xref: /freebsd/sys/dev/re/if_re.c (revision 884a2a699669ec61e2366e3e358342dbc94be24a)
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 struct rl_type re_devs[] = {
175 	{ DLINK_VENDORID, DLINK_DEVICEID_528T, 0,
176 	    "D-Link DGE-528(T) Gigabit Ethernet Adapter" },
177 	{ RT_VENDORID, RT_DEVICEID_8139, 0,
178 	    "RealTek 8139C+ 10/100BaseTX" },
179 	{ RT_VENDORID, RT_DEVICEID_8101E, 0,
180 	    "RealTek 810xE PCIe 10/100baseTX" },
181 	{ RT_VENDORID, RT_DEVICEID_8168, 0,
182 	    "RealTek 8168/8111 B/C/CP/D/DP/E PCIe Gigabit Ethernet" },
183 	{ RT_VENDORID, RT_DEVICEID_8169, 0,
184 	    "RealTek 8169/8169S/8169SB(L)/8110S/8110SB(L) Gigabit Ethernet" },
185 	{ RT_VENDORID, RT_DEVICEID_8169SC, 0,
186 	    "RealTek 8169SC/8110SC Single-chip Gigabit Ethernet" },
187 	{ COREGA_VENDORID, COREGA_DEVICEID_CGLAPCIGT, 0,
188 	    "Corega CG-LAPCIGT (RTL8169S) Gigabit Ethernet" },
189 	{ LINKSYS_VENDORID, LINKSYS_DEVICEID_EG1032, 0,
190 	    "Linksys EG1032 (RTL8169S) Gigabit Ethernet" },
191 	{ USR_VENDORID, USR_DEVICEID_997902, 0,
192 	    "US Robotics 997902 (RTL8169S) Gigabit Ethernet" }
193 };
194 
195 static struct rl_hwrev re_hwrevs[] = {
196 	{ RL_HWREV_8139, RL_8139,  "", RL_MTU },
197 	{ RL_HWREV_8139A, RL_8139, "A", RL_MTU },
198 	{ RL_HWREV_8139AG, RL_8139, "A-G", RL_MTU },
199 	{ RL_HWREV_8139B, RL_8139, "B", RL_MTU },
200 	{ RL_HWREV_8130, RL_8139, "8130", RL_MTU },
201 	{ RL_HWREV_8139C, RL_8139, "C", RL_MTU },
202 	{ RL_HWREV_8139D, RL_8139, "8139D/8100B/8100C", RL_MTU },
203 	{ RL_HWREV_8139CPLUS, RL_8139CPLUS, "C+", RL_MTU },
204 	{ RL_HWREV_8168B_SPIN1, RL_8169, "8168", RL_JUMBO_MTU },
205 	{ RL_HWREV_8169, RL_8169, "8169", RL_JUMBO_MTU },
206 	{ RL_HWREV_8169S, RL_8169, "8169S", RL_JUMBO_MTU },
207 	{ RL_HWREV_8110S, RL_8169, "8110S", RL_JUMBO_MTU },
208 	{ RL_HWREV_8169_8110SB, RL_8169, "8169SB/8110SB", RL_JUMBO_MTU },
209 	{ RL_HWREV_8169_8110SC, RL_8169, "8169SC/8110SC", RL_JUMBO_MTU },
210 	{ RL_HWREV_8169_8110SBL, RL_8169, "8169SBL/8110SBL", RL_JUMBO_MTU },
211 	{ RL_HWREV_8169_8110SCE, RL_8169, "8169SC/8110SC", RL_JUMBO_MTU },
212 	{ RL_HWREV_8100, RL_8139, "8100", RL_MTU },
213 	{ RL_HWREV_8101, RL_8139, "8101", RL_MTU },
214 	{ RL_HWREV_8100E, RL_8169, "8100E", RL_MTU },
215 	{ RL_HWREV_8101E, RL_8169, "8101E", RL_MTU },
216 	{ RL_HWREV_8102E, RL_8169, "8102E", RL_MTU },
217 	{ RL_HWREV_8102EL, RL_8169, "8102EL", RL_MTU },
218 	{ RL_HWREV_8102EL_SPIN1, RL_8169, "8102EL", RL_MTU },
219 	{ RL_HWREV_8103E, RL_8169, "8103E", RL_MTU },
220 	{ RL_HWREV_8401E, RL_8169, "8401E", RL_MTU },
221 	{ RL_HWREV_8105E, RL_8169, "8105E", RL_MTU },
222 	{ RL_HWREV_8168B_SPIN2, RL_8169, "8168", RL_JUMBO_MTU },
223 	{ RL_HWREV_8168B_SPIN3, RL_8169, "8168", RL_JUMBO_MTU },
224 	{ RL_HWREV_8168C, RL_8169, "8168C/8111C", RL_JUMBO_MTU_6K },
225 	{ RL_HWREV_8168C_SPIN2, RL_8169, "8168C/8111C", RL_JUMBO_MTU_6K },
226 	{ RL_HWREV_8168CP, RL_8169, "8168CP/8111CP", RL_JUMBO_MTU_6K },
227 	{ RL_HWREV_8168D, RL_8169, "8168D/8111D", RL_JUMBO_MTU_9K },
228 	{ RL_HWREV_8168DP, RL_8169, "8168DP/8111DP", RL_JUMBO_MTU_9K },
229 	{ RL_HWREV_8168E, RL_8169, "8168E/8111E", RL_JUMBO_MTU_9K},
230 	{ RL_HWREV_8168E_VL, RL_8169, "8168E/8111E-VL", RL_JUMBO_MTU_6K},
231 	{ 0, 0, NULL, 0 }
232 };
233 
234 static int re_probe		(device_t);
235 static int re_attach		(device_t);
236 static int re_detach		(device_t);
237 
238 static int re_encap		(struct rl_softc *, struct mbuf **);
239 
240 static void re_dma_map_addr	(void *, bus_dma_segment_t *, int, int);
241 static int re_allocmem		(device_t, struct rl_softc *);
242 static __inline void re_discard_rxbuf
243 				(struct rl_softc *, int);
244 static int re_newbuf		(struct rl_softc *, int);
245 static int re_jumbo_newbuf	(struct rl_softc *, int);
246 static int re_rx_list_init	(struct rl_softc *);
247 static int re_jrx_list_init	(struct rl_softc *);
248 static int re_tx_list_init	(struct rl_softc *);
249 #ifdef RE_FIXUP_RX
250 static __inline void re_fixup_rx
251 				(struct mbuf *);
252 #endif
253 static int re_rxeof		(struct rl_softc *, int *);
254 static void re_txeof		(struct rl_softc *);
255 #ifdef DEVICE_POLLING
256 static int re_poll		(struct ifnet *, enum poll_cmd, int);
257 static int re_poll_locked	(struct ifnet *, enum poll_cmd, int);
258 #endif
259 static int re_intr		(void *);
260 static void re_intr_msi		(void *);
261 static void re_tick		(void *);
262 static void re_int_task		(void *, int);
263 static void re_start		(struct ifnet *);
264 static void re_start_locked	(struct ifnet *);
265 static int re_ioctl		(struct ifnet *, u_long, caddr_t);
266 static void re_init		(void *);
267 static void re_init_locked	(struct rl_softc *);
268 static void re_stop		(struct rl_softc *);
269 static void re_watchdog		(struct rl_softc *);
270 static int re_suspend		(device_t);
271 static int re_resume		(device_t);
272 static int re_shutdown		(device_t);
273 static int re_ifmedia_upd	(struct ifnet *);
274 static void re_ifmedia_sts	(struct ifnet *, struct ifmediareq *);
275 
276 static void re_eeprom_putbyte	(struct rl_softc *, int);
277 static void re_eeprom_getword	(struct rl_softc *, int, u_int16_t *);
278 static void re_read_eeprom	(struct rl_softc *, caddr_t, int, int);
279 static int re_gmii_readreg	(device_t, int, int);
280 static int re_gmii_writereg	(device_t, int, int, int);
281 
282 static int re_miibus_readreg	(device_t, int, int);
283 static int re_miibus_writereg	(device_t, int, int, int);
284 static void re_miibus_statchg	(device_t);
285 
286 static void re_set_jumbo	(struct rl_softc *, int);
287 static void re_set_rxmode		(struct rl_softc *);
288 static void re_reset		(struct rl_softc *);
289 static void re_setwol		(struct rl_softc *);
290 static void re_clrwol		(struct rl_softc *);
291 
292 #ifdef RE_DIAG
293 static int re_diag		(struct rl_softc *);
294 #endif
295 
296 static void re_add_sysctls	(struct rl_softc *);
297 static int re_sysctl_stats	(SYSCTL_HANDLER_ARGS);
298 static int sysctl_int_range	(SYSCTL_HANDLER_ARGS, int, int);
299 static int sysctl_hw_re_int_mod	(SYSCTL_HANDLER_ARGS);
300 
301 static device_method_t re_methods[] = {
302 	/* Device interface */
303 	DEVMETHOD(device_probe,		re_probe),
304 	DEVMETHOD(device_attach,	re_attach),
305 	DEVMETHOD(device_detach,	re_detach),
306 	DEVMETHOD(device_suspend,	re_suspend),
307 	DEVMETHOD(device_resume,	re_resume),
308 	DEVMETHOD(device_shutdown,	re_shutdown),
309 
310 	/* bus interface */
311 	DEVMETHOD(bus_print_child,	bus_generic_print_child),
312 	DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
313 
314 	/* MII interface */
315 	DEVMETHOD(miibus_readreg,	re_miibus_readreg),
316 	DEVMETHOD(miibus_writereg,	re_miibus_writereg),
317 	DEVMETHOD(miibus_statchg,	re_miibus_statchg),
318 
319 	{ 0, 0 }
320 };
321 
322 static driver_t re_driver = {
323 	"re",
324 	re_methods,
325 	sizeof(struct rl_softc)
326 };
327 
328 static devclass_t re_devclass;
329 
330 DRIVER_MODULE(re, pci, re_driver, re_devclass, 0, 0);
331 DRIVER_MODULE(miibus, re, miibus_driver, miibus_devclass, 0, 0);
332 
333 #define EE_SET(x)					\
334 	CSR_WRITE_1(sc, RL_EECMD,			\
335 		CSR_READ_1(sc, RL_EECMD) | x)
336 
337 #define EE_CLR(x)					\
338 	CSR_WRITE_1(sc, RL_EECMD,			\
339 		CSR_READ_1(sc, RL_EECMD) & ~x)
340 
341 /*
342  * Send a read command and address to the EEPROM, check for ACK.
343  */
344 static void
345 re_eeprom_putbyte(struct rl_softc *sc, int addr)
346 {
347 	int			d, i;
348 
349 	d = addr | (RL_9346_READ << sc->rl_eewidth);
350 
351 	/*
352 	 * Feed in each bit and strobe the clock.
353 	 */
354 
355 	for (i = 1 << (sc->rl_eewidth + 3); i; i >>= 1) {
356 		if (d & i) {
357 			EE_SET(RL_EE_DATAIN);
358 		} else {
359 			EE_CLR(RL_EE_DATAIN);
360 		}
361 		DELAY(100);
362 		EE_SET(RL_EE_CLK);
363 		DELAY(150);
364 		EE_CLR(RL_EE_CLK);
365 		DELAY(100);
366 	}
367 }
368 
369 /*
370  * Read a word of data stored in the EEPROM at address 'addr.'
371  */
372 static void
373 re_eeprom_getword(struct rl_softc *sc, int addr, u_int16_t *dest)
374 {
375 	int			i;
376 	u_int16_t		word = 0;
377 
378 	/*
379 	 * Send address of word we want to read.
380 	 */
381 	re_eeprom_putbyte(sc, addr);
382 
383 	/*
384 	 * Start reading bits from EEPROM.
385 	 */
386 	for (i = 0x8000; i; i >>= 1) {
387 		EE_SET(RL_EE_CLK);
388 		DELAY(100);
389 		if (CSR_READ_1(sc, RL_EECMD) & RL_EE_DATAOUT)
390 			word |= i;
391 		EE_CLR(RL_EE_CLK);
392 		DELAY(100);
393 	}
394 
395 	*dest = word;
396 }
397 
398 /*
399  * Read a sequence of words from the EEPROM.
400  */
401 static void
402 re_read_eeprom(struct rl_softc *sc, caddr_t dest, int off, int cnt)
403 {
404 	int			i;
405 	u_int16_t		word = 0, *ptr;
406 
407 	CSR_SETBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM);
408 
409         DELAY(100);
410 
411 	for (i = 0; i < cnt; i++) {
412 		CSR_SETBIT_1(sc, RL_EECMD, RL_EE_SEL);
413 		re_eeprom_getword(sc, off + i, &word);
414 		CSR_CLRBIT_1(sc, RL_EECMD, RL_EE_SEL);
415 		ptr = (u_int16_t *)(dest + (i * 2));
416                 *ptr = word;
417 	}
418 
419 	CSR_CLRBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM);
420 }
421 
422 static int
423 re_gmii_readreg(device_t dev, int phy, int reg)
424 {
425 	struct rl_softc		*sc;
426 	u_int32_t		rval;
427 	int			i;
428 
429 	sc = device_get_softc(dev);
430 
431 	/* Let the rgephy driver read the GMEDIASTAT register */
432 
433 	if (reg == RL_GMEDIASTAT) {
434 		rval = CSR_READ_1(sc, RL_GMEDIASTAT);
435 		return (rval);
436 	}
437 
438 	CSR_WRITE_4(sc, RL_PHYAR, reg << 16);
439 
440 	for (i = 0; i < RL_PHY_TIMEOUT; i++) {
441 		rval = CSR_READ_4(sc, RL_PHYAR);
442 		if (rval & RL_PHYAR_BUSY)
443 			break;
444 		DELAY(25);
445 	}
446 
447 	if (i == RL_PHY_TIMEOUT) {
448 		device_printf(sc->rl_dev, "PHY read failed\n");
449 		return (0);
450 	}
451 
452 	/*
453 	 * Controller requires a 20us delay to process next MDIO request.
454 	 */
455 	DELAY(20);
456 
457 	return (rval & RL_PHYAR_PHYDATA);
458 }
459 
460 static int
461 re_gmii_writereg(device_t dev, int phy, int reg, int data)
462 {
463 	struct rl_softc		*sc;
464 	u_int32_t		rval;
465 	int			i;
466 
467 	sc = device_get_softc(dev);
468 
469 	CSR_WRITE_4(sc, RL_PHYAR, (reg << 16) |
470 	    (data & RL_PHYAR_PHYDATA) | RL_PHYAR_BUSY);
471 
472 	for (i = 0; i < RL_PHY_TIMEOUT; i++) {
473 		rval = CSR_READ_4(sc, RL_PHYAR);
474 		if (!(rval & RL_PHYAR_BUSY))
475 			break;
476 		DELAY(25);
477 	}
478 
479 	if (i == RL_PHY_TIMEOUT) {
480 		device_printf(sc->rl_dev, "PHY write failed\n");
481 		return (0);
482 	}
483 
484 	/*
485 	 * Controller requires a 20us delay to process next MDIO request.
486 	 */
487 	DELAY(20);
488 
489 	return (0);
490 }
491 
492 static int
493 re_miibus_readreg(device_t dev, int phy, int reg)
494 {
495 	struct rl_softc		*sc;
496 	u_int16_t		rval = 0;
497 	u_int16_t		re8139_reg = 0;
498 
499 	sc = device_get_softc(dev);
500 
501 	if (sc->rl_type == RL_8169) {
502 		rval = re_gmii_readreg(dev, phy, reg);
503 		return (rval);
504 	}
505 
506 	switch (reg) {
507 	case MII_BMCR:
508 		re8139_reg = RL_BMCR;
509 		break;
510 	case MII_BMSR:
511 		re8139_reg = RL_BMSR;
512 		break;
513 	case MII_ANAR:
514 		re8139_reg = RL_ANAR;
515 		break;
516 	case MII_ANER:
517 		re8139_reg = RL_ANER;
518 		break;
519 	case MII_ANLPAR:
520 		re8139_reg = RL_LPAR;
521 		break;
522 	case MII_PHYIDR1:
523 	case MII_PHYIDR2:
524 		return (0);
525 	/*
526 	 * Allow the rlphy driver to read the media status
527 	 * register. If we have a link partner which does not
528 	 * support NWAY, this is the register which will tell
529 	 * us the results of parallel detection.
530 	 */
531 	case RL_MEDIASTAT:
532 		rval = CSR_READ_1(sc, RL_MEDIASTAT);
533 		return (rval);
534 	default:
535 		device_printf(sc->rl_dev, "bad phy register\n");
536 		return (0);
537 	}
538 	rval = CSR_READ_2(sc, re8139_reg);
539 	if (sc->rl_type == RL_8139CPLUS && re8139_reg == RL_BMCR) {
540 		/* 8139C+ has different bit layout. */
541 		rval &= ~(BMCR_LOOP | BMCR_ISO);
542 	}
543 	return (rval);
544 }
545 
546 static int
547 re_miibus_writereg(device_t dev, int phy, int reg, int data)
548 {
549 	struct rl_softc		*sc;
550 	u_int16_t		re8139_reg = 0;
551 	int			rval = 0;
552 
553 	sc = device_get_softc(dev);
554 
555 	if (sc->rl_type == RL_8169) {
556 		rval = re_gmii_writereg(dev, phy, reg, data);
557 		return (rval);
558 	}
559 
560 	switch (reg) {
561 	case MII_BMCR:
562 		re8139_reg = RL_BMCR;
563 		if (sc->rl_type == RL_8139CPLUS) {
564 			/* 8139C+ has different bit layout. */
565 			data &= ~(BMCR_LOOP | BMCR_ISO);
566 		}
567 		break;
568 	case MII_BMSR:
569 		re8139_reg = RL_BMSR;
570 		break;
571 	case MII_ANAR:
572 		re8139_reg = RL_ANAR;
573 		break;
574 	case MII_ANER:
575 		re8139_reg = RL_ANER;
576 		break;
577 	case MII_ANLPAR:
578 		re8139_reg = RL_LPAR;
579 		break;
580 	case MII_PHYIDR1:
581 	case MII_PHYIDR2:
582 		return (0);
583 		break;
584 	default:
585 		device_printf(sc->rl_dev, "bad phy register\n");
586 		return (0);
587 	}
588 	CSR_WRITE_2(sc, re8139_reg, data);
589 	return (0);
590 }
591 
592 static void
593 re_miibus_statchg(device_t dev)
594 {
595 	struct rl_softc		*sc;
596 	struct ifnet		*ifp;
597 	struct mii_data		*mii;
598 
599 	sc = device_get_softc(dev);
600 	mii = device_get_softc(sc->rl_miibus);
601 	ifp = sc->rl_ifp;
602 	if (mii == NULL || ifp == NULL ||
603 	    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
604 		return;
605 
606 	sc->rl_flags &= ~RL_FLAG_LINK;
607 	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
608 	    (IFM_ACTIVE | IFM_AVALID)) {
609 		switch (IFM_SUBTYPE(mii->mii_media_active)) {
610 		case IFM_10_T:
611 		case IFM_100_TX:
612 			sc->rl_flags |= RL_FLAG_LINK;
613 			break;
614 		case IFM_1000_T:
615 			if ((sc->rl_flags & RL_FLAG_FASTETHER) != 0)
616 				break;
617 			sc->rl_flags |= RL_FLAG_LINK;
618 			break;
619 		default:
620 			break;
621 		}
622 	}
623 	/*
624 	 * RealTek controllers does not provide any interface to
625 	 * Tx/Rx MACs for resolved speed, duplex and flow-control
626 	 * parameters.
627 	 */
628 }
629 
630 /*
631  * Set the RX configuration and 64-bit multicast hash filter.
632  */
633 static void
634 re_set_rxmode(struct rl_softc *sc)
635 {
636 	struct ifnet		*ifp;
637 	struct ifmultiaddr	*ifma;
638 	uint32_t		hashes[2] = { 0, 0 };
639 	uint32_t		h, rxfilt;
640 
641 	RL_LOCK_ASSERT(sc);
642 
643 	ifp = sc->rl_ifp;
644 
645 	rxfilt = RL_RXCFG_CONFIG | RL_RXCFG_RX_INDIV | RL_RXCFG_RX_BROAD;
646 
647 	if (ifp->if_flags & (IFF_ALLMULTI | IFF_PROMISC)) {
648 		if (ifp->if_flags & IFF_PROMISC)
649 			rxfilt |= RL_RXCFG_RX_ALLPHYS;
650 		/*
651 		 * Unlike other hardwares, we have to explicitly set
652 		 * RL_RXCFG_RX_MULTI to receive multicast frames in
653 		 * promiscuous mode.
654 		 */
655 		rxfilt |= RL_RXCFG_RX_MULTI;
656 		hashes[0] = hashes[1] = 0xffffffff;
657 		goto done;
658 	}
659 
660 	if_maddr_rlock(ifp);
661 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
662 		if (ifma->ifma_addr->sa_family != AF_LINK)
663 			continue;
664 		h = ether_crc32_be(LLADDR((struct sockaddr_dl *)
665 		    ifma->ifma_addr), ETHER_ADDR_LEN) >> 26;
666 		if (h < 32)
667 			hashes[0] |= (1 << h);
668 		else
669 			hashes[1] |= (1 << (h - 32));
670 	}
671 	if_maddr_runlock(ifp);
672 
673 	if (hashes[0] != 0 || hashes[1] != 0) {
674 		/*
675 		 * For some unfathomable reason, RealTek decided to
676 		 * reverse the order of the multicast hash registers
677 		 * in the PCI Express parts.  This means we have to
678 		 * write the hash pattern in reverse order for those
679 		 * devices.
680 		 */
681 		if ((sc->rl_flags & RL_FLAG_PCIE) != 0) {
682 			h = bswap32(hashes[0]);
683 			hashes[0] = bswap32(hashes[1]);
684 			hashes[1] = h;
685 		}
686 		rxfilt |= RL_RXCFG_RX_MULTI;
687 	}
688 
689 done:
690 	CSR_WRITE_4(sc, RL_MAR0, hashes[0]);
691 	CSR_WRITE_4(sc, RL_MAR4, hashes[1]);
692 	CSR_WRITE_4(sc, RL_RXCFG, rxfilt);
693 }
694 
695 static void
696 re_reset(struct rl_softc *sc)
697 {
698 	int			i;
699 
700 	RL_LOCK_ASSERT(sc);
701 
702 	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RESET);
703 
704 	for (i = 0; i < RL_TIMEOUT; i++) {
705 		DELAY(10);
706 		if (!(CSR_READ_1(sc, RL_COMMAND) & RL_CMD_RESET))
707 			break;
708 	}
709 	if (i == RL_TIMEOUT)
710 		device_printf(sc->rl_dev, "reset never completed!\n");
711 
712 	if ((sc->rl_flags & RL_FLAG_MACRESET) != 0)
713 		CSR_WRITE_1(sc, 0x82, 1);
714 	if (sc->rl_hwrev->rl_rev == RL_HWREV_8169S)
715 		re_gmii_writereg(sc->rl_dev, 1, 0x0b, 0);
716 }
717 
718 #ifdef RE_DIAG
719 
720 /*
721  * The following routine is designed to test for a defect on some
722  * 32-bit 8169 cards. Some of these NICs have the REQ64# and ACK64#
723  * lines connected to the bus, however for a 32-bit only card, they
724  * should be pulled high. The result of this defect is that the
725  * NIC will not work right if you plug it into a 64-bit slot: DMA
726  * operations will be done with 64-bit transfers, which will fail
727  * because the 64-bit data lines aren't connected.
728  *
729  * There's no way to work around this (short of talking a soldering
730  * iron to the board), however we can detect it. The method we use
731  * here is to put the NIC into digital loopback mode, set the receiver
732  * to promiscuous mode, and then try to send a frame. We then compare
733  * the frame data we sent to what was received. If the data matches,
734  * then the NIC is working correctly, otherwise we know the user has
735  * a defective NIC which has been mistakenly plugged into a 64-bit PCI
736  * slot. In the latter case, there's no way the NIC can work correctly,
737  * so we print out a message on the console and abort the device attach.
738  */
739 
740 static int
741 re_diag(struct rl_softc *sc)
742 {
743 	struct ifnet		*ifp = sc->rl_ifp;
744 	struct mbuf		*m0;
745 	struct ether_header	*eh;
746 	struct rl_desc		*cur_rx;
747 	u_int16_t		status;
748 	u_int32_t		rxstat;
749 	int			total_len, i, error = 0, phyaddr;
750 	u_int8_t		dst[] = { 0x00, 'h', 'e', 'l', 'l', 'o' };
751 	u_int8_t		src[] = { 0x00, 'w', 'o', 'r', 'l', 'd' };
752 
753 	/* Allocate a single mbuf */
754 	MGETHDR(m0, M_DONTWAIT, MT_DATA);
755 	if (m0 == NULL)
756 		return (ENOBUFS);
757 
758 	RL_LOCK(sc);
759 
760 	/*
761 	 * Initialize the NIC in test mode. This sets the chip up
762 	 * so that it can send and receive frames, but performs the
763 	 * following special functions:
764 	 * - Puts receiver in promiscuous mode
765 	 * - Enables digital loopback mode
766 	 * - Leaves interrupts turned off
767 	 */
768 
769 	ifp->if_flags |= IFF_PROMISC;
770 	sc->rl_testmode = 1;
771 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
772 	re_init_locked(sc);
773 	sc->rl_flags |= RL_FLAG_LINK;
774 	if (sc->rl_type == RL_8169)
775 		phyaddr = 1;
776 	else
777 		phyaddr = 0;
778 
779 	re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_RESET);
780 	for (i = 0; i < RL_TIMEOUT; i++) {
781 		status = re_miibus_readreg(sc->rl_dev, phyaddr, MII_BMCR);
782 		if (!(status & BMCR_RESET))
783 			break;
784 	}
785 
786 	re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_LOOP);
787 	CSR_WRITE_2(sc, RL_ISR, RL_INTRS);
788 
789 	DELAY(100000);
790 
791 	/* Put some data in the mbuf */
792 
793 	eh = mtod(m0, struct ether_header *);
794 	bcopy ((char *)&dst, eh->ether_dhost, ETHER_ADDR_LEN);
795 	bcopy ((char *)&src, eh->ether_shost, ETHER_ADDR_LEN);
796 	eh->ether_type = htons(ETHERTYPE_IP);
797 	m0->m_pkthdr.len = m0->m_len = ETHER_MIN_LEN - ETHER_CRC_LEN;
798 
799 	/*
800 	 * Queue the packet, start transmission.
801 	 * Note: IF_HANDOFF() ultimately calls re_start() for us.
802 	 */
803 
804 	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
805 	RL_UNLOCK(sc);
806 	/* XXX: re_diag must not be called when in ALTQ mode */
807 	IF_HANDOFF(&ifp->if_snd, m0, ifp);
808 	RL_LOCK(sc);
809 	m0 = NULL;
810 
811 	/* Wait for it to propagate through the chip */
812 
813 	DELAY(100000);
814 	for (i = 0; i < RL_TIMEOUT; i++) {
815 		status = CSR_READ_2(sc, RL_ISR);
816 		CSR_WRITE_2(sc, RL_ISR, status);
817 		if ((status & (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK)) ==
818 		    (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK))
819 			break;
820 		DELAY(10);
821 	}
822 
823 	if (i == RL_TIMEOUT) {
824 		device_printf(sc->rl_dev,
825 		    "diagnostic failed, failed to receive packet in"
826 		    " loopback mode\n");
827 		error = EIO;
828 		goto done;
829 	}
830 
831 	/*
832 	 * The packet should have been dumped into the first
833 	 * entry in the RX DMA ring. Grab it from there.
834 	 */
835 
836 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
837 	    sc->rl_ldata.rl_rx_list_map,
838 	    BUS_DMASYNC_POSTREAD);
839 	bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag,
840 	    sc->rl_ldata.rl_rx_desc[0].rx_dmamap,
841 	    BUS_DMASYNC_POSTREAD);
842 	bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag,
843 	    sc->rl_ldata.rl_rx_desc[0].rx_dmamap);
844 
845 	m0 = sc->rl_ldata.rl_rx_desc[0].rx_m;
846 	sc->rl_ldata.rl_rx_desc[0].rx_m = NULL;
847 	eh = mtod(m0, struct ether_header *);
848 
849 	cur_rx = &sc->rl_ldata.rl_rx_list[0];
850 	total_len = RL_RXBYTES(cur_rx);
851 	rxstat = le32toh(cur_rx->rl_cmdstat);
852 
853 	if (total_len != ETHER_MIN_LEN) {
854 		device_printf(sc->rl_dev,
855 		    "diagnostic failed, received short packet\n");
856 		error = EIO;
857 		goto done;
858 	}
859 
860 	/* Test that the received packet data matches what we sent. */
861 
862 	if (bcmp((char *)&eh->ether_dhost, (char *)&dst, ETHER_ADDR_LEN) ||
863 	    bcmp((char *)&eh->ether_shost, (char *)&src, ETHER_ADDR_LEN) ||
864 	    ntohs(eh->ether_type) != ETHERTYPE_IP) {
865 		device_printf(sc->rl_dev, "WARNING, DMA FAILURE!\n");
866 		device_printf(sc->rl_dev, "expected TX data: %6D/%6D/0x%x\n",
867 		    dst, ":", src, ":", ETHERTYPE_IP);
868 		device_printf(sc->rl_dev, "received RX data: %6D/%6D/0x%x\n",
869 		    eh->ether_dhost, ":",  eh->ether_shost, ":",
870 		    ntohs(eh->ether_type));
871 		device_printf(sc->rl_dev, "You may have a defective 32-bit "
872 		    "NIC plugged into a 64-bit PCI slot.\n");
873 		device_printf(sc->rl_dev, "Please re-install the NIC in a "
874 		    "32-bit slot for proper operation.\n");
875 		device_printf(sc->rl_dev, "Read the re(4) man page for more "
876 		    "details.\n");
877 		error = EIO;
878 	}
879 
880 done:
881 	/* Turn interface off, release resources */
882 
883 	sc->rl_testmode = 0;
884 	sc->rl_flags &= ~RL_FLAG_LINK;
885 	ifp->if_flags &= ~IFF_PROMISC;
886 	re_stop(sc);
887 	if (m0 != NULL)
888 		m_freem(m0);
889 
890 	RL_UNLOCK(sc);
891 
892 	return (error);
893 }
894 
895 #endif
896 
897 /*
898  * Probe for a RealTek 8139C+/8169/8110 chip. Check the PCI vendor and device
899  * IDs against our list and return a device name if we find a match.
900  */
901 static int
902 re_probe(device_t dev)
903 {
904 	struct rl_type		*t;
905 	uint16_t		devid, vendor;
906 	uint16_t		revid, sdevid;
907 	int			i;
908 
909 	vendor = pci_get_vendor(dev);
910 	devid = pci_get_device(dev);
911 	revid = pci_get_revid(dev);
912 	sdevid = pci_get_subdevice(dev);
913 
914 	if (vendor == LINKSYS_VENDORID && devid == LINKSYS_DEVICEID_EG1032) {
915 		if (sdevid != LINKSYS_SUBDEVICE_EG1032_REV3) {
916 			/*
917 			 * Only attach to rev. 3 of the Linksys EG1032 adapter.
918 			 * Rev. 2 is supported by sk(4).
919 			 */
920 			return (ENXIO);
921 		}
922 	}
923 
924 	if (vendor == RT_VENDORID && devid == RT_DEVICEID_8139) {
925 		if (revid != 0x20) {
926 			/* 8139, let rl(4) take care of this device. */
927 			return (ENXIO);
928 		}
929 	}
930 
931 	t = re_devs;
932 	for (i = 0; i < sizeof(re_devs) / sizeof(re_devs[0]); i++, t++) {
933 		if (vendor == t->rl_vid && devid == t->rl_did) {
934 			device_set_desc(dev, t->rl_name);
935 			return (BUS_PROBE_DEFAULT);
936 		}
937 	}
938 
939 	return (ENXIO);
940 }
941 
942 /*
943  * Map a single buffer address.
944  */
945 
946 static void
947 re_dma_map_addr(void *arg, bus_dma_segment_t *segs, int nseg, int error)
948 {
949 	bus_addr_t		*addr;
950 
951 	if (error)
952 		return;
953 
954 	KASSERT(nseg == 1, ("too many DMA segments, %d should be 1", nseg));
955 	addr = arg;
956 	*addr = segs->ds_addr;
957 }
958 
959 static int
960 re_allocmem(device_t dev, struct rl_softc *sc)
961 {
962 	bus_addr_t		lowaddr;
963 	bus_size_t		rx_list_size, tx_list_size;
964 	int			error;
965 	int			i;
966 
967 	rx_list_size = sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc);
968 	tx_list_size = sc->rl_ldata.rl_tx_desc_cnt * sizeof(struct rl_desc);
969 
970 	/*
971 	 * Allocate the parent bus DMA tag appropriate for PCI.
972 	 * In order to use DAC, RL_CPLUSCMD_PCI_DAC bit of RL_CPLUS_CMD
973 	 * register should be set. However some RealTek chips are known
974 	 * to be buggy on DAC handling, therefore disable DAC by limiting
975 	 * DMA address space to 32bit. PCIe variants of RealTek chips
976 	 * may not have the limitation.
977 	 */
978 	lowaddr = BUS_SPACE_MAXADDR;
979 	if ((sc->rl_flags & RL_FLAG_PCIE) == 0)
980 		lowaddr = BUS_SPACE_MAXADDR_32BIT;
981 	error = bus_dma_tag_create(bus_get_dma_tag(dev), 1, 0,
982 	    lowaddr, BUS_SPACE_MAXADDR, NULL, NULL,
983 	    BUS_SPACE_MAXSIZE_32BIT, 0, BUS_SPACE_MAXSIZE_32BIT, 0,
984 	    NULL, NULL, &sc->rl_parent_tag);
985 	if (error) {
986 		device_printf(dev, "could not allocate parent DMA tag\n");
987 		return (error);
988 	}
989 
990 	/*
991 	 * Allocate map for TX mbufs.
992 	 */
993 	error = bus_dma_tag_create(sc->rl_parent_tag, 1, 0,
994 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL,
995 	    NULL, MCLBYTES * RL_NTXSEGS, RL_NTXSEGS, 4096, 0,
996 	    NULL, NULL, &sc->rl_ldata.rl_tx_mtag);
997 	if (error) {
998 		device_printf(dev, "could not allocate TX DMA tag\n");
999 		return (error);
1000 	}
1001 
1002 	/*
1003 	 * Allocate map for RX mbufs.
1004 	 */
1005 
1006 	if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
1007 		error = bus_dma_tag_create(sc->rl_parent_tag, sizeof(uint64_t),
1008 		    0, BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
1009 		    MJUM9BYTES, 1, MJUM9BYTES, 0, NULL, NULL,
1010 		    &sc->rl_ldata.rl_jrx_mtag);
1011 		if (error) {
1012 			device_printf(dev,
1013 			    "could not allocate jumbo RX DMA tag\n");
1014 			return (error);
1015 		}
1016 	}
1017 	error = bus_dma_tag_create(sc->rl_parent_tag, sizeof(uint64_t), 0,
1018 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
1019 	    MCLBYTES, 1, MCLBYTES, 0, NULL, NULL, &sc->rl_ldata.rl_rx_mtag);
1020 	if (error) {
1021 		device_printf(dev, "could not allocate RX DMA tag\n");
1022 		return (error);
1023 	}
1024 
1025 	/*
1026 	 * Allocate map for TX descriptor list.
1027 	 */
1028 	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
1029 	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1030 	    NULL, tx_list_size, 1, tx_list_size, 0,
1031 	    NULL, NULL, &sc->rl_ldata.rl_tx_list_tag);
1032 	if (error) {
1033 		device_printf(dev, "could not allocate TX DMA ring tag\n");
1034 		return (error);
1035 	}
1036 
1037 	/* Allocate DMA'able memory for the TX ring */
1038 
1039 	error = bus_dmamem_alloc(sc->rl_ldata.rl_tx_list_tag,
1040 	    (void **)&sc->rl_ldata.rl_tx_list,
1041 	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1042 	    &sc->rl_ldata.rl_tx_list_map);
1043 	if (error) {
1044 		device_printf(dev, "could not allocate TX DMA ring\n");
1045 		return (error);
1046 	}
1047 
1048 	/* Load the map for the TX ring. */
1049 
1050 	sc->rl_ldata.rl_tx_list_addr = 0;
1051 	error = bus_dmamap_load(sc->rl_ldata.rl_tx_list_tag,
1052 	     sc->rl_ldata.rl_tx_list_map, sc->rl_ldata.rl_tx_list,
1053 	     tx_list_size, re_dma_map_addr,
1054 	     &sc->rl_ldata.rl_tx_list_addr, BUS_DMA_NOWAIT);
1055 	if (error != 0 || sc->rl_ldata.rl_tx_list_addr == 0) {
1056 		device_printf(dev, "could not load TX DMA ring\n");
1057 		return (ENOMEM);
1058 	}
1059 
1060 	/* Create DMA maps for TX buffers */
1061 
1062 	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
1063 		error = bus_dmamap_create(sc->rl_ldata.rl_tx_mtag, 0,
1064 		    &sc->rl_ldata.rl_tx_desc[i].tx_dmamap);
1065 		if (error) {
1066 			device_printf(dev, "could not create DMA map for TX\n");
1067 			return (error);
1068 		}
1069 	}
1070 
1071 	/*
1072 	 * Allocate map for RX descriptor list.
1073 	 */
1074 	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
1075 	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1076 	    NULL, rx_list_size, 1, rx_list_size, 0,
1077 	    NULL, NULL, &sc->rl_ldata.rl_rx_list_tag);
1078 	if (error) {
1079 		device_printf(dev, "could not create RX DMA ring tag\n");
1080 		return (error);
1081 	}
1082 
1083 	/* Allocate DMA'able memory for the RX ring */
1084 
1085 	error = bus_dmamem_alloc(sc->rl_ldata.rl_rx_list_tag,
1086 	    (void **)&sc->rl_ldata.rl_rx_list,
1087 	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1088 	    &sc->rl_ldata.rl_rx_list_map);
1089 	if (error) {
1090 		device_printf(dev, "could not allocate RX DMA ring\n");
1091 		return (error);
1092 	}
1093 
1094 	/* Load the map for the RX ring. */
1095 
1096 	sc->rl_ldata.rl_rx_list_addr = 0;
1097 	error = bus_dmamap_load(sc->rl_ldata.rl_rx_list_tag,
1098 	     sc->rl_ldata.rl_rx_list_map, sc->rl_ldata.rl_rx_list,
1099 	     rx_list_size, re_dma_map_addr,
1100 	     &sc->rl_ldata.rl_rx_list_addr, BUS_DMA_NOWAIT);
1101 	if (error != 0 || sc->rl_ldata.rl_rx_list_addr == 0) {
1102 		device_printf(dev, "could not load RX DMA ring\n");
1103 		return (ENOMEM);
1104 	}
1105 
1106 	/* Create DMA maps for RX buffers */
1107 
1108 	if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
1109 		error = bus_dmamap_create(sc->rl_ldata.rl_jrx_mtag, 0,
1110 		    &sc->rl_ldata.rl_jrx_sparemap);
1111 		if (error) {
1112 			device_printf(dev,
1113 			    "could not create spare DMA map for jumbo RX\n");
1114 			return (error);
1115 		}
1116 		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1117 			error = bus_dmamap_create(sc->rl_ldata.rl_jrx_mtag, 0,
1118 			    &sc->rl_ldata.rl_jrx_desc[i].rx_dmamap);
1119 			if (error) {
1120 				device_printf(dev,
1121 				    "could not create DMA map for jumbo RX\n");
1122 				return (error);
1123 			}
1124 		}
1125 	}
1126 	error = bus_dmamap_create(sc->rl_ldata.rl_rx_mtag, 0,
1127 	    &sc->rl_ldata.rl_rx_sparemap);
1128 	if (error) {
1129 		device_printf(dev, "could not create spare DMA map for RX\n");
1130 		return (error);
1131 	}
1132 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1133 		error = bus_dmamap_create(sc->rl_ldata.rl_rx_mtag, 0,
1134 		    &sc->rl_ldata.rl_rx_desc[i].rx_dmamap);
1135 		if (error) {
1136 			device_printf(dev, "could not create DMA map for RX\n");
1137 			return (error);
1138 		}
1139 	}
1140 
1141 	/* Create DMA map for statistics. */
1142 	error = bus_dma_tag_create(sc->rl_parent_tag, RL_DUMP_ALIGN, 0,
1143 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
1144 	    sizeof(struct rl_stats), 1, sizeof(struct rl_stats), 0, NULL, NULL,
1145 	    &sc->rl_ldata.rl_stag);
1146 	if (error) {
1147 		device_printf(dev, "could not create statistics DMA tag\n");
1148 		return (error);
1149 	}
1150 	/* Allocate DMA'able memory for statistics. */
1151 	error = bus_dmamem_alloc(sc->rl_ldata.rl_stag,
1152 	    (void **)&sc->rl_ldata.rl_stats,
1153 	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1154 	    &sc->rl_ldata.rl_smap);
1155 	if (error) {
1156 		device_printf(dev,
1157 		    "could not allocate statistics DMA memory\n");
1158 		return (error);
1159 	}
1160 	/* Load the map for statistics. */
1161 	sc->rl_ldata.rl_stats_addr = 0;
1162 	error = bus_dmamap_load(sc->rl_ldata.rl_stag, sc->rl_ldata.rl_smap,
1163 	    sc->rl_ldata.rl_stats, sizeof(struct rl_stats), re_dma_map_addr,
1164 	     &sc->rl_ldata.rl_stats_addr, BUS_DMA_NOWAIT);
1165 	if (error != 0 || sc->rl_ldata.rl_stats_addr == 0) {
1166 		device_printf(dev, "could not load statistics DMA memory\n");
1167 		return (ENOMEM);
1168 	}
1169 
1170 	return (0);
1171 }
1172 
1173 /*
1174  * Attach the interface. Allocate softc structures, do ifmedia
1175  * setup and ethernet/BPF attach.
1176  */
1177 static int
1178 re_attach(device_t dev)
1179 {
1180 	u_char			eaddr[ETHER_ADDR_LEN];
1181 	u_int16_t		as[ETHER_ADDR_LEN / 2];
1182 	struct rl_softc		*sc;
1183 	struct ifnet		*ifp;
1184 	struct rl_hwrev		*hw_rev;
1185 	int			hwrev;
1186 	u_int16_t		devid, re_did = 0;
1187 	int			error = 0, i, phy, rid;
1188 	int			msic, msixc, reg;
1189 	uint8_t			cfg;
1190 
1191 	sc = device_get_softc(dev);
1192 	sc->rl_dev = dev;
1193 
1194 	mtx_init(&sc->rl_mtx, device_get_nameunit(dev), MTX_NETWORK_LOCK,
1195 	    MTX_DEF);
1196 	callout_init_mtx(&sc->rl_stat_callout, &sc->rl_mtx, 0);
1197 
1198 	/*
1199 	 * Map control/status registers.
1200 	 */
1201 	pci_enable_busmaster(dev);
1202 
1203 	devid = pci_get_device(dev);
1204 	/*
1205 	 * Prefer memory space register mapping over IO space.
1206 	 * Because RTL8169SC does not seem to work when memory mapping
1207 	 * is used always activate io mapping.
1208 	 */
1209 	if (devid == RT_DEVICEID_8169SC)
1210 		prefer_iomap = 1;
1211 	if (prefer_iomap == 0) {
1212 		sc->rl_res_id = PCIR_BAR(1);
1213 		sc->rl_res_type = SYS_RES_MEMORY;
1214 		/* RTL8168/8101E seems to use different BARs. */
1215 		if (devid == RT_DEVICEID_8168 || devid == RT_DEVICEID_8101E)
1216 			sc->rl_res_id = PCIR_BAR(2);
1217 	} else {
1218 		sc->rl_res_id = PCIR_BAR(0);
1219 		sc->rl_res_type = SYS_RES_IOPORT;
1220 	}
1221 	sc->rl_res = bus_alloc_resource_any(dev, sc->rl_res_type,
1222 	    &sc->rl_res_id, RF_ACTIVE);
1223 	if (sc->rl_res == NULL && prefer_iomap == 0) {
1224 		sc->rl_res_id = PCIR_BAR(0);
1225 		sc->rl_res_type = SYS_RES_IOPORT;
1226 		sc->rl_res = bus_alloc_resource_any(dev, sc->rl_res_type,
1227 		    &sc->rl_res_id, RF_ACTIVE);
1228 	}
1229 	if (sc->rl_res == NULL) {
1230 		device_printf(dev, "couldn't map ports/memory\n");
1231 		error = ENXIO;
1232 		goto fail;
1233 	}
1234 
1235 	sc->rl_btag = rman_get_bustag(sc->rl_res);
1236 	sc->rl_bhandle = rman_get_bushandle(sc->rl_res);
1237 
1238 	msic = pci_msi_count(dev);
1239 	msixc = pci_msix_count(dev);
1240 	if (pci_find_cap(dev, PCIY_EXPRESS, &reg) == 0)
1241 		sc->rl_flags |= RL_FLAG_PCIE;
1242 	if (bootverbose) {
1243 		device_printf(dev, "MSI count : %d\n", msic);
1244 		device_printf(dev, "MSI-X count : %d\n", msixc);
1245 	}
1246 	if (msix_disable > 0)
1247 		msixc = 0;
1248 	if (msi_disable > 0)
1249 		msic = 0;
1250 	/* Prefer MSI-X to MSI. */
1251 	if (msixc > 0) {
1252 		msixc = 1;
1253 		rid = PCIR_BAR(4);
1254 		sc->rl_res_pba = bus_alloc_resource_any(dev, SYS_RES_MEMORY,
1255 		    &rid, RF_ACTIVE);
1256 		if (sc->rl_res_pba == NULL) {
1257 			device_printf(sc->rl_dev,
1258 			    "could not allocate MSI-X PBA resource\n");
1259 		}
1260 		if (sc->rl_res_pba != NULL &&
1261 		    pci_alloc_msix(dev, &msixc) == 0) {
1262 			if (msixc == 1) {
1263 				device_printf(dev, "Using %d MSI-X message\n",
1264 				    msixc);
1265 				sc->rl_flags |= RL_FLAG_MSIX;
1266 			} else
1267 				pci_release_msi(dev);
1268 		}
1269 		if ((sc->rl_flags & RL_FLAG_MSIX) == 0) {
1270 			if (sc->rl_res_pba != NULL)
1271 				bus_release_resource(dev, SYS_RES_MEMORY, rid,
1272 				    sc->rl_res_pba);
1273 			sc->rl_res_pba = NULL;
1274 			msixc = 0;
1275 		}
1276 	}
1277 	/* Prefer MSI to INTx. */
1278 	if (msixc == 0 && msic > 0) {
1279 		msic = 1;
1280 		if (pci_alloc_msi(dev, &msic) == 0) {
1281 			if (msic == RL_MSI_MESSAGES) {
1282 				device_printf(dev, "Using %d MSI message\n",
1283 				    msic);
1284 				sc->rl_flags |= RL_FLAG_MSI;
1285 				/* Explicitly set MSI enable bit. */
1286 				CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1287 				cfg = CSR_READ_1(sc, RL_CFG2);
1288 				cfg |= RL_CFG2_MSI;
1289 				CSR_WRITE_1(sc, RL_CFG2, cfg);
1290 				CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1291 			} else
1292 				pci_release_msi(dev);
1293 		}
1294 		if ((sc->rl_flags & RL_FLAG_MSI) == 0)
1295 			msic = 0;
1296 	}
1297 
1298 	/* Allocate interrupt */
1299 	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) == 0) {
1300 		rid = 0;
1301 		sc->rl_irq[0] = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
1302 		    RF_SHAREABLE | RF_ACTIVE);
1303 		if (sc->rl_irq[0] == NULL) {
1304 			device_printf(dev, "couldn't allocate IRQ resources\n");
1305 			error = ENXIO;
1306 			goto fail;
1307 		}
1308 	} else {
1309 		for (i = 0, rid = 1; i < RL_MSI_MESSAGES; i++, rid++) {
1310 			sc->rl_irq[i] = bus_alloc_resource_any(dev,
1311 			    SYS_RES_IRQ, &rid, RF_ACTIVE);
1312 			if (sc->rl_irq[i] == NULL) {
1313 				device_printf(dev,
1314 				    "couldn't llocate IRQ resources for "
1315 				    "message %d\n", rid);
1316 				error = ENXIO;
1317 				goto fail;
1318 			}
1319 		}
1320 	}
1321 
1322 	if ((sc->rl_flags & RL_FLAG_MSI) == 0) {
1323 		CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1324 		cfg = CSR_READ_1(sc, RL_CFG2);
1325 		if ((cfg & RL_CFG2_MSI) != 0) {
1326 			device_printf(dev, "turning off MSI enable bit.\n");
1327 			cfg &= ~RL_CFG2_MSI;
1328 			CSR_WRITE_1(sc, RL_CFG2, cfg);
1329 		}
1330 		CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1331 	}
1332 
1333 	hw_rev = re_hwrevs;
1334 	hwrev = CSR_READ_4(sc, RL_TXCFG);
1335 	switch (hwrev & 0x70000000) {
1336 	case 0x00000000:
1337 	case 0x10000000:
1338 		device_printf(dev, "Chip rev. 0x%08x\n", hwrev & 0xfc800000);
1339 		hwrev &= (RL_TXCFG_HWREV | 0x80000000);
1340 		break;
1341 	default:
1342 		device_printf(dev, "Chip rev. 0x%08x\n", hwrev & 0x7c800000);
1343 		hwrev &= RL_TXCFG_HWREV;
1344 		break;
1345 	}
1346 	device_printf(dev, "MAC rev. 0x%08x\n", hwrev & 0x00700000);
1347 	while (hw_rev->rl_desc != NULL) {
1348 		if (hw_rev->rl_rev == hwrev) {
1349 			sc->rl_type = hw_rev->rl_type;
1350 			sc->rl_hwrev = hw_rev;
1351 			break;
1352 		}
1353 		hw_rev++;
1354 	}
1355 	if (hw_rev->rl_desc == NULL) {
1356 		device_printf(dev, "Unknown H/W revision: 0x%08x\n", hwrev);
1357 		error = ENXIO;
1358 		goto fail;
1359 	}
1360 
1361 	switch (hw_rev->rl_rev) {
1362 	case RL_HWREV_8139CPLUS:
1363 		sc->rl_flags |= RL_FLAG_FASTETHER | RL_FLAG_AUTOPAD;
1364 		break;
1365 	case RL_HWREV_8100E:
1366 	case RL_HWREV_8101E:
1367 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_FASTETHER;
1368 		break;
1369 	case RL_HWREV_8102E:
1370 	case RL_HWREV_8102EL:
1371 	case RL_HWREV_8102EL_SPIN1:
1372 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR | RL_FLAG_DESCV2 |
1373 		    RL_FLAG_MACSTAT | RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP |
1374 		    RL_FLAG_AUTOPAD;
1375 		break;
1376 	case RL_HWREV_8103E:
1377 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR | RL_FLAG_DESCV2 |
1378 		    RL_FLAG_MACSTAT | RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP |
1379 		    RL_FLAG_AUTOPAD | RL_FLAG_MACSLEEP;
1380 		break;
1381 	case RL_HWREV_8401E:
1382 	case RL_HWREV_8105E:
1383 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
1384 		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
1385 		    RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD;
1386 		break;
1387 	case RL_HWREV_8168B_SPIN1:
1388 	case RL_HWREV_8168B_SPIN2:
1389 		sc->rl_flags |= RL_FLAG_WOLRXENB;
1390 		/* FALLTHROUGH */
1391 	case RL_HWREV_8168B_SPIN3:
1392 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_MACSTAT;
1393 		break;
1394 	case RL_HWREV_8168C_SPIN2:
1395 		sc->rl_flags |= RL_FLAG_MACSLEEP;
1396 		/* FALLTHROUGH */
1397 	case RL_HWREV_8168C:
1398 		if ((hwrev & 0x00700000) == 0x00200000)
1399 			sc->rl_flags |= RL_FLAG_MACSLEEP;
1400 		/* FALLTHROUGH */
1401 	case RL_HWREV_8168CP:
1402 	case RL_HWREV_8168D:
1403 	case RL_HWREV_8168DP:
1404 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
1405 		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP |
1406 		    RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2;
1407 		break;
1408 	case RL_HWREV_8168E:
1409 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
1410 		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
1411 		    RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2;
1412 		break;
1413 	case RL_HWREV_8168E_VL:
1414 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
1415 		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP |
1416 		    RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2;
1417 		break;
1418 	case RL_HWREV_8169_8110SB:
1419 	case RL_HWREV_8169_8110SBL:
1420 	case RL_HWREV_8169_8110SC:
1421 	case RL_HWREV_8169_8110SCE:
1422 		sc->rl_flags |= RL_FLAG_PHYWAKE;
1423 		/* FALLTHROUGH */
1424 	case RL_HWREV_8169:
1425 	case RL_HWREV_8169S:
1426 	case RL_HWREV_8110S:
1427 		sc->rl_flags |= RL_FLAG_MACRESET;
1428 		break;
1429 	default:
1430 		break;
1431 	}
1432 
1433 	/* Reset the adapter. */
1434 	RL_LOCK(sc);
1435 	re_reset(sc);
1436 	RL_UNLOCK(sc);
1437 
1438 	/* Enable PME. */
1439 	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1440 	cfg = CSR_READ_1(sc, RL_CFG1);
1441 	cfg |= RL_CFG1_PME;
1442 	CSR_WRITE_1(sc, RL_CFG1, cfg);
1443 	cfg = CSR_READ_1(sc, RL_CFG5);
1444 	cfg &= RL_CFG5_PME_STS;
1445 	CSR_WRITE_1(sc, RL_CFG5, cfg);
1446 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1447 
1448 	if ((sc->rl_flags & RL_FLAG_PAR) != 0) {
1449 		/*
1450 		 * XXX Should have a better way to extract station
1451 		 * address from EEPROM.
1452 		 */
1453 		for (i = 0; i < ETHER_ADDR_LEN; i++)
1454 			eaddr[i] = CSR_READ_1(sc, RL_IDR0 + i);
1455 	} else {
1456 		sc->rl_eewidth = RL_9356_ADDR_LEN;
1457 		re_read_eeprom(sc, (caddr_t)&re_did, 0, 1);
1458 		if (re_did != 0x8129)
1459 			sc->rl_eewidth = RL_9346_ADDR_LEN;
1460 
1461 		/*
1462 		 * Get station address from the EEPROM.
1463 		 */
1464 		re_read_eeprom(sc, (caddr_t)as, RL_EE_EADDR, 3);
1465 		for (i = 0; i < ETHER_ADDR_LEN / 2; i++)
1466 			as[i] = le16toh(as[i]);
1467 		bcopy(as, eaddr, sizeof(eaddr));
1468 	}
1469 
1470 	if (sc->rl_type == RL_8169) {
1471 		/* Set RX length mask and number of descriptors. */
1472 		sc->rl_rxlenmask = RL_RDESC_STAT_GFRAGLEN;
1473 		sc->rl_txstart = RL_GTXSTART;
1474 		sc->rl_ldata.rl_tx_desc_cnt = RL_8169_TX_DESC_CNT;
1475 		sc->rl_ldata.rl_rx_desc_cnt = RL_8169_RX_DESC_CNT;
1476 	} else {
1477 		/* Set RX length mask and number of descriptors. */
1478 		sc->rl_rxlenmask = RL_RDESC_STAT_FRAGLEN;
1479 		sc->rl_txstart = RL_TXSTART;
1480 		sc->rl_ldata.rl_tx_desc_cnt = RL_8139_TX_DESC_CNT;
1481 		sc->rl_ldata.rl_rx_desc_cnt = RL_8139_RX_DESC_CNT;
1482 	}
1483 
1484 	error = re_allocmem(dev, sc);
1485 	if (error)
1486 		goto fail;
1487 	re_add_sysctls(sc);
1488 
1489 	ifp = sc->rl_ifp = if_alloc(IFT_ETHER);
1490 	if (ifp == NULL) {
1491 		device_printf(dev, "can not if_alloc()\n");
1492 		error = ENOSPC;
1493 		goto fail;
1494 	}
1495 
1496 	/* Take controller out of deep sleep mode. */
1497 	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
1498 		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
1499 			CSR_WRITE_1(sc, RL_GPIO,
1500 			    CSR_READ_1(sc, RL_GPIO) | 0x01);
1501 		else
1502 			CSR_WRITE_1(sc, RL_GPIO,
1503 			    CSR_READ_1(sc, RL_GPIO) & ~0x01);
1504 	}
1505 
1506 	/* Take PHY out of power down mode. */
1507 	if ((sc->rl_flags & RL_FLAG_PHYWAKE_PM) != 0) {
1508 		CSR_WRITE_1(sc, RL_PMCH, CSR_READ_1(sc, RL_PMCH) | 0x80);
1509 		if (hw_rev->rl_rev == RL_HWREV_8401E)
1510 			CSR_WRITE_1(sc, 0xD1, CSR_READ_1(sc, 0xD1) & ~0x08);
1511 	}
1512 	if ((sc->rl_flags & RL_FLAG_PHYWAKE) != 0) {
1513 		re_gmii_writereg(dev, 1, 0x1f, 0);
1514 		re_gmii_writereg(dev, 1, 0x0e, 0);
1515 	}
1516 
1517 #define	RE_PHYAD_INTERNAL	 0
1518 
1519 	/* Do MII setup. */
1520 	phy = RE_PHYAD_INTERNAL;
1521 	if (sc->rl_type == RL_8169)
1522 		phy = 1;
1523 	error = mii_attach(dev, &sc->rl_miibus, ifp, re_ifmedia_upd,
1524 	    re_ifmedia_sts, BMSR_DEFCAPMASK, phy, MII_OFFSET_ANY, MIIF_DOPAUSE);
1525 	if (error != 0) {
1526 		device_printf(dev, "attaching PHYs failed\n");
1527 		goto fail;
1528 	}
1529 
1530 	ifp->if_softc = sc;
1531 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1532 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1533 	ifp->if_ioctl = re_ioctl;
1534 	ifp->if_start = re_start;
1535 	/*
1536 	 * RTL8168/8111C generates wrong IP checksummed frame if the
1537 	 * packet has IP options so disable TX IP checksum offloading.
1538 	 */
1539 	if (sc->rl_hwrev->rl_rev == RL_HWREV_8168C ||
1540 	    sc->rl_hwrev->rl_rev == RL_HWREV_8168C_SPIN2)
1541 		ifp->if_hwassist = CSUM_TCP | CSUM_UDP;
1542 	else
1543 		ifp->if_hwassist = CSUM_IP | CSUM_TCP | CSUM_UDP;
1544 	ifp->if_hwassist |= CSUM_TSO;
1545 	ifp->if_capabilities = IFCAP_HWCSUM | IFCAP_TSO4;
1546 	ifp->if_capenable = ifp->if_capabilities;
1547 	ifp->if_init = re_init;
1548 	IFQ_SET_MAXLEN(&ifp->if_snd, RL_IFQ_MAXLEN);
1549 	ifp->if_snd.ifq_drv_maxlen = RL_IFQ_MAXLEN;
1550 	IFQ_SET_READY(&ifp->if_snd);
1551 
1552 	TASK_INIT(&sc->rl_inttask, 0, re_int_task, sc);
1553 
1554 	/*
1555 	 * Call MI attach routine.
1556 	 */
1557 	ether_ifattach(ifp, eaddr);
1558 
1559 	/* VLAN capability setup */
1560 	ifp->if_capabilities |= IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING;
1561 	if (ifp->if_capabilities & IFCAP_HWCSUM)
1562 		ifp->if_capabilities |= IFCAP_VLAN_HWCSUM;
1563 	/* Enable WOL if PM is supported. */
1564 	if (pci_find_cap(sc->rl_dev, PCIY_PMG, &reg) == 0)
1565 		ifp->if_capabilities |= IFCAP_WOL;
1566 	ifp->if_capenable = ifp->if_capabilities;
1567 	/*
1568 	 * Don't enable TSO by default.  It is known to generate
1569 	 * corrupted TCP segments(bad TCP options) under certain
1570 	 * circumtances.
1571 	 */
1572 	ifp->if_hwassist &= ~CSUM_TSO;
1573 	ifp->if_capenable &= ~(IFCAP_TSO4 | IFCAP_VLAN_HWTSO);
1574 #ifdef DEVICE_POLLING
1575 	ifp->if_capabilities |= IFCAP_POLLING;
1576 #endif
1577 	/*
1578 	 * Tell the upper layer(s) we support long frames.
1579 	 * Must appear after the call to ether_ifattach() because
1580 	 * ether_ifattach() sets ifi_hdrlen to the default value.
1581 	 */
1582 	ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
1583 
1584 #ifdef RE_DIAG
1585 	/*
1586 	 * Perform hardware diagnostic on the original RTL8169.
1587 	 * Some 32-bit cards were incorrectly wired and would
1588 	 * malfunction if plugged into a 64-bit slot.
1589 	 */
1590 
1591 	if (hwrev == RL_HWREV_8169) {
1592 		error = re_diag(sc);
1593 		if (error) {
1594 			device_printf(dev,
1595 		    	"attach aborted due to hardware diag failure\n");
1596 			ether_ifdetach(ifp);
1597 			goto fail;
1598 		}
1599 	}
1600 #endif
1601 
1602 #ifdef RE_TX_MODERATION
1603 	intr_filter = 1;
1604 #endif
1605 	/* Hook interrupt last to avoid having to lock softc */
1606 	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) != 0 &&
1607 	    intr_filter == 0) {
1608 		error = bus_setup_intr(dev, sc->rl_irq[0],
1609 		    INTR_TYPE_NET | INTR_MPSAFE, NULL, re_intr_msi, sc,
1610 		    &sc->rl_intrhand[0]);
1611 	} else {
1612 		error = bus_setup_intr(dev, sc->rl_irq[0],
1613 		    INTR_TYPE_NET | INTR_MPSAFE, re_intr, NULL, sc,
1614 		    &sc->rl_intrhand[0]);
1615 	}
1616 	if (error) {
1617 		device_printf(dev, "couldn't set up irq\n");
1618 		ether_ifdetach(ifp);
1619 	}
1620 
1621 fail:
1622 
1623 	if (error)
1624 		re_detach(dev);
1625 
1626 	return (error);
1627 }
1628 
1629 /*
1630  * Shutdown hardware and free up resources. This can be called any
1631  * time after the mutex has been initialized. It is called in both
1632  * the error case in attach and the normal detach case so it needs
1633  * to be careful about only freeing resources that have actually been
1634  * allocated.
1635  */
1636 static int
1637 re_detach(device_t dev)
1638 {
1639 	struct rl_softc		*sc;
1640 	struct ifnet		*ifp;
1641 	int			i, rid;
1642 
1643 	sc = device_get_softc(dev);
1644 	ifp = sc->rl_ifp;
1645 	KASSERT(mtx_initialized(&sc->rl_mtx), ("re mutex not initialized"));
1646 
1647 	/* These should only be active if attach succeeded */
1648 	if (device_is_attached(dev)) {
1649 #ifdef DEVICE_POLLING
1650 		if (ifp->if_capenable & IFCAP_POLLING)
1651 			ether_poll_deregister(ifp);
1652 #endif
1653 		RL_LOCK(sc);
1654 #if 0
1655 		sc->suspended = 1;
1656 #endif
1657 		re_stop(sc);
1658 		RL_UNLOCK(sc);
1659 		callout_drain(&sc->rl_stat_callout);
1660 		taskqueue_drain(taskqueue_fast, &sc->rl_inttask);
1661 		/*
1662 		 * Force off the IFF_UP flag here, in case someone
1663 		 * still had a BPF descriptor attached to this
1664 		 * interface. If they do, ether_ifdetach() will cause
1665 		 * the BPF code to try and clear the promisc mode
1666 		 * flag, which will bubble down to re_ioctl(),
1667 		 * which will try to call re_init() again. This will
1668 		 * turn the NIC back on and restart the MII ticker,
1669 		 * which will panic the system when the kernel tries
1670 		 * to invoke the re_tick() function that isn't there
1671 		 * anymore.
1672 		 */
1673 		ifp->if_flags &= ~IFF_UP;
1674 		ether_ifdetach(ifp);
1675 	}
1676 	if (sc->rl_miibus)
1677 		device_delete_child(dev, sc->rl_miibus);
1678 	bus_generic_detach(dev);
1679 
1680 	/*
1681 	 * The rest is resource deallocation, so we should already be
1682 	 * stopped here.
1683 	 */
1684 
1685 	if (sc->rl_intrhand[0] != NULL) {
1686 		bus_teardown_intr(dev, sc->rl_irq[0], sc->rl_intrhand[0]);
1687 		sc->rl_intrhand[0] = NULL;
1688 	}
1689 	if (ifp != NULL)
1690 		if_free(ifp);
1691 	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) == 0)
1692 		rid = 0;
1693 	else
1694 		rid = 1;
1695 	if (sc->rl_irq[0] != NULL) {
1696 		bus_release_resource(dev, SYS_RES_IRQ, rid, sc->rl_irq[0]);
1697 		sc->rl_irq[0] = NULL;
1698 	}
1699 	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) != 0)
1700 		pci_release_msi(dev);
1701 	if (sc->rl_res_pba) {
1702 		rid = PCIR_BAR(4);
1703 		bus_release_resource(dev, SYS_RES_MEMORY, rid, sc->rl_res_pba);
1704 	}
1705 	if (sc->rl_res)
1706 		bus_release_resource(dev, sc->rl_res_type, sc->rl_res_id,
1707 		    sc->rl_res);
1708 
1709 	/* Unload and free the RX DMA ring memory and map */
1710 
1711 	if (sc->rl_ldata.rl_rx_list_tag) {
1712 		if (sc->rl_ldata.rl_rx_list_map)
1713 			bus_dmamap_unload(sc->rl_ldata.rl_rx_list_tag,
1714 			    sc->rl_ldata.rl_rx_list_map);
1715 		if (sc->rl_ldata.rl_rx_list_map && sc->rl_ldata.rl_rx_list)
1716 			bus_dmamem_free(sc->rl_ldata.rl_rx_list_tag,
1717 			    sc->rl_ldata.rl_rx_list,
1718 			    sc->rl_ldata.rl_rx_list_map);
1719 		bus_dma_tag_destroy(sc->rl_ldata.rl_rx_list_tag);
1720 	}
1721 
1722 	/* Unload and free the TX DMA ring memory and map */
1723 
1724 	if (sc->rl_ldata.rl_tx_list_tag) {
1725 		if (sc->rl_ldata.rl_tx_list_map)
1726 			bus_dmamap_unload(sc->rl_ldata.rl_tx_list_tag,
1727 			    sc->rl_ldata.rl_tx_list_map);
1728 		if (sc->rl_ldata.rl_tx_list_map && sc->rl_ldata.rl_tx_list)
1729 			bus_dmamem_free(sc->rl_ldata.rl_tx_list_tag,
1730 			    sc->rl_ldata.rl_tx_list,
1731 			    sc->rl_ldata.rl_tx_list_map);
1732 		bus_dma_tag_destroy(sc->rl_ldata.rl_tx_list_tag);
1733 	}
1734 
1735 	/* Destroy all the RX and TX buffer maps */
1736 
1737 	if (sc->rl_ldata.rl_tx_mtag) {
1738 		for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
1739 			if (sc->rl_ldata.rl_tx_desc[i].tx_dmamap)
1740 				bus_dmamap_destroy(sc->rl_ldata.rl_tx_mtag,
1741 				    sc->rl_ldata.rl_tx_desc[i].tx_dmamap);
1742 		}
1743 		bus_dma_tag_destroy(sc->rl_ldata.rl_tx_mtag);
1744 	}
1745 	if (sc->rl_ldata.rl_rx_mtag) {
1746 		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1747 			if (sc->rl_ldata.rl_rx_desc[i].rx_dmamap)
1748 				bus_dmamap_destroy(sc->rl_ldata.rl_rx_mtag,
1749 				    sc->rl_ldata.rl_rx_desc[i].rx_dmamap);
1750 		}
1751 		if (sc->rl_ldata.rl_rx_sparemap)
1752 			bus_dmamap_destroy(sc->rl_ldata.rl_rx_mtag,
1753 			    sc->rl_ldata.rl_rx_sparemap);
1754 		bus_dma_tag_destroy(sc->rl_ldata.rl_rx_mtag);
1755 	}
1756 	if (sc->rl_ldata.rl_jrx_mtag) {
1757 		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1758 			if (sc->rl_ldata.rl_jrx_desc[i].rx_dmamap)
1759 				bus_dmamap_destroy(sc->rl_ldata.rl_jrx_mtag,
1760 				    sc->rl_ldata.rl_jrx_desc[i].rx_dmamap);
1761 		}
1762 		if (sc->rl_ldata.rl_jrx_sparemap)
1763 			bus_dmamap_destroy(sc->rl_ldata.rl_jrx_mtag,
1764 			    sc->rl_ldata.rl_jrx_sparemap);
1765 		bus_dma_tag_destroy(sc->rl_ldata.rl_jrx_mtag);
1766 	}
1767 	/* Unload and free the stats buffer and map */
1768 
1769 	if (sc->rl_ldata.rl_stag) {
1770 		if (sc->rl_ldata.rl_smap)
1771 			bus_dmamap_unload(sc->rl_ldata.rl_stag,
1772 			    sc->rl_ldata.rl_smap);
1773 		if (sc->rl_ldata.rl_smap && sc->rl_ldata.rl_stats)
1774 			bus_dmamem_free(sc->rl_ldata.rl_stag,
1775 			    sc->rl_ldata.rl_stats, sc->rl_ldata.rl_smap);
1776 		bus_dma_tag_destroy(sc->rl_ldata.rl_stag);
1777 	}
1778 
1779 	if (sc->rl_parent_tag)
1780 		bus_dma_tag_destroy(sc->rl_parent_tag);
1781 
1782 	mtx_destroy(&sc->rl_mtx);
1783 
1784 	return (0);
1785 }
1786 
1787 static __inline void
1788 re_discard_rxbuf(struct rl_softc *sc, int idx)
1789 {
1790 	struct rl_desc		*desc;
1791 	struct rl_rxdesc	*rxd;
1792 	uint32_t		cmdstat;
1793 
1794 	if (sc->rl_ifp->if_mtu > RL_MTU &&
1795 	    (sc->rl_flags & RL_FLAG_JUMBOV2) != 0)
1796 		rxd = &sc->rl_ldata.rl_jrx_desc[idx];
1797 	else
1798 		rxd = &sc->rl_ldata.rl_rx_desc[idx];
1799 	desc = &sc->rl_ldata.rl_rx_list[idx];
1800 	desc->rl_vlanctl = 0;
1801 	cmdstat = rxd->rx_size;
1802 	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
1803 		cmdstat |= RL_RDESC_CMD_EOR;
1804 	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
1805 }
1806 
1807 static int
1808 re_newbuf(struct rl_softc *sc, int idx)
1809 {
1810 	struct mbuf		*m;
1811 	struct rl_rxdesc	*rxd;
1812 	bus_dma_segment_t	segs[1];
1813 	bus_dmamap_t		map;
1814 	struct rl_desc		*desc;
1815 	uint32_t		cmdstat;
1816 	int			error, nsegs;
1817 
1818 	m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
1819 	if (m == NULL)
1820 		return (ENOBUFS);
1821 
1822 	m->m_len = m->m_pkthdr.len = MCLBYTES;
1823 #ifdef RE_FIXUP_RX
1824 	/*
1825 	 * This is part of an evil trick to deal with non-x86 platforms.
1826 	 * The RealTek chip requires RX buffers to be aligned on 64-bit
1827 	 * boundaries, but that will hose non-x86 machines. To get around
1828 	 * this, we leave some empty space at the start of each buffer
1829 	 * and for non-x86 hosts, we copy the buffer back six bytes
1830 	 * to achieve word alignment. This is slightly more efficient
1831 	 * than allocating a new buffer, copying the contents, and
1832 	 * discarding the old buffer.
1833 	 */
1834 	m_adj(m, RE_ETHER_ALIGN);
1835 #endif
1836 	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_rx_mtag,
1837 	    sc->rl_ldata.rl_rx_sparemap, m, segs, &nsegs, BUS_DMA_NOWAIT);
1838 	if (error != 0) {
1839 		m_freem(m);
1840 		return (ENOBUFS);
1841 	}
1842 	KASSERT(nsegs == 1, ("%s: %d segment returned!", __func__, nsegs));
1843 
1844 	rxd = &sc->rl_ldata.rl_rx_desc[idx];
1845 	if (rxd->rx_m != NULL) {
1846 		bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap,
1847 		    BUS_DMASYNC_POSTREAD);
1848 		bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap);
1849 	}
1850 
1851 	rxd->rx_m = m;
1852 	map = rxd->rx_dmamap;
1853 	rxd->rx_dmamap = sc->rl_ldata.rl_rx_sparemap;
1854 	rxd->rx_size = segs[0].ds_len;
1855 	sc->rl_ldata.rl_rx_sparemap = map;
1856 	bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap,
1857 	    BUS_DMASYNC_PREREAD);
1858 
1859 	desc = &sc->rl_ldata.rl_rx_list[idx];
1860 	desc->rl_vlanctl = 0;
1861 	desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[0].ds_addr));
1862 	desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[0].ds_addr));
1863 	cmdstat = segs[0].ds_len;
1864 	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
1865 		cmdstat |= RL_RDESC_CMD_EOR;
1866 	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
1867 
1868 	return (0);
1869 }
1870 
1871 static int
1872 re_jumbo_newbuf(struct rl_softc *sc, int idx)
1873 {
1874 	struct mbuf		*m;
1875 	struct rl_rxdesc	*rxd;
1876 	bus_dma_segment_t	segs[1];
1877 	bus_dmamap_t		map;
1878 	struct rl_desc		*desc;
1879 	uint32_t		cmdstat;
1880 	int			error, nsegs;
1881 
1882 	m = m_getjcl(M_DONTWAIT, MT_DATA, M_PKTHDR, MJUM9BYTES);
1883 	if (m == NULL)
1884 		return (ENOBUFS);
1885 	m->m_len = m->m_pkthdr.len = MJUM9BYTES;
1886 #ifdef RE_FIXUP_RX
1887 	m_adj(m, RE_ETHER_ALIGN);
1888 #endif
1889 	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_jrx_mtag,
1890 	    sc->rl_ldata.rl_jrx_sparemap, m, segs, &nsegs, BUS_DMA_NOWAIT);
1891 	if (error != 0) {
1892 		m_freem(m);
1893 		return (ENOBUFS);
1894 	}
1895 	KASSERT(nsegs == 1, ("%s: %d segment returned!", __func__, nsegs));
1896 
1897 	rxd = &sc->rl_ldata.rl_jrx_desc[idx];
1898 	if (rxd->rx_m != NULL) {
1899 		bus_dmamap_sync(sc->rl_ldata.rl_jrx_mtag, rxd->rx_dmamap,
1900 		    BUS_DMASYNC_POSTREAD);
1901 		bus_dmamap_unload(sc->rl_ldata.rl_jrx_mtag, rxd->rx_dmamap);
1902 	}
1903 
1904 	rxd->rx_m = m;
1905 	map = rxd->rx_dmamap;
1906 	rxd->rx_dmamap = sc->rl_ldata.rl_jrx_sparemap;
1907 	rxd->rx_size = segs[0].ds_len;
1908 	sc->rl_ldata.rl_jrx_sparemap = map;
1909 	bus_dmamap_sync(sc->rl_ldata.rl_jrx_mtag, rxd->rx_dmamap,
1910 	    BUS_DMASYNC_PREREAD);
1911 
1912 	desc = &sc->rl_ldata.rl_rx_list[idx];
1913 	desc->rl_vlanctl = 0;
1914 	desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[0].ds_addr));
1915 	desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[0].ds_addr));
1916 	cmdstat = segs[0].ds_len;
1917 	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
1918 		cmdstat |= RL_RDESC_CMD_EOR;
1919 	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
1920 
1921 	return (0);
1922 }
1923 
1924 #ifdef RE_FIXUP_RX
1925 static __inline void
1926 re_fixup_rx(struct mbuf *m)
1927 {
1928 	int                     i;
1929 	uint16_t                *src, *dst;
1930 
1931 	src = mtod(m, uint16_t *);
1932 	dst = src - (RE_ETHER_ALIGN - ETHER_ALIGN) / sizeof *src;
1933 
1934 	for (i = 0; i < (m->m_len / sizeof(uint16_t) + 1); i++)
1935 		*dst++ = *src++;
1936 
1937 	m->m_data -= RE_ETHER_ALIGN - ETHER_ALIGN;
1938 }
1939 #endif
1940 
1941 static int
1942 re_tx_list_init(struct rl_softc *sc)
1943 {
1944 	struct rl_desc		*desc;
1945 	int			i;
1946 
1947 	RL_LOCK_ASSERT(sc);
1948 
1949 	bzero(sc->rl_ldata.rl_tx_list,
1950 	    sc->rl_ldata.rl_tx_desc_cnt * sizeof(struct rl_desc));
1951 	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++)
1952 		sc->rl_ldata.rl_tx_desc[i].tx_m = NULL;
1953 	/* Set EOR. */
1954 	desc = &sc->rl_ldata.rl_tx_list[sc->rl_ldata.rl_tx_desc_cnt - 1];
1955 	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_EOR);
1956 
1957 	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
1958 	    sc->rl_ldata.rl_tx_list_map,
1959 	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1960 
1961 	sc->rl_ldata.rl_tx_prodidx = 0;
1962 	sc->rl_ldata.rl_tx_considx = 0;
1963 	sc->rl_ldata.rl_tx_free = sc->rl_ldata.rl_tx_desc_cnt;
1964 
1965 	return (0);
1966 }
1967 
1968 static int
1969 re_rx_list_init(struct rl_softc *sc)
1970 {
1971 	int			error, i;
1972 
1973 	bzero(sc->rl_ldata.rl_rx_list,
1974 	    sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc));
1975 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1976 		sc->rl_ldata.rl_rx_desc[i].rx_m = NULL;
1977 		if ((error = re_newbuf(sc, i)) != 0)
1978 			return (error);
1979 	}
1980 
1981 	/* Flush the RX descriptors */
1982 
1983 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1984 	    sc->rl_ldata.rl_rx_list_map,
1985 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
1986 
1987 	sc->rl_ldata.rl_rx_prodidx = 0;
1988 	sc->rl_head = sc->rl_tail = NULL;
1989 	sc->rl_int_rx_act = 0;
1990 
1991 	return (0);
1992 }
1993 
1994 static int
1995 re_jrx_list_init(struct rl_softc *sc)
1996 {
1997 	int			error, i;
1998 
1999 	bzero(sc->rl_ldata.rl_rx_list,
2000 	    sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc));
2001 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
2002 		sc->rl_ldata.rl_jrx_desc[i].rx_m = NULL;
2003 		if ((error = re_jumbo_newbuf(sc, i)) != 0)
2004 			return (error);
2005 	}
2006 
2007 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
2008 	    sc->rl_ldata.rl_rx_list_map,
2009 	    BUS_DMASYNC_PREWRITE | BUS_DMASYNC_PREREAD);
2010 
2011 	sc->rl_ldata.rl_rx_prodidx = 0;
2012 	sc->rl_head = sc->rl_tail = NULL;
2013 	sc->rl_int_rx_act = 0;
2014 
2015 	return (0);
2016 }
2017 
2018 /*
2019  * RX handler for C+ and 8169. For the gigE chips, we support
2020  * the reception of jumbo frames that have been fragmented
2021  * across multiple 2K mbuf cluster buffers.
2022  */
2023 static int
2024 re_rxeof(struct rl_softc *sc, int *rx_npktsp)
2025 {
2026 	struct mbuf		*m;
2027 	struct ifnet		*ifp;
2028 	int			i, rxerr, total_len;
2029 	struct rl_desc		*cur_rx;
2030 	u_int32_t		rxstat, rxvlan;
2031 	int			jumbo, maxpkt = 16, rx_npkts = 0;
2032 
2033 	RL_LOCK_ASSERT(sc);
2034 
2035 	ifp = sc->rl_ifp;
2036 	if (ifp->if_mtu > RL_MTU && (sc->rl_flags & RL_FLAG_JUMBOV2) != 0)
2037 		jumbo = 1;
2038 	else
2039 		jumbo = 0;
2040 
2041 	/* Invalidate the descriptor memory */
2042 
2043 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
2044 	    sc->rl_ldata.rl_rx_list_map,
2045 	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2046 
2047 	for (i = sc->rl_ldata.rl_rx_prodidx; maxpkt > 0;
2048 	    i = RL_RX_DESC_NXT(sc, i)) {
2049 		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
2050 			break;
2051 		cur_rx = &sc->rl_ldata.rl_rx_list[i];
2052 		rxstat = le32toh(cur_rx->rl_cmdstat);
2053 		if ((rxstat & RL_RDESC_STAT_OWN) != 0)
2054 			break;
2055 		total_len = rxstat & sc->rl_rxlenmask;
2056 		rxvlan = le32toh(cur_rx->rl_vlanctl);
2057 		if (jumbo != 0)
2058 			m = sc->rl_ldata.rl_jrx_desc[i].rx_m;
2059 		else
2060 			m = sc->rl_ldata.rl_rx_desc[i].rx_m;
2061 
2062 		if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0 &&
2063 		    (rxstat & (RL_RDESC_STAT_SOF | RL_RDESC_STAT_EOF)) !=
2064 		    (RL_RDESC_STAT_SOF | RL_RDESC_STAT_EOF)) {
2065 			/*
2066 			 * RTL8168C or later controllers do not
2067 			 * support multi-fragment packet.
2068 			 */
2069 			re_discard_rxbuf(sc, i);
2070 			continue;
2071 		} else if ((rxstat & RL_RDESC_STAT_EOF) == 0) {
2072 			if (re_newbuf(sc, i) != 0) {
2073 				/*
2074 				 * If this is part of a multi-fragment packet,
2075 				 * discard all the pieces.
2076 				 */
2077 				if (sc->rl_head != NULL) {
2078 					m_freem(sc->rl_head);
2079 					sc->rl_head = sc->rl_tail = NULL;
2080 				}
2081 				re_discard_rxbuf(sc, i);
2082 				continue;
2083 			}
2084 			m->m_len = RE_RX_DESC_BUFLEN;
2085 			if (sc->rl_head == NULL)
2086 				sc->rl_head = sc->rl_tail = m;
2087 			else {
2088 				m->m_flags &= ~M_PKTHDR;
2089 				sc->rl_tail->m_next = m;
2090 				sc->rl_tail = m;
2091 			}
2092 			continue;
2093 		}
2094 
2095 		/*
2096 		 * NOTE: for the 8139C+, the frame length field
2097 		 * is always 12 bits in size, but for the gigE chips,
2098 		 * it is 13 bits (since the max RX frame length is 16K).
2099 		 * Unfortunately, all 32 bits in the status word
2100 		 * were already used, so to make room for the extra
2101 		 * length bit, RealTek took out the 'frame alignment
2102 		 * error' bit and shifted the other status bits
2103 		 * over one slot. The OWN, EOR, FS and LS bits are
2104 		 * still in the same places. We have already extracted
2105 		 * the frame length and checked the OWN bit, so rather
2106 		 * than using an alternate bit mapping, we shift the
2107 		 * status bits one space to the right so we can evaluate
2108 		 * them using the 8169 status as though it was in the
2109 		 * same format as that of the 8139C+.
2110 		 */
2111 		if (sc->rl_type == RL_8169)
2112 			rxstat >>= 1;
2113 
2114 		/*
2115 		 * if total_len > 2^13-1, both _RXERRSUM and _GIANT will be
2116 		 * set, but if CRC is clear, it will still be a valid frame.
2117 		 */
2118 		if ((rxstat & RL_RDESC_STAT_RXERRSUM) != 0) {
2119 			rxerr = 1;
2120 			if ((sc->rl_flags & RL_FLAG_JUMBOV2) == 0 &&
2121 			    total_len > 8191 &&
2122 			    (rxstat & RL_RDESC_STAT_ERRS) == RL_RDESC_STAT_GIANT)
2123 				rxerr = 0;
2124 			if (rxerr != 0) {
2125 				ifp->if_ierrors++;
2126 				/*
2127 				 * If this is part of a multi-fragment packet,
2128 				 * discard all the pieces.
2129 				 */
2130 				if (sc->rl_head != NULL) {
2131 					m_freem(sc->rl_head);
2132 					sc->rl_head = sc->rl_tail = NULL;
2133 				}
2134 				re_discard_rxbuf(sc, i);
2135 				continue;
2136 			}
2137 		}
2138 
2139 		/*
2140 		 * If allocating a replacement mbuf fails,
2141 		 * reload the current one.
2142 		 */
2143 		if (jumbo != 0)
2144 			rxerr = re_jumbo_newbuf(sc, i);
2145 		else
2146 			rxerr = re_newbuf(sc, i);
2147 		if (rxerr != 0) {
2148 			ifp->if_iqdrops++;
2149 			if (sc->rl_head != NULL) {
2150 				m_freem(sc->rl_head);
2151 				sc->rl_head = sc->rl_tail = NULL;
2152 			}
2153 			re_discard_rxbuf(sc, i);
2154 			continue;
2155 		}
2156 
2157 		if (sc->rl_head != NULL) {
2158 			if (jumbo != 0)
2159 				m->m_len = total_len;
2160 			else {
2161 				m->m_len = total_len % RE_RX_DESC_BUFLEN;
2162 				if (m->m_len == 0)
2163 					m->m_len = RE_RX_DESC_BUFLEN;
2164 			}
2165 			/*
2166 			 * Special case: if there's 4 bytes or less
2167 			 * in this buffer, the mbuf can be discarded:
2168 			 * the last 4 bytes is the CRC, which we don't
2169 			 * care about anyway.
2170 			 */
2171 			if (m->m_len <= ETHER_CRC_LEN) {
2172 				sc->rl_tail->m_len -=
2173 				    (ETHER_CRC_LEN - m->m_len);
2174 				m_freem(m);
2175 			} else {
2176 				m->m_len -= ETHER_CRC_LEN;
2177 				m->m_flags &= ~M_PKTHDR;
2178 				sc->rl_tail->m_next = m;
2179 			}
2180 			m = sc->rl_head;
2181 			sc->rl_head = sc->rl_tail = NULL;
2182 			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
2183 		} else
2184 			m->m_pkthdr.len = m->m_len =
2185 			    (total_len - ETHER_CRC_LEN);
2186 
2187 #ifdef RE_FIXUP_RX
2188 		re_fixup_rx(m);
2189 #endif
2190 		ifp->if_ipackets++;
2191 		m->m_pkthdr.rcvif = ifp;
2192 
2193 		/* Do RX checksumming if enabled */
2194 
2195 		if (ifp->if_capenable & IFCAP_RXCSUM) {
2196 			if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
2197 				/* Check IP header checksum */
2198 				if (rxstat & RL_RDESC_STAT_PROTOID)
2199 					m->m_pkthdr.csum_flags |=
2200 					    CSUM_IP_CHECKED;
2201 				if (!(rxstat & RL_RDESC_STAT_IPSUMBAD))
2202 					m->m_pkthdr.csum_flags |=
2203 					    CSUM_IP_VALID;
2204 
2205 				/* Check TCP/UDP checksum */
2206 				if ((RL_TCPPKT(rxstat) &&
2207 				    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
2208 				    (RL_UDPPKT(rxstat) &&
2209 				     !(rxstat & RL_RDESC_STAT_UDPSUMBAD))) {
2210 					m->m_pkthdr.csum_flags |=
2211 						CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
2212 					m->m_pkthdr.csum_data = 0xffff;
2213 				}
2214 			} else {
2215 				/*
2216 				 * RTL8168C/RTL816CP/RTL8111C/RTL8111CP
2217 				 */
2218 				if ((rxstat & RL_RDESC_STAT_PROTOID) &&
2219 				    (rxvlan & RL_RDESC_IPV4))
2220 					m->m_pkthdr.csum_flags |=
2221 					    CSUM_IP_CHECKED;
2222 				if (!(rxstat & RL_RDESC_STAT_IPSUMBAD) &&
2223 				    (rxvlan & RL_RDESC_IPV4))
2224 					m->m_pkthdr.csum_flags |=
2225 					    CSUM_IP_VALID;
2226 				if (((rxstat & RL_RDESC_STAT_TCP) &&
2227 				    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
2228 				    ((rxstat & RL_RDESC_STAT_UDP) &&
2229 				    !(rxstat & RL_RDESC_STAT_UDPSUMBAD))) {
2230 					m->m_pkthdr.csum_flags |=
2231 						CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
2232 					m->m_pkthdr.csum_data = 0xffff;
2233 				}
2234 			}
2235 		}
2236 		maxpkt--;
2237 		if (rxvlan & RL_RDESC_VLANCTL_TAG) {
2238 			m->m_pkthdr.ether_vtag =
2239 			    bswap16((rxvlan & RL_RDESC_VLANCTL_DATA));
2240 			m->m_flags |= M_VLANTAG;
2241 		}
2242 		RL_UNLOCK(sc);
2243 		(*ifp->if_input)(ifp, m);
2244 		RL_LOCK(sc);
2245 		rx_npkts++;
2246 	}
2247 
2248 	/* Flush the RX DMA ring */
2249 
2250 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
2251 	    sc->rl_ldata.rl_rx_list_map,
2252 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2253 
2254 	sc->rl_ldata.rl_rx_prodidx = i;
2255 
2256 	if (rx_npktsp != NULL)
2257 		*rx_npktsp = rx_npkts;
2258 	if (maxpkt)
2259 		return (EAGAIN);
2260 
2261 	return (0);
2262 }
2263 
2264 static void
2265 re_txeof(struct rl_softc *sc)
2266 {
2267 	struct ifnet		*ifp;
2268 	struct rl_txdesc	*txd;
2269 	u_int32_t		txstat;
2270 	int			cons;
2271 
2272 	cons = sc->rl_ldata.rl_tx_considx;
2273 	if (cons == sc->rl_ldata.rl_tx_prodidx)
2274 		return;
2275 
2276 	ifp = sc->rl_ifp;
2277 	/* Invalidate the TX descriptor list */
2278 	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2279 	    sc->rl_ldata.rl_tx_list_map,
2280 	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2281 
2282 	for (; cons != sc->rl_ldata.rl_tx_prodidx;
2283 	    cons = RL_TX_DESC_NXT(sc, cons)) {
2284 		txstat = le32toh(sc->rl_ldata.rl_tx_list[cons].rl_cmdstat);
2285 		if (txstat & RL_TDESC_STAT_OWN)
2286 			break;
2287 		/*
2288 		 * We only stash mbufs in the last descriptor
2289 		 * in a fragment chain, which also happens to
2290 		 * be the only place where the TX status bits
2291 		 * are valid.
2292 		 */
2293 		if (txstat & RL_TDESC_CMD_EOF) {
2294 			txd = &sc->rl_ldata.rl_tx_desc[cons];
2295 			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
2296 			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
2297 			bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag,
2298 			    txd->tx_dmamap);
2299 			KASSERT(txd->tx_m != NULL,
2300 			    ("%s: freeing NULL mbufs!", __func__));
2301 			m_freem(txd->tx_m);
2302 			txd->tx_m = NULL;
2303 			if (txstat & (RL_TDESC_STAT_EXCESSCOL|
2304 			    RL_TDESC_STAT_COLCNT))
2305 				ifp->if_collisions++;
2306 			if (txstat & RL_TDESC_STAT_TXERRSUM)
2307 				ifp->if_oerrors++;
2308 			else
2309 				ifp->if_opackets++;
2310 		}
2311 		sc->rl_ldata.rl_tx_free++;
2312 		ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
2313 	}
2314 	sc->rl_ldata.rl_tx_considx = cons;
2315 
2316 	/* No changes made to the TX ring, so no flush needed */
2317 
2318 	if (sc->rl_ldata.rl_tx_free != sc->rl_ldata.rl_tx_desc_cnt) {
2319 #ifdef RE_TX_MODERATION
2320 		/*
2321 		 * If not all descriptors have been reaped yet, reload
2322 		 * the timer so that we will eventually get another
2323 		 * interrupt that will cause us to re-enter this routine.
2324 		 * This is done in case the transmitter has gone idle.
2325 		 */
2326 		CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2327 #endif
2328 	} else
2329 		sc->rl_watchdog_timer = 0;
2330 }
2331 
2332 static void
2333 re_tick(void *xsc)
2334 {
2335 	struct rl_softc		*sc;
2336 	struct mii_data		*mii;
2337 
2338 	sc = xsc;
2339 
2340 	RL_LOCK_ASSERT(sc);
2341 
2342 	mii = device_get_softc(sc->rl_miibus);
2343 	mii_tick(mii);
2344 	if ((sc->rl_flags & RL_FLAG_LINK) == 0)
2345 		re_miibus_statchg(sc->rl_dev);
2346 	/*
2347 	 * Reclaim transmitted frames here. Technically it is not
2348 	 * necessary to do here but it ensures periodic reclamation
2349 	 * regardless of Tx completion interrupt which seems to be
2350 	 * lost on PCIe based controllers under certain situations.
2351 	 */
2352 	re_txeof(sc);
2353 	re_watchdog(sc);
2354 	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
2355 }
2356 
2357 #ifdef DEVICE_POLLING
2358 static int
2359 re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
2360 {
2361 	struct rl_softc *sc = ifp->if_softc;
2362 	int rx_npkts = 0;
2363 
2364 	RL_LOCK(sc);
2365 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
2366 		rx_npkts = re_poll_locked(ifp, cmd, count);
2367 	RL_UNLOCK(sc);
2368 	return (rx_npkts);
2369 }
2370 
2371 static int
2372 re_poll_locked(struct ifnet *ifp, enum poll_cmd cmd, int count)
2373 {
2374 	struct rl_softc *sc = ifp->if_softc;
2375 	int rx_npkts;
2376 
2377 	RL_LOCK_ASSERT(sc);
2378 
2379 	sc->rxcycles = count;
2380 	re_rxeof(sc, &rx_npkts);
2381 	re_txeof(sc);
2382 
2383 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2384 		re_start_locked(ifp);
2385 
2386 	if (cmd == POLL_AND_CHECK_STATUS) { /* also check status register */
2387 		u_int16_t       status;
2388 
2389 		status = CSR_READ_2(sc, RL_ISR);
2390 		if (status == 0xffff)
2391 			return (rx_npkts);
2392 		if (status)
2393 			CSR_WRITE_2(sc, RL_ISR, status);
2394 		if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
2395 		    (sc->rl_flags & RL_FLAG_PCIE))
2396 			CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2397 
2398 		/*
2399 		 * XXX check behaviour on receiver stalls.
2400 		 */
2401 
2402 		if (status & RL_ISR_SYSTEM_ERR) {
2403 			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2404 			re_init_locked(sc);
2405 		}
2406 	}
2407 	return (rx_npkts);
2408 }
2409 #endif /* DEVICE_POLLING */
2410 
2411 static int
2412 re_intr(void *arg)
2413 {
2414 	struct rl_softc		*sc;
2415 	uint16_t		status;
2416 
2417 	sc = arg;
2418 
2419 	status = CSR_READ_2(sc, RL_ISR);
2420 	if (status == 0xFFFF || (status & RL_INTRS_CPLUS) == 0)
2421                 return (FILTER_STRAY);
2422 	CSR_WRITE_2(sc, RL_IMR, 0);
2423 
2424 	taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_inttask);
2425 
2426 	return (FILTER_HANDLED);
2427 }
2428 
2429 static void
2430 re_int_task(void *arg, int npending)
2431 {
2432 	struct rl_softc		*sc;
2433 	struct ifnet		*ifp;
2434 	u_int16_t		status;
2435 	int			rval = 0;
2436 
2437 	sc = arg;
2438 	ifp = sc->rl_ifp;
2439 
2440 	RL_LOCK(sc);
2441 
2442 	status = CSR_READ_2(sc, RL_ISR);
2443         CSR_WRITE_2(sc, RL_ISR, status);
2444 
2445 	if (sc->suspended ||
2446 	    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2447 		RL_UNLOCK(sc);
2448 		return;
2449 	}
2450 
2451 #ifdef DEVICE_POLLING
2452 	if  (ifp->if_capenable & IFCAP_POLLING) {
2453 		RL_UNLOCK(sc);
2454 		return;
2455 	}
2456 #endif
2457 
2458 	if (status & (RL_ISR_RX_OK|RL_ISR_RX_ERR|RL_ISR_FIFO_OFLOW))
2459 		rval = re_rxeof(sc, NULL);
2460 
2461 	/*
2462 	 * Some chips will ignore a second TX request issued
2463 	 * while an existing transmission is in progress. If
2464 	 * the transmitter goes idle but there are still
2465 	 * packets waiting to be sent, we need to restart the
2466 	 * channel here to flush them out. This only seems to
2467 	 * be required with the PCIe devices.
2468 	 */
2469 	if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
2470 	    (sc->rl_flags & RL_FLAG_PCIE))
2471 		CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2472 	if (status & (
2473 #ifdef RE_TX_MODERATION
2474 	    RL_ISR_TIMEOUT_EXPIRED|
2475 #else
2476 	    RL_ISR_TX_OK|
2477 #endif
2478 	    RL_ISR_TX_ERR|RL_ISR_TX_DESC_UNAVAIL))
2479 		re_txeof(sc);
2480 
2481 	if (status & RL_ISR_SYSTEM_ERR) {
2482 		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2483 		re_init_locked(sc);
2484 	}
2485 
2486 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2487 		re_start_locked(ifp);
2488 
2489 	RL_UNLOCK(sc);
2490 
2491         if ((CSR_READ_2(sc, RL_ISR) & RL_INTRS_CPLUS) || rval) {
2492 		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_inttask);
2493 		return;
2494 	}
2495 
2496 	CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2497 }
2498 
2499 static void
2500 re_intr_msi(void *xsc)
2501 {
2502 	struct rl_softc		*sc;
2503 	struct ifnet		*ifp;
2504 	uint16_t		intrs, status;
2505 
2506 	sc = xsc;
2507 	RL_LOCK(sc);
2508 
2509 	ifp = sc->rl_ifp;
2510 #ifdef DEVICE_POLLING
2511 	if (ifp->if_capenable & IFCAP_POLLING) {
2512 		RL_UNLOCK(sc);
2513 		return;
2514 	}
2515 #endif
2516 	/* Disable interrupts. */
2517 	CSR_WRITE_2(sc, RL_IMR, 0);
2518 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2519 		RL_UNLOCK(sc);
2520 		return;
2521 	}
2522 
2523 	intrs = RL_INTRS_CPLUS;
2524 	status = CSR_READ_2(sc, RL_ISR);
2525         CSR_WRITE_2(sc, RL_ISR, status);
2526 	if (sc->rl_int_rx_act > 0) {
2527 		intrs &= ~(RL_ISR_RX_OK | RL_ISR_RX_ERR | RL_ISR_FIFO_OFLOW |
2528 		    RL_ISR_RX_OVERRUN);
2529 		status &= ~(RL_ISR_RX_OK | RL_ISR_RX_ERR | RL_ISR_FIFO_OFLOW |
2530 		    RL_ISR_RX_OVERRUN);
2531 	}
2532 
2533 	if (status & (RL_ISR_TIMEOUT_EXPIRED | RL_ISR_RX_OK | RL_ISR_RX_ERR |
2534 	    RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN)) {
2535 		re_rxeof(sc, NULL);
2536 		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
2537 			if (sc->rl_int_rx_mod != 0 &&
2538 			    (status & (RL_ISR_RX_OK | RL_ISR_RX_ERR |
2539 			    RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN)) != 0) {
2540 				/* Rearm one-shot timer. */
2541 				CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2542 				intrs &= ~(RL_ISR_RX_OK | RL_ISR_RX_ERR |
2543 				    RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN);
2544 				sc->rl_int_rx_act = 1;
2545 			} else {
2546 				intrs |= RL_ISR_RX_OK | RL_ISR_RX_ERR |
2547 				    RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN;
2548 				sc->rl_int_rx_act = 0;
2549 			}
2550 		}
2551 	}
2552 
2553 	/*
2554 	 * Some chips will ignore a second TX request issued
2555 	 * while an existing transmission is in progress. If
2556 	 * the transmitter goes idle but there are still
2557 	 * packets waiting to be sent, we need to restart the
2558 	 * channel here to flush them out. This only seems to
2559 	 * be required with the PCIe devices.
2560 	 */
2561 	if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
2562 	    (sc->rl_flags & RL_FLAG_PCIE))
2563 		CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2564 	if (status & (RL_ISR_TX_OK | RL_ISR_TX_ERR | RL_ISR_TX_DESC_UNAVAIL))
2565 		re_txeof(sc);
2566 
2567 	if (status & RL_ISR_SYSTEM_ERR) {
2568 		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2569 		re_init_locked(sc);
2570 	}
2571 
2572 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
2573 		if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2574 			re_start_locked(ifp);
2575 		CSR_WRITE_2(sc, RL_IMR, intrs);
2576 	}
2577 	RL_UNLOCK(sc);
2578 }
2579 
2580 static int
2581 re_encap(struct rl_softc *sc, struct mbuf **m_head)
2582 {
2583 	struct rl_txdesc	*txd, *txd_last;
2584 	bus_dma_segment_t	segs[RL_NTXSEGS];
2585 	bus_dmamap_t		map;
2586 	struct mbuf		*m_new;
2587 	struct rl_desc		*desc;
2588 	int			nsegs, prod;
2589 	int			i, error, ei, si;
2590 	int			padlen;
2591 	uint32_t		cmdstat, csum_flags, vlanctl;
2592 
2593 	RL_LOCK_ASSERT(sc);
2594 	M_ASSERTPKTHDR((*m_head));
2595 
2596 	/*
2597 	 * With some of the RealTek chips, using the checksum offload
2598 	 * support in conjunction with the autopadding feature results
2599 	 * in the transmission of corrupt frames. For example, if we
2600 	 * need to send a really small IP fragment that's less than 60
2601 	 * bytes in size, and IP header checksumming is enabled, the
2602 	 * resulting ethernet frame that appears on the wire will
2603 	 * have garbled payload. To work around this, if TX IP checksum
2604 	 * offload is enabled, we always manually pad short frames out
2605 	 * to the minimum ethernet frame size.
2606 	 */
2607 	if ((sc->rl_flags & RL_FLAG_AUTOPAD) == 0 &&
2608 	    (*m_head)->m_pkthdr.len < RL_IP4CSUMTX_PADLEN &&
2609 	    ((*m_head)->m_pkthdr.csum_flags & CSUM_IP) != 0) {
2610 		padlen = RL_MIN_FRAMELEN - (*m_head)->m_pkthdr.len;
2611 		if (M_WRITABLE(*m_head) == 0) {
2612 			/* Get a writable copy. */
2613 			m_new = m_dup(*m_head, M_DONTWAIT);
2614 			m_freem(*m_head);
2615 			if (m_new == NULL) {
2616 				*m_head = NULL;
2617 				return (ENOBUFS);
2618 			}
2619 			*m_head = m_new;
2620 		}
2621 		if ((*m_head)->m_next != NULL ||
2622 		    M_TRAILINGSPACE(*m_head) < padlen) {
2623 			m_new = m_defrag(*m_head, M_DONTWAIT);
2624 			if (m_new == NULL) {
2625 				m_freem(*m_head);
2626 				*m_head = NULL;
2627 				return (ENOBUFS);
2628 			}
2629 		} else
2630 			m_new = *m_head;
2631 
2632 		/*
2633 		 * Manually pad short frames, and zero the pad space
2634 		 * to avoid leaking data.
2635 		 */
2636 		bzero(mtod(m_new, char *) + m_new->m_pkthdr.len, padlen);
2637 		m_new->m_pkthdr.len += padlen;
2638 		m_new->m_len = m_new->m_pkthdr.len;
2639 		*m_head = m_new;
2640 	}
2641 
2642 	prod = sc->rl_ldata.rl_tx_prodidx;
2643 	txd = &sc->rl_ldata.rl_tx_desc[prod];
2644 	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap,
2645 	    *m_head, segs, &nsegs, BUS_DMA_NOWAIT);
2646 	if (error == EFBIG) {
2647 		m_new = m_collapse(*m_head, M_DONTWAIT, RL_NTXSEGS);
2648 		if (m_new == NULL) {
2649 			m_freem(*m_head);
2650 			*m_head = NULL;
2651 			return (ENOBUFS);
2652 		}
2653 		*m_head = m_new;
2654 		error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_tx_mtag,
2655 		    txd->tx_dmamap, *m_head, segs, &nsegs, BUS_DMA_NOWAIT);
2656 		if (error != 0) {
2657 			m_freem(*m_head);
2658 			*m_head = NULL;
2659 			return (error);
2660 		}
2661 	} else if (error != 0)
2662 		return (error);
2663 	if (nsegs == 0) {
2664 		m_freem(*m_head);
2665 		*m_head = NULL;
2666 		return (EIO);
2667 	}
2668 
2669 	/* Check for number of available descriptors. */
2670 	if (sc->rl_ldata.rl_tx_free - nsegs <= 1) {
2671 		bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap);
2672 		return (ENOBUFS);
2673 	}
2674 
2675 	bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap,
2676 	    BUS_DMASYNC_PREWRITE);
2677 
2678 	/*
2679 	 * Set up checksum offload. Note: checksum offload bits must
2680 	 * appear in all descriptors of a multi-descriptor transmit
2681 	 * attempt. This is according to testing done with an 8169
2682 	 * chip. This is a requirement.
2683 	 */
2684 	vlanctl = 0;
2685 	csum_flags = 0;
2686 	if (((*m_head)->m_pkthdr.csum_flags & CSUM_TSO) != 0) {
2687 		if ((sc->rl_flags & RL_FLAG_DESCV2) != 0) {
2688 			csum_flags |= RL_TDESC_CMD_LGSEND;
2689 			vlanctl |= ((uint32_t)(*m_head)->m_pkthdr.tso_segsz <<
2690 			    RL_TDESC_CMD_MSSVALV2_SHIFT);
2691 		} else {
2692 			csum_flags |= RL_TDESC_CMD_LGSEND |
2693 			    ((uint32_t)(*m_head)->m_pkthdr.tso_segsz <<
2694 			    RL_TDESC_CMD_MSSVAL_SHIFT);
2695 		}
2696 	} else {
2697 		/*
2698 		 * Unconditionally enable IP checksum if TCP or UDP
2699 		 * checksum is required. Otherwise, TCP/UDP checksum
2700 		 * does't make effects.
2701 		 */
2702 		if (((*m_head)->m_pkthdr.csum_flags & RE_CSUM_FEATURES) != 0) {
2703 			if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
2704 				csum_flags |= RL_TDESC_CMD_IPCSUM;
2705 				if (((*m_head)->m_pkthdr.csum_flags &
2706 				    CSUM_TCP) != 0)
2707 					csum_flags |= RL_TDESC_CMD_TCPCSUM;
2708 				if (((*m_head)->m_pkthdr.csum_flags &
2709 				    CSUM_UDP) != 0)
2710 					csum_flags |= RL_TDESC_CMD_UDPCSUM;
2711 			} else {
2712 				vlanctl |= RL_TDESC_CMD_IPCSUMV2;
2713 				if (((*m_head)->m_pkthdr.csum_flags &
2714 				    CSUM_TCP) != 0)
2715 					vlanctl |= RL_TDESC_CMD_TCPCSUMV2;
2716 				if (((*m_head)->m_pkthdr.csum_flags &
2717 				    CSUM_UDP) != 0)
2718 					vlanctl |= RL_TDESC_CMD_UDPCSUMV2;
2719 			}
2720 		}
2721 	}
2722 
2723 	/*
2724 	 * Set up hardware VLAN tagging. Note: vlan tag info must
2725 	 * appear in all descriptors of a multi-descriptor
2726 	 * transmission attempt.
2727 	 */
2728 	if ((*m_head)->m_flags & M_VLANTAG)
2729 		vlanctl |= bswap16((*m_head)->m_pkthdr.ether_vtag) |
2730 		    RL_TDESC_VLANCTL_TAG;
2731 
2732 	si = prod;
2733 	for (i = 0; i < nsegs; i++, prod = RL_TX_DESC_NXT(sc, prod)) {
2734 		desc = &sc->rl_ldata.rl_tx_list[prod];
2735 		desc->rl_vlanctl = htole32(vlanctl);
2736 		desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[i].ds_addr));
2737 		desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[i].ds_addr));
2738 		cmdstat = segs[i].ds_len;
2739 		if (i != 0)
2740 			cmdstat |= RL_TDESC_CMD_OWN;
2741 		if (prod == sc->rl_ldata.rl_tx_desc_cnt - 1)
2742 			cmdstat |= RL_TDESC_CMD_EOR;
2743 		desc->rl_cmdstat = htole32(cmdstat | csum_flags);
2744 		sc->rl_ldata.rl_tx_free--;
2745 	}
2746 	/* Update producer index. */
2747 	sc->rl_ldata.rl_tx_prodidx = prod;
2748 
2749 	/* Set EOF on the last descriptor. */
2750 	ei = RL_TX_DESC_PRV(sc, prod);
2751 	desc = &sc->rl_ldata.rl_tx_list[ei];
2752 	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_EOF);
2753 
2754 	desc = &sc->rl_ldata.rl_tx_list[si];
2755 	/* Set SOF and transfer ownership of packet to the chip. */
2756 	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_OWN | RL_TDESC_CMD_SOF);
2757 
2758 	/*
2759 	 * Insure that the map for this transmission
2760 	 * is placed at the array index of the last descriptor
2761 	 * in this chain.  (Swap last and first dmamaps.)
2762 	 */
2763 	txd_last = &sc->rl_ldata.rl_tx_desc[ei];
2764 	map = txd->tx_dmamap;
2765 	txd->tx_dmamap = txd_last->tx_dmamap;
2766 	txd_last->tx_dmamap = map;
2767 	txd_last->tx_m = *m_head;
2768 
2769 	return (0);
2770 }
2771 
2772 static void
2773 re_start(struct ifnet *ifp)
2774 {
2775 	struct rl_softc		*sc;
2776 
2777 	sc = ifp->if_softc;
2778 	RL_LOCK(sc);
2779 	re_start_locked(ifp);
2780 	RL_UNLOCK(sc);
2781 }
2782 
2783 /*
2784  * Main transmit routine for C+ and gigE NICs.
2785  */
2786 static void
2787 re_start_locked(struct ifnet *ifp)
2788 {
2789 	struct rl_softc		*sc;
2790 	struct mbuf		*m_head;
2791 	int			queued;
2792 
2793 	sc = ifp->if_softc;
2794 
2795 	if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
2796 	    IFF_DRV_RUNNING || (sc->rl_flags & RL_FLAG_LINK) == 0)
2797 		return;
2798 
2799 	for (queued = 0; !IFQ_DRV_IS_EMPTY(&ifp->if_snd) &&
2800 	    sc->rl_ldata.rl_tx_free > 1;) {
2801 		IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
2802 		if (m_head == NULL)
2803 			break;
2804 
2805 		if (re_encap(sc, &m_head) != 0) {
2806 			if (m_head == NULL)
2807 				break;
2808 			IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
2809 			ifp->if_drv_flags |= IFF_DRV_OACTIVE;
2810 			break;
2811 		}
2812 
2813 		/*
2814 		 * If there's a BPF listener, bounce a copy of this frame
2815 		 * to him.
2816 		 */
2817 		ETHER_BPF_MTAP(ifp, m_head);
2818 
2819 		queued++;
2820 	}
2821 
2822 	if (queued == 0) {
2823 #ifdef RE_TX_MODERATION
2824 		if (sc->rl_ldata.rl_tx_free != sc->rl_ldata.rl_tx_desc_cnt)
2825 			CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2826 #endif
2827 		return;
2828 	}
2829 
2830 	/* Flush the TX descriptors */
2831 
2832 	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2833 	    sc->rl_ldata.rl_tx_list_map,
2834 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2835 
2836 	CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2837 
2838 #ifdef RE_TX_MODERATION
2839 	/*
2840 	 * Use the countdown timer for interrupt moderation.
2841 	 * 'TX done' interrupts are disabled. Instead, we reset the
2842 	 * countdown timer, which will begin counting until it hits
2843 	 * the value in the TIMERINT register, and then trigger an
2844 	 * interrupt. Each time we write to the TIMERCNT register,
2845 	 * the timer count is reset to 0.
2846 	 */
2847 	CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2848 #endif
2849 
2850 	/*
2851 	 * Set a timeout in case the chip goes out to lunch.
2852 	 */
2853 	sc->rl_watchdog_timer = 5;
2854 }
2855 
2856 static void
2857 re_set_jumbo(struct rl_softc *sc, int jumbo)
2858 {
2859 
2860 	if (sc->rl_hwrev->rl_rev == RL_HWREV_8168E_VL) {
2861 		pci_set_max_read_req(sc->rl_dev, 4096);
2862 		return;
2863 	}
2864 
2865 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG);
2866 	if (jumbo != 0) {
2867 		CSR_WRITE_1(sc, RL_CFG3, CSR_READ_1(sc, RL_CFG3) |
2868 		    RL_CFG3_JUMBO_EN0);
2869 		switch (sc->rl_hwrev->rl_rev) {
2870 		case RL_HWREV_8168DP:
2871 			break;
2872 		case RL_HWREV_8168E:
2873 			CSR_WRITE_1(sc, RL_CFG4, CSR_READ_1(sc, RL_CFG4) |
2874 			    0x01);
2875 			break;
2876 		default:
2877 			CSR_WRITE_1(sc, RL_CFG4, CSR_READ_1(sc, RL_CFG4) |
2878 			    RL_CFG4_JUMBO_EN1);
2879 		}
2880 	} else {
2881 		CSR_WRITE_1(sc, RL_CFG3, CSR_READ_1(sc, RL_CFG3) &
2882 		    ~RL_CFG3_JUMBO_EN0);
2883 		switch (sc->rl_hwrev->rl_rev) {
2884 		case RL_HWREV_8168DP:
2885 			break;
2886 		case RL_HWREV_8168E:
2887 			CSR_WRITE_1(sc, RL_CFG4, CSR_READ_1(sc, RL_CFG4) &
2888 			    ~0x01);
2889 			break;
2890 		default:
2891 			CSR_WRITE_1(sc, RL_CFG4, CSR_READ_1(sc, RL_CFG4) &
2892 			    ~RL_CFG4_JUMBO_EN1);
2893 		}
2894 	}
2895 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
2896 
2897 	switch (sc->rl_hwrev->rl_rev) {
2898 	case RL_HWREV_8168DP:
2899 		pci_set_max_read_req(sc->rl_dev, 4096);
2900 		break;
2901 	default:
2902 		if (jumbo != 0)
2903 			pci_set_max_read_req(sc->rl_dev, 512);
2904 		else
2905 			pci_set_max_read_req(sc->rl_dev, 4096);
2906 	}
2907 }
2908 
2909 static void
2910 re_init(void *xsc)
2911 {
2912 	struct rl_softc		*sc = xsc;
2913 
2914 	RL_LOCK(sc);
2915 	re_init_locked(sc);
2916 	RL_UNLOCK(sc);
2917 }
2918 
2919 static void
2920 re_init_locked(struct rl_softc *sc)
2921 {
2922 	struct ifnet		*ifp = sc->rl_ifp;
2923 	struct mii_data		*mii;
2924 	uint32_t		reg;
2925 	uint16_t		cfg;
2926 	union {
2927 		uint32_t align_dummy;
2928 		u_char eaddr[ETHER_ADDR_LEN];
2929         } eaddr;
2930 
2931 	RL_LOCK_ASSERT(sc);
2932 
2933 	mii = device_get_softc(sc->rl_miibus);
2934 
2935 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
2936 		return;
2937 
2938 	/*
2939 	 * Cancel pending I/O and free all RX/TX buffers.
2940 	 */
2941 	re_stop(sc);
2942 
2943 	/* Put controller into known state. */
2944 	re_reset(sc);
2945 
2946 	/*
2947 	 * For C+ mode, initialize the RX descriptors and mbufs.
2948 	 */
2949 	if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
2950 		if (ifp->if_mtu > RL_MTU) {
2951 			if (re_jrx_list_init(sc) != 0) {
2952 				device_printf(sc->rl_dev,
2953 				    "no memory for jumbo RX buffers\n");
2954 				re_stop(sc);
2955 				return;
2956 			}
2957 			/* Disable checksum offloading for jumbo frames. */
2958 			ifp->if_capenable &= ~(IFCAP_HWCSUM | IFCAP_TSO4);
2959 			ifp->if_hwassist &= ~(RE_CSUM_FEATURES | CSUM_TSO);
2960 		} else {
2961 			if (re_rx_list_init(sc) != 0) {
2962 				device_printf(sc->rl_dev,
2963 				    "no memory for RX buffers\n");
2964 				re_stop(sc);
2965 				return;
2966 			}
2967 		}
2968 		re_set_jumbo(sc, ifp->if_mtu > RL_MTU);
2969 	} else {
2970 		if (re_rx_list_init(sc) != 0) {
2971 			device_printf(sc->rl_dev, "no memory for RX buffers\n");
2972 			re_stop(sc);
2973 			return;
2974 		}
2975 		if ((sc->rl_flags & RL_FLAG_PCIE) != 0 &&
2976 		    pci_get_device(sc->rl_dev) != RT_DEVICEID_8101E) {
2977 			if (ifp->if_mtu > RL_MTU)
2978 				pci_set_max_read_req(sc->rl_dev, 512);
2979 			else
2980 				pci_set_max_read_req(sc->rl_dev, 4096);
2981 		}
2982 	}
2983 	re_tx_list_init(sc);
2984 
2985 	/*
2986 	 * Enable C+ RX and TX mode, as well as VLAN stripping and
2987 	 * RX checksum offload. We must configure the C+ register
2988 	 * before all others.
2989 	 */
2990 	cfg = RL_CPLUSCMD_PCI_MRW;
2991 	if ((ifp->if_capenable & IFCAP_RXCSUM) != 0)
2992 		cfg |= RL_CPLUSCMD_RXCSUM_ENB;
2993 	if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0)
2994 		cfg |= RL_CPLUSCMD_VLANSTRIP;
2995 	if ((sc->rl_flags & RL_FLAG_MACSTAT) != 0) {
2996 		cfg |= RL_CPLUSCMD_MACSTAT_DIS;
2997 		/* XXX magic. */
2998 		cfg |= 0x0001;
2999 	} else
3000 		cfg |= RL_CPLUSCMD_RXENB | RL_CPLUSCMD_TXENB;
3001 	CSR_WRITE_2(sc, RL_CPLUS_CMD, cfg);
3002 	if (sc->rl_hwrev->rl_rev == RL_HWREV_8169_8110SC ||
3003 	    sc->rl_hwrev->rl_rev == RL_HWREV_8169_8110SCE) {
3004 		reg = 0x000fff00;
3005 		if ((CSR_READ_1(sc, RL_CFG2) & RL_CFG2_PCI66MHZ) != 0)
3006 			reg |= 0x000000ff;
3007 		if (sc->rl_hwrev->rl_rev == RL_HWREV_8169_8110SCE)
3008 			reg |= 0x00f00000;
3009 		CSR_WRITE_4(sc, 0x7c, reg);
3010 		/* Disable interrupt mitigation. */
3011 		CSR_WRITE_2(sc, 0xe2, 0);
3012 	}
3013 	/*
3014 	 * Disable TSO if interface MTU size is greater than MSS
3015 	 * allowed in controller.
3016 	 */
3017 	if (ifp->if_mtu > RL_TSO_MTU && (ifp->if_capenable & IFCAP_TSO4) != 0) {
3018 		ifp->if_capenable &= ~IFCAP_TSO4;
3019 		ifp->if_hwassist &= ~CSUM_TSO;
3020 	}
3021 
3022 	/*
3023 	 * Init our MAC address.  Even though the chipset
3024 	 * documentation doesn't mention it, we need to enter "Config
3025 	 * register write enable" mode to modify the ID registers.
3026 	 */
3027 	/* Copy MAC address on stack to align. */
3028 	bcopy(IF_LLADDR(ifp), eaddr.eaddr, ETHER_ADDR_LEN);
3029 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG);
3030 	CSR_WRITE_4(sc, RL_IDR0,
3031 	    htole32(*(u_int32_t *)(&eaddr.eaddr[0])));
3032 	CSR_WRITE_4(sc, RL_IDR4,
3033 	    htole32(*(u_int32_t *)(&eaddr.eaddr[4])));
3034 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3035 
3036 	/*
3037 	 * Load the addresses of the RX and TX lists into the chip.
3038 	 */
3039 
3040 	CSR_WRITE_4(sc, RL_RXLIST_ADDR_HI,
3041 	    RL_ADDR_HI(sc->rl_ldata.rl_rx_list_addr));
3042 	CSR_WRITE_4(sc, RL_RXLIST_ADDR_LO,
3043 	    RL_ADDR_LO(sc->rl_ldata.rl_rx_list_addr));
3044 
3045 	CSR_WRITE_4(sc, RL_TXLIST_ADDR_HI,
3046 	    RL_ADDR_HI(sc->rl_ldata.rl_tx_list_addr));
3047 	CSR_WRITE_4(sc, RL_TXLIST_ADDR_LO,
3048 	    RL_ADDR_LO(sc->rl_ldata.rl_tx_list_addr));
3049 
3050 	/*
3051 	 * Enable transmit and receive.
3052 	 */
3053 	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
3054 
3055 	/*
3056 	 * Set the initial TX configuration.
3057 	 */
3058 	if (sc->rl_testmode) {
3059 		if (sc->rl_type == RL_8169)
3060 			CSR_WRITE_4(sc, RL_TXCFG,
3061 			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON);
3062 		else
3063 			CSR_WRITE_4(sc, RL_TXCFG,
3064 			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON_CPLUS);
3065 	} else
3066 		CSR_WRITE_4(sc, RL_TXCFG, RL_TXCFG_CONFIG);
3067 
3068 	CSR_WRITE_1(sc, RL_EARLY_TX_THRESH, 16);
3069 
3070 	/*
3071 	 * Set the initial RX configuration.
3072 	 */
3073 	re_set_rxmode(sc);
3074 
3075 	/* Configure interrupt moderation. */
3076 	if (sc->rl_type == RL_8169) {
3077 		/* Magic from vendor. */
3078 		CSR_WRITE_2(sc, RL_INTRMOD, 0x5100);
3079 	}
3080 
3081 #ifdef DEVICE_POLLING
3082 	/*
3083 	 * Disable interrupts if we are polling.
3084 	 */
3085 	if (ifp->if_capenable & IFCAP_POLLING)
3086 		CSR_WRITE_2(sc, RL_IMR, 0);
3087 	else	/* otherwise ... */
3088 #endif
3089 
3090 	/*
3091 	 * Enable interrupts.
3092 	 */
3093 	if (sc->rl_testmode)
3094 		CSR_WRITE_2(sc, RL_IMR, 0);
3095 	else
3096 		CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
3097 	CSR_WRITE_2(sc, RL_ISR, RL_INTRS_CPLUS);
3098 
3099 	/* Set initial TX threshold */
3100 	sc->rl_txthresh = RL_TX_THRESH_INIT;
3101 
3102 	/* Start RX/TX process. */
3103 	CSR_WRITE_4(sc, RL_MISSEDPKT, 0);
3104 #ifdef notdef
3105 	/* Enable receiver and transmitter. */
3106 	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
3107 #endif
3108 
3109 	/*
3110 	 * Initialize the timer interrupt register so that
3111 	 * a timer interrupt will be generated once the timer
3112 	 * reaches a certain number of ticks. The timer is
3113 	 * reloaded on each transmit.
3114 	 */
3115 #ifdef RE_TX_MODERATION
3116 	/*
3117 	 * Use timer interrupt register to moderate TX interrupt
3118 	 * moderation, which dramatically improves TX frame rate.
3119 	 */
3120 	if (sc->rl_type == RL_8169)
3121 		CSR_WRITE_4(sc, RL_TIMERINT_8169, 0x800);
3122 	else
3123 		CSR_WRITE_4(sc, RL_TIMERINT, 0x400);
3124 #else
3125 	/*
3126 	 * Use timer interrupt register to moderate RX interrupt
3127 	 * moderation.
3128 	 */
3129 	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) != 0 &&
3130 	    intr_filter == 0) {
3131 		if (sc->rl_type == RL_8169)
3132 			CSR_WRITE_4(sc, RL_TIMERINT_8169,
3133 			    RL_USECS(sc->rl_int_rx_mod));
3134 	} else {
3135 		if (sc->rl_type == RL_8169)
3136 			CSR_WRITE_4(sc, RL_TIMERINT_8169, RL_USECS(0));
3137 	}
3138 #endif
3139 
3140 	/*
3141 	 * For 8169 gigE NICs, set the max allowed RX packet
3142 	 * size so we can receive jumbo frames.
3143 	 */
3144 	if (sc->rl_type == RL_8169) {
3145 		if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
3146 			/*
3147 			 * For controllers that use new jumbo frame scheme,
3148 			 * set maximum size of jumbo frame depedning on
3149 			 * controller revisions.
3150 			 */
3151 			if (ifp->if_mtu > RL_MTU)
3152 				CSR_WRITE_2(sc, RL_MAXRXPKTLEN,
3153 				    sc->rl_hwrev->rl_max_mtu +
3154 				    ETHER_VLAN_ENCAP_LEN + ETHER_HDR_LEN +
3155 				    ETHER_CRC_LEN);
3156 			else
3157 				CSR_WRITE_2(sc, RL_MAXRXPKTLEN,
3158 				    RE_RX_DESC_BUFLEN);
3159 		} else if ((sc->rl_flags & RL_FLAG_PCIE) != 0 &&
3160 		    sc->rl_hwrev->rl_max_mtu == RL_MTU) {
3161 			/* RTL810x has no jumbo frame support. */
3162 			CSR_WRITE_2(sc, RL_MAXRXPKTLEN, RE_RX_DESC_BUFLEN);
3163 		} else
3164 			CSR_WRITE_2(sc, RL_MAXRXPKTLEN, 16383);
3165 	}
3166 
3167 	if (sc->rl_testmode)
3168 		return;
3169 
3170 	mii_mediachg(mii);
3171 
3172 	CSR_WRITE_1(sc, RL_CFG1, CSR_READ_1(sc, RL_CFG1) | RL_CFG1_DRVLOAD);
3173 
3174 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
3175 	ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
3176 
3177 	sc->rl_flags &= ~RL_FLAG_LINK;
3178 	sc->rl_watchdog_timer = 0;
3179 	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
3180 }
3181 
3182 /*
3183  * Set media options.
3184  */
3185 static int
3186 re_ifmedia_upd(struct ifnet *ifp)
3187 {
3188 	struct rl_softc		*sc;
3189 	struct mii_data		*mii;
3190 	int			error;
3191 
3192 	sc = ifp->if_softc;
3193 	mii = device_get_softc(sc->rl_miibus);
3194 	RL_LOCK(sc);
3195 	error = mii_mediachg(mii);
3196 	RL_UNLOCK(sc);
3197 
3198 	return (error);
3199 }
3200 
3201 /*
3202  * Report current media status.
3203  */
3204 static void
3205 re_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
3206 {
3207 	struct rl_softc		*sc;
3208 	struct mii_data		*mii;
3209 
3210 	sc = ifp->if_softc;
3211 	mii = device_get_softc(sc->rl_miibus);
3212 
3213 	RL_LOCK(sc);
3214 	mii_pollstat(mii);
3215 	RL_UNLOCK(sc);
3216 	ifmr->ifm_active = mii->mii_media_active;
3217 	ifmr->ifm_status = mii->mii_media_status;
3218 }
3219 
3220 static int
3221 re_ioctl(struct ifnet *ifp, u_long command, caddr_t data)
3222 {
3223 	struct rl_softc		*sc = ifp->if_softc;
3224 	struct ifreq		*ifr = (struct ifreq *) data;
3225 	struct mii_data		*mii;
3226 	uint32_t		rev;
3227 	int			error = 0;
3228 
3229 	switch (command) {
3230 	case SIOCSIFMTU:
3231 		if (ifr->ifr_mtu < ETHERMIN ||
3232 		    ifr->ifr_mtu > sc->rl_hwrev->rl_max_mtu) {
3233 			error = EINVAL;
3234 			break;
3235 		}
3236 		RL_LOCK(sc);
3237 		if (ifp->if_mtu != ifr->ifr_mtu) {
3238 			ifp->if_mtu = ifr->ifr_mtu;
3239 			if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0 &&
3240 			    (ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
3241 				ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3242 				re_init_locked(sc);
3243 			}
3244 			if (ifp->if_mtu > RL_TSO_MTU &&
3245 			    (ifp->if_capenable & IFCAP_TSO4) != 0) {
3246 				ifp->if_capenable &= ~(IFCAP_TSO4 |
3247 				    IFCAP_VLAN_HWTSO);
3248 				ifp->if_hwassist &= ~CSUM_TSO;
3249 			}
3250 			VLAN_CAPABILITIES(ifp);
3251 		}
3252 		RL_UNLOCK(sc);
3253 		break;
3254 	case SIOCSIFFLAGS:
3255 		RL_LOCK(sc);
3256 		if ((ifp->if_flags & IFF_UP) != 0) {
3257 			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
3258 				if (((ifp->if_flags ^ sc->rl_if_flags)
3259 				    & (IFF_PROMISC | IFF_ALLMULTI)) != 0)
3260 					re_set_rxmode(sc);
3261 			} else
3262 				re_init_locked(sc);
3263 		} else {
3264 			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
3265 				re_stop(sc);
3266 		}
3267 		sc->rl_if_flags = ifp->if_flags;
3268 		RL_UNLOCK(sc);
3269 		break;
3270 	case SIOCADDMULTI:
3271 	case SIOCDELMULTI:
3272 		RL_LOCK(sc);
3273 		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
3274 			re_set_rxmode(sc);
3275 		RL_UNLOCK(sc);
3276 		break;
3277 	case SIOCGIFMEDIA:
3278 	case SIOCSIFMEDIA:
3279 		mii = device_get_softc(sc->rl_miibus);
3280 		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, command);
3281 		break;
3282 	case SIOCSIFCAP:
3283 	    {
3284 		int mask, reinit;
3285 
3286 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
3287 		reinit = 0;
3288 #ifdef DEVICE_POLLING
3289 		if (mask & IFCAP_POLLING) {
3290 			if (ifr->ifr_reqcap & IFCAP_POLLING) {
3291 				error = ether_poll_register(re_poll, ifp);
3292 				if (error)
3293 					return (error);
3294 				RL_LOCK(sc);
3295 				/* Disable interrupts */
3296 				CSR_WRITE_2(sc, RL_IMR, 0x0000);
3297 				ifp->if_capenable |= IFCAP_POLLING;
3298 				RL_UNLOCK(sc);
3299 			} else {
3300 				error = ether_poll_deregister(ifp);
3301 				/* Enable interrupts. */
3302 				RL_LOCK(sc);
3303 				CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
3304 				ifp->if_capenable &= ~IFCAP_POLLING;
3305 				RL_UNLOCK(sc);
3306 			}
3307 		}
3308 #endif /* DEVICE_POLLING */
3309 		if ((mask & IFCAP_TXCSUM) != 0 &&
3310 		    (ifp->if_capabilities & IFCAP_TXCSUM) != 0) {
3311 			ifp->if_capenable ^= IFCAP_TXCSUM;
3312 			if ((ifp->if_capenable & IFCAP_TXCSUM) != 0) {
3313 				rev = sc->rl_hwrev->rl_rev;
3314 				if (rev == RL_HWREV_8168C ||
3315 				    rev == RL_HWREV_8168C_SPIN2)
3316 					ifp->if_hwassist |= CSUM_TCP | CSUM_UDP;
3317 				else
3318 					ifp->if_hwassist |= RE_CSUM_FEATURES;
3319 			} else
3320 				ifp->if_hwassist &= ~RE_CSUM_FEATURES;
3321 			reinit = 1;
3322 		}
3323 		if ((mask & IFCAP_RXCSUM) != 0 &&
3324 		    (ifp->if_capabilities & IFCAP_RXCSUM) != 0) {
3325 			ifp->if_capenable ^= IFCAP_RXCSUM;
3326 			reinit = 1;
3327 		}
3328 		if ((mask & IFCAP_TSO4) != 0 &&
3329 		    (ifp->if_capabilities & IFCAP_TSO) != 0) {
3330 			ifp->if_capenable ^= IFCAP_TSO4;
3331 			if ((IFCAP_TSO4 & ifp->if_capenable) != 0)
3332 				ifp->if_hwassist |= CSUM_TSO;
3333 			else
3334 				ifp->if_hwassist &= ~CSUM_TSO;
3335 			if (ifp->if_mtu > RL_TSO_MTU &&
3336 			    (ifp->if_capenable & IFCAP_TSO4) != 0) {
3337 				ifp->if_capenable &= ~IFCAP_TSO4;
3338 				ifp->if_hwassist &= ~CSUM_TSO;
3339 			}
3340 		}
3341 		if ((mask & IFCAP_VLAN_HWTSO) != 0 &&
3342 		    (ifp->if_capabilities & IFCAP_VLAN_HWTSO) != 0)
3343 			ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
3344 		if ((mask & IFCAP_VLAN_HWTAGGING) != 0 &&
3345 		    (ifp->if_capabilities & IFCAP_VLAN_HWTAGGING) != 0) {
3346 			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
3347 			/* TSO over VLAN requires VLAN hardware tagging. */
3348 			if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) == 0)
3349 				ifp->if_capenable &= ~IFCAP_VLAN_HWTSO;
3350 			reinit = 1;
3351 		}
3352 		if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0 &&
3353 		    (mask & (IFCAP_HWCSUM | IFCAP_TSO4 |
3354 		    IFCAP_VLAN_HWTSO)) != 0)
3355 				reinit = 1;
3356 		if ((mask & IFCAP_WOL) != 0 &&
3357 		    (ifp->if_capabilities & IFCAP_WOL) != 0) {
3358 			if ((mask & IFCAP_WOL_UCAST) != 0)
3359 				ifp->if_capenable ^= IFCAP_WOL_UCAST;
3360 			if ((mask & IFCAP_WOL_MCAST) != 0)
3361 				ifp->if_capenable ^= IFCAP_WOL_MCAST;
3362 			if ((mask & IFCAP_WOL_MAGIC) != 0)
3363 				ifp->if_capenable ^= IFCAP_WOL_MAGIC;
3364 		}
3365 		if (reinit && ifp->if_drv_flags & IFF_DRV_RUNNING) {
3366 			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3367 			re_init(sc);
3368 		}
3369 		VLAN_CAPABILITIES(ifp);
3370 	    }
3371 		break;
3372 	default:
3373 		error = ether_ioctl(ifp, command, data);
3374 		break;
3375 	}
3376 
3377 	return (error);
3378 }
3379 
3380 static void
3381 re_watchdog(struct rl_softc *sc)
3382 {
3383 	struct ifnet		*ifp;
3384 
3385 	RL_LOCK_ASSERT(sc);
3386 
3387 	if (sc->rl_watchdog_timer == 0 || --sc->rl_watchdog_timer != 0)
3388 		return;
3389 
3390 	ifp = sc->rl_ifp;
3391 	re_txeof(sc);
3392 	if (sc->rl_ldata.rl_tx_free == sc->rl_ldata.rl_tx_desc_cnt) {
3393 		if_printf(ifp, "watchdog timeout (missed Tx interrupts) "
3394 		    "-- recovering\n");
3395 		if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
3396 			re_start_locked(ifp);
3397 		return;
3398 	}
3399 
3400 	if_printf(ifp, "watchdog timeout\n");
3401 	ifp->if_oerrors++;
3402 
3403 	re_rxeof(sc, NULL);
3404 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3405 	re_init_locked(sc);
3406 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
3407 		re_start_locked(ifp);
3408 }
3409 
3410 /*
3411  * Stop the adapter and free any mbufs allocated to the
3412  * RX and TX lists.
3413  */
3414 static void
3415 re_stop(struct rl_softc *sc)
3416 {
3417 	int			i;
3418 	struct ifnet		*ifp;
3419 	struct rl_txdesc	*txd;
3420 	struct rl_rxdesc	*rxd;
3421 
3422 	RL_LOCK_ASSERT(sc);
3423 
3424 	ifp = sc->rl_ifp;
3425 
3426 	sc->rl_watchdog_timer = 0;
3427 	callout_stop(&sc->rl_stat_callout);
3428 	ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
3429 
3430 	if ((sc->rl_flags & RL_FLAG_CMDSTOP) != 0)
3431 		CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_STOPREQ | RL_CMD_TX_ENB |
3432 		    RL_CMD_RX_ENB);
3433 	else
3434 		CSR_WRITE_1(sc, RL_COMMAND, 0x00);
3435 	DELAY(1000);
3436 	CSR_WRITE_2(sc, RL_IMR, 0x0000);
3437 	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
3438 
3439 	if (sc->rl_head != NULL) {
3440 		m_freem(sc->rl_head);
3441 		sc->rl_head = sc->rl_tail = NULL;
3442 	}
3443 
3444 	/* Free the TX list buffers. */
3445 
3446 	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
3447 		txd = &sc->rl_ldata.rl_tx_desc[i];
3448 		if (txd->tx_m != NULL) {
3449 			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
3450 			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
3451 			bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag,
3452 			    txd->tx_dmamap);
3453 			m_freem(txd->tx_m);
3454 			txd->tx_m = NULL;
3455 		}
3456 	}
3457 
3458 	/* Free the RX list buffers. */
3459 
3460 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
3461 		rxd = &sc->rl_ldata.rl_rx_desc[i];
3462 		if (rxd->rx_m != NULL) {
3463 			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
3464 			    rxd->rx_dmamap, BUS_DMASYNC_POSTREAD);
3465 			bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag,
3466 			    rxd->rx_dmamap);
3467 			m_freem(rxd->rx_m);
3468 			rxd->rx_m = NULL;
3469 		}
3470 	}
3471 }
3472 
3473 /*
3474  * Device suspend routine.  Stop the interface and save some PCI
3475  * settings in case the BIOS doesn't restore them properly on
3476  * resume.
3477  */
3478 static int
3479 re_suspend(device_t dev)
3480 {
3481 	struct rl_softc		*sc;
3482 
3483 	sc = device_get_softc(dev);
3484 
3485 	RL_LOCK(sc);
3486 	re_stop(sc);
3487 	re_setwol(sc);
3488 	sc->suspended = 1;
3489 	RL_UNLOCK(sc);
3490 
3491 	return (0);
3492 }
3493 
3494 /*
3495  * Device resume routine.  Restore some PCI settings in case the BIOS
3496  * doesn't, re-enable busmastering, and restart the interface if
3497  * appropriate.
3498  */
3499 static int
3500 re_resume(device_t dev)
3501 {
3502 	struct rl_softc		*sc;
3503 	struct ifnet		*ifp;
3504 
3505 	sc = device_get_softc(dev);
3506 
3507 	RL_LOCK(sc);
3508 
3509 	ifp = sc->rl_ifp;
3510 	/* Take controller out of sleep mode. */
3511 	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
3512 		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
3513 			CSR_WRITE_1(sc, RL_GPIO,
3514 			    CSR_READ_1(sc, RL_GPIO) | 0x01);
3515 	}
3516 
3517 	/*
3518 	 * Clear WOL matching such that normal Rx filtering
3519 	 * wouldn't interfere with WOL patterns.
3520 	 */
3521 	re_clrwol(sc);
3522 
3523 	/* reinitialize interface if necessary */
3524 	if (ifp->if_flags & IFF_UP)
3525 		re_init_locked(sc);
3526 
3527 	sc->suspended = 0;
3528 	RL_UNLOCK(sc);
3529 
3530 	return (0);
3531 }
3532 
3533 /*
3534  * Stop all chip I/O so that the kernel's probe routines don't
3535  * get confused by errant DMAs when rebooting.
3536  */
3537 static int
3538 re_shutdown(device_t dev)
3539 {
3540 	struct rl_softc		*sc;
3541 
3542 	sc = device_get_softc(dev);
3543 
3544 	RL_LOCK(sc);
3545 	re_stop(sc);
3546 	/*
3547 	 * Mark interface as down since otherwise we will panic if
3548 	 * interrupt comes in later on, which can happen in some
3549 	 * cases.
3550 	 */
3551 	sc->rl_ifp->if_flags &= ~IFF_UP;
3552 	re_setwol(sc);
3553 	RL_UNLOCK(sc);
3554 
3555 	return (0);
3556 }
3557 
3558 static void
3559 re_setwol(struct rl_softc *sc)
3560 {
3561 	struct ifnet		*ifp;
3562 	int			pmc;
3563 	uint16_t		pmstat;
3564 	uint8_t			v;
3565 
3566 	RL_LOCK_ASSERT(sc);
3567 
3568 	if (pci_find_cap(sc->rl_dev, PCIY_PMG, &pmc) != 0)
3569 		return;
3570 
3571 	ifp = sc->rl_ifp;
3572 	/* Put controller into sleep mode. */
3573 	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
3574 		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
3575 			CSR_WRITE_1(sc, RL_GPIO,
3576 			    CSR_READ_1(sc, RL_GPIO) & ~0x01);
3577 	}
3578 	if ((ifp->if_capenable & IFCAP_WOL) != 0 &&
3579 	    (sc->rl_flags & RL_FLAG_WOLRXENB) != 0)
3580 		CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RX_ENB);
3581 	/* Enable config register write. */
3582 	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
3583 
3584 	/* Enable PME. */
3585 	v = CSR_READ_1(sc, RL_CFG1);
3586 	v &= ~RL_CFG1_PME;
3587 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3588 		v |= RL_CFG1_PME;
3589 	CSR_WRITE_1(sc, RL_CFG1, v);
3590 
3591 	v = CSR_READ_1(sc, RL_CFG3);
3592 	v &= ~(RL_CFG3_WOL_LINK | RL_CFG3_WOL_MAGIC);
3593 	if ((ifp->if_capenable & IFCAP_WOL_MAGIC) != 0)
3594 		v |= RL_CFG3_WOL_MAGIC;
3595 	CSR_WRITE_1(sc, RL_CFG3, v);
3596 
3597 	/* Config register write done. */
3598 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3599 
3600 	v = CSR_READ_1(sc, RL_CFG5);
3601 	v &= ~(RL_CFG5_WOL_BCAST | RL_CFG5_WOL_MCAST | RL_CFG5_WOL_UCAST);
3602 	v &= ~RL_CFG5_WOL_LANWAKE;
3603 	if ((ifp->if_capenable & IFCAP_WOL_UCAST) != 0)
3604 		v |= RL_CFG5_WOL_UCAST;
3605 	if ((ifp->if_capenable & IFCAP_WOL_MCAST) != 0)
3606 		v |= RL_CFG5_WOL_MCAST | RL_CFG5_WOL_BCAST;
3607 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3608 		v |= RL_CFG5_WOL_LANWAKE;
3609 	CSR_WRITE_1(sc, RL_CFG5, v);
3610 
3611 	if ((ifp->if_capenable & IFCAP_WOL) != 0 &&
3612 	    (sc->rl_flags & RL_FLAG_PHYWAKE_PM) != 0)
3613 		CSR_WRITE_1(sc, RL_PMCH, CSR_READ_1(sc, RL_PMCH) & ~0x80);
3614 	/*
3615 	 * It seems that hardware resets its link speed to 100Mbps in
3616 	 * power down mode so switching to 100Mbps in driver is not
3617 	 * needed.
3618 	 */
3619 
3620 	/* Request PME if WOL is requested. */
3621 	pmstat = pci_read_config(sc->rl_dev, pmc + PCIR_POWER_STATUS, 2);
3622 	pmstat &= ~(PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE);
3623 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3624 		pmstat |= PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE;
3625 	pci_write_config(sc->rl_dev, pmc + PCIR_POWER_STATUS, pmstat, 2);
3626 }
3627 
3628 static void
3629 re_clrwol(struct rl_softc *sc)
3630 {
3631 	int			pmc;
3632 	uint8_t			v;
3633 
3634 	RL_LOCK_ASSERT(sc);
3635 
3636 	if (pci_find_cap(sc->rl_dev, PCIY_PMG, &pmc) != 0)
3637 		return;
3638 
3639 	/* Enable config register write. */
3640 	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
3641 
3642 	v = CSR_READ_1(sc, RL_CFG3);
3643 	v &= ~(RL_CFG3_WOL_LINK | RL_CFG3_WOL_MAGIC);
3644 	CSR_WRITE_1(sc, RL_CFG3, v);
3645 
3646 	/* Config register write done. */
3647 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3648 
3649 	v = CSR_READ_1(sc, RL_CFG5);
3650 	v &= ~(RL_CFG5_WOL_BCAST | RL_CFG5_WOL_MCAST | RL_CFG5_WOL_UCAST);
3651 	v &= ~RL_CFG5_WOL_LANWAKE;
3652 	CSR_WRITE_1(sc, RL_CFG5, v);
3653 }
3654 
3655 static void
3656 re_add_sysctls(struct rl_softc *sc)
3657 {
3658 	struct sysctl_ctx_list	*ctx;
3659 	struct sysctl_oid_list	*children;
3660 	int			error;
3661 
3662 	ctx = device_get_sysctl_ctx(sc->rl_dev);
3663 	children = SYSCTL_CHILDREN(device_get_sysctl_tree(sc->rl_dev));
3664 
3665 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "stats",
3666 	    CTLTYPE_INT | CTLFLAG_RW, sc, 0, re_sysctl_stats, "I",
3667 	    "Statistics Information");
3668 	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) == 0)
3669 		return;
3670 
3671 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "int_rx_mod",
3672 	    CTLTYPE_INT | CTLFLAG_RW, &sc->rl_int_rx_mod, 0,
3673 	    sysctl_hw_re_int_mod, "I", "re RX interrupt moderation");
3674 	/* Pull in device tunables. */
3675 	sc->rl_int_rx_mod = RL_TIMER_DEFAULT;
3676 	error = resource_int_value(device_get_name(sc->rl_dev),
3677 	    device_get_unit(sc->rl_dev), "int_rx_mod", &sc->rl_int_rx_mod);
3678 	if (error == 0) {
3679 		if (sc->rl_int_rx_mod < RL_TIMER_MIN ||
3680 		    sc->rl_int_rx_mod > RL_TIMER_MAX) {
3681 			device_printf(sc->rl_dev, "int_rx_mod value out of "
3682 			    "range; using default: %d\n",
3683 			    RL_TIMER_DEFAULT);
3684 			sc->rl_int_rx_mod = RL_TIMER_DEFAULT;
3685 		}
3686 	}
3687 
3688 }
3689 
3690 static int
3691 re_sysctl_stats(SYSCTL_HANDLER_ARGS)
3692 {
3693 	struct rl_softc		*sc;
3694 	struct rl_stats		*stats;
3695 	int			error, i, result;
3696 
3697 	result = -1;
3698 	error = sysctl_handle_int(oidp, &result, 0, req);
3699 	if (error || req->newptr == NULL)
3700 		return (error);
3701 
3702 	if (result == 1) {
3703 		sc = (struct rl_softc *)arg1;
3704 		RL_LOCK(sc);
3705 		if ((sc->rl_ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
3706 			RL_UNLOCK(sc);
3707 			goto done;
3708 		}
3709 		bus_dmamap_sync(sc->rl_ldata.rl_stag,
3710 		    sc->rl_ldata.rl_smap, BUS_DMASYNC_PREREAD);
3711 		CSR_WRITE_4(sc, RL_DUMPSTATS_HI,
3712 		    RL_ADDR_HI(sc->rl_ldata.rl_stats_addr));
3713 		CSR_WRITE_4(sc, RL_DUMPSTATS_LO,
3714 		    RL_ADDR_LO(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 		    RL_DUMPSTATS_START));
3718 		for (i = RL_TIMEOUT; i > 0; i--) {
3719 			if ((CSR_READ_4(sc, RL_DUMPSTATS_LO) &
3720 			    RL_DUMPSTATS_START) == 0)
3721 				break;
3722 			DELAY(1000);
3723 		}
3724 		bus_dmamap_sync(sc->rl_ldata.rl_stag,
3725 		    sc->rl_ldata.rl_smap, BUS_DMASYNC_POSTREAD);
3726 		RL_UNLOCK(sc);
3727 		if (i == 0) {
3728 			device_printf(sc->rl_dev,
3729 			    "DUMP statistics request timedout\n");
3730 			return (ETIMEDOUT);
3731 		}
3732 done:
3733 		stats = sc->rl_ldata.rl_stats;
3734 		printf("%s statistics:\n", device_get_nameunit(sc->rl_dev));
3735 		printf("Tx frames : %ju\n",
3736 		    (uintmax_t)le64toh(stats->rl_tx_pkts));
3737 		printf("Rx frames : %ju\n",
3738 		    (uintmax_t)le64toh(stats->rl_rx_pkts));
3739 		printf("Tx errors : %ju\n",
3740 		    (uintmax_t)le64toh(stats->rl_tx_errs));
3741 		printf("Rx errors : %u\n",
3742 		    le32toh(stats->rl_rx_errs));
3743 		printf("Rx missed frames : %u\n",
3744 		    (uint32_t)le16toh(stats->rl_missed_pkts));
3745 		printf("Rx frame alignment errs : %u\n",
3746 		    (uint32_t)le16toh(stats->rl_rx_framealign_errs));
3747 		printf("Tx single collisions : %u\n",
3748 		    le32toh(stats->rl_tx_onecoll));
3749 		printf("Tx multiple collisions : %u\n",
3750 		    le32toh(stats->rl_tx_multicolls));
3751 		printf("Rx unicast frames : %ju\n",
3752 		    (uintmax_t)le64toh(stats->rl_rx_ucasts));
3753 		printf("Rx broadcast frames : %ju\n",
3754 		    (uintmax_t)le64toh(stats->rl_rx_bcasts));
3755 		printf("Rx multicast frames : %u\n",
3756 		    le32toh(stats->rl_rx_mcasts));
3757 		printf("Tx aborts : %u\n",
3758 		    (uint32_t)le16toh(stats->rl_tx_aborts));
3759 		printf("Tx underruns : %u\n",
3760 		    (uint32_t)le16toh(stats->rl_rx_underruns));
3761 	}
3762 
3763 	return (error);
3764 }
3765 
3766 static int
3767 sysctl_int_range(SYSCTL_HANDLER_ARGS, int low, int high)
3768 {
3769 	int error, value;
3770 
3771 	if (arg1 == NULL)
3772 		return (EINVAL);
3773 	value = *(int *)arg1;
3774 	error = sysctl_handle_int(oidp, &value, 0, req);
3775 	if (error || req->newptr == NULL)
3776 		return (error);
3777 	if (value < low || value > high)
3778 		return (EINVAL);
3779 	*(int *)arg1 = value;
3780 
3781 	return (0);
3782 }
3783 
3784 static int
3785 sysctl_hw_re_int_mod(SYSCTL_HANDLER_ARGS)
3786 {
3787 
3788 	return (sysctl_int_range(oidp, arg1, arg2, req, RL_TIMER_MIN,
3789 	    RL_TIMER_MAX));
3790 }
3791