xref: /netbsd-src/sys/dev/pci/if_dge.c (revision daf6c4152fcddc27c445489775ed1f66ab4ea9a9)
1 /*	$NetBSD: if_dge.c,v 1.31 2010/11/13 13:52:06 uebayasi Exp $ */
2 
3 /*
4  * Copyright (c) 2004, SUNET, Swedish University Computer Network.
5  * All rights reserved.
6  *
7  * Written by Anders Magnusson for SUNET, Swedish University Computer Network.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  * 3. All advertising materials mentioning features or use of this software
18  *    must display the following acknowledgement:
19  *	This product includes software developed for the NetBSD Project by
20  *	SUNET, Swedish University Computer Network.
21  * 4. The name of SUNET may not be used to endorse or promote products
22  *    derived from this software without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY SUNET ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
26  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
27  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL WASABI SYSTEMS, INC
28  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
29  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
30  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
31  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
32  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
33  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
34  * POSSIBILITY OF SUCH DAMAGE.
35  */
36 
37 /*
38  * Copyright (c) 2001, 2002, 2003 Wasabi Systems, Inc.
39  * All rights reserved.
40  *
41  * Written by Jason R. Thorpe for Wasabi Systems, Inc.
42  *
43  * Redistribution and use in source and binary forms, with or without
44  * modification, are permitted provided that the following conditions
45  * are met:
46  * 1. Redistributions of source code must retain the above copyright
47  *    notice, this list of conditions and the following disclaimer.
48  * 2. Redistributions in binary form must reproduce the above copyright
49  *    notice, this list of conditions and the following disclaimer in the
50  *    documentation and/or other materials provided with the distribution.
51  * 3. All advertising materials mentioning features or use of this software
52  *    must display the following acknowledgement:
53  *	This product includes software developed for the NetBSD Project by
54  *	Wasabi Systems, Inc.
55  * 4. The name of Wasabi Systems, Inc. may not be used to endorse
56  *    or promote products derived from this software without specific prior
57  *    written permission.
58  *
59  * THIS SOFTWARE IS PROVIDED BY WASABI SYSTEMS, INC. ``AS IS'' AND
60  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
61  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
62  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL WASABI SYSTEMS, INC
63  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
64  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
65  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
66  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
67  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
68  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
69  * POSSIBILITY OF SUCH DAMAGE.
70  */
71 
72 /*
73  * Device driver for the Intel 82597EX Ten Gigabit Ethernet controller.
74  *
75  * TODO (in no specific order):
76  *	HW VLAN support.
77  *	TSE offloading (needs kernel changes...)
78  *	RAIDC (receive interrupt delay adaptation)
79  *	Use memory > 4GB.
80  */
81 
82 #include <sys/cdefs.h>
83 __KERNEL_RCSID(0, "$NetBSD: if_dge.c,v 1.31 2010/11/13 13:52:06 uebayasi Exp $");
84 
85 #include "rnd.h"
86 
87 #include <sys/param.h>
88 #include <sys/systm.h>
89 #include <sys/callout.h>
90 #include <sys/mbuf.h>
91 #include <sys/malloc.h>
92 #include <sys/kernel.h>
93 #include <sys/socket.h>
94 #include <sys/ioctl.h>
95 #include <sys/errno.h>
96 #include <sys/device.h>
97 #include <sys/queue.h>
98 
99 #if NRND > 0
100 #include <sys/rnd.h>
101 #endif
102 
103 #include <net/if.h>
104 #include <net/if_dl.h>
105 #include <net/if_media.h>
106 #include <net/if_ether.h>
107 
108 #include <net/bpf.h>
109 
110 #include <netinet/in.h>			/* XXX for struct ip */
111 #include <netinet/in_systm.h>		/* XXX for struct ip */
112 #include <netinet/ip.h>			/* XXX for struct ip */
113 #include <netinet/tcp.h>		/* XXX for struct tcphdr */
114 
115 #include <sys/bus.h>
116 #include <sys/intr.h>
117 #include <machine/endian.h>
118 
119 #include <dev/mii/mii.h>
120 #include <dev/mii/miivar.h>
121 #include <dev/mii/mii_bitbang.h>
122 
123 #include <dev/pci/pcireg.h>
124 #include <dev/pci/pcivar.h>
125 #include <dev/pci/pcidevs.h>
126 
127 #include <dev/pci/if_dgereg.h>
128 
129 /*
130  * The receive engine may sometimes become off-by-one when writing back
131  * chained descriptors.	 Avoid this by allocating a large chunk of
132  * memory and use if instead (to avoid chained descriptors).
133  * This only happens with chained descriptors under heavy load.
134  */
135 #define DGE_OFFBYONE_RXBUG
136 
137 #define DGE_EVENT_COUNTERS
138 #define DGE_DEBUG
139 
140 #ifdef DGE_DEBUG
141 #define DGE_DEBUG_LINK		0x01
142 #define DGE_DEBUG_TX		0x02
143 #define DGE_DEBUG_RX		0x04
144 #define DGE_DEBUG_CKSUM		0x08
145 int	dge_debug = 0;
146 
147 #define DPRINTF(x, y)	if (dge_debug & (x)) printf y
148 #else
149 #define DPRINTF(x, y)	/* nothing */
150 #endif /* DGE_DEBUG */
151 
152 /*
153  * Transmit descriptor list size. We allow up to 100 DMA segments per
154  * packet (Intel reports of jumbo frame packets with as
155  * many as 80 DMA segments when using 16k buffers).
156  */
157 #define DGE_NTXSEGS		100
158 #define DGE_IFQUEUELEN		20000
159 #define DGE_TXQUEUELEN		2048
160 #define DGE_TXQUEUELEN_MASK	(DGE_TXQUEUELEN - 1)
161 #define DGE_TXQUEUE_GC		(DGE_TXQUEUELEN / 8)
162 #define DGE_NTXDESC		1024
163 #define DGE_NTXDESC_MASK		(DGE_NTXDESC - 1)
164 #define DGE_NEXTTX(x)		(((x) + 1) & DGE_NTXDESC_MASK)
165 #define DGE_NEXTTXS(x)		(((x) + 1) & DGE_TXQUEUELEN_MASK)
166 
167 /*
168  * Receive descriptor list size.
169  * Packet is of size MCLBYTES, and for jumbo packets buffers may
170  * be chained.	Due to the nature of the card (high-speed), keep this
171  * ring large. With 2k buffers the ring can store 400 jumbo packets,
172  * which at full speed will be received in just under 3ms.
173  */
174 #define DGE_NRXDESC		2048
175 #define DGE_NRXDESC_MASK	(DGE_NRXDESC - 1)
176 #define DGE_NEXTRX(x)		(((x) + 1) & DGE_NRXDESC_MASK)
177 /*
178  * # of descriptors between head and written descriptors.
179  * This is to work-around two erratas.
180  */
181 #define DGE_RXSPACE		10
182 #define DGE_PREVRX(x)		(((x) - DGE_RXSPACE) & DGE_NRXDESC_MASK)
183 /*
184  * Receive descriptor fetch threshholds. These are values recommended
185  * by Intel, do not touch them unless you know what you are doing.
186  */
187 #define RXDCTL_PTHRESH_VAL	128
188 #define RXDCTL_HTHRESH_VAL	16
189 #define RXDCTL_WTHRESH_VAL	16
190 
191 
192 /*
193  * Tweakable parameters; default values.
194  */
195 #define FCRTH	0x30000 /* Send XOFF water mark */
196 #define FCRTL	0x28000 /* Send XON water mark */
197 #define RDTR	0x20	/* Interrupt delay after receive, .8192us units */
198 #define TIDV	0x20	/* Interrupt delay after send, .8192us units */
199 
200 /*
201  * Control structures are DMA'd to the i82597 chip.  We allocate them in
202  * a single clump that maps to a single DMA segment to make serveral things
203  * easier.
204  */
205 struct dge_control_data {
206 	/*
207 	 * The transmit descriptors.
208 	 */
209 	struct dge_tdes wcd_txdescs[DGE_NTXDESC];
210 
211 	/*
212 	 * The receive descriptors.
213 	 */
214 	struct dge_rdes wcd_rxdescs[DGE_NRXDESC];
215 };
216 
217 #define DGE_CDOFF(x)	offsetof(struct dge_control_data, x)
218 #define DGE_CDTXOFF(x)	DGE_CDOFF(wcd_txdescs[(x)])
219 #define DGE_CDRXOFF(x)	DGE_CDOFF(wcd_rxdescs[(x)])
220 
221 /*
222  * The DGE interface have a higher max MTU size than normal jumbo frames.
223  */
224 #define DGE_MAX_MTU	16288	/* Max MTU size for this interface */
225 
226 /*
227  * Software state for transmit jobs.
228  */
229 struct dge_txsoft {
230 	struct mbuf *txs_mbuf;		/* head of our mbuf chain */
231 	bus_dmamap_t txs_dmamap;	/* our DMA map */
232 	int txs_firstdesc;		/* first descriptor in packet */
233 	int txs_lastdesc;		/* last descriptor in packet */
234 	int txs_ndesc;			/* # of descriptors used */
235 };
236 
237 /*
238  * Software state for receive buffers.	Each descriptor gets a
239  * 2k (MCLBYTES) buffer and a DMA map.	For packets which fill
240  * more than one buffer, we chain them together.
241  */
242 struct dge_rxsoft {
243 	struct mbuf *rxs_mbuf;		/* head of our mbuf chain */
244 	bus_dmamap_t rxs_dmamap;	/* our DMA map */
245 };
246 
247 /*
248  * Software state per device.
249  */
250 struct dge_softc {
251 	struct device sc_dev;		/* generic device information */
252 	bus_space_tag_t sc_st;		/* bus space tag */
253 	bus_space_handle_t sc_sh;	/* bus space handle */
254 	bus_dma_tag_t sc_dmat;		/* bus DMA tag */
255 	struct ethercom sc_ethercom;	/* ethernet common data */
256 
257 	int sc_flags;			/* flags; see below */
258 	int sc_bus_speed;		/* PCI/PCIX bus speed */
259 	int sc_pcix_offset;		/* PCIX capability register offset */
260 
261 	pci_chipset_tag_t sc_pc;
262 	pcitag_t sc_pt;
263 	int sc_mmrbc;			/* Max PCIX memory read byte count */
264 
265 	void *sc_ih;			/* interrupt cookie */
266 
267 	struct ifmedia sc_media;
268 
269 	bus_dmamap_t sc_cddmamap;	/* control data DMA map */
270 #define sc_cddma	sc_cddmamap->dm_segs[0].ds_addr
271 
272 	int		sc_align_tweak;
273 
274 	/*
275 	 * Software state for the transmit and receive descriptors.
276 	 */
277 	struct dge_txsoft sc_txsoft[DGE_TXQUEUELEN];
278 	struct dge_rxsoft sc_rxsoft[DGE_NRXDESC];
279 
280 	/*
281 	 * Control data structures.
282 	 */
283 	struct dge_control_data *sc_control_data;
284 #define sc_txdescs	sc_control_data->wcd_txdescs
285 #define sc_rxdescs	sc_control_data->wcd_rxdescs
286 
287 #ifdef DGE_EVENT_COUNTERS
288 	/* Event counters. */
289 	struct evcnt sc_ev_txsstall;	/* Tx stalled due to no txs */
290 	struct evcnt sc_ev_txdstall;	/* Tx stalled due to no txd */
291 	struct evcnt sc_ev_txforceintr; /* Tx interrupts forced */
292 	struct evcnt sc_ev_txdw;	/* Tx descriptor interrupts */
293 	struct evcnt sc_ev_txqe;	/* Tx queue empty interrupts */
294 	struct evcnt sc_ev_rxintr;	/* Rx interrupts */
295 	struct evcnt sc_ev_linkintr;	/* Link interrupts */
296 
297 	struct evcnt sc_ev_rxipsum;	/* IP checksums checked in-bound */
298 	struct evcnt sc_ev_rxtusum;	/* TCP/UDP cksums checked in-bound */
299 	struct evcnt sc_ev_txipsum;	/* IP checksums comp. out-bound */
300 	struct evcnt sc_ev_txtusum;	/* TCP/UDP cksums comp. out-bound */
301 
302 	struct evcnt sc_ev_txctx_init;	/* Tx cksum context cache initialized */
303 	struct evcnt sc_ev_txctx_hit;	/* Tx cksum context cache hit */
304 	struct evcnt sc_ev_txctx_miss;	/* Tx cksum context cache miss */
305 
306 	struct evcnt sc_ev_txseg[DGE_NTXSEGS]; /* Tx packets w/ N segments */
307 	struct evcnt sc_ev_txdrop;	/* Tx packets dropped (too many segs) */
308 #endif /* DGE_EVENT_COUNTERS */
309 
310 	int	sc_txfree;		/* number of free Tx descriptors */
311 	int	sc_txnext;		/* next ready Tx descriptor */
312 
313 	int	sc_txsfree;		/* number of free Tx jobs */
314 	int	sc_txsnext;		/* next free Tx job */
315 	int	sc_txsdirty;		/* dirty Tx jobs */
316 
317 	uint32_t sc_txctx_ipcs;		/* cached Tx IP cksum ctx */
318 	uint32_t sc_txctx_tucs;		/* cached Tx TCP/UDP cksum ctx */
319 
320 	int	sc_rxptr;		/* next ready Rx descriptor/queue ent */
321 	int	sc_rxdiscard;
322 	int	sc_rxlen;
323 	struct mbuf *sc_rxhead;
324 	struct mbuf *sc_rxtail;
325 	struct mbuf **sc_rxtailp;
326 
327 	uint32_t sc_ctrl0;		/* prototype CTRL0 register */
328 	uint32_t sc_icr;		/* prototype interrupt bits */
329 	uint32_t sc_tctl;		/* prototype TCTL register */
330 	uint32_t sc_rctl;		/* prototype RCTL register */
331 
332 	int sc_mchash_type;		/* multicast filter offset */
333 
334 	uint16_t sc_eeprom[EEPROM_SIZE];
335 
336 #if NRND > 0
337 	rndsource_element_t rnd_source; /* random source */
338 #endif
339 #ifdef DGE_OFFBYONE_RXBUG
340 	void *sc_bugbuf;
341 	SLIST_HEAD(, rxbugentry) sc_buglist;
342 	bus_dmamap_t sc_bugmap;
343 	struct rxbugentry *sc_entry;
344 #endif
345 };
346 
347 #define DGE_RXCHAIN_RESET(sc)						\
348 do {									\
349 	(sc)->sc_rxtailp = &(sc)->sc_rxhead;				\
350 	*(sc)->sc_rxtailp = NULL;					\
351 	(sc)->sc_rxlen = 0;						\
352 } while (/*CONSTCOND*/0)
353 
354 #define DGE_RXCHAIN_LINK(sc, m)						\
355 do {									\
356 	*(sc)->sc_rxtailp = (sc)->sc_rxtail = (m);			\
357 	(sc)->sc_rxtailp = &(m)->m_next;				\
358 } while (/*CONSTCOND*/0)
359 
360 /* sc_flags */
361 #define DGE_F_BUS64		0x20	/* bus is 64-bit */
362 #define DGE_F_PCIX		0x40	/* bus is PCI-X */
363 
364 #ifdef DGE_EVENT_COUNTERS
365 #define DGE_EVCNT_INCR(ev)	(ev)->ev_count++
366 #else
367 #define DGE_EVCNT_INCR(ev)	/* nothing */
368 #endif
369 
370 #define CSR_READ(sc, reg)						\
371 	bus_space_read_4((sc)->sc_st, (sc)->sc_sh, (reg))
372 #define CSR_WRITE(sc, reg, val)						\
373 	bus_space_write_4((sc)->sc_st, (sc)->sc_sh, (reg), (val))
374 
375 #define DGE_CDTXADDR(sc, x)	((sc)->sc_cddma + DGE_CDTXOFF((x)))
376 #define DGE_CDRXADDR(sc, x)	((sc)->sc_cddma + DGE_CDRXOFF((x)))
377 
378 #define DGE_CDTXSYNC(sc, x, n, ops)					\
379 do {									\
380 	int __x, __n;							\
381 									\
382 	__x = (x);							\
383 	__n = (n);							\
384 									\
385 	/* If it will wrap around, sync to the end of the ring. */	\
386 	if ((__x + __n) > DGE_NTXDESC) {				\
387 		bus_dmamap_sync((sc)->sc_dmat, (sc)->sc_cddmamap,	\
388 		    DGE_CDTXOFF(__x), sizeof(struct dge_tdes) *		\
389 		    (DGE_NTXDESC - __x), (ops));			\
390 		__n -= (DGE_NTXDESC - __x);				\
391 		__x = 0;						\
392 	}								\
393 									\
394 	/* Now sync whatever is left. */				\
395 	bus_dmamap_sync((sc)->sc_dmat, (sc)->sc_cddmamap,		\
396 	    DGE_CDTXOFF(__x), sizeof(struct dge_tdes) * __n, (ops));	\
397 } while (/*CONSTCOND*/0)
398 
399 #define DGE_CDRXSYNC(sc, x, ops)						\
400 do {									\
401 	bus_dmamap_sync((sc)->sc_dmat, (sc)->sc_cddmamap,		\
402 	   DGE_CDRXOFF((x)), sizeof(struct dge_rdes), (ops));		\
403 } while (/*CONSTCOND*/0)
404 
405 #ifdef DGE_OFFBYONE_RXBUG
406 #define DGE_INIT_RXDESC(sc, x)						\
407 do {									\
408 	struct dge_rxsoft *__rxs = &(sc)->sc_rxsoft[(x)];		\
409 	struct dge_rdes *__rxd = &(sc)->sc_rxdescs[(x)];		\
410 	struct mbuf *__m = __rxs->rxs_mbuf;				\
411 									\
412 	__rxd->dr_baddrl = htole32(sc->sc_bugmap->dm_segs[0].ds_addr +	\
413 	    (mtod((__m), char *) - (char *)sc->sc_bugbuf));		\
414 	__rxd->dr_baddrh = 0;						\
415 	__rxd->dr_len = 0;						\
416 	__rxd->dr_cksum = 0;						\
417 	__rxd->dr_status = 0;						\
418 	__rxd->dr_errors = 0;						\
419 	__rxd->dr_special = 0;						\
420 	DGE_CDRXSYNC((sc), (x), BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE); \
421 									\
422 	CSR_WRITE((sc), DGE_RDT, (x));					\
423 } while (/*CONSTCOND*/0)
424 #else
425 #define DGE_INIT_RXDESC(sc, x)						\
426 do {									\
427 	struct dge_rxsoft *__rxs = &(sc)->sc_rxsoft[(x)];		\
428 	struct dge_rdes *__rxd = &(sc)->sc_rxdescs[(x)];		\
429 	struct mbuf *__m = __rxs->rxs_mbuf;				\
430 									\
431 	/*								\
432 	 * Note: We scoot the packet forward 2 bytes in the buffer	\
433 	 * so that the payload after the Ethernet header is aligned	\
434 	 * to a 4-byte boundary.					\
435 	 *								\
436 	 * XXX BRAINDAMAGE ALERT!					\
437 	 * The stupid chip uses the same size for every buffer, which	\
438 	 * is set in the Receive Control register.  We are using the 2K \
439 	 * size option, but what we REALLY want is (2K - 2)!  For this	\
440 	 * reason, we can't "scoot" packets longer than the standard	\
441 	 * Ethernet MTU.  On strict-alignment platforms, if the total	\
442 	 * size exceeds (2K - 2) we set align_tweak to 0 and let	\
443 	 * the upper layer copy the headers.				\
444 	 */								\
445 	__m->m_data = __m->m_ext.ext_buf + (sc)->sc_align_tweak;	\
446 									\
447 	__rxd->dr_baddrl =						\
448 	    htole32(__rxs->rxs_dmamap->dm_segs[0].ds_addr +		\
449 		(sc)->sc_align_tweak);					\
450 	__rxd->dr_baddrh = 0;						\
451 	__rxd->dr_len = 0;						\
452 	__rxd->dr_cksum = 0;						\
453 	__rxd->dr_status = 0;						\
454 	__rxd->dr_errors = 0;						\
455 	__rxd->dr_special = 0;						\
456 	DGE_CDRXSYNC((sc), (x), BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE); \
457 									\
458 	CSR_WRITE((sc), DGE_RDT, (x));					\
459 } while (/*CONSTCOND*/0)
460 #endif
461 
462 #ifdef DGE_OFFBYONE_RXBUG
463 /*
464  * Allocation constants.  Much memory may be used for this.
465  */
466 #ifndef DGE_BUFFER_SIZE
467 #define DGE_BUFFER_SIZE DGE_MAX_MTU
468 #endif
469 #define DGE_NBUFFERS	(4*DGE_NRXDESC)
470 #define DGE_RXMEM	(DGE_NBUFFERS*DGE_BUFFER_SIZE)
471 
472 struct rxbugentry {
473 	SLIST_ENTRY(rxbugentry) rb_entry;
474 	int rb_slot;
475 };
476 
477 static int
478 dge_alloc_rcvmem(struct dge_softc *sc)
479 {
480 	char *ptr, *kva;
481 	bus_dma_segment_t seg;
482 	int i, rseg, state, error;
483 	struct rxbugentry *entry;
484 
485 	state = error = 0;
486 
487 	if (bus_dmamem_alloc(sc->sc_dmat, DGE_RXMEM, PAGE_SIZE, 0,
488 	     &seg, 1, &rseg, BUS_DMA_NOWAIT)) {
489 		aprint_error_dev(&sc->sc_dev, "can't alloc rx buffers\n");
490 		return ENOBUFS;
491 	}
492 
493 	state = 1;
494 	if (bus_dmamem_map(sc->sc_dmat, &seg, rseg, DGE_RXMEM, (void **)&kva,
495 	    BUS_DMA_NOWAIT)) {
496 		aprint_error_dev(&sc->sc_dev, "can't map DMA buffers (%d bytes)\n",
497 		    (int)DGE_RXMEM);
498 		error = ENOBUFS;
499 		goto out;
500 	}
501 
502 	state = 2;
503 	if (bus_dmamap_create(sc->sc_dmat, DGE_RXMEM, 1, DGE_RXMEM, 0,
504 	    BUS_DMA_NOWAIT, &sc->sc_bugmap)) {
505 		aprint_error_dev(&sc->sc_dev, "can't create DMA map\n");
506 		error = ENOBUFS;
507 		goto out;
508 	}
509 
510 	state = 3;
511 	if (bus_dmamap_load(sc->sc_dmat, sc->sc_bugmap,
512 	    kva, DGE_RXMEM, NULL, BUS_DMA_NOWAIT)) {
513 		aprint_error_dev(&sc->sc_dev, "can't load DMA map\n");
514 		error = ENOBUFS;
515 		goto out;
516 	}
517 
518 	state = 4;
519 	sc->sc_bugbuf = (void *)kva;
520 	SLIST_INIT(&sc->sc_buglist);
521 
522 	/*
523 	 * Now divide it up into DGE_BUFFER_SIZE pieces and save the addresses
524 	 * in an array.
525 	 */
526 	ptr = sc->sc_bugbuf;
527 	if ((entry = malloc(sizeof(*entry) * DGE_NBUFFERS,
528 	    M_DEVBUF, M_NOWAIT)) == NULL) {
529 		error = ENOBUFS;
530 		goto out;
531 	}
532 	sc->sc_entry = entry;
533 	for (i = 0; i < DGE_NBUFFERS; i++) {
534 		entry[i].rb_slot = i;
535 		SLIST_INSERT_HEAD(&sc->sc_buglist, &entry[i], rb_entry);
536 	}
537 out:
538 	if (error != 0) {
539 		switch (state) {
540 		case 4:
541 			bus_dmamap_unload(sc->sc_dmat, sc->sc_bugmap);
542 		case 3:
543 			bus_dmamap_destroy(sc->sc_dmat, sc->sc_bugmap);
544 		case 2:
545 			bus_dmamem_unmap(sc->sc_dmat, kva, DGE_RXMEM);
546 		case 1:
547 			bus_dmamem_free(sc->sc_dmat, &seg, rseg);
548 			break;
549 		default:
550 			break;
551 		}
552 	}
553 
554 	return error;
555 }
556 
557 /*
558  * Allocate a jumbo buffer.
559  */
560 static void *
561 dge_getbuf(struct dge_softc *sc)
562 {
563 	struct rxbugentry *entry;
564 
565 	entry = SLIST_FIRST(&sc->sc_buglist);
566 
567 	if (entry == NULL) {
568 		printf("%s: no free RX buffers\n", device_xname(&sc->sc_dev));
569 		return(NULL);
570 	}
571 
572 	SLIST_REMOVE_HEAD(&sc->sc_buglist, rb_entry);
573 	return (char *)sc->sc_bugbuf + entry->rb_slot * DGE_BUFFER_SIZE;
574 }
575 
576 /*
577  * Release a jumbo buffer.
578  */
579 static void
580 dge_freebuf(struct mbuf *m, void *buf, size_t size, void *arg)
581 {
582 	struct rxbugentry *entry;
583 	struct dge_softc *sc;
584 	int i, s;
585 
586 	/* Extract the softc struct pointer. */
587 	sc = (struct dge_softc *)arg;
588 
589 	if (sc == NULL)
590 		panic("dge_freebuf: can't find softc pointer!");
591 
592 	/* calculate the slot this buffer belongs to */
593 
594 	i = ((char *)buf - (char *)sc->sc_bugbuf) / DGE_BUFFER_SIZE;
595 
596 	if ((i < 0) || (i >= DGE_NBUFFERS))
597 		panic("dge_freebuf: asked to free buffer %d!", i);
598 
599 	s = splvm();
600 	entry = sc->sc_entry + i;
601 	SLIST_INSERT_HEAD(&sc->sc_buglist, entry, rb_entry);
602 
603 	if (__predict_true(m != NULL))
604 		pool_cache_put(mb_cache, m);
605 	splx(s);
606 }
607 #endif
608 
609 static void	dge_start(struct ifnet *);
610 static void	dge_watchdog(struct ifnet *);
611 static int	dge_ioctl(struct ifnet *, u_long, void *);
612 static int	dge_init(struct ifnet *);
613 static void	dge_stop(struct ifnet *, int);
614 
615 static bool	dge_shutdown(device_t, int);
616 
617 static void	dge_reset(struct dge_softc *);
618 static void	dge_rxdrain(struct dge_softc *);
619 static int	dge_add_rxbuf(struct dge_softc *, int);
620 
621 static void	dge_set_filter(struct dge_softc *);
622 
623 static int	dge_intr(void *);
624 static void	dge_txintr(struct dge_softc *);
625 static void	dge_rxintr(struct dge_softc *);
626 static void	dge_linkintr(struct dge_softc *, uint32_t);
627 
628 static int	dge_match(device_t, cfdata_t, void *);
629 static void	dge_attach(device_t, device_t, void *);
630 
631 static int	dge_read_eeprom(struct dge_softc *sc);
632 static int	dge_eeprom_clockin(struct dge_softc *sc);
633 static void	dge_eeprom_clockout(struct dge_softc *sc, int bit);
634 static uint16_t	dge_eeprom_word(struct dge_softc *sc, int addr);
635 static int	dge_xgmii_mediachange(struct ifnet *);
636 static void	dge_xgmii_mediastatus(struct ifnet *, struct ifmediareq *);
637 static void	dge_xgmii_reset(struct dge_softc *);
638 static void	dge_xgmii_writereg(device_t, int, int, int);
639 
640 
641 CFATTACH_DECL(dge, sizeof(struct dge_softc),
642     dge_match, dge_attach, NULL, NULL);
643 
644 #ifdef DGE_EVENT_COUNTERS
645 #if DGE_NTXSEGS > 100
646 #error Update dge_txseg_evcnt_names
647 #endif
648 static char (*dge_txseg_evcnt_names)[DGE_NTXSEGS][8 /* "txseg00" + \0 */];
649 #endif /* DGE_EVENT_COUNTERS */
650 
651 static int
652 dge_match(device_t parent, cfdata_t cf, void *aux)
653 {
654 	struct pci_attach_args *pa = aux;
655 
656 	if (PCI_VENDOR(pa->pa_id) == PCI_VENDOR_INTEL &&
657 	    PCI_PRODUCT(pa->pa_id) == PCI_PRODUCT_INTEL_82597EX)
658 		return (1);
659 
660 	return (0);
661 }
662 
663 static void
664 dge_attach(device_t parent, device_t self, void *aux)
665 {
666 	struct dge_softc *sc = device_private(self);
667 	struct pci_attach_args *pa = aux;
668 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
669 	pci_chipset_tag_t pc = pa->pa_pc;
670 	pci_intr_handle_t ih;
671 	const char *intrstr = NULL;
672 	bus_dma_segment_t seg;
673 	int i, rseg, error;
674 	uint8_t enaddr[ETHER_ADDR_LEN];
675 	pcireg_t preg, memtype;
676 	uint32_t reg;
677 
678 	sc->sc_dmat = pa->pa_dmat;
679 	sc->sc_pc = pa->pa_pc;
680 	sc->sc_pt = pa->pa_tag;
681 
682 	preg = PCI_REVISION(pci_conf_read(pc, pa->pa_tag, PCI_CLASS_REG));
683 	aprint_naive(": Ethernet controller\n");
684 	aprint_normal(": Intel i82597EX 10GbE-LR Ethernet, rev. %d\n", preg);
685 
686 	memtype = pci_mapreg_type(pa->pa_pc, pa->pa_tag, DGE_PCI_BAR);
687         if (pci_mapreg_map(pa, DGE_PCI_BAR, memtype, 0,
688             &sc->sc_st, &sc->sc_sh, NULL, NULL)) {
689                 aprint_error_dev(&sc->sc_dev, "unable to map device registers\n");
690                 return;
691         }
692 
693 	/* Enable bus mastering */
694 	preg = pci_conf_read(pc, pa->pa_tag, PCI_COMMAND_STATUS_REG);
695 	preg |= PCI_COMMAND_MASTER_ENABLE;
696 	pci_conf_write(pc, pa->pa_tag, PCI_COMMAND_STATUS_REG, preg);
697 
698 	/*
699 	 * Map and establish our interrupt.
700 	 */
701 	if (pci_intr_map(pa, &ih)) {
702 		aprint_error_dev(&sc->sc_dev, "unable to map interrupt\n");
703 		return;
704 	}
705 	intrstr = pci_intr_string(pc, ih);
706 	sc->sc_ih = pci_intr_establish(pc, ih, IPL_NET, dge_intr, sc);
707 	if (sc->sc_ih == NULL) {
708 		aprint_error_dev(&sc->sc_dev, "unable to establish interrupt");
709 		if (intrstr != NULL)
710 			aprint_error(" at %s", intrstr);
711 		aprint_error("\n");
712 		return;
713 	}
714 	aprint_normal_dev(&sc->sc_dev, "interrupting at %s\n", intrstr);
715 
716 	/*
717 	 * Determine a few things about the bus we're connected to.
718 	 */
719 	reg = CSR_READ(sc, DGE_STATUS);
720 	if (reg & STATUS_BUS64)
721 		sc->sc_flags |= DGE_F_BUS64;
722 
723 	sc->sc_flags |= DGE_F_PCIX;
724 	if (pci_get_capability(pa->pa_pc, pa->pa_tag,
725 			       PCI_CAP_PCIX,
726 			       &sc->sc_pcix_offset, NULL) == 0)
727 		aprint_error_dev(&sc->sc_dev, "unable to find PCIX "
728 		    "capability\n");
729 
730 	if (sc->sc_flags & DGE_F_PCIX) {
731 		switch (reg & STATUS_PCIX_MSK) {
732 		case STATUS_PCIX_66:
733 			sc->sc_bus_speed = 66;
734 			break;
735 		case STATUS_PCIX_100:
736 			sc->sc_bus_speed = 100;
737 			break;
738 		case STATUS_PCIX_133:
739 			sc->sc_bus_speed = 133;
740 			break;
741 		default:
742 			aprint_error_dev(&sc->sc_dev,
743 			    "unknown PCIXSPD %d; assuming 66MHz\n",
744 			    reg & STATUS_PCIX_MSK);
745 			sc->sc_bus_speed = 66;
746 		}
747 	} else
748 		sc->sc_bus_speed = (reg & STATUS_BUS64) ? 66 : 33;
749 	aprint_verbose_dev(&sc->sc_dev, "%d-bit %dMHz %s bus\n",
750 	    (sc->sc_flags & DGE_F_BUS64) ? 64 : 32, sc->sc_bus_speed,
751 	    (sc->sc_flags & DGE_F_PCIX) ? "PCIX" : "PCI");
752 
753 	/*
754 	 * Allocate the control data structures, and create and load the
755 	 * DMA map for it.
756 	 */
757 	if ((error = bus_dmamem_alloc(sc->sc_dmat,
758 	    sizeof(struct dge_control_data), PAGE_SIZE, 0, &seg, 1, &rseg,
759 	    0)) != 0) {
760 		aprint_error_dev(&sc->sc_dev,
761 		    "unable to allocate control data, error = %d\n",
762 		    error);
763 		goto fail_0;
764 	}
765 
766 	if ((error = bus_dmamem_map(sc->sc_dmat, &seg, rseg,
767 	    sizeof(struct dge_control_data), (void **)&sc->sc_control_data,
768 	    0)) != 0) {
769 		aprint_error_dev(&sc->sc_dev, "unable to map control data, error = %d\n",
770 		    error);
771 		goto fail_1;
772 	}
773 
774 	if ((error = bus_dmamap_create(sc->sc_dmat,
775 	    sizeof(struct dge_control_data), 1,
776 	    sizeof(struct dge_control_data), 0, 0, &sc->sc_cddmamap)) != 0) {
777 		aprint_error_dev(&sc->sc_dev, "unable to create control data DMA map, "
778 		    "error = %d\n", error);
779 		goto fail_2;
780 	}
781 
782 	if ((error = bus_dmamap_load(sc->sc_dmat, sc->sc_cddmamap,
783 	    sc->sc_control_data, sizeof(struct dge_control_data), NULL,
784 	    0)) != 0) {
785 		aprint_error_dev(&sc->sc_dev,
786 		    "unable to load control data DMA map, error = %d\n",
787 		    error);
788 		goto fail_3;
789 	}
790 
791 #ifdef DGE_OFFBYONE_RXBUG
792 	if (dge_alloc_rcvmem(sc) != 0)
793 		return; /* Already complained */
794 #endif
795 	/*
796 	 * Create the transmit buffer DMA maps.
797 	 */
798 	for (i = 0; i < DGE_TXQUEUELEN; i++) {
799 		if ((error = bus_dmamap_create(sc->sc_dmat, DGE_MAX_MTU,
800 		    DGE_NTXSEGS, MCLBYTES, 0, 0,
801 		    &sc->sc_txsoft[i].txs_dmamap)) != 0) {
802 			aprint_error_dev(&sc->sc_dev, "unable to create Tx DMA map %d, "
803 			    "error = %d\n", i, error);
804 			goto fail_4;
805 		}
806 	}
807 
808 	/*
809 	 * Create the receive buffer DMA maps.
810 	 */
811 	for (i = 0; i < DGE_NRXDESC; i++) {
812 #ifdef DGE_OFFBYONE_RXBUG
813 		if ((error = bus_dmamap_create(sc->sc_dmat, DGE_BUFFER_SIZE, 1,
814 		    DGE_BUFFER_SIZE, 0, 0, &sc->sc_rxsoft[i].rxs_dmamap)) != 0) {
815 #else
816 		if ((error = bus_dmamap_create(sc->sc_dmat, MCLBYTES, 1,
817 		    MCLBYTES, 0, 0, &sc->sc_rxsoft[i].rxs_dmamap)) != 0) {
818 #endif
819 			aprint_error_dev(&sc->sc_dev, "unable to create Rx DMA map %d, "
820 			    "error = %d\n", i, error);
821 			goto fail_5;
822 		}
823 		sc->sc_rxsoft[i].rxs_mbuf = NULL;
824 	}
825 
826 	/*
827 	 * Set bits in ctrl0 register.
828 	 * Should get the software defined pins out of EEPROM?
829 	 */
830 	sc->sc_ctrl0 |= CTRL0_RPE | CTRL0_TPE; /* XON/XOFF */
831 	sc->sc_ctrl0 |= CTRL0_SDP3_DIR | CTRL0_SDP2_DIR | CTRL0_SDP1_DIR |
832 	    CTRL0_SDP0_DIR | CTRL0_SDP3 | CTRL0_SDP2 | CTRL0_SDP0;
833 
834 	/*
835 	 * Reset the chip to a known state.
836 	 */
837 	dge_reset(sc);
838 
839 	/*
840 	 * Reset the PHY.
841 	 */
842 	dge_xgmii_reset(sc);
843 
844 	/*
845 	 * Read in EEPROM data.
846 	 */
847 	if (dge_read_eeprom(sc)) {
848 		aprint_error_dev(&sc->sc_dev, "couldn't read EEPROM\n");
849 		return;
850 	}
851 
852 	/*
853 	 * Get the ethernet address.
854 	 */
855 	enaddr[0] = sc->sc_eeprom[EE_ADDR01] & 0377;
856 	enaddr[1] = sc->sc_eeprom[EE_ADDR01] >> 8;
857 	enaddr[2] = sc->sc_eeprom[EE_ADDR23] & 0377;
858 	enaddr[3] = sc->sc_eeprom[EE_ADDR23] >> 8;
859 	enaddr[4] = sc->sc_eeprom[EE_ADDR45] & 0377;
860 	enaddr[5] = sc->sc_eeprom[EE_ADDR45] >> 8;
861 
862 	aprint_normal_dev(&sc->sc_dev, "Ethernet address %s\n",
863 	    ether_sprintf(enaddr));
864 
865 	/*
866 	 * Setup media stuff.
867 	 */
868         ifmedia_init(&sc->sc_media, IFM_IMASK, dge_xgmii_mediachange,
869             dge_xgmii_mediastatus);
870         ifmedia_add(&sc->sc_media, IFM_ETHER|IFM_10G_LR, 0, NULL);
871         ifmedia_set(&sc->sc_media, IFM_ETHER|IFM_10G_LR);
872 
873 	ifp = &sc->sc_ethercom.ec_if;
874 	strlcpy(ifp->if_xname, device_xname(&sc->sc_dev), IFNAMSIZ);
875 	ifp->if_softc = sc;
876 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
877 	ifp->if_ioctl = dge_ioctl;
878 	ifp->if_start = dge_start;
879 	ifp->if_watchdog = dge_watchdog;
880 	ifp->if_init = dge_init;
881 	ifp->if_stop = dge_stop;
882 	IFQ_SET_MAXLEN(&ifp->if_snd, max(DGE_IFQUEUELEN, IFQ_MAXLEN));
883 	IFQ_SET_READY(&ifp->if_snd);
884 
885 	sc->sc_ethercom.ec_capabilities |=
886 	    ETHERCAP_JUMBO_MTU | ETHERCAP_VLAN_MTU;
887 
888 	/*
889 	 * We can perform TCPv4 and UDPv4 checkums in-bound.
890 	 */
891 	ifp->if_capabilities |=
892 	    IFCAP_CSUM_IPv4_Tx | IFCAP_CSUM_IPv4_Rx |
893 	    IFCAP_CSUM_TCPv4_Tx | IFCAP_CSUM_TCPv4_Rx |
894 	    IFCAP_CSUM_UDPv4_Tx | IFCAP_CSUM_UDPv4_Rx;
895 
896 	/*
897 	 * Attach the interface.
898 	 */
899 	if_attach(ifp);
900 	ether_ifattach(ifp, enaddr);
901 #if NRND > 0
902 	rnd_attach_source(&sc->rnd_source, device_xname(&sc->sc_dev),
903 	    RND_TYPE_NET, 0);
904 #endif
905 
906 #ifdef DGE_EVENT_COUNTERS
907 	/* Fix segment event naming */
908 	if (dge_txseg_evcnt_names == NULL) {
909 		dge_txseg_evcnt_names =
910 		    malloc(sizeof(*dge_txseg_evcnt_names), M_DEVBUF, M_WAITOK);
911 		for (i = 0; i < DGE_NTXSEGS; i++)
912 			snprintf((*dge_txseg_evcnt_names)[i],
913 			    sizeof((*dge_txseg_evcnt_names)[i]), "txseg%d", i);
914 	}
915 
916 	/* Attach event counters. */
917 	evcnt_attach_dynamic(&sc->sc_ev_txsstall, EVCNT_TYPE_MISC,
918 	    NULL, device_xname(&sc->sc_dev), "txsstall");
919 	evcnt_attach_dynamic(&sc->sc_ev_txdstall, EVCNT_TYPE_MISC,
920 	    NULL, device_xname(&sc->sc_dev), "txdstall");
921 	evcnt_attach_dynamic(&sc->sc_ev_txforceintr, EVCNT_TYPE_MISC,
922 	    NULL, device_xname(&sc->sc_dev), "txforceintr");
923 	evcnt_attach_dynamic(&sc->sc_ev_txdw, EVCNT_TYPE_INTR,
924 	    NULL, device_xname(&sc->sc_dev), "txdw");
925 	evcnt_attach_dynamic(&sc->sc_ev_txqe, EVCNT_TYPE_INTR,
926 	    NULL, device_xname(&sc->sc_dev), "txqe");
927 	evcnt_attach_dynamic(&sc->sc_ev_rxintr, EVCNT_TYPE_INTR,
928 	    NULL, device_xname(&sc->sc_dev), "rxintr");
929 	evcnt_attach_dynamic(&sc->sc_ev_linkintr, EVCNT_TYPE_INTR,
930 	    NULL, device_xname(&sc->sc_dev), "linkintr");
931 
932 	evcnt_attach_dynamic(&sc->sc_ev_rxipsum, EVCNT_TYPE_MISC,
933 	    NULL, device_xname(&sc->sc_dev), "rxipsum");
934 	evcnt_attach_dynamic(&sc->sc_ev_rxtusum, EVCNT_TYPE_MISC,
935 	    NULL, device_xname(&sc->sc_dev), "rxtusum");
936 	evcnt_attach_dynamic(&sc->sc_ev_txipsum, EVCNT_TYPE_MISC,
937 	    NULL, device_xname(&sc->sc_dev), "txipsum");
938 	evcnt_attach_dynamic(&sc->sc_ev_txtusum, EVCNT_TYPE_MISC,
939 	    NULL, device_xname(&sc->sc_dev), "txtusum");
940 
941 	evcnt_attach_dynamic(&sc->sc_ev_txctx_init, EVCNT_TYPE_MISC,
942 	    NULL, device_xname(&sc->sc_dev), "txctx init");
943 	evcnt_attach_dynamic(&sc->sc_ev_txctx_hit, EVCNT_TYPE_MISC,
944 	    NULL, device_xname(&sc->sc_dev), "txctx hit");
945 	evcnt_attach_dynamic(&sc->sc_ev_txctx_miss, EVCNT_TYPE_MISC,
946 	    NULL, device_xname(&sc->sc_dev), "txctx miss");
947 
948 	for (i = 0; i < DGE_NTXSEGS; i++)
949 		evcnt_attach_dynamic(&sc->sc_ev_txseg[i], EVCNT_TYPE_MISC,
950 		    NULL, device_xname(&sc->sc_dev), (*dge_txseg_evcnt_names)[i]);
951 
952 	evcnt_attach_dynamic(&sc->sc_ev_txdrop, EVCNT_TYPE_MISC,
953 	    NULL, device_xname(&sc->sc_dev), "txdrop");
954 
955 #endif /* DGE_EVENT_COUNTERS */
956 
957 	/*
958 	 * Make sure the interface is shutdown during reboot.
959 	 */
960 	if (pmf_device_register1(self, NULL, NULL, dge_shutdown))
961 		pmf_class_network_register(self, ifp);
962 	else
963 		aprint_error_dev(self, "couldn't establish power handler\n");
964 
965 	return;
966 
967 	/*
968 	 * Free any resources we've allocated during the failed attach
969 	 * attempt.  Do this in reverse order and fall through.
970 	 */
971  fail_5:
972 	for (i = 0; i < DGE_NRXDESC; i++) {
973 		if (sc->sc_rxsoft[i].rxs_dmamap != NULL)
974 			bus_dmamap_destroy(sc->sc_dmat,
975 			    sc->sc_rxsoft[i].rxs_dmamap);
976 	}
977  fail_4:
978 	for (i = 0; i < DGE_TXQUEUELEN; i++) {
979 		if (sc->sc_txsoft[i].txs_dmamap != NULL)
980 			bus_dmamap_destroy(sc->sc_dmat,
981 			    sc->sc_txsoft[i].txs_dmamap);
982 	}
983 	bus_dmamap_unload(sc->sc_dmat, sc->sc_cddmamap);
984  fail_3:
985 	bus_dmamap_destroy(sc->sc_dmat, sc->sc_cddmamap);
986  fail_2:
987 	bus_dmamem_unmap(sc->sc_dmat, (void *)sc->sc_control_data,
988 	    sizeof(struct dge_control_data));
989  fail_1:
990 	bus_dmamem_free(sc->sc_dmat, &seg, rseg);
991  fail_0:
992 	return;
993 }
994 
995 /*
996  * dge_shutdown:
997  *
998  *	Make sure the interface is stopped at reboot time.
999  */
1000 static bool
1001 dge_shutdown(device_t self, int howto)
1002 {
1003 	struct dge_softc *sc;
1004 
1005 	sc = device_private(self);
1006 	dge_stop(&sc->sc_ethercom.ec_if, 1);
1007 
1008 	return true;
1009 }
1010 
1011 /*
1012  * dge_tx_cksum:
1013  *
1014  *	Set up TCP/IP checksumming parameters for the
1015  *	specified packet.
1016  */
1017 static int
1018 dge_tx_cksum(struct dge_softc *sc, struct dge_txsoft *txs, uint8_t *fieldsp)
1019 {
1020 	struct mbuf *m0 = txs->txs_mbuf;
1021 	struct dge_ctdes *t;
1022 	uint32_t ipcs, tucs;
1023 	struct ether_header *eh;
1024 	int offset, iphl;
1025 	uint8_t fields = 0;
1026 
1027 	/*
1028 	 * XXX It would be nice if the mbuf pkthdr had offset
1029 	 * fields for the protocol headers.
1030 	 */
1031 
1032 	eh = mtod(m0, struct ether_header *);
1033 	switch (htons(eh->ether_type)) {
1034 	case ETHERTYPE_IP:
1035 		offset = ETHER_HDR_LEN;
1036 		break;
1037 
1038 	case ETHERTYPE_VLAN:
1039 		offset = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
1040 		break;
1041 
1042 	default:
1043 		/*
1044 		 * Don't support this protocol or encapsulation.
1045 		 */
1046 		*fieldsp = 0;
1047 		return (0);
1048 	}
1049 
1050 	iphl = M_CSUM_DATA_IPv4_IPHL(m0->m_pkthdr.csum_data);
1051 
1052 	/*
1053 	 * NOTE: Even if we're not using the IP or TCP/UDP checksum
1054 	 * offload feature, if we load the context descriptor, we
1055 	 * MUST provide valid values for IPCSS and TUCSS fields.
1056 	 */
1057 
1058 	if (m0->m_pkthdr.csum_flags & M_CSUM_IPv4) {
1059 		DGE_EVCNT_INCR(&sc->sc_ev_txipsum);
1060 		fields |= TDESC_POPTS_IXSM;
1061 		ipcs = DGE_TCPIP_IPCSS(offset) |
1062 		    DGE_TCPIP_IPCSO(offset + offsetof(struct ip, ip_sum)) |
1063 		    DGE_TCPIP_IPCSE(offset + iphl - 1);
1064 	} else if (__predict_true(sc->sc_txctx_ipcs != 0xffffffff)) {
1065 		/* Use the cached value. */
1066 		ipcs = sc->sc_txctx_ipcs;
1067 	} else {
1068 		/* Just initialize it to the likely value anyway. */
1069 		ipcs = DGE_TCPIP_IPCSS(offset) |
1070 		    DGE_TCPIP_IPCSO(offset + offsetof(struct ip, ip_sum)) |
1071 		    DGE_TCPIP_IPCSE(offset + iphl - 1);
1072 	}
1073 	DPRINTF(DGE_DEBUG_CKSUM,
1074 	    ("%s: CKSUM: offset %d ipcs 0x%x\n",
1075 	    device_xname(&sc->sc_dev), offset, ipcs));
1076 
1077 	offset += iphl;
1078 
1079 	if (m0->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4)) {
1080 		DGE_EVCNT_INCR(&sc->sc_ev_txtusum);
1081 		fields |= TDESC_POPTS_TXSM;
1082 		tucs = DGE_TCPIP_TUCSS(offset) |
1083 		   DGE_TCPIP_TUCSO(offset + M_CSUM_DATA_IPv4_OFFSET(m0->m_pkthdr.csum_data)) |
1084 		   DGE_TCPIP_TUCSE(0) /* rest of packet */;
1085 	} else if (__predict_true(sc->sc_txctx_tucs != 0xffffffff)) {
1086 		/* Use the cached value. */
1087 		tucs = sc->sc_txctx_tucs;
1088 	} else {
1089 		/* Just initialize it to a valid TCP context. */
1090 		tucs = DGE_TCPIP_TUCSS(offset) |
1091 		    DGE_TCPIP_TUCSO(offset + offsetof(struct tcphdr, th_sum)) |
1092 		    DGE_TCPIP_TUCSE(0) /* rest of packet */;
1093 	}
1094 
1095 	DPRINTF(DGE_DEBUG_CKSUM,
1096 	    ("%s: CKSUM: offset %d tucs 0x%x\n",
1097 	    device_xname(&sc->sc_dev), offset, tucs));
1098 
1099 	if (sc->sc_txctx_ipcs == ipcs &&
1100 	    sc->sc_txctx_tucs == tucs) {
1101 		/* Cached context is fine. */
1102 		DGE_EVCNT_INCR(&sc->sc_ev_txctx_hit);
1103 	} else {
1104 		/* Fill in the context descriptor. */
1105 #ifdef DGE_EVENT_COUNTERS
1106 		if (sc->sc_txctx_ipcs == 0xffffffff &&
1107 		    sc->sc_txctx_tucs == 0xffffffff)
1108 			DGE_EVCNT_INCR(&sc->sc_ev_txctx_init);
1109 		else
1110 			DGE_EVCNT_INCR(&sc->sc_ev_txctx_miss);
1111 #endif
1112 		t = (struct dge_ctdes *)&sc->sc_txdescs[sc->sc_txnext];
1113 		t->dc_tcpip_ipcs = htole32(ipcs);
1114 		t->dc_tcpip_tucs = htole32(tucs);
1115 		t->dc_tcpip_cmdlen = htole32(TDESC_DTYP_CTD);
1116 		t->dc_tcpip_seg = 0;
1117 		DGE_CDTXSYNC(sc, sc->sc_txnext, 1, BUS_DMASYNC_PREWRITE);
1118 
1119 		sc->sc_txctx_ipcs = ipcs;
1120 		sc->sc_txctx_tucs = tucs;
1121 
1122 		sc->sc_txnext = DGE_NEXTTX(sc->sc_txnext);
1123 		txs->txs_ndesc++;
1124 	}
1125 
1126 	*fieldsp = fields;
1127 
1128 	return (0);
1129 }
1130 
1131 /*
1132  * dge_start:		[ifnet interface function]
1133  *
1134  *	Start packet transmission on the interface.
1135  */
1136 static void
1137 dge_start(struct ifnet *ifp)
1138 {
1139 	struct dge_softc *sc = ifp->if_softc;
1140 	struct mbuf *m0;
1141 	struct dge_txsoft *txs;
1142 	bus_dmamap_t dmamap;
1143 	int error, nexttx, lasttx = -1, ofree, seg;
1144 	uint32_t cksumcmd;
1145 	uint8_t cksumfields;
1146 
1147 	if ((ifp->if_flags & (IFF_RUNNING|IFF_OACTIVE)) != IFF_RUNNING)
1148 		return;
1149 
1150 	/*
1151 	 * Remember the previous number of free descriptors.
1152 	 */
1153 	ofree = sc->sc_txfree;
1154 
1155 	/*
1156 	 * Loop through the send queue, setting up transmit descriptors
1157 	 * until we drain the queue, or use up all available transmit
1158 	 * descriptors.
1159 	 */
1160 	for (;;) {
1161 		/* Grab a packet off the queue. */
1162 		IFQ_POLL(&ifp->if_snd, m0);
1163 		if (m0 == NULL)
1164 			break;
1165 
1166 		DPRINTF(DGE_DEBUG_TX,
1167 		    ("%s: TX: have packet to transmit: %p\n",
1168 		    device_xname(&sc->sc_dev), m0));
1169 
1170 		/* Get a work queue entry. */
1171 		if (sc->sc_txsfree < DGE_TXQUEUE_GC) {
1172 			dge_txintr(sc);
1173 			if (sc->sc_txsfree == 0) {
1174 				DPRINTF(DGE_DEBUG_TX,
1175 				    ("%s: TX: no free job descriptors\n",
1176 					device_xname(&sc->sc_dev)));
1177 				DGE_EVCNT_INCR(&sc->sc_ev_txsstall);
1178 				break;
1179 			}
1180 		}
1181 
1182 		txs = &sc->sc_txsoft[sc->sc_txsnext];
1183 		dmamap = txs->txs_dmamap;
1184 
1185 		/*
1186 		 * Load the DMA map.  If this fails, the packet either
1187 		 * didn't fit in the allotted number of segments, or we
1188 		 * were short on resources.  For the too-many-segments
1189 		 * case, we simply report an error and drop the packet,
1190 		 * since we can't sanely copy a jumbo packet to a single
1191 		 * buffer.
1192 		 */
1193 		error = bus_dmamap_load_mbuf(sc->sc_dmat, dmamap, m0,
1194 		    BUS_DMA_WRITE|BUS_DMA_NOWAIT);
1195 		if (error) {
1196 			if (error == EFBIG) {
1197 				DGE_EVCNT_INCR(&sc->sc_ev_txdrop);
1198 				printf("%s: Tx packet consumes too many "
1199 				    "DMA segments, dropping...\n",
1200 				    device_xname(&sc->sc_dev));
1201 				IFQ_DEQUEUE(&ifp->if_snd, m0);
1202 				m_freem(m0);
1203 				continue;
1204 			}
1205 			/*
1206 			 * Short on resources, just stop for now.
1207 			 */
1208 			DPRINTF(DGE_DEBUG_TX,
1209 			    ("%s: TX: dmamap load failed: %d\n",
1210 			    device_xname(&sc->sc_dev), error));
1211 			break;
1212 		}
1213 
1214 		/*
1215 		 * Ensure we have enough descriptors free to describe
1216 		 * the packet.  Note, we always reserve one descriptor
1217 		 * at the end of the ring due to the semantics of the
1218 		 * TDT register, plus one more in the event we need
1219 		 * to re-load checksum offload context.
1220 		 */
1221 		if (dmamap->dm_nsegs > (sc->sc_txfree - 2)) {
1222 			/*
1223 			 * Not enough free descriptors to transmit this
1224 			 * packet.  We haven't committed anything yet,
1225 			 * so just unload the DMA map, put the packet
1226 			 * pack on the queue, and punt.  Notify the upper
1227 			 * layer that there are no more slots left.
1228 			 */
1229 			DPRINTF(DGE_DEBUG_TX,
1230 			    ("%s: TX: need %d descriptors, have %d\n",
1231 			    device_xname(&sc->sc_dev), dmamap->dm_nsegs,
1232 			    sc->sc_txfree - 1));
1233 			ifp->if_flags |= IFF_OACTIVE;
1234 			bus_dmamap_unload(sc->sc_dmat, dmamap);
1235 			DGE_EVCNT_INCR(&sc->sc_ev_txdstall);
1236 			break;
1237 		}
1238 
1239 		IFQ_DEQUEUE(&ifp->if_snd, m0);
1240 
1241 		/*
1242 		 * WE ARE NOW COMMITTED TO TRANSMITTING THE PACKET.
1243 		 */
1244 
1245 		/* Sync the DMA map. */
1246 		bus_dmamap_sync(sc->sc_dmat, dmamap, 0, dmamap->dm_mapsize,
1247 		    BUS_DMASYNC_PREWRITE);
1248 
1249 		DPRINTF(DGE_DEBUG_TX,
1250 		    ("%s: TX: packet has %d DMA segments\n",
1251 		    device_xname(&sc->sc_dev), dmamap->dm_nsegs));
1252 
1253 		DGE_EVCNT_INCR(&sc->sc_ev_txseg[dmamap->dm_nsegs - 1]);
1254 
1255 		/*
1256 		 * Store a pointer to the packet so that we can free it
1257 		 * later.
1258 		 *
1259 		 * Initially, we consider the number of descriptors the
1260 		 * packet uses the number of DMA segments.  This may be
1261 		 * incremented by 1 if we do checksum offload (a descriptor
1262 		 * is used to set the checksum context).
1263 		 */
1264 		txs->txs_mbuf = m0;
1265 		txs->txs_firstdesc = sc->sc_txnext;
1266 		txs->txs_ndesc = dmamap->dm_nsegs;
1267 
1268 		/*
1269 		 * Set up checksum offload parameters for
1270 		 * this packet.
1271 		 */
1272 		if (m0->m_pkthdr.csum_flags &
1273 		    (M_CSUM_IPv4|M_CSUM_TCPv4|M_CSUM_UDPv4)) {
1274 			if (dge_tx_cksum(sc, txs, &cksumfields) != 0) {
1275 				/* Error message already displayed. */
1276 				bus_dmamap_unload(sc->sc_dmat, dmamap);
1277 				continue;
1278 			}
1279 		} else {
1280 			cksumfields = 0;
1281 		}
1282 
1283 		cksumcmd = TDESC_DCMD_IDE | TDESC_DTYP_DATA;
1284 
1285 		/*
1286 		 * Initialize the transmit descriptor.
1287 		 */
1288 		for (nexttx = sc->sc_txnext, seg = 0;
1289 		     seg < dmamap->dm_nsegs;
1290 		     seg++, nexttx = DGE_NEXTTX(nexttx)) {
1291 			/*
1292 			 * Note: we currently only use 32-bit DMA
1293 			 * addresses.
1294 			 */
1295 			sc->sc_txdescs[nexttx].dt_baddrh = 0;
1296 			sc->sc_txdescs[nexttx].dt_baddrl =
1297 			    htole32(dmamap->dm_segs[seg].ds_addr);
1298 			sc->sc_txdescs[nexttx].dt_ctl =
1299 			    htole32(cksumcmd | dmamap->dm_segs[seg].ds_len);
1300 			sc->sc_txdescs[nexttx].dt_status = 0;
1301 			sc->sc_txdescs[nexttx].dt_popts = cksumfields;
1302 			sc->sc_txdescs[nexttx].dt_vlan = 0;
1303 			lasttx = nexttx;
1304 
1305 			DPRINTF(DGE_DEBUG_TX,
1306 			    ("%s: TX: desc %d: low 0x%08lx, len 0x%04lx\n",
1307 			    device_xname(&sc->sc_dev), nexttx,
1308 			    (unsigned long)le32toh(dmamap->dm_segs[seg].ds_addr),
1309 			    (unsigned long)le32toh(dmamap->dm_segs[seg].ds_len)));
1310 		}
1311 
1312 		KASSERT(lasttx != -1);
1313 
1314 		/*
1315 		 * Set up the command byte on the last descriptor of
1316 		 * the packet.  If we're in the interrupt delay window,
1317 		 * delay the interrupt.
1318 		 */
1319 		sc->sc_txdescs[lasttx].dt_ctl |=
1320 		    htole32(TDESC_DCMD_EOP | TDESC_DCMD_RS);
1321 
1322 		txs->txs_lastdesc = lasttx;
1323 
1324 		DPRINTF(DGE_DEBUG_TX,
1325 		    ("%s: TX: desc %d: cmdlen 0x%08x\n", device_xname(&sc->sc_dev),
1326 		    lasttx, le32toh(sc->sc_txdescs[lasttx].dt_ctl)));
1327 
1328 		/* Sync the descriptors we're using. */
1329 		DGE_CDTXSYNC(sc, sc->sc_txnext, dmamap->dm_nsegs,
1330 		    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1331 
1332 		/* Give the packet to the chip. */
1333 		CSR_WRITE(sc, DGE_TDT, nexttx);
1334 
1335 		DPRINTF(DGE_DEBUG_TX,
1336 		    ("%s: TX: TDT -> %d\n", device_xname(&sc->sc_dev), nexttx));
1337 
1338 		DPRINTF(DGE_DEBUG_TX,
1339 		    ("%s: TX: finished transmitting packet, job %d\n",
1340 		    device_xname(&sc->sc_dev), sc->sc_txsnext));
1341 
1342 		/* Advance the tx pointer. */
1343 		sc->sc_txfree -= txs->txs_ndesc;
1344 		sc->sc_txnext = nexttx;
1345 
1346 		sc->sc_txsfree--;
1347 		sc->sc_txsnext = DGE_NEXTTXS(sc->sc_txsnext);
1348 
1349 		/* Pass the packet to any BPF listeners. */
1350 		bpf_mtap(ifp, m0);
1351 	}
1352 
1353 	if (sc->sc_txsfree == 0 || sc->sc_txfree <= 2) {
1354 		/* No more slots; notify upper layer. */
1355 		ifp->if_flags |= IFF_OACTIVE;
1356 	}
1357 
1358 	if (sc->sc_txfree != ofree) {
1359 		/* Set a watchdog timer in case the chip flakes out. */
1360 		ifp->if_timer = 5;
1361 	}
1362 }
1363 
1364 /*
1365  * dge_watchdog:		[ifnet interface function]
1366  *
1367  *	Watchdog timer handler.
1368  */
1369 static void
1370 dge_watchdog(struct ifnet *ifp)
1371 {
1372 	struct dge_softc *sc = ifp->if_softc;
1373 
1374 	/*
1375 	 * Since we're using delayed interrupts, sweep up
1376 	 * before we report an error.
1377 	 */
1378 	dge_txintr(sc);
1379 
1380 	if (sc->sc_txfree != DGE_NTXDESC) {
1381 		printf("%s: device timeout (txfree %d txsfree %d txnext %d)\n",
1382 		    device_xname(&sc->sc_dev), sc->sc_txfree, sc->sc_txsfree,
1383 		    sc->sc_txnext);
1384 		ifp->if_oerrors++;
1385 
1386 		/* Reset the interface. */
1387 		(void) dge_init(ifp);
1388 	}
1389 
1390 	/* Try to get more packets going. */
1391 	dge_start(ifp);
1392 }
1393 
1394 /*
1395  * dge_ioctl:		[ifnet interface function]
1396  *
1397  *	Handle control requests from the operator.
1398  */
1399 static int
1400 dge_ioctl(struct ifnet *ifp, u_long cmd, void *data)
1401 {
1402 	struct dge_softc *sc = ifp->if_softc;
1403 	struct ifreq *ifr = (struct ifreq *) data;
1404 	pcireg_t preg;
1405 	int s, error, mmrbc;
1406 
1407 	s = splnet();
1408 
1409 	switch (cmd) {
1410 	case SIOCSIFMEDIA:
1411 	case SIOCGIFMEDIA:
1412 		error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
1413 		break;
1414 
1415 	case SIOCSIFMTU:
1416 		if (ifr->ifr_mtu < ETHERMIN || ifr->ifr_mtu > DGE_MAX_MTU)
1417 			error = EINVAL;
1418 		else if ((error = ifioctl_common(ifp, cmd, data)) != ENETRESET)
1419 			break;
1420 		else if (ifp->if_flags & IFF_UP)
1421 			error = (*ifp->if_init)(ifp);
1422 		else
1423 			error = 0;
1424 		break;
1425 
1426         case SIOCSIFFLAGS:
1427 		if ((error = ifioctl_common(ifp, cmd, data)) != 0)
1428 			break;
1429 		/* extract link flags */
1430 		if ((ifp->if_flags & IFF_LINK0) == 0 &&
1431 		    (ifp->if_flags & IFF_LINK1) == 0)
1432 			mmrbc = PCIX_MMRBC_512;
1433 		else if ((ifp->if_flags & IFF_LINK0) == 0 &&
1434 		    (ifp->if_flags & IFF_LINK1) != 0)
1435 			mmrbc = PCIX_MMRBC_1024;
1436 		else if ((ifp->if_flags & IFF_LINK0) != 0 &&
1437 		    (ifp->if_flags & IFF_LINK1) == 0)
1438 			mmrbc = PCIX_MMRBC_2048;
1439 		else
1440 			mmrbc = PCIX_MMRBC_4096;
1441 		if (mmrbc != sc->sc_mmrbc) {
1442 			preg = pci_conf_read(sc->sc_pc, sc->sc_pt,DGE_PCIX_CMD);
1443 			preg &= ~PCIX_MMRBC_MSK;
1444 			preg |= mmrbc;
1445 			pci_conf_write(sc->sc_pc, sc->sc_pt,DGE_PCIX_CMD, preg);
1446 			sc->sc_mmrbc = mmrbc;
1447 		}
1448                 /* FALLTHROUGH */
1449 	default:
1450 		if ((error = ether_ioctl(ifp, cmd, data)) != ENETRESET)
1451 			break;
1452 
1453 		error = 0;
1454 
1455 		if (cmd == SIOCSIFCAP)
1456 			error = (*ifp->if_init)(ifp);
1457 		else if (cmd != SIOCADDMULTI && cmd != SIOCDELMULTI)
1458 			;
1459 		else if (ifp->if_flags & IFF_RUNNING) {
1460 			/*
1461 			 * Multicast list has changed; set the hardware filter
1462 			 * accordingly.
1463 			 */
1464 			dge_set_filter(sc);
1465 		}
1466 		break;
1467 	}
1468 
1469 	/* Try to get more packets going. */
1470 	dge_start(ifp);
1471 
1472 	splx(s);
1473 	return (error);
1474 }
1475 
1476 /*
1477  * dge_intr:
1478  *
1479  *	Interrupt service routine.
1480  */
1481 static int
1482 dge_intr(void *arg)
1483 {
1484 	struct dge_softc *sc = arg;
1485 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
1486 	uint32_t icr;
1487 	int wantinit, handled = 0;
1488 
1489 	for (wantinit = 0; wantinit == 0;) {
1490 		icr = CSR_READ(sc, DGE_ICR);
1491 		if ((icr & sc->sc_icr) == 0)
1492 			break;
1493 
1494 #if 0 /*NRND > 0*/
1495 		if (RND_ENABLED(&sc->rnd_source))
1496 			rnd_add_uint32(&sc->rnd_source, icr);
1497 #endif
1498 
1499 		handled = 1;
1500 
1501 #if defined(DGE_DEBUG) || defined(DGE_EVENT_COUNTERS)
1502 		if (icr & (ICR_RXDMT0|ICR_RXT0)) {
1503 			DPRINTF(DGE_DEBUG_RX,
1504 			    ("%s: RX: got Rx intr 0x%08x\n",
1505 			    device_xname(&sc->sc_dev),
1506 			    icr & (ICR_RXDMT0|ICR_RXT0)));
1507 			DGE_EVCNT_INCR(&sc->sc_ev_rxintr);
1508 		}
1509 #endif
1510 		dge_rxintr(sc);
1511 
1512 #if defined(DGE_DEBUG) || defined(DGE_EVENT_COUNTERS)
1513 		if (icr & ICR_TXDW) {
1514 			DPRINTF(DGE_DEBUG_TX,
1515 			    ("%s: TX: got TXDW interrupt\n",
1516 			    device_xname(&sc->sc_dev)));
1517 			DGE_EVCNT_INCR(&sc->sc_ev_txdw);
1518 		}
1519 		if (icr & ICR_TXQE)
1520 			DGE_EVCNT_INCR(&sc->sc_ev_txqe);
1521 #endif
1522 		dge_txintr(sc);
1523 
1524 		if (icr & (ICR_LSC|ICR_RXSEQ)) {
1525 			DGE_EVCNT_INCR(&sc->sc_ev_linkintr);
1526 			dge_linkintr(sc, icr);
1527 		}
1528 
1529 		if (icr & ICR_RXO) {
1530 			printf("%s: Receive overrun\n", device_xname(&sc->sc_dev));
1531 			wantinit = 1;
1532 		}
1533 	}
1534 
1535 	if (handled) {
1536 		if (wantinit)
1537 			dge_init(ifp);
1538 
1539 		/* Try to get more packets going. */
1540 		dge_start(ifp);
1541 	}
1542 
1543 	return (handled);
1544 }
1545 
1546 /*
1547  * dge_txintr:
1548  *
1549  *	Helper; handle transmit interrupts.
1550  */
1551 static void
1552 dge_txintr(struct dge_softc *sc)
1553 {
1554 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
1555 	struct dge_txsoft *txs;
1556 	uint8_t status;
1557 	int i;
1558 
1559 	ifp->if_flags &= ~IFF_OACTIVE;
1560 
1561 	/*
1562 	 * Go through the Tx list and free mbufs for those
1563 	 * frames which have been transmitted.
1564 	 */
1565 	for (i = sc->sc_txsdirty; sc->sc_txsfree != DGE_TXQUEUELEN;
1566 	     i = DGE_NEXTTXS(i), sc->sc_txsfree++) {
1567 		txs = &sc->sc_txsoft[i];
1568 
1569 		DPRINTF(DGE_DEBUG_TX,
1570 		    ("%s: TX: checking job %d\n", device_xname(&sc->sc_dev), i));
1571 
1572 		DGE_CDTXSYNC(sc, txs->txs_firstdesc, txs->txs_dmamap->dm_nsegs,
1573 		    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1574 
1575 		status =
1576 		    sc->sc_txdescs[txs->txs_lastdesc].dt_status;
1577 		if ((status & TDESC_STA_DD) == 0) {
1578 			DGE_CDTXSYNC(sc, txs->txs_lastdesc, 1,
1579 			    BUS_DMASYNC_PREREAD);
1580 			break;
1581 		}
1582 
1583 		DPRINTF(DGE_DEBUG_TX,
1584 		    ("%s: TX: job %d done: descs %d..%d\n",
1585 		    device_xname(&sc->sc_dev), i, txs->txs_firstdesc,
1586 		    txs->txs_lastdesc));
1587 
1588 		ifp->if_opackets++;
1589 		sc->sc_txfree += txs->txs_ndesc;
1590 		bus_dmamap_sync(sc->sc_dmat, txs->txs_dmamap,
1591 		    0, txs->txs_dmamap->dm_mapsize, BUS_DMASYNC_POSTWRITE);
1592 		bus_dmamap_unload(sc->sc_dmat, txs->txs_dmamap);
1593 		m_freem(txs->txs_mbuf);
1594 		txs->txs_mbuf = NULL;
1595 	}
1596 
1597 	/* Update the dirty transmit buffer pointer. */
1598 	sc->sc_txsdirty = i;
1599 	DPRINTF(DGE_DEBUG_TX,
1600 	    ("%s: TX: txsdirty -> %d\n", device_xname(&sc->sc_dev), i));
1601 
1602 	/*
1603 	 * If there are no more pending transmissions, cancel the watchdog
1604 	 * timer.
1605 	 */
1606 	if (sc->sc_txsfree == DGE_TXQUEUELEN)
1607 		ifp->if_timer = 0;
1608 }
1609 
1610 /*
1611  * dge_rxintr:
1612  *
1613  *	Helper; handle receive interrupts.
1614  */
1615 static void
1616 dge_rxintr(struct dge_softc *sc)
1617 {
1618 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
1619 	struct dge_rxsoft *rxs;
1620 	struct mbuf *m;
1621 	int i, len;
1622 	uint8_t status, errors;
1623 
1624 	for (i = sc->sc_rxptr;; i = DGE_NEXTRX(i)) {
1625 		rxs = &sc->sc_rxsoft[i];
1626 
1627 		DPRINTF(DGE_DEBUG_RX,
1628 		    ("%s: RX: checking descriptor %d\n",
1629 		    device_xname(&sc->sc_dev), i));
1630 
1631 		DGE_CDRXSYNC(sc, i, BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1632 
1633 		status = sc->sc_rxdescs[i].dr_status;
1634 		errors = sc->sc_rxdescs[i].dr_errors;
1635 		len = le16toh(sc->sc_rxdescs[i].dr_len);
1636 
1637 		if ((status & RDESC_STS_DD) == 0) {
1638 			/*
1639 			 * We have processed all of the receive descriptors.
1640 			 */
1641 			DGE_CDRXSYNC(sc, i, BUS_DMASYNC_PREREAD);
1642 			break;
1643 		}
1644 
1645 		if (__predict_false(sc->sc_rxdiscard)) {
1646 			DPRINTF(DGE_DEBUG_RX,
1647 			    ("%s: RX: discarding contents of descriptor %d\n",
1648 			    device_xname(&sc->sc_dev), i));
1649 			DGE_INIT_RXDESC(sc, i);
1650 			if (status & RDESC_STS_EOP) {
1651 				/* Reset our state. */
1652 				DPRINTF(DGE_DEBUG_RX,
1653 				    ("%s: RX: resetting rxdiscard -> 0\n",
1654 				    device_xname(&sc->sc_dev)));
1655 				sc->sc_rxdiscard = 0;
1656 			}
1657 			continue;
1658 		}
1659 
1660 		bus_dmamap_sync(sc->sc_dmat, rxs->rxs_dmamap, 0,
1661 		    rxs->rxs_dmamap->dm_mapsize, BUS_DMASYNC_POSTREAD);
1662 
1663 		m = rxs->rxs_mbuf;
1664 
1665 		/*
1666 		 * Add a new receive buffer to the ring.
1667 		 */
1668 		if (dge_add_rxbuf(sc, i) != 0) {
1669 			/*
1670 			 * Failed, throw away what we've done so
1671 			 * far, and discard the rest of the packet.
1672 			 */
1673 			ifp->if_ierrors++;
1674 			bus_dmamap_sync(sc->sc_dmat, rxs->rxs_dmamap, 0,
1675 			    rxs->rxs_dmamap->dm_mapsize, BUS_DMASYNC_PREREAD);
1676 			DGE_INIT_RXDESC(sc, i);
1677 			if ((status & RDESC_STS_EOP) == 0)
1678 				sc->sc_rxdiscard = 1;
1679 			if (sc->sc_rxhead != NULL)
1680 				m_freem(sc->sc_rxhead);
1681 			DGE_RXCHAIN_RESET(sc);
1682 			DPRINTF(DGE_DEBUG_RX,
1683 			    ("%s: RX: Rx buffer allocation failed, "
1684 			    "dropping packet%s\n", device_xname(&sc->sc_dev),
1685 			    sc->sc_rxdiscard ? " (discard)" : ""));
1686 			continue;
1687 		}
1688 		DGE_INIT_RXDESC(sc, DGE_PREVRX(i)); /* Write the descriptor */
1689 
1690 		DGE_RXCHAIN_LINK(sc, m);
1691 
1692 		m->m_len = len;
1693 
1694 		DPRINTF(DGE_DEBUG_RX,
1695 		    ("%s: RX: buffer at %p len %d\n",
1696 		    device_xname(&sc->sc_dev), m->m_data, len));
1697 
1698 		/*
1699 		 * If this is not the end of the packet, keep
1700 		 * looking.
1701 		 */
1702 		if ((status & RDESC_STS_EOP) == 0) {
1703 			sc->sc_rxlen += len;
1704 			DPRINTF(DGE_DEBUG_RX,
1705 			    ("%s: RX: not yet EOP, rxlen -> %d\n",
1706 			    device_xname(&sc->sc_dev), sc->sc_rxlen));
1707 			continue;
1708 		}
1709 
1710 		/*
1711 		 * Okay, we have the entire packet now...
1712 		 */
1713 		*sc->sc_rxtailp = NULL;
1714 		m = sc->sc_rxhead;
1715 		len += sc->sc_rxlen;
1716 
1717 		DGE_RXCHAIN_RESET(sc);
1718 
1719 		DPRINTF(DGE_DEBUG_RX,
1720 		    ("%s: RX: have entire packet, len -> %d\n",
1721 		    device_xname(&sc->sc_dev), len));
1722 
1723 		/*
1724 		 * If an error occurred, update stats and drop the packet.
1725 		 */
1726 		if (errors &
1727 		     (RDESC_ERR_CE|RDESC_ERR_SE|RDESC_ERR_P|RDESC_ERR_RXE)) {
1728 			ifp->if_ierrors++;
1729 			if (errors & RDESC_ERR_SE)
1730 				printf("%s: symbol error\n",
1731 				    device_xname(&sc->sc_dev));
1732 			else if (errors & RDESC_ERR_P)
1733 				printf("%s: parity error\n",
1734 				    device_xname(&sc->sc_dev));
1735 			else if (errors & RDESC_ERR_CE)
1736 				printf("%s: CRC error\n",
1737 				    device_xname(&sc->sc_dev));
1738 			m_freem(m);
1739 			continue;
1740 		}
1741 
1742 		/*
1743 		 * No errors.  Receive the packet.
1744 		 */
1745 		m->m_pkthdr.rcvif = ifp;
1746 		m->m_pkthdr.len = len;
1747 
1748 		/*
1749 		 * Set up checksum info for this packet.
1750 		 */
1751 		if (status & RDESC_STS_IPCS) {
1752 			DGE_EVCNT_INCR(&sc->sc_ev_rxipsum);
1753 			m->m_pkthdr.csum_flags |= M_CSUM_IPv4;
1754 			if (errors & RDESC_ERR_IPE)
1755 				m->m_pkthdr.csum_flags |= M_CSUM_IPv4_BAD;
1756 		}
1757 		if (status & RDESC_STS_TCPCS) {
1758 			/*
1759 			 * Note: we don't know if this was TCP or UDP,
1760 			 * so we just set both bits, and expect the
1761 			 * upper layers to deal.
1762 			 */
1763 			DGE_EVCNT_INCR(&sc->sc_ev_rxtusum);
1764 			m->m_pkthdr.csum_flags |= M_CSUM_TCPv4|M_CSUM_UDPv4;
1765 			if (errors & RDESC_ERR_TCPE)
1766 				m->m_pkthdr.csum_flags |= M_CSUM_TCP_UDP_BAD;
1767 		}
1768 
1769 		ifp->if_ipackets++;
1770 
1771 		/* Pass this up to any BPF listeners. */
1772 		bpf_mtap(ifp, m);
1773 
1774 		/* Pass it on. */
1775 		(*ifp->if_input)(ifp, m);
1776 	}
1777 
1778 	/* Update the receive pointer. */
1779 	sc->sc_rxptr = i;
1780 
1781 	DPRINTF(DGE_DEBUG_RX,
1782 	    ("%s: RX: rxptr -> %d\n", device_xname(&sc->sc_dev), i));
1783 }
1784 
1785 /*
1786  * dge_linkintr:
1787  *
1788  *	Helper; handle link interrupts.
1789  */
1790 static void
1791 dge_linkintr(struct dge_softc *sc, uint32_t icr)
1792 {
1793 	uint32_t status;
1794 
1795 	if (icr & ICR_LSC) {
1796 		status = CSR_READ(sc, DGE_STATUS);
1797 		if (status & STATUS_LINKUP) {
1798 			DPRINTF(DGE_DEBUG_LINK, ("%s: LINK: LSC -> up\n",
1799 			    device_xname(&sc->sc_dev)));
1800 		} else {
1801 			DPRINTF(DGE_DEBUG_LINK, ("%s: LINK: LSC -> down\n",
1802 			    device_xname(&sc->sc_dev)));
1803 		}
1804 	} else if (icr & ICR_RXSEQ) {
1805 		DPRINTF(DGE_DEBUG_LINK,
1806 		    ("%s: LINK: Receive sequence error\n",
1807 		    device_xname(&sc->sc_dev)));
1808 	}
1809 	/* XXX - fix errata */
1810 }
1811 
1812 /*
1813  * dge_reset:
1814  *
1815  *	Reset the i82597 chip.
1816  */
1817 static void
1818 dge_reset(struct dge_softc *sc)
1819 {
1820 	int i;
1821 
1822 	/*
1823 	 * Do a chip reset.
1824 	 */
1825 	CSR_WRITE(sc, DGE_CTRL0, CTRL0_RST | sc->sc_ctrl0);
1826 
1827 	delay(10000);
1828 
1829 	for (i = 0; i < 1000; i++) {
1830 		if ((CSR_READ(sc, DGE_CTRL0) & CTRL0_RST) == 0)
1831 			break;
1832 		delay(20);
1833 	}
1834 
1835 	if (CSR_READ(sc, DGE_CTRL0) & CTRL0_RST)
1836 		printf("%s: WARNING: reset failed to complete\n",
1837 		    device_xname(&sc->sc_dev));
1838         /*
1839          * Reset the EEPROM logic.
1840          * This will cause the chip to reread its default values,
1841 	 * which doesn't happen otherwise (errata).
1842          */
1843         CSR_WRITE(sc, DGE_CTRL1, CTRL1_EE_RST);
1844         delay(10000);
1845 }
1846 
1847 /*
1848  * dge_init:		[ifnet interface function]
1849  *
1850  *	Initialize the interface.  Must be called at splnet().
1851  */
1852 static int
1853 dge_init(struct ifnet *ifp)
1854 {
1855 	struct dge_softc *sc = ifp->if_softc;
1856 	struct dge_rxsoft *rxs;
1857 	int i, error = 0;
1858 	uint32_t reg;
1859 
1860 	/*
1861 	 * *_HDR_ALIGNED_P is constant 1 if __NO_STRICT_ALIGMENT is set.
1862 	 * There is a small but measurable benefit to avoiding the adjusment
1863 	 * of the descriptor so that the headers are aligned, for normal mtu,
1864 	 * on such platforms.  One possibility is that the DMA itself is
1865 	 * slightly more efficient if the front of the entire packet (instead
1866 	 * of the front of the headers) is aligned.
1867 	 *
1868 	 * Note we must always set align_tweak to 0 if we are using
1869 	 * jumbo frames.
1870 	 */
1871 #ifdef __NO_STRICT_ALIGNMENT
1872 	sc->sc_align_tweak = 0;
1873 #else
1874 	if ((ifp->if_mtu + ETHER_HDR_LEN + ETHER_CRC_LEN) > (MCLBYTES - 2))
1875 		sc->sc_align_tweak = 0;
1876 	else
1877 		sc->sc_align_tweak = 2;
1878 #endif /* __NO_STRICT_ALIGNMENT */
1879 
1880 	/* Cancel any pending I/O. */
1881 	dge_stop(ifp, 0);
1882 
1883 	/* Reset the chip to a known state. */
1884 	dge_reset(sc);
1885 
1886 	/* Initialize the transmit descriptor ring. */
1887 	memset(sc->sc_txdescs, 0, sizeof(sc->sc_txdescs));
1888 	DGE_CDTXSYNC(sc, 0, DGE_NTXDESC,
1889 	    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1890 	sc->sc_txfree = DGE_NTXDESC;
1891 	sc->sc_txnext = 0;
1892 
1893 	sc->sc_txctx_ipcs = 0xffffffff;
1894 	sc->sc_txctx_tucs = 0xffffffff;
1895 
1896 	CSR_WRITE(sc, DGE_TDBAH, 0);
1897 	CSR_WRITE(sc, DGE_TDBAL, DGE_CDTXADDR(sc, 0));
1898 	CSR_WRITE(sc, DGE_TDLEN, sizeof(sc->sc_txdescs));
1899 	CSR_WRITE(sc, DGE_TDH, 0);
1900 	CSR_WRITE(sc, DGE_TDT, 0);
1901 	CSR_WRITE(sc, DGE_TIDV, TIDV);
1902 
1903 #if 0
1904 	CSR_WRITE(sc, DGE_TXDCTL, TXDCTL_PTHRESH(0) |
1905 	    TXDCTL_HTHRESH(0) | TXDCTL_WTHRESH(0));
1906 #endif
1907 	CSR_WRITE(sc, DGE_RXDCTL,
1908 	    RXDCTL_PTHRESH(RXDCTL_PTHRESH_VAL) |
1909 	    RXDCTL_HTHRESH(RXDCTL_HTHRESH_VAL) |
1910 	    RXDCTL_WTHRESH(RXDCTL_WTHRESH_VAL));
1911 
1912 	/* Initialize the transmit job descriptors. */
1913 	for (i = 0; i < DGE_TXQUEUELEN; i++)
1914 		sc->sc_txsoft[i].txs_mbuf = NULL;
1915 	sc->sc_txsfree = DGE_TXQUEUELEN;
1916 	sc->sc_txsnext = 0;
1917 	sc->sc_txsdirty = 0;
1918 
1919 	/*
1920 	 * Initialize the receive descriptor and receive job
1921 	 * descriptor rings.
1922 	 */
1923 	CSR_WRITE(sc, DGE_RDBAH, 0);
1924 	CSR_WRITE(sc, DGE_RDBAL, DGE_CDRXADDR(sc, 0));
1925 	CSR_WRITE(sc, DGE_RDLEN, sizeof(sc->sc_rxdescs));
1926 	CSR_WRITE(sc, DGE_RDH, DGE_RXSPACE);
1927 	CSR_WRITE(sc, DGE_RDT, 0);
1928 	CSR_WRITE(sc, DGE_RDTR, RDTR | 0x80000000);
1929 	CSR_WRITE(sc, DGE_FCRTL, FCRTL | FCRTL_XONE);
1930 	CSR_WRITE(sc, DGE_FCRTH, FCRTH);
1931 
1932 	for (i = 0; i < DGE_NRXDESC; i++) {
1933 		rxs = &sc->sc_rxsoft[i];
1934 		if (rxs->rxs_mbuf == NULL) {
1935 			if ((error = dge_add_rxbuf(sc, i)) != 0) {
1936 				printf("%s: unable to allocate or map rx "
1937 				    "buffer %d, error = %d\n",
1938 				    device_xname(&sc->sc_dev), i, error);
1939 				/*
1940 				 * XXX Should attempt to run with fewer receive
1941 				 * XXX buffers instead of just failing.
1942 				 */
1943 				dge_rxdrain(sc);
1944 				goto out;
1945 			}
1946 		}
1947 		DGE_INIT_RXDESC(sc, i);
1948 	}
1949 	sc->sc_rxptr = DGE_RXSPACE;
1950 	sc->sc_rxdiscard = 0;
1951 	DGE_RXCHAIN_RESET(sc);
1952 
1953 	if (sc->sc_ethercom.ec_capabilities & ETHERCAP_JUMBO_MTU) {
1954 		sc->sc_ctrl0 |= CTRL0_JFE;
1955 		CSR_WRITE(sc, DGE_MFS, ETHER_MAX_LEN_JUMBO << 16);
1956 	}
1957 
1958 	/* Write the control registers. */
1959 	CSR_WRITE(sc, DGE_CTRL0, sc->sc_ctrl0);
1960 
1961 	/*
1962 	 * Set up checksum offload parameters.
1963 	 */
1964 	reg = CSR_READ(sc, DGE_RXCSUM);
1965 	if (ifp->if_capenable & IFCAP_CSUM_IPv4_Rx)
1966 		reg |= RXCSUM_IPOFL;
1967 	else
1968 		reg &= ~RXCSUM_IPOFL;
1969 	if (ifp->if_capenable & (IFCAP_CSUM_TCPv4_Rx | IFCAP_CSUM_UDPv4_Rx))
1970 		reg |= RXCSUM_IPOFL | RXCSUM_TUOFL;
1971 	else {
1972 		reg &= ~RXCSUM_TUOFL;
1973 		if ((ifp->if_capenable & IFCAP_CSUM_IPv4_Rx) == 0)
1974 			reg &= ~RXCSUM_IPOFL;
1975 	}
1976 	CSR_WRITE(sc, DGE_RXCSUM, reg);
1977 
1978 	/*
1979 	 * Set up the interrupt registers.
1980 	 */
1981 	CSR_WRITE(sc, DGE_IMC, 0xffffffffU);
1982 	sc->sc_icr = ICR_TXDW | ICR_LSC | ICR_RXSEQ | ICR_RXDMT0 |
1983 	    ICR_RXO | ICR_RXT0;
1984 
1985 	CSR_WRITE(sc, DGE_IMS, sc->sc_icr);
1986 
1987 	/*
1988 	 * Set up the transmit control register.
1989 	 */
1990 	sc->sc_tctl = TCTL_TCE|TCTL_TPDE|TCTL_TXEN;
1991 	CSR_WRITE(sc, DGE_TCTL, sc->sc_tctl);
1992 
1993 	/*
1994 	 * Set up the receive control register; we actually program
1995 	 * the register when we set the receive filter.  Use multicast
1996 	 * address offset type 0.
1997 	 */
1998 	sc->sc_mchash_type = 0;
1999 
2000 	sc->sc_rctl = RCTL_RXEN | RCTL_RDMTS_12 | RCTL_RPDA_MC |
2001 	    RCTL_CFF | RCTL_SECRC | RCTL_MO(sc->sc_mchash_type);
2002 
2003 #ifdef DGE_OFFBYONE_RXBUG
2004 	sc->sc_rctl |= RCTL_BSIZE_16k;
2005 #else
2006 	switch(MCLBYTES) {
2007 	case 2048:
2008 		sc->sc_rctl |= RCTL_BSIZE_2k;
2009 		break;
2010 	case 4096:
2011 		sc->sc_rctl |= RCTL_BSIZE_4k;
2012 		break;
2013 	case 8192:
2014 		sc->sc_rctl |= RCTL_BSIZE_8k;
2015 		break;
2016 	case 16384:
2017 		sc->sc_rctl |= RCTL_BSIZE_16k;
2018 		break;
2019 	default:
2020 		panic("dge_init: MCLBYTES %d unsupported", MCLBYTES);
2021 	}
2022 #endif
2023 
2024 	/* Set the receive filter. */
2025 	/* Also sets RCTL */
2026 	dge_set_filter(sc);
2027 
2028 	/* ...all done! */
2029 	ifp->if_flags |= IFF_RUNNING;
2030 	ifp->if_flags &= ~IFF_OACTIVE;
2031 
2032  out:
2033 	if (error)
2034 		printf("%s: interface not running\n", device_xname(&sc->sc_dev));
2035 	return (error);
2036 }
2037 
2038 /*
2039  * dge_rxdrain:
2040  *
2041  *	Drain the receive queue.
2042  */
2043 static void
2044 dge_rxdrain(struct dge_softc *sc)
2045 {
2046 	struct dge_rxsoft *rxs;
2047 	int i;
2048 
2049 	for (i = 0; i < DGE_NRXDESC; i++) {
2050 		rxs = &sc->sc_rxsoft[i];
2051 		if (rxs->rxs_mbuf != NULL) {
2052 			bus_dmamap_unload(sc->sc_dmat, rxs->rxs_dmamap);
2053 			m_freem(rxs->rxs_mbuf);
2054 			rxs->rxs_mbuf = NULL;
2055 		}
2056 	}
2057 }
2058 
2059 /*
2060  * dge_stop:		[ifnet interface function]
2061  *
2062  *	Stop transmission on the interface.
2063  */
2064 static void
2065 dge_stop(struct ifnet *ifp, int disable)
2066 {
2067 	struct dge_softc *sc = ifp->if_softc;
2068 	struct dge_txsoft *txs;
2069 	int i;
2070 
2071 	/* Stop the transmit and receive processes. */
2072 	CSR_WRITE(sc, DGE_TCTL, 0);
2073 	CSR_WRITE(sc, DGE_RCTL, 0);
2074 
2075 	/* Release any queued transmit buffers. */
2076 	for (i = 0; i < DGE_TXQUEUELEN; i++) {
2077 		txs = &sc->sc_txsoft[i];
2078 		if (txs->txs_mbuf != NULL) {
2079 			bus_dmamap_unload(sc->sc_dmat, txs->txs_dmamap);
2080 			m_freem(txs->txs_mbuf);
2081 			txs->txs_mbuf = NULL;
2082 		}
2083 	}
2084 
2085 	/* Mark the interface as down and cancel the watchdog timer. */
2086 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
2087 	ifp->if_timer = 0;
2088 
2089 	if (disable)
2090 		dge_rxdrain(sc);
2091 }
2092 
2093 /*
2094  * dge_add_rxbuf:
2095  *
2096  *	Add a receive buffer to the indiciated descriptor.
2097  */
2098 static int
2099 dge_add_rxbuf(struct dge_softc *sc, int idx)
2100 {
2101 	struct dge_rxsoft *rxs = &sc->sc_rxsoft[idx];
2102 	struct mbuf *m;
2103 	int error;
2104 #ifdef DGE_OFFBYONE_RXBUG
2105 	void *buf;
2106 #endif
2107 
2108 	MGETHDR(m, M_DONTWAIT, MT_DATA);
2109 	if (m == NULL)
2110 		return (ENOBUFS);
2111 
2112 #ifdef DGE_OFFBYONE_RXBUG
2113 	if ((buf = dge_getbuf(sc)) == NULL)
2114 		return ENOBUFS;
2115 
2116 	m->m_len = m->m_pkthdr.len = DGE_BUFFER_SIZE;
2117 	MEXTADD(m, buf, DGE_BUFFER_SIZE, M_DEVBUF, dge_freebuf, sc);
2118 	m->m_flags |= M_EXT_RW;
2119 
2120 	if (rxs->rxs_mbuf != NULL)
2121 		bus_dmamap_unload(sc->sc_dmat, rxs->rxs_dmamap);
2122 	rxs->rxs_mbuf = m;
2123 
2124 	error = bus_dmamap_load(sc->sc_dmat, rxs->rxs_dmamap, buf,
2125 	    DGE_BUFFER_SIZE, NULL, BUS_DMA_READ|BUS_DMA_NOWAIT);
2126 #else
2127 	MCLGET(m, M_DONTWAIT);
2128 	if ((m->m_flags & M_EXT) == 0) {
2129 		m_freem(m);
2130 		return (ENOBUFS);
2131 	}
2132 
2133 	if (rxs->rxs_mbuf != NULL)
2134 		bus_dmamap_unload(sc->sc_dmat, rxs->rxs_dmamap);
2135 
2136 	rxs->rxs_mbuf = m;
2137 
2138 	m->m_len = m->m_pkthdr.len = m->m_ext.ext_size;
2139 	error = bus_dmamap_load_mbuf(sc->sc_dmat, rxs->rxs_dmamap, m,
2140 	    BUS_DMA_READ|BUS_DMA_NOWAIT);
2141 #endif
2142 	if (error) {
2143 		printf("%s: unable to load rx DMA map %d, error = %d\n",
2144 		    device_xname(&sc->sc_dev), idx, error);
2145 		panic("dge_add_rxbuf");	/* XXX XXX XXX */
2146 	}
2147 	bus_dmamap_sync(sc->sc_dmat, rxs->rxs_dmamap, 0,
2148 	    rxs->rxs_dmamap->dm_mapsize, BUS_DMASYNC_PREREAD);
2149 
2150 	return (0);
2151 }
2152 
2153 /*
2154  * dge_set_ral:
2155  *
2156  *	Set an entry in the receive address list.
2157  */
2158 static void
2159 dge_set_ral(struct dge_softc *sc, const uint8_t *enaddr, int idx)
2160 {
2161 	uint32_t ral_lo, ral_hi;
2162 
2163 	if (enaddr != NULL) {
2164 		ral_lo = enaddr[0] | (enaddr[1] << 8) | (enaddr[2] << 16) |
2165 		    (enaddr[3] << 24);
2166 		ral_hi = enaddr[4] | (enaddr[5] << 8);
2167 		ral_hi |= RAH_AV;
2168 	} else {
2169 		ral_lo = 0;
2170 		ral_hi = 0;
2171 	}
2172 	CSR_WRITE(sc, RA_ADDR(DGE_RAL, idx), ral_lo);
2173 	CSR_WRITE(sc, RA_ADDR(DGE_RAH, idx), ral_hi);
2174 }
2175 
2176 /*
2177  * dge_mchash:
2178  *
2179  *	Compute the hash of the multicast address for the 4096-bit
2180  *	multicast filter.
2181  */
2182 static uint32_t
2183 dge_mchash(struct dge_softc *sc, const uint8_t *enaddr)
2184 {
2185 	static const int lo_shift[4] = { 4, 3, 2, 0 };
2186 	static const int hi_shift[4] = { 4, 5, 6, 8 };
2187 	uint32_t hash;
2188 
2189 	hash = (enaddr[4] >> lo_shift[sc->sc_mchash_type]) |
2190 	    (((uint16_t) enaddr[5]) << hi_shift[sc->sc_mchash_type]);
2191 
2192 	return (hash & 0xfff);
2193 }
2194 
2195 /*
2196  * dge_set_filter:
2197  *
2198  *	Set up the receive filter.
2199  */
2200 static void
2201 dge_set_filter(struct dge_softc *sc)
2202 {
2203 	struct ethercom *ec = &sc->sc_ethercom;
2204 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
2205 	struct ether_multi *enm;
2206 	struct ether_multistep step;
2207 	uint32_t hash, reg, bit;
2208 	int i;
2209 
2210 	sc->sc_rctl &= ~(RCTL_BAM | RCTL_UPE | RCTL_MPE);
2211 
2212 	if (ifp->if_flags & IFF_BROADCAST)
2213 		sc->sc_rctl |= RCTL_BAM;
2214 	if (ifp->if_flags & IFF_PROMISC) {
2215 		sc->sc_rctl |= RCTL_UPE;
2216 		goto allmulti;
2217 	}
2218 
2219 	/*
2220 	 * Set the station address in the first RAL slot, and
2221 	 * clear the remaining slots.
2222 	 */
2223 	dge_set_ral(sc, CLLADDR(ifp->if_sadl), 0);
2224 	for (i = 1; i < RA_TABSIZE; i++)
2225 		dge_set_ral(sc, NULL, i);
2226 
2227 	/* Clear out the multicast table. */
2228 	for (i = 0; i < MC_TABSIZE; i++)
2229 		CSR_WRITE(sc, DGE_MTA + (i << 2), 0);
2230 
2231 	ETHER_FIRST_MULTI(step, ec, enm);
2232 	while (enm != NULL) {
2233 		if (memcmp(enm->enm_addrlo, enm->enm_addrhi, ETHER_ADDR_LEN)) {
2234 			/*
2235 			 * We must listen to a range of multicast addresses.
2236 			 * For now, just accept all multicasts, rather than
2237 			 * trying to set only those filter bits needed to match
2238 			 * the range.  (At this time, the only use of address
2239 			 * ranges is for IP multicast routing, for which the
2240 			 * range is big enough to require all bits set.)
2241 			 */
2242 			goto allmulti;
2243 		}
2244 
2245 		hash = dge_mchash(sc, enm->enm_addrlo);
2246 
2247 		reg = (hash >> 5) & 0x7f;
2248 		bit = hash & 0x1f;
2249 
2250 		hash = CSR_READ(sc, DGE_MTA + (reg << 2));
2251 		hash |= 1U << bit;
2252 
2253 		CSR_WRITE(sc, DGE_MTA + (reg << 2), hash);
2254 
2255 		ETHER_NEXT_MULTI(step, enm);
2256 	}
2257 
2258 	ifp->if_flags &= ~IFF_ALLMULTI;
2259 	goto setit;
2260 
2261  allmulti:
2262 	ifp->if_flags |= IFF_ALLMULTI;
2263 	sc->sc_rctl |= RCTL_MPE;
2264 
2265  setit:
2266 	CSR_WRITE(sc, DGE_RCTL, sc->sc_rctl);
2267 }
2268 
2269 /*
2270  * Read in the EEPROM info and verify checksum.
2271  */
2272 int
2273 dge_read_eeprom(struct dge_softc *sc)
2274 {
2275 	uint16_t cksum;
2276 	int i;
2277 
2278 	cksum = 0;
2279 	for (i = 0; i < EEPROM_SIZE; i++) {
2280 		sc->sc_eeprom[i] = dge_eeprom_word(sc, i);
2281 		cksum += sc->sc_eeprom[i];
2282 	}
2283 	return cksum != EEPROM_CKSUM;
2284 }
2285 
2286 
2287 /*
2288  * Read a 16-bit word from address addr in the serial EEPROM.
2289  */
2290 uint16_t
2291 dge_eeprom_word(struct dge_softc *sc, int addr)
2292 {
2293 	uint32_t reg;
2294 	uint16_t rval = 0;
2295 	int i;
2296 
2297 	reg = CSR_READ(sc, DGE_EECD) & ~(EECD_SK|EECD_DI|EECD_CS);
2298 
2299 	/* Lower clock pulse (and data in to chip) */
2300 	CSR_WRITE(sc, DGE_EECD, reg);
2301 	/* Select chip */
2302 	CSR_WRITE(sc, DGE_EECD, reg|EECD_CS);
2303 
2304 	/* Send read command */
2305 	dge_eeprom_clockout(sc, 1);
2306 	dge_eeprom_clockout(sc, 1);
2307 	dge_eeprom_clockout(sc, 0);
2308 
2309 	/* Send address */
2310 	for (i = 5; i >= 0; i--)
2311 		dge_eeprom_clockout(sc, (addr >> i) & 1);
2312 
2313 	/* Read data */
2314 	for (i = 0; i < 16; i++) {
2315 		rval <<= 1;
2316 		rval |= dge_eeprom_clockin(sc);
2317 	}
2318 
2319 	/* Deselect chip */
2320 	CSR_WRITE(sc, DGE_EECD, reg);
2321 
2322 	return rval;
2323 }
2324 
2325 /*
2326  * Clock out a single bit to the EEPROM.
2327  */
2328 void
2329 dge_eeprom_clockout(struct dge_softc *sc, int bit)
2330 {
2331 	int reg;
2332 
2333 	reg = CSR_READ(sc, DGE_EECD) & ~(EECD_DI|EECD_SK);
2334 	if (bit)
2335 		reg |= EECD_DI;
2336 
2337 	CSR_WRITE(sc, DGE_EECD, reg);
2338 	delay(2);
2339 	CSR_WRITE(sc, DGE_EECD, reg|EECD_SK);
2340 	delay(2);
2341 	CSR_WRITE(sc, DGE_EECD, reg);
2342 	delay(2);
2343 }
2344 
2345 /*
2346  * Clock in a single bit from EEPROM.
2347  */
2348 int
2349 dge_eeprom_clockin(struct dge_softc *sc)
2350 {
2351 	int reg, rv;
2352 
2353 	reg = CSR_READ(sc, DGE_EECD) & ~(EECD_DI|EECD_DO|EECD_SK);
2354 
2355 	CSR_WRITE(sc, DGE_EECD, reg|EECD_SK); /* Raise clock */
2356 	delay(2);
2357 	rv = (CSR_READ(sc, DGE_EECD) & EECD_DO) != 0; /* Get bit */
2358 	CSR_WRITE(sc, DGE_EECD, reg); /* Lower clock */
2359 	delay(2);
2360 
2361 	return rv;
2362 }
2363 
2364 static void
2365 dge_xgmii_mediastatus(struct ifnet *ifp, struct ifmediareq *ifmr)
2366 {
2367 	struct dge_softc *sc = ifp->if_softc;
2368 
2369 	ifmr->ifm_status = IFM_AVALID;
2370 	ifmr->ifm_active = IFM_ETHER|IFM_10G_LR;
2371 
2372 	if (CSR_READ(sc, DGE_STATUS) & STATUS_LINKUP)
2373 		ifmr->ifm_status |= IFM_ACTIVE;
2374 }
2375 
2376 static inline int
2377 phwait(struct dge_softc *sc, int p, int r, int d, int type)
2378 {
2379         int i, mdic;
2380 
2381         CSR_WRITE(sc, DGE_MDIO,
2382 	    MDIO_PHY(p) | MDIO_REG(r) | MDIO_DEV(d) | type | MDIO_CMD);
2383         for (i = 0; i < 10; i++) {
2384                 delay(10);
2385                 if (((mdic = CSR_READ(sc, DGE_MDIO)) & MDIO_CMD) == 0)
2386                         break;
2387         }
2388         return mdic;
2389 }
2390 
2391 
2392 static void
2393 dge_xgmii_writereg(device_t self, int phy, int reg, int val)
2394 {
2395 	struct dge_softc *sc = device_private(self);
2396 	int mdic;
2397 
2398 	CSR_WRITE(sc, DGE_MDIRW, val);
2399 	if (((mdic = phwait(sc, phy, reg, 1, MDIO_ADDR)) & MDIO_CMD)) {
2400 		printf("%s: address cycle timeout; phy %d reg %d\n",
2401 		    device_xname(&sc->sc_dev), phy, reg);
2402 		return;
2403 	}
2404 	if (((mdic = phwait(sc, phy, reg, 1, MDIO_WRITE)) & MDIO_CMD)) {
2405 		printf("%s: read cycle timeout; phy %d reg %d\n",
2406 		    device_xname(&sc->sc_dev), phy, reg);
2407 		return;
2408 	}
2409 }
2410 
2411 static void
2412 dge_xgmii_reset(struct dge_softc *sc)
2413 {
2414 	dge_xgmii_writereg((void *)sc, 0, 0, BMCR_RESET);
2415 }
2416 
2417 static int
2418 dge_xgmii_mediachange(struct ifnet *ifp)
2419 {
2420 	return 0;
2421 }
2422