if_re.c revision 187483
133965Sjdp/*-
289857Sobrien * Copyright (c) 1997, 1998-2003
333965Sjdp *	Bill Paul <wpaul@windriver.com>.  All rights reserved.
433965Sjdp *
533965Sjdp * Redistribution and use in source and binary forms, with or without
633965Sjdp * modification, are permitted provided that the following conditions
733965Sjdp * are met:
833965Sjdp * 1. Redistributions of source code must retain the above copyright
933965Sjdp *    notice, this list of conditions and the following disclaimer.
1033965Sjdp * 2. Redistributions in binary form must reproduce the above copyright
1133965Sjdp *    notice, this list of conditions and the following disclaimer in the
1233965Sjdp *    documentation and/or other materials provided with the distribution.
1333965Sjdp * 3. All advertising materials mentioning features or use of this software
1433965Sjdp *    must display the following acknowledgement:
1533965Sjdp *	This product includes software developed by Bill Paul.
1633965Sjdp * 4. Neither the name of the author nor the names of any co-contributors
1733965Sjdp *    may be used to endorse or promote products derived from this software
18218822Sdim *    without specific prior written permission.
1933965Sjdp *
2033965Sjdp * THIS SOFTWARE IS PROVIDED BY Bill Paul AND CONTRIBUTORS ``AS IS'' AND
2133965Sjdp * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
2233965Sjdp * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
2333965Sjdp * ARE DISCLAIMED.  IN NO EVENT SHALL Bill Paul OR THE VOICES IN HIS HEAD
2433965Sjdp * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
2533965Sjdp * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
2633965Sjdp * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
2733965Sjdp * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
2833965Sjdp * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
2989857Sobrien * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
3089857Sobrien * THE POSSIBILITY OF SUCH DAMAGE.
3189857Sobrien */
3289857Sobrien
3389857Sobrien#include <sys/cdefs.h>
3433965Sjdp__FBSDID("$FreeBSD: head/sys/dev/re/if_re.c 187483 2009-01-20 20:22:28Z jkim $");
3533965Sjdp
3633965Sjdp/*
3733965Sjdp * 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/taskqueue.h>
127
128#include <net/if.h>
129#include <net/if_arp.h>
130#include <net/ethernet.h>
131#include <net/if_dl.h>
132#include <net/if_media.h>
133#include <net/if_types.h>
134#include <net/if_vlan_var.h>
135
136#include <net/bpf.h>
137
138#include <machine/bus.h>
139#include <machine/resource.h>
140#include <sys/bus.h>
141#include <sys/rman.h>
142
143#include <dev/mii/mii.h>
144#include <dev/mii/miivar.h>
145
146#include <dev/pci/pcireg.h>
147#include <dev/pci/pcivar.h>
148
149#include <pci/if_rlreg.h>
150
151MODULE_DEPEND(re, pci, 1, 1, 1);
152MODULE_DEPEND(re, ether, 1, 1, 1);
153MODULE_DEPEND(re, miibus, 1, 1, 1);
154
155/* "device miibus" required.  See GENERIC if you get errors here. */
156#include "miibus_if.h"
157
158/* Tunables. */
159static int msi_disable = 1;
160TUNABLE_INT("hw.re.msi_disable", &msi_disable);
161
162#define RE_CSUM_FEATURES    (CSUM_IP | CSUM_TCP | CSUM_UDP)
163
164/*
165 * Various supported device vendors/types and their names.
166 */
167static struct rl_type re_devs[] = {
168	{ DLINK_VENDORID, DLINK_DEVICEID_528T, 0,
169	    "D-Link DGE-528(T) Gigabit Ethernet Adapter" },
170	{ RT_VENDORID, RT_DEVICEID_8139, 0,
171	    "RealTek 8139C+ 10/100BaseTX" },
172	{ RT_VENDORID, RT_DEVICEID_8101E, 0,
173	    "RealTek 8101E/8102E/8102EL PCIe 10/100baseTX" },
174	{ RT_VENDORID, RT_DEVICEID_8168, 0,
175	    "RealTek 8168/8168B/8168C/8168CP/8168D/8111B/8111C/8111CP PCIe "
176	    "Gigabit Ethernet" },
177	{ RT_VENDORID, RT_DEVICEID_8169, 0,
178	    "RealTek 8169/8169S/8169SB(L)/8110S/8110SB(L) Gigabit Ethernet" },
179	{ RT_VENDORID, RT_DEVICEID_8169SC, 0,
180	    "RealTek 8169SC/8110SC Single-chip Gigabit Ethernet" },
181	{ COREGA_VENDORID, COREGA_DEVICEID_CGLAPCIGT, 0,
182	    "Corega CG-LAPCIGT (RTL8169S) Gigabit Ethernet" },
183	{ LINKSYS_VENDORID, LINKSYS_DEVICEID_EG1032, 0,
184	    "Linksys EG1032 (RTL8169S) Gigabit Ethernet" },
185	{ USR_VENDORID, USR_DEVICEID_997902, 0,
186	    "US Robotics 997902 (RTL8169S) Gigabit Ethernet" }
187};
188
189static struct rl_hwrev re_hwrevs[] = {
190	{ RL_HWREV_8139, RL_8139,  "" },
191	{ RL_HWREV_8139A, RL_8139, "A" },
192	{ RL_HWREV_8139AG, RL_8139, "A-G" },
193	{ RL_HWREV_8139B, RL_8139, "B" },
194	{ RL_HWREV_8130, RL_8139, "8130" },
195	{ RL_HWREV_8139C, RL_8139, "C" },
196	{ RL_HWREV_8139D, RL_8139, "8139D/8100B/8100C" },
197	{ RL_HWREV_8139CPLUS, RL_8139CPLUS, "C+"},
198	{ RL_HWREV_8168_SPIN1, RL_8169, "8168"},
199	{ RL_HWREV_8169, RL_8169, "8169"},
200	{ RL_HWREV_8169S, RL_8169, "8169S"},
201	{ RL_HWREV_8110S, RL_8169, "8110S"},
202	{ RL_HWREV_8169_8110SB, RL_8169, "8169SB/8110SB"},
203	{ RL_HWREV_8169_8110SC, RL_8169, "8169SC/8110SC"},
204	{ RL_HWREV_8169_8110SBL, RL_8169, "8169SBL/8110SBL"},
205	{ RL_HWREV_8169_8110SCE, RL_8169, "8169SC/8110SC"},
206	{ RL_HWREV_8100, RL_8139, "8100"},
207	{ RL_HWREV_8101, RL_8139, "8101"},
208	{ RL_HWREV_8100E, RL_8169, "8100E"},
209	{ RL_HWREV_8101E, RL_8169, "8101E"},
210	{ RL_HWREV_8102E, RL_8169, "8102E"},
211	{ RL_HWREV_8102EL, RL_8169, "8102EL"},
212	{ RL_HWREV_8168_SPIN2, RL_8169, "8168"},
213	{ RL_HWREV_8168_SPIN3, RL_8169, "8168"},
214	{ RL_HWREV_8168C, RL_8169, "8168C/8111C"},
215	{ RL_HWREV_8168C_SPIN2, RL_8169, "8168C/8111C"},
216	{ RL_HWREV_8168CP, RL_8169, "8168CP/8111CP"},
217	{ RL_HWREV_8168D, RL_8169, "8168D"},
218	{ 0, 0, NULL }
219};
220
221static int re_probe		(device_t);
222static int re_attach		(device_t);
223static int re_detach		(device_t);
224
225static int re_encap		(struct rl_softc *, struct mbuf **);
226
227static void re_dma_map_addr	(void *, bus_dma_segment_t *, int, int);
228static int re_allocmem		(device_t, struct rl_softc *);
229static __inline void re_discard_rxbuf
230				(struct rl_softc *, int);
231static int re_newbuf		(struct rl_softc *, int);
232static int re_rx_list_init	(struct rl_softc *);
233static int re_tx_list_init	(struct rl_softc *);
234#ifdef RE_FIXUP_RX
235static __inline void re_fixup_rx
236				(struct mbuf *);
237#endif
238static int re_rxeof		(struct rl_softc *);
239static void re_txeof		(struct rl_softc *);
240#ifdef DEVICE_POLLING
241static void re_poll		(struct ifnet *, enum poll_cmd, int);
242static void re_poll_locked	(struct ifnet *, enum poll_cmd, int);
243#endif
244static int re_intr		(void *);
245static void re_tick		(void *);
246static void re_tx_task		(void *, int);
247static void re_int_task		(void *, int);
248static void re_start		(struct ifnet *);
249static int re_ioctl		(struct ifnet *, u_long, caddr_t);
250static void re_init		(void *);
251static void re_init_locked	(struct rl_softc *);
252static void re_stop		(struct rl_softc *);
253static void re_watchdog		(struct rl_softc *);
254static int re_suspend		(device_t);
255static int re_resume		(device_t);
256static int re_shutdown		(device_t);
257static int re_ifmedia_upd	(struct ifnet *);
258static void re_ifmedia_sts	(struct ifnet *, struct ifmediareq *);
259
260static void re_eeprom_putbyte	(struct rl_softc *, int);
261static void re_eeprom_getword	(struct rl_softc *, int, u_int16_t *);
262static void re_read_eeprom	(struct rl_softc *, caddr_t, int, int);
263static int re_gmii_readreg	(device_t, int, int);
264static int re_gmii_writereg	(device_t, int, int, int);
265
266static int re_miibus_readreg	(device_t, int, int);
267static int re_miibus_writereg	(device_t, int, int, int);
268static void re_miibus_statchg	(device_t);
269
270static void re_set_rxmode		(struct rl_softc *);
271static void re_reset		(struct rl_softc *);
272static void re_setwol		(struct rl_softc *);
273static void re_clrwol		(struct rl_softc *);
274
275#ifdef RE_DIAG
276static int re_diag		(struct rl_softc *);
277#endif
278
279static device_method_t re_methods[] = {
280	/* Device interface */
281	DEVMETHOD(device_probe,		re_probe),
282	DEVMETHOD(device_attach,	re_attach),
283	DEVMETHOD(device_detach,	re_detach),
284	DEVMETHOD(device_suspend,	re_suspend),
285	DEVMETHOD(device_resume,	re_resume),
286	DEVMETHOD(device_shutdown,	re_shutdown),
287
288	/* bus interface */
289	DEVMETHOD(bus_print_child,	bus_generic_print_child),
290	DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
291
292	/* MII interface */
293	DEVMETHOD(miibus_readreg,	re_miibus_readreg),
294	DEVMETHOD(miibus_writereg,	re_miibus_writereg),
295	DEVMETHOD(miibus_statchg,	re_miibus_statchg),
296
297	{ 0, 0 }
298};
299
300static driver_t re_driver = {
301	"re",
302	re_methods,
303	sizeof(struct rl_softc)
304};
305
306static devclass_t re_devclass;
307
308DRIVER_MODULE(re, pci, re_driver, re_devclass, 0, 0);
309DRIVER_MODULE(re, cardbus, re_driver, re_devclass, 0, 0);
310DRIVER_MODULE(miibus, re, miibus_driver, miibus_devclass, 0, 0);
311
312#define EE_SET(x)					\
313	CSR_WRITE_1(sc, RL_EECMD,			\
314		CSR_READ_1(sc, RL_EECMD) | x)
315
316#define EE_CLR(x)					\
317	CSR_WRITE_1(sc, RL_EECMD,			\
318		CSR_READ_1(sc, RL_EECMD) & ~x)
319
320/*
321 * Send a read command and address to the EEPROM, check for ACK.
322 */
323static void
324re_eeprom_putbyte(struct rl_softc *sc, int addr)
325{
326	int			d, i;
327
328	d = addr | (RL_9346_READ << sc->rl_eewidth);
329
330	/*
331	 * Feed in each bit and strobe the clock.
332	 */
333
334	for (i = 1 << (sc->rl_eewidth + 3); i; i >>= 1) {
335		if (d & i) {
336			EE_SET(RL_EE_DATAIN);
337		} else {
338			EE_CLR(RL_EE_DATAIN);
339		}
340		DELAY(100);
341		EE_SET(RL_EE_CLK);
342		DELAY(150);
343		EE_CLR(RL_EE_CLK);
344		DELAY(100);
345	}
346}
347
348/*
349 * Read a word of data stored in the EEPROM at address 'addr.'
350 */
351static void
352re_eeprom_getword(struct rl_softc *sc, int addr, u_int16_t *dest)
353{
354	int			i;
355	u_int16_t		word = 0;
356
357	/*
358	 * Send address of word we want to read.
359	 */
360	re_eeprom_putbyte(sc, addr);
361
362	/*
363	 * Start reading bits from EEPROM.
364	 */
365	for (i = 0x8000; i; i >>= 1) {
366		EE_SET(RL_EE_CLK);
367		DELAY(100);
368		if (CSR_READ_1(sc, RL_EECMD) & RL_EE_DATAOUT)
369			word |= i;
370		EE_CLR(RL_EE_CLK);
371		DELAY(100);
372	}
373
374	*dest = word;
375}
376
377/*
378 * Read a sequence of words from the EEPROM.
379 */
380static void
381re_read_eeprom(struct rl_softc *sc, caddr_t dest, int off, int cnt)
382{
383	int			i;
384	u_int16_t		word = 0, *ptr;
385
386	CSR_SETBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM);
387
388        DELAY(100);
389
390	for (i = 0; i < cnt; i++) {
391		CSR_SETBIT_1(sc, RL_EECMD, RL_EE_SEL);
392		re_eeprom_getword(sc, off + i, &word);
393		CSR_CLRBIT_1(sc, RL_EECMD, RL_EE_SEL);
394		ptr = (u_int16_t *)(dest + (i * 2));
395                *ptr = word;
396	}
397
398	CSR_CLRBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM);
399}
400
401static int
402re_gmii_readreg(device_t dev, int phy, int reg)
403{
404	struct rl_softc		*sc;
405	u_int32_t		rval;
406	int			i;
407
408	if (phy != 1)
409		return (0);
410
411	sc = device_get_softc(dev);
412
413	/* Let the rgephy driver read the GMEDIASTAT register */
414
415	if (reg == RL_GMEDIASTAT) {
416		rval = CSR_READ_1(sc, RL_GMEDIASTAT);
417		return (rval);
418	}
419
420	CSR_WRITE_4(sc, RL_PHYAR, reg << 16);
421
422	for (i = 0; i < RL_PHY_TIMEOUT; i++) {
423		rval = CSR_READ_4(sc, RL_PHYAR);
424		if (rval & RL_PHYAR_BUSY)
425			break;
426		DELAY(100);
427	}
428
429	if (i == RL_PHY_TIMEOUT) {
430		device_printf(sc->rl_dev, "PHY read failed\n");
431		return (0);
432	}
433
434	return (rval & RL_PHYAR_PHYDATA);
435}
436
437static int
438re_gmii_writereg(device_t dev, int phy, int reg, int data)
439{
440	struct rl_softc		*sc;
441	u_int32_t		rval;
442	int			i;
443
444	sc = device_get_softc(dev);
445
446	CSR_WRITE_4(sc, RL_PHYAR, (reg << 16) |
447	    (data & RL_PHYAR_PHYDATA) | RL_PHYAR_BUSY);
448
449	for (i = 0; i < RL_PHY_TIMEOUT; i++) {
450		rval = CSR_READ_4(sc, RL_PHYAR);
451		if (!(rval & RL_PHYAR_BUSY))
452			break;
453		DELAY(100);
454	}
455
456	if (i == RL_PHY_TIMEOUT) {
457		device_printf(sc->rl_dev, "PHY write failed\n");
458		return (0);
459	}
460
461	return (0);
462}
463
464static int
465re_miibus_readreg(device_t dev, int phy, int reg)
466{
467	struct rl_softc		*sc;
468	u_int16_t		rval = 0;
469	u_int16_t		re8139_reg = 0;
470
471	sc = device_get_softc(dev);
472
473	if (sc->rl_type == RL_8169) {
474		rval = re_gmii_readreg(dev, phy, reg);
475		return (rval);
476	}
477
478	/* Pretend the internal PHY is only at address 0 */
479	if (phy) {
480		return (0);
481	}
482	switch (reg) {
483	case MII_BMCR:
484		re8139_reg = RL_BMCR;
485		break;
486	case MII_BMSR:
487		re8139_reg = RL_BMSR;
488		break;
489	case MII_ANAR:
490		re8139_reg = RL_ANAR;
491		break;
492	case MII_ANER:
493		re8139_reg = RL_ANER;
494		break;
495	case MII_ANLPAR:
496		re8139_reg = RL_LPAR;
497		break;
498	case MII_PHYIDR1:
499	case MII_PHYIDR2:
500		return (0);
501	/*
502	 * Allow the rlphy driver to read the media status
503	 * register. If we have a link partner which does not
504	 * support NWAY, this is the register which will tell
505	 * us the results of parallel detection.
506	 */
507	case RL_MEDIASTAT:
508		rval = CSR_READ_1(sc, RL_MEDIASTAT);
509		return (rval);
510	default:
511		device_printf(sc->rl_dev, "bad phy register\n");
512		return (0);
513	}
514	rval = CSR_READ_2(sc, re8139_reg);
515	if (sc->rl_type == RL_8139CPLUS && re8139_reg == RL_BMCR) {
516		/* 8139C+ has different bit layout. */
517		rval &= ~(BMCR_LOOP | BMCR_ISO);
518	}
519	return (rval);
520}
521
522static int
523re_miibus_writereg(device_t dev, int phy, int reg, int data)
524{
525	struct rl_softc		*sc;
526	u_int16_t		re8139_reg = 0;
527	int			rval = 0;
528
529	sc = device_get_softc(dev);
530
531	if (sc->rl_type == RL_8169) {
532		rval = re_gmii_writereg(dev, phy, reg, data);
533		return (rval);
534	}
535
536	/* Pretend the internal PHY is only at address 0 */
537	if (phy)
538		return (0);
539
540	switch (reg) {
541	case MII_BMCR:
542		re8139_reg = RL_BMCR;
543		if (sc->rl_type == RL_8139CPLUS) {
544			/* 8139C+ has different bit layout. */
545			data &= ~(BMCR_LOOP | BMCR_ISO);
546		}
547		break;
548	case MII_BMSR:
549		re8139_reg = RL_BMSR;
550		break;
551	case MII_ANAR:
552		re8139_reg = RL_ANAR;
553		break;
554	case MII_ANER:
555		re8139_reg = RL_ANER;
556		break;
557	case MII_ANLPAR:
558		re8139_reg = RL_LPAR;
559		break;
560	case MII_PHYIDR1:
561	case MII_PHYIDR2:
562		return (0);
563		break;
564	default:
565		device_printf(sc->rl_dev, "bad phy register\n");
566		return (0);
567	}
568	CSR_WRITE_2(sc, re8139_reg, data);
569	return (0);
570}
571
572static void
573re_miibus_statchg(device_t dev)
574{
575	struct rl_softc		*sc;
576	struct ifnet		*ifp;
577	struct mii_data		*mii;
578
579	sc = device_get_softc(dev);
580	mii = device_get_softc(sc->rl_miibus);
581	ifp = sc->rl_ifp;
582	if (mii == NULL || ifp == NULL ||
583	    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
584		return;
585
586	sc->rl_flags &= ~RL_FLAG_LINK;
587	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
588	    (IFM_ACTIVE | IFM_AVALID)) {
589		switch (IFM_SUBTYPE(mii->mii_media_active)) {
590		case IFM_10_T:
591		case IFM_100_TX:
592			sc->rl_flags |= RL_FLAG_LINK;
593			break;
594		case IFM_1000_T:
595			if ((sc->rl_flags & RL_FLAG_FASTETHER) != 0)
596				break;
597			sc->rl_flags |= RL_FLAG_LINK;
598			break;
599		default:
600			break;
601		}
602	}
603	/*
604	 * RealTek controllers does not provide any interface to
605	 * Tx/Rx MACs for resolved speed, duplex and flow-control
606	 * parameters.
607	 */
608}
609
610/*
611 * Set the RX configuration and 64-bit multicast hash filter.
612 */
613static void
614re_set_rxmode(struct rl_softc *sc)
615{
616	struct ifnet		*ifp;
617	struct ifmultiaddr	*ifma;
618	uint32_t		hashes[2] = { 0, 0 };
619	uint32_t		h, rxfilt;
620
621	RL_LOCK_ASSERT(sc);
622
623	ifp = sc->rl_ifp;
624
625	rxfilt = RL_RXCFG_CONFIG | RL_RXCFG_RX_INDIV | RL_RXCFG_RX_BROAD;
626
627	if (ifp->if_flags & (IFF_ALLMULTI | IFF_PROMISC)) {
628		if (ifp->if_flags & IFF_PROMISC)
629			rxfilt |= RL_RXCFG_RX_ALLPHYS;
630		/*
631		 * Unlike other hardwares, we have to explicitly set
632		 * RL_RXCFG_RX_MULTI to receive multicast frames in
633		 * promiscuous mode.
634		 */
635		rxfilt |= RL_RXCFG_RX_MULTI;
636		hashes[0] = hashes[1] = 0xffffffff;
637		goto done;
638	}
639
640	IF_ADDR_LOCK(ifp);
641	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
642		if (ifma->ifma_addr->sa_family != AF_LINK)
643			continue;
644		h = ether_crc32_be(LLADDR((struct sockaddr_dl *)
645		    ifma->ifma_addr), ETHER_ADDR_LEN) >> 26;
646		if (h < 32)
647			hashes[0] |= (1 << h);
648		else
649			hashes[1] |= (1 << (h - 32));
650	}
651	IF_ADDR_UNLOCK(ifp);
652
653	if (hashes[0] != 0 || hashes[1] != 0) {
654		/*
655		 * For some unfathomable reason, RealTek decided to
656		 * reverse the order of the multicast hash registers
657		 * in the PCI Express parts.  This means we have to
658		 * write the hash pattern in reverse order for those
659		 * devices.
660		 */
661		if ((sc->rl_flags & RL_FLAG_PCIE) != 0) {
662			h = bswap32(hashes[0]);
663			hashes[0] = bswap32(hashes[1]);
664			hashes[1] = h;
665		}
666		rxfilt |= RL_RXCFG_RX_MULTI;
667	}
668
669done:
670	CSR_WRITE_4(sc, RL_MAR0, hashes[0]);
671	CSR_WRITE_4(sc, RL_MAR4, hashes[1]);
672	CSR_WRITE_4(sc, RL_RXCFG, rxfilt);
673}
674
675static void
676re_reset(struct rl_softc *sc)
677{
678	int			i;
679
680	RL_LOCK_ASSERT(sc);
681
682	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RESET);
683
684	for (i = 0; i < RL_TIMEOUT; i++) {
685		DELAY(10);
686		if (!(CSR_READ_1(sc, RL_COMMAND) & RL_CMD_RESET))
687			break;
688	}
689	if (i == RL_TIMEOUT)
690		device_printf(sc->rl_dev, "reset never completed!\n");
691
692	if ((sc->rl_flags & RL_FLAG_MACRESET) != 0)
693		CSR_WRITE_1(sc, 0x82, 1);
694	if (sc->rl_hwrev == RL_HWREV_8169S)
695		re_gmii_writereg(sc->rl_dev, 1, 0x0b, 0);
696}
697
698#ifdef RE_DIAG
699
700/*
701 * The following routine is designed to test for a defect on some
702 * 32-bit 8169 cards. Some of these NICs have the REQ64# and ACK64#
703 * lines connected to the bus, however for a 32-bit only card, they
704 * should be pulled high. The result of this defect is that the
705 * NIC will not work right if you plug it into a 64-bit slot: DMA
706 * operations will be done with 64-bit transfers, which will fail
707 * because the 64-bit data lines aren't connected.
708 *
709 * There's no way to work around this (short of talking a soldering
710 * iron to the board), however we can detect it. The method we use
711 * here is to put the NIC into digital loopback mode, set the receiver
712 * to promiscuous mode, and then try to send a frame. We then compare
713 * the frame data we sent to what was received. If the data matches,
714 * then the NIC is working correctly, otherwise we know the user has
715 * a defective NIC which has been mistakenly plugged into a 64-bit PCI
716 * slot. In the latter case, there's no way the NIC can work correctly,
717 * so we print out a message on the console and abort the device attach.
718 */
719
720static int
721re_diag(struct rl_softc *sc)
722{
723	struct ifnet		*ifp = sc->rl_ifp;
724	struct mbuf		*m0;
725	struct ether_header	*eh;
726	struct rl_desc		*cur_rx;
727	u_int16_t		status;
728	u_int32_t		rxstat;
729	int			total_len, i, error = 0, phyaddr;
730	u_int8_t		dst[] = { 0x00, 'h', 'e', 'l', 'l', 'o' };
731	u_int8_t		src[] = { 0x00, 'w', 'o', 'r', 'l', 'd' };
732
733	/* Allocate a single mbuf */
734	MGETHDR(m0, M_DONTWAIT, MT_DATA);
735	if (m0 == NULL)
736		return (ENOBUFS);
737
738	RL_LOCK(sc);
739
740	/*
741	 * Initialize the NIC in test mode. This sets the chip up
742	 * so that it can send and receive frames, but performs the
743	 * following special functions:
744	 * - Puts receiver in promiscuous mode
745	 * - Enables digital loopback mode
746	 * - Leaves interrupts turned off
747	 */
748
749	ifp->if_flags |= IFF_PROMISC;
750	sc->rl_testmode = 1;
751	re_init_locked(sc);
752	sc->rl_flags |= RL_FLAG_LINK;
753	if (sc->rl_type == RL_8169)
754		phyaddr = 1;
755	else
756		phyaddr = 0;
757
758	re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_RESET);
759	for (i = 0; i < RL_TIMEOUT; i++) {
760		status = re_miibus_readreg(sc->rl_dev, phyaddr, MII_BMCR);
761		if (!(status & BMCR_RESET))
762			break;
763	}
764
765	re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_LOOP);
766	CSR_WRITE_2(sc, RL_ISR, RL_INTRS);
767
768	DELAY(100000);
769
770	/* Put some data in the mbuf */
771
772	eh = mtod(m0, struct ether_header *);
773	bcopy ((char *)&dst, eh->ether_dhost, ETHER_ADDR_LEN);
774	bcopy ((char *)&src, eh->ether_shost, ETHER_ADDR_LEN);
775	eh->ether_type = htons(ETHERTYPE_IP);
776	m0->m_pkthdr.len = m0->m_len = ETHER_MIN_LEN - ETHER_CRC_LEN;
777
778	/*
779	 * Queue the packet, start transmission.
780	 * Note: IF_HANDOFF() ultimately calls re_start() for us.
781	 */
782
783	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
784	RL_UNLOCK(sc);
785	/* XXX: re_diag must not be called when in ALTQ mode */
786	IF_HANDOFF(&ifp->if_snd, m0, ifp);
787	RL_LOCK(sc);
788	m0 = NULL;
789
790	/* Wait for it to propagate through the chip */
791
792	DELAY(100000);
793	for (i = 0; i < RL_TIMEOUT; i++) {
794		status = CSR_READ_2(sc, RL_ISR);
795		CSR_WRITE_2(sc, RL_ISR, status);
796		if ((status & (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK)) ==
797		    (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK))
798			break;
799		DELAY(10);
800	}
801
802	if (i == RL_TIMEOUT) {
803		device_printf(sc->rl_dev,
804		    "diagnostic failed, failed to receive packet in"
805		    " loopback mode\n");
806		error = EIO;
807		goto done;
808	}
809
810	/*
811	 * The packet should have been dumped into the first
812	 * entry in the RX DMA ring. Grab it from there.
813	 */
814
815	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
816	    sc->rl_ldata.rl_rx_list_map,
817	    BUS_DMASYNC_POSTREAD);
818	bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag,
819	    sc->rl_ldata.rl_rx_desc[0].rx_dmamap,
820	    BUS_DMASYNC_POSTREAD);
821	bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag,
822	    sc->rl_ldata.rl_rx_desc[0].rx_dmamap);
823
824	m0 = sc->rl_ldata.rl_rx_desc[0].rx_m;
825	sc->rl_ldata.rl_rx_desc[0].rx_m = NULL;
826	eh = mtod(m0, struct ether_header *);
827
828	cur_rx = &sc->rl_ldata.rl_rx_list[0];
829	total_len = RL_RXBYTES(cur_rx);
830	rxstat = le32toh(cur_rx->rl_cmdstat);
831
832	if (total_len != ETHER_MIN_LEN) {
833		device_printf(sc->rl_dev,
834		    "diagnostic failed, received short packet\n");
835		error = EIO;
836		goto done;
837	}
838
839	/* Test that the received packet data matches what we sent. */
840
841	if (bcmp((char *)&eh->ether_dhost, (char *)&dst, ETHER_ADDR_LEN) ||
842	    bcmp((char *)&eh->ether_shost, (char *)&src, ETHER_ADDR_LEN) ||
843	    ntohs(eh->ether_type) != ETHERTYPE_IP) {
844		device_printf(sc->rl_dev, "WARNING, DMA FAILURE!\n");
845		device_printf(sc->rl_dev, "expected TX data: %6D/%6D/0x%x\n",
846		    dst, ":", src, ":", ETHERTYPE_IP);
847		device_printf(sc->rl_dev, "received RX data: %6D/%6D/0x%x\n",
848		    eh->ether_dhost, ":",  eh->ether_shost, ":",
849		    ntohs(eh->ether_type));
850		device_printf(sc->rl_dev, "You may have a defective 32-bit "
851		    "NIC plugged into a 64-bit PCI slot.\n");
852		device_printf(sc->rl_dev, "Please re-install the NIC in a "
853		    "32-bit slot for proper operation.\n");
854		device_printf(sc->rl_dev, "Read the re(4) man page for more "
855		    "details.\n");
856		error = EIO;
857	}
858
859done:
860	/* Turn interface off, release resources */
861
862	sc->rl_testmode = 0;
863	sc->rl_flags &= ~RL_FLAG_LINK;
864	ifp->if_flags &= ~IFF_PROMISC;
865	re_stop(sc);
866	if (m0 != NULL)
867		m_freem(m0);
868
869	RL_UNLOCK(sc);
870
871	return (error);
872}
873
874#endif
875
876/*
877 * Probe for a RealTek 8139C+/8169/8110 chip. Check the PCI vendor and device
878 * IDs against our list and return a device name if we find a match.
879 */
880static int
881re_probe(device_t dev)
882{
883	struct rl_type		*t;
884	uint16_t		devid, vendor;
885	uint16_t		revid, sdevid;
886	int			i;
887
888	vendor = pci_get_vendor(dev);
889	devid = pci_get_device(dev);
890	revid = pci_get_revid(dev);
891	sdevid = pci_get_subdevice(dev);
892
893	if (vendor == LINKSYS_VENDORID && devid == LINKSYS_DEVICEID_EG1032) {
894		if (sdevid != LINKSYS_SUBDEVICE_EG1032_REV3) {
895			/*
896			 * Only attach to rev. 3 of the Linksys EG1032 adapter.
897			 * Rev. 2 is supported by sk(4).
898			 */
899			return (ENXIO);
900		}
901	}
902
903	if (vendor == RT_VENDORID && devid == RT_DEVICEID_8139) {
904		if (revid != 0x20) {
905			/* 8139, let rl(4) take care of this device. */
906			return (ENXIO);
907		}
908	}
909
910	t = re_devs;
911	for (i = 0; i < sizeof(re_devs) / sizeof(re_devs[0]); i++, t++) {
912		if (vendor == t->rl_vid && devid == t->rl_did) {
913			device_set_desc(dev, t->rl_name);
914			return (BUS_PROBE_DEFAULT);
915		}
916	}
917
918	return (ENXIO);
919}
920
921/*
922 * Map a single buffer address.
923 */
924
925static void
926re_dma_map_addr(void *arg, bus_dma_segment_t *segs, int nseg, int error)
927{
928	bus_addr_t		*addr;
929
930	if (error)
931		return;
932
933	KASSERT(nseg == 1, ("too many DMA segments, %d should be 1", nseg));
934	addr = arg;
935	*addr = segs->ds_addr;
936}
937
938static int
939re_allocmem(device_t dev, struct rl_softc *sc)
940{
941	bus_size_t		rx_list_size, tx_list_size;
942	int			error;
943	int			i;
944
945	rx_list_size = sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc);
946	tx_list_size = sc->rl_ldata.rl_tx_desc_cnt * sizeof(struct rl_desc);
947
948	/*
949	 * Allocate the parent bus DMA tag appropriate for PCI.
950	 * In order to use DAC, RL_CPLUSCMD_PCI_DAC bit of RL_CPLUS_CMD
951	 * register should be set. However some RealTek chips are known
952	 * to be buggy on DAC handling, therefore disable DAC by limiting
953	 * DMA address space to 32bit. PCIe variants of RealTek chips
954	 * may not have the limitation but I took safer path.
955	 */
956	error = bus_dma_tag_create(bus_get_dma_tag(dev), 1, 0,
957	    BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL, NULL,
958	    BUS_SPACE_MAXSIZE_32BIT, 0, BUS_SPACE_MAXSIZE_32BIT, 0,
959	    NULL, NULL, &sc->rl_parent_tag);
960	if (error) {
961		device_printf(dev, "could not allocate parent DMA tag\n");
962		return (error);
963	}
964
965	/*
966	 * Allocate map for TX mbufs.
967	 */
968	error = bus_dma_tag_create(sc->rl_parent_tag, 1, 0,
969	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL,
970	    NULL, MCLBYTES * RL_NTXSEGS, RL_NTXSEGS, 4096, 0,
971	    NULL, NULL, &sc->rl_ldata.rl_tx_mtag);
972	if (error) {
973		device_printf(dev, "could not allocate TX DMA tag\n");
974		return (error);
975	}
976
977	/*
978	 * Allocate map for RX mbufs.
979	 */
980
981	error = bus_dma_tag_create(sc->rl_parent_tag, sizeof(uint64_t), 0,
982	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
983	    MCLBYTES, 1, MCLBYTES, 0, NULL, NULL, &sc->rl_ldata.rl_rx_mtag);
984	if (error) {
985		device_printf(dev, "could not allocate RX DMA tag\n");
986		return (error);
987	}
988
989	/*
990	 * Allocate map for TX descriptor list.
991	 */
992	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
993	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
994	    NULL, tx_list_size, 1, tx_list_size, 0,
995	    NULL, NULL, &sc->rl_ldata.rl_tx_list_tag);
996	if (error) {
997		device_printf(dev, "could not allocate TX DMA ring tag\n");
998		return (error);
999	}
1000
1001	/* Allocate DMA'able memory for the TX ring */
1002
1003	error = bus_dmamem_alloc(sc->rl_ldata.rl_tx_list_tag,
1004	    (void **)&sc->rl_ldata.rl_tx_list,
1005	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1006	    &sc->rl_ldata.rl_tx_list_map);
1007	if (error) {
1008		device_printf(dev, "could not allocate TX DMA ring\n");
1009		return (error);
1010	}
1011
1012	/* Load the map for the TX ring. */
1013
1014	sc->rl_ldata.rl_tx_list_addr = 0;
1015	error = bus_dmamap_load(sc->rl_ldata.rl_tx_list_tag,
1016	     sc->rl_ldata.rl_tx_list_map, sc->rl_ldata.rl_tx_list,
1017	     tx_list_size, re_dma_map_addr,
1018	     &sc->rl_ldata.rl_tx_list_addr, BUS_DMA_NOWAIT);
1019	if (error != 0 || sc->rl_ldata.rl_tx_list_addr == 0) {
1020		device_printf(dev, "could not load TX DMA ring\n");
1021		return (ENOMEM);
1022	}
1023
1024	/* Create DMA maps for TX buffers */
1025
1026	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
1027		error = bus_dmamap_create(sc->rl_ldata.rl_tx_mtag, 0,
1028		    &sc->rl_ldata.rl_tx_desc[i].tx_dmamap);
1029		if (error) {
1030			device_printf(dev, "could not create DMA map for TX\n");
1031			return (error);
1032		}
1033	}
1034
1035	/*
1036	 * Allocate map for RX descriptor list.
1037	 */
1038	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
1039	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1040	    NULL, rx_list_size, 1, rx_list_size, 0,
1041	    NULL, NULL, &sc->rl_ldata.rl_rx_list_tag);
1042	if (error) {
1043		device_printf(dev, "could not create RX DMA ring tag\n");
1044		return (error);
1045	}
1046
1047	/* Allocate DMA'able memory for the RX ring */
1048
1049	error = bus_dmamem_alloc(sc->rl_ldata.rl_rx_list_tag,
1050	    (void **)&sc->rl_ldata.rl_rx_list,
1051	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1052	    &sc->rl_ldata.rl_rx_list_map);
1053	if (error) {
1054		device_printf(dev, "could not allocate RX DMA ring\n");
1055		return (error);
1056	}
1057
1058	/* Load the map for the RX ring. */
1059
1060	sc->rl_ldata.rl_rx_list_addr = 0;
1061	error = bus_dmamap_load(sc->rl_ldata.rl_rx_list_tag,
1062	     sc->rl_ldata.rl_rx_list_map, sc->rl_ldata.rl_rx_list,
1063	     rx_list_size, re_dma_map_addr,
1064	     &sc->rl_ldata.rl_rx_list_addr, BUS_DMA_NOWAIT);
1065	if (error != 0 || sc->rl_ldata.rl_rx_list_addr == 0) {
1066		device_printf(dev, "could not load RX DMA ring\n");
1067		return (ENOMEM);
1068	}
1069
1070	/* Create DMA maps for RX buffers */
1071
1072	error = bus_dmamap_create(sc->rl_ldata.rl_rx_mtag, 0,
1073	    &sc->rl_ldata.rl_rx_sparemap);
1074	if (error) {
1075		device_printf(dev, "could not create spare DMA map for RX\n");
1076		return (error);
1077	}
1078	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1079		error = bus_dmamap_create(sc->rl_ldata.rl_rx_mtag, 0,
1080		    &sc->rl_ldata.rl_rx_desc[i].rx_dmamap);
1081		if (error) {
1082			device_printf(dev, "could not create DMA map for RX\n");
1083			return (error);
1084		}
1085	}
1086
1087	return (0);
1088}
1089
1090/*
1091 * Attach the interface. Allocate softc structures, do ifmedia
1092 * setup and ethernet/BPF attach.
1093 */
1094static int
1095re_attach(device_t dev)
1096{
1097	u_char			eaddr[ETHER_ADDR_LEN];
1098	u_int16_t		as[ETHER_ADDR_LEN / 2];
1099	struct rl_softc		*sc;
1100	struct ifnet		*ifp;
1101	struct rl_hwrev		*hw_rev;
1102	int			hwrev;
1103	u_int16_t		devid, re_did = 0;
1104	int			error = 0, rid, i;
1105	int			msic, reg;
1106	uint8_t			cfg;
1107
1108	sc = device_get_softc(dev);
1109	sc->rl_dev = dev;
1110
1111	mtx_init(&sc->rl_mtx, device_get_nameunit(dev), MTX_NETWORK_LOCK,
1112	    MTX_DEF);
1113	callout_init_mtx(&sc->rl_stat_callout, &sc->rl_mtx, 0);
1114
1115	/*
1116	 * Map control/status registers.
1117	 */
1118	pci_enable_busmaster(dev);
1119
1120	devid = pci_get_device(dev);
1121	/* Prefer memory space register mapping over IO space. */
1122	sc->rl_res_id = PCIR_BAR(1);
1123	sc->rl_res_type = SYS_RES_MEMORY;
1124	/* RTL8168/8101E seems to use different BARs. */
1125	if (devid == RT_DEVICEID_8168 || devid == RT_DEVICEID_8101E)
1126		sc->rl_res_id = PCIR_BAR(2);
1127	sc->rl_res = bus_alloc_resource_any(dev, sc->rl_res_type,
1128	    &sc->rl_res_id, RF_ACTIVE);
1129
1130	if (sc->rl_res == NULL) {
1131		sc->rl_res_id = PCIR_BAR(0);
1132		sc->rl_res_type = SYS_RES_IOPORT;
1133		sc->rl_res = bus_alloc_resource_any(dev, sc->rl_res_type,
1134		    &sc->rl_res_id, RF_ACTIVE);
1135		if (sc->rl_res == NULL) {
1136			device_printf(dev, "couldn't map ports/memory\n");
1137			error = ENXIO;
1138			goto fail;
1139		}
1140	}
1141
1142	sc->rl_btag = rman_get_bustag(sc->rl_res);
1143	sc->rl_bhandle = rman_get_bushandle(sc->rl_res);
1144
1145	msic = 0;
1146	if (pci_find_extcap(dev, PCIY_EXPRESS, &reg) == 0) {
1147		sc->rl_flags |= RL_FLAG_PCIE;
1148		msic = pci_msi_count(dev);
1149		if (bootverbose)
1150			device_printf(dev, "MSI count : %d\n", msic);
1151	}
1152	if (msic == RL_MSI_MESSAGES  && msi_disable == 0) {
1153		if (pci_alloc_msi(dev, &msic) == 0) {
1154			if (msic == RL_MSI_MESSAGES) {
1155				device_printf(dev, "Using %d MSI messages\n",
1156				    msic);
1157				sc->rl_flags |= RL_FLAG_MSI;
1158				/* Explicitly set MSI enable bit. */
1159				CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1160				cfg = CSR_READ_1(sc, RL_CFG2);
1161				cfg |= RL_CFG2_MSI;
1162				CSR_WRITE_1(sc, RL_CFG2, cfg);
1163				CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1164			} else
1165				pci_release_msi(dev);
1166		}
1167	}
1168
1169	/* Allocate interrupt */
1170	if ((sc->rl_flags & RL_FLAG_MSI) == 0) {
1171		rid = 0;
1172		sc->rl_irq[0] = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
1173		    RF_SHAREABLE | RF_ACTIVE);
1174		if (sc->rl_irq[0] == NULL) {
1175			device_printf(dev, "couldn't allocate IRQ resources\n");
1176			error = ENXIO;
1177			goto fail;
1178		}
1179	} else {
1180		for (i = 0, rid = 1; i < RL_MSI_MESSAGES; i++, rid++) {
1181			sc->rl_irq[i] = bus_alloc_resource_any(dev,
1182			    SYS_RES_IRQ, &rid, RF_ACTIVE);
1183			if (sc->rl_irq[i] == NULL) {
1184				device_printf(dev,
1185				    "couldn't llocate IRQ resources for "
1186				    "message %d\n", rid);
1187				error = ENXIO;
1188				goto fail;
1189			}
1190		}
1191	}
1192
1193	if ((sc->rl_flags & RL_FLAG_MSI) == 0) {
1194		CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1195		cfg = CSR_READ_1(sc, RL_CFG2);
1196		if ((cfg & RL_CFG2_MSI) != 0) {
1197			device_printf(dev, "turning off MSI enable bit.\n");
1198			cfg &= ~RL_CFG2_MSI;
1199			CSR_WRITE_1(sc, RL_CFG2, cfg);
1200		}
1201		CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1202	}
1203
1204	/* Reset the adapter. */
1205	RL_LOCK(sc);
1206	re_reset(sc);
1207	RL_UNLOCK(sc);
1208
1209	hw_rev = re_hwrevs;
1210	hwrev = CSR_READ_4(sc, RL_TXCFG);
1211	switch (hwrev & 0x70000000) {
1212	case 0x00000000:
1213	case 0x10000000:
1214		device_printf(dev, "Chip rev. 0x%08x\n", hwrev & 0xfc800000);
1215		hwrev &= (RL_TXCFG_HWREV | 0x80000000);
1216		break;
1217	default:
1218		device_printf(dev, "Chip rev. 0x%08x\n", hwrev & 0x7c800000);
1219		hwrev &= RL_TXCFG_HWREV;
1220		break;
1221	}
1222	device_printf(dev, "MAC rev. 0x%08x\n", hwrev & 0x00700000);
1223	while (hw_rev->rl_desc != NULL) {
1224		if (hw_rev->rl_rev == hwrev) {
1225			sc->rl_type = hw_rev->rl_type;
1226			sc->rl_hwrev = hw_rev->rl_rev;
1227			break;
1228		}
1229		hw_rev++;
1230	}
1231	if (hw_rev->rl_desc == NULL) {
1232		device_printf(dev, "Unknown H/W revision: 0x%08x\n", hwrev);
1233		error = ENXIO;
1234		goto fail;
1235	}
1236
1237	switch (hw_rev->rl_rev) {
1238	case RL_HWREV_8139CPLUS:
1239		sc->rl_flags |= RL_FLAG_NOJUMBO | RL_FLAG_FASTETHER;
1240		break;
1241	case RL_HWREV_8100E:
1242	case RL_HWREV_8101E:
1243		sc->rl_flags |= RL_FLAG_NOJUMBO | RL_FLAG_PHYWAKE |
1244		    RL_FLAG_FASTETHER;
1245		break;
1246	case RL_HWREV_8102E:
1247	case RL_HWREV_8102EL:
1248		sc->rl_flags |= RL_FLAG_NOJUMBO | RL_FLAG_PHYWAKE |
1249		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
1250		    RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP;
1251		break;
1252	case RL_HWREV_8168_SPIN1:
1253	case RL_HWREV_8168_SPIN2:
1254		sc->rl_flags |= RL_FLAG_WOLRXENB;
1255		/* FALLTHROUGH */
1256	case RL_HWREV_8168_SPIN3:
1257		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_MACSTAT;
1258		break;
1259	case RL_HWREV_8168C_SPIN2:
1260		sc->rl_flags |= RL_FLAG_MACSLEEP;
1261		/* FALLTHROUGH */
1262	case RL_HWREV_8168C:
1263		if ((hwrev & 0x00700000) == 0x00200000)
1264			sc->rl_flags |= RL_FLAG_MACSLEEP;
1265		/* FALLTHROUGH */
1266	case RL_HWREV_8168CP:
1267	case RL_HWREV_8168D:
1268		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
1269		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP;
1270		/*
1271		 * These controllers support jumbo frame but it seems
1272		 * that enabling it requires touching additional magic
1273		 * registers. Depending on MAC revisions some
1274		 * controllers need to disable checksum offload. So
1275		 * disable jumbo frame until I have better idea what
1276		 * it really requires to make it support.
1277		 * RTL8168C/CP : supports up to 6KB jumbo frame.
1278		 * RTL8111C/CP : supports up to 9KB jumbo frame.
1279		 */
1280		sc->rl_flags |= RL_FLAG_NOJUMBO;
1281		break;
1282	case RL_HWREV_8169_8110SB:
1283	case RL_HWREV_8169_8110SBL:
1284	case RL_HWREV_8169_8110SC:
1285	case RL_HWREV_8169_8110SCE:
1286		sc->rl_flags |= RL_FLAG_PHYWAKE;
1287		/* FALLTHROUGH */
1288	case RL_HWREV_8169:
1289	case RL_HWREV_8169S:
1290	case RL_HWREV_8110S:
1291		sc->rl_flags |= RL_FLAG_MACRESET;
1292		break;
1293	default:
1294		break;
1295	}
1296
1297	/* Enable PME. */
1298	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1299	cfg = CSR_READ_1(sc, RL_CFG1);
1300	cfg |= RL_CFG1_PME;
1301	CSR_WRITE_1(sc, RL_CFG1, cfg);
1302	cfg = CSR_READ_1(sc, RL_CFG5);
1303	cfg &= RL_CFG5_PME_STS;
1304	CSR_WRITE_1(sc, RL_CFG5, cfg);
1305	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1306
1307	if ((sc->rl_flags & RL_FLAG_PAR) != 0) {
1308		/*
1309		 * XXX Should have a better way to extract station
1310		 * address from EEPROM.
1311		 */
1312		for (i = 0; i < ETHER_ADDR_LEN; i++)
1313			eaddr[i] = CSR_READ_1(sc, RL_IDR0 + i);
1314	} else {
1315		sc->rl_eewidth = RL_9356_ADDR_LEN;
1316		re_read_eeprom(sc, (caddr_t)&re_did, 0, 1);
1317		if (re_did != 0x8129)
1318			sc->rl_eewidth = RL_9346_ADDR_LEN;
1319
1320		/*
1321		 * Get station address from the EEPROM.
1322		 */
1323		re_read_eeprom(sc, (caddr_t)as, RL_EE_EADDR, 3);
1324		for (i = 0; i < ETHER_ADDR_LEN / 2; i++)
1325			as[i] = le16toh(as[i]);
1326		bcopy(as, eaddr, sizeof(eaddr));
1327	}
1328
1329	if (sc->rl_type == RL_8169) {
1330		/* Set RX length mask and number of descriptors. */
1331		sc->rl_rxlenmask = RL_RDESC_STAT_GFRAGLEN;
1332		sc->rl_txstart = RL_GTXSTART;
1333		sc->rl_ldata.rl_tx_desc_cnt = RL_8169_TX_DESC_CNT;
1334		sc->rl_ldata.rl_rx_desc_cnt = RL_8169_RX_DESC_CNT;
1335	} else {
1336		/* Set RX length mask and number of descriptors. */
1337		sc->rl_rxlenmask = RL_RDESC_STAT_FRAGLEN;
1338		sc->rl_txstart = RL_TXSTART;
1339		sc->rl_ldata.rl_tx_desc_cnt = RL_8139_TX_DESC_CNT;
1340		sc->rl_ldata.rl_rx_desc_cnt = RL_8139_RX_DESC_CNT;
1341	}
1342
1343	error = re_allocmem(dev, sc);
1344	if (error)
1345		goto fail;
1346
1347	ifp = sc->rl_ifp = if_alloc(IFT_ETHER);
1348	if (ifp == NULL) {
1349		device_printf(dev, "can not if_alloc()\n");
1350		error = ENOSPC;
1351		goto fail;
1352	}
1353
1354	/* Take controller out of deep sleep mode. */
1355	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
1356		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
1357			CSR_WRITE_1(sc, RL_GPIO,
1358			    CSR_READ_1(sc, RL_GPIO) | 0x01);
1359		else
1360			CSR_WRITE_1(sc, RL_GPIO,
1361			    CSR_READ_1(sc, RL_GPIO) & ~0x01);
1362	}
1363
1364	/* Take PHY out of power down mode. */
1365	if ((sc->rl_flags & RL_FLAG_PHYWAKE) != 0) {
1366		re_gmii_writereg(dev, 1, 0x1f, 0);
1367		re_gmii_writereg(dev, 1, 0x0e, 0);
1368	}
1369
1370	/* Do MII setup */
1371	if (mii_phy_probe(dev, &sc->rl_miibus,
1372	    re_ifmedia_upd, re_ifmedia_sts)) {
1373		device_printf(dev, "MII without any phy!\n");
1374		error = ENXIO;
1375		goto fail;
1376	}
1377
1378	ifp->if_softc = sc;
1379	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1380	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1381	ifp->if_ioctl = re_ioctl;
1382	ifp->if_start = re_start;
1383	ifp->if_hwassist = RE_CSUM_FEATURES;
1384	ifp->if_capabilities = IFCAP_HWCSUM;
1385	ifp->if_capenable = ifp->if_capabilities;
1386	ifp->if_init = re_init;
1387	IFQ_SET_MAXLEN(&ifp->if_snd, RL_IFQ_MAXLEN);
1388	ifp->if_snd.ifq_drv_maxlen = RL_IFQ_MAXLEN;
1389	IFQ_SET_READY(&ifp->if_snd);
1390
1391	TASK_INIT(&sc->rl_txtask, 1, re_tx_task, ifp);
1392	TASK_INIT(&sc->rl_inttask, 0, re_int_task, sc);
1393
1394	/*
1395	 * XXX
1396	 * Still have no idea how to make TSO work on 8168C, 8168CP,
1397	 * 8111C and 8111CP.
1398	 */
1399	if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
1400		ifp->if_hwassist |= CSUM_TSO;
1401		ifp->if_capabilities |= IFCAP_TSO4;
1402	}
1403
1404	/*
1405	 * Call MI attach routine.
1406	 */
1407	ether_ifattach(ifp, eaddr);
1408
1409	/* VLAN capability setup */
1410	ifp->if_capabilities |= IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING;
1411	if (ifp->if_capabilities & IFCAP_HWCSUM)
1412		ifp->if_capabilities |= IFCAP_VLAN_HWCSUM;
1413	/* Enable WOL if PM is supported. */
1414	if (pci_find_extcap(sc->rl_dev, PCIY_PMG, &reg) == 0)
1415		ifp->if_capabilities |= IFCAP_WOL;
1416	ifp->if_capenable = ifp->if_capabilities;
1417	/*
1418	 * Don't enable TSO by default. Under certain
1419	 * circumtances the controller generated corrupted
1420	 * packets in TSO size.
1421	 */
1422	ifp->if_hwassist &= ~CSUM_TSO;
1423	ifp->if_capenable &= ~IFCAP_TSO4;
1424#ifdef DEVICE_POLLING
1425	ifp->if_capabilities |= IFCAP_POLLING;
1426#endif
1427	/*
1428	 * Tell the upper layer(s) we support long frames.
1429	 * Must appear after the call to ether_ifattach() because
1430	 * ether_ifattach() sets ifi_hdrlen to the default value.
1431	 */
1432	ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
1433
1434#ifdef RE_DIAG
1435	/*
1436	 * Perform hardware diagnostic on the original RTL8169.
1437	 * Some 32-bit cards were incorrectly wired and would
1438	 * malfunction if plugged into a 64-bit slot.
1439	 */
1440
1441	if (hwrev == RL_HWREV_8169) {
1442		error = re_diag(sc);
1443		if (error) {
1444			device_printf(dev,
1445		    	"attach aborted due to hardware diag failure\n");
1446			ether_ifdetach(ifp);
1447			goto fail;
1448		}
1449	}
1450#endif
1451
1452	/* Hook interrupt last to avoid having to lock softc */
1453	if ((sc->rl_flags & RL_FLAG_MSI) == 0)
1454		error = bus_setup_intr(dev, sc->rl_irq[0],
1455		    INTR_TYPE_NET | INTR_MPSAFE, re_intr, NULL, sc,
1456		    &sc->rl_intrhand[0]);
1457	else {
1458		for (i = 0; i < RL_MSI_MESSAGES; i++) {
1459			error = bus_setup_intr(dev, sc->rl_irq[i],
1460			    INTR_TYPE_NET | INTR_MPSAFE, re_intr, NULL, sc,
1461		    	    &sc->rl_intrhand[i]);
1462			if (error != 0)
1463				break;
1464		}
1465	}
1466	if (error) {
1467		device_printf(dev, "couldn't set up irq\n");
1468		ether_ifdetach(ifp);
1469	}
1470
1471fail:
1472
1473	if (error)
1474		re_detach(dev);
1475
1476	return (error);
1477}
1478
1479/*
1480 * Shutdown hardware and free up resources. This can be called any
1481 * time after the mutex has been initialized. It is called in both
1482 * the error case in attach and the normal detach case so it needs
1483 * to be careful about only freeing resources that have actually been
1484 * allocated.
1485 */
1486static int
1487re_detach(device_t dev)
1488{
1489	struct rl_softc		*sc;
1490	struct ifnet		*ifp;
1491	int			i, rid;
1492
1493	sc = device_get_softc(dev);
1494	ifp = sc->rl_ifp;
1495	KASSERT(mtx_initialized(&sc->rl_mtx), ("re mutex not initialized"));
1496
1497	/* These should only be active if attach succeeded */
1498	if (device_is_attached(dev)) {
1499#ifdef DEVICE_POLLING
1500		if (ifp->if_capenable & IFCAP_POLLING)
1501			ether_poll_deregister(ifp);
1502#endif
1503		RL_LOCK(sc);
1504#if 0
1505		sc->suspended = 1;
1506#endif
1507		re_stop(sc);
1508		RL_UNLOCK(sc);
1509		callout_drain(&sc->rl_stat_callout);
1510		taskqueue_drain(taskqueue_fast, &sc->rl_inttask);
1511		taskqueue_drain(taskqueue_fast, &sc->rl_txtask);
1512		/*
1513		 * Force off the IFF_UP flag here, in case someone
1514		 * still had a BPF descriptor attached to this
1515		 * interface. If they do, ether_ifdetach() will cause
1516		 * the BPF code to try and clear the promisc mode
1517		 * flag, which will bubble down to re_ioctl(),
1518		 * which will try to call re_init() again. This will
1519		 * turn the NIC back on and restart the MII ticker,
1520		 * which will panic the system when the kernel tries
1521		 * to invoke the re_tick() function that isn't there
1522		 * anymore.
1523		 */
1524		ifp->if_flags &= ~IFF_UP;
1525		ether_ifdetach(ifp);
1526	}
1527	if (sc->rl_miibus)
1528		device_delete_child(dev, sc->rl_miibus);
1529	bus_generic_detach(dev);
1530
1531	/*
1532	 * The rest is resource deallocation, so we should already be
1533	 * stopped here.
1534	 */
1535
1536	for (i = 0; i < RL_MSI_MESSAGES; i++) {
1537		if (sc->rl_intrhand[i] != NULL) {
1538			bus_teardown_intr(dev, sc->rl_irq[i],
1539			    sc->rl_intrhand[i]);
1540			sc->rl_intrhand[i] = NULL;
1541		}
1542	}
1543	if (ifp != NULL)
1544		if_free(ifp);
1545	if ((sc->rl_flags & RL_FLAG_MSI) == 0) {
1546		if (sc->rl_irq[0] != NULL) {
1547			bus_release_resource(dev, SYS_RES_IRQ, 0,
1548			    sc->rl_irq[0]);
1549			sc->rl_irq[0] = NULL;
1550		}
1551	} else {
1552		for (i = 0, rid = 1; i < RL_MSI_MESSAGES; i++, rid++) {
1553			if (sc->rl_irq[i] != NULL) {
1554				bus_release_resource(dev, SYS_RES_IRQ, rid,
1555				    sc->rl_irq[i]);
1556				sc->rl_irq[i] = NULL;
1557			}
1558		}
1559		pci_release_msi(dev);
1560	}
1561	if (sc->rl_res)
1562		bus_release_resource(dev, sc->rl_res_type, sc->rl_res_id,
1563		    sc->rl_res);
1564
1565	/* Unload and free the RX DMA ring memory and map */
1566
1567	if (sc->rl_ldata.rl_rx_list_tag) {
1568		bus_dmamap_unload(sc->rl_ldata.rl_rx_list_tag,
1569		    sc->rl_ldata.rl_rx_list_map);
1570		bus_dmamem_free(sc->rl_ldata.rl_rx_list_tag,
1571		    sc->rl_ldata.rl_rx_list,
1572		    sc->rl_ldata.rl_rx_list_map);
1573		bus_dma_tag_destroy(sc->rl_ldata.rl_rx_list_tag);
1574	}
1575
1576	/* Unload and free the TX DMA ring memory and map */
1577
1578	if (sc->rl_ldata.rl_tx_list_tag) {
1579		bus_dmamap_unload(sc->rl_ldata.rl_tx_list_tag,
1580		    sc->rl_ldata.rl_tx_list_map);
1581		bus_dmamem_free(sc->rl_ldata.rl_tx_list_tag,
1582		    sc->rl_ldata.rl_tx_list,
1583		    sc->rl_ldata.rl_tx_list_map);
1584		bus_dma_tag_destroy(sc->rl_ldata.rl_tx_list_tag);
1585	}
1586
1587	/* Destroy all the RX and TX buffer maps */
1588
1589	if (sc->rl_ldata.rl_tx_mtag) {
1590		for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++)
1591			bus_dmamap_destroy(sc->rl_ldata.rl_tx_mtag,
1592			    sc->rl_ldata.rl_tx_desc[i].tx_dmamap);
1593		bus_dma_tag_destroy(sc->rl_ldata.rl_tx_mtag);
1594	}
1595	if (sc->rl_ldata.rl_rx_mtag) {
1596		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++)
1597			bus_dmamap_destroy(sc->rl_ldata.rl_rx_mtag,
1598			    sc->rl_ldata.rl_rx_desc[i].rx_dmamap);
1599		if (sc->rl_ldata.rl_rx_sparemap)
1600			bus_dmamap_destroy(sc->rl_ldata.rl_rx_mtag,
1601			    sc->rl_ldata.rl_rx_sparemap);
1602		bus_dma_tag_destroy(sc->rl_ldata.rl_rx_mtag);
1603	}
1604
1605	/* Unload and free the stats buffer and map */
1606
1607	if (sc->rl_ldata.rl_stag) {
1608		bus_dmamap_unload(sc->rl_ldata.rl_stag,
1609		    sc->rl_ldata.rl_rx_list_map);
1610		bus_dmamem_free(sc->rl_ldata.rl_stag,
1611		    sc->rl_ldata.rl_stats,
1612		    sc->rl_ldata.rl_smap);
1613		bus_dma_tag_destroy(sc->rl_ldata.rl_stag);
1614	}
1615
1616	if (sc->rl_parent_tag)
1617		bus_dma_tag_destroy(sc->rl_parent_tag);
1618
1619	mtx_destroy(&sc->rl_mtx);
1620
1621	return (0);
1622}
1623
1624static __inline void
1625re_discard_rxbuf(struct rl_softc *sc, int idx)
1626{
1627	struct rl_desc		*desc;
1628	struct rl_rxdesc	*rxd;
1629	uint32_t		cmdstat;
1630
1631	rxd = &sc->rl_ldata.rl_rx_desc[idx];
1632	desc = &sc->rl_ldata.rl_rx_list[idx];
1633	desc->rl_vlanctl = 0;
1634	cmdstat = rxd->rx_size;
1635	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
1636		cmdstat |= RL_RDESC_CMD_EOR;
1637	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
1638}
1639
1640static int
1641re_newbuf(struct rl_softc *sc, int idx)
1642{
1643	struct mbuf		*m;
1644	struct rl_rxdesc	*rxd;
1645	bus_dma_segment_t	segs[1];
1646	bus_dmamap_t		map;
1647	struct rl_desc		*desc;
1648	uint32_t		cmdstat;
1649	int			error, nsegs;
1650
1651	m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
1652	if (m == NULL)
1653		return (ENOBUFS);
1654
1655	m->m_len = m->m_pkthdr.len = MCLBYTES;
1656#ifdef RE_FIXUP_RX
1657	/*
1658	 * This is part of an evil trick to deal with non-x86 platforms.
1659	 * The RealTek chip requires RX buffers to be aligned on 64-bit
1660	 * boundaries, but that will hose non-x86 machines. To get around
1661	 * this, we leave some empty space at the start of each buffer
1662	 * and for non-x86 hosts, we copy the buffer back six bytes
1663	 * to achieve word alignment. This is slightly more efficient
1664	 * than allocating a new buffer, copying the contents, and
1665	 * discarding the old buffer.
1666	 */
1667	m_adj(m, RE_ETHER_ALIGN);
1668#endif
1669	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_rx_mtag,
1670	    sc->rl_ldata.rl_rx_sparemap, m, segs, &nsegs, BUS_DMA_NOWAIT);
1671	if (error != 0) {
1672		m_freem(m);
1673		return (ENOBUFS);
1674	}
1675	KASSERT(nsegs == 1, ("%s: %d segment returned!", __func__, nsegs));
1676
1677	rxd = &sc->rl_ldata.rl_rx_desc[idx];
1678	if (rxd->rx_m != NULL) {
1679		bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap,
1680		    BUS_DMASYNC_POSTREAD);
1681		bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap);
1682	}
1683
1684	rxd->rx_m = m;
1685	map = rxd->rx_dmamap;
1686	rxd->rx_dmamap = sc->rl_ldata.rl_rx_sparemap;
1687	rxd->rx_size = segs[0].ds_len;
1688	sc->rl_ldata.rl_rx_sparemap = map;
1689	bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap,
1690	    BUS_DMASYNC_PREREAD);
1691
1692	desc = &sc->rl_ldata.rl_rx_list[idx];
1693	desc->rl_vlanctl = 0;
1694	desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[0].ds_addr));
1695	desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[0].ds_addr));
1696	cmdstat = segs[0].ds_len;
1697	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
1698		cmdstat |= RL_RDESC_CMD_EOR;
1699	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
1700
1701	return (0);
1702}
1703
1704#ifdef RE_FIXUP_RX
1705static __inline void
1706re_fixup_rx(struct mbuf *m)
1707{
1708	int                     i;
1709	uint16_t                *src, *dst;
1710
1711	src = mtod(m, uint16_t *);
1712	dst = src - (RE_ETHER_ALIGN - ETHER_ALIGN) / sizeof *src;
1713
1714	for (i = 0; i < (m->m_len / sizeof(uint16_t) + 1); i++)
1715		*dst++ = *src++;
1716
1717	m->m_data -= RE_ETHER_ALIGN - ETHER_ALIGN;
1718}
1719#endif
1720
1721static int
1722re_tx_list_init(struct rl_softc *sc)
1723{
1724	struct rl_desc		*desc;
1725	int			i;
1726
1727	RL_LOCK_ASSERT(sc);
1728
1729	bzero(sc->rl_ldata.rl_tx_list,
1730	    sc->rl_ldata.rl_tx_desc_cnt * sizeof(struct rl_desc));
1731	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++)
1732		sc->rl_ldata.rl_tx_desc[i].tx_m = NULL;
1733	/* Set EOR. */
1734	desc = &sc->rl_ldata.rl_tx_list[sc->rl_ldata.rl_tx_desc_cnt - 1];
1735	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_EOR);
1736
1737	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
1738	    sc->rl_ldata.rl_tx_list_map,
1739	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1740
1741	sc->rl_ldata.rl_tx_prodidx = 0;
1742	sc->rl_ldata.rl_tx_considx = 0;
1743	sc->rl_ldata.rl_tx_free = sc->rl_ldata.rl_tx_desc_cnt;
1744
1745	return (0);
1746}
1747
1748static int
1749re_rx_list_init(struct rl_softc *sc)
1750{
1751	int			error, i;
1752
1753	bzero(sc->rl_ldata.rl_rx_list,
1754	    sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc));
1755	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1756		sc->rl_ldata.rl_rx_desc[i].rx_m = NULL;
1757		if ((error = re_newbuf(sc, i)) != 0)
1758			return (error);
1759	}
1760
1761	/* Flush the RX descriptors */
1762
1763	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1764	    sc->rl_ldata.rl_rx_list_map,
1765	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
1766
1767	sc->rl_ldata.rl_rx_prodidx = 0;
1768	sc->rl_head = sc->rl_tail = NULL;
1769
1770	return (0);
1771}
1772
1773/*
1774 * RX handler for C+ and 8169. For the gigE chips, we support
1775 * the reception of jumbo frames that have been fragmented
1776 * across multiple 2K mbuf cluster buffers.
1777 */
1778static int
1779re_rxeof(struct rl_softc *sc)
1780{
1781	struct mbuf		*m;
1782	struct ifnet		*ifp;
1783	int			i, total_len;
1784	struct rl_desc		*cur_rx;
1785	u_int32_t		rxstat, rxvlan;
1786	int			maxpkt = 16;
1787
1788	RL_LOCK_ASSERT(sc);
1789
1790	ifp = sc->rl_ifp;
1791
1792	/* Invalidate the descriptor memory */
1793
1794	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1795	    sc->rl_ldata.rl_rx_list_map,
1796	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
1797
1798	for (i = sc->rl_ldata.rl_rx_prodidx; maxpkt > 0;
1799	    i = RL_RX_DESC_NXT(sc, i)) {
1800		cur_rx = &sc->rl_ldata.rl_rx_list[i];
1801		rxstat = le32toh(cur_rx->rl_cmdstat);
1802		if ((rxstat & RL_RDESC_STAT_OWN) != 0)
1803			break;
1804		total_len = rxstat & sc->rl_rxlenmask;
1805		rxvlan = le32toh(cur_rx->rl_vlanctl);
1806		m = sc->rl_ldata.rl_rx_desc[i].rx_m;
1807
1808		if (!(rxstat & RL_RDESC_STAT_EOF)) {
1809			if (re_newbuf(sc, i) != 0) {
1810				/*
1811				 * If this is part of a multi-fragment packet,
1812				 * discard all the pieces.
1813				 */
1814				if (sc->rl_head != NULL) {
1815					m_freem(sc->rl_head);
1816					sc->rl_head = sc->rl_tail = NULL;
1817				}
1818				re_discard_rxbuf(sc, i);
1819				continue;
1820			}
1821			m->m_len = RE_RX_DESC_BUFLEN;
1822			if (sc->rl_head == NULL)
1823				sc->rl_head = sc->rl_tail = m;
1824			else {
1825				m->m_flags &= ~M_PKTHDR;
1826				sc->rl_tail->m_next = m;
1827				sc->rl_tail = m;
1828			}
1829			continue;
1830		}
1831
1832		/*
1833		 * NOTE: for the 8139C+, the frame length field
1834		 * is always 12 bits in size, but for the gigE chips,
1835		 * it is 13 bits (since the max RX frame length is 16K).
1836		 * Unfortunately, all 32 bits in the status word
1837		 * were already used, so to make room for the extra
1838		 * length bit, RealTek took out the 'frame alignment
1839		 * error' bit and shifted the other status bits
1840		 * over one slot. The OWN, EOR, FS and LS bits are
1841		 * still in the same places. We have already extracted
1842		 * the frame length and checked the OWN bit, so rather
1843		 * than using an alternate bit mapping, we shift the
1844		 * status bits one space to the right so we can evaluate
1845		 * them using the 8169 status as though it was in the
1846		 * same format as that of the 8139C+.
1847		 */
1848		if (sc->rl_type == RL_8169)
1849			rxstat >>= 1;
1850
1851		/*
1852		 * if total_len > 2^13-1, both _RXERRSUM and _GIANT will be
1853		 * set, but if CRC is clear, it will still be a valid frame.
1854		 */
1855		if (rxstat & RL_RDESC_STAT_RXERRSUM && !(total_len > 8191 &&
1856		    (rxstat & RL_RDESC_STAT_ERRS) == RL_RDESC_STAT_GIANT)) {
1857			ifp->if_ierrors++;
1858			/*
1859			 * If this is part of a multi-fragment packet,
1860			 * discard all the pieces.
1861			 */
1862			if (sc->rl_head != NULL) {
1863				m_freem(sc->rl_head);
1864				sc->rl_head = sc->rl_tail = NULL;
1865			}
1866			re_discard_rxbuf(sc, i);
1867			continue;
1868		}
1869
1870		/*
1871		 * If allocating a replacement mbuf fails,
1872		 * reload the current one.
1873		 */
1874
1875		if (re_newbuf(sc, i) != 0) {
1876			ifp->if_iqdrops++;
1877			if (sc->rl_head != NULL) {
1878				m_freem(sc->rl_head);
1879				sc->rl_head = sc->rl_tail = NULL;
1880			}
1881			re_discard_rxbuf(sc, i);
1882			continue;
1883		}
1884
1885		if (sc->rl_head != NULL) {
1886			m->m_len = total_len % RE_RX_DESC_BUFLEN;
1887			if (m->m_len == 0)
1888				m->m_len = RE_RX_DESC_BUFLEN;
1889			/*
1890			 * Special case: if there's 4 bytes or less
1891			 * in this buffer, the mbuf can be discarded:
1892			 * the last 4 bytes is the CRC, which we don't
1893			 * care about anyway.
1894			 */
1895			if (m->m_len <= ETHER_CRC_LEN) {
1896				sc->rl_tail->m_len -=
1897				    (ETHER_CRC_LEN - m->m_len);
1898				m_freem(m);
1899			} else {
1900				m->m_len -= ETHER_CRC_LEN;
1901				m->m_flags &= ~M_PKTHDR;
1902				sc->rl_tail->m_next = m;
1903			}
1904			m = sc->rl_head;
1905			sc->rl_head = sc->rl_tail = NULL;
1906			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
1907		} else
1908			m->m_pkthdr.len = m->m_len =
1909			    (total_len - ETHER_CRC_LEN);
1910
1911#ifdef RE_FIXUP_RX
1912		re_fixup_rx(m);
1913#endif
1914		ifp->if_ipackets++;
1915		m->m_pkthdr.rcvif = ifp;
1916
1917		/* Do RX checksumming if enabled */
1918
1919		if (ifp->if_capenable & IFCAP_RXCSUM) {
1920			if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
1921				/* Check IP header checksum */
1922				if (rxstat & RL_RDESC_STAT_PROTOID)
1923					m->m_pkthdr.csum_flags |=
1924					    CSUM_IP_CHECKED;
1925				if (!(rxstat & RL_RDESC_STAT_IPSUMBAD))
1926					m->m_pkthdr.csum_flags |=
1927					    CSUM_IP_VALID;
1928
1929				/* Check TCP/UDP checksum */
1930				if ((RL_TCPPKT(rxstat) &&
1931				    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
1932				    (RL_UDPPKT(rxstat) &&
1933				     !(rxstat & RL_RDESC_STAT_UDPSUMBAD))) {
1934					m->m_pkthdr.csum_flags |=
1935						CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
1936					m->m_pkthdr.csum_data = 0xffff;
1937				}
1938			} else {
1939				/*
1940				 * RTL8168C/RTL816CP/RTL8111C/RTL8111CP
1941				 */
1942				if ((rxstat & RL_RDESC_STAT_PROTOID) &&
1943				    (rxvlan & RL_RDESC_IPV4))
1944					m->m_pkthdr.csum_flags |=
1945					    CSUM_IP_CHECKED;
1946				if (!(rxstat & RL_RDESC_STAT_IPSUMBAD) &&
1947				    (rxvlan & RL_RDESC_IPV4))
1948					m->m_pkthdr.csum_flags |=
1949					    CSUM_IP_VALID;
1950				if (((rxstat & RL_RDESC_STAT_TCP) &&
1951				    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
1952				    ((rxstat & RL_RDESC_STAT_UDP) &&
1953				    !(rxstat & RL_RDESC_STAT_UDPSUMBAD))) {
1954					m->m_pkthdr.csum_flags |=
1955						CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
1956					m->m_pkthdr.csum_data = 0xffff;
1957				}
1958			}
1959		}
1960		maxpkt--;
1961		if (rxvlan & RL_RDESC_VLANCTL_TAG) {
1962			m->m_pkthdr.ether_vtag =
1963			    bswap16((rxvlan & RL_RDESC_VLANCTL_DATA));
1964			m->m_flags |= M_VLANTAG;
1965		}
1966		RL_UNLOCK(sc);
1967		(*ifp->if_input)(ifp, m);
1968		RL_LOCK(sc);
1969	}
1970
1971	/* Flush the RX DMA ring */
1972
1973	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1974	    sc->rl_ldata.rl_rx_list_map,
1975	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
1976
1977	sc->rl_ldata.rl_rx_prodidx = i;
1978
1979	if (maxpkt)
1980		return(EAGAIN);
1981
1982	return(0);
1983}
1984
1985static void
1986re_txeof(struct rl_softc *sc)
1987{
1988	struct ifnet		*ifp;
1989	struct rl_txdesc	*txd;
1990	u_int32_t		txstat;
1991	int			cons;
1992
1993	cons = sc->rl_ldata.rl_tx_considx;
1994	if (cons == sc->rl_ldata.rl_tx_prodidx)
1995		return;
1996
1997	ifp = sc->rl_ifp;
1998	/* Invalidate the TX descriptor list */
1999	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2000	    sc->rl_ldata.rl_tx_list_map,
2001	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2002
2003	for (; cons != sc->rl_ldata.rl_tx_prodidx;
2004	    cons = RL_TX_DESC_NXT(sc, cons)) {
2005		txstat = le32toh(sc->rl_ldata.rl_tx_list[cons].rl_cmdstat);
2006		if (txstat & RL_TDESC_STAT_OWN)
2007			break;
2008		/*
2009		 * We only stash mbufs in the last descriptor
2010		 * in a fragment chain, which also happens to
2011		 * be the only place where the TX status bits
2012		 * are valid.
2013		 */
2014		if (txstat & RL_TDESC_CMD_EOF) {
2015			txd = &sc->rl_ldata.rl_tx_desc[cons];
2016			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
2017			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
2018			bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag,
2019			    txd->tx_dmamap);
2020			KASSERT(txd->tx_m != NULL,
2021			    ("%s: freeing NULL mbufs!", __func__));
2022			m_freem(txd->tx_m);
2023			txd->tx_m = NULL;
2024			if (txstat & (RL_TDESC_STAT_EXCESSCOL|
2025			    RL_TDESC_STAT_COLCNT))
2026				ifp->if_collisions++;
2027			if (txstat & RL_TDESC_STAT_TXERRSUM)
2028				ifp->if_oerrors++;
2029			else
2030				ifp->if_opackets++;
2031		}
2032		sc->rl_ldata.rl_tx_free++;
2033		ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
2034	}
2035	sc->rl_ldata.rl_tx_considx = cons;
2036
2037	/* No changes made to the TX ring, so no flush needed */
2038
2039	if (sc->rl_ldata.rl_tx_free != sc->rl_ldata.rl_tx_desc_cnt) {
2040#ifdef RE_TX_MODERATION
2041		/*
2042		 * If not all descriptors have been reaped yet, reload
2043		 * the timer so that we will eventually get another
2044		 * interrupt that will cause us to re-enter this routine.
2045		 * This is done in case the transmitter has gone idle.
2046		 */
2047		CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2048#endif
2049	} else
2050		sc->rl_watchdog_timer = 0;
2051}
2052
2053static void
2054re_tick(void *xsc)
2055{
2056	struct rl_softc		*sc;
2057	struct mii_data		*mii;
2058
2059	sc = xsc;
2060
2061	RL_LOCK_ASSERT(sc);
2062
2063	mii = device_get_softc(sc->rl_miibus);
2064	mii_tick(mii);
2065	if ((sc->rl_flags & RL_FLAG_LINK) == 0)
2066		re_miibus_statchg(sc->rl_dev);
2067	re_watchdog(sc);
2068	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
2069}
2070
2071#ifdef DEVICE_POLLING
2072static void
2073re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
2074{
2075	struct rl_softc *sc = ifp->if_softc;
2076
2077	RL_LOCK(sc);
2078	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
2079		re_poll_locked(ifp, cmd, count);
2080	RL_UNLOCK(sc);
2081}
2082
2083static void
2084re_poll_locked(struct ifnet *ifp, enum poll_cmd cmd, int count)
2085{
2086	struct rl_softc *sc = ifp->if_softc;
2087
2088	RL_LOCK_ASSERT(sc);
2089
2090	sc->rxcycles = count;
2091	re_rxeof(sc);
2092	re_txeof(sc);
2093
2094	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2095		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_txtask);
2096
2097	if (cmd == POLL_AND_CHECK_STATUS) { /* also check status register */
2098		u_int16_t       status;
2099
2100		status = CSR_READ_2(sc, RL_ISR);
2101		if (status == 0xffff)
2102			return;
2103		if (status)
2104			CSR_WRITE_2(sc, RL_ISR, status);
2105		if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
2106		    (sc->rl_flags & RL_FLAG_PCIE))
2107			CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2108
2109		/*
2110		 * XXX check behaviour on receiver stalls.
2111		 */
2112
2113		if (status & RL_ISR_SYSTEM_ERR)
2114			re_init_locked(sc);
2115	}
2116}
2117#endif /* DEVICE_POLLING */
2118
2119static int
2120re_intr(void *arg)
2121{
2122	struct rl_softc		*sc;
2123	uint16_t		status;
2124
2125	sc = arg;
2126
2127	status = CSR_READ_2(sc, RL_ISR);
2128	if (status == 0xFFFF || (status & RL_INTRS_CPLUS) == 0)
2129                return (FILTER_STRAY);
2130	CSR_WRITE_2(sc, RL_IMR, 0);
2131
2132	taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_inttask);
2133
2134	return (FILTER_HANDLED);
2135}
2136
2137static void
2138re_int_task(void *arg, int npending)
2139{
2140	struct rl_softc		*sc;
2141	struct ifnet		*ifp;
2142	u_int16_t		status;
2143	int			rval = 0;
2144
2145	sc = arg;
2146	ifp = sc->rl_ifp;
2147
2148	RL_LOCK(sc);
2149
2150	status = CSR_READ_2(sc, RL_ISR);
2151        CSR_WRITE_2(sc, RL_ISR, status);
2152
2153	if (sc->suspended ||
2154	    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2155		RL_UNLOCK(sc);
2156		return;
2157	}
2158
2159#ifdef DEVICE_POLLING
2160	if  (ifp->if_capenable & IFCAP_POLLING) {
2161		RL_UNLOCK(sc);
2162		return;
2163	}
2164#endif
2165
2166	if (status & (RL_ISR_RX_OK|RL_ISR_RX_ERR|RL_ISR_FIFO_OFLOW))
2167		rval = re_rxeof(sc);
2168
2169	/*
2170	 * Some chips will ignore a second TX request issued
2171	 * while an existing transmission is in progress. If
2172	 * the transmitter goes idle but there are still
2173	 * packets waiting to be sent, we need to restart the
2174	 * channel here to flush them out. This only seems to
2175	 * be required with the PCIe devices.
2176	 */
2177	if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
2178	    (sc->rl_flags & RL_FLAG_PCIE))
2179		CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2180	if (status & (
2181#ifdef RE_TX_MODERATION
2182	    RL_ISR_TIMEOUT_EXPIRED|
2183#else
2184	    RL_ISR_TX_OK|
2185#endif
2186	    RL_ISR_TX_ERR|RL_ISR_TX_DESC_UNAVAIL))
2187		re_txeof(sc);
2188
2189	if (status & RL_ISR_SYSTEM_ERR)
2190		re_init_locked(sc);
2191
2192	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2193		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_txtask);
2194
2195	RL_UNLOCK(sc);
2196
2197        if ((CSR_READ_2(sc, RL_ISR) & RL_INTRS_CPLUS) || rval) {
2198		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_inttask);
2199		return;
2200	}
2201
2202	CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2203}
2204
2205static int
2206re_encap(struct rl_softc *sc, struct mbuf **m_head)
2207{
2208	struct rl_txdesc	*txd, *txd_last;
2209	bus_dma_segment_t	segs[RL_NTXSEGS];
2210	bus_dmamap_t		map;
2211	struct mbuf		*m_new;
2212	struct rl_desc		*desc;
2213	int			nsegs, prod;
2214	int			i, error, ei, si;
2215	int			padlen;
2216	uint32_t		cmdstat, csum_flags, vlanctl;
2217
2218	RL_LOCK_ASSERT(sc);
2219	M_ASSERTPKTHDR((*m_head));
2220
2221	/*
2222	 * With some of the RealTek chips, using the checksum offload
2223	 * support in conjunction with the autopadding feature results
2224	 * in the transmission of corrupt frames. For example, if we
2225	 * need to send a really small IP fragment that's less than 60
2226	 * bytes in size, and IP header checksumming is enabled, the
2227	 * resulting ethernet frame that appears on the wire will
2228	 * have garbled payload. To work around this, if TX IP checksum
2229	 * offload is enabled, we always manually pad short frames out
2230	 * to the minimum ethernet frame size.
2231	 */
2232	if ((sc->rl_flags & RL_FLAG_DESCV2) == 0 &&
2233	    (*m_head)->m_pkthdr.len < RL_IP4CSUMTX_PADLEN &&
2234	    ((*m_head)->m_pkthdr.csum_flags & CSUM_IP) != 0) {
2235		padlen = RL_MIN_FRAMELEN - (*m_head)->m_pkthdr.len;
2236		if (M_WRITABLE(*m_head) == 0) {
2237			/* Get a writable copy. */
2238			m_new = m_dup(*m_head, M_DONTWAIT);
2239			m_freem(*m_head);
2240			if (m_new == NULL) {
2241				*m_head = NULL;
2242				return (ENOBUFS);
2243			}
2244			*m_head = m_new;
2245		}
2246		if ((*m_head)->m_next != NULL ||
2247		    M_TRAILINGSPACE(*m_head) < padlen) {
2248			m_new = m_defrag(*m_head, M_DONTWAIT);
2249			if (m_new == NULL) {
2250				m_freem(*m_head);
2251				*m_head = NULL;
2252				return (ENOBUFS);
2253			}
2254		} else
2255			m_new = *m_head;
2256
2257		/*
2258		 * Manually pad short frames, and zero the pad space
2259		 * to avoid leaking data.
2260		 */
2261		bzero(mtod(m_new, char *) + m_new->m_pkthdr.len, padlen);
2262		m_new->m_pkthdr.len += padlen;
2263		m_new->m_len = m_new->m_pkthdr.len;
2264		*m_head = m_new;
2265	}
2266
2267	prod = sc->rl_ldata.rl_tx_prodidx;
2268	txd = &sc->rl_ldata.rl_tx_desc[prod];
2269	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap,
2270	    *m_head, segs, &nsegs, BUS_DMA_NOWAIT);
2271	if (error == EFBIG) {
2272		m_new = m_collapse(*m_head, M_DONTWAIT, RL_NTXSEGS);
2273		if (m_new == NULL) {
2274			m_freem(*m_head);
2275			*m_head = NULL;
2276			return (ENOBUFS);
2277		}
2278		*m_head = m_new;
2279		error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_tx_mtag,
2280		    txd->tx_dmamap, *m_head, segs, &nsegs, BUS_DMA_NOWAIT);
2281		if (error != 0) {
2282			m_freem(*m_head);
2283			*m_head = NULL;
2284			return (error);
2285		}
2286	} else if (error != 0)
2287		return (error);
2288	if (nsegs == 0) {
2289		m_freem(*m_head);
2290		*m_head = NULL;
2291		return (EIO);
2292	}
2293
2294	/* Check for number of available descriptors. */
2295	if (sc->rl_ldata.rl_tx_free - nsegs <= 1) {
2296		bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap);
2297		return (ENOBUFS);
2298	}
2299
2300	bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap,
2301	    BUS_DMASYNC_PREWRITE);
2302
2303	/*
2304	 * Set up checksum offload. Note: checksum offload bits must
2305	 * appear in all descriptors of a multi-descriptor transmit
2306	 * attempt. This is according to testing done with an 8169
2307	 * chip. This is a requirement.
2308	 */
2309	vlanctl = 0;
2310	csum_flags = 0;
2311	if (((*m_head)->m_pkthdr.csum_flags & CSUM_TSO) != 0)
2312		csum_flags = RL_TDESC_CMD_LGSEND |
2313		    ((uint32_t)(*m_head)->m_pkthdr.tso_segsz <<
2314		    RL_TDESC_CMD_MSSVAL_SHIFT);
2315	else {
2316		/*
2317		 * Unconditionally enable IP checksum if TCP or UDP
2318		 * checksum is required. Otherwise, TCP/UDP checksum
2319		 * does't make effects.
2320		 */
2321		if (((*m_head)->m_pkthdr.csum_flags & RE_CSUM_FEATURES) != 0) {
2322			if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
2323				csum_flags |= RL_TDESC_CMD_IPCSUM;
2324				if (((*m_head)->m_pkthdr.csum_flags &
2325				    CSUM_TCP) != 0)
2326					csum_flags |= RL_TDESC_CMD_TCPCSUM;
2327				if (((*m_head)->m_pkthdr.csum_flags &
2328				    CSUM_UDP) != 0)
2329					csum_flags |= RL_TDESC_CMD_UDPCSUM;
2330			} else {
2331				vlanctl |= RL_TDESC_CMD_IPCSUMV2;
2332				if (((*m_head)->m_pkthdr.csum_flags &
2333				    CSUM_TCP) != 0)
2334					vlanctl |= RL_TDESC_CMD_TCPCSUMV2;
2335				if (((*m_head)->m_pkthdr.csum_flags &
2336				    CSUM_UDP) != 0)
2337					vlanctl |= RL_TDESC_CMD_UDPCSUMV2;
2338			}
2339		}
2340	}
2341
2342	/*
2343	 * Set up hardware VLAN tagging. Note: vlan tag info must
2344	 * appear in all descriptors of a multi-descriptor
2345	 * transmission attempt.
2346	 */
2347	if ((*m_head)->m_flags & M_VLANTAG)
2348		vlanctl |= bswap16((*m_head)->m_pkthdr.ether_vtag) |
2349		    RL_TDESC_VLANCTL_TAG;
2350
2351	si = prod;
2352	for (i = 0; i < nsegs; i++, prod = RL_TX_DESC_NXT(sc, prod)) {
2353		desc = &sc->rl_ldata.rl_tx_list[prod];
2354		desc->rl_vlanctl = htole32(vlanctl);
2355		desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[i].ds_addr));
2356		desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[i].ds_addr));
2357		cmdstat = segs[i].ds_len;
2358		if (i != 0)
2359			cmdstat |= RL_TDESC_CMD_OWN;
2360		if (prod == sc->rl_ldata.rl_tx_desc_cnt - 1)
2361			cmdstat |= RL_TDESC_CMD_EOR;
2362		desc->rl_cmdstat = htole32(cmdstat | csum_flags);
2363		sc->rl_ldata.rl_tx_free--;
2364	}
2365	/* Update producer index. */
2366	sc->rl_ldata.rl_tx_prodidx = prod;
2367
2368	/* Set EOF on the last descriptor. */
2369	ei = RL_TX_DESC_PRV(sc, prod);
2370	desc = &sc->rl_ldata.rl_tx_list[ei];
2371	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_EOF);
2372
2373	desc = &sc->rl_ldata.rl_tx_list[si];
2374	/* Set SOF and transfer ownership of packet to the chip. */
2375	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_OWN | RL_TDESC_CMD_SOF);
2376
2377	/*
2378	 * Insure that the map for this transmission
2379	 * is placed at the array index of the last descriptor
2380	 * in this chain.  (Swap last and first dmamaps.)
2381	 */
2382	txd_last = &sc->rl_ldata.rl_tx_desc[ei];
2383	map = txd->tx_dmamap;
2384	txd->tx_dmamap = txd_last->tx_dmamap;
2385	txd_last->tx_dmamap = map;
2386	txd_last->tx_m = *m_head;
2387
2388	return (0);
2389}
2390
2391static void
2392re_tx_task(void *arg, int npending)
2393{
2394	struct ifnet		*ifp;
2395
2396	ifp = arg;
2397	re_start(ifp);
2398}
2399
2400/*
2401 * Main transmit routine for C+ and gigE NICs.
2402 */
2403static void
2404re_start(struct ifnet *ifp)
2405{
2406	struct rl_softc		*sc;
2407	struct mbuf		*m_head;
2408	int			queued;
2409
2410	sc = ifp->if_softc;
2411
2412	RL_LOCK(sc);
2413
2414	if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
2415	    IFF_DRV_RUNNING || (sc->rl_flags & RL_FLAG_LINK) == 0) {
2416		RL_UNLOCK(sc);
2417		return;
2418	}
2419
2420	for (queued = 0; !IFQ_DRV_IS_EMPTY(&ifp->if_snd) &&
2421	    sc->rl_ldata.rl_tx_free > 1;) {
2422		IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
2423		if (m_head == NULL)
2424			break;
2425
2426		if (re_encap(sc, &m_head) != 0) {
2427			if (m_head == NULL)
2428				break;
2429			IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
2430			ifp->if_drv_flags |= IFF_DRV_OACTIVE;
2431			break;
2432		}
2433
2434		/*
2435		 * If there's a BPF listener, bounce a copy of this frame
2436		 * to him.
2437		 */
2438		ETHER_BPF_MTAP(ifp, m_head);
2439
2440		queued++;
2441	}
2442
2443	if (queued == 0) {
2444#ifdef RE_TX_MODERATION
2445		if (sc->rl_ldata.rl_tx_free != sc->rl_ldata.rl_tx_desc_cnt)
2446			CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2447#endif
2448		RL_UNLOCK(sc);
2449		return;
2450	}
2451
2452	/* Flush the TX descriptors */
2453
2454	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2455	    sc->rl_ldata.rl_tx_list_map,
2456	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2457
2458	CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2459
2460#ifdef RE_TX_MODERATION
2461	/*
2462	 * Use the countdown timer for interrupt moderation.
2463	 * 'TX done' interrupts are disabled. Instead, we reset the
2464	 * countdown timer, which will begin counting until it hits
2465	 * the value in the TIMERINT register, and then trigger an
2466	 * interrupt. Each time we write to the TIMERCNT register,
2467	 * the timer count is reset to 0.
2468	 */
2469	CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2470#endif
2471
2472	/*
2473	 * Set a timeout in case the chip goes out to lunch.
2474	 */
2475	sc->rl_watchdog_timer = 5;
2476
2477	RL_UNLOCK(sc);
2478}
2479
2480static void
2481re_init(void *xsc)
2482{
2483	struct rl_softc		*sc = xsc;
2484
2485	RL_LOCK(sc);
2486	re_init_locked(sc);
2487	RL_UNLOCK(sc);
2488}
2489
2490static void
2491re_init_locked(struct rl_softc *sc)
2492{
2493	struct ifnet		*ifp = sc->rl_ifp;
2494	struct mii_data		*mii;
2495	uint32_t		reg;
2496	uint16_t		cfg;
2497	union {
2498		uint32_t align_dummy;
2499		u_char eaddr[ETHER_ADDR_LEN];
2500        } eaddr;
2501
2502	RL_LOCK_ASSERT(sc);
2503
2504	mii = device_get_softc(sc->rl_miibus);
2505
2506	/*
2507	 * Cancel pending I/O and free all RX/TX buffers.
2508	 */
2509	re_stop(sc);
2510
2511	/* Put controller into known state. */
2512	re_reset(sc);
2513
2514	/*
2515	 * Enable C+ RX and TX mode, as well as VLAN stripping and
2516	 * RX checksum offload. We must configure the C+ register
2517	 * before all others.
2518	 */
2519	cfg = RL_CPLUSCMD_PCI_MRW;
2520	if ((ifp->if_capenable & IFCAP_RXCSUM) != 0)
2521		cfg |= RL_CPLUSCMD_RXCSUM_ENB;
2522	if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0)
2523		cfg |= RL_CPLUSCMD_VLANSTRIP;
2524	if ((sc->rl_flags & RL_FLAG_MACSTAT) != 0) {
2525		cfg |= RL_CPLUSCMD_MACSTAT_DIS;
2526		/* XXX magic. */
2527		cfg |= 0x0001;
2528	} else
2529		cfg |= RL_CPLUSCMD_RXENB | RL_CPLUSCMD_TXENB;
2530	CSR_WRITE_2(sc, RL_CPLUS_CMD, cfg);
2531	if (sc->rl_hwrev == RL_HWREV_8169_8110SC ||
2532	    sc->rl_hwrev == RL_HWREV_8169_8110SCE) {
2533		reg = 0x000fff00;
2534		if ((CSR_READ_1(sc, RL_CFG2) & RL_CFG2_PCI66MHZ) != 0)
2535			reg |= 0x000000ff;
2536		if (sc->rl_hwrev == RL_HWREV_8169_8110SCE)
2537			reg |= 0x00f00000;
2538		CSR_WRITE_4(sc, 0x7c, reg);
2539		/* Disable interrupt mitigation. */
2540		CSR_WRITE_2(sc, 0xe2, 0);
2541	}
2542	/*
2543	 * Disable TSO if interface MTU size is greater than MSS
2544	 * allowed in controller.
2545	 */
2546	if (ifp->if_mtu > RL_TSO_MTU && (ifp->if_capenable & IFCAP_TSO4) != 0) {
2547		ifp->if_capenable &= ~IFCAP_TSO4;
2548		ifp->if_hwassist &= ~CSUM_TSO;
2549	}
2550
2551	/*
2552	 * Init our MAC address.  Even though the chipset
2553	 * documentation doesn't mention it, we need to enter "Config
2554	 * register write enable" mode to modify the ID registers.
2555	 */
2556	/* Copy MAC address on stack to align. */
2557	bcopy(IF_LLADDR(ifp), eaddr.eaddr, ETHER_ADDR_LEN);
2558	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG);
2559	CSR_WRITE_4(sc, RL_IDR0,
2560	    htole32(*(u_int32_t *)(&eaddr.eaddr[0])));
2561	CSR_WRITE_4(sc, RL_IDR4,
2562	    htole32(*(u_int32_t *)(&eaddr.eaddr[4])));
2563	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
2564
2565	/*
2566	 * For C+ mode, initialize the RX descriptors and mbufs.
2567	 */
2568	re_rx_list_init(sc);
2569	re_tx_list_init(sc);
2570
2571	/*
2572	 * Load the addresses of the RX and TX lists into the chip.
2573	 */
2574
2575	CSR_WRITE_4(sc, RL_RXLIST_ADDR_HI,
2576	    RL_ADDR_HI(sc->rl_ldata.rl_rx_list_addr));
2577	CSR_WRITE_4(sc, RL_RXLIST_ADDR_LO,
2578	    RL_ADDR_LO(sc->rl_ldata.rl_rx_list_addr));
2579
2580	CSR_WRITE_4(sc, RL_TXLIST_ADDR_HI,
2581	    RL_ADDR_HI(sc->rl_ldata.rl_tx_list_addr));
2582	CSR_WRITE_4(sc, RL_TXLIST_ADDR_LO,
2583	    RL_ADDR_LO(sc->rl_ldata.rl_tx_list_addr));
2584
2585	/*
2586	 * Enable transmit and receive.
2587	 */
2588	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
2589
2590	/*
2591	 * Set the initial TX configuration.
2592	 */
2593	if (sc->rl_testmode) {
2594		if (sc->rl_type == RL_8169)
2595			CSR_WRITE_4(sc, RL_TXCFG,
2596			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON);
2597		else
2598			CSR_WRITE_4(sc, RL_TXCFG,
2599			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON_CPLUS);
2600	} else
2601		CSR_WRITE_4(sc, RL_TXCFG, RL_TXCFG_CONFIG);
2602
2603	CSR_WRITE_1(sc, RL_EARLY_TX_THRESH, 16);
2604
2605	/*
2606	 * Set the initial RX configuration.
2607	 */
2608	re_set_rxmode(sc);
2609
2610#ifdef DEVICE_POLLING
2611	/*
2612	 * Disable interrupts if we are polling.
2613	 */
2614	if (ifp->if_capenable & IFCAP_POLLING)
2615		CSR_WRITE_2(sc, RL_IMR, 0);
2616	else	/* otherwise ... */
2617#endif
2618
2619	/*
2620	 * Enable interrupts.
2621	 */
2622	if (sc->rl_testmode)
2623		CSR_WRITE_2(sc, RL_IMR, 0);
2624	else
2625		CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2626	CSR_WRITE_2(sc, RL_ISR, RL_INTRS_CPLUS);
2627
2628	/* Set initial TX threshold */
2629	sc->rl_txthresh = RL_TX_THRESH_INIT;
2630
2631	/* Start RX/TX process. */
2632	CSR_WRITE_4(sc, RL_MISSEDPKT, 0);
2633#ifdef notdef
2634	/* Enable receiver and transmitter. */
2635	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
2636#endif
2637
2638#ifdef RE_TX_MODERATION
2639	/*
2640	 * Initialize the timer interrupt register so that
2641	 * a timer interrupt will be generated once the timer
2642	 * reaches a certain number of ticks. The timer is
2643	 * reloaded on each transmit. This gives us TX interrupt
2644	 * moderation, which dramatically improves TX frame rate.
2645	 */
2646	if (sc->rl_type == RL_8169)
2647		CSR_WRITE_4(sc, RL_TIMERINT_8169, 0x800);
2648	else
2649		CSR_WRITE_4(sc, RL_TIMERINT, 0x400);
2650#endif
2651
2652	/*
2653	 * For 8169 gigE NICs, set the max allowed RX packet
2654	 * size so we can receive jumbo frames.
2655	 */
2656	if (sc->rl_type == RL_8169)
2657		CSR_WRITE_2(sc, RL_MAXRXPKTLEN, 16383);
2658
2659	if (sc->rl_testmode)
2660		return;
2661
2662	mii_mediachg(mii);
2663
2664	CSR_WRITE_1(sc, RL_CFG1, CSR_READ_1(sc, RL_CFG1) | RL_CFG1_DRVLOAD);
2665
2666	ifp->if_drv_flags |= IFF_DRV_RUNNING;
2667	ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
2668
2669	sc->rl_flags &= ~RL_FLAG_LINK;
2670	sc->rl_watchdog_timer = 0;
2671	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
2672}
2673
2674/*
2675 * Set media options.
2676 */
2677static int
2678re_ifmedia_upd(struct ifnet *ifp)
2679{
2680	struct rl_softc		*sc;
2681	struct mii_data		*mii;
2682	int			error;
2683
2684	sc = ifp->if_softc;
2685	mii = device_get_softc(sc->rl_miibus);
2686	RL_LOCK(sc);
2687	error = mii_mediachg(mii);
2688	RL_UNLOCK(sc);
2689
2690	return (error);
2691}
2692
2693/*
2694 * Report current media status.
2695 */
2696static void
2697re_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
2698{
2699	struct rl_softc		*sc;
2700	struct mii_data		*mii;
2701
2702	sc = ifp->if_softc;
2703	mii = device_get_softc(sc->rl_miibus);
2704
2705	RL_LOCK(sc);
2706	mii_pollstat(mii);
2707	RL_UNLOCK(sc);
2708	ifmr->ifm_active = mii->mii_media_active;
2709	ifmr->ifm_status = mii->mii_media_status;
2710}
2711
2712static int
2713re_ioctl(struct ifnet *ifp, u_long command, caddr_t data)
2714{
2715	struct rl_softc		*sc = ifp->if_softc;
2716	struct ifreq		*ifr = (struct ifreq *) data;
2717	struct mii_data		*mii;
2718	int			error = 0;
2719
2720	switch (command) {
2721	case SIOCSIFMTU:
2722		if (ifr->ifr_mtu < ETHERMIN || ifr->ifr_mtu > RL_JUMBO_MTU) {
2723			error = EINVAL;
2724			break;
2725		}
2726		if ((sc->rl_flags & RL_FLAG_NOJUMBO) != 0 &&
2727		    ifr->ifr_mtu > RL_MAX_FRAMELEN) {
2728			error = EINVAL;
2729			break;
2730		}
2731		RL_LOCK(sc);
2732		if (ifp->if_mtu != ifr->ifr_mtu)
2733			ifp->if_mtu = ifr->ifr_mtu;
2734		if (ifp->if_mtu > RL_TSO_MTU &&
2735		    (ifp->if_capenable & IFCAP_TSO4) != 0) {
2736			ifp->if_capenable &= ~IFCAP_TSO4;
2737			ifp->if_hwassist &= ~CSUM_TSO;
2738		}
2739		RL_UNLOCK(sc);
2740		break;
2741	case SIOCSIFFLAGS:
2742		RL_LOCK(sc);
2743		if ((ifp->if_flags & IFF_UP) != 0) {
2744			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
2745				if (((ifp->if_flags ^ sc->rl_if_flags)
2746				    & (IFF_PROMISC | IFF_ALLMULTI)) != 0)
2747					re_set_rxmode(sc);
2748			} else
2749				re_init_locked(sc);
2750		} else {
2751			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
2752				re_stop(sc);
2753		}
2754		sc->rl_if_flags = ifp->if_flags;
2755		RL_UNLOCK(sc);
2756		break;
2757	case SIOCADDMULTI:
2758	case SIOCDELMULTI:
2759		RL_LOCK(sc);
2760		re_set_rxmode(sc);
2761		RL_UNLOCK(sc);
2762		break;
2763	case SIOCGIFMEDIA:
2764	case SIOCSIFMEDIA:
2765		mii = device_get_softc(sc->rl_miibus);
2766		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, command);
2767		break;
2768	case SIOCSIFCAP:
2769	    {
2770		int mask, reinit;
2771
2772		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
2773		reinit = 0;
2774#ifdef DEVICE_POLLING
2775		if (mask & IFCAP_POLLING) {
2776			if (ifr->ifr_reqcap & IFCAP_POLLING) {
2777				error = ether_poll_register(re_poll, ifp);
2778				if (error)
2779					return(error);
2780				RL_LOCK(sc);
2781				/* Disable interrupts */
2782				CSR_WRITE_2(sc, RL_IMR, 0x0000);
2783				ifp->if_capenable |= IFCAP_POLLING;
2784				RL_UNLOCK(sc);
2785			} else {
2786				error = ether_poll_deregister(ifp);
2787				/* Enable interrupts. */
2788				RL_LOCK(sc);
2789				CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2790				ifp->if_capenable &= ~IFCAP_POLLING;
2791				RL_UNLOCK(sc);
2792			}
2793		}
2794#endif /* DEVICE_POLLING */
2795		if (mask & IFCAP_HWCSUM) {
2796			ifp->if_capenable ^= IFCAP_HWCSUM;
2797			if (ifp->if_capenable & IFCAP_TXCSUM)
2798				ifp->if_hwassist |= RE_CSUM_FEATURES;
2799			else
2800				ifp->if_hwassist &= ~RE_CSUM_FEATURES;
2801			reinit = 1;
2802		}
2803		if (mask & IFCAP_VLAN_HWTAGGING) {
2804			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
2805			reinit = 1;
2806		}
2807		if (mask & IFCAP_TSO4) {
2808			ifp->if_capenable ^= IFCAP_TSO4;
2809			if ((IFCAP_TSO4 & ifp->if_capenable) &&
2810			    (IFCAP_TSO4 & ifp->if_capabilities))
2811				ifp->if_hwassist |= CSUM_TSO;
2812			else
2813				ifp->if_hwassist &= ~CSUM_TSO;
2814			if (ifp->if_mtu > RL_TSO_MTU &&
2815			    (ifp->if_capenable & IFCAP_TSO4) != 0) {
2816				ifp->if_capenable &= ~IFCAP_TSO4;
2817				ifp->if_hwassist &= ~CSUM_TSO;
2818			}
2819		}
2820		if ((mask & IFCAP_WOL) != 0 &&
2821		    (ifp->if_capabilities & IFCAP_WOL) != 0) {
2822			if ((mask & IFCAP_WOL_UCAST) != 0)
2823				ifp->if_capenable ^= IFCAP_WOL_UCAST;
2824			if ((mask & IFCAP_WOL_MCAST) != 0)
2825				ifp->if_capenable ^= IFCAP_WOL_MCAST;
2826			if ((mask & IFCAP_WOL_MAGIC) != 0)
2827				ifp->if_capenable ^= IFCAP_WOL_MAGIC;
2828		}
2829		if (reinit && ifp->if_drv_flags & IFF_DRV_RUNNING)
2830			re_init(sc);
2831		VLAN_CAPABILITIES(ifp);
2832	    }
2833		break;
2834	default:
2835		error = ether_ioctl(ifp, command, data);
2836		break;
2837	}
2838
2839	return (error);
2840}
2841
2842static void
2843re_watchdog(struct rl_softc *sc)
2844{
2845	struct ifnet		*ifp;
2846
2847	RL_LOCK_ASSERT(sc);
2848
2849	if (sc->rl_watchdog_timer == 0 || --sc->rl_watchdog_timer != 0)
2850		return;
2851
2852	ifp = sc->rl_ifp;
2853	re_txeof(sc);
2854	if (sc->rl_ldata.rl_tx_free == sc->rl_ldata.rl_tx_desc_cnt) {
2855		if_printf(ifp, "watchdog timeout (missed Tx interrupts) "
2856		    "-- recovering\n");
2857		if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2858			taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_txtask);
2859		return;
2860	}
2861
2862	if_printf(ifp, "watchdog timeout\n");
2863	ifp->if_oerrors++;
2864
2865	re_rxeof(sc);
2866	re_init_locked(sc);
2867	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2868		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_txtask);
2869}
2870
2871/*
2872 * Stop the adapter and free any mbufs allocated to the
2873 * RX and TX lists.
2874 */
2875static void
2876re_stop(struct rl_softc *sc)
2877{
2878	int			i;
2879	struct ifnet		*ifp;
2880	struct rl_txdesc	*txd;
2881	struct rl_rxdesc	*rxd;
2882
2883	RL_LOCK_ASSERT(sc);
2884
2885	ifp = sc->rl_ifp;
2886
2887	sc->rl_watchdog_timer = 0;
2888	callout_stop(&sc->rl_stat_callout);
2889	ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
2890
2891	if ((sc->rl_flags & RL_FLAG_CMDSTOP) != 0)
2892		CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_STOPREQ | RL_CMD_TX_ENB |
2893		    RL_CMD_RX_ENB);
2894	else
2895		CSR_WRITE_1(sc, RL_COMMAND, 0x00);
2896	DELAY(1000);
2897	CSR_WRITE_2(sc, RL_IMR, 0x0000);
2898	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
2899
2900	if (sc->rl_head != NULL) {
2901		m_freem(sc->rl_head);
2902		sc->rl_head = sc->rl_tail = NULL;
2903	}
2904
2905	/* Free the TX list buffers. */
2906
2907	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
2908		txd = &sc->rl_ldata.rl_tx_desc[i];
2909		if (txd->tx_m != NULL) {
2910			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
2911			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
2912			bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag,
2913			    txd->tx_dmamap);
2914			m_freem(txd->tx_m);
2915			txd->tx_m = NULL;
2916		}
2917	}
2918
2919	/* Free the RX list buffers. */
2920
2921	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
2922		rxd = &sc->rl_ldata.rl_rx_desc[i];
2923		if (rxd->rx_m != NULL) {
2924			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
2925			    rxd->rx_dmamap, BUS_DMASYNC_POSTREAD);
2926			bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag,
2927			    rxd->rx_dmamap);
2928			m_freem(rxd->rx_m);
2929			rxd->rx_m = NULL;
2930		}
2931	}
2932}
2933
2934/*
2935 * Device suspend routine.  Stop the interface and save some PCI
2936 * settings in case the BIOS doesn't restore them properly on
2937 * resume.
2938 */
2939static int
2940re_suspend(device_t dev)
2941{
2942	struct rl_softc		*sc;
2943
2944	sc = device_get_softc(dev);
2945
2946	RL_LOCK(sc);
2947	re_stop(sc);
2948	re_setwol(sc);
2949	sc->suspended = 1;
2950	RL_UNLOCK(sc);
2951
2952	return (0);
2953}
2954
2955/*
2956 * Device resume routine.  Restore some PCI settings in case the BIOS
2957 * doesn't, re-enable busmastering, and restart the interface if
2958 * appropriate.
2959 */
2960static int
2961re_resume(device_t dev)
2962{
2963	struct rl_softc		*sc;
2964	struct ifnet		*ifp;
2965
2966	sc = device_get_softc(dev);
2967
2968	RL_LOCK(sc);
2969
2970	ifp = sc->rl_ifp;
2971	/* Take controller out of sleep mode. */
2972	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
2973		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
2974			CSR_WRITE_1(sc, RL_GPIO,
2975			    CSR_READ_1(sc, RL_GPIO) | 0x01);
2976	}
2977
2978	/* reinitialize interface if necessary */
2979	if (ifp->if_flags & IFF_UP)
2980		re_init_locked(sc);
2981
2982	/*
2983	 * Clear WOL matching such that normal Rx filtering
2984	 * wouldn't interfere with WOL patterns.
2985	 */
2986	re_clrwol(sc);
2987	sc->suspended = 0;
2988	RL_UNLOCK(sc);
2989
2990	return (0);
2991}
2992
2993/*
2994 * Stop all chip I/O so that the kernel's probe routines don't
2995 * get confused by errant DMAs when rebooting.
2996 */
2997static int
2998re_shutdown(device_t dev)
2999{
3000	struct rl_softc		*sc;
3001
3002	sc = device_get_softc(dev);
3003
3004	RL_LOCK(sc);
3005	re_stop(sc);
3006	/*
3007	 * Mark interface as down since otherwise we will panic if
3008	 * interrupt comes in later on, which can happen in some
3009	 * cases.
3010	 */
3011	sc->rl_ifp->if_flags &= ~IFF_UP;
3012	re_setwol(sc);
3013	RL_UNLOCK(sc);
3014
3015	return (0);
3016}
3017
3018static void
3019re_setwol(struct rl_softc *sc)
3020{
3021	struct ifnet		*ifp;
3022	int			pmc;
3023	uint16_t		pmstat;
3024	uint8_t			v;
3025
3026	RL_LOCK_ASSERT(sc);
3027
3028	if (pci_find_extcap(sc->rl_dev, PCIY_PMG, &pmc) != 0)
3029		return;
3030
3031	ifp = sc->rl_ifp;
3032	/* Put controller into sleep mode. */
3033	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
3034		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
3035			CSR_WRITE_1(sc, RL_GPIO,
3036			    CSR_READ_1(sc, RL_GPIO) & ~0x01);
3037	}
3038	if ((ifp->if_capenable & IFCAP_WOL) != 0 &&
3039	    (sc->rl_flags & RL_FLAG_WOLRXENB) != 0)
3040		CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RX_ENB);
3041	/* Enable config register write. */
3042	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
3043
3044	/* Enable PME. */
3045	v = CSR_READ_1(sc, RL_CFG1);
3046	v &= ~RL_CFG1_PME;
3047	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3048		v |= RL_CFG1_PME;
3049	CSR_WRITE_1(sc, RL_CFG1, v);
3050
3051	v = CSR_READ_1(sc, RL_CFG3);
3052	v &= ~(RL_CFG3_WOL_LINK | RL_CFG3_WOL_MAGIC);
3053	if ((ifp->if_capenable & IFCAP_WOL_MAGIC) != 0)
3054		v |= RL_CFG3_WOL_MAGIC;
3055	CSR_WRITE_1(sc, RL_CFG3, v);
3056
3057	/* Config register write done. */
3058	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3059
3060	v = CSR_READ_1(sc, RL_CFG5);
3061	v &= ~(RL_CFG5_WOL_BCAST | RL_CFG5_WOL_MCAST | RL_CFG5_WOL_UCAST);
3062	v &= ~RL_CFG5_WOL_LANWAKE;
3063	if ((ifp->if_capenable & IFCAP_WOL_UCAST) != 0)
3064		v |= RL_CFG5_WOL_UCAST;
3065	if ((ifp->if_capenable & IFCAP_WOL_MCAST) != 0)
3066		v |= RL_CFG5_WOL_MCAST | RL_CFG5_WOL_BCAST;
3067	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3068		v |= RL_CFG5_WOL_LANWAKE;
3069	CSR_WRITE_1(sc, RL_CFG5, v);
3070
3071	/*
3072	 * It seems that hardware resets its link speed to 100Mbps in
3073	 * power down mode so switching to 100Mbps in driver is not
3074	 * needed.
3075	 */
3076
3077	/* Request PME if WOL is requested. */
3078	pmstat = pci_read_config(sc->rl_dev, pmc + PCIR_POWER_STATUS, 2);
3079	pmstat &= ~(PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE);
3080	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3081		pmstat |= PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE;
3082	pci_write_config(sc->rl_dev, pmc + PCIR_POWER_STATUS, pmstat, 2);
3083}
3084
3085static void
3086re_clrwol(struct rl_softc *sc)
3087{
3088	int			pmc;
3089	uint8_t			v;
3090
3091	RL_LOCK_ASSERT(sc);
3092
3093	if (pci_find_extcap(sc->rl_dev, PCIY_PMG, &pmc) != 0)
3094		return;
3095
3096	/* Enable config register write. */
3097	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
3098
3099	v = CSR_READ_1(sc, RL_CFG3);
3100	v &= ~(RL_CFG3_WOL_LINK | RL_CFG3_WOL_MAGIC);
3101	CSR_WRITE_1(sc, RL_CFG3, v);
3102
3103	/* Config register write done. */
3104	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3105
3106	v = CSR_READ_1(sc, RL_CFG5);
3107	v &= ~(RL_CFG5_WOL_BCAST | RL_CFG5_WOL_MCAST | RL_CFG5_WOL_UCAST);
3108	v &= ~RL_CFG5_WOL_LANWAKE;
3109	CSR_WRITE_1(sc, RL_CFG5, v);
3110}
3111