xref: /netbsd-src/sys/dev/pci/if_vge.c (revision ce2c90c7c172d95d2402a5b3d96d8f8e6d138a21)
1 /* $NetBSD: if_vge.c,v 1.18 2006/10/15 11:52:36 tsutsui 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 1.18 2006/10/15 11:52:36 tsutsui 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 #include <dev/pci/if_vgevar.h>
117 
118 static int vge_probe(struct device *, struct cfdata *, void *);
119 static void vge_attach(struct device *, struct device *, void *);
120 
121 static int vge_encap(struct vge_softc *, struct mbuf *, int);
122 
123 static int vge_allocmem(struct vge_softc *);
124 static int vge_newbuf(struct vge_softc *, int, struct mbuf *);
125 static int vge_rx_list_init(struct vge_softc *);
126 static int vge_tx_list_init(struct vge_softc *);
127 #ifndef __NO_STRICT_ALIGNMENT
128 static inline void vge_fixup_rx(struct mbuf *);
129 #endif
130 static void vge_rxeof(struct vge_softc *);
131 static void vge_txeof(struct vge_softc *);
132 static int vge_intr(void *);
133 static void vge_tick(void *);
134 static void vge_start(struct ifnet *);
135 static int vge_ioctl(struct ifnet *, u_long, caddr_t);
136 static int vge_init(struct ifnet *);
137 static void vge_stop(struct vge_softc *);
138 static void vge_watchdog(struct ifnet *);
139 #if VGE_POWER_MANAGEMENT
140 static int vge_suspend(struct device *);
141 static int vge_resume(struct device *);
142 #endif
143 static void vge_shutdown(void *);
144 static int vge_ifmedia_upd(struct ifnet *);
145 static void vge_ifmedia_sts(struct ifnet *, struct ifmediareq *);
146 
147 static uint16_t vge_read_eeprom(struct vge_softc *, int);
148 
149 static void vge_miipoll_start(struct vge_softc *);
150 static void vge_miipoll_stop(struct vge_softc *);
151 static int vge_miibus_readreg(struct device *, int, int);
152 static void vge_miibus_writereg(struct device *, int, int, int);
153 static void vge_miibus_statchg(struct device *);
154 
155 static void vge_cam_clear(struct vge_softc *);
156 static int vge_cam_set(struct vge_softc *, uint8_t *);
157 static void vge_setmulti(struct vge_softc *);
158 static void vge_reset(struct vge_softc *);
159 
160 #define VGE_PCI_LOIO             0x10
161 #define VGE_PCI_LOMEM            0x14
162 
163 CFATTACH_DECL(vge, sizeof(struct vge_softc),
164     vge_probe, vge_attach, NULL, NULL);
165 
166 /*
167  * Defragment mbuf chain contents to be as linear as possible.
168  * Returns new mbuf chain on success, NULL on failure. Old mbuf
169  * chain is always freed.
170  * XXX temporary until there would be generic function doing this.
171  */
172 #define m_defrag	vge_m_defrag
173 struct mbuf * vge_m_defrag(struct mbuf *, int);
174 
175 struct mbuf *
176 vge_m_defrag(struct mbuf *mold, int flags)
177 {
178 	struct mbuf *m0, *mn, *n;
179 	size_t sz = mold->m_pkthdr.len;
180 
181 #ifdef DIAGNOSTIC
182 	if ((mold->m_flags & M_PKTHDR) == 0)
183 		panic("m_defrag: not a mbuf chain header");
184 #endif
185 
186 	MGETHDR(m0, flags, MT_DATA);
187 	if (m0 == NULL)
188 		return NULL;
189 	m0->m_pkthdr.len = mold->m_pkthdr.len;
190 	mn = m0;
191 
192 	do {
193 		if (sz > MHLEN) {
194 			MCLGET(mn, M_DONTWAIT);
195 			if ((mn->m_flags & M_EXT) == 0) {
196 				m_freem(m0);
197 				return NULL;
198 			}
199 		}
200 
201 		mn->m_len = MIN(sz, MCLBYTES);
202 
203 		m_copydata(mold, mold->m_pkthdr.len - sz, mn->m_len,
204 		     mtod(mn, caddr_t));
205 
206 		sz -= mn->m_len;
207 
208 		if (sz > 0) {
209 			/* need more mbufs */
210 			MGET(n, M_NOWAIT, MT_DATA);
211 			if (n == NULL) {
212 				m_freem(m0);
213 				return NULL;
214 			}
215 
216 			mn->m_next = n;
217 			mn = n;
218 		}
219 	} while (sz > 0);
220 
221 	return m0;
222 }
223 
224 /*
225  * Read a word of data stored in the EEPROM at address 'addr.'
226  */
227 static uint16_t
228 vge_read_eeprom(struct vge_softc *sc, int addr)
229 {
230 	int i;
231 	uint16_t word = 0;
232 
233 	/*
234 	 * Enter EEPROM embedded programming mode. In order to
235 	 * access the EEPROM at all, we first have to set the
236 	 * EELOAD bit in the CHIPCFG2 register.
237 	 */
238 	CSR_SETBIT_1(sc, VGE_CHIPCFG2, VGE_CHIPCFG2_EELOAD);
239 	CSR_SETBIT_1(sc, VGE_EECSR, VGE_EECSR_EMBP/*|VGE_EECSR_ECS*/);
240 
241 	/* Select the address of the word we want to read */
242 	CSR_WRITE_1(sc, VGE_EEADDR, addr);
243 
244 	/* Issue read command */
245 	CSR_SETBIT_1(sc, VGE_EECMD, VGE_EECMD_ERD);
246 
247 	/* Wait for the done bit to be set. */
248 	for (i = 0; i < VGE_TIMEOUT; i++) {
249 		if (CSR_READ_1(sc, VGE_EECMD) & VGE_EECMD_EDONE)
250 			break;
251 	}
252 
253 	if (i == VGE_TIMEOUT) {
254 		printf("%s: EEPROM read timed out\n", sc->sc_dev.dv_xname);
255 		return 0;
256 	}
257 
258 	/* Read the result */
259 	word = CSR_READ_2(sc, VGE_EERDDAT);
260 
261 	/* Turn off EEPROM access mode. */
262 	CSR_CLRBIT_1(sc, VGE_EECSR, VGE_EECSR_EMBP/*|VGE_EECSR_ECS*/);
263 	CSR_CLRBIT_1(sc, VGE_CHIPCFG2, VGE_CHIPCFG2_EELOAD);
264 
265 	return word;
266 }
267 
268 static void
269 vge_miipoll_stop(struct vge_softc *sc)
270 {
271 	int i;
272 
273 	CSR_WRITE_1(sc, VGE_MIICMD, 0);
274 
275 	for (i = 0; i < VGE_TIMEOUT; i++) {
276 		DELAY(1);
277 		if (CSR_READ_1(sc, VGE_MIISTS) & VGE_MIISTS_IIDL)
278 			break;
279 	}
280 
281 	if (i == VGE_TIMEOUT) {
282 		printf("%s: failed to idle MII autopoll\n",
283 		    sc->sc_dev.dv_xname);
284 	}
285 }
286 
287 static void
288 vge_miipoll_start(struct vge_softc *sc)
289 {
290 	int i;
291 
292 	/* First, make sure we're idle. */
293 
294 	CSR_WRITE_1(sc, VGE_MIICMD, 0);
295 	CSR_WRITE_1(sc, VGE_MIIADDR, VGE_MIIADDR_SWMPL);
296 
297 	for (i = 0; i < VGE_TIMEOUT; i++) {
298 		DELAY(1);
299 		if (CSR_READ_1(sc, VGE_MIISTS) & VGE_MIISTS_IIDL)
300 			break;
301 	}
302 
303 	if (i == VGE_TIMEOUT) {
304 		printf("%s: failed to idle MII autopoll\n",
305 		    sc->sc_dev.dv_xname);
306 		return;
307 	}
308 
309 	/* Now enable auto poll mode. */
310 
311 	CSR_WRITE_1(sc, VGE_MIICMD, VGE_MIICMD_MAUTO);
312 
313 	/* And make sure it started. */
314 
315 	for (i = 0; i < VGE_TIMEOUT; i++) {
316 		DELAY(1);
317 		if ((CSR_READ_1(sc, VGE_MIISTS) & VGE_MIISTS_IIDL) == 0)
318 			break;
319 	}
320 
321 	if (i == VGE_TIMEOUT) {
322 		printf("%s: failed to start MII autopoll\n",
323 		    sc->sc_dev.dv_xname);
324 	}
325 }
326 
327 static int
328 vge_miibus_readreg(struct device *dev, int phy, int reg)
329 {
330 	struct vge_softc *sc;
331 	int i;
332 	uint16_t rval;
333 
334 	sc = (void *)dev;
335 	rval = 0;
336 	if (phy != (CSR_READ_1(sc, VGE_MIICFG) & 0x1F))
337 		return 0;
338 
339 	VGE_LOCK(sc);
340 	vge_miipoll_stop(sc);
341 
342 	/* Specify the register we want to read. */
343 	CSR_WRITE_1(sc, VGE_MIIADDR, reg);
344 
345 	/* Issue read command. */
346 	CSR_SETBIT_1(sc, VGE_MIICMD, VGE_MIICMD_RCMD);
347 
348 	/* Wait for the read command bit to self-clear. */
349 	for (i = 0; i < VGE_TIMEOUT; i++) {
350 		DELAY(1);
351 		if ((CSR_READ_1(sc, VGE_MIICMD) & VGE_MIICMD_RCMD) == 0)
352 			break;
353 	}
354 
355 	if (i == VGE_TIMEOUT)
356 		printf("%s: MII read timed out\n", sc->sc_dev.dv_xname);
357 	else
358 		rval = CSR_READ_2(sc, VGE_MIIDATA);
359 
360 	vge_miipoll_start(sc);
361 	VGE_UNLOCK(sc);
362 
363 	return rval;
364 }
365 
366 static void
367 vge_miibus_writereg(struct device *dev, int phy, int reg, int data)
368 {
369 	struct vge_softc *sc;
370 	int i;
371 
372 	sc = (void *)dev;
373 	if (phy != (CSR_READ_1(sc, VGE_MIICFG) & 0x1F))
374 		return;
375 
376 	VGE_LOCK(sc);
377 	vge_miipoll_stop(sc);
378 
379 	/* Specify the register we want to write. */
380 	CSR_WRITE_1(sc, VGE_MIIADDR, reg);
381 
382 	/* Specify the data we want to write. */
383 	CSR_WRITE_2(sc, VGE_MIIDATA, data);
384 
385 	/* Issue write command. */
386 	CSR_SETBIT_1(sc, VGE_MIICMD, VGE_MIICMD_WCMD);
387 
388 	/* Wait for the write command bit to self-clear. */
389 	for (i = 0; i < VGE_TIMEOUT; i++) {
390 		DELAY(1);
391 		if ((CSR_READ_1(sc, VGE_MIICMD) & VGE_MIICMD_WCMD) == 0)
392 			break;
393 	}
394 
395 	if (i == VGE_TIMEOUT) {
396 		printf("%s: MII write timed out\n", sc->sc_dev.dv_xname);
397 	}
398 
399 	vge_miipoll_start(sc);
400 	VGE_UNLOCK(sc);
401 }
402 
403 static void
404 vge_cam_clear(struct vge_softc *sc)
405 {
406 	int i;
407 
408 	/*
409 	 * Turn off all the mask bits. This tells the chip
410 	 * that none of the entries in the CAM filter are valid.
411 	 * desired entries will be enabled as we fill the filter in.
412 	 */
413 
414 	CSR_CLRBIT_1(sc, VGE_CAMCTL, VGE_CAMCTL_PAGESEL);
415 	CSR_SETBIT_1(sc, VGE_CAMCTL, VGE_PAGESEL_CAMMASK);
416 	CSR_WRITE_1(sc, VGE_CAMADDR, VGE_CAMADDR_ENABLE);
417 	for (i = 0; i < 8; i++)
418 		CSR_WRITE_1(sc, VGE_CAM0 + i, 0);
419 
420 	/* Clear the VLAN filter too. */
421 
422 	CSR_WRITE_1(sc, VGE_CAMADDR, VGE_CAMADDR_ENABLE|VGE_CAMADDR_AVSEL|0);
423 	for (i = 0; i < 8; i++)
424 		CSR_WRITE_1(sc, VGE_CAM0 + i, 0);
425 
426 	CSR_WRITE_1(sc, VGE_CAMADDR, 0);
427 	CSR_CLRBIT_1(sc, VGE_CAMCTL, VGE_CAMCTL_PAGESEL);
428 	CSR_SETBIT_1(sc, VGE_CAMCTL, VGE_PAGESEL_MAR);
429 
430 	sc->vge_camidx = 0;
431 }
432 
433 static int
434 vge_cam_set(struct vge_softc *sc, uint8_t *addr)
435 {
436 	int i, error;
437 
438 	error = 0;
439 
440 	if (sc->vge_camidx == VGE_CAM_MAXADDRS)
441 		return ENOSPC;
442 
443 	/* Select the CAM data page. */
444 	CSR_CLRBIT_1(sc, VGE_CAMCTL, VGE_CAMCTL_PAGESEL);
445 	CSR_SETBIT_1(sc, VGE_CAMCTL, VGE_PAGESEL_CAMDATA);
446 
447 	/* Set the filter entry we want to update and enable writing. */
448 	CSR_WRITE_1(sc, VGE_CAMADDR, VGE_CAMADDR_ENABLE|sc->vge_camidx);
449 
450 	/* Write the address to the CAM registers */
451 	for (i = 0; i < ETHER_ADDR_LEN; i++)
452 		CSR_WRITE_1(sc, VGE_CAM0 + i, addr[i]);
453 
454 	/* Issue a write command. */
455 	CSR_SETBIT_1(sc, VGE_CAMCTL, VGE_CAMCTL_WRITE);
456 
457 	/* Wake for it to clear. */
458 	for (i = 0; i < VGE_TIMEOUT; i++) {
459 		DELAY(1);
460 		if ((CSR_READ_1(sc, VGE_CAMCTL) & VGE_CAMCTL_WRITE) == 0)
461 			break;
462 	}
463 
464 	if (i == VGE_TIMEOUT) {
465 		printf("%s: setting CAM filter failed\n", sc->sc_dev.dv_xname);
466 		error = EIO;
467 		goto fail;
468 	}
469 
470 	/* Select the CAM mask page. */
471 	CSR_CLRBIT_1(sc, VGE_CAMCTL, VGE_CAMCTL_PAGESEL);
472 	CSR_SETBIT_1(sc, VGE_CAMCTL, VGE_PAGESEL_CAMMASK);
473 
474 	/* Set the mask bit that enables this filter. */
475 	CSR_SETBIT_1(sc, VGE_CAM0 + (sc->vge_camidx / 8),
476 	    1 << (sc->vge_camidx & 7));
477 
478 	sc->vge_camidx++;
479 
480  fail:
481 	/* Turn off access to CAM. */
482 	CSR_WRITE_1(sc, VGE_CAMADDR, 0);
483 	CSR_CLRBIT_1(sc, VGE_CAMCTL, VGE_CAMCTL_PAGESEL);
484 	CSR_SETBIT_1(sc, VGE_CAMCTL, VGE_PAGESEL_MAR);
485 
486 	return error;
487 }
488 
489 /*
490  * Program the multicast filter. We use the 64-entry CAM filter
491  * for perfect filtering. If there's more than 64 multicast addresses,
492  * we use the hash filter insted.
493  */
494 static void
495 vge_setmulti(struct vge_softc *sc)
496 {
497 	struct ifnet *ifp;
498 	int error;
499 	uint32_t h, hashes[2] = { 0, 0 };
500 	struct ether_multi *enm;
501 	struct ether_multistep step;
502 
503 	error = 0;
504 	ifp = &sc->sc_ethercom.ec_if;
505 
506 	/* First, zot all the multicast entries. */
507 	vge_cam_clear(sc);
508 	CSR_WRITE_4(sc, VGE_MAR0, 0);
509 	CSR_WRITE_4(sc, VGE_MAR1, 0);
510 	ifp->if_flags &= ~IFF_ALLMULTI;
511 
512 	/*
513 	 * If the user wants allmulti or promisc mode, enable reception
514 	 * of all multicast frames.
515 	 */
516 	if (ifp->if_flags & IFF_PROMISC) {
517  allmulti:
518 		CSR_WRITE_4(sc, VGE_MAR0, 0xFFFFFFFF);
519 		CSR_WRITE_4(sc, VGE_MAR1, 0xFFFFFFFF);
520 		ifp->if_flags |= IFF_ALLMULTI;
521 		return;
522 	}
523 
524 	/* Now program new ones */
525 	ETHER_FIRST_MULTI(step, &sc->sc_ethercom, enm);
526 	while (enm != NULL) {
527 		/*
528 		 * If multicast range, fall back to ALLMULTI.
529 		 */
530 		if (memcmp(enm->enm_addrlo, enm->enm_addrhi,
531 				ETHER_ADDR_LEN) != 0)
532 			goto allmulti;
533 
534 		error = vge_cam_set(sc, enm->enm_addrlo);
535 		if (error)
536 			break;
537 
538 		ETHER_NEXT_MULTI(step, enm);
539 	}
540 
541 	/* If there were too many addresses, use the hash filter. */
542 	if (error) {
543 		vge_cam_clear(sc);
544 
545 		ETHER_FIRST_MULTI(step, &sc->sc_ethercom, enm);
546 		while (enm != NULL) {
547 			/*
548 			 * If multicast range, fall back to ALLMULTI.
549 			 */
550 			if (memcmp(enm->enm_addrlo, enm->enm_addrhi,
551 					ETHER_ADDR_LEN) != 0)
552 				goto allmulti;
553 
554 			h = ether_crc32_be(enm->enm_addrlo,
555 			    ETHER_ADDR_LEN) >> 26;
556 			hashes[h >> 5] |= 1 << (h & 0x1f);
557 
558 			ETHER_NEXT_MULTI(step, enm);
559 		}
560 
561 		CSR_WRITE_4(sc, VGE_MAR0, hashes[0]);
562 		CSR_WRITE_4(sc, VGE_MAR1, hashes[1]);
563 	}
564 }
565 
566 static void
567 vge_reset(struct vge_softc *sc)
568 {
569 	int i;
570 
571 	CSR_WRITE_1(sc, VGE_CRS1, VGE_CR1_SOFTRESET);
572 
573 	for (i = 0; i < VGE_TIMEOUT; i++) {
574 		DELAY(5);
575 		if ((CSR_READ_1(sc, VGE_CRS1) & VGE_CR1_SOFTRESET) == 0)
576 			break;
577 	}
578 
579 	if (i == VGE_TIMEOUT) {
580 		printf("%s: soft reset timed out", sc->sc_dev.dv_xname);
581 		CSR_WRITE_1(sc, VGE_CRS3, VGE_CR3_STOP_FORCE);
582 		DELAY(2000);
583 	}
584 
585 	DELAY(5000);
586 
587 	CSR_SETBIT_1(sc, VGE_EECSR, VGE_EECSR_RELOAD);
588 
589 	for (i = 0; i < VGE_TIMEOUT; i++) {
590 		DELAY(5);
591 		if ((CSR_READ_1(sc, VGE_EECSR) & VGE_EECSR_RELOAD) == 0)
592 			break;
593 	}
594 
595 	if (i == VGE_TIMEOUT) {
596 		printf("%s: EEPROM reload timed out\n", sc->sc_dev.dv_xname);
597 		return;
598 	}
599 
600 	/*
601 	 * On some machine, the first read data from EEPROM could be
602 	 * messed up, so read one dummy data here to avoid the mess.
603 	 */
604 	(void)vge_read_eeprom(sc, 0);
605 
606 	CSR_CLRBIT_1(sc, VGE_CHIPCFG0, VGE_CHIPCFG0_PACPI);
607 }
608 
609 /*
610  * Probe for a VIA gigabit chip. Check the PCI vendor and device
611  * IDs against our list and return a device name if we find a match.
612  */
613 static int
614 vge_probe(struct device *parent __unused, struct cfdata *match __unused,
615     void *aux)
616 {
617 	struct pci_attach_args *pa = aux;
618 
619 	if (PCI_VENDOR(pa->pa_id) == PCI_VENDOR_VIATECH
620 	    && PCI_PRODUCT(pa->pa_id) == PCI_PRODUCT_VIATECH_VT612X)
621 		return 1;
622 
623 	return 0;
624 }
625 
626 static int
627 vge_allocmem(struct vge_softc *sc)
628 {
629 	int error;
630 	int nseg;
631 	int i;
632 	bus_dma_segment_t seg;
633 
634 	/*
635 	 * Allocate map for TX descriptor list.
636 	 */
637 	error = bus_dmamap_create(sc->vge_dmat,
638 	    VGE_TX_LIST_SZ, 1, VGE_TX_LIST_SZ, 0, BUS_DMA_NOWAIT,
639 	    &sc->vge_ldata.vge_tx_list_map);
640 	if (error) {
641 		printf("%s: could not allocate TX dma list map\n",
642 		    sc->sc_dev.dv_xname);
643 		return ENOMEM;
644 	}
645 
646 	/*
647 	 * Allocate memory for TX descriptor list.
648 	 */
649 
650 	error = bus_dmamem_alloc(sc->vge_dmat, VGE_TX_LIST_SZ, VGE_RING_ALIGN,
651 	    0, &seg, 1, &nseg, BUS_DMA_NOWAIT);
652 	if (error) {
653 		printf("%s: could not allocate TX ring dma memory\n",
654 		    sc->sc_dev.dv_xname);
655 		return ENOMEM;
656 	}
657 
658 	/* Map the memory to kernel VA space */
659 
660 	error = bus_dmamem_map(sc->vge_dmat, &seg, nseg, VGE_TX_LIST_SZ,
661 	     (caddr_t *)&sc->vge_ldata.vge_tx_list, BUS_DMA_NOWAIT);
662 	if (error) {
663 		printf("%s: could not map TX ring dma memory\n",
664 		    sc->sc_dev.dv_xname);
665 		return ENOMEM;
666 	}
667 
668 	/* Load the map for the TX ring. */
669 	error = bus_dmamap_load(sc->vge_dmat, sc->vge_ldata.vge_tx_list_map,
670 	    sc->vge_ldata.vge_tx_list, VGE_TX_LIST_SZ, NULL, BUS_DMA_NOWAIT);
671 	if (error) {
672 		printf("%s: could not load TX ring dma memory\n",
673 		    sc->sc_dev.dv_xname);
674 		return ENOMEM;
675 	}
676 
677 	/* Create DMA maps for TX buffers */
678 
679 	for (i = 0; i < VGE_TX_DESC_CNT; i++) {
680 		error = bus_dmamap_create(sc->vge_dmat, VGE_TX_MAXLEN,
681 		    VGE_TX_FRAGS, VGE_TX_MAXLEN, 0,
682 		    BUS_DMA_NOWAIT|BUS_DMA_ALLOCNOW,
683 		    &sc->vge_ldata.vge_tx_dmamap[i]);
684 		if (error) {
685 			printf("%s: can't create DMA map for TX\n",
686 			    sc->sc_dev.dv_xname);
687 			return ENOMEM;
688 		}
689 	}
690 
691 	/*
692 	 * Allocate map for RX descriptor list.
693 	 */
694 	error = bus_dmamap_create(sc->vge_dmat,
695 	    VGE_RX_LIST_SZ, 1, VGE_RX_LIST_SZ, 0, BUS_DMA_NOWAIT,
696 	    &sc->vge_ldata.vge_rx_list_map);
697 	if (error) {
698 		printf("%s: could not allocate RX dma list map\n",
699 		    sc->sc_dev.dv_xname);
700 		return ENOMEM;
701 	}
702 
703 	/* Allocate DMA'able memory for the RX ring */
704 
705 	error = bus_dmamem_alloc(sc->vge_dmat, VGE_RX_LIST_SZ, VGE_RING_ALIGN,
706 	    0, &seg, 1, &nseg, BUS_DMA_NOWAIT);
707 	if (error)
708 		return ENOMEM;
709 
710 	/* Map the memory to kernel VA space */
711 
712 	error = bus_dmamem_map(sc->vge_dmat, &seg, nseg, VGE_RX_LIST_SZ,
713 	     (caddr_t *)&sc->vge_ldata.vge_rx_list, BUS_DMA_NOWAIT);
714 	if (error)
715 		return ENOMEM;
716 
717 	/* Load the map for the RX ring. */
718 	error = bus_dmamap_load(sc->vge_dmat, sc->vge_ldata.vge_rx_list_map,
719 	    sc->vge_ldata.vge_rx_list, VGE_RX_LIST_SZ, NULL, BUS_DMA_NOWAIT);
720 	if (error) {
721 		printf("%s: could not load RX ring dma memory\n",
722 		    sc->sc_dev.dv_xname);
723 		return ENOMEM;
724 	}
725 
726 	/* Create DMA maps for RX buffers */
727 
728 	for (i = 0; i < VGE_RX_DESC_CNT; i++) {
729 		error = bus_dmamap_create(sc->vge_dmat, MCLBYTES,
730 		    1, MCLBYTES, 0, BUS_DMA_NOWAIT|BUS_DMA_ALLOCNOW,
731 		    &sc->vge_ldata.vge_rx_dmamap[i]);
732 		if (error) {
733 			printf("%s: can't create DMA map for RX\n",
734 			     sc->sc_dev.dv_xname);
735 			return ENOMEM;
736 		}
737 	}
738 
739 	return 0;
740 }
741 
742 /*
743  * Attach the interface. Allocate softc structures, do ifmedia
744  * setup and ethernet/BPF attach.
745  */
746 static void
747 vge_attach(struct device *parent __unused, struct device *self, void *aux)
748 {
749 	uint8_t	*eaddr;
750 	struct vge_softc *sc = (struct vge_softc *)self;
751 	struct ifnet *ifp;
752 	struct pci_attach_args *pa = aux;
753 	pci_chipset_tag_t pc = pa->pa_pc;
754 	const char *intrstr;
755 	pci_intr_handle_t ih;
756 	uint16_t val;
757 
758 	aprint_normal(": VIA VT612X Gigabit Ethernet (rev. %#x)\n",
759 		PCI_REVISION(pa->pa_class));
760 
761 	/* Make sure bus-mastering is enabled */
762         pci_conf_write(pc, pa->pa_tag, PCI_COMMAND_STATUS_REG,
763 	    pci_conf_read(pc, pa->pa_tag, PCI_COMMAND_STATUS_REG) |
764 	    PCI_COMMAND_MASTER_ENABLE);
765 
766 	/*
767 	 * Map control/status registers.
768 	 */
769 	if (pci_mapreg_map(pa, VGE_PCI_LOMEM, PCI_MAPREG_TYPE_MEM, 0,
770 	    &sc->vge_btag, &sc->vge_bhandle, NULL, NULL) != 0) {
771 		aprint_error("%s: couldn't map memory\n", sc->sc_dev.dv_xname);
772 		return;
773 	}
774 
775         /*
776          * Map and establish our interrupt.
777          */
778 	if (pci_intr_map(pa, &ih)) {
779 		aprint_error("%s: unable to map interrupt\n",
780 		    sc->sc_dev.dv_xname);
781 		return;
782 	}
783 	intrstr = pci_intr_string(pc, ih);
784 	sc->vge_intrhand = pci_intr_establish(pc, ih, IPL_NET, vge_intr, sc);
785 	if (sc->vge_intrhand == NULL) {
786 		printf("%s: unable to establish interrupt",
787 		    sc->sc_dev.dv_xname);
788 		if (intrstr != NULL)
789 			printf(" at %s", intrstr);
790 		printf("\n");
791 		return;
792 	}
793 	aprint_normal("%s: interrupting at %s\n", sc->sc_dev.dv_xname, intrstr);
794 
795 	/* Reset the adapter. */
796 	vge_reset(sc);
797 
798 	/*
799 	 * Get station address from the EEPROM.
800 	 */
801 	eaddr = sc->vge_eaddr;
802 	val = vge_read_eeprom(sc, VGE_EE_EADDR + 0);
803 	eaddr[0] = val & 0xff;
804 	eaddr[1] = val >> 8;
805 	val = vge_read_eeprom(sc, VGE_EE_EADDR + 1);
806 	eaddr[2] = val & 0xff;
807 	eaddr[3] = val >> 8;
808 	val = vge_read_eeprom(sc, VGE_EE_EADDR + 2);
809 	eaddr[4] = val & 0xff;
810 	eaddr[5] = val >> 8;
811 
812 	printf("%s: Ethernet address: %s\n", sc->sc_dev.dv_xname,
813 	    ether_sprintf(eaddr));
814 
815 	/*
816 	 * Use the 32bit tag. Hardware supports 48bit physical addresses,
817 	 * but we don't use that for now.
818 	 */
819 	sc->vge_dmat = pa->pa_dmat;
820 
821 	if (vge_allocmem(sc))
822 		return;
823 
824 	ifp = &sc->sc_ethercom.ec_if;
825 	ifp->if_softc = sc;
826 	strcpy(ifp->if_xname, sc->sc_dev.dv_xname);
827 	ifp->if_mtu = ETHERMTU;
828 	ifp->if_baudrate = IF_Gbps(1);
829 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
830 	ifp->if_ioctl = vge_ioctl;
831 	ifp->if_start = vge_start;
832 
833 	/*
834 	 * We can support 802.1Q VLAN-sized frames and jumbo
835 	 * Ethernet frames.
836 	 */
837 	sc->sc_ethercom.ec_capabilities |=
838 	    ETHERCAP_VLAN_MTU | ETHERCAP_JUMBO_MTU |
839 	    ETHERCAP_VLAN_HWTAGGING;
840 
841 	/*
842 	 * We can do IPv4/TCPv4/UDPv4 checksums in hardware.
843 	 */
844 	ifp->if_capabilities |=
845 	    IFCAP_CSUM_IPv4_Tx | IFCAP_CSUM_IPv4_Rx |
846 	    IFCAP_CSUM_TCPv4_Tx | IFCAP_CSUM_TCPv4_Rx |
847 	    IFCAP_CSUM_UDPv4_Tx | IFCAP_CSUM_UDPv4_Rx;
848 
849 #ifdef DEVICE_POLLING
850 #ifdef IFCAP_POLLING
851 	ifp->if_capabilities |= IFCAP_POLLING;
852 #endif
853 #endif
854 	ifp->if_watchdog = vge_watchdog;
855 	ifp->if_init = vge_init;
856 	IFQ_SET_MAXLEN(&ifp->if_snd, max(VGE_IFQ_MAXLEN, IFQ_MAXLEN));
857 
858 	/*
859 	 * Initialize our media structures and probe the MII.
860 	 */
861 	sc->sc_mii.mii_ifp = ifp;
862 	sc->sc_mii.mii_readreg = vge_miibus_readreg;
863 	sc->sc_mii.mii_writereg = vge_miibus_writereg;
864 	sc->sc_mii.mii_statchg = vge_miibus_statchg;
865 	ifmedia_init(&sc->sc_mii.mii_media, 0, vge_ifmedia_upd,
866 	    vge_ifmedia_sts);
867 	mii_attach(&sc->sc_dev, &sc->sc_mii, 0xffffffff, MII_PHY_ANY,
868 	    MII_OFFSET_ANY, MIIF_DOPAUSE);
869 	if (LIST_FIRST(&sc->sc_mii.mii_phys) == NULL) {
870 		ifmedia_add(&sc->sc_mii.mii_media, IFM_ETHER|IFM_NONE, 0, NULL);
871 		ifmedia_set(&sc->sc_mii.mii_media, IFM_ETHER|IFM_NONE);
872 	} else
873 		ifmedia_set(&sc->sc_mii.mii_media, IFM_ETHER|IFM_AUTO);
874 
875 	/*
876 	 * Attach the interface.
877 	 */
878 	if_attach(ifp);
879 	ether_ifattach(ifp, eaddr);
880 
881 	callout_init(&sc->vge_timeout);
882 	callout_setfunc(&sc->vge_timeout, vge_tick, sc);
883 
884 	/*
885 	 * Make sure the interface is shutdown during reboot.
886 	 */
887 	if (shutdownhook_establish(vge_shutdown, sc) == NULL) {
888 		printf("%s: WARNING: unable to establish shutdown hook\n",
889 		    sc->sc_dev.dv_xname);
890 	}
891 }
892 
893 static int
894 vge_newbuf(struct vge_softc *sc, int idx, struct mbuf *m)
895 {
896 	struct vge_rx_desc *d;
897 	struct mbuf *m_new;
898 	bus_dmamap_t map;
899 	int i;
900 
901 	m_new = NULL;
902 	if (m == NULL) {
903 		MGETHDR(m_new, M_DONTWAIT, MT_DATA);
904 		if (m_new == NULL)
905 			return ENOBUFS;
906 
907 		MCLGET(m_new, M_DONTWAIT);
908 		if ((m_new->m_flags & M_EXT) == 0) {
909 			m_freem(m_new);
910 			return ENOBUFS;
911 		}
912 
913 		m = m_new;
914 	} else
915 		m->m_data = m->m_ext.ext_buf;
916 
917 
918 #ifndef __NO_STRICT_ALIGNMENT
919 	/*
920 	 * This is part of an evil trick to deal with non-x86 platforms.
921 	 * The VIA chip requires RX buffers to be aligned on 32-bit
922 	 * boundaries, but that will hose non-x86 machines. To get around
923 	 * this, we leave some empty space at the start of each buffer
924 	 * and for non-x86 hosts, we copy the buffer back two bytes
925 	 * to achieve word alignment. This is slightly more efficient
926 	 * than allocating a new buffer, copying the contents, and
927 	 * discarding the old buffer.
928 	 */
929 	m->m_len = m->m_pkthdr.len = MCLBYTES - VGE_ETHER_ALIGN;
930 	m->m_data += VGE_ETHER_ALIGN;
931 #else
932 	m->m_len = m->m_pkthdr.len = MCLBYTES;
933 #endif
934 	map = sc->vge_ldata.vge_rx_dmamap[idx];
935 
936 	if (bus_dmamap_load_mbuf(sc->vge_dmat, map, m, BUS_DMA_NOWAIT) != 0)
937 		goto out;
938 
939 	d = &sc->vge_ldata.vge_rx_list[idx];
940 
941 	/* If this descriptor is still owned by the chip, bail. */
942 
943 	VGE_RXDESCSYNC(sc, idx, BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
944 	if (le32toh(d->vge_sts) & VGE_RDSTS_OWN) {
945 		printf("%s: tried to map busy descriptor\n",
946 		    sc->sc_dev.dv_xname);
947 		VGE_RXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD);
948 		goto out;
949 	}
950 
951 	d->vge_buflen =
952 	    htole16(VGE_BUFLEN(map->dm_segs[0].ds_len) | VGE_RXDESC_I);
953 	d->vge_addrlo = htole32(VGE_ADDR_LO(map->dm_segs[0].ds_addr));
954 	d->vge_addrhi = htole16(VGE_ADDR_HI(map->dm_segs[0].ds_addr) & 0xFFFF);
955 	d->vge_sts = 0;
956 	d->vge_ctl = 0;
957 	VGE_RXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
958 
959 	bus_dmamap_sync(sc->vge_dmat,
960 	    sc->vge_ldata.vge_rx_dmamap[idx],
961 	    0, sc->vge_ldata.vge_rx_dmamap[idx]->dm_mapsize,
962 	    BUS_DMASYNC_PREREAD);
963 
964 	/*
965 	 * Note: the manual fails to document the fact that for
966 	 * proper opration, the driver needs to replentish the RX
967 	 * DMA ring 4 descriptors at a time (rather than one at a
968 	 * time, like most chips). We can allocate the new buffers
969 	 * but we should not set the OWN bits until we're ready
970 	 * to hand back 4 of them in one shot.
971 	 */
972 
973 #define VGE_RXCHUNK 4
974 	sc->vge_rx_consumed++;
975 	if (sc->vge_rx_consumed == VGE_RXCHUNK) {
976 		for (i = idx; i != idx - sc->vge_rx_consumed; i--) {
977 			sc->vge_ldata.vge_rx_list[i].vge_sts |=
978 			    htole32(VGE_RDSTS_OWN);
979 			VGE_RXDESCSYNC(sc, i,
980 			    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
981 		}
982 		sc->vge_rx_consumed = 0;
983 	}
984 
985 	sc->vge_ldata.vge_rx_mbuf[idx] = m;
986 
987 	return 0;
988  out:
989 	if (m_new != NULL)
990 		m_freem(m_new);
991 	return ENOMEM;
992 }
993 
994 static int
995 vge_tx_list_init(struct vge_softc *sc)
996 {
997 
998 	memset((char *)sc->vge_ldata.vge_tx_list, 0, VGE_TX_LIST_SZ);
999 	bus_dmamap_sync(sc->vge_dmat, sc->vge_ldata.vge_tx_list_map,
1000 	    0, sc->vge_ldata.vge_tx_list_map->dm_mapsize,
1001 	    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1002 
1003 	memset((char *)&sc->vge_ldata.vge_tx_mbuf, 0,
1004 	    (VGE_TX_DESC_CNT * sizeof(struct mbuf *)));
1005 
1006 	sc->vge_ldata.vge_tx_prodidx = 0;
1007 	sc->vge_ldata.vge_tx_considx = 0;
1008 	sc->vge_ldata.vge_tx_free = VGE_TX_DESC_CNT;
1009 
1010 	return 0;
1011 }
1012 
1013 static int
1014 vge_rx_list_init(struct vge_softc *sc)
1015 {
1016 	int i;
1017 
1018 	memset((char *)sc->vge_ldata.vge_rx_list, 0, VGE_RX_LIST_SZ);
1019 	memset((char *)&sc->vge_ldata.vge_rx_mbuf, 0,
1020 	    (VGE_RX_DESC_CNT * sizeof(struct mbuf *)));
1021 
1022 	sc->vge_rx_consumed = 0;
1023 
1024 	for (i = 0; i < VGE_RX_DESC_CNT; i++) {
1025 		if (vge_newbuf(sc, i, NULL) == ENOBUFS)
1026 			return (ENOBUFS);
1027 	}
1028 
1029 	sc->vge_ldata.vge_rx_prodidx = 0;
1030 	sc->vge_rx_consumed = 0;
1031 	sc->vge_head = sc->vge_tail = NULL;
1032 
1033 	return 0;
1034 }
1035 
1036 #ifndef __NO_STRICT_ALIGNMENT
1037 static inline void
1038 vge_fixup_rx(struct mbuf *m)
1039 {
1040 	int i;
1041 	uint16_t *src, *dst;
1042 
1043 	src = mtod(m, uint16_t *);
1044 	dst = src - 1;
1045 
1046 	for (i = 0; i < (m->m_len / sizeof(uint16_t) + 1); i++)
1047 		*dst++ = *src++;
1048 
1049 	m->m_data -= ETHER_ALIGN;
1050 }
1051 #endif
1052 
1053 /*
1054  * RX handler. We support the reception of jumbo frames that have
1055  * been fragmented across multiple 2K mbuf cluster buffers.
1056  */
1057 static void
1058 vge_rxeof(struct vge_softc *sc)
1059 {
1060 	struct mbuf *m;
1061 	struct ifnet *ifp;
1062 	int idx, total_len, lim;
1063 	struct vge_rx_desc *cur_rx;
1064 	uint32_t rxstat, rxctl;
1065 
1066 	VGE_LOCK_ASSERT(sc);
1067 	ifp = &sc->sc_ethercom.ec_if;
1068 	idx = sc->vge_ldata.vge_rx_prodidx;
1069 	lim = 0;
1070 
1071 	/* Invalidate the descriptor memory */
1072 
1073 	for (;;) {
1074 		cur_rx = &sc->vge_ldata.vge_rx_list[idx];
1075 
1076 		VGE_RXDESCSYNC(sc, idx,
1077 		    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1078 		rxstat = le32toh(cur_rx->vge_sts);
1079 		if ((rxstat & VGE_RDSTS_OWN) != 0) {
1080 			VGE_RXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD);
1081 			break;
1082 		}
1083 
1084 #ifdef DEVICE_POLLING
1085 		if (ifp->if_flags & IFF_POLLING) {
1086 			if (sc->rxcycles <= 0)
1087 				break;
1088 			sc->rxcycles--;
1089 		}
1090 #endif /* DEVICE_POLLING */
1091 
1092 		m = sc->vge_ldata.vge_rx_mbuf[idx];
1093 		total_len = (rxstat & VGE_RDSTS_BUFSIZ) >> 16;
1094 		rxctl = le32toh(cur_rx->vge_ctl);
1095 
1096 		/* Invalidate the RX mbuf and unload its map */
1097 
1098 		bus_dmamap_sync(sc->vge_dmat,
1099 		    sc->vge_ldata.vge_rx_dmamap[idx],
1100 		    0, sc->vge_ldata.vge_rx_dmamap[idx]->dm_mapsize,
1101 		    BUS_DMASYNC_POSTREAD);
1102 		bus_dmamap_unload(sc->vge_dmat,
1103 		    sc->vge_ldata.vge_rx_dmamap[idx]);
1104 
1105 		/*
1106 		 * If the 'start of frame' bit is set, this indicates
1107 		 * either the first fragment in a multi-fragment receive,
1108 		 * or an intermediate fragment. Either way, we want to
1109 		 * accumulate the buffers.
1110 		 */
1111 		if (rxstat & VGE_RXPKT_SOF) {
1112 			m->m_len = MCLBYTES - VGE_ETHER_ALIGN;
1113 			if (sc->vge_head == NULL)
1114 				sc->vge_head = sc->vge_tail = m;
1115 			else {
1116 				m->m_flags &= ~M_PKTHDR;
1117 				sc->vge_tail->m_next = m;
1118 				sc->vge_tail = m;
1119 			}
1120 			vge_newbuf(sc, idx, NULL);
1121 			VGE_RX_DESC_INC(idx);
1122 			continue;
1123 		}
1124 
1125 		/*
1126 		 * Bad/error frames will have the RXOK bit cleared.
1127 		 * However, there's one error case we want to allow:
1128 		 * if a VLAN tagged frame arrives and the chip can't
1129 		 * match it against the CAM filter, it considers this
1130 		 * a 'VLAN CAM filter miss' and clears the 'RXOK' bit.
1131 		 * We don't want to drop the frame though: our VLAN
1132 		 * filtering is done in software.
1133 		 */
1134 		if (!(rxstat & VGE_RDSTS_RXOK) && !(rxstat & VGE_RDSTS_VIDM)
1135 		    && !(rxstat & VGE_RDSTS_CSUMERR)) {
1136 			ifp->if_ierrors++;
1137 			/*
1138 			 * If this is part of a multi-fragment packet,
1139 			 * discard all the pieces.
1140 			 */
1141 			if (sc->vge_head != NULL) {
1142 				m_freem(sc->vge_head);
1143 				sc->vge_head = sc->vge_tail = NULL;
1144 			}
1145 			vge_newbuf(sc, idx, m);
1146 			VGE_RX_DESC_INC(idx);
1147 			continue;
1148 		}
1149 
1150 		/*
1151 		 * If allocating a replacement mbuf fails,
1152 		 * reload the current one.
1153 		 */
1154 
1155 		if (vge_newbuf(sc, idx, NULL)) {
1156 			ifp->if_ierrors++;
1157 			if (sc->vge_head != NULL) {
1158 				m_freem(sc->vge_head);
1159 				sc->vge_head = sc->vge_tail = NULL;
1160 			}
1161 			vge_newbuf(sc, idx, m);
1162 			VGE_RX_DESC_INC(idx);
1163 			continue;
1164 		}
1165 
1166 		VGE_RX_DESC_INC(idx);
1167 
1168 		if (sc->vge_head != NULL) {
1169 			m->m_len = total_len % (MCLBYTES - VGE_ETHER_ALIGN);
1170 			/*
1171 			 * Special case: if there's 4 bytes or less
1172 			 * in this buffer, the mbuf can be discarded:
1173 			 * the last 4 bytes is the CRC, which we don't
1174 			 * care about anyway.
1175 			 */
1176 			if (m->m_len <= ETHER_CRC_LEN) {
1177 				sc->vge_tail->m_len -=
1178 				    (ETHER_CRC_LEN - m->m_len);
1179 				m_freem(m);
1180 			} else {
1181 				m->m_len -= ETHER_CRC_LEN;
1182 				m->m_flags &= ~M_PKTHDR;
1183 				sc->vge_tail->m_next = m;
1184 			}
1185 			m = sc->vge_head;
1186 			sc->vge_head = sc->vge_tail = NULL;
1187 			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
1188 		} else
1189 			m->m_pkthdr.len = m->m_len =
1190 			    (total_len - ETHER_CRC_LEN);
1191 
1192 #ifndef __NO_STRICT_ALIGNMENT
1193 		vge_fixup_rx(m);
1194 #endif
1195 		ifp->if_ipackets++;
1196 		m->m_pkthdr.rcvif = ifp;
1197 
1198 		/* Do RX checksumming if enabled */
1199 		if (ifp->if_csum_flags_rx & M_CSUM_IPv4) {
1200 
1201 			/* Check IP header checksum */
1202 			if (rxctl & VGE_RDCTL_IPPKT)
1203 				m->m_pkthdr.csum_flags |= M_CSUM_IPv4;
1204 			if ((rxctl & VGE_RDCTL_IPCSUMOK) == 0)
1205 				m->m_pkthdr.csum_flags |= M_CSUM_IPv4_BAD;
1206 		}
1207 
1208 		if (ifp->if_csum_flags_rx & M_CSUM_TCPv4) {
1209 			/* Check UDP checksum */
1210 			if (rxctl & VGE_RDCTL_TCPPKT)
1211 				m->m_pkthdr.csum_flags |= M_CSUM_TCPv4;
1212 
1213 			if ((rxctl & VGE_RDCTL_PROTOCSUMOK) == 0)
1214 				m->m_pkthdr.csum_flags |= M_CSUM_TCP_UDP_BAD;
1215 		}
1216 
1217 		if (ifp->if_csum_flags_rx & M_CSUM_UDPv4) {
1218 			/* Check UDP checksum */
1219 			if (rxctl & VGE_RDCTL_UDPPKT)
1220 				m->m_pkthdr.csum_flags |= M_CSUM_UDPv4;
1221 
1222 			if ((rxctl & VGE_RDCTL_PROTOCSUMOK) == 0)
1223 				m->m_pkthdr.csum_flags |= M_CSUM_TCP_UDP_BAD;
1224 		}
1225 
1226 		if (rxstat & VGE_RDSTS_VTAG)
1227 			VLAN_INPUT_TAG(ifp, m,
1228 			    ntohs((rxctl & VGE_RDCTL_VLANID)), continue);
1229 
1230 #if NBPFILTER > 0
1231 		/*
1232 		 * Handle BPF listeners.
1233 		 */
1234 		if (ifp->if_bpf)
1235 			bpf_mtap(ifp->if_bpf, m);
1236 #endif
1237 
1238 		VGE_UNLOCK(sc);
1239 		(*ifp->if_input)(ifp, m);
1240 		VGE_LOCK(sc);
1241 
1242 		lim++;
1243 		if (lim == VGE_RX_DESC_CNT)
1244 			break;
1245 
1246 	}
1247 
1248 	sc->vge_ldata.vge_rx_prodidx = idx;
1249 	CSR_WRITE_2(sc, VGE_RXDESC_RESIDUECNT, lim);
1250 }
1251 
1252 static void
1253 vge_txeof(struct vge_softc *sc)
1254 {
1255 	struct ifnet *ifp;
1256 	uint32_t txstat;
1257 	int idx;
1258 
1259 	ifp = &sc->sc_ethercom.ec_if;
1260 	idx = sc->vge_ldata.vge_tx_considx;
1261 
1262 	while (idx != sc->vge_ldata.vge_tx_prodidx) {
1263 		VGE_TXDESCSYNC(sc, idx,
1264 		    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1265 
1266 		txstat = le32toh(sc->vge_ldata.vge_tx_list[idx].vge_sts);
1267 		if (txstat & VGE_TDSTS_OWN) {
1268 			VGE_TXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD);
1269 			break;
1270 		}
1271 
1272 		m_freem(sc->vge_ldata.vge_tx_mbuf[idx]);
1273 		sc->vge_ldata.vge_tx_mbuf[idx] = NULL;
1274 		bus_dmamap_unload(sc->vge_dmat,
1275 		    sc->vge_ldata.vge_tx_dmamap[idx]);
1276 		if (txstat & (VGE_TDSTS_EXCESSCOLL|VGE_TDSTS_COLL))
1277 			ifp->if_collisions++;
1278 		if (txstat & VGE_TDSTS_TXERR)
1279 			ifp->if_oerrors++;
1280 		else
1281 			ifp->if_opackets++;
1282 
1283 		sc->vge_ldata.vge_tx_free++;
1284 		VGE_TX_DESC_INC(idx);
1285 	}
1286 
1287 	/* No changes made to the TX ring, so no flush needed */
1288 
1289 	if (idx != sc->vge_ldata.vge_tx_considx) {
1290 		sc->vge_ldata.vge_tx_considx = idx;
1291 		ifp->if_flags &= ~IFF_OACTIVE;
1292 		ifp->if_timer = 0;
1293 	}
1294 
1295 	/*
1296 	 * If not all descriptors have been released reaped yet,
1297 	 * reload the timer so that we will eventually get another
1298 	 * interrupt that will cause us to re-enter this routine.
1299 	 * This is done in case the transmitter has gone idle.
1300 	 */
1301 	if (sc->vge_ldata.vge_tx_free != VGE_TX_DESC_CNT) {
1302 		CSR_WRITE_1(sc, VGE_CRS1, VGE_CR1_TIMER0_ENABLE);
1303 	}
1304 }
1305 
1306 static void
1307 vge_tick(void *xsc)
1308 {
1309 	struct vge_softc *sc;
1310 	struct ifnet *ifp;
1311 	struct mii_data *mii;
1312 	int s;
1313 
1314 	sc = xsc;
1315 	ifp = &sc->sc_ethercom.ec_if;
1316 	mii = &sc->sc_mii;
1317 
1318 	s = splnet();
1319 
1320 	VGE_LOCK(sc);
1321 
1322 	callout_schedule(&sc->vge_timeout, hz);
1323 
1324 	mii_tick(mii);
1325 	if (sc->vge_link) {
1326 		if (!(mii->mii_media_status & IFM_ACTIVE))
1327 			sc->vge_link = 0;
1328 	} else {
1329 		if (mii->mii_media_status & IFM_ACTIVE &&
1330 		    IFM_SUBTYPE(mii->mii_media_active) != IFM_NONE) {
1331 			sc->vge_link = 1;
1332 			if (!IFQ_IS_EMPTY(&ifp->if_snd))
1333 				vge_start(ifp);
1334 		}
1335 	}
1336 
1337 	VGE_UNLOCK(sc);
1338 
1339 	splx(s);
1340 }
1341 
1342 #ifdef DEVICE_POLLING
1343 static void
1344 vge_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
1345 {
1346 	struct vge_softc *sc = ifp->if_softc;
1347 
1348 	VGE_LOCK(sc);
1349 #ifdef IFCAP_POLLING
1350 	if (!(ifp->if_capenable & IFCAP_POLLING)) {
1351 		ether_poll_deregister(ifp);
1352 		cmd = POLL_DEREGISTER;
1353 	}
1354 #endif
1355 	if (cmd == POLL_DEREGISTER) { /* final call, enable interrupts */
1356 		CSR_WRITE_4(sc, VGE_IMR, VGE_INTRS);
1357 		CSR_WRITE_4(sc, VGE_ISR, 0xFFFFFFFF);
1358 		CSR_WRITE_1(sc, VGE_CRS3, VGE_CR3_INT_GMSK);
1359 		goto done;
1360 	}
1361 
1362 	sc->rxcycles = count;
1363 	vge_rxeof(sc);
1364 	vge_txeof(sc);
1365 
1366 #if __FreeBSD_version < 502114
1367 	if (ifp->if_snd.ifq_head != NULL)
1368 #else
1369 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
1370 #endif
1371 		taskqueue_enqueue(taskqueue_swi, &sc->vge_txtask);
1372 
1373 	if (cmd == POLL_AND_CHECK_STATUS) { /* also check status register */
1374 		uint32_t status;
1375 		status = CSR_READ_4(sc, VGE_ISR);
1376 		if (status == 0xFFFFFFFF)
1377 			goto done;
1378 		if (status)
1379 			CSR_WRITE_4(sc, VGE_ISR, status);
1380 
1381 		/*
1382 		 * XXX check behaviour on receiver stalls.
1383 		 */
1384 
1385 		if (status & VGE_ISR_TXDMA_STALL ||
1386 		    status & VGE_ISR_RXDMA_STALL)
1387 			vge_init(sc);
1388 
1389 		if (status & (VGE_ISR_RXOFLOW|VGE_ISR_RXNODESC)) {
1390 			vge_rxeof(sc);
1391 			ifp->if_ierrors++;
1392 			CSR_WRITE_1(sc, VGE_RXQCSRS, VGE_RXQCSR_RUN);
1393 			CSR_WRITE_1(sc, VGE_RXQCSRS, VGE_RXQCSR_WAK);
1394 		}
1395 	}
1396  done:
1397 	VGE_UNLOCK(sc);
1398 }
1399 #endif /* DEVICE_POLLING */
1400 
1401 static int
1402 vge_intr(void *arg)
1403 {
1404 	struct vge_softc *sc;
1405 	struct ifnet *ifp;
1406 	uint32_t status;
1407 	int claim;
1408 
1409 	sc = arg;
1410 	claim = 0;
1411 	if (sc->suspended) {
1412 		return claim;
1413 	}
1414 
1415 	ifp = &sc->sc_ethercom.ec_if;
1416 
1417 	VGE_LOCK(sc);
1418 
1419 	if (!(ifp->if_flags & IFF_UP)) {
1420 		VGE_UNLOCK(sc);
1421 		return claim;
1422 	}
1423 
1424 #ifdef DEVICE_POLLING
1425 	if  (ifp->if_flags & IFF_POLLING)
1426 		goto done;
1427 	if (
1428 #ifdef IFCAP_POLLING
1429 	    (ifp->if_capenable & IFCAP_POLLING) &&
1430 #endif
1431 	    ether_poll_register(vge_poll, ifp)) { /* ok, disable interrupts */
1432 		CSR_WRITE_4(sc, VGE_IMR, 0);
1433 		CSR_WRITE_1(sc, VGE_CRC3, VGE_CR3_INT_GMSK);
1434 		vge_poll(ifp, 0, 1);
1435 		goto done;
1436 	}
1437 
1438 #endif /* DEVICE_POLLING */
1439 
1440 	/* Disable interrupts */
1441 	CSR_WRITE_1(sc, VGE_CRC3, VGE_CR3_INT_GMSK);
1442 
1443 	for (;;) {
1444 
1445 		status = CSR_READ_4(sc, VGE_ISR);
1446 		/* If the card has gone away the read returns 0xffff. */
1447 		if (status == 0xFFFFFFFF)
1448 			break;
1449 
1450 		if (status) {
1451 			claim = 1;
1452 			CSR_WRITE_4(sc, VGE_ISR, status);
1453 		}
1454 
1455 		if ((status & VGE_INTRS) == 0)
1456 			break;
1457 
1458 		if (status & (VGE_ISR_RXOK|VGE_ISR_RXOK_HIPRIO))
1459 			vge_rxeof(sc);
1460 
1461 		if (status & (VGE_ISR_RXOFLOW|VGE_ISR_RXNODESC)) {
1462 			vge_rxeof(sc);
1463 			CSR_WRITE_1(sc, VGE_RXQCSRS, VGE_RXQCSR_RUN);
1464 			CSR_WRITE_1(sc, VGE_RXQCSRS, VGE_RXQCSR_WAK);
1465 		}
1466 
1467 		if (status & (VGE_ISR_TXOK0|VGE_ISR_TIMER0))
1468 			vge_txeof(sc);
1469 
1470 		if (status & (VGE_ISR_TXDMA_STALL|VGE_ISR_RXDMA_STALL))
1471 			vge_init(ifp);
1472 
1473 		if (status & VGE_ISR_LINKSTS)
1474 			vge_tick(sc);
1475 	}
1476 
1477 	/* Re-enable interrupts */
1478 	CSR_WRITE_1(sc, VGE_CRS3, VGE_CR3_INT_GMSK);
1479 
1480 #ifdef DEVICE_POLLING
1481  done:
1482 #endif
1483 	VGE_UNLOCK(sc);
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_tx_desc *d;
1495 	struct vge_tx_frag *f;
1496 	struct mbuf *m_new;
1497 	bus_dmamap_t map;
1498 	int seg, error, flags;
1499 	struct m_tag *mtag;
1500 	size_t sz;
1501 
1502 	d = &sc->vge_ldata.vge_tx_list[idx];
1503 
1504 	/* If this descriptor is still owned by the chip, bail. */
1505 	if (sc->vge_ldata.vge_tx_free <= 2) {
1506 		VGE_TXDESCSYNC(sc, idx,
1507 		    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1508 		if (le32toh(d->vge_sts) & VGE_TDSTS_OWN) {
1509 			VGE_TXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD);
1510 			return ENOBUFS;
1511 		}
1512 	}
1513 
1514 	map = sc->vge_ldata.vge_tx_dmamap[idx];
1515 	error = bus_dmamap_load_mbuf(sc->vge_dmat, map, m_head, BUS_DMA_NOWAIT);
1516 
1517 	/* If too many segments to map, coalesce */
1518 	if (error == EFBIG) {
1519 		m_new = m_defrag(m_head, M_DONTWAIT);
1520 		if (m_new == NULL)
1521 			return (error);
1522 
1523 		error = bus_dmamap_load_mbuf(sc->vge_dmat, map,
1524 		    m_new, BUS_DMA_NOWAIT);
1525 		if (error) {
1526 			m_freem(m_new);
1527 			return error;
1528 		}
1529 
1530 		m_head = m_new;
1531 	} else if (error)
1532 		return error;
1533 
1534 	for (seg = 0, f = &d->vge_frag[0]; seg < map->dm_nsegs; seg++, f++) {
1535 		f->vge_buflen = htole16(VGE_BUFLEN(map->dm_segs[seg].ds_len));
1536 		f->vge_addrlo = htole32(VGE_ADDR_LO(map->dm_segs[seg].ds_addr));
1537 		f->vge_addrhi = htole16(VGE_ADDR_HI(map->dm_segs[seg].ds_addr));
1538 	}
1539 
1540 	/* Argh. This chip does not autopad short frames */
1541 
1542 	sz = m_head->m_pkthdr.len;
1543 	if (m_head->m_pkthdr.len < VGE_MIN_FRAMELEN) {
1544 		f->vge_buflen = htole16(VGE_BUFLEN(VGE_MIN_FRAMELEN - sz));
1545 		f->vge_addrlo = htole32(VGE_ADDR_LO(map->dm_segs[0].ds_addr));
1546 		f->vge_addrhi =
1547 		    htole16(VGE_ADDR_HI(map->dm_segs[0].ds_addr) & 0xFFFF);
1548 		sz = VGE_MIN_FRAMELEN;
1549 		seg++;
1550 	}
1551 	VGE_TXFRAGSYNC(sc, idx, seg, BUS_DMASYNC_PREWRITE);
1552 
1553 	/*
1554 	 * When telling the chip how many segments there are, we
1555 	 * must use nsegs + 1 instead of just nsegs. Darned if I
1556 	 * know why.
1557 	 */
1558 	seg++;
1559 
1560 	flags = 0;
1561 	if (m_head->m_pkthdr.csum_flags & M_CSUM_IPv4)
1562 		flags |= VGE_TDCTL_IPCSUM;
1563 	if (m_head->m_pkthdr.csum_flags & M_CSUM_TCPv4)
1564 		flags |= VGE_TDCTL_TCPCSUM;
1565 	if (m_head->m_pkthdr.csum_flags & M_CSUM_UDPv4)
1566 		flags |= VGE_TDCTL_UDPCSUM;
1567 	d->vge_sts = htole32(sz << 16);
1568 	d->vge_ctl = htole32(flags | (seg << 28) | VGE_TD_LS_NORM);
1569 
1570 	if (sz > ETHERMTU + ETHER_HDR_LEN)
1571 		d->vge_ctl |= htole32(VGE_TDCTL_JUMBO);
1572 
1573 	bus_dmamap_sync(sc->vge_dmat, map, 0, map->dm_mapsize,
1574 	    BUS_DMASYNC_PREWRITE);
1575 
1576 	sc->vge_ldata.vge_tx_mbuf[idx] = m_head;
1577 	sc->vge_ldata.vge_tx_free--;
1578 
1579 	/*
1580 	 * Set up hardware VLAN tagging.
1581 	 */
1582 
1583 	mtag = VLAN_OUTPUT_TAG(&sc->sc_ethercom, m_head);
1584 	if (mtag != NULL)
1585 		d->vge_ctl |=
1586 		    htole32(htons(VLAN_TAG_VALUE(mtag)) | VGE_TDCTL_VTAG);
1587 
1588 	d->vge_sts |= htole32(VGE_TDSTS_OWN);
1589 
1590 	VGE_TXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1591 
1592 	return 0;
1593 }
1594 
1595 /*
1596  * Main transmit routine.
1597  */
1598 
1599 static void
1600 vge_start(struct ifnet *ifp)
1601 {
1602 	struct vge_softc *sc;
1603 	struct mbuf *m_head;
1604 	int idx, pidx, error;
1605 
1606 	sc = ifp->if_softc;
1607 	VGE_LOCK(sc);
1608 
1609 	if (!sc->vge_link ||
1610 	    (ifp->if_flags & (IFF_RUNNING|IFF_OACTIVE)) != IFF_RUNNING) {
1611 		VGE_UNLOCK(sc);
1612 		return;
1613 	}
1614 
1615 	m_head = NULL;
1616 	idx = sc->vge_ldata.vge_tx_prodidx;
1617 
1618 	pidx = idx - 1;
1619 	if (pidx < 0)
1620 		pidx = VGE_TX_DESC_CNT - 1;
1621 
1622 	/*
1623 	 * Loop through the send queue, setting up transmit descriptors
1624 	 * until we drain the queue, or use up all available transmit
1625 	 * descriptors.
1626 	 */
1627 	for (;;) {
1628 		/* Grab a packet off the queue. */
1629 		IFQ_POLL(&ifp->if_snd, m_head);
1630 		if (m_head == NULL)
1631 			break;
1632 
1633 		if (sc->vge_ldata.vge_tx_mbuf[idx] != NULL) {
1634 			/*
1635 			 * Slot already used, stop for now.
1636 			 */
1637 			ifp->if_flags |= IFF_OACTIVE;
1638 			break;
1639 		}
1640 
1641 		if ((error = vge_encap(sc, m_head, idx))) {
1642 			if (error == EFBIG) {
1643 				printf("%s: Tx packet consumes too many "
1644 				    "DMA segments, dropping...\n",
1645 				    sc->sc_dev.dv_xname);
1646 				IFQ_DEQUEUE(&ifp->if_snd, m_head);
1647 				m_freem(m_head);
1648 				continue;
1649 			}
1650 
1651 			/*
1652 			 * Short on resources, just stop for now.
1653 			 */
1654 			if (error == ENOBUFS)
1655 				ifp->if_flags |= IFF_OACTIVE;
1656 			break;
1657 		}
1658 
1659 		IFQ_DEQUEUE(&ifp->if_snd, m_head);
1660 
1661 		/*
1662 		 * WE ARE NOW COMMITTED TO TRANSMITTING THE PACKET.
1663 		 */
1664 
1665 		sc->vge_ldata.vge_tx_list[pidx].vge_frag[0].vge_buflen |=
1666 		    htole16(VGE_TXDESC_Q);
1667 		VGE_TXDESCSYNC(sc, pidx,
1668 		    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1669 
1670 		if (sc->vge_ldata.vge_tx_mbuf[idx] != m_head) {
1671 			m_freem(m_head);
1672 			m_head = sc->vge_ldata.vge_tx_mbuf[idx];
1673 		}
1674 
1675 		pidx = idx;
1676 		VGE_TX_DESC_INC(idx);
1677 
1678 		/*
1679 		 * If there's a BPF listener, bounce a copy of this frame
1680 		 * to him.
1681 		 */
1682 #if NBPFILTER > 0
1683 		if (ifp->if_bpf)
1684 			bpf_mtap(ifp->if_bpf, m_head);
1685 #endif
1686 	}
1687 
1688 	if (idx == sc->vge_ldata.vge_tx_prodidx) {
1689 		VGE_UNLOCK(sc);
1690 		return;
1691 	}
1692 
1693 	/* Issue a transmit command. */
1694 	CSR_WRITE_2(sc, VGE_TXQCSRS, VGE_TXQCSR_WAK0);
1695 
1696 	sc->vge_ldata.vge_tx_prodidx = idx;
1697 
1698 	/*
1699 	 * Use the countdown timer for interrupt moderation.
1700 	 * 'TX done' interrupts are disabled. Instead, we reset the
1701 	 * countdown timer, which will begin counting until it hits
1702 	 * the value in the SSTIMER register, and then trigger an
1703 	 * interrupt. Each time we set the TIMER0_ENABLE bit, the
1704 	 * the timer count is reloaded. Only when the transmitter
1705 	 * is idle will the timer hit 0 and an interrupt fire.
1706 	 */
1707 	CSR_WRITE_1(sc, VGE_CRS1, VGE_CR1_TIMER0_ENABLE);
1708 
1709 	VGE_UNLOCK(sc);
1710 
1711 	/*
1712 	 * Set a timeout in case the chip goes out to lunch.
1713 	 */
1714 	ifp->if_timer = 5;
1715 }
1716 
1717 static int
1718 vge_init(struct ifnet *ifp)
1719 {
1720 	struct vge_softc *sc;
1721 	int i;
1722 
1723 	sc = ifp->if_softc;
1724 
1725 	VGE_LOCK(sc);
1726 
1727 	/*
1728 	 * Cancel pending I/O and free all RX/TX buffers.
1729 	 */
1730 	vge_stop(sc);
1731 	vge_reset(sc);
1732 
1733 	/*
1734 	 * Initialize the RX and TX descriptors and mbufs.
1735 	 */
1736 
1737 	vge_rx_list_init(sc);
1738 	vge_tx_list_init(sc);
1739 
1740 	/* Set our station address */
1741 	for (i = 0; i < ETHER_ADDR_LEN; i++)
1742 		CSR_WRITE_1(sc, VGE_PAR0 + i, sc->vge_eaddr[i]);
1743 
1744 	/*
1745 	 * Set receive FIFO threshold. Also allow transmission and
1746 	 * reception of VLAN tagged frames.
1747 	 */
1748 	CSR_CLRBIT_1(sc, VGE_RXCFG, VGE_RXCFG_FIFO_THR|VGE_RXCFG_VTAGOPT);
1749 	CSR_SETBIT_1(sc, VGE_RXCFG, VGE_RXFIFOTHR_128BYTES|VGE_VTAG_OPT2);
1750 
1751 	/* Set DMA burst length */
1752 	CSR_CLRBIT_1(sc, VGE_DMACFG0, VGE_DMACFG0_BURSTLEN);
1753 	CSR_SETBIT_1(sc, VGE_DMACFG0, VGE_DMABURST_128);
1754 
1755 	CSR_SETBIT_1(sc, VGE_TXCFG, VGE_TXCFG_ARB_PRIO|VGE_TXCFG_NONBLK);
1756 
1757 	/* Set collision backoff algorithm */
1758 	CSR_CLRBIT_1(sc, VGE_CHIPCFG1, VGE_CHIPCFG1_CRANDOM|
1759 	    VGE_CHIPCFG1_CAP|VGE_CHIPCFG1_MBA|VGE_CHIPCFG1_BAKOPT);
1760 	CSR_SETBIT_1(sc, VGE_CHIPCFG1, VGE_CHIPCFG1_OFSET);
1761 
1762 	/* Disable LPSEL field in priority resolution */
1763 	CSR_SETBIT_1(sc, VGE_DIAGCTL, VGE_DIAGCTL_LPSEL_DIS);
1764 
1765 	/*
1766 	 * Load the addresses of the DMA queues into the chip.
1767 	 * Note that we only use one transmit queue.
1768 	 */
1769 
1770 	CSR_WRITE_4(sc, VGE_TXDESC_ADDR_LO0,
1771 	    VGE_ADDR_LO(sc->vge_ldata.vge_tx_list_map->dm_segs[0].ds_addr));
1772 	CSR_WRITE_2(sc, VGE_TXDESCNUM, VGE_TX_DESC_CNT - 1);
1773 
1774 	CSR_WRITE_4(sc, VGE_RXDESC_ADDR_LO,
1775 	    VGE_ADDR_LO(sc->vge_ldata.vge_rx_list_map->dm_segs[0].ds_addr));
1776 	CSR_WRITE_2(sc, VGE_RXDESCNUM, VGE_RX_DESC_CNT - 1);
1777 	CSR_WRITE_2(sc, VGE_RXDESC_RESIDUECNT, VGE_RX_DESC_CNT);
1778 
1779 	/* Enable and wake up the RX descriptor queue */
1780 	CSR_WRITE_1(sc, VGE_RXQCSRS, VGE_RXQCSR_RUN);
1781 	CSR_WRITE_1(sc, VGE_RXQCSRS, VGE_RXQCSR_WAK);
1782 
1783 	/* Enable the TX descriptor queue */
1784 	CSR_WRITE_2(sc, VGE_TXQCSRS, VGE_TXQCSR_RUN0);
1785 
1786 	/* Set up the receive filter -- allow large frames for VLANs. */
1787 	CSR_WRITE_1(sc, VGE_RXCTL, VGE_RXCTL_RX_UCAST|VGE_RXCTL_RX_GIANT);
1788 
1789 	/* If we want promiscuous mode, set the allframes bit. */
1790 	if (ifp->if_flags & IFF_PROMISC) {
1791 		CSR_SETBIT_1(sc, VGE_RXCTL, VGE_RXCTL_RX_PROMISC);
1792 	}
1793 
1794 	/* Set capture broadcast bit to capture broadcast frames. */
1795 	if (ifp->if_flags & IFF_BROADCAST) {
1796 		CSR_SETBIT_1(sc, VGE_RXCTL, VGE_RXCTL_RX_BCAST);
1797 	}
1798 
1799 	/* Set multicast bit to capture multicast frames. */
1800 	if (ifp->if_flags & IFF_MULTICAST) {
1801 		CSR_SETBIT_1(sc, VGE_RXCTL, VGE_RXCTL_RX_MCAST);
1802 	}
1803 
1804 	/* Init the cam filter. */
1805 	vge_cam_clear(sc);
1806 
1807 	/* Init the multicast filter. */
1808 	vge_setmulti(sc);
1809 
1810 	/* Enable flow control */
1811 
1812 	CSR_WRITE_1(sc, VGE_CRS2, 0x8B);
1813 
1814 	/* Enable jumbo frame reception (if desired) */
1815 
1816 	/* Start the MAC. */
1817 	CSR_WRITE_1(sc, VGE_CRC0, VGE_CR0_STOP);
1818 	CSR_WRITE_1(sc, VGE_CRS1, VGE_CR1_NOPOLL);
1819 	CSR_WRITE_1(sc, VGE_CRS0,
1820 	    VGE_CR0_TX_ENABLE|VGE_CR0_RX_ENABLE|VGE_CR0_START);
1821 
1822 	/*
1823 	 * Configure one-shot timer for microsecond
1824 	 * resulution and load it for 500 usecs.
1825 	 */
1826 	CSR_SETBIT_1(sc, VGE_DIAGCTL, VGE_DIAGCTL_TIMER0_RES);
1827 	CSR_WRITE_2(sc, VGE_SSTIMER, 400);
1828 
1829 	/*
1830 	 * Configure interrupt moderation for receive. Enable
1831 	 * the holdoff counter and load it, and set the RX
1832 	 * suppression count to the number of descriptors we
1833 	 * want to allow before triggering an interrupt.
1834 	 * The holdoff timer is in units of 20 usecs.
1835 	 */
1836 
1837 #ifdef notyet
1838 	CSR_WRITE_1(sc, VGE_INTCTL1, VGE_INTCTL_TXINTSUP_DISABLE);
1839 	/* Select the interrupt holdoff timer page. */
1840 	CSR_CLRBIT_1(sc, VGE_CAMCTL, VGE_CAMCTL_PAGESEL);
1841 	CSR_SETBIT_1(sc, VGE_CAMCTL, VGE_PAGESEL_INTHLDOFF);
1842 	CSR_WRITE_1(sc, VGE_INTHOLDOFF, 10); /* ~200 usecs */
1843 
1844 	/* Enable use of the holdoff timer. */
1845 	CSR_WRITE_1(sc, VGE_CRS3, VGE_CR3_INT_HOLDOFF);
1846 	CSR_WRITE_1(sc, VGE_INTCTL1, VGE_INTCTL_SC_RELOAD);
1847 
1848 	/* Select the RX suppression threshold page. */
1849 	CSR_CLRBIT_1(sc, VGE_CAMCTL, VGE_CAMCTL_PAGESEL);
1850 	CSR_SETBIT_1(sc, VGE_CAMCTL, VGE_PAGESEL_RXSUPPTHR);
1851 	CSR_WRITE_1(sc, VGE_RXSUPPTHR, 64); /* interrupt after 64 packets */
1852 
1853 	/* Restore the page select bits. */
1854 	CSR_CLRBIT_1(sc, VGE_CAMCTL, VGE_CAMCTL_PAGESEL);
1855 	CSR_SETBIT_1(sc, VGE_CAMCTL, VGE_PAGESEL_MAR);
1856 #endif
1857 
1858 #ifdef DEVICE_POLLING
1859 	/*
1860 	 * Disable interrupts if we are polling.
1861 	 */
1862 	if (ifp->if_flags & IFF_POLLING) {
1863 		CSR_WRITE_4(sc, VGE_IMR, 0);
1864 		CSR_WRITE_1(sc, VGE_CRC3, VGE_CR3_INT_GMSK);
1865 	} else	/* otherwise ... */
1866 #endif /* DEVICE_POLLING */
1867 	{
1868 	/*
1869 	 * Enable interrupts.
1870 	 */
1871 		CSR_WRITE_4(sc, VGE_IMR, VGE_INTRS);
1872 		CSR_WRITE_4(sc, VGE_ISR, 0);
1873 		CSR_WRITE_1(sc, VGE_CRS3, VGE_CR3_INT_GMSK);
1874 	}
1875 
1876 	mii_mediachg(&sc->sc_mii);
1877 
1878 	ifp->if_flags |= IFF_RUNNING;
1879 	ifp->if_flags &= ~IFF_OACTIVE;
1880 
1881 	sc->vge_if_flags = 0;
1882 	sc->vge_link = 0;
1883 
1884 	VGE_UNLOCK(sc);
1885 
1886 	callout_schedule(&sc->vge_timeout, hz);
1887 
1888 	return 0;
1889 }
1890 
1891 /*
1892  * Set media options.
1893  */
1894 static int
1895 vge_ifmedia_upd(struct ifnet *ifp)
1896 {
1897 	struct vge_softc *sc;
1898 
1899 	sc = ifp->if_softc;
1900 	mii_mediachg(&sc->sc_mii);
1901 
1902 	return 0;
1903 }
1904 
1905 /*
1906  * Report current media status.
1907  */
1908 static void
1909 vge_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
1910 {
1911 	struct vge_softc *sc;
1912 	struct mii_data *mii;
1913 
1914 	sc = ifp->if_softc;
1915 	mii = &sc->sc_mii;
1916 
1917 	mii_pollstat(mii);
1918 	ifmr->ifm_active = mii->mii_media_active;
1919 	ifmr->ifm_status = mii->mii_media_status;
1920 }
1921 
1922 static void
1923 vge_miibus_statchg(struct device *self)
1924 {
1925 	struct vge_softc *sc;
1926 	struct mii_data *mii;
1927 	struct ifmedia_entry *ife;
1928 
1929 	sc = (void *)self;
1930 	mii = &sc->sc_mii;
1931 	ife = mii->mii_media.ifm_cur;
1932 	/*
1933 	 * If the user manually selects a media mode, we need to turn
1934 	 * on the forced MAC mode bit in the DIAGCTL register. If the
1935 	 * user happens to choose a full duplex mode, we also need to
1936 	 * set the 'force full duplex' bit. This applies only to
1937 	 * 10Mbps and 100Mbps speeds. In autoselect mode, forced MAC
1938 	 * mode is disabled, and in 1000baseT mode, full duplex is
1939 	 * always implied, so we turn on the forced mode bit but leave
1940 	 * the FDX bit cleared.
1941 	 */
1942 
1943 	switch (IFM_SUBTYPE(ife->ifm_media)) {
1944 	case IFM_AUTO:
1945 		CSR_CLRBIT_1(sc, VGE_DIAGCTL, VGE_DIAGCTL_MACFORCE);
1946 		CSR_CLRBIT_1(sc, VGE_DIAGCTL, VGE_DIAGCTL_FDXFORCE);
1947 		break;
1948 	case IFM_1000_T:
1949 		CSR_SETBIT_1(sc, VGE_DIAGCTL, VGE_DIAGCTL_MACFORCE);
1950 		CSR_CLRBIT_1(sc, VGE_DIAGCTL, VGE_DIAGCTL_FDXFORCE);
1951 		break;
1952 	case IFM_100_TX:
1953 	case IFM_10_T:
1954 		CSR_SETBIT_1(sc, VGE_DIAGCTL, VGE_DIAGCTL_MACFORCE);
1955 		if ((ife->ifm_media & IFM_GMASK) == IFM_FDX) {
1956 			CSR_SETBIT_1(sc, VGE_DIAGCTL, VGE_DIAGCTL_FDXFORCE);
1957 		} else {
1958 			CSR_CLRBIT_1(sc, VGE_DIAGCTL, VGE_DIAGCTL_FDXFORCE);
1959 		}
1960 		break;
1961 	default:
1962 		printf("%s: unknown media type: %x\n",
1963 		    sc->sc_dev.dv_xname,
1964 		    IFM_SUBTYPE(ife->ifm_media));
1965 		break;
1966 	}
1967 }
1968 
1969 static int
1970 vge_ioctl(struct ifnet *ifp, u_long command, caddr_t data)
1971 {
1972 	struct vge_softc *sc;
1973 	struct ifreq *ifr;
1974 	struct mii_data *mii;
1975 	int s, error;
1976 
1977 	sc = ifp->if_softc;
1978 	ifr = (struct ifreq *)data;
1979 	error = 0;
1980 
1981 	s = splnet();
1982 
1983 	switch (command) {
1984 	case SIOCSIFMTU:
1985 		if (ifr->ifr_mtu > VGE_JUMBO_MTU)
1986 			error = EINVAL;
1987 		ifp->if_mtu = ifr->ifr_mtu;
1988 		break;
1989 	case SIOCSIFFLAGS:
1990 		if (ifp->if_flags & IFF_UP) {
1991 			if (ifp->if_flags & IFF_RUNNING &&
1992 			    ifp->if_flags & IFF_PROMISC &&
1993 			    !(sc->vge_if_flags & IFF_PROMISC)) {
1994 				CSR_SETBIT_1(sc, VGE_RXCTL,
1995 				    VGE_RXCTL_RX_PROMISC);
1996 				vge_setmulti(sc);
1997 			} else if (ifp->if_flags & IFF_RUNNING &&
1998 			    !(ifp->if_flags & IFF_PROMISC) &&
1999 			    sc->vge_if_flags & IFF_PROMISC) {
2000 				CSR_CLRBIT_1(sc, VGE_RXCTL,
2001 				    VGE_RXCTL_RX_PROMISC);
2002 				vge_setmulti(sc);
2003                         } else
2004 				vge_init(ifp);
2005 		} else {
2006 			if (ifp->if_flags & IFF_RUNNING)
2007 				vge_stop(sc);
2008 		}
2009 		sc->vge_if_flags = ifp->if_flags;
2010 		break;
2011 	case SIOCADDMULTI:
2012 	case SIOCDELMULTI:
2013 		error = (command == SIOCADDMULTI) ?
2014 		    ether_addmulti(ifr, &sc->sc_ethercom) :
2015 		    ether_delmulti(ifr, &sc->sc_ethercom);
2016 
2017 		if (error == ENETRESET) {
2018 			/*
2019 			 * Multicast list has changed; set the hardware filter
2020 			 * accordingly.
2021 			 */
2022 			if (ifp->if_flags & IFF_RUNNING)
2023 				vge_setmulti(sc);
2024 			error = 0;
2025 		}
2026 		break;
2027 	case SIOCGIFMEDIA:
2028 	case SIOCSIFMEDIA:
2029 		mii = &sc->sc_mii;
2030 		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, command);
2031 		break;
2032 	default:
2033 		error = ether_ioctl(ifp, command, data);
2034 		break;
2035 	}
2036 
2037 	splx(s);
2038 	return error;
2039 }
2040 
2041 static void
2042 vge_watchdog(struct ifnet *ifp)
2043 {
2044 	struct vge_softc *sc;
2045 
2046 	sc = ifp->if_softc;
2047 	VGE_LOCK(sc);
2048 	printf("%s: watchdog timeout\n", sc->sc_dev.dv_xname);
2049 	ifp->if_oerrors++;
2050 
2051 	vge_txeof(sc);
2052 	vge_rxeof(sc);
2053 
2054 	vge_init(ifp);
2055 
2056 	VGE_UNLOCK(sc);
2057 }
2058 
2059 /*
2060  * Stop the adapter and free any mbufs allocated to the
2061  * RX and TX lists.
2062  */
2063 static void
2064 vge_stop(struct vge_softc *sc)
2065 {
2066 	int i;
2067 	struct ifnet *ifp;
2068 
2069 	ifp = &sc->sc_ethercom.ec_if;
2070 
2071 	VGE_LOCK(sc);
2072 	ifp->if_timer = 0;
2073 
2074 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
2075 #ifdef DEVICE_POLLING
2076 	ether_poll_deregister(ifp);
2077 #endif /* DEVICE_POLLING */
2078 
2079 	CSR_WRITE_1(sc, VGE_CRC3, VGE_CR3_INT_GMSK);
2080 	CSR_WRITE_1(sc, VGE_CRS0, VGE_CR0_STOP);
2081 	CSR_WRITE_4(sc, VGE_ISR, 0xFFFFFFFF);
2082 	CSR_WRITE_2(sc, VGE_TXQCSRC, 0xFFFF);
2083 	CSR_WRITE_1(sc, VGE_RXQCSRC, 0xFF);
2084 	CSR_WRITE_4(sc, VGE_RXDESC_ADDR_LO, 0);
2085 
2086 	if (sc->vge_head != NULL) {
2087 		m_freem(sc->vge_head);
2088 		sc->vge_head = sc->vge_tail = NULL;
2089 	}
2090 
2091 	/* Free the TX list buffers. */
2092 
2093 	for (i = 0; i < VGE_TX_DESC_CNT; i++) {
2094 		if (sc->vge_ldata.vge_tx_mbuf[i] != NULL) {
2095 			bus_dmamap_unload(sc->vge_dmat,
2096 			    sc->vge_ldata.vge_tx_dmamap[i]);
2097 			m_freem(sc->vge_ldata.vge_tx_mbuf[i]);
2098 			sc->vge_ldata.vge_tx_mbuf[i] = NULL;
2099 		}
2100 	}
2101 
2102 	/* Free the RX list buffers. */
2103 
2104 	for (i = 0; i < VGE_RX_DESC_CNT; i++) {
2105 		if (sc->vge_ldata.vge_rx_mbuf[i] != NULL) {
2106 			bus_dmamap_unload(sc->vge_dmat,
2107 			    sc->vge_ldata.vge_rx_dmamap[i]);
2108 			m_freem(sc->vge_ldata.vge_rx_mbuf[i]);
2109 			sc->vge_ldata.vge_rx_mbuf[i] = NULL;
2110 		}
2111 	}
2112 
2113 	VGE_UNLOCK(sc);
2114 }
2115 
2116 #if VGE_POWER_MANAGEMENT
2117 /*
2118  * Device suspend routine.  Stop the interface and save some PCI
2119  * settings in case the BIOS doesn't restore them properly on
2120  * resume.
2121  */
2122 static int
2123 vge_suspend(struct device *dev)
2124 {
2125 	struct vge_softc *sc;
2126 	int i;
2127 
2128 	sc = device_get_softc(dev);
2129 
2130 	vge_stop(sc);
2131 
2132         for (i = 0; i < 5; i++)
2133 		sc->saved_maps[i] = pci_read_config(dev, PCIR_MAPS + i * 4, 4);
2134 	sc->saved_biosaddr = pci_read_config(dev, PCIR_BIOS, 4);
2135 	sc->saved_intline = pci_read_config(dev, PCIR_INTLINE, 1);
2136 	sc->saved_cachelnsz = pci_read_config(dev, PCIR_CACHELNSZ, 1);
2137 	sc->saved_lattimer = pci_read_config(dev, PCIR_LATTIMER, 1);
2138 
2139 	sc->suspended = 1;
2140 
2141 	return 0;
2142 }
2143 
2144 /*
2145  * Device resume routine.  Restore some PCI settings in case the BIOS
2146  * doesn't, re-enable busmastering, and restart the interface if
2147  * appropriate.
2148  */
2149 static int
2150 vge_resume(struct device *dev)
2151 {
2152 	struct vge_softc *sc;
2153 	struct ifnet *ifp;
2154 	int i;
2155 
2156 	sc = (void *)dev;
2157 	ifp = &sc->sc_ethercom.ec_if;
2158 
2159         /* better way to do this? */
2160 	for (i = 0; i < 5; i++)
2161 		pci_write_config(dev, PCIR_MAPS + i * 4, sc->saved_maps[i], 4);
2162 	pci_write_config(dev, PCIR_BIOS, sc->saved_biosaddr, 4);
2163 	pci_write_config(dev, PCIR_INTLINE, sc->saved_intline, 1);
2164 	pci_write_config(dev, PCIR_CACHELNSZ, sc->saved_cachelnsz, 1);
2165 	pci_write_config(dev, PCIR_LATTIMER, sc->saved_lattimer, 1);
2166 
2167 	/* reenable busmastering */
2168 	pci_enable_busmaster(dev);
2169 	pci_enable_io(dev, SYS_RES_MEMORY);
2170 
2171 	/* reinitialize interface if necessary */
2172 	if (ifp->if_flags & IFF_UP)
2173 		vge_init(sc);
2174 
2175 	sc->suspended = 0;
2176 
2177 	return 0;
2178 }
2179 #endif
2180 
2181 /*
2182  * Stop all chip I/O so that the kernel's probe routines don't
2183  * get confused by errant DMAs when rebooting.
2184  */
2185 static void
2186 vge_shutdown(void *arg)
2187 {
2188 	struct vge_softc *sc;
2189 
2190 	sc = arg;
2191 	vge_stop(sc);
2192 }
2193