if_re.c revision 187417
1/*-
2 * Copyright (c) 1997, 1998-2003
3 *	Bill Paul <wpaul@windriver.com>.  All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 *    notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 *    notice, this list of conditions and the following disclaimer in the
12 *    documentation and/or other materials provided with the distribution.
13 * 3. All advertising materials mentioning features or use of this software
14 *    must display the following acknowledgement:
15 *	This product includes software developed by Bill Paul.
16 * 4. Neither the name of the author nor the names of any co-contributors
17 *    may be used to endorse or promote products derived from this software
18 *    without specific prior written permission.
19 *
20 * THIS SOFTWARE IS PROVIDED BY Bill Paul AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED.  IN NO EVENT SHALL Bill Paul OR THE VOICES IN HIS HEAD
24 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
30 * THE POSSIBILITY OF SUCH DAMAGE.
31 */
32
33#include <sys/cdefs.h>
34__FBSDID("$FreeBSD: head/sys/dev/re/if_re.c 187417 2009-01-19 02:31:27Z yongari $");
35
36/*
37 * RealTek 8139C+/8169/8169S/8110S/8168/8111/8101E PCI NIC driver
38 *
39 * Written by Bill Paul <wpaul@windriver.com>
40 * Senior Networking Software Engineer
41 * Wind River Systems
42 */
43
44/*
45 * This driver is designed to support RealTek's next generation of
46 * 10/100 and 10/100/1000 PCI ethernet controllers. There are currently
47 * seven devices in this family: the RTL8139C+, the RTL8169, the RTL8169S,
48 * RTL8110S, the RTL8168, the RTL8111 and the RTL8101E.
49 *
50 * The 8139C+ is a 10/100 ethernet chip. It is backwards compatible
51 * with the older 8139 family, however it also supports a special
52 * C+ mode of operation that provides several new performance enhancing
53 * features. These include:
54 *
55 *	o Descriptor based DMA mechanism. Each descriptor represents
56 *	  a single packet fragment. Data buffers may be aligned on
57 *	  any byte boundary.
58 *
59 *	o 64-bit DMA
60 *
61 *	o TCP/IP checksum offload for both RX and TX
62 *
63 *	o High and normal priority transmit DMA rings
64 *
65 *	o VLAN tag insertion and extraction
66 *
67 *	o TCP large send (segmentation offload)
68 *
69 * Like the 8139, the 8139C+ also has a built-in 10/100 PHY. The C+
70 * programming API is fairly straightforward. The RX filtering, EEPROM
71 * access and PHY access is the same as it is on the older 8139 series
72 * chips.
73 *
74 * The 8169 is a 64-bit 10/100/1000 gigabit ethernet MAC. It has almost the
75 * same programming API and feature set as the 8139C+ with the following
76 * differences and additions:
77 *
78 *	o 1000Mbps mode
79 *
80 *	o Jumbo frames
81 *
82 *	o GMII and TBI ports/registers for interfacing with copper
83 *	  or fiber PHYs
84 *
85 *	o RX and TX DMA rings can have up to 1024 descriptors
86 *	  (the 8139C+ allows a maximum of 64)
87 *
88 *	o Slight differences in register layout from the 8139C+
89 *
90 * The TX start and timer interrupt registers are at different locations
91 * on the 8169 than they are on the 8139C+. Also, the status word in the
92 * RX descriptor has a slightly different bit layout. The 8169 does not
93 * have a built-in PHY. Most reference boards use a Marvell 88E1000 'Alaska'
94 * copper gigE PHY.
95 *
96 * The 8169S/8110S 10/100/1000 devices have built-in copper gigE PHYs
97 * (the 'S' stands for 'single-chip'). These devices have the same
98 * programming API as the older 8169, but also have some vendor-specific
99 * registers for the on-board PHY. The 8110S is a LAN-on-motherboard
100 * part designed to be pin-compatible with the RealTek 8100 10/100 chip.
101 *
102 * This driver takes advantage of the RX and TX checksum offload and
103 * VLAN tag insertion/extraction features. It also implements TX
104 * interrupt moderation using the timer interrupt registers, which
105 * significantly reduces TX interrupt load. There is also support
106 * for jumbo frames, however the 8169/8169S/8110S can not transmit
107 * jumbo frames larger than 7440, so the max MTU possible with this
108 * driver is 7422 bytes.
109 */
110
111#ifdef HAVE_KERNEL_OPTION_HEADERS
112#include "opt_device_polling.h"
113#endif
114
115#include <sys/param.h>
116#include <sys/endian.h>
117#include <sys/systm.h>
118#include <sys/sockio.h>
119#include <sys/mbuf.h>
120#include <sys/malloc.h>
121#include <sys/module.h>
122#include <sys/kernel.h>
123#include <sys/socket.h>
124#include <sys/lock.h>
125#include <sys/mutex.h>
126#include <sys/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"},
203	{ RL_HWREV_8169_8110SC, RL_8169, "8169SC"},
204	{ RL_HWREV_8169_8110SBL, RL_8169, "8169SBL"},
205	{ RL_HWREV_8100, RL_8139, "8100"},
206	{ RL_HWREV_8101, RL_8139, "8101"},
207	{ RL_HWREV_8100E, RL_8169, "8100E"},
208	{ RL_HWREV_8101E, RL_8169, "8101E"},
209	{ RL_HWREV_8102E, RL_8169, "8102E"},
210	{ RL_HWREV_8102EL, RL_8169, "8102EL"},
211	{ RL_HWREV_8168_SPIN2, RL_8169, "8168"},
212	{ RL_HWREV_8168_SPIN3, RL_8169, "8168"},
213	{ RL_HWREV_8168C, RL_8169, "8168C/8111C"},
214	{ RL_HWREV_8168C_SPIN2, RL_8169, "8168C/8111C"},
215	{ RL_HWREV_8168CP, RL_8169, "8168CP/8111CP"},
216	{ RL_HWREV_8168D, RL_8169, "8168D"},
217	{ 0, 0, NULL }
218};
219
220static int re_probe		(device_t);
221static int re_attach		(device_t);
222static int re_detach		(device_t);
223
224static int re_encap		(struct rl_softc *, struct mbuf **);
225
226static void re_dma_map_addr	(void *, bus_dma_segment_t *, int, int);
227static int re_allocmem		(device_t, struct rl_softc *);
228static __inline void re_discard_rxbuf
229				(struct rl_softc *, int);
230static int re_newbuf		(struct rl_softc *, int);
231static int re_rx_list_init	(struct rl_softc *);
232static int re_tx_list_init	(struct rl_softc *);
233#ifdef RE_FIXUP_RX
234static __inline void re_fixup_rx
235				(struct mbuf *);
236#endif
237static int re_rxeof		(struct rl_softc *);
238static void re_txeof		(struct rl_softc *);
239#ifdef DEVICE_POLLING
240static void re_poll		(struct ifnet *, enum poll_cmd, int);
241static void re_poll_locked	(struct ifnet *, enum poll_cmd, int);
242#endif
243static int re_intr		(void *);
244static void re_tick		(void *);
245static void re_tx_task		(void *, int);
246static void re_int_task		(void *, int);
247static void re_start		(struct ifnet *);
248static int re_ioctl		(struct ifnet *, u_long, caddr_t);
249static void re_init		(void *);
250static void re_init_locked	(struct rl_softc *);
251static void re_stop		(struct rl_softc *);
252static void re_watchdog		(struct rl_softc *);
253static int re_suspend		(device_t);
254static int re_resume		(device_t);
255static int re_shutdown		(device_t);
256static int re_ifmedia_upd	(struct ifnet *);
257static void re_ifmedia_sts	(struct ifnet *, struct ifmediareq *);
258
259static void re_eeprom_putbyte	(struct rl_softc *, int);
260static void re_eeprom_getword	(struct rl_softc *, int, u_int16_t *);
261static void re_read_eeprom	(struct rl_softc *, caddr_t, int, int);
262static int re_gmii_readreg	(device_t, int, int);
263static int re_gmii_writereg	(device_t, int, int, int);
264
265static int re_miibus_readreg	(device_t, int, int);
266static int re_miibus_writereg	(device_t, int, int, int);
267static void re_miibus_statchg	(device_t);
268
269static void re_setmulti		(struct rl_softc *);
270static void re_reset		(struct rl_softc *);
271static void re_setwol		(struct rl_softc *);
272static void re_clrwol		(struct rl_softc *);
273
274#ifdef RE_DIAG
275static int re_diag		(struct rl_softc *);
276#endif
277
278static device_method_t re_methods[] = {
279	/* Device interface */
280	DEVMETHOD(device_probe,		re_probe),
281	DEVMETHOD(device_attach,	re_attach),
282	DEVMETHOD(device_detach,	re_detach),
283	DEVMETHOD(device_suspend,	re_suspend),
284	DEVMETHOD(device_resume,	re_resume),
285	DEVMETHOD(device_shutdown,	re_shutdown),
286
287	/* bus interface */
288	DEVMETHOD(bus_print_child,	bus_generic_print_child),
289	DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
290
291	/* MII interface */
292	DEVMETHOD(miibus_readreg,	re_miibus_readreg),
293	DEVMETHOD(miibus_writereg,	re_miibus_writereg),
294	DEVMETHOD(miibus_statchg,	re_miibus_statchg),
295
296	{ 0, 0 }
297};
298
299static driver_t re_driver = {
300	"re",
301	re_methods,
302	sizeof(struct rl_softc)
303};
304
305static devclass_t re_devclass;
306
307DRIVER_MODULE(re, pci, re_driver, re_devclass, 0, 0);
308DRIVER_MODULE(re, cardbus, re_driver, re_devclass, 0, 0);
309DRIVER_MODULE(miibus, re, miibus_driver, miibus_devclass, 0, 0);
310
311#define EE_SET(x)					\
312	CSR_WRITE_1(sc, RL_EECMD,			\
313		CSR_READ_1(sc, RL_EECMD) | x)
314
315#define EE_CLR(x)					\
316	CSR_WRITE_1(sc, RL_EECMD,			\
317		CSR_READ_1(sc, RL_EECMD) & ~x)
318
319/*
320 * Send a read command and address to the EEPROM, check for ACK.
321 */
322static void
323re_eeprom_putbyte(struct rl_softc *sc, int addr)
324{
325	int			d, i;
326
327	d = addr | (RL_9346_READ << sc->rl_eewidth);
328
329	/*
330	 * Feed in each bit and strobe the clock.
331	 */
332
333	for (i = 1 << (sc->rl_eewidth + 3); i; i >>= 1) {
334		if (d & i) {
335			EE_SET(RL_EE_DATAIN);
336		} else {
337			EE_CLR(RL_EE_DATAIN);
338		}
339		DELAY(100);
340		EE_SET(RL_EE_CLK);
341		DELAY(150);
342		EE_CLR(RL_EE_CLK);
343		DELAY(100);
344	}
345}
346
347/*
348 * Read a word of data stored in the EEPROM at address 'addr.'
349 */
350static void
351re_eeprom_getword(struct rl_softc *sc, int addr, u_int16_t *dest)
352{
353	int			i;
354	u_int16_t		word = 0;
355
356	/*
357	 * Send address of word we want to read.
358	 */
359	re_eeprom_putbyte(sc, addr);
360
361	/*
362	 * Start reading bits from EEPROM.
363	 */
364	for (i = 0x8000; i; i >>= 1) {
365		EE_SET(RL_EE_CLK);
366		DELAY(100);
367		if (CSR_READ_1(sc, RL_EECMD) & RL_EE_DATAOUT)
368			word |= i;
369		EE_CLR(RL_EE_CLK);
370		DELAY(100);
371	}
372
373	*dest = word;
374}
375
376/*
377 * Read a sequence of words from the EEPROM.
378 */
379static void
380re_read_eeprom(struct rl_softc *sc, caddr_t dest, int off, int cnt)
381{
382	int			i;
383	u_int16_t		word = 0, *ptr;
384
385	CSR_SETBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM);
386
387        DELAY(100);
388
389	for (i = 0; i < cnt; i++) {
390		CSR_SETBIT_1(sc, RL_EECMD, RL_EE_SEL);
391		re_eeprom_getword(sc, off + i, &word);
392		CSR_CLRBIT_1(sc, RL_EECMD, RL_EE_SEL);
393		ptr = (u_int16_t *)(dest + (i * 2));
394                *ptr = word;
395	}
396
397	CSR_CLRBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM);
398}
399
400static int
401re_gmii_readreg(device_t dev, int phy, int reg)
402{
403	struct rl_softc		*sc;
404	u_int32_t		rval;
405	int			i;
406
407	if (phy != 1)
408		return (0);
409
410	sc = device_get_softc(dev);
411
412	/* Let the rgephy driver read the GMEDIASTAT register */
413
414	if (reg == RL_GMEDIASTAT) {
415		rval = CSR_READ_1(sc, RL_GMEDIASTAT);
416		return (rval);
417	}
418
419	CSR_WRITE_4(sc, RL_PHYAR, reg << 16);
420
421	for (i = 0; i < RL_PHY_TIMEOUT; i++) {
422		rval = CSR_READ_4(sc, RL_PHYAR);
423		if (rval & RL_PHYAR_BUSY)
424			break;
425		DELAY(100);
426	}
427
428	if (i == RL_PHY_TIMEOUT) {
429		device_printf(sc->rl_dev, "PHY read failed\n");
430		return (0);
431	}
432
433	return (rval & RL_PHYAR_PHYDATA);
434}
435
436static int
437re_gmii_writereg(device_t dev, int phy, int reg, int data)
438{
439	struct rl_softc		*sc;
440	u_int32_t		rval;
441	int			i;
442
443	sc = device_get_softc(dev);
444
445	CSR_WRITE_4(sc, RL_PHYAR, (reg << 16) |
446	    (data & RL_PHYAR_PHYDATA) | RL_PHYAR_BUSY);
447
448	for (i = 0; i < RL_PHY_TIMEOUT; i++) {
449		rval = CSR_READ_4(sc, RL_PHYAR);
450		if (!(rval & RL_PHYAR_BUSY))
451			break;
452		DELAY(100);
453	}
454
455	if (i == RL_PHY_TIMEOUT) {
456		device_printf(sc->rl_dev, "PHY write failed\n");
457		return (0);
458	}
459
460	return (0);
461}
462
463static int
464re_miibus_readreg(device_t dev, int phy, int reg)
465{
466	struct rl_softc		*sc;
467	u_int16_t		rval = 0;
468	u_int16_t		re8139_reg = 0;
469
470	sc = device_get_softc(dev);
471
472	if (sc->rl_type == RL_8169) {
473		rval = re_gmii_readreg(dev, phy, reg);
474		return (rval);
475	}
476
477	/* Pretend the internal PHY is only at address 0 */
478	if (phy) {
479		return (0);
480	}
481	switch (reg) {
482	case MII_BMCR:
483		re8139_reg = RL_BMCR;
484		break;
485	case MII_BMSR:
486		re8139_reg = RL_BMSR;
487		break;
488	case MII_ANAR:
489		re8139_reg = RL_ANAR;
490		break;
491	case MII_ANER:
492		re8139_reg = RL_ANER;
493		break;
494	case MII_ANLPAR:
495		re8139_reg = RL_LPAR;
496		break;
497	case MII_PHYIDR1:
498	case MII_PHYIDR2:
499		return (0);
500	/*
501	 * Allow the rlphy driver to read the media status
502	 * register. If we have a link partner which does not
503	 * support NWAY, this is the register which will tell
504	 * us the results of parallel detection.
505	 */
506	case RL_MEDIASTAT:
507		rval = CSR_READ_1(sc, RL_MEDIASTAT);
508		return (rval);
509	default:
510		device_printf(sc->rl_dev, "bad phy register\n");
511		return (0);
512	}
513	rval = CSR_READ_2(sc, re8139_reg);
514	if (sc->rl_type == RL_8139CPLUS && re8139_reg == RL_BMCR) {
515		/* 8139C+ has different bit layout. */
516		rval &= ~(BMCR_LOOP | BMCR_ISO);
517	}
518	return (rval);
519}
520
521static int
522re_miibus_writereg(device_t dev, int phy, int reg, int data)
523{
524	struct rl_softc		*sc;
525	u_int16_t		re8139_reg = 0;
526	int			rval = 0;
527
528	sc = device_get_softc(dev);
529
530	if (sc->rl_type == RL_8169) {
531		rval = re_gmii_writereg(dev, phy, reg, data);
532		return (rval);
533	}
534
535	/* Pretend the internal PHY is only at address 0 */
536	if (phy)
537		return (0);
538
539	switch (reg) {
540	case MII_BMCR:
541		re8139_reg = RL_BMCR;
542		if (sc->rl_type == RL_8139CPLUS) {
543			/* 8139C+ has different bit layout. */
544			data &= ~(BMCR_LOOP | BMCR_ISO);
545		}
546		break;
547	case MII_BMSR:
548		re8139_reg = RL_BMSR;
549		break;
550	case MII_ANAR:
551		re8139_reg = RL_ANAR;
552		break;
553	case MII_ANER:
554		re8139_reg = RL_ANER;
555		break;
556	case MII_ANLPAR:
557		re8139_reg = RL_LPAR;
558		break;
559	case MII_PHYIDR1:
560	case MII_PHYIDR2:
561		return (0);
562		break;
563	default:
564		device_printf(sc->rl_dev, "bad phy register\n");
565		return (0);
566	}
567	CSR_WRITE_2(sc, re8139_reg, data);
568	return (0);
569}
570
571static void
572re_miibus_statchg(device_t dev)
573{
574	struct rl_softc		*sc;
575	struct ifnet		*ifp;
576	struct mii_data		*mii;
577
578	sc = device_get_softc(dev);
579	mii = device_get_softc(sc->rl_miibus);
580	ifp = sc->rl_ifp;
581	if (mii == NULL || ifp == NULL ||
582	    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
583		return;
584
585	sc->rl_flags &= ~RL_FLAG_LINK;
586	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
587	    (IFM_ACTIVE | IFM_AVALID)) {
588		switch (IFM_SUBTYPE(mii->mii_media_active)) {
589		case IFM_10_T:
590		case IFM_100_TX:
591			sc->rl_flags |= RL_FLAG_LINK;
592			break;
593		case IFM_1000_T:
594			if ((sc->rl_flags & RL_FLAG_FASTETHER) != 0)
595				break;
596			sc->rl_flags |= RL_FLAG_LINK;
597			break;
598		default:
599			break;
600		}
601	}
602	/*
603	 * RealTek controllers does not provide any interface to
604	 * Tx/Rx MACs for resolved speed, duplex and flow-control
605	 * parameters.
606	 */
607}
608
609/*
610 * Program the 64-bit multicast hash filter.
611 */
612static void
613re_setmulti(struct rl_softc *sc)
614{
615	struct ifnet		*ifp;
616	int			h = 0;
617	u_int32_t		hashes[2] = { 0, 0 };
618	struct ifmultiaddr	*ifma;
619	u_int32_t		rxfilt;
620	int			mcnt = 0;
621
622	RL_LOCK_ASSERT(sc);
623
624	ifp = sc->rl_ifp;
625
626
627	rxfilt = CSR_READ_4(sc, RL_RXCFG);
628	rxfilt &= ~(RL_RXCFG_RX_ALLPHYS | RL_RXCFG_RX_MULTI);
629	if (ifp->if_flags & IFF_ALLMULTI || ifp->if_flags & IFF_PROMISC) {
630		if (ifp->if_flags & IFF_PROMISC)
631			rxfilt |= RL_RXCFG_RX_ALLPHYS;
632		/*
633		 * Unlike other hardwares, we have to explicitly set
634		 * RL_RXCFG_RX_MULTI to receive multicast frames in
635		 * promiscuous mode.
636		 */
637		rxfilt |= RL_RXCFG_RX_MULTI;
638		CSR_WRITE_4(sc, RL_RXCFG, rxfilt);
639		CSR_WRITE_4(sc, RL_MAR0, 0xFFFFFFFF);
640		CSR_WRITE_4(sc, RL_MAR4, 0xFFFFFFFF);
641		return;
642	}
643
644	/* first, zot all the existing hash bits */
645	CSR_WRITE_4(sc, RL_MAR0, 0);
646	CSR_WRITE_4(sc, RL_MAR4, 0);
647
648	/* now program new ones */
649	IF_ADDR_LOCK(ifp);
650	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
651		if (ifma->ifma_addr->sa_family != AF_LINK)
652			continue;
653		h = ether_crc32_be(LLADDR((struct sockaddr_dl *)
654		    ifma->ifma_addr), ETHER_ADDR_LEN) >> 26;
655		if (h < 32)
656			hashes[0] |= (1 << h);
657		else
658			hashes[1] |= (1 << (h - 32));
659		mcnt++;
660	}
661	IF_ADDR_UNLOCK(ifp);
662
663	if (mcnt)
664		rxfilt |= RL_RXCFG_RX_MULTI;
665	else
666		rxfilt &= ~RL_RXCFG_RX_MULTI;
667
668	CSR_WRITE_4(sc, RL_RXCFG, rxfilt);
669
670	/*
671	 * For some unfathomable reason, RealTek decided to reverse
672	 * the order of the multicast hash registers in the PCI Express
673	 * parts. This means we have to write the hash pattern in reverse
674	 * order for those devices.
675	 */
676
677	if ((sc->rl_flags & RL_FLAG_INVMAR) != 0) {
678		CSR_WRITE_4(sc, RL_MAR0, bswap32(hashes[1]));
679		CSR_WRITE_4(sc, RL_MAR4, bswap32(hashes[0]));
680	} else {
681		CSR_WRITE_4(sc, RL_MAR0, hashes[0]);
682		CSR_WRITE_4(sc, RL_MAR4, hashes[1]);
683	}
684}
685
686static void
687re_reset(struct rl_softc *sc)
688{
689	int			i;
690
691	RL_LOCK_ASSERT(sc);
692
693	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RESET);
694
695	for (i = 0; i < RL_TIMEOUT; i++) {
696		DELAY(10);
697		if (!(CSR_READ_1(sc, RL_COMMAND) & RL_CMD_RESET))
698			break;
699	}
700	if (i == RL_TIMEOUT)
701		device_printf(sc->rl_dev, "reset never completed!\n");
702
703	if ((sc->rl_flags & RL_FLAG_PHY8169) != 0)
704		CSR_WRITE_1(sc, 0x82, 1);
705	if ((sc->rl_flags & RL_FLAG_PHY8110S) != 0) {
706		CSR_WRITE_1(sc, 0x82, 1);
707		re_gmii_writereg(sc->rl_dev, 1, 0x0B, 0);
708	}
709}
710
711#ifdef RE_DIAG
712
713/*
714 * The following routine is designed to test for a defect on some
715 * 32-bit 8169 cards. Some of these NICs have the REQ64# and ACK64#
716 * lines connected to the bus, however for a 32-bit only card, they
717 * should be pulled high. The result of this defect is that the
718 * NIC will not work right if you plug it into a 64-bit slot: DMA
719 * operations will be done with 64-bit transfers, which will fail
720 * because the 64-bit data lines aren't connected.
721 *
722 * There's no way to work around this (short of talking a soldering
723 * iron to the board), however we can detect it. The method we use
724 * here is to put the NIC into digital loopback mode, set the receiver
725 * to promiscuous mode, and then try to send a frame. We then compare
726 * the frame data we sent to what was received. If the data matches,
727 * then the NIC is working correctly, otherwise we know the user has
728 * a defective NIC which has been mistakenly plugged into a 64-bit PCI
729 * slot. In the latter case, there's no way the NIC can work correctly,
730 * so we print out a message on the console and abort the device attach.
731 */
732
733static int
734re_diag(struct rl_softc *sc)
735{
736	struct ifnet		*ifp = sc->rl_ifp;
737	struct mbuf		*m0;
738	struct ether_header	*eh;
739	struct rl_desc		*cur_rx;
740	u_int16_t		status;
741	u_int32_t		rxstat;
742	int			total_len, i, error = 0, phyaddr;
743	u_int8_t		dst[] = { 0x00, 'h', 'e', 'l', 'l', 'o' };
744	u_int8_t		src[] = { 0x00, 'w', 'o', 'r', 'l', 'd' };
745
746	/* Allocate a single mbuf */
747	MGETHDR(m0, M_DONTWAIT, MT_DATA);
748	if (m0 == NULL)
749		return (ENOBUFS);
750
751	RL_LOCK(sc);
752
753	/*
754	 * Initialize the NIC in test mode. This sets the chip up
755	 * so that it can send and receive frames, but performs the
756	 * following special functions:
757	 * - Puts receiver in promiscuous mode
758	 * - Enables digital loopback mode
759	 * - Leaves interrupts turned off
760	 */
761
762	ifp->if_flags |= IFF_PROMISC;
763	sc->rl_testmode = 1;
764	re_init_locked(sc);
765	sc->rl_flags |= RL_FLAG_LINK;
766	if (sc->rl_type == RL_8169)
767		phyaddr = 1;
768	else
769		phyaddr = 0;
770
771	re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_RESET);
772	for (i = 0; i < RL_TIMEOUT; i++) {
773		status = re_miibus_readreg(sc->rl_dev, phyaddr, MII_BMCR);
774		if (!(status & BMCR_RESET))
775			break;
776	}
777
778	re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_LOOP);
779	CSR_WRITE_2(sc, RL_ISR, RL_INTRS);
780
781	DELAY(100000);
782
783	/* Put some data in the mbuf */
784
785	eh = mtod(m0, struct ether_header *);
786	bcopy ((char *)&dst, eh->ether_dhost, ETHER_ADDR_LEN);
787	bcopy ((char *)&src, eh->ether_shost, ETHER_ADDR_LEN);
788	eh->ether_type = htons(ETHERTYPE_IP);
789	m0->m_pkthdr.len = m0->m_len = ETHER_MIN_LEN - ETHER_CRC_LEN;
790
791	/*
792	 * Queue the packet, start transmission.
793	 * Note: IF_HANDOFF() ultimately calls re_start() for us.
794	 */
795
796	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
797	RL_UNLOCK(sc);
798	/* XXX: re_diag must not be called when in ALTQ mode */
799	IF_HANDOFF(&ifp->if_snd, m0, ifp);
800	RL_LOCK(sc);
801	m0 = NULL;
802
803	/* Wait for it to propagate through the chip */
804
805	DELAY(100000);
806	for (i = 0; i < RL_TIMEOUT; i++) {
807		status = CSR_READ_2(sc, RL_ISR);
808		CSR_WRITE_2(sc, RL_ISR, status);
809		if ((status & (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK)) ==
810		    (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK))
811			break;
812		DELAY(10);
813	}
814
815	if (i == RL_TIMEOUT) {
816		device_printf(sc->rl_dev,
817		    "diagnostic failed, failed to receive packet in"
818		    " loopback mode\n");
819		error = EIO;
820		goto done;
821	}
822
823	/*
824	 * The packet should have been dumped into the first
825	 * entry in the RX DMA ring. Grab it from there.
826	 */
827
828	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
829	    sc->rl_ldata.rl_rx_list_map,
830	    BUS_DMASYNC_POSTREAD);
831	bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag,
832	    sc->rl_ldata.rl_rx_desc[0].rx_dmamap,
833	    BUS_DMASYNC_POSTREAD);
834	bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag,
835	    sc->rl_ldata.rl_rx_desc[0].rx_dmamap);
836
837	m0 = sc->rl_ldata.rl_rx_desc[0].rx_m;
838	sc->rl_ldata.rl_rx_desc[0].rx_m = NULL;
839	eh = mtod(m0, struct ether_header *);
840
841	cur_rx = &sc->rl_ldata.rl_rx_list[0];
842	total_len = RL_RXBYTES(cur_rx);
843	rxstat = le32toh(cur_rx->rl_cmdstat);
844
845	if (total_len != ETHER_MIN_LEN) {
846		device_printf(sc->rl_dev,
847		    "diagnostic failed, received short packet\n");
848		error = EIO;
849		goto done;
850	}
851
852	/* Test that the received packet data matches what we sent. */
853
854	if (bcmp((char *)&eh->ether_dhost, (char *)&dst, ETHER_ADDR_LEN) ||
855	    bcmp((char *)&eh->ether_shost, (char *)&src, ETHER_ADDR_LEN) ||
856	    ntohs(eh->ether_type) != ETHERTYPE_IP) {
857		device_printf(sc->rl_dev, "WARNING, DMA FAILURE!\n");
858		device_printf(sc->rl_dev, "expected TX data: %6D/%6D/0x%x\n",
859		    dst, ":", src, ":", ETHERTYPE_IP);
860		device_printf(sc->rl_dev, "received RX data: %6D/%6D/0x%x\n",
861		    eh->ether_dhost, ":",  eh->ether_shost, ":",
862		    ntohs(eh->ether_type));
863		device_printf(sc->rl_dev, "You may have a defective 32-bit "
864		    "NIC plugged into a 64-bit PCI slot.\n");
865		device_printf(sc->rl_dev, "Please re-install the NIC in a "
866		    "32-bit slot for proper operation.\n");
867		device_printf(sc->rl_dev, "Read the re(4) man page for more "
868		    "details.\n");
869		error = EIO;
870	}
871
872done:
873	/* Turn interface off, release resources */
874
875	sc->rl_testmode = 0;
876	sc->rl_flags &= ~RL_FLAG_LINK;
877	ifp->if_flags &= ~IFF_PROMISC;
878	re_stop(sc);
879	if (m0 != NULL)
880		m_freem(m0);
881
882	RL_UNLOCK(sc);
883
884	return (error);
885}
886
887#endif
888
889/*
890 * Probe for a RealTek 8139C+/8169/8110 chip. Check the PCI vendor and device
891 * IDs against our list and return a device name if we find a match.
892 */
893static int
894re_probe(device_t dev)
895{
896	struct rl_type		*t;
897	uint16_t		devid, vendor;
898	uint16_t		revid, sdevid;
899	int			i;
900
901	vendor = pci_get_vendor(dev);
902	devid = pci_get_device(dev);
903	revid = pci_get_revid(dev);
904	sdevid = pci_get_subdevice(dev);
905
906	if (vendor == LINKSYS_VENDORID && devid == LINKSYS_DEVICEID_EG1032) {
907		if (sdevid != LINKSYS_SUBDEVICE_EG1032_REV3) {
908			/*
909			 * Only attach to rev. 3 of the Linksys EG1032 adapter.
910			 * Rev. 2 is supported by sk(4).
911			 */
912			return (ENXIO);
913		}
914	}
915
916	if (vendor == RT_VENDORID && devid == RT_DEVICEID_8139) {
917		if (revid != 0x20) {
918			/* 8139, let rl(4) take care of this device. */
919			return (ENXIO);
920		}
921	}
922
923	t = re_devs;
924	for (i = 0; i < sizeof(re_devs) / sizeof(re_devs[0]); i++, t++) {
925		if (vendor == t->rl_vid && devid == t->rl_did) {
926			device_set_desc(dev, t->rl_name);
927			return (BUS_PROBE_DEFAULT);
928		}
929	}
930
931	return (ENXIO);
932}
933
934/*
935 * Map a single buffer address.
936 */
937
938static void
939re_dma_map_addr(void *arg, bus_dma_segment_t *segs, int nseg, int error)
940{
941	bus_addr_t		*addr;
942
943	if (error)
944		return;
945
946	KASSERT(nseg == 1, ("too many DMA segments, %d should be 1", nseg));
947	addr = arg;
948	*addr = segs->ds_addr;
949}
950
951static int
952re_allocmem(device_t dev, struct rl_softc *sc)
953{
954	bus_size_t		rx_list_size, tx_list_size;
955	int			error;
956	int			i;
957
958	rx_list_size = sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc);
959	tx_list_size = sc->rl_ldata.rl_tx_desc_cnt * sizeof(struct rl_desc);
960
961	/*
962	 * Allocate the parent bus DMA tag appropriate for PCI.
963	 * In order to use DAC, RL_CPLUSCMD_PCI_DAC bit of RL_CPLUS_CMD
964	 * register should be set. However some RealTek chips are known
965	 * to be buggy on DAC handling, therefore disable DAC by limiting
966	 * DMA address space to 32bit. PCIe variants of RealTek chips
967	 * may not have the limitation but I took safer path.
968	 */
969	error = bus_dma_tag_create(bus_get_dma_tag(dev), 1, 0,
970	    BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL, NULL,
971	    BUS_SPACE_MAXSIZE_32BIT, 0, BUS_SPACE_MAXSIZE_32BIT, 0,
972	    NULL, NULL, &sc->rl_parent_tag);
973	if (error) {
974		device_printf(dev, "could not allocate parent DMA tag\n");
975		return (error);
976	}
977
978	/*
979	 * Allocate map for TX mbufs.
980	 */
981	error = bus_dma_tag_create(sc->rl_parent_tag, 1, 0,
982	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL,
983	    NULL, MCLBYTES * RL_NTXSEGS, RL_NTXSEGS, 4096, 0,
984	    NULL, NULL, &sc->rl_ldata.rl_tx_mtag);
985	if (error) {
986		device_printf(dev, "could not allocate TX DMA tag\n");
987		return (error);
988	}
989
990	/*
991	 * Allocate map for RX mbufs.
992	 */
993
994	error = bus_dma_tag_create(sc->rl_parent_tag, sizeof(uint64_t), 0,
995	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
996	    MCLBYTES, 1, MCLBYTES, 0, NULL, NULL, &sc->rl_ldata.rl_rx_mtag);
997	if (error) {
998		device_printf(dev, "could not allocate RX DMA tag\n");
999		return (error);
1000	}
1001
1002	/*
1003	 * Allocate map for TX descriptor list.
1004	 */
1005	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
1006	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1007	    NULL, tx_list_size, 1, tx_list_size, 0,
1008	    NULL, NULL, &sc->rl_ldata.rl_tx_list_tag);
1009	if (error) {
1010		device_printf(dev, "could not allocate TX DMA ring tag\n");
1011		return (error);
1012	}
1013
1014	/* Allocate DMA'able memory for the TX ring */
1015
1016	error = bus_dmamem_alloc(sc->rl_ldata.rl_tx_list_tag,
1017	    (void **)&sc->rl_ldata.rl_tx_list,
1018	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1019	    &sc->rl_ldata.rl_tx_list_map);
1020	if (error) {
1021		device_printf(dev, "could not allocate TX DMA ring\n");
1022		return (error);
1023	}
1024
1025	/* Load the map for the TX ring. */
1026
1027	sc->rl_ldata.rl_tx_list_addr = 0;
1028	error = bus_dmamap_load(sc->rl_ldata.rl_tx_list_tag,
1029	     sc->rl_ldata.rl_tx_list_map, sc->rl_ldata.rl_tx_list,
1030	     tx_list_size, re_dma_map_addr,
1031	     &sc->rl_ldata.rl_tx_list_addr, BUS_DMA_NOWAIT);
1032	if (error != 0 || sc->rl_ldata.rl_tx_list_addr == 0) {
1033		device_printf(dev, "could not load TX DMA ring\n");
1034		return (ENOMEM);
1035	}
1036
1037	/* Create DMA maps for TX buffers */
1038
1039	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
1040		error = bus_dmamap_create(sc->rl_ldata.rl_tx_mtag, 0,
1041		    &sc->rl_ldata.rl_tx_desc[i].tx_dmamap);
1042		if (error) {
1043			device_printf(dev, "could not create DMA map for TX\n");
1044			return (error);
1045		}
1046	}
1047
1048	/*
1049	 * Allocate map for RX descriptor list.
1050	 */
1051	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
1052	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1053	    NULL, rx_list_size, 1, rx_list_size, 0,
1054	    NULL, NULL, &sc->rl_ldata.rl_rx_list_tag);
1055	if (error) {
1056		device_printf(dev, "could not create RX DMA ring tag\n");
1057		return (error);
1058	}
1059
1060	/* Allocate DMA'able memory for the RX ring */
1061
1062	error = bus_dmamem_alloc(sc->rl_ldata.rl_rx_list_tag,
1063	    (void **)&sc->rl_ldata.rl_rx_list,
1064	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1065	    &sc->rl_ldata.rl_rx_list_map);
1066	if (error) {
1067		device_printf(dev, "could not allocate RX DMA ring\n");
1068		return (error);
1069	}
1070
1071	/* Load the map for the RX ring. */
1072
1073	sc->rl_ldata.rl_rx_list_addr = 0;
1074	error = bus_dmamap_load(sc->rl_ldata.rl_rx_list_tag,
1075	     sc->rl_ldata.rl_rx_list_map, sc->rl_ldata.rl_rx_list,
1076	     rx_list_size, re_dma_map_addr,
1077	     &sc->rl_ldata.rl_rx_list_addr, BUS_DMA_NOWAIT);
1078	if (error != 0 || sc->rl_ldata.rl_rx_list_addr == 0) {
1079		device_printf(dev, "could not load RX DMA ring\n");
1080		return (ENOMEM);
1081	}
1082
1083	/* Create DMA maps for RX buffers */
1084
1085	error = bus_dmamap_create(sc->rl_ldata.rl_rx_mtag, 0,
1086	    &sc->rl_ldata.rl_rx_sparemap);
1087	if (error) {
1088		device_printf(dev, "could not create spare DMA map for RX\n");
1089		return (error);
1090	}
1091	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1092		error = bus_dmamap_create(sc->rl_ldata.rl_rx_mtag, 0,
1093		    &sc->rl_ldata.rl_rx_desc[i].rx_dmamap);
1094		if (error) {
1095			device_printf(dev, "could not create DMA map for RX\n");
1096			return (error);
1097		}
1098	}
1099
1100	return (0);
1101}
1102
1103/*
1104 * Attach the interface. Allocate softc structures, do ifmedia
1105 * setup and ethernet/BPF attach.
1106 */
1107static int
1108re_attach(device_t dev)
1109{
1110	u_char			eaddr[ETHER_ADDR_LEN];
1111	u_int16_t		as[ETHER_ADDR_LEN / 2];
1112	struct rl_softc		*sc;
1113	struct ifnet		*ifp;
1114	struct rl_hwrev		*hw_rev;
1115	int			hwrev;
1116	u_int16_t		devid, re_did = 0;
1117	int			error = 0, rid, i;
1118	int			msic, reg;
1119	uint8_t			cfg;
1120
1121	sc = device_get_softc(dev);
1122	sc->rl_dev = dev;
1123
1124	mtx_init(&sc->rl_mtx, device_get_nameunit(dev), MTX_NETWORK_LOCK,
1125	    MTX_DEF);
1126	callout_init_mtx(&sc->rl_stat_callout, &sc->rl_mtx, 0);
1127
1128	/*
1129	 * Map control/status registers.
1130	 */
1131	pci_enable_busmaster(dev);
1132
1133	devid = pci_get_device(dev);
1134	/* Prefer memory space register mapping over IO space. */
1135	sc->rl_res_id = PCIR_BAR(1);
1136	sc->rl_res_type = SYS_RES_MEMORY;
1137	/* RTL8168/8101E seems to use different BARs. */
1138	if (devid == RT_DEVICEID_8168 || devid == RT_DEVICEID_8101E)
1139		sc->rl_res_id = PCIR_BAR(2);
1140	sc->rl_res = bus_alloc_resource_any(dev, sc->rl_res_type,
1141	    &sc->rl_res_id, RF_ACTIVE);
1142
1143	if (sc->rl_res == NULL) {
1144		sc->rl_res_id = PCIR_BAR(0);
1145		sc->rl_res_type = SYS_RES_IOPORT;
1146		sc->rl_res = bus_alloc_resource_any(dev, sc->rl_res_type,
1147		    &sc->rl_res_id, RF_ACTIVE);
1148		if (sc->rl_res == NULL) {
1149			device_printf(dev, "couldn't map ports/memory\n");
1150			error = ENXIO;
1151			goto fail;
1152		}
1153	}
1154
1155	sc->rl_btag = rman_get_bustag(sc->rl_res);
1156	sc->rl_bhandle = rman_get_bushandle(sc->rl_res);
1157
1158	msic = 0;
1159	if (pci_find_extcap(dev, PCIY_EXPRESS, &reg) == 0) {
1160		sc->rl_flags |= RL_FLAG_PCIE;
1161		msic = pci_msi_count(dev);
1162		if (bootverbose)
1163			device_printf(dev, "MSI count : %d\n", msic);
1164	}
1165	if (msic == RL_MSI_MESSAGES  && msi_disable == 0) {
1166		if (pci_alloc_msi(dev, &msic) == 0) {
1167			if (msic == RL_MSI_MESSAGES) {
1168				device_printf(dev, "Using %d MSI messages\n",
1169				    msic);
1170				sc->rl_flags |= RL_FLAG_MSI;
1171				/* Explicitly set MSI enable bit. */
1172				CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1173				cfg = CSR_READ_1(sc, RL_CFG2);
1174				cfg |= RL_CFG2_MSI;
1175				CSR_WRITE_1(sc, RL_CFG2, cfg);
1176				CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1177			} else
1178				pci_release_msi(dev);
1179		}
1180	}
1181
1182	/* Allocate interrupt */
1183	if ((sc->rl_flags & RL_FLAG_MSI) == 0) {
1184		rid = 0;
1185		sc->rl_irq[0] = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
1186		    RF_SHAREABLE | RF_ACTIVE);
1187		if (sc->rl_irq[0] == NULL) {
1188			device_printf(dev, "couldn't allocate IRQ resources\n");
1189			error = ENXIO;
1190			goto fail;
1191		}
1192	} else {
1193		for (i = 0, rid = 1; i < RL_MSI_MESSAGES; i++, rid++) {
1194			sc->rl_irq[i] = bus_alloc_resource_any(dev,
1195			    SYS_RES_IRQ, &rid, RF_ACTIVE);
1196			if (sc->rl_irq[i] == NULL) {
1197				device_printf(dev,
1198				    "couldn't llocate IRQ resources for "
1199				    "message %d\n", rid);
1200				error = ENXIO;
1201				goto fail;
1202			}
1203		}
1204	}
1205
1206	if ((sc->rl_flags & RL_FLAG_MSI) == 0) {
1207		CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1208		cfg = CSR_READ_1(sc, RL_CFG2);
1209		if ((cfg & RL_CFG2_MSI) != 0) {
1210			device_printf(dev, "turning off MSI enable bit.\n");
1211			cfg &= ~RL_CFG2_MSI;
1212			CSR_WRITE_1(sc, RL_CFG2, cfg);
1213		}
1214		CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1215	}
1216
1217	/* Reset the adapter. */
1218	RL_LOCK(sc);
1219	re_reset(sc);
1220	RL_UNLOCK(sc);
1221
1222	hw_rev = re_hwrevs;
1223	hwrev = CSR_READ_4(sc, RL_TXCFG);
1224	device_printf(dev, "Chip rev. 0x%08x\n", hwrev & 0x7c800000);
1225	device_printf(dev, "MAC rev. 0x%08x\n", hwrev & 0x00700000);
1226	hwrev &= RL_TXCFG_HWREV;
1227	while (hw_rev->rl_desc != NULL) {
1228		if (hw_rev->rl_rev == hwrev) {
1229			sc->rl_type = hw_rev->rl_type;
1230			break;
1231		}
1232		hw_rev++;
1233	}
1234	if (hw_rev->rl_desc == NULL) {
1235		device_printf(dev, "Unknown H/W revision: 0x%08x\n", hwrev);
1236		error = ENXIO;
1237		goto fail;
1238	}
1239
1240	switch (hw_rev->rl_rev) {
1241	case RL_HWREV_8139CPLUS:
1242		sc->rl_flags |= RL_FLAG_NOJUMBO | RL_FLAG_FASTETHER;
1243		break;
1244	case RL_HWREV_8110S:
1245		sc->rl_flags |= RL_FLAG_PHY8110S;
1246		break;
1247	case RL_HWREV_8100E:
1248	case RL_HWREV_8101E:
1249		sc->rl_flags |= RL_FLAG_NOJUMBO | RL_FLAG_INVMAR |
1250		    RL_FLAG_PHYWAKE | RL_FLAG_FASTETHER;
1251		break;
1252	case RL_HWREV_8102E:
1253	case RL_HWREV_8102EL:
1254		sc->rl_flags |= RL_FLAG_NOJUMBO | RL_FLAG_INVMAR |
1255		    RL_FLAG_PHYWAKE | RL_FLAG_PAR | RL_FLAG_DESCV2 |
1256		    RL_FLAG_MACSTAT | RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP;
1257		break;
1258	case RL_HWREV_8168_SPIN1:
1259	case RL_HWREV_8168_SPIN2:
1260		sc->rl_flags |= RL_FLAG_WOLRXENB;
1261		/* FALLTHROUGH */
1262	case RL_HWREV_8168_SPIN3:
1263		sc->rl_flags |= RL_FLAG_INVMAR | RL_FLAG_PHYWAKE |
1264		    RL_FLAG_MACSTAT;
1265		break;
1266	case RL_HWREV_8168C_SPIN2:
1267		sc->rl_flags |= RL_FLAG_MACSLEEP;
1268		/* FALLTHROUGH */
1269	case RL_HWREV_8168C:
1270		if ((hwrev & 0x00700000) == 0x00200000)
1271			sc->rl_flags |= RL_FLAG_MACSLEEP;
1272		/* FALLTHROUGH */
1273	case RL_HWREV_8168CP:
1274	case RL_HWREV_8168D:
1275		sc->rl_flags |= RL_FLAG_INVMAR | RL_FLAG_PHYWAKE |
1276		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
1277		    RL_FLAG_CMDSTOP;
1278		/*
1279		 * These controllers support jumbo frame but it seems
1280		 * that enabling it requires touching additional magic
1281		 * registers. Depending on MAC revisions some
1282		 * controllers need to disable checksum offload. So
1283		 * disable jumbo frame until I have better idea what
1284		 * it really requires to make it support.
1285		 * RTL8168C/CP : supports up to 6KB jumbo frame.
1286		 * RTL8111C/CP : supports up to 9KB jumbo frame.
1287		 */
1288		sc->rl_flags |= RL_FLAG_NOJUMBO;
1289		break;
1290	case RL_HWREV_8169:
1291	case RL_HWREV_8169S:
1292		sc->rl_flags |= RL_FLAG_PHY8169;
1293		break;
1294	case RL_HWREV_8169_8110SB:
1295	case RL_HWREV_8169_8110SC:
1296	case RL_HWREV_8169_8110SBL:
1297		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHY8169;
1298		break;
1299	default:
1300		break;
1301	}
1302
1303	/* Enable PME. */
1304	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1305	cfg = CSR_READ_1(sc, RL_CFG1);
1306	cfg |= RL_CFG1_PME;
1307	CSR_WRITE_1(sc, RL_CFG1, cfg);
1308	cfg = CSR_READ_1(sc, RL_CFG5);
1309	cfg &= RL_CFG5_PME_STS;
1310	CSR_WRITE_1(sc, RL_CFG5, cfg);
1311	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1312
1313	if ((sc->rl_flags & RL_FLAG_PAR) != 0) {
1314		/*
1315		 * XXX Should have a better way to extract station
1316		 * address from EEPROM.
1317		 */
1318		for (i = 0; i < ETHER_ADDR_LEN; i++)
1319			eaddr[i] = CSR_READ_1(sc, RL_IDR0 + i);
1320	} else {
1321		sc->rl_eewidth = RL_9356_ADDR_LEN;
1322		re_read_eeprom(sc, (caddr_t)&re_did, 0, 1);
1323		if (re_did != 0x8129)
1324			sc->rl_eewidth = RL_9346_ADDR_LEN;
1325
1326		/*
1327		 * Get station address from the EEPROM.
1328		 */
1329		re_read_eeprom(sc, (caddr_t)as, RL_EE_EADDR, 3);
1330		for (i = 0; i < ETHER_ADDR_LEN / 2; i++)
1331			as[i] = le16toh(as[i]);
1332		bcopy(as, eaddr, sizeof(eaddr));
1333	}
1334
1335	if (sc->rl_type == RL_8169) {
1336		/* Set RX length mask and number of descriptors. */
1337		sc->rl_rxlenmask = RL_RDESC_STAT_GFRAGLEN;
1338		sc->rl_txstart = RL_GTXSTART;
1339		sc->rl_ldata.rl_tx_desc_cnt = RL_8169_TX_DESC_CNT;
1340		sc->rl_ldata.rl_rx_desc_cnt = RL_8169_RX_DESC_CNT;
1341	} else {
1342		/* Set RX length mask and number of descriptors. */
1343		sc->rl_rxlenmask = RL_RDESC_STAT_FRAGLEN;
1344		sc->rl_txstart = RL_TXSTART;
1345		sc->rl_ldata.rl_tx_desc_cnt = RL_8139_TX_DESC_CNT;
1346		sc->rl_ldata.rl_rx_desc_cnt = RL_8139_RX_DESC_CNT;
1347	}
1348
1349	error = re_allocmem(dev, sc);
1350	if (error)
1351		goto fail;
1352
1353	ifp = sc->rl_ifp = if_alloc(IFT_ETHER);
1354	if (ifp == NULL) {
1355		device_printf(dev, "can not if_alloc()\n");
1356		error = ENOSPC;
1357		goto fail;
1358	}
1359
1360	/* Take controller out of deep sleep mode. */
1361	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
1362		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
1363			CSR_WRITE_1(sc, RL_GPIO,
1364			    CSR_READ_1(sc, RL_GPIO) | 0x01);
1365		else
1366			CSR_WRITE_1(sc, RL_GPIO,
1367			    CSR_READ_1(sc, RL_GPIO) & ~0x01);
1368	}
1369
1370	/* Take PHY out of power down mode. */
1371	if ((sc->rl_flags & RL_FLAG_PHYWAKE) != 0) {
1372		re_gmii_writereg(dev, 1, 0x1f, 0);
1373		re_gmii_writereg(dev, 1, 0x0e, 0);
1374	}
1375
1376	/* Do MII setup */
1377	if (mii_phy_probe(dev, &sc->rl_miibus,
1378	    re_ifmedia_upd, re_ifmedia_sts)) {
1379		device_printf(dev, "MII without any phy!\n");
1380		error = ENXIO;
1381		goto fail;
1382	}
1383
1384	ifp->if_softc = sc;
1385	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1386	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1387	ifp->if_ioctl = re_ioctl;
1388	ifp->if_start = re_start;
1389	ifp->if_hwassist = RE_CSUM_FEATURES;
1390	ifp->if_capabilities = IFCAP_HWCSUM;
1391	ifp->if_capenable = ifp->if_capabilities;
1392	ifp->if_init = re_init;
1393	IFQ_SET_MAXLEN(&ifp->if_snd, RL_IFQ_MAXLEN);
1394	ifp->if_snd.ifq_drv_maxlen = RL_IFQ_MAXLEN;
1395	IFQ_SET_READY(&ifp->if_snd);
1396
1397	TASK_INIT(&sc->rl_txtask, 1, re_tx_task, ifp);
1398	TASK_INIT(&sc->rl_inttask, 0, re_int_task, sc);
1399
1400	/*
1401	 * XXX
1402	 * Still have no idea how to make TSO work on 8168C, 8168CP,
1403	 * 8111C and 8111CP.
1404	 */
1405	if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
1406		ifp->if_hwassist |= CSUM_TSO;
1407		ifp->if_capabilities |= IFCAP_TSO4;
1408	}
1409
1410	/*
1411	 * Call MI attach routine.
1412	 */
1413	ether_ifattach(ifp, eaddr);
1414
1415	/* VLAN capability setup */
1416	ifp->if_capabilities |= IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING;
1417	if (ifp->if_capabilities & IFCAP_HWCSUM)
1418		ifp->if_capabilities |= IFCAP_VLAN_HWCSUM;
1419	/* Enable WOL if PM is supported. */
1420	if (pci_find_extcap(sc->rl_dev, PCIY_PMG, &reg) == 0)
1421		ifp->if_capabilities |= IFCAP_WOL;
1422	ifp->if_capenable = ifp->if_capabilities;
1423	/*
1424	 * Don't enable TSO by default. Under certain
1425	 * circumtances the controller generated corrupted
1426	 * packets in TSO size.
1427	 */
1428	ifp->if_hwassist &= ~CSUM_TSO;
1429	ifp->if_capenable &= ~IFCAP_TSO4;
1430#ifdef DEVICE_POLLING
1431	ifp->if_capabilities |= IFCAP_POLLING;
1432#endif
1433	/*
1434	 * Tell the upper layer(s) we support long frames.
1435	 * Must appear after the call to ether_ifattach() because
1436	 * ether_ifattach() sets ifi_hdrlen to the default value.
1437	 */
1438	ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
1439
1440#ifdef RE_DIAG
1441	/*
1442	 * Perform hardware diagnostic on the original RTL8169.
1443	 * Some 32-bit cards were incorrectly wired and would
1444	 * malfunction if plugged into a 64-bit slot.
1445	 */
1446
1447	if (hwrev == RL_HWREV_8169) {
1448		error = re_diag(sc);
1449		if (error) {
1450			device_printf(dev,
1451		    	"attach aborted due to hardware diag failure\n");
1452			ether_ifdetach(ifp);
1453			goto fail;
1454		}
1455	}
1456#endif
1457
1458	/* Hook interrupt last to avoid having to lock softc */
1459	if ((sc->rl_flags & RL_FLAG_MSI) == 0)
1460		error = bus_setup_intr(dev, sc->rl_irq[0],
1461		    INTR_TYPE_NET | INTR_MPSAFE, re_intr, NULL, sc,
1462		    &sc->rl_intrhand[0]);
1463	else {
1464		for (i = 0; i < RL_MSI_MESSAGES; i++) {
1465			error = bus_setup_intr(dev, sc->rl_irq[i],
1466			    INTR_TYPE_NET | INTR_MPSAFE, re_intr, NULL, sc,
1467		    	    &sc->rl_intrhand[i]);
1468			if (error != 0)
1469				break;
1470		}
1471	}
1472	if (error) {
1473		device_printf(dev, "couldn't set up irq\n");
1474		ether_ifdetach(ifp);
1475	}
1476
1477fail:
1478
1479	if (error)
1480		re_detach(dev);
1481
1482	return (error);
1483}
1484
1485/*
1486 * Shutdown hardware and free up resources. This can be called any
1487 * time after the mutex has been initialized. It is called in both
1488 * the error case in attach and the normal detach case so it needs
1489 * to be careful about only freeing resources that have actually been
1490 * allocated.
1491 */
1492static int
1493re_detach(device_t dev)
1494{
1495	struct rl_softc		*sc;
1496	struct ifnet		*ifp;
1497	int			i, rid;
1498
1499	sc = device_get_softc(dev);
1500	ifp = sc->rl_ifp;
1501	KASSERT(mtx_initialized(&sc->rl_mtx), ("re mutex not initialized"));
1502
1503	/* These should only be active if attach succeeded */
1504	if (device_is_attached(dev)) {
1505#ifdef DEVICE_POLLING
1506		if (ifp->if_capenable & IFCAP_POLLING)
1507			ether_poll_deregister(ifp);
1508#endif
1509		RL_LOCK(sc);
1510#if 0
1511		sc->suspended = 1;
1512#endif
1513		re_stop(sc);
1514		RL_UNLOCK(sc);
1515		callout_drain(&sc->rl_stat_callout);
1516		taskqueue_drain(taskqueue_fast, &sc->rl_inttask);
1517		taskqueue_drain(taskqueue_fast, &sc->rl_txtask);
1518		/*
1519		 * Force off the IFF_UP flag here, in case someone
1520		 * still had a BPF descriptor attached to this
1521		 * interface. If they do, ether_ifdetach() will cause
1522		 * the BPF code to try and clear the promisc mode
1523		 * flag, which will bubble down to re_ioctl(),
1524		 * which will try to call re_init() again. This will
1525		 * turn the NIC back on and restart the MII ticker,
1526		 * which will panic the system when the kernel tries
1527		 * to invoke the re_tick() function that isn't there
1528		 * anymore.
1529		 */
1530		ifp->if_flags &= ~IFF_UP;
1531		ether_ifdetach(ifp);
1532	}
1533	if (sc->rl_miibus)
1534		device_delete_child(dev, sc->rl_miibus);
1535	bus_generic_detach(dev);
1536
1537	/*
1538	 * The rest is resource deallocation, so we should already be
1539	 * stopped here.
1540	 */
1541
1542	for (i = 0; i < RL_MSI_MESSAGES; i++) {
1543		if (sc->rl_intrhand[i] != NULL) {
1544			bus_teardown_intr(dev, sc->rl_irq[i],
1545			    sc->rl_intrhand[i]);
1546			sc->rl_intrhand[i] = NULL;
1547		}
1548	}
1549	if (ifp != NULL)
1550		if_free(ifp);
1551	if ((sc->rl_flags & RL_FLAG_MSI) == 0) {
1552		if (sc->rl_irq[0] != NULL) {
1553			bus_release_resource(dev, SYS_RES_IRQ, 0,
1554			    sc->rl_irq[0]);
1555			sc->rl_irq[0] = NULL;
1556		}
1557	} else {
1558		for (i = 0, rid = 1; i < RL_MSI_MESSAGES; i++, rid++) {
1559			if (sc->rl_irq[i] != NULL) {
1560				bus_release_resource(dev, SYS_RES_IRQ, rid,
1561				    sc->rl_irq[i]);
1562				sc->rl_irq[i] = NULL;
1563			}
1564		}
1565		pci_release_msi(dev);
1566	}
1567	if (sc->rl_res)
1568		bus_release_resource(dev, sc->rl_res_type, sc->rl_res_id,
1569		    sc->rl_res);
1570
1571	/* Unload and free the RX DMA ring memory and map */
1572
1573	if (sc->rl_ldata.rl_rx_list_tag) {
1574		bus_dmamap_unload(sc->rl_ldata.rl_rx_list_tag,
1575		    sc->rl_ldata.rl_rx_list_map);
1576		bus_dmamem_free(sc->rl_ldata.rl_rx_list_tag,
1577		    sc->rl_ldata.rl_rx_list,
1578		    sc->rl_ldata.rl_rx_list_map);
1579		bus_dma_tag_destroy(sc->rl_ldata.rl_rx_list_tag);
1580	}
1581
1582	/* Unload and free the TX DMA ring memory and map */
1583
1584	if (sc->rl_ldata.rl_tx_list_tag) {
1585		bus_dmamap_unload(sc->rl_ldata.rl_tx_list_tag,
1586		    sc->rl_ldata.rl_tx_list_map);
1587		bus_dmamem_free(sc->rl_ldata.rl_tx_list_tag,
1588		    sc->rl_ldata.rl_tx_list,
1589		    sc->rl_ldata.rl_tx_list_map);
1590		bus_dma_tag_destroy(sc->rl_ldata.rl_tx_list_tag);
1591	}
1592
1593	/* Destroy all the RX and TX buffer maps */
1594
1595	if (sc->rl_ldata.rl_tx_mtag) {
1596		for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++)
1597			bus_dmamap_destroy(sc->rl_ldata.rl_tx_mtag,
1598			    sc->rl_ldata.rl_tx_desc[i].tx_dmamap);
1599		bus_dma_tag_destroy(sc->rl_ldata.rl_tx_mtag);
1600	}
1601	if (sc->rl_ldata.rl_rx_mtag) {
1602		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++)
1603			bus_dmamap_destroy(sc->rl_ldata.rl_rx_mtag,
1604			    sc->rl_ldata.rl_rx_desc[i].rx_dmamap);
1605		if (sc->rl_ldata.rl_rx_sparemap)
1606			bus_dmamap_destroy(sc->rl_ldata.rl_rx_mtag,
1607			    sc->rl_ldata.rl_rx_sparemap);
1608		bus_dma_tag_destroy(sc->rl_ldata.rl_rx_mtag);
1609	}
1610
1611	/* Unload and free the stats buffer and map */
1612
1613	if (sc->rl_ldata.rl_stag) {
1614		bus_dmamap_unload(sc->rl_ldata.rl_stag,
1615		    sc->rl_ldata.rl_rx_list_map);
1616		bus_dmamem_free(sc->rl_ldata.rl_stag,
1617		    sc->rl_ldata.rl_stats,
1618		    sc->rl_ldata.rl_smap);
1619		bus_dma_tag_destroy(sc->rl_ldata.rl_stag);
1620	}
1621
1622	if (sc->rl_parent_tag)
1623		bus_dma_tag_destroy(sc->rl_parent_tag);
1624
1625	mtx_destroy(&sc->rl_mtx);
1626
1627	return (0);
1628}
1629
1630static __inline void
1631re_discard_rxbuf(struct rl_softc *sc, int idx)
1632{
1633	struct rl_desc		*desc;
1634	struct rl_rxdesc	*rxd;
1635	uint32_t		cmdstat;
1636
1637	rxd = &sc->rl_ldata.rl_rx_desc[idx];
1638	desc = &sc->rl_ldata.rl_rx_list[idx];
1639	desc->rl_vlanctl = 0;
1640	cmdstat = rxd->rx_size;
1641	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
1642		cmdstat |= RL_RDESC_CMD_EOR;
1643	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
1644}
1645
1646static int
1647re_newbuf(struct rl_softc *sc, int idx)
1648{
1649	struct mbuf		*m;
1650	struct rl_rxdesc	*rxd;
1651	bus_dma_segment_t	segs[1];
1652	bus_dmamap_t		map;
1653	struct rl_desc		*desc;
1654	uint32_t		cmdstat;
1655	int			error, nsegs;
1656
1657	m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
1658	if (m == NULL)
1659		return (ENOBUFS);
1660
1661	m->m_len = m->m_pkthdr.len = MCLBYTES;
1662#ifdef RE_FIXUP_RX
1663	/*
1664	 * This is part of an evil trick to deal with non-x86 platforms.
1665	 * The RealTek chip requires RX buffers to be aligned on 64-bit
1666	 * boundaries, but that will hose non-x86 machines. To get around
1667	 * this, we leave some empty space at the start of each buffer
1668	 * and for non-x86 hosts, we copy the buffer back six bytes
1669	 * to achieve word alignment. This is slightly more efficient
1670	 * than allocating a new buffer, copying the contents, and
1671	 * discarding the old buffer.
1672	 */
1673	m_adj(m, RE_ETHER_ALIGN);
1674#endif
1675	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_rx_mtag,
1676	    sc->rl_ldata.rl_rx_sparemap, m, segs, &nsegs, BUS_DMA_NOWAIT);
1677	if (error != 0) {
1678		m_freem(m);
1679		return (ENOBUFS);
1680	}
1681	KASSERT(nsegs == 1, ("%s: %d segment returned!", __func__, nsegs));
1682
1683	rxd = &sc->rl_ldata.rl_rx_desc[idx];
1684	if (rxd->rx_m != NULL) {
1685		bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap,
1686		    BUS_DMASYNC_POSTREAD);
1687		bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap);
1688	}
1689
1690	rxd->rx_m = m;
1691	map = rxd->rx_dmamap;
1692	rxd->rx_dmamap = sc->rl_ldata.rl_rx_sparemap;
1693	rxd->rx_size = segs[0].ds_len;
1694	sc->rl_ldata.rl_rx_sparemap = map;
1695	bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap,
1696	    BUS_DMASYNC_PREREAD);
1697
1698	desc = &sc->rl_ldata.rl_rx_list[idx];
1699	desc->rl_vlanctl = 0;
1700	desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[0].ds_addr));
1701	desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[0].ds_addr));
1702	cmdstat = segs[0].ds_len;
1703	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
1704		cmdstat |= RL_RDESC_CMD_EOR;
1705	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
1706
1707	return (0);
1708}
1709
1710#ifdef RE_FIXUP_RX
1711static __inline void
1712re_fixup_rx(struct mbuf *m)
1713{
1714	int                     i;
1715	uint16_t                *src, *dst;
1716
1717	src = mtod(m, uint16_t *);
1718	dst = src - (RE_ETHER_ALIGN - ETHER_ALIGN) / sizeof *src;
1719
1720	for (i = 0; i < (m->m_len / sizeof(uint16_t) + 1); i++)
1721		*dst++ = *src++;
1722
1723	m->m_data -= RE_ETHER_ALIGN - ETHER_ALIGN;
1724}
1725#endif
1726
1727static int
1728re_tx_list_init(struct rl_softc *sc)
1729{
1730	struct rl_desc		*desc;
1731	int			i;
1732
1733	RL_LOCK_ASSERT(sc);
1734
1735	bzero(sc->rl_ldata.rl_tx_list,
1736	    sc->rl_ldata.rl_tx_desc_cnt * sizeof(struct rl_desc));
1737	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++)
1738		sc->rl_ldata.rl_tx_desc[i].tx_m = NULL;
1739	/* Set EOR. */
1740	desc = &sc->rl_ldata.rl_tx_list[sc->rl_ldata.rl_tx_desc_cnt - 1];
1741	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_EOR);
1742
1743	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
1744	    sc->rl_ldata.rl_tx_list_map,
1745	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1746
1747	sc->rl_ldata.rl_tx_prodidx = 0;
1748	sc->rl_ldata.rl_tx_considx = 0;
1749	sc->rl_ldata.rl_tx_free = sc->rl_ldata.rl_tx_desc_cnt;
1750
1751	return (0);
1752}
1753
1754static int
1755re_rx_list_init(struct rl_softc *sc)
1756{
1757	int			error, i;
1758
1759	bzero(sc->rl_ldata.rl_rx_list,
1760	    sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc));
1761	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1762		sc->rl_ldata.rl_rx_desc[i].rx_m = NULL;
1763		if ((error = re_newbuf(sc, i)) != 0)
1764			return (error);
1765	}
1766
1767	/* Flush the RX descriptors */
1768
1769	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1770	    sc->rl_ldata.rl_rx_list_map,
1771	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
1772
1773	sc->rl_ldata.rl_rx_prodidx = 0;
1774	sc->rl_head = sc->rl_tail = NULL;
1775
1776	return (0);
1777}
1778
1779/*
1780 * RX handler for C+ and 8169. For the gigE chips, we support
1781 * the reception of jumbo frames that have been fragmented
1782 * across multiple 2K mbuf cluster buffers.
1783 */
1784static int
1785re_rxeof(struct rl_softc *sc)
1786{
1787	struct mbuf		*m;
1788	struct ifnet		*ifp;
1789	int			i, total_len;
1790	struct rl_desc		*cur_rx;
1791	u_int32_t		rxstat, rxvlan;
1792	int			maxpkt = 16;
1793
1794	RL_LOCK_ASSERT(sc);
1795
1796	ifp = sc->rl_ifp;
1797
1798	/* Invalidate the descriptor memory */
1799
1800	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1801	    sc->rl_ldata.rl_rx_list_map,
1802	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
1803
1804	for (i = sc->rl_ldata.rl_rx_prodidx; maxpkt > 0;
1805	    i = RL_RX_DESC_NXT(sc, i)) {
1806		cur_rx = &sc->rl_ldata.rl_rx_list[i];
1807		rxstat = le32toh(cur_rx->rl_cmdstat);
1808		if ((rxstat & RL_RDESC_STAT_OWN) != 0)
1809			break;
1810		total_len = rxstat & sc->rl_rxlenmask;
1811		rxvlan = le32toh(cur_rx->rl_vlanctl);
1812		m = sc->rl_ldata.rl_rx_desc[i].rx_m;
1813
1814		if (!(rxstat & RL_RDESC_STAT_EOF)) {
1815			if (re_newbuf(sc, i) != 0) {
1816				/*
1817				 * If this is part of a multi-fragment packet,
1818				 * discard all the pieces.
1819				 */
1820				if (sc->rl_head != NULL) {
1821					m_freem(sc->rl_head);
1822					sc->rl_head = sc->rl_tail = NULL;
1823				}
1824				re_discard_rxbuf(sc, i);
1825				continue;
1826			}
1827			m->m_len = RE_RX_DESC_BUFLEN;
1828			if (sc->rl_head == NULL)
1829				sc->rl_head = sc->rl_tail = m;
1830			else {
1831				m->m_flags &= ~M_PKTHDR;
1832				sc->rl_tail->m_next = m;
1833				sc->rl_tail = m;
1834			}
1835			continue;
1836		}
1837
1838		/*
1839		 * NOTE: for the 8139C+, the frame length field
1840		 * is always 12 bits in size, but for the gigE chips,
1841		 * it is 13 bits (since the max RX frame length is 16K).
1842		 * Unfortunately, all 32 bits in the status word
1843		 * were already used, so to make room for the extra
1844		 * length bit, RealTek took out the 'frame alignment
1845		 * error' bit and shifted the other status bits
1846		 * over one slot. The OWN, EOR, FS and LS bits are
1847		 * still in the same places. We have already extracted
1848		 * the frame length and checked the OWN bit, so rather
1849		 * than using an alternate bit mapping, we shift the
1850		 * status bits one space to the right so we can evaluate
1851		 * them using the 8169 status as though it was in the
1852		 * same format as that of the 8139C+.
1853		 */
1854		if (sc->rl_type == RL_8169)
1855			rxstat >>= 1;
1856
1857		/*
1858		 * if total_len > 2^13-1, both _RXERRSUM and _GIANT will be
1859		 * set, but if CRC is clear, it will still be a valid frame.
1860		 */
1861		if (rxstat & RL_RDESC_STAT_RXERRSUM && !(total_len > 8191 &&
1862		    (rxstat & RL_RDESC_STAT_ERRS) == RL_RDESC_STAT_GIANT)) {
1863			ifp->if_ierrors++;
1864			/*
1865			 * If this is part of a multi-fragment packet,
1866			 * discard all the pieces.
1867			 */
1868			if (sc->rl_head != NULL) {
1869				m_freem(sc->rl_head);
1870				sc->rl_head = sc->rl_tail = NULL;
1871			}
1872			re_discard_rxbuf(sc, i);
1873			continue;
1874		}
1875
1876		/*
1877		 * If allocating a replacement mbuf fails,
1878		 * reload the current one.
1879		 */
1880
1881		if (re_newbuf(sc, i) != 0) {
1882			ifp->if_iqdrops++;
1883			if (sc->rl_head != NULL) {
1884				m_freem(sc->rl_head);
1885				sc->rl_head = sc->rl_tail = NULL;
1886			}
1887			re_discard_rxbuf(sc, i);
1888			continue;
1889		}
1890
1891		if (sc->rl_head != NULL) {
1892			m->m_len = total_len % RE_RX_DESC_BUFLEN;
1893			if (m->m_len == 0)
1894				m->m_len = RE_RX_DESC_BUFLEN;
1895			/*
1896			 * Special case: if there's 4 bytes or less
1897			 * in this buffer, the mbuf can be discarded:
1898			 * the last 4 bytes is the CRC, which we don't
1899			 * care about anyway.
1900			 */
1901			if (m->m_len <= ETHER_CRC_LEN) {
1902				sc->rl_tail->m_len -=
1903				    (ETHER_CRC_LEN - m->m_len);
1904				m_freem(m);
1905			} else {
1906				m->m_len -= ETHER_CRC_LEN;
1907				m->m_flags &= ~M_PKTHDR;
1908				sc->rl_tail->m_next = m;
1909			}
1910			m = sc->rl_head;
1911			sc->rl_head = sc->rl_tail = NULL;
1912			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
1913		} else
1914			m->m_pkthdr.len = m->m_len =
1915			    (total_len - ETHER_CRC_LEN);
1916
1917#ifdef RE_FIXUP_RX
1918		re_fixup_rx(m);
1919#endif
1920		ifp->if_ipackets++;
1921		m->m_pkthdr.rcvif = ifp;
1922
1923		/* Do RX checksumming if enabled */
1924
1925		if (ifp->if_capenable & IFCAP_RXCSUM) {
1926			if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
1927				/* Check IP header checksum */
1928				if (rxstat & RL_RDESC_STAT_PROTOID)
1929					m->m_pkthdr.csum_flags |=
1930					    CSUM_IP_CHECKED;
1931				if (!(rxstat & RL_RDESC_STAT_IPSUMBAD))
1932					m->m_pkthdr.csum_flags |=
1933					    CSUM_IP_VALID;
1934
1935				/* Check TCP/UDP checksum */
1936				if ((RL_TCPPKT(rxstat) &&
1937				    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
1938				    (RL_UDPPKT(rxstat) &&
1939				     !(rxstat & RL_RDESC_STAT_UDPSUMBAD))) {
1940					m->m_pkthdr.csum_flags |=
1941						CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
1942					m->m_pkthdr.csum_data = 0xffff;
1943				}
1944			} else {
1945				/*
1946				 * RTL8168C/RTL816CP/RTL8111C/RTL8111CP
1947				 */
1948				if ((rxstat & RL_RDESC_STAT_PROTOID) &&
1949				    (rxvlan & RL_RDESC_IPV4))
1950					m->m_pkthdr.csum_flags |=
1951					    CSUM_IP_CHECKED;
1952				if (!(rxstat & RL_RDESC_STAT_IPSUMBAD) &&
1953				    (rxvlan & RL_RDESC_IPV4))
1954					m->m_pkthdr.csum_flags |=
1955					    CSUM_IP_VALID;
1956				if (((rxstat & RL_RDESC_STAT_TCP) &&
1957				    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
1958				    ((rxstat & RL_RDESC_STAT_UDP) &&
1959				    !(rxstat & RL_RDESC_STAT_UDPSUMBAD))) {
1960					m->m_pkthdr.csum_flags |=
1961						CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
1962					m->m_pkthdr.csum_data = 0xffff;
1963				}
1964			}
1965		}
1966		maxpkt--;
1967		if (rxvlan & RL_RDESC_VLANCTL_TAG) {
1968			m->m_pkthdr.ether_vtag =
1969			    bswap16((rxvlan & RL_RDESC_VLANCTL_DATA));
1970			m->m_flags |= M_VLANTAG;
1971		}
1972		RL_UNLOCK(sc);
1973		(*ifp->if_input)(ifp, m);
1974		RL_LOCK(sc);
1975	}
1976
1977	/* Flush the RX DMA ring */
1978
1979	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1980	    sc->rl_ldata.rl_rx_list_map,
1981	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
1982
1983	sc->rl_ldata.rl_rx_prodidx = i;
1984
1985	if (maxpkt)
1986		return(EAGAIN);
1987
1988	return(0);
1989}
1990
1991static void
1992re_txeof(struct rl_softc *sc)
1993{
1994	struct ifnet		*ifp;
1995	struct rl_txdesc	*txd;
1996	u_int32_t		txstat;
1997	int			cons;
1998
1999	cons = sc->rl_ldata.rl_tx_considx;
2000	if (cons == sc->rl_ldata.rl_tx_prodidx)
2001		return;
2002
2003	ifp = sc->rl_ifp;
2004	/* Invalidate the TX descriptor list */
2005	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2006	    sc->rl_ldata.rl_tx_list_map,
2007	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2008
2009	for (; cons != sc->rl_ldata.rl_tx_prodidx;
2010	    cons = RL_TX_DESC_NXT(sc, cons)) {
2011		txstat = le32toh(sc->rl_ldata.rl_tx_list[cons].rl_cmdstat);
2012		if (txstat & RL_TDESC_STAT_OWN)
2013			break;
2014		/*
2015		 * We only stash mbufs in the last descriptor
2016		 * in a fragment chain, which also happens to
2017		 * be the only place where the TX status bits
2018		 * are valid.
2019		 */
2020		if (txstat & RL_TDESC_CMD_EOF) {
2021			txd = &sc->rl_ldata.rl_tx_desc[cons];
2022			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
2023			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
2024			bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag,
2025			    txd->tx_dmamap);
2026			KASSERT(txd->tx_m != NULL,
2027			    ("%s: freeing NULL mbufs!", __func__));
2028			m_freem(txd->tx_m);
2029			txd->tx_m = NULL;
2030			if (txstat & (RL_TDESC_STAT_EXCESSCOL|
2031			    RL_TDESC_STAT_COLCNT))
2032				ifp->if_collisions++;
2033			if (txstat & RL_TDESC_STAT_TXERRSUM)
2034				ifp->if_oerrors++;
2035			else
2036				ifp->if_opackets++;
2037		}
2038		sc->rl_ldata.rl_tx_free++;
2039		ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
2040	}
2041	sc->rl_ldata.rl_tx_considx = cons;
2042
2043	/* No changes made to the TX ring, so no flush needed */
2044
2045	if (sc->rl_ldata.rl_tx_free != sc->rl_ldata.rl_tx_desc_cnt) {
2046#ifdef RE_TX_MODERATION
2047		/*
2048		 * If not all descriptors have been reaped yet, reload
2049		 * the timer so that we will eventually get another
2050		 * interrupt that will cause us to re-enter this routine.
2051		 * This is done in case the transmitter has gone idle.
2052		 */
2053		CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2054#endif
2055	} else
2056		sc->rl_watchdog_timer = 0;
2057}
2058
2059static void
2060re_tick(void *xsc)
2061{
2062	struct rl_softc		*sc;
2063	struct mii_data		*mii;
2064
2065	sc = xsc;
2066
2067	RL_LOCK_ASSERT(sc);
2068
2069	mii = device_get_softc(sc->rl_miibus);
2070	mii_tick(mii);
2071	if ((sc->rl_flags & RL_FLAG_LINK) == 0)
2072		re_miibus_statchg(sc->rl_dev);
2073	re_watchdog(sc);
2074	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
2075}
2076
2077#ifdef DEVICE_POLLING
2078static void
2079re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
2080{
2081	struct rl_softc *sc = ifp->if_softc;
2082
2083	RL_LOCK(sc);
2084	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
2085		re_poll_locked(ifp, cmd, count);
2086	RL_UNLOCK(sc);
2087}
2088
2089static void
2090re_poll_locked(struct ifnet *ifp, enum poll_cmd cmd, int count)
2091{
2092	struct rl_softc *sc = ifp->if_softc;
2093
2094	RL_LOCK_ASSERT(sc);
2095
2096	sc->rxcycles = count;
2097	re_rxeof(sc);
2098	re_txeof(sc);
2099
2100	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2101		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_txtask);
2102
2103	if (cmd == POLL_AND_CHECK_STATUS) { /* also check status register */
2104		u_int16_t       status;
2105
2106		status = CSR_READ_2(sc, RL_ISR);
2107		if (status == 0xffff)
2108			return;
2109		if (status)
2110			CSR_WRITE_2(sc, RL_ISR, status);
2111		if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
2112		    (sc->rl_flags & RL_FLAG_PCIE))
2113			CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2114
2115		/*
2116		 * XXX check behaviour on receiver stalls.
2117		 */
2118
2119		if (status & RL_ISR_SYSTEM_ERR)
2120			re_init_locked(sc);
2121	}
2122}
2123#endif /* DEVICE_POLLING */
2124
2125static int
2126re_intr(void *arg)
2127{
2128	struct rl_softc		*sc;
2129	uint16_t		status;
2130
2131	sc = arg;
2132
2133	status = CSR_READ_2(sc, RL_ISR);
2134	if (status == 0xFFFF || (status & RL_INTRS_CPLUS) == 0)
2135                return (FILTER_STRAY);
2136	CSR_WRITE_2(sc, RL_IMR, 0);
2137
2138	taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_inttask);
2139
2140	return (FILTER_HANDLED);
2141}
2142
2143static void
2144re_int_task(void *arg, int npending)
2145{
2146	struct rl_softc		*sc;
2147	struct ifnet		*ifp;
2148	u_int16_t		status;
2149	int			rval = 0;
2150
2151	sc = arg;
2152	ifp = sc->rl_ifp;
2153
2154	RL_LOCK(sc);
2155
2156	status = CSR_READ_2(sc, RL_ISR);
2157        CSR_WRITE_2(sc, RL_ISR, status);
2158
2159	if (sc->suspended ||
2160	    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2161		RL_UNLOCK(sc);
2162		return;
2163	}
2164
2165#ifdef DEVICE_POLLING
2166	if  (ifp->if_capenable & IFCAP_POLLING) {
2167		RL_UNLOCK(sc);
2168		return;
2169	}
2170#endif
2171
2172	if (status & (RL_ISR_RX_OK|RL_ISR_RX_ERR|RL_ISR_FIFO_OFLOW))
2173		rval = re_rxeof(sc);
2174
2175	/*
2176	 * Some chips will ignore a second TX request issued
2177	 * while an existing transmission is in progress. If
2178	 * the transmitter goes idle but there are still
2179	 * packets waiting to be sent, we need to restart the
2180	 * channel here to flush them out. This only seems to
2181	 * be required with the PCIe devices.
2182	 */
2183	if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
2184	    (sc->rl_flags & RL_FLAG_PCIE))
2185		CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2186	if (status & (
2187#ifdef RE_TX_MODERATION
2188	    RL_ISR_TIMEOUT_EXPIRED|
2189#else
2190	    RL_ISR_TX_OK|
2191#endif
2192	    RL_ISR_TX_ERR|RL_ISR_TX_DESC_UNAVAIL))
2193		re_txeof(sc);
2194
2195	if (status & RL_ISR_SYSTEM_ERR)
2196		re_init_locked(sc);
2197
2198	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2199		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_txtask);
2200
2201	RL_UNLOCK(sc);
2202
2203        if ((CSR_READ_2(sc, RL_ISR) & RL_INTRS_CPLUS) || rval) {
2204		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_inttask);
2205		return;
2206	}
2207
2208	CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2209}
2210
2211static int
2212re_encap(struct rl_softc *sc, struct mbuf **m_head)
2213{
2214	struct rl_txdesc	*txd, *txd_last;
2215	bus_dma_segment_t	segs[RL_NTXSEGS];
2216	bus_dmamap_t		map;
2217	struct mbuf		*m_new;
2218	struct rl_desc		*desc;
2219	int			nsegs, prod;
2220	int			i, error, ei, si;
2221	int			padlen;
2222	uint32_t		cmdstat, csum_flags, vlanctl;
2223
2224	RL_LOCK_ASSERT(sc);
2225	M_ASSERTPKTHDR((*m_head));
2226
2227	/*
2228	 * With some of the RealTek chips, using the checksum offload
2229	 * support in conjunction with the autopadding feature results
2230	 * in the transmission of corrupt frames. For example, if we
2231	 * need to send a really small IP fragment that's less than 60
2232	 * bytes in size, and IP header checksumming is enabled, the
2233	 * resulting ethernet frame that appears on the wire will
2234	 * have garbled payload. To work around this, if TX IP checksum
2235	 * offload is enabled, we always manually pad short frames out
2236	 * to the minimum ethernet frame size.
2237	 */
2238	if ((sc->rl_flags & RL_FLAG_DESCV2) == 0 &&
2239	    (*m_head)->m_pkthdr.len < RL_IP4CSUMTX_PADLEN &&
2240	    ((*m_head)->m_pkthdr.csum_flags & CSUM_IP) != 0) {
2241		padlen = RL_MIN_FRAMELEN - (*m_head)->m_pkthdr.len;
2242		if (M_WRITABLE(*m_head) == 0) {
2243			/* Get a writable copy. */
2244			m_new = m_dup(*m_head, M_DONTWAIT);
2245			m_freem(*m_head);
2246			if (m_new == NULL) {
2247				*m_head = NULL;
2248				return (ENOBUFS);
2249			}
2250			*m_head = m_new;
2251		}
2252		if ((*m_head)->m_next != NULL ||
2253		    M_TRAILINGSPACE(*m_head) < padlen) {
2254			m_new = m_defrag(*m_head, M_DONTWAIT);
2255			if (m_new == NULL) {
2256				m_freem(*m_head);
2257				*m_head = NULL;
2258				return (ENOBUFS);
2259			}
2260		} else
2261			m_new = *m_head;
2262
2263		/*
2264		 * Manually pad short frames, and zero the pad space
2265		 * to avoid leaking data.
2266		 */
2267		bzero(mtod(m_new, char *) + m_new->m_pkthdr.len, padlen);
2268		m_new->m_pkthdr.len += padlen;
2269		m_new->m_len = m_new->m_pkthdr.len;
2270		*m_head = m_new;
2271	}
2272
2273	prod = sc->rl_ldata.rl_tx_prodidx;
2274	txd = &sc->rl_ldata.rl_tx_desc[prod];
2275	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap,
2276	    *m_head, segs, &nsegs, BUS_DMA_NOWAIT);
2277	if (error == EFBIG) {
2278		m_new = m_collapse(*m_head, M_DONTWAIT, RL_NTXSEGS);
2279		if (m_new == NULL) {
2280			m_freem(*m_head);
2281			*m_head = NULL;
2282			return (ENOBUFS);
2283		}
2284		*m_head = m_new;
2285		error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_tx_mtag,
2286		    txd->tx_dmamap, *m_head, segs, &nsegs, BUS_DMA_NOWAIT);
2287		if (error != 0) {
2288			m_freem(*m_head);
2289			*m_head = NULL;
2290			return (error);
2291		}
2292	} else if (error != 0)
2293		return (error);
2294	if (nsegs == 0) {
2295		m_freem(*m_head);
2296		*m_head = NULL;
2297		return (EIO);
2298	}
2299
2300	/* Check for number of available descriptors. */
2301	if (sc->rl_ldata.rl_tx_free - nsegs <= 1) {
2302		bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap);
2303		return (ENOBUFS);
2304	}
2305
2306	bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap,
2307	    BUS_DMASYNC_PREWRITE);
2308
2309	/*
2310	 * Set up checksum offload. Note: checksum offload bits must
2311	 * appear in all descriptors of a multi-descriptor transmit
2312	 * attempt. This is according to testing done with an 8169
2313	 * chip. This is a requirement.
2314	 */
2315	vlanctl = 0;
2316	csum_flags = 0;
2317	if (((*m_head)->m_pkthdr.csum_flags & CSUM_TSO) != 0)
2318		csum_flags = RL_TDESC_CMD_LGSEND |
2319		    ((uint32_t)(*m_head)->m_pkthdr.tso_segsz <<
2320		    RL_TDESC_CMD_MSSVAL_SHIFT);
2321	else {
2322		/*
2323		 * Unconditionally enable IP checksum if TCP or UDP
2324		 * checksum is required. Otherwise, TCP/UDP checksum
2325		 * does't make effects.
2326		 */
2327		if (((*m_head)->m_pkthdr.csum_flags & RE_CSUM_FEATURES) != 0) {
2328			if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
2329				csum_flags |= RL_TDESC_CMD_IPCSUM;
2330				if (((*m_head)->m_pkthdr.csum_flags &
2331				    CSUM_TCP) != 0)
2332					csum_flags |= RL_TDESC_CMD_TCPCSUM;
2333				if (((*m_head)->m_pkthdr.csum_flags &
2334				    CSUM_UDP) != 0)
2335					csum_flags |= RL_TDESC_CMD_UDPCSUM;
2336			} else {
2337				vlanctl |= RL_TDESC_CMD_IPCSUMV2;
2338				if (((*m_head)->m_pkthdr.csum_flags &
2339				    CSUM_TCP) != 0)
2340					vlanctl |= RL_TDESC_CMD_TCPCSUMV2;
2341				if (((*m_head)->m_pkthdr.csum_flags &
2342				    CSUM_UDP) != 0)
2343					vlanctl |= RL_TDESC_CMD_UDPCSUMV2;
2344			}
2345		}
2346	}
2347
2348	/*
2349	 * Set up hardware VLAN tagging. Note: vlan tag info must
2350	 * appear in all descriptors of a multi-descriptor
2351	 * transmission attempt.
2352	 */
2353	if ((*m_head)->m_flags & M_VLANTAG)
2354		vlanctl |= bswap16((*m_head)->m_pkthdr.ether_vtag) |
2355		    RL_TDESC_VLANCTL_TAG;
2356
2357	si = prod;
2358	for (i = 0; i < nsegs; i++, prod = RL_TX_DESC_NXT(sc, prod)) {
2359		desc = &sc->rl_ldata.rl_tx_list[prod];
2360		desc->rl_vlanctl = htole32(vlanctl);
2361		desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[i].ds_addr));
2362		desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[i].ds_addr));
2363		cmdstat = segs[i].ds_len;
2364		if (i != 0)
2365			cmdstat |= RL_TDESC_CMD_OWN;
2366		if (prod == sc->rl_ldata.rl_tx_desc_cnt - 1)
2367			cmdstat |= RL_TDESC_CMD_EOR;
2368		desc->rl_cmdstat = htole32(cmdstat | csum_flags);
2369		sc->rl_ldata.rl_tx_free--;
2370	}
2371	/* Update producer index. */
2372	sc->rl_ldata.rl_tx_prodidx = prod;
2373
2374	/* Set EOF on the last descriptor. */
2375	ei = RL_TX_DESC_PRV(sc, prod);
2376	desc = &sc->rl_ldata.rl_tx_list[ei];
2377	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_EOF);
2378
2379	desc = &sc->rl_ldata.rl_tx_list[si];
2380	/* Set SOF and transfer ownership of packet to the chip. */
2381	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_OWN | RL_TDESC_CMD_SOF);
2382
2383	/*
2384	 * Insure that the map for this transmission
2385	 * is placed at the array index of the last descriptor
2386	 * in this chain.  (Swap last and first dmamaps.)
2387	 */
2388	txd_last = &sc->rl_ldata.rl_tx_desc[ei];
2389	map = txd->tx_dmamap;
2390	txd->tx_dmamap = txd_last->tx_dmamap;
2391	txd_last->tx_dmamap = map;
2392	txd_last->tx_m = *m_head;
2393
2394	return (0);
2395}
2396
2397static void
2398re_tx_task(void *arg, int npending)
2399{
2400	struct ifnet		*ifp;
2401
2402	ifp = arg;
2403	re_start(ifp);
2404}
2405
2406/*
2407 * Main transmit routine for C+ and gigE NICs.
2408 */
2409static void
2410re_start(struct ifnet *ifp)
2411{
2412	struct rl_softc		*sc;
2413	struct mbuf		*m_head;
2414	int			queued;
2415
2416	sc = ifp->if_softc;
2417
2418	RL_LOCK(sc);
2419
2420	if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
2421	    IFF_DRV_RUNNING || (sc->rl_flags & RL_FLAG_LINK) == 0) {
2422		RL_UNLOCK(sc);
2423		return;
2424	}
2425
2426	for (queued = 0; !IFQ_DRV_IS_EMPTY(&ifp->if_snd) &&
2427	    sc->rl_ldata.rl_tx_free > 1;) {
2428		IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
2429		if (m_head == NULL)
2430			break;
2431
2432		if (re_encap(sc, &m_head) != 0) {
2433			if (m_head == NULL)
2434				break;
2435			IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
2436			ifp->if_drv_flags |= IFF_DRV_OACTIVE;
2437			break;
2438		}
2439
2440		/*
2441		 * If there's a BPF listener, bounce a copy of this frame
2442		 * to him.
2443		 */
2444		ETHER_BPF_MTAP(ifp, m_head);
2445
2446		queued++;
2447	}
2448
2449	if (queued == 0) {
2450#ifdef RE_TX_MODERATION
2451		if (sc->rl_ldata.rl_tx_free != sc->rl_ldata.rl_tx_desc_cnt)
2452			CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2453#endif
2454		RL_UNLOCK(sc);
2455		return;
2456	}
2457
2458	/* Flush the TX descriptors */
2459
2460	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2461	    sc->rl_ldata.rl_tx_list_map,
2462	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2463
2464	CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2465
2466#ifdef RE_TX_MODERATION
2467	/*
2468	 * Use the countdown timer for interrupt moderation.
2469	 * 'TX done' interrupts are disabled. Instead, we reset the
2470	 * countdown timer, which will begin counting until it hits
2471	 * the value in the TIMERINT register, and then trigger an
2472	 * interrupt. Each time we write to the TIMERCNT register,
2473	 * the timer count is reset to 0.
2474	 */
2475	CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2476#endif
2477
2478	/*
2479	 * Set a timeout in case the chip goes out to lunch.
2480	 */
2481	sc->rl_watchdog_timer = 5;
2482
2483	RL_UNLOCK(sc);
2484}
2485
2486static void
2487re_init(void *xsc)
2488{
2489	struct rl_softc		*sc = xsc;
2490
2491	RL_LOCK(sc);
2492	re_init_locked(sc);
2493	RL_UNLOCK(sc);
2494}
2495
2496static void
2497re_init_locked(struct rl_softc *sc)
2498{
2499	struct ifnet		*ifp = sc->rl_ifp;
2500	struct mii_data		*mii;
2501	u_int32_t		rxcfg = 0;
2502	uint16_t		cfg;
2503	union {
2504		uint32_t align_dummy;
2505		u_char eaddr[ETHER_ADDR_LEN];
2506        } eaddr;
2507
2508	RL_LOCK_ASSERT(sc);
2509
2510	mii = device_get_softc(sc->rl_miibus);
2511
2512	/*
2513	 * Cancel pending I/O and free all RX/TX buffers.
2514	 */
2515	re_stop(sc);
2516
2517	/* Put controller into known state. */
2518	re_reset(sc);
2519
2520	/*
2521	 * Enable C+ RX and TX mode, as well as VLAN stripping and
2522	 * RX checksum offload. We must configure the C+ register
2523	 * before all others.
2524	 */
2525	cfg = RL_CPLUSCMD_PCI_MRW;
2526	if ((ifp->if_capenable & IFCAP_RXCSUM) != 0)
2527		cfg |= RL_CPLUSCMD_RXCSUM_ENB;
2528	if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0)
2529		cfg |= RL_CPLUSCMD_VLANSTRIP;
2530	if ((sc->rl_flags & RL_FLAG_MACSTAT) != 0) {
2531		cfg |= RL_CPLUSCMD_MACSTAT_DIS;
2532		/* XXX magic. */
2533		cfg |= 0x0001;
2534	} else
2535		cfg |= RL_CPLUSCMD_RXENB | RL_CPLUSCMD_TXENB;
2536	CSR_WRITE_2(sc, RL_CPLUS_CMD, cfg);
2537	/*
2538	 * Disable TSO if interface MTU size is greater than MSS
2539	 * allowed in controller.
2540	 */
2541	if (ifp->if_mtu > RL_TSO_MTU && (ifp->if_capenable & IFCAP_TSO4) != 0) {
2542		ifp->if_capenable &= ~IFCAP_TSO4;
2543		ifp->if_hwassist &= ~CSUM_TSO;
2544	}
2545
2546	/*
2547	 * Init our MAC address.  Even though the chipset
2548	 * documentation doesn't mention it, we need to enter "Config
2549	 * register write enable" mode to modify the ID registers.
2550	 */
2551	/* Copy MAC address on stack to align. */
2552	bcopy(IF_LLADDR(ifp), eaddr.eaddr, ETHER_ADDR_LEN);
2553	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG);
2554	CSR_WRITE_4(sc, RL_IDR0,
2555	    htole32(*(u_int32_t *)(&eaddr.eaddr[0])));
2556	CSR_WRITE_4(sc, RL_IDR4,
2557	    htole32(*(u_int32_t *)(&eaddr.eaddr[4])));
2558	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
2559
2560	/*
2561	 * For C+ mode, initialize the RX descriptors and mbufs.
2562	 */
2563	re_rx_list_init(sc);
2564	re_tx_list_init(sc);
2565
2566	/*
2567	 * Load the addresses of the RX and TX lists into the chip.
2568	 */
2569
2570	CSR_WRITE_4(sc, RL_RXLIST_ADDR_HI,
2571	    RL_ADDR_HI(sc->rl_ldata.rl_rx_list_addr));
2572	CSR_WRITE_4(sc, RL_RXLIST_ADDR_LO,
2573	    RL_ADDR_LO(sc->rl_ldata.rl_rx_list_addr));
2574
2575	CSR_WRITE_4(sc, RL_TXLIST_ADDR_HI,
2576	    RL_ADDR_HI(sc->rl_ldata.rl_tx_list_addr));
2577	CSR_WRITE_4(sc, RL_TXLIST_ADDR_LO,
2578	    RL_ADDR_LO(sc->rl_ldata.rl_tx_list_addr));
2579
2580	/*
2581	 * Enable transmit and receive.
2582	 */
2583	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
2584
2585	/*
2586	 * Set the initial TX and RX configuration.
2587	 */
2588	if (sc->rl_testmode) {
2589		if (sc->rl_type == RL_8169)
2590			CSR_WRITE_4(sc, RL_TXCFG,
2591			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON);
2592		else
2593			CSR_WRITE_4(sc, RL_TXCFG,
2594			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON_CPLUS);
2595	} else
2596		CSR_WRITE_4(sc, RL_TXCFG, RL_TXCFG_CONFIG);
2597
2598	CSR_WRITE_1(sc, RL_EARLY_TX_THRESH, 16);
2599
2600	CSR_WRITE_4(sc, RL_RXCFG, RL_RXCFG_CONFIG);
2601
2602	/* Set the individual bit to receive frames for this host only. */
2603	rxcfg = CSR_READ_4(sc, RL_RXCFG);
2604	rxcfg |= RL_RXCFG_RX_INDIV;
2605
2606	/* If we want promiscuous mode, set the allframes bit. */
2607	if (ifp->if_flags & IFF_PROMISC)
2608		rxcfg |= RL_RXCFG_RX_ALLPHYS;
2609	else
2610		rxcfg &= ~RL_RXCFG_RX_ALLPHYS;
2611	CSR_WRITE_4(sc, RL_RXCFG, rxcfg);
2612
2613	/*
2614	 * Set capture broadcast bit to capture broadcast frames.
2615	 */
2616	if (ifp->if_flags & IFF_BROADCAST)
2617		rxcfg |= RL_RXCFG_RX_BROAD;
2618	else
2619		rxcfg &= ~RL_RXCFG_RX_BROAD;
2620	CSR_WRITE_4(sc, RL_RXCFG, rxcfg);
2621
2622	/*
2623	 * Program the multicast filter, if necessary.
2624	 */
2625	re_setmulti(sc);
2626
2627#ifdef DEVICE_POLLING
2628	/*
2629	 * Disable interrupts if we are polling.
2630	 */
2631	if (ifp->if_capenable & IFCAP_POLLING)
2632		CSR_WRITE_2(sc, RL_IMR, 0);
2633	else	/* otherwise ... */
2634#endif
2635
2636	/*
2637	 * Enable interrupts.
2638	 */
2639	if (sc->rl_testmode)
2640		CSR_WRITE_2(sc, RL_IMR, 0);
2641	else
2642		CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2643	CSR_WRITE_2(sc, RL_ISR, RL_INTRS_CPLUS);
2644
2645	/* Set initial TX threshold */
2646	sc->rl_txthresh = RL_TX_THRESH_INIT;
2647
2648	/* Start RX/TX process. */
2649	CSR_WRITE_4(sc, RL_MISSEDPKT, 0);
2650#ifdef notdef
2651	/* Enable receiver and transmitter. */
2652	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
2653#endif
2654
2655#ifdef RE_TX_MODERATION
2656	/*
2657	 * Initialize the timer interrupt register so that
2658	 * a timer interrupt will be generated once the timer
2659	 * reaches a certain number of ticks. The timer is
2660	 * reloaded on each transmit. This gives us TX interrupt
2661	 * moderation, which dramatically improves TX frame rate.
2662	 */
2663	if (sc->rl_type == RL_8169)
2664		CSR_WRITE_4(sc, RL_TIMERINT_8169, 0x800);
2665	else
2666		CSR_WRITE_4(sc, RL_TIMERINT, 0x400);
2667#endif
2668
2669	/*
2670	 * For 8169 gigE NICs, set the max allowed RX packet
2671	 * size so we can receive jumbo frames.
2672	 */
2673	if (sc->rl_type == RL_8169)
2674		CSR_WRITE_2(sc, RL_MAXRXPKTLEN, 16383);
2675
2676	if (sc->rl_testmode)
2677		return;
2678
2679	mii_mediachg(mii);
2680
2681	CSR_WRITE_1(sc, RL_CFG1, CSR_READ_1(sc, RL_CFG1) | RL_CFG1_DRVLOAD);
2682
2683	ifp->if_drv_flags |= IFF_DRV_RUNNING;
2684	ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
2685
2686	sc->rl_flags &= ~RL_FLAG_LINK;
2687	sc->rl_watchdog_timer = 0;
2688	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
2689}
2690
2691/*
2692 * Set media options.
2693 */
2694static int
2695re_ifmedia_upd(struct ifnet *ifp)
2696{
2697	struct rl_softc		*sc;
2698	struct mii_data		*mii;
2699	int			error;
2700
2701	sc = ifp->if_softc;
2702	mii = device_get_softc(sc->rl_miibus);
2703	RL_LOCK(sc);
2704	error = mii_mediachg(mii);
2705	RL_UNLOCK(sc);
2706
2707	return (error);
2708}
2709
2710/*
2711 * Report current media status.
2712 */
2713static void
2714re_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
2715{
2716	struct rl_softc		*sc;
2717	struct mii_data		*mii;
2718
2719	sc = ifp->if_softc;
2720	mii = device_get_softc(sc->rl_miibus);
2721
2722	RL_LOCK(sc);
2723	mii_pollstat(mii);
2724	RL_UNLOCK(sc);
2725	ifmr->ifm_active = mii->mii_media_active;
2726	ifmr->ifm_status = mii->mii_media_status;
2727}
2728
2729static int
2730re_ioctl(struct ifnet *ifp, u_long command, caddr_t data)
2731{
2732	struct rl_softc		*sc = ifp->if_softc;
2733	struct ifreq		*ifr = (struct ifreq *) data;
2734	struct mii_data		*mii;
2735	int			error = 0;
2736
2737	switch (command) {
2738	case SIOCSIFMTU:
2739		if (ifr->ifr_mtu < ETHERMIN || ifr->ifr_mtu > RL_JUMBO_MTU) {
2740			error = EINVAL;
2741			break;
2742		}
2743		if ((sc->rl_flags & RL_FLAG_NOJUMBO) != 0 &&
2744		    ifr->ifr_mtu > RL_MAX_FRAMELEN) {
2745			error = EINVAL;
2746			break;
2747		}
2748		RL_LOCK(sc);
2749		if (ifp->if_mtu != ifr->ifr_mtu)
2750			ifp->if_mtu = ifr->ifr_mtu;
2751		if (ifp->if_mtu > RL_TSO_MTU &&
2752		    (ifp->if_capenable & IFCAP_TSO4) != 0) {
2753			ifp->if_capenable &= ~IFCAP_TSO4;
2754			ifp->if_hwassist &= ~CSUM_TSO;
2755		}
2756		RL_UNLOCK(sc);
2757		break;
2758	case SIOCSIFFLAGS:
2759		RL_LOCK(sc);
2760		if ((ifp->if_flags & IFF_UP) != 0) {
2761			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
2762				if (((ifp->if_flags ^ sc->rl_if_flags)
2763				    & (IFF_PROMISC | IFF_ALLMULTI)) != 0)
2764					re_setmulti(sc);
2765			} else
2766				re_init_locked(sc);
2767		} else {
2768			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
2769				re_stop(sc);
2770		}
2771		sc->rl_if_flags = ifp->if_flags;
2772		RL_UNLOCK(sc);
2773		break;
2774	case SIOCADDMULTI:
2775	case SIOCDELMULTI:
2776		RL_LOCK(sc);
2777		re_setmulti(sc);
2778		RL_UNLOCK(sc);
2779		break;
2780	case SIOCGIFMEDIA:
2781	case SIOCSIFMEDIA:
2782		mii = device_get_softc(sc->rl_miibus);
2783		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, command);
2784		break;
2785	case SIOCSIFCAP:
2786	    {
2787		int mask, reinit;
2788
2789		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
2790		reinit = 0;
2791#ifdef DEVICE_POLLING
2792		if (mask & IFCAP_POLLING) {
2793			if (ifr->ifr_reqcap & IFCAP_POLLING) {
2794				error = ether_poll_register(re_poll, ifp);
2795				if (error)
2796					return(error);
2797				RL_LOCK(sc);
2798				/* Disable interrupts */
2799				CSR_WRITE_2(sc, RL_IMR, 0x0000);
2800				ifp->if_capenable |= IFCAP_POLLING;
2801				RL_UNLOCK(sc);
2802			} else {
2803				error = ether_poll_deregister(ifp);
2804				/* Enable interrupts. */
2805				RL_LOCK(sc);
2806				CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2807				ifp->if_capenable &= ~IFCAP_POLLING;
2808				RL_UNLOCK(sc);
2809			}
2810		}
2811#endif /* DEVICE_POLLING */
2812		if (mask & IFCAP_HWCSUM) {
2813			ifp->if_capenable ^= IFCAP_HWCSUM;
2814			if (ifp->if_capenable & IFCAP_TXCSUM)
2815				ifp->if_hwassist |= RE_CSUM_FEATURES;
2816			else
2817				ifp->if_hwassist &= ~RE_CSUM_FEATURES;
2818			reinit = 1;
2819		}
2820		if (mask & IFCAP_VLAN_HWTAGGING) {
2821			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
2822			reinit = 1;
2823		}
2824		if (mask & IFCAP_TSO4) {
2825			ifp->if_capenable ^= IFCAP_TSO4;
2826			if ((IFCAP_TSO4 & ifp->if_capenable) &&
2827			    (IFCAP_TSO4 & ifp->if_capabilities))
2828				ifp->if_hwassist |= CSUM_TSO;
2829			else
2830				ifp->if_hwassist &= ~CSUM_TSO;
2831			if (ifp->if_mtu > RL_TSO_MTU &&
2832			    (ifp->if_capenable & IFCAP_TSO4) != 0) {
2833				ifp->if_capenable &= ~IFCAP_TSO4;
2834				ifp->if_hwassist &= ~CSUM_TSO;
2835			}
2836		}
2837		if ((mask & IFCAP_WOL) != 0 &&
2838		    (ifp->if_capabilities & IFCAP_WOL) != 0) {
2839			if ((mask & IFCAP_WOL_UCAST) != 0)
2840				ifp->if_capenable ^= IFCAP_WOL_UCAST;
2841			if ((mask & IFCAP_WOL_MCAST) != 0)
2842				ifp->if_capenable ^= IFCAP_WOL_MCAST;
2843			if ((mask & IFCAP_WOL_MAGIC) != 0)
2844				ifp->if_capenable ^= IFCAP_WOL_MAGIC;
2845		}
2846		if (reinit && ifp->if_drv_flags & IFF_DRV_RUNNING)
2847			re_init(sc);
2848		VLAN_CAPABILITIES(ifp);
2849	    }
2850		break;
2851	default:
2852		error = ether_ioctl(ifp, command, data);
2853		break;
2854	}
2855
2856	return (error);
2857}
2858
2859static void
2860re_watchdog(struct rl_softc *sc)
2861{
2862	struct ifnet		*ifp;
2863
2864	RL_LOCK_ASSERT(sc);
2865
2866	if (sc->rl_watchdog_timer == 0 || --sc->rl_watchdog_timer != 0)
2867		return;
2868
2869	ifp = sc->rl_ifp;
2870	re_txeof(sc);
2871	if (sc->rl_ldata.rl_tx_free == sc->rl_ldata.rl_tx_desc_cnt) {
2872		if_printf(ifp, "watchdog timeout (missed Tx interrupts) "
2873		    "-- recovering\n");
2874		if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2875			taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_txtask);
2876		return;
2877	}
2878
2879	if_printf(ifp, "watchdog timeout\n");
2880	ifp->if_oerrors++;
2881
2882	re_rxeof(sc);
2883	re_init_locked(sc);
2884	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2885		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_txtask);
2886}
2887
2888/*
2889 * Stop the adapter and free any mbufs allocated to the
2890 * RX and TX lists.
2891 */
2892static void
2893re_stop(struct rl_softc *sc)
2894{
2895	int			i;
2896	struct ifnet		*ifp;
2897	struct rl_txdesc	*txd;
2898	struct rl_rxdesc	*rxd;
2899
2900	RL_LOCK_ASSERT(sc);
2901
2902	ifp = sc->rl_ifp;
2903
2904	sc->rl_watchdog_timer = 0;
2905	callout_stop(&sc->rl_stat_callout);
2906	ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
2907
2908	if ((sc->rl_flags & RL_FLAG_CMDSTOP) != 0)
2909		CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_STOPREQ | RL_CMD_TX_ENB |
2910		    RL_CMD_RX_ENB);
2911	else
2912		CSR_WRITE_1(sc, RL_COMMAND, 0x00);
2913	DELAY(1000);
2914	CSR_WRITE_2(sc, RL_IMR, 0x0000);
2915	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
2916
2917	if (sc->rl_head != NULL) {
2918		m_freem(sc->rl_head);
2919		sc->rl_head = sc->rl_tail = NULL;
2920	}
2921
2922	/* Free the TX list buffers. */
2923
2924	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
2925		txd = &sc->rl_ldata.rl_tx_desc[i];
2926		if (txd->tx_m != NULL) {
2927			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
2928			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
2929			bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag,
2930			    txd->tx_dmamap);
2931			m_freem(txd->tx_m);
2932			txd->tx_m = NULL;
2933		}
2934	}
2935
2936	/* Free the RX list buffers. */
2937
2938	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
2939		rxd = &sc->rl_ldata.rl_rx_desc[i];
2940		if (rxd->rx_m != NULL) {
2941			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
2942			    rxd->rx_dmamap, BUS_DMASYNC_POSTREAD);
2943			bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag,
2944			    rxd->rx_dmamap);
2945			m_freem(rxd->rx_m);
2946			rxd->rx_m = NULL;
2947		}
2948	}
2949}
2950
2951/*
2952 * Device suspend routine.  Stop the interface and save some PCI
2953 * settings in case the BIOS doesn't restore them properly on
2954 * resume.
2955 */
2956static int
2957re_suspend(device_t dev)
2958{
2959	struct rl_softc		*sc;
2960
2961	sc = device_get_softc(dev);
2962
2963	RL_LOCK(sc);
2964	re_stop(sc);
2965	re_setwol(sc);
2966	sc->suspended = 1;
2967	RL_UNLOCK(sc);
2968
2969	return (0);
2970}
2971
2972/*
2973 * Device resume routine.  Restore some PCI settings in case the BIOS
2974 * doesn't, re-enable busmastering, and restart the interface if
2975 * appropriate.
2976 */
2977static int
2978re_resume(device_t dev)
2979{
2980	struct rl_softc		*sc;
2981	struct ifnet		*ifp;
2982
2983	sc = device_get_softc(dev);
2984
2985	RL_LOCK(sc);
2986
2987	ifp = sc->rl_ifp;
2988	/* Take controller out of sleep mode. */
2989	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
2990		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
2991			CSR_WRITE_1(sc, RL_GPIO,
2992			    CSR_READ_1(sc, RL_GPIO) | 0x01);
2993	}
2994
2995	/* reinitialize interface if necessary */
2996	if (ifp->if_flags & IFF_UP)
2997		re_init_locked(sc);
2998
2999	/*
3000	 * Clear WOL matching such that normal Rx filtering
3001	 * wouldn't interfere with WOL patterns.
3002	 */
3003	re_clrwol(sc);
3004	sc->suspended = 0;
3005	RL_UNLOCK(sc);
3006
3007	return (0);
3008}
3009
3010/*
3011 * Stop all chip I/O so that the kernel's probe routines don't
3012 * get confused by errant DMAs when rebooting.
3013 */
3014static int
3015re_shutdown(device_t dev)
3016{
3017	struct rl_softc		*sc;
3018
3019	sc = device_get_softc(dev);
3020
3021	RL_LOCK(sc);
3022	re_stop(sc);
3023	/*
3024	 * Mark interface as down since otherwise we will panic if
3025	 * interrupt comes in later on, which can happen in some
3026	 * cases.
3027	 */
3028	sc->rl_ifp->if_flags &= ~IFF_UP;
3029	re_setwol(sc);
3030	RL_UNLOCK(sc);
3031
3032	return (0);
3033}
3034
3035static void
3036re_setwol(struct rl_softc *sc)
3037{
3038	struct ifnet		*ifp;
3039	int			pmc;
3040	uint16_t		pmstat;
3041	uint8_t			v;
3042
3043	RL_LOCK_ASSERT(sc);
3044
3045	if (pci_find_extcap(sc->rl_dev, PCIY_PMG, &pmc) != 0)
3046		return;
3047
3048	ifp = sc->rl_ifp;
3049	/* Put controller into sleep mode. */
3050	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
3051		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
3052			CSR_WRITE_1(sc, RL_GPIO,
3053			    CSR_READ_1(sc, RL_GPIO) & ~0x01);
3054	}
3055	if ((ifp->if_capenable & IFCAP_WOL) != 0 &&
3056	    (sc->rl_flags & RL_FLAG_WOLRXENB) != 0)
3057		CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RX_ENB);
3058	/* Enable config register write. */
3059	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
3060
3061	/* Enable PME. */
3062	v = CSR_READ_1(sc, RL_CFG1);
3063	v &= ~RL_CFG1_PME;
3064	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3065		v |= RL_CFG1_PME;
3066	CSR_WRITE_1(sc, RL_CFG1, v);
3067
3068	v = CSR_READ_1(sc, RL_CFG3);
3069	v &= ~(RL_CFG3_WOL_LINK | RL_CFG3_WOL_MAGIC);
3070	if ((ifp->if_capenable & IFCAP_WOL_MAGIC) != 0)
3071		v |= RL_CFG3_WOL_MAGIC;
3072	CSR_WRITE_1(sc, RL_CFG3, v);
3073
3074	/* Config register write done. */
3075	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3076
3077	v = CSR_READ_1(sc, RL_CFG5);
3078	v &= ~(RL_CFG5_WOL_BCAST | RL_CFG5_WOL_MCAST | RL_CFG5_WOL_UCAST);
3079	v &= ~RL_CFG5_WOL_LANWAKE;
3080	if ((ifp->if_capenable & IFCAP_WOL_UCAST) != 0)
3081		v |= RL_CFG5_WOL_UCAST;
3082	if ((ifp->if_capenable & IFCAP_WOL_MCAST) != 0)
3083		v |= RL_CFG5_WOL_MCAST | RL_CFG5_WOL_BCAST;
3084	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3085		v |= RL_CFG5_WOL_LANWAKE;
3086	CSR_WRITE_1(sc, RL_CFG5, v);
3087
3088	/*
3089	 * It seems that hardware resets its link speed to 100Mbps in
3090	 * power down mode so switching to 100Mbps in driver is not
3091	 * needed.
3092	 */
3093
3094	/* Request PME if WOL is requested. */
3095	pmstat = pci_read_config(sc->rl_dev, pmc + PCIR_POWER_STATUS, 2);
3096	pmstat &= ~(PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE);
3097	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3098		pmstat |= PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE;
3099	pci_write_config(sc->rl_dev, pmc + PCIR_POWER_STATUS, pmstat, 2);
3100}
3101
3102static void
3103re_clrwol(struct rl_softc *sc)
3104{
3105	int			pmc;
3106	uint8_t			v;
3107
3108	RL_LOCK_ASSERT(sc);
3109
3110	if (pci_find_extcap(sc->rl_dev, PCIY_PMG, &pmc) != 0)
3111		return;
3112
3113	/* Enable config register write. */
3114	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
3115
3116	v = CSR_READ_1(sc, RL_CFG3);
3117	v &= ~(RL_CFG3_WOL_LINK | RL_CFG3_WOL_MAGIC);
3118	CSR_WRITE_1(sc, RL_CFG3, v);
3119
3120	/* Config register write done. */
3121	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3122
3123	v = CSR_READ_1(sc, RL_CFG5);
3124	v &= ~(RL_CFG5_WOL_BCAST | RL_CFG5_WOL_MCAST | RL_CFG5_WOL_UCAST);
3125	v &= ~RL_CFG5_WOL_LANWAKE;
3126	CSR_WRITE_1(sc, RL_CFG5, v);
3127}
3128