if_re.c revision 147256
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 147256 2005-06-10 16:49:24Z brooks $");
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	u_int32_t		*addr;
941
942	if (error)
943		return;
944
945	KASSERT(nseg == 1, ("too many DMA segments, %d should be 1", nseg));
946	addr = arg;
947	*addr = segs->ds_addr;
948}
949
950static int
951re_allocmem(dev, sc)
952	device_t		dev;
953	struct rl_softc		*sc;
954{
955	int			error;
956	int			nseg;
957	int			i;
958
959	/*
960	 * Allocate map for RX mbufs.
961	 */
962	nseg = 32;
963	error = bus_dma_tag_create(sc->rl_parent_tag, ETHER_ALIGN, 0,
964	    BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
965	    NULL, MCLBYTES * nseg, nseg, MCLBYTES, BUS_DMA_ALLOCNOW,
966	    NULL, NULL, &sc->rl_ldata.rl_mtag);
967	if (error) {
968		device_printf(dev, "could not allocate dma tag\n");
969		return (ENOMEM);
970	}
971
972	/*
973	 * Allocate map for TX descriptor list.
974	 */
975	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
976	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
977	    NULL, RL_TX_LIST_SZ, 1, RL_TX_LIST_SZ, BUS_DMA_ALLOCNOW,
978	    NULL, NULL, &sc->rl_ldata.rl_tx_list_tag);
979	if (error) {
980		device_printf(dev, "could not allocate dma tag\n");
981		return (ENOMEM);
982	}
983
984	/* Allocate DMA'able memory for the TX ring */
985
986	error = bus_dmamem_alloc(sc->rl_ldata.rl_tx_list_tag,
987	    (void **)&sc->rl_ldata.rl_tx_list, BUS_DMA_NOWAIT | BUS_DMA_ZERO,
988	    &sc->rl_ldata.rl_tx_list_map);
989	if (error)
990		return (ENOMEM);
991
992	/* Load the map for the TX ring. */
993
994	error = bus_dmamap_load(sc->rl_ldata.rl_tx_list_tag,
995	     sc->rl_ldata.rl_tx_list_map, sc->rl_ldata.rl_tx_list,
996	     RL_TX_LIST_SZ, re_dma_map_addr,
997	     &sc->rl_ldata.rl_tx_list_addr, BUS_DMA_NOWAIT);
998
999	/* Create DMA maps for TX buffers */
1000
1001	for (i = 0; i < RL_TX_DESC_CNT; i++) {
1002		error = bus_dmamap_create(sc->rl_ldata.rl_mtag, 0,
1003			    &sc->rl_ldata.rl_tx_dmamap[i]);
1004		if (error) {
1005			device_printf(dev, "can't create DMA map for TX\n");
1006			return (ENOMEM);
1007		}
1008	}
1009
1010	/*
1011	 * Allocate map for RX descriptor list.
1012	 */
1013	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
1014	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1015	    NULL, RL_RX_LIST_SZ, 1, RL_RX_LIST_SZ, BUS_DMA_ALLOCNOW,
1016	    NULL, NULL, &sc->rl_ldata.rl_rx_list_tag);
1017	if (error) {
1018		device_printf(dev, "could not allocate dma tag\n");
1019		return (ENOMEM);
1020	}
1021
1022	/* Allocate DMA'able memory for the RX ring */
1023
1024	error = bus_dmamem_alloc(sc->rl_ldata.rl_rx_list_tag,
1025	    (void **)&sc->rl_ldata.rl_rx_list, BUS_DMA_NOWAIT | BUS_DMA_ZERO,
1026	    &sc->rl_ldata.rl_rx_list_map);
1027	if (error)
1028		return (ENOMEM);
1029
1030	/* Load the map for the RX ring. */
1031
1032	error = bus_dmamap_load(sc->rl_ldata.rl_rx_list_tag,
1033	     sc->rl_ldata.rl_rx_list_map, sc->rl_ldata.rl_rx_list,
1034	     RL_RX_LIST_SZ, re_dma_map_addr,
1035	     &sc->rl_ldata.rl_rx_list_addr, BUS_DMA_NOWAIT);
1036
1037	/* Create DMA maps for RX buffers */
1038
1039	for (i = 0; i < RL_RX_DESC_CNT; i++) {
1040		error = bus_dmamap_create(sc->rl_ldata.rl_mtag, 0,
1041			    &sc->rl_ldata.rl_rx_dmamap[i]);
1042		if (error) {
1043			device_printf(dev, "can't create DMA map for RX\n");
1044			return (ENOMEM);
1045		}
1046	}
1047
1048	return (0);
1049}
1050
1051/*
1052 * Attach the interface. Allocate softc structures, do ifmedia
1053 * setup and ethernet/BPF attach.
1054 */
1055static int
1056re_attach(dev)
1057	device_t		dev;
1058{
1059	u_char			eaddr[ETHER_ADDR_LEN];
1060	u_int16_t		as[3];
1061	struct rl_softc		*sc;
1062	struct ifnet		*ifp;
1063	struct rl_hwrev		*hw_rev;
1064	int			hwrev;
1065	u_int16_t		re_did = 0;
1066	int			unit, error = 0, rid, i;
1067
1068	sc = device_get_softc(dev);
1069	unit = device_get_unit(dev);
1070
1071	mtx_init(&sc->rl_mtx, device_get_nameunit(dev), MTX_NETWORK_LOCK,
1072	    MTX_DEF);
1073	/*
1074	 * Map control/status registers.
1075	 */
1076	pci_enable_busmaster(dev);
1077
1078	rid = RL_RID;
1079	sc->rl_res = bus_alloc_resource_any(dev, RL_RES, &rid,
1080	    RF_ACTIVE);
1081
1082	if (sc->rl_res == NULL) {
1083		printf ("re%d: couldn't map ports/memory\n", unit);
1084		error = ENXIO;
1085		goto fail;
1086	}
1087
1088	sc->rl_btag = rman_get_bustag(sc->rl_res);
1089	sc->rl_bhandle = rman_get_bushandle(sc->rl_res);
1090
1091	/* Allocate interrupt */
1092	rid = 0;
1093	sc->rl_irq = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
1094	    RF_SHAREABLE | RF_ACTIVE);
1095
1096	if (sc->rl_irq == NULL) {
1097		printf("re%d: couldn't map interrupt\n", unit);
1098		error = ENXIO;
1099		goto fail;
1100	}
1101
1102	/* Reset the adapter. */
1103	RL_LOCK(sc);
1104	re_reset(sc);
1105	RL_UNLOCK(sc);
1106
1107	hw_rev = re_hwrevs;
1108	hwrev = CSR_READ_4(sc, RL_TXCFG) & RL_TXCFG_HWREV;
1109	while (hw_rev->rl_desc != NULL) {
1110		if (hw_rev->rl_rev == hwrev) {
1111			sc->rl_type = hw_rev->rl_type;
1112			break;
1113		}
1114		hw_rev++;
1115	}
1116
1117	if (sc->rl_type == RL_8169) {
1118
1119		/* Set RX length mask */
1120
1121		sc->rl_rxlenmask = RL_RDESC_STAT_GFRAGLEN;
1122
1123		/* Force station address autoload from the EEPROM */
1124
1125		CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_AUTOLOAD);
1126		for (i = 0; i < RL_TIMEOUT; i++) {
1127			if (!(CSR_READ_1(sc, RL_EECMD) & RL_EEMODE_AUTOLOAD))
1128				break;
1129			DELAY(100);
1130		}
1131		if (i == RL_TIMEOUT)
1132			printf ("re%d: eeprom autoload timed out\n", unit);
1133
1134			for (i = 0; i < ETHER_ADDR_LEN; i++)
1135				eaddr[i] = CSR_READ_1(sc, RL_IDR0 + i);
1136	} else {
1137
1138		/* Set RX length mask */
1139
1140		sc->rl_rxlenmask = RL_RDESC_STAT_FRAGLEN;
1141
1142		sc->rl_eecmd_read = RL_EECMD_READ_6BIT;
1143		re_read_eeprom(sc, (caddr_t)&re_did, 0, 1, 0);
1144		if (re_did != 0x8129)
1145			sc->rl_eecmd_read = RL_EECMD_READ_8BIT;
1146
1147		/*
1148		 * Get station address from the EEPROM.
1149		 */
1150		re_read_eeprom(sc, (caddr_t)as, RL_EE_EADDR, 3, 0);
1151		for (i = 0; i < 3; i++) {
1152			eaddr[(i * 2) + 0] = as[i] & 0xff;
1153			eaddr[(i * 2) + 1] = as[i] >> 8;
1154		}
1155	}
1156
1157	sc->rl_unit = unit;
1158
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	/* Do MII setup */
1182	if (mii_phy_probe(dev, &sc->rl_miibus,
1183	    re_ifmedia_upd, re_ifmedia_sts)) {
1184		printf("re%d: MII without any phy!\n", sc->rl_unit);
1185		error = ENXIO;
1186		goto fail;
1187	}
1188
1189	ifp = sc->rl_ifp = if_alloc(IFT_ETHER);
1190	if (ifp == NULL) {
1191		printf("re%d: can not if_alloc()\n", sc->rl_unit);
1192		error = ENOSPC;
1193		goto fail;
1194	}
1195	ifp->if_softc = sc;
1196	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1197	ifp->if_mtu = ETHERMTU;
1198	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1199	ifp->if_ioctl = re_ioctl;
1200	ifp->if_capabilities = IFCAP_VLAN_MTU;
1201	ifp->if_start = re_start;
1202	ifp->if_hwassist = /*RE_CSUM_FEATURES*/0;
1203	ifp->if_capabilities |= IFCAP_HWCSUM|IFCAP_VLAN_HWTAGGING;
1204#ifdef DEVICE_POLLING
1205	ifp->if_capabilities |= IFCAP_POLLING;
1206#endif
1207	ifp->if_watchdog = re_watchdog;
1208	ifp->if_init = re_init;
1209	if (sc->rl_type == RL_8169)
1210		ifp->if_baudrate = 1000000000;
1211	else
1212		ifp->if_baudrate = 100000000;
1213	IFQ_SET_MAXLEN(&ifp->if_snd,  RL_IFQ_MAXLEN);
1214	ifp->if_snd.ifq_drv_maxlen = RL_IFQ_MAXLEN;
1215	IFQ_SET_READY(&ifp->if_snd);
1216	ifp->if_capenable = ifp->if_capabilities & ~IFCAP_HWCSUM;
1217
1218	callout_handle_init(&sc->rl_stat_ch);
1219
1220	/*
1221	 * Call MI attach routine.
1222	 */
1223	ether_ifattach(ifp, eaddr);
1224
1225	/* Perform hardware diagnostic. */
1226	error = re_diag(sc);
1227
1228	if (error) {
1229		printf("re%d: attach aborted due to hardware diag failure\n",
1230		    unit);
1231		ether_ifdetach(ifp);
1232		if_free(ifp);
1233		goto fail;
1234	}
1235
1236	/* Hook interrupt last to avoid having to lock softc */
1237	error = bus_setup_intr(dev, sc->rl_irq, INTR_TYPE_NET | INTR_MPSAFE,
1238	    re_intr, sc, &sc->rl_intrhand);
1239	if (error) {
1240		printf("re%d: couldn't set up irq\n", unit);
1241		ether_ifdetach(ifp);
1242		if_free(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	int			attached;
1267
1268	sc = device_get_softc(dev);
1269	ifp = sc->rl_ifp;
1270	KASSERT(mtx_initialized(&sc->rl_mtx), ("re mutex not initialized"));
1271
1272	attached = device_is_attached(dev);
1273	/* These should only be active if attach succeeded */
1274	if (attached)
1275		ether_ifdetach(ifp);
1276	if (ifp == NULL)
1277		if_free(ifp);
1278
1279	RL_LOCK(sc);
1280#if 0
1281	sc->suspended = 1;
1282#endif
1283
1284	/* These should only be active if attach succeeded */
1285	if (attached) {
1286		re_stop(sc);
1287		/*
1288		 * Force off the IFF_UP flag here, in case someone
1289		 * still had a BPF descriptor attached to this
1290		 * interface. If they do, ether_ifdetach() will cause
1291		 * the BPF code to try and clear the promisc mode
1292		 * flag, which will bubble down to re_ioctl(),
1293		 * which will try to call re_init() again. This will
1294		 * turn the NIC back on and restart the MII ticker,
1295		 * which will panic the system when the kernel tries
1296		 * to invoke the re_tick() function that isn't there
1297		 * anymore.
1298		 */
1299		ifp->if_flags &= ~IFF_UP;
1300	}
1301	if (sc->rl_miibus)
1302		device_delete_child(dev, sc->rl_miibus);
1303	bus_generic_detach(dev);
1304
1305	/*
1306	 * The rest is resource deallocation, so we should already be
1307	 * stopped here.
1308	 */
1309	RL_UNLOCK(sc);
1310
1311	if (sc->rl_intrhand)
1312		bus_teardown_intr(dev, sc->rl_irq, sc->rl_intrhand);
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_flags &= ~IFF_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
1749	sc = xsc;
1750	RL_LOCK(sc);
1751	re_tick_locked(sc);
1752	RL_UNLOCK(sc);
1753}
1754
1755static void
1756re_tick_locked(sc)
1757	struct rl_softc		*sc;
1758{
1759	struct mii_data		*mii;
1760
1761	RL_LOCK_ASSERT(sc);
1762
1763	mii = device_get_softc(sc->rl_miibus);
1764
1765	mii_tick(mii);
1766
1767	sc->rl_stat_ch = timeout(re_tick, sc, hz);
1768}
1769
1770#ifdef DEVICE_POLLING
1771static void
1772re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
1773{
1774	struct rl_softc *sc = ifp->if_softc;
1775
1776	RL_LOCK(sc);
1777	re_poll_locked(ifp, cmd, count);
1778	RL_UNLOCK(sc);
1779}
1780
1781static void
1782re_poll_locked(struct ifnet *ifp, enum poll_cmd cmd, int count)
1783{
1784	struct rl_softc *sc = ifp->if_softc;
1785
1786	RL_LOCK_ASSERT(sc);
1787
1788	if (!(ifp->if_capenable & IFCAP_POLLING)) {
1789		ether_poll_deregister(ifp);
1790		cmd = POLL_DEREGISTER;
1791	}
1792	if (cmd == POLL_DEREGISTER) { /* final call, enable interrupts */
1793		CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
1794		return;
1795	}
1796
1797	sc->rxcycles = count;
1798	re_rxeof(sc);
1799	re_txeof(sc);
1800
1801	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
1802		re_start_locked(ifp);
1803
1804	if (cmd == POLL_AND_CHECK_STATUS) { /* also check status register */
1805		u_int16_t       status;
1806
1807		status = CSR_READ_2(sc, RL_ISR);
1808		if (status == 0xffff)
1809			return;
1810		if (status)
1811			CSR_WRITE_2(sc, RL_ISR, status);
1812
1813		/*
1814		 * XXX check behaviour on receiver stalls.
1815		 */
1816
1817		if (status & RL_ISR_SYSTEM_ERR) {
1818			re_reset(sc);
1819			re_init_locked(sc);
1820		}
1821	}
1822}
1823#endif /* DEVICE_POLLING */
1824
1825static void
1826re_intr(arg)
1827	void			*arg;
1828{
1829	struct rl_softc		*sc;
1830	struct ifnet		*ifp;
1831	u_int16_t		status;
1832
1833	sc = arg;
1834
1835	RL_LOCK(sc);
1836
1837	ifp = sc->rl_ifp;
1838
1839	if (sc->suspended || !(ifp->if_flags & IFF_UP))
1840		goto done_locked;
1841
1842#ifdef DEVICE_POLLING
1843	if  (ifp->if_flags & IFF_POLLING)
1844		goto done_locked;
1845	if ((ifp->if_capenable & IFCAP_POLLING) &&
1846	    ether_poll_register(re_poll, ifp)) { /* ok, disable interrupts */
1847		CSR_WRITE_2(sc, RL_IMR, 0x0000);
1848		re_poll_locked(ifp, 0, 1);
1849		goto done_locked;
1850	}
1851#endif /* DEVICE_POLLING */
1852
1853	for (;;) {
1854
1855		status = CSR_READ_2(sc, RL_ISR);
1856		/* If the card has gone away the read returns 0xffff. */
1857		if (status == 0xffff)
1858			break;
1859		if (status)
1860			CSR_WRITE_2(sc, RL_ISR, status);
1861
1862		if ((status & RL_INTRS_CPLUS) == 0)
1863			break;
1864
1865		if ((status & RL_ISR_RX_OK) ||
1866		    (status & RL_ISR_RX_ERR))
1867			re_rxeof(sc);
1868
1869		if ((status & RL_ISR_TIMEOUT_EXPIRED) ||
1870		    (status & RL_ISR_TX_ERR) ||
1871		    (status & RL_ISR_TX_DESC_UNAVAIL))
1872			re_txeof(sc);
1873
1874		if (status & RL_ISR_SYSTEM_ERR) {
1875			re_reset(sc);
1876			re_init_locked(sc);
1877		}
1878
1879		if (status & RL_ISR_LINKCHG) {
1880			untimeout(re_tick, sc, sc->rl_stat_ch);
1881			re_tick_locked(sc);
1882		}
1883	}
1884
1885	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
1886		re_start_locked(ifp);
1887
1888done_locked:
1889	RL_UNLOCK(sc);
1890}
1891
1892static int
1893re_encap(sc, m_head, idx)
1894	struct rl_softc		*sc;
1895	struct mbuf		**m_head;
1896	int			*idx;
1897{
1898	struct mbuf		*m_new = NULL;
1899	struct rl_dmaload_arg	arg;
1900	bus_dmamap_t		map;
1901	int			error;
1902	struct m_tag		*mtag;
1903
1904	RL_LOCK_ASSERT(sc);
1905
1906	if (sc->rl_ldata.rl_tx_free <= 4)
1907		return (EFBIG);
1908
1909	/*
1910	 * Set up checksum offload. Note: checksum offload bits must
1911	 * appear in all descriptors of a multi-descriptor transmit
1912	 * attempt. This is according to testing done with an 8169
1913	 * chip. This is a requirement.
1914	 */
1915
1916	arg.rl_flags = 0;
1917
1918	if ((*m_head)->m_pkthdr.csum_flags & CSUM_IP)
1919		arg.rl_flags |= RL_TDESC_CMD_IPCSUM;
1920	if ((*m_head)->m_pkthdr.csum_flags & CSUM_TCP)
1921		arg.rl_flags |= RL_TDESC_CMD_TCPCSUM;
1922	if ((*m_head)->m_pkthdr.csum_flags & CSUM_UDP)
1923		arg.rl_flags |= RL_TDESC_CMD_UDPCSUM;
1924
1925	arg.sc = sc;
1926	arg.rl_idx = *idx;
1927	arg.rl_maxsegs = sc->rl_ldata.rl_tx_free;
1928	if (arg.rl_maxsegs > 4)
1929		arg.rl_maxsegs -= 4;
1930	arg.rl_ring = sc->rl_ldata.rl_tx_list;
1931
1932	map = sc->rl_ldata.rl_tx_dmamap[*idx];
1933	error = bus_dmamap_load_mbuf(sc->rl_ldata.rl_mtag, map,
1934	    *m_head, re_dma_map_desc, &arg, BUS_DMA_NOWAIT);
1935
1936	if (error && error != EFBIG) {
1937		printf("re%d: can't map mbuf (error %d)\n", sc->rl_unit, error);
1938		return (ENOBUFS);
1939	}
1940
1941	/* Too many segments to map, coalesce into a single mbuf */
1942
1943	if (error || arg.rl_maxsegs == 0) {
1944		m_new = m_defrag(*m_head, M_DONTWAIT);
1945		if (m_new == NULL)
1946			return (ENOBUFS);
1947		else
1948			*m_head = m_new;
1949
1950		arg.sc = sc;
1951		arg.rl_idx = *idx;
1952		arg.rl_maxsegs = sc->rl_ldata.rl_tx_free;
1953		arg.rl_ring = sc->rl_ldata.rl_tx_list;
1954
1955		error = bus_dmamap_load_mbuf(sc->rl_ldata.rl_mtag, map,
1956		    *m_head, re_dma_map_desc, &arg, BUS_DMA_NOWAIT);
1957		if (error) {
1958			printf("re%d: can't map mbuf (error %d)\n",
1959			    sc->rl_unit, error);
1960			return (EFBIG);
1961		}
1962	}
1963
1964	/*
1965	 * Insure that the map for this transmission
1966	 * is placed at the array index of the last descriptor
1967	 * in this chain.  (Swap last and first dmamaps.)
1968	 */
1969	sc->rl_ldata.rl_tx_dmamap[*idx] =
1970	    sc->rl_ldata.rl_tx_dmamap[arg.rl_idx];
1971	sc->rl_ldata.rl_tx_dmamap[arg.rl_idx] = map;
1972
1973	sc->rl_ldata.rl_tx_mbuf[arg.rl_idx] = *m_head;
1974	sc->rl_ldata.rl_tx_free -= arg.rl_maxsegs;
1975
1976	/*
1977	 * Set up hardware VLAN tagging. Note: vlan tag info must
1978	 * appear in the first descriptor of a multi-descriptor
1979	 * transmission attempt.
1980	 */
1981
1982	mtag = VLAN_OUTPUT_TAG(sc->rl_ifp, *m_head);
1983	if (mtag != NULL)
1984		sc->rl_ldata.rl_tx_list[*idx].rl_vlanctl =
1985		    htole32(htons(VLAN_TAG_VALUE(mtag)) | RL_TDESC_VLANCTL_TAG);
1986
1987	/* Transfer ownership of packet to the chip. */
1988
1989	sc->rl_ldata.rl_tx_list[arg.rl_idx].rl_cmdstat |=
1990	    htole32(RL_TDESC_CMD_OWN);
1991	if (*idx != arg.rl_idx)
1992		sc->rl_ldata.rl_tx_list[*idx].rl_cmdstat |=
1993		    htole32(RL_TDESC_CMD_OWN);
1994
1995	RL_DESC_INC(arg.rl_idx);
1996	*idx = arg.rl_idx;
1997
1998	return (0);
1999}
2000
2001static void
2002re_start(ifp)
2003	struct ifnet		*ifp;
2004{
2005	struct rl_softc		*sc;
2006
2007	sc = ifp->if_softc;
2008	RL_LOCK(sc);
2009	re_start_locked(ifp);
2010	RL_UNLOCK(sc);
2011}
2012
2013/*
2014 * Main transmit routine for C+ and gigE NICs.
2015 */
2016static void
2017re_start_locked(ifp)
2018	struct ifnet		*ifp;
2019{
2020	struct rl_softc		*sc;
2021	struct mbuf		*m_head = NULL;
2022	int			idx, queued = 0;
2023
2024	sc = ifp->if_softc;
2025
2026	RL_LOCK_ASSERT(sc);
2027
2028	idx = sc->rl_ldata.rl_tx_prodidx;
2029
2030	while (sc->rl_ldata.rl_tx_mbuf[idx] == NULL) {
2031		IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
2032		if (m_head == NULL)
2033			break;
2034
2035		if (re_encap(sc, &m_head, &idx)) {
2036			IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
2037			ifp->if_flags |= IFF_OACTIVE;
2038			break;
2039		}
2040
2041		/*
2042		 * If there's a BPF listener, bounce a copy of this frame
2043		 * to him.
2044		 */
2045		BPF_MTAP(ifp, m_head);
2046
2047		queued++;
2048	}
2049
2050	if (queued == 0)
2051		return;
2052
2053	/* Flush the TX descriptors */
2054
2055	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2056	    sc->rl_ldata.rl_tx_list_map,
2057	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2058
2059	sc->rl_ldata.rl_tx_prodidx = idx;
2060
2061	/*
2062	 * RealTek put the TX poll request register in a different
2063	 * location on the 8169 gigE chip. I don't know why.
2064	 */
2065
2066	if (sc->rl_type == RL_8169)
2067		CSR_WRITE_2(sc, RL_GTXSTART, RL_TXSTART_START);
2068	else
2069		CSR_WRITE_2(sc, RL_TXSTART, RL_TXSTART_START);
2070
2071	/*
2072	 * Use the countdown timer for interrupt moderation.
2073	 * 'TX done' interrupts are disabled. Instead, we reset the
2074	 * countdown timer, which will begin counting until it hits
2075	 * the value in the TIMERINT register, and then trigger an
2076	 * interrupt. Each time we write to the TIMERCNT register,
2077	 * the timer count is reset to 0.
2078	 */
2079	CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2080
2081	/*
2082	 * Set a timeout in case the chip goes out to lunch.
2083	 */
2084	ifp->if_timer = 5;
2085}
2086
2087static void
2088re_init(xsc)
2089	void			*xsc;
2090{
2091	struct rl_softc		*sc = xsc;
2092
2093	RL_LOCK(sc);
2094	re_init_locked(sc);
2095	RL_UNLOCK(sc);
2096}
2097
2098static void
2099re_init_locked(sc)
2100	struct rl_softc		*sc;
2101{
2102	struct ifnet		*ifp = sc->rl_ifp;
2103	struct mii_data		*mii;
2104	u_int32_t		rxcfg = 0;
2105
2106	RL_LOCK_ASSERT(sc);
2107
2108	mii = device_get_softc(sc->rl_miibus);
2109
2110	/*
2111	 * Cancel pending I/O and free all RX/TX buffers.
2112	 */
2113	re_stop(sc);
2114
2115	/*
2116	 * Enable C+ RX and TX mode, as well as VLAN stripping and
2117	 * RX checksum offload. We must configure the C+ register
2118	 * before all others.
2119	 */
2120	CSR_WRITE_2(sc, RL_CPLUS_CMD, RL_CPLUSCMD_RXENB|
2121	    RL_CPLUSCMD_TXENB|RL_CPLUSCMD_PCI_MRW|
2122	    RL_CPLUSCMD_VLANSTRIP|
2123	    (ifp->if_capenable & IFCAP_RXCSUM ?
2124	    RL_CPLUSCMD_RXCSUM_ENB : 0));
2125
2126	/*
2127	 * Init our MAC address.  Even though the chipset
2128	 * documentation doesn't mention it, we need to enter "Config
2129	 * register write enable" mode to modify the ID registers.
2130	 */
2131	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG);
2132	CSR_WRITE_STREAM_4(sc, RL_IDR0,
2133	    *(u_int32_t *)(&IFP2ENADDR(sc->rl_ifp)[0]));
2134	CSR_WRITE_STREAM_4(sc, RL_IDR4,
2135	    *(u_int32_t *)(&IFP2ENADDR(sc->rl_ifp)[4]));
2136	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
2137
2138	/*
2139	 * For C+ mode, initialize the RX descriptors and mbufs.
2140	 */
2141	re_rx_list_init(sc);
2142	re_tx_list_init(sc);
2143
2144	/*
2145	 * Enable transmit and receive.
2146	 */
2147	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
2148
2149	/*
2150	 * Set the initial TX and RX configuration.
2151	 */
2152	if (sc->rl_testmode) {
2153		if (sc->rl_type == RL_8169)
2154			CSR_WRITE_4(sc, RL_TXCFG,
2155			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON);
2156		else
2157			CSR_WRITE_4(sc, RL_TXCFG,
2158			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON_CPLUS);
2159	} else
2160		CSR_WRITE_4(sc, RL_TXCFG, RL_TXCFG_CONFIG);
2161	CSR_WRITE_4(sc, RL_RXCFG, RL_RXCFG_CONFIG);
2162
2163	/* Set the individual bit to receive frames for this host only. */
2164	rxcfg = CSR_READ_4(sc, RL_RXCFG);
2165	rxcfg |= RL_RXCFG_RX_INDIV;
2166
2167	/* If we want promiscuous mode, set the allframes bit. */
2168	if (ifp->if_flags & IFF_PROMISC)
2169		rxcfg |= RL_RXCFG_RX_ALLPHYS;
2170	else
2171		rxcfg &= ~RL_RXCFG_RX_ALLPHYS;
2172	CSR_WRITE_4(sc, RL_RXCFG, rxcfg);
2173
2174	/*
2175	 * Set capture broadcast bit to capture broadcast frames.
2176	 */
2177	if (ifp->if_flags & IFF_BROADCAST)
2178		rxcfg |= RL_RXCFG_RX_BROAD;
2179	else
2180		rxcfg &= ~RL_RXCFG_RX_BROAD;
2181	CSR_WRITE_4(sc, RL_RXCFG, rxcfg);
2182
2183	/*
2184	 * Program the multicast filter, if necessary.
2185	 */
2186	re_setmulti(sc);
2187
2188#ifdef DEVICE_POLLING
2189	/*
2190	 * Disable interrupts if we are polling.
2191	 */
2192	if (ifp->if_flags & IFF_POLLING)
2193		CSR_WRITE_2(sc, RL_IMR, 0);
2194	else	/* otherwise ... */
2195#endif /* DEVICE_POLLING */
2196	/*
2197	 * Enable interrupts.
2198	 */
2199	if (sc->rl_testmode)
2200		CSR_WRITE_2(sc, RL_IMR, 0);
2201	else
2202		CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2203
2204	/* Set initial TX threshold */
2205	sc->rl_txthresh = RL_TX_THRESH_INIT;
2206
2207	/* Start RX/TX process. */
2208	CSR_WRITE_4(sc, RL_MISSEDPKT, 0);
2209#ifdef notdef
2210	/* Enable receiver and transmitter. */
2211	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
2212#endif
2213	/*
2214	 * Load the addresses of the RX and TX lists into the chip.
2215	 */
2216
2217	CSR_WRITE_4(sc, RL_RXLIST_ADDR_HI,
2218	    RL_ADDR_HI(sc->rl_ldata.rl_rx_list_addr));
2219	CSR_WRITE_4(sc, RL_RXLIST_ADDR_LO,
2220	    RL_ADDR_LO(sc->rl_ldata.rl_rx_list_addr));
2221
2222	CSR_WRITE_4(sc, RL_TXLIST_ADDR_HI,
2223	    RL_ADDR_HI(sc->rl_ldata.rl_tx_list_addr));
2224	CSR_WRITE_4(sc, RL_TXLIST_ADDR_LO,
2225	    RL_ADDR_LO(sc->rl_ldata.rl_tx_list_addr));
2226
2227	CSR_WRITE_1(sc, RL_EARLY_TX_THRESH, 16);
2228
2229	/*
2230	 * Initialize the timer interrupt register so that
2231	 * a timer interrupt will be generated once the timer
2232	 * reaches a certain number of ticks. The timer is
2233	 * reloaded on each transmit. This gives us TX interrupt
2234	 * moderation, which dramatically improves TX frame rate.
2235	 */
2236	if (sc->rl_type == RL_8169)
2237		CSR_WRITE_4(sc, RL_TIMERINT_8169, 0x800);
2238	else
2239		CSR_WRITE_4(sc, RL_TIMERINT, 0x400);
2240
2241	/*
2242	 * For 8169 gigE NICs, set the max allowed RX packet
2243	 * size so we can receive jumbo frames.
2244	 */
2245	if (sc->rl_type == RL_8169)
2246		CSR_WRITE_2(sc, RL_MAXRXPKTLEN, 16383);
2247
2248	if (sc->rl_testmode)
2249		return;
2250
2251	mii_mediachg(mii);
2252
2253	CSR_WRITE_1(sc, RL_CFG1, RL_CFG1_DRVLOAD|RL_CFG1_FULLDUPLEX);
2254
2255	ifp->if_flags |= IFF_RUNNING;
2256	ifp->if_flags &= ~IFF_OACTIVE;
2257
2258	sc->rl_stat_ch = timeout(re_tick, sc, hz);
2259}
2260
2261/*
2262 * Set media options.
2263 */
2264static int
2265re_ifmedia_upd(ifp)
2266	struct ifnet		*ifp;
2267{
2268	struct rl_softc		*sc;
2269	struct mii_data		*mii;
2270
2271	sc = ifp->if_softc;
2272	mii = device_get_softc(sc->rl_miibus);
2273	mii_mediachg(mii);
2274
2275	return (0);
2276}
2277
2278/*
2279 * Report current media status.
2280 */
2281static void
2282re_ifmedia_sts(ifp, ifmr)
2283	struct ifnet		*ifp;
2284	struct ifmediareq	*ifmr;
2285{
2286	struct rl_softc		*sc;
2287	struct mii_data		*mii;
2288
2289	sc = ifp->if_softc;
2290	mii = device_get_softc(sc->rl_miibus);
2291
2292	mii_pollstat(mii);
2293	ifmr->ifm_active = mii->mii_media_active;
2294	ifmr->ifm_status = mii->mii_media_status;
2295}
2296
2297static int
2298re_ioctl(ifp, command, data)
2299	struct ifnet		*ifp;
2300	u_long			command;
2301	caddr_t			data;
2302{
2303	struct rl_softc		*sc = ifp->if_softc;
2304	struct ifreq		*ifr = (struct ifreq *) data;
2305	struct mii_data		*mii;
2306	int			error = 0;
2307
2308	switch (command) {
2309	case SIOCSIFMTU:
2310		if (ifr->ifr_mtu > RL_JUMBO_MTU)
2311			error = EINVAL;
2312		ifp->if_mtu = ifr->ifr_mtu;
2313		break;
2314	case SIOCSIFFLAGS:
2315		RL_LOCK(sc);
2316		if (ifp->if_flags & IFF_UP)
2317			re_init_locked(sc);
2318		else if (ifp->if_flags & IFF_RUNNING)
2319			re_stop(sc);
2320		RL_UNLOCK(sc);
2321		error = 0;
2322		break;
2323	case SIOCADDMULTI:
2324	case SIOCDELMULTI:
2325		RL_LOCK(sc);
2326		re_setmulti(sc);
2327		RL_UNLOCK(sc);
2328		error = 0;
2329		break;
2330	case SIOCGIFMEDIA:
2331	case SIOCSIFMEDIA:
2332		mii = device_get_softc(sc->rl_miibus);
2333		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, command);
2334		break;
2335	case SIOCSIFCAP:
2336		ifp->if_capenable &= ~(IFCAP_HWCSUM | IFCAP_POLLING);
2337		ifp->if_capenable |=
2338		    ifr->ifr_reqcap & (IFCAP_HWCSUM | IFCAP_POLLING);
2339		if (ifp->if_capenable & IFCAP_TXCSUM)
2340			ifp->if_hwassist = RE_CSUM_FEATURES;
2341		else
2342			ifp->if_hwassist = 0;
2343		if (ifp->if_flags & IFF_RUNNING)
2344			re_init(sc);
2345		break;
2346	default:
2347		error = ether_ioctl(ifp, command, data);
2348		break;
2349	}
2350
2351	return (error);
2352}
2353
2354static void
2355re_watchdog(ifp)
2356	struct ifnet		*ifp;
2357{
2358	struct rl_softc		*sc;
2359
2360	sc = ifp->if_softc;
2361	RL_LOCK(sc);
2362	printf("re%d: watchdog timeout\n", sc->rl_unit);
2363	ifp->if_oerrors++;
2364
2365	re_txeof(sc);
2366	re_rxeof(sc);
2367	re_init_locked(sc);
2368
2369	RL_UNLOCK(sc);
2370}
2371
2372/*
2373 * Stop the adapter and free any mbufs allocated to the
2374 * RX and TX lists.
2375 */
2376static void
2377re_stop(sc)
2378	struct rl_softc		*sc;
2379{
2380	register int		i;
2381	struct ifnet		*ifp;
2382
2383	RL_LOCK_ASSERT(sc);
2384
2385	ifp = sc->rl_ifp;
2386	ifp->if_timer = 0;
2387
2388	untimeout(re_tick, sc, sc->rl_stat_ch);
2389	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
2390#ifdef DEVICE_POLLING
2391	ether_poll_deregister(ifp);
2392#endif /* DEVICE_POLLING */
2393
2394	CSR_WRITE_1(sc, RL_COMMAND, 0x00);
2395	CSR_WRITE_2(sc, RL_IMR, 0x0000);
2396
2397	if (sc->rl_head != NULL) {
2398		m_freem(sc->rl_head);
2399		sc->rl_head = sc->rl_tail = NULL;
2400	}
2401
2402	/* Free the TX list buffers. */
2403
2404	for (i = 0; i < RL_TX_DESC_CNT; i++) {
2405		if (sc->rl_ldata.rl_tx_mbuf[i] != NULL) {
2406			bus_dmamap_unload(sc->rl_ldata.rl_mtag,
2407			    sc->rl_ldata.rl_tx_dmamap[i]);
2408			m_freem(sc->rl_ldata.rl_tx_mbuf[i]);
2409			sc->rl_ldata.rl_tx_mbuf[i] = NULL;
2410		}
2411	}
2412
2413	/* Free the RX list buffers. */
2414
2415	for (i = 0; i < RL_RX_DESC_CNT; i++) {
2416		if (sc->rl_ldata.rl_rx_mbuf[i] != NULL) {
2417			bus_dmamap_unload(sc->rl_ldata.rl_mtag,
2418			    sc->rl_ldata.rl_rx_dmamap[i]);
2419			m_freem(sc->rl_ldata.rl_rx_mbuf[i]);
2420			sc->rl_ldata.rl_rx_mbuf[i] = NULL;
2421		}
2422	}
2423}
2424
2425/*
2426 * Device suspend routine.  Stop the interface and save some PCI
2427 * settings in case the BIOS doesn't restore them properly on
2428 * resume.
2429 */
2430static int
2431re_suspend(dev)
2432	device_t		dev;
2433{
2434	struct rl_softc		*sc;
2435
2436	sc = device_get_softc(dev);
2437
2438	RL_LOCK(sc);
2439	re_stop(sc);
2440	sc->suspended = 1;
2441	RL_UNLOCK(sc);
2442
2443	return (0);
2444}
2445
2446/*
2447 * Device resume routine.  Restore some PCI settings in case the BIOS
2448 * doesn't, re-enable busmastering, and restart the interface if
2449 * appropriate.
2450 */
2451static int
2452re_resume(dev)
2453	device_t		dev;
2454{
2455	struct rl_softc		*sc;
2456	struct ifnet		*ifp;
2457
2458	sc = device_get_softc(dev);
2459
2460	RL_LOCK(sc);
2461
2462	ifp = sc->rl_ifp;
2463
2464	/* reinitialize interface if necessary */
2465	if (ifp->if_flags & IFF_UP)
2466		re_init_locked(sc);
2467
2468	sc->suspended = 0;
2469	RL_UNLOCK(sc);
2470
2471	return (0);
2472}
2473
2474/*
2475 * Stop all chip I/O so that the kernel's probe routines don't
2476 * get confused by errant DMAs when rebooting.
2477 */
2478static void
2479re_shutdown(dev)
2480	device_t		dev;
2481{
2482	struct rl_softc		*sc;
2483
2484	sc = device_get_softc(dev);
2485
2486	RL_LOCK(sc);
2487	re_stop(sc);
2488	RL_UNLOCK(sc);
2489}
2490