xref: /freebsd/sys/dev/re/if_re.c (revision 6472ac3d8a86336899b6cfb789a4cd9897e3fab5)
1 /*-
2  * Copyright (c) 1997, 1998-2003
3  *	Bill Paul <wpaul@windriver.com>.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. All advertising materials mentioning features or use of this software
14  *    must display the following acknowledgement:
15  *	This product includes software developed by Bill Paul.
16  * 4. Neither the name of the author nor the names of any co-contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY Bill Paul AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL Bill Paul OR THE VOICES IN HIS HEAD
24  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
30  * THE POSSIBILITY OF SUCH DAMAGE.
31  */
32 
33 #include <sys/cdefs.h>
34 __FBSDID("$FreeBSD$");
35 
36 /*
37  * RealTek 8139C+/8169/8169S/8110S/8168/8111/8101E PCI NIC driver
38  *
39  * Written by Bill Paul <wpaul@windriver.com>
40  * Senior Networking Software Engineer
41  * Wind River Systems
42  */
43 
44 /*
45  * This driver is designed to support RealTek's next generation of
46  * 10/100 and 10/100/1000 PCI ethernet controllers. There are currently
47  * seven devices in this family: the RTL8139C+, the RTL8169, the RTL8169S,
48  * RTL8110S, the RTL8168, the RTL8111 and the RTL8101E.
49  *
50  * The 8139C+ is a 10/100 ethernet chip. It is backwards compatible
51  * with the older 8139 family, however it also supports a special
52  * C+ mode of operation that provides several new performance enhancing
53  * features. These include:
54  *
55  *	o Descriptor based DMA mechanism. Each descriptor represents
56  *	  a single packet fragment. Data buffers may be aligned on
57  *	  any byte boundary.
58  *
59  *	o 64-bit DMA
60  *
61  *	o TCP/IP checksum offload for both RX and TX
62  *
63  *	o High and normal priority transmit DMA rings
64  *
65  *	o VLAN tag insertion and extraction
66  *
67  *	o TCP large send (segmentation offload)
68  *
69  * Like the 8139, the 8139C+ also has a built-in 10/100 PHY. The C+
70  * programming API is fairly straightforward. The RX filtering, EEPROM
71  * access and PHY access is the same as it is on the older 8139 series
72  * chips.
73  *
74  * The 8169 is a 64-bit 10/100/1000 gigabit ethernet MAC. It has almost the
75  * same programming API and feature set as the 8139C+ with the following
76  * differences and additions:
77  *
78  *	o 1000Mbps mode
79  *
80  *	o Jumbo frames
81  *
82  *	o GMII and TBI ports/registers for interfacing with copper
83  *	  or fiber PHYs
84  *
85  *	o RX and TX DMA rings can have up to 1024 descriptors
86  *	  (the 8139C+ allows a maximum of 64)
87  *
88  *	o Slight differences in register layout from the 8139C+
89  *
90  * The TX start and timer interrupt registers are at different locations
91  * on the 8169 than they are on the 8139C+. Also, the status word in the
92  * RX descriptor has a slightly different bit layout. The 8169 does not
93  * have a built-in PHY. Most reference boards use a Marvell 88E1000 'Alaska'
94  * copper gigE PHY.
95  *
96  * The 8169S/8110S 10/100/1000 devices have built-in copper gigE PHYs
97  * (the 'S' stands for 'single-chip'). These devices have the same
98  * programming API as the older 8169, but also have some vendor-specific
99  * registers for the on-board PHY. The 8110S is a LAN-on-motherboard
100  * part designed to be pin-compatible with the RealTek 8100 10/100 chip.
101  *
102  * This driver takes advantage of the RX and TX checksum offload and
103  * VLAN tag insertion/extraction features. It also implements TX
104  * interrupt moderation using the timer interrupt registers, which
105  * significantly reduces TX interrupt load. There is also support
106  * for jumbo frames, however the 8169/8169S/8110S can not transmit
107  * jumbo frames larger than 7440, so the max MTU possible with this
108  * driver is 7422 bytes.
109  */
110 
111 #ifdef HAVE_KERNEL_OPTION_HEADERS
112 #include "opt_device_polling.h"
113 #endif
114 
115 #include <sys/param.h>
116 #include <sys/endian.h>
117 #include <sys/systm.h>
118 #include <sys/sockio.h>
119 #include <sys/mbuf.h>
120 #include <sys/malloc.h>
121 #include <sys/module.h>
122 #include <sys/kernel.h>
123 #include <sys/socket.h>
124 #include <sys/lock.h>
125 #include <sys/mutex.h>
126 #include <sys/sysctl.h>
127 #include <sys/taskqueue.h>
128 
129 #include <net/if.h>
130 #include <net/if_arp.h>
131 #include <net/ethernet.h>
132 #include <net/if_dl.h>
133 #include <net/if_media.h>
134 #include <net/if_types.h>
135 #include <net/if_vlan_var.h>
136 
137 #include <net/bpf.h>
138 
139 #include <machine/bus.h>
140 #include <machine/resource.h>
141 #include <sys/bus.h>
142 #include <sys/rman.h>
143 
144 #include <dev/mii/mii.h>
145 #include <dev/mii/miivar.h>
146 
147 #include <dev/pci/pcireg.h>
148 #include <dev/pci/pcivar.h>
149 
150 #include <pci/if_rlreg.h>
151 
152 MODULE_DEPEND(re, pci, 1, 1, 1);
153 MODULE_DEPEND(re, ether, 1, 1, 1);
154 MODULE_DEPEND(re, miibus, 1, 1, 1);
155 
156 /* "device miibus" required.  See GENERIC if you get errors here. */
157 #include "miibus_if.h"
158 
159 /* Tunables. */
160 static int intr_filter = 0;
161 TUNABLE_INT("hw.re.intr_filter", &intr_filter);
162 static int msi_disable = 0;
163 TUNABLE_INT("hw.re.msi_disable", &msi_disable);
164 static int msix_disable = 0;
165 TUNABLE_INT("hw.re.msix_disable", &msix_disable);
166 static int prefer_iomap = 0;
167 TUNABLE_INT("hw.re.prefer_iomap", &prefer_iomap);
168 
169 #define RE_CSUM_FEATURES    (CSUM_IP | CSUM_TCP | CSUM_UDP)
170 
171 /*
172  * Various supported device vendors/types and their names.
173  */
174 static const struct rl_type const re_devs[] = {
175 	{ DLINK_VENDORID, DLINK_DEVICEID_528T, 0,
176 	    "D-Link DGE-528(T) Gigabit Ethernet Adapter" },
177 	{ DLINK_VENDORID, DLINK_DEVICEID_530T_REVC, 0,
178 	    "D-Link DGE-530(T) Gigabit Ethernet Adapter" },
179 	{ RT_VENDORID, RT_DEVICEID_8139, 0,
180 	    "RealTek 8139C+ 10/100BaseTX" },
181 	{ RT_VENDORID, RT_DEVICEID_8101E, 0,
182 	    "RealTek 810xE PCIe 10/100baseTX" },
183 	{ RT_VENDORID, RT_DEVICEID_8168, 0,
184 	    "RealTek 8168/8111 B/C/CP/D/DP/E PCIe Gigabit Ethernet" },
185 	{ RT_VENDORID, RT_DEVICEID_8169, 0,
186 	    "RealTek 8169/8169S/8169SB(L)/8110S/8110SB(L) Gigabit Ethernet" },
187 	{ RT_VENDORID, RT_DEVICEID_8169SC, 0,
188 	    "RealTek 8169SC/8110SC Single-chip Gigabit Ethernet" },
189 	{ COREGA_VENDORID, COREGA_DEVICEID_CGLAPCIGT, 0,
190 	    "Corega CG-LAPCIGT (RTL8169S) Gigabit Ethernet" },
191 	{ LINKSYS_VENDORID, LINKSYS_DEVICEID_EG1032, 0,
192 	    "Linksys EG1032 (RTL8169S) Gigabit Ethernet" },
193 	{ USR_VENDORID, USR_DEVICEID_997902, 0,
194 	    "US Robotics 997902 (RTL8169S) Gigabit Ethernet" }
195 };
196 
197 static const struct rl_hwrev const re_hwrevs[] = {
198 	{ RL_HWREV_8139, RL_8139, "", RL_MTU },
199 	{ RL_HWREV_8139A, RL_8139, "A", RL_MTU },
200 	{ RL_HWREV_8139AG, RL_8139, "A-G", RL_MTU },
201 	{ RL_HWREV_8139B, RL_8139, "B", RL_MTU },
202 	{ RL_HWREV_8130, RL_8139, "8130", RL_MTU },
203 	{ RL_HWREV_8139C, RL_8139, "C", RL_MTU },
204 	{ RL_HWREV_8139D, RL_8139, "8139D/8100B/8100C", RL_MTU },
205 	{ RL_HWREV_8139CPLUS, RL_8139CPLUS, "C+", RL_MTU },
206 	{ RL_HWREV_8168B_SPIN1, RL_8169, "8168", RL_JUMBO_MTU },
207 	{ RL_HWREV_8169, RL_8169, "8169", RL_JUMBO_MTU },
208 	{ RL_HWREV_8169S, RL_8169, "8169S", RL_JUMBO_MTU },
209 	{ RL_HWREV_8110S, RL_8169, "8110S", RL_JUMBO_MTU },
210 	{ RL_HWREV_8169_8110SB, RL_8169, "8169SB/8110SB", RL_JUMBO_MTU },
211 	{ RL_HWREV_8169_8110SC, RL_8169, "8169SC/8110SC", RL_JUMBO_MTU },
212 	{ RL_HWREV_8169_8110SBL, RL_8169, "8169SBL/8110SBL", RL_JUMBO_MTU },
213 	{ RL_HWREV_8169_8110SCE, RL_8169, "8169SC/8110SC", RL_JUMBO_MTU },
214 	{ RL_HWREV_8100, RL_8139, "8100", RL_MTU },
215 	{ RL_HWREV_8101, RL_8139, "8101", RL_MTU },
216 	{ RL_HWREV_8100E, RL_8169, "8100E", RL_MTU },
217 	{ RL_HWREV_8101E, RL_8169, "8101E", RL_MTU },
218 	{ RL_HWREV_8102E, RL_8169, "8102E", RL_MTU },
219 	{ RL_HWREV_8102EL, RL_8169, "8102EL", RL_MTU },
220 	{ RL_HWREV_8102EL_SPIN1, RL_8169, "8102EL", RL_MTU },
221 	{ RL_HWREV_8103E, RL_8169, "8103E", RL_MTU },
222 	{ RL_HWREV_8401E, RL_8169, "8401E", RL_MTU },
223 	{ RL_HWREV_8105E, RL_8169, "8105E", RL_MTU },
224 	{ RL_HWREV_8168B_SPIN2, RL_8169, "8168", RL_JUMBO_MTU },
225 	{ RL_HWREV_8168B_SPIN3, RL_8169, "8168", RL_JUMBO_MTU },
226 	{ RL_HWREV_8168C, RL_8169, "8168C/8111C", RL_JUMBO_MTU_6K },
227 	{ RL_HWREV_8168C_SPIN2, RL_8169, "8168C/8111C", RL_JUMBO_MTU_6K },
228 	{ RL_HWREV_8168CP, RL_8169, "8168CP/8111CP", RL_JUMBO_MTU_6K },
229 	{ RL_HWREV_8168D, RL_8169, "8168D/8111D", RL_JUMBO_MTU_9K },
230 	{ RL_HWREV_8168DP, RL_8169, "8168DP/8111DP", RL_JUMBO_MTU_9K },
231 	{ RL_HWREV_8168E, RL_8169, "8168E/8111E", RL_JUMBO_MTU_9K},
232 	{ RL_HWREV_8168E_VL, RL_8169, "8168E/8111E-VL", RL_JUMBO_MTU_6K},
233 	{ 0, 0, NULL, 0 }
234 };
235 
236 static int re_probe		(device_t);
237 static int re_attach		(device_t);
238 static int re_detach		(device_t);
239 
240 static int re_encap		(struct rl_softc *, struct mbuf **);
241 
242 static void re_dma_map_addr	(void *, bus_dma_segment_t *, int, int);
243 static int re_allocmem		(device_t, struct rl_softc *);
244 static __inline void re_discard_rxbuf
245 				(struct rl_softc *, int);
246 static int re_newbuf		(struct rl_softc *, int);
247 static int re_jumbo_newbuf	(struct rl_softc *, int);
248 static int re_rx_list_init	(struct rl_softc *);
249 static int re_jrx_list_init	(struct rl_softc *);
250 static int re_tx_list_init	(struct rl_softc *);
251 #ifdef RE_FIXUP_RX
252 static __inline void re_fixup_rx
253 				(struct mbuf *);
254 #endif
255 static int re_rxeof		(struct rl_softc *, int *);
256 static void re_txeof		(struct rl_softc *);
257 #ifdef DEVICE_POLLING
258 static int re_poll		(struct ifnet *, enum poll_cmd, int);
259 static int re_poll_locked	(struct ifnet *, enum poll_cmd, int);
260 #endif
261 static int re_intr		(void *);
262 static void re_intr_msi		(void *);
263 static void re_tick		(void *);
264 static void re_int_task		(void *, int);
265 static void re_start		(struct ifnet *);
266 static void re_start_locked	(struct ifnet *);
267 static int re_ioctl		(struct ifnet *, u_long, caddr_t);
268 static void re_init		(void *);
269 static void re_init_locked	(struct rl_softc *);
270 static void re_stop		(struct rl_softc *);
271 static void re_watchdog		(struct rl_softc *);
272 static int re_suspend		(device_t);
273 static int re_resume		(device_t);
274 static int re_shutdown		(device_t);
275 static int re_ifmedia_upd	(struct ifnet *);
276 static void re_ifmedia_sts	(struct ifnet *, struct ifmediareq *);
277 
278 static void re_eeprom_putbyte	(struct rl_softc *, int);
279 static void re_eeprom_getword	(struct rl_softc *, int, u_int16_t *);
280 static void re_read_eeprom	(struct rl_softc *, caddr_t, int, int);
281 static int re_gmii_readreg	(device_t, int, int);
282 static int re_gmii_writereg	(device_t, int, int, int);
283 
284 static int re_miibus_readreg	(device_t, int, int);
285 static int re_miibus_writereg	(device_t, int, int, int);
286 static void re_miibus_statchg	(device_t);
287 
288 static void re_set_jumbo	(struct rl_softc *, int);
289 static void re_set_rxmode		(struct rl_softc *);
290 static void re_reset		(struct rl_softc *);
291 static void re_setwol		(struct rl_softc *);
292 static void re_clrwol		(struct rl_softc *);
293 
294 #ifdef RE_DIAG
295 static int re_diag		(struct rl_softc *);
296 #endif
297 
298 static void re_add_sysctls	(struct rl_softc *);
299 static int re_sysctl_stats	(SYSCTL_HANDLER_ARGS);
300 static int sysctl_int_range	(SYSCTL_HANDLER_ARGS, int, int);
301 static int sysctl_hw_re_int_mod	(SYSCTL_HANDLER_ARGS);
302 
303 static device_method_t re_methods[] = {
304 	/* Device interface */
305 	DEVMETHOD(device_probe,		re_probe),
306 	DEVMETHOD(device_attach,	re_attach),
307 	DEVMETHOD(device_detach,	re_detach),
308 	DEVMETHOD(device_suspend,	re_suspend),
309 	DEVMETHOD(device_resume,	re_resume),
310 	DEVMETHOD(device_shutdown,	re_shutdown),
311 
312 	/* bus interface */
313 	DEVMETHOD(bus_print_child,	bus_generic_print_child),
314 	DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
315 
316 	/* MII interface */
317 	DEVMETHOD(miibus_readreg,	re_miibus_readreg),
318 	DEVMETHOD(miibus_writereg,	re_miibus_writereg),
319 	DEVMETHOD(miibus_statchg,	re_miibus_statchg),
320 
321 	{ 0, 0 }
322 };
323 
324 static driver_t re_driver = {
325 	"re",
326 	re_methods,
327 	sizeof(struct rl_softc)
328 };
329 
330 static devclass_t re_devclass;
331 
332 DRIVER_MODULE(re, pci, re_driver, re_devclass, 0, 0);
333 DRIVER_MODULE(miibus, re, miibus_driver, miibus_devclass, 0, 0);
334 
335 #define EE_SET(x)					\
336 	CSR_WRITE_1(sc, RL_EECMD,			\
337 		CSR_READ_1(sc, RL_EECMD) | x)
338 
339 #define EE_CLR(x)					\
340 	CSR_WRITE_1(sc, RL_EECMD,			\
341 		CSR_READ_1(sc, RL_EECMD) & ~x)
342 
343 /*
344  * Send a read command and address to the EEPROM, check for ACK.
345  */
346 static void
347 re_eeprom_putbyte(struct rl_softc *sc, int addr)
348 {
349 	int			d, i;
350 
351 	d = addr | (RL_9346_READ << sc->rl_eewidth);
352 
353 	/*
354 	 * Feed in each bit and strobe the clock.
355 	 */
356 
357 	for (i = 1 << (sc->rl_eewidth + 3); i; i >>= 1) {
358 		if (d & i) {
359 			EE_SET(RL_EE_DATAIN);
360 		} else {
361 			EE_CLR(RL_EE_DATAIN);
362 		}
363 		DELAY(100);
364 		EE_SET(RL_EE_CLK);
365 		DELAY(150);
366 		EE_CLR(RL_EE_CLK);
367 		DELAY(100);
368 	}
369 }
370 
371 /*
372  * Read a word of data stored in the EEPROM at address 'addr.'
373  */
374 static void
375 re_eeprom_getword(struct rl_softc *sc, int addr, u_int16_t *dest)
376 {
377 	int			i;
378 	u_int16_t		word = 0;
379 
380 	/*
381 	 * Send address of word we want to read.
382 	 */
383 	re_eeprom_putbyte(sc, addr);
384 
385 	/*
386 	 * Start reading bits from EEPROM.
387 	 */
388 	for (i = 0x8000; i; i >>= 1) {
389 		EE_SET(RL_EE_CLK);
390 		DELAY(100);
391 		if (CSR_READ_1(sc, RL_EECMD) & RL_EE_DATAOUT)
392 			word |= i;
393 		EE_CLR(RL_EE_CLK);
394 		DELAY(100);
395 	}
396 
397 	*dest = word;
398 }
399 
400 /*
401  * Read a sequence of words from the EEPROM.
402  */
403 static void
404 re_read_eeprom(struct rl_softc *sc, caddr_t dest, int off, int cnt)
405 {
406 	int			i;
407 	u_int16_t		word = 0, *ptr;
408 
409 	CSR_SETBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM);
410 
411         DELAY(100);
412 
413 	for (i = 0; i < cnt; i++) {
414 		CSR_SETBIT_1(sc, RL_EECMD, RL_EE_SEL);
415 		re_eeprom_getword(sc, off + i, &word);
416 		CSR_CLRBIT_1(sc, RL_EECMD, RL_EE_SEL);
417 		ptr = (u_int16_t *)(dest + (i * 2));
418                 *ptr = word;
419 	}
420 
421 	CSR_CLRBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM);
422 }
423 
424 static int
425 re_gmii_readreg(device_t dev, int phy, int reg)
426 {
427 	struct rl_softc		*sc;
428 	u_int32_t		rval;
429 	int			i;
430 
431 	sc = device_get_softc(dev);
432 
433 	/* Let the rgephy driver read the GMEDIASTAT register */
434 
435 	if (reg == RL_GMEDIASTAT) {
436 		rval = CSR_READ_1(sc, RL_GMEDIASTAT);
437 		return (rval);
438 	}
439 
440 	CSR_WRITE_4(sc, RL_PHYAR, reg << 16);
441 
442 	for (i = 0; i < RL_PHY_TIMEOUT; i++) {
443 		rval = CSR_READ_4(sc, RL_PHYAR);
444 		if (rval & RL_PHYAR_BUSY)
445 			break;
446 		DELAY(25);
447 	}
448 
449 	if (i == RL_PHY_TIMEOUT) {
450 		device_printf(sc->rl_dev, "PHY read failed\n");
451 		return (0);
452 	}
453 
454 	/*
455 	 * Controller requires a 20us delay to process next MDIO request.
456 	 */
457 	DELAY(20);
458 
459 	return (rval & RL_PHYAR_PHYDATA);
460 }
461 
462 static int
463 re_gmii_writereg(device_t dev, int phy, int reg, int data)
464 {
465 	struct rl_softc		*sc;
466 	u_int32_t		rval;
467 	int			i;
468 
469 	sc = device_get_softc(dev);
470 
471 	CSR_WRITE_4(sc, RL_PHYAR, (reg << 16) |
472 	    (data & RL_PHYAR_PHYDATA) | RL_PHYAR_BUSY);
473 
474 	for (i = 0; i < RL_PHY_TIMEOUT; i++) {
475 		rval = CSR_READ_4(sc, RL_PHYAR);
476 		if (!(rval & RL_PHYAR_BUSY))
477 			break;
478 		DELAY(25);
479 	}
480 
481 	if (i == RL_PHY_TIMEOUT) {
482 		device_printf(sc->rl_dev, "PHY write failed\n");
483 		return (0);
484 	}
485 
486 	/*
487 	 * Controller requires a 20us delay to process next MDIO request.
488 	 */
489 	DELAY(20);
490 
491 	return (0);
492 }
493 
494 static int
495 re_miibus_readreg(device_t dev, int phy, int reg)
496 {
497 	struct rl_softc		*sc;
498 	u_int16_t		rval = 0;
499 	u_int16_t		re8139_reg = 0;
500 
501 	sc = device_get_softc(dev);
502 
503 	if (sc->rl_type == RL_8169) {
504 		rval = re_gmii_readreg(dev, phy, reg);
505 		return (rval);
506 	}
507 
508 	switch (reg) {
509 	case MII_BMCR:
510 		re8139_reg = RL_BMCR;
511 		break;
512 	case MII_BMSR:
513 		re8139_reg = RL_BMSR;
514 		break;
515 	case MII_ANAR:
516 		re8139_reg = RL_ANAR;
517 		break;
518 	case MII_ANER:
519 		re8139_reg = RL_ANER;
520 		break;
521 	case MII_ANLPAR:
522 		re8139_reg = RL_LPAR;
523 		break;
524 	case MII_PHYIDR1:
525 	case MII_PHYIDR2:
526 		return (0);
527 	/*
528 	 * Allow the rlphy driver to read the media status
529 	 * register. If we have a link partner which does not
530 	 * support NWAY, this is the register which will tell
531 	 * us the results of parallel detection.
532 	 */
533 	case RL_MEDIASTAT:
534 		rval = CSR_READ_1(sc, RL_MEDIASTAT);
535 		return (rval);
536 	default:
537 		device_printf(sc->rl_dev, "bad phy register\n");
538 		return (0);
539 	}
540 	rval = CSR_READ_2(sc, re8139_reg);
541 	if (sc->rl_type == RL_8139CPLUS && re8139_reg == RL_BMCR) {
542 		/* 8139C+ has different bit layout. */
543 		rval &= ~(BMCR_LOOP | BMCR_ISO);
544 	}
545 	return (rval);
546 }
547 
548 static int
549 re_miibus_writereg(device_t dev, int phy, int reg, int data)
550 {
551 	struct rl_softc		*sc;
552 	u_int16_t		re8139_reg = 0;
553 	int			rval = 0;
554 
555 	sc = device_get_softc(dev);
556 
557 	if (sc->rl_type == RL_8169) {
558 		rval = re_gmii_writereg(dev, phy, reg, data);
559 		return (rval);
560 	}
561 
562 	switch (reg) {
563 	case MII_BMCR:
564 		re8139_reg = RL_BMCR;
565 		if (sc->rl_type == RL_8139CPLUS) {
566 			/* 8139C+ has different bit layout. */
567 			data &= ~(BMCR_LOOP | BMCR_ISO);
568 		}
569 		break;
570 	case MII_BMSR:
571 		re8139_reg = RL_BMSR;
572 		break;
573 	case MII_ANAR:
574 		re8139_reg = RL_ANAR;
575 		break;
576 	case MII_ANER:
577 		re8139_reg = RL_ANER;
578 		break;
579 	case MII_ANLPAR:
580 		re8139_reg = RL_LPAR;
581 		break;
582 	case MII_PHYIDR1:
583 	case MII_PHYIDR2:
584 		return (0);
585 		break;
586 	default:
587 		device_printf(sc->rl_dev, "bad phy register\n");
588 		return (0);
589 	}
590 	CSR_WRITE_2(sc, re8139_reg, data);
591 	return (0);
592 }
593 
594 static void
595 re_miibus_statchg(device_t dev)
596 {
597 	struct rl_softc		*sc;
598 	struct ifnet		*ifp;
599 	struct mii_data		*mii;
600 
601 	sc = device_get_softc(dev);
602 	mii = device_get_softc(sc->rl_miibus);
603 	ifp = sc->rl_ifp;
604 	if (mii == NULL || ifp == NULL ||
605 	    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
606 		return;
607 
608 	sc->rl_flags &= ~RL_FLAG_LINK;
609 	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
610 	    (IFM_ACTIVE | IFM_AVALID)) {
611 		switch (IFM_SUBTYPE(mii->mii_media_active)) {
612 		case IFM_10_T:
613 		case IFM_100_TX:
614 			sc->rl_flags |= RL_FLAG_LINK;
615 			break;
616 		case IFM_1000_T:
617 			if ((sc->rl_flags & RL_FLAG_FASTETHER) != 0)
618 				break;
619 			sc->rl_flags |= RL_FLAG_LINK;
620 			break;
621 		default:
622 			break;
623 		}
624 	}
625 	/*
626 	 * RealTek controllers does not provide any interface to
627 	 * Tx/Rx MACs for resolved speed, duplex and flow-control
628 	 * parameters.
629 	 */
630 }
631 
632 /*
633  * Set the RX configuration and 64-bit multicast hash filter.
634  */
635 static void
636 re_set_rxmode(struct rl_softc *sc)
637 {
638 	struct ifnet		*ifp;
639 	struct ifmultiaddr	*ifma;
640 	uint32_t		hashes[2] = { 0, 0 };
641 	uint32_t		h, rxfilt;
642 
643 	RL_LOCK_ASSERT(sc);
644 
645 	ifp = sc->rl_ifp;
646 
647 	rxfilt = RL_RXCFG_CONFIG | RL_RXCFG_RX_INDIV | RL_RXCFG_RX_BROAD;
648 
649 	if (ifp->if_flags & (IFF_ALLMULTI | IFF_PROMISC)) {
650 		if (ifp->if_flags & IFF_PROMISC)
651 			rxfilt |= RL_RXCFG_RX_ALLPHYS;
652 		/*
653 		 * Unlike other hardwares, we have to explicitly set
654 		 * RL_RXCFG_RX_MULTI to receive multicast frames in
655 		 * promiscuous mode.
656 		 */
657 		rxfilt |= RL_RXCFG_RX_MULTI;
658 		hashes[0] = hashes[1] = 0xffffffff;
659 		goto done;
660 	}
661 
662 	if_maddr_rlock(ifp);
663 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
664 		if (ifma->ifma_addr->sa_family != AF_LINK)
665 			continue;
666 		h = ether_crc32_be(LLADDR((struct sockaddr_dl *)
667 		    ifma->ifma_addr), ETHER_ADDR_LEN) >> 26;
668 		if (h < 32)
669 			hashes[0] |= (1 << h);
670 		else
671 			hashes[1] |= (1 << (h - 32));
672 	}
673 	if_maddr_runlock(ifp);
674 
675 	if (hashes[0] != 0 || hashes[1] != 0) {
676 		/*
677 		 * For some unfathomable reason, RealTek decided to
678 		 * reverse the order of the multicast hash registers
679 		 * in the PCI Express parts.  This means we have to
680 		 * write the hash pattern in reverse order for those
681 		 * devices.
682 		 */
683 		if ((sc->rl_flags & RL_FLAG_PCIE) != 0) {
684 			h = bswap32(hashes[0]);
685 			hashes[0] = bswap32(hashes[1]);
686 			hashes[1] = h;
687 		}
688 		rxfilt |= RL_RXCFG_RX_MULTI;
689 	}
690 
691 done:
692 	CSR_WRITE_4(sc, RL_MAR0, hashes[0]);
693 	CSR_WRITE_4(sc, RL_MAR4, hashes[1]);
694 	CSR_WRITE_4(sc, RL_RXCFG, rxfilt);
695 }
696 
697 static void
698 re_reset(struct rl_softc *sc)
699 {
700 	int			i;
701 
702 	RL_LOCK_ASSERT(sc);
703 
704 	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RESET);
705 
706 	for (i = 0; i < RL_TIMEOUT; i++) {
707 		DELAY(10);
708 		if (!(CSR_READ_1(sc, RL_COMMAND) & RL_CMD_RESET))
709 			break;
710 	}
711 	if (i == RL_TIMEOUT)
712 		device_printf(sc->rl_dev, "reset never completed!\n");
713 
714 	if ((sc->rl_flags & RL_FLAG_MACRESET) != 0)
715 		CSR_WRITE_1(sc, 0x82, 1);
716 	if (sc->rl_hwrev->rl_rev == RL_HWREV_8169S)
717 		re_gmii_writereg(sc->rl_dev, 1, 0x0b, 0);
718 }
719 
720 #ifdef RE_DIAG
721 
722 /*
723  * The following routine is designed to test for a defect on some
724  * 32-bit 8169 cards. Some of these NICs have the REQ64# and ACK64#
725  * lines connected to the bus, however for a 32-bit only card, they
726  * should be pulled high. The result of this defect is that the
727  * NIC will not work right if you plug it into a 64-bit slot: DMA
728  * operations will be done with 64-bit transfers, which will fail
729  * because the 64-bit data lines aren't connected.
730  *
731  * There's no way to work around this (short of talking a soldering
732  * iron to the board), however we can detect it. The method we use
733  * here is to put the NIC into digital loopback mode, set the receiver
734  * to promiscuous mode, and then try to send a frame. We then compare
735  * the frame data we sent to what was received. If the data matches,
736  * then the NIC is working correctly, otherwise we know the user has
737  * a defective NIC which has been mistakenly plugged into a 64-bit PCI
738  * slot. In the latter case, there's no way the NIC can work correctly,
739  * so we print out a message on the console and abort the device attach.
740  */
741 
742 static int
743 re_diag(struct rl_softc *sc)
744 {
745 	struct ifnet		*ifp = sc->rl_ifp;
746 	struct mbuf		*m0;
747 	struct ether_header	*eh;
748 	struct rl_desc		*cur_rx;
749 	u_int16_t		status;
750 	u_int32_t		rxstat;
751 	int			total_len, i, error = 0, phyaddr;
752 	u_int8_t		dst[] = { 0x00, 'h', 'e', 'l', 'l', 'o' };
753 	u_int8_t		src[] = { 0x00, 'w', 'o', 'r', 'l', 'd' };
754 
755 	/* Allocate a single mbuf */
756 	MGETHDR(m0, M_DONTWAIT, MT_DATA);
757 	if (m0 == NULL)
758 		return (ENOBUFS);
759 
760 	RL_LOCK(sc);
761 
762 	/*
763 	 * Initialize the NIC in test mode. This sets the chip up
764 	 * so that it can send and receive frames, but performs the
765 	 * following special functions:
766 	 * - Puts receiver in promiscuous mode
767 	 * - Enables digital loopback mode
768 	 * - Leaves interrupts turned off
769 	 */
770 
771 	ifp->if_flags |= IFF_PROMISC;
772 	sc->rl_testmode = 1;
773 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
774 	re_init_locked(sc);
775 	sc->rl_flags |= RL_FLAG_LINK;
776 	if (sc->rl_type == RL_8169)
777 		phyaddr = 1;
778 	else
779 		phyaddr = 0;
780 
781 	re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_RESET);
782 	for (i = 0; i < RL_TIMEOUT; i++) {
783 		status = re_miibus_readreg(sc->rl_dev, phyaddr, MII_BMCR);
784 		if (!(status & BMCR_RESET))
785 			break;
786 	}
787 
788 	re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_LOOP);
789 	CSR_WRITE_2(sc, RL_ISR, RL_INTRS);
790 
791 	DELAY(100000);
792 
793 	/* Put some data in the mbuf */
794 
795 	eh = mtod(m0, struct ether_header *);
796 	bcopy ((char *)&dst, eh->ether_dhost, ETHER_ADDR_LEN);
797 	bcopy ((char *)&src, eh->ether_shost, ETHER_ADDR_LEN);
798 	eh->ether_type = htons(ETHERTYPE_IP);
799 	m0->m_pkthdr.len = m0->m_len = ETHER_MIN_LEN - ETHER_CRC_LEN;
800 
801 	/*
802 	 * Queue the packet, start transmission.
803 	 * Note: IF_HANDOFF() ultimately calls re_start() for us.
804 	 */
805 
806 	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
807 	RL_UNLOCK(sc);
808 	/* XXX: re_diag must not be called when in ALTQ mode */
809 	IF_HANDOFF(&ifp->if_snd, m0, ifp);
810 	RL_LOCK(sc);
811 	m0 = NULL;
812 
813 	/* Wait for it to propagate through the chip */
814 
815 	DELAY(100000);
816 	for (i = 0; i < RL_TIMEOUT; i++) {
817 		status = CSR_READ_2(sc, RL_ISR);
818 		CSR_WRITE_2(sc, RL_ISR, status);
819 		if ((status & (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK)) ==
820 		    (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK))
821 			break;
822 		DELAY(10);
823 	}
824 
825 	if (i == RL_TIMEOUT) {
826 		device_printf(sc->rl_dev,
827 		    "diagnostic failed, failed to receive packet in"
828 		    " loopback mode\n");
829 		error = EIO;
830 		goto done;
831 	}
832 
833 	/*
834 	 * The packet should have been dumped into the first
835 	 * entry in the RX DMA ring. Grab it from there.
836 	 */
837 
838 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
839 	    sc->rl_ldata.rl_rx_list_map,
840 	    BUS_DMASYNC_POSTREAD);
841 	bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag,
842 	    sc->rl_ldata.rl_rx_desc[0].rx_dmamap,
843 	    BUS_DMASYNC_POSTREAD);
844 	bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag,
845 	    sc->rl_ldata.rl_rx_desc[0].rx_dmamap);
846 
847 	m0 = sc->rl_ldata.rl_rx_desc[0].rx_m;
848 	sc->rl_ldata.rl_rx_desc[0].rx_m = NULL;
849 	eh = mtod(m0, struct ether_header *);
850 
851 	cur_rx = &sc->rl_ldata.rl_rx_list[0];
852 	total_len = RL_RXBYTES(cur_rx);
853 	rxstat = le32toh(cur_rx->rl_cmdstat);
854 
855 	if (total_len != ETHER_MIN_LEN) {
856 		device_printf(sc->rl_dev,
857 		    "diagnostic failed, received short packet\n");
858 		error = EIO;
859 		goto done;
860 	}
861 
862 	/* Test that the received packet data matches what we sent. */
863 
864 	if (bcmp((char *)&eh->ether_dhost, (char *)&dst, ETHER_ADDR_LEN) ||
865 	    bcmp((char *)&eh->ether_shost, (char *)&src, ETHER_ADDR_LEN) ||
866 	    ntohs(eh->ether_type) != ETHERTYPE_IP) {
867 		device_printf(sc->rl_dev, "WARNING, DMA FAILURE!\n");
868 		device_printf(sc->rl_dev, "expected TX data: %6D/%6D/0x%x\n",
869 		    dst, ":", src, ":", ETHERTYPE_IP);
870 		device_printf(sc->rl_dev, "received RX data: %6D/%6D/0x%x\n",
871 		    eh->ether_dhost, ":", eh->ether_shost, ":",
872 		    ntohs(eh->ether_type));
873 		device_printf(sc->rl_dev, "You may have a defective 32-bit "
874 		    "NIC plugged into a 64-bit PCI slot.\n");
875 		device_printf(sc->rl_dev, "Please re-install the NIC in a "
876 		    "32-bit slot for proper operation.\n");
877 		device_printf(sc->rl_dev, "Read the re(4) man page for more "
878 		    "details.\n");
879 		error = EIO;
880 	}
881 
882 done:
883 	/* Turn interface off, release resources */
884 
885 	sc->rl_testmode = 0;
886 	sc->rl_flags &= ~RL_FLAG_LINK;
887 	ifp->if_flags &= ~IFF_PROMISC;
888 	re_stop(sc);
889 	if (m0 != NULL)
890 		m_freem(m0);
891 
892 	RL_UNLOCK(sc);
893 
894 	return (error);
895 }
896 
897 #endif
898 
899 /*
900  * Probe for a RealTek 8139C+/8169/8110 chip. Check the PCI vendor and device
901  * IDs against our list and return a device name if we find a match.
902  */
903 static int
904 re_probe(device_t dev)
905 {
906 	const struct rl_type	*t;
907 	uint16_t		devid, vendor;
908 	uint16_t		revid, sdevid;
909 	int			i;
910 
911 	vendor = pci_get_vendor(dev);
912 	devid = pci_get_device(dev);
913 	revid = pci_get_revid(dev);
914 	sdevid = pci_get_subdevice(dev);
915 
916 	if (vendor == LINKSYS_VENDORID && devid == LINKSYS_DEVICEID_EG1032) {
917 		if (sdevid != LINKSYS_SUBDEVICE_EG1032_REV3) {
918 			/*
919 			 * Only attach to rev. 3 of the Linksys EG1032 adapter.
920 			 * Rev. 2 is supported by sk(4).
921 			 */
922 			return (ENXIO);
923 		}
924 	}
925 
926 	if (vendor == RT_VENDORID && devid == RT_DEVICEID_8139) {
927 		if (revid != 0x20) {
928 			/* 8139, let rl(4) take care of this device. */
929 			return (ENXIO);
930 		}
931 	}
932 
933 	t = re_devs;
934 	for (i = 0; i < sizeof(re_devs) / sizeof(re_devs[0]); i++, t++) {
935 		if (vendor == t->rl_vid && devid == t->rl_did) {
936 			device_set_desc(dev, t->rl_name);
937 			return (BUS_PROBE_DEFAULT);
938 		}
939 	}
940 
941 	return (ENXIO);
942 }
943 
944 /*
945  * Map a single buffer address.
946  */
947 
948 static void
949 re_dma_map_addr(void *arg, bus_dma_segment_t *segs, int nseg, int error)
950 {
951 	bus_addr_t		*addr;
952 
953 	if (error)
954 		return;
955 
956 	KASSERT(nseg == 1, ("too many DMA segments, %d should be 1", nseg));
957 	addr = arg;
958 	*addr = segs->ds_addr;
959 }
960 
961 static int
962 re_allocmem(device_t dev, struct rl_softc *sc)
963 {
964 	bus_addr_t		lowaddr;
965 	bus_size_t		rx_list_size, tx_list_size;
966 	int			error;
967 	int			i;
968 
969 	rx_list_size = sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc);
970 	tx_list_size = sc->rl_ldata.rl_tx_desc_cnt * sizeof(struct rl_desc);
971 
972 	/*
973 	 * Allocate the parent bus DMA tag appropriate for PCI.
974 	 * In order to use DAC, RL_CPLUSCMD_PCI_DAC bit of RL_CPLUS_CMD
975 	 * register should be set. However some RealTek chips are known
976 	 * to be buggy on DAC handling, therefore disable DAC by limiting
977 	 * DMA address space to 32bit. PCIe variants of RealTek chips
978 	 * may not have the limitation.
979 	 */
980 	lowaddr = BUS_SPACE_MAXADDR;
981 	if ((sc->rl_flags & RL_FLAG_PCIE) == 0)
982 		lowaddr = BUS_SPACE_MAXADDR_32BIT;
983 	error = bus_dma_tag_create(bus_get_dma_tag(dev), 1, 0,
984 	    lowaddr, BUS_SPACE_MAXADDR, NULL, NULL,
985 	    BUS_SPACE_MAXSIZE_32BIT, 0, BUS_SPACE_MAXSIZE_32BIT, 0,
986 	    NULL, NULL, &sc->rl_parent_tag);
987 	if (error) {
988 		device_printf(dev, "could not allocate parent DMA tag\n");
989 		return (error);
990 	}
991 
992 	/*
993 	 * Allocate map for TX mbufs.
994 	 */
995 	error = bus_dma_tag_create(sc->rl_parent_tag, 1, 0,
996 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL,
997 	    NULL, MCLBYTES * RL_NTXSEGS, RL_NTXSEGS, 4096, 0,
998 	    NULL, NULL, &sc->rl_ldata.rl_tx_mtag);
999 	if (error) {
1000 		device_printf(dev, "could not allocate TX DMA tag\n");
1001 		return (error);
1002 	}
1003 
1004 	/*
1005 	 * Allocate map for RX mbufs.
1006 	 */
1007 
1008 	if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
1009 		error = bus_dma_tag_create(sc->rl_parent_tag, sizeof(uint64_t),
1010 		    0, BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
1011 		    MJUM9BYTES, 1, MJUM9BYTES, 0, NULL, NULL,
1012 		    &sc->rl_ldata.rl_jrx_mtag);
1013 		if (error) {
1014 			device_printf(dev,
1015 			    "could not allocate jumbo RX DMA tag\n");
1016 			return (error);
1017 		}
1018 	}
1019 	error = bus_dma_tag_create(sc->rl_parent_tag, sizeof(uint64_t), 0,
1020 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
1021 	    MCLBYTES, 1, MCLBYTES, 0, NULL, NULL, &sc->rl_ldata.rl_rx_mtag);
1022 	if (error) {
1023 		device_printf(dev, "could not allocate RX DMA tag\n");
1024 		return (error);
1025 	}
1026 
1027 	/*
1028 	 * Allocate map for TX descriptor list.
1029 	 */
1030 	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
1031 	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1032 	    NULL, tx_list_size, 1, tx_list_size, 0,
1033 	    NULL, NULL, &sc->rl_ldata.rl_tx_list_tag);
1034 	if (error) {
1035 		device_printf(dev, "could not allocate TX DMA ring tag\n");
1036 		return (error);
1037 	}
1038 
1039 	/* Allocate DMA'able memory for the TX ring */
1040 
1041 	error = bus_dmamem_alloc(sc->rl_ldata.rl_tx_list_tag,
1042 	    (void **)&sc->rl_ldata.rl_tx_list,
1043 	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1044 	    &sc->rl_ldata.rl_tx_list_map);
1045 	if (error) {
1046 		device_printf(dev, "could not allocate TX DMA ring\n");
1047 		return (error);
1048 	}
1049 
1050 	/* Load the map for the TX ring. */
1051 
1052 	sc->rl_ldata.rl_tx_list_addr = 0;
1053 	error = bus_dmamap_load(sc->rl_ldata.rl_tx_list_tag,
1054 	     sc->rl_ldata.rl_tx_list_map, sc->rl_ldata.rl_tx_list,
1055 	     tx_list_size, re_dma_map_addr,
1056 	     &sc->rl_ldata.rl_tx_list_addr, BUS_DMA_NOWAIT);
1057 	if (error != 0 || sc->rl_ldata.rl_tx_list_addr == 0) {
1058 		device_printf(dev, "could not load TX DMA ring\n");
1059 		return (ENOMEM);
1060 	}
1061 
1062 	/* Create DMA maps for TX buffers */
1063 
1064 	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
1065 		error = bus_dmamap_create(sc->rl_ldata.rl_tx_mtag, 0,
1066 		    &sc->rl_ldata.rl_tx_desc[i].tx_dmamap);
1067 		if (error) {
1068 			device_printf(dev, "could not create DMA map for TX\n");
1069 			return (error);
1070 		}
1071 	}
1072 
1073 	/*
1074 	 * Allocate map for RX descriptor list.
1075 	 */
1076 	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
1077 	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1078 	    NULL, rx_list_size, 1, rx_list_size, 0,
1079 	    NULL, NULL, &sc->rl_ldata.rl_rx_list_tag);
1080 	if (error) {
1081 		device_printf(dev, "could not create RX DMA ring tag\n");
1082 		return (error);
1083 	}
1084 
1085 	/* Allocate DMA'able memory for the RX ring */
1086 
1087 	error = bus_dmamem_alloc(sc->rl_ldata.rl_rx_list_tag,
1088 	    (void **)&sc->rl_ldata.rl_rx_list,
1089 	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1090 	    &sc->rl_ldata.rl_rx_list_map);
1091 	if (error) {
1092 		device_printf(dev, "could not allocate RX DMA ring\n");
1093 		return (error);
1094 	}
1095 
1096 	/* Load the map for the RX ring. */
1097 
1098 	sc->rl_ldata.rl_rx_list_addr = 0;
1099 	error = bus_dmamap_load(sc->rl_ldata.rl_rx_list_tag,
1100 	     sc->rl_ldata.rl_rx_list_map, sc->rl_ldata.rl_rx_list,
1101 	     rx_list_size, re_dma_map_addr,
1102 	     &sc->rl_ldata.rl_rx_list_addr, BUS_DMA_NOWAIT);
1103 	if (error != 0 || sc->rl_ldata.rl_rx_list_addr == 0) {
1104 		device_printf(dev, "could not load RX DMA ring\n");
1105 		return (ENOMEM);
1106 	}
1107 
1108 	/* Create DMA maps for RX buffers */
1109 
1110 	if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
1111 		error = bus_dmamap_create(sc->rl_ldata.rl_jrx_mtag, 0,
1112 		    &sc->rl_ldata.rl_jrx_sparemap);
1113 		if (error) {
1114 			device_printf(dev,
1115 			    "could not create spare DMA map for jumbo RX\n");
1116 			return (error);
1117 		}
1118 		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1119 			error = bus_dmamap_create(sc->rl_ldata.rl_jrx_mtag, 0,
1120 			    &sc->rl_ldata.rl_jrx_desc[i].rx_dmamap);
1121 			if (error) {
1122 				device_printf(dev,
1123 				    "could not create DMA map for jumbo RX\n");
1124 				return (error);
1125 			}
1126 		}
1127 	}
1128 	error = bus_dmamap_create(sc->rl_ldata.rl_rx_mtag, 0,
1129 	    &sc->rl_ldata.rl_rx_sparemap);
1130 	if (error) {
1131 		device_printf(dev, "could not create spare DMA map for RX\n");
1132 		return (error);
1133 	}
1134 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1135 		error = bus_dmamap_create(sc->rl_ldata.rl_rx_mtag, 0,
1136 		    &sc->rl_ldata.rl_rx_desc[i].rx_dmamap);
1137 		if (error) {
1138 			device_printf(dev, "could not create DMA map for RX\n");
1139 			return (error);
1140 		}
1141 	}
1142 
1143 	/* Create DMA map for statistics. */
1144 	error = bus_dma_tag_create(sc->rl_parent_tag, RL_DUMP_ALIGN, 0,
1145 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
1146 	    sizeof(struct rl_stats), 1, sizeof(struct rl_stats), 0, NULL, NULL,
1147 	    &sc->rl_ldata.rl_stag);
1148 	if (error) {
1149 		device_printf(dev, "could not create statistics DMA tag\n");
1150 		return (error);
1151 	}
1152 	/* Allocate DMA'able memory for statistics. */
1153 	error = bus_dmamem_alloc(sc->rl_ldata.rl_stag,
1154 	    (void **)&sc->rl_ldata.rl_stats,
1155 	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1156 	    &sc->rl_ldata.rl_smap);
1157 	if (error) {
1158 		device_printf(dev,
1159 		    "could not allocate statistics DMA memory\n");
1160 		return (error);
1161 	}
1162 	/* Load the map for statistics. */
1163 	sc->rl_ldata.rl_stats_addr = 0;
1164 	error = bus_dmamap_load(sc->rl_ldata.rl_stag, sc->rl_ldata.rl_smap,
1165 	    sc->rl_ldata.rl_stats, sizeof(struct rl_stats), re_dma_map_addr,
1166 	     &sc->rl_ldata.rl_stats_addr, BUS_DMA_NOWAIT);
1167 	if (error != 0 || sc->rl_ldata.rl_stats_addr == 0) {
1168 		device_printf(dev, "could not load statistics DMA memory\n");
1169 		return (ENOMEM);
1170 	}
1171 
1172 	return (0);
1173 }
1174 
1175 /*
1176  * Attach the interface. Allocate softc structures, do ifmedia
1177  * setup and ethernet/BPF attach.
1178  */
1179 static int
1180 re_attach(device_t dev)
1181 {
1182 	u_char			eaddr[ETHER_ADDR_LEN];
1183 	u_int16_t		as[ETHER_ADDR_LEN / 2];
1184 	struct rl_softc		*sc;
1185 	struct ifnet		*ifp;
1186 	const struct rl_hwrev	*hw_rev;
1187 	int			hwrev;
1188 	u_int16_t		devid, re_did = 0;
1189 	int			error = 0, i, phy, rid;
1190 	int			msic, msixc, reg;
1191 	uint8_t			cfg;
1192 
1193 	sc = device_get_softc(dev);
1194 	sc->rl_dev = dev;
1195 
1196 	mtx_init(&sc->rl_mtx, device_get_nameunit(dev), MTX_NETWORK_LOCK,
1197 	    MTX_DEF);
1198 	callout_init_mtx(&sc->rl_stat_callout, &sc->rl_mtx, 0);
1199 
1200 	/*
1201 	 * Map control/status registers.
1202 	 */
1203 	pci_enable_busmaster(dev);
1204 
1205 	devid = pci_get_device(dev);
1206 	/*
1207 	 * Prefer memory space register mapping over IO space.
1208 	 * Because RTL8169SC does not seem to work when memory mapping
1209 	 * is used always activate io mapping.
1210 	 */
1211 	if (devid == RT_DEVICEID_8169SC)
1212 		prefer_iomap = 1;
1213 	if (prefer_iomap == 0) {
1214 		sc->rl_res_id = PCIR_BAR(1);
1215 		sc->rl_res_type = SYS_RES_MEMORY;
1216 		/* RTL8168/8101E seems to use different BARs. */
1217 		if (devid == RT_DEVICEID_8168 || devid == RT_DEVICEID_8101E)
1218 			sc->rl_res_id = PCIR_BAR(2);
1219 	} else {
1220 		sc->rl_res_id = PCIR_BAR(0);
1221 		sc->rl_res_type = SYS_RES_IOPORT;
1222 	}
1223 	sc->rl_res = bus_alloc_resource_any(dev, sc->rl_res_type,
1224 	    &sc->rl_res_id, RF_ACTIVE);
1225 	if (sc->rl_res == NULL && prefer_iomap == 0) {
1226 		sc->rl_res_id = PCIR_BAR(0);
1227 		sc->rl_res_type = SYS_RES_IOPORT;
1228 		sc->rl_res = bus_alloc_resource_any(dev, sc->rl_res_type,
1229 		    &sc->rl_res_id, RF_ACTIVE);
1230 	}
1231 	if (sc->rl_res == NULL) {
1232 		device_printf(dev, "couldn't map ports/memory\n");
1233 		error = ENXIO;
1234 		goto fail;
1235 	}
1236 
1237 	sc->rl_btag = rman_get_bustag(sc->rl_res);
1238 	sc->rl_bhandle = rman_get_bushandle(sc->rl_res);
1239 
1240 	msic = pci_msi_count(dev);
1241 	msixc = pci_msix_count(dev);
1242 	if (pci_find_cap(dev, PCIY_EXPRESS, &reg) == 0)
1243 		sc->rl_flags |= RL_FLAG_PCIE;
1244 	if (bootverbose) {
1245 		device_printf(dev, "MSI count : %d\n", msic);
1246 		device_printf(dev, "MSI-X count : %d\n", msixc);
1247 	}
1248 	if (msix_disable > 0)
1249 		msixc = 0;
1250 	if (msi_disable > 0)
1251 		msic = 0;
1252 	/* Prefer MSI-X to MSI. */
1253 	if (msixc > 0) {
1254 		msixc = 1;
1255 		rid = PCIR_BAR(4);
1256 		sc->rl_res_pba = bus_alloc_resource_any(dev, SYS_RES_MEMORY,
1257 		    &rid, RF_ACTIVE);
1258 		if (sc->rl_res_pba == NULL) {
1259 			device_printf(sc->rl_dev,
1260 			    "could not allocate MSI-X PBA resource\n");
1261 		}
1262 		if (sc->rl_res_pba != NULL &&
1263 		    pci_alloc_msix(dev, &msixc) == 0) {
1264 			if (msixc == 1) {
1265 				device_printf(dev, "Using %d MSI-X message\n",
1266 				    msixc);
1267 				sc->rl_flags |= RL_FLAG_MSIX;
1268 			} else
1269 				pci_release_msi(dev);
1270 		}
1271 		if ((sc->rl_flags & RL_FLAG_MSIX) == 0) {
1272 			if (sc->rl_res_pba != NULL)
1273 				bus_release_resource(dev, SYS_RES_MEMORY, rid,
1274 				    sc->rl_res_pba);
1275 			sc->rl_res_pba = NULL;
1276 			msixc = 0;
1277 		}
1278 	}
1279 	/* Prefer MSI to INTx. */
1280 	if (msixc == 0 && msic > 0) {
1281 		msic = 1;
1282 		if (pci_alloc_msi(dev, &msic) == 0) {
1283 			if (msic == RL_MSI_MESSAGES) {
1284 				device_printf(dev, "Using %d MSI message\n",
1285 				    msic);
1286 				sc->rl_flags |= RL_FLAG_MSI;
1287 				/* Explicitly set MSI enable bit. */
1288 				CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1289 				cfg = CSR_READ_1(sc, RL_CFG2);
1290 				cfg |= RL_CFG2_MSI;
1291 				CSR_WRITE_1(sc, RL_CFG2, cfg);
1292 				CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1293 			} else
1294 				pci_release_msi(dev);
1295 		}
1296 		if ((sc->rl_flags & RL_FLAG_MSI) == 0)
1297 			msic = 0;
1298 	}
1299 
1300 	/* Allocate interrupt */
1301 	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) == 0) {
1302 		rid = 0;
1303 		sc->rl_irq[0] = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
1304 		    RF_SHAREABLE | RF_ACTIVE);
1305 		if (sc->rl_irq[0] == NULL) {
1306 			device_printf(dev, "couldn't allocate IRQ resources\n");
1307 			error = ENXIO;
1308 			goto fail;
1309 		}
1310 	} else {
1311 		for (i = 0, rid = 1; i < RL_MSI_MESSAGES; i++, rid++) {
1312 			sc->rl_irq[i] = bus_alloc_resource_any(dev,
1313 			    SYS_RES_IRQ, &rid, RF_ACTIVE);
1314 			if (sc->rl_irq[i] == NULL) {
1315 				device_printf(dev,
1316 				    "couldn't llocate IRQ resources for "
1317 				    "message %d\n", rid);
1318 				error = ENXIO;
1319 				goto fail;
1320 			}
1321 		}
1322 	}
1323 
1324 	if ((sc->rl_flags & RL_FLAG_MSI) == 0) {
1325 		CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1326 		cfg = CSR_READ_1(sc, RL_CFG2);
1327 		if ((cfg & RL_CFG2_MSI) != 0) {
1328 			device_printf(dev, "turning off MSI enable bit.\n");
1329 			cfg &= ~RL_CFG2_MSI;
1330 			CSR_WRITE_1(sc, RL_CFG2, cfg);
1331 		}
1332 		CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1333 	}
1334 
1335 	hw_rev = re_hwrevs;
1336 	hwrev = CSR_READ_4(sc, RL_TXCFG);
1337 	switch (hwrev & 0x70000000) {
1338 	case 0x00000000:
1339 	case 0x10000000:
1340 		device_printf(dev, "Chip rev. 0x%08x\n", hwrev & 0xfc800000);
1341 		hwrev &= (RL_TXCFG_HWREV | 0x80000000);
1342 		break;
1343 	default:
1344 		device_printf(dev, "Chip rev. 0x%08x\n", hwrev & 0x7c800000);
1345 		hwrev &= RL_TXCFG_HWREV;
1346 		break;
1347 	}
1348 	device_printf(dev, "MAC rev. 0x%08x\n", hwrev & 0x00700000);
1349 	while (hw_rev->rl_desc != NULL) {
1350 		if (hw_rev->rl_rev == hwrev) {
1351 			sc->rl_type = hw_rev->rl_type;
1352 			sc->rl_hwrev = hw_rev;
1353 			break;
1354 		}
1355 		hw_rev++;
1356 	}
1357 	if (hw_rev->rl_desc == NULL) {
1358 		device_printf(dev, "Unknown H/W revision: 0x%08x\n", hwrev);
1359 		error = ENXIO;
1360 		goto fail;
1361 	}
1362 
1363 	switch (hw_rev->rl_rev) {
1364 	case RL_HWREV_8139CPLUS:
1365 		sc->rl_flags |= RL_FLAG_FASTETHER | RL_FLAG_AUTOPAD;
1366 		break;
1367 	case RL_HWREV_8100E:
1368 	case RL_HWREV_8101E:
1369 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_FASTETHER;
1370 		break;
1371 	case RL_HWREV_8102E:
1372 	case RL_HWREV_8102EL:
1373 	case RL_HWREV_8102EL_SPIN1:
1374 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR | RL_FLAG_DESCV2 |
1375 		    RL_FLAG_MACSTAT | RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP |
1376 		    RL_FLAG_AUTOPAD;
1377 		break;
1378 	case RL_HWREV_8103E:
1379 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR | RL_FLAG_DESCV2 |
1380 		    RL_FLAG_MACSTAT | RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP |
1381 		    RL_FLAG_AUTOPAD | RL_FLAG_MACSLEEP;
1382 		break;
1383 	case RL_HWREV_8401E:
1384 	case RL_HWREV_8105E:
1385 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
1386 		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
1387 		    RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD;
1388 		break;
1389 	case RL_HWREV_8168B_SPIN1:
1390 	case RL_HWREV_8168B_SPIN2:
1391 		sc->rl_flags |= RL_FLAG_WOLRXENB;
1392 		/* FALLTHROUGH */
1393 	case RL_HWREV_8168B_SPIN3:
1394 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_MACSTAT;
1395 		break;
1396 	case RL_HWREV_8168C_SPIN2:
1397 		sc->rl_flags |= RL_FLAG_MACSLEEP;
1398 		/* FALLTHROUGH */
1399 	case RL_HWREV_8168C:
1400 		if ((hwrev & 0x00700000) == 0x00200000)
1401 			sc->rl_flags |= RL_FLAG_MACSLEEP;
1402 		/* FALLTHROUGH */
1403 	case RL_HWREV_8168CP:
1404 	case RL_HWREV_8168D:
1405 	case RL_HWREV_8168DP:
1406 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
1407 		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP |
1408 		    RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2;
1409 		break;
1410 	case RL_HWREV_8168E:
1411 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
1412 		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
1413 		    RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2;
1414 		break;
1415 	case RL_HWREV_8168E_VL:
1416 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
1417 		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP |
1418 		    RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2;
1419 		break;
1420 	case RL_HWREV_8169_8110SB:
1421 	case RL_HWREV_8169_8110SBL:
1422 	case RL_HWREV_8169_8110SC:
1423 	case RL_HWREV_8169_8110SCE:
1424 		sc->rl_flags |= RL_FLAG_PHYWAKE;
1425 		/* FALLTHROUGH */
1426 	case RL_HWREV_8169:
1427 	case RL_HWREV_8169S:
1428 	case RL_HWREV_8110S:
1429 		sc->rl_flags |= RL_FLAG_MACRESET;
1430 		break;
1431 	default:
1432 		break;
1433 	}
1434 
1435 	/* Reset the adapter. */
1436 	RL_LOCK(sc);
1437 	re_reset(sc);
1438 	RL_UNLOCK(sc);
1439 
1440 	/* Enable PME. */
1441 	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1442 	cfg = CSR_READ_1(sc, RL_CFG1);
1443 	cfg |= RL_CFG1_PME;
1444 	CSR_WRITE_1(sc, RL_CFG1, cfg);
1445 	cfg = CSR_READ_1(sc, RL_CFG5);
1446 	cfg &= RL_CFG5_PME_STS;
1447 	CSR_WRITE_1(sc, RL_CFG5, cfg);
1448 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1449 
1450 	if ((sc->rl_flags & RL_FLAG_PAR) != 0) {
1451 		/*
1452 		 * XXX Should have a better way to extract station
1453 		 * address from EEPROM.
1454 		 */
1455 		for (i = 0; i < ETHER_ADDR_LEN; i++)
1456 			eaddr[i] = CSR_READ_1(sc, RL_IDR0 + i);
1457 	} else {
1458 		sc->rl_eewidth = RL_9356_ADDR_LEN;
1459 		re_read_eeprom(sc, (caddr_t)&re_did, 0, 1);
1460 		if (re_did != 0x8129)
1461 			sc->rl_eewidth = RL_9346_ADDR_LEN;
1462 
1463 		/*
1464 		 * Get station address from the EEPROM.
1465 		 */
1466 		re_read_eeprom(sc, (caddr_t)as, RL_EE_EADDR, 3);
1467 		for (i = 0; i < ETHER_ADDR_LEN / 2; i++)
1468 			as[i] = le16toh(as[i]);
1469 		bcopy(as, eaddr, sizeof(eaddr));
1470 	}
1471 
1472 	if (sc->rl_type == RL_8169) {
1473 		/* Set RX length mask and number of descriptors. */
1474 		sc->rl_rxlenmask = RL_RDESC_STAT_GFRAGLEN;
1475 		sc->rl_txstart = RL_GTXSTART;
1476 		sc->rl_ldata.rl_tx_desc_cnt = RL_8169_TX_DESC_CNT;
1477 		sc->rl_ldata.rl_rx_desc_cnt = RL_8169_RX_DESC_CNT;
1478 	} else {
1479 		/* Set RX length mask and number of descriptors. */
1480 		sc->rl_rxlenmask = RL_RDESC_STAT_FRAGLEN;
1481 		sc->rl_txstart = RL_TXSTART;
1482 		sc->rl_ldata.rl_tx_desc_cnt = RL_8139_TX_DESC_CNT;
1483 		sc->rl_ldata.rl_rx_desc_cnt = RL_8139_RX_DESC_CNT;
1484 	}
1485 
1486 	error = re_allocmem(dev, sc);
1487 	if (error)
1488 		goto fail;
1489 	re_add_sysctls(sc);
1490 
1491 	ifp = sc->rl_ifp = if_alloc(IFT_ETHER);
1492 	if (ifp == NULL) {
1493 		device_printf(dev, "can not if_alloc()\n");
1494 		error = ENOSPC;
1495 		goto fail;
1496 	}
1497 
1498 	/* Take controller out of deep sleep mode. */
1499 	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
1500 		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
1501 			CSR_WRITE_1(sc, RL_GPIO,
1502 			    CSR_READ_1(sc, RL_GPIO) | 0x01);
1503 		else
1504 			CSR_WRITE_1(sc, RL_GPIO,
1505 			    CSR_READ_1(sc, RL_GPIO) & ~0x01);
1506 	}
1507 
1508 	/* Take PHY out of power down mode. */
1509 	if ((sc->rl_flags & RL_FLAG_PHYWAKE_PM) != 0) {
1510 		CSR_WRITE_1(sc, RL_PMCH, CSR_READ_1(sc, RL_PMCH) | 0x80);
1511 		if (hw_rev->rl_rev == RL_HWREV_8401E)
1512 			CSR_WRITE_1(sc, 0xD1, CSR_READ_1(sc, 0xD1) & ~0x08);
1513 	}
1514 	if ((sc->rl_flags & RL_FLAG_PHYWAKE) != 0) {
1515 		re_gmii_writereg(dev, 1, 0x1f, 0);
1516 		re_gmii_writereg(dev, 1, 0x0e, 0);
1517 	}
1518 
1519 #define	RE_PHYAD_INTERNAL	 0
1520 
1521 	/* Do MII setup. */
1522 	phy = RE_PHYAD_INTERNAL;
1523 	if (sc->rl_type == RL_8169)
1524 		phy = 1;
1525 	error = mii_attach(dev, &sc->rl_miibus, ifp, re_ifmedia_upd,
1526 	    re_ifmedia_sts, BMSR_DEFCAPMASK, phy, MII_OFFSET_ANY, MIIF_DOPAUSE);
1527 	if (error != 0) {
1528 		device_printf(dev, "attaching PHYs failed\n");
1529 		goto fail;
1530 	}
1531 
1532 	ifp->if_softc = sc;
1533 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1534 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1535 	ifp->if_ioctl = re_ioctl;
1536 	ifp->if_start = re_start;
1537 	/*
1538 	 * RTL8168/8111C generates wrong IP checksummed frame if the
1539 	 * packet has IP options so disable TX IP checksum offloading.
1540 	 */
1541 	if (sc->rl_hwrev->rl_rev == RL_HWREV_8168C ||
1542 	    sc->rl_hwrev->rl_rev == RL_HWREV_8168C_SPIN2)
1543 		ifp->if_hwassist = CSUM_TCP | CSUM_UDP;
1544 	else
1545 		ifp->if_hwassist = CSUM_IP | CSUM_TCP | CSUM_UDP;
1546 	ifp->if_hwassist |= CSUM_TSO;
1547 	ifp->if_capabilities = IFCAP_HWCSUM | IFCAP_TSO4;
1548 	ifp->if_capenable = ifp->if_capabilities;
1549 	ifp->if_init = re_init;
1550 	IFQ_SET_MAXLEN(&ifp->if_snd, RL_IFQ_MAXLEN);
1551 	ifp->if_snd.ifq_drv_maxlen = RL_IFQ_MAXLEN;
1552 	IFQ_SET_READY(&ifp->if_snd);
1553 
1554 	TASK_INIT(&sc->rl_inttask, 0, re_int_task, sc);
1555 
1556 	/*
1557 	 * Call MI attach routine.
1558 	 */
1559 	ether_ifattach(ifp, eaddr);
1560 
1561 	/* VLAN capability setup */
1562 	ifp->if_capabilities |= IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING;
1563 	if (ifp->if_capabilities & IFCAP_HWCSUM)
1564 		ifp->if_capabilities |= IFCAP_VLAN_HWCSUM;
1565 	/* Enable WOL if PM is supported. */
1566 	if (pci_find_cap(sc->rl_dev, PCIY_PMG, &reg) == 0)
1567 		ifp->if_capabilities |= IFCAP_WOL;
1568 	ifp->if_capenable = ifp->if_capabilities;
1569 	/*
1570 	 * Don't enable TSO by default.  It is known to generate
1571 	 * corrupted TCP segments(bad TCP options) under certain
1572 	 * circumtances.
1573 	 */
1574 	ifp->if_hwassist &= ~CSUM_TSO;
1575 	ifp->if_capenable &= ~(IFCAP_TSO4 | IFCAP_VLAN_HWTSO);
1576 #ifdef DEVICE_POLLING
1577 	ifp->if_capabilities |= IFCAP_POLLING;
1578 #endif
1579 	/*
1580 	 * Tell the upper layer(s) we support long frames.
1581 	 * Must appear after the call to ether_ifattach() because
1582 	 * ether_ifattach() sets ifi_hdrlen to the default value.
1583 	 */
1584 	ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
1585 
1586 #ifdef RE_DIAG
1587 	/*
1588 	 * Perform hardware diagnostic on the original RTL8169.
1589 	 * Some 32-bit cards were incorrectly wired and would
1590 	 * malfunction if plugged into a 64-bit slot.
1591 	 */
1592 
1593 	if (hwrev == RL_HWREV_8169) {
1594 		error = re_diag(sc);
1595 		if (error) {
1596 			device_printf(dev,
1597 		    	"attach aborted due to hardware diag failure\n");
1598 			ether_ifdetach(ifp);
1599 			goto fail;
1600 		}
1601 	}
1602 #endif
1603 
1604 #ifdef RE_TX_MODERATION
1605 	intr_filter = 1;
1606 #endif
1607 	/* Hook interrupt last to avoid having to lock softc */
1608 	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) != 0 &&
1609 	    intr_filter == 0) {
1610 		error = bus_setup_intr(dev, sc->rl_irq[0],
1611 		    INTR_TYPE_NET | INTR_MPSAFE, NULL, re_intr_msi, sc,
1612 		    &sc->rl_intrhand[0]);
1613 	} else {
1614 		error = bus_setup_intr(dev, sc->rl_irq[0],
1615 		    INTR_TYPE_NET | INTR_MPSAFE, re_intr, NULL, sc,
1616 		    &sc->rl_intrhand[0]);
1617 	}
1618 	if (error) {
1619 		device_printf(dev, "couldn't set up irq\n");
1620 		ether_ifdetach(ifp);
1621 	}
1622 
1623 fail:
1624 
1625 	if (error)
1626 		re_detach(dev);
1627 
1628 	return (error);
1629 }
1630 
1631 /*
1632  * Shutdown hardware and free up resources. This can be called any
1633  * time after the mutex has been initialized. It is called in both
1634  * the error case in attach and the normal detach case so it needs
1635  * to be careful about only freeing resources that have actually been
1636  * allocated.
1637  */
1638 static int
1639 re_detach(device_t dev)
1640 {
1641 	struct rl_softc		*sc;
1642 	struct ifnet		*ifp;
1643 	int			i, rid;
1644 
1645 	sc = device_get_softc(dev);
1646 	ifp = sc->rl_ifp;
1647 	KASSERT(mtx_initialized(&sc->rl_mtx), ("re mutex not initialized"));
1648 
1649 	/* These should only be active if attach succeeded */
1650 	if (device_is_attached(dev)) {
1651 #ifdef DEVICE_POLLING
1652 		if (ifp->if_capenable & IFCAP_POLLING)
1653 			ether_poll_deregister(ifp);
1654 #endif
1655 		RL_LOCK(sc);
1656 #if 0
1657 		sc->suspended = 1;
1658 #endif
1659 		re_stop(sc);
1660 		RL_UNLOCK(sc);
1661 		callout_drain(&sc->rl_stat_callout);
1662 		taskqueue_drain(taskqueue_fast, &sc->rl_inttask);
1663 		/*
1664 		 * Force off the IFF_UP flag here, in case someone
1665 		 * still had a BPF descriptor attached to this
1666 		 * interface. If they do, ether_ifdetach() will cause
1667 		 * the BPF code to try and clear the promisc mode
1668 		 * flag, which will bubble down to re_ioctl(),
1669 		 * which will try to call re_init() again. This will
1670 		 * turn the NIC back on and restart the MII ticker,
1671 		 * which will panic the system when the kernel tries
1672 		 * to invoke the re_tick() function that isn't there
1673 		 * anymore.
1674 		 */
1675 		ifp->if_flags &= ~IFF_UP;
1676 		ether_ifdetach(ifp);
1677 	}
1678 	if (sc->rl_miibus)
1679 		device_delete_child(dev, sc->rl_miibus);
1680 	bus_generic_detach(dev);
1681 
1682 	/*
1683 	 * The rest is resource deallocation, so we should already be
1684 	 * stopped here.
1685 	 */
1686 
1687 	if (sc->rl_intrhand[0] != NULL) {
1688 		bus_teardown_intr(dev, sc->rl_irq[0], sc->rl_intrhand[0]);
1689 		sc->rl_intrhand[0] = NULL;
1690 	}
1691 	if (ifp != NULL)
1692 		if_free(ifp);
1693 	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) == 0)
1694 		rid = 0;
1695 	else
1696 		rid = 1;
1697 	if (sc->rl_irq[0] != NULL) {
1698 		bus_release_resource(dev, SYS_RES_IRQ, rid, sc->rl_irq[0]);
1699 		sc->rl_irq[0] = NULL;
1700 	}
1701 	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) != 0)
1702 		pci_release_msi(dev);
1703 	if (sc->rl_res_pba) {
1704 		rid = PCIR_BAR(4);
1705 		bus_release_resource(dev, SYS_RES_MEMORY, rid, sc->rl_res_pba);
1706 	}
1707 	if (sc->rl_res)
1708 		bus_release_resource(dev, sc->rl_res_type, sc->rl_res_id,
1709 		    sc->rl_res);
1710 
1711 	/* Unload and free the RX DMA ring memory and map */
1712 
1713 	if (sc->rl_ldata.rl_rx_list_tag) {
1714 		if (sc->rl_ldata.rl_rx_list_map)
1715 			bus_dmamap_unload(sc->rl_ldata.rl_rx_list_tag,
1716 			    sc->rl_ldata.rl_rx_list_map);
1717 		if (sc->rl_ldata.rl_rx_list_map && sc->rl_ldata.rl_rx_list)
1718 			bus_dmamem_free(sc->rl_ldata.rl_rx_list_tag,
1719 			    sc->rl_ldata.rl_rx_list,
1720 			    sc->rl_ldata.rl_rx_list_map);
1721 		bus_dma_tag_destroy(sc->rl_ldata.rl_rx_list_tag);
1722 	}
1723 
1724 	/* Unload and free the TX DMA ring memory and map */
1725 
1726 	if (sc->rl_ldata.rl_tx_list_tag) {
1727 		if (sc->rl_ldata.rl_tx_list_map)
1728 			bus_dmamap_unload(sc->rl_ldata.rl_tx_list_tag,
1729 			    sc->rl_ldata.rl_tx_list_map);
1730 		if (sc->rl_ldata.rl_tx_list_map && sc->rl_ldata.rl_tx_list)
1731 			bus_dmamem_free(sc->rl_ldata.rl_tx_list_tag,
1732 			    sc->rl_ldata.rl_tx_list,
1733 			    sc->rl_ldata.rl_tx_list_map);
1734 		bus_dma_tag_destroy(sc->rl_ldata.rl_tx_list_tag);
1735 	}
1736 
1737 	/* Destroy all the RX and TX buffer maps */
1738 
1739 	if (sc->rl_ldata.rl_tx_mtag) {
1740 		for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
1741 			if (sc->rl_ldata.rl_tx_desc[i].tx_dmamap)
1742 				bus_dmamap_destroy(sc->rl_ldata.rl_tx_mtag,
1743 				    sc->rl_ldata.rl_tx_desc[i].tx_dmamap);
1744 		}
1745 		bus_dma_tag_destroy(sc->rl_ldata.rl_tx_mtag);
1746 	}
1747 	if (sc->rl_ldata.rl_rx_mtag) {
1748 		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1749 			if (sc->rl_ldata.rl_rx_desc[i].rx_dmamap)
1750 				bus_dmamap_destroy(sc->rl_ldata.rl_rx_mtag,
1751 				    sc->rl_ldata.rl_rx_desc[i].rx_dmamap);
1752 		}
1753 		if (sc->rl_ldata.rl_rx_sparemap)
1754 			bus_dmamap_destroy(sc->rl_ldata.rl_rx_mtag,
1755 			    sc->rl_ldata.rl_rx_sparemap);
1756 		bus_dma_tag_destroy(sc->rl_ldata.rl_rx_mtag);
1757 	}
1758 	if (sc->rl_ldata.rl_jrx_mtag) {
1759 		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1760 			if (sc->rl_ldata.rl_jrx_desc[i].rx_dmamap)
1761 				bus_dmamap_destroy(sc->rl_ldata.rl_jrx_mtag,
1762 				    sc->rl_ldata.rl_jrx_desc[i].rx_dmamap);
1763 		}
1764 		if (sc->rl_ldata.rl_jrx_sparemap)
1765 			bus_dmamap_destroy(sc->rl_ldata.rl_jrx_mtag,
1766 			    sc->rl_ldata.rl_jrx_sparemap);
1767 		bus_dma_tag_destroy(sc->rl_ldata.rl_jrx_mtag);
1768 	}
1769 	/* Unload and free the stats buffer and map */
1770 
1771 	if (sc->rl_ldata.rl_stag) {
1772 		if (sc->rl_ldata.rl_smap)
1773 			bus_dmamap_unload(sc->rl_ldata.rl_stag,
1774 			    sc->rl_ldata.rl_smap);
1775 		if (sc->rl_ldata.rl_smap && sc->rl_ldata.rl_stats)
1776 			bus_dmamem_free(sc->rl_ldata.rl_stag,
1777 			    sc->rl_ldata.rl_stats, sc->rl_ldata.rl_smap);
1778 		bus_dma_tag_destroy(sc->rl_ldata.rl_stag);
1779 	}
1780 
1781 	if (sc->rl_parent_tag)
1782 		bus_dma_tag_destroy(sc->rl_parent_tag);
1783 
1784 	mtx_destroy(&sc->rl_mtx);
1785 
1786 	return (0);
1787 }
1788 
1789 static __inline void
1790 re_discard_rxbuf(struct rl_softc *sc, int idx)
1791 {
1792 	struct rl_desc		*desc;
1793 	struct rl_rxdesc	*rxd;
1794 	uint32_t		cmdstat;
1795 
1796 	if (sc->rl_ifp->if_mtu > RL_MTU &&
1797 	    (sc->rl_flags & RL_FLAG_JUMBOV2) != 0)
1798 		rxd = &sc->rl_ldata.rl_jrx_desc[idx];
1799 	else
1800 		rxd = &sc->rl_ldata.rl_rx_desc[idx];
1801 	desc = &sc->rl_ldata.rl_rx_list[idx];
1802 	desc->rl_vlanctl = 0;
1803 	cmdstat = rxd->rx_size;
1804 	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
1805 		cmdstat |= RL_RDESC_CMD_EOR;
1806 	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
1807 }
1808 
1809 static int
1810 re_newbuf(struct rl_softc *sc, int idx)
1811 {
1812 	struct mbuf		*m;
1813 	struct rl_rxdesc	*rxd;
1814 	bus_dma_segment_t	segs[1];
1815 	bus_dmamap_t		map;
1816 	struct rl_desc		*desc;
1817 	uint32_t		cmdstat;
1818 	int			error, nsegs;
1819 
1820 	m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
1821 	if (m == NULL)
1822 		return (ENOBUFS);
1823 
1824 	m->m_len = m->m_pkthdr.len = MCLBYTES;
1825 #ifdef RE_FIXUP_RX
1826 	/*
1827 	 * This is part of an evil trick to deal with non-x86 platforms.
1828 	 * The RealTek chip requires RX buffers to be aligned on 64-bit
1829 	 * boundaries, but that will hose non-x86 machines. To get around
1830 	 * this, we leave some empty space at the start of each buffer
1831 	 * and for non-x86 hosts, we copy the buffer back six bytes
1832 	 * to achieve word alignment. This is slightly more efficient
1833 	 * than allocating a new buffer, copying the contents, and
1834 	 * discarding the old buffer.
1835 	 */
1836 	m_adj(m, RE_ETHER_ALIGN);
1837 #endif
1838 	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_rx_mtag,
1839 	    sc->rl_ldata.rl_rx_sparemap, m, segs, &nsegs, BUS_DMA_NOWAIT);
1840 	if (error != 0) {
1841 		m_freem(m);
1842 		return (ENOBUFS);
1843 	}
1844 	KASSERT(nsegs == 1, ("%s: %d segment returned!", __func__, nsegs));
1845 
1846 	rxd = &sc->rl_ldata.rl_rx_desc[idx];
1847 	if (rxd->rx_m != NULL) {
1848 		bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap,
1849 		    BUS_DMASYNC_POSTREAD);
1850 		bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap);
1851 	}
1852 
1853 	rxd->rx_m = m;
1854 	map = rxd->rx_dmamap;
1855 	rxd->rx_dmamap = sc->rl_ldata.rl_rx_sparemap;
1856 	rxd->rx_size = segs[0].ds_len;
1857 	sc->rl_ldata.rl_rx_sparemap = map;
1858 	bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap,
1859 	    BUS_DMASYNC_PREREAD);
1860 
1861 	desc = &sc->rl_ldata.rl_rx_list[idx];
1862 	desc->rl_vlanctl = 0;
1863 	desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[0].ds_addr));
1864 	desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[0].ds_addr));
1865 	cmdstat = segs[0].ds_len;
1866 	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
1867 		cmdstat |= RL_RDESC_CMD_EOR;
1868 	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
1869 
1870 	return (0);
1871 }
1872 
1873 static int
1874 re_jumbo_newbuf(struct rl_softc *sc, int idx)
1875 {
1876 	struct mbuf		*m;
1877 	struct rl_rxdesc	*rxd;
1878 	bus_dma_segment_t	segs[1];
1879 	bus_dmamap_t		map;
1880 	struct rl_desc		*desc;
1881 	uint32_t		cmdstat;
1882 	int			error, nsegs;
1883 
1884 	m = m_getjcl(M_DONTWAIT, MT_DATA, M_PKTHDR, MJUM9BYTES);
1885 	if (m == NULL)
1886 		return (ENOBUFS);
1887 	m->m_len = m->m_pkthdr.len = MJUM9BYTES;
1888 #ifdef RE_FIXUP_RX
1889 	m_adj(m, RE_ETHER_ALIGN);
1890 #endif
1891 	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_jrx_mtag,
1892 	    sc->rl_ldata.rl_jrx_sparemap, m, segs, &nsegs, BUS_DMA_NOWAIT);
1893 	if (error != 0) {
1894 		m_freem(m);
1895 		return (ENOBUFS);
1896 	}
1897 	KASSERT(nsegs == 1, ("%s: %d segment returned!", __func__, nsegs));
1898 
1899 	rxd = &sc->rl_ldata.rl_jrx_desc[idx];
1900 	if (rxd->rx_m != NULL) {
1901 		bus_dmamap_sync(sc->rl_ldata.rl_jrx_mtag, rxd->rx_dmamap,
1902 		    BUS_DMASYNC_POSTREAD);
1903 		bus_dmamap_unload(sc->rl_ldata.rl_jrx_mtag, rxd->rx_dmamap);
1904 	}
1905 
1906 	rxd->rx_m = m;
1907 	map = rxd->rx_dmamap;
1908 	rxd->rx_dmamap = sc->rl_ldata.rl_jrx_sparemap;
1909 	rxd->rx_size = segs[0].ds_len;
1910 	sc->rl_ldata.rl_jrx_sparemap = map;
1911 	bus_dmamap_sync(sc->rl_ldata.rl_jrx_mtag, rxd->rx_dmamap,
1912 	    BUS_DMASYNC_PREREAD);
1913 
1914 	desc = &sc->rl_ldata.rl_rx_list[idx];
1915 	desc->rl_vlanctl = 0;
1916 	desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[0].ds_addr));
1917 	desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[0].ds_addr));
1918 	cmdstat = segs[0].ds_len;
1919 	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
1920 		cmdstat |= RL_RDESC_CMD_EOR;
1921 	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
1922 
1923 	return (0);
1924 }
1925 
1926 #ifdef RE_FIXUP_RX
1927 static __inline void
1928 re_fixup_rx(struct mbuf *m)
1929 {
1930 	int                     i;
1931 	uint16_t                *src, *dst;
1932 
1933 	src = mtod(m, uint16_t *);
1934 	dst = src - (RE_ETHER_ALIGN - ETHER_ALIGN) / sizeof *src;
1935 
1936 	for (i = 0; i < (m->m_len / sizeof(uint16_t) + 1); i++)
1937 		*dst++ = *src++;
1938 
1939 	m->m_data -= RE_ETHER_ALIGN - ETHER_ALIGN;
1940 }
1941 #endif
1942 
1943 static int
1944 re_tx_list_init(struct rl_softc *sc)
1945 {
1946 	struct rl_desc		*desc;
1947 	int			i;
1948 
1949 	RL_LOCK_ASSERT(sc);
1950 
1951 	bzero(sc->rl_ldata.rl_tx_list,
1952 	    sc->rl_ldata.rl_tx_desc_cnt * sizeof(struct rl_desc));
1953 	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++)
1954 		sc->rl_ldata.rl_tx_desc[i].tx_m = NULL;
1955 	/* Set EOR. */
1956 	desc = &sc->rl_ldata.rl_tx_list[sc->rl_ldata.rl_tx_desc_cnt - 1];
1957 	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_EOR);
1958 
1959 	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
1960 	    sc->rl_ldata.rl_tx_list_map,
1961 	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1962 
1963 	sc->rl_ldata.rl_tx_prodidx = 0;
1964 	sc->rl_ldata.rl_tx_considx = 0;
1965 	sc->rl_ldata.rl_tx_free = sc->rl_ldata.rl_tx_desc_cnt;
1966 
1967 	return (0);
1968 }
1969 
1970 static int
1971 re_rx_list_init(struct rl_softc *sc)
1972 {
1973 	int			error, i;
1974 
1975 	bzero(sc->rl_ldata.rl_rx_list,
1976 	    sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc));
1977 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1978 		sc->rl_ldata.rl_rx_desc[i].rx_m = NULL;
1979 		if ((error = re_newbuf(sc, i)) != 0)
1980 			return (error);
1981 	}
1982 
1983 	/* Flush the RX descriptors */
1984 
1985 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1986 	    sc->rl_ldata.rl_rx_list_map,
1987 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
1988 
1989 	sc->rl_ldata.rl_rx_prodidx = 0;
1990 	sc->rl_head = sc->rl_tail = NULL;
1991 	sc->rl_int_rx_act = 0;
1992 
1993 	return (0);
1994 }
1995 
1996 static int
1997 re_jrx_list_init(struct rl_softc *sc)
1998 {
1999 	int			error, i;
2000 
2001 	bzero(sc->rl_ldata.rl_rx_list,
2002 	    sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc));
2003 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
2004 		sc->rl_ldata.rl_jrx_desc[i].rx_m = NULL;
2005 		if ((error = re_jumbo_newbuf(sc, i)) != 0)
2006 			return (error);
2007 	}
2008 
2009 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
2010 	    sc->rl_ldata.rl_rx_list_map,
2011 	    BUS_DMASYNC_PREWRITE | BUS_DMASYNC_PREREAD);
2012 
2013 	sc->rl_ldata.rl_rx_prodidx = 0;
2014 	sc->rl_head = sc->rl_tail = NULL;
2015 	sc->rl_int_rx_act = 0;
2016 
2017 	return (0);
2018 }
2019 
2020 /*
2021  * RX handler for C+ and 8169. For the gigE chips, we support
2022  * the reception of jumbo frames that have been fragmented
2023  * across multiple 2K mbuf cluster buffers.
2024  */
2025 static int
2026 re_rxeof(struct rl_softc *sc, int *rx_npktsp)
2027 {
2028 	struct mbuf		*m;
2029 	struct ifnet		*ifp;
2030 	int			i, rxerr, total_len;
2031 	struct rl_desc		*cur_rx;
2032 	u_int32_t		rxstat, rxvlan;
2033 	int			jumbo, maxpkt = 16, rx_npkts = 0;
2034 
2035 	RL_LOCK_ASSERT(sc);
2036 
2037 	ifp = sc->rl_ifp;
2038 	if (ifp->if_mtu > RL_MTU && (sc->rl_flags & RL_FLAG_JUMBOV2) != 0)
2039 		jumbo = 1;
2040 	else
2041 		jumbo = 0;
2042 
2043 	/* Invalidate the descriptor memory */
2044 
2045 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
2046 	    sc->rl_ldata.rl_rx_list_map,
2047 	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2048 
2049 	for (i = sc->rl_ldata.rl_rx_prodidx; maxpkt > 0;
2050 	    i = RL_RX_DESC_NXT(sc, i)) {
2051 		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
2052 			break;
2053 		cur_rx = &sc->rl_ldata.rl_rx_list[i];
2054 		rxstat = le32toh(cur_rx->rl_cmdstat);
2055 		if ((rxstat & RL_RDESC_STAT_OWN) != 0)
2056 			break;
2057 		total_len = rxstat & sc->rl_rxlenmask;
2058 		rxvlan = le32toh(cur_rx->rl_vlanctl);
2059 		if (jumbo != 0)
2060 			m = sc->rl_ldata.rl_jrx_desc[i].rx_m;
2061 		else
2062 			m = sc->rl_ldata.rl_rx_desc[i].rx_m;
2063 
2064 		if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0 &&
2065 		    (rxstat & (RL_RDESC_STAT_SOF | RL_RDESC_STAT_EOF)) !=
2066 		    (RL_RDESC_STAT_SOF | RL_RDESC_STAT_EOF)) {
2067 			/*
2068 			 * RTL8168C or later controllers do not
2069 			 * support multi-fragment packet.
2070 			 */
2071 			re_discard_rxbuf(sc, i);
2072 			continue;
2073 		} else if ((rxstat & RL_RDESC_STAT_EOF) == 0) {
2074 			if (re_newbuf(sc, i) != 0) {
2075 				/*
2076 				 * If this is part of a multi-fragment packet,
2077 				 * discard all the pieces.
2078 				 */
2079 				if (sc->rl_head != NULL) {
2080 					m_freem(sc->rl_head);
2081 					sc->rl_head = sc->rl_tail = NULL;
2082 				}
2083 				re_discard_rxbuf(sc, i);
2084 				continue;
2085 			}
2086 			m->m_len = RE_RX_DESC_BUFLEN;
2087 			if (sc->rl_head == NULL)
2088 				sc->rl_head = sc->rl_tail = m;
2089 			else {
2090 				m->m_flags &= ~M_PKTHDR;
2091 				sc->rl_tail->m_next = m;
2092 				sc->rl_tail = m;
2093 			}
2094 			continue;
2095 		}
2096 
2097 		/*
2098 		 * NOTE: for the 8139C+, the frame length field
2099 		 * is always 12 bits in size, but for the gigE chips,
2100 		 * it is 13 bits (since the max RX frame length is 16K).
2101 		 * Unfortunately, all 32 bits in the status word
2102 		 * were already used, so to make room for the extra
2103 		 * length bit, RealTek took out the 'frame alignment
2104 		 * error' bit and shifted the other status bits
2105 		 * over one slot. The OWN, EOR, FS and LS bits are
2106 		 * still in the same places. We have already extracted
2107 		 * the frame length and checked the OWN bit, so rather
2108 		 * than using an alternate bit mapping, we shift the
2109 		 * status bits one space to the right so we can evaluate
2110 		 * them using the 8169 status as though it was in the
2111 		 * same format as that of the 8139C+.
2112 		 */
2113 		if (sc->rl_type == RL_8169)
2114 			rxstat >>= 1;
2115 
2116 		/*
2117 		 * if total_len > 2^13-1, both _RXERRSUM and _GIANT will be
2118 		 * set, but if CRC is clear, it will still be a valid frame.
2119 		 */
2120 		if ((rxstat & RL_RDESC_STAT_RXERRSUM) != 0) {
2121 			rxerr = 1;
2122 			if ((sc->rl_flags & RL_FLAG_JUMBOV2) == 0 &&
2123 			    total_len > 8191 &&
2124 			    (rxstat & RL_RDESC_STAT_ERRS) == RL_RDESC_STAT_GIANT)
2125 				rxerr = 0;
2126 			if (rxerr != 0) {
2127 				ifp->if_ierrors++;
2128 				/*
2129 				 * If this is part of a multi-fragment packet,
2130 				 * discard all the pieces.
2131 				 */
2132 				if (sc->rl_head != NULL) {
2133 					m_freem(sc->rl_head);
2134 					sc->rl_head = sc->rl_tail = NULL;
2135 				}
2136 				re_discard_rxbuf(sc, i);
2137 				continue;
2138 			}
2139 		}
2140 
2141 		/*
2142 		 * If allocating a replacement mbuf fails,
2143 		 * reload the current one.
2144 		 */
2145 		if (jumbo != 0)
2146 			rxerr = re_jumbo_newbuf(sc, i);
2147 		else
2148 			rxerr = re_newbuf(sc, i);
2149 		if (rxerr != 0) {
2150 			ifp->if_iqdrops++;
2151 			if (sc->rl_head != NULL) {
2152 				m_freem(sc->rl_head);
2153 				sc->rl_head = sc->rl_tail = NULL;
2154 			}
2155 			re_discard_rxbuf(sc, i);
2156 			continue;
2157 		}
2158 
2159 		if (sc->rl_head != NULL) {
2160 			if (jumbo != 0)
2161 				m->m_len = total_len;
2162 			else {
2163 				m->m_len = total_len % RE_RX_DESC_BUFLEN;
2164 				if (m->m_len == 0)
2165 					m->m_len = RE_RX_DESC_BUFLEN;
2166 			}
2167 			/*
2168 			 * Special case: if there's 4 bytes or less
2169 			 * in this buffer, the mbuf can be discarded:
2170 			 * the last 4 bytes is the CRC, which we don't
2171 			 * care about anyway.
2172 			 */
2173 			if (m->m_len <= ETHER_CRC_LEN) {
2174 				sc->rl_tail->m_len -=
2175 				    (ETHER_CRC_LEN - m->m_len);
2176 				m_freem(m);
2177 			} else {
2178 				m->m_len -= ETHER_CRC_LEN;
2179 				m->m_flags &= ~M_PKTHDR;
2180 				sc->rl_tail->m_next = m;
2181 			}
2182 			m = sc->rl_head;
2183 			sc->rl_head = sc->rl_tail = NULL;
2184 			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
2185 		} else
2186 			m->m_pkthdr.len = m->m_len =
2187 			    (total_len - ETHER_CRC_LEN);
2188 
2189 #ifdef RE_FIXUP_RX
2190 		re_fixup_rx(m);
2191 #endif
2192 		ifp->if_ipackets++;
2193 		m->m_pkthdr.rcvif = ifp;
2194 
2195 		/* Do RX checksumming if enabled */
2196 
2197 		if (ifp->if_capenable & IFCAP_RXCSUM) {
2198 			if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
2199 				/* Check IP header checksum */
2200 				if (rxstat & RL_RDESC_STAT_PROTOID)
2201 					m->m_pkthdr.csum_flags |=
2202 					    CSUM_IP_CHECKED;
2203 				if (!(rxstat & RL_RDESC_STAT_IPSUMBAD))
2204 					m->m_pkthdr.csum_flags |=
2205 					    CSUM_IP_VALID;
2206 
2207 				/* Check TCP/UDP checksum */
2208 				if ((RL_TCPPKT(rxstat) &&
2209 				    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
2210 				    (RL_UDPPKT(rxstat) &&
2211 				     !(rxstat & RL_RDESC_STAT_UDPSUMBAD))) {
2212 					m->m_pkthdr.csum_flags |=
2213 						CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
2214 					m->m_pkthdr.csum_data = 0xffff;
2215 				}
2216 			} else {
2217 				/*
2218 				 * RTL8168C/RTL816CP/RTL8111C/RTL8111CP
2219 				 */
2220 				if ((rxstat & RL_RDESC_STAT_PROTOID) &&
2221 				    (rxvlan & RL_RDESC_IPV4))
2222 					m->m_pkthdr.csum_flags |=
2223 					    CSUM_IP_CHECKED;
2224 				if (!(rxstat & RL_RDESC_STAT_IPSUMBAD) &&
2225 				    (rxvlan & RL_RDESC_IPV4))
2226 					m->m_pkthdr.csum_flags |=
2227 					    CSUM_IP_VALID;
2228 				if (((rxstat & RL_RDESC_STAT_TCP) &&
2229 				    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
2230 				    ((rxstat & RL_RDESC_STAT_UDP) &&
2231 				    !(rxstat & RL_RDESC_STAT_UDPSUMBAD))) {
2232 					m->m_pkthdr.csum_flags |=
2233 						CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
2234 					m->m_pkthdr.csum_data = 0xffff;
2235 				}
2236 			}
2237 		}
2238 		maxpkt--;
2239 		if (rxvlan & RL_RDESC_VLANCTL_TAG) {
2240 			m->m_pkthdr.ether_vtag =
2241 			    bswap16((rxvlan & RL_RDESC_VLANCTL_DATA));
2242 			m->m_flags |= M_VLANTAG;
2243 		}
2244 		RL_UNLOCK(sc);
2245 		(*ifp->if_input)(ifp, m);
2246 		RL_LOCK(sc);
2247 		rx_npkts++;
2248 	}
2249 
2250 	/* Flush the RX DMA ring */
2251 
2252 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
2253 	    sc->rl_ldata.rl_rx_list_map,
2254 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2255 
2256 	sc->rl_ldata.rl_rx_prodidx = i;
2257 
2258 	if (rx_npktsp != NULL)
2259 		*rx_npktsp = rx_npkts;
2260 	if (maxpkt)
2261 		return (EAGAIN);
2262 
2263 	return (0);
2264 }
2265 
2266 static void
2267 re_txeof(struct rl_softc *sc)
2268 {
2269 	struct ifnet		*ifp;
2270 	struct rl_txdesc	*txd;
2271 	u_int32_t		txstat;
2272 	int			cons;
2273 
2274 	cons = sc->rl_ldata.rl_tx_considx;
2275 	if (cons == sc->rl_ldata.rl_tx_prodidx)
2276 		return;
2277 
2278 	ifp = sc->rl_ifp;
2279 	/* Invalidate the TX descriptor list */
2280 	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2281 	    sc->rl_ldata.rl_tx_list_map,
2282 	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2283 
2284 	for (; cons != sc->rl_ldata.rl_tx_prodidx;
2285 	    cons = RL_TX_DESC_NXT(sc, cons)) {
2286 		txstat = le32toh(sc->rl_ldata.rl_tx_list[cons].rl_cmdstat);
2287 		if (txstat & RL_TDESC_STAT_OWN)
2288 			break;
2289 		/*
2290 		 * We only stash mbufs in the last descriptor
2291 		 * in a fragment chain, which also happens to
2292 		 * be the only place where the TX status bits
2293 		 * are valid.
2294 		 */
2295 		if (txstat & RL_TDESC_CMD_EOF) {
2296 			txd = &sc->rl_ldata.rl_tx_desc[cons];
2297 			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
2298 			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
2299 			bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag,
2300 			    txd->tx_dmamap);
2301 			KASSERT(txd->tx_m != NULL,
2302 			    ("%s: freeing NULL mbufs!", __func__));
2303 			m_freem(txd->tx_m);
2304 			txd->tx_m = NULL;
2305 			if (txstat & (RL_TDESC_STAT_EXCESSCOL|
2306 			    RL_TDESC_STAT_COLCNT))
2307 				ifp->if_collisions++;
2308 			if (txstat & RL_TDESC_STAT_TXERRSUM)
2309 				ifp->if_oerrors++;
2310 			else
2311 				ifp->if_opackets++;
2312 		}
2313 		sc->rl_ldata.rl_tx_free++;
2314 		ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
2315 	}
2316 	sc->rl_ldata.rl_tx_considx = cons;
2317 
2318 	/* No changes made to the TX ring, so no flush needed */
2319 
2320 	if (sc->rl_ldata.rl_tx_free != sc->rl_ldata.rl_tx_desc_cnt) {
2321 #ifdef RE_TX_MODERATION
2322 		/*
2323 		 * If not all descriptors have been reaped yet, reload
2324 		 * the timer so that we will eventually get another
2325 		 * interrupt that will cause us to re-enter this routine.
2326 		 * This is done in case the transmitter has gone idle.
2327 		 */
2328 		CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2329 #endif
2330 	} else
2331 		sc->rl_watchdog_timer = 0;
2332 }
2333 
2334 static void
2335 re_tick(void *xsc)
2336 {
2337 	struct rl_softc		*sc;
2338 	struct mii_data		*mii;
2339 
2340 	sc = xsc;
2341 
2342 	RL_LOCK_ASSERT(sc);
2343 
2344 	mii = device_get_softc(sc->rl_miibus);
2345 	mii_tick(mii);
2346 	if ((sc->rl_flags & RL_FLAG_LINK) == 0)
2347 		re_miibus_statchg(sc->rl_dev);
2348 	/*
2349 	 * Reclaim transmitted frames here. Technically it is not
2350 	 * necessary to do here but it ensures periodic reclamation
2351 	 * regardless of Tx completion interrupt which seems to be
2352 	 * lost on PCIe based controllers under certain situations.
2353 	 */
2354 	re_txeof(sc);
2355 	re_watchdog(sc);
2356 	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
2357 }
2358 
2359 #ifdef DEVICE_POLLING
2360 static int
2361 re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
2362 {
2363 	struct rl_softc *sc = ifp->if_softc;
2364 	int rx_npkts = 0;
2365 
2366 	RL_LOCK(sc);
2367 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
2368 		rx_npkts = re_poll_locked(ifp, cmd, count);
2369 	RL_UNLOCK(sc);
2370 	return (rx_npkts);
2371 }
2372 
2373 static int
2374 re_poll_locked(struct ifnet *ifp, enum poll_cmd cmd, int count)
2375 {
2376 	struct rl_softc *sc = ifp->if_softc;
2377 	int rx_npkts;
2378 
2379 	RL_LOCK_ASSERT(sc);
2380 
2381 	sc->rxcycles = count;
2382 	re_rxeof(sc, &rx_npkts);
2383 	re_txeof(sc);
2384 
2385 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2386 		re_start_locked(ifp);
2387 
2388 	if (cmd == POLL_AND_CHECK_STATUS) { /* also check status register */
2389 		u_int16_t       status;
2390 
2391 		status = CSR_READ_2(sc, RL_ISR);
2392 		if (status == 0xffff)
2393 			return (rx_npkts);
2394 		if (status)
2395 			CSR_WRITE_2(sc, RL_ISR, status);
2396 		if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
2397 		    (sc->rl_flags & RL_FLAG_PCIE))
2398 			CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2399 
2400 		/*
2401 		 * XXX check behaviour on receiver stalls.
2402 		 */
2403 
2404 		if (status & RL_ISR_SYSTEM_ERR) {
2405 			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2406 			re_init_locked(sc);
2407 		}
2408 	}
2409 	return (rx_npkts);
2410 }
2411 #endif /* DEVICE_POLLING */
2412 
2413 static int
2414 re_intr(void *arg)
2415 {
2416 	struct rl_softc		*sc;
2417 	uint16_t		status;
2418 
2419 	sc = arg;
2420 
2421 	status = CSR_READ_2(sc, RL_ISR);
2422 	if (status == 0xFFFF || (status & RL_INTRS_CPLUS) == 0)
2423                 return (FILTER_STRAY);
2424 	CSR_WRITE_2(sc, RL_IMR, 0);
2425 
2426 	taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_inttask);
2427 
2428 	return (FILTER_HANDLED);
2429 }
2430 
2431 static void
2432 re_int_task(void *arg, int npending)
2433 {
2434 	struct rl_softc		*sc;
2435 	struct ifnet		*ifp;
2436 	u_int16_t		status;
2437 	int			rval = 0;
2438 
2439 	sc = arg;
2440 	ifp = sc->rl_ifp;
2441 
2442 	RL_LOCK(sc);
2443 
2444 	status = CSR_READ_2(sc, RL_ISR);
2445         CSR_WRITE_2(sc, RL_ISR, status);
2446 
2447 	if (sc->suspended ||
2448 	    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2449 		RL_UNLOCK(sc);
2450 		return;
2451 	}
2452 
2453 #ifdef DEVICE_POLLING
2454 	if  (ifp->if_capenable & IFCAP_POLLING) {
2455 		RL_UNLOCK(sc);
2456 		return;
2457 	}
2458 #endif
2459 
2460 	if (status & (RL_ISR_RX_OK|RL_ISR_RX_ERR|RL_ISR_FIFO_OFLOW))
2461 		rval = re_rxeof(sc, NULL);
2462 
2463 	/*
2464 	 * Some chips will ignore a second TX request issued
2465 	 * while an existing transmission is in progress. If
2466 	 * the transmitter goes idle but there are still
2467 	 * packets waiting to be sent, we need to restart the
2468 	 * channel here to flush them out. This only seems to
2469 	 * be required with the PCIe devices.
2470 	 */
2471 	if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
2472 	    (sc->rl_flags & RL_FLAG_PCIE))
2473 		CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2474 	if (status & (
2475 #ifdef RE_TX_MODERATION
2476 	    RL_ISR_TIMEOUT_EXPIRED|
2477 #else
2478 	    RL_ISR_TX_OK|
2479 #endif
2480 	    RL_ISR_TX_ERR|RL_ISR_TX_DESC_UNAVAIL))
2481 		re_txeof(sc);
2482 
2483 	if (status & RL_ISR_SYSTEM_ERR) {
2484 		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2485 		re_init_locked(sc);
2486 	}
2487 
2488 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2489 		re_start_locked(ifp);
2490 
2491 	RL_UNLOCK(sc);
2492 
2493         if ((CSR_READ_2(sc, RL_ISR) & RL_INTRS_CPLUS) || rval) {
2494 		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_inttask);
2495 		return;
2496 	}
2497 
2498 	CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2499 }
2500 
2501 static void
2502 re_intr_msi(void *xsc)
2503 {
2504 	struct rl_softc		*sc;
2505 	struct ifnet		*ifp;
2506 	uint16_t		intrs, status;
2507 
2508 	sc = xsc;
2509 	RL_LOCK(sc);
2510 
2511 	ifp = sc->rl_ifp;
2512 #ifdef DEVICE_POLLING
2513 	if (ifp->if_capenable & IFCAP_POLLING) {
2514 		RL_UNLOCK(sc);
2515 		return;
2516 	}
2517 #endif
2518 	/* Disable interrupts. */
2519 	CSR_WRITE_2(sc, RL_IMR, 0);
2520 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2521 		RL_UNLOCK(sc);
2522 		return;
2523 	}
2524 
2525 	intrs = RL_INTRS_CPLUS;
2526 	status = CSR_READ_2(sc, RL_ISR);
2527         CSR_WRITE_2(sc, RL_ISR, status);
2528 	if (sc->rl_int_rx_act > 0) {
2529 		intrs &= ~(RL_ISR_RX_OK | RL_ISR_RX_ERR | RL_ISR_FIFO_OFLOW |
2530 		    RL_ISR_RX_OVERRUN);
2531 		status &= ~(RL_ISR_RX_OK | RL_ISR_RX_ERR | RL_ISR_FIFO_OFLOW |
2532 		    RL_ISR_RX_OVERRUN);
2533 	}
2534 
2535 	if (status & (RL_ISR_TIMEOUT_EXPIRED | RL_ISR_RX_OK | RL_ISR_RX_ERR |
2536 	    RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN)) {
2537 		re_rxeof(sc, NULL);
2538 		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
2539 			if (sc->rl_int_rx_mod != 0 &&
2540 			    (status & (RL_ISR_RX_OK | RL_ISR_RX_ERR |
2541 			    RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN)) != 0) {
2542 				/* Rearm one-shot timer. */
2543 				CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2544 				intrs &= ~(RL_ISR_RX_OK | RL_ISR_RX_ERR |
2545 				    RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN);
2546 				sc->rl_int_rx_act = 1;
2547 			} else {
2548 				intrs |= RL_ISR_RX_OK | RL_ISR_RX_ERR |
2549 				    RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN;
2550 				sc->rl_int_rx_act = 0;
2551 			}
2552 		}
2553 	}
2554 
2555 	/*
2556 	 * Some chips will ignore a second TX request issued
2557 	 * while an existing transmission is in progress. If
2558 	 * the transmitter goes idle but there are still
2559 	 * packets waiting to be sent, we need to restart the
2560 	 * channel here to flush them out. This only seems to
2561 	 * be required with the PCIe devices.
2562 	 */
2563 	if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
2564 	    (sc->rl_flags & RL_FLAG_PCIE))
2565 		CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2566 	if (status & (RL_ISR_TX_OK | RL_ISR_TX_ERR | RL_ISR_TX_DESC_UNAVAIL))
2567 		re_txeof(sc);
2568 
2569 	if (status & RL_ISR_SYSTEM_ERR) {
2570 		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2571 		re_init_locked(sc);
2572 	}
2573 
2574 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
2575 		if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2576 			re_start_locked(ifp);
2577 		CSR_WRITE_2(sc, RL_IMR, intrs);
2578 	}
2579 	RL_UNLOCK(sc);
2580 }
2581 
2582 static int
2583 re_encap(struct rl_softc *sc, struct mbuf **m_head)
2584 {
2585 	struct rl_txdesc	*txd, *txd_last;
2586 	bus_dma_segment_t	segs[RL_NTXSEGS];
2587 	bus_dmamap_t		map;
2588 	struct mbuf		*m_new;
2589 	struct rl_desc		*desc;
2590 	int			nsegs, prod;
2591 	int			i, error, ei, si;
2592 	int			padlen;
2593 	uint32_t		cmdstat, csum_flags, vlanctl;
2594 
2595 	RL_LOCK_ASSERT(sc);
2596 	M_ASSERTPKTHDR((*m_head));
2597 
2598 	/*
2599 	 * With some of the RealTek chips, using the checksum offload
2600 	 * support in conjunction with the autopadding feature results
2601 	 * in the transmission of corrupt frames. For example, if we
2602 	 * need to send a really small IP fragment that's less than 60
2603 	 * bytes in size, and IP header checksumming is enabled, the
2604 	 * resulting ethernet frame that appears on the wire will
2605 	 * have garbled payload. To work around this, if TX IP checksum
2606 	 * offload is enabled, we always manually pad short frames out
2607 	 * to the minimum ethernet frame size.
2608 	 */
2609 	if ((sc->rl_flags & RL_FLAG_AUTOPAD) == 0 &&
2610 	    (*m_head)->m_pkthdr.len < RL_IP4CSUMTX_PADLEN &&
2611 	    ((*m_head)->m_pkthdr.csum_flags & CSUM_IP) != 0) {
2612 		padlen = RL_MIN_FRAMELEN - (*m_head)->m_pkthdr.len;
2613 		if (M_WRITABLE(*m_head) == 0) {
2614 			/* Get a writable copy. */
2615 			m_new = m_dup(*m_head, M_DONTWAIT);
2616 			m_freem(*m_head);
2617 			if (m_new == NULL) {
2618 				*m_head = NULL;
2619 				return (ENOBUFS);
2620 			}
2621 			*m_head = m_new;
2622 		}
2623 		if ((*m_head)->m_next != NULL ||
2624 		    M_TRAILINGSPACE(*m_head) < padlen) {
2625 			m_new = m_defrag(*m_head, M_DONTWAIT);
2626 			if (m_new == NULL) {
2627 				m_freem(*m_head);
2628 				*m_head = NULL;
2629 				return (ENOBUFS);
2630 			}
2631 		} else
2632 			m_new = *m_head;
2633 
2634 		/*
2635 		 * Manually pad short frames, and zero the pad space
2636 		 * to avoid leaking data.
2637 		 */
2638 		bzero(mtod(m_new, char *) + m_new->m_pkthdr.len, padlen);
2639 		m_new->m_pkthdr.len += padlen;
2640 		m_new->m_len = m_new->m_pkthdr.len;
2641 		*m_head = m_new;
2642 	}
2643 
2644 	prod = sc->rl_ldata.rl_tx_prodidx;
2645 	txd = &sc->rl_ldata.rl_tx_desc[prod];
2646 	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap,
2647 	    *m_head, segs, &nsegs, BUS_DMA_NOWAIT);
2648 	if (error == EFBIG) {
2649 		m_new = m_collapse(*m_head, M_DONTWAIT, RL_NTXSEGS);
2650 		if (m_new == NULL) {
2651 			m_freem(*m_head);
2652 			*m_head = NULL;
2653 			return (ENOBUFS);
2654 		}
2655 		*m_head = m_new;
2656 		error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_tx_mtag,
2657 		    txd->tx_dmamap, *m_head, segs, &nsegs, BUS_DMA_NOWAIT);
2658 		if (error != 0) {
2659 			m_freem(*m_head);
2660 			*m_head = NULL;
2661 			return (error);
2662 		}
2663 	} else if (error != 0)
2664 		return (error);
2665 	if (nsegs == 0) {
2666 		m_freem(*m_head);
2667 		*m_head = NULL;
2668 		return (EIO);
2669 	}
2670 
2671 	/* Check for number of available descriptors. */
2672 	if (sc->rl_ldata.rl_tx_free - nsegs <= 1) {
2673 		bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap);
2674 		return (ENOBUFS);
2675 	}
2676 
2677 	bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap,
2678 	    BUS_DMASYNC_PREWRITE);
2679 
2680 	/*
2681 	 * Set up checksum offload. Note: checksum offload bits must
2682 	 * appear in all descriptors of a multi-descriptor transmit
2683 	 * attempt. This is according to testing done with an 8169
2684 	 * chip. This is a requirement.
2685 	 */
2686 	vlanctl = 0;
2687 	csum_flags = 0;
2688 	if (((*m_head)->m_pkthdr.csum_flags & CSUM_TSO) != 0) {
2689 		if ((sc->rl_flags & RL_FLAG_DESCV2) != 0) {
2690 			csum_flags |= RL_TDESC_CMD_LGSEND;
2691 			vlanctl |= ((uint32_t)(*m_head)->m_pkthdr.tso_segsz <<
2692 			    RL_TDESC_CMD_MSSVALV2_SHIFT);
2693 		} else {
2694 			csum_flags |= RL_TDESC_CMD_LGSEND |
2695 			    ((uint32_t)(*m_head)->m_pkthdr.tso_segsz <<
2696 			    RL_TDESC_CMD_MSSVAL_SHIFT);
2697 		}
2698 	} else {
2699 		/*
2700 		 * Unconditionally enable IP checksum if TCP or UDP
2701 		 * checksum is required. Otherwise, TCP/UDP checksum
2702 		 * does't make effects.
2703 		 */
2704 		if (((*m_head)->m_pkthdr.csum_flags & RE_CSUM_FEATURES) != 0) {
2705 			if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
2706 				csum_flags |= RL_TDESC_CMD_IPCSUM;
2707 				if (((*m_head)->m_pkthdr.csum_flags &
2708 				    CSUM_TCP) != 0)
2709 					csum_flags |= RL_TDESC_CMD_TCPCSUM;
2710 				if (((*m_head)->m_pkthdr.csum_flags &
2711 				    CSUM_UDP) != 0)
2712 					csum_flags |= RL_TDESC_CMD_UDPCSUM;
2713 			} else {
2714 				vlanctl |= RL_TDESC_CMD_IPCSUMV2;
2715 				if (((*m_head)->m_pkthdr.csum_flags &
2716 				    CSUM_TCP) != 0)
2717 					vlanctl |= RL_TDESC_CMD_TCPCSUMV2;
2718 				if (((*m_head)->m_pkthdr.csum_flags &
2719 				    CSUM_UDP) != 0)
2720 					vlanctl |= RL_TDESC_CMD_UDPCSUMV2;
2721 			}
2722 		}
2723 	}
2724 
2725 	/*
2726 	 * Set up hardware VLAN tagging. Note: vlan tag info must
2727 	 * appear in all descriptors of a multi-descriptor
2728 	 * transmission attempt.
2729 	 */
2730 	if ((*m_head)->m_flags & M_VLANTAG)
2731 		vlanctl |= bswap16((*m_head)->m_pkthdr.ether_vtag) |
2732 		    RL_TDESC_VLANCTL_TAG;
2733 
2734 	si = prod;
2735 	for (i = 0; i < nsegs; i++, prod = RL_TX_DESC_NXT(sc, prod)) {
2736 		desc = &sc->rl_ldata.rl_tx_list[prod];
2737 		desc->rl_vlanctl = htole32(vlanctl);
2738 		desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[i].ds_addr));
2739 		desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[i].ds_addr));
2740 		cmdstat = segs[i].ds_len;
2741 		if (i != 0)
2742 			cmdstat |= RL_TDESC_CMD_OWN;
2743 		if (prod == sc->rl_ldata.rl_tx_desc_cnt - 1)
2744 			cmdstat |= RL_TDESC_CMD_EOR;
2745 		desc->rl_cmdstat = htole32(cmdstat | csum_flags);
2746 		sc->rl_ldata.rl_tx_free--;
2747 	}
2748 	/* Update producer index. */
2749 	sc->rl_ldata.rl_tx_prodidx = prod;
2750 
2751 	/* Set EOF on the last descriptor. */
2752 	ei = RL_TX_DESC_PRV(sc, prod);
2753 	desc = &sc->rl_ldata.rl_tx_list[ei];
2754 	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_EOF);
2755 
2756 	desc = &sc->rl_ldata.rl_tx_list[si];
2757 	/* Set SOF and transfer ownership of packet to the chip. */
2758 	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_OWN | RL_TDESC_CMD_SOF);
2759 
2760 	/*
2761 	 * Insure that the map for this transmission
2762 	 * is placed at the array index of the last descriptor
2763 	 * in this chain.  (Swap last and first dmamaps.)
2764 	 */
2765 	txd_last = &sc->rl_ldata.rl_tx_desc[ei];
2766 	map = txd->tx_dmamap;
2767 	txd->tx_dmamap = txd_last->tx_dmamap;
2768 	txd_last->tx_dmamap = map;
2769 	txd_last->tx_m = *m_head;
2770 
2771 	return (0);
2772 }
2773 
2774 static void
2775 re_start(struct ifnet *ifp)
2776 {
2777 	struct rl_softc		*sc;
2778 
2779 	sc = ifp->if_softc;
2780 	RL_LOCK(sc);
2781 	re_start_locked(ifp);
2782 	RL_UNLOCK(sc);
2783 }
2784 
2785 /*
2786  * Main transmit routine for C+ and gigE NICs.
2787  */
2788 static void
2789 re_start_locked(struct ifnet *ifp)
2790 {
2791 	struct rl_softc		*sc;
2792 	struct mbuf		*m_head;
2793 	int			queued;
2794 
2795 	sc = ifp->if_softc;
2796 
2797 	if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
2798 	    IFF_DRV_RUNNING || (sc->rl_flags & RL_FLAG_LINK) == 0)
2799 		return;
2800 
2801 	for (queued = 0; !IFQ_DRV_IS_EMPTY(&ifp->if_snd) &&
2802 	    sc->rl_ldata.rl_tx_free > 1;) {
2803 		IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
2804 		if (m_head == NULL)
2805 			break;
2806 
2807 		if (re_encap(sc, &m_head) != 0) {
2808 			if (m_head == NULL)
2809 				break;
2810 			IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
2811 			ifp->if_drv_flags |= IFF_DRV_OACTIVE;
2812 			break;
2813 		}
2814 
2815 		/*
2816 		 * If there's a BPF listener, bounce a copy of this frame
2817 		 * to him.
2818 		 */
2819 		ETHER_BPF_MTAP(ifp, m_head);
2820 
2821 		queued++;
2822 	}
2823 
2824 	if (queued == 0) {
2825 #ifdef RE_TX_MODERATION
2826 		if (sc->rl_ldata.rl_tx_free != sc->rl_ldata.rl_tx_desc_cnt)
2827 			CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2828 #endif
2829 		return;
2830 	}
2831 
2832 	/* Flush the TX descriptors */
2833 
2834 	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2835 	    sc->rl_ldata.rl_tx_list_map,
2836 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2837 
2838 	CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2839 
2840 #ifdef RE_TX_MODERATION
2841 	/*
2842 	 * Use the countdown timer for interrupt moderation.
2843 	 * 'TX done' interrupts are disabled. Instead, we reset the
2844 	 * countdown timer, which will begin counting until it hits
2845 	 * the value in the TIMERINT register, and then trigger an
2846 	 * interrupt. Each time we write to the TIMERCNT register,
2847 	 * the timer count is reset to 0.
2848 	 */
2849 	CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2850 #endif
2851 
2852 	/*
2853 	 * Set a timeout in case the chip goes out to lunch.
2854 	 */
2855 	sc->rl_watchdog_timer = 5;
2856 }
2857 
2858 static void
2859 re_set_jumbo(struct rl_softc *sc, int jumbo)
2860 {
2861 
2862 	if (sc->rl_hwrev->rl_rev == RL_HWREV_8168E_VL) {
2863 		pci_set_max_read_req(sc->rl_dev, 4096);
2864 		return;
2865 	}
2866 
2867 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG);
2868 	if (jumbo != 0) {
2869 		CSR_WRITE_1(sc, RL_CFG3, CSR_READ_1(sc, RL_CFG3) |
2870 		    RL_CFG3_JUMBO_EN0);
2871 		switch (sc->rl_hwrev->rl_rev) {
2872 		case RL_HWREV_8168DP:
2873 			break;
2874 		case RL_HWREV_8168E:
2875 			CSR_WRITE_1(sc, RL_CFG4, CSR_READ_1(sc, RL_CFG4) |
2876 			    0x01);
2877 			break;
2878 		default:
2879 			CSR_WRITE_1(sc, RL_CFG4, CSR_READ_1(sc, RL_CFG4) |
2880 			    RL_CFG4_JUMBO_EN1);
2881 		}
2882 	} else {
2883 		CSR_WRITE_1(sc, RL_CFG3, CSR_READ_1(sc, RL_CFG3) &
2884 		    ~RL_CFG3_JUMBO_EN0);
2885 		switch (sc->rl_hwrev->rl_rev) {
2886 		case RL_HWREV_8168DP:
2887 			break;
2888 		case RL_HWREV_8168E:
2889 			CSR_WRITE_1(sc, RL_CFG4, CSR_READ_1(sc, RL_CFG4) &
2890 			    ~0x01);
2891 			break;
2892 		default:
2893 			CSR_WRITE_1(sc, RL_CFG4, CSR_READ_1(sc, RL_CFG4) &
2894 			    ~RL_CFG4_JUMBO_EN1);
2895 		}
2896 	}
2897 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
2898 
2899 	switch (sc->rl_hwrev->rl_rev) {
2900 	case RL_HWREV_8168DP:
2901 		pci_set_max_read_req(sc->rl_dev, 4096);
2902 		break;
2903 	default:
2904 		if (jumbo != 0)
2905 			pci_set_max_read_req(sc->rl_dev, 512);
2906 		else
2907 			pci_set_max_read_req(sc->rl_dev, 4096);
2908 	}
2909 }
2910 
2911 static void
2912 re_init(void *xsc)
2913 {
2914 	struct rl_softc		*sc = xsc;
2915 
2916 	RL_LOCK(sc);
2917 	re_init_locked(sc);
2918 	RL_UNLOCK(sc);
2919 }
2920 
2921 static void
2922 re_init_locked(struct rl_softc *sc)
2923 {
2924 	struct ifnet		*ifp = sc->rl_ifp;
2925 	struct mii_data		*mii;
2926 	uint32_t		reg;
2927 	uint16_t		cfg;
2928 	union {
2929 		uint32_t align_dummy;
2930 		u_char eaddr[ETHER_ADDR_LEN];
2931         } eaddr;
2932 
2933 	RL_LOCK_ASSERT(sc);
2934 
2935 	mii = device_get_softc(sc->rl_miibus);
2936 
2937 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
2938 		return;
2939 
2940 	/*
2941 	 * Cancel pending I/O and free all RX/TX buffers.
2942 	 */
2943 	re_stop(sc);
2944 
2945 	/* Put controller into known state. */
2946 	re_reset(sc);
2947 
2948 	/*
2949 	 * For C+ mode, initialize the RX descriptors and mbufs.
2950 	 */
2951 	if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
2952 		if (ifp->if_mtu > RL_MTU) {
2953 			if (re_jrx_list_init(sc) != 0) {
2954 				device_printf(sc->rl_dev,
2955 				    "no memory for jumbo RX buffers\n");
2956 				re_stop(sc);
2957 				return;
2958 			}
2959 			/* Disable checksum offloading for jumbo frames. */
2960 			ifp->if_capenable &= ~(IFCAP_HWCSUM | IFCAP_TSO4);
2961 			ifp->if_hwassist &= ~(RE_CSUM_FEATURES | CSUM_TSO);
2962 		} else {
2963 			if (re_rx_list_init(sc) != 0) {
2964 				device_printf(sc->rl_dev,
2965 				    "no memory for RX buffers\n");
2966 				re_stop(sc);
2967 				return;
2968 			}
2969 		}
2970 		re_set_jumbo(sc, ifp->if_mtu > RL_MTU);
2971 	} else {
2972 		if (re_rx_list_init(sc) != 0) {
2973 			device_printf(sc->rl_dev, "no memory for RX buffers\n");
2974 			re_stop(sc);
2975 			return;
2976 		}
2977 		if ((sc->rl_flags & RL_FLAG_PCIE) != 0 &&
2978 		    pci_get_device(sc->rl_dev) != RT_DEVICEID_8101E) {
2979 			if (ifp->if_mtu > RL_MTU)
2980 				pci_set_max_read_req(sc->rl_dev, 512);
2981 			else
2982 				pci_set_max_read_req(sc->rl_dev, 4096);
2983 		}
2984 	}
2985 	re_tx_list_init(sc);
2986 
2987 	/*
2988 	 * Enable C+ RX and TX mode, as well as VLAN stripping and
2989 	 * RX checksum offload. We must configure the C+ register
2990 	 * before all others.
2991 	 */
2992 	cfg = RL_CPLUSCMD_PCI_MRW;
2993 	if ((ifp->if_capenable & IFCAP_RXCSUM) != 0)
2994 		cfg |= RL_CPLUSCMD_RXCSUM_ENB;
2995 	if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0)
2996 		cfg |= RL_CPLUSCMD_VLANSTRIP;
2997 	if ((sc->rl_flags & RL_FLAG_MACSTAT) != 0) {
2998 		cfg |= RL_CPLUSCMD_MACSTAT_DIS;
2999 		/* XXX magic. */
3000 		cfg |= 0x0001;
3001 	} else
3002 		cfg |= RL_CPLUSCMD_RXENB | RL_CPLUSCMD_TXENB;
3003 	CSR_WRITE_2(sc, RL_CPLUS_CMD, cfg);
3004 	if (sc->rl_hwrev->rl_rev == RL_HWREV_8169_8110SC ||
3005 	    sc->rl_hwrev->rl_rev == RL_HWREV_8169_8110SCE) {
3006 		reg = 0x000fff00;
3007 		if ((CSR_READ_1(sc, RL_CFG2) & RL_CFG2_PCI66MHZ) != 0)
3008 			reg |= 0x000000ff;
3009 		if (sc->rl_hwrev->rl_rev == RL_HWREV_8169_8110SCE)
3010 			reg |= 0x00f00000;
3011 		CSR_WRITE_4(sc, 0x7c, reg);
3012 		/* Disable interrupt mitigation. */
3013 		CSR_WRITE_2(sc, 0xe2, 0);
3014 	}
3015 	/*
3016 	 * Disable TSO if interface MTU size is greater than MSS
3017 	 * allowed in controller.
3018 	 */
3019 	if (ifp->if_mtu > RL_TSO_MTU && (ifp->if_capenable & IFCAP_TSO4) != 0) {
3020 		ifp->if_capenable &= ~IFCAP_TSO4;
3021 		ifp->if_hwassist &= ~CSUM_TSO;
3022 	}
3023 
3024 	/*
3025 	 * Init our MAC address.  Even though the chipset
3026 	 * documentation doesn't mention it, we need to enter "Config
3027 	 * register write enable" mode to modify the ID registers.
3028 	 */
3029 	/* Copy MAC address on stack to align. */
3030 	bcopy(IF_LLADDR(ifp), eaddr.eaddr, ETHER_ADDR_LEN);
3031 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG);
3032 	CSR_WRITE_4(sc, RL_IDR0,
3033 	    htole32(*(u_int32_t *)(&eaddr.eaddr[0])));
3034 	CSR_WRITE_4(sc, RL_IDR4,
3035 	    htole32(*(u_int32_t *)(&eaddr.eaddr[4])));
3036 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3037 
3038 	/*
3039 	 * Load the addresses of the RX and TX lists into the chip.
3040 	 */
3041 
3042 	CSR_WRITE_4(sc, RL_RXLIST_ADDR_HI,
3043 	    RL_ADDR_HI(sc->rl_ldata.rl_rx_list_addr));
3044 	CSR_WRITE_4(sc, RL_RXLIST_ADDR_LO,
3045 	    RL_ADDR_LO(sc->rl_ldata.rl_rx_list_addr));
3046 
3047 	CSR_WRITE_4(sc, RL_TXLIST_ADDR_HI,
3048 	    RL_ADDR_HI(sc->rl_ldata.rl_tx_list_addr));
3049 	CSR_WRITE_4(sc, RL_TXLIST_ADDR_LO,
3050 	    RL_ADDR_LO(sc->rl_ldata.rl_tx_list_addr));
3051 
3052 	/*
3053 	 * Enable transmit and receive.
3054 	 */
3055 	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
3056 
3057 	/*
3058 	 * Set the initial TX configuration.
3059 	 */
3060 	if (sc->rl_testmode) {
3061 		if (sc->rl_type == RL_8169)
3062 			CSR_WRITE_4(sc, RL_TXCFG,
3063 			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON);
3064 		else
3065 			CSR_WRITE_4(sc, RL_TXCFG,
3066 			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON_CPLUS);
3067 	} else
3068 		CSR_WRITE_4(sc, RL_TXCFG, RL_TXCFG_CONFIG);
3069 
3070 	CSR_WRITE_1(sc, RL_EARLY_TX_THRESH, 16);
3071 
3072 	/*
3073 	 * Set the initial RX configuration.
3074 	 */
3075 	re_set_rxmode(sc);
3076 
3077 	/* Configure interrupt moderation. */
3078 	if (sc->rl_type == RL_8169) {
3079 		/* Magic from vendor. */
3080 		CSR_WRITE_2(sc, RL_INTRMOD, 0x5100);
3081 	}
3082 
3083 #ifdef DEVICE_POLLING
3084 	/*
3085 	 * Disable interrupts if we are polling.
3086 	 */
3087 	if (ifp->if_capenable & IFCAP_POLLING)
3088 		CSR_WRITE_2(sc, RL_IMR, 0);
3089 	else	/* otherwise ... */
3090 #endif
3091 
3092 	/*
3093 	 * Enable interrupts.
3094 	 */
3095 	if (sc->rl_testmode)
3096 		CSR_WRITE_2(sc, RL_IMR, 0);
3097 	else
3098 		CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
3099 	CSR_WRITE_2(sc, RL_ISR, RL_INTRS_CPLUS);
3100 
3101 	/* Set initial TX threshold */
3102 	sc->rl_txthresh = RL_TX_THRESH_INIT;
3103 
3104 	/* Start RX/TX process. */
3105 	CSR_WRITE_4(sc, RL_MISSEDPKT, 0);
3106 #ifdef notdef
3107 	/* Enable receiver and transmitter. */
3108 	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
3109 #endif
3110 
3111 	/*
3112 	 * Initialize the timer interrupt register so that
3113 	 * a timer interrupt will be generated once the timer
3114 	 * reaches a certain number of ticks. The timer is
3115 	 * reloaded on each transmit.
3116 	 */
3117 #ifdef RE_TX_MODERATION
3118 	/*
3119 	 * Use timer interrupt register to moderate TX interrupt
3120 	 * moderation, which dramatically improves TX frame rate.
3121 	 */
3122 	if (sc->rl_type == RL_8169)
3123 		CSR_WRITE_4(sc, RL_TIMERINT_8169, 0x800);
3124 	else
3125 		CSR_WRITE_4(sc, RL_TIMERINT, 0x400);
3126 #else
3127 	/*
3128 	 * Use timer interrupt register to moderate RX interrupt
3129 	 * moderation.
3130 	 */
3131 	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) != 0 &&
3132 	    intr_filter == 0) {
3133 		if (sc->rl_type == RL_8169)
3134 			CSR_WRITE_4(sc, RL_TIMERINT_8169,
3135 			    RL_USECS(sc->rl_int_rx_mod));
3136 	} else {
3137 		if (sc->rl_type == RL_8169)
3138 			CSR_WRITE_4(sc, RL_TIMERINT_8169, RL_USECS(0));
3139 	}
3140 #endif
3141 
3142 	/*
3143 	 * For 8169 gigE NICs, set the max allowed RX packet
3144 	 * size so we can receive jumbo frames.
3145 	 */
3146 	if (sc->rl_type == RL_8169) {
3147 		if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
3148 			/*
3149 			 * For controllers that use new jumbo frame scheme,
3150 			 * set maximum size of jumbo frame depedning on
3151 			 * controller revisions.
3152 			 */
3153 			if (ifp->if_mtu > RL_MTU)
3154 				CSR_WRITE_2(sc, RL_MAXRXPKTLEN,
3155 				    sc->rl_hwrev->rl_max_mtu +
3156 				    ETHER_VLAN_ENCAP_LEN + ETHER_HDR_LEN +
3157 				    ETHER_CRC_LEN);
3158 			else
3159 				CSR_WRITE_2(sc, RL_MAXRXPKTLEN,
3160 				    RE_RX_DESC_BUFLEN);
3161 		} else if ((sc->rl_flags & RL_FLAG_PCIE) != 0 &&
3162 		    sc->rl_hwrev->rl_max_mtu == RL_MTU) {
3163 			/* RTL810x has no jumbo frame support. */
3164 			CSR_WRITE_2(sc, RL_MAXRXPKTLEN, RE_RX_DESC_BUFLEN);
3165 		} else
3166 			CSR_WRITE_2(sc, RL_MAXRXPKTLEN, 16383);
3167 	}
3168 
3169 	if (sc->rl_testmode)
3170 		return;
3171 
3172 	mii_mediachg(mii);
3173 
3174 	CSR_WRITE_1(sc, RL_CFG1, CSR_READ_1(sc, RL_CFG1) | RL_CFG1_DRVLOAD);
3175 
3176 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
3177 	ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
3178 
3179 	sc->rl_flags &= ~RL_FLAG_LINK;
3180 	sc->rl_watchdog_timer = 0;
3181 	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
3182 }
3183 
3184 /*
3185  * Set media options.
3186  */
3187 static int
3188 re_ifmedia_upd(struct ifnet *ifp)
3189 {
3190 	struct rl_softc		*sc;
3191 	struct mii_data		*mii;
3192 	int			error;
3193 
3194 	sc = ifp->if_softc;
3195 	mii = device_get_softc(sc->rl_miibus);
3196 	RL_LOCK(sc);
3197 	error = mii_mediachg(mii);
3198 	RL_UNLOCK(sc);
3199 
3200 	return (error);
3201 }
3202 
3203 /*
3204  * Report current media status.
3205  */
3206 static void
3207 re_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
3208 {
3209 	struct rl_softc		*sc;
3210 	struct mii_data		*mii;
3211 
3212 	sc = ifp->if_softc;
3213 	mii = device_get_softc(sc->rl_miibus);
3214 
3215 	RL_LOCK(sc);
3216 	mii_pollstat(mii);
3217 	ifmr->ifm_active = mii->mii_media_active;
3218 	ifmr->ifm_status = mii->mii_media_status;
3219 	RL_UNLOCK(sc);
3220 }
3221 
3222 static int
3223 re_ioctl(struct ifnet *ifp, u_long command, caddr_t data)
3224 {
3225 	struct rl_softc		*sc = ifp->if_softc;
3226 	struct ifreq		*ifr = (struct ifreq *) data;
3227 	struct mii_data		*mii;
3228 	uint32_t		rev;
3229 	int			error = 0;
3230 
3231 	switch (command) {
3232 	case SIOCSIFMTU:
3233 		if (ifr->ifr_mtu < ETHERMIN ||
3234 		    ifr->ifr_mtu > sc->rl_hwrev->rl_max_mtu) {
3235 			error = EINVAL;
3236 			break;
3237 		}
3238 		RL_LOCK(sc);
3239 		if (ifp->if_mtu != ifr->ifr_mtu) {
3240 			ifp->if_mtu = ifr->ifr_mtu;
3241 			if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0 &&
3242 			    (ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
3243 				ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3244 				re_init_locked(sc);
3245 			}
3246 			if (ifp->if_mtu > RL_TSO_MTU &&
3247 			    (ifp->if_capenable & IFCAP_TSO4) != 0) {
3248 				ifp->if_capenable &= ~(IFCAP_TSO4 |
3249 				    IFCAP_VLAN_HWTSO);
3250 				ifp->if_hwassist &= ~CSUM_TSO;
3251 			}
3252 			VLAN_CAPABILITIES(ifp);
3253 		}
3254 		RL_UNLOCK(sc);
3255 		break;
3256 	case SIOCSIFFLAGS:
3257 		RL_LOCK(sc);
3258 		if ((ifp->if_flags & IFF_UP) != 0) {
3259 			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
3260 				if (((ifp->if_flags ^ sc->rl_if_flags)
3261 				    & (IFF_PROMISC | IFF_ALLMULTI)) != 0)
3262 					re_set_rxmode(sc);
3263 			} else
3264 				re_init_locked(sc);
3265 		} else {
3266 			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
3267 				re_stop(sc);
3268 		}
3269 		sc->rl_if_flags = ifp->if_flags;
3270 		RL_UNLOCK(sc);
3271 		break;
3272 	case SIOCADDMULTI:
3273 	case SIOCDELMULTI:
3274 		RL_LOCK(sc);
3275 		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
3276 			re_set_rxmode(sc);
3277 		RL_UNLOCK(sc);
3278 		break;
3279 	case SIOCGIFMEDIA:
3280 	case SIOCSIFMEDIA:
3281 		mii = device_get_softc(sc->rl_miibus);
3282 		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, command);
3283 		break;
3284 	case SIOCSIFCAP:
3285 	    {
3286 		int mask, reinit;
3287 
3288 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
3289 		reinit = 0;
3290 #ifdef DEVICE_POLLING
3291 		if (mask & IFCAP_POLLING) {
3292 			if (ifr->ifr_reqcap & IFCAP_POLLING) {
3293 				error = ether_poll_register(re_poll, ifp);
3294 				if (error)
3295 					return (error);
3296 				RL_LOCK(sc);
3297 				/* Disable interrupts */
3298 				CSR_WRITE_2(sc, RL_IMR, 0x0000);
3299 				ifp->if_capenable |= IFCAP_POLLING;
3300 				RL_UNLOCK(sc);
3301 			} else {
3302 				error = ether_poll_deregister(ifp);
3303 				/* Enable interrupts. */
3304 				RL_LOCK(sc);
3305 				CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
3306 				ifp->if_capenable &= ~IFCAP_POLLING;
3307 				RL_UNLOCK(sc);
3308 			}
3309 		}
3310 #endif /* DEVICE_POLLING */
3311 		if ((mask & IFCAP_TXCSUM) != 0 &&
3312 		    (ifp->if_capabilities & IFCAP_TXCSUM) != 0) {
3313 			ifp->if_capenable ^= IFCAP_TXCSUM;
3314 			if ((ifp->if_capenable & IFCAP_TXCSUM) != 0) {
3315 				rev = sc->rl_hwrev->rl_rev;
3316 				if (rev == RL_HWREV_8168C ||
3317 				    rev == RL_HWREV_8168C_SPIN2)
3318 					ifp->if_hwassist |= CSUM_TCP | CSUM_UDP;
3319 				else
3320 					ifp->if_hwassist |= RE_CSUM_FEATURES;
3321 			} else
3322 				ifp->if_hwassist &= ~RE_CSUM_FEATURES;
3323 			reinit = 1;
3324 		}
3325 		if ((mask & IFCAP_RXCSUM) != 0 &&
3326 		    (ifp->if_capabilities & IFCAP_RXCSUM) != 0) {
3327 			ifp->if_capenable ^= IFCAP_RXCSUM;
3328 			reinit = 1;
3329 		}
3330 		if ((mask & IFCAP_TSO4) != 0 &&
3331 		    (ifp->if_capabilities & IFCAP_TSO) != 0) {
3332 			ifp->if_capenable ^= IFCAP_TSO4;
3333 			if ((IFCAP_TSO4 & ifp->if_capenable) != 0)
3334 				ifp->if_hwassist |= CSUM_TSO;
3335 			else
3336 				ifp->if_hwassist &= ~CSUM_TSO;
3337 			if (ifp->if_mtu > RL_TSO_MTU &&
3338 			    (ifp->if_capenable & IFCAP_TSO4) != 0) {
3339 				ifp->if_capenable &= ~IFCAP_TSO4;
3340 				ifp->if_hwassist &= ~CSUM_TSO;
3341 			}
3342 		}
3343 		if ((mask & IFCAP_VLAN_HWTSO) != 0 &&
3344 		    (ifp->if_capabilities & IFCAP_VLAN_HWTSO) != 0)
3345 			ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
3346 		if ((mask & IFCAP_VLAN_HWTAGGING) != 0 &&
3347 		    (ifp->if_capabilities & IFCAP_VLAN_HWTAGGING) != 0) {
3348 			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
3349 			/* TSO over VLAN requires VLAN hardware tagging. */
3350 			if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) == 0)
3351 				ifp->if_capenable &= ~IFCAP_VLAN_HWTSO;
3352 			reinit = 1;
3353 		}
3354 		if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0 &&
3355 		    (mask & (IFCAP_HWCSUM | IFCAP_TSO4 |
3356 		    IFCAP_VLAN_HWTSO)) != 0)
3357 				reinit = 1;
3358 		if ((mask & IFCAP_WOL) != 0 &&
3359 		    (ifp->if_capabilities & IFCAP_WOL) != 0) {
3360 			if ((mask & IFCAP_WOL_UCAST) != 0)
3361 				ifp->if_capenable ^= IFCAP_WOL_UCAST;
3362 			if ((mask & IFCAP_WOL_MCAST) != 0)
3363 				ifp->if_capenable ^= IFCAP_WOL_MCAST;
3364 			if ((mask & IFCAP_WOL_MAGIC) != 0)
3365 				ifp->if_capenable ^= IFCAP_WOL_MAGIC;
3366 		}
3367 		if (reinit && ifp->if_drv_flags & IFF_DRV_RUNNING) {
3368 			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3369 			re_init(sc);
3370 		}
3371 		VLAN_CAPABILITIES(ifp);
3372 	    }
3373 		break;
3374 	default:
3375 		error = ether_ioctl(ifp, command, data);
3376 		break;
3377 	}
3378 
3379 	return (error);
3380 }
3381 
3382 static void
3383 re_watchdog(struct rl_softc *sc)
3384 {
3385 	struct ifnet		*ifp;
3386 
3387 	RL_LOCK_ASSERT(sc);
3388 
3389 	if (sc->rl_watchdog_timer == 0 || --sc->rl_watchdog_timer != 0)
3390 		return;
3391 
3392 	ifp = sc->rl_ifp;
3393 	re_txeof(sc);
3394 	if (sc->rl_ldata.rl_tx_free == sc->rl_ldata.rl_tx_desc_cnt) {
3395 		if_printf(ifp, "watchdog timeout (missed Tx interrupts) "
3396 		    "-- recovering\n");
3397 		if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
3398 			re_start_locked(ifp);
3399 		return;
3400 	}
3401 
3402 	if_printf(ifp, "watchdog timeout\n");
3403 	ifp->if_oerrors++;
3404 
3405 	re_rxeof(sc, NULL);
3406 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3407 	re_init_locked(sc);
3408 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
3409 		re_start_locked(ifp);
3410 }
3411 
3412 /*
3413  * Stop the adapter and free any mbufs allocated to the
3414  * RX and TX lists.
3415  */
3416 static void
3417 re_stop(struct rl_softc *sc)
3418 {
3419 	int			i;
3420 	struct ifnet		*ifp;
3421 	struct rl_txdesc	*txd;
3422 	struct rl_rxdesc	*rxd;
3423 
3424 	RL_LOCK_ASSERT(sc);
3425 
3426 	ifp = sc->rl_ifp;
3427 
3428 	sc->rl_watchdog_timer = 0;
3429 	callout_stop(&sc->rl_stat_callout);
3430 	ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
3431 
3432 	if ((sc->rl_flags & RL_FLAG_CMDSTOP) != 0)
3433 		CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_STOPREQ | RL_CMD_TX_ENB |
3434 		    RL_CMD_RX_ENB);
3435 	else
3436 		CSR_WRITE_1(sc, RL_COMMAND, 0x00);
3437 	DELAY(1000);
3438 	CSR_WRITE_2(sc, RL_IMR, 0x0000);
3439 	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
3440 
3441 	if (sc->rl_head != NULL) {
3442 		m_freem(sc->rl_head);
3443 		sc->rl_head = sc->rl_tail = NULL;
3444 	}
3445 
3446 	/* Free the TX list buffers. */
3447 
3448 	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
3449 		txd = &sc->rl_ldata.rl_tx_desc[i];
3450 		if (txd->tx_m != NULL) {
3451 			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
3452 			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
3453 			bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag,
3454 			    txd->tx_dmamap);
3455 			m_freem(txd->tx_m);
3456 			txd->tx_m = NULL;
3457 		}
3458 	}
3459 
3460 	/* Free the RX list buffers. */
3461 
3462 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
3463 		rxd = &sc->rl_ldata.rl_rx_desc[i];
3464 		if (rxd->rx_m != NULL) {
3465 			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
3466 			    rxd->rx_dmamap, BUS_DMASYNC_POSTREAD);
3467 			bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag,
3468 			    rxd->rx_dmamap);
3469 			m_freem(rxd->rx_m);
3470 			rxd->rx_m = NULL;
3471 		}
3472 	}
3473 }
3474 
3475 /*
3476  * Device suspend routine.  Stop the interface and save some PCI
3477  * settings in case the BIOS doesn't restore them properly on
3478  * resume.
3479  */
3480 static int
3481 re_suspend(device_t dev)
3482 {
3483 	struct rl_softc		*sc;
3484 
3485 	sc = device_get_softc(dev);
3486 
3487 	RL_LOCK(sc);
3488 	re_stop(sc);
3489 	re_setwol(sc);
3490 	sc->suspended = 1;
3491 	RL_UNLOCK(sc);
3492 
3493 	return (0);
3494 }
3495 
3496 /*
3497  * Device resume routine.  Restore some PCI settings in case the BIOS
3498  * doesn't, re-enable busmastering, and restart the interface if
3499  * appropriate.
3500  */
3501 static int
3502 re_resume(device_t dev)
3503 {
3504 	struct rl_softc		*sc;
3505 	struct ifnet		*ifp;
3506 
3507 	sc = device_get_softc(dev);
3508 
3509 	RL_LOCK(sc);
3510 
3511 	ifp = sc->rl_ifp;
3512 	/* Take controller out of sleep mode. */
3513 	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
3514 		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
3515 			CSR_WRITE_1(sc, RL_GPIO,
3516 			    CSR_READ_1(sc, RL_GPIO) | 0x01);
3517 	}
3518 
3519 	/*
3520 	 * Clear WOL matching such that normal Rx filtering
3521 	 * wouldn't interfere with WOL patterns.
3522 	 */
3523 	re_clrwol(sc);
3524 
3525 	/* reinitialize interface if necessary */
3526 	if (ifp->if_flags & IFF_UP)
3527 		re_init_locked(sc);
3528 
3529 	sc->suspended = 0;
3530 	RL_UNLOCK(sc);
3531 
3532 	return (0);
3533 }
3534 
3535 /*
3536  * Stop all chip I/O so that the kernel's probe routines don't
3537  * get confused by errant DMAs when rebooting.
3538  */
3539 static int
3540 re_shutdown(device_t dev)
3541 {
3542 	struct rl_softc		*sc;
3543 
3544 	sc = device_get_softc(dev);
3545 
3546 	RL_LOCK(sc);
3547 	re_stop(sc);
3548 	/*
3549 	 * Mark interface as down since otherwise we will panic if
3550 	 * interrupt comes in later on, which can happen in some
3551 	 * cases.
3552 	 */
3553 	sc->rl_ifp->if_flags &= ~IFF_UP;
3554 	re_setwol(sc);
3555 	RL_UNLOCK(sc);
3556 
3557 	return (0);
3558 }
3559 
3560 static void
3561 re_setwol(struct rl_softc *sc)
3562 {
3563 	struct ifnet		*ifp;
3564 	int			pmc;
3565 	uint16_t		pmstat;
3566 	uint8_t			v;
3567 
3568 	RL_LOCK_ASSERT(sc);
3569 
3570 	if (pci_find_cap(sc->rl_dev, PCIY_PMG, &pmc) != 0)
3571 		return;
3572 
3573 	ifp = sc->rl_ifp;
3574 	/* Put controller into sleep mode. */
3575 	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
3576 		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
3577 			CSR_WRITE_1(sc, RL_GPIO,
3578 			    CSR_READ_1(sc, RL_GPIO) & ~0x01);
3579 	}
3580 	if ((ifp->if_capenable & IFCAP_WOL) != 0 &&
3581 	    (sc->rl_flags & RL_FLAG_WOLRXENB) != 0)
3582 		CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RX_ENB);
3583 	/* Enable config register write. */
3584 	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
3585 
3586 	/* Enable PME. */
3587 	v = CSR_READ_1(sc, RL_CFG1);
3588 	v &= ~RL_CFG1_PME;
3589 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3590 		v |= RL_CFG1_PME;
3591 	CSR_WRITE_1(sc, RL_CFG1, v);
3592 
3593 	v = CSR_READ_1(sc, RL_CFG3);
3594 	v &= ~(RL_CFG3_WOL_LINK | RL_CFG3_WOL_MAGIC);
3595 	if ((ifp->if_capenable & IFCAP_WOL_MAGIC) != 0)
3596 		v |= RL_CFG3_WOL_MAGIC;
3597 	CSR_WRITE_1(sc, RL_CFG3, v);
3598 
3599 	/* Config register write done. */
3600 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3601 
3602 	v = CSR_READ_1(sc, RL_CFG5);
3603 	v &= ~(RL_CFG5_WOL_BCAST | RL_CFG5_WOL_MCAST | RL_CFG5_WOL_UCAST);
3604 	v &= ~RL_CFG5_WOL_LANWAKE;
3605 	if ((ifp->if_capenable & IFCAP_WOL_UCAST) != 0)
3606 		v |= RL_CFG5_WOL_UCAST;
3607 	if ((ifp->if_capenable & IFCAP_WOL_MCAST) != 0)
3608 		v |= RL_CFG5_WOL_MCAST | RL_CFG5_WOL_BCAST;
3609 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3610 		v |= RL_CFG5_WOL_LANWAKE;
3611 	CSR_WRITE_1(sc, RL_CFG5, v);
3612 
3613 	if ((ifp->if_capenable & IFCAP_WOL) != 0 &&
3614 	    (sc->rl_flags & RL_FLAG_PHYWAKE_PM) != 0)
3615 		CSR_WRITE_1(sc, RL_PMCH, CSR_READ_1(sc, RL_PMCH) & ~0x80);
3616 	/*
3617 	 * It seems that hardware resets its link speed to 100Mbps in
3618 	 * power down mode so switching to 100Mbps in driver is not
3619 	 * needed.
3620 	 */
3621 
3622 	/* Request PME if WOL is requested. */
3623 	pmstat = pci_read_config(sc->rl_dev, pmc + PCIR_POWER_STATUS, 2);
3624 	pmstat &= ~(PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE);
3625 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3626 		pmstat |= PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE;
3627 	pci_write_config(sc->rl_dev, pmc + PCIR_POWER_STATUS, pmstat, 2);
3628 }
3629 
3630 static void
3631 re_clrwol(struct rl_softc *sc)
3632 {
3633 	int			pmc;
3634 	uint8_t			v;
3635 
3636 	RL_LOCK_ASSERT(sc);
3637 
3638 	if (pci_find_cap(sc->rl_dev, PCIY_PMG, &pmc) != 0)
3639 		return;
3640 
3641 	/* Enable config register write. */
3642 	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
3643 
3644 	v = CSR_READ_1(sc, RL_CFG3);
3645 	v &= ~(RL_CFG3_WOL_LINK | RL_CFG3_WOL_MAGIC);
3646 	CSR_WRITE_1(sc, RL_CFG3, v);
3647 
3648 	/* Config register write done. */
3649 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3650 
3651 	v = CSR_READ_1(sc, RL_CFG5);
3652 	v &= ~(RL_CFG5_WOL_BCAST | RL_CFG5_WOL_MCAST | RL_CFG5_WOL_UCAST);
3653 	v &= ~RL_CFG5_WOL_LANWAKE;
3654 	CSR_WRITE_1(sc, RL_CFG5, v);
3655 }
3656 
3657 static void
3658 re_add_sysctls(struct rl_softc *sc)
3659 {
3660 	struct sysctl_ctx_list	*ctx;
3661 	struct sysctl_oid_list	*children;
3662 	int			error;
3663 
3664 	ctx = device_get_sysctl_ctx(sc->rl_dev);
3665 	children = SYSCTL_CHILDREN(device_get_sysctl_tree(sc->rl_dev));
3666 
3667 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "stats",
3668 	    CTLTYPE_INT | CTLFLAG_RW, sc, 0, re_sysctl_stats, "I",
3669 	    "Statistics Information");
3670 	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) == 0)
3671 		return;
3672 
3673 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "int_rx_mod",
3674 	    CTLTYPE_INT | CTLFLAG_RW, &sc->rl_int_rx_mod, 0,
3675 	    sysctl_hw_re_int_mod, "I", "re RX interrupt moderation");
3676 	/* Pull in device tunables. */
3677 	sc->rl_int_rx_mod = RL_TIMER_DEFAULT;
3678 	error = resource_int_value(device_get_name(sc->rl_dev),
3679 	    device_get_unit(sc->rl_dev), "int_rx_mod", &sc->rl_int_rx_mod);
3680 	if (error == 0) {
3681 		if (sc->rl_int_rx_mod < RL_TIMER_MIN ||
3682 		    sc->rl_int_rx_mod > RL_TIMER_MAX) {
3683 			device_printf(sc->rl_dev, "int_rx_mod value out of "
3684 			    "range; using default: %d\n",
3685 			    RL_TIMER_DEFAULT);
3686 			sc->rl_int_rx_mod = RL_TIMER_DEFAULT;
3687 		}
3688 	}
3689 
3690 }
3691 
3692 static int
3693 re_sysctl_stats(SYSCTL_HANDLER_ARGS)
3694 {
3695 	struct rl_softc		*sc;
3696 	struct rl_stats		*stats;
3697 	int			error, i, result;
3698 
3699 	result = -1;
3700 	error = sysctl_handle_int(oidp, &result, 0, req);
3701 	if (error || req->newptr == NULL)
3702 		return (error);
3703 
3704 	if (result == 1) {
3705 		sc = (struct rl_softc *)arg1;
3706 		RL_LOCK(sc);
3707 		if ((sc->rl_ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
3708 			RL_UNLOCK(sc);
3709 			goto done;
3710 		}
3711 		bus_dmamap_sync(sc->rl_ldata.rl_stag,
3712 		    sc->rl_ldata.rl_smap, BUS_DMASYNC_PREREAD);
3713 		CSR_WRITE_4(sc, RL_DUMPSTATS_HI,
3714 		    RL_ADDR_HI(sc->rl_ldata.rl_stats_addr));
3715 		CSR_WRITE_4(sc, RL_DUMPSTATS_LO,
3716 		    RL_ADDR_LO(sc->rl_ldata.rl_stats_addr));
3717 		CSR_WRITE_4(sc, RL_DUMPSTATS_LO,
3718 		    RL_ADDR_LO(sc->rl_ldata.rl_stats_addr |
3719 		    RL_DUMPSTATS_START));
3720 		for (i = RL_TIMEOUT; i > 0; i--) {
3721 			if ((CSR_READ_4(sc, RL_DUMPSTATS_LO) &
3722 			    RL_DUMPSTATS_START) == 0)
3723 				break;
3724 			DELAY(1000);
3725 		}
3726 		bus_dmamap_sync(sc->rl_ldata.rl_stag,
3727 		    sc->rl_ldata.rl_smap, BUS_DMASYNC_POSTREAD);
3728 		RL_UNLOCK(sc);
3729 		if (i == 0) {
3730 			device_printf(sc->rl_dev,
3731 			    "DUMP statistics request timedout\n");
3732 			return (ETIMEDOUT);
3733 		}
3734 done:
3735 		stats = sc->rl_ldata.rl_stats;
3736 		printf("%s statistics:\n", device_get_nameunit(sc->rl_dev));
3737 		printf("Tx frames : %ju\n",
3738 		    (uintmax_t)le64toh(stats->rl_tx_pkts));
3739 		printf("Rx frames : %ju\n",
3740 		    (uintmax_t)le64toh(stats->rl_rx_pkts));
3741 		printf("Tx errors : %ju\n",
3742 		    (uintmax_t)le64toh(stats->rl_tx_errs));
3743 		printf("Rx errors : %u\n",
3744 		    le32toh(stats->rl_rx_errs));
3745 		printf("Rx missed frames : %u\n",
3746 		    (uint32_t)le16toh(stats->rl_missed_pkts));
3747 		printf("Rx frame alignment errs : %u\n",
3748 		    (uint32_t)le16toh(stats->rl_rx_framealign_errs));
3749 		printf("Tx single collisions : %u\n",
3750 		    le32toh(stats->rl_tx_onecoll));
3751 		printf("Tx multiple collisions : %u\n",
3752 		    le32toh(stats->rl_tx_multicolls));
3753 		printf("Rx unicast frames : %ju\n",
3754 		    (uintmax_t)le64toh(stats->rl_rx_ucasts));
3755 		printf("Rx broadcast frames : %ju\n",
3756 		    (uintmax_t)le64toh(stats->rl_rx_bcasts));
3757 		printf("Rx multicast frames : %u\n",
3758 		    le32toh(stats->rl_rx_mcasts));
3759 		printf("Tx aborts : %u\n",
3760 		    (uint32_t)le16toh(stats->rl_tx_aborts));
3761 		printf("Tx underruns : %u\n",
3762 		    (uint32_t)le16toh(stats->rl_rx_underruns));
3763 	}
3764 
3765 	return (error);
3766 }
3767 
3768 static int
3769 sysctl_int_range(SYSCTL_HANDLER_ARGS, int low, int high)
3770 {
3771 	int error, value;
3772 
3773 	if (arg1 == NULL)
3774 		return (EINVAL);
3775 	value = *(int *)arg1;
3776 	error = sysctl_handle_int(oidp, &value, 0, req);
3777 	if (error || req->newptr == NULL)
3778 		return (error);
3779 	if (value < low || value > high)
3780 		return (EINVAL);
3781 	*(int *)arg1 = value;
3782 
3783 	return (0);
3784 }
3785 
3786 static int
3787 sysctl_hw_re_int_mod(SYSCTL_HANDLER_ARGS)
3788 {
3789 
3790 	return (sysctl_int_range(oidp, arg1, arg2, req, RL_TIMER_MIN,
3791 	    RL_TIMER_MAX));
3792 }
3793