xref: /freebsd/sys/dev/re/if_re.c (revision 09d325677d53a12c79a43664ff29871e92247629)
1 /*-
2  * Copyright (c) 1997, 1998-2003
3  *	Bill Paul <wpaul@windriver.com>.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. All advertising materials mentioning features or use of this software
14  *    must display the following acknowledgement:
15  *	This product includes software developed by Bill Paul.
16  * 4. Neither the name of the author nor the names of any co-contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY Bill Paul AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL Bill Paul OR THE VOICES IN HIS HEAD
24  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
30  * THE POSSIBILITY OF SUCH DAMAGE.
31  */
32 
33 #include <sys/cdefs.h>
34 __FBSDID("$FreeBSD$");
35 
36 /*
37  * RealTek 8139C+/8169/8169S/8110S/8168/8111/8101E PCI NIC driver
38  *
39  * Written by Bill Paul <wpaul@windriver.com>
40  * Senior Networking Software Engineer
41  * Wind River Systems
42  */
43 
44 /*
45  * This driver is designed to support RealTek's next generation of
46  * 10/100 and 10/100/1000 PCI ethernet controllers. There are currently
47  * seven devices in this family: the RTL8139C+, the RTL8169, the RTL8169S,
48  * RTL8110S, the RTL8168, the RTL8111 and the RTL8101E.
49  *
50  * The 8139C+ is a 10/100 ethernet chip. It is backwards compatible
51  * with the older 8139 family, however it also supports a special
52  * C+ mode of operation that provides several new performance enhancing
53  * features. These include:
54  *
55  *	o Descriptor based DMA mechanism. Each descriptor represents
56  *	  a single packet fragment. Data buffers may be aligned on
57  *	  any byte boundary.
58  *
59  *	o 64-bit DMA
60  *
61  *	o TCP/IP checksum offload for both RX and TX
62  *
63  *	o High and normal priority transmit DMA rings
64  *
65  *	o VLAN tag insertion and extraction
66  *
67  *	o TCP large send (segmentation offload)
68  *
69  * Like the 8139, the 8139C+ also has a built-in 10/100 PHY. The C+
70  * programming API is fairly straightforward. The RX filtering, EEPROM
71  * access and PHY access is the same as it is on the older 8139 series
72  * chips.
73  *
74  * The 8169 is a 64-bit 10/100/1000 gigabit ethernet MAC. It has almost the
75  * same programming API and feature set as the 8139C+ with the following
76  * differences and additions:
77  *
78  *	o 1000Mbps mode
79  *
80  *	o Jumbo frames
81  *
82  *	o GMII and TBI ports/registers for interfacing with copper
83  *	  or fiber PHYs
84  *
85  *	o RX and TX DMA rings can have up to 1024 descriptors
86  *	  (the 8139C+ allows a maximum of 64)
87  *
88  *	o Slight differences in register layout from the 8139C+
89  *
90  * The TX start and timer interrupt registers are at different locations
91  * on the 8169 than they are on the 8139C+. Also, the status word in the
92  * RX descriptor has a slightly different bit layout. The 8169 does not
93  * have a built-in PHY. Most reference boards use a Marvell 88E1000 'Alaska'
94  * copper gigE PHY.
95  *
96  * The 8169S/8110S 10/100/1000 devices have built-in copper gigE PHYs
97  * (the 'S' stands for 'single-chip'). These devices have the same
98  * programming API as the older 8169, but also have some vendor-specific
99  * registers for the on-board PHY. The 8110S is a LAN-on-motherboard
100  * part designed to be pin-compatible with the RealTek 8100 10/100 chip.
101  *
102  * This driver takes advantage of the RX and TX checksum offload and
103  * VLAN tag insertion/extraction features. It also implements TX
104  * interrupt moderation using the timer interrupt registers, which
105  * significantly reduces TX interrupt load. There is also support
106  * for jumbo frames, however the 8169/8169S/8110S can not transmit
107  * jumbo frames larger than 7440, so the max MTU possible with this
108  * driver is 7422 bytes.
109  */
110 
111 #ifdef HAVE_KERNEL_OPTION_HEADERS
112 #include "opt_device_polling.h"
113 #endif
114 
115 #include <sys/param.h>
116 #include <sys/endian.h>
117 #include <sys/systm.h>
118 #include <sys/sockio.h>
119 #include <sys/mbuf.h>
120 #include <sys/malloc.h>
121 #include <sys/module.h>
122 #include <sys/kernel.h>
123 #include <sys/socket.h>
124 #include <sys/lock.h>
125 #include <sys/mutex.h>
126 #include <sys/sysctl.h>
127 #include <sys/taskqueue.h>
128 
129 #include <net/if.h>
130 #include <net/if_var.h>
131 #include <net/if_arp.h>
132 #include <net/ethernet.h>
133 #include <net/if_dl.h>
134 #include <net/if_media.h>
135 #include <net/if_types.h>
136 #include <net/if_vlan_var.h>
137 
138 #include <net/bpf.h>
139 
140 #include <machine/bus.h>
141 #include <machine/resource.h>
142 #include <sys/bus.h>
143 #include <sys/rman.h>
144 
145 #include <dev/mii/mii.h>
146 #include <dev/mii/miivar.h>
147 
148 #include <dev/pci/pcireg.h>
149 #include <dev/pci/pcivar.h>
150 
151 #include <pci/if_rlreg.h>
152 
153 MODULE_DEPEND(re, pci, 1, 1, 1);
154 MODULE_DEPEND(re, ether, 1, 1, 1);
155 MODULE_DEPEND(re, miibus, 1, 1, 1);
156 
157 /* "device miibus" required.  See GENERIC if you get errors here. */
158 #include "miibus_if.h"
159 
160 /* Tunables. */
161 static int intr_filter = 0;
162 TUNABLE_INT("hw.re.intr_filter", &intr_filter);
163 static int msi_disable = 0;
164 TUNABLE_INT("hw.re.msi_disable", &msi_disable);
165 static int msix_disable = 0;
166 TUNABLE_INT("hw.re.msix_disable", &msix_disable);
167 static int prefer_iomap = 0;
168 TUNABLE_INT("hw.re.prefer_iomap", &prefer_iomap);
169 
170 #define RE_CSUM_FEATURES    (CSUM_IP | CSUM_TCP | CSUM_UDP)
171 
172 /*
173  * Various supported device vendors/types and their names.
174  */
175 static const struct rl_type re_devs[] = {
176 	{ DLINK_VENDORID, DLINK_DEVICEID_528T, 0,
177 	    "D-Link DGE-528(T) Gigabit Ethernet Adapter" },
178 	{ DLINK_VENDORID, DLINK_DEVICEID_530T_REVC, 0,
179 	    "D-Link DGE-530(T) Gigabit Ethernet Adapter" },
180 	{ RT_VENDORID, RT_DEVICEID_8139, 0,
181 	    "RealTek 8139C+ 10/100BaseTX" },
182 	{ RT_VENDORID, RT_DEVICEID_8101E, 0,
183 	    "RealTek 810xE PCIe 10/100baseTX" },
184 	{ RT_VENDORID, RT_DEVICEID_8168, 0,
185 	    "RealTek 8168/8111 B/C/CP/D/DP/E/F/G PCIe Gigabit Ethernet" },
186 	{ RT_VENDORID, RT_DEVICEID_8169, 0,
187 	    "RealTek 8169/8169S/8169SB(L)/8110S/8110SB(L) Gigabit Ethernet" },
188 	{ RT_VENDORID, RT_DEVICEID_8169SC, 0,
189 	    "RealTek 8169SC/8110SC Single-chip Gigabit Ethernet" },
190 	{ COREGA_VENDORID, COREGA_DEVICEID_CGLAPCIGT, 0,
191 	    "Corega CG-LAPCIGT (RTL8169S) Gigabit Ethernet" },
192 	{ LINKSYS_VENDORID, LINKSYS_DEVICEID_EG1032, 0,
193 	    "Linksys EG1032 (RTL8169S) Gigabit Ethernet" },
194 	{ USR_VENDORID, USR_DEVICEID_997902, 0,
195 	    "US Robotics 997902 (RTL8169S) Gigabit Ethernet" }
196 };
197 
198 static const struct rl_hwrev re_hwrevs[] = {
199 	{ RL_HWREV_8139, RL_8139, "", RL_MTU },
200 	{ RL_HWREV_8139A, RL_8139, "A", RL_MTU },
201 	{ RL_HWREV_8139AG, RL_8139, "A-G", RL_MTU },
202 	{ RL_HWREV_8139B, RL_8139, "B", RL_MTU },
203 	{ RL_HWREV_8130, RL_8139, "8130", RL_MTU },
204 	{ RL_HWREV_8139C, RL_8139, "C", RL_MTU },
205 	{ RL_HWREV_8139D, RL_8139, "8139D/8100B/8100C", RL_MTU },
206 	{ RL_HWREV_8139CPLUS, RL_8139CPLUS, "C+", RL_MTU },
207 	{ RL_HWREV_8168B_SPIN1, RL_8169, "8168", RL_JUMBO_MTU },
208 	{ RL_HWREV_8169, RL_8169, "8169", RL_JUMBO_MTU },
209 	{ RL_HWREV_8169S, RL_8169, "8169S", RL_JUMBO_MTU },
210 	{ RL_HWREV_8110S, RL_8169, "8110S", RL_JUMBO_MTU },
211 	{ RL_HWREV_8169_8110SB, RL_8169, "8169SB/8110SB", RL_JUMBO_MTU },
212 	{ RL_HWREV_8169_8110SC, RL_8169, "8169SC/8110SC", RL_JUMBO_MTU },
213 	{ RL_HWREV_8169_8110SBL, RL_8169, "8169SBL/8110SBL", RL_JUMBO_MTU },
214 	{ RL_HWREV_8169_8110SCE, RL_8169, "8169SC/8110SC", RL_JUMBO_MTU },
215 	{ RL_HWREV_8100, RL_8139, "8100", RL_MTU },
216 	{ RL_HWREV_8101, RL_8139, "8101", RL_MTU },
217 	{ RL_HWREV_8100E, RL_8169, "8100E", RL_MTU },
218 	{ RL_HWREV_8101E, RL_8169, "8101E", RL_MTU },
219 	{ RL_HWREV_8102E, RL_8169, "8102E", RL_MTU },
220 	{ RL_HWREV_8102EL, RL_8169, "8102EL", RL_MTU },
221 	{ RL_HWREV_8102EL_SPIN1, RL_8169, "8102EL", RL_MTU },
222 	{ RL_HWREV_8103E, RL_8169, "8103E", RL_MTU },
223 	{ RL_HWREV_8401E, RL_8169, "8401E", RL_MTU },
224 	{ RL_HWREV_8402, RL_8169, "8402", RL_MTU },
225 	{ RL_HWREV_8105E, RL_8169, "8105E", RL_MTU },
226 	{ RL_HWREV_8105E_SPIN1, RL_8169, "8105E", RL_MTU },
227 	{ RL_HWREV_8106E, RL_8169, "8106E", RL_MTU },
228 	{ RL_HWREV_8168B_SPIN2, RL_8169, "8168", RL_JUMBO_MTU },
229 	{ RL_HWREV_8168B_SPIN3, RL_8169, "8168", RL_JUMBO_MTU },
230 	{ RL_HWREV_8168C, RL_8169, "8168C/8111C", RL_JUMBO_MTU_6K },
231 	{ RL_HWREV_8168C_SPIN2, RL_8169, "8168C/8111C", RL_JUMBO_MTU_6K },
232 	{ RL_HWREV_8168CP, RL_8169, "8168CP/8111CP", RL_JUMBO_MTU_6K },
233 	{ RL_HWREV_8168D, RL_8169, "8168D/8111D", RL_JUMBO_MTU_9K },
234 	{ RL_HWREV_8168DP, RL_8169, "8168DP/8111DP", RL_JUMBO_MTU_9K },
235 	{ RL_HWREV_8168E, RL_8169, "8168E/8111E", RL_JUMBO_MTU_9K},
236 	{ RL_HWREV_8168E_VL, RL_8169, "8168E/8111E-VL", RL_JUMBO_MTU_6K},
237 	{ RL_HWREV_8168EP, RL_8169, "8168EP/8111EP", RL_JUMBO_MTU_9K},
238 	{ RL_HWREV_8168F, RL_8169, "8168F/8111F", RL_JUMBO_MTU_9K},
239 	{ RL_HWREV_8168G, RL_8169, "8168G/8111G", RL_JUMBO_MTU_9K},
240 	{ RL_HWREV_8168GU, RL_8169, "8168GU/8111GU", RL_JUMBO_MTU_9K},
241 	{ RL_HWREV_8411, RL_8169, "8411", RL_JUMBO_MTU_9K},
242 	{ RL_HWREV_8411B, RL_8169, "8411B", RL_JUMBO_MTU_9K},
243 	{ 0, 0, NULL, 0 }
244 };
245 
246 static int re_probe		(device_t);
247 static int re_attach		(device_t);
248 static int re_detach		(device_t);
249 
250 static int re_encap		(struct rl_softc *, struct mbuf **);
251 
252 static void re_dma_map_addr	(void *, bus_dma_segment_t *, int, int);
253 static int re_allocmem		(device_t, struct rl_softc *);
254 static __inline void re_discard_rxbuf
255 				(struct rl_softc *, int);
256 static int re_newbuf		(struct rl_softc *, int);
257 static int re_jumbo_newbuf	(struct rl_softc *, int);
258 static int re_rx_list_init	(struct rl_softc *);
259 static int re_jrx_list_init	(struct rl_softc *);
260 static int re_tx_list_init	(struct rl_softc *);
261 #ifdef RE_FIXUP_RX
262 static __inline void re_fixup_rx
263 				(struct mbuf *);
264 #endif
265 static int re_rxeof		(struct rl_softc *, int *);
266 static void re_txeof		(struct rl_softc *);
267 #ifdef DEVICE_POLLING
268 static int re_poll		(struct ifnet *, enum poll_cmd, int);
269 static int re_poll_locked	(struct ifnet *, enum poll_cmd, int);
270 #endif
271 static int re_intr		(void *);
272 static void re_intr_msi		(void *);
273 static void re_tick		(void *);
274 static void re_int_task		(void *, int);
275 static void re_start		(struct ifnet *);
276 static void re_start_locked	(struct ifnet *);
277 static int re_ioctl		(struct ifnet *, u_long, caddr_t);
278 static void re_init		(void *);
279 static void re_init_locked	(struct rl_softc *);
280 static void re_stop		(struct rl_softc *);
281 static void re_watchdog		(struct rl_softc *);
282 static int re_suspend		(device_t);
283 static int re_resume		(device_t);
284 static int re_shutdown		(device_t);
285 static int re_ifmedia_upd	(struct ifnet *);
286 static void re_ifmedia_sts	(struct ifnet *, struct ifmediareq *);
287 
288 static void re_eeprom_putbyte	(struct rl_softc *, int);
289 static void re_eeprom_getword	(struct rl_softc *, int, u_int16_t *);
290 static void re_read_eeprom	(struct rl_softc *, caddr_t, int, int);
291 static int re_gmii_readreg	(device_t, int, int);
292 static int re_gmii_writereg	(device_t, int, int, int);
293 
294 static int re_miibus_readreg	(device_t, int, int);
295 static int re_miibus_writereg	(device_t, int, int, int);
296 static void re_miibus_statchg	(device_t);
297 
298 static void re_set_jumbo	(struct rl_softc *, int);
299 static void re_set_rxmode		(struct rl_softc *);
300 static void re_reset		(struct rl_softc *);
301 static void re_setwol		(struct rl_softc *);
302 static void re_clrwol		(struct rl_softc *);
303 static void re_set_linkspeed	(struct rl_softc *);
304 
305 #ifdef DEV_NETMAP	/* see ixgbe.c for details */
306 #include <dev/netmap/if_re_netmap.h>
307 #endif /* !DEV_NETMAP */
308 
309 #ifdef RE_DIAG
310 static int re_diag		(struct rl_softc *);
311 #endif
312 
313 static void re_add_sysctls	(struct rl_softc *);
314 static int re_sysctl_stats	(SYSCTL_HANDLER_ARGS);
315 static int sysctl_int_range	(SYSCTL_HANDLER_ARGS, int, int);
316 static int sysctl_hw_re_int_mod	(SYSCTL_HANDLER_ARGS);
317 
318 static device_method_t re_methods[] = {
319 	/* Device interface */
320 	DEVMETHOD(device_probe,		re_probe),
321 	DEVMETHOD(device_attach,	re_attach),
322 	DEVMETHOD(device_detach,	re_detach),
323 	DEVMETHOD(device_suspend,	re_suspend),
324 	DEVMETHOD(device_resume,	re_resume),
325 	DEVMETHOD(device_shutdown,	re_shutdown),
326 
327 	/* MII interface */
328 	DEVMETHOD(miibus_readreg,	re_miibus_readreg),
329 	DEVMETHOD(miibus_writereg,	re_miibus_writereg),
330 	DEVMETHOD(miibus_statchg,	re_miibus_statchg),
331 
332 	DEVMETHOD_END
333 };
334 
335 static driver_t re_driver = {
336 	"re",
337 	re_methods,
338 	sizeof(struct rl_softc)
339 };
340 
341 static devclass_t re_devclass;
342 
343 DRIVER_MODULE(re, pci, re_driver, re_devclass, 0, 0);
344 DRIVER_MODULE(miibus, re, miibus_driver, miibus_devclass, 0, 0);
345 
346 #define EE_SET(x)					\
347 	CSR_WRITE_1(sc, RL_EECMD,			\
348 		CSR_READ_1(sc, RL_EECMD) | x)
349 
350 #define EE_CLR(x)					\
351 	CSR_WRITE_1(sc, RL_EECMD,			\
352 		CSR_READ_1(sc, RL_EECMD) & ~x)
353 
354 /*
355  * Send a read command and address to the EEPROM, check for ACK.
356  */
357 static void
358 re_eeprom_putbyte(struct rl_softc *sc, int addr)
359 {
360 	int			d, i;
361 
362 	d = addr | (RL_9346_READ << sc->rl_eewidth);
363 
364 	/*
365 	 * Feed in each bit and strobe the clock.
366 	 */
367 
368 	for (i = 1 << (sc->rl_eewidth + 3); i; i >>= 1) {
369 		if (d & i) {
370 			EE_SET(RL_EE_DATAIN);
371 		} else {
372 			EE_CLR(RL_EE_DATAIN);
373 		}
374 		DELAY(100);
375 		EE_SET(RL_EE_CLK);
376 		DELAY(150);
377 		EE_CLR(RL_EE_CLK);
378 		DELAY(100);
379 	}
380 }
381 
382 /*
383  * Read a word of data stored in the EEPROM at address 'addr.'
384  */
385 static void
386 re_eeprom_getword(struct rl_softc *sc, int addr, u_int16_t *dest)
387 {
388 	int			i;
389 	u_int16_t		word = 0;
390 
391 	/*
392 	 * Send address of word we want to read.
393 	 */
394 	re_eeprom_putbyte(sc, addr);
395 
396 	/*
397 	 * Start reading bits from EEPROM.
398 	 */
399 	for (i = 0x8000; i; i >>= 1) {
400 		EE_SET(RL_EE_CLK);
401 		DELAY(100);
402 		if (CSR_READ_1(sc, RL_EECMD) & RL_EE_DATAOUT)
403 			word |= i;
404 		EE_CLR(RL_EE_CLK);
405 		DELAY(100);
406 	}
407 
408 	*dest = word;
409 }
410 
411 /*
412  * Read a sequence of words from the EEPROM.
413  */
414 static void
415 re_read_eeprom(struct rl_softc *sc, caddr_t dest, int off, int cnt)
416 {
417 	int			i;
418 	u_int16_t		word = 0, *ptr;
419 
420 	CSR_SETBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM);
421 
422         DELAY(100);
423 
424 	for (i = 0; i < cnt; i++) {
425 		CSR_SETBIT_1(sc, RL_EECMD, RL_EE_SEL);
426 		re_eeprom_getword(sc, off + i, &word);
427 		CSR_CLRBIT_1(sc, RL_EECMD, RL_EE_SEL);
428 		ptr = (u_int16_t *)(dest + (i * 2));
429                 *ptr = word;
430 	}
431 
432 	CSR_CLRBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM);
433 }
434 
435 static int
436 re_gmii_readreg(device_t dev, int phy, int reg)
437 {
438 	struct rl_softc		*sc;
439 	u_int32_t		rval;
440 	int			i;
441 
442 	sc = device_get_softc(dev);
443 
444 	/* Let the rgephy driver read the GMEDIASTAT register */
445 
446 	if (reg == RL_GMEDIASTAT) {
447 		rval = CSR_READ_1(sc, RL_GMEDIASTAT);
448 		return (rval);
449 	}
450 
451 	CSR_WRITE_4(sc, RL_PHYAR, reg << 16);
452 
453 	for (i = 0; i < RL_PHY_TIMEOUT; i++) {
454 		rval = CSR_READ_4(sc, RL_PHYAR);
455 		if (rval & RL_PHYAR_BUSY)
456 			break;
457 		DELAY(25);
458 	}
459 
460 	if (i == RL_PHY_TIMEOUT) {
461 		device_printf(sc->rl_dev, "PHY read failed\n");
462 		return (0);
463 	}
464 
465 	/*
466 	 * Controller requires a 20us delay to process next MDIO request.
467 	 */
468 	DELAY(20);
469 
470 	return (rval & RL_PHYAR_PHYDATA);
471 }
472 
473 static int
474 re_gmii_writereg(device_t dev, int phy, int reg, int data)
475 {
476 	struct rl_softc		*sc;
477 	u_int32_t		rval;
478 	int			i;
479 
480 	sc = device_get_softc(dev);
481 
482 	CSR_WRITE_4(sc, RL_PHYAR, (reg << 16) |
483 	    (data & RL_PHYAR_PHYDATA) | RL_PHYAR_BUSY);
484 
485 	for (i = 0; i < RL_PHY_TIMEOUT; i++) {
486 		rval = CSR_READ_4(sc, RL_PHYAR);
487 		if (!(rval & RL_PHYAR_BUSY))
488 			break;
489 		DELAY(25);
490 	}
491 
492 	if (i == RL_PHY_TIMEOUT) {
493 		device_printf(sc->rl_dev, "PHY write failed\n");
494 		return (0);
495 	}
496 
497 	/*
498 	 * Controller requires a 20us delay to process next MDIO request.
499 	 */
500 	DELAY(20);
501 
502 	return (0);
503 }
504 
505 static int
506 re_miibus_readreg(device_t dev, int phy, int reg)
507 {
508 	struct rl_softc		*sc;
509 	u_int16_t		rval = 0;
510 	u_int16_t		re8139_reg = 0;
511 
512 	sc = device_get_softc(dev);
513 
514 	if (sc->rl_type == RL_8169) {
515 		rval = re_gmii_readreg(dev, phy, reg);
516 		return (rval);
517 	}
518 
519 	switch (reg) {
520 	case MII_BMCR:
521 		re8139_reg = RL_BMCR;
522 		break;
523 	case MII_BMSR:
524 		re8139_reg = RL_BMSR;
525 		break;
526 	case MII_ANAR:
527 		re8139_reg = RL_ANAR;
528 		break;
529 	case MII_ANER:
530 		re8139_reg = RL_ANER;
531 		break;
532 	case MII_ANLPAR:
533 		re8139_reg = RL_LPAR;
534 		break;
535 	case MII_PHYIDR1:
536 	case MII_PHYIDR2:
537 		return (0);
538 	/*
539 	 * Allow the rlphy driver to read the media status
540 	 * register. If we have a link partner which does not
541 	 * support NWAY, this is the register which will tell
542 	 * us the results of parallel detection.
543 	 */
544 	case RL_MEDIASTAT:
545 		rval = CSR_READ_1(sc, RL_MEDIASTAT);
546 		return (rval);
547 	default:
548 		device_printf(sc->rl_dev, "bad phy register\n");
549 		return (0);
550 	}
551 	rval = CSR_READ_2(sc, re8139_reg);
552 	if (sc->rl_type == RL_8139CPLUS && re8139_reg == RL_BMCR) {
553 		/* 8139C+ has different bit layout. */
554 		rval &= ~(BMCR_LOOP | BMCR_ISO);
555 	}
556 	return (rval);
557 }
558 
559 static int
560 re_miibus_writereg(device_t dev, int phy, int reg, int data)
561 {
562 	struct rl_softc		*sc;
563 	u_int16_t		re8139_reg = 0;
564 	int			rval = 0;
565 
566 	sc = device_get_softc(dev);
567 
568 	if (sc->rl_type == RL_8169) {
569 		rval = re_gmii_writereg(dev, phy, reg, data);
570 		return (rval);
571 	}
572 
573 	switch (reg) {
574 	case MII_BMCR:
575 		re8139_reg = RL_BMCR;
576 		if (sc->rl_type == RL_8139CPLUS) {
577 			/* 8139C+ has different bit layout. */
578 			data &= ~(BMCR_LOOP | BMCR_ISO);
579 		}
580 		break;
581 	case MII_BMSR:
582 		re8139_reg = RL_BMSR;
583 		break;
584 	case MII_ANAR:
585 		re8139_reg = RL_ANAR;
586 		break;
587 	case MII_ANER:
588 		re8139_reg = RL_ANER;
589 		break;
590 	case MII_ANLPAR:
591 		re8139_reg = RL_LPAR;
592 		break;
593 	case MII_PHYIDR1:
594 	case MII_PHYIDR2:
595 		return (0);
596 		break;
597 	default:
598 		device_printf(sc->rl_dev, "bad phy register\n");
599 		return (0);
600 	}
601 	CSR_WRITE_2(sc, re8139_reg, data);
602 	return (0);
603 }
604 
605 static void
606 re_miibus_statchg(device_t dev)
607 {
608 	struct rl_softc		*sc;
609 	struct ifnet		*ifp;
610 	struct mii_data		*mii;
611 
612 	sc = device_get_softc(dev);
613 	mii = device_get_softc(sc->rl_miibus);
614 	ifp = sc->rl_ifp;
615 	if (mii == NULL || ifp == NULL ||
616 	    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
617 		return;
618 
619 	sc->rl_flags &= ~RL_FLAG_LINK;
620 	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
621 	    (IFM_ACTIVE | IFM_AVALID)) {
622 		switch (IFM_SUBTYPE(mii->mii_media_active)) {
623 		case IFM_10_T:
624 		case IFM_100_TX:
625 			sc->rl_flags |= RL_FLAG_LINK;
626 			break;
627 		case IFM_1000_T:
628 			if ((sc->rl_flags & RL_FLAG_FASTETHER) != 0)
629 				break;
630 			sc->rl_flags |= RL_FLAG_LINK;
631 			break;
632 		default:
633 			break;
634 		}
635 	}
636 	/*
637 	 * RealTek controllers does not provide any interface to
638 	 * Tx/Rx MACs for resolved speed, duplex and flow-control
639 	 * parameters.
640 	 */
641 }
642 
643 /*
644  * Set the RX configuration and 64-bit multicast hash filter.
645  */
646 static void
647 re_set_rxmode(struct rl_softc *sc)
648 {
649 	struct ifnet		*ifp;
650 	struct ifmultiaddr	*ifma;
651 	uint32_t		hashes[2] = { 0, 0 };
652 	uint32_t		h, rxfilt;
653 
654 	RL_LOCK_ASSERT(sc);
655 
656 	ifp = sc->rl_ifp;
657 
658 	rxfilt = RL_RXCFG_CONFIG | RL_RXCFG_RX_INDIV | RL_RXCFG_RX_BROAD;
659 
660 	if (ifp->if_flags & (IFF_ALLMULTI | IFF_PROMISC)) {
661 		if (ifp->if_flags & IFF_PROMISC)
662 			rxfilt |= RL_RXCFG_RX_ALLPHYS;
663 		/*
664 		 * Unlike other hardwares, we have to explicitly set
665 		 * RL_RXCFG_RX_MULTI to receive multicast frames in
666 		 * promiscuous mode.
667 		 */
668 		rxfilt |= RL_RXCFG_RX_MULTI;
669 		hashes[0] = hashes[1] = 0xffffffff;
670 		goto done;
671 	}
672 
673 	if_maddr_rlock(ifp);
674 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
675 		if (ifma->ifma_addr->sa_family != AF_LINK)
676 			continue;
677 		h = ether_crc32_be(LLADDR((struct sockaddr_dl *)
678 		    ifma->ifma_addr), ETHER_ADDR_LEN) >> 26;
679 		if (h < 32)
680 			hashes[0] |= (1 << h);
681 		else
682 			hashes[1] |= (1 << (h - 32));
683 	}
684 	if_maddr_runlock(ifp);
685 
686 	if (hashes[0] != 0 || hashes[1] != 0) {
687 		/*
688 		 * For some unfathomable reason, RealTek decided to
689 		 * reverse the order of the multicast hash registers
690 		 * in the PCI Express parts.  This means we have to
691 		 * write the hash pattern in reverse order for those
692 		 * devices.
693 		 */
694 		if ((sc->rl_flags & RL_FLAG_PCIE) != 0) {
695 			h = bswap32(hashes[0]);
696 			hashes[0] = bswap32(hashes[1]);
697 			hashes[1] = h;
698 		}
699 		rxfilt |= RL_RXCFG_RX_MULTI;
700 	}
701 
702 done:
703 	CSR_WRITE_4(sc, RL_MAR0, hashes[0]);
704 	CSR_WRITE_4(sc, RL_MAR4, hashes[1]);
705 	CSR_WRITE_4(sc, RL_RXCFG, rxfilt);
706 }
707 
708 static void
709 re_reset(struct rl_softc *sc)
710 {
711 	int			i;
712 
713 	RL_LOCK_ASSERT(sc);
714 
715 	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RESET);
716 
717 	for (i = 0; i < RL_TIMEOUT; i++) {
718 		DELAY(10);
719 		if (!(CSR_READ_1(sc, RL_COMMAND) & RL_CMD_RESET))
720 			break;
721 	}
722 	if (i == RL_TIMEOUT)
723 		device_printf(sc->rl_dev, "reset never completed!\n");
724 
725 	if ((sc->rl_flags & RL_FLAG_MACRESET) != 0)
726 		CSR_WRITE_1(sc, 0x82, 1);
727 	if (sc->rl_hwrev->rl_rev == RL_HWREV_8169S)
728 		re_gmii_writereg(sc->rl_dev, 1, 0x0b, 0);
729 }
730 
731 #ifdef RE_DIAG
732 
733 /*
734  * The following routine is designed to test for a defect on some
735  * 32-bit 8169 cards. Some of these NICs have the REQ64# and ACK64#
736  * lines connected to the bus, however for a 32-bit only card, they
737  * should be pulled high. The result of this defect is that the
738  * NIC will not work right if you plug it into a 64-bit slot: DMA
739  * operations will be done with 64-bit transfers, which will fail
740  * because the 64-bit data lines aren't connected.
741  *
742  * There's no way to work around this (short of talking a soldering
743  * iron to the board), however we can detect it. The method we use
744  * here is to put the NIC into digital loopback mode, set the receiver
745  * to promiscuous mode, and then try to send a frame. We then compare
746  * the frame data we sent to what was received. If the data matches,
747  * then the NIC is working correctly, otherwise we know the user has
748  * a defective NIC which has been mistakenly plugged into a 64-bit PCI
749  * slot. In the latter case, there's no way the NIC can work correctly,
750  * so we print out a message on the console and abort the device attach.
751  */
752 
753 static int
754 re_diag(struct rl_softc *sc)
755 {
756 	struct ifnet		*ifp = sc->rl_ifp;
757 	struct mbuf		*m0;
758 	struct ether_header	*eh;
759 	struct rl_desc		*cur_rx;
760 	u_int16_t		status;
761 	u_int32_t		rxstat;
762 	int			total_len, i, error = 0, phyaddr;
763 	u_int8_t		dst[] = { 0x00, 'h', 'e', 'l', 'l', 'o' };
764 	u_int8_t		src[] = { 0x00, 'w', 'o', 'r', 'l', 'd' };
765 
766 	/* Allocate a single mbuf */
767 	MGETHDR(m0, M_NOWAIT, MT_DATA);
768 	if (m0 == NULL)
769 		return (ENOBUFS);
770 
771 	RL_LOCK(sc);
772 
773 	/*
774 	 * Initialize the NIC in test mode. This sets the chip up
775 	 * so that it can send and receive frames, but performs the
776 	 * following special functions:
777 	 * - Puts receiver in promiscuous mode
778 	 * - Enables digital loopback mode
779 	 * - Leaves interrupts turned off
780 	 */
781 
782 	ifp->if_flags |= IFF_PROMISC;
783 	sc->rl_testmode = 1;
784 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
785 	re_init_locked(sc);
786 	sc->rl_flags |= RL_FLAG_LINK;
787 	if (sc->rl_type == RL_8169)
788 		phyaddr = 1;
789 	else
790 		phyaddr = 0;
791 
792 	re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_RESET);
793 	for (i = 0; i < RL_TIMEOUT; i++) {
794 		status = re_miibus_readreg(sc->rl_dev, phyaddr, MII_BMCR);
795 		if (!(status & BMCR_RESET))
796 			break;
797 	}
798 
799 	re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_LOOP);
800 	CSR_WRITE_2(sc, RL_ISR, RL_INTRS);
801 
802 	DELAY(100000);
803 
804 	/* Put some data in the mbuf */
805 
806 	eh = mtod(m0, struct ether_header *);
807 	bcopy ((char *)&dst, eh->ether_dhost, ETHER_ADDR_LEN);
808 	bcopy ((char *)&src, eh->ether_shost, ETHER_ADDR_LEN);
809 	eh->ether_type = htons(ETHERTYPE_IP);
810 	m0->m_pkthdr.len = m0->m_len = ETHER_MIN_LEN - ETHER_CRC_LEN;
811 
812 	/*
813 	 * Queue the packet, start transmission.
814 	 * Note: IF_HANDOFF() ultimately calls re_start() for us.
815 	 */
816 
817 	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
818 	RL_UNLOCK(sc);
819 	/* XXX: re_diag must not be called when in ALTQ mode */
820 	IF_HANDOFF(&ifp->if_snd, m0, ifp);
821 	RL_LOCK(sc);
822 	m0 = NULL;
823 
824 	/* Wait for it to propagate through the chip */
825 
826 	DELAY(100000);
827 	for (i = 0; i < RL_TIMEOUT; i++) {
828 		status = CSR_READ_2(sc, RL_ISR);
829 		CSR_WRITE_2(sc, RL_ISR, status);
830 		if ((status & (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK)) ==
831 		    (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK))
832 			break;
833 		DELAY(10);
834 	}
835 
836 	if (i == RL_TIMEOUT) {
837 		device_printf(sc->rl_dev,
838 		    "diagnostic failed, failed to receive packet in"
839 		    " loopback mode\n");
840 		error = EIO;
841 		goto done;
842 	}
843 
844 	/*
845 	 * The packet should have been dumped into the first
846 	 * entry in the RX DMA ring. Grab it from there.
847 	 */
848 
849 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
850 	    sc->rl_ldata.rl_rx_list_map,
851 	    BUS_DMASYNC_POSTREAD);
852 	bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag,
853 	    sc->rl_ldata.rl_rx_desc[0].rx_dmamap,
854 	    BUS_DMASYNC_POSTREAD);
855 	bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag,
856 	    sc->rl_ldata.rl_rx_desc[0].rx_dmamap);
857 
858 	m0 = sc->rl_ldata.rl_rx_desc[0].rx_m;
859 	sc->rl_ldata.rl_rx_desc[0].rx_m = NULL;
860 	eh = mtod(m0, struct ether_header *);
861 
862 	cur_rx = &sc->rl_ldata.rl_rx_list[0];
863 	total_len = RL_RXBYTES(cur_rx);
864 	rxstat = le32toh(cur_rx->rl_cmdstat);
865 
866 	if (total_len != ETHER_MIN_LEN) {
867 		device_printf(sc->rl_dev,
868 		    "diagnostic failed, received short packet\n");
869 		error = EIO;
870 		goto done;
871 	}
872 
873 	/* Test that the received packet data matches what we sent. */
874 
875 	if (bcmp((char *)&eh->ether_dhost, (char *)&dst, ETHER_ADDR_LEN) ||
876 	    bcmp((char *)&eh->ether_shost, (char *)&src, ETHER_ADDR_LEN) ||
877 	    ntohs(eh->ether_type) != ETHERTYPE_IP) {
878 		device_printf(sc->rl_dev, "WARNING, DMA FAILURE!\n");
879 		device_printf(sc->rl_dev, "expected TX data: %6D/%6D/0x%x\n",
880 		    dst, ":", src, ":", ETHERTYPE_IP);
881 		device_printf(sc->rl_dev, "received RX data: %6D/%6D/0x%x\n",
882 		    eh->ether_dhost, ":", eh->ether_shost, ":",
883 		    ntohs(eh->ether_type));
884 		device_printf(sc->rl_dev, "You may have a defective 32-bit "
885 		    "NIC plugged into a 64-bit PCI slot.\n");
886 		device_printf(sc->rl_dev, "Please re-install the NIC in a "
887 		    "32-bit slot for proper operation.\n");
888 		device_printf(sc->rl_dev, "Read the re(4) man page for more "
889 		    "details.\n");
890 		error = EIO;
891 	}
892 
893 done:
894 	/* Turn interface off, release resources */
895 
896 	sc->rl_testmode = 0;
897 	sc->rl_flags &= ~RL_FLAG_LINK;
898 	ifp->if_flags &= ~IFF_PROMISC;
899 	re_stop(sc);
900 	if (m0 != NULL)
901 		m_freem(m0);
902 
903 	RL_UNLOCK(sc);
904 
905 	return (error);
906 }
907 
908 #endif
909 
910 /*
911  * Probe for a RealTek 8139C+/8169/8110 chip. Check the PCI vendor and device
912  * IDs against our list and return a device name if we find a match.
913  */
914 static int
915 re_probe(device_t dev)
916 {
917 	const struct rl_type	*t;
918 	uint16_t		devid, vendor;
919 	uint16_t		revid, sdevid;
920 	int			i;
921 
922 	vendor = pci_get_vendor(dev);
923 	devid = pci_get_device(dev);
924 	revid = pci_get_revid(dev);
925 	sdevid = pci_get_subdevice(dev);
926 
927 	if (vendor == LINKSYS_VENDORID && devid == LINKSYS_DEVICEID_EG1032) {
928 		if (sdevid != LINKSYS_SUBDEVICE_EG1032_REV3) {
929 			/*
930 			 * Only attach to rev. 3 of the Linksys EG1032 adapter.
931 			 * Rev. 2 is supported by sk(4).
932 			 */
933 			return (ENXIO);
934 		}
935 	}
936 
937 	if (vendor == RT_VENDORID && devid == RT_DEVICEID_8139) {
938 		if (revid != 0x20) {
939 			/* 8139, let rl(4) take care of this device. */
940 			return (ENXIO);
941 		}
942 	}
943 
944 	t = re_devs;
945 	for (i = 0; i < sizeof(re_devs) / sizeof(re_devs[0]); i++, t++) {
946 		if (vendor == t->rl_vid && devid == t->rl_did) {
947 			device_set_desc(dev, t->rl_name);
948 			return (BUS_PROBE_DEFAULT);
949 		}
950 	}
951 
952 	return (ENXIO);
953 }
954 
955 /*
956  * Map a single buffer address.
957  */
958 
959 static void
960 re_dma_map_addr(void *arg, bus_dma_segment_t *segs, int nseg, int error)
961 {
962 	bus_addr_t		*addr;
963 
964 	if (error)
965 		return;
966 
967 	KASSERT(nseg == 1, ("too many DMA segments, %d should be 1", nseg));
968 	addr = arg;
969 	*addr = segs->ds_addr;
970 }
971 
972 static int
973 re_allocmem(device_t dev, struct rl_softc *sc)
974 {
975 	bus_addr_t		lowaddr;
976 	bus_size_t		rx_list_size, tx_list_size;
977 	int			error;
978 	int			i;
979 
980 	rx_list_size = sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc);
981 	tx_list_size = sc->rl_ldata.rl_tx_desc_cnt * sizeof(struct rl_desc);
982 
983 	/*
984 	 * Allocate the parent bus DMA tag appropriate for PCI.
985 	 * In order to use DAC, RL_CPLUSCMD_PCI_DAC bit of RL_CPLUS_CMD
986 	 * register should be set. However some RealTek chips are known
987 	 * to be buggy on DAC handling, therefore disable DAC by limiting
988 	 * DMA address space to 32bit. PCIe variants of RealTek chips
989 	 * may not have the limitation.
990 	 */
991 	lowaddr = BUS_SPACE_MAXADDR;
992 	if ((sc->rl_flags & RL_FLAG_PCIE) == 0)
993 		lowaddr = BUS_SPACE_MAXADDR_32BIT;
994 	error = bus_dma_tag_create(bus_get_dma_tag(dev), 1, 0,
995 	    lowaddr, BUS_SPACE_MAXADDR, NULL, NULL,
996 	    BUS_SPACE_MAXSIZE_32BIT, 0, BUS_SPACE_MAXSIZE_32BIT, 0,
997 	    NULL, NULL, &sc->rl_parent_tag);
998 	if (error) {
999 		device_printf(dev, "could not allocate parent DMA tag\n");
1000 		return (error);
1001 	}
1002 
1003 	/*
1004 	 * Allocate map for TX mbufs.
1005 	 */
1006 	error = bus_dma_tag_create(sc->rl_parent_tag, 1, 0,
1007 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL,
1008 	    NULL, MCLBYTES * RL_NTXSEGS, RL_NTXSEGS, 4096, 0,
1009 	    NULL, NULL, &sc->rl_ldata.rl_tx_mtag);
1010 	if (error) {
1011 		device_printf(dev, "could not allocate TX DMA tag\n");
1012 		return (error);
1013 	}
1014 
1015 	/*
1016 	 * Allocate map for RX mbufs.
1017 	 */
1018 
1019 	if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
1020 		error = bus_dma_tag_create(sc->rl_parent_tag, sizeof(uint64_t),
1021 		    0, BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
1022 		    MJUM9BYTES, 1, MJUM9BYTES, 0, NULL, NULL,
1023 		    &sc->rl_ldata.rl_jrx_mtag);
1024 		if (error) {
1025 			device_printf(dev,
1026 			    "could not allocate jumbo RX DMA tag\n");
1027 			return (error);
1028 		}
1029 	}
1030 	error = bus_dma_tag_create(sc->rl_parent_tag, sizeof(uint64_t), 0,
1031 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
1032 	    MCLBYTES, 1, MCLBYTES, 0, NULL, NULL, &sc->rl_ldata.rl_rx_mtag);
1033 	if (error) {
1034 		device_printf(dev, "could not allocate RX DMA tag\n");
1035 		return (error);
1036 	}
1037 
1038 	/*
1039 	 * Allocate map for TX descriptor list.
1040 	 */
1041 	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
1042 	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1043 	    NULL, tx_list_size, 1, tx_list_size, 0,
1044 	    NULL, NULL, &sc->rl_ldata.rl_tx_list_tag);
1045 	if (error) {
1046 		device_printf(dev, "could not allocate TX DMA ring tag\n");
1047 		return (error);
1048 	}
1049 
1050 	/* Allocate DMA'able memory for the TX ring */
1051 
1052 	error = bus_dmamem_alloc(sc->rl_ldata.rl_tx_list_tag,
1053 	    (void **)&sc->rl_ldata.rl_tx_list,
1054 	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1055 	    &sc->rl_ldata.rl_tx_list_map);
1056 	if (error) {
1057 		device_printf(dev, "could not allocate TX DMA ring\n");
1058 		return (error);
1059 	}
1060 
1061 	/* Load the map for the TX ring. */
1062 
1063 	sc->rl_ldata.rl_tx_list_addr = 0;
1064 	error = bus_dmamap_load(sc->rl_ldata.rl_tx_list_tag,
1065 	     sc->rl_ldata.rl_tx_list_map, sc->rl_ldata.rl_tx_list,
1066 	     tx_list_size, re_dma_map_addr,
1067 	     &sc->rl_ldata.rl_tx_list_addr, BUS_DMA_NOWAIT);
1068 	if (error != 0 || sc->rl_ldata.rl_tx_list_addr == 0) {
1069 		device_printf(dev, "could not load TX DMA ring\n");
1070 		return (ENOMEM);
1071 	}
1072 
1073 	/* Create DMA maps for TX buffers */
1074 
1075 	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
1076 		error = bus_dmamap_create(sc->rl_ldata.rl_tx_mtag, 0,
1077 		    &sc->rl_ldata.rl_tx_desc[i].tx_dmamap);
1078 		if (error) {
1079 			device_printf(dev, "could not create DMA map for TX\n");
1080 			return (error);
1081 		}
1082 	}
1083 
1084 	/*
1085 	 * Allocate map for RX descriptor list.
1086 	 */
1087 	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
1088 	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1089 	    NULL, rx_list_size, 1, rx_list_size, 0,
1090 	    NULL, NULL, &sc->rl_ldata.rl_rx_list_tag);
1091 	if (error) {
1092 		device_printf(dev, "could not create RX DMA ring tag\n");
1093 		return (error);
1094 	}
1095 
1096 	/* Allocate DMA'able memory for the RX ring */
1097 
1098 	error = bus_dmamem_alloc(sc->rl_ldata.rl_rx_list_tag,
1099 	    (void **)&sc->rl_ldata.rl_rx_list,
1100 	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1101 	    &sc->rl_ldata.rl_rx_list_map);
1102 	if (error) {
1103 		device_printf(dev, "could not allocate RX DMA ring\n");
1104 		return (error);
1105 	}
1106 
1107 	/* Load the map for the RX ring. */
1108 
1109 	sc->rl_ldata.rl_rx_list_addr = 0;
1110 	error = bus_dmamap_load(sc->rl_ldata.rl_rx_list_tag,
1111 	     sc->rl_ldata.rl_rx_list_map, sc->rl_ldata.rl_rx_list,
1112 	     rx_list_size, re_dma_map_addr,
1113 	     &sc->rl_ldata.rl_rx_list_addr, BUS_DMA_NOWAIT);
1114 	if (error != 0 || sc->rl_ldata.rl_rx_list_addr == 0) {
1115 		device_printf(dev, "could not load RX DMA ring\n");
1116 		return (ENOMEM);
1117 	}
1118 
1119 	/* Create DMA maps for RX buffers */
1120 
1121 	if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
1122 		error = bus_dmamap_create(sc->rl_ldata.rl_jrx_mtag, 0,
1123 		    &sc->rl_ldata.rl_jrx_sparemap);
1124 		if (error) {
1125 			device_printf(dev,
1126 			    "could not create spare DMA map for jumbo RX\n");
1127 			return (error);
1128 		}
1129 		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1130 			error = bus_dmamap_create(sc->rl_ldata.rl_jrx_mtag, 0,
1131 			    &sc->rl_ldata.rl_jrx_desc[i].rx_dmamap);
1132 			if (error) {
1133 				device_printf(dev,
1134 				    "could not create DMA map for jumbo RX\n");
1135 				return (error);
1136 			}
1137 		}
1138 	}
1139 	error = bus_dmamap_create(sc->rl_ldata.rl_rx_mtag, 0,
1140 	    &sc->rl_ldata.rl_rx_sparemap);
1141 	if (error) {
1142 		device_printf(dev, "could not create spare DMA map for RX\n");
1143 		return (error);
1144 	}
1145 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1146 		error = bus_dmamap_create(sc->rl_ldata.rl_rx_mtag, 0,
1147 		    &sc->rl_ldata.rl_rx_desc[i].rx_dmamap);
1148 		if (error) {
1149 			device_printf(dev, "could not create DMA map for RX\n");
1150 			return (error);
1151 		}
1152 	}
1153 
1154 	/* Create DMA map for statistics. */
1155 	error = bus_dma_tag_create(sc->rl_parent_tag, RL_DUMP_ALIGN, 0,
1156 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
1157 	    sizeof(struct rl_stats), 1, sizeof(struct rl_stats), 0, NULL, NULL,
1158 	    &sc->rl_ldata.rl_stag);
1159 	if (error) {
1160 		device_printf(dev, "could not create statistics DMA tag\n");
1161 		return (error);
1162 	}
1163 	/* Allocate DMA'able memory for statistics. */
1164 	error = bus_dmamem_alloc(sc->rl_ldata.rl_stag,
1165 	    (void **)&sc->rl_ldata.rl_stats,
1166 	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1167 	    &sc->rl_ldata.rl_smap);
1168 	if (error) {
1169 		device_printf(dev,
1170 		    "could not allocate statistics DMA memory\n");
1171 		return (error);
1172 	}
1173 	/* Load the map for statistics. */
1174 	sc->rl_ldata.rl_stats_addr = 0;
1175 	error = bus_dmamap_load(sc->rl_ldata.rl_stag, sc->rl_ldata.rl_smap,
1176 	    sc->rl_ldata.rl_stats, sizeof(struct rl_stats), re_dma_map_addr,
1177 	     &sc->rl_ldata.rl_stats_addr, BUS_DMA_NOWAIT);
1178 	if (error != 0 || sc->rl_ldata.rl_stats_addr == 0) {
1179 		device_printf(dev, "could not load statistics DMA memory\n");
1180 		return (ENOMEM);
1181 	}
1182 
1183 	return (0);
1184 }
1185 
1186 /*
1187  * Attach the interface. Allocate softc structures, do ifmedia
1188  * setup and ethernet/BPF attach.
1189  */
1190 static int
1191 re_attach(device_t dev)
1192 {
1193 	u_char			eaddr[ETHER_ADDR_LEN];
1194 	u_int16_t		as[ETHER_ADDR_LEN / 2];
1195 	struct rl_softc		*sc;
1196 	struct ifnet		*ifp;
1197 	const struct rl_hwrev	*hw_rev;
1198 	u_int32_t		cap, ctl;
1199 	int			hwrev;
1200 	u_int16_t		devid, re_did = 0;
1201 	int			error = 0, i, phy, rid;
1202 	int			msic, msixc, reg;
1203 	uint8_t			cfg;
1204 
1205 	sc = device_get_softc(dev);
1206 	sc->rl_dev = dev;
1207 
1208 	mtx_init(&sc->rl_mtx, device_get_nameunit(dev), MTX_NETWORK_LOCK,
1209 	    MTX_DEF);
1210 	callout_init_mtx(&sc->rl_stat_callout, &sc->rl_mtx, 0);
1211 
1212 	/*
1213 	 * Map control/status registers.
1214 	 */
1215 	pci_enable_busmaster(dev);
1216 
1217 	devid = pci_get_device(dev);
1218 	/*
1219 	 * Prefer memory space register mapping over IO space.
1220 	 * Because RTL8169SC does not seem to work when memory mapping
1221 	 * is used always activate io mapping.
1222 	 */
1223 	if (devid == RT_DEVICEID_8169SC)
1224 		prefer_iomap = 1;
1225 	if (prefer_iomap == 0) {
1226 		sc->rl_res_id = PCIR_BAR(1);
1227 		sc->rl_res_type = SYS_RES_MEMORY;
1228 		/* RTL8168/8101E seems to use different BARs. */
1229 		if (devid == RT_DEVICEID_8168 || devid == RT_DEVICEID_8101E)
1230 			sc->rl_res_id = PCIR_BAR(2);
1231 	} else {
1232 		sc->rl_res_id = PCIR_BAR(0);
1233 		sc->rl_res_type = SYS_RES_IOPORT;
1234 	}
1235 	sc->rl_res = bus_alloc_resource_any(dev, sc->rl_res_type,
1236 	    &sc->rl_res_id, RF_ACTIVE);
1237 	if (sc->rl_res == NULL && prefer_iomap == 0) {
1238 		sc->rl_res_id = PCIR_BAR(0);
1239 		sc->rl_res_type = SYS_RES_IOPORT;
1240 		sc->rl_res = bus_alloc_resource_any(dev, sc->rl_res_type,
1241 		    &sc->rl_res_id, RF_ACTIVE);
1242 	}
1243 	if (sc->rl_res == NULL) {
1244 		device_printf(dev, "couldn't map ports/memory\n");
1245 		error = ENXIO;
1246 		goto fail;
1247 	}
1248 
1249 	sc->rl_btag = rman_get_bustag(sc->rl_res);
1250 	sc->rl_bhandle = rman_get_bushandle(sc->rl_res);
1251 
1252 	msic = pci_msi_count(dev);
1253 	msixc = pci_msix_count(dev);
1254 	if (pci_find_cap(dev, PCIY_EXPRESS, &reg) == 0) {
1255 		sc->rl_flags |= RL_FLAG_PCIE;
1256 		sc->rl_expcap = reg;
1257 	}
1258 	if (bootverbose) {
1259 		device_printf(dev, "MSI count : %d\n", msic);
1260 		device_printf(dev, "MSI-X count : %d\n", msixc);
1261 	}
1262 	if (msix_disable > 0)
1263 		msixc = 0;
1264 	if (msi_disable > 0)
1265 		msic = 0;
1266 	/* Prefer MSI-X to MSI. */
1267 	if (msixc > 0) {
1268 		msixc = 1;
1269 		rid = PCIR_BAR(4);
1270 		sc->rl_res_pba = bus_alloc_resource_any(dev, SYS_RES_MEMORY,
1271 		    &rid, RF_ACTIVE);
1272 		if (sc->rl_res_pba == NULL) {
1273 			device_printf(sc->rl_dev,
1274 			    "could not allocate MSI-X PBA resource\n");
1275 		}
1276 		if (sc->rl_res_pba != NULL &&
1277 		    pci_alloc_msix(dev, &msixc) == 0) {
1278 			if (msixc == 1) {
1279 				device_printf(dev, "Using %d MSI-X message\n",
1280 				    msixc);
1281 				sc->rl_flags |= RL_FLAG_MSIX;
1282 			} else
1283 				pci_release_msi(dev);
1284 		}
1285 		if ((sc->rl_flags & RL_FLAG_MSIX) == 0) {
1286 			if (sc->rl_res_pba != NULL)
1287 				bus_release_resource(dev, SYS_RES_MEMORY, rid,
1288 				    sc->rl_res_pba);
1289 			sc->rl_res_pba = NULL;
1290 			msixc = 0;
1291 		}
1292 	}
1293 	/* Prefer MSI to INTx. */
1294 	if (msixc == 0 && msic > 0) {
1295 		msic = 1;
1296 		if (pci_alloc_msi(dev, &msic) == 0) {
1297 			if (msic == RL_MSI_MESSAGES) {
1298 				device_printf(dev, "Using %d MSI message\n",
1299 				    msic);
1300 				sc->rl_flags |= RL_FLAG_MSI;
1301 				/* Explicitly set MSI enable bit. */
1302 				CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1303 				cfg = CSR_READ_1(sc, RL_CFG2);
1304 				cfg |= RL_CFG2_MSI;
1305 				CSR_WRITE_1(sc, RL_CFG2, cfg);
1306 				CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1307 			} else
1308 				pci_release_msi(dev);
1309 		}
1310 		if ((sc->rl_flags & RL_FLAG_MSI) == 0)
1311 			msic = 0;
1312 	}
1313 
1314 	/* Allocate interrupt */
1315 	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) == 0) {
1316 		rid = 0;
1317 		sc->rl_irq[0] = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
1318 		    RF_SHAREABLE | RF_ACTIVE);
1319 		if (sc->rl_irq[0] == NULL) {
1320 			device_printf(dev, "couldn't allocate IRQ resources\n");
1321 			error = ENXIO;
1322 			goto fail;
1323 		}
1324 	} else {
1325 		for (i = 0, rid = 1; i < RL_MSI_MESSAGES; i++, rid++) {
1326 			sc->rl_irq[i] = bus_alloc_resource_any(dev,
1327 			    SYS_RES_IRQ, &rid, RF_ACTIVE);
1328 			if (sc->rl_irq[i] == NULL) {
1329 				device_printf(dev,
1330 				    "couldn't allocate IRQ resources for "
1331 				    "message %d\n", rid);
1332 				error = ENXIO;
1333 				goto fail;
1334 			}
1335 		}
1336 	}
1337 
1338 	if ((sc->rl_flags & RL_FLAG_MSI) == 0) {
1339 		CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1340 		cfg = CSR_READ_1(sc, RL_CFG2);
1341 		if ((cfg & RL_CFG2_MSI) != 0) {
1342 			device_printf(dev, "turning off MSI enable bit.\n");
1343 			cfg &= ~RL_CFG2_MSI;
1344 			CSR_WRITE_1(sc, RL_CFG2, cfg);
1345 		}
1346 		CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1347 	}
1348 
1349 	/* Disable ASPM L0S/L1. */
1350 	if (sc->rl_expcap != 0) {
1351 		cap = pci_read_config(dev, sc->rl_expcap +
1352 		    PCIER_LINK_CAP, 2);
1353 		if ((cap & PCIEM_LINK_CAP_ASPM) != 0) {
1354 			ctl = pci_read_config(dev, sc->rl_expcap +
1355 			    PCIER_LINK_CTL, 2);
1356 			if ((ctl & PCIEM_LINK_CTL_ASPMC) != 0) {
1357 				ctl &= ~PCIEM_LINK_CTL_ASPMC;
1358 				pci_write_config(dev, sc->rl_expcap +
1359 				    PCIER_LINK_CTL, ctl, 2);
1360 				device_printf(dev, "ASPM disabled\n");
1361 			}
1362 		} else
1363 			device_printf(dev, "no ASPM capability\n");
1364 	}
1365 
1366 	hw_rev = re_hwrevs;
1367 	hwrev = CSR_READ_4(sc, RL_TXCFG);
1368 	switch (hwrev & 0x70000000) {
1369 	case 0x00000000:
1370 	case 0x10000000:
1371 		device_printf(dev, "Chip rev. 0x%08x\n", hwrev & 0xfc800000);
1372 		hwrev &= (RL_TXCFG_HWREV | 0x80000000);
1373 		break;
1374 	default:
1375 		device_printf(dev, "Chip rev. 0x%08x\n", hwrev & 0x7c800000);
1376 		sc->rl_macrev = hwrev & 0x00700000;
1377 		hwrev &= RL_TXCFG_HWREV;
1378 		break;
1379 	}
1380 	device_printf(dev, "MAC rev. 0x%08x\n", sc->rl_macrev);
1381 	while (hw_rev->rl_desc != NULL) {
1382 		if (hw_rev->rl_rev == hwrev) {
1383 			sc->rl_type = hw_rev->rl_type;
1384 			sc->rl_hwrev = hw_rev;
1385 			break;
1386 		}
1387 		hw_rev++;
1388 	}
1389 	if (hw_rev->rl_desc == NULL) {
1390 		device_printf(dev, "Unknown H/W revision: 0x%08x\n", hwrev);
1391 		error = ENXIO;
1392 		goto fail;
1393 	}
1394 
1395 	switch (hw_rev->rl_rev) {
1396 	case RL_HWREV_8139CPLUS:
1397 		sc->rl_flags |= RL_FLAG_FASTETHER | RL_FLAG_AUTOPAD;
1398 		break;
1399 	case RL_HWREV_8100E:
1400 	case RL_HWREV_8101E:
1401 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_FASTETHER;
1402 		break;
1403 	case RL_HWREV_8102E:
1404 	case RL_HWREV_8102EL:
1405 	case RL_HWREV_8102EL_SPIN1:
1406 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR | RL_FLAG_DESCV2 |
1407 		    RL_FLAG_MACSTAT | RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP |
1408 		    RL_FLAG_AUTOPAD;
1409 		break;
1410 	case RL_HWREV_8103E:
1411 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR | RL_FLAG_DESCV2 |
1412 		    RL_FLAG_MACSTAT | RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP |
1413 		    RL_FLAG_AUTOPAD | RL_FLAG_MACSLEEP;
1414 		break;
1415 	case RL_HWREV_8401E:
1416 	case RL_HWREV_8105E:
1417 	case RL_HWREV_8105E_SPIN1:
1418 	case RL_HWREV_8106E:
1419 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
1420 		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
1421 		    RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD;
1422 		break;
1423 	case RL_HWREV_8402:
1424 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
1425 		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
1426 		    RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD |
1427 		    RL_FLAG_CMDSTOP_WAIT_TXQ;
1428 		break;
1429 	case RL_HWREV_8168B_SPIN1:
1430 	case RL_HWREV_8168B_SPIN2:
1431 		sc->rl_flags |= RL_FLAG_WOLRXENB;
1432 		/* FALLTHROUGH */
1433 	case RL_HWREV_8168B_SPIN3:
1434 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_MACSTAT;
1435 		break;
1436 	case RL_HWREV_8168C_SPIN2:
1437 		sc->rl_flags |= RL_FLAG_MACSLEEP;
1438 		/* FALLTHROUGH */
1439 	case RL_HWREV_8168C:
1440 		if (sc->rl_macrev == 0x00200000)
1441 			sc->rl_flags |= RL_FLAG_MACSLEEP;
1442 		/* FALLTHROUGH */
1443 	case RL_HWREV_8168CP:
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 | RL_FLAG_WOL_MANLINK;
1447 		break;
1448 	case RL_HWREV_8168D:
1449 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
1450 		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
1451 		    RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 |
1452 		    RL_FLAG_WOL_MANLINK;
1453 		break;
1454 	case RL_HWREV_8168DP:
1455 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
1456 		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_AUTOPAD |
1457 		    RL_FLAG_JUMBOV2 | RL_FLAG_WAIT_TXPOLL | RL_FLAG_WOL_MANLINK;
1458 		break;
1459 	case RL_HWREV_8168E:
1460 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
1461 		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
1462 		    RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 |
1463 		    RL_FLAG_WOL_MANLINK;
1464 		break;
1465 	case RL_HWREV_8168E_VL:
1466 	case RL_HWREV_8168EP:
1467 	case RL_HWREV_8168F:
1468 	case RL_HWREV_8168G:
1469 	case RL_HWREV_8411:
1470 	case RL_HWREV_8411B:
1471 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
1472 		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP |
1473 		    RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 |
1474 		    RL_FLAG_CMDSTOP_WAIT_TXQ | RL_FLAG_WOL_MANLINK;
1475 		break;
1476 	case RL_HWREV_8168GU:
1477 		if (pci_get_device(dev) == RT_DEVICEID_8101E) {
1478 			/* RTL8106EUS */
1479 			sc->rl_flags |= RL_FLAG_FASTETHER;
1480 		} else
1481 			sc->rl_flags |= RL_FLAG_JUMBOV2 | RL_FLAG_WOL_MANLINK;
1482 
1483 		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
1484 		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP |
1485 		    RL_FLAG_AUTOPAD | RL_FLAG_CMDSTOP_WAIT_TXQ;
1486 		break;
1487 	case RL_HWREV_8169_8110SB:
1488 	case RL_HWREV_8169_8110SBL:
1489 	case RL_HWREV_8169_8110SC:
1490 	case RL_HWREV_8169_8110SCE:
1491 		sc->rl_flags |= RL_FLAG_PHYWAKE;
1492 		/* FALLTHROUGH */
1493 	case RL_HWREV_8169:
1494 	case RL_HWREV_8169S:
1495 	case RL_HWREV_8110S:
1496 		sc->rl_flags |= RL_FLAG_MACRESET;
1497 		break;
1498 	default:
1499 		break;
1500 	}
1501 
1502 	if (sc->rl_hwrev->rl_rev == RL_HWREV_8139CPLUS) {
1503 		sc->rl_cfg0 = RL_8139_CFG0;
1504 		sc->rl_cfg1 = RL_8139_CFG1;
1505 		sc->rl_cfg2 = 0;
1506 		sc->rl_cfg3 = RL_8139_CFG3;
1507 		sc->rl_cfg4 = RL_8139_CFG4;
1508 		sc->rl_cfg5 = RL_8139_CFG5;
1509 	} else {
1510 		sc->rl_cfg0 = RL_CFG0;
1511 		sc->rl_cfg1 = RL_CFG1;
1512 		sc->rl_cfg2 = RL_CFG2;
1513 		sc->rl_cfg3 = RL_CFG3;
1514 		sc->rl_cfg4 = RL_CFG4;
1515 		sc->rl_cfg5 = RL_CFG5;
1516 	}
1517 
1518 	/* Reset the adapter. */
1519 	RL_LOCK(sc);
1520 	re_reset(sc);
1521 	RL_UNLOCK(sc);
1522 
1523 	/* Enable PME. */
1524 	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1525 	cfg = CSR_READ_1(sc, sc->rl_cfg1);
1526 	cfg |= RL_CFG1_PME;
1527 	CSR_WRITE_1(sc, sc->rl_cfg1, cfg);
1528 	cfg = CSR_READ_1(sc, sc->rl_cfg5);
1529 	cfg &= RL_CFG5_PME_STS;
1530 	CSR_WRITE_1(sc, sc->rl_cfg5, cfg);
1531 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1532 
1533 	if ((sc->rl_flags & RL_FLAG_PAR) != 0) {
1534 		/*
1535 		 * XXX Should have a better way to extract station
1536 		 * address from EEPROM.
1537 		 */
1538 		for (i = 0; i < ETHER_ADDR_LEN; i++)
1539 			eaddr[i] = CSR_READ_1(sc, RL_IDR0 + i);
1540 	} else {
1541 		sc->rl_eewidth = RL_9356_ADDR_LEN;
1542 		re_read_eeprom(sc, (caddr_t)&re_did, 0, 1);
1543 		if (re_did != 0x8129)
1544 			sc->rl_eewidth = RL_9346_ADDR_LEN;
1545 
1546 		/*
1547 		 * Get station address from the EEPROM.
1548 		 */
1549 		re_read_eeprom(sc, (caddr_t)as, RL_EE_EADDR, 3);
1550 		for (i = 0; i < ETHER_ADDR_LEN / 2; i++)
1551 			as[i] = le16toh(as[i]);
1552 		bcopy(as, eaddr, ETHER_ADDR_LEN);
1553 	}
1554 
1555 	if (sc->rl_type == RL_8169) {
1556 		/* Set RX length mask and number of descriptors. */
1557 		sc->rl_rxlenmask = RL_RDESC_STAT_GFRAGLEN;
1558 		sc->rl_txstart = RL_GTXSTART;
1559 		sc->rl_ldata.rl_tx_desc_cnt = RL_8169_TX_DESC_CNT;
1560 		sc->rl_ldata.rl_rx_desc_cnt = RL_8169_RX_DESC_CNT;
1561 	} else {
1562 		/* Set RX length mask and number of descriptors. */
1563 		sc->rl_rxlenmask = RL_RDESC_STAT_FRAGLEN;
1564 		sc->rl_txstart = RL_TXSTART;
1565 		sc->rl_ldata.rl_tx_desc_cnt = RL_8139_TX_DESC_CNT;
1566 		sc->rl_ldata.rl_rx_desc_cnt = RL_8139_RX_DESC_CNT;
1567 	}
1568 
1569 	error = re_allocmem(dev, sc);
1570 	if (error)
1571 		goto fail;
1572 	re_add_sysctls(sc);
1573 
1574 	ifp = sc->rl_ifp = if_alloc(IFT_ETHER);
1575 	if (ifp == NULL) {
1576 		device_printf(dev, "can not if_alloc()\n");
1577 		error = ENOSPC;
1578 		goto fail;
1579 	}
1580 
1581 	/* Take controller out of deep sleep mode. */
1582 	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
1583 		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
1584 			CSR_WRITE_1(sc, RL_GPIO,
1585 			    CSR_READ_1(sc, RL_GPIO) | 0x01);
1586 		else
1587 			CSR_WRITE_1(sc, RL_GPIO,
1588 			    CSR_READ_1(sc, RL_GPIO) & ~0x01);
1589 	}
1590 
1591 	/* Take PHY out of power down mode. */
1592 	if ((sc->rl_flags & RL_FLAG_PHYWAKE_PM) != 0) {
1593 		CSR_WRITE_1(sc, RL_PMCH, CSR_READ_1(sc, RL_PMCH) | 0x80);
1594 		if (hw_rev->rl_rev == RL_HWREV_8401E)
1595 			CSR_WRITE_1(sc, 0xD1, CSR_READ_1(sc, 0xD1) & ~0x08);
1596 	}
1597 	if ((sc->rl_flags & RL_FLAG_PHYWAKE) != 0) {
1598 		re_gmii_writereg(dev, 1, 0x1f, 0);
1599 		re_gmii_writereg(dev, 1, 0x0e, 0);
1600 	}
1601 
1602 	ifp->if_softc = sc;
1603 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1604 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1605 	ifp->if_ioctl = re_ioctl;
1606 	ifp->if_start = re_start;
1607 	/*
1608 	 * RTL8168/8111C generates wrong IP checksummed frame if the
1609 	 * packet has IP options so disable TX IP checksum offloading.
1610 	 */
1611 	if (sc->rl_hwrev->rl_rev == RL_HWREV_8168C ||
1612 	    sc->rl_hwrev->rl_rev == RL_HWREV_8168C_SPIN2 ||
1613 	    sc->rl_hwrev->rl_rev == RL_HWREV_8168CP)
1614 		ifp->if_hwassist = CSUM_TCP | CSUM_UDP;
1615 	else
1616 		ifp->if_hwassist = CSUM_IP | CSUM_TCP | CSUM_UDP;
1617 	ifp->if_hwassist |= CSUM_TSO;
1618 	ifp->if_capabilities = IFCAP_HWCSUM | IFCAP_TSO4;
1619 	ifp->if_capenable = ifp->if_capabilities;
1620 	ifp->if_init = re_init;
1621 	IFQ_SET_MAXLEN(&ifp->if_snd, RL_IFQ_MAXLEN);
1622 	ifp->if_snd.ifq_drv_maxlen = RL_IFQ_MAXLEN;
1623 	IFQ_SET_READY(&ifp->if_snd);
1624 
1625 	TASK_INIT(&sc->rl_inttask, 0, re_int_task, sc);
1626 
1627 #define	RE_PHYAD_INTERNAL	 0
1628 
1629 	/* Do MII setup. */
1630 	phy = RE_PHYAD_INTERNAL;
1631 	if (sc->rl_type == RL_8169)
1632 		phy = 1;
1633 	error = mii_attach(dev, &sc->rl_miibus, ifp, re_ifmedia_upd,
1634 	    re_ifmedia_sts, BMSR_DEFCAPMASK, phy, MII_OFFSET_ANY, MIIF_DOPAUSE);
1635 	if (error != 0) {
1636 		device_printf(dev, "attaching PHYs failed\n");
1637 		goto fail;
1638 	}
1639 
1640 	/*
1641 	 * Call MI attach routine.
1642 	 */
1643 	ether_ifattach(ifp, eaddr);
1644 
1645 	/* VLAN capability setup */
1646 	ifp->if_capabilities |= IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING;
1647 	if (ifp->if_capabilities & IFCAP_HWCSUM)
1648 		ifp->if_capabilities |= IFCAP_VLAN_HWCSUM;
1649 	/* Enable WOL if PM is supported. */
1650 	if (pci_find_cap(sc->rl_dev, PCIY_PMG, &reg) == 0)
1651 		ifp->if_capabilities |= IFCAP_WOL;
1652 	ifp->if_capenable = ifp->if_capabilities;
1653 	ifp->if_capenable &= ~(IFCAP_WOL_UCAST | IFCAP_WOL_MCAST);
1654 	/*
1655 	 * Don't enable TSO by default.  It is known to generate
1656 	 * corrupted TCP segments(bad TCP options) under certain
1657 	 * circumstances.
1658 	 */
1659 	ifp->if_hwassist &= ~CSUM_TSO;
1660 	ifp->if_capenable &= ~(IFCAP_TSO4 | IFCAP_VLAN_HWTSO);
1661 #ifdef DEVICE_POLLING
1662 	ifp->if_capabilities |= IFCAP_POLLING;
1663 #endif
1664 	/*
1665 	 * Tell the upper layer(s) we support long frames.
1666 	 * Must appear after the call to ether_ifattach() because
1667 	 * ether_ifattach() sets ifi_hdrlen to the default value.
1668 	 */
1669 	ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
1670 
1671 #ifdef DEV_NETMAP
1672 	re_netmap_attach(sc);
1673 #endif /* DEV_NETMAP */
1674 #ifdef RE_DIAG
1675 	/*
1676 	 * Perform hardware diagnostic on the original RTL8169.
1677 	 * Some 32-bit cards were incorrectly wired and would
1678 	 * malfunction if plugged into a 64-bit slot.
1679 	 */
1680 
1681 	if (hwrev == RL_HWREV_8169) {
1682 		error = re_diag(sc);
1683 		if (error) {
1684 			device_printf(dev,
1685 		    	"attach aborted due to hardware diag failure\n");
1686 			ether_ifdetach(ifp);
1687 			goto fail;
1688 		}
1689 	}
1690 #endif
1691 
1692 #ifdef RE_TX_MODERATION
1693 	intr_filter = 1;
1694 #endif
1695 	/* Hook interrupt last to avoid having to lock softc */
1696 	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) != 0 &&
1697 	    intr_filter == 0) {
1698 		error = bus_setup_intr(dev, sc->rl_irq[0],
1699 		    INTR_TYPE_NET | INTR_MPSAFE, NULL, re_intr_msi, sc,
1700 		    &sc->rl_intrhand[0]);
1701 	} else {
1702 		error = bus_setup_intr(dev, sc->rl_irq[0],
1703 		    INTR_TYPE_NET | INTR_MPSAFE, re_intr, NULL, sc,
1704 		    &sc->rl_intrhand[0]);
1705 	}
1706 	if (error) {
1707 		device_printf(dev, "couldn't set up irq\n");
1708 		ether_ifdetach(ifp);
1709 	}
1710 
1711 fail:
1712 
1713 	if (error)
1714 		re_detach(dev);
1715 
1716 	return (error);
1717 }
1718 
1719 /*
1720  * Shutdown hardware and free up resources. This can be called any
1721  * time after the mutex has been initialized. It is called in both
1722  * the error case in attach and the normal detach case so it needs
1723  * to be careful about only freeing resources that have actually been
1724  * allocated.
1725  */
1726 static int
1727 re_detach(device_t dev)
1728 {
1729 	struct rl_softc		*sc;
1730 	struct ifnet		*ifp;
1731 	int			i, rid;
1732 
1733 	sc = device_get_softc(dev);
1734 	ifp = sc->rl_ifp;
1735 	KASSERT(mtx_initialized(&sc->rl_mtx), ("re mutex not initialized"));
1736 
1737 	/* These should only be active if attach succeeded */
1738 	if (device_is_attached(dev)) {
1739 #ifdef DEVICE_POLLING
1740 		if (ifp->if_capenable & IFCAP_POLLING)
1741 			ether_poll_deregister(ifp);
1742 #endif
1743 		RL_LOCK(sc);
1744 #if 0
1745 		sc->suspended = 1;
1746 #endif
1747 		re_stop(sc);
1748 		RL_UNLOCK(sc);
1749 		callout_drain(&sc->rl_stat_callout);
1750 		taskqueue_drain(taskqueue_fast, &sc->rl_inttask);
1751 		/*
1752 		 * Force off the IFF_UP flag here, in case someone
1753 		 * still had a BPF descriptor attached to this
1754 		 * interface. If they do, ether_ifdetach() will cause
1755 		 * the BPF code to try and clear the promisc mode
1756 		 * flag, which will bubble down to re_ioctl(),
1757 		 * which will try to call re_init() again. This will
1758 		 * turn the NIC back on and restart the MII ticker,
1759 		 * which will panic the system when the kernel tries
1760 		 * to invoke the re_tick() function that isn't there
1761 		 * anymore.
1762 		 */
1763 		ifp->if_flags &= ~IFF_UP;
1764 		ether_ifdetach(ifp);
1765 	}
1766 	if (sc->rl_miibus)
1767 		device_delete_child(dev, sc->rl_miibus);
1768 	bus_generic_detach(dev);
1769 
1770 	/*
1771 	 * The rest is resource deallocation, so we should already be
1772 	 * stopped here.
1773 	 */
1774 
1775 	if (sc->rl_intrhand[0] != NULL) {
1776 		bus_teardown_intr(dev, sc->rl_irq[0], sc->rl_intrhand[0]);
1777 		sc->rl_intrhand[0] = NULL;
1778 	}
1779 	if (ifp != NULL) {
1780 #ifdef DEV_NETMAP
1781 		netmap_detach(ifp);
1782 #endif /* DEV_NETMAP */
1783 		if_free(ifp);
1784 	}
1785 	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) == 0)
1786 		rid = 0;
1787 	else
1788 		rid = 1;
1789 	if (sc->rl_irq[0] != NULL) {
1790 		bus_release_resource(dev, SYS_RES_IRQ, rid, sc->rl_irq[0]);
1791 		sc->rl_irq[0] = NULL;
1792 	}
1793 	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) != 0)
1794 		pci_release_msi(dev);
1795 	if (sc->rl_res_pba) {
1796 		rid = PCIR_BAR(4);
1797 		bus_release_resource(dev, SYS_RES_MEMORY, rid, sc->rl_res_pba);
1798 	}
1799 	if (sc->rl_res)
1800 		bus_release_resource(dev, sc->rl_res_type, sc->rl_res_id,
1801 		    sc->rl_res);
1802 
1803 	/* Unload and free the RX DMA ring memory and map */
1804 
1805 	if (sc->rl_ldata.rl_rx_list_tag) {
1806 		if (sc->rl_ldata.rl_rx_list_map)
1807 			bus_dmamap_unload(sc->rl_ldata.rl_rx_list_tag,
1808 			    sc->rl_ldata.rl_rx_list_map);
1809 		if (sc->rl_ldata.rl_rx_list_map && sc->rl_ldata.rl_rx_list)
1810 			bus_dmamem_free(sc->rl_ldata.rl_rx_list_tag,
1811 			    sc->rl_ldata.rl_rx_list,
1812 			    sc->rl_ldata.rl_rx_list_map);
1813 		bus_dma_tag_destroy(sc->rl_ldata.rl_rx_list_tag);
1814 	}
1815 
1816 	/* Unload and free the TX DMA ring memory and map */
1817 
1818 	if (sc->rl_ldata.rl_tx_list_tag) {
1819 		if (sc->rl_ldata.rl_tx_list_map)
1820 			bus_dmamap_unload(sc->rl_ldata.rl_tx_list_tag,
1821 			    sc->rl_ldata.rl_tx_list_map);
1822 		if (sc->rl_ldata.rl_tx_list_map && sc->rl_ldata.rl_tx_list)
1823 			bus_dmamem_free(sc->rl_ldata.rl_tx_list_tag,
1824 			    sc->rl_ldata.rl_tx_list,
1825 			    sc->rl_ldata.rl_tx_list_map);
1826 		bus_dma_tag_destroy(sc->rl_ldata.rl_tx_list_tag);
1827 	}
1828 
1829 	/* Destroy all the RX and TX buffer maps */
1830 
1831 	if (sc->rl_ldata.rl_tx_mtag) {
1832 		for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
1833 			if (sc->rl_ldata.rl_tx_desc[i].tx_dmamap)
1834 				bus_dmamap_destroy(sc->rl_ldata.rl_tx_mtag,
1835 				    sc->rl_ldata.rl_tx_desc[i].tx_dmamap);
1836 		}
1837 		bus_dma_tag_destroy(sc->rl_ldata.rl_tx_mtag);
1838 	}
1839 	if (sc->rl_ldata.rl_rx_mtag) {
1840 		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1841 			if (sc->rl_ldata.rl_rx_desc[i].rx_dmamap)
1842 				bus_dmamap_destroy(sc->rl_ldata.rl_rx_mtag,
1843 				    sc->rl_ldata.rl_rx_desc[i].rx_dmamap);
1844 		}
1845 		if (sc->rl_ldata.rl_rx_sparemap)
1846 			bus_dmamap_destroy(sc->rl_ldata.rl_rx_mtag,
1847 			    sc->rl_ldata.rl_rx_sparemap);
1848 		bus_dma_tag_destroy(sc->rl_ldata.rl_rx_mtag);
1849 	}
1850 	if (sc->rl_ldata.rl_jrx_mtag) {
1851 		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1852 			if (sc->rl_ldata.rl_jrx_desc[i].rx_dmamap)
1853 				bus_dmamap_destroy(sc->rl_ldata.rl_jrx_mtag,
1854 				    sc->rl_ldata.rl_jrx_desc[i].rx_dmamap);
1855 		}
1856 		if (sc->rl_ldata.rl_jrx_sparemap)
1857 			bus_dmamap_destroy(sc->rl_ldata.rl_jrx_mtag,
1858 			    sc->rl_ldata.rl_jrx_sparemap);
1859 		bus_dma_tag_destroy(sc->rl_ldata.rl_jrx_mtag);
1860 	}
1861 	/* Unload and free the stats buffer and map */
1862 
1863 	if (sc->rl_ldata.rl_stag) {
1864 		if (sc->rl_ldata.rl_smap)
1865 			bus_dmamap_unload(sc->rl_ldata.rl_stag,
1866 			    sc->rl_ldata.rl_smap);
1867 		if (sc->rl_ldata.rl_smap && sc->rl_ldata.rl_stats)
1868 			bus_dmamem_free(sc->rl_ldata.rl_stag,
1869 			    sc->rl_ldata.rl_stats, sc->rl_ldata.rl_smap);
1870 		bus_dma_tag_destroy(sc->rl_ldata.rl_stag);
1871 	}
1872 
1873 	if (sc->rl_parent_tag)
1874 		bus_dma_tag_destroy(sc->rl_parent_tag);
1875 
1876 	mtx_destroy(&sc->rl_mtx);
1877 
1878 	return (0);
1879 }
1880 
1881 static __inline void
1882 re_discard_rxbuf(struct rl_softc *sc, int idx)
1883 {
1884 	struct rl_desc		*desc;
1885 	struct rl_rxdesc	*rxd;
1886 	uint32_t		cmdstat;
1887 
1888 	if (sc->rl_ifp->if_mtu > RL_MTU &&
1889 	    (sc->rl_flags & RL_FLAG_JUMBOV2) != 0)
1890 		rxd = &sc->rl_ldata.rl_jrx_desc[idx];
1891 	else
1892 		rxd = &sc->rl_ldata.rl_rx_desc[idx];
1893 	desc = &sc->rl_ldata.rl_rx_list[idx];
1894 	desc->rl_vlanctl = 0;
1895 	cmdstat = rxd->rx_size;
1896 	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
1897 		cmdstat |= RL_RDESC_CMD_EOR;
1898 	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
1899 }
1900 
1901 static int
1902 re_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_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
1913 	if (m == NULL)
1914 		return (ENOBUFS);
1915 
1916 	m->m_len = m->m_pkthdr.len = MCLBYTES;
1917 #ifdef RE_FIXUP_RX
1918 	/*
1919 	 * This is part of an evil trick to deal with non-x86 platforms.
1920 	 * The RealTek chip requires RX buffers to be aligned on 64-bit
1921 	 * boundaries, but that will hose non-x86 machines. To get around
1922 	 * this, we leave some empty space at the start of each buffer
1923 	 * and for non-x86 hosts, we copy the buffer back six bytes
1924 	 * to achieve word alignment. This is slightly more efficient
1925 	 * than allocating a new buffer, copying the contents, and
1926 	 * discarding the old buffer.
1927 	 */
1928 	m_adj(m, RE_ETHER_ALIGN);
1929 #endif
1930 	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_rx_mtag,
1931 	    sc->rl_ldata.rl_rx_sparemap, m, segs, &nsegs, BUS_DMA_NOWAIT);
1932 	if (error != 0) {
1933 		m_freem(m);
1934 		return (ENOBUFS);
1935 	}
1936 	KASSERT(nsegs == 1, ("%s: %d segment returned!", __func__, nsegs));
1937 
1938 	rxd = &sc->rl_ldata.rl_rx_desc[idx];
1939 	if (rxd->rx_m != NULL) {
1940 		bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap,
1941 		    BUS_DMASYNC_POSTREAD);
1942 		bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap);
1943 	}
1944 
1945 	rxd->rx_m = m;
1946 	map = rxd->rx_dmamap;
1947 	rxd->rx_dmamap = sc->rl_ldata.rl_rx_sparemap;
1948 	rxd->rx_size = segs[0].ds_len;
1949 	sc->rl_ldata.rl_rx_sparemap = map;
1950 	bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap,
1951 	    BUS_DMASYNC_PREREAD);
1952 
1953 	desc = &sc->rl_ldata.rl_rx_list[idx];
1954 	desc->rl_vlanctl = 0;
1955 	desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[0].ds_addr));
1956 	desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[0].ds_addr));
1957 	cmdstat = segs[0].ds_len;
1958 	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
1959 		cmdstat |= RL_RDESC_CMD_EOR;
1960 	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
1961 
1962 	return (0);
1963 }
1964 
1965 static int
1966 re_jumbo_newbuf(struct rl_softc *sc, int idx)
1967 {
1968 	struct mbuf		*m;
1969 	struct rl_rxdesc	*rxd;
1970 	bus_dma_segment_t	segs[1];
1971 	bus_dmamap_t		map;
1972 	struct rl_desc		*desc;
1973 	uint32_t		cmdstat;
1974 	int			error, nsegs;
1975 
1976 	m = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, MJUM9BYTES);
1977 	if (m == NULL)
1978 		return (ENOBUFS);
1979 	m->m_len = m->m_pkthdr.len = MJUM9BYTES;
1980 #ifdef RE_FIXUP_RX
1981 	m_adj(m, RE_ETHER_ALIGN);
1982 #endif
1983 	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_jrx_mtag,
1984 	    sc->rl_ldata.rl_jrx_sparemap, m, segs, &nsegs, BUS_DMA_NOWAIT);
1985 	if (error != 0) {
1986 		m_freem(m);
1987 		return (ENOBUFS);
1988 	}
1989 	KASSERT(nsegs == 1, ("%s: %d segment returned!", __func__, nsegs));
1990 
1991 	rxd = &sc->rl_ldata.rl_jrx_desc[idx];
1992 	if (rxd->rx_m != NULL) {
1993 		bus_dmamap_sync(sc->rl_ldata.rl_jrx_mtag, rxd->rx_dmamap,
1994 		    BUS_DMASYNC_POSTREAD);
1995 		bus_dmamap_unload(sc->rl_ldata.rl_jrx_mtag, rxd->rx_dmamap);
1996 	}
1997 
1998 	rxd->rx_m = m;
1999 	map = rxd->rx_dmamap;
2000 	rxd->rx_dmamap = sc->rl_ldata.rl_jrx_sparemap;
2001 	rxd->rx_size = segs[0].ds_len;
2002 	sc->rl_ldata.rl_jrx_sparemap = map;
2003 	bus_dmamap_sync(sc->rl_ldata.rl_jrx_mtag, rxd->rx_dmamap,
2004 	    BUS_DMASYNC_PREREAD);
2005 
2006 	desc = &sc->rl_ldata.rl_rx_list[idx];
2007 	desc->rl_vlanctl = 0;
2008 	desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[0].ds_addr));
2009 	desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[0].ds_addr));
2010 	cmdstat = segs[0].ds_len;
2011 	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
2012 		cmdstat |= RL_RDESC_CMD_EOR;
2013 	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
2014 
2015 	return (0);
2016 }
2017 
2018 #ifdef RE_FIXUP_RX
2019 static __inline void
2020 re_fixup_rx(struct mbuf *m)
2021 {
2022 	int                     i;
2023 	uint16_t                *src, *dst;
2024 
2025 	src = mtod(m, uint16_t *);
2026 	dst = src - (RE_ETHER_ALIGN - ETHER_ALIGN) / sizeof *src;
2027 
2028 	for (i = 0; i < (m->m_len / sizeof(uint16_t) + 1); i++)
2029 		*dst++ = *src++;
2030 
2031 	m->m_data -= RE_ETHER_ALIGN - ETHER_ALIGN;
2032 }
2033 #endif
2034 
2035 static int
2036 re_tx_list_init(struct rl_softc *sc)
2037 {
2038 	struct rl_desc		*desc;
2039 	int			i;
2040 
2041 	RL_LOCK_ASSERT(sc);
2042 
2043 	bzero(sc->rl_ldata.rl_tx_list,
2044 	    sc->rl_ldata.rl_tx_desc_cnt * sizeof(struct rl_desc));
2045 	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++)
2046 		sc->rl_ldata.rl_tx_desc[i].tx_m = NULL;
2047 #ifdef DEV_NETMAP
2048 	re_netmap_tx_init(sc);
2049 #endif /* DEV_NETMAP */
2050 	/* Set EOR. */
2051 	desc = &sc->rl_ldata.rl_tx_list[sc->rl_ldata.rl_tx_desc_cnt - 1];
2052 	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_EOR);
2053 
2054 	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2055 	    sc->rl_ldata.rl_tx_list_map,
2056 	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
2057 
2058 	sc->rl_ldata.rl_tx_prodidx = 0;
2059 	sc->rl_ldata.rl_tx_considx = 0;
2060 	sc->rl_ldata.rl_tx_free = sc->rl_ldata.rl_tx_desc_cnt;
2061 
2062 	return (0);
2063 }
2064 
2065 static int
2066 re_rx_list_init(struct rl_softc *sc)
2067 {
2068 	int			error, i;
2069 
2070 	bzero(sc->rl_ldata.rl_rx_list,
2071 	    sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc));
2072 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
2073 		sc->rl_ldata.rl_rx_desc[i].rx_m = NULL;
2074 		if ((error = re_newbuf(sc, i)) != 0)
2075 			return (error);
2076 	}
2077 #ifdef DEV_NETMAP
2078 	re_netmap_rx_init(sc);
2079 #endif /* DEV_NETMAP */
2080 
2081 	/* Flush the RX descriptors */
2082 
2083 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
2084 	    sc->rl_ldata.rl_rx_list_map,
2085 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2086 
2087 	sc->rl_ldata.rl_rx_prodidx = 0;
2088 	sc->rl_head = sc->rl_tail = NULL;
2089 	sc->rl_int_rx_act = 0;
2090 
2091 	return (0);
2092 }
2093 
2094 static int
2095 re_jrx_list_init(struct rl_softc *sc)
2096 {
2097 	int			error, i;
2098 
2099 	bzero(sc->rl_ldata.rl_rx_list,
2100 	    sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc));
2101 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
2102 		sc->rl_ldata.rl_jrx_desc[i].rx_m = NULL;
2103 		if ((error = re_jumbo_newbuf(sc, i)) != 0)
2104 			return (error);
2105 	}
2106 
2107 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
2108 	    sc->rl_ldata.rl_rx_list_map,
2109 	    BUS_DMASYNC_PREWRITE | BUS_DMASYNC_PREREAD);
2110 
2111 	sc->rl_ldata.rl_rx_prodidx = 0;
2112 	sc->rl_head = sc->rl_tail = NULL;
2113 	sc->rl_int_rx_act = 0;
2114 
2115 	return (0);
2116 }
2117 
2118 /*
2119  * RX handler for C+ and 8169. For the gigE chips, we support
2120  * the reception of jumbo frames that have been fragmented
2121  * across multiple 2K mbuf cluster buffers.
2122  */
2123 static int
2124 re_rxeof(struct rl_softc *sc, int *rx_npktsp)
2125 {
2126 	struct mbuf		*m;
2127 	struct ifnet		*ifp;
2128 	int			i, rxerr, total_len;
2129 	struct rl_desc		*cur_rx;
2130 	u_int32_t		rxstat, rxvlan;
2131 	int			jumbo, maxpkt = 16, rx_npkts = 0;
2132 
2133 	RL_LOCK_ASSERT(sc);
2134 
2135 	ifp = sc->rl_ifp;
2136 #ifdef DEV_NETMAP
2137 	if (netmap_rx_irq(ifp, 0, &rx_npkts))
2138 		return 0;
2139 #endif /* DEV_NETMAP */
2140 	if (ifp->if_mtu > RL_MTU && (sc->rl_flags & RL_FLAG_JUMBOV2) != 0)
2141 		jumbo = 1;
2142 	else
2143 		jumbo = 0;
2144 
2145 	/* Invalidate the descriptor memory */
2146 
2147 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
2148 	    sc->rl_ldata.rl_rx_list_map,
2149 	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2150 
2151 	for (i = sc->rl_ldata.rl_rx_prodidx; maxpkt > 0;
2152 	    i = RL_RX_DESC_NXT(sc, i)) {
2153 		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
2154 			break;
2155 		cur_rx = &sc->rl_ldata.rl_rx_list[i];
2156 		rxstat = le32toh(cur_rx->rl_cmdstat);
2157 		if ((rxstat & RL_RDESC_STAT_OWN) != 0)
2158 			break;
2159 		total_len = rxstat & sc->rl_rxlenmask;
2160 		rxvlan = le32toh(cur_rx->rl_vlanctl);
2161 		if (jumbo != 0)
2162 			m = sc->rl_ldata.rl_jrx_desc[i].rx_m;
2163 		else
2164 			m = sc->rl_ldata.rl_rx_desc[i].rx_m;
2165 
2166 		if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0 &&
2167 		    (rxstat & (RL_RDESC_STAT_SOF | RL_RDESC_STAT_EOF)) !=
2168 		    (RL_RDESC_STAT_SOF | RL_RDESC_STAT_EOF)) {
2169 			/*
2170 			 * RTL8168C or later controllers do not
2171 			 * support multi-fragment packet.
2172 			 */
2173 			re_discard_rxbuf(sc, i);
2174 			continue;
2175 		} else if ((rxstat & RL_RDESC_STAT_EOF) == 0) {
2176 			if (re_newbuf(sc, i) != 0) {
2177 				/*
2178 				 * If this is part of a multi-fragment packet,
2179 				 * discard all the pieces.
2180 				 */
2181 				if (sc->rl_head != NULL) {
2182 					m_freem(sc->rl_head);
2183 					sc->rl_head = sc->rl_tail = NULL;
2184 				}
2185 				re_discard_rxbuf(sc, i);
2186 				continue;
2187 			}
2188 			m->m_len = RE_RX_DESC_BUFLEN;
2189 			if (sc->rl_head == NULL)
2190 				sc->rl_head = sc->rl_tail = m;
2191 			else {
2192 				m->m_flags &= ~M_PKTHDR;
2193 				sc->rl_tail->m_next = m;
2194 				sc->rl_tail = m;
2195 			}
2196 			continue;
2197 		}
2198 
2199 		/*
2200 		 * NOTE: for the 8139C+, the frame length field
2201 		 * is always 12 bits in size, but for the gigE chips,
2202 		 * it is 13 bits (since the max RX frame length is 16K).
2203 		 * Unfortunately, all 32 bits in the status word
2204 		 * were already used, so to make room for the extra
2205 		 * length bit, RealTek took out the 'frame alignment
2206 		 * error' bit and shifted the other status bits
2207 		 * over one slot. The OWN, EOR, FS and LS bits are
2208 		 * still in the same places. We have already extracted
2209 		 * the frame length and checked the OWN bit, so rather
2210 		 * than using an alternate bit mapping, we shift the
2211 		 * status bits one space to the right so we can evaluate
2212 		 * them using the 8169 status as though it was in the
2213 		 * same format as that of the 8139C+.
2214 		 */
2215 		if (sc->rl_type == RL_8169)
2216 			rxstat >>= 1;
2217 
2218 		/*
2219 		 * if total_len > 2^13-1, both _RXERRSUM and _GIANT will be
2220 		 * set, but if CRC is clear, it will still be a valid frame.
2221 		 */
2222 		if ((rxstat & RL_RDESC_STAT_RXERRSUM) != 0) {
2223 			rxerr = 1;
2224 			if ((sc->rl_flags & RL_FLAG_JUMBOV2) == 0 &&
2225 			    total_len > 8191 &&
2226 			    (rxstat & RL_RDESC_STAT_ERRS) == RL_RDESC_STAT_GIANT)
2227 				rxerr = 0;
2228 			if (rxerr != 0) {
2229 				ifp->if_ierrors++;
2230 				/*
2231 				 * If this is part of a multi-fragment packet,
2232 				 * discard all the pieces.
2233 				 */
2234 				if (sc->rl_head != NULL) {
2235 					m_freem(sc->rl_head);
2236 					sc->rl_head = sc->rl_tail = NULL;
2237 				}
2238 				re_discard_rxbuf(sc, i);
2239 				continue;
2240 			}
2241 		}
2242 
2243 		/*
2244 		 * If allocating a replacement mbuf fails,
2245 		 * reload the current one.
2246 		 */
2247 		if (jumbo != 0)
2248 			rxerr = re_jumbo_newbuf(sc, i);
2249 		else
2250 			rxerr = re_newbuf(sc, i);
2251 		if (rxerr != 0) {
2252 			ifp->if_iqdrops++;
2253 			if (sc->rl_head != NULL) {
2254 				m_freem(sc->rl_head);
2255 				sc->rl_head = sc->rl_tail = NULL;
2256 			}
2257 			re_discard_rxbuf(sc, i);
2258 			continue;
2259 		}
2260 
2261 		if (sc->rl_head != NULL) {
2262 			if (jumbo != 0)
2263 				m->m_len = total_len;
2264 			else {
2265 				m->m_len = total_len % RE_RX_DESC_BUFLEN;
2266 				if (m->m_len == 0)
2267 					m->m_len = RE_RX_DESC_BUFLEN;
2268 			}
2269 			/*
2270 			 * Special case: if there's 4 bytes or less
2271 			 * in this buffer, the mbuf can be discarded:
2272 			 * the last 4 bytes is the CRC, which we don't
2273 			 * care about anyway.
2274 			 */
2275 			if (m->m_len <= ETHER_CRC_LEN) {
2276 				sc->rl_tail->m_len -=
2277 				    (ETHER_CRC_LEN - m->m_len);
2278 				m_freem(m);
2279 			} else {
2280 				m->m_len -= ETHER_CRC_LEN;
2281 				m->m_flags &= ~M_PKTHDR;
2282 				sc->rl_tail->m_next = m;
2283 			}
2284 			m = sc->rl_head;
2285 			sc->rl_head = sc->rl_tail = NULL;
2286 			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
2287 		} else
2288 			m->m_pkthdr.len = m->m_len =
2289 			    (total_len - ETHER_CRC_LEN);
2290 
2291 #ifdef RE_FIXUP_RX
2292 		re_fixup_rx(m);
2293 #endif
2294 		ifp->if_ipackets++;
2295 		m->m_pkthdr.rcvif = ifp;
2296 
2297 		/* Do RX checksumming if enabled */
2298 
2299 		if (ifp->if_capenable & IFCAP_RXCSUM) {
2300 			if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
2301 				/* Check IP header checksum */
2302 				if (rxstat & RL_RDESC_STAT_PROTOID)
2303 					m->m_pkthdr.csum_flags |=
2304 					    CSUM_IP_CHECKED;
2305 				if (!(rxstat & RL_RDESC_STAT_IPSUMBAD))
2306 					m->m_pkthdr.csum_flags |=
2307 					    CSUM_IP_VALID;
2308 
2309 				/* Check TCP/UDP checksum */
2310 				if ((RL_TCPPKT(rxstat) &&
2311 				    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
2312 				    (RL_UDPPKT(rxstat) &&
2313 				     !(rxstat & RL_RDESC_STAT_UDPSUMBAD))) {
2314 					m->m_pkthdr.csum_flags |=
2315 						CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
2316 					m->m_pkthdr.csum_data = 0xffff;
2317 				}
2318 			} else {
2319 				/*
2320 				 * RTL8168C/RTL816CP/RTL8111C/RTL8111CP
2321 				 */
2322 				if ((rxstat & RL_RDESC_STAT_PROTOID) &&
2323 				    (rxvlan & RL_RDESC_IPV4))
2324 					m->m_pkthdr.csum_flags |=
2325 					    CSUM_IP_CHECKED;
2326 				if (!(rxstat & RL_RDESC_STAT_IPSUMBAD) &&
2327 				    (rxvlan & RL_RDESC_IPV4))
2328 					m->m_pkthdr.csum_flags |=
2329 					    CSUM_IP_VALID;
2330 				if (((rxstat & RL_RDESC_STAT_TCP) &&
2331 				    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
2332 				    ((rxstat & RL_RDESC_STAT_UDP) &&
2333 				    !(rxstat & RL_RDESC_STAT_UDPSUMBAD))) {
2334 					m->m_pkthdr.csum_flags |=
2335 						CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
2336 					m->m_pkthdr.csum_data = 0xffff;
2337 				}
2338 			}
2339 		}
2340 		maxpkt--;
2341 		if (rxvlan & RL_RDESC_VLANCTL_TAG) {
2342 			m->m_pkthdr.ether_vtag =
2343 			    bswap16((rxvlan & RL_RDESC_VLANCTL_DATA));
2344 			m->m_flags |= M_VLANTAG;
2345 		}
2346 		RL_UNLOCK(sc);
2347 		(*ifp->if_input)(ifp, m);
2348 		RL_LOCK(sc);
2349 		rx_npkts++;
2350 	}
2351 
2352 	/* Flush the RX DMA ring */
2353 
2354 	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
2355 	    sc->rl_ldata.rl_rx_list_map,
2356 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2357 
2358 	sc->rl_ldata.rl_rx_prodidx = i;
2359 
2360 	if (rx_npktsp != NULL)
2361 		*rx_npktsp = rx_npkts;
2362 	if (maxpkt)
2363 		return (EAGAIN);
2364 
2365 	return (0);
2366 }
2367 
2368 static void
2369 re_txeof(struct rl_softc *sc)
2370 {
2371 	struct ifnet		*ifp;
2372 	struct rl_txdesc	*txd;
2373 	u_int32_t		txstat;
2374 	int			cons;
2375 
2376 	cons = sc->rl_ldata.rl_tx_considx;
2377 	if (cons == sc->rl_ldata.rl_tx_prodidx)
2378 		return;
2379 
2380 	ifp = sc->rl_ifp;
2381 #ifdef DEV_NETMAP
2382 	if (netmap_tx_irq(ifp, 0))
2383 		return;
2384 #endif /* DEV_NETMAP */
2385 	/* Invalidate the TX descriptor list */
2386 	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2387 	    sc->rl_ldata.rl_tx_list_map,
2388 	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2389 
2390 	for (; cons != sc->rl_ldata.rl_tx_prodidx;
2391 	    cons = RL_TX_DESC_NXT(sc, cons)) {
2392 		txstat = le32toh(sc->rl_ldata.rl_tx_list[cons].rl_cmdstat);
2393 		if (txstat & RL_TDESC_STAT_OWN)
2394 			break;
2395 		/*
2396 		 * We only stash mbufs in the last descriptor
2397 		 * in a fragment chain, which also happens to
2398 		 * be the only place where the TX status bits
2399 		 * are valid.
2400 		 */
2401 		if (txstat & RL_TDESC_CMD_EOF) {
2402 			txd = &sc->rl_ldata.rl_tx_desc[cons];
2403 			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
2404 			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
2405 			bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag,
2406 			    txd->tx_dmamap);
2407 			KASSERT(txd->tx_m != NULL,
2408 			    ("%s: freeing NULL mbufs!", __func__));
2409 			m_freem(txd->tx_m);
2410 			txd->tx_m = NULL;
2411 			if (txstat & (RL_TDESC_STAT_EXCESSCOL|
2412 			    RL_TDESC_STAT_COLCNT))
2413 				ifp->if_collisions++;
2414 			if (txstat & RL_TDESC_STAT_TXERRSUM)
2415 				ifp->if_oerrors++;
2416 			else
2417 				ifp->if_opackets++;
2418 		}
2419 		sc->rl_ldata.rl_tx_free++;
2420 		ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
2421 	}
2422 	sc->rl_ldata.rl_tx_considx = cons;
2423 
2424 	/* No changes made to the TX ring, so no flush needed */
2425 
2426 	if (sc->rl_ldata.rl_tx_free != sc->rl_ldata.rl_tx_desc_cnt) {
2427 #ifdef RE_TX_MODERATION
2428 		/*
2429 		 * If not all descriptors have been reaped yet, reload
2430 		 * the timer so that we will eventually get another
2431 		 * interrupt that will cause us to re-enter this routine.
2432 		 * This is done in case the transmitter has gone idle.
2433 		 */
2434 		CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2435 #endif
2436 	} else
2437 		sc->rl_watchdog_timer = 0;
2438 }
2439 
2440 static void
2441 re_tick(void *xsc)
2442 {
2443 	struct rl_softc		*sc;
2444 	struct mii_data		*mii;
2445 
2446 	sc = xsc;
2447 
2448 	RL_LOCK_ASSERT(sc);
2449 
2450 	mii = device_get_softc(sc->rl_miibus);
2451 	mii_tick(mii);
2452 	if ((sc->rl_flags & RL_FLAG_LINK) == 0)
2453 		re_miibus_statchg(sc->rl_dev);
2454 	/*
2455 	 * Reclaim transmitted frames here. Technically it is not
2456 	 * necessary to do here but it ensures periodic reclamation
2457 	 * regardless of Tx completion interrupt which seems to be
2458 	 * lost on PCIe based controllers under certain situations.
2459 	 */
2460 	re_txeof(sc);
2461 	re_watchdog(sc);
2462 	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
2463 }
2464 
2465 #ifdef DEVICE_POLLING
2466 static int
2467 re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
2468 {
2469 	struct rl_softc *sc = ifp->if_softc;
2470 	int rx_npkts = 0;
2471 
2472 	RL_LOCK(sc);
2473 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
2474 		rx_npkts = re_poll_locked(ifp, cmd, count);
2475 	RL_UNLOCK(sc);
2476 	return (rx_npkts);
2477 }
2478 
2479 static int
2480 re_poll_locked(struct ifnet *ifp, enum poll_cmd cmd, int count)
2481 {
2482 	struct rl_softc *sc = ifp->if_softc;
2483 	int rx_npkts;
2484 
2485 	RL_LOCK_ASSERT(sc);
2486 
2487 	sc->rxcycles = count;
2488 	re_rxeof(sc, &rx_npkts);
2489 	re_txeof(sc);
2490 
2491 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2492 		re_start_locked(ifp);
2493 
2494 	if (cmd == POLL_AND_CHECK_STATUS) { /* also check status register */
2495 		u_int16_t       status;
2496 
2497 		status = CSR_READ_2(sc, RL_ISR);
2498 		if (status == 0xffff)
2499 			return (rx_npkts);
2500 		if (status)
2501 			CSR_WRITE_2(sc, RL_ISR, status);
2502 		if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
2503 		    (sc->rl_flags & RL_FLAG_PCIE))
2504 			CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2505 
2506 		/*
2507 		 * XXX check behaviour on receiver stalls.
2508 		 */
2509 
2510 		if (status & RL_ISR_SYSTEM_ERR) {
2511 			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2512 			re_init_locked(sc);
2513 		}
2514 	}
2515 	return (rx_npkts);
2516 }
2517 #endif /* DEVICE_POLLING */
2518 
2519 static int
2520 re_intr(void *arg)
2521 {
2522 	struct rl_softc		*sc;
2523 	uint16_t		status;
2524 
2525 	sc = arg;
2526 
2527 	status = CSR_READ_2(sc, RL_ISR);
2528 	if (status == 0xFFFF || (status & RL_INTRS_CPLUS) == 0)
2529                 return (FILTER_STRAY);
2530 	CSR_WRITE_2(sc, RL_IMR, 0);
2531 
2532 	taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_inttask);
2533 
2534 	return (FILTER_HANDLED);
2535 }
2536 
2537 static void
2538 re_int_task(void *arg, int npending)
2539 {
2540 	struct rl_softc		*sc;
2541 	struct ifnet		*ifp;
2542 	u_int16_t		status;
2543 	int			rval = 0;
2544 
2545 	sc = arg;
2546 	ifp = sc->rl_ifp;
2547 
2548 	RL_LOCK(sc);
2549 
2550 	status = CSR_READ_2(sc, RL_ISR);
2551         CSR_WRITE_2(sc, RL_ISR, status);
2552 
2553 	if (sc->suspended ||
2554 	    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2555 		RL_UNLOCK(sc);
2556 		return;
2557 	}
2558 
2559 #ifdef DEVICE_POLLING
2560 	if  (ifp->if_capenable & IFCAP_POLLING) {
2561 		RL_UNLOCK(sc);
2562 		return;
2563 	}
2564 #endif
2565 
2566 	if (status & (RL_ISR_RX_OK|RL_ISR_RX_ERR|RL_ISR_FIFO_OFLOW))
2567 		rval = re_rxeof(sc, NULL);
2568 
2569 	/*
2570 	 * Some chips will ignore a second TX request issued
2571 	 * while an existing transmission is in progress. If
2572 	 * the transmitter goes idle but there are still
2573 	 * packets waiting to be sent, we need to restart the
2574 	 * channel here to flush them out. This only seems to
2575 	 * be required with the PCIe devices.
2576 	 */
2577 	if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
2578 	    (sc->rl_flags & RL_FLAG_PCIE))
2579 		CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2580 	if (status & (
2581 #ifdef RE_TX_MODERATION
2582 	    RL_ISR_TIMEOUT_EXPIRED|
2583 #else
2584 	    RL_ISR_TX_OK|
2585 #endif
2586 	    RL_ISR_TX_ERR|RL_ISR_TX_DESC_UNAVAIL))
2587 		re_txeof(sc);
2588 
2589 	if (status & RL_ISR_SYSTEM_ERR) {
2590 		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2591 		re_init_locked(sc);
2592 	}
2593 
2594 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2595 		re_start_locked(ifp);
2596 
2597 	RL_UNLOCK(sc);
2598 
2599         if ((CSR_READ_2(sc, RL_ISR) & RL_INTRS_CPLUS) || rval) {
2600 		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_inttask);
2601 		return;
2602 	}
2603 
2604 	CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2605 }
2606 
2607 static void
2608 re_intr_msi(void *xsc)
2609 {
2610 	struct rl_softc		*sc;
2611 	struct ifnet		*ifp;
2612 	uint16_t		intrs, status;
2613 
2614 	sc = xsc;
2615 	RL_LOCK(sc);
2616 
2617 	ifp = sc->rl_ifp;
2618 #ifdef DEVICE_POLLING
2619 	if (ifp->if_capenable & IFCAP_POLLING) {
2620 		RL_UNLOCK(sc);
2621 		return;
2622 	}
2623 #endif
2624 	/* Disable interrupts. */
2625 	CSR_WRITE_2(sc, RL_IMR, 0);
2626 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2627 		RL_UNLOCK(sc);
2628 		return;
2629 	}
2630 
2631 	intrs = RL_INTRS_CPLUS;
2632 	status = CSR_READ_2(sc, RL_ISR);
2633         CSR_WRITE_2(sc, RL_ISR, status);
2634 	if (sc->rl_int_rx_act > 0) {
2635 		intrs &= ~(RL_ISR_RX_OK | RL_ISR_RX_ERR | RL_ISR_FIFO_OFLOW |
2636 		    RL_ISR_RX_OVERRUN);
2637 		status &= ~(RL_ISR_RX_OK | RL_ISR_RX_ERR | RL_ISR_FIFO_OFLOW |
2638 		    RL_ISR_RX_OVERRUN);
2639 	}
2640 
2641 	if (status & (RL_ISR_TIMEOUT_EXPIRED | RL_ISR_RX_OK | RL_ISR_RX_ERR |
2642 	    RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN)) {
2643 		re_rxeof(sc, NULL);
2644 		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
2645 			if (sc->rl_int_rx_mod != 0 &&
2646 			    (status & (RL_ISR_RX_OK | RL_ISR_RX_ERR |
2647 			    RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN)) != 0) {
2648 				/* Rearm one-shot timer. */
2649 				CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2650 				intrs &= ~(RL_ISR_RX_OK | RL_ISR_RX_ERR |
2651 				    RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN);
2652 				sc->rl_int_rx_act = 1;
2653 			} else {
2654 				intrs |= RL_ISR_RX_OK | RL_ISR_RX_ERR |
2655 				    RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN;
2656 				sc->rl_int_rx_act = 0;
2657 			}
2658 		}
2659 	}
2660 
2661 	/*
2662 	 * Some chips will ignore a second TX request issued
2663 	 * while an existing transmission is in progress. If
2664 	 * the transmitter goes idle but there are still
2665 	 * packets waiting to be sent, we need to restart the
2666 	 * channel here to flush them out. This only seems to
2667 	 * be required with the PCIe devices.
2668 	 */
2669 	if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
2670 	    (sc->rl_flags & RL_FLAG_PCIE))
2671 		CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2672 	if (status & (RL_ISR_TX_OK | RL_ISR_TX_ERR | RL_ISR_TX_DESC_UNAVAIL))
2673 		re_txeof(sc);
2674 
2675 	if (status & RL_ISR_SYSTEM_ERR) {
2676 		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2677 		re_init_locked(sc);
2678 	}
2679 
2680 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
2681 		if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2682 			re_start_locked(ifp);
2683 		CSR_WRITE_2(sc, RL_IMR, intrs);
2684 	}
2685 	RL_UNLOCK(sc);
2686 }
2687 
2688 static int
2689 re_encap(struct rl_softc *sc, struct mbuf **m_head)
2690 {
2691 	struct rl_txdesc	*txd, *txd_last;
2692 	bus_dma_segment_t	segs[RL_NTXSEGS];
2693 	bus_dmamap_t		map;
2694 	struct mbuf		*m_new;
2695 	struct rl_desc		*desc;
2696 	int			nsegs, prod;
2697 	int			i, error, ei, si;
2698 	int			padlen;
2699 	uint32_t		cmdstat, csum_flags, vlanctl;
2700 
2701 	RL_LOCK_ASSERT(sc);
2702 	M_ASSERTPKTHDR((*m_head));
2703 
2704 	/*
2705 	 * With some of the RealTek chips, using the checksum offload
2706 	 * support in conjunction with the autopadding feature results
2707 	 * in the transmission of corrupt frames. For example, if we
2708 	 * need to send a really small IP fragment that's less than 60
2709 	 * bytes in size, and IP header checksumming is enabled, the
2710 	 * resulting ethernet frame that appears on the wire will
2711 	 * have garbled payload. To work around this, if TX IP checksum
2712 	 * offload is enabled, we always manually pad short frames out
2713 	 * to the minimum ethernet frame size.
2714 	 */
2715 	if ((sc->rl_flags & RL_FLAG_AUTOPAD) == 0 &&
2716 	    (*m_head)->m_pkthdr.len < RL_IP4CSUMTX_PADLEN &&
2717 	    ((*m_head)->m_pkthdr.csum_flags & CSUM_IP) != 0) {
2718 		padlen = RL_MIN_FRAMELEN - (*m_head)->m_pkthdr.len;
2719 		if (M_WRITABLE(*m_head) == 0) {
2720 			/* Get a writable copy. */
2721 			m_new = m_dup(*m_head, M_NOWAIT);
2722 			m_freem(*m_head);
2723 			if (m_new == NULL) {
2724 				*m_head = NULL;
2725 				return (ENOBUFS);
2726 			}
2727 			*m_head = m_new;
2728 		}
2729 		if ((*m_head)->m_next != NULL ||
2730 		    M_TRAILINGSPACE(*m_head) < padlen) {
2731 			m_new = m_defrag(*m_head, M_NOWAIT);
2732 			if (m_new == NULL) {
2733 				m_freem(*m_head);
2734 				*m_head = NULL;
2735 				return (ENOBUFS);
2736 			}
2737 		} else
2738 			m_new = *m_head;
2739 
2740 		/*
2741 		 * Manually pad short frames, and zero the pad space
2742 		 * to avoid leaking data.
2743 		 */
2744 		bzero(mtod(m_new, char *) + m_new->m_pkthdr.len, padlen);
2745 		m_new->m_pkthdr.len += padlen;
2746 		m_new->m_len = m_new->m_pkthdr.len;
2747 		*m_head = m_new;
2748 	}
2749 
2750 	prod = sc->rl_ldata.rl_tx_prodidx;
2751 	txd = &sc->rl_ldata.rl_tx_desc[prod];
2752 	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap,
2753 	    *m_head, segs, &nsegs, BUS_DMA_NOWAIT);
2754 	if (error == EFBIG) {
2755 		m_new = m_collapse(*m_head, M_NOWAIT, RL_NTXSEGS);
2756 		if (m_new == NULL) {
2757 			m_freem(*m_head);
2758 			*m_head = NULL;
2759 			return (ENOBUFS);
2760 		}
2761 		*m_head = m_new;
2762 		error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_tx_mtag,
2763 		    txd->tx_dmamap, *m_head, segs, &nsegs, BUS_DMA_NOWAIT);
2764 		if (error != 0) {
2765 			m_freem(*m_head);
2766 			*m_head = NULL;
2767 			return (error);
2768 		}
2769 	} else if (error != 0)
2770 		return (error);
2771 	if (nsegs == 0) {
2772 		m_freem(*m_head);
2773 		*m_head = NULL;
2774 		return (EIO);
2775 	}
2776 
2777 	/* Check for number of available descriptors. */
2778 	if (sc->rl_ldata.rl_tx_free - nsegs <= 1) {
2779 		bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap);
2780 		return (ENOBUFS);
2781 	}
2782 
2783 	bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap,
2784 	    BUS_DMASYNC_PREWRITE);
2785 
2786 	/*
2787 	 * Set up checksum offload. Note: checksum offload bits must
2788 	 * appear in all descriptors of a multi-descriptor transmit
2789 	 * attempt. This is according to testing done with an 8169
2790 	 * chip. This is a requirement.
2791 	 */
2792 	vlanctl = 0;
2793 	csum_flags = 0;
2794 	if (((*m_head)->m_pkthdr.csum_flags & CSUM_TSO) != 0) {
2795 		if ((sc->rl_flags & RL_FLAG_DESCV2) != 0) {
2796 			csum_flags |= RL_TDESC_CMD_LGSEND;
2797 			vlanctl |= ((uint32_t)(*m_head)->m_pkthdr.tso_segsz <<
2798 			    RL_TDESC_CMD_MSSVALV2_SHIFT);
2799 		} else {
2800 			csum_flags |= RL_TDESC_CMD_LGSEND |
2801 			    ((uint32_t)(*m_head)->m_pkthdr.tso_segsz <<
2802 			    RL_TDESC_CMD_MSSVAL_SHIFT);
2803 		}
2804 	} else {
2805 		/*
2806 		 * Unconditionally enable IP checksum if TCP or UDP
2807 		 * checksum is required. Otherwise, TCP/UDP checksum
2808 		 * doesn't make effects.
2809 		 */
2810 		if (((*m_head)->m_pkthdr.csum_flags & RE_CSUM_FEATURES) != 0) {
2811 			if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
2812 				csum_flags |= RL_TDESC_CMD_IPCSUM;
2813 				if (((*m_head)->m_pkthdr.csum_flags &
2814 				    CSUM_TCP) != 0)
2815 					csum_flags |= RL_TDESC_CMD_TCPCSUM;
2816 				if (((*m_head)->m_pkthdr.csum_flags &
2817 				    CSUM_UDP) != 0)
2818 					csum_flags |= RL_TDESC_CMD_UDPCSUM;
2819 			} else {
2820 				vlanctl |= RL_TDESC_CMD_IPCSUMV2;
2821 				if (((*m_head)->m_pkthdr.csum_flags &
2822 				    CSUM_TCP) != 0)
2823 					vlanctl |= RL_TDESC_CMD_TCPCSUMV2;
2824 				if (((*m_head)->m_pkthdr.csum_flags &
2825 				    CSUM_UDP) != 0)
2826 					vlanctl |= RL_TDESC_CMD_UDPCSUMV2;
2827 			}
2828 		}
2829 	}
2830 
2831 	/*
2832 	 * Set up hardware VLAN tagging. Note: vlan tag info must
2833 	 * appear in all descriptors of a multi-descriptor
2834 	 * transmission attempt.
2835 	 */
2836 	if ((*m_head)->m_flags & M_VLANTAG)
2837 		vlanctl |= bswap16((*m_head)->m_pkthdr.ether_vtag) |
2838 		    RL_TDESC_VLANCTL_TAG;
2839 
2840 	si = prod;
2841 	for (i = 0; i < nsegs; i++, prod = RL_TX_DESC_NXT(sc, prod)) {
2842 		desc = &sc->rl_ldata.rl_tx_list[prod];
2843 		desc->rl_vlanctl = htole32(vlanctl);
2844 		desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[i].ds_addr));
2845 		desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[i].ds_addr));
2846 		cmdstat = segs[i].ds_len;
2847 		if (i != 0)
2848 			cmdstat |= RL_TDESC_CMD_OWN;
2849 		if (prod == sc->rl_ldata.rl_tx_desc_cnt - 1)
2850 			cmdstat |= RL_TDESC_CMD_EOR;
2851 		desc->rl_cmdstat = htole32(cmdstat | csum_flags);
2852 		sc->rl_ldata.rl_tx_free--;
2853 	}
2854 	/* Update producer index. */
2855 	sc->rl_ldata.rl_tx_prodidx = prod;
2856 
2857 	/* Set EOF on the last descriptor. */
2858 	ei = RL_TX_DESC_PRV(sc, prod);
2859 	desc = &sc->rl_ldata.rl_tx_list[ei];
2860 	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_EOF);
2861 
2862 	desc = &sc->rl_ldata.rl_tx_list[si];
2863 	/* Set SOF and transfer ownership of packet to the chip. */
2864 	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_OWN | RL_TDESC_CMD_SOF);
2865 
2866 	/*
2867 	 * Insure that the map for this transmission
2868 	 * is placed at the array index of the last descriptor
2869 	 * in this chain.  (Swap last and first dmamaps.)
2870 	 */
2871 	txd_last = &sc->rl_ldata.rl_tx_desc[ei];
2872 	map = txd->tx_dmamap;
2873 	txd->tx_dmamap = txd_last->tx_dmamap;
2874 	txd_last->tx_dmamap = map;
2875 	txd_last->tx_m = *m_head;
2876 
2877 	return (0);
2878 }
2879 
2880 static void
2881 re_start(struct ifnet *ifp)
2882 {
2883 	struct rl_softc		*sc;
2884 
2885 	sc = ifp->if_softc;
2886 	RL_LOCK(sc);
2887 	re_start_locked(ifp);
2888 	RL_UNLOCK(sc);
2889 }
2890 
2891 /*
2892  * Main transmit routine for C+ and gigE NICs.
2893  */
2894 static void
2895 re_start_locked(struct ifnet *ifp)
2896 {
2897 	struct rl_softc		*sc;
2898 	struct mbuf		*m_head;
2899 	int			queued;
2900 
2901 	sc = ifp->if_softc;
2902 
2903 #ifdef DEV_NETMAP
2904 	/* XXX is this necessary ? */
2905 	if (ifp->if_capenable & IFCAP_NETMAP) {
2906 		struct netmap_kring *kring = &NA(ifp)->tx_rings[0];
2907 		if (sc->rl_ldata.rl_tx_prodidx != kring->nr_hwcur) {
2908 			/* kick the tx unit */
2909 			CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2910 #ifdef RE_TX_MODERATION
2911 			CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2912 #endif
2913 			sc->rl_watchdog_timer = 5;
2914 		}
2915 		return;
2916 	}
2917 #endif /* DEV_NETMAP */
2918 	if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
2919 	    IFF_DRV_RUNNING || (sc->rl_flags & RL_FLAG_LINK) == 0)
2920 		return;
2921 
2922 	for (queued = 0; !IFQ_DRV_IS_EMPTY(&ifp->if_snd) &&
2923 	    sc->rl_ldata.rl_tx_free > 1;) {
2924 		IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
2925 		if (m_head == NULL)
2926 			break;
2927 
2928 		if (re_encap(sc, &m_head) != 0) {
2929 			if (m_head == NULL)
2930 				break;
2931 			IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
2932 			ifp->if_drv_flags |= IFF_DRV_OACTIVE;
2933 			break;
2934 		}
2935 
2936 		/*
2937 		 * If there's a BPF listener, bounce a copy of this frame
2938 		 * to him.
2939 		 */
2940 		ETHER_BPF_MTAP(ifp, m_head);
2941 
2942 		queued++;
2943 	}
2944 
2945 	if (queued == 0) {
2946 #ifdef RE_TX_MODERATION
2947 		if (sc->rl_ldata.rl_tx_free != sc->rl_ldata.rl_tx_desc_cnt)
2948 			CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2949 #endif
2950 		return;
2951 	}
2952 
2953 	/* Flush the TX descriptors */
2954 
2955 	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2956 	    sc->rl_ldata.rl_tx_list_map,
2957 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2958 
2959 	CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2960 
2961 #ifdef RE_TX_MODERATION
2962 	/*
2963 	 * Use the countdown timer for interrupt moderation.
2964 	 * 'TX done' interrupts are disabled. Instead, we reset the
2965 	 * countdown timer, which will begin counting until it hits
2966 	 * the value in the TIMERINT register, and then trigger an
2967 	 * interrupt. Each time we write to the TIMERCNT register,
2968 	 * the timer count is reset to 0.
2969 	 */
2970 	CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2971 #endif
2972 
2973 	/*
2974 	 * Set a timeout in case the chip goes out to lunch.
2975 	 */
2976 	sc->rl_watchdog_timer = 5;
2977 }
2978 
2979 static void
2980 re_set_jumbo(struct rl_softc *sc, int jumbo)
2981 {
2982 
2983 	if (sc->rl_hwrev->rl_rev == RL_HWREV_8168E_VL) {
2984 		pci_set_max_read_req(sc->rl_dev, 4096);
2985 		return;
2986 	}
2987 
2988 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG);
2989 	if (jumbo != 0) {
2990 		CSR_WRITE_1(sc, sc->rl_cfg3, CSR_READ_1(sc, sc->rl_cfg3) |
2991 		    RL_CFG3_JUMBO_EN0);
2992 		switch (sc->rl_hwrev->rl_rev) {
2993 		case RL_HWREV_8168DP:
2994 			break;
2995 		case RL_HWREV_8168E:
2996 			CSR_WRITE_1(sc, sc->rl_cfg4,
2997 			    CSR_READ_1(sc, sc->rl_cfg4) | 0x01);
2998 			break;
2999 		default:
3000 			CSR_WRITE_1(sc, sc->rl_cfg4,
3001 			    CSR_READ_1(sc, sc->rl_cfg4) | RL_CFG4_JUMBO_EN1);
3002 		}
3003 	} else {
3004 		CSR_WRITE_1(sc, sc->rl_cfg3, CSR_READ_1(sc, sc->rl_cfg3) &
3005 		    ~RL_CFG3_JUMBO_EN0);
3006 		switch (sc->rl_hwrev->rl_rev) {
3007 		case RL_HWREV_8168DP:
3008 			break;
3009 		case RL_HWREV_8168E:
3010 			CSR_WRITE_1(sc, sc->rl_cfg4,
3011 			    CSR_READ_1(sc, sc->rl_cfg4) & ~0x01);
3012 			break;
3013 		default:
3014 			CSR_WRITE_1(sc, sc->rl_cfg4,
3015 			    CSR_READ_1(sc, sc->rl_cfg4) & ~RL_CFG4_JUMBO_EN1);
3016 		}
3017 	}
3018 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3019 
3020 	switch (sc->rl_hwrev->rl_rev) {
3021 	case RL_HWREV_8168DP:
3022 		pci_set_max_read_req(sc->rl_dev, 4096);
3023 		break;
3024 	default:
3025 		if (jumbo != 0)
3026 			pci_set_max_read_req(sc->rl_dev, 512);
3027 		else
3028 			pci_set_max_read_req(sc->rl_dev, 4096);
3029 	}
3030 }
3031 
3032 static void
3033 re_init(void *xsc)
3034 {
3035 	struct rl_softc		*sc = xsc;
3036 
3037 	RL_LOCK(sc);
3038 	re_init_locked(sc);
3039 	RL_UNLOCK(sc);
3040 }
3041 
3042 static void
3043 re_init_locked(struct rl_softc *sc)
3044 {
3045 	struct ifnet		*ifp = sc->rl_ifp;
3046 	struct mii_data		*mii;
3047 	uint32_t		reg;
3048 	uint16_t		cfg;
3049 	union {
3050 		uint32_t align_dummy;
3051 		u_char eaddr[ETHER_ADDR_LEN];
3052         } eaddr;
3053 
3054 	RL_LOCK_ASSERT(sc);
3055 
3056 	mii = device_get_softc(sc->rl_miibus);
3057 
3058 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
3059 		return;
3060 
3061 	/*
3062 	 * Cancel pending I/O and free all RX/TX buffers.
3063 	 */
3064 	re_stop(sc);
3065 
3066 	/* Put controller into known state. */
3067 	re_reset(sc);
3068 
3069 	/*
3070 	 * For C+ mode, initialize the RX descriptors and mbufs.
3071 	 */
3072 	if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
3073 		if (ifp->if_mtu > RL_MTU) {
3074 			if (re_jrx_list_init(sc) != 0) {
3075 				device_printf(sc->rl_dev,
3076 				    "no memory for jumbo RX buffers\n");
3077 				re_stop(sc);
3078 				return;
3079 			}
3080 			/* Disable checksum offloading for jumbo frames. */
3081 			ifp->if_capenable &= ~(IFCAP_HWCSUM | IFCAP_TSO4);
3082 			ifp->if_hwassist &= ~(RE_CSUM_FEATURES | CSUM_TSO);
3083 		} else {
3084 			if (re_rx_list_init(sc) != 0) {
3085 				device_printf(sc->rl_dev,
3086 				    "no memory for RX buffers\n");
3087 				re_stop(sc);
3088 				return;
3089 			}
3090 		}
3091 		re_set_jumbo(sc, ifp->if_mtu > RL_MTU);
3092 	} else {
3093 		if (re_rx_list_init(sc) != 0) {
3094 			device_printf(sc->rl_dev, "no memory for RX buffers\n");
3095 			re_stop(sc);
3096 			return;
3097 		}
3098 		if ((sc->rl_flags & RL_FLAG_PCIE) != 0 &&
3099 		    pci_get_device(sc->rl_dev) != RT_DEVICEID_8101E) {
3100 			if (ifp->if_mtu > RL_MTU)
3101 				pci_set_max_read_req(sc->rl_dev, 512);
3102 			else
3103 				pci_set_max_read_req(sc->rl_dev, 4096);
3104 		}
3105 	}
3106 	re_tx_list_init(sc);
3107 
3108 	/*
3109 	 * Enable C+ RX and TX mode, as well as VLAN stripping and
3110 	 * RX checksum offload. We must configure the C+ register
3111 	 * before all others.
3112 	 */
3113 	cfg = RL_CPLUSCMD_PCI_MRW;
3114 	if ((ifp->if_capenable & IFCAP_RXCSUM) != 0)
3115 		cfg |= RL_CPLUSCMD_RXCSUM_ENB;
3116 	if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0)
3117 		cfg |= RL_CPLUSCMD_VLANSTRIP;
3118 	if ((sc->rl_flags & RL_FLAG_MACSTAT) != 0) {
3119 		cfg |= RL_CPLUSCMD_MACSTAT_DIS;
3120 		/* XXX magic. */
3121 		cfg |= 0x0001;
3122 	} else
3123 		cfg |= RL_CPLUSCMD_RXENB | RL_CPLUSCMD_TXENB;
3124 	CSR_WRITE_2(sc, RL_CPLUS_CMD, cfg);
3125 	if (sc->rl_hwrev->rl_rev == RL_HWREV_8169_8110SC ||
3126 	    sc->rl_hwrev->rl_rev == RL_HWREV_8169_8110SCE) {
3127 		reg = 0x000fff00;
3128 		if ((CSR_READ_1(sc, sc->rl_cfg2) & RL_CFG2_PCI66MHZ) != 0)
3129 			reg |= 0x000000ff;
3130 		if (sc->rl_hwrev->rl_rev == RL_HWREV_8169_8110SCE)
3131 			reg |= 0x00f00000;
3132 		CSR_WRITE_4(sc, 0x7c, reg);
3133 		/* Disable interrupt mitigation. */
3134 		CSR_WRITE_2(sc, 0xe2, 0);
3135 	}
3136 	/*
3137 	 * Disable TSO if interface MTU size is greater than MSS
3138 	 * allowed in controller.
3139 	 */
3140 	if (ifp->if_mtu > RL_TSO_MTU && (ifp->if_capenable & IFCAP_TSO4) != 0) {
3141 		ifp->if_capenable &= ~IFCAP_TSO4;
3142 		ifp->if_hwassist &= ~CSUM_TSO;
3143 	}
3144 
3145 	/*
3146 	 * Init our MAC address.  Even though the chipset
3147 	 * documentation doesn't mention it, we need to enter "Config
3148 	 * register write enable" mode to modify the ID registers.
3149 	 */
3150 	/* Copy MAC address on stack to align. */
3151 	bcopy(IF_LLADDR(ifp), eaddr.eaddr, ETHER_ADDR_LEN);
3152 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG);
3153 	CSR_WRITE_4(sc, RL_IDR0,
3154 	    htole32(*(u_int32_t *)(&eaddr.eaddr[0])));
3155 	CSR_WRITE_4(sc, RL_IDR4,
3156 	    htole32(*(u_int32_t *)(&eaddr.eaddr[4])));
3157 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3158 
3159 	/*
3160 	 * Load the addresses of the RX and TX lists into the chip.
3161 	 */
3162 
3163 	CSR_WRITE_4(sc, RL_RXLIST_ADDR_HI,
3164 	    RL_ADDR_HI(sc->rl_ldata.rl_rx_list_addr));
3165 	CSR_WRITE_4(sc, RL_RXLIST_ADDR_LO,
3166 	    RL_ADDR_LO(sc->rl_ldata.rl_rx_list_addr));
3167 
3168 	CSR_WRITE_4(sc, RL_TXLIST_ADDR_HI,
3169 	    RL_ADDR_HI(sc->rl_ldata.rl_tx_list_addr));
3170 	CSR_WRITE_4(sc, RL_TXLIST_ADDR_LO,
3171 	    RL_ADDR_LO(sc->rl_ldata.rl_tx_list_addr));
3172 
3173 	/*
3174 	 * Enable transmit and receive.
3175 	 */
3176 	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
3177 
3178 	/*
3179 	 * Set the initial TX configuration.
3180 	 */
3181 	if (sc->rl_testmode) {
3182 		if (sc->rl_type == RL_8169)
3183 			CSR_WRITE_4(sc, RL_TXCFG,
3184 			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON);
3185 		else
3186 			CSR_WRITE_4(sc, RL_TXCFG,
3187 			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON_CPLUS);
3188 	} else
3189 		CSR_WRITE_4(sc, RL_TXCFG, RL_TXCFG_CONFIG);
3190 
3191 	CSR_WRITE_1(sc, RL_EARLY_TX_THRESH, 16);
3192 
3193 	/*
3194 	 * Set the initial RX configuration.
3195 	 */
3196 	re_set_rxmode(sc);
3197 
3198 	/* Configure interrupt moderation. */
3199 	if (sc->rl_type == RL_8169) {
3200 		/* Magic from vendor. */
3201 		CSR_WRITE_2(sc, RL_INTRMOD, 0x5100);
3202 	}
3203 
3204 #ifdef DEVICE_POLLING
3205 	/*
3206 	 * Disable interrupts if we are polling.
3207 	 */
3208 	if (ifp->if_capenable & IFCAP_POLLING)
3209 		CSR_WRITE_2(sc, RL_IMR, 0);
3210 	else	/* otherwise ... */
3211 #endif
3212 
3213 	/*
3214 	 * Enable interrupts.
3215 	 */
3216 	if (sc->rl_testmode)
3217 		CSR_WRITE_2(sc, RL_IMR, 0);
3218 	else
3219 		CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
3220 	CSR_WRITE_2(sc, RL_ISR, RL_INTRS_CPLUS);
3221 
3222 	/* Set initial TX threshold */
3223 	sc->rl_txthresh = RL_TX_THRESH_INIT;
3224 
3225 	/* Start RX/TX process. */
3226 	CSR_WRITE_4(sc, RL_MISSEDPKT, 0);
3227 #ifdef notdef
3228 	/* Enable receiver and transmitter. */
3229 	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
3230 #endif
3231 
3232 	/*
3233 	 * Initialize the timer interrupt register so that
3234 	 * a timer interrupt will be generated once the timer
3235 	 * reaches a certain number of ticks. The timer is
3236 	 * reloaded on each transmit.
3237 	 */
3238 #ifdef RE_TX_MODERATION
3239 	/*
3240 	 * Use timer interrupt register to moderate TX interrupt
3241 	 * moderation, which dramatically improves TX frame rate.
3242 	 */
3243 	if (sc->rl_type == RL_8169)
3244 		CSR_WRITE_4(sc, RL_TIMERINT_8169, 0x800);
3245 	else
3246 		CSR_WRITE_4(sc, RL_TIMERINT, 0x400);
3247 #else
3248 	/*
3249 	 * Use timer interrupt register to moderate RX interrupt
3250 	 * moderation.
3251 	 */
3252 	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) != 0 &&
3253 	    intr_filter == 0) {
3254 		if (sc->rl_type == RL_8169)
3255 			CSR_WRITE_4(sc, RL_TIMERINT_8169,
3256 			    RL_USECS(sc->rl_int_rx_mod));
3257 	} else {
3258 		if (sc->rl_type == RL_8169)
3259 			CSR_WRITE_4(sc, RL_TIMERINT_8169, RL_USECS(0));
3260 	}
3261 #endif
3262 
3263 	/*
3264 	 * For 8169 gigE NICs, set the max allowed RX packet
3265 	 * size so we can receive jumbo frames.
3266 	 */
3267 	if (sc->rl_type == RL_8169) {
3268 		if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
3269 			/*
3270 			 * For controllers that use new jumbo frame scheme,
3271 			 * set maximum size of jumbo frame depending on
3272 			 * controller revisions.
3273 			 */
3274 			if (ifp->if_mtu > RL_MTU)
3275 				CSR_WRITE_2(sc, RL_MAXRXPKTLEN,
3276 				    sc->rl_hwrev->rl_max_mtu +
3277 				    ETHER_VLAN_ENCAP_LEN + ETHER_HDR_LEN +
3278 				    ETHER_CRC_LEN);
3279 			else
3280 				CSR_WRITE_2(sc, RL_MAXRXPKTLEN,
3281 				    RE_RX_DESC_BUFLEN);
3282 		} else if ((sc->rl_flags & RL_FLAG_PCIE) != 0 &&
3283 		    sc->rl_hwrev->rl_max_mtu == RL_MTU) {
3284 			/* RTL810x has no jumbo frame support. */
3285 			CSR_WRITE_2(sc, RL_MAXRXPKTLEN, RE_RX_DESC_BUFLEN);
3286 		} else
3287 			CSR_WRITE_2(sc, RL_MAXRXPKTLEN, 16383);
3288 	}
3289 
3290 	if (sc->rl_testmode)
3291 		return;
3292 
3293 	CSR_WRITE_1(sc, sc->rl_cfg1, CSR_READ_1(sc, sc->rl_cfg1) |
3294 	    RL_CFG1_DRVLOAD);
3295 
3296 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
3297 	ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
3298 
3299 	sc->rl_flags &= ~RL_FLAG_LINK;
3300 	mii_mediachg(mii);
3301 
3302 	sc->rl_watchdog_timer = 0;
3303 	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
3304 }
3305 
3306 /*
3307  * Set media options.
3308  */
3309 static int
3310 re_ifmedia_upd(struct ifnet *ifp)
3311 {
3312 	struct rl_softc		*sc;
3313 	struct mii_data		*mii;
3314 	int			error;
3315 
3316 	sc = ifp->if_softc;
3317 	mii = device_get_softc(sc->rl_miibus);
3318 	RL_LOCK(sc);
3319 	error = mii_mediachg(mii);
3320 	RL_UNLOCK(sc);
3321 
3322 	return (error);
3323 }
3324 
3325 /*
3326  * Report current media status.
3327  */
3328 static void
3329 re_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
3330 {
3331 	struct rl_softc		*sc;
3332 	struct mii_data		*mii;
3333 
3334 	sc = ifp->if_softc;
3335 	mii = device_get_softc(sc->rl_miibus);
3336 
3337 	RL_LOCK(sc);
3338 	mii_pollstat(mii);
3339 	ifmr->ifm_active = mii->mii_media_active;
3340 	ifmr->ifm_status = mii->mii_media_status;
3341 	RL_UNLOCK(sc);
3342 }
3343 
3344 static int
3345 re_ioctl(struct ifnet *ifp, u_long command, caddr_t data)
3346 {
3347 	struct rl_softc		*sc = ifp->if_softc;
3348 	struct ifreq		*ifr = (struct ifreq *) data;
3349 	struct mii_data		*mii;
3350 	uint32_t		rev;
3351 	int			error = 0;
3352 
3353 	switch (command) {
3354 	case SIOCSIFMTU:
3355 		if (ifr->ifr_mtu < ETHERMIN ||
3356 		    ifr->ifr_mtu > sc->rl_hwrev->rl_max_mtu ||
3357 		    ((sc->rl_flags & RL_FLAG_FASTETHER) != 0 &&
3358 		    ifr->ifr_mtu > RL_MTU)) {
3359 			error = EINVAL;
3360 			break;
3361 		}
3362 		RL_LOCK(sc);
3363 		if (ifp->if_mtu != ifr->ifr_mtu) {
3364 			ifp->if_mtu = ifr->ifr_mtu;
3365 			if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0 &&
3366 			    (ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
3367 				ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3368 				re_init_locked(sc);
3369 			}
3370 			if (ifp->if_mtu > RL_TSO_MTU &&
3371 			    (ifp->if_capenable & IFCAP_TSO4) != 0) {
3372 				ifp->if_capenable &= ~(IFCAP_TSO4 |
3373 				    IFCAP_VLAN_HWTSO);
3374 				ifp->if_hwassist &= ~CSUM_TSO;
3375 			}
3376 			VLAN_CAPABILITIES(ifp);
3377 		}
3378 		RL_UNLOCK(sc);
3379 		break;
3380 	case SIOCSIFFLAGS:
3381 		RL_LOCK(sc);
3382 		if ((ifp->if_flags & IFF_UP) != 0) {
3383 			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
3384 				if (((ifp->if_flags ^ sc->rl_if_flags)
3385 				    & (IFF_PROMISC | IFF_ALLMULTI)) != 0)
3386 					re_set_rxmode(sc);
3387 			} else
3388 				re_init_locked(sc);
3389 		} else {
3390 			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
3391 				re_stop(sc);
3392 		}
3393 		sc->rl_if_flags = ifp->if_flags;
3394 		RL_UNLOCK(sc);
3395 		break;
3396 	case SIOCADDMULTI:
3397 	case SIOCDELMULTI:
3398 		RL_LOCK(sc);
3399 		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
3400 			re_set_rxmode(sc);
3401 		RL_UNLOCK(sc);
3402 		break;
3403 	case SIOCGIFMEDIA:
3404 	case SIOCSIFMEDIA:
3405 		mii = device_get_softc(sc->rl_miibus);
3406 		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, command);
3407 		break;
3408 	case SIOCSIFCAP:
3409 	    {
3410 		int mask, reinit;
3411 
3412 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
3413 		reinit = 0;
3414 #ifdef DEVICE_POLLING
3415 		if (mask & IFCAP_POLLING) {
3416 			if (ifr->ifr_reqcap & IFCAP_POLLING) {
3417 				error = ether_poll_register(re_poll, ifp);
3418 				if (error)
3419 					return (error);
3420 				RL_LOCK(sc);
3421 				/* Disable interrupts */
3422 				CSR_WRITE_2(sc, RL_IMR, 0x0000);
3423 				ifp->if_capenable |= IFCAP_POLLING;
3424 				RL_UNLOCK(sc);
3425 			} else {
3426 				error = ether_poll_deregister(ifp);
3427 				/* Enable interrupts. */
3428 				RL_LOCK(sc);
3429 				CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
3430 				ifp->if_capenable &= ~IFCAP_POLLING;
3431 				RL_UNLOCK(sc);
3432 			}
3433 		}
3434 #endif /* DEVICE_POLLING */
3435 		RL_LOCK(sc);
3436 		if ((mask & IFCAP_TXCSUM) != 0 &&
3437 		    (ifp->if_capabilities & IFCAP_TXCSUM) != 0) {
3438 			ifp->if_capenable ^= IFCAP_TXCSUM;
3439 			if ((ifp->if_capenable & IFCAP_TXCSUM) != 0) {
3440 				rev = sc->rl_hwrev->rl_rev;
3441 				if (rev == RL_HWREV_8168C ||
3442 				    rev == RL_HWREV_8168C_SPIN2 ||
3443 				    rev == RL_HWREV_8168CP)
3444 					ifp->if_hwassist |= CSUM_TCP | CSUM_UDP;
3445 				else
3446 					ifp->if_hwassist |= RE_CSUM_FEATURES;
3447 			} else
3448 				ifp->if_hwassist &= ~RE_CSUM_FEATURES;
3449 			reinit = 1;
3450 		}
3451 		if ((mask & IFCAP_RXCSUM) != 0 &&
3452 		    (ifp->if_capabilities & IFCAP_RXCSUM) != 0) {
3453 			ifp->if_capenable ^= IFCAP_RXCSUM;
3454 			reinit = 1;
3455 		}
3456 		if ((mask & IFCAP_TSO4) != 0 &&
3457 		    (ifp->if_capabilities & IFCAP_TSO4) != 0) {
3458 			ifp->if_capenable ^= IFCAP_TSO4;
3459 			if ((IFCAP_TSO4 & ifp->if_capenable) != 0)
3460 				ifp->if_hwassist |= CSUM_TSO;
3461 			else
3462 				ifp->if_hwassist &= ~CSUM_TSO;
3463 			if (ifp->if_mtu > RL_TSO_MTU &&
3464 			    (ifp->if_capenable & IFCAP_TSO4) != 0) {
3465 				ifp->if_capenable &= ~IFCAP_TSO4;
3466 				ifp->if_hwassist &= ~CSUM_TSO;
3467 			}
3468 		}
3469 		if ((mask & IFCAP_VLAN_HWTSO) != 0 &&
3470 		    (ifp->if_capabilities & IFCAP_VLAN_HWTSO) != 0)
3471 			ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
3472 		if ((mask & IFCAP_VLAN_HWTAGGING) != 0 &&
3473 		    (ifp->if_capabilities & IFCAP_VLAN_HWTAGGING) != 0) {
3474 			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
3475 			/* TSO over VLAN requires VLAN hardware tagging. */
3476 			if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) == 0)
3477 				ifp->if_capenable &= ~IFCAP_VLAN_HWTSO;
3478 			reinit = 1;
3479 		}
3480 		if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0 &&
3481 		    (mask & (IFCAP_HWCSUM | IFCAP_TSO4 |
3482 		    IFCAP_VLAN_HWTSO)) != 0)
3483 				reinit = 1;
3484 		if ((mask & IFCAP_WOL) != 0 &&
3485 		    (ifp->if_capabilities & IFCAP_WOL) != 0) {
3486 			if ((mask & IFCAP_WOL_UCAST) != 0)
3487 				ifp->if_capenable ^= IFCAP_WOL_UCAST;
3488 			if ((mask & IFCAP_WOL_MCAST) != 0)
3489 				ifp->if_capenable ^= IFCAP_WOL_MCAST;
3490 			if ((mask & IFCAP_WOL_MAGIC) != 0)
3491 				ifp->if_capenable ^= IFCAP_WOL_MAGIC;
3492 		}
3493 		if (reinit && ifp->if_drv_flags & IFF_DRV_RUNNING) {
3494 			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3495 			re_init_locked(sc);
3496 		}
3497 		RL_UNLOCK(sc);
3498 		VLAN_CAPABILITIES(ifp);
3499 	    }
3500 		break;
3501 	default:
3502 		error = ether_ioctl(ifp, command, data);
3503 		break;
3504 	}
3505 
3506 	return (error);
3507 }
3508 
3509 static void
3510 re_watchdog(struct rl_softc *sc)
3511 {
3512 	struct ifnet		*ifp;
3513 
3514 	RL_LOCK_ASSERT(sc);
3515 
3516 	if (sc->rl_watchdog_timer == 0 || --sc->rl_watchdog_timer != 0)
3517 		return;
3518 
3519 	ifp = sc->rl_ifp;
3520 	re_txeof(sc);
3521 	if (sc->rl_ldata.rl_tx_free == sc->rl_ldata.rl_tx_desc_cnt) {
3522 		if_printf(ifp, "watchdog timeout (missed Tx interrupts) "
3523 		    "-- recovering\n");
3524 		if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
3525 			re_start_locked(ifp);
3526 		return;
3527 	}
3528 
3529 	if_printf(ifp, "watchdog timeout\n");
3530 	ifp->if_oerrors++;
3531 
3532 	re_rxeof(sc, NULL);
3533 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3534 	re_init_locked(sc);
3535 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
3536 		re_start_locked(ifp);
3537 }
3538 
3539 /*
3540  * Stop the adapter and free any mbufs allocated to the
3541  * RX and TX lists.
3542  */
3543 static void
3544 re_stop(struct rl_softc *sc)
3545 {
3546 	int			i;
3547 	struct ifnet		*ifp;
3548 	struct rl_txdesc	*txd;
3549 	struct rl_rxdesc	*rxd;
3550 
3551 	RL_LOCK_ASSERT(sc);
3552 
3553 	ifp = sc->rl_ifp;
3554 
3555 	sc->rl_watchdog_timer = 0;
3556 	callout_stop(&sc->rl_stat_callout);
3557 	ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
3558 
3559 	/*
3560 	 * Disable accepting frames to put RX MAC into idle state.
3561 	 * Otherwise it's possible to get frames while stop command
3562 	 * execution is in progress and controller can DMA the frame
3563 	 * to already freed RX buffer during that period.
3564 	 */
3565 	CSR_WRITE_4(sc, RL_RXCFG, CSR_READ_4(sc, RL_RXCFG) &
3566 	    ~(RL_RXCFG_RX_ALLPHYS | RL_RXCFG_RX_INDIV | RL_RXCFG_RX_MULTI |
3567 	    RL_RXCFG_RX_BROAD));
3568 
3569 	if ((sc->rl_flags & RL_FLAG_WAIT_TXPOLL) != 0) {
3570 		for (i = RL_TIMEOUT; i > 0; i--) {
3571 			if ((CSR_READ_1(sc, sc->rl_txstart) &
3572 			    RL_TXSTART_START) == 0)
3573 				break;
3574 			DELAY(20);
3575 		}
3576 		if (i == 0)
3577 			device_printf(sc->rl_dev,
3578 			    "stopping TX poll timed out!\n");
3579 		CSR_WRITE_1(sc, RL_COMMAND, 0x00);
3580 	} else if ((sc->rl_flags & RL_FLAG_CMDSTOP) != 0) {
3581 		CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_STOPREQ | RL_CMD_TX_ENB |
3582 		    RL_CMD_RX_ENB);
3583 		if ((sc->rl_flags & RL_FLAG_CMDSTOP_WAIT_TXQ) != 0) {
3584 			for (i = RL_TIMEOUT; i > 0; i--) {
3585 				if ((CSR_READ_4(sc, RL_TXCFG) &
3586 				    RL_TXCFG_QUEUE_EMPTY) != 0)
3587 					break;
3588 				DELAY(100);
3589 			}
3590 			if (i == 0)
3591 				device_printf(sc->rl_dev,
3592 				   "stopping TXQ timed out!\n");
3593 		}
3594 	} else
3595 		CSR_WRITE_1(sc, RL_COMMAND, 0x00);
3596 	DELAY(1000);
3597 	CSR_WRITE_2(sc, RL_IMR, 0x0000);
3598 	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
3599 
3600 	if (sc->rl_head != NULL) {
3601 		m_freem(sc->rl_head);
3602 		sc->rl_head = sc->rl_tail = NULL;
3603 	}
3604 
3605 	/* Free the TX list buffers. */
3606 	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
3607 		txd = &sc->rl_ldata.rl_tx_desc[i];
3608 		if (txd->tx_m != NULL) {
3609 			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
3610 			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
3611 			bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag,
3612 			    txd->tx_dmamap);
3613 			m_freem(txd->tx_m);
3614 			txd->tx_m = NULL;
3615 		}
3616 	}
3617 
3618 	/* Free the RX list buffers. */
3619 	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
3620 		rxd = &sc->rl_ldata.rl_rx_desc[i];
3621 		if (rxd->rx_m != NULL) {
3622 			bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag,
3623 			    rxd->rx_dmamap, BUS_DMASYNC_POSTREAD);
3624 			bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag,
3625 			    rxd->rx_dmamap);
3626 			m_freem(rxd->rx_m);
3627 			rxd->rx_m = NULL;
3628 		}
3629 	}
3630 
3631 	if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
3632 		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
3633 			rxd = &sc->rl_ldata.rl_jrx_desc[i];
3634 			if (rxd->rx_m != NULL) {
3635 				bus_dmamap_sync(sc->rl_ldata.rl_jrx_mtag,
3636 				    rxd->rx_dmamap, BUS_DMASYNC_POSTREAD);
3637 				bus_dmamap_unload(sc->rl_ldata.rl_jrx_mtag,
3638 				    rxd->rx_dmamap);
3639 				m_freem(rxd->rx_m);
3640 				rxd->rx_m = NULL;
3641 			}
3642 		}
3643 	}
3644 }
3645 
3646 /*
3647  * Device suspend routine.  Stop the interface and save some PCI
3648  * settings in case the BIOS doesn't restore them properly on
3649  * resume.
3650  */
3651 static int
3652 re_suspend(device_t dev)
3653 {
3654 	struct rl_softc		*sc;
3655 
3656 	sc = device_get_softc(dev);
3657 
3658 	RL_LOCK(sc);
3659 	re_stop(sc);
3660 	re_setwol(sc);
3661 	sc->suspended = 1;
3662 	RL_UNLOCK(sc);
3663 
3664 	return (0);
3665 }
3666 
3667 /*
3668  * Device resume routine.  Restore some PCI settings in case the BIOS
3669  * doesn't, re-enable busmastering, and restart the interface if
3670  * appropriate.
3671  */
3672 static int
3673 re_resume(device_t dev)
3674 {
3675 	struct rl_softc		*sc;
3676 	struct ifnet		*ifp;
3677 
3678 	sc = device_get_softc(dev);
3679 
3680 	RL_LOCK(sc);
3681 
3682 	ifp = sc->rl_ifp;
3683 	/* Take controller out of sleep mode. */
3684 	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
3685 		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
3686 			CSR_WRITE_1(sc, RL_GPIO,
3687 			    CSR_READ_1(sc, RL_GPIO) | 0x01);
3688 	}
3689 
3690 	/*
3691 	 * Clear WOL matching such that normal Rx filtering
3692 	 * wouldn't interfere with WOL patterns.
3693 	 */
3694 	re_clrwol(sc);
3695 
3696 	/* reinitialize interface if necessary */
3697 	if (ifp->if_flags & IFF_UP)
3698 		re_init_locked(sc);
3699 
3700 	sc->suspended = 0;
3701 	RL_UNLOCK(sc);
3702 
3703 	return (0);
3704 }
3705 
3706 /*
3707  * Stop all chip I/O so that the kernel's probe routines don't
3708  * get confused by errant DMAs when rebooting.
3709  */
3710 static int
3711 re_shutdown(device_t dev)
3712 {
3713 	struct rl_softc		*sc;
3714 
3715 	sc = device_get_softc(dev);
3716 
3717 	RL_LOCK(sc);
3718 	re_stop(sc);
3719 	/*
3720 	 * Mark interface as down since otherwise we will panic if
3721 	 * interrupt comes in later on, which can happen in some
3722 	 * cases.
3723 	 */
3724 	sc->rl_ifp->if_flags &= ~IFF_UP;
3725 	re_setwol(sc);
3726 	RL_UNLOCK(sc);
3727 
3728 	return (0);
3729 }
3730 
3731 static void
3732 re_set_linkspeed(struct rl_softc *sc)
3733 {
3734 	struct mii_softc *miisc;
3735 	struct mii_data *mii;
3736 	int aneg, i, phyno;
3737 
3738 	RL_LOCK_ASSERT(sc);
3739 
3740 	mii = device_get_softc(sc->rl_miibus);
3741 	mii_pollstat(mii);
3742 	aneg = 0;
3743 	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
3744 	    (IFM_ACTIVE | IFM_AVALID)) {
3745 		switch IFM_SUBTYPE(mii->mii_media_active) {
3746 		case IFM_10_T:
3747 		case IFM_100_TX:
3748 			return;
3749 		case IFM_1000_T:
3750 			aneg++;
3751 			break;
3752 		default:
3753 			break;
3754 		}
3755 	}
3756 	miisc = LIST_FIRST(&mii->mii_phys);
3757 	phyno = miisc->mii_phy;
3758 	LIST_FOREACH(miisc, &mii->mii_phys, mii_list)
3759 		PHY_RESET(miisc);
3760 	re_miibus_writereg(sc->rl_dev, phyno, MII_100T2CR, 0);
3761 	re_miibus_writereg(sc->rl_dev, phyno,
3762 	    MII_ANAR, ANAR_TX_FD | ANAR_TX | ANAR_10_FD | ANAR_10 | ANAR_CSMA);
3763 	re_miibus_writereg(sc->rl_dev, phyno,
3764 	    MII_BMCR, BMCR_AUTOEN | BMCR_STARTNEG);
3765 	DELAY(1000);
3766 	if (aneg != 0) {
3767 		/*
3768 		 * Poll link state until re(4) get a 10/100Mbps link.
3769 		 */
3770 		for (i = 0; i < MII_ANEGTICKS_GIGE; i++) {
3771 			mii_pollstat(mii);
3772 			if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID))
3773 			    == (IFM_ACTIVE | IFM_AVALID)) {
3774 				switch (IFM_SUBTYPE(mii->mii_media_active)) {
3775 				case IFM_10_T:
3776 				case IFM_100_TX:
3777 					return;
3778 				default:
3779 					break;
3780 				}
3781 			}
3782 			RL_UNLOCK(sc);
3783 			pause("relnk", hz);
3784 			RL_LOCK(sc);
3785 		}
3786 		if (i == MII_ANEGTICKS_GIGE)
3787 			device_printf(sc->rl_dev,
3788 			    "establishing a link failed, WOL may not work!");
3789 	}
3790 	/*
3791 	 * No link, force MAC to have 100Mbps, full-duplex link.
3792 	 * MAC does not require reprogramming on resolved speed/duplex,
3793 	 * so this is just for completeness.
3794 	 */
3795 	mii->mii_media_status = IFM_AVALID | IFM_ACTIVE;
3796 	mii->mii_media_active = IFM_ETHER | IFM_100_TX | IFM_FDX;
3797 }
3798 
3799 static void
3800 re_setwol(struct rl_softc *sc)
3801 {
3802 	struct ifnet		*ifp;
3803 	int			pmc;
3804 	uint16_t		pmstat;
3805 	uint8_t			v;
3806 
3807 	RL_LOCK_ASSERT(sc);
3808 
3809 	if (pci_find_cap(sc->rl_dev, PCIY_PMG, &pmc) != 0)
3810 		return;
3811 
3812 	ifp = sc->rl_ifp;
3813 	/* Put controller into sleep mode. */
3814 	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
3815 		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
3816 			CSR_WRITE_1(sc, RL_GPIO,
3817 			    CSR_READ_1(sc, RL_GPIO) & ~0x01);
3818 	}
3819 	if ((ifp->if_capenable & IFCAP_WOL) != 0) {
3820 		re_set_rxmode(sc);
3821 		if ((sc->rl_flags & RL_FLAG_WOL_MANLINK) != 0)
3822 			re_set_linkspeed(sc);
3823 		if ((sc->rl_flags & RL_FLAG_WOLRXENB) != 0)
3824 			CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RX_ENB);
3825 	}
3826 	/* Enable config register write. */
3827 	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
3828 
3829 	/* Enable PME. */
3830 	v = CSR_READ_1(sc, sc->rl_cfg1);
3831 	v &= ~RL_CFG1_PME;
3832 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3833 		v |= RL_CFG1_PME;
3834 	CSR_WRITE_1(sc, sc->rl_cfg1, v);
3835 
3836 	v = CSR_READ_1(sc, sc->rl_cfg3);
3837 	v &= ~(RL_CFG3_WOL_LINK | RL_CFG3_WOL_MAGIC);
3838 	if ((ifp->if_capenable & IFCAP_WOL_MAGIC) != 0)
3839 		v |= RL_CFG3_WOL_MAGIC;
3840 	CSR_WRITE_1(sc, sc->rl_cfg3, v);
3841 
3842 	v = CSR_READ_1(sc, sc->rl_cfg5);
3843 	v &= ~(RL_CFG5_WOL_BCAST | RL_CFG5_WOL_MCAST | RL_CFG5_WOL_UCAST |
3844 	    RL_CFG5_WOL_LANWAKE);
3845 	if ((ifp->if_capenable & IFCAP_WOL_UCAST) != 0)
3846 		v |= RL_CFG5_WOL_UCAST;
3847 	if ((ifp->if_capenable & IFCAP_WOL_MCAST) != 0)
3848 		v |= RL_CFG5_WOL_MCAST | RL_CFG5_WOL_BCAST;
3849 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3850 		v |= RL_CFG5_WOL_LANWAKE;
3851 	CSR_WRITE_1(sc, sc->rl_cfg5, v);
3852 
3853 	/* Config register write done. */
3854 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3855 
3856 	if ((ifp->if_capenable & IFCAP_WOL) == 0 &&
3857 	    (sc->rl_flags & RL_FLAG_PHYWAKE_PM) != 0)
3858 		CSR_WRITE_1(sc, RL_PMCH, CSR_READ_1(sc, RL_PMCH) & ~0x80);
3859 	/*
3860 	 * It seems that hardware resets its link speed to 100Mbps in
3861 	 * power down mode so switching to 100Mbps in driver is not
3862 	 * needed.
3863 	 */
3864 
3865 	/* Request PME if WOL is requested. */
3866 	pmstat = pci_read_config(sc->rl_dev, pmc + PCIR_POWER_STATUS, 2);
3867 	pmstat &= ~(PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE);
3868 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3869 		pmstat |= PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE;
3870 	pci_write_config(sc->rl_dev, pmc + PCIR_POWER_STATUS, pmstat, 2);
3871 }
3872 
3873 static void
3874 re_clrwol(struct rl_softc *sc)
3875 {
3876 	int			pmc;
3877 	uint8_t			v;
3878 
3879 	RL_LOCK_ASSERT(sc);
3880 
3881 	if (pci_find_cap(sc->rl_dev, PCIY_PMG, &pmc) != 0)
3882 		return;
3883 
3884 	/* Enable config register write. */
3885 	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
3886 
3887 	v = CSR_READ_1(sc, sc->rl_cfg3);
3888 	v &= ~(RL_CFG3_WOL_LINK | RL_CFG3_WOL_MAGIC);
3889 	CSR_WRITE_1(sc, sc->rl_cfg3, v);
3890 
3891 	/* Config register write done. */
3892 	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3893 
3894 	v = CSR_READ_1(sc, sc->rl_cfg5);
3895 	v &= ~(RL_CFG5_WOL_BCAST | RL_CFG5_WOL_MCAST | RL_CFG5_WOL_UCAST);
3896 	v &= ~RL_CFG5_WOL_LANWAKE;
3897 	CSR_WRITE_1(sc, sc->rl_cfg5, v);
3898 }
3899 
3900 static void
3901 re_add_sysctls(struct rl_softc *sc)
3902 {
3903 	struct sysctl_ctx_list	*ctx;
3904 	struct sysctl_oid_list	*children;
3905 	int			error;
3906 
3907 	ctx = device_get_sysctl_ctx(sc->rl_dev);
3908 	children = SYSCTL_CHILDREN(device_get_sysctl_tree(sc->rl_dev));
3909 
3910 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "stats",
3911 	    CTLTYPE_INT | CTLFLAG_RW, sc, 0, re_sysctl_stats, "I",
3912 	    "Statistics Information");
3913 	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) == 0)
3914 		return;
3915 
3916 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "int_rx_mod",
3917 	    CTLTYPE_INT | CTLFLAG_RW, &sc->rl_int_rx_mod, 0,
3918 	    sysctl_hw_re_int_mod, "I", "re RX interrupt moderation");
3919 	/* Pull in device tunables. */
3920 	sc->rl_int_rx_mod = RL_TIMER_DEFAULT;
3921 	error = resource_int_value(device_get_name(sc->rl_dev),
3922 	    device_get_unit(sc->rl_dev), "int_rx_mod", &sc->rl_int_rx_mod);
3923 	if (error == 0) {
3924 		if (sc->rl_int_rx_mod < RL_TIMER_MIN ||
3925 		    sc->rl_int_rx_mod > RL_TIMER_MAX) {
3926 			device_printf(sc->rl_dev, "int_rx_mod value out of "
3927 			    "range; using default: %d\n",
3928 			    RL_TIMER_DEFAULT);
3929 			sc->rl_int_rx_mod = RL_TIMER_DEFAULT;
3930 		}
3931 	}
3932 
3933 }
3934 
3935 static int
3936 re_sysctl_stats(SYSCTL_HANDLER_ARGS)
3937 {
3938 	struct rl_softc		*sc;
3939 	struct rl_stats		*stats;
3940 	int			error, i, result;
3941 
3942 	result = -1;
3943 	error = sysctl_handle_int(oidp, &result, 0, req);
3944 	if (error || req->newptr == NULL)
3945 		return (error);
3946 
3947 	if (result == 1) {
3948 		sc = (struct rl_softc *)arg1;
3949 		RL_LOCK(sc);
3950 		if ((sc->rl_ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
3951 			RL_UNLOCK(sc);
3952 			goto done;
3953 		}
3954 		bus_dmamap_sync(sc->rl_ldata.rl_stag,
3955 		    sc->rl_ldata.rl_smap, BUS_DMASYNC_PREREAD);
3956 		CSR_WRITE_4(sc, RL_DUMPSTATS_HI,
3957 		    RL_ADDR_HI(sc->rl_ldata.rl_stats_addr));
3958 		CSR_WRITE_4(sc, RL_DUMPSTATS_LO,
3959 		    RL_ADDR_LO(sc->rl_ldata.rl_stats_addr));
3960 		CSR_WRITE_4(sc, RL_DUMPSTATS_LO,
3961 		    RL_ADDR_LO(sc->rl_ldata.rl_stats_addr |
3962 		    RL_DUMPSTATS_START));
3963 		for (i = RL_TIMEOUT; i > 0; i--) {
3964 			if ((CSR_READ_4(sc, RL_DUMPSTATS_LO) &
3965 			    RL_DUMPSTATS_START) == 0)
3966 				break;
3967 			DELAY(1000);
3968 		}
3969 		bus_dmamap_sync(sc->rl_ldata.rl_stag,
3970 		    sc->rl_ldata.rl_smap, BUS_DMASYNC_POSTREAD);
3971 		RL_UNLOCK(sc);
3972 		if (i == 0) {
3973 			device_printf(sc->rl_dev,
3974 			    "DUMP statistics request timed out\n");
3975 			return (ETIMEDOUT);
3976 		}
3977 done:
3978 		stats = sc->rl_ldata.rl_stats;
3979 		printf("%s statistics:\n", device_get_nameunit(sc->rl_dev));
3980 		printf("Tx frames : %ju\n",
3981 		    (uintmax_t)le64toh(stats->rl_tx_pkts));
3982 		printf("Rx frames : %ju\n",
3983 		    (uintmax_t)le64toh(stats->rl_rx_pkts));
3984 		printf("Tx errors : %ju\n",
3985 		    (uintmax_t)le64toh(stats->rl_tx_errs));
3986 		printf("Rx errors : %u\n",
3987 		    le32toh(stats->rl_rx_errs));
3988 		printf("Rx missed frames : %u\n",
3989 		    (uint32_t)le16toh(stats->rl_missed_pkts));
3990 		printf("Rx frame alignment errs : %u\n",
3991 		    (uint32_t)le16toh(stats->rl_rx_framealign_errs));
3992 		printf("Tx single collisions : %u\n",
3993 		    le32toh(stats->rl_tx_onecoll));
3994 		printf("Tx multiple collisions : %u\n",
3995 		    le32toh(stats->rl_tx_multicolls));
3996 		printf("Rx unicast frames : %ju\n",
3997 		    (uintmax_t)le64toh(stats->rl_rx_ucasts));
3998 		printf("Rx broadcast frames : %ju\n",
3999 		    (uintmax_t)le64toh(stats->rl_rx_bcasts));
4000 		printf("Rx multicast frames : %u\n",
4001 		    le32toh(stats->rl_rx_mcasts));
4002 		printf("Tx aborts : %u\n",
4003 		    (uint32_t)le16toh(stats->rl_tx_aborts));
4004 		printf("Tx underruns : %u\n",
4005 		    (uint32_t)le16toh(stats->rl_rx_underruns));
4006 	}
4007 
4008 	return (error);
4009 }
4010 
4011 static int
4012 sysctl_int_range(SYSCTL_HANDLER_ARGS, int low, int high)
4013 {
4014 	int error, value;
4015 
4016 	if (arg1 == NULL)
4017 		return (EINVAL);
4018 	value = *(int *)arg1;
4019 	error = sysctl_handle_int(oidp, &value, 0, req);
4020 	if (error || req->newptr == NULL)
4021 		return (error);
4022 	if (value < low || value > high)
4023 		return (EINVAL);
4024 	*(int *)arg1 = value;
4025 
4026 	return (0);
4027 }
4028 
4029 static int
4030 sysctl_hw_re_int_mod(SYSCTL_HANDLER_ARGS)
4031 {
4032 
4033 	return (sysctl_int_range(oidp, arg1, arg2, req, RL_TIMER_MIN,
4034 	    RL_TIMER_MAX));
4035 }
4036