xref: /freebsd/sys/dev/re/if_re.c (revision ff0ba87247820afbdfdc1b307c803f7923d0e4d3)
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_8411, RL_8169, "8411", RL_JUMBO_MTU_9K},
242 	{ RL_HWREV_8411B, RL_8169, "8411B", RL_JUMBO_MTU_9K},
243 	{ 0, 0, NULL, 0 }
244 };
245 
246 static int re_probe		(device_t);
247 static int re_attach		(device_t);
248 static int re_detach		(device_t);
249 
250 static int re_encap		(struct rl_softc *, struct mbuf **);
251 
252 static void re_dma_map_addr	(void *, bus_dma_segment_t *, int, int);
253 static int re_allocmem		(device_t, struct rl_softc *);
254 static __inline void re_discard_rxbuf
255 				(struct rl_softc *, int);
256 static int re_newbuf		(struct rl_softc *, int);
257 static int re_jumbo_newbuf	(struct rl_softc *, int);
258 static int re_rx_list_init	(struct rl_softc *);
259 static int re_jrx_list_init	(struct rl_softc *);
260 static int re_tx_list_init	(struct rl_softc *);
261 #ifdef RE_FIXUP_RX
262 static __inline void re_fixup_rx
263 				(struct mbuf *);
264 #endif
265 static int re_rxeof		(struct rl_softc *, int *);
266 static void re_txeof		(struct rl_softc *);
267 #ifdef DEVICE_POLLING
268 static int re_poll		(struct ifnet *, enum poll_cmd, int);
269 static int re_poll_locked	(struct ifnet *, enum poll_cmd, int);
270 #endif
271 static int re_intr		(void *);
272 static void re_intr_msi		(void *);
273 static void re_tick		(void *);
274 static void re_int_task		(void *, int);
275 static void re_start		(struct ifnet *);
276 static void re_start_locked	(struct ifnet *);
277 static int re_ioctl		(struct ifnet *, u_long, caddr_t);
278 static void re_init		(void *);
279 static void re_init_locked	(struct rl_softc *);
280 static void re_stop		(struct rl_softc *);
281 static void re_watchdog		(struct rl_softc *);
282 static int re_suspend		(device_t);
283 static int re_resume		(device_t);
284 static int re_shutdown		(device_t);
285 static int re_ifmedia_upd	(struct ifnet *);
286 static void re_ifmedia_sts	(struct ifnet *, struct ifmediareq *);
287 
288 static void re_eeprom_putbyte	(struct rl_softc *, int);
289 static void re_eeprom_getword	(struct rl_softc *, int, u_int16_t *);
290 static void re_read_eeprom	(struct rl_softc *, caddr_t, int, int);
291 static int re_gmii_readreg	(device_t, int, int);
292 static int re_gmii_writereg	(device_t, int, int, int);
293 
294 static int re_miibus_readreg	(device_t, int, int);
295 static int re_miibus_writereg	(device_t, int, int, int);
296 static void re_miibus_statchg	(device_t);
297 
298 static void re_set_jumbo	(struct rl_softc *, int);
299 static void re_set_rxmode		(struct rl_softc *);
300 static void re_reset		(struct rl_softc *);
301 static void re_setwol		(struct rl_softc *);
302 static void re_clrwol		(struct rl_softc *);
303 static void re_set_linkspeed	(struct rl_softc *);
304 
305 #ifdef DEV_NETMAP	/* see ixgbe.c for details */
306 #include <dev/netmap/if_re_netmap.h>
307 #endif /* !DEV_NETMAP */
308 
309 #ifdef RE_DIAG
310 static int re_diag		(struct rl_softc *);
311 #endif
312 
313 static void re_add_sysctls	(struct rl_softc *);
314 static int re_sysctl_stats	(SYSCTL_HANDLER_ARGS);
315 static int sysctl_int_range	(SYSCTL_HANDLER_ARGS, int, int);
316 static int sysctl_hw_re_int_mod	(SYSCTL_HANDLER_ARGS);
317 
318 static device_method_t re_methods[] = {
319 	/* Device interface */
320 	DEVMETHOD(device_probe,		re_probe),
321 	DEVMETHOD(device_attach,	re_attach),
322 	DEVMETHOD(device_detach,	re_detach),
323 	DEVMETHOD(device_suspend,	re_suspend),
324 	DEVMETHOD(device_resume,	re_resume),
325 	DEVMETHOD(device_shutdown,	re_shutdown),
326 
327 	/* MII interface */
328 	DEVMETHOD(miibus_readreg,	re_miibus_readreg),
329 	DEVMETHOD(miibus_writereg,	re_miibus_writereg),
330 	DEVMETHOD(miibus_statchg,	re_miibus_statchg),
331 
332 	DEVMETHOD_END
333 };
334 
335 static driver_t re_driver = {
336 	"re",
337 	re_methods,
338 	sizeof(struct rl_softc)
339 };
340 
341 static devclass_t re_devclass;
342 
343 DRIVER_MODULE(re, pci, re_driver, re_devclass, 0, 0);
344 DRIVER_MODULE(miibus, re, miibus_driver, miibus_devclass, 0, 0);
345 
346 #define EE_SET(x)					\
347 	CSR_WRITE_1(sc, RL_EECMD,			\
348 		CSR_READ_1(sc, RL_EECMD) | x)
349 
350 #define EE_CLR(x)					\
351 	CSR_WRITE_1(sc, RL_EECMD,			\
352 		CSR_READ_1(sc, RL_EECMD) & ~x)
353 
354 /*
355  * Send a read command and address to the EEPROM, check for ACK.
356  */
357 static void
358 re_eeprom_putbyte(struct rl_softc *sc, int addr)
359 {
360 	int			d, i;
361 
362 	d = addr | (RL_9346_READ << sc->rl_eewidth);
363 
364 	/*
365 	 * Feed in each bit and strobe the clock.
366 	 */
367 
368 	for (i = 1 << (sc->rl_eewidth + 3); i; i >>= 1) {
369 		if (d & i) {
370 			EE_SET(RL_EE_DATAIN);
371 		} else {
372 			EE_CLR(RL_EE_DATAIN);
373 		}
374 		DELAY(100);
375 		EE_SET(RL_EE_CLK);
376 		DELAY(150);
377 		EE_CLR(RL_EE_CLK);
378 		DELAY(100);
379 	}
380 }
381 
382 /*
383  * Read a word of data stored in the EEPROM at address 'addr.'
384  */
385 static void
386 re_eeprom_getword(struct rl_softc *sc, int addr, u_int16_t *dest)
387 {
388 	int			i;
389 	u_int16_t		word = 0;
390 
391 	/*
392 	 * Send address of word we want to read.
393 	 */
394 	re_eeprom_putbyte(sc, addr);
395 
396 	/*
397 	 * Start reading bits from EEPROM.
398 	 */
399 	for (i = 0x8000; i; i >>= 1) {
400 		EE_SET(RL_EE_CLK);
401 		DELAY(100);
402 		if (CSR_READ_1(sc, RL_EECMD) & RL_EE_DATAOUT)
403 			word |= i;
404 		EE_CLR(RL_EE_CLK);
405 		DELAY(100);
406 	}
407 
408 	*dest = word;
409 }
410 
411 /*
412  * Read a sequence of words from the EEPROM.
413  */
414 static void
415 re_read_eeprom(struct rl_softc *sc, caddr_t dest, int off, int cnt)
416 {
417 	int			i;
418 	u_int16_t		word = 0, *ptr;
419 
420 	CSR_SETBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM);
421 
422         DELAY(100);
423 
424 	for (i = 0; i < cnt; i++) {
425 		CSR_SETBIT_1(sc, RL_EECMD, RL_EE_SEL);
426 		re_eeprom_getword(sc, off + i, &word);
427 		CSR_CLRBIT_1(sc, RL_EECMD, RL_EE_SEL);
428 		ptr = (u_int16_t *)(dest + (i * 2));
429                 *ptr = word;
430 	}
431 
432 	CSR_CLRBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM);
433 }
434 
435 static int
436 re_gmii_readreg(device_t dev, int phy, int reg)
437 {
438 	struct rl_softc		*sc;
439 	u_int32_t		rval;
440 	int			i;
441 
442 	sc = device_get_softc(dev);
443 
444 	/* Let the rgephy driver read the GMEDIASTAT register */
445 
446 	if (reg == RL_GMEDIASTAT) {
447 		rval = CSR_READ_1(sc, RL_GMEDIASTAT);
448 		return (rval);
449 	}
450 
451 	CSR_WRITE_4(sc, RL_PHYAR, reg << 16);
452 
453 	for (i = 0; i < RL_PHY_TIMEOUT; i++) {
454 		rval = CSR_READ_4(sc, RL_PHYAR);
455 		if (rval & RL_PHYAR_BUSY)
456 			break;
457 		DELAY(25);
458 	}
459 
460 	if (i == RL_PHY_TIMEOUT) {
461 		device_printf(sc->rl_dev, "PHY read failed\n");
462 		return (0);
463 	}
464 
465 	/*
466 	 * Controller requires a 20us delay to process next MDIO request.
467 	 */
468 	DELAY(20);
469 
470 	return (rval & RL_PHYAR_PHYDATA);
471 }
472 
473 static int
474 re_gmii_writereg(device_t dev, int phy, int reg, int data)
475 {
476 	struct rl_softc		*sc;
477 	u_int32_t		rval;
478 	int			i;
479 
480 	sc = device_get_softc(dev);
481 
482 	CSR_WRITE_4(sc, RL_PHYAR, (reg << 16) |
483 	    (data & RL_PHYAR_PHYDATA) | RL_PHYAR_BUSY);
484 
485 	for (i = 0; i < RL_PHY_TIMEOUT; i++) {
486 		rval = CSR_READ_4(sc, RL_PHYAR);
487 		if (!(rval & RL_PHYAR_BUSY))
488 			break;
489 		DELAY(25);
490 	}
491 
492 	if (i == RL_PHY_TIMEOUT) {
493 		device_printf(sc->rl_dev, "PHY write failed\n");
494 		return (0);
495 	}
496 
497 	/*
498 	 * Controller requires a 20us delay to process next MDIO request.
499 	 */
500 	DELAY(20);
501 
502 	return (0);
503 }
504 
505 static int
506 re_miibus_readreg(device_t dev, int phy, int reg)
507 {
508 	struct rl_softc		*sc;
509 	u_int16_t		rval = 0;
510 	u_int16_t		re8139_reg = 0;
511 
512 	sc = device_get_softc(dev);
513 
514 	if (sc->rl_type == RL_8169) {
515 		rval = re_gmii_readreg(dev, phy, reg);
516 		return (rval);
517 	}
518 
519 	switch (reg) {
520 	case MII_BMCR:
521 		re8139_reg = RL_BMCR;
522 		break;
523 	case MII_BMSR:
524 		re8139_reg = RL_BMSR;
525 		break;
526 	case MII_ANAR:
527 		re8139_reg = RL_ANAR;
528 		break;
529 	case MII_ANER:
530 		re8139_reg = RL_ANER;
531 		break;
532 	case MII_ANLPAR:
533 		re8139_reg = RL_LPAR;
534 		break;
535 	case MII_PHYIDR1:
536 	case MII_PHYIDR2:
537 		return (0);
538 	/*
539 	 * Allow the rlphy driver to read the media status
540 	 * register. If we have a link partner which does not
541 	 * support NWAY, this is the register which will tell
542 	 * us the results of parallel detection.
543 	 */
544 	case RL_MEDIASTAT:
545 		rval = CSR_READ_1(sc, RL_MEDIASTAT);
546 		return (rval);
547 	default:
548 		device_printf(sc->rl_dev, "bad phy register\n");
549 		return (0);
550 	}
551 	rval = CSR_READ_2(sc, re8139_reg);
552 	if (sc->rl_type == RL_8139CPLUS && re8139_reg == RL_BMCR) {
553 		/* 8139C+ has different bit layout. */
554 		rval &= ~(BMCR_LOOP | BMCR_ISO);
555 	}
556 	return (rval);
557 }
558 
559 static int
560 re_miibus_writereg(device_t dev, int phy, int reg, int data)
561 {
562 	struct rl_softc		*sc;
563 	u_int16_t		re8139_reg = 0;
564 	int			rval = 0;
565 
566 	sc = device_get_softc(dev);
567 
568 	if (sc->rl_type == RL_8169) {
569 		rval = re_gmii_writereg(dev, phy, reg, data);
570 		return (rval);
571 	}
572 
573 	switch (reg) {
574 	case MII_BMCR:
575 		re8139_reg = RL_BMCR;
576 		if (sc->rl_type == RL_8139CPLUS) {
577 			/* 8139C+ has different bit layout. */
578 			data &= ~(BMCR_LOOP | BMCR_ISO);
579 		}
580 		break;
581 	case MII_BMSR:
582 		re8139_reg = RL_BMSR;
583 		break;
584 	case MII_ANAR:
585 		re8139_reg = RL_ANAR;
586 		break;
587 	case MII_ANER:
588 		re8139_reg = RL_ANER;
589 		break;
590 	case MII_ANLPAR:
591 		re8139_reg = RL_LPAR;
592 		break;
593 	case MII_PHYIDR1:
594 	case MII_PHYIDR2:
595 		return (0);
596 		break;
597 	default:
598 		device_printf(sc->rl_dev, "bad phy register\n");
599 		return (0);
600 	}
601 	CSR_WRITE_2(sc, re8139_reg, data);
602 	return (0);
603 }
604 
605 static void
606 re_miibus_statchg(device_t dev)
607 {
608 	struct rl_softc		*sc;
609 	struct ifnet		*ifp;
610 	struct mii_data		*mii;
611 
612 	sc = device_get_softc(dev);
613 	mii = device_get_softc(sc->rl_miibus);
614 	ifp = sc->rl_ifp;
615 	if (mii == NULL || ifp == NULL ||
616 	    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
617 		return;
618 
619 	sc->rl_flags &= ~RL_FLAG_LINK;
620 	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
621 	    (IFM_ACTIVE | IFM_AVALID)) {
622 		switch (IFM_SUBTYPE(mii->mii_media_active)) {
623 		case IFM_10_T:
624 		case IFM_100_TX:
625 			sc->rl_flags |= RL_FLAG_LINK;
626 			break;
627 		case IFM_1000_T:
628 			if ((sc->rl_flags & RL_FLAG_FASTETHER) != 0)
629 				break;
630 			sc->rl_flags |= RL_FLAG_LINK;
631 			break;
632 		default:
633 			break;
634 		}
635 	}
636 	/*
637 	 * RealTek controllers does not provide any interface to
638 	 * Tx/Rx MACs for resolved speed, duplex and flow-control
639 	 * parameters.
640 	 */
641 }
642 
643 /*
644  * Set the RX configuration and 64-bit multicast hash filter.
645  */
646 static void
647 re_set_rxmode(struct rl_softc *sc)
648 {
649 	struct ifnet		*ifp;
650 	struct ifmultiaddr	*ifma;
651 	uint32_t		hashes[2] = { 0, 0 };
652 	uint32_t		h, rxfilt;
653 
654 	RL_LOCK_ASSERT(sc);
655 
656 	ifp = sc->rl_ifp;
657 
658 	rxfilt = RL_RXCFG_CONFIG | RL_RXCFG_RX_INDIV | RL_RXCFG_RX_BROAD;
659 	if ((sc->rl_flags & RL_FLAG_EARLYOFF) != 0)
660 		rxfilt |= RL_RXCFG_EARLYOFF;
661 	else if ((sc->rl_flags & RL_FLAG_EARLYOFFV2) != 0)
662 		rxfilt |= RL_RXCFG_EARLYOFFV2;
663 
664 	if (ifp->if_flags & (IFF_ALLMULTI | IFF_PROMISC)) {
665 		if (ifp->if_flags & IFF_PROMISC)
666 			rxfilt |= RL_RXCFG_RX_ALLPHYS;
667 		/*
668 		 * Unlike other hardwares, we have to explicitly set
669 		 * RL_RXCFG_RX_MULTI to receive multicast frames in
670 		 * promiscuous mode.
671 		 */
672 		rxfilt |= RL_RXCFG_RX_MULTI;
673 		hashes[0] = hashes[1] = 0xffffffff;
674 		goto done;
675 	}
676 
677 	if_maddr_rlock(ifp);
678 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
679 		if (ifma->ifma_addr->sa_family != AF_LINK)
680 			continue;
681 		h = ether_crc32_be(LLADDR((struct sockaddr_dl *)
682 		    ifma->ifma_addr), ETHER_ADDR_LEN) >> 26;
683 		if (h < 32)
684 			hashes[0] |= (1 << h);
685 		else
686 			hashes[1] |= (1 << (h - 32));
687 	}
688 	if_maddr_runlock(ifp);
689 
690 	if (hashes[0] != 0 || hashes[1] != 0) {
691 		/*
692 		 * For some unfathomable reason, RealTek decided to
693 		 * reverse the order of the multicast hash registers
694 		 * in the PCI Express parts.  This means we have to
695 		 * write the hash pattern in reverse order for those
696 		 * devices.
697 		 */
698 		if ((sc->rl_flags & RL_FLAG_PCIE) != 0) {
699 			h = bswap32(hashes[0]);
700 			hashes[0] = bswap32(hashes[1]);
701 			hashes[1] = h;
702 		}
703 		rxfilt |= RL_RXCFG_RX_MULTI;
704 	}
705 
706 	if  (sc->rl_hwrev->rl_rev == RL_HWREV_8168F) {
707 		/* Disable multicast filtering due to silicon bug. */
708 		hashes[0] = 0xffffffff;
709 		hashes[1] = 0xffffffff;
710 	}
711 
712 done:
713 	CSR_WRITE_4(sc, RL_MAR0, hashes[0]);
714 	CSR_WRITE_4(sc, RL_MAR4, hashes[1]);
715 	CSR_WRITE_4(sc, RL_RXCFG, rxfilt);
716 }
717 
718 static void
719 re_reset(struct rl_softc *sc)
720 {
721 	int			i;
722 
723 	RL_LOCK_ASSERT(sc);
724 
725 	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RESET);
726 
727 	for (i = 0; i < RL_TIMEOUT; i++) {
728 		DELAY(10);
729 		if (!(CSR_READ_1(sc, RL_COMMAND) & RL_CMD_RESET))
730 			break;
731 	}
732 	if (i == RL_TIMEOUT)
733 		device_printf(sc->rl_dev, "reset never completed!\n");
734 
735 	if ((sc->rl_flags & RL_FLAG_MACRESET) != 0)
736 		CSR_WRITE_1(sc, 0x82, 1);
737 	if (sc->rl_hwrev->rl_rev == RL_HWREV_8169S)
738 		re_gmii_writereg(sc->rl_dev, 1, 0x0b, 0);
739 }
740 
741 #ifdef RE_DIAG
742 
743 /*
744  * The following routine is designed to test for a defect on some
745  * 32-bit 8169 cards. Some of these NICs have the REQ64# and ACK64#
746  * lines connected to the bus, however for a 32-bit only card, they
747  * should be pulled high. The result of this defect is that the
748  * NIC will not work right if you plug it into a 64-bit slot: DMA
749  * operations will be done with 64-bit transfers, which will fail
750  * because the 64-bit data lines aren't connected.
751  *
752  * There's no way to work around this (short of talking a soldering
753  * iron to the board), however we can detect it. The method we use
754  * here is to put the NIC into digital loopback mode, set the receiver
755  * to promiscuous mode, and then try to send a frame. We then compare
756  * the frame data we sent to what was received. If the data matches,
757  * then the NIC is working correctly, otherwise we know the user has
758  * a defective NIC which has been mistakenly plugged into a 64-bit PCI
759  * slot. In the latter case, there's no way the NIC can work correctly,
760  * so we print out a message on the console and abort the device attach.
761  */
762 
763 static int
764 re_diag(struct rl_softc *sc)
765 {
766 	struct ifnet		*ifp = sc->rl_ifp;
767 	struct mbuf		*m0;
768 	struct ether_header	*eh;
769 	struct rl_desc		*cur_rx;
770 	u_int16_t		status;
771 	u_int32_t		rxstat;
772 	int			total_len, i, error = 0, phyaddr;
773 	u_int8_t		dst[] = { 0x00, 'h', 'e', 'l', 'l', 'o' };
774 	u_int8_t		src[] = { 0x00, 'w', 'o', 'r', 'l', 'd' };
775 
776 	/* Allocate a single mbuf */
777 	MGETHDR(m0, M_NOWAIT, MT_DATA);
778 	if (m0 == NULL)
779 		return (ENOBUFS);
780 
781 	RL_LOCK(sc);
782 
783 	/*
784 	 * Initialize the NIC in test mode. This sets the chip up
785 	 * so that it can send and receive frames, but performs the
786 	 * following special functions:
787 	 * - Puts receiver in promiscuous mode
788 	 * - Enables digital loopback mode
789 	 * - Leaves interrupts turned off
790 	 */
791 
792 	ifp->if_flags |= IFF_PROMISC;
793 	sc->rl_testmode = 1;
794 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
795 	re_init_locked(sc);
796 	sc->rl_flags |= RL_FLAG_LINK;
797 	if (sc->rl_type == RL_8169)
798 		phyaddr = 1;
799 	else
800 		phyaddr = 0;
801 
802 	re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_RESET);
803 	for (i = 0; i < RL_TIMEOUT; i++) {
804 		status = re_miibus_readreg(sc->rl_dev, phyaddr, MII_BMCR);
805 		if (!(status & BMCR_RESET))
806 			break;
807 	}
808 
809 	re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_LOOP);
810 	CSR_WRITE_2(sc, RL_ISR, RL_INTRS);
811 
812 	DELAY(100000);
813 
814 	/* Put some data in the mbuf */
815 
816 	eh = mtod(m0, struct ether_header *);
817 	bcopy ((char *)&dst, eh->ether_dhost, ETHER_ADDR_LEN);
818 	bcopy ((char *)&src, eh->ether_shost, ETHER_ADDR_LEN);
819 	eh->ether_type = htons(ETHERTYPE_IP);
820 	m0->m_pkthdr.len = m0->m_len = ETHER_MIN_LEN - ETHER_CRC_LEN;
821 
822 	/*
823 	 * Queue the packet, start transmission.
824 	 * Note: IF_HANDOFF() ultimately calls re_start() for us.
825 	 */
826 
827 	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
828 	RL_UNLOCK(sc);
829 	/* XXX: re_diag must not be called when in ALTQ mode */
830 	IF_HANDOFF(&ifp->if_snd, m0, ifp);
831 	RL_LOCK(sc);
832 	m0 = NULL;
833 
834 	/* Wait for it to propagate through the chip */
835 
836 	DELAY(100000);
837 	for (i = 0; i < RL_TIMEOUT; i++) {
838 		status = CSR_READ_2(sc, RL_ISR);
839 		CSR_WRITE_2(sc, RL_ISR, status);
840 		if ((status & (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK)) ==
841 		    (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK))
842 			break;
843 		DELAY(10);
844 	}
845 
846 	if (i == RL_TIMEOUT) {
847 		device_printf(sc->rl_dev,
848 		    "diagnostic failed, failed to receive packet in"
849 		    " loopback mode\n");
850 		error = EIO;
851 		goto done;
852 	}
853 
854 	/*
855 	 * The packet should have been dumped into the first
856 	 * entry in the RX DMA ring. Grab it from there.
857 	 */
858 
859 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
860 	    sc->rl_ldata.rl_rx_list_map,
861 	    BUS_DMASYNC_POSTREAD);
862 	bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag,
863 	    sc->rl_ldata.rl_rx_desc[0].rx_dmamap,
864 	    BUS_DMASYNC_POSTREAD);
865 	bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag,
866 	    sc->rl_ldata.rl_rx_desc[0].rx_dmamap);
867 
868 	m0 = sc->rl_ldata.rl_rx_desc[0].rx_m;
869 	sc->rl_ldata.rl_rx_desc[0].rx_m = NULL;
870 	eh = mtod(m0, struct ether_header *);
871 
872 	cur_rx = &sc->rl_ldata.rl_rx_list[0];
873 	total_len = RL_RXBYTES(cur_rx);
874 	rxstat = le32toh(cur_rx->rl_cmdstat);
875 
876 	if (total_len != ETHER_MIN_LEN) {
877 		device_printf(sc->rl_dev,
878 		    "diagnostic failed, received short packet\n");
879 		error = EIO;
880 		goto done;
881 	}
882 
883 	/* Test that the received packet data matches what we sent. */
884 
885 	if (bcmp((char *)&eh->ether_dhost, (char *)&dst, ETHER_ADDR_LEN) ||
886 	    bcmp((char *)&eh->ether_shost, (char *)&src, ETHER_ADDR_LEN) ||
887 	    ntohs(eh->ether_type) != ETHERTYPE_IP) {
888 		device_printf(sc->rl_dev, "WARNING, DMA FAILURE!\n");
889 		device_printf(sc->rl_dev, "expected TX data: %6D/%6D/0x%x\n",
890 		    dst, ":", src, ":", ETHERTYPE_IP);
891 		device_printf(sc->rl_dev, "received RX data: %6D/%6D/0x%x\n",
892 		    eh->ether_dhost, ":", eh->ether_shost, ":",
893 		    ntohs(eh->ether_type));
894 		device_printf(sc->rl_dev, "You may have a defective 32-bit "
895 		    "NIC plugged into a 64-bit PCI slot.\n");
896 		device_printf(sc->rl_dev, "Please re-install the NIC in a "
897 		    "32-bit slot for proper operation.\n");
898 		device_printf(sc->rl_dev, "Read the re(4) man page for more "
899 		    "details.\n");
900 		error = EIO;
901 	}
902 
903 done:
904 	/* Turn interface off, release resources */
905 
906 	sc->rl_testmode = 0;
907 	sc->rl_flags &= ~RL_FLAG_LINK;
908 	ifp->if_flags &= ~IFF_PROMISC;
909 	re_stop(sc);
910 	if (m0 != NULL)
911 		m_freem(m0);
912 
913 	RL_UNLOCK(sc);
914 
915 	return (error);
916 }
917 
918 #endif
919 
920 /*
921  * Probe for a RealTek 8139C+/8169/8110 chip. Check the PCI vendor and device
922  * IDs against our list and return a device name if we find a match.
923  */
924 static int
925 re_probe(device_t dev)
926 {
927 	const struct rl_type	*t;
928 	uint16_t		devid, vendor;
929 	uint16_t		revid, sdevid;
930 	int			i;
931 
932 	vendor = pci_get_vendor(dev);
933 	devid = pci_get_device(dev);
934 	revid = pci_get_revid(dev);
935 	sdevid = pci_get_subdevice(dev);
936 
937 	if (vendor == LINKSYS_VENDORID && devid == LINKSYS_DEVICEID_EG1032) {
938 		if (sdevid != LINKSYS_SUBDEVICE_EG1032_REV3) {
939 			/*
940 			 * Only attach to rev. 3 of the Linksys EG1032 adapter.
941 			 * Rev. 2 is supported by sk(4).
942 			 */
943 			return (ENXIO);
944 		}
945 	}
946 
947 	if (vendor == RT_VENDORID && devid == RT_DEVICEID_8139) {
948 		if (revid != 0x20) {
949 			/* 8139, let rl(4) take care of this device. */
950 			return (ENXIO);
951 		}
952 	}
953 
954 	t = re_devs;
955 	for (i = 0; i < sizeof(re_devs) / sizeof(re_devs[0]); i++, t++) {
956 		if (vendor == t->rl_vid && devid == t->rl_did) {
957 			device_set_desc(dev, t->rl_name);
958 			return (BUS_PROBE_DEFAULT);
959 		}
960 	}
961 
962 	return (ENXIO);
963 }
964 
965 /*
966  * Map a single buffer address.
967  */
968 
969 static void
970 re_dma_map_addr(void *arg, bus_dma_segment_t *segs, int nseg, int error)
971 {
972 	bus_addr_t		*addr;
973 
974 	if (error)
975 		return;
976 
977 	KASSERT(nseg == 1, ("too many DMA segments, %d should be 1", nseg));
978 	addr = arg;
979 	*addr = segs->ds_addr;
980 }
981 
982 static int
983 re_allocmem(device_t dev, struct rl_softc *sc)
984 {
985 	bus_addr_t		lowaddr;
986 	bus_size_t		rx_list_size, tx_list_size;
987 	int			error;
988 	int			i;
989 
990 	rx_list_size = sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc);
991 	tx_list_size = sc->rl_ldata.rl_tx_desc_cnt * sizeof(struct rl_desc);
992 
993 	/*
994 	 * Allocate the parent bus DMA tag appropriate for PCI.
995 	 * In order to use DAC, RL_CPLUSCMD_PCI_DAC bit of RL_CPLUS_CMD
996 	 * register should be set. However some RealTek chips are known
997 	 * to be buggy on DAC handling, therefore disable DAC by limiting
998 	 * DMA address space to 32bit. PCIe variants of RealTek chips
999 	 * may not have the limitation.
1000 	 */
1001 	lowaddr = BUS_SPACE_MAXADDR;
1002 	if ((sc->rl_flags & RL_FLAG_PCIE) == 0)
1003 		lowaddr = BUS_SPACE_MAXADDR_32BIT;
1004 	error = bus_dma_tag_create(bus_get_dma_tag(dev), 1, 0,
1005 	    lowaddr, BUS_SPACE_MAXADDR, NULL, NULL,
1006 	    BUS_SPACE_MAXSIZE_32BIT, 0, BUS_SPACE_MAXSIZE_32BIT, 0,
1007 	    NULL, NULL, &sc->rl_parent_tag);
1008 	if (error) {
1009 		device_printf(dev, "could not allocate parent DMA tag\n");
1010 		return (error);
1011 	}
1012 
1013 	/*
1014 	 * Allocate map for TX mbufs.
1015 	 */
1016 	error = bus_dma_tag_create(sc->rl_parent_tag, 1, 0,
1017 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL,
1018 	    NULL, MCLBYTES * RL_NTXSEGS, RL_NTXSEGS, 4096, 0,
1019 	    NULL, NULL, &sc->rl_ldata.rl_tx_mtag);
1020 	if (error) {
1021 		device_printf(dev, "could not allocate TX DMA tag\n");
1022 		return (error);
1023 	}
1024 
1025 	/*
1026 	 * Allocate map for RX mbufs.
1027 	 */
1028 
1029 	if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
1030 		error = bus_dma_tag_create(sc->rl_parent_tag, sizeof(uint64_t),
1031 		    0, BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
1032 		    MJUM9BYTES, 1, MJUM9BYTES, 0, NULL, NULL,
1033 		    &sc->rl_ldata.rl_jrx_mtag);
1034 		if (error) {
1035 			device_printf(dev,
1036 			    "could not allocate jumbo RX DMA tag\n");
1037 			return (error);
1038 		}
1039 	}
1040 	error = bus_dma_tag_create(sc->rl_parent_tag, sizeof(uint64_t), 0,
1041 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
1042 	    MCLBYTES, 1, MCLBYTES, 0, NULL, NULL, &sc->rl_ldata.rl_rx_mtag);
1043 	if (error) {
1044 		device_printf(dev, "could not allocate RX DMA tag\n");
1045 		return (error);
1046 	}
1047 
1048 	/*
1049 	 * Allocate map for TX descriptor list.
1050 	 */
1051 	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
1052 	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1053 	    NULL, tx_list_size, 1, tx_list_size, 0,
1054 	    NULL, NULL, &sc->rl_ldata.rl_tx_list_tag);
1055 	if (error) {
1056 		device_printf(dev, "could not allocate TX DMA ring tag\n");
1057 		return (error);
1058 	}
1059 
1060 	/* Allocate DMA'able memory for the TX ring */
1061 
1062 	error = bus_dmamem_alloc(sc->rl_ldata.rl_tx_list_tag,
1063 	    (void **)&sc->rl_ldata.rl_tx_list,
1064 	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1065 	    &sc->rl_ldata.rl_tx_list_map);
1066 	if (error) {
1067 		device_printf(dev, "could not allocate TX DMA ring\n");
1068 		return (error);
1069 	}
1070 
1071 	/* Load the map for the TX ring. */
1072 
1073 	sc->rl_ldata.rl_tx_list_addr = 0;
1074 	error = bus_dmamap_load(sc->rl_ldata.rl_tx_list_tag,
1075 	     sc->rl_ldata.rl_tx_list_map, sc->rl_ldata.rl_tx_list,
1076 	     tx_list_size, re_dma_map_addr,
1077 	     &sc->rl_ldata.rl_tx_list_addr, BUS_DMA_NOWAIT);
1078 	if (error != 0 || sc->rl_ldata.rl_tx_list_addr == 0) {
1079 		device_printf(dev, "could not load TX DMA ring\n");
1080 		return (ENOMEM);
1081 	}
1082 
1083 	/* Create DMA maps for TX buffers */
1084 
1085 	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
1086 		error = bus_dmamap_create(sc->rl_ldata.rl_tx_mtag, 0,
1087 		    &sc->rl_ldata.rl_tx_desc[i].tx_dmamap);
1088 		if (error) {
1089 			device_printf(dev, "could not create DMA map for TX\n");
1090 			return (error);
1091 		}
1092 	}
1093 
1094 	/*
1095 	 * Allocate map for RX descriptor list.
1096 	 */
1097 	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
1098 	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1099 	    NULL, rx_list_size, 1, rx_list_size, 0,
1100 	    NULL, NULL, &sc->rl_ldata.rl_rx_list_tag);
1101 	if (error) {
1102 		device_printf(dev, "could not create RX DMA ring tag\n");
1103 		return (error);
1104 	}
1105 
1106 	/* Allocate DMA'able memory for the RX ring */
1107 
1108 	error = bus_dmamem_alloc(sc->rl_ldata.rl_rx_list_tag,
1109 	    (void **)&sc->rl_ldata.rl_rx_list,
1110 	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1111 	    &sc->rl_ldata.rl_rx_list_map);
1112 	if (error) {
1113 		device_printf(dev, "could not allocate RX DMA ring\n");
1114 		return (error);
1115 	}
1116 
1117 	/* Load the map for the RX ring. */
1118 
1119 	sc->rl_ldata.rl_rx_list_addr = 0;
1120 	error = bus_dmamap_load(sc->rl_ldata.rl_rx_list_tag,
1121 	     sc->rl_ldata.rl_rx_list_map, sc->rl_ldata.rl_rx_list,
1122 	     rx_list_size, re_dma_map_addr,
1123 	     &sc->rl_ldata.rl_rx_list_addr, BUS_DMA_NOWAIT);
1124 	if (error != 0 || sc->rl_ldata.rl_rx_list_addr == 0) {
1125 		device_printf(dev, "could not load RX DMA ring\n");
1126 		return (ENOMEM);
1127 	}
1128 
1129 	/* Create DMA maps for RX buffers */
1130 
1131 	if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
1132 		error = bus_dmamap_create(sc->rl_ldata.rl_jrx_mtag, 0,
1133 		    &sc->rl_ldata.rl_jrx_sparemap);
1134 		if (error) {
1135 			device_printf(dev,
1136 			    "could not create spare DMA map for jumbo RX\n");
1137 			return (error);
1138 		}
1139 		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1140 			error = bus_dmamap_create(sc->rl_ldata.rl_jrx_mtag, 0,
1141 			    &sc->rl_ldata.rl_jrx_desc[i].rx_dmamap);
1142 			if (error) {
1143 				device_printf(dev,
1144 				    "could not create DMA map for jumbo RX\n");
1145 				return (error);
1146 			}
1147 		}
1148 	}
1149 	error = bus_dmamap_create(sc->rl_ldata.rl_rx_mtag, 0,
1150 	    &sc->rl_ldata.rl_rx_sparemap);
1151 	if (error) {
1152 		device_printf(dev, "could not create spare DMA map for RX\n");
1153 		return (error);
1154 	}
1155 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1156 		error = bus_dmamap_create(sc->rl_ldata.rl_rx_mtag, 0,
1157 		    &sc->rl_ldata.rl_rx_desc[i].rx_dmamap);
1158 		if (error) {
1159 			device_printf(dev, "could not create DMA map for RX\n");
1160 			return (error);
1161 		}
1162 	}
1163 
1164 	/* Create DMA map for statistics. */
1165 	error = bus_dma_tag_create(sc->rl_parent_tag, RL_DUMP_ALIGN, 0,
1166 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
1167 	    sizeof(struct rl_stats), 1, sizeof(struct rl_stats), 0, NULL, NULL,
1168 	    &sc->rl_ldata.rl_stag);
1169 	if (error) {
1170 		device_printf(dev, "could not create statistics DMA tag\n");
1171 		return (error);
1172 	}
1173 	/* Allocate DMA'able memory for statistics. */
1174 	error = bus_dmamem_alloc(sc->rl_ldata.rl_stag,
1175 	    (void **)&sc->rl_ldata.rl_stats,
1176 	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1177 	    &sc->rl_ldata.rl_smap);
1178 	if (error) {
1179 		device_printf(dev,
1180 		    "could not allocate statistics DMA memory\n");
1181 		return (error);
1182 	}
1183 	/* Load the map for statistics. */
1184 	sc->rl_ldata.rl_stats_addr = 0;
1185 	error = bus_dmamap_load(sc->rl_ldata.rl_stag, sc->rl_ldata.rl_smap,
1186 	    sc->rl_ldata.rl_stats, sizeof(struct rl_stats), re_dma_map_addr,
1187 	     &sc->rl_ldata.rl_stats_addr, BUS_DMA_NOWAIT);
1188 	if (error != 0 || sc->rl_ldata.rl_stats_addr == 0) {
1189 		device_printf(dev, "could not load statistics DMA memory\n");
1190 		return (ENOMEM);
1191 	}
1192 
1193 	return (0);
1194 }
1195 
1196 /*
1197  * Attach the interface. Allocate softc structures, do ifmedia
1198  * setup and ethernet/BPF attach.
1199  */
1200 static int
1201 re_attach(device_t dev)
1202 {
1203 	u_char			eaddr[ETHER_ADDR_LEN];
1204 	u_int16_t		as[ETHER_ADDR_LEN / 2];
1205 	struct rl_softc		*sc;
1206 	struct ifnet		*ifp;
1207 	const struct rl_hwrev	*hw_rev;
1208 	u_int32_t		cap, ctl;
1209 	int			hwrev;
1210 	u_int16_t		devid, re_did = 0;
1211 	int			error = 0, i, phy, rid;
1212 	int			msic, msixc, reg;
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_EARLYOFFV2 | RL_FLAG_RXDV_GATED;
1493 		break;
1494 	case RL_HWREV_8168GU:
1495 		if (pci_get_device(dev) == RT_DEVICEID_8101E) {
1496 			/* RTL8106EUS */
1497 			sc->rl_flags |= RL_FLAG_FASTETHER;
1498 		} else
1499 			sc->rl_flags |= RL_FLAG_JUMBOV2 | RL_FLAG_WOL_MANLINK;
1500 
1501 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
1502 		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP |
1503 		    RL_FLAG_AUTOPAD | RL_FLAG_CMDSTOP_WAIT_TXQ |
1504 		    RL_FLAG_EARLYOFFV2 | RL_FLAG_RXDV_GATED;
1505 		break;
1506 	case RL_HWREV_8169_8110SB:
1507 	case RL_HWREV_8169_8110SBL:
1508 	case RL_HWREV_8169_8110SC:
1509 	case RL_HWREV_8169_8110SCE:
1510 		sc->rl_flags |= RL_FLAG_PHYWAKE;
1511 		/* FALLTHROUGH */
1512 	case RL_HWREV_8169:
1513 	case RL_HWREV_8169S:
1514 	case RL_HWREV_8110S:
1515 		sc->rl_flags |= RL_FLAG_MACRESET;
1516 		break;
1517 	default:
1518 		break;
1519 	}
1520 
1521 	if (sc->rl_hwrev->rl_rev == RL_HWREV_8139CPLUS) {
1522 		sc->rl_cfg0 = RL_8139_CFG0;
1523 		sc->rl_cfg1 = RL_8139_CFG1;
1524 		sc->rl_cfg2 = 0;
1525 		sc->rl_cfg3 = RL_8139_CFG3;
1526 		sc->rl_cfg4 = RL_8139_CFG4;
1527 		sc->rl_cfg5 = RL_8139_CFG5;
1528 	} else {
1529 		sc->rl_cfg0 = RL_CFG0;
1530 		sc->rl_cfg1 = RL_CFG1;
1531 		sc->rl_cfg2 = RL_CFG2;
1532 		sc->rl_cfg3 = RL_CFG3;
1533 		sc->rl_cfg4 = RL_CFG4;
1534 		sc->rl_cfg5 = RL_CFG5;
1535 	}
1536 
1537 	/* Reset the adapter. */
1538 	RL_LOCK(sc);
1539 	re_reset(sc);
1540 	RL_UNLOCK(sc);
1541 
1542 	/* Enable PME. */
1543 	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1544 	cfg = CSR_READ_1(sc, sc->rl_cfg1);
1545 	cfg |= RL_CFG1_PME;
1546 	CSR_WRITE_1(sc, sc->rl_cfg1, cfg);
1547 	cfg = CSR_READ_1(sc, sc->rl_cfg5);
1548 	cfg &= RL_CFG5_PME_STS;
1549 	CSR_WRITE_1(sc, sc->rl_cfg5, cfg);
1550 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1551 
1552 	if ((sc->rl_flags & RL_FLAG_PAR) != 0) {
1553 		/*
1554 		 * XXX Should have a better way to extract station
1555 		 * address from EEPROM.
1556 		 */
1557 		for (i = 0; i < ETHER_ADDR_LEN; i++)
1558 			eaddr[i] = CSR_READ_1(sc, RL_IDR0 + i);
1559 	} else {
1560 		sc->rl_eewidth = RL_9356_ADDR_LEN;
1561 		re_read_eeprom(sc, (caddr_t)&re_did, 0, 1);
1562 		if (re_did != 0x8129)
1563 			sc->rl_eewidth = RL_9346_ADDR_LEN;
1564 
1565 		/*
1566 		 * Get station address from the EEPROM.
1567 		 */
1568 		re_read_eeprom(sc, (caddr_t)as, RL_EE_EADDR, 3);
1569 		for (i = 0; i < ETHER_ADDR_LEN / 2; i++)
1570 			as[i] = le16toh(as[i]);
1571 		bcopy(as, eaddr, ETHER_ADDR_LEN);
1572 	}
1573 
1574 	if (sc->rl_type == RL_8169) {
1575 		/* Set RX length mask and number of descriptors. */
1576 		sc->rl_rxlenmask = RL_RDESC_STAT_GFRAGLEN;
1577 		sc->rl_txstart = RL_GTXSTART;
1578 		sc->rl_ldata.rl_tx_desc_cnt = RL_8169_TX_DESC_CNT;
1579 		sc->rl_ldata.rl_rx_desc_cnt = RL_8169_RX_DESC_CNT;
1580 	} else {
1581 		/* Set RX length mask and number of descriptors. */
1582 		sc->rl_rxlenmask = RL_RDESC_STAT_FRAGLEN;
1583 		sc->rl_txstart = RL_TXSTART;
1584 		sc->rl_ldata.rl_tx_desc_cnt = RL_8139_TX_DESC_CNT;
1585 		sc->rl_ldata.rl_rx_desc_cnt = RL_8139_RX_DESC_CNT;
1586 	}
1587 
1588 	error = re_allocmem(dev, sc);
1589 	if (error)
1590 		goto fail;
1591 	re_add_sysctls(sc);
1592 
1593 	ifp = sc->rl_ifp = if_alloc(IFT_ETHER);
1594 	if (ifp == NULL) {
1595 		device_printf(dev, "can not if_alloc()\n");
1596 		error = ENOSPC;
1597 		goto fail;
1598 	}
1599 
1600 	/* Take controller out of deep sleep mode. */
1601 	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
1602 		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
1603 			CSR_WRITE_1(sc, RL_GPIO,
1604 			    CSR_READ_1(sc, RL_GPIO) | 0x01);
1605 		else
1606 			CSR_WRITE_1(sc, RL_GPIO,
1607 			    CSR_READ_1(sc, RL_GPIO) & ~0x01);
1608 	}
1609 
1610 	/* Take PHY out of power down mode. */
1611 	if ((sc->rl_flags & RL_FLAG_PHYWAKE_PM) != 0) {
1612 		CSR_WRITE_1(sc, RL_PMCH, CSR_READ_1(sc, RL_PMCH) | 0x80);
1613 		if (hw_rev->rl_rev == RL_HWREV_8401E)
1614 			CSR_WRITE_1(sc, 0xD1, CSR_READ_1(sc, 0xD1) & ~0x08);
1615 	}
1616 	if ((sc->rl_flags & RL_FLAG_PHYWAKE) != 0) {
1617 		re_gmii_writereg(dev, 1, 0x1f, 0);
1618 		re_gmii_writereg(dev, 1, 0x0e, 0);
1619 	}
1620 
1621 	ifp->if_softc = sc;
1622 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1623 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1624 	ifp->if_ioctl = re_ioctl;
1625 	ifp->if_start = re_start;
1626 	/*
1627 	 * RTL8168/8111C generates wrong IP checksummed frame if the
1628 	 * packet has IP options so disable TX checksum offloading.
1629 	 */
1630 	if (sc->rl_hwrev->rl_rev == RL_HWREV_8168C ||
1631 	    sc->rl_hwrev->rl_rev == RL_HWREV_8168C_SPIN2 ||
1632 	    sc->rl_hwrev->rl_rev == RL_HWREV_8168CP) {
1633 		ifp->if_hwassist = 0;
1634 		ifp->if_capabilities = IFCAP_RXCSUM | IFCAP_TSO4;
1635 	} else {
1636 		ifp->if_hwassist = CSUM_IP | CSUM_TCP | CSUM_UDP;
1637 		ifp->if_capabilities = IFCAP_HWCSUM | IFCAP_TSO4;
1638 	}
1639 	ifp->if_hwassist |= CSUM_TSO;
1640 	ifp->if_capenable = ifp->if_capabilities;
1641 	ifp->if_init = re_init;
1642 	IFQ_SET_MAXLEN(&ifp->if_snd, RL_IFQ_MAXLEN);
1643 	ifp->if_snd.ifq_drv_maxlen = RL_IFQ_MAXLEN;
1644 	IFQ_SET_READY(&ifp->if_snd);
1645 
1646 	TASK_INIT(&sc->rl_inttask, 0, re_int_task, sc);
1647 
1648 #define	RE_PHYAD_INTERNAL	 0
1649 
1650 	/* Do MII setup. */
1651 	phy = RE_PHYAD_INTERNAL;
1652 	if (sc->rl_type == RL_8169)
1653 		phy = 1;
1654 	error = mii_attach(dev, &sc->rl_miibus, ifp, re_ifmedia_upd,
1655 	    re_ifmedia_sts, BMSR_DEFCAPMASK, phy, MII_OFFSET_ANY, MIIF_DOPAUSE);
1656 	if (error != 0) {
1657 		device_printf(dev, "attaching PHYs failed\n");
1658 		goto fail;
1659 	}
1660 
1661 	/*
1662 	 * Call MI attach routine.
1663 	 */
1664 	ether_ifattach(ifp, eaddr);
1665 
1666 	/* VLAN capability setup */
1667 	ifp->if_capabilities |= IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING;
1668 	if (ifp->if_capabilities & IFCAP_HWCSUM)
1669 		ifp->if_capabilities |= IFCAP_VLAN_HWCSUM;
1670 	/* Enable WOL if PM is supported. */
1671 	if (pci_find_cap(sc->rl_dev, PCIY_PMG, &reg) == 0)
1672 		ifp->if_capabilities |= IFCAP_WOL;
1673 	ifp->if_capenable = ifp->if_capabilities;
1674 	ifp->if_capenable &= ~(IFCAP_WOL_UCAST | IFCAP_WOL_MCAST);
1675 	/*
1676 	 * Don't enable TSO by default.  It is known to generate
1677 	 * corrupted TCP segments(bad TCP options) under certain
1678 	 * circumstances.
1679 	 */
1680 	ifp->if_hwassist &= ~CSUM_TSO;
1681 	ifp->if_capenable &= ~(IFCAP_TSO4 | IFCAP_VLAN_HWTSO);
1682 #ifdef DEVICE_POLLING
1683 	ifp->if_capabilities |= IFCAP_POLLING;
1684 #endif
1685 	/*
1686 	 * Tell the upper layer(s) we support long frames.
1687 	 * Must appear after the call to ether_ifattach() because
1688 	 * ether_ifattach() sets ifi_hdrlen to the default value.
1689 	 */
1690 	ifp->if_hdrlen = sizeof(struct ether_vlan_header);
1691 
1692 #ifdef DEV_NETMAP
1693 	re_netmap_attach(sc);
1694 #endif /* DEV_NETMAP */
1695 #ifdef RE_DIAG
1696 	/*
1697 	 * Perform hardware diagnostic on the original RTL8169.
1698 	 * Some 32-bit cards were incorrectly wired and would
1699 	 * malfunction if plugged into a 64-bit slot.
1700 	 */
1701 
1702 	if (hwrev == RL_HWREV_8169) {
1703 		error = re_diag(sc);
1704 		if (error) {
1705 			device_printf(dev,
1706 		    	"attach aborted due to hardware diag failure\n");
1707 			ether_ifdetach(ifp);
1708 			goto fail;
1709 		}
1710 	}
1711 #endif
1712 
1713 #ifdef RE_TX_MODERATION
1714 	intr_filter = 1;
1715 #endif
1716 	/* Hook interrupt last to avoid having to lock softc */
1717 	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) != 0 &&
1718 	    intr_filter == 0) {
1719 		error = bus_setup_intr(dev, sc->rl_irq[0],
1720 		    INTR_TYPE_NET | INTR_MPSAFE, NULL, re_intr_msi, sc,
1721 		    &sc->rl_intrhand[0]);
1722 	} else {
1723 		error = bus_setup_intr(dev, sc->rl_irq[0],
1724 		    INTR_TYPE_NET | INTR_MPSAFE, re_intr, NULL, sc,
1725 		    &sc->rl_intrhand[0]);
1726 	}
1727 	if (error) {
1728 		device_printf(dev, "couldn't set up irq\n");
1729 		ether_ifdetach(ifp);
1730 	}
1731 
1732 fail:
1733 
1734 	if (error)
1735 		re_detach(dev);
1736 
1737 	return (error);
1738 }
1739 
1740 /*
1741  * Shutdown hardware and free up resources. This can be called any
1742  * time after the mutex has been initialized. It is called in both
1743  * the error case in attach and the normal detach case so it needs
1744  * to be careful about only freeing resources that have actually been
1745  * allocated.
1746  */
1747 static int
1748 re_detach(device_t dev)
1749 {
1750 	struct rl_softc		*sc;
1751 	struct ifnet		*ifp;
1752 	int			i, rid;
1753 
1754 	sc = device_get_softc(dev);
1755 	ifp = sc->rl_ifp;
1756 	KASSERT(mtx_initialized(&sc->rl_mtx), ("re mutex not initialized"));
1757 
1758 	/* These should only be active if attach succeeded */
1759 	if (device_is_attached(dev)) {
1760 #ifdef DEVICE_POLLING
1761 		if (ifp->if_capenable & IFCAP_POLLING)
1762 			ether_poll_deregister(ifp);
1763 #endif
1764 		RL_LOCK(sc);
1765 #if 0
1766 		sc->suspended = 1;
1767 #endif
1768 		re_stop(sc);
1769 		RL_UNLOCK(sc);
1770 		callout_drain(&sc->rl_stat_callout);
1771 		taskqueue_drain(taskqueue_fast, &sc->rl_inttask);
1772 		/*
1773 		 * Force off the IFF_UP flag here, in case someone
1774 		 * still had a BPF descriptor attached to this
1775 		 * interface. If they do, ether_ifdetach() will cause
1776 		 * the BPF code to try and clear the promisc mode
1777 		 * flag, which will bubble down to re_ioctl(),
1778 		 * which will try to call re_init() again. This will
1779 		 * turn the NIC back on and restart the MII ticker,
1780 		 * which will panic the system when the kernel tries
1781 		 * to invoke the re_tick() function that isn't there
1782 		 * anymore.
1783 		 */
1784 		ifp->if_flags &= ~IFF_UP;
1785 		ether_ifdetach(ifp);
1786 	}
1787 	if (sc->rl_miibus)
1788 		device_delete_child(dev, sc->rl_miibus);
1789 	bus_generic_detach(dev);
1790 
1791 	/*
1792 	 * The rest is resource deallocation, so we should already be
1793 	 * stopped here.
1794 	 */
1795 
1796 	if (sc->rl_intrhand[0] != NULL) {
1797 		bus_teardown_intr(dev, sc->rl_irq[0], sc->rl_intrhand[0]);
1798 		sc->rl_intrhand[0] = NULL;
1799 	}
1800 	if (ifp != NULL) {
1801 #ifdef DEV_NETMAP
1802 		netmap_detach(ifp);
1803 #endif /* DEV_NETMAP */
1804 		if_free(ifp);
1805 	}
1806 	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) == 0)
1807 		rid = 0;
1808 	else
1809 		rid = 1;
1810 	if (sc->rl_irq[0] != NULL) {
1811 		bus_release_resource(dev, SYS_RES_IRQ, rid, sc->rl_irq[0]);
1812 		sc->rl_irq[0] = NULL;
1813 	}
1814 	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) != 0)
1815 		pci_release_msi(dev);
1816 	if (sc->rl_res_pba) {
1817 		rid = PCIR_BAR(4);
1818 		bus_release_resource(dev, SYS_RES_MEMORY, rid, sc->rl_res_pba);
1819 	}
1820 	if (sc->rl_res)
1821 		bus_release_resource(dev, sc->rl_res_type, sc->rl_res_id,
1822 		    sc->rl_res);
1823 
1824 	/* Unload and free the RX DMA ring memory and map */
1825 
1826 	if (sc->rl_ldata.rl_rx_list_tag) {
1827 		if (sc->rl_ldata.rl_rx_list_addr)
1828 			bus_dmamap_unload(sc->rl_ldata.rl_rx_list_tag,
1829 			    sc->rl_ldata.rl_rx_list_map);
1830 		if (sc->rl_ldata.rl_rx_list)
1831 			bus_dmamem_free(sc->rl_ldata.rl_rx_list_tag,
1832 			    sc->rl_ldata.rl_rx_list,
1833 			    sc->rl_ldata.rl_rx_list_map);
1834 		bus_dma_tag_destroy(sc->rl_ldata.rl_rx_list_tag);
1835 	}
1836 
1837 	/* Unload and free the TX DMA ring memory and map */
1838 
1839 	if (sc->rl_ldata.rl_tx_list_tag) {
1840 		if (sc->rl_ldata.rl_tx_list_addr)
1841 			bus_dmamap_unload(sc->rl_ldata.rl_tx_list_tag,
1842 			    sc->rl_ldata.rl_tx_list_map);
1843 		if (sc->rl_ldata.rl_tx_list)
1844 			bus_dmamem_free(sc->rl_ldata.rl_tx_list_tag,
1845 			    sc->rl_ldata.rl_tx_list,
1846 			    sc->rl_ldata.rl_tx_list_map);
1847 		bus_dma_tag_destroy(sc->rl_ldata.rl_tx_list_tag);
1848 	}
1849 
1850 	/* Destroy all the RX and TX buffer maps */
1851 
1852 	if (sc->rl_ldata.rl_tx_mtag) {
1853 		for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
1854 			if (sc->rl_ldata.rl_tx_desc[i].tx_dmamap)
1855 				bus_dmamap_destroy(sc->rl_ldata.rl_tx_mtag,
1856 				    sc->rl_ldata.rl_tx_desc[i].tx_dmamap);
1857 		}
1858 		bus_dma_tag_destroy(sc->rl_ldata.rl_tx_mtag);
1859 	}
1860 	if (sc->rl_ldata.rl_rx_mtag) {
1861 		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1862 			if (sc->rl_ldata.rl_rx_desc[i].rx_dmamap)
1863 				bus_dmamap_destroy(sc->rl_ldata.rl_rx_mtag,
1864 				    sc->rl_ldata.rl_rx_desc[i].rx_dmamap);
1865 		}
1866 		if (sc->rl_ldata.rl_rx_sparemap)
1867 			bus_dmamap_destroy(sc->rl_ldata.rl_rx_mtag,
1868 			    sc->rl_ldata.rl_rx_sparemap);
1869 		bus_dma_tag_destroy(sc->rl_ldata.rl_rx_mtag);
1870 	}
1871 	if (sc->rl_ldata.rl_jrx_mtag) {
1872 		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1873 			if (sc->rl_ldata.rl_jrx_desc[i].rx_dmamap)
1874 				bus_dmamap_destroy(sc->rl_ldata.rl_jrx_mtag,
1875 				    sc->rl_ldata.rl_jrx_desc[i].rx_dmamap);
1876 		}
1877 		if (sc->rl_ldata.rl_jrx_sparemap)
1878 			bus_dmamap_destroy(sc->rl_ldata.rl_jrx_mtag,
1879 			    sc->rl_ldata.rl_jrx_sparemap);
1880 		bus_dma_tag_destroy(sc->rl_ldata.rl_jrx_mtag);
1881 	}
1882 	/* Unload and free the stats buffer and map */
1883 
1884 	if (sc->rl_ldata.rl_stag) {
1885 		if (sc->rl_ldata.rl_stats_addr)
1886 			bus_dmamap_unload(sc->rl_ldata.rl_stag,
1887 			    sc->rl_ldata.rl_smap);
1888 		if (sc->rl_ldata.rl_stats)
1889 			bus_dmamem_free(sc->rl_ldata.rl_stag,
1890 			    sc->rl_ldata.rl_stats, sc->rl_ldata.rl_smap);
1891 		bus_dma_tag_destroy(sc->rl_ldata.rl_stag);
1892 	}
1893 
1894 	if (sc->rl_parent_tag)
1895 		bus_dma_tag_destroy(sc->rl_parent_tag);
1896 
1897 	mtx_destroy(&sc->rl_mtx);
1898 
1899 	return (0);
1900 }
1901 
1902 static __inline void
1903 re_discard_rxbuf(struct rl_softc *sc, int idx)
1904 {
1905 	struct rl_desc		*desc;
1906 	struct rl_rxdesc	*rxd;
1907 	uint32_t		cmdstat;
1908 
1909 	if (sc->rl_ifp->if_mtu > RL_MTU &&
1910 	    (sc->rl_flags & RL_FLAG_JUMBOV2) != 0)
1911 		rxd = &sc->rl_ldata.rl_jrx_desc[idx];
1912 	else
1913 		rxd = &sc->rl_ldata.rl_rx_desc[idx];
1914 	desc = &sc->rl_ldata.rl_rx_list[idx];
1915 	desc->rl_vlanctl = 0;
1916 	cmdstat = rxd->rx_size;
1917 	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
1918 		cmdstat |= RL_RDESC_CMD_EOR;
1919 	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
1920 }
1921 
1922 static int
1923 re_newbuf(struct rl_softc *sc, int idx)
1924 {
1925 	struct mbuf		*m;
1926 	struct rl_rxdesc	*rxd;
1927 	bus_dma_segment_t	segs[1];
1928 	bus_dmamap_t		map;
1929 	struct rl_desc		*desc;
1930 	uint32_t		cmdstat;
1931 	int			error, nsegs;
1932 
1933 	m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
1934 	if (m == NULL)
1935 		return (ENOBUFS);
1936 
1937 	m->m_len = m->m_pkthdr.len = MCLBYTES;
1938 #ifdef RE_FIXUP_RX
1939 	/*
1940 	 * This is part of an evil trick to deal with non-x86 platforms.
1941 	 * The RealTek chip requires RX buffers to be aligned on 64-bit
1942 	 * boundaries, but that will hose non-x86 machines. To get around
1943 	 * this, we leave some empty space at the start of each buffer
1944 	 * and for non-x86 hosts, we copy the buffer back six bytes
1945 	 * to achieve word alignment. This is slightly more efficient
1946 	 * than allocating a new buffer, copying the contents, and
1947 	 * discarding the old buffer.
1948 	 */
1949 	m_adj(m, RE_ETHER_ALIGN);
1950 #endif
1951 	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_rx_mtag,
1952 	    sc->rl_ldata.rl_rx_sparemap, m, segs, &nsegs, BUS_DMA_NOWAIT);
1953 	if (error != 0) {
1954 		m_freem(m);
1955 		return (ENOBUFS);
1956 	}
1957 	KASSERT(nsegs == 1, ("%s: %d segment returned!", __func__, nsegs));
1958 
1959 	rxd = &sc->rl_ldata.rl_rx_desc[idx];
1960 	if (rxd->rx_m != NULL) {
1961 		bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap,
1962 		    BUS_DMASYNC_POSTREAD);
1963 		bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap);
1964 	}
1965 
1966 	rxd->rx_m = m;
1967 	map = rxd->rx_dmamap;
1968 	rxd->rx_dmamap = sc->rl_ldata.rl_rx_sparemap;
1969 	rxd->rx_size = segs[0].ds_len;
1970 	sc->rl_ldata.rl_rx_sparemap = map;
1971 	bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap,
1972 	    BUS_DMASYNC_PREREAD);
1973 
1974 	desc = &sc->rl_ldata.rl_rx_list[idx];
1975 	desc->rl_vlanctl = 0;
1976 	desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[0].ds_addr));
1977 	desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[0].ds_addr));
1978 	cmdstat = segs[0].ds_len;
1979 	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
1980 		cmdstat |= RL_RDESC_CMD_EOR;
1981 	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
1982 
1983 	return (0);
1984 }
1985 
1986 static int
1987 re_jumbo_newbuf(struct rl_softc *sc, int idx)
1988 {
1989 	struct mbuf		*m;
1990 	struct rl_rxdesc	*rxd;
1991 	bus_dma_segment_t	segs[1];
1992 	bus_dmamap_t		map;
1993 	struct rl_desc		*desc;
1994 	uint32_t		cmdstat;
1995 	int			error, nsegs;
1996 
1997 	m = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, MJUM9BYTES);
1998 	if (m == NULL)
1999 		return (ENOBUFS);
2000 	m->m_len = m->m_pkthdr.len = MJUM9BYTES;
2001 #ifdef RE_FIXUP_RX
2002 	m_adj(m, RE_ETHER_ALIGN);
2003 #endif
2004 	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_jrx_mtag,
2005 	    sc->rl_ldata.rl_jrx_sparemap, m, segs, &nsegs, BUS_DMA_NOWAIT);
2006 	if (error != 0) {
2007 		m_freem(m);
2008 		return (ENOBUFS);
2009 	}
2010 	KASSERT(nsegs == 1, ("%s: %d segment returned!", __func__, nsegs));
2011 
2012 	rxd = &sc->rl_ldata.rl_jrx_desc[idx];
2013 	if (rxd->rx_m != NULL) {
2014 		bus_dmamap_sync(sc->rl_ldata.rl_jrx_mtag, rxd->rx_dmamap,
2015 		    BUS_DMASYNC_POSTREAD);
2016 		bus_dmamap_unload(sc->rl_ldata.rl_jrx_mtag, rxd->rx_dmamap);
2017 	}
2018 
2019 	rxd->rx_m = m;
2020 	map = rxd->rx_dmamap;
2021 	rxd->rx_dmamap = sc->rl_ldata.rl_jrx_sparemap;
2022 	rxd->rx_size = segs[0].ds_len;
2023 	sc->rl_ldata.rl_jrx_sparemap = map;
2024 	bus_dmamap_sync(sc->rl_ldata.rl_jrx_mtag, rxd->rx_dmamap,
2025 	    BUS_DMASYNC_PREREAD);
2026 
2027 	desc = &sc->rl_ldata.rl_rx_list[idx];
2028 	desc->rl_vlanctl = 0;
2029 	desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[0].ds_addr));
2030 	desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[0].ds_addr));
2031 	cmdstat = segs[0].ds_len;
2032 	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
2033 		cmdstat |= RL_RDESC_CMD_EOR;
2034 	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
2035 
2036 	return (0);
2037 }
2038 
2039 #ifdef RE_FIXUP_RX
2040 static __inline void
2041 re_fixup_rx(struct mbuf *m)
2042 {
2043 	int                     i;
2044 	uint16_t                *src, *dst;
2045 
2046 	src = mtod(m, uint16_t *);
2047 	dst = src - (RE_ETHER_ALIGN - ETHER_ALIGN) / sizeof *src;
2048 
2049 	for (i = 0; i < (m->m_len / sizeof(uint16_t) + 1); i++)
2050 		*dst++ = *src++;
2051 
2052 	m->m_data -= RE_ETHER_ALIGN - ETHER_ALIGN;
2053 }
2054 #endif
2055 
2056 static int
2057 re_tx_list_init(struct rl_softc *sc)
2058 {
2059 	struct rl_desc		*desc;
2060 	int			i;
2061 
2062 	RL_LOCK_ASSERT(sc);
2063 
2064 	bzero(sc->rl_ldata.rl_tx_list,
2065 	    sc->rl_ldata.rl_tx_desc_cnt * sizeof(struct rl_desc));
2066 	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++)
2067 		sc->rl_ldata.rl_tx_desc[i].tx_m = NULL;
2068 #ifdef DEV_NETMAP
2069 	re_netmap_tx_init(sc);
2070 #endif /* DEV_NETMAP */
2071 	/* Set EOR. */
2072 	desc = &sc->rl_ldata.rl_tx_list[sc->rl_ldata.rl_tx_desc_cnt - 1];
2073 	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_EOR);
2074 
2075 	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2076 	    sc->rl_ldata.rl_tx_list_map,
2077 	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
2078 
2079 	sc->rl_ldata.rl_tx_prodidx = 0;
2080 	sc->rl_ldata.rl_tx_considx = 0;
2081 	sc->rl_ldata.rl_tx_free = sc->rl_ldata.rl_tx_desc_cnt;
2082 
2083 	return (0);
2084 }
2085 
2086 static int
2087 re_rx_list_init(struct rl_softc *sc)
2088 {
2089 	int			error, i;
2090 
2091 	bzero(sc->rl_ldata.rl_rx_list,
2092 	    sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc));
2093 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
2094 		sc->rl_ldata.rl_rx_desc[i].rx_m = NULL;
2095 		if ((error = re_newbuf(sc, i)) != 0)
2096 			return (error);
2097 	}
2098 #ifdef DEV_NETMAP
2099 	re_netmap_rx_init(sc);
2100 #endif /* DEV_NETMAP */
2101 
2102 	/* Flush the RX descriptors */
2103 
2104 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
2105 	    sc->rl_ldata.rl_rx_list_map,
2106 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2107 
2108 	sc->rl_ldata.rl_rx_prodidx = 0;
2109 	sc->rl_head = sc->rl_tail = NULL;
2110 	sc->rl_int_rx_act = 0;
2111 
2112 	return (0);
2113 }
2114 
2115 static int
2116 re_jrx_list_init(struct rl_softc *sc)
2117 {
2118 	int			error, i;
2119 
2120 	bzero(sc->rl_ldata.rl_rx_list,
2121 	    sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc));
2122 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
2123 		sc->rl_ldata.rl_jrx_desc[i].rx_m = NULL;
2124 		if ((error = re_jumbo_newbuf(sc, i)) != 0)
2125 			return (error);
2126 	}
2127 
2128 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
2129 	    sc->rl_ldata.rl_rx_list_map,
2130 	    BUS_DMASYNC_PREWRITE | BUS_DMASYNC_PREREAD);
2131 
2132 	sc->rl_ldata.rl_rx_prodidx = 0;
2133 	sc->rl_head = sc->rl_tail = NULL;
2134 	sc->rl_int_rx_act = 0;
2135 
2136 	return (0);
2137 }
2138 
2139 /*
2140  * RX handler for C+ and 8169. For the gigE chips, we support
2141  * the reception of jumbo frames that have been fragmented
2142  * across multiple 2K mbuf cluster buffers.
2143  */
2144 static int
2145 re_rxeof(struct rl_softc *sc, int *rx_npktsp)
2146 {
2147 	struct mbuf		*m;
2148 	struct ifnet		*ifp;
2149 	int			i, rxerr, total_len;
2150 	struct rl_desc		*cur_rx;
2151 	u_int32_t		rxstat, rxvlan;
2152 	int			jumbo, maxpkt = 16, rx_npkts = 0;
2153 
2154 	RL_LOCK_ASSERT(sc);
2155 
2156 	ifp = sc->rl_ifp;
2157 #ifdef DEV_NETMAP
2158 	if (netmap_rx_irq(ifp, 0, &rx_npkts))
2159 		return 0;
2160 #endif /* DEV_NETMAP */
2161 	if (ifp->if_mtu > RL_MTU && (sc->rl_flags & RL_FLAG_JUMBOV2) != 0)
2162 		jumbo = 1;
2163 	else
2164 		jumbo = 0;
2165 
2166 	/* Invalidate the descriptor memory */
2167 
2168 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
2169 	    sc->rl_ldata.rl_rx_list_map,
2170 	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2171 
2172 	for (i = sc->rl_ldata.rl_rx_prodidx; maxpkt > 0;
2173 	    i = RL_RX_DESC_NXT(sc, i)) {
2174 		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
2175 			break;
2176 		cur_rx = &sc->rl_ldata.rl_rx_list[i];
2177 		rxstat = le32toh(cur_rx->rl_cmdstat);
2178 		if ((rxstat & RL_RDESC_STAT_OWN) != 0)
2179 			break;
2180 		total_len = rxstat & sc->rl_rxlenmask;
2181 		rxvlan = le32toh(cur_rx->rl_vlanctl);
2182 		if (jumbo != 0)
2183 			m = sc->rl_ldata.rl_jrx_desc[i].rx_m;
2184 		else
2185 			m = sc->rl_ldata.rl_rx_desc[i].rx_m;
2186 
2187 		if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0 &&
2188 		    (rxstat & (RL_RDESC_STAT_SOF | RL_RDESC_STAT_EOF)) !=
2189 		    (RL_RDESC_STAT_SOF | RL_RDESC_STAT_EOF)) {
2190 			/*
2191 			 * RTL8168C or later controllers do not
2192 			 * support multi-fragment packet.
2193 			 */
2194 			re_discard_rxbuf(sc, i);
2195 			continue;
2196 		} else if ((rxstat & RL_RDESC_STAT_EOF) == 0) {
2197 			if (re_newbuf(sc, i) != 0) {
2198 				/*
2199 				 * If this is part of a multi-fragment packet,
2200 				 * discard all the pieces.
2201 				 */
2202 				if (sc->rl_head != NULL) {
2203 					m_freem(sc->rl_head);
2204 					sc->rl_head = sc->rl_tail = NULL;
2205 				}
2206 				re_discard_rxbuf(sc, i);
2207 				continue;
2208 			}
2209 			m->m_len = RE_RX_DESC_BUFLEN;
2210 			if (sc->rl_head == NULL)
2211 				sc->rl_head = sc->rl_tail = m;
2212 			else {
2213 				m->m_flags &= ~M_PKTHDR;
2214 				sc->rl_tail->m_next = m;
2215 				sc->rl_tail = m;
2216 			}
2217 			continue;
2218 		}
2219 
2220 		/*
2221 		 * NOTE: for the 8139C+, the frame length field
2222 		 * is always 12 bits in size, but for the gigE chips,
2223 		 * it is 13 bits (since the max RX frame length is 16K).
2224 		 * Unfortunately, all 32 bits in the status word
2225 		 * were already used, so to make room for the extra
2226 		 * length bit, RealTek took out the 'frame alignment
2227 		 * error' bit and shifted the other status bits
2228 		 * over one slot. The OWN, EOR, FS and LS bits are
2229 		 * still in the same places. We have already extracted
2230 		 * the frame length and checked the OWN bit, so rather
2231 		 * than using an alternate bit mapping, we shift the
2232 		 * status bits one space to the right so we can evaluate
2233 		 * them using the 8169 status as though it was in the
2234 		 * same format as that of the 8139C+.
2235 		 */
2236 		if (sc->rl_type == RL_8169)
2237 			rxstat >>= 1;
2238 
2239 		/*
2240 		 * if total_len > 2^13-1, both _RXERRSUM and _GIANT will be
2241 		 * set, but if CRC is clear, it will still be a valid frame.
2242 		 */
2243 		if ((rxstat & RL_RDESC_STAT_RXERRSUM) != 0) {
2244 			rxerr = 1;
2245 			if ((sc->rl_flags & RL_FLAG_JUMBOV2) == 0 &&
2246 			    total_len > 8191 &&
2247 			    (rxstat & RL_RDESC_STAT_ERRS) == RL_RDESC_STAT_GIANT)
2248 				rxerr = 0;
2249 			if (rxerr != 0) {
2250 				if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
2251 				/*
2252 				 * If this is part of a multi-fragment packet,
2253 				 * discard all the pieces.
2254 				 */
2255 				if (sc->rl_head != NULL) {
2256 					m_freem(sc->rl_head);
2257 					sc->rl_head = sc->rl_tail = NULL;
2258 				}
2259 				re_discard_rxbuf(sc, i);
2260 				continue;
2261 			}
2262 		}
2263 
2264 		/*
2265 		 * If allocating a replacement mbuf fails,
2266 		 * reload the current one.
2267 		 */
2268 		if (jumbo != 0)
2269 			rxerr = re_jumbo_newbuf(sc, i);
2270 		else
2271 			rxerr = re_newbuf(sc, i);
2272 		if (rxerr != 0) {
2273 			if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
2274 			if (sc->rl_head != NULL) {
2275 				m_freem(sc->rl_head);
2276 				sc->rl_head = sc->rl_tail = NULL;
2277 			}
2278 			re_discard_rxbuf(sc, i);
2279 			continue;
2280 		}
2281 
2282 		if (sc->rl_head != NULL) {
2283 			if (jumbo != 0)
2284 				m->m_len = total_len;
2285 			else {
2286 				m->m_len = total_len % RE_RX_DESC_BUFLEN;
2287 				if (m->m_len == 0)
2288 					m->m_len = RE_RX_DESC_BUFLEN;
2289 			}
2290 			/*
2291 			 * Special case: if there's 4 bytes or less
2292 			 * in this buffer, the mbuf can be discarded:
2293 			 * the last 4 bytes is the CRC, which we don't
2294 			 * care about anyway.
2295 			 */
2296 			if (m->m_len <= ETHER_CRC_LEN) {
2297 				sc->rl_tail->m_len -=
2298 				    (ETHER_CRC_LEN - m->m_len);
2299 				m_freem(m);
2300 			} else {
2301 				m->m_len -= ETHER_CRC_LEN;
2302 				m->m_flags &= ~M_PKTHDR;
2303 				sc->rl_tail->m_next = m;
2304 			}
2305 			m = sc->rl_head;
2306 			sc->rl_head = sc->rl_tail = NULL;
2307 			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
2308 		} else
2309 			m->m_pkthdr.len = m->m_len =
2310 			    (total_len - ETHER_CRC_LEN);
2311 
2312 #ifdef RE_FIXUP_RX
2313 		re_fixup_rx(m);
2314 #endif
2315 		if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
2316 		m->m_pkthdr.rcvif = ifp;
2317 
2318 		/* Do RX checksumming if enabled */
2319 
2320 		if (ifp->if_capenable & IFCAP_RXCSUM) {
2321 			if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
2322 				/* Check IP header checksum */
2323 				if (rxstat & RL_RDESC_STAT_PROTOID)
2324 					m->m_pkthdr.csum_flags |=
2325 					    CSUM_IP_CHECKED;
2326 				if (!(rxstat & RL_RDESC_STAT_IPSUMBAD))
2327 					m->m_pkthdr.csum_flags |=
2328 					    CSUM_IP_VALID;
2329 
2330 				/* Check TCP/UDP checksum */
2331 				if ((RL_TCPPKT(rxstat) &&
2332 				    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
2333 				    (RL_UDPPKT(rxstat) &&
2334 				     !(rxstat & RL_RDESC_STAT_UDPSUMBAD))) {
2335 					m->m_pkthdr.csum_flags |=
2336 						CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
2337 					m->m_pkthdr.csum_data = 0xffff;
2338 				}
2339 			} else {
2340 				/*
2341 				 * RTL8168C/RTL816CP/RTL8111C/RTL8111CP
2342 				 */
2343 				if ((rxstat & RL_RDESC_STAT_PROTOID) &&
2344 				    (rxvlan & RL_RDESC_IPV4))
2345 					m->m_pkthdr.csum_flags |=
2346 					    CSUM_IP_CHECKED;
2347 				if (!(rxstat & RL_RDESC_STAT_IPSUMBAD) &&
2348 				    (rxvlan & RL_RDESC_IPV4))
2349 					m->m_pkthdr.csum_flags |=
2350 					    CSUM_IP_VALID;
2351 				if (((rxstat & RL_RDESC_STAT_TCP) &&
2352 				    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
2353 				    ((rxstat & RL_RDESC_STAT_UDP) &&
2354 				    !(rxstat & RL_RDESC_STAT_UDPSUMBAD))) {
2355 					m->m_pkthdr.csum_flags |=
2356 						CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
2357 					m->m_pkthdr.csum_data = 0xffff;
2358 				}
2359 			}
2360 		}
2361 		maxpkt--;
2362 		if (rxvlan & RL_RDESC_VLANCTL_TAG) {
2363 			m->m_pkthdr.ether_vtag =
2364 			    bswap16((rxvlan & RL_RDESC_VLANCTL_DATA));
2365 			m->m_flags |= M_VLANTAG;
2366 		}
2367 		RL_UNLOCK(sc);
2368 		(*ifp->if_input)(ifp, m);
2369 		RL_LOCK(sc);
2370 		rx_npkts++;
2371 	}
2372 
2373 	/* Flush the RX DMA ring */
2374 
2375 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
2376 	    sc->rl_ldata.rl_rx_list_map,
2377 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2378 
2379 	sc->rl_ldata.rl_rx_prodidx = i;
2380 
2381 	if (rx_npktsp != NULL)
2382 		*rx_npktsp = rx_npkts;
2383 	if (maxpkt)
2384 		return (EAGAIN);
2385 
2386 	return (0);
2387 }
2388 
2389 static void
2390 re_txeof(struct rl_softc *sc)
2391 {
2392 	struct ifnet		*ifp;
2393 	struct rl_txdesc	*txd;
2394 	u_int32_t		txstat;
2395 	int			cons;
2396 
2397 	cons = sc->rl_ldata.rl_tx_considx;
2398 	if (cons == sc->rl_ldata.rl_tx_prodidx)
2399 		return;
2400 
2401 	ifp = sc->rl_ifp;
2402 #ifdef DEV_NETMAP
2403 	if (netmap_tx_irq(ifp, 0))
2404 		return;
2405 #endif /* DEV_NETMAP */
2406 	/* Invalidate the TX descriptor list */
2407 	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2408 	    sc->rl_ldata.rl_tx_list_map,
2409 	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2410 
2411 	for (; cons != sc->rl_ldata.rl_tx_prodidx;
2412 	    cons = RL_TX_DESC_NXT(sc, cons)) {
2413 		txstat = le32toh(sc->rl_ldata.rl_tx_list[cons].rl_cmdstat);
2414 		if (txstat & RL_TDESC_STAT_OWN)
2415 			break;
2416 		/*
2417 		 * We only stash mbufs in the last descriptor
2418 		 * in a fragment chain, which also happens to
2419 		 * be the only place where the TX status bits
2420 		 * are valid.
2421 		 */
2422 		if (txstat & RL_TDESC_CMD_EOF) {
2423 			txd = &sc->rl_ldata.rl_tx_desc[cons];
2424 			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
2425 			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
2426 			bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag,
2427 			    txd->tx_dmamap);
2428 			KASSERT(txd->tx_m != NULL,
2429 			    ("%s: freeing NULL mbufs!", __func__));
2430 			m_freem(txd->tx_m);
2431 			txd->tx_m = NULL;
2432 			if (txstat & (RL_TDESC_STAT_EXCESSCOL|
2433 			    RL_TDESC_STAT_COLCNT))
2434 				if_inc_counter(ifp, IFCOUNTER_COLLISIONS, 1);
2435 			if (txstat & RL_TDESC_STAT_TXERRSUM)
2436 				if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
2437 			else
2438 				if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
2439 		}
2440 		sc->rl_ldata.rl_tx_free++;
2441 		ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
2442 	}
2443 	sc->rl_ldata.rl_tx_considx = cons;
2444 
2445 	/* No changes made to the TX ring, so no flush needed */
2446 
2447 	if (sc->rl_ldata.rl_tx_free != sc->rl_ldata.rl_tx_desc_cnt) {
2448 #ifdef RE_TX_MODERATION
2449 		/*
2450 		 * If not all descriptors have been reaped yet, reload
2451 		 * the timer so that we will eventually get another
2452 		 * interrupt that will cause us to re-enter this routine.
2453 		 * This is done in case the transmitter has gone idle.
2454 		 */
2455 		CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2456 #endif
2457 	} else
2458 		sc->rl_watchdog_timer = 0;
2459 }
2460 
2461 static void
2462 re_tick(void *xsc)
2463 {
2464 	struct rl_softc		*sc;
2465 	struct mii_data		*mii;
2466 
2467 	sc = xsc;
2468 
2469 	RL_LOCK_ASSERT(sc);
2470 
2471 	mii = device_get_softc(sc->rl_miibus);
2472 	mii_tick(mii);
2473 	if ((sc->rl_flags & RL_FLAG_LINK) == 0)
2474 		re_miibus_statchg(sc->rl_dev);
2475 	/*
2476 	 * Reclaim transmitted frames here. Technically it is not
2477 	 * necessary to do here but it ensures periodic reclamation
2478 	 * regardless of Tx completion interrupt which seems to be
2479 	 * lost on PCIe based controllers under certain situations.
2480 	 */
2481 	re_txeof(sc);
2482 	re_watchdog(sc);
2483 	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
2484 }
2485 
2486 #ifdef DEVICE_POLLING
2487 static int
2488 re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
2489 {
2490 	struct rl_softc *sc = ifp->if_softc;
2491 	int rx_npkts = 0;
2492 
2493 	RL_LOCK(sc);
2494 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
2495 		rx_npkts = re_poll_locked(ifp, cmd, count);
2496 	RL_UNLOCK(sc);
2497 	return (rx_npkts);
2498 }
2499 
2500 static int
2501 re_poll_locked(struct ifnet *ifp, enum poll_cmd cmd, int count)
2502 {
2503 	struct rl_softc *sc = ifp->if_softc;
2504 	int rx_npkts;
2505 
2506 	RL_LOCK_ASSERT(sc);
2507 
2508 	sc->rxcycles = count;
2509 	re_rxeof(sc, &rx_npkts);
2510 	re_txeof(sc);
2511 
2512 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2513 		re_start_locked(ifp);
2514 
2515 	if (cmd == POLL_AND_CHECK_STATUS) { /* also check status register */
2516 		u_int16_t       status;
2517 
2518 		status = CSR_READ_2(sc, RL_ISR);
2519 		if (status == 0xffff)
2520 			return (rx_npkts);
2521 		if (status)
2522 			CSR_WRITE_2(sc, RL_ISR, status);
2523 		if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
2524 		    (sc->rl_flags & RL_FLAG_PCIE))
2525 			CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2526 
2527 		/*
2528 		 * XXX check behaviour on receiver stalls.
2529 		 */
2530 
2531 		if (status & RL_ISR_SYSTEM_ERR) {
2532 			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2533 			re_init_locked(sc);
2534 		}
2535 	}
2536 	return (rx_npkts);
2537 }
2538 #endif /* DEVICE_POLLING */
2539 
2540 static int
2541 re_intr(void *arg)
2542 {
2543 	struct rl_softc		*sc;
2544 	uint16_t		status;
2545 
2546 	sc = arg;
2547 
2548 	status = CSR_READ_2(sc, RL_ISR);
2549 	if (status == 0xFFFF || (status & RL_INTRS_CPLUS) == 0)
2550                 return (FILTER_STRAY);
2551 	CSR_WRITE_2(sc, RL_IMR, 0);
2552 
2553 	taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_inttask);
2554 
2555 	return (FILTER_HANDLED);
2556 }
2557 
2558 static void
2559 re_int_task(void *arg, int npending)
2560 {
2561 	struct rl_softc		*sc;
2562 	struct ifnet		*ifp;
2563 	u_int16_t		status;
2564 	int			rval = 0;
2565 
2566 	sc = arg;
2567 	ifp = sc->rl_ifp;
2568 
2569 	RL_LOCK(sc);
2570 
2571 	status = CSR_READ_2(sc, RL_ISR);
2572         CSR_WRITE_2(sc, RL_ISR, status);
2573 
2574 	if (sc->suspended ||
2575 	    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2576 		RL_UNLOCK(sc);
2577 		return;
2578 	}
2579 
2580 #ifdef DEVICE_POLLING
2581 	if  (ifp->if_capenable & IFCAP_POLLING) {
2582 		RL_UNLOCK(sc);
2583 		return;
2584 	}
2585 #endif
2586 
2587 	if (status & (RL_ISR_RX_OK|RL_ISR_RX_ERR|RL_ISR_FIFO_OFLOW))
2588 		rval = re_rxeof(sc, NULL);
2589 
2590 	/*
2591 	 * Some chips will ignore a second TX request issued
2592 	 * while an existing transmission is in progress. If
2593 	 * the transmitter goes idle but there are still
2594 	 * packets waiting to be sent, we need to restart the
2595 	 * channel here to flush them out. This only seems to
2596 	 * be required with the PCIe devices.
2597 	 */
2598 	if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
2599 	    (sc->rl_flags & RL_FLAG_PCIE))
2600 		CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2601 	if (status & (
2602 #ifdef RE_TX_MODERATION
2603 	    RL_ISR_TIMEOUT_EXPIRED|
2604 #else
2605 	    RL_ISR_TX_OK|
2606 #endif
2607 	    RL_ISR_TX_ERR|RL_ISR_TX_DESC_UNAVAIL))
2608 		re_txeof(sc);
2609 
2610 	if (status & RL_ISR_SYSTEM_ERR) {
2611 		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2612 		re_init_locked(sc);
2613 	}
2614 
2615 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2616 		re_start_locked(ifp);
2617 
2618 	RL_UNLOCK(sc);
2619 
2620         if ((CSR_READ_2(sc, RL_ISR) & RL_INTRS_CPLUS) || rval) {
2621 		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_inttask);
2622 		return;
2623 	}
2624 
2625 	CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2626 }
2627 
2628 static void
2629 re_intr_msi(void *xsc)
2630 {
2631 	struct rl_softc		*sc;
2632 	struct ifnet		*ifp;
2633 	uint16_t		intrs, status;
2634 
2635 	sc = xsc;
2636 	RL_LOCK(sc);
2637 
2638 	ifp = sc->rl_ifp;
2639 #ifdef DEVICE_POLLING
2640 	if (ifp->if_capenable & IFCAP_POLLING) {
2641 		RL_UNLOCK(sc);
2642 		return;
2643 	}
2644 #endif
2645 	/* Disable interrupts. */
2646 	CSR_WRITE_2(sc, RL_IMR, 0);
2647 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2648 		RL_UNLOCK(sc);
2649 		return;
2650 	}
2651 
2652 	intrs = RL_INTRS_CPLUS;
2653 	status = CSR_READ_2(sc, RL_ISR);
2654         CSR_WRITE_2(sc, RL_ISR, status);
2655 	if (sc->rl_int_rx_act > 0) {
2656 		intrs &= ~(RL_ISR_RX_OK | RL_ISR_RX_ERR | RL_ISR_FIFO_OFLOW |
2657 		    RL_ISR_RX_OVERRUN);
2658 		status &= ~(RL_ISR_RX_OK | RL_ISR_RX_ERR | RL_ISR_FIFO_OFLOW |
2659 		    RL_ISR_RX_OVERRUN);
2660 	}
2661 
2662 	if (status & (RL_ISR_TIMEOUT_EXPIRED | RL_ISR_RX_OK | RL_ISR_RX_ERR |
2663 	    RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN)) {
2664 		re_rxeof(sc, NULL);
2665 		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
2666 			if (sc->rl_int_rx_mod != 0 &&
2667 			    (status & (RL_ISR_RX_OK | RL_ISR_RX_ERR |
2668 			    RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN)) != 0) {
2669 				/* Rearm one-shot timer. */
2670 				CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2671 				intrs &= ~(RL_ISR_RX_OK | RL_ISR_RX_ERR |
2672 				    RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN);
2673 				sc->rl_int_rx_act = 1;
2674 			} else {
2675 				intrs |= RL_ISR_RX_OK | RL_ISR_RX_ERR |
2676 				    RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN;
2677 				sc->rl_int_rx_act = 0;
2678 			}
2679 		}
2680 	}
2681 
2682 	/*
2683 	 * Some chips will ignore a second TX request issued
2684 	 * while an existing transmission is in progress. If
2685 	 * the transmitter goes idle but there are still
2686 	 * packets waiting to be sent, we need to restart the
2687 	 * channel here to flush them out. This only seems to
2688 	 * be required with the PCIe devices.
2689 	 */
2690 	if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
2691 	    (sc->rl_flags & RL_FLAG_PCIE))
2692 		CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2693 	if (status & (RL_ISR_TX_OK | RL_ISR_TX_ERR | RL_ISR_TX_DESC_UNAVAIL))
2694 		re_txeof(sc);
2695 
2696 	if (status & RL_ISR_SYSTEM_ERR) {
2697 		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2698 		re_init_locked(sc);
2699 	}
2700 
2701 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
2702 		if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2703 			re_start_locked(ifp);
2704 		CSR_WRITE_2(sc, RL_IMR, intrs);
2705 	}
2706 	RL_UNLOCK(sc);
2707 }
2708 
2709 static int
2710 re_encap(struct rl_softc *sc, struct mbuf **m_head)
2711 {
2712 	struct rl_txdesc	*txd, *txd_last;
2713 	bus_dma_segment_t	segs[RL_NTXSEGS];
2714 	bus_dmamap_t		map;
2715 	struct mbuf		*m_new;
2716 	struct rl_desc		*desc;
2717 	int			nsegs, prod;
2718 	int			i, error, ei, si;
2719 	int			padlen;
2720 	uint32_t		cmdstat, csum_flags, vlanctl;
2721 
2722 	RL_LOCK_ASSERT(sc);
2723 	M_ASSERTPKTHDR((*m_head));
2724 
2725 	/*
2726 	 * With some of the RealTek chips, using the checksum offload
2727 	 * support in conjunction with the autopadding feature results
2728 	 * in the transmission of corrupt frames. For example, if we
2729 	 * need to send a really small IP fragment that's less than 60
2730 	 * bytes in size, and IP header checksumming is enabled, the
2731 	 * resulting ethernet frame that appears on the wire will
2732 	 * have garbled payload. To work around this, if TX IP checksum
2733 	 * offload is enabled, we always manually pad short frames out
2734 	 * to the minimum ethernet frame size.
2735 	 */
2736 	if ((sc->rl_flags & RL_FLAG_AUTOPAD) == 0 &&
2737 	    (*m_head)->m_pkthdr.len < RL_IP4CSUMTX_PADLEN &&
2738 	    ((*m_head)->m_pkthdr.csum_flags & CSUM_IP) != 0) {
2739 		padlen = RL_MIN_FRAMELEN - (*m_head)->m_pkthdr.len;
2740 		if (M_WRITABLE(*m_head) == 0) {
2741 			/* Get a writable copy. */
2742 			m_new = m_dup(*m_head, M_NOWAIT);
2743 			m_freem(*m_head);
2744 			if (m_new == NULL) {
2745 				*m_head = NULL;
2746 				return (ENOBUFS);
2747 			}
2748 			*m_head = m_new;
2749 		}
2750 		if ((*m_head)->m_next != NULL ||
2751 		    M_TRAILINGSPACE(*m_head) < padlen) {
2752 			m_new = m_defrag(*m_head, M_NOWAIT);
2753 			if (m_new == NULL) {
2754 				m_freem(*m_head);
2755 				*m_head = NULL;
2756 				return (ENOBUFS);
2757 			}
2758 		} else
2759 			m_new = *m_head;
2760 
2761 		/*
2762 		 * Manually pad short frames, and zero the pad space
2763 		 * to avoid leaking data.
2764 		 */
2765 		bzero(mtod(m_new, char *) + m_new->m_pkthdr.len, padlen);
2766 		m_new->m_pkthdr.len += padlen;
2767 		m_new->m_len = m_new->m_pkthdr.len;
2768 		*m_head = m_new;
2769 	}
2770 
2771 	prod = sc->rl_ldata.rl_tx_prodidx;
2772 	txd = &sc->rl_ldata.rl_tx_desc[prod];
2773 	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap,
2774 	    *m_head, segs, &nsegs, BUS_DMA_NOWAIT);
2775 	if (error == EFBIG) {
2776 		m_new = m_collapse(*m_head, M_NOWAIT, RL_NTXSEGS);
2777 		if (m_new == NULL) {
2778 			m_freem(*m_head);
2779 			*m_head = NULL;
2780 			return (ENOBUFS);
2781 		}
2782 		*m_head = m_new;
2783 		error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_tx_mtag,
2784 		    txd->tx_dmamap, *m_head, segs, &nsegs, BUS_DMA_NOWAIT);
2785 		if (error != 0) {
2786 			m_freem(*m_head);
2787 			*m_head = NULL;
2788 			return (error);
2789 		}
2790 	} else if (error != 0)
2791 		return (error);
2792 	if (nsegs == 0) {
2793 		m_freem(*m_head);
2794 		*m_head = NULL;
2795 		return (EIO);
2796 	}
2797 
2798 	/* Check for number of available descriptors. */
2799 	if (sc->rl_ldata.rl_tx_free - nsegs <= 1) {
2800 		bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap);
2801 		return (ENOBUFS);
2802 	}
2803 
2804 	bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap,
2805 	    BUS_DMASYNC_PREWRITE);
2806 
2807 	/*
2808 	 * Set up checksum offload. Note: checksum offload bits must
2809 	 * appear in all descriptors of a multi-descriptor transmit
2810 	 * attempt. This is according to testing done with an 8169
2811 	 * chip. This is a requirement.
2812 	 */
2813 	vlanctl = 0;
2814 	csum_flags = 0;
2815 	if (((*m_head)->m_pkthdr.csum_flags & CSUM_TSO) != 0) {
2816 		if ((sc->rl_flags & RL_FLAG_DESCV2) != 0) {
2817 			csum_flags |= RL_TDESC_CMD_LGSEND;
2818 			vlanctl |= ((uint32_t)(*m_head)->m_pkthdr.tso_segsz <<
2819 			    RL_TDESC_CMD_MSSVALV2_SHIFT);
2820 		} else {
2821 			csum_flags |= RL_TDESC_CMD_LGSEND |
2822 			    ((uint32_t)(*m_head)->m_pkthdr.tso_segsz <<
2823 			    RL_TDESC_CMD_MSSVAL_SHIFT);
2824 		}
2825 	} else {
2826 		/*
2827 		 * Unconditionally enable IP checksum if TCP or UDP
2828 		 * checksum is required. Otherwise, TCP/UDP checksum
2829 		 * doesn't make effects.
2830 		 */
2831 		if (((*m_head)->m_pkthdr.csum_flags & RE_CSUM_FEATURES) != 0) {
2832 			if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
2833 				csum_flags |= RL_TDESC_CMD_IPCSUM;
2834 				if (((*m_head)->m_pkthdr.csum_flags &
2835 				    CSUM_TCP) != 0)
2836 					csum_flags |= RL_TDESC_CMD_TCPCSUM;
2837 				if (((*m_head)->m_pkthdr.csum_flags &
2838 				    CSUM_UDP) != 0)
2839 					csum_flags |= RL_TDESC_CMD_UDPCSUM;
2840 			} else {
2841 				vlanctl |= RL_TDESC_CMD_IPCSUMV2;
2842 				if (((*m_head)->m_pkthdr.csum_flags &
2843 				    CSUM_TCP) != 0)
2844 					vlanctl |= RL_TDESC_CMD_TCPCSUMV2;
2845 				if (((*m_head)->m_pkthdr.csum_flags &
2846 				    CSUM_UDP) != 0)
2847 					vlanctl |= RL_TDESC_CMD_UDPCSUMV2;
2848 			}
2849 		}
2850 	}
2851 
2852 	/*
2853 	 * Set up hardware VLAN tagging. Note: vlan tag info must
2854 	 * appear in all descriptors of a multi-descriptor
2855 	 * transmission attempt.
2856 	 */
2857 	if ((*m_head)->m_flags & M_VLANTAG)
2858 		vlanctl |= bswap16((*m_head)->m_pkthdr.ether_vtag) |
2859 		    RL_TDESC_VLANCTL_TAG;
2860 
2861 	si = prod;
2862 	for (i = 0; i < nsegs; i++, prod = RL_TX_DESC_NXT(sc, prod)) {
2863 		desc = &sc->rl_ldata.rl_tx_list[prod];
2864 		desc->rl_vlanctl = htole32(vlanctl);
2865 		desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[i].ds_addr));
2866 		desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[i].ds_addr));
2867 		cmdstat = segs[i].ds_len;
2868 		if (i != 0)
2869 			cmdstat |= RL_TDESC_CMD_OWN;
2870 		if (prod == sc->rl_ldata.rl_tx_desc_cnt - 1)
2871 			cmdstat |= RL_TDESC_CMD_EOR;
2872 		desc->rl_cmdstat = htole32(cmdstat | csum_flags);
2873 		sc->rl_ldata.rl_tx_free--;
2874 	}
2875 	/* Update producer index. */
2876 	sc->rl_ldata.rl_tx_prodidx = prod;
2877 
2878 	/* Set EOF on the last descriptor. */
2879 	ei = RL_TX_DESC_PRV(sc, prod);
2880 	desc = &sc->rl_ldata.rl_tx_list[ei];
2881 	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_EOF);
2882 
2883 	desc = &sc->rl_ldata.rl_tx_list[si];
2884 	/* Set SOF and transfer ownership of packet to the chip. */
2885 	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_OWN | RL_TDESC_CMD_SOF);
2886 
2887 	/*
2888 	 * Insure that the map for this transmission
2889 	 * is placed at the array index of the last descriptor
2890 	 * in this chain.  (Swap last and first dmamaps.)
2891 	 */
2892 	txd_last = &sc->rl_ldata.rl_tx_desc[ei];
2893 	map = txd->tx_dmamap;
2894 	txd->tx_dmamap = txd_last->tx_dmamap;
2895 	txd_last->tx_dmamap = map;
2896 	txd_last->tx_m = *m_head;
2897 
2898 	return (0);
2899 }
2900 
2901 static void
2902 re_start(struct ifnet *ifp)
2903 {
2904 	struct rl_softc		*sc;
2905 
2906 	sc = ifp->if_softc;
2907 	RL_LOCK(sc);
2908 	re_start_locked(ifp);
2909 	RL_UNLOCK(sc);
2910 }
2911 
2912 /*
2913  * Main transmit routine for C+ and gigE NICs.
2914  */
2915 static void
2916 re_start_locked(struct ifnet *ifp)
2917 {
2918 	struct rl_softc		*sc;
2919 	struct mbuf		*m_head;
2920 	int			queued;
2921 
2922 	sc = ifp->if_softc;
2923 
2924 #ifdef DEV_NETMAP
2925 	/* XXX is this necessary ? */
2926 	if (ifp->if_capenable & IFCAP_NETMAP) {
2927 		struct netmap_kring *kring = &NA(ifp)->tx_rings[0];
2928 		if (sc->rl_ldata.rl_tx_prodidx != kring->nr_hwcur) {
2929 			/* kick the tx unit */
2930 			CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2931 #ifdef RE_TX_MODERATION
2932 			CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2933 #endif
2934 			sc->rl_watchdog_timer = 5;
2935 		}
2936 		return;
2937 	}
2938 #endif /* DEV_NETMAP */
2939 	if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
2940 	    IFF_DRV_RUNNING || (sc->rl_flags & RL_FLAG_LINK) == 0)
2941 		return;
2942 
2943 	for (queued = 0; !IFQ_DRV_IS_EMPTY(&ifp->if_snd) &&
2944 	    sc->rl_ldata.rl_tx_free > 1;) {
2945 		IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
2946 		if (m_head == NULL)
2947 			break;
2948 
2949 		if (re_encap(sc, &m_head) != 0) {
2950 			if (m_head == NULL)
2951 				break;
2952 			IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
2953 			ifp->if_drv_flags |= IFF_DRV_OACTIVE;
2954 			break;
2955 		}
2956 
2957 		/*
2958 		 * If there's a BPF listener, bounce a copy of this frame
2959 		 * to him.
2960 		 */
2961 		ETHER_BPF_MTAP(ifp, m_head);
2962 
2963 		queued++;
2964 	}
2965 
2966 	if (queued == 0) {
2967 #ifdef RE_TX_MODERATION
2968 		if (sc->rl_ldata.rl_tx_free != sc->rl_ldata.rl_tx_desc_cnt)
2969 			CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2970 #endif
2971 		return;
2972 	}
2973 
2974 	/* Flush the TX descriptors */
2975 
2976 	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2977 	    sc->rl_ldata.rl_tx_list_map,
2978 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2979 
2980 	CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2981 
2982 #ifdef RE_TX_MODERATION
2983 	/*
2984 	 * Use the countdown timer for interrupt moderation.
2985 	 * 'TX done' interrupts are disabled. Instead, we reset the
2986 	 * countdown timer, which will begin counting until it hits
2987 	 * the value in the TIMERINT register, and then trigger an
2988 	 * interrupt. Each time we write to the TIMERCNT register,
2989 	 * the timer count is reset to 0.
2990 	 */
2991 	CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2992 #endif
2993 
2994 	/*
2995 	 * Set a timeout in case the chip goes out to lunch.
2996 	 */
2997 	sc->rl_watchdog_timer = 5;
2998 }
2999 
3000 static void
3001 re_set_jumbo(struct rl_softc *sc, int jumbo)
3002 {
3003 
3004 	if (sc->rl_hwrev->rl_rev == RL_HWREV_8168E_VL) {
3005 		pci_set_max_read_req(sc->rl_dev, 4096);
3006 		return;
3007 	}
3008 
3009 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG);
3010 	if (jumbo != 0) {
3011 		CSR_WRITE_1(sc, sc->rl_cfg3, CSR_READ_1(sc, sc->rl_cfg3) |
3012 		    RL_CFG3_JUMBO_EN0);
3013 		switch (sc->rl_hwrev->rl_rev) {
3014 		case RL_HWREV_8168DP:
3015 			break;
3016 		case RL_HWREV_8168E:
3017 			CSR_WRITE_1(sc, sc->rl_cfg4,
3018 			    CSR_READ_1(sc, sc->rl_cfg4) | 0x01);
3019 			break;
3020 		default:
3021 			CSR_WRITE_1(sc, sc->rl_cfg4,
3022 			    CSR_READ_1(sc, sc->rl_cfg4) | RL_CFG4_JUMBO_EN1);
3023 		}
3024 	} else {
3025 		CSR_WRITE_1(sc, sc->rl_cfg3, CSR_READ_1(sc, sc->rl_cfg3) &
3026 		    ~RL_CFG3_JUMBO_EN0);
3027 		switch (sc->rl_hwrev->rl_rev) {
3028 		case RL_HWREV_8168DP:
3029 			break;
3030 		case RL_HWREV_8168E:
3031 			CSR_WRITE_1(sc, sc->rl_cfg4,
3032 			    CSR_READ_1(sc, sc->rl_cfg4) & ~0x01);
3033 			break;
3034 		default:
3035 			CSR_WRITE_1(sc, sc->rl_cfg4,
3036 			    CSR_READ_1(sc, sc->rl_cfg4) & ~RL_CFG4_JUMBO_EN1);
3037 		}
3038 	}
3039 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3040 
3041 	switch (sc->rl_hwrev->rl_rev) {
3042 	case RL_HWREV_8168DP:
3043 		pci_set_max_read_req(sc->rl_dev, 4096);
3044 		break;
3045 	default:
3046 		if (jumbo != 0)
3047 			pci_set_max_read_req(sc->rl_dev, 512);
3048 		else
3049 			pci_set_max_read_req(sc->rl_dev, 4096);
3050 	}
3051 }
3052 
3053 static void
3054 re_init(void *xsc)
3055 {
3056 	struct rl_softc		*sc = xsc;
3057 
3058 	RL_LOCK(sc);
3059 	re_init_locked(sc);
3060 	RL_UNLOCK(sc);
3061 }
3062 
3063 static void
3064 re_init_locked(struct rl_softc *sc)
3065 {
3066 	struct ifnet		*ifp = sc->rl_ifp;
3067 	struct mii_data		*mii;
3068 	uint32_t		reg;
3069 	uint16_t		cfg;
3070 	union {
3071 		uint32_t align_dummy;
3072 		u_char eaddr[ETHER_ADDR_LEN];
3073         } eaddr;
3074 
3075 	RL_LOCK_ASSERT(sc);
3076 
3077 	mii = device_get_softc(sc->rl_miibus);
3078 
3079 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
3080 		return;
3081 
3082 	/*
3083 	 * Cancel pending I/O and free all RX/TX buffers.
3084 	 */
3085 	re_stop(sc);
3086 
3087 	/* Put controller into known state. */
3088 	re_reset(sc);
3089 
3090 	/*
3091 	 * For C+ mode, initialize the RX descriptors and mbufs.
3092 	 */
3093 	if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
3094 		if (ifp->if_mtu > RL_MTU) {
3095 			if (re_jrx_list_init(sc) != 0) {
3096 				device_printf(sc->rl_dev,
3097 				    "no memory for jumbo RX buffers\n");
3098 				re_stop(sc);
3099 				return;
3100 			}
3101 			/* Disable checksum offloading for jumbo frames. */
3102 			ifp->if_capenable &= ~(IFCAP_HWCSUM | IFCAP_TSO4);
3103 			ifp->if_hwassist &= ~(RE_CSUM_FEATURES | CSUM_TSO);
3104 		} else {
3105 			if (re_rx_list_init(sc) != 0) {
3106 				device_printf(sc->rl_dev,
3107 				    "no memory for RX buffers\n");
3108 				re_stop(sc);
3109 				return;
3110 			}
3111 		}
3112 		re_set_jumbo(sc, ifp->if_mtu > RL_MTU);
3113 	} else {
3114 		if (re_rx_list_init(sc) != 0) {
3115 			device_printf(sc->rl_dev, "no memory for RX buffers\n");
3116 			re_stop(sc);
3117 			return;
3118 		}
3119 		if ((sc->rl_flags & RL_FLAG_PCIE) != 0 &&
3120 		    pci_get_device(sc->rl_dev) != RT_DEVICEID_8101E) {
3121 			if (ifp->if_mtu > RL_MTU)
3122 				pci_set_max_read_req(sc->rl_dev, 512);
3123 			else
3124 				pci_set_max_read_req(sc->rl_dev, 4096);
3125 		}
3126 	}
3127 	re_tx_list_init(sc);
3128 
3129 	/*
3130 	 * Enable C+ RX and TX mode, as well as VLAN stripping and
3131 	 * RX checksum offload. We must configure the C+ register
3132 	 * before all others.
3133 	 */
3134 	cfg = RL_CPLUSCMD_PCI_MRW;
3135 	if ((ifp->if_capenable & IFCAP_RXCSUM) != 0)
3136 		cfg |= RL_CPLUSCMD_RXCSUM_ENB;
3137 	if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0)
3138 		cfg |= RL_CPLUSCMD_VLANSTRIP;
3139 	if ((sc->rl_flags & RL_FLAG_MACSTAT) != 0) {
3140 		cfg |= RL_CPLUSCMD_MACSTAT_DIS;
3141 		/* XXX magic. */
3142 		cfg |= 0x0001;
3143 	} else
3144 		cfg |= RL_CPLUSCMD_RXENB | RL_CPLUSCMD_TXENB;
3145 	CSR_WRITE_2(sc, RL_CPLUS_CMD, cfg);
3146 	if (sc->rl_hwrev->rl_rev == RL_HWREV_8169_8110SC ||
3147 	    sc->rl_hwrev->rl_rev == RL_HWREV_8169_8110SCE) {
3148 		reg = 0x000fff00;
3149 		if ((CSR_READ_1(sc, sc->rl_cfg2) & RL_CFG2_PCI66MHZ) != 0)
3150 			reg |= 0x000000ff;
3151 		if (sc->rl_hwrev->rl_rev == RL_HWREV_8169_8110SCE)
3152 			reg |= 0x00f00000;
3153 		CSR_WRITE_4(sc, 0x7c, reg);
3154 		/* Disable interrupt mitigation. */
3155 		CSR_WRITE_2(sc, 0xe2, 0);
3156 	}
3157 	/*
3158 	 * Disable TSO if interface MTU size is greater than MSS
3159 	 * allowed in controller.
3160 	 */
3161 	if (ifp->if_mtu > RL_TSO_MTU && (ifp->if_capenable & IFCAP_TSO4) != 0) {
3162 		ifp->if_capenable &= ~IFCAP_TSO4;
3163 		ifp->if_hwassist &= ~CSUM_TSO;
3164 	}
3165 
3166 	/*
3167 	 * Init our MAC address.  Even though the chipset
3168 	 * documentation doesn't mention it, we need to enter "Config
3169 	 * register write enable" mode to modify the ID registers.
3170 	 */
3171 	/* Copy MAC address on stack to align. */
3172 	bcopy(IF_LLADDR(ifp), eaddr.eaddr, ETHER_ADDR_LEN);
3173 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG);
3174 	CSR_WRITE_4(sc, RL_IDR0,
3175 	    htole32(*(u_int32_t *)(&eaddr.eaddr[0])));
3176 	CSR_WRITE_4(sc, RL_IDR4,
3177 	    htole32(*(u_int32_t *)(&eaddr.eaddr[4])));
3178 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3179 
3180 	/*
3181 	 * Load the addresses of the RX and TX lists into the chip.
3182 	 */
3183 
3184 	CSR_WRITE_4(sc, RL_RXLIST_ADDR_HI,
3185 	    RL_ADDR_HI(sc->rl_ldata.rl_rx_list_addr));
3186 	CSR_WRITE_4(sc, RL_RXLIST_ADDR_LO,
3187 	    RL_ADDR_LO(sc->rl_ldata.rl_rx_list_addr));
3188 
3189 	CSR_WRITE_4(sc, RL_TXLIST_ADDR_HI,
3190 	    RL_ADDR_HI(sc->rl_ldata.rl_tx_list_addr));
3191 	CSR_WRITE_4(sc, RL_TXLIST_ADDR_LO,
3192 	    RL_ADDR_LO(sc->rl_ldata.rl_tx_list_addr));
3193 
3194 	if ((sc->rl_flags & RL_FLAG_RXDV_GATED) != 0)
3195 		CSR_WRITE_4(sc, RL_MISC, CSR_READ_4(sc, RL_MISC) &
3196 		    ~0x00080000);
3197 
3198 	/*
3199 	 * Enable transmit and receive.
3200 	 */
3201 	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
3202 
3203 	/*
3204 	 * Set the initial TX configuration.
3205 	 */
3206 	if (sc->rl_testmode) {
3207 		if (sc->rl_type == RL_8169)
3208 			CSR_WRITE_4(sc, RL_TXCFG,
3209 			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON);
3210 		else
3211 			CSR_WRITE_4(sc, RL_TXCFG,
3212 			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON_CPLUS);
3213 	} else
3214 		CSR_WRITE_4(sc, RL_TXCFG, RL_TXCFG_CONFIG);
3215 
3216 	CSR_WRITE_1(sc, RL_EARLY_TX_THRESH, 16);
3217 
3218 	/*
3219 	 * Set the initial RX configuration.
3220 	 */
3221 	re_set_rxmode(sc);
3222 
3223 	/* Configure interrupt moderation. */
3224 	if (sc->rl_type == RL_8169) {
3225 		/* Magic from vendor. */
3226 		CSR_WRITE_2(sc, RL_INTRMOD, 0x5100);
3227 	}
3228 
3229 #ifdef DEVICE_POLLING
3230 	/*
3231 	 * Disable interrupts if we are polling.
3232 	 */
3233 	if (ifp->if_capenable & IFCAP_POLLING)
3234 		CSR_WRITE_2(sc, RL_IMR, 0);
3235 	else	/* otherwise ... */
3236 #endif
3237 
3238 	/*
3239 	 * Enable interrupts.
3240 	 */
3241 	if (sc->rl_testmode)
3242 		CSR_WRITE_2(sc, RL_IMR, 0);
3243 	else
3244 		CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
3245 	CSR_WRITE_2(sc, RL_ISR, RL_INTRS_CPLUS);
3246 
3247 	/* Set initial TX threshold */
3248 	sc->rl_txthresh = RL_TX_THRESH_INIT;
3249 
3250 	/* Start RX/TX process. */
3251 	CSR_WRITE_4(sc, RL_MISSEDPKT, 0);
3252 #ifdef notdef
3253 	/* Enable receiver and transmitter. */
3254 	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
3255 #endif
3256 
3257 	/*
3258 	 * Initialize the timer interrupt register so that
3259 	 * a timer interrupt will be generated once the timer
3260 	 * reaches a certain number of ticks. The timer is
3261 	 * reloaded on each transmit.
3262 	 */
3263 #ifdef RE_TX_MODERATION
3264 	/*
3265 	 * Use timer interrupt register to moderate TX interrupt
3266 	 * moderation, which dramatically improves TX frame rate.
3267 	 */
3268 	if (sc->rl_type == RL_8169)
3269 		CSR_WRITE_4(sc, RL_TIMERINT_8169, 0x800);
3270 	else
3271 		CSR_WRITE_4(sc, RL_TIMERINT, 0x400);
3272 #else
3273 	/*
3274 	 * Use timer interrupt register to moderate RX interrupt
3275 	 * moderation.
3276 	 */
3277 	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) != 0 &&
3278 	    intr_filter == 0) {
3279 		if (sc->rl_type == RL_8169)
3280 			CSR_WRITE_4(sc, RL_TIMERINT_8169,
3281 			    RL_USECS(sc->rl_int_rx_mod));
3282 	} else {
3283 		if (sc->rl_type == RL_8169)
3284 			CSR_WRITE_4(sc, RL_TIMERINT_8169, RL_USECS(0));
3285 	}
3286 #endif
3287 
3288 	/*
3289 	 * For 8169 gigE NICs, set the max allowed RX packet
3290 	 * size so we can receive jumbo frames.
3291 	 */
3292 	if (sc->rl_type == RL_8169) {
3293 		if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
3294 			/*
3295 			 * For controllers that use new jumbo frame scheme,
3296 			 * set maximum size of jumbo frame depending on
3297 			 * controller revisions.
3298 			 */
3299 			if (ifp->if_mtu > RL_MTU)
3300 				CSR_WRITE_2(sc, RL_MAXRXPKTLEN,
3301 				    sc->rl_hwrev->rl_max_mtu +
3302 				    ETHER_VLAN_ENCAP_LEN + ETHER_HDR_LEN +
3303 				    ETHER_CRC_LEN);
3304 			else
3305 				CSR_WRITE_2(sc, RL_MAXRXPKTLEN,
3306 				    RE_RX_DESC_BUFLEN);
3307 		} else if ((sc->rl_flags & RL_FLAG_PCIE) != 0 &&
3308 		    sc->rl_hwrev->rl_max_mtu == RL_MTU) {
3309 			/* RTL810x has no jumbo frame support. */
3310 			CSR_WRITE_2(sc, RL_MAXRXPKTLEN, RE_RX_DESC_BUFLEN);
3311 		} else
3312 			CSR_WRITE_2(sc, RL_MAXRXPKTLEN, 16383);
3313 	}
3314 
3315 	if (sc->rl_testmode)
3316 		return;
3317 
3318 	CSR_WRITE_1(sc, sc->rl_cfg1, CSR_READ_1(sc, sc->rl_cfg1) |
3319 	    RL_CFG1_DRVLOAD);
3320 
3321 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
3322 	ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
3323 
3324 	sc->rl_flags &= ~RL_FLAG_LINK;
3325 	mii_mediachg(mii);
3326 
3327 	sc->rl_watchdog_timer = 0;
3328 	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
3329 }
3330 
3331 /*
3332  * Set media options.
3333  */
3334 static int
3335 re_ifmedia_upd(struct ifnet *ifp)
3336 {
3337 	struct rl_softc		*sc;
3338 	struct mii_data		*mii;
3339 	int			error;
3340 
3341 	sc = ifp->if_softc;
3342 	mii = device_get_softc(sc->rl_miibus);
3343 	RL_LOCK(sc);
3344 	error = mii_mediachg(mii);
3345 	RL_UNLOCK(sc);
3346 
3347 	return (error);
3348 }
3349 
3350 /*
3351  * Report current media status.
3352  */
3353 static void
3354 re_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
3355 {
3356 	struct rl_softc		*sc;
3357 	struct mii_data		*mii;
3358 
3359 	sc = ifp->if_softc;
3360 	mii = device_get_softc(sc->rl_miibus);
3361 
3362 	RL_LOCK(sc);
3363 	mii_pollstat(mii);
3364 	ifmr->ifm_active = mii->mii_media_active;
3365 	ifmr->ifm_status = mii->mii_media_status;
3366 	RL_UNLOCK(sc);
3367 }
3368 
3369 static int
3370 re_ioctl(struct ifnet *ifp, u_long command, caddr_t data)
3371 {
3372 	struct rl_softc		*sc = ifp->if_softc;
3373 	struct ifreq		*ifr = (struct ifreq *) data;
3374 	struct mii_data		*mii;
3375 	int			error = 0;
3376 
3377 	switch (command) {
3378 	case SIOCSIFMTU:
3379 		if (ifr->ifr_mtu < ETHERMIN ||
3380 		    ifr->ifr_mtu > sc->rl_hwrev->rl_max_mtu ||
3381 		    ((sc->rl_flags & RL_FLAG_FASTETHER) != 0 &&
3382 		    ifr->ifr_mtu > RL_MTU)) {
3383 			error = EINVAL;
3384 			break;
3385 		}
3386 		RL_LOCK(sc);
3387 		if (ifp->if_mtu != ifr->ifr_mtu) {
3388 			ifp->if_mtu = ifr->ifr_mtu;
3389 			if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0 &&
3390 			    (ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
3391 				ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3392 				re_init_locked(sc);
3393 			}
3394 			if (ifp->if_mtu > RL_TSO_MTU &&
3395 			    (ifp->if_capenable & IFCAP_TSO4) != 0) {
3396 				ifp->if_capenable &= ~(IFCAP_TSO4 |
3397 				    IFCAP_VLAN_HWTSO);
3398 				ifp->if_hwassist &= ~CSUM_TSO;
3399 			}
3400 			VLAN_CAPABILITIES(ifp);
3401 		}
3402 		RL_UNLOCK(sc);
3403 		break;
3404 	case SIOCSIFFLAGS:
3405 		RL_LOCK(sc);
3406 		if ((ifp->if_flags & IFF_UP) != 0) {
3407 			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
3408 				if (((ifp->if_flags ^ sc->rl_if_flags)
3409 				    & (IFF_PROMISC | IFF_ALLMULTI)) != 0)
3410 					re_set_rxmode(sc);
3411 			} else
3412 				re_init_locked(sc);
3413 		} else {
3414 			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
3415 				re_stop(sc);
3416 		}
3417 		sc->rl_if_flags = ifp->if_flags;
3418 		RL_UNLOCK(sc);
3419 		break;
3420 	case SIOCADDMULTI:
3421 	case SIOCDELMULTI:
3422 		RL_LOCK(sc);
3423 		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
3424 			re_set_rxmode(sc);
3425 		RL_UNLOCK(sc);
3426 		break;
3427 	case SIOCGIFMEDIA:
3428 	case SIOCSIFMEDIA:
3429 		mii = device_get_softc(sc->rl_miibus);
3430 		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, command);
3431 		break;
3432 	case SIOCSIFCAP:
3433 	    {
3434 		int mask, reinit;
3435 
3436 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
3437 		reinit = 0;
3438 #ifdef DEVICE_POLLING
3439 		if (mask & IFCAP_POLLING) {
3440 			if (ifr->ifr_reqcap & IFCAP_POLLING) {
3441 				error = ether_poll_register(re_poll, ifp);
3442 				if (error)
3443 					return (error);
3444 				RL_LOCK(sc);
3445 				/* Disable interrupts */
3446 				CSR_WRITE_2(sc, RL_IMR, 0x0000);
3447 				ifp->if_capenable |= IFCAP_POLLING;
3448 				RL_UNLOCK(sc);
3449 			} else {
3450 				error = ether_poll_deregister(ifp);
3451 				/* Enable interrupts. */
3452 				RL_LOCK(sc);
3453 				CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
3454 				ifp->if_capenable &= ~IFCAP_POLLING;
3455 				RL_UNLOCK(sc);
3456 			}
3457 		}
3458 #endif /* DEVICE_POLLING */
3459 		RL_LOCK(sc);
3460 		if ((mask & IFCAP_TXCSUM) != 0 &&
3461 		    (ifp->if_capabilities & IFCAP_TXCSUM) != 0) {
3462 			ifp->if_capenable ^= IFCAP_TXCSUM;
3463 			if ((ifp->if_capenable & IFCAP_TXCSUM) != 0)
3464 				ifp->if_hwassist |= RE_CSUM_FEATURES;
3465 			else
3466 				ifp->if_hwassist &= ~RE_CSUM_FEATURES;
3467 			reinit = 1;
3468 		}
3469 		if ((mask & IFCAP_RXCSUM) != 0 &&
3470 		    (ifp->if_capabilities & IFCAP_RXCSUM) != 0) {
3471 			ifp->if_capenable ^= IFCAP_RXCSUM;
3472 			reinit = 1;
3473 		}
3474 		if ((mask & IFCAP_TSO4) != 0 &&
3475 		    (ifp->if_capabilities & IFCAP_TSO4) != 0) {
3476 			ifp->if_capenable ^= IFCAP_TSO4;
3477 			if ((IFCAP_TSO4 & ifp->if_capenable) != 0)
3478 				ifp->if_hwassist |= CSUM_TSO;
3479 			else
3480 				ifp->if_hwassist &= ~CSUM_TSO;
3481 			if (ifp->if_mtu > RL_TSO_MTU &&
3482 			    (ifp->if_capenable & IFCAP_TSO4) != 0) {
3483 				ifp->if_capenable &= ~IFCAP_TSO4;
3484 				ifp->if_hwassist &= ~CSUM_TSO;
3485 			}
3486 		}
3487 		if ((mask & IFCAP_VLAN_HWTSO) != 0 &&
3488 		    (ifp->if_capabilities & IFCAP_VLAN_HWTSO) != 0)
3489 			ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
3490 		if ((mask & IFCAP_VLAN_HWTAGGING) != 0 &&
3491 		    (ifp->if_capabilities & IFCAP_VLAN_HWTAGGING) != 0) {
3492 			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
3493 			/* TSO over VLAN requires VLAN hardware tagging. */
3494 			if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) == 0)
3495 				ifp->if_capenable &= ~IFCAP_VLAN_HWTSO;
3496 			reinit = 1;
3497 		}
3498 		if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0 &&
3499 		    (mask & (IFCAP_HWCSUM | IFCAP_TSO4 |
3500 		    IFCAP_VLAN_HWTSO)) != 0)
3501 				reinit = 1;
3502 		if ((mask & IFCAP_WOL) != 0 &&
3503 		    (ifp->if_capabilities & IFCAP_WOL) != 0) {
3504 			if ((mask & IFCAP_WOL_UCAST) != 0)
3505 				ifp->if_capenable ^= IFCAP_WOL_UCAST;
3506 			if ((mask & IFCAP_WOL_MCAST) != 0)
3507 				ifp->if_capenable ^= IFCAP_WOL_MCAST;
3508 			if ((mask & IFCAP_WOL_MAGIC) != 0)
3509 				ifp->if_capenable ^= IFCAP_WOL_MAGIC;
3510 		}
3511 		if (reinit && ifp->if_drv_flags & IFF_DRV_RUNNING) {
3512 			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3513 			re_init_locked(sc);
3514 		}
3515 		RL_UNLOCK(sc);
3516 		VLAN_CAPABILITIES(ifp);
3517 	    }
3518 		break;
3519 	default:
3520 		error = ether_ioctl(ifp, command, data);
3521 		break;
3522 	}
3523 
3524 	return (error);
3525 }
3526 
3527 static void
3528 re_watchdog(struct rl_softc *sc)
3529 {
3530 	struct ifnet		*ifp;
3531 
3532 	RL_LOCK_ASSERT(sc);
3533 
3534 	if (sc->rl_watchdog_timer == 0 || --sc->rl_watchdog_timer != 0)
3535 		return;
3536 
3537 	ifp = sc->rl_ifp;
3538 	re_txeof(sc);
3539 	if (sc->rl_ldata.rl_tx_free == sc->rl_ldata.rl_tx_desc_cnt) {
3540 		if_printf(ifp, "watchdog timeout (missed Tx interrupts) "
3541 		    "-- recovering\n");
3542 		if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
3543 			re_start_locked(ifp);
3544 		return;
3545 	}
3546 
3547 	if_printf(ifp, "watchdog timeout\n");
3548 	if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
3549 
3550 	re_rxeof(sc, NULL);
3551 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3552 	re_init_locked(sc);
3553 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
3554 		re_start_locked(ifp);
3555 }
3556 
3557 /*
3558  * Stop the adapter and free any mbufs allocated to the
3559  * RX and TX lists.
3560  */
3561 static void
3562 re_stop(struct rl_softc *sc)
3563 {
3564 	int			i;
3565 	struct ifnet		*ifp;
3566 	struct rl_txdesc	*txd;
3567 	struct rl_rxdesc	*rxd;
3568 
3569 	RL_LOCK_ASSERT(sc);
3570 
3571 	ifp = sc->rl_ifp;
3572 
3573 	sc->rl_watchdog_timer = 0;
3574 	callout_stop(&sc->rl_stat_callout);
3575 	ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
3576 
3577 	/*
3578 	 * Disable accepting frames to put RX MAC into idle state.
3579 	 * Otherwise it's possible to get frames while stop command
3580 	 * execution is in progress and controller can DMA the frame
3581 	 * to already freed RX buffer during that period.
3582 	 */
3583 	CSR_WRITE_4(sc, RL_RXCFG, CSR_READ_4(sc, RL_RXCFG) &
3584 	    ~(RL_RXCFG_RX_ALLPHYS | RL_RXCFG_RX_INDIV | RL_RXCFG_RX_MULTI |
3585 	    RL_RXCFG_RX_BROAD));
3586 
3587 	if ((sc->rl_flags & RL_FLAG_WAIT_TXPOLL) != 0) {
3588 		for (i = RL_TIMEOUT; i > 0; i--) {
3589 			if ((CSR_READ_1(sc, sc->rl_txstart) &
3590 			    RL_TXSTART_START) == 0)
3591 				break;
3592 			DELAY(20);
3593 		}
3594 		if (i == 0)
3595 			device_printf(sc->rl_dev,
3596 			    "stopping TX poll timed out!\n");
3597 		CSR_WRITE_1(sc, RL_COMMAND, 0x00);
3598 	} else if ((sc->rl_flags & RL_FLAG_CMDSTOP) != 0) {
3599 		CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_STOPREQ | RL_CMD_TX_ENB |
3600 		    RL_CMD_RX_ENB);
3601 		if ((sc->rl_flags & RL_FLAG_CMDSTOP_WAIT_TXQ) != 0) {
3602 			for (i = RL_TIMEOUT; i > 0; i--) {
3603 				if ((CSR_READ_4(sc, RL_TXCFG) &
3604 				    RL_TXCFG_QUEUE_EMPTY) != 0)
3605 					break;
3606 				DELAY(100);
3607 			}
3608 			if (i == 0)
3609 				device_printf(sc->rl_dev,
3610 				   "stopping TXQ timed out!\n");
3611 		}
3612 	} else
3613 		CSR_WRITE_1(sc, RL_COMMAND, 0x00);
3614 	DELAY(1000);
3615 	CSR_WRITE_2(sc, RL_IMR, 0x0000);
3616 	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
3617 
3618 	if (sc->rl_head != NULL) {
3619 		m_freem(sc->rl_head);
3620 		sc->rl_head = sc->rl_tail = NULL;
3621 	}
3622 
3623 	/* Free the TX list buffers. */
3624 	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
3625 		txd = &sc->rl_ldata.rl_tx_desc[i];
3626 		if (txd->tx_m != NULL) {
3627 			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
3628 			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
3629 			bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag,
3630 			    txd->tx_dmamap);
3631 			m_freem(txd->tx_m);
3632 			txd->tx_m = NULL;
3633 		}
3634 	}
3635 
3636 	/* Free the RX list buffers. */
3637 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
3638 		rxd = &sc->rl_ldata.rl_rx_desc[i];
3639 		if (rxd->rx_m != NULL) {
3640 			bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag,
3641 			    rxd->rx_dmamap, BUS_DMASYNC_POSTREAD);
3642 			bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag,
3643 			    rxd->rx_dmamap);
3644 			m_freem(rxd->rx_m);
3645 			rxd->rx_m = NULL;
3646 		}
3647 	}
3648 
3649 	if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
3650 		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
3651 			rxd = &sc->rl_ldata.rl_jrx_desc[i];
3652 			if (rxd->rx_m != NULL) {
3653 				bus_dmamap_sync(sc->rl_ldata.rl_jrx_mtag,
3654 				    rxd->rx_dmamap, BUS_DMASYNC_POSTREAD);
3655 				bus_dmamap_unload(sc->rl_ldata.rl_jrx_mtag,
3656 				    rxd->rx_dmamap);
3657 				m_freem(rxd->rx_m);
3658 				rxd->rx_m = NULL;
3659 			}
3660 		}
3661 	}
3662 }
3663 
3664 /*
3665  * Device suspend routine.  Stop the interface and save some PCI
3666  * settings in case the BIOS doesn't restore them properly on
3667  * resume.
3668  */
3669 static int
3670 re_suspend(device_t dev)
3671 {
3672 	struct rl_softc		*sc;
3673 
3674 	sc = device_get_softc(dev);
3675 
3676 	RL_LOCK(sc);
3677 	re_stop(sc);
3678 	re_setwol(sc);
3679 	sc->suspended = 1;
3680 	RL_UNLOCK(sc);
3681 
3682 	return (0);
3683 }
3684 
3685 /*
3686  * Device resume routine.  Restore some PCI settings in case the BIOS
3687  * doesn't, re-enable busmastering, and restart the interface if
3688  * appropriate.
3689  */
3690 static int
3691 re_resume(device_t dev)
3692 {
3693 	struct rl_softc		*sc;
3694 	struct ifnet		*ifp;
3695 
3696 	sc = device_get_softc(dev);
3697 
3698 	RL_LOCK(sc);
3699 
3700 	ifp = sc->rl_ifp;
3701 	/* Take controller out of sleep mode. */
3702 	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
3703 		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
3704 			CSR_WRITE_1(sc, RL_GPIO,
3705 			    CSR_READ_1(sc, RL_GPIO) | 0x01);
3706 	}
3707 
3708 	/*
3709 	 * Clear WOL matching such that normal Rx filtering
3710 	 * wouldn't interfere with WOL patterns.
3711 	 */
3712 	re_clrwol(sc);
3713 
3714 	/* reinitialize interface if necessary */
3715 	if (ifp->if_flags & IFF_UP)
3716 		re_init_locked(sc);
3717 
3718 	sc->suspended = 0;
3719 	RL_UNLOCK(sc);
3720 
3721 	return (0);
3722 }
3723 
3724 /*
3725  * Stop all chip I/O so that the kernel's probe routines don't
3726  * get confused by errant DMAs when rebooting.
3727  */
3728 static int
3729 re_shutdown(device_t dev)
3730 {
3731 	struct rl_softc		*sc;
3732 
3733 	sc = device_get_softc(dev);
3734 
3735 	RL_LOCK(sc);
3736 	re_stop(sc);
3737 	/*
3738 	 * Mark interface as down since otherwise we will panic if
3739 	 * interrupt comes in later on, which can happen in some
3740 	 * cases.
3741 	 */
3742 	sc->rl_ifp->if_flags &= ~IFF_UP;
3743 	re_setwol(sc);
3744 	RL_UNLOCK(sc);
3745 
3746 	return (0);
3747 }
3748 
3749 static void
3750 re_set_linkspeed(struct rl_softc *sc)
3751 {
3752 	struct mii_softc *miisc;
3753 	struct mii_data *mii;
3754 	int aneg, i, phyno;
3755 
3756 	RL_LOCK_ASSERT(sc);
3757 
3758 	mii = device_get_softc(sc->rl_miibus);
3759 	mii_pollstat(mii);
3760 	aneg = 0;
3761 	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
3762 	    (IFM_ACTIVE | IFM_AVALID)) {
3763 		switch IFM_SUBTYPE(mii->mii_media_active) {
3764 		case IFM_10_T:
3765 		case IFM_100_TX:
3766 			return;
3767 		case IFM_1000_T:
3768 			aneg++;
3769 			break;
3770 		default:
3771 			break;
3772 		}
3773 	}
3774 	miisc = LIST_FIRST(&mii->mii_phys);
3775 	phyno = miisc->mii_phy;
3776 	LIST_FOREACH(miisc, &mii->mii_phys, mii_list)
3777 		PHY_RESET(miisc);
3778 	re_miibus_writereg(sc->rl_dev, phyno, MII_100T2CR, 0);
3779 	re_miibus_writereg(sc->rl_dev, phyno,
3780 	    MII_ANAR, ANAR_TX_FD | ANAR_TX | ANAR_10_FD | ANAR_10 | ANAR_CSMA);
3781 	re_miibus_writereg(sc->rl_dev, phyno,
3782 	    MII_BMCR, BMCR_AUTOEN | BMCR_STARTNEG);
3783 	DELAY(1000);
3784 	if (aneg != 0) {
3785 		/*
3786 		 * Poll link state until re(4) get a 10/100Mbps link.
3787 		 */
3788 		for (i = 0; i < MII_ANEGTICKS_GIGE; i++) {
3789 			mii_pollstat(mii);
3790 			if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID))
3791 			    == (IFM_ACTIVE | IFM_AVALID)) {
3792 				switch (IFM_SUBTYPE(mii->mii_media_active)) {
3793 				case IFM_10_T:
3794 				case IFM_100_TX:
3795 					return;
3796 				default:
3797 					break;
3798 				}
3799 			}
3800 			RL_UNLOCK(sc);
3801 			pause("relnk", hz);
3802 			RL_LOCK(sc);
3803 		}
3804 		if (i == MII_ANEGTICKS_GIGE)
3805 			device_printf(sc->rl_dev,
3806 			    "establishing a link failed, WOL may not work!");
3807 	}
3808 	/*
3809 	 * No link, force MAC to have 100Mbps, full-duplex link.
3810 	 * MAC does not require reprogramming on resolved speed/duplex,
3811 	 * so this is just for completeness.
3812 	 */
3813 	mii->mii_media_status = IFM_AVALID | IFM_ACTIVE;
3814 	mii->mii_media_active = IFM_ETHER | IFM_100_TX | IFM_FDX;
3815 }
3816 
3817 static void
3818 re_setwol(struct rl_softc *sc)
3819 {
3820 	struct ifnet		*ifp;
3821 	int			pmc;
3822 	uint16_t		pmstat;
3823 	uint8_t			v;
3824 
3825 	RL_LOCK_ASSERT(sc);
3826 
3827 	if (pci_find_cap(sc->rl_dev, PCIY_PMG, &pmc) != 0)
3828 		return;
3829 
3830 	ifp = sc->rl_ifp;
3831 	/* Put controller into sleep mode. */
3832 	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
3833 		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
3834 			CSR_WRITE_1(sc, RL_GPIO,
3835 			    CSR_READ_1(sc, RL_GPIO) & ~0x01);
3836 	}
3837 	if ((ifp->if_capenable & IFCAP_WOL) != 0) {
3838 		re_set_rxmode(sc);
3839 		if ((sc->rl_flags & RL_FLAG_WOL_MANLINK) != 0)
3840 			re_set_linkspeed(sc);
3841 		if ((sc->rl_flags & RL_FLAG_WOLRXENB) != 0)
3842 			CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RX_ENB);
3843 	}
3844 	/* Enable config register write. */
3845 	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
3846 
3847 	/* Enable PME. */
3848 	v = CSR_READ_1(sc, sc->rl_cfg1);
3849 	v &= ~RL_CFG1_PME;
3850 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3851 		v |= RL_CFG1_PME;
3852 	CSR_WRITE_1(sc, sc->rl_cfg1, v);
3853 
3854 	v = CSR_READ_1(sc, sc->rl_cfg3);
3855 	v &= ~(RL_CFG3_WOL_LINK | RL_CFG3_WOL_MAGIC);
3856 	if ((ifp->if_capenable & IFCAP_WOL_MAGIC) != 0)
3857 		v |= RL_CFG3_WOL_MAGIC;
3858 	CSR_WRITE_1(sc, sc->rl_cfg3, v);
3859 
3860 	v = CSR_READ_1(sc, sc->rl_cfg5);
3861 	v &= ~(RL_CFG5_WOL_BCAST | RL_CFG5_WOL_MCAST | RL_CFG5_WOL_UCAST |
3862 	    RL_CFG5_WOL_LANWAKE);
3863 	if ((ifp->if_capenable & IFCAP_WOL_UCAST) != 0)
3864 		v |= RL_CFG5_WOL_UCAST;
3865 	if ((ifp->if_capenable & IFCAP_WOL_MCAST) != 0)
3866 		v |= RL_CFG5_WOL_MCAST | RL_CFG5_WOL_BCAST;
3867 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3868 		v |= RL_CFG5_WOL_LANWAKE;
3869 	CSR_WRITE_1(sc, sc->rl_cfg5, v);
3870 
3871 	/* Config register write done. */
3872 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3873 
3874 	if ((ifp->if_capenable & IFCAP_WOL) == 0 &&
3875 	    (sc->rl_flags & RL_FLAG_PHYWAKE_PM) != 0)
3876 		CSR_WRITE_1(sc, RL_PMCH, CSR_READ_1(sc, RL_PMCH) & ~0x80);
3877 	/*
3878 	 * It seems that hardware resets its link speed to 100Mbps in
3879 	 * power down mode so switching to 100Mbps in driver is not
3880 	 * needed.
3881 	 */
3882 
3883 	/* Request PME if WOL is requested. */
3884 	pmstat = pci_read_config(sc->rl_dev, pmc + PCIR_POWER_STATUS, 2);
3885 	pmstat &= ~(PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE);
3886 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3887 		pmstat |= PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE;
3888 	pci_write_config(sc->rl_dev, pmc + PCIR_POWER_STATUS, pmstat, 2);
3889 }
3890 
3891 static void
3892 re_clrwol(struct rl_softc *sc)
3893 {
3894 	int			pmc;
3895 	uint8_t			v;
3896 
3897 	RL_LOCK_ASSERT(sc);
3898 
3899 	if (pci_find_cap(sc->rl_dev, PCIY_PMG, &pmc) != 0)
3900 		return;
3901 
3902 	/* Enable config register write. */
3903 	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
3904 
3905 	v = CSR_READ_1(sc, sc->rl_cfg3);
3906 	v &= ~(RL_CFG3_WOL_LINK | RL_CFG3_WOL_MAGIC);
3907 	CSR_WRITE_1(sc, sc->rl_cfg3, v);
3908 
3909 	/* Config register write done. */
3910 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3911 
3912 	v = CSR_READ_1(sc, sc->rl_cfg5);
3913 	v &= ~(RL_CFG5_WOL_BCAST | RL_CFG5_WOL_MCAST | RL_CFG5_WOL_UCAST);
3914 	v &= ~RL_CFG5_WOL_LANWAKE;
3915 	CSR_WRITE_1(sc, sc->rl_cfg5, v);
3916 }
3917 
3918 static void
3919 re_add_sysctls(struct rl_softc *sc)
3920 {
3921 	struct sysctl_ctx_list	*ctx;
3922 	struct sysctl_oid_list	*children;
3923 	int			error;
3924 
3925 	ctx = device_get_sysctl_ctx(sc->rl_dev);
3926 	children = SYSCTL_CHILDREN(device_get_sysctl_tree(sc->rl_dev));
3927 
3928 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "stats",
3929 	    CTLTYPE_INT | CTLFLAG_RW, sc, 0, re_sysctl_stats, "I",
3930 	    "Statistics Information");
3931 	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) == 0)
3932 		return;
3933 
3934 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "int_rx_mod",
3935 	    CTLTYPE_INT | CTLFLAG_RW, &sc->rl_int_rx_mod, 0,
3936 	    sysctl_hw_re_int_mod, "I", "re RX interrupt moderation");
3937 	/* Pull in device tunables. */
3938 	sc->rl_int_rx_mod = RL_TIMER_DEFAULT;
3939 	error = resource_int_value(device_get_name(sc->rl_dev),
3940 	    device_get_unit(sc->rl_dev), "int_rx_mod", &sc->rl_int_rx_mod);
3941 	if (error == 0) {
3942 		if (sc->rl_int_rx_mod < RL_TIMER_MIN ||
3943 		    sc->rl_int_rx_mod > RL_TIMER_MAX) {
3944 			device_printf(sc->rl_dev, "int_rx_mod value out of "
3945 			    "range; using default: %d\n",
3946 			    RL_TIMER_DEFAULT);
3947 			sc->rl_int_rx_mod = RL_TIMER_DEFAULT;
3948 		}
3949 	}
3950 
3951 }
3952 
3953 static int
3954 re_sysctl_stats(SYSCTL_HANDLER_ARGS)
3955 {
3956 	struct rl_softc		*sc;
3957 	struct rl_stats		*stats;
3958 	int			error, i, result;
3959 
3960 	result = -1;
3961 	error = sysctl_handle_int(oidp, &result, 0, req);
3962 	if (error || req->newptr == NULL)
3963 		return (error);
3964 
3965 	if (result == 1) {
3966 		sc = (struct rl_softc *)arg1;
3967 		RL_LOCK(sc);
3968 		if ((sc->rl_ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
3969 			RL_UNLOCK(sc);
3970 			goto done;
3971 		}
3972 		bus_dmamap_sync(sc->rl_ldata.rl_stag,
3973 		    sc->rl_ldata.rl_smap, BUS_DMASYNC_PREREAD);
3974 		CSR_WRITE_4(sc, RL_DUMPSTATS_HI,
3975 		    RL_ADDR_HI(sc->rl_ldata.rl_stats_addr));
3976 		CSR_WRITE_4(sc, RL_DUMPSTATS_LO,
3977 		    RL_ADDR_LO(sc->rl_ldata.rl_stats_addr));
3978 		CSR_WRITE_4(sc, RL_DUMPSTATS_LO,
3979 		    RL_ADDR_LO(sc->rl_ldata.rl_stats_addr |
3980 		    RL_DUMPSTATS_START));
3981 		for (i = RL_TIMEOUT; i > 0; i--) {
3982 			if ((CSR_READ_4(sc, RL_DUMPSTATS_LO) &
3983 			    RL_DUMPSTATS_START) == 0)
3984 				break;
3985 			DELAY(1000);
3986 		}
3987 		bus_dmamap_sync(sc->rl_ldata.rl_stag,
3988 		    sc->rl_ldata.rl_smap, BUS_DMASYNC_POSTREAD);
3989 		RL_UNLOCK(sc);
3990 		if (i == 0) {
3991 			device_printf(sc->rl_dev,
3992 			    "DUMP statistics request timed out\n");
3993 			return (ETIMEDOUT);
3994 		}
3995 done:
3996 		stats = sc->rl_ldata.rl_stats;
3997 		printf("%s statistics:\n", device_get_nameunit(sc->rl_dev));
3998 		printf("Tx frames : %ju\n",
3999 		    (uintmax_t)le64toh(stats->rl_tx_pkts));
4000 		printf("Rx frames : %ju\n",
4001 		    (uintmax_t)le64toh(stats->rl_rx_pkts));
4002 		printf("Tx errors : %ju\n",
4003 		    (uintmax_t)le64toh(stats->rl_tx_errs));
4004 		printf("Rx errors : %u\n",
4005 		    le32toh(stats->rl_rx_errs));
4006 		printf("Rx missed frames : %u\n",
4007 		    (uint32_t)le16toh(stats->rl_missed_pkts));
4008 		printf("Rx frame alignment errs : %u\n",
4009 		    (uint32_t)le16toh(stats->rl_rx_framealign_errs));
4010 		printf("Tx single collisions : %u\n",
4011 		    le32toh(stats->rl_tx_onecoll));
4012 		printf("Tx multiple collisions : %u\n",
4013 		    le32toh(stats->rl_tx_multicolls));
4014 		printf("Rx unicast frames : %ju\n",
4015 		    (uintmax_t)le64toh(stats->rl_rx_ucasts));
4016 		printf("Rx broadcast frames : %ju\n",
4017 		    (uintmax_t)le64toh(stats->rl_rx_bcasts));
4018 		printf("Rx multicast frames : %u\n",
4019 		    le32toh(stats->rl_rx_mcasts));
4020 		printf("Tx aborts : %u\n",
4021 		    (uint32_t)le16toh(stats->rl_tx_aborts));
4022 		printf("Tx underruns : %u\n",
4023 		    (uint32_t)le16toh(stats->rl_rx_underruns));
4024 	}
4025 
4026 	return (error);
4027 }
4028 
4029 static int
4030 sysctl_int_range(SYSCTL_HANDLER_ARGS, int low, int high)
4031 {
4032 	int error, value;
4033 
4034 	if (arg1 == NULL)
4035 		return (EINVAL);
4036 	value = *(int *)arg1;
4037 	error = sysctl_handle_int(oidp, &value, 0, req);
4038 	if (error || req->newptr == NULL)
4039 		return (error);
4040 	if (value < low || value > high)
4041 		return (EINVAL);
4042 	*(int *)arg1 = value;
4043 
4044 	return (0);
4045 }
4046 
4047 static int
4048 sysctl_hw_re_int_mod(SYSCTL_HANDLER_ARGS)
4049 {
4050 
4051 	return (sysctl_int_range(oidp, arg1, arg2, req, RL_TIMER_MIN,
4052 	    RL_TIMER_MAX));
4053 }
4054