if_re.c revision 143481
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 143481 2005-03-13 01:54:41Z mlaier $");
35
36/*
37 * RealTek 8139C+/8169/8169S/8110S 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 * four devices in this family: the RTL8139C+, the RTL8169, the RTL8169S
48 * and the RTL8110S.
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#include <sys/param.h>
112#include <sys/endian.h>
113#include <sys/systm.h>
114#include <sys/sockio.h>
115#include <sys/mbuf.h>
116#include <sys/malloc.h>
117#include <sys/module.h>
118#include <sys/kernel.h>
119#include <sys/socket.h>
120
121#include <net/if.h>
122#include <net/if_arp.h>
123#include <net/ethernet.h>
124#include <net/if_dl.h>
125#include <net/if_media.h>
126#include <net/if_vlan_var.h>
127
128#include <net/bpf.h>
129
130#include <machine/bus_pio.h>
131#include <machine/bus_memio.h>
132#include <machine/bus.h>
133#include <machine/resource.h>
134#include <sys/bus.h>
135#include <sys/rman.h>
136
137#include <dev/mii/mii.h>
138#include <dev/mii/miivar.h>
139
140#include <dev/pci/pcireg.h>
141#include <dev/pci/pcivar.h>
142
143MODULE_DEPEND(re, pci, 1, 1, 1);
144MODULE_DEPEND(re, ether, 1, 1, 1);
145MODULE_DEPEND(re, miibus, 1, 1, 1);
146
147/* "controller miibus0" required.  See GENERIC if you get errors here. */
148#include "miibus_if.h"
149
150/*
151 * Default to using PIO access for this driver.
152 */
153#define RE_USEIOSPACE
154
155#include <pci/if_rlreg.h>
156
157#define RE_CSUM_FEATURES    (CSUM_IP | CSUM_TCP | CSUM_UDP)
158
159/*
160 * Various supported device vendors/types and their names.
161 */
162static struct rl_type re_devs[] = {
163	{ RT_VENDORID, RT_DEVICEID_8139, RL_HWREV_8139CPLUS,
164		"RealTek 8139C+ 10/100BaseTX" },
165	{ RT_VENDORID, RT_DEVICEID_8169, RL_HWREV_8169,
166		"RealTek 8169 Gigabit Ethernet" },
167	{ RT_VENDORID, RT_DEVICEID_8169, RL_HWREV_8169S,
168		"RealTek 8169S Single-chip Gigabit Ethernet" },
169	{ RT_VENDORID, RT_DEVICEID_8169, RL_HWREV_8169SB,
170		"RealTek 8169SB Single-chip Gigabit Ethernet" },
171	{ RT_VENDORID, RT_DEVICEID_8169, RL_HWREV_8110S,
172		"RealTek 8110S Single-chip Gigabit Ethernet" },
173	{ COREGA_VENDORID, COREGA_DEVICEID_CGLAPCIGT, RL_HWREV_8169S,
174		"Corega CG-LAPCIGT (RTL8169S) Gigabit Ethernet" },
175	{ 0, 0, 0, NULL }
176};
177
178static struct rl_hwrev re_hwrevs[] = {
179	{ RL_HWREV_8139, RL_8139,  "" },
180	{ RL_HWREV_8139A, RL_8139, "A" },
181	{ RL_HWREV_8139AG, RL_8139, "A-G" },
182	{ RL_HWREV_8139B, RL_8139, "B" },
183	{ RL_HWREV_8130, RL_8139, "8130" },
184	{ RL_HWREV_8139C, RL_8139, "C" },
185	{ RL_HWREV_8139D, RL_8139, "8139D/8100B/8100C" },
186	{ RL_HWREV_8139CPLUS, RL_8139CPLUS, "C+"},
187	{ RL_HWREV_8169, RL_8169, "8169"},
188	{ RL_HWREV_8169S, RL_8169, "8169S"},
189	{ RL_HWREV_8169SB, RL_8169, "8169SB"},
190	{ RL_HWREV_8110S, RL_8169, "8110S"},
191	{ RL_HWREV_8100, RL_8139, "8100"},
192	{ RL_HWREV_8101, RL_8139, "8101"},
193	{ 0, 0, NULL }
194};
195
196static int re_probe		(device_t);
197static int re_attach		(device_t);
198static int re_detach		(device_t);
199
200static int re_encap		(struct rl_softc *, struct mbuf **, int *);
201
202static void re_dma_map_addr	(void *, bus_dma_segment_t *, int, int);
203static void re_dma_map_desc	(void *, bus_dma_segment_t *, int,
204				    bus_size_t, int);
205static int re_allocmem		(device_t, struct rl_softc *);
206static int re_newbuf		(struct rl_softc *, int, struct mbuf *);
207static int re_rx_list_init	(struct rl_softc *);
208static int re_tx_list_init	(struct rl_softc *);
209#ifdef RE_FIXUP_RX
210static __inline void re_fixup_rx
211				(struct mbuf *);
212#endif
213static void re_rxeof		(struct rl_softc *);
214static void re_txeof		(struct rl_softc *);
215#ifdef DEVICE_POLLING
216static void re_poll		(struct ifnet *, enum poll_cmd, int);
217static void re_poll_locked	(struct ifnet *, enum poll_cmd, int);
218#endif
219static void re_intr		(void *);
220static void re_tick		(void *);
221static void re_tick_locked	(struct rl_softc *);
222static void re_start		(struct ifnet *);
223static void re_start_locked	(struct ifnet *);
224static int re_ioctl		(struct ifnet *, u_long, caddr_t);
225static void re_init		(void *);
226static void re_init_locked	(struct rl_softc *);
227static void re_stop		(struct rl_softc *);
228static void re_watchdog		(struct ifnet *);
229static int re_suspend		(device_t);
230static int re_resume		(device_t);
231static void re_shutdown		(device_t);
232static int re_ifmedia_upd	(struct ifnet *);
233static void re_ifmedia_sts	(struct ifnet *, struct ifmediareq *);
234
235static void re_eeprom_putbyte	(struct rl_softc *, int);
236static void re_eeprom_getword	(struct rl_softc *, int, u_int16_t *);
237static void re_read_eeprom	(struct rl_softc *, caddr_t, int, int, int);
238static int re_gmii_readreg	(device_t, int, int);
239static int re_gmii_writereg	(device_t, int, int, int);
240
241static int re_miibus_readreg	(device_t, int, int);
242static int re_miibus_writereg	(device_t, int, int, int);
243static void re_miibus_statchg	(device_t);
244
245static void re_setmulti		(struct rl_softc *);
246static void re_reset		(struct rl_softc *);
247
248static int re_diag		(struct rl_softc *);
249
250#ifdef RE_USEIOSPACE
251#define RL_RES			SYS_RES_IOPORT
252#define RL_RID			RL_PCI_LOIO
253#else
254#define RL_RES			SYS_RES_MEMORY
255#define RL_RID			RL_PCI_LOMEM
256#endif
257
258static device_method_t re_methods[] = {
259	/* Device interface */
260	DEVMETHOD(device_probe,		re_probe),
261	DEVMETHOD(device_attach,	re_attach),
262	DEVMETHOD(device_detach,	re_detach),
263	DEVMETHOD(device_suspend,	re_suspend),
264	DEVMETHOD(device_resume,	re_resume),
265	DEVMETHOD(device_shutdown,	re_shutdown),
266
267	/* bus interface */
268	DEVMETHOD(bus_print_child,	bus_generic_print_child),
269	DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
270
271	/* MII interface */
272	DEVMETHOD(miibus_readreg,	re_miibus_readreg),
273	DEVMETHOD(miibus_writereg,	re_miibus_writereg),
274	DEVMETHOD(miibus_statchg,	re_miibus_statchg),
275
276	{ 0, 0 }
277};
278
279static driver_t re_driver = {
280	"re",
281	re_methods,
282	sizeof(struct rl_softc)
283};
284
285static devclass_t re_devclass;
286
287DRIVER_MODULE(re, pci, re_driver, re_devclass, 0, 0);
288DRIVER_MODULE(re, cardbus, re_driver, re_devclass, 0, 0);
289DRIVER_MODULE(miibus, re, miibus_driver, miibus_devclass, 0, 0);
290
291#define EE_SET(x)					\
292	CSR_WRITE_1(sc, RL_EECMD,			\
293		CSR_READ_1(sc, RL_EECMD) | x)
294
295#define EE_CLR(x)					\
296	CSR_WRITE_1(sc, RL_EECMD,			\
297		CSR_READ_1(sc, RL_EECMD) & ~x)
298
299/*
300 * Send a read command and address to the EEPROM, check for ACK.
301 */
302static void
303re_eeprom_putbyte(sc, addr)
304	struct rl_softc		*sc;
305	int			addr;
306{
307	register int		d, i;
308
309	d = addr | sc->rl_eecmd_read;
310
311	/*
312	 * Feed in each bit and strobe the clock.
313	 */
314	for (i = 0x400; i; i >>= 1) {
315		if (d & i) {
316			EE_SET(RL_EE_DATAIN);
317		} else {
318			EE_CLR(RL_EE_DATAIN);
319		}
320		DELAY(100);
321		EE_SET(RL_EE_CLK);
322		DELAY(150);
323		EE_CLR(RL_EE_CLK);
324		DELAY(100);
325	}
326}
327
328/*
329 * Read a word of data stored in the EEPROM at address 'addr.'
330 */
331static void
332re_eeprom_getword(sc, addr, dest)
333	struct rl_softc		*sc;
334	int			addr;
335	u_int16_t		*dest;
336{
337	register int		i;
338	u_int16_t		word = 0;
339
340	/* Enter EEPROM access mode. */
341	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_PROGRAM|RL_EE_SEL);
342
343	/*
344	 * Send address of word we want to read.
345	 */
346	re_eeprom_putbyte(sc, addr);
347
348	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_PROGRAM|RL_EE_SEL);
349
350	/*
351	 * Start reading bits from EEPROM.
352	 */
353	for (i = 0x8000; i; i >>= 1) {
354		EE_SET(RL_EE_CLK);
355		DELAY(100);
356		if (CSR_READ_1(sc, RL_EECMD) & RL_EE_DATAOUT)
357			word |= i;
358		EE_CLR(RL_EE_CLK);
359		DELAY(100);
360	}
361
362	/* Turn off EEPROM access mode. */
363	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
364
365	*dest = word;
366}
367
368/*
369 * Read a sequence of words from the EEPROM.
370 */
371static void
372re_read_eeprom(sc, dest, off, cnt, swap)
373	struct rl_softc		*sc;
374	caddr_t			dest;
375	int			off;
376	int			cnt;
377	int			swap;
378{
379	int			i;
380	u_int16_t		word = 0, *ptr;
381
382	for (i = 0; i < cnt; i++) {
383		re_eeprom_getword(sc, off + i, &word);
384		ptr = (u_int16_t *)(dest + (i * 2));
385		if (swap)
386			*ptr = ntohs(word);
387		else
388			*ptr = word;
389	}
390}
391
392static int
393re_gmii_readreg(dev, phy, reg)
394	device_t		dev;
395	int			phy, reg;
396{
397	struct rl_softc		*sc;
398	u_int32_t		rval;
399	int			i;
400
401	if (phy != 1)
402		return (0);
403
404	sc = device_get_softc(dev);
405
406	/* Let the rgephy driver read the GMEDIASTAT register */
407
408	if (reg == RL_GMEDIASTAT) {
409		rval = CSR_READ_1(sc, RL_GMEDIASTAT);
410		return (rval);
411	}
412
413	CSR_WRITE_4(sc, RL_PHYAR, reg << 16);
414	DELAY(1000);
415
416	for (i = 0; i < RL_TIMEOUT; i++) {
417		rval = CSR_READ_4(sc, RL_PHYAR);
418		if (rval & RL_PHYAR_BUSY)
419			break;
420		DELAY(100);
421	}
422
423	if (i == RL_TIMEOUT) {
424		printf ("re%d: PHY read failed\n", sc->rl_unit);
425		return (0);
426	}
427
428	return (rval & RL_PHYAR_PHYDATA);
429}
430
431static int
432re_gmii_writereg(dev, phy, reg, data)
433	device_t		dev;
434	int			phy, reg, data;
435{
436	struct rl_softc		*sc;
437	u_int32_t		rval;
438	int			i;
439
440	sc = device_get_softc(dev);
441
442	CSR_WRITE_4(sc, RL_PHYAR, (reg << 16) |
443	    (data & RL_PHYAR_PHYDATA) | RL_PHYAR_BUSY);
444	DELAY(1000);
445
446	for (i = 0; i < RL_TIMEOUT; i++) {
447		rval = CSR_READ_4(sc, RL_PHYAR);
448		if (!(rval & RL_PHYAR_BUSY))
449			break;
450		DELAY(100);
451	}
452
453	if (i == RL_TIMEOUT) {
454		printf ("re%d: PHY write failed\n", sc->rl_unit);
455		return (0);
456	}
457
458	return (0);
459}
460
461static int
462re_miibus_readreg(dev, phy, reg)
463	device_t		dev;
464	int			phy, reg;
465{
466	struct rl_softc		*sc;
467	u_int16_t		rval = 0;
468	u_int16_t		re8139_reg = 0;
469
470	sc = device_get_softc(dev);
471
472	if (sc->rl_type == RL_8169) {
473		rval = re_gmii_readreg(dev, phy, reg);
474		return (rval);
475	}
476
477	/* Pretend the internal PHY is only at address 0 */
478	if (phy) {
479		return (0);
480	}
481	switch (reg) {
482	case MII_BMCR:
483		re8139_reg = RL_BMCR;
484		break;
485	case MII_BMSR:
486		re8139_reg = RL_BMSR;
487		break;
488	case MII_ANAR:
489		re8139_reg = RL_ANAR;
490		break;
491	case MII_ANER:
492		re8139_reg = RL_ANER;
493		break;
494	case MII_ANLPAR:
495		re8139_reg = RL_LPAR;
496		break;
497	case MII_PHYIDR1:
498	case MII_PHYIDR2:
499		return (0);
500	/*
501	 * Allow the rlphy driver to read the media status
502	 * register. If we have a link partner which does not
503	 * support NWAY, this is the register which will tell
504	 * us the results of parallel detection.
505	 */
506	case RL_MEDIASTAT:
507		rval = CSR_READ_1(sc, RL_MEDIASTAT);
508		return (rval);
509	default:
510		printf("re%d: bad phy register\n", sc->rl_unit);
511		return (0);
512	}
513	rval = CSR_READ_2(sc, re8139_reg);
514	return (rval);
515}
516
517static int
518re_miibus_writereg(dev, phy, reg, data)
519	device_t		dev;
520	int			phy, reg, data;
521{
522	struct rl_softc		*sc;
523	u_int16_t		re8139_reg = 0;
524	int			rval = 0;
525
526	sc = device_get_softc(dev);
527
528	if (sc->rl_type == RL_8169) {
529		rval = re_gmii_writereg(dev, phy, reg, data);
530		return (rval);
531	}
532
533	/* Pretend the internal PHY is only at address 0 */
534	if (phy)
535		return (0);
536
537	switch (reg) {
538	case MII_BMCR:
539		re8139_reg = RL_BMCR;
540		break;
541	case MII_BMSR:
542		re8139_reg = RL_BMSR;
543		break;
544	case MII_ANAR:
545		re8139_reg = RL_ANAR;
546		break;
547	case MII_ANER:
548		re8139_reg = RL_ANER;
549		break;
550	case MII_ANLPAR:
551		re8139_reg = RL_LPAR;
552		break;
553	case MII_PHYIDR1:
554	case MII_PHYIDR2:
555		return (0);
556		break;
557	default:
558		printf("re%d: bad phy register\n", sc->rl_unit);
559		return (0);
560	}
561	CSR_WRITE_2(sc, re8139_reg, data);
562	return (0);
563}
564
565static void
566re_miibus_statchg(dev)
567	device_t		dev;
568{
569
570}
571
572/*
573 * Program the 64-bit multicast hash filter.
574 */
575static void
576re_setmulti(sc)
577	struct rl_softc		*sc;
578{
579	struct ifnet		*ifp;
580	int			h = 0;
581	u_int32_t		hashes[2] = { 0, 0 };
582	struct ifmultiaddr	*ifma;
583	u_int32_t		rxfilt;
584	int			mcnt = 0;
585
586	RL_LOCK_ASSERT(sc);
587
588	ifp = &sc->arpcom.ac_if;
589
590	rxfilt = CSR_READ_4(sc, RL_RXCFG);
591
592	if (ifp->if_flags & IFF_ALLMULTI || ifp->if_flags & IFF_PROMISC) {
593		rxfilt |= RL_RXCFG_RX_MULTI;
594		CSR_WRITE_4(sc, RL_RXCFG, rxfilt);
595		CSR_WRITE_4(sc, RL_MAR0, 0xFFFFFFFF);
596		CSR_WRITE_4(sc, RL_MAR4, 0xFFFFFFFF);
597		return;
598	}
599
600	/* first, zot all the existing hash bits */
601	CSR_WRITE_4(sc, RL_MAR0, 0);
602	CSR_WRITE_4(sc, RL_MAR4, 0);
603
604	/* now program new ones */
605	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
606		if (ifma->ifma_addr->sa_family != AF_LINK)
607			continue;
608		h = ether_crc32_be(LLADDR((struct sockaddr_dl *)
609		    ifma->ifma_addr), ETHER_ADDR_LEN) >> 26;
610		if (h < 32)
611			hashes[0] |= (1 << h);
612		else
613			hashes[1] |= (1 << (h - 32));
614		mcnt++;
615	}
616
617	if (mcnt)
618		rxfilt |= RL_RXCFG_RX_MULTI;
619	else
620		rxfilt &= ~RL_RXCFG_RX_MULTI;
621
622	CSR_WRITE_4(sc, RL_RXCFG, rxfilt);
623	CSR_WRITE_4(sc, RL_MAR0, hashes[0]);
624	CSR_WRITE_4(sc, RL_MAR4, hashes[1]);
625}
626
627static void
628re_reset(sc)
629	struct rl_softc		*sc;
630{
631	register int		i;
632
633	RL_LOCK_ASSERT(sc);
634
635	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RESET);
636
637	for (i = 0; i < RL_TIMEOUT; i++) {
638		DELAY(10);
639		if (!(CSR_READ_1(sc, RL_COMMAND) & RL_CMD_RESET))
640			break;
641	}
642	if (i == RL_TIMEOUT)
643		printf("re%d: reset never completed!\n", sc->rl_unit);
644
645	CSR_WRITE_1(sc, 0x82, 1);
646}
647
648/*
649 * The following routine is designed to test for a defect on some
650 * 32-bit 8169 cards. Some of these NICs have the REQ64# and ACK64#
651 * lines connected to the bus, however for a 32-bit only card, they
652 * should be pulled high. The result of this defect is that the
653 * NIC will not work right if you plug it into a 64-bit slot: DMA
654 * operations will be done with 64-bit transfers, which will fail
655 * because the 64-bit data lines aren't connected.
656 *
657 * There's no way to work around this (short of talking a soldering
658 * iron to the board), however we can detect it. The method we use
659 * here is to put the NIC into digital loopback mode, set the receiver
660 * to promiscuous mode, and then try to send a frame. We then compare
661 * the frame data we sent to what was received. If the data matches,
662 * then the NIC is working correctly, otherwise we know the user has
663 * a defective NIC which has been mistakenly plugged into a 64-bit PCI
664 * slot. In the latter case, there's no way the NIC can work correctly,
665 * so we print out a message on the console and abort the device attach.
666 */
667
668static int
669re_diag(sc)
670	struct rl_softc		*sc;
671{
672	struct ifnet		*ifp = &sc->arpcom.ac_if;
673	struct mbuf		*m0;
674	struct ether_header	*eh;
675	struct rl_desc		*cur_rx;
676	u_int16_t		status;
677	u_int32_t		rxstat;
678	int			total_len, i, error = 0;
679	u_int8_t		dst[] = { 0x00, 'h', 'e', 'l', 'l', 'o' };
680	u_int8_t		src[] = { 0x00, 'w', 'o', 'r', 'l', 'd' };
681
682	/* Allocate a single mbuf */
683	MGETHDR(m0, M_DONTWAIT, MT_DATA);
684	if (m0 == NULL)
685		return (ENOBUFS);
686
687	RL_LOCK(sc);
688
689	/*
690	 * Initialize the NIC in test mode. This sets the chip up
691	 * so that it can send and receive frames, but performs the
692	 * following special functions:
693	 * - Puts receiver in promiscuous mode
694	 * - Enables digital loopback mode
695	 * - Leaves interrupts turned off
696	 */
697
698	ifp->if_flags |= IFF_PROMISC;
699	sc->rl_testmode = 1;
700	re_init_locked(sc);
701	re_stop(sc);
702	DELAY(100000);
703	re_init_locked(sc);
704
705	/* Put some data in the mbuf */
706
707	eh = mtod(m0, struct ether_header *);
708	bcopy ((char *)&dst, eh->ether_dhost, ETHER_ADDR_LEN);
709	bcopy ((char *)&src, eh->ether_shost, ETHER_ADDR_LEN);
710	eh->ether_type = htons(ETHERTYPE_IP);
711	m0->m_pkthdr.len = m0->m_len = ETHER_MIN_LEN - ETHER_CRC_LEN;
712
713	/*
714	 * Queue the packet, start transmission.
715	 * Note: IF_HANDOFF() ultimately calls re_start() for us.
716	 */
717
718	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
719	RL_UNLOCK(sc);
720	/* XXX: re_diag must not be called when in ALTQ mode */
721	IF_HANDOFF(&ifp->if_snd, m0, ifp);
722	RL_LOCK(sc);
723	m0 = NULL;
724
725	/* Wait for it to propagate through the chip */
726
727	DELAY(100000);
728	for (i = 0; i < RL_TIMEOUT; i++) {
729		status = CSR_READ_2(sc, RL_ISR);
730		if ((status & (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK)) ==
731		    (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK))
732			break;
733		DELAY(10);
734	}
735
736	if (i == RL_TIMEOUT) {
737		printf("re%d: diagnostic failed, failed to receive packet "
738		    "in loopback mode\n", sc->rl_unit);
739		error = EIO;
740		goto done;
741	}
742
743	/*
744	 * The packet should have been dumped into the first
745	 * entry in the RX DMA ring. Grab it from there.
746	 */
747
748	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
749	    sc->rl_ldata.rl_rx_list_map,
750	    BUS_DMASYNC_POSTREAD);
751	bus_dmamap_sync(sc->rl_ldata.rl_mtag,
752	    sc->rl_ldata.rl_rx_dmamap[0],
753	    BUS_DMASYNC_POSTWRITE);
754	bus_dmamap_unload(sc->rl_ldata.rl_mtag,
755	    sc->rl_ldata.rl_rx_dmamap[0]);
756
757	m0 = sc->rl_ldata.rl_rx_mbuf[0];
758	sc->rl_ldata.rl_rx_mbuf[0] = NULL;
759	eh = mtod(m0, struct ether_header *);
760
761	cur_rx = &sc->rl_ldata.rl_rx_list[0];
762	total_len = RL_RXBYTES(cur_rx);
763	rxstat = le32toh(cur_rx->rl_cmdstat);
764
765	if (total_len != ETHER_MIN_LEN) {
766		printf("re%d: diagnostic failed, received short packet\n",
767		    sc->rl_unit);
768		error = EIO;
769		goto done;
770	}
771
772	/* Test that the received packet data matches what we sent. */
773
774	if (bcmp((char *)&eh->ether_dhost, (char *)&dst, ETHER_ADDR_LEN) ||
775	    bcmp((char *)&eh->ether_shost, (char *)&src, ETHER_ADDR_LEN) ||
776	    ntohs(eh->ether_type) != ETHERTYPE_IP) {
777		printf("re%d: WARNING, DMA FAILURE!\n", sc->rl_unit);
778		printf("re%d: expected TX data: %6D/%6D/0x%x\n", sc->rl_unit,
779		    dst, ":", src, ":", ETHERTYPE_IP);
780		printf("re%d: received RX data: %6D/%6D/0x%x\n", sc->rl_unit,
781		    eh->ether_dhost, ":",  eh->ether_shost, ":",
782		    ntohs(eh->ether_type));
783		printf("re%d: You may have a defective 32-bit NIC plugged "
784		    "into a 64-bit PCI slot.\n", sc->rl_unit);
785		printf("re%d: Please re-install the NIC in a 32-bit slot "
786		    "for proper operation.\n", sc->rl_unit);
787		printf("re%d: Read the re(4) man page for more details.\n",
788		    sc->rl_unit);
789		error = EIO;
790	}
791
792done:
793	/* Turn interface off, release resources */
794
795	sc->rl_testmode = 0;
796	ifp->if_flags &= ~IFF_PROMISC;
797	re_stop(sc);
798	if (m0 != NULL)
799		m_freem(m0);
800
801	RL_UNLOCK(sc);
802
803	return (error);
804}
805
806/*
807 * Probe for a RealTek 8139C+/8169/8110 chip. Check the PCI vendor and device
808 * IDs against our list and return a device name if we find a match.
809 */
810static int
811re_probe(dev)
812	device_t		dev;
813{
814	struct rl_type		*t;
815	struct rl_softc		*sc;
816	int			rid;
817	u_int32_t		hwrev;
818
819	t = re_devs;
820	sc = device_get_softc(dev);
821
822	while (t->rl_name != NULL) {
823		if ((pci_get_vendor(dev) == t->rl_vid) &&
824		    (pci_get_device(dev) == t->rl_did)) {
825
826			/*
827			 * Temporarily map the I/O space
828			 * so we can read the chip ID register.
829			 */
830			rid = RL_RID;
831			sc->rl_res = bus_alloc_resource_any(dev, RL_RES, &rid,
832			    RF_ACTIVE);
833			if (sc->rl_res == NULL) {
834				device_printf(dev,
835				    "couldn't map ports/memory\n");
836				return (ENXIO);
837			}
838			sc->rl_btag = rman_get_bustag(sc->rl_res);
839			sc->rl_bhandle = rman_get_bushandle(sc->rl_res);
840			hwrev = CSR_READ_4(sc, RL_TXCFG) & RL_TXCFG_HWREV;
841			bus_release_resource(dev, RL_RES,
842			    RL_RID, sc->rl_res);
843			if (t->rl_basetype == hwrev) {
844				device_set_desc(dev, t->rl_name);
845				return (BUS_PROBE_DEFAULT);
846			}
847		}
848		t++;
849	}
850
851	return (ENXIO);
852}
853
854/*
855 * This routine takes the segment list provided as the result of
856 * a bus_dma_map_load() operation and assigns the addresses/lengths
857 * to RealTek DMA descriptors. This can be called either by the RX
858 * code or the TX code. In the RX case, we'll probably wind up mapping
859 * at most one segment. For the TX case, there could be any number of
860 * segments since TX packets may span multiple mbufs. In either case,
861 * if the number of segments is larger than the rl_maxsegs limit
862 * specified by the caller, we abort the mapping operation. Sadly,
863 * whoever designed the buffer mapping API did not provide a way to
864 * return an error from here, so we have to fake it a bit.
865 */
866
867static void
868re_dma_map_desc(arg, segs, nseg, mapsize, error)
869	void			*arg;
870	bus_dma_segment_t	*segs;
871	int			nseg;
872	bus_size_t		mapsize;
873	int			error;
874{
875	struct rl_dmaload_arg	*ctx;
876	struct rl_desc		*d = NULL;
877	int			i = 0, idx;
878
879	if (error)
880		return;
881
882	ctx = arg;
883
884	/* Signal error to caller if there's too many segments */
885	if (nseg > ctx->rl_maxsegs) {
886		ctx->rl_maxsegs = 0;
887		return;
888	}
889
890	/*
891	 * Map the segment array into descriptors. Note that we set the
892	 * start-of-frame and end-of-frame markers for either TX or RX, but
893	 * they really only have meaning in the TX case. (In the RX case,
894	 * it's the chip that tells us where packets begin and end.)
895	 * We also keep track of the end of the ring and set the
896	 * end-of-ring bits as needed, and we set the ownership bits
897	 * in all except the very first descriptor. (The caller will
898	 * set this descriptor later when it start transmission or
899	 * reception.)
900	 */
901	idx = ctx->rl_idx;
902	for (;;) {
903		u_int32_t		cmdstat;
904		d = &ctx->rl_ring[idx];
905		if (le32toh(d->rl_cmdstat) & RL_RDESC_STAT_OWN) {
906			ctx->rl_maxsegs = 0;
907			return;
908		}
909		cmdstat = segs[i].ds_len;
910		d->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[i].ds_addr));
911		d->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[i].ds_addr));
912		if (i == 0)
913			cmdstat |= RL_TDESC_CMD_SOF;
914		else
915			cmdstat |= RL_TDESC_CMD_OWN;
916		if (idx == (RL_RX_DESC_CNT - 1))
917			cmdstat |= RL_TDESC_CMD_EOR;
918		d->rl_cmdstat = htole32(cmdstat | ctx->rl_flags);
919		i++;
920		if (i == nseg)
921			break;
922		RL_DESC_INC(idx);
923	}
924
925	d->rl_cmdstat |= htole32(RL_TDESC_CMD_EOF);
926	ctx->rl_maxsegs = nseg;
927	ctx->rl_idx = idx;
928}
929
930/*
931 * Map a single buffer address.
932 */
933
934static void
935re_dma_map_addr(arg, segs, nseg, error)
936	void			*arg;
937	bus_dma_segment_t	*segs;
938	int			nseg;
939	int			error;
940{
941	u_int32_t		*addr;
942
943	if (error)
944		return;
945
946	KASSERT(nseg == 1, ("too many DMA segments, %d should be 1", nseg));
947	addr = arg;
948	*addr = segs->ds_addr;
949}
950
951static int
952re_allocmem(dev, sc)
953	device_t		dev;
954	struct rl_softc		*sc;
955{
956	int			error;
957	int			nseg;
958	int			i;
959
960	/*
961	 * Allocate map for RX mbufs.
962	 */
963	nseg = 32;
964	error = bus_dma_tag_create(sc->rl_parent_tag, ETHER_ALIGN, 0,
965	    BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
966	    NULL, MCLBYTES * nseg, nseg, MCLBYTES, BUS_DMA_ALLOCNOW,
967	    NULL, NULL, &sc->rl_ldata.rl_mtag);
968	if (error) {
969		device_printf(dev, "could not allocate dma tag\n");
970		return (ENOMEM);
971	}
972
973	/*
974	 * Allocate map for TX descriptor list.
975	 */
976	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
977	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
978	    NULL, RL_TX_LIST_SZ, 1, RL_TX_LIST_SZ, BUS_DMA_ALLOCNOW,
979	    NULL, NULL, &sc->rl_ldata.rl_tx_list_tag);
980	if (error) {
981		device_printf(dev, "could not allocate dma tag\n");
982		return (ENOMEM);
983	}
984
985	/* Allocate DMA'able memory for the TX ring */
986
987	error = bus_dmamem_alloc(sc->rl_ldata.rl_tx_list_tag,
988	    (void **)&sc->rl_ldata.rl_tx_list, BUS_DMA_NOWAIT | BUS_DMA_ZERO,
989	    &sc->rl_ldata.rl_tx_list_map);
990	if (error)
991		return (ENOMEM);
992
993	/* Load the map for the TX ring. */
994
995	error = bus_dmamap_load(sc->rl_ldata.rl_tx_list_tag,
996	     sc->rl_ldata.rl_tx_list_map, sc->rl_ldata.rl_tx_list,
997	     RL_TX_LIST_SZ, re_dma_map_addr,
998	     &sc->rl_ldata.rl_tx_list_addr, BUS_DMA_NOWAIT);
999
1000	/* Create DMA maps for TX buffers */
1001
1002	for (i = 0; i < RL_TX_DESC_CNT; i++) {
1003		error = bus_dmamap_create(sc->rl_ldata.rl_mtag, 0,
1004			    &sc->rl_ldata.rl_tx_dmamap[i]);
1005		if (error) {
1006			device_printf(dev, "can't create DMA map for TX\n");
1007			return (ENOMEM);
1008		}
1009	}
1010
1011	/*
1012	 * Allocate map for RX descriptor list.
1013	 */
1014	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
1015	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1016	    NULL, RL_RX_LIST_SZ, 1, RL_RX_LIST_SZ, BUS_DMA_ALLOCNOW,
1017	    NULL, NULL, &sc->rl_ldata.rl_rx_list_tag);
1018	if (error) {
1019		device_printf(dev, "could not allocate dma tag\n");
1020		return (ENOMEM);
1021	}
1022
1023	/* Allocate DMA'able memory for the RX ring */
1024
1025	error = bus_dmamem_alloc(sc->rl_ldata.rl_rx_list_tag,
1026	    (void **)&sc->rl_ldata.rl_rx_list, BUS_DMA_NOWAIT | BUS_DMA_ZERO,
1027	    &sc->rl_ldata.rl_rx_list_map);
1028	if (error)
1029		return (ENOMEM);
1030
1031	/* Load the map for the RX ring. */
1032
1033	error = bus_dmamap_load(sc->rl_ldata.rl_rx_list_tag,
1034	     sc->rl_ldata.rl_rx_list_map, sc->rl_ldata.rl_rx_list,
1035	     RL_RX_LIST_SZ, re_dma_map_addr,
1036	     &sc->rl_ldata.rl_rx_list_addr, BUS_DMA_NOWAIT);
1037
1038	/* Create DMA maps for RX buffers */
1039
1040	for (i = 0; i < RL_RX_DESC_CNT; i++) {
1041		error = bus_dmamap_create(sc->rl_ldata.rl_mtag, 0,
1042			    &sc->rl_ldata.rl_rx_dmamap[i]);
1043		if (error) {
1044			device_printf(dev, "can't create DMA map for RX\n");
1045			return (ENOMEM);
1046		}
1047	}
1048
1049	return (0);
1050}
1051
1052/*
1053 * Attach the interface. Allocate softc structures, do ifmedia
1054 * setup and ethernet/BPF attach.
1055 */
1056static int
1057re_attach(dev)
1058	device_t		dev;
1059{
1060	u_char			eaddr[ETHER_ADDR_LEN];
1061	u_int16_t		as[3];
1062	struct rl_softc		*sc;
1063	struct ifnet		*ifp;
1064	struct rl_hwrev		*hw_rev;
1065	int			hwrev;
1066	u_int16_t		re_did = 0;
1067	int			unit, error = 0, rid, i;
1068
1069	sc = device_get_softc(dev);
1070	unit = device_get_unit(dev);
1071
1072	mtx_init(&sc->rl_mtx, device_get_nameunit(dev), MTX_NETWORK_LOCK,
1073	    MTX_DEF);
1074	/*
1075	 * Map control/status registers.
1076	 */
1077	pci_enable_busmaster(dev);
1078
1079	rid = RL_RID;
1080	sc->rl_res = bus_alloc_resource_any(dev, RL_RES, &rid,
1081	    RF_ACTIVE);
1082
1083	if (sc->rl_res == NULL) {
1084		printf ("re%d: couldn't map ports/memory\n", unit);
1085		error = ENXIO;
1086		goto fail;
1087	}
1088
1089	sc->rl_btag = rman_get_bustag(sc->rl_res);
1090	sc->rl_bhandle = rman_get_bushandle(sc->rl_res);
1091
1092	/* Allocate interrupt */
1093	rid = 0;
1094	sc->rl_irq = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
1095	    RF_SHAREABLE | RF_ACTIVE);
1096
1097	if (sc->rl_irq == NULL) {
1098		printf("re%d: couldn't map interrupt\n", unit);
1099		error = ENXIO;
1100		goto fail;
1101	}
1102
1103	/* Reset the adapter. */
1104	RL_LOCK(sc);
1105	re_reset(sc);
1106	RL_UNLOCK(sc);
1107
1108	hw_rev = re_hwrevs;
1109	hwrev = CSR_READ_4(sc, RL_TXCFG) & RL_TXCFG_HWREV;
1110	while (hw_rev->rl_desc != NULL) {
1111		if (hw_rev->rl_rev == hwrev) {
1112			sc->rl_type = hw_rev->rl_type;
1113			break;
1114		}
1115		hw_rev++;
1116	}
1117
1118	if (sc->rl_type == RL_8169) {
1119
1120		/* Set RX length mask */
1121
1122		sc->rl_rxlenmask = RL_RDESC_STAT_GFRAGLEN;
1123
1124		/* Force station address autoload from the EEPROM */
1125
1126		CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_AUTOLOAD);
1127		for (i = 0; i < RL_TIMEOUT; i++) {
1128			if (!(CSR_READ_1(sc, RL_EECMD) & RL_EEMODE_AUTOLOAD))
1129				break;
1130			DELAY(100);
1131		}
1132		if (i == RL_TIMEOUT)
1133			printf ("re%d: eeprom autoload timed out\n", unit);
1134
1135			for (i = 0; i < ETHER_ADDR_LEN; i++)
1136				eaddr[i] = CSR_READ_1(sc, RL_IDR0 + i);
1137	} else {
1138
1139		/* Set RX length mask */
1140
1141		sc->rl_rxlenmask = RL_RDESC_STAT_FRAGLEN;
1142
1143		sc->rl_eecmd_read = RL_EECMD_READ_6BIT;
1144		re_read_eeprom(sc, (caddr_t)&re_did, 0, 1, 0);
1145		if (re_did != 0x8129)
1146			sc->rl_eecmd_read = RL_EECMD_READ_8BIT;
1147
1148		/*
1149		 * Get station address from the EEPROM.
1150		 */
1151		re_read_eeprom(sc, (caddr_t)as, RL_EE_EADDR, 3, 0);
1152		for (i = 0; i < 3; i++) {
1153			eaddr[(i * 2) + 0] = as[i] & 0xff;
1154			eaddr[(i * 2) + 1] = as[i] >> 8;
1155		}
1156	}
1157
1158	sc->rl_unit = unit;
1159	bcopy(eaddr, (char *)&sc->arpcom.ac_enaddr, ETHER_ADDR_LEN);
1160
1161	/*
1162	 * Allocate the parent bus DMA tag appropriate for PCI.
1163	 */
1164#define RL_NSEG_NEW 32
1165	error = bus_dma_tag_create(NULL,	/* parent */
1166			1, 0,			/* alignment, boundary */
1167			BUS_SPACE_MAXADDR_32BIT,/* lowaddr */
1168			BUS_SPACE_MAXADDR,	/* highaddr */
1169			NULL, NULL,		/* filter, filterarg */
1170			MAXBSIZE, RL_NSEG_NEW,	/* maxsize, nsegments */
1171			BUS_SPACE_MAXSIZE_32BIT,/* maxsegsize */
1172			BUS_DMA_ALLOCNOW,	/* flags */
1173			NULL, NULL,		/* lockfunc, lockarg */
1174			&sc->rl_parent_tag);
1175	if (error)
1176		goto fail;
1177
1178	error = re_allocmem(dev, sc);
1179
1180	if (error)
1181		goto fail;
1182
1183	/* Do MII setup */
1184	if (mii_phy_probe(dev, &sc->rl_miibus,
1185	    re_ifmedia_upd, re_ifmedia_sts)) {
1186		printf("re%d: MII without any phy!\n", sc->rl_unit);
1187		error = ENXIO;
1188		goto fail;
1189	}
1190
1191	ifp = &sc->arpcom.ac_if;
1192	ifp->if_softc = sc;
1193	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1194	ifp->if_mtu = ETHERMTU;
1195	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1196	ifp->if_ioctl = re_ioctl;
1197	ifp->if_capabilities = IFCAP_VLAN_MTU;
1198	ifp->if_start = re_start;
1199	ifp->if_hwassist = /*RE_CSUM_FEATURES*/0;
1200	ifp->if_capabilities |= IFCAP_HWCSUM|IFCAP_VLAN_HWTAGGING;
1201#ifdef DEVICE_POLLING
1202	ifp->if_capabilities |= IFCAP_POLLING;
1203#endif
1204	ifp->if_watchdog = re_watchdog;
1205	ifp->if_init = re_init;
1206	if (sc->rl_type == RL_8169)
1207		ifp->if_baudrate = 1000000000;
1208	else
1209		ifp->if_baudrate = 100000000;
1210	IFQ_SET_MAXLEN(&ifp->if_snd,  RL_IFQ_MAXLEN);
1211	ifp->if_snd.ifq_drv_maxlen = RL_IFQ_MAXLEN;
1212	IFQ_SET_READY(&ifp->if_snd);
1213	ifp->if_capenable = ifp->if_capabilities & ~IFCAP_HWCSUM;
1214
1215	callout_handle_init(&sc->rl_stat_ch);
1216
1217	/*
1218	 * Call MI attach routine.
1219	 */
1220	ether_ifattach(ifp, eaddr);
1221
1222	/* Perform hardware diagnostic. */
1223	error = re_diag(sc);
1224
1225	if (error) {
1226		printf("re%d: attach aborted due to hardware diag failure\n",
1227		    unit);
1228		ether_ifdetach(ifp);
1229		goto fail;
1230	}
1231
1232	/* Hook interrupt last to avoid having to lock softc */
1233	error = bus_setup_intr(dev, sc->rl_irq, INTR_TYPE_NET | INTR_MPSAFE,
1234	    re_intr, sc, &sc->rl_intrhand);
1235	if (error) {
1236		printf("re%d: couldn't set up irq\n", unit);
1237		ether_ifdetach(ifp);
1238	}
1239
1240fail:
1241	if (error)
1242		re_detach(dev);
1243
1244	return (error);
1245}
1246
1247/*
1248 * Shutdown hardware and free up resources. This can be called any
1249 * time after the mutex has been initialized. It is called in both
1250 * the error case in attach and the normal detach case so it needs
1251 * to be careful about only freeing resources that have actually been
1252 * allocated.
1253 */
1254static int
1255re_detach(dev)
1256	device_t		dev;
1257{
1258	struct rl_softc		*sc;
1259	struct ifnet		*ifp;
1260	int			i;
1261	int			attached;
1262
1263	sc = device_get_softc(dev);
1264	ifp = &sc->arpcom.ac_if;
1265	KASSERT(mtx_initialized(&sc->rl_mtx), ("re mutex not initialized"));
1266
1267	attached = device_is_attached(dev);
1268	/* These should only be active if attach succeeded */
1269	if (attached)
1270		ether_ifdetach(ifp);
1271
1272	RL_LOCK(sc);
1273#if 0
1274	sc->suspended = 1;
1275#endif
1276
1277	/* These should only be active if attach succeeded */
1278	if (attached) {
1279		re_stop(sc);
1280		/*
1281		 * Force off the IFF_UP flag here, in case someone
1282		 * still had a BPF descriptor attached to this
1283		 * interface. If they do, ether_ifdetach() will cause
1284		 * the BPF code to try and clear the promisc mode
1285		 * flag, which will bubble down to re_ioctl(),
1286		 * which will try to call re_init() again. This will
1287		 * turn the NIC back on and restart the MII ticker,
1288		 * which will panic the system when the kernel tries
1289		 * to invoke the re_tick() function that isn't there
1290		 * anymore.
1291		 */
1292		ifp->if_flags &= ~IFF_UP;
1293	}
1294	if (sc->rl_miibus)
1295		device_delete_child(dev, sc->rl_miibus);
1296	bus_generic_detach(dev);
1297
1298	/*
1299	 * The rest is resource deallocation, so we should already be
1300	 * stopped here.
1301	 */
1302	RL_UNLOCK(sc);
1303
1304	if (sc->rl_intrhand)
1305		bus_teardown_intr(dev, sc->rl_irq, sc->rl_intrhand);
1306	if (sc->rl_irq)
1307		bus_release_resource(dev, SYS_RES_IRQ, 0, sc->rl_irq);
1308	if (sc->rl_res)
1309		bus_release_resource(dev, RL_RES, RL_RID, sc->rl_res);
1310
1311
1312	/* Unload and free the RX DMA ring memory and map */
1313
1314	if (sc->rl_ldata.rl_rx_list_tag) {
1315		bus_dmamap_unload(sc->rl_ldata.rl_rx_list_tag,
1316		    sc->rl_ldata.rl_rx_list_map);
1317		bus_dmamem_free(sc->rl_ldata.rl_rx_list_tag,
1318		    sc->rl_ldata.rl_rx_list,
1319		    sc->rl_ldata.rl_rx_list_map);
1320		bus_dma_tag_destroy(sc->rl_ldata.rl_rx_list_tag);
1321	}
1322
1323	/* Unload and free the TX DMA ring memory and map */
1324
1325	if (sc->rl_ldata.rl_tx_list_tag) {
1326		bus_dmamap_unload(sc->rl_ldata.rl_tx_list_tag,
1327		    sc->rl_ldata.rl_tx_list_map);
1328		bus_dmamem_free(sc->rl_ldata.rl_tx_list_tag,
1329		    sc->rl_ldata.rl_tx_list,
1330		    sc->rl_ldata.rl_tx_list_map);
1331		bus_dma_tag_destroy(sc->rl_ldata.rl_tx_list_tag);
1332	}
1333
1334	/* Destroy all the RX and TX buffer maps */
1335
1336	if (sc->rl_ldata.rl_mtag) {
1337		for (i = 0; i < RL_TX_DESC_CNT; i++)
1338			bus_dmamap_destroy(sc->rl_ldata.rl_mtag,
1339			    sc->rl_ldata.rl_tx_dmamap[i]);
1340		for (i = 0; i < RL_RX_DESC_CNT; i++)
1341			bus_dmamap_destroy(sc->rl_ldata.rl_mtag,
1342			    sc->rl_ldata.rl_rx_dmamap[i]);
1343		bus_dma_tag_destroy(sc->rl_ldata.rl_mtag);
1344	}
1345
1346	/* Unload and free the stats buffer and map */
1347
1348	if (sc->rl_ldata.rl_stag) {
1349		bus_dmamap_unload(sc->rl_ldata.rl_stag,
1350		    sc->rl_ldata.rl_rx_list_map);
1351		bus_dmamem_free(sc->rl_ldata.rl_stag,
1352		    sc->rl_ldata.rl_stats,
1353		    sc->rl_ldata.rl_smap);
1354		bus_dma_tag_destroy(sc->rl_ldata.rl_stag);
1355	}
1356
1357	if (sc->rl_parent_tag)
1358		bus_dma_tag_destroy(sc->rl_parent_tag);
1359
1360	mtx_destroy(&sc->rl_mtx);
1361
1362	return (0);
1363}
1364
1365static int
1366re_newbuf(sc, idx, m)
1367	struct rl_softc		*sc;
1368	int			idx;
1369	struct mbuf		*m;
1370{
1371	struct rl_dmaload_arg	arg;
1372	struct mbuf		*n = NULL;
1373	int			error;
1374
1375	if (m == NULL) {
1376		n = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
1377		if (n == NULL)
1378			return (ENOBUFS);
1379		m = n;
1380	} else
1381		m->m_data = m->m_ext.ext_buf;
1382
1383	m->m_len = m->m_pkthdr.len = MCLBYTES;
1384#ifdef RE_FIXUP_RX
1385	/*
1386	 * This is part of an evil trick to deal with non-x86 platforms.
1387	 * The RealTek chip requires RX buffers to be aligned on 64-bit
1388	 * boundaries, but that will hose non-x86 machines. To get around
1389	 * this, we leave some empty space at the start of each buffer
1390	 * and for non-x86 hosts, we copy the buffer back six bytes
1391	 * to achieve word alignment. This is slightly more efficient
1392	 * than allocating a new buffer, copying the contents, and
1393	 * discarding the old buffer.
1394	 */
1395	m_adj(m, RE_ETHER_ALIGN);
1396#endif
1397	arg.sc = sc;
1398	arg.rl_idx = idx;
1399	arg.rl_maxsegs = 1;
1400	arg.rl_flags = 0;
1401	arg.rl_ring = sc->rl_ldata.rl_rx_list;
1402
1403	error = bus_dmamap_load_mbuf(sc->rl_ldata.rl_mtag,
1404	    sc->rl_ldata.rl_rx_dmamap[idx], m, re_dma_map_desc,
1405	    &arg, BUS_DMA_NOWAIT);
1406	if (error || arg.rl_maxsegs != 1) {
1407		if (n != NULL)
1408			m_freem(n);
1409		return (ENOMEM);
1410	}
1411
1412	sc->rl_ldata.rl_rx_list[idx].rl_cmdstat |= htole32(RL_RDESC_CMD_OWN);
1413	sc->rl_ldata.rl_rx_mbuf[idx] = m;
1414
1415	bus_dmamap_sync(sc->rl_ldata.rl_mtag,
1416	    sc->rl_ldata.rl_rx_dmamap[idx],
1417	    BUS_DMASYNC_PREREAD);
1418
1419	return (0);
1420}
1421
1422#ifdef RE_FIXUP_RX
1423static __inline void
1424re_fixup_rx(m)
1425	struct mbuf		*m;
1426{
1427	int                     i;
1428	uint16_t                *src, *dst;
1429
1430	src = mtod(m, uint16_t *);
1431	dst = src - (RE_ETHER_ALIGN - ETHER_ALIGN) / sizeof *src;
1432
1433	for (i = 0; i < (m->m_len / sizeof(uint16_t) + 1); i++)
1434		*dst++ = *src++;
1435
1436	m->m_data -= RE_ETHER_ALIGN - ETHER_ALIGN;
1437
1438	return;
1439}
1440#endif
1441
1442static int
1443re_tx_list_init(sc)
1444	struct rl_softc		*sc;
1445{
1446
1447	RL_LOCK_ASSERT(sc);
1448
1449	bzero ((char *)sc->rl_ldata.rl_tx_list, RL_TX_LIST_SZ);
1450	bzero ((char *)&sc->rl_ldata.rl_tx_mbuf,
1451	    (RL_TX_DESC_CNT * sizeof(struct mbuf *)));
1452
1453	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
1454	    sc->rl_ldata.rl_tx_list_map, BUS_DMASYNC_PREWRITE);
1455	sc->rl_ldata.rl_tx_prodidx = 0;
1456	sc->rl_ldata.rl_tx_considx = 0;
1457	sc->rl_ldata.rl_tx_free = RL_TX_DESC_CNT;
1458
1459	return (0);
1460}
1461
1462static int
1463re_rx_list_init(sc)
1464	struct rl_softc		*sc;
1465{
1466	int			i;
1467
1468	bzero ((char *)sc->rl_ldata.rl_rx_list, RL_RX_LIST_SZ);
1469	bzero ((char *)&sc->rl_ldata.rl_rx_mbuf,
1470	    (RL_RX_DESC_CNT * sizeof(struct mbuf *)));
1471
1472	for (i = 0; i < RL_RX_DESC_CNT; i++) {
1473		if (re_newbuf(sc, i, NULL) == ENOBUFS)
1474			return (ENOBUFS);
1475	}
1476
1477	/* Flush the RX descriptors */
1478
1479	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1480	    sc->rl_ldata.rl_rx_list_map,
1481	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
1482
1483	sc->rl_ldata.rl_rx_prodidx = 0;
1484	sc->rl_head = sc->rl_tail = NULL;
1485
1486	return (0);
1487}
1488
1489/*
1490 * RX handler for C+ and 8169. For the gigE chips, we support
1491 * the reception of jumbo frames that have been fragmented
1492 * across multiple 2K mbuf cluster buffers.
1493 */
1494static void
1495re_rxeof(sc)
1496	struct rl_softc		*sc;
1497{
1498	struct mbuf		*m;
1499	struct ifnet		*ifp;
1500	int			i, total_len;
1501	struct rl_desc		*cur_rx;
1502	u_int32_t		rxstat, rxvlan;
1503
1504	RL_LOCK_ASSERT(sc);
1505
1506	ifp = &sc->arpcom.ac_if;
1507	i = sc->rl_ldata.rl_rx_prodidx;
1508
1509	/* Invalidate the descriptor memory */
1510
1511	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1512	    sc->rl_ldata.rl_rx_list_map,
1513	    BUS_DMASYNC_POSTREAD);
1514
1515	while (!RL_OWN(&sc->rl_ldata.rl_rx_list[i])) {
1516		cur_rx = &sc->rl_ldata.rl_rx_list[i];
1517		m = sc->rl_ldata.rl_rx_mbuf[i];
1518		total_len = RL_RXBYTES(cur_rx);
1519		rxstat = le32toh(cur_rx->rl_cmdstat);
1520		rxvlan = le32toh(cur_rx->rl_vlanctl);
1521
1522		/* Invalidate the RX mbuf and unload its map */
1523
1524		bus_dmamap_sync(sc->rl_ldata.rl_mtag,
1525		    sc->rl_ldata.rl_rx_dmamap[i],
1526		    BUS_DMASYNC_POSTWRITE);
1527		bus_dmamap_unload(sc->rl_ldata.rl_mtag,
1528		    sc->rl_ldata.rl_rx_dmamap[i]);
1529
1530		if (!(rxstat & RL_RDESC_STAT_EOF)) {
1531			m->m_len = RE_RX_DESC_BUFLEN;
1532			if (sc->rl_head == NULL)
1533				sc->rl_head = sc->rl_tail = m;
1534			else {
1535				m->m_flags &= ~M_PKTHDR;
1536				sc->rl_tail->m_next = m;
1537				sc->rl_tail = m;
1538			}
1539			re_newbuf(sc, i, NULL);
1540			RL_DESC_INC(i);
1541			continue;
1542		}
1543
1544		/*
1545		 * NOTE: for the 8139C+, the frame length field
1546		 * is always 12 bits in size, but for the gigE chips,
1547		 * it is 13 bits (since the max RX frame length is 16K).
1548		 * Unfortunately, all 32 bits in the status word
1549		 * were already used, so to make room for the extra
1550		 * length bit, RealTek took out the 'frame alignment
1551		 * error' bit and shifted the other status bits
1552		 * over one slot. The OWN, EOR, FS and LS bits are
1553		 * still in the same places. We have already extracted
1554		 * the frame length and checked the OWN bit, so rather
1555		 * than using an alternate bit mapping, we shift the
1556		 * status bits one space to the right so we can evaluate
1557		 * them using the 8169 status as though it was in the
1558		 * same format as that of the 8139C+.
1559		 */
1560		if (sc->rl_type == RL_8169)
1561			rxstat >>= 1;
1562
1563		/*
1564		 * if total_len > 2^13-1, both _RXERRSUM and _GIANT will be
1565		 * set, but if CRC is clear, it will still be a valid frame.
1566		 */
1567		if (rxstat & RL_RDESC_STAT_RXERRSUM && !(total_len > 8191 &&
1568		    (rxstat & RL_RDESC_STAT_ERRS) == RL_RDESC_STAT_GIANT)) {
1569			ifp->if_ierrors++;
1570			/*
1571			 * If this is part of a multi-fragment packet,
1572			 * discard all the pieces.
1573			 */
1574			if (sc->rl_head != NULL) {
1575				m_freem(sc->rl_head);
1576				sc->rl_head = sc->rl_tail = NULL;
1577			}
1578			re_newbuf(sc, i, m);
1579			RL_DESC_INC(i);
1580			continue;
1581		}
1582
1583		/*
1584		 * If allocating a replacement mbuf fails,
1585		 * reload the current one.
1586		 */
1587
1588		if (re_newbuf(sc, i, NULL)) {
1589			ifp->if_ierrors++;
1590			if (sc->rl_head != NULL) {
1591				m_freem(sc->rl_head);
1592				sc->rl_head = sc->rl_tail = NULL;
1593			}
1594			re_newbuf(sc, i, m);
1595			RL_DESC_INC(i);
1596			continue;
1597		}
1598
1599		RL_DESC_INC(i);
1600
1601		if (sc->rl_head != NULL) {
1602			m->m_len = total_len % RE_RX_DESC_BUFLEN;
1603			if (m->m_len == 0)
1604				m->m_len = RE_RX_DESC_BUFLEN;
1605			/*
1606			 * Special case: if there's 4 bytes or less
1607			 * in this buffer, the mbuf can be discarded:
1608			 * the last 4 bytes is the CRC, which we don't
1609			 * care about anyway.
1610			 */
1611			if (m->m_len <= ETHER_CRC_LEN) {
1612				sc->rl_tail->m_len -=
1613				    (ETHER_CRC_LEN - m->m_len);
1614				m_freem(m);
1615			} else {
1616				m->m_len -= ETHER_CRC_LEN;
1617				m->m_flags &= ~M_PKTHDR;
1618				sc->rl_tail->m_next = m;
1619			}
1620			m = sc->rl_head;
1621			sc->rl_head = sc->rl_tail = NULL;
1622			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
1623		} else
1624			m->m_pkthdr.len = m->m_len =
1625			    (total_len - ETHER_CRC_LEN);
1626
1627#ifdef RE_FIXUP_RX
1628		re_fixup_rx(m);
1629#endif
1630		ifp->if_ipackets++;
1631		m->m_pkthdr.rcvif = ifp;
1632
1633		/* Do RX checksumming if enabled */
1634
1635		if (ifp->if_capenable & IFCAP_RXCSUM) {
1636
1637			/* Check IP header checksum */
1638			if (rxstat & RL_RDESC_STAT_PROTOID)
1639				m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED;
1640			if (!(rxstat & RL_RDESC_STAT_IPSUMBAD))
1641				m->m_pkthdr.csum_flags |= CSUM_IP_VALID;
1642
1643			/* Check TCP/UDP checksum */
1644			if ((RL_TCPPKT(rxstat) &&
1645			    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
1646			    (RL_UDPPKT(rxstat) &&
1647			    !(rxstat & RL_RDESC_STAT_UDPSUMBAD))) {
1648				m->m_pkthdr.csum_flags |=
1649				    CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
1650				m->m_pkthdr.csum_data = 0xffff;
1651			}
1652		}
1653
1654		if (rxvlan & RL_RDESC_VLANCTL_TAG)
1655			VLAN_INPUT_TAG(ifp, m,
1656			    ntohs((rxvlan & RL_RDESC_VLANCTL_DATA)), continue);
1657		RL_UNLOCK(sc);
1658		(*ifp->if_input)(ifp, m);
1659		RL_LOCK(sc);
1660	}
1661
1662	/* Flush the RX DMA ring */
1663
1664	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1665	    sc->rl_ldata.rl_rx_list_map,
1666	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
1667
1668	sc->rl_ldata.rl_rx_prodidx = i;
1669}
1670
1671static void
1672re_txeof(sc)
1673	struct rl_softc		*sc;
1674{
1675	struct ifnet		*ifp;
1676	u_int32_t		txstat;
1677	int			idx;
1678
1679	ifp = &sc->arpcom.ac_if;
1680	idx = sc->rl_ldata.rl_tx_considx;
1681
1682	/* Invalidate the TX descriptor list */
1683
1684	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
1685	    sc->rl_ldata.rl_tx_list_map,
1686	    BUS_DMASYNC_POSTREAD);
1687
1688	while (idx != sc->rl_ldata.rl_tx_prodidx) {
1689
1690		txstat = le32toh(sc->rl_ldata.rl_tx_list[idx].rl_cmdstat);
1691		if (txstat & RL_TDESC_CMD_OWN)
1692			break;
1693
1694		/*
1695		 * We only stash mbufs in the last descriptor
1696		 * in a fragment chain, which also happens to
1697		 * be the only place where the TX status bits
1698		 * are valid.
1699		 */
1700
1701		if (txstat & RL_TDESC_CMD_EOF) {
1702			m_freem(sc->rl_ldata.rl_tx_mbuf[idx]);
1703			sc->rl_ldata.rl_tx_mbuf[idx] = NULL;
1704			bus_dmamap_unload(sc->rl_ldata.rl_mtag,
1705			    sc->rl_ldata.rl_tx_dmamap[idx]);
1706			if (txstat & (RL_TDESC_STAT_EXCESSCOL|
1707			    RL_TDESC_STAT_COLCNT))
1708				ifp->if_collisions++;
1709			if (txstat & RL_TDESC_STAT_TXERRSUM)
1710				ifp->if_oerrors++;
1711			else
1712				ifp->if_opackets++;
1713		}
1714		sc->rl_ldata.rl_tx_free++;
1715		RL_DESC_INC(idx);
1716	}
1717
1718	/* No changes made to the TX ring, so no flush needed */
1719
1720	if (idx != sc->rl_ldata.rl_tx_considx) {
1721		sc->rl_ldata.rl_tx_considx = idx;
1722		ifp->if_flags &= ~IFF_OACTIVE;
1723		ifp->if_timer = 0;
1724	}
1725
1726	/*
1727	 * If not all descriptors have been released reaped yet,
1728	 * reload the timer so that we will eventually get another
1729	 * interrupt that will cause us to re-enter this routine.
1730	 * This is done in case the transmitter has gone idle.
1731	 */
1732	if (sc->rl_ldata.rl_tx_free != RL_TX_DESC_CNT)
1733		CSR_WRITE_4(sc, RL_TIMERCNT, 1);
1734}
1735
1736static void
1737re_tick(xsc)
1738	void			*xsc;
1739{
1740	struct rl_softc		*sc;
1741
1742	sc = xsc;
1743	RL_LOCK(sc);
1744	re_tick_locked(sc);
1745	RL_UNLOCK(sc);
1746}
1747
1748static void
1749re_tick_locked(sc)
1750	struct rl_softc		*sc;
1751{
1752	struct mii_data		*mii;
1753
1754	RL_LOCK_ASSERT(sc);
1755
1756	mii = device_get_softc(sc->rl_miibus);
1757
1758	mii_tick(mii);
1759
1760	sc->rl_stat_ch = timeout(re_tick, sc, hz);
1761}
1762
1763#ifdef DEVICE_POLLING
1764static void
1765re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
1766{
1767	struct rl_softc *sc = ifp->if_softc;
1768
1769	RL_LOCK(sc);
1770	re_poll_locked(ifp, cmd, count);
1771	RL_UNLOCK(sc);
1772}
1773
1774static void
1775re_poll_locked(struct ifnet *ifp, enum poll_cmd cmd, int count)
1776{
1777	struct rl_softc *sc = ifp->if_softc;
1778
1779	RL_LOCK_ASSERT(sc);
1780
1781	if (!(ifp->if_capenable & IFCAP_POLLING)) {
1782		ether_poll_deregister(ifp);
1783		cmd = POLL_DEREGISTER;
1784	}
1785	if (cmd == POLL_DEREGISTER) { /* final call, enable interrupts */
1786		CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
1787		return;
1788	}
1789
1790	sc->rxcycles = count;
1791	re_rxeof(sc);
1792	re_txeof(sc);
1793
1794	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
1795		re_start_locked(ifp);
1796
1797	if (cmd == POLL_AND_CHECK_STATUS) { /* also check status register */
1798		u_int16_t       status;
1799
1800		status = CSR_READ_2(sc, RL_ISR);
1801		if (status == 0xffff)
1802			return;
1803		if (status)
1804			CSR_WRITE_2(sc, RL_ISR, status);
1805
1806		/*
1807		 * XXX check behaviour on receiver stalls.
1808		 */
1809
1810		if (status & RL_ISR_SYSTEM_ERR) {
1811			re_reset(sc);
1812			re_init_locked(sc);
1813		}
1814	}
1815}
1816#endif /* DEVICE_POLLING */
1817
1818static void
1819re_intr(arg)
1820	void			*arg;
1821{
1822	struct rl_softc		*sc;
1823	struct ifnet		*ifp;
1824	u_int16_t		status;
1825
1826	sc = arg;
1827
1828	RL_LOCK(sc);
1829
1830	ifp = &sc->arpcom.ac_if;
1831
1832	if (sc->suspended || !(ifp->if_flags & IFF_UP))
1833		goto done_locked;
1834
1835#ifdef DEVICE_POLLING
1836	if  (ifp->if_flags & IFF_POLLING)
1837		goto done_locked;
1838	if ((ifp->if_capenable & IFCAP_POLLING) &&
1839	    ether_poll_register(re_poll, ifp)) { /* ok, disable interrupts */
1840		CSR_WRITE_2(sc, RL_IMR, 0x0000);
1841		re_poll_locked(ifp, 0, 1);
1842		goto done_locked;
1843	}
1844#endif /* DEVICE_POLLING */
1845
1846	for (;;) {
1847
1848		status = CSR_READ_2(sc, RL_ISR);
1849		/* If the card has gone away the read returns 0xffff. */
1850		if (status == 0xffff)
1851			break;
1852		if (status)
1853			CSR_WRITE_2(sc, RL_ISR, status);
1854
1855		if ((status & RL_INTRS_CPLUS) == 0)
1856			break;
1857
1858		if ((status & RL_ISR_RX_OK) ||
1859		    (status & RL_ISR_RX_ERR))
1860			re_rxeof(sc);
1861
1862		if ((status & RL_ISR_TIMEOUT_EXPIRED) ||
1863		    (status & RL_ISR_TX_ERR) ||
1864		    (status & RL_ISR_TX_DESC_UNAVAIL))
1865			re_txeof(sc);
1866
1867		if (status & RL_ISR_SYSTEM_ERR) {
1868			re_reset(sc);
1869			re_init_locked(sc);
1870		}
1871
1872		if (status & RL_ISR_LINKCHG) {
1873			untimeout(re_tick, sc, sc->rl_stat_ch);
1874			re_tick_locked(sc);
1875		}
1876	}
1877
1878	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
1879		re_start_locked(ifp);
1880
1881done_locked:
1882	RL_UNLOCK(sc);
1883}
1884
1885static int
1886re_encap(sc, m_head, idx)
1887	struct rl_softc		*sc;
1888	struct mbuf		**m_head;
1889	int			*idx;
1890{
1891	struct mbuf		*m_new = NULL;
1892	struct rl_dmaload_arg	arg;
1893	bus_dmamap_t		map;
1894	int			error;
1895	struct m_tag		*mtag;
1896
1897	RL_LOCK_ASSERT(sc);
1898
1899	if (sc->rl_ldata.rl_tx_free <= 4)
1900		return (EFBIG);
1901
1902	/*
1903	 * Set up checksum offload. Note: checksum offload bits must
1904	 * appear in all descriptors of a multi-descriptor transmit
1905	 * attempt. This is according to testing done with an 8169
1906	 * chip. This is a requirement.
1907	 */
1908
1909	arg.rl_flags = 0;
1910
1911	if ((*m_head)->m_pkthdr.csum_flags & CSUM_IP)
1912		arg.rl_flags |= RL_TDESC_CMD_IPCSUM;
1913	if ((*m_head)->m_pkthdr.csum_flags & CSUM_TCP)
1914		arg.rl_flags |= RL_TDESC_CMD_TCPCSUM;
1915	if ((*m_head)->m_pkthdr.csum_flags & CSUM_UDP)
1916		arg.rl_flags |= RL_TDESC_CMD_UDPCSUM;
1917
1918	arg.sc = sc;
1919	arg.rl_idx = *idx;
1920	arg.rl_maxsegs = sc->rl_ldata.rl_tx_free;
1921	if (arg.rl_maxsegs > 4)
1922		arg.rl_maxsegs -= 4;
1923	arg.rl_ring = sc->rl_ldata.rl_tx_list;
1924
1925	map = sc->rl_ldata.rl_tx_dmamap[*idx];
1926	error = bus_dmamap_load_mbuf(sc->rl_ldata.rl_mtag, map,
1927	    *m_head, re_dma_map_desc, &arg, BUS_DMA_NOWAIT);
1928
1929	if (error && error != EFBIG) {
1930		printf("re%d: can't map mbuf (error %d)\n", sc->rl_unit, error);
1931		return (ENOBUFS);
1932	}
1933
1934	/* Too many segments to map, coalesce into a single mbuf */
1935
1936	if (error || arg.rl_maxsegs == 0) {
1937		m_new = m_defrag(*m_head, M_DONTWAIT);
1938		if (m_new == NULL)
1939			return (ENOBUFS);
1940		else
1941			*m_head = m_new;
1942
1943		arg.sc = sc;
1944		arg.rl_idx = *idx;
1945		arg.rl_maxsegs = sc->rl_ldata.rl_tx_free;
1946		arg.rl_ring = sc->rl_ldata.rl_tx_list;
1947
1948		error = bus_dmamap_load_mbuf(sc->rl_ldata.rl_mtag, map,
1949		    *m_head, re_dma_map_desc, &arg, BUS_DMA_NOWAIT);
1950		if (error) {
1951			printf("re%d: can't map mbuf (error %d)\n",
1952			    sc->rl_unit, error);
1953			return (EFBIG);
1954		}
1955	}
1956
1957	/*
1958	 * Insure that the map for this transmission
1959	 * is placed at the array index of the last descriptor
1960	 * in this chain.  (Swap last and first dmamaps.)
1961	 */
1962	sc->rl_ldata.rl_tx_dmamap[*idx] =
1963	    sc->rl_ldata.rl_tx_dmamap[arg.rl_idx];
1964	sc->rl_ldata.rl_tx_dmamap[arg.rl_idx] = map;
1965
1966	sc->rl_ldata.rl_tx_mbuf[arg.rl_idx] = *m_head;
1967	sc->rl_ldata.rl_tx_free -= arg.rl_maxsegs;
1968
1969	/*
1970	 * Set up hardware VLAN tagging. Note: vlan tag info must
1971	 * appear in the first descriptor of a multi-descriptor
1972	 * transmission attempt.
1973	 */
1974
1975	mtag = VLAN_OUTPUT_TAG(&sc->arpcom.ac_if, *m_head);
1976	if (mtag != NULL)
1977		sc->rl_ldata.rl_tx_list[*idx].rl_vlanctl =
1978		    htole32(htons(VLAN_TAG_VALUE(mtag)) | RL_TDESC_VLANCTL_TAG);
1979
1980	/* Transfer ownership of packet to the chip. */
1981
1982	sc->rl_ldata.rl_tx_list[arg.rl_idx].rl_cmdstat |=
1983	    htole32(RL_TDESC_CMD_OWN);
1984	if (*idx != arg.rl_idx)
1985		sc->rl_ldata.rl_tx_list[*idx].rl_cmdstat |=
1986		    htole32(RL_TDESC_CMD_OWN);
1987
1988	RL_DESC_INC(arg.rl_idx);
1989	*idx = arg.rl_idx;
1990
1991	return (0);
1992}
1993
1994static void
1995re_start(ifp)
1996	struct ifnet		*ifp;
1997{
1998	struct rl_softc		*sc;
1999
2000	sc = ifp->if_softc;
2001	RL_LOCK(sc);
2002	re_start_locked(ifp);
2003	RL_UNLOCK(sc);
2004}
2005
2006/*
2007 * Main transmit routine for C+ and gigE NICs.
2008 */
2009static void
2010re_start_locked(ifp)
2011	struct ifnet		*ifp;
2012{
2013	struct rl_softc		*sc;
2014	struct mbuf		*m_head = NULL;
2015	int			idx, queued = 0;
2016
2017	sc = ifp->if_softc;
2018
2019	RL_LOCK_ASSERT(sc);
2020
2021	idx = sc->rl_ldata.rl_tx_prodidx;
2022
2023	while (sc->rl_ldata.rl_tx_mbuf[idx] == NULL) {
2024		IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
2025		if (m_head == NULL)
2026			break;
2027
2028		if (re_encap(sc, &m_head, &idx)) {
2029			IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
2030			ifp->if_flags |= IFF_OACTIVE;
2031			break;
2032		}
2033
2034		/*
2035		 * If there's a BPF listener, bounce a copy of this frame
2036		 * to him.
2037		 */
2038		BPF_MTAP(ifp, m_head);
2039
2040		queued++;
2041	}
2042
2043	if (queued == 0)
2044		return;
2045
2046	/* Flush the TX descriptors */
2047
2048	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2049	    sc->rl_ldata.rl_tx_list_map,
2050	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2051
2052	sc->rl_ldata.rl_tx_prodidx = idx;
2053
2054	/*
2055	 * RealTek put the TX poll request register in a different
2056	 * location on the 8169 gigE chip. I don't know why.
2057	 */
2058
2059	if (sc->rl_type == RL_8169)
2060		CSR_WRITE_2(sc, RL_GTXSTART, RL_TXSTART_START);
2061	else
2062		CSR_WRITE_2(sc, RL_TXSTART, RL_TXSTART_START);
2063
2064	/*
2065	 * Use the countdown timer for interrupt moderation.
2066	 * 'TX done' interrupts are disabled. Instead, we reset the
2067	 * countdown timer, which will begin counting until it hits
2068	 * the value in the TIMERINT register, and then trigger an
2069	 * interrupt. Each time we write to the TIMERCNT register,
2070	 * the timer count is reset to 0.
2071	 */
2072	CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2073
2074	/*
2075	 * Set a timeout in case the chip goes out to lunch.
2076	 */
2077	ifp->if_timer = 5;
2078}
2079
2080static void
2081re_init(xsc)
2082	void			*xsc;
2083{
2084	struct rl_softc		*sc = xsc;
2085
2086	RL_LOCK(sc);
2087	re_init_locked(sc);
2088	RL_UNLOCK(sc);
2089}
2090
2091static void
2092re_init_locked(sc)
2093	struct rl_softc		*sc;
2094{
2095	struct ifnet		*ifp = &sc->arpcom.ac_if;
2096	struct mii_data		*mii;
2097	u_int32_t		rxcfg = 0;
2098
2099	RL_LOCK_ASSERT(sc);
2100
2101	mii = device_get_softc(sc->rl_miibus);
2102
2103	/*
2104	 * Cancel pending I/O and free all RX/TX buffers.
2105	 */
2106	re_stop(sc);
2107
2108	/*
2109	 * Enable C+ RX and TX mode, as well as VLAN stripping and
2110	 * RX checksum offload. We must configure the C+ register
2111	 * before all others.
2112	 */
2113	CSR_WRITE_2(sc, RL_CPLUS_CMD, RL_CPLUSCMD_RXENB|
2114	    RL_CPLUSCMD_TXENB|RL_CPLUSCMD_PCI_MRW|
2115	    RL_CPLUSCMD_VLANSTRIP|
2116	    (ifp->if_capenable & IFCAP_RXCSUM ?
2117	    RL_CPLUSCMD_RXCSUM_ENB : 0));
2118
2119	/*
2120	 * Init our MAC address.  Even though the chipset
2121	 * documentation doesn't mention it, we need to enter "Config
2122	 * register write enable" mode to modify the ID registers.
2123	 */
2124	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG);
2125	CSR_WRITE_STREAM_4(sc, RL_IDR0,
2126	    *(u_int32_t *)(&sc->arpcom.ac_enaddr[0]));
2127	CSR_WRITE_STREAM_4(sc, RL_IDR4,
2128	    *(u_int32_t *)(&sc->arpcom.ac_enaddr[4]));
2129	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
2130
2131	/*
2132	 * For C+ mode, initialize the RX descriptors and mbufs.
2133	 */
2134	re_rx_list_init(sc);
2135	re_tx_list_init(sc);
2136
2137	/*
2138	 * Enable transmit and receive.
2139	 */
2140	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
2141
2142	/*
2143	 * Set the initial TX and RX configuration.
2144	 */
2145	if (sc->rl_testmode) {
2146		if (sc->rl_type == RL_8169)
2147			CSR_WRITE_4(sc, RL_TXCFG,
2148			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON);
2149		else
2150			CSR_WRITE_4(sc, RL_TXCFG,
2151			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON_CPLUS);
2152	} else
2153		CSR_WRITE_4(sc, RL_TXCFG, RL_TXCFG_CONFIG);
2154	CSR_WRITE_4(sc, RL_RXCFG, RL_RXCFG_CONFIG);
2155
2156	/* Set the individual bit to receive frames for this host only. */
2157	rxcfg = CSR_READ_4(sc, RL_RXCFG);
2158	rxcfg |= RL_RXCFG_RX_INDIV;
2159
2160	/* If we want promiscuous mode, set the allframes bit. */
2161	if (ifp->if_flags & IFF_PROMISC)
2162		rxcfg |= RL_RXCFG_RX_ALLPHYS;
2163	else
2164		rxcfg &= ~RL_RXCFG_RX_ALLPHYS;
2165	CSR_WRITE_4(sc, RL_RXCFG, rxcfg);
2166
2167	/*
2168	 * Set capture broadcast bit to capture broadcast frames.
2169	 */
2170	if (ifp->if_flags & IFF_BROADCAST)
2171		rxcfg |= RL_RXCFG_RX_BROAD;
2172	else
2173		rxcfg &= ~RL_RXCFG_RX_BROAD;
2174	CSR_WRITE_4(sc, RL_RXCFG, rxcfg);
2175
2176	/*
2177	 * Program the multicast filter, if necessary.
2178	 */
2179	re_setmulti(sc);
2180
2181#ifdef DEVICE_POLLING
2182	/*
2183	 * Disable interrupts if we are polling.
2184	 */
2185	if (ifp->if_flags & IFF_POLLING)
2186		CSR_WRITE_2(sc, RL_IMR, 0);
2187	else	/* otherwise ... */
2188#endif /* DEVICE_POLLING */
2189	/*
2190	 * Enable interrupts.
2191	 */
2192	if (sc->rl_testmode)
2193		CSR_WRITE_2(sc, RL_IMR, 0);
2194	else
2195		CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2196
2197	/* Set initial TX threshold */
2198	sc->rl_txthresh = RL_TX_THRESH_INIT;
2199
2200	/* Start RX/TX process. */
2201	CSR_WRITE_4(sc, RL_MISSEDPKT, 0);
2202#ifdef notdef
2203	/* Enable receiver and transmitter. */
2204	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
2205#endif
2206	/*
2207	 * Load the addresses of the RX and TX lists into the chip.
2208	 */
2209
2210	CSR_WRITE_4(sc, RL_RXLIST_ADDR_HI,
2211	    RL_ADDR_HI(sc->rl_ldata.rl_rx_list_addr));
2212	CSR_WRITE_4(sc, RL_RXLIST_ADDR_LO,
2213	    RL_ADDR_LO(sc->rl_ldata.rl_rx_list_addr));
2214
2215	CSR_WRITE_4(sc, RL_TXLIST_ADDR_HI,
2216	    RL_ADDR_HI(sc->rl_ldata.rl_tx_list_addr));
2217	CSR_WRITE_4(sc, RL_TXLIST_ADDR_LO,
2218	    RL_ADDR_LO(sc->rl_ldata.rl_tx_list_addr));
2219
2220	CSR_WRITE_1(sc, RL_EARLY_TX_THRESH, 16);
2221
2222	/*
2223	 * Initialize the timer interrupt register so that
2224	 * a timer interrupt will be generated once the timer
2225	 * reaches a certain number of ticks. The timer is
2226	 * reloaded on each transmit. This gives us TX interrupt
2227	 * moderation, which dramatically improves TX frame rate.
2228	 */
2229	if (sc->rl_type == RL_8169)
2230		CSR_WRITE_4(sc, RL_TIMERINT_8169, 0x800);
2231	else
2232		CSR_WRITE_4(sc, RL_TIMERINT, 0x400);
2233
2234	/*
2235	 * For 8169 gigE NICs, set the max allowed RX packet
2236	 * size so we can receive jumbo frames.
2237	 */
2238	if (sc->rl_type == RL_8169)
2239		CSR_WRITE_2(sc, RL_MAXRXPKTLEN, 16383);
2240
2241	if (sc->rl_testmode)
2242		return;
2243
2244	mii_mediachg(mii);
2245
2246	CSR_WRITE_1(sc, RL_CFG1, RL_CFG1_DRVLOAD|RL_CFG1_FULLDUPLEX);
2247
2248	ifp->if_flags |= IFF_RUNNING;
2249	ifp->if_flags &= ~IFF_OACTIVE;
2250
2251	sc->rl_stat_ch = timeout(re_tick, sc, hz);
2252}
2253
2254/*
2255 * Set media options.
2256 */
2257static int
2258re_ifmedia_upd(ifp)
2259	struct ifnet		*ifp;
2260{
2261	struct rl_softc		*sc;
2262	struct mii_data		*mii;
2263
2264	sc = ifp->if_softc;
2265	mii = device_get_softc(sc->rl_miibus);
2266	mii_mediachg(mii);
2267
2268	return (0);
2269}
2270
2271/*
2272 * Report current media status.
2273 */
2274static void
2275re_ifmedia_sts(ifp, ifmr)
2276	struct ifnet		*ifp;
2277	struct ifmediareq	*ifmr;
2278{
2279	struct rl_softc		*sc;
2280	struct mii_data		*mii;
2281
2282	sc = ifp->if_softc;
2283	mii = device_get_softc(sc->rl_miibus);
2284
2285	mii_pollstat(mii);
2286	ifmr->ifm_active = mii->mii_media_active;
2287	ifmr->ifm_status = mii->mii_media_status;
2288}
2289
2290static int
2291re_ioctl(ifp, command, data)
2292	struct ifnet		*ifp;
2293	u_long			command;
2294	caddr_t			data;
2295{
2296	struct rl_softc		*sc = ifp->if_softc;
2297	struct ifreq		*ifr = (struct ifreq *) data;
2298	struct mii_data		*mii;
2299	int			error = 0;
2300
2301	switch (command) {
2302	case SIOCSIFMTU:
2303		if (ifr->ifr_mtu > RL_JUMBO_MTU)
2304			error = EINVAL;
2305		ifp->if_mtu = ifr->ifr_mtu;
2306		break;
2307	case SIOCSIFFLAGS:
2308		RL_LOCK(sc);
2309		if (ifp->if_flags & IFF_UP)
2310			re_init_locked(sc);
2311		else if (ifp->if_flags & IFF_RUNNING)
2312			re_stop(sc);
2313		RL_UNLOCK(sc);
2314		error = 0;
2315		break;
2316	case SIOCADDMULTI:
2317	case SIOCDELMULTI:
2318		RL_LOCK(sc);
2319		re_setmulti(sc);
2320		RL_UNLOCK(sc);
2321		error = 0;
2322		break;
2323	case SIOCGIFMEDIA:
2324	case SIOCSIFMEDIA:
2325		mii = device_get_softc(sc->rl_miibus);
2326		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, command);
2327		break;
2328	case SIOCSIFCAP:
2329		ifp->if_capenable &= ~(IFCAP_HWCSUM | IFCAP_POLLING);
2330		ifp->if_capenable |=
2331		    ifr->ifr_reqcap & (IFCAP_HWCSUM | IFCAP_POLLING);
2332		if (ifp->if_capenable & IFCAP_TXCSUM)
2333			ifp->if_hwassist = RE_CSUM_FEATURES;
2334		else
2335			ifp->if_hwassist = 0;
2336		if (ifp->if_flags & IFF_RUNNING)
2337			re_init(sc);
2338		break;
2339	default:
2340		error = ether_ioctl(ifp, command, data);
2341		break;
2342	}
2343
2344	return (error);
2345}
2346
2347static void
2348re_watchdog(ifp)
2349	struct ifnet		*ifp;
2350{
2351	struct rl_softc		*sc;
2352
2353	sc = ifp->if_softc;
2354	RL_LOCK(sc);
2355	printf("re%d: watchdog timeout\n", sc->rl_unit);
2356	ifp->if_oerrors++;
2357
2358	re_txeof(sc);
2359	re_rxeof(sc);
2360	re_init_locked(sc);
2361
2362	RL_UNLOCK(sc);
2363}
2364
2365/*
2366 * Stop the adapter and free any mbufs allocated to the
2367 * RX and TX lists.
2368 */
2369static void
2370re_stop(sc)
2371	struct rl_softc		*sc;
2372{
2373	register int		i;
2374	struct ifnet		*ifp;
2375
2376	RL_LOCK_ASSERT(sc);
2377
2378	ifp = &sc->arpcom.ac_if;
2379	ifp->if_timer = 0;
2380
2381	untimeout(re_tick, sc, sc->rl_stat_ch);
2382	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
2383#ifdef DEVICE_POLLING
2384	ether_poll_deregister(ifp);
2385#endif /* DEVICE_POLLING */
2386
2387	CSR_WRITE_1(sc, RL_COMMAND, 0x00);
2388	CSR_WRITE_2(sc, RL_IMR, 0x0000);
2389
2390	if (sc->rl_head != NULL) {
2391		m_freem(sc->rl_head);
2392		sc->rl_head = sc->rl_tail = NULL;
2393	}
2394
2395	/* Free the TX list buffers. */
2396
2397	for (i = 0; i < RL_TX_DESC_CNT; i++) {
2398		if (sc->rl_ldata.rl_tx_mbuf[i] != NULL) {
2399			bus_dmamap_unload(sc->rl_ldata.rl_mtag,
2400			    sc->rl_ldata.rl_tx_dmamap[i]);
2401			m_freem(sc->rl_ldata.rl_tx_mbuf[i]);
2402			sc->rl_ldata.rl_tx_mbuf[i] = NULL;
2403		}
2404	}
2405
2406	/* Free the RX list buffers. */
2407
2408	for (i = 0; i < RL_RX_DESC_CNT; i++) {
2409		if (sc->rl_ldata.rl_rx_mbuf[i] != NULL) {
2410			bus_dmamap_unload(sc->rl_ldata.rl_mtag,
2411			    sc->rl_ldata.rl_rx_dmamap[i]);
2412			m_freem(sc->rl_ldata.rl_rx_mbuf[i]);
2413			sc->rl_ldata.rl_rx_mbuf[i] = NULL;
2414		}
2415	}
2416}
2417
2418/*
2419 * Device suspend routine.  Stop the interface and save some PCI
2420 * settings in case the BIOS doesn't restore them properly on
2421 * resume.
2422 */
2423static int
2424re_suspend(dev)
2425	device_t		dev;
2426{
2427	struct rl_softc		*sc;
2428
2429	sc = device_get_softc(dev);
2430
2431	RL_LOCK(sc);
2432	re_stop(sc);
2433	sc->suspended = 1;
2434	RL_UNLOCK(sc);
2435
2436	return (0);
2437}
2438
2439/*
2440 * Device resume routine.  Restore some PCI settings in case the BIOS
2441 * doesn't, re-enable busmastering, and restart the interface if
2442 * appropriate.
2443 */
2444static int
2445re_resume(dev)
2446	device_t		dev;
2447{
2448	struct rl_softc		*sc;
2449	struct ifnet		*ifp;
2450
2451	sc = device_get_softc(dev);
2452
2453	RL_LOCK(sc);
2454
2455	ifp = &sc->arpcom.ac_if;
2456
2457	/* reinitialize interface if necessary */
2458	if (ifp->if_flags & IFF_UP)
2459		re_init_locked(sc);
2460
2461	sc->suspended = 0;
2462	RL_UNLOCK(sc);
2463
2464	return (0);
2465}
2466
2467/*
2468 * Stop all chip I/O so that the kernel's probe routines don't
2469 * get confused by errant DMAs when rebooting.
2470 */
2471static void
2472re_shutdown(dev)
2473	device_t		dev;
2474{
2475	struct rl_softc		*sc;
2476
2477	sc = device_get_softc(dev);
2478
2479	RL_LOCK(sc);
2480	re_stop(sc);
2481	RL_UNLOCK(sc);
2482}
2483