if_re.c revision 150968
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 150968 2005-10-05 10:09:17Z glebius $");
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#ifdef HAVE_KERNEL_OPTION_HEADERS
112#include "opt_device_polling.h"
113#endif
114
115#include <sys/param.h>
116#include <sys/endian.h>
117#include <sys/systm.h>
118#include <sys/sockio.h>
119#include <sys/mbuf.h>
120#include <sys/malloc.h>
121#include <sys/module.h>
122#include <sys/kernel.h>
123#include <sys/socket.h>
124
125#include <net/if.h>
126#include <net/if_arp.h>
127#include <net/ethernet.h>
128#include <net/if_dl.h>
129#include <net/if_media.h>
130#include <net/if_types.h>
131#include <net/if_vlan_var.h>
132
133#include <net/bpf.h>
134
135#include <machine/bus.h>
136#include <machine/resource.h>
137#include <sys/bus.h>
138#include <sys/rman.h>
139
140#include <dev/mii/mii.h>
141#include <dev/mii/miivar.h>
142
143#include <dev/pci/pcireg.h>
144#include <dev/pci/pcivar.h>
145
146MODULE_DEPEND(re, pci, 1, 1, 1);
147MODULE_DEPEND(re, ether, 1, 1, 1);
148MODULE_DEPEND(re, miibus, 1, 1, 1);
149
150/* "controller miibus0" required.  See GENERIC if you get errors here. */
151#include "miibus_if.h"
152
153/*
154 * Default to using PIO access for this driver.
155 */
156#define RE_USEIOSPACE
157
158#include <pci/if_rlreg.h>
159
160#define RE_CSUM_FEATURES    (CSUM_IP | CSUM_TCP | CSUM_UDP)
161
162/*
163 * Various supported device vendors/types and their names.
164 */
165static struct rl_type re_devs[] = {
166	{ DLINK_VENDORID, DLINK_DEVICEID_528T, RL_HWREV_8169S,
167		"D-Link DGE-528(T) Gigabit Ethernet Adapter" },
168	{ RT_VENDORID, RT_DEVICEID_8139, RL_HWREV_8139CPLUS,
169		"RealTek 8139C+ 10/100BaseTX" },
170	{ RT_VENDORID, RT_DEVICEID_8169, RL_HWREV_8169,
171		"RealTek 8169 Gigabit Ethernet" },
172	{ RT_VENDORID, RT_DEVICEID_8169, RL_HWREV_8169S,
173		"RealTek 8169S Single-chip Gigabit Ethernet" },
174	{ RT_VENDORID, RT_DEVICEID_8169, RL_HWREV_8169SB,
175		"RealTek 8169SB Single-chip Gigabit Ethernet" },
176	{ RT_VENDORID, RT_DEVICEID_8169, RL_HWREV_8110S,
177		"RealTek 8110S Single-chip Gigabit Ethernet" },
178	{ COREGA_VENDORID, COREGA_DEVICEID_CGLAPCIGT, RL_HWREV_8169S,
179		"Corega CG-LAPCIGT (RTL8169S) Gigabit Ethernet" },
180	{ 0, 0, 0, NULL }
181};
182
183static struct rl_hwrev re_hwrevs[] = {
184	{ RL_HWREV_8139, RL_8139,  "" },
185	{ RL_HWREV_8139A, RL_8139, "A" },
186	{ RL_HWREV_8139AG, RL_8139, "A-G" },
187	{ RL_HWREV_8139B, RL_8139, "B" },
188	{ RL_HWREV_8130, RL_8139, "8130" },
189	{ RL_HWREV_8139C, RL_8139, "C" },
190	{ RL_HWREV_8139D, RL_8139, "8139D/8100B/8100C" },
191	{ RL_HWREV_8139CPLUS, RL_8139CPLUS, "C+"},
192	{ RL_HWREV_8169, RL_8169, "8169"},
193	{ RL_HWREV_8169S, RL_8169, "8169S"},
194	{ RL_HWREV_8169SB, RL_8169, "8169SB"},
195	{ RL_HWREV_8110S, RL_8169, "8110S"},
196	{ RL_HWREV_8100, RL_8139, "8100"},
197	{ RL_HWREV_8101, RL_8139, "8101"},
198	{ 0, 0, NULL }
199};
200
201static int re_probe		(device_t);
202static int re_attach		(device_t);
203static int re_detach		(device_t);
204
205static int re_encap		(struct rl_softc *, struct mbuf **, int *);
206
207static void re_dma_map_addr	(void *, bus_dma_segment_t *, int, int);
208static void re_dma_map_desc	(void *, bus_dma_segment_t *, int,
209				    bus_size_t, int);
210static int re_allocmem		(device_t, struct rl_softc *);
211static int re_newbuf		(struct rl_softc *, int, struct mbuf *);
212static int re_rx_list_init	(struct rl_softc *);
213static int re_tx_list_init	(struct rl_softc *);
214#ifdef RE_FIXUP_RX
215static __inline void re_fixup_rx
216				(struct mbuf *);
217#endif
218static void re_rxeof		(struct rl_softc *);
219static void re_txeof		(struct rl_softc *);
220#ifdef DEVICE_POLLING
221static void re_poll		(struct ifnet *, enum poll_cmd, int);
222static void re_poll_locked	(struct ifnet *, enum poll_cmd, int);
223#endif
224static void re_intr		(void *);
225static void re_tick		(void *);
226static void re_start		(struct ifnet *);
227static void re_start_locked	(struct ifnet *);
228static int re_ioctl		(struct ifnet *, u_long, caddr_t);
229static void re_init		(void *);
230static void re_init_locked	(struct rl_softc *);
231static void re_stop		(struct rl_softc *);
232static void re_watchdog		(struct ifnet *);
233static int re_suspend		(device_t);
234static int re_resume		(device_t);
235static void re_shutdown		(device_t);
236static int re_ifmedia_upd	(struct ifnet *);
237static void re_ifmedia_sts	(struct ifnet *, struct ifmediareq *);
238
239static void re_eeprom_putbyte	(struct rl_softc *, int);
240static void re_eeprom_getword	(struct rl_softc *, int, u_int16_t *);
241static void re_read_eeprom	(struct rl_softc *, caddr_t, int, int, int);
242static int re_gmii_readreg	(device_t, int, int);
243static int re_gmii_writereg	(device_t, int, int, int);
244
245static int re_miibus_readreg	(device_t, int, int);
246static int re_miibus_writereg	(device_t, int, int, int);
247static void re_miibus_statchg	(device_t);
248
249static void re_setmulti		(struct rl_softc *);
250static void re_reset		(struct rl_softc *);
251
252static int re_diag		(struct rl_softc *);
253
254#ifdef RE_USEIOSPACE
255#define RL_RES			SYS_RES_IOPORT
256#define RL_RID			RL_PCI_LOIO
257#else
258#define RL_RES			SYS_RES_MEMORY
259#define RL_RID			RL_PCI_LOMEM
260#endif
261
262static device_method_t re_methods[] = {
263	/* Device interface */
264	DEVMETHOD(device_probe,		re_probe),
265	DEVMETHOD(device_attach,	re_attach),
266	DEVMETHOD(device_detach,	re_detach),
267	DEVMETHOD(device_suspend,	re_suspend),
268	DEVMETHOD(device_resume,	re_resume),
269	DEVMETHOD(device_shutdown,	re_shutdown),
270
271	/* bus interface */
272	DEVMETHOD(bus_print_child,	bus_generic_print_child),
273	DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
274
275	/* MII interface */
276	DEVMETHOD(miibus_readreg,	re_miibus_readreg),
277	DEVMETHOD(miibus_writereg,	re_miibus_writereg),
278	DEVMETHOD(miibus_statchg,	re_miibus_statchg),
279
280	{ 0, 0 }
281};
282
283static driver_t re_driver = {
284	"re",
285	re_methods,
286	sizeof(struct rl_softc)
287};
288
289static devclass_t re_devclass;
290
291DRIVER_MODULE(re, pci, re_driver, re_devclass, 0, 0);
292DRIVER_MODULE(re, cardbus, re_driver, re_devclass, 0, 0);
293DRIVER_MODULE(miibus, re, miibus_driver, miibus_devclass, 0, 0);
294
295#define EE_SET(x)					\
296	CSR_WRITE_1(sc, RL_EECMD,			\
297		CSR_READ_1(sc, RL_EECMD) | x)
298
299#define EE_CLR(x)					\
300	CSR_WRITE_1(sc, RL_EECMD,			\
301		CSR_READ_1(sc, RL_EECMD) & ~x)
302
303/*
304 * Send a read command and address to the EEPROM, check for ACK.
305 */
306static void
307re_eeprom_putbyte(sc, addr)
308	struct rl_softc		*sc;
309	int			addr;
310{
311	register int		d, i;
312
313	d = addr | sc->rl_eecmd_read;
314
315	/*
316	 * Feed in each bit and strobe the clock.
317	 */
318	for (i = 0x400; i; i >>= 1) {
319		if (d & i) {
320			EE_SET(RL_EE_DATAIN);
321		} else {
322			EE_CLR(RL_EE_DATAIN);
323		}
324		DELAY(100);
325		EE_SET(RL_EE_CLK);
326		DELAY(150);
327		EE_CLR(RL_EE_CLK);
328		DELAY(100);
329	}
330}
331
332/*
333 * Read a word of data stored in the EEPROM at address 'addr.'
334 */
335static void
336re_eeprom_getword(sc, addr, dest)
337	struct rl_softc		*sc;
338	int			addr;
339	u_int16_t		*dest;
340{
341	register int		i;
342	u_int16_t		word = 0;
343
344	/* Enter EEPROM access mode. */
345	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_PROGRAM|RL_EE_SEL);
346
347	/*
348	 * Send address of word we want to read.
349	 */
350	re_eeprom_putbyte(sc, addr);
351
352	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_PROGRAM|RL_EE_SEL);
353
354	/*
355	 * Start reading bits from EEPROM.
356	 */
357	for (i = 0x8000; i; i >>= 1) {
358		EE_SET(RL_EE_CLK);
359		DELAY(100);
360		if (CSR_READ_1(sc, RL_EECMD) & RL_EE_DATAOUT)
361			word |= i;
362		EE_CLR(RL_EE_CLK);
363		DELAY(100);
364	}
365
366	/* Turn off EEPROM access mode. */
367	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
368
369	*dest = word;
370}
371
372/*
373 * Read a sequence of words from the EEPROM.
374 */
375static void
376re_read_eeprom(sc, dest, off, cnt, swap)
377	struct rl_softc		*sc;
378	caddr_t			dest;
379	int			off;
380	int			cnt;
381	int			swap;
382{
383	int			i;
384	u_int16_t		word = 0, *ptr;
385
386	for (i = 0; i < cnt; i++) {
387		re_eeprom_getword(sc, off + i, &word);
388		ptr = (u_int16_t *)(dest + (i * 2));
389		if (swap)
390			*ptr = ntohs(word);
391		else
392			*ptr = word;
393	}
394}
395
396static int
397re_gmii_readreg(dev, phy, reg)
398	device_t		dev;
399	int			phy, reg;
400{
401	struct rl_softc		*sc;
402	u_int32_t		rval;
403	int			i;
404
405	if (phy != 1)
406		return (0);
407
408	sc = device_get_softc(dev);
409
410	/* Let the rgephy driver read the GMEDIASTAT register */
411
412	if (reg == RL_GMEDIASTAT) {
413		rval = CSR_READ_1(sc, RL_GMEDIASTAT);
414		return (rval);
415	}
416
417	CSR_WRITE_4(sc, RL_PHYAR, reg << 16);
418	DELAY(1000);
419
420	for (i = 0; i < RL_TIMEOUT; i++) {
421		rval = CSR_READ_4(sc, RL_PHYAR);
422		if (rval & RL_PHYAR_BUSY)
423			break;
424		DELAY(100);
425	}
426
427	if (i == RL_TIMEOUT) {
428		if_printf(sc->rl_ifp, "PHY read failed\n");
429		return (0);
430	}
431
432	return (rval & RL_PHYAR_PHYDATA);
433}
434
435static int
436re_gmii_writereg(dev, phy, reg, data)
437	device_t		dev;
438	int			phy, reg, data;
439{
440	struct rl_softc		*sc;
441	u_int32_t		rval;
442	int			i;
443
444	sc = device_get_softc(dev);
445
446	CSR_WRITE_4(sc, RL_PHYAR, (reg << 16) |
447	    (data & RL_PHYAR_PHYDATA) | RL_PHYAR_BUSY);
448	DELAY(1000);
449
450	for (i = 0; i < RL_TIMEOUT; i++) {
451		rval = CSR_READ_4(sc, RL_PHYAR);
452		if (!(rval & RL_PHYAR_BUSY))
453			break;
454		DELAY(100);
455	}
456
457	if (i == RL_TIMEOUT) {
458		if_printf(sc->rl_ifp, "PHY write failed\n");
459		return (0);
460	}
461
462	return (0);
463}
464
465static int
466re_miibus_readreg(dev, phy, reg)
467	device_t		dev;
468	int			phy, reg;
469{
470	struct rl_softc		*sc;
471	u_int16_t		rval = 0;
472	u_int16_t		re8139_reg = 0;
473
474	sc = device_get_softc(dev);
475
476	if (sc->rl_type == RL_8169) {
477		rval = re_gmii_readreg(dev, phy, reg);
478		return (rval);
479	}
480
481	/* Pretend the internal PHY is only at address 0 */
482	if (phy) {
483		return (0);
484	}
485	switch (reg) {
486	case MII_BMCR:
487		re8139_reg = RL_BMCR;
488		break;
489	case MII_BMSR:
490		re8139_reg = RL_BMSR;
491		break;
492	case MII_ANAR:
493		re8139_reg = RL_ANAR;
494		break;
495	case MII_ANER:
496		re8139_reg = RL_ANER;
497		break;
498	case MII_ANLPAR:
499		re8139_reg = RL_LPAR;
500		break;
501	case MII_PHYIDR1:
502	case MII_PHYIDR2:
503		return (0);
504	/*
505	 * Allow the rlphy driver to read the media status
506	 * register. If we have a link partner which does not
507	 * support NWAY, this is the register which will tell
508	 * us the results of parallel detection.
509	 */
510	case RL_MEDIASTAT:
511		rval = CSR_READ_1(sc, RL_MEDIASTAT);
512		return (rval);
513	default:
514		if_printf(sc->rl_ifp, "bad phy register\n");
515		return (0);
516	}
517	rval = CSR_READ_2(sc, re8139_reg);
518	return (rval);
519}
520
521static int
522re_miibus_writereg(dev, phy, reg, data)
523	device_t		dev;
524	int			phy, reg, data;
525{
526	struct rl_softc		*sc;
527	u_int16_t		re8139_reg = 0;
528	int			rval = 0;
529
530	sc = device_get_softc(dev);
531
532	if (sc->rl_type == RL_8169) {
533		rval = re_gmii_writereg(dev, phy, reg, data);
534		return (rval);
535	}
536
537	/* Pretend the internal PHY is only at address 0 */
538	if (phy)
539		return (0);
540
541	switch (reg) {
542	case MII_BMCR:
543		re8139_reg = RL_BMCR;
544		break;
545	case MII_BMSR:
546		re8139_reg = RL_BMSR;
547		break;
548	case MII_ANAR:
549		re8139_reg = RL_ANAR;
550		break;
551	case MII_ANER:
552		re8139_reg = RL_ANER;
553		break;
554	case MII_ANLPAR:
555		re8139_reg = RL_LPAR;
556		break;
557	case MII_PHYIDR1:
558	case MII_PHYIDR2:
559		return (0);
560		break;
561	default:
562		if_printf(sc->rl_ifp, "bad phy register\n");
563		return (0);
564	}
565	CSR_WRITE_2(sc, re8139_reg, data);
566	return (0);
567}
568
569static void
570re_miibus_statchg(dev)
571	device_t		dev;
572{
573
574}
575
576/*
577 * Program the 64-bit multicast hash filter.
578 */
579static void
580re_setmulti(sc)
581	struct rl_softc		*sc;
582{
583	struct ifnet		*ifp;
584	int			h = 0;
585	u_int32_t		hashes[2] = { 0, 0 };
586	struct ifmultiaddr	*ifma;
587	u_int32_t		rxfilt;
588	int			mcnt = 0;
589
590	RL_LOCK_ASSERT(sc);
591
592	ifp = sc->rl_ifp;
593
594	rxfilt = CSR_READ_4(sc, RL_RXCFG);
595
596	if (ifp->if_flags & IFF_ALLMULTI || ifp->if_flags & IFF_PROMISC) {
597		rxfilt |= RL_RXCFG_RX_MULTI;
598		CSR_WRITE_4(sc, RL_RXCFG, rxfilt);
599		CSR_WRITE_4(sc, RL_MAR0, 0xFFFFFFFF);
600		CSR_WRITE_4(sc, RL_MAR4, 0xFFFFFFFF);
601		return;
602	}
603
604	/* first, zot all the existing hash bits */
605	CSR_WRITE_4(sc, RL_MAR0, 0);
606	CSR_WRITE_4(sc, RL_MAR4, 0);
607
608	/* now program new ones */
609	IF_ADDR_LOCK(ifp);
610	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
611		if (ifma->ifma_addr->sa_family != AF_LINK)
612			continue;
613		h = ether_crc32_be(LLADDR((struct sockaddr_dl *)
614		    ifma->ifma_addr), ETHER_ADDR_LEN) >> 26;
615		if (h < 32)
616			hashes[0] |= (1 << h);
617		else
618			hashes[1] |= (1 << (h - 32));
619		mcnt++;
620	}
621	IF_ADDR_UNLOCK(ifp);
622
623	if (mcnt)
624		rxfilt |= RL_RXCFG_RX_MULTI;
625	else
626		rxfilt &= ~RL_RXCFG_RX_MULTI;
627
628	CSR_WRITE_4(sc, RL_RXCFG, rxfilt);
629	CSR_WRITE_4(sc, RL_MAR0, hashes[0]);
630	CSR_WRITE_4(sc, RL_MAR4, hashes[1]);
631}
632
633static void
634re_reset(sc)
635	struct rl_softc		*sc;
636{
637	register int		i;
638
639	RL_LOCK_ASSERT(sc);
640
641	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RESET);
642
643	for (i = 0; i < RL_TIMEOUT; i++) {
644		DELAY(10);
645		if (!(CSR_READ_1(sc, RL_COMMAND) & RL_CMD_RESET))
646			break;
647	}
648	if (i == RL_TIMEOUT)
649		if_printf(sc->rl_ifp, "reset never completed!\n");
650
651	CSR_WRITE_1(sc, 0x82, 1);
652}
653
654/*
655 * The following routine is designed to test for a defect on some
656 * 32-bit 8169 cards. Some of these NICs have the REQ64# and ACK64#
657 * lines connected to the bus, however for a 32-bit only card, they
658 * should be pulled high. The result of this defect is that the
659 * NIC will not work right if you plug it into a 64-bit slot: DMA
660 * operations will be done with 64-bit transfers, which will fail
661 * because the 64-bit data lines aren't connected.
662 *
663 * There's no way to work around this (short of talking a soldering
664 * iron to the board), however we can detect it. The method we use
665 * here is to put the NIC into digital loopback mode, set the receiver
666 * to promiscuous mode, and then try to send a frame. We then compare
667 * the frame data we sent to what was received. If the data matches,
668 * then the NIC is working correctly, otherwise we know the user has
669 * a defective NIC which has been mistakenly plugged into a 64-bit PCI
670 * slot. In the latter case, there's no way the NIC can work correctly,
671 * so we print out a message on the console and abort the device attach.
672 */
673
674static int
675re_diag(sc)
676	struct rl_softc		*sc;
677{
678	struct ifnet		*ifp = sc->rl_ifp;
679	struct mbuf		*m0;
680	struct ether_header	*eh;
681	struct rl_desc		*cur_rx;
682	u_int16_t		status;
683	u_int32_t		rxstat;
684	int			total_len, i, error = 0;
685	u_int8_t		dst[] = { 0x00, 'h', 'e', 'l', 'l', 'o' };
686	u_int8_t		src[] = { 0x00, 'w', 'o', 'r', 'l', 'd' };
687
688	/* Allocate a single mbuf */
689	MGETHDR(m0, M_DONTWAIT, MT_DATA);
690	if (m0 == NULL)
691		return (ENOBUFS);
692
693	RL_LOCK(sc);
694
695	/*
696	 * Initialize the NIC in test mode. This sets the chip up
697	 * so that it can send and receive frames, but performs the
698	 * following special functions:
699	 * - Puts receiver in promiscuous mode
700	 * - Enables digital loopback mode
701	 * - Leaves interrupts turned off
702	 */
703
704	ifp->if_flags |= IFF_PROMISC;
705	sc->rl_testmode = 1;
706	re_init_locked(sc);
707	re_stop(sc);
708	DELAY(100000);
709	re_init_locked(sc);
710
711	/* Put some data in the mbuf */
712
713	eh = mtod(m0, struct ether_header *);
714	bcopy ((char *)&dst, eh->ether_dhost, ETHER_ADDR_LEN);
715	bcopy ((char *)&src, eh->ether_shost, ETHER_ADDR_LEN);
716	eh->ether_type = htons(ETHERTYPE_IP);
717	m0->m_pkthdr.len = m0->m_len = ETHER_MIN_LEN - ETHER_CRC_LEN;
718
719	/*
720	 * Queue the packet, start transmission.
721	 * Note: IF_HANDOFF() ultimately calls re_start() for us.
722	 */
723
724	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
725	RL_UNLOCK(sc);
726	/* XXX: re_diag must not be called when in ALTQ mode */
727	IF_HANDOFF(&ifp->if_snd, m0, ifp);
728	RL_LOCK(sc);
729	m0 = NULL;
730
731	/* Wait for it to propagate through the chip */
732
733	DELAY(100000);
734	for (i = 0; i < RL_TIMEOUT; i++) {
735		status = CSR_READ_2(sc, RL_ISR);
736		if ((status & (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK)) ==
737		    (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK))
738			break;
739		DELAY(10);
740	}
741
742	if (i == RL_TIMEOUT) {
743		if_printf(ifp, "diagnostic failed, failed to receive packet "
744		    "in loopback mode\n");
745		error = EIO;
746		goto done;
747	}
748
749	/*
750	 * The packet should have been dumped into the first
751	 * entry in the RX DMA ring. Grab it from there.
752	 */
753
754	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
755	    sc->rl_ldata.rl_rx_list_map,
756	    BUS_DMASYNC_POSTREAD);
757	bus_dmamap_sync(sc->rl_ldata.rl_mtag,
758	    sc->rl_ldata.rl_rx_dmamap[0],
759	    BUS_DMASYNC_POSTWRITE);
760	bus_dmamap_unload(sc->rl_ldata.rl_mtag,
761	    sc->rl_ldata.rl_rx_dmamap[0]);
762
763	m0 = sc->rl_ldata.rl_rx_mbuf[0];
764	sc->rl_ldata.rl_rx_mbuf[0] = NULL;
765	eh = mtod(m0, struct ether_header *);
766
767	cur_rx = &sc->rl_ldata.rl_rx_list[0];
768	total_len = RL_RXBYTES(cur_rx);
769	rxstat = le32toh(cur_rx->rl_cmdstat);
770
771	if (total_len != ETHER_MIN_LEN) {
772		if_printf(ifp, "diagnostic failed, received short packet\n");
773		error = EIO;
774		goto done;
775	}
776
777	/* Test that the received packet data matches what we sent. */
778
779	if (bcmp((char *)&eh->ether_dhost, (char *)&dst, ETHER_ADDR_LEN) ||
780	    bcmp((char *)&eh->ether_shost, (char *)&src, ETHER_ADDR_LEN) ||
781	    ntohs(eh->ether_type) != ETHERTYPE_IP) {
782		if_printf(ifp, "WARNING, DMA FAILURE!\n");
783		if_printf(ifp, "expected TX data: %6D/%6D/0x%x\n",
784		    dst, ":", src, ":", ETHERTYPE_IP);
785		if_printf(ifp, "received RX data: %6D/%6D/0x%x\n",
786		    eh->ether_dhost, ":",  eh->ether_shost, ":",
787		    ntohs(eh->ether_type));
788		if_printf(ifp, "You may have a defective 32-bit NIC plugged "
789		    "into a 64-bit PCI slot.\n");
790		if_printf(ifp, "Please re-install the NIC in a 32-bit slot "
791		    "for proper operation.\n");
792		if_printf(ifp, "Read the re(4) man page for more details.\n");
793		error = EIO;
794	}
795
796done:
797	/* Turn interface off, release resources */
798
799	sc->rl_testmode = 0;
800	ifp->if_flags &= ~IFF_PROMISC;
801	re_stop(sc);
802	if (m0 != NULL)
803		m_freem(m0);
804
805	RL_UNLOCK(sc);
806
807	return (error);
808}
809
810/*
811 * Probe for a RealTek 8139C+/8169/8110 chip. Check the PCI vendor and device
812 * IDs against our list and return a device name if we find a match.
813 */
814static int
815re_probe(dev)
816	device_t		dev;
817{
818	struct rl_type		*t;
819	struct rl_softc		*sc;
820	int			rid;
821	u_int32_t		hwrev;
822
823	t = re_devs;
824	sc = device_get_softc(dev);
825
826	while (t->rl_name != NULL) {
827		if ((pci_get_vendor(dev) == t->rl_vid) &&
828		    (pci_get_device(dev) == t->rl_did)) {
829
830			/*
831			 * Temporarily map the I/O space
832			 * so we can read the chip ID register.
833			 */
834			rid = RL_RID;
835			sc->rl_res = bus_alloc_resource_any(dev, RL_RES, &rid,
836			    RF_ACTIVE);
837			if (sc->rl_res == NULL) {
838				device_printf(dev,
839				    "couldn't map ports/memory\n");
840				return (ENXIO);
841			}
842			sc->rl_btag = rman_get_bustag(sc->rl_res);
843			sc->rl_bhandle = rman_get_bushandle(sc->rl_res);
844			hwrev = CSR_READ_4(sc, RL_TXCFG) & RL_TXCFG_HWREV;
845			bus_release_resource(dev, RL_RES,
846			    RL_RID, sc->rl_res);
847			if (t->rl_basetype == hwrev) {
848				device_set_desc(dev, t->rl_name);
849				return (BUS_PROBE_DEFAULT);
850			}
851		}
852		t++;
853	}
854
855	return (ENXIO);
856}
857
858/*
859 * This routine takes the segment list provided as the result of
860 * a bus_dma_map_load() operation and assigns the addresses/lengths
861 * to RealTek DMA descriptors. This can be called either by the RX
862 * code or the TX code. In the RX case, we'll probably wind up mapping
863 * at most one segment. For the TX case, there could be any number of
864 * segments since TX packets may span multiple mbufs. In either case,
865 * if the number of segments is larger than the rl_maxsegs limit
866 * specified by the caller, we abort the mapping operation. Sadly,
867 * whoever designed the buffer mapping API did not provide a way to
868 * return an error from here, so we have to fake it a bit.
869 */
870
871static void
872re_dma_map_desc(arg, segs, nseg, mapsize, error)
873	void			*arg;
874	bus_dma_segment_t	*segs;
875	int			nseg;
876	bus_size_t		mapsize;
877	int			error;
878{
879	struct rl_dmaload_arg	*ctx;
880	struct rl_desc		*d = NULL;
881	int			i = 0, idx;
882
883	if (error)
884		return;
885
886	ctx = arg;
887
888	/* Signal error to caller if there's too many segments */
889	if (nseg > ctx->rl_maxsegs) {
890		ctx->rl_maxsegs = 0;
891		return;
892	}
893
894	/*
895	 * Map the segment array into descriptors. Note that we set the
896	 * start-of-frame and end-of-frame markers for either TX or RX, but
897	 * they really only have meaning in the TX case. (In the RX case,
898	 * it's the chip that tells us where packets begin and end.)
899	 * We also keep track of the end of the ring and set the
900	 * end-of-ring bits as needed, and we set the ownership bits
901	 * in all except the very first descriptor. (The caller will
902	 * set this descriptor later when it start transmission or
903	 * reception.)
904	 */
905	idx = ctx->rl_idx;
906	for (;;) {
907		u_int32_t		cmdstat;
908		d = &ctx->rl_ring[idx];
909		if (le32toh(d->rl_cmdstat) & RL_RDESC_STAT_OWN) {
910			ctx->rl_maxsegs = 0;
911			return;
912		}
913		cmdstat = segs[i].ds_len;
914		d->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[i].ds_addr));
915		d->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[i].ds_addr));
916		if (i == 0)
917			cmdstat |= RL_TDESC_CMD_SOF;
918		else
919			cmdstat |= RL_TDESC_CMD_OWN;
920		if (idx == (RL_RX_DESC_CNT - 1))
921			cmdstat |= RL_TDESC_CMD_EOR;
922		d->rl_cmdstat = htole32(cmdstat | ctx->rl_flags);
923		i++;
924		if (i == nseg)
925			break;
926		RL_DESC_INC(idx);
927	}
928
929	d->rl_cmdstat |= htole32(RL_TDESC_CMD_EOF);
930	ctx->rl_maxsegs = nseg;
931	ctx->rl_idx = idx;
932}
933
934/*
935 * Map a single buffer address.
936 */
937
938static void
939re_dma_map_addr(arg, segs, nseg, error)
940	void			*arg;
941	bus_dma_segment_t	*segs;
942	int			nseg;
943	int			error;
944{
945	bus_addr_t		*addr;
946
947	if (error)
948		return;
949
950	KASSERT(nseg == 1, ("too many DMA segments, %d should be 1", nseg));
951	addr = arg;
952	*addr = segs->ds_addr;
953}
954
955static int
956re_allocmem(dev, sc)
957	device_t		dev;
958	struct rl_softc		*sc;
959{
960	int			error;
961	int			nseg;
962	int			i;
963
964	/*
965	 * Allocate map for RX mbufs.
966	 */
967	nseg = 32;
968	error = bus_dma_tag_create(sc->rl_parent_tag, ETHER_ALIGN, 0,
969	    BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
970	    NULL, MCLBYTES * nseg, nseg, MCLBYTES, BUS_DMA_ALLOCNOW,
971	    NULL, NULL, &sc->rl_ldata.rl_mtag);
972	if (error) {
973		device_printf(dev, "could not allocate dma tag\n");
974		return (ENOMEM);
975	}
976
977	/*
978	 * Allocate map for TX descriptor list.
979	 */
980	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
981	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
982	    NULL, RL_TX_LIST_SZ, 1, RL_TX_LIST_SZ, BUS_DMA_ALLOCNOW,
983	    NULL, NULL, &sc->rl_ldata.rl_tx_list_tag);
984	if (error) {
985		device_printf(dev, "could not allocate dma tag\n");
986		return (ENOMEM);
987	}
988
989	/* Allocate DMA'able memory for the TX ring */
990
991	error = bus_dmamem_alloc(sc->rl_ldata.rl_tx_list_tag,
992	    (void **)&sc->rl_ldata.rl_tx_list, BUS_DMA_NOWAIT | BUS_DMA_ZERO,
993	    &sc->rl_ldata.rl_tx_list_map);
994	if (error)
995		return (ENOMEM);
996
997	/* Load the map for the TX ring. */
998
999	error = bus_dmamap_load(sc->rl_ldata.rl_tx_list_tag,
1000	     sc->rl_ldata.rl_tx_list_map, sc->rl_ldata.rl_tx_list,
1001	     RL_TX_LIST_SZ, re_dma_map_addr,
1002	     &sc->rl_ldata.rl_tx_list_addr, BUS_DMA_NOWAIT);
1003
1004	/* Create DMA maps for TX buffers */
1005
1006	for (i = 0; i < RL_TX_DESC_CNT; i++) {
1007		error = bus_dmamap_create(sc->rl_ldata.rl_mtag, 0,
1008			    &sc->rl_ldata.rl_tx_dmamap[i]);
1009		if (error) {
1010			device_printf(dev, "can't create DMA map for TX\n");
1011			return (ENOMEM);
1012		}
1013	}
1014
1015	/*
1016	 * Allocate map for RX descriptor list.
1017	 */
1018	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
1019	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1020	    NULL, RL_RX_LIST_SZ, 1, RL_RX_LIST_SZ, BUS_DMA_ALLOCNOW,
1021	    NULL, NULL, &sc->rl_ldata.rl_rx_list_tag);
1022	if (error) {
1023		device_printf(dev, "could not allocate dma tag\n");
1024		return (ENOMEM);
1025	}
1026
1027	/* Allocate DMA'able memory for the RX ring */
1028
1029	error = bus_dmamem_alloc(sc->rl_ldata.rl_rx_list_tag,
1030	    (void **)&sc->rl_ldata.rl_rx_list, BUS_DMA_NOWAIT | BUS_DMA_ZERO,
1031	    &sc->rl_ldata.rl_rx_list_map);
1032	if (error)
1033		return (ENOMEM);
1034
1035	/* Load the map for the RX ring. */
1036
1037	error = bus_dmamap_load(sc->rl_ldata.rl_rx_list_tag,
1038	     sc->rl_ldata.rl_rx_list_map, sc->rl_ldata.rl_rx_list,
1039	     RL_RX_LIST_SZ, re_dma_map_addr,
1040	     &sc->rl_ldata.rl_rx_list_addr, BUS_DMA_NOWAIT);
1041
1042	/* Create DMA maps for RX buffers */
1043
1044	for (i = 0; i < RL_RX_DESC_CNT; i++) {
1045		error = bus_dmamap_create(sc->rl_ldata.rl_mtag, 0,
1046			    &sc->rl_ldata.rl_rx_dmamap[i]);
1047		if (error) {
1048			device_printf(dev, "can't create DMA map for RX\n");
1049			return (ENOMEM);
1050		}
1051	}
1052
1053	return (0);
1054}
1055
1056/*
1057 * Attach the interface. Allocate softc structures, do ifmedia
1058 * setup and ethernet/BPF attach.
1059 */
1060static int
1061re_attach(dev)
1062	device_t		dev;
1063{
1064	u_char			eaddr[ETHER_ADDR_LEN];
1065	u_int16_t		as[3];
1066	struct rl_softc		*sc;
1067	struct ifnet		*ifp;
1068	struct rl_hwrev		*hw_rev;
1069	int			hwrev;
1070	u_int16_t		re_did = 0;
1071	int			error = 0, rid, i;
1072
1073	sc = device_get_softc(dev);
1074
1075	mtx_init(&sc->rl_mtx, device_get_nameunit(dev), MTX_NETWORK_LOCK,
1076	    MTX_DEF);
1077	callout_init_mtx(&sc->rl_stat_callout, &sc->rl_mtx, 0);
1078
1079	/*
1080	 * Map control/status registers.
1081	 */
1082	pci_enable_busmaster(dev);
1083
1084	rid = RL_RID;
1085	sc->rl_res = bus_alloc_resource_any(dev, RL_RES, &rid,
1086	    RF_ACTIVE);
1087
1088	if (sc->rl_res == NULL) {
1089		device_printf(dev, "couldn't map ports/memory\n");
1090		error = ENXIO;
1091		goto fail;
1092	}
1093
1094	sc->rl_btag = rman_get_bustag(sc->rl_res);
1095	sc->rl_bhandle = rman_get_bushandle(sc->rl_res);
1096
1097	/* Allocate interrupt */
1098	rid = 0;
1099	sc->rl_irq = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
1100	    RF_SHAREABLE | RF_ACTIVE);
1101
1102	if (sc->rl_irq == NULL) {
1103		device_printf(dev, "couldn't map interrupt\n");
1104		error = ENXIO;
1105		goto fail;
1106	}
1107
1108	/* Reset the adapter. */
1109	RL_LOCK(sc);
1110	re_reset(sc);
1111	RL_UNLOCK(sc);
1112
1113	hw_rev = re_hwrevs;
1114	hwrev = CSR_READ_4(sc, RL_TXCFG) & RL_TXCFG_HWREV;
1115	while (hw_rev->rl_desc != NULL) {
1116		if (hw_rev->rl_rev == hwrev) {
1117			sc->rl_type = hw_rev->rl_type;
1118			break;
1119		}
1120		hw_rev++;
1121	}
1122
1123	if (sc->rl_type == RL_8169) {
1124
1125		/* Set RX length mask */
1126
1127		sc->rl_rxlenmask = RL_RDESC_STAT_GFRAGLEN;
1128
1129		/* Force station address autoload from the EEPROM */
1130
1131		CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_AUTOLOAD);
1132		for (i = 0; i < RL_TIMEOUT; i++) {
1133			if (!(CSR_READ_1(sc, RL_EECMD) & RL_EEMODE_AUTOLOAD))
1134				break;
1135			DELAY(100);
1136		}
1137		if (i == RL_TIMEOUT)
1138			device_printf(dev, "eeprom autoload timed out\n");
1139
1140			for (i = 0; i < ETHER_ADDR_LEN; i++)
1141				eaddr[i] = CSR_READ_1(sc, RL_IDR0 + i);
1142	} else {
1143
1144		/* Set RX length mask */
1145
1146		sc->rl_rxlenmask = RL_RDESC_STAT_FRAGLEN;
1147
1148		sc->rl_eecmd_read = RL_EECMD_READ_6BIT;
1149		re_read_eeprom(sc, (caddr_t)&re_did, 0, 1, 0);
1150		if (re_did != 0x8129)
1151			sc->rl_eecmd_read = RL_EECMD_READ_8BIT;
1152
1153		/*
1154		 * Get station address from the EEPROM.
1155		 */
1156		re_read_eeprom(sc, (caddr_t)as, RL_EE_EADDR, 3, 0);
1157		for (i = 0; i < 3; i++) {
1158			eaddr[(i * 2) + 0] = as[i] & 0xff;
1159			eaddr[(i * 2) + 1] = as[i] >> 8;
1160		}
1161	}
1162
1163	/*
1164	 * Allocate the parent bus DMA tag appropriate for PCI.
1165	 */
1166#define RL_NSEG_NEW 32
1167	error = bus_dma_tag_create(NULL,	/* parent */
1168			1, 0,			/* alignment, boundary */
1169			BUS_SPACE_MAXADDR_32BIT,/* lowaddr */
1170			BUS_SPACE_MAXADDR,	/* highaddr */
1171			NULL, NULL,		/* filter, filterarg */
1172			MAXBSIZE, RL_NSEG_NEW,	/* maxsize, nsegments */
1173			BUS_SPACE_MAXSIZE_32BIT,/* maxsegsize */
1174			BUS_DMA_ALLOCNOW,	/* flags */
1175			NULL, NULL,		/* lockfunc, lockarg */
1176			&sc->rl_parent_tag);
1177	if (error)
1178		goto fail;
1179
1180	error = re_allocmem(dev, sc);
1181
1182	if (error)
1183		goto fail;
1184
1185	ifp = sc->rl_ifp = if_alloc(IFT_ETHER);
1186	if (ifp == NULL) {
1187		device_printf(dev, "can not if_alloc()\n");
1188		error = ENOSPC;
1189		goto fail;
1190	}
1191
1192	/* Do MII setup */
1193	if (mii_phy_probe(dev, &sc->rl_miibus,
1194	    re_ifmedia_upd, re_ifmedia_sts)) {
1195		device_printf(dev, "MII without any phy!\n");
1196		error = ENXIO;
1197		goto fail;
1198	}
1199
1200	ifp->if_softc = sc;
1201	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1202	ifp->if_mtu = ETHERMTU;
1203	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1204	ifp->if_ioctl = re_ioctl;
1205	ifp->if_capabilities = IFCAP_VLAN_MTU;
1206	ifp->if_start = re_start;
1207	ifp->if_hwassist = /*RE_CSUM_FEATURES*/0;
1208	ifp->if_capabilities |= IFCAP_HWCSUM|IFCAP_VLAN_HWTAGGING;
1209	ifp->if_capenable = ifp->if_capabilities & ~IFCAP_HWCSUM;
1210#ifdef DEVICE_POLLING
1211	ifp->if_capabilities |= IFCAP_POLLING;
1212#endif
1213	ifp->if_watchdog = re_watchdog;
1214	ifp->if_init = re_init;
1215	if (sc->rl_type == RL_8169)
1216		ifp->if_baudrate = 1000000000;
1217	else
1218		ifp->if_baudrate = 100000000;
1219	IFQ_SET_MAXLEN(&ifp->if_snd,  RL_IFQ_MAXLEN);
1220	ifp->if_snd.ifq_drv_maxlen = RL_IFQ_MAXLEN;
1221	IFQ_SET_READY(&ifp->if_snd);
1222
1223	/*
1224	 * Call MI attach routine.
1225	 */
1226	ether_ifattach(ifp, eaddr);
1227
1228	/* Perform hardware diagnostic. */
1229	error = re_diag(sc);
1230
1231	if (error) {
1232		device_printf(dev, "attach aborted due to hardware diag failure\n");
1233		ether_ifdetach(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		device_printf(dev, "couldn't set up irq\n");
1242		ether_ifdetach(ifp);
1243	}
1244
1245fail:
1246	if (error)
1247		re_detach(dev);
1248
1249	return (error);
1250}
1251
1252/*
1253 * Shutdown hardware and free up resources. This can be called any
1254 * time after the mutex has been initialized. It is called in both
1255 * the error case in attach and the normal detach case so it needs
1256 * to be careful about only freeing resources that have actually been
1257 * allocated.
1258 */
1259static int
1260re_detach(dev)
1261	device_t		dev;
1262{
1263	struct rl_softc		*sc;
1264	struct ifnet		*ifp;
1265	int			i;
1266
1267	sc = device_get_softc(dev);
1268	ifp = sc->rl_ifp;
1269	KASSERT(mtx_initialized(&sc->rl_mtx), ("re mutex not initialized"));
1270
1271#ifdef DEVICE_POLLING
1272	if (ifp->if_capenable & IFCAP_POLLING)
1273		ether_poll_deregister(ifp);
1274#endif
1275
1276	/* These should only be active if attach succeeded */
1277	if (device_is_attached(dev)) {
1278		RL_LOCK(sc);
1279#if 0
1280		sc->suspended = 1;
1281#endif
1282		re_stop(sc);
1283		RL_UNLOCK(sc);
1284		callout_drain(&sc->rl_stat_callout);
1285		/*
1286		 * Force off the IFF_UP flag here, in case someone
1287		 * still had a BPF descriptor attached to this
1288		 * interface. If they do, ether_ifdetach() will cause
1289		 * the BPF code to try and clear the promisc mode
1290		 * flag, which will bubble down to re_ioctl(),
1291		 * which will try to call re_init() again. This will
1292		 * turn the NIC back on and restart the MII ticker,
1293		 * which will panic the system when the kernel tries
1294		 * to invoke the re_tick() function that isn't there
1295		 * anymore.
1296		 */
1297		ifp->if_flags &= ~IFF_UP;
1298		ether_ifdetach(ifp);
1299	}
1300	if (sc->rl_miibus)
1301		device_delete_child(dev, sc->rl_miibus);
1302	bus_generic_detach(dev);
1303
1304	/*
1305	 * The rest is resource deallocation, so we should already be
1306	 * stopped here.
1307	 */
1308
1309	if (sc->rl_intrhand)
1310		bus_teardown_intr(dev, sc->rl_irq, sc->rl_intrhand);
1311	if (ifp != NULL)
1312		if_free(ifp);
1313	if (sc->rl_irq)
1314		bus_release_resource(dev, SYS_RES_IRQ, 0, sc->rl_irq);
1315	if (sc->rl_res)
1316		bus_release_resource(dev, RL_RES, RL_RID, sc->rl_res);
1317
1318
1319	/* Unload and free the RX DMA ring memory and map */
1320
1321	if (sc->rl_ldata.rl_rx_list_tag) {
1322		bus_dmamap_unload(sc->rl_ldata.rl_rx_list_tag,
1323		    sc->rl_ldata.rl_rx_list_map);
1324		bus_dmamem_free(sc->rl_ldata.rl_rx_list_tag,
1325		    sc->rl_ldata.rl_rx_list,
1326		    sc->rl_ldata.rl_rx_list_map);
1327		bus_dma_tag_destroy(sc->rl_ldata.rl_rx_list_tag);
1328	}
1329
1330	/* Unload and free the TX DMA ring memory and map */
1331
1332	if (sc->rl_ldata.rl_tx_list_tag) {
1333		bus_dmamap_unload(sc->rl_ldata.rl_tx_list_tag,
1334		    sc->rl_ldata.rl_tx_list_map);
1335		bus_dmamem_free(sc->rl_ldata.rl_tx_list_tag,
1336		    sc->rl_ldata.rl_tx_list,
1337		    sc->rl_ldata.rl_tx_list_map);
1338		bus_dma_tag_destroy(sc->rl_ldata.rl_tx_list_tag);
1339	}
1340
1341	/* Destroy all the RX and TX buffer maps */
1342
1343	if (sc->rl_ldata.rl_mtag) {
1344		for (i = 0; i < RL_TX_DESC_CNT; i++)
1345			bus_dmamap_destroy(sc->rl_ldata.rl_mtag,
1346			    sc->rl_ldata.rl_tx_dmamap[i]);
1347		for (i = 0; i < RL_RX_DESC_CNT; i++)
1348			bus_dmamap_destroy(sc->rl_ldata.rl_mtag,
1349			    sc->rl_ldata.rl_rx_dmamap[i]);
1350		bus_dma_tag_destroy(sc->rl_ldata.rl_mtag);
1351	}
1352
1353	/* Unload and free the stats buffer and map */
1354
1355	if (sc->rl_ldata.rl_stag) {
1356		bus_dmamap_unload(sc->rl_ldata.rl_stag,
1357		    sc->rl_ldata.rl_rx_list_map);
1358		bus_dmamem_free(sc->rl_ldata.rl_stag,
1359		    sc->rl_ldata.rl_stats,
1360		    sc->rl_ldata.rl_smap);
1361		bus_dma_tag_destroy(sc->rl_ldata.rl_stag);
1362	}
1363
1364	if (sc->rl_parent_tag)
1365		bus_dma_tag_destroy(sc->rl_parent_tag);
1366
1367	mtx_destroy(&sc->rl_mtx);
1368
1369	return (0);
1370}
1371
1372static int
1373re_newbuf(sc, idx, m)
1374	struct rl_softc		*sc;
1375	int			idx;
1376	struct mbuf		*m;
1377{
1378	struct rl_dmaload_arg	arg;
1379	struct mbuf		*n = NULL;
1380	int			error;
1381
1382	if (m == NULL) {
1383		n = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
1384		if (n == NULL)
1385			return (ENOBUFS);
1386		m = n;
1387	} else
1388		m->m_data = m->m_ext.ext_buf;
1389
1390	m->m_len = m->m_pkthdr.len = MCLBYTES;
1391#ifdef RE_FIXUP_RX
1392	/*
1393	 * This is part of an evil trick to deal with non-x86 platforms.
1394	 * The RealTek chip requires RX buffers to be aligned on 64-bit
1395	 * boundaries, but that will hose non-x86 machines. To get around
1396	 * this, we leave some empty space at the start of each buffer
1397	 * and for non-x86 hosts, we copy the buffer back six bytes
1398	 * to achieve word alignment. This is slightly more efficient
1399	 * than allocating a new buffer, copying the contents, and
1400	 * discarding the old buffer.
1401	 */
1402	m_adj(m, RE_ETHER_ALIGN);
1403#endif
1404	arg.sc = sc;
1405	arg.rl_idx = idx;
1406	arg.rl_maxsegs = 1;
1407	arg.rl_flags = 0;
1408	arg.rl_ring = sc->rl_ldata.rl_rx_list;
1409
1410	error = bus_dmamap_load_mbuf(sc->rl_ldata.rl_mtag,
1411	    sc->rl_ldata.rl_rx_dmamap[idx], m, re_dma_map_desc,
1412	    &arg, BUS_DMA_NOWAIT);
1413	if (error || arg.rl_maxsegs != 1) {
1414		if (n != NULL)
1415			m_freem(n);
1416		return (ENOMEM);
1417	}
1418
1419	sc->rl_ldata.rl_rx_list[idx].rl_cmdstat |= htole32(RL_RDESC_CMD_OWN);
1420	sc->rl_ldata.rl_rx_mbuf[idx] = m;
1421
1422	bus_dmamap_sync(sc->rl_ldata.rl_mtag,
1423	    sc->rl_ldata.rl_rx_dmamap[idx],
1424	    BUS_DMASYNC_PREREAD);
1425
1426	return (0);
1427}
1428
1429#ifdef RE_FIXUP_RX
1430static __inline void
1431re_fixup_rx(m)
1432	struct mbuf		*m;
1433{
1434	int                     i;
1435	uint16_t                *src, *dst;
1436
1437	src = mtod(m, uint16_t *);
1438	dst = src - (RE_ETHER_ALIGN - ETHER_ALIGN) / sizeof *src;
1439
1440	for (i = 0; i < (m->m_len / sizeof(uint16_t) + 1); i++)
1441		*dst++ = *src++;
1442
1443	m->m_data -= RE_ETHER_ALIGN - ETHER_ALIGN;
1444
1445	return;
1446}
1447#endif
1448
1449static int
1450re_tx_list_init(sc)
1451	struct rl_softc		*sc;
1452{
1453
1454	RL_LOCK_ASSERT(sc);
1455
1456	bzero ((char *)sc->rl_ldata.rl_tx_list, RL_TX_LIST_SZ);
1457	bzero ((char *)&sc->rl_ldata.rl_tx_mbuf,
1458	    (RL_TX_DESC_CNT * sizeof(struct mbuf *)));
1459
1460	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
1461	    sc->rl_ldata.rl_tx_list_map, BUS_DMASYNC_PREWRITE);
1462	sc->rl_ldata.rl_tx_prodidx = 0;
1463	sc->rl_ldata.rl_tx_considx = 0;
1464	sc->rl_ldata.rl_tx_free = RL_TX_DESC_CNT;
1465
1466	return (0);
1467}
1468
1469static int
1470re_rx_list_init(sc)
1471	struct rl_softc		*sc;
1472{
1473	int			i;
1474
1475	bzero ((char *)sc->rl_ldata.rl_rx_list, RL_RX_LIST_SZ);
1476	bzero ((char *)&sc->rl_ldata.rl_rx_mbuf,
1477	    (RL_RX_DESC_CNT * sizeof(struct mbuf *)));
1478
1479	for (i = 0; i < RL_RX_DESC_CNT; i++) {
1480		if (re_newbuf(sc, i, NULL) == ENOBUFS)
1481			return (ENOBUFS);
1482	}
1483
1484	/* Flush the RX descriptors */
1485
1486	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1487	    sc->rl_ldata.rl_rx_list_map,
1488	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
1489
1490	sc->rl_ldata.rl_rx_prodidx = 0;
1491	sc->rl_head = sc->rl_tail = NULL;
1492
1493	return (0);
1494}
1495
1496/*
1497 * RX handler for C+ and 8169. For the gigE chips, we support
1498 * the reception of jumbo frames that have been fragmented
1499 * across multiple 2K mbuf cluster buffers.
1500 */
1501static void
1502re_rxeof(sc)
1503	struct rl_softc		*sc;
1504{
1505	struct mbuf		*m;
1506	struct ifnet		*ifp;
1507	int			i, total_len;
1508	struct rl_desc		*cur_rx;
1509	u_int32_t		rxstat, rxvlan;
1510
1511	RL_LOCK_ASSERT(sc);
1512
1513	ifp = sc->rl_ifp;
1514	i = sc->rl_ldata.rl_rx_prodidx;
1515
1516	/* Invalidate the descriptor memory */
1517
1518	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1519	    sc->rl_ldata.rl_rx_list_map,
1520	    BUS_DMASYNC_POSTREAD);
1521
1522	while (!RL_OWN(&sc->rl_ldata.rl_rx_list[i])) {
1523		cur_rx = &sc->rl_ldata.rl_rx_list[i];
1524		m = sc->rl_ldata.rl_rx_mbuf[i];
1525		total_len = RL_RXBYTES(cur_rx);
1526		rxstat = le32toh(cur_rx->rl_cmdstat);
1527		rxvlan = le32toh(cur_rx->rl_vlanctl);
1528
1529		/* Invalidate the RX mbuf and unload its map */
1530
1531		bus_dmamap_sync(sc->rl_ldata.rl_mtag,
1532		    sc->rl_ldata.rl_rx_dmamap[i],
1533		    BUS_DMASYNC_POSTWRITE);
1534		bus_dmamap_unload(sc->rl_ldata.rl_mtag,
1535		    sc->rl_ldata.rl_rx_dmamap[i]);
1536
1537		if (!(rxstat & RL_RDESC_STAT_EOF)) {
1538			m->m_len = RE_RX_DESC_BUFLEN;
1539			if (sc->rl_head == NULL)
1540				sc->rl_head = sc->rl_tail = m;
1541			else {
1542				m->m_flags &= ~M_PKTHDR;
1543				sc->rl_tail->m_next = m;
1544				sc->rl_tail = m;
1545			}
1546			re_newbuf(sc, i, NULL);
1547			RL_DESC_INC(i);
1548			continue;
1549		}
1550
1551		/*
1552		 * NOTE: for the 8139C+, the frame length field
1553		 * is always 12 bits in size, but for the gigE chips,
1554		 * it is 13 bits (since the max RX frame length is 16K).
1555		 * Unfortunately, all 32 bits in the status word
1556		 * were already used, so to make room for the extra
1557		 * length bit, RealTek took out the 'frame alignment
1558		 * error' bit and shifted the other status bits
1559		 * over one slot. The OWN, EOR, FS and LS bits are
1560		 * still in the same places. We have already extracted
1561		 * the frame length and checked the OWN bit, so rather
1562		 * than using an alternate bit mapping, we shift the
1563		 * status bits one space to the right so we can evaluate
1564		 * them using the 8169 status as though it was in the
1565		 * same format as that of the 8139C+.
1566		 */
1567		if (sc->rl_type == RL_8169)
1568			rxstat >>= 1;
1569
1570		/*
1571		 * if total_len > 2^13-1, both _RXERRSUM and _GIANT will be
1572		 * set, but if CRC is clear, it will still be a valid frame.
1573		 */
1574		if (rxstat & RL_RDESC_STAT_RXERRSUM && !(total_len > 8191 &&
1575		    (rxstat & RL_RDESC_STAT_ERRS) == RL_RDESC_STAT_GIANT)) {
1576			ifp->if_ierrors++;
1577			/*
1578			 * If this is part of a multi-fragment packet,
1579			 * discard all the pieces.
1580			 */
1581			if (sc->rl_head != NULL) {
1582				m_freem(sc->rl_head);
1583				sc->rl_head = sc->rl_tail = NULL;
1584			}
1585			re_newbuf(sc, i, m);
1586			RL_DESC_INC(i);
1587			continue;
1588		}
1589
1590		/*
1591		 * If allocating a replacement mbuf fails,
1592		 * reload the current one.
1593		 */
1594
1595		if (re_newbuf(sc, i, NULL)) {
1596			ifp->if_ierrors++;
1597			if (sc->rl_head != NULL) {
1598				m_freem(sc->rl_head);
1599				sc->rl_head = sc->rl_tail = NULL;
1600			}
1601			re_newbuf(sc, i, m);
1602			RL_DESC_INC(i);
1603			continue;
1604		}
1605
1606		RL_DESC_INC(i);
1607
1608		if (sc->rl_head != NULL) {
1609			m->m_len = total_len % RE_RX_DESC_BUFLEN;
1610			if (m->m_len == 0)
1611				m->m_len = RE_RX_DESC_BUFLEN;
1612			/*
1613			 * Special case: if there's 4 bytes or less
1614			 * in this buffer, the mbuf can be discarded:
1615			 * the last 4 bytes is the CRC, which we don't
1616			 * care about anyway.
1617			 */
1618			if (m->m_len <= ETHER_CRC_LEN) {
1619				sc->rl_tail->m_len -=
1620				    (ETHER_CRC_LEN - m->m_len);
1621				m_freem(m);
1622			} else {
1623				m->m_len -= ETHER_CRC_LEN;
1624				m->m_flags &= ~M_PKTHDR;
1625				sc->rl_tail->m_next = m;
1626			}
1627			m = sc->rl_head;
1628			sc->rl_head = sc->rl_tail = NULL;
1629			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
1630		} else
1631			m->m_pkthdr.len = m->m_len =
1632			    (total_len - ETHER_CRC_LEN);
1633
1634#ifdef RE_FIXUP_RX
1635		re_fixup_rx(m);
1636#endif
1637		ifp->if_ipackets++;
1638		m->m_pkthdr.rcvif = ifp;
1639
1640		/* Do RX checksumming if enabled */
1641
1642		if (ifp->if_capenable & IFCAP_RXCSUM) {
1643
1644			/* Check IP header checksum */
1645			if (rxstat & RL_RDESC_STAT_PROTOID)
1646				m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED;
1647			if (!(rxstat & RL_RDESC_STAT_IPSUMBAD))
1648				m->m_pkthdr.csum_flags |= CSUM_IP_VALID;
1649
1650			/* Check TCP/UDP checksum */
1651			if ((RL_TCPPKT(rxstat) &&
1652			    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
1653			    (RL_UDPPKT(rxstat) &&
1654			    !(rxstat & RL_RDESC_STAT_UDPSUMBAD))) {
1655				m->m_pkthdr.csum_flags |=
1656				    CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
1657				m->m_pkthdr.csum_data = 0xffff;
1658			}
1659		}
1660
1661		if (rxvlan & RL_RDESC_VLANCTL_TAG)
1662			VLAN_INPUT_TAG(ifp, m,
1663			    ntohs((rxvlan & RL_RDESC_VLANCTL_DATA)), continue);
1664		RL_UNLOCK(sc);
1665		(*ifp->if_input)(ifp, m);
1666		RL_LOCK(sc);
1667	}
1668
1669	/* Flush the RX DMA ring */
1670
1671	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1672	    sc->rl_ldata.rl_rx_list_map,
1673	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
1674
1675	sc->rl_ldata.rl_rx_prodidx = i;
1676}
1677
1678static void
1679re_txeof(sc)
1680	struct rl_softc		*sc;
1681{
1682	struct ifnet		*ifp;
1683	u_int32_t		txstat;
1684	int			idx;
1685
1686	ifp = sc->rl_ifp;
1687	idx = sc->rl_ldata.rl_tx_considx;
1688
1689	/* Invalidate the TX descriptor list */
1690
1691	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
1692	    sc->rl_ldata.rl_tx_list_map,
1693	    BUS_DMASYNC_POSTREAD);
1694
1695	while (idx != sc->rl_ldata.rl_tx_prodidx) {
1696
1697		txstat = le32toh(sc->rl_ldata.rl_tx_list[idx].rl_cmdstat);
1698		if (txstat & RL_TDESC_CMD_OWN)
1699			break;
1700
1701		/*
1702		 * We only stash mbufs in the last descriptor
1703		 * in a fragment chain, which also happens to
1704		 * be the only place where the TX status bits
1705		 * are valid.
1706		 */
1707
1708		if (txstat & RL_TDESC_CMD_EOF) {
1709			m_freem(sc->rl_ldata.rl_tx_mbuf[idx]);
1710			sc->rl_ldata.rl_tx_mbuf[idx] = NULL;
1711			bus_dmamap_unload(sc->rl_ldata.rl_mtag,
1712			    sc->rl_ldata.rl_tx_dmamap[idx]);
1713			if (txstat & (RL_TDESC_STAT_EXCESSCOL|
1714			    RL_TDESC_STAT_COLCNT))
1715				ifp->if_collisions++;
1716			if (txstat & RL_TDESC_STAT_TXERRSUM)
1717				ifp->if_oerrors++;
1718			else
1719				ifp->if_opackets++;
1720		}
1721		sc->rl_ldata.rl_tx_free++;
1722		RL_DESC_INC(idx);
1723	}
1724
1725	/* No changes made to the TX ring, so no flush needed */
1726
1727	if (idx != sc->rl_ldata.rl_tx_considx) {
1728		sc->rl_ldata.rl_tx_considx = idx;
1729		ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1730		ifp->if_timer = 0;
1731	}
1732
1733	/*
1734	 * If not all descriptors have been released reaped yet,
1735	 * reload the timer so that we will eventually get another
1736	 * interrupt that will cause us to re-enter this routine.
1737	 * This is done in case the transmitter has gone idle.
1738	 */
1739	if (sc->rl_ldata.rl_tx_free != RL_TX_DESC_CNT)
1740		CSR_WRITE_4(sc, RL_TIMERCNT, 1);
1741}
1742
1743static void
1744re_tick(xsc)
1745	void			*xsc;
1746{
1747	struct rl_softc		*sc;
1748	struct mii_data		*mii;
1749
1750	sc = xsc;
1751
1752	RL_LOCK_ASSERT(sc);
1753
1754	mii = device_get_softc(sc->rl_miibus);
1755
1756	mii_tick(mii);
1757
1758	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
1759}
1760
1761#ifdef DEVICE_POLLING
1762static void
1763re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
1764{
1765	struct rl_softc *sc = ifp->if_softc;
1766
1767	RL_LOCK(sc);
1768	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1769		re_poll_locked(ifp, cmd, count);
1770	RL_UNLOCK(sc);
1771}
1772
1773static void
1774re_poll_locked(struct ifnet *ifp, enum poll_cmd cmd, int count)
1775{
1776	struct rl_softc *sc = ifp->if_softc;
1777
1778	RL_LOCK_ASSERT(sc);
1779
1780	sc->rxcycles = count;
1781	re_rxeof(sc);
1782	re_txeof(sc);
1783
1784	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
1785		re_start_locked(ifp);
1786
1787	if (cmd == POLL_AND_CHECK_STATUS) { /* also check status register */
1788		u_int16_t       status;
1789
1790		status = CSR_READ_2(sc, RL_ISR);
1791		if (status == 0xffff)
1792			return;
1793		if (status)
1794			CSR_WRITE_2(sc, RL_ISR, status);
1795
1796		/*
1797		 * XXX check behaviour on receiver stalls.
1798		 */
1799
1800		if (status & RL_ISR_SYSTEM_ERR) {
1801			re_reset(sc);
1802			re_init_locked(sc);
1803		}
1804	}
1805}
1806#endif /* DEVICE_POLLING */
1807
1808static void
1809re_intr(arg)
1810	void			*arg;
1811{
1812	struct rl_softc		*sc;
1813	struct ifnet		*ifp;
1814	u_int16_t		status;
1815
1816	sc = arg;
1817
1818	RL_LOCK(sc);
1819
1820	ifp = sc->rl_ifp;
1821
1822	if (sc->suspended || !(ifp->if_flags & IFF_UP))
1823		goto done_locked;
1824
1825#ifdef DEVICE_POLLING
1826	if  (ifp->if_capenable & IFCAP_POLLING)
1827		goto done_locked;
1828#endif
1829
1830	for (;;) {
1831
1832		status = CSR_READ_2(sc, RL_ISR);
1833		/* If the card has gone away the read returns 0xffff. */
1834		if (status == 0xffff)
1835			break;
1836		if (status)
1837			CSR_WRITE_2(sc, RL_ISR, status);
1838
1839		if ((status & RL_INTRS_CPLUS) == 0)
1840			break;
1841
1842		if ((status & RL_ISR_RX_OK) ||
1843		    (status & RL_ISR_RX_ERR))
1844			re_rxeof(sc);
1845
1846		if ((status & RL_ISR_TIMEOUT_EXPIRED) ||
1847		    (status & RL_ISR_TX_ERR) ||
1848		    (status & RL_ISR_TX_DESC_UNAVAIL))
1849			re_txeof(sc);
1850
1851		if (status & RL_ISR_SYSTEM_ERR) {
1852			re_reset(sc);
1853			re_init_locked(sc);
1854		}
1855
1856		if (status & RL_ISR_LINKCHG) {
1857			callout_stop(&sc->rl_stat_callout);
1858			re_tick(sc);
1859		}
1860	}
1861
1862	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
1863		re_start_locked(ifp);
1864
1865done_locked:
1866	RL_UNLOCK(sc);
1867}
1868
1869static int
1870re_encap(sc, m_head, idx)
1871	struct rl_softc		*sc;
1872	struct mbuf		**m_head;
1873	int			*idx;
1874{
1875	struct mbuf		*m_new = NULL;
1876	struct rl_dmaload_arg	arg;
1877	bus_dmamap_t		map;
1878	int			error;
1879	struct m_tag		*mtag;
1880
1881	RL_LOCK_ASSERT(sc);
1882
1883	if (sc->rl_ldata.rl_tx_free <= 4)
1884		return (EFBIG);
1885
1886	/*
1887	 * Set up checksum offload. Note: checksum offload bits must
1888	 * appear in all descriptors of a multi-descriptor transmit
1889	 * attempt. This is according to testing done with an 8169
1890	 * chip. This is a requirement.
1891	 */
1892
1893	arg.rl_flags = 0;
1894
1895	if ((*m_head)->m_pkthdr.csum_flags & CSUM_IP)
1896		arg.rl_flags |= RL_TDESC_CMD_IPCSUM;
1897	if ((*m_head)->m_pkthdr.csum_flags & CSUM_TCP)
1898		arg.rl_flags |= RL_TDESC_CMD_TCPCSUM;
1899	if ((*m_head)->m_pkthdr.csum_flags & CSUM_UDP)
1900		arg.rl_flags |= RL_TDESC_CMD_UDPCSUM;
1901
1902	arg.sc = sc;
1903	arg.rl_idx = *idx;
1904	arg.rl_maxsegs = sc->rl_ldata.rl_tx_free;
1905	if (arg.rl_maxsegs > 4)
1906		arg.rl_maxsegs -= 4;
1907	arg.rl_ring = sc->rl_ldata.rl_tx_list;
1908
1909	map = sc->rl_ldata.rl_tx_dmamap[*idx];
1910	error = bus_dmamap_load_mbuf(sc->rl_ldata.rl_mtag, map,
1911	    *m_head, re_dma_map_desc, &arg, BUS_DMA_NOWAIT);
1912
1913	if (error && error != EFBIG) {
1914		if_printf(sc->rl_ifp, "can't map mbuf (error %d)\n", error);
1915		return (ENOBUFS);
1916	}
1917
1918	/* Too many segments to map, coalesce into a single mbuf */
1919
1920	if (error || arg.rl_maxsegs == 0) {
1921		m_new = m_defrag(*m_head, M_DONTWAIT);
1922		if (m_new == NULL)
1923			return (ENOBUFS);
1924		else
1925			*m_head = m_new;
1926
1927		arg.sc = sc;
1928		arg.rl_idx = *idx;
1929		arg.rl_maxsegs = sc->rl_ldata.rl_tx_free;
1930		arg.rl_ring = sc->rl_ldata.rl_tx_list;
1931
1932		error = bus_dmamap_load_mbuf(sc->rl_ldata.rl_mtag, map,
1933		    *m_head, re_dma_map_desc, &arg, BUS_DMA_NOWAIT);
1934		if (error) {
1935			if_printf(sc->rl_ifp, "can't map mbuf (error %d)\n",
1936			    error);
1937			return (EFBIG);
1938		}
1939	}
1940
1941	/*
1942	 * Insure that the map for this transmission
1943	 * is placed at the array index of the last descriptor
1944	 * in this chain.  (Swap last and first dmamaps.)
1945	 */
1946	sc->rl_ldata.rl_tx_dmamap[*idx] =
1947	    sc->rl_ldata.rl_tx_dmamap[arg.rl_idx];
1948	sc->rl_ldata.rl_tx_dmamap[arg.rl_idx] = map;
1949
1950	sc->rl_ldata.rl_tx_mbuf[arg.rl_idx] = *m_head;
1951	sc->rl_ldata.rl_tx_free -= arg.rl_maxsegs;
1952
1953	/*
1954	 * Set up hardware VLAN tagging. Note: vlan tag info must
1955	 * appear in the first descriptor of a multi-descriptor
1956	 * transmission attempt.
1957	 */
1958
1959	mtag = VLAN_OUTPUT_TAG(sc->rl_ifp, *m_head);
1960	if (mtag != NULL)
1961		sc->rl_ldata.rl_tx_list[*idx].rl_vlanctl =
1962		    htole32(htons(VLAN_TAG_VALUE(mtag)) | RL_TDESC_VLANCTL_TAG);
1963
1964	/* Transfer ownership of packet to the chip. */
1965
1966	sc->rl_ldata.rl_tx_list[arg.rl_idx].rl_cmdstat |=
1967	    htole32(RL_TDESC_CMD_OWN);
1968	if (*idx != arg.rl_idx)
1969		sc->rl_ldata.rl_tx_list[*idx].rl_cmdstat |=
1970		    htole32(RL_TDESC_CMD_OWN);
1971
1972	RL_DESC_INC(arg.rl_idx);
1973	*idx = arg.rl_idx;
1974
1975	return (0);
1976}
1977
1978static void
1979re_start(ifp)
1980	struct ifnet		*ifp;
1981{
1982	struct rl_softc		*sc;
1983
1984	sc = ifp->if_softc;
1985	RL_LOCK(sc);
1986	re_start_locked(ifp);
1987	RL_UNLOCK(sc);
1988}
1989
1990/*
1991 * Main transmit routine for C+ and gigE NICs.
1992 */
1993static void
1994re_start_locked(ifp)
1995	struct ifnet		*ifp;
1996{
1997	struct rl_softc		*sc;
1998	struct mbuf		*m_head = NULL;
1999	int			idx, queued = 0;
2000
2001	sc = ifp->if_softc;
2002
2003	RL_LOCK_ASSERT(sc);
2004
2005	idx = sc->rl_ldata.rl_tx_prodidx;
2006
2007	while (sc->rl_ldata.rl_tx_mbuf[idx] == NULL) {
2008		IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
2009		if (m_head == NULL)
2010			break;
2011
2012		if (re_encap(sc, &m_head, &idx)) {
2013			IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
2014			ifp->if_drv_flags |= IFF_DRV_OACTIVE;
2015			break;
2016		}
2017
2018		/*
2019		 * If there's a BPF listener, bounce a copy of this frame
2020		 * to him.
2021		 */
2022		BPF_MTAP(ifp, m_head);
2023
2024		queued++;
2025	}
2026
2027	if (queued == 0)
2028		return;
2029
2030	/* Flush the TX descriptors */
2031
2032	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2033	    sc->rl_ldata.rl_tx_list_map,
2034	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2035
2036	sc->rl_ldata.rl_tx_prodidx = idx;
2037
2038	/*
2039	 * RealTek put the TX poll request register in a different
2040	 * location on the 8169 gigE chip. I don't know why.
2041	 */
2042
2043	if (sc->rl_type == RL_8169)
2044		CSR_WRITE_2(sc, RL_GTXSTART, RL_TXSTART_START);
2045	else
2046		CSR_WRITE_2(sc, RL_TXSTART, RL_TXSTART_START);
2047
2048	/*
2049	 * Use the countdown timer for interrupt moderation.
2050	 * 'TX done' interrupts are disabled. Instead, we reset the
2051	 * countdown timer, which will begin counting until it hits
2052	 * the value in the TIMERINT register, and then trigger an
2053	 * interrupt. Each time we write to the TIMERCNT register,
2054	 * the timer count is reset to 0.
2055	 */
2056	CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2057
2058	/*
2059	 * Set a timeout in case the chip goes out to lunch.
2060	 */
2061	ifp->if_timer = 5;
2062}
2063
2064static void
2065re_init(xsc)
2066	void			*xsc;
2067{
2068	struct rl_softc		*sc = xsc;
2069
2070	RL_LOCK(sc);
2071	re_init_locked(sc);
2072	RL_UNLOCK(sc);
2073}
2074
2075static void
2076re_init_locked(sc)
2077	struct rl_softc		*sc;
2078{
2079	struct ifnet		*ifp = sc->rl_ifp;
2080	struct mii_data		*mii;
2081	u_int32_t		rxcfg = 0;
2082
2083	RL_LOCK_ASSERT(sc);
2084
2085	mii = device_get_softc(sc->rl_miibus);
2086
2087	/*
2088	 * Cancel pending I/O and free all RX/TX buffers.
2089	 */
2090	re_stop(sc);
2091
2092	/*
2093	 * Enable C+ RX and TX mode, as well as VLAN stripping and
2094	 * RX checksum offload. We must configure the C+ register
2095	 * before all others.
2096	 */
2097	CSR_WRITE_2(sc, RL_CPLUS_CMD, RL_CPLUSCMD_RXENB|
2098	    RL_CPLUSCMD_TXENB|RL_CPLUSCMD_PCI_MRW|
2099	    RL_CPLUSCMD_VLANSTRIP|
2100	    (ifp->if_capenable & IFCAP_RXCSUM ?
2101	    RL_CPLUSCMD_RXCSUM_ENB : 0));
2102
2103	/*
2104	 * Init our MAC address.  Even though the chipset
2105	 * documentation doesn't mention it, we need to enter "Config
2106	 * register write enable" mode to modify the ID registers.
2107	 */
2108	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG);
2109	CSR_WRITE_STREAM_4(sc, RL_IDR0,
2110	    *(u_int32_t *)(&IFP2ENADDR(sc->rl_ifp)[0]));
2111	CSR_WRITE_STREAM_4(sc, RL_IDR4,
2112	    *(u_int32_t *)(&IFP2ENADDR(sc->rl_ifp)[4]));
2113	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
2114
2115	/*
2116	 * For C+ mode, initialize the RX descriptors and mbufs.
2117	 */
2118	re_rx_list_init(sc);
2119	re_tx_list_init(sc);
2120
2121	/*
2122	 * Enable transmit and receive.
2123	 */
2124	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
2125
2126	/*
2127	 * Set the initial TX and RX configuration.
2128	 */
2129	if (sc->rl_testmode) {
2130		if (sc->rl_type == RL_8169)
2131			CSR_WRITE_4(sc, RL_TXCFG,
2132			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON);
2133		else
2134			CSR_WRITE_4(sc, RL_TXCFG,
2135			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON_CPLUS);
2136	} else
2137		CSR_WRITE_4(sc, RL_TXCFG, RL_TXCFG_CONFIG);
2138	CSR_WRITE_4(sc, RL_RXCFG, RL_RXCFG_CONFIG);
2139
2140	/* Set the individual bit to receive frames for this host only. */
2141	rxcfg = CSR_READ_4(sc, RL_RXCFG);
2142	rxcfg |= RL_RXCFG_RX_INDIV;
2143
2144	/* If we want promiscuous mode, set the allframes bit. */
2145	if (ifp->if_flags & IFF_PROMISC)
2146		rxcfg |= RL_RXCFG_RX_ALLPHYS;
2147	else
2148		rxcfg &= ~RL_RXCFG_RX_ALLPHYS;
2149	CSR_WRITE_4(sc, RL_RXCFG, rxcfg);
2150
2151	/*
2152	 * Set capture broadcast bit to capture broadcast frames.
2153	 */
2154	if (ifp->if_flags & IFF_BROADCAST)
2155		rxcfg |= RL_RXCFG_RX_BROAD;
2156	else
2157		rxcfg &= ~RL_RXCFG_RX_BROAD;
2158	CSR_WRITE_4(sc, RL_RXCFG, rxcfg);
2159
2160	/*
2161	 * Program the multicast filter, if necessary.
2162	 */
2163	re_setmulti(sc);
2164
2165#ifdef DEVICE_POLLING
2166	/*
2167	 * Disable interrupts if we are polling.
2168	 */
2169	if (ifp->if_capenable & IFCAP_POLLING)
2170		CSR_WRITE_2(sc, RL_IMR, 0);
2171	else	/* otherwise ... */
2172#endif
2173	/*
2174	 * Enable interrupts.
2175	 */
2176	if (sc->rl_testmode)
2177		CSR_WRITE_2(sc, RL_IMR, 0);
2178	else
2179		CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2180
2181	/* Set initial TX threshold */
2182	sc->rl_txthresh = RL_TX_THRESH_INIT;
2183
2184	/* Start RX/TX process. */
2185	CSR_WRITE_4(sc, RL_MISSEDPKT, 0);
2186#ifdef notdef
2187	/* Enable receiver and transmitter. */
2188	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
2189#endif
2190	/*
2191	 * Load the addresses of the RX and TX lists into the chip.
2192	 */
2193
2194	CSR_WRITE_4(sc, RL_RXLIST_ADDR_HI,
2195	    RL_ADDR_HI(sc->rl_ldata.rl_rx_list_addr));
2196	CSR_WRITE_4(sc, RL_RXLIST_ADDR_LO,
2197	    RL_ADDR_LO(sc->rl_ldata.rl_rx_list_addr));
2198
2199	CSR_WRITE_4(sc, RL_TXLIST_ADDR_HI,
2200	    RL_ADDR_HI(sc->rl_ldata.rl_tx_list_addr));
2201	CSR_WRITE_4(sc, RL_TXLIST_ADDR_LO,
2202	    RL_ADDR_LO(sc->rl_ldata.rl_tx_list_addr));
2203
2204	CSR_WRITE_1(sc, RL_EARLY_TX_THRESH, 16);
2205
2206	/*
2207	 * Initialize the timer interrupt register so that
2208	 * a timer interrupt will be generated once the timer
2209	 * reaches a certain number of ticks. The timer is
2210	 * reloaded on each transmit. This gives us TX interrupt
2211	 * moderation, which dramatically improves TX frame rate.
2212	 */
2213	if (sc->rl_type == RL_8169)
2214		CSR_WRITE_4(sc, RL_TIMERINT_8169, 0x800);
2215	else
2216		CSR_WRITE_4(sc, RL_TIMERINT, 0x400);
2217
2218	/*
2219	 * For 8169 gigE NICs, set the max allowed RX packet
2220	 * size so we can receive jumbo frames.
2221	 */
2222	if (sc->rl_type == RL_8169)
2223		CSR_WRITE_2(sc, RL_MAXRXPKTLEN, 16383);
2224
2225	if (sc->rl_testmode)
2226		return;
2227
2228	mii_mediachg(mii);
2229
2230	CSR_WRITE_1(sc, RL_CFG1, RL_CFG1_DRVLOAD|RL_CFG1_FULLDUPLEX);
2231
2232	ifp->if_drv_flags |= IFF_DRV_RUNNING;
2233	ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
2234
2235	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
2236}
2237
2238/*
2239 * Set media options.
2240 */
2241static int
2242re_ifmedia_upd(ifp)
2243	struct ifnet		*ifp;
2244{
2245	struct rl_softc		*sc;
2246	struct mii_data		*mii;
2247
2248	sc = ifp->if_softc;
2249	mii = device_get_softc(sc->rl_miibus);
2250	RL_LOCK(sc);
2251	mii_mediachg(mii);
2252	RL_UNLOCK(sc);
2253
2254	return (0);
2255}
2256
2257/*
2258 * Report current media status.
2259 */
2260static void
2261re_ifmedia_sts(ifp, ifmr)
2262	struct ifnet		*ifp;
2263	struct ifmediareq	*ifmr;
2264{
2265	struct rl_softc		*sc;
2266	struct mii_data		*mii;
2267
2268	sc = ifp->if_softc;
2269	mii = device_get_softc(sc->rl_miibus);
2270
2271	RL_LOCK(sc);
2272	mii_pollstat(mii);
2273	RL_UNLOCK(sc);
2274	ifmr->ifm_active = mii->mii_media_active;
2275	ifmr->ifm_status = mii->mii_media_status;
2276}
2277
2278static int
2279re_ioctl(ifp, command, data)
2280	struct ifnet		*ifp;
2281	u_long			command;
2282	caddr_t			data;
2283{
2284	struct rl_softc		*sc = ifp->if_softc;
2285	struct ifreq		*ifr = (struct ifreq *) data;
2286	struct mii_data		*mii;
2287	int			error = 0;
2288
2289	switch (command) {
2290	case SIOCSIFMTU:
2291		RL_LOCK(sc);
2292		if (ifr->ifr_mtu > RL_JUMBO_MTU)
2293			error = EINVAL;
2294		ifp->if_mtu = ifr->ifr_mtu;
2295		RL_UNLOCK(sc);
2296		break;
2297	case SIOCSIFFLAGS:
2298		RL_LOCK(sc);
2299		if (ifp->if_flags & IFF_UP)
2300			re_init_locked(sc);
2301		else if (ifp->if_drv_flags & IFF_DRV_RUNNING)
2302			re_stop(sc);
2303		RL_UNLOCK(sc);
2304		break;
2305	case SIOCADDMULTI:
2306	case SIOCDELMULTI:
2307		RL_LOCK(sc);
2308		re_setmulti(sc);
2309		RL_UNLOCK(sc);
2310		break;
2311	case SIOCGIFMEDIA:
2312	case SIOCSIFMEDIA:
2313		mii = device_get_softc(sc->rl_miibus);
2314		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, command);
2315		break;
2316	case SIOCSIFCAP:
2317	    {
2318		int mask = ifr->ifr_reqcap ^ ifp->if_capenable;
2319#ifdef DEVICE_POLLING
2320		if (mask & IFCAP_POLLING) {
2321			if (ifr->ifr_reqcap & IFCAP_POLLING) {
2322				error = ether_poll_register(re_poll, ifp);
2323				if (error)
2324					return(error);
2325				RL_LOCK(sc);
2326				/* Disable interrupts */
2327				CSR_WRITE_2(sc, RL_IMR, 0x0000);
2328				ifp->if_capenable |= IFCAP_POLLING;
2329				RL_UNLOCK(sc);
2330
2331			} else {
2332				error = ether_poll_deregister(ifp);
2333				/* Enable interrupts. */
2334				RL_LOCK(sc);
2335				CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2336				ifp->if_capenable &= ~IFCAP_POLLING;
2337				RL_UNLOCK(sc);
2338			}
2339		}
2340#endif /* DEVICE_POLLING */
2341		if (mask & IFCAP_HWCSUM) {
2342			RL_LOCK(sc);
2343			ifp->if_capenable |= ifr->ifr_reqcap & IFCAP_HWCSUM;
2344			if (ifp->if_capenable & IFCAP_TXCSUM)
2345				ifp->if_hwassist = RE_CSUM_FEATURES;
2346			else
2347				ifp->if_hwassist = 0;
2348			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
2349				re_init_locked(sc);
2350			RL_UNLOCK(sc);
2351		}
2352	    }
2353		break;
2354	default:
2355		error = ether_ioctl(ifp, command, data);
2356		break;
2357	}
2358
2359	return (error);
2360}
2361
2362static void
2363re_watchdog(ifp)
2364	struct ifnet		*ifp;
2365{
2366	struct rl_softc		*sc;
2367
2368	sc = ifp->if_softc;
2369	RL_LOCK(sc);
2370	if_printf(ifp, "watchdog timeout\n");
2371	ifp->if_oerrors++;
2372
2373	re_txeof(sc);
2374	re_rxeof(sc);
2375	re_init_locked(sc);
2376
2377	RL_UNLOCK(sc);
2378}
2379
2380/*
2381 * Stop the adapter and free any mbufs allocated to the
2382 * RX and TX lists.
2383 */
2384static void
2385re_stop(sc)
2386	struct rl_softc		*sc;
2387{
2388	register int		i;
2389	struct ifnet		*ifp;
2390
2391	RL_LOCK_ASSERT(sc);
2392
2393	ifp = sc->rl_ifp;
2394	ifp->if_timer = 0;
2395
2396	callout_stop(&sc->rl_stat_callout);
2397	ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
2398
2399	CSR_WRITE_1(sc, RL_COMMAND, 0x00);
2400	CSR_WRITE_2(sc, RL_IMR, 0x0000);
2401
2402	if (sc->rl_head != NULL) {
2403		m_freem(sc->rl_head);
2404		sc->rl_head = sc->rl_tail = NULL;
2405	}
2406
2407	/* Free the TX list buffers. */
2408
2409	for (i = 0; i < RL_TX_DESC_CNT; i++) {
2410		if (sc->rl_ldata.rl_tx_mbuf[i] != NULL) {
2411			bus_dmamap_unload(sc->rl_ldata.rl_mtag,
2412			    sc->rl_ldata.rl_tx_dmamap[i]);
2413			m_freem(sc->rl_ldata.rl_tx_mbuf[i]);
2414			sc->rl_ldata.rl_tx_mbuf[i] = NULL;
2415		}
2416	}
2417
2418	/* Free the RX list buffers. */
2419
2420	for (i = 0; i < RL_RX_DESC_CNT; i++) {
2421		if (sc->rl_ldata.rl_rx_mbuf[i] != NULL) {
2422			bus_dmamap_unload(sc->rl_ldata.rl_mtag,
2423			    sc->rl_ldata.rl_rx_dmamap[i]);
2424			m_freem(sc->rl_ldata.rl_rx_mbuf[i]);
2425			sc->rl_ldata.rl_rx_mbuf[i] = NULL;
2426		}
2427	}
2428}
2429
2430/*
2431 * Device suspend routine.  Stop the interface and save some PCI
2432 * settings in case the BIOS doesn't restore them properly on
2433 * resume.
2434 */
2435static int
2436re_suspend(dev)
2437	device_t		dev;
2438{
2439	struct rl_softc		*sc;
2440
2441	sc = device_get_softc(dev);
2442
2443	RL_LOCK(sc);
2444	re_stop(sc);
2445	sc->suspended = 1;
2446	RL_UNLOCK(sc);
2447
2448	return (0);
2449}
2450
2451/*
2452 * Device resume routine.  Restore some PCI settings in case the BIOS
2453 * doesn't, re-enable busmastering, and restart the interface if
2454 * appropriate.
2455 */
2456static int
2457re_resume(dev)
2458	device_t		dev;
2459{
2460	struct rl_softc		*sc;
2461	struct ifnet		*ifp;
2462
2463	sc = device_get_softc(dev);
2464
2465	RL_LOCK(sc);
2466
2467	ifp = sc->rl_ifp;
2468
2469	/* reinitialize interface if necessary */
2470	if (ifp->if_flags & IFF_UP)
2471		re_init_locked(sc);
2472
2473	sc->suspended = 0;
2474	RL_UNLOCK(sc);
2475
2476	return (0);
2477}
2478
2479/*
2480 * Stop all chip I/O so that the kernel's probe routines don't
2481 * get confused by errant DMAs when rebooting.
2482 */
2483static void
2484re_shutdown(dev)
2485	device_t		dev;
2486{
2487	struct rl_softc		*sc;
2488
2489	sc = device_get_softc(dev);
2490
2491	RL_LOCK(sc);
2492	re_stop(sc);
2493	/*
2494	 * Mark interface as down since otherwise we will panic if
2495	 * interrupt comes in later on, which can happen in some
2496	 * cases.
2497	 */
2498	sc->rl_ifp->if_flags &= ~IFF_UP;
2499	RL_UNLOCK(sc);
2500}
2501