1/* $NetBSD: if_vge.c,v 1.51 2010/04/05 07:20:27 joerg Exp $ */
2
3/*-
4 * Copyright (c) 2004
5 *	Bill Paul <wpaul@windriver.com>.  All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 *    notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 *    notice, this list of conditions and the following disclaimer in the
14 *    documentation and/or other materials provided with the distribution.
15 * 3. All advertising materials mentioning features or use of this software
16 *    must display the following acknowledgement:
17 *	This product includes software developed by Bill Paul.
18 * 4. Neither the name of the author nor the names of any co-contributors
19 *    may be used to endorse or promote products derived from this software
20 *    without specific prior written permission.
21 *
22 * THIS SOFTWARE IS PROVIDED BY Bill Paul AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED.  IN NO EVENT SHALL Bill Paul OR THE VOICES IN HIS HEAD
26 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
27 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
28 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
29 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
30 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
31 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
32 * THE POSSIBILITY OF SUCH DAMAGE.
33 *
34 * FreeBSD: src/sys/dev/vge/if_vge.c,v 1.5 2005/02/07 19:39:29 glebius Exp
35 */
36
37#include <sys/cdefs.h>
38__KERNEL_RCSID(0, "$NetBSD: if_vge.c,v 1.51 2010/04/05 07:20:27 joerg Exp $");
39
40/*
41 * VIA Networking Technologies VT612x PCI gigabit ethernet NIC driver.
42 *
43 * Written by Bill Paul <wpaul@windriver.com>
44 * Senior Networking Software Engineer
45 * Wind River Systems
46 */
47
48/*
49 * The VIA Networking VT6122 is a 32bit, 33/66 MHz PCI device that
50 * combines a tri-speed ethernet MAC and PHY, with the following
51 * features:
52 *
53 *	o Jumbo frame support up to 16K
54 *	o Transmit and receive flow control
55 *	o IPv4 checksum offload
56 *	o VLAN tag insertion and stripping
57 *	o TCP large send
58 *	o 64-bit multicast hash table filter
59 *	o 64 entry CAM filter
60 *	o 16K RX FIFO and 48K TX FIFO memory
61 *	o Interrupt moderation
62 *
63 * The VT6122 supports up to four transmit DMA queues. The descriptors
64 * in the transmit ring can address up to 7 data fragments; frames which
65 * span more than 7 data buffers must be coalesced, but in general the
66 * BSD TCP/IP stack rarely generates frames more than 2 or 3 fragments
67 * long. The receive descriptors address only a single buffer.
68 *
69 * There are two peculiar design issues with the VT6122. One is that
70 * receive data buffers must be aligned on a 32-bit boundary. This is
71 * not a problem where the VT6122 is used as a LOM device in x86-based
72 * systems, but on architectures that generate unaligned access traps, we
73 * have to do some copying.
74 *
75 * The other issue has to do with the way 64-bit addresses are handled.
76 * The DMA descriptors only allow you to specify 48 bits of addressing
77 * information. The remaining 16 bits are specified using one of the
78 * I/O registers. If you only have a 32-bit system, then this isn't
79 * an issue, but if you have a 64-bit system and more than 4GB of
80 * memory, you must have to make sure your network data buffers reside
81 * in the same 48-bit 'segment.'
82 *
83 * Special thanks to Ryan Fu at VIA Networking for providing documentation
84 * and sample NICs for testing.
85 */
86
87
88#include <sys/param.h>
89#include <sys/endian.h>
90#include <sys/systm.h>
91#include <sys/device.h>
92#include <sys/sockio.h>
93#include <sys/mbuf.h>
94#include <sys/malloc.h>
95#include <sys/kernel.h>
96#include <sys/socket.h>
97
98#include <net/if.h>
99#include <net/if_arp.h>
100#include <net/if_ether.h>
101#include <net/if_dl.h>
102#include <net/if_media.h>
103
104#include <net/bpf.h>
105
106#include <sys/bus.h>
107
108#include <dev/mii/mii.h>
109#include <dev/mii/miivar.h>
110
111#include <dev/pci/pcireg.h>
112#include <dev/pci/pcivar.h>
113#include <dev/pci/pcidevs.h>
114
115#include <dev/pci/if_vgereg.h>
116
117#define VGE_IFQ_MAXLEN		64
118
119#define VGE_RING_ALIGN		256
120
121#define VGE_NTXDESC		256
122#define VGE_NTXDESC_MASK	(VGE_NTXDESC - 1)
123#define VGE_NEXT_TXDESC(x)	((x + 1) & VGE_NTXDESC_MASK)
124#define VGE_PREV_TXDESC(x)	((x - 1) & VGE_NTXDESC_MASK)
125
126#define VGE_NRXDESC		256	/* Must be a multiple of 4!! */
127#define VGE_NRXDESC_MASK	(VGE_NRXDESC - 1)
128#define VGE_NEXT_RXDESC(x)	((x + 1) & VGE_NRXDESC_MASK)
129#define VGE_PREV_RXDESC(x)	((x - 1) & VGE_NRXDESC_MASK)
130
131#define VGE_ADDR_LO(y)		((uint64_t)(y) & 0xFFFFFFFF)
132#define VGE_ADDR_HI(y)		((uint64_t)(y) >> 32)
133#define VGE_BUFLEN(y)		((y) & 0x7FFF)
134#define ETHER_PAD_LEN		(ETHER_MIN_LEN - ETHER_CRC_LEN)
135
136#define VGE_POWER_MANAGEMENT	0	/* disabled for now */
137
138/*
139 * Mbuf adjust factor to force 32-bit alignment of IP header.
140 * Drivers should pad ETHER_ALIGN bytes when setting up a
141 * RX mbuf so the upper layers get the IP header properly aligned
142 * past the 14-byte Ethernet header.
143 *
144 * See also comment in vge_encap().
145 */
146#define ETHER_ALIGN		2
147
148#ifdef __NO_STRICT_ALIGNMENT
149#define VGE_RX_BUFSIZE		MCLBYTES
150#else
151#define VGE_RX_PAD		sizeof(uint32_t)
152#define VGE_RX_BUFSIZE		(MCLBYTES - VGE_RX_PAD)
153#endif
154
155/*
156 * Control structures are DMA'd to the vge chip. We allocate them in
157 * a single clump that maps to a single DMA segment to make several things
158 * easier.
159 */
160struct vge_control_data {
161	/* TX descriptors */
162	struct vge_txdesc	vcd_txdescs[VGE_NTXDESC];
163	/* RX descriptors */
164	struct vge_rxdesc	vcd_rxdescs[VGE_NRXDESC];
165	/* dummy data for TX padding */
166	uint8_t			vcd_pad[ETHER_PAD_LEN];
167};
168
169#define VGE_CDOFF(x)	offsetof(struct vge_control_data, x)
170#define VGE_CDTXOFF(x)	VGE_CDOFF(vcd_txdescs[(x)])
171#define VGE_CDRXOFF(x)	VGE_CDOFF(vcd_rxdescs[(x)])
172#define VGE_CDPADOFF()	VGE_CDOFF(vcd_pad[0])
173
174/*
175 * Software state for TX jobs.
176 */
177struct vge_txsoft {
178	struct mbuf	*txs_mbuf;		/* head of our mbuf chain */
179	bus_dmamap_t	txs_dmamap;		/* our DMA map */
180};
181
182/*
183 * Software state for RX jobs.
184 */
185struct vge_rxsoft {
186	struct mbuf	*rxs_mbuf;		/* head of our mbuf chain */
187	bus_dmamap_t	rxs_dmamap;		/* our DMA map */
188};
189
190
191struct vge_softc {
192	device_t		sc_dev;
193
194	bus_space_tag_t		sc_bst;		/* bus space tag */
195	bus_space_handle_t	sc_bsh;		/* bus space handle */
196	bus_dma_tag_t		sc_dmat;
197
198	struct ethercom		sc_ethercom;	/* interface info */
199	uint8_t			sc_eaddr[ETHER_ADDR_LEN];
200
201	void			*sc_intrhand;
202	struct mii_data		sc_mii;
203	uint8_t			sc_type;
204	int			sc_if_flags;
205	int			sc_link;
206	int			sc_camidx;
207	callout_t		sc_timeout;
208
209	bus_dmamap_t		sc_cddmamap;
210#define sc_cddma		sc_cddmamap->dm_segs[0].ds_addr
211
212	struct vge_txsoft	sc_txsoft[VGE_NTXDESC];
213	struct vge_rxsoft	sc_rxsoft[VGE_NRXDESC];
214	struct vge_control_data	*sc_control_data;
215#define sc_txdescs		sc_control_data->vcd_txdescs
216#define sc_rxdescs		sc_control_data->vcd_rxdescs
217
218	int			sc_tx_prodidx;
219	int			sc_tx_considx;
220	int			sc_tx_free;
221
222	struct mbuf		*sc_rx_mhead;
223	struct mbuf		*sc_rx_mtail;
224	int			sc_rx_prodidx;
225	int			sc_rx_consumed;
226
227	int			sc_suspended;	/* 0 = normal  1 = suspended */
228	uint32_t		sc_saved_maps[5];	/* pci data */
229	uint32_t		sc_saved_biosaddr;
230	uint8_t			sc_saved_intline;
231	uint8_t			sc_saved_cachelnsz;
232	uint8_t			sc_saved_lattimer;
233};
234
235#define VGE_CDTXADDR(sc, x)	((sc)->sc_cddma + VGE_CDTXOFF(x))
236#define VGE_CDRXADDR(sc, x)	((sc)->sc_cddma + VGE_CDRXOFF(x))
237#define VGE_CDPADADDR(sc)	((sc)->sc_cddma + VGE_CDPADOFF())
238
239#define VGE_TXDESCSYNC(sc, idx, ops)					\
240	bus_dmamap_sync((sc)->sc_dmat,(sc)->sc_cddmamap,		\
241	    VGE_CDTXOFF(idx),						\
242	    offsetof(struct vge_txdesc, td_frag[0]),			\
243	    (ops))
244#define VGE_TXFRAGSYNC(sc, idx, nsegs, ops)				\
245	bus_dmamap_sync((sc)->sc_dmat, (sc)->sc_cddmamap,		\
246	    VGE_CDTXOFF(idx) +						\
247	    offsetof(struct vge_txdesc, td_frag[0]),			\
248	    sizeof(struct vge_txfrag) * (nsegs),			\
249	    (ops))
250#define VGE_RXDESCSYNC(sc, idx, ops)					\
251	bus_dmamap_sync((sc)->sc_dmat, (sc)->sc_cddmamap,		\
252	    VGE_CDRXOFF(idx),						\
253	    sizeof(struct vge_rxdesc),					\
254	    (ops))
255
256/*
257 * register space access macros
258 */
259#define CSR_WRITE_4(sc, reg, val)	\
260	bus_space_write_4((sc)->sc_bst, (sc)->sc_bsh, (reg), (val))
261#define CSR_WRITE_2(sc, reg, val)	\
262	bus_space_write_2((sc)->sc_bst, (sc)->sc_bsh, (reg), (val))
263#define CSR_WRITE_1(sc, reg, val)	\
264	bus_space_write_1((sc)->sc_bst, (sc)->sc_bsh, (reg), (val))
265
266#define CSR_READ_4(sc, reg)		\
267	bus_space_read_4((sc)->sc_bst, (sc)->sc_bsh, (reg))
268#define CSR_READ_2(sc, reg)		\
269	bus_space_read_2((sc)->sc_bst, (sc)->sc_bsh, (reg))
270#define CSR_READ_1(sc, reg)		\
271	bus_space_read_1((sc)->sc_bst, (sc)->sc_bsh, (reg))
272
273#define CSR_SETBIT_1(sc, reg, x)	\
274	CSR_WRITE_1((sc), (reg), CSR_READ_1((sc), (reg)) | (x))
275#define CSR_SETBIT_2(sc, reg, x)	\
276	CSR_WRITE_2((sc), (reg), CSR_READ_2((sc), (reg)) | (x))
277#define CSR_SETBIT_4(sc, reg, x)	\
278	CSR_WRITE_4((sc), (reg), CSR_READ_4((sc), (reg)) | (x))
279
280#define CSR_CLRBIT_1(sc, reg, x)	\
281	CSR_WRITE_1((sc), (reg), CSR_READ_1((sc), (reg)) & ~(x))
282#define CSR_CLRBIT_2(sc, reg, x)	\
283	CSR_WRITE_2((sc), (reg), CSR_READ_2((sc), (reg)) & ~(x))
284#define CSR_CLRBIT_4(sc, reg, x)	\
285	CSR_WRITE_4((sc), (reg), CSR_READ_4((sc), (reg)) & ~(x))
286
287#define VGE_TIMEOUT		10000
288
289#define VGE_PCI_LOIO             0x10
290#define VGE_PCI_LOMEM            0x14
291
292static inline void vge_set_txaddr(struct vge_txfrag *, bus_addr_t);
293static inline void vge_set_rxaddr(struct vge_rxdesc *, bus_addr_t);
294
295static int vge_ifflags_cb(struct ethercom *);
296
297static int vge_match(device_t, cfdata_t, void *);
298static void vge_attach(device_t, device_t, void *);
299
300static int vge_encap(struct vge_softc *, struct mbuf *, int);
301
302static int vge_allocmem(struct vge_softc *);
303static int vge_newbuf(struct vge_softc *, int, struct mbuf *);
304#ifndef __NO_STRICT_ALIGNMENT
305static inline void vge_fixup_rx(struct mbuf *);
306#endif
307static void vge_rxeof(struct vge_softc *);
308static void vge_txeof(struct vge_softc *);
309static int vge_intr(void *);
310static void vge_tick(void *);
311static void vge_start(struct ifnet *);
312static int vge_ioctl(struct ifnet *, u_long, void *);
313static int vge_init(struct ifnet *);
314static void vge_stop(struct ifnet *, int);
315static void vge_watchdog(struct ifnet *);
316#if VGE_POWER_MANAGEMENT
317static int vge_suspend(device_t);
318static int vge_resume(device_t);
319#endif
320static bool vge_shutdown(device_t, int);
321
322static uint16_t vge_read_eeprom(struct vge_softc *, int);
323
324static void vge_miipoll_start(struct vge_softc *);
325static void vge_miipoll_stop(struct vge_softc *);
326static int vge_miibus_readreg(device_t, int, int);
327static void vge_miibus_writereg(device_t, int, int, int);
328static void vge_miibus_statchg(device_t);
329
330static void vge_cam_clear(struct vge_softc *);
331static int vge_cam_set(struct vge_softc *, uint8_t *);
332static void vge_setmulti(struct vge_softc *);
333static void vge_reset(struct vge_softc *);
334
335CFATTACH_DECL_NEW(vge, sizeof(struct vge_softc),
336    vge_match, vge_attach, NULL, NULL);
337
338static inline void
339vge_set_txaddr(struct vge_txfrag *f, bus_addr_t daddr)
340{
341
342	f->tf_addrlo = htole32((uint32_t)daddr);
343	if (sizeof(bus_addr_t) == sizeof(uint64_t))
344		f->tf_addrhi = htole16(((uint64_t)daddr >> 32) & 0xFFFF);
345	else
346		f->tf_addrhi = 0;
347}
348
349static inline void
350vge_set_rxaddr(struct vge_rxdesc *rxd, bus_addr_t daddr)
351{
352
353	rxd->rd_addrlo = htole32((uint32_t)daddr);
354	if (sizeof(bus_addr_t) == sizeof(uint64_t))
355		rxd->rd_addrhi = htole16(((uint64_t)daddr >> 32) & 0xFFFF);
356	else
357		rxd->rd_addrhi = 0;
358}
359
360/*
361 * Defragment mbuf chain contents to be as linear as possible.
362 * Returns new mbuf chain on success, NULL on failure. Old mbuf
363 * chain is always freed.
364 * XXX temporary until there would be generic function doing this.
365 */
366#define m_defrag	vge_m_defrag
367struct mbuf * vge_m_defrag(struct mbuf *, int);
368
369struct mbuf *
370vge_m_defrag(struct mbuf *mold, int flags)
371{
372	struct mbuf *m0, *mn, *n;
373	size_t sz = mold->m_pkthdr.len;
374
375#ifdef DIAGNOSTIC
376	if ((mold->m_flags & M_PKTHDR) == 0)
377		panic("m_defrag: not a mbuf chain header");
378#endif
379
380	MGETHDR(m0, flags, MT_DATA);
381	if (m0 == NULL)
382		return NULL;
383	m0->m_pkthdr.len = mold->m_pkthdr.len;
384	mn = m0;
385
386	do {
387		if (sz > MHLEN) {
388			MCLGET(mn, M_DONTWAIT);
389			if ((mn->m_flags & M_EXT) == 0) {
390				m_freem(m0);
391				return NULL;
392			}
393		}
394
395		mn->m_len = MIN(sz, MCLBYTES);
396
397		m_copydata(mold, mold->m_pkthdr.len - sz, mn->m_len,
398		     mtod(mn, void *));
399
400		sz -= mn->m_len;
401
402		if (sz > 0) {
403			/* need more mbufs */
404			MGET(n, M_NOWAIT, MT_DATA);
405			if (n == NULL) {
406				m_freem(m0);
407				return NULL;
408			}
409
410			mn->m_next = n;
411			mn = n;
412		}
413	} while (sz > 0);
414
415	return m0;
416}
417
418/*
419 * Read a word of data stored in the EEPROM at address 'addr.'
420 */
421static uint16_t
422vge_read_eeprom(struct vge_softc *sc, int addr)
423{
424	int i;
425	uint16_t word = 0;
426
427	/*
428	 * Enter EEPROM embedded programming mode. In order to
429	 * access the EEPROM at all, we first have to set the
430	 * EELOAD bit in the CHIPCFG2 register.
431	 */
432	CSR_SETBIT_1(sc, VGE_CHIPCFG2, VGE_CHIPCFG2_EELOAD);
433	CSR_SETBIT_1(sc, VGE_EECSR, VGE_EECSR_EMBP/*|VGE_EECSR_ECS*/);
434
435	/* Select the address of the word we want to read */
436	CSR_WRITE_1(sc, VGE_EEADDR, addr);
437
438	/* Issue read command */
439	CSR_SETBIT_1(sc, VGE_EECMD, VGE_EECMD_ERD);
440
441	/* Wait for the done bit to be set. */
442	for (i = 0; i < VGE_TIMEOUT; i++) {
443		if (CSR_READ_1(sc, VGE_EECMD) & VGE_EECMD_EDONE)
444			break;
445	}
446
447	if (i == VGE_TIMEOUT) {
448		printf("%s: EEPROM read timed out\n", device_xname(sc->sc_dev));
449		return 0;
450	}
451
452	/* Read the result */
453	word = CSR_READ_2(sc, VGE_EERDDAT);
454
455	/* Turn off EEPROM access mode. */
456	CSR_CLRBIT_1(sc, VGE_EECSR, VGE_EECSR_EMBP/*|VGE_EECSR_ECS*/);
457	CSR_CLRBIT_1(sc, VGE_CHIPCFG2, VGE_CHIPCFG2_EELOAD);
458
459	return word;
460}
461
462static void
463vge_miipoll_stop(struct vge_softc *sc)
464{
465	int i;
466
467	CSR_WRITE_1(sc, VGE_MIICMD, 0);
468
469	for (i = 0; i < VGE_TIMEOUT; i++) {
470		DELAY(1);
471		if (CSR_READ_1(sc, VGE_MIISTS) & VGE_MIISTS_IIDL)
472			break;
473	}
474
475	if (i == VGE_TIMEOUT) {
476		printf("%s: failed to idle MII autopoll\n",
477		    device_xname(sc->sc_dev));
478	}
479}
480
481static void
482vge_miipoll_start(struct vge_softc *sc)
483{
484	int i;
485
486	/* First, make sure we're idle. */
487
488	CSR_WRITE_1(sc, VGE_MIICMD, 0);
489	CSR_WRITE_1(sc, VGE_MIIADDR, VGE_MIIADDR_SWMPL);
490
491	for (i = 0; i < VGE_TIMEOUT; i++) {
492		DELAY(1);
493		if (CSR_READ_1(sc, VGE_MIISTS) & VGE_MIISTS_IIDL)
494			break;
495	}
496
497	if (i == VGE_TIMEOUT) {
498		printf("%s: failed to idle MII autopoll\n",
499		    device_xname(sc->sc_dev));
500		return;
501	}
502
503	/* Now enable auto poll mode. */
504
505	CSR_WRITE_1(sc, VGE_MIICMD, VGE_MIICMD_MAUTO);
506
507	/* And make sure it started. */
508
509	for (i = 0; i < VGE_TIMEOUT; i++) {
510		DELAY(1);
511		if ((CSR_READ_1(sc, VGE_MIISTS) & VGE_MIISTS_IIDL) == 0)
512			break;
513	}
514
515	if (i == VGE_TIMEOUT) {
516		printf("%s: failed to start MII autopoll\n",
517		    device_xname(sc->sc_dev));
518	}
519}
520
521static int
522vge_miibus_readreg(device_t dev, int phy, int reg)
523{
524	struct vge_softc *sc;
525	int i, s;
526	uint16_t rval;
527
528	sc = device_private(dev);
529	rval = 0;
530	if (phy != (CSR_READ_1(sc, VGE_MIICFG) & 0x1F))
531		return 0;
532
533	s = splnet();
534	vge_miipoll_stop(sc);
535
536	/* Specify the register we want to read. */
537	CSR_WRITE_1(sc, VGE_MIIADDR, reg);
538
539	/* Issue read command. */
540	CSR_SETBIT_1(sc, VGE_MIICMD, VGE_MIICMD_RCMD);
541
542	/* Wait for the read command bit to self-clear. */
543	for (i = 0; i < VGE_TIMEOUT; i++) {
544		DELAY(1);
545		if ((CSR_READ_1(sc, VGE_MIICMD) & VGE_MIICMD_RCMD) == 0)
546			break;
547	}
548
549	if (i == VGE_TIMEOUT)
550		printf("%s: MII read timed out\n", device_xname(sc->sc_dev));
551	else
552		rval = CSR_READ_2(sc, VGE_MIIDATA);
553
554	vge_miipoll_start(sc);
555	splx(s);
556
557	return rval;
558}
559
560static void
561vge_miibus_writereg(device_t dev, int phy, int reg, int data)
562{
563	struct vge_softc *sc;
564	int i, s;
565
566	sc = device_private(dev);
567	if (phy != (CSR_READ_1(sc, VGE_MIICFG) & 0x1F))
568		return;
569
570	s = splnet();
571	vge_miipoll_stop(sc);
572
573	/* Specify the register we want to write. */
574	CSR_WRITE_1(sc, VGE_MIIADDR, reg);
575
576	/* Specify the data we want to write. */
577	CSR_WRITE_2(sc, VGE_MIIDATA, data);
578
579	/* Issue write command. */
580	CSR_SETBIT_1(sc, VGE_MIICMD, VGE_MIICMD_WCMD);
581
582	/* Wait for the write command bit to self-clear. */
583	for (i = 0; i < VGE_TIMEOUT; i++) {
584		DELAY(1);
585		if ((CSR_READ_1(sc, VGE_MIICMD) & VGE_MIICMD_WCMD) == 0)
586			break;
587	}
588
589	if (i == VGE_TIMEOUT) {
590		printf("%s: MII write timed out\n", device_xname(sc->sc_dev));
591	}
592
593	vge_miipoll_start(sc);
594	splx(s);
595}
596
597static void
598vge_cam_clear(struct vge_softc *sc)
599{
600	int i;
601
602	/*
603	 * Turn off all the mask bits. This tells the chip
604	 * that none of the entries in the CAM filter are valid.
605	 * desired entries will be enabled as we fill the filter in.
606	 */
607
608	CSR_CLRBIT_1(sc, VGE_CAMCTL, VGE_CAMCTL_PAGESEL);
609	CSR_SETBIT_1(sc, VGE_CAMCTL, VGE_PAGESEL_CAMMASK);
610	CSR_WRITE_1(sc, VGE_CAMADDR, VGE_CAMADDR_ENABLE);
611	for (i = 0; i < 8; i++)
612		CSR_WRITE_1(sc, VGE_CAM0 + i, 0);
613
614	/* Clear the VLAN filter too. */
615
616	CSR_WRITE_1(sc, VGE_CAMADDR, VGE_CAMADDR_ENABLE|VGE_CAMADDR_AVSEL|0);
617	for (i = 0; i < 8; i++)
618		CSR_WRITE_1(sc, VGE_CAM0 + i, 0);
619
620	CSR_WRITE_1(sc, VGE_CAMADDR, 0);
621	CSR_CLRBIT_1(sc, VGE_CAMCTL, VGE_CAMCTL_PAGESEL);
622	CSR_SETBIT_1(sc, VGE_CAMCTL, VGE_PAGESEL_MAR);
623
624	sc->sc_camidx = 0;
625}
626
627static int
628vge_cam_set(struct vge_softc *sc, uint8_t *addr)
629{
630	int i, error;
631
632	error = 0;
633
634	if (sc->sc_camidx == VGE_CAM_MAXADDRS)
635		return ENOSPC;
636
637	/* Select the CAM data page. */
638	CSR_CLRBIT_1(sc, VGE_CAMCTL, VGE_CAMCTL_PAGESEL);
639	CSR_SETBIT_1(sc, VGE_CAMCTL, VGE_PAGESEL_CAMDATA);
640
641	/* Set the filter entry we want to update and enable writing. */
642	CSR_WRITE_1(sc, VGE_CAMADDR, VGE_CAMADDR_ENABLE | sc->sc_camidx);
643
644	/* Write the address to the CAM registers */
645	for (i = 0; i < ETHER_ADDR_LEN; i++)
646		CSR_WRITE_1(sc, VGE_CAM0 + i, addr[i]);
647
648	/* Issue a write command. */
649	CSR_SETBIT_1(sc, VGE_CAMCTL, VGE_CAMCTL_WRITE);
650
651	/* Wake for it to clear. */
652	for (i = 0; i < VGE_TIMEOUT; i++) {
653		DELAY(1);
654		if ((CSR_READ_1(sc, VGE_CAMCTL) & VGE_CAMCTL_WRITE) == 0)
655			break;
656	}
657
658	if (i == VGE_TIMEOUT) {
659		printf("%s: setting CAM filter failed\n",
660		    device_xname(sc->sc_dev));
661		error = EIO;
662		goto fail;
663	}
664
665	/* Select the CAM mask page. */
666	CSR_CLRBIT_1(sc, VGE_CAMCTL, VGE_CAMCTL_PAGESEL);
667	CSR_SETBIT_1(sc, VGE_CAMCTL, VGE_PAGESEL_CAMMASK);
668
669	/* Set the mask bit that enables this filter. */
670	CSR_SETBIT_1(sc, VGE_CAM0 + (sc->sc_camidx / 8),
671	    1 << (sc->sc_camidx & 7));
672
673	sc->sc_camidx++;
674
675 fail:
676	/* Turn off access to CAM. */
677	CSR_WRITE_1(sc, VGE_CAMADDR, 0);
678	CSR_CLRBIT_1(sc, VGE_CAMCTL, VGE_CAMCTL_PAGESEL);
679	CSR_SETBIT_1(sc, VGE_CAMCTL, VGE_PAGESEL_MAR);
680
681	return error;
682}
683
684/*
685 * Program the multicast filter. We use the 64-entry CAM filter
686 * for perfect filtering. If there's more than 64 multicast addresses,
687 * we use the hash filter instead.
688 */
689static void
690vge_setmulti(struct vge_softc *sc)
691{
692	struct ifnet *ifp;
693	int error;
694	uint32_t h, hashes[2] = { 0, 0 };
695	struct ether_multi *enm;
696	struct ether_multistep step;
697
698	error = 0;
699	ifp = &sc->sc_ethercom.ec_if;
700
701	/* First, zot all the multicast entries. */
702	vge_cam_clear(sc);
703	CSR_WRITE_4(sc, VGE_MAR0, 0);
704	CSR_WRITE_4(sc, VGE_MAR1, 0);
705	ifp->if_flags &= ~IFF_ALLMULTI;
706
707	/*
708	 * If the user wants allmulti or promisc mode, enable reception
709	 * of all multicast frames.
710	 */
711	if (ifp->if_flags & IFF_PROMISC) {
712 allmulti:
713		CSR_WRITE_4(sc, VGE_MAR0, 0xFFFFFFFF);
714		CSR_WRITE_4(sc, VGE_MAR1, 0xFFFFFFFF);
715		ifp->if_flags |= IFF_ALLMULTI;
716		return;
717	}
718
719	/* Now program new ones */
720	ETHER_FIRST_MULTI(step, &sc->sc_ethercom, enm);
721	while (enm != NULL) {
722		/*
723		 * If multicast range, fall back to ALLMULTI.
724		 */
725		if (memcmp(enm->enm_addrlo, enm->enm_addrhi,
726				ETHER_ADDR_LEN) != 0)
727			goto allmulti;
728
729		error = vge_cam_set(sc, enm->enm_addrlo);
730		if (error)
731			break;
732
733		ETHER_NEXT_MULTI(step, enm);
734	}
735
736	/* If there were too many addresses, use the hash filter. */
737	if (error) {
738		vge_cam_clear(sc);
739
740		ETHER_FIRST_MULTI(step, &sc->sc_ethercom, enm);
741		while (enm != NULL) {
742			/*
743			 * If multicast range, fall back to ALLMULTI.
744			 */
745			if (memcmp(enm->enm_addrlo, enm->enm_addrhi,
746					ETHER_ADDR_LEN) != 0)
747				goto allmulti;
748
749			h = ether_crc32_be(enm->enm_addrlo,
750			    ETHER_ADDR_LEN) >> 26;
751			hashes[h >> 5] |= 1 << (h & 0x1f);
752
753			ETHER_NEXT_MULTI(step, enm);
754		}
755
756		CSR_WRITE_4(sc, VGE_MAR0, hashes[0]);
757		CSR_WRITE_4(sc, VGE_MAR1, hashes[1]);
758	}
759}
760
761static void
762vge_reset(struct vge_softc *sc)
763{
764	int i;
765
766	CSR_WRITE_1(sc, VGE_CRS1, VGE_CR1_SOFTRESET);
767
768	for (i = 0; i < VGE_TIMEOUT; i++) {
769		DELAY(5);
770		if ((CSR_READ_1(sc, VGE_CRS1) & VGE_CR1_SOFTRESET) == 0)
771			break;
772	}
773
774	if (i == VGE_TIMEOUT) {
775		printf("%s: soft reset timed out", device_xname(sc->sc_dev));
776		CSR_WRITE_1(sc, VGE_CRS3, VGE_CR3_STOP_FORCE);
777		DELAY(2000);
778	}
779
780	DELAY(5000);
781
782	CSR_SETBIT_1(sc, VGE_EECSR, VGE_EECSR_RELOAD);
783
784	for (i = 0; i < VGE_TIMEOUT; i++) {
785		DELAY(5);
786		if ((CSR_READ_1(sc, VGE_EECSR) & VGE_EECSR_RELOAD) == 0)
787			break;
788	}
789
790	if (i == VGE_TIMEOUT) {
791		printf("%s: EEPROM reload timed out\n",
792		    device_xname(sc->sc_dev));
793		return;
794	}
795
796	/*
797	 * On some machine, the first read data from EEPROM could be
798	 * messed up, so read one dummy data here to avoid the mess.
799	 */
800	(void)vge_read_eeprom(sc, 0);
801
802	CSR_CLRBIT_1(sc, VGE_CHIPCFG0, VGE_CHIPCFG0_PACPI);
803}
804
805/*
806 * Probe for a VIA gigabit chip. Check the PCI vendor and device
807 * IDs against our list and return a device name if we find a match.
808 */
809static int
810vge_match(device_t parent, cfdata_t match, void *aux)
811{
812	struct pci_attach_args *pa = aux;
813
814	if (PCI_VENDOR(pa->pa_id) == PCI_VENDOR_VIATECH
815	    && PCI_PRODUCT(pa->pa_id) == PCI_PRODUCT_VIATECH_VT612X)
816		return 1;
817
818	return 0;
819}
820
821static int
822vge_allocmem(struct vge_softc *sc)
823{
824	int error;
825	int nseg;
826	int i;
827	bus_dma_segment_t seg;
828
829	/*
830	 * Allocate memory for control data.
831	 */
832
833	error = bus_dmamem_alloc(sc->sc_dmat, sizeof(struct vge_control_data),
834	     VGE_RING_ALIGN, 0, &seg, 1, &nseg, BUS_DMA_NOWAIT);
835	if (error) {
836		aprint_error_dev(sc->sc_dev,
837		    "could not allocate control data dma memory\n");
838		goto fail_1;
839	}
840
841	/* Map the memory to kernel VA space */
842
843	error = bus_dmamem_map(sc->sc_dmat, &seg, nseg,
844	    sizeof(struct vge_control_data), (void **)&sc->sc_control_data,
845	    BUS_DMA_NOWAIT);
846	if (error) {
847		aprint_error_dev(sc->sc_dev,
848		    "could not map control data dma memory\n");
849		goto fail_2;
850	}
851	memset(sc->sc_control_data, 0, sizeof(struct vge_control_data));
852
853	/*
854	 * Create map for control data.
855	 */
856	error = bus_dmamap_create(sc->sc_dmat,
857	    sizeof(struct vge_control_data), 1,
858	    sizeof(struct vge_control_data), 0, BUS_DMA_NOWAIT,
859	    &sc->sc_cddmamap);
860	if (error) {
861		aprint_error_dev(sc->sc_dev,
862		    "could not create control data dmamap\n");
863		goto fail_3;
864	}
865
866	/* Load the map for the control data. */
867	error = bus_dmamap_load(sc->sc_dmat, sc->sc_cddmamap,
868	    sc->sc_control_data, sizeof(struct vge_control_data), NULL,
869	    BUS_DMA_NOWAIT);
870	if (error) {
871		aprint_error_dev(sc->sc_dev,
872		    "could not load control data dma memory\n");
873		goto fail_4;
874	}
875
876	/* Create DMA maps for TX buffers */
877
878	for (i = 0; i < VGE_NTXDESC; i++) {
879		error = bus_dmamap_create(sc->sc_dmat, VGE_TX_MAXLEN,
880		    VGE_TX_FRAGS, VGE_TX_MAXLEN, 0, BUS_DMA_NOWAIT,
881		    &sc->sc_txsoft[i].txs_dmamap);
882		if (error) {
883			aprint_error_dev(sc->sc_dev,
884			    "can't create DMA map for TX descs\n");
885			goto fail_5;
886		}
887	}
888
889	/* Create DMA maps for RX buffers */
890
891	for (i = 0; i < VGE_NRXDESC; i++) {
892		error = bus_dmamap_create(sc->sc_dmat, MCLBYTES,
893		    1, MCLBYTES, 0, BUS_DMA_NOWAIT,
894		    &sc->sc_rxsoft[i].rxs_dmamap);
895		if (error) {
896			aprint_error_dev(sc->sc_dev,
897			    "can't create DMA map for RX descs\n");
898			goto fail_6;
899		}
900		sc->sc_rxsoft[i].rxs_mbuf = NULL;
901	}
902
903	return 0;
904
905 fail_6:
906	for (i = 0; i < VGE_NRXDESC; i++) {
907		if (sc->sc_rxsoft[i].rxs_dmamap != NULL)
908			bus_dmamap_destroy(sc->sc_dmat,
909			    sc->sc_rxsoft[i].rxs_dmamap);
910	}
911 fail_5:
912	for (i = 0; i < VGE_NTXDESC; i++) {
913		if (sc->sc_txsoft[i].txs_dmamap != NULL)
914			bus_dmamap_destroy(sc->sc_dmat,
915			    sc->sc_txsoft[i].txs_dmamap);
916	}
917	bus_dmamap_unload(sc->sc_dmat, sc->sc_cddmamap);
918 fail_4:
919	bus_dmamap_destroy(sc->sc_dmat, sc->sc_cddmamap);
920 fail_3:
921	bus_dmamem_unmap(sc->sc_dmat, (void *)sc->sc_control_data,
922	    sizeof(struct vge_control_data));
923 fail_2:
924	bus_dmamem_free(sc->sc_dmat, &seg, nseg);
925 fail_1:
926	return ENOMEM;
927}
928
929/*
930 * Attach the interface. Allocate softc structures, do ifmedia
931 * setup and ethernet/BPF attach.
932 */
933static void
934vge_attach(device_t parent, device_t self, void *aux)
935{
936	uint8_t	*eaddr;
937	struct vge_softc *sc = device_private(self);
938	struct ifnet *ifp;
939	struct pci_attach_args *pa = aux;
940	pci_chipset_tag_t pc = pa->pa_pc;
941	const char *intrstr;
942	pci_intr_handle_t ih;
943	uint16_t val;
944
945	sc->sc_dev = self;
946
947	pci_aprint_devinfo_fancy(pa, NULL, "VIA VT612X Gigabit Ethernet", 1);
948
949	/* Make sure bus-mastering is enabled */
950        pci_conf_write(pc, pa->pa_tag, PCI_COMMAND_STATUS_REG,
951	    pci_conf_read(pc, pa->pa_tag, PCI_COMMAND_STATUS_REG) |
952	    PCI_COMMAND_MASTER_ENABLE);
953
954	/*
955	 * Map control/status registers.
956	 */
957	if (pci_mapreg_map(pa, VGE_PCI_LOMEM, PCI_MAPREG_TYPE_MEM, 0,
958	    &sc->sc_bst, &sc->sc_bsh, NULL, NULL) != 0) {
959		aprint_error_dev(self, "couldn't map memory\n");
960		return;
961	}
962
963        /*
964         * Map and establish our interrupt.
965         */
966	if (pci_intr_map(pa, &ih)) {
967		aprint_error_dev(self, "unable to map interrupt\n");
968		return;
969	}
970	intrstr = pci_intr_string(pc, ih);
971	sc->sc_intrhand = pci_intr_establish(pc, ih, IPL_NET, vge_intr, sc);
972	if (sc->sc_intrhand == NULL) {
973		aprint_error_dev(self, "unable to establish interrupt");
974		if (intrstr != NULL)
975			aprint_error(" at %s", intrstr);
976		aprint_error("\n");
977		return;
978	}
979	aprint_normal_dev(self, "interrupting at %s\n", intrstr);
980
981	/* Reset the adapter. */
982	vge_reset(sc);
983
984	/*
985	 * Get station address from the EEPROM.
986	 */
987	eaddr = sc->sc_eaddr;
988	val = vge_read_eeprom(sc, VGE_EE_EADDR + 0);
989	eaddr[0] = val & 0xff;
990	eaddr[1] = val >> 8;
991	val = vge_read_eeprom(sc, VGE_EE_EADDR + 1);
992	eaddr[2] = val & 0xff;
993	eaddr[3] = val >> 8;
994	val = vge_read_eeprom(sc, VGE_EE_EADDR + 2);
995	eaddr[4] = val & 0xff;
996	eaddr[5] = val >> 8;
997
998	aprint_normal_dev(self, "Ethernet address: %s\n",
999	    ether_sprintf(eaddr));
1000
1001	/*
1002	 * Use the 32bit tag. Hardware supports 48bit physical addresses,
1003	 * but we don't use that for now.
1004	 */
1005	sc->sc_dmat = pa->pa_dmat;
1006
1007	if (vge_allocmem(sc) != 0)
1008		return;
1009
1010	ifp = &sc->sc_ethercom.ec_if;
1011	ifp->if_softc = sc;
1012	strlcpy(ifp->if_xname, device_xname(self), IFNAMSIZ);
1013	ifp->if_mtu = ETHERMTU;
1014	ifp->if_baudrate = IF_Gbps(1);
1015	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1016	ifp->if_ioctl = vge_ioctl;
1017	ifp->if_start = vge_start;
1018	ifp->if_init = vge_init;
1019	ifp->if_stop = vge_stop;
1020
1021	/*
1022	 * We can support 802.1Q VLAN-sized frames and jumbo
1023	 * Ethernet frames.
1024	 */
1025	sc->sc_ethercom.ec_capabilities |=
1026	    ETHERCAP_VLAN_MTU | ETHERCAP_JUMBO_MTU |
1027	    ETHERCAP_VLAN_HWTAGGING;
1028
1029	/*
1030	 * We can do IPv4/TCPv4/UDPv4 checksums in hardware.
1031	 */
1032	ifp->if_capabilities |=
1033	    IFCAP_CSUM_IPv4_Tx | IFCAP_CSUM_IPv4_Rx |
1034	    IFCAP_CSUM_TCPv4_Tx | IFCAP_CSUM_TCPv4_Rx |
1035	    IFCAP_CSUM_UDPv4_Tx | IFCAP_CSUM_UDPv4_Rx;
1036
1037#ifdef DEVICE_POLLING
1038#ifdef IFCAP_POLLING
1039	ifp->if_capabilities |= IFCAP_POLLING;
1040#endif
1041#endif
1042	ifp->if_watchdog = vge_watchdog;
1043	IFQ_SET_MAXLEN(&ifp->if_snd, max(VGE_IFQ_MAXLEN, IFQ_MAXLEN));
1044	IFQ_SET_READY(&ifp->if_snd);
1045
1046	/*
1047	 * Initialize our media structures and probe the MII.
1048	 */
1049	sc->sc_mii.mii_ifp = ifp;
1050	sc->sc_mii.mii_readreg = vge_miibus_readreg;
1051	sc->sc_mii.mii_writereg = vge_miibus_writereg;
1052	sc->sc_mii.mii_statchg = vge_miibus_statchg;
1053
1054	sc->sc_ethercom.ec_mii = &sc->sc_mii;
1055	ifmedia_init(&sc->sc_mii.mii_media, 0, ether_mediachange,
1056	    ether_mediastatus);
1057	mii_attach(self, &sc->sc_mii, 0xffffffff, MII_PHY_ANY,
1058	    MII_OFFSET_ANY, MIIF_DOPAUSE);
1059	if (LIST_FIRST(&sc->sc_mii.mii_phys) == NULL) {
1060		ifmedia_add(&sc->sc_mii.mii_media, IFM_ETHER|IFM_NONE, 0, NULL);
1061		ifmedia_set(&sc->sc_mii.mii_media, IFM_ETHER|IFM_NONE);
1062	} else
1063		ifmedia_set(&sc->sc_mii.mii_media, IFM_ETHER|IFM_AUTO);
1064
1065	/*
1066	 * Attach the interface.
1067	 */
1068	if_attach(ifp);
1069	ether_ifattach(ifp, eaddr);
1070	ether_set_ifflags_cb(&sc->sc_ethercom, vge_ifflags_cb);
1071
1072	callout_init(&sc->sc_timeout, 0);
1073	callout_setfunc(&sc->sc_timeout, vge_tick, sc);
1074
1075	/*
1076	 * Make sure the interface is shutdown during reboot.
1077	 */
1078	if (pmf_device_register1(self, NULL, NULL, vge_shutdown))
1079		pmf_class_network_register(self, ifp);
1080	else
1081		aprint_error_dev(self, "couldn't establish power handler\n");
1082}
1083
1084static int
1085vge_newbuf(struct vge_softc *sc, int idx, struct mbuf *m)
1086{
1087	struct mbuf *m_new;
1088	struct vge_rxdesc *rxd;
1089	struct vge_rxsoft *rxs;
1090	bus_dmamap_t map;
1091	int i;
1092#ifdef DIAGNOSTIC
1093	uint32_t rd_sts;
1094#endif
1095
1096	m_new = NULL;
1097	if (m == NULL) {
1098		MGETHDR(m_new, M_DONTWAIT, MT_DATA);
1099		if (m_new == NULL)
1100			return ENOBUFS;
1101
1102		MCLGET(m_new, M_DONTWAIT);
1103		if ((m_new->m_flags & M_EXT) == 0) {
1104			m_freem(m_new);
1105			return ENOBUFS;
1106		}
1107
1108		m = m_new;
1109	} else
1110		m->m_data = m->m_ext.ext_buf;
1111
1112
1113	/*
1114	 * This is part of an evil trick to deal with non-x86 platforms.
1115	 * The VIA chip requires RX buffers to be aligned on 32-bit
1116	 * boundaries, but that will hose non-x86 machines. To get around
1117	 * this, we leave some empty space at the start of each buffer
1118	 * and for non-x86 hosts, we copy the buffer back two bytes
1119	 * to achieve word alignment. This is slightly more efficient
1120	 * than allocating a new buffer, copying the contents, and
1121	 * discarding the old buffer.
1122	 */
1123	m->m_len = m->m_pkthdr.len = VGE_RX_BUFSIZE;
1124#ifndef __NO_STRICT_ALIGNMENT
1125	m->m_data += VGE_RX_PAD;
1126#endif
1127	rxs = &sc->sc_rxsoft[idx];
1128	map = rxs->rxs_dmamap;
1129
1130	if (bus_dmamap_load_mbuf(sc->sc_dmat, map, m, BUS_DMA_NOWAIT) != 0)
1131		goto out;
1132
1133	rxd = &sc->sc_rxdescs[idx];
1134
1135#ifdef DIAGNOSTIC
1136	/* If this descriptor is still owned by the chip, bail. */
1137	VGE_RXDESCSYNC(sc, idx, BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1138	rd_sts = le32toh(rxd->rd_sts);
1139	VGE_RXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD);
1140	if (rd_sts & VGE_RDSTS_OWN) {
1141		panic("%s: tried to map busy RX descriptor",
1142		    device_xname(sc->sc_dev));
1143	}
1144#endif
1145
1146	rxs->rxs_mbuf = m;
1147	bus_dmamap_sync(sc->sc_dmat, map, 0, map->dm_mapsize,
1148	    BUS_DMASYNC_PREREAD);
1149
1150	rxd->rd_buflen =
1151	    htole16(VGE_BUFLEN(map->dm_segs[0].ds_len) | VGE_RXDESC_I);
1152	vge_set_rxaddr(rxd, map->dm_segs[0].ds_addr);
1153	rxd->rd_sts = 0;
1154	rxd->rd_ctl = 0;
1155	VGE_RXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1156
1157	/*
1158	 * Note: the manual fails to document the fact that for
1159	 * proper opration, the driver needs to replentish the RX
1160	 * DMA ring 4 descriptors at a time (rather than one at a
1161	 * time, like most chips). We can allocate the new buffers
1162	 * but we should not set the OWN bits until we're ready
1163	 * to hand back 4 of them in one shot.
1164	 */
1165
1166#define VGE_RXCHUNK 4
1167	sc->sc_rx_consumed++;
1168	if (sc->sc_rx_consumed == VGE_RXCHUNK) {
1169		for (i = idx; i != idx - VGE_RXCHUNK; i--) {
1170			KASSERT(i >= 0);
1171			sc->sc_rxdescs[i].rd_sts |= htole32(VGE_RDSTS_OWN);
1172			VGE_RXDESCSYNC(sc, i,
1173			    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1174		}
1175		sc->sc_rx_consumed = 0;
1176	}
1177
1178	return 0;
1179 out:
1180	if (m_new != NULL)
1181		m_freem(m_new);
1182	return ENOMEM;
1183}
1184
1185#ifndef __NO_STRICT_ALIGNMENT
1186static inline void
1187vge_fixup_rx(struct mbuf *m)
1188{
1189	int i;
1190	uint16_t *src, *dst;
1191
1192	src = mtod(m, uint16_t *);
1193	dst = src - 1;
1194
1195	for (i = 0; i < (m->m_len / sizeof(uint16_t) + 1); i++)
1196		*dst++ = *src++;
1197
1198	m->m_data -= ETHER_ALIGN;
1199}
1200#endif
1201
1202/*
1203 * RX handler. We support the reception of jumbo frames that have
1204 * been fragmented across multiple 2K mbuf cluster buffers.
1205 */
1206static void
1207vge_rxeof(struct vge_softc *sc)
1208{
1209	struct mbuf *m;
1210	struct ifnet *ifp;
1211	int idx, total_len, lim;
1212	struct vge_rxdesc *cur_rxd;
1213	struct vge_rxsoft *rxs;
1214	uint32_t rxstat, rxctl;
1215
1216	ifp = &sc->sc_ethercom.ec_if;
1217	lim = 0;
1218
1219	/* Invalidate the descriptor memory */
1220
1221	for (idx = sc->sc_rx_prodidx;; idx = VGE_NEXT_RXDESC(idx)) {
1222		cur_rxd = &sc->sc_rxdescs[idx];
1223
1224		VGE_RXDESCSYNC(sc, idx,
1225		    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1226		rxstat = le32toh(cur_rxd->rd_sts);
1227		if ((rxstat & VGE_RDSTS_OWN) != 0) {
1228			VGE_RXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD);
1229			break;
1230		}
1231
1232		rxctl = le32toh(cur_rxd->rd_ctl);
1233		rxs = &sc->sc_rxsoft[idx];
1234		m = rxs->rxs_mbuf;
1235		total_len = (rxstat & VGE_RDSTS_BUFSIZ) >> 16;
1236
1237		/* Invalidate the RX mbuf and unload its map */
1238
1239		bus_dmamap_sync(sc->sc_dmat, rxs->rxs_dmamap,
1240		    0, rxs->rxs_dmamap->dm_mapsize, BUS_DMASYNC_POSTREAD);
1241		bus_dmamap_unload(sc->sc_dmat, rxs->rxs_dmamap);
1242
1243		/*
1244		 * If the 'start of frame' bit is set, this indicates
1245		 * either the first fragment in a multi-fragment receive,
1246		 * or an intermediate fragment. Either way, we want to
1247		 * accumulate the buffers.
1248		 */
1249		if (rxstat & VGE_RXPKT_SOF) {
1250			m->m_len = VGE_RX_BUFSIZE;
1251			if (sc->sc_rx_mhead == NULL)
1252				sc->sc_rx_mhead = sc->sc_rx_mtail = m;
1253			else {
1254				m->m_flags &= ~M_PKTHDR;
1255				sc->sc_rx_mtail->m_next = m;
1256				sc->sc_rx_mtail = m;
1257			}
1258			vge_newbuf(sc, idx, NULL);
1259			continue;
1260		}
1261
1262		/*
1263		 * Bad/error frames will have the RXOK bit cleared.
1264		 * However, there's one error case we want to allow:
1265		 * if a VLAN tagged frame arrives and the chip can't
1266		 * match it against the CAM filter, it considers this
1267		 * a 'VLAN CAM filter miss' and clears the 'RXOK' bit.
1268		 * We don't want to drop the frame though: our VLAN
1269		 * filtering is done in software.
1270		 */
1271		if ((rxstat & VGE_RDSTS_RXOK) == 0 &&
1272		    (rxstat & VGE_RDSTS_VIDM) == 0 &&
1273		    (rxstat & VGE_RDSTS_CSUMERR) == 0) {
1274			ifp->if_ierrors++;
1275			/*
1276			 * If this is part of a multi-fragment packet,
1277			 * discard all the pieces.
1278			 */
1279			if (sc->sc_rx_mhead != NULL) {
1280				m_freem(sc->sc_rx_mhead);
1281				sc->sc_rx_mhead = sc->sc_rx_mtail = NULL;
1282			}
1283			vge_newbuf(sc, idx, m);
1284			continue;
1285		}
1286
1287		/*
1288		 * If allocating a replacement mbuf fails,
1289		 * reload the current one.
1290		 */
1291
1292		if (vge_newbuf(sc, idx, NULL)) {
1293			ifp->if_ierrors++;
1294			if (sc->sc_rx_mhead != NULL) {
1295				m_freem(sc->sc_rx_mhead);
1296				sc->sc_rx_mhead = sc->sc_rx_mtail = NULL;
1297			}
1298			vge_newbuf(sc, idx, m);
1299			continue;
1300		}
1301
1302		if (sc->sc_rx_mhead != NULL) {
1303			m->m_len = total_len % VGE_RX_BUFSIZE;
1304			/*
1305			 * Special case: if there's 4 bytes or less
1306			 * in this buffer, the mbuf can be discarded:
1307			 * the last 4 bytes is the CRC, which we don't
1308			 * care about anyway.
1309			 */
1310			if (m->m_len <= ETHER_CRC_LEN) {
1311				sc->sc_rx_mtail->m_len -=
1312				    (ETHER_CRC_LEN - m->m_len);
1313				m_freem(m);
1314			} else {
1315				m->m_len -= ETHER_CRC_LEN;
1316				m->m_flags &= ~M_PKTHDR;
1317				sc->sc_rx_mtail->m_next = m;
1318			}
1319			m = sc->sc_rx_mhead;
1320			sc->sc_rx_mhead = sc->sc_rx_mtail = NULL;
1321			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
1322		} else
1323			m->m_pkthdr.len = m->m_len = total_len - ETHER_CRC_LEN;
1324
1325#ifndef __NO_STRICT_ALIGNMENT
1326		vge_fixup_rx(m);
1327#endif
1328		ifp->if_ipackets++;
1329		m->m_pkthdr.rcvif = ifp;
1330
1331		/* Do RX checksumming if enabled */
1332		if (ifp->if_csum_flags_rx & M_CSUM_IPv4) {
1333
1334			/* Check IP header checksum */
1335			if (rxctl & VGE_RDCTL_IPPKT)
1336				m->m_pkthdr.csum_flags |= M_CSUM_IPv4;
1337			if ((rxctl & VGE_RDCTL_IPCSUMOK) == 0)
1338				m->m_pkthdr.csum_flags |= M_CSUM_IPv4_BAD;
1339		}
1340
1341		if (ifp->if_csum_flags_rx & M_CSUM_TCPv4) {
1342			/* Check UDP checksum */
1343			if (rxctl & VGE_RDCTL_TCPPKT)
1344				m->m_pkthdr.csum_flags |= M_CSUM_TCPv4;
1345
1346			if ((rxctl & VGE_RDCTL_PROTOCSUMOK) == 0)
1347				m->m_pkthdr.csum_flags |= M_CSUM_TCP_UDP_BAD;
1348		}
1349
1350		if (ifp->if_csum_flags_rx & M_CSUM_UDPv4) {
1351			/* Check UDP checksum */
1352			if (rxctl & VGE_RDCTL_UDPPKT)
1353				m->m_pkthdr.csum_flags |= M_CSUM_UDPv4;
1354
1355			if ((rxctl & VGE_RDCTL_PROTOCSUMOK) == 0)
1356				m->m_pkthdr.csum_flags |= M_CSUM_TCP_UDP_BAD;
1357		}
1358
1359		if (rxstat & VGE_RDSTS_VTAG) {
1360			/*
1361			 * We use bswap16() here because:
1362			 * On LE machines, tag is stored in BE as stream data.
1363			 * On BE machines, tag is stored in BE as stream data
1364			 *  but it was already swapped by le32toh() above.
1365			 */
1366			VLAN_INPUT_TAG(ifp, m,
1367			    bswap16(rxctl & VGE_RDCTL_VLANID), continue);
1368		}
1369
1370		/*
1371		 * Handle BPF listeners.
1372		 */
1373		bpf_mtap(ifp, m);
1374
1375		(*ifp->if_input)(ifp, m);
1376
1377		lim++;
1378		if (lim == VGE_NRXDESC)
1379			break;
1380	}
1381
1382	sc->sc_rx_prodidx = idx;
1383	CSR_WRITE_2(sc, VGE_RXDESC_RESIDUECNT, lim);
1384}
1385
1386static void
1387vge_txeof(struct vge_softc *sc)
1388{
1389	struct ifnet *ifp;
1390	struct vge_txsoft *txs;
1391	uint32_t txstat;
1392	int idx;
1393
1394	ifp = &sc->sc_ethercom.ec_if;
1395
1396	for (idx = sc->sc_tx_considx;
1397	    sc->sc_tx_free < VGE_NTXDESC;
1398	    idx = VGE_NEXT_TXDESC(idx), sc->sc_tx_free++) {
1399		VGE_TXDESCSYNC(sc, idx,
1400		    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1401		txstat = le32toh(sc->sc_txdescs[idx].td_sts);
1402		VGE_TXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD);
1403		if (txstat & VGE_TDSTS_OWN) {
1404			break;
1405		}
1406
1407		txs = &sc->sc_txsoft[idx];
1408		m_freem(txs->txs_mbuf);
1409		txs->txs_mbuf = NULL;
1410		bus_dmamap_sync(sc->sc_dmat, txs->txs_dmamap, 0,
1411		    txs->txs_dmamap->dm_mapsize, BUS_DMASYNC_POSTWRITE);
1412		bus_dmamap_unload(sc->sc_dmat, txs->txs_dmamap);
1413		if (txstat & (VGE_TDSTS_EXCESSCOLL|VGE_TDSTS_COLL))
1414			ifp->if_collisions++;
1415		if (txstat & VGE_TDSTS_TXERR)
1416			ifp->if_oerrors++;
1417		else
1418			ifp->if_opackets++;
1419	}
1420
1421	sc->sc_tx_considx = idx;
1422
1423	if (sc->sc_tx_free > 0) {
1424		ifp->if_flags &= ~IFF_OACTIVE;
1425	}
1426
1427	/*
1428	 * If not all descriptors have been released reaped yet,
1429	 * reload the timer so that we will eventually get another
1430	 * interrupt that will cause us to re-enter this routine.
1431	 * This is done in case the transmitter has gone idle.
1432	 */
1433	if (sc->sc_tx_free < VGE_NTXDESC)
1434		CSR_WRITE_1(sc, VGE_CRS1, VGE_CR1_TIMER0_ENABLE);
1435	else
1436		ifp->if_timer = 0;
1437}
1438
1439static void
1440vge_tick(void *arg)
1441{
1442	struct vge_softc *sc;
1443	struct ifnet *ifp;
1444	struct mii_data *mii;
1445	int s;
1446
1447	sc = arg;
1448	ifp = &sc->sc_ethercom.ec_if;
1449	mii = &sc->sc_mii;
1450
1451	s = splnet();
1452
1453	callout_schedule(&sc->sc_timeout, hz);
1454
1455	mii_tick(mii);
1456	if (sc->sc_link) {
1457		if ((mii->mii_media_status & IFM_ACTIVE) == 0)
1458			sc->sc_link = 0;
1459	} else {
1460		if (mii->mii_media_status & IFM_ACTIVE &&
1461		    IFM_SUBTYPE(mii->mii_media_active) != IFM_NONE) {
1462			sc->sc_link = 1;
1463			if (!IFQ_IS_EMPTY(&ifp->if_snd))
1464				vge_start(ifp);
1465		}
1466	}
1467
1468	splx(s);
1469}
1470
1471static int
1472vge_intr(void *arg)
1473{
1474	struct vge_softc *sc;
1475	struct ifnet *ifp;
1476	uint32_t status;
1477	int claim;
1478
1479	sc = arg;
1480	claim = 0;
1481	if (sc->sc_suspended) {
1482		return claim;
1483	}
1484
1485	ifp = &sc->sc_ethercom.ec_if;
1486
1487	if ((ifp->if_flags & IFF_UP) == 0) {
1488		return claim;
1489	}
1490
1491	/* Disable interrupts */
1492	CSR_WRITE_1(sc, VGE_CRC3, VGE_CR3_INT_GMSK);
1493
1494	for (;;) {
1495
1496		status = CSR_READ_4(sc, VGE_ISR);
1497		/* If the card has gone away the read returns 0xffffffff. */
1498		if (status == 0xFFFFFFFF)
1499			break;
1500
1501		if (status) {
1502			claim = 1;
1503			CSR_WRITE_4(sc, VGE_ISR, status);
1504		}
1505
1506		if ((status & VGE_INTRS) == 0)
1507			break;
1508
1509		if (status & (VGE_ISR_RXOK|VGE_ISR_RXOK_HIPRIO))
1510			vge_rxeof(sc);
1511
1512		if (status & (VGE_ISR_RXOFLOW|VGE_ISR_RXNODESC)) {
1513			vge_rxeof(sc);
1514			CSR_WRITE_1(sc, VGE_RXQCSRS, VGE_RXQCSR_RUN);
1515			CSR_WRITE_1(sc, VGE_RXQCSRS, VGE_RXQCSR_WAK);
1516		}
1517
1518		if (status & (VGE_ISR_TXOK0|VGE_ISR_TIMER0))
1519			vge_txeof(sc);
1520
1521		if (status & (VGE_ISR_TXDMA_STALL|VGE_ISR_RXDMA_STALL))
1522			vge_init(ifp);
1523
1524		if (status & VGE_ISR_LINKSTS)
1525			vge_tick(sc);
1526	}
1527
1528	/* Re-enable interrupts */
1529	CSR_WRITE_1(sc, VGE_CRS3, VGE_CR3_INT_GMSK);
1530
1531	if (claim && !IFQ_IS_EMPTY(&ifp->if_snd))
1532		vge_start(ifp);
1533
1534	return claim;
1535}
1536
1537static int
1538vge_encap(struct vge_softc *sc, struct mbuf *m_head, int idx)
1539{
1540	struct vge_txsoft *txs;
1541	struct vge_txdesc *txd;
1542	struct vge_txfrag *f;
1543	struct mbuf *m_new;
1544	bus_dmamap_t map;
1545	int m_csumflags, seg, error, flags;
1546	struct m_tag *mtag;
1547	size_t sz;
1548	uint32_t td_sts, td_ctl;
1549
1550	KASSERT(sc->sc_tx_free > 0);
1551
1552	txd = &sc->sc_txdescs[idx];
1553
1554#ifdef DIAGNOSTIC
1555	/* If this descriptor is still owned by the chip, bail. */
1556	VGE_TXDESCSYNC(sc, idx,
1557	    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1558	td_sts = le32toh(txd->td_sts);
1559	VGE_TXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD);
1560	if (td_sts & VGE_TDSTS_OWN) {
1561		return ENOBUFS;
1562	}
1563#endif
1564
1565	/*
1566	 * Preserve m_pkthdr.csum_flags here since m_head might be
1567	 * updated by m_defrag()
1568	 */
1569	m_csumflags = m_head->m_pkthdr.csum_flags;
1570
1571	txs = &sc->sc_txsoft[idx];
1572	map = txs->txs_dmamap;
1573	error = bus_dmamap_load_mbuf(sc->sc_dmat, map, m_head, BUS_DMA_NOWAIT);
1574
1575	/* If too many segments to map, coalesce */
1576	if (error == EFBIG ||
1577	    (m_head->m_pkthdr.len < ETHER_PAD_LEN &&
1578	     map->dm_nsegs == VGE_TX_FRAGS)) {
1579		m_new = m_defrag(m_head, M_DONTWAIT);
1580		if (m_new == NULL)
1581			return EFBIG;
1582
1583		error = bus_dmamap_load_mbuf(sc->sc_dmat, map,
1584		    m_new, BUS_DMA_NOWAIT);
1585		if (error) {
1586			m_freem(m_new);
1587			return error;
1588		}
1589
1590		m_head = m_new;
1591	} else if (error)
1592		return error;
1593
1594	txs->txs_mbuf = m_head;
1595
1596	bus_dmamap_sync(sc->sc_dmat, map, 0, map->dm_mapsize,
1597	    BUS_DMASYNC_PREWRITE);
1598
1599	for (seg = 0, f = &txd->td_frag[0]; seg < map->dm_nsegs; seg++, f++) {
1600		f->tf_buflen = htole16(VGE_BUFLEN(map->dm_segs[seg].ds_len));
1601		vge_set_txaddr(f, map->dm_segs[seg].ds_addr);
1602	}
1603
1604	/* Argh. This chip does not autopad short frames */
1605	sz = m_head->m_pkthdr.len;
1606	if (sz < ETHER_PAD_LEN) {
1607		f->tf_buflen = htole16(VGE_BUFLEN(ETHER_PAD_LEN - sz));
1608		vge_set_txaddr(f, VGE_CDPADADDR(sc));
1609		sz = ETHER_PAD_LEN;
1610		seg++;
1611	}
1612	VGE_TXFRAGSYNC(sc, idx, seg, BUS_DMASYNC_PREWRITE);
1613
1614	/*
1615	 * When telling the chip how many segments there are, we
1616	 * must use nsegs + 1 instead of just nsegs. Darned if I
1617	 * know why.
1618	 */
1619	seg++;
1620
1621	flags = 0;
1622	if (m_csumflags & M_CSUM_IPv4)
1623		flags |= VGE_TDCTL_IPCSUM;
1624	if (m_csumflags & M_CSUM_TCPv4)
1625		flags |= VGE_TDCTL_TCPCSUM;
1626	if (m_csumflags & M_CSUM_UDPv4)
1627		flags |= VGE_TDCTL_UDPCSUM;
1628	td_sts = sz << 16;
1629	td_ctl = flags | (seg << 28) | VGE_TD_LS_NORM;
1630
1631	if (sz > ETHERMTU + ETHER_HDR_LEN)
1632		td_ctl |= VGE_TDCTL_JUMBO;
1633
1634	/*
1635	 * Set up hardware VLAN tagging.
1636	 */
1637	mtag = VLAN_OUTPUT_TAG(&sc->sc_ethercom, m_head);
1638	if (mtag != NULL) {
1639		/*
1640		 * No need htons() here since vge(4) chip assumes
1641		 * that tags are written in little endian and
1642		 * we already use htole32() here.
1643		 */
1644		td_ctl |= VLAN_TAG_VALUE(mtag) | VGE_TDCTL_VTAG;
1645	}
1646	txd->td_ctl = htole32(td_ctl);
1647	txd->td_sts = htole32(td_sts);
1648	VGE_TXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1649
1650	txd->td_sts = htole32(VGE_TDSTS_OWN | td_sts);
1651	VGE_TXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1652
1653	sc->sc_tx_free--;
1654
1655	return 0;
1656}
1657
1658/*
1659 * Main transmit routine.
1660 */
1661
1662static void
1663vge_start(struct ifnet *ifp)
1664{
1665	struct vge_softc *sc;
1666	struct vge_txsoft *txs;
1667	struct mbuf *m_head;
1668	int idx, pidx, ofree, error;
1669
1670	sc = ifp->if_softc;
1671
1672	if (!sc->sc_link ||
1673	    (ifp->if_flags & (IFF_RUNNING|IFF_OACTIVE)) != IFF_RUNNING) {
1674		return;
1675	}
1676
1677	m_head = NULL;
1678	idx = sc->sc_tx_prodidx;
1679	pidx = VGE_PREV_TXDESC(idx);
1680	ofree = sc->sc_tx_free;
1681
1682	/*
1683	 * Loop through the send queue, setting up transmit descriptors
1684	 * until we drain the queue, or use up all available transmit
1685	 * descriptors.
1686	 */
1687	for (;;) {
1688		/* Grab a packet off the queue. */
1689		IFQ_POLL(&ifp->if_snd, m_head);
1690		if (m_head == NULL)
1691			break;
1692
1693		if (sc->sc_tx_free == 0) {
1694			/*
1695			 * All slots used, stop for now.
1696			 */
1697			ifp->if_flags |= IFF_OACTIVE;
1698			break;
1699		}
1700
1701		txs = &sc->sc_txsoft[idx];
1702		KASSERT(txs->txs_mbuf == NULL);
1703
1704		if ((error = vge_encap(sc, m_head, idx))) {
1705			if (error == EFBIG) {
1706				printf("%s: Tx packet consumes too many "
1707				    "DMA segments, dropping...\n",
1708				    device_xname(sc->sc_dev));
1709				IFQ_DEQUEUE(&ifp->if_snd, m_head);
1710				m_freem(m_head);
1711				continue;
1712			}
1713
1714			/*
1715			 * Short on resources, just stop for now.
1716			 */
1717			if (error == ENOBUFS)
1718				ifp->if_flags |= IFF_OACTIVE;
1719			break;
1720		}
1721
1722		IFQ_DEQUEUE(&ifp->if_snd, m_head);
1723
1724		/*
1725		 * WE ARE NOW COMMITTED TO TRANSMITTING THE PACKET.
1726		 */
1727
1728		sc->sc_txdescs[pidx].td_frag[0].tf_buflen |=
1729		    htole16(VGE_TXDESC_Q);
1730		VGE_TXFRAGSYNC(sc, pidx, 1,
1731		    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1732
1733		if (txs->txs_mbuf != m_head) {
1734			m_freem(m_head);
1735			m_head = txs->txs_mbuf;
1736		}
1737
1738		pidx = idx;
1739		idx = VGE_NEXT_TXDESC(idx);
1740
1741		/*
1742		 * If there's a BPF listener, bounce a copy of this frame
1743		 * to him.
1744		 */
1745		bpf_mtap(ifp, m_head);
1746	}
1747
1748	if (sc->sc_tx_free < ofree) {
1749		/* TX packet queued */
1750
1751		sc->sc_tx_prodidx = idx;
1752
1753		/* Issue a transmit command. */
1754		CSR_WRITE_2(sc, VGE_TXQCSRS, VGE_TXQCSR_WAK0);
1755
1756		/*
1757		 * Use the countdown timer for interrupt moderation.
1758		 * 'TX done' interrupts are disabled. Instead, we reset the
1759		 * countdown timer, which will begin counting until it hits
1760		 * the value in the SSTIMER register, and then trigger an
1761		 * interrupt. Each time we set the TIMER0_ENABLE bit, the
1762		 * the timer count is reloaded. Only when the transmitter
1763		 * is idle will the timer hit 0 and an interrupt fire.
1764		 */
1765		CSR_WRITE_1(sc, VGE_CRS1, VGE_CR1_TIMER0_ENABLE);
1766
1767		/*
1768		 * Set a timeout in case the chip goes out to lunch.
1769		 */
1770		ifp->if_timer = 5;
1771	}
1772}
1773
1774static int
1775vge_init(struct ifnet *ifp)
1776{
1777	struct vge_softc *sc;
1778	int i, rc = 0;
1779
1780	sc = ifp->if_softc;
1781
1782	/*
1783	 * Cancel pending I/O and free all RX/TX buffers.
1784	 */
1785	vge_stop(ifp, 0);
1786	vge_reset(sc);
1787
1788	/* Initialize the RX descriptors and mbufs. */
1789	memset(sc->sc_rxdescs, 0, sizeof(sc->sc_rxdescs));
1790	sc->sc_rx_consumed = 0;
1791	for (i = 0; i < VGE_NRXDESC; i++) {
1792		if (vge_newbuf(sc, i, NULL) == ENOBUFS) {
1793			printf("%s: unable to allocate or map rx buffer\n",
1794			    device_xname(sc->sc_dev));
1795			return 1; /* XXX */
1796		}
1797	}
1798	sc->sc_rx_prodidx = 0;
1799	sc->sc_rx_mhead = sc->sc_rx_mtail = NULL;
1800
1801	/* Initialize the  TX descriptors and mbufs. */
1802	memset(sc->sc_txdescs, 0, sizeof(sc->sc_txdescs));
1803	bus_dmamap_sync(sc->sc_dmat, sc->sc_cddmamap,
1804	    VGE_CDTXOFF(0), sizeof(sc->sc_txdescs),
1805	    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1806	for (i = 0; i < VGE_NTXDESC; i++)
1807		sc->sc_txsoft[i].txs_mbuf = NULL;
1808
1809	sc->sc_tx_prodidx = 0;
1810	sc->sc_tx_considx = 0;
1811	sc->sc_tx_free = VGE_NTXDESC;
1812
1813	/* Set our station address */
1814	for (i = 0; i < ETHER_ADDR_LEN; i++)
1815		CSR_WRITE_1(sc, VGE_PAR0 + i, sc->sc_eaddr[i]);
1816
1817	/*
1818	 * Set receive FIFO threshold. Also allow transmission and
1819	 * reception of VLAN tagged frames.
1820	 */
1821	CSR_CLRBIT_1(sc, VGE_RXCFG, VGE_RXCFG_FIFO_THR|VGE_RXCFG_VTAGOPT);
1822	CSR_SETBIT_1(sc, VGE_RXCFG, VGE_RXFIFOTHR_128BYTES|VGE_VTAG_OPT2);
1823
1824	/* Set DMA burst length */
1825	CSR_CLRBIT_1(sc, VGE_DMACFG0, VGE_DMACFG0_BURSTLEN);
1826	CSR_SETBIT_1(sc, VGE_DMACFG0, VGE_DMABURST_128);
1827
1828	CSR_SETBIT_1(sc, VGE_TXCFG, VGE_TXCFG_ARB_PRIO|VGE_TXCFG_NONBLK);
1829
1830	/* Set collision backoff algorithm */
1831	CSR_CLRBIT_1(sc, VGE_CHIPCFG1, VGE_CHIPCFG1_CRANDOM|
1832	    VGE_CHIPCFG1_CAP|VGE_CHIPCFG1_MBA|VGE_CHIPCFG1_BAKOPT);
1833	CSR_SETBIT_1(sc, VGE_CHIPCFG1, VGE_CHIPCFG1_OFSET);
1834
1835	/* Disable LPSEL field in priority resolution */
1836	CSR_SETBIT_1(sc, VGE_DIAGCTL, VGE_DIAGCTL_LPSEL_DIS);
1837
1838	/*
1839	 * Load the addresses of the DMA queues into the chip.
1840	 * Note that we only use one transmit queue.
1841	 */
1842
1843	CSR_WRITE_4(sc, VGE_TXDESC_ADDR_LO0, VGE_ADDR_LO(VGE_CDTXADDR(sc, 0)));
1844	CSR_WRITE_2(sc, VGE_TXDESCNUM, VGE_NTXDESC - 1);
1845
1846	CSR_WRITE_4(sc, VGE_RXDESC_ADDR_LO, VGE_ADDR_LO(VGE_CDRXADDR(sc, 0)));
1847	CSR_WRITE_2(sc, VGE_RXDESCNUM, VGE_NRXDESC - 1);
1848	CSR_WRITE_2(sc, VGE_RXDESC_RESIDUECNT, VGE_NRXDESC);
1849
1850	/* Enable and wake up the RX descriptor queue */
1851	CSR_WRITE_1(sc, VGE_RXQCSRS, VGE_RXQCSR_RUN);
1852	CSR_WRITE_1(sc, VGE_RXQCSRS, VGE_RXQCSR_WAK);
1853
1854	/* Enable the TX descriptor queue */
1855	CSR_WRITE_2(sc, VGE_TXQCSRS, VGE_TXQCSR_RUN0);
1856
1857	/* Set up the receive filter -- allow large frames for VLANs. */
1858	CSR_WRITE_1(sc, VGE_RXCTL, VGE_RXCTL_RX_UCAST|VGE_RXCTL_RX_GIANT);
1859
1860	/* If we want promiscuous mode, set the allframes bit. */
1861	if (ifp->if_flags & IFF_PROMISC) {
1862		CSR_SETBIT_1(sc, VGE_RXCTL, VGE_RXCTL_RX_PROMISC);
1863	}
1864
1865	/* Set capture broadcast bit to capture broadcast frames. */
1866	if (ifp->if_flags & IFF_BROADCAST) {
1867		CSR_SETBIT_1(sc, VGE_RXCTL, VGE_RXCTL_RX_BCAST);
1868	}
1869
1870	/* Set multicast bit to capture multicast frames. */
1871	if (ifp->if_flags & IFF_MULTICAST) {
1872		CSR_SETBIT_1(sc, VGE_RXCTL, VGE_RXCTL_RX_MCAST);
1873	}
1874
1875	/* Init the cam filter. */
1876	vge_cam_clear(sc);
1877
1878	/* Init the multicast filter. */
1879	vge_setmulti(sc);
1880
1881	/* Enable flow control */
1882
1883	CSR_WRITE_1(sc, VGE_CRS2, 0x8B);
1884
1885	/* Enable jumbo frame reception (if desired) */
1886
1887	/* Start the MAC. */
1888	CSR_WRITE_1(sc, VGE_CRC0, VGE_CR0_STOP);
1889	CSR_WRITE_1(sc, VGE_CRS1, VGE_CR1_NOPOLL);
1890	CSR_WRITE_1(sc, VGE_CRS0,
1891	    VGE_CR0_TX_ENABLE|VGE_CR0_RX_ENABLE|VGE_CR0_START);
1892
1893	/*
1894	 * Configure one-shot timer for microsecond
1895	 * resulution and load it for 500 usecs.
1896	 */
1897	CSR_SETBIT_1(sc, VGE_DIAGCTL, VGE_DIAGCTL_TIMER0_RES);
1898	CSR_WRITE_2(sc, VGE_SSTIMER, 400);
1899
1900	/*
1901	 * Configure interrupt moderation for receive. Enable
1902	 * the holdoff counter and load it, and set the RX
1903	 * suppression count to the number of descriptors we
1904	 * want to allow before triggering an interrupt.
1905	 * The holdoff timer is in units of 20 usecs.
1906	 */
1907
1908#ifdef notyet
1909	CSR_WRITE_1(sc, VGE_INTCTL1, VGE_INTCTL_TXINTSUP_DISABLE);
1910	/* Select the interrupt holdoff timer page. */
1911	CSR_CLRBIT_1(sc, VGE_CAMCTL, VGE_CAMCTL_PAGESEL);
1912	CSR_SETBIT_1(sc, VGE_CAMCTL, VGE_PAGESEL_INTHLDOFF);
1913	CSR_WRITE_1(sc, VGE_INTHOLDOFF, 10); /* ~200 usecs */
1914
1915	/* Enable use of the holdoff timer. */
1916	CSR_WRITE_1(sc, VGE_CRS3, VGE_CR3_INT_HOLDOFF);
1917	CSR_WRITE_1(sc, VGE_INTCTL1, VGE_INTCTL_SC_RELOAD);
1918
1919	/* Select the RX suppression threshold page. */
1920	CSR_CLRBIT_1(sc, VGE_CAMCTL, VGE_CAMCTL_PAGESEL);
1921	CSR_SETBIT_1(sc, VGE_CAMCTL, VGE_PAGESEL_RXSUPPTHR);
1922	CSR_WRITE_1(sc, VGE_RXSUPPTHR, 64); /* interrupt after 64 packets */
1923
1924	/* Restore the page select bits. */
1925	CSR_CLRBIT_1(sc, VGE_CAMCTL, VGE_CAMCTL_PAGESEL);
1926	CSR_SETBIT_1(sc, VGE_CAMCTL, VGE_PAGESEL_MAR);
1927#endif
1928
1929#ifdef DEVICE_POLLING
1930	/*
1931	 * Disable interrupts if we are polling.
1932	 */
1933	if (ifp->if_flags & IFF_POLLING) {
1934		CSR_WRITE_4(sc, VGE_IMR, 0);
1935		CSR_WRITE_1(sc, VGE_CRC3, VGE_CR3_INT_GMSK);
1936	} else	/* otherwise ... */
1937#endif /* DEVICE_POLLING */
1938	{
1939	/*
1940	 * Enable interrupts.
1941	 */
1942		CSR_WRITE_4(sc, VGE_IMR, VGE_INTRS);
1943		CSR_WRITE_4(sc, VGE_ISR, 0);
1944		CSR_WRITE_1(sc, VGE_CRS3, VGE_CR3_INT_GMSK);
1945	}
1946
1947	if ((rc = ether_mediachange(ifp)) != 0)
1948		goto out;
1949
1950	ifp->if_flags |= IFF_RUNNING;
1951	ifp->if_flags &= ~IFF_OACTIVE;
1952
1953	sc->sc_if_flags = 0;
1954	sc->sc_link = 0;
1955
1956	callout_schedule(&sc->sc_timeout, hz);
1957
1958out:
1959	return rc;
1960}
1961
1962static void
1963vge_miibus_statchg(device_t self)
1964{
1965	struct vge_softc *sc;
1966	struct mii_data *mii;
1967	struct ifmedia_entry *ife;
1968
1969	sc = device_private(self);
1970	mii = &sc->sc_mii;
1971	ife = mii->mii_media.ifm_cur;
1972	/*
1973	 * If the user manually selects a media mode, we need to turn
1974	 * on the forced MAC mode bit in the DIAGCTL register. If the
1975	 * user happens to choose a full duplex mode, we also need to
1976	 * set the 'force full duplex' bit. This applies only to
1977	 * 10Mbps and 100Mbps speeds. In autoselect mode, forced MAC
1978	 * mode is disabled, and in 1000baseT mode, full duplex is
1979	 * always implied, so we turn on the forced mode bit but leave
1980	 * the FDX bit cleared.
1981	 */
1982
1983	switch (IFM_SUBTYPE(ife->ifm_media)) {
1984	case IFM_AUTO:
1985		CSR_CLRBIT_1(sc, VGE_DIAGCTL, VGE_DIAGCTL_MACFORCE);
1986		CSR_CLRBIT_1(sc, VGE_DIAGCTL, VGE_DIAGCTL_FDXFORCE);
1987		break;
1988	case IFM_1000_T:
1989		CSR_SETBIT_1(sc, VGE_DIAGCTL, VGE_DIAGCTL_MACFORCE);
1990		CSR_CLRBIT_1(sc, VGE_DIAGCTL, VGE_DIAGCTL_FDXFORCE);
1991		break;
1992	case IFM_100_TX:
1993	case IFM_10_T:
1994		CSR_SETBIT_1(sc, VGE_DIAGCTL, VGE_DIAGCTL_MACFORCE);
1995		if ((ife->ifm_media & IFM_GMASK) == IFM_FDX) {
1996			CSR_SETBIT_1(sc, VGE_DIAGCTL, VGE_DIAGCTL_FDXFORCE);
1997		} else {
1998			CSR_CLRBIT_1(sc, VGE_DIAGCTL, VGE_DIAGCTL_FDXFORCE);
1999		}
2000		break;
2001	default:
2002		printf("%s: unknown media type: %x\n",
2003		    device_xname(sc->sc_dev),
2004		    IFM_SUBTYPE(ife->ifm_media));
2005		break;
2006	}
2007}
2008
2009static int
2010vge_ifflags_cb(struct ethercom *ec)
2011{
2012	struct ifnet *ifp = &ec->ec_if;
2013	struct vge_softc *sc = ifp->if_softc;
2014	int change = ifp->if_flags ^ sc->sc_if_flags;
2015
2016	if ((change & ~(IFF_CANTCHANGE|IFF_DEBUG)) != 0)
2017		return ENETRESET;
2018	else if ((change & IFF_PROMISC) == 0)
2019		return 0;
2020
2021	if ((ifp->if_flags & IFF_PROMISC) == 0)
2022		CSR_CLRBIT_1(sc, VGE_RXCTL, VGE_RXCTL_RX_PROMISC);
2023	else
2024		CSR_SETBIT_1(sc, VGE_RXCTL, VGE_RXCTL_RX_PROMISC);
2025	vge_setmulti(sc);
2026	return 0;
2027}
2028
2029static int
2030vge_ioctl(struct ifnet *ifp, u_long command, void *data)
2031{
2032	struct vge_softc *sc;
2033	struct ifreq *ifr;
2034	int s, error;
2035
2036	sc = ifp->if_softc;
2037	ifr = (struct ifreq *)data;
2038	error = 0;
2039
2040	s = splnet();
2041
2042	if ((error = ether_ioctl(ifp, command, data)) == ENETRESET) {
2043		error = 0;
2044		if (command != SIOCADDMULTI && command != SIOCDELMULTI)
2045			;
2046		else if (ifp->if_flags & IFF_RUNNING) {
2047			/*
2048			 * Multicast list has changed; set the hardware filter
2049			 * accordingly.
2050			 */
2051			vge_setmulti(sc);
2052		}
2053	}
2054	sc->sc_if_flags = ifp->if_flags;
2055
2056	splx(s);
2057	return error;
2058}
2059
2060static void
2061vge_watchdog(struct ifnet *ifp)
2062{
2063	struct vge_softc *sc;
2064	int s;
2065
2066	sc = ifp->if_softc;
2067	s = splnet();
2068	printf("%s: watchdog timeout\n", device_xname(sc->sc_dev));
2069	ifp->if_oerrors++;
2070
2071	vge_txeof(sc);
2072	vge_rxeof(sc);
2073
2074	vge_init(ifp);
2075
2076	splx(s);
2077}
2078
2079/*
2080 * Stop the adapter and free any mbufs allocated to the
2081 * RX and TX lists.
2082 */
2083static void
2084vge_stop(struct ifnet *ifp, int disable)
2085{
2086	struct vge_softc *sc = ifp->if_softc;
2087	struct vge_txsoft *txs;
2088	struct vge_rxsoft *rxs;
2089	int i, s;
2090
2091	s = splnet();
2092	ifp->if_timer = 0;
2093
2094	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
2095#ifdef DEVICE_POLLING
2096	ether_poll_deregister(ifp);
2097#endif /* DEVICE_POLLING */
2098
2099	CSR_WRITE_1(sc, VGE_CRC3, VGE_CR3_INT_GMSK);
2100	CSR_WRITE_1(sc, VGE_CRS0, VGE_CR0_STOP);
2101	CSR_WRITE_4(sc, VGE_ISR, 0xFFFFFFFF);
2102	CSR_WRITE_2(sc, VGE_TXQCSRC, 0xFFFF);
2103	CSR_WRITE_1(sc, VGE_RXQCSRC, 0xFF);
2104	CSR_WRITE_4(sc, VGE_RXDESC_ADDR_LO, 0);
2105
2106	if (sc->sc_rx_mhead != NULL) {
2107		m_freem(sc->sc_rx_mhead);
2108		sc->sc_rx_mhead = sc->sc_rx_mtail = NULL;
2109	}
2110
2111	/* Free the TX list buffers. */
2112
2113	for (i = 0; i < VGE_NTXDESC; i++) {
2114		txs = &sc->sc_txsoft[i];
2115		if (txs->txs_mbuf != NULL) {
2116			bus_dmamap_unload(sc->sc_dmat, txs->txs_dmamap);
2117			m_freem(txs->txs_mbuf);
2118			txs->txs_mbuf = NULL;
2119		}
2120	}
2121
2122	/* Free the RX list buffers. */
2123
2124	for (i = 0; i < VGE_NRXDESC; i++) {
2125		rxs = &sc->sc_rxsoft[i];
2126		if (rxs->rxs_mbuf != NULL) {
2127			bus_dmamap_unload(sc->sc_dmat, rxs->rxs_dmamap);
2128			m_freem(rxs->rxs_mbuf);
2129			rxs->rxs_mbuf = NULL;
2130		}
2131	}
2132
2133	splx(s);
2134}
2135
2136#if VGE_POWER_MANAGEMENT
2137/*
2138 * Device suspend routine.  Stop the interface and save some PCI
2139 * settings in case the BIOS doesn't restore them properly on
2140 * resume.
2141 */
2142static int
2143vge_suspend(device_t dev)
2144{
2145	struct vge_softc *sc;
2146	int i;
2147
2148	sc = device_get_softc(dev);
2149
2150	vge_stop(sc);
2151
2152        for (i = 0; i < 5; i++)
2153		sc->sc_saved_maps[i] =
2154		    pci_read_config(dev, PCIR_MAPS + i * 4, 4);
2155	sc->sc_saved_biosaddr = pci_read_config(dev, PCIR_BIOS, 4);
2156	sc->sc_saved_intline = pci_read_config(dev, PCIR_INTLINE, 1);
2157	sc->sc_saved_cachelnsz = pci_read_config(dev, PCIR_CACHELNSZ, 1);
2158	sc->sc_saved_lattimer = pci_read_config(dev, PCIR_LATTIMER, 1);
2159
2160	sc->suspended = 1;
2161
2162	return 0;
2163}
2164
2165/*
2166 * Device resume routine.  Restore some PCI settings in case the BIOS
2167 * doesn't, re-enable busmastering, and restart the interface if
2168 * appropriate.
2169 */
2170static int
2171vge_resume(device_t dev)
2172{
2173	struct vge_softc *sc;
2174	struct ifnet *ifp;
2175	int i;
2176
2177	sc = device_private(dev);
2178	ifp = &sc->sc_ethercom.ec_if;
2179
2180        /* better way to do this? */
2181	for (i = 0; i < 5; i++)
2182		pci_write_config(dev, PCIR_MAPS + i * 4,
2183		    sc->sc_saved_maps[i], 4);
2184	pci_write_config(dev, PCIR_BIOS, sc->sc_saved_biosaddr, 4);
2185	pci_write_config(dev, PCIR_INTLINE, sc->sc_saved_intline, 1);
2186	pci_write_config(dev, PCIR_CACHELNSZ, sc->sc_saved_cachelnsz, 1);
2187	pci_write_config(dev, PCIR_LATTIMER, sc->sc_saved_lattimer, 1);
2188
2189	/* reenable busmastering */
2190	pci_enable_busmaster(dev);
2191	pci_enable_io(dev, SYS_RES_MEMORY);
2192
2193	/* reinitialize interface if necessary */
2194	if (ifp->if_flags & IFF_UP)
2195		vge_init(sc);
2196
2197	sc->suspended = 0;
2198
2199	return 0;
2200}
2201#endif
2202
2203/*
2204 * Stop all chip I/O so that the kernel's probe routines don't
2205 * get confused by errant DMAs when rebooting.
2206 */
2207static bool
2208vge_shutdown(device_t self, int howto)
2209{
2210	struct vge_softc *sc;
2211
2212	sc = device_private(self);
2213	vge_stop(&sc->sc_ethercom.ec_if, 1);
2214
2215	return true;
2216}
2217