xref: /netbsd-src/sys/dev/pci/if_vge.c (revision c0179c282a5968435315a82f4128c61372c68fc3)
1 /* $NetBSD: if_vge.c,v 1.27 2006/11/16 01:33:09 christos 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 1.27 2006/11/16 01:33:09 christos 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 #include "bpfilter.h"
88 
89 #include <sys/param.h>
90 #include <sys/endian.h>
91 #include <sys/systm.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 <machine/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_JUMBO_MTU		9000
118 
119 #define VGE_IFQ_MAXLEN		64
120 
121 #define VGE_RING_ALIGN		256
122 
123 #define VGE_NTXDESC		256
124 #define VGE_NTXDESC_MASK	(VGE_NTXDESC - 1)
125 #define VGE_NEXT_TXDESC(x)	((x + 1) & VGE_NTXDESC_MASK)
126 
127 #define VGE_NRXDESC		256	/* Must be a multiple of 4!! */
128 #define VGE_NRXDESC_MASK	(VGE_NRXDESC - 1)
129 #define VGE_NEXT_RXDESC(x)	((x + 1) & VGE_NRXDESC_MASK)
130 #define VGE_PREV_RXDESC(x)	((x - 1) & VGE_NRXDESC_MASK)
131 
132 #define VGE_ADDR_LO(y)		((uint64_t)(y) & 0xFFFFFFFF)
133 #define VGE_ADDR_HI(y)		((uint64_t)(y) >> 32)
134 #define VGE_BUFLEN(y)		((y) & 0x7FFF)
135 #define ETHER_PAD_LEN		(ETHER_MIN_LEN - ETHER_CRC_LEN)
136 
137 #ifdef __NO_STRICT_ALIGNMENT
138 #define VGE_RX_PAD		sizeof(uint32_t)
139 #else
140 #define VGE_RX_PAD		0
141 #endif
142 
143 /*
144  * Control structures are DMA'd to the vge chip. We allocate them in
145  * a single clump that maps to a single DMA segment to make several things
146  * easier.
147  */
148 struct vge_control_data {
149 	/* TX descriptors */
150 	struct vge_txdesc	vcd_txdescs[VGE_NTXDESC];
151 	/* RX descriptors */
152 	struct vge_rxdesc	vcd_rxdescs[VGE_NRXDESC];
153 	/* dummy data for TX padding */
154 	uint8_t			vcd_pad[ETHER_PAD_LEN];
155 };
156 
157 #define VGE_CDOFF(x)	offsetof(struct vge_control_data, x)
158 #define VGE_CDTXOFF(x)	VGE_CDOFF(vcd_txdescs[(x)])
159 #define VGE_CDRXOFF(x)	VGE_CDOFF(vcd_rxdescs[(x)])
160 #define VGE_CDPADOFF()	VGE_CDOFF(vcd_pad[0])
161 
162 /*
163  * Software state for TX jobs.
164  */
165 struct vge_txsoft {
166 	struct mbuf	*txs_mbuf;		/* head of our mbuf chain */
167 	bus_dmamap_t	txs_dmamap;		/* our DMA map */
168 };
169 
170 /*
171  * Software state for RX jobs.
172  */
173 struct vge_rxsoft {
174 	struct mbuf	*rxs_mbuf;		/* head of our mbuf chain */
175 	bus_dmamap_t	rxs_dmamap;		/* our DMA map */
176 };
177 
178 
179 struct vge_softc {
180 	struct device		sc_dev;
181 
182 	bus_space_tag_t		sc_bst;		/* bus space tag */
183 	bus_space_handle_t	sc_bsh;		/* bus space handle */
184 	bus_dma_tag_t		sc_dmat;
185 
186 	struct ethercom		sc_ethercom;	/* interface info */
187 	uint8_t			sc_eaddr[ETHER_ADDR_LEN];
188 
189 	void			*sc_intrhand;
190 	struct mii_data		sc_mii;
191 	uint8_t			sc_type;
192 	int			sc_if_flags;
193 	int			sc_link;
194 	int			sc_camidx;
195 	struct callout		sc_timeout;
196 
197 	bus_dmamap_t		sc_cddmamap;
198 #define sc_cddma		sc_cddmamap->dm_segs[0].ds_addr
199 
200 	struct vge_txsoft	sc_txsoft[VGE_NTXDESC];
201 	struct vge_rxsoft	sc_rxsoft[VGE_NRXDESC];
202 	struct vge_control_data	*sc_control_data;
203 #define sc_txdescs		sc_control_data->vcd_txdescs
204 #define sc_rxdescs		sc_control_data->vcd_rxdescs
205 
206 	int			sc_tx_prodidx;
207 	int			sc_tx_considx;
208 	int			sc_tx_free;
209 
210 	struct mbuf		*sc_rx_mhead;
211 	struct mbuf		*sc_rx_mtail;
212 	int			sc_rx_prodidx;
213 	int			sc_rx_consumed;
214 
215 	int			sc_suspended;	/* 0 = normal  1 = suspended */
216 	uint32_t		sc_saved_maps[5];	/* pci data */
217 	uint32_t		sc_saved_biosaddr;
218 	uint8_t			sc_saved_intline;
219 	uint8_t			sc_saved_cachelnsz;
220 	uint8_t			sc_saved_lattimer;
221 };
222 
223 #define VGE_CDTXADDR(sc, x)	((sc)->sc_cddma + VGE_CDTXOFF(x))
224 #define VGE_CDRXADDR(sc, x)	((sc)->sc_cddma + VGE_CDRXOFF(x))
225 #define VGE_CDPADADDR(sc)	((sc)->sc_cddma + VGE_CDPADOFF())
226 
227 #define VGE_TXDESCSYNC(sc, idx, ops)					\
228 	bus_dmamap_sync((sc)->sc_dmat,(sc)->sc_cddmamap,		\
229 	    VGE_CDTXOFF(idx),						\
230 	    offsetof(struct vge_txdesc, td_frag[0]),			\
231 	    (ops))
232 #define VGE_TXFRAGSYNC(sc, idx, nsegs, ops)				\
233 	bus_dmamap_sync((sc)->sc_dmat, (sc)->sc_cddmamap,		\
234 	    VGE_CDTXOFF(idx) +						\
235 	    offsetof(struct vge_txdesc, td_frag[0]),			\
236 	    sizeof(struct vge_txfrag) * (nsegs),			\
237 	    (ops))
238 #define VGE_RXDESCSYNC(sc, idx, ops)					\
239 	bus_dmamap_sync((sc)->sc_dmat, (sc)->sc_cddmamap,		\
240 	    VGE_CDRXOFF(idx),						\
241 	    sizeof(struct vge_rxdesc),					\
242 	    (ops))
243 
244 /*
245  * Mbuf adjust factor to force 32-bit alignment of IP header.
246  * Drivers should do m_adj(m, ETHER_ALIGN) when setting up a
247  * receive so the upper layers get the IP header properly aligned
248  * past the 14-byte Ethernet header.
249  */
250 #define	ETHER_ALIGN	2
251 
252 #define	VGE_POWER_MANAGEMENT	0	/* disabled for now */
253 
254 /*
255  * register space access macros
256  */
257 #define CSR_WRITE_4(sc, reg, val)	\
258 	bus_space_write_4((sc)->sc_bst, (sc)->sc_bsh, (reg), (val))
259 #define CSR_WRITE_2(sc, reg, val)	\
260 	bus_space_write_2((sc)->sc_bst, (sc)->sc_bsh, (reg), (val))
261 #define CSR_WRITE_1(sc, reg, val)	\
262 	bus_space_write_1((sc)->sc_bst, (sc)->sc_bsh, (reg), (val))
263 
264 #define CSR_READ_4(sc, reg)		\
265 	bus_space_read_4((sc)->sc_bst, (sc)->sc_bsh, (reg))
266 #define CSR_READ_2(sc, reg)		\
267 	bus_space_read_2((sc)->sc_bst, (sc)->sc_bsh, (reg))
268 #define CSR_READ_1(sc, reg)		\
269 	bus_space_read_1((sc)->sc_bst, (sc)->sc_bsh, (reg))
270 
271 #define CSR_SETBIT_1(sc, reg, x)	\
272 	CSR_WRITE_1((sc), (reg), CSR_READ_1((sc), (reg)) | (x))
273 #define CSR_SETBIT_2(sc, reg, x)	\
274 	CSR_WRITE_2((sc), (reg), CSR_READ_2((sc), (reg)) | (x))
275 #define CSR_SETBIT_4(sc, reg, x)	\
276 	CSR_WRITE_4((sc), (reg), CSR_READ_4((sc), (reg)) | (x))
277 
278 #define CSR_CLRBIT_1(sc, reg, x)	\
279 	CSR_WRITE_1((sc), (reg), CSR_READ_1((sc), (reg)) & ~(x))
280 #define CSR_CLRBIT_2(sc, reg, x)	\
281 	CSR_WRITE_2((sc), (reg), CSR_READ_2((sc), (reg)) & ~(x))
282 #define CSR_CLRBIT_4(sc, reg, x)	\
283 	CSR_WRITE_4((sc), (reg), CSR_READ_4((sc), (reg)) & ~(x))
284 
285 #define VGE_TIMEOUT		10000
286 
287 #define VGE_PCI_LOIO             0x10
288 #define VGE_PCI_LOMEM            0x14
289 
290 static int vge_probe(struct device *, struct cfdata *, void *);
291 static void vge_attach(struct device *, struct device *, void *);
292 
293 static int vge_encap(struct vge_softc *, struct mbuf *, int);
294 
295 static int vge_allocmem(struct vge_softc *);
296 static int vge_newbuf(struct vge_softc *, int, struct mbuf *);
297 #ifndef __NO_STRICT_ALIGNMENT
298 static inline void vge_fixup_rx(struct mbuf *);
299 #endif
300 static void vge_rxeof(struct vge_softc *);
301 static void vge_txeof(struct vge_softc *);
302 static int vge_intr(void *);
303 static void vge_tick(void *);
304 static void vge_start(struct ifnet *);
305 static int vge_ioctl(struct ifnet *, u_long, caddr_t);
306 static int vge_init(struct ifnet *);
307 static void vge_stop(struct vge_softc *);
308 static void vge_watchdog(struct ifnet *);
309 #if VGE_POWER_MANAGEMENT
310 static int vge_suspend(struct device *);
311 static int vge_resume(struct device *);
312 #endif
313 static void vge_shutdown(void *);
314 static int vge_ifmedia_upd(struct ifnet *);
315 static void vge_ifmedia_sts(struct ifnet *, struct ifmediareq *);
316 
317 static uint16_t vge_read_eeprom(struct vge_softc *, int);
318 
319 static void vge_miipoll_start(struct vge_softc *);
320 static void vge_miipoll_stop(struct vge_softc *);
321 static int vge_miibus_readreg(struct device *, int, int);
322 static void vge_miibus_writereg(struct device *, int, int, int);
323 static void vge_miibus_statchg(struct device *);
324 
325 static void vge_cam_clear(struct vge_softc *);
326 static int vge_cam_set(struct vge_softc *, uint8_t *);
327 static void vge_setmulti(struct vge_softc *);
328 static void vge_reset(struct vge_softc *);
329 
330 CFATTACH_DECL(vge, sizeof(struct vge_softc),
331     vge_probe, vge_attach, NULL, NULL);
332 
333 /*
334  * Defragment mbuf chain contents to be as linear as possible.
335  * Returns new mbuf chain on success, NULL on failure. Old mbuf
336  * chain is always freed.
337  * XXX temporary until there would be generic function doing this.
338  */
339 #define m_defrag	vge_m_defrag
340 struct mbuf * vge_m_defrag(struct mbuf *, int);
341 
342 struct mbuf *
343 vge_m_defrag(struct mbuf *mold, int flags)
344 {
345 	struct mbuf *m0, *mn, *n;
346 	size_t sz = mold->m_pkthdr.len;
347 
348 #ifdef DIAGNOSTIC
349 	if ((mold->m_flags & M_PKTHDR) == 0)
350 		panic("m_defrag: not a mbuf chain header");
351 #endif
352 
353 	MGETHDR(m0, flags, MT_DATA);
354 	if (m0 == NULL)
355 		return NULL;
356 	m0->m_pkthdr.len = mold->m_pkthdr.len;
357 	mn = m0;
358 
359 	do {
360 		if (sz > MHLEN) {
361 			MCLGET(mn, M_DONTWAIT);
362 			if ((mn->m_flags & M_EXT) == 0) {
363 				m_freem(m0);
364 				return NULL;
365 			}
366 		}
367 
368 		mn->m_len = MIN(sz, MCLBYTES);
369 
370 		m_copydata(mold, mold->m_pkthdr.len - sz, mn->m_len,
371 		     mtod(mn, caddr_t));
372 
373 		sz -= mn->m_len;
374 
375 		if (sz > 0) {
376 			/* need more mbufs */
377 			MGET(n, M_NOWAIT, MT_DATA);
378 			if (n == NULL) {
379 				m_freem(m0);
380 				return NULL;
381 			}
382 
383 			mn->m_next = n;
384 			mn = n;
385 		}
386 	} while (sz > 0);
387 
388 	return m0;
389 }
390 
391 /*
392  * Read a word of data stored in the EEPROM at address 'addr.'
393  */
394 static uint16_t
395 vge_read_eeprom(struct vge_softc *sc, int addr)
396 {
397 	int i;
398 	uint16_t word = 0;
399 
400 	/*
401 	 * Enter EEPROM embedded programming mode. In order to
402 	 * access the EEPROM at all, we first have to set the
403 	 * EELOAD bit in the CHIPCFG2 register.
404 	 */
405 	CSR_SETBIT_1(sc, VGE_CHIPCFG2, VGE_CHIPCFG2_EELOAD);
406 	CSR_SETBIT_1(sc, VGE_EECSR, VGE_EECSR_EMBP/*|VGE_EECSR_ECS*/);
407 
408 	/* Select the address of the word we want to read */
409 	CSR_WRITE_1(sc, VGE_EEADDR, addr);
410 
411 	/* Issue read command */
412 	CSR_SETBIT_1(sc, VGE_EECMD, VGE_EECMD_ERD);
413 
414 	/* Wait for the done bit to be set. */
415 	for (i = 0; i < VGE_TIMEOUT; i++) {
416 		if (CSR_READ_1(sc, VGE_EECMD) & VGE_EECMD_EDONE)
417 			break;
418 	}
419 
420 	if (i == VGE_TIMEOUT) {
421 		aprint_error("%s: EEPROM read timed out\n",
422 		    sc->sc_dev.dv_xname);
423 		return 0;
424 	}
425 
426 	/* Read the result */
427 	word = CSR_READ_2(sc, VGE_EERDDAT);
428 
429 	/* Turn off EEPROM access mode. */
430 	CSR_CLRBIT_1(sc, VGE_EECSR, VGE_EECSR_EMBP/*|VGE_EECSR_ECS*/);
431 	CSR_CLRBIT_1(sc, VGE_CHIPCFG2, VGE_CHIPCFG2_EELOAD);
432 
433 	return word;
434 }
435 
436 static void
437 vge_miipoll_stop(struct vge_softc *sc)
438 {
439 	int i;
440 
441 	CSR_WRITE_1(sc, VGE_MIICMD, 0);
442 
443 	for (i = 0; i < VGE_TIMEOUT; i++) {
444 		DELAY(1);
445 		if (CSR_READ_1(sc, VGE_MIISTS) & VGE_MIISTS_IIDL)
446 			break;
447 	}
448 
449 	if (i == VGE_TIMEOUT) {
450 		aprint_error("%s: failed to idle MII autopoll\n",
451 		    sc->sc_dev.dv_xname);
452 	}
453 }
454 
455 static void
456 vge_miipoll_start(struct vge_softc *sc)
457 {
458 	int i;
459 
460 	/* First, make sure we're idle. */
461 
462 	CSR_WRITE_1(sc, VGE_MIICMD, 0);
463 	CSR_WRITE_1(sc, VGE_MIIADDR, VGE_MIIADDR_SWMPL);
464 
465 	for (i = 0; i < VGE_TIMEOUT; i++) {
466 		DELAY(1);
467 		if (CSR_READ_1(sc, VGE_MIISTS) & VGE_MIISTS_IIDL)
468 			break;
469 	}
470 
471 	if (i == VGE_TIMEOUT) {
472 		aprint_error("%s: failed to idle MII autopoll\n",
473 		    sc->sc_dev.dv_xname);
474 		return;
475 	}
476 
477 	/* Now enable auto poll mode. */
478 
479 	CSR_WRITE_1(sc, VGE_MIICMD, VGE_MIICMD_MAUTO);
480 
481 	/* And make sure it started. */
482 
483 	for (i = 0; i < VGE_TIMEOUT; i++) {
484 		DELAY(1);
485 		if ((CSR_READ_1(sc, VGE_MIISTS) & VGE_MIISTS_IIDL) == 0)
486 			break;
487 	}
488 
489 	if (i == VGE_TIMEOUT) {
490 		aprint_error("%s: failed to start MII autopoll\n",
491 		    sc->sc_dev.dv_xname);
492 	}
493 }
494 
495 static int
496 vge_miibus_readreg(struct device *dev, int phy, int reg)
497 {
498 	struct vge_softc *sc;
499 	int i, s;
500 	uint16_t rval;
501 
502 	sc = (void *)dev;
503 	rval = 0;
504 	if (phy != (CSR_READ_1(sc, VGE_MIICFG) & 0x1F))
505 		return 0;
506 
507 	s = splnet();
508 	vge_miipoll_stop(sc);
509 
510 	/* Specify the register we want to read. */
511 	CSR_WRITE_1(sc, VGE_MIIADDR, reg);
512 
513 	/* Issue read command. */
514 	CSR_SETBIT_1(sc, VGE_MIICMD, VGE_MIICMD_RCMD);
515 
516 	/* Wait for the read command bit to self-clear. */
517 	for (i = 0; i < VGE_TIMEOUT; i++) {
518 		DELAY(1);
519 		if ((CSR_READ_1(sc, VGE_MIICMD) & VGE_MIICMD_RCMD) == 0)
520 			break;
521 	}
522 
523 	if (i == VGE_TIMEOUT)
524 		aprint_error("%s: MII read timed out\n", sc->sc_dev.dv_xname);
525 	else
526 		rval = CSR_READ_2(sc, VGE_MIIDATA);
527 
528 	vge_miipoll_start(sc);
529 	splx(s);
530 
531 	return rval;
532 }
533 
534 static void
535 vge_miibus_writereg(struct device *dev, int phy, int reg, int data)
536 {
537 	struct vge_softc *sc;
538 	int i, s;
539 
540 	sc = (void *)dev;
541 	if (phy != (CSR_READ_1(sc, VGE_MIICFG) & 0x1F))
542 		return;
543 
544 	s = splnet();
545 	vge_miipoll_stop(sc);
546 
547 	/* Specify the register we want to write. */
548 	CSR_WRITE_1(sc, VGE_MIIADDR, reg);
549 
550 	/* Specify the data we want to write. */
551 	CSR_WRITE_2(sc, VGE_MIIDATA, data);
552 
553 	/* Issue write command. */
554 	CSR_SETBIT_1(sc, VGE_MIICMD, VGE_MIICMD_WCMD);
555 
556 	/* Wait for the write command bit to self-clear. */
557 	for (i = 0; i < VGE_TIMEOUT; i++) {
558 		DELAY(1);
559 		if ((CSR_READ_1(sc, VGE_MIICMD) & VGE_MIICMD_WCMD) == 0)
560 			break;
561 	}
562 
563 	if (i == VGE_TIMEOUT) {
564 		aprint_error("%s: MII write timed out\n", sc->sc_dev.dv_xname);
565 	}
566 
567 	vge_miipoll_start(sc);
568 	splx(s);
569 }
570 
571 static void
572 vge_cam_clear(struct vge_softc *sc)
573 {
574 	int i;
575 
576 	/*
577 	 * Turn off all the mask bits. This tells the chip
578 	 * that none of the entries in the CAM filter are valid.
579 	 * desired entries will be enabled as we fill the filter in.
580 	 */
581 
582 	CSR_CLRBIT_1(sc, VGE_CAMCTL, VGE_CAMCTL_PAGESEL);
583 	CSR_SETBIT_1(sc, VGE_CAMCTL, VGE_PAGESEL_CAMMASK);
584 	CSR_WRITE_1(sc, VGE_CAMADDR, VGE_CAMADDR_ENABLE);
585 	for (i = 0; i < 8; i++)
586 		CSR_WRITE_1(sc, VGE_CAM0 + i, 0);
587 
588 	/* Clear the VLAN filter too. */
589 
590 	CSR_WRITE_1(sc, VGE_CAMADDR, VGE_CAMADDR_ENABLE|VGE_CAMADDR_AVSEL|0);
591 	for (i = 0; i < 8; i++)
592 		CSR_WRITE_1(sc, VGE_CAM0 + i, 0);
593 
594 	CSR_WRITE_1(sc, VGE_CAMADDR, 0);
595 	CSR_CLRBIT_1(sc, VGE_CAMCTL, VGE_CAMCTL_PAGESEL);
596 	CSR_SETBIT_1(sc, VGE_CAMCTL, VGE_PAGESEL_MAR);
597 
598 	sc->sc_camidx = 0;
599 }
600 
601 static int
602 vge_cam_set(struct vge_softc *sc, uint8_t *addr)
603 {
604 	int i, error;
605 
606 	error = 0;
607 
608 	if (sc->sc_camidx == VGE_CAM_MAXADDRS)
609 		return ENOSPC;
610 
611 	/* Select the CAM data page. */
612 	CSR_CLRBIT_1(sc, VGE_CAMCTL, VGE_CAMCTL_PAGESEL);
613 	CSR_SETBIT_1(sc, VGE_CAMCTL, VGE_PAGESEL_CAMDATA);
614 
615 	/* Set the filter entry we want to update and enable writing. */
616 	CSR_WRITE_1(sc, VGE_CAMADDR, VGE_CAMADDR_ENABLE | sc->sc_camidx);
617 
618 	/* Write the address to the CAM registers */
619 	for (i = 0; i < ETHER_ADDR_LEN; i++)
620 		CSR_WRITE_1(sc, VGE_CAM0 + i, addr[i]);
621 
622 	/* Issue a write command. */
623 	CSR_SETBIT_1(sc, VGE_CAMCTL, VGE_CAMCTL_WRITE);
624 
625 	/* Wake for it to clear. */
626 	for (i = 0; i < VGE_TIMEOUT; i++) {
627 		DELAY(1);
628 		if ((CSR_READ_1(sc, VGE_CAMCTL) & VGE_CAMCTL_WRITE) == 0)
629 			break;
630 	}
631 
632 	if (i == VGE_TIMEOUT) {
633 		aprint_error("%s: setting CAM filter failed\n",
634 		    sc->sc_dev.dv_xname);
635 		error = EIO;
636 		goto fail;
637 	}
638 
639 	/* Select the CAM mask page. */
640 	CSR_CLRBIT_1(sc, VGE_CAMCTL, VGE_CAMCTL_PAGESEL);
641 	CSR_SETBIT_1(sc, VGE_CAMCTL, VGE_PAGESEL_CAMMASK);
642 
643 	/* Set the mask bit that enables this filter. */
644 	CSR_SETBIT_1(sc, VGE_CAM0 + (sc->sc_camidx / 8),
645 	    1 << (sc->sc_camidx & 7));
646 
647 	sc->sc_camidx++;
648 
649  fail:
650 	/* Turn off access to CAM. */
651 	CSR_WRITE_1(sc, VGE_CAMADDR, 0);
652 	CSR_CLRBIT_1(sc, VGE_CAMCTL, VGE_CAMCTL_PAGESEL);
653 	CSR_SETBIT_1(sc, VGE_CAMCTL, VGE_PAGESEL_MAR);
654 
655 	return error;
656 }
657 
658 /*
659  * Program the multicast filter. We use the 64-entry CAM filter
660  * for perfect filtering. If there's more than 64 multicast addresses,
661  * we use the hash filter instead.
662  */
663 static void
664 vge_setmulti(struct vge_softc *sc)
665 {
666 	struct ifnet *ifp;
667 	int error;
668 	uint32_t h, hashes[2] = { 0, 0 };
669 	struct ether_multi *enm;
670 	struct ether_multistep step;
671 
672 	error = 0;
673 	ifp = &sc->sc_ethercom.ec_if;
674 
675 	/* First, zot all the multicast entries. */
676 	vge_cam_clear(sc);
677 	CSR_WRITE_4(sc, VGE_MAR0, 0);
678 	CSR_WRITE_4(sc, VGE_MAR1, 0);
679 	ifp->if_flags &= ~IFF_ALLMULTI;
680 
681 	/*
682 	 * If the user wants allmulti or promisc mode, enable reception
683 	 * of all multicast frames.
684 	 */
685 	if (ifp->if_flags & IFF_PROMISC) {
686  allmulti:
687 		CSR_WRITE_4(sc, VGE_MAR0, 0xFFFFFFFF);
688 		CSR_WRITE_4(sc, VGE_MAR1, 0xFFFFFFFF);
689 		ifp->if_flags |= IFF_ALLMULTI;
690 		return;
691 	}
692 
693 	/* Now program new ones */
694 	ETHER_FIRST_MULTI(step, &sc->sc_ethercom, enm);
695 	while (enm != NULL) {
696 		/*
697 		 * If multicast range, fall back to ALLMULTI.
698 		 */
699 		if (memcmp(enm->enm_addrlo, enm->enm_addrhi,
700 				ETHER_ADDR_LEN) != 0)
701 			goto allmulti;
702 
703 		error = vge_cam_set(sc, enm->enm_addrlo);
704 		if (error)
705 			break;
706 
707 		ETHER_NEXT_MULTI(step, enm);
708 	}
709 
710 	/* If there were too many addresses, use the hash filter. */
711 	if (error) {
712 		vge_cam_clear(sc);
713 
714 		ETHER_FIRST_MULTI(step, &sc->sc_ethercom, enm);
715 		while (enm != NULL) {
716 			/*
717 			 * If multicast range, fall back to ALLMULTI.
718 			 */
719 			if (memcmp(enm->enm_addrlo, enm->enm_addrhi,
720 					ETHER_ADDR_LEN) != 0)
721 				goto allmulti;
722 
723 			h = ether_crc32_be(enm->enm_addrlo,
724 			    ETHER_ADDR_LEN) >> 26;
725 			hashes[h >> 5] |= 1 << (h & 0x1f);
726 
727 			ETHER_NEXT_MULTI(step, enm);
728 		}
729 
730 		CSR_WRITE_4(sc, VGE_MAR0, hashes[0]);
731 		CSR_WRITE_4(sc, VGE_MAR1, hashes[1]);
732 	}
733 }
734 
735 static void
736 vge_reset(struct vge_softc *sc)
737 {
738 	int i;
739 
740 	CSR_WRITE_1(sc, VGE_CRS1, VGE_CR1_SOFTRESET);
741 
742 	for (i = 0; i < VGE_TIMEOUT; i++) {
743 		DELAY(5);
744 		if ((CSR_READ_1(sc, VGE_CRS1) & VGE_CR1_SOFTRESET) == 0)
745 			break;
746 	}
747 
748 	if (i == VGE_TIMEOUT) {
749 		aprint_error("%s: soft reset timed out", sc->sc_dev.dv_xname);
750 		CSR_WRITE_1(sc, VGE_CRS3, VGE_CR3_STOP_FORCE);
751 		DELAY(2000);
752 	}
753 
754 	DELAY(5000);
755 
756 	CSR_SETBIT_1(sc, VGE_EECSR, VGE_EECSR_RELOAD);
757 
758 	for (i = 0; i < VGE_TIMEOUT; i++) {
759 		DELAY(5);
760 		if ((CSR_READ_1(sc, VGE_EECSR) & VGE_EECSR_RELOAD) == 0)
761 			break;
762 	}
763 
764 	if (i == VGE_TIMEOUT) {
765 		aprint_error("%s: EEPROM reload timed out\n",
766 		    sc->sc_dev.dv_xname);
767 		return;
768 	}
769 
770 	/*
771 	 * On some machine, the first read data from EEPROM could be
772 	 * messed up, so read one dummy data here to avoid the mess.
773 	 */
774 	(void)vge_read_eeprom(sc, 0);
775 
776 	CSR_CLRBIT_1(sc, VGE_CHIPCFG0, VGE_CHIPCFG0_PACPI);
777 }
778 
779 /*
780  * Probe for a VIA gigabit chip. Check the PCI vendor and device
781  * IDs against our list and return a device name if we find a match.
782  */
783 static int
784 vge_probe(struct device *parent, struct cfdata *match,
785     void *aux)
786 {
787 	struct pci_attach_args *pa = aux;
788 
789 	if (PCI_VENDOR(pa->pa_id) == PCI_VENDOR_VIATECH
790 	    && PCI_PRODUCT(pa->pa_id) == PCI_PRODUCT_VIATECH_VT612X)
791 		return 1;
792 
793 	return 0;
794 }
795 
796 static int
797 vge_allocmem(struct vge_softc *sc)
798 {
799 	int error;
800 	int nseg;
801 	int i;
802 	bus_dma_segment_t seg;
803 
804 	/*
805 	 * Allocate memory for control data.
806 	 */
807 
808 	error = bus_dmamem_alloc(sc->sc_dmat, sizeof(struct vge_control_data),
809 	     VGE_RING_ALIGN, 0, &seg, 1, &nseg, BUS_DMA_NOWAIT);
810 	if (error) {
811 		aprint_error("%s: could not allocate control data dma memory\n",
812 		    sc->sc_dev.dv_xname);
813 		return ENOMEM;
814 	}
815 
816 	/* Map the memory to kernel VA space */
817 
818 	error = bus_dmamem_map(sc->sc_dmat, &seg, nseg,
819 	    sizeof(struct vge_control_data), (caddr_t *)&sc->sc_control_data,
820 	    BUS_DMA_NOWAIT);
821 	if (error) {
822 		aprint_error("%s: could not map control data dma memory\n",
823 		    sc->sc_dev.dv_xname);
824 		return ENOMEM;
825 	}
826 	memset(sc->sc_control_data, 0, sizeof(struct vge_control_data));
827 
828 	/*
829 	 * Create map for control data.
830 	 */
831 	error = bus_dmamap_create(sc->sc_dmat,
832 	    sizeof(struct vge_control_data), 1,
833 	    sizeof(struct vge_control_data), 0, BUS_DMA_NOWAIT,
834 	    &sc->sc_cddmamap);
835 	if (error) {
836 		aprint_error("%s: could not create control data dmamap\n",
837 		    sc->sc_dev.dv_xname);
838 		return ENOMEM;
839 	}
840 
841 	/* Load the map for the control data. */
842 	error = bus_dmamap_load(sc->sc_dmat, sc->sc_cddmamap,
843 	    sc->sc_control_data, sizeof(struct vge_control_data), NULL,
844 	    BUS_DMA_NOWAIT);
845 	if (error) {
846 		aprint_error("%s: could not load control data dma memory\n",
847 		    sc->sc_dev.dv_xname);
848 		return ENOMEM;
849 	}
850 
851 	/* Create DMA maps for TX buffers */
852 
853 	for (i = 0; i < VGE_NTXDESC; i++) {
854 		error = bus_dmamap_create(sc->sc_dmat, VGE_TX_MAXLEN,
855 		    VGE_TX_FRAGS, VGE_TX_MAXLEN, 0, BUS_DMA_NOWAIT,
856 		    &sc->sc_txsoft[i].txs_dmamap);
857 		if (error) {
858 			aprint_error("%s: can't create DMA map for TX descs\n",
859 			    sc->sc_dev.dv_xname);
860 			return ENOMEM;
861 		}
862 	}
863 
864 	/* Create DMA maps for RX buffers */
865 
866 	for (i = 0; i < VGE_NRXDESC; i++) {
867 		error = bus_dmamap_create(sc->sc_dmat, MCLBYTES,
868 		    1, MCLBYTES, 0, BUS_DMA_NOWAIT,
869 		    &sc->sc_rxsoft[i].rxs_dmamap);
870 		if (error) {
871 			aprint_error("%s: can't create DMA map for RX descs\n",
872 			    sc->sc_dev.dv_xname);
873 			return ENOMEM;
874 		}
875 		sc->sc_rxsoft[i].rxs_mbuf = NULL;
876 	}
877 
878 	return 0;
879 }
880 
881 /*
882  * Attach the interface. Allocate softc structures, do ifmedia
883  * setup and ethernet/BPF attach.
884  */
885 static void
886 vge_attach(struct device *parent, struct device *self, void *aux)
887 {
888 	uint8_t	*eaddr;
889 	struct vge_softc *sc = (void *)self;
890 	struct ifnet *ifp;
891 	struct pci_attach_args *pa = aux;
892 	pci_chipset_tag_t pc = pa->pa_pc;
893 	const char *intrstr;
894 	pci_intr_handle_t ih;
895 	uint16_t val;
896 
897 	aprint_normal(": VIA VT612X Gigabit Ethernet (rev. %#x)\n",
898 		PCI_REVISION(pa->pa_class));
899 
900 	/* Make sure bus-mastering is enabled */
901         pci_conf_write(pc, pa->pa_tag, PCI_COMMAND_STATUS_REG,
902 	    pci_conf_read(pc, pa->pa_tag, PCI_COMMAND_STATUS_REG) |
903 	    PCI_COMMAND_MASTER_ENABLE);
904 
905 	/*
906 	 * Map control/status registers.
907 	 */
908 	if (pci_mapreg_map(pa, VGE_PCI_LOMEM, PCI_MAPREG_TYPE_MEM, 0,
909 	    &sc->sc_bst, &sc->sc_bsh, NULL, NULL) != 0) {
910 		aprint_error("%s: couldn't map memory\n", sc->sc_dev.dv_xname);
911 		return;
912 	}
913 
914         /*
915          * Map and establish our interrupt.
916          */
917 	if (pci_intr_map(pa, &ih)) {
918 		aprint_error("%s: unable to map interrupt\n",
919 		    sc->sc_dev.dv_xname);
920 		return;
921 	}
922 	intrstr = pci_intr_string(pc, ih);
923 	sc->sc_intrhand = pci_intr_establish(pc, ih, IPL_NET, vge_intr, sc);
924 	if (sc->sc_intrhand == NULL) {
925 		aprint_error("%s: unable to establish interrupt",
926 		    sc->sc_dev.dv_xname);
927 		if (intrstr != NULL)
928 			aprint_error(" at %s", intrstr);
929 		aprint_error("\n");
930 		return;
931 	}
932 	aprint_normal("%s: interrupting at %s\n", sc->sc_dev.dv_xname, intrstr);
933 
934 	/* Reset the adapter. */
935 	vge_reset(sc);
936 
937 	/*
938 	 * Get station address from the EEPROM.
939 	 */
940 	eaddr = sc->sc_eaddr;
941 	val = vge_read_eeprom(sc, VGE_EE_EADDR + 0);
942 	eaddr[0] = val & 0xff;
943 	eaddr[1] = val >> 8;
944 	val = vge_read_eeprom(sc, VGE_EE_EADDR + 1);
945 	eaddr[2] = val & 0xff;
946 	eaddr[3] = val >> 8;
947 	val = vge_read_eeprom(sc, VGE_EE_EADDR + 2);
948 	eaddr[4] = val & 0xff;
949 	eaddr[5] = val >> 8;
950 
951 	aprint_normal("%s: Ethernet address: %s\n", sc->sc_dev.dv_xname,
952 	    ether_sprintf(eaddr));
953 
954 	/*
955 	 * Use the 32bit tag. Hardware supports 48bit physical addresses,
956 	 * but we don't use that for now.
957 	 */
958 	sc->sc_dmat = pa->pa_dmat;
959 
960 	if (vge_allocmem(sc))
961 		return;
962 
963 	ifp = &sc->sc_ethercom.ec_if;
964 	ifp->if_softc = sc;
965 	strcpy(ifp->if_xname, sc->sc_dev.dv_xname);
966 	ifp->if_mtu = ETHERMTU;
967 	ifp->if_baudrate = IF_Gbps(1);
968 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
969 	ifp->if_ioctl = vge_ioctl;
970 	ifp->if_start = vge_start;
971 
972 	/*
973 	 * We can support 802.1Q VLAN-sized frames and jumbo
974 	 * Ethernet frames.
975 	 */
976 	sc->sc_ethercom.ec_capabilities |=
977 	    ETHERCAP_VLAN_MTU | ETHERCAP_JUMBO_MTU |
978 	    ETHERCAP_VLAN_HWTAGGING;
979 
980 	/*
981 	 * We can do IPv4/TCPv4/UDPv4 checksums in hardware.
982 	 */
983 	ifp->if_capabilities |=
984 	    IFCAP_CSUM_IPv4_Tx | IFCAP_CSUM_IPv4_Rx |
985 	    IFCAP_CSUM_TCPv4_Tx | IFCAP_CSUM_TCPv4_Rx |
986 	    IFCAP_CSUM_UDPv4_Tx | IFCAP_CSUM_UDPv4_Rx;
987 
988 #ifdef DEVICE_POLLING
989 #ifdef IFCAP_POLLING
990 	ifp->if_capabilities |= IFCAP_POLLING;
991 #endif
992 #endif
993 	ifp->if_watchdog = vge_watchdog;
994 	ifp->if_init = vge_init;
995 	IFQ_SET_MAXLEN(&ifp->if_snd, max(VGE_IFQ_MAXLEN, IFQ_MAXLEN));
996 
997 	/*
998 	 * Initialize our media structures and probe the MII.
999 	 */
1000 	sc->sc_mii.mii_ifp = ifp;
1001 	sc->sc_mii.mii_readreg = vge_miibus_readreg;
1002 	sc->sc_mii.mii_writereg = vge_miibus_writereg;
1003 	sc->sc_mii.mii_statchg = vge_miibus_statchg;
1004 	ifmedia_init(&sc->sc_mii.mii_media, 0, vge_ifmedia_upd,
1005 	    vge_ifmedia_sts);
1006 	mii_attach(&sc->sc_dev, &sc->sc_mii, 0xffffffff, MII_PHY_ANY,
1007 	    MII_OFFSET_ANY, MIIF_DOPAUSE);
1008 	if (LIST_FIRST(&sc->sc_mii.mii_phys) == NULL) {
1009 		ifmedia_add(&sc->sc_mii.mii_media, IFM_ETHER|IFM_NONE, 0, NULL);
1010 		ifmedia_set(&sc->sc_mii.mii_media, IFM_ETHER|IFM_NONE);
1011 	} else
1012 		ifmedia_set(&sc->sc_mii.mii_media, IFM_ETHER|IFM_AUTO);
1013 
1014 	/*
1015 	 * Attach the interface.
1016 	 */
1017 	if_attach(ifp);
1018 	ether_ifattach(ifp, eaddr);
1019 
1020 	callout_init(&sc->sc_timeout);
1021 	callout_setfunc(&sc->sc_timeout, vge_tick, sc);
1022 
1023 	/*
1024 	 * Make sure the interface is shutdown during reboot.
1025 	 */
1026 	if (shutdownhook_establish(vge_shutdown, sc) == NULL) {
1027 		aprint_error("%s: WARNING: unable to establish shutdown hook\n",
1028 		    sc->sc_dev.dv_xname);
1029 	}
1030 }
1031 
1032 static int
1033 vge_newbuf(struct vge_softc *sc, int idx, struct mbuf *m)
1034 {
1035 	struct mbuf *m_new;
1036 	struct vge_rxdesc *rxd;
1037 	struct vge_rxsoft *rxs;
1038 	bus_dmamap_t map;
1039 	int i;
1040 
1041 	m_new = NULL;
1042 	if (m == NULL) {
1043 		MGETHDR(m_new, M_DONTWAIT, MT_DATA);
1044 		if (m_new == NULL)
1045 			return ENOBUFS;
1046 
1047 		MCLGET(m_new, M_DONTWAIT);
1048 		if ((m_new->m_flags & M_EXT) == 0) {
1049 			m_freem(m_new);
1050 			return ENOBUFS;
1051 		}
1052 
1053 		m = m_new;
1054 	} else
1055 		m->m_data = m->m_ext.ext_buf;
1056 
1057 
1058 #ifndef __NO_STRICT_ALIGNMENT
1059 	/*
1060 	 * This is part of an evil trick to deal with non-x86 platforms.
1061 	 * The VIA chip requires RX buffers to be aligned on 32-bit
1062 	 * boundaries, but that will hose non-x86 machines. To get around
1063 	 * this, we leave some empty space at the start of each buffer
1064 	 * and for non-x86 hosts, we copy the buffer back two bytes
1065 	 * to achieve word alignment. This is slightly more efficient
1066 	 * than allocating a new buffer, copying the contents, and
1067 	 * discarding the old buffer.
1068 	 */
1069 	m->m_len = m->m_pkthdr.len = MCLBYTES - VGE_RX_PAD;
1070 	m->m_data += VGE_RX_PAD;
1071 #else
1072 	m->m_len = m->m_pkthdr.len = MCLBYTES;
1073 #endif
1074 	rxs = &sc->sc_rxsoft[idx];
1075 	map = rxs->rxs_dmamap;
1076 
1077 	if (bus_dmamap_load_mbuf(sc->sc_dmat, map, m, BUS_DMA_NOWAIT) != 0)
1078 		goto out;
1079 
1080 	rxd = &sc->sc_rxdescs[idx];
1081 
1082 	/* If this descriptor is still owned by the chip, bail. */
1083 
1084 	VGE_RXDESCSYNC(sc, idx, BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1085 	if (le32toh(rxd->rd_sts) & VGE_RDSTS_OWN) {
1086 		aprint_error("%s: tried to map busy RX descriptor\n",
1087 		    sc->sc_dev.dv_xname);
1088 		VGE_RXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD);
1089 		goto out;
1090 	}
1091 
1092 	rxs->rxs_mbuf = m;
1093 	bus_dmamap_sync(sc->sc_dmat, map, 0, map->dm_mapsize,
1094 	    BUS_DMASYNC_PREREAD);
1095 
1096 	rxd->rd_buflen =
1097 	    htole16(VGE_BUFLEN(map->dm_segs[0].ds_len) | VGE_RXDESC_I);
1098 	rxd->rd_addrlo = htole32(VGE_ADDR_LO(map->dm_segs[0].ds_addr));
1099 	rxd->rd_addrhi = htole16(VGE_ADDR_HI(map->dm_segs[0].ds_addr) & 0xFFFF);
1100 	rxd->rd_sts = 0;
1101 	rxd->rd_ctl = 0;
1102 	VGE_RXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1103 
1104 	/*
1105 	 * Note: the manual fails to document the fact that for
1106 	 * proper opration, the driver needs to replentish the RX
1107 	 * DMA ring 4 descriptors at a time (rather than one at a
1108 	 * time, like most chips). We can allocate the new buffers
1109 	 * but we should not set the OWN bits until we're ready
1110 	 * to hand back 4 of them in one shot.
1111 	 */
1112 
1113 #define VGE_RXCHUNK 4
1114 	sc->sc_rx_consumed++;
1115 	if (sc->sc_rx_consumed == VGE_RXCHUNK) {
1116 		for (i = idx; i != idx - VGE_RXCHUNK; i--) {
1117 			KASSERT(i >= 0);
1118 			sc->sc_rxdescs[i].rd_sts |= htole32(VGE_RDSTS_OWN);
1119 			VGE_RXDESCSYNC(sc, i,
1120 			    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1121 		}
1122 		sc->sc_rx_consumed = 0;
1123 	}
1124 
1125 	return 0;
1126  out:
1127 	if (m_new != NULL)
1128 		m_freem(m_new);
1129 	return ENOMEM;
1130 }
1131 
1132 #ifndef __NO_STRICT_ALIGNMENT
1133 static inline void
1134 vge_fixup_rx(struct mbuf *m)
1135 {
1136 	int i;
1137 	uint16_t *src, *dst;
1138 
1139 	src = mtod(m, uint16_t *);
1140 	dst = src - 1;
1141 
1142 	for (i = 0; i < (m->m_len / sizeof(uint16_t) + 1); i++)
1143 		*dst++ = *src++;
1144 
1145 	m->m_data -= ETHER_ALIGN;
1146 }
1147 #endif
1148 
1149 /*
1150  * RX handler. We support the reception of jumbo frames that have
1151  * been fragmented across multiple 2K mbuf cluster buffers.
1152  */
1153 static void
1154 vge_rxeof(struct vge_softc *sc)
1155 {
1156 	struct mbuf *m;
1157 	struct ifnet *ifp;
1158 	int idx, total_len, lim;
1159 	struct vge_rxdesc *cur_rxd;
1160 	struct vge_rxsoft *rxs;
1161 	uint32_t rxstat, rxctl;
1162 
1163 	ifp = &sc->sc_ethercom.ec_if;
1164 	lim = 0;
1165 
1166 	/* Invalidate the descriptor memory */
1167 
1168 	for (idx = sc->sc_rx_prodidx;; idx = VGE_NEXT_RXDESC(idx)) {
1169 		cur_rxd = &sc->sc_rxdescs[idx];
1170 
1171 		VGE_RXDESCSYNC(sc, idx,
1172 		    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1173 		rxstat = le32toh(cur_rxd->rd_sts);
1174 		if ((rxstat & VGE_RDSTS_OWN) != 0) {
1175 			VGE_RXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD);
1176 			break;
1177 		}
1178 
1179 		rxctl = le32toh(cur_rxd->rd_ctl);
1180 		rxs = &sc->sc_rxsoft[idx];
1181 		m = rxs->rxs_mbuf;
1182 		total_len = (rxstat & VGE_RDSTS_BUFSIZ) >> 16;
1183 
1184 		/* Invalidate the RX mbuf and unload its map */
1185 
1186 		bus_dmamap_sync(sc->sc_dmat, rxs->rxs_dmamap,
1187 		    0, rxs->rxs_dmamap->dm_mapsize, BUS_DMASYNC_POSTREAD);
1188 		bus_dmamap_unload(sc->sc_dmat, rxs->rxs_dmamap);
1189 
1190 		/*
1191 		 * If the 'start of frame' bit is set, this indicates
1192 		 * either the first fragment in a multi-fragment receive,
1193 		 * or an intermediate fragment. Either way, we want to
1194 		 * accumulate the buffers.
1195 		 */
1196 		if (rxstat & VGE_RXPKT_SOF) {
1197 			m->m_len = MCLBYTES - VGE_RX_PAD;
1198 			if (sc->sc_rx_mhead == NULL)
1199 				sc->sc_rx_mhead = sc->sc_rx_mtail = m;
1200 			else {
1201 				m->m_flags &= ~M_PKTHDR;
1202 				sc->sc_rx_mtail->m_next = m;
1203 				sc->sc_rx_mtail = m;
1204 			}
1205 			vge_newbuf(sc, idx, NULL);
1206 			continue;
1207 		}
1208 
1209 		/*
1210 		 * Bad/error frames will have the RXOK bit cleared.
1211 		 * However, there's one error case we want to allow:
1212 		 * if a VLAN tagged frame arrives and the chip can't
1213 		 * match it against the CAM filter, it considers this
1214 		 * a 'VLAN CAM filter miss' and clears the 'RXOK' bit.
1215 		 * We don't want to drop the frame though: our VLAN
1216 		 * filtering is done in software.
1217 		 */
1218 		if (!(rxstat & VGE_RDSTS_RXOK) && !(rxstat & VGE_RDSTS_VIDM)
1219 		    && !(rxstat & VGE_RDSTS_CSUMERR)) {
1220 			ifp->if_ierrors++;
1221 			/*
1222 			 * If this is part of a multi-fragment packet,
1223 			 * discard all the pieces.
1224 			 */
1225 			if (sc->sc_rx_mhead != NULL) {
1226 				m_freem(sc->sc_rx_mhead);
1227 				sc->sc_rx_mhead = sc->sc_rx_mtail = NULL;
1228 			}
1229 			vge_newbuf(sc, idx, m);
1230 			continue;
1231 		}
1232 
1233 		/*
1234 		 * If allocating a replacement mbuf fails,
1235 		 * reload the current one.
1236 		 */
1237 
1238 		if (vge_newbuf(sc, idx, NULL)) {
1239 			ifp->if_ierrors++;
1240 			if (sc->sc_rx_mhead != NULL) {
1241 				m_freem(sc->sc_rx_mhead);
1242 				sc->sc_rx_mhead = sc->sc_rx_mtail = NULL;
1243 			}
1244 			vge_newbuf(sc, idx, m);
1245 			continue;
1246 		}
1247 
1248 		if (sc->sc_rx_mhead != NULL) {
1249 			m->m_len = total_len % (MCLBYTES - VGE_RX_PAD);
1250 			/*
1251 			 * Special case: if there's 4 bytes or less
1252 			 * in this buffer, the mbuf can be discarded:
1253 			 * the last 4 bytes is the CRC, which we don't
1254 			 * care about anyway.
1255 			 */
1256 			if (m->m_len <= ETHER_CRC_LEN) {
1257 				sc->sc_rx_mtail->m_len -=
1258 				    (ETHER_CRC_LEN - m->m_len);
1259 				m_freem(m);
1260 			} else {
1261 				m->m_len -= ETHER_CRC_LEN;
1262 				m->m_flags &= ~M_PKTHDR;
1263 				sc->sc_rx_mtail->m_next = m;
1264 			}
1265 			m = sc->sc_rx_mhead;
1266 			sc->sc_rx_mhead = sc->sc_rx_mtail = NULL;
1267 			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
1268 		} else
1269 			m->m_pkthdr.len = m->m_len = total_len - ETHER_CRC_LEN;
1270 
1271 #ifndef __NO_STRICT_ALIGNMENT
1272 		vge_fixup_rx(m);
1273 #endif
1274 		ifp->if_ipackets++;
1275 		m->m_pkthdr.rcvif = ifp;
1276 
1277 		/* Do RX checksumming if enabled */
1278 		if (ifp->if_csum_flags_rx & M_CSUM_IPv4) {
1279 
1280 			/* Check IP header checksum */
1281 			if (rxctl & VGE_RDCTL_IPPKT)
1282 				m->m_pkthdr.csum_flags |= M_CSUM_IPv4;
1283 			if ((rxctl & VGE_RDCTL_IPCSUMOK) == 0)
1284 				m->m_pkthdr.csum_flags |= M_CSUM_IPv4_BAD;
1285 		}
1286 
1287 		if (ifp->if_csum_flags_rx & M_CSUM_TCPv4) {
1288 			/* Check UDP checksum */
1289 			if (rxctl & VGE_RDCTL_TCPPKT)
1290 				m->m_pkthdr.csum_flags |= M_CSUM_TCPv4;
1291 
1292 			if ((rxctl & VGE_RDCTL_PROTOCSUMOK) == 0)
1293 				m->m_pkthdr.csum_flags |= M_CSUM_TCP_UDP_BAD;
1294 		}
1295 
1296 		if (ifp->if_csum_flags_rx & M_CSUM_UDPv4) {
1297 			/* Check UDP checksum */
1298 			if (rxctl & VGE_RDCTL_UDPPKT)
1299 				m->m_pkthdr.csum_flags |= M_CSUM_UDPv4;
1300 
1301 			if ((rxctl & VGE_RDCTL_PROTOCSUMOK) == 0)
1302 				m->m_pkthdr.csum_flags |= M_CSUM_TCP_UDP_BAD;
1303 		}
1304 
1305 		if (rxstat & VGE_RDSTS_VTAG) {
1306 			/*
1307 			 * We use bswap16() here because:
1308 			 * On LE machines, tag is stored in BE as stream data.
1309 			 * On BE machines, tag is stored in BE as stream data
1310 			 *  but it was already swapped by le32toh() above.
1311 			 */
1312 			VLAN_INPUT_TAG(ifp, m,
1313 			    bswap16(rxctl & VGE_RDCTL_VLANID), continue);
1314 		}
1315 
1316 #if NBPFILTER > 0
1317 		/*
1318 		 * Handle BPF listeners.
1319 		 */
1320 		if (ifp->if_bpf)
1321 			bpf_mtap(ifp->if_bpf, m);
1322 #endif
1323 
1324 		(*ifp->if_input)(ifp, m);
1325 
1326 		lim++;
1327 		if (lim == VGE_NRXDESC)
1328 			break;
1329 
1330 	}
1331 
1332 	sc->sc_rx_prodidx = idx;
1333 	CSR_WRITE_2(sc, VGE_RXDESC_RESIDUECNT, lim);
1334 }
1335 
1336 static void
1337 vge_txeof(struct vge_softc *sc)
1338 {
1339 	struct ifnet *ifp;
1340 	struct vge_txsoft *txs;
1341 	uint32_t txstat;
1342 	int idx;
1343 
1344 	ifp = &sc->sc_ethercom.ec_if;
1345 	idx = sc->sc_tx_considx;
1346 
1347 	for (idx = sc->sc_tx_considx;
1348 	    idx != sc->sc_tx_prodidx;
1349 	    idx = VGE_NEXT_TXDESC(idx)) {
1350 		VGE_TXDESCSYNC(sc, idx,
1351 		    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1352 		txstat = le32toh(sc->sc_txdescs[idx].td_sts);
1353 		if (txstat & VGE_TDSTS_OWN) {
1354 			VGE_TXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD);
1355 			break;
1356 		}
1357 
1358 		txs = &sc->sc_txsoft[idx];
1359 		m_freem(txs->txs_mbuf);
1360 		txs->txs_mbuf = NULL;
1361 		bus_dmamap_sync(sc->sc_dmat, txs->txs_dmamap, 0,
1362 		    txs->txs_dmamap->dm_mapsize, BUS_DMASYNC_POSTWRITE);
1363 		bus_dmamap_unload(sc->sc_dmat, txs->txs_dmamap);
1364 		if (txstat & (VGE_TDSTS_EXCESSCOLL|VGE_TDSTS_COLL))
1365 			ifp->if_collisions++;
1366 		if (txstat & VGE_TDSTS_TXERR)
1367 			ifp->if_oerrors++;
1368 		else
1369 			ifp->if_opackets++;
1370 
1371 		sc->sc_tx_free++;
1372 	}
1373 
1374 	/* No changes made to the TX ring, so no flush needed */
1375 
1376 	if (idx != sc->sc_tx_considx) {
1377 		sc->sc_tx_considx = idx;
1378 		ifp->if_flags &= ~IFF_OACTIVE;
1379 	}
1380 
1381 	/*
1382 	 * If not all descriptors have been released reaped yet,
1383 	 * reload the timer so that we will eventually get another
1384 	 * interrupt that will cause us to re-enter this routine.
1385 	 * This is done in case the transmitter has gone idle.
1386 	 */
1387 	if (sc->sc_tx_free != VGE_NTXDESC)
1388 		CSR_WRITE_1(sc, VGE_CRS1, VGE_CR1_TIMER0_ENABLE);
1389 	else
1390 		ifp->if_timer = 0;
1391 }
1392 
1393 static void
1394 vge_tick(void *xsc)
1395 {
1396 	struct vge_softc *sc;
1397 	struct ifnet *ifp;
1398 	struct mii_data *mii;
1399 	int s;
1400 
1401 	sc = xsc;
1402 	ifp = &sc->sc_ethercom.ec_if;
1403 	mii = &sc->sc_mii;
1404 
1405 	s = splnet();
1406 
1407 	callout_schedule(&sc->sc_timeout, hz);
1408 
1409 	mii_tick(mii);
1410 	if (sc->sc_link) {
1411 		if (!(mii->mii_media_status & IFM_ACTIVE))
1412 			sc->sc_link = 0;
1413 	} else {
1414 		if (mii->mii_media_status & IFM_ACTIVE &&
1415 		    IFM_SUBTYPE(mii->mii_media_active) != IFM_NONE) {
1416 			sc->sc_link = 1;
1417 			if (!IFQ_IS_EMPTY(&ifp->if_snd))
1418 				vge_start(ifp);
1419 		}
1420 	}
1421 
1422 	splx(s);
1423 }
1424 
1425 static int
1426 vge_intr(void *arg)
1427 {
1428 	struct vge_softc *sc;
1429 	struct ifnet *ifp;
1430 	uint32_t status;
1431 	int claim;
1432 
1433 	sc = arg;
1434 	claim = 0;
1435 	if (sc->sc_suspended) {
1436 		return claim;
1437 	}
1438 
1439 	ifp = &sc->sc_ethercom.ec_if;
1440 
1441 	if (!(ifp->if_flags & IFF_UP)) {
1442 		return claim;
1443 	}
1444 
1445 	/* Disable interrupts */
1446 	CSR_WRITE_1(sc, VGE_CRC3, VGE_CR3_INT_GMSK);
1447 
1448 	for (;;) {
1449 
1450 		status = CSR_READ_4(sc, VGE_ISR);
1451 		/* If the card has gone away the read returns 0xffff. */
1452 		if (status == 0xFFFFFFFF)
1453 			break;
1454 
1455 		if (status) {
1456 			claim = 1;
1457 			CSR_WRITE_4(sc, VGE_ISR, status);
1458 		}
1459 
1460 		if ((status & VGE_INTRS) == 0)
1461 			break;
1462 
1463 		if (status & (VGE_ISR_RXOK|VGE_ISR_RXOK_HIPRIO))
1464 			vge_rxeof(sc);
1465 
1466 		if (status & (VGE_ISR_RXOFLOW|VGE_ISR_RXNODESC)) {
1467 			vge_rxeof(sc);
1468 			CSR_WRITE_1(sc, VGE_RXQCSRS, VGE_RXQCSR_RUN);
1469 			CSR_WRITE_1(sc, VGE_RXQCSRS, VGE_RXQCSR_WAK);
1470 		}
1471 
1472 		if (status & (VGE_ISR_TXOK0|VGE_ISR_TIMER0))
1473 			vge_txeof(sc);
1474 
1475 		if (status & (VGE_ISR_TXDMA_STALL|VGE_ISR_RXDMA_STALL))
1476 			vge_init(ifp);
1477 
1478 		if (status & VGE_ISR_LINKSTS)
1479 			vge_tick(sc);
1480 	}
1481 
1482 	/* Re-enable interrupts */
1483 	CSR_WRITE_1(sc, VGE_CRS3, VGE_CR3_INT_GMSK);
1484 
1485 	if (!IFQ_IS_EMPTY(&ifp->if_snd))
1486 		vge_start(ifp);
1487 
1488 	return claim;
1489 }
1490 
1491 static int
1492 vge_encap(struct vge_softc *sc, struct mbuf *m_head, int idx)
1493 {
1494 	struct vge_txsoft *txs;
1495 	struct vge_txdesc *txd;
1496 	struct vge_txfrag *f;
1497 	struct mbuf *m_new;
1498 	bus_dmamap_t map;
1499 	int m_csumflags, seg, error, flags;
1500 	struct m_tag *mtag;
1501 	size_t sz;
1502 
1503 	KASSERT(sc->sc_tx_free > 0);
1504 
1505 	txd = &sc->sc_txdescs[idx];
1506 
1507 #ifdef DIAGNOSTIC
1508 	/* If this descriptor is still owned by the chip, bail. */
1509 	VGE_TXDESCSYNC(sc, idx,
1510 	    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1511 	if (le32toh(txd->td_sts) & VGE_TDSTS_OWN) {
1512 		VGE_TXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD);
1513 		return ENOBUFS;
1514 	}
1515 #endif
1516 
1517 	/*
1518 	 * Preserve m_pkthdr.csum_flags here since m_head might be
1519 	 * updated by m_defrag()
1520 	 */
1521 	m_csumflags = m_head->m_pkthdr.csum_flags;
1522 
1523 	txs = &sc->sc_txsoft[idx];
1524 	map = txs->txs_dmamap;
1525 	error = bus_dmamap_load_mbuf(sc->sc_dmat, map, m_head, BUS_DMA_NOWAIT);
1526 
1527 	/* If too many segments to map, coalesce */
1528 	if (error == EFBIG ||
1529 	    (m_head->m_pkthdr.len < ETHER_PAD_LEN &&
1530 	     map->dm_nsegs == VGE_TX_FRAGS)) {
1531 		m_new = m_defrag(m_head, M_DONTWAIT);
1532 		if (m_new == NULL)
1533 			return EFBIG;
1534 
1535 		error = bus_dmamap_load_mbuf(sc->sc_dmat, map,
1536 		    m_new, BUS_DMA_NOWAIT);
1537 		if (error) {
1538 			m_freem(m_new);
1539 			return error;
1540 		}
1541 
1542 		m_head = m_new;
1543 	} else if (error)
1544 		return error;
1545 
1546 	txs->txs_mbuf = m_head;
1547 
1548 	bus_dmamap_sync(sc->sc_dmat, map, 0, map->dm_mapsize,
1549 	    BUS_DMASYNC_PREWRITE);
1550 
1551 	for (seg = 0, f = &txd->td_frag[0]; seg < map->dm_nsegs; seg++, f++) {
1552 		f->tf_buflen = htole16(VGE_BUFLEN(map->dm_segs[seg].ds_len));
1553 		f->tf_addrlo = htole32(VGE_ADDR_LO(map->dm_segs[seg].ds_addr));
1554 		f->tf_addrhi = htole16(VGE_ADDR_HI(map->dm_segs[seg].ds_addr));
1555 	}
1556 
1557 	/* Argh. This chip does not autopad short frames */
1558 	sz = m_head->m_pkthdr.len;
1559 	if (sz < ETHER_PAD_LEN) {
1560 		f->tf_buflen = htole16(VGE_BUFLEN(ETHER_PAD_LEN - sz));
1561 		f->tf_addrlo = htole32(VGE_ADDR_LO(VGE_CDPADADDR(sc)));
1562 		f->tf_addrhi = htole16(VGE_ADDR_HI(VGE_CDPADADDR(sc)) & 0xFFFF);
1563 		sz = ETHER_PAD_LEN;
1564 		seg++;
1565 	}
1566 	VGE_TXFRAGSYNC(sc, idx, seg, BUS_DMASYNC_PREWRITE);
1567 
1568 	/*
1569 	 * When telling the chip how many segments there are, we
1570 	 * must use nsegs + 1 instead of just nsegs. Darned if I
1571 	 * know why.
1572 	 */
1573 	seg++;
1574 
1575 	flags = 0;
1576 	if (m_csumflags & M_CSUM_IPv4)
1577 		flags |= VGE_TDCTL_IPCSUM;
1578 	if (m_csumflags & M_CSUM_TCPv4)
1579 		flags |= VGE_TDCTL_TCPCSUM;
1580 	if (m_csumflags & M_CSUM_UDPv4)
1581 		flags |= VGE_TDCTL_UDPCSUM;
1582 	txd->td_sts = htole32(sz << 16);
1583 	txd->td_ctl = htole32(flags | (seg << 28) | VGE_TD_LS_NORM);
1584 
1585 	if (sz > ETHERMTU + ETHER_HDR_LEN)
1586 		txd->td_ctl |= htole32(VGE_TDCTL_JUMBO);
1587 
1588 	/*
1589 	 * Set up hardware VLAN tagging.
1590 	 */
1591 
1592 	mtag = VLAN_OUTPUT_TAG(&sc->sc_ethercom, m_head);
1593 	if (mtag != NULL) {
1594 		/*
1595 		 * No need htons() here since vge(4) chip assumes
1596 		 * that tags are written in little endian and
1597 		 * we already use htole32() here.
1598 		 */
1599 		txd->td_ctl |= htole32(VLAN_TAG_VALUE(mtag) | VGE_TDCTL_VTAG);
1600 	}
1601 
1602 	txd->td_sts |= htole32(VGE_TDSTS_OWN);
1603 	VGE_TXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1604 
1605 	sc->sc_tx_free--;
1606 
1607 	return 0;
1608 }
1609 
1610 /*
1611  * Main transmit routine.
1612  */
1613 
1614 static void
1615 vge_start(struct ifnet *ifp)
1616 {
1617 	struct vge_softc *sc;
1618 	struct vge_txsoft *txs;
1619 	struct mbuf *m_head;
1620 	int idx, pidx, error;
1621 
1622 	sc = ifp->if_softc;
1623 
1624 	if (!sc->sc_link ||
1625 	    (ifp->if_flags & (IFF_RUNNING|IFF_OACTIVE)) != IFF_RUNNING) {
1626 		return;
1627 	}
1628 
1629 	m_head = NULL;
1630 	idx = sc->sc_tx_prodidx;
1631 
1632 	pidx = (idx - 1) & VGE_NTXDESC_MASK;
1633 
1634 	/*
1635 	 * Loop through the send queue, setting up transmit descriptors
1636 	 * until we drain the queue, or use up all available transmit
1637 	 * descriptors.
1638 	 */
1639 	for (;;) {
1640 		/* Grab a packet off the queue. */
1641 		IFQ_POLL(&ifp->if_snd, m_head);
1642 		if (m_head == NULL)
1643 			break;
1644 
1645 		txs = &sc->sc_txsoft[idx];
1646 
1647 		if (txs->txs_mbuf != NULL) {
1648 			/*
1649 			 * Slot already used, stop for now.
1650 			 */
1651 			ifp->if_flags |= IFF_OACTIVE;
1652 			break;
1653 		}
1654 
1655 		if ((error = vge_encap(sc, m_head, idx))) {
1656 			if (error == EFBIG) {
1657 				aprint_error("%s: Tx packet consumes too many "
1658 				    "DMA segments, dropping...\n",
1659 				    sc->sc_dev.dv_xname);
1660 				IFQ_DEQUEUE(&ifp->if_snd, m_head);
1661 				m_freem(m_head);
1662 				continue;
1663 			}
1664 
1665 			/*
1666 			 * Short on resources, just stop for now.
1667 			 */
1668 			if (error == ENOBUFS)
1669 				ifp->if_flags |= IFF_OACTIVE;
1670 			break;
1671 		}
1672 
1673 		IFQ_DEQUEUE(&ifp->if_snd, m_head);
1674 
1675 		/*
1676 		 * WE ARE NOW COMMITTED TO TRANSMITTING THE PACKET.
1677 		 */
1678 
1679 		sc->sc_txdescs[pidx].td_frag[0].tf_buflen |=
1680 		    htole16(VGE_TXDESC_Q);
1681 		VGE_TXFRAGSYNC(sc, pidx, 1,
1682 		    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1683 
1684 		if (txs->txs_mbuf != m_head) {
1685 			m_freem(m_head);
1686 			m_head = txs->txs_mbuf;
1687 		}
1688 
1689 		pidx = idx;
1690 		idx = VGE_NEXT_TXDESC(idx);
1691 
1692 		/*
1693 		 * If there's a BPF listener, bounce a copy of this frame
1694 		 * to him.
1695 		 */
1696 #if NBPFILTER > 0
1697 		if (ifp->if_bpf)
1698 			bpf_mtap(ifp->if_bpf, m_head);
1699 #endif
1700 	}
1701 
1702 	if (idx == sc->sc_tx_prodidx) {
1703 		return;
1704 	}
1705 
1706 	/* Issue a transmit command. */
1707 	CSR_WRITE_2(sc, VGE_TXQCSRS, VGE_TXQCSR_WAK0);
1708 
1709 	sc->sc_tx_prodidx = idx;
1710 
1711 	/*
1712 	 * Use the countdown timer for interrupt moderation.
1713 	 * 'TX done' interrupts are disabled. Instead, we reset the
1714 	 * countdown timer, which will begin counting until it hits
1715 	 * the value in the SSTIMER register, and then trigger an
1716 	 * interrupt. Each time we set the TIMER0_ENABLE bit, the
1717 	 * the timer count is reloaded. Only when the transmitter
1718 	 * is idle will the timer hit 0 and an interrupt fire.
1719 	 */
1720 	CSR_WRITE_1(sc, VGE_CRS1, VGE_CR1_TIMER0_ENABLE);
1721 
1722 	/*
1723 	 * Set a timeout in case the chip goes out to lunch.
1724 	 */
1725 	ifp->if_timer = 5;
1726 }
1727 
1728 static int
1729 vge_init(struct ifnet *ifp)
1730 {
1731 	struct vge_softc *sc;
1732 	struct vge_rxsoft *rxs;
1733 	int i;
1734 
1735 	sc = ifp->if_softc;
1736 
1737 	/*
1738 	 * Cancel pending I/O and free all RX/TX buffers.
1739 	 */
1740 	vge_stop(sc);
1741 	vge_reset(sc);
1742 
1743 	/* Initialize the RX descriptors and mbufs. */
1744 	memset(sc->sc_rxdescs, 0, sizeof(sc->sc_rxdescs));
1745 	for (i = 0; i < VGE_NRXDESC; i++) {
1746 		rxs = &sc->sc_rxsoft[i];
1747 		if (rxs->rxs_mbuf) {
1748 			m_freem(rxs->rxs_mbuf);
1749 			rxs->rxs_mbuf = NULL;
1750 		}
1751 		if (rxs->rxs_dmamap)
1752 			bus_dmamap_unload(sc->sc_dmat, rxs->rxs_dmamap);
1753 		if (vge_newbuf(sc, i, NULL) == ENOBUFS) {
1754 			aprint_error("%s: unable to allocate or map "
1755 			    "rx buffer\n", sc->sc_dev.dv_xname);
1756 			return 1; /* XXX */
1757 		}
1758 	}
1759 	sc->sc_rx_prodidx = 0;
1760 	sc->sc_rx_consumed = 0;
1761 	sc->sc_rx_mhead = sc->sc_rx_mtail = NULL;
1762 
1763 	/* Initialize the  TX descriptors and mbufs. */
1764 	memset(sc->sc_txdescs, 0, sizeof(sc->sc_txdescs));
1765 	bus_dmamap_sync(sc->sc_dmat, sc->sc_cddmamap,
1766 	    VGE_CDTXOFF(0), sizeof(sc->sc_txdescs),
1767 	    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1768 	for (i = 0; i < VGE_NTXDESC; i++)
1769 		sc->sc_txsoft[i].txs_mbuf = NULL;
1770 
1771 	sc->sc_tx_prodidx = 0;
1772 	sc->sc_tx_considx = 0;
1773 	sc->sc_tx_free = VGE_NTXDESC;
1774 
1775 	/* Set our station address */
1776 	for (i = 0; i < ETHER_ADDR_LEN; i++)
1777 		CSR_WRITE_1(sc, VGE_PAR0 + i, sc->sc_eaddr[i]);
1778 
1779 	/*
1780 	 * Set receive FIFO threshold. Also allow transmission and
1781 	 * reception of VLAN tagged frames.
1782 	 */
1783 	CSR_CLRBIT_1(sc, VGE_RXCFG, VGE_RXCFG_FIFO_THR|VGE_RXCFG_VTAGOPT);
1784 	CSR_SETBIT_1(sc, VGE_RXCFG, VGE_RXFIFOTHR_128BYTES|VGE_VTAG_OPT2);
1785 
1786 	/* Set DMA burst length */
1787 	CSR_CLRBIT_1(sc, VGE_DMACFG0, VGE_DMACFG0_BURSTLEN);
1788 	CSR_SETBIT_1(sc, VGE_DMACFG0, VGE_DMABURST_128);
1789 
1790 	CSR_SETBIT_1(sc, VGE_TXCFG, VGE_TXCFG_ARB_PRIO|VGE_TXCFG_NONBLK);
1791 
1792 	/* Set collision backoff algorithm */
1793 	CSR_CLRBIT_1(sc, VGE_CHIPCFG1, VGE_CHIPCFG1_CRANDOM|
1794 	    VGE_CHIPCFG1_CAP|VGE_CHIPCFG1_MBA|VGE_CHIPCFG1_BAKOPT);
1795 	CSR_SETBIT_1(sc, VGE_CHIPCFG1, VGE_CHIPCFG1_OFSET);
1796 
1797 	/* Disable LPSEL field in priority resolution */
1798 	CSR_SETBIT_1(sc, VGE_DIAGCTL, VGE_DIAGCTL_LPSEL_DIS);
1799 
1800 	/*
1801 	 * Load the addresses of the DMA queues into the chip.
1802 	 * Note that we only use one transmit queue.
1803 	 */
1804 
1805 	CSR_WRITE_4(sc, VGE_TXDESC_ADDR_LO0, VGE_ADDR_LO(VGE_CDTXADDR(sc, 0)));
1806 	CSR_WRITE_2(sc, VGE_TXDESCNUM, VGE_NTXDESC - 1);
1807 
1808 	CSR_WRITE_4(sc, VGE_RXDESC_ADDR_LO, VGE_ADDR_LO(VGE_CDRXADDR(sc, 0)));
1809 	CSR_WRITE_2(sc, VGE_RXDESCNUM, VGE_NRXDESC - 1);
1810 	CSR_WRITE_2(sc, VGE_RXDESC_RESIDUECNT, VGE_NRXDESC);
1811 
1812 	/* Enable and wake up the RX descriptor queue */
1813 	CSR_WRITE_1(sc, VGE_RXQCSRS, VGE_RXQCSR_RUN);
1814 	CSR_WRITE_1(sc, VGE_RXQCSRS, VGE_RXQCSR_WAK);
1815 
1816 	/* Enable the TX descriptor queue */
1817 	CSR_WRITE_2(sc, VGE_TXQCSRS, VGE_TXQCSR_RUN0);
1818 
1819 	/* Set up the receive filter -- allow large frames for VLANs. */
1820 	CSR_WRITE_1(sc, VGE_RXCTL, VGE_RXCTL_RX_UCAST|VGE_RXCTL_RX_GIANT);
1821 
1822 	/* If we want promiscuous mode, set the allframes bit. */
1823 	if (ifp->if_flags & IFF_PROMISC) {
1824 		CSR_SETBIT_1(sc, VGE_RXCTL, VGE_RXCTL_RX_PROMISC);
1825 	}
1826 
1827 	/* Set capture broadcast bit to capture broadcast frames. */
1828 	if (ifp->if_flags & IFF_BROADCAST) {
1829 		CSR_SETBIT_1(sc, VGE_RXCTL, VGE_RXCTL_RX_BCAST);
1830 	}
1831 
1832 	/* Set multicast bit to capture multicast frames. */
1833 	if (ifp->if_flags & IFF_MULTICAST) {
1834 		CSR_SETBIT_1(sc, VGE_RXCTL, VGE_RXCTL_RX_MCAST);
1835 	}
1836 
1837 	/* Init the cam filter. */
1838 	vge_cam_clear(sc);
1839 
1840 	/* Init the multicast filter. */
1841 	vge_setmulti(sc);
1842 
1843 	/* Enable flow control */
1844 
1845 	CSR_WRITE_1(sc, VGE_CRS2, 0x8B);
1846 
1847 	/* Enable jumbo frame reception (if desired) */
1848 
1849 	/* Start the MAC. */
1850 	CSR_WRITE_1(sc, VGE_CRC0, VGE_CR0_STOP);
1851 	CSR_WRITE_1(sc, VGE_CRS1, VGE_CR1_NOPOLL);
1852 	CSR_WRITE_1(sc, VGE_CRS0,
1853 	    VGE_CR0_TX_ENABLE|VGE_CR0_RX_ENABLE|VGE_CR0_START);
1854 
1855 	/*
1856 	 * Configure one-shot timer for microsecond
1857 	 * resulution and load it for 500 usecs.
1858 	 */
1859 	CSR_SETBIT_1(sc, VGE_DIAGCTL, VGE_DIAGCTL_TIMER0_RES);
1860 	CSR_WRITE_2(sc, VGE_SSTIMER, 400);
1861 
1862 	/*
1863 	 * Configure interrupt moderation for receive. Enable
1864 	 * the holdoff counter and load it, and set the RX
1865 	 * suppression count to the number of descriptors we
1866 	 * want to allow before triggering an interrupt.
1867 	 * The holdoff timer is in units of 20 usecs.
1868 	 */
1869 
1870 #ifdef notyet
1871 	CSR_WRITE_1(sc, VGE_INTCTL1, VGE_INTCTL_TXINTSUP_DISABLE);
1872 	/* Select the interrupt holdoff timer page. */
1873 	CSR_CLRBIT_1(sc, VGE_CAMCTL, VGE_CAMCTL_PAGESEL);
1874 	CSR_SETBIT_1(sc, VGE_CAMCTL, VGE_PAGESEL_INTHLDOFF);
1875 	CSR_WRITE_1(sc, VGE_INTHOLDOFF, 10); /* ~200 usecs */
1876 
1877 	/* Enable use of the holdoff timer. */
1878 	CSR_WRITE_1(sc, VGE_CRS3, VGE_CR3_INT_HOLDOFF);
1879 	CSR_WRITE_1(sc, VGE_INTCTL1, VGE_INTCTL_SC_RELOAD);
1880 
1881 	/* Select the RX suppression threshold page. */
1882 	CSR_CLRBIT_1(sc, VGE_CAMCTL, VGE_CAMCTL_PAGESEL);
1883 	CSR_SETBIT_1(sc, VGE_CAMCTL, VGE_PAGESEL_RXSUPPTHR);
1884 	CSR_WRITE_1(sc, VGE_RXSUPPTHR, 64); /* interrupt after 64 packets */
1885 
1886 	/* Restore the page select bits. */
1887 	CSR_CLRBIT_1(sc, VGE_CAMCTL, VGE_CAMCTL_PAGESEL);
1888 	CSR_SETBIT_1(sc, VGE_CAMCTL, VGE_PAGESEL_MAR);
1889 #endif
1890 
1891 #ifdef DEVICE_POLLING
1892 	/*
1893 	 * Disable interrupts if we are polling.
1894 	 */
1895 	if (ifp->if_flags & IFF_POLLING) {
1896 		CSR_WRITE_4(sc, VGE_IMR, 0);
1897 		CSR_WRITE_1(sc, VGE_CRC3, VGE_CR3_INT_GMSK);
1898 	} else	/* otherwise ... */
1899 #endif /* DEVICE_POLLING */
1900 	{
1901 	/*
1902 	 * Enable interrupts.
1903 	 */
1904 		CSR_WRITE_4(sc, VGE_IMR, VGE_INTRS);
1905 		CSR_WRITE_4(sc, VGE_ISR, 0);
1906 		CSR_WRITE_1(sc, VGE_CRS3, VGE_CR3_INT_GMSK);
1907 	}
1908 
1909 	mii_mediachg(&sc->sc_mii);
1910 
1911 	ifp->if_flags |= IFF_RUNNING;
1912 	ifp->if_flags &= ~IFF_OACTIVE;
1913 
1914 	sc->sc_if_flags = 0;
1915 	sc->sc_link = 0;
1916 
1917 	callout_schedule(&sc->sc_timeout, hz);
1918 
1919 	return 0;
1920 }
1921 
1922 /*
1923  * Set media options.
1924  */
1925 static int
1926 vge_ifmedia_upd(struct ifnet *ifp)
1927 {
1928 	struct vge_softc *sc;
1929 
1930 	sc = ifp->if_softc;
1931 	mii_mediachg(&sc->sc_mii);
1932 
1933 	return 0;
1934 }
1935 
1936 /*
1937  * Report current media status.
1938  */
1939 static void
1940 vge_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
1941 {
1942 	struct vge_softc *sc;
1943 	struct mii_data *mii;
1944 
1945 	sc = ifp->if_softc;
1946 	mii = &sc->sc_mii;
1947 
1948 	mii_pollstat(mii);
1949 	ifmr->ifm_active = mii->mii_media_active;
1950 	ifmr->ifm_status = mii->mii_media_status;
1951 }
1952 
1953 static void
1954 vge_miibus_statchg(struct device *self)
1955 {
1956 	struct vge_softc *sc;
1957 	struct mii_data *mii;
1958 	struct ifmedia_entry *ife;
1959 
1960 	sc = (void *)self;
1961 	mii = &sc->sc_mii;
1962 	ife = mii->mii_media.ifm_cur;
1963 	/*
1964 	 * If the user manually selects a media mode, we need to turn
1965 	 * on the forced MAC mode bit in the DIAGCTL register. If the
1966 	 * user happens to choose a full duplex mode, we also need to
1967 	 * set the 'force full duplex' bit. This applies only to
1968 	 * 10Mbps and 100Mbps speeds. In autoselect mode, forced MAC
1969 	 * mode is disabled, and in 1000baseT mode, full duplex is
1970 	 * always implied, so we turn on the forced mode bit but leave
1971 	 * the FDX bit cleared.
1972 	 */
1973 
1974 	switch (IFM_SUBTYPE(ife->ifm_media)) {
1975 	case IFM_AUTO:
1976 		CSR_CLRBIT_1(sc, VGE_DIAGCTL, VGE_DIAGCTL_MACFORCE);
1977 		CSR_CLRBIT_1(sc, VGE_DIAGCTL, VGE_DIAGCTL_FDXFORCE);
1978 		break;
1979 	case IFM_1000_T:
1980 		CSR_SETBIT_1(sc, VGE_DIAGCTL, VGE_DIAGCTL_MACFORCE);
1981 		CSR_CLRBIT_1(sc, VGE_DIAGCTL, VGE_DIAGCTL_FDXFORCE);
1982 		break;
1983 	case IFM_100_TX:
1984 	case IFM_10_T:
1985 		CSR_SETBIT_1(sc, VGE_DIAGCTL, VGE_DIAGCTL_MACFORCE);
1986 		if ((ife->ifm_media & IFM_GMASK) == IFM_FDX) {
1987 			CSR_SETBIT_1(sc, VGE_DIAGCTL, VGE_DIAGCTL_FDXFORCE);
1988 		} else {
1989 			CSR_CLRBIT_1(sc, VGE_DIAGCTL, VGE_DIAGCTL_FDXFORCE);
1990 		}
1991 		break;
1992 	default:
1993 		aprint_error("%s: unknown media type: %x\n",
1994 		    sc->sc_dev.dv_xname,
1995 		    IFM_SUBTYPE(ife->ifm_media));
1996 		break;
1997 	}
1998 }
1999 
2000 static int
2001 vge_ioctl(struct ifnet *ifp, u_long command, caddr_t data)
2002 {
2003 	struct vge_softc *sc;
2004 	struct ifreq *ifr;
2005 	struct mii_data *mii;
2006 	int s, error;
2007 
2008 	sc = ifp->if_softc;
2009 	ifr = (struct ifreq *)data;
2010 	error = 0;
2011 
2012 	s = splnet();
2013 
2014 	switch (command) {
2015 	case SIOCSIFMTU:
2016 		if (ifr->ifr_mtu > VGE_JUMBO_MTU)
2017 			error = EINVAL;
2018 		ifp->if_mtu = ifr->ifr_mtu;
2019 		break;
2020 	case SIOCSIFFLAGS:
2021 		if (ifp->if_flags & IFF_UP) {
2022 			if (ifp->if_flags & IFF_RUNNING &&
2023 			    ifp->if_flags & IFF_PROMISC &&
2024 			    !(sc->sc_if_flags & IFF_PROMISC)) {
2025 				CSR_SETBIT_1(sc, VGE_RXCTL,
2026 				    VGE_RXCTL_RX_PROMISC);
2027 				vge_setmulti(sc);
2028 			} else if (ifp->if_flags & IFF_RUNNING &&
2029 			    !(ifp->if_flags & IFF_PROMISC) &&
2030 			    sc->sc_if_flags & IFF_PROMISC) {
2031 				CSR_CLRBIT_1(sc, VGE_RXCTL,
2032 				    VGE_RXCTL_RX_PROMISC);
2033 				vge_setmulti(sc);
2034                         } else
2035 				vge_init(ifp);
2036 		} else {
2037 			if (ifp->if_flags & IFF_RUNNING)
2038 				vge_stop(sc);
2039 		}
2040 		sc->sc_if_flags = ifp->if_flags;
2041 		break;
2042 	case SIOCADDMULTI:
2043 	case SIOCDELMULTI:
2044 		error = (command == SIOCADDMULTI) ?
2045 		    ether_addmulti(ifr, &sc->sc_ethercom) :
2046 		    ether_delmulti(ifr, &sc->sc_ethercom);
2047 
2048 		if (error == ENETRESET) {
2049 			/*
2050 			 * Multicast list has changed; set the hardware filter
2051 			 * accordingly.
2052 			 */
2053 			if (ifp->if_flags & IFF_RUNNING)
2054 				vge_setmulti(sc);
2055 			error = 0;
2056 		}
2057 		break;
2058 	case SIOCGIFMEDIA:
2059 	case SIOCSIFMEDIA:
2060 		mii = &sc->sc_mii;
2061 		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, command);
2062 		break;
2063 	default:
2064 		error = ether_ioctl(ifp, command, data);
2065 		break;
2066 	}
2067 
2068 	splx(s);
2069 	return error;
2070 }
2071 
2072 static void
2073 vge_watchdog(struct ifnet *ifp)
2074 {
2075 	struct vge_softc *sc;
2076 	int s;
2077 
2078 	sc = ifp->if_softc;
2079 	s = splnet();
2080 	aprint_error("%s: watchdog timeout\n", sc->sc_dev.dv_xname);
2081 	ifp->if_oerrors++;
2082 
2083 	vge_txeof(sc);
2084 	vge_rxeof(sc);
2085 
2086 	vge_init(ifp);
2087 
2088 	splx(s);
2089 }
2090 
2091 /*
2092  * Stop the adapter and free any mbufs allocated to the
2093  * RX and TX lists.
2094  */
2095 static void
2096 vge_stop(struct vge_softc *sc)
2097 {
2098 	struct ifnet *ifp;
2099 	struct vge_txsoft *txs;
2100 	struct vge_rxsoft *rxs;
2101 	int i, s;
2102 
2103 	ifp = &sc->sc_ethercom.ec_if;
2104 
2105 	s = splnet();
2106 	ifp->if_timer = 0;
2107 
2108 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
2109 #ifdef DEVICE_POLLING
2110 	ether_poll_deregister(ifp);
2111 #endif /* DEVICE_POLLING */
2112 
2113 	CSR_WRITE_1(sc, VGE_CRC3, VGE_CR3_INT_GMSK);
2114 	CSR_WRITE_1(sc, VGE_CRS0, VGE_CR0_STOP);
2115 	CSR_WRITE_4(sc, VGE_ISR, 0xFFFFFFFF);
2116 	CSR_WRITE_2(sc, VGE_TXQCSRC, 0xFFFF);
2117 	CSR_WRITE_1(sc, VGE_RXQCSRC, 0xFF);
2118 	CSR_WRITE_4(sc, VGE_RXDESC_ADDR_LO, 0);
2119 
2120 	if (sc->sc_rx_mhead != NULL) {
2121 		m_freem(sc->sc_rx_mhead);
2122 		sc->sc_rx_mhead = sc->sc_rx_mtail = NULL;
2123 	}
2124 
2125 	/* Free the TX list buffers. */
2126 
2127 	for (i = 0; i < VGE_NTXDESC; i++) {
2128 		txs = &sc->sc_txsoft[i];
2129 		if (txs->txs_mbuf != NULL) {
2130 			bus_dmamap_unload(sc->sc_dmat, txs->txs_dmamap);
2131 			m_freem(txs->txs_mbuf);
2132 			txs->txs_mbuf = NULL;
2133 		}
2134 	}
2135 
2136 	/* Free the RX list buffers. */
2137 
2138 	for (i = 0; i < VGE_NRXDESC; i++) {
2139 		rxs = &sc->sc_rxsoft[i];
2140 		if (rxs->rxs_mbuf != NULL) {
2141 			bus_dmamap_unload(sc->sc_dmat, rxs->rxs_dmamap);
2142 			m_freem(rxs->rxs_mbuf);
2143 			rxs->rxs_mbuf = NULL;
2144 		}
2145 	}
2146 
2147 	splx(s);
2148 }
2149 
2150 #if VGE_POWER_MANAGEMENT
2151 /*
2152  * Device suspend routine.  Stop the interface and save some PCI
2153  * settings in case the BIOS doesn't restore them properly on
2154  * resume.
2155  */
2156 static int
2157 vge_suspend(struct device *dev)
2158 {
2159 	struct vge_softc *sc;
2160 	int i;
2161 
2162 	sc = device_get_softc(dev);
2163 
2164 	vge_stop(sc);
2165 
2166         for (i = 0; i < 5; i++)
2167 		sc->sc_saved_maps[i] =
2168 		    pci_read_config(dev, PCIR_MAPS + i * 4, 4);
2169 	sc->sc_saved_biosaddr = pci_read_config(dev, PCIR_BIOS, 4);
2170 	sc->sc_saved_intline = pci_read_config(dev, PCIR_INTLINE, 1);
2171 	sc->sc_saved_cachelnsz = pci_read_config(dev, PCIR_CACHELNSZ, 1);
2172 	sc->sc_saved_lattimer = pci_read_config(dev, PCIR_LATTIMER, 1);
2173 
2174 	sc->suspended = 1;
2175 
2176 	return 0;
2177 }
2178 
2179 /*
2180  * Device resume routine.  Restore some PCI settings in case the BIOS
2181  * doesn't, re-enable busmastering, and restart the interface if
2182  * appropriate.
2183  */
2184 static int
2185 vge_resume(struct device *dev)
2186 {
2187 	struct vge_softc *sc;
2188 	struct ifnet *ifp;
2189 	int i;
2190 
2191 	sc = (void *)dev;
2192 	ifp = &sc->sc_ethercom.ec_if;
2193 
2194         /* better way to do this? */
2195 	for (i = 0; i < 5; i++)
2196 		pci_write_config(dev, PCIR_MAPS + i * 4,
2197 		    sc->sc_saved_maps[i], 4);
2198 	pci_write_config(dev, PCIR_BIOS, sc->sc_saved_biosaddr, 4);
2199 	pci_write_config(dev, PCIR_INTLINE, sc->sc_saved_intline, 1);
2200 	pci_write_config(dev, PCIR_CACHELNSZ, sc->sc_saved_cachelnsz, 1);
2201 	pci_write_config(dev, PCIR_LATTIMER, sc->sc_saved_lattimer, 1);
2202 
2203 	/* reenable busmastering */
2204 	pci_enable_busmaster(dev);
2205 	pci_enable_io(dev, SYS_RES_MEMORY);
2206 
2207 	/* reinitialize interface if necessary */
2208 	if (ifp->if_flags & IFF_UP)
2209 		vge_init(sc);
2210 
2211 	sc->suspended = 0;
2212 
2213 	return 0;
2214 }
2215 #endif
2216 
2217 /*
2218  * Stop all chip I/O so that the kernel's probe routines don't
2219  * get confused by errant DMAs when rebooting.
2220  */
2221 static void
2222 vge_shutdown(void *arg)
2223 {
2224 	struct vge_softc *sc;
2225 
2226 	sc = arg;
2227 	vge_stop(sc);
2228 }
2229