xref: /netbsd-src/sys/dev/pci/if_dge.c (revision 267197ec1eebfcb9810ea27a89625b6ddf68e3e7)
1 /*	$NetBSD: if_dge.c,v 1.19 2008/02/07 01:21:56 dyoung 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 1.19 2008/02/07 01:21:56 dyoung 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 <sys/bus.h>
121 #include <sys/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 	void *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 	char *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, (void **)&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 = (void *)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 (char *)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, void *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 = ((char *)buf - (char *)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(mb_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, void *);
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), (void **)&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_Tx | IFCAP_CSUM_IPv4_Rx |
903 	    IFCAP_CSUM_TCPv4_Tx | IFCAP_CSUM_TCPv4_Rx |
904 	    IFCAP_CSUM_UDPv4_Tx | IFCAP_CSUM_UDPv4_Rx;
905 
906 	/*
907 	 * Attach the interface.
908 	 */
909 	if_attach(ifp);
910 	ether_ifattach(ifp, enaddr);
911 #if NRND > 0
912 	rnd_attach_source(&sc->rnd_source, sc->sc_dev.dv_xname,
913 	    RND_TYPE_NET, 0);
914 #endif
915 
916 #ifdef DGE_EVENT_COUNTERS
917 	/* Fix segment event naming */
918 	if (dge_txseg_evcnt_names == NULL) {
919 		dge_txseg_evcnt_names =
920 		    malloc(sizeof(*dge_txseg_evcnt_names), M_DEVBUF, M_WAITOK);
921 		for (i = 0; i < DGE_NTXSEGS; i++)
922 			snprintf((*dge_txseg_evcnt_names)[i],
923 			    sizeof((*dge_txseg_evcnt_names)[i]), "txseg%d", i);
924 	}
925 
926 	/* Attach event counters. */
927 	evcnt_attach_dynamic(&sc->sc_ev_txsstall, EVCNT_TYPE_MISC,
928 	    NULL, sc->sc_dev.dv_xname, "txsstall");
929 	evcnt_attach_dynamic(&sc->sc_ev_txdstall, EVCNT_TYPE_MISC,
930 	    NULL, sc->sc_dev.dv_xname, "txdstall");
931 	evcnt_attach_dynamic(&sc->sc_ev_txforceintr, EVCNT_TYPE_MISC,
932 	    NULL, sc->sc_dev.dv_xname, "txforceintr");
933 	evcnt_attach_dynamic(&sc->sc_ev_txdw, EVCNT_TYPE_INTR,
934 	    NULL, sc->sc_dev.dv_xname, "txdw");
935 	evcnt_attach_dynamic(&sc->sc_ev_txqe, EVCNT_TYPE_INTR,
936 	    NULL, sc->sc_dev.dv_xname, "txqe");
937 	evcnt_attach_dynamic(&sc->sc_ev_rxintr, EVCNT_TYPE_INTR,
938 	    NULL, sc->sc_dev.dv_xname, "rxintr");
939 	evcnt_attach_dynamic(&sc->sc_ev_linkintr, EVCNT_TYPE_INTR,
940 	    NULL, sc->sc_dev.dv_xname, "linkintr");
941 
942 	evcnt_attach_dynamic(&sc->sc_ev_rxipsum, EVCNT_TYPE_MISC,
943 	    NULL, sc->sc_dev.dv_xname, "rxipsum");
944 	evcnt_attach_dynamic(&sc->sc_ev_rxtusum, EVCNT_TYPE_MISC,
945 	    NULL, sc->sc_dev.dv_xname, "rxtusum");
946 	evcnt_attach_dynamic(&sc->sc_ev_txipsum, EVCNT_TYPE_MISC,
947 	    NULL, sc->sc_dev.dv_xname, "txipsum");
948 	evcnt_attach_dynamic(&sc->sc_ev_txtusum, EVCNT_TYPE_MISC,
949 	    NULL, sc->sc_dev.dv_xname, "txtusum");
950 
951 	evcnt_attach_dynamic(&sc->sc_ev_txctx_init, EVCNT_TYPE_MISC,
952 	    NULL, sc->sc_dev.dv_xname, "txctx init");
953 	evcnt_attach_dynamic(&sc->sc_ev_txctx_hit, EVCNT_TYPE_MISC,
954 	    NULL, sc->sc_dev.dv_xname, "txctx hit");
955 	evcnt_attach_dynamic(&sc->sc_ev_txctx_miss, EVCNT_TYPE_MISC,
956 	    NULL, sc->sc_dev.dv_xname, "txctx miss");
957 
958 	for (i = 0; i < DGE_NTXSEGS; i++)
959 		evcnt_attach_dynamic(&sc->sc_ev_txseg[i], EVCNT_TYPE_MISC,
960 		    NULL, sc->sc_dev.dv_xname, (*dge_txseg_evcnt_names)[i]);
961 
962 	evcnt_attach_dynamic(&sc->sc_ev_txdrop, EVCNT_TYPE_MISC,
963 	    NULL, sc->sc_dev.dv_xname, "txdrop");
964 
965 #endif /* DGE_EVENT_COUNTERS */
966 
967 	/*
968 	 * Make sure the interface is shutdown during reboot.
969 	 */
970 	sc->sc_sdhook = shutdownhook_establish(dge_shutdown, sc);
971 	if (sc->sc_sdhook == NULL)
972 		aprint_error("%s: WARNING: unable to establish shutdown hook\n",
973 		    sc->sc_dev.dv_xname);
974 	return;
975 
976 	/*
977 	 * Free any resources we've allocated during the failed attach
978 	 * attempt.  Do this in reverse order and fall through.
979 	 */
980  fail_5:
981 	for (i = 0; i < DGE_NRXDESC; i++) {
982 		if (sc->sc_rxsoft[i].rxs_dmamap != NULL)
983 			bus_dmamap_destroy(sc->sc_dmat,
984 			    sc->sc_rxsoft[i].rxs_dmamap);
985 	}
986  fail_4:
987 	for (i = 0; i < DGE_TXQUEUELEN; i++) {
988 		if (sc->sc_txsoft[i].txs_dmamap != NULL)
989 			bus_dmamap_destroy(sc->sc_dmat,
990 			    sc->sc_txsoft[i].txs_dmamap);
991 	}
992 	bus_dmamap_unload(sc->sc_dmat, sc->sc_cddmamap);
993  fail_3:
994 	bus_dmamap_destroy(sc->sc_dmat, sc->sc_cddmamap);
995  fail_2:
996 	bus_dmamem_unmap(sc->sc_dmat, (void *)sc->sc_control_data,
997 	    sizeof(struct dge_control_data));
998  fail_1:
999 	bus_dmamem_free(sc->sc_dmat, &seg, rseg);
1000  fail_0:
1001 	return;
1002 }
1003 
1004 /*
1005  * dge_shutdown:
1006  *
1007  *	Make sure the interface is stopped at reboot time.
1008  */
1009 static void
1010 dge_shutdown(void *arg)
1011 {
1012 	struct dge_softc *sc = arg;
1013 
1014 	dge_stop(&sc->sc_ethercom.ec_if, 1);
1015 }
1016 
1017 /*
1018  * dge_tx_cksum:
1019  *
1020  *	Set up TCP/IP checksumming parameters for the
1021  *	specified packet.
1022  */
1023 static int
1024 dge_tx_cksum(struct dge_softc *sc, struct dge_txsoft *txs, uint8_t *fieldsp)
1025 {
1026 	struct mbuf *m0 = txs->txs_mbuf;
1027 	struct dge_ctdes *t;
1028 	uint32_t ipcs, tucs;
1029 	struct ether_header *eh;
1030 	int offset, iphl;
1031 	uint8_t fields = 0;
1032 
1033 	/*
1034 	 * XXX It would be nice if the mbuf pkthdr had offset
1035 	 * fields for the protocol headers.
1036 	 */
1037 
1038 	eh = mtod(m0, struct ether_header *);
1039 	switch (htons(eh->ether_type)) {
1040 	case ETHERTYPE_IP:
1041 		offset = ETHER_HDR_LEN;
1042 		break;
1043 
1044 	case ETHERTYPE_VLAN:
1045 		offset = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
1046 		break;
1047 
1048 	default:
1049 		/*
1050 		 * Don't support this protocol or encapsulation.
1051 		 */
1052 		*fieldsp = 0;
1053 		return (0);
1054 	}
1055 
1056 	iphl = M_CSUM_DATA_IPv4_IPHL(m0->m_pkthdr.csum_data);
1057 
1058 	/*
1059 	 * NOTE: Even if we're not using the IP or TCP/UDP checksum
1060 	 * offload feature, if we load the context descriptor, we
1061 	 * MUST provide valid values for IPCSS and TUCSS fields.
1062 	 */
1063 
1064 	if (m0->m_pkthdr.csum_flags & M_CSUM_IPv4) {
1065 		DGE_EVCNT_INCR(&sc->sc_ev_txipsum);
1066 		fields |= TDESC_POPTS_IXSM;
1067 		ipcs = DGE_TCPIP_IPCSS(offset) |
1068 		    DGE_TCPIP_IPCSO(offset + offsetof(struct ip, ip_sum)) |
1069 		    DGE_TCPIP_IPCSE(offset + iphl - 1);
1070 	} else if (__predict_true(sc->sc_txctx_ipcs != 0xffffffff)) {
1071 		/* Use the cached value. */
1072 		ipcs = sc->sc_txctx_ipcs;
1073 	} else {
1074 		/* Just initialize it to the likely value anyway. */
1075 		ipcs = DGE_TCPIP_IPCSS(offset) |
1076 		    DGE_TCPIP_IPCSO(offset + offsetof(struct ip, ip_sum)) |
1077 		    DGE_TCPIP_IPCSE(offset + iphl - 1);
1078 	}
1079 	DPRINTF(DGE_DEBUG_CKSUM,
1080 	    ("%s: CKSUM: offset %d ipcs 0x%x\n",
1081 	    sc->sc_dev.dv_xname, offset, ipcs));
1082 
1083 	offset += iphl;
1084 
1085 	if (m0->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4)) {
1086 		DGE_EVCNT_INCR(&sc->sc_ev_txtusum);
1087 		fields |= TDESC_POPTS_TXSM;
1088 		tucs = DGE_TCPIP_TUCSS(offset) |
1089 		   DGE_TCPIP_TUCSO(offset + M_CSUM_DATA_IPv4_OFFSET(m0->m_pkthdr.csum_data)) |
1090 		   DGE_TCPIP_TUCSE(0) /* rest of packet */;
1091 	} else if (__predict_true(sc->sc_txctx_tucs != 0xffffffff)) {
1092 		/* Use the cached value. */
1093 		tucs = sc->sc_txctx_tucs;
1094 	} else {
1095 		/* Just initialize it to a valid TCP context. */
1096 		tucs = DGE_TCPIP_TUCSS(offset) |
1097 		    DGE_TCPIP_TUCSO(offset + offsetof(struct tcphdr, th_sum)) |
1098 		    DGE_TCPIP_TUCSE(0) /* rest of packet */;
1099 	}
1100 
1101 	DPRINTF(DGE_DEBUG_CKSUM,
1102 	    ("%s: CKSUM: offset %d tucs 0x%x\n",
1103 	    sc->sc_dev.dv_xname, offset, tucs));
1104 
1105 	if (sc->sc_txctx_ipcs == ipcs &&
1106 	    sc->sc_txctx_tucs == tucs) {
1107 		/* Cached context is fine. */
1108 		DGE_EVCNT_INCR(&sc->sc_ev_txctx_hit);
1109 	} else {
1110 		/* Fill in the context descriptor. */
1111 #ifdef DGE_EVENT_COUNTERS
1112 		if (sc->sc_txctx_ipcs == 0xffffffff &&
1113 		    sc->sc_txctx_tucs == 0xffffffff)
1114 			DGE_EVCNT_INCR(&sc->sc_ev_txctx_init);
1115 		else
1116 			DGE_EVCNT_INCR(&sc->sc_ev_txctx_miss);
1117 #endif
1118 		t = (struct dge_ctdes *)&sc->sc_txdescs[sc->sc_txnext];
1119 		t->dc_tcpip_ipcs = htole32(ipcs);
1120 		t->dc_tcpip_tucs = htole32(tucs);
1121 		t->dc_tcpip_cmdlen = htole32(TDESC_DTYP_CTD);
1122 		t->dc_tcpip_seg = 0;
1123 		DGE_CDTXSYNC(sc, sc->sc_txnext, 1, BUS_DMASYNC_PREWRITE);
1124 
1125 		sc->sc_txctx_ipcs = ipcs;
1126 		sc->sc_txctx_tucs = tucs;
1127 
1128 		sc->sc_txnext = DGE_NEXTTX(sc->sc_txnext);
1129 		txs->txs_ndesc++;
1130 	}
1131 
1132 	*fieldsp = fields;
1133 
1134 	return (0);
1135 }
1136 
1137 /*
1138  * dge_start:		[ifnet interface function]
1139  *
1140  *	Start packet transmission on the interface.
1141  */
1142 static void
1143 dge_start(struct ifnet *ifp)
1144 {
1145 	struct dge_softc *sc = ifp->if_softc;
1146 	struct mbuf *m0;
1147 	struct dge_txsoft *txs;
1148 	bus_dmamap_t dmamap;
1149 	int error, nexttx, lasttx = -1, ofree, seg;
1150 	uint32_t cksumcmd;
1151 	uint8_t cksumfields;
1152 
1153 	if ((ifp->if_flags & (IFF_RUNNING|IFF_OACTIVE)) != IFF_RUNNING)
1154 		return;
1155 
1156 	/*
1157 	 * Remember the previous number of free descriptors.
1158 	 */
1159 	ofree = sc->sc_txfree;
1160 
1161 	/*
1162 	 * Loop through the send queue, setting up transmit descriptors
1163 	 * until we drain the queue, or use up all available transmit
1164 	 * descriptors.
1165 	 */
1166 	for (;;) {
1167 		/* Grab a packet off the queue. */
1168 		IFQ_POLL(&ifp->if_snd, m0);
1169 		if (m0 == NULL)
1170 			break;
1171 
1172 		DPRINTF(DGE_DEBUG_TX,
1173 		    ("%s: TX: have packet to transmit: %p\n",
1174 		    sc->sc_dev.dv_xname, m0));
1175 
1176 		/* Get a work queue entry. */
1177 		if (sc->sc_txsfree < DGE_TXQUEUE_GC) {
1178 			dge_txintr(sc);
1179 			if (sc->sc_txsfree == 0) {
1180 				DPRINTF(DGE_DEBUG_TX,
1181 				    ("%s: TX: no free job descriptors\n",
1182 					sc->sc_dev.dv_xname));
1183 				DGE_EVCNT_INCR(&sc->sc_ev_txsstall);
1184 				break;
1185 			}
1186 		}
1187 
1188 		txs = &sc->sc_txsoft[sc->sc_txsnext];
1189 		dmamap = txs->txs_dmamap;
1190 
1191 		/*
1192 		 * Load the DMA map.  If this fails, the packet either
1193 		 * didn't fit in the allotted number of segments, or we
1194 		 * were short on resources.  For the too-many-segments
1195 		 * case, we simply report an error and drop the packet,
1196 		 * since we can't sanely copy a jumbo packet to a single
1197 		 * buffer.
1198 		 */
1199 		error = bus_dmamap_load_mbuf(sc->sc_dmat, dmamap, m0,
1200 		    BUS_DMA_WRITE|BUS_DMA_NOWAIT);
1201 		if (error) {
1202 			if (error == EFBIG) {
1203 				DGE_EVCNT_INCR(&sc->sc_ev_txdrop);
1204 				printf("%s: Tx packet consumes too many "
1205 				    "DMA segments, dropping...\n",
1206 				    sc->sc_dev.dv_xname);
1207 				IFQ_DEQUEUE(&ifp->if_snd, m0);
1208 				m_freem(m0);
1209 				continue;
1210 			}
1211 			/*
1212 			 * Short on resources, just stop for now.
1213 			 */
1214 			DPRINTF(DGE_DEBUG_TX,
1215 			    ("%s: TX: dmamap load failed: %d\n",
1216 			    sc->sc_dev.dv_xname, error));
1217 			break;
1218 		}
1219 
1220 		/*
1221 		 * Ensure we have enough descriptors free to describe
1222 		 * the packet.  Note, we always reserve one descriptor
1223 		 * at the end of the ring due to the semantics of the
1224 		 * TDT register, plus one more in the event we need
1225 		 * to re-load checksum offload context.
1226 		 */
1227 		if (dmamap->dm_nsegs > (sc->sc_txfree - 2)) {
1228 			/*
1229 			 * Not enough free descriptors to transmit this
1230 			 * packet.  We haven't committed anything yet,
1231 			 * so just unload the DMA map, put the packet
1232 			 * pack on the queue, and punt.  Notify the upper
1233 			 * layer that there are no more slots left.
1234 			 */
1235 			DPRINTF(DGE_DEBUG_TX,
1236 			    ("%s: TX: need %d descriptors, have %d\n",
1237 			    sc->sc_dev.dv_xname, dmamap->dm_nsegs,
1238 			    sc->sc_txfree - 1));
1239 			ifp->if_flags |= IFF_OACTIVE;
1240 			bus_dmamap_unload(sc->sc_dmat, dmamap);
1241 			DGE_EVCNT_INCR(&sc->sc_ev_txdstall);
1242 			break;
1243 		}
1244 
1245 		IFQ_DEQUEUE(&ifp->if_snd, m0);
1246 
1247 		/*
1248 		 * WE ARE NOW COMMITTED TO TRANSMITTING THE PACKET.
1249 		 */
1250 
1251 		/* Sync the DMA map. */
1252 		bus_dmamap_sync(sc->sc_dmat, dmamap, 0, dmamap->dm_mapsize,
1253 		    BUS_DMASYNC_PREWRITE);
1254 
1255 		DPRINTF(DGE_DEBUG_TX,
1256 		    ("%s: TX: packet has %d DMA segments\n",
1257 		    sc->sc_dev.dv_xname, dmamap->dm_nsegs));
1258 
1259 		DGE_EVCNT_INCR(&sc->sc_ev_txseg[dmamap->dm_nsegs - 1]);
1260 
1261 		/*
1262 		 * Store a pointer to the packet so that we can free it
1263 		 * later.
1264 		 *
1265 		 * Initially, we consider the number of descriptors the
1266 		 * packet uses the number of DMA segments.  This may be
1267 		 * incremented by 1 if we do checksum offload (a descriptor
1268 		 * is used to set the checksum context).
1269 		 */
1270 		txs->txs_mbuf = m0;
1271 		txs->txs_firstdesc = sc->sc_txnext;
1272 		txs->txs_ndesc = dmamap->dm_nsegs;
1273 
1274 		/*
1275 		 * Set up checksum offload parameters for
1276 		 * this packet.
1277 		 */
1278 		if (m0->m_pkthdr.csum_flags &
1279 		    (M_CSUM_IPv4|M_CSUM_TCPv4|M_CSUM_UDPv4)) {
1280 			if (dge_tx_cksum(sc, txs, &cksumfields) != 0) {
1281 				/* Error message already displayed. */
1282 				bus_dmamap_unload(sc->sc_dmat, dmamap);
1283 				continue;
1284 			}
1285 		} else {
1286 			cksumfields = 0;
1287 		}
1288 
1289 		cksumcmd = TDESC_DCMD_IDE | TDESC_DTYP_DATA;
1290 
1291 		/*
1292 		 * Initialize the transmit descriptor.
1293 		 */
1294 		for (nexttx = sc->sc_txnext, seg = 0;
1295 		     seg < dmamap->dm_nsegs;
1296 		     seg++, nexttx = DGE_NEXTTX(nexttx)) {
1297 			/*
1298 			 * Note: we currently only use 32-bit DMA
1299 			 * addresses.
1300 			 */
1301 			sc->sc_txdescs[nexttx].dt_baddrh = 0;
1302 			sc->sc_txdescs[nexttx].dt_baddrl =
1303 			    htole32(dmamap->dm_segs[seg].ds_addr);
1304 			sc->sc_txdescs[nexttx].dt_ctl =
1305 			    htole32(cksumcmd | dmamap->dm_segs[seg].ds_len);
1306 			sc->sc_txdescs[nexttx].dt_status = 0;
1307 			sc->sc_txdescs[nexttx].dt_popts = cksumfields;
1308 			sc->sc_txdescs[nexttx].dt_vlan = 0;
1309 			lasttx = nexttx;
1310 
1311 			DPRINTF(DGE_DEBUG_TX,
1312 			    ("%s: TX: desc %d: low 0x%08lx, len 0x%04lx\n",
1313 			    sc->sc_dev.dv_xname, nexttx,
1314 			    le32toh(dmamap->dm_segs[seg].ds_addr),
1315 			    le32toh(dmamap->dm_segs[seg].ds_len)));
1316 		}
1317 
1318 		KASSERT(lasttx != -1);
1319 
1320 		/*
1321 		 * Set up the command byte on the last descriptor of
1322 		 * the packet.  If we're in the interrupt delay window,
1323 		 * delay the interrupt.
1324 		 */
1325 		sc->sc_txdescs[lasttx].dt_ctl |=
1326 		    htole32(TDESC_DCMD_EOP | TDESC_DCMD_RS);
1327 
1328 		txs->txs_lastdesc = lasttx;
1329 
1330 		DPRINTF(DGE_DEBUG_TX,
1331 		    ("%s: TX: desc %d: cmdlen 0x%08x\n", sc->sc_dev.dv_xname,
1332 		    lasttx, le32toh(sc->sc_txdescs[lasttx].dt_ctl)));
1333 
1334 		/* Sync the descriptors we're using. */
1335 		DGE_CDTXSYNC(sc, sc->sc_txnext, dmamap->dm_nsegs,
1336 		    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1337 
1338 		/* Give the packet to the chip. */
1339 		CSR_WRITE(sc, DGE_TDT, nexttx);
1340 
1341 		DPRINTF(DGE_DEBUG_TX,
1342 		    ("%s: TX: TDT -> %d\n", sc->sc_dev.dv_xname, nexttx));
1343 
1344 		DPRINTF(DGE_DEBUG_TX,
1345 		    ("%s: TX: finished transmitting packet, job %d\n",
1346 		    sc->sc_dev.dv_xname, sc->sc_txsnext));
1347 
1348 		/* Advance the tx pointer. */
1349 		sc->sc_txfree -= txs->txs_ndesc;
1350 		sc->sc_txnext = nexttx;
1351 
1352 		sc->sc_txsfree--;
1353 		sc->sc_txsnext = DGE_NEXTTXS(sc->sc_txsnext);
1354 
1355 #if NBPFILTER > 0
1356 		/* Pass the packet to any BPF listeners. */
1357 		if (ifp->if_bpf)
1358 			bpf_mtap(ifp->if_bpf, m0);
1359 #endif /* NBPFILTER > 0 */
1360 	}
1361 
1362 	if (sc->sc_txsfree == 0 || sc->sc_txfree <= 2) {
1363 		/* No more slots; notify upper layer. */
1364 		ifp->if_flags |= IFF_OACTIVE;
1365 	}
1366 
1367 	if (sc->sc_txfree != ofree) {
1368 		/* Set a watchdog timer in case the chip flakes out. */
1369 		ifp->if_timer = 5;
1370 	}
1371 }
1372 
1373 /*
1374  * dge_watchdog:		[ifnet interface function]
1375  *
1376  *	Watchdog timer handler.
1377  */
1378 static void
1379 dge_watchdog(struct ifnet *ifp)
1380 {
1381 	struct dge_softc *sc = ifp->if_softc;
1382 
1383 	/*
1384 	 * Since we're using delayed interrupts, sweep up
1385 	 * before we report an error.
1386 	 */
1387 	dge_txintr(sc);
1388 
1389 	if (sc->sc_txfree != DGE_NTXDESC) {
1390 		printf("%s: device timeout (txfree %d txsfree %d txnext %d)\n",
1391 		    sc->sc_dev.dv_xname, sc->sc_txfree, sc->sc_txsfree,
1392 		    sc->sc_txnext);
1393 		ifp->if_oerrors++;
1394 
1395 		/* Reset the interface. */
1396 		(void) dge_init(ifp);
1397 	}
1398 
1399 	/* Try to get more packets going. */
1400 	dge_start(ifp);
1401 }
1402 
1403 /*
1404  * dge_ioctl:		[ifnet interface function]
1405  *
1406  *	Handle control requests from the operator.
1407  */
1408 static int
1409 dge_ioctl(struct ifnet *ifp, u_long cmd, void *data)
1410 {
1411 	struct dge_softc *sc = ifp->if_softc;
1412 	struct ifreq *ifr = (struct ifreq *) data;
1413 	pcireg_t preg;
1414 	int s, error, mmrbc;
1415 
1416 	s = splnet();
1417 
1418 	switch (cmd) {
1419 	case SIOCSIFMEDIA:
1420 	case SIOCGIFMEDIA:
1421 		error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
1422 		break;
1423 
1424 	case SIOCSIFMTU:
1425 		if (ifr->ifr_mtu < ETHERMIN || ifr->ifr_mtu > DGE_MAX_MTU)
1426 			error = EINVAL;
1427 		else if ((error = ifioctl_common(ifp, cmd, data)) != ENETRESET)
1428 			break;
1429 		else if (ifp->if_flags & IFF_UP)
1430 			error = (*ifp->if_init)(ifp);
1431 		else
1432 			error = 0;
1433 		break;
1434 
1435         case SIOCSIFFLAGS:
1436 		/* extract link flags */
1437 		if ((ifp->if_flags & IFF_LINK0) == 0 &&
1438 		    (ifp->if_flags & IFF_LINK1) == 0)
1439 			mmrbc = PCIX_MMRBC_512;
1440 		else if ((ifp->if_flags & IFF_LINK0) == 0 &&
1441 		    (ifp->if_flags & IFF_LINK1) != 0)
1442 			mmrbc = PCIX_MMRBC_1024;
1443 		else if ((ifp->if_flags & IFF_LINK0) != 0 &&
1444 		    (ifp->if_flags & IFF_LINK1) == 0)
1445 			mmrbc = PCIX_MMRBC_2048;
1446 		else
1447 			mmrbc = PCIX_MMRBC_4096;
1448 		if (mmrbc != sc->sc_mmrbc) {
1449 			preg = pci_conf_read(sc->sc_pc, sc->sc_pt,DGE_PCIX_CMD);
1450 			preg &= ~PCIX_MMRBC_MSK;
1451 			preg |= mmrbc;
1452 			pci_conf_write(sc->sc_pc, sc->sc_pt,DGE_PCIX_CMD, preg);
1453 			sc->sc_mmrbc = mmrbc;
1454 		}
1455                 /* FALLTHROUGH */
1456 	default:
1457 		if ((error = ether_ioctl(ifp, cmd, data)) != ENETRESET)
1458 			break;
1459 
1460 		error = 0;
1461 
1462 		if (cmd == SIOCSIFCAP)
1463 			error = (*ifp->if_init)(ifp);
1464 		else if (cmd != SIOCADDMULTI && cmd != SIOCDELMULTI)
1465 			;
1466 		else if (ifp->if_flags & IFF_RUNNING) {
1467 			/*
1468 			 * Multicast list has changed; set the hardware filter
1469 			 * accordingly.
1470 			 */
1471 			dge_set_filter(sc);
1472 		}
1473 		break;
1474 	}
1475 
1476 	/* Try to get more packets going. */
1477 	dge_start(ifp);
1478 
1479 	splx(s);
1480 	return (error);
1481 }
1482 
1483 /*
1484  * dge_intr:
1485  *
1486  *	Interrupt service routine.
1487  */
1488 static int
1489 dge_intr(void *arg)
1490 {
1491 	struct dge_softc *sc = arg;
1492 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
1493 	uint32_t icr;
1494 	int wantinit, handled = 0;
1495 
1496 	for (wantinit = 0; wantinit == 0;) {
1497 		icr = CSR_READ(sc, DGE_ICR);
1498 		if ((icr & sc->sc_icr) == 0)
1499 			break;
1500 
1501 #if 0 /*NRND > 0*/
1502 		if (RND_ENABLED(&sc->rnd_source))
1503 			rnd_add_uint32(&sc->rnd_source, icr);
1504 #endif
1505 
1506 		handled = 1;
1507 
1508 #if defined(DGE_DEBUG) || defined(DGE_EVENT_COUNTERS)
1509 		if (icr & (ICR_RXDMT0|ICR_RXT0)) {
1510 			DPRINTF(DGE_DEBUG_RX,
1511 			    ("%s: RX: got Rx intr 0x%08x\n",
1512 			    sc->sc_dev.dv_xname,
1513 			    icr & (ICR_RXDMT0|ICR_RXT0)));
1514 			DGE_EVCNT_INCR(&sc->sc_ev_rxintr);
1515 		}
1516 #endif
1517 		dge_rxintr(sc);
1518 
1519 #if defined(DGE_DEBUG) || defined(DGE_EVENT_COUNTERS)
1520 		if (icr & ICR_TXDW) {
1521 			DPRINTF(DGE_DEBUG_TX,
1522 			    ("%s: TX: got TXDW interrupt\n",
1523 			    sc->sc_dev.dv_xname));
1524 			DGE_EVCNT_INCR(&sc->sc_ev_txdw);
1525 		}
1526 		if (icr & ICR_TXQE)
1527 			DGE_EVCNT_INCR(&sc->sc_ev_txqe);
1528 #endif
1529 		dge_txintr(sc);
1530 
1531 		if (icr & (ICR_LSC|ICR_RXSEQ)) {
1532 			DGE_EVCNT_INCR(&sc->sc_ev_linkintr);
1533 			dge_linkintr(sc, icr);
1534 		}
1535 
1536 		if (icr & ICR_RXO) {
1537 			printf("%s: Receive overrun\n", sc->sc_dev.dv_xname);
1538 			wantinit = 1;
1539 		}
1540 	}
1541 
1542 	if (handled) {
1543 		if (wantinit)
1544 			dge_init(ifp);
1545 
1546 		/* Try to get more packets going. */
1547 		dge_start(ifp);
1548 	}
1549 
1550 	return (handled);
1551 }
1552 
1553 /*
1554  * dge_txintr:
1555  *
1556  *	Helper; handle transmit interrupts.
1557  */
1558 static void
1559 dge_txintr(struct dge_softc *sc)
1560 {
1561 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
1562 	struct dge_txsoft *txs;
1563 	uint8_t status;
1564 	int i;
1565 
1566 	ifp->if_flags &= ~IFF_OACTIVE;
1567 
1568 	/*
1569 	 * Go through the Tx list and free mbufs for those
1570 	 * frames which have been transmitted.
1571 	 */
1572 	for (i = sc->sc_txsdirty; sc->sc_txsfree != DGE_TXQUEUELEN;
1573 	     i = DGE_NEXTTXS(i), sc->sc_txsfree++) {
1574 		txs = &sc->sc_txsoft[i];
1575 
1576 		DPRINTF(DGE_DEBUG_TX,
1577 		    ("%s: TX: checking job %d\n", sc->sc_dev.dv_xname, i));
1578 
1579 		DGE_CDTXSYNC(sc, txs->txs_firstdesc, txs->txs_dmamap->dm_nsegs,
1580 		    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1581 
1582 		status =
1583 		    sc->sc_txdescs[txs->txs_lastdesc].dt_status;
1584 		if ((status & TDESC_STA_DD) == 0) {
1585 			DGE_CDTXSYNC(sc, txs->txs_lastdesc, 1,
1586 			    BUS_DMASYNC_PREREAD);
1587 			break;
1588 		}
1589 
1590 		DPRINTF(DGE_DEBUG_TX,
1591 		    ("%s: TX: job %d done: descs %d..%d\n",
1592 		    sc->sc_dev.dv_xname, i, txs->txs_firstdesc,
1593 		    txs->txs_lastdesc));
1594 
1595 		ifp->if_opackets++;
1596 		sc->sc_txfree += txs->txs_ndesc;
1597 		bus_dmamap_sync(sc->sc_dmat, txs->txs_dmamap,
1598 		    0, txs->txs_dmamap->dm_mapsize, BUS_DMASYNC_POSTWRITE);
1599 		bus_dmamap_unload(sc->sc_dmat, txs->txs_dmamap);
1600 		m_freem(txs->txs_mbuf);
1601 		txs->txs_mbuf = NULL;
1602 	}
1603 
1604 	/* Update the dirty transmit buffer pointer. */
1605 	sc->sc_txsdirty = i;
1606 	DPRINTF(DGE_DEBUG_TX,
1607 	    ("%s: TX: txsdirty -> %d\n", sc->sc_dev.dv_xname, i));
1608 
1609 	/*
1610 	 * If there are no more pending transmissions, cancel the watchdog
1611 	 * timer.
1612 	 */
1613 	if (sc->sc_txsfree == DGE_TXQUEUELEN)
1614 		ifp->if_timer = 0;
1615 }
1616 
1617 /*
1618  * dge_rxintr:
1619  *
1620  *	Helper; handle receive interrupts.
1621  */
1622 static void
1623 dge_rxintr(struct dge_softc *sc)
1624 {
1625 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
1626 	struct dge_rxsoft *rxs;
1627 	struct mbuf *m;
1628 	int i, len;
1629 	uint8_t status, errors;
1630 
1631 	for (i = sc->sc_rxptr;; i = DGE_NEXTRX(i)) {
1632 		rxs = &sc->sc_rxsoft[i];
1633 
1634 		DPRINTF(DGE_DEBUG_RX,
1635 		    ("%s: RX: checking descriptor %d\n",
1636 		    sc->sc_dev.dv_xname, i));
1637 
1638 		DGE_CDRXSYNC(sc, i, BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1639 
1640 		status = sc->sc_rxdescs[i].dr_status;
1641 		errors = sc->sc_rxdescs[i].dr_errors;
1642 		len = le16toh(sc->sc_rxdescs[i].dr_len);
1643 
1644 		if ((status & RDESC_STS_DD) == 0) {
1645 			/*
1646 			 * We have processed all of the receive descriptors.
1647 			 */
1648 			DGE_CDRXSYNC(sc, i, BUS_DMASYNC_PREREAD);
1649 			break;
1650 		}
1651 
1652 		if (__predict_false(sc->sc_rxdiscard)) {
1653 			DPRINTF(DGE_DEBUG_RX,
1654 			    ("%s: RX: discarding contents of descriptor %d\n",
1655 			    sc->sc_dev.dv_xname, i));
1656 			DGE_INIT_RXDESC(sc, i);
1657 			if (status & RDESC_STS_EOP) {
1658 				/* Reset our state. */
1659 				DPRINTF(DGE_DEBUG_RX,
1660 				    ("%s: RX: resetting rxdiscard -> 0\n",
1661 				    sc->sc_dev.dv_xname));
1662 				sc->sc_rxdiscard = 0;
1663 			}
1664 			continue;
1665 		}
1666 
1667 		bus_dmamap_sync(sc->sc_dmat, rxs->rxs_dmamap, 0,
1668 		    rxs->rxs_dmamap->dm_mapsize, BUS_DMASYNC_POSTREAD);
1669 
1670 		m = rxs->rxs_mbuf;
1671 
1672 		/*
1673 		 * Add a new receive buffer to the ring.
1674 		 */
1675 		if (dge_add_rxbuf(sc, i) != 0) {
1676 			/*
1677 			 * Failed, throw away what we've done so
1678 			 * far, and discard the rest of the packet.
1679 			 */
1680 			ifp->if_ierrors++;
1681 			bus_dmamap_sync(sc->sc_dmat, rxs->rxs_dmamap, 0,
1682 			    rxs->rxs_dmamap->dm_mapsize, BUS_DMASYNC_PREREAD);
1683 			DGE_INIT_RXDESC(sc, i);
1684 			if ((status & RDESC_STS_EOP) == 0)
1685 				sc->sc_rxdiscard = 1;
1686 			if (sc->sc_rxhead != NULL)
1687 				m_freem(sc->sc_rxhead);
1688 			DGE_RXCHAIN_RESET(sc);
1689 			DPRINTF(DGE_DEBUG_RX,
1690 			    ("%s: RX: Rx buffer allocation failed, "
1691 			    "dropping packet%s\n", sc->sc_dev.dv_xname,
1692 			    sc->sc_rxdiscard ? " (discard)" : ""));
1693 			continue;
1694 		}
1695 		DGE_INIT_RXDESC(sc, DGE_PREVRX(i)); /* Write the descriptor */
1696 
1697 		DGE_RXCHAIN_LINK(sc, m);
1698 
1699 		m->m_len = len;
1700 
1701 		DPRINTF(DGE_DEBUG_RX,
1702 		    ("%s: RX: buffer at %p len %d\n",
1703 		    sc->sc_dev.dv_xname, m->m_data, len));
1704 
1705 		/*
1706 		 * If this is not the end of the packet, keep
1707 		 * looking.
1708 		 */
1709 		if ((status & RDESC_STS_EOP) == 0) {
1710 			sc->sc_rxlen += len;
1711 			DPRINTF(DGE_DEBUG_RX,
1712 			    ("%s: RX: not yet EOP, rxlen -> %d\n",
1713 			    sc->sc_dev.dv_xname, sc->sc_rxlen));
1714 			continue;
1715 		}
1716 
1717 		/*
1718 		 * Okay, we have the entire packet now...
1719 		 */
1720 		*sc->sc_rxtailp = NULL;
1721 		m = sc->sc_rxhead;
1722 		len += sc->sc_rxlen;
1723 
1724 		DGE_RXCHAIN_RESET(sc);
1725 
1726 		DPRINTF(DGE_DEBUG_RX,
1727 		    ("%s: RX: have entire packet, len -> %d\n",
1728 		    sc->sc_dev.dv_xname, len));
1729 
1730 		/*
1731 		 * If an error occurred, update stats and drop the packet.
1732 		 */
1733 		if (errors &
1734 		     (RDESC_ERR_CE|RDESC_ERR_SE|RDESC_ERR_P|RDESC_ERR_RXE)) {
1735 			ifp->if_ierrors++;
1736 			if (errors & RDESC_ERR_SE)
1737 				printf("%s: symbol error\n",
1738 				    sc->sc_dev.dv_xname);
1739 			else if (errors & RDESC_ERR_P)
1740 				printf("%s: parity error\n",
1741 				    sc->sc_dev.dv_xname);
1742 			else if (errors & RDESC_ERR_CE)
1743 				printf("%s: CRC error\n",
1744 				    sc->sc_dev.dv_xname);
1745 			m_freem(m);
1746 			continue;
1747 		}
1748 
1749 		/*
1750 		 * No errors.  Receive the packet.
1751 		 */
1752 		m->m_pkthdr.rcvif = ifp;
1753 		m->m_pkthdr.len = len;
1754 
1755 		/*
1756 		 * Set up checksum info for this packet.
1757 		 */
1758 		if (status & RDESC_STS_IPCS) {
1759 			DGE_EVCNT_INCR(&sc->sc_ev_rxipsum);
1760 			m->m_pkthdr.csum_flags |= M_CSUM_IPv4;
1761 			if (errors & RDESC_ERR_IPE)
1762 				m->m_pkthdr.csum_flags |= M_CSUM_IPv4_BAD;
1763 		}
1764 		if (status & RDESC_STS_TCPCS) {
1765 			/*
1766 			 * Note: we don't know if this was TCP or UDP,
1767 			 * so we just set both bits, and expect the
1768 			 * upper layers to deal.
1769 			 */
1770 			DGE_EVCNT_INCR(&sc->sc_ev_rxtusum);
1771 			m->m_pkthdr.csum_flags |= M_CSUM_TCPv4|M_CSUM_UDPv4;
1772 			if (errors & RDESC_ERR_TCPE)
1773 				m->m_pkthdr.csum_flags |= M_CSUM_TCP_UDP_BAD;
1774 		}
1775 
1776 		ifp->if_ipackets++;
1777 
1778 #if NBPFILTER > 0
1779 		/* Pass this up to any BPF listeners. */
1780 		if (ifp->if_bpf)
1781 			bpf_mtap(ifp->if_bpf, m);
1782 #endif /* NBPFILTER > 0 */
1783 
1784 		/* Pass it on. */
1785 		(*ifp->if_input)(ifp, m);
1786 	}
1787 
1788 	/* Update the receive pointer. */
1789 	sc->sc_rxptr = i;
1790 
1791 	DPRINTF(DGE_DEBUG_RX,
1792 	    ("%s: RX: rxptr -> %d\n", sc->sc_dev.dv_xname, i));
1793 }
1794 
1795 /*
1796  * dge_linkintr:
1797  *
1798  *	Helper; handle link interrupts.
1799  */
1800 static void
1801 dge_linkintr(struct dge_softc *sc, uint32_t icr)
1802 {
1803 	uint32_t status;
1804 
1805 	if (icr & ICR_LSC) {
1806 		status = CSR_READ(sc, DGE_STATUS);
1807 		if (status & STATUS_LINKUP) {
1808 			DPRINTF(DGE_DEBUG_LINK, ("%s: LINK: LSC -> up\n",
1809 			    sc->sc_dev.dv_xname));
1810 		} else {
1811 			DPRINTF(DGE_DEBUG_LINK, ("%s: LINK: LSC -> down\n",
1812 			    sc->sc_dev.dv_xname));
1813 		}
1814 	} else if (icr & ICR_RXSEQ) {
1815 		DPRINTF(DGE_DEBUG_LINK,
1816 		    ("%s: LINK: Receive sequence error\n",
1817 		    sc->sc_dev.dv_xname));
1818 	}
1819 	/* XXX - fix errata */
1820 }
1821 
1822 /*
1823  * dge_reset:
1824  *
1825  *	Reset the i82597 chip.
1826  */
1827 static void
1828 dge_reset(struct dge_softc *sc)
1829 {
1830 	int i;
1831 
1832 	/*
1833 	 * Do a chip reset.
1834 	 */
1835 	CSR_WRITE(sc, DGE_CTRL0, CTRL0_RST | sc->sc_ctrl0);
1836 
1837 	delay(10000);
1838 
1839 	for (i = 0; i < 1000; i++) {
1840 		if ((CSR_READ(sc, DGE_CTRL0) & CTRL0_RST) == 0)
1841 			break;
1842 		delay(20);
1843 	}
1844 
1845 	if (CSR_READ(sc, DGE_CTRL0) & CTRL0_RST)
1846 		printf("%s: WARNING: reset failed to complete\n",
1847 		    sc->sc_dev.dv_xname);
1848         /*
1849          * Reset the EEPROM logic.
1850          * This will cause the chip to reread its default values,
1851 	 * which doesn't happen otherwise (errata).
1852          */
1853         CSR_WRITE(sc, DGE_CTRL1, CTRL1_EE_RST);
1854         delay(10000);
1855 }
1856 
1857 /*
1858  * dge_init:		[ifnet interface function]
1859  *
1860  *	Initialize the interface.  Must be called at splnet().
1861  */
1862 static int
1863 dge_init(struct ifnet *ifp)
1864 {
1865 	struct dge_softc *sc = ifp->if_softc;
1866 	struct dge_rxsoft *rxs;
1867 	int i, error = 0;
1868 	uint32_t reg;
1869 
1870 	/*
1871 	 * *_HDR_ALIGNED_P is constant 1 if __NO_STRICT_ALIGMENT is set.
1872 	 * There is a small but measurable benefit to avoiding the adjusment
1873 	 * of the descriptor so that the headers are aligned, for normal mtu,
1874 	 * on such platforms.  One possibility is that the DMA itself is
1875 	 * slightly more efficient if the front of the entire packet (instead
1876 	 * of the front of the headers) is aligned.
1877 	 *
1878 	 * Note we must always set align_tweak to 0 if we are using
1879 	 * jumbo frames.
1880 	 */
1881 #ifdef __NO_STRICT_ALIGNMENT
1882 	sc->sc_align_tweak = 0;
1883 #else
1884 	if ((ifp->if_mtu + ETHER_HDR_LEN + ETHER_CRC_LEN) > (MCLBYTES - 2))
1885 		sc->sc_align_tweak = 0;
1886 	else
1887 		sc->sc_align_tweak = 2;
1888 #endif /* __NO_STRICT_ALIGNMENT */
1889 
1890 	/* Cancel any pending I/O. */
1891 	dge_stop(ifp, 0);
1892 
1893 	/* Reset the chip to a known state. */
1894 	dge_reset(sc);
1895 
1896 	/* Initialize the transmit descriptor ring. */
1897 	memset(sc->sc_txdescs, 0, sizeof(sc->sc_txdescs));
1898 	DGE_CDTXSYNC(sc, 0, DGE_NTXDESC,
1899 	    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1900 	sc->sc_txfree = DGE_NTXDESC;
1901 	sc->sc_txnext = 0;
1902 
1903 	sc->sc_txctx_ipcs = 0xffffffff;
1904 	sc->sc_txctx_tucs = 0xffffffff;
1905 
1906 	CSR_WRITE(sc, DGE_TDBAH, 0);
1907 	CSR_WRITE(sc, DGE_TDBAL, DGE_CDTXADDR(sc, 0));
1908 	CSR_WRITE(sc, DGE_TDLEN, sizeof(sc->sc_txdescs));
1909 	CSR_WRITE(sc, DGE_TDH, 0);
1910 	CSR_WRITE(sc, DGE_TDT, 0);
1911 	CSR_WRITE(sc, DGE_TIDV, TIDV);
1912 
1913 #if 0
1914 	CSR_WRITE(sc, DGE_TXDCTL, TXDCTL_PTHRESH(0) |
1915 	    TXDCTL_HTHRESH(0) | TXDCTL_WTHRESH(0));
1916 #endif
1917 	CSR_WRITE(sc, DGE_RXDCTL,
1918 	    RXDCTL_PTHRESH(RXDCTL_PTHRESH_VAL) |
1919 	    RXDCTL_HTHRESH(RXDCTL_HTHRESH_VAL) |
1920 	    RXDCTL_WTHRESH(RXDCTL_WTHRESH_VAL));
1921 
1922 	/* Initialize the transmit job descriptors. */
1923 	for (i = 0; i < DGE_TXQUEUELEN; i++)
1924 		sc->sc_txsoft[i].txs_mbuf = NULL;
1925 	sc->sc_txsfree = DGE_TXQUEUELEN;
1926 	sc->sc_txsnext = 0;
1927 	sc->sc_txsdirty = 0;
1928 
1929 	/*
1930 	 * Initialize the receive descriptor and receive job
1931 	 * descriptor rings.
1932 	 */
1933 	CSR_WRITE(sc, DGE_RDBAH, 0);
1934 	CSR_WRITE(sc, DGE_RDBAL, DGE_CDRXADDR(sc, 0));
1935 	CSR_WRITE(sc, DGE_RDLEN, sizeof(sc->sc_rxdescs));
1936 	CSR_WRITE(sc, DGE_RDH, DGE_RXSPACE);
1937 	CSR_WRITE(sc, DGE_RDT, 0);
1938 	CSR_WRITE(sc, DGE_RDTR, RDTR | 0x80000000);
1939 	CSR_WRITE(sc, DGE_FCRTL, FCRTL | FCRTL_XONE);
1940 	CSR_WRITE(sc, DGE_FCRTH, FCRTH);
1941 
1942 	for (i = 0; i < DGE_NRXDESC; i++) {
1943 		rxs = &sc->sc_rxsoft[i];
1944 		if (rxs->rxs_mbuf == NULL) {
1945 			if ((error = dge_add_rxbuf(sc, i)) != 0) {
1946 				printf("%s: unable to allocate or map rx "
1947 				    "buffer %d, error = %d\n",
1948 				    sc->sc_dev.dv_xname, i, error);
1949 				/*
1950 				 * XXX Should attempt to run with fewer receive
1951 				 * XXX buffers instead of just failing.
1952 				 */
1953 				dge_rxdrain(sc);
1954 				goto out;
1955 			}
1956 		}
1957 		DGE_INIT_RXDESC(sc, i);
1958 	}
1959 	sc->sc_rxptr = DGE_RXSPACE;
1960 	sc->sc_rxdiscard = 0;
1961 	DGE_RXCHAIN_RESET(sc);
1962 
1963 	if (sc->sc_ethercom.ec_capabilities & ETHERCAP_JUMBO_MTU) {
1964 		sc->sc_ctrl0 |= CTRL0_JFE;
1965 		CSR_WRITE(sc, DGE_MFS, ETHER_MAX_LEN_JUMBO << 16);
1966 	}
1967 
1968 	/* Write the control registers. */
1969 	CSR_WRITE(sc, DGE_CTRL0, sc->sc_ctrl0);
1970 
1971 	/*
1972 	 * Set up checksum offload parameters.
1973 	 */
1974 	reg = CSR_READ(sc, DGE_RXCSUM);
1975 	if (ifp->if_capenable & IFCAP_CSUM_IPv4_Rx)
1976 		reg |= RXCSUM_IPOFL;
1977 	else
1978 		reg &= ~RXCSUM_IPOFL;
1979 	if (ifp->if_capenable & (IFCAP_CSUM_TCPv4_Rx | IFCAP_CSUM_UDPv4_Rx))
1980 		reg |= RXCSUM_IPOFL | RXCSUM_TUOFL;
1981 	else {
1982 		reg &= ~RXCSUM_TUOFL;
1983 		if ((ifp->if_capenable & IFCAP_CSUM_IPv4_Rx) == 0)
1984 			reg &= ~RXCSUM_IPOFL;
1985 	}
1986 	CSR_WRITE(sc, DGE_RXCSUM, reg);
1987 
1988 	/*
1989 	 * Set up the interrupt registers.
1990 	 */
1991 	CSR_WRITE(sc, DGE_IMC, 0xffffffffU);
1992 	sc->sc_icr = ICR_TXDW | ICR_LSC | ICR_RXSEQ | ICR_RXDMT0 |
1993 	    ICR_RXO | ICR_RXT0;
1994 
1995 	CSR_WRITE(sc, DGE_IMS, sc->sc_icr);
1996 
1997 	/*
1998 	 * Set up the transmit control register.
1999 	 */
2000 	sc->sc_tctl = TCTL_TCE|TCTL_TPDE|TCTL_TXEN;
2001 	CSR_WRITE(sc, DGE_TCTL, sc->sc_tctl);
2002 
2003 	/*
2004 	 * Set up the receive control register; we actually program
2005 	 * the register when we set the receive filter.  Use multicast
2006 	 * address offset type 0.
2007 	 */
2008 	sc->sc_mchash_type = 0;
2009 
2010 	sc->sc_rctl = RCTL_RXEN | RCTL_RDMTS_12 | RCTL_RPDA_MC |
2011 	    RCTL_CFF | RCTL_SECRC | RCTL_MO(sc->sc_mchash_type);
2012 
2013 #ifdef DGE_OFFBYONE_RXBUG
2014 	sc->sc_rctl |= RCTL_BSIZE_16k;
2015 #else
2016 	switch(MCLBYTES) {
2017 	case 2048:
2018 		sc->sc_rctl |= RCTL_BSIZE_2k;
2019 		break;
2020 	case 4096:
2021 		sc->sc_rctl |= RCTL_BSIZE_4k;
2022 		break;
2023 	case 8192:
2024 		sc->sc_rctl |= RCTL_BSIZE_8k;
2025 		break;
2026 	case 16384:
2027 		sc->sc_rctl |= RCTL_BSIZE_16k;
2028 		break;
2029 	default:
2030 		panic("dge_init: MCLBYTES %d unsupported", MCLBYTES);
2031 	}
2032 #endif
2033 
2034 	/* Set the receive filter. */
2035 	/* Also sets RCTL */
2036 	dge_set_filter(sc);
2037 
2038 	/* ...all done! */
2039 	ifp->if_flags |= IFF_RUNNING;
2040 	ifp->if_flags &= ~IFF_OACTIVE;
2041 
2042  out:
2043 	if (error)
2044 		printf("%s: interface not running\n", sc->sc_dev.dv_xname);
2045 	return (error);
2046 }
2047 
2048 /*
2049  * dge_rxdrain:
2050  *
2051  *	Drain the receive queue.
2052  */
2053 static void
2054 dge_rxdrain(struct dge_softc *sc)
2055 {
2056 	struct dge_rxsoft *rxs;
2057 	int i;
2058 
2059 	for (i = 0; i < DGE_NRXDESC; i++) {
2060 		rxs = &sc->sc_rxsoft[i];
2061 		if (rxs->rxs_mbuf != NULL) {
2062 			bus_dmamap_unload(sc->sc_dmat, rxs->rxs_dmamap);
2063 			m_freem(rxs->rxs_mbuf);
2064 			rxs->rxs_mbuf = NULL;
2065 		}
2066 	}
2067 }
2068 
2069 /*
2070  * dge_stop:		[ifnet interface function]
2071  *
2072  *	Stop transmission on the interface.
2073  */
2074 static void
2075 dge_stop(struct ifnet *ifp, int disable)
2076 {
2077 	struct dge_softc *sc = ifp->if_softc;
2078 	struct dge_txsoft *txs;
2079 	int i;
2080 
2081 	/* Stop the transmit and receive processes. */
2082 	CSR_WRITE(sc, DGE_TCTL, 0);
2083 	CSR_WRITE(sc, DGE_RCTL, 0);
2084 
2085 	/* Release any queued transmit buffers. */
2086 	for (i = 0; i < DGE_TXQUEUELEN; i++) {
2087 		txs = &sc->sc_txsoft[i];
2088 		if (txs->txs_mbuf != NULL) {
2089 			bus_dmamap_unload(sc->sc_dmat, txs->txs_dmamap);
2090 			m_freem(txs->txs_mbuf);
2091 			txs->txs_mbuf = NULL;
2092 		}
2093 	}
2094 
2095 	if (disable)
2096 		dge_rxdrain(sc);
2097 
2098 	/* Mark the interface as down and cancel the watchdog timer. */
2099 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
2100 	ifp->if_timer = 0;
2101 }
2102 
2103 /*
2104  * dge_add_rxbuf:
2105  *
2106  *	Add a receive buffer to the indiciated descriptor.
2107  */
2108 static int
2109 dge_add_rxbuf(struct dge_softc *sc, int idx)
2110 {
2111 	struct dge_rxsoft *rxs = &sc->sc_rxsoft[idx];
2112 	struct mbuf *m;
2113 	int error;
2114 #ifdef DGE_OFFBYONE_RXBUG
2115 	void *buf;
2116 #endif
2117 
2118 	MGETHDR(m, M_DONTWAIT, MT_DATA);
2119 	if (m == NULL)
2120 		return (ENOBUFS);
2121 
2122 #ifdef DGE_OFFBYONE_RXBUG
2123 	if ((buf = dge_getbuf(sc)) == NULL)
2124 		return ENOBUFS;
2125 
2126 	m->m_len = m->m_pkthdr.len = DGE_BUFFER_SIZE;
2127 	MEXTADD(m, buf, DGE_BUFFER_SIZE, M_DEVBUF, dge_freebuf, sc);
2128 	m->m_flags |= M_EXT_RW;
2129 
2130 	if (rxs->rxs_mbuf != NULL)
2131 		bus_dmamap_unload(sc->sc_dmat, rxs->rxs_dmamap);
2132 	rxs->rxs_mbuf = m;
2133 
2134 	error = bus_dmamap_load(sc->sc_dmat, rxs->rxs_dmamap, buf,
2135 	    DGE_BUFFER_SIZE, NULL, BUS_DMA_READ|BUS_DMA_NOWAIT);
2136 #else
2137 	MCLGET(m, M_DONTWAIT);
2138 	if ((m->m_flags & M_EXT) == 0) {
2139 		m_freem(m);
2140 		return (ENOBUFS);
2141 	}
2142 
2143 	if (rxs->rxs_mbuf != NULL)
2144 		bus_dmamap_unload(sc->sc_dmat, rxs->rxs_dmamap);
2145 
2146 	rxs->rxs_mbuf = m;
2147 
2148 	m->m_len = m->m_pkthdr.len = m->m_ext.ext_size;
2149 	error = bus_dmamap_load_mbuf(sc->sc_dmat, rxs->rxs_dmamap, m,
2150 	    BUS_DMA_READ|BUS_DMA_NOWAIT);
2151 #endif
2152 	if (error) {
2153 		printf("%s: unable to load rx DMA map %d, error = %d\n",
2154 		    sc->sc_dev.dv_xname, idx, error);
2155 		panic("dge_add_rxbuf");	/* XXX XXX XXX */
2156 	}
2157 	bus_dmamap_sync(sc->sc_dmat, rxs->rxs_dmamap, 0,
2158 	    rxs->rxs_dmamap->dm_mapsize, BUS_DMASYNC_PREREAD);
2159 
2160 	return (0);
2161 }
2162 
2163 /*
2164  * dge_set_ral:
2165  *
2166  *	Set an entry in the receive address list.
2167  */
2168 static void
2169 dge_set_ral(struct dge_softc *sc, const uint8_t *enaddr, int idx)
2170 {
2171 	uint32_t ral_lo, ral_hi;
2172 
2173 	if (enaddr != NULL) {
2174 		ral_lo = enaddr[0] | (enaddr[1] << 8) | (enaddr[2] << 16) |
2175 		    (enaddr[3] << 24);
2176 		ral_hi = enaddr[4] | (enaddr[5] << 8);
2177 		ral_hi |= RAH_AV;
2178 	} else {
2179 		ral_lo = 0;
2180 		ral_hi = 0;
2181 	}
2182 	CSR_WRITE(sc, RA_ADDR(DGE_RAL, idx), ral_lo);
2183 	CSR_WRITE(sc, RA_ADDR(DGE_RAH, idx), ral_hi);
2184 }
2185 
2186 /*
2187  * dge_mchash:
2188  *
2189  *	Compute the hash of the multicast address for the 4096-bit
2190  *	multicast filter.
2191  */
2192 static uint32_t
2193 dge_mchash(struct dge_softc *sc, const uint8_t *enaddr)
2194 {
2195 	static const int lo_shift[4] = { 4, 3, 2, 0 };
2196 	static const int hi_shift[4] = { 4, 5, 6, 8 };
2197 	uint32_t hash;
2198 
2199 	hash = (enaddr[4] >> lo_shift[sc->sc_mchash_type]) |
2200 	    (((uint16_t) enaddr[5]) << hi_shift[sc->sc_mchash_type]);
2201 
2202 	return (hash & 0xfff);
2203 }
2204 
2205 /*
2206  * dge_set_filter:
2207  *
2208  *	Set up the receive filter.
2209  */
2210 static void
2211 dge_set_filter(struct dge_softc *sc)
2212 {
2213 	struct ethercom *ec = &sc->sc_ethercom;
2214 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
2215 	struct ether_multi *enm;
2216 	struct ether_multistep step;
2217 	uint32_t hash, reg, bit;
2218 	int i;
2219 
2220 	sc->sc_rctl &= ~(RCTL_BAM | RCTL_UPE | RCTL_MPE);
2221 
2222 	if (ifp->if_flags & IFF_BROADCAST)
2223 		sc->sc_rctl |= RCTL_BAM;
2224 	if (ifp->if_flags & IFF_PROMISC) {
2225 		sc->sc_rctl |= RCTL_UPE;
2226 		goto allmulti;
2227 	}
2228 
2229 	/*
2230 	 * Set the station address in the first RAL slot, and
2231 	 * clear the remaining slots.
2232 	 */
2233 	dge_set_ral(sc, CLLADDR(ifp->if_sadl), 0);
2234 	for (i = 1; i < RA_TABSIZE; i++)
2235 		dge_set_ral(sc, NULL, i);
2236 
2237 	/* Clear out the multicast table. */
2238 	for (i = 0; i < MC_TABSIZE; i++)
2239 		CSR_WRITE(sc, DGE_MTA + (i << 2), 0);
2240 
2241 	ETHER_FIRST_MULTI(step, ec, enm);
2242 	while (enm != NULL) {
2243 		if (memcmp(enm->enm_addrlo, enm->enm_addrhi, ETHER_ADDR_LEN)) {
2244 			/*
2245 			 * We must listen to a range of multicast addresses.
2246 			 * For now, just accept all multicasts, rather than
2247 			 * trying to set only those filter bits needed to match
2248 			 * the range.  (At this time, the only use of address
2249 			 * ranges is for IP multicast routing, for which the
2250 			 * range is big enough to require all bits set.)
2251 			 */
2252 			goto allmulti;
2253 		}
2254 
2255 		hash = dge_mchash(sc, enm->enm_addrlo);
2256 
2257 		reg = (hash >> 5) & 0x7f;
2258 		bit = hash & 0x1f;
2259 
2260 		hash = CSR_READ(sc, DGE_MTA + (reg << 2));
2261 		hash |= 1U << bit;
2262 
2263 		CSR_WRITE(sc, DGE_MTA + (reg << 2), hash);
2264 
2265 		ETHER_NEXT_MULTI(step, enm);
2266 	}
2267 
2268 	ifp->if_flags &= ~IFF_ALLMULTI;
2269 	goto setit;
2270 
2271  allmulti:
2272 	ifp->if_flags |= IFF_ALLMULTI;
2273 	sc->sc_rctl |= RCTL_MPE;
2274 
2275  setit:
2276 	CSR_WRITE(sc, DGE_RCTL, sc->sc_rctl);
2277 }
2278 
2279 /*
2280  * Read in the EEPROM info and verify checksum.
2281  */
2282 int
2283 dge_read_eeprom(struct dge_softc *sc)
2284 {
2285 	uint16_t cksum;
2286 	int i;
2287 
2288 	cksum = 0;
2289 	for (i = 0; i < EEPROM_SIZE; i++) {
2290 		sc->sc_eeprom[i] = dge_eeprom_word(sc, i);
2291 		cksum += sc->sc_eeprom[i];
2292 	}
2293 	return cksum != EEPROM_CKSUM;
2294 }
2295 
2296 
2297 /*
2298  * Read a 16-bit word from address addr in the serial EEPROM.
2299  */
2300 uint16_t
2301 dge_eeprom_word(struct dge_softc *sc, int addr)
2302 {
2303 	uint32_t reg;
2304 	uint16_t rval = 0;
2305 	int i;
2306 
2307 	reg = CSR_READ(sc, DGE_EECD) & ~(EECD_SK|EECD_DI|EECD_CS);
2308 
2309 	/* Lower clock pulse (and data in to chip) */
2310 	CSR_WRITE(sc, DGE_EECD, reg);
2311 	/* Select chip */
2312 	CSR_WRITE(sc, DGE_EECD, reg|EECD_CS);
2313 
2314 	/* Send read command */
2315 	dge_eeprom_clockout(sc, 1);
2316 	dge_eeprom_clockout(sc, 1);
2317 	dge_eeprom_clockout(sc, 0);
2318 
2319 	/* Send address */
2320 	for (i = 5; i >= 0; i--)
2321 		dge_eeprom_clockout(sc, (addr >> i) & 1);
2322 
2323 	/* Read data */
2324 	for (i = 0; i < 16; i++) {
2325 		rval <<= 1;
2326 		rval |= dge_eeprom_clockin(sc);
2327 	}
2328 
2329 	/* Deselect chip */
2330 	CSR_WRITE(sc, DGE_EECD, reg);
2331 
2332 	return rval;
2333 }
2334 
2335 /*
2336  * Clock out a single bit to the EEPROM.
2337  */
2338 void
2339 dge_eeprom_clockout(struct dge_softc *sc, int bit)
2340 {
2341 	int reg;
2342 
2343 	reg = CSR_READ(sc, DGE_EECD) & ~(EECD_DI|EECD_SK);
2344 	if (bit)
2345 		reg |= EECD_DI;
2346 
2347 	CSR_WRITE(sc, DGE_EECD, reg);
2348 	delay(2);
2349 	CSR_WRITE(sc, DGE_EECD, reg|EECD_SK);
2350 	delay(2);
2351 	CSR_WRITE(sc, DGE_EECD, reg);
2352 	delay(2);
2353 }
2354 
2355 /*
2356  * Clock in a single bit from EEPROM.
2357  */
2358 int
2359 dge_eeprom_clockin(struct dge_softc *sc)
2360 {
2361 	int reg, rv;
2362 
2363 	reg = CSR_READ(sc, DGE_EECD) & ~(EECD_DI|EECD_DO|EECD_SK);
2364 
2365 	CSR_WRITE(sc, DGE_EECD, reg|EECD_SK); /* Raise clock */
2366 	delay(2);
2367 	rv = (CSR_READ(sc, DGE_EECD) & EECD_DO) != 0; /* Get bit */
2368 	CSR_WRITE(sc, DGE_EECD, reg); /* Lower clock */
2369 	delay(2);
2370 
2371 	return rv;
2372 }
2373 
2374 static void
2375 dge_xgmii_mediastatus(struct ifnet *ifp, struct ifmediareq *ifmr)
2376 {
2377 	struct dge_softc *sc = ifp->if_softc;
2378 
2379 	ifmr->ifm_status = IFM_AVALID;
2380 	ifmr->ifm_active = IFM_ETHER|IFM_10G_LR;
2381 
2382 	if (CSR_READ(sc, DGE_STATUS) & STATUS_LINKUP)
2383 		ifmr->ifm_status |= IFM_ACTIVE;
2384 }
2385 
2386 static inline int
2387 phwait(struct dge_softc *sc, int p, int r, int d, int type)
2388 {
2389         int i, mdic;
2390 
2391         CSR_WRITE(sc, DGE_MDIO,
2392 	    MDIO_PHY(p) | MDIO_REG(r) | MDIO_DEV(d) | type | MDIO_CMD);
2393         for (i = 0; i < 10; i++) {
2394                 delay(10);
2395                 if (((mdic = CSR_READ(sc, DGE_MDIO)) & MDIO_CMD) == 0)
2396                         break;
2397         }
2398         return mdic;
2399 }
2400 
2401 
2402 static void
2403 dge_xgmii_writereg(struct device *self, int phy, int reg, int val)
2404 {
2405 	struct dge_softc *sc = (void *) self;
2406 	int mdic;
2407 
2408 	CSR_WRITE(sc, DGE_MDIRW, val);
2409 	if (((mdic = phwait(sc, phy, reg, 1, MDIO_ADDR)) & MDIO_CMD)) {
2410 		printf("%s: address cycle timeout; phy %d reg %d\n",
2411 		    sc->sc_dev.dv_xname, phy, reg);
2412 		return;
2413 	}
2414 	if (((mdic = phwait(sc, phy, reg, 1, MDIO_WRITE)) & MDIO_CMD)) {
2415 		printf("%s: read cycle timeout; phy %d reg %d\n",
2416 		    sc->sc_dev.dv_xname, phy, reg);
2417 		return;
2418 	}
2419 }
2420 
2421 static void
2422 dge_xgmii_reset(struct dge_softc *sc)
2423 {
2424 	dge_xgmii_writereg((void *)sc, 0, 0, BMCR_RESET);
2425 }
2426 
2427 static int
2428 dge_xgmii_mediachange(struct ifnet *ifp)
2429 {
2430 	return 0;
2431 }
2432