xref: /freebsd/sys/dev/re/if_re.c (revision 9034852c84a13f0e3b5527e1c886ca94b2863b2b)
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_var.h>
131 #include <net/if_arp.h>
132 #include <net/ethernet.h>
133 #include <net/if_dl.h>
134 #include <net/if_media.h>
135 #include <net/if_types.h>
136 #include <net/if_vlan_var.h>
137 
138 #include <net/bpf.h>
139 
140 #include <machine/bus.h>
141 #include <machine/resource.h>
142 #include <sys/bus.h>
143 #include <sys/rman.h>
144 
145 #include <dev/mii/mii.h>
146 #include <dev/mii/miivar.h>
147 
148 #include <dev/pci/pcireg.h>
149 #include <dev/pci/pcivar.h>
150 
151 #include <dev/rl/if_rlreg.h>
152 
153 MODULE_DEPEND(re, pci, 1, 1, 1);
154 MODULE_DEPEND(re, ether, 1, 1, 1);
155 MODULE_DEPEND(re, miibus, 1, 1, 1);
156 
157 /* "device miibus" required.  See GENERIC if you get errors here. */
158 #include "miibus_if.h"
159 
160 /* Tunables. */
161 static int intr_filter = 0;
162 TUNABLE_INT("hw.re.intr_filter", &intr_filter);
163 static int msi_disable = 0;
164 TUNABLE_INT("hw.re.msi_disable", &msi_disable);
165 static int msix_disable = 0;
166 TUNABLE_INT("hw.re.msix_disable", &msix_disable);
167 static int prefer_iomap = 0;
168 TUNABLE_INT("hw.re.prefer_iomap", &prefer_iomap);
169 
170 #define RE_CSUM_FEATURES    (CSUM_IP | CSUM_TCP | CSUM_UDP)
171 
172 /*
173  * Various supported device vendors/types and their names.
174  */
175 static const struct rl_type re_devs[] = {
176 	{ DLINK_VENDORID, DLINK_DEVICEID_528T, 0,
177 	    "D-Link DGE-528(T) Gigabit Ethernet Adapter" },
178 	{ DLINK_VENDORID, DLINK_DEVICEID_530T_REVC, 0,
179 	    "D-Link DGE-530(T) Gigabit Ethernet Adapter" },
180 	{ RT_VENDORID, RT_DEVICEID_8139, 0,
181 	    "RealTek 8139C+ 10/100BaseTX" },
182 	{ RT_VENDORID, RT_DEVICEID_8101E, 0,
183 	    "RealTek 810xE PCIe 10/100baseTX" },
184 	{ RT_VENDORID, RT_DEVICEID_8168, 0,
185 	    "RealTek 8168/8111 B/C/CP/D/DP/E/F/G PCIe Gigabit Ethernet" },
186 	{ RT_VENDORID, RT_DEVICEID_8169, 0,
187 	    "RealTek 8169/8169S/8169SB(L)/8110S/8110SB(L) Gigabit Ethernet" },
188 	{ RT_VENDORID, RT_DEVICEID_8169SC, 0,
189 	    "RealTek 8169SC/8110SC Single-chip Gigabit Ethernet" },
190 	{ COREGA_VENDORID, COREGA_DEVICEID_CGLAPCIGT, 0,
191 	    "Corega CG-LAPCIGT (RTL8169S) Gigabit Ethernet" },
192 	{ LINKSYS_VENDORID, LINKSYS_DEVICEID_EG1032, 0,
193 	    "Linksys EG1032 (RTL8169S) Gigabit Ethernet" },
194 	{ USR_VENDORID, USR_DEVICEID_997902, 0,
195 	    "US Robotics 997902 (RTL8169S) Gigabit Ethernet" }
196 };
197 
198 static const struct rl_hwrev re_hwrevs[] = {
199 	{ RL_HWREV_8139, RL_8139, "", RL_MTU },
200 	{ RL_HWREV_8139A, RL_8139, "A", RL_MTU },
201 	{ RL_HWREV_8139AG, RL_8139, "A-G", RL_MTU },
202 	{ RL_HWREV_8139B, RL_8139, "B", RL_MTU },
203 	{ RL_HWREV_8130, RL_8139, "8130", RL_MTU },
204 	{ RL_HWREV_8139C, RL_8139, "C", RL_MTU },
205 	{ RL_HWREV_8139D, RL_8139, "8139D/8100B/8100C", RL_MTU },
206 	{ RL_HWREV_8139CPLUS, RL_8139CPLUS, "C+", RL_MTU },
207 	{ RL_HWREV_8168B_SPIN1, RL_8169, "8168", RL_JUMBO_MTU },
208 	{ RL_HWREV_8169, RL_8169, "8169", RL_JUMBO_MTU },
209 	{ RL_HWREV_8169S, RL_8169, "8169S", RL_JUMBO_MTU },
210 	{ RL_HWREV_8110S, RL_8169, "8110S", RL_JUMBO_MTU },
211 	{ RL_HWREV_8169_8110SB, RL_8169, "8169SB/8110SB", RL_JUMBO_MTU },
212 	{ RL_HWREV_8169_8110SC, RL_8169, "8169SC/8110SC", RL_JUMBO_MTU },
213 	{ RL_HWREV_8169_8110SBL, RL_8169, "8169SBL/8110SBL", RL_JUMBO_MTU },
214 	{ RL_HWREV_8169_8110SCE, RL_8169, "8169SC/8110SC", RL_JUMBO_MTU },
215 	{ RL_HWREV_8100, RL_8139, "8100", RL_MTU },
216 	{ RL_HWREV_8101, RL_8139, "8101", RL_MTU },
217 	{ RL_HWREV_8100E, RL_8169, "8100E", RL_MTU },
218 	{ RL_HWREV_8101E, RL_8169, "8101E", RL_MTU },
219 	{ RL_HWREV_8102E, RL_8169, "8102E", RL_MTU },
220 	{ RL_HWREV_8102EL, RL_8169, "8102EL", RL_MTU },
221 	{ RL_HWREV_8102EL_SPIN1, RL_8169, "8102EL", RL_MTU },
222 	{ RL_HWREV_8103E, RL_8169, "8103E", RL_MTU },
223 	{ RL_HWREV_8401E, RL_8169, "8401E", RL_MTU },
224 	{ RL_HWREV_8402, RL_8169, "8402", RL_MTU },
225 	{ RL_HWREV_8105E, RL_8169, "8105E", RL_MTU },
226 	{ RL_HWREV_8105E_SPIN1, RL_8169, "8105E", RL_MTU },
227 	{ RL_HWREV_8106E, RL_8169, "8106E", RL_MTU },
228 	{ RL_HWREV_8168B_SPIN2, RL_8169, "8168", RL_JUMBO_MTU },
229 	{ RL_HWREV_8168B_SPIN3, RL_8169, "8168", RL_JUMBO_MTU },
230 	{ RL_HWREV_8168C, RL_8169, "8168C/8111C", RL_JUMBO_MTU_6K },
231 	{ RL_HWREV_8168C_SPIN2, RL_8169, "8168C/8111C", RL_JUMBO_MTU_6K },
232 	{ RL_HWREV_8168CP, RL_8169, "8168CP/8111CP", RL_JUMBO_MTU_6K },
233 	{ RL_HWREV_8168D, RL_8169, "8168D/8111D", RL_JUMBO_MTU_9K },
234 	{ RL_HWREV_8168DP, RL_8169, "8168DP/8111DP", RL_JUMBO_MTU_9K },
235 	{ RL_HWREV_8168E, RL_8169, "8168E/8111E", RL_JUMBO_MTU_9K},
236 	{ RL_HWREV_8168E_VL, RL_8169, "8168E/8111E-VL", RL_JUMBO_MTU_6K},
237 	{ RL_HWREV_8168EP, RL_8169, "8168EP/8111EP", RL_JUMBO_MTU_9K},
238 	{ RL_HWREV_8168F, RL_8169, "8168F/8111F", RL_JUMBO_MTU_9K},
239 	{ RL_HWREV_8168G, RL_8169, "8168G/8111G", RL_JUMBO_MTU_9K},
240 	{ RL_HWREV_8168GU, RL_8169, "8168GU/8111GU", RL_JUMBO_MTU_9K},
241 	{ RL_HWREV_8168H, RL_8169, "8168H/8111H", RL_JUMBO_MTU_9K},
242 	{ RL_HWREV_8411, RL_8169, "8411", RL_JUMBO_MTU_9K},
243 	{ RL_HWREV_8411B, RL_8169, "8411B", RL_JUMBO_MTU_9K},
244 	{ 0, 0, NULL, 0 }
245 };
246 
247 static int re_probe		(device_t);
248 static int re_attach		(device_t);
249 static int re_detach		(device_t);
250 
251 static int re_encap		(struct rl_softc *, struct mbuf **);
252 
253 static void re_dma_map_addr	(void *, bus_dma_segment_t *, int, int);
254 static int re_allocmem		(device_t, struct rl_softc *);
255 static __inline void re_discard_rxbuf
256 				(struct rl_softc *, int);
257 static int re_newbuf		(struct rl_softc *, int);
258 static int re_jumbo_newbuf	(struct rl_softc *, int);
259 static int re_rx_list_init	(struct rl_softc *);
260 static int re_jrx_list_init	(struct rl_softc *);
261 static int re_tx_list_init	(struct rl_softc *);
262 #ifdef RE_FIXUP_RX
263 static __inline void re_fixup_rx
264 				(struct mbuf *);
265 #endif
266 static int re_rxeof		(struct rl_softc *, int *);
267 static void re_txeof		(struct rl_softc *);
268 #ifdef DEVICE_POLLING
269 static int re_poll		(struct ifnet *, enum poll_cmd, int);
270 static int re_poll_locked	(struct ifnet *, enum poll_cmd, int);
271 #endif
272 static int re_intr		(void *);
273 static void re_intr_msi		(void *);
274 static void re_tick		(void *);
275 static void re_int_task		(void *, int);
276 static void re_start		(struct ifnet *);
277 static void re_start_locked	(struct ifnet *);
278 static int re_ioctl		(struct ifnet *, u_long, caddr_t);
279 static void re_init		(void *);
280 static void re_init_locked	(struct rl_softc *);
281 static void re_stop		(struct rl_softc *);
282 static void re_watchdog		(struct rl_softc *);
283 static int re_suspend		(device_t);
284 static int re_resume		(device_t);
285 static int re_shutdown		(device_t);
286 static int re_ifmedia_upd	(struct ifnet *);
287 static void re_ifmedia_sts	(struct ifnet *, struct ifmediareq *);
288 
289 static void re_eeprom_putbyte	(struct rl_softc *, int);
290 static void re_eeprom_getword	(struct rl_softc *, int, u_int16_t *);
291 static void re_read_eeprom	(struct rl_softc *, caddr_t, int, int);
292 static int re_gmii_readreg	(device_t, int, int);
293 static int re_gmii_writereg	(device_t, int, int, int);
294 
295 static int re_miibus_readreg	(device_t, int, int);
296 static int re_miibus_writereg	(device_t, int, int, int);
297 static void re_miibus_statchg	(device_t);
298 
299 static void re_set_jumbo	(struct rl_softc *, int);
300 static void re_set_rxmode		(struct rl_softc *);
301 static void re_reset		(struct rl_softc *);
302 static void re_setwol		(struct rl_softc *);
303 static void re_clrwol		(struct rl_softc *);
304 static void re_set_linkspeed	(struct rl_softc *);
305 
306 #ifdef DEV_NETMAP	/* see ixgbe.c for details */
307 #include <dev/netmap/if_re_netmap.h>
308 MODULE_DEPEND(re, netmap, 1, 1, 1);
309 #endif /* !DEV_NETMAP */
310 
311 #ifdef RE_DIAG
312 static int re_diag		(struct rl_softc *);
313 #endif
314 
315 static void re_add_sysctls	(struct rl_softc *);
316 static int re_sysctl_stats	(SYSCTL_HANDLER_ARGS);
317 static int sysctl_int_range	(SYSCTL_HANDLER_ARGS, int, int);
318 static int sysctl_hw_re_int_mod	(SYSCTL_HANDLER_ARGS);
319 
320 static device_method_t re_methods[] = {
321 	/* Device interface */
322 	DEVMETHOD(device_probe,		re_probe),
323 	DEVMETHOD(device_attach,	re_attach),
324 	DEVMETHOD(device_detach,	re_detach),
325 	DEVMETHOD(device_suspend,	re_suspend),
326 	DEVMETHOD(device_resume,	re_resume),
327 	DEVMETHOD(device_shutdown,	re_shutdown),
328 
329 	/* MII interface */
330 	DEVMETHOD(miibus_readreg,	re_miibus_readreg),
331 	DEVMETHOD(miibus_writereg,	re_miibus_writereg),
332 	DEVMETHOD(miibus_statchg,	re_miibus_statchg),
333 
334 	DEVMETHOD_END
335 };
336 
337 static driver_t re_driver = {
338 	"re",
339 	re_methods,
340 	sizeof(struct rl_softc)
341 };
342 
343 static devclass_t re_devclass;
344 
345 DRIVER_MODULE(re, pci, re_driver, re_devclass, 0, 0);
346 DRIVER_MODULE(miibus, re, miibus_driver, miibus_devclass, 0, 0);
347 
348 #define EE_SET(x)					\
349 	CSR_WRITE_1(sc, RL_EECMD,			\
350 		CSR_READ_1(sc, RL_EECMD) | x)
351 
352 #define EE_CLR(x)					\
353 	CSR_WRITE_1(sc, RL_EECMD,			\
354 		CSR_READ_1(sc, RL_EECMD) & ~x)
355 
356 /*
357  * Send a read command and address to the EEPROM, check for ACK.
358  */
359 static void
360 re_eeprom_putbyte(struct rl_softc *sc, int addr)
361 {
362 	int			d, i;
363 
364 	d = addr | (RL_9346_READ << sc->rl_eewidth);
365 
366 	/*
367 	 * Feed in each bit and strobe the clock.
368 	 */
369 
370 	for (i = 1 << (sc->rl_eewidth + 3); i; i >>= 1) {
371 		if (d & i) {
372 			EE_SET(RL_EE_DATAIN);
373 		} else {
374 			EE_CLR(RL_EE_DATAIN);
375 		}
376 		DELAY(100);
377 		EE_SET(RL_EE_CLK);
378 		DELAY(150);
379 		EE_CLR(RL_EE_CLK);
380 		DELAY(100);
381 	}
382 }
383 
384 /*
385  * Read a word of data stored in the EEPROM at address 'addr.'
386  */
387 static void
388 re_eeprom_getword(struct rl_softc *sc, int addr, u_int16_t *dest)
389 {
390 	int			i;
391 	u_int16_t		word = 0;
392 
393 	/*
394 	 * Send address of word we want to read.
395 	 */
396 	re_eeprom_putbyte(sc, addr);
397 
398 	/*
399 	 * Start reading bits from EEPROM.
400 	 */
401 	for (i = 0x8000; i; i >>= 1) {
402 		EE_SET(RL_EE_CLK);
403 		DELAY(100);
404 		if (CSR_READ_1(sc, RL_EECMD) & RL_EE_DATAOUT)
405 			word |= i;
406 		EE_CLR(RL_EE_CLK);
407 		DELAY(100);
408 	}
409 
410 	*dest = word;
411 }
412 
413 /*
414  * Read a sequence of words from the EEPROM.
415  */
416 static void
417 re_read_eeprom(struct rl_softc *sc, caddr_t dest, int off, int cnt)
418 {
419 	int			i;
420 	u_int16_t		word = 0, *ptr;
421 
422 	CSR_SETBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM);
423 
424         DELAY(100);
425 
426 	for (i = 0; i < cnt; i++) {
427 		CSR_SETBIT_1(sc, RL_EECMD, RL_EE_SEL);
428 		re_eeprom_getword(sc, off + i, &word);
429 		CSR_CLRBIT_1(sc, RL_EECMD, RL_EE_SEL);
430 		ptr = (u_int16_t *)(dest + (i * 2));
431                 *ptr = word;
432 	}
433 
434 	CSR_CLRBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM);
435 }
436 
437 static int
438 re_gmii_readreg(device_t dev, int phy, int reg)
439 {
440 	struct rl_softc		*sc;
441 	u_int32_t		rval;
442 	int			i;
443 
444 	sc = device_get_softc(dev);
445 
446 	/* Let the rgephy driver read the GMEDIASTAT register */
447 
448 	if (reg == RL_GMEDIASTAT) {
449 		rval = CSR_READ_1(sc, RL_GMEDIASTAT);
450 		return (rval);
451 	}
452 
453 	CSR_WRITE_4(sc, RL_PHYAR, reg << 16);
454 
455 	for (i = 0; i < RL_PHY_TIMEOUT; i++) {
456 		rval = CSR_READ_4(sc, RL_PHYAR);
457 		if (rval & RL_PHYAR_BUSY)
458 			break;
459 		DELAY(25);
460 	}
461 
462 	if (i == RL_PHY_TIMEOUT) {
463 		device_printf(sc->rl_dev, "PHY read failed\n");
464 		return (0);
465 	}
466 
467 	/*
468 	 * Controller requires a 20us delay to process next MDIO request.
469 	 */
470 	DELAY(20);
471 
472 	return (rval & RL_PHYAR_PHYDATA);
473 }
474 
475 static int
476 re_gmii_writereg(device_t dev, int phy, int reg, int data)
477 {
478 	struct rl_softc		*sc;
479 	u_int32_t		rval;
480 	int			i;
481 
482 	sc = device_get_softc(dev);
483 
484 	CSR_WRITE_4(sc, RL_PHYAR, (reg << 16) |
485 	    (data & RL_PHYAR_PHYDATA) | RL_PHYAR_BUSY);
486 
487 	for (i = 0; i < RL_PHY_TIMEOUT; i++) {
488 		rval = CSR_READ_4(sc, RL_PHYAR);
489 		if (!(rval & RL_PHYAR_BUSY))
490 			break;
491 		DELAY(25);
492 	}
493 
494 	if (i == RL_PHY_TIMEOUT) {
495 		device_printf(sc->rl_dev, "PHY write failed\n");
496 		return (0);
497 	}
498 
499 	/*
500 	 * Controller requires a 20us delay to process next MDIO request.
501 	 */
502 	DELAY(20);
503 
504 	return (0);
505 }
506 
507 static int
508 re_miibus_readreg(device_t dev, int phy, int reg)
509 {
510 	struct rl_softc		*sc;
511 	u_int16_t		rval = 0;
512 	u_int16_t		re8139_reg = 0;
513 
514 	sc = device_get_softc(dev);
515 
516 	if (sc->rl_type == RL_8169) {
517 		rval = re_gmii_readreg(dev, phy, reg);
518 		return (rval);
519 	}
520 
521 	switch (reg) {
522 	case MII_BMCR:
523 		re8139_reg = RL_BMCR;
524 		break;
525 	case MII_BMSR:
526 		re8139_reg = RL_BMSR;
527 		break;
528 	case MII_ANAR:
529 		re8139_reg = RL_ANAR;
530 		break;
531 	case MII_ANER:
532 		re8139_reg = RL_ANER;
533 		break;
534 	case MII_ANLPAR:
535 		re8139_reg = RL_LPAR;
536 		break;
537 	case MII_PHYIDR1:
538 	case MII_PHYIDR2:
539 		return (0);
540 	/*
541 	 * Allow the rlphy driver to read the media status
542 	 * register. If we have a link partner which does not
543 	 * support NWAY, this is the register which will tell
544 	 * us the results of parallel detection.
545 	 */
546 	case RL_MEDIASTAT:
547 		rval = CSR_READ_1(sc, RL_MEDIASTAT);
548 		return (rval);
549 	default:
550 		device_printf(sc->rl_dev, "bad phy register\n");
551 		return (0);
552 	}
553 	rval = CSR_READ_2(sc, re8139_reg);
554 	if (sc->rl_type == RL_8139CPLUS && re8139_reg == RL_BMCR) {
555 		/* 8139C+ has different bit layout. */
556 		rval &= ~(BMCR_LOOP | BMCR_ISO);
557 	}
558 	return (rval);
559 }
560 
561 static int
562 re_miibus_writereg(device_t dev, int phy, int reg, int data)
563 {
564 	struct rl_softc		*sc;
565 	u_int16_t		re8139_reg = 0;
566 	int			rval = 0;
567 
568 	sc = device_get_softc(dev);
569 
570 	if (sc->rl_type == RL_8169) {
571 		rval = re_gmii_writereg(dev, phy, reg, data);
572 		return (rval);
573 	}
574 
575 	switch (reg) {
576 	case MII_BMCR:
577 		re8139_reg = RL_BMCR;
578 		if (sc->rl_type == RL_8139CPLUS) {
579 			/* 8139C+ has different bit layout. */
580 			data &= ~(BMCR_LOOP | BMCR_ISO);
581 		}
582 		break;
583 	case MII_BMSR:
584 		re8139_reg = RL_BMSR;
585 		break;
586 	case MII_ANAR:
587 		re8139_reg = RL_ANAR;
588 		break;
589 	case MII_ANER:
590 		re8139_reg = RL_ANER;
591 		break;
592 	case MII_ANLPAR:
593 		re8139_reg = RL_LPAR;
594 		break;
595 	case MII_PHYIDR1:
596 	case MII_PHYIDR2:
597 		return (0);
598 		break;
599 	default:
600 		device_printf(sc->rl_dev, "bad phy register\n");
601 		return (0);
602 	}
603 	CSR_WRITE_2(sc, re8139_reg, data);
604 	return (0);
605 }
606 
607 static void
608 re_miibus_statchg(device_t dev)
609 {
610 	struct rl_softc		*sc;
611 	struct ifnet		*ifp;
612 	struct mii_data		*mii;
613 
614 	sc = device_get_softc(dev);
615 	mii = device_get_softc(sc->rl_miibus);
616 	ifp = sc->rl_ifp;
617 	if (mii == NULL || ifp == NULL ||
618 	    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
619 		return;
620 
621 	sc->rl_flags &= ~RL_FLAG_LINK;
622 	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
623 	    (IFM_ACTIVE | IFM_AVALID)) {
624 		switch (IFM_SUBTYPE(mii->mii_media_active)) {
625 		case IFM_10_T:
626 		case IFM_100_TX:
627 			sc->rl_flags |= RL_FLAG_LINK;
628 			break;
629 		case IFM_1000_T:
630 			if ((sc->rl_flags & RL_FLAG_FASTETHER) != 0)
631 				break;
632 			sc->rl_flags |= RL_FLAG_LINK;
633 			break;
634 		default:
635 			break;
636 		}
637 	}
638 	/*
639 	 * RealTek controllers does not provide any interface to
640 	 * Tx/Rx MACs for resolved speed, duplex and flow-control
641 	 * parameters.
642 	 */
643 }
644 
645 /*
646  * Set the RX configuration and 64-bit multicast hash filter.
647  */
648 static void
649 re_set_rxmode(struct rl_softc *sc)
650 {
651 	struct ifnet		*ifp;
652 	struct ifmultiaddr	*ifma;
653 	uint32_t		hashes[2] = { 0, 0 };
654 	uint32_t		h, rxfilt;
655 
656 	RL_LOCK_ASSERT(sc);
657 
658 	ifp = sc->rl_ifp;
659 
660 	rxfilt = RL_RXCFG_CONFIG | RL_RXCFG_RX_INDIV | RL_RXCFG_RX_BROAD;
661 	if ((sc->rl_flags & RL_FLAG_EARLYOFF) != 0)
662 		rxfilt |= RL_RXCFG_EARLYOFF;
663 	else if ((sc->rl_flags & RL_FLAG_EARLYOFFV2) != 0)
664 		rxfilt |= RL_RXCFG_EARLYOFFV2;
665 
666 	if (ifp->if_flags & (IFF_ALLMULTI | IFF_PROMISC)) {
667 		if (ifp->if_flags & IFF_PROMISC)
668 			rxfilt |= RL_RXCFG_RX_ALLPHYS;
669 		/*
670 		 * Unlike other hardwares, we have to explicitly set
671 		 * RL_RXCFG_RX_MULTI to receive multicast frames in
672 		 * promiscuous mode.
673 		 */
674 		rxfilt |= RL_RXCFG_RX_MULTI;
675 		hashes[0] = hashes[1] = 0xffffffff;
676 		goto done;
677 	}
678 
679 	if_maddr_rlock(ifp);
680 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
681 		if (ifma->ifma_addr->sa_family != AF_LINK)
682 			continue;
683 		h = ether_crc32_be(LLADDR((struct sockaddr_dl *)
684 		    ifma->ifma_addr), ETHER_ADDR_LEN) >> 26;
685 		if (h < 32)
686 			hashes[0] |= (1 << h);
687 		else
688 			hashes[1] |= (1 << (h - 32));
689 	}
690 	if_maddr_runlock(ifp);
691 
692 	if (hashes[0] != 0 || hashes[1] != 0) {
693 		/*
694 		 * For some unfathomable reason, RealTek decided to
695 		 * reverse the order of the multicast hash registers
696 		 * in the PCI Express parts.  This means we have to
697 		 * write the hash pattern in reverse order for those
698 		 * devices.
699 		 */
700 		if ((sc->rl_flags & RL_FLAG_PCIE) != 0) {
701 			h = bswap32(hashes[0]);
702 			hashes[0] = bswap32(hashes[1]);
703 			hashes[1] = h;
704 		}
705 		rxfilt |= RL_RXCFG_RX_MULTI;
706 	}
707 
708 	if  (sc->rl_hwrev->rl_rev == RL_HWREV_8168F) {
709 		/* Disable multicast filtering due to silicon bug. */
710 		hashes[0] = 0xffffffff;
711 		hashes[1] = 0xffffffff;
712 	}
713 
714 done:
715 	CSR_WRITE_4(sc, RL_MAR0, hashes[0]);
716 	CSR_WRITE_4(sc, RL_MAR4, hashes[1]);
717 	CSR_WRITE_4(sc, RL_RXCFG, rxfilt);
718 }
719 
720 static void
721 re_reset(struct rl_softc *sc)
722 {
723 	int			i;
724 
725 	RL_LOCK_ASSERT(sc);
726 
727 	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RESET);
728 
729 	for (i = 0; i < RL_TIMEOUT; i++) {
730 		DELAY(10);
731 		if (!(CSR_READ_1(sc, RL_COMMAND) & RL_CMD_RESET))
732 			break;
733 	}
734 	if (i == RL_TIMEOUT)
735 		device_printf(sc->rl_dev, "reset never completed!\n");
736 
737 	if ((sc->rl_flags & RL_FLAG_MACRESET) != 0)
738 		CSR_WRITE_1(sc, 0x82, 1);
739 	if (sc->rl_hwrev->rl_rev == RL_HWREV_8169S)
740 		re_gmii_writereg(sc->rl_dev, 1, 0x0b, 0);
741 }
742 
743 #ifdef RE_DIAG
744 
745 /*
746  * The following routine is designed to test for a defect on some
747  * 32-bit 8169 cards. Some of these NICs have the REQ64# and ACK64#
748  * lines connected to the bus, however for a 32-bit only card, they
749  * should be pulled high. The result of this defect is that the
750  * NIC will not work right if you plug it into a 64-bit slot: DMA
751  * operations will be done with 64-bit transfers, which will fail
752  * because the 64-bit data lines aren't connected.
753  *
754  * There's no way to work around this (short of talking a soldering
755  * iron to the board), however we can detect it. The method we use
756  * here is to put the NIC into digital loopback mode, set the receiver
757  * to promiscuous mode, and then try to send a frame. We then compare
758  * the frame data we sent to what was received. If the data matches,
759  * then the NIC is working correctly, otherwise we know the user has
760  * a defective NIC which has been mistakenly plugged into a 64-bit PCI
761  * slot. In the latter case, there's no way the NIC can work correctly,
762  * so we print out a message on the console and abort the device attach.
763  */
764 
765 static int
766 re_diag(struct rl_softc *sc)
767 {
768 	struct ifnet		*ifp = sc->rl_ifp;
769 	struct mbuf		*m0;
770 	struct ether_header	*eh;
771 	struct rl_desc		*cur_rx;
772 	u_int16_t		status;
773 	u_int32_t		rxstat;
774 	int			total_len, i, error = 0, phyaddr;
775 	u_int8_t		dst[] = { 0x00, 'h', 'e', 'l', 'l', 'o' };
776 	u_int8_t		src[] = { 0x00, 'w', 'o', 'r', 'l', 'd' };
777 
778 	/* Allocate a single mbuf */
779 	MGETHDR(m0, M_NOWAIT, MT_DATA);
780 	if (m0 == NULL)
781 		return (ENOBUFS);
782 
783 	RL_LOCK(sc);
784 
785 	/*
786 	 * Initialize the NIC in test mode. This sets the chip up
787 	 * so that it can send and receive frames, but performs the
788 	 * following special functions:
789 	 * - Puts receiver in promiscuous mode
790 	 * - Enables digital loopback mode
791 	 * - Leaves interrupts turned off
792 	 */
793 
794 	ifp->if_flags |= IFF_PROMISC;
795 	sc->rl_testmode = 1;
796 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
797 	re_init_locked(sc);
798 	sc->rl_flags |= RL_FLAG_LINK;
799 	if (sc->rl_type == RL_8169)
800 		phyaddr = 1;
801 	else
802 		phyaddr = 0;
803 
804 	re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_RESET);
805 	for (i = 0; i < RL_TIMEOUT; i++) {
806 		status = re_miibus_readreg(sc->rl_dev, phyaddr, MII_BMCR);
807 		if (!(status & BMCR_RESET))
808 			break;
809 	}
810 
811 	re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_LOOP);
812 	CSR_WRITE_2(sc, RL_ISR, RL_INTRS);
813 
814 	DELAY(100000);
815 
816 	/* Put some data in the mbuf */
817 
818 	eh = mtod(m0, struct ether_header *);
819 	bcopy ((char *)&dst, eh->ether_dhost, ETHER_ADDR_LEN);
820 	bcopy ((char *)&src, eh->ether_shost, ETHER_ADDR_LEN);
821 	eh->ether_type = htons(ETHERTYPE_IP);
822 	m0->m_pkthdr.len = m0->m_len = ETHER_MIN_LEN - ETHER_CRC_LEN;
823 
824 	/*
825 	 * Queue the packet, start transmission.
826 	 * Note: IF_HANDOFF() ultimately calls re_start() for us.
827 	 */
828 
829 	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
830 	RL_UNLOCK(sc);
831 	/* XXX: re_diag must not be called when in ALTQ mode */
832 	IF_HANDOFF(&ifp->if_snd, m0, ifp);
833 	RL_LOCK(sc);
834 	m0 = NULL;
835 
836 	/* Wait for it to propagate through the chip */
837 
838 	DELAY(100000);
839 	for (i = 0; i < RL_TIMEOUT; i++) {
840 		status = CSR_READ_2(sc, RL_ISR);
841 		CSR_WRITE_2(sc, RL_ISR, status);
842 		if ((status & (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK)) ==
843 		    (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK))
844 			break;
845 		DELAY(10);
846 	}
847 
848 	if (i == RL_TIMEOUT) {
849 		device_printf(sc->rl_dev,
850 		    "diagnostic failed, failed to receive packet in"
851 		    " loopback mode\n");
852 		error = EIO;
853 		goto done;
854 	}
855 
856 	/*
857 	 * The packet should have been dumped into the first
858 	 * entry in the RX DMA ring. Grab it from there.
859 	 */
860 
861 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
862 	    sc->rl_ldata.rl_rx_list_map,
863 	    BUS_DMASYNC_POSTREAD);
864 	bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag,
865 	    sc->rl_ldata.rl_rx_desc[0].rx_dmamap,
866 	    BUS_DMASYNC_POSTREAD);
867 	bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag,
868 	    sc->rl_ldata.rl_rx_desc[0].rx_dmamap);
869 
870 	m0 = sc->rl_ldata.rl_rx_desc[0].rx_m;
871 	sc->rl_ldata.rl_rx_desc[0].rx_m = NULL;
872 	eh = mtod(m0, struct ether_header *);
873 
874 	cur_rx = &sc->rl_ldata.rl_rx_list[0];
875 	total_len = RL_RXBYTES(cur_rx);
876 	rxstat = le32toh(cur_rx->rl_cmdstat);
877 
878 	if (total_len != ETHER_MIN_LEN) {
879 		device_printf(sc->rl_dev,
880 		    "diagnostic failed, received short packet\n");
881 		error = EIO;
882 		goto done;
883 	}
884 
885 	/* Test that the received packet data matches what we sent. */
886 
887 	if (bcmp((char *)&eh->ether_dhost, (char *)&dst, ETHER_ADDR_LEN) ||
888 	    bcmp((char *)&eh->ether_shost, (char *)&src, ETHER_ADDR_LEN) ||
889 	    ntohs(eh->ether_type) != ETHERTYPE_IP) {
890 		device_printf(sc->rl_dev, "WARNING, DMA FAILURE!\n");
891 		device_printf(sc->rl_dev, "expected TX data: %6D/%6D/0x%x\n",
892 		    dst, ":", src, ":", ETHERTYPE_IP);
893 		device_printf(sc->rl_dev, "received RX data: %6D/%6D/0x%x\n",
894 		    eh->ether_dhost, ":", eh->ether_shost, ":",
895 		    ntohs(eh->ether_type));
896 		device_printf(sc->rl_dev, "You may have a defective 32-bit "
897 		    "NIC plugged into a 64-bit PCI slot.\n");
898 		device_printf(sc->rl_dev, "Please re-install the NIC in a "
899 		    "32-bit slot for proper operation.\n");
900 		device_printf(sc->rl_dev, "Read the re(4) man page for more "
901 		    "details.\n");
902 		error = EIO;
903 	}
904 
905 done:
906 	/* Turn interface off, release resources */
907 
908 	sc->rl_testmode = 0;
909 	sc->rl_flags &= ~RL_FLAG_LINK;
910 	ifp->if_flags &= ~IFF_PROMISC;
911 	re_stop(sc);
912 	if (m0 != NULL)
913 		m_freem(m0);
914 
915 	RL_UNLOCK(sc);
916 
917 	return (error);
918 }
919 
920 #endif
921 
922 /*
923  * Probe for a RealTek 8139C+/8169/8110 chip. Check the PCI vendor and device
924  * IDs against our list and return a device name if we find a match.
925  */
926 static int
927 re_probe(device_t dev)
928 {
929 	const struct rl_type	*t;
930 	uint16_t		devid, vendor;
931 	uint16_t		revid, sdevid;
932 	int			i;
933 
934 	vendor = pci_get_vendor(dev);
935 	devid = pci_get_device(dev);
936 	revid = pci_get_revid(dev);
937 	sdevid = pci_get_subdevice(dev);
938 
939 	if (vendor == LINKSYS_VENDORID && devid == LINKSYS_DEVICEID_EG1032) {
940 		if (sdevid != LINKSYS_SUBDEVICE_EG1032_REV3) {
941 			/*
942 			 * Only attach to rev. 3 of the Linksys EG1032 adapter.
943 			 * Rev. 2 is supported by sk(4).
944 			 */
945 			return (ENXIO);
946 		}
947 	}
948 
949 	if (vendor == RT_VENDORID && devid == RT_DEVICEID_8139) {
950 		if (revid != 0x20) {
951 			/* 8139, let rl(4) take care of this device. */
952 			return (ENXIO);
953 		}
954 	}
955 
956 	t = re_devs;
957 	for (i = 0; i < sizeof(re_devs) / sizeof(re_devs[0]); i++, t++) {
958 		if (vendor == t->rl_vid && devid == t->rl_did) {
959 			device_set_desc(dev, t->rl_name);
960 			return (BUS_PROBE_DEFAULT);
961 		}
962 	}
963 
964 	return (ENXIO);
965 }
966 
967 /*
968  * Map a single buffer address.
969  */
970 
971 static void
972 re_dma_map_addr(void *arg, bus_dma_segment_t *segs, int nseg, int error)
973 {
974 	bus_addr_t		*addr;
975 
976 	if (error)
977 		return;
978 
979 	KASSERT(nseg == 1, ("too many DMA segments, %d should be 1", nseg));
980 	addr = arg;
981 	*addr = segs->ds_addr;
982 }
983 
984 static int
985 re_allocmem(device_t dev, struct rl_softc *sc)
986 {
987 	bus_addr_t		lowaddr;
988 	bus_size_t		rx_list_size, tx_list_size;
989 	int			error;
990 	int			i;
991 
992 	rx_list_size = sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc);
993 	tx_list_size = sc->rl_ldata.rl_tx_desc_cnt * sizeof(struct rl_desc);
994 
995 	/*
996 	 * Allocate the parent bus DMA tag appropriate for PCI.
997 	 * In order to use DAC, RL_CPLUSCMD_PCI_DAC bit of RL_CPLUS_CMD
998 	 * register should be set. However some RealTek chips are known
999 	 * to be buggy on DAC handling, therefore disable DAC by limiting
1000 	 * DMA address space to 32bit. PCIe variants of RealTek chips
1001 	 * may not have the limitation.
1002 	 */
1003 	lowaddr = BUS_SPACE_MAXADDR;
1004 	if ((sc->rl_flags & RL_FLAG_PCIE) == 0)
1005 		lowaddr = BUS_SPACE_MAXADDR_32BIT;
1006 	error = bus_dma_tag_create(bus_get_dma_tag(dev), 1, 0,
1007 	    lowaddr, BUS_SPACE_MAXADDR, NULL, NULL,
1008 	    BUS_SPACE_MAXSIZE_32BIT, 0, BUS_SPACE_MAXSIZE_32BIT, 0,
1009 	    NULL, NULL, &sc->rl_parent_tag);
1010 	if (error) {
1011 		device_printf(dev, "could not allocate parent DMA tag\n");
1012 		return (error);
1013 	}
1014 
1015 	/*
1016 	 * Allocate map for TX mbufs.
1017 	 */
1018 	error = bus_dma_tag_create(sc->rl_parent_tag, 1, 0,
1019 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL,
1020 	    NULL, MCLBYTES * RL_NTXSEGS, RL_NTXSEGS, 4096, 0,
1021 	    NULL, NULL, &sc->rl_ldata.rl_tx_mtag);
1022 	if (error) {
1023 		device_printf(dev, "could not allocate TX DMA tag\n");
1024 		return (error);
1025 	}
1026 
1027 	/*
1028 	 * Allocate map for RX mbufs.
1029 	 */
1030 
1031 	if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
1032 		error = bus_dma_tag_create(sc->rl_parent_tag, sizeof(uint64_t),
1033 		    0, BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
1034 		    MJUM9BYTES, 1, MJUM9BYTES, 0, NULL, NULL,
1035 		    &sc->rl_ldata.rl_jrx_mtag);
1036 		if (error) {
1037 			device_printf(dev,
1038 			    "could not allocate jumbo RX DMA tag\n");
1039 			return (error);
1040 		}
1041 	}
1042 	error = bus_dma_tag_create(sc->rl_parent_tag, sizeof(uint64_t), 0,
1043 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
1044 	    MCLBYTES, 1, MCLBYTES, 0, NULL, NULL, &sc->rl_ldata.rl_rx_mtag);
1045 	if (error) {
1046 		device_printf(dev, "could not allocate RX DMA tag\n");
1047 		return (error);
1048 	}
1049 
1050 	/*
1051 	 * Allocate map for TX descriptor list.
1052 	 */
1053 	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
1054 	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1055 	    NULL, tx_list_size, 1, tx_list_size, 0,
1056 	    NULL, NULL, &sc->rl_ldata.rl_tx_list_tag);
1057 	if (error) {
1058 		device_printf(dev, "could not allocate TX DMA ring tag\n");
1059 		return (error);
1060 	}
1061 
1062 	/* Allocate DMA'able memory for the TX ring */
1063 
1064 	error = bus_dmamem_alloc(sc->rl_ldata.rl_tx_list_tag,
1065 	    (void **)&sc->rl_ldata.rl_tx_list,
1066 	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1067 	    &sc->rl_ldata.rl_tx_list_map);
1068 	if (error) {
1069 		device_printf(dev, "could not allocate TX DMA ring\n");
1070 		return (error);
1071 	}
1072 
1073 	/* Load the map for the TX ring. */
1074 
1075 	sc->rl_ldata.rl_tx_list_addr = 0;
1076 	error = bus_dmamap_load(sc->rl_ldata.rl_tx_list_tag,
1077 	     sc->rl_ldata.rl_tx_list_map, sc->rl_ldata.rl_tx_list,
1078 	     tx_list_size, re_dma_map_addr,
1079 	     &sc->rl_ldata.rl_tx_list_addr, BUS_DMA_NOWAIT);
1080 	if (error != 0 || sc->rl_ldata.rl_tx_list_addr == 0) {
1081 		device_printf(dev, "could not load TX DMA ring\n");
1082 		return (ENOMEM);
1083 	}
1084 
1085 	/* Create DMA maps for TX buffers */
1086 
1087 	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
1088 		error = bus_dmamap_create(sc->rl_ldata.rl_tx_mtag, 0,
1089 		    &sc->rl_ldata.rl_tx_desc[i].tx_dmamap);
1090 		if (error) {
1091 			device_printf(dev, "could not create DMA map for TX\n");
1092 			return (error);
1093 		}
1094 	}
1095 
1096 	/*
1097 	 * Allocate map for RX descriptor list.
1098 	 */
1099 	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
1100 	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1101 	    NULL, rx_list_size, 1, rx_list_size, 0,
1102 	    NULL, NULL, &sc->rl_ldata.rl_rx_list_tag);
1103 	if (error) {
1104 		device_printf(dev, "could not create RX DMA ring tag\n");
1105 		return (error);
1106 	}
1107 
1108 	/* Allocate DMA'able memory for the RX ring */
1109 
1110 	error = bus_dmamem_alloc(sc->rl_ldata.rl_rx_list_tag,
1111 	    (void **)&sc->rl_ldata.rl_rx_list,
1112 	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1113 	    &sc->rl_ldata.rl_rx_list_map);
1114 	if (error) {
1115 		device_printf(dev, "could not allocate RX DMA ring\n");
1116 		return (error);
1117 	}
1118 
1119 	/* Load the map for the RX ring. */
1120 
1121 	sc->rl_ldata.rl_rx_list_addr = 0;
1122 	error = bus_dmamap_load(sc->rl_ldata.rl_rx_list_tag,
1123 	     sc->rl_ldata.rl_rx_list_map, sc->rl_ldata.rl_rx_list,
1124 	     rx_list_size, re_dma_map_addr,
1125 	     &sc->rl_ldata.rl_rx_list_addr, BUS_DMA_NOWAIT);
1126 	if (error != 0 || sc->rl_ldata.rl_rx_list_addr == 0) {
1127 		device_printf(dev, "could not load RX DMA ring\n");
1128 		return (ENOMEM);
1129 	}
1130 
1131 	/* Create DMA maps for RX buffers */
1132 
1133 	if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
1134 		error = bus_dmamap_create(sc->rl_ldata.rl_jrx_mtag, 0,
1135 		    &sc->rl_ldata.rl_jrx_sparemap);
1136 		if (error) {
1137 			device_printf(dev,
1138 			    "could not create spare DMA map for jumbo RX\n");
1139 			return (error);
1140 		}
1141 		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1142 			error = bus_dmamap_create(sc->rl_ldata.rl_jrx_mtag, 0,
1143 			    &sc->rl_ldata.rl_jrx_desc[i].rx_dmamap);
1144 			if (error) {
1145 				device_printf(dev,
1146 				    "could not create DMA map for jumbo RX\n");
1147 				return (error);
1148 			}
1149 		}
1150 	}
1151 	error = bus_dmamap_create(sc->rl_ldata.rl_rx_mtag, 0,
1152 	    &sc->rl_ldata.rl_rx_sparemap);
1153 	if (error) {
1154 		device_printf(dev, "could not create spare DMA map for RX\n");
1155 		return (error);
1156 	}
1157 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1158 		error = bus_dmamap_create(sc->rl_ldata.rl_rx_mtag, 0,
1159 		    &sc->rl_ldata.rl_rx_desc[i].rx_dmamap);
1160 		if (error) {
1161 			device_printf(dev, "could not create DMA map for RX\n");
1162 			return (error);
1163 		}
1164 	}
1165 
1166 	/* Create DMA map for statistics. */
1167 	error = bus_dma_tag_create(sc->rl_parent_tag, RL_DUMP_ALIGN, 0,
1168 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
1169 	    sizeof(struct rl_stats), 1, sizeof(struct rl_stats), 0, NULL, NULL,
1170 	    &sc->rl_ldata.rl_stag);
1171 	if (error) {
1172 		device_printf(dev, "could not create statistics DMA tag\n");
1173 		return (error);
1174 	}
1175 	/* Allocate DMA'able memory for statistics. */
1176 	error = bus_dmamem_alloc(sc->rl_ldata.rl_stag,
1177 	    (void **)&sc->rl_ldata.rl_stats,
1178 	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1179 	    &sc->rl_ldata.rl_smap);
1180 	if (error) {
1181 		device_printf(dev,
1182 		    "could not allocate statistics DMA memory\n");
1183 		return (error);
1184 	}
1185 	/* Load the map for statistics. */
1186 	sc->rl_ldata.rl_stats_addr = 0;
1187 	error = bus_dmamap_load(sc->rl_ldata.rl_stag, sc->rl_ldata.rl_smap,
1188 	    sc->rl_ldata.rl_stats, sizeof(struct rl_stats), re_dma_map_addr,
1189 	     &sc->rl_ldata.rl_stats_addr, BUS_DMA_NOWAIT);
1190 	if (error != 0 || sc->rl_ldata.rl_stats_addr == 0) {
1191 		device_printf(dev, "could not load statistics DMA memory\n");
1192 		return (ENOMEM);
1193 	}
1194 
1195 	return (0);
1196 }
1197 
1198 /*
1199  * Attach the interface. Allocate softc structures, do ifmedia
1200  * setup and ethernet/BPF attach.
1201  */
1202 static int
1203 re_attach(device_t dev)
1204 {
1205 	u_char			eaddr[ETHER_ADDR_LEN];
1206 	u_int16_t		as[ETHER_ADDR_LEN / 2];
1207 	struct rl_softc		*sc;
1208 	struct ifnet		*ifp;
1209 	const struct rl_hwrev	*hw_rev;
1210 	u_int32_t		cap, ctl;
1211 	int			hwrev;
1212 	u_int16_t		devid, re_did = 0;
1213 	int			error = 0, i, phy, rid;
1214 	int			msic, msixc, reg;
1215 	uint8_t			cfg;
1216 
1217 	sc = device_get_softc(dev);
1218 	sc->rl_dev = dev;
1219 
1220 	mtx_init(&sc->rl_mtx, device_get_nameunit(dev), MTX_NETWORK_LOCK,
1221 	    MTX_DEF);
1222 	callout_init_mtx(&sc->rl_stat_callout, &sc->rl_mtx, 0);
1223 
1224 	/*
1225 	 * Map control/status registers.
1226 	 */
1227 	pci_enable_busmaster(dev);
1228 
1229 	devid = pci_get_device(dev);
1230 	/*
1231 	 * Prefer memory space register mapping over IO space.
1232 	 * Because RTL8169SC does not seem to work when memory mapping
1233 	 * is used always activate io mapping.
1234 	 */
1235 	if (devid == RT_DEVICEID_8169SC)
1236 		prefer_iomap = 1;
1237 	if (prefer_iomap == 0) {
1238 		sc->rl_res_id = PCIR_BAR(1);
1239 		sc->rl_res_type = SYS_RES_MEMORY;
1240 		/* RTL8168/8101E seems to use different BARs. */
1241 		if (devid == RT_DEVICEID_8168 || devid == RT_DEVICEID_8101E)
1242 			sc->rl_res_id = PCIR_BAR(2);
1243 	} else {
1244 		sc->rl_res_id = PCIR_BAR(0);
1245 		sc->rl_res_type = SYS_RES_IOPORT;
1246 	}
1247 	sc->rl_res = bus_alloc_resource_any(dev, sc->rl_res_type,
1248 	    &sc->rl_res_id, RF_ACTIVE);
1249 	if (sc->rl_res == NULL && prefer_iomap == 0) {
1250 		sc->rl_res_id = PCIR_BAR(0);
1251 		sc->rl_res_type = SYS_RES_IOPORT;
1252 		sc->rl_res = bus_alloc_resource_any(dev, sc->rl_res_type,
1253 		    &sc->rl_res_id, RF_ACTIVE);
1254 	}
1255 	if (sc->rl_res == NULL) {
1256 		device_printf(dev, "couldn't map ports/memory\n");
1257 		error = ENXIO;
1258 		goto fail;
1259 	}
1260 
1261 	sc->rl_btag = rman_get_bustag(sc->rl_res);
1262 	sc->rl_bhandle = rman_get_bushandle(sc->rl_res);
1263 
1264 	msic = pci_msi_count(dev);
1265 	msixc = pci_msix_count(dev);
1266 	if (pci_find_cap(dev, PCIY_EXPRESS, &reg) == 0) {
1267 		sc->rl_flags |= RL_FLAG_PCIE;
1268 		sc->rl_expcap = reg;
1269 	}
1270 	if (bootverbose) {
1271 		device_printf(dev, "MSI count : %d\n", msic);
1272 		device_printf(dev, "MSI-X count : %d\n", msixc);
1273 	}
1274 	if (msix_disable > 0)
1275 		msixc = 0;
1276 	if (msi_disable > 0)
1277 		msic = 0;
1278 	/* Prefer MSI-X to MSI. */
1279 	if (msixc > 0) {
1280 		msixc = RL_MSI_MESSAGES;
1281 		rid = PCIR_BAR(4);
1282 		sc->rl_res_pba = bus_alloc_resource_any(dev, SYS_RES_MEMORY,
1283 		    &rid, RF_ACTIVE);
1284 		if (sc->rl_res_pba == NULL) {
1285 			device_printf(sc->rl_dev,
1286 			    "could not allocate MSI-X PBA resource\n");
1287 		}
1288 		if (sc->rl_res_pba != NULL &&
1289 		    pci_alloc_msix(dev, &msixc) == 0) {
1290 			if (msixc == RL_MSI_MESSAGES) {
1291 				device_printf(dev, "Using %d MSI-X message\n",
1292 				    msixc);
1293 				sc->rl_flags |= RL_FLAG_MSIX;
1294 			} else
1295 				pci_release_msi(dev);
1296 		}
1297 		if ((sc->rl_flags & RL_FLAG_MSIX) == 0) {
1298 			if (sc->rl_res_pba != NULL)
1299 				bus_release_resource(dev, SYS_RES_MEMORY, rid,
1300 				    sc->rl_res_pba);
1301 			sc->rl_res_pba = NULL;
1302 			msixc = 0;
1303 		}
1304 	}
1305 	/* Prefer MSI to INTx. */
1306 	if (msixc == 0 && msic > 0) {
1307 		msic = RL_MSI_MESSAGES;
1308 		if (pci_alloc_msi(dev, &msic) == 0) {
1309 			if (msic == RL_MSI_MESSAGES) {
1310 				device_printf(dev, "Using %d MSI message\n",
1311 				    msic);
1312 				sc->rl_flags |= RL_FLAG_MSI;
1313 				/* Explicitly set MSI enable bit. */
1314 				CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1315 				cfg = CSR_READ_1(sc, RL_CFG2);
1316 				cfg |= RL_CFG2_MSI;
1317 				CSR_WRITE_1(sc, RL_CFG2, cfg);
1318 				CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1319 			} else
1320 				pci_release_msi(dev);
1321 		}
1322 		if ((sc->rl_flags & RL_FLAG_MSI) == 0)
1323 			msic = 0;
1324 	}
1325 
1326 	/* Allocate interrupt */
1327 	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) == 0) {
1328 		rid = 0;
1329 		sc->rl_irq[0] = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
1330 		    RF_SHAREABLE | RF_ACTIVE);
1331 		if (sc->rl_irq[0] == NULL) {
1332 			device_printf(dev, "couldn't allocate IRQ resources\n");
1333 			error = ENXIO;
1334 			goto fail;
1335 		}
1336 	} else {
1337 		for (i = 0, rid = 1; i < RL_MSI_MESSAGES; i++, rid++) {
1338 			sc->rl_irq[i] = bus_alloc_resource_any(dev,
1339 			    SYS_RES_IRQ, &rid, RF_ACTIVE);
1340 			if (sc->rl_irq[i] == NULL) {
1341 				device_printf(dev,
1342 				    "couldn't allocate IRQ resources for "
1343 				    "message %d\n", rid);
1344 				error = ENXIO;
1345 				goto fail;
1346 			}
1347 		}
1348 	}
1349 
1350 	if ((sc->rl_flags & RL_FLAG_MSI) == 0) {
1351 		CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1352 		cfg = CSR_READ_1(sc, RL_CFG2);
1353 		if ((cfg & RL_CFG2_MSI) != 0) {
1354 			device_printf(dev, "turning off MSI enable bit.\n");
1355 			cfg &= ~RL_CFG2_MSI;
1356 			CSR_WRITE_1(sc, RL_CFG2, cfg);
1357 		}
1358 		CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1359 	}
1360 
1361 	/* Disable ASPM L0S/L1. */
1362 	if (sc->rl_expcap != 0) {
1363 		cap = pci_read_config(dev, sc->rl_expcap +
1364 		    PCIER_LINK_CAP, 2);
1365 		if ((cap & PCIEM_LINK_CAP_ASPM) != 0) {
1366 			ctl = pci_read_config(dev, sc->rl_expcap +
1367 			    PCIER_LINK_CTL, 2);
1368 			if ((ctl & PCIEM_LINK_CTL_ASPMC) != 0) {
1369 				ctl &= ~PCIEM_LINK_CTL_ASPMC;
1370 				pci_write_config(dev, sc->rl_expcap +
1371 				    PCIER_LINK_CTL, ctl, 2);
1372 				device_printf(dev, "ASPM disabled\n");
1373 			}
1374 		} else
1375 			device_printf(dev, "no ASPM capability\n");
1376 	}
1377 
1378 	hw_rev = re_hwrevs;
1379 	hwrev = CSR_READ_4(sc, RL_TXCFG);
1380 	switch (hwrev & 0x70000000) {
1381 	case 0x00000000:
1382 	case 0x10000000:
1383 		device_printf(dev, "Chip rev. 0x%08x\n", hwrev & 0xfc800000);
1384 		hwrev &= (RL_TXCFG_HWREV | 0x80000000);
1385 		break;
1386 	default:
1387 		device_printf(dev, "Chip rev. 0x%08x\n", hwrev & 0x7c800000);
1388 		sc->rl_macrev = hwrev & 0x00700000;
1389 		hwrev &= RL_TXCFG_HWREV;
1390 		break;
1391 	}
1392 	device_printf(dev, "MAC rev. 0x%08x\n", sc->rl_macrev);
1393 	while (hw_rev->rl_desc != NULL) {
1394 		if (hw_rev->rl_rev == hwrev) {
1395 			sc->rl_type = hw_rev->rl_type;
1396 			sc->rl_hwrev = hw_rev;
1397 			break;
1398 		}
1399 		hw_rev++;
1400 	}
1401 	if (hw_rev->rl_desc == NULL) {
1402 		device_printf(dev, "Unknown H/W revision: 0x%08x\n", hwrev);
1403 		error = ENXIO;
1404 		goto fail;
1405 	}
1406 
1407 	switch (hw_rev->rl_rev) {
1408 	case RL_HWREV_8139CPLUS:
1409 		sc->rl_flags |= RL_FLAG_FASTETHER | RL_FLAG_AUTOPAD;
1410 		break;
1411 	case RL_HWREV_8100E:
1412 	case RL_HWREV_8101E:
1413 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_FASTETHER;
1414 		break;
1415 	case RL_HWREV_8102E:
1416 	case RL_HWREV_8102EL:
1417 	case RL_HWREV_8102EL_SPIN1:
1418 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR | RL_FLAG_DESCV2 |
1419 		    RL_FLAG_MACSTAT | RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP |
1420 		    RL_FLAG_AUTOPAD;
1421 		break;
1422 	case RL_HWREV_8103E:
1423 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR | RL_FLAG_DESCV2 |
1424 		    RL_FLAG_MACSTAT | RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP |
1425 		    RL_FLAG_AUTOPAD | RL_FLAG_MACSLEEP;
1426 		break;
1427 	case RL_HWREV_8401E:
1428 	case RL_HWREV_8105E:
1429 	case RL_HWREV_8105E_SPIN1:
1430 	case RL_HWREV_8106E:
1431 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
1432 		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
1433 		    RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD;
1434 		break;
1435 	case RL_HWREV_8402:
1436 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
1437 		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
1438 		    RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD |
1439 		    RL_FLAG_CMDSTOP_WAIT_TXQ;
1440 		break;
1441 	case RL_HWREV_8168B_SPIN1:
1442 	case RL_HWREV_8168B_SPIN2:
1443 		sc->rl_flags |= RL_FLAG_WOLRXENB;
1444 		/* FALLTHROUGH */
1445 	case RL_HWREV_8168B_SPIN3:
1446 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_MACSTAT;
1447 		break;
1448 	case RL_HWREV_8168C_SPIN2:
1449 		sc->rl_flags |= RL_FLAG_MACSLEEP;
1450 		/* FALLTHROUGH */
1451 	case RL_HWREV_8168C:
1452 		if (sc->rl_macrev == 0x00200000)
1453 			sc->rl_flags |= RL_FLAG_MACSLEEP;
1454 		/* FALLTHROUGH */
1455 	case RL_HWREV_8168CP:
1456 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
1457 		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP |
1458 		    RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 | RL_FLAG_WOL_MANLINK;
1459 		break;
1460 	case RL_HWREV_8168D:
1461 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
1462 		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
1463 		    RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 |
1464 		    RL_FLAG_WOL_MANLINK;
1465 		break;
1466 	case RL_HWREV_8168DP:
1467 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
1468 		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_AUTOPAD |
1469 		    RL_FLAG_JUMBOV2 | RL_FLAG_WAIT_TXPOLL | RL_FLAG_WOL_MANLINK;
1470 		break;
1471 	case RL_HWREV_8168E:
1472 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
1473 		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
1474 		    RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 |
1475 		    RL_FLAG_WOL_MANLINK;
1476 		break;
1477 	case RL_HWREV_8168E_VL:
1478 	case RL_HWREV_8168F:
1479 		sc->rl_flags |= RL_FLAG_EARLYOFF;
1480 		/* FALLTHROUGH */
1481 	case RL_HWREV_8411:
1482 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
1483 		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP |
1484 		    RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 |
1485 		    RL_FLAG_CMDSTOP_WAIT_TXQ | RL_FLAG_WOL_MANLINK;
1486 		break;
1487 	case RL_HWREV_8168EP:
1488 	case RL_HWREV_8168G:
1489 	case RL_HWREV_8168H:
1490 	case RL_HWREV_8411B:
1491 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
1492 		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP |
1493 		    RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 |
1494 		    RL_FLAG_CMDSTOP_WAIT_TXQ | RL_FLAG_WOL_MANLINK |
1495 		    RL_FLAG_EARLYOFFV2 | RL_FLAG_RXDV_GATED;
1496 		break;
1497 	case RL_HWREV_8168GU:
1498 		if (pci_get_device(dev) == RT_DEVICEID_8101E) {
1499 			/* RTL8106EUS */
1500 			sc->rl_flags |= RL_FLAG_FASTETHER;
1501 		} else
1502 			sc->rl_flags |= RL_FLAG_JUMBOV2 | RL_FLAG_WOL_MANLINK;
1503 
1504 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
1505 		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP |
1506 		    RL_FLAG_AUTOPAD | RL_FLAG_CMDSTOP_WAIT_TXQ |
1507 		    RL_FLAG_EARLYOFFV2 | RL_FLAG_RXDV_GATED;
1508 		break;
1509 	case RL_HWREV_8169_8110SB:
1510 	case RL_HWREV_8169_8110SBL:
1511 	case RL_HWREV_8169_8110SC:
1512 	case RL_HWREV_8169_8110SCE:
1513 		sc->rl_flags |= RL_FLAG_PHYWAKE;
1514 		/* FALLTHROUGH */
1515 	case RL_HWREV_8169:
1516 	case RL_HWREV_8169S:
1517 	case RL_HWREV_8110S:
1518 		sc->rl_flags |= RL_FLAG_MACRESET;
1519 		break;
1520 	default:
1521 		break;
1522 	}
1523 
1524 	if (sc->rl_hwrev->rl_rev == RL_HWREV_8139CPLUS) {
1525 		sc->rl_cfg0 = RL_8139_CFG0;
1526 		sc->rl_cfg1 = RL_8139_CFG1;
1527 		sc->rl_cfg2 = 0;
1528 		sc->rl_cfg3 = RL_8139_CFG3;
1529 		sc->rl_cfg4 = RL_8139_CFG4;
1530 		sc->rl_cfg5 = RL_8139_CFG5;
1531 	} else {
1532 		sc->rl_cfg0 = RL_CFG0;
1533 		sc->rl_cfg1 = RL_CFG1;
1534 		sc->rl_cfg2 = RL_CFG2;
1535 		sc->rl_cfg3 = RL_CFG3;
1536 		sc->rl_cfg4 = RL_CFG4;
1537 		sc->rl_cfg5 = RL_CFG5;
1538 	}
1539 
1540 	/* Reset the adapter. */
1541 	RL_LOCK(sc);
1542 	re_reset(sc);
1543 	RL_UNLOCK(sc);
1544 
1545 	/* Enable PME. */
1546 	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1547 	cfg = CSR_READ_1(sc, sc->rl_cfg1);
1548 	cfg |= RL_CFG1_PME;
1549 	CSR_WRITE_1(sc, sc->rl_cfg1, cfg);
1550 	cfg = CSR_READ_1(sc, sc->rl_cfg5);
1551 	cfg &= RL_CFG5_PME_STS;
1552 	CSR_WRITE_1(sc, sc->rl_cfg5, cfg);
1553 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1554 
1555 	if ((sc->rl_flags & RL_FLAG_PAR) != 0) {
1556 		/*
1557 		 * XXX Should have a better way to extract station
1558 		 * address from EEPROM.
1559 		 */
1560 		for (i = 0; i < ETHER_ADDR_LEN; i++)
1561 			eaddr[i] = CSR_READ_1(sc, RL_IDR0 + i);
1562 	} else {
1563 		sc->rl_eewidth = RL_9356_ADDR_LEN;
1564 		re_read_eeprom(sc, (caddr_t)&re_did, 0, 1);
1565 		if (re_did != 0x8129)
1566 			sc->rl_eewidth = RL_9346_ADDR_LEN;
1567 
1568 		/*
1569 		 * Get station address from the EEPROM.
1570 		 */
1571 		re_read_eeprom(sc, (caddr_t)as, RL_EE_EADDR, 3);
1572 		for (i = 0; i < ETHER_ADDR_LEN / 2; i++)
1573 			as[i] = le16toh(as[i]);
1574 		bcopy(as, eaddr, ETHER_ADDR_LEN);
1575 	}
1576 
1577 	if (sc->rl_type == RL_8169) {
1578 		/* Set RX length mask and number of descriptors. */
1579 		sc->rl_rxlenmask = RL_RDESC_STAT_GFRAGLEN;
1580 		sc->rl_txstart = RL_GTXSTART;
1581 		sc->rl_ldata.rl_tx_desc_cnt = RL_8169_TX_DESC_CNT;
1582 		sc->rl_ldata.rl_rx_desc_cnt = RL_8169_RX_DESC_CNT;
1583 	} else {
1584 		/* Set RX length mask and number of descriptors. */
1585 		sc->rl_rxlenmask = RL_RDESC_STAT_FRAGLEN;
1586 		sc->rl_txstart = RL_TXSTART;
1587 		sc->rl_ldata.rl_tx_desc_cnt = RL_8139_TX_DESC_CNT;
1588 		sc->rl_ldata.rl_rx_desc_cnt = RL_8139_RX_DESC_CNT;
1589 	}
1590 
1591 	error = re_allocmem(dev, sc);
1592 	if (error)
1593 		goto fail;
1594 	re_add_sysctls(sc);
1595 
1596 	ifp = sc->rl_ifp = if_alloc(IFT_ETHER);
1597 	if (ifp == NULL) {
1598 		device_printf(dev, "can not if_alloc()\n");
1599 		error = ENOSPC;
1600 		goto fail;
1601 	}
1602 
1603 	/* Take controller out of deep sleep mode. */
1604 	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
1605 		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
1606 			CSR_WRITE_1(sc, RL_GPIO,
1607 			    CSR_READ_1(sc, RL_GPIO) | 0x01);
1608 		else
1609 			CSR_WRITE_1(sc, RL_GPIO,
1610 			    CSR_READ_1(sc, RL_GPIO) & ~0x01);
1611 	}
1612 
1613 	/* Take PHY out of power down mode. */
1614 	if ((sc->rl_flags & RL_FLAG_PHYWAKE_PM) != 0) {
1615 		CSR_WRITE_1(sc, RL_PMCH, CSR_READ_1(sc, RL_PMCH) | 0x80);
1616 		if (hw_rev->rl_rev == RL_HWREV_8401E)
1617 			CSR_WRITE_1(sc, 0xD1, CSR_READ_1(sc, 0xD1) & ~0x08);
1618 	}
1619 	if ((sc->rl_flags & RL_FLAG_PHYWAKE) != 0) {
1620 		re_gmii_writereg(dev, 1, 0x1f, 0);
1621 		re_gmii_writereg(dev, 1, 0x0e, 0);
1622 	}
1623 
1624 	ifp->if_softc = sc;
1625 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1626 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1627 	ifp->if_ioctl = re_ioctl;
1628 	ifp->if_start = re_start;
1629 	/*
1630 	 * RTL8168/8111C generates wrong IP checksummed frame if the
1631 	 * packet has IP options so disable TX checksum offloading.
1632 	 */
1633 	if (sc->rl_hwrev->rl_rev == RL_HWREV_8168C ||
1634 	    sc->rl_hwrev->rl_rev == RL_HWREV_8168C_SPIN2 ||
1635 	    sc->rl_hwrev->rl_rev == RL_HWREV_8168CP) {
1636 		ifp->if_hwassist = 0;
1637 		ifp->if_capabilities = IFCAP_RXCSUM | IFCAP_TSO4;
1638 	} else {
1639 		ifp->if_hwassist = CSUM_IP | CSUM_TCP | CSUM_UDP;
1640 		ifp->if_capabilities = IFCAP_HWCSUM | IFCAP_TSO4;
1641 	}
1642 	ifp->if_hwassist |= CSUM_TSO;
1643 	ifp->if_capenable = ifp->if_capabilities;
1644 	ifp->if_init = re_init;
1645 	IFQ_SET_MAXLEN(&ifp->if_snd, RL_IFQ_MAXLEN);
1646 	ifp->if_snd.ifq_drv_maxlen = RL_IFQ_MAXLEN;
1647 	IFQ_SET_READY(&ifp->if_snd);
1648 
1649 	TASK_INIT(&sc->rl_inttask, 0, re_int_task, sc);
1650 
1651 #define	RE_PHYAD_INTERNAL	 0
1652 
1653 	/* Do MII setup. */
1654 	phy = RE_PHYAD_INTERNAL;
1655 	if (sc->rl_type == RL_8169)
1656 		phy = 1;
1657 	error = mii_attach(dev, &sc->rl_miibus, ifp, re_ifmedia_upd,
1658 	    re_ifmedia_sts, BMSR_DEFCAPMASK, phy, MII_OFFSET_ANY, MIIF_DOPAUSE);
1659 	if (error != 0) {
1660 		device_printf(dev, "attaching PHYs failed\n");
1661 		goto fail;
1662 	}
1663 
1664 	/*
1665 	 * Call MI attach routine.
1666 	 */
1667 	ether_ifattach(ifp, eaddr);
1668 
1669 	/* VLAN capability setup */
1670 	ifp->if_capabilities |= IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING;
1671 	if (ifp->if_capabilities & IFCAP_HWCSUM)
1672 		ifp->if_capabilities |= IFCAP_VLAN_HWCSUM;
1673 	/* Enable WOL if PM is supported. */
1674 	if (pci_find_cap(sc->rl_dev, PCIY_PMG, &reg) == 0)
1675 		ifp->if_capabilities |= IFCAP_WOL;
1676 	ifp->if_capenable = ifp->if_capabilities;
1677 	ifp->if_capenable &= ~(IFCAP_WOL_UCAST | IFCAP_WOL_MCAST);
1678 	/*
1679 	 * Don't enable TSO by default.  It is known to generate
1680 	 * corrupted TCP segments(bad TCP options) under certain
1681 	 * circumstances.
1682 	 */
1683 	ifp->if_hwassist &= ~CSUM_TSO;
1684 	ifp->if_capenable &= ~(IFCAP_TSO4 | IFCAP_VLAN_HWTSO);
1685 #ifdef DEVICE_POLLING
1686 	ifp->if_capabilities |= IFCAP_POLLING;
1687 #endif
1688 	/*
1689 	 * Tell the upper layer(s) we support long frames.
1690 	 * Must appear after the call to ether_ifattach() because
1691 	 * ether_ifattach() sets ifi_hdrlen to the default value.
1692 	 */
1693 	ifp->if_hdrlen = sizeof(struct ether_vlan_header);
1694 
1695 #ifdef DEV_NETMAP
1696 	re_netmap_attach(sc);
1697 #endif /* DEV_NETMAP */
1698 #ifdef RE_DIAG
1699 	/*
1700 	 * Perform hardware diagnostic on the original RTL8169.
1701 	 * Some 32-bit cards were incorrectly wired and would
1702 	 * malfunction if plugged into a 64-bit slot.
1703 	 */
1704 
1705 	if (hwrev == RL_HWREV_8169) {
1706 		error = re_diag(sc);
1707 		if (error) {
1708 			device_printf(dev,
1709 		    	"attach aborted due to hardware diag failure\n");
1710 			ether_ifdetach(ifp);
1711 			goto fail;
1712 		}
1713 	}
1714 #endif
1715 
1716 #ifdef RE_TX_MODERATION
1717 	intr_filter = 1;
1718 #endif
1719 	/* Hook interrupt last to avoid having to lock softc */
1720 	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) != 0 &&
1721 	    intr_filter == 0) {
1722 		error = bus_setup_intr(dev, sc->rl_irq[0],
1723 		    INTR_TYPE_NET | INTR_MPSAFE, NULL, re_intr_msi, sc,
1724 		    &sc->rl_intrhand[0]);
1725 	} else {
1726 		error = bus_setup_intr(dev, sc->rl_irq[0],
1727 		    INTR_TYPE_NET | INTR_MPSAFE, re_intr, NULL, sc,
1728 		    &sc->rl_intrhand[0]);
1729 	}
1730 	if (error) {
1731 		device_printf(dev, "couldn't set up irq\n");
1732 		ether_ifdetach(ifp);
1733 	}
1734 
1735 fail:
1736 
1737 	if (error)
1738 		re_detach(dev);
1739 
1740 	return (error);
1741 }
1742 
1743 /*
1744  * Shutdown hardware and free up resources. This can be called any
1745  * time after the mutex has been initialized. It is called in both
1746  * the error case in attach and the normal detach case so it needs
1747  * to be careful about only freeing resources that have actually been
1748  * allocated.
1749  */
1750 static int
1751 re_detach(device_t dev)
1752 {
1753 	struct rl_softc		*sc;
1754 	struct ifnet		*ifp;
1755 	int			i, rid;
1756 
1757 	sc = device_get_softc(dev);
1758 	ifp = sc->rl_ifp;
1759 	KASSERT(mtx_initialized(&sc->rl_mtx), ("re mutex not initialized"));
1760 
1761 	/* These should only be active if attach succeeded */
1762 	if (device_is_attached(dev)) {
1763 #ifdef DEVICE_POLLING
1764 		if (ifp->if_capenable & IFCAP_POLLING)
1765 			ether_poll_deregister(ifp);
1766 #endif
1767 		RL_LOCK(sc);
1768 #if 0
1769 		sc->suspended = 1;
1770 #endif
1771 		re_stop(sc);
1772 		RL_UNLOCK(sc);
1773 		callout_drain(&sc->rl_stat_callout);
1774 		taskqueue_drain(taskqueue_fast, &sc->rl_inttask);
1775 		/*
1776 		 * Force off the IFF_UP flag here, in case someone
1777 		 * still had a BPF descriptor attached to this
1778 		 * interface. If they do, ether_ifdetach() will cause
1779 		 * the BPF code to try and clear the promisc mode
1780 		 * flag, which will bubble down to re_ioctl(),
1781 		 * which will try to call re_init() again. This will
1782 		 * turn the NIC back on and restart the MII ticker,
1783 		 * which will panic the system when the kernel tries
1784 		 * to invoke the re_tick() function that isn't there
1785 		 * anymore.
1786 		 */
1787 		ifp->if_flags &= ~IFF_UP;
1788 		ether_ifdetach(ifp);
1789 	}
1790 	if (sc->rl_miibus)
1791 		device_delete_child(dev, sc->rl_miibus);
1792 	bus_generic_detach(dev);
1793 
1794 	/*
1795 	 * The rest is resource deallocation, so we should already be
1796 	 * stopped here.
1797 	 */
1798 
1799 	if (sc->rl_intrhand[0] != NULL) {
1800 		bus_teardown_intr(dev, sc->rl_irq[0], sc->rl_intrhand[0]);
1801 		sc->rl_intrhand[0] = NULL;
1802 	}
1803 	if (ifp != NULL) {
1804 #ifdef DEV_NETMAP
1805 		netmap_detach(ifp);
1806 #endif /* DEV_NETMAP */
1807 		if_free(ifp);
1808 	}
1809 	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) == 0)
1810 		rid = 0;
1811 	else
1812 		rid = 1;
1813 	if (sc->rl_irq[0] != NULL) {
1814 		bus_release_resource(dev, SYS_RES_IRQ, rid, sc->rl_irq[0]);
1815 		sc->rl_irq[0] = NULL;
1816 	}
1817 	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) != 0)
1818 		pci_release_msi(dev);
1819 	if (sc->rl_res_pba) {
1820 		rid = PCIR_BAR(4);
1821 		bus_release_resource(dev, SYS_RES_MEMORY, rid, sc->rl_res_pba);
1822 	}
1823 	if (sc->rl_res)
1824 		bus_release_resource(dev, sc->rl_res_type, sc->rl_res_id,
1825 		    sc->rl_res);
1826 
1827 	/* Unload and free the RX DMA ring memory and map */
1828 
1829 	if (sc->rl_ldata.rl_rx_list_tag) {
1830 		if (sc->rl_ldata.rl_rx_list_addr)
1831 			bus_dmamap_unload(sc->rl_ldata.rl_rx_list_tag,
1832 			    sc->rl_ldata.rl_rx_list_map);
1833 		if (sc->rl_ldata.rl_rx_list)
1834 			bus_dmamem_free(sc->rl_ldata.rl_rx_list_tag,
1835 			    sc->rl_ldata.rl_rx_list,
1836 			    sc->rl_ldata.rl_rx_list_map);
1837 		bus_dma_tag_destroy(sc->rl_ldata.rl_rx_list_tag);
1838 	}
1839 
1840 	/* Unload and free the TX DMA ring memory and map */
1841 
1842 	if (sc->rl_ldata.rl_tx_list_tag) {
1843 		if (sc->rl_ldata.rl_tx_list_addr)
1844 			bus_dmamap_unload(sc->rl_ldata.rl_tx_list_tag,
1845 			    sc->rl_ldata.rl_tx_list_map);
1846 		if (sc->rl_ldata.rl_tx_list)
1847 			bus_dmamem_free(sc->rl_ldata.rl_tx_list_tag,
1848 			    sc->rl_ldata.rl_tx_list,
1849 			    sc->rl_ldata.rl_tx_list_map);
1850 		bus_dma_tag_destroy(sc->rl_ldata.rl_tx_list_tag);
1851 	}
1852 
1853 	/* Destroy all the RX and TX buffer maps */
1854 
1855 	if (sc->rl_ldata.rl_tx_mtag) {
1856 		for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
1857 			if (sc->rl_ldata.rl_tx_desc[i].tx_dmamap)
1858 				bus_dmamap_destroy(sc->rl_ldata.rl_tx_mtag,
1859 				    sc->rl_ldata.rl_tx_desc[i].tx_dmamap);
1860 		}
1861 		bus_dma_tag_destroy(sc->rl_ldata.rl_tx_mtag);
1862 	}
1863 	if (sc->rl_ldata.rl_rx_mtag) {
1864 		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1865 			if (sc->rl_ldata.rl_rx_desc[i].rx_dmamap)
1866 				bus_dmamap_destroy(sc->rl_ldata.rl_rx_mtag,
1867 				    sc->rl_ldata.rl_rx_desc[i].rx_dmamap);
1868 		}
1869 		if (sc->rl_ldata.rl_rx_sparemap)
1870 			bus_dmamap_destroy(sc->rl_ldata.rl_rx_mtag,
1871 			    sc->rl_ldata.rl_rx_sparemap);
1872 		bus_dma_tag_destroy(sc->rl_ldata.rl_rx_mtag);
1873 	}
1874 	if (sc->rl_ldata.rl_jrx_mtag) {
1875 		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1876 			if (sc->rl_ldata.rl_jrx_desc[i].rx_dmamap)
1877 				bus_dmamap_destroy(sc->rl_ldata.rl_jrx_mtag,
1878 				    sc->rl_ldata.rl_jrx_desc[i].rx_dmamap);
1879 		}
1880 		if (sc->rl_ldata.rl_jrx_sparemap)
1881 			bus_dmamap_destroy(sc->rl_ldata.rl_jrx_mtag,
1882 			    sc->rl_ldata.rl_jrx_sparemap);
1883 		bus_dma_tag_destroy(sc->rl_ldata.rl_jrx_mtag);
1884 	}
1885 	/* Unload and free the stats buffer and map */
1886 
1887 	if (sc->rl_ldata.rl_stag) {
1888 		if (sc->rl_ldata.rl_stats_addr)
1889 			bus_dmamap_unload(sc->rl_ldata.rl_stag,
1890 			    sc->rl_ldata.rl_smap);
1891 		if (sc->rl_ldata.rl_stats)
1892 			bus_dmamem_free(sc->rl_ldata.rl_stag,
1893 			    sc->rl_ldata.rl_stats, sc->rl_ldata.rl_smap);
1894 		bus_dma_tag_destroy(sc->rl_ldata.rl_stag);
1895 	}
1896 
1897 	if (sc->rl_parent_tag)
1898 		bus_dma_tag_destroy(sc->rl_parent_tag);
1899 
1900 	mtx_destroy(&sc->rl_mtx);
1901 
1902 	return (0);
1903 }
1904 
1905 static __inline void
1906 re_discard_rxbuf(struct rl_softc *sc, int idx)
1907 {
1908 	struct rl_desc		*desc;
1909 	struct rl_rxdesc	*rxd;
1910 	uint32_t		cmdstat;
1911 
1912 	if (sc->rl_ifp->if_mtu > RL_MTU &&
1913 	    (sc->rl_flags & RL_FLAG_JUMBOV2) != 0)
1914 		rxd = &sc->rl_ldata.rl_jrx_desc[idx];
1915 	else
1916 		rxd = &sc->rl_ldata.rl_rx_desc[idx];
1917 	desc = &sc->rl_ldata.rl_rx_list[idx];
1918 	desc->rl_vlanctl = 0;
1919 	cmdstat = rxd->rx_size;
1920 	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
1921 		cmdstat |= RL_RDESC_CMD_EOR;
1922 	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
1923 }
1924 
1925 static int
1926 re_newbuf(struct rl_softc *sc, int idx)
1927 {
1928 	struct mbuf		*m;
1929 	struct rl_rxdesc	*rxd;
1930 	bus_dma_segment_t	segs[1];
1931 	bus_dmamap_t		map;
1932 	struct rl_desc		*desc;
1933 	uint32_t		cmdstat;
1934 	int			error, nsegs;
1935 
1936 	m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
1937 	if (m == NULL)
1938 		return (ENOBUFS);
1939 
1940 	m->m_len = m->m_pkthdr.len = MCLBYTES;
1941 #ifdef RE_FIXUP_RX
1942 	/*
1943 	 * This is part of an evil trick to deal with non-x86 platforms.
1944 	 * The RealTek chip requires RX buffers to be aligned on 64-bit
1945 	 * boundaries, but that will hose non-x86 machines. To get around
1946 	 * this, we leave some empty space at the start of each buffer
1947 	 * and for non-x86 hosts, we copy the buffer back six bytes
1948 	 * to achieve word alignment. This is slightly more efficient
1949 	 * than allocating a new buffer, copying the contents, and
1950 	 * discarding the old buffer.
1951 	 */
1952 	m_adj(m, RE_ETHER_ALIGN);
1953 #endif
1954 	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_rx_mtag,
1955 	    sc->rl_ldata.rl_rx_sparemap, m, segs, &nsegs, BUS_DMA_NOWAIT);
1956 	if (error != 0) {
1957 		m_freem(m);
1958 		return (ENOBUFS);
1959 	}
1960 	KASSERT(nsegs == 1, ("%s: %d segment returned!", __func__, nsegs));
1961 
1962 	rxd = &sc->rl_ldata.rl_rx_desc[idx];
1963 	if (rxd->rx_m != NULL) {
1964 		bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap,
1965 		    BUS_DMASYNC_POSTREAD);
1966 		bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap);
1967 	}
1968 
1969 	rxd->rx_m = m;
1970 	map = rxd->rx_dmamap;
1971 	rxd->rx_dmamap = sc->rl_ldata.rl_rx_sparemap;
1972 	rxd->rx_size = segs[0].ds_len;
1973 	sc->rl_ldata.rl_rx_sparemap = map;
1974 	bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap,
1975 	    BUS_DMASYNC_PREREAD);
1976 
1977 	desc = &sc->rl_ldata.rl_rx_list[idx];
1978 	desc->rl_vlanctl = 0;
1979 	desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[0].ds_addr));
1980 	desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[0].ds_addr));
1981 	cmdstat = segs[0].ds_len;
1982 	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
1983 		cmdstat |= RL_RDESC_CMD_EOR;
1984 	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
1985 
1986 	return (0);
1987 }
1988 
1989 static int
1990 re_jumbo_newbuf(struct rl_softc *sc, int idx)
1991 {
1992 	struct mbuf		*m;
1993 	struct rl_rxdesc	*rxd;
1994 	bus_dma_segment_t	segs[1];
1995 	bus_dmamap_t		map;
1996 	struct rl_desc		*desc;
1997 	uint32_t		cmdstat;
1998 	int			error, nsegs;
1999 
2000 	m = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, MJUM9BYTES);
2001 	if (m == NULL)
2002 		return (ENOBUFS);
2003 	m->m_len = m->m_pkthdr.len = MJUM9BYTES;
2004 #ifdef RE_FIXUP_RX
2005 	m_adj(m, RE_ETHER_ALIGN);
2006 #endif
2007 	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_jrx_mtag,
2008 	    sc->rl_ldata.rl_jrx_sparemap, m, segs, &nsegs, BUS_DMA_NOWAIT);
2009 	if (error != 0) {
2010 		m_freem(m);
2011 		return (ENOBUFS);
2012 	}
2013 	KASSERT(nsegs == 1, ("%s: %d segment returned!", __func__, nsegs));
2014 
2015 	rxd = &sc->rl_ldata.rl_jrx_desc[idx];
2016 	if (rxd->rx_m != NULL) {
2017 		bus_dmamap_sync(sc->rl_ldata.rl_jrx_mtag, rxd->rx_dmamap,
2018 		    BUS_DMASYNC_POSTREAD);
2019 		bus_dmamap_unload(sc->rl_ldata.rl_jrx_mtag, rxd->rx_dmamap);
2020 	}
2021 
2022 	rxd->rx_m = m;
2023 	map = rxd->rx_dmamap;
2024 	rxd->rx_dmamap = sc->rl_ldata.rl_jrx_sparemap;
2025 	rxd->rx_size = segs[0].ds_len;
2026 	sc->rl_ldata.rl_jrx_sparemap = map;
2027 	bus_dmamap_sync(sc->rl_ldata.rl_jrx_mtag, rxd->rx_dmamap,
2028 	    BUS_DMASYNC_PREREAD);
2029 
2030 	desc = &sc->rl_ldata.rl_rx_list[idx];
2031 	desc->rl_vlanctl = 0;
2032 	desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[0].ds_addr));
2033 	desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[0].ds_addr));
2034 	cmdstat = segs[0].ds_len;
2035 	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
2036 		cmdstat |= RL_RDESC_CMD_EOR;
2037 	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
2038 
2039 	return (0);
2040 }
2041 
2042 #ifdef RE_FIXUP_RX
2043 static __inline void
2044 re_fixup_rx(struct mbuf *m)
2045 {
2046 	int                     i;
2047 	uint16_t                *src, *dst;
2048 
2049 	src = mtod(m, uint16_t *);
2050 	dst = src - (RE_ETHER_ALIGN - ETHER_ALIGN) / sizeof *src;
2051 
2052 	for (i = 0; i < (m->m_len / sizeof(uint16_t) + 1); i++)
2053 		*dst++ = *src++;
2054 
2055 	m->m_data -= RE_ETHER_ALIGN - ETHER_ALIGN;
2056 }
2057 #endif
2058 
2059 static int
2060 re_tx_list_init(struct rl_softc *sc)
2061 {
2062 	struct rl_desc		*desc;
2063 	int			i;
2064 
2065 	RL_LOCK_ASSERT(sc);
2066 
2067 	bzero(sc->rl_ldata.rl_tx_list,
2068 	    sc->rl_ldata.rl_tx_desc_cnt * sizeof(struct rl_desc));
2069 	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++)
2070 		sc->rl_ldata.rl_tx_desc[i].tx_m = NULL;
2071 #ifdef DEV_NETMAP
2072 	re_netmap_tx_init(sc);
2073 #endif /* DEV_NETMAP */
2074 	/* Set EOR. */
2075 	desc = &sc->rl_ldata.rl_tx_list[sc->rl_ldata.rl_tx_desc_cnt - 1];
2076 	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_EOR);
2077 
2078 	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2079 	    sc->rl_ldata.rl_tx_list_map,
2080 	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
2081 
2082 	sc->rl_ldata.rl_tx_prodidx = 0;
2083 	sc->rl_ldata.rl_tx_considx = 0;
2084 	sc->rl_ldata.rl_tx_free = sc->rl_ldata.rl_tx_desc_cnt;
2085 
2086 	return (0);
2087 }
2088 
2089 static int
2090 re_rx_list_init(struct rl_softc *sc)
2091 {
2092 	int			error, i;
2093 
2094 	bzero(sc->rl_ldata.rl_rx_list,
2095 	    sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc));
2096 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
2097 		sc->rl_ldata.rl_rx_desc[i].rx_m = NULL;
2098 		if ((error = re_newbuf(sc, i)) != 0)
2099 			return (error);
2100 	}
2101 #ifdef DEV_NETMAP
2102 	re_netmap_rx_init(sc);
2103 #endif /* DEV_NETMAP */
2104 
2105 	/* Flush the RX descriptors */
2106 
2107 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
2108 	    sc->rl_ldata.rl_rx_list_map,
2109 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2110 
2111 	sc->rl_ldata.rl_rx_prodidx = 0;
2112 	sc->rl_head = sc->rl_tail = NULL;
2113 	sc->rl_int_rx_act = 0;
2114 
2115 	return (0);
2116 }
2117 
2118 static int
2119 re_jrx_list_init(struct rl_softc *sc)
2120 {
2121 	int			error, i;
2122 
2123 	bzero(sc->rl_ldata.rl_rx_list,
2124 	    sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc));
2125 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
2126 		sc->rl_ldata.rl_jrx_desc[i].rx_m = NULL;
2127 		if ((error = re_jumbo_newbuf(sc, i)) != 0)
2128 			return (error);
2129 	}
2130 
2131 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
2132 	    sc->rl_ldata.rl_rx_list_map,
2133 	    BUS_DMASYNC_PREWRITE | BUS_DMASYNC_PREREAD);
2134 
2135 	sc->rl_ldata.rl_rx_prodidx = 0;
2136 	sc->rl_head = sc->rl_tail = NULL;
2137 	sc->rl_int_rx_act = 0;
2138 
2139 	return (0);
2140 }
2141 
2142 /*
2143  * RX handler for C+ and 8169. For the gigE chips, we support
2144  * the reception of jumbo frames that have been fragmented
2145  * across multiple 2K mbuf cluster buffers.
2146  */
2147 static int
2148 re_rxeof(struct rl_softc *sc, int *rx_npktsp)
2149 {
2150 	struct mbuf		*m;
2151 	struct ifnet		*ifp;
2152 	int			i, rxerr, total_len;
2153 	struct rl_desc		*cur_rx;
2154 	u_int32_t		rxstat, rxvlan;
2155 	int			jumbo, maxpkt = 16, rx_npkts = 0;
2156 
2157 	RL_LOCK_ASSERT(sc);
2158 
2159 	ifp = sc->rl_ifp;
2160 #ifdef DEV_NETMAP
2161 	if (netmap_rx_irq(ifp, 0, &rx_npkts))
2162 		return 0;
2163 #endif /* DEV_NETMAP */
2164 	if (ifp->if_mtu > RL_MTU && (sc->rl_flags & RL_FLAG_JUMBOV2) != 0)
2165 		jumbo = 1;
2166 	else
2167 		jumbo = 0;
2168 
2169 	/* Invalidate the descriptor memory */
2170 
2171 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
2172 	    sc->rl_ldata.rl_rx_list_map,
2173 	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2174 
2175 	for (i = sc->rl_ldata.rl_rx_prodidx; maxpkt > 0;
2176 	    i = RL_RX_DESC_NXT(sc, i)) {
2177 		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
2178 			break;
2179 		cur_rx = &sc->rl_ldata.rl_rx_list[i];
2180 		rxstat = le32toh(cur_rx->rl_cmdstat);
2181 		if ((rxstat & RL_RDESC_STAT_OWN) != 0)
2182 			break;
2183 		total_len = rxstat & sc->rl_rxlenmask;
2184 		rxvlan = le32toh(cur_rx->rl_vlanctl);
2185 		if (jumbo != 0)
2186 			m = sc->rl_ldata.rl_jrx_desc[i].rx_m;
2187 		else
2188 			m = sc->rl_ldata.rl_rx_desc[i].rx_m;
2189 
2190 		if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0 &&
2191 		    (rxstat & (RL_RDESC_STAT_SOF | RL_RDESC_STAT_EOF)) !=
2192 		    (RL_RDESC_STAT_SOF | RL_RDESC_STAT_EOF)) {
2193 			/*
2194 			 * RTL8168C or later controllers do not
2195 			 * support multi-fragment packet.
2196 			 */
2197 			re_discard_rxbuf(sc, i);
2198 			continue;
2199 		} else if ((rxstat & RL_RDESC_STAT_EOF) == 0) {
2200 			if (re_newbuf(sc, i) != 0) {
2201 				/*
2202 				 * If this is part of a multi-fragment packet,
2203 				 * discard all the pieces.
2204 				 */
2205 				if (sc->rl_head != NULL) {
2206 					m_freem(sc->rl_head);
2207 					sc->rl_head = sc->rl_tail = NULL;
2208 				}
2209 				re_discard_rxbuf(sc, i);
2210 				continue;
2211 			}
2212 			m->m_len = RE_RX_DESC_BUFLEN;
2213 			if (sc->rl_head == NULL)
2214 				sc->rl_head = sc->rl_tail = m;
2215 			else {
2216 				m->m_flags &= ~M_PKTHDR;
2217 				sc->rl_tail->m_next = m;
2218 				sc->rl_tail = m;
2219 			}
2220 			continue;
2221 		}
2222 
2223 		/*
2224 		 * NOTE: for the 8139C+, the frame length field
2225 		 * is always 12 bits in size, but for the gigE chips,
2226 		 * it is 13 bits (since the max RX frame length is 16K).
2227 		 * Unfortunately, all 32 bits in the status word
2228 		 * were already used, so to make room for the extra
2229 		 * length bit, RealTek took out the 'frame alignment
2230 		 * error' bit and shifted the other status bits
2231 		 * over one slot. The OWN, EOR, FS and LS bits are
2232 		 * still in the same places. We have already extracted
2233 		 * the frame length and checked the OWN bit, so rather
2234 		 * than using an alternate bit mapping, we shift the
2235 		 * status bits one space to the right so we can evaluate
2236 		 * them using the 8169 status as though it was in the
2237 		 * same format as that of the 8139C+.
2238 		 */
2239 		if (sc->rl_type == RL_8169)
2240 			rxstat >>= 1;
2241 
2242 		/*
2243 		 * if total_len > 2^13-1, both _RXERRSUM and _GIANT will be
2244 		 * set, but if CRC is clear, it will still be a valid frame.
2245 		 */
2246 		if ((rxstat & RL_RDESC_STAT_RXERRSUM) != 0) {
2247 			rxerr = 1;
2248 			if ((sc->rl_flags & RL_FLAG_JUMBOV2) == 0 &&
2249 			    total_len > 8191 &&
2250 			    (rxstat & RL_RDESC_STAT_ERRS) == RL_RDESC_STAT_GIANT)
2251 				rxerr = 0;
2252 			if (rxerr != 0) {
2253 				if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
2254 				/*
2255 				 * If this is part of a multi-fragment packet,
2256 				 * discard all the pieces.
2257 				 */
2258 				if (sc->rl_head != NULL) {
2259 					m_freem(sc->rl_head);
2260 					sc->rl_head = sc->rl_tail = NULL;
2261 				}
2262 				re_discard_rxbuf(sc, i);
2263 				continue;
2264 			}
2265 		}
2266 
2267 		/*
2268 		 * If allocating a replacement mbuf fails,
2269 		 * reload the current one.
2270 		 */
2271 		if (jumbo != 0)
2272 			rxerr = re_jumbo_newbuf(sc, i);
2273 		else
2274 			rxerr = re_newbuf(sc, i);
2275 		if (rxerr != 0) {
2276 			if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
2277 			if (sc->rl_head != NULL) {
2278 				m_freem(sc->rl_head);
2279 				sc->rl_head = sc->rl_tail = NULL;
2280 			}
2281 			re_discard_rxbuf(sc, i);
2282 			continue;
2283 		}
2284 
2285 		if (sc->rl_head != NULL) {
2286 			if (jumbo != 0)
2287 				m->m_len = total_len;
2288 			else {
2289 				m->m_len = total_len % RE_RX_DESC_BUFLEN;
2290 				if (m->m_len == 0)
2291 					m->m_len = RE_RX_DESC_BUFLEN;
2292 			}
2293 			/*
2294 			 * Special case: if there's 4 bytes or less
2295 			 * in this buffer, the mbuf can be discarded:
2296 			 * the last 4 bytes is the CRC, which we don't
2297 			 * care about anyway.
2298 			 */
2299 			if (m->m_len <= ETHER_CRC_LEN) {
2300 				sc->rl_tail->m_len -=
2301 				    (ETHER_CRC_LEN - m->m_len);
2302 				m_freem(m);
2303 			} else {
2304 				m->m_len -= ETHER_CRC_LEN;
2305 				m->m_flags &= ~M_PKTHDR;
2306 				sc->rl_tail->m_next = m;
2307 			}
2308 			m = sc->rl_head;
2309 			sc->rl_head = sc->rl_tail = NULL;
2310 			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
2311 		} else
2312 			m->m_pkthdr.len = m->m_len =
2313 			    (total_len - ETHER_CRC_LEN);
2314 
2315 #ifdef RE_FIXUP_RX
2316 		re_fixup_rx(m);
2317 #endif
2318 		if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
2319 		m->m_pkthdr.rcvif = ifp;
2320 
2321 		/* Do RX checksumming if enabled */
2322 
2323 		if (ifp->if_capenable & IFCAP_RXCSUM) {
2324 			if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
2325 				/* Check IP header checksum */
2326 				if (rxstat & RL_RDESC_STAT_PROTOID)
2327 					m->m_pkthdr.csum_flags |=
2328 					    CSUM_IP_CHECKED;
2329 				if (!(rxstat & RL_RDESC_STAT_IPSUMBAD))
2330 					m->m_pkthdr.csum_flags |=
2331 					    CSUM_IP_VALID;
2332 
2333 				/* Check TCP/UDP checksum */
2334 				if ((RL_TCPPKT(rxstat) &&
2335 				    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
2336 				    (RL_UDPPKT(rxstat) &&
2337 				     !(rxstat & RL_RDESC_STAT_UDPSUMBAD))) {
2338 					m->m_pkthdr.csum_flags |=
2339 						CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
2340 					m->m_pkthdr.csum_data = 0xffff;
2341 				}
2342 			} else {
2343 				/*
2344 				 * RTL8168C/RTL816CP/RTL8111C/RTL8111CP
2345 				 */
2346 				if ((rxstat & RL_RDESC_STAT_PROTOID) &&
2347 				    (rxvlan & RL_RDESC_IPV4))
2348 					m->m_pkthdr.csum_flags |=
2349 					    CSUM_IP_CHECKED;
2350 				if (!(rxstat & RL_RDESC_STAT_IPSUMBAD) &&
2351 				    (rxvlan & RL_RDESC_IPV4))
2352 					m->m_pkthdr.csum_flags |=
2353 					    CSUM_IP_VALID;
2354 				if (((rxstat & RL_RDESC_STAT_TCP) &&
2355 				    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
2356 				    ((rxstat & RL_RDESC_STAT_UDP) &&
2357 				    !(rxstat & RL_RDESC_STAT_UDPSUMBAD))) {
2358 					m->m_pkthdr.csum_flags |=
2359 						CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
2360 					m->m_pkthdr.csum_data = 0xffff;
2361 				}
2362 			}
2363 		}
2364 		maxpkt--;
2365 		if (rxvlan & RL_RDESC_VLANCTL_TAG) {
2366 			m->m_pkthdr.ether_vtag =
2367 			    bswap16((rxvlan & RL_RDESC_VLANCTL_DATA));
2368 			m->m_flags |= M_VLANTAG;
2369 		}
2370 		RL_UNLOCK(sc);
2371 		(*ifp->if_input)(ifp, m);
2372 		RL_LOCK(sc);
2373 		rx_npkts++;
2374 	}
2375 
2376 	/* Flush the RX DMA ring */
2377 
2378 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
2379 	    sc->rl_ldata.rl_rx_list_map,
2380 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2381 
2382 	sc->rl_ldata.rl_rx_prodidx = i;
2383 
2384 	if (rx_npktsp != NULL)
2385 		*rx_npktsp = rx_npkts;
2386 	if (maxpkt)
2387 		return (EAGAIN);
2388 
2389 	return (0);
2390 }
2391 
2392 static void
2393 re_txeof(struct rl_softc *sc)
2394 {
2395 	struct ifnet		*ifp;
2396 	struct rl_txdesc	*txd;
2397 	u_int32_t		txstat;
2398 	int			cons;
2399 
2400 	cons = sc->rl_ldata.rl_tx_considx;
2401 	if (cons == sc->rl_ldata.rl_tx_prodidx)
2402 		return;
2403 
2404 	ifp = sc->rl_ifp;
2405 #ifdef DEV_NETMAP
2406 	if (netmap_tx_irq(ifp, 0))
2407 		return;
2408 #endif /* DEV_NETMAP */
2409 	/* Invalidate the TX descriptor list */
2410 	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2411 	    sc->rl_ldata.rl_tx_list_map,
2412 	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2413 
2414 	for (; cons != sc->rl_ldata.rl_tx_prodidx;
2415 	    cons = RL_TX_DESC_NXT(sc, cons)) {
2416 		txstat = le32toh(sc->rl_ldata.rl_tx_list[cons].rl_cmdstat);
2417 		if (txstat & RL_TDESC_STAT_OWN)
2418 			break;
2419 		/*
2420 		 * We only stash mbufs in the last descriptor
2421 		 * in a fragment chain, which also happens to
2422 		 * be the only place where the TX status bits
2423 		 * are valid.
2424 		 */
2425 		if (txstat & RL_TDESC_CMD_EOF) {
2426 			txd = &sc->rl_ldata.rl_tx_desc[cons];
2427 			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
2428 			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
2429 			bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag,
2430 			    txd->tx_dmamap);
2431 			KASSERT(txd->tx_m != NULL,
2432 			    ("%s: freeing NULL mbufs!", __func__));
2433 			m_freem(txd->tx_m);
2434 			txd->tx_m = NULL;
2435 			if (txstat & (RL_TDESC_STAT_EXCESSCOL|
2436 			    RL_TDESC_STAT_COLCNT))
2437 				if_inc_counter(ifp, IFCOUNTER_COLLISIONS, 1);
2438 			if (txstat & RL_TDESC_STAT_TXERRSUM)
2439 				if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
2440 			else
2441 				if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
2442 		}
2443 		sc->rl_ldata.rl_tx_free++;
2444 		ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
2445 	}
2446 	sc->rl_ldata.rl_tx_considx = cons;
2447 
2448 	/* No changes made to the TX ring, so no flush needed */
2449 
2450 	if (sc->rl_ldata.rl_tx_free != sc->rl_ldata.rl_tx_desc_cnt) {
2451 #ifdef RE_TX_MODERATION
2452 		/*
2453 		 * If not all descriptors have been reaped yet, reload
2454 		 * the timer so that we will eventually get another
2455 		 * interrupt that will cause us to re-enter this routine.
2456 		 * This is done in case the transmitter has gone idle.
2457 		 */
2458 		CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2459 #endif
2460 	} else
2461 		sc->rl_watchdog_timer = 0;
2462 }
2463 
2464 static void
2465 re_tick(void *xsc)
2466 {
2467 	struct rl_softc		*sc;
2468 	struct mii_data		*mii;
2469 
2470 	sc = xsc;
2471 
2472 	RL_LOCK_ASSERT(sc);
2473 
2474 	mii = device_get_softc(sc->rl_miibus);
2475 	mii_tick(mii);
2476 	if ((sc->rl_flags & RL_FLAG_LINK) == 0)
2477 		re_miibus_statchg(sc->rl_dev);
2478 	/*
2479 	 * Reclaim transmitted frames here. Technically it is not
2480 	 * necessary to do here but it ensures periodic reclamation
2481 	 * regardless of Tx completion interrupt which seems to be
2482 	 * lost on PCIe based controllers under certain situations.
2483 	 */
2484 	re_txeof(sc);
2485 	re_watchdog(sc);
2486 	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
2487 }
2488 
2489 #ifdef DEVICE_POLLING
2490 static int
2491 re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
2492 {
2493 	struct rl_softc *sc = ifp->if_softc;
2494 	int rx_npkts = 0;
2495 
2496 	RL_LOCK(sc);
2497 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
2498 		rx_npkts = re_poll_locked(ifp, cmd, count);
2499 	RL_UNLOCK(sc);
2500 	return (rx_npkts);
2501 }
2502 
2503 static int
2504 re_poll_locked(struct ifnet *ifp, enum poll_cmd cmd, int count)
2505 {
2506 	struct rl_softc *sc = ifp->if_softc;
2507 	int rx_npkts;
2508 
2509 	RL_LOCK_ASSERT(sc);
2510 
2511 	sc->rxcycles = count;
2512 	re_rxeof(sc, &rx_npkts);
2513 	re_txeof(sc);
2514 
2515 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2516 		re_start_locked(ifp);
2517 
2518 	if (cmd == POLL_AND_CHECK_STATUS) { /* also check status register */
2519 		u_int16_t       status;
2520 
2521 		status = CSR_READ_2(sc, RL_ISR);
2522 		if (status == 0xffff)
2523 			return (rx_npkts);
2524 		if (status)
2525 			CSR_WRITE_2(sc, RL_ISR, status);
2526 		if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
2527 		    (sc->rl_flags & RL_FLAG_PCIE))
2528 			CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2529 
2530 		/*
2531 		 * XXX check behaviour on receiver stalls.
2532 		 */
2533 
2534 		if (status & RL_ISR_SYSTEM_ERR) {
2535 			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2536 			re_init_locked(sc);
2537 		}
2538 	}
2539 	return (rx_npkts);
2540 }
2541 #endif /* DEVICE_POLLING */
2542 
2543 static int
2544 re_intr(void *arg)
2545 {
2546 	struct rl_softc		*sc;
2547 	uint16_t		status;
2548 
2549 	sc = arg;
2550 
2551 	status = CSR_READ_2(sc, RL_ISR);
2552 	if (status == 0xFFFF || (status & RL_INTRS_CPLUS) == 0)
2553                 return (FILTER_STRAY);
2554 	CSR_WRITE_2(sc, RL_IMR, 0);
2555 
2556 	taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_inttask);
2557 
2558 	return (FILTER_HANDLED);
2559 }
2560 
2561 static void
2562 re_int_task(void *arg, int npending)
2563 {
2564 	struct rl_softc		*sc;
2565 	struct ifnet		*ifp;
2566 	u_int16_t		status;
2567 	int			rval = 0;
2568 
2569 	sc = arg;
2570 	ifp = sc->rl_ifp;
2571 
2572 	RL_LOCK(sc);
2573 
2574 	status = CSR_READ_2(sc, RL_ISR);
2575         CSR_WRITE_2(sc, RL_ISR, status);
2576 
2577 	if (sc->suspended ||
2578 	    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2579 		RL_UNLOCK(sc);
2580 		return;
2581 	}
2582 
2583 #ifdef DEVICE_POLLING
2584 	if  (ifp->if_capenable & IFCAP_POLLING) {
2585 		RL_UNLOCK(sc);
2586 		return;
2587 	}
2588 #endif
2589 
2590 	if (status & (RL_ISR_RX_OK|RL_ISR_RX_ERR|RL_ISR_FIFO_OFLOW))
2591 		rval = re_rxeof(sc, NULL);
2592 
2593 	/*
2594 	 * Some chips will ignore a second TX request issued
2595 	 * while an existing transmission is in progress. If
2596 	 * the transmitter goes idle but there are still
2597 	 * packets waiting to be sent, we need to restart the
2598 	 * channel here to flush them out. This only seems to
2599 	 * be required with the PCIe devices.
2600 	 */
2601 	if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
2602 	    (sc->rl_flags & RL_FLAG_PCIE))
2603 		CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2604 	if (status & (
2605 #ifdef RE_TX_MODERATION
2606 	    RL_ISR_TIMEOUT_EXPIRED|
2607 #else
2608 	    RL_ISR_TX_OK|
2609 #endif
2610 	    RL_ISR_TX_ERR|RL_ISR_TX_DESC_UNAVAIL))
2611 		re_txeof(sc);
2612 
2613 	if (status & RL_ISR_SYSTEM_ERR) {
2614 		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2615 		re_init_locked(sc);
2616 	}
2617 
2618 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2619 		re_start_locked(ifp);
2620 
2621 	RL_UNLOCK(sc);
2622 
2623         if ((CSR_READ_2(sc, RL_ISR) & RL_INTRS_CPLUS) || rval) {
2624 		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_inttask);
2625 		return;
2626 	}
2627 
2628 	CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2629 }
2630 
2631 static void
2632 re_intr_msi(void *xsc)
2633 {
2634 	struct rl_softc		*sc;
2635 	struct ifnet		*ifp;
2636 	uint16_t		intrs, status;
2637 
2638 	sc = xsc;
2639 	RL_LOCK(sc);
2640 
2641 	ifp = sc->rl_ifp;
2642 #ifdef DEVICE_POLLING
2643 	if (ifp->if_capenable & IFCAP_POLLING) {
2644 		RL_UNLOCK(sc);
2645 		return;
2646 	}
2647 #endif
2648 	/* Disable interrupts. */
2649 	CSR_WRITE_2(sc, RL_IMR, 0);
2650 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2651 		RL_UNLOCK(sc);
2652 		return;
2653 	}
2654 
2655 	intrs = RL_INTRS_CPLUS;
2656 	status = CSR_READ_2(sc, RL_ISR);
2657         CSR_WRITE_2(sc, RL_ISR, status);
2658 	if (sc->rl_int_rx_act > 0) {
2659 		intrs &= ~(RL_ISR_RX_OK | RL_ISR_RX_ERR | RL_ISR_FIFO_OFLOW |
2660 		    RL_ISR_RX_OVERRUN);
2661 		status &= ~(RL_ISR_RX_OK | RL_ISR_RX_ERR | RL_ISR_FIFO_OFLOW |
2662 		    RL_ISR_RX_OVERRUN);
2663 	}
2664 
2665 	if (status & (RL_ISR_TIMEOUT_EXPIRED | RL_ISR_RX_OK | RL_ISR_RX_ERR |
2666 	    RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN)) {
2667 		re_rxeof(sc, NULL);
2668 		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
2669 			if (sc->rl_int_rx_mod != 0 &&
2670 			    (status & (RL_ISR_RX_OK | RL_ISR_RX_ERR |
2671 			    RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN)) != 0) {
2672 				/* Rearm one-shot timer. */
2673 				CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2674 				intrs &= ~(RL_ISR_RX_OK | RL_ISR_RX_ERR |
2675 				    RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN);
2676 				sc->rl_int_rx_act = 1;
2677 			} else {
2678 				intrs |= RL_ISR_RX_OK | RL_ISR_RX_ERR |
2679 				    RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN;
2680 				sc->rl_int_rx_act = 0;
2681 			}
2682 		}
2683 	}
2684 
2685 	/*
2686 	 * Some chips will ignore a second TX request issued
2687 	 * while an existing transmission is in progress. If
2688 	 * the transmitter goes idle but there are still
2689 	 * packets waiting to be sent, we need to restart the
2690 	 * channel here to flush them out. This only seems to
2691 	 * be required with the PCIe devices.
2692 	 */
2693 	if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
2694 	    (sc->rl_flags & RL_FLAG_PCIE))
2695 		CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2696 	if (status & (RL_ISR_TX_OK | RL_ISR_TX_ERR | RL_ISR_TX_DESC_UNAVAIL))
2697 		re_txeof(sc);
2698 
2699 	if (status & RL_ISR_SYSTEM_ERR) {
2700 		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2701 		re_init_locked(sc);
2702 	}
2703 
2704 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
2705 		if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2706 			re_start_locked(ifp);
2707 		CSR_WRITE_2(sc, RL_IMR, intrs);
2708 	}
2709 	RL_UNLOCK(sc);
2710 }
2711 
2712 static int
2713 re_encap(struct rl_softc *sc, struct mbuf **m_head)
2714 {
2715 	struct rl_txdesc	*txd, *txd_last;
2716 	bus_dma_segment_t	segs[RL_NTXSEGS];
2717 	bus_dmamap_t		map;
2718 	struct mbuf		*m_new;
2719 	struct rl_desc		*desc;
2720 	int			nsegs, prod;
2721 	int			i, error, ei, si;
2722 	int			padlen;
2723 	uint32_t		cmdstat, csum_flags, vlanctl;
2724 
2725 	RL_LOCK_ASSERT(sc);
2726 	M_ASSERTPKTHDR((*m_head));
2727 
2728 	/*
2729 	 * With some of the RealTek chips, using the checksum offload
2730 	 * support in conjunction with the autopadding feature results
2731 	 * in the transmission of corrupt frames. For example, if we
2732 	 * need to send a really small IP fragment that's less than 60
2733 	 * bytes in size, and IP header checksumming is enabled, the
2734 	 * resulting ethernet frame that appears on the wire will
2735 	 * have garbled payload. To work around this, if TX IP checksum
2736 	 * offload is enabled, we always manually pad short frames out
2737 	 * to the minimum ethernet frame size.
2738 	 */
2739 	if ((sc->rl_flags & RL_FLAG_AUTOPAD) == 0 &&
2740 	    (*m_head)->m_pkthdr.len < RL_IP4CSUMTX_PADLEN &&
2741 	    ((*m_head)->m_pkthdr.csum_flags & CSUM_IP) != 0) {
2742 		padlen = RL_MIN_FRAMELEN - (*m_head)->m_pkthdr.len;
2743 		if (M_WRITABLE(*m_head) == 0) {
2744 			/* Get a writable copy. */
2745 			m_new = m_dup(*m_head, M_NOWAIT);
2746 			m_freem(*m_head);
2747 			if (m_new == NULL) {
2748 				*m_head = NULL;
2749 				return (ENOBUFS);
2750 			}
2751 			*m_head = m_new;
2752 		}
2753 		if ((*m_head)->m_next != NULL ||
2754 		    M_TRAILINGSPACE(*m_head) < padlen) {
2755 			m_new = m_defrag(*m_head, M_NOWAIT);
2756 			if (m_new == NULL) {
2757 				m_freem(*m_head);
2758 				*m_head = NULL;
2759 				return (ENOBUFS);
2760 			}
2761 		} else
2762 			m_new = *m_head;
2763 
2764 		/*
2765 		 * Manually pad short frames, and zero the pad space
2766 		 * to avoid leaking data.
2767 		 */
2768 		bzero(mtod(m_new, char *) + m_new->m_pkthdr.len, padlen);
2769 		m_new->m_pkthdr.len += padlen;
2770 		m_new->m_len = m_new->m_pkthdr.len;
2771 		*m_head = m_new;
2772 	}
2773 
2774 	prod = sc->rl_ldata.rl_tx_prodidx;
2775 	txd = &sc->rl_ldata.rl_tx_desc[prod];
2776 	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap,
2777 	    *m_head, segs, &nsegs, BUS_DMA_NOWAIT);
2778 	if (error == EFBIG) {
2779 		m_new = m_collapse(*m_head, M_NOWAIT, RL_NTXSEGS);
2780 		if (m_new == NULL) {
2781 			m_freem(*m_head);
2782 			*m_head = NULL;
2783 			return (ENOBUFS);
2784 		}
2785 		*m_head = m_new;
2786 		error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_tx_mtag,
2787 		    txd->tx_dmamap, *m_head, segs, &nsegs, BUS_DMA_NOWAIT);
2788 		if (error != 0) {
2789 			m_freem(*m_head);
2790 			*m_head = NULL;
2791 			return (error);
2792 		}
2793 	} else if (error != 0)
2794 		return (error);
2795 	if (nsegs == 0) {
2796 		m_freem(*m_head);
2797 		*m_head = NULL;
2798 		return (EIO);
2799 	}
2800 
2801 	/* Check for number of available descriptors. */
2802 	if (sc->rl_ldata.rl_tx_free - nsegs <= 1) {
2803 		bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap);
2804 		return (ENOBUFS);
2805 	}
2806 
2807 	bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap,
2808 	    BUS_DMASYNC_PREWRITE);
2809 
2810 	/*
2811 	 * Set up checksum offload. Note: checksum offload bits must
2812 	 * appear in all descriptors of a multi-descriptor transmit
2813 	 * attempt. This is according to testing done with an 8169
2814 	 * chip. This is a requirement.
2815 	 */
2816 	vlanctl = 0;
2817 	csum_flags = 0;
2818 	if (((*m_head)->m_pkthdr.csum_flags & CSUM_TSO) != 0) {
2819 		if ((sc->rl_flags & RL_FLAG_DESCV2) != 0) {
2820 			csum_flags |= RL_TDESC_CMD_LGSEND;
2821 			vlanctl |= ((uint32_t)(*m_head)->m_pkthdr.tso_segsz <<
2822 			    RL_TDESC_CMD_MSSVALV2_SHIFT);
2823 		} else {
2824 			csum_flags |= RL_TDESC_CMD_LGSEND |
2825 			    ((uint32_t)(*m_head)->m_pkthdr.tso_segsz <<
2826 			    RL_TDESC_CMD_MSSVAL_SHIFT);
2827 		}
2828 	} else {
2829 		/*
2830 		 * Unconditionally enable IP checksum if TCP or UDP
2831 		 * checksum is required. Otherwise, TCP/UDP checksum
2832 		 * doesn't make effects.
2833 		 */
2834 		if (((*m_head)->m_pkthdr.csum_flags & RE_CSUM_FEATURES) != 0) {
2835 			if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
2836 				csum_flags |= RL_TDESC_CMD_IPCSUM;
2837 				if (((*m_head)->m_pkthdr.csum_flags &
2838 				    CSUM_TCP) != 0)
2839 					csum_flags |= RL_TDESC_CMD_TCPCSUM;
2840 				if (((*m_head)->m_pkthdr.csum_flags &
2841 				    CSUM_UDP) != 0)
2842 					csum_flags |= RL_TDESC_CMD_UDPCSUM;
2843 			} else {
2844 				vlanctl |= RL_TDESC_CMD_IPCSUMV2;
2845 				if (((*m_head)->m_pkthdr.csum_flags &
2846 				    CSUM_TCP) != 0)
2847 					vlanctl |= RL_TDESC_CMD_TCPCSUMV2;
2848 				if (((*m_head)->m_pkthdr.csum_flags &
2849 				    CSUM_UDP) != 0)
2850 					vlanctl |= RL_TDESC_CMD_UDPCSUMV2;
2851 			}
2852 		}
2853 	}
2854 
2855 	/*
2856 	 * Set up hardware VLAN tagging. Note: vlan tag info must
2857 	 * appear in all descriptors of a multi-descriptor
2858 	 * transmission attempt.
2859 	 */
2860 	if ((*m_head)->m_flags & M_VLANTAG)
2861 		vlanctl |= bswap16((*m_head)->m_pkthdr.ether_vtag) |
2862 		    RL_TDESC_VLANCTL_TAG;
2863 
2864 	si = prod;
2865 	for (i = 0; i < nsegs; i++, prod = RL_TX_DESC_NXT(sc, prod)) {
2866 		desc = &sc->rl_ldata.rl_tx_list[prod];
2867 		desc->rl_vlanctl = htole32(vlanctl);
2868 		desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[i].ds_addr));
2869 		desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[i].ds_addr));
2870 		cmdstat = segs[i].ds_len;
2871 		if (i != 0)
2872 			cmdstat |= RL_TDESC_CMD_OWN;
2873 		if (prod == sc->rl_ldata.rl_tx_desc_cnt - 1)
2874 			cmdstat |= RL_TDESC_CMD_EOR;
2875 		desc->rl_cmdstat = htole32(cmdstat | csum_flags);
2876 		sc->rl_ldata.rl_tx_free--;
2877 	}
2878 	/* Update producer index. */
2879 	sc->rl_ldata.rl_tx_prodidx = prod;
2880 
2881 	/* Set EOF on the last descriptor. */
2882 	ei = RL_TX_DESC_PRV(sc, prod);
2883 	desc = &sc->rl_ldata.rl_tx_list[ei];
2884 	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_EOF);
2885 
2886 	desc = &sc->rl_ldata.rl_tx_list[si];
2887 	/* Set SOF and transfer ownership of packet to the chip. */
2888 	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_OWN | RL_TDESC_CMD_SOF);
2889 
2890 	/*
2891 	 * Insure that the map for this transmission
2892 	 * is placed at the array index of the last descriptor
2893 	 * in this chain.  (Swap last and first dmamaps.)
2894 	 */
2895 	txd_last = &sc->rl_ldata.rl_tx_desc[ei];
2896 	map = txd->tx_dmamap;
2897 	txd->tx_dmamap = txd_last->tx_dmamap;
2898 	txd_last->tx_dmamap = map;
2899 	txd_last->tx_m = *m_head;
2900 
2901 	return (0);
2902 }
2903 
2904 static void
2905 re_start(struct ifnet *ifp)
2906 {
2907 	struct rl_softc		*sc;
2908 
2909 	sc = ifp->if_softc;
2910 	RL_LOCK(sc);
2911 	re_start_locked(ifp);
2912 	RL_UNLOCK(sc);
2913 }
2914 
2915 /*
2916  * Main transmit routine for C+ and gigE NICs.
2917  */
2918 static void
2919 re_start_locked(struct ifnet *ifp)
2920 {
2921 	struct rl_softc		*sc;
2922 	struct mbuf		*m_head;
2923 	int			queued;
2924 
2925 	sc = ifp->if_softc;
2926 
2927 #ifdef DEV_NETMAP
2928 	/* XXX is this necessary ? */
2929 	if (ifp->if_capenable & IFCAP_NETMAP) {
2930 		struct netmap_kring *kring = &NA(ifp)->tx_rings[0];
2931 		if (sc->rl_ldata.rl_tx_prodidx != kring->nr_hwcur) {
2932 			/* kick the tx unit */
2933 			CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2934 #ifdef RE_TX_MODERATION
2935 			CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2936 #endif
2937 			sc->rl_watchdog_timer = 5;
2938 		}
2939 		return;
2940 	}
2941 #endif /* DEV_NETMAP */
2942 	if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
2943 	    IFF_DRV_RUNNING || (sc->rl_flags & RL_FLAG_LINK) == 0)
2944 		return;
2945 
2946 	for (queued = 0; !IFQ_DRV_IS_EMPTY(&ifp->if_snd) &&
2947 	    sc->rl_ldata.rl_tx_free > 1;) {
2948 		IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
2949 		if (m_head == NULL)
2950 			break;
2951 
2952 		if (re_encap(sc, &m_head) != 0) {
2953 			if (m_head == NULL)
2954 				break;
2955 			IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
2956 			ifp->if_drv_flags |= IFF_DRV_OACTIVE;
2957 			break;
2958 		}
2959 
2960 		/*
2961 		 * If there's a BPF listener, bounce a copy of this frame
2962 		 * to him.
2963 		 */
2964 		ETHER_BPF_MTAP(ifp, m_head);
2965 
2966 		queued++;
2967 	}
2968 
2969 	if (queued == 0) {
2970 #ifdef RE_TX_MODERATION
2971 		if (sc->rl_ldata.rl_tx_free != sc->rl_ldata.rl_tx_desc_cnt)
2972 			CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2973 #endif
2974 		return;
2975 	}
2976 
2977 	/* Flush the TX descriptors */
2978 
2979 	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2980 	    sc->rl_ldata.rl_tx_list_map,
2981 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2982 
2983 	CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2984 
2985 #ifdef RE_TX_MODERATION
2986 	/*
2987 	 * Use the countdown timer for interrupt moderation.
2988 	 * 'TX done' interrupts are disabled. Instead, we reset the
2989 	 * countdown timer, which will begin counting until it hits
2990 	 * the value in the TIMERINT register, and then trigger an
2991 	 * interrupt. Each time we write to the TIMERCNT register,
2992 	 * the timer count is reset to 0.
2993 	 */
2994 	CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2995 #endif
2996 
2997 	/*
2998 	 * Set a timeout in case the chip goes out to lunch.
2999 	 */
3000 	sc->rl_watchdog_timer = 5;
3001 }
3002 
3003 static void
3004 re_set_jumbo(struct rl_softc *sc, int jumbo)
3005 {
3006 
3007 	if (sc->rl_hwrev->rl_rev == RL_HWREV_8168E_VL) {
3008 		pci_set_max_read_req(sc->rl_dev, 4096);
3009 		return;
3010 	}
3011 
3012 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG);
3013 	if (jumbo != 0) {
3014 		CSR_WRITE_1(sc, sc->rl_cfg3, CSR_READ_1(sc, sc->rl_cfg3) |
3015 		    RL_CFG3_JUMBO_EN0);
3016 		switch (sc->rl_hwrev->rl_rev) {
3017 		case RL_HWREV_8168DP:
3018 			break;
3019 		case RL_HWREV_8168E:
3020 			CSR_WRITE_1(sc, sc->rl_cfg4,
3021 			    CSR_READ_1(sc, sc->rl_cfg4) | 0x01);
3022 			break;
3023 		default:
3024 			CSR_WRITE_1(sc, sc->rl_cfg4,
3025 			    CSR_READ_1(sc, sc->rl_cfg4) | RL_CFG4_JUMBO_EN1);
3026 		}
3027 	} else {
3028 		CSR_WRITE_1(sc, sc->rl_cfg3, CSR_READ_1(sc, sc->rl_cfg3) &
3029 		    ~RL_CFG3_JUMBO_EN0);
3030 		switch (sc->rl_hwrev->rl_rev) {
3031 		case RL_HWREV_8168DP:
3032 			break;
3033 		case RL_HWREV_8168E:
3034 			CSR_WRITE_1(sc, sc->rl_cfg4,
3035 			    CSR_READ_1(sc, sc->rl_cfg4) & ~0x01);
3036 			break;
3037 		default:
3038 			CSR_WRITE_1(sc, sc->rl_cfg4,
3039 			    CSR_READ_1(sc, sc->rl_cfg4) & ~RL_CFG4_JUMBO_EN1);
3040 		}
3041 	}
3042 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3043 
3044 	switch (sc->rl_hwrev->rl_rev) {
3045 	case RL_HWREV_8168DP:
3046 		pci_set_max_read_req(sc->rl_dev, 4096);
3047 		break;
3048 	default:
3049 		if (jumbo != 0)
3050 			pci_set_max_read_req(sc->rl_dev, 512);
3051 		else
3052 			pci_set_max_read_req(sc->rl_dev, 4096);
3053 	}
3054 }
3055 
3056 static void
3057 re_init(void *xsc)
3058 {
3059 	struct rl_softc		*sc = xsc;
3060 
3061 	RL_LOCK(sc);
3062 	re_init_locked(sc);
3063 	RL_UNLOCK(sc);
3064 }
3065 
3066 static void
3067 re_init_locked(struct rl_softc *sc)
3068 {
3069 	struct ifnet		*ifp = sc->rl_ifp;
3070 	struct mii_data		*mii;
3071 	uint32_t		reg;
3072 	uint16_t		cfg;
3073 	union {
3074 		uint32_t align_dummy;
3075 		u_char eaddr[ETHER_ADDR_LEN];
3076         } eaddr;
3077 
3078 	RL_LOCK_ASSERT(sc);
3079 
3080 	mii = device_get_softc(sc->rl_miibus);
3081 
3082 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
3083 		return;
3084 
3085 	/*
3086 	 * Cancel pending I/O and free all RX/TX buffers.
3087 	 */
3088 	re_stop(sc);
3089 
3090 	/* Put controller into known state. */
3091 	re_reset(sc);
3092 
3093 	/*
3094 	 * For C+ mode, initialize the RX descriptors and mbufs.
3095 	 */
3096 	if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
3097 		if (ifp->if_mtu > RL_MTU) {
3098 			if (re_jrx_list_init(sc) != 0) {
3099 				device_printf(sc->rl_dev,
3100 				    "no memory for jumbo RX buffers\n");
3101 				re_stop(sc);
3102 				return;
3103 			}
3104 			/* Disable checksum offloading for jumbo frames. */
3105 			ifp->if_capenable &= ~(IFCAP_HWCSUM | IFCAP_TSO4);
3106 			ifp->if_hwassist &= ~(RE_CSUM_FEATURES | CSUM_TSO);
3107 		} else {
3108 			if (re_rx_list_init(sc) != 0) {
3109 				device_printf(sc->rl_dev,
3110 				    "no memory for RX buffers\n");
3111 				re_stop(sc);
3112 				return;
3113 			}
3114 		}
3115 		re_set_jumbo(sc, ifp->if_mtu > RL_MTU);
3116 	} else {
3117 		if (re_rx_list_init(sc) != 0) {
3118 			device_printf(sc->rl_dev, "no memory for RX buffers\n");
3119 			re_stop(sc);
3120 			return;
3121 		}
3122 		if ((sc->rl_flags & RL_FLAG_PCIE) != 0 &&
3123 		    pci_get_device(sc->rl_dev) != RT_DEVICEID_8101E) {
3124 			if (ifp->if_mtu > RL_MTU)
3125 				pci_set_max_read_req(sc->rl_dev, 512);
3126 			else
3127 				pci_set_max_read_req(sc->rl_dev, 4096);
3128 		}
3129 	}
3130 	re_tx_list_init(sc);
3131 
3132 	/*
3133 	 * Enable C+ RX and TX mode, as well as VLAN stripping and
3134 	 * RX checksum offload. We must configure the C+ register
3135 	 * before all others.
3136 	 */
3137 	cfg = RL_CPLUSCMD_PCI_MRW;
3138 	if ((ifp->if_capenable & IFCAP_RXCSUM) != 0)
3139 		cfg |= RL_CPLUSCMD_RXCSUM_ENB;
3140 	if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0)
3141 		cfg |= RL_CPLUSCMD_VLANSTRIP;
3142 	if ((sc->rl_flags & RL_FLAG_MACSTAT) != 0) {
3143 		cfg |= RL_CPLUSCMD_MACSTAT_DIS;
3144 		/* XXX magic. */
3145 		cfg |= 0x0001;
3146 	} else
3147 		cfg |= RL_CPLUSCMD_RXENB | RL_CPLUSCMD_TXENB;
3148 	CSR_WRITE_2(sc, RL_CPLUS_CMD, cfg);
3149 	if (sc->rl_hwrev->rl_rev == RL_HWREV_8169_8110SC ||
3150 	    sc->rl_hwrev->rl_rev == RL_HWREV_8169_8110SCE) {
3151 		reg = 0x000fff00;
3152 		if ((CSR_READ_1(sc, sc->rl_cfg2) & RL_CFG2_PCI66MHZ) != 0)
3153 			reg |= 0x000000ff;
3154 		if (sc->rl_hwrev->rl_rev == RL_HWREV_8169_8110SCE)
3155 			reg |= 0x00f00000;
3156 		CSR_WRITE_4(sc, 0x7c, reg);
3157 		/* Disable interrupt mitigation. */
3158 		CSR_WRITE_2(sc, 0xe2, 0);
3159 	}
3160 	/*
3161 	 * Disable TSO if interface MTU size is greater than MSS
3162 	 * allowed in controller.
3163 	 */
3164 	if (ifp->if_mtu > RL_TSO_MTU && (ifp->if_capenable & IFCAP_TSO4) != 0) {
3165 		ifp->if_capenable &= ~IFCAP_TSO4;
3166 		ifp->if_hwassist &= ~CSUM_TSO;
3167 	}
3168 
3169 	/*
3170 	 * Init our MAC address.  Even though the chipset
3171 	 * documentation doesn't mention it, we need to enter "Config
3172 	 * register write enable" mode to modify the ID registers.
3173 	 */
3174 	/* Copy MAC address on stack to align. */
3175 	bcopy(IF_LLADDR(ifp), eaddr.eaddr, ETHER_ADDR_LEN);
3176 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG);
3177 	CSR_WRITE_4(sc, RL_IDR0,
3178 	    htole32(*(u_int32_t *)(&eaddr.eaddr[0])));
3179 	CSR_WRITE_4(sc, RL_IDR4,
3180 	    htole32(*(u_int32_t *)(&eaddr.eaddr[4])));
3181 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3182 
3183 	/*
3184 	 * Load the addresses of the RX and TX lists into the chip.
3185 	 */
3186 
3187 	CSR_WRITE_4(sc, RL_RXLIST_ADDR_HI,
3188 	    RL_ADDR_HI(sc->rl_ldata.rl_rx_list_addr));
3189 	CSR_WRITE_4(sc, RL_RXLIST_ADDR_LO,
3190 	    RL_ADDR_LO(sc->rl_ldata.rl_rx_list_addr));
3191 
3192 	CSR_WRITE_4(sc, RL_TXLIST_ADDR_HI,
3193 	    RL_ADDR_HI(sc->rl_ldata.rl_tx_list_addr));
3194 	CSR_WRITE_4(sc, RL_TXLIST_ADDR_LO,
3195 	    RL_ADDR_LO(sc->rl_ldata.rl_tx_list_addr));
3196 
3197 	if ((sc->rl_flags & RL_FLAG_RXDV_GATED) != 0)
3198 		CSR_WRITE_4(sc, RL_MISC, CSR_READ_4(sc, RL_MISC) &
3199 		    ~0x00080000);
3200 
3201 	/*
3202 	 * Set the initial TX configuration.
3203 	 */
3204 	if (sc->rl_testmode) {
3205 		if (sc->rl_type == RL_8169)
3206 			CSR_WRITE_4(sc, RL_TXCFG,
3207 			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON);
3208 		else
3209 			CSR_WRITE_4(sc, RL_TXCFG,
3210 			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON_CPLUS);
3211 	} else
3212 		CSR_WRITE_4(sc, RL_TXCFG, RL_TXCFG_CONFIG);
3213 
3214 	CSR_WRITE_1(sc, RL_EARLY_TX_THRESH, 16);
3215 
3216 	/*
3217 	 * Set the initial RX configuration.
3218 	 */
3219 	re_set_rxmode(sc);
3220 
3221 	/* Configure interrupt moderation. */
3222 	if (sc->rl_type == RL_8169) {
3223 		/* Magic from vendor. */
3224 		CSR_WRITE_2(sc, RL_INTRMOD, 0x5100);
3225 	}
3226 
3227 	/*
3228 	 * Enable transmit and receive.
3229 	 */
3230 	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB | RL_CMD_RX_ENB);
3231 
3232 #ifdef DEVICE_POLLING
3233 	/*
3234 	 * Disable interrupts if we are polling.
3235 	 */
3236 	if (ifp->if_capenable & IFCAP_POLLING)
3237 		CSR_WRITE_2(sc, RL_IMR, 0);
3238 	else	/* otherwise ... */
3239 #endif
3240 
3241 	/*
3242 	 * Enable interrupts.
3243 	 */
3244 	if (sc->rl_testmode)
3245 		CSR_WRITE_2(sc, RL_IMR, 0);
3246 	else
3247 		CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
3248 	CSR_WRITE_2(sc, RL_ISR, RL_INTRS_CPLUS);
3249 
3250 	/* Set initial TX threshold */
3251 	sc->rl_txthresh = RL_TX_THRESH_INIT;
3252 
3253 	/* Start RX/TX process. */
3254 	CSR_WRITE_4(sc, RL_MISSEDPKT, 0);
3255 
3256 	/*
3257 	 * Initialize the timer interrupt register so that
3258 	 * a timer interrupt will be generated once the timer
3259 	 * reaches a certain number of ticks. The timer is
3260 	 * reloaded on each transmit.
3261 	 */
3262 #ifdef RE_TX_MODERATION
3263 	/*
3264 	 * Use timer interrupt register to moderate TX interrupt
3265 	 * moderation, which dramatically improves TX frame rate.
3266 	 */
3267 	if (sc->rl_type == RL_8169)
3268 		CSR_WRITE_4(sc, RL_TIMERINT_8169, 0x800);
3269 	else
3270 		CSR_WRITE_4(sc, RL_TIMERINT, 0x400);
3271 #else
3272 	/*
3273 	 * Use timer interrupt register to moderate RX interrupt
3274 	 * moderation.
3275 	 */
3276 	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) != 0 &&
3277 	    intr_filter == 0) {
3278 		if (sc->rl_type == RL_8169)
3279 			CSR_WRITE_4(sc, RL_TIMERINT_8169,
3280 			    RL_USECS(sc->rl_int_rx_mod));
3281 	} else {
3282 		if (sc->rl_type == RL_8169)
3283 			CSR_WRITE_4(sc, RL_TIMERINT_8169, RL_USECS(0));
3284 	}
3285 #endif
3286 
3287 	/*
3288 	 * For 8169 gigE NICs, set the max allowed RX packet
3289 	 * size so we can receive jumbo frames.
3290 	 */
3291 	if (sc->rl_type == RL_8169) {
3292 		if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
3293 			/*
3294 			 * For controllers that use new jumbo frame scheme,
3295 			 * set maximum size of jumbo frame depending on
3296 			 * controller revisions.
3297 			 */
3298 			if (ifp->if_mtu > RL_MTU)
3299 				CSR_WRITE_2(sc, RL_MAXRXPKTLEN,
3300 				    sc->rl_hwrev->rl_max_mtu +
3301 				    ETHER_VLAN_ENCAP_LEN + ETHER_HDR_LEN +
3302 				    ETHER_CRC_LEN);
3303 			else
3304 				CSR_WRITE_2(sc, RL_MAXRXPKTLEN,
3305 				    RE_RX_DESC_BUFLEN);
3306 		} else if ((sc->rl_flags & RL_FLAG_PCIE) != 0 &&
3307 		    sc->rl_hwrev->rl_max_mtu == RL_MTU) {
3308 			/* RTL810x has no jumbo frame support. */
3309 			CSR_WRITE_2(sc, RL_MAXRXPKTLEN, RE_RX_DESC_BUFLEN);
3310 		} else
3311 			CSR_WRITE_2(sc, RL_MAXRXPKTLEN, 16383);
3312 	}
3313 
3314 	if (sc->rl_testmode)
3315 		return;
3316 
3317 	CSR_WRITE_1(sc, sc->rl_cfg1, CSR_READ_1(sc, sc->rl_cfg1) |
3318 	    RL_CFG1_DRVLOAD);
3319 
3320 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
3321 	ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
3322 
3323 	sc->rl_flags &= ~RL_FLAG_LINK;
3324 	mii_mediachg(mii);
3325 
3326 	sc->rl_watchdog_timer = 0;
3327 	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
3328 }
3329 
3330 /*
3331  * Set media options.
3332  */
3333 static int
3334 re_ifmedia_upd(struct ifnet *ifp)
3335 {
3336 	struct rl_softc		*sc;
3337 	struct mii_data		*mii;
3338 	int			error;
3339 
3340 	sc = ifp->if_softc;
3341 	mii = device_get_softc(sc->rl_miibus);
3342 	RL_LOCK(sc);
3343 	error = mii_mediachg(mii);
3344 	RL_UNLOCK(sc);
3345 
3346 	return (error);
3347 }
3348 
3349 /*
3350  * Report current media status.
3351  */
3352 static void
3353 re_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
3354 {
3355 	struct rl_softc		*sc;
3356 	struct mii_data		*mii;
3357 
3358 	sc = ifp->if_softc;
3359 	mii = device_get_softc(sc->rl_miibus);
3360 
3361 	RL_LOCK(sc);
3362 	mii_pollstat(mii);
3363 	ifmr->ifm_active = mii->mii_media_active;
3364 	ifmr->ifm_status = mii->mii_media_status;
3365 	RL_UNLOCK(sc);
3366 }
3367 
3368 static int
3369 re_ioctl(struct ifnet *ifp, u_long command, caddr_t data)
3370 {
3371 	struct rl_softc		*sc = ifp->if_softc;
3372 	struct ifreq		*ifr = (struct ifreq *) data;
3373 	struct mii_data		*mii;
3374 	int			error = 0;
3375 
3376 	switch (command) {
3377 	case SIOCSIFMTU:
3378 		if (ifr->ifr_mtu < ETHERMIN ||
3379 		    ifr->ifr_mtu > sc->rl_hwrev->rl_max_mtu ||
3380 		    ((sc->rl_flags & RL_FLAG_FASTETHER) != 0 &&
3381 		    ifr->ifr_mtu > RL_MTU)) {
3382 			error = EINVAL;
3383 			break;
3384 		}
3385 		RL_LOCK(sc);
3386 		if (ifp->if_mtu != ifr->ifr_mtu) {
3387 			ifp->if_mtu = ifr->ifr_mtu;
3388 			if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0 &&
3389 			    (ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
3390 				ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3391 				re_init_locked(sc);
3392 			}
3393 			if (ifp->if_mtu > RL_TSO_MTU &&
3394 			    (ifp->if_capenable & IFCAP_TSO4) != 0) {
3395 				ifp->if_capenable &= ~(IFCAP_TSO4 |
3396 				    IFCAP_VLAN_HWTSO);
3397 				ifp->if_hwassist &= ~CSUM_TSO;
3398 			}
3399 			VLAN_CAPABILITIES(ifp);
3400 		}
3401 		RL_UNLOCK(sc);
3402 		break;
3403 	case SIOCSIFFLAGS:
3404 		RL_LOCK(sc);
3405 		if ((ifp->if_flags & IFF_UP) != 0) {
3406 			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
3407 				if (((ifp->if_flags ^ sc->rl_if_flags)
3408 				    & (IFF_PROMISC | IFF_ALLMULTI)) != 0)
3409 					re_set_rxmode(sc);
3410 			} else
3411 				re_init_locked(sc);
3412 		} else {
3413 			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
3414 				re_stop(sc);
3415 		}
3416 		sc->rl_if_flags = ifp->if_flags;
3417 		RL_UNLOCK(sc);
3418 		break;
3419 	case SIOCADDMULTI:
3420 	case SIOCDELMULTI:
3421 		RL_LOCK(sc);
3422 		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
3423 			re_set_rxmode(sc);
3424 		RL_UNLOCK(sc);
3425 		break;
3426 	case SIOCGIFMEDIA:
3427 	case SIOCSIFMEDIA:
3428 		mii = device_get_softc(sc->rl_miibus);
3429 		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, command);
3430 		break;
3431 	case SIOCSIFCAP:
3432 	    {
3433 		int mask, reinit;
3434 
3435 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
3436 		reinit = 0;
3437 #ifdef DEVICE_POLLING
3438 		if (mask & IFCAP_POLLING) {
3439 			if (ifr->ifr_reqcap & IFCAP_POLLING) {
3440 				error = ether_poll_register(re_poll, ifp);
3441 				if (error)
3442 					return (error);
3443 				RL_LOCK(sc);
3444 				/* Disable interrupts */
3445 				CSR_WRITE_2(sc, RL_IMR, 0x0000);
3446 				ifp->if_capenable |= IFCAP_POLLING;
3447 				RL_UNLOCK(sc);
3448 			} else {
3449 				error = ether_poll_deregister(ifp);
3450 				/* Enable interrupts. */
3451 				RL_LOCK(sc);
3452 				CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
3453 				ifp->if_capenable &= ~IFCAP_POLLING;
3454 				RL_UNLOCK(sc);
3455 			}
3456 		}
3457 #endif /* DEVICE_POLLING */
3458 		RL_LOCK(sc);
3459 		if ((mask & IFCAP_TXCSUM) != 0 &&
3460 		    (ifp->if_capabilities & IFCAP_TXCSUM) != 0) {
3461 			ifp->if_capenable ^= IFCAP_TXCSUM;
3462 			if ((ifp->if_capenable & IFCAP_TXCSUM) != 0)
3463 				ifp->if_hwassist |= RE_CSUM_FEATURES;
3464 			else
3465 				ifp->if_hwassist &= ~RE_CSUM_FEATURES;
3466 			reinit = 1;
3467 		}
3468 		if ((mask & IFCAP_RXCSUM) != 0 &&
3469 		    (ifp->if_capabilities & IFCAP_RXCSUM) != 0) {
3470 			ifp->if_capenable ^= IFCAP_RXCSUM;
3471 			reinit = 1;
3472 		}
3473 		if ((mask & IFCAP_TSO4) != 0 &&
3474 		    (ifp->if_capabilities & IFCAP_TSO4) != 0) {
3475 			ifp->if_capenable ^= IFCAP_TSO4;
3476 			if ((IFCAP_TSO4 & ifp->if_capenable) != 0)
3477 				ifp->if_hwassist |= CSUM_TSO;
3478 			else
3479 				ifp->if_hwassist &= ~CSUM_TSO;
3480 			if (ifp->if_mtu > RL_TSO_MTU &&
3481 			    (ifp->if_capenable & IFCAP_TSO4) != 0) {
3482 				ifp->if_capenable &= ~IFCAP_TSO4;
3483 				ifp->if_hwassist &= ~CSUM_TSO;
3484 			}
3485 		}
3486 		if ((mask & IFCAP_VLAN_HWTSO) != 0 &&
3487 		    (ifp->if_capabilities & IFCAP_VLAN_HWTSO) != 0)
3488 			ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
3489 		if ((mask & IFCAP_VLAN_HWTAGGING) != 0 &&
3490 		    (ifp->if_capabilities & IFCAP_VLAN_HWTAGGING) != 0) {
3491 			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
3492 			/* TSO over VLAN requires VLAN hardware tagging. */
3493 			if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) == 0)
3494 				ifp->if_capenable &= ~IFCAP_VLAN_HWTSO;
3495 			reinit = 1;
3496 		}
3497 		if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0 &&
3498 		    (mask & (IFCAP_HWCSUM | IFCAP_TSO4 |
3499 		    IFCAP_VLAN_HWTSO)) != 0)
3500 				reinit = 1;
3501 		if ((mask & IFCAP_WOL) != 0 &&
3502 		    (ifp->if_capabilities & IFCAP_WOL) != 0) {
3503 			if ((mask & IFCAP_WOL_UCAST) != 0)
3504 				ifp->if_capenable ^= IFCAP_WOL_UCAST;
3505 			if ((mask & IFCAP_WOL_MCAST) != 0)
3506 				ifp->if_capenable ^= IFCAP_WOL_MCAST;
3507 			if ((mask & IFCAP_WOL_MAGIC) != 0)
3508 				ifp->if_capenable ^= IFCAP_WOL_MAGIC;
3509 		}
3510 		if (reinit && ifp->if_drv_flags & IFF_DRV_RUNNING) {
3511 			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3512 			re_init_locked(sc);
3513 		}
3514 		RL_UNLOCK(sc);
3515 		VLAN_CAPABILITIES(ifp);
3516 	    }
3517 		break;
3518 	default:
3519 		error = ether_ioctl(ifp, command, data);
3520 		break;
3521 	}
3522 
3523 	return (error);
3524 }
3525 
3526 static void
3527 re_watchdog(struct rl_softc *sc)
3528 {
3529 	struct ifnet		*ifp;
3530 
3531 	RL_LOCK_ASSERT(sc);
3532 
3533 	if (sc->rl_watchdog_timer == 0 || --sc->rl_watchdog_timer != 0)
3534 		return;
3535 
3536 	ifp = sc->rl_ifp;
3537 	re_txeof(sc);
3538 	if (sc->rl_ldata.rl_tx_free == sc->rl_ldata.rl_tx_desc_cnt) {
3539 		if_printf(ifp, "watchdog timeout (missed Tx interrupts) "
3540 		    "-- recovering\n");
3541 		if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
3542 			re_start_locked(ifp);
3543 		return;
3544 	}
3545 
3546 	if_printf(ifp, "watchdog timeout\n");
3547 	if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
3548 
3549 	re_rxeof(sc, NULL);
3550 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3551 	re_init_locked(sc);
3552 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
3553 		re_start_locked(ifp);
3554 }
3555 
3556 /*
3557  * Stop the adapter and free any mbufs allocated to the
3558  * RX and TX lists.
3559  */
3560 static void
3561 re_stop(struct rl_softc *sc)
3562 {
3563 	int			i;
3564 	struct ifnet		*ifp;
3565 	struct rl_txdesc	*txd;
3566 	struct rl_rxdesc	*rxd;
3567 
3568 	RL_LOCK_ASSERT(sc);
3569 
3570 	ifp = sc->rl_ifp;
3571 
3572 	sc->rl_watchdog_timer = 0;
3573 	callout_stop(&sc->rl_stat_callout);
3574 	ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
3575 
3576 	/*
3577 	 * Disable accepting frames to put RX MAC into idle state.
3578 	 * Otherwise it's possible to get frames while stop command
3579 	 * execution is in progress and controller can DMA the frame
3580 	 * to already freed RX buffer during that period.
3581 	 */
3582 	CSR_WRITE_4(sc, RL_RXCFG, CSR_READ_4(sc, RL_RXCFG) &
3583 	    ~(RL_RXCFG_RX_ALLPHYS | RL_RXCFG_RX_INDIV | RL_RXCFG_RX_MULTI |
3584 	    RL_RXCFG_RX_BROAD));
3585 
3586 	if ((sc->rl_flags & RL_FLAG_WAIT_TXPOLL) != 0) {
3587 		for (i = RL_TIMEOUT; i > 0; i--) {
3588 			if ((CSR_READ_1(sc, sc->rl_txstart) &
3589 			    RL_TXSTART_START) == 0)
3590 				break;
3591 			DELAY(20);
3592 		}
3593 		if (i == 0)
3594 			device_printf(sc->rl_dev,
3595 			    "stopping TX poll timed out!\n");
3596 		CSR_WRITE_1(sc, RL_COMMAND, 0x00);
3597 	} else if ((sc->rl_flags & RL_FLAG_CMDSTOP) != 0) {
3598 		CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_STOPREQ | RL_CMD_TX_ENB |
3599 		    RL_CMD_RX_ENB);
3600 		if ((sc->rl_flags & RL_FLAG_CMDSTOP_WAIT_TXQ) != 0) {
3601 			for (i = RL_TIMEOUT; i > 0; i--) {
3602 				if ((CSR_READ_4(sc, RL_TXCFG) &
3603 				    RL_TXCFG_QUEUE_EMPTY) != 0)
3604 					break;
3605 				DELAY(100);
3606 			}
3607 			if (i == 0)
3608 				device_printf(sc->rl_dev,
3609 				   "stopping TXQ timed out!\n");
3610 		}
3611 	} else
3612 		CSR_WRITE_1(sc, RL_COMMAND, 0x00);
3613 	DELAY(1000);
3614 	CSR_WRITE_2(sc, RL_IMR, 0x0000);
3615 	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
3616 
3617 	if (sc->rl_head != NULL) {
3618 		m_freem(sc->rl_head);
3619 		sc->rl_head = sc->rl_tail = NULL;
3620 	}
3621 
3622 	/* Free the TX list buffers. */
3623 	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
3624 		txd = &sc->rl_ldata.rl_tx_desc[i];
3625 		if (txd->tx_m != NULL) {
3626 			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
3627 			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
3628 			bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag,
3629 			    txd->tx_dmamap);
3630 			m_freem(txd->tx_m);
3631 			txd->tx_m = NULL;
3632 		}
3633 	}
3634 
3635 	/* Free the RX list buffers. */
3636 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
3637 		rxd = &sc->rl_ldata.rl_rx_desc[i];
3638 		if (rxd->rx_m != NULL) {
3639 			bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag,
3640 			    rxd->rx_dmamap, BUS_DMASYNC_POSTREAD);
3641 			bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag,
3642 			    rxd->rx_dmamap);
3643 			m_freem(rxd->rx_m);
3644 			rxd->rx_m = NULL;
3645 		}
3646 	}
3647 
3648 	if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
3649 		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
3650 			rxd = &sc->rl_ldata.rl_jrx_desc[i];
3651 			if (rxd->rx_m != NULL) {
3652 				bus_dmamap_sync(sc->rl_ldata.rl_jrx_mtag,
3653 				    rxd->rx_dmamap, BUS_DMASYNC_POSTREAD);
3654 				bus_dmamap_unload(sc->rl_ldata.rl_jrx_mtag,
3655 				    rxd->rx_dmamap);
3656 				m_freem(rxd->rx_m);
3657 				rxd->rx_m = NULL;
3658 			}
3659 		}
3660 	}
3661 }
3662 
3663 /*
3664  * Device suspend routine.  Stop the interface and save some PCI
3665  * settings in case the BIOS doesn't restore them properly on
3666  * resume.
3667  */
3668 static int
3669 re_suspend(device_t dev)
3670 {
3671 	struct rl_softc		*sc;
3672 
3673 	sc = device_get_softc(dev);
3674 
3675 	RL_LOCK(sc);
3676 	re_stop(sc);
3677 	re_setwol(sc);
3678 	sc->suspended = 1;
3679 	RL_UNLOCK(sc);
3680 
3681 	return (0);
3682 }
3683 
3684 /*
3685  * Device resume routine.  Restore some PCI settings in case the BIOS
3686  * doesn't, re-enable busmastering, and restart the interface if
3687  * appropriate.
3688  */
3689 static int
3690 re_resume(device_t dev)
3691 {
3692 	struct rl_softc		*sc;
3693 	struct ifnet		*ifp;
3694 
3695 	sc = device_get_softc(dev);
3696 
3697 	RL_LOCK(sc);
3698 
3699 	ifp = sc->rl_ifp;
3700 	/* Take controller out of sleep mode. */
3701 	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
3702 		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
3703 			CSR_WRITE_1(sc, RL_GPIO,
3704 			    CSR_READ_1(sc, RL_GPIO) | 0x01);
3705 	}
3706 
3707 	/*
3708 	 * Clear WOL matching such that normal Rx filtering
3709 	 * wouldn't interfere with WOL patterns.
3710 	 */
3711 	re_clrwol(sc);
3712 
3713 	/* reinitialize interface if necessary */
3714 	if (ifp->if_flags & IFF_UP)
3715 		re_init_locked(sc);
3716 
3717 	sc->suspended = 0;
3718 	RL_UNLOCK(sc);
3719 
3720 	return (0);
3721 }
3722 
3723 /*
3724  * Stop all chip I/O so that the kernel's probe routines don't
3725  * get confused by errant DMAs when rebooting.
3726  */
3727 static int
3728 re_shutdown(device_t dev)
3729 {
3730 	struct rl_softc		*sc;
3731 
3732 	sc = device_get_softc(dev);
3733 
3734 	RL_LOCK(sc);
3735 	re_stop(sc);
3736 	/*
3737 	 * Mark interface as down since otherwise we will panic if
3738 	 * interrupt comes in later on, which can happen in some
3739 	 * cases.
3740 	 */
3741 	sc->rl_ifp->if_flags &= ~IFF_UP;
3742 	re_setwol(sc);
3743 	RL_UNLOCK(sc);
3744 
3745 	return (0);
3746 }
3747 
3748 static void
3749 re_set_linkspeed(struct rl_softc *sc)
3750 {
3751 	struct mii_softc *miisc;
3752 	struct mii_data *mii;
3753 	int aneg, i, phyno;
3754 
3755 	RL_LOCK_ASSERT(sc);
3756 
3757 	mii = device_get_softc(sc->rl_miibus);
3758 	mii_pollstat(mii);
3759 	aneg = 0;
3760 	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
3761 	    (IFM_ACTIVE | IFM_AVALID)) {
3762 		switch IFM_SUBTYPE(mii->mii_media_active) {
3763 		case IFM_10_T:
3764 		case IFM_100_TX:
3765 			return;
3766 		case IFM_1000_T:
3767 			aneg++;
3768 			break;
3769 		default:
3770 			break;
3771 		}
3772 	}
3773 	miisc = LIST_FIRST(&mii->mii_phys);
3774 	phyno = miisc->mii_phy;
3775 	LIST_FOREACH(miisc, &mii->mii_phys, mii_list)
3776 		PHY_RESET(miisc);
3777 	re_miibus_writereg(sc->rl_dev, phyno, MII_100T2CR, 0);
3778 	re_miibus_writereg(sc->rl_dev, phyno,
3779 	    MII_ANAR, ANAR_TX_FD | ANAR_TX | ANAR_10_FD | ANAR_10 | ANAR_CSMA);
3780 	re_miibus_writereg(sc->rl_dev, phyno,
3781 	    MII_BMCR, BMCR_AUTOEN | BMCR_STARTNEG);
3782 	DELAY(1000);
3783 	if (aneg != 0) {
3784 		/*
3785 		 * Poll link state until re(4) get a 10/100Mbps link.
3786 		 */
3787 		for (i = 0; i < MII_ANEGTICKS_GIGE; i++) {
3788 			mii_pollstat(mii);
3789 			if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID))
3790 			    == (IFM_ACTIVE | IFM_AVALID)) {
3791 				switch (IFM_SUBTYPE(mii->mii_media_active)) {
3792 				case IFM_10_T:
3793 				case IFM_100_TX:
3794 					return;
3795 				default:
3796 					break;
3797 				}
3798 			}
3799 			RL_UNLOCK(sc);
3800 			pause("relnk", hz);
3801 			RL_LOCK(sc);
3802 		}
3803 		if (i == MII_ANEGTICKS_GIGE)
3804 			device_printf(sc->rl_dev,
3805 			    "establishing a link failed, WOL may not work!");
3806 	}
3807 	/*
3808 	 * No link, force MAC to have 100Mbps, full-duplex link.
3809 	 * MAC does not require reprogramming on resolved speed/duplex,
3810 	 * so this is just for completeness.
3811 	 */
3812 	mii->mii_media_status = IFM_AVALID | IFM_ACTIVE;
3813 	mii->mii_media_active = IFM_ETHER | IFM_100_TX | IFM_FDX;
3814 }
3815 
3816 static void
3817 re_setwol(struct rl_softc *sc)
3818 {
3819 	struct ifnet		*ifp;
3820 	int			pmc;
3821 	uint16_t		pmstat;
3822 	uint8_t			v;
3823 
3824 	RL_LOCK_ASSERT(sc);
3825 
3826 	if (pci_find_cap(sc->rl_dev, PCIY_PMG, &pmc) != 0)
3827 		return;
3828 
3829 	ifp = sc->rl_ifp;
3830 	/* Put controller into sleep mode. */
3831 	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
3832 		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
3833 			CSR_WRITE_1(sc, RL_GPIO,
3834 			    CSR_READ_1(sc, RL_GPIO) & ~0x01);
3835 	}
3836 	if ((ifp->if_capenable & IFCAP_WOL) != 0) {
3837 		re_set_rxmode(sc);
3838 		if ((sc->rl_flags & RL_FLAG_WOL_MANLINK) != 0)
3839 			re_set_linkspeed(sc);
3840 		if ((sc->rl_flags & RL_FLAG_WOLRXENB) != 0)
3841 			CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RX_ENB);
3842 	}
3843 	/* Enable config register write. */
3844 	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
3845 
3846 	/* Enable PME. */
3847 	v = CSR_READ_1(sc, sc->rl_cfg1);
3848 	v &= ~RL_CFG1_PME;
3849 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3850 		v |= RL_CFG1_PME;
3851 	CSR_WRITE_1(sc, sc->rl_cfg1, v);
3852 
3853 	v = CSR_READ_1(sc, sc->rl_cfg3);
3854 	v &= ~(RL_CFG3_WOL_LINK | RL_CFG3_WOL_MAGIC);
3855 	if ((ifp->if_capenable & IFCAP_WOL_MAGIC) != 0)
3856 		v |= RL_CFG3_WOL_MAGIC;
3857 	CSR_WRITE_1(sc, sc->rl_cfg3, v);
3858 
3859 	v = CSR_READ_1(sc, sc->rl_cfg5);
3860 	v &= ~(RL_CFG5_WOL_BCAST | RL_CFG5_WOL_MCAST | RL_CFG5_WOL_UCAST |
3861 	    RL_CFG5_WOL_LANWAKE);
3862 	if ((ifp->if_capenable & IFCAP_WOL_UCAST) != 0)
3863 		v |= RL_CFG5_WOL_UCAST;
3864 	if ((ifp->if_capenable & IFCAP_WOL_MCAST) != 0)
3865 		v |= RL_CFG5_WOL_MCAST | RL_CFG5_WOL_BCAST;
3866 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3867 		v |= RL_CFG5_WOL_LANWAKE;
3868 	CSR_WRITE_1(sc, sc->rl_cfg5, v);
3869 
3870 	/* Config register write done. */
3871 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3872 
3873 	if ((ifp->if_capenable & IFCAP_WOL) == 0 &&
3874 	    (sc->rl_flags & RL_FLAG_PHYWAKE_PM) != 0)
3875 		CSR_WRITE_1(sc, RL_PMCH, CSR_READ_1(sc, RL_PMCH) & ~0x80);
3876 	/*
3877 	 * It seems that hardware resets its link speed to 100Mbps in
3878 	 * power down mode so switching to 100Mbps in driver is not
3879 	 * needed.
3880 	 */
3881 
3882 	/* Request PME if WOL is requested. */
3883 	pmstat = pci_read_config(sc->rl_dev, pmc + PCIR_POWER_STATUS, 2);
3884 	pmstat &= ~(PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE);
3885 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3886 		pmstat |= PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE;
3887 	pci_write_config(sc->rl_dev, pmc + PCIR_POWER_STATUS, pmstat, 2);
3888 }
3889 
3890 static void
3891 re_clrwol(struct rl_softc *sc)
3892 {
3893 	int			pmc;
3894 	uint8_t			v;
3895 
3896 	RL_LOCK_ASSERT(sc);
3897 
3898 	if (pci_find_cap(sc->rl_dev, PCIY_PMG, &pmc) != 0)
3899 		return;
3900 
3901 	/* Enable config register write. */
3902 	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
3903 
3904 	v = CSR_READ_1(sc, sc->rl_cfg3);
3905 	v &= ~(RL_CFG3_WOL_LINK | RL_CFG3_WOL_MAGIC);
3906 	CSR_WRITE_1(sc, sc->rl_cfg3, v);
3907 
3908 	/* Config register write done. */
3909 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3910 
3911 	v = CSR_READ_1(sc, sc->rl_cfg5);
3912 	v &= ~(RL_CFG5_WOL_BCAST | RL_CFG5_WOL_MCAST | RL_CFG5_WOL_UCAST);
3913 	v &= ~RL_CFG5_WOL_LANWAKE;
3914 	CSR_WRITE_1(sc, sc->rl_cfg5, v);
3915 }
3916 
3917 static void
3918 re_add_sysctls(struct rl_softc *sc)
3919 {
3920 	struct sysctl_ctx_list	*ctx;
3921 	struct sysctl_oid_list	*children;
3922 	int			error;
3923 
3924 	ctx = device_get_sysctl_ctx(sc->rl_dev);
3925 	children = SYSCTL_CHILDREN(device_get_sysctl_tree(sc->rl_dev));
3926 
3927 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "stats",
3928 	    CTLTYPE_INT | CTLFLAG_RW, sc, 0, re_sysctl_stats, "I",
3929 	    "Statistics Information");
3930 	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) == 0)
3931 		return;
3932 
3933 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "int_rx_mod",
3934 	    CTLTYPE_INT | CTLFLAG_RW, &sc->rl_int_rx_mod, 0,
3935 	    sysctl_hw_re_int_mod, "I", "re RX interrupt moderation");
3936 	/* Pull in device tunables. */
3937 	sc->rl_int_rx_mod = RL_TIMER_DEFAULT;
3938 	error = resource_int_value(device_get_name(sc->rl_dev),
3939 	    device_get_unit(sc->rl_dev), "int_rx_mod", &sc->rl_int_rx_mod);
3940 	if (error == 0) {
3941 		if (sc->rl_int_rx_mod < RL_TIMER_MIN ||
3942 		    sc->rl_int_rx_mod > RL_TIMER_MAX) {
3943 			device_printf(sc->rl_dev, "int_rx_mod value out of "
3944 			    "range; using default: %d\n",
3945 			    RL_TIMER_DEFAULT);
3946 			sc->rl_int_rx_mod = RL_TIMER_DEFAULT;
3947 		}
3948 	}
3949 
3950 }
3951 
3952 static int
3953 re_sysctl_stats(SYSCTL_HANDLER_ARGS)
3954 {
3955 	struct rl_softc		*sc;
3956 	struct rl_stats		*stats;
3957 	int			error, i, result;
3958 
3959 	result = -1;
3960 	error = sysctl_handle_int(oidp, &result, 0, req);
3961 	if (error || req->newptr == NULL)
3962 		return (error);
3963 
3964 	if (result == 1) {
3965 		sc = (struct rl_softc *)arg1;
3966 		RL_LOCK(sc);
3967 		if ((sc->rl_ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
3968 			RL_UNLOCK(sc);
3969 			goto done;
3970 		}
3971 		bus_dmamap_sync(sc->rl_ldata.rl_stag,
3972 		    sc->rl_ldata.rl_smap, BUS_DMASYNC_PREREAD);
3973 		CSR_WRITE_4(sc, RL_DUMPSTATS_HI,
3974 		    RL_ADDR_HI(sc->rl_ldata.rl_stats_addr));
3975 		CSR_WRITE_4(sc, RL_DUMPSTATS_LO,
3976 		    RL_ADDR_LO(sc->rl_ldata.rl_stats_addr));
3977 		CSR_WRITE_4(sc, RL_DUMPSTATS_LO,
3978 		    RL_ADDR_LO(sc->rl_ldata.rl_stats_addr |
3979 		    RL_DUMPSTATS_START));
3980 		for (i = RL_TIMEOUT; i > 0; i--) {
3981 			if ((CSR_READ_4(sc, RL_DUMPSTATS_LO) &
3982 			    RL_DUMPSTATS_START) == 0)
3983 				break;
3984 			DELAY(1000);
3985 		}
3986 		bus_dmamap_sync(sc->rl_ldata.rl_stag,
3987 		    sc->rl_ldata.rl_smap, BUS_DMASYNC_POSTREAD);
3988 		RL_UNLOCK(sc);
3989 		if (i == 0) {
3990 			device_printf(sc->rl_dev,
3991 			    "DUMP statistics request timed out\n");
3992 			return (ETIMEDOUT);
3993 		}
3994 done:
3995 		stats = sc->rl_ldata.rl_stats;
3996 		printf("%s statistics:\n", device_get_nameunit(sc->rl_dev));
3997 		printf("Tx frames : %ju\n",
3998 		    (uintmax_t)le64toh(stats->rl_tx_pkts));
3999 		printf("Rx frames : %ju\n",
4000 		    (uintmax_t)le64toh(stats->rl_rx_pkts));
4001 		printf("Tx errors : %ju\n",
4002 		    (uintmax_t)le64toh(stats->rl_tx_errs));
4003 		printf("Rx errors : %u\n",
4004 		    le32toh(stats->rl_rx_errs));
4005 		printf("Rx missed frames : %u\n",
4006 		    (uint32_t)le16toh(stats->rl_missed_pkts));
4007 		printf("Rx frame alignment errs : %u\n",
4008 		    (uint32_t)le16toh(stats->rl_rx_framealign_errs));
4009 		printf("Tx single collisions : %u\n",
4010 		    le32toh(stats->rl_tx_onecoll));
4011 		printf("Tx multiple collisions : %u\n",
4012 		    le32toh(stats->rl_tx_multicolls));
4013 		printf("Rx unicast frames : %ju\n",
4014 		    (uintmax_t)le64toh(stats->rl_rx_ucasts));
4015 		printf("Rx broadcast frames : %ju\n",
4016 		    (uintmax_t)le64toh(stats->rl_rx_bcasts));
4017 		printf("Rx multicast frames : %u\n",
4018 		    le32toh(stats->rl_rx_mcasts));
4019 		printf("Tx aborts : %u\n",
4020 		    (uint32_t)le16toh(stats->rl_tx_aborts));
4021 		printf("Tx underruns : %u\n",
4022 		    (uint32_t)le16toh(stats->rl_rx_underruns));
4023 	}
4024 
4025 	return (error);
4026 }
4027 
4028 static int
4029 sysctl_int_range(SYSCTL_HANDLER_ARGS, int low, int high)
4030 {
4031 	int error, value;
4032 
4033 	if (arg1 == NULL)
4034 		return (EINVAL);
4035 	value = *(int *)arg1;
4036 	error = sysctl_handle_int(oidp, &value, 0, req);
4037 	if (error || req->newptr == NULL)
4038 		return (error);
4039 	if (value < low || value > high)
4040 		return (EINVAL);
4041 	*(int *)arg1 = value;
4042 
4043 	return (0);
4044 }
4045 
4046 static int
4047 sysctl_hw_re_int_mod(SYSCTL_HANDLER_ARGS)
4048 {
4049 
4050 	return (sysctl_int_range(oidp, arg1, arg2, req, RL_TIMER_MIN,
4051 	    RL_TIMER_MAX));
4052 }
4053