xref: /netbsd-src/sys/dev/pci/if_dge.c (revision aa73cae19608873cc4d1f712c4a0f8f8435f1ffa)
1 /*	$NetBSD: if_dge.c,v 1.10 2005/02/27 00:27:32 perry 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 1.10 2005/02/27 00:27:32 perry Exp $");
84 
85 #include "bpfilter.h"
86 #include "rnd.h"
87 
88 #include <sys/param.h>
89 #include <sys/systm.h>
90 #include <sys/callout.h>
91 #include <sys/mbuf.h>
92 #include <sys/malloc.h>
93 #include <sys/kernel.h>
94 #include <sys/socket.h>
95 #include <sys/ioctl.h>
96 #include <sys/errno.h>
97 #include <sys/device.h>
98 #include <sys/queue.h>
99 
100 #include <uvm/uvm_extern.h>		/* for PAGE_SIZE */
101 
102 #if NRND > 0
103 #include <sys/rnd.h>
104 #endif
105 
106 #include <net/if.h>
107 #include <net/if_dl.h>
108 #include <net/if_media.h>
109 #include <net/if_ether.h>
110 
111 #if NBPFILTER > 0
112 #include <net/bpf.h>
113 #endif
114 
115 #include <netinet/in.h>			/* XXX for struct ip */
116 #include <netinet/in_systm.h>		/* XXX for struct ip */
117 #include <netinet/ip.h>			/* XXX for struct ip */
118 #include <netinet/tcp.h>		/* XXX for struct tcphdr */
119 
120 #include <machine/bus.h>
121 #include <machine/intr.h>
122 #include <machine/endian.h>
123 
124 #include <dev/mii/mii.h>
125 #include <dev/mii/miivar.h>
126 #include <dev/mii/mii_bitbang.h>
127 
128 #include <dev/pci/pcireg.h>
129 #include <dev/pci/pcivar.h>
130 #include <dev/pci/pcidevs.h>
131 
132 #include <dev/pci/if_dgereg.h>
133 
134 /*
135  * The receive engine may sometimes become off-by-one when writing back
136  * chained descriptors.	 Avoid this by allocating a large chunk of
137  * memory and use if instead (to avoid chained descriptors).
138  * This only happens with chained descriptors under heavy load.
139  */
140 #define DGE_OFFBYONE_RXBUG
141 
142 #define DGE_EVENT_COUNTERS
143 #define DGE_DEBUG
144 
145 #ifdef DGE_DEBUG
146 #define DGE_DEBUG_LINK		0x01
147 #define DGE_DEBUG_TX		0x02
148 #define DGE_DEBUG_RX		0x04
149 #define DGE_DEBUG_CKSUM		0x08
150 int	dge_debug = 0;
151 
152 #define DPRINTF(x, y)	if (dge_debug & (x)) printf y
153 #else
154 #define DPRINTF(x, y)	/* nothing */
155 #endif /* DGE_DEBUG */
156 
157 /*
158  * Transmit descriptor list size. We allow up to 100 DMA segments per
159  * packet (Intel reports of jumbo frame packets with as
160  * many as 80 DMA segments when using 16k buffers).
161  */
162 #define DGE_NTXSEGS		100
163 #define DGE_IFQUEUELEN		20000
164 #define DGE_TXQUEUELEN		2048
165 #define DGE_TXQUEUELEN_MASK	(DGE_TXQUEUELEN - 1)
166 #define DGE_TXQUEUE_GC		(DGE_TXQUEUELEN / 8)
167 #define DGE_NTXDESC		1024
168 #define DGE_NTXDESC_MASK		(DGE_NTXDESC - 1)
169 #define DGE_NEXTTX(x)		(((x) + 1) & DGE_NTXDESC_MASK)
170 #define DGE_NEXTTXS(x)		(((x) + 1) & DGE_TXQUEUELEN_MASK)
171 
172 /*
173  * Receive descriptor list size.
174  * Packet is of size MCLBYTES, and for jumbo packets buffers may
175  * be chained.	Due to the nature of the card (high-speed), keep this
176  * ring large. With 2k buffers the ring can store 400 jumbo packets,
177  * which at full speed will be received in just under 3ms.
178  */
179 #define DGE_NRXDESC		2048
180 #define DGE_NRXDESC_MASK	(DGE_NRXDESC - 1)
181 #define DGE_NEXTRX(x)		(((x) + 1) & DGE_NRXDESC_MASK)
182 /*
183  * # of descriptors between head and written descriptors.
184  * This is to work-around two erratas.
185  */
186 #define DGE_RXSPACE		10
187 #define DGE_PREVRX(x)		(((x) - DGE_RXSPACE) & DGE_NRXDESC_MASK)
188 /*
189  * Receive descriptor fetch threshholds. These are values recommended
190  * by Intel, do not touch them unless you know what you are doing.
191  */
192 #define RXDCTL_PTHRESH_VAL	128
193 #define RXDCTL_HTHRESH_VAL	16
194 #define RXDCTL_WTHRESH_VAL	16
195 
196 
197 /*
198  * Tweakable parameters; default values.
199  */
200 #define FCRTH	0x30000 /* Send XOFF water mark */
201 #define FCRTL	0x28000 /* Send XON water mark */
202 #define RDTR	0x20	/* Interrupt delay after receive, .8192us units */
203 #define TIDV	0x20	/* Interrupt delay after send, .8192us units */
204 
205 /*
206  * Control structures are DMA'd to the i82597 chip.  We allocate them in
207  * a single clump that maps to a single DMA segment to make serveral things
208  * easier.
209  */
210 struct dge_control_data {
211 	/*
212 	 * The transmit descriptors.
213 	 */
214 	struct dge_tdes wcd_txdescs[DGE_NTXDESC];
215 
216 	/*
217 	 * The receive descriptors.
218 	 */
219 	struct dge_rdes wcd_rxdescs[DGE_NRXDESC];
220 };
221 
222 #define DGE_CDOFF(x)	offsetof(struct dge_control_data, x)
223 #define DGE_CDTXOFF(x)	DGE_CDOFF(wcd_txdescs[(x)])
224 #define DGE_CDRXOFF(x)	DGE_CDOFF(wcd_rxdescs[(x)])
225 
226 /*
227  * The DGE interface have a higher max MTU size than normal jumbo frames.
228  */
229 #define DGE_MAX_MTU	16288	/* Max MTU size for this interface */
230 
231 /*
232  * Software state for transmit jobs.
233  */
234 struct dge_txsoft {
235 	struct mbuf *txs_mbuf;		/* head of our mbuf chain */
236 	bus_dmamap_t txs_dmamap;	/* our DMA map */
237 	int txs_firstdesc;		/* first descriptor in packet */
238 	int txs_lastdesc;		/* last descriptor in packet */
239 	int txs_ndesc;			/* # of descriptors used */
240 };
241 
242 /*
243  * Software state for receive buffers.	Each descriptor gets a
244  * 2k (MCLBYTES) buffer and a DMA map.	For packets which fill
245  * more than one buffer, we chain them together.
246  */
247 struct dge_rxsoft {
248 	struct mbuf *rxs_mbuf;		/* head of our mbuf chain */
249 	bus_dmamap_t rxs_dmamap;	/* our DMA map */
250 };
251 
252 /*
253  * Software state per device.
254  */
255 struct dge_softc {
256 	struct device sc_dev;		/* generic device information */
257 	bus_space_tag_t sc_st;		/* bus space tag */
258 	bus_space_handle_t sc_sh;	/* bus space handle */
259 	bus_dma_tag_t sc_dmat;		/* bus DMA tag */
260 	struct ethercom sc_ethercom;	/* ethernet common data */
261 	void *sc_sdhook;		/* shutdown hook */
262 
263 	int sc_flags;			/* flags; see below */
264 	int sc_bus_speed;		/* PCI/PCIX bus speed */
265 	int sc_pcix_offset;		/* PCIX capability register offset */
266 
267 	pci_chipset_tag_t sc_pc;
268 	pcitag_t sc_pt;
269 	int sc_mmrbc;			/* Max PCIX memory read byte count */
270 
271 	void *sc_ih;			/* interrupt cookie */
272 
273 	struct ifmedia sc_media;
274 
275 	bus_dmamap_t sc_cddmamap;	/* control data DMA map */
276 #define sc_cddma	sc_cddmamap->dm_segs[0].ds_addr
277 
278 	int		sc_align_tweak;
279 
280 	/*
281 	 * Software state for the transmit and receive descriptors.
282 	 */
283 	struct dge_txsoft sc_txsoft[DGE_TXQUEUELEN];
284 	struct dge_rxsoft sc_rxsoft[DGE_NRXDESC];
285 
286 	/*
287 	 * Control data structures.
288 	 */
289 	struct dge_control_data *sc_control_data;
290 #define sc_txdescs	sc_control_data->wcd_txdescs
291 #define sc_rxdescs	sc_control_data->wcd_rxdescs
292 
293 #ifdef DGE_EVENT_COUNTERS
294 	/* Event counters. */
295 	struct evcnt sc_ev_txsstall;	/* Tx stalled due to no txs */
296 	struct evcnt sc_ev_txdstall;	/* Tx stalled due to no txd */
297 	struct evcnt sc_ev_txforceintr; /* Tx interrupts forced */
298 	struct evcnt sc_ev_txdw;	/* Tx descriptor interrupts */
299 	struct evcnt sc_ev_txqe;	/* Tx queue empty interrupts */
300 	struct evcnt sc_ev_rxintr;	/* Rx interrupts */
301 	struct evcnt sc_ev_linkintr;	/* Link interrupts */
302 
303 	struct evcnt sc_ev_rxipsum;	/* IP checksums checked in-bound */
304 	struct evcnt sc_ev_rxtusum;	/* TCP/UDP cksums checked in-bound */
305 	struct evcnt sc_ev_txipsum;	/* IP checksums comp. out-bound */
306 	struct evcnt sc_ev_txtusum;	/* TCP/UDP cksums comp. out-bound */
307 
308 	struct evcnt sc_ev_txctx_init;	/* Tx cksum context cache initialized */
309 	struct evcnt sc_ev_txctx_hit;	/* Tx cksum context cache hit */
310 	struct evcnt sc_ev_txctx_miss;	/* Tx cksum context cache miss */
311 
312 	struct evcnt sc_ev_txseg[DGE_NTXSEGS]; /* Tx packets w/ N segments */
313 	struct evcnt sc_ev_txdrop;	/* Tx packets dropped (too many segs) */
314 #endif /* DGE_EVENT_COUNTERS */
315 
316 	int	sc_txfree;		/* number of free Tx descriptors */
317 	int	sc_txnext;		/* next ready Tx descriptor */
318 
319 	int	sc_txsfree;		/* number of free Tx jobs */
320 	int	sc_txsnext;		/* next free Tx job */
321 	int	sc_txsdirty;		/* dirty Tx jobs */
322 
323 	uint32_t sc_txctx_ipcs;		/* cached Tx IP cksum ctx */
324 	uint32_t sc_txctx_tucs;		/* cached Tx TCP/UDP cksum ctx */
325 
326 	int	sc_rxptr;		/* next ready Rx descriptor/queue ent */
327 	int	sc_rxdiscard;
328 	int	sc_rxlen;
329 	struct mbuf *sc_rxhead;
330 	struct mbuf *sc_rxtail;
331 	struct mbuf **sc_rxtailp;
332 
333 	uint32_t sc_ctrl0;		/* prototype CTRL0 register */
334 	uint32_t sc_icr;		/* prototype interrupt bits */
335 	uint32_t sc_tctl;		/* prototype TCTL register */
336 	uint32_t sc_rctl;		/* prototype RCTL register */
337 
338 	int sc_mchash_type;		/* multicast filter offset */
339 
340 	uint16_t sc_eeprom[EEPROM_SIZE];
341 
342 #if NRND > 0
343 	rndsource_element_t rnd_source; /* random source */
344 #endif
345 #ifdef DGE_OFFBYONE_RXBUG
346 	caddr_t sc_bugbuf;
347 	SLIST_HEAD(, rxbugentry) sc_buglist;
348 	bus_dmamap_t sc_bugmap;
349 	struct rxbugentry *sc_entry;
350 #endif
351 };
352 
353 #define DGE_RXCHAIN_RESET(sc)						\
354 do {									\
355 	(sc)->sc_rxtailp = &(sc)->sc_rxhead;				\
356 	*(sc)->sc_rxtailp = NULL;					\
357 	(sc)->sc_rxlen = 0;						\
358 } while (/*CONSTCOND*/0)
359 
360 #define DGE_RXCHAIN_LINK(sc, m)						\
361 do {									\
362 	*(sc)->sc_rxtailp = (sc)->sc_rxtail = (m);			\
363 	(sc)->sc_rxtailp = &(m)->m_next;				\
364 } while (/*CONSTCOND*/0)
365 
366 /* sc_flags */
367 #define DGE_F_BUS64		0x20	/* bus is 64-bit */
368 #define DGE_F_PCIX		0x40	/* bus is PCI-X */
369 
370 #ifdef DGE_EVENT_COUNTERS
371 #define DGE_EVCNT_INCR(ev)	(ev)->ev_count++
372 #else
373 #define DGE_EVCNT_INCR(ev)	/* nothing */
374 #endif
375 
376 #define CSR_READ(sc, reg)						\
377 	bus_space_read_4((sc)->sc_st, (sc)->sc_sh, (reg))
378 #define CSR_WRITE(sc, reg, val)						\
379 	bus_space_write_4((sc)->sc_st, (sc)->sc_sh, (reg), (val))
380 
381 #define DGE_CDTXADDR(sc, x)	((sc)->sc_cddma + DGE_CDTXOFF((x)))
382 #define DGE_CDRXADDR(sc, x)	((sc)->sc_cddma + DGE_CDRXOFF((x)))
383 
384 #define DGE_CDTXSYNC(sc, x, n, ops)					\
385 do {									\
386 	int __x, __n;							\
387 									\
388 	__x = (x);							\
389 	__n = (n);							\
390 									\
391 	/* If it will wrap around, sync to the end of the ring. */	\
392 	if ((__x + __n) > DGE_NTXDESC) {				\
393 		bus_dmamap_sync((sc)->sc_dmat, (sc)->sc_cddmamap,	\
394 		    DGE_CDTXOFF(__x), sizeof(struct dge_tdes) *		\
395 		    (DGE_NTXDESC - __x), (ops));			\
396 		__n -= (DGE_NTXDESC - __x);				\
397 		__x = 0;						\
398 	}								\
399 									\
400 	/* Now sync whatever is left. */				\
401 	bus_dmamap_sync((sc)->sc_dmat, (sc)->sc_cddmamap,		\
402 	    DGE_CDTXOFF(__x), sizeof(struct dge_tdes) * __n, (ops));	\
403 } while (/*CONSTCOND*/0)
404 
405 #define DGE_CDRXSYNC(sc, x, ops)						\
406 do {									\
407 	bus_dmamap_sync((sc)->sc_dmat, (sc)->sc_cddmamap,		\
408 	   DGE_CDRXOFF((x)), sizeof(struct dge_rdes), (ops));		\
409 } while (/*CONSTCOND*/0)
410 
411 #ifdef DGE_OFFBYONE_RXBUG
412 #define DGE_INIT_RXDESC(sc, x)						\
413 do {									\
414 	struct dge_rxsoft *__rxs = &(sc)->sc_rxsoft[(x)];		\
415 	struct dge_rdes *__rxd = &(sc)->sc_rxdescs[(x)];		\
416 	struct mbuf *__m = __rxs->rxs_mbuf;				\
417 									\
418 	__rxd->dr_baddrl = htole32(sc->sc_bugmap->dm_segs[0].ds_addr +	\
419 	    (mtod((__m), char *) - (char *)sc->sc_bugbuf));		\
420 	__rxd->dr_baddrh = 0;						\
421 	__rxd->dr_len = 0;						\
422 	__rxd->dr_cksum = 0;						\
423 	__rxd->dr_status = 0;						\
424 	__rxd->dr_errors = 0;						\
425 	__rxd->dr_special = 0;						\
426 	DGE_CDRXSYNC((sc), (x), BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE); \
427 									\
428 	CSR_WRITE((sc), DGE_RDT, (x));					\
429 } while (/*CONSTCOND*/0)
430 #else
431 #define DGE_INIT_RXDESC(sc, x)						\
432 do {									\
433 	struct dge_rxsoft *__rxs = &(sc)->sc_rxsoft[(x)];		\
434 	struct dge_rdes *__rxd = &(sc)->sc_rxdescs[(x)];		\
435 	struct mbuf *__m = __rxs->rxs_mbuf;				\
436 									\
437 	/*								\
438 	 * Note: We scoot the packet forward 2 bytes in the buffer	\
439 	 * so that the payload after the Ethernet header is aligned	\
440 	 * to a 4-byte boundary.					\
441 	 *								\
442 	 * XXX BRAINDAMAGE ALERT!					\
443 	 * The stupid chip uses the same size for every buffer, which	\
444 	 * is set in the Receive Control register.  We are using the 2K \
445 	 * size option, but what we REALLY want is (2K - 2)!  For this	\
446 	 * reason, we can't "scoot" packets longer than the standard	\
447 	 * Ethernet MTU.  On strict-alignment platforms, if the total	\
448 	 * size exceeds (2K - 2) we set align_tweak to 0 and let	\
449 	 * the upper layer copy the headers.				\
450 	 */								\
451 	__m->m_data = __m->m_ext.ext_buf + (sc)->sc_align_tweak;	\
452 									\
453 	__rxd->dr_baddrl =						\
454 	    htole32(__rxs->rxs_dmamap->dm_segs[0].ds_addr +		\
455 		(sc)->sc_align_tweak);					\
456 	__rxd->dr_baddrh = 0;						\
457 	__rxd->dr_len = 0;						\
458 	__rxd->dr_cksum = 0;						\
459 	__rxd->dr_status = 0;						\
460 	__rxd->dr_errors = 0;						\
461 	__rxd->dr_special = 0;						\
462 	DGE_CDRXSYNC((sc), (x), BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE); \
463 									\
464 	CSR_WRITE((sc), DGE_RDT, (x));					\
465 } while (/*CONSTCOND*/0)
466 #endif
467 
468 #ifdef DGE_OFFBYONE_RXBUG
469 /*
470  * Allocation constants.  Much memory may be used for this.
471  */
472 #ifndef DGE_BUFFER_SIZE
473 #define DGE_BUFFER_SIZE DGE_MAX_MTU
474 #endif
475 #define DGE_NBUFFERS	(4*DGE_NRXDESC)
476 #define DGE_RXMEM	(DGE_NBUFFERS*DGE_BUFFER_SIZE)
477 
478 struct rxbugentry {
479 	SLIST_ENTRY(rxbugentry) rb_entry;
480 	int rb_slot;
481 };
482 
483 static int
484 dge_alloc_rcvmem(struct dge_softc *sc)
485 {
486 	caddr_t	ptr, kva;
487 	bus_dma_segment_t seg;
488 	int i, rseg, state, error;
489 	struct rxbugentry *entry;
490 
491 	state = error = 0;
492 
493 	if (bus_dmamem_alloc(sc->sc_dmat, DGE_RXMEM, PAGE_SIZE, 0,
494 	     &seg, 1, &rseg, BUS_DMA_NOWAIT)) {
495 		printf("%s: can't alloc rx buffers\n", sc->sc_dev.dv_xname);
496 		return ENOBUFS;
497 	}
498 
499 	state = 1;
500 	if (bus_dmamem_map(sc->sc_dmat, &seg, rseg, DGE_RXMEM, &kva,
501 	    BUS_DMA_NOWAIT)) {
502 		printf("%s: can't map DMA buffers (%d bytes)\n",
503 		    sc->sc_dev.dv_xname, (int)DGE_RXMEM);
504 		error = ENOBUFS;
505 		goto out;
506 	}
507 
508 	state = 2;
509 	if (bus_dmamap_create(sc->sc_dmat, DGE_RXMEM, 1, DGE_RXMEM, 0,
510 	    BUS_DMA_NOWAIT, &sc->sc_bugmap)) {
511 		printf("%s: can't create DMA map\n", sc->sc_dev.dv_xname);
512 		error = ENOBUFS;
513 		goto out;
514 	}
515 
516 	state = 3;
517 	if (bus_dmamap_load(sc->sc_dmat, sc->sc_bugmap,
518 	    kva, DGE_RXMEM, NULL, BUS_DMA_NOWAIT)) {
519 		printf("%s: can't load DMA map\n", sc->sc_dev.dv_xname);
520 		error = ENOBUFS;
521 		goto out;
522 	}
523 
524 	state = 4;
525 	sc->sc_bugbuf = (caddr_t)kva;
526 	SLIST_INIT(&sc->sc_buglist);
527 
528 	/*
529 	 * Now divide it up into DGE_BUFFER_SIZE pieces and save the addresses
530 	 * in an array.
531 	 */
532 	ptr = sc->sc_bugbuf;
533 	if ((entry = malloc(sizeof(*entry) * DGE_NBUFFERS,
534 	    M_DEVBUF, M_NOWAIT)) == NULL) {
535 		error = ENOBUFS;
536 		goto out;
537 	}
538 	sc->sc_entry = entry;
539 	for (i = 0; i < DGE_NBUFFERS; i++) {
540 		entry[i].rb_slot = i;
541 		SLIST_INSERT_HEAD(&sc->sc_buglist, &entry[i], rb_entry);
542 	}
543 out:
544 	if (error != 0) {
545 		switch (state) {
546 		case 4:
547 			bus_dmamap_unload(sc->sc_dmat, sc->sc_bugmap);
548 		case 3:
549 			bus_dmamap_destroy(sc->sc_dmat, sc->sc_bugmap);
550 		case 2:
551 			bus_dmamem_unmap(sc->sc_dmat, kva, DGE_RXMEM);
552 		case 1:
553 			bus_dmamem_free(sc->sc_dmat, &seg, rseg);
554 			break;
555 		default:
556 			break;
557 		}
558 	}
559 
560 	return error;
561 }
562 
563 /*
564  * Allocate a jumbo buffer.
565  */
566 static void *
567 dge_getbuf(struct dge_softc *sc)
568 {
569 	struct rxbugentry *entry;
570 
571 	entry = SLIST_FIRST(&sc->sc_buglist);
572 
573 	if (entry == NULL) {
574 		printf("%s: no free RX buffers\n", sc->sc_dev.dv_xname);
575 		return(NULL);
576 	}
577 
578 	SLIST_REMOVE_HEAD(&sc->sc_buglist, rb_entry);
579 	return sc->sc_bugbuf + entry->rb_slot * DGE_BUFFER_SIZE;
580 }
581 
582 /*
583  * Release a jumbo buffer.
584  */
585 static void
586 dge_freebuf(struct mbuf *m, caddr_t buf, size_t size, void *arg)
587 {
588 	struct rxbugentry *entry;
589 	struct dge_softc *sc;
590 	int i, s;
591 
592 	/* Extract the softc struct pointer. */
593 	sc = (struct dge_softc *)arg;
594 
595 	if (sc == NULL)
596 		panic("dge_freebuf: can't find softc pointer!");
597 
598 	/* calculate the slot this buffer belongs to */
599 
600 	i = (buf - sc->sc_bugbuf) / DGE_BUFFER_SIZE;
601 
602 	if ((i < 0) || (i >= DGE_NBUFFERS))
603 		panic("dge_freebuf: asked to free buffer %d!", i);
604 
605 	s = splvm();
606 	entry = sc->sc_entry + i;
607 	SLIST_INSERT_HEAD(&sc->sc_buglist, entry, rb_entry);
608 
609 	if (__predict_true(m != NULL))
610 		pool_cache_put(&mbpool_cache, m);
611 	splx(s);
612 }
613 #endif
614 
615 static void	dge_start(struct ifnet *);
616 static void	dge_watchdog(struct ifnet *);
617 static int	dge_ioctl(struct ifnet *, u_long, caddr_t);
618 static int	dge_init(struct ifnet *);
619 static void	dge_stop(struct ifnet *, int);
620 
621 static void	dge_shutdown(void *);
622 
623 static void	dge_reset(struct dge_softc *);
624 static void	dge_rxdrain(struct dge_softc *);
625 static int	dge_add_rxbuf(struct dge_softc *, int);
626 
627 static void	dge_set_filter(struct dge_softc *);
628 
629 static int	dge_intr(void *);
630 static void	dge_txintr(struct dge_softc *);
631 static void	dge_rxintr(struct dge_softc *);
632 static void	dge_linkintr(struct dge_softc *, uint32_t);
633 
634 static int	dge_match(struct device *, struct cfdata *, void *);
635 static void	dge_attach(struct device *, struct device *, void *);
636 
637 static int	dge_read_eeprom(struct dge_softc *sc);
638 static int	dge_eeprom_clockin(struct dge_softc *sc);
639 static void	dge_eeprom_clockout(struct dge_softc *sc, int bit);
640 static uint16_t	dge_eeprom_word(struct dge_softc *sc, int addr);
641 static int	dge_xgmii_mediachange(struct ifnet *);
642 static void	dge_xgmii_mediastatus(struct ifnet *, struct ifmediareq *);
643 static void	dge_xgmii_reset(struct dge_softc *);
644 static void	dge_xgmii_writereg(struct device *, int, int, int);
645 
646 
647 CFATTACH_DECL(dge, sizeof(struct dge_softc),
648     dge_match, dge_attach, NULL, NULL);
649 
650 #ifdef DGE_EVENT_COUNTERS
651 #if DGE_NTXSEGS > 100
652 #error Update dge_txseg_evcnt_names
653 #endif
654 static char (*dge_txseg_evcnt_names)[DGE_NTXSEGS][8 /* "txseg00" + \0 */];
655 #endif /* DGE_EVENT_COUNTERS */
656 
657 static int
658 dge_match(struct device *parent, struct cfdata *cf, void *aux)
659 {
660 	struct pci_attach_args *pa = aux;
661 
662 	if (PCI_VENDOR(pa->pa_id) == PCI_VENDOR_INTEL &&
663 	    PCI_PRODUCT(pa->pa_id) == PCI_PRODUCT_INTEL_82597EX)
664 		return (1);
665 
666 	return (0);
667 }
668 
669 static void
670 dge_attach(struct device *parent, struct device *self, void *aux)
671 {
672 	struct dge_softc *sc = (void *) self;
673 	struct pci_attach_args *pa = aux;
674 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
675 	pci_chipset_tag_t pc = pa->pa_pc;
676 	pci_intr_handle_t ih;
677 	const char *intrstr = NULL;
678 	bus_dma_segment_t seg;
679 	int i, rseg, error;
680 	uint8_t enaddr[ETHER_ADDR_LEN];
681 	pcireg_t preg, memtype;
682 	uint32_t reg;
683 
684 	sc->sc_dmat = pa->pa_dmat;
685 	sc->sc_pc = pa->pa_pc;
686 	sc->sc_pt = pa->pa_tag;
687 
688 	preg = PCI_REVISION(pci_conf_read(pc, pa->pa_tag, PCI_CLASS_REG));
689 	aprint_naive(": Ethernet controller\n");
690 	aprint_normal(": Intel i82597EX 10GbE-LR Ethernet, rev. %d\n", preg);
691 
692 	memtype = pci_mapreg_type(pa->pa_pc, pa->pa_tag, DGE_PCI_BAR);
693         if (pci_mapreg_map(pa, DGE_PCI_BAR, memtype, 0,
694             &sc->sc_st, &sc->sc_sh, NULL, NULL)) {
695                 aprint_error("%s: unable to map device registers\n",
696                     sc->sc_dev.dv_xname);
697                 return;
698         }
699 
700 	/* Enable bus mastering */
701 	preg = pci_conf_read(pc, pa->pa_tag, PCI_COMMAND_STATUS_REG);
702 	preg |= PCI_COMMAND_MASTER_ENABLE;
703 	pci_conf_write(pc, pa->pa_tag, PCI_COMMAND_STATUS_REG, preg);
704 
705 	/*
706 	 * Map and establish our interrupt.
707 	 */
708 	if (pci_intr_map(pa, &ih)) {
709 		aprint_error("%s: unable to map interrupt\n",
710 		    sc->sc_dev.dv_xname);
711 		return;
712 	}
713 	intrstr = pci_intr_string(pc, ih);
714 	sc->sc_ih = pci_intr_establish(pc, ih, IPL_NET, dge_intr, sc);
715 	if (sc->sc_ih == NULL) {
716 		aprint_error("%s: unable to establish interrupt",
717 		    sc->sc_dev.dv_xname);
718 		if (intrstr != NULL)
719 			aprint_normal(" at %s", intrstr);
720 		aprint_normal("\n");
721 		return;
722 	}
723 	aprint_normal("%s: interrupting at %s\n", sc->sc_dev.dv_xname, intrstr);
724 
725 	/*
726 	 * Determine a few things about the bus we're connected to.
727 	 */
728 	reg = CSR_READ(sc, DGE_STATUS);
729 	if (reg & STATUS_BUS64)
730 		sc->sc_flags |= DGE_F_BUS64;
731 
732 	sc->sc_flags |= DGE_F_PCIX;
733 	if (pci_get_capability(pa->pa_pc, pa->pa_tag,
734 			       PCI_CAP_PCIX,
735 			       &sc->sc_pcix_offset, NULL) == 0)
736 		aprint_error("%s: unable to find PCIX "
737 		    "capability\n", sc->sc_dev.dv_xname);
738 
739 	if (sc->sc_flags & DGE_F_PCIX) {
740 		switch (reg & STATUS_PCIX_MSK) {
741 		case STATUS_PCIX_66:
742 			sc->sc_bus_speed = 66;
743 			break;
744 		case STATUS_PCIX_100:
745 			sc->sc_bus_speed = 100;
746 			break;
747 		case STATUS_PCIX_133:
748 			sc->sc_bus_speed = 133;
749 			break;
750 		default:
751 			aprint_error(
752 			    "%s: unknown PCIXSPD %d; assuming 66MHz\n",
753 			    sc->sc_dev.dv_xname,
754 			    reg & STATUS_PCIX_MSK);
755 			sc->sc_bus_speed = 66;
756 		}
757 	} else
758 		sc->sc_bus_speed = (reg & STATUS_BUS64) ? 66 : 33;
759 	aprint_verbose("%s: %d-bit %dMHz %s bus\n", sc->sc_dev.dv_xname,
760 	    (sc->sc_flags & DGE_F_BUS64) ? 64 : 32, sc->sc_bus_speed,
761 	    (sc->sc_flags & DGE_F_PCIX) ? "PCIX" : "PCI");
762 
763 	/*
764 	 * Allocate the control data structures, and create and load the
765 	 * DMA map for it.
766 	 */
767 	if ((error = bus_dmamem_alloc(sc->sc_dmat,
768 	    sizeof(struct dge_control_data), PAGE_SIZE, 0, &seg, 1, &rseg,
769 	    0)) != 0) {
770 		aprint_error(
771 		    "%s: unable to allocate control data, error = %d\n",
772 		    sc->sc_dev.dv_xname, error);
773 		goto fail_0;
774 	}
775 
776 	if ((error = bus_dmamem_map(sc->sc_dmat, &seg, rseg,
777 	    sizeof(struct dge_control_data), (caddr_t *)&sc->sc_control_data,
778 	    0)) != 0) {
779 		aprint_error("%s: unable to map control data, error = %d\n",
780 		    sc->sc_dev.dv_xname, error);
781 		goto fail_1;
782 	}
783 
784 	if ((error = bus_dmamap_create(sc->sc_dmat,
785 	    sizeof(struct dge_control_data), 1,
786 	    sizeof(struct dge_control_data), 0, 0, &sc->sc_cddmamap)) != 0) {
787 		aprint_error("%s: unable to create control data DMA map, "
788 		    "error = %d\n", sc->sc_dev.dv_xname, error);
789 		goto fail_2;
790 	}
791 
792 	if ((error = bus_dmamap_load(sc->sc_dmat, sc->sc_cddmamap,
793 	    sc->sc_control_data, sizeof(struct dge_control_data), NULL,
794 	    0)) != 0) {
795 		aprint_error(
796 		    "%s: unable to load control data DMA map, error = %d\n",
797 		    sc->sc_dev.dv_xname, error);
798 		goto fail_3;
799 	}
800 
801 #ifdef DGE_OFFBYONE_RXBUG
802 	if (dge_alloc_rcvmem(sc) != 0)
803 		return; /* Already complained */
804 #endif
805 	/*
806 	 * Create the transmit buffer DMA maps.
807 	 */
808 	for (i = 0; i < DGE_TXQUEUELEN; i++) {
809 		if ((error = bus_dmamap_create(sc->sc_dmat, DGE_MAX_MTU,
810 		    DGE_NTXSEGS, MCLBYTES, 0, 0,
811 		    &sc->sc_txsoft[i].txs_dmamap)) != 0) {
812 			aprint_error("%s: unable to create Tx DMA map %d, "
813 			    "error = %d\n", sc->sc_dev.dv_xname, i, error);
814 			goto fail_4;
815 		}
816 	}
817 
818 	/*
819 	 * Create the receive buffer DMA maps.
820 	 */
821 	for (i = 0; i < DGE_NRXDESC; i++) {
822 #ifdef DGE_OFFBYONE_RXBUG
823 		if ((error = bus_dmamap_create(sc->sc_dmat, DGE_BUFFER_SIZE, 1,
824 		    DGE_BUFFER_SIZE, 0, 0, &sc->sc_rxsoft[i].rxs_dmamap)) != 0) {
825 #else
826 		if ((error = bus_dmamap_create(sc->sc_dmat, MCLBYTES, 1,
827 		    MCLBYTES, 0, 0, &sc->sc_rxsoft[i].rxs_dmamap)) != 0) {
828 #endif
829 			aprint_error("%s: unable to create Rx DMA map %d, "
830 			    "error = %d\n", sc->sc_dev.dv_xname, i, error);
831 			goto fail_5;
832 		}
833 		sc->sc_rxsoft[i].rxs_mbuf = NULL;
834 	}
835 
836 	/*
837 	 * Set bits in ctrl0 register.
838 	 * Should get the software defined pins out of EEPROM?
839 	 */
840 	sc->sc_ctrl0 |= CTRL0_RPE | CTRL0_TPE; /* XON/XOFF */
841 	sc->sc_ctrl0 |= CTRL0_SDP3_DIR | CTRL0_SDP2_DIR | CTRL0_SDP1_DIR |
842 	    CTRL0_SDP0_DIR | CTRL0_SDP3 | CTRL0_SDP2 | CTRL0_SDP0;
843 
844 	/*
845 	 * Reset the chip to a known state.
846 	 */
847 	dge_reset(sc);
848 
849 	/*
850 	 * Reset the PHY.
851 	 */
852 	dge_xgmii_reset(sc);
853 
854 	/*
855 	 * Read in EEPROM data.
856 	 */
857 	if (dge_read_eeprom(sc)) {
858 		aprint_error("%s: couldn't read EEPROM\n", sc->sc_dev.dv_xname);
859 		return;
860 	}
861 
862 	/*
863 	 * Get the ethernet address.
864 	 */
865 	enaddr[0] = sc->sc_eeprom[EE_ADDR01] & 0377;
866 	enaddr[1] = sc->sc_eeprom[EE_ADDR01] >> 8;
867 	enaddr[2] = sc->sc_eeprom[EE_ADDR23] & 0377;
868 	enaddr[3] = sc->sc_eeprom[EE_ADDR23] >> 8;
869 	enaddr[4] = sc->sc_eeprom[EE_ADDR45] & 0377;
870 	enaddr[5] = sc->sc_eeprom[EE_ADDR45] >> 8;
871 
872 	aprint_normal("%s: Ethernet address %s\n", sc->sc_dev.dv_xname,
873 	    ether_sprintf(enaddr));
874 
875 	/*
876 	 * Setup media stuff.
877 	 */
878         ifmedia_init(&sc->sc_media, IFM_IMASK, dge_xgmii_mediachange,
879             dge_xgmii_mediastatus);
880         ifmedia_add(&sc->sc_media, IFM_ETHER|IFM_10G_LR, 0, NULL);
881         ifmedia_set(&sc->sc_media, IFM_ETHER|IFM_10G_LR);
882 
883 	ifp = &sc->sc_ethercom.ec_if;
884 	strcpy(ifp->if_xname, sc->sc_dev.dv_xname);
885 	ifp->if_softc = sc;
886 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
887 	ifp->if_ioctl = dge_ioctl;
888 	ifp->if_start = dge_start;
889 	ifp->if_watchdog = dge_watchdog;
890 	ifp->if_init = dge_init;
891 	ifp->if_stop = dge_stop;
892 	IFQ_SET_MAXLEN(&ifp->if_snd, max(DGE_IFQUEUELEN, IFQ_MAXLEN));
893 	IFQ_SET_READY(&ifp->if_snd);
894 
895 	sc->sc_ethercom.ec_capabilities |=
896 	    ETHERCAP_JUMBO_MTU | ETHERCAP_VLAN_MTU;
897 
898 	/*
899 	 * We can perform TCPv4 and UDPv4 checkums in-bound.
900 	 */
901 	ifp->if_capabilities |=
902 	    IFCAP_CSUM_IPv4 | IFCAP_CSUM_TCPv4 | IFCAP_CSUM_UDPv4;
903 
904 	/*
905 	 * Attach the interface.
906 	 */
907 	if_attach(ifp);
908 	ether_ifattach(ifp, enaddr);
909 #if NRND > 0
910 	rnd_attach_source(&sc->rnd_source, sc->sc_dev.dv_xname,
911 	    RND_TYPE_NET, 0);
912 #endif
913 
914 #ifdef DGE_EVENT_COUNTERS
915 	/* Fix segment event naming */
916 	if (dge_txseg_evcnt_names == NULL) {
917 		dge_txseg_evcnt_names =
918 		    malloc(sizeof(*dge_txseg_evcnt_names), M_DEVBUF, M_WAITOK);
919 		for (i = 0; i < DGE_NTXSEGS; i++)
920 			snprintf((*dge_txseg_evcnt_names)[i],
921 			    sizeof((*dge_txseg_evcnt_names)[i]), "txseg%d", i);
922 	}
923 
924 	/* Attach event counters. */
925 	evcnt_attach_dynamic(&sc->sc_ev_txsstall, EVCNT_TYPE_MISC,
926 	    NULL, sc->sc_dev.dv_xname, "txsstall");
927 	evcnt_attach_dynamic(&sc->sc_ev_txdstall, EVCNT_TYPE_MISC,
928 	    NULL, sc->sc_dev.dv_xname, "txdstall");
929 	evcnt_attach_dynamic(&sc->sc_ev_txforceintr, EVCNT_TYPE_MISC,
930 	    NULL, sc->sc_dev.dv_xname, "txforceintr");
931 	evcnt_attach_dynamic(&sc->sc_ev_txdw, EVCNT_TYPE_INTR,
932 	    NULL, sc->sc_dev.dv_xname, "txdw");
933 	evcnt_attach_dynamic(&sc->sc_ev_txqe, EVCNT_TYPE_INTR,
934 	    NULL, sc->sc_dev.dv_xname, "txqe");
935 	evcnt_attach_dynamic(&sc->sc_ev_rxintr, EVCNT_TYPE_INTR,
936 	    NULL, sc->sc_dev.dv_xname, "rxintr");
937 	evcnt_attach_dynamic(&sc->sc_ev_linkintr, EVCNT_TYPE_INTR,
938 	    NULL, sc->sc_dev.dv_xname, "linkintr");
939 
940 	evcnt_attach_dynamic(&sc->sc_ev_rxipsum, EVCNT_TYPE_MISC,
941 	    NULL, sc->sc_dev.dv_xname, "rxipsum");
942 	evcnt_attach_dynamic(&sc->sc_ev_rxtusum, EVCNT_TYPE_MISC,
943 	    NULL, sc->sc_dev.dv_xname, "rxtusum");
944 	evcnt_attach_dynamic(&sc->sc_ev_txipsum, EVCNT_TYPE_MISC,
945 	    NULL, sc->sc_dev.dv_xname, "txipsum");
946 	evcnt_attach_dynamic(&sc->sc_ev_txtusum, EVCNT_TYPE_MISC,
947 	    NULL, sc->sc_dev.dv_xname, "txtusum");
948 
949 	evcnt_attach_dynamic(&sc->sc_ev_txctx_init, EVCNT_TYPE_MISC,
950 	    NULL, sc->sc_dev.dv_xname, "txctx init");
951 	evcnt_attach_dynamic(&sc->sc_ev_txctx_hit, EVCNT_TYPE_MISC,
952 	    NULL, sc->sc_dev.dv_xname, "txctx hit");
953 	evcnt_attach_dynamic(&sc->sc_ev_txctx_miss, EVCNT_TYPE_MISC,
954 	    NULL, sc->sc_dev.dv_xname, "txctx miss");
955 
956 	for (i = 0; i < DGE_NTXSEGS; i++)
957 		evcnt_attach_dynamic(&sc->sc_ev_txseg[i], EVCNT_TYPE_MISC,
958 		    NULL, sc->sc_dev.dv_xname, (*dge_txseg_evcnt_names)[i]);
959 
960 	evcnt_attach_dynamic(&sc->sc_ev_txdrop, EVCNT_TYPE_MISC,
961 	    NULL, sc->sc_dev.dv_xname, "txdrop");
962 
963 #endif /* DGE_EVENT_COUNTERS */
964 
965 	/*
966 	 * Make sure the interface is shutdown during reboot.
967 	 */
968 	sc->sc_sdhook = shutdownhook_establish(dge_shutdown, sc);
969 	if (sc->sc_sdhook == NULL)
970 		aprint_error("%s: WARNING: unable to establish shutdown hook\n",
971 		    sc->sc_dev.dv_xname);
972 	return;
973 
974 	/*
975 	 * Free any resources we've allocated during the failed attach
976 	 * attempt.  Do this in reverse order and fall through.
977 	 */
978  fail_5:
979 	for (i = 0; i < DGE_NRXDESC; i++) {
980 		if (sc->sc_rxsoft[i].rxs_dmamap != NULL)
981 			bus_dmamap_destroy(sc->sc_dmat,
982 			    sc->sc_rxsoft[i].rxs_dmamap);
983 	}
984  fail_4:
985 	for (i = 0; i < DGE_TXQUEUELEN; i++) {
986 		if (sc->sc_txsoft[i].txs_dmamap != NULL)
987 			bus_dmamap_destroy(sc->sc_dmat,
988 			    sc->sc_txsoft[i].txs_dmamap);
989 	}
990 	bus_dmamap_unload(sc->sc_dmat, sc->sc_cddmamap);
991  fail_3:
992 	bus_dmamap_destroy(sc->sc_dmat, sc->sc_cddmamap);
993  fail_2:
994 	bus_dmamem_unmap(sc->sc_dmat, (caddr_t)sc->sc_control_data,
995 	    sizeof(struct dge_control_data));
996  fail_1:
997 	bus_dmamem_free(sc->sc_dmat, &seg, rseg);
998  fail_0:
999 	return;
1000 }
1001 
1002 /*
1003  * dge_shutdown:
1004  *
1005  *	Make sure the interface is stopped at reboot time.
1006  */
1007 static void
1008 dge_shutdown(void *arg)
1009 {
1010 	struct dge_softc *sc = arg;
1011 
1012 	dge_stop(&sc->sc_ethercom.ec_if, 1);
1013 }
1014 
1015 /*
1016  * dge_tx_cksum:
1017  *
1018  *	Set up TCP/IP checksumming parameters for the
1019  *	specified packet.
1020  */
1021 static int
1022 dge_tx_cksum(struct dge_softc *sc, struct dge_txsoft *txs, uint8_t *fieldsp)
1023 {
1024 	struct mbuf *m0 = txs->txs_mbuf;
1025 	struct dge_ctdes *t;
1026 	uint32_t ipcs, tucs;
1027 	struct ether_header *eh;
1028 	int offset, iphl;
1029 	uint8_t fields = 0;
1030 
1031 	/*
1032 	 * XXX It would be nice if the mbuf pkthdr had offset
1033 	 * fields for the protocol headers.
1034 	 */
1035 
1036 	eh = mtod(m0, struct ether_header *);
1037 	switch (htons(eh->ether_type)) {
1038 	case ETHERTYPE_IP:
1039 		offset = ETHER_HDR_LEN;
1040 		break;
1041 
1042 	case ETHERTYPE_VLAN:
1043 		offset = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
1044 		break;
1045 
1046 	default:
1047 		/*
1048 		 * Don't support this protocol or encapsulation.
1049 		 */
1050 		*fieldsp = 0;
1051 		return (0);
1052 	}
1053 
1054 	iphl = M_CSUM_DATA_IPv4_IPHL(m0->m_pkthdr.csum_data);
1055 
1056 	/*
1057 	 * NOTE: Even if we're not using the IP or TCP/UDP checksum
1058 	 * offload feature, if we load the context descriptor, we
1059 	 * MUST provide valid values for IPCSS and TUCSS fields.
1060 	 */
1061 
1062 	if (m0->m_pkthdr.csum_flags & M_CSUM_IPv4) {
1063 		DGE_EVCNT_INCR(&sc->sc_ev_txipsum);
1064 		fields |= TDESC_POPTS_IXSM;
1065 		ipcs = DGE_TCPIP_IPCSS(offset) |
1066 		    DGE_TCPIP_IPCSO(offset + offsetof(struct ip, ip_sum)) |
1067 		    DGE_TCPIP_IPCSE(offset + iphl - 1);
1068 	} else if (__predict_true(sc->sc_txctx_ipcs != 0xffffffff)) {
1069 		/* Use the cached value. */
1070 		ipcs = sc->sc_txctx_ipcs;
1071 	} else {
1072 		/* Just initialize it to the likely value anyway. */
1073 		ipcs = DGE_TCPIP_IPCSS(offset) |
1074 		    DGE_TCPIP_IPCSO(offset + offsetof(struct ip, ip_sum)) |
1075 		    DGE_TCPIP_IPCSE(offset + iphl - 1);
1076 	}
1077 	DPRINTF(DGE_DEBUG_CKSUM,
1078 	    ("%s: CKSUM: offset %d ipcs 0x%x\n",
1079 	    sc->sc_dev.dv_xname, offset, ipcs));
1080 
1081 	offset += iphl;
1082 
1083 	if (m0->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4)) {
1084 		DGE_EVCNT_INCR(&sc->sc_ev_txtusum);
1085 		fields |= TDESC_POPTS_TXSM;
1086 		tucs = DGE_TCPIP_TUCSS(offset) |
1087 		   DGE_TCPIP_TUCSO(offset + M_CSUM_DATA_IPv4_OFFSET(m0->m_pkthdr.csum_data)) |
1088 		   DGE_TCPIP_TUCSE(0) /* rest of packet */;
1089 	} else if (__predict_true(sc->sc_txctx_tucs != 0xffffffff)) {
1090 		/* Use the cached value. */
1091 		tucs = sc->sc_txctx_tucs;
1092 	} else {
1093 		/* Just initialize it to a valid TCP context. */
1094 		tucs = DGE_TCPIP_TUCSS(offset) |
1095 		    DGE_TCPIP_TUCSO(offset + offsetof(struct tcphdr, th_sum)) |
1096 		    DGE_TCPIP_TUCSE(0) /* rest of packet */;
1097 	}
1098 
1099 	DPRINTF(DGE_DEBUG_CKSUM,
1100 	    ("%s: CKSUM: offset %d tucs 0x%x\n",
1101 	    sc->sc_dev.dv_xname, offset, tucs));
1102 
1103 	if (sc->sc_txctx_ipcs == ipcs &&
1104 	    sc->sc_txctx_tucs == tucs) {
1105 		/* Cached context is fine. */
1106 		DGE_EVCNT_INCR(&sc->sc_ev_txctx_hit);
1107 	} else {
1108 		/* Fill in the context descriptor. */
1109 #ifdef DGE_EVENT_COUNTERS
1110 		if (sc->sc_txctx_ipcs == 0xffffffff &&
1111 		    sc->sc_txctx_tucs == 0xffffffff)
1112 			DGE_EVCNT_INCR(&sc->sc_ev_txctx_init);
1113 		else
1114 			DGE_EVCNT_INCR(&sc->sc_ev_txctx_miss);
1115 #endif
1116 		t = (struct dge_ctdes *)&sc->sc_txdescs[sc->sc_txnext];
1117 		t->dc_tcpip_ipcs = htole32(ipcs);
1118 		t->dc_tcpip_tucs = htole32(tucs);
1119 		t->dc_tcpip_cmdlen = htole32(TDESC_DTYP_CTD);
1120 		t->dc_tcpip_seg = 0;
1121 		DGE_CDTXSYNC(sc, sc->sc_txnext, 1, BUS_DMASYNC_PREWRITE);
1122 
1123 		sc->sc_txctx_ipcs = ipcs;
1124 		sc->sc_txctx_tucs = tucs;
1125 
1126 		sc->sc_txnext = DGE_NEXTTX(sc->sc_txnext);
1127 		txs->txs_ndesc++;
1128 	}
1129 
1130 	*fieldsp = fields;
1131 
1132 	return (0);
1133 }
1134 
1135 /*
1136  * dge_start:		[ifnet interface function]
1137  *
1138  *	Start packet transmission on the interface.
1139  */
1140 static void
1141 dge_start(struct ifnet *ifp)
1142 {
1143 	struct dge_softc *sc = ifp->if_softc;
1144 	struct mbuf *m0;
1145 	struct dge_txsoft *txs;
1146 	bus_dmamap_t dmamap;
1147 	int error, nexttx, lasttx = -1, ofree, seg;
1148 	uint32_t cksumcmd;
1149 	uint8_t cksumfields;
1150 
1151 	if ((ifp->if_flags & (IFF_RUNNING|IFF_OACTIVE)) != IFF_RUNNING)
1152 		return;
1153 
1154 	/*
1155 	 * Remember the previous number of free descriptors.
1156 	 */
1157 	ofree = sc->sc_txfree;
1158 
1159 	/*
1160 	 * Loop through the send queue, setting up transmit descriptors
1161 	 * until we drain the queue, or use up all available transmit
1162 	 * descriptors.
1163 	 */
1164 	for (;;) {
1165 		/* Grab a packet off the queue. */
1166 		IFQ_POLL(&ifp->if_snd, m0);
1167 		if (m0 == NULL)
1168 			break;
1169 
1170 		DPRINTF(DGE_DEBUG_TX,
1171 		    ("%s: TX: have packet to transmit: %p\n",
1172 		    sc->sc_dev.dv_xname, m0));
1173 
1174 		/* Get a work queue entry. */
1175 		if (sc->sc_txsfree < DGE_TXQUEUE_GC) {
1176 			dge_txintr(sc);
1177 			if (sc->sc_txsfree == 0) {
1178 				DPRINTF(DGE_DEBUG_TX,
1179 				    ("%s: TX: no free job descriptors\n",
1180 					sc->sc_dev.dv_xname));
1181 				DGE_EVCNT_INCR(&sc->sc_ev_txsstall);
1182 				break;
1183 			}
1184 		}
1185 
1186 		txs = &sc->sc_txsoft[sc->sc_txsnext];
1187 		dmamap = txs->txs_dmamap;
1188 
1189 		/*
1190 		 * Load the DMA map.  If this fails, the packet either
1191 		 * didn't fit in the allotted number of segments, or we
1192 		 * were short on resources.  For the too-many-segments
1193 		 * case, we simply report an error and drop the packet,
1194 		 * since we can't sanely copy a jumbo packet to a single
1195 		 * buffer.
1196 		 */
1197 		error = bus_dmamap_load_mbuf(sc->sc_dmat, dmamap, m0,
1198 		    BUS_DMA_WRITE|BUS_DMA_NOWAIT);
1199 		if (error) {
1200 			if (error == EFBIG) {
1201 				DGE_EVCNT_INCR(&sc->sc_ev_txdrop);
1202 				printf("%s: Tx packet consumes too many "
1203 				    "DMA segments, dropping...\n",
1204 				    sc->sc_dev.dv_xname);
1205 				IFQ_DEQUEUE(&ifp->if_snd, m0);
1206 				m_freem(m0);
1207 				continue;
1208 			}
1209 			/*
1210 			 * Short on resources, just stop for now.
1211 			 */
1212 			DPRINTF(DGE_DEBUG_TX,
1213 			    ("%s: TX: dmamap load failed: %d\n",
1214 			    sc->sc_dev.dv_xname, error));
1215 			break;
1216 		}
1217 
1218 		/*
1219 		 * Ensure we have enough descriptors free to describe
1220 		 * the packet.  Note, we always reserve one descriptor
1221 		 * at the end of the ring due to the semantics of the
1222 		 * TDT register, plus one more in the event we need
1223 		 * to re-load checksum offload context.
1224 		 */
1225 		if (dmamap->dm_nsegs > (sc->sc_txfree - 2)) {
1226 			/*
1227 			 * Not enough free descriptors to transmit this
1228 			 * packet.  We haven't committed anything yet,
1229 			 * so just unload the DMA map, put the packet
1230 			 * pack on the queue, and punt.  Notify the upper
1231 			 * layer that there are no more slots left.
1232 			 */
1233 			DPRINTF(DGE_DEBUG_TX,
1234 			    ("%s: TX: need %d descriptors, have %d\n",
1235 			    sc->sc_dev.dv_xname, dmamap->dm_nsegs,
1236 			    sc->sc_txfree - 1));
1237 			ifp->if_flags |= IFF_OACTIVE;
1238 			bus_dmamap_unload(sc->sc_dmat, dmamap);
1239 			DGE_EVCNT_INCR(&sc->sc_ev_txdstall);
1240 			break;
1241 		}
1242 
1243 		IFQ_DEQUEUE(&ifp->if_snd, m0);
1244 
1245 		/*
1246 		 * WE ARE NOW COMMITTED TO TRANSMITTING THE PACKET.
1247 		 */
1248 
1249 		/* Sync the DMA map. */
1250 		bus_dmamap_sync(sc->sc_dmat, dmamap, 0, dmamap->dm_mapsize,
1251 		    BUS_DMASYNC_PREWRITE);
1252 
1253 		DPRINTF(DGE_DEBUG_TX,
1254 		    ("%s: TX: packet has %d DMA segments\n",
1255 		    sc->sc_dev.dv_xname, dmamap->dm_nsegs));
1256 
1257 		DGE_EVCNT_INCR(&sc->sc_ev_txseg[dmamap->dm_nsegs - 1]);
1258 
1259 		/*
1260 		 * Store a pointer to the packet so that we can free it
1261 		 * later.
1262 		 *
1263 		 * Initially, we consider the number of descriptors the
1264 		 * packet uses the number of DMA segments.  This may be
1265 		 * incremented by 1 if we do checksum offload (a descriptor
1266 		 * is used to set the checksum context).
1267 		 */
1268 		txs->txs_mbuf = m0;
1269 		txs->txs_firstdesc = sc->sc_txnext;
1270 		txs->txs_ndesc = dmamap->dm_nsegs;
1271 
1272 		/*
1273 		 * Set up checksum offload parameters for
1274 		 * this packet.
1275 		 */
1276 		if (m0->m_pkthdr.csum_flags &
1277 		    (M_CSUM_IPv4|M_CSUM_TCPv4|M_CSUM_UDPv4)) {
1278 			if (dge_tx_cksum(sc, txs, &cksumfields) != 0) {
1279 				/* Error message already displayed. */
1280 				bus_dmamap_unload(sc->sc_dmat, dmamap);
1281 				continue;
1282 			}
1283 		} else {
1284 			cksumfields = 0;
1285 		}
1286 
1287 		cksumcmd = TDESC_DCMD_IDE | TDESC_DTYP_DATA;
1288 
1289 		/*
1290 		 * Initialize the transmit descriptor.
1291 		 */
1292 		for (nexttx = sc->sc_txnext, seg = 0;
1293 		     seg < dmamap->dm_nsegs;
1294 		     seg++, nexttx = DGE_NEXTTX(nexttx)) {
1295 			/*
1296 			 * Note: we currently only use 32-bit DMA
1297 			 * addresses.
1298 			 */
1299 			sc->sc_txdescs[nexttx].dt_baddrh = 0;
1300 			sc->sc_txdescs[nexttx].dt_baddrl =
1301 			    htole32(dmamap->dm_segs[seg].ds_addr);
1302 			sc->sc_txdescs[nexttx].dt_ctl =
1303 			    htole32(cksumcmd | dmamap->dm_segs[seg].ds_len);
1304 			sc->sc_txdescs[nexttx].dt_status = 0;
1305 			sc->sc_txdescs[nexttx].dt_popts = cksumfields;
1306 			sc->sc_txdescs[nexttx].dt_vlan = 0;
1307 			lasttx = nexttx;
1308 
1309 			DPRINTF(DGE_DEBUG_TX,
1310 			    ("%s: TX: desc %d: low 0x%08lx, len 0x%04lx\n",
1311 			    sc->sc_dev.dv_xname, nexttx,
1312 			    le32toh(dmamap->dm_segs[seg].ds_addr),
1313 			    le32toh(dmamap->dm_segs[seg].ds_len)));
1314 		}
1315 
1316 		KASSERT(lasttx != -1);
1317 
1318 		/*
1319 		 * Set up the command byte on the last descriptor of
1320 		 * the packet.  If we're in the interrupt delay window,
1321 		 * delay the interrupt.
1322 		 */
1323 		sc->sc_txdescs[lasttx].dt_ctl |=
1324 		    htole32(TDESC_DCMD_EOP | TDESC_DCMD_RS);
1325 
1326 		txs->txs_lastdesc = lasttx;
1327 
1328 		DPRINTF(DGE_DEBUG_TX,
1329 		    ("%s: TX: desc %d: cmdlen 0x%08x\n", sc->sc_dev.dv_xname,
1330 		    lasttx, le32toh(sc->sc_txdescs[lasttx].dt_ctl)));
1331 
1332 		/* Sync the descriptors we're using. */
1333 		DGE_CDTXSYNC(sc, sc->sc_txnext, dmamap->dm_nsegs,
1334 		    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1335 
1336 		/* Give the packet to the chip. */
1337 		CSR_WRITE(sc, DGE_TDT, nexttx);
1338 
1339 		DPRINTF(DGE_DEBUG_TX,
1340 		    ("%s: TX: TDT -> %d\n", sc->sc_dev.dv_xname, nexttx));
1341 
1342 		DPRINTF(DGE_DEBUG_TX,
1343 		    ("%s: TX: finished transmitting packet, job %d\n",
1344 		    sc->sc_dev.dv_xname, sc->sc_txsnext));
1345 
1346 		/* Advance the tx pointer. */
1347 		sc->sc_txfree -= txs->txs_ndesc;
1348 		sc->sc_txnext = nexttx;
1349 
1350 		sc->sc_txsfree--;
1351 		sc->sc_txsnext = DGE_NEXTTXS(sc->sc_txsnext);
1352 
1353 #if NBPFILTER > 0
1354 		/* Pass the packet to any BPF listeners. */
1355 		if (ifp->if_bpf)
1356 			bpf_mtap(ifp->if_bpf, m0);
1357 #endif /* NBPFILTER > 0 */
1358 	}
1359 
1360 	if (sc->sc_txsfree == 0 || sc->sc_txfree <= 2) {
1361 		/* No more slots; notify upper layer. */
1362 		ifp->if_flags |= IFF_OACTIVE;
1363 	}
1364 
1365 	if (sc->sc_txfree != ofree) {
1366 		/* Set a watchdog timer in case the chip flakes out. */
1367 		ifp->if_timer = 5;
1368 	}
1369 }
1370 
1371 /*
1372  * dge_watchdog:		[ifnet interface function]
1373  *
1374  *	Watchdog timer handler.
1375  */
1376 static void
1377 dge_watchdog(struct ifnet *ifp)
1378 {
1379 	struct dge_softc *sc = ifp->if_softc;
1380 
1381 	/*
1382 	 * Since we're using delayed interrupts, sweep up
1383 	 * before we report an error.
1384 	 */
1385 	dge_txintr(sc);
1386 
1387 	if (sc->sc_txfree != DGE_NTXDESC) {
1388 		printf("%s: device timeout (txfree %d txsfree %d txnext %d)\n",
1389 		    sc->sc_dev.dv_xname, sc->sc_txfree, sc->sc_txsfree,
1390 		    sc->sc_txnext);
1391 		ifp->if_oerrors++;
1392 
1393 		/* Reset the interface. */
1394 		(void) dge_init(ifp);
1395 	}
1396 
1397 	/* Try to get more packets going. */
1398 	dge_start(ifp);
1399 }
1400 
1401 /*
1402  * dge_ioctl:		[ifnet interface function]
1403  *
1404  *	Handle control requests from the operator.
1405  */
1406 static int
1407 dge_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1408 {
1409 	struct dge_softc *sc = ifp->if_softc;
1410 	struct ifreq *ifr = (struct ifreq *) data;
1411 	pcireg_t preg;
1412 	int s, error, mmrbc;
1413 
1414 	s = splnet();
1415 
1416 	switch (cmd) {
1417 	case SIOCSIFMEDIA:
1418 	case SIOCGIFMEDIA:
1419 		error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
1420 		break;
1421 
1422 	case SIOCSIFMTU:
1423 		if (ifr->ifr_mtu < ETHERMIN || ifr->ifr_mtu > DGE_MAX_MTU) {
1424 			error = EINVAL;
1425 		} else {
1426 			error = 0;
1427 			ifp->if_mtu = ifr->ifr_mtu;
1428 			if (ifp->if_flags & IFF_UP)
1429 				error = (*ifp->if_init)(ifp);
1430 		}
1431 		break;
1432 
1433         case SIOCSIFFLAGS:
1434 		/* extract link flags */
1435 		if ((ifp->if_flags & IFF_LINK0) == 0 &&
1436 		    (ifp->if_flags & IFF_LINK1) == 0)
1437 			mmrbc = PCIX_MMRBC_512;
1438 		else if ((ifp->if_flags & IFF_LINK0) == 0 &&
1439 		    (ifp->if_flags & IFF_LINK1) != 0)
1440 			mmrbc = PCIX_MMRBC_1024;
1441 		else if ((ifp->if_flags & IFF_LINK0) != 0 &&
1442 		    (ifp->if_flags & IFF_LINK1) == 0)
1443 			mmrbc = PCIX_MMRBC_2048;
1444 		else
1445 			mmrbc = PCIX_MMRBC_4096;
1446 		if (mmrbc != sc->sc_mmrbc) {
1447 			preg = pci_conf_read(sc->sc_pc, sc->sc_pt,DGE_PCIX_CMD);
1448 			preg &= ~PCIX_MMRBC_MSK;
1449 			preg |= mmrbc;
1450 			pci_conf_write(sc->sc_pc, sc->sc_pt,DGE_PCIX_CMD, preg);
1451 			sc->sc_mmrbc = mmrbc;
1452 		}
1453                 /* FALLTHROUGH */
1454 	default:
1455 		error = ether_ioctl(ifp, cmd, data);
1456 		if (error == ENETRESET) {
1457 			/*
1458 			 * Multicast list has changed; set the hardware filter
1459 			 * accordingly.
1460 			 */
1461 			if (ifp->if_flags & IFF_RUNNING)
1462 				dge_set_filter(sc);
1463 			error = 0;
1464 		}
1465 		break;
1466 	}
1467 
1468 	/* Try to get more packets going. */
1469 	dge_start(ifp);
1470 
1471 	splx(s);
1472 	return (error);
1473 }
1474 
1475 /*
1476  * dge_intr:
1477  *
1478  *	Interrupt service routine.
1479  */
1480 static int
1481 dge_intr(void *arg)
1482 {
1483 	struct dge_softc *sc = arg;
1484 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
1485 	uint32_t icr;
1486 	int wantinit, handled = 0;
1487 
1488 	for (wantinit = 0; wantinit == 0;) {
1489 		icr = CSR_READ(sc, DGE_ICR);
1490 		if ((icr & sc->sc_icr) == 0)
1491 			break;
1492 
1493 #if 0 /*NRND > 0*/
1494 		if (RND_ENABLED(&sc->rnd_source))
1495 			rnd_add_uint32(&sc->rnd_source, icr);
1496 #endif
1497 
1498 		handled = 1;
1499 
1500 #if defined(DGE_DEBUG) || defined(DGE_EVENT_COUNTERS)
1501 		if (icr & (ICR_RXDMT0|ICR_RXT0)) {
1502 			DPRINTF(DGE_DEBUG_RX,
1503 			    ("%s: RX: got Rx intr 0x%08x\n",
1504 			    sc->sc_dev.dv_xname,
1505 			    icr & (ICR_RXDMT0|ICR_RXT0)));
1506 			DGE_EVCNT_INCR(&sc->sc_ev_rxintr);
1507 		}
1508 #endif
1509 		dge_rxintr(sc);
1510 
1511 #if defined(DGE_DEBUG) || defined(DGE_EVENT_COUNTERS)
1512 		if (icr & ICR_TXDW) {
1513 			DPRINTF(DGE_DEBUG_TX,
1514 			    ("%s: TX: got TXDW interrupt\n",
1515 			    sc->sc_dev.dv_xname));
1516 			DGE_EVCNT_INCR(&sc->sc_ev_txdw);
1517 		}
1518 		if (icr & ICR_TXQE)
1519 			DGE_EVCNT_INCR(&sc->sc_ev_txqe);
1520 #endif
1521 		dge_txintr(sc);
1522 
1523 		if (icr & (ICR_LSC|ICR_RXSEQ)) {
1524 			DGE_EVCNT_INCR(&sc->sc_ev_linkintr);
1525 			dge_linkintr(sc, icr);
1526 		}
1527 
1528 		if (icr & ICR_RXO) {
1529 			printf("%s: Receive overrun\n", sc->sc_dev.dv_xname);
1530 			wantinit = 1;
1531 		}
1532 	}
1533 
1534 	if (handled) {
1535 		if (wantinit)
1536 			dge_init(ifp);
1537 
1538 		/* Try to get more packets going. */
1539 		dge_start(ifp);
1540 	}
1541 
1542 	return (handled);
1543 }
1544 
1545 /*
1546  * dge_txintr:
1547  *
1548  *	Helper; handle transmit interrupts.
1549  */
1550 static void
1551 dge_txintr(struct dge_softc *sc)
1552 {
1553 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
1554 	struct dge_txsoft *txs;
1555 	uint8_t status;
1556 	int i;
1557 
1558 	ifp->if_flags &= ~IFF_OACTIVE;
1559 
1560 	/*
1561 	 * Go through the Tx list and free mbufs for those
1562 	 * frames which have been transmitted.
1563 	 */
1564 	for (i = sc->sc_txsdirty; sc->sc_txsfree != DGE_TXQUEUELEN;
1565 	     i = DGE_NEXTTXS(i), sc->sc_txsfree++) {
1566 		txs = &sc->sc_txsoft[i];
1567 
1568 		DPRINTF(DGE_DEBUG_TX,
1569 		    ("%s: TX: checking job %d\n", sc->sc_dev.dv_xname, i));
1570 
1571 		DGE_CDTXSYNC(sc, txs->txs_firstdesc, txs->txs_dmamap->dm_nsegs,
1572 		    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1573 
1574 		status =
1575 		    sc->sc_txdescs[txs->txs_lastdesc].dt_status;
1576 		if ((status & TDESC_STA_DD) == 0) {
1577 			DGE_CDTXSYNC(sc, txs->txs_lastdesc, 1,
1578 			    BUS_DMASYNC_PREREAD);
1579 			break;
1580 		}
1581 
1582 		DPRINTF(DGE_DEBUG_TX,
1583 		    ("%s: TX: job %d done: descs %d..%d\n",
1584 		    sc->sc_dev.dv_xname, i, txs->txs_firstdesc,
1585 		    txs->txs_lastdesc));
1586 
1587 		ifp->if_opackets++;
1588 		sc->sc_txfree += txs->txs_ndesc;
1589 		bus_dmamap_sync(sc->sc_dmat, txs->txs_dmamap,
1590 		    0, txs->txs_dmamap->dm_mapsize, BUS_DMASYNC_POSTWRITE);
1591 		bus_dmamap_unload(sc->sc_dmat, txs->txs_dmamap);
1592 		m_freem(txs->txs_mbuf);
1593 		txs->txs_mbuf = NULL;
1594 	}
1595 
1596 	/* Update the dirty transmit buffer pointer. */
1597 	sc->sc_txsdirty = i;
1598 	DPRINTF(DGE_DEBUG_TX,
1599 	    ("%s: TX: txsdirty -> %d\n", sc->sc_dev.dv_xname, i));
1600 
1601 	/*
1602 	 * If there are no more pending transmissions, cancel the watchdog
1603 	 * timer.
1604 	 */
1605 	if (sc->sc_txsfree == DGE_TXQUEUELEN)
1606 		ifp->if_timer = 0;
1607 }
1608 
1609 /*
1610  * dge_rxintr:
1611  *
1612  *	Helper; handle receive interrupts.
1613  */
1614 static void
1615 dge_rxintr(struct dge_softc *sc)
1616 {
1617 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
1618 	struct dge_rxsoft *rxs;
1619 	struct mbuf *m;
1620 	int i, len;
1621 	uint8_t status, errors;
1622 
1623 	for (i = sc->sc_rxptr;; i = DGE_NEXTRX(i)) {
1624 		rxs = &sc->sc_rxsoft[i];
1625 
1626 		DPRINTF(DGE_DEBUG_RX,
1627 		    ("%s: RX: checking descriptor %d\n",
1628 		    sc->sc_dev.dv_xname, i));
1629 
1630 		DGE_CDRXSYNC(sc, i, BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1631 
1632 		status = sc->sc_rxdescs[i].dr_status;
1633 		errors = sc->sc_rxdescs[i].dr_errors;
1634 		len = le16toh(sc->sc_rxdescs[i].dr_len);
1635 
1636 		if ((status & RDESC_STS_DD) == 0) {
1637 			/*
1638 			 * We have processed all of the receive descriptors.
1639 			 */
1640 			DGE_CDRXSYNC(sc, i, BUS_DMASYNC_PREREAD);
1641 			break;
1642 		}
1643 
1644 		if (__predict_false(sc->sc_rxdiscard)) {
1645 			DPRINTF(DGE_DEBUG_RX,
1646 			    ("%s: RX: discarding contents of descriptor %d\n",
1647 			    sc->sc_dev.dv_xname, i));
1648 			DGE_INIT_RXDESC(sc, i);
1649 			if (status & RDESC_STS_EOP) {
1650 				/* Reset our state. */
1651 				DPRINTF(DGE_DEBUG_RX,
1652 				    ("%s: RX: resetting rxdiscard -> 0\n",
1653 				    sc->sc_dev.dv_xname));
1654 				sc->sc_rxdiscard = 0;
1655 			}
1656 			continue;
1657 		}
1658 
1659 		bus_dmamap_sync(sc->sc_dmat, rxs->rxs_dmamap, 0,
1660 		    rxs->rxs_dmamap->dm_mapsize, BUS_DMASYNC_POSTREAD);
1661 
1662 		m = rxs->rxs_mbuf;
1663 
1664 		/*
1665 		 * Add a new receive buffer to the ring.
1666 		 */
1667 		if (dge_add_rxbuf(sc, i) != 0) {
1668 			/*
1669 			 * Failed, throw away what we've done so
1670 			 * far, and discard the rest of the packet.
1671 			 */
1672 			ifp->if_ierrors++;
1673 			bus_dmamap_sync(sc->sc_dmat, rxs->rxs_dmamap, 0,
1674 			    rxs->rxs_dmamap->dm_mapsize, BUS_DMASYNC_PREREAD);
1675 			DGE_INIT_RXDESC(sc, i);
1676 			if ((status & RDESC_STS_EOP) == 0)
1677 				sc->sc_rxdiscard = 1;
1678 			if (sc->sc_rxhead != NULL)
1679 				m_freem(sc->sc_rxhead);
1680 			DGE_RXCHAIN_RESET(sc);
1681 			DPRINTF(DGE_DEBUG_RX,
1682 			    ("%s: RX: Rx buffer allocation failed, "
1683 			    "dropping packet%s\n", sc->sc_dev.dv_xname,
1684 			    sc->sc_rxdiscard ? " (discard)" : ""));
1685 			continue;
1686 		}
1687 		DGE_INIT_RXDESC(sc, DGE_PREVRX(i)); /* Write the descriptor */
1688 
1689 		DGE_RXCHAIN_LINK(sc, m);
1690 
1691 		m->m_len = len;
1692 
1693 		DPRINTF(DGE_DEBUG_RX,
1694 		    ("%s: RX: buffer at %p len %d\n",
1695 		    sc->sc_dev.dv_xname, m->m_data, len));
1696 
1697 		/*
1698 		 * If this is not the end of the packet, keep
1699 		 * looking.
1700 		 */
1701 		if ((status & RDESC_STS_EOP) == 0) {
1702 			sc->sc_rxlen += len;
1703 			DPRINTF(DGE_DEBUG_RX,
1704 			    ("%s: RX: not yet EOP, rxlen -> %d\n",
1705 			    sc->sc_dev.dv_xname, sc->sc_rxlen));
1706 			continue;
1707 		}
1708 
1709 		/*
1710 		 * Okay, we have the entire packet now...
1711 		 */
1712 		*sc->sc_rxtailp = NULL;
1713 		m = sc->sc_rxhead;
1714 		len += sc->sc_rxlen;
1715 
1716 		DGE_RXCHAIN_RESET(sc);
1717 
1718 		DPRINTF(DGE_DEBUG_RX,
1719 		    ("%s: RX: have entire packet, len -> %d\n",
1720 		    sc->sc_dev.dv_xname, len));
1721 
1722 		/*
1723 		 * If an error occurred, update stats and drop the packet.
1724 		 */
1725 		if (errors &
1726 		     (RDESC_ERR_CE|RDESC_ERR_SE|RDESC_ERR_P|RDESC_ERR_RXE)) {
1727 			ifp->if_ierrors++;
1728 			if (errors & RDESC_ERR_SE)
1729 				printf("%s: symbol error\n",
1730 				    sc->sc_dev.dv_xname);
1731 			else if (errors & RDESC_ERR_P)
1732 				printf("%s: parity error\n",
1733 				    sc->sc_dev.dv_xname);
1734 			else if (errors & RDESC_ERR_CE)
1735 				printf("%s: CRC error\n",
1736 				    sc->sc_dev.dv_xname);
1737 			m_freem(m);
1738 			continue;
1739 		}
1740 
1741 		/*
1742 		 * No errors.  Receive the packet.
1743 		 */
1744 		m->m_pkthdr.rcvif = ifp;
1745 		m->m_pkthdr.len = len;
1746 
1747 		/*
1748 		 * Set up checksum info for this packet.
1749 		 */
1750 		if (status & RDESC_STS_IPCS) {
1751 			DGE_EVCNT_INCR(&sc->sc_ev_rxipsum);
1752 			m->m_pkthdr.csum_flags |= M_CSUM_IPv4;
1753 			if (errors & RDESC_ERR_IPE)
1754 				m->m_pkthdr.csum_flags |= M_CSUM_IPv4_BAD;
1755 		}
1756 		if (status & RDESC_STS_TCPCS) {
1757 			/*
1758 			 * Note: we don't know if this was TCP or UDP,
1759 			 * so we just set both bits, and expect the
1760 			 * upper layers to deal.
1761 			 */
1762 			DGE_EVCNT_INCR(&sc->sc_ev_rxtusum);
1763 			m->m_pkthdr.csum_flags |= M_CSUM_TCPv4|M_CSUM_UDPv4;
1764 			if (errors & RDESC_ERR_TCPE)
1765 				m->m_pkthdr.csum_flags |= M_CSUM_TCP_UDP_BAD;
1766 		}
1767 
1768 		ifp->if_ipackets++;
1769 
1770 #if NBPFILTER > 0
1771 		/* Pass this up to any BPF listeners. */
1772 		if (ifp->if_bpf)
1773 			bpf_mtap(ifp->if_bpf, m);
1774 #endif /* NBPFILTER > 0 */
1775 
1776 		/* Pass it on. */
1777 		(*ifp->if_input)(ifp, m);
1778 	}
1779 
1780 	/* Update the receive pointer. */
1781 	sc->sc_rxptr = i;
1782 
1783 	DPRINTF(DGE_DEBUG_RX,
1784 	    ("%s: RX: rxptr -> %d\n", sc->sc_dev.dv_xname, i));
1785 }
1786 
1787 /*
1788  * dge_linkintr:
1789  *
1790  *	Helper; handle link interrupts.
1791  */
1792 static void
1793 dge_linkintr(struct dge_softc *sc, uint32_t icr)
1794 {
1795 	uint32_t status;
1796 
1797 	if (icr & ICR_LSC) {
1798 		status = CSR_READ(sc, DGE_STATUS);
1799 		if (status & STATUS_LINKUP) {
1800 			DPRINTF(DGE_DEBUG_LINK, ("%s: LINK: LSC -> up\n",
1801 			    sc->sc_dev.dv_xname));
1802 		} else {
1803 			DPRINTF(DGE_DEBUG_LINK, ("%s: LINK: LSC -> down\n",
1804 			    sc->sc_dev.dv_xname));
1805 		}
1806 	} else if (icr & ICR_RXSEQ) {
1807 		DPRINTF(DGE_DEBUG_LINK,
1808 		    ("%s: LINK: Receive sequence error\n",
1809 		    sc->sc_dev.dv_xname));
1810 	}
1811 	/* XXX - fix errata */
1812 }
1813 
1814 /*
1815  * dge_reset:
1816  *
1817  *	Reset the i82597 chip.
1818  */
1819 static void
1820 dge_reset(struct dge_softc *sc)
1821 {
1822 	int i;
1823 
1824 	/*
1825 	 * Do a chip reset.
1826 	 */
1827 	CSR_WRITE(sc, DGE_CTRL0, CTRL0_RST | sc->sc_ctrl0);
1828 
1829 	delay(10000);
1830 
1831 	for (i = 0; i < 1000; i++) {
1832 		if ((CSR_READ(sc, DGE_CTRL0) & CTRL0_RST) == 0)
1833 			break;
1834 		delay(20);
1835 	}
1836 
1837 	if (CSR_READ(sc, DGE_CTRL0) & CTRL0_RST)
1838 		printf("%s: WARNING: reset failed to complete\n",
1839 		    sc->sc_dev.dv_xname);
1840         /*
1841          * Reset the EEPROM logic.
1842          * This will cause the chip to reread its default values,
1843 	 * which doesn't happen otherwise (errata).
1844          */
1845         CSR_WRITE(sc, DGE_CTRL1, CTRL1_EE_RST);
1846         delay(10000);
1847 }
1848 
1849 /*
1850  * dge_init:		[ifnet interface function]
1851  *
1852  *	Initialize the interface.  Must be called at splnet().
1853  */
1854 static int
1855 dge_init(struct ifnet *ifp)
1856 {
1857 	struct dge_softc *sc = ifp->if_softc;
1858 	struct dge_rxsoft *rxs;
1859 	int i, error = 0;
1860 	uint32_t reg;
1861 
1862 	/*
1863 	 * *_HDR_ALIGNED_P is constant 1 if __NO_STRICT_ALIGMENT is set.
1864 	 * There is a small but measurable benefit to avoiding the adjusment
1865 	 * of the descriptor so that the headers are aligned, for normal mtu,
1866 	 * on such platforms.  One possibility is that the DMA itself is
1867 	 * slightly more efficient if the front of the entire packet (instead
1868 	 * of the front of the headers) is aligned.
1869 	 *
1870 	 * Note we must always set align_tweak to 0 if we are using
1871 	 * jumbo frames.
1872 	 */
1873 #ifdef __NO_STRICT_ALIGNMENT
1874 	sc->sc_align_tweak = 0;
1875 #else
1876 	if ((ifp->if_mtu + ETHER_HDR_LEN + ETHER_CRC_LEN) > (MCLBYTES - 2))
1877 		sc->sc_align_tweak = 0;
1878 	else
1879 		sc->sc_align_tweak = 2;
1880 #endif /* __NO_STRICT_ALIGNMENT */
1881 
1882 	/* Cancel any pending I/O. */
1883 	dge_stop(ifp, 0);
1884 
1885 	/* Reset the chip to a known state. */
1886 	dge_reset(sc);
1887 
1888 	/* Initialize the transmit descriptor ring. */
1889 	memset(sc->sc_txdescs, 0, sizeof(sc->sc_txdescs));
1890 	DGE_CDTXSYNC(sc, 0, DGE_NTXDESC,
1891 	    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1892 	sc->sc_txfree = DGE_NTXDESC;
1893 	sc->sc_txnext = 0;
1894 
1895 	sc->sc_txctx_ipcs = 0xffffffff;
1896 	sc->sc_txctx_tucs = 0xffffffff;
1897 
1898 	CSR_WRITE(sc, DGE_TDBAH, 0);
1899 	CSR_WRITE(sc, DGE_TDBAL, DGE_CDTXADDR(sc, 0));
1900 	CSR_WRITE(sc, DGE_TDLEN, sizeof(sc->sc_txdescs));
1901 	CSR_WRITE(sc, DGE_TDH, 0);
1902 	CSR_WRITE(sc, DGE_TDT, 0);
1903 	CSR_WRITE(sc, DGE_TIDV, TIDV);
1904 
1905 #if 0
1906 	CSR_WRITE(sc, DGE_TXDCTL, TXDCTL_PTHRESH(0) |
1907 	    TXDCTL_HTHRESH(0) | TXDCTL_WTHRESH(0));
1908 #endif
1909 	CSR_WRITE(sc, DGE_RXDCTL,
1910 	    RXDCTL_PTHRESH(RXDCTL_PTHRESH_VAL) |
1911 	    RXDCTL_HTHRESH(RXDCTL_HTHRESH_VAL) |
1912 	    RXDCTL_WTHRESH(RXDCTL_WTHRESH_VAL));
1913 
1914 	/* Initialize the transmit job descriptors. */
1915 	for (i = 0; i < DGE_TXQUEUELEN; i++)
1916 		sc->sc_txsoft[i].txs_mbuf = NULL;
1917 	sc->sc_txsfree = DGE_TXQUEUELEN;
1918 	sc->sc_txsnext = 0;
1919 	sc->sc_txsdirty = 0;
1920 
1921 	/*
1922 	 * Initialize the receive descriptor and receive job
1923 	 * descriptor rings.
1924 	 */
1925 	CSR_WRITE(sc, DGE_RDBAH, 0);
1926 	CSR_WRITE(sc, DGE_RDBAL, DGE_CDRXADDR(sc, 0));
1927 	CSR_WRITE(sc, DGE_RDLEN, sizeof(sc->sc_rxdescs));
1928 	CSR_WRITE(sc, DGE_RDH, DGE_RXSPACE);
1929 	CSR_WRITE(sc, DGE_RDT, 0);
1930 	CSR_WRITE(sc, DGE_RDTR, RDTR | 0x80000000);
1931 	CSR_WRITE(sc, DGE_FCRTL, FCRTL | FCRTL_XONE);
1932 	CSR_WRITE(sc, DGE_FCRTH, FCRTH);
1933 
1934 	for (i = 0; i < DGE_NRXDESC; i++) {
1935 		rxs = &sc->sc_rxsoft[i];
1936 		if (rxs->rxs_mbuf == NULL) {
1937 			if ((error = dge_add_rxbuf(sc, i)) != 0) {
1938 				printf("%s: unable to allocate or map rx "
1939 				    "buffer %d, error = %d\n",
1940 				    sc->sc_dev.dv_xname, i, error);
1941 				/*
1942 				 * XXX Should attempt to run with fewer receive
1943 				 * XXX buffers instead of just failing.
1944 				 */
1945 				dge_rxdrain(sc);
1946 				goto out;
1947 			}
1948 		}
1949 		DGE_INIT_RXDESC(sc, i);
1950 	}
1951 	sc->sc_rxptr = DGE_RXSPACE;
1952 	sc->sc_rxdiscard = 0;
1953 	DGE_RXCHAIN_RESET(sc);
1954 
1955 	if (sc->sc_ethercom.ec_capabilities & ETHERCAP_JUMBO_MTU) {
1956 		sc->sc_ctrl0 |= CTRL0_JFE;
1957 		CSR_WRITE(sc, DGE_MFS, ETHER_MAX_LEN_JUMBO << 16);
1958 	}
1959 
1960 	/* Write the control registers. */
1961 	CSR_WRITE(sc, DGE_CTRL0, sc->sc_ctrl0);
1962 
1963 	/*
1964 	 * Set up checksum offload parameters.
1965 	 */
1966 	reg = CSR_READ(sc, DGE_RXCSUM);
1967 	if (ifp->if_capenable & IFCAP_CSUM_IPv4)
1968 		reg |= RXCSUM_IPOFL;
1969 	else
1970 		reg &= ~RXCSUM_IPOFL;
1971 	if (ifp->if_capenable & (IFCAP_CSUM_TCPv4 | IFCAP_CSUM_UDPv4))
1972 		reg |= RXCSUM_IPOFL | RXCSUM_TUOFL;
1973 	else {
1974 		reg &= ~RXCSUM_TUOFL;
1975 		if ((ifp->if_capenable & IFCAP_CSUM_IPv4) == 0)
1976 			reg &= ~RXCSUM_IPOFL;
1977 	}
1978 	CSR_WRITE(sc, DGE_RXCSUM, reg);
1979 
1980 	/*
1981 	 * Set up the interrupt registers.
1982 	 */
1983 	CSR_WRITE(sc, DGE_IMC, 0xffffffffU);
1984 	sc->sc_icr = ICR_TXDW | ICR_LSC | ICR_RXSEQ | ICR_RXDMT0 |
1985 	    ICR_RXO | ICR_RXT0;
1986 
1987 	CSR_WRITE(sc, DGE_IMS, sc->sc_icr);
1988 
1989 	/*
1990 	 * Set up the transmit control register.
1991 	 */
1992 	sc->sc_tctl = TCTL_TCE|TCTL_TPDE|TCTL_TXEN;
1993 	CSR_WRITE(sc, DGE_TCTL, sc->sc_tctl);
1994 
1995 	/*
1996 	 * Set up the receive control register; we actually program
1997 	 * the register when we set the receive filter.  Use multicast
1998 	 * address offset type 0.
1999 	 */
2000 	sc->sc_mchash_type = 0;
2001 
2002 	sc->sc_rctl = RCTL_RXEN | RCTL_RDMTS_12 | RCTL_RPDA_MC |
2003 	    RCTL_CFF | RCTL_SECRC | RCTL_MO(sc->sc_mchash_type);
2004 
2005 #ifdef DGE_OFFBYONE_RXBUG
2006 	sc->sc_rctl |= RCTL_BSIZE_16k;
2007 #else
2008 	switch(MCLBYTES) {
2009 	case 2048:
2010 		sc->sc_rctl |= RCTL_BSIZE_2k;
2011 		break;
2012 	case 4096:
2013 		sc->sc_rctl |= RCTL_BSIZE_4k;
2014 		break;
2015 	case 8192:
2016 		sc->sc_rctl |= RCTL_BSIZE_8k;
2017 		break;
2018 	case 16384:
2019 		sc->sc_rctl |= RCTL_BSIZE_16k;
2020 		break;
2021 	default:
2022 		panic("dge_init: MCLBYTES %d unsupported", MCLBYTES);
2023 	}
2024 #endif
2025 
2026 	/* Set the receive filter. */
2027 	/* Also sets RCTL */
2028 	dge_set_filter(sc);
2029 
2030 	/* ...all done! */
2031 	ifp->if_flags |= IFF_RUNNING;
2032 	ifp->if_flags &= ~IFF_OACTIVE;
2033 
2034  out:
2035 	if (error)
2036 		printf("%s: interface not running\n", sc->sc_dev.dv_xname);
2037 	return (error);
2038 }
2039 
2040 /*
2041  * dge_rxdrain:
2042  *
2043  *	Drain the receive queue.
2044  */
2045 static void
2046 dge_rxdrain(struct dge_softc *sc)
2047 {
2048 	struct dge_rxsoft *rxs;
2049 	int i;
2050 
2051 	for (i = 0; i < DGE_NRXDESC; i++) {
2052 		rxs = &sc->sc_rxsoft[i];
2053 		if (rxs->rxs_mbuf != NULL) {
2054 			bus_dmamap_unload(sc->sc_dmat, rxs->rxs_dmamap);
2055 			m_freem(rxs->rxs_mbuf);
2056 			rxs->rxs_mbuf = NULL;
2057 		}
2058 	}
2059 }
2060 
2061 /*
2062  * dge_stop:		[ifnet interface function]
2063  *
2064  *	Stop transmission on the interface.
2065  */
2066 static void
2067 dge_stop(struct ifnet *ifp, int disable)
2068 {
2069 	struct dge_softc *sc = ifp->if_softc;
2070 	struct dge_txsoft *txs;
2071 	int i;
2072 
2073 	/* Stop the transmit and receive processes. */
2074 	CSR_WRITE(sc, DGE_TCTL, 0);
2075 	CSR_WRITE(sc, DGE_RCTL, 0);
2076 
2077 	/* Release any queued transmit buffers. */
2078 	for (i = 0; i < DGE_TXQUEUELEN; i++) {
2079 		txs = &sc->sc_txsoft[i];
2080 		if (txs->txs_mbuf != NULL) {
2081 			bus_dmamap_unload(sc->sc_dmat, txs->txs_dmamap);
2082 			m_freem(txs->txs_mbuf);
2083 			txs->txs_mbuf = NULL;
2084 		}
2085 	}
2086 
2087 	if (disable)
2088 		dge_rxdrain(sc);
2089 
2090 	/* Mark the interface as down and cancel the watchdog timer. */
2091 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
2092 	ifp->if_timer = 0;
2093 }
2094 
2095 /*
2096  * dge_add_rxbuf:
2097  *
2098  *	Add a receive buffer to the indiciated descriptor.
2099  */
2100 static int
2101 dge_add_rxbuf(struct dge_softc *sc, int idx)
2102 {
2103 	struct dge_rxsoft *rxs = &sc->sc_rxsoft[idx];
2104 	struct mbuf *m;
2105 	int error;
2106 #ifdef DGE_OFFBYONE_RXBUG
2107 	caddr_t buf;
2108 #endif
2109 
2110 	MGETHDR(m, M_DONTWAIT, MT_DATA);
2111 	if (m == NULL)
2112 		return (ENOBUFS);
2113 
2114 #ifdef DGE_OFFBYONE_RXBUG
2115 	if ((buf = dge_getbuf(sc)) == NULL)
2116 		return ENOBUFS;
2117 
2118 	m->m_len = m->m_pkthdr.len = DGE_BUFFER_SIZE;
2119 	MEXTADD(m, buf, DGE_BUFFER_SIZE, M_DEVBUF, dge_freebuf, sc);
2120 	m->m_flags |= M_EXT_RW;
2121 
2122 	if (rxs->rxs_mbuf != NULL)
2123 		bus_dmamap_unload(sc->sc_dmat, rxs->rxs_dmamap);
2124 	rxs->rxs_mbuf = m;
2125 
2126 	error = bus_dmamap_load(sc->sc_dmat, rxs->rxs_dmamap, buf,
2127 	    DGE_BUFFER_SIZE, NULL, BUS_DMA_READ|BUS_DMA_NOWAIT);
2128 #else
2129 	MCLGET(m, M_DONTWAIT);
2130 	if ((m->m_flags & M_EXT) == 0) {
2131 		m_freem(m);
2132 		return (ENOBUFS);
2133 	}
2134 
2135 	if (rxs->rxs_mbuf != NULL)
2136 		bus_dmamap_unload(sc->sc_dmat, rxs->rxs_dmamap);
2137 
2138 	rxs->rxs_mbuf = m;
2139 
2140 	m->m_len = m->m_pkthdr.len = m->m_ext.ext_size;
2141 	error = bus_dmamap_load_mbuf(sc->sc_dmat, rxs->rxs_dmamap, m,
2142 	    BUS_DMA_READ|BUS_DMA_NOWAIT);
2143 #endif
2144 	if (error) {
2145 		printf("%s: unable to load rx DMA map %d, error = %d\n",
2146 		    sc->sc_dev.dv_xname, idx, error);
2147 		panic("dge_add_rxbuf");	/* XXX XXX XXX */
2148 	}
2149 	bus_dmamap_sync(sc->sc_dmat, rxs->rxs_dmamap, 0,
2150 	    rxs->rxs_dmamap->dm_mapsize, BUS_DMASYNC_PREREAD);
2151 
2152 	return (0);
2153 }
2154 
2155 /*
2156  * dge_set_ral:
2157  *
2158  *	Set an entry in the receive address list.
2159  */
2160 static void
2161 dge_set_ral(struct dge_softc *sc, const uint8_t *enaddr, int idx)
2162 {
2163 	uint32_t ral_lo, ral_hi;
2164 
2165 	if (enaddr != NULL) {
2166 		ral_lo = enaddr[0] | (enaddr[1] << 8) | (enaddr[2] << 16) |
2167 		    (enaddr[3] << 24);
2168 		ral_hi = enaddr[4] | (enaddr[5] << 8);
2169 		ral_hi |= RAH_AV;
2170 	} else {
2171 		ral_lo = 0;
2172 		ral_hi = 0;
2173 	}
2174 	CSR_WRITE(sc, RA_ADDR(DGE_RAL, idx), ral_lo);
2175 	CSR_WRITE(sc, RA_ADDR(DGE_RAH, idx), ral_hi);
2176 }
2177 
2178 /*
2179  * dge_mchash:
2180  *
2181  *	Compute the hash of the multicast address for the 4096-bit
2182  *	multicast filter.
2183  */
2184 static uint32_t
2185 dge_mchash(struct dge_softc *sc, const uint8_t *enaddr)
2186 {
2187 	static const int lo_shift[4] = { 4, 3, 2, 0 };
2188 	static const int hi_shift[4] = { 4, 5, 6, 8 };
2189 	uint32_t hash;
2190 
2191 	hash = (enaddr[4] >> lo_shift[sc->sc_mchash_type]) |
2192 	    (((uint16_t) enaddr[5]) << hi_shift[sc->sc_mchash_type]);
2193 
2194 	return (hash & 0xfff);
2195 }
2196 
2197 /*
2198  * dge_set_filter:
2199  *
2200  *	Set up the receive filter.
2201  */
2202 static void
2203 dge_set_filter(struct dge_softc *sc)
2204 {
2205 	struct ethercom *ec = &sc->sc_ethercom;
2206 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
2207 	struct ether_multi *enm;
2208 	struct ether_multistep step;
2209 	uint32_t hash, reg, bit;
2210 	int i;
2211 
2212 	sc->sc_rctl &= ~(RCTL_BAM | RCTL_UPE | RCTL_MPE);
2213 
2214 	if (ifp->if_flags & IFF_BROADCAST)
2215 		sc->sc_rctl |= RCTL_BAM;
2216 	if (ifp->if_flags & IFF_PROMISC) {
2217 		sc->sc_rctl |= RCTL_UPE;
2218 		goto allmulti;
2219 	}
2220 
2221 	/*
2222 	 * Set the station address in the first RAL slot, and
2223 	 * clear the remaining slots.
2224 	 */
2225 	dge_set_ral(sc, LLADDR(ifp->if_sadl), 0);
2226 	for (i = 1; i < RA_TABSIZE; i++)
2227 		dge_set_ral(sc, NULL, i);
2228 
2229 	/* Clear out the multicast table. */
2230 	for (i = 0; i < MC_TABSIZE; i++)
2231 		CSR_WRITE(sc, DGE_MTA + (i << 2), 0);
2232 
2233 	ETHER_FIRST_MULTI(step, ec, enm);
2234 	while (enm != NULL) {
2235 		if (memcmp(enm->enm_addrlo, enm->enm_addrhi, ETHER_ADDR_LEN)) {
2236 			/*
2237 			 * We must listen to a range of multicast addresses.
2238 			 * For now, just accept all multicasts, rather than
2239 			 * trying to set only those filter bits needed to match
2240 			 * the range.  (At this time, the only use of address
2241 			 * ranges is for IP multicast routing, for which the
2242 			 * range is big enough to require all bits set.)
2243 			 */
2244 			goto allmulti;
2245 		}
2246 
2247 		hash = dge_mchash(sc, enm->enm_addrlo);
2248 
2249 		reg = (hash >> 5) & 0x7f;
2250 		bit = hash & 0x1f;
2251 
2252 		hash = CSR_READ(sc, DGE_MTA + (reg << 2));
2253 		hash |= 1U << bit;
2254 
2255 		CSR_WRITE(sc, DGE_MTA + (reg << 2), hash);
2256 
2257 		ETHER_NEXT_MULTI(step, enm);
2258 	}
2259 
2260 	ifp->if_flags &= ~IFF_ALLMULTI;
2261 	goto setit;
2262 
2263  allmulti:
2264 	ifp->if_flags |= IFF_ALLMULTI;
2265 	sc->sc_rctl |= RCTL_MPE;
2266 
2267  setit:
2268 	CSR_WRITE(sc, DGE_RCTL, sc->sc_rctl);
2269 }
2270 
2271 /*
2272  * Read in the EEPROM info and verify checksum.
2273  */
2274 int
2275 dge_read_eeprom(struct dge_softc *sc)
2276 {
2277 	uint16_t cksum;
2278 	int i;
2279 
2280 	cksum = 0;
2281 	for (i = 0; i < EEPROM_SIZE; i++) {
2282 		sc->sc_eeprom[i] = dge_eeprom_word(sc, i);
2283 		cksum += sc->sc_eeprom[i];
2284 	}
2285 	return cksum != EEPROM_CKSUM;
2286 }
2287 
2288 
2289 /*
2290  * Read a 16-bit word from address addr in the serial EEPROM.
2291  */
2292 uint16_t
2293 dge_eeprom_word(struct dge_softc *sc, int addr)
2294 {
2295 	uint32_t reg;
2296 	uint16_t rval = 0;
2297 	int i;
2298 
2299 	reg = CSR_READ(sc, DGE_EECD) & ~(EECD_SK|EECD_DI|EECD_CS);
2300 
2301 	/* Lower clock pulse (and data in to chip) */
2302 	CSR_WRITE(sc, DGE_EECD, reg);
2303 	/* Select chip */
2304 	CSR_WRITE(sc, DGE_EECD, reg|EECD_CS);
2305 
2306 	/* Send read command */
2307 	dge_eeprom_clockout(sc, 1);
2308 	dge_eeprom_clockout(sc, 1);
2309 	dge_eeprom_clockout(sc, 0);
2310 
2311 	/* Send address */
2312 	for (i = 5; i >= 0; i--)
2313 		dge_eeprom_clockout(sc, (addr >> i) & 1);
2314 
2315 	/* Read data */
2316 	for (i = 0; i < 16; i++) {
2317 		rval <<= 1;
2318 		rval |= dge_eeprom_clockin(sc);
2319 	}
2320 
2321 	/* Deselect chip */
2322 	CSR_WRITE(sc, DGE_EECD, reg);
2323 
2324 	return rval;
2325 }
2326 
2327 /*
2328  * Clock out a single bit to the EEPROM.
2329  */
2330 void
2331 dge_eeprom_clockout(struct dge_softc *sc, int bit)
2332 {
2333 	int reg;
2334 
2335 	reg = CSR_READ(sc, DGE_EECD) & ~(EECD_DI|EECD_SK);
2336 	if (bit)
2337 		reg |= EECD_DI;
2338 
2339 	CSR_WRITE(sc, DGE_EECD, reg);
2340 	delay(2);
2341 	CSR_WRITE(sc, DGE_EECD, reg|EECD_SK);
2342 	delay(2);
2343 	CSR_WRITE(sc, DGE_EECD, reg);
2344 	delay(2);
2345 }
2346 
2347 /*
2348  * Clock in a single bit from EEPROM.
2349  */
2350 int
2351 dge_eeprom_clockin(struct dge_softc *sc)
2352 {
2353 	int reg, rv;
2354 
2355 	reg = CSR_READ(sc, DGE_EECD) & ~(EECD_DI|EECD_DO|EECD_SK);
2356 
2357 	CSR_WRITE(sc, DGE_EECD, reg|EECD_SK); /* Raise clock */
2358 	delay(2);
2359 	rv = (CSR_READ(sc, DGE_EECD) & EECD_DO) != 0; /* Get bit */
2360 	CSR_WRITE(sc, DGE_EECD, reg); /* Lower clock */
2361 	delay(2);
2362 
2363 	return rv;
2364 }
2365 
2366 static void
2367 dge_xgmii_mediastatus(struct ifnet *ifp, struct ifmediareq *ifmr)
2368 {
2369 	struct dge_softc *sc = ifp->if_softc;
2370 
2371 	ifmr->ifm_status = IFM_AVALID;
2372 	ifmr->ifm_active = IFM_ETHER|IFM_10G_LR;
2373 
2374 	if (CSR_READ(sc, DGE_STATUS) & STATUS_LINKUP)
2375 		ifmr->ifm_status |= IFM_ACTIVE;
2376 }
2377 
2378 static inline int
2379 phwait(struct dge_softc *sc, int p, int r, int d, int type)
2380 {
2381         int i, mdic;
2382 
2383         CSR_WRITE(sc, DGE_MDIO,
2384 	    MDIO_PHY(p) | MDIO_REG(r) | MDIO_DEV(d) | type | MDIO_CMD);
2385         for (i = 0; i < 10; i++) {
2386                 delay(10);
2387                 if (((mdic = CSR_READ(sc, DGE_MDIO)) & MDIO_CMD) == 0)
2388                         break;
2389         }
2390         return mdic;
2391 }
2392 
2393 
2394 static void
2395 dge_xgmii_writereg(struct device *self, int phy, int reg, int val)
2396 {
2397 	struct dge_softc *sc = (void *) self;
2398 	int mdic;
2399 
2400 	CSR_WRITE(sc, DGE_MDIRW, val);
2401 	if (((mdic = phwait(sc, phy, reg, 1, MDIO_ADDR)) & MDIO_CMD)) {
2402 		printf("%s: address cycle timeout; phy %d reg %d\n",
2403 		    sc->sc_dev.dv_xname, phy, reg);
2404 		return;
2405 	}
2406 	if (((mdic = phwait(sc, phy, reg, 1, MDIO_WRITE)) & MDIO_CMD)) {
2407 		printf("%s: read cycle timeout; phy %d reg %d\n",
2408 		    sc->sc_dev.dv_xname, phy, reg);
2409 		return;
2410 	}
2411 }
2412 
2413 static void
2414 dge_xgmii_reset(struct dge_softc *sc)
2415 {
2416 	dge_xgmii_writereg((void *)sc, 0, 0, BMCR_RESET);
2417 }
2418 
2419 static int
2420 dge_xgmii_mediachange(struct ifnet *ifp)
2421 {
2422 	return 0;
2423 }
2424