xref: /freebsd/sys/dev/re/if_re.c (revision 5dae51da3da0cc94d17bd67b308fad304ebec7e0)
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 	{ NCUBE_VENDORID, RT_DEVICEID_8168, 0,
187 	    "TP-Link TG-3468 v2 (RTL8168) Gigabit Ethernet" },
188 	{ RT_VENDORID, RT_DEVICEID_8169, 0,
189 	    "RealTek 8169/8169S/8169SB(L)/8110S/8110SB(L) Gigabit Ethernet" },
190 	{ RT_VENDORID, RT_DEVICEID_8169SC, 0,
191 	    "RealTek 8169SC/8110SC Single-chip Gigabit Ethernet" },
192 	{ COREGA_VENDORID, COREGA_DEVICEID_CGLAPCIGT, 0,
193 	    "Corega CG-LAPCIGT (RTL8169S) Gigabit Ethernet" },
194 	{ LINKSYS_VENDORID, LINKSYS_DEVICEID_EG1032, 0,
195 	    "Linksys EG1032 (RTL8169S) Gigabit Ethernet" },
196 	{ USR_VENDORID, USR_DEVICEID_997902, 0,
197 	    "US Robotics 997902 (RTL8169S) Gigabit Ethernet" }
198 };
199 
200 static const struct rl_hwrev re_hwrevs[] = {
201 	{ RL_HWREV_8139, RL_8139, "", RL_MTU },
202 	{ RL_HWREV_8139A, RL_8139, "A", RL_MTU },
203 	{ RL_HWREV_8139AG, RL_8139, "A-G", RL_MTU },
204 	{ RL_HWREV_8139B, RL_8139, "B", RL_MTU },
205 	{ RL_HWREV_8130, RL_8139, "8130", RL_MTU },
206 	{ RL_HWREV_8139C, RL_8139, "C", RL_MTU },
207 	{ RL_HWREV_8139D, RL_8139, "8139D/8100B/8100C", RL_MTU },
208 	{ RL_HWREV_8139CPLUS, RL_8139CPLUS, "C+", RL_MTU },
209 	{ RL_HWREV_8168B_SPIN1, RL_8169, "8168", RL_JUMBO_MTU },
210 	{ RL_HWREV_8169, RL_8169, "8169", RL_JUMBO_MTU },
211 	{ RL_HWREV_8169S, RL_8169, "8169S", RL_JUMBO_MTU },
212 	{ RL_HWREV_8110S, RL_8169, "8110S", RL_JUMBO_MTU },
213 	{ RL_HWREV_8169_8110SB, RL_8169, "8169SB/8110SB", RL_JUMBO_MTU },
214 	{ RL_HWREV_8169_8110SC, RL_8169, "8169SC/8110SC", RL_JUMBO_MTU },
215 	{ RL_HWREV_8169_8110SBL, RL_8169, "8169SBL/8110SBL", RL_JUMBO_MTU },
216 	{ RL_HWREV_8169_8110SCE, RL_8169, "8169SC/8110SC", RL_JUMBO_MTU },
217 	{ RL_HWREV_8100, RL_8139, "8100", RL_MTU },
218 	{ RL_HWREV_8101, RL_8139, "8101", RL_MTU },
219 	{ RL_HWREV_8100E, RL_8169, "8100E", RL_MTU },
220 	{ RL_HWREV_8101E, RL_8169, "8101E", RL_MTU },
221 	{ RL_HWREV_8102E, RL_8169, "8102E", RL_MTU },
222 	{ RL_HWREV_8102EL, RL_8169, "8102EL", RL_MTU },
223 	{ RL_HWREV_8102EL_SPIN1, RL_8169, "8102EL", RL_MTU },
224 	{ RL_HWREV_8103E, RL_8169, "8103E", RL_MTU },
225 	{ RL_HWREV_8401E, RL_8169, "8401E", RL_MTU },
226 	{ RL_HWREV_8402, RL_8169, "8402", RL_MTU },
227 	{ RL_HWREV_8105E, RL_8169, "8105E", RL_MTU },
228 	{ RL_HWREV_8105E_SPIN1, RL_8169, "8105E", RL_MTU },
229 	{ RL_HWREV_8106E, RL_8169, "8106E", RL_MTU },
230 	{ RL_HWREV_8168B_SPIN2, RL_8169, "8168", RL_JUMBO_MTU },
231 	{ RL_HWREV_8168B_SPIN3, RL_8169, "8168", RL_JUMBO_MTU },
232 	{ RL_HWREV_8168C, RL_8169, "8168C/8111C", RL_JUMBO_MTU_6K },
233 	{ RL_HWREV_8168C_SPIN2, RL_8169, "8168C/8111C", RL_JUMBO_MTU_6K },
234 	{ RL_HWREV_8168CP, RL_8169, "8168CP/8111CP", RL_JUMBO_MTU_6K },
235 	{ RL_HWREV_8168D, RL_8169, "8168D/8111D", RL_JUMBO_MTU_9K },
236 	{ RL_HWREV_8168DP, RL_8169, "8168DP/8111DP", RL_JUMBO_MTU_9K },
237 	{ RL_HWREV_8168E, RL_8169, "8168E/8111E", RL_JUMBO_MTU_9K},
238 	{ RL_HWREV_8168E_VL, RL_8169, "8168E/8111E-VL", RL_JUMBO_MTU_6K},
239 	{ RL_HWREV_8168EP, RL_8169, "8168EP/8111EP", RL_JUMBO_MTU_9K},
240 	{ RL_HWREV_8168F, RL_8169, "8168F/8111F", RL_JUMBO_MTU_9K},
241 	{ RL_HWREV_8168G, RL_8169, "8168G/8111G", RL_JUMBO_MTU_9K},
242 	{ RL_HWREV_8168GU, RL_8169, "8168GU/8111GU", RL_JUMBO_MTU_9K},
243 	{ RL_HWREV_8168H, RL_8169, "8168H/8111H", RL_JUMBO_MTU_9K},
244 	{ RL_HWREV_8411, RL_8169, "8411", RL_JUMBO_MTU_9K},
245 	{ RL_HWREV_8411B, RL_8169, "8411B", RL_JUMBO_MTU_9K},
246 	{ 0, 0, NULL, 0 }
247 };
248 
249 static int re_probe		(device_t);
250 static int re_attach		(device_t);
251 static int re_detach		(device_t);
252 
253 static int re_encap		(struct rl_softc *, struct mbuf **);
254 
255 static void re_dma_map_addr	(void *, bus_dma_segment_t *, int, int);
256 static int re_allocmem		(device_t, struct rl_softc *);
257 static __inline void re_discard_rxbuf
258 				(struct rl_softc *, int);
259 static int re_newbuf		(struct rl_softc *, int);
260 static int re_jumbo_newbuf	(struct rl_softc *, int);
261 static int re_rx_list_init	(struct rl_softc *);
262 static int re_jrx_list_init	(struct rl_softc *);
263 static int re_tx_list_init	(struct rl_softc *);
264 #ifdef RE_FIXUP_RX
265 static __inline void re_fixup_rx
266 				(struct mbuf *);
267 #endif
268 static int re_rxeof		(struct rl_softc *, int *);
269 static void re_txeof		(struct rl_softc *);
270 #ifdef DEVICE_POLLING
271 static int re_poll		(struct ifnet *, enum poll_cmd, int);
272 static int re_poll_locked	(struct ifnet *, enum poll_cmd, int);
273 #endif
274 static int re_intr		(void *);
275 static void re_intr_msi		(void *);
276 static void re_tick		(void *);
277 static void re_int_task		(void *, int);
278 static void re_start		(struct ifnet *);
279 static void re_start_locked	(struct ifnet *);
280 static int re_ioctl		(struct ifnet *, u_long, caddr_t);
281 static void re_init		(void *);
282 static void re_init_locked	(struct rl_softc *);
283 static void re_stop		(struct rl_softc *);
284 static void re_watchdog		(struct rl_softc *);
285 static int re_suspend		(device_t);
286 static int re_resume		(device_t);
287 static int re_shutdown		(device_t);
288 static int re_ifmedia_upd	(struct ifnet *);
289 static void re_ifmedia_sts	(struct ifnet *, struct ifmediareq *);
290 
291 static void re_eeprom_putbyte	(struct rl_softc *, int);
292 static void re_eeprom_getword	(struct rl_softc *, int, u_int16_t *);
293 static void re_read_eeprom	(struct rl_softc *, caddr_t, int, int);
294 static int re_gmii_readreg	(device_t, int, int);
295 static int re_gmii_writereg	(device_t, int, int, int);
296 
297 static int re_miibus_readreg	(device_t, int, int);
298 static int re_miibus_writereg	(device_t, int, int, int);
299 static void re_miibus_statchg	(device_t);
300 
301 static void re_set_jumbo	(struct rl_softc *, int);
302 static void re_set_rxmode		(struct rl_softc *);
303 static void re_reset		(struct rl_softc *);
304 static void re_setwol		(struct rl_softc *);
305 static void re_clrwol		(struct rl_softc *);
306 static void re_set_linkspeed	(struct rl_softc *);
307 
308 #ifdef DEV_NETMAP	/* see ixgbe.c for details */
309 #include <dev/netmap/if_re_netmap.h>
310 MODULE_DEPEND(re, netmap, 1, 1, 1);
311 #endif /* !DEV_NETMAP */
312 
313 #ifdef RE_DIAG
314 static int re_diag		(struct rl_softc *);
315 #endif
316 
317 static void re_add_sysctls	(struct rl_softc *);
318 static int re_sysctl_stats	(SYSCTL_HANDLER_ARGS);
319 static int sysctl_int_range	(SYSCTL_HANDLER_ARGS, int, int);
320 static int sysctl_hw_re_int_mod	(SYSCTL_HANDLER_ARGS);
321 
322 static device_method_t re_methods[] = {
323 	/* Device interface */
324 	DEVMETHOD(device_probe,		re_probe),
325 	DEVMETHOD(device_attach,	re_attach),
326 	DEVMETHOD(device_detach,	re_detach),
327 	DEVMETHOD(device_suspend,	re_suspend),
328 	DEVMETHOD(device_resume,	re_resume),
329 	DEVMETHOD(device_shutdown,	re_shutdown),
330 
331 	/* MII interface */
332 	DEVMETHOD(miibus_readreg,	re_miibus_readreg),
333 	DEVMETHOD(miibus_writereg,	re_miibus_writereg),
334 	DEVMETHOD(miibus_statchg,	re_miibus_statchg),
335 
336 	DEVMETHOD_END
337 };
338 
339 static driver_t re_driver = {
340 	"re",
341 	re_methods,
342 	sizeof(struct rl_softc)
343 };
344 
345 static devclass_t re_devclass;
346 
347 DRIVER_MODULE(re, pci, re_driver, re_devclass, 0, 0);
348 DRIVER_MODULE(miibus, re, miibus_driver, miibus_devclass, 0, 0);
349 
350 #define EE_SET(x)					\
351 	CSR_WRITE_1(sc, RL_EECMD,			\
352 		CSR_READ_1(sc, RL_EECMD) | x)
353 
354 #define EE_CLR(x)					\
355 	CSR_WRITE_1(sc, RL_EECMD,			\
356 		CSR_READ_1(sc, RL_EECMD) & ~x)
357 
358 /*
359  * Send a read command and address to the EEPROM, check for ACK.
360  */
361 static void
362 re_eeprom_putbyte(struct rl_softc *sc, int addr)
363 {
364 	int			d, i;
365 
366 	d = addr | (RL_9346_READ << sc->rl_eewidth);
367 
368 	/*
369 	 * Feed in each bit and strobe the clock.
370 	 */
371 
372 	for (i = 1 << (sc->rl_eewidth + 3); i; i >>= 1) {
373 		if (d & i) {
374 			EE_SET(RL_EE_DATAIN);
375 		} else {
376 			EE_CLR(RL_EE_DATAIN);
377 		}
378 		DELAY(100);
379 		EE_SET(RL_EE_CLK);
380 		DELAY(150);
381 		EE_CLR(RL_EE_CLK);
382 		DELAY(100);
383 	}
384 }
385 
386 /*
387  * Read a word of data stored in the EEPROM at address 'addr.'
388  */
389 static void
390 re_eeprom_getword(struct rl_softc *sc, int addr, u_int16_t *dest)
391 {
392 	int			i;
393 	u_int16_t		word = 0;
394 
395 	/*
396 	 * Send address of word we want to read.
397 	 */
398 	re_eeprom_putbyte(sc, addr);
399 
400 	/*
401 	 * Start reading bits from EEPROM.
402 	 */
403 	for (i = 0x8000; i; i >>= 1) {
404 		EE_SET(RL_EE_CLK);
405 		DELAY(100);
406 		if (CSR_READ_1(sc, RL_EECMD) & RL_EE_DATAOUT)
407 			word |= i;
408 		EE_CLR(RL_EE_CLK);
409 		DELAY(100);
410 	}
411 
412 	*dest = word;
413 }
414 
415 /*
416  * Read a sequence of words from the EEPROM.
417  */
418 static void
419 re_read_eeprom(struct rl_softc *sc, caddr_t dest, int off, int cnt)
420 {
421 	int			i;
422 	u_int16_t		word = 0, *ptr;
423 
424 	CSR_SETBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM);
425 
426         DELAY(100);
427 
428 	for (i = 0; i < cnt; i++) {
429 		CSR_SETBIT_1(sc, RL_EECMD, RL_EE_SEL);
430 		re_eeprom_getword(sc, off + i, &word);
431 		CSR_CLRBIT_1(sc, RL_EECMD, RL_EE_SEL);
432 		ptr = (u_int16_t *)(dest + (i * 2));
433                 *ptr = word;
434 	}
435 
436 	CSR_CLRBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM);
437 }
438 
439 static int
440 re_gmii_readreg(device_t dev, int phy, int reg)
441 {
442 	struct rl_softc		*sc;
443 	u_int32_t		rval;
444 	int			i;
445 
446 	sc = device_get_softc(dev);
447 
448 	/* Let the rgephy driver read the GMEDIASTAT register */
449 
450 	if (reg == RL_GMEDIASTAT) {
451 		rval = CSR_READ_1(sc, RL_GMEDIASTAT);
452 		return (rval);
453 	}
454 
455 	CSR_WRITE_4(sc, RL_PHYAR, reg << 16);
456 
457 	for (i = 0; i < RL_PHY_TIMEOUT; i++) {
458 		rval = CSR_READ_4(sc, RL_PHYAR);
459 		if (rval & RL_PHYAR_BUSY)
460 			break;
461 		DELAY(25);
462 	}
463 
464 	if (i == RL_PHY_TIMEOUT) {
465 		device_printf(sc->rl_dev, "PHY read failed\n");
466 		return (0);
467 	}
468 
469 	/*
470 	 * Controller requires a 20us delay to process next MDIO request.
471 	 */
472 	DELAY(20);
473 
474 	return (rval & RL_PHYAR_PHYDATA);
475 }
476 
477 static int
478 re_gmii_writereg(device_t dev, int phy, int reg, int data)
479 {
480 	struct rl_softc		*sc;
481 	u_int32_t		rval;
482 	int			i;
483 
484 	sc = device_get_softc(dev);
485 
486 	CSR_WRITE_4(sc, RL_PHYAR, (reg << 16) |
487 	    (data & RL_PHYAR_PHYDATA) | RL_PHYAR_BUSY);
488 
489 	for (i = 0; i < RL_PHY_TIMEOUT; i++) {
490 		rval = CSR_READ_4(sc, RL_PHYAR);
491 		if (!(rval & RL_PHYAR_BUSY))
492 			break;
493 		DELAY(25);
494 	}
495 
496 	if (i == RL_PHY_TIMEOUT) {
497 		device_printf(sc->rl_dev, "PHY write failed\n");
498 		return (0);
499 	}
500 
501 	/*
502 	 * Controller requires a 20us delay to process next MDIO request.
503 	 */
504 	DELAY(20);
505 
506 	return (0);
507 }
508 
509 static int
510 re_miibus_readreg(device_t dev, int phy, int reg)
511 {
512 	struct rl_softc		*sc;
513 	u_int16_t		rval = 0;
514 	u_int16_t		re8139_reg = 0;
515 
516 	sc = device_get_softc(dev);
517 
518 	if (sc->rl_type == RL_8169) {
519 		rval = re_gmii_readreg(dev, phy, reg);
520 		return (rval);
521 	}
522 
523 	switch (reg) {
524 	case MII_BMCR:
525 		re8139_reg = RL_BMCR;
526 		break;
527 	case MII_BMSR:
528 		re8139_reg = RL_BMSR;
529 		break;
530 	case MII_ANAR:
531 		re8139_reg = RL_ANAR;
532 		break;
533 	case MII_ANER:
534 		re8139_reg = RL_ANER;
535 		break;
536 	case MII_ANLPAR:
537 		re8139_reg = RL_LPAR;
538 		break;
539 	case MII_PHYIDR1:
540 	case MII_PHYIDR2:
541 		return (0);
542 	/*
543 	 * Allow the rlphy driver to read the media status
544 	 * register. If we have a link partner which does not
545 	 * support NWAY, this is the register which will tell
546 	 * us the results of parallel detection.
547 	 */
548 	case RL_MEDIASTAT:
549 		rval = CSR_READ_1(sc, RL_MEDIASTAT);
550 		return (rval);
551 	default:
552 		device_printf(sc->rl_dev, "bad phy register\n");
553 		return (0);
554 	}
555 	rval = CSR_READ_2(sc, re8139_reg);
556 	if (sc->rl_type == RL_8139CPLUS && re8139_reg == RL_BMCR) {
557 		/* 8139C+ has different bit layout. */
558 		rval &= ~(BMCR_LOOP | BMCR_ISO);
559 	}
560 	return (rval);
561 }
562 
563 static int
564 re_miibus_writereg(device_t dev, int phy, int reg, int data)
565 {
566 	struct rl_softc		*sc;
567 	u_int16_t		re8139_reg = 0;
568 	int			rval = 0;
569 
570 	sc = device_get_softc(dev);
571 
572 	if (sc->rl_type == RL_8169) {
573 		rval = re_gmii_writereg(dev, phy, reg, data);
574 		return (rval);
575 	}
576 
577 	switch (reg) {
578 	case MII_BMCR:
579 		re8139_reg = RL_BMCR;
580 		if (sc->rl_type == RL_8139CPLUS) {
581 			/* 8139C+ has different bit layout. */
582 			data &= ~(BMCR_LOOP | BMCR_ISO);
583 		}
584 		break;
585 	case MII_BMSR:
586 		re8139_reg = RL_BMSR;
587 		break;
588 	case MII_ANAR:
589 		re8139_reg = RL_ANAR;
590 		break;
591 	case MII_ANER:
592 		re8139_reg = RL_ANER;
593 		break;
594 	case MII_ANLPAR:
595 		re8139_reg = RL_LPAR;
596 		break;
597 	case MII_PHYIDR1:
598 	case MII_PHYIDR2:
599 		return (0);
600 		break;
601 	default:
602 		device_printf(sc->rl_dev, "bad phy register\n");
603 		return (0);
604 	}
605 	CSR_WRITE_2(sc, re8139_reg, data);
606 	return (0);
607 }
608 
609 static void
610 re_miibus_statchg(device_t dev)
611 {
612 	struct rl_softc		*sc;
613 	struct ifnet		*ifp;
614 	struct mii_data		*mii;
615 
616 	sc = device_get_softc(dev);
617 	mii = device_get_softc(sc->rl_miibus);
618 	ifp = sc->rl_ifp;
619 	if (mii == NULL || ifp == NULL ||
620 	    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
621 		return;
622 
623 	sc->rl_flags &= ~RL_FLAG_LINK;
624 	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
625 	    (IFM_ACTIVE | IFM_AVALID)) {
626 		switch (IFM_SUBTYPE(mii->mii_media_active)) {
627 		case IFM_10_T:
628 		case IFM_100_TX:
629 			sc->rl_flags |= RL_FLAG_LINK;
630 			break;
631 		case IFM_1000_T:
632 			if ((sc->rl_flags & RL_FLAG_FASTETHER) != 0)
633 				break;
634 			sc->rl_flags |= RL_FLAG_LINK;
635 			break;
636 		default:
637 			break;
638 		}
639 	}
640 	/*
641 	 * RealTek controllers do not provide any interface to the RX/TX
642 	 * MACs for resolved speed, duplex and flow-control parameters.
643 	 */
644 }
645 
646 /*
647  * Set the RX configuration and 64-bit multicast hash filter.
648  */
649 static void
650 re_set_rxmode(struct rl_softc *sc)
651 {
652 	struct ifnet		*ifp;
653 	struct ifmultiaddr	*ifma;
654 	uint32_t		hashes[2] = { 0, 0 };
655 	uint32_t		h, rxfilt;
656 
657 	RL_LOCK_ASSERT(sc);
658 
659 	ifp = sc->rl_ifp;
660 
661 	rxfilt = RL_RXCFG_CONFIG | RL_RXCFG_RX_INDIV | RL_RXCFG_RX_BROAD;
662 	if ((sc->rl_flags & RL_FLAG_EARLYOFF) != 0)
663 		rxfilt |= RL_RXCFG_EARLYOFF;
664 	else if ((sc->rl_flags & RL_FLAG_8168G_PLUS) != 0)
665 		rxfilt |= RL_RXCFG_EARLYOFFV2;
666 
667 	if (ifp->if_flags & (IFF_ALLMULTI | IFF_PROMISC)) {
668 		if (ifp->if_flags & IFF_PROMISC)
669 			rxfilt |= RL_RXCFG_RX_ALLPHYS;
670 		/*
671 		 * Unlike other hardwares, we have to explicitly set
672 		 * RL_RXCFG_RX_MULTI to receive multicast frames in
673 		 * promiscuous mode.
674 		 */
675 		rxfilt |= RL_RXCFG_RX_MULTI;
676 		hashes[0] = hashes[1] = 0xffffffff;
677 		goto done;
678 	}
679 
680 	if_maddr_rlock(ifp);
681 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
682 		if (ifma->ifma_addr->sa_family != AF_LINK)
683 			continue;
684 		h = ether_crc32_be(LLADDR((struct sockaddr_dl *)
685 		    ifma->ifma_addr), ETHER_ADDR_LEN) >> 26;
686 		if (h < 32)
687 			hashes[0] |= (1 << h);
688 		else
689 			hashes[1] |= (1 << (h - 32));
690 	}
691 	if_maddr_runlock(ifp);
692 
693 	if (hashes[0] != 0 || hashes[1] != 0) {
694 		/*
695 		 * For some unfathomable reason, RealTek decided to
696 		 * reverse the order of the multicast hash registers
697 		 * in the PCI Express parts.  This means we have to
698 		 * write the hash pattern in reverse order for those
699 		 * devices.
700 		 */
701 		if ((sc->rl_flags & RL_FLAG_PCIE) != 0) {
702 			h = bswap32(hashes[0]);
703 			hashes[0] = bswap32(hashes[1]);
704 			hashes[1] = h;
705 		}
706 		rxfilt |= RL_RXCFG_RX_MULTI;
707 	}
708 
709 	if  (sc->rl_hwrev->rl_rev == RL_HWREV_8168F) {
710 		/* Disable multicast filtering due to silicon bug. */
711 		hashes[0] = 0xffffffff;
712 		hashes[1] = 0xffffffff;
713 	}
714 
715 done:
716 	CSR_WRITE_4(sc, RL_MAR0, hashes[0]);
717 	CSR_WRITE_4(sc, RL_MAR4, hashes[1]);
718 	CSR_WRITE_4(sc, RL_RXCFG, rxfilt);
719 }
720 
721 static void
722 re_reset(struct rl_softc *sc)
723 {
724 	int			i;
725 
726 	RL_LOCK_ASSERT(sc);
727 
728 	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RESET);
729 
730 	for (i = 0; i < RL_TIMEOUT; i++) {
731 		DELAY(10);
732 		if (!(CSR_READ_1(sc, RL_COMMAND) & RL_CMD_RESET))
733 			break;
734 	}
735 	if (i == RL_TIMEOUT)
736 		device_printf(sc->rl_dev, "reset never completed!\n");
737 
738 	if ((sc->rl_flags & RL_FLAG_MACRESET) != 0)
739 		CSR_WRITE_1(sc, 0x82, 1);
740 	if (sc->rl_hwrev->rl_rev == RL_HWREV_8169S)
741 		re_gmii_writereg(sc->rl_dev, 1, 0x0b, 0);
742 }
743 
744 #ifdef RE_DIAG
745 
746 /*
747  * The following routine is designed to test for a defect on some
748  * 32-bit 8169 cards. Some of these NICs have the REQ64# and ACK64#
749  * lines connected to the bus, however for a 32-bit only card, they
750  * should be pulled high. The result of this defect is that the
751  * NIC will not work right if you plug it into a 64-bit slot: DMA
752  * operations will be done with 64-bit transfers, which will fail
753  * because the 64-bit data lines aren't connected.
754  *
755  * There's no way to work around this (short of talking a soldering
756  * iron to the board), however we can detect it. The method we use
757  * here is to put the NIC into digital loopback mode, set the receiver
758  * to promiscuous mode, and then try to send a frame. We then compare
759  * the frame data we sent to what was received. If the data matches,
760  * then the NIC is working correctly, otherwise we know the user has
761  * a defective NIC which has been mistakenly plugged into a 64-bit PCI
762  * slot. In the latter case, there's no way the NIC can work correctly,
763  * so we print out a message on the console and abort the device attach.
764  */
765 
766 static int
767 re_diag(struct rl_softc *sc)
768 {
769 	struct ifnet		*ifp = sc->rl_ifp;
770 	struct mbuf		*m0;
771 	struct ether_header	*eh;
772 	struct rl_desc		*cur_rx;
773 	u_int16_t		status;
774 	u_int32_t		rxstat;
775 	int			total_len, i, error = 0, phyaddr;
776 	u_int8_t		dst[] = { 0x00, 'h', 'e', 'l', 'l', 'o' };
777 	u_int8_t		src[] = { 0x00, 'w', 'o', 'r', 'l', 'd' };
778 
779 	/* Allocate a single mbuf */
780 	MGETHDR(m0, M_NOWAIT, MT_DATA);
781 	if (m0 == NULL)
782 		return (ENOBUFS);
783 
784 	RL_LOCK(sc);
785 
786 	/*
787 	 * Initialize the NIC in test mode. This sets the chip up
788 	 * so that it can send and receive frames, but performs the
789 	 * following special functions:
790 	 * - Puts receiver in promiscuous mode
791 	 * - Enables digital loopback mode
792 	 * - Leaves interrupts turned off
793 	 */
794 
795 	ifp->if_flags |= IFF_PROMISC;
796 	sc->rl_testmode = 1;
797 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
798 	re_init_locked(sc);
799 	sc->rl_flags |= RL_FLAG_LINK;
800 	if (sc->rl_type == RL_8169)
801 		phyaddr = 1;
802 	else
803 		phyaddr = 0;
804 
805 	re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_RESET);
806 	for (i = 0; i < RL_TIMEOUT; i++) {
807 		status = re_miibus_readreg(sc->rl_dev, phyaddr, MII_BMCR);
808 		if (!(status & BMCR_RESET))
809 			break;
810 	}
811 
812 	re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_LOOP);
813 	CSR_WRITE_2(sc, RL_ISR, RL_INTRS);
814 
815 	DELAY(100000);
816 
817 	/* Put some data in the mbuf */
818 
819 	eh = mtod(m0, struct ether_header *);
820 	bcopy ((char *)&dst, eh->ether_dhost, ETHER_ADDR_LEN);
821 	bcopy ((char *)&src, eh->ether_shost, ETHER_ADDR_LEN);
822 	eh->ether_type = htons(ETHERTYPE_IP);
823 	m0->m_pkthdr.len = m0->m_len = ETHER_MIN_LEN - ETHER_CRC_LEN;
824 
825 	/*
826 	 * Queue the packet, start transmission.
827 	 * Note: IF_HANDOFF() ultimately calls re_start() for us.
828 	 */
829 
830 	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
831 	RL_UNLOCK(sc);
832 	/* XXX: re_diag must not be called when in ALTQ mode */
833 	IF_HANDOFF(&ifp->if_snd, m0, ifp);
834 	RL_LOCK(sc);
835 	m0 = NULL;
836 
837 	/* Wait for it to propagate through the chip */
838 
839 	DELAY(100000);
840 	for (i = 0; i < RL_TIMEOUT; i++) {
841 		status = CSR_READ_2(sc, RL_ISR);
842 		CSR_WRITE_2(sc, RL_ISR, status);
843 		if ((status & (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK)) ==
844 		    (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK))
845 			break;
846 		DELAY(10);
847 	}
848 
849 	if (i == RL_TIMEOUT) {
850 		device_printf(sc->rl_dev,
851 		    "diagnostic failed, failed to receive packet in"
852 		    " loopback mode\n");
853 		error = EIO;
854 		goto done;
855 	}
856 
857 	/*
858 	 * The packet should have been dumped into the first
859 	 * entry in the RX DMA ring. Grab it from there.
860 	 */
861 
862 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
863 	    sc->rl_ldata.rl_rx_list_map,
864 	    BUS_DMASYNC_POSTREAD);
865 	bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag,
866 	    sc->rl_ldata.rl_rx_desc[0].rx_dmamap,
867 	    BUS_DMASYNC_POSTREAD);
868 	bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag,
869 	    sc->rl_ldata.rl_rx_desc[0].rx_dmamap);
870 
871 	m0 = sc->rl_ldata.rl_rx_desc[0].rx_m;
872 	sc->rl_ldata.rl_rx_desc[0].rx_m = NULL;
873 	eh = mtod(m0, struct ether_header *);
874 
875 	cur_rx = &sc->rl_ldata.rl_rx_list[0];
876 	total_len = RL_RXBYTES(cur_rx);
877 	rxstat = le32toh(cur_rx->rl_cmdstat);
878 
879 	if (total_len != ETHER_MIN_LEN) {
880 		device_printf(sc->rl_dev,
881 		    "diagnostic failed, received short packet\n");
882 		error = EIO;
883 		goto done;
884 	}
885 
886 	/* Test that the received packet data matches what we sent. */
887 
888 	if (bcmp((char *)&eh->ether_dhost, (char *)&dst, ETHER_ADDR_LEN) ||
889 	    bcmp((char *)&eh->ether_shost, (char *)&src, ETHER_ADDR_LEN) ||
890 	    ntohs(eh->ether_type) != ETHERTYPE_IP) {
891 		device_printf(sc->rl_dev, "WARNING, DMA FAILURE!\n");
892 		device_printf(sc->rl_dev, "expected TX data: %6D/%6D/0x%x\n",
893 		    dst, ":", src, ":", ETHERTYPE_IP);
894 		device_printf(sc->rl_dev, "received RX data: %6D/%6D/0x%x\n",
895 		    eh->ether_dhost, ":", eh->ether_shost, ":",
896 		    ntohs(eh->ether_type));
897 		device_printf(sc->rl_dev, "You may have a defective 32-bit "
898 		    "NIC plugged into a 64-bit PCI slot.\n");
899 		device_printf(sc->rl_dev, "Please re-install the NIC in a "
900 		    "32-bit slot for proper operation.\n");
901 		device_printf(sc->rl_dev, "Read the re(4) man page for more "
902 		    "details.\n");
903 		error = EIO;
904 	}
905 
906 done:
907 	/* Turn interface off, release resources */
908 
909 	sc->rl_testmode = 0;
910 	sc->rl_flags &= ~RL_FLAG_LINK;
911 	ifp->if_flags &= ~IFF_PROMISC;
912 	re_stop(sc);
913 	if (m0 != NULL)
914 		m_freem(m0);
915 
916 	RL_UNLOCK(sc);
917 
918 	return (error);
919 }
920 
921 #endif
922 
923 /*
924  * Probe for a RealTek 8139C+/8169/8110 chip. Check the PCI vendor and device
925  * IDs against our list and return a device name if we find a match.
926  */
927 static int
928 re_probe(device_t dev)
929 {
930 	const struct rl_type	*t;
931 	uint16_t		devid, vendor;
932 	uint16_t		revid, sdevid;
933 	int			i;
934 
935 	vendor = pci_get_vendor(dev);
936 	devid = pci_get_device(dev);
937 	revid = pci_get_revid(dev);
938 	sdevid = pci_get_subdevice(dev);
939 
940 	if (vendor == LINKSYS_VENDORID && devid == LINKSYS_DEVICEID_EG1032) {
941 		if (sdevid != LINKSYS_SUBDEVICE_EG1032_REV3) {
942 			/*
943 			 * Only attach to rev. 3 of the Linksys EG1032 adapter.
944 			 * Rev. 2 is supported by sk(4).
945 			 */
946 			return (ENXIO);
947 		}
948 	}
949 
950 	if (vendor == RT_VENDORID && devid == RT_DEVICEID_8139) {
951 		if (revid != 0x20) {
952 			/* 8139, let rl(4) take care of this device. */
953 			return (ENXIO);
954 		}
955 	}
956 
957 	t = re_devs;
958 	for (i = 0; i < nitems(re_devs); i++, t++) {
959 		if (vendor == t->rl_vid && devid == t->rl_did) {
960 			device_set_desc(dev, t->rl_name);
961 			return (BUS_PROBE_DEFAULT);
962 		}
963 	}
964 
965 	return (ENXIO);
966 }
967 
968 /*
969  * Map a single buffer address.
970  */
971 
972 static void
973 re_dma_map_addr(void *arg, bus_dma_segment_t *segs, int nseg, int error)
974 {
975 	bus_addr_t		*addr;
976 
977 	if (error)
978 		return;
979 
980 	KASSERT(nseg == 1, ("too many DMA segments, %d should be 1", nseg));
981 	addr = arg;
982 	*addr = segs->ds_addr;
983 }
984 
985 static int
986 re_allocmem(device_t dev, struct rl_softc *sc)
987 {
988 	bus_addr_t		lowaddr;
989 	bus_size_t		rx_list_size, tx_list_size;
990 	int			error;
991 	int			i;
992 
993 	rx_list_size = sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc);
994 	tx_list_size = sc->rl_ldata.rl_tx_desc_cnt * sizeof(struct rl_desc);
995 
996 	/*
997 	 * Allocate the parent bus DMA tag appropriate for PCI.
998 	 * In order to use DAC, RL_CPLUSCMD_PCI_DAC bit of RL_CPLUS_CMD
999 	 * register should be set. However some RealTek chips are known
1000 	 * to be buggy on DAC handling, therefore disable DAC by limiting
1001 	 * DMA address space to 32bit. PCIe variants of RealTek chips
1002 	 * may not have the limitation.
1003 	 */
1004 	lowaddr = BUS_SPACE_MAXADDR;
1005 	if ((sc->rl_flags & RL_FLAG_PCIE) == 0)
1006 		lowaddr = BUS_SPACE_MAXADDR_32BIT;
1007 	error = bus_dma_tag_create(bus_get_dma_tag(dev), 1, 0,
1008 	    lowaddr, BUS_SPACE_MAXADDR, NULL, NULL,
1009 	    BUS_SPACE_MAXSIZE_32BIT, 0, BUS_SPACE_MAXSIZE_32BIT, 0,
1010 	    NULL, NULL, &sc->rl_parent_tag);
1011 	if (error) {
1012 		device_printf(dev, "could not allocate parent DMA tag\n");
1013 		return (error);
1014 	}
1015 
1016 	/*
1017 	 * Allocate map for TX mbufs.
1018 	 */
1019 	error = bus_dma_tag_create(sc->rl_parent_tag, 1, 0,
1020 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL,
1021 	    NULL, MCLBYTES * RL_NTXSEGS, RL_NTXSEGS, 4096, 0,
1022 	    NULL, NULL, &sc->rl_ldata.rl_tx_mtag);
1023 	if (error) {
1024 		device_printf(dev, "could not allocate TX DMA tag\n");
1025 		return (error);
1026 	}
1027 
1028 	/*
1029 	 * Allocate map for RX mbufs.
1030 	 */
1031 
1032 	if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
1033 		error = bus_dma_tag_create(sc->rl_parent_tag, sizeof(uint64_t),
1034 		    0, BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
1035 		    MJUM9BYTES, 1, MJUM9BYTES, 0, NULL, NULL,
1036 		    &sc->rl_ldata.rl_jrx_mtag);
1037 		if (error) {
1038 			device_printf(dev,
1039 			    "could not allocate jumbo RX DMA tag\n");
1040 			return (error);
1041 		}
1042 	}
1043 	error = bus_dma_tag_create(sc->rl_parent_tag, sizeof(uint64_t), 0,
1044 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
1045 	    MCLBYTES, 1, MCLBYTES, 0, NULL, NULL, &sc->rl_ldata.rl_rx_mtag);
1046 	if (error) {
1047 		device_printf(dev, "could not allocate RX DMA tag\n");
1048 		return (error);
1049 	}
1050 
1051 	/*
1052 	 * Allocate map for TX descriptor list.
1053 	 */
1054 	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
1055 	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1056 	    NULL, tx_list_size, 1, tx_list_size, 0,
1057 	    NULL, NULL, &sc->rl_ldata.rl_tx_list_tag);
1058 	if (error) {
1059 		device_printf(dev, "could not allocate TX DMA ring tag\n");
1060 		return (error);
1061 	}
1062 
1063 	/* Allocate DMA'able memory for the TX ring */
1064 
1065 	error = bus_dmamem_alloc(sc->rl_ldata.rl_tx_list_tag,
1066 	    (void **)&sc->rl_ldata.rl_tx_list,
1067 	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1068 	    &sc->rl_ldata.rl_tx_list_map);
1069 	if (error) {
1070 		device_printf(dev, "could not allocate TX DMA ring\n");
1071 		return (error);
1072 	}
1073 
1074 	/* Load the map for the TX ring. */
1075 
1076 	sc->rl_ldata.rl_tx_list_addr = 0;
1077 	error = bus_dmamap_load(sc->rl_ldata.rl_tx_list_tag,
1078 	     sc->rl_ldata.rl_tx_list_map, sc->rl_ldata.rl_tx_list,
1079 	     tx_list_size, re_dma_map_addr,
1080 	     &sc->rl_ldata.rl_tx_list_addr, BUS_DMA_NOWAIT);
1081 	if (error != 0 || sc->rl_ldata.rl_tx_list_addr == 0) {
1082 		device_printf(dev, "could not load TX DMA ring\n");
1083 		return (ENOMEM);
1084 	}
1085 
1086 	/* Create DMA maps for TX buffers */
1087 
1088 	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
1089 		error = bus_dmamap_create(sc->rl_ldata.rl_tx_mtag, 0,
1090 		    &sc->rl_ldata.rl_tx_desc[i].tx_dmamap);
1091 		if (error) {
1092 			device_printf(dev, "could not create DMA map for TX\n");
1093 			return (error);
1094 		}
1095 	}
1096 
1097 	/*
1098 	 * Allocate map for RX descriptor list.
1099 	 */
1100 	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
1101 	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1102 	    NULL, rx_list_size, 1, rx_list_size, 0,
1103 	    NULL, NULL, &sc->rl_ldata.rl_rx_list_tag);
1104 	if (error) {
1105 		device_printf(dev, "could not create RX DMA ring tag\n");
1106 		return (error);
1107 	}
1108 
1109 	/* Allocate DMA'able memory for the RX ring */
1110 
1111 	error = bus_dmamem_alloc(sc->rl_ldata.rl_rx_list_tag,
1112 	    (void **)&sc->rl_ldata.rl_rx_list,
1113 	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1114 	    &sc->rl_ldata.rl_rx_list_map);
1115 	if (error) {
1116 		device_printf(dev, "could not allocate RX DMA ring\n");
1117 		return (error);
1118 	}
1119 
1120 	/* Load the map for the RX ring. */
1121 
1122 	sc->rl_ldata.rl_rx_list_addr = 0;
1123 	error = bus_dmamap_load(sc->rl_ldata.rl_rx_list_tag,
1124 	     sc->rl_ldata.rl_rx_list_map, sc->rl_ldata.rl_rx_list,
1125 	     rx_list_size, re_dma_map_addr,
1126 	     &sc->rl_ldata.rl_rx_list_addr, BUS_DMA_NOWAIT);
1127 	if (error != 0 || sc->rl_ldata.rl_rx_list_addr == 0) {
1128 		device_printf(dev, "could not load RX DMA ring\n");
1129 		return (ENOMEM);
1130 	}
1131 
1132 	/* Create DMA maps for RX buffers */
1133 
1134 	if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
1135 		error = bus_dmamap_create(sc->rl_ldata.rl_jrx_mtag, 0,
1136 		    &sc->rl_ldata.rl_jrx_sparemap);
1137 		if (error) {
1138 			device_printf(dev,
1139 			    "could not create spare DMA map for jumbo RX\n");
1140 			return (error);
1141 		}
1142 		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1143 			error = bus_dmamap_create(sc->rl_ldata.rl_jrx_mtag, 0,
1144 			    &sc->rl_ldata.rl_jrx_desc[i].rx_dmamap);
1145 			if (error) {
1146 				device_printf(dev,
1147 				    "could not create DMA map for jumbo RX\n");
1148 				return (error);
1149 			}
1150 		}
1151 	}
1152 	error = bus_dmamap_create(sc->rl_ldata.rl_rx_mtag, 0,
1153 	    &sc->rl_ldata.rl_rx_sparemap);
1154 	if (error) {
1155 		device_printf(dev, "could not create spare DMA map for RX\n");
1156 		return (error);
1157 	}
1158 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1159 		error = bus_dmamap_create(sc->rl_ldata.rl_rx_mtag, 0,
1160 		    &sc->rl_ldata.rl_rx_desc[i].rx_dmamap);
1161 		if (error) {
1162 			device_printf(dev, "could not create DMA map for RX\n");
1163 			return (error);
1164 		}
1165 	}
1166 
1167 	/* Create DMA map for statistics. */
1168 	error = bus_dma_tag_create(sc->rl_parent_tag, RL_DUMP_ALIGN, 0,
1169 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
1170 	    sizeof(struct rl_stats), 1, sizeof(struct rl_stats), 0, NULL, NULL,
1171 	    &sc->rl_ldata.rl_stag);
1172 	if (error) {
1173 		device_printf(dev, "could not create statistics DMA tag\n");
1174 		return (error);
1175 	}
1176 	/* Allocate DMA'able memory for statistics. */
1177 	error = bus_dmamem_alloc(sc->rl_ldata.rl_stag,
1178 	    (void **)&sc->rl_ldata.rl_stats,
1179 	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1180 	    &sc->rl_ldata.rl_smap);
1181 	if (error) {
1182 		device_printf(dev,
1183 		    "could not allocate statistics DMA memory\n");
1184 		return (error);
1185 	}
1186 	/* Load the map for statistics. */
1187 	sc->rl_ldata.rl_stats_addr = 0;
1188 	error = bus_dmamap_load(sc->rl_ldata.rl_stag, sc->rl_ldata.rl_smap,
1189 	    sc->rl_ldata.rl_stats, sizeof(struct rl_stats), re_dma_map_addr,
1190 	     &sc->rl_ldata.rl_stats_addr, BUS_DMA_NOWAIT);
1191 	if (error != 0 || sc->rl_ldata.rl_stats_addr == 0) {
1192 		device_printf(dev, "could not load statistics DMA memory\n");
1193 		return (ENOMEM);
1194 	}
1195 
1196 	return (0);
1197 }
1198 
1199 /*
1200  * Attach the interface. Allocate softc structures, do ifmedia
1201  * setup and ethernet/BPF attach.
1202  */
1203 static int
1204 re_attach(device_t dev)
1205 {
1206 	u_char			eaddr[ETHER_ADDR_LEN];
1207 	u_int16_t		as[ETHER_ADDR_LEN / 2];
1208 	struct rl_softc		*sc;
1209 	struct ifnet		*ifp;
1210 	const struct rl_hwrev	*hw_rev;
1211 	int			capmask, error = 0, hwrev, i, msic, msixc,
1212 				phy, reg, rid;
1213 	u_int32_t		cap, ctl;
1214 	u_int16_t		devid, re_did = 0;
1215 	uint8_t			cfg;
1216 
1217 	sc = device_get_softc(dev);
1218 	sc->rl_dev = dev;
1219 
1220 	mtx_init(&sc->rl_mtx, device_get_nameunit(dev), MTX_NETWORK_LOCK,
1221 	    MTX_DEF);
1222 	callout_init_mtx(&sc->rl_stat_callout, &sc->rl_mtx, 0);
1223 
1224 	/*
1225 	 * Map control/status registers.
1226 	 */
1227 	pci_enable_busmaster(dev);
1228 
1229 	devid = pci_get_device(dev);
1230 	/*
1231 	 * Prefer memory space register mapping over IO space.
1232 	 * Because RTL8169SC does not seem to work when memory mapping
1233 	 * is used always activate io mapping.
1234 	 */
1235 	if (devid == RT_DEVICEID_8169SC)
1236 		prefer_iomap = 1;
1237 	if (prefer_iomap == 0) {
1238 		sc->rl_res_id = PCIR_BAR(1);
1239 		sc->rl_res_type = SYS_RES_MEMORY;
1240 		/* RTL8168/8101E seems to use different BARs. */
1241 		if (devid == RT_DEVICEID_8168 || devid == RT_DEVICEID_8101E)
1242 			sc->rl_res_id = PCIR_BAR(2);
1243 	} else {
1244 		sc->rl_res_id = PCIR_BAR(0);
1245 		sc->rl_res_type = SYS_RES_IOPORT;
1246 	}
1247 	sc->rl_res = bus_alloc_resource_any(dev, sc->rl_res_type,
1248 	    &sc->rl_res_id, RF_ACTIVE);
1249 	if (sc->rl_res == NULL && prefer_iomap == 0) {
1250 		sc->rl_res_id = PCIR_BAR(0);
1251 		sc->rl_res_type = SYS_RES_IOPORT;
1252 		sc->rl_res = bus_alloc_resource_any(dev, sc->rl_res_type,
1253 		    &sc->rl_res_id, RF_ACTIVE);
1254 	}
1255 	if (sc->rl_res == NULL) {
1256 		device_printf(dev, "couldn't map ports/memory\n");
1257 		error = ENXIO;
1258 		goto fail;
1259 	}
1260 
1261 	sc->rl_btag = rman_get_bustag(sc->rl_res);
1262 	sc->rl_bhandle = rman_get_bushandle(sc->rl_res);
1263 
1264 	msic = pci_msi_count(dev);
1265 	msixc = pci_msix_count(dev);
1266 	if (pci_find_cap(dev, PCIY_EXPRESS, &reg) == 0) {
1267 		sc->rl_flags |= RL_FLAG_PCIE;
1268 		sc->rl_expcap = reg;
1269 	}
1270 	if (bootverbose) {
1271 		device_printf(dev, "MSI count : %d\n", msic);
1272 		device_printf(dev, "MSI-X count : %d\n", msixc);
1273 	}
1274 	if (msix_disable > 0)
1275 		msixc = 0;
1276 	if (msi_disable > 0)
1277 		msic = 0;
1278 	/* Prefer MSI-X to MSI. */
1279 	if (msixc > 0) {
1280 		msixc = RL_MSI_MESSAGES;
1281 		rid = PCIR_BAR(4);
1282 		sc->rl_res_pba = bus_alloc_resource_any(dev, SYS_RES_MEMORY,
1283 		    &rid, RF_ACTIVE);
1284 		if (sc->rl_res_pba == NULL) {
1285 			device_printf(sc->rl_dev,
1286 			    "could not allocate MSI-X PBA resource\n");
1287 		}
1288 		if (sc->rl_res_pba != NULL &&
1289 		    pci_alloc_msix(dev, &msixc) == 0) {
1290 			if (msixc == RL_MSI_MESSAGES) {
1291 				device_printf(dev, "Using %d MSI-X message\n",
1292 				    msixc);
1293 				sc->rl_flags |= RL_FLAG_MSIX;
1294 			} else
1295 				pci_release_msi(dev);
1296 		}
1297 		if ((sc->rl_flags & RL_FLAG_MSIX) == 0) {
1298 			if (sc->rl_res_pba != NULL)
1299 				bus_release_resource(dev, SYS_RES_MEMORY, rid,
1300 				    sc->rl_res_pba);
1301 			sc->rl_res_pba = NULL;
1302 			msixc = 0;
1303 		}
1304 	}
1305 	/* Prefer MSI to INTx. */
1306 	if (msixc == 0 && msic > 0) {
1307 		msic = RL_MSI_MESSAGES;
1308 		if (pci_alloc_msi(dev, &msic) == 0) {
1309 			if (msic == RL_MSI_MESSAGES) {
1310 				device_printf(dev, "Using %d MSI message\n",
1311 				    msic);
1312 				sc->rl_flags |= RL_FLAG_MSI;
1313 				/* Explicitly set MSI enable bit. */
1314 				CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1315 				cfg = CSR_READ_1(sc, RL_CFG2);
1316 				cfg |= RL_CFG2_MSI;
1317 				CSR_WRITE_1(sc, RL_CFG2, cfg);
1318 				CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1319 			} else
1320 				pci_release_msi(dev);
1321 		}
1322 		if ((sc->rl_flags & RL_FLAG_MSI) == 0)
1323 			msic = 0;
1324 	}
1325 
1326 	/* Allocate interrupt */
1327 	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) == 0) {
1328 		rid = 0;
1329 		sc->rl_irq[0] = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
1330 		    RF_SHAREABLE | RF_ACTIVE);
1331 		if (sc->rl_irq[0] == NULL) {
1332 			device_printf(dev, "couldn't allocate IRQ resources\n");
1333 			error = ENXIO;
1334 			goto fail;
1335 		}
1336 	} else {
1337 		for (i = 0, rid = 1; i < RL_MSI_MESSAGES; i++, rid++) {
1338 			sc->rl_irq[i] = bus_alloc_resource_any(dev,
1339 			    SYS_RES_IRQ, &rid, RF_ACTIVE);
1340 			if (sc->rl_irq[i] == NULL) {
1341 				device_printf(dev,
1342 				    "couldn't allocate IRQ resources for "
1343 				    "message %d\n", rid);
1344 				error = ENXIO;
1345 				goto fail;
1346 			}
1347 		}
1348 	}
1349 
1350 	if ((sc->rl_flags & RL_FLAG_MSI) == 0) {
1351 		CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1352 		cfg = CSR_READ_1(sc, RL_CFG2);
1353 		if ((cfg & RL_CFG2_MSI) != 0) {
1354 			device_printf(dev, "turning off MSI enable bit.\n");
1355 			cfg &= ~RL_CFG2_MSI;
1356 			CSR_WRITE_1(sc, RL_CFG2, cfg);
1357 		}
1358 		CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1359 	}
1360 
1361 	/* Disable ASPM L0S/L1 and CLKREQ. */
1362 	if (sc->rl_expcap != 0) {
1363 		cap = pci_read_config(dev, sc->rl_expcap +
1364 		    PCIER_LINK_CAP, 2);
1365 		if ((cap & PCIEM_LINK_CAP_ASPM) != 0) {
1366 			ctl = pci_read_config(dev, sc->rl_expcap +
1367 			    PCIER_LINK_CTL, 2);
1368 			if ((ctl & (PCIEM_LINK_CTL_ECPM |
1369 			    PCIEM_LINK_CTL_ASPMC))!= 0) {
1370 				ctl &= ~(PCIEM_LINK_CTL_ECPM |
1371 				    PCIEM_LINK_CTL_ASPMC);
1372 				pci_write_config(dev, sc->rl_expcap +
1373 				    PCIER_LINK_CTL, ctl, 2);
1374 				device_printf(dev, "ASPM disabled\n");
1375 			}
1376 		} else
1377 			device_printf(dev, "no ASPM capability\n");
1378 	}
1379 
1380 	hw_rev = re_hwrevs;
1381 	hwrev = CSR_READ_4(sc, RL_TXCFG);
1382 	switch (hwrev & 0x70000000) {
1383 	case 0x00000000:
1384 	case 0x10000000:
1385 		device_printf(dev, "Chip rev. 0x%08x\n", hwrev & 0xfc800000);
1386 		hwrev &= (RL_TXCFG_HWREV | 0x80000000);
1387 		break;
1388 	default:
1389 		device_printf(dev, "Chip rev. 0x%08x\n", hwrev & 0x7c800000);
1390 		sc->rl_macrev = hwrev & 0x00700000;
1391 		hwrev &= RL_TXCFG_HWREV;
1392 		break;
1393 	}
1394 	device_printf(dev, "MAC rev. 0x%08x\n", sc->rl_macrev);
1395 	while (hw_rev->rl_desc != NULL) {
1396 		if (hw_rev->rl_rev == hwrev) {
1397 			sc->rl_type = hw_rev->rl_type;
1398 			sc->rl_hwrev = hw_rev;
1399 			break;
1400 		}
1401 		hw_rev++;
1402 	}
1403 	if (hw_rev->rl_desc == NULL) {
1404 		device_printf(dev, "Unknown H/W revision: 0x%08x\n", hwrev);
1405 		error = ENXIO;
1406 		goto fail;
1407 	}
1408 
1409 	switch (hw_rev->rl_rev) {
1410 	case RL_HWREV_8139CPLUS:
1411 		sc->rl_flags |= RL_FLAG_FASTETHER | RL_FLAG_AUTOPAD;
1412 		break;
1413 	case RL_HWREV_8100E:
1414 	case RL_HWREV_8101E:
1415 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_FASTETHER;
1416 		break;
1417 	case RL_HWREV_8102E:
1418 	case RL_HWREV_8102EL:
1419 	case RL_HWREV_8102EL_SPIN1:
1420 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR | RL_FLAG_DESCV2 |
1421 		    RL_FLAG_MACSTAT | RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP |
1422 		    RL_FLAG_AUTOPAD;
1423 		break;
1424 	case RL_HWREV_8103E:
1425 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR | RL_FLAG_DESCV2 |
1426 		    RL_FLAG_MACSTAT | RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP |
1427 		    RL_FLAG_AUTOPAD | RL_FLAG_MACSLEEP;
1428 		break;
1429 	case RL_HWREV_8401E:
1430 	case RL_HWREV_8105E:
1431 	case RL_HWREV_8105E_SPIN1:
1432 	case RL_HWREV_8106E:
1433 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
1434 		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
1435 		    RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD;
1436 		break;
1437 	case RL_HWREV_8402:
1438 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
1439 		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
1440 		    RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD |
1441 		    RL_FLAG_CMDSTOP_WAIT_TXQ;
1442 		break;
1443 	case RL_HWREV_8168B_SPIN1:
1444 	case RL_HWREV_8168B_SPIN2:
1445 		sc->rl_flags |= RL_FLAG_WOLRXENB;
1446 		/* FALLTHROUGH */
1447 	case RL_HWREV_8168B_SPIN3:
1448 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_MACSTAT;
1449 		break;
1450 	case RL_HWREV_8168C_SPIN2:
1451 		sc->rl_flags |= RL_FLAG_MACSLEEP;
1452 		/* FALLTHROUGH */
1453 	case RL_HWREV_8168C:
1454 		if (sc->rl_macrev == 0x00200000)
1455 			sc->rl_flags |= RL_FLAG_MACSLEEP;
1456 		/* FALLTHROUGH */
1457 	case RL_HWREV_8168CP:
1458 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
1459 		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP |
1460 		    RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 | RL_FLAG_WOL_MANLINK;
1461 		break;
1462 	case RL_HWREV_8168D:
1463 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
1464 		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
1465 		    RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 |
1466 		    RL_FLAG_WOL_MANLINK;
1467 		break;
1468 	case RL_HWREV_8168DP:
1469 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
1470 		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_AUTOPAD |
1471 		    RL_FLAG_JUMBOV2 | RL_FLAG_WAIT_TXPOLL | RL_FLAG_WOL_MANLINK;
1472 		break;
1473 	case RL_HWREV_8168E:
1474 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
1475 		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
1476 		    RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 |
1477 		    RL_FLAG_WOL_MANLINK;
1478 		break;
1479 	case RL_HWREV_8168E_VL:
1480 	case RL_HWREV_8168F:
1481 		sc->rl_flags |= RL_FLAG_EARLYOFF;
1482 		/* FALLTHROUGH */
1483 	case RL_HWREV_8411:
1484 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
1485 		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP |
1486 		    RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 |
1487 		    RL_FLAG_CMDSTOP_WAIT_TXQ | RL_FLAG_WOL_MANLINK;
1488 		break;
1489 	case RL_HWREV_8168EP:
1490 	case RL_HWREV_8168G:
1491 	case RL_HWREV_8411B:
1492 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
1493 		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP |
1494 		    RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 |
1495 		    RL_FLAG_CMDSTOP_WAIT_TXQ | RL_FLAG_WOL_MANLINK |
1496 		    RL_FLAG_8168G_PLUS;
1497 		break;
1498 	case RL_HWREV_8168GU:
1499 	case RL_HWREV_8168H:
1500 		if (pci_get_device(dev) == RT_DEVICEID_8101E) {
1501 			/* RTL8106E(US), RTL8107E */
1502 			sc->rl_flags |= RL_FLAG_FASTETHER;
1503 		} else
1504 			sc->rl_flags |= RL_FLAG_JUMBOV2 | RL_FLAG_WOL_MANLINK;
1505 
1506 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
1507 		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP |
1508 		    RL_FLAG_AUTOPAD | RL_FLAG_CMDSTOP_WAIT_TXQ |
1509 		    RL_FLAG_8168G_PLUS;
1510 		break;
1511 	case RL_HWREV_8169_8110SB:
1512 	case RL_HWREV_8169_8110SBL:
1513 	case RL_HWREV_8169_8110SC:
1514 	case RL_HWREV_8169_8110SCE:
1515 		sc->rl_flags |= RL_FLAG_PHYWAKE;
1516 		/* FALLTHROUGH */
1517 	case RL_HWREV_8169:
1518 	case RL_HWREV_8169S:
1519 	case RL_HWREV_8110S:
1520 		sc->rl_flags |= RL_FLAG_MACRESET;
1521 		break;
1522 	default:
1523 		break;
1524 	}
1525 
1526 	if (sc->rl_hwrev->rl_rev == RL_HWREV_8139CPLUS) {
1527 		sc->rl_cfg0 = RL_8139_CFG0;
1528 		sc->rl_cfg1 = RL_8139_CFG1;
1529 		sc->rl_cfg2 = 0;
1530 		sc->rl_cfg3 = RL_8139_CFG3;
1531 		sc->rl_cfg4 = RL_8139_CFG4;
1532 		sc->rl_cfg5 = RL_8139_CFG5;
1533 	} else {
1534 		sc->rl_cfg0 = RL_CFG0;
1535 		sc->rl_cfg1 = RL_CFG1;
1536 		sc->rl_cfg2 = RL_CFG2;
1537 		sc->rl_cfg3 = RL_CFG3;
1538 		sc->rl_cfg4 = RL_CFG4;
1539 		sc->rl_cfg5 = RL_CFG5;
1540 	}
1541 
1542 	/* Reset the adapter. */
1543 	RL_LOCK(sc);
1544 	re_reset(sc);
1545 	RL_UNLOCK(sc);
1546 
1547 	/* Enable PME. */
1548 	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1549 	cfg = CSR_READ_1(sc, sc->rl_cfg1);
1550 	cfg |= RL_CFG1_PME;
1551 	CSR_WRITE_1(sc, sc->rl_cfg1, cfg);
1552 	cfg = CSR_READ_1(sc, sc->rl_cfg5);
1553 	cfg &= RL_CFG5_PME_STS;
1554 	CSR_WRITE_1(sc, sc->rl_cfg5, cfg);
1555 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1556 
1557 	if ((sc->rl_flags & RL_FLAG_PAR) != 0) {
1558 		/*
1559 		 * XXX Should have a better way to extract station
1560 		 * address from EEPROM.
1561 		 */
1562 		for (i = 0; i < ETHER_ADDR_LEN; i++)
1563 			eaddr[i] = CSR_READ_1(sc, RL_IDR0 + i);
1564 	} else {
1565 		sc->rl_eewidth = RL_9356_ADDR_LEN;
1566 		re_read_eeprom(sc, (caddr_t)&re_did, 0, 1);
1567 		if (re_did != 0x8129)
1568 			sc->rl_eewidth = RL_9346_ADDR_LEN;
1569 
1570 		/*
1571 		 * Get station address from the EEPROM.
1572 		 */
1573 		re_read_eeprom(sc, (caddr_t)as, RL_EE_EADDR, 3);
1574 		for (i = 0; i < ETHER_ADDR_LEN / 2; i++)
1575 			as[i] = le16toh(as[i]);
1576 		bcopy(as, eaddr, ETHER_ADDR_LEN);
1577 	}
1578 
1579 	if (sc->rl_type == RL_8169) {
1580 		/* Set RX length mask and number of descriptors. */
1581 		sc->rl_rxlenmask = RL_RDESC_STAT_GFRAGLEN;
1582 		sc->rl_txstart = RL_GTXSTART;
1583 		sc->rl_ldata.rl_tx_desc_cnt = RL_8169_TX_DESC_CNT;
1584 		sc->rl_ldata.rl_rx_desc_cnt = RL_8169_RX_DESC_CNT;
1585 	} else {
1586 		/* Set RX length mask and number of descriptors. */
1587 		sc->rl_rxlenmask = RL_RDESC_STAT_FRAGLEN;
1588 		sc->rl_txstart = RL_TXSTART;
1589 		sc->rl_ldata.rl_tx_desc_cnt = RL_8139_TX_DESC_CNT;
1590 		sc->rl_ldata.rl_rx_desc_cnt = RL_8139_RX_DESC_CNT;
1591 	}
1592 
1593 	error = re_allocmem(dev, sc);
1594 	if (error)
1595 		goto fail;
1596 	re_add_sysctls(sc);
1597 
1598 	ifp = sc->rl_ifp = if_alloc(IFT_ETHER);
1599 	if (ifp == NULL) {
1600 		device_printf(dev, "can not if_alloc()\n");
1601 		error = ENOSPC;
1602 		goto fail;
1603 	}
1604 
1605 	/* Take controller out of deep sleep mode. */
1606 	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
1607 		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
1608 			CSR_WRITE_1(sc, RL_GPIO,
1609 			    CSR_READ_1(sc, RL_GPIO) | 0x01);
1610 		else
1611 			CSR_WRITE_1(sc, RL_GPIO,
1612 			    CSR_READ_1(sc, RL_GPIO) & ~0x01);
1613 	}
1614 
1615 	/* Take PHY out of power down mode. */
1616 	if ((sc->rl_flags & RL_FLAG_PHYWAKE_PM) != 0) {
1617 		CSR_WRITE_1(sc, RL_PMCH, CSR_READ_1(sc, RL_PMCH) | 0x80);
1618 		if (hw_rev->rl_rev == RL_HWREV_8401E)
1619 			CSR_WRITE_1(sc, 0xD1, CSR_READ_1(sc, 0xD1) & ~0x08);
1620 	}
1621 	if ((sc->rl_flags & RL_FLAG_PHYWAKE) != 0) {
1622 		re_gmii_writereg(dev, 1, 0x1f, 0);
1623 		re_gmii_writereg(dev, 1, 0x0e, 0);
1624 	}
1625 
1626 	ifp->if_softc = sc;
1627 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1628 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1629 	ifp->if_ioctl = re_ioctl;
1630 	ifp->if_start = re_start;
1631 	/*
1632 	 * RTL8168/8111C generates wrong IP checksummed frame if the
1633 	 * packet has IP options so disable TX checksum offloading.
1634 	 */
1635 	if (sc->rl_hwrev->rl_rev == RL_HWREV_8168C ||
1636 	    sc->rl_hwrev->rl_rev == RL_HWREV_8168C_SPIN2 ||
1637 	    sc->rl_hwrev->rl_rev == RL_HWREV_8168CP) {
1638 		ifp->if_hwassist = 0;
1639 		ifp->if_capabilities = IFCAP_RXCSUM | IFCAP_TSO4;
1640 	} else {
1641 		ifp->if_hwassist = CSUM_IP | CSUM_TCP | CSUM_UDP;
1642 		ifp->if_capabilities = IFCAP_HWCSUM | IFCAP_TSO4;
1643 	}
1644 	ifp->if_hwassist |= CSUM_TSO;
1645 	ifp->if_capenable = ifp->if_capabilities;
1646 	ifp->if_init = re_init;
1647 	IFQ_SET_MAXLEN(&ifp->if_snd, RL_IFQ_MAXLEN);
1648 	ifp->if_snd.ifq_drv_maxlen = RL_IFQ_MAXLEN;
1649 	IFQ_SET_READY(&ifp->if_snd);
1650 
1651 	TASK_INIT(&sc->rl_inttask, 0, re_int_task, sc);
1652 
1653 #define	RE_PHYAD_INTERNAL	 0
1654 
1655 	/* Do MII setup. */
1656 	phy = RE_PHYAD_INTERNAL;
1657 	if (sc->rl_type == RL_8169)
1658 		phy = 1;
1659 	capmask = BMSR_DEFCAPMASK;
1660 	if ((sc->rl_flags & RL_FLAG_FASTETHER) != 0)
1661 		 capmask &= ~BMSR_EXTSTAT;
1662 	error = mii_attach(dev, &sc->rl_miibus, ifp, re_ifmedia_upd,
1663 	    re_ifmedia_sts, capmask, phy, MII_OFFSET_ANY, MIIF_DOPAUSE);
1664 	if (error != 0) {
1665 		device_printf(dev, "attaching PHYs failed\n");
1666 		goto fail;
1667 	}
1668 
1669 	/*
1670 	 * Call MI attach routine.
1671 	 */
1672 	ether_ifattach(ifp, eaddr);
1673 
1674 	/* VLAN capability setup */
1675 	ifp->if_capabilities |= IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING;
1676 	if (ifp->if_capabilities & IFCAP_HWCSUM)
1677 		ifp->if_capabilities |= IFCAP_VLAN_HWCSUM;
1678 	/* Enable WOL if PM is supported. */
1679 	if (pci_find_cap(sc->rl_dev, PCIY_PMG, &reg) == 0)
1680 		ifp->if_capabilities |= IFCAP_WOL;
1681 	ifp->if_capenable = ifp->if_capabilities;
1682 	ifp->if_capenable &= ~(IFCAP_WOL_UCAST | IFCAP_WOL_MCAST);
1683 	/*
1684 	 * Don't enable TSO by default.  It is known to generate
1685 	 * corrupted TCP segments(bad TCP options) under certain
1686 	 * circumstances.
1687 	 */
1688 	ifp->if_hwassist &= ~CSUM_TSO;
1689 	ifp->if_capenable &= ~(IFCAP_TSO4 | IFCAP_VLAN_HWTSO);
1690 #ifdef DEVICE_POLLING
1691 	ifp->if_capabilities |= IFCAP_POLLING;
1692 #endif
1693 	/*
1694 	 * Tell the upper layer(s) we support long frames.
1695 	 * Must appear after the call to ether_ifattach() because
1696 	 * ether_ifattach() sets ifi_hdrlen to the default value.
1697 	 */
1698 	ifp->if_hdrlen = sizeof(struct ether_vlan_header);
1699 
1700 #ifdef DEV_NETMAP
1701 	re_netmap_attach(sc);
1702 #endif /* DEV_NETMAP */
1703 
1704 #ifdef RE_DIAG
1705 	/*
1706 	 * Perform hardware diagnostic on the original RTL8169.
1707 	 * Some 32-bit cards were incorrectly wired and would
1708 	 * malfunction if plugged into a 64-bit slot.
1709 	 */
1710 	if (hwrev == RL_HWREV_8169) {
1711 		error = re_diag(sc);
1712 		if (error) {
1713 			device_printf(dev,
1714 		    	"attach aborted due to hardware diag failure\n");
1715 			ether_ifdetach(ifp);
1716 			goto fail;
1717 		}
1718 	}
1719 #endif
1720 
1721 #ifdef RE_TX_MODERATION
1722 	intr_filter = 1;
1723 #endif
1724 	/* Hook interrupt last to avoid having to lock softc */
1725 	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) != 0 &&
1726 	    intr_filter == 0) {
1727 		error = bus_setup_intr(dev, sc->rl_irq[0],
1728 		    INTR_TYPE_NET | INTR_MPSAFE, NULL, re_intr_msi, sc,
1729 		    &sc->rl_intrhand[0]);
1730 	} else {
1731 		error = bus_setup_intr(dev, sc->rl_irq[0],
1732 		    INTR_TYPE_NET | INTR_MPSAFE, re_intr, NULL, sc,
1733 		    &sc->rl_intrhand[0]);
1734 	}
1735 	if (error) {
1736 		device_printf(dev, "couldn't set up irq\n");
1737 		ether_ifdetach(ifp);
1738 	}
1739 
1740 fail:
1741 	if (error)
1742 		re_detach(dev);
1743 
1744 	return (error);
1745 }
1746 
1747 /*
1748  * Shutdown hardware and free up resources. This can be called any
1749  * time after the mutex has been initialized. It is called in both
1750  * the error case in attach and the normal detach case so it needs
1751  * to be careful about only freeing resources that have actually been
1752  * allocated.
1753  */
1754 static int
1755 re_detach(device_t dev)
1756 {
1757 	struct rl_softc		*sc;
1758 	struct ifnet		*ifp;
1759 	int			i, rid;
1760 
1761 	sc = device_get_softc(dev);
1762 	ifp = sc->rl_ifp;
1763 	KASSERT(mtx_initialized(&sc->rl_mtx), ("re mutex not initialized"));
1764 
1765 	/* These should only be active if attach succeeded */
1766 	if (device_is_attached(dev)) {
1767 #ifdef DEVICE_POLLING
1768 		if (ifp->if_capenable & IFCAP_POLLING)
1769 			ether_poll_deregister(ifp);
1770 #endif
1771 		RL_LOCK(sc);
1772 #if 0
1773 		sc->suspended = 1;
1774 #endif
1775 		re_stop(sc);
1776 		RL_UNLOCK(sc);
1777 		callout_drain(&sc->rl_stat_callout);
1778 		taskqueue_drain(taskqueue_fast, &sc->rl_inttask);
1779 		/*
1780 		 * Force off the IFF_UP flag here, in case someone
1781 		 * still had a BPF descriptor attached to this
1782 		 * interface. If they do, ether_ifdetach() will cause
1783 		 * the BPF code to try and clear the promisc mode
1784 		 * flag, which will bubble down to re_ioctl(),
1785 		 * which will try to call re_init() again. This will
1786 		 * turn the NIC back on and restart the MII ticker,
1787 		 * which will panic the system when the kernel tries
1788 		 * to invoke the re_tick() function that isn't there
1789 		 * anymore.
1790 		 */
1791 		ifp->if_flags &= ~IFF_UP;
1792 		ether_ifdetach(ifp);
1793 	}
1794 	if (sc->rl_miibus)
1795 		device_delete_child(dev, sc->rl_miibus);
1796 	bus_generic_detach(dev);
1797 
1798 	/*
1799 	 * The rest is resource deallocation, so we should already be
1800 	 * stopped here.
1801 	 */
1802 
1803 	if (sc->rl_intrhand[0] != NULL) {
1804 		bus_teardown_intr(dev, sc->rl_irq[0], sc->rl_intrhand[0]);
1805 		sc->rl_intrhand[0] = NULL;
1806 	}
1807 	if (ifp != NULL) {
1808 #ifdef DEV_NETMAP
1809 		netmap_detach(ifp);
1810 #endif /* DEV_NETMAP */
1811 		if_free(ifp);
1812 	}
1813 	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) == 0)
1814 		rid = 0;
1815 	else
1816 		rid = 1;
1817 	if (sc->rl_irq[0] != NULL) {
1818 		bus_release_resource(dev, SYS_RES_IRQ, rid, sc->rl_irq[0]);
1819 		sc->rl_irq[0] = NULL;
1820 	}
1821 	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) != 0)
1822 		pci_release_msi(dev);
1823 	if (sc->rl_res_pba) {
1824 		rid = PCIR_BAR(4);
1825 		bus_release_resource(dev, SYS_RES_MEMORY, rid, sc->rl_res_pba);
1826 	}
1827 	if (sc->rl_res)
1828 		bus_release_resource(dev, sc->rl_res_type, sc->rl_res_id,
1829 		    sc->rl_res);
1830 
1831 	/* Unload and free the RX DMA ring memory and map */
1832 
1833 	if (sc->rl_ldata.rl_rx_list_tag) {
1834 		if (sc->rl_ldata.rl_rx_list_addr)
1835 			bus_dmamap_unload(sc->rl_ldata.rl_rx_list_tag,
1836 			    sc->rl_ldata.rl_rx_list_map);
1837 		if (sc->rl_ldata.rl_rx_list)
1838 			bus_dmamem_free(sc->rl_ldata.rl_rx_list_tag,
1839 			    sc->rl_ldata.rl_rx_list,
1840 			    sc->rl_ldata.rl_rx_list_map);
1841 		bus_dma_tag_destroy(sc->rl_ldata.rl_rx_list_tag);
1842 	}
1843 
1844 	/* Unload and free the TX DMA ring memory and map */
1845 
1846 	if (sc->rl_ldata.rl_tx_list_tag) {
1847 		if (sc->rl_ldata.rl_tx_list_addr)
1848 			bus_dmamap_unload(sc->rl_ldata.rl_tx_list_tag,
1849 			    sc->rl_ldata.rl_tx_list_map);
1850 		if (sc->rl_ldata.rl_tx_list)
1851 			bus_dmamem_free(sc->rl_ldata.rl_tx_list_tag,
1852 			    sc->rl_ldata.rl_tx_list,
1853 			    sc->rl_ldata.rl_tx_list_map);
1854 		bus_dma_tag_destroy(sc->rl_ldata.rl_tx_list_tag);
1855 	}
1856 
1857 	/* Destroy all the RX and TX buffer maps */
1858 
1859 	if (sc->rl_ldata.rl_tx_mtag) {
1860 		for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
1861 			if (sc->rl_ldata.rl_tx_desc[i].tx_dmamap)
1862 				bus_dmamap_destroy(sc->rl_ldata.rl_tx_mtag,
1863 				    sc->rl_ldata.rl_tx_desc[i].tx_dmamap);
1864 		}
1865 		bus_dma_tag_destroy(sc->rl_ldata.rl_tx_mtag);
1866 	}
1867 	if (sc->rl_ldata.rl_rx_mtag) {
1868 		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1869 			if (sc->rl_ldata.rl_rx_desc[i].rx_dmamap)
1870 				bus_dmamap_destroy(sc->rl_ldata.rl_rx_mtag,
1871 				    sc->rl_ldata.rl_rx_desc[i].rx_dmamap);
1872 		}
1873 		if (sc->rl_ldata.rl_rx_sparemap)
1874 			bus_dmamap_destroy(sc->rl_ldata.rl_rx_mtag,
1875 			    sc->rl_ldata.rl_rx_sparemap);
1876 		bus_dma_tag_destroy(sc->rl_ldata.rl_rx_mtag);
1877 	}
1878 	if (sc->rl_ldata.rl_jrx_mtag) {
1879 		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1880 			if (sc->rl_ldata.rl_jrx_desc[i].rx_dmamap)
1881 				bus_dmamap_destroy(sc->rl_ldata.rl_jrx_mtag,
1882 				    sc->rl_ldata.rl_jrx_desc[i].rx_dmamap);
1883 		}
1884 		if (sc->rl_ldata.rl_jrx_sparemap)
1885 			bus_dmamap_destroy(sc->rl_ldata.rl_jrx_mtag,
1886 			    sc->rl_ldata.rl_jrx_sparemap);
1887 		bus_dma_tag_destroy(sc->rl_ldata.rl_jrx_mtag);
1888 	}
1889 	/* Unload and free the stats buffer and map */
1890 
1891 	if (sc->rl_ldata.rl_stag) {
1892 		if (sc->rl_ldata.rl_stats_addr)
1893 			bus_dmamap_unload(sc->rl_ldata.rl_stag,
1894 			    sc->rl_ldata.rl_smap);
1895 		if (sc->rl_ldata.rl_stats)
1896 			bus_dmamem_free(sc->rl_ldata.rl_stag,
1897 			    sc->rl_ldata.rl_stats, sc->rl_ldata.rl_smap);
1898 		bus_dma_tag_destroy(sc->rl_ldata.rl_stag);
1899 	}
1900 
1901 	if (sc->rl_parent_tag)
1902 		bus_dma_tag_destroy(sc->rl_parent_tag);
1903 
1904 	mtx_destroy(&sc->rl_mtx);
1905 
1906 	return (0);
1907 }
1908 
1909 static __inline void
1910 re_discard_rxbuf(struct rl_softc *sc, int idx)
1911 {
1912 	struct rl_desc		*desc;
1913 	struct rl_rxdesc	*rxd;
1914 	uint32_t		cmdstat;
1915 
1916 	if (sc->rl_ifp->if_mtu > RL_MTU &&
1917 	    (sc->rl_flags & RL_FLAG_JUMBOV2) != 0)
1918 		rxd = &sc->rl_ldata.rl_jrx_desc[idx];
1919 	else
1920 		rxd = &sc->rl_ldata.rl_rx_desc[idx];
1921 	desc = &sc->rl_ldata.rl_rx_list[idx];
1922 	desc->rl_vlanctl = 0;
1923 	cmdstat = rxd->rx_size;
1924 	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
1925 		cmdstat |= RL_RDESC_CMD_EOR;
1926 	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
1927 }
1928 
1929 static int
1930 re_newbuf(struct rl_softc *sc, int idx)
1931 {
1932 	struct mbuf		*m;
1933 	struct rl_rxdesc	*rxd;
1934 	bus_dma_segment_t	segs[1];
1935 	bus_dmamap_t		map;
1936 	struct rl_desc		*desc;
1937 	uint32_t		cmdstat;
1938 	int			error, nsegs;
1939 
1940 	m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
1941 	if (m == NULL)
1942 		return (ENOBUFS);
1943 
1944 	m->m_len = m->m_pkthdr.len = MCLBYTES;
1945 #ifdef RE_FIXUP_RX
1946 	/*
1947 	 * This is part of an evil trick to deal with non-x86 platforms.
1948 	 * The RealTek chip requires RX buffers to be aligned on 64-bit
1949 	 * boundaries, but that will hose non-x86 machines. To get around
1950 	 * this, we leave some empty space at the start of each buffer
1951 	 * and for non-x86 hosts, we copy the buffer back six bytes
1952 	 * to achieve word alignment. This is slightly more efficient
1953 	 * than allocating a new buffer, copying the contents, and
1954 	 * discarding the old buffer.
1955 	 */
1956 	m_adj(m, RE_ETHER_ALIGN);
1957 #endif
1958 	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_rx_mtag,
1959 	    sc->rl_ldata.rl_rx_sparemap, m, segs, &nsegs, BUS_DMA_NOWAIT);
1960 	if (error != 0) {
1961 		m_freem(m);
1962 		return (ENOBUFS);
1963 	}
1964 	KASSERT(nsegs == 1, ("%s: %d segment returned!", __func__, nsegs));
1965 
1966 	rxd = &sc->rl_ldata.rl_rx_desc[idx];
1967 	if (rxd->rx_m != NULL) {
1968 		bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap,
1969 		    BUS_DMASYNC_POSTREAD);
1970 		bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap);
1971 	}
1972 
1973 	rxd->rx_m = m;
1974 	map = rxd->rx_dmamap;
1975 	rxd->rx_dmamap = sc->rl_ldata.rl_rx_sparemap;
1976 	rxd->rx_size = segs[0].ds_len;
1977 	sc->rl_ldata.rl_rx_sparemap = map;
1978 	bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap,
1979 	    BUS_DMASYNC_PREREAD);
1980 
1981 	desc = &sc->rl_ldata.rl_rx_list[idx];
1982 	desc->rl_vlanctl = 0;
1983 	desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[0].ds_addr));
1984 	desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[0].ds_addr));
1985 	cmdstat = segs[0].ds_len;
1986 	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
1987 		cmdstat |= RL_RDESC_CMD_EOR;
1988 	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
1989 
1990 	return (0);
1991 }
1992 
1993 static int
1994 re_jumbo_newbuf(struct rl_softc *sc, int idx)
1995 {
1996 	struct mbuf		*m;
1997 	struct rl_rxdesc	*rxd;
1998 	bus_dma_segment_t	segs[1];
1999 	bus_dmamap_t		map;
2000 	struct rl_desc		*desc;
2001 	uint32_t		cmdstat;
2002 	int			error, nsegs;
2003 
2004 	m = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, MJUM9BYTES);
2005 	if (m == NULL)
2006 		return (ENOBUFS);
2007 	m->m_len = m->m_pkthdr.len = MJUM9BYTES;
2008 #ifdef RE_FIXUP_RX
2009 	m_adj(m, RE_ETHER_ALIGN);
2010 #endif
2011 	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_jrx_mtag,
2012 	    sc->rl_ldata.rl_jrx_sparemap, m, segs, &nsegs, BUS_DMA_NOWAIT);
2013 	if (error != 0) {
2014 		m_freem(m);
2015 		return (ENOBUFS);
2016 	}
2017 	KASSERT(nsegs == 1, ("%s: %d segment returned!", __func__, nsegs));
2018 
2019 	rxd = &sc->rl_ldata.rl_jrx_desc[idx];
2020 	if (rxd->rx_m != NULL) {
2021 		bus_dmamap_sync(sc->rl_ldata.rl_jrx_mtag, rxd->rx_dmamap,
2022 		    BUS_DMASYNC_POSTREAD);
2023 		bus_dmamap_unload(sc->rl_ldata.rl_jrx_mtag, rxd->rx_dmamap);
2024 	}
2025 
2026 	rxd->rx_m = m;
2027 	map = rxd->rx_dmamap;
2028 	rxd->rx_dmamap = sc->rl_ldata.rl_jrx_sparemap;
2029 	rxd->rx_size = segs[0].ds_len;
2030 	sc->rl_ldata.rl_jrx_sparemap = map;
2031 	bus_dmamap_sync(sc->rl_ldata.rl_jrx_mtag, rxd->rx_dmamap,
2032 	    BUS_DMASYNC_PREREAD);
2033 
2034 	desc = &sc->rl_ldata.rl_rx_list[idx];
2035 	desc->rl_vlanctl = 0;
2036 	desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[0].ds_addr));
2037 	desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[0].ds_addr));
2038 	cmdstat = segs[0].ds_len;
2039 	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
2040 		cmdstat |= RL_RDESC_CMD_EOR;
2041 	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
2042 
2043 	return (0);
2044 }
2045 
2046 #ifdef RE_FIXUP_RX
2047 static __inline void
2048 re_fixup_rx(struct mbuf *m)
2049 {
2050 	int                     i;
2051 	uint16_t                *src, *dst;
2052 
2053 	src = mtod(m, uint16_t *);
2054 	dst = src - (RE_ETHER_ALIGN - ETHER_ALIGN) / sizeof *src;
2055 
2056 	for (i = 0; i < (m->m_len / sizeof(uint16_t) + 1); i++)
2057 		*dst++ = *src++;
2058 
2059 	m->m_data -= RE_ETHER_ALIGN - ETHER_ALIGN;
2060 }
2061 #endif
2062 
2063 static int
2064 re_tx_list_init(struct rl_softc *sc)
2065 {
2066 	struct rl_desc		*desc;
2067 	int			i;
2068 
2069 	RL_LOCK_ASSERT(sc);
2070 
2071 	bzero(sc->rl_ldata.rl_tx_list,
2072 	    sc->rl_ldata.rl_tx_desc_cnt * sizeof(struct rl_desc));
2073 	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++)
2074 		sc->rl_ldata.rl_tx_desc[i].tx_m = NULL;
2075 #ifdef DEV_NETMAP
2076 	re_netmap_tx_init(sc);
2077 #endif /* DEV_NETMAP */
2078 	/* Set EOR. */
2079 	desc = &sc->rl_ldata.rl_tx_list[sc->rl_ldata.rl_tx_desc_cnt - 1];
2080 	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_EOR);
2081 
2082 	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2083 	    sc->rl_ldata.rl_tx_list_map,
2084 	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
2085 
2086 	sc->rl_ldata.rl_tx_prodidx = 0;
2087 	sc->rl_ldata.rl_tx_considx = 0;
2088 	sc->rl_ldata.rl_tx_free = sc->rl_ldata.rl_tx_desc_cnt;
2089 
2090 	return (0);
2091 }
2092 
2093 static int
2094 re_rx_list_init(struct rl_softc *sc)
2095 {
2096 	int			error, i;
2097 
2098 	bzero(sc->rl_ldata.rl_rx_list,
2099 	    sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc));
2100 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
2101 		sc->rl_ldata.rl_rx_desc[i].rx_m = NULL;
2102 		if ((error = re_newbuf(sc, i)) != 0)
2103 			return (error);
2104 	}
2105 #ifdef DEV_NETMAP
2106 	re_netmap_rx_init(sc);
2107 #endif /* DEV_NETMAP */
2108 
2109 	/* Flush the RX descriptors */
2110 
2111 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
2112 	    sc->rl_ldata.rl_rx_list_map,
2113 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2114 
2115 	sc->rl_ldata.rl_rx_prodidx = 0;
2116 	sc->rl_head = sc->rl_tail = NULL;
2117 	sc->rl_int_rx_act = 0;
2118 
2119 	return (0);
2120 }
2121 
2122 static int
2123 re_jrx_list_init(struct rl_softc *sc)
2124 {
2125 	int			error, i;
2126 
2127 	bzero(sc->rl_ldata.rl_rx_list,
2128 	    sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc));
2129 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
2130 		sc->rl_ldata.rl_jrx_desc[i].rx_m = NULL;
2131 		if ((error = re_jumbo_newbuf(sc, i)) != 0)
2132 			return (error);
2133 	}
2134 
2135 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
2136 	    sc->rl_ldata.rl_rx_list_map,
2137 	    BUS_DMASYNC_PREWRITE | BUS_DMASYNC_PREREAD);
2138 
2139 	sc->rl_ldata.rl_rx_prodidx = 0;
2140 	sc->rl_head = sc->rl_tail = NULL;
2141 	sc->rl_int_rx_act = 0;
2142 
2143 	return (0);
2144 }
2145 
2146 /*
2147  * RX handler for C+ and 8169. For the gigE chips, we support
2148  * the reception of jumbo frames that have been fragmented
2149  * across multiple 2K mbuf cluster buffers.
2150  */
2151 static int
2152 re_rxeof(struct rl_softc *sc, int *rx_npktsp)
2153 {
2154 	struct mbuf		*m;
2155 	struct ifnet		*ifp;
2156 	int			i, rxerr, total_len;
2157 	struct rl_desc		*cur_rx;
2158 	u_int32_t		rxstat, rxvlan;
2159 	int			jumbo, maxpkt = 16, rx_npkts = 0;
2160 
2161 	RL_LOCK_ASSERT(sc);
2162 
2163 	ifp = sc->rl_ifp;
2164 #ifdef DEV_NETMAP
2165 	if (netmap_rx_irq(ifp, 0, &rx_npkts))
2166 		return 0;
2167 #endif /* DEV_NETMAP */
2168 	if (ifp->if_mtu > RL_MTU && (sc->rl_flags & RL_FLAG_JUMBOV2) != 0)
2169 		jumbo = 1;
2170 	else
2171 		jumbo = 0;
2172 
2173 	/* Invalidate the descriptor memory */
2174 
2175 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
2176 	    sc->rl_ldata.rl_rx_list_map,
2177 	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2178 
2179 	for (i = sc->rl_ldata.rl_rx_prodidx; maxpkt > 0;
2180 	    i = RL_RX_DESC_NXT(sc, i)) {
2181 		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
2182 			break;
2183 		cur_rx = &sc->rl_ldata.rl_rx_list[i];
2184 		rxstat = le32toh(cur_rx->rl_cmdstat);
2185 		if ((rxstat & RL_RDESC_STAT_OWN) != 0)
2186 			break;
2187 		total_len = rxstat & sc->rl_rxlenmask;
2188 		rxvlan = le32toh(cur_rx->rl_vlanctl);
2189 		if (jumbo != 0)
2190 			m = sc->rl_ldata.rl_jrx_desc[i].rx_m;
2191 		else
2192 			m = sc->rl_ldata.rl_rx_desc[i].rx_m;
2193 
2194 		if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0 &&
2195 		    (rxstat & (RL_RDESC_STAT_SOF | RL_RDESC_STAT_EOF)) !=
2196 		    (RL_RDESC_STAT_SOF | RL_RDESC_STAT_EOF)) {
2197 			/*
2198 			 * RTL8168C or later controllers do not
2199 			 * support multi-fragment packet.
2200 			 */
2201 			re_discard_rxbuf(sc, i);
2202 			continue;
2203 		} else if ((rxstat & RL_RDESC_STAT_EOF) == 0) {
2204 			if (re_newbuf(sc, i) != 0) {
2205 				/*
2206 				 * If this is part of a multi-fragment packet,
2207 				 * discard all the pieces.
2208 				 */
2209 				if (sc->rl_head != NULL) {
2210 					m_freem(sc->rl_head);
2211 					sc->rl_head = sc->rl_tail = NULL;
2212 				}
2213 				re_discard_rxbuf(sc, i);
2214 				continue;
2215 			}
2216 			m->m_len = RE_RX_DESC_BUFLEN;
2217 			if (sc->rl_head == NULL)
2218 				sc->rl_head = sc->rl_tail = m;
2219 			else {
2220 				m->m_flags &= ~M_PKTHDR;
2221 				sc->rl_tail->m_next = m;
2222 				sc->rl_tail = m;
2223 			}
2224 			continue;
2225 		}
2226 
2227 		/*
2228 		 * NOTE: for the 8139C+, the frame length field
2229 		 * is always 12 bits in size, but for the gigE chips,
2230 		 * it is 13 bits (since the max RX frame length is 16K).
2231 		 * Unfortunately, all 32 bits in the status word
2232 		 * were already used, so to make room for the extra
2233 		 * length bit, RealTek took out the 'frame alignment
2234 		 * error' bit and shifted the other status bits
2235 		 * over one slot. The OWN, EOR, FS and LS bits are
2236 		 * still in the same places. We have already extracted
2237 		 * the frame length and checked the OWN bit, so rather
2238 		 * than using an alternate bit mapping, we shift the
2239 		 * status bits one space to the right so we can evaluate
2240 		 * them using the 8169 status as though it was in the
2241 		 * same format as that of the 8139C+.
2242 		 */
2243 		if (sc->rl_type == RL_8169)
2244 			rxstat >>= 1;
2245 
2246 		/*
2247 		 * if total_len > 2^13-1, both _RXERRSUM and _GIANT will be
2248 		 * set, but if CRC is clear, it will still be a valid frame.
2249 		 */
2250 		if ((rxstat & RL_RDESC_STAT_RXERRSUM) != 0) {
2251 			rxerr = 1;
2252 			if ((sc->rl_flags & RL_FLAG_JUMBOV2) == 0 &&
2253 			    total_len > 8191 &&
2254 			    (rxstat & RL_RDESC_STAT_ERRS) == RL_RDESC_STAT_GIANT)
2255 				rxerr = 0;
2256 			if (rxerr != 0) {
2257 				if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
2258 				/*
2259 				 * If this is part of a multi-fragment packet,
2260 				 * discard all the pieces.
2261 				 */
2262 				if (sc->rl_head != NULL) {
2263 					m_freem(sc->rl_head);
2264 					sc->rl_head = sc->rl_tail = NULL;
2265 				}
2266 				re_discard_rxbuf(sc, i);
2267 				continue;
2268 			}
2269 		}
2270 
2271 		/*
2272 		 * If allocating a replacement mbuf fails,
2273 		 * reload the current one.
2274 		 */
2275 		if (jumbo != 0)
2276 			rxerr = re_jumbo_newbuf(sc, i);
2277 		else
2278 			rxerr = re_newbuf(sc, i);
2279 		if (rxerr != 0) {
2280 			if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
2281 			if (sc->rl_head != NULL) {
2282 				m_freem(sc->rl_head);
2283 				sc->rl_head = sc->rl_tail = NULL;
2284 			}
2285 			re_discard_rxbuf(sc, i);
2286 			continue;
2287 		}
2288 
2289 		if (sc->rl_head != NULL) {
2290 			if (jumbo != 0)
2291 				m->m_len = total_len;
2292 			else {
2293 				m->m_len = total_len % RE_RX_DESC_BUFLEN;
2294 				if (m->m_len == 0)
2295 					m->m_len = RE_RX_DESC_BUFLEN;
2296 			}
2297 			/*
2298 			 * Special case: if there's 4 bytes or less
2299 			 * in this buffer, the mbuf can be discarded:
2300 			 * the last 4 bytes is the CRC, which we don't
2301 			 * care about anyway.
2302 			 */
2303 			if (m->m_len <= ETHER_CRC_LEN) {
2304 				sc->rl_tail->m_len -=
2305 				    (ETHER_CRC_LEN - m->m_len);
2306 				m_freem(m);
2307 			} else {
2308 				m->m_len -= ETHER_CRC_LEN;
2309 				m->m_flags &= ~M_PKTHDR;
2310 				sc->rl_tail->m_next = m;
2311 			}
2312 			m = sc->rl_head;
2313 			sc->rl_head = sc->rl_tail = NULL;
2314 			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
2315 		} else
2316 			m->m_pkthdr.len = m->m_len =
2317 			    (total_len - ETHER_CRC_LEN);
2318 
2319 #ifdef RE_FIXUP_RX
2320 		re_fixup_rx(m);
2321 #endif
2322 		if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
2323 		m->m_pkthdr.rcvif = ifp;
2324 
2325 		/* Do RX checksumming if enabled */
2326 
2327 		if (ifp->if_capenable & IFCAP_RXCSUM) {
2328 			if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
2329 				/* Check IP header checksum */
2330 				if (rxstat & RL_RDESC_STAT_PROTOID)
2331 					m->m_pkthdr.csum_flags |=
2332 					    CSUM_IP_CHECKED;
2333 				if (!(rxstat & RL_RDESC_STAT_IPSUMBAD))
2334 					m->m_pkthdr.csum_flags |=
2335 					    CSUM_IP_VALID;
2336 
2337 				/* Check TCP/UDP checksum */
2338 				if ((RL_TCPPKT(rxstat) &&
2339 				    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
2340 				    (RL_UDPPKT(rxstat) &&
2341 				     !(rxstat & RL_RDESC_STAT_UDPSUMBAD))) {
2342 					m->m_pkthdr.csum_flags |=
2343 						CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
2344 					m->m_pkthdr.csum_data = 0xffff;
2345 				}
2346 			} else {
2347 				/*
2348 				 * RTL8168C/RTL816CP/RTL8111C/RTL8111CP
2349 				 */
2350 				if ((rxstat & RL_RDESC_STAT_PROTOID) &&
2351 				    (rxvlan & RL_RDESC_IPV4))
2352 					m->m_pkthdr.csum_flags |=
2353 					    CSUM_IP_CHECKED;
2354 				if (!(rxstat & RL_RDESC_STAT_IPSUMBAD) &&
2355 				    (rxvlan & RL_RDESC_IPV4))
2356 					m->m_pkthdr.csum_flags |=
2357 					    CSUM_IP_VALID;
2358 				if (((rxstat & RL_RDESC_STAT_TCP) &&
2359 				    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
2360 				    ((rxstat & RL_RDESC_STAT_UDP) &&
2361 				    !(rxstat & RL_RDESC_STAT_UDPSUMBAD))) {
2362 					m->m_pkthdr.csum_flags |=
2363 						CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
2364 					m->m_pkthdr.csum_data = 0xffff;
2365 				}
2366 			}
2367 		}
2368 		maxpkt--;
2369 		if (rxvlan & RL_RDESC_VLANCTL_TAG) {
2370 			m->m_pkthdr.ether_vtag =
2371 			    bswap16((rxvlan & RL_RDESC_VLANCTL_DATA));
2372 			m->m_flags |= M_VLANTAG;
2373 		}
2374 		RL_UNLOCK(sc);
2375 		(*ifp->if_input)(ifp, m);
2376 		RL_LOCK(sc);
2377 		rx_npkts++;
2378 	}
2379 
2380 	/* Flush the RX DMA ring */
2381 
2382 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
2383 	    sc->rl_ldata.rl_rx_list_map,
2384 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2385 
2386 	sc->rl_ldata.rl_rx_prodidx = i;
2387 
2388 	if (rx_npktsp != NULL)
2389 		*rx_npktsp = rx_npkts;
2390 	if (maxpkt)
2391 		return (EAGAIN);
2392 
2393 	return (0);
2394 }
2395 
2396 static void
2397 re_txeof(struct rl_softc *sc)
2398 {
2399 	struct ifnet		*ifp;
2400 	struct rl_txdesc	*txd;
2401 	u_int32_t		txstat;
2402 	int			cons;
2403 
2404 	cons = sc->rl_ldata.rl_tx_considx;
2405 	if (cons == sc->rl_ldata.rl_tx_prodidx)
2406 		return;
2407 
2408 	ifp = sc->rl_ifp;
2409 #ifdef DEV_NETMAP
2410 	if (netmap_tx_irq(ifp, 0))
2411 		return;
2412 #endif /* DEV_NETMAP */
2413 	/* Invalidate the TX descriptor list */
2414 	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2415 	    sc->rl_ldata.rl_tx_list_map,
2416 	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2417 
2418 	for (; cons != sc->rl_ldata.rl_tx_prodidx;
2419 	    cons = RL_TX_DESC_NXT(sc, cons)) {
2420 		txstat = le32toh(sc->rl_ldata.rl_tx_list[cons].rl_cmdstat);
2421 		if (txstat & RL_TDESC_STAT_OWN)
2422 			break;
2423 		/*
2424 		 * We only stash mbufs in the last descriptor
2425 		 * in a fragment chain, which also happens to
2426 		 * be the only place where the TX status bits
2427 		 * are valid.
2428 		 */
2429 		if (txstat & RL_TDESC_CMD_EOF) {
2430 			txd = &sc->rl_ldata.rl_tx_desc[cons];
2431 			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
2432 			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
2433 			bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag,
2434 			    txd->tx_dmamap);
2435 			KASSERT(txd->tx_m != NULL,
2436 			    ("%s: freeing NULL mbufs!", __func__));
2437 			m_freem(txd->tx_m);
2438 			txd->tx_m = NULL;
2439 			if (txstat & (RL_TDESC_STAT_EXCESSCOL|
2440 			    RL_TDESC_STAT_COLCNT))
2441 				if_inc_counter(ifp, IFCOUNTER_COLLISIONS, 1);
2442 			if (txstat & RL_TDESC_STAT_TXERRSUM)
2443 				if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
2444 			else
2445 				if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
2446 		}
2447 		sc->rl_ldata.rl_tx_free++;
2448 		ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
2449 	}
2450 	sc->rl_ldata.rl_tx_considx = cons;
2451 
2452 	/* No changes made to the TX ring, so no flush needed */
2453 
2454 	if (sc->rl_ldata.rl_tx_free != sc->rl_ldata.rl_tx_desc_cnt) {
2455 #ifdef RE_TX_MODERATION
2456 		/*
2457 		 * If not all descriptors have been reaped yet, reload
2458 		 * the timer so that we will eventually get another
2459 		 * interrupt that will cause us to re-enter this routine.
2460 		 * This is done in case the transmitter has gone idle.
2461 		 */
2462 		CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2463 #endif
2464 	} else
2465 		sc->rl_watchdog_timer = 0;
2466 }
2467 
2468 static void
2469 re_tick(void *xsc)
2470 {
2471 	struct rl_softc		*sc;
2472 	struct mii_data		*mii;
2473 
2474 	sc = xsc;
2475 
2476 	RL_LOCK_ASSERT(sc);
2477 
2478 	mii = device_get_softc(sc->rl_miibus);
2479 	mii_tick(mii);
2480 	if ((sc->rl_flags & RL_FLAG_LINK) == 0)
2481 		re_miibus_statchg(sc->rl_dev);
2482 	/*
2483 	 * Reclaim transmitted frames here. Technically it is not
2484 	 * necessary to do here but it ensures periodic reclamation
2485 	 * regardless of Tx completion interrupt which seems to be
2486 	 * lost on PCIe based controllers under certain situations.
2487 	 */
2488 	re_txeof(sc);
2489 	re_watchdog(sc);
2490 	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
2491 }
2492 
2493 #ifdef DEVICE_POLLING
2494 static int
2495 re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
2496 {
2497 	struct rl_softc *sc = ifp->if_softc;
2498 	int rx_npkts = 0;
2499 
2500 	RL_LOCK(sc);
2501 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
2502 		rx_npkts = re_poll_locked(ifp, cmd, count);
2503 	RL_UNLOCK(sc);
2504 	return (rx_npkts);
2505 }
2506 
2507 static int
2508 re_poll_locked(struct ifnet *ifp, enum poll_cmd cmd, int count)
2509 {
2510 	struct rl_softc *sc = ifp->if_softc;
2511 	int rx_npkts;
2512 
2513 	RL_LOCK_ASSERT(sc);
2514 
2515 	sc->rxcycles = count;
2516 	re_rxeof(sc, &rx_npkts);
2517 	re_txeof(sc);
2518 
2519 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2520 		re_start_locked(ifp);
2521 
2522 	if (cmd == POLL_AND_CHECK_STATUS) { /* also check status register */
2523 		u_int16_t       status;
2524 
2525 		status = CSR_READ_2(sc, RL_ISR);
2526 		if (status == 0xffff)
2527 			return (rx_npkts);
2528 		if (status)
2529 			CSR_WRITE_2(sc, RL_ISR, status);
2530 		if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
2531 		    (sc->rl_flags & RL_FLAG_PCIE))
2532 			CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2533 
2534 		/*
2535 		 * XXX check behaviour on receiver stalls.
2536 		 */
2537 
2538 		if (status & RL_ISR_SYSTEM_ERR) {
2539 			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2540 			re_init_locked(sc);
2541 		}
2542 	}
2543 	return (rx_npkts);
2544 }
2545 #endif /* DEVICE_POLLING */
2546 
2547 static int
2548 re_intr(void *arg)
2549 {
2550 	struct rl_softc		*sc;
2551 	uint16_t		status;
2552 
2553 	sc = arg;
2554 
2555 	status = CSR_READ_2(sc, RL_ISR);
2556 	if (status == 0xFFFF || (status & RL_INTRS_CPLUS) == 0)
2557                 return (FILTER_STRAY);
2558 	CSR_WRITE_2(sc, RL_IMR, 0);
2559 
2560 	taskqueue_enqueue(taskqueue_fast, &sc->rl_inttask);
2561 
2562 	return (FILTER_HANDLED);
2563 }
2564 
2565 static void
2566 re_int_task(void *arg, int npending)
2567 {
2568 	struct rl_softc		*sc;
2569 	struct ifnet		*ifp;
2570 	u_int16_t		status;
2571 	int			rval = 0;
2572 
2573 	sc = arg;
2574 	ifp = sc->rl_ifp;
2575 
2576 	RL_LOCK(sc);
2577 
2578 	status = CSR_READ_2(sc, RL_ISR);
2579         CSR_WRITE_2(sc, RL_ISR, status);
2580 
2581 	if (sc->suspended ||
2582 	    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2583 		RL_UNLOCK(sc);
2584 		return;
2585 	}
2586 
2587 #ifdef DEVICE_POLLING
2588 	if  (ifp->if_capenable & IFCAP_POLLING) {
2589 		RL_UNLOCK(sc);
2590 		return;
2591 	}
2592 #endif
2593 
2594 	if (status & (RL_ISR_RX_OK|RL_ISR_RX_ERR|RL_ISR_FIFO_OFLOW))
2595 		rval = re_rxeof(sc, NULL);
2596 
2597 	/*
2598 	 * Some chips will ignore a second TX request issued
2599 	 * while an existing transmission is in progress. If
2600 	 * the transmitter goes idle but there are still
2601 	 * packets waiting to be sent, we need to restart the
2602 	 * channel here to flush them out. This only seems to
2603 	 * be required with the PCIe devices.
2604 	 */
2605 	if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
2606 	    (sc->rl_flags & RL_FLAG_PCIE))
2607 		CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2608 	if (status & (
2609 #ifdef RE_TX_MODERATION
2610 	    RL_ISR_TIMEOUT_EXPIRED|
2611 #else
2612 	    RL_ISR_TX_OK|
2613 #endif
2614 	    RL_ISR_TX_ERR|RL_ISR_TX_DESC_UNAVAIL))
2615 		re_txeof(sc);
2616 
2617 	if (status & RL_ISR_SYSTEM_ERR) {
2618 		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2619 		re_init_locked(sc);
2620 	}
2621 
2622 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2623 		re_start_locked(ifp);
2624 
2625 	RL_UNLOCK(sc);
2626 
2627         if ((CSR_READ_2(sc, RL_ISR) & RL_INTRS_CPLUS) || rval) {
2628 		taskqueue_enqueue(taskqueue_fast, &sc->rl_inttask);
2629 		return;
2630 	}
2631 
2632 	CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2633 }
2634 
2635 static void
2636 re_intr_msi(void *xsc)
2637 {
2638 	struct rl_softc		*sc;
2639 	struct ifnet		*ifp;
2640 	uint16_t		intrs, status;
2641 
2642 	sc = xsc;
2643 	RL_LOCK(sc);
2644 
2645 	ifp = sc->rl_ifp;
2646 #ifdef DEVICE_POLLING
2647 	if (ifp->if_capenable & IFCAP_POLLING) {
2648 		RL_UNLOCK(sc);
2649 		return;
2650 	}
2651 #endif
2652 	/* Disable interrupts. */
2653 	CSR_WRITE_2(sc, RL_IMR, 0);
2654 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2655 		RL_UNLOCK(sc);
2656 		return;
2657 	}
2658 
2659 	intrs = RL_INTRS_CPLUS;
2660 	status = CSR_READ_2(sc, RL_ISR);
2661         CSR_WRITE_2(sc, RL_ISR, status);
2662 	if (sc->rl_int_rx_act > 0) {
2663 		intrs &= ~(RL_ISR_RX_OK | RL_ISR_RX_ERR | RL_ISR_FIFO_OFLOW |
2664 		    RL_ISR_RX_OVERRUN);
2665 		status &= ~(RL_ISR_RX_OK | RL_ISR_RX_ERR | RL_ISR_FIFO_OFLOW |
2666 		    RL_ISR_RX_OVERRUN);
2667 	}
2668 
2669 	if (status & (RL_ISR_TIMEOUT_EXPIRED | RL_ISR_RX_OK | RL_ISR_RX_ERR |
2670 	    RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN)) {
2671 		re_rxeof(sc, NULL);
2672 		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
2673 			if (sc->rl_int_rx_mod != 0 &&
2674 			    (status & (RL_ISR_RX_OK | RL_ISR_RX_ERR |
2675 			    RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN)) != 0) {
2676 				/* Rearm one-shot timer. */
2677 				CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2678 				intrs &= ~(RL_ISR_RX_OK | RL_ISR_RX_ERR |
2679 				    RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN);
2680 				sc->rl_int_rx_act = 1;
2681 			} else {
2682 				intrs |= RL_ISR_RX_OK | RL_ISR_RX_ERR |
2683 				    RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN;
2684 				sc->rl_int_rx_act = 0;
2685 			}
2686 		}
2687 	}
2688 
2689 	/*
2690 	 * Some chips will ignore a second TX request issued
2691 	 * while an existing transmission is in progress. If
2692 	 * the transmitter goes idle but there are still
2693 	 * packets waiting to be sent, we need to restart the
2694 	 * channel here to flush them out. This only seems to
2695 	 * be required with the PCIe devices.
2696 	 */
2697 	if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
2698 	    (sc->rl_flags & RL_FLAG_PCIE))
2699 		CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2700 	if (status & (RL_ISR_TX_OK | RL_ISR_TX_ERR | RL_ISR_TX_DESC_UNAVAIL))
2701 		re_txeof(sc);
2702 
2703 	if (status & RL_ISR_SYSTEM_ERR) {
2704 		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2705 		re_init_locked(sc);
2706 	}
2707 
2708 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
2709 		if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2710 			re_start_locked(ifp);
2711 		CSR_WRITE_2(sc, RL_IMR, intrs);
2712 	}
2713 	RL_UNLOCK(sc);
2714 }
2715 
2716 static int
2717 re_encap(struct rl_softc *sc, struct mbuf **m_head)
2718 {
2719 	struct rl_txdesc	*txd, *txd_last;
2720 	bus_dma_segment_t	segs[RL_NTXSEGS];
2721 	bus_dmamap_t		map;
2722 	struct mbuf		*m_new;
2723 	struct rl_desc		*desc;
2724 	int			nsegs, prod;
2725 	int			i, error, ei, si;
2726 	int			padlen;
2727 	uint32_t		cmdstat, csum_flags, vlanctl;
2728 
2729 	RL_LOCK_ASSERT(sc);
2730 	M_ASSERTPKTHDR((*m_head));
2731 
2732 	/*
2733 	 * With some of the RealTek chips, using the checksum offload
2734 	 * support in conjunction with the autopadding feature results
2735 	 * in the transmission of corrupt frames. For example, if we
2736 	 * need to send a really small IP fragment that's less than 60
2737 	 * bytes in size, and IP header checksumming is enabled, the
2738 	 * resulting ethernet frame that appears on the wire will
2739 	 * have garbled payload. To work around this, if TX IP checksum
2740 	 * offload is enabled, we always manually pad short frames out
2741 	 * to the minimum ethernet frame size.
2742 	 */
2743 	if ((sc->rl_flags & RL_FLAG_AUTOPAD) == 0 &&
2744 	    (*m_head)->m_pkthdr.len < RL_IP4CSUMTX_PADLEN &&
2745 	    ((*m_head)->m_pkthdr.csum_flags & CSUM_IP) != 0) {
2746 		padlen = RL_MIN_FRAMELEN - (*m_head)->m_pkthdr.len;
2747 		if (M_WRITABLE(*m_head) == 0) {
2748 			/* Get a writable copy. */
2749 			m_new = m_dup(*m_head, M_NOWAIT);
2750 			m_freem(*m_head);
2751 			if (m_new == NULL) {
2752 				*m_head = NULL;
2753 				return (ENOBUFS);
2754 			}
2755 			*m_head = m_new;
2756 		}
2757 		if ((*m_head)->m_next != NULL ||
2758 		    M_TRAILINGSPACE(*m_head) < padlen) {
2759 			m_new = m_defrag(*m_head, M_NOWAIT);
2760 			if (m_new == NULL) {
2761 				m_freem(*m_head);
2762 				*m_head = NULL;
2763 				return (ENOBUFS);
2764 			}
2765 		} else
2766 			m_new = *m_head;
2767 
2768 		/*
2769 		 * Manually pad short frames, and zero the pad space
2770 		 * to avoid leaking data.
2771 		 */
2772 		bzero(mtod(m_new, char *) + m_new->m_pkthdr.len, padlen);
2773 		m_new->m_pkthdr.len += padlen;
2774 		m_new->m_len = m_new->m_pkthdr.len;
2775 		*m_head = m_new;
2776 	}
2777 
2778 	prod = sc->rl_ldata.rl_tx_prodidx;
2779 	txd = &sc->rl_ldata.rl_tx_desc[prod];
2780 	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap,
2781 	    *m_head, segs, &nsegs, BUS_DMA_NOWAIT);
2782 	if (error == EFBIG) {
2783 		m_new = m_collapse(*m_head, M_NOWAIT, RL_NTXSEGS);
2784 		if (m_new == NULL) {
2785 			m_freem(*m_head);
2786 			*m_head = NULL;
2787 			return (ENOBUFS);
2788 		}
2789 		*m_head = m_new;
2790 		error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_tx_mtag,
2791 		    txd->tx_dmamap, *m_head, segs, &nsegs, BUS_DMA_NOWAIT);
2792 		if (error != 0) {
2793 			m_freem(*m_head);
2794 			*m_head = NULL;
2795 			return (error);
2796 		}
2797 	} else if (error != 0)
2798 		return (error);
2799 	if (nsegs == 0) {
2800 		m_freem(*m_head);
2801 		*m_head = NULL;
2802 		return (EIO);
2803 	}
2804 
2805 	/* Check for number of available descriptors. */
2806 	if (sc->rl_ldata.rl_tx_free - nsegs <= 1) {
2807 		bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap);
2808 		return (ENOBUFS);
2809 	}
2810 
2811 	bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap,
2812 	    BUS_DMASYNC_PREWRITE);
2813 
2814 	/*
2815 	 * Set up checksum offload. Note: checksum offload bits must
2816 	 * appear in all descriptors of a multi-descriptor transmit
2817 	 * attempt. This is according to testing done with an 8169
2818 	 * chip. This is a requirement.
2819 	 */
2820 	vlanctl = 0;
2821 	csum_flags = 0;
2822 	if (((*m_head)->m_pkthdr.csum_flags & CSUM_TSO) != 0) {
2823 		if ((sc->rl_flags & RL_FLAG_DESCV2) != 0) {
2824 			csum_flags |= RL_TDESC_CMD_LGSEND;
2825 			vlanctl |= ((uint32_t)(*m_head)->m_pkthdr.tso_segsz <<
2826 			    RL_TDESC_CMD_MSSVALV2_SHIFT);
2827 		} else {
2828 			csum_flags |= RL_TDESC_CMD_LGSEND |
2829 			    ((uint32_t)(*m_head)->m_pkthdr.tso_segsz <<
2830 			    RL_TDESC_CMD_MSSVAL_SHIFT);
2831 		}
2832 	} else {
2833 		/*
2834 		 * Unconditionally enable IP checksum if TCP or UDP
2835 		 * checksum is required. Otherwise, TCP/UDP checksum
2836 		 * doesn't make effects.
2837 		 */
2838 		if (((*m_head)->m_pkthdr.csum_flags & RE_CSUM_FEATURES) != 0) {
2839 			if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
2840 				csum_flags |= RL_TDESC_CMD_IPCSUM;
2841 				if (((*m_head)->m_pkthdr.csum_flags &
2842 				    CSUM_TCP) != 0)
2843 					csum_flags |= RL_TDESC_CMD_TCPCSUM;
2844 				if (((*m_head)->m_pkthdr.csum_flags &
2845 				    CSUM_UDP) != 0)
2846 					csum_flags |= RL_TDESC_CMD_UDPCSUM;
2847 			} else {
2848 				vlanctl |= RL_TDESC_CMD_IPCSUMV2;
2849 				if (((*m_head)->m_pkthdr.csum_flags &
2850 				    CSUM_TCP) != 0)
2851 					vlanctl |= RL_TDESC_CMD_TCPCSUMV2;
2852 				if (((*m_head)->m_pkthdr.csum_flags &
2853 				    CSUM_UDP) != 0)
2854 					vlanctl |= RL_TDESC_CMD_UDPCSUMV2;
2855 			}
2856 		}
2857 	}
2858 
2859 	/*
2860 	 * Set up hardware VLAN tagging. Note: vlan tag info must
2861 	 * appear in all descriptors of a multi-descriptor
2862 	 * transmission attempt.
2863 	 */
2864 	if ((*m_head)->m_flags & M_VLANTAG)
2865 		vlanctl |= bswap16((*m_head)->m_pkthdr.ether_vtag) |
2866 		    RL_TDESC_VLANCTL_TAG;
2867 
2868 	si = prod;
2869 	for (i = 0; i < nsegs; i++, prod = RL_TX_DESC_NXT(sc, prod)) {
2870 		desc = &sc->rl_ldata.rl_tx_list[prod];
2871 		desc->rl_vlanctl = htole32(vlanctl);
2872 		desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[i].ds_addr));
2873 		desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[i].ds_addr));
2874 		cmdstat = segs[i].ds_len;
2875 		if (i != 0)
2876 			cmdstat |= RL_TDESC_CMD_OWN;
2877 		if (prod == sc->rl_ldata.rl_tx_desc_cnt - 1)
2878 			cmdstat |= RL_TDESC_CMD_EOR;
2879 		desc->rl_cmdstat = htole32(cmdstat | csum_flags);
2880 		sc->rl_ldata.rl_tx_free--;
2881 	}
2882 	/* Update producer index. */
2883 	sc->rl_ldata.rl_tx_prodidx = prod;
2884 
2885 	/* Set EOF on the last descriptor. */
2886 	ei = RL_TX_DESC_PRV(sc, prod);
2887 	desc = &sc->rl_ldata.rl_tx_list[ei];
2888 	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_EOF);
2889 
2890 	desc = &sc->rl_ldata.rl_tx_list[si];
2891 	/* Set SOF and transfer ownership of packet to the chip. */
2892 	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_OWN | RL_TDESC_CMD_SOF);
2893 
2894 	/*
2895 	 * Insure that the map for this transmission
2896 	 * is placed at the array index of the last descriptor
2897 	 * in this chain.  (Swap last and first dmamaps.)
2898 	 */
2899 	txd_last = &sc->rl_ldata.rl_tx_desc[ei];
2900 	map = txd->tx_dmamap;
2901 	txd->tx_dmamap = txd_last->tx_dmamap;
2902 	txd_last->tx_dmamap = map;
2903 	txd_last->tx_m = *m_head;
2904 
2905 	return (0);
2906 }
2907 
2908 static void
2909 re_start(struct ifnet *ifp)
2910 {
2911 	struct rl_softc		*sc;
2912 
2913 	sc = ifp->if_softc;
2914 	RL_LOCK(sc);
2915 	re_start_locked(ifp);
2916 	RL_UNLOCK(sc);
2917 }
2918 
2919 /*
2920  * Main transmit routine for C+ and gigE NICs.
2921  */
2922 static void
2923 re_start_locked(struct ifnet *ifp)
2924 {
2925 	struct rl_softc		*sc;
2926 	struct mbuf		*m_head;
2927 	int			queued;
2928 
2929 	sc = ifp->if_softc;
2930 
2931 #ifdef DEV_NETMAP
2932 	/* XXX is this necessary ? */
2933 	if (ifp->if_capenable & IFCAP_NETMAP) {
2934 		struct netmap_kring *kring = &NA(ifp)->tx_rings[0];
2935 		if (sc->rl_ldata.rl_tx_prodidx != kring->nr_hwcur) {
2936 			/* kick the tx unit */
2937 			CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2938 #ifdef RE_TX_MODERATION
2939 			CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2940 #endif
2941 			sc->rl_watchdog_timer = 5;
2942 		}
2943 		return;
2944 	}
2945 #endif /* DEV_NETMAP */
2946 
2947 	if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
2948 	    IFF_DRV_RUNNING || (sc->rl_flags & RL_FLAG_LINK) == 0)
2949 		return;
2950 
2951 	for (queued = 0; !IFQ_DRV_IS_EMPTY(&ifp->if_snd) &&
2952 	    sc->rl_ldata.rl_tx_free > 1;) {
2953 		IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
2954 		if (m_head == NULL)
2955 			break;
2956 
2957 		if (re_encap(sc, &m_head) != 0) {
2958 			if (m_head == NULL)
2959 				break;
2960 			IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
2961 			ifp->if_drv_flags |= IFF_DRV_OACTIVE;
2962 			break;
2963 		}
2964 
2965 		/*
2966 		 * If there's a BPF listener, bounce a copy of this frame
2967 		 * to him.
2968 		 */
2969 		ETHER_BPF_MTAP(ifp, m_head);
2970 
2971 		queued++;
2972 	}
2973 
2974 	if (queued == 0) {
2975 #ifdef RE_TX_MODERATION
2976 		if (sc->rl_ldata.rl_tx_free != sc->rl_ldata.rl_tx_desc_cnt)
2977 			CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2978 #endif
2979 		return;
2980 	}
2981 
2982 	/* Flush the TX descriptors */
2983 
2984 	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2985 	    sc->rl_ldata.rl_tx_list_map,
2986 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2987 
2988 	CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2989 
2990 #ifdef RE_TX_MODERATION
2991 	/*
2992 	 * Use the countdown timer for interrupt moderation.
2993 	 * 'TX done' interrupts are disabled. Instead, we reset the
2994 	 * countdown timer, which will begin counting until it hits
2995 	 * the value in the TIMERINT register, and then trigger an
2996 	 * interrupt. Each time we write to the TIMERCNT register,
2997 	 * the timer count is reset to 0.
2998 	 */
2999 	CSR_WRITE_4(sc, RL_TIMERCNT, 1);
3000 #endif
3001 
3002 	/*
3003 	 * Set a timeout in case the chip goes out to lunch.
3004 	 */
3005 	sc->rl_watchdog_timer = 5;
3006 }
3007 
3008 static void
3009 re_set_jumbo(struct rl_softc *sc, int jumbo)
3010 {
3011 
3012 	if (sc->rl_hwrev->rl_rev == RL_HWREV_8168E_VL) {
3013 		pci_set_max_read_req(sc->rl_dev, 4096);
3014 		return;
3015 	}
3016 
3017 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG);
3018 	if (jumbo != 0) {
3019 		CSR_WRITE_1(sc, sc->rl_cfg3, CSR_READ_1(sc, sc->rl_cfg3) |
3020 		    RL_CFG3_JUMBO_EN0);
3021 		switch (sc->rl_hwrev->rl_rev) {
3022 		case RL_HWREV_8168DP:
3023 			break;
3024 		case RL_HWREV_8168E:
3025 			CSR_WRITE_1(sc, sc->rl_cfg4,
3026 			    CSR_READ_1(sc, sc->rl_cfg4) | 0x01);
3027 			break;
3028 		default:
3029 			CSR_WRITE_1(sc, sc->rl_cfg4,
3030 			    CSR_READ_1(sc, sc->rl_cfg4) | RL_CFG4_JUMBO_EN1);
3031 		}
3032 	} else {
3033 		CSR_WRITE_1(sc, sc->rl_cfg3, CSR_READ_1(sc, sc->rl_cfg3) &
3034 		    ~RL_CFG3_JUMBO_EN0);
3035 		switch (sc->rl_hwrev->rl_rev) {
3036 		case RL_HWREV_8168DP:
3037 			break;
3038 		case RL_HWREV_8168E:
3039 			CSR_WRITE_1(sc, sc->rl_cfg4,
3040 			    CSR_READ_1(sc, sc->rl_cfg4) & ~0x01);
3041 			break;
3042 		default:
3043 			CSR_WRITE_1(sc, sc->rl_cfg4,
3044 			    CSR_READ_1(sc, sc->rl_cfg4) & ~RL_CFG4_JUMBO_EN1);
3045 		}
3046 	}
3047 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3048 
3049 	switch (sc->rl_hwrev->rl_rev) {
3050 	case RL_HWREV_8168DP:
3051 		pci_set_max_read_req(sc->rl_dev, 4096);
3052 		break;
3053 	default:
3054 		if (jumbo != 0)
3055 			pci_set_max_read_req(sc->rl_dev, 512);
3056 		else
3057 			pci_set_max_read_req(sc->rl_dev, 4096);
3058 	}
3059 }
3060 
3061 static void
3062 re_init(void *xsc)
3063 {
3064 	struct rl_softc		*sc = xsc;
3065 
3066 	RL_LOCK(sc);
3067 	re_init_locked(sc);
3068 	RL_UNLOCK(sc);
3069 }
3070 
3071 static void
3072 re_init_locked(struct rl_softc *sc)
3073 {
3074 	struct ifnet		*ifp = sc->rl_ifp;
3075 	struct mii_data		*mii;
3076 	uint32_t		reg;
3077 	uint16_t		cfg;
3078 	union {
3079 		uint32_t align_dummy;
3080 		u_char eaddr[ETHER_ADDR_LEN];
3081         } eaddr;
3082 
3083 	RL_LOCK_ASSERT(sc);
3084 
3085 	mii = device_get_softc(sc->rl_miibus);
3086 
3087 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
3088 		return;
3089 
3090 	/*
3091 	 * Cancel pending I/O and free all RX/TX buffers.
3092 	 */
3093 	re_stop(sc);
3094 
3095 	/* Put controller into known state. */
3096 	re_reset(sc);
3097 
3098 	/*
3099 	 * For C+ mode, initialize the RX descriptors and mbufs.
3100 	 */
3101 	if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
3102 		if (ifp->if_mtu > RL_MTU) {
3103 			if (re_jrx_list_init(sc) != 0) {
3104 				device_printf(sc->rl_dev,
3105 				    "no memory for jumbo RX buffers\n");
3106 				re_stop(sc);
3107 				return;
3108 			}
3109 			/* Disable checksum offloading for jumbo frames. */
3110 			ifp->if_capenable &= ~(IFCAP_HWCSUM | IFCAP_TSO4);
3111 			ifp->if_hwassist &= ~(RE_CSUM_FEATURES | CSUM_TSO);
3112 		} else {
3113 			if (re_rx_list_init(sc) != 0) {
3114 				device_printf(sc->rl_dev,
3115 				    "no memory for RX buffers\n");
3116 				re_stop(sc);
3117 				return;
3118 			}
3119 		}
3120 		re_set_jumbo(sc, ifp->if_mtu > RL_MTU);
3121 	} else {
3122 		if (re_rx_list_init(sc) != 0) {
3123 			device_printf(sc->rl_dev, "no memory for RX buffers\n");
3124 			re_stop(sc);
3125 			return;
3126 		}
3127 		if ((sc->rl_flags & RL_FLAG_PCIE) != 0 &&
3128 		    pci_get_device(sc->rl_dev) != RT_DEVICEID_8101E) {
3129 			if (ifp->if_mtu > RL_MTU)
3130 				pci_set_max_read_req(sc->rl_dev, 512);
3131 			else
3132 				pci_set_max_read_req(sc->rl_dev, 4096);
3133 		}
3134 	}
3135 	re_tx_list_init(sc);
3136 
3137 	/*
3138 	 * Enable C+ RX and TX mode, as well as VLAN stripping and
3139 	 * RX checksum offload. We must configure the C+ register
3140 	 * before all others.
3141 	 */
3142 	cfg = RL_CPLUSCMD_PCI_MRW;
3143 	if ((ifp->if_capenable & IFCAP_RXCSUM) != 0)
3144 		cfg |= RL_CPLUSCMD_RXCSUM_ENB;
3145 	if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0)
3146 		cfg |= RL_CPLUSCMD_VLANSTRIP;
3147 	if ((sc->rl_flags & RL_FLAG_MACSTAT) != 0) {
3148 		cfg |= RL_CPLUSCMD_MACSTAT_DIS;
3149 		/* XXX magic. */
3150 		cfg |= 0x0001;
3151 	} else
3152 		cfg |= RL_CPLUSCMD_RXENB | RL_CPLUSCMD_TXENB;
3153 	CSR_WRITE_2(sc, RL_CPLUS_CMD, cfg);
3154 	if (sc->rl_hwrev->rl_rev == RL_HWREV_8169_8110SC ||
3155 	    sc->rl_hwrev->rl_rev == RL_HWREV_8169_8110SCE) {
3156 		reg = 0x000fff00;
3157 		if ((CSR_READ_1(sc, sc->rl_cfg2) & RL_CFG2_PCI66MHZ) != 0)
3158 			reg |= 0x000000ff;
3159 		if (sc->rl_hwrev->rl_rev == RL_HWREV_8169_8110SCE)
3160 			reg |= 0x00f00000;
3161 		CSR_WRITE_4(sc, 0x7c, reg);
3162 		/* Disable interrupt mitigation. */
3163 		CSR_WRITE_2(sc, 0xe2, 0);
3164 	}
3165 	/*
3166 	 * Disable TSO if interface MTU size is greater than MSS
3167 	 * allowed in controller.
3168 	 */
3169 	if (ifp->if_mtu > RL_TSO_MTU && (ifp->if_capenable & IFCAP_TSO4) != 0) {
3170 		ifp->if_capenable &= ~IFCAP_TSO4;
3171 		ifp->if_hwassist &= ~CSUM_TSO;
3172 	}
3173 
3174 	/*
3175 	 * Init our MAC address.  Even though the chipset
3176 	 * documentation doesn't mention it, we need to enter "Config
3177 	 * register write enable" mode to modify the ID registers.
3178 	 */
3179 	/* Copy MAC address on stack to align. */
3180 	bcopy(IF_LLADDR(ifp), eaddr.eaddr, ETHER_ADDR_LEN);
3181 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG);
3182 	CSR_WRITE_4(sc, RL_IDR0,
3183 	    htole32(*(u_int32_t *)(&eaddr.eaddr[0])));
3184 	CSR_WRITE_4(sc, RL_IDR4,
3185 	    htole32(*(u_int32_t *)(&eaddr.eaddr[4])));
3186 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3187 
3188 	/*
3189 	 * Load the addresses of the RX and TX lists into the chip.
3190 	 */
3191 
3192 	CSR_WRITE_4(sc, RL_RXLIST_ADDR_HI,
3193 	    RL_ADDR_HI(sc->rl_ldata.rl_rx_list_addr));
3194 	CSR_WRITE_4(sc, RL_RXLIST_ADDR_LO,
3195 	    RL_ADDR_LO(sc->rl_ldata.rl_rx_list_addr));
3196 
3197 	CSR_WRITE_4(sc, RL_TXLIST_ADDR_HI,
3198 	    RL_ADDR_HI(sc->rl_ldata.rl_tx_list_addr));
3199 	CSR_WRITE_4(sc, RL_TXLIST_ADDR_LO,
3200 	    RL_ADDR_LO(sc->rl_ldata.rl_tx_list_addr));
3201 
3202 	if ((sc->rl_flags & RL_FLAG_8168G_PLUS) != 0) {
3203 		/* Disable RXDV gate. */
3204 		CSR_WRITE_4(sc, RL_MISC, CSR_READ_4(sc, RL_MISC) &
3205 		    ~0x00080000);
3206 	}
3207 
3208 	/*
3209 	 * Enable transmit and receive for pre-RTL8168G controllers.
3210 	 * RX/TX MACs should be enabled before RX/TX configuration.
3211 	 */
3212 	if ((sc->rl_flags & RL_FLAG_8168G_PLUS) == 0)
3213 		CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB | RL_CMD_RX_ENB);
3214 
3215 	/*
3216 	 * Set the initial TX configuration.
3217 	 */
3218 	if (sc->rl_testmode) {
3219 		if (sc->rl_type == RL_8169)
3220 			CSR_WRITE_4(sc, RL_TXCFG,
3221 			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON);
3222 		else
3223 			CSR_WRITE_4(sc, RL_TXCFG,
3224 			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON_CPLUS);
3225 	} else
3226 		CSR_WRITE_4(sc, RL_TXCFG, RL_TXCFG_CONFIG);
3227 
3228 	CSR_WRITE_1(sc, RL_EARLY_TX_THRESH, 16);
3229 
3230 	/*
3231 	 * Set the initial RX configuration.
3232 	 */
3233 	re_set_rxmode(sc);
3234 
3235 	/* Configure interrupt moderation. */
3236 	if (sc->rl_type == RL_8169) {
3237 		/* Magic from vendor. */
3238 		CSR_WRITE_2(sc, RL_INTRMOD, 0x5100);
3239 	}
3240 
3241 	/*
3242 	 * Enable transmit and receive for RTL8168G and later controllers.
3243 	 * RX/TX MACs should be enabled after RX/TX configuration.
3244 	 */
3245 	if ((sc->rl_flags & RL_FLAG_8168G_PLUS) != 0)
3246 		CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB | RL_CMD_RX_ENB);
3247 
3248 #ifdef DEVICE_POLLING
3249 	/*
3250 	 * Disable interrupts if we are polling.
3251 	 */
3252 	if (ifp->if_capenable & IFCAP_POLLING)
3253 		CSR_WRITE_2(sc, RL_IMR, 0);
3254 	else	/* otherwise ... */
3255 #endif
3256 
3257 	/*
3258 	 * Enable interrupts.
3259 	 */
3260 	if (sc->rl_testmode)
3261 		CSR_WRITE_2(sc, RL_IMR, 0);
3262 	else
3263 		CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
3264 	CSR_WRITE_2(sc, RL_ISR, RL_INTRS_CPLUS);
3265 
3266 	/* Set initial TX threshold */
3267 	sc->rl_txthresh = RL_TX_THRESH_INIT;
3268 
3269 	/* Start RX/TX process. */
3270 	CSR_WRITE_4(sc, RL_MISSEDPKT, 0);
3271 
3272 	/*
3273 	 * Initialize the timer interrupt register so that
3274 	 * a timer interrupt will be generated once the timer
3275 	 * reaches a certain number of ticks. The timer is
3276 	 * reloaded on each transmit.
3277 	 */
3278 #ifdef RE_TX_MODERATION
3279 	/*
3280 	 * Use timer interrupt register to moderate TX interrupt
3281 	 * moderation, which dramatically improves TX frame rate.
3282 	 */
3283 	if (sc->rl_type == RL_8169)
3284 		CSR_WRITE_4(sc, RL_TIMERINT_8169, 0x800);
3285 	else
3286 		CSR_WRITE_4(sc, RL_TIMERINT, 0x400);
3287 #else
3288 	/*
3289 	 * Use timer interrupt register to moderate RX interrupt
3290 	 * moderation.
3291 	 */
3292 	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) != 0 &&
3293 	    intr_filter == 0) {
3294 		if (sc->rl_type == RL_8169)
3295 			CSR_WRITE_4(sc, RL_TIMERINT_8169,
3296 			    RL_USECS(sc->rl_int_rx_mod));
3297 	} else {
3298 		if (sc->rl_type == RL_8169)
3299 			CSR_WRITE_4(sc, RL_TIMERINT_8169, RL_USECS(0));
3300 	}
3301 #endif
3302 
3303 	/*
3304 	 * For 8169 gigE NICs, set the max allowed RX packet
3305 	 * size so we can receive jumbo frames.
3306 	 */
3307 	if (sc->rl_type == RL_8169) {
3308 		if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
3309 			/*
3310 			 * For controllers that use new jumbo frame scheme,
3311 			 * set maximum size of jumbo frame depending on
3312 			 * controller revisions.
3313 			 */
3314 			if (ifp->if_mtu > RL_MTU)
3315 				CSR_WRITE_2(sc, RL_MAXRXPKTLEN,
3316 				    sc->rl_hwrev->rl_max_mtu +
3317 				    ETHER_VLAN_ENCAP_LEN + ETHER_HDR_LEN +
3318 				    ETHER_CRC_LEN);
3319 			else
3320 				CSR_WRITE_2(sc, RL_MAXRXPKTLEN,
3321 				    RE_RX_DESC_BUFLEN);
3322 		} else if ((sc->rl_flags & RL_FLAG_PCIE) != 0 &&
3323 		    sc->rl_hwrev->rl_max_mtu == RL_MTU) {
3324 			/* RTL810x has no jumbo frame support. */
3325 			CSR_WRITE_2(sc, RL_MAXRXPKTLEN, RE_RX_DESC_BUFLEN);
3326 		} else
3327 			CSR_WRITE_2(sc, RL_MAXRXPKTLEN, 16383);
3328 	}
3329 
3330 	if (sc->rl_testmode)
3331 		return;
3332 
3333 	CSR_WRITE_1(sc, sc->rl_cfg1, CSR_READ_1(sc, sc->rl_cfg1) |
3334 	    RL_CFG1_DRVLOAD);
3335 
3336 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
3337 	ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
3338 
3339 	sc->rl_flags &= ~RL_FLAG_LINK;
3340 	mii_mediachg(mii);
3341 
3342 	sc->rl_watchdog_timer = 0;
3343 	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
3344 }
3345 
3346 /*
3347  * Set media options.
3348  */
3349 static int
3350 re_ifmedia_upd(struct ifnet *ifp)
3351 {
3352 	struct rl_softc		*sc;
3353 	struct mii_data		*mii;
3354 	int			error;
3355 
3356 	sc = ifp->if_softc;
3357 	mii = device_get_softc(sc->rl_miibus);
3358 	RL_LOCK(sc);
3359 	error = mii_mediachg(mii);
3360 	RL_UNLOCK(sc);
3361 
3362 	return (error);
3363 }
3364 
3365 /*
3366  * Report current media status.
3367  */
3368 static void
3369 re_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
3370 {
3371 	struct rl_softc		*sc;
3372 	struct mii_data		*mii;
3373 
3374 	sc = ifp->if_softc;
3375 	mii = device_get_softc(sc->rl_miibus);
3376 
3377 	RL_LOCK(sc);
3378 	mii_pollstat(mii);
3379 	ifmr->ifm_active = mii->mii_media_active;
3380 	ifmr->ifm_status = mii->mii_media_status;
3381 	RL_UNLOCK(sc);
3382 }
3383 
3384 static int
3385 re_ioctl(struct ifnet *ifp, u_long command, caddr_t data)
3386 {
3387 	struct rl_softc		*sc = ifp->if_softc;
3388 	struct ifreq		*ifr = (struct ifreq *) data;
3389 	struct mii_data		*mii;
3390 	int			error = 0;
3391 
3392 	switch (command) {
3393 	case SIOCSIFMTU:
3394 		if (ifr->ifr_mtu < ETHERMIN ||
3395 		    ifr->ifr_mtu > sc->rl_hwrev->rl_max_mtu ||
3396 		    ((sc->rl_flags & RL_FLAG_FASTETHER) != 0 &&
3397 		    ifr->ifr_mtu > RL_MTU)) {
3398 			error = EINVAL;
3399 			break;
3400 		}
3401 		RL_LOCK(sc);
3402 		if (ifp->if_mtu != ifr->ifr_mtu) {
3403 			ifp->if_mtu = ifr->ifr_mtu;
3404 			if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0 &&
3405 			    (ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
3406 				ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3407 				re_init_locked(sc);
3408 			}
3409 			if (ifp->if_mtu > RL_TSO_MTU &&
3410 			    (ifp->if_capenable & IFCAP_TSO4) != 0) {
3411 				ifp->if_capenable &= ~(IFCAP_TSO4 |
3412 				    IFCAP_VLAN_HWTSO);
3413 				ifp->if_hwassist &= ~CSUM_TSO;
3414 			}
3415 			VLAN_CAPABILITIES(ifp);
3416 		}
3417 		RL_UNLOCK(sc);
3418 		break;
3419 	case SIOCSIFFLAGS:
3420 		RL_LOCK(sc);
3421 		if ((ifp->if_flags & IFF_UP) != 0) {
3422 			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
3423 				if (((ifp->if_flags ^ sc->rl_if_flags)
3424 				    & (IFF_PROMISC | IFF_ALLMULTI)) != 0)
3425 					re_set_rxmode(sc);
3426 			} else
3427 				re_init_locked(sc);
3428 		} else {
3429 			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
3430 				re_stop(sc);
3431 		}
3432 		sc->rl_if_flags = ifp->if_flags;
3433 		RL_UNLOCK(sc);
3434 		break;
3435 	case SIOCADDMULTI:
3436 	case SIOCDELMULTI:
3437 		RL_LOCK(sc);
3438 		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
3439 			re_set_rxmode(sc);
3440 		RL_UNLOCK(sc);
3441 		break;
3442 	case SIOCGIFMEDIA:
3443 	case SIOCSIFMEDIA:
3444 		mii = device_get_softc(sc->rl_miibus);
3445 		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, command);
3446 		break;
3447 	case SIOCSIFCAP:
3448 	    {
3449 		int mask, reinit;
3450 
3451 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
3452 		reinit = 0;
3453 #ifdef DEVICE_POLLING
3454 		if (mask & IFCAP_POLLING) {
3455 			if (ifr->ifr_reqcap & IFCAP_POLLING) {
3456 				error = ether_poll_register(re_poll, ifp);
3457 				if (error)
3458 					return (error);
3459 				RL_LOCK(sc);
3460 				/* Disable interrupts */
3461 				CSR_WRITE_2(sc, RL_IMR, 0x0000);
3462 				ifp->if_capenable |= IFCAP_POLLING;
3463 				RL_UNLOCK(sc);
3464 			} else {
3465 				error = ether_poll_deregister(ifp);
3466 				/* Enable interrupts. */
3467 				RL_LOCK(sc);
3468 				CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
3469 				ifp->if_capenable &= ~IFCAP_POLLING;
3470 				RL_UNLOCK(sc);
3471 			}
3472 		}
3473 #endif /* DEVICE_POLLING */
3474 		RL_LOCK(sc);
3475 		if ((mask & IFCAP_TXCSUM) != 0 &&
3476 		    (ifp->if_capabilities & IFCAP_TXCSUM) != 0) {
3477 			ifp->if_capenable ^= IFCAP_TXCSUM;
3478 			if ((ifp->if_capenable & IFCAP_TXCSUM) != 0)
3479 				ifp->if_hwassist |= RE_CSUM_FEATURES;
3480 			else
3481 				ifp->if_hwassist &= ~RE_CSUM_FEATURES;
3482 			reinit = 1;
3483 		}
3484 		if ((mask & IFCAP_RXCSUM) != 0 &&
3485 		    (ifp->if_capabilities & IFCAP_RXCSUM) != 0) {
3486 			ifp->if_capenable ^= IFCAP_RXCSUM;
3487 			reinit = 1;
3488 		}
3489 		if ((mask & IFCAP_TSO4) != 0 &&
3490 		    (ifp->if_capabilities & IFCAP_TSO4) != 0) {
3491 			ifp->if_capenable ^= IFCAP_TSO4;
3492 			if ((IFCAP_TSO4 & ifp->if_capenable) != 0)
3493 				ifp->if_hwassist |= CSUM_TSO;
3494 			else
3495 				ifp->if_hwassist &= ~CSUM_TSO;
3496 			if (ifp->if_mtu > RL_TSO_MTU &&
3497 			    (ifp->if_capenable & IFCAP_TSO4) != 0) {
3498 				ifp->if_capenable &= ~IFCAP_TSO4;
3499 				ifp->if_hwassist &= ~CSUM_TSO;
3500 			}
3501 		}
3502 		if ((mask & IFCAP_VLAN_HWTSO) != 0 &&
3503 		    (ifp->if_capabilities & IFCAP_VLAN_HWTSO) != 0)
3504 			ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
3505 		if ((mask & IFCAP_VLAN_HWTAGGING) != 0 &&
3506 		    (ifp->if_capabilities & IFCAP_VLAN_HWTAGGING) != 0) {
3507 			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
3508 			/* TSO over VLAN requires VLAN hardware tagging. */
3509 			if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) == 0)
3510 				ifp->if_capenable &= ~IFCAP_VLAN_HWTSO;
3511 			reinit = 1;
3512 		}
3513 		if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0 &&
3514 		    (mask & (IFCAP_HWCSUM | IFCAP_TSO4 |
3515 		    IFCAP_VLAN_HWTSO)) != 0)
3516 				reinit = 1;
3517 		if ((mask & IFCAP_WOL) != 0 &&
3518 		    (ifp->if_capabilities & IFCAP_WOL) != 0) {
3519 			if ((mask & IFCAP_WOL_UCAST) != 0)
3520 				ifp->if_capenable ^= IFCAP_WOL_UCAST;
3521 			if ((mask & IFCAP_WOL_MCAST) != 0)
3522 				ifp->if_capenable ^= IFCAP_WOL_MCAST;
3523 			if ((mask & IFCAP_WOL_MAGIC) != 0)
3524 				ifp->if_capenable ^= IFCAP_WOL_MAGIC;
3525 		}
3526 		if (reinit && ifp->if_drv_flags & IFF_DRV_RUNNING) {
3527 			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3528 			re_init_locked(sc);
3529 		}
3530 		RL_UNLOCK(sc);
3531 		VLAN_CAPABILITIES(ifp);
3532 	    }
3533 		break;
3534 	default:
3535 		error = ether_ioctl(ifp, command, data);
3536 		break;
3537 	}
3538 
3539 	return (error);
3540 }
3541 
3542 static void
3543 re_watchdog(struct rl_softc *sc)
3544 {
3545 	struct ifnet		*ifp;
3546 
3547 	RL_LOCK_ASSERT(sc);
3548 
3549 	if (sc->rl_watchdog_timer == 0 || --sc->rl_watchdog_timer != 0)
3550 		return;
3551 
3552 	ifp = sc->rl_ifp;
3553 	re_txeof(sc);
3554 	if (sc->rl_ldata.rl_tx_free == sc->rl_ldata.rl_tx_desc_cnt) {
3555 		if_printf(ifp, "watchdog timeout (missed Tx interrupts) "
3556 		    "-- recovering\n");
3557 		if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
3558 			re_start_locked(ifp);
3559 		return;
3560 	}
3561 
3562 	if_printf(ifp, "watchdog timeout\n");
3563 	if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
3564 
3565 	re_rxeof(sc, NULL);
3566 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3567 	re_init_locked(sc);
3568 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
3569 		re_start_locked(ifp);
3570 }
3571 
3572 /*
3573  * Stop the adapter and free any mbufs allocated to the
3574  * RX and TX lists.
3575  */
3576 static void
3577 re_stop(struct rl_softc *sc)
3578 {
3579 	int			i;
3580 	struct ifnet		*ifp;
3581 	struct rl_txdesc	*txd;
3582 	struct rl_rxdesc	*rxd;
3583 
3584 	RL_LOCK_ASSERT(sc);
3585 
3586 	ifp = sc->rl_ifp;
3587 
3588 	sc->rl_watchdog_timer = 0;
3589 	callout_stop(&sc->rl_stat_callout);
3590 	ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
3591 
3592 	/*
3593 	 * Disable accepting frames to put RX MAC into idle state.
3594 	 * Otherwise it's possible to get frames while stop command
3595 	 * execution is in progress and controller can DMA the frame
3596 	 * to already freed RX buffer during that period.
3597 	 */
3598 	CSR_WRITE_4(sc, RL_RXCFG, CSR_READ_4(sc, RL_RXCFG) &
3599 	    ~(RL_RXCFG_RX_ALLPHYS | RL_RXCFG_RX_INDIV | RL_RXCFG_RX_MULTI |
3600 	    RL_RXCFG_RX_BROAD));
3601 
3602 	if ((sc->rl_flags & RL_FLAG_8168G_PLUS) != 0) {
3603 		/* Enable RXDV gate. */
3604 		CSR_WRITE_4(sc, RL_MISC, CSR_READ_4(sc, RL_MISC) |
3605 		    0x00080000);
3606 	}
3607 
3608 	if ((sc->rl_flags & RL_FLAG_WAIT_TXPOLL) != 0) {
3609 		for (i = RL_TIMEOUT; i > 0; i--) {
3610 			if ((CSR_READ_1(sc, sc->rl_txstart) &
3611 			    RL_TXSTART_START) == 0)
3612 				break;
3613 			DELAY(20);
3614 		}
3615 		if (i == 0)
3616 			device_printf(sc->rl_dev,
3617 			    "stopping TX poll timed out!\n");
3618 		CSR_WRITE_1(sc, RL_COMMAND, 0x00);
3619 	} else if ((sc->rl_flags & RL_FLAG_CMDSTOP) != 0) {
3620 		CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_STOPREQ | RL_CMD_TX_ENB |
3621 		    RL_CMD_RX_ENB);
3622 		if ((sc->rl_flags & RL_FLAG_CMDSTOP_WAIT_TXQ) != 0) {
3623 			for (i = RL_TIMEOUT; i > 0; i--) {
3624 				if ((CSR_READ_4(sc, RL_TXCFG) &
3625 				    RL_TXCFG_QUEUE_EMPTY) != 0)
3626 					break;
3627 				DELAY(100);
3628 			}
3629 			if (i == 0)
3630 				device_printf(sc->rl_dev,
3631 				   "stopping TXQ timed out!\n");
3632 		}
3633 	} else
3634 		CSR_WRITE_1(sc, RL_COMMAND, 0x00);
3635 	DELAY(1000);
3636 	CSR_WRITE_2(sc, RL_IMR, 0x0000);
3637 	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
3638 
3639 	if (sc->rl_head != NULL) {
3640 		m_freem(sc->rl_head);
3641 		sc->rl_head = sc->rl_tail = NULL;
3642 	}
3643 
3644 	/* Free the TX list buffers. */
3645 	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
3646 		txd = &sc->rl_ldata.rl_tx_desc[i];
3647 		if (txd->tx_m != NULL) {
3648 			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
3649 			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
3650 			bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag,
3651 			    txd->tx_dmamap);
3652 			m_freem(txd->tx_m);
3653 			txd->tx_m = NULL;
3654 		}
3655 	}
3656 
3657 	/* Free the RX list buffers. */
3658 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
3659 		rxd = &sc->rl_ldata.rl_rx_desc[i];
3660 		if (rxd->rx_m != NULL) {
3661 			bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag,
3662 			    rxd->rx_dmamap, BUS_DMASYNC_POSTREAD);
3663 			bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag,
3664 			    rxd->rx_dmamap);
3665 			m_freem(rxd->rx_m);
3666 			rxd->rx_m = NULL;
3667 		}
3668 	}
3669 
3670 	if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
3671 		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
3672 			rxd = &sc->rl_ldata.rl_jrx_desc[i];
3673 			if (rxd->rx_m != NULL) {
3674 				bus_dmamap_sync(sc->rl_ldata.rl_jrx_mtag,
3675 				    rxd->rx_dmamap, BUS_DMASYNC_POSTREAD);
3676 				bus_dmamap_unload(sc->rl_ldata.rl_jrx_mtag,
3677 				    rxd->rx_dmamap);
3678 				m_freem(rxd->rx_m);
3679 				rxd->rx_m = NULL;
3680 			}
3681 		}
3682 	}
3683 }
3684 
3685 /*
3686  * Device suspend routine.  Stop the interface and save some PCI
3687  * settings in case the BIOS doesn't restore them properly on
3688  * resume.
3689  */
3690 static int
3691 re_suspend(device_t dev)
3692 {
3693 	struct rl_softc		*sc;
3694 
3695 	sc = device_get_softc(dev);
3696 
3697 	RL_LOCK(sc);
3698 	re_stop(sc);
3699 	re_setwol(sc);
3700 	sc->suspended = 1;
3701 	RL_UNLOCK(sc);
3702 
3703 	return (0);
3704 }
3705 
3706 /*
3707  * Device resume routine.  Restore some PCI settings in case the BIOS
3708  * doesn't, re-enable busmastering, and restart the interface if
3709  * appropriate.
3710  */
3711 static int
3712 re_resume(device_t dev)
3713 {
3714 	struct rl_softc		*sc;
3715 	struct ifnet		*ifp;
3716 
3717 	sc = device_get_softc(dev);
3718 
3719 	RL_LOCK(sc);
3720 
3721 	ifp = sc->rl_ifp;
3722 	/* Take controller out of sleep mode. */
3723 	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
3724 		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
3725 			CSR_WRITE_1(sc, RL_GPIO,
3726 			    CSR_READ_1(sc, RL_GPIO) | 0x01);
3727 	}
3728 
3729 	/*
3730 	 * Clear WOL matching such that normal Rx filtering
3731 	 * wouldn't interfere with WOL patterns.
3732 	 */
3733 	re_clrwol(sc);
3734 
3735 	/* reinitialize interface if necessary */
3736 	if (ifp->if_flags & IFF_UP)
3737 		re_init_locked(sc);
3738 
3739 	sc->suspended = 0;
3740 	RL_UNLOCK(sc);
3741 
3742 	return (0);
3743 }
3744 
3745 /*
3746  * Stop all chip I/O so that the kernel's probe routines don't
3747  * get confused by errant DMAs when rebooting.
3748  */
3749 static int
3750 re_shutdown(device_t dev)
3751 {
3752 	struct rl_softc		*sc;
3753 
3754 	sc = device_get_softc(dev);
3755 
3756 	RL_LOCK(sc);
3757 	re_stop(sc);
3758 	/*
3759 	 * Mark interface as down since otherwise we will panic if
3760 	 * interrupt comes in later on, which can happen in some
3761 	 * cases.
3762 	 */
3763 	sc->rl_ifp->if_flags &= ~IFF_UP;
3764 	re_setwol(sc);
3765 	RL_UNLOCK(sc);
3766 
3767 	return (0);
3768 }
3769 
3770 static void
3771 re_set_linkspeed(struct rl_softc *sc)
3772 {
3773 	struct mii_softc *miisc;
3774 	struct mii_data *mii;
3775 	int aneg, i, phyno;
3776 
3777 	RL_LOCK_ASSERT(sc);
3778 
3779 	mii = device_get_softc(sc->rl_miibus);
3780 	mii_pollstat(mii);
3781 	aneg = 0;
3782 	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
3783 	    (IFM_ACTIVE | IFM_AVALID)) {
3784 		switch IFM_SUBTYPE(mii->mii_media_active) {
3785 		case IFM_10_T:
3786 		case IFM_100_TX:
3787 			return;
3788 		case IFM_1000_T:
3789 			aneg++;
3790 			break;
3791 		default:
3792 			break;
3793 		}
3794 	}
3795 	miisc = LIST_FIRST(&mii->mii_phys);
3796 	phyno = miisc->mii_phy;
3797 	LIST_FOREACH(miisc, &mii->mii_phys, mii_list)
3798 		PHY_RESET(miisc);
3799 	re_miibus_writereg(sc->rl_dev, phyno, MII_100T2CR, 0);
3800 	re_miibus_writereg(sc->rl_dev, phyno,
3801 	    MII_ANAR, ANAR_TX_FD | ANAR_TX | ANAR_10_FD | ANAR_10 | ANAR_CSMA);
3802 	re_miibus_writereg(sc->rl_dev, phyno,
3803 	    MII_BMCR, BMCR_AUTOEN | BMCR_STARTNEG);
3804 	DELAY(1000);
3805 	if (aneg != 0) {
3806 		/*
3807 		 * Poll link state until re(4) get a 10/100Mbps link.
3808 		 */
3809 		for (i = 0; i < MII_ANEGTICKS_GIGE; i++) {
3810 			mii_pollstat(mii);
3811 			if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID))
3812 			    == (IFM_ACTIVE | IFM_AVALID)) {
3813 				switch (IFM_SUBTYPE(mii->mii_media_active)) {
3814 				case IFM_10_T:
3815 				case IFM_100_TX:
3816 					return;
3817 				default:
3818 					break;
3819 				}
3820 			}
3821 			RL_UNLOCK(sc);
3822 			pause("relnk", hz);
3823 			RL_LOCK(sc);
3824 		}
3825 		if (i == MII_ANEGTICKS_GIGE)
3826 			device_printf(sc->rl_dev,
3827 			    "establishing a link failed, WOL may not work!");
3828 	}
3829 	/*
3830 	 * No link, force MAC to have 100Mbps, full-duplex link.
3831 	 * MAC does not require reprogramming on resolved speed/duplex,
3832 	 * so this is just for completeness.
3833 	 */
3834 	mii->mii_media_status = IFM_AVALID | IFM_ACTIVE;
3835 	mii->mii_media_active = IFM_ETHER | IFM_100_TX | IFM_FDX;
3836 }
3837 
3838 static void
3839 re_setwol(struct rl_softc *sc)
3840 {
3841 	struct ifnet		*ifp;
3842 	int			pmc;
3843 	uint16_t		pmstat;
3844 	uint8_t			v;
3845 
3846 	RL_LOCK_ASSERT(sc);
3847 
3848 	if (pci_find_cap(sc->rl_dev, PCIY_PMG, &pmc) != 0)
3849 		return;
3850 
3851 	ifp = sc->rl_ifp;
3852 	/* Put controller into sleep mode. */
3853 	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
3854 		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
3855 			CSR_WRITE_1(sc, RL_GPIO,
3856 			    CSR_READ_1(sc, RL_GPIO) & ~0x01);
3857 	}
3858 	if ((ifp->if_capenable & IFCAP_WOL) != 0) {
3859 		if ((sc->rl_flags & RL_FLAG_8168G_PLUS) != 0) {
3860 			/* Disable RXDV gate. */
3861 			CSR_WRITE_4(sc, RL_MISC, CSR_READ_4(sc, RL_MISC) &
3862 			    ~0x00080000);
3863 		}
3864 		re_set_rxmode(sc);
3865 		if ((sc->rl_flags & RL_FLAG_WOL_MANLINK) != 0)
3866 			re_set_linkspeed(sc);
3867 		if ((sc->rl_flags & RL_FLAG_WOLRXENB) != 0)
3868 			CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RX_ENB);
3869 	}
3870 	/* Enable config register write. */
3871 	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
3872 
3873 	/* Enable PME. */
3874 	v = CSR_READ_1(sc, sc->rl_cfg1);
3875 	v &= ~RL_CFG1_PME;
3876 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3877 		v |= RL_CFG1_PME;
3878 	CSR_WRITE_1(sc, sc->rl_cfg1, v);
3879 
3880 	v = CSR_READ_1(sc, sc->rl_cfg3);
3881 	v &= ~(RL_CFG3_WOL_LINK | RL_CFG3_WOL_MAGIC);
3882 	if ((ifp->if_capenable & IFCAP_WOL_MAGIC) != 0)
3883 		v |= RL_CFG3_WOL_MAGIC;
3884 	CSR_WRITE_1(sc, sc->rl_cfg3, v);
3885 
3886 	v = CSR_READ_1(sc, sc->rl_cfg5);
3887 	v &= ~(RL_CFG5_WOL_BCAST | RL_CFG5_WOL_MCAST | RL_CFG5_WOL_UCAST |
3888 	    RL_CFG5_WOL_LANWAKE);
3889 	if ((ifp->if_capenable & IFCAP_WOL_UCAST) != 0)
3890 		v |= RL_CFG5_WOL_UCAST;
3891 	if ((ifp->if_capenable & IFCAP_WOL_MCAST) != 0)
3892 		v |= RL_CFG5_WOL_MCAST | RL_CFG5_WOL_BCAST;
3893 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3894 		v |= RL_CFG5_WOL_LANWAKE;
3895 	CSR_WRITE_1(sc, sc->rl_cfg5, v);
3896 
3897 	/* Config register write done. */
3898 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3899 
3900 	if ((ifp->if_capenable & IFCAP_WOL) == 0 &&
3901 	    (sc->rl_flags & RL_FLAG_PHYWAKE_PM) != 0)
3902 		CSR_WRITE_1(sc, RL_PMCH, CSR_READ_1(sc, RL_PMCH) & ~0x80);
3903 	/*
3904 	 * It seems that hardware resets its link speed to 100Mbps in
3905 	 * power down mode so switching to 100Mbps in driver is not
3906 	 * needed.
3907 	 */
3908 
3909 	/* Request PME if WOL is requested. */
3910 	pmstat = pci_read_config(sc->rl_dev, pmc + PCIR_POWER_STATUS, 2);
3911 	pmstat &= ~(PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE);
3912 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3913 		pmstat |= PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE;
3914 	pci_write_config(sc->rl_dev, pmc + PCIR_POWER_STATUS, pmstat, 2);
3915 }
3916 
3917 static void
3918 re_clrwol(struct rl_softc *sc)
3919 {
3920 	int			pmc;
3921 	uint8_t			v;
3922 
3923 	RL_LOCK_ASSERT(sc);
3924 
3925 	if (pci_find_cap(sc->rl_dev, PCIY_PMG, &pmc) != 0)
3926 		return;
3927 
3928 	/* Enable config register write. */
3929 	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
3930 
3931 	v = CSR_READ_1(sc, sc->rl_cfg3);
3932 	v &= ~(RL_CFG3_WOL_LINK | RL_CFG3_WOL_MAGIC);
3933 	CSR_WRITE_1(sc, sc->rl_cfg3, v);
3934 
3935 	/* Config register write done. */
3936 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3937 
3938 	v = CSR_READ_1(sc, sc->rl_cfg5);
3939 	v &= ~(RL_CFG5_WOL_BCAST | RL_CFG5_WOL_MCAST | RL_CFG5_WOL_UCAST);
3940 	v &= ~RL_CFG5_WOL_LANWAKE;
3941 	CSR_WRITE_1(sc, sc->rl_cfg5, v);
3942 }
3943 
3944 static void
3945 re_add_sysctls(struct rl_softc *sc)
3946 {
3947 	struct sysctl_ctx_list	*ctx;
3948 	struct sysctl_oid_list	*children;
3949 	int			error;
3950 
3951 	ctx = device_get_sysctl_ctx(sc->rl_dev);
3952 	children = SYSCTL_CHILDREN(device_get_sysctl_tree(sc->rl_dev));
3953 
3954 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "stats",
3955 	    CTLTYPE_INT | CTLFLAG_RW, sc, 0, re_sysctl_stats, "I",
3956 	    "Statistics Information");
3957 	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) == 0)
3958 		return;
3959 
3960 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "int_rx_mod",
3961 	    CTLTYPE_INT | CTLFLAG_RW, &sc->rl_int_rx_mod, 0,
3962 	    sysctl_hw_re_int_mod, "I", "re RX interrupt moderation");
3963 	/* Pull in device tunables. */
3964 	sc->rl_int_rx_mod = RL_TIMER_DEFAULT;
3965 	error = resource_int_value(device_get_name(sc->rl_dev),
3966 	    device_get_unit(sc->rl_dev), "int_rx_mod", &sc->rl_int_rx_mod);
3967 	if (error == 0) {
3968 		if (sc->rl_int_rx_mod < RL_TIMER_MIN ||
3969 		    sc->rl_int_rx_mod > RL_TIMER_MAX) {
3970 			device_printf(sc->rl_dev, "int_rx_mod value out of "
3971 			    "range; using default: %d\n",
3972 			    RL_TIMER_DEFAULT);
3973 			sc->rl_int_rx_mod = RL_TIMER_DEFAULT;
3974 		}
3975 	}
3976 }
3977 
3978 static int
3979 re_sysctl_stats(SYSCTL_HANDLER_ARGS)
3980 {
3981 	struct rl_softc		*sc;
3982 	struct rl_stats		*stats;
3983 	int			error, i, result;
3984 
3985 	result = -1;
3986 	error = sysctl_handle_int(oidp, &result, 0, req);
3987 	if (error || req->newptr == NULL)
3988 		return (error);
3989 
3990 	if (result == 1) {
3991 		sc = (struct rl_softc *)arg1;
3992 		RL_LOCK(sc);
3993 		if ((sc->rl_ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
3994 			RL_UNLOCK(sc);
3995 			goto done;
3996 		}
3997 		bus_dmamap_sync(sc->rl_ldata.rl_stag,
3998 		    sc->rl_ldata.rl_smap, BUS_DMASYNC_PREREAD);
3999 		CSR_WRITE_4(sc, RL_DUMPSTATS_HI,
4000 		    RL_ADDR_HI(sc->rl_ldata.rl_stats_addr));
4001 		CSR_WRITE_4(sc, RL_DUMPSTATS_LO,
4002 		    RL_ADDR_LO(sc->rl_ldata.rl_stats_addr));
4003 		CSR_WRITE_4(sc, RL_DUMPSTATS_LO,
4004 		    RL_ADDR_LO(sc->rl_ldata.rl_stats_addr |
4005 		    RL_DUMPSTATS_START));
4006 		for (i = RL_TIMEOUT; i > 0; i--) {
4007 			if ((CSR_READ_4(sc, RL_DUMPSTATS_LO) &
4008 			    RL_DUMPSTATS_START) == 0)
4009 				break;
4010 			DELAY(1000);
4011 		}
4012 		bus_dmamap_sync(sc->rl_ldata.rl_stag,
4013 		    sc->rl_ldata.rl_smap, BUS_DMASYNC_POSTREAD);
4014 		RL_UNLOCK(sc);
4015 		if (i == 0) {
4016 			device_printf(sc->rl_dev,
4017 			    "DUMP statistics request timed out\n");
4018 			return (ETIMEDOUT);
4019 		}
4020 done:
4021 		stats = sc->rl_ldata.rl_stats;
4022 		printf("%s statistics:\n", device_get_nameunit(sc->rl_dev));
4023 		printf("Tx frames : %ju\n",
4024 		    (uintmax_t)le64toh(stats->rl_tx_pkts));
4025 		printf("Rx frames : %ju\n",
4026 		    (uintmax_t)le64toh(stats->rl_rx_pkts));
4027 		printf("Tx errors : %ju\n",
4028 		    (uintmax_t)le64toh(stats->rl_tx_errs));
4029 		printf("Rx errors : %u\n",
4030 		    le32toh(stats->rl_rx_errs));
4031 		printf("Rx missed frames : %u\n",
4032 		    (uint32_t)le16toh(stats->rl_missed_pkts));
4033 		printf("Rx frame alignment errs : %u\n",
4034 		    (uint32_t)le16toh(stats->rl_rx_framealign_errs));
4035 		printf("Tx single collisions : %u\n",
4036 		    le32toh(stats->rl_tx_onecoll));
4037 		printf("Tx multiple collisions : %u\n",
4038 		    le32toh(stats->rl_tx_multicolls));
4039 		printf("Rx unicast frames : %ju\n",
4040 		    (uintmax_t)le64toh(stats->rl_rx_ucasts));
4041 		printf("Rx broadcast frames : %ju\n",
4042 		    (uintmax_t)le64toh(stats->rl_rx_bcasts));
4043 		printf("Rx multicast frames : %u\n",
4044 		    le32toh(stats->rl_rx_mcasts));
4045 		printf("Tx aborts : %u\n",
4046 		    (uint32_t)le16toh(stats->rl_tx_aborts));
4047 		printf("Tx underruns : %u\n",
4048 		    (uint32_t)le16toh(stats->rl_rx_underruns));
4049 	}
4050 
4051 	return (error);
4052 }
4053 
4054 static int
4055 sysctl_int_range(SYSCTL_HANDLER_ARGS, int low, int high)
4056 {
4057 	int error, value;
4058 
4059 	if (arg1 == NULL)
4060 		return (EINVAL);
4061 	value = *(int *)arg1;
4062 	error = sysctl_handle_int(oidp, &value, 0, req);
4063 	if (error || req->newptr == NULL)
4064 		return (error);
4065 	if (value < low || value > high)
4066 		return (EINVAL);
4067 	*(int *)arg1 = value;
4068 
4069 	return (0);
4070 }
4071 
4072 static int
4073 sysctl_hw_re_int_mod(SYSCTL_HANDLER_ARGS)
4074 {
4075 
4076 	return (sysctl_int_range(oidp, arg1, arg2, req, RL_TIMER_MIN,
4077 	    RL_TIMER_MAX));
4078 }
4079