868 
1869 	if (m->m_pkthdr.csum_flags & CSUM_IP)
1870 		arg.re_flags |= RE_TDESC_CMD_IPCSUM;
1871 	if (m->m_pkthdr.csum_flags & CSUM_TCP)
1872 		arg.re_flags |= RE_TDESC_CMD_TCPCSUM;
1873 	if (m->m_pkthdr.csum_flags & CSUM_UDP)
1874 		arg.re_flags |= RE_TDESC_CMD_UDPCSUM;
1875 
1876 	arg.sc = sc;
1877 	arg.re_idx = *idx;
1878 	arg.re_maxsegs = sc->re_ldata.re_tx_free;
1879 	if (arg.re_maxsegs > 4)
1880 		arg.re_maxsegs -= 4;
1881 	arg.re_ring = sc->re_ldata.re_tx_list;
1882 
1883 	map = sc->re_ldata.re_tx_dmamap[*idx];
1884 
1885 	/*
1886 	 * With some of the RealTek chips, using the checksum offload
1887 	 * support in conjunction with the autopadding feature results
1888 	 * in the transmission of corrupt frames. For example, if we
1889 	 * need to send a really small IP fragment that's less than 60
1890 	 * bytes in size, and IP header checksumming is enabled, the
1891 	 * resulting ethernet frame that appears on the wire will
1892 	 * have garbled payload. To work around this, if TX checksum
1893 	 * offload is enabled, we always manually pad short frames out
1894 	 * to the minimum ethernet frame size. We do this by pretending
1895 	 * the mbuf chain has too many fragments so the coalescing code
1896 	 * below can assemble the packet into a single buffer that's
1897 	 * padded out to the mininum frame size.
1898 	 *
1899 	 * Note: this appears unnecessary for TCP, and doing it for TCP
1900 	 * with PCIe adapters seems to result in bad checksums.
1901 	 */
1902 	if (arg.re_flags && !(arg.re_flags & RE_TDESC_CMD_TCPCSUM) &&
1903 	    m->m_pkthdr.len < RE_MIN_FRAMELEN) {
1904 		error = EFBIG;
1905 	} else {
1906 		error = bus_dmamap_load_mbuf(sc->re_ldata.re_mtag, map,
1907 		    m, re_dma_map_desc, &arg, BUS_DMA_NOWAIT);
1908 	}
1909 
1910 	if (error && error != EFBIG) {
1911 		if_printf(ifp, "can't map mbuf (error %d)\n", error);
1912 		return(ENOBUFS);
1913 	}
1914 
1915 	/* Too many segments to map, coalesce into a single mbuf */
1916 
1917 	if (error || arg.re_maxsegs == 0) {
1918 		m_new = m_defrag_nofree(m, MB_DONTWAIT);
1919 		if (m_new == NULL) {
1920 			return(1);
1921 		} else {
1922 			m = m_new;
1923 			*m_head = m;
1924 		}
1925 
1926 		/*
1927 		 * Manually pad short frames, and zero the pad space
1928 		 * to avoid leaking data.
1929 		 */
1930 		if (m_new->m_pkthdr.len < RE_MIN_FRAMELEN) {
1931 			bzero(mtod(m_new, char *) + m_new->m_pkthdr.len,
1932 			    RE_MIN_FRAMELEN - m_new->m_pkthdr.len);
1933 			m_new->m_pkthdr.len += RE_MIN_FRAMELEN -
1934 			    m_new->m_pkthdr.len;
1935 			m_new->m_len = m_new->m_pkthdr.len;
1936 		}
1937 
1938 		*called_defrag = 1;
1939 		arg.sc = sc;
1940 		arg.re_idx = *idx;
1941 		arg.re_maxsegs = sc->re_ldata.re_tx_free;
1942 		arg.re_ring = sc->re_ldata.re_tx_list;
1943 
1944 		error = bus_dmamap_load_mbuf(sc->re_ldata.re_mtag, map,
1945 		    m, re_dma_map_desc, &arg, BUS_DMA_NOWAIT);
1946 		if (error) {
1947 			m_freem(m);
1948 			if_printf(ifp, "can't map mbuf (error %d)\n", error);
1949 			return(EFBIG);
1950 		}
1951 	}
1952 
1953 	/*
1954 	 * Insure that the map for this transmission
1955 	 * is placed at the array index of the last descriptor
1956 	 * in this chain.
1957 	 */
1958 	sc->re_ldata.re_tx_dmamap[*idx] =
1959 	    sc->re_ldata.re_tx_dmamap[arg.re_idx];
1960 	sc->re_ldata.re_tx_dmamap[arg.re_idx] = map;
1961 
1962 	sc->re_ldata.re_tx_mbuf[arg.re_idx] = m;
1963 	sc->re_ldata.re_tx_free -= arg.re_maxsegs;
1964 
1965 	/*
1966 	 * Set up hardware VLAN tagging. Note: vlan tag info must
1967 	 * appear in the first descriptor of a multi-descriptor
1968 	 * transmission attempt.
1969 	 */
1970 
1971 	if (m->m_flags & M_VLANTAG) {
1972 		sc->re_ldata.re_tx_list[*idx].re_vlanctl =
1973 		    htole32(htobe16(m->m_pkthdr.ether_vlantag) |
1974 		    	    RE_TDESC_VLANCTL_TAG);
1975 	}
1976 
1977 	/* Transfer ownership of packet to the chip. */
1978 
1979 	sc->re_ldata.re_tx_list[arg.re_idx].re_cmdstat |=
1980 	    htole32(RE_TDESC_CMD_OWN);
1981 	if (*idx != arg.re_idx)
1982 		sc->re_ldata.re_tx_list[*idx].re_cmdstat |=
1983 		    htole32(RE_TDESC_CMD_OWN);
1984 
1985 	RE_DESC_INC(arg.re_idx);
1986 	*idx = arg.re_idx;
1987 
1988 	return(0);
1989 }
1990 
1991 /*
1992  * Main transmit routine for C+ and gigE NICs.
1993  */
1994 
1995 static void
1996 re_start(struct ifnet *ifp)
1997 {
1998 	struct re_softc	*sc = ifp->if_softc;
1999 	struct mbuf *m_head;
2000 	struct mbuf *m_head2;
2001 	int called_defrag, idx, need_trans;
2002 
2003 	ASSERT_SERIALIZED(ifp->if_serializer);
2004 
2005 	if (!sc->re_link) {
2006 		ifq_purge(&ifp->if_snd);
2007 		return;
2008 	}
2009 
2010 	if ((ifp->if_flags & (IFF_OACTIVE | IFF_RUNNING)) != IFF_RUNNING)
2011 		return;
2012 
2013 	idx = sc->re_ldata.re_tx_prodidx;
2014 
2015 	need_trans = 0;
2016 	while (sc->re_ldata.re_tx_mbuf[idx] == NULL) {
2017 		if (sc->re_ldata.re_tx_free <= 4) {
2018 			ifp->if_flags |= IFF_OACTIVE;
2019 			break;
2020 		}
2021 
2022 		m_head = ifq_dequeue(&ifp->if_snd, NULL);
2023 		if (m_head == NULL)
2024 			break;
2025 
2026 		m_head2 = m_head;
2027 		if (re_encap(sc, &m_head2, &idx, &called_defrag)) {
2028 			/*
2029 			 * If we could not encapsulate the defragged packet,
2030 			 * the returned m_head2 is garbage and we must dequeue
2031 			 * and throw away the original packet.
2032 			 */
2033 			if (called_defrag)
2034 				m_freem(m_head);
2035 			ifp->if_flags |= IFF_OACTIVE;
2036 			break;
2037 		}
2038 
2039 		/*
2040 		 * Clean out the packet we encapsulated.  If we defragged
2041 		 * the packet the m_head2 is the one that got encapsulated
2042 		 * and the original must be thrown away.  Otherwise m_head2
2043 		 * *IS* the original.
2044 		 */
2045 		if (called_defrag)
2046 			m_freem(m_head);
2047 		need_trans = 1;
2048 
2049 		/*
2050 		 * If there's a BPF listener, bounce a copy of this frame
2051 		 * to him.
2052 		 */
2053 		ETHER_BPF_MTAP(ifp, m_head2);
2054 	}
2055 
2056 	if (!need_trans) {
2057 		if (RE_TX_MODERATION_IS_ENABLED(sc) &&
2058 		    sc->re_ldata.re_tx_free != RE_TX_DESC_CNT)
2059 			CSR_WRITE_4(sc, RE_TIMERCNT, 1);
2060 		return;
2061 	}
2062 
2063 	/* Flush the TX descriptors */
2064 	bus_dmamap_sync(sc->re_ldata.re_tx_list_tag,
2065 			sc->re_ldata.re_tx_list_map, BUS_DMASYNC_PREWRITE);
2066 
2067 	sc->re_ldata.re_tx_prodidx = idx;
2068 
2069 	/*
2070 	 * RealTek put the TX poll request register in a different
2071 	 * location on the 8169 gigE chip. I don't know why.
2072 	 */
2073 	CSR_WRITE_1(sc, sc->re_txstart, RE_TXSTART_START);
2074 
2075 	if (RE_TX_MODERATION_IS_ENABLED(sc)) {
2076 		/*
2077 		 * Use the countdown timer for interrupt moderation.
2078 		 * 'TX done' interrupts are disabled. Instead, we reset the
2079 		 * countdown timer, which will begin counting until it hits
2080 		 * the value in the TIMERINT register, and then trigger an
2081 		 * interrupt. Each time we write to the TIMERCNT register,
2082 		 * the timer count is reset to 0.
2083 		 */
2084 		CSR_WRITE_4(sc, RE_TIMERCNT, 1);
2085 	}
2086 
2087 	/*
2088 	 * Set a timeout in case the chip goes out to lunch.
2089 	 */
2090 	ifp->if_timer = 5;
2091 }
2092 
2093 static void
2094 re_init(void *xsc)
2095 {
2096 	struct re_softc *sc = xsc;
2097 	struct ifnet *ifp = &sc->arpcom.ac_if;
2098 	struct mii_data *mii;
2099 	uint32_t rxcfg = 0;
2100 
2101 	ASSERT_SERIALIZED(ifp->if_serializer);
2102 
2103 	mii = device_get_softc(sc->re_miibus);
2104 
2105 	/*
2106 	 * Cancel pending I/O and free all RX/TX buffers.
2107 	 */
2108 	re_stop(sc);
2109 
2110 	/*
2111 	 * Enable C+ RX and TX mode, as well as VLAN stripping and
2112 	 * RX checksum offload. We must configure the C+ register
2113 	 * before all others.
2114 	 */
2115 	CSR_WRITE_2(sc, RE_CPLUS_CMD, RE_CPLUSCMD_RXENB | RE_CPLUSCMD_TXENB |
2116 		    RE_CPLUSCMD_PCI_MRW | RE_CPLUSCMD_VLANSTRIP |
2117 		    (ifp->if_capenable & IFCAP_RXCSUM ?
2118 		     RE_CPLUSCMD_RXCSUM_ENB : 0));
2119 
2120 	/*
2121 	 * Init our MAC address.  Even though the chipset
2122 	 * documentation doesn't mention it, we need to enter "Config
2123 	 * register write enable" mode to modify the ID registers.
2124 	 */
2125 	CSR_WRITE_1(sc, RE_EECMD, RE_EEMODE_WRITECFG);
2126 	CSR_WRITE_4(sc, RE_IDR0,
2127 	    htole32(*(uint32_t *)(&sc->arpcom.ac_enaddr[0])));
2128 	CSR_WRITE_2(sc, RE_IDR4,
2129 	    htole16(*(uint16_t *)(&sc->arpcom.ac_enaddr[4])));
2130 	CSR_WRITE_1(sc, RE_EECMD, RE_EEMODE_OFF);
2131 
2132 	/*
2133 	 * For C+ mode, initialize the RX descriptors and mbufs.
2134 	 */
2135 	re_rx_list_init(sc);
2136 	re_tx_list_init(sc);
2137 
2138 	/*
2139 	 * Load the addresses of the RX and TX lists into the chip.
2140 	 */
2141 	CSR_WRITE_4(sc, RE_RXLIST_ADDR_HI,
2142 	    RE_ADDR_HI(sc->re_ldata.re_rx_list_addr));
2143 	CSR_WRITE_4(sc, RE_RXLIST_ADDR_LO,
2144 	    RE_ADDR_LO(sc->re_ldata.re_rx_list_addr));
2145 
2146 	CSR_WRITE_4(sc, RE_TXLIST_ADDR_HI,
2147 	    RE_ADDR_HI(sc->re_ldata.re_tx_list_addr));
2148 	CSR_WRITE_4(sc, RE_TXLIST_ADDR_LO,
2149 	    RE_ADDR_LO(sc->re_ldata.re_tx_list_addr));
2150 
2151 	/*
2152 	 * Enable transmit and receive.
2153 	 */
2154 	CSR_WRITE_1(sc, RE_COMMAND, RE_CMD_TX_ENB|RE_CMD_RX_ENB);
2155 
2156 	/*
2157 	 * Set the initial TX and RX configuration.
2158 	 */
2159 	if (sc->re_testmode) {
2160 		if (sc->re_type == RE_8169)
2161 			CSR_WRITE_4(sc, RE_TXCFG,
2162 				    RE_TXCFG_CONFIG | RE_LOOPTEST_ON);
2163 		else
2164 			CSR_WRITE_4(sc, RE_TXCFG,
2165 				    RE_TXCFG_CONFIG | RE_LOOPTEST_ON_CPLUS);
2166 	} else
2167 		CSR_WRITE_4(sc, RE_TXCFG, RE_TXCFG_CONFIG);
2168 
2169 	CSR_WRITE_1(sc, RE_EARLY_TX_THRESH, 16);
2170 
2171 	CSR_WRITE_4(sc, RE_RXCFG, RE_RXCFG_CONFIG);
2172 
2173 	/* Set the individual bit to receive frames for this host only. */
2174 	rxcfg = CSR_READ_4(sc, RE_RXCFG);
2175 	rxcfg |= RE_RXCFG_RX_INDIV;
2176 
2177 	/* If we want promiscuous mode, set the allframes bit. */
2178 	if (ifp->if_flags & IFF_PROMISC) {
2179 		rxcfg |= RE_RXCFG_RX_ALLPHYS;
2180 		CSR_WRITE_4(sc, RE_RXCFG, rxcfg);
2181 	} else {
2182 		rxcfg &= ~RE_RXCFG_RX_ALLPHYS;
2183 		CSR_WRITE_4(sc, RE_RXCFG, rxcfg);
2184 	}
2185 
2186 	/*
2187 	 * Set capture broadcast bit to capture broadcast frames.
2188 	 */
2189 	if (ifp->if_flags & IFF_BROADCAST) {
2190 		rxcfg |= RE_RXCFG_RX_BROAD;
2191 		CSR_WRITE_4(sc, RE_RXCFG, rxcfg);
2192 	} else {
2193 		rxcfg &= ~RE_RXCFG_RX_BROAD;
2194 		CSR_WRITE_4(sc, RE_RXCFG, rxcfg);
2195 	}
2196 
2197 	/*
2198 	 * Program the multicast filter, if necessary.
2199 	 */
2200 	re_setmulti(sc);
2201 
2202 #ifdef DEVICE_POLLING
2203 	/*
2204 	 * Disable interrupts if we are polling.
2205 	 */
2206 	if (ifp->if_flags & IFF_POLLING)
2207 		CSR_WRITE_2(sc, RE_IMR, 0);
2208 	else	/* otherwise ... */
2209 #endif /* DEVICE_POLLING */
2210 	/*
2211 	 * Enable interrupts.
2212 	 */
2213 	if (sc->re_testmode)
2214 		CSR_WRITE_2(sc, RE_IMR, 0);
2215 	else
2216 		CSR_WRITE_2(sc, RE_IMR, sc->re_intrs);
2217 	CSR_WRITE_2(sc, RE_ISR, sc->re_intrs);
2218 
2219 	/* Set initial TX threshold */
2220 	sc->re_txthresh = RE_TX_THRESH_INIT;
2221 
2222 	/* Start RX/TX process. */
2223 	if (sc->re_flags & RE_F_HASMPC)
2224 		CSR_WRITE_4(sc, RE_MISSEDPKT, 0);
2225 #ifdef notdef
2226 	/* Enable receiver and transmitter. */
2227 	CSR_WRITE_1(sc, RE_COMMAND, RE_CMD_TX_ENB|RE_CMD_RX_ENB);
2228 #endif
2229 
2230 	if (RE_TX_MODERATION_IS_ENABLED(sc)) {
2231 		/*
2232 		 * Initialize the timer interrupt register so that
2233 		 * a timer interrupt will be generated once the timer
2234 		 * reaches a certain number of ticks. The timer is
2235 		 * reloaded on each transmit. This gives us TX interrupt
2236 		 * moderation, which dramatically improves TX frame rate.
2237 		 */
2238 		if (sc->re_type == RE_8169)
2239 			CSR_WRITE_4(sc, RE_TIMERINT_8169, 0x800);
2240 		else
2241 			CSR_WRITE_4(sc, RE_TIMERINT, 0x400);
2242 	}
2243 
2244 	/*
2245 	 * For 8169 gigE NICs, set the max allowed RX packet
2246 	 * size so we can receive jumbo frames.
2247 	 */
2248 	if (sc->re_type == RE_8169)
2249 		CSR_WRITE_2(sc, RE_MAXRXPKTLEN, 16383);
2250 
2251 	if (sc->re_testmode) {
2252 		return;
2253 	}
2254 
2255 	mii_mediachg(mii);
2256 
2257 	CSR_WRITE_1(sc, RE_CFG1, RE_CFG1_DRVLOAD|RE_CFG1_FULLDUPLEX);
2258 
2259 	ifp->if_flags |= IFF_RUNNING;
2260 	ifp->if_flags &= ~IFF_OACTIVE;
2261 
2262 	sc->re_link = 0;
2263 	callout_reset(&sc->re_timer, hz, re_tick, sc);
2264 }
2265 
2266 /*
2267  * Set media options.
2268  */
2269 static int
2270 re_ifmedia_upd(struct ifnet *ifp)
2271 {
2272 	struct re_softc *sc = ifp->if_softc;
2273 	struct mii_data *mii;
2274 
2275 	ASSERT_SERIALIZED(ifp->if_serializer);
2276 
2277 	mii = device_get_softc(sc->re_miibus);
2278 	mii_mediachg(mii);
2279 
2280 	return(0);
2281 }
2282 
2283 /*
2284  * Report current media status.
2285  */
2286 static void
2287 re_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
2288 {
2289 	struct re_softc *sc = ifp->if_softc;
2290 	struct mii_data *mii;
2291 
2292 	ASSERT_SERIALIZED(ifp->if_serializer);
2293 
2294 	mii = device_get_softc(sc->re_miibus);
2295 
2296 	mii_pollstat(mii);
2297 	ifmr->ifm_active = mii->mii_media_active;
2298 	ifmr->ifm_status = mii->mii_media_status;
2299 }
2300 
2301 static int
2302 re_ioctl(struct ifnet *ifp, u_long command, caddr_t data, struct ucred *cr)
2303 {
2304 	struct re_softc *sc = ifp->if_softc;
2305 	struct ifreq *ifr = (struct ifreq *) data;
2306 	struct mii_data *mii;
2307 	int error = 0;
2308 
2309 	ASSERT_SERIALIZED(ifp->if_serializer);
2310 
2311 	switch(command) {
2312 	case SIOCSIFMTU:
2313 		if (ifr->ifr_mtu > RE_JUMBO_MTU)
2314 			error = EINVAL;
2315 		ifp->if_mtu = ifr->ifr_mtu;
2316 		break;
2317 	case SIOCSIFFLAGS:
2318 		if (ifp->if_flags & IFF_UP)
2319 			re_init(sc);
2320 		else if (ifp->if_flags & IFF_RUNNING)
2321 			re_stop(sc);
2322 		break;
2323 	case SIOCADDMULTI:
2324 	case SIOCDELMULTI:
2325 		re_setmulti(sc);
2326 		error = 0;
2327 		break;
2328 	case SIOCGIFMEDIA:
2329 	case SIOCSIFMEDIA:
2330 		mii = device_get_softc(sc->re_miibus);
2331 		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, command);
2332 		break;
2333 	case SIOCSIFCAP:
2334 		ifp->if_capenable &= ~(IFCAP_HWCSUM);
2335 		ifp->if_capenable |=
2336 		    ifr->ifr_reqcap & (IFCAP_HWCSUM);
2337 		if (ifp->if_capenable & IFCAP_TXCSUM)
2338 			ifp->if_hwassist = RE_CSUM_FEATURES;
2339 		else
2340 			ifp->if_hwassist = 0;
2341 		if (ifp->if_flags & IFF_RUNNING)
2342 			re_init(sc);
2343 		break;
2344 	default:
2345 		error = ether_ioctl(ifp, command, data);
2346 		break;
2347 	}
2348 	return(error);
2349 }
2350 
2351 static void
2352 re_watchdog(struct ifnet *ifp)
2353 {
2354 	struct re_softc *sc = ifp->if_softc;
2355 
2356 	ASSERT_SERIALIZED(ifp->if_serializer);
2357 
2358 	if_printf(ifp, "watchdog timeout\n");
2359 
2360 	ifp->if_oerrors++;
2361 
2362 	re_txeof(sc);
2363 	re_rxeof(sc);
2364 
2365 	re_init(sc);
2366 
2367 	if (!ifq_is_empty(&ifp->if_snd))
2368 		if_devstart(ifp);
2369 }
2370 
2371 /*
2372  * Stop the adapter and free any mbufs allocated to the
2373  * RX and TX lists.
2374  */
2375 static void
2376 re_stop(struct re_softc *sc)
2377 {
2378 	struct ifnet *ifp = &sc->arpcom.ac_if;
2379 	int i;
2380 
2381 	ASSERT_SERIALIZED(ifp->if_serializer);
2382 
2383 	ifp->if_timer = 0;
2384 	callout_stop(&sc->re_timer);
2385 
2386 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
2387 
2388 	CSR_WRITE_1(sc, RE_COMMAND, 0x00);
2389 	CSR_WRITE_2(sc, RE_IMR, 0x0000);
2390 	CSR_WRITE_2(sc, RE_ISR, 0xFFFF);
2391 
2392 	if (sc->re_head != NULL) {
2393 		m_freem(sc->re_head);
2394 		sc->re_head = sc->re_tail = NULL;
2395 	}
2396 
2397 	/* Free the TX list buffers. */
2398 	for (i = 0; i < RE_TX_DESC_CNT; i++) {
2399 		if (sc->re_ldata.re_tx_mbuf[i] != NULL) {
2400 			bus_dmamap_unload(sc->re_ldata.re_mtag,
2401 					  sc->re_ldata.re_tx_dmamap[i]);
2402 			m_freem(sc->re_ldata.re_tx_mbuf[i]);
2403 			sc->re_ldata.re_tx_mbuf[i] = NULL;
2404 		}
2405 	}
2406 
2407 	/* Free the RX list buffers. */
2408 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
2409 		if (sc->re_ldata.re_rx_mbuf[i] != NULL) {
2410 			bus_dmamap_unload(sc->re_ldata.re_mtag,
2411 					  sc->re_ldata.re_rx_dmamap[i]);
2412 			m_freem(sc->re_ldata.re_rx_mbuf[i]);
2413 			sc->re_ldata.re_rx_mbuf[i] = NULL;
2414 		}
2415 	}
2416 }
2417 
2418 /*
2419  * Device suspend routine.  Stop the interface and save some PCI
2420  * settings in case the BIOS doesn't restore them properly on
2421  * resume.
2422  */
2423 static int
2424 re_suspend(device_t dev)
2425 {
2426 #ifndef BURN_BRIDGES
2427 	int i;
2428 #endif
2429 	struct re_softc *sc = device_get_softc(dev);
2430 	struct ifnet *ifp = &sc->arpcom.ac_if;
2431 
2432 	lwkt_serialize_enter(ifp->if_serializer);
2433 
2434 	re_stop(sc);
2435 
2436 #ifndef BURN_BRIDGES
2437 	for (i = 0; i < 5; i++)
2438 		sc->saved_maps[i] = pci_read_config(dev, PCIR_MAPS + i * 4, 4);
2439 	sc->saved_biosaddr = pci_read_config(dev, PCIR_BIOS, 4);
2440 	sc->saved_intline = pci_read_config(dev, PCIR_INTLINE, 1);
2441 	sc->saved_cachelnsz = pci_read_config(dev, PCIR_CACHELNSZ, 1);
2442 	sc->saved_lattimer = pci_read_config(dev, PCIR_LATTIMER, 1);
2443 #endif
2444 
2445 	sc->suspended = 1;
2446 
2447 	lwkt_serialize_exit(ifp->if_serializer);
2448 
2449 	return (0);
2450 }
2451 
2452 /*
2453  * Device resume routine.  Restore some PCI settings in case the BIOS
2454  * doesn't, re-enable busmastering, and restart the interface if
2455  * appropriate.
2456  */
2457 static int
2458 re_resume(device_t dev)
2459 {
2460 	struct re_softc *sc = device_get_softc(dev);
2461 	struct ifnet *ifp = &sc->arpcom.ac_if;
2462 #ifndef BURN_BRIDGES
2463 	int i;
2464 #endif
2465 
2466 	lwkt_serialize_enter(ifp->if_serializer);
2467 
2468 #ifndef BURN_BRIDGES
2469 	/* better way to do this? */
2470 	for (i = 0; i < 5; i++)
2471 		pci_write_config(dev, PCIR_MAPS + i * 4, sc->saved_maps[i], 4);
2472 	pci_write_config(dev, PCIR_BIOS, sc->saved_biosaddr, 4);
2473 	pci_write_config(dev, PCIR_INTLINE, sc->saved_intline, 1);
2474 	pci_write_config(dev, PCIR_CACHELNSZ, sc->saved_cachelnsz, 1);
2475 	pci_write_config(dev, PCIR_LATTIMER, sc->saved_lattimer, 1);
2476 
2477 	/* reenable busmastering */
2478 	pci_enable_busmaster(dev);
2479 	pci_enable_io(dev, SYS_RES_IOPORT);
2480 #endif
2481 
2482 	/* reinitialize interface if necessary */
2483 	if (ifp->if_flags & IFF_UP)
2484 		re_init(sc);
2485 
2486 	sc->suspended = 0;
2487 
2488 	lwkt_serialize_exit(ifp->if_serializer);
2489 
2490 	return (0);
2491 }
2492 
2493 /*
2494  * Stop all chip I/O so that the kernel's probe routines don't
2495  * get confused by errant DMAs when rebooting.
2496  */
2497 static void
2498 re_shutdown(device_t dev)
2499 {
2500 	struct re_softc *sc = device_get_softc(dev);
2501 	struct ifnet *ifp = &sc->arpcom.ac_if;
2502 
2503 	lwkt_serialize_enter(ifp->if_serializer);
2504 	re_stop(sc);
2505 	lwkt_serialize_exit(ifp->if_serializer);
2506 }
2507 
2508 static int
2509 re_sysctl_tx_moderation(SYSCTL_HANDLER_ARGS)
2510 {
2511 	struct re_softc *sc = arg1;
2512 	struct ifnet *ifp = &sc->arpcom.ac_if;
2513 	int error = 0, mod, mod_old;
2514 
2515 	lwkt_serialize_enter(ifp->if_serializer);
2516 
2517 	mod_old = mod = RE_TX_MODERATION_IS_ENABLED(sc);
2518 
2519 	error = sysctl_handle_int(oidp, &mod, 0, req);
2520 	if (error || req->newptr == NULL || mod == mod_old)
2521 		goto back;
2522 	if (mod != 0 && mod != 1) {
2523 		error = EINVAL;
2524 		goto back;
2525 	}
2526 
2527 	if (mod)
2528 		RE_ENABLE_TX_MODERATION(sc);
2529 	else
2530 		RE_DISABLE_TX_MODERATION(sc);
2531 
2532 	if ((ifp->if_flags & (IFF_RUNNING | IFF_UP)) == (IFF_RUNNING | IFF_UP))
2533 		re_init(sc);
2534 back:
2535 	lwkt_serialize_exit(ifp->if_serializer);
2536 	return error;
2537 }
2538