if_re.c revision 185752
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 185752 2008-12-08 01:44:18Z yongari $");
35
36/*
37 * RealTek 8139C+/8169/8169S/8110S/8168/8111/8101E PCI NIC driver
38 *
39 * Written by Bill Paul <wpaul@windriver.com>
40 * Senior Networking Software Engineer
41 * Wind River Systems
42 */
43
44/*
45 * This driver is designed to support RealTek's next generation of
46 * 10/100 and 10/100/1000 PCI ethernet controllers. There are currently
47 * seven devices in this family: the RTL8139C+, the RTL8169, the RTL8169S,
48 * RTL8110S, the RTL8168, the RTL8111 and the RTL8101E.
49 *
50 * The 8139C+ is a 10/100 ethernet chip. It is backwards compatible
51 * with the older 8139 family, however it also supports a special
52 * C+ mode of operation that provides several new performance enhancing
53 * features. These include:
54 *
55 *	o Descriptor based DMA mechanism. Each descriptor represents
56 *	  a single packet fragment. Data buffers may be aligned on
57 *	  any byte boundary.
58 *
59 *	o 64-bit DMA
60 *
61 *	o TCP/IP checksum offload for both RX and TX
62 *
63 *	o High and normal priority transmit DMA rings
64 *
65 *	o VLAN tag insertion and extraction
66 *
67 *	o TCP large send (segmentation offload)
68 *
69 * Like the 8139, the 8139C+ also has a built-in 10/100 PHY. The C+
70 * programming API is fairly straightforward. The RX filtering, EEPROM
71 * access and PHY access is the same as it is on the older 8139 series
72 * chips.
73 *
74 * The 8169 is a 64-bit 10/100/1000 gigabit ethernet MAC. It has almost the
75 * same programming API and feature set as the 8139C+ with the following
76 * differences and additions:
77 *
78 *	o 1000Mbps mode
79 *
80 *	o Jumbo frames
81 *
82 *	o GMII and TBI ports/registers for interfacing with copper
83 *	  or fiber PHYs
84 *
85 *	o RX and TX DMA rings can have up to 1024 descriptors
86 *	  (the 8139C+ allows a maximum of 64)
87 *
88 *	o Slight differences in register layout from the 8139C+
89 *
90 * The TX start and timer interrupt registers are at different locations
91 * on the 8169 than they are on the 8139C+. Also, the status word in the
92 * RX descriptor has a slightly different bit layout. The 8169 does not
93 * have a built-in PHY. Most reference boards use a Marvell 88E1000 'Alaska'
94 * copper gigE PHY.
95 *
96 * The 8169S/8110S 10/100/1000 devices have built-in copper gigE PHYs
97 * (the 'S' stands for 'single-chip'). These devices have the same
98 * programming API as the older 8169, but also have some vendor-specific
99 * registers for the on-board PHY. The 8110S is a LAN-on-motherboard
100 * part designed to be pin-compatible with the RealTek 8100 10/100 chip.
101 *
102 * This driver takes advantage of the RX and TX checksum offload and
103 * VLAN tag insertion/extraction features. It also implements TX
104 * interrupt moderation using the timer interrupt registers, which
105 * significantly reduces TX interrupt load. There is also support
106 * for jumbo frames, however the 8169/8169S/8110S can not transmit
107 * jumbo frames larger than 7440, so the max MTU possible with this
108 * driver is 7422 bytes.
109 */
110
111#ifdef HAVE_KERNEL_OPTION_HEADERS
112#include "opt_device_polling.h"
113#endif
114
115#include <sys/param.h>
116#include <sys/endian.h>
117#include <sys/systm.h>
118#include <sys/sockio.h>
119#include <sys/mbuf.h>
120#include <sys/malloc.h>
121#include <sys/module.h>
122#include <sys/kernel.h>
123#include <sys/socket.h>
124#include <sys/lock.h>
125#include <sys/mutex.h>
126#include <sys/taskqueue.h>
127
128#include <net/if.h>
129#include <net/if_arp.h>
130#include <net/ethernet.h>
131#include <net/if_dl.h>
132#include <net/if_media.h>
133#include <net/if_types.h>
134#include <net/if_vlan_var.h>
135
136#include <net/bpf.h>
137
138#include <machine/bus.h>
139#include <machine/resource.h>
140#include <sys/bus.h>
141#include <sys/rman.h>
142
143#include <dev/mii/mii.h>
144#include <dev/mii/miivar.h>
145
146#include <dev/pci/pcireg.h>
147#include <dev/pci/pcivar.h>
148
149#include <pci/if_rlreg.h>
150
151MODULE_DEPEND(re, pci, 1, 1, 1);
152MODULE_DEPEND(re, ether, 1, 1, 1);
153MODULE_DEPEND(re, miibus, 1, 1, 1);
154
155/* "device miibus" required.  See GENERIC if you get errors here. */
156#include "miibus_if.h"
157
158/* Tunables. */
159static int msi_disable = 1;
160TUNABLE_INT("hw.re.msi_disable", &msi_disable);
161
162#define RE_CSUM_FEATURES    (CSUM_IP | CSUM_TCP | CSUM_UDP)
163
164/*
165 * Various supported device vendors/types and their names.
166 */
167static struct rl_type re_devs[] = {
168	{ DLINK_VENDORID, DLINK_DEVICEID_528T, 0,
169	    "D-Link DGE-528(T) Gigabit Ethernet Adapter" },
170	{ RT_VENDORID, RT_DEVICEID_8139, 0,
171	    "RealTek 8139C+ 10/100BaseTX" },
172	{ RT_VENDORID, RT_DEVICEID_8101E, 0,
173	    "RealTek 8101E/8102E/8102EL PCIe 10/100baseTX" },
174	{ RT_VENDORID, RT_DEVICEID_8168, 0,
175	    "RealTek 8168/8168B/8168C/8168CP/8168D/8111B/8111C/8111CP PCIe "
176	    "Gigabit Ethernet" },
177	{ RT_VENDORID, RT_DEVICEID_8169, 0,
178	    "RealTek 8169/8169S/8169SB(L)/8110S/8110SB(L) Gigabit Ethernet" },
179	{ RT_VENDORID, RT_DEVICEID_8169SC, 0,
180	    "RealTek 8169SC/8110SC Single-chip Gigabit Ethernet" },
181	{ COREGA_VENDORID, COREGA_DEVICEID_CGLAPCIGT, 0,
182	    "Corega CG-LAPCIGT (RTL8169S) Gigabit Ethernet" },
183	{ LINKSYS_VENDORID, LINKSYS_DEVICEID_EG1032, 0,
184	    "Linksys EG1032 (RTL8169S) Gigabit Ethernet" },
185	{ USR_VENDORID, USR_DEVICEID_997902, 0,
186	    "US Robotics 997902 (RTL8169S) Gigabit Ethernet" }
187};
188
189static struct rl_hwrev re_hwrevs[] = {
190	{ RL_HWREV_8139, RL_8139,  "" },
191	{ RL_HWREV_8139A, RL_8139, "A" },
192	{ RL_HWREV_8139AG, RL_8139, "A-G" },
193	{ RL_HWREV_8139B, RL_8139, "B" },
194	{ RL_HWREV_8130, RL_8139, "8130" },
195	{ RL_HWREV_8139C, RL_8139, "C" },
196	{ RL_HWREV_8139D, RL_8139, "8139D/8100B/8100C" },
197	{ RL_HWREV_8139CPLUS, RL_8139CPLUS, "C+"},
198	{ RL_HWREV_8168_SPIN1, RL_8169, "8168"},
199	{ RL_HWREV_8169, RL_8169, "8169"},
200	{ RL_HWREV_8169S, RL_8169, "8169S"},
201	{ RL_HWREV_8110S, RL_8169, "8110S"},
202	{ RL_HWREV_8169_8110SB, RL_8169, "8169SB"},
203	{ RL_HWREV_8169_8110SC, RL_8169, "8169SC"},
204	{ RL_HWREV_8169_8110SBL, RL_8169, "8169SBL"},
205	{ RL_HWREV_8100, RL_8139, "8100"},
206	{ RL_HWREV_8101, RL_8139, "8101"},
207	{ RL_HWREV_8100E, RL_8169, "8100E"},
208	{ RL_HWREV_8101E, RL_8169, "8101E"},
209	{ RL_HWREV_8102E, RL_8169, "8102E"},
210	{ RL_HWREV_8102EL, RL_8169, "8102EL"},
211	{ RL_HWREV_8168_SPIN2, RL_8169, "8168"},
212	{ RL_HWREV_8168_SPIN3, RL_8169, "8168"},
213	{ RL_HWREV_8168C, RL_8169, "8168C/8111C"},
214	{ RL_HWREV_8168C_SPIN2, RL_8169, "8168C/8111C"},
215	{ RL_HWREV_8168CP, RL_8169, "8168CP/8111CP"},
216	{ RL_HWREV_8168D, RL_8169, "8168D"},
217	{ 0, 0, NULL }
218};
219
220static int re_probe		(device_t);
221static int re_attach		(device_t);
222static int re_detach		(device_t);
223
224static int re_encap		(struct rl_softc *, struct mbuf **);
225
226static void re_dma_map_addr	(void *, bus_dma_segment_t *, int, int);
227static int re_allocmem		(device_t, struct rl_softc *);
228static __inline void re_discard_rxbuf
229				(struct rl_softc *, int);
230static int re_newbuf		(struct rl_softc *, int);
231static int re_rx_list_init	(struct rl_softc *);
232static int re_tx_list_init	(struct rl_softc *);
233#ifdef RE_FIXUP_RX
234static __inline void re_fixup_rx
235				(struct mbuf *);
236#endif
237static int re_rxeof		(struct rl_softc *);
238static void re_txeof		(struct rl_softc *);
239#ifdef DEVICE_POLLING
240static void re_poll		(struct ifnet *, enum poll_cmd, int);
241static void re_poll_locked	(struct ifnet *, enum poll_cmd, int);
242#endif
243static int re_intr		(void *);
244static void re_tick		(void *);
245static void re_tx_task		(void *, int);
246static void re_int_task		(void *, int);
247static void re_start		(struct ifnet *);
248static int re_ioctl		(struct ifnet *, u_long, caddr_t);
249static void re_init		(void *);
250static void re_init_locked	(struct rl_softc *);
251static void re_stop		(struct rl_softc *);
252static void re_watchdog		(struct rl_softc *);
253static int re_suspend		(device_t);
254static int re_resume		(device_t);
255static int re_shutdown		(device_t);
256static int re_ifmedia_upd	(struct ifnet *);
257static void re_ifmedia_sts	(struct ifnet *, struct ifmediareq *);
258
259static void re_eeprom_putbyte	(struct rl_softc *, int);
260static void re_eeprom_getword	(struct rl_softc *, int, u_int16_t *);
261static void re_read_eeprom	(struct rl_softc *, caddr_t, int, int);
262static int re_gmii_readreg	(device_t, int, int);
263static int re_gmii_writereg	(device_t, int, int, int);
264
265static int re_miibus_readreg	(device_t, int, int);
266static int re_miibus_writereg	(device_t, int, int, int);
267static void re_miibus_statchg	(device_t);
268
269static void re_setmulti		(struct rl_softc *);
270static void re_reset		(struct rl_softc *);
271static void re_setwol		(struct rl_softc *);
272static void re_clrwol		(struct rl_softc *);
273
274#ifdef RE_DIAG
275static int re_diag		(struct rl_softc *);
276#endif
277
278static device_method_t re_methods[] = {
279	/* Device interface */
280	DEVMETHOD(device_probe,		re_probe),
281	DEVMETHOD(device_attach,	re_attach),
282	DEVMETHOD(device_detach,	re_detach),
283	DEVMETHOD(device_suspend,	re_suspend),
284	DEVMETHOD(device_resume,	re_resume),
285	DEVMETHOD(device_shutdown,	re_shutdown),
286
287	/* bus interface */
288	DEVMETHOD(bus_print_child,	bus_generic_print_child),
289	DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
290
291	/* MII interface */
292	DEVMETHOD(miibus_readreg,	re_miibus_readreg),
293	DEVMETHOD(miibus_writereg,	re_miibus_writereg),
294	DEVMETHOD(miibus_statchg,	re_miibus_statchg),
295
296	{ 0, 0 }
297};
298
299static driver_t re_driver = {
300	"re",
301	re_methods,
302	sizeof(struct rl_softc)
303};
304
305static devclass_t re_devclass;
306
307DRIVER_MODULE(re, pci, re_driver, re_devclass, 0, 0);
308DRIVER_MODULE(re, cardbus, re_driver, re_devclass, 0, 0);
309DRIVER_MODULE(miibus, re, miibus_driver, miibus_devclass, 0, 0);
310
311#define EE_SET(x)					\
312	CSR_WRITE_1(sc, RL_EECMD,			\
313		CSR_READ_1(sc, RL_EECMD) | x)
314
315#define EE_CLR(x)					\
316	CSR_WRITE_1(sc, RL_EECMD,			\
317		CSR_READ_1(sc, RL_EECMD) & ~x)
318
319/*
320 * Send a read command and address to the EEPROM, check for ACK.
321 */
322static void
323re_eeprom_putbyte(struct rl_softc *sc, int addr)
324{
325	int			d, i;
326
327	d = addr | (RL_9346_READ << sc->rl_eewidth);
328
329	/*
330	 * Feed in each bit and strobe the clock.
331	 */
332
333	for (i = 1 << (sc->rl_eewidth + 3); i; i >>= 1) {
334		if (d & i) {
335			EE_SET(RL_EE_DATAIN);
336		} else {
337			EE_CLR(RL_EE_DATAIN);
338		}
339		DELAY(100);
340		EE_SET(RL_EE_CLK);
341		DELAY(150);
342		EE_CLR(RL_EE_CLK);
343		DELAY(100);
344	}
345}
346
347/*
348 * Read a word of data stored in the EEPROM at address 'addr.'
349 */
350static void
351re_eeprom_getword(struct rl_softc *sc, int addr, u_int16_t *dest)
352{
353	int			i;
354	u_int16_t		word = 0;
355
356	/*
357	 * Send address of word we want to read.
358	 */
359	re_eeprom_putbyte(sc, addr);
360
361	/*
362	 * Start reading bits from EEPROM.
363	 */
364	for (i = 0x8000; i; i >>= 1) {
365		EE_SET(RL_EE_CLK);
366		DELAY(100);
367		if (CSR_READ_1(sc, RL_EECMD) & RL_EE_DATAOUT)
368			word |= i;
369		EE_CLR(RL_EE_CLK);
370		DELAY(100);
371	}
372
373	*dest = word;
374}
375
376/*
377 * Read a sequence of words from the EEPROM.
378 */
379static void
380re_read_eeprom(struct rl_softc *sc, caddr_t dest, int off, int cnt)
381{
382	int			i;
383	u_int16_t		word = 0, *ptr;
384
385	CSR_SETBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM);
386
387        DELAY(100);
388
389	for (i = 0; i < cnt; i++) {
390		CSR_SETBIT_1(sc, RL_EECMD, RL_EE_SEL);
391		re_eeprom_getword(sc, off + i, &word);
392		CSR_CLRBIT_1(sc, RL_EECMD, RL_EE_SEL);
393		ptr = (u_int16_t *)(dest + (i * 2));
394                *ptr = word;
395	}
396
397	CSR_CLRBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM);
398}
399
400static int
401re_gmii_readreg(device_t dev, int phy, int reg)
402{
403	struct rl_softc		*sc;
404	u_int32_t		rval;
405	int			i;
406
407	if (phy != 1)
408		return (0);
409
410	sc = device_get_softc(dev);
411
412	/* Let the rgephy driver read the GMEDIASTAT register */
413
414	if (reg == RL_GMEDIASTAT) {
415		rval = CSR_READ_1(sc, RL_GMEDIASTAT);
416		return (rval);
417	}
418
419	CSR_WRITE_4(sc, RL_PHYAR, reg << 16);
420	DELAY(1000);
421
422	for (i = 0; i < RL_TIMEOUT; i++) {
423		rval = CSR_READ_4(sc, RL_PHYAR);
424		if (rval & RL_PHYAR_BUSY)
425			break;
426		DELAY(100);
427	}
428
429	if (i == RL_TIMEOUT) {
430		device_printf(sc->rl_dev, "PHY read failed\n");
431		return (0);
432	}
433
434	return (rval & RL_PHYAR_PHYDATA);
435}
436
437static int
438re_gmii_writereg(device_t dev, int phy, int reg, int data)
439{
440	struct rl_softc		*sc;
441	u_int32_t		rval;
442	int			i;
443
444	sc = device_get_softc(dev);
445
446	CSR_WRITE_4(sc, RL_PHYAR, (reg << 16) |
447	    (data & RL_PHYAR_PHYDATA) | RL_PHYAR_BUSY);
448	DELAY(1000);
449
450	for (i = 0; i < RL_TIMEOUT; i++) {
451		rval = CSR_READ_4(sc, RL_PHYAR);
452		if (!(rval & RL_PHYAR_BUSY))
453			break;
454		DELAY(100);
455	}
456
457	if (i == RL_TIMEOUT) {
458		device_printf(sc->rl_dev, "PHY write failed\n");
459		return (0);
460	}
461
462	return (0);
463}
464
465static int
466re_miibus_readreg(device_t dev, int phy, int reg)
467{
468	struct rl_softc		*sc;
469	u_int16_t		rval = 0;
470	u_int16_t		re8139_reg = 0;
471
472	sc = device_get_softc(dev);
473
474	if (sc->rl_type == RL_8169) {
475		rval = re_gmii_readreg(dev, phy, reg);
476		return (rval);
477	}
478
479	/* Pretend the internal PHY is only at address 0 */
480	if (phy) {
481		return (0);
482	}
483	switch (reg) {
484	case MII_BMCR:
485		re8139_reg = RL_BMCR;
486		break;
487	case MII_BMSR:
488		re8139_reg = RL_BMSR;
489		break;
490	case MII_ANAR:
491		re8139_reg = RL_ANAR;
492		break;
493	case MII_ANER:
494		re8139_reg = RL_ANER;
495		break;
496	case MII_ANLPAR:
497		re8139_reg = RL_LPAR;
498		break;
499	case MII_PHYIDR1:
500	case MII_PHYIDR2:
501		return (0);
502	/*
503	 * Allow the rlphy driver to read the media status
504	 * register. If we have a link partner which does not
505	 * support NWAY, this is the register which will tell
506	 * us the results of parallel detection.
507	 */
508	case RL_MEDIASTAT:
509		rval = CSR_READ_1(sc, RL_MEDIASTAT);
510		return (rval);
511	default:
512		device_printf(sc->rl_dev, "bad phy register\n");
513		return (0);
514	}
515	rval = CSR_READ_2(sc, re8139_reg);
516	if (sc->rl_type == RL_8139CPLUS && re8139_reg == RL_BMCR) {
517		/* 8139C+ has different bit layout. */
518		rval &= ~(BMCR_LOOP | BMCR_ISO);
519	}
520	return (rval);
521}
522
523static int
524re_miibus_writereg(device_t dev, int phy, int reg, int data)
525{
526	struct rl_softc		*sc;
527	u_int16_t		re8139_reg = 0;
528	int			rval = 0;
529
530	sc = device_get_softc(dev);
531
532	if (sc->rl_type == RL_8169) {
533		rval = re_gmii_writereg(dev, phy, reg, data);
534		return (rval);
535	}
536
537	/* Pretend the internal PHY is only at address 0 */
538	if (phy)
539		return (0);
540
541	switch (reg) {
542	case MII_BMCR:
543		re8139_reg = RL_BMCR;
544		if (sc->rl_type == RL_8139CPLUS) {
545			/* 8139C+ has different bit layout. */
546			data &= ~(BMCR_LOOP | BMCR_ISO);
547		}
548		break;
549	case MII_BMSR:
550		re8139_reg = RL_BMSR;
551		break;
552	case MII_ANAR:
553		re8139_reg = RL_ANAR;
554		break;
555	case MII_ANER:
556		re8139_reg = RL_ANER;
557		break;
558	case MII_ANLPAR:
559		re8139_reg = RL_LPAR;
560		break;
561	case MII_PHYIDR1:
562	case MII_PHYIDR2:
563		return (0);
564		break;
565	default:
566		device_printf(sc->rl_dev, "bad phy register\n");
567		return (0);
568	}
569	CSR_WRITE_2(sc, re8139_reg, data);
570	return (0);
571}
572
573static void
574re_miibus_statchg(device_t dev)
575{
576
577}
578
579/*
580 * Program the 64-bit multicast hash filter.
581 */
582static void
583re_setmulti(struct rl_softc *sc)
584{
585	struct ifnet		*ifp;
586	int			h = 0;
587	u_int32_t		hashes[2] = { 0, 0 };
588	struct ifmultiaddr	*ifma;
589	u_int32_t		rxfilt;
590	int			mcnt = 0;
591
592	RL_LOCK_ASSERT(sc);
593
594	ifp = sc->rl_ifp;
595
596
597	rxfilt = CSR_READ_4(sc, RL_RXCFG);
598	rxfilt &= ~(RL_RXCFG_RX_ALLPHYS | RL_RXCFG_RX_MULTI);
599	if (ifp->if_flags & IFF_ALLMULTI || ifp->if_flags & IFF_PROMISC) {
600		if (ifp->if_flags & IFF_PROMISC)
601			rxfilt |= RL_RXCFG_RX_ALLPHYS;
602		/*
603		 * Unlike other hardwares, we have to explicitly set
604		 * RL_RXCFG_RX_MULTI to receive multicast frames in
605		 * promiscuous mode.
606		 */
607		rxfilt |= RL_RXCFG_RX_MULTI;
608		CSR_WRITE_4(sc, RL_RXCFG, rxfilt);
609		CSR_WRITE_4(sc, RL_MAR0, 0xFFFFFFFF);
610		CSR_WRITE_4(sc, RL_MAR4, 0xFFFFFFFF);
611		return;
612	}
613
614	/* first, zot all the existing hash bits */
615	CSR_WRITE_4(sc, RL_MAR0, 0);
616	CSR_WRITE_4(sc, RL_MAR4, 0);
617
618	/* now program new ones */
619	IF_ADDR_LOCK(ifp);
620	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
621		if (ifma->ifma_addr->sa_family != AF_LINK)
622			continue;
623		h = ether_crc32_be(LLADDR((struct sockaddr_dl *)
624		    ifma->ifma_addr), ETHER_ADDR_LEN) >> 26;
625		if (h < 32)
626			hashes[0] |= (1 << h);
627		else
628			hashes[1] |= (1 << (h - 32));
629		mcnt++;
630	}
631	IF_ADDR_UNLOCK(ifp);
632
633	if (mcnt)
634		rxfilt |= RL_RXCFG_RX_MULTI;
635	else
636		rxfilt &= ~RL_RXCFG_RX_MULTI;
637
638	CSR_WRITE_4(sc, RL_RXCFG, rxfilt);
639
640	/*
641	 * For some unfathomable reason, RealTek decided to reverse
642	 * the order of the multicast hash registers in the PCI Express
643	 * parts. This means we have to write the hash pattern in reverse
644	 * order for those devices.
645	 */
646
647	if ((sc->rl_flags & RL_FLAG_INVMAR) != 0) {
648		CSR_WRITE_4(sc, RL_MAR0, bswap32(hashes[1]));
649		CSR_WRITE_4(sc, RL_MAR4, bswap32(hashes[0]));
650	} else {
651		CSR_WRITE_4(sc, RL_MAR0, hashes[0]);
652		CSR_WRITE_4(sc, RL_MAR4, hashes[1]);
653	}
654}
655
656static void
657re_reset(struct rl_softc *sc)
658{
659	int			i;
660
661	RL_LOCK_ASSERT(sc);
662
663	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RESET);
664
665	for (i = 0; i < RL_TIMEOUT; i++) {
666		DELAY(10);
667		if (!(CSR_READ_1(sc, RL_COMMAND) & RL_CMD_RESET))
668			break;
669	}
670	if (i == RL_TIMEOUT)
671		device_printf(sc->rl_dev, "reset never completed!\n");
672
673	CSR_WRITE_1(sc, 0x82, 1);
674}
675
676#ifdef RE_DIAG
677
678/*
679 * The following routine is designed to test for a defect on some
680 * 32-bit 8169 cards. Some of these NICs have the REQ64# and ACK64#
681 * lines connected to the bus, however for a 32-bit only card, they
682 * should be pulled high. The result of this defect is that the
683 * NIC will not work right if you plug it into a 64-bit slot: DMA
684 * operations will be done with 64-bit transfers, which will fail
685 * because the 64-bit data lines aren't connected.
686 *
687 * There's no way to work around this (short of talking a soldering
688 * iron to the board), however we can detect it. The method we use
689 * here is to put the NIC into digital loopback mode, set the receiver
690 * to promiscuous mode, and then try to send a frame. We then compare
691 * the frame data we sent to what was received. If the data matches,
692 * then the NIC is working correctly, otherwise we know the user has
693 * a defective NIC which has been mistakenly plugged into a 64-bit PCI
694 * slot. In the latter case, there's no way the NIC can work correctly,
695 * so we print out a message on the console and abort the device attach.
696 */
697
698static int
699re_diag(struct rl_softc *sc)
700{
701	struct ifnet		*ifp = sc->rl_ifp;
702	struct mbuf		*m0;
703	struct ether_header	*eh;
704	struct rl_desc		*cur_rx;
705	u_int16_t		status;
706	u_int32_t		rxstat;
707	int			total_len, i, error = 0, phyaddr;
708	u_int8_t		dst[] = { 0x00, 'h', 'e', 'l', 'l', 'o' };
709	u_int8_t		src[] = { 0x00, 'w', 'o', 'r', 'l', 'd' };
710
711	/* Allocate a single mbuf */
712	MGETHDR(m0, M_DONTWAIT, MT_DATA);
713	if (m0 == NULL)
714		return (ENOBUFS);
715
716	RL_LOCK(sc);
717
718	/*
719	 * Initialize the NIC in test mode. This sets the chip up
720	 * so that it can send and receive frames, but performs the
721	 * following special functions:
722	 * - Puts receiver in promiscuous mode
723	 * - Enables digital loopback mode
724	 * - Leaves interrupts turned off
725	 */
726
727	ifp->if_flags |= IFF_PROMISC;
728	sc->rl_testmode = 1;
729	re_reset(sc);
730	re_init_locked(sc);
731	sc->rl_flags |= RL_FLAG_LINK;
732	if (sc->rl_type == RL_8169)
733		phyaddr = 1;
734	else
735		phyaddr = 0;
736
737	re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_RESET);
738	for (i = 0; i < RL_TIMEOUT; i++) {
739		status = re_miibus_readreg(sc->rl_dev, phyaddr, MII_BMCR);
740		if (!(status & BMCR_RESET))
741			break;
742	}
743
744	re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_LOOP);
745	CSR_WRITE_2(sc, RL_ISR, RL_INTRS);
746
747	DELAY(100000);
748
749	/* Put some data in the mbuf */
750
751	eh = mtod(m0, struct ether_header *);
752	bcopy ((char *)&dst, eh->ether_dhost, ETHER_ADDR_LEN);
753	bcopy ((char *)&src, eh->ether_shost, ETHER_ADDR_LEN);
754	eh->ether_type = htons(ETHERTYPE_IP);
755	m0->m_pkthdr.len = m0->m_len = ETHER_MIN_LEN - ETHER_CRC_LEN;
756
757	/*
758	 * Queue the packet, start transmission.
759	 * Note: IF_HANDOFF() ultimately calls re_start() for us.
760	 */
761
762	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
763	RL_UNLOCK(sc);
764	/* XXX: re_diag must not be called when in ALTQ mode */
765	IF_HANDOFF(&ifp->if_snd, m0, ifp);
766	RL_LOCK(sc);
767	m0 = NULL;
768
769	/* Wait for it to propagate through the chip */
770
771	DELAY(100000);
772	for (i = 0; i < RL_TIMEOUT; i++) {
773		status = CSR_READ_2(sc, RL_ISR);
774		CSR_WRITE_2(sc, RL_ISR, status);
775		if ((status & (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK)) ==
776		    (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK))
777			break;
778		DELAY(10);
779	}
780
781	if (i == RL_TIMEOUT) {
782		device_printf(sc->rl_dev,
783		    "diagnostic failed, failed to receive packet in"
784		    " loopback mode\n");
785		error = EIO;
786		goto done;
787	}
788
789	/*
790	 * The packet should have been dumped into the first
791	 * entry in the RX DMA ring. Grab it from there.
792	 */
793
794	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
795	    sc->rl_ldata.rl_rx_list_map,
796	    BUS_DMASYNC_POSTREAD);
797	bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag,
798	    sc->rl_ldata.rl_rx_desc[0].rx_dmamap,
799	    BUS_DMASYNC_POSTREAD);
800	bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag,
801	    sc->rl_ldata.rl_rx_desc[0].rx_dmamap);
802
803	m0 = sc->rl_ldata.rl_rx_desc[0].rx_m;
804	sc->rl_ldata.rl_rx_desc[0].rx_m = NULL;
805	eh = mtod(m0, struct ether_header *);
806
807	cur_rx = &sc->rl_ldata.rl_rx_list[0];
808	total_len = RL_RXBYTES(cur_rx);
809	rxstat = le32toh(cur_rx->rl_cmdstat);
810
811	if (total_len != ETHER_MIN_LEN) {
812		device_printf(sc->rl_dev,
813		    "diagnostic failed, received short packet\n");
814		error = EIO;
815		goto done;
816	}
817
818	/* Test that the received packet data matches what we sent. */
819
820	if (bcmp((char *)&eh->ether_dhost, (char *)&dst, ETHER_ADDR_LEN) ||
821	    bcmp((char *)&eh->ether_shost, (char *)&src, ETHER_ADDR_LEN) ||
822	    ntohs(eh->ether_type) != ETHERTYPE_IP) {
823		device_printf(sc->rl_dev, "WARNING, DMA FAILURE!\n");
824		device_printf(sc->rl_dev, "expected TX data: %6D/%6D/0x%x\n",
825		    dst, ":", src, ":", ETHERTYPE_IP);
826		device_printf(sc->rl_dev, "received RX data: %6D/%6D/0x%x\n",
827		    eh->ether_dhost, ":",  eh->ether_shost, ":",
828		    ntohs(eh->ether_type));
829		device_printf(sc->rl_dev, "You may have a defective 32-bit "
830		    "NIC plugged into a 64-bit PCI slot.\n");
831		device_printf(sc->rl_dev, "Please re-install the NIC in a "
832		    "32-bit slot for proper operation.\n");
833		device_printf(sc->rl_dev, "Read the re(4) man page for more "
834		    "details.\n");
835		error = EIO;
836	}
837
838done:
839	/* Turn interface off, release resources */
840
841	sc->rl_testmode = 0;
842	sc->rl_flags &= ~RL_FLAG_LINK;
843	ifp->if_flags &= ~IFF_PROMISC;
844	re_stop(sc);
845	if (m0 != NULL)
846		m_freem(m0);
847
848	RL_UNLOCK(sc);
849
850	return (error);
851}
852
853#endif
854
855/*
856 * Probe for a RealTek 8139C+/8169/8110 chip. Check the PCI vendor and device
857 * IDs against our list and return a device name if we find a match.
858 */
859static int
860re_probe(device_t dev)
861{
862	struct rl_type		*t;
863	uint16_t		devid, vendor;
864	uint16_t		revid, sdevid;
865	int			i;
866
867	vendor = pci_get_vendor(dev);
868	devid = pci_get_device(dev);
869	revid = pci_get_revid(dev);
870	sdevid = pci_get_subdevice(dev);
871
872	if (vendor == LINKSYS_VENDORID && devid == LINKSYS_DEVICEID_EG1032) {
873		if (sdevid != LINKSYS_SUBDEVICE_EG1032_REV3) {
874			/*
875			 * Only attach to rev. 3 of the Linksys EG1032 adapter.
876			 * Rev. 2 is supported by sk(4).
877			 */
878			return (ENXIO);
879		}
880	}
881
882	if (vendor == RT_VENDORID && devid == RT_DEVICEID_8139) {
883		if (revid != 0x20) {
884			/* 8139, let rl(4) take care of this device. */
885			return (ENXIO);
886		}
887	}
888
889	t = re_devs;
890	for (i = 0; i < sizeof(re_devs) / sizeof(re_devs[0]); i++, t++) {
891		if (vendor == t->rl_vid && devid == t->rl_did) {
892			device_set_desc(dev, t->rl_name);
893			return (BUS_PROBE_DEFAULT);
894		}
895	}
896
897	return (ENXIO);
898}
899
900/*
901 * Map a single buffer address.
902 */
903
904static void
905re_dma_map_addr(void *arg, bus_dma_segment_t *segs, int nseg, int error)
906{
907	bus_addr_t		*addr;
908
909	if (error)
910		return;
911
912	KASSERT(nseg == 1, ("too many DMA segments, %d should be 1", nseg));
913	addr = arg;
914	*addr = segs->ds_addr;
915}
916
917static int
918re_allocmem(device_t dev, struct rl_softc *sc)
919{
920	bus_size_t		rx_list_size, tx_list_size;
921	int			error;
922	int			i;
923
924	rx_list_size = sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc);
925	tx_list_size = sc->rl_ldata.rl_tx_desc_cnt * sizeof(struct rl_desc);
926
927	/*
928	 * Allocate the parent bus DMA tag appropriate for PCI.
929	 * In order to use DAC, RL_CPLUSCMD_PCI_DAC bit of RL_CPLUS_CMD
930	 * register should be set. However some RealTek chips are known
931	 * to be buggy on DAC handling, therefore disable DAC by limiting
932	 * DMA address space to 32bit. PCIe variants of RealTek chips
933	 * may not have the limitation but I took safer path.
934	 */
935	error = bus_dma_tag_create(bus_get_dma_tag(dev), 1, 0,
936	    BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL, NULL,
937	    BUS_SPACE_MAXSIZE_32BIT, 0, BUS_SPACE_MAXSIZE_32BIT, 0,
938	    NULL, NULL, &sc->rl_parent_tag);
939	if (error) {
940		device_printf(dev, "could not allocate parent DMA tag\n");
941		return (error);
942	}
943
944	/*
945	 * Allocate map for TX mbufs.
946	 */
947	error = bus_dma_tag_create(sc->rl_parent_tag, 1, 0,
948	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL,
949	    NULL, MCLBYTES * RL_NTXSEGS, RL_NTXSEGS, 4096, 0,
950	    NULL, NULL, &sc->rl_ldata.rl_tx_mtag);
951	if (error) {
952		device_printf(dev, "could not allocate TX DMA tag\n");
953		return (error);
954	}
955
956	/*
957	 * Allocate map for RX mbufs.
958	 */
959
960	error = bus_dma_tag_create(sc->rl_parent_tag, sizeof(uint64_t), 0,
961	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
962	    MCLBYTES, 1, MCLBYTES, 0, NULL, NULL, &sc->rl_ldata.rl_rx_mtag);
963	if (error) {
964		device_printf(dev, "could not allocate RX DMA tag\n");
965		return (error);
966	}
967
968	/*
969	 * Allocate map for TX descriptor list.
970	 */
971	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
972	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
973	    NULL, tx_list_size, 1, tx_list_size, 0,
974	    NULL, NULL, &sc->rl_ldata.rl_tx_list_tag);
975	if (error) {
976		device_printf(dev, "could not allocate TX DMA ring tag\n");
977		return (error);
978	}
979
980	/* Allocate DMA'able memory for the TX ring */
981
982	error = bus_dmamem_alloc(sc->rl_ldata.rl_tx_list_tag,
983	    (void **)&sc->rl_ldata.rl_tx_list,
984	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
985	    &sc->rl_ldata.rl_tx_list_map);
986	if (error) {
987		device_printf(dev, "could not allocate TX DMA ring\n");
988		return (error);
989	}
990
991	/* Load the map for the TX ring. */
992
993	sc->rl_ldata.rl_tx_list_addr = 0;
994	error = bus_dmamap_load(sc->rl_ldata.rl_tx_list_tag,
995	     sc->rl_ldata.rl_tx_list_map, sc->rl_ldata.rl_tx_list,
996	     tx_list_size, re_dma_map_addr,
997	     &sc->rl_ldata.rl_tx_list_addr, BUS_DMA_NOWAIT);
998	if (error != 0 || sc->rl_ldata.rl_tx_list_addr == 0) {
999		device_printf(dev, "could not load TX DMA ring\n");
1000		return (ENOMEM);
1001	}
1002
1003	/* Create DMA maps for TX buffers */
1004
1005	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
1006		error = bus_dmamap_create(sc->rl_ldata.rl_tx_mtag, 0,
1007		    &sc->rl_ldata.rl_tx_desc[i].tx_dmamap);
1008		if (error) {
1009			device_printf(dev, "could not create DMA map for TX\n");
1010			return (error);
1011		}
1012	}
1013
1014	/*
1015	 * Allocate map for RX descriptor list.
1016	 */
1017	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
1018	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1019	    NULL, rx_list_size, 1, rx_list_size, 0,
1020	    NULL, NULL, &sc->rl_ldata.rl_rx_list_tag);
1021	if (error) {
1022		device_printf(dev, "could not create RX DMA ring tag\n");
1023		return (error);
1024	}
1025
1026	/* Allocate DMA'able memory for the RX ring */
1027
1028	error = bus_dmamem_alloc(sc->rl_ldata.rl_rx_list_tag,
1029	    (void **)&sc->rl_ldata.rl_rx_list,
1030	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1031	    &sc->rl_ldata.rl_rx_list_map);
1032	if (error) {
1033		device_printf(dev, "could not allocate RX DMA ring\n");
1034		return (error);
1035	}
1036
1037	/* Load the map for the RX ring. */
1038
1039	sc->rl_ldata.rl_rx_list_addr = 0;
1040	error = bus_dmamap_load(sc->rl_ldata.rl_rx_list_tag,
1041	     sc->rl_ldata.rl_rx_list_map, sc->rl_ldata.rl_rx_list,
1042	     rx_list_size, re_dma_map_addr,
1043	     &sc->rl_ldata.rl_rx_list_addr, BUS_DMA_NOWAIT);
1044	if (error != 0 || sc->rl_ldata.rl_rx_list_addr == 0) {
1045		device_printf(dev, "could not load RX DMA ring\n");
1046		return (ENOMEM);
1047	}
1048
1049	/* Create DMA maps for RX buffers */
1050
1051	error = bus_dmamap_create(sc->rl_ldata.rl_rx_mtag, 0,
1052	    &sc->rl_ldata.rl_rx_sparemap);
1053	if (error) {
1054		device_printf(dev, "could not create spare DMA map for RX\n");
1055		return (error);
1056	}
1057	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1058		error = bus_dmamap_create(sc->rl_ldata.rl_rx_mtag, 0,
1059		    &sc->rl_ldata.rl_rx_desc[i].rx_dmamap);
1060		if (error) {
1061			device_printf(dev, "could not create DMA map for RX\n");
1062			return (error);
1063		}
1064	}
1065
1066	return (0);
1067}
1068
1069/*
1070 * Attach the interface. Allocate softc structures, do ifmedia
1071 * setup and ethernet/BPF attach.
1072 */
1073static int
1074re_attach(device_t dev)
1075{
1076	u_char			eaddr[ETHER_ADDR_LEN];
1077	u_int16_t		as[ETHER_ADDR_LEN / 2];
1078	struct rl_softc		*sc;
1079	struct ifnet		*ifp;
1080	struct rl_hwrev		*hw_rev;
1081	int			hwrev;
1082	u_int16_t		devid, re_did = 0;
1083	int			error = 0, rid, i;
1084	int			msic, reg;
1085	uint8_t			cfg;
1086
1087	sc = device_get_softc(dev);
1088	sc->rl_dev = dev;
1089
1090	mtx_init(&sc->rl_mtx, device_get_nameunit(dev), MTX_NETWORK_LOCK,
1091	    MTX_DEF);
1092	callout_init_mtx(&sc->rl_stat_callout, &sc->rl_mtx, 0);
1093
1094	/*
1095	 * Map control/status registers.
1096	 */
1097	pci_enable_busmaster(dev);
1098
1099	devid = pci_get_device(dev);
1100	/* Prefer memory space register mapping over IO space. */
1101	sc->rl_res_id = PCIR_BAR(1);
1102	sc->rl_res_type = SYS_RES_MEMORY;
1103	/* RTL8168/8101E seems to use different BARs. */
1104	if (devid == RT_DEVICEID_8168 || devid == RT_DEVICEID_8101E)
1105		sc->rl_res_id = PCIR_BAR(2);
1106	sc->rl_res = bus_alloc_resource_any(dev, sc->rl_res_type,
1107	    &sc->rl_res_id, RF_ACTIVE);
1108
1109	if (sc->rl_res == NULL) {
1110		sc->rl_res_id = PCIR_BAR(0);
1111		sc->rl_res_type = SYS_RES_IOPORT;
1112		sc->rl_res = bus_alloc_resource_any(dev, sc->rl_res_type,
1113		    &sc->rl_res_id, RF_ACTIVE);
1114		if (sc->rl_res == NULL) {
1115			device_printf(dev, "couldn't map ports/memory\n");
1116			error = ENXIO;
1117			goto fail;
1118		}
1119	}
1120
1121	sc->rl_btag = rman_get_bustag(sc->rl_res);
1122	sc->rl_bhandle = rman_get_bushandle(sc->rl_res);
1123
1124	msic = 0;
1125	if (pci_find_extcap(dev, PCIY_EXPRESS, &reg) == 0) {
1126		msic = pci_msi_count(dev);
1127		if (bootverbose)
1128			device_printf(dev, "MSI count : %d\n", msic);
1129	}
1130	if (msic == RL_MSI_MESSAGES  && msi_disable == 0) {
1131		if (pci_alloc_msi(dev, &msic) == 0) {
1132			if (msic == RL_MSI_MESSAGES) {
1133				device_printf(dev, "Using %d MSI messages\n",
1134				    msic);
1135				sc->rl_flags |= RL_FLAG_MSI;
1136				/* Explicitly set MSI enable bit. */
1137				CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1138				cfg = CSR_READ_1(sc, RL_CFG2);
1139				cfg |= RL_CFG2_MSI;
1140				CSR_WRITE_1(sc, RL_CFG2, cfg);
1141				CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1142			} else
1143				pci_release_msi(dev);
1144		}
1145	}
1146
1147	/* Allocate interrupt */
1148	if ((sc->rl_flags & RL_FLAG_MSI) == 0) {
1149		rid = 0;
1150		sc->rl_irq[0] = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
1151		    RF_SHAREABLE | RF_ACTIVE);
1152		if (sc->rl_irq[0] == NULL) {
1153			device_printf(dev, "couldn't allocate IRQ resources\n");
1154			error = ENXIO;
1155			goto fail;
1156		}
1157	} else {
1158		for (i = 0, rid = 1; i < RL_MSI_MESSAGES; i++, rid++) {
1159			sc->rl_irq[i] = bus_alloc_resource_any(dev,
1160			    SYS_RES_IRQ, &rid, RF_ACTIVE);
1161			if (sc->rl_irq[i] == NULL) {
1162				device_printf(dev,
1163				    "couldn't llocate IRQ resources for "
1164				    "message %d\n", rid);
1165				error = ENXIO;
1166				goto fail;
1167			}
1168		}
1169	}
1170
1171	if ((sc->rl_flags & RL_FLAG_MSI) == 0) {
1172		CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1173		cfg = CSR_READ_1(sc, RL_CFG2);
1174		if ((cfg & RL_CFG2_MSI) != 0) {
1175			device_printf(dev, "turning off MSI enable bit.\n");
1176			cfg &= ~RL_CFG2_MSI;
1177			CSR_WRITE_1(sc, RL_CFG2, cfg);
1178		}
1179		CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1180	}
1181
1182	/* Reset the adapter. */
1183	RL_LOCK(sc);
1184	re_reset(sc);
1185	RL_UNLOCK(sc);
1186
1187	hw_rev = re_hwrevs;
1188	hwrev = CSR_READ_4(sc, RL_TXCFG);
1189	device_printf(dev, "Chip rev. 0x%08x\n", hwrev & 0x7c800000);
1190	device_printf(dev, "MAC rev. 0x%08x\n", hwrev & 0x00700000);
1191	hwrev &= RL_TXCFG_HWREV;
1192	while (hw_rev->rl_desc != NULL) {
1193		if (hw_rev->rl_rev == hwrev) {
1194			sc->rl_type = hw_rev->rl_type;
1195			break;
1196		}
1197		hw_rev++;
1198	}
1199	if (hw_rev->rl_desc == NULL) {
1200		device_printf(dev, "Unknown H/W revision: 0x%08x\n", hwrev);
1201		error = ENXIO;
1202		goto fail;
1203	}
1204
1205	switch (hw_rev->rl_rev) {
1206	case RL_HWREV_8139CPLUS:
1207		sc->rl_flags |= RL_FLAG_NOJUMBO;
1208		break;
1209	case RL_HWREV_8100E:
1210	case RL_HWREV_8101E:
1211		sc->rl_flags |= RL_FLAG_NOJUMBO | RL_FLAG_INVMAR |
1212		    RL_FLAG_PHYWAKE;
1213		break;
1214	case RL_HWREV_8102E:
1215	case RL_HWREV_8102EL:
1216		sc->rl_flags |= RL_FLAG_NOJUMBO | RL_FLAG_INVMAR |
1217		    RL_FLAG_PHYWAKE | RL_FLAG_PAR | RL_FLAG_DESCV2 |
1218		    RL_FLAG_MACSTAT;
1219		break;
1220	case RL_HWREV_8168_SPIN1:
1221	case RL_HWREV_8168_SPIN2:
1222	case RL_HWREV_8168_SPIN3:
1223		sc->rl_flags |= RL_FLAG_INVMAR | RL_FLAG_PHYWAKE |
1224		    RL_FLAG_MACSTAT;
1225		break;
1226	case RL_HWREV_8168C:
1227	case RL_HWREV_8168C_SPIN2:
1228	case RL_HWREV_8168CP:
1229	case RL_HWREV_8168D:
1230		sc->rl_flags |= RL_FLAG_INVMAR | RL_FLAG_PHYWAKE |
1231		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT;
1232		/*
1233		 * These controllers support jumbo frame but it seems
1234		 * that enabling it requires touching additional magic
1235		 * registers. Depending on MAC revisions some
1236		 * controllers need to disable checksum offload. So
1237		 * disable jumbo frame until I have better idea what
1238		 * it really requires to make it support.
1239		 * RTL8168C/CP : supports up to 6KB jumbo frame.
1240		 * RTL8111C/CP : supports up to 9KB jumbo frame.
1241		 */
1242		sc->rl_flags |= RL_FLAG_NOJUMBO;
1243		break;
1244	case RL_HWREV_8169_8110SB:
1245	case RL_HWREV_8169_8110SC:
1246	case RL_HWREV_8169_8110SBL:
1247		sc->rl_flags |= RL_FLAG_PHYWAKE;
1248		break;
1249	default:
1250		break;
1251	}
1252
1253	/* Enable PME. */
1254	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1255	cfg = CSR_READ_1(sc, RL_CFG1);
1256	cfg |= RL_CFG1_PME;
1257	CSR_WRITE_1(sc, RL_CFG1, cfg);
1258	cfg = CSR_READ_1(sc, RL_CFG5);
1259	cfg &= RL_CFG5_PME_STS;
1260	CSR_WRITE_1(sc, RL_CFG5, cfg);
1261	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1262
1263	if ((sc->rl_flags & RL_FLAG_PAR) != 0) {
1264		/*
1265		 * XXX Should have a better way to extract station
1266		 * address from EEPROM.
1267		 */
1268		for (i = 0; i < ETHER_ADDR_LEN; i++)
1269			eaddr[i] = CSR_READ_1(sc, RL_IDR0 + i);
1270	} else {
1271		sc->rl_eewidth = RL_9356_ADDR_LEN;
1272		re_read_eeprom(sc, (caddr_t)&re_did, 0, 1);
1273		if (re_did != 0x8129)
1274			sc->rl_eewidth = RL_9346_ADDR_LEN;
1275
1276		/*
1277		 * Get station address from the EEPROM.
1278		 */
1279		re_read_eeprom(sc, (caddr_t)as, RL_EE_EADDR, 3);
1280		for (i = 0; i < ETHER_ADDR_LEN / 2; i++)
1281			as[i] = le16toh(as[i]);
1282		bcopy(as, eaddr, sizeof(eaddr));
1283	}
1284
1285	if (sc->rl_type == RL_8169) {
1286		/* Set RX length mask and number of descriptors. */
1287		sc->rl_rxlenmask = RL_RDESC_STAT_GFRAGLEN;
1288		sc->rl_txstart = RL_GTXSTART;
1289		sc->rl_ldata.rl_tx_desc_cnt = RL_8169_TX_DESC_CNT;
1290		sc->rl_ldata.rl_rx_desc_cnt = RL_8169_RX_DESC_CNT;
1291	} else {
1292		/* Set RX length mask and number of descriptors. */
1293		sc->rl_rxlenmask = RL_RDESC_STAT_FRAGLEN;
1294		sc->rl_txstart = RL_TXSTART;
1295		sc->rl_ldata.rl_tx_desc_cnt = RL_8139_TX_DESC_CNT;
1296		sc->rl_ldata.rl_rx_desc_cnt = RL_8139_RX_DESC_CNT;
1297	}
1298
1299	error = re_allocmem(dev, sc);
1300	if (error)
1301		goto fail;
1302
1303	ifp = sc->rl_ifp = if_alloc(IFT_ETHER);
1304	if (ifp == NULL) {
1305		device_printf(dev, "can not if_alloc()\n");
1306		error = ENOSPC;
1307		goto fail;
1308	}
1309
1310	/* Take PHY out of power down mode. */
1311	if ((sc->rl_flags & RL_FLAG_PHYWAKE) != 0) {
1312		re_gmii_writereg(dev, 1, 0x1f, 0);
1313		re_gmii_writereg(dev, 1, 0x0e, 0);
1314	}
1315
1316	/* Do MII setup */
1317	if (mii_phy_probe(dev, &sc->rl_miibus,
1318	    re_ifmedia_upd, re_ifmedia_sts)) {
1319		device_printf(dev, "MII without any phy!\n");
1320		error = ENXIO;
1321		goto fail;
1322	}
1323
1324	ifp->if_softc = sc;
1325	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1326	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1327	ifp->if_ioctl = re_ioctl;
1328	ifp->if_start = re_start;
1329	ifp->if_hwassist = RE_CSUM_FEATURES;
1330	ifp->if_capabilities = IFCAP_HWCSUM;
1331	ifp->if_capenable = ifp->if_capabilities;
1332	ifp->if_init = re_init;
1333	IFQ_SET_MAXLEN(&ifp->if_snd, RL_IFQ_MAXLEN);
1334	ifp->if_snd.ifq_drv_maxlen = RL_IFQ_MAXLEN;
1335	IFQ_SET_READY(&ifp->if_snd);
1336
1337	TASK_INIT(&sc->rl_txtask, 1, re_tx_task, ifp);
1338	TASK_INIT(&sc->rl_inttask, 0, re_int_task, sc);
1339
1340	/*
1341	 * XXX
1342	 * Still have no idea how to make TSO work on 8168C, 8168CP,
1343	 * 8111C and 8111CP.
1344	 */
1345	if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
1346		ifp->if_hwassist |= CSUM_TSO;
1347		ifp->if_capabilities |= IFCAP_TSO4;
1348	}
1349
1350	/*
1351	 * Call MI attach routine.
1352	 */
1353	ether_ifattach(ifp, eaddr);
1354
1355	/* VLAN capability setup */
1356	ifp->if_capabilities |= IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING;
1357	if (ifp->if_capabilities & IFCAP_HWCSUM)
1358		ifp->if_capabilities |= IFCAP_VLAN_HWCSUM;
1359	/* Enable WOL if PM is supported. */
1360	if (pci_find_extcap(sc->rl_dev, PCIY_PMG, &reg) == 0)
1361		ifp->if_capabilities |= IFCAP_WOL;
1362	ifp->if_capenable = ifp->if_capabilities;
1363	/*
1364	 * Don't enable TSO by default. Under certain
1365	 * circumtances the controller generated corrupted
1366	 * packets in TSO size.
1367	 */
1368	ifp->if_hwassist &= ~CSUM_TSO;
1369	ifp->if_capenable &= ~IFCAP_TSO4;
1370#ifdef DEVICE_POLLING
1371	ifp->if_capabilities |= IFCAP_POLLING;
1372#endif
1373	/*
1374	 * Tell the upper layer(s) we support long frames.
1375	 * Must appear after the call to ether_ifattach() because
1376	 * ether_ifattach() sets ifi_hdrlen to the default value.
1377	 */
1378	ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
1379
1380#ifdef RE_DIAG
1381	/*
1382	 * Perform hardware diagnostic on the original RTL8169.
1383	 * Some 32-bit cards were incorrectly wired and would
1384	 * malfunction if plugged into a 64-bit slot.
1385	 */
1386
1387	if (hwrev == RL_HWREV_8169) {
1388		error = re_diag(sc);
1389		if (error) {
1390			device_printf(dev,
1391		    	"attach aborted due to hardware diag failure\n");
1392			ether_ifdetach(ifp);
1393			goto fail;
1394		}
1395	}
1396#endif
1397
1398	/* Hook interrupt last to avoid having to lock softc */
1399	if ((sc->rl_flags & RL_FLAG_MSI) == 0)
1400		error = bus_setup_intr(dev, sc->rl_irq[0],
1401		    INTR_TYPE_NET | INTR_MPSAFE, re_intr, NULL, sc,
1402		    &sc->rl_intrhand[0]);
1403	else {
1404		for (i = 0; i < RL_MSI_MESSAGES; i++) {
1405			error = bus_setup_intr(dev, sc->rl_irq[i],
1406			    INTR_TYPE_NET | INTR_MPSAFE, re_intr, NULL, sc,
1407		    	    &sc->rl_intrhand[i]);
1408			if (error != 0)
1409				break;
1410		}
1411	}
1412	if (error) {
1413		device_printf(dev, "couldn't set up irq\n");
1414		ether_ifdetach(ifp);
1415	}
1416
1417fail:
1418
1419	if (error)
1420		re_detach(dev);
1421
1422	return (error);
1423}
1424
1425/*
1426 * Shutdown hardware and free up resources. This can be called any
1427 * time after the mutex has been initialized. It is called in both
1428 * the error case in attach and the normal detach case so it needs
1429 * to be careful about only freeing resources that have actually been
1430 * allocated.
1431 */
1432static int
1433re_detach(device_t dev)
1434{
1435	struct rl_softc		*sc;
1436	struct ifnet		*ifp;
1437	int			i, rid;
1438
1439	sc = device_get_softc(dev);
1440	ifp = sc->rl_ifp;
1441	KASSERT(mtx_initialized(&sc->rl_mtx), ("re mutex not initialized"));
1442
1443	/* These should only be active if attach succeeded */
1444	if (device_is_attached(dev)) {
1445#ifdef DEVICE_POLLING
1446		if (ifp->if_capenable & IFCAP_POLLING)
1447			ether_poll_deregister(ifp);
1448#endif
1449		RL_LOCK(sc);
1450#if 0
1451		sc->suspended = 1;
1452#endif
1453		re_stop(sc);
1454		RL_UNLOCK(sc);
1455		callout_drain(&sc->rl_stat_callout);
1456		taskqueue_drain(taskqueue_fast, &sc->rl_inttask);
1457		taskqueue_drain(taskqueue_fast, &sc->rl_txtask);
1458		/*
1459		 * Force off the IFF_UP flag here, in case someone
1460		 * still had a BPF descriptor attached to this
1461		 * interface. If they do, ether_ifdetach() will cause
1462		 * the BPF code to try and clear the promisc mode
1463		 * flag, which will bubble down to re_ioctl(),
1464		 * which will try to call re_init() again. This will
1465		 * turn the NIC back on and restart the MII ticker,
1466		 * which will panic the system when the kernel tries
1467		 * to invoke the re_tick() function that isn't there
1468		 * anymore.
1469		 */
1470		ifp->if_flags &= ~IFF_UP;
1471		ether_ifdetach(ifp);
1472	}
1473	if (sc->rl_miibus)
1474		device_delete_child(dev, sc->rl_miibus);
1475	bus_generic_detach(dev);
1476
1477	/*
1478	 * The rest is resource deallocation, so we should already be
1479	 * stopped here.
1480	 */
1481
1482	for (i = 0; i < RL_MSI_MESSAGES; i++) {
1483		if (sc->rl_intrhand[i] != NULL) {
1484			bus_teardown_intr(dev, sc->rl_irq[i],
1485			    sc->rl_intrhand[i]);
1486			sc->rl_intrhand[i] = NULL;
1487		}
1488	}
1489	if (ifp != NULL)
1490		if_free(ifp);
1491	if ((sc->rl_flags & RL_FLAG_MSI) == 0) {
1492		if (sc->rl_irq[0] != NULL) {
1493			bus_release_resource(dev, SYS_RES_IRQ, 0,
1494			    sc->rl_irq[0]);
1495			sc->rl_irq[0] = NULL;
1496		}
1497	} else {
1498		for (i = 0, rid = 1; i < RL_MSI_MESSAGES; i++, rid++) {
1499			if (sc->rl_irq[i] != NULL) {
1500				bus_release_resource(dev, SYS_RES_IRQ, rid,
1501				    sc->rl_irq[i]);
1502				sc->rl_irq[i] = NULL;
1503			}
1504		}
1505		pci_release_msi(dev);
1506	}
1507	if (sc->rl_res)
1508		bus_release_resource(dev, sc->rl_res_type, sc->rl_res_id,
1509		    sc->rl_res);
1510
1511	/* Unload and free the RX DMA ring memory and map */
1512
1513	if (sc->rl_ldata.rl_rx_list_tag) {
1514		bus_dmamap_unload(sc->rl_ldata.rl_rx_list_tag,
1515		    sc->rl_ldata.rl_rx_list_map);
1516		bus_dmamem_free(sc->rl_ldata.rl_rx_list_tag,
1517		    sc->rl_ldata.rl_rx_list,
1518		    sc->rl_ldata.rl_rx_list_map);
1519		bus_dma_tag_destroy(sc->rl_ldata.rl_rx_list_tag);
1520	}
1521
1522	/* Unload and free the TX DMA ring memory and map */
1523
1524	if (sc->rl_ldata.rl_tx_list_tag) {
1525		bus_dmamap_unload(sc->rl_ldata.rl_tx_list_tag,
1526		    sc->rl_ldata.rl_tx_list_map);
1527		bus_dmamem_free(sc->rl_ldata.rl_tx_list_tag,
1528		    sc->rl_ldata.rl_tx_list,
1529		    sc->rl_ldata.rl_tx_list_map);
1530		bus_dma_tag_destroy(sc->rl_ldata.rl_tx_list_tag);
1531	}
1532
1533	/* Destroy all the RX and TX buffer maps */
1534
1535	if (sc->rl_ldata.rl_tx_mtag) {
1536		for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++)
1537			bus_dmamap_destroy(sc->rl_ldata.rl_tx_mtag,
1538			    sc->rl_ldata.rl_tx_desc[i].tx_dmamap);
1539		bus_dma_tag_destroy(sc->rl_ldata.rl_tx_mtag);
1540	}
1541	if (sc->rl_ldata.rl_rx_mtag) {
1542		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++)
1543			bus_dmamap_destroy(sc->rl_ldata.rl_rx_mtag,
1544			    sc->rl_ldata.rl_rx_desc[i].rx_dmamap);
1545		if (sc->rl_ldata.rl_rx_sparemap)
1546			bus_dmamap_destroy(sc->rl_ldata.rl_rx_mtag,
1547			    sc->rl_ldata.rl_rx_sparemap);
1548		bus_dma_tag_destroy(sc->rl_ldata.rl_rx_mtag);
1549	}
1550
1551	/* Unload and free the stats buffer and map */
1552
1553	if (sc->rl_ldata.rl_stag) {
1554		bus_dmamap_unload(sc->rl_ldata.rl_stag,
1555		    sc->rl_ldata.rl_rx_list_map);
1556		bus_dmamem_free(sc->rl_ldata.rl_stag,
1557		    sc->rl_ldata.rl_stats,
1558		    sc->rl_ldata.rl_smap);
1559		bus_dma_tag_destroy(sc->rl_ldata.rl_stag);
1560	}
1561
1562	if (sc->rl_parent_tag)
1563		bus_dma_tag_destroy(sc->rl_parent_tag);
1564
1565	mtx_destroy(&sc->rl_mtx);
1566
1567	return (0);
1568}
1569
1570static __inline void
1571re_discard_rxbuf(struct rl_softc *sc, int idx)
1572{
1573	struct rl_desc		*desc;
1574	struct rl_rxdesc	*rxd;
1575	uint32_t		cmdstat;
1576
1577	rxd = &sc->rl_ldata.rl_rx_desc[idx];
1578	desc = &sc->rl_ldata.rl_rx_list[idx];
1579	desc->rl_vlanctl = 0;
1580	cmdstat = rxd->rx_size;
1581	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
1582		cmdstat |= RL_RDESC_CMD_EOR;
1583	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
1584}
1585
1586static int
1587re_newbuf(struct rl_softc *sc, int idx)
1588{
1589	struct mbuf		*m;
1590	struct rl_rxdesc	*rxd;
1591	bus_dma_segment_t	segs[1];
1592	bus_dmamap_t		map;
1593	struct rl_desc		*desc;
1594	uint32_t		cmdstat;
1595	int			error, nsegs;
1596
1597	m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
1598	if (m == NULL)
1599		return (ENOBUFS);
1600
1601	m->m_len = m->m_pkthdr.len = MCLBYTES;
1602#ifdef RE_FIXUP_RX
1603	/*
1604	 * This is part of an evil trick to deal with non-x86 platforms.
1605	 * The RealTek chip requires RX buffers to be aligned on 64-bit
1606	 * boundaries, but that will hose non-x86 machines. To get around
1607	 * this, we leave some empty space at the start of each buffer
1608	 * and for non-x86 hosts, we copy the buffer back six bytes
1609	 * to achieve word alignment. This is slightly more efficient
1610	 * than allocating a new buffer, copying the contents, and
1611	 * discarding the old buffer.
1612	 */
1613	m_adj(m, RE_ETHER_ALIGN);
1614#endif
1615	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_rx_mtag,
1616	    sc->rl_ldata.rl_rx_sparemap, m, segs, &nsegs, BUS_DMA_NOWAIT);
1617	if (error != 0) {
1618		m_freem(m);
1619		return (ENOBUFS);
1620	}
1621	KASSERT(nsegs == 1, ("%s: %d segment returned!", __func__, nsegs));
1622
1623	rxd = &sc->rl_ldata.rl_rx_desc[idx];
1624	if (rxd->rx_m != NULL) {
1625		bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap,
1626		    BUS_DMASYNC_POSTREAD);
1627		bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap);
1628	}
1629
1630	rxd->rx_m = m;
1631	map = rxd->rx_dmamap;
1632	rxd->rx_dmamap = sc->rl_ldata.rl_rx_sparemap;
1633	rxd->rx_size = segs[0].ds_len;
1634	sc->rl_ldata.rl_rx_sparemap = map;
1635	bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap,
1636	    BUS_DMASYNC_PREREAD);
1637
1638	desc = &sc->rl_ldata.rl_rx_list[idx];
1639	desc->rl_vlanctl = 0;
1640	desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[0].ds_addr));
1641	desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[0].ds_addr));
1642	cmdstat = segs[0].ds_len;
1643	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
1644		cmdstat |= RL_RDESC_CMD_EOR;
1645	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
1646
1647	return (0);
1648}
1649
1650#ifdef RE_FIXUP_RX
1651static __inline void
1652re_fixup_rx(struct mbuf *m)
1653{
1654	int                     i;
1655	uint16_t                *src, *dst;
1656
1657	src = mtod(m, uint16_t *);
1658	dst = src - (RE_ETHER_ALIGN - ETHER_ALIGN) / sizeof *src;
1659
1660	for (i = 0; i < (m->m_len / sizeof(uint16_t) + 1); i++)
1661		*dst++ = *src++;
1662
1663	m->m_data -= RE_ETHER_ALIGN - ETHER_ALIGN;
1664}
1665#endif
1666
1667static int
1668re_tx_list_init(struct rl_softc *sc)
1669{
1670	struct rl_desc		*desc;
1671	int			i;
1672
1673	RL_LOCK_ASSERT(sc);
1674
1675	bzero(sc->rl_ldata.rl_tx_list,
1676	    sc->rl_ldata.rl_tx_desc_cnt * sizeof(struct rl_desc));
1677	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++)
1678		sc->rl_ldata.rl_tx_desc[i].tx_m = NULL;
1679	/* Set EOR. */
1680	desc = &sc->rl_ldata.rl_tx_list[sc->rl_ldata.rl_tx_desc_cnt - 1];
1681	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_EOR);
1682
1683	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
1684	    sc->rl_ldata.rl_tx_list_map,
1685	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1686
1687	sc->rl_ldata.rl_tx_prodidx = 0;
1688	sc->rl_ldata.rl_tx_considx = 0;
1689	sc->rl_ldata.rl_tx_free = sc->rl_ldata.rl_tx_desc_cnt;
1690
1691	return (0);
1692}
1693
1694static int
1695re_rx_list_init(struct rl_softc *sc)
1696{
1697	int			error, i;
1698
1699	bzero(sc->rl_ldata.rl_rx_list,
1700	    sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc));
1701	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1702		sc->rl_ldata.rl_rx_desc[i].rx_m = NULL;
1703		if ((error = re_newbuf(sc, i)) != 0)
1704			return (error);
1705	}
1706
1707	/* Flush the RX descriptors */
1708
1709	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1710	    sc->rl_ldata.rl_rx_list_map,
1711	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
1712
1713	sc->rl_ldata.rl_rx_prodidx = 0;
1714	sc->rl_head = sc->rl_tail = NULL;
1715
1716	return (0);
1717}
1718
1719/*
1720 * RX handler for C+ and 8169. For the gigE chips, we support
1721 * the reception of jumbo frames that have been fragmented
1722 * across multiple 2K mbuf cluster buffers.
1723 */
1724static int
1725re_rxeof(struct rl_softc *sc)
1726{
1727	struct mbuf		*m;
1728	struct ifnet		*ifp;
1729	int			i, total_len;
1730	struct rl_desc		*cur_rx;
1731	u_int32_t		rxstat, rxvlan;
1732	int			maxpkt = 16;
1733
1734	RL_LOCK_ASSERT(sc);
1735
1736	ifp = sc->rl_ifp;
1737
1738	/* Invalidate the descriptor memory */
1739
1740	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1741	    sc->rl_ldata.rl_rx_list_map,
1742	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
1743
1744	for (i = sc->rl_ldata.rl_rx_prodidx; maxpkt > 0;
1745	    i = RL_RX_DESC_NXT(sc, i)) {
1746		cur_rx = &sc->rl_ldata.rl_rx_list[i];
1747		rxstat = le32toh(cur_rx->rl_cmdstat);
1748		if ((rxstat & RL_RDESC_STAT_OWN) != 0)
1749			break;
1750		total_len = rxstat & sc->rl_rxlenmask;
1751		rxvlan = le32toh(cur_rx->rl_vlanctl);
1752		m = sc->rl_ldata.rl_rx_desc[i].rx_m;
1753
1754		if (!(rxstat & RL_RDESC_STAT_EOF)) {
1755			if (re_newbuf(sc, i) != 0) {
1756				/*
1757				 * If this is part of a multi-fragment packet,
1758				 * discard all the pieces.
1759				 */
1760				if (sc->rl_head != NULL) {
1761					m_freem(sc->rl_head);
1762					sc->rl_head = sc->rl_tail = NULL;
1763				}
1764				re_discard_rxbuf(sc, i);
1765				continue;
1766			}
1767			m->m_len = RE_RX_DESC_BUFLEN;
1768			if (sc->rl_head == NULL)
1769				sc->rl_head = sc->rl_tail = m;
1770			else {
1771				m->m_flags &= ~M_PKTHDR;
1772				sc->rl_tail->m_next = m;
1773				sc->rl_tail = m;
1774			}
1775			continue;
1776		}
1777
1778		/*
1779		 * NOTE: for the 8139C+, the frame length field
1780		 * is always 12 bits in size, but for the gigE chips,
1781		 * it is 13 bits (since the max RX frame length is 16K).
1782		 * Unfortunately, all 32 bits in the status word
1783		 * were already used, so to make room for the extra
1784		 * length bit, RealTek took out the 'frame alignment
1785		 * error' bit and shifted the other status bits
1786		 * over one slot. The OWN, EOR, FS and LS bits are
1787		 * still in the same places. We have already extracted
1788		 * the frame length and checked the OWN bit, so rather
1789		 * than using an alternate bit mapping, we shift the
1790		 * status bits one space to the right so we can evaluate
1791		 * them using the 8169 status as though it was in the
1792		 * same format as that of the 8139C+.
1793		 */
1794		if (sc->rl_type == RL_8169)
1795			rxstat >>= 1;
1796
1797		/*
1798		 * if total_len > 2^13-1, both _RXERRSUM and _GIANT will be
1799		 * set, but if CRC is clear, it will still be a valid frame.
1800		 */
1801		if (rxstat & RL_RDESC_STAT_RXERRSUM && !(total_len > 8191 &&
1802		    (rxstat & RL_RDESC_STAT_ERRS) == RL_RDESC_STAT_GIANT)) {
1803			ifp->if_ierrors++;
1804			/*
1805			 * If this is part of a multi-fragment packet,
1806			 * discard all the pieces.
1807			 */
1808			if (sc->rl_head != NULL) {
1809				m_freem(sc->rl_head);
1810				sc->rl_head = sc->rl_tail = NULL;
1811			}
1812			re_discard_rxbuf(sc, i);
1813			continue;
1814		}
1815
1816		/*
1817		 * If allocating a replacement mbuf fails,
1818		 * reload the current one.
1819		 */
1820
1821		if (re_newbuf(sc, i) != 0) {
1822			ifp->if_iqdrops++;
1823			if (sc->rl_head != NULL) {
1824				m_freem(sc->rl_head);
1825				sc->rl_head = sc->rl_tail = NULL;
1826			}
1827			re_discard_rxbuf(sc, i);
1828			continue;
1829		}
1830
1831		if (sc->rl_head != NULL) {
1832			m->m_len = total_len % RE_RX_DESC_BUFLEN;
1833			if (m->m_len == 0)
1834				m->m_len = RE_RX_DESC_BUFLEN;
1835			/*
1836			 * Special case: if there's 4 bytes or less
1837			 * in this buffer, the mbuf can be discarded:
1838			 * the last 4 bytes is the CRC, which we don't
1839			 * care about anyway.
1840			 */
1841			if (m->m_len <= ETHER_CRC_LEN) {
1842				sc->rl_tail->m_len -=
1843				    (ETHER_CRC_LEN - m->m_len);
1844				m_freem(m);
1845			} else {
1846				m->m_len -= ETHER_CRC_LEN;
1847				m->m_flags &= ~M_PKTHDR;
1848				sc->rl_tail->m_next = m;
1849			}
1850			m = sc->rl_head;
1851			sc->rl_head = sc->rl_tail = NULL;
1852			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
1853		} else
1854			m->m_pkthdr.len = m->m_len =
1855			    (total_len - ETHER_CRC_LEN);
1856
1857#ifdef RE_FIXUP_RX
1858		re_fixup_rx(m);
1859#endif
1860		ifp->if_ipackets++;
1861		m->m_pkthdr.rcvif = ifp;
1862
1863		/* Do RX checksumming if enabled */
1864
1865		if (ifp->if_capenable & IFCAP_RXCSUM) {
1866			if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
1867				/* Check IP header checksum */
1868				if (rxstat & RL_RDESC_STAT_PROTOID)
1869					m->m_pkthdr.csum_flags |=
1870					    CSUM_IP_CHECKED;
1871				if (!(rxstat & RL_RDESC_STAT_IPSUMBAD))
1872					m->m_pkthdr.csum_flags |=
1873					    CSUM_IP_VALID;
1874
1875				/* Check TCP/UDP checksum */
1876				if ((RL_TCPPKT(rxstat) &&
1877				    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
1878				    (RL_UDPPKT(rxstat) &&
1879				     !(rxstat & RL_RDESC_STAT_UDPSUMBAD))) {
1880					m->m_pkthdr.csum_flags |=
1881						CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
1882					m->m_pkthdr.csum_data = 0xffff;
1883				}
1884			} else {
1885				/*
1886				 * RTL8168C/RTL816CP/RTL8111C/RTL8111CP
1887				 */
1888				if ((rxstat & RL_RDESC_STAT_PROTOID) &&
1889				    (rxvlan & RL_RDESC_IPV4))
1890					m->m_pkthdr.csum_flags |=
1891					    CSUM_IP_CHECKED;
1892				if (!(rxstat & RL_RDESC_STAT_IPSUMBAD) &&
1893				    (rxvlan & RL_RDESC_IPV4))
1894					m->m_pkthdr.csum_flags |=
1895					    CSUM_IP_VALID;
1896				if (((rxstat & RL_RDESC_STAT_TCP) &&
1897				    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
1898				    ((rxstat & RL_RDESC_STAT_UDP) &&
1899				    !(rxstat & RL_RDESC_STAT_UDPSUMBAD))) {
1900					m->m_pkthdr.csum_flags |=
1901						CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
1902					m->m_pkthdr.csum_data = 0xffff;
1903				}
1904			}
1905		}
1906		maxpkt--;
1907		if (rxvlan & RL_RDESC_VLANCTL_TAG) {
1908			m->m_pkthdr.ether_vtag =
1909			    bswap16((rxvlan & RL_RDESC_VLANCTL_DATA));
1910			m->m_flags |= M_VLANTAG;
1911		}
1912		RL_UNLOCK(sc);
1913		(*ifp->if_input)(ifp, m);
1914		RL_LOCK(sc);
1915	}
1916
1917	/* Flush the RX DMA ring */
1918
1919	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1920	    sc->rl_ldata.rl_rx_list_map,
1921	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
1922
1923	sc->rl_ldata.rl_rx_prodidx = i;
1924
1925	if (maxpkt)
1926		return(EAGAIN);
1927
1928	return(0);
1929}
1930
1931static void
1932re_txeof(struct rl_softc *sc)
1933{
1934	struct ifnet		*ifp;
1935	struct rl_txdesc	*txd;
1936	u_int32_t		txstat;
1937	int			cons;
1938
1939	cons = sc->rl_ldata.rl_tx_considx;
1940	if (cons == sc->rl_ldata.rl_tx_prodidx)
1941		return;
1942
1943	ifp = sc->rl_ifp;
1944	/* Invalidate the TX descriptor list */
1945	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
1946	    sc->rl_ldata.rl_tx_list_map,
1947	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
1948
1949	for (; cons != sc->rl_ldata.rl_tx_prodidx;
1950	    cons = RL_TX_DESC_NXT(sc, cons)) {
1951		txstat = le32toh(sc->rl_ldata.rl_tx_list[cons].rl_cmdstat);
1952		if (txstat & RL_TDESC_STAT_OWN)
1953			break;
1954		/*
1955		 * We only stash mbufs in the last descriptor
1956		 * in a fragment chain, which also happens to
1957		 * be the only place where the TX status bits
1958		 * are valid.
1959		 */
1960		if (txstat & RL_TDESC_CMD_EOF) {
1961			txd = &sc->rl_ldata.rl_tx_desc[cons];
1962			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
1963			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
1964			bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag,
1965			    txd->tx_dmamap);
1966			KASSERT(txd->tx_m != NULL,
1967			    ("%s: freeing NULL mbufs!", __func__));
1968			m_freem(txd->tx_m);
1969			txd->tx_m = NULL;
1970			if (txstat & (RL_TDESC_STAT_EXCESSCOL|
1971			    RL_TDESC_STAT_COLCNT))
1972				ifp->if_collisions++;
1973			if (txstat & RL_TDESC_STAT_TXERRSUM)
1974				ifp->if_oerrors++;
1975			else
1976				ifp->if_opackets++;
1977		}
1978		sc->rl_ldata.rl_tx_free++;
1979		ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1980	}
1981	sc->rl_ldata.rl_tx_considx = cons;
1982
1983	/* No changes made to the TX ring, so no flush needed */
1984
1985	if (sc->rl_ldata.rl_tx_free != sc->rl_ldata.rl_tx_desc_cnt) {
1986		/*
1987		 * Some chips will ignore a second TX request issued
1988		 * while an existing transmission is in progress. If
1989		 * the transmitter goes idle but there are still
1990		 * packets waiting to be sent, we need to restart the
1991		 * channel here to flush them out. This only seems to
1992		 * be required with the PCIe devices.
1993		 */
1994		CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
1995
1996#ifdef RE_TX_MODERATION
1997		/*
1998		 * If not all descriptors have been reaped yet, reload
1999		 * the timer so that we will eventually get another
2000		 * interrupt that will cause us to re-enter this routine.
2001		 * This is done in case the transmitter has gone idle.
2002		 */
2003		CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2004#endif
2005	} else
2006		sc->rl_watchdog_timer = 0;
2007}
2008
2009static void
2010re_tick(void *xsc)
2011{
2012	struct rl_softc		*sc;
2013	struct mii_data		*mii;
2014	struct ifnet		*ifp;
2015
2016	sc = xsc;
2017	ifp = sc->rl_ifp;
2018
2019	RL_LOCK_ASSERT(sc);
2020
2021	re_watchdog(sc);
2022
2023	mii = device_get_softc(sc->rl_miibus);
2024	mii_tick(mii);
2025	if ((sc->rl_flags & RL_FLAG_LINK) != 0) {
2026		if (!(mii->mii_media_status & IFM_ACTIVE))
2027			sc->rl_flags &= ~RL_FLAG_LINK;
2028	} else {
2029		if (mii->mii_media_status & IFM_ACTIVE &&
2030		    IFM_SUBTYPE(mii->mii_media_active) != IFM_NONE) {
2031			sc->rl_flags |= RL_FLAG_LINK;
2032			if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2033				taskqueue_enqueue_fast(taskqueue_fast,
2034				    &sc->rl_txtask);
2035		}
2036	}
2037
2038	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
2039}
2040
2041#ifdef DEVICE_POLLING
2042static void
2043re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
2044{
2045	struct rl_softc *sc = ifp->if_softc;
2046
2047	RL_LOCK(sc);
2048	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
2049		re_poll_locked(ifp, cmd, count);
2050	RL_UNLOCK(sc);
2051}
2052
2053static void
2054re_poll_locked(struct ifnet *ifp, enum poll_cmd cmd, int count)
2055{
2056	struct rl_softc *sc = ifp->if_softc;
2057
2058	RL_LOCK_ASSERT(sc);
2059
2060	sc->rxcycles = count;
2061	re_rxeof(sc);
2062	re_txeof(sc);
2063
2064	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2065		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_txtask);
2066
2067	if (cmd == POLL_AND_CHECK_STATUS) { /* also check status register */
2068		u_int16_t       status;
2069
2070		status = CSR_READ_2(sc, RL_ISR);
2071		if (status == 0xffff)
2072			return;
2073		if (status)
2074			CSR_WRITE_2(sc, RL_ISR, status);
2075
2076		/*
2077		 * XXX check behaviour on receiver stalls.
2078		 */
2079
2080		if (status & RL_ISR_SYSTEM_ERR) {
2081			re_reset(sc);
2082			re_init_locked(sc);
2083		}
2084	}
2085}
2086#endif /* DEVICE_POLLING */
2087
2088static int
2089re_intr(void *arg)
2090{
2091	struct rl_softc		*sc;
2092	uint16_t		status;
2093
2094	sc = arg;
2095
2096	status = CSR_READ_2(sc, RL_ISR);
2097	if (status == 0xFFFF || (status & RL_INTRS_CPLUS) == 0)
2098                return (FILTER_STRAY);
2099	CSR_WRITE_2(sc, RL_IMR, 0);
2100
2101	taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_inttask);
2102
2103	return (FILTER_HANDLED);
2104}
2105
2106static void
2107re_int_task(void *arg, int npending)
2108{
2109	struct rl_softc		*sc;
2110	struct ifnet		*ifp;
2111	u_int16_t		status;
2112	int			rval = 0;
2113
2114	sc = arg;
2115	ifp = sc->rl_ifp;
2116
2117	RL_LOCK(sc);
2118
2119	status = CSR_READ_2(sc, RL_ISR);
2120        CSR_WRITE_2(sc, RL_ISR, status);
2121
2122	if (sc->suspended ||
2123	    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2124		RL_UNLOCK(sc);
2125		return;
2126	}
2127
2128#ifdef DEVICE_POLLING
2129	if  (ifp->if_capenable & IFCAP_POLLING) {
2130		RL_UNLOCK(sc);
2131		return;
2132	}
2133#endif
2134
2135	if (status & (RL_ISR_RX_OK|RL_ISR_RX_ERR|RL_ISR_FIFO_OFLOW))
2136		rval = re_rxeof(sc);
2137
2138	if (status & (
2139#ifdef RE_TX_MODERATION
2140	    RL_ISR_TIMEOUT_EXPIRED|
2141#else
2142	    RL_ISR_TX_OK|
2143#endif
2144	    RL_ISR_TX_ERR|RL_ISR_TX_DESC_UNAVAIL))
2145		re_txeof(sc);
2146
2147	if (status & RL_ISR_SYSTEM_ERR) {
2148		re_reset(sc);
2149		re_init_locked(sc);
2150	}
2151
2152	if (status & RL_ISR_LINKCHG) {
2153		callout_stop(&sc->rl_stat_callout);
2154		re_tick(sc);
2155	}
2156
2157	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2158		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_txtask);
2159
2160	RL_UNLOCK(sc);
2161
2162        if ((CSR_READ_2(sc, RL_ISR) & RL_INTRS_CPLUS) || rval) {
2163		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_inttask);
2164		return;
2165	}
2166
2167	CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2168}
2169
2170static int
2171re_encap(struct rl_softc *sc, struct mbuf **m_head)
2172{
2173	struct rl_txdesc	*txd, *txd_last;
2174	bus_dma_segment_t	segs[RL_NTXSEGS];
2175	bus_dmamap_t		map;
2176	struct mbuf		*m_new;
2177	struct rl_desc		*desc;
2178	int			nsegs, prod;
2179	int			i, error, ei, si;
2180	int			padlen;
2181	uint32_t		cmdstat, csum_flags, vlanctl;
2182
2183	RL_LOCK_ASSERT(sc);
2184	M_ASSERTPKTHDR((*m_head));
2185
2186	/*
2187	 * With some of the RealTek chips, using the checksum offload
2188	 * support in conjunction with the autopadding feature results
2189	 * in the transmission of corrupt frames. For example, if we
2190	 * need to send a really small IP fragment that's less than 60
2191	 * bytes in size, and IP header checksumming is enabled, the
2192	 * resulting ethernet frame that appears on the wire will
2193	 * have garbled payload. To work around this, if TX IP checksum
2194	 * offload is enabled, we always manually pad short frames out
2195	 * to the minimum ethernet frame size.
2196	 */
2197	if ((sc->rl_flags & RL_FLAG_DESCV2) == 0 &&
2198	    (*m_head)->m_pkthdr.len < RL_IP4CSUMTX_PADLEN &&
2199	    ((*m_head)->m_pkthdr.csum_flags & CSUM_IP) != 0) {
2200		padlen = RL_MIN_FRAMELEN - (*m_head)->m_pkthdr.len;
2201		if (M_WRITABLE(*m_head) == 0) {
2202			/* Get a writable copy. */
2203			m_new = m_dup(*m_head, M_DONTWAIT);
2204			m_freem(*m_head);
2205			if (m_new == NULL) {
2206				*m_head = NULL;
2207				return (ENOBUFS);
2208			}
2209			*m_head = m_new;
2210		}
2211		if ((*m_head)->m_next != NULL ||
2212		    M_TRAILINGSPACE(*m_head) < padlen) {
2213			m_new = m_defrag(*m_head, M_DONTWAIT);
2214			if (m_new == NULL) {
2215				m_freem(*m_head);
2216				*m_head = NULL;
2217				return (ENOBUFS);
2218			}
2219		} else
2220			m_new = *m_head;
2221
2222		/*
2223		 * Manually pad short frames, and zero the pad space
2224		 * to avoid leaking data.
2225		 */
2226		bzero(mtod(m_new, char *) + m_new->m_pkthdr.len, padlen);
2227		m_new->m_pkthdr.len += padlen;
2228		m_new->m_len = m_new->m_pkthdr.len;
2229		*m_head = m_new;
2230	}
2231
2232	prod = sc->rl_ldata.rl_tx_prodidx;
2233	txd = &sc->rl_ldata.rl_tx_desc[prod];
2234	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap,
2235	    *m_head, segs, &nsegs, BUS_DMA_NOWAIT);
2236	if (error == EFBIG) {
2237		m_new = m_collapse(*m_head, M_DONTWAIT, RL_NTXSEGS);
2238		if (m_new == NULL) {
2239			m_freem(*m_head);
2240			*m_head = NULL;
2241			return (ENOBUFS);
2242		}
2243		*m_head = m_new;
2244		error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_tx_mtag,
2245		    txd->tx_dmamap, *m_head, segs, &nsegs, BUS_DMA_NOWAIT);
2246		if (error != 0) {
2247			m_freem(*m_head);
2248			*m_head = NULL;
2249			return (error);
2250		}
2251	} else if (error != 0)
2252		return (error);
2253	if (nsegs == 0) {
2254		m_freem(*m_head);
2255		*m_head = NULL;
2256		return (EIO);
2257	}
2258
2259	/* Check for number of available descriptors. */
2260	if (sc->rl_ldata.rl_tx_free - nsegs <= 1) {
2261		bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap);
2262		return (ENOBUFS);
2263	}
2264
2265	bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap,
2266	    BUS_DMASYNC_PREWRITE);
2267
2268	/*
2269	 * Set up checksum offload. Note: checksum offload bits must
2270	 * appear in all descriptors of a multi-descriptor transmit
2271	 * attempt. This is according to testing done with an 8169
2272	 * chip. This is a requirement.
2273	 */
2274	vlanctl = 0;
2275	csum_flags = 0;
2276	if (((*m_head)->m_pkthdr.csum_flags & CSUM_TSO) != 0)
2277		csum_flags = RL_TDESC_CMD_LGSEND |
2278		    ((uint32_t)(*m_head)->m_pkthdr.tso_segsz <<
2279		    RL_TDESC_CMD_MSSVAL_SHIFT);
2280	else {
2281		/*
2282		 * Unconditionally enable IP checksum if TCP or UDP
2283		 * checksum is required. Otherwise, TCP/UDP checksum
2284		 * does't make effects.
2285		 */
2286		if (((*m_head)->m_pkthdr.csum_flags & RE_CSUM_FEATURES) != 0) {
2287			if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
2288				csum_flags |= RL_TDESC_CMD_IPCSUM;
2289				if (((*m_head)->m_pkthdr.csum_flags &
2290				    CSUM_TCP) != 0)
2291					csum_flags |= RL_TDESC_CMD_TCPCSUM;
2292				if (((*m_head)->m_pkthdr.csum_flags &
2293				    CSUM_UDP) != 0)
2294					csum_flags |= RL_TDESC_CMD_UDPCSUM;
2295			} else {
2296				vlanctl |= RL_TDESC_CMD_IPCSUMV2;
2297				if (((*m_head)->m_pkthdr.csum_flags &
2298				    CSUM_TCP) != 0)
2299					vlanctl |= RL_TDESC_CMD_TCPCSUMV2;
2300				if (((*m_head)->m_pkthdr.csum_flags &
2301				    CSUM_UDP) != 0)
2302					vlanctl |= RL_TDESC_CMD_UDPCSUMV2;
2303			}
2304		}
2305	}
2306
2307	/*
2308	 * Set up hardware VLAN tagging. Note: vlan tag info must
2309	 * appear in all descriptors of a multi-descriptor
2310	 * transmission attempt.
2311	 */
2312	if ((*m_head)->m_flags & M_VLANTAG)
2313		vlanctl |= bswap16((*m_head)->m_pkthdr.ether_vtag) |
2314		    RL_TDESC_VLANCTL_TAG;
2315
2316	si = prod;
2317	for (i = 0; i < nsegs; i++, prod = RL_TX_DESC_NXT(sc, prod)) {
2318		desc = &sc->rl_ldata.rl_tx_list[prod];
2319		desc->rl_vlanctl = htole32(vlanctl);
2320		desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[i].ds_addr));
2321		desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[i].ds_addr));
2322		cmdstat = segs[i].ds_len;
2323		if (i != 0)
2324			cmdstat |= RL_TDESC_CMD_OWN;
2325		if (prod == sc->rl_ldata.rl_tx_desc_cnt - 1)
2326			cmdstat |= RL_TDESC_CMD_EOR;
2327		desc->rl_cmdstat = htole32(cmdstat | csum_flags);
2328		sc->rl_ldata.rl_tx_free--;
2329	}
2330	/* Update producer index. */
2331	sc->rl_ldata.rl_tx_prodidx = prod;
2332
2333	/* Set EOF on the last descriptor. */
2334	ei = RL_TX_DESC_PRV(sc, prod);
2335	desc = &sc->rl_ldata.rl_tx_list[ei];
2336	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_EOF);
2337
2338	desc = &sc->rl_ldata.rl_tx_list[si];
2339	/* Set SOF and transfer ownership of packet to the chip. */
2340	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_OWN | RL_TDESC_CMD_SOF);
2341
2342	/*
2343	 * Insure that the map for this transmission
2344	 * is placed at the array index of the last descriptor
2345	 * in this chain.  (Swap last and first dmamaps.)
2346	 */
2347	txd_last = &sc->rl_ldata.rl_tx_desc[ei];
2348	map = txd->tx_dmamap;
2349	txd->tx_dmamap = txd_last->tx_dmamap;
2350	txd_last->tx_dmamap = map;
2351	txd_last->tx_m = *m_head;
2352
2353	return (0);
2354}
2355
2356static void
2357re_tx_task(void *arg, int npending)
2358{
2359	struct ifnet		*ifp;
2360
2361	ifp = arg;
2362	re_start(ifp);
2363}
2364
2365/*
2366 * Main transmit routine for C+ and gigE NICs.
2367 */
2368static void
2369re_start(struct ifnet *ifp)
2370{
2371	struct rl_softc		*sc;
2372	struct mbuf		*m_head;
2373	int			queued;
2374
2375	sc = ifp->if_softc;
2376
2377	RL_LOCK(sc);
2378
2379	if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
2380	    IFF_DRV_RUNNING || (sc->rl_flags & RL_FLAG_LINK) == 0) {
2381		RL_UNLOCK(sc);
2382		return;
2383	}
2384
2385	for (queued = 0; !IFQ_DRV_IS_EMPTY(&ifp->if_snd) &&
2386	    sc->rl_ldata.rl_tx_free > 1;) {
2387		IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
2388		if (m_head == NULL)
2389			break;
2390
2391		if (re_encap(sc, &m_head) != 0) {
2392			if (m_head == NULL)
2393				break;
2394			IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
2395			ifp->if_drv_flags |= IFF_DRV_OACTIVE;
2396			break;
2397		}
2398
2399		/*
2400		 * If there's a BPF listener, bounce a copy of this frame
2401		 * to him.
2402		 */
2403		ETHER_BPF_MTAP(ifp, m_head);
2404
2405		queued++;
2406	}
2407
2408	if (queued == 0) {
2409#ifdef RE_TX_MODERATION
2410		if (sc->rl_ldata.rl_tx_free != sc->rl_ldata.rl_tx_desc_cnt)
2411			CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2412#endif
2413		RL_UNLOCK(sc);
2414		return;
2415	}
2416
2417	/* Flush the TX descriptors */
2418
2419	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2420	    sc->rl_ldata.rl_tx_list_map,
2421	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2422
2423	CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2424
2425#ifdef RE_TX_MODERATION
2426	/*
2427	 * Use the countdown timer for interrupt moderation.
2428	 * 'TX done' interrupts are disabled. Instead, we reset the
2429	 * countdown timer, which will begin counting until it hits
2430	 * the value in the TIMERINT register, and then trigger an
2431	 * interrupt. Each time we write to the TIMERCNT register,
2432	 * the timer count is reset to 0.
2433	 */
2434	CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2435#endif
2436
2437	/*
2438	 * Set a timeout in case the chip goes out to lunch.
2439	 */
2440	sc->rl_watchdog_timer = 5;
2441
2442	RL_UNLOCK(sc);
2443}
2444
2445static void
2446re_init(void *xsc)
2447{
2448	struct rl_softc		*sc = xsc;
2449
2450	RL_LOCK(sc);
2451	re_init_locked(sc);
2452	RL_UNLOCK(sc);
2453}
2454
2455static void
2456re_init_locked(struct rl_softc *sc)
2457{
2458	struct ifnet		*ifp = sc->rl_ifp;
2459	struct mii_data		*mii;
2460	u_int32_t		rxcfg = 0;
2461	uint16_t		cfg;
2462	union {
2463		uint32_t align_dummy;
2464		u_char eaddr[ETHER_ADDR_LEN];
2465        } eaddr;
2466
2467	RL_LOCK_ASSERT(sc);
2468
2469	mii = device_get_softc(sc->rl_miibus);
2470
2471	/*
2472	 * Cancel pending I/O and free all RX/TX buffers.
2473	 */
2474	re_stop(sc);
2475
2476	/*
2477	 * Enable C+ RX and TX mode, as well as VLAN stripping and
2478	 * RX checksum offload. We must configure the C+ register
2479	 * before all others.
2480	 */
2481	cfg = RL_CPLUSCMD_PCI_MRW;
2482	if ((ifp->if_capenable & IFCAP_RXCSUM) != 0)
2483		cfg |= RL_CPLUSCMD_RXCSUM_ENB;
2484	if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0)
2485		cfg |= RL_CPLUSCMD_VLANSTRIP;
2486	if ((sc->rl_flags & RL_FLAG_MACSTAT) != 0) {
2487		cfg |= RL_CPLUSCMD_MACSTAT_DIS;
2488		/* XXX magic. */
2489		cfg |= 0x0001;
2490	} else
2491		cfg |= RL_CPLUSCMD_RXENB | RL_CPLUSCMD_TXENB;
2492	CSR_WRITE_2(sc, RL_CPLUS_CMD, cfg);
2493	/*
2494	 * Disable TSO if interface MTU size is greater than MSS
2495	 * allowed in controller.
2496	 */
2497	if (ifp->if_mtu > RL_TSO_MTU && (ifp->if_capenable & IFCAP_TSO4) != 0) {
2498		ifp->if_capenable &= ~IFCAP_TSO4;
2499		ifp->if_hwassist &= ~CSUM_TSO;
2500	}
2501
2502	/*
2503	 * Init our MAC address.  Even though the chipset
2504	 * documentation doesn't mention it, we need to enter "Config
2505	 * register write enable" mode to modify the ID registers.
2506	 */
2507	/* Copy MAC address on stack to align. */
2508	bcopy(IF_LLADDR(ifp), eaddr.eaddr, ETHER_ADDR_LEN);
2509	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG);
2510	CSR_WRITE_4(sc, RL_IDR0,
2511	    htole32(*(u_int32_t *)(&eaddr.eaddr[0])));
2512	CSR_WRITE_4(sc, RL_IDR4,
2513	    htole32(*(u_int32_t *)(&eaddr.eaddr[4])));
2514	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
2515
2516	/*
2517	 * For C+ mode, initialize the RX descriptors and mbufs.
2518	 */
2519	re_rx_list_init(sc);
2520	re_tx_list_init(sc);
2521
2522	/*
2523	 * Load the addresses of the RX and TX lists into the chip.
2524	 */
2525
2526	CSR_WRITE_4(sc, RL_RXLIST_ADDR_HI,
2527	    RL_ADDR_HI(sc->rl_ldata.rl_rx_list_addr));
2528	CSR_WRITE_4(sc, RL_RXLIST_ADDR_LO,
2529	    RL_ADDR_LO(sc->rl_ldata.rl_rx_list_addr));
2530
2531	CSR_WRITE_4(sc, RL_TXLIST_ADDR_HI,
2532	    RL_ADDR_HI(sc->rl_ldata.rl_tx_list_addr));
2533	CSR_WRITE_4(sc, RL_TXLIST_ADDR_LO,
2534	    RL_ADDR_LO(sc->rl_ldata.rl_tx_list_addr));
2535
2536	/*
2537	 * Enable transmit and receive.
2538	 */
2539	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
2540
2541	/*
2542	 * Set the initial TX and RX configuration.
2543	 */
2544	if (sc->rl_testmode) {
2545		if (sc->rl_type == RL_8169)
2546			CSR_WRITE_4(sc, RL_TXCFG,
2547			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON);
2548		else
2549			CSR_WRITE_4(sc, RL_TXCFG,
2550			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON_CPLUS);
2551	} else
2552		CSR_WRITE_4(sc, RL_TXCFG, RL_TXCFG_CONFIG);
2553
2554	CSR_WRITE_1(sc, RL_EARLY_TX_THRESH, 16);
2555
2556	CSR_WRITE_4(sc, RL_RXCFG, RL_RXCFG_CONFIG);
2557
2558	/* Set the individual bit to receive frames for this host only. */
2559	rxcfg = CSR_READ_4(sc, RL_RXCFG);
2560	rxcfg |= RL_RXCFG_RX_INDIV;
2561
2562	/* If we want promiscuous mode, set the allframes bit. */
2563	if (ifp->if_flags & IFF_PROMISC)
2564		rxcfg |= RL_RXCFG_RX_ALLPHYS;
2565	else
2566		rxcfg &= ~RL_RXCFG_RX_ALLPHYS;
2567	CSR_WRITE_4(sc, RL_RXCFG, rxcfg);
2568
2569	/*
2570	 * Set capture broadcast bit to capture broadcast frames.
2571	 */
2572	if (ifp->if_flags & IFF_BROADCAST)
2573		rxcfg |= RL_RXCFG_RX_BROAD;
2574	else
2575		rxcfg &= ~RL_RXCFG_RX_BROAD;
2576	CSR_WRITE_4(sc, RL_RXCFG, rxcfg);
2577
2578	/*
2579	 * Program the multicast filter, if necessary.
2580	 */
2581	re_setmulti(sc);
2582
2583#ifdef DEVICE_POLLING
2584	/*
2585	 * Disable interrupts if we are polling.
2586	 */
2587	if (ifp->if_capenable & IFCAP_POLLING)
2588		CSR_WRITE_2(sc, RL_IMR, 0);
2589	else	/* otherwise ... */
2590#endif
2591
2592	/*
2593	 * Enable interrupts.
2594	 */
2595	if (sc->rl_testmode)
2596		CSR_WRITE_2(sc, RL_IMR, 0);
2597	else
2598		CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2599	CSR_WRITE_2(sc, RL_ISR, RL_INTRS_CPLUS);
2600
2601	/* Set initial TX threshold */
2602	sc->rl_txthresh = RL_TX_THRESH_INIT;
2603
2604	/* Start RX/TX process. */
2605	CSR_WRITE_4(sc, RL_MISSEDPKT, 0);
2606#ifdef notdef
2607	/* Enable receiver and transmitter. */
2608	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
2609#endif
2610
2611#ifdef RE_TX_MODERATION
2612	/*
2613	 * Initialize the timer interrupt register so that
2614	 * a timer interrupt will be generated once the timer
2615	 * reaches a certain number of ticks. The timer is
2616	 * reloaded on each transmit. This gives us TX interrupt
2617	 * moderation, which dramatically improves TX frame rate.
2618	 */
2619	if (sc->rl_type == RL_8169)
2620		CSR_WRITE_4(sc, RL_TIMERINT_8169, 0x800);
2621	else
2622		CSR_WRITE_4(sc, RL_TIMERINT, 0x400);
2623#endif
2624
2625	/*
2626	 * For 8169 gigE NICs, set the max allowed RX packet
2627	 * size so we can receive jumbo frames.
2628	 */
2629	if (sc->rl_type == RL_8169)
2630		CSR_WRITE_2(sc, RL_MAXRXPKTLEN, 16383);
2631
2632	if (sc->rl_testmode)
2633		return;
2634
2635	mii_mediachg(mii);
2636
2637	CSR_WRITE_1(sc, RL_CFG1, CSR_READ_1(sc, RL_CFG1) | RL_CFG1_DRVLOAD);
2638
2639	ifp->if_drv_flags |= IFF_DRV_RUNNING;
2640	ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
2641
2642	sc->rl_flags &= ~RL_FLAG_LINK;
2643	sc->rl_watchdog_timer = 0;
2644	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
2645}
2646
2647/*
2648 * Set media options.
2649 */
2650static int
2651re_ifmedia_upd(struct ifnet *ifp)
2652{
2653	struct rl_softc		*sc;
2654	struct mii_data		*mii;
2655	int			error;
2656
2657	sc = ifp->if_softc;
2658	mii = device_get_softc(sc->rl_miibus);
2659	RL_LOCK(sc);
2660	error = mii_mediachg(mii);
2661	RL_UNLOCK(sc);
2662
2663	return (error);
2664}
2665
2666/*
2667 * Report current media status.
2668 */
2669static void
2670re_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
2671{
2672	struct rl_softc		*sc;
2673	struct mii_data		*mii;
2674
2675	sc = ifp->if_softc;
2676	mii = device_get_softc(sc->rl_miibus);
2677
2678	RL_LOCK(sc);
2679	mii_pollstat(mii);
2680	RL_UNLOCK(sc);
2681	ifmr->ifm_active = mii->mii_media_active;
2682	ifmr->ifm_status = mii->mii_media_status;
2683}
2684
2685static int
2686re_ioctl(struct ifnet *ifp, u_long command, caddr_t data)
2687{
2688	struct rl_softc		*sc = ifp->if_softc;
2689	struct ifreq		*ifr = (struct ifreq *) data;
2690	struct mii_data		*mii;
2691	int			error = 0;
2692
2693	switch (command) {
2694	case SIOCSIFMTU:
2695		if (ifr->ifr_mtu < ETHERMIN || ifr->ifr_mtu > RL_JUMBO_MTU) {
2696			error = EINVAL;
2697			break;
2698		}
2699		if ((sc->rl_flags & RL_FLAG_NOJUMBO) != 0 &&
2700		    ifr->ifr_mtu > RL_MAX_FRAMELEN) {
2701			error = EINVAL;
2702			break;
2703		}
2704		RL_LOCK(sc);
2705		if (ifp->if_mtu != ifr->ifr_mtu)
2706			ifp->if_mtu = ifr->ifr_mtu;
2707		if (ifp->if_mtu > RL_TSO_MTU &&
2708		    (ifp->if_capenable & IFCAP_TSO4) != 0) {
2709			ifp->if_capenable &= ~IFCAP_TSO4;
2710			ifp->if_hwassist &= ~CSUM_TSO;
2711		}
2712		RL_UNLOCK(sc);
2713		break;
2714	case SIOCSIFFLAGS:
2715		RL_LOCK(sc);
2716		if ((ifp->if_flags & IFF_UP) != 0) {
2717			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
2718				if (((ifp->if_flags ^ sc->rl_if_flags)
2719				    & (IFF_PROMISC | IFF_ALLMULTI)) != 0)
2720					re_setmulti(sc);
2721			} else
2722				re_init_locked(sc);
2723		} else {
2724			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
2725				re_stop(sc);
2726		}
2727		sc->rl_if_flags = ifp->if_flags;
2728		RL_UNLOCK(sc);
2729		break;
2730	case SIOCADDMULTI:
2731	case SIOCDELMULTI:
2732		RL_LOCK(sc);
2733		re_setmulti(sc);
2734		RL_UNLOCK(sc);
2735		break;
2736	case SIOCGIFMEDIA:
2737	case SIOCSIFMEDIA:
2738		mii = device_get_softc(sc->rl_miibus);
2739		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, command);
2740		break;
2741	case SIOCSIFCAP:
2742	    {
2743		int mask, reinit;
2744
2745		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
2746		reinit = 0;
2747#ifdef DEVICE_POLLING
2748		if (mask & IFCAP_POLLING) {
2749			if (ifr->ifr_reqcap & IFCAP_POLLING) {
2750				error = ether_poll_register(re_poll, ifp);
2751				if (error)
2752					return(error);
2753				RL_LOCK(sc);
2754				/* Disable interrupts */
2755				CSR_WRITE_2(sc, RL_IMR, 0x0000);
2756				ifp->if_capenable |= IFCAP_POLLING;
2757				RL_UNLOCK(sc);
2758			} else {
2759				error = ether_poll_deregister(ifp);
2760				/* Enable interrupts. */
2761				RL_LOCK(sc);
2762				CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2763				ifp->if_capenable &= ~IFCAP_POLLING;
2764				RL_UNLOCK(sc);
2765			}
2766		}
2767#endif /* DEVICE_POLLING */
2768		if (mask & IFCAP_HWCSUM) {
2769			ifp->if_capenable ^= IFCAP_HWCSUM;
2770			if (ifp->if_capenable & IFCAP_TXCSUM)
2771				ifp->if_hwassist |= RE_CSUM_FEATURES;
2772			else
2773				ifp->if_hwassist &= ~RE_CSUM_FEATURES;
2774			reinit = 1;
2775		}
2776		if (mask & IFCAP_VLAN_HWTAGGING) {
2777			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
2778			reinit = 1;
2779		}
2780		if (mask & IFCAP_TSO4) {
2781			ifp->if_capenable ^= IFCAP_TSO4;
2782			if ((IFCAP_TSO4 & ifp->if_capenable) &&
2783			    (IFCAP_TSO4 & ifp->if_capabilities))
2784				ifp->if_hwassist |= CSUM_TSO;
2785			else
2786				ifp->if_hwassist &= ~CSUM_TSO;
2787			if (ifp->if_mtu > RL_TSO_MTU &&
2788			    (ifp->if_capenable & IFCAP_TSO4) != 0) {
2789				ifp->if_capenable &= ~IFCAP_TSO4;
2790				ifp->if_hwassist &= ~CSUM_TSO;
2791			}
2792		}
2793		if ((mask & IFCAP_WOL) != 0 &&
2794		    (ifp->if_capabilities & IFCAP_WOL) != 0) {
2795			if ((mask & IFCAP_WOL_UCAST) != 0)
2796				ifp->if_capenable ^= IFCAP_WOL_UCAST;
2797			if ((mask & IFCAP_WOL_MCAST) != 0)
2798				ifp->if_capenable ^= IFCAP_WOL_MCAST;
2799			if ((mask & IFCAP_WOL_MAGIC) != 0)
2800				ifp->if_capenable ^= IFCAP_WOL_MAGIC;
2801		}
2802		if (reinit && ifp->if_drv_flags & IFF_DRV_RUNNING)
2803			re_init(sc);
2804		VLAN_CAPABILITIES(ifp);
2805	    }
2806		break;
2807	default:
2808		error = ether_ioctl(ifp, command, data);
2809		break;
2810	}
2811
2812	return (error);
2813}
2814
2815static void
2816re_watchdog(struct rl_softc *sc)
2817{
2818
2819	RL_LOCK_ASSERT(sc);
2820
2821	if (sc->rl_watchdog_timer == 0 || --sc->rl_watchdog_timer != 0)
2822		return;
2823
2824	device_printf(sc->rl_dev, "watchdog timeout\n");
2825	sc->rl_ifp->if_oerrors++;
2826
2827	re_txeof(sc);
2828	re_rxeof(sc);
2829	re_init_locked(sc);
2830}
2831
2832/*
2833 * Stop the adapter and free any mbufs allocated to the
2834 * RX and TX lists.
2835 */
2836static void
2837re_stop(struct rl_softc *sc)
2838{
2839	int			i;
2840	struct ifnet		*ifp;
2841	struct rl_txdesc	*txd;
2842	struct rl_rxdesc	*rxd;
2843
2844	RL_LOCK_ASSERT(sc);
2845
2846	ifp = sc->rl_ifp;
2847
2848	sc->rl_watchdog_timer = 0;
2849	callout_stop(&sc->rl_stat_callout);
2850	ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
2851
2852	CSR_WRITE_1(sc, RL_COMMAND, 0x00);
2853	CSR_WRITE_2(sc, RL_IMR, 0x0000);
2854	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
2855
2856	if (sc->rl_head != NULL) {
2857		m_freem(sc->rl_head);
2858		sc->rl_head = sc->rl_tail = NULL;
2859	}
2860
2861	/* Free the TX list buffers. */
2862
2863	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
2864		txd = &sc->rl_ldata.rl_tx_desc[i];
2865		if (txd->tx_m != NULL) {
2866			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
2867			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
2868			bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag,
2869			    txd->tx_dmamap);
2870			m_freem(txd->tx_m);
2871			txd->tx_m = NULL;
2872		}
2873	}
2874
2875	/* Free the RX list buffers. */
2876
2877	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
2878		rxd = &sc->rl_ldata.rl_rx_desc[i];
2879		if (rxd->rx_m != NULL) {
2880			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
2881			    rxd->rx_dmamap, BUS_DMASYNC_POSTREAD);
2882			bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag,
2883			    rxd->rx_dmamap);
2884			m_freem(rxd->rx_m);
2885			rxd->rx_m = NULL;
2886		}
2887	}
2888}
2889
2890/*
2891 * Device suspend routine.  Stop the interface and save some PCI
2892 * settings in case the BIOS doesn't restore them properly on
2893 * resume.
2894 */
2895static int
2896re_suspend(device_t dev)
2897{
2898	struct rl_softc		*sc;
2899
2900	sc = device_get_softc(dev);
2901
2902	RL_LOCK(sc);
2903	re_stop(sc);
2904	re_setwol(sc);
2905	sc->suspended = 1;
2906	RL_UNLOCK(sc);
2907
2908	return (0);
2909}
2910
2911/*
2912 * Device resume routine.  Restore some PCI settings in case the BIOS
2913 * doesn't, re-enable busmastering, and restart the interface if
2914 * appropriate.
2915 */
2916static int
2917re_resume(device_t dev)
2918{
2919	struct rl_softc		*sc;
2920	struct ifnet		*ifp;
2921
2922	sc = device_get_softc(dev);
2923
2924	RL_LOCK(sc);
2925
2926	ifp = sc->rl_ifp;
2927
2928	/* reinitialize interface if necessary */
2929	if (ifp->if_flags & IFF_UP)
2930		re_init_locked(sc);
2931
2932	/*
2933	 * Clear WOL matching such that normal Rx filtering
2934	 * wouldn't interfere with WOL patterns.
2935	 */
2936	re_clrwol(sc);
2937	sc->suspended = 0;
2938	RL_UNLOCK(sc);
2939
2940	return (0);
2941}
2942
2943/*
2944 * Stop all chip I/O so that the kernel's probe routines don't
2945 * get confused by errant DMAs when rebooting.
2946 */
2947static int
2948re_shutdown(device_t dev)
2949{
2950	struct rl_softc		*sc;
2951
2952	sc = device_get_softc(dev);
2953
2954	RL_LOCK(sc);
2955	re_stop(sc);
2956	/*
2957	 * Mark interface as down since otherwise we will panic if
2958	 * interrupt comes in later on, which can happen in some
2959	 * cases.
2960	 */
2961	sc->rl_ifp->if_flags &= ~IFF_UP;
2962	re_setwol(sc);
2963	RL_UNLOCK(sc);
2964
2965	return (0);
2966}
2967
2968static void
2969re_setwol(struct rl_softc *sc)
2970{
2971	struct ifnet		*ifp;
2972	int			pmc;
2973	uint16_t		pmstat;
2974	uint8_t			v;
2975
2976	RL_LOCK_ASSERT(sc);
2977
2978	if (pci_find_extcap(sc->rl_dev, PCIY_PMG, &pmc) != 0)
2979		return;
2980
2981	ifp = sc->rl_ifp;
2982	/* Enable config register write. */
2983	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
2984
2985	/* Enable PME. */
2986	v = CSR_READ_1(sc, RL_CFG1);
2987	v &= ~RL_CFG1_PME;
2988	if ((ifp->if_capenable & IFCAP_WOL) != 0)
2989		v |= RL_CFG1_PME;
2990	CSR_WRITE_1(sc, RL_CFG1, v);
2991
2992	v = CSR_READ_1(sc, RL_CFG3);
2993	v &= ~(RL_CFG3_WOL_LINK | RL_CFG3_WOL_MAGIC);
2994	if ((ifp->if_capenable & IFCAP_WOL_MAGIC) != 0)
2995		v |= RL_CFG3_WOL_MAGIC;
2996	CSR_WRITE_1(sc, RL_CFG3, v);
2997
2998	/* Config register write done. */
2999	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3000
3001	v = CSR_READ_1(sc, RL_CFG5);
3002	v &= ~(RL_CFG5_WOL_BCAST | RL_CFG5_WOL_MCAST | RL_CFG5_WOL_UCAST);
3003	v &= ~RL_CFG5_WOL_LANWAKE;
3004	if ((ifp->if_capenable & IFCAP_WOL_UCAST) != 0)
3005		v |= RL_CFG5_WOL_UCAST;
3006	if ((ifp->if_capenable & IFCAP_WOL_MCAST) != 0)
3007		v |= RL_CFG5_WOL_MCAST | RL_CFG5_WOL_BCAST;
3008	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3009		v |= RL_CFG5_WOL_LANWAKE;
3010	CSR_WRITE_1(sc, RL_CFG5, v);
3011
3012	/*
3013	 * It seems that hardware resets its link speed to 100Mbps in
3014	 * power down mode so switching to 100Mbps in driver is not
3015	 * needed.
3016	 */
3017
3018	/* Request PME if WOL is requested. */
3019	pmstat = pci_read_config(sc->rl_dev, pmc + PCIR_POWER_STATUS, 2);
3020	pmstat &= ~(PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE);
3021	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3022		pmstat |= PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE;
3023	pci_write_config(sc->rl_dev, pmc + PCIR_POWER_STATUS, pmstat, 2);
3024}
3025
3026static void
3027re_clrwol(struct rl_softc *sc)
3028{
3029	int			pmc;
3030	uint8_t			v;
3031
3032	RL_LOCK_ASSERT(sc);
3033
3034	if (pci_find_extcap(sc->rl_dev, PCIY_PMG, &pmc) != 0)
3035		return;
3036
3037	/* Enable config register write. */
3038	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
3039
3040	v = CSR_READ_1(sc, RL_CFG3);
3041	v &= ~(RL_CFG3_WOL_LINK | RL_CFG3_WOL_MAGIC);
3042	CSR_WRITE_1(sc, RL_CFG3, v);
3043
3044	/* Config register write done. */
3045	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3046
3047	v = CSR_READ_1(sc, RL_CFG5);
3048	v &= ~(RL_CFG5_WOL_BCAST | RL_CFG5_WOL_MCAST | RL_CFG5_WOL_UCAST);
3049	v &= ~RL_CFG5_WOL_LANWAKE;
3050	CSR_WRITE_1(sc, RL_CFG5, v);
3051}
3052