xref: /netbsd-src/sys/dev/pci/if_dge.c (revision 3816d47b2c42fcd6e549e3407f842a5b1a1d23ad)
1 /*	$NetBSD: if_dge.c,v 1.28 2009/11/26 15:17:09 njoly 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 1.28 2009/11/26 15:17:09 njoly 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 
262 	int sc_flags;			/* flags; see below */
263 	int sc_bus_speed;		/* PCI/PCIX bus speed */
264 	int sc_pcix_offset;		/* PCIX capability register offset */
265 
266 	pci_chipset_tag_t sc_pc;
267 	pcitag_t sc_pt;
268 	int sc_mmrbc;			/* Max PCIX memory read byte count */
269 
270 	void *sc_ih;			/* interrupt cookie */
271 
272 	struct ifmedia sc_media;
273 
274 	bus_dmamap_t sc_cddmamap;	/* control data DMA map */
275 #define sc_cddma	sc_cddmamap->dm_segs[0].ds_addr
276 
277 	int		sc_align_tweak;
278 
279 	/*
280 	 * Software state for the transmit and receive descriptors.
281 	 */
282 	struct dge_txsoft sc_txsoft[DGE_TXQUEUELEN];
283 	struct dge_rxsoft sc_rxsoft[DGE_NRXDESC];
284 
285 	/*
286 	 * Control data structures.
287 	 */
288 	struct dge_control_data *sc_control_data;
289 #define sc_txdescs	sc_control_data->wcd_txdescs
290 #define sc_rxdescs	sc_control_data->wcd_rxdescs
291 
292 #ifdef DGE_EVENT_COUNTERS
293 	/* Event counters. */
294 	struct evcnt sc_ev_txsstall;	/* Tx stalled due to no txs */
295 	struct evcnt sc_ev_txdstall;	/* Tx stalled due to no txd */
296 	struct evcnt sc_ev_txforceintr; /* Tx interrupts forced */
297 	struct evcnt sc_ev_txdw;	/* Tx descriptor interrupts */
298 	struct evcnt sc_ev_txqe;	/* Tx queue empty interrupts */
299 	struct evcnt sc_ev_rxintr;	/* Rx interrupts */
300 	struct evcnt sc_ev_linkintr;	/* Link interrupts */
301 
302 	struct evcnt sc_ev_rxipsum;	/* IP checksums checked in-bound */
303 	struct evcnt sc_ev_rxtusum;	/* TCP/UDP cksums checked in-bound */
304 	struct evcnt sc_ev_txipsum;	/* IP checksums comp. out-bound */
305 	struct evcnt sc_ev_txtusum;	/* TCP/UDP cksums comp. out-bound */
306 
307 	struct evcnt sc_ev_txctx_init;	/* Tx cksum context cache initialized */
308 	struct evcnt sc_ev_txctx_hit;	/* Tx cksum context cache hit */
309 	struct evcnt sc_ev_txctx_miss;	/* Tx cksum context cache miss */
310 
311 	struct evcnt sc_ev_txseg[DGE_NTXSEGS]; /* Tx packets w/ N segments */
312 	struct evcnt sc_ev_txdrop;	/* Tx packets dropped (too many segs) */
313 #endif /* DGE_EVENT_COUNTERS */
314 
315 	int	sc_txfree;		/* number of free Tx descriptors */
316 	int	sc_txnext;		/* next ready Tx descriptor */
317 
318 	int	sc_txsfree;		/* number of free Tx jobs */
319 	int	sc_txsnext;		/* next free Tx job */
320 	int	sc_txsdirty;		/* dirty Tx jobs */
321 
322 	uint32_t sc_txctx_ipcs;		/* cached Tx IP cksum ctx */
323 	uint32_t sc_txctx_tucs;		/* cached Tx TCP/UDP cksum ctx */
324 
325 	int	sc_rxptr;		/* next ready Rx descriptor/queue ent */
326 	int	sc_rxdiscard;
327 	int	sc_rxlen;
328 	struct mbuf *sc_rxhead;
329 	struct mbuf *sc_rxtail;
330 	struct mbuf **sc_rxtailp;
331 
332 	uint32_t sc_ctrl0;		/* prototype CTRL0 register */
333 	uint32_t sc_icr;		/* prototype interrupt bits */
334 	uint32_t sc_tctl;		/* prototype TCTL register */
335 	uint32_t sc_rctl;		/* prototype RCTL register */
336 
337 	int sc_mchash_type;		/* multicast filter offset */
338 
339 	uint16_t sc_eeprom[EEPROM_SIZE];
340 
341 #if NRND > 0
342 	rndsource_element_t rnd_source; /* random source */
343 #endif
344 #ifdef DGE_OFFBYONE_RXBUG
345 	void *sc_bugbuf;
346 	SLIST_HEAD(, rxbugentry) sc_buglist;
347 	bus_dmamap_t sc_bugmap;
348 	struct rxbugentry *sc_entry;
349 #endif
350 };
351 
352 #define DGE_RXCHAIN_RESET(sc)						\
353 do {									\
354 	(sc)->sc_rxtailp = &(sc)->sc_rxhead;				\
355 	*(sc)->sc_rxtailp = NULL;					\
356 	(sc)->sc_rxlen = 0;						\
357 } while (/*CONSTCOND*/0)
358 
359 #define DGE_RXCHAIN_LINK(sc, m)						\
360 do {									\
361 	*(sc)->sc_rxtailp = (sc)->sc_rxtail = (m);			\
362 	(sc)->sc_rxtailp = &(m)->m_next;				\
363 } while (/*CONSTCOND*/0)
364 
365 /* sc_flags */
366 #define DGE_F_BUS64		0x20	/* bus is 64-bit */
367 #define DGE_F_PCIX		0x40	/* bus is PCI-X */
368 
369 #ifdef DGE_EVENT_COUNTERS
370 #define DGE_EVCNT_INCR(ev)	(ev)->ev_count++
371 #else
372 #define DGE_EVCNT_INCR(ev)	/* nothing */
373 #endif
374 
375 #define CSR_READ(sc, reg)						\
376 	bus_space_read_4((sc)->sc_st, (sc)->sc_sh, (reg))
377 #define CSR_WRITE(sc, reg, val)						\
378 	bus_space_write_4((sc)->sc_st, (sc)->sc_sh, (reg), (val))
379 
380 #define DGE_CDTXADDR(sc, x)	((sc)->sc_cddma + DGE_CDTXOFF((x)))
381 #define DGE_CDRXADDR(sc, x)	((sc)->sc_cddma + DGE_CDRXOFF((x)))
382 
383 #define DGE_CDTXSYNC(sc, x, n, ops)					\
384 do {									\
385 	int __x, __n;							\
386 									\
387 	__x = (x);							\
388 	__n = (n);							\
389 									\
390 	/* If it will wrap around, sync to the end of the ring. */	\
391 	if ((__x + __n) > DGE_NTXDESC) {				\
392 		bus_dmamap_sync((sc)->sc_dmat, (sc)->sc_cddmamap,	\
393 		    DGE_CDTXOFF(__x), sizeof(struct dge_tdes) *		\
394 		    (DGE_NTXDESC - __x), (ops));			\
395 		__n -= (DGE_NTXDESC - __x);				\
396 		__x = 0;						\
397 	}								\
398 									\
399 	/* Now sync whatever is left. */				\
400 	bus_dmamap_sync((sc)->sc_dmat, (sc)->sc_cddmamap,		\
401 	    DGE_CDTXOFF(__x), sizeof(struct dge_tdes) * __n, (ops));	\
402 } while (/*CONSTCOND*/0)
403 
404 #define DGE_CDRXSYNC(sc, x, ops)						\
405 do {									\
406 	bus_dmamap_sync((sc)->sc_dmat, (sc)->sc_cddmamap,		\
407 	   DGE_CDRXOFF((x)), sizeof(struct dge_rdes), (ops));		\
408 } while (/*CONSTCOND*/0)
409 
410 #ifdef DGE_OFFBYONE_RXBUG
411 #define DGE_INIT_RXDESC(sc, x)						\
412 do {									\
413 	struct dge_rxsoft *__rxs = &(sc)->sc_rxsoft[(x)];		\
414 	struct dge_rdes *__rxd = &(sc)->sc_rxdescs[(x)];		\
415 	struct mbuf *__m = __rxs->rxs_mbuf;				\
416 									\
417 	__rxd->dr_baddrl = htole32(sc->sc_bugmap->dm_segs[0].ds_addr +	\
418 	    (mtod((__m), char *) - (char *)sc->sc_bugbuf));		\
419 	__rxd->dr_baddrh = 0;						\
420 	__rxd->dr_len = 0;						\
421 	__rxd->dr_cksum = 0;						\
422 	__rxd->dr_status = 0;						\
423 	__rxd->dr_errors = 0;						\
424 	__rxd->dr_special = 0;						\
425 	DGE_CDRXSYNC((sc), (x), BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE); \
426 									\
427 	CSR_WRITE((sc), DGE_RDT, (x));					\
428 } while (/*CONSTCOND*/0)
429 #else
430 #define DGE_INIT_RXDESC(sc, x)						\
431 do {									\
432 	struct dge_rxsoft *__rxs = &(sc)->sc_rxsoft[(x)];		\
433 	struct dge_rdes *__rxd = &(sc)->sc_rxdescs[(x)];		\
434 	struct mbuf *__m = __rxs->rxs_mbuf;				\
435 									\
436 	/*								\
437 	 * Note: We scoot the packet forward 2 bytes in the buffer	\
438 	 * so that the payload after the Ethernet header is aligned	\
439 	 * to a 4-byte boundary.					\
440 	 *								\
441 	 * XXX BRAINDAMAGE ALERT!					\
442 	 * The stupid chip uses the same size for every buffer, which	\
443 	 * is set in the Receive Control register.  We are using the 2K \
444 	 * size option, but what we REALLY want is (2K - 2)!  For this	\
445 	 * reason, we can't "scoot" packets longer than the standard	\
446 	 * Ethernet MTU.  On strict-alignment platforms, if the total	\
447 	 * size exceeds (2K - 2) we set align_tweak to 0 and let	\
448 	 * the upper layer copy the headers.				\
449 	 */								\
450 	__m->m_data = __m->m_ext.ext_buf + (sc)->sc_align_tweak;	\
451 									\
452 	__rxd->dr_baddrl =						\
453 	    htole32(__rxs->rxs_dmamap->dm_segs[0].ds_addr +		\
454 		(sc)->sc_align_tweak);					\
455 	__rxd->dr_baddrh = 0;						\
456 	__rxd->dr_len = 0;						\
457 	__rxd->dr_cksum = 0;						\
458 	__rxd->dr_status = 0;						\
459 	__rxd->dr_errors = 0;						\
460 	__rxd->dr_special = 0;						\
461 	DGE_CDRXSYNC((sc), (x), BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE); \
462 									\
463 	CSR_WRITE((sc), DGE_RDT, (x));					\
464 } while (/*CONSTCOND*/0)
465 #endif
466 
467 #ifdef DGE_OFFBYONE_RXBUG
468 /*
469  * Allocation constants.  Much memory may be used for this.
470  */
471 #ifndef DGE_BUFFER_SIZE
472 #define DGE_BUFFER_SIZE DGE_MAX_MTU
473 #endif
474 #define DGE_NBUFFERS	(4*DGE_NRXDESC)
475 #define DGE_RXMEM	(DGE_NBUFFERS*DGE_BUFFER_SIZE)
476 
477 struct rxbugentry {
478 	SLIST_ENTRY(rxbugentry) rb_entry;
479 	int rb_slot;
480 };
481 
482 static int
483 dge_alloc_rcvmem(struct dge_softc *sc)
484 {
485 	char *ptr, *kva;
486 	bus_dma_segment_t seg;
487 	int i, rseg, state, error;
488 	struct rxbugentry *entry;
489 
490 	state = error = 0;
491 
492 	if (bus_dmamem_alloc(sc->sc_dmat, DGE_RXMEM, PAGE_SIZE, 0,
493 	     &seg, 1, &rseg, BUS_DMA_NOWAIT)) {
494 		aprint_error_dev(&sc->sc_dev, "can't alloc rx buffers\n");
495 		return ENOBUFS;
496 	}
497 
498 	state = 1;
499 	if (bus_dmamem_map(sc->sc_dmat, &seg, rseg, DGE_RXMEM, (void **)&kva,
500 	    BUS_DMA_NOWAIT)) {
501 		aprint_error_dev(&sc->sc_dev, "can't map DMA buffers (%d bytes)\n",
502 		    (int)DGE_RXMEM);
503 		error = ENOBUFS;
504 		goto out;
505 	}
506 
507 	state = 2;
508 	if (bus_dmamap_create(sc->sc_dmat, DGE_RXMEM, 1, DGE_RXMEM, 0,
509 	    BUS_DMA_NOWAIT, &sc->sc_bugmap)) {
510 		aprint_error_dev(&sc->sc_dev, "can't create DMA map\n");
511 		error = ENOBUFS;
512 		goto out;
513 	}
514 
515 	state = 3;
516 	if (bus_dmamap_load(sc->sc_dmat, sc->sc_bugmap,
517 	    kva, DGE_RXMEM, NULL, BUS_DMA_NOWAIT)) {
518 		aprint_error_dev(&sc->sc_dev, "can't load DMA map\n");
519 		error = ENOBUFS;
520 		goto out;
521 	}
522 
523 	state = 4;
524 	sc->sc_bugbuf = (void *)kva;
525 	SLIST_INIT(&sc->sc_buglist);
526 
527 	/*
528 	 * Now divide it up into DGE_BUFFER_SIZE pieces and save the addresses
529 	 * in an array.
530 	 */
531 	ptr = sc->sc_bugbuf;
532 	if ((entry = malloc(sizeof(*entry) * DGE_NBUFFERS,
533 	    M_DEVBUF, M_NOWAIT)) == NULL) {
534 		error = ENOBUFS;
535 		goto out;
536 	}
537 	sc->sc_entry = entry;
538 	for (i = 0; i < DGE_NBUFFERS; i++) {
539 		entry[i].rb_slot = i;
540 		SLIST_INSERT_HEAD(&sc->sc_buglist, &entry[i], rb_entry);
541 	}
542 out:
543 	if (error != 0) {
544 		switch (state) {
545 		case 4:
546 			bus_dmamap_unload(sc->sc_dmat, sc->sc_bugmap);
547 		case 3:
548 			bus_dmamap_destroy(sc->sc_dmat, sc->sc_bugmap);
549 		case 2:
550 			bus_dmamem_unmap(sc->sc_dmat, kva, DGE_RXMEM);
551 		case 1:
552 			bus_dmamem_free(sc->sc_dmat, &seg, rseg);
553 			break;
554 		default:
555 			break;
556 		}
557 	}
558 
559 	return error;
560 }
561 
562 /*
563  * Allocate a jumbo buffer.
564  */
565 static void *
566 dge_getbuf(struct dge_softc *sc)
567 {
568 	struct rxbugentry *entry;
569 
570 	entry = SLIST_FIRST(&sc->sc_buglist);
571 
572 	if (entry == NULL) {
573 		printf("%s: no free RX buffers\n", device_xname(&sc->sc_dev));
574 		return(NULL);
575 	}
576 
577 	SLIST_REMOVE_HEAD(&sc->sc_buglist, rb_entry);
578 	return (char *)sc->sc_bugbuf + entry->rb_slot * DGE_BUFFER_SIZE;
579 }
580 
581 /*
582  * Release a jumbo buffer.
583  */
584 static void
585 dge_freebuf(struct mbuf *m, void *buf, size_t size, void *arg)
586 {
587 	struct rxbugentry *entry;
588 	struct dge_softc *sc;
589 	int i, s;
590 
591 	/* Extract the softc struct pointer. */
592 	sc = (struct dge_softc *)arg;
593 
594 	if (sc == NULL)
595 		panic("dge_freebuf: can't find softc pointer!");
596 
597 	/* calculate the slot this buffer belongs to */
598 
599 	i = ((char *)buf - (char *)sc->sc_bugbuf) / DGE_BUFFER_SIZE;
600 
601 	if ((i < 0) || (i >= DGE_NBUFFERS))
602 		panic("dge_freebuf: asked to free buffer %d!", i);
603 
604 	s = splvm();
605 	entry = sc->sc_entry + i;
606 	SLIST_INSERT_HEAD(&sc->sc_buglist, entry, rb_entry);
607 
608 	if (__predict_true(m != NULL))
609 		pool_cache_put(mb_cache, m);
610 	splx(s);
611 }
612 #endif
613 
614 static void	dge_start(struct ifnet *);
615 static void	dge_watchdog(struct ifnet *);
616 static int	dge_ioctl(struct ifnet *, u_long, void *);
617 static int	dge_init(struct ifnet *);
618 static void	dge_stop(struct ifnet *, int);
619 
620 static bool	dge_shutdown(device_t, int);
621 
622 static void	dge_reset(struct dge_softc *);
623 static void	dge_rxdrain(struct dge_softc *);
624 static int	dge_add_rxbuf(struct dge_softc *, int);
625 
626 static void	dge_set_filter(struct dge_softc *);
627 
628 static int	dge_intr(void *);
629 static void	dge_txintr(struct dge_softc *);
630 static void	dge_rxintr(struct dge_softc *);
631 static void	dge_linkintr(struct dge_softc *, uint32_t);
632 
633 static int	dge_match(device_t, cfdata_t, void *);
634 static void	dge_attach(device_t, device_t, void *);
635 
636 static int	dge_read_eeprom(struct dge_softc *sc);
637 static int	dge_eeprom_clockin(struct dge_softc *sc);
638 static void	dge_eeprom_clockout(struct dge_softc *sc, int bit);
639 static uint16_t	dge_eeprom_word(struct dge_softc *sc, int addr);
640 static int	dge_xgmii_mediachange(struct ifnet *);
641 static void	dge_xgmii_mediastatus(struct ifnet *, struct ifmediareq *);
642 static void	dge_xgmii_reset(struct dge_softc *);
643 static void	dge_xgmii_writereg(device_t, int, int, int);
644 
645 
646 CFATTACH_DECL(dge, sizeof(struct dge_softc),
647     dge_match, dge_attach, NULL, NULL);
648 
649 #ifdef DGE_EVENT_COUNTERS
650 #if DGE_NTXSEGS > 100
651 #error Update dge_txseg_evcnt_names
652 #endif
653 static char (*dge_txseg_evcnt_names)[DGE_NTXSEGS][8 /* "txseg00" + \0 */];
654 #endif /* DGE_EVENT_COUNTERS */
655 
656 static int
657 dge_match(device_t parent, cfdata_t cf, void *aux)
658 {
659 	struct pci_attach_args *pa = aux;
660 
661 	if (PCI_VENDOR(pa->pa_id) == PCI_VENDOR_INTEL &&
662 	    PCI_PRODUCT(pa->pa_id) == PCI_PRODUCT_INTEL_82597EX)
663 		return (1);
664 
665 	return (0);
666 }
667 
668 static void
669 dge_attach(device_t parent, device_t self, void *aux)
670 {
671 	struct dge_softc *sc = device_private(self);
672 	struct pci_attach_args *pa = aux;
673 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
674 	pci_chipset_tag_t pc = pa->pa_pc;
675 	pci_intr_handle_t ih;
676 	const char *intrstr = NULL;
677 	bus_dma_segment_t seg;
678 	int i, rseg, error;
679 	uint8_t enaddr[ETHER_ADDR_LEN];
680 	pcireg_t preg, memtype;
681 	uint32_t reg;
682 
683 	sc->sc_dmat = pa->pa_dmat;
684 	sc->sc_pc = pa->pa_pc;
685 	sc->sc_pt = pa->pa_tag;
686 
687 	preg = PCI_REVISION(pci_conf_read(pc, pa->pa_tag, PCI_CLASS_REG));
688 	aprint_naive(": Ethernet controller\n");
689 	aprint_normal(": Intel i82597EX 10GbE-LR Ethernet, rev. %d\n", preg);
690 
691 	memtype = pci_mapreg_type(pa->pa_pc, pa->pa_tag, DGE_PCI_BAR);
692         if (pci_mapreg_map(pa, DGE_PCI_BAR, memtype, 0,
693             &sc->sc_st, &sc->sc_sh, NULL, NULL)) {
694                 aprint_error_dev(&sc->sc_dev, "unable to map device registers\n");
695                 return;
696         }
697 
698 	/* Enable bus mastering */
699 	preg = pci_conf_read(pc, pa->pa_tag, PCI_COMMAND_STATUS_REG);
700 	preg |= PCI_COMMAND_MASTER_ENABLE;
701 	pci_conf_write(pc, pa->pa_tag, PCI_COMMAND_STATUS_REG, preg);
702 
703 	/*
704 	 * Map and establish our interrupt.
705 	 */
706 	if (pci_intr_map(pa, &ih)) {
707 		aprint_error_dev(&sc->sc_dev, "unable to map interrupt\n");
708 		return;
709 	}
710 	intrstr = pci_intr_string(pc, ih);
711 	sc->sc_ih = pci_intr_establish(pc, ih, IPL_NET, dge_intr, sc);
712 	if (sc->sc_ih == NULL) {
713 		aprint_error_dev(&sc->sc_dev, "unable to establish interrupt");
714 		if (intrstr != NULL)
715 			aprint_error(" at %s", intrstr);
716 		aprint_error("\n");
717 		return;
718 	}
719 	aprint_normal_dev(&sc->sc_dev, "interrupting at %s\n", intrstr);
720 
721 	/*
722 	 * Determine a few things about the bus we're connected to.
723 	 */
724 	reg = CSR_READ(sc, DGE_STATUS);
725 	if (reg & STATUS_BUS64)
726 		sc->sc_flags |= DGE_F_BUS64;
727 
728 	sc->sc_flags |= DGE_F_PCIX;
729 	if (pci_get_capability(pa->pa_pc, pa->pa_tag,
730 			       PCI_CAP_PCIX,
731 			       &sc->sc_pcix_offset, NULL) == 0)
732 		aprint_error_dev(&sc->sc_dev, "unable to find PCIX "
733 		    "capability\n");
734 
735 	if (sc->sc_flags & DGE_F_PCIX) {
736 		switch (reg & STATUS_PCIX_MSK) {
737 		case STATUS_PCIX_66:
738 			sc->sc_bus_speed = 66;
739 			break;
740 		case STATUS_PCIX_100:
741 			sc->sc_bus_speed = 100;
742 			break;
743 		case STATUS_PCIX_133:
744 			sc->sc_bus_speed = 133;
745 			break;
746 		default:
747 			aprint_error_dev(&sc->sc_dev,
748 			    "unknown PCIXSPD %d; assuming 66MHz\n",
749 			    reg & STATUS_PCIX_MSK);
750 			sc->sc_bus_speed = 66;
751 		}
752 	} else
753 		sc->sc_bus_speed = (reg & STATUS_BUS64) ? 66 : 33;
754 	aprint_verbose_dev(&sc->sc_dev, "%d-bit %dMHz %s bus\n",
755 	    (sc->sc_flags & DGE_F_BUS64) ? 64 : 32, sc->sc_bus_speed,
756 	    (sc->sc_flags & DGE_F_PCIX) ? "PCIX" : "PCI");
757 
758 	/*
759 	 * Allocate the control data structures, and create and load the
760 	 * DMA map for it.
761 	 */
762 	if ((error = bus_dmamem_alloc(sc->sc_dmat,
763 	    sizeof(struct dge_control_data), PAGE_SIZE, 0, &seg, 1, &rseg,
764 	    0)) != 0) {
765 		aprint_error_dev(&sc->sc_dev,
766 		    "unable to allocate control data, error = %d\n",
767 		    error);
768 		goto fail_0;
769 	}
770 
771 	if ((error = bus_dmamem_map(sc->sc_dmat, &seg, rseg,
772 	    sizeof(struct dge_control_data), (void **)&sc->sc_control_data,
773 	    0)) != 0) {
774 		aprint_error_dev(&sc->sc_dev, "unable to map control data, error = %d\n",
775 		    error);
776 		goto fail_1;
777 	}
778 
779 	if ((error = bus_dmamap_create(sc->sc_dmat,
780 	    sizeof(struct dge_control_data), 1,
781 	    sizeof(struct dge_control_data), 0, 0, &sc->sc_cddmamap)) != 0) {
782 		aprint_error_dev(&sc->sc_dev, "unable to create control data DMA map, "
783 		    "error = %d\n", error);
784 		goto fail_2;
785 	}
786 
787 	if ((error = bus_dmamap_load(sc->sc_dmat, sc->sc_cddmamap,
788 	    sc->sc_control_data, sizeof(struct dge_control_data), NULL,
789 	    0)) != 0) {
790 		aprint_error_dev(&sc->sc_dev,
791 		    "unable to load control data DMA map, error = %d\n",
792 		    error);
793 		goto fail_3;
794 	}
795 
796 #ifdef DGE_OFFBYONE_RXBUG
797 	if (dge_alloc_rcvmem(sc) != 0)
798 		return; /* Already complained */
799 #endif
800 	/*
801 	 * Create the transmit buffer DMA maps.
802 	 */
803 	for (i = 0; i < DGE_TXQUEUELEN; i++) {
804 		if ((error = bus_dmamap_create(sc->sc_dmat, DGE_MAX_MTU,
805 		    DGE_NTXSEGS, MCLBYTES, 0, 0,
806 		    &sc->sc_txsoft[i].txs_dmamap)) != 0) {
807 			aprint_error_dev(&sc->sc_dev, "unable to create Tx DMA map %d, "
808 			    "error = %d\n", i, error);
809 			goto fail_4;
810 		}
811 	}
812 
813 	/*
814 	 * Create the receive buffer DMA maps.
815 	 */
816 	for (i = 0; i < DGE_NRXDESC; i++) {
817 #ifdef DGE_OFFBYONE_RXBUG
818 		if ((error = bus_dmamap_create(sc->sc_dmat, DGE_BUFFER_SIZE, 1,
819 		    DGE_BUFFER_SIZE, 0, 0, &sc->sc_rxsoft[i].rxs_dmamap)) != 0) {
820 #else
821 		if ((error = bus_dmamap_create(sc->sc_dmat, MCLBYTES, 1,
822 		    MCLBYTES, 0, 0, &sc->sc_rxsoft[i].rxs_dmamap)) != 0) {
823 #endif
824 			aprint_error_dev(&sc->sc_dev, "unable to create Rx DMA map %d, "
825 			    "error = %d\n", i, error);
826 			goto fail_5;
827 		}
828 		sc->sc_rxsoft[i].rxs_mbuf = NULL;
829 	}
830 
831 	/*
832 	 * Set bits in ctrl0 register.
833 	 * Should get the software defined pins out of EEPROM?
834 	 */
835 	sc->sc_ctrl0 |= CTRL0_RPE | CTRL0_TPE; /* XON/XOFF */
836 	sc->sc_ctrl0 |= CTRL0_SDP3_DIR | CTRL0_SDP2_DIR | CTRL0_SDP1_DIR |
837 	    CTRL0_SDP0_DIR | CTRL0_SDP3 | CTRL0_SDP2 | CTRL0_SDP0;
838 
839 	/*
840 	 * Reset the chip to a known state.
841 	 */
842 	dge_reset(sc);
843 
844 	/*
845 	 * Reset the PHY.
846 	 */
847 	dge_xgmii_reset(sc);
848 
849 	/*
850 	 * Read in EEPROM data.
851 	 */
852 	if (dge_read_eeprom(sc)) {
853 		aprint_error_dev(&sc->sc_dev, "couldn't read EEPROM\n");
854 		return;
855 	}
856 
857 	/*
858 	 * Get the ethernet address.
859 	 */
860 	enaddr[0] = sc->sc_eeprom[EE_ADDR01] & 0377;
861 	enaddr[1] = sc->sc_eeprom[EE_ADDR01] >> 8;
862 	enaddr[2] = sc->sc_eeprom[EE_ADDR23] & 0377;
863 	enaddr[3] = sc->sc_eeprom[EE_ADDR23] >> 8;
864 	enaddr[4] = sc->sc_eeprom[EE_ADDR45] & 0377;
865 	enaddr[5] = sc->sc_eeprom[EE_ADDR45] >> 8;
866 
867 	aprint_normal_dev(&sc->sc_dev, "Ethernet address %s\n",
868 	    ether_sprintf(enaddr));
869 
870 	/*
871 	 * Setup media stuff.
872 	 */
873         ifmedia_init(&sc->sc_media, IFM_IMASK, dge_xgmii_mediachange,
874             dge_xgmii_mediastatus);
875         ifmedia_add(&sc->sc_media, IFM_ETHER|IFM_10G_LR, 0, NULL);
876         ifmedia_set(&sc->sc_media, IFM_ETHER|IFM_10G_LR);
877 
878 	ifp = &sc->sc_ethercom.ec_if;
879 	strlcpy(ifp->if_xname, device_xname(&sc->sc_dev), IFNAMSIZ);
880 	ifp->if_softc = sc;
881 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
882 	ifp->if_ioctl = dge_ioctl;
883 	ifp->if_start = dge_start;
884 	ifp->if_watchdog = dge_watchdog;
885 	ifp->if_init = dge_init;
886 	ifp->if_stop = dge_stop;
887 	IFQ_SET_MAXLEN(&ifp->if_snd, max(DGE_IFQUEUELEN, IFQ_MAXLEN));
888 	IFQ_SET_READY(&ifp->if_snd);
889 
890 	sc->sc_ethercom.ec_capabilities |=
891 	    ETHERCAP_JUMBO_MTU | ETHERCAP_VLAN_MTU;
892 
893 	/*
894 	 * We can perform TCPv4 and UDPv4 checkums in-bound.
895 	 */
896 	ifp->if_capabilities |=
897 	    IFCAP_CSUM_IPv4_Tx | IFCAP_CSUM_IPv4_Rx |
898 	    IFCAP_CSUM_TCPv4_Tx | IFCAP_CSUM_TCPv4_Rx |
899 	    IFCAP_CSUM_UDPv4_Tx | IFCAP_CSUM_UDPv4_Rx;
900 
901 	/*
902 	 * Attach the interface.
903 	 */
904 	if_attach(ifp);
905 	ether_ifattach(ifp, enaddr);
906 #if NRND > 0
907 	rnd_attach_source(&sc->rnd_source, device_xname(&sc->sc_dev),
908 	    RND_TYPE_NET, 0);
909 #endif
910 
911 #ifdef DGE_EVENT_COUNTERS
912 	/* Fix segment event naming */
913 	if (dge_txseg_evcnt_names == NULL) {
914 		dge_txseg_evcnt_names =
915 		    malloc(sizeof(*dge_txseg_evcnt_names), M_DEVBUF, M_WAITOK);
916 		for (i = 0; i < DGE_NTXSEGS; i++)
917 			snprintf((*dge_txseg_evcnt_names)[i],
918 			    sizeof((*dge_txseg_evcnt_names)[i]), "txseg%d", i);
919 	}
920 
921 	/* Attach event counters. */
922 	evcnt_attach_dynamic(&sc->sc_ev_txsstall, EVCNT_TYPE_MISC,
923 	    NULL, device_xname(&sc->sc_dev), "txsstall");
924 	evcnt_attach_dynamic(&sc->sc_ev_txdstall, EVCNT_TYPE_MISC,
925 	    NULL, device_xname(&sc->sc_dev), "txdstall");
926 	evcnt_attach_dynamic(&sc->sc_ev_txforceintr, EVCNT_TYPE_MISC,
927 	    NULL, device_xname(&sc->sc_dev), "txforceintr");
928 	evcnt_attach_dynamic(&sc->sc_ev_txdw, EVCNT_TYPE_INTR,
929 	    NULL, device_xname(&sc->sc_dev), "txdw");
930 	evcnt_attach_dynamic(&sc->sc_ev_txqe, EVCNT_TYPE_INTR,
931 	    NULL, device_xname(&sc->sc_dev), "txqe");
932 	evcnt_attach_dynamic(&sc->sc_ev_rxintr, EVCNT_TYPE_INTR,
933 	    NULL, device_xname(&sc->sc_dev), "rxintr");
934 	evcnt_attach_dynamic(&sc->sc_ev_linkintr, EVCNT_TYPE_INTR,
935 	    NULL, device_xname(&sc->sc_dev), "linkintr");
936 
937 	evcnt_attach_dynamic(&sc->sc_ev_rxipsum, EVCNT_TYPE_MISC,
938 	    NULL, device_xname(&sc->sc_dev), "rxipsum");
939 	evcnt_attach_dynamic(&sc->sc_ev_rxtusum, EVCNT_TYPE_MISC,
940 	    NULL, device_xname(&sc->sc_dev), "rxtusum");
941 	evcnt_attach_dynamic(&sc->sc_ev_txipsum, EVCNT_TYPE_MISC,
942 	    NULL, device_xname(&sc->sc_dev), "txipsum");
943 	evcnt_attach_dynamic(&sc->sc_ev_txtusum, EVCNT_TYPE_MISC,
944 	    NULL, device_xname(&sc->sc_dev), "txtusum");
945 
946 	evcnt_attach_dynamic(&sc->sc_ev_txctx_init, EVCNT_TYPE_MISC,
947 	    NULL, device_xname(&sc->sc_dev), "txctx init");
948 	evcnt_attach_dynamic(&sc->sc_ev_txctx_hit, EVCNT_TYPE_MISC,
949 	    NULL, device_xname(&sc->sc_dev), "txctx hit");
950 	evcnt_attach_dynamic(&sc->sc_ev_txctx_miss, EVCNT_TYPE_MISC,
951 	    NULL, device_xname(&sc->sc_dev), "txctx miss");
952 
953 	for (i = 0; i < DGE_NTXSEGS; i++)
954 		evcnt_attach_dynamic(&sc->sc_ev_txseg[i], EVCNT_TYPE_MISC,
955 		    NULL, device_xname(&sc->sc_dev), (*dge_txseg_evcnt_names)[i]);
956 
957 	evcnt_attach_dynamic(&sc->sc_ev_txdrop, EVCNT_TYPE_MISC,
958 	    NULL, device_xname(&sc->sc_dev), "txdrop");
959 
960 #endif /* DGE_EVENT_COUNTERS */
961 
962 	/*
963 	 * Make sure the interface is shutdown during reboot.
964 	 */
965 	if (pmf_device_register1(self, NULL, NULL, dge_shutdown))
966 		pmf_class_network_register(self, ifp);
967 	else
968 		aprint_error_dev(self, "couldn't establish power handler\n");
969 
970 	return;
971 
972 	/*
973 	 * Free any resources we've allocated during the failed attach
974 	 * attempt.  Do this in reverse order and fall through.
975 	 */
976  fail_5:
977 	for (i = 0; i < DGE_NRXDESC; i++) {
978 		if (sc->sc_rxsoft[i].rxs_dmamap != NULL)
979 			bus_dmamap_destroy(sc->sc_dmat,
980 			    sc->sc_rxsoft[i].rxs_dmamap);
981 	}
982  fail_4:
983 	for (i = 0; i < DGE_TXQUEUELEN; i++) {
984 		if (sc->sc_txsoft[i].txs_dmamap != NULL)
985 			bus_dmamap_destroy(sc->sc_dmat,
986 			    sc->sc_txsoft[i].txs_dmamap);
987 	}
988 	bus_dmamap_unload(sc->sc_dmat, sc->sc_cddmamap);
989  fail_3:
990 	bus_dmamap_destroy(sc->sc_dmat, sc->sc_cddmamap);
991  fail_2:
992 	bus_dmamem_unmap(sc->sc_dmat, (void *)sc->sc_control_data,
993 	    sizeof(struct dge_control_data));
994  fail_1:
995 	bus_dmamem_free(sc->sc_dmat, &seg, rseg);
996  fail_0:
997 	return;
998 }
999 
1000 /*
1001  * dge_shutdown:
1002  *
1003  *	Make sure the interface is stopped at reboot time.
1004  */
1005 static bool
1006 dge_shutdown(device_t self, int howto)
1007 {
1008 	struct dge_softc *sc;
1009 
1010 	sc = device_private(self);
1011 	dge_stop(&sc->sc_ethercom.ec_if, 1);
1012 
1013 	return true;
1014 }
1015 
1016 /*
1017  * dge_tx_cksum:
1018  *
1019  *	Set up TCP/IP checksumming parameters for the
1020  *	specified packet.
1021  */
1022 static int
1023 dge_tx_cksum(struct dge_softc *sc, struct dge_txsoft *txs, uint8_t *fieldsp)
1024 {
1025 	struct mbuf *m0 = txs->txs_mbuf;
1026 	struct dge_ctdes *t;
1027 	uint32_t ipcs, tucs;
1028 	struct ether_header *eh;
1029 	int offset, iphl;
1030 	uint8_t fields = 0;
1031 
1032 	/*
1033 	 * XXX It would be nice if the mbuf pkthdr had offset
1034 	 * fields for the protocol headers.
1035 	 */
1036 
1037 	eh = mtod(m0, struct ether_header *);
1038 	switch (htons(eh->ether_type)) {
1039 	case ETHERTYPE_IP:
1040 		offset = ETHER_HDR_LEN;
1041 		break;
1042 
1043 	case ETHERTYPE_VLAN:
1044 		offset = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
1045 		break;
1046 
1047 	default:
1048 		/*
1049 		 * Don't support this protocol or encapsulation.
1050 		 */
1051 		*fieldsp = 0;
1052 		return (0);
1053 	}
1054 
1055 	iphl = M_CSUM_DATA_IPv4_IPHL(m0->m_pkthdr.csum_data);
1056 
1057 	/*
1058 	 * NOTE: Even if we're not using the IP or TCP/UDP checksum
1059 	 * offload feature, if we load the context descriptor, we
1060 	 * MUST provide valid values for IPCSS and TUCSS fields.
1061 	 */
1062 
1063 	if (m0->m_pkthdr.csum_flags & M_CSUM_IPv4) {
1064 		DGE_EVCNT_INCR(&sc->sc_ev_txipsum);
1065 		fields |= TDESC_POPTS_IXSM;
1066 		ipcs = DGE_TCPIP_IPCSS(offset) |
1067 		    DGE_TCPIP_IPCSO(offset + offsetof(struct ip, ip_sum)) |
1068 		    DGE_TCPIP_IPCSE(offset + iphl - 1);
1069 	} else if (__predict_true(sc->sc_txctx_ipcs != 0xffffffff)) {
1070 		/* Use the cached value. */
1071 		ipcs = sc->sc_txctx_ipcs;
1072 	} else {
1073 		/* Just initialize it to the likely value anyway. */
1074 		ipcs = DGE_TCPIP_IPCSS(offset) |
1075 		    DGE_TCPIP_IPCSO(offset + offsetof(struct ip, ip_sum)) |
1076 		    DGE_TCPIP_IPCSE(offset + iphl - 1);
1077 	}
1078 	DPRINTF(DGE_DEBUG_CKSUM,
1079 	    ("%s: CKSUM: offset %d ipcs 0x%x\n",
1080 	    device_xname(&sc->sc_dev), offset, ipcs));
1081 
1082 	offset += iphl;
1083 
1084 	if (m0->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4)) {
1085 		DGE_EVCNT_INCR(&sc->sc_ev_txtusum);
1086 		fields |= TDESC_POPTS_TXSM;
1087 		tucs = DGE_TCPIP_TUCSS(offset) |
1088 		   DGE_TCPIP_TUCSO(offset + M_CSUM_DATA_IPv4_OFFSET(m0->m_pkthdr.csum_data)) |
1089 		   DGE_TCPIP_TUCSE(0) /* rest of packet */;
1090 	} else if (__predict_true(sc->sc_txctx_tucs != 0xffffffff)) {
1091 		/* Use the cached value. */
1092 		tucs = sc->sc_txctx_tucs;
1093 	} else {
1094 		/* Just initialize it to a valid TCP context. */
1095 		tucs = DGE_TCPIP_TUCSS(offset) |
1096 		    DGE_TCPIP_TUCSO(offset + offsetof(struct tcphdr, th_sum)) |
1097 		    DGE_TCPIP_TUCSE(0) /* rest of packet */;
1098 	}
1099 
1100 	DPRINTF(DGE_DEBUG_CKSUM,
1101 	    ("%s: CKSUM: offset %d tucs 0x%x\n",
1102 	    device_xname(&sc->sc_dev), offset, tucs));
1103 
1104 	if (sc->sc_txctx_ipcs == ipcs &&
1105 	    sc->sc_txctx_tucs == tucs) {
1106 		/* Cached context is fine. */
1107 		DGE_EVCNT_INCR(&sc->sc_ev_txctx_hit);
1108 	} else {
1109 		/* Fill in the context descriptor. */
1110 #ifdef DGE_EVENT_COUNTERS
1111 		if (sc->sc_txctx_ipcs == 0xffffffff &&
1112 		    sc->sc_txctx_tucs == 0xffffffff)
1113 			DGE_EVCNT_INCR(&sc->sc_ev_txctx_init);
1114 		else
1115 			DGE_EVCNT_INCR(&sc->sc_ev_txctx_miss);
1116 #endif
1117 		t = (struct dge_ctdes *)&sc->sc_txdescs[sc->sc_txnext];
1118 		t->dc_tcpip_ipcs = htole32(ipcs);
1119 		t->dc_tcpip_tucs = htole32(tucs);
1120 		t->dc_tcpip_cmdlen = htole32(TDESC_DTYP_CTD);
1121 		t->dc_tcpip_seg = 0;
1122 		DGE_CDTXSYNC(sc, sc->sc_txnext, 1, BUS_DMASYNC_PREWRITE);
1123 
1124 		sc->sc_txctx_ipcs = ipcs;
1125 		sc->sc_txctx_tucs = tucs;
1126 
1127 		sc->sc_txnext = DGE_NEXTTX(sc->sc_txnext);
1128 		txs->txs_ndesc++;
1129 	}
1130 
1131 	*fieldsp = fields;
1132 
1133 	return (0);
1134 }
1135 
1136 /*
1137  * dge_start:		[ifnet interface function]
1138  *
1139  *	Start packet transmission on the interface.
1140  */
1141 static void
1142 dge_start(struct ifnet *ifp)
1143 {
1144 	struct dge_softc *sc = ifp->if_softc;
1145 	struct mbuf *m0;
1146 	struct dge_txsoft *txs;
1147 	bus_dmamap_t dmamap;
1148 	int error, nexttx, lasttx = -1, ofree, seg;
1149 	uint32_t cksumcmd;
1150 	uint8_t cksumfields;
1151 
1152 	if ((ifp->if_flags & (IFF_RUNNING|IFF_OACTIVE)) != IFF_RUNNING)
1153 		return;
1154 
1155 	/*
1156 	 * Remember the previous number of free descriptors.
1157 	 */
1158 	ofree = sc->sc_txfree;
1159 
1160 	/*
1161 	 * Loop through the send queue, setting up transmit descriptors
1162 	 * until we drain the queue, or use up all available transmit
1163 	 * descriptors.
1164 	 */
1165 	for (;;) {
1166 		/* Grab a packet off the queue. */
1167 		IFQ_POLL(&ifp->if_snd, m0);
1168 		if (m0 == NULL)
1169 			break;
1170 
1171 		DPRINTF(DGE_DEBUG_TX,
1172 		    ("%s: TX: have packet to transmit: %p\n",
1173 		    device_xname(&sc->sc_dev), m0));
1174 
1175 		/* Get a work queue entry. */
1176 		if (sc->sc_txsfree < DGE_TXQUEUE_GC) {
1177 			dge_txintr(sc);
1178 			if (sc->sc_txsfree == 0) {
1179 				DPRINTF(DGE_DEBUG_TX,
1180 				    ("%s: TX: no free job descriptors\n",
1181 					device_xname(&sc->sc_dev)));
1182 				DGE_EVCNT_INCR(&sc->sc_ev_txsstall);
1183 				break;
1184 			}
1185 		}
1186 
1187 		txs = &sc->sc_txsoft[sc->sc_txsnext];
1188 		dmamap = txs->txs_dmamap;
1189 
1190 		/*
1191 		 * Load the DMA map.  If this fails, the packet either
1192 		 * didn't fit in the allotted number of segments, or we
1193 		 * were short on resources.  For the too-many-segments
1194 		 * case, we simply report an error and drop the packet,
1195 		 * since we can't sanely copy a jumbo packet to a single
1196 		 * buffer.
1197 		 */
1198 		error = bus_dmamap_load_mbuf(sc->sc_dmat, dmamap, m0,
1199 		    BUS_DMA_WRITE|BUS_DMA_NOWAIT);
1200 		if (error) {
1201 			if (error == EFBIG) {
1202 				DGE_EVCNT_INCR(&sc->sc_ev_txdrop);
1203 				printf("%s: Tx packet consumes too many "
1204 				    "DMA segments, dropping...\n",
1205 				    device_xname(&sc->sc_dev));
1206 				IFQ_DEQUEUE(&ifp->if_snd, m0);
1207 				m_freem(m0);
1208 				continue;
1209 			}
1210 			/*
1211 			 * Short on resources, just stop for now.
1212 			 */
1213 			DPRINTF(DGE_DEBUG_TX,
1214 			    ("%s: TX: dmamap load failed: %d\n",
1215 			    device_xname(&sc->sc_dev), error));
1216 			break;
1217 		}
1218 
1219 		/*
1220 		 * Ensure we have enough descriptors free to describe
1221 		 * the packet.  Note, we always reserve one descriptor
1222 		 * at the end of the ring due to the semantics of the
1223 		 * TDT register, plus one more in the event we need
1224 		 * to re-load checksum offload context.
1225 		 */
1226 		if (dmamap->dm_nsegs > (sc->sc_txfree - 2)) {
1227 			/*
1228 			 * Not enough free descriptors to transmit this
1229 			 * packet.  We haven't committed anything yet,
1230 			 * so just unload the DMA map, put the packet
1231 			 * pack on the queue, and punt.  Notify the upper
1232 			 * layer that there are no more slots left.
1233 			 */
1234 			DPRINTF(DGE_DEBUG_TX,
1235 			    ("%s: TX: need %d descriptors, have %d\n",
1236 			    device_xname(&sc->sc_dev), dmamap->dm_nsegs,
1237 			    sc->sc_txfree - 1));
1238 			ifp->if_flags |= IFF_OACTIVE;
1239 			bus_dmamap_unload(sc->sc_dmat, dmamap);
1240 			DGE_EVCNT_INCR(&sc->sc_ev_txdstall);
1241 			break;
1242 		}
1243 
1244 		IFQ_DEQUEUE(&ifp->if_snd, m0);
1245 
1246 		/*
1247 		 * WE ARE NOW COMMITTED TO TRANSMITTING THE PACKET.
1248 		 */
1249 
1250 		/* Sync the DMA map. */
1251 		bus_dmamap_sync(sc->sc_dmat, dmamap, 0, dmamap->dm_mapsize,
1252 		    BUS_DMASYNC_PREWRITE);
1253 
1254 		DPRINTF(DGE_DEBUG_TX,
1255 		    ("%s: TX: packet has %d DMA segments\n",
1256 		    device_xname(&sc->sc_dev), dmamap->dm_nsegs));
1257 
1258 		DGE_EVCNT_INCR(&sc->sc_ev_txseg[dmamap->dm_nsegs - 1]);
1259 
1260 		/*
1261 		 * Store a pointer to the packet so that we can free it
1262 		 * later.
1263 		 *
1264 		 * Initially, we consider the number of descriptors the
1265 		 * packet uses the number of DMA segments.  This may be
1266 		 * incremented by 1 if we do checksum offload (a descriptor
1267 		 * is used to set the checksum context).
1268 		 */
1269 		txs->txs_mbuf = m0;
1270 		txs->txs_firstdesc = sc->sc_txnext;
1271 		txs->txs_ndesc = dmamap->dm_nsegs;
1272 
1273 		/*
1274 		 * Set up checksum offload parameters for
1275 		 * this packet.
1276 		 */
1277 		if (m0->m_pkthdr.csum_flags &
1278 		    (M_CSUM_IPv4|M_CSUM_TCPv4|M_CSUM_UDPv4)) {
1279 			if (dge_tx_cksum(sc, txs, &cksumfields) != 0) {
1280 				/* Error message already displayed. */
1281 				bus_dmamap_unload(sc->sc_dmat, dmamap);
1282 				continue;
1283 			}
1284 		} else {
1285 			cksumfields = 0;
1286 		}
1287 
1288 		cksumcmd = TDESC_DCMD_IDE | TDESC_DTYP_DATA;
1289 
1290 		/*
1291 		 * Initialize the transmit descriptor.
1292 		 */
1293 		for (nexttx = sc->sc_txnext, seg = 0;
1294 		     seg < dmamap->dm_nsegs;
1295 		     seg++, nexttx = DGE_NEXTTX(nexttx)) {
1296 			/*
1297 			 * Note: we currently only use 32-bit DMA
1298 			 * addresses.
1299 			 */
1300 			sc->sc_txdescs[nexttx].dt_baddrh = 0;
1301 			sc->sc_txdescs[nexttx].dt_baddrl =
1302 			    htole32(dmamap->dm_segs[seg].ds_addr);
1303 			sc->sc_txdescs[nexttx].dt_ctl =
1304 			    htole32(cksumcmd | dmamap->dm_segs[seg].ds_len);
1305 			sc->sc_txdescs[nexttx].dt_status = 0;
1306 			sc->sc_txdescs[nexttx].dt_popts = cksumfields;
1307 			sc->sc_txdescs[nexttx].dt_vlan = 0;
1308 			lasttx = nexttx;
1309 
1310 			DPRINTF(DGE_DEBUG_TX,
1311 			    ("%s: TX: desc %d: low 0x%08lx, len 0x%04lx\n",
1312 			    device_xname(&sc->sc_dev), nexttx,
1313 			    (unsigned long)le32toh(dmamap->dm_segs[seg].ds_addr),
1314 			    (unsigned long)le32toh(dmamap->dm_segs[seg].ds_len)));
1315 		}
1316 
1317 		KASSERT(lasttx != -1);
1318 
1319 		/*
1320 		 * Set up the command byte on the last descriptor of
1321 		 * the packet.  If we're in the interrupt delay window,
1322 		 * delay the interrupt.
1323 		 */
1324 		sc->sc_txdescs[lasttx].dt_ctl |=
1325 		    htole32(TDESC_DCMD_EOP | TDESC_DCMD_RS);
1326 
1327 		txs->txs_lastdesc = lasttx;
1328 
1329 		DPRINTF(DGE_DEBUG_TX,
1330 		    ("%s: TX: desc %d: cmdlen 0x%08x\n", device_xname(&sc->sc_dev),
1331 		    lasttx, le32toh(sc->sc_txdescs[lasttx].dt_ctl)));
1332 
1333 		/* Sync the descriptors we're using. */
1334 		DGE_CDTXSYNC(sc, sc->sc_txnext, dmamap->dm_nsegs,
1335 		    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1336 
1337 		/* Give the packet to the chip. */
1338 		CSR_WRITE(sc, DGE_TDT, nexttx);
1339 
1340 		DPRINTF(DGE_DEBUG_TX,
1341 		    ("%s: TX: TDT -> %d\n", device_xname(&sc->sc_dev), nexttx));
1342 
1343 		DPRINTF(DGE_DEBUG_TX,
1344 		    ("%s: TX: finished transmitting packet, job %d\n",
1345 		    device_xname(&sc->sc_dev), sc->sc_txsnext));
1346 
1347 		/* Advance the tx pointer. */
1348 		sc->sc_txfree -= txs->txs_ndesc;
1349 		sc->sc_txnext = nexttx;
1350 
1351 		sc->sc_txsfree--;
1352 		sc->sc_txsnext = DGE_NEXTTXS(sc->sc_txsnext);
1353 
1354 #if NBPFILTER > 0
1355 		/* Pass the packet to any BPF listeners. */
1356 		if (ifp->if_bpf)
1357 			bpf_mtap(ifp->if_bpf, m0);
1358 #endif /* NBPFILTER > 0 */
1359 	}
1360 
1361 	if (sc->sc_txsfree == 0 || sc->sc_txfree <= 2) {
1362 		/* No more slots; notify upper layer. */
1363 		ifp->if_flags |= IFF_OACTIVE;
1364 	}
1365 
1366 	if (sc->sc_txfree != ofree) {
1367 		/* Set a watchdog timer in case the chip flakes out. */
1368 		ifp->if_timer = 5;
1369 	}
1370 }
1371 
1372 /*
1373  * dge_watchdog:		[ifnet interface function]
1374  *
1375  *	Watchdog timer handler.
1376  */
1377 static void
1378 dge_watchdog(struct ifnet *ifp)
1379 {
1380 	struct dge_softc *sc = ifp->if_softc;
1381 
1382 	/*
1383 	 * Since we're using delayed interrupts, sweep up
1384 	 * before we report an error.
1385 	 */
1386 	dge_txintr(sc);
1387 
1388 	if (sc->sc_txfree != DGE_NTXDESC) {
1389 		printf("%s: device timeout (txfree %d txsfree %d txnext %d)\n",
1390 		    device_xname(&sc->sc_dev), sc->sc_txfree, sc->sc_txsfree,
1391 		    sc->sc_txnext);
1392 		ifp->if_oerrors++;
1393 
1394 		/* Reset the interface. */
1395 		(void) dge_init(ifp);
1396 	}
1397 
1398 	/* Try to get more packets going. */
1399 	dge_start(ifp);
1400 }
1401 
1402 /*
1403  * dge_ioctl:		[ifnet interface function]
1404  *
1405  *	Handle control requests from the operator.
1406  */
1407 static int
1408 dge_ioctl(struct ifnet *ifp, u_long cmd, void *data)
1409 {
1410 	struct dge_softc *sc = ifp->if_softc;
1411 	struct ifreq *ifr = (struct ifreq *) data;
1412 	pcireg_t preg;
1413 	int s, error, mmrbc;
1414 
1415 	s = splnet();
1416 
1417 	switch (cmd) {
1418 	case SIOCSIFMEDIA:
1419 	case SIOCGIFMEDIA:
1420 		error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
1421 		break;
1422 
1423 	case SIOCSIFMTU:
1424 		if (ifr->ifr_mtu < ETHERMIN || ifr->ifr_mtu > DGE_MAX_MTU)
1425 			error = EINVAL;
1426 		else if ((error = ifioctl_common(ifp, cmd, data)) != ENETRESET)
1427 			break;
1428 		else if (ifp->if_flags & IFF_UP)
1429 			error = (*ifp->if_init)(ifp);
1430 		else
1431 			error = 0;
1432 		break;
1433 
1434         case SIOCSIFFLAGS:
1435 		if ((error = ifioctl_common(ifp, cmd, data)) != 0)
1436 			break;
1437 		/* extract link flags */
1438 		if ((ifp->if_flags & IFF_LINK0) == 0 &&
1439 		    (ifp->if_flags & IFF_LINK1) == 0)
1440 			mmrbc = PCIX_MMRBC_512;
1441 		else if ((ifp->if_flags & IFF_LINK0) == 0 &&
1442 		    (ifp->if_flags & IFF_LINK1) != 0)
1443 			mmrbc = PCIX_MMRBC_1024;
1444 		else if ((ifp->if_flags & IFF_LINK0) != 0 &&
1445 		    (ifp->if_flags & IFF_LINK1) == 0)
1446 			mmrbc = PCIX_MMRBC_2048;
1447 		else
1448 			mmrbc = PCIX_MMRBC_4096;
1449 		if (mmrbc != sc->sc_mmrbc) {
1450 			preg = pci_conf_read(sc->sc_pc, sc->sc_pt,DGE_PCIX_CMD);
1451 			preg &= ~PCIX_MMRBC_MSK;
1452 			preg |= mmrbc;
1453 			pci_conf_write(sc->sc_pc, sc->sc_pt,DGE_PCIX_CMD, preg);
1454 			sc->sc_mmrbc = mmrbc;
1455 		}
1456                 /* FALLTHROUGH */
1457 	default:
1458 		if ((error = ether_ioctl(ifp, cmd, data)) != ENETRESET)
1459 			break;
1460 
1461 		error = 0;
1462 
1463 		if (cmd == SIOCSIFCAP)
1464 			error = (*ifp->if_init)(ifp);
1465 		else if (cmd != SIOCADDMULTI && cmd != SIOCDELMULTI)
1466 			;
1467 		else if (ifp->if_flags & IFF_RUNNING) {
1468 			/*
1469 			 * Multicast list has changed; set the hardware filter
1470 			 * accordingly.
1471 			 */
1472 			dge_set_filter(sc);
1473 		}
1474 		break;
1475 	}
1476 
1477 	/* Try to get more packets going. */
1478 	dge_start(ifp);
1479 
1480 	splx(s);
1481 	return (error);
1482 }
1483 
1484 /*
1485  * dge_intr:
1486  *
1487  *	Interrupt service routine.
1488  */
1489 static int
1490 dge_intr(void *arg)
1491 {
1492 	struct dge_softc *sc = arg;
1493 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
1494 	uint32_t icr;
1495 	int wantinit, handled = 0;
1496 
1497 	for (wantinit = 0; wantinit == 0;) {
1498 		icr = CSR_READ(sc, DGE_ICR);
1499 		if ((icr & sc->sc_icr) == 0)
1500 			break;
1501 
1502 #if 0 /*NRND > 0*/
1503 		if (RND_ENABLED(&sc->rnd_source))
1504 			rnd_add_uint32(&sc->rnd_source, icr);
1505 #endif
1506 
1507 		handled = 1;
1508 
1509 #if defined(DGE_DEBUG) || defined(DGE_EVENT_COUNTERS)
1510 		if (icr & (ICR_RXDMT0|ICR_RXT0)) {
1511 			DPRINTF(DGE_DEBUG_RX,
1512 			    ("%s: RX: got Rx intr 0x%08x\n",
1513 			    device_xname(&sc->sc_dev),
1514 			    icr & (ICR_RXDMT0|ICR_RXT0)));
1515 			DGE_EVCNT_INCR(&sc->sc_ev_rxintr);
1516 		}
1517 #endif
1518 		dge_rxintr(sc);
1519 
1520 #if defined(DGE_DEBUG) || defined(DGE_EVENT_COUNTERS)
1521 		if (icr & ICR_TXDW) {
1522 			DPRINTF(DGE_DEBUG_TX,
1523 			    ("%s: TX: got TXDW interrupt\n",
1524 			    device_xname(&sc->sc_dev)));
1525 			DGE_EVCNT_INCR(&sc->sc_ev_txdw);
1526 		}
1527 		if (icr & ICR_TXQE)
1528 			DGE_EVCNT_INCR(&sc->sc_ev_txqe);
1529 #endif
1530 		dge_txintr(sc);
1531 
1532 		if (icr & (ICR_LSC|ICR_RXSEQ)) {
1533 			DGE_EVCNT_INCR(&sc->sc_ev_linkintr);
1534 			dge_linkintr(sc, icr);
1535 		}
1536 
1537 		if (icr & ICR_RXO) {
1538 			printf("%s: Receive overrun\n", device_xname(&sc->sc_dev));
1539 			wantinit = 1;
1540 		}
1541 	}
1542 
1543 	if (handled) {
1544 		if (wantinit)
1545 			dge_init(ifp);
1546 
1547 		/* Try to get more packets going. */
1548 		dge_start(ifp);
1549 	}
1550 
1551 	return (handled);
1552 }
1553 
1554 /*
1555  * dge_txintr:
1556  *
1557  *	Helper; handle transmit interrupts.
1558  */
1559 static void
1560 dge_txintr(struct dge_softc *sc)
1561 {
1562 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
1563 	struct dge_txsoft *txs;
1564 	uint8_t status;
1565 	int i;
1566 
1567 	ifp->if_flags &= ~IFF_OACTIVE;
1568 
1569 	/*
1570 	 * Go through the Tx list and free mbufs for those
1571 	 * frames which have been transmitted.
1572 	 */
1573 	for (i = sc->sc_txsdirty; sc->sc_txsfree != DGE_TXQUEUELEN;
1574 	     i = DGE_NEXTTXS(i), sc->sc_txsfree++) {
1575 		txs = &sc->sc_txsoft[i];
1576 
1577 		DPRINTF(DGE_DEBUG_TX,
1578 		    ("%s: TX: checking job %d\n", device_xname(&sc->sc_dev), i));
1579 
1580 		DGE_CDTXSYNC(sc, txs->txs_firstdesc, txs->txs_dmamap->dm_nsegs,
1581 		    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1582 
1583 		status =
1584 		    sc->sc_txdescs[txs->txs_lastdesc].dt_status;
1585 		if ((status & TDESC_STA_DD) == 0) {
1586 			DGE_CDTXSYNC(sc, txs->txs_lastdesc, 1,
1587 			    BUS_DMASYNC_PREREAD);
1588 			break;
1589 		}
1590 
1591 		DPRINTF(DGE_DEBUG_TX,
1592 		    ("%s: TX: job %d done: descs %d..%d\n",
1593 		    device_xname(&sc->sc_dev), i, txs->txs_firstdesc,
1594 		    txs->txs_lastdesc));
1595 
1596 		ifp->if_opackets++;
1597 		sc->sc_txfree += txs->txs_ndesc;
1598 		bus_dmamap_sync(sc->sc_dmat, txs->txs_dmamap,
1599 		    0, txs->txs_dmamap->dm_mapsize, BUS_DMASYNC_POSTWRITE);
1600 		bus_dmamap_unload(sc->sc_dmat, txs->txs_dmamap);
1601 		m_freem(txs->txs_mbuf);
1602 		txs->txs_mbuf = NULL;
1603 	}
1604 
1605 	/* Update the dirty transmit buffer pointer. */
1606 	sc->sc_txsdirty = i;
1607 	DPRINTF(DGE_DEBUG_TX,
1608 	    ("%s: TX: txsdirty -> %d\n", device_xname(&sc->sc_dev), i));
1609 
1610 	/*
1611 	 * If there are no more pending transmissions, cancel the watchdog
1612 	 * timer.
1613 	 */
1614 	if (sc->sc_txsfree == DGE_TXQUEUELEN)
1615 		ifp->if_timer = 0;
1616 }
1617 
1618 /*
1619  * dge_rxintr:
1620  *
1621  *	Helper; handle receive interrupts.
1622  */
1623 static void
1624 dge_rxintr(struct dge_softc *sc)
1625 {
1626 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
1627 	struct dge_rxsoft *rxs;
1628 	struct mbuf *m;
1629 	int i, len;
1630 	uint8_t status, errors;
1631 
1632 	for (i = sc->sc_rxptr;; i = DGE_NEXTRX(i)) {
1633 		rxs = &sc->sc_rxsoft[i];
1634 
1635 		DPRINTF(DGE_DEBUG_RX,
1636 		    ("%s: RX: checking descriptor %d\n",
1637 		    device_xname(&sc->sc_dev), i));
1638 
1639 		DGE_CDRXSYNC(sc, i, BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1640 
1641 		status = sc->sc_rxdescs[i].dr_status;
1642 		errors = sc->sc_rxdescs[i].dr_errors;
1643 		len = le16toh(sc->sc_rxdescs[i].dr_len);
1644 
1645 		if ((status & RDESC_STS_DD) == 0) {
1646 			/*
1647 			 * We have processed all of the receive descriptors.
1648 			 */
1649 			DGE_CDRXSYNC(sc, i, BUS_DMASYNC_PREREAD);
1650 			break;
1651 		}
1652 
1653 		if (__predict_false(sc->sc_rxdiscard)) {
1654 			DPRINTF(DGE_DEBUG_RX,
1655 			    ("%s: RX: discarding contents of descriptor %d\n",
1656 			    device_xname(&sc->sc_dev), i));
1657 			DGE_INIT_RXDESC(sc, i);
1658 			if (status & RDESC_STS_EOP) {
1659 				/* Reset our state. */
1660 				DPRINTF(DGE_DEBUG_RX,
1661 				    ("%s: RX: resetting rxdiscard -> 0\n",
1662 				    device_xname(&sc->sc_dev)));
1663 				sc->sc_rxdiscard = 0;
1664 			}
1665 			continue;
1666 		}
1667 
1668 		bus_dmamap_sync(sc->sc_dmat, rxs->rxs_dmamap, 0,
1669 		    rxs->rxs_dmamap->dm_mapsize, BUS_DMASYNC_POSTREAD);
1670 
1671 		m = rxs->rxs_mbuf;
1672 
1673 		/*
1674 		 * Add a new receive buffer to the ring.
1675 		 */
1676 		if (dge_add_rxbuf(sc, i) != 0) {
1677 			/*
1678 			 * Failed, throw away what we've done so
1679 			 * far, and discard the rest of the packet.
1680 			 */
1681 			ifp->if_ierrors++;
1682 			bus_dmamap_sync(sc->sc_dmat, rxs->rxs_dmamap, 0,
1683 			    rxs->rxs_dmamap->dm_mapsize, BUS_DMASYNC_PREREAD);
1684 			DGE_INIT_RXDESC(sc, i);
1685 			if ((status & RDESC_STS_EOP) == 0)
1686 				sc->sc_rxdiscard = 1;
1687 			if (sc->sc_rxhead != NULL)
1688 				m_freem(sc->sc_rxhead);
1689 			DGE_RXCHAIN_RESET(sc);
1690 			DPRINTF(DGE_DEBUG_RX,
1691 			    ("%s: RX: Rx buffer allocation failed, "
1692 			    "dropping packet%s\n", device_xname(&sc->sc_dev),
1693 			    sc->sc_rxdiscard ? " (discard)" : ""));
1694 			continue;
1695 		}
1696 		DGE_INIT_RXDESC(sc, DGE_PREVRX(i)); /* Write the descriptor */
1697 
1698 		DGE_RXCHAIN_LINK(sc, m);
1699 
1700 		m->m_len = len;
1701 
1702 		DPRINTF(DGE_DEBUG_RX,
1703 		    ("%s: RX: buffer at %p len %d\n",
1704 		    device_xname(&sc->sc_dev), m->m_data, len));
1705 
1706 		/*
1707 		 * If this is not the end of the packet, keep
1708 		 * looking.
1709 		 */
1710 		if ((status & RDESC_STS_EOP) == 0) {
1711 			sc->sc_rxlen += len;
1712 			DPRINTF(DGE_DEBUG_RX,
1713 			    ("%s: RX: not yet EOP, rxlen -> %d\n",
1714 			    device_xname(&sc->sc_dev), sc->sc_rxlen));
1715 			continue;
1716 		}
1717 
1718 		/*
1719 		 * Okay, we have the entire packet now...
1720 		 */
1721 		*sc->sc_rxtailp = NULL;
1722 		m = sc->sc_rxhead;
1723 		len += sc->sc_rxlen;
1724 
1725 		DGE_RXCHAIN_RESET(sc);
1726 
1727 		DPRINTF(DGE_DEBUG_RX,
1728 		    ("%s: RX: have entire packet, len -> %d\n",
1729 		    device_xname(&sc->sc_dev), len));
1730 
1731 		/*
1732 		 * If an error occurred, update stats and drop the packet.
1733 		 */
1734 		if (errors &
1735 		     (RDESC_ERR_CE|RDESC_ERR_SE|RDESC_ERR_P|RDESC_ERR_RXE)) {
1736 			ifp->if_ierrors++;
1737 			if (errors & RDESC_ERR_SE)
1738 				printf("%s: symbol error\n",
1739 				    device_xname(&sc->sc_dev));
1740 			else if (errors & RDESC_ERR_P)
1741 				printf("%s: parity error\n",
1742 				    device_xname(&sc->sc_dev));
1743 			else if (errors & RDESC_ERR_CE)
1744 				printf("%s: CRC error\n",
1745 				    device_xname(&sc->sc_dev));
1746 			m_freem(m);
1747 			continue;
1748 		}
1749 
1750 		/*
1751 		 * No errors.  Receive the packet.
1752 		 */
1753 		m->m_pkthdr.rcvif = ifp;
1754 		m->m_pkthdr.len = len;
1755 
1756 		/*
1757 		 * Set up checksum info for this packet.
1758 		 */
1759 		if (status & RDESC_STS_IPCS) {
1760 			DGE_EVCNT_INCR(&sc->sc_ev_rxipsum);
1761 			m->m_pkthdr.csum_flags |= M_CSUM_IPv4;
1762 			if (errors & RDESC_ERR_IPE)
1763 				m->m_pkthdr.csum_flags |= M_CSUM_IPv4_BAD;
1764 		}
1765 		if (status & RDESC_STS_TCPCS) {
1766 			/*
1767 			 * Note: we don't know if this was TCP or UDP,
1768 			 * so we just set both bits, and expect the
1769 			 * upper layers to deal.
1770 			 */
1771 			DGE_EVCNT_INCR(&sc->sc_ev_rxtusum);
1772 			m->m_pkthdr.csum_flags |= M_CSUM_TCPv4|M_CSUM_UDPv4;
1773 			if (errors & RDESC_ERR_TCPE)
1774 				m->m_pkthdr.csum_flags |= M_CSUM_TCP_UDP_BAD;
1775 		}
1776 
1777 		ifp->if_ipackets++;
1778 
1779 #if NBPFILTER > 0
1780 		/* Pass this up to any BPF listeners. */
1781 		if (ifp->if_bpf)
1782 			bpf_mtap(ifp->if_bpf, m);
1783 #endif /* NBPFILTER > 0 */
1784 
1785 		/* Pass it on. */
1786 		(*ifp->if_input)(ifp, m);
1787 	}
1788 
1789 	/* Update the receive pointer. */
1790 	sc->sc_rxptr = i;
1791 
1792 	DPRINTF(DGE_DEBUG_RX,
1793 	    ("%s: RX: rxptr -> %d\n", device_xname(&sc->sc_dev), i));
1794 }
1795 
1796 /*
1797  * dge_linkintr:
1798  *
1799  *	Helper; handle link interrupts.
1800  */
1801 static void
1802 dge_linkintr(struct dge_softc *sc, uint32_t icr)
1803 {
1804 	uint32_t status;
1805 
1806 	if (icr & ICR_LSC) {
1807 		status = CSR_READ(sc, DGE_STATUS);
1808 		if (status & STATUS_LINKUP) {
1809 			DPRINTF(DGE_DEBUG_LINK, ("%s: LINK: LSC -> up\n",
1810 			    device_xname(&sc->sc_dev)));
1811 		} else {
1812 			DPRINTF(DGE_DEBUG_LINK, ("%s: LINK: LSC -> down\n",
1813 			    device_xname(&sc->sc_dev)));
1814 		}
1815 	} else if (icr & ICR_RXSEQ) {
1816 		DPRINTF(DGE_DEBUG_LINK,
1817 		    ("%s: LINK: Receive sequence error\n",
1818 		    device_xname(&sc->sc_dev)));
1819 	}
1820 	/* XXX - fix errata */
1821 }
1822 
1823 /*
1824  * dge_reset:
1825  *
1826  *	Reset the i82597 chip.
1827  */
1828 static void
1829 dge_reset(struct dge_softc *sc)
1830 {
1831 	int i;
1832 
1833 	/*
1834 	 * Do a chip reset.
1835 	 */
1836 	CSR_WRITE(sc, DGE_CTRL0, CTRL0_RST | sc->sc_ctrl0);
1837 
1838 	delay(10000);
1839 
1840 	for (i = 0; i < 1000; i++) {
1841 		if ((CSR_READ(sc, DGE_CTRL0) & CTRL0_RST) == 0)
1842 			break;
1843 		delay(20);
1844 	}
1845 
1846 	if (CSR_READ(sc, DGE_CTRL0) & CTRL0_RST)
1847 		printf("%s: WARNING: reset failed to complete\n",
1848 		    device_xname(&sc->sc_dev));
1849         /*
1850          * Reset the EEPROM logic.
1851          * This will cause the chip to reread its default values,
1852 	 * which doesn't happen otherwise (errata).
1853          */
1854         CSR_WRITE(sc, DGE_CTRL1, CTRL1_EE_RST);
1855         delay(10000);
1856 }
1857 
1858 /*
1859  * dge_init:		[ifnet interface function]
1860  *
1861  *	Initialize the interface.  Must be called at splnet().
1862  */
1863 static int
1864 dge_init(struct ifnet *ifp)
1865 {
1866 	struct dge_softc *sc = ifp->if_softc;
1867 	struct dge_rxsoft *rxs;
1868 	int i, error = 0;
1869 	uint32_t reg;
1870 
1871 	/*
1872 	 * *_HDR_ALIGNED_P is constant 1 if __NO_STRICT_ALIGMENT is set.
1873 	 * There is a small but measurable benefit to avoiding the adjusment
1874 	 * of the descriptor so that the headers are aligned, for normal mtu,
1875 	 * on such platforms.  One possibility is that the DMA itself is
1876 	 * slightly more efficient if the front of the entire packet (instead
1877 	 * of the front of the headers) is aligned.
1878 	 *
1879 	 * Note we must always set align_tweak to 0 if we are using
1880 	 * jumbo frames.
1881 	 */
1882 #ifdef __NO_STRICT_ALIGNMENT
1883 	sc->sc_align_tweak = 0;
1884 #else
1885 	if ((ifp->if_mtu + ETHER_HDR_LEN + ETHER_CRC_LEN) > (MCLBYTES - 2))
1886 		sc->sc_align_tweak = 0;
1887 	else
1888 		sc->sc_align_tweak = 2;
1889 #endif /* __NO_STRICT_ALIGNMENT */
1890 
1891 	/* Cancel any pending I/O. */
1892 	dge_stop(ifp, 0);
1893 
1894 	/* Reset the chip to a known state. */
1895 	dge_reset(sc);
1896 
1897 	/* Initialize the transmit descriptor ring. */
1898 	memset(sc->sc_txdescs, 0, sizeof(sc->sc_txdescs));
1899 	DGE_CDTXSYNC(sc, 0, DGE_NTXDESC,
1900 	    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1901 	sc->sc_txfree = DGE_NTXDESC;
1902 	sc->sc_txnext = 0;
1903 
1904 	sc->sc_txctx_ipcs = 0xffffffff;
1905 	sc->sc_txctx_tucs = 0xffffffff;
1906 
1907 	CSR_WRITE(sc, DGE_TDBAH, 0);
1908 	CSR_WRITE(sc, DGE_TDBAL, DGE_CDTXADDR(sc, 0));
1909 	CSR_WRITE(sc, DGE_TDLEN, sizeof(sc->sc_txdescs));
1910 	CSR_WRITE(sc, DGE_TDH, 0);
1911 	CSR_WRITE(sc, DGE_TDT, 0);
1912 	CSR_WRITE(sc, DGE_TIDV, TIDV);
1913 
1914 #if 0
1915 	CSR_WRITE(sc, DGE_TXDCTL, TXDCTL_PTHRESH(0) |
1916 	    TXDCTL_HTHRESH(0) | TXDCTL_WTHRESH(0));
1917 #endif
1918 	CSR_WRITE(sc, DGE_RXDCTL,
1919 	    RXDCTL_PTHRESH(RXDCTL_PTHRESH_VAL) |
1920 	    RXDCTL_HTHRESH(RXDCTL_HTHRESH_VAL) |
1921 	    RXDCTL_WTHRESH(RXDCTL_WTHRESH_VAL));
1922 
1923 	/* Initialize the transmit job descriptors. */
1924 	for (i = 0; i < DGE_TXQUEUELEN; i++)
1925 		sc->sc_txsoft[i].txs_mbuf = NULL;
1926 	sc->sc_txsfree = DGE_TXQUEUELEN;
1927 	sc->sc_txsnext = 0;
1928 	sc->sc_txsdirty = 0;
1929 
1930 	/*
1931 	 * Initialize the receive descriptor and receive job
1932 	 * descriptor rings.
1933 	 */
1934 	CSR_WRITE(sc, DGE_RDBAH, 0);
1935 	CSR_WRITE(sc, DGE_RDBAL, DGE_CDRXADDR(sc, 0));
1936 	CSR_WRITE(sc, DGE_RDLEN, sizeof(sc->sc_rxdescs));
1937 	CSR_WRITE(sc, DGE_RDH, DGE_RXSPACE);
1938 	CSR_WRITE(sc, DGE_RDT, 0);
1939 	CSR_WRITE(sc, DGE_RDTR, RDTR | 0x80000000);
1940 	CSR_WRITE(sc, DGE_FCRTL, FCRTL | FCRTL_XONE);
1941 	CSR_WRITE(sc, DGE_FCRTH, FCRTH);
1942 
1943 	for (i = 0; i < DGE_NRXDESC; i++) {
1944 		rxs = &sc->sc_rxsoft[i];
1945 		if (rxs->rxs_mbuf == NULL) {
1946 			if ((error = dge_add_rxbuf(sc, i)) != 0) {
1947 				printf("%s: unable to allocate or map rx "
1948 				    "buffer %d, error = %d\n",
1949 				    device_xname(&sc->sc_dev), i, error);
1950 				/*
1951 				 * XXX Should attempt to run with fewer receive
1952 				 * XXX buffers instead of just failing.
1953 				 */
1954 				dge_rxdrain(sc);
1955 				goto out;
1956 			}
1957 		}
1958 		DGE_INIT_RXDESC(sc, i);
1959 	}
1960 	sc->sc_rxptr = DGE_RXSPACE;
1961 	sc->sc_rxdiscard = 0;
1962 	DGE_RXCHAIN_RESET(sc);
1963 
1964 	if (sc->sc_ethercom.ec_capabilities & ETHERCAP_JUMBO_MTU) {
1965 		sc->sc_ctrl0 |= CTRL0_JFE;
1966 		CSR_WRITE(sc, DGE_MFS, ETHER_MAX_LEN_JUMBO << 16);
1967 	}
1968 
1969 	/* Write the control registers. */
1970 	CSR_WRITE(sc, DGE_CTRL0, sc->sc_ctrl0);
1971 
1972 	/*
1973 	 * Set up checksum offload parameters.
1974 	 */
1975 	reg = CSR_READ(sc, DGE_RXCSUM);
1976 	if (ifp->if_capenable & IFCAP_CSUM_IPv4_Rx)
1977 		reg |= RXCSUM_IPOFL;
1978 	else
1979 		reg &= ~RXCSUM_IPOFL;
1980 	if (ifp->if_capenable & (IFCAP_CSUM_TCPv4_Rx | IFCAP_CSUM_UDPv4_Rx))
1981 		reg |= RXCSUM_IPOFL | RXCSUM_TUOFL;
1982 	else {
1983 		reg &= ~RXCSUM_TUOFL;
1984 		if ((ifp->if_capenable & IFCAP_CSUM_IPv4_Rx) == 0)
1985 			reg &= ~RXCSUM_IPOFL;
1986 	}
1987 	CSR_WRITE(sc, DGE_RXCSUM, reg);
1988 
1989 	/*
1990 	 * Set up the interrupt registers.
1991 	 */
1992 	CSR_WRITE(sc, DGE_IMC, 0xffffffffU);
1993 	sc->sc_icr = ICR_TXDW | ICR_LSC | ICR_RXSEQ | ICR_RXDMT0 |
1994 	    ICR_RXO | ICR_RXT0;
1995 
1996 	CSR_WRITE(sc, DGE_IMS, sc->sc_icr);
1997 
1998 	/*
1999 	 * Set up the transmit control register.
2000 	 */
2001 	sc->sc_tctl = TCTL_TCE|TCTL_TPDE|TCTL_TXEN;
2002 	CSR_WRITE(sc, DGE_TCTL, sc->sc_tctl);
2003 
2004 	/*
2005 	 * Set up the receive control register; we actually program
2006 	 * the register when we set the receive filter.  Use multicast
2007 	 * address offset type 0.
2008 	 */
2009 	sc->sc_mchash_type = 0;
2010 
2011 	sc->sc_rctl = RCTL_RXEN | RCTL_RDMTS_12 | RCTL_RPDA_MC |
2012 	    RCTL_CFF | RCTL_SECRC | RCTL_MO(sc->sc_mchash_type);
2013 
2014 #ifdef DGE_OFFBYONE_RXBUG
2015 	sc->sc_rctl |= RCTL_BSIZE_16k;
2016 #else
2017 	switch(MCLBYTES) {
2018 	case 2048:
2019 		sc->sc_rctl |= RCTL_BSIZE_2k;
2020 		break;
2021 	case 4096:
2022 		sc->sc_rctl |= RCTL_BSIZE_4k;
2023 		break;
2024 	case 8192:
2025 		sc->sc_rctl |= RCTL_BSIZE_8k;
2026 		break;
2027 	case 16384:
2028 		sc->sc_rctl |= RCTL_BSIZE_16k;
2029 		break;
2030 	default:
2031 		panic("dge_init: MCLBYTES %d unsupported", MCLBYTES);
2032 	}
2033 #endif
2034 
2035 	/* Set the receive filter. */
2036 	/* Also sets RCTL */
2037 	dge_set_filter(sc);
2038 
2039 	/* ...all done! */
2040 	ifp->if_flags |= IFF_RUNNING;
2041 	ifp->if_flags &= ~IFF_OACTIVE;
2042 
2043  out:
2044 	if (error)
2045 		printf("%s: interface not running\n", device_xname(&sc->sc_dev));
2046 	return (error);
2047 }
2048 
2049 /*
2050  * dge_rxdrain:
2051  *
2052  *	Drain the receive queue.
2053  */
2054 static void
2055 dge_rxdrain(struct dge_softc *sc)
2056 {
2057 	struct dge_rxsoft *rxs;
2058 	int i;
2059 
2060 	for (i = 0; i < DGE_NRXDESC; i++) {
2061 		rxs = &sc->sc_rxsoft[i];
2062 		if (rxs->rxs_mbuf != NULL) {
2063 			bus_dmamap_unload(sc->sc_dmat, rxs->rxs_dmamap);
2064 			m_freem(rxs->rxs_mbuf);
2065 			rxs->rxs_mbuf = NULL;
2066 		}
2067 	}
2068 }
2069 
2070 /*
2071  * dge_stop:		[ifnet interface function]
2072  *
2073  *	Stop transmission on the interface.
2074  */
2075 static void
2076 dge_stop(struct ifnet *ifp, int disable)
2077 {
2078 	struct dge_softc *sc = ifp->if_softc;
2079 	struct dge_txsoft *txs;
2080 	int i;
2081 
2082 	/* Stop the transmit and receive processes. */
2083 	CSR_WRITE(sc, DGE_TCTL, 0);
2084 	CSR_WRITE(sc, DGE_RCTL, 0);
2085 
2086 	/* Release any queued transmit buffers. */
2087 	for (i = 0; i < DGE_TXQUEUELEN; i++) {
2088 		txs = &sc->sc_txsoft[i];
2089 		if (txs->txs_mbuf != NULL) {
2090 			bus_dmamap_unload(sc->sc_dmat, txs->txs_dmamap);
2091 			m_freem(txs->txs_mbuf);
2092 			txs->txs_mbuf = NULL;
2093 		}
2094 	}
2095 
2096 	/* Mark the interface as down and cancel the watchdog timer. */
2097 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
2098 	ifp->if_timer = 0;
2099 
2100 	if (disable)
2101 		dge_rxdrain(sc);
2102 }
2103 
2104 /*
2105  * dge_add_rxbuf:
2106  *
2107  *	Add a receive buffer to the indiciated descriptor.
2108  */
2109 static int
2110 dge_add_rxbuf(struct dge_softc *sc, int idx)
2111 {
2112 	struct dge_rxsoft *rxs = &sc->sc_rxsoft[idx];
2113 	struct mbuf *m;
2114 	int error;
2115 #ifdef DGE_OFFBYONE_RXBUG
2116 	void *buf;
2117 #endif
2118 
2119 	MGETHDR(m, M_DONTWAIT, MT_DATA);
2120 	if (m == NULL)
2121 		return (ENOBUFS);
2122 
2123 #ifdef DGE_OFFBYONE_RXBUG
2124 	if ((buf = dge_getbuf(sc)) == NULL)
2125 		return ENOBUFS;
2126 
2127 	m->m_len = m->m_pkthdr.len = DGE_BUFFER_SIZE;
2128 	MEXTADD(m, buf, DGE_BUFFER_SIZE, M_DEVBUF, dge_freebuf, sc);
2129 	m->m_flags |= M_EXT_RW;
2130 
2131 	if (rxs->rxs_mbuf != NULL)
2132 		bus_dmamap_unload(sc->sc_dmat, rxs->rxs_dmamap);
2133 	rxs->rxs_mbuf = m;
2134 
2135 	error = bus_dmamap_load(sc->sc_dmat, rxs->rxs_dmamap, buf,
2136 	    DGE_BUFFER_SIZE, NULL, BUS_DMA_READ|BUS_DMA_NOWAIT);
2137 #else
2138 	MCLGET(m, M_DONTWAIT);
2139 	if ((m->m_flags & M_EXT) == 0) {
2140 		m_freem(m);
2141 		return (ENOBUFS);
2142 	}
2143 
2144 	if (rxs->rxs_mbuf != NULL)
2145 		bus_dmamap_unload(sc->sc_dmat, rxs->rxs_dmamap);
2146 
2147 	rxs->rxs_mbuf = m;
2148 
2149 	m->m_len = m->m_pkthdr.len = m->m_ext.ext_size;
2150 	error = bus_dmamap_load_mbuf(sc->sc_dmat, rxs->rxs_dmamap, m,
2151 	    BUS_DMA_READ|BUS_DMA_NOWAIT);
2152 #endif
2153 	if (error) {
2154 		printf("%s: unable to load rx DMA map %d, error = %d\n",
2155 		    device_xname(&sc->sc_dev), idx, error);
2156 		panic("dge_add_rxbuf");	/* XXX XXX XXX */
2157 	}
2158 	bus_dmamap_sync(sc->sc_dmat, rxs->rxs_dmamap, 0,
2159 	    rxs->rxs_dmamap->dm_mapsize, BUS_DMASYNC_PREREAD);
2160 
2161 	return (0);
2162 }
2163 
2164 /*
2165  * dge_set_ral:
2166  *
2167  *	Set an entry in the receive address list.
2168  */
2169 static void
2170 dge_set_ral(struct dge_softc *sc, const uint8_t *enaddr, int idx)
2171 {
2172 	uint32_t ral_lo, ral_hi;
2173 
2174 	if (enaddr != NULL) {
2175 		ral_lo = enaddr[0] | (enaddr[1] << 8) | (enaddr[2] << 16) |
2176 		    (enaddr[3] << 24);
2177 		ral_hi = enaddr[4] | (enaddr[5] << 8);
2178 		ral_hi |= RAH_AV;
2179 	} else {
2180 		ral_lo = 0;
2181 		ral_hi = 0;
2182 	}
2183 	CSR_WRITE(sc, RA_ADDR(DGE_RAL, idx), ral_lo);
2184 	CSR_WRITE(sc, RA_ADDR(DGE_RAH, idx), ral_hi);
2185 }
2186 
2187 /*
2188  * dge_mchash:
2189  *
2190  *	Compute the hash of the multicast address for the 4096-bit
2191  *	multicast filter.
2192  */
2193 static uint32_t
2194 dge_mchash(struct dge_softc *sc, const uint8_t *enaddr)
2195 {
2196 	static const int lo_shift[4] = { 4, 3, 2, 0 };
2197 	static const int hi_shift[4] = { 4, 5, 6, 8 };
2198 	uint32_t hash;
2199 
2200 	hash = (enaddr[4] >> lo_shift[sc->sc_mchash_type]) |
2201 	    (((uint16_t) enaddr[5]) << hi_shift[sc->sc_mchash_type]);
2202 
2203 	return (hash & 0xfff);
2204 }
2205 
2206 /*
2207  * dge_set_filter:
2208  *
2209  *	Set up the receive filter.
2210  */
2211 static void
2212 dge_set_filter(struct dge_softc *sc)
2213 {
2214 	struct ethercom *ec = &sc->sc_ethercom;
2215 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
2216 	struct ether_multi *enm;
2217 	struct ether_multistep step;
2218 	uint32_t hash, reg, bit;
2219 	int i;
2220 
2221 	sc->sc_rctl &= ~(RCTL_BAM | RCTL_UPE | RCTL_MPE);
2222 
2223 	if (ifp->if_flags & IFF_BROADCAST)
2224 		sc->sc_rctl |= RCTL_BAM;
2225 	if (ifp->if_flags & IFF_PROMISC) {
2226 		sc->sc_rctl |= RCTL_UPE;
2227 		goto allmulti;
2228 	}
2229 
2230 	/*
2231 	 * Set the station address in the first RAL slot, and
2232 	 * clear the remaining slots.
2233 	 */
2234 	dge_set_ral(sc, CLLADDR(ifp->if_sadl), 0);
2235 	for (i = 1; i < RA_TABSIZE; i++)
2236 		dge_set_ral(sc, NULL, i);
2237 
2238 	/* Clear out the multicast table. */
2239 	for (i = 0; i < MC_TABSIZE; i++)
2240 		CSR_WRITE(sc, DGE_MTA + (i << 2), 0);
2241 
2242 	ETHER_FIRST_MULTI(step, ec, enm);
2243 	while (enm != NULL) {
2244 		if (memcmp(enm->enm_addrlo, enm->enm_addrhi, ETHER_ADDR_LEN)) {
2245 			/*
2246 			 * We must listen to a range of multicast addresses.
2247 			 * For now, just accept all multicasts, rather than
2248 			 * trying to set only those filter bits needed to match
2249 			 * the range.  (At this time, the only use of address
2250 			 * ranges is for IP multicast routing, for which the
2251 			 * range is big enough to require all bits set.)
2252 			 */
2253 			goto allmulti;
2254 		}
2255 
2256 		hash = dge_mchash(sc, enm->enm_addrlo);
2257 
2258 		reg = (hash >> 5) & 0x7f;
2259 		bit = hash & 0x1f;
2260 
2261 		hash = CSR_READ(sc, DGE_MTA + (reg << 2));
2262 		hash |= 1U << bit;
2263 
2264 		CSR_WRITE(sc, DGE_MTA + (reg << 2), hash);
2265 
2266 		ETHER_NEXT_MULTI(step, enm);
2267 	}
2268 
2269 	ifp->if_flags &= ~IFF_ALLMULTI;
2270 	goto setit;
2271 
2272  allmulti:
2273 	ifp->if_flags |= IFF_ALLMULTI;
2274 	sc->sc_rctl |= RCTL_MPE;
2275 
2276  setit:
2277 	CSR_WRITE(sc, DGE_RCTL, sc->sc_rctl);
2278 }
2279 
2280 /*
2281  * Read in the EEPROM info and verify checksum.
2282  */
2283 int
2284 dge_read_eeprom(struct dge_softc *sc)
2285 {
2286 	uint16_t cksum;
2287 	int i;
2288 
2289 	cksum = 0;
2290 	for (i = 0; i < EEPROM_SIZE; i++) {
2291 		sc->sc_eeprom[i] = dge_eeprom_word(sc, i);
2292 		cksum += sc->sc_eeprom[i];
2293 	}
2294 	return cksum != EEPROM_CKSUM;
2295 }
2296 
2297 
2298 /*
2299  * Read a 16-bit word from address addr in the serial EEPROM.
2300  */
2301 uint16_t
2302 dge_eeprom_word(struct dge_softc *sc, int addr)
2303 {
2304 	uint32_t reg;
2305 	uint16_t rval = 0;
2306 	int i;
2307 
2308 	reg = CSR_READ(sc, DGE_EECD) & ~(EECD_SK|EECD_DI|EECD_CS);
2309 
2310 	/* Lower clock pulse (and data in to chip) */
2311 	CSR_WRITE(sc, DGE_EECD, reg);
2312 	/* Select chip */
2313 	CSR_WRITE(sc, DGE_EECD, reg|EECD_CS);
2314 
2315 	/* Send read command */
2316 	dge_eeprom_clockout(sc, 1);
2317 	dge_eeprom_clockout(sc, 1);
2318 	dge_eeprom_clockout(sc, 0);
2319 
2320 	/* Send address */
2321 	for (i = 5; i >= 0; i--)
2322 		dge_eeprom_clockout(sc, (addr >> i) & 1);
2323 
2324 	/* Read data */
2325 	for (i = 0; i < 16; i++) {
2326 		rval <<= 1;
2327 		rval |= dge_eeprom_clockin(sc);
2328 	}
2329 
2330 	/* Deselect chip */
2331 	CSR_WRITE(sc, DGE_EECD, reg);
2332 
2333 	return rval;
2334 }
2335 
2336 /*
2337  * Clock out a single bit to the EEPROM.
2338  */
2339 void
2340 dge_eeprom_clockout(struct dge_softc *sc, int bit)
2341 {
2342 	int reg;
2343 
2344 	reg = CSR_READ(sc, DGE_EECD) & ~(EECD_DI|EECD_SK);
2345 	if (bit)
2346 		reg |= EECD_DI;
2347 
2348 	CSR_WRITE(sc, DGE_EECD, reg);
2349 	delay(2);
2350 	CSR_WRITE(sc, DGE_EECD, reg|EECD_SK);
2351 	delay(2);
2352 	CSR_WRITE(sc, DGE_EECD, reg);
2353 	delay(2);
2354 }
2355 
2356 /*
2357  * Clock in a single bit from EEPROM.
2358  */
2359 int
2360 dge_eeprom_clockin(struct dge_softc *sc)
2361 {
2362 	int reg, rv;
2363 
2364 	reg = CSR_READ(sc, DGE_EECD) & ~(EECD_DI|EECD_DO|EECD_SK);
2365 
2366 	CSR_WRITE(sc, DGE_EECD, reg|EECD_SK); /* Raise clock */
2367 	delay(2);
2368 	rv = (CSR_READ(sc, DGE_EECD) & EECD_DO) != 0; /* Get bit */
2369 	CSR_WRITE(sc, DGE_EECD, reg); /* Lower clock */
2370 	delay(2);
2371 
2372 	return rv;
2373 }
2374 
2375 static void
2376 dge_xgmii_mediastatus(struct ifnet *ifp, struct ifmediareq *ifmr)
2377 {
2378 	struct dge_softc *sc = ifp->if_softc;
2379 
2380 	ifmr->ifm_status = IFM_AVALID;
2381 	ifmr->ifm_active = IFM_ETHER|IFM_10G_LR;
2382 
2383 	if (CSR_READ(sc, DGE_STATUS) & STATUS_LINKUP)
2384 		ifmr->ifm_status |= IFM_ACTIVE;
2385 }
2386 
2387 static inline int
2388 phwait(struct dge_softc *sc, int p, int r, int d, int type)
2389 {
2390         int i, mdic;
2391 
2392         CSR_WRITE(sc, DGE_MDIO,
2393 	    MDIO_PHY(p) | MDIO_REG(r) | MDIO_DEV(d) | type | MDIO_CMD);
2394         for (i = 0; i < 10; i++) {
2395                 delay(10);
2396                 if (((mdic = CSR_READ(sc, DGE_MDIO)) & MDIO_CMD) == 0)
2397                         break;
2398         }
2399         return mdic;
2400 }
2401 
2402 
2403 static void
2404 dge_xgmii_writereg(device_t self, int phy, int reg, int val)
2405 {
2406 	struct dge_softc *sc = device_private(self);
2407 	int mdic;
2408 
2409 	CSR_WRITE(sc, DGE_MDIRW, val);
2410 	if (((mdic = phwait(sc, phy, reg, 1, MDIO_ADDR)) & MDIO_CMD)) {
2411 		printf("%s: address cycle timeout; phy %d reg %d\n",
2412 		    device_xname(&sc->sc_dev), phy, reg);
2413 		return;
2414 	}
2415 	if (((mdic = phwait(sc, phy, reg, 1, MDIO_WRITE)) & MDIO_CMD)) {
2416 		printf("%s: read cycle timeout; phy %d reg %d\n",
2417 		    device_xname(&sc->sc_dev), phy, reg);
2418 		return;
2419 	}
2420 }
2421 
2422 static void
2423 dge_xgmii_reset(struct dge_softc *sc)
2424 {
2425 	dge_xgmii_writereg((void *)sc, 0, 0, BMCR_RESET);
2426 }
2427 
2428 static int
2429 dge_xgmii_mediachange(struct ifnet *ifp)
2430 {
2431 	return 0;
2432 }
2433