xref: /freebsd/sys/dev/re/if_re.c (revision 6574b8ed19b093f0af09501d2c9676c28993cb97)
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 <pci/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_8411, RL_8169, "8411", RL_JUMBO_MTU_9K},
242 	{ RL_HWREV_8411B, RL_8169, "8411B", RL_JUMBO_MTU_9K},
243 	{ 0, 0, NULL, 0 }
244 };
245 
246 static int re_probe		(device_t);
247 static int re_attach		(device_t);
248 static int re_detach		(device_t);
249 
250 static int re_encap		(struct rl_softc *, struct mbuf **);
251 
252 static void re_dma_map_addr	(void *, bus_dma_segment_t *, int, int);
253 static int re_allocmem		(device_t, struct rl_softc *);
254 static __inline void re_discard_rxbuf
255 				(struct rl_softc *, int);
256 static int re_newbuf		(struct rl_softc *, int);
257 static int re_jumbo_newbuf	(struct rl_softc *, int);
258 static int re_rx_list_init	(struct rl_softc *);
259 static int re_jrx_list_init	(struct rl_softc *);
260 static int re_tx_list_init	(struct rl_softc *);
261 #ifdef RE_FIXUP_RX
262 static __inline void re_fixup_rx
263 				(struct mbuf *);
264 #endif
265 static int re_rxeof		(struct rl_softc *, int *);
266 static void re_txeof		(struct rl_softc *);
267 #ifdef DEVICE_POLLING
268 static int re_poll		(struct ifnet *, enum poll_cmd, int);
269 static int re_poll_locked	(struct ifnet *, enum poll_cmd, int);
270 #endif
271 static int re_intr		(void *);
272 static void re_intr_msi		(void *);
273 static void re_tick		(void *);
274 static void re_int_task		(void *, int);
275 static void re_start		(struct ifnet *);
276 static void re_start_locked	(struct ifnet *);
277 static int re_ioctl		(struct ifnet *, u_long, caddr_t);
278 static void re_init		(void *);
279 static void re_init_locked	(struct rl_softc *);
280 static void re_stop		(struct rl_softc *);
281 static void re_watchdog		(struct rl_softc *);
282 static int re_suspend		(device_t);
283 static int re_resume		(device_t);
284 static int re_shutdown		(device_t);
285 static int re_ifmedia_upd	(struct ifnet *);
286 static void re_ifmedia_sts	(struct ifnet *, struct ifmediareq *);
287 
288 static void re_eeprom_putbyte	(struct rl_softc *, int);
289 static void re_eeprom_getword	(struct rl_softc *, int, u_int16_t *);
290 static void re_read_eeprom	(struct rl_softc *, caddr_t, int, int);
291 static int re_gmii_readreg	(device_t, int, int);
292 static int re_gmii_writereg	(device_t, int, int, int);
293 
294 static int re_miibus_readreg	(device_t, int, int);
295 static int re_miibus_writereg	(device_t, int, int, int);
296 static void re_miibus_statchg	(device_t);
297 
298 static void re_set_jumbo	(struct rl_softc *, int);
299 static void re_set_rxmode		(struct rl_softc *);
300 static void re_reset		(struct rl_softc *);
301 static void re_setwol		(struct rl_softc *);
302 static void re_clrwol		(struct rl_softc *);
303 static void re_set_linkspeed	(struct rl_softc *);
304 
305 #ifdef DEV_NETMAP	/* see ixgbe.c for details */
306 #include <dev/netmap/if_re_netmap.h>
307 #endif /* !DEV_NETMAP */
308 
309 #ifdef RE_DIAG
310 static int re_diag		(struct rl_softc *);
311 #endif
312 
313 static void re_add_sysctls	(struct rl_softc *);
314 static int re_sysctl_stats	(SYSCTL_HANDLER_ARGS);
315 static int sysctl_int_range	(SYSCTL_HANDLER_ARGS, int, int);
316 static int sysctl_hw_re_int_mod	(SYSCTL_HANDLER_ARGS);
317 
318 static device_method_t re_methods[] = {
319 	/* Device interface */
320 	DEVMETHOD(device_probe,		re_probe),
321 	DEVMETHOD(device_attach,	re_attach),
322 	DEVMETHOD(device_detach,	re_detach),
323 	DEVMETHOD(device_suspend,	re_suspend),
324 	DEVMETHOD(device_resume,	re_resume),
325 	DEVMETHOD(device_shutdown,	re_shutdown),
326 
327 	/* MII interface */
328 	DEVMETHOD(miibus_readreg,	re_miibus_readreg),
329 	DEVMETHOD(miibus_writereg,	re_miibus_writereg),
330 	DEVMETHOD(miibus_statchg,	re_miibus_statchg),
331 
332 	DEVMETHOD_END
333 };
334 
335 static driver_t re_driver = {
336 	"re",
337 	re_methods,
338 	sizeof(struct rl_softc)
339 };
340 
341 static devclass_t re_devclass;
342 
343 DRIVER_MODULE(re, pci, re_driver, re_devclass, 0, 0);
344 DRIVER_MODULE(miibus, re, miibus_driver, miibus_devclass, 0, 0);
345 
346 #define EE_SET(x)					\
347 	CSR_WRITE_1(sc, RL_EECMD,			\
348 		CSR_READ_1(sc, RL_EECMD) | x)
349 
350 #define EE_CLR(x)					\
351 	CSR_WRITE_1(sc, RL_EECMD,			\
352 		CSR_READ_1(sc, RL_EECMD) & ~x)
353 
354 /*
355  * Send a read command and address to the EEPROM, check for ACK.
356  */
357 static void
358 re_eeprom_putbyte(struct rl_softc *sc, int addr)
359 {
360 	int			d, i;
361 
362 	d = addr | (RL_9346_READ << sc->rl_eewidth);
363 
364 	/*
365 	 * Feed in each bit and strobe the clock.
366 	 */
367 
368 	for (i = 1 << (sc->rl_eewidth + 3); i; i >>= 1) {
369 		if (d & i) {
370 			EE_SET(RL_EE_DATAIN);
371 		} else {
372 			EE_CLR(RL_EE_DATAIN);
373 		}
374 		DELAY(100);
375 		EE_SET(RL_EE_CLK);
376 		DELAY(150);
377 		EE_CLR(RL_EE_CLK);
378 		DELAY(100);
379 	}
380 }
381 
382 /*
383  * Read a word of data stored in the EEPROM at address 'addr.'
384  */
385 static void
386 re_eeprom_getword(struct rl_softc *sc, int addr, u_int16_t *dest)
387 {
388 	int			i;
389 	u_int16_t		word = 0;
390 
391 	/*
392 	 * Send address of word we want to read.
393 	 */
394 	re_eeprom_putbyte(sc, addr);
395 
396 	/*
397 	 * Start reading bits from EEPROM.
398 	 */
399 	for (i = 0x8000; i; i >>= 1) {
400 		EE_SET(RL_EE_CLK);
401 		DELAY(100);
402 		if (CSR_READ_1(sc, RL_EECMD) & RL_EE_DATAOUT)
403 			word |= i;
404 		EE_CLR(RL_EE_CLK);
405 		DELAY(100);
406 	}
407 
408 	*dest = word;
409 }
410 
411 /*
412  * Read a sequence of words from the EEPROM.
413  */
414 static void
415 re_read_eeprom(struct rl_softc *sc, caddr_t dest, int off, int cnt)
416 {
417 	int			i;
418 	u_int16_t		word = 0, *ptr;
419 
420 	CSR_SETBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM);
421 
422         DELAY(100);
423 
424 	for (i = 0; i < cnt; i++) {
425 		CSR_SETBIT_1(sc, RL_EECMD, RL_EE_SEL);
426 		re_eeprom_getword(sc, off + i, &word);
427 		CSR_CLRBIT_1(sc, RL_EECMD, RL_EE_SEL);
428 		ptr = (u_int16_t *)(dest + (i * 2));
429                 *ptr = word;
430 	}
431 
432 	CSR_CLRBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM);
433 }
434 
435 static int
436 re_gmii_readreg(device_t dev, int phy, int reg)
437 {
438 	struct rl_softc		*sc;
439 	u_int32_t		rval;
440 	int			i;
441 
442 	sc = device_get_softc(dev);
443 
444 	/* Let the rgephy driver read the GMEDIASTAT register */
445 
446 	if (reg == RL_GMEDIASTAT) {
447 		rval = CSR_READ_1(sc, RL_GMEDIASTAT);
448 		return (rval);
449 	}
450 
451 	CSR_WRITE_4(sc, RL_PHYAR, reg << 16);
452 
453 	for (i = 0; i < RL_PHY_TIMEOUT; i++) {
454 		rval = CSR_READ_4(sc, RL_PHYAR);
455 		if (rval & RL_PHYAR_BUSY)
456 			break;
457 		DELAY(25);
458 	}
459 
460 	if (i == RL_PHY_TIMEOUT) {
461 		device_printf(sc->rl_dev, "PHY read failed\n");
462 		return (0);
463 	}
464 
465 	/*
466 	 * Controller requires a 20us delay to process next MDIO request.
467 	 */
468 	DELAY(20);
469 
470 	return (rval & RL_PHYAR_PHYDATA);
471 }
472 
473 static int
474 re_gmii_writereg(device_t dev, int phy, int reg, int data)
475 {
476 	struct rl_softc		*sc;
477 	u_int32_t		rval;
478 	int			i;
479 
480 	sc = device_get_softc(dev);
481 
482 	CSR_WRITE_4(sc, RL_PHYAR, (reg << 16) |
483 	    (data & RL_PHYAR_PHYDATA) | RL_PHYAR_BUSY);
484 
485 	for (i = 0; i < RL_PHY_TIMEOUT; i++) {
486 		rval = CSR_READ_4(sc, RL_PHYAR);
487 		if (!(rval & RL_PHYAR_BUSY))
488 			break;
489 		DELAY(25);
490 	}
491 
492 	if (i == RL_PHY_TIMEOUT) {
493 		device_printf(sc->rl_dev, "PHY write failed\n");
494 		return (0);
495 	}
496 
497 	/*
498 	 * Controller requires a 20us delay to process next MDIO request.
499 	 */
500 	DELAY(20);
501 
502 	return (0);
503 }
504 
505 static int
506 re_miibus_readreg(device_t dev, int phy, int reg)
507 {
508 	struct rl_softc		*sc;
509 	u_int16_t		rval = 0;
510 	u_int16_t		re8139_reg = 0;
511 
512 	sc = device_get_softc(dev);
513 
514 	if (sc->rl_type == RL_8169) {
515 		rval = re_gmii_readreg(dev, phy, reg);
516 		return (rval);
517 	}
518 
519 	switch (reg) {
520 	case MII_BMCR:
521 		re8139_reg = RL_BMCR;
522 		break;
523 	case MII_BMSR:
524 		re8139_reg = RL_BMSR;
525 		break;
526 	case MII_ANAR:
527 		re8139_reg = RL_ANAR;
528 		break;
529 	case MII_ANER:
530 		re8139_reg = RL_ANER;
531 		break;
532 	case MII_ANLPAR:
533 		re8139_reg = RL_LPAR;
534 		break;
535 	case MII_PHYIDR1:
536 	case MII_PHYIDR2:
537 		return (0);
538 	/*
539 	 * Allow the rlphy driver to read the media status
540 	 * register. If we have a link partner which does not
541 	 * support NWAY, this is the register which will tell
542 	 * us the results of parallel detection.
543 	 */
544 	case RL_MEDIASTAT:
545 		rval = CSR_READ_1(sc, RL_MEDIASTAT);
546 		return (rval);
547 	default:
548 		device_printf(sc->rl_dev, "bad phy register\n");
549 		return (0);
550 	}
551 	rval = CSR_READ_2(sc, re8139_reg);
552 	if (sc->rl_type == RL_8139CPLUS && re8139_reg == RL_BMCR) {
553 		/* 8139C+ has different bit layout. */
554 		rval &= ~(BMCR_LOOP | BMCR_ISO);
555 	}
556 	return (rval);
557 }
558 
559 static int
560 re_miibus_writereg(device_t dev, int phy, int reg, int data)
561 {
562 	struct rl_softc		*sc;
563 	u_int16_t		re8139_reg = 0;
564 	int			rval = 0;
565 
566 	sc = device_get_softc(dev);
567 
568 	if (sc->rl_type == RL_8169) {
569 		rval = re_gmii_writereg(dev, phy, reg, data);
570 		return (rval);
571 	}
572 
573 	switch (reg) {
574 	case MII_BMCR:
575 		re8139_reg = RL_BMCR;
576 		if (sc->rl_type == RL_8139CPLUS) {
577 			/* 8139C+ has different bit layout. */
578 			data &= ~(BMCR_LOOP | BMCR_ISO);
579 		}
580 		break;
581 	case MII_BMSR:
582 		re8139_reg = RL_BMSR;
583 		break;
584 	case MII_ANAR:
585 		re8139_reg = RL_ANAR;
586 		break;
587 	case MII_ANER:
588 		re8139_reg = RL_ANER;
589 		break;
590 	case MII_ANLPAR:
591 		re8139_reg = RL_LPAR;
592 		break;
593 	case MII_PHYIDR1:
594 	case MII_PHYIDR2:
595 		return (0);
596 		break;
597 	default:
598 		device_printf(sc->rl_dev, "bad phy register\n");
599 		return (0);
600 	}
601 	CSR_WRITE_2(sc, re8139_reg, data);
602 	return (0);
603 }
604 
605 static void
606 re_miibus_statchg(device_t dev)
607 {
608 	struct rl_softc		*sc;
609 	struct ifnet		*ifp;
610 	struct mii_data		*mii;
611 
612 	sc = device_get_softc(dev);
613 	mii = device_get_softc(sc->rl_miibus);
614 	ifp = sc->rl_ifp;
615 	if (mii == NULL || ifp == NULL ||
616 	    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
617 		return;
618 
619 	sc->rl_flags &= ~RL_FLAG_LINK;
620 	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
621 	    (IFM_ACTIVE | IFM_AVALID)) {
622 		switch (IFM_SUBTYPE(mii->mii_media_active)) {
623 		case IFM_10_T:
624 		case IFM_100_TX:
625 			sc->rl_flags |= RL_FLAG_LINK;
626 			break;
627 		case IFM_1000_T:
628 			if ((sc->rl_flags & RL_FLAG_FASTETHER) != 0)
629 				break;
630 			sc->rl_flags |= RL_FLAG_LINK;
631 			break;
632 		default:
633 			break;
634 		}
635 	}
636 	/*
637 	 * RealTek controllers does not provide any interface to
638 	 * Tx/Rx MACs for resolved speed, duplex and flow-control
639 	 * parameters.
640 	 */
641 }
642 
643 /*
644  * Set the RX configuration and 64-bit multicast hash filter.
645  */
646 static void
647 re_set_rxmode(struct rl_softc *sc)
648 {
649 	struct ifnet		*ifp;
650 	struct ifmultiaddr	*ifma;
651 	uint32_t		hashes[2] = { 0, 0 };
652 	uint32_t		h, rxfilt;
653 
654 	RL_LOCK_ASSERT(sc);
655 
656 	ifp = sc->rl_ifp;
657 
658 	rxfilt = RL_RXCFG_CONFIG | RL_RXCFG_RX_INDIV | RL_RXCFG_RX_BROAD;
659 	if ((sc->rl_flags & RL_FLAG_EARLYOFF) != 0)
660 		rxfilt |= RL_RXCFG_EARLYOFF;
661 	else if ((sc->rl_flags & RL_FLAG_EARLYOFFV2) != 0)
662 		rxfilt |= RL_RXCFG_EARLYOFFV2;
663 
664 	if (ifp->if_flags & (IFF_ALLMULTI | IFF_PROMISC)) {
665 		if (ifp->if_flags & IFF_PROMISC)
666 			rxfilt |= RL_RXCFG_RX_ALLPHYS;
667 		/*
668 		 * Unlike other hardwares, we have to explicitly set
669 		 * RL_RXCFG_RX_MULTI to receive multicast frames in
670 		 * promiscuous mode.
671 		 */
672 		rxfilt |= RL_RXCFG_RX_MULTI;
673 		hashes[0] = hashes[1] = 0xffffffff;
674 		goto done;
675 	}
676 
677 	if_maddr_rlock(ifp);
678 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
679 		if (ifma->ifma_addr->sa_family != AF_LINK)
680 			continue;
681 		h = ether_crc32_be(LLADDR((struct sockaddr_dl *)
682 		    ifma->ifma_addr), ETHER_ADDR_LEN) >> 26;
683 		if (h < 32)
684 			hashes[0] |= (1 << h);
685 		else
686 			hashes[1] |= (1 << (h - 32));
687 	}
688 	if_maddr_runlock(ifp);
689 
690 	if (hashes[0] != 0 || hashes[1] != 0) {
691 		/*
692 		 * For some unfathomable reason, RealTek decided to
693 		 * reverse the order of the multicast hash registers
694 		 * in the PCI Express parts.  This means we have to
695 		 * write the hash pattern in reverse order for those
696 		 * devices.
697 		 */
698 		if ((sc->rl_flags & RL_FLAG_PCIE) != 0) {
699 			h = bswap32(hashes[0]);
700 			hashes[0] = bswap32(hashes[1]);
701 			hashes[1] = h;
702 		}
703 		rxfilt |= RL_RXCFG_RX_MULTI;
704 	}
705 
706 done:
707 	CSR_WRITE_4(sc, RL_MAR0, hashes[0]);
708 	CSR_WRITE_4(sc, RL_MAR4, hashes[1]);
709 	CSR_WRITE_4(sc, RL_RXCFG, rxfilt);
710 }
711 
712 static void
713 re_reset(struct rl_softc *sc)
714 {
715 	int			i;
716 
717 	RL_LOCK_ASSERT(sc);
718 
719 	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RESET);
720 
721 	for (i = 0; i < RL_TIMEOUT; i++) {
722 		DELAY(10);
723 		if (!(CSR_READ_1(sc, RL_COMMAND) & RL_CMD_RESET))
724 			break;
725 	}
726 	if (i == RL_TIMEOUT)
727 		device_printf(sc->rl_dev, "reset never completed!\n");
728 
729 	if ((sc->rl_flags & RL_FLAG_MACRESET) != 0)
730 		CSR_WRITE_1(sc, 0x82, 1);
731 	if (sc->rl_hwrev->rl_rev == RL_HWREV_8169S)
732 		re_gmii_writereg(sc->rl_dev, 1, 0x0b, 0);
733 }
734 
735 #ifdef RE_DIAG
736 
737 /*
738  * The following routine is designed to test for a defect on some
739  * 32-bit 8169 cards. Some of these NICs have the REQ64# and ACK64#
740  * lines connected to the bus, however for a 32-bit only card, they
741  * should be pulled high. The result of this defect is that the
742  * NIC will not work right if you plug it into a 64-bit slot: DMA
743  * operations will be done with 64-bit transfers, which will fail
744  * because the 64-bit data lines aren't connected.
745  *
746  * There's no way to work around this (short of talking a soldering
747  * iron to the board), however we can detect it. The method we use
748  * here is to put the NIC into digital loopback mode, set the receiver
749  * to promiscuous mode, and then try to send a frame. We then compare
750  * the frame data we sent to what was received. If the data matches,
751  * then the NIC is working correctly, otherwise we know the user has
752  * a defective NIC which has been mistakenly plugged into a 64-bit PCI
753  * slot. In the latter case, there's no way the NIC can work correctly,
754  * so we print out a message on the console and abort the device attach.
755  */
756 
757 static int
758 re_diag(struct rl_softc *sc)
759 {
760 	struct ifnet		*ifp = sc->rl_ifp;
761 	struct mbuf		*m0;
762 	struct ether_header	*eh;
763 	struct rl_desc		*cur_rx;
764 	u_int16_t		status;
765 	u_int32_t		rxstat;
766 	int			total_len, i, error = 0, phyaddr;
767 	u_int8_t		dst[] = { 0x00, 'h', 'e', 'l', 'l', 'o' };
768 	u_int8_t		src[] = { 0x00, 'w', 'o', 'r', 'l', 'd' };
769 
770 	/* Allocate a single mbuf */
771 	MGETHDR(m0, M_NOWAIT, MT_DATA);
772 	if (m0 == NULL)
773 		return (ENOBUFS);
774 
775 	RL_LOCK(sc);
776 
777 	/*
778 	 * Initialize the NIC in test mode. This sets the chip up
779 	 * so that it can send and receive frames, but performs the
780 	 * following special functions:
781 	 * - Puts receiver in promiscuous mode
782 	 * - Enables digital loopback mode
783 	 * - Leaves interrupts turned off
784 	 */
785 
786 	ifp->if_flags |= IFF_PROMISC;
787 	sc->rl_testmode = 1;
788 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
789 	re_init_locked(sc);
790 	sc->rl_flags |= RL_FLAG_LINK;
791 	if (sc->rl_type == RL_8169)
792 		phyaddr = 1;
793 	else
794 		phyaddr = 0;
795 
796 	re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_RESET);
797 	for (i = 0; i < RL_TIMEOUT; i++) {
798 		status = re_miibus_readreg(sc->rl_dev, phyaddr, MII_BMCR);
799 		if (!(status & BMCR_RESET))
800 			break;
801 	}
802 
803 	re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_LOOP);
804 	CSR_WRITE_2(sc, RL_ISR, RL_INTRS);
805 
806 	DELAY(100000);
807 
808 	/* Put some data in the mbuf */
809 
810 	eh = mtod(m0, struct ether_header *);
811 	bcopy ((char *)&dst, eh->ether_dhost, ETHER_ADDR_LEN);
812 	bcopy ((char *)&src, eh->ether_shost, ETHER_ADDR_LEN);
813 	eh->ether_type = htons(ETHERTYPE_IP);
814 	m0->m_pkthdr.len = m0->m_len = ETHER_MIN_LEN - ETHER_CRC_LEN;
815 
816 	/*
817 	 * Queue the packet, start transmission.
818 	 * Note: IF_HANDOFF() ultimately calls re_start() for us.
819 	 */
820 
821 	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
822 	RL_UNLOCK(sc);
823 	/* XXX: re_diag must not be called when in ALTQ mode */
824 	IF_HANDOFF(&ifp->if_snd, m0, ifp);
825 	RL_LOCK(sc);
826 	m0 = NULL;
827 
828 	/* Wait for it to propagate through the chip */
829 
830 	DELAY(100000);
831 	for (i = 0; i < RL_TIMEOUT; i++) {
832 		status = CSR_READ_2(sc, RL_ISR);
833 		CSR_WRITE_2(sc, RL_ISR, status);
834 		if ((status & (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK)) ==
835 		    (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK))
836 			break;
837 		DELAY(10);
838 	}
839 
840 	if (i == RL_TIMEOUT) {
841 		device_printf(sc->rl_dev,
842 		    "diagnostic failed, failed to receive packet in"
843 		    " loopback mode\n");
844 		error = EIO;
845 		goto done;
846 	}
847 
848 	/*
849 	 * The packet should have been dumped into the first
850 	 * entry in the RX DMA ring. Grab it from there.
851 	 */
852 
853 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
854 	    sc->rl_ldata.rl_rx_list_map,
855 	    BUS_DMASYNC_POSTREAD);
856 	bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag,
857 	    sc->rl_ldata.rl_rx_desc[0].rx_dmamap,
858 	    BUS_DMASYNC_POSTREAD);
859 	bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag,
860 	    sc->rl_ldata.rl_rx_desc[0].rx_dmamap);
861 
862 	m0 = sc->rl_ldata.rl_rx_desc[0].rx_m;
863 	sc->rl_ldata.rl_rx_desc[0].rx_m = NULL;
864 	eh = mtod(m0, struct ether_header *);
865 
866 	cur_rx = &sc->rl_ldata.rl_rx_list[0];
867 	total_len = RL_RXBYTES(cur_rx);
868 	rxstat = le32toh(cur_rx->rl_cmdstat);
869 
870 	if (total_len != ETHER_MIN_LEN) {
871 		device_printf(sc->rl_dev,
872 		    "diagnostic failed, received short packet\n");
873 		error = EIO;
874 		goto done;
875 	}
876 
877 	/* Test that the received packet data matches what we sent. */
878 
879 	if (bcmp((char *)&eh->ether_dhost, (char *)&dst, ETHER_ADDR_LEN) ||
880 	    bcmp((char *)&eh->ether_shost, (char *)&src, ETHER_ADDR_LEN) ||
881 	    ntohs(eh->ether_type) != ETHERTYPE_IP) {
882 		device_printf(sc->rl_dev, "WARNING, DMA FAILURE!\n");
883 		device_printf(sc->rl_dev, "expected TX data: %6D/%6D/0x%x\n",
884 		    dst, ":", src, ":", ETHERTYPE_IP);
885 		device_printf(sc->rl_dev, "received RX data: %6D/%6D/0x%x\n",
886 		    eh->ether_dhost, ":", eh->ether_shost, ":",
887 		    ntohs(eh->ether_type));
888 		device_printf(sc->rl_dev, "You may have a defective 32-bit "
889 		    "NIC plugged into a 64-bit PCI slot.\n");
890 		device_printf(sc->rl_dev, "Please re-install the NIC in a "
891 		    "32-bit slot for proper operation.\n");
892 		device_printf(sc->rl_dev, "Read the re(4) man page for more "
893 		    "details.\n");
894 		error = EIO;
895 	}
896 
897 done:
898 	/* Turn interface off, release resources */
899 
900 	sc->rl_testmode = 0;
901 	sc->rl_flags &= ~RL_FLAG_LINK;
902 	ifp->if_flags &= ~IFF_PROMISC;
903 	re_stop(sc);
904 	if (m0 != NULL)
905 		m_freem(m0);
906 
907 	RL_UNLOCK(sc);
908 
909 	return (error);
910 }
911 
912 #endif
913 
914 /*
915  * Probe for a RealTek 8139C+/8169/8110 chip. Check the PCI vendor and device
916  * IDs against our list and return a device name if we find a match.
917  */
918 static int
919 re_probe(device_t dev)
920 {
921 	const struct rl_type	*t;
922 	uint16_t		devid, vendor;
923 	uint16_t		revid, sdevid;
924 	int			i;
925 
926 	vendor = pci_get_vendor(dev);
927 	devid = pci_get_device(dev);
928 	revid = pci_get_revid(dev);
929 	sdevid = pci_get_subdevice(dev);
930 
931 	if (vendor == LINKSYS_VENDORID && devid == LINKSYS_DEVICEID_EG1032) {
932 		if (sdevid != LINKSYS_SUBDEVICE_EG1032_REV3) {
933 			/*
934 			 * Only attach to rev. 3 of the Linksys EG1032 adapter.
935 			 * Rev. 2 is supported by sk(4).
936 			 */
937 			return (ENXIO);
938 		}
939 	}
940 
941 	if (vendor == RT_VENDORID && devid == RT_DEVICEID_8139) {
942 		if (revid != 0x20) {
943 			/* 8139, let rl(4) take care of this device. */
944 			return (ENXIO);
945 		}
946 	}
947 
948 	t = re_devs;
949 	for (i = 0; i < sizeof(re_devs) / sizeof(re_devs[0]); i++, t++) {
950 		if (vendor == t->rl_vid && devid == t->rl_did) {
951 			device_set_desc(dev, t->rl_name);
952 			return (BUS_PROBE_DEFAULT);
953 		}
954 	}
955 
956 	return (ENXIO);
957 }
958 
959 /*
960  * Map a single buffer address.
961  */
962 
963 static void
964 re_dma_map_addr(void *arg, bus_dma_segment_t *segs, int nseg, int error)
965 {
966 	bus_addr_t		*addr;
967 
968 	if (error)
969 		return;
970 
971 	KASSERT(nseg == 1, ("too many DMA segments, %d should be 1", nseg));
972 	addr = arg;
973 	*addr = segs->ds_addr;
974 }
975 
976 static int
977 re_allocmem(device_t dev, struct rl_softc *sc)
978 {
979 	bus_addr_t		lowaddr;
980 	bus_size_t		rx_list_size, tx_list_size;
981 	int			error;
982 	int			i;
983 
984 	rx_list_size = sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc);
985 	tx_list_size = sc->rl_ldata.rl_tx_desc_cnt * sizeof(struct rl_desc);
986 
987 	/*
988 	 * Allocate the parent bus DMA tag appropriate for PCI.
989 	 * In order to use DAC, RL_CPLUSCMD_PCI_DAC bit of RL_CPLUS_CMD
990 	 * register should be set. However some RealTek chips are known
991 	 * to be buggy on DAC handling, therefore disable DAC by limiting
992 	 * DMA address space to 32bit. PCIe variants of RealTek chips
993 	 * may not have the limitation.
994 	 */
995 	lowaddr = BUS_SPACE_MAXADDR;
996 	if ((sc->rl_flags & RL_FLAG_PCIE) == 0)
997 		lowaddr = BUS_SPACE_MAXADDR_32BIT;
998 	error = bus_dma_tag_create(bus_get_dma_tag(dev), 1, 0,
999 	    lowaddr, BUS_SPACE_MAXADDR, NULL, NULL,
1000 	    BUS_SPACE_MAXSIZE_32BIT, 0, BUS_SPACE_MAXSIZE_32BIT, 0,
1001 	    NULL, NULL, &sc->rl_parent_tag);
1002 	if (error) {
1003 		device_printf(dev, "could not allocate parent DMA tag\n");
1004 		return (error);
1005 	}
1006 
1007 	/*
1008 	 * Allocate map for TX mbufs.
1009 	 */
1010 	error = bus_dma_tag_create(sc->rl_parent_tag, 1, 0,
1011 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL,
1012 	    NULL, MCLBYTES * RL_NTXSEGS, RL_NTXSEGS, 4096, 0,
1013 	    NULL, NULL, &sc->rl_ldata.rl_tx_mtag);
1014 	if (error) {
1015 		device_printf(dev, "could not allocate TX DMA tag\n");
1016 		return (error);
1017 	}
1018 
1019 	/*
1020 	 * Allocate map for RX mbufs.
1021 	 */
1022 
1023 	if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
1024 		error = bus_dma_tag_create(sc->rl_parent_tag, sizeof(uint64_t),
1025 		    0, BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
1026 		    MJUM9BYTES, 1, MJUM9BYTES, 0, NULL, NULL,
1027 		    &sc->rl_ldata.rl_jrx_mtag);
1028 		if (error) {
1029 			device_printf(dev,
1030 			    "could not allocate jumbo RX DMA tag\n");
1031 			return (error);
1032 		}
1033 	}
1034 	error = bus_dma_tag_create(sc->rl_parent_tag, sizeof(uint64_t), 0,
1035 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
1036 	    MCLBYTES, 1, MCLBYTES, 0, NULL, NULL, &sc->rl_ldata.rl_rx_mtag);
1037 	if (error) {
1038 		device_printf(dev, "could not allocate RX DMA tag\n");
1039 		return (error);
1040 	}
1041 
1042 	/*
1043 	 * Allocate map for TX descriptor list.
1044 	 */
1045 	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
1046 	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1047 	    NULL, tx_list_size, 1, tx_list_size, 0,
1048 	    NULL, NULL, &sc->rl_ldata.rl_tx_list_tag);
1049 	if (error) {
1050 		device_printf(dev, "could not allocate TX DMA ring tag\n");
1051 		return (error);
1052 	}
1053 
1054 	/* Allocate DMA'able memory for the TX ring */
1055 
1056 	error = bus_dmamem_alloc(sc->rl_ldata.rl_tx_list_tag,
1057 	    (void **)&sc->rl_ldata.rl_tx_list,
1058 	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1059 	    &sc->rl_ldata.rl_tx_list_map);
1060 	if (error) {
1061 		device_printf(dev, "could not allocate TX DMA ring\n");
1062 		return (error);
1063 	}
1064 
1065 	/* Load the map for the TX ring. */
1066 
1067 	sc->rl_ldata.rl_tx_list_addr = 0;
1068 	error = bus_dmamap_load(sc->rl_ldata.rl_tx_list_tag,
1069 	     sc->rl_ldata.rl_tx_list_map, sc->rl_ldata.rl_tx_list,
1070 	     tx_list_size, re_dma_map_addr,
1071 	     &sc->rl_ldata.rl_tx_list_addr, BUS_DMA_NOWAIT);
1072 	if (error != 0 || sc->rl_ldata.rl_tx_list_addr == 0) {
1073 		device_printf(dev, "could not load TX DMA ring\n");
1074 		return (ENOMEM);
1075 	}
1076 
1077 	/* Create DMA maps for TX buffers */
1078 
1079 	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
1080 		error = bus_dmamap_create(sc->rl_ldata.rl_tx_mtag, 0,
1081 		    &sc->rl_ldata.rl_tx_desc[i].tx_dmamap);
1082 		if (error) {
1083 			device_printf(dev, "could not create DMA map for TX\n");
1084 			return (error);
1085 		}
1086 	}
1087 
1088 	/*
1089 	 * Allocate map for RX descriptor list.
1090 	 */
1091 	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
1092 	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1093 	    NULL, rx_list_size, 1, rx_list_size, 0,
1094 	    NULL, NULL, &sc->rl_ldata.rl_rx_list_tag);
1095 	if (error) {
1096 		device_printf(dev, "could not create RX DMA ring tag\n");
1097 		return (error);
1098 	}
1099 
1100 	/* Allocate DMA'able memory for the RX ring */
1101 
1102 	error = bus_dmamem_alloc(sc->rl_ldata.rl_rx_list_tag,
1103 	    (void **)&sc->rl_ldata.rl_rx_list,
1104 	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1105 	    &sc->rl_ldata.rl_rx_list_map);
1106 	if (error) {
1107 		device_printf(dev, "could not allocate RX DMA ring\n");
1108 		return (error);
1109 	}
1110 
1111 	/* Load the map for the RX ring. */
1112 
1113 	sc->rl_ldata.rl_rx_list_addr = 0;
1114 	error = bus_dmamap_load(sc->rl_ldata.rl_rx_list_tag,
1115 	     sc->rl_ldata.rl_rx_list_map, sc->rl_ldata.rl_rx_list,
1116 	     rx_list_size, re_dma_map_addr,
1117 	     &sc->rl_ldata.rl_rx_list_addr, BUS_DMA_NOWAIT);
1118 	if (error != 0 || sc->rl_ldata.rl_rx_list_addr == 0) {
1119 		device_printf(dev, "could not load RX DMA ring\n");
1120 		return (ENOMEM);
1121 	}
1122 
1123 	/* Create DMA maps for RX buffers */
1124 
1125 	if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
1126 		error = bus_dmamap_create(sc->rl_ldata.rl_jrx_mtag, 0,
1127 		    &sc->rl_ldata.rl_jrx_sparemap);
1128 		if (error) {
1129 			device_printf(dev,
1130 			    "could not create spare DMA map for jumbo RX\n");
1131 			return (error);
1132 		}
1133 		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1134 			error = bus_dmamap_create(sc->rl_ldata.rl_jrx_mtag, 0,
1135 			    &sc->rl_ldata.rl_jrx_desc[i].rx_dmamap);
1136 			if (error) {
1137 				device_printf(dev,
1138 				    "could not create DMA map for jumbo RX\n");
1139 				return (error);
1140 			}
1141 		}
1142 	}
1143 	error = bus_dmamap_create(sc->rl_ldata.rl_rx_mtag, 0,
1144 	    &sc->rl_ldata.rl_rx_sparemap);
1145 	if (error) {
1146 		device_printf(dev, "could not create spare DMA map for RX\n");
1147 		return (error);
1148 	}
1149 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1150 		error = bus_dmamap_create(sc->rl_ldata.rl_rx_mtag, 0,
1151 		    &sc->rl_ldata.rl_rx_desc[i].rx_dmamap);
1152 		if (error) {
1153 			device_printf(dev, "could not create DMA map for RX\n");
1154 			return (error);
1155 		}
1156 	}
1157 
1158 	/* Create DMA map for statistics. */
1159 	error = bus_dma_tag_create(sc->rl_parent_tag, RL_DUMP_ALIGN, 0,
1160 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
1161 	    sizeof(struct rl_stats), 1, sizeof(struct rl_stats), 0, NULL, NULL,
1162 	    &sc->rl_ldata.rl_stag);
1163 	if (error) {
1164 		device_printf(dev, "could not create statistics DMA tag\n");
1165 		return (error);
1166 	}
1167 	/* Allocate DMA'able memory for statistics. */
1168 	error = bus_dmamem_alloc(sc->rl_ldata.rl_stag,
1169 	    (void **)&sc->rl_ldata.rl_stats,
1170 	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1171 	    &sc->rl_ldata.rl_smap);
1172 	if (error) {
1173 		device_printf(dev,
1174 		    "could not allocate statistics DMA memory\n");
1175 		return (error);
1176 	}
1177 	/* Load the map for statistics. */
1178 	sc->rl_ldata.rl_stats_addr = 0;
1179 	error = bus_dmamap_load(sc->rl_ldata.rl_stag, sc->rl_ldata.rl_smap,
1180 	    sc->rl_ldata.rl_stats, sizeof(struct rl_stats), re_dma_map_addr,
1181 	     &sc->rl_ldata.rl_stats_addr, BUS_DMA_NOWAIT);
1182 	if (error != 0 || sc->rl_ldata.rl_stats_addr == 0) {
1183 		device_printf(dev, "could not load statistics DMA memory\n");
1184 		return (ENOMEM);
1185 	}
1186 
1187 	return (0);
1188 }
1189 
1190 /*
1191  * Attach the interface. Allocate softc structures, do ifmedia
1192  * setup and ethernet/BPF attach.
1193  */
1194 static int
1195 re_attach(device_t dev)
1196 {
1197 	u_char			eaddr[ETHER_ADDR_LEN];
1198 	u_int16_t		as[ETHER_ADDR_LEN / 2];
1199 	struct rl_softc		*sc;
1200 	struct ifnet		*ifp;
1201 	const struct rl_hwrev	*hw_rev;
1202 	u_int32_t		cap, ctl;
1203 	int			hwrev;
1204 	u_int16_t		devid, re_did = 0;
1205 	int			error = 0, i, phy, rid;
1206 	int			msic, msixc, reg;
1207 	uint8_t			cfg;
1208 
1209 	sc = device_get_softc(dev);
1210 	sc->rl_dev = dev;
1211 
1212 	mtx_init(&sc->rl_mtx, device_get_nameunit(dev), MTX_NETWORK_LOCK,
1213 	    MTX_DEF);
1214 	callout_init_mtx(&sc->rl_stat_callout, &sc->rl_mtx, 0);
1215 
1216 	/*
1217 	 * Map control/status registers.
1218 	 */
1219 	pci_enable_busmaster(dev);
1220 
1221 	devid = pci_get_device(dev);
1222 	/*
1223 	 * Prefer memory space register mapping over IO space.
1224 	 * Because RTL8169SC does not seem to work when memory mapping
1225 	 * is used always activate io mapping.
1226 	 */
1227 	if (devid == RT_DEVICEID_8169SC)
1228 		prefer_iomap = 1;
1229 	if (prefer_iomap == 0) {
1230 		sc->rl_res_id = PCIR_BAR(1);
1231 		sc->rl_res_type = SYS_RES_MEMORY;
1232 		/* RTL8168/8101E seems to use different BARs. */
1233 		if (devid == RT_DEVICEID_8168 || devid == RT_DEVICEID_8101E)
1234 			sc->rl_res_id = PCIR_BAR(2);
1235 	} else {
1236 		sc->rl_res_id = PCIR_BAR(0);
1237 		sc->rl_res_type = SYS_RES_IOPORT;
1238 	}
1239 	sc->rl_res = bus_alloc_resource_any(dev, sc->rl_res_type,
1240 	    &sc->rl_res_id, RF_ACTIVE);
1241 	if (sc->rl_res == NULL && prefer_iomap == 0) {
1242 		sc->rl_res_id = PCIR_BAR(0);
1243 		sc->rl_res_type = SYS_RES_IOPORT;
1244 		sc->rl_res = bus_alloc_resource_any(dev, sc->rl_res_type,
1245 		    &sc->rl_res_id, RF_ACTIVE);
1246 	}
1247 	if (sc->rl_res == NULL) {
1248 		device_printf(dev, "couldn't map ports/memory\n");
1249 		error = ENXIO;
1250 		goto fail;
1251 	}
1252 
1253 	sc->rl_btag = rman_get_bustag(sc->rl_res);
1254 	sc->rl_bhandle = rman_get_bushandle(sc->rl_res);
1255 
1256 	msic = pci_msi_count(dev);
1257 	msixc = pci_msix_count(dev);
1258 	if (pci_find_cap(dev, PCIY_EXPRESS, &reg) == 0) {
1259 		sc->rl_flags |= RL_FLAG_PCIE;
1260 		sc->rl_expcap = reg;
1261 	}
1262 	if (bootverbose) {
1263 		device_printf(dev, "MSI count : %d\n", msic);
1264 		device_printf(dev, "MSI-X count : %d\n", msixc);
1265 	}
1266 	if (msix_disable > 0)
1267 		msixc = 0;
1268 	if (msi_disable > 0)
1269 		msic = 0;
1270 	/* Prefer MSI-X to MSI. */
1271 	if (msixc > 0) {
1272 		msixc = RL_MSI_MESSAGES;
1273 		rid = PCIR_BAR(4);
1274 		sc->rl_res_pba = bus_alloc_resource_any(dev, SYS_RES_MEMORY,
1275 		    &rid, RF_ACTIVE);
1276 		if (sc->rl_res_pba == NULL) {
1277 			device_printf(sc->rl_dev,
1278 			    "could not allocate MSI-X PBA resource\n");
1279 		}
1280 		if (sc->rl_res_pba != NULL &&
1281 		    pci_alloc_msix(dev, &msixc) == 0) {
1282 			if (msixc == RL_MSI_MESSAGES) {
1283 				device_printf(dev, "Using %d MSI-X message\n",
1284 				    msixc);
1285 				sc->rl_flags |= RL_FLAG_MSIX;
1286 			} else
1287 				pci_release_msi(dev);
1288 		}
1289 		if ((sc->rl_flags & RL_FLAG_MSIX) == 0) {
1290 			if (sc->rl_res_pba != NULL)
1291 				bus_release_resource(dev, SYS_RES_MEMORY, rid,
1292 				    sc->rl_res_pba);
1293 			sc->rl_res_pba = NULL;
1294 			msixc = 0;
1295 		}
1296 	}
1297 	/* Prefer MSI to INTx. */
1298 	if (msixc == 0 && msic > 0) {
1299 		msic = RL_MSI_MESSAGES;
1300 		if (pci_alloc_msi(dev, &msic) == 0) {
1301 			if (msic == RL_MSI_MESSAGES) {
1302 				device_printf(dev, "Using %d MSI message\n",
1303 				    msic);
1304 				sc->rl_flags |= RL_FLAG_MSI;
1305 				/* Explicitly set MSI enable bit. */
1306 				CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1307 				cfg = CSR_READ_1(sc, RL_CFG2);
1308 				cfg |= RL_CFG2_MSI;
1309 				CSR_WRITE_1(sc, RL_CFG2, cfg);
1310 				CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1311 			} else
1312 				pci_release_msi(dev);
1313 		}
1314 		if ((sc->rl_flags & RL_FLAG_MSI) == 0)
1315 			msic = 0;
1316 	}
1317 
1318 	/* Allocate interrupt */
1319 	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) == 0) {
1320 		rid = 0;
1321 		sc->rl_irq[0] = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
1322 		    RF_SHAREABLE | RF_ACTIVE);
1323 		if (sc->rl_irq[0] == NULL) {
1324 			device_printf(dev, "couldn't allocate IRQ resources\n");
1325 			error = ENXIO;
1326 			goto fail;
1327 		}
1328 	} else {
1329 		for (i = 0, rid = 1; i < RL_MSI_MESSAGES; i++, rid++) {
1330 			sc->rl_irq[i] = bus_alloc_resource_any(dev,
1331 			    SYS_RES_IRQ, &rid, RF_ACTIVE);
1332 			if (sc->rl_irq[i] == NULL) {
1333 				device_printf(dev,
1334 				    "couldn't allocate IRQ resources for "
1335 				    "message %d\n", rid);
1336 				error = ENXIO;
1337 				goto fail;
1338 			}
1339 		}
1340 	}
1341 
1342 	if ((sc->rl_flags & RL_FLAG_MSI) == 0) {
1343 		CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1344 		cfg = CSR_READ_1(sc, RL_CFG2);
1345 		if ((cfg & RL_CFG2_MSI) != 0) {
1346 			device_printf(dev, "turning off MSI enable bit.\n");
1347 			cfg &= ~RL_CFG2_MSI;
1348 			CSR_WRITE_1(sc, RL_CFG2, cfg);
1349 		}
1350 		CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1351 	}
1352 
1353 	/* Disable ASPM L0S/L1. */
1354 	if (sc->rl_expcap != 0) {
1355 		cap = pci_read_config(dev, sc->rl_expcap +
1356 		    PCIER_LINK_CAP, 2);
1357 		if ((cap & PCIEM_LINK_CAP_ASPM) != 0) {
1358 			ctl = pci_read_config(dev, sc->rl_expcap +
1359 			    PCIER_LINK_CTL, 2);
1360 			if ((ctl & PCIEM_LINK_CTL_ASPMC) != 0) {
1361 				ctl &= ~PCIEM_LINK_CTL_ASPMC;
1362 				pci_write_config(dev, sc->rl_expcap +
1363 				    PCIER_LINK_CTL, ctl, 2);
1364 				device_printf(dev, "ASPM disabled\n");
1365 			}
1366 		} else
1367 			device_printf(dev, "no ASPM capability\n");
1368 	}
1369 
1370 	hw_rev = re_hwrevs;
1371 	hwrev = CSR_READ_4(sc, RL_TXCFG);
1372 	switch (hwrev & 0x70000000) {
1373 	case 0x00000000:
1374 	case 0x10000000:
1375 		device_printf(dev, "Chip rev. 0x%08x\n", hwrev & 0xfc800000);
1376 		hwrev &= (RL_TXCFG_HWREV | 0x80000000);
1377 		break;
1378 	default:
1379 		device_printf(dev, "Chip rev. 0x%08x\n", hwrev & 0x7c800000);
1380 		sc->rl_macrev = hwrev & 0x00700000;
1381 		hwrev &= RL_TXCFG_HWREV;
1382 		break;
1383 	}
1384 	device_printf(dev, "MAC rev. 0x%08x\n", sc->rl_macrev);
1385 	while (hw_rev->rl_desc != NULL) {
1386 		if (hw_rev->rl_rev == hwrev) {
1387 			sc->rl_type = hw_rev->rl_type;
1388 			sc->rl_hwrev = hw_rev;
1389 			break;
1390 		}
1391 		hw_rev++;
1392 	}
1393 	if (hw_rev->rl_desc == NULL) {
1394 		device_printf(dev, "Unknown H/W revision: 0x%08x\n", hwrev);
1395 		error = ENXIO;
1396 		goto fail;
1397 	}
1398 
1399 	switch (hw_rev->rl_rev) {
1400 	case RL_HWREV_8139CPLUS:
1401 		sc->rl_flags |= RL_FLAG_FASTETHER | RL_FLAG_AUTOPAD;
1402 		break;
1403 	case RL_HWREV_8100E:
1404 	case RL_HWREV_8101E:
1405 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_FASTETHER;
1406 		break;
1407 	case RL_HWREV_8102E:
1408 	case RL_HWREV_8102EL:
1409 	case RL_HWREV_8102EL_SPIN1:
1410 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR | RL_FLAG_DESCV2 |
1411 		    RL_FLAG_MACSTAT | RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP |
1412 		    RL_FLAG_AUTOPAD;
1413 		break;
1414 	case RL_HWREV_8103E:
1415 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR | RL_FLAG_DESCV2 |
1416 		    RL_FLAG_MACSTAT | RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP |
1417 		    RL_FLAG_AUTOPAD | RL_FLAG_MACSLEEP;
1418 		break;
1419 	case RL_HWREV_8401E:
1420 	case RL_HWREV_8105E:
1421 	case RL_HWREV_8105E_SPIN1:
1422 	case RL_HWREV_8106E:
1423 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
1424 		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
1425 		    RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD;
1426 		break;
1427 	case RL_HWREV_8402:
1428 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
1429 		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
1430 		    RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD |
1431 		    RL_FLAG_CMDSTOP_WAIT_TXQ;
1432 		break;
1433 	case RL_HWREV_8168B_SPIN1:
1434 	case RL_HWREV_8168B_SPIN2:
1435 		sc->rl_flags |= RL_FLAG_WOLRXENB;
1436 		/* FALLTHROUGH */
1437 	case RL_HWREV_8168B_SPIN3:
1438 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_MACSTAT;
1439 		break;
1440 	case RL_HWREV_8168C_SPIN2:
1441 		sc->rl_flags |= RL_FLAG_MACSLEEP;
1442 		/* FALLTHROUGH */
1443 	case RL_HWREV_8168C:
1444 		if (sc->rl_macrev == 0x00200000)
1445 			sc->rl_flags |= RL_FLAG_MACSLEEP;
1446 		/* FALLTHROUGH */
1447 	case RL_HWREV_8168CP:
1448 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
1449 		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP |
1450 		    RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 | RL_FLAG_WOL_MANLINK;
1451 		break;
1452 	case RL_HWREV_8168D:
1453 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
1454 		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
1455 		    RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 |
1456 		    RL_FLAG_WOL_MANLINK;
1457 		break;
1458 	case RL_HWREV_8168DP:
1459 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
1460 		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_AUTOPAD |
1461 		    RL_FLAG_JUMBOV2 | RL_FLAG_WAIT_TXPOLL | RL_FLAG_WOL_MANLINK;
1462 		break;
1463 	case RL_HWREV_8168E:
1464 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
1465 		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
1466 		    RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 |
1467 		    RL_FLAG_WOL_MANLINK;
1468 		break;
1469 	case RL_HWREV_8168E_VL:
1470 	case RL_HWREV_8168F:
1471 		sc->rl_flags |= RL_FLAG_EARLYOFF;
1472 		/* FALLTHROUGH */
1473 	case RL_HWREV_8411:
1474 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
1475 		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP |
1476 		    RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 |
1477 		    RL_FLAG_CMDSTOP_WAIT_TXQ | RL_FLAG_WOL_MANLINK;
1478 		break;
1479 	case RL_HWREV_8168EP:
1480 	case RL_HWREV_8168G:
1481 	case RL_HWREV_8411B:
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 		    RL_FLAG_EARLYOFFV2 | RL_FLAG_RXDV_GATED;
1487 		break;
1488 	case RL_HWREV_8168GU:
1489 		if (pci_get_device(dev) == RT_DEVICEID_8101E) {
1490 			/* RTL8106EUS */
1491 			sc->rl_flags |= RL_FLAG_FASTETHER;
1492 		} else
1493 			sc->rl_flags |= RL_FLAG_JUMBOV2 | RL_FLAG_WOL_MANLINK;
1494 
1495 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
1496 		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP |
1497 		    RL_FLAG_AUTOPAD | RL_FLAG_CMDSTOP_WAIT_TXQ |
1498 		    RL_FLAG_EARLYOFFV2 | RL_FLAG_RXDV_GATED;
1499 		break;
1500 	case RL_HWREV_8169_8110SB:
1501 	case RL_HWREV_8169_8110SBL:
1502 	case RL_HWREV_8169_8110SC:
1503 	case RL_HWREV_8169_8110SCE:
1504 		sc->rl_flags |= RL_FLAG_PHYWAKE;
1505 		/* FALLTHROUGH */
1506 	case RL_HWREV_8169:
1507 	case RL_HWREV_8169S:
1508 	case RL_HWREV_8110S:
1509 		sc->rl_flags |= RL_FLAG_MACRESET;
1510 		break;
1511 	default:
1512 		break;
1513 	}
1514 
1515 	if (sc->rl_hwrev->rl_rev == RL_HWREV_8139CPLUS) {
1516 		sc->rl_cfg0 = RL_8139_CFG0;
1517 		sc->rl_cfg1 = RL_8139_CFG1;
1518 		sc->rl_cfg2 = 0;
1519 		sc->rl_cfg3 = RL_8139_CFG3;
1520 		sc->rl_cfg4 = RL_8139_CFG4;
1521 		sc->rl_cfg5 = RL_8139_CFG5;
1522 	} else {
1523 		sc->rl_cfg0 = RL_CFG0;
1524 		sc->rl_cfg1 = RL_CFG1;
1525 		sc->rl_cfg2 = RL_CFG2;
1526 		sc->rl_cfg3 = RL_CFG3;
1527 		sc->rl_cfg4 = RL_CFG4;
1528 		sc->rl_cfg5 = RL_CFG5;
1529 	}
1530 
1531 	/* Reset the adapter. */
1532 	RL_LOCK(sc);
1533 	re_reset(sc);
1534 	RL_UNLOCK(sc);
1535 
1536 	/* Enable PME. */
1537 	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1538 	cfg = CSR_READ_1(sc, sc->rl_cfg1);
1539 	cfg |= RL_CFG1_PME;
1540 	CSR_WRITE_1(sc, sc->rl_cfg1, cfg);
1541 	cfg = CSR_READ_1(sc, sc->rl_cfg5);
1542 	cfg &= RL_CFG5_PME_STS;
1543 	CSR_WRITE_1(sc, sc->rl_cfg5, cfg);
1544 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1545 
1546 	if ((sc->rl_flags & RL_FLAG_PAR) != 0) {
1547 		/*
1548 		 * XXX Should have a better way to extract station
1549 		 * address from EEPROM.
1550 		 */
1551 		for (i = 0; i < ETHER_ADDR_LEN; i++)
1552 			eaddr[i] = CSR_READ_1(sc, RL_IDR0 + i);
1553 	} else {
1554 		sc->rl_eewidth = RL_9356_ADDR_LEN;
1555 		re_read_eeprom(sc, (caddr_t)&re_did, 0, 1);
1556 		if (re_did != 0x8129)
1557 			sc->rl_eewidth = RL_9346_ADDR_LEN;
1558 
1559 		/*
1560 		 * Get station address from the EEPROM.
1561 		 */
1562 		re_read_eeprom(sc, (caddr_t)as, RL_EE_EADDR, 3);
1563 		for (i = 0; i < ETHER_ADDR_LEN / 2; i++)
1564 			as[i] = le16toh(as[i]);
1565 		bcopy(as, eaddr, ETHER_ADDR_LEN);
1566 	}
1567 
1568 	if (sc->rl_type == RL_8169) {
1569 		/* Set RX length mask and number of descriptors. */
1570 		sc->rl_rxlenmask = RL_RDESC_STAT_GFRAGLEN;
1571 		sc->rl_txstart = RL_GTXSTART;
1572 		sc->rl_ldata.rl_tx_desc_cnt = RL_8169_TX_DESC_CNT;
1573 		sc->rl_ldata.rl_rx_desc_cnt = RL_8169_RX_DESC_CNT;
1574 	} else {
1575 		/* Set RX length mask and number of descriptors. */
1576 		sc->rl_rxlenmask = RL_RDESC_STAT_FRAGLEN;
1577 		sc->rl_txstart = RL_TXSTART;
1578 		sc->rl_ldata.rl_tx_desc_cnt = RL_8139_TX_DESC_CNT;
1579 		sc->rl_ldata.rl_rx_desc_cnt = RL_8139_RX_DESC_CNT;
1580 	}
1581 
1582 	error = re_allocmem(dev, sc);
1583 	if (error)
1584 		goto fail;
1585 	re_add_sysctls(sc);
1586 
1587 	ifp = sc->rl_ifp = if_alloc(IFT_ETHER);
1588 	if (ifp == NULL) {
1589 		device_printf(dev, "can not if_alloc()\n");
1590 		error = ENOSPC;
1591 		goto fail;
1592 	}
1593 
1594 	/* Take controller out of deep sleep mode. */
1595 	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
1596 		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
1597 			CSR_WRITE_1(sc, RL_GPIO,
1598 			    CSR_READ_1(sc, RL_GPIO) | 0x01);
1599 		else
1600 			CSR_WRITE_1(sc, RL_GPIO,
1601 			    CSR_READ_1(sc, RL_GPIO) & ~0x01);
1602 	}
1603 
1604 	/* Take PHY out of power down mode. */
1605 	if ((sc->rl_flags & RL_FLAG_PHYWAKE_PM) != 0) {
1606 		CSR_WRITE_1(sc, RL_PMCH, CSR_READ_1(sc, RL_PMCH) | 0x80);
1607 		if (hw_rev->rl_rev == RL_HWREV_8401E)
1608 			CSR_WRITE_1(sc, 0xD1, CSR_READ_1(sc, 0xD1) & ~0x08);
1609 	}
1610 	if ((sc->rl_flags & RL_FLAG_PHYWAKE) != 0) {
1611 		re_gmii_writereg(dev, 1, 0x1f, 0);
1612 		re_gmii_writereg(dev, 1, 0x0e, 0);
1613 	}
1614 
1615 	ifp->if_softc = sc;
1616 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1617 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1618 	ifp->if_ioctl = re_ioctl;
1619 	ifp->if_start = re_start;
1620 	/*
1621 	 * RTL8168/8111C generates wrong IP checksummed frame if the
1622 	 * packet has IP options so disable TX checksum offloading.
1623 	 */
1624 	if (sc->rl_hwrev->rl_rev == RL_HWREV_8168C ||
1625 	    sc->rl_hwrev->rl_rev == RL_HWREV_8168C_SPIN2 ||
1626 	    sc->rl_hwrev->rl_rev == RL_HWREV_8168CP) {
1627 		ifp->if_hwassist = 0;
1628 		ifp->if_capabilities = IFCAP_RXCSUM | IFCAP_TSO4;
1629 	} else {
1630 		ifp->if_hwassist = CSUM_IP | CSUM_TCP | CSUM_UDP;
1631 		ifp->if_capabilities = IFCAP_HWCSUM | IFCAP_TSO4;
1632 	}
1633 	ifp->if_hwassist |= CSUM_TSO;
1634 	ifp->if_capenable = ifp->if_capabilities;
1635 	ifp->if_init = re_init;
1636 	IFQ_SET_MAXLEN(&ifp->if_snd, RL_IFQ_MAXLEN);
1637 	ifp->if_snd.ifq_drv_maxlen = RL_IFQ_MAXLEN;
1638 	IFQ_SET_READY(&ifp->if_snd);
1639 
1640 	TASK_INIT(&sc->rl_inttask, 0, re_int_task, sc);
1641 
1642 #define	RE_PHYAD_INTERNAL	 0
1643 
1644 	/* Do MII setup. */
1645 	phy = RE_PHYAD_INTERNAL;
1646 	if (sc->rl_type == RL_8169)
1647 		phy = 1;
1648 	error = mii_attach(dev, &sc->rl_miibus, ifp, re_ifmedia_upd,
1649 	    re_ifmedia_sts, BMSR_DEFCAPMASK, phy, MII_OFFSET_ANY, MIIF_DOPAUSE);
1650 	if (error != 0) {
1651 		device_printf(dev, "attaching PHYs failed\n");
1652 		goto fail;
1653 	}
1654 
1655 	/*
1656 	 * Call MI attach routine.
1657 	 */
1658 	ether_ifattach(ifp, eaddr);
1659 
1660 	/* VLAN capability setup */
1661 	ifp->if_capabilities |= IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING;
1662 	if (ifp->if_capabilities & IFCAP_HWCSUM)
1663 		ifp->if_capabilities |= IFCAP_VLAN_HWCSUM;
1664 	/* Enable WOL if PM is supported. */
1665 	if (pci_find_cap(sc->rl_dev, PCIY_PMG, &reg) == 0)
1666 		ifp->if_capabilities |= IFCAP_WOL;
1667 	ifp->if_capenable = ifp->if_capabilities;
1668 	ifp->if_capenable &= ~(IFCAP_WOL_UCAST | IFCAP_WOL_MCAST);
1669 	/*
1670 	 * Don't enable TSO by default.  It is known to generate
1671 	 * corrupted TCP segments(bad TCP options) under certain
1672 	 * circumstances.
1673 	 */
1674 	ifp->if_hwassist &= ~CSUM_TSO;
1675 	ifp->if_capenable &= ~(IFCAP_TSO4 | IFCAP_VLAN_HWTSO);
1676 #ifdef DEVICE_POLLING
1677 	ifp->if_capabilities |= IFCAP_POLLING;
1678 #endif
1679 	/*
1680 	 * Tell the upper layer(s) we support long frames.
1681 	 * Must appear after the call to ether_ifattach() because
1682 	 * ether_ifattach() sets ifi_hdrlen to the default value.
1683 	 */
1684 	ifp->if_hdrlen = sizeof(struct ether_vlan_header);
1685 
1686 #ifdef DEV_NETMAP
1687 	re_netmap_attach(sc);
1688 #endif /* DEV_NETMAP */
1689 #ifdef RE_DIAG
1690 	/*
1691 	 * Perform hardware diagnostic on the original RTL8169.
1692 	 * Some 32-bit cards were incorrectly wired and would
1693 	 * malfunction if plugged into a 64-bit slot.
1694 	 */
1695 
1696 	if (hwrev == RL_HWREV_8169) {
1697 		error = re_diag(sc);
1698 		if (error) {
1699 			device_printf(dev,
1700 		    	"attach aborted due to hardware diag failure\n");
1701 			ether_ifdetach(ifp);
1702 			goto fail;
1703 		}
1704 	}
1705 #endif
1706 
1707 #ifdef RE_TX_MODERATION
1708 	intr_filter = 1;
1709 #endif
1710 	/* Hook interrupt last to avoid having to lock softc */
1711 	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) != 0 &&
1712 	    intr_filter == 0) {
1713 		error = bus_setup_intr(dev, sc->rl_irq[0],
1714 		    INTR_TYPE_NET | INTR_MPSAFE, NULL, re_intr_msi, sc,
1715 		    &sc->rl_intrhand[0]);
1716 	} else {
1717 		error = bus_setup_intr(dev, sc->rl_irq[0],
1718 		    INTR_TYPE_NET | INTR_MPSAFE, re_intr, NULL, sc,
1719 		    &sc->rl_intrhand[0]);
1720 	}
1721 	if (error) {
1722 		device_printf(dev, "couldn't set up irq\n");
1723 		ether_ifdetach(ifp);
1724 	}
1725 
1726 fail:
1727 
1728 	if (error)
1729 		re_detach(dev);
1730 
1731 	return (error);
1732 }
1733 
1734 /*
1735  * Shutdown hardware and free up resources. This can be called any
1736  * time after the mutex has been initialized. It is called in both
1737  * the error case in attach and the normal detach case so it needs
1738  * to be careful about only freeing resources that have actually been
1739  * allocated.
1740  */
1741 static int
1742 re_detach(device_t dev)
1743 {
1744 	struct rl_softc		*sc;
1745 	struct ifnet		*ifp;
1746 	int			i, rid;
1747 
1748 	sc = device_get_softc(dev);
1749 	ifp = sc->rl_ifp;
1750 	KASSERT(mtx_initialized(&sc->rl_mtx), ("re mutex not initialized"));
1751 
1752 	/* These should only be active if attach succeeded */
1753 	if (device_is_attached(dev)) {
1754 #ifdef DEVICE_POLLING
1755 		if (ifp->if_capenable & IFCAP_POLLING)
1756 			ether_poll_deregister(ifp);
1757 #endif
1758 		RL_LOCK(sc);
1759 #if 0
1760 		sc->suspended = 1;
1761 #endif
1762 		re_stop(sc);
1763 		RL_UNLOCK(sc);
1764 		callout_drain(&sc->rl_stat_callout);
1765 		taskqueue_drain(taskqueue_fast, &sc->rl_inttask);
1766 		/*
1767 		 * Force off the IFF_UP flag here, in case someone
1768 		 * still had a BPF descriptor attached to this
1769 		 * interface. If they do, ether_ifdetach() will cause
1770 		 * the BPF code to try and clear the promisc mode
1771 		 * flag, which will bubble down to re_ioctl(),
1772 		 * which will try to call re_init() again. This will
1773 		 * turn the NIC back on and restart the MII ticker,
1774 		 * which will panic the system when the kernel tries
1775 		 * to invoke the re_tick() function that isn't there
1776 		 * anymore.
1777 		 */
1778 		ifp->if_flags &= ~IFF_UP;
1779 		ether_ifdetach(ifp);
1780 	}
1781 	if (sc->rl_miibus)
1782 		device_delete_child(dev, sc->rl_miibus);
1783 	bus_generic_detach(dev);
1784 
1785 	/*
1786 	 * The rest is resource deallocation, so we should already be
1787 	 * stopped here.
1788 	 */
1789 
1790 	if (sc->rl_intrhand[0] != NULL) {
1791 		bus_teardown_intr(dev, sc->rl_irq[0], sc->rl_intrhand[0]);
1792 		sc->rl_intrhand[0] = NULL;
1793 	}
1794 	if (ifp != NULL) {
1795 #ifdef DEV_NETMAP
1796 		netmap_detach(ifp);
1797 #endif /* DEV_NETMAP */
1798 		if_free(ifp);
1799 	}
1800 	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) == 0)
1801 		rid = 0;
1802 	else
1803 		rid = 1;
1804 	if (sc->rl_irq[0] != NULL) {
1805 		bus_release_resource(dev, SYS_RES_IRQ, rid, sc->rl_irq[0]);
1806 		sc->rl_irq[0] = NULL;
1807 	}
1808 	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) != 0)
1809 		pci_release_msi(dev);
1810 	if (sc->rl_res_pba) {
1811 		rid = PCIR_BAR(4);
1812 		bus_release_resource(dev, SYS_RES_MEMORY, rid, sc->rl_res_pba);
1813 	}
1814 	if (sc->rl_res)
1815 		bus_release_resource(dev, sc->rl_res_type, sc->rl_res_id,
1816 		    sc->rl_res);
1817 
1818 	/* Unload and free the RX DMA ring memory and map */
1819 
1820 	if (sc->rl_ldata.rl_rx_list_tag) {
1821 		if (sc->rl_ldata.rl_rx_list_addr)
1822 			bus_dmamap_unload(sc->rl_ldata.rl_rx_list_tag,
1823 			    sc->rl_ldata.rl_rx_list_map);
1824 		if (sc->rl_ldata.rl_rx_list)
1825 			bus_dmamem_free(sc->rl_ldata.rl_rx_list_tag,
1826 			    sc->rl_ldata.rl_rx_list,
1827 			    sc->rl_ldata.rl_rx_list_map);
1828 		bus_dma_tag_destroy(sc->rl_ldata.rl_rx_list_tag);
1829 	}
1830 
1831 	/* Unload and free the TX DMA ring memory and map */
1832 
1833 	if (sc->rl_ldata.rl_tx_list_tag) {
1834 		if (sc->rl_ldata.rl_tx_list_addr)
1835 			bus_dmamap_unload(sc->rl_ldata.rl_tx_list_tag,
1836 			    sc->rl_ldata.rl_tx_list_map);
1837 		if (sc->rl_ldata.rl_tx_list)
1838 			bus_dmamem_free(sc->rl_ldata.rl_tx_list_tag,
1839 			    sc->rl_ldata.rl_tx_list,
1840 			    sc->rl_ldata.rl_tx_list_map);
1841 		bus_dma_tag_destroy(sc->rl_ldata.rl_tx_list_tag);
1842 	}
1843 
1844 	/* Destroy all the RX and TX buffer maps */
1845 
1846 	if (sc->rl_ldata.rl_tx_mtag) {
1847 		for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
1848 			if (sc->rl_ldata.rl_tx_desc[i].tx_dmamap)
1849 				bus_dmamap_destroy(sc->rl_ldata.rl_tx_mtag,
1850 				    sc->rl_ldata.rl_tx_desc[i].tx_dmamap);
1851 		}
1852 		bus_dma_tag_destroy(sc->rl_ldata.rl_tx_mtag);
1853 	}
1854 	if (sc->rl_ldata.rl_rx_mtag) {
1855 		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1856 			if (sc->rl_ldata.rl_rx_desc[i].rx_dmamap)
1857 				bus_dmamap_destroy(sc->rl_ldata.rl_rx_mtag,
1858 				    sc->rl_ldata.rl_rx_desc[i].rx_dmamap);
1859 		}
1860 		if (sc->rl_ldata.rl_rx_sparemap)
1861 			bus_dmamap_destroy(sc->rl_ldata.rl_rx_mtag,
1862 			    sc->rl_ldata.rl_rx_sparemap);
1863 		bus_dma_tag_destroy(sc->rl_ldata.rl_rx_mtag);
1864 	}
1865 	if (sc->rl_ldata.rl_jrx_mtag) {
1866 		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1867 			if (sc->rl_ldata.rl_jrx_desc[i].rx_dmamap)
1868 				bus_dmamap_destroy(sc->rl_ldata.rl_jrx_mtag,
1869 				    sc->rl_ldata.rl_jrx_desc[i].rx_dmamap);
1870 		}
1871 		if (sc->rl_ldata.rl_jrx_sparemap)
1872 			bus_dmamap_destroy(sc->rl_ldata.rl_jrx_mtag,
1873 			    sc->rl_ldata.rl_jrx_sparemap);
1874 		bus_dma_tag_destroy(sc->rl_ldata.rl_jrx_mtag);
1875 	}
1876 	/* Unload and free the stats buffer and map */
1877 
1878 	if (sc->rl_ldata.rl_stag) {
1879 		if (sc->rl_ldata.rl_stats_addr)
1880 			bus_dmamap_unload(sc->rl_ldata.rl_stag,
1881 			    sc->rl_ldata.rl_smap);
1882 		if (sc->rl_ldata.rl_stats)
1883 			bus_dmamem_free(sc->rl_ldata.rl_stag,
1884 			    sc->rl_ldata.rl_stats, sc->rl_ldata.rl_smap);
1885 		bus_dma_tag_destroy(sc->rl_ldata.rl_stag);
1886 	}
1887 
1888 	if (sc->rl_parent_tag)
1889 		bus_dma_tag_destroy(sc->rl_parent_tag);
1890 
1891 	mtx_destroy(&sc->rl_mtx);
1892 
1893 	return (0);
1894 }
1895 
1896 static __inline void
1897 re_discard_rxbuf(struct rl_softc *sc, int idx)
1898 {
1899 	struct rl_desc		*desc;
1900 	struct rl_rxdesc	*rxd;
1901 	uint32_t		cmdstat;
1902 
1903 	if (sc->rl_ifp->if_mtu > RL_MTU &&
1904 	    (sc->rl_flags & RL_FLAG_JUMBOV2) != 0)
1905 		rxd = &sc->rl_ldata.rl_jrx_desc[idx];
1906 	else
1907 		rxd = &sc->rl_ldata.rl_rx_desc[idx];
1908 	desc = &sc->rl_ldata.rl_rx_list[idx];
1909 	desc->rl_vlanctl = 0;
1910 	cmdstat = rxd->rx_size;
1911 	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
1912 		cmdstat |= RL_RDESC_CMD_EOR;
1913 	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
1914 }
1915 
1916 static int
1917 re_newbuf(struct rl_softc *sc, int idx)
1918 {
1919 	struct mbuf		*m;
1920 	struct rl_rxdesc	*rxd;
1921 	bus_dma_segment_t	segs[1];
1922 	bus_dmamap_t		map;
1923 	struct rl_desc		*desc;
1924 	uint32_t		cmdstat;
1925 	int			error, nsegs;
1926 
1927 	m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
1928 	if (m == NULL)
1929 		return (ENOBUFS);
1930 
1931 	m->m_len = m->m_pkthdr.len = MCLBYTES;
1932 #ifdef RE_FIXUP_RX
1933 	/*
1934 	 * This is part of an evil trick to deal with non-x86 platforms.
1935 	 * The RealTek chip requires RX buffers to be aligned on 64-bit
1936 	 * boundaries, but that will hose non-x86 machines. To get around
1937 	 * this, we leave some empty space at the start of each buffer
1938 	 * and for non-x86 hosts, we copy the buffer back six bytes
1939 	 * to achieve word alignment. This is slightly more efficient
1940 	 * than allocating a new buffer, copying the contents, and
1941 	 * discarding the old buffer.
1942 	 */
1943 	m_adj(m, RE_ETHER_ALIGN);
1944 #endif
1945 	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_rx_mtag,
1946 	    sc->rl_ldata.rl_rx_sparemap, m, segs, &nsegs, BUS_DMA_NOWAIT);
1947 	if (error != 0) {
1948 		m_freem(m);
1949 		return (ENOBUFS);
1950 	}
1951 	KASSERT(nsegs == 1, ("%s: %d segment returned!", __func__, nsegs));
1952 
1953 	rxd = &sc->rl_ldata.rl_rx_desc[idx];
1954 	if (rxd->rx_m != NULL) {
1955 		bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap,
1956 		    BUS_DMASYNC_POSTREAD);
1957 		bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap);
1958 	}
1959 
1960 	rxd->rx_m = m;
1961 	map = rxd->rx_dmamap;
1962 	rxd->rx_dmamap = sc->rl_ldata.rl_rx_sparemap;
1963 	rxd->rx_size = segs[0].ds_len;
1964 	sc->rl_ldata.rl_rx_sparemap = map;
1965 	bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap,
1966 	    BUS_DMASYNC_PREREAD);
1967 
1968 	desc = &sc->rl_ldata.rl_rx_list[idx];
1969 	desc->rl_vlanctl = 0;
1970 	desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[0].ds_addr));
1971 	desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[0].ds_addr));
1972 	cmdstat = segs[0].ds_len;
1973 	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
1974 		cmdstat |= RL_RDESC_CMD_EOR;
1975 	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
1976 
1977 	return (0);
1978 }
1979 
1980 static int
1981 re_jumbo_newbuf(struct rl_softc *sc, int idx)
1982 {
1983 	struct mbuf		*m;
1984 	struct rl_rxdesc	*rxd;
1985 	bus_dma_segment_t	segs[1];
1986 	bus_dmamap_t		map;
1987 	struct rl_desc		*desc;
1988 	uint32_t		cmdstat;
1989 	int			error, nsegs;
1990 
1991 	m = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, MJUM9BYTES);
1992 	if (m == NULL)
1993 		return (ENOBUFS);
1994 	m->m_len = m->m_pkthdr.len = MJUM9BYTES;
1995 #ifdef RE_FIXUP_RX
1996 	m_adj(m, RE_ETHER_ALIGN);
1997 #endif
1998 	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_jrx_mtag,
1999 	    sc->rl_ldata.rl_jrx_sparemap, m, segs, &nsegs, BUS_DMA_NOWAIT);
2000 	if (error != 0) {
2001 		m_freem(m);
2002 		return (ENOBUFS);
2003 	}
2004 	KASSERT(nsegs == 1, ("%s: %d segment returned!", __func__, nsegs));
2005 
2006 	rxd = &sc->rl_ldata.rl_jrx_desc[idx];
2007 	if (rxd->rx_m != NULL) {
2008 		bus_dmamap_sync(sc->rl_ldata.rl_jrx_mtag, rxd->rx_dmamap,
2009 		    BUS_DMASYNC_POSTREAD);
2010 		bus_dmamap_unload(sc->rl_ldata.rl_jrx_mtag, rxd->rx_dmamap);
2011 	}
2012 
2013 	rxd->rx_m = m;
2014 	map = rxd->rx_dmamap;
2015 	rxd->rx_dmamap = sc->rl_ldata.rl_jrx_sparemap;
2016 	rxd->rx_size = segs[0].ds_len;
2017 	sc->rl_ldata.rl_jrx_sparemap = map;
2018 	bus_dmamap_sync(sc->rl_ldata.rl_jrx_mtag, rxd->rx_dmamap,
2019 	    BUS_DMASYNC_PREREAD);
2020 
2021 	desc = &sc->rl_ldata.rl_rx_list[idx];
2022 	desc->rl_vlanctl = 0;
2023 	desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[0].ds_addr));
2024 	desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[0].ds_addr));
2025 	cmdstat = segs[0].ds_len;
2026 	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
2027 		cmdstat |= RL_RDESC_CMD_EOR;
2028 	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
2029 
2030 	return (0);
2031 }
2032 
2033 #ifdef RE_FIXUP_RX
2034 static __inline void
2035 re_fixup_rx(struct mbuf *m)
2036 {
2037 	int                     i;
2038 	uint16_t                *src, *dst;
2039 
2040 	src = mtod(m, uint16_t *);
2041 	dst = src - (RE_ETHER_ALIGN - ETHER_ALIGN) / sizeof *src;
2042 
2043 	for (i = 0; i < (m->m_len / sizeof(uint16_t) + 1); i++)
2044 		*dst++ = *src++;
2045 
2046 	m->m_data -= RE_ETHER_ALIGN - ETHER_ALIGN;
2047 }
2048 #endif
2049 
2050 static int
2051 re_tx_list_init(struct rl_softc *sc)
2052 {
2053 	struct rl_desc		*desc;
2054 	int			i;
2055 
2056 	RL_LOCK_ASSERT(sc);
2057 
2058 	bzero(sc->rl_ldata.rl_tx_list,
2059 	    sc->rl_ldata.rl_tx_desc_cnt * sizeof(struct rl_desc));
2060 	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++)
2061 		sc->rl_ldata.rl_tx_desc[i].tx_m = NULL;
2062 #ifdef DEV_NETMAP
2063 	re_netmap_tx_init(sc);
2064 #endif /* DEV_NETMAP */
2065 	/* Set EOR. */
2066 	desc = &sc->rl_ldata.rl_tx_list[sc->rl_ldata.rl_tx_desc_cnt - 1];
2067 	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_EOR);
2068 
2069 	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2070 	    sc->rl_ldata.rl_tx_list_map,
2071 	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
2072 
2073 	sc->rl_ldata.rl_tx_prodidx = 0;
2074 	sc->rl_ldata.rl_tx_considx = 0;
2075 	sc->rl_ldata.rl_tx_free = sc->rl_ldata.rl_tx_desc_cnt;
2076 
2077 	return (0);
2078 }
2079 
2080 static int
2081 re_rx_list_init(struct rl_softc *sc)
2082 {
2083 	int			error, i;
2084 
2085 	bzero(sc->rl_ldata.rl_rx_list,
2086 	    sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc));
2087 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
2088 		sc->rl_ldata.rl_rx_desc[i].rx_m = NULL;
2089 		if ((error = re_newbuf(sc, i)) != 0)
2090 			return (error);
2091 	}
2092 #ifdef DEV_NETMAP
2093 	re_netmap_rx_init(sc);
2094 #endif /* DEV_NETMAP */
2095 
2096 	/* Flush the RX descriptors */
2097 
2098 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
2099 	    sc->rl_ldata.rl_rx_list_map,
2100 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2101 
2102 	sc->rl_ldata.rl_rx_prodidx = 0;
2103 	sc->rl_head = sc->rl_tail = NULL;
2104 	sc->rl_int_rx_act = 0;
2105 
2106 	return (0);
2107 }
2108 
2109 static int
2110 re_jrx_list_init(struct rl_softc *sc)
2111 {
2112 	int			error, i;
2113 
2114 	bzero(sc->rl_ldata.rl_rx_list,
2115 	    sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc));
2116 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
2117 		sc->rl_ldata.rl_jrx_desc[i].rx_m = NULL;
2118 		if ((error = re_jumbo_newbuf(sc, i)) != 0)
2119 			return (error);
2120 	}
2121 
2122 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
2123 	    sc->rl_ldata.rl_rx_list_map,
2124 	    BUS_DMASYNC_PREWRITE | BUS_DMASYNC_PREREAD);
2125 
2126 	sc->rl_ldata.rl_rx_prodidx = 0;
2127 	sc->rl_head = sc->rl_tail = NULL;
2128 	sc->rl_int_rx_act = 0;
2129 
2130 	return (0);
2131 }
2132 
2133 /*
2134  * RX handler for C+ and 8169. For the gigE chips, we support
2135  * the reception of jumbo frames that have been fragmented
2136  * across multiple 2K mbuf cluster buffers.
2137  */
2138 static int
2139 re_rxeof(struct rl_softc *sc, int *rx_npktsp)
2140 {
2141 	struct mbuf		*m;
2142 	struct ifnet		*ifp;
2143 	int			i, rxerr, total_len;
2144 	struct rl_desc		*cur_rx;
2145 	u_int32_t		rxstat, rxvlan;
2146 	int			jumbo, maxpkt = 16, rx_npkts = 0;
2147 
2148 	RL_LOCK_ASSERT(sc);
2149 
2150 	ifp = sc->rl_ifp;
2151 #ifdef DEV_NETMAP
2152 	if (netmap_rx_irq(ifp, 0, &rx_npkts))
2153 		return 0;
2154 #endif /* DEV_NETMAP */
2155 	if (ifp->if_mtu > RL_MTU && (sc->rl_flags & RL_FLAG_JUMBOV2) != 0)
2156 		jumbo = 1;
2157 	else
2158 		jumbo = 0;
2159 
2160 	/* Invalidate the descriptor memory */
2161 
2162 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
2163 	    sc->rl_ldata.rl_rx_list_map,
2164 	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2165 
2166 	for (i = sc->rl_ldata.rl_rx_prodidx; maxpkt > 0;
2167 	    i = RL_RX_DESC_NXT(sc, i)) {
2168 		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
2169 			break;
2170 		cur_rx = &sc->rl_ldata.rl_rx_list[i];
2171 		rxstat = le32toh(cur_rx->rl_cmdstat);
2172 		if ((rxstat & RL_RDESC_STAT_OWN) != 0)
2173 			break;
2174 		total_len = rxstat & sc->rl_rxlenmask;
2175 		rxvlan = le32toh(cur_rx->rl_vlanctl);
2176 		if (jumbo != 0)
2177 			m = sc->rl_ldata.rl_jrx_desc[i].rx_m;
2178 		else
2179 			m = sc->rl_ldata.rl_rx_desc[i].rx_m;
2180 
2181 		if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0 &&
2182 		    (rxstat & (RL_RDESC_STAT_SOF | RL_RDESC_STAT_EOF)) !=
2183 		    (RL_RDESC_STAT_SOF | RL_RDESC_STAT_EOF)) {
2184 			/*
2185 			 * RTL8168C or later controllers do not
2186 			 * support multi-fragment packet.
2187 			 */
2188 			re_discard_rxbuf(sc, i);
2189 			continue;
2190 		} else if ((rxstat & RL_RDESC_STAT_EOF) == 0) {
2191 			if (re_newbuf(sc, i) != 0) {
2192 				/*
2193 				 * If this is part of a multi-fragment packet,
2194 				 * discard all the pieces.
2195 				 */
2196 				if (sc->rl_head != NULL) {
2197 					m_freem(sc->rl_head);
2198 					sc->rl_head = sc->rl_tail = NULL;
2199 				}
2200 				re_discard_rxbuf(sc, i);
2201 				continue;
2202 			}
2203 			m->m_len = RE_RX_DESC_BUFLEN;
2204 			if (sc->rl_head == NULL)
2205 				sc->rl_head = sc->rl_tail = m;
2206 			else {
2207 				m->m_flags &= ~M_PKTHDR;
2208 				sc->rl_tail->m_next = m;
2209 				sc->rl_tail = m;
2210 			}
2211 			continue;
2212 		}
2213 
2214 		/*
2215 		 * NOTE: for the 8139C+, the frame length field
2216 		 * is always 12 bits in size, but for the gigE chips,
2217 		 * it is 13 bits (since the max RX frame length is 16K).
2218 		 * Unfortunately, all 32 bits in the status word
2219 		 * were already used, so to make room for the extra
2220 		 * length bit, RealTek took out the 'frame alignment
2221 		 * error' bit and shifted the other status bits
2222 		 * over one slot. The OWN, EOR, FS and LS bits are
2223 		 * still in the same places. We have already extracted
2224 		 * the frame length and checked the OWN bit, so rather
2225 		 * than using an alternate bit mapping, we shift the
2226 		 * status bits one space to the right so we can evaluate
2227 		 * them using the 8169 status as though it was in the
2228 		 * same format as that of the 8139C+.
2229 		 */
2230 		if (sc->rl_type == RL_8169)
2231 			rxstat >>= 1;
2232 
2233 		/*
2234 		 * if total_len > 2^13-1, both _RXERRSUM and _GIANT will be
2235 		 * set, but if CRC is clear, it will still be a valid frame.
2236 		 */
2237 		if ((rxstat & RL_RDESC_STAT_RXERRSUM) != 0) {
2238 			rxerr = 1;
2239 			if ((sc->rl_flags & RL_FLAG_JUMBOV2) == 0 &&
2240 			    total_len > 8191 &&
2241 			    (rxstat & RL_RDESC_STAT_ERRS) == RL_RDESC_STAT_GIANT)
2242 				rxerr = 0;
2243 			if (rxerr != 0) {
2244 				ifp->if_ierrors++;
2245 				/*
2246 				 * If this is part of a multi-fragment packet,
2247 				 * discard all the pieces.
2248 				 */
2249 				if (sc->rl_head != NULL) {
2250 					m_freem(sc->rl_head);
2251 					sc->rl_head = sc->rl_tail = NULL;
2252 				}
2253 				re_discard_rxbuf(sc, i);
2254 				continue;
2255 			}
2256 		}
2257 
2258 		/*
2259 		 * If allocating a replacement mbuf fails,
2260 		 * reload the current one.
2261 		 */
2262 		if (jumbo != 0)
2263 			rxerr = re_jumbo_newbuf(sc, i);
2264 		else
2265 			rxerr = re_newbuf(sc, i);
2266 		if (rxerr != 0) {
2267 			ifp->if_iqdrops++;
2268 			if (sc->rl_head != NULL) {
2269 				m_freem(sc->rl_head);
2270 				sc->rl_head = sc->rl_tail = NULL;
2271 			}
2272 			re_discard_rxbuf(sc, i);
2273 			continue;
2274 		}
2275 
2276 		if (sc->rl_head != NULL) {
2277 			if (jumbo != 0)
2278 				m->m_len = total_len;
2279 			else {
2280 				m->m_len = total_len % RE_RX_DESC_BUFLEN;
2281 				if (m->m_len == 0)
2282 					m->m_len = RE_RX_DESC_BUFLEN;
2283 			}
2284 			/*
2285 			 * Special case: if there's 4 bytes or less
2286 			 * in this buffer, the mbuf can be discarded:
2287 			 * the last 4 bytes is the CRC, which we don't
2288 			 * care about anyway.
2289 			 */
2290 			if (m->m_len <= ETHER_CRC_LEN) {
2291 				sc->rl_tail->m_len -=
2292 				    (ETHER_CRC_LEN - m->m_len);
2293 				m_freem(m);
2294 			} else {
2295 				m->m_len -= ETHER_CRC_LEN;
2296 				m->m_flags &= ~M_PKTHDR;
2297 				sc->rl_tail->m_next = m;
2298 			}
2299 			m = sc->rl_head;
2300 			sc->rl_head = sc->rl_tail = NULL;
2301 			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
2302 		} else
2303 			m->m_pkthdr.len = m->m_len =
2304 			    (total_len - ETHER_CRC_LEN);
2305 
2306 #ifdef RE_FIXUP_RX
2307 		re_fixup_rx(m);
2308 #endif
2309 		ifp->if_ipackets++;
2310 		m->m_pkthdr.rcvif = ifp;
2311 
2312 		/* Do RX checksumming if enabled */
2313 
2314 		if (ifp->if_capenable & IFCAP_RXCSUM) {
2315 			if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
2316 				/* Check IP header checksum */
2317 				if (rxstat & RL_RDESC_STAT_PROTOID)
2318 					m->m_pkthdr.csum_flags |=
2319 					    CSUM_IP_CHECKED;
2320 				if (!(rxstat & RL_RDESC_STAT_IPSUMBAD))
2321 					m->m_pkthdr.csum_flags |=
2322 					    CSUM_IP_VALID;
2323 
2324 				/* Check TCP/UDP checksum */
2325 				if ((RL_TCPPKT(rxstat) &&
2326 				    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
2327 				    (RL_UDPPKT(rxstat) &&
2328 				     !(rxstat & RL_RDESC_STAT_UDPSUMBAD))) {
2329 					m->m_pkthdr.csum_flags |=
2330 						CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
2331 					m->m_pkthdr.csum_data = 0xffff;
2332 				}
2333 			} else {
2334 				/*
2335 				 * RTL8168C/RTL816CP/RTL8111C/RTL8111CP
2336 				 */
2337 				if ((rxstat & RL_RDESC_STAT_PROTOID) &&
2338 				    (rxvlan & RL_RDESC_IPV4))
2339 					m->m_pkthdr.csum_flags |=
2340 					    CSUM_IP_CHECKED;
2341 				if (!(rxstat & RL_RDESC_STAT_IPSUMBAD) &&
2342 				    (rxvlan & RL_RDESC_IPV4))
2343 					m->m_pkthdr.csum_flags |=
2344 					    CSUM_IP_VALID;
2345 				if (((rxstat & RL_RDESC_STAT_TCP) &&
2346 				    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
2347 				    ((rxstat & RL_RDESC_STAT_UDP) &&
2348 				    !(rxstat & RL_RDESC_STAT_UDPSUMBAD))) {
2349 					m->m_pkthdr.csum_flags |=
2350 						CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
2351 					m->m_pkthdr.csum_data = 0xffff;
2352 				}
2353 			}
2354 		}
2355 		maxpkt--;
2356 		if (rxvlan & RL_RDESC_VLANCTL_TAG) {
2357 			m->m_pkthdr.ether_vtag =
2358 			    bswap16((rxvlan & RL_RDESC_VLANCTL_DATA));
2359 			m->m_flags |= M_VLANTAG;
2360 		}
2361 		RL_UNLOCK(sc);
2362 		(*ifp->if_input)(ifp, m);
2363 		RL_LOCK(sc);
2364 		rx_npkts++;
2365 	}
2366 
2367 	/* Flush the RX DMA ring */
2368 
2369 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
2370 	    sc->rl_ldata.rl_rx_list_map,
2371 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2372 
2373 	sc->rl_ldata.rl_rx_prodidx = i;
2374 
2375 	if (rx_npktsp != NULL)
2376 		*rx_npktsp = rx_npkts;
2377 	if (maxpkt)
2378 		return (EAGAIN);
2379 
2380 	return (0);
2381 }
2382 
2383 static void
2384 re_txeof(struct rl_softc *sc)
2385 {
2386 	struct ifnet		*ifp;
2387 	struct rl_txdesc	*txd;
2388 	u_int32_t		txstat;
2389 	int			cons;
2390 
2391 	cons = sc->rl_ldata.rl_tx_considx;
2392 	if (cons == sc->rl_ldata.rl_tx_prodidx)
2393 		return;
2394 
2395 	ifp = sc->rl_ifp;
2396 #ifdef DEV_NETMAP
2397 	if (netmap_tx_irq(ifp, 0))
2398 		return;
2399 #endif /* DEV_NETMAP */
2400 	/* Invalidate the TX descriptor list */
2401 	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2402 	    sc->rl_ldata.rl_tx_list_map,
2403 	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2404 
2405 	for (; cons != sc->rl_ldata.rl_tx_prodidx;
2406 	    cons = RL_TX_DESC_NXT(sc, cons)) {
2407 		txstat = le32toh(sc->rl_ldata.rl_tx_list[cons].rl_cmdstat);
2408 		if (txstat & RL_TDESC_STAT_OWN)
2409 			break;
2410 		/*
2411 		 * We only stash mbufs in the last descriptor
2412 		 * in a fragment chain, which also happens to
2413 		 * be the only place where the TX status bits
2414 		 * are valid.
2415 		 */
2416 		if (txstat & RL_TDESC_CMD_EOF) {
2417 			txd = &sc->rl_ldata.rl_tx_desc[cons];
2418 			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
2419 			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
2420 			bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag,
2421 			    txd->tx_dmamap);
2422 			KASSERT(txd->tx_m != NULL,
2423 			    ("%s: freeing NULL mbufs!", __func__));
2424 			m_freem(txd->tx_m);
2425 			txd->tx_m = NULL;
2426 			if (txstat & (RL_TDESC_STAT_EXCESSCOL|
2427 			    RL_TDESC_STAT_COLCNT))
2428 				ifp->if_collisions++;
2429 			if (txstat & RL_TDESC_STAT_TXERRSUM)
2430 				ifp->if_oerrors++;
2431 			else
2432 				ifp->if_opackets++;
2433 		}
2434 		sc->rl_ldata.rl_tx_free++;
2435 		ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
2436 	}
2437 	sc->rl_ldata.rl_tx_considx = cons;
2438 
2439 	/* No changes made to the TX ring, so no flush needed */
2440 
2441 	if (sc->rl_ldata.rl_tx_free != sc->rl_ldata.rl_tx_desc_cnt) {
2442 #ifdef RE_TX_MODERATION
2443 		/*
2444 		 * If not all descriptors have been reaped yet, reload
2445 		 * the timer so that we will eventually get another
2446 		 * interrupt that will cause us to re-enter this routine.
2447 		 * This is done in case the transmitter has gone idle.
2448 		 */
2449 		CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2450 #endif
2451 	} else
2452 		sc->rl_watchdog_timer = 0;
2453 }
2454 
2455 static void
2456 re_tick(void *xsc)
2457 {
2458 	struct rl_softc		*sc;
2459 	struct mii_data		*mii;
2460 
2461 	sc = xsc;
2462 
2463 	RL_LOCK_ASSERT(sc);
2464 
2465 	mii = device_get_softc(sc->rl_miibus);
2466 	mii_tick(mii);
2467 	if ((sc->rl_flags & RL_FLAG_LINK) == 0)
2468 		re_miibus_statchg(sc->rl_dev);
2469 	/*
2470 	 * Reclaim transmitted frames here. Technically it is not
2471 	 * necessary to do here but it ensures periodic reclamation
2472 	 * regardless of Tx completion interrupt which seems to be
2473 	 * lost on PCIe based controllers under certain situations.
2474 	 */
2475 	re_txeof(sc);
2476 	re_watchdog(sc);
2477 	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
2478 }
2479 
2480 #ifdef DEVICE_POLLING
2481 static int
2482 re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
2483 {
2484 	struct rl_softc *sc = ifp->if_softc;
2485 	int rx_npkts = 0;
2486 
2487 	RL_LOCK(sc);
2488 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
2489 		rx_npkts = re_poll_locked(ifp, cmd, count);
2490 	RL_UNLOCK(sc);
2491 	return (rx_npkts);
2492 }
2493 
2494 static int
2495 re_poll_locked(struct ifnet *ifp, enum poll_cmd cmd, int count)
2496 {
2497 	struct rl_softc *sc = ifp->if_softc;
2498 	int rx_npkts;
2499 
2500 	RL_LOCK_ASSERT(sc);
2501 
2502 	sc->rxcycles = count;
2503 	re_rxeof(sc, &rx_npkts);
2504 	re_txeof(sc);
2505 
2506 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2507 		re_start_locked(ifp);
2508 
2509 	if (cmd == POLL_AND_CHECK_STATUS) { /* also check status register */
2510 		u_int16_t       status;
2511 
2512 		status = CSR_READ_2(sc, RL_ISR);
2513 		if (status == 0xffff)
2514 			return (rx_npkts);
2515 		if (status)
2516 			CSR_WRITE_2(sc, RL_ISR, status);
2517 		if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
2518 		    (sc->rl_flags & RL_FLAG_PCIE))
2519 			CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2520 
2521 		/*
2522 		 * XXX check behaviour on receiver stalls.
2523 		 */
2524 
2525 		if (status & RL_ISR_SYSTEM_ERR) {
2526 			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2527 			re_init_locked(sc);
2528 		}
2529 	}
2530 	return (rx_npkts);
2531 }
2532 #endif /* DEVICE_POLLING */
2533 
2534 static int
2535 re_intr(void *arg)
2536 {
2537 	struct rl_softc		*sc;
2538 	uint16_t		status;
2539 
2540 	sc = arg;
2541 
2542 	status = CSR_READ_2(sc, RL_ISR);
2543 	if (status == 0xFFFF || (status & RL_INTRS_CPLUS) == 0)
2544                 return (FILTER_STRAY);
2545 	CSR_WRITE_2(sc, RL_IMR, 0);
2546 
2547 	taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_inttask);
2548 
2549 	return (FILTER_HANDLED);
2550 }
2551 
2552 static void
2553 re_int_task(void *arg, int npending)
2554 {
2555 	struct rl_softc		*sc;
2556 	struct ifnet		*ifp;
2557 	u_int16_t		status;
2558 	int			rval = 0;
2559 
2560 	sc = arg;
2561 	ifp = sc->rl_ifp;
2562 
2563 	RL_LOCK(sc);
2564 
2565 	status = CSR_READ_2(sc, RL_ISR);
2566         CSR_WRITE_2(sc, RL_ISR, status);
2567 
2568 	if (sc->suspended ||
2569 	    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2570 		RL_UNLOCK(sc);
2571 		return;
2572 	}
2573 
2574 #ifdef DEVICE_POLLING
2575 	if  (ifp->if_capenable & IFCAP_POLLING) {
2576 		RL_UNLOCK(sc);
2577 		return;
2578 	}
2579 #endif
2580 
2581 	if (status & (RL_ISR_RX_OK|RL_ISR_RX_ERR|RL_ISR_FIFO_OFLOW))
2582 		rval = re_rxeof(sc, NULL);
2583 
2584 	/*
2585 	 * Some chips will ignore a second TX request issued
2586 	 * while an existing transmission is in progress. If
2587 	 * the transmitter goes idle but there are still
2588 	 * packets waiting to be sent, we need to restart the
2589 	 * channel here to flush them out. This only seems to
2590 	 * be required with the PCIe devices.
2591 	 */
2592 	if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
2593 	    (sc->rl_flags & RL_FLAG_PCIE))
2594 		CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2595 	if (status & (
2596 #ifdef RE_TX_MODERATION
2597 	    RL_ISR_TIMEOUT_EXPIRED|
2598 #else
2599 	    RL_ISR_TX_OK|
2600 #endif
2601 	    RL_ISR_TX_ERR|RL_ISR_TX_DESC_UNAVAIL))
2602 		re_txeof(sc);
2603 
2604 	if (status & RL_ISR_SYSTEM_ERR) {
2605 		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2606 		re_init_locked(sc);
2607 	}
2608 
2609 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2610 		re_start_locked(ifp);
2611 
2612 	RL_UNLOCK(sc);
2613 
2614         if ((CSR_READ_2(sc, RL_ISR) & RL_INTRS_CPLUS) || rval) {
2615 		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_inttask);
2616 		return;
2617 	}
2618 
2619 	CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2620 }
2621 
2622 static void
2623 re_intr_msi(void *xsc)
2624 {
2625 	struct rl_softc		*sc;
2626 	struct ifnet		*ifp;
2627 	uint16_t		intrs, status;
2628 
2629 	sc = xsc;
2630 	RL_LOCK(sc);
2631 
2632 	ifp = sc->rl_ifp;
2633 #ifdef DEVICE_POLLING
2634 	if (ifp->if_capenable & IFCAP_POLLING) {
2635 		RL_UNLOCK(sc);
2636 		return;
2637 	}
2638 #endif
2639 	/* Disable interrupts. */
2640 	CSR_WRITE_2(sc, RL_IMR, 0);
2641 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2642 		RL_UNLOCK(sc);
2643 		return;
2644 	}
2645 
2646 	intrs = RL_INTRS_CPLUS;
2647 	status = CSR_READ_2(sc, RL_ISR);
2648         CSR_WRITE_2(sc, RL_ISR, status);
2649 	if (sc->rl_int_rx_act > 0) {
2650 		intrs &= ~(RL_ISR_RX_OK | RL_ISR_RX_ERR | RL_ISR_FIFO_OFLOW |
2651 		    RL_ISR_RX_OVERRUN);
2652 		status &= ~(RL_ISR_RX_OK | RL_ISR_RX_ERR | RL_ISR_FIFO_OFLOW |
2653 		    RL_ISR_RX_OVERRUN);
2654 	}
2655 
2656 	if (status & (RL_ISR_TIMEOUT_EXPIRED | RL_ISR_RX_OK | RL_ISR_RX_ERR |
2657 	    RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN)) {
2658 		re_rxeof(sc, NULL);
2659 		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
2660 			if (sc->rl_int_rx_mod != 0 &&
2661 			    (status & (RL_ISR_RX_OK | RL_ISR_RX_ERR |
2662 			    RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN)) != 0) {
2663 				/* Rearm one-shot timer. */
2664 				CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2665 				intrs &= ~(RL_ISR_RX_OK | RL_ISR_RX_ERR |
2666 				    RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN);
2667 				sc->rl_int_rx_act = 1;
2668 			} else {
2669 				intrs |= RL_ISR_RX_OK | RL_ISR_RX_ERR |
2670 				    RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN;
2671 				sc->rl_int_rx_act = 0;
2672 			}
2673 		}
2674 	}
2675 
2676 	/*
2677 	 * Some chips will ignore a second TX request issued
2678 	 * while an existing transmission is in progress. If
2679 	 * the transmitter goes idle but there are still
2680 	 * packets waiting to be sent, we need to restart the
2681 	 * channel here to flush them out. This only seems to
2682 	 * be required with the PCIe devices.
2683 	 */
2684 	if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
2685 	    (sc->rl_flags & RL_FLAG_PCIE))
2686 		CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2687 	if (status & (RL_ISR_TX_OK | RL_ISR_TX_ERR | RL_ISR_TX_DESC_UNAVAIL))
2688 		re_txeof(sc);
2689 
2690 	if (status & RL_ISR_SYSTEM_ERR) {
2691 		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2692 		re_init_locked(sc);
2693 	}
2694 
2695 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
2696 		if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2697 			re_start_locked(ifp);
2698 		CSR_WRITE_2(sc, RL_IMR, intrs);
2699 	}
2700 	RL_UNLOCK(sc);
2701 }
2702 
2703 static int
2704 re_encap(struct rl_softc *sc, struct mbuf **m_head)
2705 {
2706 	struct rl_txdesc	*txd, *txd_last;
2707 	bus_dma_segment_t	segs[RL_NTXSEGS];
2708 	bus_dmamap_t		map;
2709 	struct mbuf		*m_new;
2710 	struct rl_desc		*desc;
2711 	int			nsegs, prod;
2712 	int			i, error, ei, si;
2713 	int			padlen;
2714 	uint32_t		cmdstat, csum_flags, vlanctl;
2715 
2716 	RL_LOCK_ASSERT(sc);
2717 	M_ASSERTPKTHDR((*m_head));
2718 
2719 	/*
2720 	 * With some of the RealTek chips, using the checksum offload
2721 	 * support in conjunction with the autopadding feature results
2722 	 * in the transmission of corrupt frames. For example, if we
2723 	 * need to send a really small IP fragment that's less than 60
2724 	 * bytes in size, and IP header checksumming is enabled, the
2725 	 * resulting ethernet frame that appears on the wire will
2726 	 * have garbled payload. To work around this, if TX IP checksum
2727 	 * offload is enabled, we always manually pad short frames out
2728 	 * to the minimum ethernet frame size.
2729 	 */
2730 	if ((sc->rl_flags & RL_FLAG_AUTOPAD) == 0 &&
2731 	    (*m_head)->m_pkthdr.len < RL_IP4CSUMTX_PADLEN &&
2732 	    ((*m_head)->m_pkthdr.csum_flags & CSUM_IP) != 0) {
2733 		padlen = RL_MIN_FRAMELEN - (*m_head)->m_pkthdr.len;
2734 		if (M_WRITABLE(*m_head) == 0) {
2735 			/* Get a writable copy. */
2736 			m_new = m_dup(*m_head, M_NOWAIT);
2737 			m_freem(*m_head);
2738 			if (m_new == NULL) {
2739 				*m_head = NULL;
2740 				return (ENOBUFS);
2741 			}
2742 			*m_head = m_new;
2743 		}
2744 		if ((*m_head)->m_next != NULL ||
2745 		    M_TRAILINGSPACE(*m_head) < padlen) {
2746 			m_new = m_defrag(*m_head, M_NOWAIT);
2747 			if (m_new == NULL) {
2748 				m_freem(*m_head);
2749 				*m_head = NULL;
2750 				return (ENOBUFS);
2751 			}
2752 		} else
2753 			m_new = *m_head;
2754 
2755 		/*
2756 		 * Manually pad short frames, and zero the pad space
2757 		 * to avoid leaking data.
2758 		 */
2759 		bzero(mtod(m_new, char *) + m_new->m_pkthdr.len, padlen);
2760 		m_new->m_pkthdr.len += padlen;
2761 		m_new->m_len = m_new->m_pkthdr.len;
2762 		*m_head = m_new;
2763 	}
2764 
2765 	prod = sc->rl_ldata.rl_tx_prodidx;
2766 	txd = &sc->rl_ldata.rl_tx_desc[prod];
2767 	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap,
2768 	    *m_head, segs, &nsegs, BUS_DMA_NOWAIT);
2769 	if (error == EFBIG) {
2770 		m_new = m_collapse(*m_head, M_NOWAIT, RL_NTXSEGS);
2771 		if (m_new == NULL) {
2772 			m_freem(*m_head);
2773 			*m_head = NULL;
2774 			return (ENOBUFS);
2775 		}
2776 		*m_head = m_new;
2777 		error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_tx_mtag,
2778 		    txd->tx_dmamap, *m_head, segs, &nsegs, BUS_DMA_NOWAIT);
2779 		if (error != 0) {
2780 			m_freem(*m_head);
2781 			*m_head = NULL;
2782 			return (error);
2783 		}
2784 	} else if (error != 0)
2785 		return (error);
2786 	if (nsegs == 0) {
2787 		m_freem(*m_head);
2788 		*m_head = NULL;
2789 		return (EIO);
2790 	}
2791 
2792 	/* Check for number of available descriptors. */
2793 	if (sc->rl_ldata.rl_tx_free - nsegs <= 1) {
2794 		bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap);
2795 		return (ENOBUFS);
2796 	}
2797 
2798 	bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap,
2799 	    BUS_DMASYNC_PREWRITE);
2800 
2801 	/*
2802 	 * Set up checksum offload. Note: checksum offload bits must
2803 	 * appear in all descriptors of a multi-descriptor transmit
2804 	 * attempt. This is according to testing done with an 8169
2805 	 * chip. This is a requirement.
2806 	 */
2807 	vlanctl = 0;
2808 	csum_flags = 0;
2809 	if (((*m_head)->m_pkthdr.csum_flags & CSUM_TSO) != 0) {
2810 		if ((sc->rl_flags & RL_FLAG_DESCV2) != 0) {
2811 			csum_flags |= RL_TDESC_CMD_LGSEND;
2812 			vlanctl |= ((uint32_t)(*m_head)->m_pkthdr.tso_segsz <<
2813 			    RL_TDESC_CMD_MSSVALV2_SHIFT);
2814 		} else {
2815 			csum_flags |= RL_TDESC_CMD_LGSEND |
2816 			    ((uint32_t)(*m_head)->m_pkthdr.tso_segsz <<
2817 			    RL_TDESC_CMD_MSSVAL_SHIFT);
2818 		}
2819 	} else {
2820 		/*
2821 		 * Unconditionally enable IP checksum if TCP or UDP
2822 		 * checksum is required. Otherwise, TCP/UDP checksum
2823 		 * doesn't make effects.
2824 		 */
2825 		if (((*m_head)->m_pkthdr.csum_flags & RE_CSUM_FEATURES) != 0) {
2826 			if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
2827 				csum_flags |= RL_TDESC_CMD_IPCSUM;
2828 				if (((*m_head)->m_pkthdr.csum_flags &
2829 				    CSUM_TCP) != 0)
2830 					csum_flags |= RL_TDESC_CMD_TCPCSUM;
2831 				if (((*m_head)->m_pkthdr.csum_flags &
2832 				    CSUM_UDP) != 0)
2833 					csum_flags |= RL_TDESC_CMD_UDPCSUM;
2834 			} else {
2835 				vlanctl |= RL_TDESC_CMD_IPCSUMV2;
2836 				if (((*m_head)->m_pkthdr.csum_flags &
2837 				    CSUM_TCP) != 0)
2838 					vlanctl |= RL_TDESC_CMD_TCPCSUMV2;
2839 				if (((*m_head)->m_pkthdr.csum_flags &
2840 				    CSUM_UDP) != 0)
2841 					vlanctl |= RL_TDESC_CMD_UDPCSUMV2;
2842 			}
2843 		}
2844 	}
2845 
2846 	/*
2847 	 * Set up hardware VLAN tagging. Note: vlan tag info must
2848 	 * appear in all descriptors of a multi-descriptor
2849 	 * transmission attempt.
2850 	 */
2851 	if ((*m_head)->m_flags & M_VLANTAG)
2852 		vlanctl |= bswap16((*m_head)->m_pkthdr.ether_vtag) |
2853 		    RL_TDESC_VLANCTL_TAG;
2854 
2855 	si = prod;
2856 	for (i = 0; i < nsegs; i++, prod = RL_TX_DESC_NXT(sc, prod)) {
2857 		desc = &sc->rl_ldata.rl_tx_list[prod];
2858 		desc->rl_vlanctl = htole32(vlanctl);
2859 		desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[i].ds_addr));
2860 		desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[i].ds_addr));
2861 		cmdstat = segs[i].ds_len;
2862 		if (i != 0)
2863 			cmdstat |= RL_TDESC_CMD_OWN;
2864 		if (prod == sc->rl_ldata.rl_tx_desc_cnt - 1)
2865 			cmdstat |= RL_TDESC_CMD_EOR;
2866 		desc->rl_cmdstat = htole32(cmdstat | csum_flags);
2867 		sc->rl_ldata.rl_tx_free--;
2868 	}
2869 	/* Update producer index. */
2870 	sc->rl_ldata.rl_tx_prodidx = prod;
2871 
2872 	/* Set EOF on the last descriptor. */
2873 	ei = RL_TX_DESC_PRV(sc, prod);
2874 	desc = &sc->rl_ldata.rl_tx_list[ei];
2875 	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_EOF);
2876 
2877 	desc = &sc->rl_ldata.rl_tx_list[si];
2878 	/* Set SOF and transfer ownership of packet to the chip. */
2879 	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_OWN | RL_TDESC_CMD_SOF);
2880 
2881 	/*
2882 	 * Insure that the map for this transmission
2883 	 * is placed at the array index of the last descriptor
2884 	 * in this chain.  (Swap last and first dmamaps.)
2885 	 */
2886 	txd_last = &sc->rl_ldata.rl_tx_desc[ei];
2887 	map = txd->tx_dmamap;
2888 	txd->tx_dmamap = txd_last->tx_dmamap;
2889 	txd_last->tx_dmamap = map;
2890 	txd_last->tx_m = *m_head;
2891 
2892 	return (0);
2893 }
2894 
2895 static void
2896 re_start(struct ifnet *ifp)
2897 {
2898 	struct rl_softc		*sc;
2899 
2900 	sc = ifp->if_softc;
2901 	RL_LOCK(sc);
2902 	re_start_locked(ifp);
2903 	RL_UNLOCK(sc);
2904 }
2905 
2906 /*
2907  * Main transmit routine for C+ and gigE NICs.
2908  */
2909 static void
2910 re_start_locked(struct ifnet *ifp)
2911 {
2912 	struct rl_softc		*sc;
2913 	struct mbuf		*m_head;
2914 	int			queued;
2915 
2916 	sc = ifp->if_softc;
2917 
2918 #ifdef DEV_NETMAP
2919 	/* XXX is this necessary ? */
2920 	if (ifp->if_capenable & IFCAP_NETMAP) {
2921 		struct netmap_kring *kring = &NA(ifp)->tx_rings[0];
2922 		if (sc->rl_ldata.rl_tx_prodidx != kring->nr_hwcur) {
2923 			/* kick the tx unit */
2924 			CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2925 #ifdef RE_TX_MODERATION
2926 			CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2927 #endif
2928 			sc->rl_watchdog_timer = 5;
2929 		}
2930 		return;
2931 	}
2932 #endif /* DEV_NETMAP */
2933 	if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
2934 	    IFF_DRV_RUNNING || (sc->rl_flags & RL_FLAG_LINK) == 0)
2935 		return;
2936 
2937 	for (queued = 0; !IFQ_DRV_IS_EMPTY(&ifp->if_snd) &&
2938 	    sc->rl_ldata.rl_tx_free > 1;) {
2939 		IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
2940 		if (m_head == NULL)
2941 			break;
2942 
2943 		if (re_encap(sc, &m_head) != 0) {
2944 			if (m_head == NULL)
2945 				break;
2946 			IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
2947 			ifp->if_drv_flags |= IFF_DRV_OACTIVE;
2948 			break;
2949 		}
2950 
2951 		/*
2952 		 * If there's a BPF listener, bounce a copy of this frame
2953 		 * to him.
2954 		 */
2955 		ETHER_BPF_MTAP(ifp, m_head);
2956 
2957 		queued++;
2958 	}
2959 
2960 	if (queued == 0) {
2961 #ifdef RE_TX_MODERATION
2962 		if (sc->rl_ldata.rl_tx_free != sc->rl_ldata.rl_tx_desc_cnt)
2963 			CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2964 #endif
2965 		return;
2966 	}
2967 
2968 	/* Flush the TX descriptors */
2969 
2970 	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2971 	    sc->rl_ldata.rl_tx_list_map,
2972 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2973 
2974 	CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2975 
2976 #ifdef RE_TX_MODERATION
2977 	/*
2978 	 * Use the countdown timer for interrupt moderation.
2979 	 * 'TX done' interrupts are disabled. Instead, we reset the
2980 	 * countdown timer, which will begin counting until it hits
2981 	 * the value in the TIMERINT register, and then trigger an
2982 	 * interrupt. Each time we write to the TIMERCNT register,
2983 	 * the timer count is reset to 0.
2984 	 */
2985 	CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2986 #endif
2987 
2988 	/*
2989 	 * Set a timeout in case the chip goes out to lunch.
2990 	 */
2991 	sc->rl_watchdog_timer = 5;
2992 }
2993 
2994 static void
2995 re_set_jumbo(struct rl_softc *sc, int jumbo)
2996 {
2997 
2998 	if (sc->rl_hwrev->rl_rev == RL_HWREV_8168E_VL) {
2999 		pci_set_max_read_req(sc->rl_dev, 4096);
3000 		return;
3001 	}
3002 
3003 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG);
3004 	if (jumbo != 0) {
3005 		CSR_WRITE_1(sc, sc->rl_cfg3, CSR_READ_1(sc, sc->rl_cfg3) |
3006 		    RL_CFG3_JUMBO_EN0);
3007 		switch (sc->rl_hwrev->rl_rev) {
3008 		case RL_HWREV_8168DP:
3009 			break;
3010 		case RL_HWREV_8168E:
3011 			CSR_WRITE_1(sc, sc->rl_cfg4,
3012 			    CSR_READ_1(sc, sc->rl_cfg4) | 0x01);
3013 			break;
3014 		default:
3015 			CSR_WRITE_1(sc, sc->rl_cfg4,
3016 			    CSR_READ_1(sc, sc->rl_cfg4) | RL_CFG4_JUMBO_EN1);
3017 		}
3018 	} else {
3019 		CSR_WRITE_1(sc, sc->rl_cfg3, CSR_READ_1(sc, sc->rl_cfg3) &
3020 		    ~RL_CFG3_JUMBO_EN0);
3021 		switch (sc->rl_hwrev->rl_rev) {
3022 		case RL_HWREV_8168DP:
3023 			break;
3024 		case RL_HWREV_8168E:
3025 			CSR_WRITE_1(sc, sc->rl_cfg4,
3026 			    CSR_READ_1(sc, sc->rl_cfg4) & ~0x01);
3027 			break;
3028 		default:
3029 			CSR_WRITE_1(sc, sc->rl_cfg4,
3030 			    CSR_READ_1(sc, sc->rl_cfg4) & ~RL_CFG4_JUMBO_EN1);
3031 		}
3032 	}
3033 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3034 
3035 	switch (sc->rl_hwrev->rl_rev) {
3036 	case RL_HWREV_8168DP:
3037 		pci_set_max_read_req(sc->rl_dev, 4096);
3038 		break;
3039 	default:
3040 		if (jumbo != 0)
3041 			pci_set_max_read_req(sc->rl_dev, 512);
3042 		else
3043 			pci_set_max_read_req(sc->rl_dev, 4096);
3044 	}
3045 }
3046 
3047 static void
3048 re_init(void *xsc)
3049 {
3050 	struct rl_softc		*sc = xsc;
3051 
3052 	RL_LOCK(sc);
3053 	re_init_locked(sc);
3054 	RL_UNLOCK(sc);
3055 }
3056 
3057 static void
3058 re_init_locked(struct rl_softc *sc)
3059 {
3060 	struct ifnet		*ifp = sc->rl_ifp;
3061 	struct mii_data		*mii;
3062 	uint32_t		reg;
3063 	uint16_t		cfg;
3064 	union {
3065 		uint32_t align_dummy;
3066 		u_char eaddr[ETHER_ADDR_LEN];
3067         } eaddr;
3068 
3069 	RL_LOCK_ASSERT(sc);
3070 
3071 	mii = device_get_softc(sc->rl_miibus);
3072 
3073 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
3074 		return;
3075 
3076 	/*
3077 	 * Cancel pending I/O and free all RX/TX buffers.
3078 	 */
3079 	re_stop(sc);
3080 
3081 	/* Put controller into known state. */
3082 	re_reset(sc);
3083 
3084 	/*
3085 	 * For C+ mode, initialize the RX descriptors and mbufs.
3086 	 */
3087 	if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
3088 		if (ifp->if_mtu > RL_MTU) {
3089 			if (re_jrx_list_init(sc) != 0) {
3090 				device_printf(sc->rl_dev,
3091 				    "no memory for jumbo RX buffers\n");
3092 				re_stop(sc);
3093 				return;
3094 			}
3095 			/* Disable checksum offloading for jumbo frames. */
3096 			ifp->if_capenable &= ~(IFCAP_HWCSUM | IFCAP_TSO4);
3097 			ifp->if_hwassist &= ~(RE_CSUM_FEATURES | CSUM_TSO);
3098 		} else {
3099 			if (re_rx_list_init(sc) != 0) {
3100 				device_printf(sc->rl_dev,
3101 				    "no memory for RX buffers\n");
3102 				re_stop(sc);
3103 				return;
3104 			}
3105 		}
3106 		re_set_jumbo(sc, ifp->if_mtu > RL_MTU);
3107 	} else {
3108 		if (re_rx_list_init(sc) != 0) {
3109 			device_printf(sc->rl_dev, "no memory for RX buffers\n");
3110 			re_stop(sc);
3111 			return;
3112 		}
3113 		if ((sc->rl_flags & RL_FLAG_PCIE) != 0 &&
3114 		    pci_get_device(sc->rl_dev) != RT_DEVICEID_8101E) {
3115 			if (ifp->if_mtu > RL_MTU)
3116 				pci_set_max_read_req(sc->rl_dev, 512);
3117 			else
3118 				pci_set_max_read_req(sc->rl_dev, 4096);
3119 		}
3120 	}
3121 	re_tx_list_init(sc);
3122 
3123 	/*
3124 	 * Enable C+ RX and TX mode, as well as VLAN stripping and
3125 	 * RX checksum offload. We must configure the C+ register
3126 	 * before all others.
3127 	 */
3128 	cfg = RL_CPLUSCMD_PCI_MRW;
3129 	if ((ifp->if_capenable & IFCAP_RXCSUM) != 0)
3130 		cfg |= RL_CPLUSCMD_RXCSUM_ENB;
3131 	if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0)
3132 		cfg |= RL_CPLUSCMD_VLANSTRIP;
3133 	if ((sc->rl_flags & RL_FLAG_MACSTAT) != 0) {
3134 		cfg |= RL_CPLUSCMD_MACSTAT_DIS;
3135 		/* XXX magic. */
3136 		cfg |= 0x0001;
3137 	} else
3138 		cfg |= RL_CPLUSCMD_RXENB | RL_CPLUSCMD_TXENB;
3139 	CSR_WRITE_2(sc, RL_CPLUS_CMD, cfg);
3140 	if (sc->rl_hwrev->rl_rev == RL_HWREV_8169_8110SC ||
3141 	    sc->rl_hwrev->rl_rev == RL_HWREV_8169_8110SCE) {
3142 		reg = 0x000fff00;
3143 		if ((CSR_READ_1(sc, sc->rl_cfg2) & RL_CFG2_PCI66MHZ) != 0)
3144 			reg |= 0x000000ff;
3145 		if (sc->rl_hwrev->rl_rev == RL_HWREV_8169_8110SCE)
3146 			reg |= 0x00f00000;
3147 		CSR_WRITE_4(sc, 0x7c, reg);
3148 		/* Disable interrupt mitigation. */
3149 		CSR_WRITE_2(sc, 0xe2, 0);
3150 	}
3151 	/*
3152 	 * Disable TSO if interface MTU size is greater than MSS
3153 	 * allowed in controller.
3154 	 */
3155 	if (ifp->if_mtu > RL_TSO_MTU && (ifp->if_capenable & IFCAP_TSO4) != 0) {
3156 		ifp->if_capenable &= ~IFCAP_TSO4;
3157 		ifp->if_hwassist &= ~CSUM_TSO;
3158 	}
3159 
3160 	/*
3161 	 * Init our MAC address.  Even though the chipset
3162 	 * documentation doesn't mention it, we need to enter "Config
3163 	 * register write enable" mode to modify the ID registers.
3164 	 */
3165 	/* Copy MAC address on stack to align. */
3166 	bcopy(IF_LLADDR(ifp), eaddr.eaddr, ETHER_ADDR_LEN);
3167 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG);
3168 	CSR_WRITE_4(sc, RL_IDR0,
3169 	    htole32(*(u_int32_t *)(&eaddr.eaddr[0])));
3170 	CSR_WRITE_4(sc, RL_IDR4,
3171 	    htole32(*(u_int32_t *)(&eaddr.eaddr[4])));
3172 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3173 
3174 	/*
3175 	 * Load the addresses of the RX and TX lists into the chip.
3176 	 */
3177 
3178 	CSR_WRITE_4(sc, RL_RXLIST_ADDR_HI,
3179 	    RL_ADDR_HI(sc->rl_ldata.rl_rx_list_addr));
3180 	CSR_WRITE_4(sc, RL_RXLIST_ADDR_LO,
3181 	    RL_ADDR_LO(sc->rl_ldata.rl_rx_list_addr));
3182 
3183 	CSR_WRITE_4(sc, RL_TXLIST_ADDR_HI,
3184 	    RL_ADDR_HI(sc->rl_ldata.rl_tx_list_addr));
3185 	CSR_WRITE_4(sc, RL_TXLIST_ADDR_LO,
3186 	    RL_ADDR_LO(sc->rl_ldata.rl_tx_list_addr));
3187 
3188 	if ((sc->rl_flags & RL_FLAG_RXDV_GATED) != 0)
3189 		CSR_WRITE_4(sc, RL_MISC, CSR_READ_4(sc, RL_MISC) &
3190 		    ~0x00080000);
3191 
3192 	/*
3193 	 * Enable transmit and receive.
3194 	 */
3195 	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
3196 
3197 	/*
3198 	 * Set the initial TX configuration.
3199 	 */
3200 	if (sc->rl_testmode) {
3201 		if (sc->rl_type == RL_8169)
3202 			CSR_WRITE_4(sc, RL_TXCFG,
3203 			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON);
3204 		else
3205 			CSR_WRITE_4(sc, RL_TXCFG,
3206 			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON_CPLUS);
3207 	} else
3208 		CSR_WRITE_4(sc, RL_TXCFG, RL_TXCFG_CONFIG);
3209 
3210 	CSR_WRITE_1(sc, RL_EARLY_TX_THRESH, 16);
3211 
3212 	/*
3213 	 * Set the initial RX configuration.
3214 	 */
3215 	re_set_rxmode(sc);
3216 
3217 	/* Configure interrupt moderation. */
3218 	if (sc->rl_type == RL_8169) {
3219 		/* Magic from vendor. */
3220 		CSR_WRITE_2(sc, RL_INTRMOD, 0x5100);
3221 	}
3222 
3223 #ifdef DEVICE_POLLING
3224 	/*
3225 	 * Disable interrupts if we are polling.
3226 	 */
3227 	if (ifp->if_capenable & IFCAP_POLLING)
3228 		CSR_WRITE_2(sc, RL_IMR, 0);
3229 	else	/* otherwise ... */
3230 #endif
3231 
3232 	/*
3233 	 * Enable interrupts.
3234 	 */
3235 	if (sc->rl_testmode)
3236 		CSR_WRITE_2(sc, RL_IMR, 0);
3237 	else
3238 		CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
3239 	CSR_WRITE_2(sc, RL_ISR, RL_INTRS_CPLUS);
3240 
3241 	/* Set initial TX threshold */
3242 	sc->rl_txthresh = RL_TX_THRESH_INIT;
3243 
3244 	/* Start RX/TX process. */
3245 	CSR_WRITE_4(sc, RL_MISSEDPKT, 0);
3246 #ifdef notdef
3247 	/* Enable receiver and transmitter. */
3248 	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
3249 #endif
3250 
3251 	/*
3252 	 * Initialize the timer interrupt register so that
3253 	 * a timer interrupt will be generated once the timer
3254 	 * reaches a certain number of ticks. The timer is
3255 	 * reloaded on each transmit.
3256 	 */
3257 #ifdef RE_TX_MODERATION
3258 	/*
3259 	 * Use timer interrupt register to moderate TX interrupt
3260 	 * moderation, which dramatically improves TX frame rate.
3261 	 */
3262 	if (sc->rl_type == RL_8169)
3263 		CSR_WRITE_4(sc, RL_TIMERINT_8169, 0x800);
3264 	else
3265 		CSR_WRITE_4(sc, RL_TIMERINT, 0x400);
3266 #else
3267 	/*
3268 	 * Use timer interrupt register to moderate RX interrupt
3269 	 * moderation.
3270 	 */
3271 	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) != 0 &&
3272 	    intr_filter == 0) {
3273 		if (sc->rl_type == RL_8169)
3274 			CSR_WRITE_4(sc, RL_TIMERINT_8169,
3275 			    RL_USECS(sc->rl_int_rx_mod));
3276 	} else {
3277 		if (sc->rl_type == RL_8169)
3278 			CSR_WRITE_4(sc, RL_TIMERINT_8169, RL_USECS(0));
3279 	}
3280 #endif
3281 
3282 	/*
3283 	 * For 8169 gigE NICs, set the max allowed RX packet
3284 	 * size so we can receive jumbo frames.
3285 	 */
3286 	if (sc->rl_type == RL_8169) {
3287 		if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
3288 			/*
3289 			 * For controllers that use new jumbo frame scheme,
3290 			 * set maximum size of jumbo frame depending on
3291 			 * controller revisions.
3292 			 */
3293 			if (ifp->if_mtu > RL_MTU)
3294 				CSR_WRITE_2(sc, RL_MAXRXPKTLEN,
3295 				    sc->rl_hwrev->rl_max_mtu +
3296 				    ETHER_VLAN_ENCAP_LEN + ETHER_HDR_LEN +
3297 				    ETHER_CRC_LEN);
3298 			else
3299 				CSR_WRITE_2(sc, RL_MAXRXPKTLEN,
3300 				    RE_RX_DESC_BUFLEN);
3301 		} else if ((sc->rl_flags & RL_FLAG_PCIE) != 0 &&
3302 		    sc->rl_hwrev->rl_max_mtu == RL_MTU) {
3303 			/* RTL810x has no jumbo frame support. */
3304 			CSR_WRITE_2(sc, RL_MAXRXPKTLEN, RE_RX_DESC_BUFLEN);
3305 		} else
3306 			CSR_WRITE_2(sc, RL_MAXRXPKTLEN, 16383);
3307 	}
3308 
3309 	if (sc->rl_testmode)
3310 		return;
3311 
3312 	CSR_WRITE_1(sc, sc->rl_cfg1, CSR_READ_1(sc, sc->rl_cfg1) |
3313 	    RL_CFG1_DRVLOAD);
3314 
3315 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
3316 	ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
3317 
3318 	sc->rl_flags &= ~RL_FLAG_LINK;
3319 	mii_mediachg(mii);
3320 
3321 	sc->rl_watchdog_timer = 0;
3322 	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
3323 }
3324 
3325 /*
3326  * Set media options.
3327  */
3328 static int
3329 re_ifmedia_upd(struct ifnet *ifp)
3330 {
3331 	struct rl_softc		*sc;
3332 	struct mii_data		*mii;
3333 	int			error;
3334 
3335 	sc = ifp->if_softc;
3336 	mii = device_get_softc(sc->rl_miibus);
3337 	RL_LOCK(sc);
3338 	error = mii_mediachg(mii);
3339 	RL_UNLOCK(sc);
3340 
3341 	return (error);
3342 }
3343 
3344 /*
3345  * Report current media status.
3346  */
3347 static void
3348 re_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
3349 {
3350 	struct rl_softc		*sc;
3351 	struct mii_data		*mii;
3352 
3353 	sc = ifp->if_softc;
3354 	mii = device_get_softc(sc->rl_miibus);
3355 
3356 	RL_LOCK(sc);
3357 	mii_pollstat(mii);
3358 	ifmr->ifm_active = mii->mii_media_active;
3359 	ifmr->ifm_status = mii->mii_media_status;
3360 	RL_UNLOCK(sc);
3361 }
3362 
3363 static int
3364 re_ioctl(struct ifnet *ifp, u_long command, caddr_t data)
3365 {
3366 	struct rl_softc		*sc = ifp->if_softc;
3367 	struct ifreq		*ifr = (struct ifreq *) data;
3368 	struct mii_data		*mii;
3369 	int			error = 0;
3370 
3371 	switch (command) {
3372 	case SIOCSIFMTU:
3373 		if (ifr->ifr_mtu < ETHERMIN ||
3374 		    ifr->ifr_mtu > sc->rl_hwrev->rl_max_mtu ||
3375 		    ((sc->rl_flags & RL_FLAG_FASTETHER) != 0 &&
3376 		    ifr->ifr_mtu > RL_MTU)) {
3377 			error = EINVAL;
3378 			break;
3379 		}
3380 		RL_LOCK(sc);
3381 		if (ifp->if_mtu != ifr->ifr_mtu) {
3382 			ifp->if_mtu = ifr->ifr_mtu;
3383 			if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0 &&
3384 			    (ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
3385 				ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3386 				re_init_locked(sc);
3387 			}
3388 			if (ifp->if_mtu > RL_TSO_MTU &&
3389 			    (ifp->if_capenable & IFCAP_TSO4) != 0) {
3390 				ifp->if_capenable &= ~(IFCAP_TSO4 |
3391 				    IFCAP_VLAN_HWTSO);
3392 				ifp->if_hwassist &= ~CSUM_TSO;
3393 			}
3394 			VLAN_CAPABILITIES(ifp);
3395 		}
3396 		RL_UNLOCK(sc);
3397 		break;
3398 	case SIOCSIFFLAGS:
3399 		RL_LOCK(sc);
3400 		if ((ifp->if_flags & IFF_UP) != 0) {
3401 			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
3402 				if (((ifp->if_flags ^ sc->rl_if_flags)
3403 				    & (IFF_PROMISC | IFF_ALLMULTI)) != 0)
3404 					re_set_rxmode(sc);
3405 			} else
3406 				re_init_locked(sc);
3407 		} else {
3408 			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
3409 				re_stop(sc);
3410 		}
3411 		sc->rl_if_flags = ifp->if_flags;
3412 		RL_UNLOCK(sc);
3413 		break;
3414 	case SIOCADDMULTI:
3415 	case SIOCDELMULTI:
3416 		RL_LOCK(sc);
3417 		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
3418 			re_set_rxmode(sc);
3419 		RL_UNLOCK(sc);
3420 		break;
3421 	case SIOCGIFMEDIA:
3422 	case SIOCSIFMEDIA:
3423 		mii = device_get_softc(sc->rl_miibus);
3424 		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, command);
3425 		break;
3426 	case SIOCSIFCAP:
3427 	    {
3428 		int mask, reinit;
3429 
3430 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
3431 		reinit = 0;
3432 #ifdef DEVICE_POLLING
3433 		if (mask & IFCAP_POLLING) {
3434 			if (ifr->ifr_reqcap & IFCAP_POLLING) {
3435 				error = ether_poll_register(re_poll, ifp);
3436 				if (error)
3437 					return (error);
3438 				RL_LOCK(sc);
3439 				/* Disable interrupts */
3440 				CSR_WRITE_2(sc, RL_IMR, 0x0000);
3441 				ifp->if_capenable |= IFCAP_POLLING;
3442 				RL_UNLOCK(sc);
3443 			} else {
3444 				error = ether_poll_deregister(ifp);
3445 				/* Enable interrupts. */
3446 				RL_LOCK(sc);
3447 				CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
3448 				ifp->if_capenable &= ~IFCAP_POLLING;
3449 				RL_UNLOCK(sc);
3450 			}
3451 		}
3452 #endif /* DEVICE_POLLING */
3453 		RL_LOCK(sc);
3454 		if ((mask & IFCAP_TXCSUM) != 0 &&
3455 		    (ifp->if_capabilities & IFCAP_TXCSUM) != 0) {
3456 			ifp->if_capenable ^= IFCAP_TXCSUM;
3457 			if ((ifp->if_capenable & IFCAP_TXCSUM) != 0)
3458 				ifp->if_hwassist |= RE_CSUM_FEATURES;
3459 			else
3460 				ifp->if_hwassist &= ~RE_CSUM_FEATURES;
3461 			reinit = 1;
3462 		}
3463 		if ((mask & IFCAP_RXCSUM) != 0 &&
3464 		    (ifp->if_capabilities & IFCAP_RXCSUM) != 0) {
3465 			ifp->if_capenable ^= IFCAP_RXCSUM;
3466 			reinit = 1;
3467 		}
3468 		if ((mask & IFCAP_TSO4) != 0 &&
3469 		    (ifp->if_capabilities & IFCAP_TSO4) != 0) {
3470 			ifp->if_capenable ^= IFCAP_TSO4;
3471 			if ((IFCAP_TSO4 & ifp->if_capenable) != 0)
3472 				ifp->if_hwassist |= CSUM_TSO;
3473 			else
3474 				ifp->if_hwassist &= ~CSUM_TSO;
3475 			if (ifp->if_mtu > RL_TSO_MTU &&
3476 			    (ifp->if_capenable & IFCAP_TSO4) != 0) {
3477 				ifp->if_capenable &= ~IFCAP_TSO4;
3478 				ifp->if_hwassist &= ~CSUM_TSO;
3479 			}
3480 		}
3481 		if ((mask & IFCAP_VLAN_HWTSO) != 0 &&
3482 		    (ifp->if_capabilities & IFCAP_VLAN_HWTSO) != 0)
3483 			ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
3484 		if ((mask & IFCAP_VLAN_HWTAGGING) != 0 &&
3485 		    (ifp->if_capabilities & IFCAP_VLAN_HWTAGGING) != 0) {
3486 			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
3487 			/* TSO over VLAN requires VLAN hardware tagging. */
3488 			if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) == 0)
3489 				ifp->if_capenable &= ~IFCAP_VLAN_HWTSO;
3490 			reinit = 1;
3491 		}
3492 		if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0 &&
3493 		    (mask & (IFCAP_HWCSUM | IFCAP_TSO4 |
3494 		    IFCAP_VLAN_HWTSO)) != 0)
3495 				reinit = 1;
3496 		if ((mask & IFCAP_WOL) != 0 &&
3497 		    (ifp->if_capabilities & IFCAP_WOL) != 0) {
3498 			if ((mask & IFCAP_WOL_UCAST) != 0)
3499 				ifp->if_capenable ^= IFCAP_WOL_UCAST;
3500 			if ((mask & IFCAP_WOL_MCAST) != 0)
3501 				ifp->if_capenable ^= IFCAP_WOL_MCAST;
3502 			if ((mask & IFCAP_WOL_MAGIC) != 0)
3503 				ifp->if_capenable ^= IFCAP_WOL_MAGIC;
3504 		}
3505 		if (reinit && ifp->if_drv_flags & IFF_DRV_RUNNING) {
3506 			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3507 			re_init_locked(sc);
3508 		}
3509 		RL_UNLOCK(sc);
3510 		VLAN_CAPABILITIES(ifp);
3511 	    }
3512 		break;
3513 	default:
3514 		error = ether_ioctl(ifp, command, data);
3515 		break;
3516 	}
3517 
3518 	return (error);
3519 }
3520 
3521 static void
3522 re_watchdog(struct rl_softc *sc)
3523 {
3524 	struct ifnet		*ifp;
3525 
3526 	RL_LOCK_ASSERT(sc);
3527 
3528 	if (sc->rl_watchdog_timer == 0 || --sc->rl_watchdog_timer != 0)
3529 		return;
3530 
3531 	ifp = sc->rl_ifp;
3532 	re_txeof(sc);
3533 	if (sc->rl_ldata.rl_tx_free == sc->rl_ldata.rl_tx_desc_cnt) {
3534 		if_printf(ifp, "watchdog timeout (missed Tx interrupts) "
3535 		    "-- recovering\n");
3536 		if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
3537 			re_start_locked(ifp);
3538 		return;
3539 	}
3540 
3541 	if_printf(ifp, "watchdog timeout\n");
3542 	ifp->if_oerrors++;
3543 
3544 	re_rxeof(sc, NULL);
3545 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3546 	re_init_locked(sc);
3547 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
3548 		re_start_locked(ifp);
3549 }
3550 
3551 /*
3552  * Stop the adapter and free any mbufs allocated to the
3553  * RX and TX lists.
3554  */
3555 static void
3556 re_stop(struct rl_softc *sc)
3557 {
3558 	int			i;
3559 	struct ifnet		*ifp;
3560 	struct rl_txdesc	*txd;
3561 	struct rl_rxdesc	*rxd;
3562 
3563 	RL_LOCK_ASSERT(sc);
3564 
3565 	ifp = sc->rl_ifp;
3566 
3567 	sc->rl_watchdog_timer = 0;
3568 	callout_stop(&sc->rl_stat_callout);
3569 	ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
3570 
3571 	/*
3572 	 * Disable accepting frames to put RX MAC into idle state.
3573 	 * Otherwise it's possible to get frames while stop command
3574 	 * execution is in progress and controller can DMA the frame
3575 	 * to already freed RX buffer during that period.
3576 	 */
3577 	CSR_WRITE_4(sc, RL_RXCFG, CSR_READ_4(sc, RL_RXCFG) &
3578 	    ~(RL_RXCFG_RX_ALLPHYS | RL_RXCFG_RX_INDIV | RL_RXCFG_RX_MULTI |
3579 	    RL_RXCFG_RX_BROAD));
3580 
3581 	if ((sc->rl_flags & RL_FLAG_WAIT_TXPOLL) != 0) {
3582 		for (i = RL_TIMEOUT; i > 0; i--) {
3583 			if ((CSR_READ_1(sc, sc->rl_txstart) &
3584 			    RL_TXSTART_START) == 0)
3585 				break;
3586 			DELAY(20);
3587 		}
3588 		if (i == 0)
3589 			device_printf(sc->rl_dev,
3590 			    "stopping TX poll timed out!\n");
3591 		CSR_WRITE_1(sc, RL_COMMAND, 0x00);
3592 	} else if ((sc->rl_flags & RL_FLAG_CMDSTOP) != 0) {
3593 		CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_STOPREQ | RL_CMD_TX_ENB |
3594 		    RL_CMD_RX_ENB);
3595 		if ((sc->rl_flags & RL_FLAG_CMDSTOP_WAIT_TXQ) != 0) {
3596 			for (i = RL_TIMEOUT; i > 0; i--) {
3597 				if ((CSR_READ_4(sc, RL_TXCFG) &
3598 				    RL_TXCFG_QUEUE_EMPTY) != 0)
3599 					break;
3600 				DELAY(100);
3601 			}
3602 			if (i == 0)
3603 				device_printf(sc->rl_dev,
3604 				   "stopping TXQ timed out!\n");
3605 		}
3606 	} else
3607 		CSR_WRITE_1(sc, RL_COMMAND, 0x00);
3608 	DELAY(1000);
3609 	CSR_WRITE_2(sc, RL_IMR, 0x0000);
3610 	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
3611 
3612 	if (sc->rl_head != NULL) {
3613 		m_freem(sc->rl_head);
3614 		sc->rl_head = sc->rl_tail = NULL;
3615 	}
3616 
3617 	/* Free the TX list buffers. */
3618 	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
3619 		txd = &sc->rl_ldata.rl_tx_desc[i];
3620 		if (txd->tx_m != NULL) {
3621 			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
3622 			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
3623 			bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag,
3624 			    txd->tx_dmamap);
3625 			m_freem(txd->tx_m);
3626 			txd->tx_m = NULL;
3627 		}
3628 	}
3629 
3630 	/* Free the RX list buffers. */
3631 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
3632 		rxd = &sc->rl_ldata.rl_rx_desc[i];
3633 		if (rxd->rx_m != NULL) {
3634 			bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag,
3635 			    rxd->rx_dmamap, BUS_DMASYNC_POSTREAD);
3636 			bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag,
3637 			    rxd->rx_dmamap);
3638 			m_freem(rxd->rx_m);
3639 			rxd->rx_m = NULL;
3640 		}
3641 	}
3642 
3643 	if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
3644 		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
3645 			rxd = &sc->rl_ldata.rl_jrx_desc[i];
3646 			if (rxd->rx_m != NULL) {
3647 				bus_dmamap_sync(sc->rl_ldata.rl_jrx_mtag,
3648 				    rxd->rx_dmamap, BUS_DMASYNC_POSTREAD);
3649 				bus_dmamap_unload(sc->rl_ldata.rl_jrx_mtag,
3650 				    rxd->rx_dmamap);
3651 				m_freem(rxd->rx_m);
3652 				rxd->rx_m = NULL;
3653 			}
3654 		}
3655 	}
3656 }
3657 
3658 /*
3659  * Device suspend routine.  Stop the interface and save some PCI
3660  * settings in case the BIOS doesn't restore them properly on
3661  * resume.
3662  */
3663 static int
3664 re_suspend(device_t dev)
3665 {
3666 	struct rl_softc		*sc;
3667 
3668 	sc = device_get_softc(dev);
3669 
3670 	RL_LOCK(sc);
3671 	re_stop(sc);
3672 	re_setwol(sc);
3673 	sc->suspended = 1;
3674 	RL_UNLOCK(sc);
3675 
3676 	return (0);
3677 }
3678 
3679 /*
3680  * Device resume routine.  Restore some PCI settings in case the BIOS
3681  * doesn't, re-enable busmastering, and restart the interface if
3682  * appropriate.
3683  */
3684 static int
3685 re_resume(device_t dev)
3686 {
3687 	struct rl_softc		*sc;
3688 	struct ifnet		*ifp;
3689 
3690 	sc = device_get_softc(dev);
3691 
3692 	RL_LOCK(sc);
3693 
3694 	ifp = sc->rl_ifp;
3695 	/* Take controller out of sleep mode. */
3696 	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
3697 		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
3698 			CSR_WRITE_1(sc, RL_GPIO,
3699 			    CSR_READ_1(sc, RL_GPIO) | 0x01);
3700 	}
3701 
3702 	/*
3703 	 * Clear WOL matching such that normal Rx filtering
3704 	 * wouldn't interfere with WOL patterns.
3705 	 */
3706 	re_clrwol(sc);
3707 
3708 	/* reinitialize interface if necessary */
3709 	if (ifp->if_flags & IFF_UP)
3710 		re_init_locked(sc);
3711 
3712 	sc->suspended = 0;
3713 	RL_UNLOCK(sc);
3714 
3715 	return (0);
3716 }
3717 
3718 /*
3719  * Stop all chip I/O so that the kernel's probe routines don't
3720  * get confused by errant DMAs when rebooting.
3721  */
3722 static int
3723 re_shutdown(device_t dev)
3724 {
3725 	struct rl_softc		*sc;
3726 
3727 	sc = device_get_softc(dev);
3728 
3729 	RL_LOCK(sc);
3730 	re_stop(sc);
3731 	/*
3732 	 * Mark interface as down since otherwise we will panic if
3733 	 * interrupt comes in later on, which can happen in some
3734 	 * cases.
3735 	 */
3736 	sc->rl_ifp->if_flags &= ~IFF_UP;
3737 	re_setwol(sc);
3738 	RL_UNLOCK(sc);
3739 
3740 	return (0);
3741 }
3742 
3743 static void
3744 re_set_linkspeed(struct rl_softc *sc)
3745 {
3746 	struct mii_softc *miisc;
3747 	struct mii_data *mii;
3748 	int aneg, i, phyno;
3749 
3750 	RL_LOCK_ASSERT(sc);
3751 
3752 	mii = device_get_softc(sc->rl_miibus);
3753 	mii_pollstat(mii);
3754 	aneg = 0;
3755 	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
3756 	    (IFM_ACTIVE | IFM_AVALID)) {
3757 		switch IFM_SUBTYPE(mii->mii_media_active) {
3758 		case IFM_10_T:
3759 		case IFM_100_TX:
3760 			return;
3761 		case IFM_1000_T:
3762 			aneg++;
3763 			break;
3764 		default:
3765 			break;
3766 		}
3767 	}
3768 	miisc = LIST_FIRST(&mii->mii_phys);
3769 	phyno = miisc->mii_phy;
3770 	LIST_FOREACH(miisc, &mii->mii_phys, mii_list)
3771 		PHY_RESET(miisc);
3772 	re_miibus_writereg(sc->rl_dev, phyno, MII_100T2CR, 0);
3773 	re_miibus_writereg(sc->rl_dev, phyno,
3774 	    MII_ANAR, ANAR_TX_FD | ANAR_TX | ANAR_10_FD | ANAR_10 | ANAR_CSMA);
3775 	re_miibus_writereg(sc->rl_dev, phyno,
3776 	    MII_BMCR, BMCR_AUTOEN | BMCR_STARTNEG);
3777 	DELAY(1000);
3778 	if (aneg != 0) {
3779 		/*
3780 		 * Poll link state until re(4) get a 10/100Mbps link.
3781 		 */
3782 		for (i = 0; i < MII_ANEGTICKS_GIGE; i++) {
3783 			mii_pollstat(mii);
3784 			if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID))
3785 			    == (IFM_ACTIVE | IFM_AVALID)) {
3786 				switch (IFM_SUBTYPE(mii->mii_media_active)) {
3787 				case IFM_10_T:
3788 				case IFM_100_TX:
3789 					return;
3790 				default:
3791 					break;
3792 				}
3793 			}
3794 			RL_UNLOCK(sc);
3795 			pause("relnk", hz);
3796 			RL_LOCK(sc);
3797 		}
3798 		if (i == MII_ANEGTICKS_GIGE)
3799 			device_printf(sc->rl_dev,
3800 			    "establishing a link failed, WOL may not work!");
3801 	}
3802 	/*
3803 	 * No link, force MAC to have 100Mbps, full-duplex link.
3804 	 * MAC does not require reprogramming on resolved speed/duplex,
3805 	 * so this is just for completeness.
3806 	 */
3807 	mii->mii_media_status = IFM_AVALID | IFM_ACTIVE;
3808 	mii->mii_media_active = IFM_ETHER | IFM_100_TX | IFM_FDX;
3809 }
3810 
3811 static void
3812 re_setwol(struct rl_softc *sc)
3813 {
3814 	struct ifnet		*ifp;
3815 	int			pmc;
3816 	uint16_t		pmstat;
3817 	uint8_t			v;
3818 
3819 	RL_LOCK_ASSERT(sc);
3820 
3821 	if (pci_find_cap(sc->rl_dev, PCIY_PMG, &pmc) != 0)
3822 		return;
3823 
3824 	ifp = sc->rl_ifp;
3825 	/* Put controller into sleep mode. */
3826 	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
3827 		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
3828 			CSR_WRITE_1(sc, RL_GPIO,
3829 			    CSR_READ_1(sc, RL_GPIO) & ~0x01);
3830 	}
3831 	if ((ifp->if_capenable & IFCAP_WOL) != 0) {
3832 		re_set_rxmode(sc);
3833 		if ((sc->rl_flags & RL_FLAG_WOL_MANLINK) != 0)
3834 			re_set_linkspeed(sc);
3835 		if ((sc->rl_flags & RL_FLAG_WOLRXENB) != 0)
3836 			CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RX_ENB);
3837 	}
3838 	/* Enable config register write. */
3839 	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
3840 
3841 	/* Enable PME. */
3842 	v = CSR_READ_1(sc, sc->rl_cfg1);
3843 	v &= ~RL_CFG1_PME;
3844 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3845 		v |= RL_CFG1_PME;
3846 	CSR_WRITE_1(sc, sc->rl_cfg1, v);
3847 
3848 	v = CSR_READ_1(sc, sc->rl_cfg3);
3849 	v &= ~(RL_CFG3_WOL_LINK | RL_CFG3_WOL_MAGIC);
3850 	if ((ifp->if_capenable & IFCAP_WOL_MAGIC) != 0)
3851 		v |= RL_CFG3_WOL_MAGIC;
3852 	CSR_WRITE_1(sc, sc->rl_cfg3, v);
3853 
3854 	v = CSR_READ_1(sc, sc->rl_cfg5);
3855 	v &= ~(RL_CFG5_WOL_BCAST | RL_CFG5_WOL_MCAST | RL_CFG5_WOL_UCAST |
3856 	    RL_CFG5_WOL_LANWAKE);
3857 	if ((ifp->if_capenable & IFCAP_WOL_UCAST) != 0)
3858 		v |= RL_CFG5_WOL_UCAST;
3859 	if ((ifp->if_capenable & IFCAP_WOL_MCAST) != 0)
3860 		v |= RL_CFG5_WOL_MCAST | RL_CFG5_WOL_BCAST;
3861 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3862 		v |= RL_CFG5_WOL_LANWAKE;
3863 	CSR_WRITE_1(sc, sc->rl_cfg5, v);
3864 
3865 	/* Config register write done. */
3866 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3867 
3868 	if ((ifp->if_capenable & IFCAP_WOL) == 0 &&
3869 	    (sc->rl_flags & RL_FLAG_PHYWAKE_PM) != 0)
3870 		CSR_WRITE_1(sc, RL_PMCH, CSR_READ_1(sc, RL_PMCH) & ~0x80);
3871 	/*
3872 	 * It seems that hardware resets its link speed to 100Mbps in
3873 	 * power down mode so switching to 100Mbps in driver is not
3874 	 * needed.
3875 	 */
3876 
3877 	/* Request PME if WOL is requested. */
3878 	pmstat = pci_read_config(sc->rl_dev, pmc + PCIR_POWER_STATUS, 2);
3879 	pmstat &= ~(PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE);
3880 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3881 		pmstat |= PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE;
3882 	pci_write_config(sc->rl_dev, pmc + PCIR_POWER_STATUS, pmstat, 2);
3883 }
3884 
3885 static void
3886 re_clrwol(struct rl_softc *sc)
3887 {
3888 	int			pmc;
3889 	uint8_t			v;
3890 
3891 	RL_LOCK_ASSERT(sc);
3892 
3893 	if (pci_find_cap(sc->rl_dev, PCIY_PMG, &pmc) != 0)
3894 		return;
3895 
3896 	/* Enable config register write. */
3897 	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
3898 
3899 	v = CSR_READ_1(sc, sc->rl_cfg3);
3900 	v &= ~(RL_CFG3_WOL_LINK | RL_CFG3_WOL_MAGIC);
3901 	CSR_WRITE_1(sc, sc->rl_cfg3, v);
3902 
3903 	/* Config register write done. */
3904 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3905 
3906 	v = CSR_READ_1(sc, sc->rl_cfg5);
3907 	v &= ~(RL_CFG5_WOL_BCAST | RL_CFG5_WOL_MCAST | RL_CFG5_WOL_UCAST);
3908 	v &= ~RL_CFG5_WOL_LANWAKE;
3909 	CSR_WRITE_1(sc, sc->rl_cfg5, v);
3910 }
3911 
3912 static void
3913 re_add_sysctls(struct rl_softc *sc)
3914 {
3915 	struct sysctl_ctx_list	*ctx;
3916 	struct sysctl_oid_list	*children;
3917 	int			error;
3918 
3919 	ctx = device_get_sysctl_ctx(sc->rl_dev);
3920 	children = SYSCTL_CHILDREN(device_get_sysctl_tree(sc->rl_dev));
3921 
3922 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "stats",
3923 	    CTLTYPE_INT | CTLFLAG_RW, sc, 0, re_sysctl_stats, "I",
3924 	    "Statistics Information");
3925 	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) == 0)
3926 		return;
3927 
3928 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "int_rx_mod",
3929 	    CTLTYPE_INT | CTLFLAG_RW, &sc->rl_int_rx_mod, 0,
3930 	    sysctl_hw_re_int_mod, "I", "re RX interrupt moderation");
3931 	/* Pull in device tunables. */
3932 	sc->rl_int_rx_mod = RL_TIMER_DEFAULT;
3933 	error = resource_int_value(device_get_name(sc->rl_dev),
3934 	    device_get_unit(sc->rl_dev), "int_rx_mod", &sc->rl_int_rx_mod);
3935 	if (error == 0) {
3936 		if (sc->rl_int_rx_mod < RL_TIMER_MIN ||
3937 		    sc->rl_int_rx_mod > RL_TIMER_MAX) {
3938 			device_printf(sc->rl_dev, "int_rx_mod value out of "
3939 			    "range; using default: %d\n",
3940 			    RL_TIMER_DEFAULT);
3941 			sc->rl_int_rx_mod = RL_TIMER_DEFAULT;
3942 		}
3943 	}
3944 
3945 }
3946 
3947 static int
3948 re_sysctl_stats(SYSCTL_HANDLER_ARGS)
3949 {
3950 	struct rl_softc		*sc;
3951 	struct rl_stats		*stats;
3952 	int			error, i, result;
3953 
3954 	result = -1;
3955 	error = sysctl_handle_int(oidp, &result, 0, req);
3956 	if (error || req->newptr == NULL)
3957 		return (error);
3958 
3959 	if (result == 1) {
3960 		sc = (struct rl_softc *)arg1;
3961 		RL_LOCK(sc);
3962 		if ((sc->rl_ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
3963 			RL_UNLOCK(sc);
3964 			goto done;
3965 		}
3966 		bus_dmamap_sync(sc->rl_ldata.rl_stag,
3967 		    sc->rl_ldata.rl_smap, BUS_DMASYNC_PREREAD);
3968 		CSR_WRITE_4(sc, RL_DUMPSTATS_HI,
3969 		    RL_ADDR_HI(sc->rl_ldata.rl_stats_addr));
3970 		CSR_WRITE_4(sc, RL_DUMPSTATS_LO,
3971 		    RL_ADDR_LO(sc->rl_ldata.rl_stats_addr));
3972 		CSR_WRITE_4(sc, RL_DUMPSTATS_LO,
3973 		    RL_ADDR_LO(sc->rl_ldata.rl_stats_addr |
3974 		    RL_DUMPSTATS_START));
3975 		for (i = RL_TIMEOUT; i > 0; i--) {
3976 			if ((CSR_READ_4(sc, RL_DUMPSTATS_LO) &
3977 			    RL_DUMPSTATS_START) == 0)
3978 				break;
3979 			DELAY(1000);
3980 		}
3981 		bus_dmamap_sync(sc->rl_ldata.rl_stag,
3982 		    sc->rl_ldata.rl_smap, BUS_DMASYNC_POSTREAD);
3983 		RL_UNLOCK(sc);
3984 		if (i == 0) {
3985 			device_printf(sc->rl_dev,
3986 			    "DUMP statistics request timed out\n");
3987 			return (ETIMEDOUT);
3988 		}
3989 done:
3990 		stats = sc->rl_ldata.rl_stats;
3991 		printf("%s statistics:\n", device_get_nameunit(sc->rl_dev));
3992 		printf("Tx frames : %ju\n",
3993 		    (uintmax_t)le64toh(stats->rl_tx_pkts));
3994 		printf("Rx frames : %ju\n",
3995 		    (uintmax_t)le64toh(stats->rl_rx_pkts));
3996 		printf("Tx errors : %ju\n",
3997 		    (uintmax_t)le64toh(stats->rl_tx_errs));
3998 		printf("Rx errors : %u\n",
3999 		    le32toh(stats->rl_rx_errs));
4000 		printf("Rx missed frames : %u\n",
4001 		    (uint32_t)le16toh(stats->rl_missed_pkts));
4002 		printf("Rx frame alignment errs : %u\n",
4003 		    (uint32_t)le16toh(stats->rl_rx_framealign_errs));
4004 		printf("Tx single collisions : %u\n",
4005 		    le32toh(stats->rl_tx_onecoll));
4006 		printf("Tx multiple collisions : %u\n",
4007 		    le32toh(stats->rl_tx_multicolls));
4008 		printf("Rx unicast frames : %ju\n",
4009 		    (uintmax_t)le64toh(stats->rl_rx_ucasts));
4010 		printf("Rx broadcast frames : %ju\n",
4011 		    (uintmax_t)le64toh(stats->rl_rx_bcasts));
4012 		printf("Rx multicast frames : %u\n",
4013 		    le32toh(stats->rl_rx_mcasts));
4014 		printf("Tx aborts : %u\n",
4015 		    (uint32_t)le16toh(stats->rl_tx_aborts));
4016 		printf("Tx underruns : %u\n",
4017 		    (uint32_t)le16toh(stats->rl_rx_underruns));
4018 	}
4019 
4020 	return (error);
4021 }
4022 
4023 static int
4024 sysctl_int_range(SYSCTL_HANDLER_ARGS, int low, int high)
4025 {
4026 	int error, value;
4027 
4028 	if (arg1 == NULL)
4029 		return (EINVAL);
4030 	value = *(int *)arg1;
4031 	error = sysctl_handle_int(oidp, &value, 0, req);
4032 	if (error || req->newptr == NULL)
4033 		return (error);
4034 	if (value < low || value > high)
4035 		return (EINVAL);
4036 	*(int *)arg1 = value;
4037 
4038 	return (0);
4039 }
4040 
4041 static int
4042 sysctl_hw_re_int_mod(SYSCTL_HANDLER_ARGS)
4043 {
4044 
4045 	return (sysctl_int_range(oidp, arg1, arg2, req, RL_TIMER_MIN,
4046 	    RL_TIMER_MAX));
4047 }
4048