if_re.c revision 162321
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 162321 2006-09-15 15:16:12Z glebius $");
35
36/*
37 * RealTek 8139C+/8169/8169S/8110S/8168/8111/8101E 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 * seven devices in this family: the RTL8139C+, the RTL8169, the RTL8169S,
48 * RTL8110S, the RTL8168, the RTL8111 and the RTL8101E.
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#include <sys/lock.h>
125#include <sys/mutex.h>
126#include <sys/taskqueue.h>
127
128#include <net/if.h>
129#include <net/if_arp.h>
130#include <net/ethernet.h>
131#include <net/if_dl.h>
132#include <net/if_media.h>
133#include <net/if_types.h>
134#include <net/if_vlan_var.h>
135
136#include <net/bpf.h>
137
138#include <machine/bus.h>
139#include <machine/resource.h>
140#include <sys/bus.h>
141#include <sys/rman.h>
142
143#include <dev/mii/mii.h>
144#include <dev/mii/miivar.h>
145
146#include <dev/pci/pcireg.h>
147#include <dev/pci/pcivar.h>
148
149MODULE_DEPEND(re, pci, 1, 1, 1);
150MODULE_DEPEND(re, ether, 1, 1, 1);
151MODULE_DEPEND(re, miibus, 1, 1, 1);
152
153/* "device miibus" required.  See GENERIC if you get errors here. */
154#include "miibus_if.h"
155
156/*
157 * Default to using PIO access for this driver.
158 */
159#define RE_USEIOSPACE
160
161#include <pci/if_rlreg.h>
162
163#define RE_CSUM_FEATURES    (CSUM_IP | CSUM_TCP | CSUM_UDP)
164
165/*
166 * Various supported device vendors/types and their names.
167 */
168static struct rl_type re_devs[] = {
169	{ DLINK_VENDORID, DLINK_DEVICEID_528T, RL_HWREV_8169S,
170		"D-Link DGE-528(T) Gigabit Ethernet Adapter" },
171	{ RT_VENDORID, RT_DEVICEID_8139, RL_HWREV_8139CPLUS,
172		"RealTek 8139C+ 10/100BaseTX" },
173	{ RT_VENDORID, RT_DEVICEID_8101E, RL_HWREV_8101E,
174		"RealTek 8101E PCIe 10/100baseTX" },
175	{ RT_VENDORID, RT_DEVICEID_8168, RL_HWREV_8168_SPIN1,
176		"RealTek 8168/8111B PCIe Gigabit Ethernet" },
177	{ RT_VENDORID, RT_DEVICEID_8168, RL_HWREV_8168_SPIN2,
178		"RealTek 8168/8111B PCIe Gigabit Ethernet" },
179	{ RT_VENDORID, RT_DEVICEID_8169, RL_HWREV_8169,
180		"RealTek 8169 Gigabit Ethernet" },
181	{ RT_VENDORID, RT_DEVICEID_8169, RL_HWREV_8169S,
182		"RealTek 8169S Single-chip Gigabit Ethernet" },
183	{ RT_VENDORID, RT_DEVICEID_8169, RL_HWREV_8169_8110SB,
184		"RealTek 8169SB/8110SB Single-chip Gigabit Ethernet" },
185	{ RT_VENDORID, RT_DEVICEID_8169SC, RL_HWREV_8169_8110SC,
186		"RealTek 8169SC/8110SC Single-chip Gigabit Ethernet" },
187	{ RT_VENDORID, RT_DEVICEID_8169, RL_HWREV_8110S,
188		"RealTek 8110S Single-chip Gigabit Ethernet" },
189	{ COREGA_VENDORID, COREGA_DEVICEID_CGLAPCIGT, RL_HWREV_8169S,
190		"Corega CG-LAPCIGT (RTL8169S) Gigabit Ethernet" },
191	{ LINKSYS_VENDORID, LINKSYS_DEVICEID_EG1032, RL_HWREV_8169S,
192		"Linksys EG1032 (RTL8169S) Gigabit Ethernet" },
193	{ USR_VENDORID, USR_DEVICEID_997902, RL_HWREV_8169S,
194		"US Robotics 997902 (RTL8169S) Gigabit Ethernet" },
195	{ 0, 0, 0, NULL }
196};
197
198static struct rl_hwrev re_hwrevs[] = {
199	{ RL_HWREV_8139, RL_8139,  "" },
200	{ RL_HWREV_8139A, RL_8139, "A" },
201	{ RL_HWREV_8139AG, RL_8139, "A-G" },
202	{ RL_HWREV_8139B, RL_8139, "B" },
203	{ RL_HWREV_8130, RL_8139, "8130" },
204	{ RL_HWREV_8139C, RL_8139, "C" },
205	{ RL_HWREV_8139D, RL_8139, "8139D/8100B/8100C" },
206	{ RL_HWREV_8139CPLUS, RL_8139CPLUS, "C+"},
207	{ RL_HWREV_8168_SPIN1, RL_8169, "8168"},
208	{ RL_HWREV_8169, RL_8169, "8169"},
209	{ RL_HWREV_8169S, RL_8169, "8169S"},
210	{ RL_HWREV_8110S, RL_8169, "8110S"},
211	{ RL_HWREV_8169_8110SB, RL_8169, "8169SB"},
212	{ RL_HWREV_8169_8110SC, RL_8169, "8169SC"},
213	{ RL_HWREV_8100, RL_8139, "8100"},
214	{ RL_HWREV_8101, RL_8139, "8101"},
215	{ RL_HWREV_8100E, RL_8169, "8100E"},
216	{ RL_HWREV_8101E, RL_8169, "8101E"},
217	{ RL_HWREV_8168_SPIN2, RL_8169, "8168"},
218	{ 0, 0, NULL }
219};
220
221static int re_probe		(device_t);
222static int re_attach		(device_t);
223static int re_detach		(device_t);
224
225static int re_encap		(struct rl_softc *, struct mbuf **, int *);
226
227static void re_dma_map_addr	(void *, bus_dma_segment_t *, int, int);
228static void re_dma_map_desc	(void *, bus_dma_segment_t *, int,
229				    bus_size_t, int);
230static int re_allocmem		(device_t, struct rl_softc *);
231static int re_newbuf		(struct rl_softc *, int, struct mbuf *);
232static int re_rx_list_init	(struct rl_softc *);
233static int re_tx_list_init	(struct rl_softc *);
234#ifdef RE_FIXUP_RX
235static __inline void re_fixup_rx
236				(struct mbuf *);
237#endif
238static int re_rxeof		(struct rl_softc *);
239static void re_txeof		(struct rl_softc *);
240#ifdef DEVICE_POLLING
241static void re_poll		(struct ifnet *, enum poll_cmd, int);
242static void re_poll_locked	(struct ifnet *, enum poll_cmd, int);
243#endif
244static void re_intr		(void *);
245static void re_tick		(void *);
246static void re_tx_task		(void *, int);
247static void re_int_task		(void *, int);
248static void re_start		(struct ifnet *);
249static int re_ioctl		(struct ifnet *, u_long, caddr_t);
250static void re_init		(void *);
251static void re_init_locked	(struct rl_softc *);
252static void re_stop		(struct rl_softc *);
253static void re_watchdog		(struct ifnet *);
254static int re_suspend		(device_t);
255static int re_resume		(device_t);
256static void re_shutdown		(device_t);
257static int re_ifmedia_upd	(struct ifnet *);
258static void re_ifmedia_sts	(struct ifnet *, struct ifmediareq *);
259
260static void re_eeprom_putbyte	(struct rl_softc *, int);
261static void re_eeprom_getword	(struct rl_softc *, int, u_int16_t *);
262static void re_read_eeprom	(struct rl_softc *, caddr_t, int, int);
263static int re_gmii_readreg	(device_t, int, int);
264static int re_gmii_writereg	(device_t, int, int, int);
265
266static int re_miibus_readreg	(device_t, int, int);
267static int re_miibus_writereg	(device_t, int, int, int);
268static void re_miibus_statchg	(device_t);
269
270static void re_setmulti		(struct rl_softc *);
271static void re_reset		(struct rl_softc *);
272
273#ifdef RE_DIAG
274static int re_diag		(struct rl_softc *);
275#endif
276
277#ifdef RE_USEIOSPACE
278#define RL_RES			SYS_RES_IOPORT
279#define RL_RID			RL_PCI_LOIO
280#else
281#define RL_RES			SYS_RES_MEMORY
282#define RL_RID			RL_PCI_LOMEM
283#endif
284
285static device_method_t re_methods[] = {
286	/* Device interface */
287	DEVMETHOD(device_probe,		re_probe),
288	DEVMETHOD(device_attach,	re_attach),
289	DEVMETHOD(device_detach,	re_detach),
290	DEVMETHOD(device_suspend,	re_suspend),
291	DEVMETHOD(device_resume,	re_resume),
292	DEVMETHOD(device_shutdown,	re_shutdown),
293
294	/* bus interface */
295	DEVMETHOD(bus_print_child,	bus_generic_print_child),
296	DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
297
298	/* MII interface */
299	DEVMETHOD(miibus_readreg,	re_miibus_readreg),
300	DEVMETHOD(miibus_writereg,	re_miibus_writereg),
301	DEVMETHOD(miibus_statchg,	re_miibus_statchg),
302
303	{ 0, 0 }
304};
305
306static driver_t re_driver = {
307	"re",
308	re_methods,
309	sizeof(struct rl_softc)
310};
311
312static devclass_t re_devclass;
313
314DRIVER_MODULE(re, pci, re_driver, re_devclass, 0, 0);
315DRIVER_MODULE(re, cardbus, re_driver, re_devclass, 0, 0);
316DRIVER_MODULE(miibus, re, miibus_driver, miibus_devclass, 0, 0);
317
318#define EE_SET(x)					\
319	CSR_WRITE_1(sc, RL_EECMD,			\
320		CSR_READ_1(sc, RL_EECMD) | x)
321
322#define EE_CLR(x)					\
323	CSR_WRITE_1(sc, RL_EECMD,			\
324		CSR_READ_1(sc, RL_EECMD) & ~x)
325
326/*
327 * Send a read command and address to the EEPROM, check for ACK.
328 */
329static void
330re_eeprom_putbyte(sc, addr)
331	struct rl_softc		*sc;
332	int			addr;
333{
334	register int		d, i;
335
336	d = addr | (RL_9346_READ << sc->rl_eewidth);
337
338	/*
339	 * Feed in each bit and strobe the clock.
340	 */
341
342	for (i = 1 << (sc->rl_eewidth + 3); i; i >>= 1) {
343		if (d & i) {
344			EE_SET(RL_EE_DATAIN);
345		} else {
346			EE_CLR(RL_EE_DATAIN);
347		}
348		DELAY(100);
349		EE_SET(RL_EE_CLK);
350		DELAY(150);
351		EE_CLR(RL_EE_CLK);
352		DELAY(100);
353	}
354
355	return;
356}
357
358/*
359 * Read a word of data stored in the EEPROM at address 'addr.'
360 */
361static void
362re_eeprom_getword(sc, addr, dest)
363	struct rl_softc		*sc;
364	int			addr;
365	u_int16_t		*dest;
366{
367	register int		i;
368	u_int16_t		word = 0;
369
370	/*
371	 * Send address of word we want to read.
372	 */
373	re_eeprom_putbyte(sc, addr);
374
375	/*
376	 * Start reading bits from EEPROM.
377	 */
378	for (i = 0x8000; i; i >>= 1) {
379		EE_SET(RL_EE_CLK);
380		DELAY(100);
381		if (CSR_READ_1(sc, RL_EECMD) & RL_EE_DATAOUT)
382			word |= i;
383		EE_CLR(RL_EE_CLK);
384		DELAY(100);
385	}
386
387	*dest = word;
388
389	return;
390}
391
392/*
393 * Read a sequence of words from the EEPROM.
394 */
395static void
396re_read_eeprom(sc, dest, off, cnt)
397	struct rl_softc		*sc;
398	caddr_t			dest;
399	int			off;
400	int			cnt;
401{
402	int			i;
403	u_int16_t		word = 0, *ptr;
404
405	CSR_SETBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM);
406
407        DELAY(100);
408
409	for (i = 0; i < cnt; i++) {
410		CSR_SETBIT_1(sc, RL_EECMD, RL_EE_SEL);
411		re_eeprom_getword(sc, off + i, &word);
412		CSR_CLRBIT_1(sc, RL_EECMD, RL_EE_SEL);
413		ptr = (u_int16_t *)(dest + (i * 2));
414                *ptr = word;
415	}
416
417	CSR_CLRBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM);
418
419	return;
420}
421
422static int
423re_gmii_readreg(dev, phy, reg)
424	device_t		dev;
425	int			phy, reg;
426{
427	struct rl_softc		*sc;
428	u_int32_t		rval;
429	int			i;
430
431	if (phy != 1)
432		return (0);
433
434	sc = device_get_softc(dev);
435
436	/* Let the rgephy driver read the GMEDIASTAT register */
437
438	if (reg == RL_GMEDIASTAT) {
439		rval = CSR_READ_1(sc, RL_GMEDIASTAT);
440		return (rval);
441	}
442
443	CSR_WRITE_4(sc, RL_PHYAR, reg << 16);
444	DELAY(1000);
445
446	for (i = 0; i < RL_TIMEOUT; i++) {
447		rval = CSR_READ_4(sc, RL_PHYAR);
448		if (rval & RL_PHYAR_BUSY)
449			break;
450		DELAY(100);
451	}
452
453	if (i == RL_TIMEOUT) {
454		device_printf(sc->rl_dev, "PHY read failed\n");
455		return (0);
456	}
457
458	return (rval & RL_PHYAR_PHYDATA);
459}
460
461static int
462re_gmii_writereg(dev, phy, reg, data)
463	device_t		dev;
464	int			phy, reg, data;
465{
466	struct rl_softc		*sc;
467	u_int32_t		rval;
468	int			i;
469
470	sc = device_get_softc(dev);
471
472	CSR_WRITE_4(sc, RL_PHYAR, (reg << 16) |
473	    (data & RL_PHYAR_PHYDATA) | RL_PHYAR_BUSY);
474	DELAY(1000);
475
476	for (i = 0; i < RL_TIMEOUT; i++) {
477		rval = CSR_READ_4(sc, RL_PHYAR);
478		if (!(rval & RL_PHYAR_BUSY))
479			break;
480		DELAY(100);
481	}
482
483	if (i == RL_TIMEOUT) {
484		device_printf(sc->rl_dev, "PHY write failed\n");
485		return (0);
486	}
487
488	return (0);
489}
490
491static int
492re_miibus_readreg(dev, phy, reg)
493	device_t		dev;
494	int			phy, reg;
495{
496	struct rl_softc		*sc;
497	u_int16_t		rval = 0;
498	u_int16_t		re8139_reg = 0;
499
500	sc = device_get_softc(dev);
501
502	if (sc->rl_type == RL_8169) {
503		rval = re_gmii_readreg(dev, phy, reg);
504		return (rval);
505	}
506
507	/* Pretend the internal PHY is only at address 0 */
508	if (phy) {
509		return (0);
510	}
511	switch (reg) {
512	case MII_BMCR:
513		re8139_reg = RL_BMCR;
514		break;
515	case MII_BMSR:
516		re8139_reg = RL_BMSR;
517		break;
518	case MII_ANAR:
519		re8139_reg = RL_ANAR;
520		break;
521	case MII_ANER:
522		re8139_reg = RL_ANER;
523		break;
524	case MII_ANLPAR:
525		re8139_reg = RL_LPAR;
526		break;
527	case MII_PHYIDR1:
528	case MII_PHYIDR2:
529		return (0);
530	/*
531	 * Allow the rlphy driver to read the media status
532	 * register. If we have a link partner which does not
533	 * support NWAY, this is the register which will tell
534	 * us the results of parallel detection.
535	 */
536	case RL_MEDIASTAT:
537		rval = CSR_READ_1(sc, RL_MEDIASTAT);
538		return (rval);
539	default:
540		device_printf(sc->rl_dev, "bad phy register\n");
541		return (0);
542	}
543	rval = CSR_READ_2(sc, re8139_reg);
544	if (sc->rl_type == RL_8139CPLUS && re8139_reg == RL_BMCR) {
545		/* 8139C+ has different bit layout. */
546		rval &= ~(BMCR_LOOP | BMCR_ISO);
547	}
548	return (rval);
549}
550
551static int
552re_miibus_writereg(dev, phy, reg, data)
553	device_t		dev;
554	int			phy, reg, data;
555{
556	struct rl_softc		*sc;
557	u_int16_t		re8139_reg = 0;
558	int			rval = 0;
559
560	sc = device_get_softc(dev);
561
562	if (sc->rl_type == RL_8169) {
563		rval = re_gmii_writereg(dev, phy, reg, data);
564		return (rval);
565	}
566
567	/* Pretend the internal PHY is only at address 0 */
568	if (phy)
569		return (0);
570
571	switch (reg) {
572	case MII_BMCR:
573		re8139_reg = RL_BMCR;
574		if (sc->rl_type == RL_8139CPLUS) {
575			/* 8139C+ has different bit layout. */
576			data &= ~(BMCR_LOOP | BMCR_ISO);
577		}
578		break;
579	case MII_BMSR:
580		re8139_reg = RL_BMSR;
581		break;
582	case MII_ANAR:
583		re8139_reg = RL_ANAR;
584		break;
585	case MII_ANER:
586		re8139_reg = RL_ANER;
587		break;
588	case MII_ANLPAR:
589		re8139_reg = RL_LPAR;
590		break;
591	case MII_PHYIDR1:
592	case MII_PHYIDR2:
593		return (0);
594		break;
595	default:
596		device_printf(sc->rl_dev, "bad phy register\n");
597		return (0);
598	}
599	CSR_WRITE_2(sc, re8139_reg, data);
600	return (0);
601}
602
603static void
604re_miibus_statchg(dev)
605	device_t		dev;
606{
607
608}
609
610/*
611 * Program the 64-bit multicast hash filter.
612 */
613static void
614re_setmulti(sc)
615	struct rl_softc		*sc;
616{
617	struct ifnet		*ifp;
618	int			h = 0;
619	u_int32_t		hashes[2] = { 0, 0 };
620	struct ifmultiaddr	*ifma;
621	u_int32_t		rxfilt;
622	int			mcnt = 0;
623
624	RL_LOCK_ASSERT(sc);
625
626	ifp = sc->rl_ifp;
627
628	rxfilt = CSR_READ_4(sc, RL_RXCFG);
629
630	if (ifp->if_flags & IFF_ALLMULTI || ifp->if_flags & IFF_PROMISC) {
631		rxfilt |= RL_RXCFG_RX_MULTI;
632		CSR_WRITE_4(sc, RL_RXCFG, rxfilt);
633		CSR_WRITE_4(sc, RL_MAR0, 0xFFFFFFFF);
634		CSR_WRITE_4(sc, RL_MAR4, 0xFFFFFFFF);
635		return;
636	}
637
638	/* first, zot all the existing hash bits */
639	CSR_WRITE_4(sc, RL_MAR0, 0);
640	CSR_WRITE_4(sc, RL_MAR4, 0);
641
642	/* now program new ones */
643	IF_ADDR_LOCK(ifp);
644	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
645		if (ifma->ifma_addr->sa_family != AF_LINK)
646			continue;
647		h = ether_crc32_be(LLADDR((struct sockaddr_dl *)
648		    ifma->ifma_addr), ETHER_ADDR_LEN) >> 26;
649		if (h < 32)
650			hashes[0] |= (1 << h);
651		else
652			hashes[1] |= (1 << (h - 32));
653		mcnt++;
654	}
655	IF_ADDR_UNLOCK(ifp);
656
657	if (mcnt)
658		rxfilt |= RL_RXCFG_RX_MULTI;
659	else
660		rxfilt &= ~RL_RXCFG_RX_MULTI;
661
662	CSR_WRITE_4(sc, RL_RXCFG, rxfilt);
663	CSR_WRITE_4(sc, RL_MAR0, hashes[0]);
664	CSR_WRITE_4(sc, RL_MAR4, hashes[1]);
665}
666
667static void
668re_reset(sc)
669	struct rl_softc		*sc;
670{
671	register int		i;
672
673	RL_LOCK_ASSERT(sc);
674
675	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RESET);
676
677	for (i = 0; i < RL_TIMEOUT; i++) {
678		DELAY(10);
679		if (!(CSR_READ_1(sc, RL_COMMAND) & RL_CMD_RESET))
680			break;
681	}
682	if (i == RL_TIMEOUT)
683		device_printf(sc->rl_dev, "reset never completed!\n");
684
685	CSR_WRITE_1(sc, 0x82, 1);
686}
687
688#ifdef RE_DIAG
689
690/*
691 * The following routine is designed to test for a defect on some
692 * 32-bit 8169 cards. Some of these NICs have the REQ64# and ACK64#
693 * lines connected to the bus, however for a 32-bit only card, they
694 * should be pulled high. The result of this defect is that the
695 * NIC will not work right if you plug it into a 64-bit slot: DMA
696 * operations will be done with 64-bit transfers, which will fail
697 * because the 64-bit data lines aren't connected.
698 *
699 * There's no way to work around this (short of talking a soldering
700 * iron to the board), however we can detect it. The method we use
701 * here is to put the NIC into digital loopback mode, set the receiver
702 * to promiscuous mode, and then try to send a frame. We then compare
703 * the frame data we sent to what was received. If the data matches,
704 * then the NIC is working correctly, otherwise we know the user has
705 * a defective NIC which has been mistakenly plugged into a 64-bit PCI
706 * slot. In the latter case, there's no way the NIC can work correctly,
707 * so we print out a message on the console and abort the device attach.
708 */
709
710static int
711re_diag(sc)
712	struct rl_softc		*sc;
713{
714	struct ifnet		*ifp = sc->rl_ifp;
715	struct mbuf		*m0;
716	struct ether_header	*eh;
717	struct rl_desc		*cur_rx;
718	u_int16_t		status;
719	u_int32_t		rxstat;
720	int			total_len, i, error = 0, phyaddr;
721	u_int8_t		dst[] = { 0x00, 'h', 'e', 'l', 'l', 'o' };
722	u_int8_t		src[] = { 0x00, 'w', 'o', 'r', 'l', 'd' };
723
724	/* Allocate a single mbuf */
725	MGETHDR(m0, M_DONTWAIT, MT_DATA);
726	if (m0 == NULL)
727		return (ENOBUFS);
728
729	RL_LOCK(sc);
730
731	/*
732	 * Initialize the NIC in test mode. This sets the chip up
733	 * so that it can send and receive frames, but performs the
734	 * following special functions:
735	 * - Puts receiver in promiscuous mode
736	 * - Enables digital loopback mode
737	 * - Leaves interrupts turned off
738	 */
739
740	ifp->if_flags |= IFF_PROMISC;
741	sc->rl_testmode = 1;
742	re_reset(sc);
743	re_init_locked(sc);
744	sc->rl_link = 1;
745	if (sc->rl_type == RL_8169)
746		phyaddr = 1;
747	else
748		phyaddr = 0;
749
750	re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_RESET);
751	for (i = 0; i < RL_TIMEOUT; i++) {
752		status = re_miibus_readreg(sc->rl_dev, phyaddr, MII_BMCR);
753		if (!(status & BMCR_RESET))
754			break;
755	}
756
757	re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_LOOP);
758	CSR_WRITE_2(sc, RL_ISR, RL_INTRS);
759
760	DELAY(100000);
761
762	/* Put some data in the mbuf */
763
764	eh = mtod(m0, struct ether_header *);
765	bcopy ((char *)&dst, eh->ether_dhost, ETHER_ADDR_LEN);
766	bcopy ((char *)&src, eh->ether_shost, ETHER_ADDR_LEN);
767	eh->ether_type = htons(ETHERTYPE_IP);
768	m0->m_pkthdr.len = m0->m_len = ETHER_MIN_LEN - ETHER_CRC_LEN;
769
770	/*
771	 * Queue the packet, start transmission.
772	 * Note: IF_HANDOFF() ultimately calls re_start() for us.
773	 */
774
775	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
776	RL_UNLOCK(sc);
777	/* XXX: re_diag must not be called when in ALTQ mode */
778	IF_HANDOFF(&ifp->if_snd, m0, ifp);
779	RL_LOCK(sc);
780	m0 = NULL;
781
782	/* Wait for it to propagate through the chip */
783
784	DELAY(100000);
785	for (i = 0; i < RL_TIMEOUT; i++) {
786		status = CSR_READ_2(sc, RL_ISR);
787		CSR_WRITE_2(sc, RL_ISR, status);
788		if ((status & (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK)) ==
789		    (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK))
790			break;
791		DELAY(10);
792	}
793
794	if (i == RL_TIMEOUT) {
795		device_printf(sc->rl_dev,
796		    "diagnostic failed, failed to receive packet in"
797		    " loopback mode\n");
798		error = EIO;
799		goto done;
800	}
801
802	/*
803	 * The packet should have been dumped into the first
804	 * entry in the RX DMA ring. Grab it from there.
805	 */
806
807	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
808	    sc->rl_ldata.rl_rx_list_map,
809	    BUS_DMASYNC_POSTREAD);
810	bus_dmamap_sync(sc->rl_ldata.rl_mtag,
811	    sc->rl_ldata.rl_rx_dmamap[0],
812	    BUS_DMASYNC_POSTWRITE);
813	bus_dmamap_unload(sc->rl_ldata.rl_mtag,
814	    sc->rl_ldata.rl_rx_dmamap[0]);
815
816	m0 = sc->rl_ldata.rl_rx_mbuf[0];
817	sc->rl_ldata.rl_rx_mbuf[0] = NULL;
818	eh = mtod(m0, struct ether_header *);
819
820	cur_rx = &sc->rl_ldata.rl_rx_list[0];
821	total_len = RL_RXBYTES(cur_rx);
822	rxstat = le32toh(cur_rx->rl_cmdstat);
823
824	if (total_len != ETHER_MIN_LEN) {
825		device_printf(sc->rl_dev,
826		    "diagnostic failed, received short packet\n");
827		error = EIO;
828		goto done;
829	}
830
831	/* Test that the received packet data matches what we sent. */
832
833	if (bcmp((char *)&eh->ether_dhost, (char *)&dst, ETHER_ADDR_LEN) ||
834	    bcmp((char *)&eh->ether_shost, (char *)&src, ETHER_ADDR_LEN) ||
835	    ntohs(eh->ether_type) != ETHERTYPE_IP) {
836		device_printf(sc->rl_dev, "WARNING, DMA FAILURE!\n");
837		device_printf(sc->rl_dev, "expected TX data: %6D/%6D/0x%x\n",
838		    dst, ":", src, ":", ETHERTYPE_IP);
839		device_printf(sc->rl_dev, "received RX data: %6D/%6D/0x%x\n",
840		    eh->ether_dhost, ":",  eh->ether_shost, ":",
841		    ntohs(eh->ether_type));
842		device_printf(sc->rl_dev, "You may have a defective 32-bit "
843		    "NIC plugged into a 64-bit PCI slot.\n");
844		device_printf(sc->rl_dev, "Please re-install the NIC in a "
845		    "32-bit slot for proper operation.\n");
846		device_printf(sc->rl_dev, "Read the re(4) man page for more "
847		    "details.\n");
848		error = EIO;
849	}
850
851done:
852	/* Turn interface off, release resources */
853
854	sc->rl_testmode = 0;
855	sc->rl_link = 0;
856	ifp->if_flags &= ~IFF_PROMISC;
857	re_stop(sc);
858	if (m0 != NULL)
859		m_freem(m0);
860
861	RL_UNLOCK(sc);
862
863	return (error);
864}
865
866#endif
867
868/*
869 * Probe for a RealTek 8139C+/8169/8110 chip. Check the PCI vendor and device
870 * IDs against our list and return a device name if we find a match.
871 */
872static int
873re_probe(dev)
874	device_t		dev;
875{
876	struct rl_type		*t;
877	struct rl_softc		*sc;
878	int			rid;
879	u_int32_t		hwrev;
880
881	t = re_devs;
882	sc = device_get_softc(dev);
883
884	while (t->rl_name != NULL) {
885		if ((pci_get_vendor(dev) == t->rl_vid) &&
886		    (pci_get_device(dev) == t->rl_did)) {
887			/*
888			 * Only attach to rev. 3 of the Linksys EG1032 adapter.
889			 * Rev. 2 i supported by sk(4).
890			 */
891			if ((t->rl_vid == LINKSYS_VENDORID) &&
892				(t->rl_did == LINKSYS_DEVICEID_EG1032) &&
893				(pci_get_subdevice(dev) !=
894				LINKSYS_SUBDEVICE_EG1032_REV3)) {
895				t++;
896				continue;
897			}
898
899			/*
900			 * Temporarily map the I/O space
901			 * so we can read the chip ID register.
902			 */
903			rid = RL_RID;
904			sc->rl_res = bus_alloc_resource_any(dev, RL_RES, &rid,
905			    RF_ACTIVE);
906			if (sc->rl_res == NULL) {
907				device_printf(dev,
908				    "couldn't map ports/memory\n");
909				return (ENXIO);
910			}
911			sc->rl_btag = rman_get_bustag(sc->rl_res);
912			sc->rl_bhandle = rman_get_bushandle(sc->rl_res);
913			hwrev = CSR_READ_4(sc, RL_TXCFG) & RL_TXCFG_HWREV;
914			bus_release_resource(dev, RL_RES,
915			    RL_RID, sc->rl_res);
916			if (t->rl_basetype == hwrev) {
917				device_set_desc(dev, t->rl_name);
918				return (BUS_PROBE_DEFAULT);
919			}
920		}
921		t++;
922	}
923
924	return (ENXIO);
925}
926
927/*
928 * This routine takes the segment list provided as the result of
929 * a bus_dma_map_load() operation and assigns the addresses/lengths
930 * to RealTek DMA descriptors. This can be called either by the RX
931 * code or the TX code. In the RX case, we'll probably wind up mapping
932 * at most one segment. For the TX case, there could be any number of
933 * segments since TX packets may span multiple mbufs. In either case,
934 * if the number of segments is larger than the rl_maxsegs limit
935 * specified by the caller, we abort the mapping operation. Sadly,
936 * whoever designed the buffer mapping API did not provide a way to
937 * return an error from here, so we have to fake it a bit.
938 */
939
940static void
941re_dma_map_desc(arg, segs, nseg, mapsize, error)
942	void			*arg;
943	bus_dma_segment_t	*segs;
944	int			nseg;
945	bus_size_t		mapsize;
946	int			error;
947{
948	struct rl_dmaload_arg	*ctx;
949	struct rl_desc		*d = NULL;
950	int			i = 0, idx;
951	u_int32_t		cmdstat;
952	int			totlen = 0;
953
954	if (error)
955		return;
956
957	ctx = arg;
958
959	/* Signal error to caller if there's too many segments */
960	if (nseg > ctx->rl_maxsegs) {
961		ctx->rl_maxsegs = 0;
962		return;
963	}
964
965	/*
966	 * Map the segment array into descriptors. Note that we set the
967	 * start-of-frame and end-of-frame markers for either TX or RX, but
968	 * they really only have meaning in the TX case. (In the RX case,
969	 * it's the chip that tells us where packets begin and end.)
970	 * We also keep track of the end of the ring and set the
971	 * end-of-ring bits as needed, and we set the ownership bits
972	 * in all except the very first descriptor. (The caller will
973	 * set this descriptor later when it start transmission or
974	 * reception.)
975	 */
976	idx = ctx->rl_idx;
977	for (;;) {
978		d = &ctx->rl_ring[idx];
979		if (le32toh(d->rl_cmdstat) & RL_RDESC_STAT_OWN) {
980			ctx->rl_maxsegs = 0;
981			return;
982		}
983		cmdstat = segs[i].ds_len;
984		totlen += segs[i].ds_len;
985		d->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[i].ds_addr));
986		d->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[i].ds_addr));
987		if (i == 0)
988			cmdstat |= RL_TDESC_CMD_SOF;
989		else
990			cmdstat |= RL_TDESC_CMD_OWN;
991		if (idx == (RL_RX_DESC_CNT - 1))
992			cmdstat |= RL_TDESC_CMD_EOR;
993		d->rl_cmdstat = htole32(cmdstat | ctx->rl_flags);
994		i++;
995		if (i == nseg)
996			break;
997		RL_DESC_INC(idx);
998	}
999
1000	d->rl_cmdstat |= htole32(RL_TDESC_CMD_EOF);
1001	ctx->rl_maxsegs = nseg;
1002	ctx->rl_idx = idx;
1003}
1004
1005/*
1006 * Map a single buffer address.
1007 */
1008
1009static void
1010re_dma_map_addr(arg, segs, nseg, error)
1011	void			*arg;
1012	bus_dma_segment_t	*segs;
1013	int			nseg;
1014	int			error;
1015{
1016	bus_addr_t		*addr;
1017
1018	if (error)
1019		return;
1020
1021	KASSERT(nseg == 1, ("too many DMA segments, %d should be 1", nseg));
1022	addr = arg;
1023	*addr = segs->ds_addr;
1024}
1025
1026static int
1027re_allocmem(dev, sc)
1028	device_t		dev;
1029	struct rl_softc		*sc;
1030{
1031	int			error;
1032	int			nseg;
1033	int			i;
1034
1035	/*
1036	 * Allocate map for RX mbufs.
1037	 */
1038	nseg = 32;
1039	error = bus_dma_tag_create(sc->rl_parent_tag, ETHER_ALIGN, 0,
1040	    BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1041	    NULL, MCLBYTES * nseg, nseg, MCLBYTES, BUS_DMA_ALLOCNOW,
1042	    NULL, NULL, &sc->rl_ldata.rl_mtag);
1043	if (error) {
1044		device_printf(dev, "could not allocate dma tag\n");
1045		return (ENOMEM);
1046	}
1047
1048	/*
1049	 * Allocate map for TX descriptor list.
1050	 */
1051	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
1052	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1053	    NULL, RL_TX_LIST_SZ, 1, RL_TX_LIST_SZ, BUS_DMA_ALLOCNOW,
1054	    NULL, NULL, &sc->rl_ldata.rl_tx_list_tag);
1055	if (error) {
1056		device_printf(dev, "could not allocate dma tag\n");
1057		return (ENOMEM);
1058	}
1059
1060	/* Allocate DMA'able memory for the TX ring */
1061
1062	error = bus_dmamem_alloc(sc->rl_ldata.rl_tx_list_tag,
1063	    (void **)&sc->rl_ldata.rl_tx_list, BUS_DMA_NOWAIT | BUS_DMA_ZERO,
1064	    &sc->rl_ldata.rl_tx_list_map);
1065	if (error)
1066		return (ENOMEM);
1067
1068	/* Load the map for the TX ring. */
1069
1070	error = bus_dmamap_load(sc->rl_ldata.rl_tx_list_tag,
1071	     sc->rl_ldata.rl_tx_list_map, sc->rl_ldata.rl_tx_list,
1072	     RL_TX_LIST_SZ, re_dma_map_addr,
1073	     &sc->rl_ldata.rl_tx_list_addr, BUS_DMA_NOWAIT);
1074
1075	/* Create DMA maps for TX buffers */
1076
1077	for (i = 0; i < RL_TX_DESC_CNT; i++) {
1078		error = bus_dmamap_create(sc->rl_ldata.rl_mtag, 0,
1079			    &sc->rl_ldata.rl_tx_dmamap[i]);
1080		if (error) {
1081			device_printf(dev, "can't create DMA map for TX\n");
1082			return (ENOMEM);
1083		}
1084	}
1085
1086	/*
1087	 * Allocate map for RX descriptor list.
1088	 */
1089	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
1090	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1091	    NULL, RL_RX_LIST_SZ, 1, RL_RX_LIST_SZ, BUS_DMA_ALLOCNOW,
1092	    NULL, NULL, &sc->rl_ldata.rl_rx_list_tag);
1093	if (error) {
1094		device_printf(dev, "could not allocate dma tag\n");
1095		return (ENOMEM);
1096	}
1097
1098	/* Allocate DMA'able memory for the RX ring */
1099
1100	error = bus_dmamem_alloc(sc->rl_ldata.rl_rx_list_tag,
1101	    (void **)&sc->rl_ldata.rl_rx_list, BUS_DMA_NOWAIT | BUS_DMA_ZERO,
1102	    &sc->rl_ldata.rl_rx_list_map);
1103	if (error)
1104		return (ENOMEM);
1105
1106	/* Load the map for the RX ring. */
1107
1108	error = bus_dmamap_load(sc->rl_ldata.rl_rx_list_tag,
1109	     sc->rl_ldata.rl_rx_list_map, sc->rl_ldata.rl_rx_list,
1110	     RL_RX_LIST_SZ, re_dma_map_addr,
1111	     &sc->rl_ldata.rl_rx_list_addr, BUS_DMA_NOWAIT);
1112
1113	/* Create DMA maps for RX buffers */
1114
1115	for (i = 0; i < RL_RX_DESC_CNT; i++) {
1116		error = bus_dmamap_create(sc->rl_ldata.rl_mtag, 0,
1117			    &sc->rl_ldata.rl_rx_dmamap[i]);
1118		if (error) {
1119			device_printf(dev, "can't create DMA map for RX\n");
1120			return (ENOMEM);
1121		}
1122	}
1123
1124	return (0);
1125}
1126
1127/*
1128 * Attach the interface. Allocate softc structures, do ifmedia
1129 * setup and ethernet/BPF attach.
1130 */
1131static int
1132re_attach(dev)
1133	device_t		dev;
1134{
1135	u_char			eaddr[ETHER_ADDR_LEN];
1136	u_int16_t		as[ETHER_ADDR_LEN / 2];
1137	struct rl_softc		*sc;
1138	struct ifnet		*ifp;
1139	struct rl_hwrev		*hw_rev;
1140	int			hwrev;
1141	u_int16_t		re_did = 0;
1142	int			error = 0, rid, i;
1143
1144	sc = device_get_softc(dev);
1145	sc->rl_dev = dev;
1146
1147	mtx_init(&sc->rl_mtx, device_get_nameunit(dev), MTX_NETWORK_LOCK,
1148	    MTX_DEF);
1149	callout_init_mtx(&sc->rl_stat_callout, &sc->rl_mtx, 0);
1150
1151	/*
1152	 * Map control/status registers.
1153	 */
1154	pci_enable_busmaster(dev);
1155
1156	rid = RL_RID;
1157	sc->rl_res = bus_alloc_resource_any(dev, RL_RES, &rid,
1158	    RF_ACTIVE);
1159
1160	if (sc->rl_res == NULL) {
1161		device_printf(dev, "couldn't map ports/memory\n");
1162		error = ENXIO;
1163		goto fail;
1164	}
1165
1166	sc->rl_btag = rman_get_bustag(sc->rl_res);
1167	sc->rl_bhandle = rman_get_bushandle(sc->rl_res);
1168
1169	/* Allocate interrupt */
1170	rid = 0;
1171	sc->rl_irq = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
1172	    RF_SHAREABLE | RF_ACTIVE);
1173
1174	if (sc->rl_irq == NULL) {
1175		device_printf(dev, "couldn't map interrupt\n");
1176		error = ENXIO;
1177		goto fail;
1178	}
1179
1180	/* Reset the adapter. */
1181	RL_LOCK(sc);
1182	re_reset(sc);
1183	RL_UNLOCK(sc);
1184
1185	hw_rev = re_hwrevs;
1186	hwrev = CSR_READ_4(sc, RL_TXCFG) & RL_TXCFG_HWREV;
1187	while (hw_rev->rl_desc != NULL) {
1188		if (hw_rev->rl_rev == hwrev) {
1189			sc->rl_type = hw_rev->rl_type;
1190			break;
1191		}
1192		hw_rev++;
1193	}
1194
1195	sc->rl_eewidth = 6;
1196	re_read_eeprom(sc, (caddr_t)&re_did, 0, 1);
1197	if (re_did != 0x8129)
1198	        sc->rl_eewidth = 8;
1199
1200	/*
1201	 * Get station address from the EEPROM.
1202	 */
1203	re_read_eeprom(sc, (caddr_t)as, RL_EE_EADDR, 3);
1204	for (i = 0; i < ETHER_ADDR_LEN / 2; i++)
1205		as[i] = le16toh(as[i]);
1206	bcopy(as, eaddr, sizeof(eaddr));
1207
1208	if (sc->rl_type == RL_8169) {
1209		/* Set RX length mask */
1210		sc->rl_rxlenmask = RL_RDESC_STAT_GFRAGLEN;
1211		sc->rl_txstart = RL_GTXSTART;
1212	} else {
1213		/* Set RX length mask */
1214		sc->rl_rxlenmask = RL_RDESC_STAT_FRAGLEN;
1215		sc->rl_txstart = RL_TXSTART;
1216	}
1217
1218	/*
1219	 * Allocate the parent bus DMA tag appropriate for PCI.
1220	 */
1221#define RL_NSEG_NEW 32
1222	error = bus_dma_tag_create(NULL,	/* parent */
1223			1, 0,			/* alignment, boundary */
1224			BUS_SPACE_MAXADDR_32BIT,/* lowaddr */
1225			BUS_SPACE_MAXADDR,	/* highaddr */
1226			NULL, NULL,		/* filter, filterarg */
1227			MAXBSIZE, RL_NSEG_NEW,	/* maxsize, nsegments */
1228			BUS_SPACE_MAXSIZE_32BIT,/* maxsegsize */
1229			BUS_DMA_ALLOCNOW,	/* flags */
1230			NULL, NULL,		/* lockfunc, lockarg */
1231			&sc->rl_parent_tag);
1232	if (error)
1233		goto fail;
1234
1235	error = re_allocmem(dev, sc);
1236
1237	if (error)
1238		goto fail;
1239
1240	ifp = sc->rl_ifp = if_alloc(IFT_ETHER);
1241	if (ifp == NULL) {
1242		device_printf(dev, "can not if_alloc()\n");
1243		error = ENOSPC;
1244		goto fail;
1245	}
1246
1247	/* Do MII setup */
1248	if (mii_phy_probe(dev, &sc->rl_miibus,
1249	    re_ifmedia_upd, re_ifmedia_sts)) {
1250		device_printf(dev, "MII without any phy!\n");
1251		error = ENXIO;
1252		goto fail;
1253	}
1254
1255	ifp->if_softc = sc;
1256	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1257	ifp->if_mtu = ETHERMTU;
1258	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1259	ifp->if_ioctl = re_ioctl;
1260	ifp->if_capabilities = IFCAP_VLAN_MTU;
1261	ifp->if_start = re_start;
1262	ifp->if_hwassist = RE_CSUM_FEATURES;
1263	ifp->if_capabilities |= IFCAP_HWCSUM|IFCAP_VLAN_HWTAGGING;
1264	ifp->if_capenable = ifp->if_capabilities;
1265#ifdef DEVICE_POLLING
1266	ifp->if_capabilities |= IFCAP_POLLING;
1267#endif
1268	ifp->if_watchdog = re_watchdog;
1269	ifp->if_init = re_init;
1270	IFQ_SET_MAXLEN(&ifp->if_snd, RL_IFQ_MAXLEN);
1271	ifp->if_snd.ifq_drv_maxlen = RL_IFQ_MAXLEN;
1272	IFQ_SET_READY(&ifp->if_snd);
1273
1274	TASK_INIT(&sc->rl_txtask, 1, re_tx_task, ifp);
1275	TASK_INIT(&sc->rl_inttask, 0, re_int_task, sc);
1276
1277	/*
1278	 * Call MI attach routine.
1279	 */
1280	ether_ifattach(ifp, eaddr);
1281
1282#ifdef RE_DIAG
1283	/*
1284	 * Perform hardware diagnostic on the original RTL8169.
1285	 * Some 32-bit cards were incorrectly wired and would
1286	 * malfunction if plugged into a 64-bit slot.
1287	 */
1288
1289	if (hwrev == RL_HWREV_8169) {
1290		error = re_diag(sc);
1291		if (error) {
1292			device_printf(dev,
1293		    	"attach aborted due to hardware diag failure\n");
1294			ether_ifdetach(ifp);
1295			goto fail;
1296		}
1297	}
1298#endif
1299
1300	/* Hook interrupt last to avoid having to lock softc */
1301	error = bus_setup_intr(dev, sc->rl_irq, INTR_TYPE_NET | INTR_MPSAFE |
1302	    INTR_FAST, re_intr, sc, &sc->rl_intrhand);
1303	if (error) {
1304		device_printf(dev, "couldn't set up irq\n");
1305		ether_ifdetach(ifp);
1306	}
1307
1308fail:
1309
1310	if (error)
1311		re_detach(dev);
1312
1313	return (error);
1314}
1315
1316/*
1317 * Shutdown hardware and free up resources. This can be called any
1318 * time after the mutex has been initialized. It is called in both
1319 * the error case in attach and the normal detach case so it needs
1320 * to be careful about only freeing resources that have actually been
1321 * allocated.
1322 */
1323static int
1324re_detach(dev)
1325	device_t		dev;
1326{
1327	struct rl_softc		*sc;
1328	struct ifnet		*ifp;
1329	int			i;
1330
1331	sc = device_get_softc(dev);
1332	ifp = sc->rl_ifp;
1333	KASSERT(mtx_initialized(&sc->rl_mtx), ("re mutex not initialized"));
1334
1335#ifdef DEVICE_POLLING
1336	if (ifp->if_capenable & IFCAP_POLLING)
1337		ether_poll_deregister(ifp);
1338#endif
1339	/* These should only be active if attach succeeded */
1340	if (device_is_attached(dev)) {
1341		RL_LOCK(sc);
1342#if 0
1343		sc->suspended = 1;
1344#endif
1345		re_stop(sc);
1346		RL_UNLOCK(sc);
1347		callout_drain(&sc->rl_stat_callout);
1348		/*
1349		 * Force off the IFF_UP flag here, in case someone
1350		 * still had a BPF descriptor attached to this
1351		 * interface. If they do, ether_ifdetach() will cause
1352		 * the BPF code to try and clear the promisc mode
1353		 * flag, which will bubble down to re_ioctl(),
1354		 * which will try to call re_init() again. This will
1355		 * turn the NIC back on and restart the MII ticker,
1356		 * which will panic the system when the kernel tries
1357		 * to invoke the re_tick() function that isn't there
1358		 * anymore.
1359		 */
1360		ifp->if_flags &= ~IFF_UP;
1361		ether_ifdetach(ifp);
1362	}
1363	if (sc->rl_miibus)
1364		device_delete_child(dev, sc->rl_miibus);
1365	bus_generic_detach(dev);
1366
1367	/*
1368	 * The rest is resource deallocation, so we should already be
1369	 * stopped here.
1370	 */
1371
1372	if (sc->rl_intrhand)
1373		bus_teardown_intr(dev, sc->rl_irq, sc->rl_intrhand);
1374	if (ifp != NULL)
1375		if_free(ifp);
1376	if (sc->rl_irq)
1377		bus_release_resource(dev, SYS_RES_IRQ, 0, sc->rl_irq);
1378	if (sc->rl_res)
1379		bus_release_resource(dev, RL_RES, RL_RID, sc->rl_res);
1380
1381	/* Yield the CPU long enough for any tasks to drain */
1382
1383        tsleep(sc, PPAUSE, "rewait", hz);
1384
1385	/* Unload and free the RX DMA ring memory and map */
1386
1387	if (sc->rl_ldata.rl_rx_list_tag) {
1388		bus_dmamap_unload(sc->rl_ldata.rl_rx_list_tag,
1389		    sc->rl_ldata.rl_rx_list_map);
1390		bus_dmamem_free(sc->rl_ldata.rl_rx_list_tag,
1391		    sc->rl_ldata.rl_rx_list,
1392		    sc->rl_ldata.rl_rx_list_map);
1393		bus_dma_tag_destroy(sc->rl_ldata.rl_rx_list_tag);
1394	}
1395
1396	/* Unload and free the TX DMA ring memory and map */
1397
1398	if (sc->rl_ldata.rl_tx_list_tag) {
1399		bus_dmamap_unload(sc->rl_ldata.rl_tx_list_tag,
1400		    sc->rl_ldata.rl_tx_list_map);
1401		bus_dmamem_free(sc->rl_ldata.rl_tx_list_tag,
1402		    sc->rl_ldata.rl_tx_list,
1403		    sc->rl_ldata.rl_tx_list_map);
1404		bus_dma_tag_destroy(sc->rl_ldata.rl_tx_list_tag);
1405	}
1406
1407	/* Destroy all the RX and TX buffer maps */
1408
1409	if (sc->rl_ldata.rl_mtag) {
1410		for (i = 0; i < RL_TX_DESC_CNT; i++)
1411			bus_dmamap_destroy(sc->rl_ldata.rl_mtag,
1412			    sc->rl_ldata.rl_tx_dmamap[i]);
1413		for (i = 0; i < RL_RX_DESC_CNT; i++)
1414			bus_dmamap_destroy(sc->rl_ldata.rl_mtag,
1415			    sc->rl_ldata.rl_rx_dmamap[i]);
1416		bus_dma_tag_destroy(sc->rl_ldata.rl_mtag);
1417	}
1418
1419	/* Unload and free the stats buffer and map */
1420
1421	if (sc->rl_ldata.rl_stag) {
1422		bus_dmamap_unload(sc->rl_ldata.rl_stag,
1423		    sc->rl_ldata.rl_rx_list_map);
1424		bus_dmamem_free(sc->rl_ldata.rl_stag,
1425		    sc->rl_ldata.rl_stats,
1426		    sc->rl_ldata.rl_smap);
1427		bus_dma_tag_destroy(sc->rl_ldata.rl_stag);
1428	}
1429
1430	if (sc->rl_parent_tag)
1431		bus_dma_tag_destroy(sc->rl_parent_tag);
1432
1433	mtx_destroy(&sc->rl_mtx);
1434
1435	return (0);
1436}
1437
1438static int
1439re_newbuf(sc, idx, m)
1440	struct rl_softc		*sc;
1441	int			idx;
1442	struct mbuf		*m;
1443{
1444	struct rl_dmaload_arg	arg;
1445	struct mbuf		*n = NULL;
1446	int			error;
1447
1448	if (m == NULL) {
1449		n = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
1450		if (n == NULL)
1451			return (ENOBUFS);
1452		m = n;
1453	} else
1454		m->m_data = m->m_ext.ext_buf;
1455
1456	m->m_len = m->m_pkthdr.len = MCLBYTES;
1457#ifdef RE_FIXUP_RX
1458	/*
1459	 * This is part of an evil trick to deal with non-x86 platforms.
1460	 * The RealTek chip requires RX buffers to be aligned on 64-bit
1461	 * boundaries, but that will hose non-x86 machines. To get around
1462	 * this, we leave some empty space at the start of each buffer
1463	 * and for non-x86 hosts, we copy the buffer back six bytes
1464	 * to achieve word alignment. This is slightly more efficient
1465	 * than allocating a new buffer, copying the contents, and
1466	 * discarding the old buffer.
1467	 */
1468	m_adj(m, RE_ETHER_ALIGN);
1469#endif
1470	arg.sc = sc;
1471	arg.rl_idx = idx;
1472	arg.rl_maxsegs = 1;
1473	arg.rl_flags = 0;
1474	arg.rl_ring = sc->rl_ldata.rl_rx_list;
1475
1476	error = bus_dmamap_load_mbuf(sc->rl_ldata.rl_mtag,
1477	    sc->rl_ldata.rl_rx_dmamap[idx], m, re_dma_map_desc,
1478	    &arg, BUS_DMA_NOWAIT);
1479	if (error || arg.rl_maxsegs != 1) {
1480		if (n != NULL)
1481			m_freem(n);
1482		return (ENOMEM);
1483	}
1484
1485	sc->rl_ldata.rl_rx_list[idx].rl_cmdstat |= htole32(RL_RDESC_CMD_OWN);
1486	sc->rl_ldata.rl_rx_mbuf[idx] = m;
1487
1488	bus_dmamap_sync(sc->rl_ldata.rl_mtag,
1489	    sc->rl_ldata.rl_rx_dmamap[idx],
1490	    BUS_DMASYNC_PREREAD);
1491
1492	return (0);
1493}
1494
1495#ifdef RE_FIXUP_RX
1496static __inline void
1497re_fixup_rx(m)
1498	struct mbuf		*m;
1499{
1500	int                     i;
1501	uint16_t                *src, *dst;
1502
1503	src = mtod(m, uint16_t *);
1504	dst = src - (RE_ETHER_ALIGN - ETHER_ALIGN) / sizeof *src;
1505
1506	for (i = 0; i < (m->m_len / sizeof(uint16_t) + 1); i++)
1507		*dst++ = *src++;
1508
1509	m->m_data -= RE_ETHER_ALIGN - ETHER_ALIGN;
1510
1511	return;
1512}
1513#endif
1514
1515static int
1516re_tx_list_init(sc)
1517	struct rl_softc		*sc;
1518{
1519
1520	RL_LOCK_ASSERT(sc);
1521
1522	bzero ((char *)sc->rl_ldata.rl_tx_list, RL_TX_LIST_SZ);
1523	bzero ((char *)&sc->rl_ldata.rl_tx_mbuf,
1524	    (RL_TX_DESC_CNT * sizeof(struct mbuf *)));
1525
1526	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
1527	    sc->rl_ldata.rl_tx_list_map, BUS_DMASYNC_PREWRITE);
1528	sc->rl_ldata.rl_tx_prodidx = 0;
1529	sc->rl_ldata.rl_tx_considx = 0;
1530	sc->rl_ldata.rl_tx_free = RL_TX_DESC_CNT;
1531
1532	return (0);
1533}
1534
1535static int
1536re_rx_list_init(sc)
1537	struct rl_softc		*sc;
1538{
1539	int			i;
1540
1541	bzero ((char *)sc->rl_ldata.rl_rx_list, RL_RX_LIST_SZ);
1542	bzero ((char *)&sc->rl_ldata.rl_rx_mbuf,
1543	    (RL_RX_DESC_CNT * sizeof(struct mbuf *)));
1544
1545	for (i = 0; i < RL_RX_DESC_CNT; i++) {
1546		if (re_newbuf(sc, i, NULL) == ENOBUFS)
1547			return (ENOBUFS);
1548	}
1549
1550	/* Flush the RX descriptors */
1551
1552	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1553	    sc->rl_ldata.rl_rx_list_map,
1554	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
1555
1556	sc->rl_ldata.rl_rx_prodidx = 0;
1557	sc->rl_head = sc->rl_tail = NULL;
1558
1559	return (0);
1560}
1561
1562/*
1563 * RX handler for C+ and 8169. For the gigE chips, we support
1564 * the reception of jumbo frames that have been fragmented
1565 * across multiple 2K mbuf cluster buffers.
1566 */
1567static int
1568re_rxeof(sc)
1569	struct rl_softc		*sc;
1570{
1571	struct mbuf		*m;
1572	struct ifnet		*ifp;
1573	int			i, total_len;
1574	struct rl_desc		*cur_rx;
1575	u_int32_t		rxstat, rxvlan;
1576	int			maxpkt = 16;
1577
1578	RL_LOCK_ASSERT(sc);
1579
1580	ifp = sc->rl_ifp;
1581	i = sc->rl_ldata.rl_rx_prodidx;
1582
1583	/* Invalidate the descriptor memory */
1584
1585	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1586	    sc->rl_ldata.rl_rx_list_map,
1587	    BUS_DMASYNC_POSTREAD);
1588
1589	while (!RL_OWN(&sc->rl_ldata.rl_rx_list[i]) && maxpkt) {
1590		cur_rx = &sc->rl_ldata.rl_rx_list[i];
1591		m = sc->rl_ldata.rl_rx_mbuf[i];
1592		total_len = RL_RXBYTES(cur_rx);
1593		rxstat = le32toh(cur_rx->rl_cmdstat);
1594		rxvlan = le32toh(cur_rx->rl_vlanctl);
1595
1596		/* Invalidate the RX mbuf and unload its map */
1597
1598		bus_dmamap_sync(sc->rl_ldata.rl_mtag,
1599		    sc->rl_ldata.rl_rx_dmamap[i],
1600		    BUS_DMASYNC_POSTWRITE);
1601		bus_dmamap_unload(sc->rl_ldata.rl_mtag,
1602		    sc->rl_ldata.rl_rx_dmamap[i]);
1603
1604		if (!(rxstat & RL_RDESC_STAT_EOF)) {
1605			m->m_len = RE_RX_DESC_BUFLEN;
1606			if (sc->rl_head == NULL)
1607				sc->rl_head = sc->rl_tail = m;
1608			else {
1609				m->m_flags &= ~M_PKTHDR;
1610				sc->rl_tail->m_next = m;
1611				sc->rl_tail = m;
1612			}
1613			re_newbuf(sc, i, NULL);
1614			RL_DESC_INC(i);
1615			continue;
1616		}
1617
1618		/*
1619		 * NOTE: for the 8139C+, the frame length field
1620		 * is always 12 bits in size, but for the gigE chips,
1621		 * it is 13 bits (since the max RX frame length is 16K).
1622		 * Unfortunately, all 32 bits in the status word
1623		 * were already used, so to make room for the extra
1624		 * length bit, RealTek took out the 'frame alignment
1625		 * error' bit and shifted the other status bits
1626		 * over one slot. The OWN, EOR, FS and LS bits are
1627		 * still in the same places. We have already extracted
1628		 * the frame length and checked the OWN bit, so rather
1629		 * than using an alternate bit mapping, we shift the
1630		 * status bits one space to the right so we can evaluate
1631		 * them using the 8169 status as though it was in the
1632		 * same format as that of the 8139C+.
1633		 */
1634		if (sc->rl_type == RL_8169)
1635			rxstat >>= 1;
1636
1637		/*
1638		 * if total_len > 2^13-1, both _RXERRSUM and _GIANT will be
1639		 * set, but if CRC is clear, it will still be a valid frame.
1640		 */
1641		if (rxstat & RL_RDESC_STAT_RXERRSUM && !(total_len > 8191 &&
1642		    (rxstat & RL_RDESC_STAT_ERRS) == RL_RDESC_STAT_GIANT)) {
1643			ifp->if_ierrors++;
1644			/*
1645			 * If this is part of a multi-fragment packet,
1646			 * discard all the pieces.
1647			 */
1648			if (sc->rl_head != NULL) {
1649				m_freem(sc->rl_head);
1650				sc->rl_head = sc->rl_tail = NULL;
1651			}
1652			re_newbuf(sc, i, m);
1653			RL_DESC_INC(i);
1654			continue;
1655		}
1656
1657		/*
1658		 * If allocating a replacement mbuf fails,
1659		 * reload the current one.
1660		 */
1661
1662		if (re_newbuf(sc, i, NULL)) {
1663			ifp->if_ierrors++;
1664			if (sc->rl_head != NULL) {
1665				m_freem(sc->rl_head);
1666				sc->rl_head = sc->rl_tail = NULL;
1667			}
1668			re_newbuf(sc, i, m);
1669			RL_DESC_INC(i);
1670			continue;
1671		}
1672
1673		RL_DESC_INC(i);
1674
1675		if (sc->rl_head != NULL) {
1676			m->m_len = total_len % RE_RX_DESC_BUFLEN;
1677			if (m->m_len == 0)
1678				m->m_len = RE_RX_DESC_BUFLEN;
1679			/*
1680			 * Special case: if there's 4 bytes or less
1681			 * in this buffer, the mbuf can be discarded:
1682			 * the last 4 bytes is the CRC, which we don't
1683			 * care about anyway.
1684			 */
1685			if (m->m_len <= ETHER_CRC_LEN) {
1686				sc->rl_tail->m_len -=
1687				    (ETHER_CRC_LEN - m->m_len);
1688				m_freem(m);
1689			} else {
1690				m->m_len -= ETHER_CRC_LEN;
1691				m->m_flags &= ~M_PKTHDR;
1692				sc->rl_tail->m_next = m;
1693			}
1694			m = sc->rl_head;
1695			sc->rl_head = sc->rl_tail = NULL;
1696			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
1697		} else
1698			m->m_pkthdr.len = m->m_len =
1699			    (total_len - ETHER_CRC_LEN);
1700
1701#ifdef RE_FIXUP_RX
1702		re_fixup_rx(m);
1703#endif
1704		ifp->if_ipackets++;
1705		m->m_pkthdr.rcvif = ifp;
1706
1707		/* Do RX checksumming if enabled */
1708
1709		if (ifp->if_capenable & IFCAP_RXCSUM) {
1710
1711			/* Check IP header checksum */
1712			if (rxstat & RL_RDESC_STAT_PROTOID)
1713				m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED;
1714			if (!(rxstat & RL_RDESC_STAT_IPSUMBAD))
1715				m->m_pkthdr.csum_flags |= CSUM_IP_VALID;
1716
1717			/* Check TCP/UDP checksum */
1718			if ((RL_TCPPKT(rxstat) &&
1719			    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
1720			    (RL_UDPPKT(rxstat) &&
1721			    !(rxstat & RL_RDESC_STAT_UDPSUMBAD))) {
1722				m->m_pkthdr.csum_flags |=
1723				    CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
1724				m->m_pkthdr.csum_data = 0xffff;
1725			}
1726		}
1727		maxpkt--;
1728		if (rxvlan & RL_RDESC_VLANCTL_TAG) {
1729			VLAN_INPUT_TAG(ifp, m,
1730			    ntohs((rxvlan & RL_RDESC_VLANCTL_DATA)));
1731			if (m == NULL)
1732				continue;
1733		}
1734		RL_UNLOCK(sc);
1735		(*ifp->if_input)(ifp, m);
1736		RL_LOCK(sc);
1737	}
1738
1739	/* Flush the RX DMA ring */
1740
1741	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1742	    sc->rl_ldata.rl_rx_list_map,
1743	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
1744
1745	sc->rl_ldata.rl_rx_prodidx = i;
1746
1747	if (maxpkt)
1748		return(EAGAIN);
1749
1750	return(0);
1751}
1752
1753static void
1754re_txeof(sc)
1755	struct rl_softc		*sc;
1756{
1757	struct ifnet		*ifp;
1758	u_int32_t		txstat;
1759	int			idx;
1760
1761	ifp = sc->rl_ifp;
1762	idx = sc->rl_ldata.rl_tx_considx;
1763
1764	/* Invalidate the TX descriptor list */
1765
1766	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
1767	    sc->rl_ldata.rl_tx_list_map,
1768	    BUS_DMASYNC_POSTREAD);
1769
1770	while (sc->rl_ldata.rl_tx_free < RL_TX_DESC_CNT) {
1771
1772		txstat = le32toh(sc->rl_ldata.rl_tx_list[idx].rl_cmdstat);
1773		if (txstat & RL_TDESC_CMD_OWN)
1774			break;
1775
1776		sc->rl_ldata.rl_tx_list[idx].rl_bufaddr_lo = 0;
1777
1778		/*
1779		 * We only stash mbufs in the last descriptor
1780		 * in a fragment chain, which also happens to
1781		 * be the only place where the TX status bits
1782		 * are valid.
1783		 */
1784
1785		if (txstat & RL_TDESC_CMD_EOF) {
1786			m_freem(sc->rl_ldata.rl_tx_mbuf[idx]);
1787			sc->rl_ldata.rl_tx_mbuf[idx] = NULL;
1788			bus_dmamap_unload(sc->rl_ldata.rl_mtag,
1789			    sc->rl_ldata.rl_tx_dmamap[idx]);
1790			if (txstat & (RL_TDESC_STAT_EXCESSCOL|
1791			    RL_TDESC_STAT_COLCNT))
1792				ifp->if_collisions++;
1793			if (txstat & RL_TDESC_STAT_TXERRSUM)
1794				ifp->if_oerrors++;
1795			else
1796				ifp->if_opackets++;
1797		}
1798		sc->rl_ldata.rl_tx_free++;
1799		RL_DESC_INC(idx);
1800	}
1801
1802	/* No changes made to the TX ring, so no flush needed */
1803
1804	if (sc->rl_ldata.rl_tx_free) {
1805		sc->rl_ldata.rl_tx_considx = idx;
1806		ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1807		ifp->if_timer = 0;
1808	}
1809
1810	/*
1811	 * Some chips will ignore a second TX request issued while an
1812	 * existing transmission is in progress. If the transmitter goes
1813	 * idle but there are still packets waiting to be sent, we need
1814	 * to restart the channel here to flush them out. This only seems
1815	 * to be required with the PCIe devices.
1816	 */
1817
1818	if (sc->rl_ldata.rl_tx_free < RL_TX_DESC_CNT)
1819	    CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
1820
1821#ifdef RE_TX_MODERATION
1822	/*
1823	 * If not all descriptors have been released reaped yet,
1824	 * reload the timer so that we will eventually get another
1825	 * interrupt that will cause us to re-enter this routine.
1826	 * This is done in case the transmitter has gone idle.
1827	 */
1828	if (sc->rl_ldata.rl_tx_free != RL_TX_DESC_CNT)
1829		CSR_WRITE_4(sc, RL_TIMERCNT, 1);
1830#endif
1831
1832}
1833
1834static void
1835re_tick(xsc)
1836	void			*xsc;
1837{
1838	struct rl_softc		*sc;
1839	struct mii_data		*mii;
1840	struct ifnet		*ifp;
1841
1842	sc = xsc;
1843	ifp = sc->rl_ifp;
1844
1845	RL_LOCK_ASSERT(sc);
1846
1847	mii = device_get_softc(sc->rl_miibus);
1848
1849	mii_tick(mii);
1850	if (sc->rl_link) {
1851		if (!(mii->mii_media_status & IFM_ACTIVE))
1852			sc->rl_link = 0;
1853	} else {
1854		if (mii->mii_media_status & IFM_ACTIVE &&
1855		    IFM_SUBTYPE(mii->mii_media_active) != IFM_NONE) {
1856			sc->rl_link = 1;
1857			if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
1858				taskqueue_enqueue_fast(taskqueue_fast,
1859				    &sc->rl_txtask);
1860		}
1861	}
1862
1863	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
1864}
1865
1866#ifdef DEVICE_POLLING
1867static void
1868re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
1869{
1870	struct rl_softc *sc = ifp->if_softc;
1871
1872	RL_LOCK(sc);
1873	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1874		re_poll_locked(ifp, cmd, count);
1875	RL_UNLOCK(sc);
1876}
1877
1878static void
1879re_poll_locked(struct ifnet *ifp, enum poll_cmd cmd, int count)
1880{
1881	struct rl_softc *sc = ifp->if_softc;
1882
1883	RL_LOCK_ASSERT(sc);
1884
1885	sc->rxcycles = count;
1886	re_rxeof(sc);
1887	re_txeof(sc);
1888
1889	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
1890		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_txtask);
1891
1892	if (cmd == POLL_AND_CHECK_STATUS) { /* also check status register */
1893		u_int16_t       status;
1894
1895		status = CSR_READ_2(sc, RL_ISR);
1896		if (status == 0xffff)
1897			return;
1898		if (status)
1899			CSR_WRITE_2(sc, RL_ISR, status);
1900
1901		/*
1902		 * XXX check behaviour on receiver stalls.
1903		 */
1904
1905		if (status & RL_ISR_SYSTEM_ERR) {
1906			re_reset(sc);
1907			re_init_locked(sc);
1908		}
1909	}
1910}
1911#endif /* DEVICE_POLLING */
1912
1913static void
1914re_intr(arg)
1915	void			*arg;
1916{
1917	struct rl_softc		*sc;
1918	struct ifnet		*ifp;
1919	uint16_t		status;
1920
1921	sc = arg;
1922	ifp = sc->rl_ifp;
1923
1924	status = CSR_READ_2(sc, RL_ISR);
1925	if (status == 0xFFFF || (status & RL_INTRS_CPLUS) == 0)
1926                return;
1927	CSR_WRITE_2(sc, RL_IMR, 0);
1928
1929	taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_inttask);
1930
1931	return;
1932}
1933
1934static void
1935re_int_task(arg, npending)
1936	void			*arg;
1937	int			npending;
1938{
1939	struct rl_softc		*sc;
1940	struct ifnet		*ifp;
1941	u_int16_t		status;
1942	int			rval = 0;
1943
1944	sc = arg;
1945	ifp = sc->rl_ifp;
1946
1947	RL_LOCK(sc);
1948
1949	status = CSR_READ_2(sc, RL_ISR);
1950        CSR_WRITE_2(sc, RL_ISR, status);
1951
1952	if (sc->suspended || !(ifp->if_flags & IFF_UP)) {
1953		RL_UNLOCK(sc);
1954		return;
1955	}
1956
1957#ifdef DEVICE_POLLING
1958	if  (ifp->if_capenable & IFCAP_POLLING) {
1959		RL_UNLOCK(sc);
1960		return;
1961	}
1962#endif
1963
1964	if (status & (RL_ISR_RX_OK|RL_ISR_RX_ERR|RL_ISR_FIFO_OFLOW))
1965		rval = re_rxeof(sc);
1966
1967#ifdef RE_TX_MODERATION
1968	if (status & (RL_ISR_TIMEOUT_EXPIRED|
1969#else
1970	if (status & (RL_ISR_TX_OK|
1971#endif
1972	    RL_ISR_TX_ERR|RL_ISR_TX_DESC_UNAVAIL))
1973		re_txeof(sc);
1974
1975	if (status & RL_ISR_SYSTEM_ERR) {
1976		re_reset(sc);
1977		re_init_locked(sc);
1978	}
1979
1980	if (status & RL_ISR_LINKCHG) {
1981		callout_stop(&sc->rl_stat_callout);
1982		re_tick(sc);
1983	}
1984
1985	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
1986		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_txtask);
1987
1988	RL_UNLOCK(sc);
1989
1990        if ((CSR_READ_2(sc, RL_ISR) & RL_INTRS_CPLUS) || rval) {
1991		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_inttask);
1992		return;
1993	}
1994
1995	CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
1996
1997	return;
1998}
1999
2000static int
2001re_encap(sc, m_head, idx)
2002	struct rl_softc		*sc;
2003	struct mbuf		**m_head;
2004	int			*idx;
2005{
2006	struct mbuf		*m_new = NULL;
2007	struct rl_dmaload_arg	arg;
2008	bus_dmamap_t		map;
2009	int			error;
2010	struct m_tag		*mtag;
2011
2012	RL_LOCK_ASSERT(sc);
2013
2014	if (sc->rl_ldata.rl_tx_free <= 4)
2015		return (EFBIG);
2016
2017	/*
2018	 * Set up checksum offload. Note: checksum offload bits must
2019	 * appear in all descriptors of a multi-descriptor transmit
2020	 * attempt. This is according to testing done with an 8169
2021	 * chip. This is a requirement.
2022	 */
2023
2024	arg.rl_flags = 0;
2025
2026	if ((*m_head)->m_pkthdr.csum_flags & CSUM_IP)
2027		arg.rl_flags |= RL_TDESC_CMD_IPCSUM;
2028	if ((*m_head)->m_pkthdr.csum_flags & CSUM_TCP)
2029		arg.rl_flags |= RL_TDESC_CMD_TCPCSUM;
2030	if ((*m_head)->m_pkthdr.csum_flags & CSUM_UDP)
2031		arg.rl_flags |= RL_TDESC_CMD_UDPCSUM;
2032
2033	arg.sc = sc;
2034	arg.rl_idx = *idx;
2035	arg.rl_maxsegs = sc->rl_ldata.rl_tx_free;
2036	if (arg.rl_maxsegs > 4)
2037		arg.rl_maxsegs -= 4;
2038	arg.rl_ring = sc->rl_ldata.rl_tx_list;
2039
2040	map = sc->rl_ldata.rl_tx_dmamap[*idx];
2041
2042	/*
2043	 * With some of the RealTek chips, using the checksum offload
2044	 * support in conjunction with the autopadding feature results
2045	 * in the transmission of corrupt frames. For example, if we
2046	 * need to send a really small IP fragment that's less than 60
2047	 * bytes in size, and IP header checksumming is enabled, the
2048	 * resulting ethernet frame that appears on the wire will
2049	 * have garbled payload. To work around this, if TX checksum
2050	 * offload is enabled, we always manually pad short frames out
2051	 * to the minimum ethernet frame size. We do this by pretending
2052	 * the mbuf chain has too many fragments so the coalescing code
2053	 * below can assemble the packet into a single buffer that's
2054	 * padded out to the mininum frame size.
2055	 */
2056
2057	if (arg.rl_flags && (*m_head)->m_pkthdr.len < RL_MIN_FRAMELEN)
2058		error = EFBIG;
2059	else
2060		error = bus_dmamap_load_mbuf(sc->rl_ldata.rl_mtag, map,
2061		    *m_head, re_dma_map_desc, &arg, BUS_DMA_NOWAIT);
2062
2063	if (error && error != EFBIG) {
2064		device_printf(sc->rl_dev, "can't map mbuf (error %d)\n", error);
2065		return (ENOBUFS);
2066	}
2067
2068	/* Too many segments to map, coalesce into a single mbuf */
2069
2070	if (error || arg.rl_maxsegs == 0) {
2071		m_new = m_defrag(*m_head, M_DONTWAIT);
2072		if (m_new == NULL)
2073			return (ENOBUFS);
2074		else
2075			*m_head = m_new;
2076
2077		/*
2078		 * Manually pad short frames, and zero the pad space
2079		 * to avoid leaking data.
2080		 */
2081
2082		if (m_new->m_pkthdr.len < RL_MIN_FRAMELEN) {
2083			bzero(mtod(m_new, char *) + m_new->m_pkthdr.len,
2084			    RL_MIN_FRAMELEN - m_new->m_pkthdr.len);
2085			m_new->m_pkthdr.len += RL_MIN_FRAMELEN -
2086			    m_new->m_pkthdr.len;
2087			m_new->m_len = m_new->m_pkthdr.len;
2088		}
2089
2090		arg.sc = sc;
2091		arg.rl_idx = *idx;
2092		arg.rl_maxsegs = sc->rl_ldata.rl_tx_free;
2093		arg.rl_ring = sc->rl_ldata.rl_tx_list;
2094
2095		error = bus_dmamap_load_mbuf(sc->rl_ldata.rl_mtag, map,
2096		    *m_head, re_dma_map_desc, &arg, BUS_DMA_NOWAIT);
2097		if (error) {
2098			device_printf(sc->rl_dev, "can't map mbuf (error %d)\n",
2099			    error);
2100			return (EFBIG);
2101		}
2102	}
2103
2104	/*
2105	 * Insure that the map for this transmission
2106	 * is placed at the array index of the last descriptor
2107	 * in this chain.  (Swap last and first dmamaps.)
2108	 */
2109	sc->rl_ldata.rl_tx_dmamap[*idx] =
2110	    sc->rl_ldata.rl_tx_dmamap[arg.rl_idx];
2111	sc->rl_ldata.rl_tx_dmamap[arg.rl_idx] = map;
2112
2113	sc->rl_ldata.rl_tx_mbuf[arg.rl_idx] = *m_head;
2114	sc->rl_ldata.rl_tx_free -= arg.rl_maxsegs;
2115
2116	/*
2117	 * Set up hardware VLAN tagging. Note: vlan tag info must
2118	 * appear in the first descriptor of a multi-descriptor
2119	 * transmission attempt.
2120	 */
2121
2122	mtag = VLAN_OUTPUT_TAG(sc->rl_ifp, *m_head);
2123	if (mtag != NULL)
2124		sc->rl_ldata.rl_tx_list[*idx].rl_vlanctl =
2125		    htole32(htons(VLAN_TAG_VALUE(mtag)) | RL_TDESC_VLANCTL_TAG);
2126
2127	/* Transfer ownership of packet to the chip. */
2128
2129	sc->rl_ldata.rl_tx_list[arg.rl_idx].rl_cmdstat |=
2130	    htole32(RL_TDESC_CMD_OWN);
2131	if (*idx != arg.rl_idx)
2132		sc->rl_ldata.rl_tx_list[*idx].rl_cmdstat |=
2133		    htole32(RL_TDESC_CMD_OWN);
2134
2135        RL_DESC_INC(arg.rl_idx);
2136	*idx = arg.rl_idx;
2137
2138	return (0);
2139}
2140
2141static void
2142re_tx_task(arg, npending)
2143	void			*arg;
2144	int			npending;
2145{
2146	struct ifnet		*ifp;
2147
2148	ifp = arg;
2149	re_start(ifp);
2150
2151	return;
2152}
2153
2154/*
2155 * Main transmit routine for C+ and gigE NICs.
2156 */
2157static void
2158re_start(ifp)
2159	struct ifnet		*ifp;
2160{
2161	struct rl_softc		*sc;
2162	struct mbuf		*m_head = NULL;
2163	int			idx, queued = 0;
2164
2165	sc = ifp->if_softc;
2166
2167	RL_LOCK(sc);
2168
2169	if (!sc->rl_link || ifp->if_drv_flags & IFF_DRV_OACTIVE) {
2170		RL_UNLOCK(sc);
2171		return;
2172	}
2173
2174	idx = sc->rl_ldata.rl_tx_prodidx;
2175
2176	while (sc->rl_ldata.rl_tx_mbuf[idx] == NULL) {
2177		IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
2178		if (m_head == NULL)
2179			break;
2180
2181		if (re_encap(sc, &m_head, &idx)) {
2182			IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
2183			ifp->if_drv_flags |= IFF_DRV_OACTIVE;
2184			break;
2185		}
2186
2187		/*
2188		 * If there's a BPF listener, bounce a copy of this frame
2189		 * to him.
2190		 */
2191		BPF_MTAP(ifp, m_head);
2192
2193		queued++;
2194	}
2195
2196	if (queued == 0) {
2197#ifdef RE_TX_MODERATION
2198		if (sc->rl_ldata.rl_tx_free != RL_TX_DESC_CNT)
2199			CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2200#endif
2201		RL_UNLOCK(sc);
2202		return;
2203	}
2204
2205	/* Flush the TX descriptors */
2206
2207	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2208	    sc->rl_ldata.rl_tx_list_map,
2209	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2210
2211	sc->rl_ldata.rl_tx_prodidx = idx;
2212
2213	/*
2214	 * RealTek put the TX poll request register in a different
2215	 * location on the 8169 gigE chip. I don't know why.
2216	 */
2217
2218	CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2219
2220#ifdef RE_TX_MODERATION
2221	/*
2222	 * Use the countdown timer for interrupt moderation.
2223	 * 'TX done' interrupts are disabled. Instead, we reset the
2224	 * countdown timer, which will begin counting until it hits
2225	 * the value in the TIMERINT register, and then trigger an
2226	 * interrupt. Each time we write to the TIMERCNT register,
2227	 * the timer count is reset to 0.
2228	 */
2229	CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2230#endif
2231
2232	/*
2233	 * Set a timeout in case the chip goes out to lunch.
2234	 */
2235
2236	ifp->if_timer = 5;
2237
2238	RL_UNLOCK(sc);
2239
2240	return;
2241}
2242
2243static void
2244re_init(xsc)
2245	void			*xsc;
2246{
2247	struct rl_softc		*sc = xsc;
2248
2249	RL_LOCK(sc);
2250	re_init_locked(sc);
2251	RL_UNLOCK(sc);
2252}
2253
2254static void
2255re_init_locked(sc)
2256	struct rl_softc		*sc;
2257{
2258	struct ifnet		*ifp = sc->rl_ifp;
2259	struct mii_data		*mii;
2260	u_int32_t		rxcfg = 0;
2261	union {
2262		uint32_t align_dummy;
2263		u_char eaddr[ETHER_ADDR_LEN];
2264        } eaddr;
2265
2266	RL_LOCK_ASSERT(sc);
2267
2268	mii = device_get_softc(sc->rl_miibus);
2269
2270	/*
2271	 * Cancel pending I/O and free all RX/TX buffers.
2272	 */
2273	re_stop(sc);
2274
2275	/*
2276	 * Enable C+ RX and TX mode, as well as VLAN stripping and
2277	 * RX checksum offload. We must configure the C+ register
2278	 * before all others.
2279	 */
2280	CSR_WRITE_2(sc, RL_CPLUS_CMD, RL_CPLUSCMD_RXENB|
2281	    RL_CPLUSCMD_TXENB|RL_CPLUSCMD_PCI_MRW|
2282	    RL_CPLUSCMD_VLANSTRIP|RL_CPLUSCMD_RXCSUM_ENB);
2283
2284	/*
2285	 * Init our MAC address.  Even though the chipset
2286	 * documentation doesn't mention it, we need to enter "Config
2287	 * register write enable" mode to modify the ID registers.
2288	 */
2289	/* Copy MAC address on stack to align. */
2290	bcopy(IF_LLADDR(ifp), eaddr.eaddr, ETHER_ADDR_LEN);
2291	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG);
2292	CSR_WRITE_4(sc, RL_IDR0,
2293	    htole32(*(u_int32_t *)(&eaddr.eaddr[0])));
2294	CSR_WRITE_4(sc, RL_IDR4,
2295	    htole32(*(u_int32_t *)(&eaddr.eaddr[4])));
2296	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
2297
2298	/*
2299	 * For C+ mode, initialize the RX descriptors and mbufs.
2300	 */
2301	re_rx_list_init(sc);
2302	re_tx_list_init(sc);
2303
2304	/*
2305	 * Enable transmit and receive.
2306	 */
2307	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
2308
2309	/*
2310	 * Set the initial TX and RX configuration.
2311	 */
2312	if (sc->rl_testmode) {
2313		if (sc->rl_type == RL_8169)
2314			CSR_WRITE_4(sc, RL_TXCFG,
2315			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON);
2316		else
2317			CSR_WRITE_4(sc, RL_TXCFG,
2318			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON_CPLUS);
2319	} else
2320		CSR_WRITE_4(sc, RL_TXCFG, RL_TXCFG_CONFIG);
2321	CSR_WRITE_4(sc, RL_RXCFG, RL_RXCFG_CONFIG);
2322
2323	/* Set the individual bit to receive frames for this host only. */
2324	rxcfg = CSR_READ_4(sc, RL_RXCFG);
2325	rxcfg |= RL_RXCFG_RX_INDIV;
2326
2327	/* If we want promiscuous mode, set the allframes bit. */
2328	if (ifp->if_flags & IFF_PROMISC)
2329		rxcfg |= RL_RXCFG_RX_ALLPHYS;
2330	else
2331		rxcfg &= ~RL_RXCFG_RX_ALLPHYS;
2332	CSR_WRITE_4(sc, RL_RXCFG, rxcfg);
2333
2334	/*
2335	 * Set capture broadcast bit to capture broadcast frames.
2336	 */
2337	if (ifp->if_flags & IFF_BROADCAST)
2338		rxcfg |= RL_RXCFG_RX_BROAD;
2339	else
2340		rxcfg &= ~RL_RXCFG_RX_BROAD;
2341	CSR_WRITE_4(sc, RL_RXCFG, rxcfg);
2342
2343	/*
2344	 * Program the multicast filter, if necessary.
2345	 */
2346	re_setmulti(sc);
2347
2348#ifdef DEVICE_POLLING
2349	/*
2350	 * Disable interrupts if we are polling.
2351	 */
2352	if (ifp->if_capenable & IFCAP_POLLING)
2353		CSR_WRITE_2(sc, RL_IMR, 0);
2354	else	/* otherwise ... */
2355#endif
2356
2357	/*
2358	 * Enable interrupts.
2359	 */
2360	if (sc->rl_testmode)
2361		CSR_WRITE_2(sc, RL_IMR, 0);
2362	else
2363		CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2364	CSR_WRITE_2(sc, RL_ISR, RL_INTRS_CPLUS);
2365
2366	/* Set initial TX threshold */
2367	sc->rl_txthresh = RL_TX_THRESH_INIT;
2368
2369	/* Start RX/TX process. */
2370	CSR_WRITE_4(sc, RL_MISSEDPKT, 0);
2371#ifdef notdef
2372	/* Enable receiver and transmitter. */
2373	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
2374#endif
2375	/*
2376	 * Load the addresses of the RX and TX lists into the chip.
2377	 */
2378
2379	CSR_WRITE_4(sc, RL_RXLIST_ADDR_HI,
2380	    RL_ADDR_HI(sc->rl_ldata.rl_rx_list_addr));
2381	CSR_WRITE_4(sc, RL_RXLIST_ADDR_LO,
2382	    RL_ADDR_LO(sc->rl_ldata.rl_rx_list_addr));
2383
2384	CSR_WRITE_4(sc, RL_TXLIST_ADDR_HI,
2385	    RL_ADDR_HI(sc->rl_ldata.rl_tx_list_addr));
2386	CSR_WRITE_4(sc, RL_TXLIST_ADDR_LO,
2387	    RL_ADDR_LO(sc->rl_ldata.rl_tx_list_addr));
2388
2389	CSR_WRITE_1(sc, RL_EARLY_TX_THRESH, 16);
2390
2391#ifdef RE_TX_MODERATION
2392	/*
2393	 * Initialize the timer interrupt register so that
2394	 * a timer interrupt will be generated once the timer
2395	 * reaches a certain number of ticks. The timer is
2396	 * reloaded on each transmit. This gives us TX interrupt
2397	 * moderation, which dramatically improves TX frame rate.
2398	 */
2399	if (sc->rl_type == RL_8169)
2400		CSR_WRITE_4(sc, RL_TIMERINT_8169, 0x800);
2401	else
2402		CSR_WRITE_4(sc, RL_TIMERINT, 0x400);
2403#endif
2404
2405	/*
2406	 * For 8169 gigE NICs, set the max allowed RX packet
2407	 * size so we can receive jumbo frames.
2408	 */
2409	if (sc->rl_type == RL_8169)
2410		CSR_WRITE_2(sc, RL_MAXRXPKTLEN, 16383);
2411
2412	if (sc->rl_testmode)
2413		return;
2414
2415	mii_mediachg(mii);
2416
2417	CSR_WRITE_1(sc, RL_CFG1, RL_CFG1_DRVLOAD|RL_CFG1_FULLDUPLEX);
2418
2419	ifp->if_drv_flags |= IFF_DRV_RUNNING;
2420	ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
2421
2422
2423	sc->rl_link = 0;
2424
2425	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
2426}
2427
2428/*
2429 * Set media options.
2430 */
2431static int
2432re_ifmedia_upd(ifp)
2433	struct ifnet		*ifp;
2434{
2435	struct rl_softc		*sc;
2436	struct mii_data		*mii;
2437
2438	sc = ifp->if_softc;
2439	mii = device_get_softc(sc->rl_miibus);
2440	RL_LOCK(sc);
2441	mii_mediachg(mii);
2442	RL_UNLOCK(sc);
2443
2444	return (0);
2445}
2446
2447/*
2448 * Report current media status.
2449 */
2450static void
2451re_ifmedia_sts(ifp, ifmr)
2452	struct ifnet		*ifp;
2453	struct ifmediareq	*ifmr;
2454{
2455	struct rl_softc		*sc;
2456	struct mii_data		*mii;
2457
2458	sc = ifp->if_softc;
2459	mii = device_get_softc(sc->rl_miibus);
2460
2461	RL_LOCK(sc);
2462	mii_pollstat(mii);
2463	RL_UNLOCK(sc);
2464	ifmr->ifm_active = mii->mii_media_active;
2465	ifmr->ifm_status = mii->mii_media_status;
2466}
2467
2468static int
2469re_ioctl(ifp, command, data)
2470	struct ifnet		*ifp;
2471	u_long			command;
2472	caddr_t			data;
2473{
2474	struct rl_softc		*sc = ifp->if_softc;
2475	struct ifreq		*ifr = (struct ifreq *) data;
2476	struct mii_data		*mii;
2477	int			error = 0;
2478
2479	switch (command) {
2480	case SIOCSIFMTU:
2481		RL_LOCK(sc);
2482		if (ifr->ifr_mtu > RL_JUMBO_MTU)
2483			error = EINVAL;
2484		ifp->if_mtu = ifr->ifr_mtu;
2485		RL_UNLOCK(sc);
2486		break;
2487	case SIOCSIFFLAGS:
2488		RL_LOCK(sc);
2489		if (ifp->if_flags & IFF_UP)
2490			re_init_locked(sc);
2491		else if (ifp->if_drv_flags & IFF_DRV_RUNNING)
2492			re_stop(sc);
2493		RL_UNLOCK(sc);
2494		break;
2495	case SIOCADDMULTI:
2496	case SIOCDELMULTI:
2497		RL_LOCK(sc);
2498		re_setmulti(sc);
2499		RL_UNLOCK(sc);
2500		break;
2501	case SIOCGIFMEDIA:
2502	case SIOCSIFMEDIA:
2503		mii = device_get_softc(sc->rl_miibus);
2504		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, command);
2505		break;
2506	case SIOCSIFCAP:
2507	    {
2508		int mask, reinit;
2509
2510		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
2511		reinit = 0;
2512#ifdef DEVICE_POLLING
2513		if (mask & IFCAP_POLLING) {
2514			if (ifr->ifr_reqcap & IFCAP_POLLING) {
2515				error = ether_poll_register(re_poll, ifp);
2516				if (error)
2517					return(error);
2518				RL_LOCK(sc);
2519				/* Disable interrupts */
2520				CSR_WRITE_2(sc, RL_IMR, 0x0000);
2521				ifp->if_capenable |= IFCAP_POLLING;
2522				RL_UNLOCK(sc);
2523
2524			} else {
2525				error = ether_poll_deregister(ifp);
2526				/* Enable interrupts. */
2527				RL_LOCK(sc);
2528				CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2529				ifp->if_capenable &= ~IFCAP_POLLING;
2530				RL_UNLOCK(sc);
2531			}
2532		}
2533#endif /* DEVICE_POLLING */
2534		if (mask & IFCAP_HWCSUM) {
2535			ifp->if_capenable ^= IFCAP_HWCSUM;
2536			if (ifp->if_capenable & IFCAP_TXCSUM)
2537				ifp->if_hwassist = RE_CSUM_FEATURES;
2538			else
2539				ifp->if_hwassist = 0;
2540			reinit = 1;
2541		}
2542		if (mask & IFCAP_VLAN_HWTAGGING) {
2543			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
2544			reinit = 1;
2545		}
2546		if (reinit && ifp->if_drv_flags & IFF_DRV_RUNNING)
2547			re_init(sc);
2548	    }
2549		break;
2550	default:
2551		error = ether_ioctl(ifp, command, data);
2552		break;
2553	}
2554
2555	return (error);
2556}
2557
2558static void
2559re_watchdog(ifp)
2560	struct ifnet		*ifp;
2561{
2562	struct rl_softc		*sc;
2563
2564	sc = ifp->if_softc;
2565	RL_LOCK(sc);
2566	if_printf(ifp, "watchdog timeout\n");
2567	ifp->if_oerrors++;
2568
2569	re_txeof(sc);
2570	re_rxeof(sc);
2571	re_init_locked(sc);
2572
2573	RL_UNLOCK(sc);
2574}
2575
2576/*
2577 * Stop the adapter and free any mbufs allocated to the
2578 * RX and TX lists.
2579 */
2580static void
2581re_stop(sc)
2582	struct rl_softc		*sc;
2583{
2584	register int		i;
2585	struct ifnet		*ifp;
2586
2587	RL_LOCK_ASSERT(sc);
2588
2589	ifp = sc->rl_ifp;
2590	ifp->if_timer = 0;
2591
2592	callout_stop(&sc->rl_stat_callout);
2593	ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
2594
2595	CSR_WRITE_1(sc, RL_COMMAND, 0x00);
2596	CSR_WRITE_2(sc, RL_IMR, 0x0000);
2597	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
2598
2599	if (sc->rl_head != NULL) {
2600		m_freem(sc->rl_head);
2601		sc->rl_head = sc->rl_tail = NULL;
2602	}
2603
2604	/* Free the TX list buffers. */
2605
2606	for (i = 0; i < RL_TX_DESC_CNT; i++) {
2607		if (sc->rl_ldata.rl_tx_mbuf[i] != NULL) {
2608			bus_dmamap_unload(sc->rl_ldata.rl_mtag,
2609			    sc->rl_ldata.rl_tx_dmamap[i]);
2610			m_freem(sc->rl_ldata.rl_tx_mbuf[i]);
2611			sc->rl_ldata.rl_tx_mbuf[i] = NULL;
2612		}
2613	}
2614
2615	/* Free the RX list buffers. */
2616
2617	for (i = 0; i < RL_RX_DESC_CNT; i++) {
2618		if (sc->rl_ldata.rl_rx_mbuf[i] != NULL) {
2619			bus_dmamap_unload(sc->rl_ldata.rl_mtag,
2620			    sc->rl_ldata.rl_rx_dmamap[i]);
2621			m_freem(sc->rl_ldata.rl_rx_mbuf[i]);
2622			sc->rl_ldata.rl_rx_mbuf[i] = NULL;
2623		}
2624	}
2625}
2626
2627/*
2628 * Device suspend routine.  Stop the interface and save some PCI
2629 * settings in case the BIOS doesn't restore them properly on
2630 * resume.
2631 */
2632static int
2633re_suspend(dev)
2634	device_t		dev;
2635{
2636	struct rl_softc		*sc;
2637
2638	sc = device_get_softc(dev);
2639
2640	RL_LOCK(sc);
2641	re_stop(sc);
2642	sc->suspended = 1;
2643	RL_UNLOCK(sc);
2644
2645	return (0);
2646}
2647
2648/*
2649 * Device resume routine.  Restore some PCI settings in case the BIOS
2650 * doesn't, re-enable busmastering, and restart the interface if
2651 * appropriate.
2652 */
2653static int
2654re_resume(dev)
2655	device_t		dev;
2656{
2657	struct rl_softc		*sc;
2658	struct ifnet		*ifp;
2659
2660	sc = device_get_softc(dev);
2661
2662	RL_LOCK(sc);
2663
2664	ifp = sc->rl_ifp;
2665
2666	/* reinitialize interface if necessary */
2667	if (ifp->if_flags & IFF_UP)
2668		re_init_locked(sc);
2669
2670	sc->suspended = 0;
2671	RL_UNLOCK(sc);
2672
2673	return (0);
2674}
2675
2676/*
2677 * Stop all chip I/O so that the kernel's probe routines don't
2678 * get confused by errant DMAs when rebooting.
2679 */
2680static void
2681re_shutdown(dev)
2682	device_t		dev;
2683{
2684	struct rl_softc		*sc;
2685
2686	sc = device_get_softc(dev);
2687
2688	RL_LOCK(sc);
2689	re_stop(sc);
2690	/*
2691	 * Mark interface as down since otherwise we will panic if
2692	 * interrupt comes in later on, which can happen in some
2693	 * cases.
2694	 */
2695	sc->rl_ifp->if_flags &= ~IFF_UP;
2696	RL_UNLOCK(sc);
2697}
2698