xref: /freebsd/sys/dev/re/if_re.c (revision 10b59a9b4add0320d52c15ce057dd697261e7dfc)
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_arp.h>
131 #include <net/ethernet.h>
132 #include <net/if_dl.h>
133 #include <net/if_media.h>
134 #include <net/if_types.h>
135 #include <net/if_vlan_var.h>
136 
137 #include <net/bpf.h>
138 
139 #include <machine/bus.h>
140 #include <machine/resource.h>
141 #include <sys/bus.h>
142 #include <sys/rman.h>
143 
144 #include <dev/mii/mii.h>
145 #include <dev/mii/miivar.h>
146 
147 #include <dev/pci/pcireg.h>
148 #include <dev/pci/pcivar.h>
149 
150 #include <pci/if_rlreg.h>
151 
152 MODULE_DEPEND(re, pci, 1, 1, 1);
153 MODULE_DEPEND(re, ether, 1, 1, 1);
154 MODULE_DEPEND(re, miibus, 1, 1, 1);
155 
156 /* "device miibus" required.  See GENERIC if you get errors here. */
157 #include "miibus_if.h"
158 
159 /* Tunables. */
160 static int intr_filter = 0;
161 TUNABLE_INT("hw.re.intr_filter", &intr_filter);
162 static int msi_disable = 0;
163 TUNABLE_INT("hw.re.msi_disable", &msi_disable);
164 static int msix_disable = 0;
165 TUNABLE_INT("hw.re.msix_disable", &msix_disable);
166 static int prefer_iomap = 0;
167 TUNABLE_INT("hw.re.prefer_iomap", &prefer_iomap);
168 
169 #define RE_CSUM_FEATURES    (CSUM_IP | CSUM_TCP | CSUM_UDP)
170 
171 /*
172  * Various supported device vendors/types and their names.
173  */
174 static const struct rl_type const re_devs[] = {
175 	{ DLINK_VENDORID, DLINK_DEVICEID_528T, 0,
176 	    "D-Link DGE-528(T) Gigabit Ethernet Adapter" },
177 	{ DLINK_VENDORID, DLINK_DEVICEID_530T_REVC, 0,
178 	    "D-Link DGE-530(T) Gigabit Ethernet Adapter" },
179 	{ RT_VENDORID, RT_DEVICEID_8139, 0,
180 	    "RealTek 8139C+ 10/100BaseTX" },
181 	{ RT_VENDORID, RT_DEVICEID_8101E, 0,
182 	    "RealTek 810xE PCIe 10/100baseTX" },
183 	{ RT_VENDORID, RT_DEVICEID_8168, 0,
184 	    "RealTek 8168/8111 B/C/CP/D/DP/E/F PCIe Gigabit Ethernet" },
185 	{ RT_VENDORID, RT_DEVICEID_8169, 0,
186 	    "RealTek 8169/8169S/8169SB(L)/8110S/8110SB(L) Gigabit Ethernet" },
187 	{ RT_VENDORID, RT_DEVICEID_8169SC, 0,
188 	    "RealTek 8169SC/8110SC Single-chip Gigabit Ethernet" },
189 	{ COREGA_VENDORID, COREGA_DEVICEID_CGLAPCIGT, 0,
190 	    "Corega CG-LAPCIGT (RTL8169S) Gigabit Ethernet" },
191 	{ LINKSYS_VENDORID, LINKSYS_DEVICEID_EG1032, 0,
192 	    "Linksys EG1032 (RTL8169S) Gigabit Ethernet" },
193 	{ USR_VENDORID, USR_DEVICEID_997902, 0,
194 	    "US Robotics 997902 (RTL8169S) Gigabit Ethernet" }
195 };
196 
197 static const struct rl_hwrev const re_hwrevs[] = {
198 	{ RL_HWREV_8139, RL_8139, "", RL_MTU },
199 	{ RL_HWREV_8139A, RL_8139, "A", RL_MTU },
200 	{ RL_HWREV_8139AG, RL_8139, "A-G", RL_MTU },
201 	{ RL_HWREV_8139B, RL_8139, "B", RL_MTU },
202 	{ RL_HWREV_8130, RL_8139, "8130", RL_MTU },
203 	{ RL_HWREV_8139C, RL_8139, "C", RL_MTU },
204 	{ RL_HWREV_8139D, RL_8139, "8139D/8100B/8100C", RL_MTU },
205 	{ RL_HWREV_8139CPLUS, RL_8139CPLUS, "C+", RL_MTU },
206 	{ RL_HWREV_8168B_SPIN1, RL_8169, "8168", RL_JUMBO_MTU },
207 	{ RL_HWREV_8169, RL_8169, "8169", RL_JUMBO_MTU },
208 	{ RL_HWREV_8169S, RL_8169, "8169S", RL_JUMBO_MTU },
209 	{ RL_HWREV_8110S, RL_8169, "8110S", RL_JUMBO_MTU },
210 	{ RL_HWREV_8169_8110SB, RL_8169, "8169SB/8110SB", RL_JUMBO_MTU },
211 	{ RL_HWREV_8169_8110SC, RL_8169, "8169SC/8110SC", RL_JUMBO_MTU },
212 	{ RL_HWREV_8169_8110SBL, RL_8169, "8169SBL/8110SBL", RL_JUMBO_MTU },
213 	{ RL_HWREV_8169_8110SCE, RL_8169, "8169SC/8110SC", RL_JUMBO_MTU },
214 	{ RL_HWREV_8100, RL_8139, "8100", RL_MTU },
215 	{ RL_HWREV_8101, RL_8139, "8101", RL_MTU },
216 	{ RL_HWREV_8100E, RL_8169, "8100E", RL_MTU },
217 	{ RL_HWREV_8101E, RL_8169, "8101E", RL_MTU },
218 	{ RL_HWREV_8102E, RL_8169, "8102E", RL_MTU },
219 	{ RL_HWREV_8102EL, RL_8169, "8102EL", RL_MTU },
220 	{ RL_HWREV_8102EL_SPIN1, RL_8169, "8102EL", RL_MTU },
221 	{ RL_HWREV_8103E, RL_8169, "8103E", RL_MTU },
222 	{ RL_HWREV_8401E, RL_8169, "8401E", RL_MTU },
223 	{ RL_HWREV_8402, RL_8169, "8402", RL_MTU },
224 	{ RL_HWREV_8105E, RL_8169, "8105E", RL_MTU },
225 	{ RL_HWREV_8105E_SPIN1, RL_8169, "8105E", RL_MTU },
226 	{ RL_HWREV_8168B_SPIN2, RL_8169, "8168", RL_JUMBO_MTU },
227 	{ RL_HWREV_8168B_SPIN3, RL_8169, "8168", RL_JUMBO_MTU },
228 	{ RL_HWREV_8168C, RL_8169, "8168C/8111C", RL_JUMBO_MTU_6K },
229 	{ RL_HWREV_8168C_SPIN2, RL_8169, "8168C/8111C", RL_JUMBO_MTU_6K },
230 	{ RL_HWREV_8168CP, RL_8169, "8168CP/8111CP", RL_JUMBO_MTU_6K },
231 	{ RL_HWREV_8168D, RL_8169, "8168D/8111D", RL_JUMBO_MTU_9K },
232 	{ RL_HWREV_8168DP, RL_8169, "8168DP/8111DP", RL_JUMBO_MTU_9K },
233 	{ RL_HWREV_8168E, RL_8169, "8168E/8111E", RL_JUMBO_MTU_9K},
234 	{ RL_HWREV_8168E_VL, RL_8169, "8168E/8111E-VL", RL_JUMBO_MTU_6K},
235 	{ RL_HWREV_8168F, RL_8169, "8168F/8111F", RL_JUMBO_MTU_9K},
236 	{ RL_HWREV_8411, RL_8169, "8411", RL_JUMBO_MTU_9K},
237 	{ 0, 0, NULL, 0 }
238 };
239 
240 static int re_probe		(device_t);
241 static int re_attach		(device_t);
242 static int re_detach		(device_t);
243 
244 static int re_encap		(struct rl_softc *, struct mbuf **);
245 
246 static void re_dma_map_addr	(void *, bus_dma_segment_t *, int, int);
247 static int re_allocmem		(device_t, struct rl_softc *);
248 static __inline void re_discard_rxbuf
249 				(struct rl_softc *, int);
250 static int re_newbuf		(struct rl_softc *, int);
251 static int re_jumbo_newbuf	(struct rl_softc *, int);
252 static int re_rx_list_init	(struct rl_softc *);
253 static int re_jrx_list_init	(struct rl_softc *);
254 static int re_tx_list_init	(struct rl_softc *);
255 #ifdef RE_FIXUP_RX
256 static __inline void re_fixup_rx
257 				(struct mbuf *);
258 #endif
259 static int re_rxeof		(struct rl_softc *, int *);
260 static void re_txeof		(struct rl_softc *);
261 #ifdef DEVICE_POLLING
262 static int re_poll		(struct ifnet *, enum poll_cmd, int);
263 static int re_poll_locked	(struct ifnet *, enum poll_cmd, int);
264 #endif
265 static int re_intr		(void *);
266 static void re_intr_msi		(void *);
267 static void re_tick		(void *);
268 static void re_int_task		(void *, int);
269 static void re_start		(struct ifnet *);
270 static void re_start_locked	(struct ifnet *);
271 static int re_ioctl		(struct ifnet *, u_long, caddr_t);
272 static void re_init		(void *);
273 static void re_init_locked	(struct rl_softc *);
274 static void re_stop		(struct rl_softc *);
275 static void re_watchdog		(struct rl_softc *);
276 static int re_suspend		(device_t);
277 static int re_resume		(device_t);
278 static int re_shutdown		(device_t);
279 static int re_ifmedia_upd	(struct ifnet *);
280 static void re_ifmedia_sts	(struct ifnet *, struct ifmediareq *);
281 
282 static void re_eeprom_putbyte	(struct rl_softc *, int);
283 static void re_eeprom_getword	(struct rl_softc *, int, u_int16_t *);
284 static void re_read_eeprom	(struct rl_softc *, caddr_t, int, int);
285 static int re_gmii_readreg	(device_t, int, int);
286 static int re_gmii_writereg	(device_t, int, int, int);
287 
288 static int re_miibus_readreg	(device_t, int, int);
289 static int re_miibus_writereg	(device_t, int, int, int);
290 static void re_miibus_statchg	(device_t);
291 
292 static void re_set_jumbo	(struct rl_softc *, int);
293 static void re_set_rxmode		(struct rl_softc *);
294 static void re_reset		(struct rl_softc *);
295 static void re_setwol		(struct rl_softc *);
296 static void re_clrwol		(struct rl_softc *);
297 
298 #ifdef RE_DIAG
299 static int re_diag		(struct rl_softc *);
300 #endif
301 
302 static void re_add_sysctls	(struct rl_softc *);
303 static int re_sysctl_stats	(SYSCTL_HANDLER_ARGS);
304 static int sysctl_int_range	(SYSCTL_HANDLER_ARGS, int, int);
305 static int sysctl_hw_re_int_mod	(SYSCTL_HANDLER_ARGS);
306 
307 static device_method_t re_methods[] = {
308 	/* Device interface */
309 	DEVMETHOD(device_probe,		re_probe),
310 	DEVMETHOD(device_attach,	re_attach),
311 	DEVMETHOD(device_detach,	re_detach),
312 	DEVMETHOD(device_suspend,	re_suspend),
313 	DEVMETHOD(device_resume,	re_resume),
314 	DEVMETHOD(device_shutdown,	re_shutdown),
315 
316 	/* bus interface */
317 	DEVMETHOD(bus_print_child,	bus_generic_print_child),
318 	DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
319 
320 	/* MII interface */
321 	DEVMETHOD(miibus_readreg,	re_miibus_readreg),
322 	DEVMETHOD(miibus_writereg,	re_miibus_writereg),
323 	DEVMETHOD(miibus_statchg,	re_miibus_statchg),
324 
325 	{ 0, 0 }
326 };
327 
328 static driver_t re_driver = {
329 	"re",
330 	re_methods,
331 	sizeof(struct rl_softc)
332 };
333 
334 static devclass_t re_devclass;
335 
336 DRIVER_MODULE(re, pci, re_driver, re_devclass, 0, 0);
337 DRIVER_MODULE(miibus, re, miibus_driver, miibus_devclass, 0, 0);
338 
339 #define EE_SET(x)					\
340 	CSR_WRITE_1(sc, RL_EECMD,			\
341 		CSR_READ_1(sc, RL_EECMD) | x)
342 
343 #define EE_CLR(x)					\
344 	CSR_WRITE_1(sc, RL_EECMD,			\
345 		CSR_READ_1(sc, RL_EECMD) & ~x)
346 
347 /*
348  * Send a read command and address to the EEPROM, check for ACK.
349  */
350 static void
351 re_eeprom_putbyte(struct rl_softc *sc, int addr)
352 {
353 	int			d, i;
354 
355 	d = addr | (RL_9346_READ << sc->rl_eewidth);
356 
357 	/*
358 	 * Feed in each bit and strobe the clock.
359 	 */
360 
361 	for (i = 1 << (sc->rl_eewidth + 3); i; i >>= 1) {
362 		if (d & i) {
363 			EE_SET(RL_EE_DATAIN);
364 		} else {
365 			EE_CLR(RL_EE_DATAIN);
366 		}
367 		DELAY(100);
368 		EE_SET(RL_EE_CLK);
369 		DELAY(150);
370 		EE_CLR(RL_EE_CLK);
371 		DELAY(100);
372 	}
373 }
374 
375 /*
376  * Read a word of data stored in the EEPROM at address 'addr.'
377  */
378 static void
379 re_eeprom_getword(struct rl_softc *sc, int addr, u_int16_t *dest)
380 {
381 	int			i;
382 	u_int16_t		word = 0;
383 
384 	/*
385 	 * Send address of word we want to read.
386 	 */
387 	re_eeprom_putbyte(sc, addr);
388 
389 	/*
390 	 * Start reading bits from EEPROM.
391 	 */
392 	for (i = 0x8000; i; i >>= 1) {
393 		EE_SET(RL_EE_CLK);
394 		DELAY(100);
395 		if (CSR_READ_1(sc, RL_EECMD) & RL_EE_DATAOUT)
396 			word |= i;
397 		EE_CLR(RL_EE_CLK);
398 		DELAY(100);
399 	}
400 
401 	*dest = word;
402 }
403 
404 /*
405  * Read a sequence of words from the EEPROM.
406  */
407 static void
408 re_read_eeprom(struct rl_softc *sc, caddr_t dest, int off, int cnt)
409 {
410 	int			i;
411 	u_int16_t		word = 0, *ptr;
412 
413 	CSR_SETBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM);
414 
415         DELAY(100);
416 
417 	for (i = 0; i < cnt; i++) {
418 		CSR_SETBIT_1(sc, RL_EECMD, RL_EE_SEL);
419 		re_eeprom_getword(sc, off + i, &word);
420 		CSR_CLRBIT_1(sc, RL_EECMD, RL_EE_SEL);
421 		ptr = (u_int16_t *)(dest + (i * 2));
422                 *ptr = word;
423 	}
424 
425 	CSR_CLRBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM);
426 }
427 
428 static int
429 re_gmii_readreg(device_t dev, int phy, int reg)
430 {
431 	struct rl_softc		*sc;
432 	u_int32_t		rval;
433 	int			i;
434 
435 	sc = device_get_softc(dev);
436 
437 	/* Let the rgephy driver read the GMEDIASTAT register */
438 
439 	if (reg == RL_GMEDIASTAT) {
440 		rval = CSR_READ_1(sc, RL_GMEDIASTAT);
441 		return (rval);
442 	}
443 
444 	CSR_WRITE_4(sc, RL_PHYAR, reg << 16);
445 
446 	for (i = 0; i < RL_PHY_TIMEOUT; i++) {
447 		rval = CSR_READ_4(sc, RL_PHYAR);
448 		if (rval & RL_PHYAR_BUSY)
449 			break;
450 		DELAY(25);
451 	}
452 
453 	if (i == RL_PHY_TIMEOUT) {
454 		device_printf(sc->rl_dev, "PHY read failed\n");
455 		return (0);
456 	}
457 
458 	/*
459 	 * Controller requires a 20us delay to process next MDIO request.
460 	 */
461 	DELAY(20);
462 
463 	return (rval & RL_PHYAR_PHYDATA);
464 }
465 
466 static int
467 re_gmii_writereg(device_t dev, int phy, int reg, int data)
468 {
469 	struct rl_softc		*sc;
470 	u_int32_t		rval;
471 	int			i;
472 
473 	sc = device_get_softc(dev);
474 
475 	CSR_WRITE_4(sc, RL_PHYAR, (reg << 16) |
476 	    (data & RL_PHYAR_PHYDATA) | RL_PHYAR_BUSY);
477 
478 	for (i = 0; i < RL_PHY_TIMEOUT; i++) {
479 		rval = CSR_READ_4(sc, RL_PHYAR);
480 		if (!(rval & RL_PHYAR_BUSY))
481 			break;
482 		DELAY(25);
483 	}
484 
485 	if (i == RL_PHY_TIMEOUT) {
486 		device_printf(sc->rl_dev, "PHY write failed\n");
487 		return (0);
488 	}
489 
490 	/*
491 	 * Controller requires a 20us delay to process next MDIO request.
492 	 */
493 	DELAY(20);
494 
495 	return (0);
496 }
497 
498 static int
499 re_miibus_readreg(device_t dev, int phy, int reg)
500 {
501 	struct rl_softc		*sc;
502 	u_int16_t		rval = 0;
503 	u_int16_t		re8139_reg = 0;
504 
505 	sc = device_get_softc(dev);
506 
507 	if (sc->rl_type == RL_8169) {
508 		rval = re_gmii_readreg(dev, phy, reg);
509 		return (rval);
510 	}
511 
512 	switch (reg) {
513 	case MII_BMCR:
514 		re8139_reg = RL_BMCR;
515 		break;
516 	case MII_BMSR:
517 		re8139_reg = RL_BMSR;
518 		break;
519 	case MII_ANAR:
520 		re8139_reg = RL_ANAR;
521 		break;
522 	case MII_ANER:
523 		re8139_reg = RL_ANER;
524 		break;
525 	case MII_ANLPAR:
526 		re8139_reg = RL_LPAR;
527 		break;
528 	case MII_PHYIDR1:
529 	case MII_PHYIDR2:
530 		return (0);
531 	/*
532 	 * Allow the rlphy driver to read the media status
533 	 * register. If we have a link partner which does not
534 	 * support NWAY, this is the register which will tell
535 	 * us the results of parallel detection.
536 	 */
537 	case RL_MEDIASTAT:
538 		rval = CSR_READ_1(sc, RL_MEDIASTAT);
539 		return (rval);
540 	default:
541 		device_printf(sc->rl_dev, "bad phy register\n");
542 		return (0);
543 	}
544 	rval = CSR_READ_2(sc, re8139_reg);
545 	if (sc->rl_type == RL_8139CPLUS && re8139_reg == RL_BMCR) {
546 		/* 8139C+ has different bit layout. */
547 		rval &= ~(BMCR_LOOP | BMCR_ISO);
548 	}
549 	return (rval);
550 }
551 
552 static int
553 re_miibus_writereg(device_t dev, int phy, int reg, int data)
554 {
555 	struct rl_softc		*sc;
556 	u_int16_t		re8139_reg = 0;
557 	int			rval = 0;
558 
559 	sc = device_get_softc(dev);
560 
561 	if (sc->rl_type == RL_8169) {
562 		rval = re_gmii_writereg(dev, phy, reg, data);
563 		return (rval);
564 	}
565 
566 	switch (reg) {
567 	case MII_BMCR:
568 		re8139_reg = RL_BMCR;
569 		if (sc->rl_type == RL_8139CPLUS) {
570 			/* 8139C+ has different bit layout. */
571 			data &= ~(BMCR_LOOP | BMCR_ISO);
572 		}
573 		break;
574 	case MII_BMSR:
575 		re8139_reg = RL_BMSR;
576 		break;
577 	case MII_ANAR:
578 		re8139_reg = RL_ANAR;
579 		break;
580 	case MII_ANER:
581 		re8139_reg = RL_ANER;
582 		break;
583 	case MII_ANLPAR:
584 		re8139_reg = RL_LPAR;
585 		break;
586 	case MII_PHYIDR1:
587 	case MII_PHYIDR2:
588 		return (0);
589 		break;
590 	default:
591 		device_printf(sc->rl_dev, "bad phy register\n");
592 		return (0);
593 	}
594 	CSR_WRITE_2(sc, re8139_reg, data);
595 	return (0);
596 }
597 
598 static void
599 re_miibus_statchg(device_t dev)
600 {
601 	struct rl_softc		*sc;
602 	struct ifnet		*ifp;
603 	struct mii_data		*mii;
604 
605 	sc = device_get_softc(dev);
606 	mii = device_get_softc(sc->rl_miibus);
607 	ifp = sc->rl_ifp;
608 	if (mii == NULL || ifp == NULL ||
609 	    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
610 		return;
611 
612 	sc->rl_flags &= ~RL_FLAG_LINK;
613 	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
614 	    (IFM_ACTIVE | IFM_AVALID)) {
615 		switch (IFM_SUBTYPE(mii->mii_media_active)) {
616 		case IFM_10_T:
617 		case IFM_100_TX:
618 			sc->rl_flags |= RL_FLAG_LINK;
619 			break;
620 		case IFM_1000_T:
621 			if ((sc->rl_flags & RL_FLAG_FASTETHER) != 0)
622 				break;
623 			sc->rl_flags |= RL_FLAG_LINK;
624 			break;
625 		default:
626 			break;
627 		}
628 	}
629 	/*
630 	 * RealTek controllers does not provide any interface to
631 	 * Tx/Rx MACs for resolved speed, duplex and flow-control
632 	 * parameters.
633 	 */
634 }
635 
636 /*
637  * Set the RX configuration and 64-bit multicast hash filter.
638  */
639 static void
640 re_set_rxmode(struct rl_softc *sc)
641 {
642 	struct ifnet		*ifp;
643 	struct ifmultiaddr	*ifma;
644 	uint32_t		hashes[2] = { 0, 0 };
645 	uint32_t		h, rxfilt;
646 
647 	RL_LOCK_ASSERT(sc);
648 
649 	ifp = sc->rl_ifp;
650 
651 	rxfilt = RL_RXCFG_CONFIG | RL_RXCFG_RX_INDIV | RL_RXCFG_RX_BROAD;
652 
653 	if (ifp->if_flags & (IFF_ALLMULTI | IFF_PROMISC)) {
654 		if (ifp->if_flags & IFF_PROMISC)
655 			rxfilt |= RL_RXCFG_RX_ALLPHYS;
656 		/*
657 		 * Unlike other hardwares, we have to explicitly set
658 		 * RL_RXCFG_RX_MULTI to receive multicast frames in
659 		 * promiscuous mode.
660 		 */
661 		rxfilt |= RL_RXCFG_RX_MULTI;
662 		hashes[0] = hashes[1] = 0xffffffff;
663 		goto done;
664 	}
665 
666 	if_maddr_rlock(ifp);
667 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
668 		if (ifma->ifma_addr->sa_family != AF_LINK)
669 			continue;
670 		h = ether_crc32_be(LLADDR((struct sockaddr_dl *)
671 		    ifma->ifma_addr), ETHER_ADDR_LEN) >> 26;
672 		if (h < 32)
673 			hashes[0] |= (1 << h);
674 		else
675 			hashes[1] |= (1 << (h - 32));
676 	}
677 	if_maddr_runlock(ifp);
678 
679 	if (hashes[0] != 0 || hashes[1] != 0) {
680 		/*
681 		 * For some unfathomable reason, RealTek decided to
682 		 * reverse the order of the multicast hash registers
683 		 * in the PCI Express parts.  This means we have to
684 		 * write the hash pattern in reverse order for those
685 		 * devices.
686 		 */
687 		if ((sc->rl_flags & RL_FLAG_PCIE) != 0) {
688 			h = bswap32(hashes[0]);
689 			hashes[0] = bswap32(hashes[1]);
690 			hashes[1] = h;
691 		}
692 		rxfilt |= RL_RXCFG_RX_MULTI;
693 	}
694 
695 done:
696 	CSR_WRITE_4(sc, RL_MAR0, hashes[0]);
697 	CSR_WRITE_4(sc, RL_MAR4, hashes[1]);
698 	CSR_WRITE_4(sc, RL_RXCFG, rxfilt);
699 }
700 
701 static void
702 re_reset(struct rl_softc *sc)
703 {
704 	int			i;
705 
706 	RL_LOCK_ASSERT(sc);
707 
708 	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RESET);
709 
710 	for (i = 0; i < RL_TIMEOUT; i++) {
711 		DELAY(10);
712 		if (!(CSR_READ_1(sc, RL_COMMAND) & RL_CMD_RESET))
713 			break;
714 	}
715 	if (i == RL_TIMEOUT)
716 		device_printf(sc->rl_dev, "reset never completed!\n");
717 
718 	if ((sc->rl_flags & RL_FLAG_MACRESET) != 0)
719 		CSR_WRITE_1(sc, 0x82, 1);
720 	if (sc->rl_hwrev->rl_rev == RL_HWREV_8169S)
721 		re_gmii_writereg(sc->rl_dev, 1, 0x0b, 0);
722 }
723 
724 #ifdef RE_DIAG
725 
726 /*
727  * The following routine is designed to test for a defect on some
728  * 32-bit 8169 cards. Some of these NICs have the REQ64# and ACK64#
729  * lines connected to the bus, however for a 32-bit only card, they
730  * should be pulled high. The result of this defect is that the
731  * NIC will not work right if you plug it into a 64-bit slot: DMA
732  * operations will be done with 64-bit transfers, which will fail
733  * because the 64-bit data lines aren't connected.
734  *
735  * There's no way to work around this (short of talking a soldering
736  * iron to the board), however we can detect it. The method we use
737  * here is to put the NIC into digital loopback mode, set the receiver
738  * to promiscuous mode, and then try to send a frame. We then compare
739  * the frame data we sent to what was received. If the data matches,
740  * then the NIC is working correctly, otherwise we know the user has
741  * a defective NIC which has been mistakenly plugged into a 64-bit PCI
742  * slot. In the latter case, there's no way the NIC can work correctly,
743  * so we print out a message on the console and abort the device attach.
744  */
745 
746 static int
747 re_diag(struct rl_softc *sc)
748 {
749 	struct ifnet		*ifp = sc->rl_ifp;
750 	struct mbuf		*m0;
751 	struct ether_header	*eh;
752 	struct rl_desc		*cur_rx;
753 	u_int16_t		status;
754 	u_int32_t		rxstat;
755 	int			total_len, i, error = 0, phyaddr;
756 	u_int8_t		dst[] = { 0x00, 'h', 'e', 'l', 'l', 'o' };
757 	u_int8_t		src[] = { 0x00, 'w', 'o', 'r', 'l', 'd' };
758 
759 	/* Allocate a single mbuf */
760 	MGETHDR(m0, M_DONTWAIT, MT_DATA);
761 	if (m0 == NULL)
762 		return (ENOBUFS);
763 
764 	RL_LOCK(sc);
765 
766 	/*
767 	 * Initialize the NIC in test mode. This sets the chip up
768 	 * so that it can send and receive frames, but performs the
769 	 * following special functions:
770 	 * - Puts receiver in promiscuous mode
771 	 * - Enables digital loopback mode
772 	 * - Leaves interrupts turned off
773 	 */
774 
775 	ifp->if_flags |= IFF_PROMISC;
776 	sc->rl_testmode = 1;
777 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
778 	re_init_locked(sc);
779 	sc->rl_flags |= RL_FLAG_LINK;
780 	if (sc->rl_type == RL_8169)
781 		phyaddr = 1;
782 	else
783 		phyaddr = 0;
784 
785 	re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_RESET);
786 	for (i = 0; i < RL_TIMEOUT; i++) {
787 		status = re_miibus_readreg(sc->rl_dev, phyaddr, MII_BMCR);
788 		if (!(status & BMCR_RESET))
789 			break;
790 	}
791 
792 	re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_LOOP);
793 	CSR_WRITE_2(sc, RL_ISR, RL_INTRS);
794 
795 	DELAY(100000);
796 
797 	/* Put some data in the mbuf */
798 
799 	eh = mtod(m0, struct ether_header *);
800 	bcopy ((char *)&dst, eh->ether_dhost, ETHER_ADDR_LEN);
801 	bcopy ((char *)&src, eh->ether_shost, ETHER_ADDR_LEN);
802 	eh->ether_type = htons(ETHERTYPE_IP);
803 	m0->m_pkthdr.len = m0->m_len = ETHER_MIN_LEN - ETHER_CRC_LEN;
804 
805 	/*
806 	 * Queue the packet, start transmission.
807 	 * Note: IF_HANDOFF() ultimately calls re_start() for us.
808 	 */
809 
810 	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
811 	RL_UNLOCK(sc);
812 	/* XXX: re_diag must not be called when in ALTQ mode */
813 	IF_HANDOFF(&ifp->if_snd, m0, ifp);
814 	RL_LOCK(sc);
815 	m0 = NULL;
816 
817 	/* Wait for it to propagate through the chip */
818 
819 	DELAY(100000);
820 	for (i = 0; i < RL_TIMEOUT; i++) {
821 		status = CSR_READ_2(sc, RL_ISR);
822 		CSR_WRITE_2(sc, RL_ISR, status);
823 		if ((status & (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK)) ==
824 		    (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK))
825 			break;
826 		DELAY(10);
827 	}
828 
829 	if (i == RL_TIMEOUT) {
830 		device_printf(sc->rl_dev,
831 		    "diagnostic failed, failed to receive packet in"
832 		    " loopback mode\n");
833 		error = EIO;
834 		goto done;
835 	}
836 
837 	/*
838 	 * The packet should have been dumped into the first
839 	 * entry in the RX DMA ring. Grab it from there.
840 	 */
841 
842 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
843 	    sc->rl_ldata.rl_rx_list_map,
844 	    BUS_DMASYNC_POSTREAD);
845 	bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag,
846 	    sc->rl_ldata.rl_rx_desc[0].rx_dmamap,
847 	    BUS_DMASYNC_POSTREAD);
848 	bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag,
849 	    sc->rl_ldata.rl_rx_desc[0].rx_dmamap);
850 
851 	m0 = sc->rl_ldata.rl_rx_desc[0].rx_m;
852 	sc->rl_ldata.rl_rx_desc[0].rx_m = NULL;
853 	eh = mtod(m0, struct ether_header *);
854 
855 	cur_rx = &sc->rl_ldata.rl_rx_list[0];
856 	total_len = RL_RXBYTES(cur_rx);
857 	rxstat = le32toh(cur_rx->rl_cmdstat);
858 
859 	if (total_len != ETHER_MIN_LEN) {
860 		device_printf(sc->rl_dev,
861 		    "diagnostic failed, received short packet\n");
862 		error = EIO;
863 		goto done;
864 	}
865 
866 	/* Test that the received packet data matches what we sent. */
867 
868 	if (bcmp((char *)&eh->ether_dhost, (char *)&dst, ETHER_ADDR_LEN) ||
869 	    bcmp((char *)&eh->ether_shost, (char *)&src, ETHER_ADDR_LEN) ||
870 	    ntohs(eh->ether_type) != ETHERTYPE_IP) {
871 		device_printf(sc->rl_dev, "WARNING, DMA FAILURE!\n");
872 		device_printf(sc->rl_dev, "expected TX data: %6D/%6D/0x%x\n",
873 		    dst, ":", src, ":", ETHERTYPE_IP);
874 		device_printf(sc->rl_dev, "received RX data: %6D/%6D/0x%x\n",
875 		    eh->ether_dhost, ":", eh->ether_shost, ":",
876 		    ntohs(eh->ether_type));
877 		device_printf(sc->rl_dev, "You may have a defective 32-bit "
878 		    "NIC plugged into a 64-bit PCI slot.\n");
879 		device_printf(sc->rl_dev, "Please re-install the NIC in a "
880 		    "32-bit slot for proper operation.\n");
881 		device_printf(sc->rl_dev, "Read the re(4) man page for more "
882 		    "details.\n");
883 		error = EIO;
884 	}
885 
886 done:
887 	/* Turn interface off, release resources */
888 
889 	sc->rl_testmode = 0;
890 	sc->rl_flags &= ~RL_FLAG_LINK;
891 	ifp->if_flags &= ~IFF_PROMISC;
892 	re_stop(sc);
893 	if (m0 != NULL)
894 		m_freem(m0);
895 
896 	RL_UNLOCK(sc);
897 
898 	return (error);
899 }
900 
901 #endif
902 
903 /*
904  * Probe for a RealTek 8139C+/8169/8110 chip. Check the PCI vendor and device
905  * IDs against our list and return a device name if we find a match.
906  */
907 static int
908 re_probe(device_t dev)
909 {
910 	const struct rl_type	*t;
911 	uint16_t		devid, vendor;
912 	uint16_t		revid, sdevid;
913 	int			i;
914 
915 	vendor = pci_get_vendor(dev);
916 	devid = pci_get_device(dev);
917 	revid = pci_get_revid(dev);
918 	sdevid = pci_get_subdevice(dev);
919 
920 	if (vendor == LINKSYS_VENDORID && devid == LINKSYS_DEVICEID_EG1032) {
921 		if (sdevid != LINKSYS_SUBDEVICE_EG1032_REV3) {
922 			/*
923 			 * Only attach to rev. 3 of the Linksys EG1032 adapter.
924 			 * Rev. 2 is supported by sk(4).
925 			 */
926 			return (ENXIO);
927 		}
928 	}
929 
930 	if (vendor == RT_VENDORID && devid == RT_DEVICEID_8139) {
931 		if (revid != 0x20) {
932 			/* 8139, let rl(4) take care of this device. */
933 			return (ENXIO);
934 		}
935 	}
936 
937 	t = re_devs;
938 	for (i = 0; i < sizeof(re_devs) / sizeof(re_devs[0]); i++, t++) {
939 		if (vendor == t->rl_vid && devid == t->rl_did) {
940 			device_set_desc(dev, t->rl_name);
941 			return (BUS_PROBE_DEFAULT);
942 		}
943 	}
944 
945 	return (ENXIO);
946 }
947 
948 /*
949  * Map a single buffer address.
950  */
951 
952 static void
953 re_dma_map_addr(void *arg, bus_dma_segment_t *segs, int nseg, int error)
954 {
955 	bus_addr_t		*addr;
956 
957 	if (error)
958 		return;
959 
960 	KASSERT(nseg == 1, ("too many DMA segments, %d should be 1", nseg));
961 	addr = arg;
962 	*addr = segs->ds_addr;
963 }
964 
965 static int
966 re_allocmem(device_t dev, struct rl_softc *sc)
967 {
968 	bus_addr_t		lowaddr;
969 	bus_size_t		rx_list_size, tx_list_size;
970 	int			error;
971 	int			i;
972 
973 	rx_list_size = sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc);
974 	tx_list_size = sc->rl_ldata.rl_tx_desc_cnt * sizeof(struct rl_desc);
975 
976 	/*
977 	 * Allocate the parent bus DMA tag appropriate for PCI.
978 	 * In order to use DAC, RL_CPLUSCMD_PCI_DAC bit of RL_CPLUS_CMD
979 	 * register should be set. However some RealTek chips are known
980 	 * to be buggy on DAC handling, therefore disable DAC by limiting
981 	 * DMA address space to 32bit. PCIe variants of RealTek chips
982 	 * may not have the limitation.
983 	 */
984 	lowaddr = BUS_SPACE_MAXADDR;
985 	if ((sc->rl_flags & RL_FLAG_PCIE) == 0)
986 		lowaddr = BUS_SPACE_MAXADDR_32BIT;
987 	error = bus_dma_tag_create(bus_get_dma_tag(dev), 1, 0,
988 	    lowaddr, BUS_SPACE_MAXADDR, NULL, NULL,
989 	    BUS_SPACE_MAXSIZE_32BIT, 0, BUS_SPACE_MAXSIZE_32BIT, 0,
990 	    NULL, NULL, &sc->rl_parent_tag);
991 	if (error) {
992 		device_printf(dev, "could not allocate parent DMA tag\n");
993 		return (error);
994 	}
995 
996 	/*
997 	 * Allocate map for TX mbufs.
998 	 */
999 	error = bus_dma_tag_create(sc->rl_parent_tag, 1, 0,
1000 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL,
1001 	    NULL, MCLBYTES * RL_NTXSEGS, RL_NTXSEGS, 4096, 0,
1002 	    NULL, NULL, &sc->rl_ldata.rl_tx_mtag);
1003 	if (error) {
1004 		device_printf(dev, "could not allocate TX DMA tag\n");
1005 		return (error);
1006 	}
1007 
1008 	/*
1009 	 * Allocate map for RX mbufs.
1010 	 */
1011 
1012 	if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
1013 		error = bus_dma_tag_create(sc->rl_parent_tag, sizeof(uint64_t),
1014 		    0, BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
1015 		    MJUM9BYTES, 1, MJUM9BYTES, 0, NULL, NULL,
1016 		    &sc->rl_ldata.rl_jrx_mtag);
1017 		if (error) {
1018 			device_printf(dev,
1019 			    "could not allocate jumbo RX DMA tag\n");
1020 			return (error);
1021 		}
1022 	}
1023 	error = bus_dma_tag_create(sc->rl_parent_tag, sizeof(uint64_t), 0,
1024 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
1025 	    MCLBYTES, 1, MCLBYTES, 0, NULL, NULL, &sc->rl_ldata.rl_rx_mtag);
1026 	if (error) {
1027 		device_printf(dev, "could not allocate RX DMA tag\n");
1028 		return (error);
1029 	}
1030 
1031 	/*
1032 	 * Allocate map for TX descriptor list.
1033 	 */
1034 	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
1035 	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1036 	    NULL, tx_list_size, 1, tx_list_size, 0,
1037 	    NULL, NULL, &sc->rl_ldata.rl_tx_list_tag);
1038 	if (error) {
1039 		device_printf(dev, "could not allocate TX DMA ring tag\n");
1040 		return (error);
1041 	}
1042 
1043 	/* Allocate DMA'able memory for the TX ring */
1044 
1045 	error = bus_dmamem_alloc(sc->rl_ldata.rl_tx_list_tag,
1046 	    (void **)&sc->rl_ldata.rl_tx_list,
1047 	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1048 	    &sc->rl_ldata.rl_tx_list_map);
1049 	if (error) {
1050 		device_printf(dev, "could not allocate TX DMA ring\n");
1051 		return (error);
1052 	}
1053 
1054 	/* Load the map for the TX ring. */
1055 
1056 	sc->rl_ldata.rl_tx_list_addr = 0;
1057 	error = bus_dmamap_load(sc->rl_ldata.rl_tx_list_tag,
1058 	     sc->rl_ldata.rl_tx_list_map, sc->rl_ldata.rl_tx_list,
1059 	     tx_list_size, re_dma_map_addr,
1060 	     &sc->rl_ldata.rl_tx_list_addr, BUS_DMA_NOWAIT);
1061 	if (error != 0 || sc->rl_ldata.rl_tx_list_addr == 0) {
1062 		device_printf(dev, "could not load TX DMA ring\n");
1063 		return (ENOMEM);
1064 	}
1065 
1066 	/* Create DMA maps for TX buffers */
1067 
1068 	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
1069 		error = bus_dmamap_create(sc->rl_ldata.rl_tx_mtag, 0,
1070 		    &sc->rl_ldata.rl_tx_desc[i].tx_dmamap);
1071 		if (error) {
1072 			device_printf(dev, "could not create DMA map for TX\n");
1073 			return (error);
1074 		}
1075 	}
1076 
1077 	/*
1078 	 * Allocate map for RX descriptor list.
1079 	 */
1080 	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
1081 	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1082 	    NULL, rx_list_size, 1, rx_list_size, 0,
1083 	    NULL, NULL, &sc->rl_ldata.rl_rx_list_tag);
1084 	if (error) {
1085 		device_printf(dev, "could not create RX DMA ring tag\n");
1086 		return (error);
1087 	}
1088 
1089 	/* Allocate DMA'able memory for the RX ring */
1090 
1091 	error = bus_dmamem_alloc(sc->rl_ldata.rl_rx_list_tag,
1092 	    (void **)&sc->rl_ldata.rl_rx_list,
1093 	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1094 	    &sc->rl_ldata.rl_rx_list_map);
1095 	if (error) {
1096 		device_printf(dev, "could not allocate RX DMA ring\n");
1097 		return (error);
1098 	}
1099 
1100 	/* Load the map for the RX ring. */
1101 
1102 	sc->rl_ldata.rl_rx_list_addr = 0;
1103 	error = bus_dmamap_load(sc->rl_ldata.rl_rx_list_tag,
1104 	     sc->rl_ldata.rl_rx_list_map, sc->rl_ldata.rl_rx_list,
1105 	     rx_list_size, re_dma_map_addr,
1106 	     &sc->rl_ldata.rl_rx_list_addr, BUS_DMA_NOWAIT);
1107 	if (error != 0 || sc->rl_ldata.rl_rx_list_addr == 0) {
1108 		device_printf(dev, "could not load RX DMA ring\n");
1109 		return (ENOMEM);
1110 	}
1111 
1112 	/* Create DMA maps for RX buffers */
1113 
1114 	if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
1115 		error = bus_dmamap_create(sc->rl_ldata.rl_jrx_mtag, 0,
1116 		    &sc->rl_ldata.rl_jrx_sparemap);
1117 		if (error) {
1118 			device_printf(dev,
1119 			    "could not create spare DMA map for jumbo RX\n");
1120 			return (error);
1121 		}
1122 		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1123 			error = bus_dmamap_create(sc->rl_ldata.rl_jrx_mtag, 0,
1124 			    &sc->rl_ldata.rl_jrx_desc[i].rx_dmamap);
1125 			if (error) {
1126 				device_printf(dev,
1127 				    "could not create DMA map for jumbo RX\n");
1128 				return (error);
1129 			}
1130 		}
1131 	}
1132 	error = bus_dmamap_create(sc->rl_ldata.rl_rx_mtag, 0,
1133 	    &sc->rl_ldata.rl_rx_sparemap);
1134 	if (error) {
1135 		device_printf(dev, "could not create spare DMA map for RX\n");
1136 		return (error);
1137 	}
1138 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1139 		error = bus_dmamap_create(sc->rl_ldata.rl_rx_mtag, 0,
1140 		    &sc->rl_ldata.rl_rx_desc[i].rx_dmamap);
1141 		if (error) {
1142 			device_printf(dev, "could not create DMA map for RX\n");
1143 			return (error);
1144 		}
1145 	}
1146 
1147 	/* Create DMA map for statistics. */
1148 	error = bus_dma_tag_create(sc->rl_parent_tag, RL_DUMP_ALIGN, 0,
1149 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
1150 	    sizeof(struct rl_stats), 1, sizeof(struct rl_stats), 0, NULL, NULL,
1151 	    &sc->rl_ldata.rl_stag);
1152 	if (error) {
1153 		device_printf(dev, "could not create statistics DMA tag\n");
1154 		return (error);
1155 	}
1156 	/* Allocate DMA'able memory for statistics. */
1157 	error = bus_dmamem_alloc(sc->rl_ldata.rl_stag,
1158 	    (void **)&sc->rl_ldata.rl_stats,
1159 	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1160 	    &sc->rl_ldata.rl_smap);
1161 	if (error) {
1162 		device_printf(dev,
1163 		    "could not allocate statistics DMA memory\n");
1164 		return (error);
1165 	}
1166 	/* Load the map for statistics. */
1167 	sc->rl_ldata.rl_stats_addr = 0;
1168 	error = bus_dmamap_load(sc->rl_ldata.rl_stag, sc->rl_ldata.rl_smap,
1169 	    sc->rl_ldata.rl_stats, sizeof(struct rl_stats), re_dma_map_addr,
1170 	     &sc->rl_ldata.rl_stats_addr, BUS_DMA_NOWAIT);
1171 	if (error != 0 || sc->rl_ldata.rl_stats_addr == 0) {
1172 		device_printf(dev, "could not load statistics DMA memory\n");
1173 		return (ENOMEM);
1174 	}
1175 
1176 	return (0);
1177 }
1178 
1179 /*
1180  * Attach the interface. Allocate softc structures, do ifmedia
1181  * setup and ethernet/BPF attach.
1182  */
1183 static int
1184 re_attach(device_t dev)
1185 {
1186 	u_char			eaddr[ETHER_ADDR_LEN];
1187 	u_int16_t		as[ETHER_ADDR_LEN / 2];
1188 	struct rl_softc		*sc;
1189 	struct ifnet		*ifp;
1190 	const struct rl_hwrev	*hw_rev;
1191 	u_int32_t		cap, ctl;
1192 	int			hwrev;
1193 	u_int16_t		devid, re_did = 0;
1194 	int			error = 0, i, phy, rid;
1195 	int			msic, msixc, reg;
1196 	uint8_t			cfg;
1197 
1198 	sc = device_get_softc(dev);
1199 	sc->rl_dev = dev;
1200 
1201 	mtx_init(&sc->rl_mtx, device_get_nameunit(dev), MTX_NETWORK_LOCK,
1202 	    MTX_DEF);
1203 	callout_init_mtx(&sc->rl_stat_callout, &sc->rl_mtx, 0);
1204 
1205 	/*
1206 	 * Map control/status registers.
1207 	 */
1208 	pci_enable_busmaster(dev);
1209 
1210 	devid = pci_get_device(dev);
1211 	/*
1212 	 * Prefer memory space register mapping over IO space.
1213 	 * Because RTL8169SC does not seem to work when memory mapping
1214 	 * is used always activate io mapping.
1215 	 */
1216 	if (devid == RT_DEVICEID_8169SC)
1217 		prefer_iomap = 1;
1218 	if (prefer_iomap == 0) {
1219 		sc->rl_res_id = PCIR_BAR(1);
1220 		sc->rl_res_type = SYS_RES_MEMORY;
1221 		/* RTL8168/8101E seems to use different BARs. */
1222 		if (devid == RT_DEVICEID_8168 || devid == RT_DEVICEID_8101E)
1223 			sc->rl_res_id = PCIR_BAR(2);
1224 	} else {
1225 		sc->rl_res_id = PCIR_BAR(0);
1226 		sc->rl_res_type = SYS_RES_IOPORT;
1227 	}
1228 	sc->rl_res = bus_alloc_resource_any(dev, sc->rl_res_type,
1229 	    &sc->rl_res_id, RF_ACTIVE);
1230 	if (sc->rl_res == NULL && prefer_iomap == 0) {
1231 		sc->rl_res_id = PCIR_BAR(0);
1232 		sc->rl_res_type = SYS_RES_IOPORT;
1233 		sc->rl_res = bus_alloc_resource_any(dev, sc->rl_res_type,
1234 		    &sc->rl_res_id, RF_ACTIVE);
1235 	}
1236 	if (sc->rl_res == NULL) {
1237 		device_printf(dev, "couldn't map ports/memory\n");
1238 		error = ENXIO;
1239 		goto fail;
1240 	}
1241 
1242 	sc->rl_btag = rman_get_bustag(sc->rl_res);
1243 	sc->rl_bhandle = rman_get_bushandle(sc->rl_res);
1244 
1245 	msic = pci_msi_count(dev);
1246 	msixc = pci_msix_count(dev);
1247 	if (pci_find_cap(dev, PCIY_EXPRESS, &reg) == 0) {
1248 		sc->rl_flags |= RL_FLAG_PCIE;
1249 		sc->rl_expcap = reg;
1250 	}
1251 	if (bootverbose) {
1252 		device_printf(dev, "MSI count : %d\n", msic);
1253 		device_printf(dev, "MSI-X count : %d\n", msixc);
1254 	}
1255 	if (msix_disable > 0)
1256 		msixc = 0;
1257 	if (msi_disable > 0)
1258 		msic = 0;
1259 	/* Prefer MSI-X to MSI. */
1260 	if (msixc > 0) {
1261 		msixc = 1;
1262 		rid = PCIR_BAR(4);
1263 		sc->rl_res_pba = bus_alloc_resource_any(dev, SYS_RES_MEMORY,
1264 		    &rid, RF_ACTIVE);
1265 		if (sc->rl_res_pba == NULL) {
1266 			device_printf(sc->rl_dev,
1267 			    "could not allocate MSI-X PBA resource\n");
1268 		}
1269 		if (sc->rl_res_pba != NULL &&
1270 		    pci_alloc_msix(dev, &msixc) == 0) {
1271 			if (msixc == 1) {
1272 				device_printf(dev, "Using %d MSI-X message\n",
1273 				    msixc);
1274 				sc->rl_flags |= RL_FLAG_MSIX;
1275 			} else
1276 				pci_release_msi(dev);
1277 		}
1278 		if ((sc->rl_flags & RL_FLAG_MSIX) == 0) {
1279 			if (sc->rl_res_pba != NULL)
1280 				bus_release_resource(dev, SYS_RES_MEMORY, rid,
1281 				    sc->rl_res_pba);
1282 			sc->rl_res_pba = NULL;
1283 			msixc = 0;
1284 		}
1285 	}
1286 	/* Prefer MSI to INTx. */
1287 	if (msixc == 0 && msic > 0) {
1288 		msic = 1;
1289 		if (pci_alloc_msi(dev, &msic) == 0) {
1290 			if (msic == RL_MSI_MESSAGES) {
1291 				device_printf(dev, "Using %d MSI message\n",
1292 				    msic);
1293 				sc->rl_flags |= RL_FLAG_MSI;
1294 				/* Explicitly set MSI enable bit. */
1295 				CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1296 				cfg = CSR_READ_1(sc, RL_CFG2);
1297 				cfg |= RL_CFG2_MSI;
1298 				CSR_WRITE_1(sc, RL_CFG2, cfg);
1299 				CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1300 			} else
1301 				pci_release_msi(dev);
1302 		}
1303 		if ((sc->rl_flags & RL_FLAG_MSI) == 0)
1304 			msic = 0;
1305 	}
1306 
1307 	/* Allocate interrupt */
1308 	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) == 0) {
1309 		rid = 0;
1310 		sc->rl_irq[0] = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
1311 		    RF_SHAREABLE | RF_ACTIVE);
1312 		if (sc->rl_irq[0] == NULL) {
1313 			device_printf(dev, "couldn't allocate IRQ resources\n");
1314 			error = ENXIO;
1315 			goto fail;
1316 		}
1317 	} else {
1318 		for (i = 0, rid = 1; i < RL_MSI_MESSAGES; i++, rid++) {
1319 			sc->rl_irq[i] = bus_alloc_resource_any(dev,
1320 			    SYS_RES_IRQ, &rid, RF_ACTIVE);
1321 			if (sc->rl_irq[i] == NULL) {
1322 				device_printf(dev,
1323 				    "couldn't llocate IRQ resources for "
1324 				    "message %d\n", rid);
1325 				error = ENXIO;
1326 				goto fail;
1327 			}
1328 		}
1329 	}
1330 
1331 	if ((sc->rl_flags & RL_FLAG_MSI) == 0) {
1332 		CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1333 		cfg = CSR_READ_1(sc, RL_CFG2);
1334 		if ((cfg & RL_CFG2_MSI) != 0) {
1335 			device_printf(dev, "turning off MSI enable bit.\n");
1336 			cfg &= ~RL_CFG2_MSI;
1337 			CSR_WRITE_1(sc, RL_CFG2, cfg);
1338 		}
1339 		CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1340 	}
1341 
1342 	/* Disable ASPM L0S/L1. */
1343 	if (sc->rl_expcap != 0) {
1344 		cap = pci_read_config(dev, sc->rl_expcap +
1345 		    PCIR_EXPRESS_LINK_CAP, 2);
1346 		if ((cap & PCIM_LINK_CAP_ASPM) != 0) {
1347 			ctl = pci_read_config(dev, sc->rl_expcap +
1348 			    PCIR_EXPRESS_LINK_CTL, 2);
1349 			if ((ctl & 0x0003) != 0) {
1350 				ctl &= ~0x0003;
1351 				pci_write_config(dev, sc->rl_expcap +
1352 				    PCIR_EXPRESS_LINK_CTL, ctl, 2);
1353 				device_printf(dev, "ASPM disabled\n");
1354 			}
1355 		} else
1356 			device_printf(dev, "no ASPM capability\n");
1357 	}
1358 
1359 	hw_rev = re_hwrevs;
1360 	hwrev = CSR_READ_4(sc, RL_TXCFG);
1361 	switch (hwrev & 0x70000000) {
1362 	case 0x00000000:
1363 	case 0x10000000:
1364 		device_printf(dev, "Chip rev. 0x%08x\n", hwrev & 0xfc800000);
1365 		hwrev &= (RL_TXCFG_HWREV | 0x80000000);
1366 		break;
1367 	default:
1368 		device_printf(dev, "Chip rev. 0x%08x\n", hwrev & 0x7c800000);
1369 		hwrev &= RL_TXCFG_HWREV;
1370 		break;
1371 	}
1372 	device_printf(dev, "MAC rev. 0x%08x\n", hwrev & 0x00700000);
1373 	while (hw_rev->rl_desc != NULL) {
1374 		if (hw_rev->rl_rev == hwrev) {
1375 			sc->rl_type = hw_rev->rl_type;
1376 			sc->rl_hwrev = hw_rev;
1377 			break;
1378 		}
1379 		hw_rev++;
1380 	}
1381 	if (hw_rev->rl_desc == NULL) {
1382 		device_printf(dev, "Unknown H/W revision: 0x%08x\n", hwrev);
1383 		error = ENXIO;
1384 		goto fail;
1385 	}
1386 
1387 	switch (hw_rev->rl_rev) {
1388 	case RL_HWREV_8139CPLUS:
1389 		sc->rl_flags |= RL_FLAG_FASTETHER | RL_FLAG_AUTOPAD;
1390 		break;
1391 	case RL_HWREV_8100E:
1392 	case RL_HWREV_8101E:
1393 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_FASTETHER;
1394 		break;
1395 	case RL_HWREV_8102E:
1396 	case RL_HWREV_8102EL:
1397 	case RL_HWREV_8102EL_SPIN1:
1398 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR | RL_FLAG_DESCV2 |
1399 		    RL_FLAG_MACSTAT | RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP |
1400 		    RL_FLAG_AUTOPAD;
1401 		break;
1402 	case RL_HWREV_8103E:
1403 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR | RL_FLAG_DESCV2 |
1404 		    RL_FLAG_MACSTAT | RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP |
1405 		    RL_FLAG_AUTOPAD | RL_FLAG_MACSLEEP;
1406 		break;
1407 	case RL_HWREV_8401E:
1408 	case RL_HWREV_8402:
1409 	case RL_HWREV_8105E:
1410 	case RL_HWREV_8105E_SPIN1:
1411 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
1412 		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
1413 		    RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD;
1414 		break;
1415 	case RL_HWREV_8168B_SPIN1:
1416 	case RL_HWREV_8168B_SPIN2:
1417 		sc->rl_flags |= RL_FLAG_WOLRXENB;
1418 		/* FALLTHROUGH */
1419 	case RL_HWREV_8168B_SPIN3:
1420 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_MACSTAT;
1421 		break;
1422 	case RL_HWREV_8168C_SPIN2:
1423 		sc->rl_flags |= RL_FLAG_MACSLEEP;
1424 		/* FALLTHROUGH */
1425 	case RL_HWREV_8168C:
1426 		if ((hwrev & 0x00700000) == 0x00200000)
1427 			sc->rl_flags |= RL_FLAG_MACSLEEP;
1428 		/* FALLTHROUGH */
1429 	case RL_HWREV_8168CP:
1430 	case RL_HWREV_8168D:
1431 	case RL_HWREV_8168DP:
1432 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
1433 		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP |
1434 		    RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2;
1435 		break;
1436 	case RL_HWREV_8168E:
1437 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
1438 		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
1439 		    RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2;
1440 		break;
1441 	case RL_HWREV_8168E_VL:
1442 	case RL_HWREV_8168F:
1443 	case RL_HWREV_8411:
1444 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
1445 		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP |
1446 		    RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2;
1447 		break;
1448 	case RL_HWREV_8169_8110SB:
1449 	case RL_HWREV_8169_8110SBL:
1450 	case RL_HWREV_8169_8110SC:
1451 	case RL_HWREV_8169_8110SCE:
1452 		sc->rl_flags |= RL_FLAG_PHYWAKE;
1453 		/* FALLTHROUGH */
1454 	case RL_HWREV_8169:
1455 	case RL_HWREV_8169S:
1456 	case RL_HWREV_8110S:
1457 		sc->rl_flags |= RL_FLAG_MACRESET;
1458 		break;
1459 	default:
1460 		break;
1461 	}
1462 
1463 	/* Reset the adapter. */
1464 	RL_LOCK(sc);
1465 	re_reset(sc);
1466 	RL_UNLOCK(sc);
1467 
1468 	/* Enable PME. */
1469 	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1470 	cfg = CSR_READ_1(sc, RL_CFG1);
1471 	cfg |= RL_CFG1_PME;
1472 	CSR_WRITE_1(sc, RL_CFG1, cfg);
1473 	cfg = CSR_READ_1(sc, RL_CFG5);
1474 	cfg &= RL_CFG5_PME_STS;
1475 	CSR_WRITE_1(sc, RL_CFG5, cfg);
1476 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1477 
1478 	if ((sc->rl_flags & RL_FLAG_PAR) != 0) {
1479 		/*
1480 		 * XXX Should have a better way to extract station
1481 		 * address from EEPROM.
1482 		 */
1483 		for (i = 0; i < ETHER_ADDR_LEN; i++)
1484 			eaddr[i] = CSR_READ_1(sc, RL_IDR0 + i);
1485 	} else {
1486 		sc->rl_eewidth = RL_9356_ADDR_LEN;
1487 		re_read_eeprom(sc, (caddr_t)&re_did, 0, 1);
1488 		if (re_did != 0x8129)
1489 			sc->rl_eewidth = RL_9346_ADDR_LEN;
1490 
1491 		/*
1492 		 * Get station address from the EEPROM.
1493 		 */
1494 		re_read_eeprom(sc, (caddr_t)as, RL_EE_EADDR, 3);
1495 		for (i = 0; i < ETHER_ADDR_LEN / 2; i++)
1496 			as[i] = le16toh(as[i]);
1497 		bcopy(as, eaddr, sizeof(eaddr));
1498 	}
1499 
1500 	if (sc->rl_type == RL_8169) {
1501 		/* Set RX length mask and number of descriptors. */
1502 		sc->rl_rxlenmask = RL_RDESC_STAT_GFRAGLEN;
1503 		sc->rl_txstart = RL_GTXSTART;
1504 		sc->rl_ldata.rl_tx_desc_cnt = RL_8169_TX_DESC_CNT;
1505 		sc->rl_ldata.rl_rx_desc_cnt = RL_8169_RX_DESC_CNT;
1506 	} else {
1507 		/* Set RX length mask and number of descriptors. */
1508 		sc->rl_rxlenmask = RL_RDESC_STAT_FRAGLEN;
1509 		sc->rl_txstart = RL_TXSTART;
1510 		sc->rl_ldata.rl_tx_desc_cnt = RL_8139_TX_DESC_CNT;
1511 		sc->rl_ldata.rl_rx_desc_cnt = RL_8139_RX_DESC_CNT;
1512 	}
1513 
1514 	error = re_allocmem(dev, sc);
1515 	if (error)
1516 		goto fail;
1517 	re_add_sysctls(sc);
1518 
1519 	ifp = sc->rl_ifp = if_alloc(IFT_ETHER);
1520 	if (ifp == NULL) {
1521 		device_printf(dev, "can not if_alloc()\n");
1522 		error = ENOSPC;
1523 		goto fail;
1524 	}
1525 
1526 	/* Take controller out of deep sleep mode. */
1527 	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
1528 		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
1529 			CSR_WRITE_1(sc, RL_GPIO,
1530 			    CSR_READ_1(sc, RL_GPIO) | 0x01);
1531 		else
1532 			CSR_WRITE_1(sc, RL_GPIO,
1533 			    CSR_READ_1(sc, RL_GPIO) & ~0x01);
1534 	}
1535 
1536 	/* Take PHY out of power down mode. */
1537 	if ((sc->rl_flags & RL_FLAG_PHYWAKE_PM) != 0) {
1538 		CSR_WRITE_1(sc, RL_PMCH, CSR_READ_1(sc, RL_PMCH) | 0x80);
1539 		if (hw_rev->rl_rev == RL_HWREV_8401E)
1540 			CSR_WRITE_1(sc, 0xD1, CSR_READ_1(sc, 0xD1) & ~0x08);
1541 	}
1542 	if ((sc->rl_flags & RL_FLAG_PHYWAKE) != 0) {
1543 		re_gmii_writereg(dev, 1, 0x1f, 0);
1544 		re_gmii_writereg(dev, 1, 0x0e, 0);
1545 	}
1546 
1547 #define	RE_PHYAD_INTERNAL	 0
1548 
1549 	/* Do MII setup. */
1550 	phy = RE_PHYAD_INTERNAL;
1551 	if (sc->rl_type == RL_8169)
1552 		phy = 1;
1553 	error = mii_attach(dev, &sc->rl_miibus, ifp, re_ifmedia_upd,
1554 	    re_ifmedia_sts, BMSR_DEFCAPMASK, phy, MII_OFFSET_ANY, MIIF_DOPAUSE);
1555 	if (error != 0) {
1556 		device_printf(dev, "attaching PHYs failed\n");
1557 		goto fail;
1558 	}
1559 
1560 	ifp->if_softc = sc;
1561 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1562 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1563 	ifp->if_ioctl = re_ioctl;
1564 	ifp->if_start = re_start;
1565 	/*
1566 	 * RTL8168/8111C generates wrong IP checksummed frame if the
1567 	 * packet has IP options so disable TX IP checksum offloading.
1568 	 */
1569 	if (sc->rl_hwrev->rl_rev == RL_HWREV_8168C ||
1570 	    sc->rl_hwrev->rl_rev == RL_HWREV_8168C_SPIN2)
1571 		ifp->if_hwassist = CSUM_TCP | CSUM_UDP;
1572 	else
1573 		ifp->if_hwassist = CSUM_IP | CSUM_TCP | CSUM_UDP;
1574 	ifp->if_hwassist |= CSUM_TSO;
1575 	ifp->if_capabilities = IFCAP_HWCSUM | IFCAP_TSO4;
1576 	ifp->if_capenable = ifp->if_capabilities;
1577 	ifp->if_init = re_init;
1578 	IFQ_SET_MAXLEN(&ifp->if_snd, RL_IFQ_MAXLEN);
1579 	ifp->if_snd.ifq_drv_maxlen = RL_IFQ_MAXLEN;
1580 	IFQ_SET_READY(&ifp->if_snd);
1581 
1582 	TASK_INIT(&sc->rl_inttask, 0, re_int_task, sc);
1583 
1584 	/*
1585 	 * Call MI attach routine.
1586 	 */
1587 	ether_ifattach(ifp, eaddr);
1588 
1589 	/* VLAN capability setup */
1590 	ifp->if_capabilities |= IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING;
1591 	if (ifp->if_capabilities & IFCAP_HWCSUM)
1592 		ifp->if_capabilities |= IFCAP_VLAN_HWCSUM;
1593 	/* Enable WOL if PM is supported. */
1594 	if (pci_find_cap(sc->rl_dev, PCIY_PMG, &reg) == 0)
1595 		ifp->if_capabilities |= IFCAP_WOL;
1596 	ifp->if_capenable = ifp->if_capabilities;
1597 	/*
1598 	 * Don't enable TSO by default.  It is known to generate
1599 	 * corrupted TCP segments(bad TCP options) under certain
1600 	 * circumtances.
1601 	 */
1602 	ifp->if_hwassist &= ~CSUM_TSO;
1603 	ifp->if_capenable &= ~(IFCAP_TSO4 | IFCAP_VLAN_HWTSO);
1604 #ifdef DEVICE_POLLING
1605 	ifp->if_capabilities |= IFCAP_POLLING;
1606 #endif
1607 	/*
1608 	 * Tell the upper layer(s) we support long frames.
1609 	 * Must appear after the call to ether_ifattach() because
1610 	 * ether_ifattach() sets ifi_hdrlen to the default value.
1611 	 */
1612 	ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
1613 
1614 #ifdef RE_DIAG
1615 	/*
1616 	 * Perform hardware diagnostic on the original RTL8169.
1617 	 * Some 32-bit cards were incorrectly wired and would
1618 	 * malfunction if plugged into a 64-bit slot.
1619 	 */
1620 
1621 	if (hwrev == RL_HWREV_8169) {
1622 		error = re_diag(sc);
1623 		if (error) {
1624 			device_printf(dev,
1625 		    	"attach aborted due to hardware diag failure\n");
1626 			ether_ifdetach(ifp);
1627 			goto fail;
1628 		}
1629 	}
1630 #endif
1631 
1632 #ifdef RE_TX_MODERATION
1633 	intr_filter = 1;
1634 #endif
1635 	/* Hook interrupt last to avoid having to lock softc */
1636 	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) != 0 &&
1637 	    intr_filter == 0) {
1638 		error = bus_setup_intr(dev, sc->rl_irq[0],
1639 		    INTR_TYPE_NET | INTR_MPSAFE, NULL, re_intr_msi, sc,
1640 		    &sc->rl_intrhand[0]);
1641 	} else {
1642 		error = bus_setup_intr(dev, sc->rl_irq[0],
1643 		    INTR_TYPE_NET | INTR_MPSAFE, re_intr, NULL, sc,
1644 		    &sc->rl_intrhand[0]);
1645 	}
1646 	if (error) {
1647 		device_printf(dev, "couldn't set up irq\n");
1648 		ether_ifdetach(ifp);
1649 	}
1650 
1651 fail:
1652 
1653 	if (error)
1654 		re_detach(dev);
1655 
1656 	return (error);
1657 }
1658 
1659 /*
1660  * Shutdown hardware and free up resources. This can be called any
1661  * time after the mutex has been initialized. It is called in both
1662  * the error case in attach and the normal detach case so it needs
1663  * to be careful about only freeing resources that have actually been
1664  * allocated.
1665  */
1666 static int
1667 re_detach(device_t dev)
1668 {
1669 	struct rl_softc		*sc;
1670 	struct ifnet		*ifp;
1671 	int			i, rid;
1672 
1673 	sc = device_get_softc(dev);
1674 	ifp = sc->rl_ifp;
1675 	KASSERT(mtx_initialized(&sc->rl_mtx), ("re mutex not initialized"));
1676 
1677 	/* These should only be active if attach succeeded */
1678 	if (device_is_attached(dev)) {
1679 #ifdef DEVICE_POLLING
1680 		if (ifp->if_capenable & IFCAP_POLLING)
1681 			ether_poll_deregister(ifp);
1682 #endif
1683 		RL_LOCK(sc);
1684 #if 0
1685 		sc->suspended = 1;
1686 #endif
1687 		re_stop(sc);
1688 		RL_UNLOCK(sc);
1689 		callout_drain(&sc->rl_stat_callout);
1690 		taskqueue_drain(taskqueue_fast, &sc->rl_inttask);
1691 		/*
1692 		 * Force off the IFF_UP flag here, in case someone
1693 		 * still had a BPF descriptor attached to this
1694 		 * interface. If they do, ether_ifdetach() will cause
1695 		 * the BPF code to try and clear the promisc mode
1696 		 * flag, which will bubble down to re_ioctl(),
1697 		 * which will try to call re_init() again. This will
1698 		 * turn the NIC back on and restart the MII ticker,
1699 		 * which will panic the system when the kernel tries
1700 		 * to invoke the re_tick() function that isn't there
1701 		 * anymore.
1702 		 */
1703 		ifp->if_flags &= ~IFF_UP;
1704 		ether_ifdetach(ifp);
1705 	}
1706 	if (sc->rl_miibus)
1707 		device_delete_child(dev, sc->rl_miibus);
1708 	bus_generic_detach(dev);
1709 
1710 	/*
1711 	 * The rest is resource deallocation, so we should already be
1712 	 * stopped here.
1713 	 */
1714 
1715 	if (sc->rl_intrhand[0] != NULL) {
1716 		bus_teardown_intr(dev, sc->rl_irq[0], sc->rl_intrhand[0]);
1717 		sc->rl_intrhand[0] = NULL;
1718 	}
1719 	if (ifp != NULL)
1720 		if_free(ifp);
1721 	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) == 0)
1722 		rid = 0;
1723 	else
1724 		rid = 1;
1725 	if (sc->rl_irq[0] != NULL) {
1726 		bus_release_resource(dev, SYS_RES_IRQ, rid, sc->rl_irq[0]);
1727 		sc->rl_irq[0] = NULL;
1728 	}
1729 	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) != 0)
1730 		pci_release_msi(dev);
1731 	if (sc->rl_res_pba) {
1732 		rid = PCIR_BAR(4);
1733 		bus_release_resource(dev, SYS_RES_MEMORY, rid, sc->rl_res_pba);
1734 	}
1735 	if (sc->rl_res)
1736 		bus_release_resource(dev, sc->rl_res_type, sc->rl_res_id,
1737 		    sc->rl_res);
1738 
1739 	/* Unload and free the RX DMA ring memory and map */
1740 
1741 	if (sc->rl_ldata.rl_rx_list_tag) {
1742 		if (sc->rl_ldata.rl_rx_list_map)
1743 			bus_dmamap_unload(sc->rl_ldata.rl_rx_list_tag,
1744 			    sc->rl_ldata.rl_rx_list_map);
1745 		if (sc->rl_ldata.rl_rx_list_map && sc->rl_ldata.rl_rx_list)
1746 			bus_dmamem_free(sc->rl_ldata.rl_rx_list_tag,
1747 			    sc->rl_ldata.rl_rx_list,
1748 			    sc->rl_ldata.rl_rx_list_map);
1749 		bus_dma_tag_destroy(sc->rl_ldata.rl_rx_list_tag);
1750 	}
1751 
1752 	/* Unload and free the TX DMA ring memory and map */
1753 
1754 	if (sc->rl_ldata.rl_tx_list_tag) {
1755 		if (sc->rl_ldata.rl_tx_list_map)
1756 			bus_dmamap_unload(sc->rl_ldata.rl_tx_list_tag,
1757 			    sc->rl_ldata.rl_tx_list_map);
1758 		if (sc->rl_ldata.rl_tx_list_map && sc->rl_ldata.rl_tx_list)
1759 			bus_dmamem_free(sc->rl_ldata.rl_tx_list_tag,
1760 			    sc->rl_ldata.rl_tx_list,
1761 			    sc->rl_ldata.rl_tx_list_map);
1762 		bus_dma_tag_destroy(sc->rl_ldata.rl_tx_list_tag);
1763 	}
1764 
1765 	/* Destroy all the RX and TX buffer maps */
1766 
1767 	if (sc->rl_ldata.rl_tx_mtag) {
1768 		for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
1769 			if (sc->rl_ldata.rl_tx_desc[i].tx_dmamap)
1770 				bus_dmamap_destroy(sc->rl_ldata.rl_tx_mtag,
1771 				    sc->rl_ldata.rl_tx_desc[i].tx_dmamap);
1772 		}
1773 		bus_dma_tag_destroy(sc->rl_ldata.rl_tx_mtag);
1774 	}
1775 	if (sc->rl_ldata.rl_rx_mtag) {
1776 		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1777 			if (sc->rl_ldata.rl_rx_desc[i].rx_dmamap)
1778 				bus_dmamap_destroy(sc->rl_ldata.rl_rx_mtag,
1779 				    sc->rl_ldata.rl_rx_desc[i].rx_dmamap);
1780 		}
1781 		if (sc->rl_ldata.rl_rx_sparemap)
1782 			bus_dmamap_destroy(sc->rl_ldata.rl_rx_mtag,
1783 			    sc->rl_ldata.rl_rx_sparemap);
1784 		bus_dma_tag_destroy(sc->rl_ldata.rl_rx_mtag);
1785 	}
1786 	if (sc->rl_ldata.rl_jrx_mtag) {
1787 		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1788 			if (sc->rl_ldata.rl_jrx_desc[i].rx_dmamap)
1789 				bus_dmamap_destroy(sc->rl_ldata.rl_jrx_mtag,
1790 				    sc->rl_ldata.rl_jrx_desc[i].rx_dmamap);
1791 		}
1792 		if (sc->rl_ldata.rl_jrx_sparemap)
1793 			bus_dmamap_destroy(sc->rl_ldata.rl_jrx_mtag,
1794 			    sc->rl_ldata.rl_jrx_sparemap);
1795 		bus_dma_tag_destroy(sc->rl_ldata.rl_jrx_mtag);
1796 	}
1797 	/* Unload and free the stats buffer and map */
1798 
1799 	if (sc->rl_ldata.rl_stag) {
1800 		if (sc->rl_ldata.rl_smap)
1801 			bus_dmamap_unload(sc->rl_ldata.rl_stag,
1802 			    sc->rl_ldata.rl_smap);
1803 		if (sc->rl_ldata.rl_smap && sc->rl_ldata.rl_stats)
1804 			bus_dmamem_free(sc->rl_ldata.rl_stag,
1805 			    sc->rl_ldata.rl_stats, sc->rl_ldata.rl_smap);
1806 		bus_dma_tag_destroy(sc->rl_ldata.rl_stag);
1807 	}
1808 
1809 	if (sc->rl_parent_tag)
1810 		bus_dma_tag_destroy(sc->rl_parent_tag);
1811 
1812 	mtx_destroy(&sc->rl_mtx);
1813 
1814 	return (0);
1815 }
1816 
1817 static __inline void
1818 re_discard_rxbuf(struct rl_softc *sc, int idx)
1819 {
1820 	struct rl_desc		*desc;
1821 	struct rl_rxdesc	*rxd;
1822 	uint32_t		cmdstat;
1823 
1824 	if (sc->rl_ifp->if_mtu > RL_MTU &&
1825 	    (sc->rl_flags & RL_FLAG_JUMBOV2) != 0)
1826 		rxd = &sc->rl_ldata.rl_jrx_desc[idx];
1827 	else
1828 		rxd = &sc->rl_ldata.rl_rx_desc[idx];
1829 	desc = &sc->rl_ldata.rl_rx_list[idx];
1830 	desc->rl_vlanctl = 0;
1831 	cmdstat = rxd->rx_size;
1832 	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
1833 		cmdstat |= RL_RDESC_CMD_EOR;
1834 	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
1835 }
1836 
1837 static int
1838 re_newbuf(struct rl_softc *sc, int idx)
1839 {
1840 	struct mbuf		*m;
1841 	struct rl_rxdesc	*rxd;
1842 	bus_dma_segment_t	segs[1];
1843 	bus_dmamap_t		map;
1844 	struct rl_desc		*desc;
1845 	uint32_t		cmdstat;
1846 	int			error, nsegs;
1847 
1848 	m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
1849 	if (m == NULL)
1850 		return (ENOBUFS);
1851 
1852 	m->m_len = m->m_pkthdr.len = MCLBYTES;
1853 #ifdef RE_FIXUP_RX
1854 	/*
1855 	 * This is part of an evil trick to deal with non-x86 platforms.
1856 	 * The RealTek chip requires RX buffers to be aligned on 64-bit
1857 	 * boundaries, but that will hose non-x86 machines. To get around
1858 	 * this, we leave some empty space at the start of each buffer
1859 	 * and for non-x86 hosts, we copy the buffer back six bytes
1860 	 * to achieve word alignment. This is slightly more efficient
1861 	 * than allocating a new buffer, copying the contents, and
1862 	 * discarding the old buffer.
1863 	 */
1864 	m_adj(m, RE_ETHER_ALIGN);
1865 #endif
1866 	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_rx_mtag,
1867 	    sc->rl_ldata.rl_rx_sparemap, m, segs, &nsegs, BUS_DMA_NOWAIT);
1868 	if (error != 0) {
1869 		m_freem(m);
1870 		return (ENOBUFS);
1871 	}
1872 	KASSERT(nsegs == 1, ("%s: %d segment returned!", __func__, nsegs));
1873 
1874 	rxd = &sc->rl_ldata.rl_rx_desc[idx];
1875 	if (rxd->rx_m != NULL) {
1876 		bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap,
1877 		    BUS_DMASYNC_POSTREAD);
1878 		bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap);
1879 	}
1880 
1881 	rxd->rx_m = m;
1882 	map = rxd->rx_dmamap;
1883 	rxd->rx_dmamap = sc->rl_ldata.rl_rx_sparemap;
1884 	rxd->rx_size = segs[0].ds_len;
1885 	sc->rl_ldata.rl_rx_sparemap = map;
1886 	bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap,
1887 	    BUS_DMASYNC_PREREAD);
1888 
1889 	desc = &sc->rl_ldata.rl_rx_list[idx];
1890 	desc->rl_vlanctl = 0;
1891 	desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[0].ds_addr));
1892 	desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[0].ds_addr));
1893 	cmdstat = segs[0].ds_len;
1894 	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
1895 		cmdstat |= RL_RDESC_CMD_EOR;
1896 	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
1897 
1898 	return (0);
1899 }
1900 
1901 static int
1902 re_jumbo_newbuf(struct rl_softc *sc, int idx)
1903 {
1904 	struct mbuf		*m;
1905 	struct rl_rxdesc	*rxd;
1906 	bus_dma_segment_t	segs[1];
1907 	bus_dmamap_t		map;
1908 	struct rl_desc		*desc;
1909 	uint32_t		cmdstat;
1910 	int			error, nsegs;
1911 
1912 	m = m_getjcl(M_DONTWAIT, MT_DATA, M_PKTHDR, MJUM9BYTES);
1913 	if (m == NULL)
1914 		return (ENOBUFS);
1915 	m->m_len = m->m_pkthdr.len = MJUM9BYTES;
1916 #ifdef RE_FIXUP_RX
1917 	m_adj(m, RE_ETHER_ALIGN);
1918 #endif
1919 	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_jrx_mtag,
1920 	    sc->rl_ldata.rl_jrx_sparemap, m, segs, &nsegs, BUS_DMA_NOWAIT);
1921 	if (error != 0) {
1922 		m_freem(m);
1923 		return (ENOBUFS);
1924 	}
1925 	KASSERT(nsegs == 1, ("%s: %d segment returned!", __func__, nsegs));
1926 
1927 	rxd = &sc->rl_ldata.rl_jrx_desc[idx];
1928 	if (rxd->rx_m != NULL) {
1929 		bus_dmamap_sync(sc->rl_ldata.rl_jrx_mtag, rxd->rx_dmamap,
1930 		    BUS_DMASYNC_POSTREAD);
1931 		bus_dmamap_unload(sc->rl_ldata.rl_jrx_mtag, rxd->rx_dmamap);
1932 	}
1933 
1934 	rxd->rx_m = m;
1935 	map = rxd->rx_dmamap;
1936 	rxd->rx_dmamap = sc->rl_ldata.rl_jrx_sparemap;
1937 	rxd->rx_size = segs[0].ds_len;
1938 	sc->rl_ldata.rl_jrx_sparemap = map;
1939 	bus_dmamap_sync(sc->rl_ldata.rl_jrx_mtag, rxd->rx_dmamap,
1940 	    BUS_DMASYNC_PREREAD);
1941 
1942 	desc = &sc->rl_ldata.rl_rx_list[idx];
1943 	desc->rl_vlanctl = 0;
1944 	desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[0].ds_addr));
1945 	desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[0].ds_addr));
1946 	cmdstat = segs[0].ds_len;
1947 	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
1948 		cmdstat |= RL_RDESC_CMD_EOR;
1949 	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
1950 
1951 	return (0);
1952 }
1953 
1954 #ifdef RE_FIXUP_RX
1955 static __inline void
1956 re_fixup_rx(struct mbuf *m)
1957 {
1958 	int                     i;
1959 	uint16_t                *src, *dst;
1960 
1961 	src = mtod(m, uint16_t *);
1962 	dst = src - (RE_ETHER_ALIGN - ETHER_ALIGN) / sizeof *src;
1963 
1964 	for (i = 0; i < (m->m_len / sizeof(uint16_t) + 1); i++)
1965 		*dst++ = *src++;
1966 
1967 	m->m_data -= RE_ETHER_ALIGN - ETHER_ALIGN;
1968 }
1969 #endif
1970 
1971 static int
1972 re_tx_list_init(struct rl_softc *sc)
1973 {
1974 	struct rl_desc		*desc;
1975 	int			i;
1976 
1977 	RL_LOCK_ASSERT(sc);
1978 
1979 	bzero(sc->rl_ldata.rl_tx_list,
1980 	    sc->rl_ldata.rl_tx_desc_cnt * sizeof(struct rl_desc));
1981 	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++)
1982 		sc->rl_ldata.rl_tx_desc[i].tx_m = NULL;
1983 	/* Set EOR. */
1984 	desc = &sc->rl_ldata.rl_tx_list[sc->rl_ldata.rl_tx_desc_cnt - 1];
1985 	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_EOR);
1986 
1987 	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
1988 	    sc->rl_ldata.rl_tx_list_map,
1989 	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1990 
1991 	sc->rl_ldata.rl_tx_prodidx = 0;
1992 	sc->rl_ldata.rl_tx_considx = 0;
1993 	sc->rl_ldata.rl_tx_free = sc->rl_ldata.rl_tx_desc_cnt;
1994 
1995 	return (0);
1996 }
1997 
1998 static int
1999 re_rx_list_init(struct rl_softc *sc)
2000 {
2001 	int			error, i;
2002 
2003 	bzero(sc->rl_ldata.rl_rx_list,
2004 	    sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc));
2005 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
2006 		sc->rl_ldata.rl_rx_desc[i].rx_m = NULL;
2007 		if ((error = re_newbuf(sc, i)) != 0)
2008 			return (error);
2009 	}
2010 
2011 	/* Flush the RX descriptors */
2012 
2013 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
2014 	    sc->rl_ldata.rl_rx_list_map,
2015 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2016 
2017 	sc->rl_ldata.rl_rx_prodidx = 0;
2018 	sc->rl_head = sc->rl_tail = NULL;
2019 	sc->rl_int_rx_act = 0;
2020 
2021 	return (0);
2022 }
2023 
2024 static int
2025 re_jrx_list_init(struct rl_softc *sc)
2026 {
2027 	int			error, i;
2028 
2029 	bzero(sc->rl_ldata.rl_rx_list,
2030 	    sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc));
2031 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
2032 		sc->rl_ldata.rl_jrx_desc[i].rx_m = NULL;
2033 		if ((error = re_jumbo_newbuf(sc, i)) != 0)
2034 			return (error);
2035 	}
2036 
2037 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
2038 	    sc->rl_ldata.rl_rx_list_map,
2039 	    BUS_DMASYNC_PREWRITE | BUS_DMASYNC_PREREAD);
2040 
2041 	sc->rl_ldata.rl_rx_prodidx = 0;
2042 	sc->rl_head = sc->rl_tail = NULL;
2043 	sc->rl_int_rx_act = 0;
2044 
2045 	return (0);
2046 }
2047 
2048 /*
2049  * RX handler for C+ and 8169. For the gigE chips, we support
2050  * the reception of jumbo frames that have been fragmented
2051  * across multiple 2K mbuf cluster buffers.
2052  */
2053 static int
2054 re_rxeof(struct rl_softc *sc, int *rx_npktsp)
2055 {
2056 	struct mbuf		*m;
2057 	struct ifnet		*ifp;
2058 	int			i, rxerr, total_len;
2059 	struct rl_desc		*cur_rx;
2060 	u_int32_t		rxstat, rxvlan;
2061 	int			jumbo, maxpkt = 16, rx_npkts = 0;
2062 
2063 	RL_LOCK_ASSERT(sc);
2064 
2065 	ifp = sc->rl_ifp;
2066 	if (ifp->if_mtu > RL_MTU && (sc->rl_flags & RL_FLAG_JUMBOV2) != 0)
2067 		jumbo = 1;
2068 	else
2069 		jumbo = 0;
2070 
2071 	/* Invalidate the descriptor memory */
2072 
2073 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
2074 	    sc->rl_ldata.rl_rx_list_map,
2075 	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2076 
2077 	for (i = sc->rl_ldata.rl_rx_prodidx; maxpkt > 0;
2078 	    i = RL_RX_DESC_NXT(sc, i)) {
2079 		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
2080 			break;
2081 		cur_rx = &sc->rl_ldata.rl_rx_list[i];
2082 		rxstat = le32toh(cur_rx->rl_cmdstat);
2083 		if ((rxstat & RL_RDESC_STAT_OWN) != 0)
2084 			break;
2085 		total_len = rxstat & sc->rl_rxlenmask;
2086 		rxvlan = le32toh(cur_rx->rl_vlanctl);
2087 		if (jumbo != 0)
2088 			m = sc->rl_ldata.rl_jrx_desc[i].rx_m;
2089 		else
2090 			m = sc->rl_ldata.rl_rx_desc[i].rx_m;
2091 
2092 		if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0 &&
2093 		    (rxstat & (RL_RDESC_STAT_SOF | RL_RDESC_STAT_EOF)) !=
2094 		    (RL_RDESC_STAT_SOF | RL_RDESC_STAT_EOF)) {
2095 			/*
2096 			 * RTL8168C or later controllers do not
2097 			 * support multi-fragment packet.
2098 			 */
2099 			re_discard_rxbuf(sc, i);
2100 			continue;
2101 		} else if ((rxstat & RL_RDESC_STAT_EOF) == 0) {
2102 			if (re_newbuf(sc, i) != 0) {
2103 				/*
2104 				 * If this is part of a multi-fragment packet,
2105 				 * discard all the pieces.
2106 				 */
2107 				if (sc->rl_head != NULL) {
2108 					m_freem(sc->rl_head);
2109 					sc->rl_head = sc->rl_tail = NULL;
2110 				}
2111 				re_discard_rxbuf(sc, i);
2112 				continue;
2113 			}
2114 			m->m_len = RE_RX_DESC_BUFLEN;
2115 			if (sc->rl_head == NULL)
2116 				sc->rl_head = sc->rl_tail = m;
2117 			else {
2118 				m->m_flags &= ~M_PKTHDR;
2119 				sc->rl_tail->m_next = m;
2120 				sc->rl_tail = m;
2121 			}
2122 			continue;
2123 		}
2124 
2125 		/*
2126 		 * NOTE: for the 8139C+, the frame length field
2127 		 * is always 12 bits in size, but for the gigE chips,
2128 		 * it is 13 bits (since the max RX frame length is 16K).
2129 		 * Unfortunately, all 32 bits in the status word
2130 		 * were already used, so to make room for the extra
2131 		 * length bit, RealTek took out the 'frame alignment
2132 		 * error' bit and shifted the other status bits
2133 		 * over one slot. The OWN, EOR, FS and LS bits are
2134 		 * still in the same places. We have already extracted
2135 		 * the frame length and checked the OWN bit, so rather
2136 		 * than using an alternate bit mapping, we shift the
2137 		 * status bits one space to the right so we can evaluate
2138 		 * them using the 8169 status as though it was in the
2139 		 * same format as that of the 8139C+.
2140 		 */
2141 		if (sc->rl_type == RL_8169)
2142 			rxstat >>= 1;
2143 
2144 		/*
2145 		 * if total_len > 2^13-1, both _RXERRSUM and _GIANT will be
2146 		 * set, but if CRC is clear, it will still be a valid frame.
2147 		 */
2148 		if ((rxstat & RL_RDESC_STAT_RXERRSUM) != 0) {
2149 			rxerr = 1;
2150 			if ((sc->rl_flags & RL_FLAG_JUMBOV2) == 0 &&
2151 			    total_len > 8191 &&
2152 			    (rxstat & RL_RDESC_STAT_ERRS) == RL_RDESC_STAT_GIANT)
2153 				rxerr = 0;
2154 			if (rxerr != 0) {
2155 				ifp->if_ierrors++;
2156 				/*
2157 				 * If this is part of a multi-fragment packet,
2158 				 * discard all the pieces.
2159 				 */
2160 				if (sc->rl_head != NULL) {
2161 					m_freem(sc->rl_head);
2162 					sc->rl_head = sc->rl_tail = NULL;
2163 				}
2164 				re_discard_rxbuf(sc, i);
2165 				continue;
2166 			}
2167 		}
2168 
2169 		/*
2170 		 * If allocating a replacement mbuf fails,
2171 		 * reload the current one.
2172 		 */
2173 		if (jumbo != 0)
2174 			rxerr = re_jumbo_newbuf(sc, i);
2175 		else
2176 			rxerr = re_newbuf(sc, i);
2177 		if (rxerr != 0) {
2178 			ifp->if_iqdrops++;
2179 			if (sc->rl_head != NULL) {
2180 				m_freem(sc->rl_head);
2181 				sc->rl_head = sc->rl_tail = NULL;
2182 			}
2183 			re_discard_rxbuf(sc, i);
2184 			continue;
2185 		}
2186 
2187 		if (sc->rl_head != NULL) {
2188 			if (jumbo != 0)
2189 				m->m_len = total_len;
2190 			else {
2191 				m->m_len = total_len % RE_RX_DESC_BUFLEN;
2192 				if (m->m_len == 0)
2193 					m->m_len = RE_RX_DESC_BUFLEN;
2194 			}
2195 			/*
2196 			 * Special case: if there's 4 bytes or less
2197 			 * in this buffer, the mbuf can be discarded:
2198 			 * the last 4 bytes is the CRC, which we don't
2199 			 * care about anyway.
2200 			 */
2201 			if (m->m_len <= ETHER_CRC_LEN) {
2202 				sc->rl_tail->m_len -=
2203 				    (ETHER_CRC_LEN - m->m_len);
2204 				m_freem(m);
2205 			} else {
2206 				m->m_len -= ETHER_CRC_LEN;
2207 				m->m_flags &= ~M_PKTHDR;
2208 				sc->rl_tail->m_next = m;
2209 			}
2210 			m = sc->rl_head;
2211 			sc->rl_head = sc->rl_tail = NULL;
2212 			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
2213 		} else
2214 			m->m_pkthdr.len = m->m_len =
2215 			    (total_len - ETHER_CRC_LEN);
2216 
2217 #ifdef RE_FIXUP_RX
2218 		re_fixup_rx(m);
2219 #endif
2220 		ifp->if_ipackets++;
2221 		m->m_pkthdr.rcvif = ifp;
2222 
2223 		/* Do RX checksumming if enabled */
2224 
2225 		if (ifp->if_capenable & IFCAP_RXCSUM) {
2226 			if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
2227 				/* Check IP header checksum */
2228 				if (rxstat & RL_RDESC_STAT_PROTOID)
2229 					m->m_pkthdr.csum_flags |=
2230 					    CSUM_IP_CHECKED;
2231 				if (!(rxstat & RL_RDESC_STAT_IPSUMBAD))
2232 					m->m_pkthdr.csum_flags |=
2233 					    CSUM_IP_VALID;
2234 
2235 				/* Check TCP/UDP checksum */
2236 				if ((RL_TCPPKT(rxstat) &&
2237 				    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
2238 				    (RL_UDPPKT(rxstat) &&
2239 				     !(rxstat & RL_RDESC_STAT_UDPSUMBAD))) {
2240 					m->m_pkthdr.csum_flags |=
2241 						CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
2242 					m->m_pkthdr.csum_data = 0xffff;
2243 				}
2244 			} else {
2245 				/*
2246 				 * RTL8168C/RTL816CP/RTL8111C/RTL8111CP
2247 				 */
2248 				if ((rxstat & RL_RDESC_STAT_PROTOID) &&
2249 				    (rxvlan & RL_RDESC_IPV4))
2250 					m->m_pkthdr.csum_flags |=
2251 					    CSUM_IP_CHECKED;
2252 				if (!(rxstat & RL_RDESC_STAT_IPSUMBAD) &&
2253 				    (rxvlan & RL_RDESC_IPV4))
2254 					m->m_pkthdr.csum_flags |=
2255 					    CSUM_IP_VALID;
2256 				if (((rxstat & RL_RDESC_STAT_TCP) &&
2257 				    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
2258 				    ((rxstat & RL_RDESC_STAT_UDP) &&
2259 				    !(rxstat & RL_RDESC_STAT_UDPSUMBAD))) {
2260 					m->m_pkthdr.csum_flags |=
2261 						CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
2262 					m->m_pkthdr.csum_data = 0xffff;
2263 				}
2264 			}
2265 		}
2266 		maxpkt--;
2267 		if (rxvlan & RL_RDESC_VLANCTL_TAG) {
2268 			m->m_pkthdr.ether_vtag =
2269 			    bswap16((rxvlan & RL_RDESC_VLANCTL_DATA));
2270 			m->m_flags |= M_VLANTAG;
2271 		}
2272 		RL_UNLOCK(sc);
2273 		(*ifp->if_input)(ifp, m);
2274 		RL_LOCK(sc);
2275 		rx_npkts++;
2276 	}
2277 
2278 	/* Flush the RX DMA ring */
2279 
2280 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
2281 	    sc->rl_ldata.rl_rx_list_map,
2282 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2283 
2284 	sc->rl_ldata.rl_rx_prodidx = i;
2285 
2286 	if (rx_npktsp != NULL)
2287 		*rx_npktsp = rx_npkts;
2288 	if (maxpkt)
2289 		return (EAGAIN);
2290 
2291 	return (0);
2292 }
2293 
2294 static void
2295 re_txeof(struct rl_softc *sc)
2296 {
2297 	struct ifnet		*ifp;
2298 	struct rl_txdesc	*txd;
2299 	u_int32_t		txstat;
2300 	int			cons;
2301 
2302 	cons = sc->rl_ldata.rl_tx_considx;
2303 	if (cons == sc->rl_ldata.rl_tx_prodidx)
2304 		return;
2305 
2306 	ifp = sc->rl_ifp;
2307 	/* Invalidate the TX descriptor list */
2308 	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2309 	    sc->rl_ldata.rl_tx_list_map,
2310 	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2311 
2312 	for (; cons != sc->rl_ldata.rl_tx_prodidx;
2313 	    cons = RL_TX_DESC_NXT(sc, cons)) {
2314 		txstat = le32toh(sc->rl_ldata.rl_tx_list[cons].rl_cmdstat);
2315 		if (txstat & RL_TDESC_STAT_OWN)
2316 			break;
2317 		/*
2318 		 * We only stash mbufs in the last descriptor
2319 		 * in a fragment chain, which also happens to
2320 		 * be the only place where the TX status bits
2321 		 * are valid.
2322 		 */
2323 		if (txstat & RL_TDESC_CMD_EOF) {
2324 			txd = &sc->rl_ldata.rl_tx_desc[cons];
2325 			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
2326 			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
2327 			bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag,
2328 			    txd->tx_dmamap);
2329 			KASSERT(txd->tx_m != NULL,
2330 			    ("%s: freeing NULL mbufs!", __func__));
2331 			m_freem(txd->tx_m);
2332 			txd->tx_m = NULL;
2333 			if (txstat & (RL_TDESC_STAT_EXCESSCOL|
2334 			    RL_TDESC_STAT_COLCNT))
2335 				ifp->if_collisions++;
2336 			if (txstat & RL_TDESC_STAT_TXERRSUM)
2337 				ifp->if_oerrors++;
2338 			else
2339 				ifp->if_opackets++;
2340 		}
2341 		sc->rl_ldata.rl_tx_free++;
2342 		ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
2343 	}
2344 	sc->rl_ldata.rl_tx_considx = cons;
2345 
2346 	/* No changes made to the TX ring, so no flush needed */
2347 
2348 	if (sc->rl_ldata.rl_tx_free != sc->rl_ldata.rl_tx_desc_cnt) {
2349 #ifdef RE_TX_MODERATION
2350 		/*
2351 		 * If not all descriptors have been reaped yet, reload
2352 		 * the timer so that we will eventually get another
2353 		 * interrupt that will cause us to re-enter this routine.
2354 		 * This is done in case the transmitter has gone idle.
2355 		 */
2356 		CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2357 #endif
2358 	} else
2359 		sc->rl_watchdog_timer = 0;
2360 }
2361 
2362 static void
2363 re_tick(void *xsc)
2364 {
2365 	struct rl_softc		*sc;
2366 	struct mii_data		*mii;
2367 
2368 	sc = xsc;
2369 
2370 	RL_LOCK_ASSERT(sc);
2371 
2372 	mii = device_get_softc(sc->rl_miibus);
2373 	mii_tick(mii);
2374 	if ((sc->rl_flags & RL_FLAG_LINK) == 0)
2375 		re_miibus_statchg(sc->rl_dev);
2376 	/*
2377 	 * Reclaim transmitted frames here. Technically it is not
2378 	 * necessary to do here but it ensures periodic reclamation
2379 	 * regardless of Tx completion interrupt which seems to be
2380 	 * lost on PCIe based controllers under certain situations.
2381 	 */
2382 	re_txeof(sc);
2383 	re_watchdog(sc);
2384 	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
2385 }
2386 
2387 #ifdef DEVICE_POLLING
2388 static int
2389 re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
2390 {
2391 	struct rl_softc *sc = ifp->if_softc;
2392 	int rx_npkts = 0;
2393 
2394 	RL_LOCK(sc);
2395 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
2396 		rx_npkts = re_poll_locked(ifp, cmd, count);
2397 	RL_UNLOCK(sc);
2398 	return (rx_npkts);
2399 }
2400 
2401 static int
2402 re_poll_locked(struct ifnet *ifp, enum poll_cmd cmd, int count)
2403 {
2404 	struct rl_softc *sc = ifp->if_softc;
2405 	int rx_npkts;
2406 
2407 	RL_LOCK_ASSERT(sc);
2408 
2409 	sc->rxcycles = count;
2410 	re_rxeof(sc, &rx_npkts);
2411 	re_txeof(sc);
2412 
2413 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2414 		re_start_locked(ifp);
2415 
2416 	if (cmd == POLL_AND_CHECK_STATUS) { /* also check status register */
2417 		u_int16_t       status;
2418 
2419 		status = CSR_READ_2(sc, RL_ISR);
2420 		if (status == 0xffff)
2421 			return (rx_npkts);
2422 		if (status)
2423 			CSR_WRITE_2(sc, RL_ISR, status);
2424 		if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
2425 		    (sc->rl_flags & RL_FLAG_PCIE))
2426 			CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2427 
2428 		/*
2429 		 * XXX check behaviour on receiver stalls.
2430 		 */
2431 
2432 		if (status & RL_ISR_SYSTEM_ERR) {
2433 			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2434 			re_init_locked(sc);
2435 		}
2436 	}
2437 	return (rx_npkts);
2438 }
2439 #endif /* DEVICE_POLLING */
2440 
2441 static int
2442 re_intr(void *arg)
2443 {
2444 	struct rl_softc		*sc;
2445 	uint16_t		status;
2446 
2447 	sc = arg;
2448 
2449 	status = CSR_READ_2(sc, RL_ISR);
2450 	if (status == 0xFFFF || (status & RL_INTRS_CPLUS) == 0)
2451                 return (FILTER_STRAY);
2452 	CSR_WRITE_2(sc, RL_IMR, 0);
2453 
2454 	taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_inttask);
2455 
2456 	return (FILTER_HANDLED);
2457 }
2458 
2459 static void
2460 re_int_task(void *arg, int npending)
2461 {
2462 	struct rl_softc		*sc;
2463 	struct ifnet		*ifp;
2464 	u_int16_t		status;
2465 	int			rval = 0;
2466 
2467 	sc = arg;
2468 	ifp = sc->rl_ifp;
2469 
2470 	RL_LOCK(sc);
2471 
2472 	status = CSR_READ_2(sc, RL_ISR);
2473         CSR_WRITE_2(sc, RL_ISR, status);
2474 
2475 	if (sc->suspended ||
2476 	    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2477 		RL_UNLOCK(sc);
2478 		return;
2479 	}
2480 
2481 #ifdef DEVICE_POLLING
2482 	if  (ifp->if_capenable & IFCAP_POLLING) {
2483 		RL_UNLOCK(sc);
2484 		return;
2485 	}
2486 #endif
2487 
2488 	if (status & (RL_ISR_RX_OK|RL_ISR_RX_ERR|RL_ISR_FIFO_OFLOW))
2489 		rval = re_rxeof(sc, NULL);
2490 
2491 	/*
2492 	 * Some chips will ignore a second TX request issued
2493 	 * while an existing transmission is in progress. If
2494 	 * the transmitter goes idle but there are still
2495 	 * packets waiting to be sent, we need to restart the
2496 	 * channel here to flush them out. This only seems to
2497 	 * be required with the PCIe devices.
2498 	 */
2499 	if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
2500 	    (sc->rl_flags & RL_FLAG_PCIE))
2501 		CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2502 	if (status & (
2503 #ifdef RE_TX_MODERATION
2504 	    RL_ISR_TIMEOUT_EXPIRED|
2505 #else
2506 	    RL_ISR_TX_OK|
2507 #endif
2508 	    RL_ISR_TX_ERR|RL_ISR_TX_DESC_UNAVAIL))
2509 		re_txeof(sc);
2510 
2511 	if (status & RL_ISR_SYSTEM_ERR) {
2512 		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2513 		re_init_locked(sc);
2514 	}
2515 
2516 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2517 		re_start_locked(ifp);
2518 
2519 	RL_UNLOCK(sc);
2520 
2521         if ((CSR_READ_2(sc, RL_ISR) & RL_INTRS_CPLUS) || rval) {
2522 		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_inttask);
2523 		return;
2524 	}
2525 
2526 	CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2527 }
2528 
2529 static void
2530 re_intr_msi(void *xsc)
2531 {
2532 	struct rl_softc		*sc;
2533 	struct ifnet		*ifp;
2534 	uint16_t		intrs, status;
2535 
2536 	sc = xsc;
2537 	RL_LOCK(sc);
2538 
2539 	ifp = sc->rl_ifp;
2540 #ifdef DEVICE_POLLING
2541 	if (ifp->if_capenable & IFCAP_POLLING) {
2542 		RL_UNLOCK(sc);
2543 		return;
2544 	}
2545 #endif
2546 	/* Disable interrupts. */
2547 	CSR_WRITE_2(sc, RL_IMR, 0);
2548 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2549 		RL_UNLOCK(sc);
2550 		return;
2551 	}
2552 
2553 	intrs = RL_INTRS_CPLUS;
2554 	status = CSR_READ_2(sc, RL_ISR);
2555         CSR_WRITE_2(sc, RL_ISR, status);
2556 	if (sc->rl_int_rx_act > 0) {
2557 		intrs &= ~(RL_ISR_RX_OK | RL_ISR_RX_ERR | RL_ISR_FIFO_OFLOW |
2558 		    RL_ISR_RX_OVERRUN);
2559 		status &= ~(RL_ISR_RX_OK | RL_ISR_RX_ERR | RL_ISR_FIFO_OFLOW |
2560 		    RL_ISR_RX_OVERRUN);
2561 	}
2562 
2563 	if (status & (RL_ISR_TIMEOUT_EXPIRED | RL_ISR_RX_OK | RL_ISR_RX_ERR |
2564 	    RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN)) {
2565 		re_rxeof(sc, NULL);
2566 		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
2567 			if (sc->rl_int_rx_mod != 0 &&
2568 			    (status & (RL_ISR_RX_OK | RL_ISR_RX_ERR |
2569 			    RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN)) != 0) {
2570 				/* Rearm one-shot timer. */
2571 				CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2572 				intrs &= ~(RL_ISR_RX_OK | RL_ISR_RX_ERR |
2573 				    RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN);
2574 				sc->rl_int_rx_act = 1;
2575 			} else {
2576 				intrs |= RL_ISR_RX_OK | RL_ISR_RX_ERR |
2577 				    RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN;
2578 				sc->rl_int_rx_act = 0;
2579 			}
2580 		}
2581 	}
2582 
2583 	/*
2584 	 * Some chips will ignore a second TX request issued
2585 	 * while an existing transmission is in progress. If
2586 	 * the transmitter goes idle but there are still
2587 	 * packets waiting to be sent, we need to restart the
2588 	 * channel here to flush them out. This only seems to
2589 	 * be required with the PCIe devices.
2590 	 */
2591 	if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
2592 	    (sc->rl_flags & RL_FLAG_PCIE))
2593 		CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2594 	if (status & (RL_ISR_TX_OK | RL_ISR_TX_ERR | RL_ISR_TX_DESC_UNAVAIL))
2595 		re_txeof(sc);
2596 
2597 	if (status & RL_ISR_SYSTEM_ERR) {
2598 		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2599 		re_init_locked(sc);
2600 	}
2601 
2602 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
2603 		if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2604 			re_start_locked(ifp);
2605 		CSR_WRITE_2(sc, RL_IMR, intrs);
2606 	}
2607 	RL_UNLOCK(sc);
2608 }
2609 
2610 static int
2611 re_encap(struct rl_softc *sc, struct mbuf **m_head)
2612 {
2613 	struct rl_txdesc	*txd, *txd_last;
2614 	bus_dma_segment_t	segs[RL_NTXSEGS];
2615 	bus_dmamap_t		map;
2616 	struct mbuf		*m_new;
2617 	struct rl_desc		*desc;
2618 	int			nsegs, prod;
2619 	int			i, error, ei, si;
2620 	int			padlen;
2621 	uint32_t		cmdstat, csum_flags, vlanctl;
2622 
2623 	RL_LOCK_ASSERT(sc);
2624 	M_ASSERTPKTHDR((*m_head));
2625 
2626 	/*
2627 	 * With some of the RealTek chips, using the checksum offload
2628 	 * support in conjunction with the autopadding feature results
2629 	 * in the transmission of corrupt frames. For example, if we
2630 	 * need to send a really small IP fragment that's less than 60
2631 	 * bytes in size, and IP header checksumming is enabled, the
2632 	 * resulting ethernet frame that appears on the wire will
2633 	 * have garbled payload. To work around this, if TX IP checksum
2634 	 * offload is enabled, we always manually pad short frames out
2635 	 * to the minimum ethernet frame size.
2636 	 */
2637 	if ((sc->rl_flags & RL_FLAG_AUTOPAD) == 0 &&
2638 	    (*m_head)->m_pkthdr.len < RL_IP4CSUMTX_PADLEN &&
2639 	    ((*m_head)->m_pkthdr.csum_flags & CSUM_IP) != 0) {
2640 		padlen = RL_MIN_FRAMELEN - (*m_head)->m_pkthdr.len;
2641 		if (M_WRITABLE(*m_head) == 0) {
2642 			/* Get a writable copy. */
2643 			m_new = m_dup(*m_head, M_DONTWAIT);
2644 			m_freem(*m_head);
2645 			if (m_new == NULL) {
2646 				*m_head = NULL;
2647 				return (ENOBUFS);
2648 			}
2649 			*m_head = m_new;
2650 		}
2651 		if ((*m_head)->m_next != NULL ||
2652 		    M_TRAILINGSPACE(*m_head) < padlen) {
2653 			m_new = m_defrag(*m_head, M_DONTWAIT);
2654 			if (m_new == NULL) {
2655 				m_freem(*m_head);
2656 				*m_head = NULL;
2657 				return (ENOBUFS);
2658 			}
2659 		} else
2660 			m_new = *m_head;
2661 
2662 		/*
2663 		 * Manually pad short frames, and zero the pad space
2664 		 * to avoid leaking data.
2665 		 */
2666 		bzero(mtod(m_new, char *) + m_new->m_pkthdr.len, padlen);
2667 		m_new->m_pkthdr.len += padlen;
2668 		m_new->m_len = m_new->m_pkthdr.len;
2669 		*m_head = m_new;
2670 	}
2671 
2672 	prod = sc->rl_ldata.rl_tx_prodidx;
2673 	txd = &sc->rl_ldata.rl_tx_desc[prod];
2674 	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap,
2675 	    *m_head, segs, &nsegs, BUS_DMA_NOWAIT);
2676 	if (error == EFBIG) {
2677 		m_new = m_collapse(*m_head, M_DONTWAIT, RL_NTXSEGS);
2678 		if (m_new == NULL) {
2679 			m_freem(*m_head);
2680 			*m_head = NULL;
2681 			return (ENOBUFS);
2682 		}
2683 		*m_head = m_new;
2684 		error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_tx_mtag,
2685 		    txd->tx_dmamap, *m_head, segs, &nsegs, BUS_DMA_NOWAIT);
2686 		if (error != 0) {
2687 			m_freem(*m_head);
2688 			*m_head = NULL;
2689 			return (error);
2690 		}
2691 	} else if (error != 0)
2692 		return (error);
2693 	if (nsegs == 0) {
2694 		m_freem(*m_head);
2695 		*m_head = NULL;
2696 		return (EIO);
2697 	}
2698 
2699 	/* Check for number of available descriptors. */
2700 	if (sc->rl_ldata.rl_tx_free - nsegs <= 1) {
2701 		bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap);
2702 		return (ENOBUFS);
2703 	}
2704 
2705 	bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap,
2706 	    BUS_DMASYNC_PREWRITE);
2707 
2708 	/*
2709 	 * Set up checksum offload. Note: checksum offload bits must
2710 	 * appear in all descriptors of a multi-descriptor transmit
2711 	 * attempt. This is according to testing done with an 8169
2712 	 * chip. This is a requirement.
2713 	 */
2714 	vlanctl = 0;
2715 	csum_flags = 0;
2716 	if (((*m_head)->m_pkthdr.csum_flags & CSUM_TSO) != 0) {
2717 		if ((sc->rl_flags & RL_FLAG_DESCV2) != 0) {
2718 			csum_flags |= RL_TDESC_CMD_LGSEND;
2719 			vlanctl |= ((uint32_t)(*m_head)->m_pkthdr.tso_segsz <<
2720 			    RL_TDESC_CMD_MSSVALV2_SHIFT);
2721 		} else {
2722 			csum_flags |= RL_TDESC_CMD_LGSEND |
2723 			    ((uint32_t)(*m_head)->m_pkthdr.tso_segsz <<
2724 			    RL_TDESC_CMD_MSSVAL_SHIFT);
2725 		}
2726 	} else {
2727 		/*
2728 		 * Unconditionally enable IP checksum if TCP or UDP
2729 		 * checksum is required. Otherwise, TCP/UDP checksum
2730 		 * does't make effects.
2731 		 */
2732 		if (((*m_head)->m_pkthdr.csum_flags & RE_CSUM_FEATURES) != 0) {
2733 			if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
2734 				csum_flags |= RL_TDESC_CMD_IPCSUM;
2735 				if (((*m_head)->m_pkthdr.csum_flags &
2736 				    CSUM_TCP) != 0)
2737 					csum_flags |= RL_TDESC_CMD_TCPCSUM;
2738 				if (((*m_head)->m_pkthdr.csum_flags &
2739 				    CSUM_UDP) != 0)
2740 					csum_flags |= RL_TDESC_CMD_UDPCSUM;
2741 			} else {
2742 				vlanctl |= RL_TDESC_CMD_IPCSUMV2;
2743 				if (((*m_head)->m_pkthdr.csum_flags &
2744 				    CSUM_TCP) != 0)
2745 					vlanctl |= RL_TDESC_CMD_TCPCSUMV2;
2746 				if (((*m_head)->m_pkthdr.csum_flags &
2747 				    CSUM_UDP) != 0)
2748 					vlanctl |= RL_TDESC_CMD_UDPCSUMV2;
2749 			}
2750 		}
2751 	}
2752 
2753 	/*
2754 	 * Set up hardware VLAN tagging. Note: vlan tag info must
2755 	 * appear in all descriptors of a multi-descriptor
2756 	 * transmission attempt.
2757 	 */
2758 	if ((*m_head)->m_flags & M_VLANTAG)
2759 		vlanctl |= bswap16((*m_head)->m_pkthdr.ether_vtag) |
2760 		    RL_TDESC_VLANCTL_TAG;
2761 
2762 	si = prod;
2763 	for (i = 0; i < nsegs; i++, prod = RL_TX_DESC_NXT(sc, prod)) {
2764 		desc = &sc->rl_ldata.rl_tx_list[prod];
2765 		desc->rl_vlanctl = htole32(vlanctl);
2766 		desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[i].ds_addr));
2767 		desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[i].ds_addr));
2768 		cmdstat = segs[i].ds_len;
2769 		if (i != 0)
2770 			cmdstat |= RL_TDESC_CMD_OWN;
2771 		if (prod == sc->rl_ldata.rl_tx_desc_cnt - 1)
2772 			cmdstat |= RL_TDESC_CMD_EOR;
2773 		desc->rl_cmdstat = htole32(cmdstat | csum_flags);
2774 		sc->rl_ldata.rl_tx_free--;
2775 	}
2776 	/* Update producer index. */
2777 	sc->rl_ldata.rl_tx_prodidx = prod;
2778 
2779 	/* Set EOF on the last descriptor. */
2780 	ei = RL_TX_DESC_PRV(sc, prod);
2781 	desc = &sc->rl_ldata.rl_tx_list[ei];
2782 	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_EOF);
2783 
2784 	desc = &sc->rl_ldata.rl_tx_list[si];
2785 	/* Set SOF and transfer ownership of packet to the chip. */
2786 	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_OWN | RL_TDESC_CMD_SOF);
2787 
2788 	/*
2789 	 * Insure that the map for this transmission
2790 	 * is placed at the array index of the last descriptor
2791 	 * in this chain.  (Swap last and first dmamaps.)
2792 	 */
2793 	txd_last = &sc->rl_ldata.rl_tx_desc[ei];
2794 	map = txd->tx_dmamap;
2795 	txd->tx_dmamap = txd_last->tx_dmamap;
2796 	txd_last->tx_dmamap = map;
2797 	txd_last->tx_m = *m_head;
2798 
2799 	return (0);
2800 }
2801 
2802 static void
2803 re_start(struct ifnet *ifp)
2804 {
2805 	struct rl_softc		*sc;
2806 
2807 	sc = ifp->if_softc;
2808 	RL_LOCK(sc);
2809 	re_start_locked(ifp);
2810 	RL_UNLOCK(sc);
2811 }
2812 
2813 /*
2814  * Main transmit routine for C+ and gigE NICs.
2815  */
2816 static void
2817 re_start_locked(struct ifnet *ifp)
2818 {
2819 	struct rl_softc		*sc;
2820 	struct mbuf		*m_head;
2821 	int			queued;
2822 
2823 	sc = ifp->if_softc;
2824 
2825 	if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
2826 	    IFF_DRV_RUNNING || (sc->rl_flags & RL_FLAG_LINK) == 0)
2827 		return;
2828 
2829 	for (queued = 0; !IFQ_DRV_IS_EMPTY(&ifp->if_snd) &&
2830 	    sc->rl_ldata.rl_tx_free > 1;) {
2831 		IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
2832 		if (m_head == NULL)
2833 			break;
2834 
2835 		if (re_encap(sc, &m_head) != 0) {
2836 			if (m_head == NULL)
2837 				break;
2838 			IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
2839 			ifp->if_drv_flags |= IFF_DRV_OACTIVE;
2840 			break;
2841 		}
2842 
2843 		/*
2844 		 * If there's a BPF listener, bounce a copy of this frame
2845 		 * to him.
2846 		 */
2847 		ETHER_BPF_MTAP(ifp, m_head);
2848 
2849 		queued++;
2850 	}
2851 
2852 	if (queued == 0) {
2853 #ifdef RE_TX_MODERATION
2854 		if (sc->rl_ldata.rl_tx_free != sc->rl_ldata.rl_tx_desc_cnt)
2855 			CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2856 #endif
2857 		return;
2858 	}
2859 
2860 	/* Flush the TX descriptors */
2861 
2862 	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2863 	    sc->rl_ldata.rl_tx_list_map,
2864 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2865 
2866 	CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2867 
2868 #ifdef RE_TX_MODERATION
2869 	/*
2870 	 * Use the countdown timer for interrupt moderation.
2871 	 * 'TX done' interrupts are disabled. Instead, we reset the
2872 	 * countdown timer, which will begin counting until it hits
2873 	 * the value in the TIMERINT register, and then trigger an
2874 	 * interrupt. Each time we write to the TIMERCNT register,
2875 	 * the timer count is reset to 0.
2876 	 */
2877 	CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2878 #endif
2879 
2880 	/*
2881 	 * Set a timeout in case the chip goes out to lunch.
2882 	 */
2883 	sc->rl_watchdog_timer = 5;
2884 }
2885 
2886 static void
2887 re_set_jumbo(struct rl_softc *sc, int jumbo)
2888 {
2889 
2890 	if (sc->rl_hwrev->rl_rev == RL_HWREV_8168E_VL) {
2891 		pci_set_max_read_req(sc->rl_dev, 4096);
2892 		return;
2893 	}
2894 
2895 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG);
2896 	if (jumbo != 0) {
2897 		CSR_WRITE_1(sc, RL_CFG3, CSR_READ_1(sc, RL_CFG3) |
2898 		    RL_CFG3_JUMBO_EN0);
2899 		switch (sc->rl_hwrev->rl_rev) {
2900 		case RL_HWREV_8168DP:
2901 			break;
2902 		case RL_HWREV_8168E:
2903 			CSR_WRITE_1(sc, RL_CFG4, CSR_READ_1(sc, RL_CFG4) |
2904 			    0x01);
2905 			break;
2906 		default:
2907 			CSR_WRITE_1(sc, RL_CFG4, CSR_READ_1(sc, RL_CFG4) |
2908 			    RL_CFG4_JUMBO_EN1);
2909 		}
2910 	} else {
2911 		CSR_WRITE_1(sc, RL_CFG3, CSR_READ_1(sc, RL_CFG3) &
2912 		    ~RL_CFG3_JUMBO_EN0);
2913 		switch (sc->rl_hwrev->rl_rev) {
2914 		case RL_HWREV_8168DP:
2915 			break;
2916 		case RL_HWREV_8168E:
2917 			CSR_WRITE_1(sc, RL_CFG4, CSR_READ_1(sc, RL_CFG4) &
2918 			    ~0x01);
2919 			break;
2920 		default:
2921 			CSR_WRITE_1(sc, RL_CFG4, CSR_READ_1(sc, RL_CFG4) &
2922 			    ~RL_CFG4_JUMBO_EN1);
2923 		}
2924 	}
2925 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
2926 
2927 	switch (sc->rl_hwrev->rl_rev) {
2928 	case RL_HWREV_8168DP:
2929 		pci_set_max_read_req(sc->rl_dev, 4096);
2930 		break;
2931 	default:
2932 		if (jumbo != 0)
2933 			pci_set_max_read_req(sc->rl_dev, 512);
2934 		else
2935 			pci_set_max_read_req(sc->rl_dev, 4096);
2936 	}
2937 }
2938 
2939 static void
2940 re_init(void *xsc)
2941 {
2942 	struct rl_softc		*sc = xsc;
2943 
2944 	RL_LOCK(sc);
2945 	re_init_locked(sc);
2946 	RL_UNLOCK(sc);
2947 }
2948 
2949 static void
2950 re_init_locked(struct rl_softc *sc)
2951 {
2952 	struct ifnet		*ifp = sc->rl_ifp;
2953 	struct mii_data		*mii;
2954 	uint32_t		reg;
2955 	uint16_t		cfg;
2956 	union {
2957 		uint32_t align_dummy;
2958 		u_char eaddr[ETHER_ADDR_LEN];
2959         } eaddr;
2960 
2961 	RL_LOCK_ASSERT(sc);
2962 
2963 	mii = device_get_softc(sc->rl_miibus);
2964 
2965 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
2966 		return;
2967 
2968 	/*
2969 	 * Cancel pending I/O and free all RX/TX buffers.
2970 	 */
2971 	re_stop(sc);
2972 
2973 	/* Put controller into known state. */
2974 	re_reset(sc);
2975 
2976 	/*
2977 	 * For C+ mode, initialize the RX descriptors and mbufs.
2978 	 */
2979 	if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
2980 		if (ifp->if_mtu > RL_MTU) {
2981 			if (re_jrx_list_init(sc) != 0) {
2982 				device_printf(sc->rl_dev,
2983 				    "no memory for jumbo RX buffers\n");
2984 				re_stop(sc);
2985 				return;
2986 			}
2987 			/* Disable checksum offloading for jumbo frames. */
2988 			ifp->if_capenable &= ~(IFCAP_HWCSUM | IFCAP_TSO4);
2989 			ifp->if_hwassist &= ~(RE_CSUM_FEATURES | CSUM_TSO);
2990 		} else {
2991 			if (re_rx_list_init(sc) != 0) {
2992 				device_printf(sc->rl_dev,
2993 				    "no memory for RX buffers\n");
2994 				re_stop(sc);
2995 				return;
2996 			}
2997 		}
2998 		re_set_jumbo(sc, ifp->if_mtu > RL_MTU);
2999 	} else {
3000 		if (re_rx_list_init(sc) != 0) {
3001 			device_printf(sc->rl_dev, "no memory for RX buffers\n");
3002 			re_stop(sc);
3003 			return;
3004 		}
3005 		if ((sc->rl_flags & RL_FLAG_PCIE) != 0 &&
3006 		    pci_get_device(sc->rl_dev) != RT_DEVICEID_8101E) {
3007 			if (ifp->if_mtu > RL_MTU)
3008 				pci_set_max_read_req(sc->rl_dev, 512);
3009 			else
3010 				pci_set_max_read_req(sc->rl_dev, 4096);
3011 		}
3012 	}
3013 	re_tx_list_init(sc);
3014 
3015 	/*
3016 	 * Enable C+ RX and TX mode, as well as VLAN stripping and
3017 	 * RX checksum offload. We must configure the C+ register
3018 	 * before all others.
3019 	 */
3020 	cfg = RL_CPLUSCMD_PCI_MRW;
3021 	if ((ifp->if_capenable & IFCAP_RXCSUM) != 0)
3022 		cfg |= RL_CPLUSCMD_RXCSUM_ENB;
3023 	if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0)
3024 		cfg |= RL_CPLUSCMD_VLANSTRIP;
3025 	if ((sc->rl_flags & RL_FLAG_MACSTAT) != 0) {
3026 		cfg |= RL_CPLUSCMD_MACSTAT_DIS;
3027 		/* XXX magic. */
3028 		cfg |= 0x0001;
3029 	} else
3030 		cfg |= RL_CPLUSCMD_RXENB | RL_CPLUSCMD_TXENB;
3031 	CSR_WRITE_2(sc, RL_CPLUS_CMD, cfg);
3032 	if (sc->rl_hwrev->rl_rev == RL_HWREV_8169_8110SC ||
3033 	    sc->rl_hwrev->rl_rev == RL_HWREV_8169_8110SCE) {
3034 		reg = 0x000fff00;
3035 		if ((CSR_READ_1(sc, RL_CFG2) & RL_CFG2_PCI66MHZ) != 0)
3036 			reg |= 0x000000ff;
3037 		if (sc->rl_hwrev->rl_rev == RL_HWREV_8169_8110SCE)
3038 			reg |= 0x00f00000;
3039 		CSR_WRITE_4(sc, 0x7c, reg);
3040 		/* Disable interrupt mitigation. */
3041 		CSR_WRITE_2(sc, 0xe2, 0);
3042 	}
3043 	/*
3044 	 * Disable TSO if interface MTU size is greater than MSS
3045 	 * allowed in controller.
3046 	 */
3047 	if (ifp->if_mtu > RL_TSO_MTU && (ifp->if_capenable & IFCAP_TSO4) != 0) {
3048 		ifp->if_capenable &= ~IFCAP_TSO4;
3049 		ifp->if_hwassist &= ~CSUM_TSO;
3050 	}
3051 
3052 	/*
3053 	 * Init our MAC address.  Even though the chipset
3054 	 * documentation doesn't mention it, we need to enter "Config
3055 	 * register write enable" mode to modify the ID registers.
3056 	 */
3057 	/* Copy MAC address on stack to align. */
3058 	bcopy(IF_LLADDR(ifp), eaddr.eaddr, ETHER_ADDR_LEN);
3059 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG);
3060 	CSR_WRITE_4(sc, RL_IDR0,
3061 	    htole32(*(u_int32_t *)(&eaddr.eaddr[0])));
3062 	CSR_WRITE_4(sc, RL_IDR4,
3063 	    htole32(*(u_int32_t *)(&eaddr.eaddr[4])));
3064 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3065 
3066 	/*
3067 	 * Load the addresses of the RX and TX lists into the chip.
3068 	 */
3069 
3070 	CSR_WRITE_4(sc, RL_RXLIST_ADDR_HI,
3071 	    RL_ADDR_HI(sc->rl_ldata.rl_rx_list_addr));
3072 	CSR_WRITE_4(sc, RL_RXLIST_ADDR_LO,
3073 	    RL_ADDR_LO(sc->rl_ldata.rl_rx_list_addr));
3074 
3075 	CSR_WRITE_4(sc, RL_TXLIST_ADDR_HI,
3076 	    RL_ADDR_HI(sc->rl_ldata.rl_tx_list_addr));
3077 	CSR_WRITE_4(sc, RL_TXLIST_ADDR_LO,
3078 	    RL_ADDR_LO(sc->rl_ldata.rl_tx_list_addr));
3079 
3080 	/*
3081 	 * Enable transmit and receive.
3082 	 */
3083 	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
3084 
3085 	/*
3086 	 * Set the initial TX configuration.
3087 	 */
3088 	if (sc->rl_testmode) {
3089 		if (sc->rl_type == RL_8169)
3090 			CSR_WRITE_4(sc, RL_TXCFG,
3091 			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON);
3092 		else
3093 			CSR_WRITE_4(sc, RL_TXCFG,
3094 			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON_CPLUS);
3095 	} else
3096 		CSR_WRITE_4(sc, RL_TXCFG, RL_TXCFG_CONFIG);
3097 
3098 	CSR_WRITE_1(sc, RL_EARLY_TX_THRESH, 16);
3099 
3100 	/*
3101 	 * Set the initial RX configuration.
3102 	 */
3103 	re_set_rxmode(sc);
3104 
3105 	/* Configure interrupt moderation. */
3106 	if (sc->rl_type == RL_8169) {
3107 		/* Magic from vendor. */
3108 		CSR_WRITE_2(sc, RL_INTRMOD, 0x5100);
3109 	}
3110 
3111 #ifdef DEVICE_POLLING
3112 	/*
3113 	 * Disable interrupts if we are polling.
3114 	 */
3115 	if (ifp->if_capenable & IFCAP_POLLING)
3116 		CSR_WRITE_2(sc, RL_IMR, 0);
3117 	else	/* otherwise ... */
3118 #endif
3119 
3120 	/*
3121 	 * Enable interrupts.
3122 	 */
3123 	if (sc->rl_testmode)
3124 		CSR_WRITE_2(sc, RL_IMR, 0);
3125 	else
3126 		CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
3127 	CSR_WRITE_2(sc, RL_ISR, RL_INTRS_CPLUS);
3128 
3129 	/* Set initial TX threshold */
3130 	sc->rl_txthresh = RL_TX_THRESH_INIT;
3131 
3132 	/* Start RX/TX process. */
3133 	CSR_WRITE_4(sc, RL_MISSEDPKT, 0);
3134 #ifdef notdef
3135 	/* Enable receiver and transmitter. */
3136 	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
3137 #endif
3138 
3139 	/*
3140 	 * Initialize the timer interrupt register so that
3141 	 * a timer interrupt will be generated once the timer
3142 	 * reaches a certain number of ticks. The timer is
3143 	 * reloaded on each transmit.
3144 	 */
3145 #ifdef RE_TX_MODERATION
3146 	/*
3147 	 * Use timer interrupt register to moderate TX interrupt
3148 	 * moderation, which dramatically improves TX frame rate.
3149 	 */
3150 	if (sc->rl_type == RL_8169)
3151 		CSR_WRITE_4(sc, RL_TIMERINT_8169, 0x800);
3152 	else
3153 		CSR_WRITE_4(sc, RL_TIMERINT, 0x400);
3154 #else
3155 	/*
3156 	 * Use timer interrupt register to moderate RX interrupt
3157 	 * moderation.
3158 	 */
3159 	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) != 0 &&
3160 	    intr_filter == 0) {
3161 		if (sc->rl_type == RL_8169)
3162 			CSR_WRITE_4(sc, RL_TIMERINT_8169,
3163 			    RL_USECS(sc->rl_int_rx_mod));
3164 	} else {
3165 		if (sc->rl_type == RL_8169)
3166 			CSR_WRITE_4(sc, RL_TIMERINT_8169, RL_USECS(0));
3167 	}
3168 #endif
3169 
3170 	/*
3171 	 * For 8169 gigE NICs, set the max allowed RX packet
3172 	 * size so we can receive jumbo frames.
3173 	 */
3174 	if (sc->rl_type == RL_8169) {
3175 		if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
3176 			/*
3177 			 * For controllers that use new jumbo frame scheme,
3178 			 * set maximum size of jumbo frame depedning on
3179 			 * controller revisions.
3180 			 */
3181 			if (ifp->if_mtu > RL_MTU)
3182 				CSR_WRITE_2(sc, RL_MAXRXPKTLEN,
3183 				    sc->rl_hwrev->rl_max_mtu +
3184 				    ETHER_VLAN_ENCAP_LEN + ETHER_HDR_LEN +
3185 				    ETHER_CRC_LEN);
3186 			else
3187 				CSR_WRITE_2(sc, RL_MAXRXPKTLEN,
3188 				    RE_RX_DESC_BUFLEN);
3189 		} else if ((sc->rl_flags & RL_FLAG_PCIE) != 0 &&
3190 		    sc->rl_hwrev->rl_max_mtu == RL_MTU) {
3191 			/* RTL810x has no jumbo frame support. */
3192 			CSR_WRITE_2(sc, RL_MAXRXPKTLEN, RE_RX_DESC_BUFLEN);
3193 		} else
3194 			CSR_WRITE_2(sc, RL_MAXRXPKTLEN, 16383);
3195 	}
3196 
3197 	if (sc->rl_testmode)
3198 		return;
3199 
3200 	mii_mediachg(mii);
3201 
3202 	CSR_WRITE_1(sc, RL_CFG1, CSR_READ_1(sc, RL_CFG1) | RL_CFG1_DRVLOAD);
3203 
3204 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
3205 	ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
3206 
3207 	sc->rl_flags &= ~RL_FLAG_LINK;
3208 	sc->rl_watchdog_timer = 0;
3209 	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
3210 }
3211 
3212 /*
3213  * Set media options.
3214  */
3215 static int
3216 re_ifmedia_upd(struct ifnet *ifp)
3217 {
3218 	struct rl_softc		*sc;
3219 	struct mii_data		*mii;
3220 	int			error;
3221 
3222 	sc = ifp->if_softc;
3223 	mii = device_get_softc(sc->rl_miibus);
3224 	RL_LOCK(sc);
3225 	error = mii_mediachg(mii);
3226 	RL_UNLOCK(sc);
3227 
3228 	return (error);
3229 }
3230 
3231 /*
3232  * Report current media status.
3233  */
3234 static void
3235 re_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
3236 {
3237 	struct rl_softc		*sc;
3238 	struct mii_data		*mii;
3239 
3240 	sc = ifp->if_softc;
3241 	mii = device_get_softc(sc->rl_miibus);
3242 
3243 	RL_LOCK(sc);
3244 	mii_pollstat(mii);
3245 	ifmr->ifm_active = mii->mii_media_active;
3246 	ifmr->ifm_status = mii->mii_media_status;
3247 	RL_UNLOCK(sc);
3248 }
3249 
3250 static int
3251 re_ioctl(struct ifnet *ifp, u_long command, caddr_t data)
3252 {
3253 	struct rl_softc		*sc = ifp->if_softc;
3254 	struct ifreq		*ifr = (struct ifreq *) data;
3255 	struct mii_data		*mii;
3256 	uint32_t		rev;
3257 	int			error = 0;
3258 
3259 	switch (command) {
3260 	case SIOCSIFMTU:
3261 		if (ifr->ifr_mtu < ETHERMIN ||
3262 		    ifr->ifr_mtu > sc->rl_hwrev->rl_max_mtu) {
3263 			error = EINVAL;
3264 			break;
3265 		}
3266 		RL_LOCK(sc);
3267 		if (ifp->if_mtu != ifr->ifr_mtu) {
3268 			ifp->if_mtu = ifr->ifr_mtu;
3269 			if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0 &&
3270 			    (ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
3271 				ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3272 				re_init_locked(sc);
3273 			}
3274 			if (ifp->if_mtu > RL_TSO_MTU &&
3275 			    (ifp->if_capenable & IFCAP_TSO4) != 0) {
3276 				ifp->if_capenable &= ~(IFCAP_TSO4 |
3277 				    IFCAP_VLAN_HWTSO);
3278 				ifp->if_hwassist &= ~CSUM_TSO;
3279 			}
3280 			VLAN_CAPABILITIES(ifp);
3281 		}
3282 		RL_UNLOCK(sc);
3283 		break;
3284 	case SIOCSIFFLAGS:
3285 		RL_LOCK(sc);
3286 		if ((ifp->if_flags & IFF_UP) != 0) {
3287 			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
3288 				if (((ifp->if_flags ^ sc->rl_if_flags)
3289 				    & (IFF_PROMISC | IFF_ALLMULTI)) != 0)
3290 					re_set_rxmode(sc);
3291 			} else
3292 				re_init_locked(sc);
3293 		} else {
3294 			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
3295 				re_stop(sc);
3296 		}
3297 		sc->rl_if_flags = ifp->if_flags;
3298 		RL_UNLOCK(sc);
3299 		break;
3300 	case SIOCADDMULTI:
3301 	case SIOCDELMULTI:
3302 		RL_LOCK(sc);
3303 		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
3304 			re_set_rxmode(sc);
3305 		RL_UNLOCK(sc);
3306 		break;
3307 	case SIOCGIFMEDIA:
3308 	case SIOCSIFMEDIA:
3309 		mii = device_get_softc(sc->rl_miibus);
3310 		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, command);
3311 		break;
3312 	case SIOCSIFCAP:
3313 	    {
3314 		int mask, reinit;
3315 
3316 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
3317 		reinit = 0;
3318 #ifdef DEVICE_POLLING
3319 		if (mask & IFCAP_POLLING) {
3320 			if (ifr->ifr_reqcap & IFCAP_POLLING) {
3321 				error = ether_poll_register(re_poll, ifp);
3322 				if (error)
3323 					return (error);
3324 				RL_LOCK(sc);
3325 				/* Disable interrupts */
3326 				CSR_WRITE_2(sc, RL_IMR, 0x0000);
3327 				ifp->if_capenable |= IFCAP_POLLING;
3328 				RL_UNLOCK(sc);
3329 			} else {
3330 				error = ether_poll_deregister(ifp);
3331 				/* Enable interrupts. */
3332 				RL_LOCK(sc);
3333 				CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
3334 				ifp->if_capenable &= ~IFCAP_POLLING;
3335 				RL_UNLOCK(sc);
3336 			}
3337 		}
3338 #endif /* DEVICE_POLLING */
3339 		RL_LOCK(sc);
3340 		if ((mask & IFCAP_TXCSUM) != 0 &&
3341 		    (ifp->if_capabilities & IFCAP_TXCSUM) != 0) {
3342 			ifp->if_capenable ^= IFCAP_TXCSUM;
3343 			if ((ifp->if_capenable & IFCAP_TXCSUM) != 0) {
3344 				rev = sc->rl_hwrev->rl_rev;
3345 				if (rev == RL_HWREV_8168C ||
3346 				    rev == RL_HWREV_8168C_SPIN2)
3347 					ifp->if_hwassist |= CSUM_TCP | CSUM_UDP;
3348 				else
3349 					ifp->if_hwassist |= RE_CSUM_FEATURES;
3350 			} else
3351 				ifp->if_hwassist &= ~RE_CSUM_FEATURES;
3352 			reinit = 1;
3353 		}
3354 		if ((mask & IFCAP_RXCSUM) != 0 &&
3355 		    (ifp->if_capabilities & IFCAP_RXCSUM) != 0) {
3356 			ifp->if_capenable ^= IFCAP_RXCSUM;
3357 			reinit = 1;
3358 		}
3359 		if ((mask & IFCAP_TSO4) != 0 &&
3360 		    (ifp->if_capabilities & IFCAP_TSO) != 0) {
3361 			ifp->if_capenable ^= IFCAP_TSO4;
3362 			if ((IFCAP_TSO4 & ifp->if_capenable) != 0)
3363 				ifp->if_hwassist |= CSUM_TSO;
3364 			else
3365 				ifp->if_hwassist &= ~CSUM_TSO;
3366 			if (ifp->if_mtu > RL_TSO_MTU &&
3367 			    (ifp->if_capenable & IFCAP_TSO4) != 0) {
3368 				ifp->if_capenable &= ~IFCAP_TSO4;
3369 				ifp->if_hwassist &= ~CSUM_TSO;
3370 			}
3371 		}
3372 		if ((mask & IFCAP_VLAN_HWTSO) != 0 &&
3373 		    (ifp->if_capabilities & IFCAP_VLAN_HWTSO) != 0)
3374 			ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
3375 		if ((mask & IFCAP_VLAN_HWTAGGING) != 0 &&
3376 		    (ifp->if_capabilities & IFCAP_VLAN_HWTAGGING) != 0) {
3377 			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
3378 			/* TSO over VLAN requires VLAN hardware tagging. */
3379 			if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) == 0)
3380 				ifp->if_capenable &= ~IFCAP_VLAN_HWTSO;
3381 			reinit = 1;
3382 		}
3383 		if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0 &&
3384 		    (mask & (IFCAP_HWCSUM | IFCAP_TSO4 |
3385 		    IFCAP_VLAN_HWTSO)) != 0)
3386 				reinit = 1;
3387 		if ((mask & IFCAP_WOL) != 0 &&
3388 		    (ifp->if_capabilities & IFCAP_WOL) != 0) {
3389 			if ((mask & IFCAP_WOL_UCAST) != 0)
3390 				ifp->if_capenable ^= IFCAP_WOL_UCAST;
3391 			if ((mask & IFCAP_WOL_MCAST) != 0)
3392 				ifp->if_capenable ^= IFCAP_WOL_MCAST;
3393 			if ((mask & IFCAP_WOL_MAGIC) != 0)
3394 				ifp->if_capenable ^= IFCAP_WOL_MAGIC;
3395 		}
3396 		if (reinit && ifp->if_drv_flags & IFF_DRV_RUNNING) {
3397 			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3398 			re_init_locked(sc);
3399 		}
3400 		RL_UNLOCK(sc);
3401 		VLAN_CAPABILITIES(ifp);
3402 	    }
3403 		break;
3404 	default:
3405 		error = ether_ioctl(ifp, command, data);
3406 		break;
3407 	}
3408 
3409 	return (error);
3410 }
3411 
3412 static void
3413 re_watchdog(struct rl_softc *sc)
3414 {
3415 	struct ifnet		*ifp;
3416 
3417 	RL_LOCK_ASSERT(sc);
3418 
3419 	if (sc->rl_watchdog_timer == 0 || --sc->rl_watchdog_timer != 0)
3420 		return;
3421 
3422 	ifp = sc->rl_ifp;
3423 	re_txeof(sc);
3424 	if (sc->rl_ldata.rl_tx_free == sc->rl_ldata.rl_tx_desc_cnt) {
3425 		if_printf(ifp, "watchdog timeout (missed Tx interrupts) "
3426 		    "-- recovering\n");
3427 		if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
3428 			re_start_locked(ifp);
3429 		return;
3430 	}
3431 
3432 	if_printf(ifp, "watchdog timeout\n");
3433 	ifp->if_oerrors++;
3434 
3435 	re_rxeof(sc, NULL);
3436 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3437 	re_init_locked(sc);
3438 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
3439 		re_start_locked(ifp);
3440 }
3441 
3442 /*
3443  * Stop the adapter and free any mbufs allocated to the
3444  * RX and TX lists.
3445  */
3446 static void
3447 re_stop(struct rl_softc *sc)
3448 {
3449 	int			i;
3450 	struct ifnet		*ifp;
3451 	struct rl_txdesc	*txd;
3452 	struct rl_rxdesc	*rxd;
3453 
3454 	RL_LOCK_ASSERT(sc);
3455 
3456 	ifp = sc->rl_ifp;
3457 
3458 	sc->rl_watchdog_timer = 0;
3459 	callout_stop(&sc->rl_stat_callout);
3460 	ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
3461 
3462 	if ((sc->rl_flags & RL_FLAG_CMDSTOP) != 0)
3463 		CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_STOPREQ | RL_CMD_TX_ENB |
3464 		    RL_CMD_RX_ENB);
3465 	else
3466 		CSR_WRITE_1(sc, RL_COMMAND, 0x00);
3467 	DELAY(1000);
3468 	CSR_WRITE_2(sc, RL_IMR, 0x0000);
3469 	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
3470 
3471 	if (sc->rl_head != NULL) {
3472 		m_freem(sc->rl_head);
3473 		sc->rl_head = sc->rl_tail = NULL;
3474 	}
3475 
3476 	/* Free the TX list buffers. */
3477 
3478 	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
3479 		txd = &sc->rl_ldata.rl_tx_desc[i];
3480 		if (txd->tx_m != NULL) {
3481 			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
3482 			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
3483 			bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag,
3484 			    txd->tx_dmamap);
3485 			m_freem(txd->tx_m);
3486 			txd->tx_m = NULL;
3487 		}
3488 	}
3489 
3490 	/* Free the RX list buffers. */
3491 
3492 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
3493 		rxd = &sc->rl_ldata.rl_rx_desc[i];
3494 		if (rxd->rx_m != NULL) {
3495 			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
3496 			    rxd->rx_dmamap, BUS_DMASYNC_POSTREAD);
3497 			bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag,
3498 			    rxd->rx_dmamap);
3499 			m_freem(rxd->rx_m);
3500 			rxd->rx_m = NULL;
3501 		}
3502 	}
3503 }
3504 
3505 /*
3506  * Device suspend routine.  Stop the interface and save some PCI
3507  * settings in case the BIOS doesn't restore them properly on
3508  * resume.
3509  */
3510 static int
3511 re_suspend(device_t dev)
3512 {
3513 	struct rl_softc		*sc;
3514 
3515 	sc = device_get_softc(dev);
3516 
3517 	RL_LOCK(sc);
3518 	re_stop(sc);
3519 	re_setwol(sc);
3520 	sc->suspended = 1;
3521 	RL_UNLOCK(sc);
3522 
3523 	return (0);
3524 }
3525 
3526 /*
3527  * Device resume routine.  Restore some PCI settings in case the BIOS
3528  * doesn't, re-enable busmastering, and restart the interface if
3529  * appropriate.
3530  */
3531 static int
3532 re_resume(device_t dev)
3533 {
3534 	struct rl_softc		*sc;
3535 	struct ifnet		*ifp;
3536 
3537 	sc = device_get_softc(dev);
3538 
3539 	RL_LOCK(sc);
3540 
3541 	ifp = sc->rl_ifp;
3542 	/* Take controller out of sleep mode. */
3543 	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
3544 		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
3545 			CSR_WRITE_1(sc, RL_GPIO,
3546 			    CSR_READ_1(sc, RL_GPIO) | 0x01);
3547 	}
3548 
3549 	/*
3550 	 * Clear WOL matching such that normal Rx filtering
3551 	 * wouldn't interfere with WOL patterns.
3552 	 */
3553 	re_clrwol(sc);
3554 
3555 	/* reinitialize interface if necessary */
3556 	if (ifp->if_flags & IFF_UP)
3557 		re_init_locked(sc);
3558 
3559 	sc->suspended = 0;
3560 	RL_UNLOCK(sc);
3561 
3562 	return (0);
3563 }
3564 
3565 /*
3566  * Stop all chip I/O so that the kernel's probe routines don't
3567  * get confused by errant DMAs when rebooting.
3568  */
3569 static int
3570 re_shutdown(device_t dev)
3571 {
3572 	struct rl_softc		*sc;
3573 
3574 	sc = device_get_softc(dev);
3575 
3576 	RL_LOCK(sc);
3577 	re_stop(sc);
3578 	/*
3579 	 * Mark interface as down since otherwise we will panic if
3580 	 * interrupt comes in later on, which can happen in some
3581 	 * cases.
3582 	 */
3583 	sc->rl_ifp->if_flags &= ~IFF_UP;
3584 	re_setwol(sc);
3585 	RL_UNLOCK(sc);
3586 
3587 	return (0);
3588 }
3589 
3590 static void
3591 re_setwol(struct rl_softc *sc)
3592 {
3593 	struct ifnet		*ifp;
3594 	int			pmc;
3595 	uint16_t		pmstat;
3596 	uint8_t			v;
3597 
3598 	RL_LOCK_ASSERT(sc);
3599 
3600 	if (pci_find_cap(sc->rl_dev, PCIY_PMG, &pmc) != 0)
3601 		return;
3602 
3603 	ifp = sc->rl_ifp;
3604 	/* Put controller into sleep mode. */
3605 	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
3606 		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
3607 			CSR_WRITE_1(sc, RL_GPIO,
3608 			    CSR_READ_1(sc, RL_GPIO) & ~0x01);
3609 	}
3610 	if ((ifp->if_capenable & IFCAP_WOL) != 0 &&
3611 	    (sc->rl_flags & RL_FLAG_WOLRXENB) != 0)
3612 		CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RX_ENB);
3613 	/* Enable config register write. */
3614 	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
3615 
3616 	/* Enable PME. */
3617 	v = CSR_READ_1(sc, RL_CFG1);
3618 	v &= ~RL_CFG1_PME;
3619 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3620 		v |= RL_CFG1_PME;
3621 	CSR_WRITE_1(sc, RL_CFG1, v);
3622 
3623 	v = CSR_READ_1(sc, RL_CFG3);
3624 	v &= ~(RL_CFG3_WOL_LINK | RL_CFG3_WOL_MAGIC);
3625 	if ((ifp->if_capenable & IFCAP_WOL_MAGIC) != 0)
3626 		v |= RL_CFG3_WOL_MAGIC;
3627 	CSR_WRITE_1(sc, RL_CFG3, v);
3628 
3629 	/* Config register write done. */
3630 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3631 
3632 	v = CSR_READ_1(sc, RL_CFG5);
3633 	v &= ~(RL_CFG5_WOL_BCAST | RL_CFG5_WOL_MCAST | RL_CFG5_WOL_UCAST);
3634 	v &= ~RL_CFG5_WOL_LANWAKE;
3635 	if ((ifp->if_capenable & IFCAP_WOL_UCAST) != 0)
3636 		v |= RL_CFG5_WOL_UCAST;
3637 	if ((ifp->if_capenable & IFCAP_WOL_MCAST) != 0)
3638 		v |= RL_CFG5_WOL_MCAST | RL_CFG5_WOL_BCAST;
3639 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3640 		v |= RL_CFG5_WOL_LANWAKE;
3641 	CSR_WRITE_1(sc, RL_CFG5, v);
3642 
3643 	if ((ifp->if_capenable & IFCAP_WOL) != 0 &&
3644 	    (sc->rl_flags & RL_FLAG_PHYWAKE_PM) != 0)
3645 		CSR_WRITE_1(sc, RL_PMCH, CSR_READ_1(sc, RL_PMCH) & ~0x80);
3646 	/*
3647 	 * It seems that hardware resets its link speed to 100Mbps in
3648 	 * power down mode so switching to 100Mbps in driver is not
3649 	 * needed.
3650 	 */
3651 
3652 	/* Request PME if WOL is requested. */
3653 	pmstat = pci_read_config(sc->rl_dev, pmc + PCIR_POWER_STATUS, 2);
3654 	pmstat &= ~(PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE);
3655 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3656 		pmstat |= PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE;
3657 	pci_write_config(sc->rl_dev, pmc + PCIR_POWER_STATUS, pmstat, 2);
3658 }
3659 
3660 static void
3661 re_clrwol(struct rl_softc *sc)
3662 {
3663 	int			pmc;
3664 	uint8_t			v;
3665 
3666 	RL_LOCK_ASSERT(sc);
3667 
3668 	if (pci_find_cap(sc->rl_dev, PCIY_PMG, &pmc) != 0)
3669 		return;
3670 
3671 	/* Enable config register write. */
3672 	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
3673 
3674 	v = CSR_READ_1(sc, RL_CFG3);
3675 	v &= ~(RL_CFG3_WOL_LINK | RL_CFG3_WOL_MAGIC);
3676 	CSR_WRITE_1(sc, RL_CFG3, v);
3677 
3678 	/* Config register write done. */
3679 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3680 
3681 	v = CSR_READ_1(sc, RL_CFG5);
3682 	v &= ~(RL_CFG5_WOL_BCAST | RL_CFG5_WOL_MCAST | RL_CFG5_WOL_UCAST);
3683 	v &= ~RL_CFG5_WOL_LANWAKE;
3684 	CSR_WRITE_1(sc, RL_CFG5, v);
3685 }
3686 
3687 static void
3688 re_add_sysctls(struct rl_softc *sc)
3689 {
3690 	struct sysctl_ctx_list	*ctx;
3691 	struct sysctl_oid_list	*children;
3692 	int			error;
3693 
3694 	ctx = device_get_sysctl_ctx(sc->rl_dev);
3695 	children = SYSCTL_CHILDREN(device_get_sysctl_tree(sc->rl_dev));
3696 
3697 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "stats",
3698 	    CTLTYPE_INT | CTLFLAG_RW, sc, 0, re_sysctl_stats, "I",
3699 	    "Statistics Information");
3700 	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) == 0)
3701 		return;
3702 
3703 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "int_rx_mod",
3704 	    CTLTYPE_INT | CTLFLAG_RW, &sc->rl_int_rx_mod, 0,
3705 	    sysctl_hw_re_int_mod, "I", "re RX interrupt moderation");
3706 	/* Pull in device tunables. */
3707 	sc->rl_int_rx_mod = RL_TIMER_DEFAULT;
3708 	error = resource_int_value(device_get_name(sc->rl_dev),
3709 	    device_get_unit(sc->rl_dev), "int_rx_mod", &sc->rl_int_rx_mod);
3710 	if (error == 0) {
3711 		if (sc->rl_int_rx_mod < RL_TIMER_MIN ||
3712 		    sc->rl_int_rx_mod > RL_TIMER_MAX) {
3713 			device_printf(sc->rl_dev, "int_rx_mod value out of "
3714 			    "range; using default: %d\n",
3715 			    RL_TIMER_DEFAULT);
3716 			sc->rl_int_rx_mod = RL_TIMER_DEFAULT;
3717 		}
3718 	}
3719 
3720 }
3721 
3722 static int
3723 re_sysctl_stats(SYSCTL_HANDLER_ARGS)
3724 {
3725 	struct rl_softc		*sc;
3726 	struct rl_stats		*stats;
3727 	int			error, i, result;
3728 
3729 	result = -1;
3730 	error = sysctl_handle_int(oidp, &result, 0, req);
3731 	if (error || req->newptr == NULL)
3732 		return (error);
3733 
3734 	if (result == 1) {
3735 		sc = (struct rl_softc *)arg1;
3736 		RL_LOCK(sc);
3737 		if ((sc->rl_ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
3738 			RL_UNLOCK(sc);
3739 			goto done;
3740 		}
3741 		bus_dmamap_sync(sc->rl_ldata.rl_stag,
3742 		    sc->rl_ldata.rl_smap, BUS_DMASYNC_PREREAD);
3743 		CSR_WRITE_4(sc, RL_DUMPSTATS_HI,
3744 		    RL_ADDR_HI(sc->rl_ldata.rl_stats_addr));
3745 		CSR_WRITE_4(sc, RL_DUMPSTATS_LO,
3746 		    RL_ADDR_LO(sc->rl_ldata.rl_stats_addr));
3747 		CSR_WRITE_4(sc, RL_DUMPSTATS_LO,
3748 		    RL_ADDR_LO(sc->rl_ldata.rl_stats_addr |
3749 		    RL_DUMPSTATS_START));
3750 		for (i = RL_TIMEOUT; i > 0; i--) {
3751 			if ((CSR_READ_4(sc, RL_DUMPSTATS_LO) &
3752 			    RL_DUMPSTATS_START) == 0)
3753 				break;
3754 			DELAY(1000);
3755 		}
3756 		bus_dmamap_sync(sc->rl_ldata.rl_stag,
3757 		    sc->rl_ldata.rl_smap, BUS_DMASYNC_POSTREAD);
3758 		RL_UNLOCK(sc);
3759 		if (i == 0) {
3760 			device_printf(sc->rl_dev,
3761 			    "DUMP statistics request timedout\n");
3762 			return (ETIMEDOUT);
3763 		}
3764 done:
3765 		stats = sc->rl_ldata.rl_stats;
3766 		printf("%s statistics:\n", device_get_nameunit(sc->rl_dev));
3767 		printf("Tx frames : %ju\n",
3768 		    (uintmax_t)le64toh(stats->rl_tx_pkts));
3769 		printf("Rx frames : %ju\n",
3770 		    (uintmax_t)le64toh(stats->rl_rx_pkts));
3771 		printf("Tx errors : %ju\n",
3772 		    (uintmax_t)le64toh(stats->rl_tx_errs));
3773 		printf("Rx errors : %u\n",
3774 		    le32toh(stats->rl_rx_errs));
3775 		printf("Rx missed frames : %u\n",
3776 		    (uint32_t)le16toh(stats->rl_missed_pkts));
3777 		printf("Rx frame alignment errs : %u\n",
3778 		    (uint32_t)le16toh(stats->rl_rx_framealign_errs));
3779 		printf("Tx single collisions : %u\n",
3780 		    le32toh(stats->rl_tx_onecoll));
3781 		printf("Tx multiple collisions : %u\n",
3782 		    le32toh(stats->rl_tx_multicolls));
3783 		printf("Rx unicast frames : %ju\n",
3784 		    (uintmax_t)le64toh(stats->rl_rx_ucasts));
3785 		printf("Rx broadcast frames : %ju\n",
3786 		    (uintmax_t)le64toh(stats->rl_rx_bcasts));
3787 		printf("Rx multicast frames : %u\n",
3788 		    le32toh(stats->rl_rx_mcasts));
3789 		printf("Tx aborts : %u\n",
3790 		    (uint32_t)le16toh(stats->rl_tx_aborts));
3791 		printf("Tx underruns : %u\n",
3792 		    (uint32_t)le16toh(stats->rl_rx_underruns));
3793 	}
3794 
3795 	return (error);
3796 }
3797 
3798 static int
3799 sysctl_int_range(SYSCTL_HANDLER_ARGS, int low, int high)
3800 {
3801 	int error, value;
3802 
3803 	if (arg1 == NULL)
3804 		return (EINVAL);
3805 	value = *(int *)arg1;
3806 	error = sysctl_handle_int(oidp, &value, 0, req);
3807 	if (error || req->newptr == NULL)
3808 		return (error);
3809 	if (value < low || value > high)
3810 		return (EINVAL);
3811 	*(int *)arg1 = value;
3812 
3813 	return (0);
3814 }
3815 
3816 static int
3817 sysctl_hw_re_int_mod(SYSCTL_HANDLER_ARGS)
3818 {
3819 
3820 	return (sysctl_int_range(oidp, arg1, arg2, req, RL_TIMER_MIN,
3821 	    RL_TIMER_MAX));
3822 }
3823