xref: /netbsd-src/sys/dev/pci/if_vge.c (revision bdc22b2e01993381dcefeff2bc9b56ca75a4235c)
1 /* $NetBSD: if_vge.c,v 1.63 2018/06/26 06:48:01 msaitoh 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 1.63 2018/06/26 06:48:01 msaitoh Exp $");
39 
40 /*
41  * VIA Networking Technologies VT612x PCI gigabit ethernet NIC driver.
42  *
43  * Written by Bill Paul <wpaul@windriver.com>
44  * Senior Networking Software Engineer
45  * Wind River Systems
46  */
47 
48 /*
49  * The VIA Networking VT6122 is a 32bit, 33/66 MHz PCI device that
50  * combines a tri-speed ethernet MAC and PHY, with the following
51  * features:
52  *
53  *	o Jumbo frame support up to 16K
54  *	o Transmit and receive flow control
55  *	o IPv4 checksum offload
56  *	o VLAN tag insertion and stripping
57  *	o TCP large send
58  *	o 64-bit multicast hash table filter
59  *	o 64 entry CAM filter
60  *	o 16K RX FIFO and 48K TX FIFO memory
61  *	o Interrupt moderation
62  *
63  * The VT6122 supports up to four transmit DMA queues. The descriptors
64  * in the transmit ring can address up to 7 data fragments; frames which
65  * span more than 7 data buffers must be coalesced, but in general the
66  * BSD TCP/IP stack rarely generates frames more than 2 or 3 fragments
67  * long. The receive descriptors address only a single buffer.
68  *
69  * There are two peculiar design issues with the VT6122. One is that
70  * receive data buffers must be aligned on a 32-bit boundary. This is
71  * not a problem where the VT6122 is used as a LOM device in x86-based
72  * systems, but on architectures that generate unaligned access traps, we
73  * have to do some copying.
74  *
75  * The other issue has to do with the way 64-bit addresses are handled.
76  * The DMA descriptors only allow you to specify 48 bits of addressing
77  * information. The remaining 16 bits are specified using one of the
78  * I/O registers. If you only have a 32-bit system, then this isn't
79  * an issue, but if you have a 64-bit system and more than 4GB of
80  * memory, you must have to make sure your network data buffers reside
81  * in the same 48-bit 'segment.'
82  *
83  * Special thanks to Ryan Fu at VIA Networking for providing documentation
84  * and sample NICs for testing.
85  */
86 
87 
88 #include <sys/param.h>
89 #include <sys/endian.h>
90 #include <sys/systm.h>
91 #include <sys/device.h>
92 #include <sys/sockio.h>
93 #include <sys/mbuf.h>
94 #include <sys/malloc.h>
95 #include <sys/kernel.h>
96 #include <sys/socket.h>
97 
98 #include <net/if.h>
99 #include <net/if_arp.h>
100 #include <net/if_ether.h>
101 #include <net/if_dl.h>
102 #include <net/if_media.h>
103 
104 #include <net/bpf.h>
105 
106 #include <sys/bus.h>
107 
108 #include <dev/mii/mii.h>
109 #include <dev/mii/miivar.h>
110 
111 #include <dev/pci/pcireg.h>
112 #include <dev/pci/pcivar.h>
113 #include <dev/pci/pcidevs.h>
114 
115 #include <dev/pci/if_vgereg.h>
116 
117 #define VGE_IFQ_MAXLEN		64
118 
119 #define VGE_RING_ALIGN		256
120 
121 #define VGE_NTXDESC		256
122 #define VGE_NTXDESC_MASK	(VGE_NTXDESC - 1)
123 #define VGE_NEXT_TXDESC(x)	((x + 1) & VGE_NTXDESC_MASK)
124 #define VGE_PREV_TXDESC(x)	((x - 1) & VGE_NTXDESC_MASK)
125 
126 #define VGE_NRXDESC		256	/* Must be a multiple of 4!! */
127 #define VGE_NRXDESC_MASK	(VGE_NRXDESC - 1)
128 #define VGE_NEXT_RXDESC(x)	((x + 1) & VGE_NRXDESC_MASK)
129 #define VGE_PREV_RXDESC(x)	((x - 1) & VGE_NRXDESC_MASK)
130 
131 #define VGE_ADDR_LO(y)		((uint64_t)(y) & 0xFFFFFFFF)
132 #define VGE_ADDR_HI(y)		((uint64_t)(y) >> 32)
133 #define VGE_BUFLEN(y)		((y) & 0x7FFF)
134 #define ETHER_PAD_LEN		(ETHER_MIN_LEN - ETHER_CRC_LEN)
135 
136 #define VGE_POWER_MANAGEMENT	0	/* disabled for now */
137 
138 /*
139  * Mbuf adjust factor to force 32-bit alignment of IP header.
140  * Drivers should pad ETHER_ALIGN bytes when setting up a
141  * RX mbuf so the upper layers get the IP header properly aligned
142  * past the 14-byte Ethernet header.
143  *
144  * See also comment in vge_encap().
145  */
146 #define ETHER_ALIGN		2
147 
148 #ifdef __NO_STRICT_ALIGNMENT
149 #define VGE_RX_BUFSIZE		MCLBYTES
150 #else
151 #define VGE_RX_PAD		sizeof(uint32_t)
152 #define VGE_RX_BUFSIZE		(MCLBYTES - VGE_RX_PAD)
153 #endif
154 
155 /*
156  * Control structures are DMA'd to the vge chip. We allocate them in
157  * a single clump that maps to a single DMA segment to make several things
158  * easier.
159  */
160 struct vge_control_data {
161 	/* TX descriptors */
162 	struct vge_txdesc	vcd_txdescs[VGE_NTXDESC];
163 	/* RX descriptors */
164 	struct vge_rxdesc	vcd_rxdescs[VGE_NRXDESC];
165 	/* dummy data for TX padding */
166 	uint8_t			vcd_pad[ETHER_PAD_LEN];
167 };
168 
169 #define VGE_CDOFF(x)	offsetof(struct vge_control_data, x)
170 #define VGE_CDTXOFF(x)	VGE_CDOFF(vcd_txdescs[(x)])
171 #define VGE_CDRXOFF(x)	VGE_CDOFF(vcd_rxdescs[(x)])
172 #define VGE_CDPADOFF()	VGE_CDOFF(vcd_pad[0])
173 
174 /*
175  * Software state for TX jobs.
176  */
177 struct vge_txsoft {
178 	struct mbuf	*txs_mbuf;		/* head of our mbuf chain */
179 	bus_dmamap_t	txs_dmamap;		/* our DMA map */
180 };
181 
182 /*
183  * Software state for RX jobs.
184  */
185 struct vge_rxsoft {
186 	struct mbuf	*rxs_mbuf;		/* head of our mbuf chain */
187 	bus_dmamap_t	rxs_dmamap;		/* our DMA map */
188 };
189 
190 
191 struct vge_softc {
192 	device_t		sc_dev;
193 
194 	bus_space_tag_t		sc_bst;		/* bus space tag */
195 	bus_space_handle_t	sc_bsh;		/* bus space handle */
196 	bus_dma_tag_t		sc_dmat;
197 
198 	struct ethercom		sc_ethercom;	/* interface info */
199 	uint8_t			sc_eaddr[ETHER_ADDR_LEN];
200 
201 	void			*sc_intrhand;
202 	struct mii_data		sc_mii;
203 	uint8_t			sc_type;
204 	int			sc_if_flags;
205 	int			sc_link;
206 	int			sc_camidx;
207 	callout_t		sc_timeout;
208 
209 	bus_dmamap_t		sc_cddmamap;
210 #define sc_cddma		sc_cddmamap->dm_segs[0].ds_addr
211 
212 	struct vge_txsoft	sc_txsoft[VGE_NTXDESC];
213 	struct vge_rxsoft	sc_rxsoft[VGE_NRXDESC];
214 	struct vge_control_data	*sc_control_data;
215 #define sc_txdescs		sc_control_data->vcd_txdescs
216 #define sc_rxdescs		sc_control_data->vcd_rxdescs
217 
218 	int			sc_tx_prodidx;
219 	int			sc_tx_considx;
220 	int			sc_tx_free;
221 
222 	struct mbuf		*sc_rx_mhead;
223 	struct mbuf		*sc_rx_mtail;
224 	int			sc_rx_prodidx;
225 	int			sc_rx_consumed;
226 
227 	int			sc_suspended;	/* 0 = normal  1 = suspended */
228 	uint32_t		sc_saved_maps[5];	/* pci data */
229 	uint32_t		sc_saved_biosaddr;
230 	uint8_t			sc_saved_intline;
231 	uint8_t			sc_saved_cachelnsz;
232 	uint8_t			sc_saved_lattimer;
233 };
234 
235 #define VGE_CDTXADDR(sc, x)	((sc)->sc_cddma + VGE_CDTXOFF(x))
236 #define VGE_CDRXADDR(sc, x)	((sc)->sc_cddma + VGE_CDRXOFF(x))
237 #define VGE_CDPADADDR(sc)	((sc)->sc_cddma + VGE_CDPADOFF())
238 
239 #define VGE_TXDESCSYNC(sc, idx, ops)					\
240 	bus_dmamap_sync((sc)->sc_dmat,(sc)->sc_cddmamap,		\
241 	    VGE_CDTXOFF(idx),						\
242 	    offsetof(struct vge_txdesc, td_frag[0]),			\
243 	    (ops))
244 #define VGE_TXFRAGSYNC(sc, idx, nsegs, ops)				\
245 	bus_dmamap_sync((sc)->sc_dmat, (sc)->sc_cddmamap,		\
246 	    VGE_CDTXOFF(idx) +						\
247 	    offsetof(struct vge_txdesc, td_frag[0]),			\
248 	    sizeof(struct vge_txfrag) * (nsegs),			\
249 	    (ops))
250 #define VGE_RXDESCSYNC(sc, idx, ops)					\
251 	bus_dmamap_sync((sc)->sc_dmat, (sc)->sc_cddmamap,		\
252 	    VGE_CDRXOFF(idx),						\
253 	    sizeof(struct vge_rxdesc),					\
254 	    (ops))
255 
256 /*
257  * register space access macros
258  */
259 #define CSR_WRITE_4(sc, reg, val)	\
260 	bus_space_write_4((sc)->sc_bst, (sc)->sc_bsh, (reg), (val))
261 #define CSR_WRITE_2(sc, reg, val)	\
262 	bus_space_write_2((sc)->sc_bst, (sc)->sc_bsh, (reg), (val))
263 #define CSR_WRITE_1(sc, reg, val)	\
264 	bus_space_write_1((sc)->sc_bst, (sc)->sc_bsh, (reg), (val))
265 
266 #define CSR_READ_4(sc, reg)		\
267 	bus_space_read_4((sc)->sc_bst, (sc)->sc_bsh, (reg))
268 #define CSR_READ_2(sc, reg)		\
269 	bus_space_read_2((sc)->sc_bst, (sc)->sc_bsh, (reg))
270 #define CSR_READ_1(sc, reg)		\
271 	bus_space_read_1((sc)->sc_bst, (sc)->sc_bsh, (reg))
272 
273 #define CSR_SETBIT_1(sc, reg, x)	\
274 	CSR_WRITE_1((sc), (reg), CSR_READ_1((sc), (reg)) | (x))
275 #define CSR_SETBIT_2(sc, reg, x)	\
276 	CSR_WRITE_2((sc), (reg), CSR_READ_2((sc), (reg)) | (x))
277 #define CSR_SETBIT_4(sc, reg, x)	\
278 	CSR_WRITE_4((sc), (reg), CSR_READ_4((sc), (reg)) | (x))
279 
280 #define CSR_CLRBIT_1(sc, reg, x)	\
281 	CSR_WRITE_1((sc), (reg), CSR_READ_1((sc), (reg)) & ~(x))
282 #define CSR_CLRBIT_2(sc, reg, x)	\
283 	CSR_WRITE_2((sc), (reg), CSR_READ_2((sc), (reg)) & ~(x))
284 #define CSR_CLRBIT_4(sc, reg, x)	\
285 	CSR_WRITE_4((sc), (reg), CSR_READ_4((sc), (reg)) & ~(x))
286 
287 #define VGE_TIMEOUT		10000
288 
289 #define VGE_PCI_LOIO             0x10
290 #define VGE_PCI_LOMEM            0x14
291 
292 static inline void vge_set_txaddr(struct vge_txfrag *, bus_addr_t);
293 static inline void vge_set_rxaddr(struct vge_rxdesc *, bus_addr_t);
294 
295 static int vge_ifflags_cb(struct ethercom *);
296 
297 static int vge_match(device_t, cfdata_t, void *);
298 static void vge_attach(device_t, device_t, void *);
299 
300 static int vge_encap(struct vge_softc *, struct mbuf *, int);
301 
302 static int vge_allocmem(struct vge_softc *);
303 static int vge_newbuf(struct vge_softc *, int, struct mbuf *);
304 #ifndef __NO_STRICT_ALIGNMENT
305 static inline void vge_fixup_rx(struct mbuf *);
306 #endif
307 static void vge_rxeof(struct vge_softc *);
308 static void vge_txeof(struct vge_softc *);
309 static int vge_intr(void *);
310 static void vge_tick(void *);
311 static void vge_start(struct ifnet *);
312 static int vge_ioctl(struct ifnet *, u_long, void *);
313 static int vge_init(struct ifnet *);
314 static void vge_stop(struct ifnet *, int);
315 static void vge_watchdog(struct ifnet *);
316 #if VGE_POWER_MANAGEMENT
317 static int vge_suspend(device_t);
318 static int vge_resume(device_t);
319 #endif
320 static bool vge_shutdown(device_t, int);
321 
322 static uint16_t vge_read_eeprom(struct vge_softc *, int);
323 
324 static void vge_miipoll_start(struct vge_softc *);
325 static void vge_miipoll_stop(struct vge_softc *);
326 static int vge_miibus_readreg(device_t, int, int);
327 static void vge_miibus_writereg(device_t, int, int, int);
328 static void vge_miibus_statchg(struct ifnet *);
329 
330 static void vge_cam_clear(struct vge_softc *);
331 static int vge_cam_set(struct vge_softc *, uint8_t *);
332 static void vge_setmulti(struct vge_softc *);
333 static void vge_reset(struct vge_softc *);
334 
335 CFATTACH_DECL_NEW(vge, sizeof(struct vge_softc),
336     vge_match, vge_attach, NULL, NULL);
337 
338 static inline void
339 vge_set_txaddr(struct vge_txfrag *f, bus_addr_t daddr)
340 {
341 
342 	f->tf_addrlo = htole32((uint32_t)daddr);
343 	if (sizeof(bus_addr_t) == sizeof(uint64_t))
344 		f->tf_addrhi = htole16(((uint64_t)daddr >> 32) & 0xFFFF);
345 	else
346 		f->tf_addrhi = 0;
347 }
348 
349 static inline void
350 vge_set_rxaddr(struct vge_rxdesc *rxd, bus_addr_t daddr)
351 {
352 
353 	rxd->rd_addrlo = htole32((uint32_t)daddr);
354 	if (sizeof(bus_addr_t) == sizeof(uint64_t))
355 		rxd->rd_addrhi = htole16(((uint64_t)daddr >> 32) & 0xFFFF);
356 	else
357 		rxd->rd_addrhi = 0;
358 }
359 
360 /*
361  * Read a word of data stored in the EEPROM at address 'addr.'
362  */
363 static uint16_t
364 vge_read_eeprom(struct vge_softc *sc, int addr)
365 {
366 	int i;
367 	uint16_t word = 0;
368 
369 	/*
370 	 * Enter EEPROM embedded programming mode. In order to
371 	 * access the EEPROM at all, we first have to set the
372 	 * EELOAD bit in the CHIPCFG2 register.
373 	 */
374 	CSR_SETBIT_1(sc, VGE_CHIPCFG2, VGE_CHIPCFG2_EELOAD);
375 	CSR_SETBIT_1(sc, VGE_EECSR, VGE_EECSR_EMBP/*|VGE_EECSR_ECS*/);
376 
377 	/* Select the address of the word we want to read */
378 	CSR_WRITE_1(sc, VGE_EEADDR, addr);
379 
380 	/* Issue read command */
381 	CSR_SETBIT_1(sc, VGE_EECMD, VGE_EECMD_ERD);
382 
383 	/* Wait for the done bit to be set. */
384 	for (i = 0; i < VGE_TIMEOUT; i++) {
385 		if (CSR_READ_1(sc, VGE_EECMD) & VGE_EECMD_EDONE)
386 			break;
387 	}
388 
389 	if (i == VGE_TIMEOUT) {
390 		printf("%s: EEPROM read timed out\n", device_xname(sc->sc_dev));
391 		return 0;
392 	}
393 
394 	/* Read the result */
395 	word = CSR_READ_2(sc, VGE_EERDDAT);
396 
397 	/* Turn off EEPROM access mode. */
398 	CSR_CLRBIT_1(sc, VGE_EECSR, VGE_EECSR_EMBP/*|VGE_EECSR_ECS*/);
399 	CSR_CLRBIT_1(sc, VGE_CHIPCFG2, VGE_CHIPCFG2_EELOAD);
400 
401 	return word;
402 }
403 
404 static void
405 vge_miipoll_stop(struct vge_softc *sc)
406 {
407 	int i;
408 
409 	CSR_WRITE_1(sc, VGE_MIICMD, 0);
410 
411 	for (i = 0; i < VGE_TIMEOUT; i++) {
412 		DELAY(1);
413 		if (CSR_READ_1(sc, VGE_MIISTS) & VGE_MIISTS_IIDL)
414 			break;
415 	}
416 
417 	if (i == VGE_TIMEOUT) {
418 		printf("%s: failed to idle MII autopoll\n",
419 		    device_xname(sc->sc_dev));
420 	}
421 }
422 
423 static void
424 vge_miipoll_start(struct vge_softc *sc)
425 {
426 	int i;
427 
428 	/* First, make sure we're idle. */
429 
430 	CSR_WRITE_1(sc, VGE_MIICMD, 0);
431 	CSR_WRITE_1(sc, VGE_MIIADDR, VGE_MIIADDR_SWMPL);
432 
433 	for (i = 0; i < VGE_TIMEOUT; i++) {
434 		DELAY(1);
435 		if (CSR_READ_1(sc, VGE_MIISTS) & VGE_MIISTS_IIDL)
436 			break;
437 	}
438 
439 	if (i == VGE_TIMEOUT) {
440 		printf("%s: failed to idle MII autopoll\n",
441 		    device_xname(sc->sc_dev));
442 		return;
443 	}
444 
445 	/* Now enable auto poll mode. */
446 
447 	CSR_WRITE_1(sc, VGE_MIICMD, VGE_MIICMD_MAUTO);
448 
449 	/* And make sure it started. */
450 
451 	for (i = 0; i < VGE_TIMEOUT; i++) {
452 		DELAY(1);
453 		if ((CSR_READ_1(sc, VGE_MIISTS) & VGE_MIISTS_IIDL) == 0)
454 			break;
455 	}
456 
457 	if (i == VGE_TIMEOUT) {
458 		printf("%s: failed to start MII autopoll\n",
459 		    device_xname(sc->sc_dev));
460 	}
461 }
462 
463 static int
464 vge_miibus_readreg(device_t dev, int phy, int reg)
465 {
466 	struct vge_softc *sc;
467 	int i, s;
468 	uint16_t rval;
469 
470 	sc = device_private(dev);
471 	rval = 0;
472 	if (phy != (CSR_READ_1(sc, VGE_MIICFG) & 0x1F))
473 		return 0;
474 
475 	s = splnet();
476 	vge_miipoll_stop(sc);
477 
478 	/* Specify the register we want to read. */
479 	CSR_WRITE_1(sc, VGE_MIIADDR, reg);
480 
481 	/* Issue read command. */
482 	CSR_SETBIT_1(sc, VGE_MIICMD, VGE_MIICMD_RCMD);
483 
484 	/* Wait for the read command bit to self-clear. */
485 	for (i = 0; i < VGE_TIMEOUT; i++) {
486 		DELAY(1);
487 		if ((CSR_READ_1(sc, VGE_MIICMD) & VGE_MIICMD_RCMD) == 0)
488 			break;
489 	}
490 
491 	if (i == VGE_TIMEOUT)
492 		printf("%s: MII read timed out\n", device_xname(sc->sc_dev));
493 	else
494 		rval = CSR_READ_2(sc, VGE_MIIDATA);
495 
496 	vge_miipoll_start(sc);
497 	splx(s);
498 
499 	return rval;
500 }
501 
502 static void
503 vge_miibus_writereg(device_t dev, int phy, int reg, int data)
504 {
505 	struct vge_softc *sc;
506 	int i, s;
507 
508 	sc = device_private(dev);
509 	if (phy != (CSR_READ_1(sc, VGE_MIICFG) & 0x1F))
510 		return;
511 
512 	s = splnet();
513 	vge_miipoll_stop(sc);
514 
515 	/* Specify the register we want to write. */
516 	CSR_WRITE_1(sc, VGE_MIIADDR, reg);
517 
518 	/* Specify the data we want to write. */
519 	CSR_WRITE_2(sc, VGE_MIIDATA, data);
520 
521 	/* Issue write command. */
522 	CSR_SETBIT_1(sc, VGE_MIICMD, VGE_MIICMD_WCMD);
523 
524 	/* Wait for the write command bit to self-clear. */
525 	for (i = 0; i < VGE_TIMEOUT; i++) {
526 		DELAY(1);
527 		if ((CSR_READ_1(sc, VGE_MIICMD) & VGE_MIICMD_WCMD) == 0)
528 			break;
529 	}
530 
531 	if (i == VGE_TIMEOUT) {
532 		printf("%s: MII write timed out\n", device_xname(sc->sc_dev));
533 	}
534 
535 	vge_miipoll_start(sc);
536 	splx(s);
537 }
538 
539 static void
540 vge_cam_clear(struct vge_softc *sc)
541 {
542 	int i;
543 
544 	/*
545 	 * Turn off all the mask bits. This tells the chip
546 	 * that none of the entries in the CAM filter are valid.
547 	 * desired entries will be enabled as we fill the filter in.
548 	 */
549 
550 	CSR_CLRBIT_1(sc, VGE_CAMCTL, VGE_CAMCTL_PAGESEL);
551 	CSR_SETBIT_1(sc, VGE_CAMCTL, VGE_PAGESEL_CAMMASK);
552 	CSR_WRITE_1(sc, VGE_CAMADDR, VGE_CAMADDR_ENABLE);
553 	for (i = 0; i < 8; i++)
554 		CSR_WRITE_1(sc, VGE_CAM0 + i, 0);
555 
556 	/* Clear the VLAN filter too. */
557 
558 	CSR_WRITE_1(sc, VGE_CAMADDR, VGE_CAMADDR_ENABLE|VGE_CAMADDR_AVSEL|0);
559 	for (i = 0; i < 8; i++)
560 		CSR_WRITE_1(sc, VGE_CAM0 + i, 0);
561 
562 	CSR_WRITE_1(sc, VGE_CAMADDR, 0);
563 	CSR_CLRBIT_1(sc, VGE_CAMCTL, VGE_CAMCTL_PAGESEL);
564 	CSR_SETBIT_1(sc, VGE_CAMCTL, VGE_PAGESEL_MAR);
565 
566 	sc->sc_camidx = 0;
567 }
568 
569 static int
570 vge_cam_set(struct vge_softc *sc, uint8_t *addr)
571 {
572 	int i, error;
573 
574 	error = 0;
575 
576 	if (sc->sc_camidx == VGE_CAM_MAXADDRS)
577 		return ENOSPC;
578 
579 	/* Select the CAM data page. */
580 	CSR_CLRBIT_1(sc, VGE_CAMCTL, VGE_CAMCTL_PAGESEL);
581 	CSR_SETBIT_1(sc, VGE_CAMCTL, VGE_PAGESEL_CAMDATA);
582 
583 	/* Set the filter entry we want to update and enable writing. */
584 	CSR_WRITE_1(sc, VGE_CAMADDR, VGE_CAMADDR_ENABLE | sc->sc_camidx);
585 
586 	/* Write the address to the CAM registers */
587 	for (i = 0; i < ETHER_ADDR_LEN; i++)
588 		CSR_WRITE_1(sc, VGE_CAM0 + i, addr[i]);
589 
590 	/* Issue a write command. */
591 	CSR_SETBIT_1(sc, VGE_CAMCTL, VGE_CAMCTL_WRITE);
592 
593 	/* Wake for it to clear. */
594 	for (i = 0; i < VGE_TIMEOUT; i++) {
595 		DELAY(1);
596 		if ((CSR_READ_1(sc, VGE_CAMCTL) & VGE_CAMCTL_WRITE) == 0)
597 			break;
598 	}
599 
600 	if (i == VGE_TIMEOUT) {
601 		printf("%s: setting CAM filter failed\n",
602 		    device_xname(sc->sc_dev));
603 		error = EIO;
604 		goto fail;
605 	}
606 
607 	/* Select the CAM mask page. */
608 	CSR_CLRBIT_1(sc, VGE_CAMCTL, VGE_CAMCTL_PAGESEL);
609 	CSR_SETBIT_1(sc, VGE_CAMCTL, VGE_PAGESEL_CAMMASK);
610 
611 	/* Set the mask bit that enables this filter. */
612 	CSR_SETBIT_1(sc, VGE_CAM0 + (sc->sc_camidx / 8),
613 	    1 << (sc->sc_camidx & 7));
614 
615 	sc->sc_camidx++;
616 
617  fail:
618 	/* Turn off access to CAM. */
619 	CSR_WRITE_1(sc, VGE_CAMADDR, 0);
620 	CSR_CLRBIT_1(sc, VGE_CAMCTL, VGE_CAMCTL_PAGESEL);
621 	CSR_SETBIT_1(sc, VGE_CAMCTL, VGE_PAGESEL_MAR);
622 
623 	return error;
624 }
625 
626 /*
627  * Program the multicast filter. We use the 64-entry CAM filter
628  * for perfect filtering. If there's more than 64 multicast addresses,
629  * we use the hash filter instead.
630  */
631 static void
632 vge_setmulti(struct vge_softc *sc)
633 {
634 	struct ifnet *ifp;
635 	int error;
636 	uint32_t h, hashes[2] = { 0, 0 };
637 	struct ether_multi *enm;
638 	struct ether_multistep step;
639 
640 	error = 0;
641 	ifp = &sc->sc_ethercom.ec_if;
642 
643 	/* First, zot all the multicast entries. */
644 	vge_cam_clear(sc);
645 	CSR_WRITE_4(sc, VGE_MAR0, 0);
646 	CSR_WRITE_4(sc, VGE_MAR1, 0);
647 	ifp->if_flags &= ~IFF_ALLMULTI;
648 
649 	/*
650 	 * If the user wants allmulti or promisc mode, enable reception
651 	 * of all multicast frames.
652 	 */
653 	if (ifp->if_flags & IFF_PROMISC) {
654  allmulti:
655 		CSR_WRITE_4(sc, VGE_MAR0, 0xFFFFFFFF);
656 		CSR_WRITE_4(sc, VGE_MAR1, 0xFFFFFFFF);
657 		ifp->if_flags |= IFF_ALLMULTI;
658 		return;
659 	}
660 
661 	/* Now program new ones */
662 	ETHER_FIRST_MULTI(step, &sc->sc_ethercom, enm);
663 	while (enm != NULL) {
664 		/*
665 		 * If multicast range, fall back to ALLMULTI.
666 		 */
667 		if (memcmp(enm->enm_addrlo, enm->enm_addrhi,
668 				ETHER_ADDR_LEN) != 0)
669 			goto allmulti;
670 
671 		error = vge_cam_set(sc, enm->enm_addrlo);
672 		if (error)
673 			break;
674 
675 		ETHER_NEXT_MULTI(step, enm);
676 	}
677 
678 	/* If there were too many addresses, use the hash filter. */
679 	if (error) {
680 		vge_cam_clear(sc);
681 
682 		ETHER_FIRST_MULTI(step, &sc->sc_ethercom, enm);
683 		while (enm != NULL) {
684 			/*
685 			 * If multicast range, fall back to ALLMULTI.
686 			 */
687 			if (memcmp(enm->enm_addrlo, enm->enm_addrhi,
688 					ETHER_ADDR_LEN) != 0)
689 				goto allmulti;
690 
691 			h = ether_crc32_be(enm->enm_addrlo,
692 			    ETHER_ADDR_LEN) >> 26;
693 			hashes[h >> 5] |= 1 << (h & 0x1f);
694 
695 			ETHER_NEXT_MULTI(step, enm);
696 		}
697 
698 		CSR_WRITE_4(sc, VGE_MAR0, hashes[0]);
699 		CSR_WRITE_4(sc, VGE_MAR1, hashes[1]);
700 	}
701 }
702 
703 static void
704 vge_reset(struct vge_softc *sc)
705 {
706 	int i;
707 
708 	CSR_WRITE_1(sc, VGE_CRS1, VGE_CR1_SOFTRESET);
709 
710 	for (i = 0; i < VGE_TIMEOUT; i++) {
711 		DELAY(5);
712 		if ((CSR_READ_1(sc, VGE_CRS1) & VGE_CR1_SOFTRESET) == 0)
713 			break;
714 	}
715 
716 	if (i == VGE_TIMEOUT) {
717 		printf("%s: soft reset timed out", device_xname(sc->sc_dev));
718 		CSR_WRITE_1(sc, VGE_CRS3, VGE_CR3_STOP_FORCE);
719 		DELAY(2000);
720 	}
721 
722 	DELAY(5000);
723 
724 	CSR_SETBIT_1(sc, VGE_EECSR, VGE_EECSR_RELOAD);
725 
726 	for (i = 0; i < VGE_TIMEOUT; i++) {
727 		DELAY(5);
728 		if ((CSR_READ_1(sc, VGE_EECSR) & VGE_EECSR_RELOAD) == 0)
729 			break;
730 	}
731 
732 	if (i == VGE_TIMEOUT) {
733 		printf("%s: EEPROM reload timed out\n",
734 		    device_xname(sc->sc_dev));
735 		return;
736 	}
737 
738 	/*
739 	 * On some machine, the first read data from EEPROM could be
740 	 * messed up, so read one dummy data here to avoid the mess.
741 	 */
742 	(void)vge_read_eeprom(sc, 0);
743 
744 	CSR_CLRBIT_1(sc, VGE_CHIPCFG0, VGE_CHIPCFG0_PACPI);
745 }
746 
747 /*
748  * Probe for a VIA gigabit chip. Check the PCI vendor and device
749  * IDs against our list and return a device name if we find a match.
750  */
751 static int
752 vge_match(device_t parent, cfdata_t match, void *aux)
753 {
754 	struct pci_attach_args *pa = aux;
755 
756 	if (PCI_VENDOR(pa->pa_id) == PCI_VENDOR_VIATECH
757 	    && PCI_PRODUCT(pa->pa_id) == PCI_PRODUCT_VIATECH_VT612X)
758 		return 1;
759 
760 	return 0;
761 }
762 
763 static int
764 vge_allocmem(struct vge_softc *sc)
765 {
766 	int error;
767 	int nseg;
768 	int i;
769 	bus_dma_segment_t seg;
770 
771 	/*
772 	 * Allocate memory for control data.
773 	 */
774 
775 	error = bus_dmamem_alloc(sc->sc_dmat, sizeof(struct vge_control_data),
776 	     VGE_RING_ALIGN, 0, &seg, 1, &nseg, BUS_DMA_NOWAIT);
777 	if (error) {
778 		aprint_error_dev(sc->sc_dev,
779 		    "could not allocate control data dma memory\n");
780 		goto fail_1;
781 	}
782 
783 	/* Map the memory to kernel VA space */
784 
785 	error = bus_dmamem_map(sc->sc_dmat, &seg, nseg,
786 	    sizeof(struct vge_control_data), (void **)&sc->sc_control_data,
787 	    BUS_DMA_NOWAIT);
788 	if (error) {
789 		aprint_error_dev(sc->sc_dev,
790 		    "could not map control data dma memory\n");
791 		goto fail_2;
792 	}
793 	memset(sc->sc_control_data, 0, sizeof(struct vge_control_data));
794 
795 	/*
796 	 * Create map for control data.
797 	 */
798 	error = bus_dmamap_create(sc->sc_dmat,
799 	    sizeof(struct vge_control_data), 1,
800 	    sizeof(struct vge_control_data), 0, BUS_DMA_NOWAIT,
801 	    &sc->sc_cddmamap);
802 	if (error) {
803 		aprint_error_dev(sc->sc_dev,
804 		    "could not create control data dmamap\n");
805 		goto fail_3;
806 	}
807 
808 	/* Load the map for the control data. */
809 	error = bus_dmamap_load(sc->sc_dmat, sc->sc_cddmamap,
810 	    sc->sc_control_data, sizeof(struct vge_control_data), NULL,
811 	    BUS_DMA_NOWAIT);
812 	if (error) {
813 		aprint_error_dev(sc->sc_dev,
814 		    "could not load control data dma memory\n");
815 		goto fail_4;
816 	}
817 
818 	/* Create DMA maps for TX buffers */
819 
820 	for (i = 0; i < VGE_NTXDESC; i++) {
821 		error = bus_dmamap_create(sc->sc_dmat, VGE_TX_MAXLEN,
822 		    VGE_TX_FRAGS, VGE_TX_MAXLEN, 0, BUS_DMA_NOWAIT,
823 		    &sc->sc_txsoft[i].txs_dmamap);
824 		if (error) {
825 			aprint_error_dev(sc->sc_dev,
826 			    "can't create DMA map for TX descs\n");
827 			goto fail_5;
828 		}
829 	}
830 
831 	/* Create DMA maps for RX buffers */
832 
833 	for (i = 0; i < VGE_NRXDESC; i++) {
834 		error = bus_dmamap_create(sc->sc_dmat, MCLBYTES,
835 		    1, MCLBYTES, 0, BUS_DMA_NOWAIT,
836 		    &sc->sc_rxsoft[i].rxs_dmamap);
837 		if (error) {
838 			aprint_error_dev(sc->sc_dev,
839 			    "can't create DMA map for RX descs\n");
840 			goto fail_6;
841 		}
842 		sc->sc_rxsoft[i].rxs_mbuf = NULL;
843 	}
844 
845 	return 0;
846 
847  fail_6:
848 	for (i = 0; i < VGE_NRXDESC; i++) {
849 		if (sc->sc_rxsoft[i].rxs_dmamap != NULL)
850 			bus_dmamap_destroy(sc->sc_dmat,
851 			    sc->sc_rxsoft[i].rxs_dmamap);
852 	}
853  fail_5:
854 	for (i = 0; i < VGE_NTXDESC; i++) {
855 		if (sc->sc_txsoft[i].txs_dmamap != NULL)
856 			bus_dmamap_destroy(sc->sc_dmat,
857 			    sc->sc_txsoft[i].txs_dmamap);
858 	}
859 	bus_dmamap_unload(sc->sc_dmat, sc->sc_cddmamap);
860  fail_4:
861 	bus_dmamap_destroy(sc->sc_dmat, sc->sc_cddmamap);
862  fail_3:
863 	bus_dmamem_unmap(sc->sc_dmat, (void *)sc->sc_control_data,
864 	    sizeof(struct vge_control_data));
865  fail_2:
866 	bus_dmamem_free(sc->sc_dmat, &seg, nseg);
867  fail_1:
868 	return ENOMEM;
869 }
870 
871 /*
872  * Attach the interface. Allocate softc structures, do ifmedia
873  * setup and ethernet/BPF attach.
874  */
875 static void
876 vge_attach(device_t parent, device_t self, void *aux)
877 {
878 	uint8_t	*eaddr;
879 	struct vge_softc *sc = device_private(self);
880 	struct ifnet *ifp;
881 	struct pci_attach_args *pa = aux;
882 	pci_chipset_tag_t pc = pa->pa_pc;
883 	const char *intrstr;
884 	pci_intr_handle_t ih;
885 	uint16_t val;
886 	char intrbuf[PCI_INTRSTR_LEN];
887 
888 	sc->sc_dev = self;
889 
890 	pci_aprint_devinfo_fancy(pa, NULL, "VIA VT612X Gigabit Ethernet", 1);
891 
892 	/* Make sure bus-mastering is enabled */
893         pci_conf_write(pc, pa->pa_tag, PCI_COMMAND_STATUS_REG,
894 	    pci_conf_read(pc, pa->pa_tag, PCI_COMMAND_STATUS_REG) |
895 	    PCI_COMMAND_MASTER_ENABLE);
896 
897 	/*
898 	 * Map control/status registers.
899 	 */
900 	if (pci_mapreg_map(pa, VGE_PCI_LOMEM, PCI_MAPREG_TYPE_MEM, 0,
901 	    &sc->sc_bst, &sc->sc_bsh, NULL, NULL) != 0) {
902 		aprint_error_dev(self, "couldn't map memory\n");
903 		return;
904 	}
905 
906         /*
907          * Map and establish our interrupt.
908          */
909 	if (pci_intr_map(pa, &ih)) {
910 		aprint_error_dev(self, "unable to map interrupt\n");
911 		return;
912 	}
913 	intrstr = pci_intr_string(pc, ih, intrbuf, sizeof(intrbuf));
914 	sc->sc_intrhand = pci_intr_establish(pc, ih, IPL_NET, vge_intr, sc);
915 	if (sc->sc_intrhand == NULL) {
916 		aprint_error_dev(self, "unable to establish interrupt");
917 		if (intrstr != NULL)
918 			aprint_error(" at %s", intrstr);
919 		aprint_error("\n");
920 		return;
921 	}
922 	aprint_normal_dev(self, "interrupting at %s\n", intrstr);
923 
924 	/* Reset the adapter. */
925 	vge_reset(sc);
926 
927 	/*
928 	 * Get station address from the EEPROM.
929 	 */
930 	eaddr = sc->sc_eaddr;
931 	val = vge_read_eeprom(sc, VGE_EE_EADDR + 0);
932 	eaddr[0] = val & 0xff;
933 	eaddr[1] = val >> 8;
934 	val = vge_read_eeprom(sc, VGE_EE_EADDR + 1);
935 	eaddr[2] = val & 0xff;
936 	eaddr[3] = val >> 8;
937 	val = vge_read_eeprom(sc, VGE_EE_EADDR + 2);
938 	eaddr[4] = val & 0xff;
939 	eaddr[5] = val >> 8;
940 
941 	aprint_normal_dev(self, "Ethernet address: %s\n",
942 	    ether_sprintf(eaddr));
943 
944 	/*
945 	 * Use the 32bit tag. Hardware supports 48bit physical addresses,
946 	 * but we don't use that for now.
947 	 */
948 	sc->sc_dmat = pa->pa_dmat;
949 
950 	if (vge_allocmem(sc) != 0)
951 		return;
952 
953 	ifp = &sc->sc_ethercom.ec_if;
954 	ifp->if_softc = sc;
955 	strlcpy(ifp->if_xname, device_xname(self), IFNAMSIZ);
956 	ifp->if_mtu = ETHERMTU;
957 	ifp->if_baudrate = IF_Gbps(1);
958 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
959 	ifp->if_ioctl = vge_ioctl;
960 	ifp->if_start = vge_start;
961 	ifp->if_init = vge_init;
962 	ifp->if_stop = vge_stop;
963 
964 	/*
965 	 * We can support 802.1Q VLAN-sized frames and jumbo
966 	 * Ethernet frames.
967 	 */
968 	sc->sc_ethercom.ec_capabilities |=
969 	    ETHERCAP_VLAN_MTU | ETHERCAP_JUMBO_MTU |
970 	    ETHERCAP_VLAN_HWTAGGING;
971 
972 	/*
973 	 * We can do IPv4/TCPv4/UDPv4 checksums in hardware.
974 	 */
975 	ifp->if_capabilities |=
976 	    IFCAP_CSUM_IPv4_Tx | IFCAP_CSUM_IPv4_Rx |
977 	    IFCAP_CSUM_TCPv4_Tx | IFCAP_CSUM_TCPv4_Rx |
978 	    IFCAP_CSUM_UDPv4_Tx | IFCAP_CSUM_UDPv4_Rx;
979 
980 #ifdef DEVICE_POLLING
981 #ifdef IFCAP_POLLING
982 	ifp->if_capabilities |= IFCAP_POLLING;
983 #endif
984 #endif
985 	ifp->if_watchdog = vge_watchdog;
986 	IFQ_SET_MAXLEN(&ifp->if_snd, max(VGE_IFQ_MAXLEN, IFQ_MAXLEN));
987 	IFQ_SET_READY(&ifp->if_snd);
988 
989 	/*
990 	 * Initialize our media structures and probe the MII.
991 	 */
992 	sc->sc_mii.mii_ifp = ifp;
993 	sc->sc_mii.mii_readreg = vge_miibus_readreg;
994 	sc->sc_mii.mii_writereg = vge_miibus_writereg;
995 	sc->sc_mii.mii_statchg = vge_miibus_statchg;
996 
997 	sc->sc_ethercom.ec_mii = &sc->sc_mii;
998 	ifmedia_init(&sc->sc_mii.mii_media, 0, ether_mediachange,
999 	    ether_mediastatus);
1000 	mii_attach(self, &sc->sc_mii, 0xffffffff, MII_PHY_ANY,
1001 	    MII_OFFSET_ANY, MIIF_DOPAUSE);
1002 	if (LIST_FIRST(&sc->sc_mii.mii_phys) == NULL) {
1003 		ifmedia_add(&sc->sc_mii.mii_media, IFM_ETHER|IFM_NONE, 0, NULL);
1004 		ifmedia_set(&sc->sc_mii.mii_media, IFM_ETHER|IFM_NONE);
1005 	} else
1006 		ifmedia_set(&sc->sc_mii.mii_media, IFM_ETHER|IFM_AUTO);
1007 
1008 	/*
1009 	 * Attach the interface.
1010 	 */
1011 	if_attach(ifp);
1012 	if_deferred_start_init(ifp, NULL);
1013 	ether_ifattach(ifp, eaddr);
1014 	ether_set_ifflags_cb(&sc->sc_ethercom, vge_ifflags_cb);
1015 
1016 	callout_init(&sc->sc_timeout, 0);
1017 	callout_setfunc(&sc->sc_timeout, vge_tick, sc);
1018 
1019 	/*
1020 	 * Make sure the interface is shutdown during reboot.
1021 	 */
1022 	if (pmf_device_register1(self, NULL, NULL, vge_shutdown))
1023 		pmf_class_network_register(self, ifp);
1024 	else
1025 		aprint_error_dev(self, "couldn't establish power handler\n");
1026 }
1027 
1028 static int
1029 vge_newbuf(struct vge_softc *sc, int idx, struct mbuf *m)
1030 {
1031 	struct mbuf *m_new;
1032 	struct vge_rxdesc *rxd;
1033 	struct vge_rxsoft *rxs;
1034 	bus_dmamap_t map;
1035 	int i;
1036 #ifdef DIAGNOSTIC
1037 	uint32_t rd_sts;
1038 #endif
1039 
1040 	m_new = NULL;
1041 	if (m == NULL) {
1042 		MGETHDR(m_new, M_DONTWAIT, MT_DATA);
1043 		if (m_new == NULL)
1044 			return ENOBUFS;
1045 
1046 		MCLGET(m_new, M_DONTWAIT);
1047 		if ((m_new->m_flags & M_EXT) == 0) {
1048 			m_freem(m_new);
1049 			return ENOBUFS;
1050 		}
1051 
1052 		m = m_new;
1053 	} else
1054 		m->m_data = m->m_ext.ext_buf;
1055 
1056 
1057 	/*
1058 	 * This is part of an evil trick to deal with non-x86 platforms.
1059 	 * The VIA chip requires RX buffers to be aligned on 32-bit
1060 	 * boundaries, but that will hose non-x86 machines. To get around
1061 	 * this, we leave some empty space at the start of each buffer
1062 	 * and for non-x86 hosts, we copy the buffer back two bytes
1063 	 * to achieve word alignment. This is slightly more efficient
1064 	 * than allocating a new buffer, copying the contents, and
1065 	 * discarding the old buffer.
1066 	 */
1067 	m->m_len = m->m_pkthdr.len = VGE_RX_BUFSIZE;
1068 #ifndef __NO_STRICT_ALIGNMENT
1069 	m->m_data += VGE_RX_PAD;
1070 #endif
1071 	rxs = &sc->sc_rxsoft[idx];
1072 	map = rxs->rxs_dmamap;
1073 
1074 	if (bus_dmamap_load_mbuf(sc->sc_dmat, map, m, BUS_DMA_NOWAIT) != 0)
1075 		goto out;
1076 
1077 	rxd = &sc->sc_rxdescs[idx];
1078 
1079 #ifdef DIAGNOSTIC
1080 	/* If this descriptor is still owned by the chip, bail. */
1081 	VGE_RXDESCSYNC(sc, idx, BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1082 	rd_sts = le32toh(rxd->rd_sts);
1083 	VGE_RXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD);
1084 	if (rd_sts & VGE_RDSTS_OWN) {
1085 		panic("%s: tried to map busy RX descriptor",
1086 		    device_xname(sc->sc_dev));
1087 	}
1088 #endif
1089 
1090 	rxs->rxs_mbuf = m;
1091 	bus_dmamap_sync(sc->sc_dmat, map, 0, map->dm_mapsize,
1092 	    BUS_DMASYNC_PREREAD);
1093 
1094 	rxd->rd_buflen =
1095 	    htole16(VGE_BUFLEN(map->dm_segs[0].ds_len) | VGE_RXDESC_I);
1096 	vge_set_rxaddr(rxd, map->dm_segs[0].ds_addr);
1097 	rxd->rd_sts = 0;
1098 	rxd->rd_ctl = 0;
1099 	VGE_RXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1100 
1101 	/*
1102 	 * Note: the manual fails to document the fact that for
1103 	 * proper opration, the driver needs to replentish the RX
1104 	 * DMA ring 4 descriptors at a time (rather than one at a
1105 	 * time, like most chips). We can allocate the new buffers
1106 	 * but we should not set the OWN bits until we're ready
1107 	 * to hand back 4 of them in one shot.
1108 	 */
1109 
1110 #define VGE_RXCHUNK 4
1111 	sc->sc_rx_consumed++;
1112 	if (sc->sc_rx_consumed == VGE_RXCHUNK) {
1113 		for (i = idx; i != idx - VGE_RXCHUNK; i--) {
1114 			KASSERT(i >= 0);
1115 			sc->sc_rxdescs[i].rd_sts |= htole32(VGE_RDSTS_OWN);
1116 			VGE_RXDESCSYNC(sc, i,
1117 			    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1118 		}
1119 		sc->sc_rx_consumed = 0;
1120 	}
1121 
1122 	return 0;
1123  out:
1124 	if (m_new != NULL)
1125 		m_freem(m_new);
1126 	return ENOMEM;
1127 }
1128 
1129 #ifndef __NO_STRICT_ALIGNMENT
1130 static inline void
1131 vge_fixup_rx(struct mbuf *m)
1132 {
1133 	int i;
1134 	uint16_t *src, *dst;
1135 
1136 	src = mtod(m, uint16_t *);
1137 	dst = src - 1;
1138 
1139 	for (i = 0; i < (m->m_len / sizeof(uint16_t) + 1); i++)
1140 		*dst++ = *src++;
1141 
1142 	m->m_data -= ETHER_ALIGN;
1143 }
1144 #endif
1145 
1146 /*
1147  * RX handler. We support the reception of jumbo frames that have
1148  * been fragmented across multiple 2K mbuf cluster buffers.
1149  */
1150 static void
1151 vge_rxeof(struct vge_softc *sc)
1152 {
1153 	struct mbuf *m;
1154 	struct ifnet *ifp;
1155 	int idx, total_len, lim;
1156 	struct vge_rxdesc *cur_rxd;
1157 	struct vge_rxsoft *rxs;
1158 	uint32_t rxstat, rxctl;
1159 
1160 	ifp = &sc->sc_ethercom.ec_if;
1161 	lim = 0;
1162 
1163 	/* Invalidate the descriptor memory */
1164 
1165 	for (idx = sc->sc_rx_prodidx;; idx = VGE_NEXT_RXDESC(idx)) {
1166 		cur_rxd = &sc->sc_rxdescs[idx];
1167 
1168 		VGE_RXDESCSYNC(sc, idx,
1169 		    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1170 		rxstat = le32toh(cur_rxd->rd_sts);
1171 		if ((rxstat & VGE_RDSTS_OWN) != 0) {
1172 			VGE_RXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD);
1173 			break;
1174 		}
1175 
1176 		rxctl = le32toh(cur_rxd->rd_ctl);
1177 		rxs = &sc->sc_rxsoft[idx];
1178 		m = rxs->rxs_mbuf;
1179 		total_len = (rxstat & VGE_RDSTS_BUFSIZ) >> 16;
1180 
1181 		/* Invalidate the RX mbuf and unload its map */
1182 
1183 		bus_dmamap_sync(sc->sc_dmat, rxs->rxs_dmamap,
1184 		    0, rxs->rxs_dmamap->dm_mapsize, BUS_DMASYNC_POSTREAD);
1185 		bus_dmamap_unload(sc->sc_dmat, rxs->rxs_dmamap);
1186 
1187 		/*
1188 		 * If the 'start of frame' bit is set, this indicates
1189 		 * either the first fragment in a multi-fragment receive,
1190 		 * or an intermediate fragment. Either way, we want to
1191 		 * accumulate the buffers.
1192 		 */
1193 		if (rxstat & VGE_RXPKT_SOF) {
1194 			m->m_len = VGE_RX_BUFSIZE;
1195 			if (sc->sc_rx_mhead == NULL)
1196 				sc->sc_rx_mhead = sc->sc_rx_mtail = m;
1197 			else {
1198 				m->m_flags &= ~M_PKTHDR;
1199 				sc->sc_rx_mtail->m_next = m;
1200 				sc->sc_rx_mtail = m;
1201 			}
1202 			vge_newbuf(sc, idx, NULL);
1203 			continue;
1204 		}
1205 
1206 		/*
1207 		 * Bad/error frames will have the RXOK bit cleared.
1208 		 * However, there's one error case we want to allow:
1209 		 * if a VLAN tagged frame arrives and the chip can't
1210 		 * match it against the CAM filter, it considers this
1211 		 * a 'VLAN CAM filter miss' and clears the 'RXOK' bit.
1212 		 * We don't want to drop the frame though: our VLAN
1213 		 * filtering is done in software.
1214 		 */
1215 		if ((rxstat & VGE_RDSTS_RXOK) == 0 &&
1216 		    (rxstat & VGE_RDSTS_VIDM) == 0 &&
1217 		    (rxstat & VGE_RDSTS_CSUMERR) == 0) {
1218 			ifp->if_ierrors++;
1219 			/*
1220 			 * If this is part of a multi-fragment packet,
1221 			 * discard all the pieces.
1222 			 */
1223 			if (sc->sc_rx_mhead != NULL) {
1224 				m_freem(sc->sc_rx_mhead);
1225 				sc->sc_rx_mhead = sc->sc_rx_mtail = NULL;
1226 			}
1227 			vge_newbuf(sc, idx, m);
1228 			continue;
1229 		}
1230 
1231 		/*
1232 		 * If allocating a replacement mbuf fails,
1233 		 * reload the current one.
1234 		 */
1235 
1236 		if (vge_newbuf(sc, idx, NULL)) {
1237 			ifp->if_ierrors++;
1238 			if (sc->sc_rx_mhead != NULL) {
1239 				m_freem(sc->sc_rx_mhead);
1240 				sc->sc_rx_mhead = sc->sc_rx_mtail = NULL;
1241 			}
1242 			vge_newbuf(sc, idx, m);
1243 			continue;
1244 		}
1245 
1246 		if (sc->sc_rx_mhead != NULL) {
1247 			m->m_len = total_len % VGE_RX_BUFSIZE;
1248 			/*
1249 			 * Special case: if there's 4 bytes or less
1250 			 * in this buffer, the mbuf can be discarded:
1251 			 * the last 4 bytes is the CRC, which we don't
1252 			 * care about anyway.
1253 			 */
1254 			if (m->m_len <= ETHER_CRC_LEN) {
1255 				sc->sc_rx_mtail->m_len -=
1256 				    (ETHER_CRC_LEN - m->m_len);
1257 				m_freem(m);
1258 			} else {
1259 				m->m_len -= ETHER_CRC_LEN;
1260 				m->m_flags &= ~M_PKTHDR;
1261 				sc->sc_rx_mtail->m_next = m;
1262 			}
1263 			m = sc->sc_rx_mhead;
1264 			sc->sc_rx_mhead = sc->sc_rx_mtail = NULL;
1265 			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
1266 		} else
1267 			m->m_pkthdr.len = m->m_len = total_len - ETHER_CRC_LEN;
1268 
1269 #ifndef __NO_STRICT_ALIGNMENT
1270 		vge_fixup_rx(m);
1271 #endif
1272 		m_set_rcvif(m, ifp);
1273 
1274 		/* Do RX checksumming if enabled */
1275 		if (ifp->if_csum_flags_rx & M_CSUM_IPv4) {
1276 
1277 			/* Check IP header checksum */
1278 			if (rxctl & VGE_RDCTL_IPPKT)
1279 				m->m_pkthdr.csum_flags |= M_CSUM_IPv4;
1280 			if ((rxctl & VGE_RDCTL_IPCSUMOK) == 0)
1281 				m->m_pkthdr.csum_flags |= M_CSUM_IPv4_BAD;
1282 		}
1283 
1284 		if (ifp->if_csum_flags_rx & M_CSUM_TCPv4) {
1285 			/* Check UDP checksum */
1286 			if (rxctl & VGE_RDCTL_TCPPKT)
1287 				m->m_pkthdr.csum_flags |= M_CSUM_TCPv4;
1288 
1289 			if ((rxctl & VGE_RDCTL_PROTOCSUMOK) == 0)
1290 				m->m_pkthdr.csum_flags |= M_CSUM_TCP_UDP_BAD;
1291 		}
1292 
1293 		if (ifp->if_csum_flags_rx & M_CSUM_UDPv4) {
1294 			/* Check UDP checksum */
1295 			if (rxctl & VGE_RDCTL_UDPPKT)
1296 				m->m_pkthdr.csum_flags |= M_CSUM_UDPv4;
1297 
1298 			if ((rxctl & VGE_RDCTL_PROTOCSUMOK) == 0)
1299 				m->m_pkthdr.csum_flags |= M_CSUM_TCP_UDP_BAD;
1300 		}
1301 
1302 		if (rxstat & VGE_RDSTS_VTAG) {
1303 			/*
1304 			 * We use bswap16() here because:
1305 			 * On LE machines, tag is stored in BE as stream data.
1306 			 * On BE machines, tag is stored in BE as stream data
1307 			 *  but it was already swapped by le32toh() above.
1308 			 */
1309 			vlan_set_tag(m, bswap16(rxctl & VGE_RDCTL_VLANID));
1310 		}
1311 
1312 		if_percpuq_enqueue(ifp->if_percpuq, m);
1313 
1314 		lim++;
1315 		if (lim == VGE_NRXDESC)
1316 			break;
1317 	}
1318 
1319 	sc->sc_rx_prodidx = idx;
1320 	CSR_WRITE_2(sc, VGE_RXDESC_RESIDUECNT, lim);
1321 }
1322 
1323 static void
1324 vge_txeof(struct vge_softc *sc)
1325 {
1326 	struct ifnet *ifp;
1327 	struct vge_txsoft *txs;
1328 	uint32_t txstat;
1329 	int idx;
1330 
1331 	ifp = &sc->sc_ethercom.ec_if;
1332 
1333 	for (idx = sc->sc_tx_considx;
1334 	    sc->sc_tx_free < VGE_NTXDESC;
1335 	    idx = VGE_NEXT_TXDESC(idx), sc->sc_tx_free++) {
1336 		VGE_TXDESCSYNC(sc, idx,
1337 		    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1338 		txstat = le32toh(sc->sc_txdescs[idx].td_sts);
1339 		VGE_TXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD);
1340 		if (txstat & VGE_TDSTS_OWN) {
1341 			break;
1342 		}
1343 
1344 		txs = &sc->sc_txsoft[idx];
1345 		m_freem(txs->txs_mbuf);
1346 		txs->txs_mbuf = NULL;
1347 		bus_dmamap_sync(sc->sc_dmat, txs->txs_dmamap, 0,
1348 		    txs->txs_dmamap->dm_mapsize, BUS_DMASYNC_POSTWRITE);
1349 		bus_dmamap_unload(sc->sc_dmat, txs->txs_dmamap);
1350 		if (txstat & (VGE_TDSTS_EXCESSCOLL|VGE_TDSTS_COLL))
1351 			ifp->if_collisions++;
1352 		if (txstat & VGE_TDSTS_TXERR)
1353 			ifp->if_oerrors++;
1354 		else
1355 			ifp->if_opackets++;
1356 	}
1357 
1358 	sc->sc_tx_considx = idx;
1359 
1360 	if (sc->sc_tx_free > 0) {
1361 		ifp->if_flags &= ~IFF_OACTIVE;
1362 	}
1363 
1364 	/*
1365 	 * If not all descriptors have been released reaped yet,
1366 	 * reload the timer so that we will eventually get another
1367 	 * interrupt that will cause us to re-enter this routine.
1368 	 * This is done in case the transmitter has gone idle.
1369 	 */
1370 	if (sc->sc_tx_free < VGE_NTXDESC)
1371 		CSR_WRITE_1(sc, VGE_CRS1, VGE_CR1_TIMER0_ENABLE);
1372 	else
1373 		ifp->if_timer = 0;
1374 }
1375 
1376 static void
1377 vge_tick(void *arg)
1378 {
1379 	struct vge_softc *sc;
1380 	struct ifnet *ifp;
1381 	struct mii_data *mii;
1382 	int s;
1383 
1384 	sc = arg;
1385 	ifp = &sc->sc_ethercom.ec_if;
1386 	mii = &sc->sc_mii;
1387 
1388 	s = splnet();
1389 
1390 	callout_schedule(&sc->sc_timeout, hz);
1391 
1392 	mii_tick(mii);
1393 	if (sc->sc_link) {
1394 		if ((mii->mii_media_status & IFM_ACTIVE) == 0)
1395 			sc->sc_link = 0;
1396 	} else {
1397 		if (mii->mii_media_status & IFM_ACTIVE &&
1398 		    IFM_SUBTYPE(mii->mii_media_active) != IFM_NONE) {
1399 			sc->sc_link = 1;
1400 			if (!IFQ_IS_EMPTY(&ifp->if_snd))
1401 				vge_start(ifp);
1402 		}
1403 	}
1404 
1405 	splx(s);
1406 }
1407 
1408 static int
1409 vge_intr(void *arg)
1410 {
1411 	struct vge_softc *sc;
1412 	struct ifnet *ifp;
1413 	uint32_t status;
1414 	int claim;
1415 
1416 	sc = arg;
1417 	claim = 0;
1418 	if (sc->sc_suspended) {
1419 		return claim;
1420 	}
1421 
1422 	ifp = &sc->sc_ethercom.ec_if;
1423 
1424 	if ((ifp->if_flags & IFF_UP) == 0) {
1425 		return claim;
1426 	}
1427 
1428 	/* Disable interrupts */
1429 	CSR_WRITE_1(sc, VGE_CRC3, VGE_CR3_INT_GMSK);
1430 
1431 	for (;;) {
1432 
1433 		status = CSR_READ_4(sc, VGE_ISR);
1434 		/* If the card has gone away the read returns 0xffffffff. */
1435 		if (status == 0xFFFFFFFF)
1436 			break;
1437 
1438 		if (status) {
1439 			claim = 1;
1440 			CSR_WRITE_4(sc, VGE_ISR, status);
1441 		}
1442 
1443 		if ((status & VGE_INTRS) == 0)
1444 			break;
1445 
1446 		if (status & (VGE_ISR_RXOK|VGE_ISR_RXOK_HIPRIO))
1447 			vge_rxeof(sc);
1448 
1449 		if (status & (VGE_ISR_RXOFLOW|VGE_ISR_RXNODESC)) {
1450 			vge_rxeof(sc);
1451 			CSR_WRITE_1(sc, VGE_RXQCSRS, VGE_RXQCSR_RUN);
1452 			CSR_WRITE_1(sc, VGE_RXQCSRS, VGE_RXQCSR_WAK);
1453 		}
1454 
1455 		if (status & (VGE_ISR_TXOK0|VGE_ISR_TIMER0))
1456 			vge_txeof(sc);
1457 
1458 		if (status & (VGE_ISR_TXDMA_STALL|VGE_ISR_RXDMA_STALL))
1459 			vge_init(ifp);
1460 
1461 		if (status & VGE_ISR_LINKSTS)
1462 			vge_tick(sc);
1463 	}
1464 
1465 	/* Re-enable interrupts */
1466 	CSR_WRITE_1(sc, VGE_CRS3, VGE_CR3_INT_GMSK);
1467 
1468 	if (claim)
1469 		if_schedule_deferred_start(ifp);
1470 
1471 	return claim;
1472 }
1473 
1474 static int
1475 vge_encap(struct vge_softc *sc, struct mbuf *m_head, int idx)
1476 {
1477 	struct vge_txsoft *txs;
1478 	struct vge_txdesc *txd;
1479 	struct vge_txfrag *f;
1480 	struct mbuf *m_new;
1481 	bus_dmamap_t map;
1482 	int m_csumflags, seg, error, flags;
1483 	size_t sz;
1484 	uint32_t td_sts, td_ctl;
1485 
1486 	KASSERT(sc->sc_tx_free > 0);
1487 
1488 	txd = &sc->sc_txdescs[idx];
1489 
1490 #ifdef DIAGNOSTIC
1491 	/* If this descriptor is still owned by the chip, bail. */
1492 	VGE_TXDESCSYNC(sc, idx,
1493 	    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1494 	td_sts = le32toh(txd->td_sts);
1495 	VGE_TXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD);
1496 	if (td_sts & VGE_TDSTS_OWN) {
1497 		return ENOBUFS;
1498 	}
1499 #endif
1500 
1501 	/*
1502 	 * Preserve m_pkthdr.csum_flags here since m_head might be
1503 	 * updated by m_defrag()
1504 	 */
1505 	m_csumflags = m_head->m_pkthdr.csum_flags;
1506 
1507 	txs = &sc->sc_txsoft[idx];
1508 	map = txs->txs_dmamap;
1509 	error = bus_dmamap_load_mbuf(sc->sc_dmat, map, m_head, BUS_DMA_NOWAIT);
1510 
1511 	/* If too many segments to map, coalesce */
1512 	if (error == EFBIG ||
1513 	    (m_head->m_pkthdr.len < ETHER_PAD_LEN &&
1514 	     map->dm_nsegs == VGE_TX_FRAGS)) {
1515 		m_new = m_defrag(m_head, M_DONTWAIT);
1516 		if (m_new == NULL)
1517 			return EFBIG;
1518 
1519 		error = bus_dmamap_load_mbuf(sc->sc_dmat, map,
1520 		    m_new, BUS_DMA_NOWAIT);
1521 		if (error) {
1522 			m_freem(m_new);
1523 			return error;
1524 		}
1525 
1526 		m_head = m_new;
1527 	} else if (error)
1528 		return error;
1529 
1530 	txs->txs_mbuf = m_head;
1531 
1532 	bus_dmamap_sync(sc->sc_dmat, map, 0, map->dm_mapsize,
1533 	    BUS_DMASYNC_PREWRITE);
1534 
1535 	for (seg = 0, f = &txd->td_frag[0]; seg < map->dm_nsegs; seg++, f++) {
1536 		f->tf_buflen = htole16(VGE_BUFLEN(map->dm_segs[seg].ds_len));
1537 		vge_set_txaddr(f, map->dm_segs[seg].ds_addr);
1538 	}
1539 
1540 	/* Argh. This chip does not autopad short frames */
1541 	sz = m_head->m_pkthdr.len;
1542 	if (sz < ETHER_PAD_LEN) {
1543 		f->tf_buflen = htole16(VGE_BUFLEN(ETHER_PAD_LEN - sz));
1544 		vge_set_txaddr(f, VGE_CDPADADDR(sc));
1545 		sz = ETHER_PAD_LEN;
1546 		seg++;
1547 	}
1548 	VGE_TXFRAGSYNC(sc, idx, seg, BUS_DMASYNC_PREWRITE);
1549 
1550 	/*
1551 	 * When telling the chip how many segments there are, we
1552 	 * must use nsegs + 1 instead of just nsegs. Darned if I
1553 	 * know why.
1554 	 */
1555 	seg++;
1556 
1557 	flags = 0;
1558 	if (m_csumflags & M_CSUM_IPv4)
1559 		flags |= VGE_TDCTL_IPCSUM;
1560 	if (m_csumflags & M_CSUM_TCPv4)
1561 		flags |= VGE_TDCTL_TCPCSUM;
1562 	if (m_csumflags & M_CSUM_UDPv4)
1563 		flags |= VGE_TDCTL_UDPCSUM;
1564 	td_sts = sz << 16;
1565 	td_ctl = flags | (seg << 28) | VGE_TD_LS_NORM;
1566 
1567 	if (sz > ETHERMTU + ETHER_HDR_LEN)
1568 		td_ctl |= VGE_TDCTL_JUMBO;
1569 
1570 	/*
1571 	 * Set up hardware VLAN tagging.
1572 	 */
1573 	if (vlan_has_tag(m_head)) {
1574 		/*
1575 		 * No need htons() here since vge(4) chip assumes
1576 		 * that tags are written in little endian and
1577 		 * we already use htole32() here.
1578 		 */
1579 		td_ctl |= vlan_get_tag(m_head) | VGE_TDCTL_VTAG;
1580 	}
1581 	txd->td_ctl = htole32(td_ctl);
1582 	txd->td_sts = htole32(td_sts);
1583 	VGE_TXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1584 
1585 	txd->td_sts = htole32(VGE_TDSTS_OWN | td_sts);
1586 	VGE_TXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1587 
1588 	sc->sc_tx_free--;
1589 
1590 	return 0;
1591 }
1592 
1593 /*
1594  * Main transmit routine.
1595  */
1596 
1597 static void
1598 vge_start(struct ifnet *ifp)
1599 {
1600 	struct vge_softc *sc;
1601 	struct vge_txsoft *txs;
1602 	struct mbuf *m_head;
1603 	int idx, pidx, ofree, error;
1604 
1605 	sc = ifp->if_softc;
1606 
1607 	if (!sc->sc_link ||
1608 	    (ifp->if_flags & (IFF_RUNNING|IFF_OACTIVE)) != IFF_RUNNING) {
1609 		return;
1610 	}
1611 
1612 	m_head = NULL;
1613 	idx = sc->sc_tx_prodidx;
1614 	pidx = VGE_PREV_TXDESC(idx);
1615 	ofree = sc->sc_tx_free;
1616 
1617 	/*
1618 	 * Loop through the send queue, setting up transmit descriptors
1619 	 * until we drain the queue, or use up all available transmit
1620 	 * descriptors.
1621 	 */
1622 	for (;;) {
1623 		/* Grab a packet off the queue. */
1624 		IFQ_POLL(&ifp->if_snd, m_head);
1625 		if (m_head == NULL)
1626 			break;
1627 
1628 		if (sc->sc_tx_free == 0) {
1629 			/*
1630 			 * All slots used, stop for now.
1631 			 */
1632 			ifp->if_flags |= IFF_OACTIVE;
1633 			break;
1634 		}
1635 
1636 		txs = &sc->sc_txsoft[idx];
1637 		KASSERT(txs->txs_mbuf == NULL);
1638 
1639 		if ((error = vge_encap(sc, m_head, idx))) {
1640 			if (error == EFBIG) {
1641 				printf("%s: Tx packet consumes too many "
1642 				    "DMA segments, dropping...\n",
1643 				    device_xname(sc->sc_dev));
1644 				IFQ_DEQUEUE(&ifp->if_snd, m_head);
1645 				m_freem(m_head);
1646 				continue;
1647 			}
1648 
1649 			/*
1650 			 * Short on resources, just stop for now.
1651 			 */
1652 			if (error == ENOBUFS)
1653 				ifp->if_flags |= IFF_OACTIVE;
1654 			break;
1655 		}
1656 
1657 		IFQ_DEQUEUE(&ifp->if_snd, m_head);
1658 
1659 		/*
1660 		 * WE ARE NOW COMMITTED TO TRANSMITTING THE PACKET.
1661 		 */
1662 
1663 		sc->sc_txdescs[pidx].td_frag[0].tf_buflen |=
1664 		    htole16(VGE_TXDESC_Q);
1665 		VGE_TXFRAGSYNC(sc, pidx, 1,
1666 		    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1667 
1668 		if (txs->txs_mbuf != m_head) {
1669 			m_freem(m_head);
1670 			m_head = txs->txs_mbuf;
1671 		}
1672 
1673 		pidx = idx;
1674 		idx = VGE_NEXT_TXDESC(idx);
1675 
1676 		/*
1677 		 * If there's a BPF listener, bounce a copy of this frame
1678 		 * to him.
1679 		 */
1680 		bpf_mtap(ifp, m_head, BPF_D_OUT);
1681 	}
1682 
1683 	if (sc->sc_tx_free < ofree) {
1684 		/* TX packet queued */
1685 
1686 		sc->sc_tx_prodidx = idx;
1687 
1688 		/* Issue a transmit command. */
1689 		CSR_WRITE_2(sc, VGE_TXQCSRS, VGE_TXQCSR_WAK0);
1690 
1691 		/*
1692 		 * Use the countdown timer for interrupt moderation.
1693 		 * 'TX done' interrupts are disabled. Instead, we reset the
1694 		 * countdown timer, which will begin counting until it hits
1695 		 * the value in the SSTIMER register, and then trigger an
1696 		 * interrupt. Each time we set the TIMER0_ENABLE bit, the
1697 		 * the timer count is reloaded. Only when the transmitter
1698 		 * is idle will the timer hit 0 and an interrupt fire.
1699 		 */
1700 		CSR_WRITE_1(sc, VGE_CRS1, VGE_CR1_TIMER0_ENABLE);
1701 
1702 		/*
1703 		 * Set a timeout in case the chip goes out to lunch.
1704 		 */
1705 		ifp->if_timer = 5;
1706 	}
1707 }
1708 
1709 static int
1710 vge_init(struct ifnet *ifp)
1711 {
1712 	struct vge_softc *sc;
1713 	int i, rc = 0;
1714 
1715 	sc = ifp->if_softc;
1716 
1717 	/*
1718 	 * Cancel pending I/O and free all RX/TX buffers.
1719 	 */
1720 	vge_stop(ifp, 0);
1721 	vge_reset(sc);
1722 
1723 	/* Initialize the RX descriptors and mbufs. */
1724 	memset(sc->sc_rxdescs, 0, sizeof(sc->sc_rxdescs));
1725 	sc->sc_rx_consumed = 0;
1726 	for (i = 0; i < VGE_NRXDESC; i++) {
1727 		if (vge_newbuf(sc, i, NULL) == ENOBUFS) {
1728 			printf("%s: unable to allocate or map rx buffer\n",
1729 			    device_xname(sc->sc_dev));
1730 			return 1; /* XXX */
1731 		}
1732 	}
1733 	sc->sc_rx_prodidx = 0;
1734 	sc->sc_rx_mhead = sc->sc_rx_mtail = NULL;
1735 
1736 	/* Initialize the  TX descriptors and mbufs. */
1737 	memset(sc->sc_txdescs, 0, sizeof(sc->sc_txdescs));
1738 	bus_dmamap_sync(sc->sc_dmat, sc->sc_cddmamap,
1739 	    VGE_CDTXOFF(0), sizeof(sc->sc_txdescs),
1740 	    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1741 	for (i = 0; i < VGE_NTXDESC; i++)
1742 		sc->sc_txsoft[i].txs_mbuf = NULL;
1743 
1744 	sc->sc_tx_prodidx = 0;
1745 	sc->sc_tx_considx = 0;
1746 	sc->sc_tx_free = VGE_NTXDESC;
1747 
1748 	/* Set our station address */
1749 	for (i = 0; i < ETHER_ADDR_LEN; i++)
1750 		CSR_WRITE_1(sc, VGE_PAR0 + i, sc->sc_eaddr[i]);
1751 
1752 	/*
1753 	 * Set receive FIFO threshold. Also allow transmission and
1754 	 * reception of VLAN tagged frames.
1755 	 */
1756 	CSR_CLRBIT_1(sc, VGE_RXCFG, VGE_RXCFG_FIFO_THR|VGE_RXCFG_VTAGOPT);
1757 	CSR_SETBIT_1(sc, VGE_RXCFG, VGE_RXFIFOTHR_128BYTES|VGE_VTAG_OPT2);
1758 
1759 	/* Set DMA burst length */
1760 	CSR_CLRBIT_1(sc, VGE_DMACFG0, VGE_DMACFG0_BURSTLEN);
1761 	CSR_SETBIT_1(sc, VGE_DMACFG0, VGE_DMABURST_128);
1762 
1763 	CSR_SETBIT_1(sc, VGE_TXCFG, VGE_TXCFG_ARB_PRIO|VGE_TXCFG_NONBLK);
1764 
1765 	/* Set collision backoff algorithm */
1766 	CSR_CLRBIT_1(sc, VGE_CHIPCFG1, VGE_CHIPCFG1_CRANDOM|
1767 	    VGE_CHIPCFG1_CAP|VGE_CHIPCFG1_MBA|VGE_CHIPCFG1_BAKOPT);
1768 	CSR_SETBIT_1(sc, VGE_CHIPCFG1, VGE_CHIPCFG1_OFSET);
1769 
1770 	/* Disable LPSEL field in priority resolution */
1771 	CSR_SETBIT_1(sc, VGE_DIAGCTL, VGE_DIAGCTL_LPSEL_DIS);
1772 
1773 	/*
1774 	 * Load the addresses of the DMA queues into the chip.
1775 	 * Note that we only use one transmit queue.
1776 	 */
1777 
1778 	CSR_WRITE_4(sc, VGE_TXDESC_ADDR_LO0, VGE_ADDR_LO(VGE_CDTXADDR(sc, 0)));
1779 	CSR_WRITE_2(sc, VGE_TXDESCNUM, VGE_NTXDESC - 1);
1780 
1781 	CSR_WRITE_4(sc, VGE_RXDESC_ADDR_LO, VGE_ADDR_LO(VGE_CDRXADDR(sc, 0)));
1782 	CSR_WRITE_2(sc, VGE_RXDESCNUM, VGE_NRXDESC - 1);
1783 	CSR_WRITE_2(sc, VGE_RXDESC_RESIDUECNT, VGE_NRXDESC);
1784 
1785 	/* Enable and wake up the RX descriptor queue */
1786 	CSR_WRITE_1(sc, VGE_RXQCSRS, VGE_RXQCSR_RUN);
1787 	CSR_WRITE_1(sc, VGE_RXQCSRS, VGE_RXQCSR_WAK);
1788 
1789 	/* Enable the TX descriptor queue */
1790 	CSR_WRITE_2(sc, VGE_TXQCSRS, VGE_TXQCSR_RUN0);
1791 
1792 	/* Set up the receive filter -- allow large frames for VLANs. */
1793 	CSR_WRITE_1(sc, VGE_RXCTL, VGE_RXCTL_RX_UCAST|VGE_RXCTL_RX_GIANT);
1794 
1795 	/* If we want promiscuous mode, set the allframes bit. */
1796 	if (ifp->if_flags & IFF_PROMISC) {
1797 		CSR_SETBIT_1(sc, VGE_RXCTL, VGE_RXCTL_RX_PROMISC);
1798 	}
1799 
1800 	/* Set capture broadcast bit to capture broadcast frames. */
1801 	if (ifp->if_flags & IFF_BROADCAST) {
1802 		CSR_SETBIT_1(sc, VGE_RXCTL, VGE_RXCTL_RX_BCAST);
1803 	}
1804 
1805 	/* Set multicast bit to capture multicast frames. */
1806 	if (ifp->if_flags & IFF_MULTICAST) {
1807 		CSR_SETBIT_1(sc, VGE_RXCTL, VGE_RXCTL_RX_MCAST);
1808 	}
1809 
1810 	/* Init the cam filter. */
1811 	vge_cam_clear(sc);
1812 
1813 	/* Init the multicast filter. */
1814 	vge_setmulti(sc);
1815 
1816 	/* Enable flow control */
1817 
1818 	CSR_WRITE_1(sc, VGE_CRS2, 0x8B);
1819 
1820 	/* Enable jumbo frame reception (if desired) */
1821 
1822 	/* Start the MAC. */
1823 	CSR_WRITE_1(sc, VGE_CRC0, VGE_CR0_STOP);
1824 	CSR_WRITE_1(sc, VGE_CRS1, VGE_CR1_NOPOLL);
1825 	CSR_WRITE_1(sc, VGE_CRS0,
1826 	    VGE_CR0_TX_ENABLE|VGE_CR0_RX_ENABLE|VGE_CR0_START);
1827 
1828 	/*
1829 	 * Configure one-shot timer for microsecond
1830 	 * resulution and load it for 500 usecs.
1831 	 */
1832 	CSR_SETBIT_1(sc, VGE_DIAGCTL, VGE_DIAGCTL_TIMER0_RES);
1833 	CSR_WRITE_2(sc, VGE_SSTIMER, 400);
1834 
1835 	/*
1836 	 * Configure interrupt moderation for receive. Enable
1837 	 * the holdoff counter and load it, and set the RX
1838 	 * suppression count to the number of descriptors we
1839 	 * want to allow before triggering an interrupt.
1840 	 * The holdoff timer is in units of 20 usecs.
1841 	 */
1842 
1843 #ifdef notyet
1844 	CSR_WRITE_1(sc, VGE_INTCTL1, VGE_INTCTL_TXINTSUP_DISABLE);
1845 	/* Select the interrupt holdoff timer page. */
1846 	CSR_CLRBIT_1(sc, VGE_CAMCTL, VGE_CAMCTL_PAGESEL);
1847 	CSR_SETBIT_1(sc, VGE_CAMCTL, VGE_PAGESEL_INTHLDOFF);
1848 	CSR_WRITE_1(sc, VGE_INTHOLDOFF, 10); /* ~200 usecs */
1849 
1850 	/* Enable use of the holdoff timer. */
1851 	CSR_WRITE_1(sc, VGE_CRS3, VGE_CR3_INT_HOLDOFF);
1852 	CSR_WRITE_1(sc, VGE_INTCTL1, VGE_INTCTL_SC_RELOAD);
1853 
1854 	/* Select the RX suppression threshold page. */
1855 	CSR_CLRBIT_1(sc, VGE_CAMCTL, VGE_CAMCTL_PAGESEL);
1856 	CSR_SETBIT_1(sc, VGE_CAMCTL, VGE_PAGESEL_RXSUPPTHR);
1857 	CSR_WRITE_1(sc, VGE_RXSUPPTHR, 64); /* interrupt after 64 packets */
1858 
1859 	/* Restore the page select bits. */
1860 	CSR_CLRBIT_1(sc, VGE_CAMCTL, VGE_CAMCTL_PAGESEL);
1861 	CSR_SETBIT_1(sc, VGE_CAMCTL, VGE_PAGESEL_MAR);
1862 #endif
1863 
1864 #ifdef DEVICE_POLLING
1865 	/*
1866 	 * Disable interrupts if we are polling.
1867 	 */
1868 	if (ifp->if_flags & IFF_POLLING) {
1869 		CSR_WRITE_4(sc, VGE_IMR, 0);
1870 		CSR_WRITE_1(sc, VGE_CRC3, VGE_CR3_INT_GMSK);
1871 	} else	/* otherwise ... */
1872 #endif /* DEVICE_POLLING */
1873 	{
1874 	/*
1875 	 * Enable interrupts.
1876 	 */
1877 		CSR_WRITE_4(sc, VGE_IMR, VGE_INTRS);
1878 		CSR_WRITE_4(sc, VGE_ISR, 0);
1879 		CSR_WRITE_1(sc, VGE_CRS3, VGE_CR3_INT_GMSK);
1880 	}
1881 
1882 	if ((rc = ether_mediachange(ifp)) != 0)
1883 		goto out;
1884 
1885 	ifp->if_flags |= IFF_RUNNING;
1886 	ifp->if_flags &= ~IFF_OACTIVE;
1887 
1888 	sc->sc_if_flags = 0;
1889 	sc->sc_link = 0;
1890 
1891 	callout_schedule(&sc->sc_timeout, hz);
1892 
1893 out:
1894 	return rc;
1895 }
1896 
1897 static void
1898 vge_miibus_statchg(struct ifnet *ifp)
1899 {
1900 	struct vge_softc *sc = ifp->if_softc;
1901 	struct mii_data *mii = &sc->sc_mii;
1902 	struct ifmedia_entry *ife = mii->mii_media.ifm_cur;
1903 
1904 	/*
1905 	 * If the user manually selects a media mode, we need to turn
1906 	 * on the forced MAC mode bit in the DIAGCTL register. If the
1907 	 * user happens to choose a full duplex mode, we also need to
1908 	 * set the 'force full duplex' bit. This applies only to
1909 	 * 10Mbps and 100Mbps speeds. In autoselect mode, forced MAC
1910 	 * mode is disabled, and in 1000baseT mode, full duplex is
1911 	 * always implied, so we turn on the forced mode bit but leave
1912 	 * the FDX bit cleared.
1913 	 */
1914 
1915 	switch (IFM_SUBTYPE(ife->ifm_media)) {
1916 	case IFM_AUTO:
1917 		CSR_CLRBIT_1(sc, VGE_DIAGCTL, VGE_DIAGCTL_MACFORCE);
1918 		CSR_CLRBIT_1(sc, VGE_DIAGCTL, VGE_DIAGCTL_FDXFORCE);
1919 		break;
1920 	case IFM_1000_T:
1921 		CSR_SETBIT_1(sc, VGE_DIAGCTL, VGE_DIAGCTL_MACFORCE);
1922 		CSR_CLRBIT_1(sc, VGE_DIAGCTL, VGE_DIAGCTL_FDXFORCE);
1923 		break;
1924 	case IFM_100_TX:
1925 	case IFM_10_T:
1926 		CSR_SETBIT_1(sc, VGE_DIAGCTL, VGE_DIAGCTL_MACFORCE);
1927 		if ((ife->ifm_media & IFM_GMASK) == IFM_FDX) {
1928 			CSR_SETBIT_1(sc, VGE_DIAGCTL, VGE_DIAGCTL_FDXFORCE);
1929 		} else {
1930 			CSR_CLRBIT_1(sc, VGE_DIAGCTL, VGE_DIAGCTL_FDXFORCE);
1931 		}
1932 		break;
1933 	default:
1934 		printf("%s: unknown media type: %x\n",
1935 		    device_xname(sc->sc_dev),
1936 		    IFM_SUBTYPE(ife->ifm_media));
1937 		break;
1938 	}
1939 }
1940 
1941 static int
1942 vge_ifflags_cb(struct ethercom *ec)
1943 {
1944 	struct ifnet *ifp = &ec->ec_if;
1945 	struct vge_softc *sc = ifp->if_softc;
1946 	int change = ifp->if_flags ^ sc->sc_if_flags;
1947 
1948 	if ((change & ~(IFF_CANTCHANGE|IFF_DEBUG)) != 0)
1949 		return ENETRESET;
1950 	else if ((change & IFF_PROMISC) == 0)
1951 		return 0;
1952 
1953 	if ((ifp->if_flags & IFF_PROMISC) == 0)
1954 		CSR_CLRBIT_1(sc, VGE_RXCTL, VGE_RXCTL_RX_PROMISC);
1955 	else
1956 		CSR_SETBIT_1(sc, VGE_RXCTL, VGE_RXCTL_RX_PROMISC);
1957 	vge_setmulti(sc);
1958 	return 0;
1959 }
1960 
1961 static int
1962 vge_ioctl(struct ifnet *ifp, u_long command, void *data)
1963 {
1964 	struct vge_softc *sc;
1965 	int s, error;
1966 
1967 	sc = ifp->if_softc;
1968 	error = 0;
1969 
1970 	s = splnet();
1971 
1972 	if ((error = ether_ioctl(ifp, command, data)) == ENETRESET) {
1973 		error = 0;
1974 		if (command != SIOCADDMULTI && command != SIOCDELMULTI)
1975 			;
1976 		else if (ifp->if_flags & IFF_RUNNING) {
1977 			/*
1978 			 * Multicast list has changed; set the hardware filter
1979 			 * accordingly.
1980 			 */
1981 			vge_setmulti(sc);
1982 		}
1983 	}
1984 	sc->sc_if_flags = ifp->if_flags;
1985 
1986 	splx(s);
1987 	return error;
1988 }
1989 
1990 static void
1991 vge_watchdog(struct ifnet *ifp)
1992 {
1993 	struct vge_softc *sc;
1994 	int s;
1995 
1996 	sc = ifp->if_softc;
1997 	s = splnet();
1998 	printf("%s: watchdog timeout\n", device_xname(sc->sc_dev));
1999 	ifp->if_oerrors++;
2000 
2001 	vge_txeof(sc);
2002 	vge_rxeof(sc);
2003 
2004 	vge_init(ifp);
2005 
2006 	splx(s);
2007 }
2008 
2009 /*
2010  * Stop the adapter and free any mbufs allocated to the
2011  * RX and TX lists.
2012  */
2013 static void
2014 vge_stop(struct ifnet *ifp, int disable)
2015 {
2016 	struct vge_softc *sc = ifp->if_softc;
2017 	struct vge_txsoft *txs;
2018 	struct vge_rxsoft *rxs;
2019 	int i, s;
2020 
2021 	s = splnet();
2022 	ifp->if_timer = 0;
2023 
2024 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
2025 #ifdef DEVICE_POLLING
2026 	ether_poll_deregister(ifp);
2027 #endif /* DEVICE_POLLING */
2028 
2029 	CSR_WRITE_1(sc, VGE_CRC3, VGE_CR3_INT_GMSK);
2030 	CSR_WRITE_1(sc, VGE_CRS0, VGE_CR0_STOP);
2031 	CSR_WRITE_4(sc, VGE_ISR, 0xFFFFFFFF);
2032 	CSR_WRITE_2(sc, VGE_TXQCSRC, 0xFFFF);
2033 	CSR_WRITE_1(sc, VGE_RXQCSRC, 0xFF);
2034 	CSR_WRITE_4(sc, VGE_RXDESC_ADDR_LO, 0);
2035 
2036 	if (sc->sc_rx_mhead != NULL) {
2037 		m_freem(sc->sc_rx_mhead);
2038 		sc->sc_rx_mhead = sc->sc_rx_mtail = NULL;
2039 	}
2040 
2041 	/* Free the TX list buffers. */
2042 
2043 	for (i = 0; i < VGE_NTXDESC; i++) {
2044 		txs = &sc->sc_txsoft[i];
2045 		if (txs->txs_mbuf != NULL) {
2046 			bus_dmamap_unload(sc->sc_dmat, txs->txs_dmamap);
2047 			m_freem(txs->txs_mbuf);
2048 			txs->txs_mbuf = NULL;
2049 		}
2050 	}
2051 
2052 	/* Free the RX list buffers. */
2053 
2054 	for (i = 0; i < VGE_NRXDESC; i++) {
2055 		rxs = &sc->sc_rxsoft[i];
2056 		if (rxs->rxs_mbuf != NULL) {
2057 			bus_dmamap_unload(sc->sc_dmat, rxs->rxs_dmamap);
2058 			m_freem(rxs->rxs_mbuf);
2059 			rxs->rxs_mbuf = NULL;
2060 		}
2061 	}
2062 
2063 	splx(s);
2064 }
2065 
2066 #if VGE_POWER_MANAGEMENT
2067 /*
2068  * Device suspend routine.  Stop the interface and save some PCI
2069  * settings in case the BIOS doesn't restore them properly on
2070  * resume.
2071  */
2072 static int
2073 vge_suspend(device_t dev)
2074 {
2075 	struct vge_softc *sc;
2076 	int i;
2077 
2078 	sc = device_get_softc(dev);
2079 
2080 	vge_stop(sc);
2081 
2082         for (i = 0; i < 5; i++)
2083 		sc->sc_saved_maps[i] =
2084 		    pci_read_config(dev, PCIR_MAPS + i * 4, 4);
2085 	sc->sc_saved_biosaddr = pci_read_config(dev, PCIR_BIOS, 4);
2086 	sc->sc_saved_intline = pci_read_config(dev, PCIR_INTLINE, 1);
2087 	sc->sc_saved_cachelnsz = pci_read_config(dev, PCIR_CACHELNSZ, 1);
2088 	sc->sc_saved_lattimer = pci_read_config(dev, PCIR_LATTIMER, 1);
2089 
2090 	sc->suspended = 1;
2091 
2092 	return 0;
2093 }
2094 
2095 /*
2096  * Device resume routine.  Restore some PCI settings in case the BIOS
2097  * doesn't, re-enable busmastering, and restart the interface if
2098  * appropriate.
2099  */
2100 static int
2101 vge_resume(device_t dev)
2102 {
2103 	struct vge_softc *sc;
2104 	struct ifnet *ifp;
2105 	int i;
2106 
2107 	sc = device_private(dev);
2108 	ifp = &sc->sc_ethercom.ec_if;
2109 
2110         /* better way to do this? */
2111 	for (i = 0; i < 5; i++)
2112 		pci_write_config(dev, PCIR_MAPS + i * 4,
2113 		    sc->sc_saved_maps[i], 4);
2114 	pci_write_config(dev, PCIR_BIOS, sc->sc_saved_biosaddr, 4);
2115 	pci_write_config(dev, PCIR_INTLINE, sc->sc_saved_intline, 1);
2116 	pci_write_config(dev, PCIR_CACHELNSZ, sc->sc_saved_cachelnsz, 1);
2117 	pci_write_config(dev, PCIR_LATTIMER, sc->sc_saved_lattimer, 1);
2118 
2119 	/* reenable busmastering */
2120 	pci_enable_busmaster(dev);
2121 	pci_enable_io(dev, SYS_RES_MEMORY);
2122 
2123 	/* reinitialize interface if necessary */
2124 	if (ifp->if_flags & IFF_UP)
2125 		vge_init(sc);
2126 
2127 	sc->suspended = 0;
2128 
2129 	return 0;
2130 }
2131 #endif
2132 
2133 /*
2134  * Stop all chip I/O so that the kernel's probe routines don't
2135  * get confused by errant DMAs when rebooting.
2136  */
2137 static bool
2138 vge_shutdown(device_t self, int howto)
2139 {
2140 	struct vge_softc *sc;
2141 
2142 	sc = device_private(self);
2143 	vge_stop(&sc->sc_ethercom.ec_if, 1);
2144 
2145 	return true;
2146 }
2147