xref: /freebsd/sys/dev/re/if_re.c (revision faf139cc5dd3396181c11922bc6685c0c59b7b24)
1 /*-
2  * Copyright (c) 1997, 1998-2003
3  *	Bill Paul <wpaul@windriver.com>.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. All advertising materials mentioning features or use of this software
14  *    must display the following acknowledgement:
15  *	This product includes software developed by Bill Paul.
16  * 4. Neither the name of the author nor the names of any co-contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY Bill Paul AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL Bill Paul OR THE VOICES IN HIS HEAD
24  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
30  * THE POSSIBILITY OF SUCH DAMAGE.
31  */
32 
33 #include <sys/cdefs.h>
34 __FBSDID("$FreeBSD$");
35 
36 /*
37  * RealTek 8139C+/8169/8169S/8110S/8168/8111/8101E PCI NIC driver
38  *
39  * Written by Bill Paul <wpaul@windriver.com>
40  * Senior Networking Software Engineer
41  * Wind River Systems
42  */
43 
44 /*
45  * This driver is designed to support RealTek's next generation of
46  * 10/100 and 10/100/1000 PCI ethernet controllers. There are currently
47  * seven devices in this family: the RTL8139C+, the RTL8169, the RTL8169S,
48  * RTL8110S, the RTL8168, the RTL8111 and the RTL8101E.
49  *
50  * The 8139C+ is a 10/100 ethernet chip. It is backwards compatible
51  * with the older 8139 family, however it also supports a special
52  * C+ mode of operation that provides several new performance enhancing
53  * features. These include:
54  *
55  *	o Descriptor based DMA mechanism. Each descriptor represents
56  *	  a single packet fragment. Data buffers may be aligned on
57  *	  any byte boundary.
58  *
59  *	o 64-bit DMA
60  *
61  *	o TCP/IP checksum offload for both RX and TX
62  *
63  *	o High and normal priority transmit DMA rings
64  *
65  *	o VLAN tag insertion and extraction
66  *
67  *	o TCP large send (segmentation offload)
68  *
69  * Like the 8139, the 8139C+ also has a built-in 10/100 PHY. The C+
70  * programming API is fairly straightforward. The RX filtering, EEPROM
71  * access and PHY access is the same as it is on the older 8139 series
72  * chips.
73  *
74  * The 8169 is a 64-bit 10/100/1000 gigabit ethernet MAC. It has almost the
75  * same programming API and feature set as the 8139C+ with the following
76  * differences and additions:
77  *
78  *	o 1000Mbps mode
79  *
80  *	o Jumbo frames
81  *
82  *	o GMII and TBI ports/registers for interfacing with copper
83  *	  or fiber PHYs
84  *
85  *	o RX and TX DMA rings can have up to 1024 descriptors
86  *	  (the 8139C+ allows a maximum of 64)
87  *
88  *	o Slight differences in register layout from the 8139C+
89  *
90  * The TX start and timer interrupt registers are at different locations
91  * on the 8169 than they are on the 8139C+. Also, the status word in the
92  * RX descriptor has a slightly different bit layout. The 8169 does not
93  * have a built-in PHY. Most reference boards use a Marvell 88E1000 'Alaska'
94  * copper gigE PHY.
95  *
96  * The 8169S/8110S 10/100/1000 devices have built-in copper gigE PHYs
97  * (the 'S' stands for 'single-chip'). These devices have the same
98  * programming API as the older 8169, but also have some vendor-specific
99  * registers for the on-board PHY. The 8110S is a LAN-on-motherboard
100  * part designed to be pin-compatible with the RealTek 8100 10/100 chip.
101  *
102  * This driver takes advantage of the RX and TX checksum offload and
103  * VLAN tag insertion/extraction features. It also implements TX
104  * interrupt moderation using the timer interrupt registers, which
105  * significantly reduces TX interrupt load. There is also support
106  * for jumbo frames, however the 8169/8169S/8110S can not transmit
107  * jumbo frames larger than 7440, so the max MTU possible with this
108  * driver is 7422 bytes.
109  */
110 
111 #ifdef HAVE_KERNEL_OPTION_HEADERS
112 #include "opt_device_polling.h"
113 #endif
114 
115 #include <sys/param.h>
116 #include <sys/endian.h>
117 #include <sys/systm.h>
118 #include <sys/sockio.h>
119 #include <sys/mbuf.h>
120 #include <sys/malloc.h>
121 #include <sys/module.h>
122 #include <sys/kernel.h>
123 #include <sys/socket.h>
124 #include <sys/lock.h>
125 #include <sys/mutex.h>
126 #include <sys/sysctl.h>
127 #include <sys/taskqueue.h>
128 
129 #include <net/if.h>
130 #include <net/if_var.h>
131 #include <net/if_arp.h>
132 #include <net/ethernet.h>
133 #include <net/if_dl.h>
134 #include <net/if_media.h>
135 #include <net/if_types.h>
136 #include <net/if_vlan_var.h>
137 
138 #include <net/bpf.h>
139 
140 #include <machine/bus.h>
141 #include <machine/resource.h>
142 #include <sys/bus.h>
143 #include <sys/rman.h>
144 
145 #include <dev/mii/mii.h>
146 #include <dev/mii/miivar.h>
147 
148 #include <dev/pci/pcireg.h>
149 #include <dev/pci/pcivar.h>
150 
151 #include <dev/rl/if_rlreg.h>
152 
153 MODULE_DEPEND(re, pci, 1, 1, 1);
154 MODULE_DEPEND(re, ether, 1, 1, 1);
155 MODULE_DEPEND(re, miibus, 1, 1, 1);
156 
157 /* "device miibus" required.  See GENERIC if you get errors here. */
158 #include "miibus_if.h"
159 
160 /* Tunables. */
161 static int intr_filter = 0;
162 TUNABLE_INT("hw.re.intr_filter", &intr_filter);
163 static int msi_disable = 0;
164 TUNABLE_INT("hw.re.msi_disable", &msi_disable);
165 static int msix_disable = 0;
166 TUNABLE_INT("hw.re.msix_disable", &msix_disable);
167 static int prefer_iomap = 0;
168 TUNABLE_INT("hw.re.prefer_iomap", &prefer_iomap);
169 
170 #define RE_CSUM_FEATURES    (CSUM_IP | CSUM_TCP | CSUM_UDP)
171 
172 /*
173  * Various supported device vendors/types and their names.
174  */
175 static const struct rl_type re_devs[] = {
176 	{ DLINK_VENDORID, DLINK_DEVICEID_528T, 0,
177 	    "D-Link DGE-528(T) Gigabit Ethernet Adapter" },
178 	{ DLINK_VENDORID, DLINK_DEVICEID_530T_REVC, 0,
179 	    "D-Link DGE-530(T) Gigabit Ethernet Adapter" },
180 	{ RT_VENDORID, RT_DEVICEID_8139, 0,
181 	    "RealTek 8139C+ 10/100BaseTX" },
182 	{ RT_VENDORID, RT_DEVICEID_8101E, 0,
183 	    "RealTek 810xE PCIe 10/100baseTX" },
184 	{ RT_VENDORID, RT_DEVICEID_8168, 0,
185 	    "RealTek 8168/8111 B/C/CP/D/DP/E/F/G PCIe Gigabit Ethernet" },
186 	{ RT_VENDORID, RT_DEVICEID_8169, 0,
187 	    "RealTek 8169/8169S/8169SB(L)/8110S/8110SB(L) Gigabit Ethernet" },
188 	{ RT_VENDORID, RT_DEVICEID_8169SC, 0,
189 	    "RealTek 8169SC/8110SC Single-chip Gigabit Ethernet" },
190 	{ COREGA_VENDORID, COREGA_DEVICEID_CGLAPCIGT, 0,
191 	    "Corega CG-LAPCIGT (RTL8169S) Gigabit Ethernet" },
192 	{ LINKSYS_VENDORID, LINKSYS_DEVICEID_EG1032, 0,
193 	    "Linksys EG1032 (RTL8169S) Gigabit Ethernet" },
194 	{ USR_VENDORID, USR_DEVICEID_997902, 0,
195 	    "US Robotics 997902 (RTL8169S) Gigabit Ethernet" }
196 };
197 
198 static const struct rl_hwrev re_hwrevs[] = {
199 	{ RL_HWREV_8139, RL_8139, "", RL_MTU },
200 	{ RL_HWREV_8139A, RL_8139, "A", RL_MTU },
201 	{ RL_HWREV_8139AG, RL_8139, "A-G", RL_MTU },
202 	{ RL_HWREV_8139B, RL_8139, "B", RL_MTU },
203 	{ RL_HWREV_8130, RL_8139, "8130", RL_MTU },
204 	{ RL_HWREV_8139C, RL_8139, "C", RL_MTU },
205 	{ RL_HWREV_8139D, RL_8139, "8139D/8100B/8100C", RL_MTU },
206 	{ RL_HWREV_8139CPLUS, RL_8139CPLUS, "C+", RL_MTU },
207 	{ RL_HWREV_8168B_SPIN1, RL_8169, "8168", RL_JUMBO_MTU },
208 	{ RL_HWREV_8169, RL_8169, "8169", RL_JUMBO_MTU },
209 	{ RL_HWREV_8169S, RL_8169, "8169S", RL_JUMBO_MTU },
210 	{ RL_HWREV_8110S, RL_8169, "8110S", RL_JUMBO_MTU },
211 	{ RL_HWREV_8169_8110SB, RL_8169, "8169SB/8110SB", RL_JUMBO_MTU },
212 	{ RL_HWREV_8169_8110SC, RL_8169, "8169SC/8110SC", RL_JUMBO_MTU },
213 	{ RL_HWREV_8169_8110SBL, RL_8169, "8169SBL/8110SBL", RL_JUMBO_MTU },
214 	{ RL_HWREV_8169_8110SCE, RL_8169, "8169SC/8110SC", RL_JUMBO_MTU },
215 	{ RL_HWREV_8100, RL_8139, "8100", RL_MTU },
216 	{ RL_HWREV_8101, RL_8139, "8101", RL_MTU },
217 	{ RL_HWREV_8100E, RL_8169, "8100E", RL_MTU },
218 	{ RL_HWREV_8101E, RL_8169, "8101E", RL_MTU },
219 	{ RL_HWREV_8102E, RL_8169, "8102E", RL_MTU },
220 	{ RL_HWREV_8102EL, RL_8169, "8102EL", RL_MTU },
221 	{ RL_HWREV_8102EL_SPIN1, RL_8169, "8102EL", RL_MTU },
222 	{ RL_HWREV_8103E, RL_8169, "8103E", RL_MTU },
223 	{ RL_HWREV_8401E, RL_8169, "8401E", RL_MTU },
224 	{ RL_HWREV_8402, RL_8169, "8402", RL_MTU },
225 	{ RL_HWREV_8105E, RL_8169, "8105E", RL_MTU },
226 	{ RL_HWREV_8105E_SPIN1, RL_8169, "8105E", RL_MTU },
227 	{ RL_HWREV_8106E, RL_8169, "8106E", RL_MTU },
228 	{ RL_HWREV_8168B_SPIN2, RL_8169, "8168", RL_JUMBO_MTU },
229 	{ RL_HWREV_8168B_SPIN3, RL_8169, "8168", RL_JUMBO_MTU },
230 	{ RL_HWREV_8168C, RL_8169, "8168C/8111C", RL_JUMBO_MTU_6K },
231 	{ RL_HWREV_8168C_SPIN2, RL_8169, "8168C/8111C", RL_JUMBO_MTU_6K },
232 	{ RL_HWREV_8168CP, RL_8169, "8168CP/8111CP", RL_JUMBO_MTU_6K },
233 	{ RL_HWREV_8168D, RL_8169, "8168D/8111D", RL_JUMBO_MTU_9K },
234 	{ RL_HWREV_8168DP, RL_8169, "8168DP/8111DP", RL_JUMBO_MTU_9K },
235 	{ RL_HWREV_8168E, RL_8169, "8168E/8111E", RL_JUMBO_MTU_9K},
236 	{ RL_HWREV_8168E_VL, RL_8169, "8168E/8111E-VL", RL_JUMBO_MTU_6K},
237 	{ RL_HWREV_8168EP, RL_8169, "8168EP/8111EP", RL_JUMBO_MTU_9K},
238 	{ RL_HWREV_8168F, RL_8169, "8168F/8111F", RL_JUMBO_MTU_9K},
239 	{ RL_HWREV_8168G, RL_8169, "8168G/8111G", RL_JUMBO_MTU_9K},
240 	{ RL_HWREV_8168GU, RL_8169, "8168GU/8111GU", RL_JUMBO_MTU_9K},
241 	{ RL_HWREV_8168H, RL_8169, "8168H/8111H", RL_JUMBO_MTU_9K},
242 	{ RL_HWREV_8411, RL_8169, "8411", RL_JUMBO_MTU_9K},
243 	{ RL_HWREV_8411B, RL_8169, "8411B", RL_JUMBO_MTU_9K},
244 	{ 0, 0, NULL, 0 }
245 };
246 
247 static int re_probe		(device_t);
248 static int re_attach		(device_t);
249 static int re_detach		(device_t);
250 
251 static int re_encap		(struct rl_softc *, struct mbuf **);
252 
253 static void re_dma_map_addr	(void *, bus_dma_segment_t *, int, int);
254 static int re_allocmem		(device_t, struct rl_softc *);
255 static __inline void re_discard_rxbuf
256 				(struct rl_softc *, int);
257 static int re_newbuf		(struct rl_softc *, int);
258 static int re_jumbo_newbuf	(struct rl_softc *, int);
259 static int re_rx_list_init	(struct rl_softc *);
260 static int re_jrx_list_init	(struct rl_softc *);
261 static int re_tx_list_init	(struct rl_softc *);
262 #ifdef RE_FIXUP_RX
263 static __inline void re_fixup_rx
264 				(struct mbuf *);
265 #endif
266 static int re_rxeof		(struct rl_softc *, int *);
267 static void re_txeof		(struct rl_softc *);
268 #ifdef DEVICE_POLLING
269 static int re_poll		(struct ifnet *, enum poll_cmd, int);
270 static int re_poll_locked	(struct ifnet *, enum poll_cmd, int);
271 #endif
272 static int re_intr		(void *);
273 static void re_intr_msi		(void *);
274 static void re_tick		(void *);
275 static void re_int_task		(void *, int);
276 static void re_start		(struct ifnet *);
277 static void re_start_locked	(struct ifnet *);
278 static int re_ioctl		(struct ifnet *, u_long, caddr_t);
279 static void re_init		(void *);
280 static void re_init_locked	(struct rl_softc *);
281 static void re_stop		(struct rl_softc *);
282 static void re_watchdog		(struct rl_softc *);
283 static int re_suspend		(device_t);
284 static int re_resume		(device_t);
285 static int re_shutdown		(device_t);
286 static int re_ifmedia_upd	(struct ifnet *);
287 static void re_ifmedia_sts	(struct ifnet *, struct ifmediareq *);
288 
289 static void re_eeprom_putbyte	(struct rl_softc *, int);
290 static void re_eeprom_getword	(struct rl_softc *, int, u_int16_t *);
291 static void re_read_eeprom	(struct rl_softc *, caddr_t, int, int);
292 static int re_gmii_readreg	(device_t, int, int);
293 static int re_gmii_writereg	(device_t, int, int, int);
294 
295 static int re_miibus_readreg	(device_t, int, int);
296 static int re_miibus_writereg	(device_t, int, int, int);
297 static void re_miibus_statchg	(device_t);
298 
299 static void re_set_jumbo	(struct rl_softc *, int);
300 static void re_set_rxmode		(struct rl_softc *);
301 static void re_reset		(struct rl_softc *);
302 static void re_setwol		(struct rl_softc *);
303 static void re_clrwol		(struct rl_softc *);
304 static void re_set_linkspeed	(struct rl_softc *);
305 
306 #ifdef DEV_NETMAP	/* see ixgbe.c for details */
307 #include <dev/netmap/if_re_netmap.h>
308 MODULE_DEPEND(re, netmap, 1, 1, 1);
309 #endif /* !DEV_NETMAP */
310 
311 #ifdef RE_DIAG
312 static int re_diag		(struct rl_softc *);
313 #endif
314 
315 static void re_add_sysctls	(struct rl_softc *);
316 static int re_sysctl_stats	(SYSCTL_HANDLER_ARGS);
317 static int sysctl_int_range	(SYSCTL_HANDLER_ARGS, int, int);
318 static int sysctl_hw_re_int_mod	(SYSCTL_HANDLER_ARGS);
319 
320 static device_method_t re_methods[] = {
321 	/* Device interface */
322 	DEVMETHOD(device_probe,		re_probe),
323 	DEVMETHOD(device_attach,	re_attach),
324 	DEVMETHOD(device_detach,	re_detach),
325 	DEVMETHOD(device_suspend,	re_suspend),
326 	DEVMETHOD(device_resume,	re_resume),
327 	DEVMETHOD(device_shutdown,	re_shutdown),
328 
329 	/* MII interface */
330 	DEVMETHOD(miibus_readreg,	re_miibus_readreg),
331 	DEVMETHOD(miibus_writereg,	re_miibus_writereg),
332 	DEVMETHOD(miibus_statchg,	re_miibus_statchg),
333 
334 	DEVMETHOD_END
335 };
336 
337 static driver_t re_driver = {
338 	"re",
339 	re_methods,
340 	sizeof(struct rl_softc)
341 };
342 
343 static devclass_t re_devclass;
344 
345 DRIVER_MODULE(re, pci, re_driver, re_devclass, 0, 0);
346 DRIVER_MODULE(miibus, re, miibus_driver, miibus_devclass, 0, 0);
347 
348 #define EE_SET(x)					\
349 	CSR_WRITE_1(sc, RL_EECMD,			\
350 		CSR_READ_1(sc, RL_EECMD) | x)
351 
352 #define EE_CLR(x)					\
353 	CSR_WRITE_1(sc, RL_EECMD,			\
354 		CSR_READ_1(sc, RL_EECMD) & ~x)
355 
356 /*
357  * Send a read command and address to the EEPROM, check for ACK.
358  */
359 static void
360 re_eeprom_putbyte(struct rl_softc *sc, int addr)
361 {
362 	int			d, i;
363 
364 	d = addr | (RL_9346_READ << sc->rl_eewidth);
365 
366 	/*
367 	 * Feed in each bit and strobe the clock.
368 	 */
369 
370 	for (i = 1 << (sc->rl_eewidth + 3); i; i >>= 1) {
371 		if (d & i) {
372 			EE_SET(RL_EE_DATAIN);
373 		} else {
374 			EE_CLR(RL_EE_DATAIN);
375 		}
376 		DELAY(100);
377 		EE_SET(RL_EE_CLK);
378 		DELAY(150);
379 		EE_CLR(RL_EE_CLK);
380 		DELAY(100);
381 	}
382 }
383 
384 /*
385  * Read a word of data stored in the EEPROM at address 'addr.'
386  */
387 static void
388 re_eeprom_getword(struct rl_softc *sc, int addr, u_int16_t *dest)
389 {
390 	int			i;
391 	u_int16_t		word = 0;
392 
393 	/*
394 	 * Send address of word we want to read.
395 	 */
396 	re_eeprom_putbyte(sc, addr);
397 
398 	/*
399 	 * Start reading bits from EEPROM.
400 	 */
401 	for (i = 0x8000; i; i >>= 1) {
402 		EE_SET(RL_EE_CLK);
403 		DELAY(100);
404 		if (CSR_READ_1(sc, RL_EECMD) & RL_EE_DATAOUT)
405 			word |= i;
406 		EE_CLR(RL_EE_CLK);
407 		DELAY(100);
408 	}
409 
410 	*dest = word;
411 }
412 
413 /*
414  * Read a sequence of words from the EEPROM.
415  */
416 static void
417 re_read_eeprom(struct rl_softc *sc, caddr_t dest, int off, int cnt)
418 {
419 	int			i;
420 	u_int16_t		word = 0, *ptr;
421 
422 	CSR_SETBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM);
423 
424         DELAY(100);
425 
426 	for (i = 0; i < cnt; i++) {
427 		CSR_SETBIT_1(sc, RL_EECMD, RL_EE_SEL);
428 		re_eeprom_getword(sc, off + i, &word);
429 		CSR_CLRBIT_1(sc, RL_EECMD, RL_EE_SEL);
430 		ptr = (u_int16_t *)(dest + (i * 2));
431                 *ptr = word;
432 	}
433 
434 	CSR_CLRBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM);
435 }
436 
437 static int
438 re_gmii_readreg(device_t dev, int phy, int reg)
439 {
440 	struct rl_softc		*sc;
441 	u_int32_t		rval;
442 	int			i;
443 
444 	sc = device_get_softc(dev);
445 
446 	/* Let the rgephy driver read the GMEDIASTAT register */
447 
448 	if (reg == RL_GMEDIASTAT) {
449 		rval = CSR_READ_1(sc, RL_GMEDIASTAT);
450 		return (rval);
451 	}
452 
453 	CSR_WRITE_4(sc, RL_PHYAR, reg << 16);
454 
455 	for (i = 0; i < RL_PHY_TIMEOUT; i++) {
456 		rval = CSR_READ_4(sc, RL_PHYAR);
457 		if (rval & RL_PHYAR_BUSY)
458 			break;
459 		DELAY(25);
460 	}
461 
462 	if (i == RL_PHY_TIMEOUT) {
463 		device_printf(sc->rl_dev, "PHY read failed\n");
464 		return (0);
465 	}
466 
467 	/*
468 	 * Controller requires a 20us delay to process next MDIO request.
469 	 */
470 	DELAY(20);
471 
472 	return (rval & RL_PHYAR_PHYDATA);
473 }
474 
475 static int
476 re_gmii_writereg(device_t dev, int phy, int reg, int data)
477 {
478 	struct rl_softc		*sc;
479 	u_int32_t		rval;
480 	int			i;
481 
482 	sc = device_get_softc(dev);
483 
484 	CSR_WRITE_4(sc, RL_PHYAR, (reg << 16) |
485 	    (data & RL_PHYAR_PHYDATA) | RL_PHYAR_BUSY);
486 
487 	for (i = 0; i < RL_PHY_TIMEOUT; i++) {
488 		rval = CSR_READ_4(sc, RL_PHYAR);
489 		if (!(rval & RL_PHYAR_BUSY))
490 			break;
491 		DELAY(25);
492 	}
493 
494 	if (i == RL_PHY_TIMEOUT) {
495 		device_printf(sc->rl_dev, "PHY write failed\n");
496 		return (0);
497 	}
498 
499 	/*
500 	 * Controller requires a 20us delay to process next MDIO request.
501 	 */
502 	DELAY(20);
503 
504 	return (0);
505 }
506 
507 static int
508 re_miibus_readreg(device_t dev, int phy, int reg)
509 {
510 	struct rl_softc		*sc;
511 	u_int16_t		rval = 0;
512 	u_int16_t		re8139_reg = 0;
513 
514 	sc = device_get_softc(dev);
515 
516 	if (sc->rl_type == RL_8169) {
517 		rval = re_gmii_readreg(dev, phy, reg);
518 		return (rval);
519 	}
520 
521 	switch (reg) {
522 	case MII_BMCR:
523 		re8139_reg = RL_BMCR;
524 		break;
525 	case MII_BMSR:
526 		re8139_reg = RL_BMSR;
527 		break;
528 	case MII_ANAR:
529 		re8139_reg = RL_ANAR;
530 		break;
531 	case MII_ANER:
532 		re8139_reg = RL_ANER;
533 		break;
534 	case MII_ANLPAR:
535 		re8139_reg = RL_LPAR;
536 		break;
537 	case MII_PHYIDR1:
538 	case MII_PHYIDR2:
539 		return (0);
540 	/*
541 	 * Allow the rlphy driver to read the media status
542 	 * register. If we have a link partner which does not
543 	 * support NWAY, this is the register which will tell
544 	 * us the results of parallel detection.
545 	 */
546 	case RL_MEDIASTAT:
547 		rval = CSR_READ_1(sc, RL_MEDIASTAT);
548 		return (rval);
549 	default:
550 		device_printf(sc->rl_dev, "bad phy register\n");
551 		return (0);
552 	}
553 	rval = CSR_READ_2(sc, re8139_reg);
554 	if (sc->rl_type == RL_8139CPLUS && re8139_reg == RL_BMCR) {
555 		/* 8139C+ has different bit layout. */
556 		rval &= ~(BMCR_LOOP | BMCR_ISO);
557 	}
558 	return (rval);
559 }
560 
561 static int
562 re_miibus_writereg(device_t dev, int phy, int reg, int data)
563 {
564 	struct rl_softc		*sc;
565 	u_int16_t		re8139_reg = 0;
566 	int			rval = 0;
567 
568 	sc = device_get_softc(dev);
569 
570 	if (sc->rl_type == RL_8169) {
571 		rval = re_gmii_writereg(dev, phy, reg, data);
572 		return (rval);
573 	}
574 
575 	switch (reg) {
576 	case MII_BMCR:
577 		re8139_reg = RL_BMCR;
578 		if (sc->rl_type == RL_8139CPLUS) {
579 			/* 8139C+ has different bit layout. */
580 			data &= ~(BMCR_LOOP | BMCR_ISO);
581 		}
582 		break;
583 	case MII_BMSR:
584 		re8139_reg = RL_BMSR;
585 		break;
586 	case MII_ANAR:
587 		re8139_reg = RL_ANAR;
588 		break;
589 	case MII_ANER:
590 		re8139_reg = RL_ANER;
591 		break;
592 	case MII_ANLPAR:
593 		re8139_reg = RL_LPAR;
594 		break;
595 	case MII_PHYIDR1:
596 	case MII_PHYIDR2:
597 		return (0);
598 		break;
599 	default:
600 		device_printf(sc->rl_dev, "bad phy register\n");
601 		return (0);
602 	}
603 	CSR_WRITE_2(sc, re8139_reg, data);
604 	return (0);
605 }
606 
607 static void
608 re_miibus_statchg(device_t dev)
609 {
610 	struct rl_softc		*sc;
611 	struct ifnet		*ifp;
612 	struct mii_data		*mii;
613 
614 	sc = device_get_softc(dev);
615 	mii = device_get_softc(sc->rl_miibus);
616 	ifp = sc->rl_ifp;
617 	if (mii == NULL || ifp == NULL ||
618 	    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
619 		return;
620 
621 	sc->rl_flags &= ~RL_FLAG_LINK;
622 	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
623 	    (IFM_ACTIVE | IFM_AVALID)) {
624 		switch (IFM_SUBTYPE(mii->mii_media_active)) {
625 		case IFM_10_T:
626 		case IFM_100_TX:
627 			sc->rl_flags |= RL_FLAG_LINK;
628 			break;
629 		case IFM_1000_T:
630 			if ((sc->rl_flags & RL_FLAG_FASTETHER) != 0)
631 				break;
632 			sc->rl_flags |= RL_FLAG_LINK;
633 			break;
634 		default:
635 			break;
636 		}
637 	}
638 	/*
639 	 * RealTek controllers do not provide any interface to the RX/TX
640 	 * MACs for resolved speed, duplex and flow-control parameters.
641 	 */
642 }
643 
644 /*
645  * Set the RX configuration and 64-bit multicast hash filter.
646  */
647 static void
648 re_set_rxmode(struct rl_softc *sc)
649 {
650 	struct ifnet		*ifp;
651 	struct ifmultiaddr	*ifma;
652 	uint32_t		hashes[2] = { 0, 0 };
653 	uint32_t		h, rxfilt;
654 
655 	RL_LOCK_ASSERT(sc);
656 
657 	ifp = sc->rl_ifp;
658 
659 	rxfilt = RL_RXCFG_CONFIG | RL_RXCFG_RX_INDIV | RL_RXCFG_RX_BROAD;
660 	if ((sc->rl_flags & RL_FLAG_EARLYOFF) != 0)
661 		rxfilt |= RL_RXCFG_EARLYOFF;
662 	else if ((sc->rl_flags & RL_FLAG_8168G_PLUS) != 0)
663 		rxfilt |= RL_RXCFG_EARLYOFFV2;
664 
665 	if (ifp->if_flags & (IFF_ALLMULTI | IFF_PROMISC)) {
666 		if (ifp->if_flags & IFF_PROMISC)
667 			rxfilt |= RL_RXCFG_RX_ALLPHYS;
668 		/*
669 		 * Unlike other hardwares, we have to explicitly set
670 		 * RL_RXCFG_RX_MULTI to receive multicast frames in
671 		 * promiscuous mode.
672 		 */
673 		rxfilt |= RL_RXCFG_RX_MULTI;
674 		hashes[0] = hashes[1] = 0xffffffff;
675 		goto done;
676 	}
677 
678 	if_maddr_rlock(ifp);
679 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
680 		if (ifma->ifma_addr->sa_family != AF_LINK)
681 			continue;
682 		h = ether_crc32_be(LLADDR((struct sockaddr_dl *)
683 		    ifma->ifma_addr), ETHER_ADDR_LEN) >> 26;
684 		if (h < 32)
685 			hashes[0] |= (1 << h);
686 		else
687 			hashes[1] |= (1 << (h - 32));
688 	}
689 	if_maddr_runlock(ifp);
690 
691 	if (hashes[0] != 0 || hashes[1] != 0) {
692 		/*
693 		 * For some unfathomable reason, RealTek decided to
694 		 * reverse the order of the multicast hash registers
695 		 * in the PCI Express parts.  This means we have to
696 		 * write the hash pattern in reverse order for those
697 		 * devices.
698 		 */
699 		if ((sc->rl_flags & RL_FLAG_PCIE) != 0) {
700 			h = bswap32(hashes[0]);
701 			hashes[0] = bswap32(hashes[1]);
702 			hashes[1] = h;
703 		}
704 		rxfilt |= RL_RXCFG_RX_MULTI;
705 	}
706 
707 	if  (sc->rl_hwrev->rl_rev == RL_HWREV_8168F) {
708 		/* Disable multicast filtering due to silicon bug. */
709 		hashes[0] = 0xffffffff;
710 		hashes[1] = 0xffffffff;
711 	}
712 
713 done:
714 	CSR_WRITE_4(sc, RL_MAR0, hashes[0]);
715 	CSR_WRITE_4(sc, RL_MAR4, hashes[1]);
716 	CSR_WRITE_4(sc, RL_RXCFG, rxfilt);
717 }
718 
719 static void
720 re_reset(struct rl_softc *sc)
721 {
722 	int			i;
723 
724 	RL_LOCK_ASSERT(sc);
725 
726 	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RESET);
727 
728 	for (i = 0; i < RL_TIMEOUT; i++) {
729 		DELAY(10);
730 		if (!(CSR_READ_1(sc, RL_COMMAND) & RL_CMD_RESET))
731 			break;
732 	}
733 	if (i == RL_TIMEOUT)
734 		device_printf(sc->rl_dev, "reset never completed!\n");
735 
736 	if ((sc->rl_flags & RL_FLAG_MACRESET) != 0)
737 		CSR_WRITE_1(sc, 0x82, 1);
738 	if (sc->rl_hwrev->rl_rev == RL_HWREV_8169S)
739 		re_gmii_writereg(sc->rl_dev, 1, 0x0b, 0);
740 }
741 
742 #ifdef RE_DIAG
743 
744 /*
745  * The following routine is designed to test for a defect on some
746  * 32-bit 8169 cards. Some of these NICs have the REQ64# and ACK64#
747  * lines connected to the bus, however for a 32-bit only card, they
748  * should be pulled high. The result of this defect is that the
749  * NIC will not work right if you plug it into a 64-bit slot: DMA
750  * operations will be done with 64-bit transfers, which will fail
751  * because the 64-bit data lines aren't connected.
752  *
753  * There's no way to work around this (short of talking a soldering
754  * iron to the board), however we can detect it. The method we use
755  * here is to put the NIC into digital loopback mode, set the receiver
756  * to promiscuous mode, and then try to send a frame. We then compare
757  * the frame data we sent to what was received. If the data matches,
758  * then the NIC is working correctly, otherwise we know the user has
759  * a defective NIC which has been mistakenly plugged into a 64-bit PCI
760  * slot. In the latter case, there's no way the NIC can work correctly,
761  * so we print out a message on the console and abort the device attach.
762  */
763 
764 static int
765 re_diag(struct rl_softc *sc)
766 {
767 	struct ifnet		*ifp = sc->rl_ifp;
768 	struct mbuf		*m0;
769 	struct ether_header	*eh;
770 	struct rl_desc		*cur_rx;
771 	u_int16_t		status;
772 	u_int32_t		rxstat;
773 	int			total_len, i, error = 0, phyaddr;
774 	u_int8_t		dst[] = { 0x00, 'h', 'e', 'l', 'l', 'o' };
775 	u_int8_t		src[] = { 0x00, 'w', 'o', 'r', 'l', 'd' };
776 
777 	/* Allocate a single mbuf */
778 	MGETHDR(m0, M_NOWAIT, MT_DATA);
779 	if (m0 == NULL)
780 		return (ENOBUFS);
781 
782 	RL_LOCK(sc);
783 
784 	/*
785 	 * Initialize the NIC in test mode. This sets the chip up
786 	 * so that it can send and receive frames, but performs the
787 	 * following special functions:
788 	 * - Puts receiver in promiscuous mode
789 	 * - Enables digital loopback mode
790 	 * - Leaves interrupts turned off
791 	 */
792 
793 	ifp->if_flags |= IFF_PROMISC;
794 	sc->rl_testmode = 1;
795 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
796 	re_init_locked(sc);
797 	sc->rl_flags |= RL_FLAG_LINK;
798 	if (sc->rl_type == RL_8169)
799 		phyaddr = 1;
800 	else
801 		phyaddr = 0;
802 
803 	re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_RESET);
804 	for (i = 0; i < RL_TIMEOUT; i++) {
805 		status = re_miibus_readreg(sc->rl_dev, phyaddr, MII_BMCR);
806 		if (!(status & BMCR_RESET))
807 			break;
808 	}
809 
810 	re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_LOOP);
811 	CSR_WRITE_2(sc, RL_ISR, RL_INTRS);
812 
813 	DELAY(100000);
814 
815 	/* Put some data in the mbuf */
816 
817 	eh = mtod(m0, struct ether_header *);
818 	bcopy ((char *)&dst, eh->ether_dhost, ETHER_ADDR_LEN);
819 	bcopy ((char *)&src, eh->ether_shost, ETHER_ADDR_LEN);
820 	eh->ether_type = htons(ETHERTYPE_IP);
821 	m0->m_pkthdr.len = m0->m_len = ETHER_MIN_LEN - ETHER_CRC_LEN;
822 
823 	/*
824 	 * Queue the packet, start transmission.
825 	 * Note: IF_HANDOFF() ultimately calls re_start() for us.
826 	 */
827 
828 	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
829 	RL_UNLOCK(sc);
830 	/* XXX: re_diag must not be called when in ALTQ mode */
831 	IF_HANDOFF(&ifp->if_snd, m0, ifp);
832 	RL_LOCK(sc);
833 	m0 = NULL;
834 
835 	/* Wait for it to propagate through the chip */
836 
837 	DELAY(100000);
838 	for (i = 0; i < RL_TIMEOUT; i++) {
839 		status = CSR_READ_2(sc, RL_ISR);
840 		CSR_WRITE_2(sc, RL_ISR, status);
841 		if ((status & (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK)) ==
842 		    (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK))
843 			break;
844 		DELAY(10);
845 	}
846 
847 	if (i == RL_TIMEOUT) {
848 		device_printf(sc->rl_dev,
849 		    "diagnostic failed, failed to receive packet in"
850 		    " loopback mode\n");
851 		error = EIO;
852 		goto done;
853 	}
854 
855 	/*
856 	 * The packet should have been dumped into the first
857 	 * entry in the RX DMA ring. Grab it from there.
858 	 */
859 
860 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
861 	    sc->rl_ldata.rl_rx_list_map,
862 	    BUS_DMASYNC_POSTREAD);
863 	bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag,
864 	    sc->rl_ldata.rl_rx_desc[0].rx_dmamap,
865 	    BUS_DMASYNC_POSTREAD);
866 	bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag,
867 	    sc->rl_ldata.rl_rx_desc[0].rx_dmamap);
868 
869 	m0 = sc->rl_ldata.rl_rx_desc[0].rx_m;
870 	sc->rl_ldata.rl_rx_desc[0].rx_m = NULL;
871 	eh = mtod(m0, struct ether_header *);
872 
873 	cur_rx = &sc->rl_ldata.rl_rx_list[0];
874 	total_len = RL_RXBYTES(cur_rx);
875 	rxstat = le32toh(cur_rx->rl_cmdstat);
876 
877 	if (total_len != ETHER_MIN_LEN) {
878 		device_printf(sc->rl_dev,
879 		    "diagnostic failed, received short packet\n");
880 		error = EIO;
881 		goto done;
882 	}
883 
884 	/* Test that the received packet data matches what we sent. */
885 
886 	if (bcmp((char *)&eh->ether_dhost, (char *)&dst, ETHER_ADDR_LEN) ||
887 	    bcmp((char *)&eh->ether_shost, (char *)&src, ETHER_ADDR_LEN) ||
888 	    ntohs(eh->ether_type) != ETHERTYPE_IP) {
889 		device_printf(sc->rl_dev, "WARNING, DMA FAILURE!\n");
890 		device_printf(sc->rl_dev, "expected TX data: %6D/%6D/0x%x\n",
891 		    dst, ":", src, ":", ETHERTYPE_IP);
892 		device_printf(sc->rl_dev, "received RX data: %6D/%6D/0x%x\n",
893 		    eh->ether_dhost, ":", eh->ether_shost, ":",
894 		    ntohs(eh->ether_type));
895 		device_printf(sc->rl_dev, "You may have a defective 32-bit "
896 		    "NIC plugged into a 64-bit PCI slot.\n");
897 		device_printf(sc->rl_dev, "Please re-install the NIC in a "
898 		    "32-bit slot for proper operation.\n");
899 		device_printf(sc->rl_dev, "Read the re(4) man page for more "
900 		    "details.\n");
901 		error = EIO;
902 	}
903 
904 done:
905 	/* Turn interface off, release resources */
906 
907 	sc->rl_testmode = 0;
908 	sc->rl_flags &= ~RL_FLAG_LINK;
909 	ifp->if_flags &= ~IFF_PROMISC;
910 	re_stop(sc);
911 	if (m0 != NULL)
912 		m_freem(m0);
913 
914 	RL_UNLOCK(sc);
915 
916 	return (error);
917 }
918 
919 #endif
920 
921 /*
922  * Probe for a RealTek 8139C+/8169/8110 chip. Check the PCI vendor and device
923  * IDs against our list and return a device name if we find a match.
924  */
925 static int
926 re_probe(device_t dev)
927 {
928 	const struct rl_type	*t;
929 	uint16_t		devid, vendor;
930 	uint16_t		revid, sdevid;
931 	int			i;
932 
933 	vendor = pci_get_vendor(dev);
934 	devid = pci_get_device(dev);
935 	revid = pci_get_revid(dev);
936 	sdevid = pci_get_subdevice(dev);
937 
938 	if (vendor == LINKSYS_VENDORID && devid == LINKSYS_DEVICEID_EG1032) {
939 		if (sdevid != LINKSYS_SUBDEVICE_EG1032_REV3) {
940 			/*
941 			 * Only attach to rev. 3 of the Linksys EG1032 adapter.
942 			 * Rev. 2 is supported by sk(4).
943 			 */
944 			return (ENXIO);
945 		}
946 	}
947 
948 	if (vendor == RT_VENDORID && devid == RT_DEVICEID_8139) {
949 		if (revid != 0x20) {
950 			/* 8139, let rl(4) take care of this device. */
951 			return (ENXIO);
952 		}
953 	}
954 
955 	t = re_devs;
956 	for (i = 0; i < sizeof(re_devs) / sizeof(re_devs[0]); i++, t++) {
957 		if (vendor == t->rl_vid && devid == t->rl_did) {
958 			device_set_desc(dev, t->rl_name);
959 			return (BUS_PROBE_DEFAULT);
960 		}
961 	}
962 
963 	return (ENXIO);
964 }
965 
966 /*
967  * Map a single buffer address.
968  */
969 
970 static void
971 re_dma_map_addr(void *arg, bus_dma_segment_t *segs, int nseg, int error)
972 {
973 	bus_addr_t		*addr;
974 
975 	if (error)
976 		return;
977 
978 	KASSERT(nseg == 1, ("too many DMA segments, %d should be 1", nseg));
979 	addr = arg;
980 	*addr = segs->ds_addr;
981 }
982 
983 static int
984 re_allocmem(device_t dev, struct rl_softc *sc)
985 {
986 	bus_addr_t		lowaddr;
987 	bus_size_t		rx_list_size, tx_list_size;
988 	int			error;
989 	int			i;
990 
991 	rx_list_size = sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc);
992 	tx_list_size = sc->rl_ldata.rl_tx_desc_cnt * sizeof(struct rl_desc);
993 
994 	/*
995 	 * Allocate the parent bus DMA tag appropriate for PCI.
996 	 * In order to use DAC, RL_CPLUSCMD_PCI_DAC bit of RL_CPLUS_CMD
997 	 * register should be set. However some RealTek chips are known
998 	 * to be buggy on DAC handling, therefore disable DAC by limiting
999 	 * DMA address space to 32bit. PCIe variants of RealTek chips
1000 	 * may not have the limitation.
1001 	 */
1002 	lowaddr = BUS_SPACE_MAXADDR;
1003 	if ((sc->rl_flags & RL_FLAG_PCIE) == 0)
1004 		lowaddr = BUS_SPACE_MAXADDR_32BIT;
1005 	error = bus_dma_tag_create(bus_get_dma_tag(dev), 1, 0,
1006 	    lowaddr, BUS_SPACE_MAXADDR, NULL, NULL,
1007 	    BUS_SPACE_MAXSIZE_32BIT, 0, BUS_SPACE_MAXSIZE_32BIT, 0,
1008 	    NULL, NULL, &sc->rl_parent_tag);
1009 	if (error) {
1010 		device_printf(dev, "could not allocate parent DMA tag\n");
1011 		return (error);
1012 	}
1013 
1014 	/*
1015 	 * Allocate map for TX mbufs.
1016 	 */
1017 	error = bus_dma_tag_create(sc->rl_parent_tag, 1, 0,
1018 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL,
1019 	    NULL, MCLBYTES * RL_NTXSEGS, RL_NTXSEGS, 4096, 0,
1020 	    NULL, NULL, &sc->rl_ldata.rl_tx_mtag);
1021 	if (error) {
1022 		device_printf(dev, "could not allocate TX DMA tag\n");
1023 		return (error);
1024 	}
1025 
1026 	/*
1027 	 * Allocate map for RX mbufs.
1028 	 */
1029 
1030 	if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
1031 		error = bus_dma_tag_create(sc->rl_parent_tag, sizeof(uint64_t),
1032 		    0, BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
1033 		    MJUM9BYTES, 1, MJUM9BYTES, 0, NULL, NULL,
1034 		    &sc->rl_ldata.rl_jrx_mtag);
1035 		if (error) {
1036 			device_printf(dev,
1037 			    "could not allocate jumbo RX DMA tag\n");
1038 			return (error);
1039 		}
1040 	}
1041 	error = bus_dma_tag_create(sc->rl_parent_tag, sizeof(uint64_t), 0,
1042 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
1043 	    MCLBYTES, 1, MCLBYTES, 0, NULL, NULL, &sc->rl_ldata.rl_rx_mtag);
1044 	if (error) {
1045 		device_printf(dev, "could not allocate RX DMA tag\n");
1046 		return (error);
1047 	}
1048 
1049 	/*
1050 	 * Allocate map for TX descriptor list.
1051 	 */
1052 	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
1053 	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1054 	    NULL, tx_list_size, 1, tx_list_size, 0,
1055 	    NULL, NULL, &sc->rl_ldata.rl_tx_list_tag);
1056 	if (error) {
1057 		device_printf(dev, "could not allocate TX DMA ring tag\n");
1058 		return (error);
1059 	}
1060 
1061 	/* Allocate DMA'able memory for the TX ring */
1062 
1063 	error = bus_dmamem_alloc(sc->rl_ldata.rl_tx_list_tag,
1064 	    (void **)&sc->rl_ldata.rl_tx_list,
1065 	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1066 	    &sc->rl_ldata.rl_tx_list_map);
1067 	if (error) {
1068 		device_printf(dev, "could not allocate TX DMA ring\n");
1069 		return (error);
1070 	}
1071 
1072 	/* Load the map for the TX ring. */
1073 
1074 	sc->rl_ldata.rl_tx_list_addr = 0;
1075 	error = bus_dmamap_load(sc->rl_ldata.rl_tx_list_tag,
1076 	     sc->rl_ldata.rl_tx_list_map, sc->rl_ldata.rl_tx_list,
1077 	     tx_list_size, re_dma_map_addr,
1078 	     &sc->rl_ldata.rl_tx_list_addr, BUS_DMA_NOWAIT);
1079 	if (error != 0 || sc->rl_ldata.rl_tx_list_addr == 0) {
1080 		device_printf(dev, "could not load TX DMA ring\n");
1081 		return (ENOMEM);
1082 	}
1083 
1084 	/* Create DMA maps for TX buffers */
1085 
1086 	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
1087 		error = bus_dmamap_create(sc->rl_ldata.rl_tx_mtag, 0,
1088 		    &sc->rl_ldata.rl_tx_desc[i].tx_dmamap);
1089 		if (error) {
1090 			device_printf(dev, "could not create DMA map for TX\n");
1091 			return (error);
1092 		}
1093 	}
1094 
1095 	/*
1096 	 * Allocate map for RX descriptor list.
1097 	 */
1098 	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
1099 	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1100 	    NULL, rx_list_size, 1, rx_list_size, 0,
1101 	    NULL, NULL, &sc->rl_ldata.rl_rx_list_tag);
1102 	if (error) {
1103 		device_printf(dev, "could not create RX DMA ring tag\n");
1104 		return (error);
1105 	}
1106 
1107 	/* Allocate DMA'able memory for the RX ring */
1108 
1109 	error = bus_dmamem_alloc(sc->rl_ldata.rl_rx_list_tag,
1110 	    (void **)&sc->rl_ldata.rl_rx_list,
1111 	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1112 	    &sc->rl_ldata.rl_rx_list_map);
1113 	if (error) {
1114 		device_printf(dev, "could not allocate RX DMA ring\n");
1115 		return (error);
1116 	}
1117 
1118 	/* Load the map for the RX ring. */
1119 
1120 	sc->rl_ldata.rl_rx_list_addr = 0;
1121 	error = bus_dmamap_load(sc->rl_ldata.rl_rx_list_tag,
1122 	     sc->rl_ldata.rl_rx_list_map, sc->rl_ldata.rl_rx_list,
1123 	     rx_list_size, re_dma_map_addr,
1124 	     &sc->rl_ldata.rl_rx_list_addr, BUS_DMA_NOWAIT);
1125 	if (error != 0 || sc->rl_ldata.rl_rx_list_addr == 0) {
1126 		device_printf(dev, "could not load RX DMA ring\n");
1127 		return (ENOMEM);
1128 	}
1129 
1130 	/* Create DMA maps for RX buffers */
1131 
1132 	if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
1133 		error = bus_dmamap_create(sc->rl_ldata.rl_jrx_mtag, 0,
1134 		    &sc->rl_ldata.rl_jrx_sparemap);
1135 		if (error) {
1136 			device_printf(dev,
1137 			    "could not create spare DMA map for jumbo RX\n");
1138 			return (error);
1139 		}
1140 		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1141 			error = bus_dmamap_create(sc->rl_ldata.rl_jrx_mtag, 0,
1142 			    &sc->rl_ldata.rl_jrx_desc[i].rx_dmamap);
1143 			if (error) {
1144 				device_printf(dev,
1145 				    "could not create DMA map for jumbo RX\n");
1146 				return (error);
1147 			}
1148 		}
1149 	}
1150 	error = bus_dmamap_create(sc->rl_ldata.rl_rx_mtag, 0,
1151 	    &sc->rl_ldata.rl_rx_sparemap);
1152 	if (error) {
1153 		device_printf(dev, "could not create spare DMA map for RX\n");
1154 		return (error);
1155 	}
1156 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1157 		error = bus_dmamap_create(sc->rl_ldata.rl_rx_mtag, 0,
1158 		    &sc->rl_ldata.rl_rx_desc[i].rx_dmamap);
1159 		if (error) {
1160 			device_printf(dev, "could not create DMA map for RX\n");
1161 			return (error);
1162 		}
1163 	}
1164 
1165 	/* Create DMA map for statistics. */
1166 	error = bus_dma_tag_create(sc->rl_parent_tag, RL_DUMP_ALIGN, 0,
1167 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
1168 	    sizeof(struct rl_stats), 1, sizeof(struct rl_stats), 0, NULL, NULL,
1169 	    &sc->rl_ldata.rl_stag);
1170 	if (error) {
1171 		device_printf(dev, "could not create statistics DMA tag\n");
1172 		return (error);
1173 	}
1174 	/* Allocate DMA'able memory for statistics. */
1175 	error = bus_dmamem_alloc(sc->rl_ldata.rl_stag,
1176 	    (void **)&sc->rl_ldata.rl_stats,
1177 	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1178 	    &sc->rl_ldata.rl_smap);
1179 	if (error) {
1180 		device_printf(dev,
1181 		    "could not allocate statistics DMA memory\n");
1182 		return (error);
1183 	}
1184 	/* Load the map for statistics. */
1185 	sc->rl_ldata.rl_stats_addr = 0;
1186 	error = bus_dmamap_load(sc->rl_ldata.rl_stag, sc->rl_ldata.rl_smap,
1187 	    sc->rl_ldata.rl_stats, sizeof(struct rl_stats), re_dma_map_addr,
1188 	     &sc->rl_ldata.rl_stats_addr, BUS_DMA_NOWAIT);
1189 	if (error != 0 || sc->rl_ldata.rl_stats_addr == 0) {
1190 		device_printf(dev, "could not load statistics DMA memory\n");
1191 		return (ENOMEM);
1192 	}
1193 
1194 	return (0);
1195 }
1196 
1197 /*
1198  * Attach the interface. Allocate softc structures, do ifmedia
1199  * setup and ethernet/BPF attach.
1200  */
1201 static int
1202 re_attach(device_t dev)
1203 {
1204 	u_char			eaddr[ETHER_ADDR_LEN];
1205 	u_int16_t		as[ETHER_ADDR_LEN / 2];
1206 	struct rl_softc		*sc;
1207 	struct ifnet		*ifp;
1208 	const struct rl_hwrev	*hw_rev;
1209 	int			capmask, error = 0, hwrev, i, msic, msixc,
1210 				phy, reg, rid;
1211 	u_int32_t		cap, ctl;
1212 	u_int16_t		devid, re_did = 0;
1213 	uint8_t			cfg;
1214 
1215 	sc = device_get_softc(dev);
1216 	sc->rl_dev = dev;
1217 
1218 	mtx_init(&sc->rl_mtx, device_get_nameunit(dev), MTX_NETWORK_LOCK,
1219 	    MTX_DEF);
1220 	callout_init_mtx(&sc->rl_stat_callout, &sc->rl_mtx, 0);
1221 
1222 	/*
1223 	 * Map control/status registers.
1224 	 */
1225 	pci_enable_busmaster(dev);
1226 
1227 	devid = pci_get_device(dev);
1228 	/*
1229 	 * Prefer memory space register mapping over IO space.
1230 	 * Because RTL8169SC does not seem to work when memory mapping
1231 	 * is used always activate io mapping.
1232 	 */
1233 	if (devid == RT_DEVICEID_8169SC)
1234 		prefer_iomap = 1;
1235 	if (prefer_iomap == 0) {
1236 		sc->rl_res_id = PCIR_BAR(1);
1237 		sc->rl_res_type = SYS_RES_MEMORY;
1238 		/* RTL8168/8101E seems to use different BARs. */
1239 		if (devid == RT_DEVICEID_8168 || devid == RT_DEVICEID_8101E)
1240 			sc->rl_res_id = PCIR_BAR(2);
1241 	} else {
1242 		sc->rl_res_id = PCIR_BAR(0);
1243 		sc->rl_res_type = SYS_RES_IOPORT;
1244 	}
1245 	sc->rl_res = bus_alloc_resource_any(dev, sc->rl_res_type,
1246 	    &sc->rl_res_id, RF_ACTIVE);
1247 	if (sc->rl_res == NULL && prefer_iomap == 0) {
1248 		sc->rl_res_id = PCIR_BAR(0);
1249 		sc->rl_res_type = SYS_RES_IOPORT;
1250 		sc->rl_res = bus_alloc_resource_any(dev, sc->rl_res_type,
1251 		    &sc->rl_res_id, RF_ACTIVE);
1252 	}
1253 	if (sc->rl_res == NULL) {
1254 		device_printf(dev, "couldn't map ports/memory\n");
1255 		error = ENXIO;
1256 		goto fail;
1257 	}
1258 
1259 	sc->rl_btag = rman_get_bustag(sc->rl_res);
1260 	sc->rl_bhandle = rman_get_bushandle(sc->rl_res);
1261 
1262 	msic = pci_msi_count(dev);
1263 	msixc = pci_msix_count(dev);
1264 	if (pci_find_cap(dev, PCIY_EXPRESS, &reg) == 0) {
1265 		sc->rl_flags |= RL_FLAG_PCIE;
1266 		sc->rl_expcap = reg;
1267 	}
1268 	if (bootverbose) {
1269 		device_printf(dev, "MSI count : %d\n", msic);
1270 		device_printf(dev, "MSI-X count : %d\n", msixc);
1271 	}
1272 	if (msix_disable > 0)
1273 		msixc = 0;
1274 	if (msi_disable > 0)
1275 		msic = 0;
1276 	/* Prefer MSI-X to MSI. */
1277 	if (msixc > 0) {
1278 		msixc = RL_MSI_MESSAGES;
1279 		rid = PCIR_BAR(4);
1280 		sc->rl_res_pba = bus_alloc_resource_any(dev, SYS_RES_MEMORY,
1281 		    &rid, RF_ACTIVE);
1282 		if (sc->rl_res_pba == NULL) {
1283 			device_printf(sc->rl_dev,
1284 			    "could not allocate MSI-X PBA resource\n");
1285 		}
1286 		if (sc->rl_res_pba != NULL &&
1287 		    pci_alloc_msix(dev, &msixc) == 0) {
1288 			if (msixc == RL_MSI_MESSAGES) {
1289 				device_printf(dev, "Using %d MSI-X message\n",
1290 				    msixc);
1291 				sc->rl_flags |= RL_FLAG_MSIX;
1292 			} else
1293 				pci_release_msi(dev);
1294 		}
1295 		if ((sc->rl_flags & RL_FLAG_MSIX) == 0) {
1296 			if (sc->rl_res_pba != NULL)
1297 				bus_release_resource(dev, SYS_RES_MEMORY, rid,
1298 				    sc->rl_res_pba);
1299 			sc->rl_res_pba = NULL;
1300 			msixc = 0;
1301 		}
1302 	}
1303 	/* Prefer MSI to INTx. */
1304 	if (msixc == 0 && msic > 0) {
1305 		msic = RL_MSI_MESSAGES;
1306 		if (pci_alloc_msi(dev, &msic) == 0) {
1307 			if (msic == RL_MSI_MESSAGES) {
1308 				device_printf(dev, "Using %d MSI message\n",
1309 				    msic);
1310 				sc->rl_flags |= RL_FLAG_MSI;
1311 				/* Explicitly set MSI enable bit. */
1312 				CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1313 				cfg = CSR_READ_1(sc, RL_CFG2);
1314 				cfg |= RL_CFG2_MSI;
1315 				CSR_WRITE_1(sc, RL_CFG2, cfg);
1316 				CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1317 			} else
1318 				pci_release_msi(dev);
1319 		}
1320 		if ((sc->rl_flags & RL_FLAG_MSI) == 0)
1321 			msic = 0;
1322 	}
1323 
1324 	/* Allocate interrupt */
1325 	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) == 0) {
1326 		rid = 0;
1327 		sc->rl_irq[0] = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
1328 		    RF_SHAREABLE | RF_ACTIVE);
1329 		if (sc->rl_irq[0] == NULL) {
1330 			device_printf(dev, "couldn't allocate IRQ resources\n");
1331 			error = ENXIO;
1332 			goto fail;
1333 		}
1334 	} else {
1335 		for (i = 0, rid = 1; i < RL_MSI_MESSAGES; i++, rid++) {
1336 			sc->rl_irq[i] = bus_alloc_resource_any(dev,
1337 			    SYS_RES_IRQ, &rid, RF_ACTIVE);
1338 			if (sc->rl_irq[i] == NULL) {
1339 				device_printf(dev,
1340 				    "couldn't allocate IRQ resources for "
1341 				    "message %d\n", rid);
1342 				error = ENXIO;
1343 				goto fail;
1344 			}
1345 		}
1346 	}
1347 
1348 	if ((sc->rl_flags & RL_FLAG_MSI) == 0) {
1349 		CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1350 		cfg = CSR_READ_1(sc, RL_CFG2);
1351 		if ((cfg & RL_CFG2_MSI) != 0) {
1352 			device_printf(dev, "turning off MSI enable bit.\n");
1353 			cfg &= ~RL_CFG2_MSI;
1354 			CSR_WRITE_1(sc, RL_CFG2, cfg);
1355 		}
1356 		CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1357 	}
1358 
1359 	/* Disable ASPM L0S/L1. */
1360 	if (sc->rl_expcap != 0) {
1361 		cap = pci_read_config(dev, sc->rl_expcap +
1362 		    PCIER_LINK_CAP, 2);
1363 		if ((cap & PCIEM_LINK_CAP_ASPM) != 0) {
1364 			ctl = pci_read_config(dev, sc->rl_expcap +
1365 			    PCIER_LINK_CTL, 2);
1366 			if ((ctl & PCIEM_LINK_CTL_ASPMC) != 0) {
1367 				ctl &= ~PCIEM_LINK_CTL_ASPMC;
1368 				pci_write_config(dev, sc->rl_expcap +
1369 				    PCIER_LINK_CTL, ctl, 2);
1370 				device_printf(dev, "ASPM disabled\n");
1371 			}
1372 		} else
1373 			device_printf(dev, "no ASPM capability\n");
1374 	}
1375 
1376 	hw_rev = re_hwrevs;
1377 	hwrev = CSR_READ_4(sc, RL_TXCFG);
1378 	switch (hwrev & 0x70000000) {
1379 	case 0x00000000:
1380 	case 0x10000000:
1381 		device_printf(dev, "Chip rev. 0x%08x\n", hwrev & 0xfc800000);
1382 		hwrev &= (RL_TXCFG_HWREV | 0x80000000);
1383 		break;
1384 	default:
1385 		device_printf(dev, "Chip rev. 0x%08x\n", hwrev & 0x7c800000);
1386 		sc->rl_macrev = hwrev & 0x00700000;
1387 		hwrev &= RL_TXCFG_HWREV;
1388 		break;
1389 	}
1390 	device_printf(dev, "MAC rev. 0x%08x\n", sc->rl_macrev);
1391 	while (hw_rev->rl_desc != NULL) {
1392 		if (hw_rev->rl_rev == hwrev) {
1393 			sc->rl_type = hw_rev->rl_type;
1394 			sc->rl_hwrev = hw_rev;
1395 			break;
1396 		}
1397 		hw_rev++;
1398 	}
1399 	if (hw_rev->rl_desc == NULL) {
1400 		device_printf(dev, "Unknown H/W revision: 0x%08x\n", hwrev);
1401 		error = ENXIO;
1402 		goto fail;
1403 	}
1404 
1405 	switch (hw_rev->rl_rev) {
1406 	case RL_HWREV_8139CPLUS:
1407 		sc->rl_flags |= RL_FLAG_FASTETHER | RL_FLAG_AUTOPAD;
1408 		break;
1409 	case RL_HWREV_8100E:
1410 	case RL_HWREV_8101E:
1411 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_FASTETHER;
1412 		break;
1413 	case RL_HWREV_8102E:
1414 	case RL_HWREV_8102EL:
1415 	case RL_HWREV_8102EL_SPIN1:
1416 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR | RL_FLAG_DESCV2 |
1417 		    RL_FLAG_MACSTAT | RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP |
1418 		    RL_FLAG_AUTOPAD;
1419 		break;
1420 	case RL_HWREV_8103E:
1421 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR | RL_FLAG_DESCV2 |
1422 		    RL_FLAG_MACSTAT | RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP |
1423 		    RL_FLAG_AUTOPAD | RL_FLAG_MACSLEEP;
1424 		break;
1425 	case RL_HWREV_8401E:
1426 	case RL_HWREV_8105E:
1427 	case RL_HWREV_8105E_SPIN1:
1428 	case RL_HWREV_8106E:
1429 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
1430 		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
1431 		    RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD;
1432 		break;
1433 	case RL_HWREV_8402:
1434 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
1435 		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
1436 		    RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD |
1437 		    RL_FLAG_CMDSTOP_WAIT_TXQ;
1438 		break;
1439 	case RL_HWREV_8168B_SPIN1:
1440 	case RL_HWREV_8168B_SPIN2:
1441 		sc->rl_flags |= RL_FLAG_WOLRXENB;
1442 		/* FALLTHROUGH */
1443 	case RL_HWREV_8168B_SPIN3:
1444 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_MACSTAT;
1445 		break;
1446 	case RL_HWREV_8168C_SPIN2:
1447 		sc->rl_flags |= RL_FLAG_MACSLEEP;
1448 		/* FALLTHROUGH */
1449 	case RL_HWREV_8168C:
1450 		if (sc->rl_macrev == 0x00200000)
1451 			sc->rl_flags |= RL_FLAG_MACSLEEP;
1452 		/* FALLTHROUGH */
1453 	case RL_HWREV_8168CP:
1454 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
1455 		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP |
1456 		    RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 | RL_FLAG_WOL_MANLINK;
1457 		break;
1458 	case RL_HWREV_8168D:
1459 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
1460 		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
1461 		    RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 |
1462 		    RL_FLAG_WOL_MANLINK;
1463 		break;
1464 	case RL_HWREV_8168DP:
1465 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
1466 		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_AUTOPAD |
1467 		    RL_FLAG_JUMBOV2 | RL_FLAG_WAIT_TXPOLL | RL_FLAG_WOL_MANLINK;
1468 		break;
1469 	case RL_HWREV_8168E:
1470 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
1471 		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
1472 		    RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 |
1473 		    RL_FLAG_WOL_MANLINK;
1474 		break;
1475 	case RL_HWREV_8168E_VL:
1476 	case RL_HWREV_8168F:
1477 		sc->rl_flags |= RL_FLAG_EARLYOFF;
1478 		/* FALLTHROUGH */
1479 	case RL_HWREV_8411:
1480 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
1481 		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP |
1482 		    RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 |
1483 		    RL_FLAG_CMDSTOP_WAIT_TXQ | RL_FLAG_WOL_MANLINK;
1484 		break;
1485 	case RL_HWREV_8168EP:
1486 	case RL_HWREV_8168G:
1487 	case RL_HWREV_8411B:
1488 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
1489 		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP |
1490 		    RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 |
1491 		    RL_FLAG_CMDSTOP_WAIT_TXQ | RL_FLAG_WOL_MANLINK |
1492 		    RL_FLAG_8168G_PLUS;
1493 		break;
1494 	case RL_HWREV_8168GU:
1495 	case RL_HWREV_8168H:
1496 		if (pci_get_device(dev) == RT_DEVICEID_8101E) {
1497 			/* RTL8106E(US), RTL8107E */
1498 			sc->rl_flags |= RL_FLAG_FASTETHER;
1499 		} else
1500 			sc->rl_flags |= RL_FLAG_JUMBOV2 | RL_FLAG_WOL_MANLINK;
1501 
1502 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
1503 		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP |
1504 		    RL_FLAG_AUTOPAD | RL_FLAG_CMDSTOP_WAIT_TXQ |
1505 		    RL_FLAG_8168G_PLUS;
1506 		break;
1507 	case RL_HWREV_8169_8110SB:
1508 	case RL_HWREV_8169_8110SBL:
1509 	case RL_HWREV_8169_8110SC:
1510 	case RL_HWREV_8169_8110SCE:
1511 		sc->rl_flags |= RL_FLAG_PHYWAKE;
1512 		/* FALLTHROUGH */
1513 	case RL_HWREV_8169:
1514 	case RL_HWREV_8169S:
1515 	case RL_HWREV_8110S:
1516 		sc->rl_flags |= RL_FLAG_MACRESET;
1517 		break;
1518 	default:
1519 		break;
1520 	}
1521 
1522 	if (sc->rl_hwrev->rl_rev == RL_HWREV_8139CPLUS) {
1523 		sc->rl_cfg0 = RL_8139_CFG0;
1524 		sc->rl_cfg1 = RL_8139_CFG1;
1525 		sc->rl_cfg2 = 0;
1526 		sc->rl_cfg3 = RL_8139_CFG3;
1527 		sc->rl_cfg4 = RL_8139_CFG4;
1528 		sc->rl_cfg5 = RL_8139_CFG5;
1529 	} else {
1530 		sc->rl_cfg0 = RL_CFG0;
1531 		sc->rl_cfg1 = RL_CFG1;
1532 		sc->rl_cfg2 = RL_CFG2;
1533 		sc->rl_cfg3 = RL_CFG3;
1534 		sc->rl_cfg4 = RL_CFG4;
1535 		sc->rl_cfg5 = RL_CFG5;
1536 	}
1537 
1538 	/* Reset the adapter. */
1539 	RL_LOCK(sc);
1540 	re_reset(sc);
1541 	RL_UNLOCK(sc);
1542 
1543 	/* Enable PME. */
1544 	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1545 	cfg = CSR_READ_1(sc, sc->rl_cfg1);
1546 	cfg |= RL_CFG1_PME;
1547 	CSR_WRITE_1(sc, sc->rl_cfg1, cfg);
1548 	cfg = CSR_READ_1(sc, sc->rl_cfg5);
1549 	cfg &= RL_CFG5_PME_STS;
1550 	CSR_WRITE_1(sc, sc->rl_cfg5, cfg);
1551 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1552 
1553 	if ((sc->rl_flags & RL_FLAG_PAR) != 0) {
1554 		/*
1555 		 * XXX Should have a better way to extract station
1556 		 * address from EEPROM.
1557 		 */
1558 		for (i = 0; i < ETHER_ADDR_LEN; i++)
1559 			eaddr[i] = CSR_READ_1(sc, RL_IDR0 + i);
1560 	} else {
1561 		sc->rl_eewidth = RL_9356_ADDR_LEN;
1562 		re_read_eeprom(sc, (caddr_t)&re_did, 0, 1);
1563 		if (re_did != 0x8129)
1564 			sc->rl_eewidth = RL_9346_ADDR_LEN;
1565 
1566 		/*
1567 		 * Get station address from the EEPROM.
1568 		 */
1569 		re_read_eeprom(sc, (caddr_t)as, RL_EE_EADDR, 3);
1570 		for (i = 0; i < ETHER_ADDR_LEN / 2; i++)
1571 			as[i] = le16toh(as[i]);
1572 		bcopy(as, eaddr, ETHER_ADDR_LEN);
1573 	}
1574 
1575 	if (sc->rl_type == RL_8169) {
1576 		/* Set RX length mask and number of descriptors. */
1577 		sc->rl_rxlenmask = RL_RDESC_STAT_GFRAGLEN;
1578 		sc->rl_txstart = RL_GTXSTART;
1579 		sc->rl_ldata.rl_tx_desc_cnt = RL_8169_TX_DESC_CNT;
1580 		sc->rl_ldata.rl_rx_desc_cnt = RL_8169_RX_DESC_CNT;
1581 	} else {
1582 		/* Set RX length mask and number of descriptors. */
1583 		sc->rl_rxlenmask = RL_RDESC_STAT_FRAGLEN;
1584 		sc->rl_txstart = RL_TXSTART;
1585 		sc->rl_ldata.rl_tx_desc_cnt = RL_8139_TX_DESC_CNT;
1586 		sc->rl_ldata.rl_rx_desc_cnt = RL_8139_RX_DESC_CNT;
1587 	}
1588 
1589 	error = re_allocmem(dev, sc);
1590 	if (error)
1591 		goto fail;
1592 	re_add_sysctls(sc);
1593 
1594 	ifp = sc->rl_ifp = if_alloc(IFT_ETHER);
1595 	if (ifp == NULL) {
1596 		device_printf(dev, "can not if_alloc()\n");
1597 		error = ENOSPC;
1598 		goto fail;
1599 	}
1600 
1601 	/* Take controller out of deep sleep mode. */
1602 	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
1603 		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
1604 			CSR_WRITE_1(sc, RL_GPIO,
1605 			    CSR_READ_1(sc, RL_GPIO) | 0x01);
1606 		else
1607 			CSR_WRITE_1(sc, RL_GPIO,
1608 			    CSR_READ_1(sc, RL_GPIO) & ~0x01);
1609 	}
1610 
1611 	/* Take PHY out of power down mode. */
1612 	if ((sc->rl_flags & RL_FLAG_PHYWAKE_PM) != 0) {
1613 		CSR_WRITE_1(sc, RL_PMCH, CSR_READ_1(sc, RL_PMCH) | 0x80);
1614 		if (hw_rev->rl_rev == RL_HWREV_8401E)
1615 			CSR_WRITE_1(sc, 0xD1, CSR_READ_1(sc, 0xD1) & ~0x08);
1616 	}
1617 	if ((sc->rl_flags & RL_FLAG_PHYWAKE) != 0) {
1618 		re_gmii_writereg(dev, 1, 0x1f, 0);
1619 		re_gmii_writereg(dev, 1, 0x0e, 0);
1620 	}
1621 
1622 	ifp->if_softc = sc;
1623 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1624 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1625 	ifp->if_ioctl = re_ioctl;
1626 	ifp->if_start = re_start;
1627 	/*
1628 	 * RTL8168/8111C generates wrong IP checksummed frame if the
1629 	 * packet has IP options so disable TX checksum offloading.
1630 	 */
1631 	if (sc->rl_hwrev->rl_rev == RL_HWREV_8168C ||
1632 	    sc->rl_hwrev->rl_rev == RL_HWREV_8168C_SPIN2 ||
1633 	    sc->rl_hwrev->rl_rev == RL_HWREV_8168CP) {
1634 		ifp->if_hwassist = 0;
1635 		ifp->if_capabilities = IFCAP_RXCSUM | IFCAP_TSO4;
1636 	} else {
1637 		ifp->if_hwassist = CSUM_IP | CSUM_TCP | CSUM_UDP;
1638 		ifp->if_capabilities = IFCAP_HWCSUM | IFCAP_TSO4;
1639 	}
1640 	ifp->if_hwassist |= CSUM_TSO;
1641 	ifp->if_capenable = ifp->if_capabilities;
1642 	ifp->if_init = re_init;
1643 	IFQ_SET_MAXLEN(&ifp->if_snd, RL_IFQ_MAXLEN);
1644 	ifp->if_snd.ifq_drv_maxlen = RL_IFQ_MAXLEN;
1645 	IFQ_SET_READY(&ifp->if_snd);
1646 
1647 	TASK_INIT(&sc->rl_inttask, 0, re_int_task, sc);
1648 
1649 #define	RE_PHYAD_INTERNAL	 0
1650 
1651 	/* Do MII setup. */
1652 	phy = RE_PHYAD_INTERNAL;
1653 	if (sc->rl_type == RL_8169)
1654 		phy = 1;
1655 	capmask = BMSR_DEFCAPMASK;
1656 	if ((sc->rl_flags & RL_FLAG_FASTETHER) != 0)
1657 		 capmask &= ~BMSR_EXTSTAT;
1658 	error = mii_attach(dev, &sc->rl_miibus, ifp, re_ifmedia_upd,
1659 	    re_ifmedia_sts, capmask, phy, MII_OFFSET_ANY, MIIF_DOPAUSE);
1660 	if (error != 0) {
1661 		device_printf(dev, "attaching PHYs failed\n");
1662 		goto fail;
1663 	}
1664 
1665 	/*
1666 	 * Call MI attach routine.
1667 	 */
1668 	ether_ifattach(ifp, eaddr);
1669 
1670 	/* VLAN capability setup */
1671 	ifp->if_capabilities |= IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING;
1672 	if (ifp->if_capabilities & IFCAP_HWCSUM)
1673 		ifp->if_capabilities |= IFCAP_VLAN_HWCSUM;
1674 	/* Enable WOL if PM is supported. */
1675 	if (pci_find_cap(sc->rl_dev, PCIY_PMG, &reg) == 0)
1676 		ifp->if_capabilities |= IFCAP_WOL;
1677 	ifp->if_capenable = ifp->if_capabilities;
1678 	ifp->if_capenable &= ~(IFCAP_WOL_UCAST | IFCAP_WOL_MCAST);
1679 	/*
1680 	 * Don't enable TSO by default.  It is known to generate
1681 	 * corrupted TCP segments(bad TCP options) under certain
1682 	 * circumstances.
1683 	 */
1684 	ifp->if_hwassist &= ~CSUM_TSO;
1685 	ifp->if_capenable &= ~(IFCAP_TSO4 | IFCAP_VLAN_HWTSO);
1686 #ifdef DEVICE_POLLING
1687 	ifp->if_capabilities |= IFCAP_POLLING;
1688 #endif
1689 	/*
1690 	 * Tell the upper layer(s) we support long frames.
1691 	 * Must appear after the call to ether_ifattach() because
1692 	 * ether_ifattach() sets ifi_hdrlen to the default value.
1693 	 */
1694 	ifp->if_hdrlen = sizeof(struct ether_vlan_header);
1695 
1696 #ifdef DEV_NETMAP
1697 	re_netmap_attach(sc);
1698 #endif /* DEV_NETMAP */
1699 
1700 #ifdef RE_DIAG
1701 	/*
1702 	 * Perform hardware diagnostic on the original RTL8169.
1703 	 * Some 32-bit cards were incorrectly wired and would
1704 	 * malfunction if plugged into a 64-bit slot.
1705 	 */
1706 	if (hwrev == RL_HWREV_8169) {
1707 		error = re_diag(sc);
1708 		if (error) {
1709 			device_printf(dev,
1710 		    	"attach aborted due to hardware diag failure\n");
1711 			ether_ifdetach(ifp);
1712 			goto fail;
1713 		}
1714 	}
1715 #endif
1716 
1717 #ifdef RE_TX_MODERATION
1718 	intr_filter = 1;
1719 #endif
1720 	/* Hook interrupt last to avoid having to lock softc */
1721 	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) != 0 &&
1722 	    intr_filter == 0) {
1723 		error = bus_setup_intr(dev, sc->rl_irq[0],
1724 		    INTR_TYPE_NET | INTR_MPSAFE, NULL, re_intr_msi, sc,
1725 		    &sc->rl_intrhand[0]);
1726 	} else {
1727 		error = bus_setup_intr(dev, sc->rl_irq[0],
1728 		    INTR_TYPE_NET | INTR_MPSAFE, re_intr, NULL, sc,
1729 		    &sc->rl_intrhand[0]);
1730 	}
1731 	if (error) {
1732 		device_printf(dev, "couldn't set up irq\n");
1733 		ether_ifdetach(ifp);
1734 	}
1735 
1736 fail:
1737 	if (error)
1738 		re_detach(dev);
1739 
1740 	return (error);
1741 }
1742 
1743 /*
1744  * Shutdown hardware and free up resources. This can be called any
1745  * time after the mutex has been initialized. It is called in both
1746  * the error case in attach and the normal detach case so it needs
1747  * to be careful about only freeing resources that have actually been
1748  * allocated.
1749  */
1750 static int
1751 re_detach(device_t dev)
1752 {
1753 	struct rl_softc		*sc;
1754 	struct ifnet		*ifp;
1755 	int			i, rid;
1756 
1757 	sc = device_get_softc(dev);
1758 	ifp = sc->rl_ifp;
1759 	KASSERT(mtx_initialized(&sc->rl_mtx), ("re mutex not initialized"));
1760 
1761 	/* These should only be active if attach succeeded */
1762 	if (device_is_attached(dev)) {
1763 #ifdef DEVICE_POLLING
1764 		if (ifp->if_capenable & IFCAP_POLLING)
1765 			ether_poll_deregister(ifp);
1766 #endif
1767 		RL_LOCK(sc);
1768 #if 0
1769 		sc->suspended = 1;
1770 #endif
1771 		re_stop(sc);
1772 		RL_UNLOCK(sc);
1773 		callout_drain(&sc->rl_stat_callout);
1774 		taskqueue_drain(taskqueue_fast, &sc->rl_inttask);
1775 		/*
1776 		 * Force off the IFF_UP flag here, in case someone
1777 		 * still had a BPF descriptor attached to this
1778 		 * interface. If they do, ether_ifdetach() will cause
1779 		 * the BPF code to try and clear the promisc mode
1780 		 * flag, which will bubble down to re_ioctl(),
1781 		 * which will try to call re_init() again. This will
1782 		 * turn the NIC back on and restart the MII ticker,
1783 		 * which will panic the system when the kernel tries
1784 		 * to invoke the re_tick() function that isn't there
1785 		 * anymore.
1786 		 */
1787 		ifp->if_flags &= ~IFF_UP;
1788 		ether_ifdetach(ifp);
1789 	}
1790 	if (sc->rl_miibus)
1791 		device_delete_child(dev, sc->rl_miibus);
1792 	bus_generic_detach(dev);
1793 
1794 	/*
1795 	 * The rest is resource deallocation, so we should already be
1796 	 * stopped here.
1797 	 */
1798 
1799 	if (sc->rl_intrhand[0] != NULL) {
1800 		bus_teardown_intr(dev, sc->rl_irq[0], sc->rl_intrhand[0]);
1801 		sc->rl_intrhand[0] = NULL;
1802 	}
1803 	if (ifp != NULL) {
1804 #ifdef DEV_NETMAP
1805 		netmap_detach(ifp);
1806 #endif /* DEV_NETMAP */
1807 		if_free(ifp);
1808 	}
1809 	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) == 0)
1810 		rid = 0;
1811 	else
1812 		rid = 1;
1813 	if (sc->rl_irq[0] != NULL) {
1814 		bus_release_resource(dev, SYS_RES_IRQ, rid, sc->rl_irq[0]);
1815 		sc->rl_irq[0] = NULL;
1816 	}
1817 	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) != 0)
1818 		pci_release_msi(dev);
1819 	if (sc->rl_res_pba) {
1820 		rid = PCIR_BAR(4);
1821 		bus_release_resource(dev, SYS_RES_MEMORY, rid, sc->rl_res_pba);
1822 	}
1823 	if (sc->rl_res)
1824 		bus_release_resource(dev, sc->rl_res_type, sc->rl_res_id,
1825 		    sc->rl_res);
1826 
1827 	/* Unload and free the RX DMA ring memory and map */
1828 
1829 	if (sc->rl_ldata.rl_rx_list_tag) {
1830 		if (sc->rl_ldata.rl_rx_list_addr)
1831 			bus_dmamap_unload(sc->rl_ldata.rl_rx_list_tag,
1832 			    sc->rl_ldata.rl_rx_list_map);
1833 		if (sc->rl_ldata.rl_rx_list)
1834 			bus_dmamem_free(sc->rl_ldata.rl_rx_list_tag,
1835 			    sc->rl_ldata.rl_rx_list,
1836 			    sc->rl_ldata.rl_rx_list_map);
1837 		bus_dma_tag_destroy(sc->rl_ldata.rl_rx_list_tag);
1838 	}
1839 
1840 	/* Unload and free the TX DMA ring memory and map */
1841 
1842 	if (sc->rl_ldata.rl_tx_list_tag) {
1843 		if (sc->rl_ldata.rl_tx_list_addr)
1844 			bus_dmamap_unload(sc->rl_ldata.rl_tx_list_tag,
1845 			    sc->rl_ldata.rl_tx_list_map);
1846 		if (sc->rl_ldata.rl_tx_list)
1847 			bus_dmamem_free(sc->rl_ldata.rl_tx_list_tag,
1848 			    sc->rl_ldata.rl_tx_list,
1849 			    sc->rl_ldata.rl_tx_list_map);
1850 		bus_dma_tag_destroy(sc->rl_ldata.rl_tx_list_tag);
1851 	}
1852 
1853 	/* Destroy all the RX and TX buffer maps */
1854 
1855 	if (sc->rl_ldata.rl_tx_mtag) {
1856 		for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
1857 			if (sc->rl_ldata.rl_tx_desc[i].tx_dmamap)
1858 				bus_dmamap_destroy(sc->rl_ldata.rl_tx_mtag,
1859 				    sc->rl_ldata.rl_tx_desc[i].tx_dmamap);
1860 		}
1861 		bus_dma_tag_destroy(sc->rl_ldata.rl_tx_mtag);
1862 	}
1863 	if (sc->rl_ldata.rl_rx_mtag) {
1864 		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1865 			if (sc->rl_ldata.rl_rx_desc[i].rx_dmamap)
1866 				bus_dmamap_destroy(sc->rl_ldata.rl_rx_mtag,
1867 				    sc->rl_ldata.rl_rx_desc[i].rx_dmamap);
1868 		}
1869 		if (sc->rl_ldata.rl_rx_sparemap)
1870 			bus_dmamap_destroy(sc->rl_ldata.rl_rx_mtag,
1871 			    sc->rl_ldata.rl_rx_sparemap);
1872 		bus_dma_tag_destroy(sc->rl_ldata.rl_rx_mtag);
1873 	}
1874 	if (sc->rl_ldata.rl_jrx_mtag) {
1875 		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1876 			if (sc->rl_ldata.rl_jrx_desc[i].rx_dmamap)
1877 				bus_dmamap_destroy(sc->rl_ldata.rl_jrx_mtag,
1878 				    sc->rl_ldata.rl_jrx_desc[i].rx_dmamap);
1879 		}
1880 		if (sc->rl_ldata.rl_jrx_sparemap)
1881 			bus_dmamap_destroy(sc->rl_ldata.rl_jrx_mtag,
1882 			    sc->rl_ldata.rl_jrx_sparemap);
1883 		bus_dma_tag_destroy(sc->rl_ldata.rl_jrx_mtag);
1884 	}
1885 	/* Unload and free the stats buffer and map */
1886 
1887 	if (sc->rl_ldata.rl_stag) {
1888 		if (sc->rl_ldata.rl_stats_addr)
1889 			bus_dmamap_unload(sc->rl_ldata.rl_stag,
1890 			    sc->rl_ldata.rl_smap);
1891 		if (sc->rl_ldata.rl_stats)
1892 			bus_dmamem_free(sc->rl_ldata.rl_stag,
1893 			    sc->rl_ldata.rl_stats, sc->rl_ldata.rl_smap);
1894 		bus_dma_tag_destroy(sc->rl_ldata.rl_stag);
1895 	}
1896 
1897 	if (sc->rl_parent_tag)
1898 		bus_dma_tag_destroy(sc->rl_parent_tag);
1899 
1900 	mtx_destroy(&sc->rl_mtx);
1901 
1902 	return (0);
1903 }
1904 
1905 static __inline void
1906 re_discard_rxbuf(struct rl_softc *sc, int idx)
1907 {
1908 	struct rl_desc		*desc;
1909 	struct rl_rxdesc	*rxd;
1910 	uint32_t		cmdstat;
1911 
1912 	if (sc->rl_ifp->if_mtu > RL_MTU &&
1913 	    (sc->rl_flags & RL_FLAG_JUMBOV2) != 0)
1914 		rxd = &sc->rl_ldata.rl_jrx_desc[idx];
1915 	else
1916 		rxd = &sc->rl_ldata.rl_rx_desc[idx];
1917 	desc = &sc->rl_ldata.rl_rx_list[idx];
1918 	desc->rl_vlanctl = 0;
1919 	cmdstat = rxd->rx_size;
1920 	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
1921 		cmdstat |= RL_RDESC_CMD_EOR;
1922 	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
1923 }
1924 
1925 static int
1926 re_newbuf(struct rl_softc *sc, int idx)
1927 {
1928 	struct mbuf		*m;
1929 	struct rl_rxdesc	*rxd;
1930 	bus_dma_segment_t	segs[1];
1931 	bus_dmamap_t		map;
1932 	struct rl_desc		*desc;
1933 	uint32_t		cmdstat;
1934 	int			error, nsegs;
1935 
1936 	m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
1937 	if (m == NULL)
1938 		return (ENOBUFS);
1939 
1940 	m->m_len = m->m_pkthdr.len = MCLBYTES;
1941 #ifdef RE_FIXUP_RX
1942 	/*
1943 	 * This is part of an evil trick to deal with non-x86 platforms.
1944 	 * The RealTek chip requires RX buffers to be aligned on 64-bit
1945 	 * boundaries, but that will hose non-x86 machines. To get around
1946 	 * this, we leave some empty space at the start of each buffer
1947 	 * and for non-x86 hosts, we copy the buffer back six bytes
1948 	 * to achieve word alignment. This is slightly more efficient
1949 	 * than allocating a new buffer, copying the contents, and
1950 	 * discarding the old buffer.
1951 	 */
1952 	m_adj(m, RE_ETHER_ALIGN);
1953 #endif
1954 	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_rx_mtag,
1955 	    sc->rl_ldata.rl_rx_sparemap, m, segs, &nsegs, BUS_DMA_NOWAIT);
1956 	if (error != 0) {
1957 		m_freem(m);
1958 		return (ENOBUFS);
1959 	}
1960 	KASSERT(nsegs == 1, ("%s: %d segment returned!", __func__, nsegs));
1961 
1962 	rxd = &sc->rl_ldata.rl_rx_desc[idx];
1963 	if (rxd->rx_m != NULL) {
1964 		bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap,
1965 		    BUS_DMASYNC_POSTREAD);
1966 		bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap);
1967 	}
1968 
1969 	rxd->rx_m = m;
1970 	map = rxd->rx_dmamap;
1971 	rxd->rx_dmamap = sc->rl_ldata.rl_rx_sparemap;
1972 	rxd->rx_size = segs[0].ds_len;
1973 	sc->rl_ldata.rl_rx_sparemap = map;
1974 	bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap,
1975 	    BUS_DMASYNC_PREREAD);
1976 
1977 	desc = &sc->rl_ldata.rl_rx_list[idx];
1978 	desc->rl_vlanctl = 0;
1979 	desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[0].ds_addr));
1980 	desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[0].ds_addr));
1981 	cmdstat = segs[0].ds_len;
1982 	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
1983 		cmdstat |= RL_RDESC_CMD_EOR;
1984 	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
1985 
1986 	return (0);
1987 }
1988 
1989 static int
1990 re_jumbo_newbuf(struct rl_softc *sc, int idx)
1991 {
1992 	struct mbuf		*m;
1993 	struct rl_rxdesc	*rxd;
1994 	bus_dma_segment_t	segs[1];
1995 	bus_dmamap_t		map;
1996 	struct rl_desc		*desc;
1997 	uint32_t		cmdstat;
1998 	int			error, nsegs;
1999 
2000 	m = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, MJUM9BYTES);
2001 	if (m == NULL)
2002 		return (ENOBUFS);
2003 	m->m_len = m->m_pkthdr.len = MJUM9BYTES;
2004 #ifdef RE_FIXUP_RX
2005 	m_adj(m, RE_ETHER_ALIGN);
2006 #endif
2007 	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_jrx_mtag,
2008 	    sc->rl_ldata.rl_jrx_sparemap, m, segs, &nsegs, BUS_DMA_NOWAIT);
2009 	if (error != 0) {
2010 		m_freem(m);
2011 		return (ENOBUFS);
2012 	}
2013 	KASSERT(nsegs == 1, ("%s: %d segment returned!", __func__, nsegs));
2014 
2015 	rxd = &sc->rl_ldata.rl_jrx_desc[idx];
2016 	if (rxd->rx_m != NULL) {
2017 		bus_dmamap_sync(sc->rl_ldata.rl_jrx_mtag, rxd->rx_dmamap,
2018 		    BUS_DMASYNC_POSTREAD);
2019 		bus_dmamap_unload(sc->rl_ldata.rl_jrx_mtag, rxd->rx_dmamap);
2020 	}
2021 
2022 	rxd->rx_m = m;
2023 	map = rxd->rx_dmamap;
2024 	rxd->rx_dmamap = sc->rl_ldata.rl_jrx_sparemap;
2025 	rxd->rx_size = segs[0].ds_len;
2026 	sc->rl_ldata.rl_jrx_sparemap = map;
2027 	bus_dmamap_sync(sc->rl_ldata.rl_jrx_mtag, rxd->rx_dmamap,
2028 	    BUS_DMASYNC_PREREAD);
2029 
2030 	desc = &sc->rl_ldata.rl_rx_list[idx];
2031 	desc->rl_vlanctl = 0;
2032 	desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[0].ds_addr));
2033 	desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[0].ds_addr));
2034 	cmdstat = segs[0].ds_len;
2035 	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
2036 		cmdstat |= RL_RDESC_CMD_EOR;
2037 	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
2038 
2039 	return (0);
2040 }
2041 
2042 #ifdef RE_FIXUP_RX
2043 static __inline void
2044 re_fixup_rx(struct mbuf *m)
2045 {
2046 	int                     i;
2047 	uint16_t                *src, *dst;
2048 
2049 	src = mtod(m, uint16_t *);
2050 	dst = src - (RE_ETHER_ALIGN - ETHER_ALIGN) / sizeof *src;
2051 
2052 	for (i = 0; i < (m->m_len / sizeof(uint16_t) + 1); i++)
2053 		*dst++ = *src++;
2054 
2055 	m->m_data -= RE_ETHER_ALIGN - ETHER_ALIGN;
2056 }
2057 #endif
2058 
2059 static int
2060 re_tx_list_init(struct rl_softc *sc)
2061 {
2062 	struct rl_desc		*desc;
2063 	int			i;
2064 
2065 	RL_LOCK_ASSERT(sc);
2066 
2067 	bzero(sc->rl_ldata.rl_tx_list,
2068 	    sc->rl_ldata.rl_tx_desc_cnt * sizeof(struct rl_desc));
2069 	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++)
2070 		sc->rl_ldata.rl_tx_desc[i].tx_m = NULL;
2071 #ifdef DEV_NETMAP
2072 	re_netmap_tx_init(sc);
2073 #endif /* DEV_NETMAP */
2074 	/* Set EOR. */
2075 	desc = &sc->rl_ldata.rl_tx_list[sc->rl_ldata.rl_tx_desc_cnt - 1];
2076 	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_EOR);
2077 
2078 	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2079 	    sc->rl_ldata.rl_tx_list_map,
2080 	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
2081 
2082 	sc->rl_ldata.rl_tx_prodidx = 0;
2083 	sc->rl_ldata.rl_tx_considx = 0;
2084 	sc->rl_ldata.rl_tx_free = sc->rl_ldata.rl_tx_desc_cnt;
2085 
2086 	return (0);
2087 }
2088 
2089 static int
2090 re_rx_list_init(struct rl_softc *sc)
2091 {
2092 	int			error, i;
2093 
2094 	bzero(sc->rl_ldata.rl_rx_list,
2095 	    sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc));
2096 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
2097 		sc->rl_ldata.rl_rx_desc[i].rx_m = NULL;
2098 		if ((error = re_newbuf(sc, i)) != 0)
2099 			return (error);
2100 	}
2101 #ifdef DEV_NETMAP
2102 	re_netmap_rx_init(sc);
2103 #endif /* DEV_NETMAP */
2104 
2105 	/* Flush the RX descriptors */
2106 
2107 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
2108 	    sc->rl_ldata.rl_rx_list_map,
2109 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2110 
2111 	sc->rl_ldata.rl_rx_prodidx = 0;
2112 	sc->rl_head = sc->rl_tail = NULL;
2113 	sc->rl_int_rx_act = 0;
2114 
2115 	return (0);
2116 }
2117 
2118 static int
2119 re_jrx_list_init(struct rl_softc *sc)
2120 {
2121 	int			error, i;
2122 
2123 	bzero(sc->rl_ldata.rl_rx_list,
2124 	    sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc));
2125 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
2126 		sc->rl_ldata.rl_jrx_desc[i].rx_m = NULL;
2127 		if ((error = re_jumbo_newbuf(sc, i)) != 0)
2128 			return (error);
2129 	}
2130 
2131 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
2132 	    sc->rl_ldata.rl_rx_list_map,
2133 	    BUS_DMASYNC_PREWRITE | BUS_DMASYNC_PREREAD);
2134 
2135 	sc->rl_ldata.rl_rx_prodidx = 0;
2136 	sc->rl_head = sc->rl_tail = NULL;
2137 	sc->rl_int_rx_act = 0;
2138 
2139 	return (0);
2140 }
2141 
2142 /*
2143  * RX handler for C+ and 8169. For the gigE chips, we support
2144  * the reception of jumbo frames that have been fragmented
2145  * across multiple 2K mbuf cluster buffers.
2146  */
2147 static int
2148 re_rxeof(struct rl_softc *sc, int *rx_npktsp)
2149 {
2150 	struct mbuf		*m;
2151 	struct ifnet		*ifp;
2152 	int			i, rxerr, total_len;
2153 	struct rl_desc		*cur_rx;
2154 	u_int32_t		rxstat, rxvlan;
2155 	int			jumbo, maxpkt = 16, rx_npkts = 0;
2156 
2157 	RL_LOCK_ASSERT(sc);
2158 
2159 	ifp = sc->rl_ifp;
2160 #ifdef DEV_NETMAP
2161 	if (netmap_rx_irq(ifp, 0, &rx_npkts))
2162 		return 0;
2163 #endif /* DEV_NETMAP */
2164 	if (ifp->if_mtu > RL_MTU && (sc->rl_flags & RL_FLAG_JUMBOV2) != 0)
2165 		jumbo = 1;
2166 	else
2167 		jumbo = 0;
2168 
2169 	/* Invalidate the descriptor memory */
2170 
2171 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
2172 	    sc->rl_ldata.rl_rx_list_map,
2173 	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2174 
2175 	for (i = sc->rl_ldata.rl_rx_prodidx; maxpkt > 0;
2176 	    i = RL_RX_DESC_NXT(sc, i)) {
2177 		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
2178 			break;
2179 		cur_rx = &sc->rl_ldata.rl_rx_list[i];
2180 		rxstat = le32toh(cur_rx->rl_cmdstat);
2181 		if ((rxstat & RL_RDESC_STAT_OWN) != 0)
2182 			break;
2183 		total_len = rxstat & sc->rl_rxlenmask;
2184 		rxvlan = le32toh(cur_rx->rl_vlanctl);
2185 		if (jumbo != 0)
2186 			m = sc->rl_ldata.rl_jrx_desc[i].rx_m;
2187 		else
2188 			m = sc->rl_ldata.rl_rx_desc[i].rx_m;
2189 
2190 		if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0 &&
2191 		    (rxstat & (RL_RDESC_STAT_SOF | RL_RDESC_STAT_EOF)) !=
2192 		    (RL_RDESC_STAT_SOF | RL_RDESC_STAT_EOF)) {
2193 			/*
2194 			 * RTL8168C or later controllers do not
2195 			 * support multi-fragment packet.
2196 			 */
2197 			re_discard_rxbuf(sc, i);
2198 			continue;
2199 		} else if ((rxstat & RL_RDESC_STAT_EOF) == 0) {
2200 			if (re_newbuf(sc, i) != 0) {
2201 				/*
2202 				 * If this is part of a multi-fragment packet,
2203 				 * discard all the pieces.
2204 				 */
2205 				if (sc->rl_head != NULL) {
2206 					m_freem(sc->rl_head);
2207 					sc->rl_head = sc->rl_tail = NULL;
2208 				}
2209 				re_discard_rxbuf(sc, i);
2210 				continue;
2211 			}
2212 			m->m_len = RE_RX_DESC_BUFLEN;
2213 			if (sc->rl_head == NULL)
2214 				sc->rl_head = sc->rl_tail = m;
2215 			else {
2216 				m->m_flags &= ~M_PKTHDR;
2217 				sc->rl_tail->m_next = m;
2218 				sc->rl_tail = m;
2219 			}
2220 			continue;
2221 		}
2222 
2223 		/*
2224 		 * NOTE: for the 8139C+, the frame length field
2225 		 * is always 12 bits in size, but for the gigE chips,
2226 		 * it is 13 bits (since the max RX frame length is 16K).
2227 		 * Unfortunately, all 32 bits in the status word
2228 		 * were already used, so to make room for the extra
2229 		 * length bit, RealTek took out the 'frame alignment
2230 		 * error' bit and shifted the other status bits
2231 		 * over one slot. The OWN, EOR, FS and LS bits are
2232 		 * still in the same places. We have already extracted
2233 		 * the frame length and checked the OWN bit, so rather
2234 		 * than using an alternate bit mapping, we shift the
2235 		 * status bits one space to the right so we can evaluate
2236 		 * them using the 8169 status as though it was in the
2237 		 * same format as that of the 8139C+.
2238 		 */
2239 		if (sc->rl_type == RL_8169)
2240 			rxstat >>= 1;
2241 
2242 		/*
2243 		 * if total_len > 2^13-1, both _RXERRSUM and _GIANT will be
2244 		 * set, but if CRC is clear, it will still be a valid frame.
2245 		 */
2246 		if ((rxstat & RL_RDESC_STAT_RXERRSUM) != 0) {
2247 			rxerr = 1;
2248 			if ((sc->rl_flags & RL_FLAG_JUMBOV2) == 0 &&
2249 			    total_len > 8191 &&
2250 			    (rxstat & RL_RDESC_STAT_ERRS) == RL_RDESC_STAT_GIANT)
2251 				rxerr = 0;
2252 			if (rxerr != 0) {
2253 				if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
2254 				/*
2255 				 * If this is part of a multi-fragment packet,
2256 				 * discard all the pieces.
2257 				 */
2258 				if (sc->rl_head != NULL) {
2259 					m_freem(sc->rl_head);
2260 					sc->rl_head = sc->rl_tail = NULL;
2261 				}
2262 				re_discard_rxbuf(sc, i);
2263 				continue;
2264 			}
2265 		}
2266 
2267 		/*
2268 		 * If allocating a replacement mbuf fails,
2269 		 * reload the current one.
2270 		 */
2271 		if (jumbo != 0)
2272 			rxerr = re_jumbo_newbuf(sc, i);
2273 		else
2274 			rxerr = re_newbuf(sc, i);
2275 		if (rxerr != 0) {
2276 			if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
2277 			if (sc->rl_head != NULL) {
2278 				m_freem(sc->rl_head);
2279 				sc->rl_head = sc->rl_tail = NULL;
2280 			}
2281 			re_discard_rxbuf(sc, i);
2282 			continue;
2283 		}
2284 
2285 		if (sc->rl_head != NULL) {
2286 			if (jumbo != 0)
2287 				m->m_len = total_len;
2288 			else {
2289 				m->m_len = total_len % RE_RX_DESC_BUFLEN;
2290 				if (m->m_len == 0)
2291 					m->m_len = RE_RX_DESC_BUFLEN;
2292 			}
2293 			/*
2294 			 * Special case: if there's 4 bytes or less
2295 			 * in this buffer, the mbuf can be discarded:
2296 			 * the last 4 bytes is the CRC, which we don't
2297 			 * care about anyway.
2298 			 */
2299 			if (m->m_len <= ETHER_CRC_LEN) {
2300 				sc->rl_tail->m_len -=
2301 				    (ETHER_CRC_LEN - m->m_len);
2302 				m_freem(m);
2303 			} else {
2304 				m->m_len -= ETHER_CRC_LEN;
2305 				m->m_flags &= ~M_PKTHDR;
2306 				sc->rl_tail->m_next = m;
2307 			}
2308 			m = sc->rl_head;
2309 			sc->rl_head = sc->rl_tail = NULL;
2310 			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
2311 		} else
2312 			m->m_pkthdr.len = m->m_len =
2313 			    (total_len - ETHER_CRC_LEN);
2314 
2315 #ifdef RE_FIXUP_RX
2316 		re_fixup_rx(m);
2317 #endif
2318 		if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
2319 		m->m_pkthdr.rcvif = ifp;
2320 
2321 		/* Do RX checksumming if enabled */
2322 
2323 		if (ifp->if_capenable & IFCAP_RXCSUM) {
2324 			if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
2325 				/* Check IP header checksum */
2326 				if (rxstat & RL_RDESC_STAT_PROTOID)
2327 					m->m_pkthdr.csum_flags |=
2328 					    CSUM_IP_CHECKED;
2329 				if (!(rxstat & RL_RDESC_STAT_IPSUMBAD))
2330 					m->m_pkthdr.csum_flags |=
2331 					    CSUM_IP_VALID;
2332 
2333 				/* Check TCP/UDP checksum */
2334 				if ((RL_TCPPKT(rxstat) &&
2335 				    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
2336 				    (RL_UDPPKT(rxstat) &&
2337 				     !(rxstat & RL_RDESC_STAT_UDPSUMBAD))) {
2338 					m->m_pkthdr.csum_flags |=
2339 						CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
2340 					m->m_pkthdr.csum_data = 0xffff;
2341 				}
2342 			} else {
2343 				/*
2344 				 * RTL8168C/RTL816CP/RTL8111C/RTL8111CP
2345 				 */
2346 				if ((rxstat & RL_RDESC_STAT_PROTOID) &&
2347 				    (rxvlan & RL_RDESC_IPV4))
2348 					m->m_pkthdr.csum_flags |=
2349 					    CSUM_IP_CHECKED;
2350 				if (!(rxstat & RL_RDESC_STAT_IPSUMBAD) &&
2351 				    (rxvlan & RL_RDESC_IPV4))
2352 					m->m_pkthdr.csum_flags |=
2353 					    CSUM_IP_VALID;
2354 				if (((rxstat & RL_RDESC_STAT_TCP) &&
2355 				    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
2356 				    ((rxstat & RL_RDESC_STAT_UDP) &&
2357 				    !(rxstat & RL_RDESC_STAT_UDPSUMBAD))) {
2358 					m->m_pkthdr.csum_flags |=
2359 						CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
2360 					m->m_pkthdr.csum_data = 0xffff;
2361 				}
2362 			}
2363 		}
2364 		maxpkt--;
2365 		if (rxvlan & RL_RDESC_VLANCTL_TAG) {
2366 			m->m_pkthdr.ether_vtag =
2367 			    bswap16((rxvlan & RL_RDESC_VLANCTL_DATA));
2368 			m->m_flags |= M_VLANTAG;
2369 		}
2370 		RL_UNLOCK(sc);
2371 		(*ifp->if_input)(ifp, m);
2372 		RL_LOCK(sc);
2373 		rx_npkts++;
2374 	}
2375 
2376 	/* Flush the RX DMA ring */
2377 
2378 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
2379 	    sc->rl_ldata.rl_rx_list_map,
2380 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2381 
2382 	sc->rl_ldata.rl_rx_prodidx = i;
2383 
2384 	if (rx_npktsp != NULL)
2385 		*rx_npktsp = rx_npkts;
2386 	if (maxpkt)
2387 		return (EAGAIN);
2388 
2389 	return (0);
2390 }
2391 
2392 static void
2393 re_txeof(struct rl_softc *sc)
2394 {
2395 	struct ifnet		*ifp;
2396 	struct rl_txdesc	*txd;
2397 	u_int32_t		txstat;
2398 	int			cons;
2399 
2400 	cons = sc->rl_ldata.rl_tx_considx;
2401 	if (cons == sc->rl_ldata.rl_tx_prodidx)
2402 		return;
2403 
2404 	ifp = sc->rl_ifp;
2405 #ifdef DEV_NETMAP
2406 	if (netmap_tx_irq(ifp, 0))
2407 		return;
2408 #endif /* DEV_NETMAP */
2409 	/* Invalidate the TX descriptor list */
2410 	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2411 	    sc->rl_ldata.rl_tx_list_map,
2412 	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2413 
2414 	for (; cons != sc->rl_ldata.rl_tx_prodidx;
2415 	    cons = RL_TX_DESC_NXT(sc, cons)) {
2416 		txstat = le32toh(sc->rl_ldata.rl_tx_list[cons].rl_cmdstat);
2417 		if (txstat & RL_TDESC_STAT_OWN)
2418 			break;
2419 		/*
2420 		 * We only stash mbufs in the last descriptor
2421 		 * in a fragment chain, which also happens to
2422 		 * be the only place where the TX status bits
2423 		 * are valid.
2424 		 */
2425 		if (txstat & RL_TDESC_CMD_EOF) {
2426 			txd = &sc->rl_ldata.rl_tx_desc[cons];
2427 			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
2428 			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
2429 			bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag,
2430 			    txd->tx_dmamap);
2431 			KASSERT(txd->tx_m != NULL,
2432 			    ("%s: freeing NULL mbufs!", __func__));
2433 			m_freem(txd->tx_m);
2434 			txd->tx_m = NULL;
2435 			if (txstat & (RL_TDESC_STAT_EXCESSCOL|
2436 			    RL_TDESC_STAT_COLCNT))
2437 				if_inc_counter(ifp, IFCOUNTER_COLLISIONS, 1);
2438 			if (txstat & RL_TDESC_STAT_TXERRSUM)
2439 				if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
2440 			else
2441 				if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
2442 		}
2443 		sc->rl_ldata.rl_tx_free++;
2444 		ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
2445 	}
2446 	sc->rl_ldata.rl_tx_considx = cons;
2447 
2448 	/* No changes made to the TX ring, so no flush needed */
2449 
2450 	if (sc->rl_ldata.rl_tx_free != sc->rl_ldata.rl_tx_desc_cnt) {
2451 #ifdef RE_TX_MODERATION
2452 		/*
2453 		 * If not all descriptors have been reaped yet, reload
2454 		 * the timer so that we will eventually get another
2455 		 * interrupt that will cause us to re-enter this routine.
2456 		 * This is done in case the transmitter has gone idle.
2457 		 */
2458 		CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2459 #endif
2460 	} else
2461 		sc->rl_watchdog_timer = 0;
2462 }
2463 
2464 static void
2465 re_tick(void *xsc)
2466 {
2467 	struct rl_softc		*sc;
2468 	struct mii_data		*mii;
2469 
2470 	sc = xsc;
2471 
2472 	RL_LOCK_ASSERT(sc);
2473 
2474 	mii = device_get_softc(sc->rl_miibus);
2475 	mii_tick(mii);
2476 	if ((sc->rl_flags & RL_FLAG_LINK) == 0)
2477 		re_miibus_statchg(sc->rl_dev);
2478 	/*
2479 	 * Reclaim transmitted frames here. Technically it is not
2480 	 * necessary to do here but it ensures periodic reclamation
2481 	 * regardless of Tx completion interrupt which seems to be
2482 	 * lost on PCIe based controllers under certain situations.
2483 	 */
2484 	re_txeof(sc);
2485 	re_watchdog(sc);
2486 	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
2487 }
2488 
2489 #ifdef DEVICE_POLLING
2490 static int
2491 re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
2492 {
2493 	struct rl_softc *sc = ifp->if_softc;
2494 	int rx_npkts = 0;
2495 
2496 	RL_LOCK(sc);
2497 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
2498 		rx_npkts = re_poll_locked(ifp, cmd, count);
2499 	RL_UNLOCK(sc);
2500 	return (rx_npkts);
2501 }
2502 
2503 static int
2504 re_poll_locked(struct ifnet *ifp, enum poll_cmd cmd, int count)
2505 {
2506 	struct rl_softc *sc = ifp->if_softc;
2507 	int rx_npkts;
2508 
2509 	RL_LOCK_ASSERT(sc);
2510 
2511 	sc->rxcycles = count;
2512 	re_rxeof(sc, &rx_npkts);
2513 	re_txeof(sc);
2514 
2515 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2516 		re_start_locked(ifp);
2517 
2518 	if (cmd == POLL_AND_CHECK_STATUS) { /* also check status register */
2519 		u_int16_t       status;
2520 
2521 		status = CSR_READ_2(sc, RL_ISR);
2522 		if (status == 0xffff)
2523 			return (rx_npkts);
2524 		if (status)
2525 			CSR_WRITE_2(sc, RL_ISR, status);
2526 		if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
2527 		    (sc->rl_flags & RL_FLAG_PCIE))
2528 			CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2529 
2530 		/*
2531 		 * XXX check behaviour on receiver stalls.
2532 		 */
2533 
2534 		if (status & RL_ISR_SYSTEM_ERR) {
2535 			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2536 			re_init_locked(sc);
2537 		}
2538 	}
2539 	return (rx_npkts);
2540 }
2541 #endif /* DEVICE_POLLING */
2542 
2543 static int
2544 re_intr(void *arg)
2545 {
2546 	struct rl_softc		*sc;
2547 	uint16_t		status;
2548 
2549 	sc = arg;
2550 
2551 	status = CSR_READ_2(sc, RL_ISR);
2552 	if (status == 0xFFFF || (status & RL_INTRS_CPLUS) == 0)
2553                 return (FILTER_STRAY);
2554 	CSR_WRITE_2(sc, RL_IMR, 0);
2555 
2556 	taskqueue_enqueue(taskqueue_fast, &sc->rl_inttask);
2557 
2558 	return (FILTER_HANDLED);
2559 }
2560 
2561 static void
2562 re_int_task(void *arg, int npending)
2563 {
2564 	struct rl_softc		*sc;
2565 	struct ifnet		*ifp;
2566 	u_int16_t		status;
2567 	int			rval = 0;
2568 
2569 	sc = arg;
2570 	ifp = sc->rl_ifp;
2571 
2572 	RL_LOCK(sc);
2573 
2574 	status = CSR_READ_2(sc, RL_ISR);
2575         CSR_WRITE_2(sc, RL_ISR, status);
2576 
2577 	if (sc->suspended ||
2578 	    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2579 		RL_UNLOCK(sc);
2580 		return;
2581 	}
2582 
2583 #ifdef DEVICE_POLLING
2584 	if  (ifp->if_capenable & IFCAP_POLLING) {
2585 		RL_UNLOCK(sc);
2586 		return;
2587 	}
2588 #endif
2589 
2590 	if (status & (RL_ISR_RX_OK|RL_ISR_RX_ERR|RL_ISR_FIFO_OFLOW))
2591 		rval = re_rxeof(sc, NULL);
2592 
2593 	/*
2594 	 * Some chips will ignore a second TX request issued
2595 	 * while an existing transmission is in progress. If
2596 	 * the transmitter goes idle but there are still
2597 	 * packets waiting to be sent, we need to restart the
2598 	 * channel here to flush them out. This only seems to
2599 	 * be required with the PCIe devices.
2600 	 */
2601 	if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
2602 	    (sc->rl_flags & RL_FLAG_PCIE))
2603 		CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2604 	if (status & (
2605 #ifdef RE_TX_MODERATION
2606 	    RL_ISR_TIMEOUT_EXPIRED|
2607 #else
2608 	    RL_ISR_TX_OK|
2609 #endif
2610 	    RL_ISR_TX_ERR|RL_ISR_TX_DESC_UNAVAIL))
2611 		re_txeof(sc);
2612 
2613 	if (status & RL_ISR_SYSTEM_ERR) {
2614 		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2615 		re_init_locked(sc);
2616 	}
2617 
2618 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2619 		re_start_locked(ifp);
2620 
2621 	RL_UNLOCK(sc);
2622 
2623         if ((CSR_READ_2(sc, RL_ISR) & RL_INTRS_CPLUS) || rval) {
2624 		taskqueue_enqueue(taskqueue_fast, &sc->rl_inttask);
2625 		return;
2626 	}
2627 
2628 	CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2629 }
2630 
2631 static void
2632 re_intr_msi(void *xsc)
2633 {
2634 	struct rl_softc		*sc;
2635 	struct ifnet		*ifp;
2636 	uint16_t		intrs, status;
2637 
2638 	sc = xsc;
2639 	RL_LOCK(sc);
2640 
2641 	ifp = sc->rl_ifp;
2642 #ifdef DEVICE_POLLING
2643 	if (ifp->if_capenable & IFCAP_POLLING) {
2644 		RL_UNLOCK(sc);
2645 		return;
2646 	}
2647 #endif
2648 	/* Disable interrupts. */
2649 	CSR_WRITE_2(sc, RL_IMR, 0);
2650 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2651 		RL_UNLOCK(sc);
2652 		return;
2653 	}
2654 
2655 	intrs = RL_INTRS_CPLUS;
2656 	status = CSR_READ_2(sc, RL_ISR);
2657         CSR_WRITE_2(sc, RL_ISR, status);
2658 	if (sc->rl_int_rx_act > 0) {
2659 		intrs &= ~(RL_ISR_RX_OK | RL_ISR_RX_ERR | RL_ISR_FIFO_OFLOW |
2660 		    RL_ISR_RX_OVERRUN);
2661 		status &= ~(RL_ISR_RX_OK | RL_ISR_RX_ERR | RL_ISR_FIFO_OFLOW |
2662 		    RL_ISR_RX_OVERRUN);
2663 	}
2664 
2665 	if (status & (RL_ISR_TIMEOUT_EXPIRED | RL_ISR_RX_OK | RL_ISR_RX_ERR |
2666 	    RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN)) {
2667 		re_rxeof(sc, NULL);
2668 		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
2669 			if (sc->rl_int_rx_mod != 0 &&
2670 			    (status & (RL_ISR_RX_OK | RL_ISR_RX_ERR |
2671 			    RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN)) != 0) {
2672 				/* Rearm one-shot timer. */
2673 				CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2674 				intrs &= ~(RL_ISR_RX_OK | RL_ISR_RX_ERR |
2675 				    RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN);
2676 				sc->rl_int_rx_act = 1;
2677 			} else {
2678 				intrs |= RL_ISR_RX_OK | RL_ISR_RX_ERR |
2679 				    RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN;
2680 				sc->rl_int_rx_act = 0;
2681 			}
2682 		}
2683 	}
2684 
2685 	/*
2686 	 * Some chips will ignore a second TX request issued
2687 	 * while an existing transmission is in progress. If
2688 	 * the transmitter goes idle but there are still
2689 	 * packets waiting to be sent, we need to restart the
2690 	 * channel here to flush them out. This only seems to
2691 	 * be required with the PCIe devices.
2692 	 */
2693 	if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
2694 	    (sc->rl_flags & RL_FLAG_PCIE))
2695 		CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2696 	if (status & (RL_ISR_TX_OK | RL_ISR_TX_ERR | RL_ISR_TX_DESC_UNAVAIL))
2697 		re_txeof(sc);
2698 
2699 	if (status & RL_ISR_SYSTEM_ERR) {
2700 		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2701 		re_init_locked(sc);
2702 	}
2703 
2704 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
2705 		if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2706 			re_start_locked(ifp);
2707 		CSR_WRITE_2(sc, RL_IMR, intrs);
2708 	}
2709 	RL_UNLOCK(sc);
2710 }
2711 
2712 static int
2713 re_encap(struct rl_softc *sc, struct mbuf **m_head)
2714 {
2715 	struct rl_txdesc	*txd, *txd_last;
2716 	bus_dma_segment_t	segs[RL_NTXSEGS];
2717 	bus_dmamap_t		map;
2718 	struct mbuf		*m_new;
2719 	struct rl_desc		*desc;
2720 	int			nsegs, prod;
2721 	int			i, error, ei, si;
2722 	int			padlen;
2723 	uint32_t		cmdstat, csum_flags, vlanctl;
2724 
2725 	RL_LOCK_ASSERT(sc);
2726 	M_ASSERTPKTHDR((*m_head));
2727 
2728 	/*
2729 	 * With some of the RealTek chips, using the checksum offload
2730 	 * support in conjunction with the autopadding feature results
2731 	 * in the transmission of corrupt frames. For example, if we
2732 	 * need to send a really small IP fragment that's less than 60
2733 	 * bytes in size, and IP header checksumming is enabled, the
2734 	 * resulting ethernet frame that appears on the wire will
2735 	 * have garbled payload. To work around this, if TX IP checksum
2736 	 * offload is enabled, we always manually pad short frames out
2737 	 * to the minimum ethernet frame size.
2738 	 */
2739 	if ((sc->rl_flags & RL_FLAG_AUTOPAD) == 0 &&
2740 	    (*m_head)->m_pkthdr.len < RL_IP4CSUMTX_PADLEN &&
2741 	    ((*m_head)->m_pkthdr.csum_flags & CSUM_IP) != 0) {
2742 		padlen = RL_MIN_FRAMELEN - (*m_head)->m_pkthdr.len;
2743 		if (M_WRITABLE(*m_head) == 0) {
2744 			/* Get a writable copy. */
2745 			m_new = m_dup(*m_head, M_NOWAIT);
2746 			m_freem(*m_head);
2747 			if (m_new == NULL) {
2748 				*m_head = NULL;
2749 				return (ENOBUFS);
2750 			}
2751 			*m_head = m_new;
2752 		}
2753 		if ((*m_head)->m_next != NULL ||
2754 		    M_TRAILINGSPACE(*m_head) < padlen) {
2755 			m_new = m_defrag(*m_head, M_NOWAIT);
2756 			if (m_new == NULL) {
2757 				m_freem(*m_head);
2758 				*m_head = NULL;
2759 				return (ENOBUFS);
2760 			}
2761 		} else
2762 			m_new = *m_head;
2763 
2764 		/*
2765 		 * Manually pad short frames, and zero the pad space
2766 		 * to avoid leaking data.
2767 		 */
2768 		bzero(mtod(m_new, char *) + m_new->m_pkthdr.len, padlen);
2769 		m_new->m_pkthdr.len += padlen;
2770 		m_new->m_len = m_new->m_pkthdr.len;
2771 		*m_head = m_new;
2772 	}
2773 
2774 	prod = sc->rl_ldata.rl_tx_prodidx;
2775 	txd = &sc->rl_ldata.rl_tx_desc[prod];
2776 	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap,
2777 	    *m_head, segs, &nsegs, BUS_DMA_NOWAIT);
2778 	if (error == EFBIG) {
2779 		m_new = m_collapse(*m_head, M_NOWAIT, RL_NTXSEGS);
2780 		if (m_new == NULL) {
2781 			m_freem(*m_head);
2782 			*m_head = NULL;
2783 			return (ENOBUFS);
2784 		}
2785 		*m_head = m_new;
2786 		error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_tx_mtag,
2787 		    txd->tx_dmamap, *m_head, segs, &nsegs, BUS_DMA_NOWAIT);
2788 		if (error != 0) {
2789 			m_freem(*m_head);
2790 			*m_head = NULL;
2791 			return (error);
2792 		}
2793 	} else if (error != 0)
2794 		return (error);
2795 	if (nsegs == 0) {
2796 		m_freem(*m_head);
2797 		*m_head = NULL;
2798 		return (EIO);
2799 	}
2800 
2801 	/* Check for number of available descriptors. */
2802 	if (sc->rl_ldata.rl_tx_free - nsegs <= 1) {
2803 		bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap);
2804 		return (ENOBUFS);
2805 	}
2806 
2807 	bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap,
2808 	    BUS_DMASYNC_PREWRITE);
2809 
2810 	/*
2811 	 * Set up checksum offload. Note: checksum offload bits must
2812 	 * appear in all descriptors of a multi-descriptor transmit
2813 	 * attempt. This is according to testing done with an 8169
2814 	 * chip. This is a requirement.
2815 	 */
2816 	vlanctl = 0;
2817 	csum_flags = 0;
2818 	if (((*m_head)->m_pkthdr.csum_flags & CSUM_TSO) != 0) {
2819 		if ((sc->rl_flags & RL_FLAG_DESCV2) != 0) {
2820 			csum_flags |= RL_TDESC_CMD_LGSEND;
2821 			vlanctl |= ((uint32_t)(*m_head)->m_pkthdr.tso_segsz <<
2822 			    RL_TDESC_CMD_MSSVALV2_SHIFT);
2823 		} else {
2824 			csum_flags |= RL_TDESC_CMD_LGSEND |
2825 			    ((uint32_t)(*m_head)->m_pkthdr.tso_segsz <<
2826 			    RL_TDESC_CMD_MSSVAL_SHIFT);
2827 		}
2828 	} else {
2829 		/*
2830 		 * Unconditionally enable IP checksum if TCP or UDP
2831 		 * checksum is required. Otherwise, TCP/UDP checksum
2832 		 * doesn't make effects.
2833 		 */
2834 		if (((*m_head)->m_pkthdr.csum_flags & RE_CSUM_FEATURES) != 0) {
2835 			if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
2836 				csum_flags |= RL_TDESC_CMD_IPCSUM;
2837 				if (((*m_head)->m_pkthdr.csum_flags &
2838 				    CSUM_TCP) != 0)
2839 					csum_flags |= RL_TDESC_CMD_TCPCSUM;
2840 				if (((*m_head)->m_pkthdr.csum_flags &
2841 				    CSUM_UDP) != 0)
2842 					csum_flags |= RL_TDESC_CMD_UDPCSUM;
2843 			} else {
2844 				vlanctl |= RL_TDESC_CMD_IPCSUMV2;
2845 				if (((*m_head)->m_pkthdr.csum_flags &
2846 				    CSUM_TCP) != 0)
2847 					vlanctl |= RL_TDESC_CMD_TCPCSUMV2;
2848 				if (((*m_head)->m_pkthdr.csum_flags &
2849 				    CSUM_UDP) != 0)
2850 					vlanctl |= RL_TDESC_CMD_UDPCSUMV2;
2851 			}
2852 		}
2853 	}
2854 
2855 	/*
2856 	 * Set up hardware VLAN tagging. Note: vlan tag info must
2857 	 * appear in all descriptors of a multi-descriptor
2858 	 * transmission attempt.
2859 	 */
2860 	if ((*m_head)->m_flags & M_VLANTAG)
2861 		vlanctl |= bswap16((*m_head)->m_pkthdr.ether_vtag) |
2862 		    RL_TDESC_VLANCTL_TAG;
2863 
2864 	si = prod;
2865 	for (i = 0; i < nsegs; i++, prod = RL_TX_DESC_NXT(sc, prod)) {
2866 		desc = &sc->rl_ldata.rl_tx_list[prod];
2867 		desc->rl_vlanctl = htole32(vlanctl);
2868 		desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[i].ds_addr));
2869 		desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[i].ds_addr));
2870 		cmdstat = segs[i].ds_len;
2871 		if (i != 0)
2872 			cmdstat |= RL_TDESC_CMD_OWN;
2873 		if (prod == sc->rl_ldata.rl_tx_desc_cnt - 1)
2874 			cmdstat |= RL_TDESC_CMD_EOR;
2875 		desc->rl_cmdstat = htole32(cmdstat | csum_flags);
2876 		sc->rl_ldata.rl_tx_free--;
2877 	}
2878 	/* Update producer index. */
2879 	sc->rl_ldata.rl_tx_prodidx = prod;
2880 
2881 	/* Set EOF on the last descriptor. */
2882 	ei = RL_TX_DESC_PRV(sc, prod);
2883 	desc = &sc->rl_ldata.rl_tx_list[ei];
2884 	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_EOF);
2885 
2886 	desc = &sc->rl_ldata.rl_tx_list[si];
2887 	/* Set SOF and transfer ownership of packet to the chip. */
2888 	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_OWN | RL_TDESC_CMD_SOF);
2889 
2890 	/*
2891 	 * Insure that the map for this transmission
2892 	 * is placed at the array index of the last descriptor
2893 	 * in this chain.  (Swap last and first dmamaps.)
2894 	 */
2895 	txd_last = &sc->rl_ldata.rl_tx_desc[ei];
2896 	map = txd->tx_dmamap;
2897 	txd->tx_dmamap = txd_last->tx_dmamap;
2898 	txd_last->tx_dmamap = map;
2899 	txd_last->tx_m = *m_head;
2900 
2901 	return (0);
2902 }
2903 
2904 static void
2905 re_start(struct ifnet *ifp)
2906 {
2907 	struct rl_softc		*sc;
2908 
2909 	sc = ifp->if_softc;
2910 	RL_LOCK(sc);
2911 	re_start_locked(ifp);
2912 	RL_UNLOCK(sc);
2913 }
2914 
2915 /*
2916  * Main transmit routine for C+ and gigE NICs.
2917  */
2918 static void
2919 re_start_locked(struct ifnet *ifp)
2920 {
2921 	struct rl_softc		*sc;
2922 	struct mbuf		*m_head;
2923 	int			queued;
2924 
2925 	sc = ifp->if_softc;
2926 
2927 #ifdef DEV_NETMAP
2928 	/* XXX is this necessary ? */
2929 	if (ifp->if_capenable & IFCAP_NETMAP) {
2930 		struct netmap_kring *kring = &NA(ifp)->tx_rings[0];
2931 		if (sc->rl_ldata.rl_tx_prodidx != kring->nr_hwcur) {
2932 			/* kick the tx unit */
2933 			CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2934 #ifdef RE_TX_MODERATION
2935 			CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2936 #endif
2937 			sc->rl_watchdog_timer = 5;
2938 		}
2939 		return;
2940 	}
2941 #endif /* DEV_NETMAP */
2942 
2943 	if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
2944 	    IFF_DRV_RUNNING || (sc->rl_flags & RL_FLAG_LINK) == 0)
2945 		return;
2946 
2947 	for (queued = 0; !IFQ_DRV_IS_EMPTY(&ifp->if_snd) &&
2948 	    sc->rl_ldata.rl_tx_free > 1;) {
2949 		IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
2950 		if (m_head == NULL)
2951 			break;
2952 
2953 		if (re_encap(sc, &m_head) != 0) {
2954 			if (m_head == NULL)
2955 				break;
2956 			IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
2957 			ifp->if_drv_flags |= IFF_DRV_OACTIVE;
2958 			break;
2959 		}
2960 
2961 		/*
2962 		 * If there's a BPF listener, bounce a copy of this frame
2963 		 * to him.
2964 		 */
2965 		ETHER_BPF_MTAP(ifp, m_head);
2966 
2967 		queued++;
2968 	}
2969 
2970 	if (queued == 0) {
2971 #ifdef RE_TX_MODERATION
2972 		if (sc->rl_ldata.rl_tx_free != sc->rl_ldata.rl_tx_desc_cnt)
2973 			CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2974 #endif
2975 		return;
2976 	}
2977 
2978 	/* Flush the TX descriptors */
2979 
2980 	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2981 	    sc->rl_ldata.rl_tx_list_map,
2982 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2983 
2984 	CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2985 
2986 #ifdef RE_TX_MODERATION
2987 	/*
2988 	 * Use the countdown timer for interrupt moderation.
2989 	 * 'TX done' interrupts are disabled. Instead, we reset the
2990 	 * countdown timer, which will begin counting until it hits
2991 	 * the value in the TIMERINT register, and then trigger an
2992 	 * interrupt. Each time we write to the TIMERCNT register,
2993 	 * the timer count is reset to 0.
2994 	 */
2995 	CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2996 #endif
2997 
2998 	/*
2999 	 * Set a timeout in case the chip goes out to lunch.
3000 	 */
3001 	sc->rl_watchdog_timer = 5;
3002 }
3003 
3004 static void
3005 re_set_jumbo(struct rl_softc *sc, int jumbo)
3006 {
3007 
3008 	if (sc->rl_hwrev->rl_rev == RL_HWREV_8168E_VL) {
3009 		pci_set_max_read_req(sc->rl_dev, 4096);
3010 		return;
3011 	}
3012 
3013 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG);
3014 	if (jumbo != 0) {
3015 		CSR_WRITE_1(sc, sc->rl_cfg3, CSR_READ_1(sc, sc->rl_cfg3) |
3016 		    RL_CFG3_JUMBO_EN0);
3017 		switch (sc->rl_hwrev->rl_rev) {
3018 		case RL_HWREV_8168DP:
3019 			break;
3020 		case RL_HWREV_8168E:
3021 			CSR_WRITE_1(sc, sc->rl_cfg4,
3022 			    CSR_READ_1(sc, sc->rl_cfg4) | 0x01);
3023 			break;
3024 		default:
3025 			CSR_WRITE_1(sc, sc->rl_cfg4,
3026 			    CSR_READ_1(sc, sc->rl_cfg4) | RL_CFG4_JUMBO_EN1);
3027 		}
3028 	} else {
3029 		CSR_WRITE_1(sc, sc->rl_cfg3, CSR_READ_1(sc, sc->rl_cfg3) &
3030 		    ~RL_CFG3_JUMBO_EN0);
3031 		switch (sc->rl_hwrev->rl_rev) {
3032 		case RL_HWREV_8168DP:
3033 			break;
3034 		case RL_HWREV_8168E:
3035 			CSR_WRITE_1(sc, sc->rl_cfg4,
3036 			    CSR_READ_1(sc, sc->rl_cfg4) & ~0x01);
3037 			break;
3038 		default:
3039 			CSR_WRITE_1(sc, sc->rl_cfg4,
3040 			    CSR_READ_1(sc, sc->rl_cfg4) & ~RL_CFG4_JUMBO_EN1);
3041 		}
3042 	}
3043 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3044 
3045 	switch (sc->rl_hwrev->rl_rev) {
3046 	case RL_HWREV_8168DP:
3047 		pci_set_max_read_req(sc->rl_dev, 4096);
3048 		break;
3049 	default:
3050 		if (jumbo != 0)
3051 			pci_set_max_read_req(sc->rl_dev, 512);
3052 		else
3053 			pci_set_max_read_req(sc->rl_dev, 4096);
3054 	}
3055 }
3056 
3057 static void
3058 re_init(void *xsc)
3059 {
3060 	struct rl_softc		*sc = xsc;
3061 
3062 	RL_LOCK(sc);
3063 	re_init_locked(sc);
3064 	RL_UNLOCK(sc);
3065 }
3066 
3067 static void
3068 re_init_locked(struct rl_softc *sc)
3069 {
3070 	struct ifnet		*ifp = sc->rl_ifp;
3071 	struct mii_data		*mii;
3072 	uint32_t		reg;
3073 	uint16_t		cfg;
3074 	union {
3075 		uint32_t align_dummy;
3076 		u_char eaddr[ETHER_ADDR_LEN];
3077         } eaddr;
3078 
3079 	RL_LOCK_ASSERT(sc);
3080 
3081 	mii = device_get_softc(sc->rl_miibus);
3082 
3083 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
3084 		return;
3085 
3086 	/*
3087 	 * Cancel pending I/O and free all RX/TX buffers.
3088 	 */
3089 	re_stop(sc);
3090 
3091 	/* Put controller into known state. */
3092 	re_reset(sc);
3093 
3094 	/*
3095 	 * For C+ mode, initialize the RX descriptors and mbufs.
3096 	 */
3097 	if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
3098 		if (ifp->if_mtu > RL_MTU) {
3099 			if (re_jrx_list_init(sc) != 0) {
3100 				device_printf(sc->rl_dev,
3101 				    "no memory for jumbo RX buffers\n");
3102 				re_stop(sc);
3103 				return;
3104 			}
3105 			/* Disable checksum offloading for jumbo frames. */
3106 			ifp->if_capenable &= ~(IFCAP_HWCSUM | IFCAP_TSO4);
3107 			ifp->if_hwassist &= ~(RE_CSUM_FEATURES | CSUM_TSO);
3108 		} else {
3109 			if (re_rx_list_init(sc) != 0) {
3110 				device_printf(sc->rl_dev,
3111 				    "no memory for RX buffers\n");
3112 				re_stop(sc);
3113 				return;
3114 			}
3115 		}
3116 		re_set_jumbo(sc, ifp->if_mtu > RL_MTU);
3117 	} else {
3118 		if (re_rx_list_init(sc) != 0) {
3119 			device_printf(sc->rl_dev, "no memory for RX buffers\n");
3120 			re_stop(sc);
3121 			return;
3122 		}
3123 		if ((sc->rl_flags & RL_FLAG_PCIE) != 0 &&
3124 		    pci_get_device(sc->rl_dev) != RT_DEVICEID_8101E) {
3125 			if (ifp->if_mtu > RL_MTU)
3126 				pci_set_max_read_req(sc->rl_dev, 512);
3127 			else
3128 				pci_set_max_read_req(sc->rl_dev, 4096);
3129 		}
3130 	}
3131 	re_tx_list_init(sc);
3132 
3133 	/*
3134 	 * Enable C+ RX and TX mode, as well as VLAN stripping and
3135 	 * RX checksum offload. We must configure the C+ register
3136 	 * before all others.
3137 	 */
3138 	cfg = RL_CPLUSCMD_PCI_MRW;
3139 	if ((ifp->if_capenable & IFCAP_RXCSUM) != 0)
3140 		cfg |= RL_CPLUSCMD_RXCSUM_ENB;
3141 	if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0)
3142 		cfg |= RL_CPLUSCMD_VLANSTRIP;
3143 	if ((sc->rl_flags & RL_FLAG_MACSTAT) != 0) {
3144 		cfg |= RL_CPLUSCMD_MACSTAT_DIS;
3145 		/* XXX magic. */
3146 		cfg |= 0x0001;
3147 	} else
3148 		cfg |= RL_CPLUSCMD_RXENB | RL_CPLUSCMD_TXENB;
3149 	CSR_WRITE_2(sc, RL_CPLUS_CMD, cfg);
3150 	if (sc->rl_hwrev->rl_rev == RL_HWREV_8169_8110SC ||
3151 	    sc->rl_hwrev->rl_rev == RL_HWREV_8169_8110SCE) {
3152 		reg = 0x000fff00;
3153 		if ((CSR_READ_1(sc, sc->rl_cfg2) & RL_CFG2_PCI66MHZ) != 0)
3154 			reg |= 0x000000ff;
3155 		if (sc->rl_hwrev->rl_rev == RL_HWREV_8169_8110SCE)
3156 			reg |= 0x00f00000;
3157 		CSR_WRITE_4(sc, 0x7c, reg);
3158 		/* Disable interrupt mitigation. */
3159 		CSR_WRITE_2(sc, 0xe2, 0);
3160 	}
3161 	/*
3162 	 * Disable TSO if interface MTU size is greater than MSS
3163 	 * allowed in controller.
3164 	 */
3165 	if (ifp->if_mtu > RL_TSO_MTU && (ifp->if_capenable & IFCAP_TSO4) != 0) {
3166 		ifp->if_capenable &= ~IFCAP_TSO4;
3167 		ifp->if_hwassist &= ~CSUM_TSO;
3168 	}
3169 
3170 	/*
3171 	 * Init our MAC address.  Even though the chipset
3172 	 * documentation doesn't mention it, we need to enter "Config
3173 	 * register write enable" mode to modify the ID registers.
3174 	 */
3175 	/* Copy MAC address on stack to align. */
3176 	bcopy(IF_LLADDR(ifp), eaddr.eaddr, ETHER_ADDR_LEN);
3177 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG);
3178 	CSR_WRITE_4(sc, RL_IDR0,
3179 	    htole32(*(u_int32_t *)(&eaddr.eaddr[0])));
3180 	CSR_WRITE_4(sc, RL_IDR4,
3181 	    htole32(*(u_int32_t *)(&eaddr.eaddr[4])));
3182 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3183 
3184 	/*
3185 	 * Load the addresses of the RX and TX lists into the chip.
3186 	 */
3187 
3188 	CSR_WRITE_4(sc, RL_RXLIST_ADDR_HI,
3189 	    RL_ADDR_HI(sc->rl_ldata.rl_rx_list_addr));
3190 	CSR_WRITE_4(sc, RL_RXLIST_ADDR_LO,
3191 	    RL_ADDR_LO(sc->rl_ldata.rl_rx_list_addr));
3192 
3193 	CSR_WRITE_4(sc, RL_TXLIST_ADDR_HI,
3194 	    RL_ADDR_HI(sc->rl_ldata.rl_tx_list_addr));
3195 	CSR_WRITE_4(sc, RL_TXLIST_ADDR_LO,
3196 	    RL_ADDR_LO(sc->rl_ldata.rl_tx_list_addr));
3197 
3198 	if ((sc->rl_flags & RL_FLAG_8168G_PLUS) != 0) {
3199 		/* Disable RXDV gate. */
3200 		CSR_WRITE_4(sc, RL_MISC, CSR_READ_4(sc, RL_MISC) &
3201 		    ~0x00080000);
3202 	}
3203 
3204 	/*
3205 	 * Enable transmit and receive for pre-RTL8168G controllers.
3206 	 * RX/TX MACs should be enabled before RX/TX configuration.
3207 	 */
3208 	if ((sc->rl_flags & RL_FLAG_8168G_PLUS) == 0)
3209 		CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB | RL_CMD_RX_ENB);
3210 
3211 	/*
3212 	 * Set the initial TX configuration.
3213 	 */
3214 	if (sc->rl_testmode) {
3215 		if (sc->rl_type == RL_8169)
3216 			CSR_WRITE_4(sc, RL_TXCFG,
3217 			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON);
3218 		else
3219 			CSR_WRITE_4(sc, RL_TXCFG,
3220 			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON_CPLUS);
3221 	} else
3222 		CSR_WRITE_4(sc, RL_TXCFG, RL_TXCFG_CONFIG);
3223 
3224 	CSR_WRITE_1(sc, RL_EARLY_TX_THRESH, 16);
3225 
3226 	/*
3227 	 * Set the initial RX configuration.
3228 	 */
3229 	re_set_rxmode(sc);
3230 
3231 	/* Configure interrupt moderation. */
3232 	if (sc->rl_type == RL_8169) {
3233 		/* Magic from vendor. */
3234 		CSR_WRITE_2(sc, RL_INTRMOD, 0x5100);
3235 	}
3236 
3237 	/*
3238 	 * Enable transmit and receive for RTL8168G and later controllers.
3239 	 * RX/TX MACs should be enabled after RX/TX configuration.
3240 	 */
3241 	if ((sc->rl_flags & RL_FLAG_8168G_PLUS) != 0)
3242 		CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB | RL_CMD_RX_ENB);
3243 
3244 #ifdef DEVICE_POLLING
3245 	/*
3246 	 * Disable interrupts if we are polling.
3247 	 */
3248 	if (ifp->if_capenable & IFCAP_POLLING)
3249 		CSR_WRITE_2(sc, RL_IMR, 0);
3250 	else	/* otherwise ... */
3251 #endif
3252 
3253 	/*
3254 	 * Enable interrupts.
3255 	 */
3256 	if (sc->rl_testmode)
3257 		CSR_WRITE_2(sc, RL_IMR, 0);
3258 	else
3259 		CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
3260 	CSR_WRITE_2(sc, RL_ISR, RL_INTRS_CPLUS);
3261 
3262 	/* Set initial TX threshold */
3263 	sc->rl_txthresh = RL_TX_THRESH_INIT;
3264 
3265 	/* Start RX/TX process. */
3266 	CSR_WRITE_4(sc, RL_MISSEDPKT, 0);
3267 
3268 	/*
3269 	 * Initialize the timer interrupt register so that
3270 	 * a timer interrupt will be generated once the timer
3271 	 * reaches a certain number of ticks. The timer is
3272 	 * reloaded on each transmit.
3273 	 */
3274 #ifdef RE_TX_MODERATION
3275 	/*
3276 	 * Use timer interrupt register to moderate TX interrupt
3277 	 * moderation, which dramatically improves TX frame rate.
3278 	 */
3279 	if (sc->rl_type == RL_8169)
3280 		CSR_WRITE_4(sc, RL_TIMERINT_8169, 0x800);
3281 	else
3282 		CSR_WRITE_4(sc, RL_TIMERINT, 0x400);
3283 #else
3284 	/*
3285 	 * Use timer interrupt register to moderate RX interrupt
3286 	 * moderation.
3287 	 */
3288 	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) != 0 &&
3289 	    intr_filter == 0) {
3290 		if (sc->rl_type == RL_8169)
3291 			CSR_WRITE_4(sc, RL_TIMERINT_8169,
3292 			    RL_USECS(sc->rl_int_rx_mod));
3293 	} else {
3294 		if (sc->rl_type == RL_8169)
3295 			CSR_WRITE_4(sc, RL_TIMERINT_8169, RL_USECS(0));
3296 	}
3297 #endif
3298 
3299 	/*
3300 	 * For 8169 gigE NICs, set the max allowed RX packet
3301 	 * size so we can receive jumbo frames.
3302 	 */
3303 	if (sc->rl_type == RL_8169) {
3304 		if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
3305 			/*
3306 			 * For controllers that use new jumbo frame scheme,
3307 			 * set maximum size of jumbo frame depending on
3308 			 * controller revisions.
3309 			 */
3310 			if (ifp->if_mtu > RL_MTU)
3311 				CSR_WRITE_2(sc, RL_MAXRXPKTLEN,
3312 				    sc->rl_hwrev->rl_max_mtu +
3313 				    ETHER_VLAN_ENCAP_LEN + ETHER_HDR_LEN +
3314 				    ETHER_CRC_LEN);
3315 			else
3316 				CSR_WRITE_2(sc, RL_MAXRXPKTLEN,
3317 				    RE_RX_DESC_BUFLEN);
3318 		} else if ((sc->rl_flags & RL_FLAG_PCIE) != 0 &&
3319 		    sc->rl_hwrev->rl_max_mtu == RL_MTU) {
3320 			/* RTL810x has no jumbo frame support. */
3321 			CSR_WRITE_2(sc, RL_MAXRXPKTLEN, RE_RX_DESC_BUFLEN);
3322 		} else
3323 			CSR_WRITE_2(sc, RL_MAXRXPKTLEN, 16383);
3324 	}
3325 
3326 	if (sc->rl_testmode)
3327 		return;
3328 
3329 	CSR_WRITE_1(sc, sc->rl_cfg1, CSR_READ_1(sc, sc->rl_cfg1) |
3330 	    RL_CFG1_DRVLOAD);
3331 
3332 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
3333 	ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
3334 
3335 	sc->rl_flags &= ~RL_FLAG_LINK;
3336 	mii_mediachg(mii);
3337 
3338 	sc->rl_watchdog_timer = 0;
3339 	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
3340 }
3341 
3342 /*
3343  * Set media options.
3344  */
3345 static int
3346 re_ifmedia_upd(struct ifnet *ifp)
3347 {
3348 	struct rl_softc		*sc;
3349 	struct mii_data		*mii;
3350 	int			error;
3351 
3352 	sc = ifp->if_softc;
3353 	mii = device_get_softc(sc->rl_miibus);
3354 	RL_LOCK(sc);
3355 	error = mii_mediachg(mii);
3356 	RL_UNLOCK(sc);
3357 
3358 	return (error);
3359 }
3360 
3361 /*
3362  * Report current media status.
3363  */
3364 static void
3365 re_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
3366 {
3367 	struct rl_softc		*sc;
3368 	struct mii_data		*mii;
3369 
3370 	sc = ifp->if_softc;
3371 	mii = device_get_softc(sc->rl_miibus);
3372 
3373 	RL_LOCK(sc);
3374 	mii_pollstat(mii);
3375 	ifmr->ifm_active = mii->mii_media_active;
3376 	ifmr->ifm_status = mii->mii_media_status;
3377 	RL_UNLOCK(sc);
3378 }
3379 
3380 static int
3381 re_ioctl(struct ifnet *ifp, u_long command, caddr_t data)
3382 {
3383 	struct rl_softc		*sc = ifp->if_softc;
3384 	struct ifreq		*ifr = (struct ifreq *) data;
3385 	struct mii_data		*mii;
3386 	int			error = 0;
3387 
3388 	switch (command) {
3389 	case SIOCSIFMTU:
3390 		if (ifr->ifr_mtu < ETHERMIN ||
3391 		    ifr->ifr_mtu > sc->rl_hwrev->rl_max_mtu ||
3392 		    ((sc->rl_flags & RL_FLAG_FASTETHER) != 0 &&
3393 		    ifr->ifr_mtu > RL_MTU)) {
3394 			error = EINVAL;
3395 			break;
3396 		}
3397 		RL_LOCK(sc);
3398 		if (ifp->if_mtu != ifr->ifr_mtu) {
3399 			ifp->if_mtu = ifr->ifr_mtu;
3400 			if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0 &&
3401 			    (ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
3402 				ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3403 				re_init_locked(sc);
3404 			}
3405 			if (ifp->if_mtu > RL_TSO_MTU &&
3406 			    (ifp->if_capenable & IFCAP_TSO4) != 0) {
3407 				ifp->if_capenable &= ~(IFCAP_TSO4 |
3408 				    IFCAP_VLAN_HWTSO);
3409 				ifp->if_hwassist &= ~CSUM_TSO;
3410 			}
3411 			VLAN_CAPABILITIES(ifp);
3412 		}
3413 		RL_UNLOCK(sc);
3414 		break;
3415 	case SIOCSIFFLAGS:
3416 		RL_LOCK(sc);
3417 		if ((ifp->if_flags & IFF_UP) != 0) {
3418 			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
3419 				if (((ifp->if_flags ^ sc->rl_if_flags)
3420 				    & (IFF_PROMISC | IFF_ALLMULTI)) != 0)
3421 					re_set_rxmode(sc);
3422 			} else
3423 				re_init_locked(sc);
3424 		} else {
3425 			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
3426 				re_stop(sc);
3427 		}
3428 		sc->rl_if_flags = ifp->if_flags;
3429 		RL_UNLOCK(sc);
3430 		break;
3431 	case SIOCADDMULTI:
3432 	case SIOCDELMULTI:
3433 		RL_LOCK(sc);
3434 		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
3435 			re_set_rxmode(sc);
3436 		RL_UNLOCK(sc);
3437 		break;
3438 	case SIOCGIFMEDIA:
3439 	case SIOCSIFMEDIA:
3440 		mii = device_get_softc(sc->rl_miibus);
3441 		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, command);
3442 		break;
3443 	case SIOCSIFCAP:
3444 	    {
3445 		int mask, reinit;
3446 
3447 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
3448 		reinit = 0;
3449 #ifdef DEVICE_POLLING
3450 		if (mask & IFCAP_POLLING) {
3451 			if (ifr->ifr_reqcap & IFCAP_POLLING) {
3452 				error = ether_poll_register(re_poll, ifp);
3453 				if (error)
3454 					return (error);
3455 				RL_LOCK(sc);
3456 				/* Disable interrupts */
3457 				CSR_WRITE_2(sc, RL_IMR, 0x0000);
3458 				ifp->if_capenable |= IFCAP_POLLING;
3459 				RL_UNLOCK(sc);
3460 			} else {
3461 				error = ether_poll_deregister(ifp);
3462 				/* Enable interrupts. */
3463 				RL_LOCK(sc);
3464 				CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
3465 				ifp->if_capenable &= ~IFCAP_POLLING;
3466 				RL_UNLOCK(sc);
3467 			}
3468 		}
3469 #endif /* DEVICE_POLLING */
3470 		RL_LOCK(sc);
3471 		if ((mask & IFCAP_TXCSUM) != 0 &&
3472 		    (ifp->if_capabilities & IFCAP_TXCSUM) != 0) {
3473 			ifp->if_capenable ^= IFCAP_TXCSUM;
3474 			if ((ifp->if_capenable & IFCAP_TXCSUM) != 0)
3475 				ifp->if_hwassist |= RE_CSUM_FEATURES;
3476 			else
3477 				ifp->if_hwassist &= ~RE_CSUM_FEATURES;
3478 			reinit = 1;
3479 		}
3480 		if ((mask & IFCAP_RXCSUM) != 0 &&
3481 		    (ifp->if_capabilities & IFCAP_RXCSUM) != 0) {
3482 			ifp->if_capenable ^= IFCAP_RXCSUM;
3483 			reinit = 1;
3484 		}
3485 		if ((mask & IFCAP_TSO4) != 0 &&
3486 		    (ifp->if_capabilities & IFCAP_TSO4) != 0) {
3487 			ifp->if_capenable ^= IFCAP_TSO4;
3488 			if ((IFCAP_TSO4 & ifp->if_capenable) != 0)
3489 				ifp->if_hwassist |= CSUM_TSO;
3490 			else
3491 				ifp->if_hwassist &= ~CSUM_TSO;
3492 			if (ifp->if_mtu > RL_TSO_MTU &&
3493 			    (ifp->if_capenable & IFCAP_TSO4) != 0) {
3494 				ifp->if_capenable &= ~IFCAP_TSO4;
3495 				ifp->if_hwassist &= ~CSUM_TSO;
3496 			}
3497 		}
3498 		if ((mask & IFCAP_VLAN_HWTSO) != 0 &&
3499 		    (ifp->if_capabilities & IFCAP_VLAN_HWTSO) != 0)
3500 			ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
3501 		if ((mask & IFCAP_VLAN_HWTAGGING) != 0 &&
3502 		    (ifp->if_capabilities & IFCAP_VLAN_HWTAGGING) != 0) {
3503 			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
3504 			/* TSO over VLAN requires VLAN hardware tagging. */
3505 			if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) == 0)
3506 				ifp->if_capenable &= ~IFCAP_VLAN_HWTSO;
3507 			reinit = 1;
3508 		}
3509 		if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0 &&
3510 		    (mask & (IFCAP_HWCSUM | IFCAP_TSO4 |
3511 		    IFCAP_VLAN_HWTSO)) != 0)
3512 				reinit = 1;
3513 		if ((mask & IFCAP_WOL) != 0 &&
3514 		    (ifp->if_capabilities & IFCAP_WOL) != 0) {
3515 			if ((mask & IFCAP_WOL_UCAST) != 0)
3516 				ifp->if_capenable ^= IFCAP_WOL_UCAST;
3517 			if ((mask & IFCAP_WOL_MCAST) != 0)
3518 				ifp->if_capenable ^= IFCAP_WOL_MCAST;
3519 			if ((mask & IFCAP_WOL_MAGIC) != 0)
3520 				ifp->if_capenable ^= IFCAP_WOL_MAGIC;
3521 		}
3522 		if (reinit && ifp->if_drv_flags & IFF_DRV_RUNNING) {
3523 			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3524 			re_init_locked(sc);
3525 		}
3526 		RL_UNLOCK(sc);
3527 		VLAN_CAPABILITIES(ifp);
3528 	    }
3529 		break;
3530 	default:
3531 		error = ether_ioctl(ifp, command, data);
3532 		break;
3533 	}
3534 
3535 	return (error);
3536 }
3537 
3538 static void
3539 re_watchdog(struct rl_softc *sc)
3540 {
3541 	struct ifnet		*ifp;
3542 
3543 	RL_LOCK_ASSERT(sc);
3544 
3545 	if (sc->rl_watchdog_timer == 0 || --sc->rl_watchdog_timer != 0)
3546 		return;
3547 
3548 	ifp = sc->rl_ifp;
3549 	re_txeof(sc);
3550 	if (sc->rl_ldata.rl_tx_free == sc->rl_ldata.rl_tx_desc_cnt) {
3551 		if_printf(ifp, "watchdog timeout (missed Tx interrupts) "
3552 		    "-- recovering\n");
3553 		if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
3554 			re_start_locked(ifp);
3555 		return;
3556 	}
3557 
3558 	if_printf(ifp, "watchdog timeout\n");
3559 	if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
3560 
3561 	re_rxeof(sc, NULL);
3562 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3563 	re_init_locked(sc);
3564 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
3565 		re_start_locked(ifp);
3566 }
3567 
3568 /*
3569  * Stop the adapter and free any mbufs allocated to the
3570  * RX and TX lists.
3571  */
3572 static void
3573 re_stop(struct rl_softc *sc)
3574 {
3575 	int			i;
3576 	struct ifnet		*ifp;
3577 	struct rl_txdesc	*txd;
3578 	struct rl_rxdesc	*rxd;
3579 
3580 	RL_LOCK_ASSERT(sc);
3581 
3582 	ifp = sc->rl_ifp;
3583 
3584 	sc->rl_watchdog_timer = 0;
3585 	callout_stop(&sc->rl_stat_callout);
3586 	ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
3587 
3588 	/*
3589 	 * Disable accepting frames to put RX MAC into idle state.
3590 	 * Otherwise it's possible to get frames while stop command
3591 	 * execution is in progress and controller can DMA the frame
3592 	 * to already freed RX buffer during that period.
3593 	 */
3594 	CSR_WRITE_4(sc, RL_RXCFG, CSR_READ_4(sc, RL_RXCFG) &
3595 	    ~(RL_RXCFG_RX_ALLPHYS | RL_RXCFG_RX_INDIV | RL_RXCFG_RX_MULTI |
3596 	    RL_RXCFG_RX_BROAD));
3597 
3598 	if ((sc->rl_flags & RL_FLAG_8168G_PLUS) != 0) {
3599 		/* Enable RXDV gate. */
3600 		CSR_WRITE_4(sc, RL_MISC, CSR_READ_4(sc, RL_MISC) |
3601 		    0x00080000);
3602 	}
3603 
3604 	if ((sc->rl_flags & RL_FLAG_WAIT_TXPOLL) != 0) {
3605 		for (i = RL_TIMEOUT; i > 0; i--) {
3606 			if ((CSR_READ_1(sc, sc->rl_txstart) &
3607 			    RL_TXSTART_START) == 0)
3608 				break;
3609 			DELAY(20);
3610 		}
3611 		if (i == 0)
3612 			device_printf(sc->rl_dev,
3613 			    "stopping TX poll timed out!\n");
3614 		CSR_WRITE_1(sc, RL_COMMAND, 0x00);
3615 	} else if ((sc->rl_flags & RL_FLAG_CMDSTOP) != 0) {
3616 		CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_STOPREQ | RL_CMD_TX_ENB |
3617 		    RL_CMD_RX_ENB);
3618 		if ((sc->rl_flags & RL_FLAG_CMDSTOP_WAIT_TXQ) != 0) {
3619 			for (i = RL_TIMEOUT; i > 0; i--) {
3620 				if ((CSR_READ_4(sc, RL_TXCFG) &
3621 				    RL_TXCFG_QUEUE_EMPTY) != 0)
3622 					break;
3623 				DELAY(100);
3624 			}
3625 			if (i == 0)
3626 				device_printf(sc->rl_dev,
3627 				   "stopping TXQ timed out!\n");
3628 		}
3629 	} else
3630 		CSR_WRITE_1(sc, RL_COMMAND, 0x00);
3631 	DELAY(1000);
3632 	CSR_WRITE_2(sc, RL_IMR, 0x0000);
3633 	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
3634 
3635 	if (sc->rl_head != NULL) {
3636 		m_freem(sc->rl_head);
3637 		sc->rl_head = sc->rl_tail = NULL;
3638 	}
3639 
3640 	/* Free the TX list buffers. */
3641 	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
3642 		txd = &sc->rl_ldata.rl_tx_desc[i];
3643 		if (txd->tx_m != NULL) {
3644 			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
3645 			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
3646 			bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag,
3647 			    txd->tx_dmamap);
3648 			m_freem(txd->tx_m);
3649 			txd->tx_m = NULL;
3650 		}
3651 	}
3652 
3653 	/* Free the RX list buffers. */
3654 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
3655 		rxd = &sc->rl_ldata.rl_rx_desc[i];
3656 		if (rxd->rx_m != NULL) {
3657 			bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag,
3658 			    rxd->rx_dmamap, BUS_DMASYNC_POSTREAD);
3659 			bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag,
3660 			    rxd->rx_dmamap);
3661 			m_freem(rxd->rx_m);
3662 			rxd->rx_m = NULL;
3663 		}
3664 	}
3665 
3666 	if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
3667 		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
3668 			rxd = &sc->rl_ldata.rl_jrx_desc[i];
3669 			if (rxd->rx_m != NULL) {
3670 				bus_dmamap_sync(sc->rl_ldata.rl_jrx_mtag,
3671 				    rxd->rx_dmamap, BUS_DMASYNC_POSTREAD);
3672 				bus_dmamap_unload(sc->rl_ldata.rl_jrx_mtag,
3673 				    rxd->rx_dmamap);
3674 				m_freem(rxd->rx_m);
3675 				rxd->rx_m = NULL;
3676 			}
3677 		}
3678 	}
3679 }
3680 
3681 /*
3682  * Device suspend routine.  Stop the interface and save some PCI
3683  * settings in case the BIOS doesn't restore them properly on
3684  * resume.
3685  */
3686 static int
3687 re_suspend(device_t dev)
3688 {
3689 	struct rl_softc		*sc;
3690 
3691 	sc = device_get_softc(dev);
3692 
3693 	RL_LOCK(sc);
3694 	re_stop(sc);
3695 	re_setwol(sc);
3696 	sc->suspended = 1;
3697 	RL_UNLOCK(sc);
3698 
3699 	return (0);
3700 }
3701 
3702 /*
3703  * Device resume routine.  Restore some PCI settings in case the BIOS
3704  * doesn't, re-enable busmastering, and restart the interface if
3705  * appropriate.
3706  */
3707 static int
3708 re_resume(device_t dev)
3709 {
3710 	struct rl_softc		*sc;
3711 	struct ifnet		*ifp;
3712 
3713 	sc = device_get_softc(dev);
3714 
3715 	RL_LOCK(sc);
3716 
3717 	ifp = sc->rl_ifp;
3718 	/* Take controller out of sleep mode. */
3719 	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
3720 		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
3721 			CSR_WRITE_1(sc, RL_GPIO,
3722 			    CSR_READ_1(sc, RL_GPIO) | 0x01);
3723 	}
3724 
3725 	/*
3726 	 * Clear WOL matching such that normal Rx filtering
3727 	 * wouldn't interfere with WOL patterns.
3728 	 */
3729 	re_clrwol(sc);
3730 
3731 	/* reinitialize interface if necessary */
3732 	if (ifp->if_flags & IFF_UP)
3733 		re_init_locked(sc);
3734 
3735 	sc->suspended = 0;
3736 	RL_UNLOCK(sc);
3737 
3738 	return (0);
3739 }
3740 
3741 /*
3742  * Stop all chip I/O so that the kernel's probe routines don't
3743  * get confused by errant DMAs when rebooting.
3744  */
3745 static int
3746 re_shutdown(device_t dev)
3747 {
3748 	struct rl_softc		*sc;
3749 
3750 	sc = device_get_softc(dev);
3751 
3752 	RL_LOCK(sc);
3753 	re_stop(sc);
3754 	/*
3755 	 * Mark interface as down since otherwise we will panic if
3756 	 * interrupt comes in later on, which can happen in some
3757 	 * cases.
3758 	 */
3759 	sc->rl_ifp->if_flags &= ~IFF_UP;
3760 	re_setwol(sc);
3761 	RL_UNLOCK(sc);
3762 
3763 	return (0);
3764 }
3765 
3766 static void
3767 re_set_linkspeed(struct rl_softc *sc)
3768 {
3769 	struct mii_softc *miisc;
3770 	struct mii_data *mii;
3771 	int aneg, i, phyno;
3772 
3773 	RL_LOCK_ASSERT(sc);
3774 
3775 	mii = device_get_softc(sc->rl_miibus);
3776 	mii_pollstat(mii);
3777 	aneg = 0;
3778 	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
3779 	    (IFM_ACTIVE | IFM_AVALID)) {
3780 		switch IFM_SUBTYPE(mii->mii_media_active) {
3781 		case IFM_10_T:
3782 		case IFM_100_TX:
3783 			return;
3784 		case IFM_1000_T:
3785 			aneg++;
3786 			break;
3787 		default:
3788 			break;
3789 		}
3790 	}
3791 	miisc = LIST_FIRST(&mii->mii_phys);
3792 	phyno = miisc->mii_phy;
3793 	LIST_FOREACH(miisc, &mii->mii_phys, mii_list)
3794 		PHY_RESET(miisc);
3795 	re_miibus_writereg(sc->rl_dev, phyno, MII_100T2CR, 0);
3796 	re_miibus_writereg(sc->rl_dev, phyno,
3797 	    MII_ANAR, ANAR_TX_FD | ANAR_TX | ANAR_10_FD | ANAR_10 | ANAR_CSMA);
3798 	re_miibus_writereg(sc->rl_dev, phyno,
3799 	    MII_BMCR, BMCR_AUTOEN | BMCR_STARTNEG);
3800 	DELAY(1000);
3801 	if (aneg != 0) {
3802 		/*
3803 		 * Poll link state until re(4) get a 10/100Mbps link.
3804 		 */
3805 		for (i = 0; i < MII_ANEGTICKS_GIGE; i++) {
3806 			mii_pollstat(mii);
3807 			if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID))
3808 			    == (IFM_ACTIVE | IFM_AVALID)) {
3809 				switch (IFM_SUBTYPE(mii->mii_media_active)) {
3810 				case IFM_10_T:
3811 				case IFM_100_TX:
3812 					return;
3813 				default:
3814 					break;
3815 				}
3816 			}
3817 			RL_UNLOCK(sc);
3818 			pause("relnk", hz);
3819 			RL_LOCK(sc);
3820 		}
3821 		if (i == MII_ANEGTICKS_GIGE)
3822 			device_printf(sc->rl_dev,
3823 			    "establishing a link failed, WOL may not work!");
3824 	}
3825 	/*
3826 	 * No link, force MAC to have 100Mbps, full-duplex link.
3827 	 * MAC does not require reprogramming on resolved speed/duplex,
3828 	 * so this is just for completeness.
3829 	 */
3830 	mii->mii_media_status = IFM_AVALID | IFM_ACTIVE;
3831 	mii->mii_media_active = IFM_ETHER | IFM_100_TX | IFM_FDX;
3832 }
3833 
3834 static void
3835 re_setwol(struct rl_softc *sc)
3836 {
3837 	struct ifnet		*ifp;
3838 	int			pmc;
3839 	uint16_t		pmstat;
3840 	uint8_t			v;
3841 
3842 	RL_LOCK_ASSERT(sc);
3843 
3844 	if (pci_find_cap(sc->rl_dev, PCIY_PMG, &pmc) != 0)
3845 		return;
3846 
3847 	ifp = sc->rl_ifp;
3848 	/* Put controller into sleep mode. */
3849 	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
3850 		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
3851 			CSR_WRITE_1(sc, RL_GPIO,
3852 			    CSR_READ_1(sc, RL_GPIO) & ~0x01);
3853 	}
3854 	if ((ifp->if_capenable & IFCAP_WOL) != 0) {
3855 		if ((sc->rl_flags & RL_FLAG_8168G_PLUS) != 0) {
3856 			/* Disable RXDV gate. */
3857 			CSR_WRITE_4(sc, RL_MISC, CSR_READ_4(sc, RL_MISC) &
3858 			    ~0x00080000);
3859 		}
3860 		re_set_rxmode(sc);
3861 		if ((sc->rl_flags & RL_FLAG_WOL_MANLINK) != 0)
3862 			re_set_linkspeed(sc);
3863 		if ((sc->rl_flags & RL_FLAG_WOLRXENB) != 0)
3864 			CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RX_ENB);
3865 	}
3866 	/* Enable config register write. */
3867 	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
3868 
3869 	/* Enable PME. */
3870 	v = CSR_READ_1(sc, sc->rl_cfg1);
3871 	v &= ~RL_CFG1_PME;
3872 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3873 		v |= RL_CFG1_PME;
3874 	CSR_WRITE_1(sc, sc->rl_cfg1, v);
3875 
3876 	v = CSR_READ_1(sc, sc->rl_cfg3);
3877 	v &= ~(RL_CFG3_WOL_LINK | RL_CFG3_WOL_MAGIC);
3878 	if ((ifp->if_capenable & IFCAP_WOL_MAGIC) != 0)
3879 		v |= RL_CFG3_WOL_MAGIC;
3880 	CSR_WRITE_1(sc, sc->rl_cfg3, v);
3881 
3882 	v = CSR_READ_1(sc, sc->rl_cfg5);
3883 	v &= ~(RL_CFG5_WOL_BCAST | RL_CFG5_WOL_MCAST | RL_CFG5_WOL_UCAST |
3884 	    RL_CFG5_WOL_LANWAKE);
3885 	if ((ifp->if_capenable & IFCAP_WOL_UCAST) != 0)
3886 		v |= RL_CFG5_WOL_UCAST;
3887 	if ((ifp->if_capenable & IFCAP_WOL_MCAST) != 0)
3888 		v |= RL_CFG5_WOL_MCAST | RL_CFG5_WOL_BCAST;
3889 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3890 		v |= RL_CFG5_WOL_LANWAKE;
3891 	CSR_WRITE_1(sc, sc->rl_cfg5, v);
3892 
3893 	/* Config register write done. */
3894 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3895 
3896 	if ((ifp->if_capenable & IFCAP_WOL) == 0 &&
3897 	    (sc->rl_flags & RL_FLAG_PHYWAKE_PM) != 0)
3898 		CSR_WRITE_1(sc, RL_PMCH, CSR_READ_1(sc, RL_PMCH) & ~0x80);
3899 	/*
3900 	 * It seems that hardware resets its link speed to 100Mbps in
3901 	 * power down mode so switching to 100Mbps in driver is not
3902 	 * needed.
3903 	 */
3904 
3905 	/* Request PME if WOL is requested. */
3906 	pmstat = pci_read_config(sc->rl_dev, pmc + PCIR_POWER_STATUS, 2);
3907 	pmstat &= ~(PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE);
3908 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3909 		pmstat |= PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE;
3910 	pci_write_config(sc->rl_dev, pmc + PCIR_POWER_STATUS, pmstat, 2);
3911 }
3912 
3913 static void
3914 re_clrwol(struct rl_softc *sc)
3915 {
3916 	int			pmc;
3917 	uint8_t			v;
3918 
3919 	RL_LOCK_ASSERT(sc);
3920 
3921 	if (pci_find_cap(sc->rl_dev, PCIY_PMG, &pmc) != 0)
3922 		return;
3923 
3924 	/* Enable config register write. */
3925 	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
3926 
3927 	v = CSR_READ_1(sc, sc->rl_cfg3);
3928 	v &= ~(RL_CFG3_WOL_LINK | RL_CFG3_WOL_MAGIC);
3929 	CSR_WRITE_1(sc, sc->rl_cfg3, v);
3930 
3931 	/* Config register write done. */
3932 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3933 
3934 	v = CSR_READ_1(sc, sc->rl_cfg5);
3935 	v &= ~(RL_CFG5_WOL_BCAST | RL_CFG5_WOL_MCAST | RL_CFG5_WOL_UCAST);
3936 	v &= ~RL_CFG5_WOL_LANWAKE;
3937 	CSR_WRITE_1(sc, sc->rl_cfg5, v);
3938 }
3939 
3940 static void
3941 re_add_sysctls(struct rl_softc *sc)
3942 {
3943 	struct sysctl_ctx_list	*ctx;
3944 	struct sysctl_oid_list	*children;
3945 	int			error;
3946 
3947 	ctx = device_get_sysctl_ctx(sc->rl_dev);
3948 	children = SYSCTL_CHILDREN(device_get_sysctl_tree(sc->rl_dev));
3949 
3950 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "stats",
3951 	    CTLTYPE_INT | CTLFLAG_RW, sc, 0, re_sysctl_stats, "I",
3952 	    "Statistics Information");
3953 	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) == 0)
3954 		return;
3955 
3956 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "int_rx_mod",
3957 	    CTLTYPE_INT | CTLFLAG_RW, &sc->rl_int_rx_mod, 0,
3958 	    sysctl_hw_re_int_mod, "I", "re RX interrupt moderation");
3959 	/* Pull in device tunables. */
3960 	sc->rl_int_rx_mod = RL_TIMER_DEFAULT;
3961 	error = resource_int_value(device_get_name(sc->rl_dev),
3962 	    device_get_unit(sc->rl_dev), "int_rx_mod", &sc->rl_int_rx_mod);
3963 	if (error == 0) {
3964 		if (sc->rl_int_rx_mod < RL_TIMER_MIN ||
3965 		    sc->rl_int_rx_mod > RL_TIMER_MAX) {
3966 			device_printf(sc->rl_dev, "int_rx_mod value out of "
3967 			    "range; using default: %d\n",
3968 			    RL_TIMER_DEFAULT);
3969 			sc->rl_int_rx_mod = RL_TIMER_DEFAULT;
3970 		}
3971 	}
3972 }
3973 
3974 static int
3975 re_sysctl_stats(SYSCTL_HANDLER_ARGS)
3976 {
3977 	struct rl_softc		*sc;
3978 	struct rl_stats		*stats;
3979 	int			error, i, result;
3980 
3981 	result = -1;
3982 	error = sysctl_handle_int(oidp, &result, 0, req);
3983 	if (error || req->newptr == NULL)
3984 		return (error);
3985 
3986 	if (result == 1) {
3987 		sc = (struct rl_softc *)arg1;
3988 		RL_LOCK(sc);
3989 		if ((sc->rl_ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
3990 			RL_UNLOCK(sc);
3991 			goto done;
3992 		}
3993 		bus_dmamap_sync(sc->rl_ldata.rl_stag,
3994 		    sc->rl_ldata.rl_smap, BUS_DMASYNC_PREREAD);
3995 		CSR_WRITE_4(sc, RL_DUMPSTATS_HI,
3996 		    RL_ADDR_HI(sc->rl_ldata.rl_stats_addr));
3997 		CSR_WRITE_4(sc, RL_DUMPSTATS_LO,
3998 		    RL_ADDR_LO(sc->rl_ldata.rl_stats_addr));
3999 		CSR_WRITE_4(sc, RL_DUMPSTATS_LO,
4000 		    RL_ADDR_LO(sc->rl_ldata.rl_stats_addr |
4001 		    RL_DUMPSTATS_START));
4002 		for (i = RL_TIMEOUT; i > 0; i--) {
4003 			if ((CSR_READ_4(sc, RL_DUMPSTATS_LO) &
4004 			    RL_DUMPSTATS_START) == 0)
4005 				break;
4006 			DELAY(1000);
4007 		}
4008 		bus_dmamap_sync(sc->rl_ldata.rl_stag,
4009 		    sc->rl_ldata.rl_smap, BUS_DMASYNC_POSTREAD);
4010 		RL_UNLOCK(sc);
4011 		if (i == 0) {
4012 			device_printf(sc->rl_dev,
4013 			    "DUMP statistics request timed out\n");
4014 			return (ETIMEDOUT);
4015 		}
4016 done:
4017 		stats = sc->rl_ldata.rl_stats;
4018 		printf("%s statistics:\n", device_get_nameunit(sc->rl_dev));
4019 		printf("Tx frames : %ju\n",
4020 		    (uintmax_t)le64toh(stats->rl_tx_pkts));
4021 		printf("Rx frames : %ju\n",
4022 		    (uintmax_t)le64toh(stats->rl_rx_pkts));
4023 		printf("Tx errors : %ju\n",
4024 		    (uintmax_t)le64toh(stats->rl_tx_errs));
4025 		printf("Rx errors : %u\n",
4026 		    le32toh(stats->rl_rx_errs));
4027 		printf("Rx missed frames : %u\n",
4028 		    (uint32_t)le16toh(stats->rl_missed_pkts));
4029 		printf("Rx frame alignment errs : %u\n",
4030 		    (uint32_t)le16toh(stats->rl_rx_framealign_errs));
4031 		printf("Tx single collisions : %u\n",
4032 		    le32toh(stats->rl_tx_onecoll));
4033 		printf("Tx multiple collisions : %u\n",
4034 		    le32toh(stats->rl_tx_multicolls));
4035 		printf("Rx unicast frames : %ju\n",
4036 		    (uintmax_t)le64toh(stats->rl_rx_ucasts));
4037 		printf("Rx broadcast frames : %ju\n",
4038 		    (uintmax_t)le64toh(stats->rl_rx_bcasts));
4039 		printf("Rx multicast frames : %u\n",
4040 		    le32toh(stats->rl_rx_mcasts));
4041 		printf("Tx aborts : %u\n",
4042 		    (uint32_t)le16toh(stats->rl_tx_aborts));
4043 		printf("Tx underruns : %u\n",
4044 		    (uint32_t)le16toh(stats->rl_rx_underruns));
4045 	}
4046 
4047 	return (error);
4048 }
4049 
4050 static int
4051 sysctl_int_range(SYSCTL_HANDLER_ARGS, int low, int high)
4052 {
4053 	int error, value;
4054 
4055 	if (arg1 == NULL)
4056 		return (EINVAL);
4057 	value = *(int *)arg1;
4058 	error = sysctl_handle_int(oidp, &value, 0, req);
4059 	if (error || req->newptr == NULL)
4060 		return (error);
4061 	if (value < low || value > high)
4062 		return (EINVAL);
4063 	*(int *)arg1 = value;
4064 
4065 	return (0);
4066 }
4067 
4068 static int
4069 sysctl_hw_re_int_mod(SYSCTL_HANDLER_ARGS)
4070 {
4071 
4072 	return (sysctl_int_range(oidp, arg1, arg2, req, RL_TIMER_MIN,
4073 	    RL_TIMER_MAX));
4074 }
4075