if_re.c revision 147714
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 147714 2005-07-01 04:12:42Z yongari $");
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_types.h>
127#include <net/if_vlan_var.h>
128
129#include <net/bpf.h>
130
131#include <machine/bus.h>
132#include <machine/resource.h>
133#include <sys/bus.h>
134#include <sys/rman.h>
135
136#include <dev/mii/mii.h>
137#include <dev/mii/miivar.h>
138
139#include <dev/pci/pcireg.h>
140#include <dev/pci/pcivar.h>
141
142MODULE_DEPEND(re, pci, 1, 1, 1);
143MODULE_DEPEND(re, ether, 1, 1, 1);
144MODULE_DEPEND(re, miibus, 1, 1, 1);
145
146/* "controller miibus0" required.  See GENERIC if you get errors here. */
147#include "miibus_if.h"
148
149/*
150 * Default to using PIO access for this driver.
151 */
152#define RE_USEIOSPACE
153
154#include <pci/if_rlreg.h>
155
156#define RE_CSUM_FEATURES    (CSUM_IP | CSUM_TCP | CSUM_UDP)
157
158/*
159 * Various supported device vendors/types and their names.
160 */
161static struct rl_type re_devs[] = {
162	{ RT_VENDORID, RT_DEVICEID_8139, RL_HWREV_8139CPLUS,
163		"RealTek 8139C+ 10/100BaseTX" },
164	{ RT_VENDORID, RT_DEVICEID_8169, RL_HWREV_8169,
165		"RealTek 8169 Gigabit Ethernet" },
166	{ RT_VENDORID, RT_DEVICEID_8169, RL_HWREV_8169S,
167		"RealTek 8169S Single-chip Gigabit Ethernet" },
168	{ RT_VENDORID, RT_DEVICEID_8169, RL_HWREV_8169SB,
169		"RealTek 8169SB Single-chip Gigabit Ethernet" },
170	{ RT_VENDORID, RT_DEVICEID_8169, RL_HWREV_8110S,
171		"RealTek 8110S Single-chip Gigabit Ethernet" },
172	{ COREGA_VENDORID, COREGA_DEVICEID_CGLAPCIGT, RL_HWREV_8169S,
173		"Corega CG-LAPCIGT (RTL8169S) Gigabit Ethernet" },
174	{ 0, 0, 0, NULL }
175};
176
177static struct rl_hwrev re_hwrevs[] = {
178	{ RL_HWREV_8139, RL_8139,  "" },
179	{ RL_HWREV_8139A, RL_8139, "A" },
180	{ RL_HWREV_8139AG, RL_8139, "A-G" },
181	{ RL_HWREV_8139B, RL_8139, "B" },
182	{ RL_HWREV_8130, RL_8139, "8130" },
183	{ RL_HWREV_8139C, RL_8139, "C" },
184	{ RL_HWREV_8139D, RL_8139, "8139D/8100B/8100C" },
185	{ RL_HWREV_8139CPLUS, RL_8139CPLUS, "C+"},
186	{ RL_HWREV_8169, RL_8169, "8169"},
187	{ RL_HWREV_8169S, RL_8169, "8169S"},
188	{ RL_HWREV_8169SB, RL_8169, "8169SB"},
189	{ RL_HWREV_8110S, RL_8169, "8110S"},
190	{ RL_HWREV_8100, RL_8139, "8100"},
191	{ RL_HWREV_8101, RL_8139, "8101"},
192	{ 0, 0, NULL }
193};
194
195static int re_probe		(device_t);
196static int re_attach		(device_t);
197static int re_detach		(device_t);
198
199static int re_encap		(struct rl_softc *, struct mbuf **, int *);
200
201static void re_dma_map_addr	(void *, bus_dma_segment_t *, int, int);
202static void re_dma_map_desc	(void *, bus_dma_segment_t *, int,
203				    bus_size_t, int);
204static int re_allocmem		(device_t, struct rl_softc *);
205static int re_newbuf		(struct rl_softc *, int, struct mbuf *);
206static int re_rx_list_init	(struct rl_softc *);
207static int re_tx_list_init	(struct rl_softc *);
208#ifdef RE_FIXUP_RX
209static __inline void re_fixup_rx
210				(struct mbuf *);
211#endif
212static void re_rxeof		(struct rl_softc *);
213static void re_txeof		(struct rl_softc *);
214#ifdef DEVICE_POLLING
215static void re_poll		(struct ifnet *, enum poll_cmd, int);
216static void re_poll_locked	(struct ifnet *, enum poll_cmd, int);
217#endif
218static void re_intr		(void *);
219static void re_tick		(void *);
220static void re_tick_locked	(struct rl_softc *);
221static void re_start		(struct ifnet *);
222static void re_start_locked	(struct ifnet *);
223static int re_ioctl		(struct ifnet *, u_long, caddr_t);
224static void re_init		(void *);
225static void re_init_locked	(struct rl_softc *);
226static void re_stop		(struct rl_softc *);
227static void re_watchdog		(struct ifnet *);
228static int re_suspend		(device_t);
229static int re_resume		(device_t);
230static void re_shutdown		(device_t);
231static int re_ifmedia_upd	(struct ifnet *);
232static void re_ifmedia_sts	(struct ifnet *, struct ifmediareq *);
233
234static void re_eeprom_putbyte	(struct rl_softc *, int);
235static void re_eeprom_getword	(struct rl_softc *, int, u_int16_t *);
236static void re_read_eeprom	(struct rl_softc *, caddr_t, int, int, int);
237static int re_gmii_readreg	(device_t, int, int);
238static int re_gmii_writereg	(device_t, int, int, int);
239
240static int re_miibus_readreg	(device_t, int, int);
241static int re_miibus_writereg	(device_t, int, int, int);
242static void re_miibus_statchg	(device_t);
243
244static void re_setmulti		(struct rl_softc *);
245static void re_reset		(struct rl_softc *);
246
247static int re_diag		(struct rl_softc *);
248
249#ifdef RE_USEIOSPACE
250#define RL_RES			SYS_RES_IOPORT
251#define RL_RID			RL_PCI_LOIO
252#else
253#define RL_RES			SYS_RES_MEMORY
254#define RL_RID			RL_PCI_LOMEM
255#endif
256
257static device_method_t re_methods[] = {
258	/* Device interface */
259	DEVMETHOD(device_probe,		re_probe),
260	DEVMETHOD(device_attach,	re_attach),
261	DEVMETHOD(device_detach,	re_detach),
262	DEVMETHOD(device_suspend,	re_suspend),
263	DEVMETHOD(device_resume,	re_resume),
264	DEVMETHOD(device_shutdown,	re_shutdown),
265
266	/* bus interface */
267	DEVMETHOD(bus_print_child,	bus_generic_print_child),
268	DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
269
270	/* MII interface */
271	DEVMETHOD(miibus_readreg,	re_miibus_readreg),
272	DEVMETHOD(miibus_writereg,	re_miibus_writereg),
273	DEVMETHOD(miibus_statchg,	re_miibus_statchg),
274
275	{ 0, 0 }
276};
277
278static driver_t re_driver = {
279	"re",
280	re_methods,
281	sizeof(struct rl_softc)
282};
283
284static devclass_t re_devclass;
285
286DRIVER_MODULE(re, pci, re_driver, re_devclass, 0, 0);
287DRIVER_MODULE(re, cardbus, re_driver, re_devclass, 0, 0);
288DRIVER_MODULE(miibus, re, miibus_driver, miibus_devclass, 0, 0);
289
290#define EE_SET(x)					\
291	CSR_WRITE_1(sc, RL_EECMD,			\
292		CSR_READ_1(sc, RL_EECMD) | x)
293
294#define EE_CLR(x)					\
295	CSR_WRITE_1(sc, RL_EECMD,			\
296		CSR_READ_1(sc, RL_EECMD) & ~x)
297
298/*
299 * Send a read command and address to the EEPROM, check for ACK.
300 */
301static void
302re_eeprom_putbyte(sc, addr)
303	struct rl_softc		*sc;
304	int			addr;
305{
306	register int		d, i;
307
308	d = addr | sc->rl_eecmd_read;
309
310	/*
311	 * Feed in each bit and strobe the clock.
312	 */
313	for (i = 0x400; i; i >>= 1) {
314		if (d & i) {
315			EE_SET(RL_EE_DATAIN);
316		} else {
317			EE_CLR(RL_EE_DATAIN);
318		}
319		DELAY(100);
320		EE_SET(RL_EE_CLK);
321		DELAY(150);
322		EE_CLR(RL_EE_CLK);
323		DELAY(100);
324	}
325}
326
327/*
328 * Read a word of data stored in the EEPROM at address 'addr.'
329 */
330static void
331re_eeprom_getword(sc, addr, dest)
332	struct rl_softc		*sc;
333	int			addr;
334	u_int16_t		*dest;
335{
336	register int		i;
337	u_int16_t		word = 0;
338
339	/* Enter EEPROM access mode. */
340	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_PROGRAM|RL_EE_SEL);
341
342	/*
343	 * Send address of word we want to read.
344	 */
345	re_eeprom_putbyte(sc, addr);
346
347	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_PROGRAM|RL_EE_SEL);
348
349	/*
350	 * Start reading bits from EEPROM.
351	 */
352	for (i = 0x8000; i; i >>= 1) {
353		EE_SET(RL_EE_CLK);
354		DELAY(100);
355		if (CSR_READ_1(sc, RL_EECMD) & RL_EE_DATAOUT)
356			word |= i;
357		EE_CLR(RL_EE_CLK);
358		DELAY(100);
359	}
360
361	/* Turn off EEPROM access mode. */
362	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
363
364	*dest = word;
365}
366
367/*
368 * Read a sequence of words from the EEPROM.
369 */
370static void
371re_read_eeprom(sc, dest, off, cnt, swap)
372	struct rl_softc		*sc;
373	caddr_t			dest;
374	int			off;
375	int			cnt;
376	int			swap;
377{
378	int			i;
379	u_int16_t		word = 0, *ptr;
380
381	for (i = 0; i < cnt; i++) {
382		re_eeprom_getword(sc, off + i, &word);
383		ptr = (u_int16_t *)(dest + (i * 2));
384		if (swap)
385			*ptr = ntohs(word);
386		else
387			*ptr = word;
388	}
389}
390
391static int
392re_gmii_readreg(dev, phy, reg)
393	device_t		dev;
394	int			phy, reg;
395{
396	struct rl_softc		*sc;
397	u_int32_t		rval;
398	int			i;
399
400	if (phy != 1)
401		return (0);
402
403	sc = device_get_softc(dev);
404
405	/* Let the rgephy driver read the GMEDIASTAT register */
406
407	if (reg == RL_GMEDIASTAT) {
408		rval = CSR_READ_1(sc, RL_GMEDIASTAT);
409		return (rval);
410	}
411
412	CSR_WRITE_4(sc, RL_PHYAR, reg << 16);
413	DELAY(1000);
414
415	for (i = 0; i < RL_TIMEOUT; i++) {
416		rval = CSR_READ_4(sc, RL_PHYAR);
417		if (rval & RL_PHYAR_BUSY)
418			break;
419		DELAY(100);
420	}
421
422	if (i == RL_TIMEOUT) {
423		printf ("re%d: PHY read failed\n", sc->rl_unit);
424		return (0);
425	}
426
427	return (rval & RL_PHYAR_PHYDATA);
428}
429
430static int
431re_gmii_writereg(dev, phy, reg, data)
432	device_t		dev;
433	int			phy, reg, data;
434{
435	struct rl_softc		*sc;
436	u_int32_t		rval;
437	int			i;
438
439	sc = device_get_softc(dev);
440
441	CSR_WRITE_4(sc, RL_PHYAR, (reg << 16) |
442	    (data & RL_PHYAR_PHYDATA) | RL_PHYAR_BUSY);
443	DELAY(1000);
444
445	for (i = 0; i < RL_TIMEOUT; i++) {
446		rval = CSR_READ_4(sc, RL_PHYAR);
447		if (!(rval & RL_PHYAR_BUSY))
448			break;
449		DELAY(100);
450	}
451
452	if (i == RL_TIMEOUT) {
453		printf ("re%d: PHY write failed\n", sc->rl_unit);
454		return (0);
455	}
456
457	return (0);
458}
459
460static int
461re_miibus_readreg(dev, phy, reg)
462	device_t		dev;
463	int			phy, reg;
464{
465	struct rl_softc		*sc;
466	u_int16_t		rval = 0;
467	u_int16_t		re8139_reg = 0;
468
469	sc = device_get_softc(dev);
470
471	if (sc->rl_type == RL_8169) {
472		rval = re_gmii_readreg(dev, phy, reg);
473		return (rval);
474	}
475
476	/* Pretend the internal PHY is only at address 0 */
477	if (phy) {
478		return (0);
479	}
480	switch (reg) {
481	case MII_BMCR:
482		re8139_reg = RL_BMCR;
483		break;
484	case MII_BMSR:
485		re8139_reg = RL_BMSR;
486		break;
487	case MII_ANAR:
488		re8139_reg = RL_ANAR;
489		break;
490	case MII_ANER:
491		re8139_reg = RL_ANER;
492		break;
493	case MII_ANLPAR:
494		re8139_reg = RL_LPAR;
495		break;
496	case MII_PHYIDR1:
497	case MII_PHYIDR2:
498		return (0);
499	/*
500	 * Allow the rlphy driver to read the media status
501	 * register. If we have a link partner which does not
502	 * support NWAY, this is the register which will tell
503	 * us the results of parallel detection.
504	 */
505	case RL_MEDIASTAT:
506		rval = CSR_READ_1(sc, RL_MEDIASTAT);
507		return (rval);
508	default:
509		printf("re%d: bad phy register\n", sc->rl_unit);
510		return (0);
511	}
512	rval = CSR_READ_2(sc, re8139_reg);
513	return (rval);
514}
515
516static int
517re_miibus_writereg(dev, phy, reg, data)
518	device_t		dev;
519	int			phy, reg, data;
520{
521	struct rl_softc		*sc;
522	u_int16_t		re8139_reg = 0;
523	int			rval = 0;
524
525	sc = device_get_softc(dev);
526
527	if (sc->rl_type == RL_8169) {
528		rval = re_gmii_writereg(dev, phy, reg, data);
529		return (rval);
530	}
531
532	/* Pretend the internal PHY is only at address 0 */
533	if (phy)
534		return (0);
535
536	switch (reg) {
537	case MII_BMCR:
538		re8139_reg = RL_BMCR;
539		break;
540	case MII_BMSR:
541		re8139_reg = RL_BMSR;
542		break;
543	case MII_ANAR:
544		re8139_reg = RL_ANAR;
545		break;
546	case MII_ANER:
547		re8139_reg = RL_ANER;
548		break;
549	case MII_ANLPAR:
550		re8139_reg = RL_LPAR;
551		break;
552	case MII_PHYIDR1:
553	case MII_PHYIDR2:
554		return (0);
555		break;
556	default:
557		printf("re%d: bad phy register\n", sc->rl_unit);
558		return (0);
559	}
560	CSR_WRITE_2(sc, re8139_reg, data);
561	return (0);
562}
563
564static void
565re_miibus_statchg(dev)
566	device_t		dev;
567{
568
569}
570
571/*
572 * Program the 64-bit multicast hash filter.
573 */
574static void
575re_setmulti(sc)
576	struct rl_softc		*sc;
577{
578	struct ifnet		*ifp;
579	int			h = 0;
580	u_int32_t		hashes[2] = { 0, 0 };
581	struct ifmultiaddr	*ifma;
582	u_int32_t		rxfilt;
583	int			mcnt = 0;
584
585	RL_LOCK_ASSERT(sc);
586
587	ifp = sc->rl_ifp;
588
589	rxfilt = CSR_READ_4(sc, RL_RXCFG);
590
591	if (ifp->if_flags & IFF_ALLMULTI || ifp->if_flags & IFF_PROMISC) {
592		rxfilt |= RL_RXCFG_RX_MULTI;
593		CSR_WRITE_4(sc, RL_RXCFG, rxfilt);
594		CSR_WRITE_4(sc, RL_MAR0, 0xFFFFFFFF);
595		CSR_WRITE_4(sc, RL_MAR4, 0xFFFFFFFF);
596		return;
597	}
598
599	/* first, zot all the existing hash bits */
600	CSR_WRITE_4(sc, RL_MAR0, 0);
601	CSR_WRITE_4(sc, RL_MAR4, 0);
602
603	/* now program new ones */
604	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
605		if (ifma->ifma_addr->sa_family != AF_LINK)
606			continue;
607		h = ether_crc32_be(LLADDR((struct sockaddr_dl *)
608		    ifma->ifma_addr), ETHER_ADDR_LEN) >> 26;
609		if (h < 32)
610			hashes[0] |= (1 << h);
611		else
612			hashes[1] |= (1 << (h - 32));
613		mcnt++;
614	}
615
616	if (mcnt)
617		rxfilt |= RL_RXCFG_RX_MULTI;
618	else
619		rxfilt &= ~RL_RXCFG_RX_MULTI;
620
621	CSR_WRITE_4(sc, RL_RXCFG, rxfilt);
622	CSR_WRITE_4(sc, RL_MAR0, hashes[0]);
623	CSR_WRITE_4(sc, RL_MAR4, hashes[1]);
624}
625
626static void
627re_reset(sc)
628	struct rl_softc		*sc;
629{
630	register int		i;
631
632	RL_LOCK_ASSERT(sc);
633
634	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RESET);
635
636	for (i = 0; i < RL_TIMEOUT; i++) {
637		DELAY(10);
638		if (!(CSR_READ_1(sc, RL_COMMAND) & RL_CMD_RESET))
639			break;
640	}
641	if (i == RL_TIMEOUT)
642		printf("re%d: reset never completed!\n", sc->rl_unit);
643
644	CSR_WRITE_1(sc, 0x82, 1);
645}
646
647/*
648 * The following routine is designed to test for a defect on some
649 * 32-bit 8169 cards. Some of these NICs have the REQ64# and ACK64#
650 * lines connected to the bus, however for a 32-bit only card, they
651 * should be pulled high. The result of this defect is that the
652 * NIC will not work right if you plug it into a 64-bit slot: DMA
653 * operations will be done with 64-bit transfers, which will fail
654 * because the 64-bit data lines aren't connected.
655 *
656 * There's no way to work around this (short of talking a soldering
657 * iron to the board), however we can detect it. The method we use
658 * here is to put the NIC into digital loopback mode, set the receiver
659 * to promiscuous mode, and then try to send a frame. We then compare
660 * the frame data we sent to what was received. If the data matches,
661 * then the NIC is working correctly, otherwise we know the user has
662 * a defective NIC which has been mistakenly plugged into a 64-bit PCI
663 * slot. In the latter case, there's no way the NIC can work correctly,
664 * so we print out a message on the console and abort the device attach.
665 */
666
667static int
668re_diag(sc)
669	struct rl_softc		*sc;
670{
671	struct ifnet		*ifp = sc->rl_ifp;
672	struct mbuf		*m0;
673	struct ether_header	*eh;
674	struct rl_desc		*cur_rx;
675	u_int16_t		status;
676	u_int32_t		rxstat;
677	int			total_len, i, error = 0;
678	u_int8_t		dst[] = { 0x00, 'h', 'e', 'l', 'l', 'o' };
679	u_int8_t		src[] = { 0x00, 'w', 'o', 'r', 'l', 'd' };
680
681	/* Allocate a single mbuf */
682	MGETHDR(m0, M_DONTWAIT, MT_DATA);
683	if (m0 == NULL)
684		return (ENOBUFS);
685
686	RL_LOCK(sc);
687
688	/*
689	 * Initialize the NIC in test mode. This sets the chip up
690	 * so that it can send and receive frames, but performs the
691	 * following special functions:
692	 * - Puts receiver in promiscuous mode
693	 * - Enables digital loopback mode
694	 * - Leaves interrupts turned off
695	 */
696
697	ifp->if_flags |= IFF_PROMISC;
698	sc->rl_testmode = 1;
699	re_init_locked(sc);
700	re_stop(sc);
701	DELAY(100000);
702	re_init_locked(sc);
703
704	/* Put some data in the mbuf */
705
706	eh = mtod(m0, struct ether_header *);
707	bcopy ((char *)&dst, eh->ether_dhost, ETHER_ADDR_LEN);
708	bcopy ((char *)&src, eh->ether_shost, ETHER_ADDR_LEN);
709	eh->ether_type = htons(ETHERTYPE_IP);
710	m0->m_pkthdr.len = m0->m_len = ETHER_MIN_LEN - ETHER_CRC_LEN;
711
712	/*
713	 * Queue the packet, start transmission.
714	 * Note: IF_HANDOFF() ultimately calls re_start() for us.
715	 */
716
717	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
718	RL_UNLOCK(sc);
719	/* XXX: re_diag must not be called when in ALTQ mode */
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	bus_addr_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
1159	/*
1160	 * Allocate the parent bus DMA tag appropriate for PCI.
1161	 */
1162#define RL_NSEG_NEW 32
1163	error = bus_dma_tag_create(NULL,	/* parent */
1164			1, 0,			/* alignment, boundary */
1165			BUS_SPACE_MAXADDR_32BIT,/* lowaddr */
1166			BUS_SPACE_MAXADDR,	/* highaddr */
1167			NULL, NULL,		/* filter, filterarg */
1168			MAXBSIZE, RL_NSEG_NEW,	/* maxsize, nsegments */
1169			BUS_SPACE_MAXSIZE_32BIT,/* maxsegsize */
1170			BUS_DMA_ALLOCNOW,	/* flags */
1171			NULL, NULL,		/* lockfunc, lockarg */
1172			&sc->rl_parent_tag);
1173	if (error)
1174		goto fail;
1175
1176	error = re_allocmem(dev, sc);
1177
1178	if (error)
1179		goto fail;
1180
1181	ifp = sc->rl_ifp = if_alloc(IFT_ETHER);
1182	if (ifp == NULL) {
1183		printf("re%d: can not if_alloc()\n", sc->rl_unit);
1184		error = ENOSPC;
1185		goto fail;
1186	}
1187
1188	/* Do MII setup */
1189	if (mii_phy_probe(dev, &sc->rl_miibus,
1190	    re_ifmedia_upd, re_ifmedia_sts)) {
1191		printf("re%d: MII without any phy!\n", sc->rl_unit);
1192		error = ENXIO;
1193		goto fail;
1194	}
1195
1196	ifp->if_softc = sc;
1197	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1198	ifp->if_mtu = ETHERMTU;
1199	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1200	ifp->if_ioctl = re_ioctl;
1201	ifp->if_capabilities = IFCAP_VLAN_MTU;
1202	ifp->if_start = re_start;
1203	ifp->if_hwassist = /*RE_CSUM_FEATURES*/0;
1204	ifp->if_capabilities |= IFCAP_HWCSUM|IFCAP_VLAN_HWTAGGING;
1205#ifdef DEVICE_POLLING
1206	ifp->if_capabilities |= IFCAP_POLLING;
1207#endif
1208	ifp->if_watchdog = re_watchdog;
1209	ifp->if_init = re_init;
1210	if (sc->rl_type == RL_8169)
1211		ifp->if_baudrate = 1000000000;
1212	else
1213		ifp->if_baudrate = 100000000;
1214	IFQ_SET_MAXLEN(&ifp->if_snd,  RL_IFQ_MAXLEN);
1215	ifp->if_snd.ifq_drv_maxlen = RL_IFQ_MAXLEN;
1216	IFQ_SET_READY(&ifp->if_snd);
1217	ifp->if_capenable = ifp->if_capabilities & ~IFCAP_HWCSUM;
1218
1219	callout_handle_init(&sc->rl_stat_ch);
1220
1221	/*
1222	 * Call MI attach routine.
1223	 */
1224	ether_ifattach(ifp, eaddr);
1225
1226	/* Perform hardware diagnostic. */
1227	error = re_diag(sc);
1228
1229	if (error) {
1230		printf("re%d: attach aborted due to hardware diag failure\n",
1231		    unit);
1232		ether_ifdetach(ifp);
1233		if_free(ifp);
1234		goto fail;
1235	}
1236
1237	/* Hook interrupt last to avoid having to lock softc */
1238	error = bus_setup_intr(dev, sc->rl_irq, INTR_TYPE_NET | INTR_MPSAFE,
1239	    re_intr, sc, &sc->rl_intrhand);
1240	if (error) {
1241		printf("re%d: couldn't set up irq\n", unit);
1242		ether_ifdetach(ifp);
1243		if_free(ifp);
1244	}
1245
1246fail:
1247	if (error)
1248		re_detach(dev);
1249
1250	return (error);
1251}
1252
1253/*
1254 * Shutdown hardware and free up resources. This can be called any
1255 * time after the mutex has been initialized. It is called in both
1256 * the error case in attach and the normal detach case so it needs
1257 * to be careful about only freeing resources that have actually been
1258 * allocated.
1259 */
1260static int
1261re_detach(dev)
1262	device_t		dev;
1263{
1264	struct rl_softc		*sc;
1265	struct ifnet		*ifp;
1266	int			i;
1267	int			attached;
1268
1269	sc = device_get_softc(dev);
1270	ifp = sc->rl_ifp;
1271	KASSERT(mtx_initialized(&sc->rl_mtx), ("re mutex not initialized"));
1272
1273	attached = device_is_attached(dev);
1274	/* These should only be active if attach succeeded */
1275	if (attached)
1276		ether_ifdetach(ifp);
1277	if (ifp == NULL)
1278		if_free(ifp);
1279
1280	RL_LOCK(sc);
1281#if 0
1282	sc->suspended = 1;
1283#endif
1284
1285	/* These should only be active if attach succeeded */
1286	if (attached) {
1287		re_stop(sc);
1288		/*
1289		 * Force off the IFF_UP flag here, in case someone
1290		 * still had a BPF descriptor attached to this
1291		 * interface. If they do, ether_ifdetach() will cause
1292		 * the BPF code to try and clear the promisc mode
1293		 * flag, which will bubble down to re_ioctl(),
1294		 * which will try to call re_init() again. This will
1295		 * turn the NIC back on and restart the MII ticker,
1296		 * which will panic the system when the kernel tries
1297		 * to invoke the re_tick() function that isn't there
1298		 * anymore.
1299		 */
1300		ifp->if_flags &= ~IFF_UP;
1301	}
1302	if (sc->rl_miibus)
1303		device_delete_child(dev, sc->rl_miibus);
1304	bus_generic_detach(dev);
1305
1306	/*
1307	 * The rest is resource deallocation, so we should already be
1308	 * stopped here.
1309	 */
1310	RL_UNLOCK(sc);
1311
1312	if (sc->rl_intrhand)
1313		bus_teardown_intr(dev, sc->rl_irq, sc->rl_intrhand);
1314	if (sc->rl_irq)
1315		bus_release_resource(dev, SYS_RES_IRQ, 0, sc->rl_irq);
1316	if (sc->rl_res)
1317		bus_release_resource(dev, RL_RES, RL_RID, sc->rl_res);
1318
1319
1320	/* Unload and free the RX DMA ring memory and map */
1321
1322	if (sc->rl_ldata.rl_rx_list_tag) {
1323		bus_dmamap_unload(sc->rl_ldata.rl_rx_list_tag,
1324		    sc->rl_ldata.rl_rx_list_map);
1325		bus_dmamem_free(sc->rl_ldata.rl_rx_list_tag,
1326		    sc->rl_ldata.rl_rx_list,
1327		    sc->rl_ldata.rl_rx_list_map);
1328		bus_dma_tag_destroy(sc->rl_ldata.rl_rx_list_tag);
1329	}
1330
1331	/* Unload and free the TX DMA ring memory and map */
1332
1333	if (sc->rl_ldata.rl_tx_list_tag) {
1334		bus_dmamap_unload(sc->rl_ldata.rl_tx_list_tag,
1335		    sc->rl_ldata.rl_tx_list_map);
1336		bus_dmamem_free(sc->rl_ldata.rl_tx_list_tag,
1337		    sc->rl_ldata.rl_tx_list,
1338		    sc->rl_ldata.rl_tx_list_map);
1339		bus_dma_tag_destroy(sc->rl_ldata.rl_tx_list_tag);
1340	}
1341
1342	/* Destroy all the RX and TX buffer maps */
1343
1344	if (sc->rl_ldata.rl_mtag) {
1345		for (i = 0; i < RL_TX_DESC_CNT; i++)
1346			bus_dmamap_destroy(sc->rl_ldata.rl_mtag,
1347			    sc->rl_ldata.rl_tx_dmamap[i]);
1348		for (i = 0; i < RL_RX_DESC_CNT; i++)
1349			bus_dmamap_destroy(sc->rl_ldata.rl_mtag,
1350			    sc->rl_ldata.rl_rx_dmamap[i]);
1351		bus_dma_tag_destroy(sc->rl_ldata.rl_mtag);
1352	}
1353
1354	/* Unload and free the stats buffer and map */
1355
1356	if (sc->rl_ldata.rl_stag) {
1357		bus_dmamap_unload(sc->rl_ldata.rl_stag,
1358		    sc->rl_ldata.rl_rx_list_map);
1359		bus_dmamem_free(sc->rl_ldata.rl_stag,
1360		    sc->rl_ldata.rl_stats,
1361		    sc->rl_ldata.rl_smap);
1362		bus_dma_tag_destroy(sc->rl_ldata.rl_stag);
1363	}
1364
1365	if (sc->rl_parent_tag)
1366		bus_dma_tag_destroy(sc->rl_parent_tag);
1367
1368	mtx_destroy(&sc->rl_mtx);
1369
1370	return (0);
1371}
1372
1373static int
1374re_newbuf(sc, idx, m)
1375	struct rl_softc		*sc;
1376	int			idx;
1377	struct mbuf		*m;
1378{
1379	struct rl_dmaload_arg	arg;
1380	struct mbuf		*n = NULL;
1381	int			error;
1382
1383	if (m == NULL) {
1384		n = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
1385		if (n == NULL)
1386			return (ENOBUFS);
1387		m = n;
1388	} else
1389		m->m_data = m->m_ext.ext_buf;
1390
1391	m->m_len = m->m_pkthdr.len = MCLBYTES;
1392#ifdef RE_FIXUP_RX
1393	/*
1394	 * This is part of an evil trick to deal with non-x86 platforms.
1395	 * The RealTek chip requires RX buffers to be aligned on 64-bit
1396	 * boundaries, but that will hose non-x86 machines. To get around
1397	 * this, we leave some empty space at the start of each buffer
1398	 * and for non-x86 hosts, we copy the buffer back six bytes
1399	 * to achieve word alignment. This is slightly more efficient
1400	 * than allocating a new buffer, copying the contents, and
1401	 * discarding the old buffer.
1402	 */
1403	m_adj(m, RE_ETHER_ALIGN);
1404#endif
1405	arg.sc = sc;
1406	arg.rl_idx = idx;
1407	arg.rl_maxsegs = 1;
1408	arg.rl_flags = 0;
1409	arg.rl_ring = sc->rl_ldata.rl_rx_list;
1410
1411	error = bus_dmamap_load_mbuf(sc->rl_ldata.rl_mtag,
1412	    sc->rl_ldata.rl_rx_dmamap[idx], m, re_dma_map_desc,
1413	    &arg, BUS_DMA_NOWAIT);
1414	if (error || arg.rl_maxsegs != 1) {
1415		if (n != NULL)
1416			m_freem(n);
1417		return (ENOMEM);
1418	}
1419
1420	sc->rl_ldata.rl_rx_list[idx].rl_cmdstat |= htole32(RL_RDESC_CMD_OWN);
1421	sc->rl_ldata.rl_rx_mbuf[idx] = m;
1422
1423	bus_dmamap_sync(sc->rl_ldata.rl_mtag,
1424	    sc->rl_ldata.rl_rx_dmamap[idx],
1425	    BUS_DMASYNC_PREREAD);
1426
1427	return (0);
1428}
1429
1430#ifdef RE_FIXUP_RX
1431static __inline void
1432re_fixup_rx(m)
1433	struct mbuf		*m;
1434{
1435	int                     i;
1436	uint16_t                *src, *dst;
1437
1438	src = mtod(m, uint16_t *);
1439	dst = src - (RE_ETHER_ALIGN - ETHER_ALIGN) / sizeof *src;
1440
1441	for (i = 0; i < (m->m_len / sizeof(uint16_t) + 1); i++)
1442		*dst++ = *src++;
1443
1444	m->m_data -= RE_ETHER_ALIGN - ETHER_ALIGN;
1445
1446	return;
1447}
1448#endif
1449
1450static int
1451re_tx_list_init(sc)
1452	struct rl_softc		*sc;
1453{
1454
1455	RL_LOCK_ASSERT(sc);
1456
1457	bzero ((char *)sc->rl_ldata.rl_tx_list, RL_TX_LIST_SZ);
1458	bzero ((char *)&sc->rl_ldata.rl_tx_mbuf,
1459	    (RL_TX_DESC_CNT * sizeof(struct mbuf *)));
1460
1461	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
1462	    sc->rl_ldata.rl_tx_list_map, BUS_DMASYNC_PREWRITE);
1463	sc->rl_ldata.rl_tx_prodidx = 0;
1464	sc->rl_ldata.rl_tx_considx = 0;
1465	sc->rl_ldata.rl_tx_free = RL_TX_DESC_CNT;
1466
1467	return (0);
1468}
1469
1470static int
1471re_rx_list_init(sc)
1472	struct rl_softc		*sc;
1473{
1474	int			i;
1475
1476	bzero ((char *)sc->rl_ldata.rl_rx_list, RL_RX_LIST_SZ);
1477	bzero ((char *)&sc->rl_ldata.rl_rx_mbuf,
1478	    (RL_RX_DESC_CNT * sizeof(struct mbuf *)));
1479
1480	for (i = 0; i < RL_RX_DESC_CNT; i++) {
1481		if (re_newbuf(sc, i, NULL) == ENOBUFS)
1482			return (ENOBUFS);
1483	}
1484
1485	/* Flush the RX descriptors */
1486
1487	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1488	    sc->rl_ldata.rl_rx_list_map,
1489	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
1490
1491	sc->rl_ldata.rl_rx_prodidx = 0;
1492	sc->rl_head = sc->rl_tail = NULL;
1493
1494	return (0);
1495}
1496
1497/*
1498 * RX handler for C+ and 8169. For the gigE chips, we support
1499 * the reception of jumbo frames that have been fragmented
1500 * across multiple 2K mbuf cluster buffers.
1501 */
1502static void
1503re_rxeof(sc)
1504	struct rl_softc		*sc;
1505{
1506	struct mbuf		*m;
1507	struct ifnet		*ifp;
1508	int			i, total_len;
1509	struct rl_desc		*cur_rx;
1510	u_int32_t		rxstat, rxvlan;
1511
1512	RL_LOCK_ASSERT(sc);
1513
1514	ifp = sc->rl_ifp;
1515	i = sc->rl_ldata.rl_rx_prodidx;
1516
1517	/* Invalidate the descriptor memory */
1518
1519	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1520	    sc->rl_ldata.rl_rx_list_map,
1521	    BUS_DMASYNC_POSTREAD);
1522
1523	while (!RL_OWN(&sc->rl_ldata.rl_rx_list[i])) {
1524		cur_rx = &sc->rl_ldata.rl_rx_list[i];
1525		m = sc->rl_ldata.rl_rx_mbuf[i];
1526		total_len = RL_RXBYTES(cur_rx);
1527		rxstat = le32toh(cur_rx->rl_cmdstat);
1528		rxvlan = le32toh(cur_rx->rl_vlanctl);
1529
1530		/* Invalidate the RX mbuf and unload its map */
1531
1532		bus_dmamap_sync(sc->rl_ldata.rl_mtag,
1533		    sc->rl_ldata.rl_rx_dmamap[i],
1534		    BUS_DMASYNC_POSTWRITE);
1535		bus_dmamap_unload(sc->rl_ldata.rl_mtag,
1536		    sc->rl_ldata.rl_rx_dmamap[i]);
1537
1538		if (!(rxstat & RL_RDESC_STAT_EOF)) {
1539			m->m_len = RE_RX_DESC_BUFLEN;
1540			if (sc->rl_head == NULL)
1541				sc->rl_head = sc->rl_tail = m;
1542			else {
1543				m->m_flags &= ~M_PKTHDR;
1544				sc->rl_tail->m_next = m;
1545				sc->rl_tail = m;
1546			}
1547			re_newbuf(sc, i, NULL);
1548			RL_DESC_INC(i);
1549			continue;
1550		}
1551
1552		/*
1553		 * NOTE: for the 8139C+, the frame length field
1554		 * is always 12 bits in size, but for the gigE chips,
1555		 * it is 13 bits (since the max RX frame length is 16K).
1556		 * Unfortunately, all 32 bits in the status word
1557		 * were already used, so to make room for the extra
1558		 * length bit, RealTek took out the 'frame alignment
1559		 * error' bit and shifted the other status bits
1560		 * over one slot. The OWN, EOR, FS and LS bits are
1561		 * still in the same places. We have already extracted
1562		 * the frame length and checked the OWN bit, so rather
1563		 * than using an alternate bit mapping, we shift the
1564		 * status bits one space to the right so we can evaluate
1565		 * them using the 8169 status as though it was in the
1566		 * same format as that of the 8139C+.
1567		 */
1568		if (sc->rl_type == RL_8169)
1569			rxstat >>= 1;
1570
1571		/*
1572		 * if total_len > 2^13-1, both _RXERRSUM and _GIANT will be
1573		 * set, but if CRC is clear, it will still be a valid frame.
1574		 */
1575		if (rxstat & RL_RDESC_STAT_RXERRSUM && !(total_len > 8191 &&
1576		    (rxstat & RL_RDESC_STAT_ERRS) == RL_RDESC_STAT_GIANT)) {
1577			ifp->if_ierrors++;
1578			/*
1579			 * If this is part of a multi-fragment packet,
1580			 * discard all the pieces.
1581			 */
1582			if (sc->rl_head != NULL) {
1583				m_freem(sc->rl_head);
1584				sc->rl_head = sc->rl_tail = NULL;
1585			}
1586			re_newbuf(sc, i, m);
1587			RL_DESC_INC(i);
1588			continue;
1589		}
1590
1591		/*
1592		 * If allocating a replacement mbuf fails,
1593		 * reload the current one.
1594		 */
1595
1596		if (re_newbuf(sc, i, NULL)) {
1597			ifp->if_ierrors++;
1598			if (sc->rl_head != NULL) {
1599				m_freem(sc->rl_head);
1600				sc->rl_head = sc->rl_tail = NULL;
1601			}
1602			re_newbuf(sc, i, m);
1603			RL_DESC_INC(i);
1604			continue;
1605		}
1606
1607		RL_DESC_INC(i);
1608
1609		if (sc->rl_head != NULL) {
1610			m->m_len = total_len % RE_RX_DESC_BUFLEN;
1611			if (m->m_len == 0)
1612				m->m_len = RE_RX_DESC_BUFLEN;
1613			/*
1614			 * Special case: if there's 4 bytes or less
1615			 * in this buffer, the mbuf can be discarded:
1616			 * the last 4 bytes is the CRC, which we don't
1617			 * care about anyway.
1618			 */
1619			if (m->m_len <= ETHER_CRC_LEN) {
1620				sc->rl_tail->m_len -=
1621				    (ETHER_CRC_LEN - m->m_len);
1622				m_freem(m);
1623			} else {
1624				m->m_len -= ETHER_CRC_LEN;
1625				m->m_flags &= ~M_PKTHDR;
1626				sc->rl_tail->m_next = m;
1627			}
1628			m = sc->rl_head;
1629			sc->rl_head = sc->rl_tail = NULL;
1630			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
1631		} else
1632			m->m_pkthdr.len = m->m_len =
1633			    (total_len - ETHER_CRC_LEN);
1634
1635#ifdef RE_FIXUP_RX
1636		re_fixup_rx(m);
1637#endif
1638		ifp->if_ipackets++;
1639		m->m_pkthdr.rcvif = ifp;
1640
1641		/* Do RX checksumming if enabled */
1642
1643		if (ifp->if_capenable & IFCAP_RXCSUM) {
1644
1645			/* Check IP header checksum */
1646			if (rxstat & RL_RDESC_STAT_PROTOID)
1647				m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED;
1648			if (!(rxstat & RL_RDESC_STAT_IPSUMBAD))
1649				m->m_pkthdr.csum_flags |= CSUM_IP_VALID;
1650
1651			/* Check TCP/UDP checksum */
1652			if ((RL_TCPPKT(rxstat) &&
1653			    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
1654			    (RL_UDPPKT(rxstat) &&
1655			    !(rxstat & RL_RDESC_STAT_UDPSUMBAD))) {
1656				m->m_pkthdr.csum_flags |=
1657				    CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
1658				m->m_pkthdr.csum_data = 0xffff;
1659			}
1660		}
1661
1662		if (rxvlan & RL_RDESC_VLANCTL_TAG)
1663			VLAN_INPUT_TAG(ifp, m,
1664			    ntohs((rxvlan & RL_RDESC_VLANCTL_DATA)), continue);
1665		RL_UNLOCK(sc);
1666		(*ifp->if_input)(ifp, m);
1667		RL_LOCK(sc);
1668	}
1669
1670	/* Flush the RX DMA ring */
1671
1672	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1673	    sc->rl_ldata.rl_rx_list_map,
1674	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
1675
1676	sc->rl_ldata.rl_rx_prodidx = i;
1677}
1678
1679static void
1680re_txeof(sc)
1681	struct rl_softc		*sc;
1682{
1683	struct ifnet		*ifp;
1684	u_int32_t		txstat;
1685	int			idx;
1686
1687	ifp = sc->rl_ifp;
1688	idx = sc->rl_ldata.rl_tx_considx;
1689
1690	/* Invalidate the TX descriptor list */
1691
1692	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
1693	    sc->rl_ldata.rl_tx_list_map,
1694	    BUS_DMASYNC_POSTREAD);
1695
1696	while (idx != sc->rl_ldata.rl_tx_prodidx) {
1697
1698		txstat = le32toh(sc->rl_ldata.rl_tx_list[idx].rl_cmdstat);
1699		if (txstat & RL_TDESC_CMD_OWN)
1700			break;
1701
1702		/*
1703		 * We only stash mbufs in the last descriptor
1704		 * in a fragment chain, which also happens to
1705		 * be the only place where the TX status bits
1706		 * are valid.
1707		 */
1708
1709		if (txstat & RL_TDESC_CMD_EOF) {
1710			m_freem(sc->rl_ldata.rl_tx_mbuf[idx]);
1711			sc->rl_ldata.rl_tx_mbuf[idx] = NULL;
1712			bus_dmamap_unload(sc->rl_ldata.rl_mtag,
1713			    sc->rl_ldata.rl_tx_dmamap[idx]);
1714			if (txstat & (RL_TDESC_STAT_EXCESSCOL|
1715			    RL_TDESC_STAT_COLCNT))
1716				ifp->if_collisions++;
1717			if (txstat & RL_TDESC_STAT_TXERRSUM)
1718				ifp->if_oerrors++;
1719			else
1720				ifp->if_opackets++;
1721		}
1722		sc->rl_ldata.rl_tx_free++;
1723		RL_DESC_INC(idx);
1724	}
1725
1726	/* No changes made to the TX ring, so no flush needed */
1727
1728	if (idx != sc->rl_ldata.rl_tx_considx) {
1729		sc->rl_ldata.rl_tx_considx = idx;
1730		ifp->if_flags &= ~IFF_OACTIVE;
1731		ifp->if_timer = 0;
1732	}
1733
1734	/*
1735	 * If not all descriptors have been released reaped yet,
1736	 * reload the timer so that we will eventually get another
1737	 * interrupt that will cause us to re-enter this routine.
1738	 * This is done in case the transmitter has gone idle.
1739	 */
1740	if (sc->rl_ldata.rl_tx_free != RL_TX_DESC_CNT)
1741		CSR_WRITE_4(sc, RL_TIMERCNT, 1);
1742}
1743
1744static void
1745re_tick(xsc)
1746	void			*xsc;
1747{
1748	struct rl_softc		*sc;
1749
1750	sc = xsc;
1751	RL_LOCK(sc);
1752	re_tick_locked(sc);
1753	RL_UNLOCK(sc);
1754}
1755
1756static void
1757re_tick_locked(sc)
1758	struct rl_softc		*sc;
1759{
1760	struct mii_data		*mii;
1761
1762	RL_LOCK_ASSERT(sc);
1763
1764	mii = device_get_softc(sc->rl_miibus);
1765
1766	mii_tick(mii);
1767
1768	sc->rl_stat_ch = timeout(re_tick, sc, hz);
1769}
1770
1771#ifdef DEVICE_POLLING
1772static void
1773re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
1774{
1775	struct rl_softc *sc = ifp->if_softc;
1776
1777	RL_LOCK(sc);
1778	re_poll_locked(ifp, cmd, count);
1779	RL_UNLOCK(sc);
1780}
1781
1782static void
1783re_poll_locked(struct ifnet *ifp, enum poll_cmd cmd, int count)
1784{
1785	struct rl_softc *sc = ifp->if_softc;
1786
1787	RL_LOCK_ASSERT(sc);
1788
1789	if (!(ifp->if_capenable & IFCAP_POLLING)) {
1790		ether_poll_deregister(ifp);
1791		cmd = POLL_DEREGISTER;
1792	}
1793	if (cmd == POLL_DEREGISTER) { /* final call, enable interrupts */
1794		CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
1795		return;
1796	}
1797
1798	sc->rxcycles = count;
1799	re_rxeof(sc);
1800	re_txeof(sc);
1801
1802	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
1803		re_start_locked(ifp);
1804
1805	if (cmd == POLL_AND_CHECK_STATUS) { /* also check status register */
1806		u_int16_t       status;
1807
1808		status = CSR_READ_2(sc, RL_ISR);
1809		if (status == 0xffff)
1810			return;
1811		if (status)
1812			CSR_WRITE_2(sc, RL_ISR, status);
1813
1814		/*
1815		 * XXX check behaviour on receiver stalls.
1816		 */
1817
1818		if (status & RL_ISR_SYSTEM_ERR) {
1819			re_reset(sc);
1820			re_init_locked(sc);
1821		}
1822	}
1823}
1824#endif /* DEVICE_POLLING */
1825
1826static void
1827re_intr(arg)
1828	void			*arg;
1829{
1830	struct rl_softc		*sc;
1831	struct ifnet		*ifp;
1832	u_int16_t		status;
1833
1834	sc = arg;
1835
1836	RL_LOCK(sc);
1837
1838	ifp = sc->rl_ifp;
1839
1840	if (sc->suspended || !(ifp->if_flags & IFF_UP))
1841		goto done_locked;
1842
1843#ifdef DEVICE_POLLING
1844	if  (ifp->if_flags & IFF_POLLING)
1845		goto done_locked;
1846	if ((ifp->if_capenable & IFCAP_POLLING) &&
1847	    ether_poll_register(re_poll, ifp)) { /* ok, disable interrupts */
1848		CSR_WRITE_2(sc, RL_IMR, 0x0000);
1849		re_poll_locked(ifp, 0, 1);
1850		goto done_locked;
1851	}
1852#endif /* DEVICE_POLLING */
1853
1854	for (;;) {
1855
1856		status = CSR_READ_2(sc, RL_ISR);
1857		/* If the card has gone away the read returns 0xffff. */
1858		if (status == 0xffff)
1859			break;
1860		if (status)
1861			CSR_WRITE_2(sc, RL_ISR, status);
1862
1863		if ((status & RL_INTRS_CPLUS) == 0)
1864			break;
1865
1866		if ((status & RL_ISR_RX_OK) ||
1867		    (status & RL_ISR_RX_ERR))
1868			re_rxeof(sc);
1869
1870		if ((status & RL_ISR_TIMEOUT_EXPIRED) ||
1871		    (status & RL_ISR_TX_ERR) ||
1872		    (status & RL_ISR_TX_DESC_UNAVAIL))
1873			re_txeof(sc);
1874
1875		if (status & RL_ISR_SYSTEM_ERR) {
1876			re_reset(sc);
1877			re_init_locked(sc);
1878		}
1879
1880		if (status & RL_ISR_LINKCHG) {
1881			untimeout(re_tick, sc, sc->rl_stat_ch);
1882			re_tick_locked(sc);
1883		}
1884	}
1885
1886	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
1887		re_start_locked(ifp);
1888
1889done_locked:
1890	RL_UNLOCK(sc);
1891}
1892
1893static int
1894re_encap(sc, m_head, idx)
1895	struct rl_softc		*sc;
1896	struct mbuf		**m_head;
1897	int			*idx;
1898{
1899	struct mbuf		*m_new = NULL;
1900	struct rl_dmaload_arg	arg;
1901	bus_dmamap_t		map;
1902	int			error;
1903	struct m_tag		*mtag;
1904
1905	RL_LOCK_ASSERT(sc);
1906
1907	if (sc->rl_ldata.rl_tx_free <= 4)
1908		return (EFBIG);
1909
1910	/*
1911	 * Set up checksum offload. Note: checksum offload bits must
1912	 * appear in all descriptors of a multi-descriptor transmit
1913	 * attempt. This is according to testing done with an 8169
1914	 * chip. This is a requirement.
1915	 */
1916
1917	arg.rl_flags = 0;
1918
1919	if ((*m_head)->m_pkthdr.csum_flags & CSUM_IP)
1920		arg.rl_flags |= RL_TDESC_CMD_IPCSUM;
1921	if ((*m_head)->m_pkthdr.csum_flags & CSUM_TCP)
1922		arg.rl_flags |= RL_TDESC_CMD_TCPCSUM;
1923	if ((*m_head)->m_pkthdr.csum_flags & CSUM_UDP)
1924		arg.rl_flags |= RL_TDESC_CMD_UDPCSUM;
1925
1926	arg.sc = sc;
1927	arg.rl_idx = *idx;
1928	arg.rl_maxsegs = sc->rl_ldata.rl_tx_free;
1929	if (arg.rl_maxsegs > 4)
1930		arg.rl_maxsegs -= 4;
1931	arg.rl_ring = sc->rl_ldata.rl_tx_list;
1932
1933	map = sc->rl_ldata.rl_tx_dmamap[*idx];
1934	error = bus_dmamap_load_mbuf(sc->rl_ldata.rl_mtag, map,
1935	    *m_head, re_dma_map_desc, &arg, BUS_DMA_NOWAIT);
1936
1937	if (error && error != EFBIG) {
1938		printf("re%d: can't map mbuf (error %d)\n", sc->rl_unit, error);
1939		return (ENOBUFS);
1940	}
1941
1942	/* Too many segments to map, coalesce into a single mbuf */
1943
1944	if (error || arg.rl_maxsegs == 0) {
1945		m_new = m_defrag(*m_head, M_DONTWAIT);
1946		if (m_new == NULL)
1947			return (ENOBUFS);
1948		else
1949			*m_head = m_new;
1950
1951		arg.sc = sc;
1952		arg.rl_idx = *idx;
1953		arg.rl_maxsegs = sc->rl_ldata.rl_tx_free;
1954		arg.rl_ring = sc->rl_ldata.rl_tx_list;
1955
1956		error = bus_dmamap_load_mbuf(sc->rl_ldata.rl_mtag, map,
1957		    *m_head, re_dma_map_desc, &arg, BUS_DMA_NOWAIT);
1958		if (error) {
1959			printf("re%d: can't map mbuf (error %d)\n",
1960			    sc->rl_unit, error);
1961			return (EFBIG);
1962		}
1963	}
1964
1965	/*
1966	 * Insure that the map for this transmission
1967	 * is placed at the array index of the last descriptor
1968	 * in this chain.  (Swap last and first dmamaps.)
1969	 */
1970	sc->rl_ldata.rl_tx_dmamap[*idx] =
1971	    sc->rl_ldata.rl_tx_dmamap[arg.rl_idx];
1972	sc->rl_ldata.rl_tx_dmamap[arg.rl_idx] = map;
1973
1974	sc->rl_ldata.rl_tx_mbuf[arg.rl_idx] = *m_head;
1975	sc->rl_ldata.rl_tx_free -= arg.rl_maxsegs;
1976
1977	/*
1978	 * Set up hardware VLAN tagging. Note: vlan tag info must
1979	 * appear in the first descriptor of a multi-descriptor
1980	 * transmission attempt.
1981	 */
1982
1983	mtag = VLAN_OUTPUT_TAG(sc->rl_ifp, *m_head);
1984	if (mtag != NULL)
1985		sc->rl_ldata.rl_tx_list[*idx].rl_vlanctl =
1986		    htole32(htons(VLAN_TAG_VALUE(mtag)) | RL_TDESC_VLANCTL_TAG);
1987
1988	/* Transfer ownership of packet to the chip. */
1989
1990	sc->rl_ldata.rl_tx_list[arg.rl_idx].rl_cmdstat |=
1991	    htole32(RL_TDESC_CMD_OWN);
1992	if (*idx != arg.rl_idx)
1993		sc->rl_ldata.rl_tx_list[*idx].rl_cmdstat |=
1994		    htole32(RL_TDESC_CMD_OWN);
1995
1996	RL_DESC_INC(arg.rl_idx);
1997	*idx = arg.rl_idx;
1998
1999	return (0);
2000}
2001
2002static void
2003re_start(ifp)
2004	struct ifnet		*ifp;
2005{
2006	struct rl_softc		*sc;
2007
2008	sc = ifp->if_softc;
2009	RL_LOCK(sc);
2010	re_start_locked(ifp);
2011	RL_UNLOCK(sc);
2012}
2013
2014/*
2015 * Main transmit routine for C+ and gigE NICs.
2016 */
2017static void
2018re_start_locked(ifp)
2019	struct ifnet		*ifp;
2020{
2021	struct rl_softc		*sc;
2022	struct mbuf		*m_head = NULL;
2023	int			idx, queued = 0;
2024
2025	sc = ifp->if_softc;
2026
2027	RL_LOCK_ASSERT(sc);
2028
2029	idx = sc->rl_ldata.rl_tx_prodidx;
2030
2031	while (sc->rl_ldata.rl_tx_mbuf[idx] == NULL) {
2032		IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
2033		if (m_head == NULL)
2034			break;
2035
2036		if (re_encap(sc, &m_head, &idx)) {
2037			IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
2038			ifp->if_flags |= IFF_OACTIVE;
2039			break;
2040		}
2041
2042		/*
2043		 * If there's a BPF listener, bounce a copy of this frame
2044		 * to him.
2045		 */
2046		BPF_MTAP(ifp, m_head);
2047
2048		queued++;
2049	}
2050
2051	if (queued == 0)
2052		return;
2053
2054	/* Flush the TX descriptors */
2055
2056	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2057	    sc->rl_ldata.rl_tx_list_map,
2058	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2059
2060	sc->rl_ldata.rl_tx_prodidx = idx;
2061
2062	/*
2063	 * RealTek put the TX poll request register in a different
2064	 * location on the 8169 gigE chip. I don't know why.
2065	 */
2066
2067	if (sc->rl_type == RL_8169)
2068		CSR_WRITE_2(sc, RL_GTXSTART, RL_TXSTART_START);
2069	else
2070		CSR_WRITE_2(sc, RL_TXSTART, RL_TXSTART_START);
2071
2072	/*
2073	 * Use the countdown timer for interrupt moderation.
2074	 * 'TX done' interrupts are disabled. Instead, we reset the
2075	 * countdown timer, which will begin counting until it hits
2076	 * the value in the TIMERINT register, and then trigger an
2077	 * interrupt. Each time we write to the TIMERCNT register,
2078	 * the timer count is reset to 0.
2079	 */
2080	CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2081
2082	/*
2083	 * Set a timeout in case the chip goes out to lunch.
2084	 */
2085	ifp->if_timer = 5;
2086}
2087
2088static void
2089re_init(xsc)
2090	void			*xsc;
2091{
2092	struct rl_softc		*sc = xsc;
2093
2094	RL_LOCK(sc);
2095	re_init_locked(sc);
2096	RL_UNLOCK(sc);
2097}
2098
2099static void
2100re_init_locked(sc)
2101	struct rl_softc		*sc;
2102{
2103	struct ifnet		*ifp = sc->rl_ifp;
2104	struct mii_data		*mii;
2105	u_int32_t		rxcfg = 0;
2106
2107	RL_LOCK_ASSERT(sc);
2108
2109	mii = device_get_softc(sc->rl_miibus);
2110
2111	/*
2112	 * Cancel pending I/O and free all RX/TX buffers.
2113	 */
2114	re_stop(sc);
2115
2116	/*
2117	 * Enable C+ RX and TX mode, as well as VLAN stripping and
2118	 * RX checksum offload. We must configure the C+ register
2119	 * before all others.
2120	 */
2121	CSR_WRITE_2(sc, RL_CPLUS_CMD, RL_CPLUSCMD_RXENB|
2122	    RL_CPLUSCMD_TXENB|RL_CPLUSCMD_PCI_MRW|
2123	    RL_CPLUSCMD_VLANSTRIP|
2124	    (ifp->if_capenable & IFCAP_RXCSUM ?
2125	    RL_CPLUSCMD_RXCSUM_ENB : 0));
2126
2127	/*
2128	 * Init our MAC address.  Even though the chipset
2129	 * documentation doesn't mention it, we need to enter "Config
2130	 * register write enable" mode to modify the ID registers.
2131	 */
2132	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG);
2133	CSR_WRITE_STREAM_4(sc, RL_IDR0,
2134	    *(u_int32_t *)(&IFP2ENADDR(sc->rl_ifp)[0]));
2135	CSR_WRITE_STREAM_4(sc, RL_IDR4,
2136	    *(u_int32_t *)(&IFP2ENADDR(sc->rl_ifp)[4]));
2137	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
2138
2139	/*
2140	 * For C+ mode, initialize the RX descriptors and mbufs.
2141	 */
2142	re_rx_list_init(sc);
2143	re_tx_list_init(sc);
2144
2145	/*
2146	 * Enable transmit and receive.
2147	 */
2148	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
2149
2150	/*
2151	 * Set the initial TX and RX configuration.
2152	 */
2153	if (sc->rl_testmode) {
2154		if (sc->rl_type == RL_8169)
2155			CSR_WRITE_4(sc, RL_TXCFG,
2156			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON);
2157		else
2158			CSR_WRITE_4(sc, RL_TXCFG,
2159			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON_CPLUS);
2160	} else
2161		CSR_WRITE_4(sc, RL_TXCFG, RL_TXCFG_CONFIG);
2162	CSR_WRITE_4(sc, RL_RXCFG, RL_RXCFG_CONFIG);
2163
2164	/* Set the individual bit to receive frames for this host only. */
2165	rxcfg = CSR_READ_4(sc, RL_RXCFG);
2166	rxcfg |= RL_RXCFG_RX_INDIV;
2167
2168	/* If we want promiscuous mode, set the allframes bit. */
2169	if (ifp->if_flags & IFF_PROMISC)
2170		rxcfg |= RL_RXCFG_RX_ALLPHYS;
2171	else
2172		rxcfg &= ~RL_RXCFG_RX_ALLPHYS;
2173	CSR_WRITE_4(sc, RL_RXCFG, rxcfg);
2174
2175	/*
2176	 * Set capture broadcast bit to capture broadcast frames.
2177	 */
2178	if (ifp->if_flags & IFF_BROADCAST)
2179		rxcfg |= RL_RXCFG_RX_BROAD;
2180	else
2181		rxcfg &= ~RL_RXCFG_RX_BROAD;
2182	CSR_WRITE_4(sc, RL_RXCFG, rxcfg);
2183
2184	/*
2185	 * Program the multicast filter, if necessary.
2186	 */
2187	re_setmulti(sc);
2188
2189#ifdef DEVICE_POLLING
2190	/*
2191	 * Disable interrupts if we are polling.
2192	 */
2193	if (ifp->if_flags & IFF_POLLING)
2194		CSR_WRITE_2(sc, RL_IMR, 0);
2195	else	/* otherwise ... */
2196#endif /* DEVICE_POLLING */
2197	/*
2198	 * Enable interrupts.
2199	 */
2200	if (sc->rl_testmode)
2201		CSR_WRITE_2(sc, RL_IMR, 0);
2202	else
2203		CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2204
2205	/* Set initial TX threshold */
2206	sc->rl_txthresh = RL_TX_THRESH_INIT;
2207
2208	/* Start RX/TX process. */
2209	CSR_WRITE_4(sc, RL_MISSEDPKT, 0);
2210#ifdef notdef
2211	/* Enable receiver and transmitter. */
2212	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
2213#endif
2214	/*
2215	 * Load the addresses of the RX and TX lists into the chip.
2216	 */
2217
2218	CSR_WRITE_4(sc, RL_RXLIST_ADDR_HI,
2219	    RL_ADDR_HI(sc->rl_ldata.rl_rx_list_addr));
2220	CSR_WRITE_4(sc, RL_RXLIST_ADDR_LO,
2221	    RL_ADDR_LO(sc->rl_ldata.rl_rx_list_addr));
2222
2223	CSR_WRITE_4(sc, RL_TXLIST_ADDR_HI,
2224	    RL_ADDR_HI(sc->rl_ldata.rl_tx_list_addr));
2225	CSR_WRITE_4(sc, RL_TXLIST_ADDR_LO,
2226	    RL_ADDR_LO(sc->rl_ldata.rl_tx_list_addr));
2227
2228	CSR_WRITE_1(sc, RL_EARLY_TX_THRESH, 16);
2229
2230	/*
2231	 * Initialize the timer interrupt register so that
2232	 * a timer interrupt will be generated once the timer
2233	 * reaches a certain number of ticks. The timer is
2234	 * reloaded on each transmit. This gives us TX interrupt
2235	 * moderation, which dramatically improves TX frame rate.
2236	 */
2237	if (sc->rl_type == RL_8169)
2238		CSR_WRITE_4(sc, RL_TIMERINT_8169, 0x800);
2239	else
2240		CSR_WRITE_4(sc, RL_TIMERINT, 0x400);
2241
2242	/*
2243	 * For 8169 gigE NICs, set the max allowed RX packet
2244	 * size so we can receive jumbo frames.
2245	 */
2246	if (sc->rl_type == RL_8169)
2247		CSR_WRITE_2(sc, RL_MAXRXPKTLEN, 16383);
2248
2249	if (sc->rl_testmode)
2250		return;
2251
2252	mii_mediachg(mii);
2253
2254	CSR_WRITE_1(sc, RL_CFG1, RL_CFG1_DRVLOAD|RL_CFG1_FULLDUPLEX);
2255
2256	ifp->if_flags |= IFF_RUNNING;
2257	ifp->if_flags &= ~IFF_OACTIVE;
2258
2259	sc->rl_stat_ch = timeout(re_tick, sc, hz);
2260}
2261
2262/*
2263 * Set media options.
2264 */
2265static int
2266re_ifmedia_upd(ifp)
2267	struct ifnet		*ifp;
2268{
2269	struct rl_softc		*sc;
2270	struct mii_data		*mii;
2271
2272	sc = ifp->if_softc;
2273	mii = device_get_softc(sc->rl_miibus);
2274	mii_mediachg(mii);
2275
2276	return (0);
2277}
2278
2279/*
2280 * Report current media status.
2281 */
2282static void
2283re_ifmedia_sts(ifp, ifmr)
2284	struct ifnet		*ifp;
2285	struct ifmediareq	*ifmr;
2286{
2287	struct rl_softc		*sc;
2288	struct mii_data		*mii;
2289
2290	sc = ifp->if_softc;
2291	mii = device_get_softc(sc->rl_miibus);
2292
2293	mii_pollstat(mii);
2294	ifmr->ifm_active = mii->mii_media_active;
2295	ifmr->ifm_status = mii->mii_media_status;
2296}
2297
2298static int
2299re_ioctl(ifp, command, data)
2300	struct ifnet		*ifp;
2301	u_long			command;
2302	caddr_t			data;
2303{
2304	struct rl_softc		*sc = ifp->if_softc;
2305	struct ifreq		*ifr = (struct ifreq *) data;
2306	struct mii_data		*mii;
2307	int			error = 0;
2308
2309	switch (command) {
2310	case SIOCSIFMTU:
2311		if (ifr->ifr_mtu > RL_JUMBO_MTU)
2312			error = EINVAL;
2313		ifp->if_mtu = ifr->ifr_mtu;
2314		break;
2315	case SIOCSIFFLAGS:
2316		RL_LOCK(sc);
2317		if (ifp->if_flags & IFF_UP)
2318			re_init_locked(sc);
2319		else if (ifp->if_flags & IFF_RUNNING)
2320			re_stop(sc);
2321		RL_UNLOCK(sc);
2322		error = 0;
2323		break;
2324	case SIOCADDMULTI:
2325	case SIOCDELMULTI:
2326		RL_LOCK(sc);
2327		re_setmulti(sc);
2328		RL_UNLOCK(sc);
2329		error = 0;
2330		break;
2331	case SIOCGIFMEDIA:
2332	case SIOCSIFMEDIA:
2333		mii = device_get_softc(sc->rl_miibus);
2334		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, command);
2335		break;
2336	case SIOCSIFCAP:
2337		ifp->if_capenable &= ~(IFCAP_HWCSUM | IFCAP_POLLING);
2338		ifp->if_capenable |=
2339		    ifr->ifr_reqcap & (IFCAP_HWCSUM | IFCAP_POLLING);
2340		if (ifp->if_capenable & IFCAP_TXCSUM)
2341			ifp->if_hwassist = RE_CSUM_FEATURES;
2342		else
2343			ifp->if_hwassist = 0;
2344		if (ifp->if_flags & IFF_RUNNING)
2345			re_init(sc);
2346		break;
2347	default:
2348		error = ether_ioctl(ifp, command, data);
2349		break;
2350	}
2351
2352	return (error);
2353}
2354
2355static void
2356re_watchdog(ifp)
2357	struct ifnet		*ifp;
2358{
2359	struct rl_softc		*sc;
2360
2361	sc = ifp->if_softc;
2362	RL_LOCK(sc);
2363	printf("re%d: watchdog timeout\n", sc->rl_unit);
2364	ifp->if_oerrors++;
2365
2366	re_txeof(sc);
2367	re_rxeof(sc);
2368	re_init_locked(sc);
2369
2370	RL_UNLOCK(sc);
2371}
2372
2373/*
2374 * Stop the adapter and free any mbufs allocated to the
2375 * RX and TX lists.
2376 */
2377static void
2378re_stop(sc)
2379	struct rl_softc		*sc;
2380{
2381	register int		i;
2382	struct ifnet		*ifp;
2383
2384	RL_LOCK_ASSERT(sc);
2385
2386	ifp = sc->rl_ifp;
2387	ifp->if_timer = 0;
2388
2389	untimeout(re_tick, sc, sc->rl_stat_ch);
2390	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
2391#ifdef DEVICE_POLLING
2392	ether_poll_deregister(ifp);
2393#endif /* DEVICE_POLLING */
2394
2395	CSR_WRITE_1(sc, RL_COMMAND, 0x00);
2396	CSR_WRITE_2(sc, RL_IMR, 0x0000);
2397
2398	if (sc->rl_head != NULL) {
2399		m_freem(sc->rl_head);
2400		sc->rl_head = sc->rl_tail = NULL;
2401	}
2402
2403	/* Free the TX list buffers. */
2404
2405	for (i = 0; i < RL_TX_DESC_CNT; i++) {
2406		if (sc->rl_ldata.rl_tx_mbuf[i] != NULL) {
2407			bus_dmamap_unload(sc->rl_ldata.rl_mtag,
2408			    sc->rl_ldata.rl_tx_dmamap[i]);
2409			m_freem(sc->rl_ldata.rl_tx_mbuf[i]);
2410			sc->rl_ldata.rl_tx_mbuf[i] = NULL;
2411		}
2412	}
2413
2414	/* Free the RX list buffers. */
2415
2416	for (i = 0; i < RL_RX_DESC_CNT; i++) {
2417		if (sc->rl_ldata.rl_rx_mbuf[i] != NULL) {
2418			bus_dmamap_unload(sc->rl_ldata.rl_mtag,
2419			    sc->rl_ldata.rl_rx_dmamap[i]);
2420			m_freem(sc->rl_ldata.rl_rx_mbuf[i]);
2421			sc->rl_ldata.rl_rx_mbuf[i] = NULL;
2422		}
2423	}
2424}
2425
2426/*
2427 * Device suspend routine.  Stop the interface and save some PCI
2428 * settings in case the BIOS doesn't restore them properly on
2429 * resume.
2430 */
2431static int
2432re_suspend(dev)
2433	device_t		dev;
2434{
2435	struct rl_softc		*sc;
2436
2437	sc = device_get_softc(dev);
2438
2439	RL_LOCK(sc);
2440	re_stop(sc);
2441	sc->suspended = 1;
2442	RL_UNLOCK(sc);
2443
2444	return (0);
2445}
2446
2447/*
2448 * Device resume routine.  Restore some PCI settings in case the BIOS
2449 * doesn't, re-enable busmastering, and restart the interface if
2450 * appropriate.
2451 */
2452static int
2453re_resume(dev)
2454	device_t		dev;
2455{
2456	struct rl_softc		*sc;
2457	struct ifnet		*ifp;
2458
2459	sc = device_get_softc(dev);
2460
2461	RL_LOCK(sc);
2462
2463	ifp = sc->rl_ifp;
2464
2465	/* reinitialize interface if necessary */
2466	if (ifp->if_flags & IFF_UP)
2467		re_init_locked(sc);
2468
2469	sc->suspended = 0;
2470	RL_UNLOCK(sc);
2471
2472	return (0);
2473}
2474
2475/*
2476 * Stop all chip I/O so that the kernel's probe routines don't
2477 * get confused by errant DMAs when rebooting.
2478 */
2479static void
2480re_shutdown(dev)
2481	device_t		dev;
2482{
2483	struct rl_softc		*sc;
2484
2485	sc = device_get_softc(dev);
2486
2487	RL_LOCK(sc);
2488	re_stop(sc);
2489	RL_UNLOCK(sc);
2490}
2491