if_re.c revision 199565
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 199565 2009-11-19 23:14:40Z 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 = 0;
160TUNABLE_INT("hw.re.msi_disable", &msi_disable);
161static int prefer_iomap = 0;
162TUNABLE_INT("hw.re.prefer_iomap", &prefer_iomap);
163
164#define RE_CSUM_FEATURES    (CSUM_IP | CSUM_TCP | CSUM_UDP)
165
166/*
167 * Various supported device vendors/types and their names.
168 */
169static struct rl_type re_devs[] = {
170	{ DLINK_VENDORID, DLINK_DEVICEID_528T, 0,
171	    "D-Link DGE-528(T) Gigabit Ethernet Adapter" },
172	{ RT_VENDORID, RT_DEVICEID_8139, 0,
173	    "RealTek 8139C+ 10/100BaseTX" },
174	{ RT_VENDORID, RT_DEVICEID_8101E, 0,
175	    "RealTek 8101E/8102E/8102EL PCIe 10/100baseTX" },
176	{ RT_VENDORID, RT_DEVICEID_8168, 0,
177	    "RealTek 8168/8168B/8168C/8168CP/8168D/8168DP/"
178	    "8111B/8111C/8111CP/8111DP PCIe Gigabit Ethernet" },
179	{ RT_VENDORID, RT_DEVICEID_8169, 0,
180	    "RealTek 8169/8169S/8169SB(L)/8110S/8110SB(L) Gigabit Ethernet" },
181	{ RT_VENDORID, RT_DEVICEID_8169SC, 0,
182	    "RealTek 8169SC/8110SC Single-chip Gigabit Ethernet" },
183	{ COREGA_VENDORID, COREGA_DEVICEID_CGLAPCIGT, 0,
184	    "Corega CG-LAPCIGT (RTL8169S) Gigabit Ethernet" },
185	{ LINKSYS_VENDORID, LINKSYS_DEVICEID_EG1032, 0,
186	    "Linksys EG1032 (RTL8169S) Gigabit Ethernet" },
187	{ USR_VENDORID, USR_DEVICEID_997902, 0,
188	    "US Robotics 997902 (RTL8169S) Gigabit Ethernet" }
189};
190
191static struct rl_hwrev re_hwrevs[] = {
192	{ RL_HWREV_8139, RL_8139,  "" },
193	{ RL_HWREV_8139A, RL_8139, "A" },
194	{ RL_HWREV_8139AG, RL_8139, "A-G" },
195	{ RL_HWREV_8139B, RL_8139, "B" },
196	{ RL_HWREV_8130, RL_8139, "8130" },
197	{ RL_HWREV_8139C, RL_8139, "C" },
198	{ RL_HWREV_8139D, RL_8139, "8139D/8100B/8100C" },
199	{ RL_HWREV_8139CPLUS, RL_8139CPLUS, "C+"},
200	{ RL_HWREV_8168_SPIN1, RL_8169, "8168"},
201	{ RL_HWREV_8169, RL_8169, "8169"},
202	{ RL_HWREV_8169S, RL_8169, "8169S"},
203	{ RL_HWREV_8110S, RL_8169, "8110S"},
204	{ RL_HWREV_8169_8110SB, RL_8169, "8169SB/8110SB"},
205	{ RL_HWREV_8169_8110SC, RL_8169, "8169SC/8110SC"},
206	{ RL_HWREV_8169_8110SBL, RL_8169, "8169SBL/8110SBL"},
207	{ RL_HWREV_8169_8110SCE, RL_8169, "8169SC/8110SC"},
208	{ RL_HWREV_8100, RL_8139, "8100"},
209	{ RL_HWREV_8101, RL_8139, "8101"},
210	{ RL_HWREV_8100E, RL_8169, "8100E"},
211	{ RL_HWREV_8101E, RL_8169, "8101E"},
212	{ RL_HWREV_8102E, RL_8169, "8102E"},
213	{ RL_HWREV_8102EL, RL_8169, "8102EL"},
214	{ RL_HWREV_8102EL_SPIN1, RL_8169, "8102EL"},
215	{ RL_HWREV_8168_SPIN2, RL_8169, "8168"},
216	{ RL_HWREV_8168_SPIN3, RL_8169, "8168"},
217	{ RL_HWREV_8168C, RL_8169, "8168C/8111C"},
218	{ RL_HWREV_8168C_SPIN2, RL_8169, "8168C/8111C"},
219	{ RL_HWREV_8168CP, RL_8169, "8168CP/8111CP"},
220	{ RL_HWREV_8168D, RL_8169, "8168D/8111D"},
221	{ RL_HWREV_8168DP, RL_8169, "8168DP/8111DP"},
222	{ 0, 0, NULL }
223};
224
225static int re_probe		(device_t);
226static int re_attach		(device_t);
227static int re_detach		(device_t);
228
229static int re_encap		(struct rl_softc *, struct mbuf **);
230
231static void re_dma_map_addr	(void *, bus_dma_segment_t *, int, int);
232static int re_allocmem		(device_t, struct rl_softc *);
233static __inline void re_discard_rxbuf
234				(struct rl_softc *, int);
235static int re_newbuf		(struct rl_softc *, int);
236static int re_rx_list_init	(struct rl_softc *);
237static int re_tx_list_init	(struct rl_softc *);
238#ifdef RE_FIXUP_RX
239static __inline void re_fixup_rx
240				(struct mbuf *);
241#endif
242static int re_rxeof		(struct rl_softc *, int *);
243static void re_txeof		(struct rl_softc *);
244#ifdef DEVICE_POLLING
245static int re_poll		(struct ifnet *, enum poll_cmd, int);
246static int re_poll_locked	(struct ifnet *, enum poll_cmd, int);
247#endif
248static int re_intr		(void *);
249static void re_tick		(void *);
250static void re_tx_task		(void *, int);
251static void re_int_task		(void *, int);
252static void re_start		(struct ifnet *);
253static int re_ioctl		(struct ifnet *, u_long, caddr_t);
254static void re_init		(void *);
255static void re_init_locked	(struct rl_softc *);
256static void re_stop		(struct rl_softc *);
257static void re_watchdog		(struct rl_softc *);
258static int re_suspend		(device_t);
259static int re_resume		(device_t);
260static int re_shutdown		(device_t);
261static int re_ifmedia_upd	(struct ifnet *);
262static void re_ifmedia_sts	(struct ifnet *, struct ifmediareq *);
263
264static void re_eeprom_putbyte	(struct rl_softc *, int);
265static void re_eeprom_getword	(struct rl_softc *, int, u_int16_t *);
266static void re_read_eeprom	(struct rl_softc *, caddr_t, int, int);
267static int re_gmii_readreg	(device_t, int, int);
268static int re_gmii_writereg	(device_t, int, int, int);
269
270static int re_miibus_readreg	(device_t, int, int);
271static int re_miibus_writereg	(device_t, int, int, int);
272static void re_miibus_statchg	(device_t);
273
274static void re_set_rxmode		(struct rl_softc *);
275static void re_reset		(struct rl_softc *);
276static void re_setwol		(struct rl_softc *);
277static void re_clrwol		(struct rl_softc *);
278
279#ifdef RE_DIAG
280static int re_diag		(struct rl_softc *);
281#endif
282
283static device_method_t re_methods[] = {
284	/* Device interface */
285	DEVMETHOD(device_probe,		re_probe),
286	DEVMETHOD(device_attach,	re_attach),
287	DEVMETHOD(device_detach,	re_detach),
288	DEVMETHOD(device_suspend,	re_suspend),
289	DEVMETHOD(device_resume,	re_resume),
290	DEVMETHOD(device_shutdown,	re_shutdown),
291
292	/* bus interface */
293	DEVMETHOD(bus_print_child,	bus_generic_print_child),
294	DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
295
296	/* MII interface */
297	DEVMETHOD(miibus_readreg,	re_miibus_readreg),
298	DEVMETHOD(miibus_writereg,	re_miibus_writereg),
299	DEVMETHOD(miibus_statchg,	re_miibus_statchg),
300
301	{ 0, 0 }
302};
303
304static driver_t re_driver = {
305	"re",
306	re_methods,
307	sizeof(struct rl_softc)
308};
309
310static devclass_t re_devclass;
311
312DRIVER_MODULE(re, pci, re_driver, re_devclass, 0, 0);
313DRIVER_MODULE(miibus, re, miibus_driver, miibus_devclass, 0, 0);
314
315#define EE_SET(x)					\
316	CSR_WRITE_1(sc, RL_EECMD,			\
317		CSR_READ_1(sc, RL_EECMD) | x)
318
319#define EE_CLR(x)					\
320	CSR_WRITE_1(sc, RL_EECMD,			\
321		CSR_READ_1(sc, RL_EECMD) & ~x)
322
323/*
324 * Send a read command and address to the EEPROM, check for ACK.
325 */
326static void
327re_eeprom_putbyte(struct rl_softc *sc, int addr)
328{
329	int			d, i;
330
331	d = addr | (RL_9346_READ << sc->rl_eewidth);
332
333	/*
334	 * Feed in each bit and strobe the clock.
335	 */
336
337	for (i = 1 << (sc->rl_eewidth + 3); i; i >>= 1) {
338		if (d & i) {
339			EE_SET(RL_EE_DATAIN);
340		} else {
341			EE_CLR(RL_EE_DATAIN);
342		}
343		DELAY(100);
344		EE_SET(RL_EE_CLK);
345		DELAY(150);
346		EE_CLR(RL_EE_CLK);
347		DELAY(100);
348	}
349}
350
351/*
352 * Read a word of data stored in the EEPROM at address 'addr.'
353 */
354static void
355re_eeprom_getword(struct rl_softc *sc, int addr, u_int16_t *dest)
356{
357	int			i;
358	u_int16_t		word = 0;
359
360	/*
361	 * Send address of word we want to read.
362	 */
363	re_eeprom_putbyte(sc, addr);
364
365	/*
366	 * Start reading bits from EEPROM.
367	 */
368	for (i = 0x8000; i; i >>= 1) {
369		EE_SET(RL_EE_CLK);
370		DELAY(100);
371		if (CSR_READ_1(sc, RL_EECMD) & RL_EE_DATAOUT)
372			word |= i;
373		EE_CLR(RL_EE_CLK);
374		DELAY(100);
375	}
376
377	*dest = word;
378}
379
380/*
381 * Read a sequence of words from the EEPROM.
382 */
383static void
384re_read_eeprom(struct rl_softc *sc, caddr_t dest, int off, int cnt)
385{
386	int			i;
387	u_int16_t		word = 0, *ptr;
388
389	CSR_SETBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM);
390
391        DELAY(100);
392
393	for (i = 0; i < cnt; i++) {
394		CSR_SETBIT_1(sc, RL_EECMD, RL_EE_SEL);
395		re_eeprom_getword(sc, off + i, &word);
396		CSR_CLRBIT_1(sc, RL_EECMD, RL_EE_SEL);
397		ptr = (u_int16_t *)(dest + (i * 2));
398                *ptr = word;
399	}
400
401	CSR_CLRBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM);
402}
403
404static int
405re_gmii_readreg(device_t dev, int phy, int reg)
406{
407	struct rl_softc		*sc;
408	u_int32_t		rval;
409	int			i;
410
411	if (phy != 1)
412		return (0);
413
414	sc = device_get_softc(dev);
415
416	/* Let the rgephy driver read the GMEDIASTAT register */
417
418	if (reg == RL_GMEDIASTAT) {
419		rval = CSR_READ_1(sc, RL_GMEDIASTAT);
420		return (rval);
421	}
422
423	CSR_WRITE_4(sc, RL_PHYAR, reg << 16);
424	DELAY(1000);
425
426	for (i = 0; i < RL_PHY_TIMEOUT; i++) {
427		rval = CSR_READ_4(sc, RL_PHYAR);
428		if (rval & RL_PHYAR_BUSY)
429			break;
430		DELAY(100);
431	}
432
433	if (i == RL_PHY_TIMEOUT) {
434		device_printf(sc->rl_dev, "PHY read failed\n");
435		return (0);
436	}
437
438	return (rval & RL_PHYAR_PHYDATA);
439}
440
441static int
442re_gmii_writereg(device_t dev, int phy, int reg, int data)
443{
444	struct rl_softc		*sc;
445	u_int32_t		rval;
446	int			i;
447
448	sc = device_get_softc(dev);
449
450	CSR_WRITE_4(sc, RL_PHYAR, (reg << 16) |
451	    (data & RL_PHYAR_PHYDATA) | RL_PHYAR_BUSY);
452	DELAY(1000);
453
454	for (i = 0; i < RL_PHY_TIMEOUT; i++) {
455		rval = CSR_READ_4(sc, RL_PHYAR);
456		if (!(rval & RL_PHYAR_BUSY))
457			break;
458		DELAY(100);
459	}
460
461	if (i == RL_PHY_TIMEOUT) {
462		device_printf(sc->rl_dev, "PHY write failed\n");
463		return (0);
464	}
465
466	return (0);
467}
468
469static int
470re_miibus_readreg(device_t dev, int phy, int reg)
471{
472	struct rl_softc		*sc;
473	u_int16_t		rval = 0;
474	u_int16_t		re8139_reg = 0;
475
476	sc = device_get_softc(dev);
477
478	if (sc->rl_type == RL_8169) {
479		rval = re_gmii_readreg(dev, phy, reg);
480		return (rval);
481	}
482
483	/* Pretend the internal PHY is only at address 0 */
484	if (phy) {
485		return (0);
486	}
487	switch (reg) {
488	case MII_BMCR:
489		re8139_reg = RL_BMCR;
490		break;
491	case MII_BMSR:
492		re8139_reg = RL_BMSR;
493		break;
494	case MII_ANAR:
495		re8139_reg = RL_ANAR;
496		break;
497	case MII_ANER:
498		re8139_reg = RL_ANER;
499		break;
500	case MII_ANLPAR:
501		re8139_reg = RL_LPAR;
502		break;
503	case MII_PHYIDR1:
504	case MII_PHYIDR2:
505		return (0);
506	/*
507	 * Allow the rlphy driver to read the media status
508	 * register. If we have a link partner which does not
509	 * support NWAY, this is the register which will tell
510	 * us the results of parallel detection.
511	 */
512	case RL_MEDIASTAT:
513		rval = CSR_READ_1(sc, RL_MEDIASTAT);
514		return (rval);
515	default:
516		device_printf(sc->rl_dev, "bad phy register\n");
517		return (0);
518	}
519	rval = CSR_READ_2(sc, re8139_reg);
520	if (sc->rl_type == RL_8139CPLUS && re8139_reg == RL_BMCR) {
521		/* 8139C+ has different bit layout. */
522		rval &= ~(BMCR_LOOP | BMCR_ISO);
523	}
524	return (rval);
525}
526
527static int
528re_miibus_writereg(device_t dev, int phy, int reg, int data)
529{
530	struct rl_softc		*sc;
531	u_int16_t		re8139_reg = 0;
532	int			rval = 0;
533
534	sc = device_get_softc(dev);
535
536	if (sc->rl_type == RL_8169) {
537		rval = re_gmii_writereg(dev, phy, reg, data);
538		return (rval);
539	}
540
541	/* Pretend the internal PHY is only at address 0 */
542	if (phy)
543		return (0);
544
545	switch (reg) {
546	case MII_BMCR:
547		re8139_reg = RL_BMCR;
548		if (sc->rl_type == RL_8139CPLUS) {
549			/* 8139C+ has different bit layout. */
550			data &= ~(BMCR_LOOP | BMCR_ISO);
551		}
552		break;
553	case MII_BMSR:
554		re8139_reg = RL_BMSR;
555		break;
556	case MII_ANAR:
557		re8139_reg = RL_ANAR;
558		break;
559	case MII_ANER:
560		re8139_reg = RL_ANER;
561		break;
562	case MII_ANLPAR:
563		re8139_reg = RL_LPAR;
564		break;
565	case MII_PHYIDR1:
566	case MII_PHYIDR2:
567		return (0);
568		break;
569	default:
570		device_printf(sc->rl_dev, "bad phy register\n");
571		return (0);
572	}
573	CSR_WRITE_2(sc, re8139_reg, data);
574	return (0);
575}
576
577static void
578re_miibus_statchg(device_t dev)
579{
580	struct rl_softc		*sc;
581	struct ifnet		*ifp;
582	struct mii_data		*mii;
583
584	sc = device_get_softc(dev);
585	mii = device_get_softc(sc->rl_miibus);
586	ifp = sc->rl_ifp;
587	if (mii == NULL || ifp == NULL ||
588	    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
589		return;
590
591	sc->rl_flags &= ~RL_FLAG_LINK;
592	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
593	    (IFM_ACTIVE | IFM_AVALID)) {
594		switch (IFM_SUBTYPE(mii->mii_media_active)) {
595		case IFM_10_T:
596		case IFM_100_TX:
597			sc->rl_flags |= RL_FLAG_LINK;
598			break;
599		case IFM_1000_T:
600			if ((sc->rl_flags & RL_FLAG_FASTETHER) != 0)
601				break;
602			sc->rl_flags |= RL_FLAG_LINK;
603			break;
604		default:
605			break;
606		}
607	}
608	/*
609	 * RealTek controllers does not provide any interface to
610	 * Tx/Rx MACs for resolved speed, duplex and flow-control
611	 * parameters.
612	 */
613}
614
615/*
616 * Set the RX configuration and 64-bit multicast hash filter.
617 */
618static void
619re_set_rxmode(struct rl_softc *sc)
620{
621	struct ifnet		*ifp;
622	struct ifmultiaddr	*ifma;
623	uint32_t		hashes[2] = { 0, 0 };
624	uint32_t		h, rxfilt;
625
626	RL_LOCK_ASSERT(sc);
627
628	ifp = sc->rl_ifp;
629
630	rxfilt = RL_RXCFG_CONFIG | RL_RXCFG_RX_INDIV | RL_RXCFG_RX_BROAD;
631
632	if (ifp->if_flags & (IFF_ALLMULTI | IFF_PROMISC)) {
633		if (ifp->if_flags & IFF_PROMISC)
634			rxfilt |= RL_RXCFG_RX_ALLPHYS;
635		/*
636		 * Unlike other hardwares, we have to explicitly set
637		 * RL_RXCFG_RX_MULTI to receive multicast frames in
638		 * promiscuous mode.
639		 */
640		rxfilt |= RL_RXCFG_RX_MULTI;
641		hashes[0] = hashes[1] = 0xffffffff;
642		goto done;
643	}
644
645	if_maddr_rlock(ifp);
646	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
647		if (ifma->ifma_addr->sa_family != AF_LINK)
648			continue;
649		h = ether_crc32_be(LLADDR((struct sockaddr_dl *)
650		    ifma->ifma_addr), ETHER_ADDR_LEN) >> 26;
651		if (h < 32)
652			hashes[0] |= (1 << h);
653		else
654			hashes[1] |= (1 << (h - 32));
655	}
656	if_maddr_runlock(ifp);
657
658	if (hashes[0] != 0 || hashes[1] != 0) {
659		/*
660		 * For some unfathomable reason, RealTek decided to
661		 * reverse the order of the multicast hash registers
662		 * in the PCI Express parts.  This means we have to
663		 * write the hash pattern in reverse order for those
664		 * devices.
665		 */
666		if ((sc->rl_flags & RL_FLAG_PCIE) != 0) {
667			h = bswap32(hashes[0]);
668			hashes[0] = bswap32(hashes[1]);
669			hashes[1] = h;
670		}
671		rxfilt |= RL_RXCFG_RX_MULTI;
672	}
673
674done:
675	CSR_WRITE_4(sc, RL_MAR0, hashes[0]);
676	CSR_WRITE_4(sc, RL_MAR4, hashes[1]);
677	CSR_WRITE_4(sc, RL_RXCFG, rxfilt);
678}
679
680static void
681re_reset(struct rl_softc *sc)
682{
683	int			i;
684
685	RL_LOCK_ASSERT(sc);
686
687	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RESET);
688
689	for (i = 0; i < RL_TIMEOUT; i++) {
690		DELAY(10);
691		if (!(CSR_READ_1(sc, RL_COMMAND) & RL_CMD_RESET))
692			break;
693	}
694	if (i == RL_TIMEOUT)
695		device_printf(sc->rl_dev, "reset never completed!\n");
696
697	if ((sc->rl_flags & RL_FLAG_MACRESET) != 0)
698		CSR_WRITE_1(sc, 0x82, 1);
699	if (sc->rl_hwrev == RL_HWREV_8169S)
700		re_gmii_writereg(sc->rl_dev, 1, 0x0b, 0);
701}
702
703#ifdef RE_DIAG
704
705/*
706 * The following routine is designed to test for a defect on some
707 * 32-bit 8169 cards. Some of these NICs have the REQ64# and ACK64#
708 * lines connected to the bus, however for a 32-bit only card, they
709 * should be pulled high. The result of this defect is that the
710 * NIC will not work right if you plug it into a 64-bit slot: DMA
711 * operations will be done with 64-bit transfers, which will fail
712 * because the 64-bit data lines aren't connected.
713 *
714 * There's no way to work around this (short of talking a soldering
715 * iron to the board), however we can detect it. The method we use
716 * here is to put the NIC into digital loopback mode, set the receiver
717 * to promiscuous mode, and then try to send a frame. We then compare
718 * the frame data we sent to what was received. If the data matches,
719 * then the NIC is working correctly, otherwise we know the user has
720 * a defective NIC which has been mistakenly plugged into a 64-bit PCI
721 * slot. In the latter case, there's no way the NIC can work correctly,
722 * so we print out a message on the console and abort the device attach.
723 */
724
725static int
726re_diag(struct rl_softc *sc)
727{
728	struct ifnet		*ifp = sc->rl_ifp;
729	struct mbuf		*m0;
730	struct ether_header	*eh;
731	struct rl_desc		*cur_rx;
732	u_int16_t		status;
733	u_int32_t		rxstat;
734	int			total_len, i, error = 0, phyaddr;
735	u_int8_t		dst[] = { 0x00, 'h', 'e', 'l', 'l', 'o' };
736	u_int8_t		src[] = { 0x00, 'w', 'o', 'r', 'l', 'd' };
737
738	/* Allocate a single mbuf */
739	MGETHDR(m0, M_DONTWAIT, MT_DATA);
740	if (m0 == NULL)
741		return (ENOBUFS);
742
743	RL_LOCK(sc);
744
745	/*
746	 * Initialize the NIC in test mode. This sets the chip up
747	 * so that it can send and receive frames, but performs the
748	 * following special functions:
749	 * - Puts receiver in promiscuous mode
750	 * - Enables digital loopback mode
751	 * - Leaves interrupts turned off
752	 */
753
754	ifp->if_flags |= IFF_PROMISC;
755	sc->rl_testmode = 1;
756	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
757	re_init_locked(sc);
758	sc->rl_flags |= RL_FLAG_LINK;
759	if (sc->rl_type == RL_8169)
760		phyaddr = 1;
761	else
762		phyaddr = 0;
763
764	re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_RESET);
765	for (i = 0; i < RL_TIMEOUT; i++) {
766		status = re_miibus_readreg(sc->rl_dev, phyaddr, MII_BMCR);
767		if (!(status & BMCR_RESET))
768			break;
769	}
770
771	re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_LOOP);
772	CSR_WRITE_2(sc, RL_ISR, RL_INTRS);
773
774	DELAY(100000);
775
776	/* Put some data in the mbuf */
777
778	eh = mtod(m0, struct ether_header *);
779	bcopy ((char *)&dst, eh->ether_dhost, ETHER_ADDR_LEN);
780	bcopy ((char *)&src, eh->ether_shost, ETHER_ADDR_LEN);
781	eh->ether_type = htons(ETHERTYPE_IP);
782	m0->m_pkthdr.len = m0->m_len = ETHER_MIN_LEN - ETHER_CRC_LEN;
783
784	/*
785	 * Queue the packet, start transmission.
786	 * Note: IF_HANDOFF() ultimately calls re_start() for us.
787	 */
788
789	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
790	RL_UNLOCK(sc);
791	/* XXX: re_diag must not be called when in ALTQ mode */
792	IF_HANDOFF(&ifp->if_snd, m0, ifp);
793	RL_LOCK(sc);
794	m0 = NULL;
795
796	/* Wait for it to propagate through the chip */
797
798	DELAY(100000);
799	for (i = 0; i < RL_TIMEOUT; i++) {
800		status = CSR_READ_2(sc, RL_ISR);
801		CSR_WRITE_2(sc, RL_ISR, status);
802		if ((status & (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK)) ==
803		    (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK))
804			break;
805		DELAY(10);
806	}
807
808	if (i == RL_TIMEOUT) {
809		device_printf(sc->rl_dev,
810		    "diagnostic failed, failed to receive packet in"
811		    " loopback mode\n");
812		error = EIO;
813		goto done;
814	}
815
816	/*
817	 * The packet should have been dumped into the first
818	 * entry in the RX DMA ring. Grab it from there.
819	 */
820
821	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
822	    sc->rl_ldata.rl_rx_list_map,
823	    BUS_DMASYNC_POSTREAD);
824	bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag,
825	    sc->rl_ldata.rl_rx_desc[0].rx_dmamap,
826	    BUS_DMASYNC_POSTREAD);
827	bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag,
828	    sc->rl_ldata.rl_rx_desc[0].rx_dmamap);
829
830	m0 = sc->rl_ldata.rl_rx_desc[0].rx_m;
831	sc->rl_ldata.rl_rx_desc[0].rx_m = NULL;
832	eh = mtod(m0, struct ether_header *);
833
834	cur_rx = &sc->rl_ldata.rl_rx_list[0];
835	total_len = RL_RXBYTES(cur_rx);
836	rxstat = le32toh(cur_rx->rl_cmdstat);
837
838	if (total_len != ETHER_MIN_LEN) {
839		device_printf(sc->rl_dev,
840		    "diagnostic failed, received short packet\n");
841		error = EIO;
842		goto done;
843	}
844
845	/* Test that the received packet data matches what we sent. */
846
847	if (bcmp((char *)&eh->ether_dhost, (char *)&dst, ETHER_ADDR_LEN) ||
848	    bcmp((char *)&eh->ether_shost, (char *)&src, ETHER_ADDR_LEN) ||
849	    ntohs(eh->ether_type) != ETHERTYPE_IP) {
850		device_printf(sc->rl_dev, "WARNING, DMA FAILURE!\n");
851		device_printf(sc->rl_dev, "expected TX data: %6D/%6D/0x%x\n",
852		    dst, ":", src, ":", ETHERTYPE_IP);
853		device_printf(sc->rl_dev, "received RX data: %6D/%6D/0x%x\n",
854		    eh->ether_dhost, ":",  eh->ether_shost, ":",
855		    ntohs(eh->ether_type));
856		device_printf(sc->rl_dev, "You may have a defective 32-bit "
857		    "NIC plugged into a 64-bit PCI slot.\n");
858		device_printf(sc->rl_dev, "Please re-install the NIC in a "
859		    "32-bit slot for proper operation.\n");
860		device_printf(sc->rl_dev, "Read the re(4) man page for more "
861		    "details.\n");
862		error = EIO;
863	}
864
865done:
866	/* Turn interface off, release resources */
867
868	sc->rl_testmode = 0;
869	sc->rl_flags &= ~RL_FLAG_LINK;
870	ifp->if_flags &= ~IFF_PROMISC;
871	re_stop(sc);
872	if (m0 != NULL)
873		m_freem(m0);
874
875	RL_UNLOCK(sc);
876
877	return (error);
878}
879
880#endif
881
882/*
883 * Probe for a RealTek 8139C+/8169/8110 chip. Check the PCI vendor and device
884 * IDs against our list and return a device name if we find a match.
885 */
886static int
887re_probe(device_t dev)
888{
889	struct rl_type		*t;
890	uint16_t		devid, vendor;
891	uint16_t		revid, sdevid;
892	int			i;
893
894	vendor = pci_get_vendor(dev);
895	devid = pci_get_device(dev);
896	revid = pci_get_revid(dev);
897	sdevid = pci_get_subdevice(dev);
898
899	if (vendor == LINKSYS_VENDORID && devid == LINKSYS_DEVICEID_EG1032) {
900		if (sdevid != LINKSYS_SUBDEVICE_EG1032_REV3) {
901			/*
902			 * Only attach to rev. 3 of the Linksys EG1032 adapter.
903			 * Rev. 2 is supported by sk(4).
904			 */
905			return (ENXIO);
906		}
907	}
908
909	if (vendor == RT_VENDORID && devid == RT_DEVICEID_8139) {
910		if (revid != 0x20) {
911			/* 8139, let rl(4) take care of this device. */
912			return (ENXIO);
913		}
914	}
915
916	t = re_devs;
917	for (i = 0; i < sizeof(re_devs) / sizeof(re_devs[0]); i++, t++) {
918		if (vendor == t->rl_vid && devid == t->rl_did) {
919			device_set_desc(dev, t->rl_name);
920			return (BUS_PROBE_DEFAULT);
921		}
922	}
923
924	return (ENXIO);
925}
926
927/*
928 * Map a single buffer address.
929 */
930
931static void
932re_dma_map_addr(void *arg, bus_dma_segment_t *segs, int nseg, int error)
933{
934	bus_addr_t		*addr;
935
936	if (error)
937		return;
938
939	KASSERT(nseg == 1, ("too many DMA segments, %d should be 1", nseg));
940	addr = arg;
941	*addr = segs->ds_addr;
942}
943
944static int
945re_allocmem(device_t dev, struct rl_softc *sc)
946{
947	bus_size_t		rx_list_size, tx_list_size;
948	int			error;
949	int			i;
950
951	rx_list_size = sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc);
952	tx_list_size = sc->rl_ldata.rl_tx_desc_cnt * sizeof(struct rl_desc);
953
954	/*
955	 * Allocate the parent bus DMA tag appropriate for PCI.
956	 * In order to use DAC, RL_CPLUSCMD_PCI_DAC bit of RL_CPLUS_CMD
957	 * register should be set. However some RealTek chips are known
958	 * to be buggy on DAC handling, therefore disable DAC by limiting
959	 * DMA address space to 32bit. PCIe variants of RealTek chips
960	 * may not have the limitation but I took safer path.
961	 */
962	error = bus_dma_tag_create(bus_get_dma_tag(dev), 1, 0,
963	    BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL, NULL,
964	    BUS_SPACE_MAXSIZE_32BIT, 0, BUS_SPACE_MAXSIZE_32BIT, 0,
965	    NULL, NULL, &sc->rl_parent_tag);
966	if (error) {
967		device_printf(dev, "could not allocate parent DMA tag\n");
968		return (error);
969	}
970
971	/*
972	 * Allocate map for TX mbufs.
973	 */
974	error = bus_dma_tag_create(sc->rl_parent_tag, 1, 0,
975	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL,
976	    NULL, MCLBYTES * RL_NTXSEGS, RL_NTXSEGS, 4096, 0,
977	    NULL, NULL, &sc->rl_ldata.rl_tx_mtag);
978	if (error) {
979		device_printf(dev, "could not allocate TX DMA tag\n");
980		return (error);
981	}
982
983	/*
984	 * Allocate map for RX mbufs.
985	 */
986
987	error = bus_dma_tag_create(sc->rl_parent_tag, sizeof(uint64_t), 0,
988	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
989	    MCLBYTES, 1, MCLBYTES, 0, NULL, NULL, &sc->rl_ldata.rl_rx_mtag);
990	if (error) {
991		device_printf(dev, "could not allocate RX DMA tag\n");
992		return (error);
993	}
994
995	/*
996	 * Allocate map for TX descriptor list.
997	 */
998	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
999	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1000	    NULL, tx_list_size, 1, tx_list_size, 0,
1001	    NULL, NULL, &sc->rl_ldata.rl_tx_list_tag);
1002	if (error) {
1003		device_printf(dev, "could not allocate TX DMA ring tag\n");
1004		return (error);
1005	}
1006
1007	/* Allocate DMA'able memory for the TX ring */
1008
1009	error = bus_dmamem_alloc(sc->rl_ldata.rl_tx_list_tag,
1010	    (void **)&sc->rl_ldata.rl_tx_list,
1011	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1012	    &sc->rl_ldata.rl_tx_list_map);
1013	if (error) {
1014		device_printf(dev, "could not allocate TX DMA ring\n");
1015		return (error);
1016	}
1017
1018	/* Load the map for the TX ring. */
1019
1020	sc->rl_ldata.rl_tx_list_addr = 0;
1021	error = bus_dmamap_load(sc->rl_ldata.rl_tx_list_tag,
1022	     sc->rl_ldata.rl_tx_list_map, sc->rl_ldata.rl_tx_list,
1023	     tx_list_size, re_dma_map_addr,
1024	     &sc->rl_ldata.rl_tx_list_addr, BUS_DMA_NOWAIT);
1025	if (error != 0 || sc->rl_ldata.rl_tx_list_addr == 0) {
1026		device_printf(dev, "could not load TX DMA ring\n");
1027		return (ENOMEM);
1028	}
1029
1030	/* Create DMA maps for TX buffers */
1031
1032	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
1033		error = bus_dmamap_create(sc->rl_ldata.rl_tx_mtag, 0,
1034		    &sc->rl_ldata.rl_tx_desc[i].tx_dmamap);
1035		if (error) {
1036			device_printf(dev, "could not create DMA map for TX\n");
1037			return (error);
1038		}
1039	}
1040
1041	/*
1042	 * Allocate map for RX descriptor list.
1043	 */
1044	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
1045	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1046	    NULL, rx_list_size, 1, rx_list_size, 0,
1047	    NULL, NULL, &sc->rl_ldata.rl_rx_list_tag);
1048	if (error) {
1049		device_printf(dev, "could not create RX DMA ring tag\n");
1050		return (error);
1051	}
1052
1053	/* Allocate DMA'able memory for the RX ring */
1054
1055	error = bus_dmamem_alloc(sc->rl_ldata.rl_rx_list_tag,
1056	    (void **)&sc->rl_ldata.rl_rx_list,
1057	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1058	    &sc->rl_ldata.rl_rx_list_map);
1059	if (error) {
1060		device_printf(dev, "could not allocate RX DMA ring\n");
1061		return (error);
1062	}
1063
1064	/* Load the map for the RX ring. */
1065
1066	sc->rl_ldata.rl_rx_list_addr = 0;
1067	error = bus_dmamap_load(sc->rl_ldata.rl_rx_list_tag,
1068	     sc->rl_ldata.rl_rx_list_map, sc->rl_ldata.rl_rx_list,
1069	     rx_list_size, re_dma_map_addr,
1070	     &sc->rl_ldata.rl_rx_list_addr, BUS_DMA_NOWAIT);
1071	if (error != 0 || sc->rl_ldata.rl_rx_list_addr == 0) {
1072		device_printf(dev, "could not load RX DMA ring\n");
1073		return (ENOMEM);
1074	}
1075
1076	/* Create DMA maps for RX buffers */
1077
1078	error = bus_dmamap_create(sc->rl_ldata.rl_rx_mtag, 0,
1079	    &sc->rl_ldata.rl_rx_sparemap);
1080	if (error) {
1081		device_printf(dev, "could not create spare DMA map for RX\n");
1082		return (error);
1083	}
1084	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1085		error = bus_dmamap_create(sc->rl_ldata.rl_rx_mtag, 0,
1086		    &sc->rl_ldata.rl_rx_desc[i].rx_dmamap);
1087		if (error) {
1088			device_printf(dev, "could not create DMA map for RX\n");
1089			return (error);
1090		}
1091	}
1092
1093	return (0);
1094}
1095
1096/*
1097 * Attach the interface. Allocate softc structures, do ifmedia
1098 * setup and ethernet/BPF attach.
1099 */
1100static int
1101re_attach(device_t dev)
1102{
1103	u_char			eaddr[ETHER_ADDR_LEN];
1104	u_int16_t		as[ETHER_ADDR_LEN / 2];
1105	struct rl_softc		*sc;
1106	struct ifnet		*ifp;
1107	struct rl_hwrev		*hw_rev;
1108	int			hwrev;
1109	u_int16_t		devid, re_did = 0;
1110	int			error = 0, rid, i;
1111	int			msic, reg;
1112	uint8_t			cfg;
1113
1114	sc = device_get_softc(dev);
1115	sc->rl_dev = dev;
1116
1117	mtx_init(&sc->rl_mtx, device_get_nameunit(dev), MTX_NETWORK_LOCK,
1118	    MTX_DEF);
1119	callout_init_mtx(&sc->rl_stat_callout, &sc->rl_mtx, 0);
1120
1121	/*
1122	 * Map control/status registers.
1123	 */
1124	pci_enable_busmaster(dev);
1125
1126	devid = pci_get_device(dev);
1127	/*
1128	 * Prefer memory space register mapping over IO space.
1129	 * Because RTL8169SC does not seem to work when memory mapping
1130	 * is used always activate io mapping.
1131	 */
1132	if (devid == RT_DEVICEID_8169SC)
1133		prefer_iomap = 1;
1134	if (prefer_iomap == 0) {
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	} else {
1141		sc->rl_res_id = PCIR_BAR(0);
1142		sc->rl_res_type = SYS_RES_IOPORT;
1143	}
1144	sc->rl_res = bus_alloc_resource_any(dev, sc->rl_res_type,
1145	    &sc->rl_res_id, RF_ACTIVE);
1146	if (sc->rl_res == NULL && prefer_iomap == 0) {
1147		sc->rl_res_id = PCIR_BAR(0);
1148		sc->rl_res_type = SYS_RES_IOPORT;
1149		sc->rl_res = bus_alloc_resource_any(dev, sc->rl_res_type,
1150		    &sc->rl_res_id, RF_ACTIVE);
1151	}
1152	if (sc->rl_res == NULL) {
1153		device_printf(dev, "couldn't map ports/memory\n");
1154		error = ENXIO;
1155		goto fail;
1156	}
1157
1158	sc->rl_btag = rman_get_bustag(sc->rl_res);
1159	sc->rl_bhandle = rman_get_bushandle(sc->rl_res);
1160
1161	msic = 0;
1162	if (pci_find_extcap(dev, PCIY_EXPRESS, &reg) == 0) {
1163		sc->rl_flags |= RL_FLAG_PCIE;
1164		msic = pci_msi_count(dev);
1165		if (bootverbose)
1166			device_printf(dev, "MSI count : %d\n", msic);
1167	}
1168	if (msic > 0 && msi_disable == 0) {
1169		msic = 1;
1170		if (pci_alloc_msi(dev, &msic) == 0) {
1171			if (msic == RL_MSI_MESSAGES) {
1172				device_printf(dev, "Using %d MSI messages\n",
1173				    msic);
1174				sc->rl_flags |= RL_FLAG_MSI;
1175				/* Explicitly set MSI enable bit. */
1176				CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1177				cfg = CSR_READ_1(sc, RL_CFG2);
1178				cfg |= RL_CFG2_MSI;
1179				CSR_WRITE_1(sc, RL_CFG2, cfg);
1180				CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1181			} else
1182				pci_release_msi(dev);
1183		}
1184	}
1185
1186	/* Allocate interrupt */
1187	if ((sc->rl_flags & RL_FLAG_MSI) == 0) {
1188		rid = 0;
1189		sc->rl_irq[0] = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
1190		    RF_SHAREABLE | RF_ACTIVE);
1191		if (sc->rl_irq[0] == NULL) {
1192			device_printf(dev, "couldn't allocate IRQ resources\n");
1193			error = ENXIO;
1194			goto fail;
1195		}
1196	} else {
1197		for (i = 0, rid = 1; i < RL_MSI_MESSAGES; i++, rid++) {
1198			sc->rl_irq[i] = bus_alloc_resource_any(dev,
1199			    SYS_RES_IRQ, &rid, RF_ACTIVE);
1200			if (sc->rl_irq[i] == NULL) {
1201				device_printf(dev,
1202				    "couldn't llocate IRQ resources for "
1203				    "message %d\n", rid);
1204				error = ENXIO;
1205				goto fail;
1206			}
1207		}
1208	}
1209
1210	if ((sc->rl_flags & RL_FLAG_MSI) == 0) {
1211		CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1212		cfg = CSR_READ_1(sc, RL_CFG2);
1213		if ((cfg & RL_CFG2_MSI) != 0) {
1214			device_printf(dev, "turning off MSI enable bit.\n");
1215			cfg &= ~RL_CFG2_MSI;
1216			CSR_WRITE_1(sc, RL_CFG2, cfg);
1217		}
1218		CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1219	}
1220
1221	/* Reset the adapter. */
1222	RL_LOCK(sc);
1223	re_reset(sc);
1224	RL_UNLOCK(sc);
1225
1226	hw_rev = re_hwrevs;
1227	hwrev = CSR_READ_4(sc, RL_TXCFG);
1228	switch (hwrev & 0x70000000) {
1229	case 0x00000000:
1230	case 0x10000000:
1231		device_printf(dev, "Chip rev. 0x%08x\n", hwrev & 0xfc800000);
1232		hwrev &= (RL_TXCFG_HWREV | 0x80000000);
1233		break;
1234	default:
1235		device_printf(dev, "Chip rev. 0x%08x\n", hwrev & 0x7c800000);
1236		hwrev &= RL_TXCFG_HWREV;
1237		break;
1238	}
1239	device_printf(dev, "MAC rev. 0x%08x\n", hwrev & 0x00700000);
1240	while (hw_rev->rl_desc != NULL) {
1241		if (hw_rev->rl_rev == hwrev) {
1242			sc->rl_type = hw_rev->rl_type;
1243			sc->rl_hwrev = hw_rev->rl_rev;
1244			break;
1245		}
1246		hw_rev++;
1247	}
1248	if (hw_rev->rl_desc == NULL) {
1249		device_printf(dev, "Unknown H/W revision: 0x%08x\n", hwrev);
1250		error = ENXIO;
1251		goto fail;
1252	}
1253
1254	switch (hw_rev->rl_rev) {
1255	case RL_HWREV_8139CPLUS:
1256		sc->rl_flags |= RL_FLAG_NOJUMBO | RL_FLAG_FASTETHER |
1257		    RL_FLAG_AUTOPAD;
1258		break;
1259	case RL_HWREV_8100E:
1260	case RL_HWREV_8101E:
1261		sc->rl_flags |= RL_FLAG_NOJUMBO | RL_FLAG_PHYWAKE |
1262		    RL_FLAG_FASTETHER;
1263		break;
1264	case RL_HWREV_8102E:
1265	case RL_HWREV_8102EL:
1266	case RL_HWREV_8102EL_SPIN1:
1267		sc->rl_flags |= RL_FLAG_NOJUMBO | RL_FLAG_PHYWAKE |
1268		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
1269		    RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD;
1270		break;
1271	case RL_HWREV_8168_SPIN1:
1272	case RL_HWREV_8168_SPIN2:
1273		sc->rl_flags |= RL_FLAG_WOLRXENB;
1274		/* FALLTHROUGH */
1275	case RL_HWREV_8168_SPIN3:
1276		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_MACSTAT;
1277		break;
1278	case RL_HWREV_8168C_SPIN2:
1279		sc->rl_flags |= RL_FLAG_MACSLEEP;
1280		/* FALLTHROUGH */
1281	case RL_HWREV_8168C:
1282		if ((hwrev & 0x00700000) == 0x00200000)
1283			sc->rl_flags |= RL_FLAG_MACSLEEP;
1284		/* FALLTHROUGH */
1285	case RL_HWREV_8168CP:
1286	case RL_HWREV_8168D:
1287	case RL_HWREV_8168DP:
1288		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
1289		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP |
1290		    RL_FLAG_AUTOPAD;
1291		/*
1292		 * These controllers support jumbo frame but it seems
1293		 * that enabling it requires touching additional magic
1294		 * registers. Depending on MAC revisions some
1295		 * controllers need to disable checksum offload. So
1296		 * disable jumbo frame until I have better idea what
1297		 * it really requires to make it support.
1298		 * RTL8168C/CP : supports up to 6KB jumbo frame.
1299		 * RTL8111C/CP : supports up to 9KB jumbo frame.
1300		 */
1301		sc->rl_flags |= RL_FLAG_NOJUMBO;
1302		break;
1303	case RL_HWREV_8169_8110SB:
1304	case RL_HWREV_8169_8110SBL:
1305	case RL_HWREV_8169_8110SC:
1306	case RL_HWREV_8169_8110SCE:
1307		sc->rl_flags |= RL_FLAG_PHYWAKE;
1308		/* FALLTHROUGH */
1309	case RL_HWREV_8169:
1310	case RL_HWREV_8169S:
1311	case RL_HWREV_8110S:
1312		sc->rl_flags |= RL_FLAG_MACRESET;
1313		break;
1314	default:
1315		break;
1316	}
1317
1318	/* Enable PME. */
1319	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1320	cfg = CSR_READ_1(sc, RL_CFG1);
1321	cfg |= RL_CFG1_PME;
1322	CSR_WRITE_1(sc, RL_CFG1, cfg);
1323	cfg = CSR_READ_1(sc, RL_CFG5);
1324	cfg &= RL_CFG5_PME_STS;
1325	CSR_WRITE_1(sc, RL_CFG5, cfg);
1326	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1327
1328	if ((sc->rl_flags & RL_FLAG_PAR) != 0) {
1329		/*
1330		 * XXX Should have a better way to extract station
1331		 * address from EEPROM.
1332		 */
1333		for (i = 0; i < ETHER_ADDR_LEN; i++)
1334			eaddr[i] = CSR_READ_1(sc, RL_IDR0 + i);
1335	} else {
1336		sc->rl_eewidth = RL_9356_ADDR_LEN;
1337		re_read_eeprom(sc, (caddr_t)&re_did, 0, 1);
1338		if (re_did != 0x8129)
1339			sc->rl_eewidth = RL_9346_ADDR_LEN;
1340
1341		/*
1342		 * Get station address from the EEPROM.
1343		 */
1344		re_read_eeprom(sc, (caddr_t)as, RL_EE_EADDR, 3);
1345		for (i = 0; i < ETHER_ADDR_LEN / 2; i++)
1346			as[i] = le16toh(as[i]);
1347		bcopy(as, eaddr, sizeof(eaddr));
1348	}
1349
1350	if (sc->rl_type == RL_8169) {
1351		/* Set RX length mask and number of descriptors. */
1352		sc->rl_rxlenmask = RL_RDESC_STAT_GFRAGLEN;
1353		sc->rl_txstart = RL_GTXSTART;
1354		sc->rl_ldata.rl_tx_desc_cnt = RL_8169_TX_DESC_CNT;
1355		sc->rl_ldata.rl_rx_desc_cnt = RL_8169_RX_DESC_CNT;
1356	} else {
1357		/* Set RX length mask and number of descriptors. */
1358		sc->rl_rxlenmask = RL_RDESC_STAT_FRAGLEN;
1359		sc->rl_txstart = RL_TXSTART;
1360		sc->rl_ldata.rl_tx_desc_cnt = RL_8139_TX_DESC_CNT;
1361		sc->rl_ldata.rl_rx_desc_cnt = RL_8139_RX_DESC_CNT;
1362	}
1363
1364	error = re_allocmem(dev, sc);
1365	if (error)
1366		goto fail;
1367
1368	ifp = sc->rl_ifp = if_alloc(IFT_ETHER);
1369	if (ifp == NULL) {
1370		device_printf(dev, "can not if_alloc()\n");
1371		error = ENOSPC;
1372		goto fail;
1373	}
1374
1375	/* Take controller out of deep sleep mode. */
1376	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
1377		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
1378			CSR_WRITE_1(sc, RL_GPIO,
1379			    CSR_READ_1(sc, RL_GPIO) | 0x01);
1380		else
1381			CSR_WRITE_1(sc, RL_GPIO,
1382			    CSR_READ_1(sc, RL_GPIO) & ~0x01);
1383	}
1384
1385	/* Take PHY out of power down mode. */
1386	if ((sc->rl_flags & RL_FLAG_PHYWAKE) != 0) {
1387		re_gmii_writereg(dev, 1, 0x1f, 0);
1388		re_gmii_writereg(dev, 1, 0x0e, 0);
1389	}
1390
1391	/* Do MII setup */
1392	if (mii_phy_probe(dev, &sc->rl_miibus,
1393	    re_ifmedia_upd, re_ifmedia_sts)) {
1394		device_printf(dev, "MII without any phy!\n");
1395		error = ENXIO;
1396		goto fail;
1397	}
1398
1399	ifp->if_softc = sc;
1400	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1401	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1402	ifp->if_ioctl = re_ioctl;
1403	ifp->if_start = re_start;
1404	ifp->if_hwassist = RE_CSUM_FEATURES;
1405	ifp->if_capabilities = IFCAP_HWCSUM;
1406	ifp->if_capenable = ifp->if_capabilities;
1407	ifp->if_init = re_init;
1408	IFQ_SET_MAXLEN(&ifp->if_snd, RL_IFQ_MAXLEN);
1409	ifp->if_snd.ifq_drv_maxlen = RL_IFQ_MAXLEN;
1410	IFQ_SET_READY(&ifp->if_snd);
1411
1412	TASK_INIT(&sc->rl_txtask, 1, re_tx_task, ifp);
1413	TASK_INIT(&sc->rl_inttask, 0, re_int_task, sc);
1414
1415	/*
1416	 * XXX
1417	 * Still have no idea how to make TSO work on 8168C, 8168CP,
1418	 * 8111C and 8111CP.
1419	 */
1420	if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
1421		ifp->if_hwassist |= CSUM_TSO;
1422		ifp->if_capabilities |= IFCAP_TSO4;
1423	}
1424
1425	/*
1426	 * Call MI attach routine.
1427	 */
1428	ether_ifattach(ifp, eaddr);
1429
1430	/* VLAN capability setup */
1431	ifp->if_capabilities |= IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING;
1432	if (ifp->if_capabilities & IFCAP_HWCSUM)
1433		ifp->if_capabilities |= IFCAP_VLAN_HWCSUM;
1434	/* Enable WOL if PM is supported. */
1435	if (pci_find_extcap(sc->rl_dev, PCIY_PMG, &reg) == 0)
1436		ifp->if_capabilities |= IFCAP_WOL;
1437	ifp->if_capenable = ifp->if_capabilities;
1438	/*
1439	 * Don't enable TSO by default. Under certain
1440	 * circumtances the controller generated corrupted
1441	 * packets in TSO size.
1442	 */
1443	ifp->if_hwassist &= ~CSUM_TSO;
1444	ifp->if_capenable &= ~IFCAP_TSO4;
1445#ifdef DEVICE_POLLING
1446	ifp->if_capabilities |= IFCAP_POLLING;
1447#endif
1448	/*
1449	 * Tell the upper layer(s) we support long frames.
1450	 * Must appear after the call to ether_ifattach() because
1451	 * ether_ifattach() sets ifi_hdrlen to the default value.
1452	 */
1453	ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
1454
1455#ifdef RE_DIAG
1456	/*
1457	 * Perform hardware diagnostic on the original RTL8169.
1458	 * Some 32-bit cards were incorrectly wired and would
1459	 * malfunction if plugged into a 64-bit slot.
1460	 */
1461
1462	if (hwrev == RL_HWREV_8169) {
1463		error = re_diag(sc);
1464		if (error) {
1465			device_printf(dev,
1466		    	"attach aborted due to hardware diag failure\n");
1467			ether_ifdetach(ifp);
1468			goto fail;
1469		}
1470	}
1471#endif
1472
1473	/* Hook interrupt last to avoid having to lock softc */
1474	if ((sc->rl_flags & RL_FLAG_MSI) == 0)
1475		error = bus_setup_intr(dev, sc->rl_irq[0],
1476		    INTR_TYPE_NET | INTR_MPSAFE, re_intr, NULL, sc,
1477		    &sc->rl_intrhand[0]);
1478	else {
1479		for (i = 0; i < RL_MSI_MESSAGES; i++) {
1480			error = bus_setup_intr(dev, sc->rl_irq[i],
1481			    INTR_TYPE_NET | INTR_MPSAFE, re_intr, NULL, sc,
1482		    	    &sc->rl_intrhand[i]);
1483			if (error != 0)
1484				break;
1485		}
1486	}
1487	if (error) {
1488		device_printf(dev, "couldn't set up irq\n");
1489		ether_ifdetach(ifp);
1490	}
1491
1492fail:
1493
1494	if (error)
1495		re_detach(dev);
1496
1497	return (error);
1498}
1499
1500/*
1501 * Shutdown hardware and free up resources. This can be called any
1502 * time after the mutex has been initialized. It is called in both
1503 * the error case in attach and the normal detach case so it needs
1504 * to be careful about only freeing resources that have actually been
1505 * allocated.
1506 */
1507static int
1508re_detach(device_t dev)
1509{
1510	struct rl_softc		*sc;
1511	struct ifnet		*ifp;
1512	int			i, rid;
1513
1514	sc = device_get_softc(dev);
1515	ifp = sc->rl_ifp;
1516	KASSERT(mtx_initialized(&sc->rl_mtx), ("re mutex not initialized"));
1517
1518	/* These should only be active if attach succeeded */
1519	if (device_is_attached(dev)) {
1520#ifdef DEVICE_POLLING
1521		if (ifp->if_capenable & IFCAP_POLLING)
1522			ether_poll_deregister(ifp);
1523#endif
1524		RL_LOCK(sc);
1525#if 0
1526		sc->suspended = 1;
1527#endif
1528		re_stop(sc);
1529		RL_UNLOCK(sc);
1530		callout_drain(&sc->rl_stat_callout);
1531		taskqueue_drain(taskqueue_fast, &sc->rl_inttask);
1532		taskqueue_drain(taskqueue_fast, &sc->rl_txtask);
1533		/*
1534		 * Force off the IFF_UP flag here, in case someone
1535		 * still had a BPF descriptor attached to this
1536		 * interface. If they do, ether_ifdetach() will cause
1537		 * the BPF code to try and clear the promisc mode
1538		 * flag, which will bubble down to re_ioctl(),
1539		 * which will try to call re_init() again. This will
1540		 * turn the NIC back on and restart the MII ticker,
1541		 * which will panic the system when the kernel tries
1542		 * to invoke the re_tick() function that isn't there
1543		 * anymore.
1544		 */
1545		ifp->if_flags &= ~IFF_UP;
1546		ether_ifdetach(ifp);
1547	}
1548	if (sc->rl_miibus)
1549		device_delete_child(dev, sc->rl_miibus);
1550	bus_generic_detach(dev);
1551
1552	/*
1553	 * The rest is resource deallocation, so we should already be
1554	 * stopped here.
1555	 */
1556
1557	for (i = 0; i < RL_MSI_MESSAGES; i++) {
1558		if (sc->rl_intrhand[i] != NULL) {
1559			bus_teardown_intr(dev, sc->rl_irq[i],
1560			    sc->rl_intrhand[i]);
1561			sc->rl_intrhand[i] = NULL;
1562		}
1563	}
1564	if (ifp != NULL)
1565		if_free(ifp);
1566	if ((sc->rl_flags & RL_FLAG_MSI) == 0) {
1567		if (sc->rl_irq[0] != NULL) {
1568			bus_release_resource(dev, SYS_RES_IRQ, 0,
1569			    sc->rl_irq[0]);
1570			sc->rl_irq[0] = NULL;
1571		}
1572	} else {
1573		for (i = 0, rid = 1; i < RL_MSI_MESSAGES; i++, rid++) {
1574			if (sc->rl_irq[i] != NULL) {
1575				bus_release_resource(dev, SYS_RES_IRQ, rid,
1576				    sc->rl_irq[i]);
1577				sc->rl_irq[i] = NULL;
1578			}
1579		}
1580		pci_release_msi(dev);
1581	}
1582	if (sc->rl_res)
1583		bus_release_resource(dev, sc->rl_res_type, sc->rl_res_id,
1584		    sc->rl_res);
1585
1586	/* Unload and free the RX DMA ring memory and map */
1587
1588	if (sc->rl_ldata.rl_rx_list_tag) {
1589		bus_dmamap_unload(sc->rl_ldata.rl_rx_list_tag,
1590		    sc->rl_ldata.rl_rx_list_map);
1591		bus_dmamem_free(sc->rl_ldata.rl_rx_list_tag,
1592		    sc->rl_ldata.rl_rx_list,
1593		    sc->rl_ldata.rl_rx_list_map);
1594		bus_dma_tag_destroy(sc->rl_ldata.rl_rx_list_tag);
1595	}
1596
1597	/* Unload and free the TX DMA ring memory and map */
1598
1599	if (sc->rl_ldata.rl_tx_list_tag) {
1600		bus_dmamap_unload(sc->rl_ldata.rl_tx_list_tag,
1601		    sc->rl_ldata.rl_tx_list_map);
1602		bus_dmamem_free(sc->rl_ldata.rl_tx_list_tag,
1603		    sc->rl_ldata.rl_tx_list,
1604		    sc->rl_ldata.rl_tx_list_map);
1605		bus_dma_tag_destroy(sc->rl_ldata.rl_tx_list_tag);
1606	}
1607
1608	/* Destroy all the RX and TX buffer maps */
1609
1610	if (sc->rl_ldata.rl_tx_mtag) {
1611		for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++)
1612			bus_dmamap_destroy(sc->rl_ldata.rl_tx_mtag,
1613			    sc->rl_ldata.rl_tx_desc[i].tx_dmamap);
1614		bus_dma_tag_destroy(sc->rl_ldata.rl_tx_mtag);
1615	}
1616	if (sc->rl_ldata.rl_rx_mtag) {
1617		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++)
1618			bus_dmamap_destroy(sc->rl_ldata.rl_rx_mtag,
1619			    sc->rl_ldata.rl_rx_desc[i].rx_dmamap);
1620		if (sc->rl_ldata.rl_rx_sparemap)
1621			bus_dmamap_destroy(sc->rl_ldata.rl_rx_mtag,
1622			    sc->rl_ldata.rl_rx_sparemap);
1623		bus_dma_tag_destroy(sc->rl_ldata.rl_rx_mtag);
1624	}
1625
1626	/* Unload and free the stats buffer and map */
1627
1628	if (sc->rl_ldata.rl_stag) {
1629		bus_dmamap_unload(sc->rl_ldata.rl_stag,
1630		    sc->rl_ldata.rl_rx_list_map);
1631		bus_dmamem_free(sc->rl_ldata.rl_stag,
1632		    sc->rl_ldata.rl_stats,
1633		    sc->rl_ldata.rl_smap);
1634		bus_dma_tag_destroy(sc->rl_ldata.rl_stag);
1635	}
1636
1637	if (sc->rl_parent_tag)
1638		bus_dma_tag_destroy(sc->rl_parent_tag);
1639
1640	mtx_destroy(&sc->rl_mtx);
1641
1642	return (0);
1643}
1644
1645static __inline void
1646re_discard_rxbuf(struct rl_softc *sc, int idx)
1647{
1648	struct rl_desc		*desc;
1649	struct rl_rxdesc	*rxd;
1650	uint32_t		cmdstat;
1651
1652	rxd = &sc->rl_ldata.rl_rx_desc[idx];
1653	desc = &sc->rl_ldata.rl_rx_list[idx];
1654	desc->rl_vlanctl = 0;
1655	cmdstat = rxd->rx_size;
1656	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
1657		cmdstat |= RL_RDESC_CMD_EOR;
1658	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
1659}
1660
1661static int
1662re_newbuf(struct rl_softc *sc, int idx)
1663{
1664	struct mbuf		*m;
1665	struct rl_rxdesc	*rxd;
1666	bus_dma_segment_t	segs[1];
1667	bus_dmamap_t		map;
1668	struct rl_desc		*desc;
1669	uint32_t		cmdstat;
1670	int			error, nsegs;
1671
1672	m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
1673	if (m == NULL)
1674		return (ENOBUFS);
1675
1676	m->m_len = m->m_pkthdr.len = MCLBYTES;
1677#ifdef RE_FIXUP_RX
1678	/*
1679	 * This is part of an evil trick to deal with non-x86 platforms.
1680	 * The RealTek chip requires RX buffers to be aligned on 64-bit
1681	 * boundaries, but that will hose non-x86 machines. To get around
1682	 * this, we leave some empty space at the start of each buffer
1683	 * and for non-x86 hosts, we copy the buffer back six bytes
1684	 * to achieve word alignment. This is slightly more efficient
1685	 * than allocating a new buffer, copying the contents, and
1686	 * discarding the old buffer.
1687	 */
1688	m_adj(m, RE_ETHER_ALIGN);
1689#endif
1690	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_rx_mtag,
1691	    sc->rl_ldata.rl_rx_sparemap, m, segs, &nsegs, BUS_DMA_NOWAIT);
1692	if (error != 0) {
1693		m_freem(m);
1694		return (ENOBUFS);
1695	}
1696	KASSERT(nsegs == 1, ("%s: %d segment returned!", __func__, nsegs));
1697
1698	rxd = &sc->rl_ldata.rl_rx_desc[idx];
1699	if (rxd->rx_m != NULL) {
1700		bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap,
1701		    BUS_DMASYNC_POSTREAD);
1702		bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap);
1703	}
1704
1705	rxd->rx_m = m;
1706	map = rxd->rx_dmamap;
1707	rxd->rx_dmamap = sc->rl_ldata.rl_rx_sparemap;
1708	rxd->rx_size = segs[0].ds_len;
1709	sc->rl_ldata.rl_rx_sparemap = map;
1710	bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap,
1711	    BUS_DMASYNC_PREREAD);
1712
1713	desc = &sc->rl_ldata.rl_rx_list[idx];
1714	desc->rl_vlanctl = 0;
1715	desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[0].ds_addr));
1716	desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[0].ds_addr));
1717	cmdstat = segs[0].ds_len;
1718	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
1719		cmdstat |= RL_RDESC_CMD_EOR;
1720	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
1721
1722	return (0);
1723}
1724
1725#ifdef RE_FIXUP_RX
1726static __inline void
1727re_fixup_rx(struct mbuf *m)
1728{
1729	int                     i;
1730	uint16_t                *src, *dst;
1731
1732	src = mtod(m, uint16_t *);
1733	dst = src - (RE_ETHER_ALIGN - ETHER_ALIGN) / sizeof *src;
1734
1735	for (i = 0; i < (m->m_len / sizeof(uint16_t) + 1); i++)
1736		*dst++ = *src++;
1737
1738	m->m_data -= RE_ETHER_ALIGN - ETHER_ALIGN;
1739}
1740#endif
1741
1742static int
1743re_tx_list_init(struct rl_softc *sc)
1744{
1745	struct rl_desc		*desc;
1746	int			i;
1747
1748	RL_LOCK_ASSERT(sc);
1749
1750	bzero(sc->rl_ldata.rl_tx_list,
1751	    sc->rl_ldata.rl_tx_desc_cnt * sizeof(struct rl_desc));
1752	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++)
1753		sc->rl_ldata.rl_tx_desc[i].tx_m = NULL;
1754	/* Set EOR. */
1755	desc = &sc->rl_ldata.rl_tx_list[sc->rl_ldata.rl_tx_desc_cnt - 1];
1756	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_EOR);
1757
1758	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
1759	    sc->rl_ldata.rl_tx_list_map,
1760	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1761
1762	sc->rl_ldata.rl_tx_prodidx = 0;
1763	sc->rl_ldata.rl_tx_considx = 0;
1764	sc->rl_ldata.rl_tx_free = sc->rl_ldata.rl_tx_desc_cnt;
1765
1766	return (0);
1767}
1768
1769static int
1770re_rx_list_init(struct rl_softc *sc)
1771{
1772	int			error, i;
1773
1774	bzero(sc->rl_ldata.rl_rx_list,
1775	    sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc));
1776	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1777		sc->rl_ldata.rl_rx_desc[i].rx_m = NULL;
1778		if ((error = re_newbuf(sc, i)) != 0)
1779			return (error);
1780	}
1781
1782	/* Flush the RX descriptors */
1783
1784	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1785	    sc->rl_ldata.rl_rx_list_map,
1786	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
1787
1788	sc->rl_ldata.rl_rx_prodidx = 0;
1789	sc->rl_head = sc->rl_tail = NULL;
1790
1791	return (0);
1792}
1793
1794/*
1795 * RX handler for C+ and 8169. For the gigE chips, we support
1796 * the reception of jumbo frames that have been fragmented
1797 * across multiple 2K mbuf cluster buffers.
1798 */
1799static int
1800re_rxeof(struct rl_softc *sc, int *rx_npktsp)
1801{
1802	struct mbuf		*m;
1803	struct ifnet		*ifp;
1804	int			i, total_len;
1805	struct rl_desc		*cur_rx;
1806	u_int32_t		rxstat, rxvlan;
1807	int			maxpkt = 16, rx_npkts = 0;
1808
1809	RL_LOCK_ASSERT(sc);
1810
1811	ifp = sc->rl_ifp;
1812
1813	/* Invalidate the descriptor memory */
1814
1815	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1816	    sc->rl_ldata.rl_rx_list_map,
1817	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
1818
1819	for (i = sc->rl_ldata.rl_rx_prodidx; maxpkt > 0;
1820	    i = RL_RX_DESC_NXT(sc, i)) {
1821		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1822			break;
1823		cur_rx = &sc->rl_ldata.rl_rx_list[i];
1824		rxstat = le32toh(cur_rx->rl_cmdstat);
1825		if ((rxstat & RL_RDESC_STAT_OWN) != 0)
1826			break;
1827		total_len = rxstat & sc->rl_rxlenmask;
1828		rxvlan = le32toh(cur_rx->rl_vlanctl);
1829		m = sc->rl_ldata.rl_rx_desc[i].rx_m;
1830
1831		if (!(rxstat & RL_RDESC_STAT_EOF)) {
1832			if (re_newbuf(sc, i) != 0) {
1833				/*
1834				 * If this is part of a multi-fragment packet,
1835				 * discard all the pieces.
1836				 */
1837				if (sc->rl_head != NULL) {
1838					m_freem(sc->rl_head);
1839					sc->rl_head = sc->rl_tail = NULL;
1840				}
1841				re_discard_rxbuf(sc, i);
1842				continue;
1843			}
1844			m->m_len = RE_RX_DESC_BUFLEN;
1845			if (sc->rl_head == NULL)
1846				sc->rl_head = sc->rl_tail = m;
1847			else {
1848				m->m_flags &= ~M_PKTHDR;
1849				sc->rl_tail->m_next = m;
1850				sc->rl_tail = m;
1851			}
1852			continue;
1853		}
1854
1855		/*
1856		 * NOTE: for the 8139C+, the frame length field
1857		 * is always 12 bits in size, but for the gigE chips,
1858		 * it is 13 bits (since the max RX frame length is 16K).
1859		 * Unfortunately, all 32 bits in the status word
1860		 * were already used, so to make room for the extra
1861		 * length bit, RealTek took out the 'frame alignment
1862		 * error' bit and shifted the other status bits
1863		 * over one slot. The OWN, EOR, FS and LS bits are
1864		 * still in the same places. We have already extracted
1865		 * the frame length and checked the OWN bit, so rather
1866		 * than using an alternate bit mapping, we shift the
1867		 * status bits one space to the right so we can evaluate
1868		 * them using the 8169 status as though it was in the
1869		 * same format as that of the 8139C+.
1870		 */
1871		if (sc->rl_type == RL_8169)
1872			rxstat >>= 1;
1873
1874		/*
1875		 * if total_len > 2^13-1, both _RXERRSUM and _GIANT will be
1876		 * set, but if CRC is clear, it will still be a valid frame.
1877		 */
1878		if (rxstat & RL_RDESC_STAT_RXERRSUM && !(total_len > 8191 &&
1879		    (rxstat & RL_RDESC_STAT_ERRS) == RL_RDESC_STAT_GIANT)) {
1880			ifp->if_ierrors++;
1881			/*
1882			 * If this is part of a multi-fragment packet,
1883			 * discard all the pieces.
1884			 */
1885			if (sc->rl_head != NULL) {
1886				m_freem(sc->rl_head);
1887				sc->rl_head = sc->rl_tail = NULL;
1888			}
1889			re_discard_rxbuf(sc, i);
1890			continue;
1891		}
1892
1893		/*
1894		 * If allocating a replacement mbuf fails,
1895		 * reload the current one.
1896		 */
1897
1898		if (re_newbuf(sc, i) != 0) {
1899			ifp->if_iqdrops++;
1900			if (sc->rl_head != NULL) {
1901				m_freem(sc->rl_head);
1902				sc->rl_head = sc->rl_tail = NULL;
1903			}
1904			re_discard_rxbuf(sc, i);
1905			continue;
1906		}
1907
1908		if (sc->rl_head != NULL) {
1909			m->m_len = total_len % RE_RX_DESC_BUFLEN;
1910			if (m->m_len == 0)
1911				m->m_len = RE_RX_DESC_BUFLEN;
1912			/*
1913			 * Special case: if there's 4 bytes or less
1914			 * in this buffer, the mbuf can be discarded:
1915			 * the last 4 bytes is the CRC, which we don't
1916			 * care about anyway.
1917			 */
1918			if (m->m_len <= ETHER_CRC_LEN) {
1919				sc->rl_tail->m_len -=
1920				    (ETHER_CRC_LEN - m->m_len);
1921				m_freem(m);
1922			} else {
1923				m->m_len -= ETHER_CRC_LEN;
1924				m->m_flags &= ~M_PKTHDR;
1925				sc->rl_tail->m_next = m;
1926			}
1927			m = sc->rl_head;
1928			sc->rl_head = sc->rl_tail = NULL;
1929			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
1930		} else
1931			m->m_pkthdr.len = m->m_len =
1932			    (total_len - ETHER_CRC_LEN);
1933
1934#ifdef RE_FIXUP_RX
1935		re_fixup_rx(m);
1936#endif
1937		ifp->if_ipackets++;
1938		m->m_pkthdr.rcvif = ifp;
1939
1940		/* Do RX checksumming if enabled */
1941
1942		if (ifp->if_capenable & IFCAP_RXCSUM) {
1943			if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
1944				/* Check IP header checksum */
1945				if (rxstat & RL_RDESC_STAT_PROTOID)
1946					m->m_pkthdr.csum_flags |=
1947					    CSUM_IP_CHECKED;
1948				if (!(rxstat & RL_RDESC_STAT_IPSUMBAD))
1949					m->m_pkthdr.csum_flags |=
1950					    CSUM_IP_VALID;
1951
1952				/* Check TCP/UDP checksum */
1953				if ((RL_TCPPKT(rxstat) &&
1954				    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
1955				    (RL_UDPPKT(rxstat) &&
1956				     !(rxstat & RL_RDESC_STAT_UDPSUMBAD))) {
1957					m->m_pkthdr.csum_flags |=
1958						CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
1959					m->m_pkthdr.csum_data = 0xffff;
1960				}
1961			} else {
1962				/*
1963				 * RTL8168C/RTL816CP/RTL8111C/RTL8111CP
1964				 */
1965				if ((rxstat & RL_RDESC_STAT_PROTOID) &&
1966				    (rxvlan & RL_RDESC_IPV4))
1967					m->m_pkthdr.csum_flags |=
1968					    CSUM_IP_CHECKED;
1969				if (!(rxstat & RL_RDESC_STAT_IPSUMBAD) &&
1970				    (rxvlan & RL_RDESC_IPV4))
1971					m->m_pkthdr.csum_flags |=
1972					    CSUM_IP_VALID;
1973				if (((rxstat & RL_RDESC_STAT_TCP) &&
1974				    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
1975				    ((rxstat & RL_RDESC_STAT_UDP) &&
1976				    !(rxstat & RL_RDESC_STAT_UDPSUMBAD))) {
1977					m->m_pkthdr.csum_flags |=
1978						CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
1979					m->m_pkthdr.csum_data = 0xffff;
1980				}
1981			}
1982		}
1983		maxpkt--;
1984		if (rxvlan & RL_RDESC_VLANCTL_TAG) {
1985			m->m_pkthdr.ether_vtag =
1986			    bswap16((rxvlan & RL_RDESC_VLANCTL_DATA));
1987			m->m_flags |= M_VLANTAG;
1988		}
1989		RL_UNLOCK(sc);
1990		(*ifp->if_input)(ifp, m);
1991		RL_LOCK(sc);
1992		rx_npkts++;
1993	}
1994
1995	/* Flush the RX DMA ring */
1996
1997	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1998	    sc->rl_ldata.rl_rx_list_map,
1999	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2000
2001	sc->rl_ldata.rl_rx_prodidx = i;
2002
2003	if (rx_npktsp != NULL)
2004		*rx_npktsp = rx_npkts;
2005	if (maxpkt)
2006		return(EAGAIN);
2007
2008	return(0);
2009}
2010
2011static void
2012re_txeof(struct rl_softc *sc)
2013{
2014	struct ifnet		*ifp;
2015	struct rl_txdesc	*txd;
2016	u_int32_t		txstat;
2017	int			cons;
2018
2019	cons = sc->rl_ldata.rl_tx_considx;
2020	if (cons == sc->rl_ldata.rl_tx_prodidx)
2021		return;
2022
2023	ifp = sc->rl_ifp;
2024	/* Invalidate the TX descriptor list */
2025	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2026	    sc->rl_ldata.rl_tx_list_map,
2027	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2028
2029	for (; cons != sc->rl_ldata.rl_tx_prodidx;
2030	    cons = RL_TX_DESC_NXT(sc, cons)) {
2031		txstat = le32toh(sc->rl_ldata.rl_tx_list[cons].rl_cmdstat);
2032		if (txstat & RL_TDESC_STAT_OWN)
2033			break;
2034		/*
2035		 * We only stash mbufs in the last descriptor
2036		 * in a fragment chain, which also happens to
2037		 * be the only place where the TX status bits
2038		 * are valid.
2039		 */
2040		if (txstat & RL_TDESC_CMD_EOF) {
2041			txd = &sc->rl_ldata.rl_tx_desc[cons];
2042			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
2043			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
2044			bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag,
2045			    txd->tx_dmamap);
2046			KASSERT(txd->tx_m != NULL,
2047			    ("%s: freeing NULL mbufs!", __func__));
2048			m_freem(txd->tx_m);
2049			txd->tx_m = NULL;
2050			if (txstat & (RL_TDESC_STAT_EXCESSCOL|
2051			    RL_TDESC_STAT_COLCNT))
2052				ifp->if_collisions++;
2053			if (txstat & RL_TDESC_STAT_TXERRSUM)
2054				ifp->if_oerrors++;
2055			else
2056				ifp->if_opackets++;
2057		}
2058		sc->rl_ldata.rl_tx_free++;
2059		ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
2060	}
2061	sc->rl_ldata.rl_tx_considx = cons;
2062
2063	/* No changes made to the TX ring, so no flush needed */
2064
2065	if (sc->rl_ldata.rl_tx_free != sc->rl_ldata.rl_tx_desc_cnt) {
2066#ifdef RE_TX_MODERATION
2067		/*
2068		 * If not all descriptors have been reaped yet, reload
2069		 * the timer so that we will eventually get another
2070		 * interrupt that will cause us to re-enter this routine.
2071		 * This is done in case the transmitter has gone idle.
2072		 */
2073		CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2074#endif
2075	} else
2076		sc->rl_watchdog_timer = 0;
2077}
2078
2079static void
2080re_tick(void *xsc)
2081{
2082	struct rl_softc		*sc;
2083	struct mii_data		*mii;
2084
2085	sc = xsc;
2086
2087	RL_LOCK_ASSERT(sc);
2088
2089	mii = device_get_softc(sc->rl_miibus);
2090	mii_tick(mii);
2091	if ((sc->rl_flags & RL_FLAG_LINK) == 0)
2092		re_miibus_statchg(sc->rl_dev);
2093	/*
2094	 * Reclaim transmitted frames here. Technically it is not
2095	 * necessary to do here but it ensures periodic reclamation
2096	 * regardless of Tx completion interrupt which seems to be
2097	 * lost on PCIe based controllers under certain situations.
2098	 */
2099	re_txeof(sc);
2100	re_watchdog(sc);
2101	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
2102}
2103
2104#ifdef DEVICE_POLLING
2105static int
2106re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
2107{
2108	struct rl_softc *sc = ifp->if_softc;
2109	int rx_npkts = 0;
2110
2111	RL_LOCK(sc);
2112	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
2113		rx_npkts = re_poll_locked(ifp, cmd, count);
2114	RL_UNLOCK(sc);
2115	return (rx_npkts);
2116}
2117
2118static int
2119re_poll_locked(struct ifnet *ifp, enum poll_cmd cmd, int count)
2120{
2121	struct rl_softc *sc = ifp->if_softc;
2122	int rx_npkts;
2123
2124	RL_LOCK_ASSERT(sc);
2125
2126	sc->rxcycles = count;
2127	re_rxeof(sc, &rx_npkts);
2128	re_txeof(sc);
2129
2130	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2131		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_txtask);
2132
2133	if (cmd == POLL_AND_CHECK_STATUS) { /* also check status register */
2134		u_int16_t       status;
2135
2136		status = CSR_READ_2(sc, RL_ISR);
2137		if (status == 0xffff)
2138			return (rx_npkts);
2139		if (status)
2140			CSR_WRITE_2(sc, RL_ISR, status);
2141		if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
2142		    (sc->rl_flags & RL_FLAG_PCIE))
2143			CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2144
2145		/*
2146		 * XXX check behaviour on receiver stalls.
2147		 */
2148
2149		if (status & RL_ISR_SYSTEM_ERR) {
2150			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2151			re_init_locked(sc);
2152		}
2153	}
2154	return (rx_npkts);
2155}
2156#endif /* DEVICE_POLLING */
2157
2158static int
2159re_intr(void *arg)
2160{
2161	struct rl_softc		*sc;
2162	uint16_t		status;
2163
2164	sc = arg;
2165
2166	status = CSR_READ_2(sc, RL_ISR);
2167	if (status == 0xFFFF || (status & RL_INTRS_CPLUS) == 0)
2168                return (FILTER_STRAY);
2169	CSR_WRITE_2(sc, RL_IMR, 0);
2170
2171	taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_inttask);
2172
2173	return (FILTER_HANDLED);
2174}
2175
2176static void
2177re_int_task(void *arg, int npending)
2178{
2179	struct rl_softc		*sc;
2180	struct ifnet		*ifp;
2181	u_int16_t		status;
2182	int			rval = 0;
2183
2184	sc = arg;
2185	ifp = sc->rl_ifp;
2186
2187	RL_LOCK(sc);
2188
2189	status = CSR_READ_2(sc, RL_ISR);
2190        CSR_WRITE_2(sc, RL_ISR, status);
2191
2192	if (sc->suspended ||
2193	    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2194		RL_UNLOCK(sc);
2195		return;
2196	}
2197
2198#ifdef DEVICE_POLLING
2199	if  (ifp->if_capenable & IFCAP_POLLING) {
2200		RL_UNLOCK(sc);
2201		return;
2202	}
2203#endif
2204
2205	if (status & (RL_ISR_RX_OK|RL_ISR_RX_ERR|RL_ISR_FIFO_OFLOW))
2206		rval = re_rxeof(sc, NULL);
2207
2208	/*
2209	 * Some chips will ignore a second TX request issued
2210	 * while an existing transmission is in progress. If
2211	 * the transmitter goes idle but there are still
2212	 * packets waiting to be sent, we need to restart the
2213	 * channel here to flush them out. This only seems to
2214	 * be required with the PCIe devices.
2215	 */
2216	if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
2217	    (sc->rl_flags & RL_FLAG_PCIE))
2218		CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2219	if (status & (
2220#ifdef RE_TX_MODERATION
2221	    RL_ISR_TIMEOUT_EXPIRED|
2222#else
2223	    RL_ISR_TX_OK|
2224#endif
2225	    RL_ISR_TX_ERR|RL_ISR_TX_DESC_UNAVAIL))
2226		re_txeof(sc);
2227
2228	if (status & RL_ISR_SYSTEM_ERR) {
2229		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2230		re_init_locked(sc);
2231	}
2232
2233	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2234		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_txtask);
2235
2236	RL_UNLOCK(sc);
2237
2238        if ((CSR_READ_2(sc, RL_ISR) & RL_INTRS_CPLUS) || rval) {
2239		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_inttask);
2240		return;
2241	}
2242
2243	CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2244}
2245
2246static int
2247re_encap(struct rl_softc *sc, struct mbuf **m_head)
2248{
2249	struct rl_txdesc	*txd, *txd_last;
2250	bus_dma_segment_t	segs[RL_NTXSEGS];
2251	bus_dmamap_t		map;
2252	struct mbuf		*m_new;
2253	struct rl_desc		*desc;
2254	int			nsegs, prod;
2255	int			i, error, ei, si;
2256	int			padlen;
2257	uint32_t		cmdstat, csum_flags, vlanctl;
2258
2259	RL_LOCK_ASSERT(sc);
2260	M_ASSERTPKTHDR((*m_head));
2261
2262	/*
2263	 * With some of the RealTek chips, using the checksum offload
2264	 * support in conjunction with the autopadding feature results
2265	 * in the transmission of corrupt frames. For example, if we
2266	 * need to send a really small IP fragment that's less than 60
2267	 * bytes in size, and IP header checksumming is enabled, the
2268	 * resulting ethernet frame that appears on the wire will
2269	 * have garbled payload. To work around this, if TX IP checksum
2270	 * offload is enabled, we always manually pad short frames out
2271	 * to the minimum ethernet frame size.
2272	 */
2273	if ((sc->rl_flags & RL_FLAG_AUTOPAD) == 0 &&
2274	    (*m_head)->m_pkthdr.len < RL_IP4CSUMTX_PADLEN &&
2275	    ((*m_head)->m_pkthdr.csum_flags & CSUM_IP) != 0) {
2276		padlen = RL_MIN_FRAMELEN - (*m_head)->m_pkthdr.len;
2277		if (M_WRITABLE(*m_head) == 0) {
2278			/* Get a writable copy. */
2279			m_new = m_dup(*m_head, M_DONTWAIT);
2280			m_freem(*m_head);
2281			if (m_new == NULL) {
2282				*m_head = NULL;
2283				return (ENOBUFS);
2284			}
2285			*m_head = m_new;
2286		}
2287		if ((*m_head)->m_next != NULL ||
2288		    M_TRAILINGSPACE(*m_head) < padlen) {
2289			m_new = m_defrag(*m_head, M_DONTWAIT);
2290			if (m_new == NULL) {
2291				m_freem(*m_head);
2292				*m_head = NULL;
2293				return (ENOBUFS);
2294			}
2295		} else
2296			m_new = *m_head;
2297
2298		/*
2299		 * Manually pad short frames, and zero the pad space
2300		 * to avoid leaking data.
2301		 */
2302		bzero(mtod(m_new, char *) + m_new->m_pkthdr.len, padlen);
2303		m_new->m_pkthdr.len += padlen;
2304		m_new->m_len = m_new->m_pkthdr.len;
2305		*m_head = m_new;
2306	}
2307
2308	prod = sc->rl_ldata.rl_tx_prodidx;
2309	txd = &sc->rl_ldata.rl_tx_desc[prod];
2310	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap,
2311	    *m_head, segs, &nsegs, BUS_DMA_NOWAIT);
2312	if (error == EFBIG) {
2313		m_new = m_collapse(*m_head, M_DONTWAIT, RL_NTXSEGS);
2314		if (m_new == NULL) {
2315			m_freem(*m_head);
2316			*m_head = NULL;
2317			return (ENOBUFS);
2318		}
2319		*m_head = m_new;
2320		error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_tx_mtag,
2321		    txd->tx_dmamap, *m_head, segs, &nsegs, BUS_DMA_NOWAIT);
2322		if (error != 0) {
2323			m_freem(*m_head);
2324			*m_head = NULL;
2325			return (error);
2326		}
2327	} else if (error != 0)
2328		return (error);
2329	if (nsegs == 0) {
2330		m_freem(*m_head);
2331		*m_head = NULL;
2332		return (EIO);
2333	}
2334
2335	/* Check for number of available descriptors. */
2336	if (sc->rl_ldata.rl_tx_free - nsegs <= 1) {
2337		bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap);
2338		return (ENOBUFS);
2339	}
2340
2341	bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap,
2342	    BUS_DMASYNC_PREWRITE);
2343
2344	/*
2345	 * Set up checksum offload. Note: checksum offload bits must
2346	 * appear in all descriptors of a multi-descriptor transmit
2347	 * attempt. This is according to testing done with an 8169
2348	 * chip. This is a requirement.
2349	 */
2350	vlanctl = 0;
2351	csum_flags = 0;
2352	if (((*m_head)->m_pkthdr.csum_flags & CSUM_TSO) != 0)
2353		csum_flags = RL_TDESC_CMD_LGSEND |
2354		    ((uint32_t)(*m_head)->m_pkthdr.tso_segsz <<
2355		    RL_TDESC_CMD_MSSVAL_SHIFT);
2356	else {
2357		/*
2358		 * Unconditionally enable IP checksum if TCP or UDP
2359		 * checksum is required. Otherwise, TCP/UDP checksum
2360		 * does't make effects.
2361		 */
2362		if (((*m_head)->m_pkthdr.csum_flags & RE_CSUM_FEATURES) != 0) {
2363			if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
2364				csum_flags |= RL_TDESC_CMD_IPCSUM;
2365				if (((*m_head)->m_pkthdr.csum_flags &
2366				    CSUM_TCP) != 0)
2367					csum_flags |= RL_TDESC_CMD_TCPCSUM;
2368				if (((*m_head)->m_pkthdr.csum_flags &
2369				    CSUM_UDP) != 0)
2370					csum_flags |= RL_TDESC_CMD_UDPCSUM;
2371			} else {
2372				vlanctl |= RL_TDESC_CMD_IPCSUMV2;
2373				if (((*m_head)->m_pkthdr.csum_flags &
2374				    CSUM_TCP) != 0)
2375					vlanctl |= RL_TDESC_CMD_TCPCSUMV2;
2376				if (((*m_head)->m_pkthdr.csum_flags &
2377				    CSUM_UDP) != 0)
2378					vlanctl |= RL_TDESC_CMD_UDPCSUMV2;
2379			}
2380		}
2381	}
2382
2383	/*
2384	 * Set up hardware VLAN tagging. Note: vlan tag info must
2385	 * appear in all descriptors of a multi-descriptor
2386	 * transmission attempt.
2387	 */
2388	if ((*m_head)->m_flags & M_VLANTAG)
2389		vlanctl |= bswap16((*m_head)->m_pkthdr.ether_vtag) |
2390		    RL_TDESC_VLANCTL_TAG;
2391
2392	si = prod;
2393	for (i = 0; i < nsegs; i++, prod = RL_TX_DESC_NXT(sc, prod)) {
2394		desc = &sc->rl_ldata.rl_tx_list[prod];
2395		desc->rl_vlanctl = htole32(vlanctl);
2396		desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[i].ds_addr));
2397		desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[i].ds_addr));
2398		cmdstat = segs[i].ds_len;
2399		if (i != 0)
2400			cmdstat |= RL_TDESC_CMD_OWN;
2401		if (prod == sc->rl_ldata.rl_tx_desc_cnt - 1)
2402			cmdstat |= RL_TDESC_CMD_EOR;
2403		desc->rl_cmdstat = htole32(cmdstat | csum_flags);
2404		sc->rl_ldata.rl_tx_free--;
2405	}
2406	/* Update producer index. */
2407	sc->rl_ldata.rl_tx_prodidx = prod;
2408
2409	/* Set EOF on the last descriptor. */
2410	ei = RL_TX_DESC_PRV(sc, prod);
2411	desc = &sc->rl_ldata.rl_tx_list[ei];
2412	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_EOF);
2413
2414	desc = &sc->rl_ldata.rl_tx_list[si];
2415	/* Set SOF and transfer ownership of packet to the chip. */
2416	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_OWN | RL_TDESC_CMD_SOF);
2417
2418	/*
2419	 * Insure that the map for this transmission
2420	 * is placed at the array index of the last descriptor
2421	 * in this chain.  (Swap last and first dmamaps.)
2422	 */
2423	txd_last = &sc->rl_ldata.rl_tx_desc[ei];
2424	map = txd->tx_dmamap;
2425	txd->tx_dmamap = txd_last->tx_dmamap;
2426	txd_last->tx_dmamap = map;
2427	txd_last->tx_m = *m_head;
2428
2429	return (0);
2430}
2431
2432static void
2433re_tx_task(void *arg, int npending)
2434{
2435	struct ifnet		*ifp;
2436
2437	ifp = arg;
2438	re_start(ifp);
2439}
2440
2441/*
2442 * Main transmit routine for C+ and gigE NICs.
2443 */
2444static void
2445re_start(struct ifnet *ifp)
2446{
2447	struct rl_softc		*sc;
2448	struct mbuf		*m_head;
2449	int			queued;
2450
2451	sc = ifp->if_softc;
2452
2453	RL_LOCK(sc);
2454
2455	if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
2456	    IFF_DRV_RUNNING || (sc->rl_flags & RL_FLAG_LINK) == 0) {
2457		RL_UNLOCK(sc);
2458		return;
2459	}
2460
2461	for (queued = 0; !IFQ_DRV_IS_EMPTY(&ifp->if_snd) &&
2462	    sc->rl_ldata.rl_tx_free > 1;) {
2463		IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
2464		if (m_head == NULL)
2465			break;
2466
2467		if (re_encap(sc, &m_head) != 0) {
2468			if (m_head == NULL)
2469				break;
2470			IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
2471			ifp->if_drv_flags |= IFF_DRV_OACTIVE;
2472			break;
2473		}
2474
2475		/*
2476		 * If there's a BPF listener, bounce a copy of this frame
2477		 * to him.
2478		 */
2479		ETHER_BPF_MTAP(ifp, m_head);
2480
2481		queued++;
2482	}
2483
2484	if (queued == 0) {
2485#ifdef RE_TX_MODERATION
2486		if (sc->rl_ldata.rl_tx_free != sc->rl_ldata.rl_tx_desc_cnt)
2487			CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2488#endif
2489		RL_UNLOCK(sc);
2490		return;
2491	}
2492
2493	/* Flush the TX descriptors */
2494
2495	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2496	    sc->rl_ldata.rl_tx_list_map,
2497	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2498
2499	CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2500
2501#ifdef RE_TX_MODERATION
2502	/*
2503	 * Use the countdown timer for interrupt moderation.
2504	 * 'TX done' interrupts are disabled. Instead, we reset the
2505	 * countdown timer, which will begin counting until it hits
2506	 * the value in the TIMERINT register, and then trigger an
2507	 * interrupt. Each time we write to the TIMERCNT register,
2508	 * the timer count is reset to 0.
2509	 */
2510	CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2511#endif
2512
2513	/*
2514	 * Set a timeout in case the chip goes out to lunch.
2515	 */
2516	sc->rl_watchdog_timer = 5;
2517
2518	RL_UNLOCK(sc);
2519}
2520
2521static void
2522re_init(void *xsc)
2523{
2524	struct rl_softc		*sc = xsc;
2525
2526	RL_LOCK(sc);
2527	re_init_locked(sc);
2528	RL_UNLOCK(sc);
2529}
2530
2531static void
2532re_init_locked(struct rl_softc *sc)
2533{
2534	struct ifnet		*ifp = sc->rl_ifp;
2535	struct mii_data		*mii;
2536	uint32_t		reg;
2537	uint16_t		cfg;
2538	union {
2539		uint32_t align_dummy;
2540		u_char eaddr[ETHER_ADDR_LEN];
2541        } eaddr;
2542
2543	RL_LOCK_ASSERT(sc);
2544
2545	mii = device_get_softc(sc->rl_miibus);
2546
2547	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
2548		return;
2549
2550	/*
2551	 * Cancel pending I/O and free all RX/TX buffers.
2552	 */
2553	re_stop(sc);
2554
2555	/* Put controller into known state. */
2556	re_reset(sc);
2557
2558	/*
2559	 * Enable C+ RX and TX mode, as well as VLAN stripping and
2560	 * RX checksum offload. We must configure the C+ register
2561	 * before all others.
2562	 */
2563	cfg = RL_CPLUSCMD_PCI_MRW;
2564	if ((ifp->if_capenable & IFCAP_RXCSUM) != 0)
2565		cfg |= RL_CPLUSCMD_RXCSUM_ENB;
2566	if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0)
2567		cfg |= RL_CPLUSCMD_VLANSTRIP;
2568	if ((sc->rl_flags & RL_FLAG_MACSTAT) != 0) {
2569		cfg |= RL_CPLUSCMD_MACSTAT_DIS;
2570		/* XXX magic. */
2571		cfg |= 0x0001;
2572	} else
2573		cfg |= RL_CPLUSCMD_RXENB | RL_CPLUSCMD_TXENB;
2574	CSR_WRITE_2(sc, RL_CPLUS_CMD, cfg);
2575	if (sc->rl_hwrev == RL_HWREV_8169_8110SC ||
2576	    sc->rl_hwrev == RL_HWREV_8169_8110SCE) {
2577		reg = 0x000fff00;
2578		if ((CSR_READ_1(sc, RL_CFG2) & RL_CFG2_PCI66MHZ) != 0)
2579			reg |= 0x000000ff;
2580		if (sc->rl_hwrev == RL_HWREV_8169_8110SCE)
2581			reg |= 0x00f00000;
2582		CSR_WRITE_4(sc, 0x7c, reg);
2583		/* Disable interrupt mitigation. */
2584		CSR_WRITE_2(sc, 0xe2, 0);
2585	}
2586	/*
2587	 * Disable TSO if interface MTU size is greater than MSS
2588	 * allowed in controller.
2589	 */
2590	if (ifp->if_mtu > RL_TSO_MTU && (ifp->if_capenable & IFCAP_TSO4) != 0) {
2591		ifp->if_capenable &= ~IFCAP_TSO4;
2592		ifp->if_hwassist &= ~CSUM_TSO;
2593	}
2594
2595	/*
2596	 * Init our MAC address.  Even though the chipset
2597	 * documentation doesn't mention it, we need to enter "Config
2598	 * register write enable" mode to modify the ID registers.
2599	 */
2600	/* Copy MAC address on stack to align. */
2601	bcopy(IF_LLADDR(ifp), eaddr.eaddr, ETHER_ADDR_LEN);
2602	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG);
2603	CSR_WRITE_4(sc, RL_IDR0,
2604	    htole32(*(u_int32_t *)(&eaddr.eaddr[0])));
2605	CSR_WRITE_4(sc, RL_IDR4,
2606	    htole32(*(u_int32_t *)(&eaddr.eaddr[4])));
2607	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
2608
2609	/*
2610	 * For C+ mode, initialize the RX descriptors and mbufs.
2611	 */
2612	re_rx_list_init(sc);
2613	re_tx_list_init(sc);
2614
2615	/*
2616	 * Load the addresses of the RX and TX lists into the chip.
2617	 */
2618
2619	CSR_WRITE_4(sc, RL_RXLIST_ADDR_HI,
2620	    RL_ADDR_HI(sc->rl_ldata.rl_rx_list_addr));
2621	CSR_WRITE_4(sc, RL_RXLIST_ADDR_LO,
2622	    RL_ADDR_LO(sc->rl_ldata.rl_rx_list_addr));
2623
2624	CSR_WRITE_4(sc, RL_TXLIST_ADDR_HI,
2625	    RL_ADDR_HI(sc->rl_ldata.rl_tx_list_addr));
2626	CSR_WRITE_4(sc, RL_TXLIST_ADDR_LO,
2627	    RL_ADDR_LO(sc->rl_ldata.rl_tx_list_addr));
2628
2629	/*
2630	 * Enable transmit and receive.
2631	 */
2632	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
2633
2634	/*
2635	 * Set the initial TX configuration.
2636	 */
2637	if (sc->rl_testmode) {
2638		if (sc->rl_type == RL_8169)
2639			CSR_WRITE_4(sc, RL_TXCFG,
2640			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON);
2641		else
2642			CSR_WRITE_4(sc, RL_TXCFG,
2643			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON_CPLUS);
2644	} else
2645		CSR_WRITE_4(sc, RL_TXCFG, RL_TXCFG_CONFIG);
2646
2647	CSR_WRITE_1(sc, RL_EARLY_TX_THRESH, 16);
2648
2649	/*
2650	 * Set the initial RX configuration.
2651	 */
2652	re_set_rxmode(sc);
2653
2654#ifdef DEVICE_POLLING
2655	/*
2656	 * Disable interrupts if we are polling.
2657	 */
2658	if (ifp->if_capenable & IFCAP_POLLING)
2659		CSR_WRITE_2(sc, RL_IMR, 0);
2660	else	/* otherwise ... */
2661#endif
2662
2663	/*
2664	 * Enable interrupts.
2665	 */
2666	if (sc->rl_testmode)
2667		CSR_WRITE_2(sc, RL_IMR, 0);
2668	else
2669		CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2670	CSR_WRITE_2(sc, RL_ISR, RL_INTRS_CPLUS);
2671
2672	/* Set initial TX threshold */
2673	sc->rl_txthresh = RL_TX_THRESH_INIT;
2674
2675	/* Start RX/TX process. */
2676	CSR_WRITE_4(sc, RL_MISSEDPKT, 0);
2677#ifdef notdef
2678	/* Enable receiver and transmitter. */
2679	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
2680#endif
2681
2682#ifdef RE_TX_MODERATION
2683	/*
2684	 * Initialize the timer interrupt register so that
2685	 * a timer interrupt will be generated once the timer
2686	 * reaches a certain number of ticks. The timer is
2687	 * reloaded on each transmit. This gives us TX interrupt
2688	 * moderation, which dramatically improves TX frame rate.
2689	 */
2690	if (sc->rl_type == RL_8169)
2691		CSR_WRITE_4(sc, RL_TIMERINT_8169, 0x800);
2692	else
2693		CSR_WRITE_4(sc, RL_TIMERINT, 0x400);
2694#endif
2695
2696	/*
2697	 * For 8169 gigE NICs, set the max allowed RX packet
2698	 * size so we can receive jumbo frames.
2699	 */
2700	if (sc->rl_type == RL_8169)
2701		CSR_WRITE_2(sc, RL_MAXRXPKTLEN, 16383);
2702
2703	if (sc->rl_testmode)
2704		return;
2705
2706	mii_mediachg(mii);
2707
2708	CSR_WRITE_1(sc, RL_CFG1, CSR_READ_1(sc, RL_CFG1) | RL_CFG1_DRVLOAD);
2709
2710	ifp->if_drv_flags |= IFF_DRV_RUNNING;
2711	ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
2712
2713	sc->rl_flags &= ~RL_FLAG_LINK;
2714	sc->rl_watchdog_timer = 0;
2715	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
2716}
2717
2718/*
2719 * Set media options.
2720 */
2721static int
2722re_ifmedia_upd(struct ifnet *ifp)
2723{
2724	struct rl_softc		*sc;
2725	struct mii_data		*mii;
2726	int			error;
2727
2728	sc = ifp->if_softc;
2729	mii = device_get_softc(sc->rl_miibus);
2730	RL_LOCK(sc);
2731	error = mii_mediachg(mii);
2732	RL_UNLOCK(sc);
2733
2734	return (error);
2735}
2736
2737/*
2738 * Report current media status.
2739 */
2740static void
2741re_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
2742{
2743	struct rl_softc		*sc;
2744	struct mii_data		*mii;
2745
2746	sc = ifp->if_softc;
2747	mii = device_get_softc(sc->rl_miibus);
2748
2749	RL_LOCK(sc);
2750	mii_pollstat(mii);
2751	RL_UNLOCK(sc);
2752	ifmr->ifm_active = mii->mii_media_active;
2753	ifmr->ifm_status = mii->mii_media_status;
2754}
2755
2756static int
2757re_ioctl(struct ifnet *ifp, u_long command, caddr_t data)
2758{
2759	struct rl_softc		*sc = ifp->if_softc;
2760	struct ifreq		*ifr = (struct ifreq *) data;
2761	struct mii_data		*mii;
2762	int			error = 0;
2763
2764	switch (command) {
2765	case SIOCSIFMTU:
2766		if (ifr->ifr_mtu < ETHERMIN || ifr->ifr_mtu > RL_JUMBO_MTU) {
2767			error = EINVAL;
2768			break;
2769		}
2770		if ((sc->rl_flags & RL_FLAG_NOJUMBO) != 0 &&
2771		    ifr->ifr_mtu > RL_MAX_FRAMELEN) {
2772			error = EINVAL;
2773			break;
2774		}
2775		RL_LOCK(sc);
2776		if (ifp->if_mtu != ifr->ifr_mtu)
2777			ifp->if_mtu = ifr->ifr_mtu;
2778		if (ifp->if_mtu > RL_TSO_MTU &&
2779		    (ifp->if_capenable & IFCAP_TSO4) != 0) {
2780			ifp->if_capenable &= ~IFCAP_TSO4;
2781			ifp->if_hwassist &= ~CSUM_TSO;
2782		}
2783		RL_UNLOCK(sc);
2784		break;
2785	case SIOCSIFFLAGS:
2786		RL_LOCK(sc);
2787		if ((ifp->if_flags & IFF_UP) != 0) {
2788			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
2789				if (((ifp->if_flags ^ sc->rl_if_flags)
2790				    & (IFF_PROMISC | IFF_ALLMULTI)) != 0)
2791					re_set_rxmode(sc);
2792			} else
2793				re_init_locked(sc);
2794		} else {
2795			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
2796				re_stop(sc);
2797		}
2798		sc->rl_if_flags = ifp->if_flags;
2799		RL_UNLOCK(sc);
2800		break;
2801	case SIOCADDMULTI:
2802	case SIOCDELMULTI:
2803		RL_LOCK(sc);
2804		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
2805			re_set_rxmode(sc);
2806		RL_UNLOCK(sc);
2807		break;
2808	case SIOCGIFMEDIA:
2809	case SIOCSIFMEDIA:
2810		mii = device_get_softc(sc->rl_miibus);
2811		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, command);
2812		break;
2813	case SIOCSIFCAP:
2814	    {
2815		int mask, reinit;
2816
2817		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
2818		reinit = 0;
2819#ifdef DEVICE_POLLING
2820		if (mask & IFCAP_POLLING) {
2821			if (ifr->ifr_reqcap & IFCAP_POLLING) {
2822				error = ether_poll_register(re_poll, ifp);
2823				if (error)
2824					return(error);
2825				RL_LOCK(sc);
2826				/* Disable interrupts */
2827				CSR_WRITE_2(sc, RL_IMR, 0x0000);
2828				ifp->if_capenable |= IFCAP_POLLING;
2829				RL_UNLOCK(sc);
2830			} else {
2831				error = ether_poll_deregister(ifp);
2832				/* Enable interrupts. */
2833				RL_LOCK(sc);
2834				CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2835				ifp->if_capenable &= ~IFCAP_POLLING;
2836				RL_UNLOCK(sc);
2837			}
2838		}
2839#endif /* DEVICE_POLLING */
2840		if (mask & IFCAP_HWCSUM) {
2841			ifp->if_capenable ^= IFCAP_HWCSUM;
2842			if (ifp->if_capenable & IFCAP_TXCSUM)
2843				ifp->if_hwassist |= RE_CSUM_FEATURES;
2844			else
2845				ifp->if_hwassist &= ~RE_CSUM_FEATURES;
2846			reinit = 1;
2847		}
2848		if (mask & IFCAP_VLAN_HWTAGGING) {
2849			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
2850			reinit = 1;
2851		}
2852		if (mask & IFCAP_TSO4) {
2853			ifp->if_capenable ^= IFCAP_TSO4;
2854			if ((IFCAP_TSO4 & ifp->if_capenable) &&
2855			    (IFCAP_TSO4 & ifp->if_capabilities))
2856				ifp->if_hwassist |= CSUM_TSO;
2857			else
2858				ifp->if_hwassist &= ~CSUM_TSO;
2859			if (ifp->if_mtu > RL_TSO_MTU &&
2860			    (ifp->if_capenable & IFCAP_TSO4) != 0) {
2861				ifp->if_capenable &= ~IFCAP_TSO4;
2862				ifp->if_hwassist &= ~CSUM_TSO;
2863			}
2864		}
2865		if ((mask & IFCAP_WOL) != 0 &&
2866		    (ifp->if_capabilities & IFCAP_WOL) != 0) {
2867			if ((mask & IFCAP_WOL_UCAST) != 0)
2868				ifp->if_capenable ^= IFCAP_WOL_UCAST;
2869			if ((mask & IFCAP_WOL_MCAST) != 0)
2870				ifp->if_capenable ^= IFCAP_WOL_MCAST;
2871			if ((mask & IFCAP_WOL_MAGIC) != 0)
2872				ifp->if_capenable ^= IFCAP_WOL_MAGIC;
2873		}
2874		if (reinit && ifp->if_drv_flags & IFF_DRV_RUNNING) {
2875			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2876			re_init(sc);
2877		}
2878		VLAN_CAPABILITIES(ifp);
2879	    }
2880		break;
2881	default:
2882		error = ether_ioctl(ifp, command, data);
2883		break;
2884	}
2885
2886	return (error);
2887}
2888
2889static void
2890re_watchdog(struct rl_softc *sc)
2891{
2892	struct ifnet		*ifp;
2893
2894	RL_LOCK_ASSERT(sc);
2895
2896	if (sc->rl_watchdog_timer == 0 || --sc->rl_watchdog_timer != 0)
2897		return;
2898
2899	ifp = sc->rl_ifp;
2900	re_txeof(sc);
2901	if (sc->rl_ldata.rl_tx_free == sc->rl_ldata.rl_tx_desc_cnt) {
2902		if_printf(ifp, "watchdog timeout (missed Tx interrupts) "
2903		    "-- recovering\n");
2904		if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2905			taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_txtask);
2906		return;
2907	}
2908
2909	if_printf(ifp, "watchdog timeout\n");
2910	ifp->if_oerrors++;
2911
2912	re_rxeof(sc, NULL);
2913	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2914	re_init_locked(sc);
2915	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2916		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_txtask);
2917}
2918
2919/*
2920 * Stop the adapter and free any mbufs allocated to the
2921 * RX and TX lists.
2922 */
2923static void
2924re_stop(struct rl_softc *sc)
2925{
2926	int			i;
2927	struct ifnet		*ifp;
2928	struct rl_txdesc	*txd;
2929	struct rl_rxdesc	*rxd;
2930
2931	RL_LOCK_ASSERT(sc);
2932
2933	ifp = sc->rl_ifp;
2934
2935	sc->rl_watchdog_timer = 0;
2936	callout_stop(&sc->rl_stat_callout);
2937	ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
2938
2939	if ((sc->rl_flags & RL_FLAG_CMDSTOP) != 0)
2940		CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_STOPREQ | RL_CMD_TX_ENB |
2941		    RL_CMD_RX_ENB);
2942	else
2943		CSR_WRITE_1(sc, RL_COMMAND, 0x00);
2944	DELAY(1000);
2945	CSR_WRITE_2(sc, RL_IMR, 0x0000);
2946	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
2947
2948	if (sc->rl_head != NULL) {
2949		m_freem(sc->rl_head);
2950		sc->rl_head = sc->rl_tail = NULL;
2951	}
2952
2953	/* Free the TX list buffers. */
2954
2955	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
2956		txd = &sc->rl_ldata.rl_tx_desc[i];
2957		if (txd->tx_m != NULL) {
2958			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
2959			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
2960			bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag,
2961			    txd->tx_dmamap);
2962			m_freem(txd->tx_m);
2963			txd->tx_m = NULL;
2964		}
2965	}
2966
2967	/* Free the RX list buffers. */
2968
2969	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
2970		rxd = &sc->rl_ldata.rl_rx_desc[i];
2971		if (rxd->rx_m != NULL) {
2972			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
2973			    rxd->rx_dmamap, BUS_DMASYNC_POSTREAD);
2974			bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag,
2975			    rxd->rx_dmamap);
2976			m_freem(rxd->rx_m);
2977			rxd->rx_m = NULL;
2978		}
2979	}
2980}
2981
2982/*
2983 * Device suspend routine.  Stop the interface and save some PCI
2984 * settings in case the BIOS doesn't restore them properly on
2985 * resume.
2986 */
2987static int
2988re_suspend(device_t dev)
2989{
2990	struct rl_softc		*sc;
2991
2992	sc = device_get_softc(dev);
2993
2994	RL_LOCK(sc);
2995	re_stop(sc);
2996	re_setwol(sc);
2997	sc->suspended = 1;
2998	RL_UNLOCK(sc);
2999
3000	return (0);
3001}
3002
3003/*
3004 * Device resume routine.  Restore some PCI settings in case the BIOS
3005 * doesn't, re-enable busmastering, and restart the interface if
3006 * appropriate.
3007 */
3008static int
3009re_resume(device_t dev)
3010{
3011	struct rl_softc		*sc;
3012	struct ifnet		*ifp;
3013
3014	sc = device_get_softc(dev);
3015
3016	RL_LOCK(sc);
3017
3018	ifp = sc->rl_ifp;
3019	/* Take controller out of sleep mode. */
3020	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
3021		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
3022			CSR_WRITE_1(sc, RL_GPIO,
3023			    CSR_READ_1(sc, RL_GPIO) | 0x01);
3024	}
3025
3026	/*
3027	 * Clear WOL matching such that normal Rx filtering
3028	 * wouldn't interfere with WOL patterns.
3029	 */
3030	re_clrwol(sc);
3031
3032	/* reinitialize interface if necessary */
3033	if (ifp->if_flags & IFF_UP)
3034		re_init_locked(sc);
3035
3036	sc->suspended = 0;
3037	RL_UNLOCK(sc);
3038
3039	return (0);
3040}
3041
3042/*
3043 * Stop all chip I/O so that the kernel's probe routines don't
3044 * get confused by errant DMAs when rebooting.
3045 */
3046static int
3047re_shutdown(device_t dev)
3048{
3049	struct rl_softc		*sc;
3050
3051	sc = device_get_softc(dev);
3052
3053	RL_LOCK(sc);
3054	re_stop(sc);
3055	/*
3056	 * Mark interface as down since otherwise we will panic if
3057	 * interrupt comes in later on, which can happen in some
3058	 * cases.
3059	 */
3060	sc->rl_ifp->if_flags &= ~IFF_UP;
3061	re_setwol(sc);
3062	RL_UNLOCK(sc);
3063
3064	return (0);
3065}
3066
3067static void
3068re_setwol(struct rl_softc *sc)
3069{
3070	struct ifnet		*ifp;
3071	int			pmc;
3072	uint16_t		pmstat;
3073	uint8_t			v;
3074
3075	RL_LOCK_ASSERT(sc);
3076
3077	if (pci_find_extcap(sc->rl_dev, PCIY_PMG, &pmc) != 0)
3078		return;
3079
3080	ifp = sc->rl_ifp;
3081	/* Put controller into sleep mode. */
3082	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
3083		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
3084			CSR_WRITE_1(sc, RL_GPIO,
3085			    CSR_READ_1(sc, RL_GPIO) & ~0x01);
3086	}
3087	if ((ifp->if_capenable & IFCAP_WOL) != 0 &&
3088	    (sc->rl_flags & RL_FLAG_WOLRXENB) != 0)
3089		CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RX_ENB);
3090	/* Enable config register write. */
3091	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
3092
3093	/* Enable PME. */
3094	v = CSR_READ_1(sc, RL_CFG1);
3095	v &= ~RL_CFG1_PME;
3096	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3097		v |= RL_CFG1_PME;
3098	CSR_WRITE_1(sc, RL_CFG1, v);
3099
3100	v = CSR_READ_1(sc, RL_CFG3);
3101	v &= ~(RL_CFG3_WOL_LINK | RL_CFG3_WOL_MAGIC);
3102	if ((ifp->if_capenable & IFCAP_WOL_MAGIC) != 0)
3103		v |= RL_CFG3_WOL_MAGIC;
3104	CSR_WRITE_1(sc, RL_CFG3, v);
3105
3106	/* Config register write done. */
3107	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3108
3109	v = CSR_READ_1(sc, RL_CFG5);
3110	v &= ~(RL_CFG5_WOL_BCAST | RL_CFG5_WOL_MCAST | RL_CFG5_WOL_UCAST);
3111	v &= ~RL_CFG5_WOL_LANWAKE;
3112	if ((ifp->if_capenable & IFCAP_WOL_UCAST) != 0)
3113		v |= RL_CFG5_WOL_UCAST;
3114	if ((ifp->if_capenable & IFCAP_WOL_MCAST) != 0)
3115		v |= RL_CFG5_WOL_MCAST | RL_CFG5_WOL_BCAST;
3116	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3117		v |= RL_CFG5_WOL_LANWAKE;
3118	CSR_WRITE_1(sc, RL_CFG5, v);
3119
3120	/*
3121	 * It seems that hardware resets its link speed to 100Mbps in
3122	 * power down mode so switching to 100Mbps in driver is not
3123	 * needed.
3124	 */
3125
3126	/* Request PME if WOL is requested. */
3127	pmstat = pci_read_config(sc->rl_dev, pmc + PCIR_POWER_STATUS, 2);
3128	pmstat &= ~(PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE);
3129	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3130		pmstat |= PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE;
3131	pci_write_config(sc->rl_dev, pmc + PCIR_POWER_STATUS, pmstat, 2);
3132}
3133
3134static void
3135re_clrwol(struct rl_softc *sc)
3136{
3137	int			pmc;
3138	uint8_t			v;
3139
3140	RL_LOCK_ASSERT(sc);
3141
3142	if (pci_find_extcap(sc->rl_dev, PCIY_PMG, &pmc) != 0)
3143		return;
3144
3145	/* Enable config register write. */
3146	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
3147
3148	v = CSR_READ_1(sc, RL_CFG3);
3149	v &= ~(RL_CFG3_WOL_LINK | RL_CFG3_WOL_MAGIC);
3150	CSR_WRITE_1(sc, RL_CFG3, v);
3151
3152	/* Config register write done. */
3153	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3154
3155	v = CSR_READ_1(sc, RL_CFG5);
3156	v &= ~(RL_CFG5_WOL_BCAST | RL_CFG5_WOL_MCAST | RL_CFG5_WOL_UCAST);
3157	v &= ~RL_CFG5_WOL_LANWAKE;
3158	CSR_WRITE_1(sc, RL_CFG5, v);
3159}
3160