if_re.c revision 142890
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 142890 2005-03-01 08:58:06Z imp $");
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	IF_HANDOFF(&ifp->if_snd, m0, ifp);
721	RL_LOCK(sc);
722	m0 = NULL;
723
724	/* Wait for it to propagate through the chip */
725
726	DELAY(100000);
727	for (i = 0; i < RL_TIMEOUT; i++) {
728		status = CSR_READ_2(sc, RL_ISR);
729		if ((status & (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK)) ==
730		    (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK))
731			break;
732		DELAY(10);
733	}
734
735	if (i == RL_TIMEOUT) {
736		printf("re%d: diagnostic failed, failed to receive packet "
737		    "in loopback mode\n", sc->rl_unit);
738		error = EIO;
739		goto done;
740	}
741
742	/*
743	 * The packet should have been dumped into the first
744	 * entry in the RX DMA ring. Grab it from there.
745	 */
746
747	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
748	    sc->rl_ldata.rl_rx_list_map,
749	    BUS_DMASYNC_POSTREAD);
750	bus_dmamap_sync(sc->rl_ldata.rl_mtag,
751	    sc->rl_ldata.rl_rx_dmamap[0],
752	    BUS_DMASYNC_POSTWRITE);
753	bus_dmamap_unload(sc->rl_ldata.rl_mtag,
754	    sc->rl_ldata.rl_rx_dmamap[0]);
755
756	m0 = sc->rl_ldata.rl_rx_mbuf[0];
757	sc->rl_ldata.rl_rx_mbuf[0] = NULL;
758	eh = mtod(m0, struct ether_header *);
759
760	cur_rx = &sc->rl_ldata.rl_rx_list[0];
761	total_len = RL_RXBYTES(cur_rx);
762	rxstat = le32toh(cur_rx->rl_cmdstat);
763
764	if (total_len != ETHER_MIN_LEN) {
765		printf("re%d: diagnostic failed, received short packet\n",
766		    sc->rl_unit);
767		error = EIO;
768		goto done;
769	}
770
771	/* Test that the received packet data matches what we sent. */
772
773	if (bcmp((char *)&eh->ether_dhost, (char *)&dst, ETHER_ADDR_LEN) ||
774	    bcmp((char *)&eh->ether_shost, (char *)&src, ETHER_ADDR_LEN) ||
775	    ntohs(eh->ether_type) != ETHERTYPE_IP) {
776		printf("re%d: WARNING, DMA FAILURE!\n", sc->rl_unit);
777		printf("re%d: expected TX data: %6D/%6D/0x%x\n", sc->rl_unit,
778		    dst, ":", src, ":", ETHERTYPE_IP);
779		printf("re%d: received RX data: %6D/%6D/0x%x\n", sc->rl_unit,
780		    eh->ether_dhost, ":",  eh->ether_shost, ":",
781		    ntohs(eh->ether_type));
782		printf("re%d: You may have a defective 32-bit NIC plugged "
783		    "into a 64-bit PCI slot.\n", sc->rl_unit);
784		printf("re%d: Please re-install the NIC in a 32-bit slot "
785		    "for proper operation.\n", sc->rl_unit);
786		printf("re%d: Read the re(4) man page for more details.\n",
787		    sc->rl_unit);
788		error = EIO;
789	}
790
791done:
792	/* Turn interface off, release resources */
793
794	sc->rl_testmode = 0;
795	ifp->if_flags &= ~IFF_PROMISC;
796	re_stop(sc);
797	if (m0 != NULL)
798		m_freem(m0);
799
800	RL_UNLOCK(sc);
801
802	return (error);
803}
804
805/*
806 * Probe for a RealTek 8139C+/8169/8110 chip. Check the PCI vendor and device
807 * IDs against our list and return a device name if we find a match.
808 */
809static int
810re_probe(dev)
811	device_t		dev;
812{
813	struct rl_type		*t;
814	struct rl_softc		*sc;
815	int			rid;
816	u_int32_t		hwrev;
817
818	t = re_devs;
819	sc = device_get_softc(dev);
820
821	while (t->rl_name != NULL) {
822		if ((pci_get_vendor(dev) == t->rl_vid) &&
823		    (pci_get_device(dev) == t->rl_did)) {
824
825			/*
826			 * Temporarily map the I/O space
827			 * so we can read the chip ID register.
828			 */
829			rid = RL_RID;
830			sc->rl_res = bus_alloc_resource_any(dev, RL_RES, &rid,
831			    RF_ACTIVE);
832			if (sc->rl_res == NULL) {
833				device_printf(dev,
834				    "couldn't map ports/memory\n");
835				return (ENXIO);
836			}
837			sc->rl_btag = rman_get_bustag(sc->rl_res);
838			sc->rl_bhandle = rman_get_bushandle(sc->rl_res);
839			hwrev = CSR_READ_4(sc, RL_TXCFG) & RL_TXCFG_HWREV;
840			bus_release_resource(dev, RL_RES,
841			    RL_RID, sc->rl_res);
842			if (t->rl_basetype == hwrev) {
843				device_set_desc(dev, t->rl_name);
844				return (BUS_PROBE_DEFAULT);
845			}
846		}
847		t++;
848	}
849
850	return (ENXIO);
851}
852
853/*
854 * This routine takes the segment list provided as the result of
855 * a bus_dma_map_load() operation and assigns the addresses/lengths
856 * to RealTek DMA descriptors. This can be called either by the RX
857 * code or the TX code. In the RX case, we'll probably wind up mapping
858 * at most one segment. For the TX case, there could be any number of
859 * segments since TX packets may span multiple mbufs. In either case,
860 * if the number of segments is larger than the rl_maxsegs limit
861 * specified by the caller, we abort the mapping operation. Sadly,
862 * whoever designed the buffer mapping API did not provide a way to
863 * return an error from here, so we have to fake it a bit.
864 */
865
866static void
867re_dma_map_desc(arg, segs, nseg, mapsize, error)
868	void			*arg;
869	bus_dma_segment_t	*segs;
870	int			nseg;
871	bus_size_t		mapsize;
872	int			error;
873{
874	struct rl_dmaload_arg	*ctx;
875	struct rl_desc		*d = NULL;
876	int			i = 0, idx;
877
878	if (error)
879		return;
880
881	ctx = arg;
882
883	/* Signal error to caller if there's too many segments */
884	if (nseg > ctx->rl_maxsegs) {
885		ctx->rl_maxsegs = 0;
886		return;
887	}
888
889	/*
890	 * Map the segment array into descriptors. Note that we set the
891	 * start-of-frame and end-of-frame markers for either TX or RX, but
892	 * they really only have meaning in the TX case. (In the RX case,
893	 * it's the chip that tells us where packets begin and end.)
894	 * We also keep track of the end of the ring and set the
895	 * end-of-ring bits as needed, and we set the ownership bits
896	 * in all except the very first descriptor. (The caller will
897	 * set this descriptor later when it start transmission or
898	 * reception.)
899	 */
900	idx = ctx->rl_idx;
901	for (;;) {
902		u_int32_t		cmdstat;
903		d = &ctx->rl_ring[idx];
904		if (le32toh(d->rl_cmdstat) & RL_RDESC_STAT_OWN) {
905			ctx->rl_maxsegs = 0;
906			return;
907		}
908		cmdstat = segs[i].ds_len;
909		d->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[i].ds_addr));
910		d->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[i].ds_addr));
911		if (i == 0)
912			cmdstat |= RL_TDESC_CMD_SOF;
913		else
914			cmdstat |= RL_TDESC_CMD_OWN;
915		if (idx == (RL_RX_DESC_CNT - 1))
916			cmdstat |= RL_TDESC_CMD_EOR;
917		d->rl_cmdstat = htole32(cmdstat | ctx->rl_flags);
918		i++;
919		if (i == nseg)
920			break;
921		RL_DESC_INC(idx);
922	}
923
924	d->rl_cmdstat |= htole32(RL_TDESC_CMD_EOF);
925	ctx->rl_maxsegs = nseg;
926	ctx->rl_idx = idx;
927}
928
929/*
930 * Map a single buffer address.
931 */
932
933static void
934re_dma_map_addr(arg, segs, nseg, error)
935	void			*arg;
936	bus_dma_segment_t	*segs;
937	int			nseg;
938	int			error;
939{
940	u_int32_t		*addr;
941
942	if (error)
943		return;
944
945	KASSERT(nseg == 1, ("too many DMA segments, %d should be 1", nseg));
946	addr = arg;
947	*addr = segs->ds_addr;
948}
949
950static int
951re_allocmem(dev, sc)
952	device_t		dev;
953	struct rl_softc		*sc;
954{
955	int			error;
956	int			nseg;
957	int			i;
958
959	/*
960	 * Allocate map for RX mbufs.
961	 */
962	nseg = 32;
963	error = bus_dma_tag_create(sc->rl_parent_tag, ETHER_ALIGN, 0,
964	    BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
965	    NULL, MCLBYTES * nseg, nseg, MCLBYTES, BUS_DMA_ALLOCNOW,
966	    NULL, NULL, &sc->rl_ldata.rl_mtag);
967	if (error) {
968		device_printf(dev, "could not allocate dma tag\n");
969		return (ENOMEM);
970	}
971
972	/*
973	 * Allocate map for TX descriptor list.
974	 */
975	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
976	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
977	    NULL, RL_TX_LIST_SZ, 1, RL_TX_LIST_SZ, BUS_DMA_ALLOCNOW,
978	    NULL, NULL, &sc->rl_ldata.rl_tx_list_tag);
979	if (error) {
980		device_printf(dev, "could not allocate dma tag\n");
981		return (ENOMEM);
982	}
983
984	/* Allocate DMA'able memory for the TX ring */
985
986	error = bus_dmamem_alloc(sc->rl_ldata.rl_tx_list_tag,
987	    (void **)&sc->rl_ldata.rl_tx_list, BUS_DMA_NOWAIT | BUS_DMA_ZERO,
988	    &sc->rl_ldata.rl_tx_list_map);
989	if (error)
990		return (ENOMEM);
991
992	/* Load the map for the TX ring. */
993
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	     RL_TX_LIST_SZ, re_dma_map_addr,
997	     &sc->rl_ldata.rl_tx_list_addr, BUS_DMA_NOWAIT);
998
999	/* Create DMA maps for TX buffers */
1000
1001	for (i = 0; i < RL_TX_DESC_CNT; i++) {
1002		error = bus_dmamap_create(sc->rl_ldata.rl_mtag, 0,
1003			    &sc->rl_ldata.rl_tx_dmamap[i]);
1004		if (error) {
1005			device_printf(dev, "can't create DMA map for TX\n");
1006			return (ENOMEM);
1007		}
1008	}
1009
1010	/*
1011	 * Allocate map for RX descriptor list.
1012	 */
1013	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
1014	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1015	    NULL, RL_RX_LIST_SZ, 1, RL_RX_LIST_SZ, BUS_DMA_ALLOCNOW,
1016	    NULL, NULL, &sc->rl_ldata.rl_rx_list_tag);
1017	if (error) {
1018		device_printf(dev, "could not allocate dma tag\n");
1019		return (ENOMEM);
1020	}
1021
1022	/* Allocate DMA'able memory for the RX ring */
1023
1024	error = bus_dmamem_alloc(sc->rl_ldata.rl_rx_list_tag,
1025	    (void **)&sc->rl_ldata.rl_rx_list, BUS_DMA_NOWAIT | BUS_DMA_ZERO,
1026	    &sc->rl_ldata.rl_rx_list_map);
1027	if (error)
1028		return (ENOMEM);
1029
1030	/* Load the map for the RX ring. */
1031
1032	error = bus_dmamap_load(sc->rl_ldata.rl_rx_list_tag,
1033	     sc->rl_ldata.rl_rx_list_map, sc->rl_ldata.rl_rx_list,
1034	     RL_RX_LIST_SZ, re_dma_map_addr,
1035	     &sc->rl_ldata.rl_rx_list_addr, BUS_DMA_NOWAIT);
1036
1037	/* Create DMA maps for RX buffers */
1038
1039	for (i = 0; i < RL_RX_DESC_CNT; i++) {
1040		error = bus_dmamap_create(sc->rl_ldata.rl_mtag, 0,
1041			    &sc->rl_ldata.rl_rx_dmamap[i]);
1042		if (error) {
1043			device_printf(dev, "can't create DMA map for RX\n");
1044			return (ENOMEM);
1045		}
1046	}
1047
1048	return (0);
1049}
1050
1051/*
1052 * Attach the interface. Allocate softc structures, do ifmedia
1053 * setup and ethernet/BPF attach.
1054 */
1055static int
1056re_attach(dev)
1057	device_t		dev;
1058{
1059	u_char			eaddr[ETHER_ADDR_LEN];
1060	u_int16_t		as[3];
1061	struct rl_softc		*sc;
1062	struct ifnet		*ifp;
1063	struct rl_hwrev		*hw_rev;
1064	int			hwrev;
1065	u_int16_t		re_did = 0;
1066	int			unit, error = 0, rid, i;
1067
1068	sc = device_get_softc(dev);
1069	unit = device_get_unit(dev);
1070
1071	mtx_init(&sc->rl_mtx, device_get_nameunit(dev), MTX_NETWORK_LOCK,
1072	    MTX_DEF);
1073	/*
1074	 * Map control/status registers.
1075	 */
1076	pci_enable_busmaster(dev);
1077
1078	rid = RL_RID;
1079	sc->rl_res = bus_alloc_resource_any(dev, RL_RES, &rid,
1080	    RF_ACTIVE);
1081
1082	if (sc->rl_res == NULL) {
1083		printf ("re%d: couldn't map ports/memory\n", unit);
1084		error = ENXIO;
1085		goto fail;
1086	}
1087
1088	sc->rl_btag = rman_get_bustag(sc->rl_res);
1089	sc->rl_bhandle = rman_get_bushandle(sc->rl_res);
1090
1091	/* Allocate interrupt */
1092	rid = 0;
1093	sc->rl_irq = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
1094	    RF_SHAREABLE | RF_ACTIVE);
1095
1096	if (sc->rl_irq == NULL) {
1097		printf("re%d: couldn't map interrupt\n", unit);
1098		error = ENXIO;
1099		goto fail;
1100	}
1101
1102	/* Reset the adapter. */
1103	RL_LOCK(sc);
1104	re_reset(sc);
1105	RL_UNLOCK(sc);
1106
1107	hw_rev = re_hwrevs;
1108	hwrev = CSR_READ_4(sc, RL_TXCFG) & RL_TXCFG_HWREV;
1109	while (hw_rev->rl_desc != NULL) {
1110		if (hw_rev->rl_rev == hwrev) {
1111			sc->rl_type = hw_rev->rl_type;
1112			break;
1113		}
1114		hw_rev++;
1115	}
1116
1117	if (sc->rl_type == RL_8169) {
1118
1119		/* Set RX length mask */
1120
1121		sc->rl_rxlenmask = RL_RDESC_STAT_GFRAGLEN;
1122
1123		/* Force station address autoload from the EEPROM */
1124
1125		CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_AUTOLOAD);
1126		for (i = 0; i < RL_TIMEOUT; i++) {
1127			if (!(CSR_READ_1(sc, RL_EECMD) & RL_EEMODE_AUTOLOAD))
1128				break;
1129			DELAY(100);
1130		}
1131		if (i == RL_TIMEOUT)
1132			printf ("re%d: eeprom autoload timed out\n", unit);
1133
1134			for (i = 0; i < ETHER_ADDR_LEN; i++)
1135				eaddr[i] = CSR_READ_1(sc, RL_IDR0 + i);
1136	} else {
1137
1138		/* Set RX length mask */
1139
1140		sc->rl_rxlenmask = RL_RDESC_STAT_FRAGLEN;
1141
1142		sc->rl_eecmd_read = RL_EECMD_READ_6BIT;
1143		re_read_eeprom(sc, (caddr_t)&re_did, 0, 1, 0);
1144		if (re_did != 0x8129)
1145			sc->rl_eecmd_read = RL_EECMD_READ_8BIT;
1146
1147		/*
1148		 * Get station address from the EEPROM.
1149		 */
1150		re_read_eeprom(sc, (caddr_t)as, RL_EE_EADDR, 3, 0);
1151		for (i = 0; i < 3; i++) {
1152			eaddr[(i * 2) + 0] = as[i] & 0xff;
1153			eaddr[(i * 2) + 1] = as[i] >> 8;
1154		}
1155	}
1156
1157	sc->rl_unit = unit;
1158	bcopy(eaddr, (char *)&sc->arpcom.ac_enaddr, ETHER_ADDR_LEN);
1159
1160	/*
1161	 * Allocate the parent bus DMA tag appropriate for PCI.
1162	 */
1163#define RL_NSEG_NEW 32
1164	error = bus_dma_tag_create(NULL,	/* parent */
1165			1, 0,			/* alignment, boundary */
1166			BUS_SPACE_MAXADDR_32BIT,/* lowaddr */
1167			BUS_SPACE_MAXADDR,	/* highaddr */
1168			NULL, NULL,		/* filter, filterarg */
1169			MAXBSIZE, RL_NSEG_NEW,	/* maxsize, nsegments */
1170			BUS_SPACE_MAXSIZE_32BIT,/* maxsegsize */
1171			BUS_DMA_ALLOCNOW,	/* flags */
1172			NULL, NULL,		/* lockfunc, lockarg */
1173			&sc->rl_parent_tag);
1174	if (error)
1175		goto fail;
1176
1177	error = re_allocmem(dev, sc);
1178
1179	if (error)
1180		goto fail;
1181
1182	/* Do MII setup */
1183	if (mii_phy_probe(dev, &sc->rl_miibus,
1184	    re_ifmedia_upd, re_ifmedia_sts)) {
1185		printf("re%d: MII without any phy!\n", sc->rl_unit);
1186		error = ENXIO;
1187		goto fail;
1188	}
1189
1190	ifp = &sc->arpcom.ac_if;
1191	ifp->if_softc = sc;
1192	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1193	ifp->if_mtu = ETHERMTU;
1194	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1195	ifp->if_ioctl = re_ioctl;
1196	ifp->if_capabilities = IFCAP_VLAN_MTU;
1197	ifp->if_start = re_start;
1198	ifp->if_hwassist = /*RE_CSUM_FEATURES*/0;
1199	ifp->if_capabilities |= IFCAP_HWCSUM|IFCAP_VLAN_HWTAGGING;
1200#ifdef DEVICE_POLLING
1201	ifp->if_capabilities |= IFCAP_POLLING;
1202#endif
1203	ifp->if_watchdog = re_watchdog;
1204	ifp->if_init = re_init;
1205	if (sc->rl_type == RL_8169)
1206		ifp->if_baudrate = 1000000000;
1207	else
1208		ifp->if_baudrate = 100000000;
1209	ifp->if_snd.ifq_maxlen = RL_IFQ_MAXLEN;
1210	ifp->if_capenable = ifp->if_capabilities & ~IFCAP_HWCSUM;
1211
1212	callout_handle_init(&sc->rl_stat_ch);
1213
1214	/*
1215	 * Call MI attach routine.
1216	 */
1217	ether_ifattach(ifp, eaddr);
1218
1219	/* Perform hardware diagnostic. */
1220	error = re_diag(sc);
1221
1222	if (error) {
1223		printf("re%d: attach aborted due to hardware diag failure\n",
1224		    unit);
1225		ether_ifdetach(ifp);
1226		goto fail;
1227	}
1228
1229	/* Hook interrupt last to avoid having to lock softc */
1230	error = bus_setup_intr(dev, sc->rl_irq, INTR_TYPE_NET | INTR_MPSAFE,
1231	    re_intr, sc, &sc->rl_intrhand);
1232	if (error) {
1233		printf("re%d: couldn't set up irq\n", unit);
1234		ether_ifdetach(ifp);
1235	}
1236
1237fail:
1238	if (error)
1239		re_detach(dev);
1240
1241	return (error);
1242}
1243
1244/*
1245 * Shutdown hardware and free up resources. This can be called any
1246 * time after the mutex has been initialized. It is called in both
1247 * the error case in attach and the normal detach case so it needs
1248 * to be careful about only freeing resources that have actually been
1249 * allocated.
1250 */
1251static int
1252re_detach(dev)
1253	device_t		dev;
1254{
1255	struct rl_softc		*sc;
1256	struct ifnet		*ifp;
1257	int			i;
1258	int			attached;
1259
1260	sc = device_get_softc(dev);
1261	ifp = &sc->arpcom.ac_if;
1262	KASSERT(mtx_initialized(&sc->rl_mtx), ("re mutex not initialized"));
1263
1264	attached = device_is_attached(dev);
1265	/* These should only be active if attach succeeded */
1266	if (attached)
1267		ether_ifdetach(ifp);
1268
1269	RL_LOCK(sc);
1270#if 0
1271	sc->suspended = 1;
1272#endif
1273
1274	/* These should only be active if attach succeeded */
1275	if (attached) {
1276		re_stop(sc);
1277		/*
1278		 * Force off the IFF_UP flag here, in case someone
1279		 * still had a BPF descriptor attached to this
1280		 * interface. If they do, ether_ifdetach() will cause
1281		 * the BPF code to try and clear the promisc mode
1282		 * flag, which will bubble down to re_ioctl(),
1283		 * which will try to call re_init() again. This will
1284		 * turn the NIC back on and restart the MII ticker,
1285		 * which will panic the system when the kernel tries
1286		 * to invoke the re_tick() function that isn't there
1287		 * anymore.
1288		 */
1289		ifp->if_flags &= ~IFF_UP;
1290	}
1291	if (sc->rl_miibus)
1292		device_delete_child(dev, sc->rl_miibus);
1293	bus_generic_detach(dev);
1294
1295	/*
1296	 * The rest is resource deallocation, so we should already be
1297	 * stopped here.
1298	 */
1299	RL_UNLOCK(sc);
1300
1301	if (sc->rl_intrhand)
1302		bus_teardown_intr(dev, sc->rl_irq, sc->rl_intrhand);
1303	if (sc->rl_irq)
1304		bus_release_resource(dev, SYS_RES_IRQ, 0, sc->rl_irq);
1305	if (sc->rl_res)
1306		bus_release_resource(dev, RL_RES, RL_RID, sc->rl_res);
1307
1308
1309	/* Unload and free the RX DMA ring memory and map */
1310
1311	if (sc->rl_ldata.rl_rx_list_tag) {
1312		bus_dmamap_unload(sc->rl_ldata.rl_rx_list_tag,
1313		    sc->rl_ldata.rl_rx_list_map);
1314		bus_dmamem_free(sc->rl_ldata.rl_rx_list_tag,
1315		    sc->rl_ldata.rl_rx_list,
1316		    sc->rl_ldata.rl_rx_list_map);
1317		bus_dma_tag_destroy(sc->rl_ldata.rl_rx_list_tag);
1318	}
1319
1320	/* Unload and free the TX DMA ring memory and map */
1321
1322	if (sc->rl_ldata.rl_tx_list_tag) {
1323		bus_dmamap_unload(sc->rl_ldata.rl_tx_list_tag,
1324		    sc->rl_ldata.rl_tx_list_map);
1325		bus_dmamem_free(sc->rl_ldata.rl_tx_list_tag,
1326		    sc->rl_ldata.rl_tx_list,
1327		    sc->rl_ldata.rl_tx_list_map);
1328		bus_dma_tag_destroy(sc->rl_ldata.rl_tx_list_tag);
1329	}
1330
1331	/* Destroy all the RX and TX buffer maps */
1332
1333	if (sc->rl_ldata.rl_mtag) {
1334		for (i = 0; i < RL_TX_DESC_CNT; i++)
1335			bus_dmamap_destroy(sc->rl_ldata.rl_mtag,
1336			    sc->rl_ldata.rl_tx_dmamap[i]);
1337		for (i = 0; i < RL_RX_DESC_CNT; i++)
1338			bus_dmamap_destroy(sc->rl_ldata.rl_mtag,
1339			    sc->rl_ldata.rl_rx_dmamap[i]);
1340		bus_dma_tag_destroy(sc->rl_ldata.rl_mtag);
1341	}
1342
1343	/* Unload and free the stats buffer and map */
1344
1345	if (sc->rl_ldata.rl_stag) {
1346		bus_dmamap_unload(sc->rl_ldata.rl_stag,
1347		    sc->rl_ldata.rl_rx_list_map);
1348		bus_dmamem_free(sc->rl_ldata.rl_stag,
1349		    sc->rl_ldata.rl_stats,
1350		    sc->rl_ldata.rl_smap);
1351		bus_dma_tag_destroy(sc->rl_ldata.rl_stag);
1352	}
1353
1354	if (sc->rl_parent_tag)
1355		bus_dma_tag_destroy(sc->rl_parent_tag);
1356
1357	mtx_destroy(&sc->rl_mtx);
1358
1359	return (0);
1360}
1361
1362static int
1363re_newbuf(sc, idx, m)
1364	struct rl_softc		*sc;
1365	int			idx;
1366	struct mbuf		*m;
1367{
1368	struct rl_dmaload_arg	arg;
1369	struct mbuf		*n = NULL;
1370	int			error;
1371
1372	if (m == NULL) {
1373		n = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
1374		if (n == NULL)
1375			return (ENOBUFS);
1376		m = n;
1377	} else
1378		m->m_data = m->m_ext.ext_buf;
1379
1380	m->m_len = m->m_pkthdr.len = MCLBYTES;
1381#ifdef RE_FIXUP_RX
1382	/*
1383	 * This is part of an evil trick to deal with non-x86 platforms.
1384	 * The RealTek chip requires RX buffers to be aligned on 64-bit
1385	 * boundaries, but that will hose non-x86 machines. To get around
1386	 * this, we leave some empty space at the start of each buffer
1387	 * and for non-x86 hosts, we copy the buffer back six bytes
1388	 * to achieve word alignment. This is slightly more efficient
1389	 * than allocating a new buffer, copying the contents, and
1390	 * discarding the old buffer.
1391	 */
1392	m_adj(m, RE_ETHER_ALIGN);
1393#endif
1394	arg.sc = sc;
1395	arg.rl_idx = idx;
1396	arg.rl_maxsegs = 1;
1397	arg.rl_flags = 0;
1398	arg.rl_ring = sc->rl_ldata.rl_rx_list;
1399
1400	error = bus_dmamap_load_mbuf(sc->rl_ldata.rl_mtag,
1401	    sc->rl_ldata.rl_rx_dmamap[idx], m, re_dma_map_desc,
1402	    &arg, BUS_DMA_NOWAIT);
1403	if (error || arg.rl_maxsegs != 1) {
1404		if (n != NULL)
1405			m_freem(n);
1406		return (ENOMEM);
1407	}
1408
1409	sc->rl_ldata.rl_rx_list[idx].rl_cmdstat |= htole32(RL_RDESC_CMD_OWN);
1410	sc->rl_ldata.rl_rx_mbuf[idx] = m;
1411
1412	bus_dmamap_sync(sc->rl_ldata.rl_mtag,
1413	    sc->rl_ldata.rl_rx_dmamap[idx],
1414	    BUS_DMASYNC_PREREAD);
1415
1416	return (0);
1417}
1418
1419#ifdef RE_FIXUP_RX
1420static __inline void
1421re_fixup_rx(m)
1422	struct mbuf		*m;
1423{
1424	int                     i;
1425	uint16_t                *src, *dst;
1426
1427	src = mtod(m, uint16_t *);
1428	dst = src - (RE_ETHER_ALIGN - ETHER_ALIGN) / sizeof *src;
1429
1430	for (i = 0; i < (m->m_len / sizeof(uint16_t) + 1); i++)
1431		*dst++ = *src++;
1432
1433	m->m_data -= RE_ETHER_ALIGN - ETHER_ALIGN;
1434
1435	return;
1436}
1437#endif
1438
1439static int
1440re_tx_list_init(sc)
1441	struct rl_softc		*sc;
1442{
1443
1444	RL_LOCK_ASSERT(sc);
1445
1446	bzero ((char *)sc->rl_ldata.rl_tx_list, RL_TX_LIST_SZ);
1447	bzero ((char *)&sc->rl_ldata.rl_tx_mbuf,
1448	    (RL_TX_DESC_CNT * sizeof(struct mbuf *)));
1449
1450	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
1451	    sc->rl_ldata.rl_tx_list_map, BUS_DMASYNC_PREWRITE);
1452	sc->rl_ldata.rl_tx_prodidx = 0;
1453	sc->rl_ldata.rl_tx_considx = 0;
1454	sc->rl_ldata.rl_tx_free = RL_TX_DESC_CNT;
1455
1456	return (0);
1457}
1458
1459static int
1460re_rx_list_init(sc)
1461	struct rl_softc		*sc;
1462{
1463	int			i;
1464
1465	bzero ((char *)sc->rl_ldata.rl_rx_list, RL_RX_LIST_SZ);
1466	bzero ((char *)&sc->rl_ldata.rl_rx_mbuf,
1467	    (RL_RX_DESC_CNT * sizeof(struct mbuf *)));
1468
1469	for (i = 0; i < RL_RX_DESC_CNT; i++) {
1470		if (re_newbuf(sc, i, NULL) == ENOBUFS)
1471			return (ENOBUFS);
1472	}
1473
1474	/* Flush the RX descriptors */
1475
1476	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1477	    sc->rl_ldata.rl_rx_list_map,
1478	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
1479
1480	sc->rl_ldata.rl_rx_prodidx = 0;
1481	sc->rl_head = sc->rl_tail = NULL;
1482
1483	return (0);
1484}
1485
1486/*
1487 * RX handler for C+ and 8169. For the gigE chips, we support
1488 * the reception of jumbo frames that have been fragmented
1489 * across multiple 2K mbuf cluster buffers.
1490 */
1491static void
1492re_rxeof(sc)
1493	struct rl_softc		*sc;
1494{
1495	struct mbuf		*m;
1496	struct ifnet		*ifp;
1497	int			i, total_len;
1498	struct rl_desc		*cur_rx;
1499	u_int32_t		rxstat, rxvlan;
1500
1501	RL_LOCK_ASSERT(sc);
1502
1503	ifp = &sc->arpcom.ac_if;
1504	i = sc->rl_ldata.rl_rx_prodidx;
1505
1506	/* Invalidate the descriptor memory */
1507
1508	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1509	    sc->rl_ldata.rl_rx_list_map,
1510	    BUS_DMASYNC_POSTREAD);
1511
1512	while (!RL_OWN(&sc->rl_ldata.rl_rx_list[i])) {
1513		cur_rx = &sc->rl_ldata.rl_rx_list[i];
1514		m = sc->rl_ldata.rl_rx_mbuf[i];
1515		total_len = RL_RXBYTES(cur_rx);
1516		rxstat = le32toh(cur_rx->rl_cmdstat);
1517		rxvlan = le32toh(cur_rx->rl_vlanctl);
1518
1519		/* Invalidate the RX mbuf and unload its map */
1520
1521		bus_dmamap_sync(sc->rl_ldata.rl_mtag,
1522		    sc->rl_ldata.rl_rx_dmamap[i],
1523		    BUS_DMASYNC_POSTWRITE);
1524		bus_dmamap_unload(sc->rl_ldata.rl_mtag,
1525		    sc->rl_ldata.rl_rx_dmamap[i]);
1526
1527		if (!(rxstat & RL_RDESC_STAT_EOF)) {
1528			m->m_len = RE_RX_DESC_BUFLEN;
1529			if (sc->rl_head == NULL)
1530				sc->rl_head = sc->rl_tail = m;
1531			else {
1532				m->m_flags &= ~M_PKTHDR;
1533				sc->rl_tail->m_next = m;
1534				sc->rl_tail = m;
1535			}
1536			re_newbuf(sc, i, NULL);
1537			RL_DESC_INC(i);
1538			continue;
1539		}
1540
1541		/*
1542		 * NOTE: for the 8139C+, the frame length field
1543		 * is always 12 bits in size, but for the gigE chips,
1544		 * it is 13 bits (since the max RX frame length is 16K).
1545		 * Unfortunately, all 32 bits in the status word
1546		 * were already used, so to make room for the extra
1547		 * length bit, RealTek took out the 'frame alignment
1548		 * error' bit and shifted the other status bits
1549		 * over one slot. The OWN, EOR, FS and LS bits are
1550		 * still in the same places. We have already extracted
1551		 * the frame length and checked the OWN bit, so rather
1552		 * than using an alternate bit mapping, we shift the
1553		 * status bits one space to the right so we can evaluate
1554		 * them using the 8169 status as though it was in the
1555		 * same format as that of the 8139C+.
1556		 */
1557		if (sc->rl_type == RL_8169)
1558			rxstat >>= 1;
1559
1560		/*
1561		 * if total_len > 2^13-1, both _RXERRSUM and _GIANT will be
1562		 * set, but if CRC is clear, it will still be a valid frame.
1563		 */
1564		if (rxstat & RL_RDESC_STAT_RXERRSUM && !(total_len > 8191 &&
1565		    (rxstat & RL_RDESC_STAT_ERRS) == RL_RDESC_STAT_GIANT)) {
1566			ifp->if_ierrors++;
1567			/*
1568			 * If this is part of a multi-fragment packet,
1569			 * discard all the pieces.
1570			 */
1571			if (sc->rl_head != NULL) {
1572				m_freem(sc->rl_head);
1573				sc->rl_head = sc->rl_tail = NULL;
1574			}
1575			re_newbuf(sc, i, m);
1576			RL_DESC_INC(i);
1577			continue;
1578		}
1579
1580		/*
1581		 * If allocating a replacement mbuf fails,
1582		 * reload the current one.
1583		 */
1584
1585		if (re_newbuf(sc, i, NULL)) {
1586			ifp->if_ierrors++;
1587			if (sc->rl_head != NULL) {
1588				m_freem(sc->rl_head);
1589				sc->rl_head = sc->rl_tail = NULL;
1590			}
1591			re_newbuf(sc, i, m);
1592			RL_DESC_INC(i);
1593			continue;
1594		}
1595
1596		RL_DESC_INC(i);
1597
1598		if (sc->rl_head != NULL) {
1599			m->m_len = total_len % RE_RX_DESC_BUFLEN;
1600			if (m->m_len == 0)
1601				m->m_len = RE_RX_DESC_BUFLEN;
1602			/*
1603			 * Special case: if there's 4 bytes or less
1604			 * in this buffer, the mbuf can be discarded:
1605			 * the last 4 bytes is the CRC, which we don't
1606			 * care about anyway.
1607			 */
1608			if (m->m_len <= ETHER_CRC_LEN) {
1609				sc->rl_tail->m_len -=
1610				    (ETHER_CRC_LEN - m->m_len);
1611				m_freem(m);
1612			} else {
1613				m->m_len -= ETHER_CRC_LEN;
1614				m->m_flags &= ~M_PKTHDR;
1615				sc->rl_tail->m_next = m;
1616			}
1617			m = sc->rl_head;
1618			sc->rl_head = sc->rl_tail = NULL;
1619			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
1620		} else
1621			m->m_pkthdr.len = m->m_len =
1622			    (total_len - ETHER_CRC_LEN);
1623
1624#ifdef RE_FIXUP_RX
1625		re_fixup_rx(m);
1626#endif
1627		ifp->if_ipackets++;
1628		m->m_pkthdr.rcvif = ifp;
1629
1630		/* Do RX checksumming if enabled */
1631
1632		if (ifp->if_capenable & IFCAP_RXCSUM) {
1633
1634			/* Check IP header checksum */
1635			if (rxstat & RL_RDESC_STAT_PROTOID)
1636				m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED;
1637			if (!(rxstat & RL_RDESC_STAT_IPSUMBAD))
1638				m->m_pkthdr.csum_flags |= CSUM_IP_VALID;
1639
1640			/* Check TCP/UDP checksum */
1641			if ((RL_TCPPKT(rxstat) &&
1642			    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
1643			    (RL_UDPPKT(rxstat) &&
1644			    !(rxstat & RL_RDESC_STAT_UDPSUMBAD))) {
1645				m->m_pkthdr.csum_flags |=
1646				    CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
1647				m->m_pkthdr.csum_data = 0xffff;
1648			}
1649		}
1650
1651		if (rxvlan & RL_RDESC_VLANCTL_TAG)
1652			VLAN_INPUT_TAG(ifp, m,
1653			    ntohs((rxvlan & RL_RDESC_VLANCTL_DATA)), continue);
1654		RL_UNLOCK(sc);
1655		(*ifp->if_input)(ifp, m);
1656		RL_LOCK(sc);
1657	}
1658
1659	/* Flush the RX DMA ring */
1660
1661	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1662	    sc->rl_ldata.rl_rx_list_map,
1663	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
1664
1665	sc->rl_ldata.rl_rx_prodidx = i;
1666}
1667
1668static void
1669re_txeof(sc)
1670	struct rl_softc		*sc;
1671{
1672	struct ifnet		*ifp;
1673	u_int32_t		txstat;
1674	int			idx;
1675
1676	ifp = &sc->arpcom.ac_if;
1677	idx = sc->rl_ldata.rl_tx_considx;
1678
1679	/* Invalidate the TX descriptor list */
1680
1681	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
1682	    sc->rl_ldata.rl_tx_list_map,
1683	    BUS_DMASYNC_POSTREAD);
1684
1685	while (idx != sc->rl_ldata.rl_tx_prodidx) {
1686
1687		txstat = le32toh(sc->rl_ldata.rl_tx_list[idx].rl_cmdstat);
1688		if (txstat & RL_TDESC_CMD_OWN)
1689			break;
1690
1691		/*
1692		 * We only stash mbufs in the last descriptor
1693		 * in a fragment chain, which also happens to
1694		 * be the only place where the TX status bits
1695		 * are valid.
1696		 */
1697
1698		if (txstat & RL_TDESC_CMD_EOF) {
1699			m_freem(sc->rl_ldata.rl_tx_mbuf[idx]);
1700			sc->rl_ldata.rl_tx_mbuf[idx] = NULL;
1701			bus_dmamap_unload(sc->rl_ldata.rl_mtag,
1702			    sc->rl_ldata.rl_tx_dmamap[idx]);
1703			if (txstat & (RL_TDESC_STAT_EXCESSCOL|
1704			    RL_TDESC_STAT_COLCNT))
1705				ifp->if_collisions++;
1706			if (txstat & RL_TDESC_STAT_TXERRSUM)
1707				ifp->if_oerrors++;
1708			else
1709				ifp->if_opackets++;
1710		}
1711		sc->rl_ldata.rl_tx_free++;
1712		RL_DESC_INC(idx);
1713	}
1714
1715	/* No changes made to the TX ring, so no flush needed */
1716
1717	if (idx != sc->rl_ldata.rl_tx_considx) {
1718		sc->rl_ldata.rl_tx_considx = idx;
1719		ifp->if_flags &= ~IFF_OACTIVE;
1720		ifp->if_timer = 0;
1721	}
1722
1723	/*
1724	 * If not all descriptors have been released reaped yet,
1725	 * reload the timer so that we will eventually get another
1726	 * interrupt that will cause us to re-enter this routine.
1727	 * This is done in case the transmitter has gone idle.
1728	 */
1729	if (sc->rl_ldata.rl_tx_free != RL_TX_DESC_CNT)
1730		CSR_WRITE_4(sc, RL_TIMERCNT, 1);
1731}
1732
1733static void
1734re_tick(xsc)
1735	void			*xsc;
1736{
1737	struct rl_softc		*sc;
1738
1739	sc = xsc;
1740	RL_LOCK(sc);
1741	re_tick_locked(sc);
1742	RL_UNLOCK(sc);
1743}
1744
1745static void
1746re_tick_locked(sc)
1747	struct rl_softc		*sc;
1748{
1749	struct mii_data		*mii;
1750
1751	RL_LOCK_ASSERT(sc);
1752
1753	mii = device_get_softc(sc->rl_miibus);
1754
1755	mii_tick(mii);
1756
1757	sc->rl_stat_ch = timeout(re_tick, sc, hz);
1758}
1759
1760#ifdef DEVICE_POLLING
1761static void
1762re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
1763{
1764	struct rl_softc *sc = ifp->if_softc;
1765
1766	RL_LOCK(sc);
1767	re_poll_locked(ifp, cmd, count);
1768	RL_UNLOCK(sc);
1769}
1770
1771static void
1772re_poll_locked(struct ifnet *ifp, enum poll_cmd cmd, int count)
1773{
1774	struct rl_softc *sc = ifp->if_softc;
1775
1776	RL_LOCK_ASSERT(sc);
1777
1778	if (!(ifp->if_capenable & IFCAP_POLLING)) {
1779		ether_poll_deregister(ifp);
1780		cmd = POLL_DEREGISTER;
1781	}
1782	if (cmd == POLL_DEREGISTER) { /* final call, enable interrupts */
1783		CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
1784		return;
1785	}
1786
1787	sc->rxcycles = count;
1788	re_rxeof(sc);
1789	re_txeof(sc);
1790
1791	if (ifp->if_snd.ifq_head != NULL)
1792		re_start_locked(ifp);
1793
1794	if (cmd == POLL_AND_CHECK_STATUS) { /* also check status register */
1795		u_int16_t       status;
1796
1797		status = CSR_READ_2(sc, RL_ISR);
1798		if (status == 0xffff)
1799			return;
1800		if (status)
1801			CSR_WRITE_2(sc, RL_ISR, status);
1802
1803		/*
1804		 * XXX check behaviour on receiver stalls.
1805		 */
1806
1807		if (status & RL_ISR_SYSTEM_ERR) {
1808			re_reset(sc);
1809			re_init_locked(sc);
1810		}
1811	}
1812}
1813#endif /* DEVICE_POLLING */
1814
1815static void
1816re_intr(arg)
1817	void			*arg;
1818{
1819	struct rl_softc		*sc;
1820	struct ifnet		*ifp;
1821	u_int16_t		status;
1822
1823	sc = arg;
1824
1825	RL_LOCK(sc);
1826
1827	ifp = &sc->arpcom.ac_if;
1828
1829	if (sc->suspended || !(ifp->if_flags & IFF_UP))
1830		goto done_locked;
1831
1832#ifdef DEVICE_POLLING
1833	if  (ifp->if_flags & IFF_POLLING)
1834		goto done_locked;
1835	if ((ifp->if_capenable & IFCAP_POLLING) &&
1836	    ether_poll_register(re_poll, ifp)) { /* ok, disable interrupts */
1837		CSR_WRITE_2(sc, RL_IMR, 0x0000);
1838		re_poll_locked(ifp, 0, 1);
1839		goto done_locked;
1840	}
1841#endif /* DEVICE_POLLING */
1842
1843	for (;;) {
1844
1845		status = CSR_READ_2(sc, RL_ISR);
1846		/* If the card has gone away the read returns 0xffff. */
1847		if (status == 0xffff)
1848			break;
1849		if (status)
1850			CSR_WRITE_2(sc, RL_ISR, status);
1851
1852		if ((status & RL_INTRS_CPLUS) == 0)
1853			break;
1854
1855		if ((status & RL_ISR_RX_OK) ||
1856		    (status & RL_ISR_RX_ERR))
1857			re_rxeof(sc);
1858
1859		if ((status & RL_ISR_TIMEOUT_EXPIRED) ||
1860		    (status & RL_ISR_TX_ERR) ||
1861		    (status & RL_ISR_TX_DESC_UNAVAIL))
1862			re_txeof(sc);
1863
1864		if (status & RL_ISR_SYSTEM_ERR) {
1865			re_reset(sc);
1866			re_init_locked(sc);
1867		}
1868
1869		if (status & RL_ISR_LINKCHG) {
1870			untimeout(re_tick, sc, sc->rl_stat_ch);
1871			re_tick_locked(sc);
1872		}
1873	}
1874
1875	if (ifp->if_snd.ifq_head != NULL)
1876		re_start_locked(ifp);
1877
1878done_locked:
1879	RL_UNLOCK(sc);
1880}
1881
1882static int
1883re_encap(sc, m_head, idx)
1884	struct rl_softc		*sc;
1885	struct mbuf		**m_head;
1886	int			*idx;
1887{
1888	struct mbuf		*m_new = NULL;
1889	struct rl_dmaload_arg	arg;
1890	bus_dmamap_t		map;
1891	int			error;
1892	struct m_tag		*mtag;
1893
1894	RL_LOCK_ASSERT(sc);
1895
1896	if (sc->rl_ldata.rl_tx_free <= 4)
1897		return (EFBIG);
1898
1899	/*
1900	 * Set up checksum offload. Note: checksum offload bits must
1901	 * appear in all descriptors of a multi-descriptor transmit
1902	 * attempt. This is according to testing done with an 8169
1903	 * chip. This is a requirement.
1904	 */
1905
1906	arg.rl_flags = 0;
1907
1908	if ((*m_head)->m_pkthdr.csum_flags & CSUM_IP)
1909		arg.rl_flags |= RL_TDESC_CMD_IPCSUM;
1910	if ((*m_head)->m_pkthdr.csum_flags & CSUM_TCP)
1911		arg.rl_flags |= RL_TDESC_CMD_TCPCSUM;
1912	if ((*m_head)->m_pkthdr.csum_flags & CSUM_UDP)
1913		arg.rl_flags |= RL_TDESC_CMD_UDPCSUM;
1914
1915	arg.sc = sc;
1916	arg.rl_idx = *idx;
1917	arg.rl_maxsegs = sc->rl_ldata.rl_tx_free;
1918	if (arg.rl_maxsegs > 4)
1919		arg.rl_maxsegs -= 4;
1920	arg.rl_ring = sc->rl_ldata.rl_tx_list;
1921
1922	map = sc->rl_ldata.rl_tx_dmamap[*idx];
1923	error = bus_dmamap_load_mbuf(sc->rl_ldata.rl_mtag, map,
1924	    *m_head, re_dma_map_desc, &arg, BUS_DMA_NOWAIT);
1925
1926	if (error && error != EFBIG) {
1927		printf("re%d: can't map mbuf (error %d)\n", sc->rl_unit, error);
1928		return (ENOBUFS);
1929	}
1930
1931	/* Too many segments to map, coalesce into a single mbuf */
1932
1933	if (error || arg.rl_maxsegs == 0) {
1934		m_new = m_defrag(*m_head, M_DONTWAIT);
1935		if (m_new == NULL)
1936			return (ENOBUFS);
1937		else
1938			*m_head = m_new;
1939
1940		arg.sc = sc;
1941		arg.rl_idx = *idx;
1942		arg.rl_maxsegs = sc->rl_ldata.rl_tx_free;
1943		arg.rl_ring = sc->rl_ldata.rl_tx_list;
1944
1945		error = bus_dmamap_load_mbuf(sc->rl_ldata.rl_mtag, map,
1946		    *m_head, re_dma_map_desc, &arg, BUS_DMA_NOWAIT);
1947		if (error) {
1948			printf("re%d: can't map mbuf (error %d)\n",
1949			    sc->rl_unit, error);
1950			return (EFBIG);
1951		}
1952	}
1953
1954	/*
1955	 * Insure that the map for this transmission
1956	 * is placed at the array index of the last descriptor
1957	 * in this chain.  (Swap last and first dmamaps.)
1958	 */
1959	sc->rl_ldata.rl_tx_dmamap[*idx] =
1960	    sc->rl_ldata.rl_tx_dmamap[arg.rl_idx];
1961	sc->rl_ldata.rl_tx_dmamap[arg.rl_idx] = map;
1962
1963	sc->rl_ldata.rl_tx_mbuf[arg.rl_idx] = *m_head;
1964	sc->rl_ldata.rl_tx_free -= arg.rl_maxsegs;
1965
1966	/*
1967	 * Set up hardware VLAN tagging. Note: vlan tag info must
1968	 * appear in the first descriptor of a multi-descriptor
1969	 * transmission attempt.
1970	 */
1971
1972	mtag = VLAN_OUTPUT_TAG(&sc->arpcom.ac_if, *m_head);
1973	if (mtag != NULL)
1974		sc->rl_ldata.rl_tx_list[*idx].rl_vlanctl =
1975		    htole32(htons(VLAN_TAG_VALUE(mtag)) | RL_TDESC_VLANCTL_TAG);
1976
1977	/* Transfer ownership of packet to the chip. */
1978
1979	sc->rl_ldata.rl_tx_list[arg.rl_idx].rl_cmdstat |=
1980	    htole32(RL_TDESC_CMD_OWN);
1981	if (*idx != arg.rl_idx)
1982		sc->rl_ldata.rl_tx_list[*idx].rl_cmdstat |=
1983		    htole32(RL_TDESC_CMD_OWN);
1984
1985	RL_DESC_INC(arg.rl_idx);
1986	*idx = arg.rl_idx;
1987
1988	return (0);
1989}
1990
1991static void
1992re_start(ifp)
1993	struct ifnet		*ifp;
1994{
1995	struct rl_softc		*sc;
1996
1997	sc = ifp->if_softc;
1998	RL_LOCK(sc);
1999	re_start_locked(ifp);
2000	RL_UNLOCK(sc);
2001}
2002
2003/*
2004 * Main transmit routine for C+ and gigE NICs.
2005 */
2006static void
2007re_start_locked(ifp)
2008	struct ifnet		*ifp;
2009{
2010	struct rl_softc		*sc;
2011	struct mbuf		*m_head = NULL;
2012	int			idx;
2013
2014	sc = ifp->if_softc;
2015
2016	RL_LOCK_ASSERT(sc);
2017
2018	idx = sc->rl_ldata.rl_tx_prodidx;
2019
2020	while (sc->rl_ldata.rl_tx_mbuf[idx] == NULL) {
2021		IF_DEQUEUE(&ifp->if_snd, m_head);
2022		if (m_head == NULL)
2023			break;
2024
2025		if (re_encap(sc, &m_head, &idx)) {
2026			IF_PREPEND(&ifp->if_snd, m_head);
2027			ifp->if_flags |= IFF_OACTIVE;
2028			break;
2029		}
2030
2031		/*
2032		 * If there's a BPF listener, bounce a copy of this frame
2033		 * to him.
2034		 */
2035		BPF_MTAP(ifp, m_head);
2036	}
2037
2038	/* Flush the TX descriptors */
2039
2040	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2041	    sc->rl_ldata.rl_tx_list_map,
2042	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2043
2044	sc->rl_ldata.rl_tx_prodidx = idx;
2045
2046	/*
2047	 * RealTek put the TX poll request register in a different
2048	 * location on the 8169 gigE chip. I don't know why.
2049	 */
2050
2051	if (sc->rl_type == RL_8169)
2052		CSR_WRITE_2(sc, RL_GTXSTART, RL_TXSTART_START);
2053	else
2054		CSR_WRITE_2(sc, RL_TXSTART, RL_TXSTART_START);
2055
2056	/*
2057	 * Use the countdown timer for interrupt moderation.
2058	 * 'TX done' interrupts are disabled. Instead, we reset the
2059	 * countdown timer, which will begin counting until it hits
2060	 * the value in the TIMERINT register, and then trigger an
2061	 * interrupt. Each time we write to the TIMERCNT register,
2062	 * the timer count is reset to 0.
2063	 */
2064	CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2065
2066	/*
2067	 * Set a timeout in case the chip goes out to lunch.
2068	 */
2069	ifp->if_timer = 5;
2070}
2071
2072static void
2073re_init(xsc)
2074	void			*xsc;
2075{
2076	struct rl_softc		*sc = xsc;
2077
2078	RL_LOCK(sc);
2079	re_init_locked(sc);
2080	RL_UNLOCK(sc);
2081}
2082
2083static void
2084re_init_locked(sc)
2085	struct rl_softc		*sc;
2086{
2087	struct ifnet		*ifp = &sc->arpcom.ac_if;
2088	struct mii_data		*mii;
2089	u_int32_t		rxcfg = 0;
2090
2091	RL_LOCK_ASSERT(sc);
2092
2093	mii = device_get_softc(sc->rl_miibus);
2094
2095	/*
2096	 * Cancel pending I/O and free all RX/TX buffers.
2097	 */
2098	re_stop(sc);
2099
2100	/*
2101	 * Enable C+ RX and TX mode, as well as VLAN stripping and
2102	 * RX checksum offload. We must configure the C+ register
2103	 * before all others.
2104	 */
2105	CSR_WRITE_2(sc, RL_CPLUS_CMD, RL_CPLUSCMD_RXENB|
2106	    RL_CPLUSCMD_TXENB|RL_CPLUSCMD_PCI_MRW|
2107	    RL_CPLUSCMD_VLANSTRIP|
2108	    (ifp->if_capenable & IFCAP_RXCSUM ?
2109	    RL_CPLUSCMD_RXCSUM_ENB : 0));
2110
2111	/*
2112	 * Init our MAC address.  Even though the chipset
2113	 * documentation doesn't mention it, we need to enter "Config
2114	 * register write enable" mode to modify the ID registers.
2115	 */
2116	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG);
2117	CSR_WRITE_STREAM_4(sc, RL_IDR0,
2118	    *(u_int32_t *)(&sc->arpcom.ac_enaddr[0]));
2119	CSR_WRITE_STREAM_4(sc, RL_IDR4,
2120	    *(u_int32_t *)(&sc->arpcom.ac_enaddr[4]));
2121	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
2122
2123	/*
2124	 * For C+ mode, initialize the RX descriptors and mbufs.
2125	 */
2126	re_rx_list_init(sc);
2127	re_tx_list_init(sc);
2128
2129	/*
2130	 * Enable transmit and receive.
2131	 */
2132	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
2133
2134	/*
2135	 * Set the initial TX and RX configuration.
2136	 */
2137	if (sc->rl_testmode) {
2138		if (sc->rl_type == RL_8169)
2139			CSR_WRITE_4(sc, RL_TXCFG,
2140			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON);
2141		else
2142			CSR_WRITE_4(sc, RL_TXCFG,
2143			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON_CPLUS);
2144	} else
2145		CSR_WRITE_4(sc, RL_TXCFG, RL_TXCFG_CONFIG);
2146	CSR_WRITE_4(sc, RL_RXCFG, RL_RXCFG_CONFIG);
2147
2148	/* Set the individual bit to receive frames for this host only. */
2149	rxcfg = CSR_READ_4(sc, RL_RXCFG);
2150	rxcfg |= RL_RXCFG_RX_INDIV;
2151
2152	/* If we want promiscuous mode, set the allframes bit. */
2153	if (ifp->if_flags & IFF_PROMISC)
2154		rxcfg |= RL_RXCFG_RX_ALLPHYS;
2155	else
2156		rxcfg &= ~RL_RXCFG_RX_ALLPHYS;
2157	CSR_WRITE_4(sc, RL_RXCFG, rxcfg);
2158
2159	/*
2160	 * Set capture broadcast bit to capture broadcast frames.
2161	 */
2162	if (ifp->if_flags & IFF_BROADCAST)
2163		rxcfg |= RL_RXCFG_RX_BROAD;
2164	else
2165		rxcfg &= ~RL_RXCFG_RX_BROAD;
2166	CSR_WRITE_4(sc, RL_RXCFG, rxcfg);
2167
2168	/*
2169	 * Program the multicast filter, if necessary.
2170	 */
2171	re_setmulti(sc);
2172
2173#ifdef DEVICE_POLLING
2174	/*
2175	 * Disable interrupts if we are polling.
2176	 */
2177	if (ifp->if_flags & IFF_POLLING)
2178		CSR_WRITE_2(sc, RL_IMR, 0);
2179	else	/* otherwise ... */
2180#endif /* DEVICE_POLLING */
2181	/*
2182	 * Enable interrupts.
2183	 */
2184	if (sc->rl_testmode)
2185		CSR_WRITE_2(sc, RL_IMR, 0);
2186	else
2187		CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2188
2189	/* Set initial TX threshold */
2190	sc->rl_txthresh = RL_TX_THRESH_INIT;
2191
2192	/* Start RX/TX process. */
2193	CSR_WRITE_4(sc, RL_MISSEDPKT, 0);
2194#ifdef notdef
2195	/* Enable receiver and transmitter. */
2196	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
2197#endif
2198	/*
2199	 * Load the addresses of the RX and TX lists into the chip.
2200	 */
2201
2202	CSR_WRITE_4(sc, RL_RXLIST_ADDR_HI,
2203	    RL_ADDR_HI(sc->rl_ldata.rl_rx_list_addr));
2204	CSR_WRITE_4(sc, RL_RXLIST_ADDR_LO,
2205	    RL_ADDR_LO(sc->rl_ldata.rl_rx_list_addr));
2206
2207	CSR_WRITE_4(sc, RL_TXLIST_ADDR_HI,
2208	    RL_ADDR_HI(sc->rl_ldata.rl_tx_list_addr));
2209	CSR_WRITE_4(sc, RL_TXLIST_ADDR_LO,
2210	    RL_ADDR_LO(sc->rl_ldata.rl_tx_list_addr));
2211
2212	CSR_WRITE_1(sc, RL_EARLY_TX_THRESH, 16);
2213
2214	/*
2215	 * Initialize the timer interrupt register so that
2216	 * a timer interrupt will be generated once the timer
2217	 * reaches a certain number of ticks. The timer is
2218	 * reloaded on each transmit. This gives us TX interrupt
2219	 * moderation, which dramatically improves TX frame rate.
2220	 */
2221	if (sc->rl_type == RL_8169)
2222		CSR_WRITE_4(sc, RL_TIMERINT_8169, 0x800);
2223	else
2224		CSR_WRITE_4(sc, RL_TIMERINT, 0x400);
2225
2226	/*
2227	 * For 8169 gigE NICs, set the max allowed RX packet
2228	 * size so we can receive jumbo frames.
2229	 */
2230	if (sc->rl_type == RL_8169)
2231		CSR_WRITE_2(sc, RL_MAXRXPKTLEN, 16383);
2232
2233	if (sc->rl_testmode)
2234		return;
2235
2236	mii_mediachg(mii);
2237
2238	CSR_WRITE_1(sc, RL_CFG1, RL_CFG1_DRVLOAD|RL_CFG1_FULLDUPLEX);
2239
2240	ifp->if_flags |= IFF_RUNNING;
2241	ifp->if_flags &= ~IFF_OACTIVE;
2242
2243	sc->rl_stat_ch = timeout(re_tick, sc, hz);
2244}
2245
2246/*
2247 * Set media options.
2248 */
2249static int
2250re_ifmedia_upd(ifp)
2251	struct ifnet		*ifp;
2252{
2253	struct rl_softc		*sc;
2254	struct mii_data		*mii;
2255
2256	sc = ifp->if_softc;
2257	mii = device_get_softc(sc->rl_miibus);
2258	mii_mediachg(mii);
2259
2260	return (0);
2261}
2262
2263/*
2264 * Report current media status.
2265 */
2266static void
2267re_ifmedia_sts(ifp, ifmr)
2268	struct ifnet		*ifp;
2269	struct ifmediareq	*ifmr;
2270{
2271	struct rl_softc		*sc;
2272	struct mii_data		*mii;
2273
2274	sc = ifp->if_softc;
2275	mii = device_get_softc(sc->rl_miibus);
2276
2277	mii_pollstat(mii);
2278	ifmr->ifm_active = mii->mii_media_active;
2279	ifmr->ifm_status = mii->mii_media_status;
2280}
2281
2282static int
2283re_ioctl(ifp, command, data)
2284	struct ifnet		*ifp;
2285	u_long			command;
2286	caddr_t			data;
2287{
2288	struct rl_softc		*sc = ifp->if_softc;
2289	struct ifreq		*ifr = (struct ifreq *) data;
2290	struct mii_data		*mii;
2291	int			error = 0;
2292
2293	switch (command) {
2294	case SIOCSIFMTU:
2295		if (ifr->ifr_mtu > RL_JUMBO_MTU)
2296			error = EINVAL;
2297		ifp->if_mtu = ifr->ifr_mtu;
2298		break;
2299	case SIOCSIFFLAGS:
2300		RL_LOCK(sc);
2301		if (ifp->if_flags & IFF_UP)
2302			re_init_locked(sc);
2303		else if (ifp->if_flags & IFF_RUNNING)
2304			re_stop(sc);
2305		RL_UNLOCK(sc);
2306		error = 0;
2307		break;
2308	case SIOCADDMULTI:
2309	case SIOCDELMULTI:
2310		RL_LOCK(sc);
2311		re_setmulti(sc);
2312		RL_UNLOCK(sc);
2313		error = 0;
2314		break;
2315	case SIOCGIFMEDIA:
2316	case SIOCSIFMEDIA:
2317		mii = device_get_softc(sc->rl_miibus);
2318		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, command);
2319		break;
2320	case SIOCSIFCAP:
2321		ifp->if_capenable &= ~(IFCAP_HWCSUM | IFCAP_POLLING);
2322		ifp->if_capenable |=
2323		    ifr->ifr_reqcap & (IFCAP_HWCSUM | IFCAP_POLLING);
2324		if (ifp->if_capenable & IFCAP_TXCSUM)
2325			ifp->if_hwassist = RE_CSUM_FEATURES;
2326		else
2327			ifp->if_hwassist = 0;
2328		if (ifp->if_flags & IFF_RUNNING)
2329			re_init(sc);
2330		break;
2331	default:
2332		error = ether_ioctl(ifp, command, data);
2333		break;
2334	}
2335
2336	return (error);
2337}
2338
2339static void
2340re_watchdog(ifp)
2341	struct ifnet		*ifp;
2342{
2343	struct rl_softc		*sc;
2344
2345	sc = ifp->if_softc;
2346	RL_LOCK(sc);
2347	printf("re%d: watchdog timeout\n", sc->rl_unit);
2348	ifp->if_oerrors++;
2349
2350	re_txeof(sc);
2351	re_rxeof(sc);
2352	re_init_locked(sc);
2353
2354	RL_UNLOCK(sc);
2355}
2356
2357/*
2358 * Stop the adapter and free any mbufs allocated to the
2359 * RX and TX lists.
2360 */
2361static void
2362re_stop(sc)
2363	struct rl_softc		*sc;
2364{
2365	register int		i;
2366	struct ifnet		*ifp;
2367
2368	RL_LOCK_ASSERT(sc);
2369
2370	ifp = &sc->arpcom.ac_if;
2371	ifp->if_timer = 0;
2372
2373	untimeout(re_tick, sc, sc->rl_stat_ch);
2374	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
2375#ifdef DEVICE_POLLING
2376	ether_poll_deregister(ifp);
2377#endif /* DEVICE_POLLING */
2378
2379	CSR_WRITE_1(sc, RL_COMMAND, 0x00);
2380	CSR_WRITE_2(sc, RL_IMR, 0x0000);
2381
2382	if (sc->rl_head != NULL) {
2383		m_freem(sc->rl_head);
2384		sc->rl_head = sc->rl_tail = NULL;
2385	}
2386
2387	/* Free the TX list buffers. */
2388
2389	for (i = 0; i < RL_TX_DESC_CNT; i++) {
2390		if (sc->rl_ldata.rl_tx_mbuf[i] != NULL) {
2391			bus_dmamap_unload(sc->rl_ldata.rl_mtag,
2392			    sc->rl_ldata.rl_tx_dmamap[i]);
2393			m_freem(sc->rl_ldata.rl_tx_mbuf[i]);
2394			sc->rl_ldata.rl_tx_mbuf[i] = NULL;
2395		}
2396	}
2397
2398	/* Free the RX list buffers. */
2399
2400	for (i = 0; i < RL_RX_DESC_CNT; i++) {
2401		if (sc->rl_ldata.rl_rx_mbuf[i] != NULL) {
2402			bus_dmamap_unload(sc->rl_ldata.rl_mtag,
2403			    sc->rl_ldata.rl_rx_dmamap[i]);
2404			m_freem(sc->rl_ldata.rl_rx_mbuf[i]);
2405			sc->rl_ldata.rl_rx_mbuf[i] = NULL;
2406		}
2407	}
2408}
2409
2410/*
2411 * Device suspend routine.  Stop the interface and save some PCI
2412 * settings in case the BIOS doesn't restore them properly on
2413 * resume.
2414 */
2415static int
2416re_suspend(dev)
2417	device_t		dev;
2418{
2419	struct rl_softc		*sc;
2420
2421	sc = device_get_softc(dev);
2422
2423	RL_LOCK(sc);
2424	re_stop(sc);
2425	sc->suspended = 1;
2426	RL_UNLOCK(sc);
2427
2428	return (0);
2429}
2430
2431/*
2432 * Device resume routine.  Restore some PCI settings in case the BIOS
2433 * doesn't, re-enable busmastering, and restart the interface if
2434 * appropriate.
2435 */
2436static int
2437re_resume(dev)
2438	device_t		dev;
2439{
2440	struct rl_softc		*sc;
2441	struct ifnet		*ifp;
2442
2443	sc = device_get_softc(dev);
2444
2445	RL_LOCK(sc);
2446
2447	ifp = &sc->arpcom.ac_if;
2448
2449	/* reinitialize interface if necessary */
2450	if (ifp->if_flags & IFF_UP)
2451		re_init_locked(sc);
2452
2453	sc->suspended = 0;
2454	RL_UNLOCK(sc);
2455
2456	return (0);
2457}
2458
2459/*
2460 * Stop all chip I/O so that the kernel's probe routines don't
2461 * get confused by errant DMAs when rebooting.
2462 */
2463static void
2464re_shutdown(dev)
2465	device_t		dev;
2466{
2467	struct rl_softc		*sc;
2468
2469	sc = device_get_softc(dev);
2470
2471	RL_LOCK(sc);
2472	re_stop(sc);
2473	RL_UNLOCK(sc);
2474}
2475