if_rl.c revision 118978
1/*
2 * Copyright (c) 1997, 1998-2003
3 *	Bill Paul <wpaul@ctr.columbia.edu>.  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/*
34 * RealTek 8129/8139/8139C+/8169 PCI NIC driver
35 *
36 * Supports several extremely cheap PCI 10/100 and 10/100/1000 adapters
37 * based on RealTek chipsets. Datasheets can be obtained from
38 * www.realtek.com.tw.
39 *
40 * Written by Bill Paul <wpaul@windriver.com>
41 * Senior Networking Software Engineer
42 * Wind River Systems
43 */
44
45/*
46 * The RealTek 8139 PCI NIC redefines the meaning of 'low end.' This is
47 * probably the worst PCI ethernet controller ever made, with the possible
48 * exception of the FEAST chip made by SMC. The 8139 supports bus-master
49 * DMA, but it has a terrible interface that nullifies any performance
50 * gains that bus-master DMA usually offers.
51 *
52 * For transmission, the chip offers a series of four TX descriptor
53 * registers. Each transmit frame must be in a contiguous buffer, aligned
54 * on a longword (32-bit) boundary. This means we almost always have to
55 * do mbuf copies in order to transmit a frame, except in the unlikely
56 * case where a) the packet fits into a single mbuf, and b) the packet
57 * is 32-bit aligned within the mbuf's data area. The presence of only
58 * four descriptor registers means that we can never have more than four
59 * packets queued for transmission at any one time.
60 *
61 * Reception is not much better. The driver has to allocate a single large
62 * buffer area (up to 64K in size) into which the chip will DMA received
63 * frames. Because we don't know where within this region received packets
64 * will begin or end, we have no choice but to copy data from the buffer
65 * area into mbufs in order to pass the packets up to the higher protocol
66 * levels.
67 *
68 * It's impossible given this rotten design to really achieve decent
69 * performance at 100Mbps, unless you happen to have a 400Mhz PII or
70 * some equally overmuscled CPU to drive it.
71 *
72 * On the bright side, the 8139 does have a built-in PHY, although
73 * rather than using an MDIO serial interface like most other NICs, the
74 * PHY registers are directly accessible through the 8139's register
75 * space. The 8139 supports autonegotiation, as well as a 64-bit multicast
76 * filter.
77 *
78 * The 8129 chip is an older version of the 8139 that uses an external PHY
79 * chip. The 8129 has a serial MDIO interface for accessing the MII where
80 * the 8139 lets you directly access the on-board PHY registers. We need
81 * to select which interface to use depending on the chip type.
82 *
83 * Fast forward a few years. RealTek now has a new chip called the
84 * 8139C+ which at long last implements descriptor-based DMA. Not
85 * only that, it supports RX and TX TCP/IP checksum offload, VLAN
86 * tagging and insertion, TCP large send and 64-bit addressing.
87 * Better still, it allows arbitrary byte alignments for RX and
88 * TX buffers, meaning no copying is necessary on any architecture.
89 * There are a few limitations however: the RX and TX descriptor
90 * rings must be aligned on 256 byte boundaries, they must be in
91 * contiguous RAM, and each ring can have a maximum of 64 descriptors.
92 * There are two TX descriptor queues: one normal priority and one
93 * high. Descriptor ring addresses and DMA buffer addresses are
94 * 64 bits wide. The 8139C+ is also backwards compatible with the
95 * 8139, so the chip will still function with older drivers: C+
96 * mode has to be enabled by setting the appropriate bits in the C+
97 * command register. The PHY access mechanism appears to be unchanged.
98 *
99 * The 8169 is a 10/100/1000 ethernet MAC. It has almost the same
100 * programming API as the C+ mode of the 8139C+, with a couple of
101 * minor changes and additions: TX start register and timer interrupt
102 * register are located at different offsets, and there are additional
103 * registers for GMII PHY status and control, as well as TBI-mode
104 * status and control. There is also a maximum RX packet size
105 * register to allow the chip to receive jumbo frames. The 8169
106 * can only be programmed in C+ mode: the old 8139 programming
107 * method isn't supported with this chip. Also, RealTek has a LOM
108 * (LAN On Motherboard) gigabit MAC chip called the RTL8110S which
109 * I believe to be register compatible with the 8169. Unlike the
110 * 8139C+, the 8169 can have up to 1024 descriptors per DMA ring.
111 * The reference 8169 board design uses a Marvell 88E1000 'Alaska'
112 * copper PHY.
113 *
114 * The 8169S and 8110S are newer versions of the 8169. Available
115 * in both 32-bit and 64-bit forms, these devices have built-in
116 * copper 10/100/1000 PHYs. The 8110S is a lan-on-motherboard chip
117 * that is pin-for-pin compatible with the 8100. Unfortunately,
118 * RealTek has not released programming manuals for the 8169S and
119 * 8110S yet. The datasheet for the original 8169 provides most
120 * of the information, but you must refer to RealTek's 8169 Linux
121 * driver to fill in the gaps. Mostly, it appears that the built-in
122 * PHY requires some special initialization. The original 8169
123 * datasheet and the 8139C+ datasheet can be obtained from
124 * http://www.freebsd.org/~wpaul/RealTek.
125 *
126 * This driver now supports both the old 8139 and new 8139C+
127 * programming models. We detect the 8139C+ by looking for the
128 * corresponding hardware rev bits, and we detect the 8169 by its
129 * PCI ID. Two new NIC type codes, RL_8139CPLUS and RL_8169 have
130 * been added to distinguish the chips at runtime. Separate RX and
131 * TX handling routines have been added to handle C+ mode, which
132 * are selected via function pointers that are initialized during
133 * the driver attach phase.
134 */
135
136#include <sys/cdefs.h>
137__FBSDID("$FreeBSD: head/sys/pci/if_rl.c 118978 2003-08-15 22:47:55Z wpaul $");
138
139#include <sys/param.h>
140#include <sys/endian.h>
141#include <sys/systm.h>
142#include <sys/sockio.h>
143#include <sys/mbuf.h>
144#include <sys/malloc.h>
145#include <sys/kernel.h>
146#include <sys/socket.h>
147
148#include <net/if.h>
149#include <net/if_arp.h>
150#include <net/ethernet.h>
151#include <net/if_dl.h>
152#include <net/if_media.h>
153#include <net/if_vlan_var.h>
154
155#include <net/bpf.h>
156
157#include <machine/bus_pio.h>
158#include <machine/bus_memio.h>
159#include <machine/bus.h>
160#include <machine/resource.h>
161#include <sys/bus.h>
162#include <sys/rman.h>
163
164#include <dev/mii/mii.h>
165#include <dev/mii/miivar.h>
166
167#include <pci/pcireg.h>
168#include <pci/pcivar.h>
169
170MODULE_DEPEND(rl, pci, 1, 1, 1);
171MODULE_DEPEND(rl, ether, 1, 1, 1);
172MODULE_DEPEND(rl, miibus, 1, 1, 1);
173
174/* "controller miibus0" required.  See GENERIC if you get errors here. */
175#include "miibus_if.h"
176
177/*
178 * Default to using PIO access for this driver. On SMP systems,
179 * there appear to be problems with memory mapped mode: it looks like
180 * doing too many memory mapped access back to back in rapid succession
181 * can hang the bus. I'm inclined to blame this on crummy design/construction
182 * on the part of RealTek. Memory mapped mode does appear to work on
183 * uniprocessor systems though.
184 */
185#define RL_USEIOSPACE
186
187#include <pci/if_rlreg.h>
188
189#define RL_CSUM_FEATURES    (CSUM_IP | CSUM_TCP | CSUM_UDP)
190
191/*
192 * Various supported device vendors/types and their names.
193 */
194static struct rl_type rl_devs[] = {
195	{ RT_VENDORID, RT_DEVICEID_8129, RL_8129,
196		"RealTek 8129 10/100BaseTX" },
197	{ RT_VENDORID, RT_DEVICEID_8139, RL_8139,
198		"RealTek 8139 10/100BaseTX" },
199	{ RT_VENDORID, RT_DEVICEID_8169, RL_8169,
200		"RealTek 8169 10/100/1000BaseTX" },
201	{ RT_VENDORID, RT_DEVICEID_8138, RL_8139,
202		"RealTek 8139 10/100BaseTX CardBus" },
203	{ RT_VENDORID, RT_DEVICEID_8100, RL_8139,
204		"RealTek 8100 10/100BaseTX" },
205	{ ACCTON_VENDORID, ACCTON_DEVICEID_5030, RL_8139,
206		"Accton MPX 5030/5038 10/100BaseTX" },
207	{ DELTA_VENDORID, DELTA_DEVICEID_8139, RL_8139,
208		"Delta Electronics 8139 10/100BaseTX" },
209	{ ADDTRON_VENDORID, ADDTRON_DEVICEID_8139, RL_8139,
210		"Addtron Technolgy 8139 10/100BaseTX" },
211	{ DLINK_VENDORID, DLINK_DEVICEID_530TXPLUS, RL_8139,
212		"D-Link DFE-530TX+ 10/100BaseTX" },
213	{ DLINK_VENDORID, DLINK_DEVICEID_690TXD, RL_8139,
214		"D-Link DFE-690TXD 10/100BaseTX" },
215	{ NORTEL_VENDORID, ACCTON_DEVICEID_5030, RL_8139,
216		"Nortel Networks 10/100BaseTX" },
217	{ COREGA_VENDORID, COREGA_DEVICEID_FETHERCBTXD, RL_8139,
218		"Corega FEther CB-TXD" },
219	{ COREGA_VENDORID, COREGA_DEVICEID_FETHERIICBTXD, RL_8139,
220		"Corega FEtherII CB-TXD" },
221		/* XXX what type of realtek is PEPPERCON_DEVICEID_ROLF ? */
222	{ PEPPERCON_VENDORID, PEPPERCON_DEVICEID_ROLF, RL_8139,
223		"Peppercon AG ROL-F" },
224	{ PLANEX_VENDORID, PLANEX_DEVICEID_FNW3800TX, RL_8139,
225		"Planex FNW-3800-TX" },
226	{ CP_VENDORID, RT_DEVICEID_8139, RL_8139,
227		"Compaq HNE-300" },
228	{ LEVEL1_VENDORID, LEVEL1_DEVICEID_FPC0106TX, RL_8139,
229		"LevelOne FPC-0106TX" },
230	{ EDIMAX_VENDORID, EDIMAX_DEVICEID_EP4103DL, RL_8139,
231		"Edimax EP-4103DL CardBus" },
232	{ 0, 0, 0, NULL }
233};
234
235static struct rl_hwrev rl_hwrevs[] = {
236	{ RL_HWREV_8139, RL_8139,  "" },
237	{ RL_HWREV_8139A, RL_8139, "A" },
238	{ RL_HWREV_8139AG, RL_8139, "A-G" },
239	{ RL_HWREV_8139B, RL_8139, "B" },
240	{ RL_HWREV_8130, RL_8139, "8130" },
241	{ RL_HWREV_8139C, RL_8139, "C" },
242	{ RL_HWREV_8139D, RL_8139, "8139D/8100B/8100C" },
243	{ RL_HWREV_8139CPLUS, RL_8139CPLUS, "C+"},
244	{ RL_HWREV_8169, RL_8169, "8169"},
245	{ RL_HWREV_8110, RL_8169, "8169S/8110S"},
246	{ RL_HWREV_8100, RL_8139, "8100"},
247	{ RL_HWREV_8101, RL_8139, "8101"},
248	{ 0, 0, NULL }
249};
250
251static int rl_probe		(device_t);
252static int rl_attach		(device_t);
253static int rl_detach		(device_t);
254
255static int rl_encap		(struct rl_softc *, struct mbuf *);
256static int rl_encapcplus	(struct rl_softc *, struct mbuf *, int *);
257
258static void rl_dma_map_addr	(void *, bus_dma_segment_t *, int, int);
259static void rl_dma_map_desc	(void *, bus_dma_segment_t *, int,
260				    bus_size_t, int);
261static int rl_allocmem		(device_t, struct rl_softc *);
262static int rl_allocmemcplus	(device_t, struct rl_softc *);
263static int rl_newbuf		(struct rl_softc *, int, struct mbuf *);
264static int rl_rx_list_init	(struct rl_softc *);
265static int rl_tx_list_init	(struct rl_softc *);
266static void rl_rxeof		(struct rl_softc *);
267static void rl_rxeofcplus	(struct rl_softc *);
268static void rl_txeof		(struct rl_softc *);
269static void rl_txeofcplus	(struct rl_softc *);
270static void rl_intr		(void *);
271static void rl_intrcplus	(void *);
272static void rl_tick		(void *);
273static void rl_start		(struct ifnet *);
274static void rl_startcplus	(struct ifnet *);
275static int rl_ioctl		(struct ifnet *, u_long, caddr_t);
276static void rl_init		(void *);
277static void rl_stop		(struct rl_softc *);
278static void rl_watchdog		(struct ifnet *);
279static int rl_suspend		(device_t);
280static int rl_resume		(device_t);
281static void rl_shutdown		(device_t);
282static int rl_ifmedia_upd	(struct ifnet *);
283static void rl_ifmedia_sts	(struct ifnet *, struct ifmediareq *);
284
285static void rl_eeprom_putbyte	(struct rl_softc *, int);
286static void rl_eeprom_getword	(struct rl_softc *, int, u_int16_t *);
287static void rl_read_eeprom	(struct rl_softc *, caddr_t, int, int, int);
288static void rl_mii_sync		(struct rl_softc *);
289static void rl_mii_send		(struct rl_softc *, u_int32_t, int);
290static int rl_mii_readreg	(struct rl_softc *, struct rl_mii_frame *);
291static int rl_mii_writereg	(struct rl_softc *, struct rl_mii_frame *);
292static int rl_gmii_readreg	(device_t, int, int);
293static int rl_gmii_writereg	(device_t, int, int, int);
294
295static int rl_miibus_readreg	(device_t, int, int);
296static int rl_miibus_writereg	(device_t, int, int, int);
297static void rl_miibus_statchg	(device_t);
298
299static u_int8_t rl_calchash	(caddr_t);
300static void rl_setmulti		(struct rl_softc *);
301static void rl_reset		(struct rl_softc *);
302static int rl_list_tx_init	(struct rl_softc *);
303
304static void rl_dma_map_rxbuf	(void *, bus_dma_segment_t *, int, int);
305static void rl_dma_map_txbuf	(void *, bus_dma_segment_t *, int, int);
306
307#ifdef RL_USEIOSPACE
308#define RL_RES			SYS_RES_IOPORT
309#define RL_RID			RL_PCI_LOIO
310#else
311#define RL_RES			SYS_RES_MEMORY
312#define RL_RID			RL_PCI_LOMEM
313#endif
314
315static device_method_t rl_methods[] = {
316	/* Device interface */
317	DEVMETHOD(device_probe,		rl_probe),
318	DEVMETHOD(device_attach,	rl_attach),
319	DEVMETHOD(device_detach,	rl_detach),
320	DEVMETHOD(device_suspend,	rl_suspend),
321	DEVMETHOD(device_resume,	rl_resume),
322	DEVMETHOD(device_shutdown,	rl_shutdown),
323
324	/* bus interface */
325	DEVMETHOD(bus_print_child,	bus_generic_print_child),
326	DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
327
328	/* MII interface */
329	DEVMETHOD(miibus_readreg,	rl_miibus_readreg),
330	DEVMETHOD(miibus_writereg,	rl_miibus_writereg),
331	DEVMETHOD(miibus_statchg,	rl_miibus_statchg),
332
333	{ 0, 0 }
334};
335
336static driver_t rl_driver = {
337	"rl",
338	rl_methods,
339	sizeof(struct rl_softc)
340};
341
342static devclass_t rl_devclass;
343
344DRIVER_MODULE(rl, pci, rl_driver, rl_devclass, 0, 0);
345DRIVER_MODULE(rl, cardbus, rl_driver, rl_devclass, 0, 0);
346DRIVER_MODULE(miibus, rl, miibus_driver, miibus_devclass, 0, 0);
347
348#define EE_SET(x)					\
349	CSR_WRITE_1(sc, RL_EECMD,			\
350		CSR_READ_1(sc, RL_EECMD) | x)
351
352#define EE_CLR(x)					\
353	CSR_WRITE_1(sc, RL_EECMD,			\
354		CSR_READ_1(sc, RL_EECMD) & ~x)
355
356static void
357rl_dma_map_rxbuf(arg, segs, nseg, error)
358	void *arg;
359	bus_dma_segment_t *segs;
360	int nseg, error;
361{
362	struct rl_softc *sc;
363
364	sc = arg;
365	CSR_WRITE_4(sc, RL_RXADDR, segs->ds_addr & 0xFFFFFFFF);
366
367	return;
368}
369
370static void
371rl_dma_map_txbuf(arg, segs, nseg, error)
372	void *arg;
373	bus_dma_segment_t *segs;
374	int nseg, error;
375{
376	struct rl_softc *sc;
377
378	sc = arg;
379	CSR_WRITE_4(sc, RL_CUR_TXADDR(sc), segs->ds_addr & 0xFFFFFFFF);
380
381	return;
382}
383
384/*
385 * Send a read command and address to the EEPROM, check for ACK.
386 */
387static void
388rl_eeprom_putbyte(sc, addr)
389	struct rl_softc		*sc;
390	int			addr;
391{
392	register int		d, i;
393
394	d = addr | sc->rl_eecmd_read;
395
396	/*
397	 * Feed in each bit and strobe the clock.
398	 */
399	for (i = 0x400; i; i >>= 1) {
400		if (d & i) {
401			EE_SET(RL_EE_DATAIN);
402		} else {
403			EE_CLR(RL_EE_DATAIN);
404		}
405		DELAY(100);
406		EE_SET(RL_EE_CLK);
407		DELAY(150);
408		EE_CLR(RL_EE_CLK);
409		DELAY(100);
410	}
411
412	return;
413}
414
415/*
416 * Read a word of data stored in the EEPROM at address 'addr.'
417 */
418static void
419rl_eeprom_getword(sc, addr, dest)
420	struct rl_softc		*sc;
421	int			addr;
422	u_int16_t		*dest;
423{
424	register int		i;
425	u_int16_t		word = 0;
426
427	/* Enter EEPROM access mode. */
428	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_PROGRAM|RL_EE_SEL);
429
430	/*
431	 * Send address of word we want to read.
432	 */
433	rl_eeprom_putbyte(sc, addr);
434
435	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_PROGRAM|RL_EE_SEL);
436
437	/*
438	 * Start reading bits from EEPROM.
439	 */
440	for (i = 0x8000; i; i >>= 1) {
441		EE_SET(RL_EE_CLK);
442		DELAY(100);
443		if (CSR_READ_1(sc, RL_EECMD) & RL_EE_DATAOUT)
444			word |= i;
445		EE_CLR(RL_EE_CLK);
446		DELAY(100);
447	}
448
449	/* Turn off EEPROM access mode. */
450	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
451
452	*dest = word;
453
454	return;
455}
456
457/*
458 * Read a sequence of words from the EEPROM.
459 */
460static void
461rl_read_eeprom(sc, dest, off, cnt, swap)
462	struct rl_softc		*sc;
463	caddr_t			dest;
464	int			off;
465	int			cnt;
466	int			swap;
467{
468	int			i;
469	u_int16_t		word = 0, *ptr;
470
471	for (i = 0; i < cnt; i++) {
472		rl_eeprom_getword(sc, off + i, &word);
473		ptr = (u_int16_t *)(dest + (i * 2));
474		if (swap)
475			*ptr = ntohs(word);
476		else
477			*ptr = word;
478	}
479
480	return;
481}
482
483
484/*
485 * MII access routines are provided for the 8129, which
486 * doesn't have a built-in PHY. For the 8139, we fake things
487 * up by diverting rl_phy_readreg()/rl_phy_writereg() to the
488 * direct access PHY registers.
489 */
490#define MII_SET(x)					\
491	CSR_WRITE_1(sc, RL_MII,				\
492		CSR_READ_1(sc, RL_MII) | (x))
493
494#define MII_CLR(x)					\
495	CSR_WRITE_1(sc, RL_MII,				\
496		CSR_READ_1(sc, RL_MII) & ~(x))
497
498/*
499 * Sync the PHYs by setting data bit and strobing the clock 32 times.
500 */
501static void
502rl_mii_sync(sc)
503	struct rl_softc		*sc;
504{
505	register int		i;
506
507	MII_SET(RL_MII_DIR|RL_MII_DATAOUT);
508
509	for (i = 0; i < 32; i++) {
510		MII_SET(RL_MII_CLK);
511		DELAY(1);
512		MII_CLR(RL_MII_CLK);
513		DELAY(1);
514	}
515
516	return;
517}
518
519/*
520 * Clock a series of bits through the MII.
521 */
522static void
523rl_mii_send(sc, bits, cnt)
524	struct rl_softc		*sc;
525	u_int32_t		bits;
526	int			cnt;
527{
528	int			i;
529
530	MII_CLR(RL_MII_CLK);
531
532	for (i = (0x1 << (cnt - 1)); i; i >>= 1) {
533		if (bits & i) {
534			MII_SET(RL_MII_DATAOUT);
535		} else {
536			MII_CLR(RL_MII_DATAOUT);
537		}
538		DELAY(1);
539		MII_CLR(RL_MII_CLK);
540		DELAY(1);
541		MII_SET(RL_MII_CLK);
542	}
543}
544
545/*
546 * Read an PHY register through the MII.
547 */
548static int
549rl_mii_readreg(sc, frame)
550	struct rl_softc		*sc;
551	struct rl_mii_frame	*frame;
552
553{
554	int			i, ack;
555
556	RL_LOCK(sc);
557
558	/*
559	 * Set up frame for RX.
560	 */
561	frame->mii_stdelim = RL_MII_STARTDELIM;
562	frame->mii_opcode = RL_MII_READOP;
563	frame->mii_turnaround = 0;
564	frame->mii_data = 0;
565
566	CSR_WRITE_2(sc, RL_MII, 0);
567
568	/*
569	 * Turn on data xmit.
570	 */
571	MII_SET(RL_MII_DIR);
572
573	rl_mii_sync(sc);
574
575	/*
576	 * Send command/address info.
577	 */
578	rl_mii_send(sc, frame->mii_stdelim, 2);
579	rl_mii_send(sc, frame->mii_opcode, 2);
580	rl_mii_send(sc, frame->mii_phyaddr, 5);
581	rl_mii_send(sc, frame->mii_regaddr, 5);
582
583	/* Idle bit */
584	MII_CLR((RL_MII_CLK|RL_MII_DATAOUT));
585	DELAY(1);
586	MII_SET(RL_MII_CLK);
587	DELAY(1);
588
589	/* Turn off xmit. */
590	MII_CLR(RL_MII_DIR);
591
592	/* Check for ack */
593	MII_CLR(RL_MII_CLK);
594	DELAY(1);
595	ack = CSR_READ_2(sc, RL_MII) & RL_MII_DATAIN;
596	MII_SET(RL_MII_CLK);
597	DELAY(1);
598
599	/*
600	 * Now try reading data bits. If the ack failed, we still
601	 * need to clock through 16 cycles to keep the PHY(s) in sync.
602	 */
603	if (ack) {
604		for(i = 0; i < 16; i++) {
605			MII_CLR(RL_MII_CLK);
606			DELAY(1);
607			MII_SET(RL_MII_CLK);
608			DELAY(1);
609		}
610		goto fail;
611	}
612
613	for (i = 0x8000; i; i >>= 1) {
614		MII_CLR(RL_MII_CLK);
615		DELAY(1);
616		if (!ack) {
617			if (CSR_READ_2(sc, RL_MII) & RL_MII_DATAIN)
618				frame->mii_data |= i;
619			DELAY(1);
620		}
621		MII_SET(RL_MII_CLK);
622		DELAY(1);
623	}
624
625fail:
626
627	MII_CLR(RL_MII_CLK);
628	DELAY(1);
629	MII_SET(RL_MII_CLK);
630	DELAY(1);
631
632	RL_UNLOCK(sc);
633
634	if (ack)
635		return(1);
636	return(0);
637}
638
639/*
640 * Write to a PHY register through the MII.
641 */
642static int
643rl_mii_writereg(sc, frame)
644	struct rl_softc		*sc;
645	struct rl_mii_frame	*frame;
646
647{
648	RL_LOCK(sc);
649
650	/*
651	 * Set up frame for TX.
652	 */
653
654	frame->mii_stdelim = RL_MII_STARTDELIM;
655	frame->mii_opcode = RL_MII_WRITEOP;
656	frame->mii_turnaround = RL_MII_TURNAROUND;
657
658	/*
659	 * Turn on data output.
660	 */
661	MII_SET(RL_MII_DIR);
662
663	rl_mii_sync(sc);
664
665	rl_mii_send(sc, frame->mii_stdelim, 2);
666	rl_mii_send(sc, frame->mii_opcode, 2);
667	rl_mii_send(sc, frame->mii_phyaddr, 5);
668	rl_mii_send(sc, frame->mii_regaddr, 5);
669	rl_mii_send(sc, frame->mii_turnaround, 2);
670	rl_mii_send(sc, frame->mii_data, 16);
671
672	/* Idle bit. */
673	MII_SET(RL_MII_CLK);
674	DELAY(1);
675	MII_CLR(RL_MII_CLK);
676	DELAY(1);
677
678	/*
679	 * Turn off xmit.
680	 */
681	MII_CLR(RL_MII_DIR);
682
683	RL_UNLOCK(sc);
684
685	return(0);
686}
687
688static int
689rl_gmii_readreg(dev, phy, reg)
690	device_t		dev;
691	int			phy, reg;
692{
693	struct rl_softc		*sc;
694	u_int32_t		rval;
695	int			i;
696
697	if (phy != 1)
698		return(0);
699
700	sc = device_get_softc(dev);
701
702	CSR_WRITE_4(sc, RL_PHYAR, reg << 16);
703	DELAY(1000);
704
705	for (i = 0; i < RL_TIMEOUT; i++) {
706		rval = CSR_READ_4(sc, RL_PHYAR);
707		if (rval & RL_PHYAR_BUSY)
708			break;
709		DELAY(100);
710	}
711
712	if (i == RL_TIMEOUT) {
713		printf ("rl%d: PHY read failed\n", sc->rl_unit);
714		return (0);
715	}
716
717	return (rval & RL_PHYAR_PHYDATA);
718}
719
720static int
721rl_gmii_writereg(dev, phy, reg, data)
722	device_t		dev;
723	int			phy, reg, data;
724{
725	struct rl_softc		*sc;
726	u_int32_t		rval;
727	int			i;
728
729	if (phy > 0)
730		return(0);
731
732	sc = device_get_softc(dev);
733
734	CSR_WRITE_4(sc, RL_PHYAR, (reg << 16) |
735	    (data | RL_PHYAR_PHYDATA) | RL_PHYAR_BUSY);
736	DELAY(1000);
737
738	for (i = 0; i < RL_TIMEOUT; i++) {
739		rval = CSR_READ_4(sc, RL_PHYAR);
740		if (!(rval & RL_PHYAR_BUSY))
741			break;
742		DELAY(100);
743	}
744
745	if (i == RL_TIMEOUT) {
746		printf ("rl%d: PHY write failed\n", sc->rl_unit);
747		return (0);
748	}
749
750	return (0);
751}
752
753static int
754rl_miibus_readreg(dev, phy, reg)
755	device_t		dev;
756	int			phy, reg;
757{
758	struct rl_softc		*sc;
759	struct rl_mii_frame	frame;
760	u_int16_t		rval = 0;
761	u_int16_t		rl8139_reg = 0;
762
763	sc = device_get_softc(dev);
764	RL_LOCK(sc);
765
766	if (sc->rl_type == RL_8169) {
767		rval = rl_gmii_readreg(dev, phy, reg);
768		RL_UNLOCK(sc);
769		return (rval);
770	}
771
772	if (sc->rl_type == RL_8139 || sc->rl_type == RL_8139CPLUS) {
773		/* Pretend the internal PHY is only at address 0 */
774		if (phy) {
775			RL_UNLOCK(sc);
776			return(0);
777		}
778		switch(reg) {
779		case MII_BMCR:
780			rl8139_reg = RL_BMCR;
781			break;
782		case MII_BMSR:
783			rl8139_reg = RL_BMSR;
784			break;
785		case MII_ANAR:
786			rl8139_reg = RL_ANAR;
787			break;
788		case MII_ANER:
789			rl8139_reg = RL_ANER;
790			break;
791		case MII_ANLPAR:
792			rl8139_reg = RL_LPAR;
793			break;
794		case MII_PHYIDR1:
795		case MII_PHYIDR2:
796			RL_UNLOCK(sc);
797			return(0);
798		/*
799		 * Allow the rlphy driver to read the media status
800		 * register. If we have a link partner which does not
801		 * support NWAY, this is the register which will tell
802		 * us the results of parallel detection.
803		 */
804		case RL_MEDIASTAT:
805			rval = CSR_READ_1(sc, RL_MEDIASTAT);
806			RL_UNLOCK(sc);
807			return(rval);
808		default:
809			printf("rl%d: bad phy register\n", sc->rl_unit);
810			RL_UNLOCK(sc);
811			return(0);
812		}
813		rval = CSR_READ_2(sc, rl8139_reg);
814		RL_UNLOCK(sc);
815		return(rval);
816	}
817
818	bzero((char *)&frame, sizeof(frame));
819
820	frame.mii_phyaddr = phy;
821	frame.mii_regaddr = reg;
822	rl_mii_readreg(sc, &frame);
823	RL_UNLOCK(sc);
824
825	return(frame.mii_data);
826}
827
828static int
829rl_miibus_writereg(dev, phy, reg, data)
830	device_t		dev;
831	int			phy, reg, data;
832{
833	struct rl_softc		*sc;
834	struct rl_mii_frame	frame;
835	u_int16_t		rl8139_reg = 0;
836	int			rval = 0;
837
838	sc = device_get_softc(dev);
839	RL_LOCK(sc);
840
841	if (sc->rl_type == RL_8169) {
842		rval = rl_gmii_writereg(dev, phy, reg, data);
843		RL_UNLOCK(sc);
844		return (rval);
845	}
846
847	if (sc->rl_type == RL_8139 || sc->rl_type == RL_8139CPLUS) {
848		/* Pretend the internal PHY is only at address 0 */
849		if (phy) {
850			RL_UNLOCK(sc);
851			return(0);
852		}
853		switch(reg) {
854		case MII_BMCR:
855			rl8139_reg = RL_BMCR;
856			break;
857		case MII_BMSR:
858			rl8139_reg = RL_BMSR;
859			break;
860		case MII_ANAR:
861			rl8139_reg = RL_ANAR;
862			break;
863		case MII_ANER:
864			rl8139_reg = RL_ANER;
865			break;
866		case MII_ANLPAR:
867			rl8139_reg = RL_LPAR;
868			break;
869		case MII_PHYIDR1:
870		case MII_PHYIDR2:
871			RL_UNLOCK(sc);
872			return(0);
873			break;
874		default:
875			printf("rl%d: bad phy register\n", sc->rl_unit);
876			RL_UNLOCK(sc);
877			return(0);
878		}
879		CSR_WRITE_2(sc, rl8139_reg, data);
880		RL_UNLOCK(sc);
881		return(0);
882	}
883
884	bzero((char *)&frame, sizeof(frame));
885
886	frame.mii_phyaddr = phy;
887	frame.mii_regaddr = reg;
888	frame.mii_data = data;
889
890	rl_mii_writereg(sc, &frame);
891
892	RL_UNLOCK(sc);
893	return(0);
894}
895
896static void
897rl_miibus_statchg(dev)
898	device_t		dev;
899{
900	return;
901}
902
903/*
904 * Calculate CRC of a multicast group address, return the upper 6 bits.
905 */
906static u_int8_t
907rl_calchash(addr)
908	caddr_t			addr;
909{
910	u_int32_t		crc, carry;
911	int			i, j;
912	u_int8_t		c;
913
914	/* Compute CRC for the address value. */
915	crc = 0xFFFFFFFF; /* initial value */
916
917	for (i = 0; i < 6; i++) {
918		c = *(addr + i);
919		for (j = 0; j < 8; j++) {
920			carry = ((crc & 0x80000000) ? 1 : 0) ^ (c & 0x01);
921			crc <<= 1;
922			c >>= 1;
923			if (carry)
924				crc = (crc ^ 0x04c11db6) | carry;
925		}
926	}
927
928	/* return the filter bit position */
929	return(crc >> 26);
930}
931
932/*
933 * Program the 64-bit multicast hash filter.
934 */
935static void
936rl_setmulti(sc)
937	struct rl_softc		*sc;
938{
939	struct ifnet		*ifp;
940	int			h = 0;
941	u_int32_t		hashes[2] = { 0, 0 };
942	struct ifmultiaddr	*ifma;
943	u_int32_t		rxfilt;
944	int			mcnt = 0;
945
946	ifp = &sc->arpcom.ac_if;
947
948	rxfilt = CSR_READ_4(sc, RL_RXCFG);
949
950	if (ifp->if_flags & IFF_ALLMULTI || ifp->if_flags & IFF_PROMISC) {
951		rxfilt |= RL_RXCFG_RX_MULTI;
952		CSR_WRITE_4(sc, RL_RXCFG, rxfilt);
953		CSR_WRITE_4(sc, RL_MAR0, 0xFFFFFFFF);
954		CSR_WRITE_4(sc, RL_MAR4, 0xFFFFFFFF);
955		return;
956	}
957
958	/* first, zot all the existing hash bits */
959	CSR_WRITE_4(sc, RL_MAR0, 0);
960	CSR_WRITE_4(sc, RL_MAR4, 0);
961
962	/* now program new ones */
963	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
964		if (ifma->ifma_addr->sa_family != AF_LINK)
965			continue;
966		h = rl_calchash(LLADDR((struct sockaddr_dl *)ifma->ifma_addr));
967		if (h < 32)
968			hashes[0] |= (1 << h);
969		else
970			hashes[1] |= (1 << (h - 32));
971		mcnt++;
972	}
973
974	if (mcnt)
975		rxfilt |= RL_RXCFG_RX_MULTI;
976	else
977		rxfilt &= ~RL_RXCFG_RX_MULTI;
978
979	CSR_WRITE_4(sc, RL_RXCFG, rxfilt);
980	CSR_WRITE_4(sc, RL_MAR0, hashes[0]);
981	CSR_WRITE_4(sc, RL_MAR4, hashes[1]);
982
983	return;
984}
985
986static void
987rl_reset(sc)
988	struct rl_softc		*sc;
989{
990	register int		i;
991
992	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RESET);
993
994	for (i = 0; i < RL_TIMEOUT; i++) {
995		DELAY(10);
996		if (!(CSR_READ_1(sc, RL_COMMAND) & RL_CMD_RESET))
997			break;
998	}
999	if (i == RL_TIMEOUT)
1000		printf("rl%d: reset never completed!\n", sc->rl_unit);
1001
1002	CSR_WRITE_1(sc, 0x82, 1);
1003
1004	return;
1005}
1006
1007/*
1008 * Probe for a RealTek 8129/8139 chip. Check the PCI vendor and device
1009 * IDs against our list and return a device name if we find a match.
1010 */
1011static int
1012rl_probe(dev)
1013	device_t		dev;
1014{
1015	struct rl_type		*t;
1016	struct rl_softc		*sc;
1017	struct rl_hwrev		*hw_rev;
1018	int			rid;
1019	u_int32_t		hwrev;
1020	char			desc[64];
1021
1022	t = rl_devs;
1023	sc = device_get_softc(dev);
1024
1025	while(t->rl_name != NULL) {
1026		if ((pci_get_vendor(dev) == t->rl_vid) &&
1027		    (pci_get_device(dev) == t->rl_did)) {
1028
1029			/*
1030			 * Temporarily map the I/O space
1031			 * so we can read the chip ID register.
1032			 */
1033			rid = RL_RID;
1034			sc->rl_res = bus_alloc_resource(dev, RL_RES, &rid,
1035			    0, ~0, 1, RF_ACTIVE);
1036			if (sc->rl_res == NULL) {
1037				device_printf(dev,
1038				    "couldn't map ports/memory\n");
1039				return(ENXIO);
1040			}
1041			sc->rl_btag = rman_get_bustag(sc->rl_res);
1042			sc->rl_bhandle = rman_get_bushandle(sc->rl_res);
1043			mtx_init(&sc->rl_mtx,
1044			    device_get_nameunit(dev),
1045			    MTX_NETWORK_LOCK, MTX_DEF);
1046			RL_LOCK(sc);
1047			if (t->rl_basetype == RL_8139) {
1048				hwrev = CSR_READ_4(sc, RL_TXCFG) &
1049				    RL_TXCFG_HWREV;
1050				hw_rev = rl_hwrevs;
1051				while (hw_rev->rl_desc != NULL) {
1052					if (hw_rev->rl_rev == hwrev) {
1053						sprintf(desc, "%s, rev. %s",
1054						    t->rl_name,
1055						    hw_rev->rl_desc);
1056						sc->rl_type = hw_rev->rl_type;
1057						break;
1058					}
1059					hw_rev++;
1060				}
1061				if (hw_rev->rl_desc == NULL)
1062					sprintf(desc, "%s, rev. %s",
1063					    t->rl_name, "unknown");
1064			} else
1065				sprintf(desc, "%s", t->rl_name);
1066			bus_release_resource(dev, RL_RES,
1067			    RL_RID, sc->rl_res);
1068			RL_UNLOCK(sc);
1069			mtx_destroy(&sc->rl_mtx);
1070			device_set_desc_copy(dev, desc);
1071			return(0);
1072		}
1073		t++;
1074	}
1075
1076	return(ENXIO);
1077}
1078
1079/*
1080 * This routine takes the segment list provided as the result of
1081 * a bus_dma_map_load() operation and assigns the addresses/lengths
1082 * to RealTek DMA descriptors. This can be called either by the RX
1083 * code or the TX code. In the RX case, we'll probably wind up mapping
1084 * at most one segment. For the TX case, there could be any number of
1085 * segments since TX packets may span multiple mbufs. In either case,
1086 * if the number of segments is larger than the rl_maxsegs limit
1087 * specified by the caller, we abort the mapping operation. Sadly,
1088 * whoever designed the buffer mapping API did not provide a way to
1089 * return an error from here, so we have to fake it a bit.
1090 */
1091
1092static void
1093rl_dma_map_desc(arg, segs, nseg, mapsize, error)
1094	void			*arg;
1095	bus_dma_segment_t	*segs;
1096	int			nseg;
1097	bus_size_t		mapsize;
1098	int			error;
1099{
1100	struct rl_dmaload_arg	*ctx;
1101	struct rl_desc		*d = NULL;
1102	int			i = 0, idx;
1103
1104	if (error)
1105		return;
1106
1107	ctx = arg;
1108
1109	/* Signal error to caller if there's too many segments */
1110	if (nseg > ctx->rl_maxsegs) {
1111		ctx->rl_maxsegs = 0;
1112		return;
1113	}
1114
1115	/*
1116	 * Map the segment array into descriptors. Note that we set the
1117	 * start-of-frame and end-of-frame markers for either TX or RX, but
1118	 * they really only have meaning in the TX case. (In the RX case,
1119	 * it's the chip that tells us where packets begin and end.)
1120	 * We also keep track of the end of the ring and set the
1121	 * end-of-ring bits as needed, and we set the ownership bits
1122	 * in all except the very first descriptor. (The caller will
1123	 * set this descriptor later when it start transmission or
1124	 * reception.)
1125	 */
1126	idx = ctx->rl_idx;
1127	while(1) {
1128		u_int32_t		cmdstat;
1129		d = &ctx->rl_ring[idx];
1130		if (le32toh(d->rl_cmdstat) & RL_RDESC_STAT_OWN) {
1131			ctx->rl_maxsegs = 0;
1132			return;
1133		}
1134		cmdstat = segs[i].ds_len;
1135		d->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[i].ds_addr));
1136		d->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[i].ds_addr));
1137		if (i == 0)
1138			cmdstat |= RL_TDESC_CMD_SOF;
1139		else
1140			cmdstat |= RL_TDESC_CMD_OWN;
1141		if (idx == (RL_RX_DESC_CNT - 1))
1142			cmdstat |= RL_TDESC_CMD_EOR;
1143		d->rl_cmdstat = htole32(cmdstat | ctx->rl_flags);
1144		i++;
1145		if (i == nseg)
1146			break;
1147		RL_DESC_INC(idx);
1148	}
1149
1150	d->rl_cmdstat |= htole32(RL_TDESC_CMD_EOF);
1151	ctx->rl_maxsegs = nseg;
1152	ctx->rl_idx = idx;
1153
1154	return;
1155}
1156
1157/*
1158 * Map a single buffer address.
1159 */
1160
1161static void
1162rl_dma_map_addr(arg, segs, nseg, error)
1163	void			*arg;
1164	bus_dma_segment_t	*segs;
1165	int			nseg;
1166	int			error;
1167{
1168	u_int32_t		*addr;
1169
1170	if (error)
1171		return;
1172
1173	KASSERT(nseg == 1, ("too many DMA segments, %d should be 1", nseg));
1174	addr = arg;
1175	*addr = segs->ds_addr;
1176
1177	return;
1178}
1179
1180static int
1181rl_allocmem(dev, sc)
1182	device_t		dev;
1183	struct rl_softc		*sc;
1184{
1185	int error;
1186
1187	/*
1188	 * Now allocate a tag for the DMA descriptor lists.
1189	 * All of our lists are allocated as a contiguous block
1190	 * of memory.
1191	 */
1192	error = bus_dma_tag_create(sc->rl_parent_tag,	/* parent */
1193			1, 0,			/* alignment, boundary */
1194			BUS_SPACE_MAXADDR,	/* lowaddr */
1195			BUS_SPACE_MAXADDR,	/* highaddr */
1196			NULL, NULL,		/* filter, filterarg */
1197			RL_RXBUFLEN + 1518, 1,	/* maxsize,nsegments */
1198			BUS_SPACE_MAXSIZE_32BIT,/* maxsegsize */
1199			0,			/* flags */
1200			NULL, NULL,		/* lockfunc, lockarg */
1201			&sc->rl_tag);
1202	if (error)
1203		return(error);
1204
1205	/*
1206	 * Now allocate a chunk of DMA-able memory based on the
1207	 * tag we just created.
1208	 */
1209	error = bus_dmamem_alloc(sc->rl_tag,
1210	    (void **)&sc->rl_cdata.rl_rx_buf, BUS_DMA_NOWAIT,
1211	    &sc->rl_cdata.rl_rx_dmamap);
1212
1213	if (error) {
1214		printf("rl%d: no memory for list buffers!\n", sc->rl_unit);
1215		bus_dma_tag_destroy(sc->rl_tag);
1216		sc->rl_tag = NULL;
1217		return(error);
1218	}
1219
1220	/* Leave a few bytes before the start of the RX ring buffer. */
1221	sc->rl_cdata.rl_rx_buf_ptr = sc->rl_cdata.rl_rx_buf;
1222	sc->rl_cdata.rl_rx_buf += sizeof(u_int64_t);
1223
1224	return(0);
1225}
1226
1227static int
1228rl_allocmemcplus(dev, sc)
1229	device_t		dev;
1230	struct rl_softc		*sc;
1231{
1232	int			error;
1233	int			nseg;
1234	int			i;
1235
1236	/*
1237	 * Allocate map for RX mbufs.
1238	 */
1239	nseg = 32;
1240	error = bus_dma_tag_create(sc->rl_parent_tag, ETHER_ALIGN, 0,
1241	    BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1242	    NULL, MCLBYTES * nseg, nseg, MCLBYTES, 0, NULL, NULL,
1243	    &sc->rl_ldata.rl_mtag);
1244	if (error) {
1245		device_printf(dev, "could not allocate dma tag\n");
1246		return (ENOMEM);
1247	}
1248
1249	/*
1250	 * Allocate map for TX descriptor list.
1251	 */
1252	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
1253	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1254            NULL, RL_TX_LIST_SZ, 1, RL_TX_LIST_SZ, 0, NULL, NULL,
1255	    &sc->rl_ldata.rl_tx_list_tag);
1256	if (error) {
1257		device_printf(dev, "could not allocate dma tag\n");
1258		return (ENOMEM);
1259	}
1260
1261	/* Allocate DMA'able memory for the TX ring */
1262
1263        error = bus_dmamem_alloc(sc->rl_ldata.rl_tx_list_tag,
1264	    (void **)&sc->rl_ldata.rl_tx_list, BUS_DMA_NOWAIT | BUS_DMA_ZERO,
1265            &sc->rl_ldata.rl_tx_list_map);
1266        if (error)
1267                return (ENOMEM);
1268
1269	/* Load the map for the TX ring. */
1270
1271	error = bus_dmamap_load(sc->rl_ldata.rl_tx_list_tag,
1272	     sc->rl_ldata.rl_tx_list_map, sc->rl_ldata.rl_tx_list,
1273	     RL_TX_LIST_SZ, rl_dma_map_addr,
1274	     &sc->rl_ldata.rl_tx_list_addr, BUS_DMA_NOWAIT);
1275
1276	/* Create DMA maps for TX buffers */
1277
1278	for (i = 0; i < RL_TX_DESC_CNT; i++) {
1279		error = bus_dmamap_create(sc->rl_ldata.rl_mtag, 0,
1280			    &sc->rl_ldata.rl_tx_dmamap[i]);
1281		if (error) {
1282			device_printf(dev, "can't create DMA map for TX\n");
1283			return(ENOMEM);
1284		}
1285	}
1286
1287	/*
1288	 * Allocate map for RX descriptor list.
1289	 */
1290	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
1291	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1292            NULL, RL_TX_LIST_SZ, 1, RL_TX_LIST_SZ, 0, NULL, NULL,
1293	    &sc->rl_ldata.rl_rx_list_tag);
1294	if (error) {
1295		device_printf(dev, "could not allocate dma tag\n");
1296		return (ENOMEM);
1297	}
1298
1299	/* Allocate DMA'able memory for the RX ring */
1300
1301        error = bus_dmamem_alloc(sc->rl_ldata.rl_rx_list_tag,
1302	    (void **)&sc->rl_ldata.rl_rx_list, BUS_DMA_NOWAIT | BUS_DMA_ZERO,
1303            &sc->rl_ldata.rl_rx_list_map);
1304        if (error)
1305                return (ENOMEM);
1306
1307	/* Load the map for the RX ring. */
1308
1309	error = bus_dmamap_load(sc->rl_ldata.rl_rx_list_tag,
1310	     sc->rl_ldata.rl_rx_list_map, sc->rl_ldata.rl_rx_list,
1311	     RL_TX_LIST_SZ, rl_dma_map_addr,
1312	     &sc->rl_ldata.rl_rx_list_addr, BUS_DMA_NOWAIT);
1313
1314	/* Create DMA maps for RX buffers */
1315
1316	for (i = 0; i < RL_RX_DESC_CNT; i++) {
1317		error = bus_dmamap_create(sc->rl_ldata.rl_mtag, 0,
1318			    &sc->rl_ldata.rl_rx_dmamap[i]);
1319		if (error) {
1320			device_printf(dev, "can't create DMA map for RX\n");
1321			return(ENOMEM);
1322		}
1323	}
1324
1325	return(0);
1326}
1327
1328/*
1329 * Attach the interface. Allocate softc structures, do ifmedia
1330 * setup and ethernet/BPF attach.
1331 */
1332static int
1333rl_attach(dev)
1334	device_t		dev;
1335{
1336	u_char			eaddr[ETHER_ADDR_LEN];
1337	u_int16_t		as[3];
1338	struct rl_softc		*sc;
1339	struct ifnet		*ifp;
1340	struct rl_type		*t;
1341	struct rl_hwrev		*hw_rev;
1342	int			hwrev;
1343	u_int16_t		rl_did = 0;
1344	int			unit, error = 0, rid, i;
1345
1346	sc = device_get_softc(dev);
1347	unit = device_get_unit(dev);
1348
1349	mtx_init(&sc->rl_mtx, device_get_nameunit(dev), MTX_NETWORK_LOCK,
1350	    MTX_DEF | MTX_RECURSE);
1351#ifndef BURN_BRIDGES
1352	/*
1353	 * Handle power management nonsense.
1354	 */
1355
1356	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D0) {
1357		u_int32_t		iobase, membase, irq;
1358
1359		/* Save important PCI config data. */
1360		iobase = pci_read_config(dev, RL_PCI_LOIO, 4);
1361		membase = pci_read_config(dev, RL_PCI_LOMEM, 4);
1362		irq = pci_read_config(dev, RL_PCI_INTLINE, 4);
1363
1364		/* Reset the power state. */
1365		printf("rl%d: chip is is in D%d power mode "
1366		    "-- setting to D0\n", unit,
1367		    pci_get_powerstate(dev));
1368
1369		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
1370
1371		/* Restore PCI config data. */
1372		pci_write_config(dev, RL_PCI_LOIO, iobase, 4);
1373		pci_write_config(dev, RL_PCI_LOMEM, membase, 4);
1374		pci_write_config(dev, RL_PCI_INTLINE, irq, 4);
1375	}
1376#endif
1377	/*
1378	 * Map control/status registers.
1379	 */
1380	pci_enable_busmaster(dev);
1381
1382	rid = RL_RID;
1383	sc->rl_res = bus_alloc_resource(dev, RL_RES, &rid,
1384	    0, ~0, 1, RF_ACTIVE);
1385
1386	if (sc->rl_res == NULL) {
1387		printf ("rl%d: couldn't map ports/memory\n", unit);
1388		error = ENXIO;
1389		goto fail;
1390	}
1391
1392#ifdef notdef
1393	/* Detect the Realtek 8139B. For some reason, this chip is very
1394	 * unstable when left to autoselect the media
1395	 * The best workaround is to set the device to the required
1396	 * media type or to set it to the 10 Meg speed.
1397	 */
1398
1399	if ((rman_get_end(sc->rl_res)-rman_get_start(sc->rl_res))==0xff) {
1400		printf("rl%d: Realtek 8139B detected. Warning,"
1401		    " this may be unstable in autoselect mode\n", unit);
1402	}
1403#endif
1404
1405	sc->rl_btag = rman_get_bustag(sc->rl_res);
1406	sc->rl_bhandle = rman_get_bushandle(sc->rl_res);
1407
1408	/* Allocate interrupt */
1409	rid = 0;
1410	sc->rl_irq = bus_alloc_resource(dev, SYS_RES_IRQ, &rid, 0, ~0, 1,
1411	    RF_SHAREABLE | RF_ACTIVE);
1412
1413	if (sc->rl_irq == NULL) {
1414		printf("rl%d: couldn't map interrupt\n", unit);
1415		error = ENXIO;
1416		goto fail;
1417	}
1418
1419	/* Reset the adapter. */
1420	rl_reset(sc);
1421	sc->rl_eecmd_read = RL_EECMD_READ_6BIT;
1422	rl_read_eeprom(sc, (caddr_t)&rl_did, 0, 1, 0);
1423	if (rl_did != 0x8129)
1424		sc->rl_eecmd_read = RL_EECMD_READ_8BIT;
1425
1426	/*
1427	 * Get station address from the EEPROM.
1428	 */
1429	rl_read_eeprom(sc, (caddr_t)as, RL_EE_EADDR, 3, 0);
1430	for (i = 0; i < 3; i++) {
1431		eaddr[(i * 2) + 0] = as[i] & 0xff;
1432		eaddr[(i * 2) + 1] = as[i] >> 8;
1433	}
1434
1435	/*
1436	 * A RealTek chip was detected. Inform the world.
1437	 */
1438	printf("rl%d: Ethernet address: %6D\n", unit, eaddr, ":");
1439
1440	sc->rl_unit = unit;
1441	bcopy(eaddr, (char *)&sc->arpcom.ac_enaddr, ETHER_ADDR_LEN);
1442
1443	/*
1444	 * Now read the exact device type from the EEPROM to find
1445	 * out if it's an 8129 or 8139.
1446	 */
1447	rl_read_eeprom(sc, (caddr_t)&rl_did, RL_EE_PCI_DID, 1, 0);
1448
1449	t = rl_devs;
1450	while(t->rl_name != NULL) {
1451		if (rl_did == t->rl_did) {
1452			sc->rl_type = t->rl_basetype;
1453			break;
1454		}
1455		t++;
1456	}
1457	if (t->rl_name == NULL) {
1458		printf("rl%d: unknown device ID: %x\n", unit, rl_did);
1459		error = ENXIO;
1460		goto fail;
1461	}
1462	if (sc->rl_type == RL_8139) {
1463		hw_rev = rl_hwrevs;
1464		hwrev = CSR_READ_4(sc, RL_TXCFG) & RL_TXCFG_HWREV;
1465		while (hw_rev->rl_desc != NULL) {
1466			if (hw_rev->rl_rev == hwrev) {
1467				sc->rl_type = hw_rev->rl_type;
1468				break;
1469			}
1470			hw_rev++;
1471		}
1472		if (hw_rev->rl_desc == NULL) {
1473			printf("rl%d: unknown hwrev: %x\n", unit, hwrev);
1474		}
1475	} else if (rl_did == RT_DEVICEID_8129) {
1476		sc->rl_type = RL_8129;
1477	} else if (rl_did == RT_DEVICEID_8169) {
1478		sc->rl_type = RL_8169;
1479	}
1480
1481	/*
1482	 * Allocate the parent bus DMA tag appropriate for PCI.
1483	 */
1484#define RL_NSEG_NEW 32
1485	error = bus_dma_tag_create(NULL,	/* parent */
1486			1, 0,			/* alignment, boundary */
1487			BUS_SPACE_MAXADDR_32BIT,/* lowaddr */
1488			BUS_SPACE_MAXADDR,	/* highaddr */
1489			NULL, NULL,		/* filter, filterarg */
1490			MAXBSIZE, RL_NSEG_NEW,	/* maxsize, nsegments */
1491			BUS_SPACE_MAXSIZE_32BIT,/* maxsegsize */
1492			BUS_DMA_ALLOCNOW,	/* flags */
1493			NULL, NULL,		/* lockfunc, lockarg */
1494			&sc->rl_parent_tag);
1495	if (error)
1496		goto fail;
1497
1498	/*
1499	 * If this is an 8139C+ or 8169 chip, we have to allocate
1500	 * our busdma tags/memory differently. We need to allocate
1501	 * a chunk of DMA'able memory for the RX and TX descriptor
1502	 * lists.
1503	 */
1504	if (sc->rl_type == RL_8139CPLUS || sc->rl_type == RL_8169)
1505		error = rl_allocmemcplus(dev, sc);
1506	else
1507		error = rl_allocmem(dev, sc);
1508
1509	if (error)
1510		goto fail;
1511
1512	/* Do MII setup */
1513	if (mii_phy_probe(dev, &sc->rl_miibus,
1514	    rl_ifmedia_upd, rl_ifmedia_sts)) {
1515		printf("rl%d: MII without any phy!\n", sc->rl_unit);
1516		error = ENXIO;
1517		goto fail;
1518	}
1519
1520	ifp = &sc->arpcom.ac_if;
1521	ifp->if_softc = sc;
1522	ifp->if_unit = unit;
1523	ifp->if_name = "rl";
1524	ifp->if_mtu = ETHERMTU;
1525	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1526	ifp->if_ioctl = rl_ioctl;
1527	ifp->if_output = ether_output;
1528	ifp->if_capabilities = IFCAP_VLAN_MTU;
1529	if (RL_ISCPLUS(sc)) {
1530		ifp->if_start = rl_startcplus;
1531		ifp->if_hwassist = RL_CSUM_FEATURES;
1532		ifp->if_capabilities |= IFCAP_HWCSUM|IFCAP_VLAN_HWTAGGING;
1533	} else
1534		ifp->if_start = rl_start;
1535	ifp->if_watchdog = rl_watchdog;
1536	ifp->if_init = rl_init;
1537	ifp->if_baudrate = 10000000;
1538	ifp->if_snd.ifq_maxlen = RL_IFQ_MAXLEN;
1539	ifp->if_capenable = ifp->if_capabilities;
1540
1541	callout_handle_init(&sc->rl_stat_ch);
1542
1543	/*
1544	 * Call MI attach routine.
1545	 */
1546	ether_ifattach(ifp, eaddr);
1547
1548	/* Hook interrupt last to avoid having to lock softc */
1549	error = bus_setup_intr(dev, sc->rl_irq, INTR_TYPE_NET,
1550	    RL_ISCPLUS(sc) ? rl_intrcplus : rl_intr, sc, &sc->rl_intrhand);
1551
1552	if (error) {
1553		printf("rl%d: couldn't set up irq\n", unit);
1554		ether_ifdetach(ifp);
1555		goto fail;
1556	}
1557
1558fail:
1559	if (error)
1560		rl_detach(dev);
1561
1562	return (error);
1563}
1564
1565/*
1566 * Shutdown hardware and free up resources. This can be called any
1567 * time after the mutex has been initialized. It is called in both
1568 * the error case in attach and the normal detach case so it needs
1569 * to be careful about only freeing resources that have actually been
1570 * allocated.
1571 */
1572static int
1573rl_detach(dev)
1574	device_t		dev;
1575{
1576	struct rl_softc		*sc;
1577	struct ifnet		*ifp;
1578	int			i;
1579
1580	sc = device_get_softc(dev);
1581	KASSERT(mtx_initialized(&sc->rl_mtx), ("rl mutex not initialized"));
1582	RL_LOCK(sc);
1583	ifp = &sc->arpcom.ac_if;
1584
1585	/* These should only be active if attach succeeded */
1586	if (device_is_attached(dev)) {
1587		rl_stop(sc);
1588		ether_ifdetach(ifp);
1589	}
1590	if (sc->rl_miibus)
1591		device_delete_child(dev, sc->rl_miibus);
1592	bus_generic_detach(dev);
1593
1594	if (sc->rl_intrhand)
1595		bus_teardown_intr(dev, sc->rl_irq, sc->rl_intrhand);
1596	if (sc->rl_irq)
1597		bus_release_resource(dev, SYS_RES_IRQ, 0, sc->rl_irq);
1598	if (sc->rl_res)
1599		bus_release_resource(dev, RL_RES, RL_RID, sc->rl_res);
1600
1601	if (RL_ISCPLUS(sc)) {
1602
1603		/* Unload and free the RX DMA ring memory and map */
1604
1605		if (sc->rl_ldata.rl_rx_list_tag) {
1606			bus_dmamap_unload(sc->rl_ldata.rl_rx_list_tag,
1607			    sc->rl_ldata.rl_rx_list_map);
1608			bus_dmamem_free(sc->rl_ldata.rl_rx_list_tag,
1609			    sc->rl_ldata.rl_rx_list,
1610			    sc->rl_ldata.rl_rx_list_map);
1611			bus_dma_tag_destroy(sc->rl_ldata.rl_rx_list_tag);
1612		}
1613
1614		/* Unload and free the TX DMA ring memory and map */
1615
1616		if (sc->rl_ldata.rl_tx_list_tag) {
1617			bus_dmamap_unload(sc->rl_ldata.rl_tx_list_tag,
1618			    sc->rl_ldata.rl_tx_list_map);
1619			bus_dmamem_free(sc->rl_ldata.rl_tx_list_tag,
1620			    sc->rl_ldata.rl_tx_list,
1621			    sc->rl_ldata.rl_tx_list_map);
1622			bus_dma_tag_destroy(sc->rl_ldata.rl_tx_list_tag);
1623		}
1624
1625		/* Destroy all the RX and TX buffer maps */
1626
1627		if (sc->rl_ldata.rl_mtag) {
1628			for (i = 0; i < RL_TX_DESC_CNT; i++)
1629				bus_dmamap_destroy(sc->rl_ldata.rl_mtag,
1630				    sc->rl_ldata.rl_tx_dmamap[i]);
1631			for (i = 0; i < RL_RX_DESC_CNT; i++)
1632				bus_dmamap_destroy(sc->rl_ldata.rl_mtag,
1633				    sc->rl_ldata.rl_rx_dmamap[i]);
1634			bus_dma_tag_destroy(sc->rl_ldata.rl_mtag);
1635		}
1636
1637		/* Unload and free the stats buffer and map */
1638
1639		if (sc->rl_ldata.rl_stag) {
1640			bus_dmamap_unload(sc->rl_ldata.rl_stag,
1641			    sc->rl_ldata.rl_rx_list_map);
1642			bus_dmamem_free(sc->rl_ldata.rl_stag,
1643			    sc->rl_ldata.rl_stats,
1644			    sc->rl_ldata.rl_smap);
1645			bus_dma_tag_destroy(sc->rl_ldata.rl_stag);
1646		}
1647
1648	} else {
1649		if (sc->rl_tag) {
1650			bus_dmamap_unload(sc->rl_tag,
1651			    sc->rl_cdata.rl_rx_dmamap);
1652			bus_dmamem_free(sc->rl_tag, sc->rl_cdata.rl_rx_buf,
1653			    sc->rl_cdata.rl_rx_dmamap);
1654			bus_dma_tag_destroy(sc->rl_tag);
1655		}
1656	}
1657
1658	if (sc->rl_parent_tag)
1659		bus_dma_tag_destroy(sc->rl_parent_tag);
1660
1661	RL_UNLOCK(sc);
1662	mtx_destroy(&sc->rl_mtx);
1663
1664	return(0);
1665}
1666
1667/*
1668 * Initialize the transmit descriptors.
1669 */
1670static int
1671rl_list_tx_init(sc)
1672	struct rl_softc		*sc;
1673{
1674	struct rl_chain_data	*cd;
1675	int			i;
1676
1677	cd = &sc->rl_cdata;
1678	for (i = 0; i < RL_TX_LIST_CNT; i++) {
1679		cd->rl_tx_chain[i] = NULL;
1680		CSR_WRITE_4(sc,
1681		    RL_TXADDR0 + (i * sizeof(u_int32_t)), 0x0000000);
1682	}
1683
1684	sc->rl_cdata.cur_tx = 0;
1685	sc->rl_cdata.last_tx = 0;
1686
1687	return(0);
1688}
1689
1690static int
1691rl_newbuf (sc, idx, m)
1692	struct rl_softc		*sc;
1693	int			idx;
1694	struct mbuf		*m;
1695{
1696	struct rl_dmaload_arg	arg;
1697	struct mbuf		*n = NULL;
1698	int			error;
1699
1700	if (m == NULL) {
1701		n = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
1702		if (n == NULL)
1703			return(ENOBUFS);
1704		m = n;
1705	} else
1706		m->m_data = m->m_ext.ext_buf;
1707
1708	/*
1709	 * Initialize mbuf length fields and fixup
1710	 * alignment so that the frame payload is
1711	 * longword aligned.
1712	 */
1713	m->m_len = m->m_pkthdr.len = 1536;
1714	m_adj(m, ETHER_ALIGN);
1715
1716	arg.sc = sc;
1717	arg.rl_idx = idx;
1718	arg.rl_maxsegs = 1;
1719	arg.rl_flags = 0;
1720	arg.rl_ring = sc->rl_ldata.rl_rx_list;
1721
1722        error = bus_dmamap_load_mbuf(sc->rl_ldata.rl_mtag,
1723	    sc->rl_ldata.rl_rx_dmamap[idx], m, rl_dma_map_desc,
1724	    &arg, BUS_DMA_NOWAIT);
1725	if (error || arg.rl_maxsegs != 1) {
1726		if (n != NULL)
1727			m_freem(n);
1728		return (ENOMEM);
1729	}
1730
1731	sc->rl_ldata.rl_rx_list[idx].rl_cmdstat |= htole32(RL_RDESC_CMD_OWN);
1732	sc->rl_ldata.rl_rx_mbuf[idx] = m;
1733
1734        bus_dmamap_sync(sc->rl_ldata.rl_mtag,
1735	    sc->rl_ldata.rl_rx_dmamap[idx],
1736	    BUS_DMASYNC_PREREAD);
1737
1738	return(0);
1739}
1740
1741static int
1742rl_tx_list_init(sc)
1743	struct rl_softc		*sc;
1744{
1745	bzero ((char *)sc->rl_ldata.rl_tx_list, RL_TX_LIST_SZ);
1746	bzero ((char *)&sc->rl_ldata.rl_tx_mbuf,
1747	    (RL_TX_DESC_CNT * sizeof(struct mbuf *)));
1748
1749	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
1750	    sc->rl_ldata.rl_tx_list_map, BUS_DMASYNC_PREWRITE);
1751	sc->rl_ldata.rl_tx_prodidx = 0;
1752	sc->rl_ldata.rl_tx_considx = 0;
1753	sc->rl_ldata.rl_tx_free = RL_TX_DESC_CNT;
1754
1755	return(0);
1756}
1757
1758static int
1759rl_rx_list_init(sc)
1760	struct rl_softc		*sc;
1761{
1762	int			i;
1763
1764	bzero ((char *)sc->rl_ldata.rl_rx_list, RL_RX_LIST_SZ);
1765	bzero ((char *)&sc->rl_ldata.rl_rx_mbuf,
1766	    (RL_RX_DESC_CNT * sizeof(struct mbuf *)));
1767
1768	for (i = 0; i < RL_RX_DESC_CNT; i++) {
1769		if (rl_newbuf(sc, i, NULL) == ENOBUFS)
1770			return(ENOBUFS);
1771	}
1772
1773	/* Flush the RX descriptors */
1774
1775	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1776	    sc->rl_ldata.rl_rx_list_map,
1777	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
1778
1779	sc->rl_ldata.rl_rx_prodidx = 0;
1780
1781	return(0);
1782}
1783
1784/*
1785 * RX handler for C+. This is pretty much like any other
1786 * descriptor-based RX handler.
1787 */
1788static void
1789rl_rxeofcplus(sc)
1790	struct rl_softc		*sc;
1791{
1792	struct mbuf		*m;
1793	struct ifnet		*ifp;
1794	int			i, total_len;
1795	struct rl_desc		*cur_rx;
1796	u_int32_t		rxstat, rxvlan;
1797
1798	ifp = &sc->arpcom.ac_if;
1799	i = sc->rl_ldata.rl_rx_prodidx;
1800
1801	/* Invalidate the descriptor memory */
1802
1803	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1804	    sc->rl_ldata.rl_rx_list_map,
1805	    BUS_DMASYNC_POSTREAD);
1806
1807	while (!RL_OWN(&sc->rl_ldata.rl_rx_list[i])) {
1808
1809		cur_rx = &sc->rl_ldata.rl_rx_list[i];
1810		m = sc->rl_ldata.rl_rx_mbuf[i];
1811		total_len = RL_RXBYTES(cur_rx) - ETHER_CRC_LEN;
1812		rxstat = le32toh(cur_rx->rl_cmdstat);
1813		rxvlan = le32toh(cur_rx->rl_vlanctl);
1814
1815		/* Invalidate the RX mbuf and unload its map */
1816
1817		bus_dmamap_sync(sc->rl_ldata.rl_mtag,
1818		    sc->rl_ldata.rl_rx_dmamap[i],
1819		    BUS_DMASYNC_POSTREAD);
1820		bus_dmamap_unload(sc->rl_ldata.rl_mtag,
1821		    sc->rl_ldata.rl_rx_dmamap[i]);
1822
1823		/*
1824		 * NOTE: For some reason that I can't comprehend,
1825		 * the RealTek engineers decided not to implement
1826		 * the 'frame alignment error' bit in the 8169's
1827		 * status word. Unfortunately, rather than simply
1828		 * mark the bit as 'reserved,' they took it away
1829		 * completely and shifted the other status bits
1830		 * over one slot. The OWN, EOR, FS and LS bits are
1831		 * still in the same places, as is the frame length
1832		 * field. We have already extracted the frame length
1833		 * and checked the OWN bit, so to work around this
1834		 * problem, we shift the status bits one space to
1835		 * the right so that we can evaluate everything else
1836		 * correctly.
1837		 */
1838		if (sc->rl_type == RL_8169)
1839			rxstat >>= 1;
1840
1841		if (rxstat & RL_RDESC_STAT_RXERRSUM) {
1842			ifp->if_ierrors++;
1843			rl_newbuf(sc, i, m);
1844			RL_DESC_INC(i);
1845			continue;
1846		}
1847
1848		/*
1849		 * If allocating a replacement mbuf fails,
1850		 * reload the current one.
1851		 */
1852
1853		if (rl_newbuf(sc, i, NULL)) {
1854			ifp->if_ierrors++;
1855			rl_newbuf(sc, i, m);
1856			RL_DESC_INC(i);
1857			continue;
1858		}
1859
1860		RL_DESC_INC(i);
1861
1862		ifp->if_ipackets++;
1863		m->m_pkthdr.len = m->m_len = total_len;
1864		m->m_pkthdr.rcvif = ifp;
1865
1866		/* Do RX checksumming if enabled */
1867
1868		if (ifp->if_capenable & IFCAP_RXCSUM) {
1869
1870			/* Check IP header checksum */
1871			if (rxstat & RL_RDESC_STAT_PROTOID)
1872				m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED;
1873			if (!(rxstat & RL_RDESC_STAT_IPSUMBAD))
1874				m->m_pkthdr.csum_flags |= CSUM_IP_VALID;
1875
1876			/* Check TCP/UDP checksum */
1877			if ((RL_TCPPKT(rxstat) &&
1878			    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
1879			    (RL_UDPPKT(rxstat) &&
1880			    !(rxstat & RL_RDESC_STAT_UDPSUMBAD))) {
1881				m->m_pkthdr.csum_flags |=
1882				    CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
1883				m->m_pkthdr.csum_data = 0xffff;
1884			}
1885		}
1886
1887		if (rxvlan & RL_RDESC_VLANCTL_TAG)
1888			VLAN_INPUT_TAG(ifp, m,
1889			    ntohs((rxvlan & RL_RDESC_VLANCTL_DATA)), continue);
1890		(*ifp->if_input)(ifp, m);
1891	}
1892
1893	/* Flush the RX DMA ring */
1894
1895	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
1896	    sc->rl_ldata.rl_rx_list_map,
1897	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
1898
1899	sc->rl_ldata.rl_rx_prodidx = i;
1900
1901	return;
1902}
1903
1904/*
1905 * A frame has been uploaded: pass the resulting mbuf chain up to
1906 * the higher level protocols.
1907 *
1908 * You know there's something wrong with a PCI bus-master chip design
1909 * when you have to use m_devget().
1910 *
1911 * The receive operation is badly documented in the datasheet, so I'll
1912 * attempt to document it here. The driver provides a buffer area and
1913 * places its base address in the RX buffer start address register.
1914 * The chip then begins copying frames into the RX buffer. Each frame
1915 * is preceded by a 32-bit RX status word which specifies the length
1916 * of the frame and certain other status bits. Each frame (starting with
1917 * the status word) is also 32-bit aligned. The frame length is in the
1918 * first 16 bits of the status word; the lower 15 bits correspond with
1919 * the 'rx status register' mentioned in the datasheet.
1920 *
1921 * Note: to make the Alpha happy, the frame payload needs to be aligned
1922 * on a 32-bit boundary. To achieve this, we pass RL_ETHER_ALIGN (2 bytes)
1923 * as the offset argument to m_devget().
1924 */
1925static void
1926rl_rxeof(sc)
1927	struct rl_softc		*sc;
1928{
1929	struct mbuf		*m;
1930	struct ifnet		*ifp;
1931	int			total_len = 0;
1932	u_int32_t		rxstat;
1933	caddr_t			rxbufpos;
1934	int			wrap = 0;
1935	u_int16_t		cur_rx;
1936	u_int16_t		limit;
1937	u_int16_t		rx_bytes = 0, max_bytes;
1938
1939	ifp = &sc->arpcom.ac_if;
1940
1941	bus_dmamap_sync(sc->rl_tag, sc->rl_cdata.rl_rx_dmamap,
1942	    BUS_DMASYNC_POSTREAD);
1943
1944	cur_rx = (CSR_READ_2(sc, RL_CURRXADDR) + 16) % RL_RXBUFLEN;
1945
1946	/* Do not try to read past this point. */
1947	limit = CSR_READ_2(sc, RL_CURRXBUF) % RL_RXBUFLEN;
1948
1949	if (limit < cur_rx)
1950		max_bytes = (RL_RXBUFLEN - cur_rx) + limit;
1951	else
1952		max_bytes = limit - cur_rx;
1953
1954	while((CSR_READ_1(sc, RL_COMMAND) & RL_CMD_EMPTY_RXBUF) == 0) {
1955#ifdef DEVICE_POLLING
1956		if (ifp->if_flags & IFF_POLLING) {
1957			if (sc->rxcycles <= 0)
1958				break;
1959			sc->rxcycles--;
1960		}
1961#endif /* DEVICE_POLLING */
1962		rxbufpos = sc->rl_cdata.rl_rx_buf + cur_rx;
1963		rxstat = le32toh(*(u_int32_t *)rxbufpos);
1964
1965		/*
1966		 * Here's a totally undocumented fact for you. When the
1967		 * RealTek chip is in the process of copying a packet into
1968		 * RAM for you, the length will be 0xfff0. If you spot a
1969		 * packet header with this value, you need to stop. The
1970		 * datasheet makes absolutely no mention of this and
1971		 * RealTek should be shot for this.
1972		 */
1973		if ((u_int16_t)(rxstat >> 16) == RL_RXSTAT_UNFINISHED)
1974			break;
1975
1976		if (!(rxstat & RL_RXSTAT_RXOK)) {
1977			ifp->if_ierrors++;
1978			rl_init(sc);
1979			return;
1980		}
1981
1982		/* No errors; receive the packet. */
1983		total_len = rxstat >> 16;
1984		rx_bytes += total_len + 4;
1985
1986		/*
1987		 * XXX The RealTek chip includes the CRC with every
1988		 * received frame, and there's no way to turn this
1989		 * behavior off (at least, I can't find anything in
1990		 * the manual that explains how to do it) so we have
1991		 * to trim off the CRC manually.
1992		 */
1993		total_len -= ETHER_CRC_LEN;
1994
1995		/*
1996		 * Avoid trying to read more bytes than we know
1997		 * the chip has prepared for us.
1998		 */
1999		if (rx_bytes > max_bytes)
2000			break;
2001
2002		rxbufpos = sc->rl_cdata.rl_rx_buf +
2003			((cur_rx + sizeof(u_int32_t)) % RL_RXBUFLEN);
2004
2005		if (rxbufpos == (sc->rl_cdata.rl_rx_buf + RL_RXBUFLEN))
2006			rxbufpos = sc->rl_cdata.rl_rx_buf;
2007
2008		wrap = (sc->rl_cdata.rl_rx_buf + RL_RXBUFLEN) - rxbufpos;
2009
2010		if (total_len > wrap) {
2011			m = m_devget(rxbufpos, total_len, RL_ETHER_ALIGN, ifp,
2012			    NULL);
2013			if (m == NULL) {
2014				ifp->if_ierrors++;
2015			} else {
2016				m_copyback(m, wrap, total_len - wrap,
2017					sc->rl_cdata.rl_rx_buf);
2018			}
2019			cur_rx = (total_len - wrap + ETHER_CRC_LEN);
2020		} else {
2021			m = m_devget(rxbufpos, total_len, RL_ETHER_ALIGN, ifp,
2022			    NULL);
2023			if (m == NULL) {
2024				ifp->if_ierrors++;
2025			}
2026			cur_rx += total_len + 4 + ETHER_CRC_LEN;
2027		}
2028
2029		/*
2030		 * Round up to 32-bit boundary.
2031		 */
2032		cur_rx = (cur_rx + 3) & ~3;
2033		CSR_WRITE_2(sc, RL_CURRXADDR, cur_rx - 16);
2034
2035		if (m == NULL)
2036			continue;
2037
2038		ifp->if_ipackets++;
2039		(*ifp->if_input)(ifp, m);
2040	}
2041
2042	return;
2043}
2044
2045static void
2046rl_txeofcplus(sc)
2047	struct rl_softc		*sc;
2048{
2049	struct ifnet		*ifp;
2050	u_int32_t		txstat;
2051	int			idx;
2052
2053	ifp = &sc->arpcom.ac_if;
2054	idx = sc->rl_ldata.rl_tx_considx;
2055
2056	/* Invalidate the TX descriptor list */
2057
2058	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2059	    sc->rl_ldata.rl_tx_list_map,
2060	    BUS_DMASYNC_POSTREAD);
2061
2062	while (idx != sc->rl_ldata.rl_tx_prodidx) {
2063
2064		txstat = le32toh(sc->rl_ldata.rl_tx_list[idx].rl_cmdstat);
2065		if (txstat & RL_TDESC_CMD_OWN)
2066			break;
2067
2068		/*
2069		 * We only stash mbufs in the last descriptor
2070		 * in a fragment chain, which also happens to
2071		 * be the only place where the TX status bits
2072		 * are valid.
2073		 */
2074
2075		if (txstat & RL_TDESC_CMD_EOF) {
2076			m_freem(sc->rl_ldata.rl_tx_mbuf[idx]);
2077			sc->rl_ldata.rl_tx_mbuf[idx] = NULL;
2078			bus_dmamap_unload(sc->rl_ldata.rl_mtag,
2079			    sc->rl_ldata.rl_tx_dmamap[idx]);
2080			if (txstat & (RL_TDESC_STAT_EXCESSCOL|
2081			    RL_TDESC_STAT_COLCNT))
2082				ifp->if_collisions++;
2083			if (txstat & RL_TDESC_STAT_TXERRSUM)
2084				ifp->if_oerrors++;
2085			else
2086				ifp->if_opackets++;
2087		}
2088		sc->rl_ldata.rl_tx_free++;
2089		RL_DESC_INC(idx);
2090	}
2091
2092	/* No changes made to the TX ring, so no flush needed */
2093
2094	if (idx != sc->rl_ldata.rl_tx_considx) {
2095		sc->rl_ldata.rl_tx_considx = idx;
2096		ifp->if_flags &= ~IFF_OACTIVE;
2097		ifp->if_timer = 0;
2098	}
2099
2100	return;
2101}
2102
2103/*
2104 * A frame was downloaded to the chip. It's safe for us to clean up
2105 * the list buffers.
2106 */
2107static void
2108rl_txeof(sc)
2109	struct rl_softc		*sc;
2110{
2111	struct ifnet		*ifp;
2112	u_int32_t		txstat;
2113
2114	ifp = &sc->arpcom.ac_if;
2115
2116	/*
2117	 * Go through our tx list and free mbufs for those
2118	 * frames that have been uploaded.
2119	 */
2120	do {
2121		txstat = CSR_READ_4(sc, RL_LAST_TXSTAT(sc));
2122		if (!(txstat & (RL_TXSTAT_TX_OK|
2123		    RL_TXSTAT_TX_UNDERRUN|RL_TXSTAT_TXABRT)))
2124			break;
2125
2126		ifp->if_collisions += (txstat & RL_TXSTAT_COLLCNT) >> 24;
2127
2128		if (RL_LAST_TXMBUF(sc) != NULL) {
2129			bus_dmamap_unload(sc->rl_tag, RL_LAST_DMAMAP(sc));
2130			bus_dmamap_destroy(sc->rl_tag, RL_LAST_DMAMAP(sc));
2131			m_freem(RL_LAST_TXMBUF(sc));
2132			RL_LAST_TXMBUF(sc) = NULL;
2133		}
2134		if (txstat & RL_TXSTAT_TX_OK)
2135			ifp->if_opackets++;
2136		else {
2137			int			oldthresh;
2138			ifp->if_oerrors++;
2139			if ((txstat & RL_TXSTAT_TXABRT) ||
2140			    (txstat & RL_TXSTAT_OUTOFWIN))
2141				CSR_WRITE_4(sc, RL_TXCFG, RL_TXCFG_CONFIG);
2142			oldthresh = sc->rl_txthresh;
2143			/* error recovery */
2144			rl_reset(sc);
2145			rl_init(sc);
2146			/*
2147			 * If there was a transmit underrun,
2148			 * bump the TX threshold.
2149			 */
2150			if (txstat & RL_TXSTAT_TX_UNDERRUN)
2151				sc->rl_txthresh = oldthresh + 32;
2152			return;
2153		}
2154		RL_INC(sc->rl_cdata.last_tx);
2155		ifp->if_flags &= ~IFF_OACTIVE;
2156	} while (sc->rl_cdata.last_tx != sc->rl_cdata.cur_tx);
2157
2158	ifp->if_timer =
2159	    (sc->rl_cdata.last_tx == sc->rl_cdata.cur_tx) ? 0 : 5;
2160
2161	return;
2162}
2163
2164static void
2165rl_tick(xsc)
2166	void			*xsc;
2167{
2168	struct rl_softc		*sc;
2169	struct mii_data		*mii;
2170
2171	sc = xsc;
2172	RL_LOCK(sc);
2173	mii = device_get_softc(sc->rl_miibus);
2174
2175	mii_tick(mii);
2176
2177	sc->rl_stat_ch = timeout(rl_tick, sc, hz);
2178	RL_UNLOCK(sc);
2179
2180	return;
2181}
2182
2183#ifdef DEVICE_POLLING
2184static void
2185rl_poll (struct ifnet *ifp, enum poll_cmd cmd, int count)
2186{
2187	struct rl_softc *sc = ifp->if_softc;
2188
2189	RL_LOCK(sc);
2190	if (cmd == POLL_DEREGISTER) { /* final call, enable interrupts */
2191		if (RL_ISCPLUS(sc))
2192			CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2193		else
2194			CSR_WRITE_2(sc, RL_IMR, RL_INTRS);
2195		goto done;
2196	}
2197
2198	sc->rxcycles = count;
2199	if (RL_ISCPLUS(sc)) {
2200		rl_rxeofcplus(sc);
2201		rl_txeofcplus(sc);
2202	} else {
2203		rl_rxeof(sc);
2204		rl_txeof(sc);
2205	}
2206
2207	if (ifp->if_snd.ifq_head != NULL)
2208		(*ifp->if_start)(ifp);
2209
2210	if (cmd == POLL_AND_CHECK_STATUS) { /* also check status register */
2211		u_int16_t       status;
2212
2213		status = CSR_READ_2(sc, RL_ISR);
2214		if (status == 0xffff)
2215			goto done;
2216		if (status)
2217			CSR_WRITE_2(sc, RL_ISR, status);
2218
2219		/*
2220		 * XXX check behaviour on receiver stalls.
2221		 */
2222
2223		if (status & RL_ISR_SYSTEM_ERR) {
2224			rl_reset(sc);
2225			rl_init(sc);
2226		}
2227	}
2228done:
2229	RL_UNLOCK(sc);
2230}
2231#endif /* DEVICE_POLLING */
2232
2233static void
2234rl_intrcplus(arg)
2235	void			*arg;
2236{
2237	struct rl_softc		*sc;
2238	struct ifnet		*ifp;
2239	u_int16_t		status;
2240
2241	sc = arg;
2242
2243	if (sc->suspended) {
2244		return;
2245	}
2246
2247	RL_LOCK(sc);
2248	ifp = &sc->arpcom.ac_if;
2249
2250#ifdef DEVICE_POLLING
2251	if  (ifp->if_flags & IFF_POLLING)
2252		goto done;
2253	if (ether_poll_register(rl_poll, ifp)) { /* ok, disable interrupts */
2254		CSR_WRITE_2(sc, RL_IMR, 0x0000);
2255		rl_poll(ifp, 0, 1);
2256		goto done;
2257	}
2258#endif /* DEVICE_POLLING */
2259
2260	for (;;) {
2261
2262		status = CSR_READ_2(sc, RL_ISR);
2263		/* If the card has gone away the read returns 0xffff. */
2264		if (status == 0xffff)
2265			break;
2266		if (status)
2267			CSR_WRITE_2(sc, RL_ISR, status);
2268
2269		if ((status & RL_INTRS_CPLUS) == 0)
2270			break;
2271
2272		if (status & RL_ISR_RX_OK)
2273			rl_rxeofcplus(sc);
2274
2275		if (status & RL_ISR_RX_ERR)
2276			rl_rxeofcplus(sc);
2277
2278		if ((status & RL_ISR_TIMEOUT_EXPIRED) ||
2279		    (status & RL_ISR_TX_ERR) ||
2280		    (status & RL_ISR_TX_DESC_UNAVAIL))
2281			rl_txeofcplus(sc);
2282
2283		if (status & RL_ISR_SYSTEM_ERR) {
2284			rl_reset(sc);
2285			rl_init(sc);
2286		}
2287
2288	}
2289
2290	if (ifp->if_snd.ifq_head != NULL)
2291		(*ifp->if_start)(ifp);
2292
2293#ifdef DEVICE_POLLING
2294done:
2295#endif
2296	RL_UNLOCK(sc);
2297
2298	return;
2299}
2300
2301static void
2302rl_intr(arg)
2303	void			*arg;
2304{
2305	struct rl_softc		*sc;
2306	struct ifnet		*ifp;
2307	u_int16_t		status;
2308
2309	sc = arg;
2310
2311	if (sc->suspended) {
2312		return;
2313	}
2314
2315	RL_LOCK(sc);
2316	ifp = &sc->arpcom.ac_if;
2317
2318#ifdef DEVICE_POLLING
2319	if  (ifp->if_flags & IFF_POLLING)
2320		goto done;
2321	if (ether_poll_register(rl_poll, ifp)) { /* ok, disable interrupts */
2322		CSR_WRITE_2(sc, RL_IMR, 0x0000);
2323		rl_poll(ifp, 0, 1);
2324		goto done;
2325	}
2326#endif /* DEVICE_POLLING */
2327
2328	for (;;) {
2329
2330		status = CSR_READ_2(sc, RL_ISR);
2331		/* If the card has gone away the read returns 0xffff. */
2332		if (status == 0xffff)
2333			break;
2334		if (status)
2335			CSR_WRITE_2(sc, RL_ISR, status);
2336
2337		if ((status & RL_INTRS) == 0)
2338			break;
2339
2340		if (status & RL_ISR_RX_OK)
2341			rl_rxeof(sc);
2342
2343		if (status & RL_ISR_RX_ERR)
2344			rl_rxeof(sc);
2345
2346		if ((status & RL_ISR_TX_OK) || (status & RL_ISR_TX_ERR))
2347			rl_txeof(sc);
2348
2349		if (status & RL_ISR_SYSTEM_ERR) {
2350			rl_reset(sc);
2351			rl_init(sc);
2352		}
2353
2354	}
2355
2356	if (ifp->if_snd.ifq_head != NULL)
2357		(*ifp->if_start)(ifp);
2358
2359#ifdef DEVICE_POLLING
2360done:
2361#endif
2362	RL_UNLOCK(sc);
2363
2364	return;
2365}
2366
2367static int
2368rl_encapcplus(sc, m_head, idx)
2369	struct rl_softc		*sc;
2370	struct mbuf		*m_head;
2371	int			*idx;
2372{
2373	struct mbuf		*m_new = NULL;
2374	struct rl_dmaload_arg	arg;
2375	bus_dmamap_t		map;
2376	int			error;
2377	struct m_tag		*mtag;
2378
2379	if (sc->rl_ldata.rl_tx_free < 4)
2380		return(EFBIG);
2381
2382	/*
2383	 * Set up checksum offload. Note: checksum offload bits must
2384	 * appear in all descriptors of a multi-descriptor transmit
2385	 * attempt. (This is according to testing done with an 8169
2386	 * chip. I'm not sure if this is a requirement or a bug.)
2387	 */
2388
2389	arg.rl_flags = 0;
2390
2391	if (m_head->m_pkthdr.csum_flags & CSUM_IP)
2392		arg.rl_flags |= RL_TDESC_CMD_IPCSUM;
2393	if (m_head->m_pkthdr.csum_flags & CSUM_TCP)
2394		arg.rl_flags |= RL_TDESC_CMD_TCPCSUM;
2395	if (m_head->m_pkthdr.csum_flags & CSUM_UDP)
2396		arg.rl_flags |= RL_TDESC_CMD_UDPCSUM;
2397
2398	arg.sc = sc;
2399	arg.rl_idx = *idx;
2400	arg.rl_maxsegs = sc->rl_ldata.rl_tx_free;
2401	arg.rl_ring = sc->rl_ldata.rl_tx_list;
2402
2403	map = sc->rl_ldata.rl_tx_dmamap[*idx];
2404	error = bus_dmamap_load_mbuf(sc->rl_ldata.rl_mtag, map,
2405	    m_head, rl_dma_map_desc, &arg, BUS_DMA_NOWAIT);
2406
2407	if (error && error != EFBIG) {
2408		printf("rl%d: can't map mbuf (error %d)\n", sc->rl_unit, error);
2409		return(ENOBUFS);
2410	}
2411
2412	/* Too many segments to map, coalesce into a single mbuf */
2413
2414	if (error || arg.rl_maxsegs == 0) {
2415		m_new = m_defrag(m_head, M_DONTWAIT);
2416		if (m_new == NULL)
2417			return(1);
2418		else
2419			m_head = m_new;
2420
2421		arg.sc = sc;
2422		arg.rl_idx = *idx;
2423		arg.rl_maxsegs = sc->rl_ldata.rl_tx_free;
2424		arg.rl_ring = sc->rl_ldata.rl_tx_list;
2425
2426		error = bus_dmamap_load_mbuf(sc->rl_ldata.rl_mtag, map,
2427		    m_head, rl_dma_map_desc, &arg, BUS_DMA_NOWAIT);
2428		if (error) {
2429			printf("rl%d: can't map mbuf (error %d)\n",
2430			    sc->rl_unit, error);
2431			return(EFBIG);
2432		}
2433	}
2434
2435	/*
2436	 * Insure that the map for this transmission
2437	 * is placed at the array index of the last descriptor
2438	 * in this chain.
2439	 */
2440	sc->rl_ldata.rl_tx_dmamap[*idx] =
2441	    sc->rl_ldata.rl_tx_dmamap[arg.rl_idx];
2442	sc->rl_ldata.rl_tx_dmamap[arg.rl_idx] = map;
2443
2444	sc->rl_ldata.rl_tx_mbuf[arg.rl_idx] = m_head;
2445	sc->rl_ldata.rl_tx_free -= arg.rl_maxsegs;
2446
2447	/*
2448	 * Set up hardware VLAN tagging. Note: vlan tag info must
2449	 * appear in the first descriptor of a multi-descriptor
2450	 * transmission attempt.
2451	 */
2452
2453	mtag = VLAN_OUTPUT_TAG(&sc->arpcom.ac_if, m_head);
2454	if (mtag != NULL)
2455		sc->rl_ldata.rl_tx_list[*idx].rl_vlanctl =
2456		    htole32(htons(VLAN_TAG_VALUE(mtag)) | RL_TDESC_VLANCTL_TAG);
2457
2458	/* Transfer ownership of packet to the chip. */
2459
2460	sc->rl_ldata.rl_tx_list[arg.rl_idx].rl_cmdstat |=
2461	    htole32(RL_TDESC_CMD_OWN);
2462	if (*idx != arg.rl_idx)
2463		sc->rl_ldata.rl_tx_list[*idx].rl_cmdstat |=
2464		    htole32(RL_TDESC_CMD_OWN);
2465
2466	RL_DESC_INC(arg.rl_idx);
2467	*idx = arg.rl_idx;
2468
2469	return(0);
2470}
2471
2472/*
2473 * Main transmit routine for C+ and gigE NICs.
2474 */
2475
2476static void
2477rl_startcplus(ifp)
2478	struct ifnet		*ifp;
2479{
2480	struct rl_softc		*sc;
2481	struct mbuf		*m_head = NULL;
2482	int			idx;
2483
2484	sc = ifp->if_softc;
2485	RL_LOCK(sc);
2486
2487	idx = sc->rl_ldata.rl_tx_prodidx;
2488
2489	while (sc->rl_ldata.rl_tx_mbuf[idx] == NULL) {
2490		IF_DEQUEUE(&ifp->if_snd, m_head);
2491		if (m_head == NULL)
2492			break;
2493
2494		if (rl_encapcplus(sc, m_head, &idx)) {
2495			IF_PREPEND(&ifp->if_snd, m_head);
2496			ifp->if_flags |= IFF_OACTIVE;
2497			break;
2498		}
2499
2500		/*
2501		 * If there's a BPF listener, bounce a copy of this frame
2502		 * to him.
2503		 */
2504		BPF_MTAP(ifp, m_head);
2505	}
2506
2507	/* Flush the TX descriptors */
2508
2509	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2510	    sc->rl_ldata.rl_tx_list_map,
2511	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2512
2513	sc->rl_ldata.rl_tx_prodidx = idx;
2514
2515	/*
2516	 * RealTek put the TX poll request register in a different
2517	 * location on the 8169 gigE chip. I don't know why.
2518	 */
2519
2520	if (sc->rl_type == RL_8169)
2521		CSR_WRITE_2(sc, RL_GTXSTART, RL_TXSTART_START);
2522	else
2523		CSR_WRITE_2(sc, RL_TXSTART, RL_TXSTART_START);
2524
2525	/*
2526	 * Use the countdown timer for interrupt moderation.
2527	 * 'TX done' interrupts are disabled. Instead, we reset the
2528	 * countdown timer, which will begin counting until it hits
2529	 * the value in the TIMERINT register, and then trigger an
2530	 * interrupt. Each time we write to the TIMERCNT register,
2531	 * the timer count is reset to 0.
2532	 */
2533	CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2534
2535	RL_UNLOCK(sc);
2536
2537	/*
2538	 * Set a timeout in case the chip goes out to lunch.
2539	 */
2540	ifp->if_timer = 5;
2541
2542	return;
2543}
2544
2545/*
2546 * Encapsulate an mbuf chain in a descriptor by coupling the mbuf data
2547 * pointers to the fragment pointers.
2548 */
2549static int
2550rl_encap(sc, m_head)
2551	struct rl_softc		*sc;
2552	struct mbuf		*m_head;
2553{
2554	struct mbuf		*m_new = NULL;
2555
2556	/*
2557	 * The RealTek is brain damaged and wants longword-aligned
2558	 * TX buffers, plus we can only have one fragment buffer
2559	 * per packet. We have to copy pretty much all the time.
2560	 */
2561	m_new = m_defrag(m_head, M_DONTWAIT);
2562
2563	if (m_new == NULL) {
2564		m_freem(m_head);
2565		return(1);
2566	}
2567	m_head = m_new;
2568
2569	/* Pad frames to at least 60 bytes. */
2570	if (m_head->m_pkthdr.len < RL_MIN_FRAMELEN) {
2571		/*
2572		 * Make security concious people happy: zero out the
2573		 * bytes in the pad area, since we don't know what
2574		 * this mbuf cluster buffer's previous user might
2575		 * have left in it.
2576		 */
2577		bzero(mtod(m_head, char *) + m_head->m_pkthdr.len,
2578		     RL_MIN_FRAMELEN - m_head->m_pkthdr.len);
2579		m_head->m_pkthdr.len +=
2580		    (RL_MIN_FRAMELEN - m_head->m_pkthdr.len);
2581		m_head->m_len = m_head->m_pkthdr.len;
2582	}
2583
2584	RL_CUR_TXMBUF(sc) = m_head;
2585
2586	return(0);
2587}
2588
2589/*
2590 * Main transmit routine.
2591 */
2592
2593static void
2594rl_start(ifp)
2595	struct ifnet		*ifp;
2596{
2597	struct rl_softc		*sc;
2598	struct mbuf		*m_head = NULL;
2599
2600	sc = ifp->if_softc;
2601	RL_LOCK(sc);
2602
2603	while(RL_CUR_TXMBUF(sc) == NULL) {
2604		IF_DEQUEUE(&ifp->if_snd, m_head);
2605		if (m_head == NULL)
2606			break;
2607
2608		if (rl_encap(sc, m_head)) {
2609			break;
2610		}
2611
2612		/*
2613		 * If there's a BPF listener, bounce a copy of this frame
2614		 * to him.
2615		 */
2616		BPF_MTAP(ifp, RL_CUR_TXMBUF(sc));
2617
2618		/*
2619		 * Transmit the frame.
2620		 */
2621		bus_dmamap_create(sc->rl_tag, 0, &RL_CUR_DMAMAP(sc));
2622		bus_dmamap_load(sc->rl_tag, RL_CUR_DMAMAP(sc),
2623		    mtod(RL_CUR_TXMBUF(sc), void *),
2624		    RL_CUR_TXMBUF(sc)->m_pkthdr.len, rl_dma_map_txbuf,
2625		    sc, BUS_DMA_NOWAIT);
2626		bus_dmamap_sync(sc->rl_tag, RL_CUR_DMAMAP(sc),
2627		    BUS_DMASYNC_PREREAD);
2628		CSR_WRITE_4(sc, RL_CUR_TXSTAT(sc),
2629		    RL_TXTHRESH(sc->rl_txthresh) |
2630		    RL_CUR_TXMBUF(sc)->m_pkthdr.len);
2631
2632		RL_INC(sc->rl_cdata.cur_tx);
2633
2634		/*
2635		 * Set a timeout in case the chip goes out to lunch.
2636		 */
2637		ifp->if_timer = 5;
2638	}
2639
2640	/*
2641	 * We broke out of the loop because all our TX slots are
2642	 * full. Mark the NIC as busy until it drains some of the
2643	 * packets from the queue.
2644	 */
2645	if (RL_CUR_TXMBUF(sc) != NULL)
2646		ifp->if_flags |= IFF_OACTIVE;
2647
2648	RL_UNLOCK(sc);
2649
2650	return;
2651}
2652
2653static void
2654rl_init(xsc)
2655	void			*xsc;
2656{
2657	struct rl_softc		*sc = xsc;
2658	struct ifnet		*ifp = &sc->arpcom.ac_if;
2659	struct mii_data		*mii;
2660	u_int32_t		rxcfg = 0;
2661
2662	RL_LOCK(sc);
2663	mii = device_get_softc(sc->rl_miibus);
2664
2665	/*
2666	 * Cancel pending I/O and free all RX/TX buffers.
2667	 */
2668	rl_stop(sc);
2669
2670	/*
2671	 * Init our MAC address.  Even though the chipset
2672	 * documentation doesn't mention it, we need to enter "Config
2673	 * register write enable" mode to modify the ID registers.
2674	 */
2675	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG);
2676	CSR_WRITE_4(sc, RL_IDR0, *(u_int32_t *)(&sc->arpcom.ac_enaddr[0]));
2677	CSR_WRITE_4(sc, RL_IDR4, *(u_int32_t *)(&sc->arpcom.ac_enaddr[4]));
2678	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
2679
2680	/*
2681	 * For C+ mode, initialize the RX descriptors and mbufs.
2682	 */
2683	if (RL_ISCPLUS(sc)) {
2684		rl_rx_list_init(sc);
2685		rl_tx_list_init(sc);
2686	} else {
2687
2688		/* Init the RX buffer pointer register. */
2689		bus_dmamap_load(sc->rl_tag, sc->rl_cdata.rl_rx_dmamap,
2690		    sc->rl_cdata.rl_rx_buf, RL_RXBUFLEN,
2691		    rl_dma_map_rxbuf, sc, BUS_DMA_NOWAIT);
2692		bus_dmamap_sync(sc->rl_tag, sc->rl_cdata.rl_rx_dmamap,
2693		    BUS_DMASYNC_PREWRITE);
2694
2695		/* Init TX descriptors. */
2696		rl_list_tx_init(sc);
2697	}
2698
2699	/*
2700	 * Enable transmit and receive.
2701	 */
2702	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
2703
2704	/*
2705	 * Set the initial TX and RX configuration.
2706	 */
2707	CSR_WRITE_4(sc, RL_TXCFG, RL_TXCFG_CONFIG);
2708	CSR_WRITE_4(sc, RL_RXCFG, RL_RXCFG_CONFIG);
2709
2710	/* Set the individual bit to receive frames for this host only. */
2711	rxcfg = CSR_READ_4(sc, RL_RXCFG);
2712	rxcfg |= RL_RXCFG_RX_INDIV;
2713
2714	/* If we want promiscuous mode, set the allframes bit. */
2715	if (ifp->if_flags & IFF_PROMISC) {
2716		rxcfg |= RL_RXCFG_RX_ALLPHYS;
2717		CSR_WRITE_4(sc, RL_RXCFG, rxcfg);
2718	} else {
2719		rxcfg &= ~RL_RXCFG_RX_ALLPHYS;
2720		CSR_WRITE_4(sc, RL_RXCFG, rxcfg);
2721	}
2722
2723	/*
2724	 * Set capture broadcast bit to capture broadcast frames.
2725	 */
2726	if (ifp->if_flags & IFF_BROADCAST) {
2727		rxcfg |= RL_RXCFG_RX_BROAD;
2728		CSR_WRITE_4(sc, RL_RXCFG, rxcfg);
2729	} else {
2730		rxcfg &= ~RL_RXCFG_RX_BROAD;
2731		CSR_WRITE_4(sc, RL_RXCFG, rxcfg);
2732	}
2733
2734	/*
2735	 * Program the multicast filter, if necessary.
2736	 */
2737	rl_setmulti(sc);
2738
2739#ifdef DEVICE_POLLING
2740	/*
2741	 * Disable interrupts if we are polling.
2742	 */
2743	if (ifp->if_flags & IFF_POLLING)
2744		CSR_WRITE_2(sc, RL_IMR, 0);
2745	else	/* otherwise ... */
2746#endif /* DEVICE_POLLING */
2747	/*
2748	 * Enable interrupts.
2749	 */
2750	if (RL_ISCPLUS(sc))
2751		CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2752	else
2753		CSR_WRITE_2(sc, RL_IMR, RL_INTRS);
2754
2755	/* Set initial TX threshold */
2756	sc->rl_txthresh = RL_TX_THRESH_INIT;
2757
2758	/* Start RX/TX process. */
2759	CSR_WRITE_4(sc, RL_MISSEDPKT, 0);
2760#ifdef notdef
2761	/* Enable receiver and transmitter. */
2762	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
2763#endif
2764	/*
2765	 * If this is a C+ capable chip, enable C+ RX and TX mode,
2766	 * and load the addresses of the RX and TX lists into the chip.
2767	 */
2768	if (RL_ISCPLUS(sc)) {
2769		CSR_WRITE_2(sc, RL_CPLUS_CMD, RL_CPLUSCMD_RXENB|
2770		    RL_CPLUSCMD_TXENB|RL_CPLUSCMD_PCI_MRW|
2771		    RL_CPLUSCMD_VLANSTRIP|
2772		    (ifp->if_capenable & IFCAP_RXCSUM ?
2773		    RL_CPLUSCMD_RXCSUM_ENB : 0));
2774
2775		CSR_WRITE_4(sc, RL_RXLIST_ADDR_HI,
2776		    RL_ADDR_HI(sc->rl_ldata.rl_rx_list_addr));
2777		CSR_WRITE_4(sc, RL_RXLIST_ADDR_LO,
2778		    RL_ADDR_LO(sc->rl_ldata.rl_rx_list_addr));
2779
2780		CSR_WRITE_4(sc, RL_TXLIST_ADDR_HI,
2781		    RL_ADDR_HI(sc->rl_ldata.rl_tx_list_addr));
2782		CSR_WRITE_4(sc, RL_TXLIST_ADDR_LO,
2783		    RL_ADDR_LO(sc->rl_ldata.rl_tx_list_addr));
2784
2785		CSR_WRITE_1(sc, RL_EARLY_TX_THRESH, RL_EARLYTXTHRESH_CNT);
2786
2787		/*
2788		 * Initialize the timer interrupt register so that
2789		 * a timer interrupt will be generated once the timer
2790		 * reaches a certain number of ticks. The timer is
2791		 * reloaded on each transmit. This gives us TX interrupt
2792		 * moderation, which dramatically improves TX frame rate.
2793		 */
2794
2795		if (sc->rl_type == RL_8169)
2796			CSR_WRITE_4(sc, RL_TIMERINT_8169, 0x800);
2797		else
2798			CSR_WRITE_4(sc, RL_TIMERINT, 0x400);
2799
2800		/*
2801		 * For 8169 gigE NICs, set the max allowed RX packet
2802		 * size so we can receive jumbo frames.
2803		 */
2804		if (sc->rl_type == RL_8169)
2805			CSR_WRITE_2(sc, RL_MAXRXPKTLEN, RL_PKTSZ(16384));
2806
2807	}
2808
2809	mii_mediachg(mii);
2810
2811	CSR_WRITE_1(sc, RL_CFG1, RL_CFG1_DRVLOAD|RL_CFG1_FULLDUPLEX);
2812
2813	ifp->if_flags |= IFF_RUNNING;
2814	ifp->if_flags &= ~IFF_OACTIVE;
2815
2816	sc->rl_stat_ch = timeout(rl_tick, sc, hz);
2817	RL_UNLOCK(sc);
2818
2819	return;
2820}
2821
2822/*
2823 * Set media options.
2824 */
2825static int
2826rl_ifmedia_upd(ifp)
2827	struct ifnet		*ifp;
2828{
2829	struct rl_softc		*sc;
2830	struct mii_data		*mii;
2831
2832	sc = ifp->if_softc;
2833	mii = device_get_softc(sc->rl_miibus);
2834	mii_mediachg(mii);
2835
2836	return(0);
2837}
2838
2839/*
2840 * Report current media status.
2841 */
2842static void
2843rl_ifmedia_sts(ifp, ifmr)
2844	struct ifnet		*ifp;
2845	struct ifmediareq	*ifmr;
2846{
2847	struct rl_softc		*sc;
2848	struct mii_data		*mii;
2849
2850	sc = ifp->if_softc;
2851	mii = device_get_softc(sc->rl_miibus);
2852
2853	mii_pollstat(mii);
2854	ifmr->ifm_active = mii->mii_media_active;
2855	ifmr->ifm_status = mii->mii_media_status;
2856
2857	return;
2858}
2859
2860static int
2861rl_ioctl(ifp, command, data)
2862	struct ifnet		*ifp;
2863	u_long			command;
2864	caddr_t			data;
2865{
2866	struct rl_softc		*sc = ifp->if_softc;
2867	struct ifreq		*ifr = (struct ifreq *) data;
2868	struct mii_data		*mii;
2869	int			error = 0;
2870
2871	RL_LOCK(sc);
2872
2873	switch(command) {
2874	case SIOCSIFFLAGS:
2875		if (ifp->if_flags & IFF_UP) {
2876			rl_init(sc);
2877		} else {
2878			if (ifp->if_flags & IFF_RUNNING)
2879				rl_stop(sc);
2880		}
2881		error = 0;
2882		break;
2883	case SIOCADDMULTI:
2884	case SIOCDELMULTI:
2885		rl_setmulti(sc);
2886		error = 0;
2887		break;
2888	case SIOCGIFMEDIA:
2889	case SIOCSIFMEDIA:
2890		mii = device_get_softc(sc->rl_miibus);
2891		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, command);
2892		break;
2893	case SIOCSIFCAP:
2894		ifp->if_capenable = ifr->ifr_reqcap;
2895		if (ifp->if_capenable & IFCAP_TXCSUM)
2896			ifp->if_hwassist = RL_CSUM_FEATURES;
2897		else
2898			ifp->if_hwassist = 0;
2899		if (ifp->if_flags & IFF_RUNNING)
2900			rl_init(sc);
2901		break;
2902	default:
2903		error = ether_ioctl(ifp, command, data);
2904		break;
2905	}
2906
2907	RL_UNLOCK(sc);
2908
2909	return(error);
2910}
2911
2912static void
2913rl_watchdog(ifp)
2914	struct ifnet		*ifp;
2915{
2916	struct rl_softc		*sc;
2917
2918	sc = ifp->if_softc;
2919	RL_LOCK(sc);
2920	printf("rl%d: watchdog timeout\n", sc->rl_unit);
2921	ifp->if_oerrors++;
2922
2923	if (RL_ISCPLUS(sc)) {
2924		rl_txeofcplus(sc);
2925		rl_rxeofcplus(sc);
2926	} else {
2927		rl_txeof(sc);
2928		rl_rxeof(sc);
2929	}
2930
2931	rl_init(sc);
2932
2933	RL_UNLOCK(sc);
2934
2935	return;
2936}
2937
2938/*
2939 * Stop the adapter and free any mbufs allocated to the
2940 * RX and TX lists.
2941 */
2942static void
2943rl_stop(sc)
2944	struct rl_softc		*sc;
2945{
2946	register int		i;
2947	struct ifnet		*ifp;
2948
2949	RL_LOCK(sc);
2950	ifp = &sc->arpcom.ac_if;
2951	ifp->if_timer = 0;
2952
2953	untimeout(rl_tick, sc, sc->rl_stat_ch);
2954	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
2955#ifdef DEVICE_POLLING
2956	ether_poll_deregister(ifp);
2957#endif /* DEVICE_POLLING */
2958
2959	CSR_WRITE_1(sc, RL_COMMAND, 0x00);
2960	CSR_WRITE_2(sc, RL_IMR, 0x0000);
2961
2962	if (RL_ISCPLUS(sc)) {
2963
2964		/* Free the TX list buffers. */
2965
2966		for (i = 0; i < RL_TX_DESC_CNT; i++) {
2967			if (sc->rl_ldata.rl_tx_mbuf[i] != NULL) {
2968				bus_dmamap_unload(sc->rl_ldata.rl_mtag,
2969				    sc->rl_ldata.rl_tx_dmamap[i]);
2970				m_freem(sc->rl_ldata.rl_tx_mbuf[i]);
2971				sc->rl_ldata.rl_tx_mbuf[i] = NULL;
2972			}
2973		}
2974
2975		/* Free the RX list buffers. */
2976
2977		for (i = 0; i < RL_RX_DESC_CNT; i++) {
2978			if (sc->rl_ldata.rl_rx_mbuf[i] != NULL) {
2979				bus_dmamap_unload(sc->rl_ldata.rl_mtag,
2980				    sc->rl_ldata.rl_rx_dmamap[i]);
2981				m_freem(sc->rl_ldata.rl_rx_mbuf[i]);
2982				sc->rl_ldata.rl_rx_mbuf[i] = NULL;
2983			}
2984		}
2985
2986	} else {
2987
2988		bus_dmamap_unload(sc->rl_tag, sc->rl_cdata.rl_rx_dmamap);
2989
2990		/*
2991		 * Free the TX list buffers.
2992		 */
2993		for (i = 0; i < RL_TX_LIST_CNT; i++) {
2994			if (sc->rl_cdata.rl_tx_chain[i] != NULL) {
2995				bus_dmamap_unload(sc->rl_tag,
2996				    sc->rl_cdata.rl_tx_dmamap[i]);
2997				bus_dmamap_destroy(sc->rl_tag,
2998				    sc->rl_cdata.rl_tx_dmamap[i]);
2999				m_freem(sc->rl_cdata.rl_tx_chain[i]);
3000				sc->rl_cdata.rl_tx_chain[i] = NULL;
3001				CSR_WRITE_4(sc, RL_TXADDR0 + i, 0x0000000);
3002			}
3003		}
3004	}
3005
3006	RL_UNLOCK(sc);
3007	return;
3008}
3009
3010/*
3011 * Device suspend routine.  Stop the interface and save some PCI
3012 * settings in case the BIOS doesn't restore them properly on
3013 * resume.
3014 */
3015static int
3016rl_suspend(dev)
3017	device_t		dev;
3018{
3019	register int		i;
3020	struct rl_softc		*sc;
3021
3022	sc = device_get_softc(dev);
3023
3024	rl_stop(sc);
3025
3026	for (i = 0; i < 5; i++)
3027		sc->saved_maps[i] = pci_read_config(dev, PCIR_MAPS + i * 4, 4);
3028	sc->saved_biosaddr = pci_read_config(dev, PCIR_BIOS, 4);
3029	sc->saved_intline = pci_read_config(dev, PCIR_INTLINE, 1);
3030	sc->saved_cachelnsz = pci_read_config(dev, PCIR_CACHELNSZ, 1);
3031	sc->saved_lattimer = pci_read_config(dev, PCIR_LATTIMER, 1);
3032
3033	sc->suspended = 1;
3034
3035	return (0);
3036}
3037
3038/*
3039 * Device resume routine.  Restore some PCI settings in case the BIOS
3040 * doesn't, re-enable busmastering, and restart the interface if
3041 * appropriate.
3042 */
3043static int
3044rl_resume(dev)
3045	device_t		dev;
3046{
3047	register int		i;
3048	struct rl_softc		*sc;
3049	struct ifnet		*ifp;
3050
3051	sc = device_get_softc(dev);
3052	ifp = &sc->arpcom.ac_if;
3053
3054	/* better way to do this? */
3055	for (i = 0; i < 5; i++)
3056		pci_write_config(dev, PCIR_MAPS + i * 4, sc->saved_maps[i], 4);
3057	pci_write_config(dev, PCIR_BIOS, sc->saved_biosaddr, 4);
3058	pci_write_config(dev, PCIR_INTLINE, sc->saved_intline, 1);
3059	pci_write_config(dev, PCIR_CACHELNSZ, sc->saved_cachelnsz, 1);
3060	pci_write_config(dev, PCIR_LATTIMER, sc->saved_lattimer, 1);
3061
3062	/* reenable busmastering */
3063	pci_enable_busmaster(dev);
3064	pci_enable_io(dev, RL_RES);
3065
3066	/* reinitialize interface if necessary */
3067	if (ifp->if_flags & IFF_UP)
3068		rl_init(sc);
3069
3070	sc->suspended = 0;
3071
3072	return (0);
3073}
3074
3075/*
3076 * Stop all chip I/O so that the kernel's probe routines don't
3077 * get confused by errant DMAs when rebooting.
3078 */
3079static void
3080rl_shutdown(dev)
3081	device_t		dev;
3082{
3083	struct rl_softc		*sc;
3084
3085	sc = device_get_softc(dev);
3086
3087	rl_stop(sc);
3088
3089	return;
3090}
3091