1 /*	$NetBSD: if_dge.c,v 1.45 2016/07/07 06:55:41 msaitoh 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 1.45 2016/07/07 06:55:41 msaitoh Exp $");
84 
85 
86 
87 #include <sys/param.h>
88 #include <sys/systm.h>
89 #include <sys/callout.h>
90 #include <sys/mbuf.h>
91 #include <sys/malloc.h>
92 #include <sys/kernel.h>
93 #include <sys/socket.h>
94 #include <sys/ioctl.h>
95 #include <sys/errno.h>
96 #include <sys/device.h>
97 #include <sys/queue.h>
98 
99 #include <sys/rndsource.h>
100 
101 #include <net/if.h>
102 #include <net/if_dl.h>
103 #include <net/if_media.h>
104 #include <net/if_ether.h>
105 
106 #include <net/bpf.h>
107 
108 #include <netinet/in.h>			/* XXX for struct ip */
109 #include <netinet/in_systm.h>		/* XXX for struct ip */
110 #include <netinet/ip.h>			/* XXX for struct ip */
111 #include <netinet/tcp.h>		/* XXX for struct tcphdr */
112 
113 #include <sys/bus.h>
114 #include <sys/intr.h>
115 #include <machine/endian.h>
116 
117 #include <dev/mii/mii.h>
118 #include <dev/mii/miivar.h>
119 #include <dev/mii/mii_bitbang.h>
120 
121 #include <dev/pci/pcireg.h>
122 #include <dev/pci/pcivar.h>
123 #include <dev/pci/pcidevs.h>
124 
125 #include <dev/pci/if_dgereg.h>
126 
127 /*
128  * The receive engine may sometimes become off-by-one when writing back
129  * chained descriptors.	 Avoid this by allocating a large chunk of
130  * memory and use if instead (to avoid chained descriptors).
131  * This only happens with chained descriptors under heavy load.
132  */
133 #define DGE_OFFBYONE_RXBUG
134 
135 #define DGE_EVENT_COUNTERS
136 #define DGE_DEBUG
137 
138 #ifdef DGE_DEBUG
139 #define DGE_DEBUG_LINK		0x01
140 #define DGE_DEBUG_TX		0x02
141 #define DGE_DEBUG_RX		0x04
142 #define DGE_DEBUG_CKSUM		0x08
143 int	dge_debug = 0;
144 
145 #define DPRINTF(x, y)	if (dge_debug & (x)) printf y
146 #else
147 #define DPRINTF(x, y)	/* nothing */
148 #endif /* DGE_DEBUG */
149 
150 /*
151  * Transmit descriptor list size. We allow up to 100 DMA segments per
152  * packet (Intel reports of jumbo frame packets with as
153  * many as 80 DMA segments when using 16k buffers).
154  */
155 #define DGE_NTXSEGS		100
156 #define DGE_IFQUEUELEN		20000
157 #define DGE_TXQUEUELEN		2048
158 #define DGE_TXQUEUELEN_MASK	(DGE_TXQUEUELEN - 1)
159 #define DGE_TXQUEUE_GC		(DGE_TXQUEUELEN / 8)
160 #define DGE_NTXDESC		1024
161 #define DGE_NTXDESC_MASK		(DGE_NTXDESC - 1)
162 #define DGE_NEXTTX(x)		(((x) + 1) & DGE_NTXDESC_MASK)
163 #define DGE_NEXTTXS(x)		(((x) + 1) & DGE_TXQUEUELEN_MASK)
164 
165 /*
166  * Receive descriptor list size.
167  * Packet is of size MCLBYTES, and for jumbo packets buffers may
168  * be chained.	Due to the nature of the card (high-speed), keep this
169  * ring large. With 2k buffers the ring can store 400 jumbo packets,
170  * which at full speed will be received in just under 3ms.
171  */
172 #define DGE_NRXDESC		2048
173 #define DGE_NRXDESC_MASK	(DGE_NRXDESC - 1)
174 #define DGE_NEXTRX(x)		(((x) + 1) & DGE_NRXDESC_MASK)
175 /*
176  * # of descriptors between head and written descriptors.
177  * This is to work-around two erratas.
178  */
179 #define DGE_RXSPACE		10
180 #define DGE_PREVRX(x)		(((x) - DGE_RXSPACE) & DGE_NRXDESC_MASK)
181 /*
182  * Receive descriptor fetch threshholds. These are values recommended
183  * by Intel, do not touch them unless you know what you are doing.
184  */
185 #define RXDCTL_PTHRESH_VAL	128
186 #define RXDCTL_HTHRESH_VAL	16
187 #define RXDCTL_WTHRESH_VAL	16
188 
189 
190 /*
191  * Tweakable parameters; default values.
192  */
193 #define FCRTH	0x30000 /* Send XOFF water mark */
194 #define FCRTL	0x28000 /* Send XON water mark */
195 #define RDTR	0x20	/* Interrupt delay after receive, .8192us units */
196 #define TIDV	0x20	/* Interrupt delay after send, .8192us units */
197 
198 /*
199  * Control structures are DMA'd to the i82597 chip.  We allocate them in
200  * a single clump that maps to a single DMA segment to make serveral things
201  * easier.
202  */
203 struct dge_control_data {
204 	/*
205 	 * The transmit descriptors.
206 	 */
207 	struct dge_tdes wcd_txdescs[DGE_NTXDESC];
208 
209 	/*
210 	 * The receive descriptors.
211 	 */
212 	struct dge_rdes wcd_rxdescs[DGE_NRXDESC];
213 };
214 
215 #define DGE_CDOFF(x)	offsetof(struct dge_control_data, x)
216 #define DGE_CDTXOFF(x)	DGE_CDOFF(wcd_txdescs[(x)])
217 #define DGE_CDRXOFF(x)	DGE_CDOFF(wcd_rxdescs[(x)])
218 
219 /*
220  * The DGE interface have a higher max MTU size than normal jumbo frames.
221  */
222 #define DGE_MAX_MTU	16288	/* Max MTU size for this interface */
223 
224 /*
225  * Software state for transmit jobs.
226  */
227 struct dge_txsoft {
228 	struct mbuf *txs_mbuf;		/* head of our mbuf chain */
229 	bus_dmamap_t txs_dmamap;	/* our DMA map */
230 	int txs_firstdesc;		/* first descriptor in packet */
231 	int txs_lastdesc;		/* last descriptor in packet */
232 	int txs_ndesc;			/* # of descriptors used */
233 };
234 
235 /*
236  * Software state for receive buffers.	Each descriptor gets a
237  * 2k (MCLBYTES) buffer and a DMA map.	For packets which fill
238  * more than one buffer, we chain them together.
239  */
240 struct dge_rxsoft {
241 	struct mbuf *rxs_mbuf;		/* head of our mbuf chain */
242 	bus_dmamap_t rxs_dmamap;	/* our DMA map */
243 };
244 
245 /*
246  * Software state per device.
247  */
248 struct dge_softc {
249 	device_t sc_dev;		/* generic device information */
250 	bus_space_tag_t sc_st;		/* bus space tag */
251 	bus_space_handle_t sc_sh;	/* bus space handle */
252 	bus_dma_tag_t sc_dmat;		/* bus DMA tag */
253 	struct ethercom sc_ethercom;	/* ethernet common data */
254 
255 	int sc_flags;			/* flags; see below */
256 	int sc_bus_speed;		/* PCI/PCIX bus speed */
257 	int sc_pcix_offset;		/* PCIX capability register offset */
258 
259 	const struct dge_product *sc_dgep; /* Pointer to the dge_product entry */
260 	pci_chipset_tag_t sc_pc;
261 	pcitag_t sc_pt;
262 	int sc_mmrbc;			/* Max PCIX memory read byte count */
263 
264 	void *sc_ih;			/* interrupt cookie */
265 
266 	struct ifmedia sc_media;
267 
268 	bus_dmamap_t sc_cddmamap;	/* control data DMA map */
269 #define sc_cddma	sc_cddmamap->dm_segs[0].ds_addr
270 
271 	int		sc_align_tweak;
272 
273 	/*
274 	 * Software state for the transmit and receive descriptors.
275 	 */
276 	struct dge_txsoft sc_txsoft[DGE_TXQUEUELEN];
277 	struct dge_rxsoft sc_rxsoft[DGE_NRXDESC];
278 
279 	/*
280 	 * Control data structures.
281 	 */
282 	struct dge_control_data *sc_control_data;
283 #define sc_txdescs	sc_control_data->wcd_txdescs
284 #define sc_rxdescs	sc_control_data->wcd_rxdescs
285 
286 #ifdef DGE_EVENT_COUNTERS
287 	/* Event counters. */
288 	struct evcnt sc_ev_txsstall;	/* Tx stalled due to no txs */
289 	struct evcnt sc_ev_txdstall;	/* Tx stalled due to no txd */
290 	struct evcnt sc_ev_txforceintr; /* Tx interrupts forced */
291 	struct evcnt sc_ev_txdw;	/* Tx descriptor interrupts */
292 	struct evcnt sc_ev_txqe;	/* Tx queue empty interrupts */
293 	struct evcnt sc_ev_rxintr;	/* Rx interrupts */
294 	struct evcnt sc_ev_linkintr;	/* Link interrupts */
295 
296 	struct evcnt sc_ev_rxipsum;	/* IP checksums checked in-bound */
297 	struct evcnt sc_ev_rxtusum;	/* TCP/UDP cksums checked in-bound */
298 	struct evcnt sc_ev_txipsum;	/* IP checksums comp. out-bound */
299 	struct evcnt sc_ev_txtusum;	/* TCP/UDP cksums comp. out-bound */
300 
301 	struct evcnt sc_ev_txctx_init;	/* Tx cksum context cache initialized */
302 	struct evcnt sc_ev_txctx_hit;	/* Tx cksum context cache hit */
303 	struct evcnt sc_ev_txctx_miss;	/* Tx cksum context cache miss */
304 
305 	struct evcnt sc_ev_txseg[DGE_NTXSEGS]; /* Tx packets w/ N segments */
306 	struct evcnt sc_ev_txdrop;	/* Tx packets dropped (too many segs) */
307 #endif /* DGE_EVENT_COUNTERS */
308 
309 	int	sc_txfree;		/* number of free Tx descriptors */
310 	int	sc_txnext;		/* next ready Tx descriptor */
311 
312 	int	sc_txsfree;		/* number of free Tx jobs */
313 	int	sc_txsnext;		/* next free Tx job */
314 	int	sc_txsdirty;		/* dirty Tx jobs */
315 
316 	uint32_t sc_txctx_ipcs;		/* cached Tx IP cksum ctx */
317 	uint32_t sc_txctx_tucs;		/* cached Tx TCP/UDP cksum ctx */
318 
319 	int	sc_rxptr;		/* next ready Rx descriptor/queue ent */
320 	int	sc_rxdiscard;
321 	int	sc_rxlen;
322 	struct mbuf *sc_rxhead;
323 	struct mbuf *sc_rxtail;
324 	struct mbuf **sc_rxtailp;
325 
326 	uint32_t sc_ctrl0;		/* prototype CTRL0 register */
327 	uint32_t sc_icr;		/* prototype interrupt bits */
328 	uint32_t sc_tctl;		/* prototype TCTL register */
329 	uint32_t sc_rctl;		/* prototype RCTL register */
330 
331 	int sc_mchash_type;		/* multicast filter offset */
332 
333 	uint16_t sc_eeprom[EEPROM_SIZE];
334 
335 	krndsource_t rnd_source; /* random source */
336 #ifdef DGE_OFFBYONE_RXBUG
337 	void *sc_bugbuf;
338 	SLIST_HEAD(, rxbugentry) sc_buglist;
339 	bus_dmamap_t sc_bugmap;
340 	struct rxbugentry *sc_entry;
341 #endif
342 };
343 
344 #define DGE_RXCHAIN_RESET(sc)						\
345 do {									\
346 	(sc)->sc_rxtailp = &(sc)->sc_rxhead;				\
347 	*(sc)->sc_rxtailp = NULL;					\
348 	(sc)->sc_rxlen = 0;						\
349 } while (/*CONSTCOND*/0)
350 
351 #define DGE_RXCHAIN_LINK(sc, m)						\
352 do {									\
353 	*(sc)->sc_rxtailp = (sc)->sc_rxtail = (m);			\
354 	(sc)->sc_rxtailp = &(m)->m_next;				\
355 } while (/*CONSTCOND*/0)
356 
357 /* sc_flags */
358 #define DGE_F_BUS64		0x20	/* bus is 64-bit */
359 #define DGE_F_PCIX		0x40	/* bus is PCI-X */
360 
361 #ifdef DGE_EVENT_COUNTERS
362 #define DGE_EVCNT_INCR(ev)	(ev)->ev_count++
363 #else
364 #define DGE_EVCNT_INCR(ev)	/* nothing */
365 #endif
366 
367 #define CSR_READ(sc, reg)						\
368 	bus_space_read_4((sc)->sc_st, (sc)->sc_sh, (reg))
369 #define CSR_WRITE(sc, reg, val)						\
370 	bus_space_write_4((sc)->sc_st, (sc)->sc_sh, (reg), (val))
371 
372 #define DGE_CDTXADDR(sc, x)	((sc)->sc_cddma + DGE_CDTXOFF((x)))
373 #define DGE_CDRXADDR(sc, x)	((sc)->sc_cddma + DGE_CDRXOFF((x)))
374 
375 #define DGE_CDTXSYNC(sc, x, n, ops)					\
376 do {									\
377 	int __x, __n;							\
378 									\
379 	__x = (x);							\
380 	__n = (n);							\
381 									\
382 	/* If it will wrap around, sync to the end of the ring. */	\
383 	if ((__x + __n) > DGE_NTXDESC) {				\
384 		bus_dmamap_sync((sc)->sc_dmat, (sc)->sc_cddmamap,	\
385 		    DGE_CDTXOFF(__x), sizeof(struct dge_tdes) *		\
386 		    (DGE_NTXDESC - __x), (ops));			\
387 		__n -= (DGE_NTXDESC - __x);				\
388 		__x = 0;						\
389 	}								\
390 									\
391 	/* Now sync whatever is left. */				\
392 	bus_dmamap_sync((sc)->sc_dmat, (sc)->sc_cddmamap,		\
393 	    DGE_CDTXOFF(__x), sizeof(struct dge_tdes) * __n, (ops));	\
394 } while (/*CONSTCOND*/0)
395 
396 #define DGE_CDRXSYNC(sc, x, ops)						\
397 do {									\
398 	bus_dmamap_sync((sc)->sc_dmat, (sc)->sc_cddmamap,		\
399 	   DGE_CDRXOFF((x)), sizeof(struct dge_rdes), (ops));		\
400 } while (/*CONSTCOND*/0)
401 
402 #ifdef DGE_OFFBYONE_RXBUG
403 #define DGE_INIT_RXDESC(sc, x)						\
404 do {									\
405 	struct dge_rxsoft *__rxs = &(sc)->sc_rxsoft[(x)];		\
406 	struct dge_rdes *__rxd = &(sc)->sc_rxdescs[(x)];		\
407 	struct mbuf *__m = __rxs->rxs_mbuf;				\
408 									\
409 	__rxd->dr_baddrl = htole32(sc->sc_bugmap->dm_segs[0].ds_addr +	\
410 	    (mtod((__m), char *) - (char *)sc->sc_bugbuf));		\
411 	__rxd->dr_baddrh = 0;						\
412 	__rxd->dr_len = 0;						\
413 	__rxd->dr_cksum = 0;						\
414 	__rxd->dr_status = 0;						\
415 	__rxd->dr_errors = 0;						\
416 	__rxd->dr_special = 0;						\
417 	DGE_CDRXSYNC((sc), (x), BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE); \
418 									\
419 	CSR_WRITE((sc), DGE_RDT, (x));					\
420 } while (/*CONSTCOND*/0)
421 #else
422 #define DGE_INIT_RXDESC(sc, x)						\
423 do {									\
424 	struct dge_rxsoft *__rxs = &(sc)->sc_rxsoft[(x)];		\
425 	struct dge_rdes *__rxd = &(sc)->sc_rxdescs[(x)];		\
426 	struct mbuf *__m = __rxs->rxs_mbuf;				\
427 									\
428 	/*								\
429 	 * Note: We scoot the packet forward 2 bytes in the buffer	\
430 	 * so that the payload after the Ethernet header is aligned	\
431 	 * to a 4-byte boundary.					\
432 	 *								\
433 	 * XXX BRAINDAMAGE ALERT!					\
434 	 * The stupid chip uses the same size for every buffer, which	\
435 	 * is set in the Receive Control register.  We are using the 2K \
436 	 * size option, but what we REALLY want is (2K - 2)!  For this	\
437 	 * reason, we can't "scoot" packets longer than the standard	\
438 	 * Ethernet MTU.  On strict-alignment platforms, if the total	\
439 	 * size exceeds (2K - 2) we set align_tweak to 0 and let	\
440 	 * the upper layer copy the headers.				\
441 	 */								\
442 	__m->m_data = __m->m_ext.ext_buf + (sc)->sc_align_tweak;	\
443 									\
444 	__rxd->dr_baddrl =						\
445 	    htole32(__rxs->rxs_dmamap->dm_segs[0].ds_addr +		\
446 		(sc)->sc_align_tweak);					\
447 	__rxd->dr_baddrh = 0;						\
448 	__rxd->dr_len = 0;						\
449 	__rxd->dr_cksum = 0;						\
450 	__rxd->dr_status = 0;						\
451 	__rxd->dr_errors = 0;						\
452 	__rxd->dr_special = 0;						\
453 	DGE_CDRXSYNC((sc), (x), BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE); \
454 									\
455 	CSR_WRITE((sc), DGE_RDT, (x));					\
456 } while (/*CONSTCOND*/0)
457 #endif
458 
459 #ifdef DGE_OFFBYONE_RXBUG
460 /*
461  * Allocation constants.  Much memory may be used for this.
462  */
463 #ifndef DGE_BUFFER_SIZE
464 #define DGE_BUFFER_SIZE DGE_MAX_MTU
465 #endif
466 #define DGE_NBUFFERS	(4*DGE_NRXDESC)
467 #define DGE_RXMEM	(DGE_NBUFFERS*DGE_BUFFER_SIZE)
468 
469 struct rxbugentry {
470 	SLIST_ENTRY(rxbugentry) rb_entry;
471 	int rb_slot;
472 };
473 
474 static int
dge_alloc_rcvmem(struct dge_softc * sc)475 dge_alloc_rcvmem(struct dge_softc *sc)
476 {
477 	char *kva;
478 	bus_dma_segment_t seg;
479 	int i, rseg, state, error;
480 	struct rxbugentry *entry;
481 
482 	state = error = 0;
483 
484 	if (bus_dmamem_alloc(sc->sc_dmat, DGE_RXMEM, PAGE_SIZE, 0,
485 	     &seg, 1, &rseg, BUS_DMA_NOWAIT)) {
486 		aprint_error_dev(sc->sc_dev, "can't alloc rx buffers\n");
487 		return ENOBUFS;
488 	}
489 
490 	state = 1;
491 	if (bus_dmamem_map(sc->sc_dmat, &seg, rseg, DGE_RXMEM, (void **)&kva,
492 	    BUS_DMA_NOWAIT)) {
493 		aprint_error_dev(sc->sc_dev, "can't map DMA buffers (%d bytes)\n",
494 		    (int)DGE_RXMEM);
495 		error = ENOBUFS;
496 		goto out;
497 	}
498 
499 	state = 2;
500 	if (bus_dmamap_create(sc->sc_dmat, DGE_RXMEM, 1, DGE_RXMEM, 0,
501 	    BUS_DMA_NOWAIT, &sc->sc_bugmap)) {
502 		aprint_error_dev(sc->sc_dev, "can't create DMA map\n");
503 		error = ENOBUFS;
504 		goto out;
505 	}
506 
507 	state = 3;
508 	if (bus_dmamap_load(sc->sc_dmat, sc->sc_bugmap,
509 	    kva, DGE_RXMEM, NULL, BUS_DMA_NOWAIT)) {
510 		aprint_error_dev(sc->sc_dev, "can't load DMA map\n");
511 		error = ENOBUFS;
512 		goto out;
513 	}
514 
515 	state = 4;
516 	sc->sc_bugbuf = (void *)kva;
517 	SLIST_INIT(&sc->sc_buglist);
518 
519 	/*
520 	 * Now divide it up into DGE_BUFFER_SIZE pieces and save the addresses
521 	 * in an array.
522 	 */
523 	if ((entry = malloc(sizeof(*entry) * DGE_NBUFFERS,
524 	    M_DEVBUF, M_NOWAIT)) == NULL) {
525 		error = ENOBUFS;
526 		goto out;
527 	}
528 	sc->sc_entry = entry;
529 	for (i = 0; i < DGE_NBUFFERS; i++) {
530 		entry[i].rb_slot = i;
531 		SLIST_INSERT_HEAD(&sc->sc_buglist, &entry[i], rb_entry);
532 	}
533 out:
534 	if (error != 0) {
535 		switch (state) {
536 		case 4:
537 			bus_dmamap_unload(sc->sc_dmat, sc->sc_bugmap);
538 		case 3:
539 			bus_dmamap_destroy(sc->sc_dmat, sc->sc_bugmap);
540 		case 2:
541 			bus_dmamem_unmap(sc->sc_dmat, kva, DGE_RXMEM);
542 		case 1:
543 			bus_dmamem_free(sc->sc_dmat, &seg, rseg);
544 			break;
545 		default:
546 			break;
547 		}
548 	}
549 
550 	return error;
551 }
552 
553 /*
554  * Allocate a jumbo buffer.
555  */
556 static void *
dge_getbuf(struct dge_softc * sc)557 dge_getbuf(struct dge_softc *sc)
558 {
559 	struct rxbugentry *entry;
560 
561 	entry = SLIST_FIRST(&sc->sc_buglist);
562 
563 	if (entry == NULL) {
564 		printf("%s: no free RX buffers\n", device_xname(sc->sc_dev));
565 		return(NULL);
566 	}
567 
568 	SLIST_REMOVE_HEAD(&sc->sc_buglist, rb_entry);
569 	return (char *)sc->sc_bugbuf + entry->rb_slot * DGE_BUFFER_SIZE;
570 }
571 
572 /*
573  * Release a jumbo buffer.
574  */
575 static void
dge_freebuf(struct mbuf * m,void * buf,size_t size,void * arg)576 dge_freebuf(struct mbuf *m, void *buf, size_t size, void *arg)
577 {
578 	struct rxbugentry *entry;
579 	struct dge_softc *sc;
580 	int i, s;
581 
582 	/* Extract the softc struct pointer. */
583 	sc = (struct dge_softc *)arg;
584 
585 	if (sc == NULL)
586 		panic("dge_freebuf: can't find softc pointer!");
587 
588 	/* calculate the slot this buffer belongs to */
589 
590 	i = ((char *)buf - (char *)sc->sc_bugbuf) / DGE_BUFFER_SIZE;
591 
592 	if ((i < 0) || (i >= DGE_NBUFFERS))
593 		panic("dge_freebuf: asked to free buffer %d!", i);
594 
595 	s = splvm();
596 	entry = sc->sc_entry + i;
597 	SLIST_INSERT_HEAD(&sc->sc_buglist, entry, rb_entry);
598 
599 	if (__predict_true(m != NULL))
600 		pool_cache_put(mb_cache, m);
601 	splx(s);
602 }
603 #endif
604 
605 static void	dge_start(struct ifnet *);
606 static void	dge_watchdog(struct ifnet *);
607 static int	dge_ioctl(struct ifnet *, u_long, void *);
608 static int	dge_init(struct ifnet *);
609 static void	dge_stop(struct ifnet *, int);
610 
611 static bool	dge_shutdown(device_t, int);
612 
613 static void	dge_reset(struct dge_softc *);
614 static void	dge_rxdrain(struct dge_softc *);
615 static int	dge_add_rxbuf(struct dge_softc *, int);
616 
617 static void	dge_set_filter(struct dge_softc *);
618 
619 static int	dge_intr(void *);
620 static void	dge_txintr(struct dge_softc *);
621 static void	dge_rxintr(struct dge_softc *);
622 static void	dge_linkintr(struct dge_softc *, uint32_t);
623 
624 static int	dge_match(device_t, cfdata_t, void *);
625 static void	dge_attach(device_t, device_t, void *);
626 
627 static int	dge_read_eeprom(struct dge_softc *sc);
628 static int	dge_eeprom_clockin(struct dge_softc *sc);
629 static void	dge_eeprom_clockout(struct dge_softc *sc, int bit);
630 static uint16_t	dge_eeprom_word(struct dge_softc *sc, int addr);
631 static int	dge_xgmii_mediachange(struct ifnet *);
632 static void	dge_xgmii_mediastatus(struct ifnet *, struct ifmediareq *);
633 static void	dge_xgmii_reset(struct dge_softc *);
634 static void	dge_xgmii_writereg(struct dge_softc *, int, int, int);
635 
636 
637 CFATTACH_DECL_NEW(dge, sizeof(struct dge_softc),
638     dge_match, dge_attach, NULL, NULL);
639 
640 #ifdef DGE_EVENT_COUNTERS
641 #if DGE_NTXSEGS > 100
642 #error Update dge_txseg_evcnt_names
643 #endif
644 static char (*dge_txseg_evcnt_names)[DGE_NTXSEGS][8 /* "txseg00" + \0 */];
645 #endif /* DGE_EVENT_COUNTERS */
646 
647 /*
648  * Devices supported by this driver.
649  */
650 static const struct dge_product {
651   pci_vendor_id_t      dgep_vendor;
652   pci_product_id_t  dgep_product;
653   const char     *dgep_name;
654   int         dgep_flags;
655 #define DGEP_F_10G_LR     0x01
656 #define DGEP_F_10G_SR     0x02
657 } dge_products[] = {
658   { PCI_VENDOR_INTEL,  PCI_PRODUCT_INTEL_82597EX,
659     "Intel i82597EX 10GbE-LR Ethernet",
660     DGEP_F_10G_LR },
661 
662   { PCI_VENDOR_INTEL,  PCI_PRODUCT_INTEL_82597EX_SR,
663     "Intel i82597EX 10GbE-SR Ethernet",
664     DGEP_F_10G_SR },
665 
666   { 0,        0,
667     NULL,
668     0 },
669 };
670 
671 static const struct dge_product *
dge_lookup(const struct pci_attach_args * pa)672 dge_lookup(const struct pci_attach_args *pa)
673 {
674 	const struct dge_product *dgep;
675 
676 	for (dgep = dge_products; dgep->dgep_name != NULL; dgep++) {
677 		if (PCI_VENDOR(pa->pa_id) == dgep->dgep_vendor &&
678 		    PCI_PRODUCT(pa->pa_id) == dgep->dgep_product)
679 			return dgep;
680 		}
681 	return NULL;
682 }
683 
684 static int
dge_match(device_t parent,cfdata_t cf,void * aux)685 dge_match(device_t parent, cfdata_t cf, void *aux)
686 {
687 	struct pci_attach_args *pa = aux;
688 
689 	if (dge_lookup(pa) != NULL)
690 		return (1);
691 
692 	return (0);
693 }
694 
695 static void
dge_attach(device_t parent,device_t self,void * aux)696 dge_attach(device_t parent, device_t self, void *aux)
697 {
698 	struct dge_softc *sc = device_private(self);
699 	struct pci_attach_args *pa = aux;
700 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
701 	pci_chipset_tag_t pc = pa->pa_pc;
702 	pci_intr_handle_t ih;
703 	const char *intrstr = NULL;
704 	bus_dma_segment_t seg;
705 	int i, rseg, error;
706 	uint8_t enaddr[ETHER_ADDR_LEN];
707 	pcireg_t preg, memtype;
708 	uint32_t reg;
709 	char intrbuf[PCI_INTRSTR_LEN];
710 	const struct dge_product *dgep;
711 
712 	sc->sc_dgep = dgep = dge_lookup(pa);
713 	if (dgep == NULL) {
714 		printf("\n");
715 		panic("dge_attach: impossible");
716 	}
717 
718 	sc->sc_dev = self;
719 	sc->sc_dmat = pa->pa_dmat;
720 	sc->sc_pc = pa->pa_pc;
721 	sc->sc_pt = pa->pa_tag;
722 
723 	pci_aprint_devinfo_fancy(pa, "Ethernet controller",
724 		dgep->dgep_name, 1);
725 
726 	memtype = pci_mapreg_type(pa->pa_pc, pa->pa_tag, DGE_PCI_BAR);
727         if (pci_mapreg_map(pa, DGE_PCI_BAR, memtype, 0,
728             &sc->sc_st, &sc->sc_sh, NULL, NULL)) {
729                 aprint_error_dev(sc->sc_dev,
730 		    "unable to map device registers\n");
731                 return;
732         }
733 
734 	/* Enable bus mastering */
735 	preg = pci_conf_read(pc, pa->pa_tag, PCI_COMMAND_STATUS_REG);
736 	preg |= PCI_COMMAND_MASTER_ENABLE;
737 	pci_conf_write(pc, pa->pa_tag, PCI_COMMAND_STATUS_REG, preg);
738 
739 	/*
740 	 * Map and establish our interrupt.
741 	 */
742 	if (pci_intr_map(pa, &ih)) {
743 		aprint_error_dev(sc->sc_dev, "unable to map interrupt\n");
744 		return;
745 	}
746 	intrstr = pci_intr_string(pc, ih, intrbuf, sizeof(intrbuf));
747 	sc->sc_ih = pci_intr_establish(pc, ih, IPL_NET, dge_intr, sc);
748 	if (sc->sc_ih == NULL) {
749 		aprint_error_dev(sc->sc_dev, "unable to establish interrupt");
750 		if (intrstr != NULL)
751 			aprint_error(" at %s", intrstr);
752 		aprint_error("\n");
753 		return;
754 	}
755 	aprint_normal_dev(sc->sc_dev, "interrupting at %s\n", intrstr);
756 
757 	/*
758 	 * Determine a few things about the bus we're connected to.
759 	 */
760 	reg = CSR_READ(sc, DGE_STATUS);
761 	if (reg & STATUS_BUS64)
762 		sc->sc_flags |= DGE_F_BUS64;
763 
764 	sc->sc_flags |= DGE_F_PCIX;
765 	if (pci_get_capability(pa->pa_pc, pa->pa_tag,
766 			       PCI_CAP_PCIX,
767 			       &sc->sc_pcix_offset, NULL) == 0)
768 		aprint_error_dev(sc->sc_dev, "unable to find PCIX "
769 		    "capability\n");
770 
771 	if (sc->sc_flags & DGE_F_PCIX) {
772 		switch (reg & STATUS_PCIX_MSK) {
773 		case STATUS_PCIX_66:
774 			sc->sc_bus_speed = 66;
775 			break;
776 		case STATUS_PCIX_100:
777 			sc->sc_bus_speed = 100;
778 			break;
779 		case STATUS_PCIX_133:
780 			sc->sc_bus_speed = 133;
781 			break;
782 		default:
783 			aprint_error_dev(sc->sc_dev,
784 			    "unknown PCIXSPD %d; assuming 66MHz\n",
785 			    reg & STATUS_PCIX_MSK);
786 			sc->sc_bus_speed = 66;
787 		}
788 	} else
789 		sc->sc_bus_speed = (reg & STATUS_BUS64) ? 66 : 33;
790 	aprint_verbose_dev(sc->sc_dev, "%d-bit %dMHz %s bus\n",
791 	    (sc->sc_flags & DGE_F_BUS64) ? 64 : 32, sc->sc_bus_speed,
792 	    (sc->sc_flags & DGE_F_PCIX) ? "PCIX" : "PCI");
793 
794 	/*
795 	 * Allocate the control data structures, and create and load the
796 	 * DMA map for it.
797 	 */
798 	if ((error = bus_dmamem_alloc(sc->sc_dmat,
799 	    sizeof(struct dge_control_data), PAGE_SIZE, 0, &seg, 1, &rseg,
800 	    0)) != 0) {
801 		aprint_error_dev(sc->sc_dev,
802 		    "unable to allocate control data, error = %d\n",
803 		    error);
804 		goto fail_0;
805 	}
806 
807 	if ((error = bus_dmamem_map(sc->sc_dmat, &seg, rseg,
808 	    sizeof(struct dge_control_data), (void **)&sc->sc_control_data,
809 	    0)) != 0) {
810 		aprint_error_dev(sc->sc_dev, "unable to map control data, error = %d\n",
811 		    error);
812 		goto fail_1;
813 	}
814 
815 	if ((error = bus_dmamap_create(sc->sc_dmat,
816 	    sizeof(struct dge_control_data), 1,
817 	    sizeof(struct dge_control_data), 0, 0, &sc->sc_cddmamap)) != 0) {
818 		aprint_error_dev(sc->sc_dev, "unable to create control data DMA map, "
819 		    "error = %d\n", error);
820 		goto fail_2;
821 	}
822 
823 	if ((error = bus_dmamap_load(sc->sc_dmat, sc->sc_cddmamap,
824 	    sc->sc_control_data, sizeof(struct dge_control_data), NULL,
825 	    0)) != 0) {
826 		aprint_error_dev(sc->sc_dev,
827 		    "unable to load control data DMA map, error = %d\n",
828 		    error);
829 		goto fail_3;
830 	}
831 
832 #ifdef DGE_OFFBYONE_RXBUG
833 	if (dge_alloc_rcvmem(sc) != 0)
834 		return; /* Already complained */
835 #endif
836 	/*
837 	 * Create the transmit buffer DMA maps.
838 	 */
839 	for (i = 0; i < DGE_TXQUEUELEN; i++) {
840 		if ((error = bus_dmamap_create(sc->sc_dmat, DGE_MAX_MTU,
841 		    DGE_NTXSEGS, MCLBYTES, 0, 0,
842 		    &sc->sc_txsoft[i].txs_dmamap)) != 0) {
843 			aprint_error_dev(sc->sc_dev, "unable to create Tx DMA map %d, "
844 			    "error = %d\n", i, error);
845 			goto fail_4;
846 		}
847 	}
848 
849 	/*
850 	 * Create the receive buffer DMA maps.
851 	 */
852 	for (i = 0; i < DGE_NRXDESC; i++) {
853 #ifdef DGE_OFFBYONE_RXBUG
854 		if ((error = bus_dmamap_create(sc->sc_dmat, DGE_BUFFER_SIZE, 1,
855 		    DGE_BUFFER_SIZE, 0, 0, &sc->sc_rxsoft[i].rxs_dmamap)) != 0) {
856 #else
857 		if ((error = bus_dmamap_create(sc->sc_dmat, MCLBYTES, 1,
858 		    MCLBYTES, 0, 0, &sc->sc_rxsoft[i].rxs_dmamap)) != 0) {
859 #endif
860 			aprint_error_dev(sc->sc_dev, "unable to create Rx DMA map %d, "
861 			    "error = %d\n", i, error);
862 			goto fail_5;
863 		}
864 		sc->sc_rxsoft[i].rxs_mbuf = NULL;
865 	}
866 
867 	/*
868 	 * Set bits in ctrl0 register.
869 	 * Should get the software defined pins out of EEPROM?
870 	 */
871 	sc->sc_ctrl0 |= CTRL0_RPE | CTRL0_TPE; /* XON/XOFF */
872 	sc->sc_ctrl0 |= CTRL0_SDP3_DIR | CTRL0_SDP2_DIR | CTRL0_SDP1_DIR |
873 	    CTRL0_SDP0_DIR | CTRL0_SDP3 | CTRL0_SDP2 | CTRL0_SDP0;
874 
875 	/*
876 	 * Reset the chip to a known state.
877 	 */
878 	dge_reset(sc);
879 
880 	/*
881 	 * Reset the PHY.
882 	 */
883 	dge_xgmii_reset(sc);
884 
885 	/*
886 	 * Read in EEPROM data.
887 	 */
888 	if (dge_read_eeprom(sc)) {
889 		aprint_error_dev(sc->sc_dev, "couldn't read EEPROM\n");
890 		return;
891 	}
892 
893 	/*
894 	 * Get the ethernet address.
895 	 */
896 	enaddr[0] = sc->sc_eeprom[EE_ADDR01] & 0377;
897 	enaddr[1] = sc->sc_eeprom[EE_ADDR01] >> 8;
898 	enaddr[2] = sc->sc_eeprom[EE_ADDR23] & 0377;
899 	enaddr[3] = sc->sc_eeprom[EE_ADDR23] >> 8;
900 	enaddr[4] = sc->sc_eeprom[EE_ADDR45] & 0377;
901 	enaddr[5] = sc->sc_eeprom[EE_ADDR45] >> 8;
902 
903 	aprint_normal_dev(sc->sc_dev, "Ethernet address %s\n",
904 	    ether_sprintf(enaddr));
905 
906 	/*
907 	 * Setup media stuff.
908 	 */
909         ifmedia_init(&sc->sc_media, IFM_IMASK, dge_xgmii_mediachange,
910             dge_xgmii_mediastatus);
911 	if (dgep->dgep_flags & DGEP_F_10G_SR) {
912 		ifmedia_add(&sc->sc_media, IFM_ETHER|IFM_10G_SR, 0, NULL);
913 		ifmedia_set(&sc->sc_media, IFM_ETHER|IFM_10G_SR);
914 	} else { /* XXX default is LR */
915 		ifmedia_add(&sc->sc_media, IFM_ETHER|IFM_10G_LR, 0, NULL);
916 		ifmedia_set(&sc->sc_media, IFM_ETHER|IFM_10G_LR);
917 	}
918 
919 	ifp = &sc->sc_ethercom.ec_if;
920 	strlcpy(ifp->if_xname, device_xname(sc->sc_dev), IFNAMSIZ);
921 	ifp->if_softc = sc;
922 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
923 	ifp->if_ioctl = dge_ioctl;
924 	ifp->if_start = dge_start;
925 	ifp->if_watchdog = dge_watchdog;
926 	ifp->if_init = dge_init;
927 	ifp->if_stop = dge_stop;
928 	IFQ_SET_MAXLEN(&ifp->if_snd, max(DGE_IFQUEUELEN, IFQ_MAXLEN));
929 	IFQ_SET_READY(&ifp->if_snd);
930 
931 	sc->sc_ethercom.ec_capabilities |=
932 	    ETHERCAP_JUMBO_MTU | ETHERCAP_VLAN_MTU;
933 
934 	/*
935 	 * We can perform TCPv4 and UDPv4 checkums in-bound.
936 	 */
937 	ifp->if_capabilities |=
938 	    IFCAP_CSUM_IPv4_Tx | IFCAP_CSUM_IPv4_Rx |
939 	    IFCAP_CSUM_TCPv4_Tx | IFCAP_CSUM_TCPv4_Rx |
940 	    IFCAP_CSUM_UDPv4_Tx | IFCAP_CSUM_UDPv4_Rx;
941 
942 	/*
943 	 * Attach the interface.
944 	 */
945 	if_attach(ifp);
946 	ether_ifattach(ifp, enaddr);
947 	rnd_attach_source(&sc->rnd_source, device_xname(sc->sc_dev),
948 	    RND_TYPE_NET, RND_FLAG_DEFAULT);
949 
950 #ifdef DGE_EVENT_COUNTERS
951 	/* Fix segment event naming */
952 	if (dge_txseg_evcnt_names == NULL) {
953 		dge_txseg_evcnt_names =
954 		    malloc(sizeof(*dge_txseg_evcnt_names), M_DEVBUF, M_WAITOK);
955 		for (i = 0; i < DGE_NTXSEGS; i++)
956 			snprintf((*dge_txseg_evcnt_names)[i],
957 			    sizeof((*dge_txseg_evcnt_names)[i]), "txseg%d", i);
958 	}
959 
960 	/* Attach event counters. */
961 	evcnt_attach_dynamic(&sc->sc_ev_txsstall, EVCNT_TYPE_MISC,
962 	    NULL, device_xname(sc->sc_dev), "txsstall");
963 	evcnt_attach_dynamic(&sc->sc_ev_txdstall, EVCNT_TYPE_MISC,
964 	    NULL, device_xname(sc->sc_dev), "txdstall");
965 	evcnt_attach_dynamic(&sc->sc_ev_txforceintr, EVCNT_TYPE_MISC,
966 	    NULL, device_xname(sc->sc_dev), "txforceintr");
967 	evcnt_attach_dynamic(&sc->sc_ev_txdw, EVCNT_TYPE_INTR,
968 	    NULL, device_xname(sc->sc_dev), "txdw");
969 	evcnt_attach_dynamic(&sc->sc_ev_txqe, EVCNT_TYPE_INTR,
970 	    NULL, device_xname(sc->sc_dev), "txqe");
971 	evcnt_attach_dynamic(&sc->sc_ev_rxintr, EVCNT_TYPE_INTR,
972 	    NULL, device_xname(sc->sc_dev), "rxintr");
973 	evcnt_attach_dynamic(&sc->sc_ev_linkintr, EVCNT_TYPE_INTR,
974 	    NULL, device_xname(sc->sc_dev), "linkintr");
975 
976 	evcnt_attach_dynamic(&sc->sc_ev_rxipsum, EVCNT_TYPE_MISC,
977 	    NULL, device_xname(sc->sc_dev), "rxipsum");
978 	evcnt_attach_dynamic(&sc->sc_ev_rxtusum, EVCNT_TYPE_MISC,
979 	    NULL, device_xname(sc->sc_dev), "rxtusum");
980 	evcnt_attach_dynamic(&sc->sc_ev_txipsum, EVCNT_TYPE_MISC,
981 	    NULL, device_xname(sc->sc_dev), "txipsum");
982 	evcnt_attach_dynamic(&sc->sc_ev_txtusum, EVCNT_TYPE_MISC,
983 	    NULL, device_xname(sc->sc_dev), "txtusum");
984 
985 	evcnt_attach_dynamic(&sc->sc_ev_txctx_init, EVCNT_TYPE_MISC,
986 	    NULL, device_xname(sc->sc_dev), "txctx init");
987 	evcnt_attach_dynamic(&sc->sc_ev_txctx_hit, EVCNT_TYPE_MISC,
988 	    NULL, device_xname(sc->sc_dev), "txctx hit");
989 	evcnt_attach_dynamic(&sc->sc_ev_txctx_miss, EVCNT_TYPE_MISC,
990 	    NULL, device_xname(sc->sc_dev), "txctx miss");
991 
992 	for (i = 0; i < DGE_NTXSEGS; i++)
993 		evcnt_attach_dynamic(&sc->sc_ev_txseg[i], EVCNT_TYPE_MISC,
994 		    NULL, device_xname(sc->sc_dev), (*dge_txseg_evcnt_names)[i]);
995 
996 	evcnt_attach_dynamic(&sc->sc_ev_txdrop, EVCNT_TYPE_MISC,
997 	    NULL, device_xname(sc->sc_dev), "txdrop");
998 
999 #endif /* DGE_EVENT_COUNTERS */
1000 
1001 	/*
1002 	 * Make sure the interface is shutdown during reboot.
1003 	 */
1004 	if (pmf_device_register1(self, NULL, NULL, dge_shutdown))
1005 		pmf_class_network_register(self, ifp);
1006 	else
1007 		aprint_error_dev(self, "couldn't establish power handler\n");
1008 
1009 	return;
1010 
1011 	/*
1012 	 * Free any resources we've allocated during the failed attach
1013 	 * attempt.  Do this in reverse order and fall through.
1014 	 */
1015  fail_5:
1016 	for (i = 0; i < DGE_NRXDESC; i++) {
1017 		if (sc->sc_rxsoft[i].rxs_dmamap != NULL)
1018 			bus_dmamap_destroy(sc->sc_dmat,
1019 			    sc->sc_rxsoft[i].rxs_dmamap);
1020 	}
1021  fail_4:
1022 	for (i = 0; i < DGE_TXQUEUELEN; i++) {
1023 		if (sc->sc_txsoft[i].txs_dmamap != NULL)
1024 			bus_dmamap_destroy(sc->sc_dmat,
1025 			    sc->sc_txsoft[i].txs_dmamap);
1026 	}
1027 	bus_dmamap_unload(sc->sc_dmat, sc->sc_cddmamap);
1028  fail_3:
1029 	bus_dmamap_destroy(sc->sc_dmat, sc->sc_cddmamap);
1030  fail_2:
1031 	bus_dmamem_unmap(sc->sc_dmat, (void *)sc->sc_control_data,
1032 	    sizeof(struct dge_control_data));
1033  fail_1:
1034 	bus_dmamem_free(sc->sc_dmat, &seg, rseg);
1035  fail_0:
1036 	return;
1037 }
1038 
1039 /*
1040  * dge_shutdown:
1041  *
1042  *	Make sure the interface is stopped at reboot time.
1043  */
1044 static bool
1045 dge_shutdown(device_t self, int howto)
1046 {
1047 	struct dge_softc *sc;
1048 
1049 	sc = device_private(self);
1050 	dge_stop(&sc->sc_ethercom.ec_if, 1);
1051 
1052 	return true;
1053 }
1054 
1055 /*
1056  * dge_tx_cksum:
1057  *
1058  *	Set up TCP/IP checksumming parameters for the
1059  *	specified packet.
1060  */
1061 static int
1062 dge_tx_cksum(struct dge_softc *sc, struct dge_txsoft *txs, uint8_t *fieldsp)
1063 {
1064 	struct mbuf *m0 = txs->txs_mbuf;
1065 	struct dge_ctdes *t;
1066 	uint32_t ipcs, tucs;
1067 	struct ether_header *eh;
1068 	int offset, iphl;
1069 	uint8_t fields = 0;
1070 
1071 	/*
1072 	 * XXX It would be nice if the mbuf pkthdr had offset
1073 	 * fields for the protocol headers.
1074 	 */
1075 
1076 	eh = mtod(m0, struct ether_header *);
1077 	switch (htons(eh->ether_type)) {
1078 	case ETHERTYPE_IP:
1079 		offset = ETHER_HDR_LEN;
1080 		break;
1081 
1082 	case ETHERTYPE_VLAN:
1083 		offset = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
1084 		break;
1085 
1086 	default:
1087 		/*
1088 		 * Don't support this protocol or encapsulation.
1089 		 */
1090 		*fieldsp = 0;
1091 		return (0);
1092 	}
1093 
1094 	iphl = M_CSUM_DATA_IPv4_IPHL(m0->m_pkthdr.csum_data);
1095 
1096 	/*
1097 	 * NOTE: Even if we're not using the IP or TCP/UDP checksum
1098 	 * offload feature, if we load the context descriptor, we
1099 	 * MUST provide valid values for IPCSS and TUCSS fields.
1100 	 */
1101 
1102 	if (m0->m_pkthdr.csum_flags & M_CSUM_IPv4) {
1103 		DGE_EVCNT_INCR(&sc->sc_ev_txipsum);
1104 		fields |= TDESC_POPTS_IXSM;
1105 		ipcs = DGE_TCPIP_IPCSS(offset) |
1106 		    DGE_TCPIP_IPCSO(offset + offsetof(struct ip, ip_sum)) |
1107 		    DGE_TCPIP_IPCSE(offset + iphl - 1);
1108 	} else if (__predict_true(sc->sc_txctx_ipcs != 0xffffffff)) {
1109 		/* Use the cached value. */
1110 		ipcs = sc->sc_txctx_ipcs;
1111 	} else {
1112 		/* Just initialize it to the likely value anyway. */
1113 		ipcs = DGE_TCPIP_IPCSS(offset) |
1114 		    DGE_TCPIP_IPCSO(offset + offsetof(struct ip, ip_sum)) |
1115 		    DGE_TCPIP_IPCSE(offset + iphl - 1);
1116 	}
1117 	DPRINTF(DGE_DEBUG_CKSUM,
1118 	    ("%s: CKSUM: offset %d ipcs 0x%x\n",
1119 	    device_xname(sc->sc_dev), offset, ipcs));
1120 
1121 	offset += iphl;
1122 
1123 	if (m0->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4)) {
1124 		DGE_EVCNT_INCR(&sc->sc_ev_txtusum);
1125 		fields |= TDESC_POPTS_TXSM;
1126 		tucs = DGE_TCPIP_TUCSS(offset) |
1127 		   DGE_TCPIP_TUCSO(offset + M_CSUM_DATA_IPv4_OFFSET(m0->m_pkthdr.csum_data)) |
1128 		   DGE_TCPIP_TUCSE(0) /* rest of packet */;
1129 	} else if (__predict_true(sc->sc_txctx_tucs != 0xffffffff)) {
1130 		/* Use the cached value. */
1131 		tucs = sc->sc_txctx_tucs;
1132 	} else {
1133 		/* Just initialize it to a valid TCP context. */
1134 		tucs = DGE_TCPIP_TUCSS(offset) |
1135 		    DGE_TCPIP_TUCSO(offset + offsetof(struct tcphdr, th_sum)) |
1136 		    DGE_TCPIP_TUCSE(0) /* rest of packet */;
1137 	}
1138 
1139 	DPRINTF(DGE_DEBUG_CKSUM,
1140 	    ("%s: CKSUM: offset %d tucs 0x%x\n",
1141 	    device_xname(sc->sc_dev), offset, tucs));
1142 
1143 	if (sc->sc_txctx_ipcs == ipcs &&
1144 	    sc->sc_txctx_tucs == tucs) {
1145 		/* Cached context is fine. */
1146 		DGE_EVCNT_INCR(&sc->sc_ev_txctx_hit);
1147 	} else {
1148 		/* Fill in the context descriptor. */
1149 #ifdef DGE_EVENT_COUNTERS
1150 		if (sc->sc_txctx_ipcs == 0xffffffff &&
1151 		    sc->sc_txctx_tucs == 0xffffffff)
1152 			DGE_EVCNT_INCR(&sc->sc_ev_txctx_init);
1153 		else
1154 			DGE_EVCNT_INCR(&sc->sc_ev_txctx_miss);
1155 #endif
1156 		t = (struct dge_ctdes *)&sc->sc_txdescs[sc->sc_txnext];
1157 		t->dc_tcpip_ipcs = htole32(ipcs);
1158 		t->dc_tcpip_tucs = htole32(tucs);
1159 		t->dc_tcpip_cmdlen = htole32(TDESC_DTYP_CTD);
1160 		t->dc_tcpip_seg = 0;
1161 		DGE_CDTXSYNC(sc, sc->sc_txnext, 1, BUS_DMASYNC_PREWRITE);
1162 
1163 		sc->sc_txctx_ipcs = ipcs;
1164 		sc->sc_txctx_tucs = tucs;
1165 
1166 		sc->sc_txnext = DGE_NEXTTX(sc->sc_txnext);
1167 		txs->txs_ndesc++;
1168 	}
1169 
1170 	*fieldsp = fields;
1171 
1172 	return (0);
1173 }
1174 
1175 /*
1176  * dge_start:		[ifnet interface function]
1177  *
1178  *	Start packet transmission on the interface.
1179  */
1180 static void
1181 dge_start(struct ifnet *ifp)
1182 {
1183 	struct dge_softc *sc = ifp->if_softc;
1184 	struct mbuf *m0;
1185 	struct dge_txsoft *txs;
1186 	bus_dmamap_t dmamap;
1187 	int error, nexttx, lasttx = -1, ofree, seg;
1188 	uint32_t cksumcmd;
1189 	uint8_t cksumfields;
1190 
1191 	if ((ifp->if_flags & (IFF_RUNNING|IFF_OACTIVE)) != IFF_RUNNING)
1192 		return;
1193 
1194 	/*
1195 	 * Remember the previous number of free descriptors.
1196 	 */
1197 	ofree = sc->sc_txfree;
1198 
1199 	/*
1200 	 * Loop through the send queue, setting up transmit descriptors
1201 	 * until we drain the queue, or use up all available transmit
1202 	 * descriptors.
1203 	 */
1204 	for (;;) {
1205 		/* Grab a packet off the queue. */
1206 		IFQ_POLL(&ifp->if_snd, m0);
1207 		if (m0 == NULL)
1208 			break;
1209 
1210 		DPRINTF(DGE_DEBUG_TX,
1211 		    ("%s: TX: have packet to transmit: %p\n",
1212 		    device_xname(sc->sc_dev), m0));
1213 
1214 		/* Get a work queue entry. */
1215 		if (sc->sc_txsfree < DGE_TXQUEUE_GC) {
1216 			dge_txintr(sc);
1217 			if (sc->sc_txsfree == 0) {
1218 				DPRINTF(DGE_DEBUG_TX,
1219 				    ("%s: TX: no free job descriptors\n",
1220 					device_xname(sc->sc_dev)));
1221 				DGE_EVCNT_INCR(&sc->sc_ev_txsstall);
1222 				break;
1223 			}
1224 		}
1225 
1226 		txs = &sc->sc_txsoft[sc->sc_txsnext];
1227 		dmamap = txs->txs_dmamap;
1228 
1229 		/*
1230 		 * Load the DMA map.  If this fails, the packet either
1231 		 * didn't fit in the allotted number of segments, or we
1232 		 * were short on resources.  For the too-many-segments
1233 		 * case, we simply report an error and drop the packet,
1234 		 * since we can't sanely copy a jumbo packet to a single
1235 		 * buffer.
1236 		 */
1237 		error = bus_dmamap_load_mbuf(sc->sc_dmat, dmamap, m0,
1238 		    BUS_DMA_WRITE|BUS_DMA_NOWAIT);
1239 		if (error) {
1240 			if (error == EFBIG) {
1241 				DGE_EVCNT_INCR(&sc->sc_ev_txdrop);
1242 				printf("%s: Tx packet consumes too many "
1243 				    "DMA segments, dropping...\n",
1244 				    device_xname(sc->sc_dev));
1245 				IFQ_DEQUEUE(&ifp->if_snd, m0);
1246 				m_freem(m0);
1247 				continue;
1248 			}
1249 			/*
1250 			 * Short on resources, just stop for now.
1251 			 */
1252 			DPRINTF(DGE_DEBUG_TX,
1253 			    ("%s: TX: dmamap load failed: %d\n",
1254 			    device_xname(sc->sc_dev), error));
1255 			break;
1256 		}
1257 
1258 		/*
1259 		 * Ensure we have enough descriptors free to describe
1260 		 * the packet.  Note, we always reserve one descriptor
1261 		 * at the end of the ring due to the semantics of the
1262 		 * TDT register, plus one more in the event we need
1263 		 * to re-load checksum offload context.
1264 		 */
1265 		if (dmamap->dm_nsegs > (sc->sc_txfree - 2)) {
1266 			/*
1267 			 * Not enough free descriptors to transmit this
1268 			 * packet.  We haven't committed anything yet,
1269 			 * so just unload the DMA map, put the packet
1270 			 * pack on the queue, and punt.  Notify the upper
1271 			 * layer that there are no more slots left.
1272 			 */
1273 			DPRINTF(DGE_DEBUG_TX,
1274 			    ("%s: TX: need %d descriptors, have %d\n",
1275 			    device_xname(sc->sc_dev), dmamap->dm_nsegs,
1276 			    sc->sc_txfree - 1));
1277 			ifp->if_flags |= IFF_OACTIVE;
1278 			bus_dmamap_unload(sc->sc_dmat, dmamap);
1279 			DGE_EVCNT_INCR(&sc->sc_ev_txdstall);
1280 			break;
1281 		}
1282 
1283 		IFQ_DEQUEUE(&ifp->if_snd, m0);
1284 
1285 		/*
1286 		 * WE ARE NOW COMMITTED TO TRANSMITTING THE PACKET.
1287 		 */
1288 
1289 		/* Sync the DMA map. */
1290 		bus_dmamap_sync(sc->sc_dmat, dmamap, 0, dmamap->dm_mapsize,
1291 		    BUS_DMASYNC_PREWRITE);
1292 
1293 		DPRINTF(DGE_DEBUG_TX,
1294 		    ("%s: TX: packet has %d DMA segments\n",
1295 		    device_xname(sc->sc_dev), dmamap->dm_nsegs));
1296 
1297 		DGE_EVCNT_INCR(&sc->sc_ev_txseg[dmamap->dm_nsegs - 1]);
1298 
1299 		/*
1300 		 * Store a pointer to the packet so that we can free it
1301 		 * later.
1302 		 *
1303 		 * Initially, we consider the number of descriptors the
1304 		 * packet uses the number of DMA segments.  This may be
1305 		 * incremented by 1 if we do checksum offload (a descriptor
1306 		 * is used to set the checksum context).
1307 		 */
1308 		txs->txs_mbuf = m0;
1309 		txs->txs_firstdesc = sc->sc_txnext;
1310 		txs->txs_ndesc = dmamap->dm_nsegs;
1311 
1312 		/*
1313 		 * Set up checksum offload parameters for
1314 		 * this packet.
1315 		 */
1316 		if (m0->m_pkthdr.csum_flags &
1317 		    (M_CSUM_IPv4|M_CSUM_TCPv4|M_CSUM_UDPv4)) {
1318 			if (dge_tx_cksum(sc, txs, &cksumfields) != 0) {
1319 				/* Error message already displayed. */
1320 				bus_dmamap_unload(sc->sc_dmat, dmamap);
1321 				continue;
1322 			}
1323 		} else {
1324 			cksumfields = 0;
1325 		}
1326 
1327 		cksumcmd = TDESC_DCMD_IDE | TDESC_DTYP_DATA;
1328 
1329 		/*
1330 		 * Initialize the transmit descriptor.
1331 		 */
1332 		for (nexttx = sc->sc_txnext, seg = 0;
1333 		     seg < dmamap->dm_nsegs;
1334 		     seg++, nexttx = DGE_NEXTTX(nexttx)) {
1335 			/*
1336 			 * Note: we currently only use 32-bit DMA
1337 			 * addresses.
1338 			 */
1339 			sc->sc_txdescs[nexttx].dt_baddrh = 0;
1340 			sc->sc_txdescs[nexttx].dt_baddrl =
1341 			    htole32(dmamap->dm_segs[seg].ds_addr);
1342 			sc->sc_txdescs[nexttx].dt_ctl =
1343 			    htole32(cksumcmd | dmamap->dm_segs[seg].ds_len);
1344 			sc->sc_txdescs[nexttx].dt_status = 0;
1345 			sc->sc_txdescs[nexttx].dt_popts = cksumfields;
1346 			sc->sc_txdescs[nexttx].dt_vlan = 0;
1347 			lasttx = nexttx;
1348 
1349 			DPRINTF(DGE_DEBUG_TX,
1350 			    ("%s: TX: desc %d: low 0x%08lx, len 0x%04lx\n",
1351 			    device_xname(sc->sc_dev), nexttx,
1352 			    (unsigned long)le32toh(dmamap->dm_segs[seg].ds_addr),
1353 			    (unsigned long)le32toh(dmamap->dm_segs[seg].ds_len)));
1354 		}
1355 
1356 		KASSERT(lasttx != -1);
1357 
1358 		/*
1359 		 * Set up the command byte on the last descriptor of
1360 		 * the packet.  If we're in the interrupt delay window,
1361 		 * delay the interrupt.
1362 		 */
1363 		sc->sc_txdescs[lasttx].dt_ctl |=
1364 		    htole32(TDESC_DCMD_EOP | TDESC_DCMD_RS);
1365 
1366 		txs->txs_lastdesc = lasttx;
1367 
1368 		DPRINTF(DGE_DEBUG_TX,
1369 		    ("%s: TX: desc %d: cmdlen 0x%08x\n", device_xname(sc->sc_dev),
1370 		    lasttx, le32toh(sc->sc_txdescs[lasttx].dt_ctl)));
1371 
1372 		/* Sync the descriptors we're using. */
1373 		DGE_CDTXSYNC(sc, sc->sc_txnext, dmamap->dm_nsegs,
1374 		    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1375 
1376 		/* Give the packet to the chip. */
1377 		CSR_WRITE(sc, DGE_TDT, nexttx);
1378 
1379 		DPRINTF(DGE_DEBUG_TX,
1380 		    ("%s: TX: TDT -> %d\n", device_xname(sc->sc_dev), nexttx));
1381 
1382 		DPRINTF(DGE_DEBUG_TX,
1383 		    ("%s: TX: finished transmitting packet, job %d\n",
1384 		    device_xname(sc->sc_dev), sc->sc_txsnext));
1385 
1386 		/* Advance the tx pointer. */
1387 		sc->sc_txfree -= txs->txs_ndesc;
1388 		sc->sc_txnext = nexttx;
1389 
1390 		sc->sc_txsfree--;
1391 		sc->sc_txsnext = DGE_NEXTTXS(sc->sc_txsnext);
1392 
1393 		/* Pass the packet to any BPF listeners. */
1394 		bpf_mtap(ifp, m0);
1395 	}
1396 
1397 	if (sc->sc_txsfree == 0 || sc->sc_txfree <= 2) {
1398 		/* No more slots; notify upper layer. */
1399 		ifp->if_flags |= IFF_OACTIVE;
1400 	}
1401 
1402 	if (sc->sc_txfree != ofree) {
1403 		/* Set a watchdog timer in case the chip flakes out. */
1404 		ifp->if_timer = 5;
1405 	}
1406 }
1407 
1408 /*
1409  * dge_watchdog:		[ifnet interface function]
1410  *
1411  *	Watchdog timer handler.
1412  */
1413 static void
1414 dge_watchdog(struct ifnet *ifp)
1415 {
1416 	struct dge_softc *sc = ifp->if_softc;
1417 
1418 	/*
1419 	 * Since we're using delayed interrupts, sweep up
1420 	 * before we report an error.
1421 	 */
1422 	dge_txintr(sc);
1423 
1424 	if (sc->sc_txfree != DGE_NTXDESC) {
1425 		printf("%s: device timeout (txfree %d txsfree %d txnext %d)\n",
1426 		    device_xname(sc->sc_dev), sc->sc_txfree, sc->sc_txsfree,
1427 		    sc->sc_txnext);
1428 		ifp->if_oerrors++;
1429 
1430 		/* Reset the interface. */
1431 		(void) dge_init(ifp);
1432 	}
1433 
1434 	/* Try to get more packets going. */
1435 	dge_start(ifp);
1436 }
1437 
1438 /*
1439  * dge_ioctl:		[ifnet interface function]
1440  *
1441  *	Handle control requests from the operator.
1442  */
1443 static int
1444 dge_ioctl(struct ifnet *ifp, u_long cmd, void *data)
1445 {
1446 	struct dge_softc *sc = ifp->if_softc;
1447 	struct ifreq *ifr = (struct ifreq *) data;
1448 	pcireg_t preg;
1449 	int s, error, mmrbc;
1450 
1451 	s = splnet();
1452 
1453 	switch (cmd) {
1454 	case SIOCSIFMEDIA:
1455 	case SIOCGIFMEDIA:
1456 		error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
1457 		break;
1458 
1459 	case SIOCSIFMTU:
1460 		if (ifr->ifr_mtu < ETHERMIN || ifr->ifr_mtu > DGE_MAX_MTU)
1461 			error = EINVAL;
1462 		else if ((error = ifioctl_common(ifp, cmd, data)) != ENETRESET)
1463 			break;
1464 		else if (ifp->if_flags & IFF_UP)
1465 			error = (*ifp->if_init)(ifp);
1466 		else
1467 			error = 0;
1468 		break;
1469 
1470         case SIOCSIFFLAGS:
1471 		if ((error = ifioctl_common(ifp, cmd, data)) != 0)
1472 			break;
1473 		/* extract link flags */
1474 		if ((ifp->if_flags & IFF_LINK0) == 0 &&
1475 		    (ifp->if_flags & IFF_LINK1) == 0)
1476 			mmrbc = PCIX_MMRBC_512;
1477 		else if ((ifp->if_flags & IFF_LINK0) == 0 &&
1478 		    (ifp->if_flags & IFF_LINK1) != 0)
1479 			mmrbc = PCIX_MMRBC_1024;
1480 		else if ((ifp->if_flags & IFF_LINK0) != 0 &&
1481 		    (ifp->if_flags & IFF_LINK1) == 0)
1482 			mmrbc = PCIX_MMRBC_2048;
1483 		else
1484 			mmrbc = PCIX_MMRBC_4096;
1485 		if (mmrbc != sc->sc_mmrbc) {
1486 			preg = pci_conf_read(sc->sc_pc, sc->sc_pt,DGE_PCIX_CMD);
1487 			preg &= ~PCIX_MMRBC_MSK;
1488 			preg |= mmrbc;
1489 			pci_conf_write(sc->sc_pc, sc->sc_pt,DGE_PCIX_CMD, preg);
1490 			sc->sc_mmrbc = mmrbc;
1491 		}
1492                 /* FALLTHROUGH */
1493 	default:
1494 		if ((error = ether_ioctl(ifp, cmd, data)) != ENETRESET)
1495 			break;
1496 
1497 		error = 0;
1498 
1499 		if (cmd == SIOCSIFCAP)
1500 			error = (*ifp->if_init)(ifp);
1501 		else if (cmd != SIOCADDMULTI && cmd != SIOCDELMULTI)
1502 			;
1503 		else if (ifp->if_flags & IFF_RUNNING) {
1504 			/*
1505 			 * Multicast list has changed; set the hardware filter
1506 			 * accordingly.
1507 			 */
1508 			dge_set_filter(sc);
1509 		}
1510 		break;
1511 	}
1512 
1513 	/* Try to get more packets going. */
1514 	dge_start(ifp);
1515 
1516 	splx(s);
1517 	return (error);
1518 }
1519 
1520 /*
1521  * dge_intr:
1522  *
1523  *	Interrupt service routine.
1524  */
1525 static int
1526 dge_intr(void *arg)
1527 {
1528 	struct dge_softc *sc = arg;
1529 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
1530 	uint32_t icr;
1531 	int wantinit, handled = 0;
1532 
1533 	for (wantinit = 0; wantinit == 0;) {
1534 		icr = CSR_READ(sc, DGE_ICR);
1535 		if ((icr & sc->sc_icr) == 0)
1536 			break;
1537 
1538 		rnd_add_uint32(&sc->rnd_source, icr);
1539 
1540 		handled = 1;
1541 
1542 #if defined(DGE_DEBUG) || defined(DGE_EVENT_COUNTERS)
1543 		if (icr & (ICR_RXDMT0|ICR_RXT0)) {
1544 			DPRINTF(DGE_DEBUG_RX,
1545 			    ("%s: RX: got Rx intr 0x%08x\n",
1546 			    device_xname(sc->sc_dev),
1547 			    icr & (ICR_RXDMT0|ICR_RXT0)));
1548 			DGE_EVCNT_INCR(&sc->sc_ev_rxintr);
1549 		}
1550 #endif
1551 		dge_rxintr(sc);
1552 
1553 #if defined(DGE_DEBUG) || defined(DGE_EVENT_COUNTERS)
1554 		if (icr & ICR_TXDW) {
1555 			DPRINTF(DGE_DEBUG_TX,
1556 			    ("%s: TX: got TXDW interrupt\n",
1557 			    device_xname(sc->sc_dev)));
1558 			DGE_EVCNT_INCR(&sc->sc_ev_txdw);
1559 		}
1560 		if (icr & ICR_TXQE)
1561 			DGE_EVCNT_INCR(&sc->sc_ev_txqe);
1562 #endif
1563 		dge_txintr(sc);
1564 
1565 		if (icr & (ICR_LSC|ICR_RXSEQ)) {
1566 			DGE_EVCNT_INCR(&sc->sc_ev_linkintr);
1567 			dge_linkintr(sc, icr);
1568 		}
1569 
1570 		if (icr & ICR_RXO) {
1571 			printf("%s: Receive overrun\n", device_xname(sc->sc_dev));
1572 			wantinit = 1;
1573 		}
1574 	}
1575 
1576 	if (handled) {
1577 		if (wantinit)
1578 			dge_init(ifp);
1579 
1580 		/* Try to get more packets going. */
1581 		dge_start(ifp);
1582 	}
1583 
1584 	return (handled);
1585 }
1586 
1587 /*
1588  * dge_txintr:
1589  *
1590  *	Helper; handle transmit interrupts.
1591  */
1592 static void
1593 dge_txintr(struct dge_softc *sc)
1594 {
1595 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
1596 	struct dge_txsoft *txs;
1597 	uint8_t status;
1598 	int i;
1599 
1600 	ifp->if_flags &= ~IFF_OACTIVE;
1601 
1602 	/*
1603 	 * Go through the Tx list and free mbufs for those
1604 	 * frames which have been transmitted.
1605 	 */
1606 	for (i = sc->sc_txsdirty; sc->sc_txsfree != DGE_TXQUEUELEN;
1607 	     i = DGE_NEXTTXS(i), sc->sc_txsfree++) {
1608 		txs = &sc->sc_txsoft[i];
1609 
1610 		DPRINTF(DGE_DEBUG_TX,
1611 		    ("%s: TX: checking job %d\n", device_xname(sc->sc_dev), i));
1612 
1613 		DGE_CDTXSYNC(sc, txs->txs_firstdesc, txs->txs_dmamap->dm_nsegs,
1614 		    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1615 
1616 		status =
1617 		    sc->sc_txdescs[txs->txs_lastdesc].dt_status;
1618 		if ((status & TDESC_STA_DD) == 0) {
1619 			DGE_CDTXSYNC(sc, txs->txs_lastdesc, 1,
1620 			    BUS_DMASYNC_PREREAD);
1621 			break;
1622 		}
1623 
1624 		DPRINTF(DGE_DEBUG_TX,
1625 		    ("%s: TX: job %d done: descs %d..%d\n",
1626 		    device_xname(sc->sc_dev), i, txs->txs_firstdesc,
1627 		    txs->txs_lastdesc));
1628 
1629 		ifp->if_opackets++;
1630 		sc->sc_txfree += txs->txs_ndesc;
1631 		bus_dmamap_sync(sc->sc_dmat, txs->txs_dmamap,
1632 		    0, txs->txs_dmamap->dm_mapsize, BUS_DMASYNC_POSTWRITE);
1633 		bus_dmamap_unload(sc->sc_dmat, txs->txs_dmamap);
1634 		m_freem(txs->txs_mbuf);
1635 		txs->txs_mbuf = NULL;
1636 	}
1637 
1638 	/* Update the dirty transmit buffer pointer. */
1639 	sc->sc_txsdirty = i;
1640 	DPRINTF(DGE_DEBUG_TX,
1641 	    ("%s: TX: txsdirty -> %d\n", device_xname(sc->sc_dev), i));
1642 
1643 	/*
1644 	 * If there are no more pending transmissions, cancel the watchdog
1645 	 * timer.
1646 	 */
1647 	if (sc->sc_txsfree == DGE_TXQUEUELEN)
1648 		ifp->if_timer = 0;
1649 }
1650 
1651 /*
1652  * dge_rxintr:
1653  *
1654  *	Helper; handle receive interrupts.
1655  */
1656 static void
1657 dge_rxintr(struct dge_softc *sc)
1658 {
1659 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
1660 	struct dge_rxsoft *rxs;
1661 	struct mbuf *m;
1662 	int i, len;
1663 	uint8_t status, errors;
1664 
1665 	for (i = sc->sc_rxptr;; i = DGE_NEXTRX(i)) {
1666 		rxs = &sc->sc_rxsoft[i];
1667 
1668 		DPRINTF(DGE_DEBUG_RX,
1669 		    ("%s: RX: checking descriptor %d\n",
1670 		    device_xname(sc->sc_dev), i));
1671 
1672 		DGE_CDRXSYNC(sc, i, BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1673 
1674 		status = sc->sc_rxdescs[i].dr_status;
1675 		errors = sc->sc_rxdescs[i].dr_errors;
1676 		len = le16toh(sc->sc_rxdescs[i].dr_len);
1677 
1678 		if ((status & RDESC_STS_DD) == 0) {
1679 			/*
1680 			 * We have processed all of the receive descriptors.
1681 			 */
1682 			DGE_CDRXSYNC(sc, i, BUS_DMASYNC_PREREAD);
1683 			break;
1684 		}
1685 
1686 		if (__predict_false(sc->sc_rxdiscard)) {
1687 			DPRINTF(DGE_DEBUG_RX,
1688 			    ("%s: RX: discarding contents of descriptor %d\n",
1689 			    device_xname(sc->sc_dev), i));
1690 			DGE_INIT_RXDESC(sc, i);
1691 			if (status & RDESC_STS_EOP) {
1692 				/* Reset our state. */
1693 				DPRINTF(DGE_DEBUG_RX,
1694 				    ("%s: RX: resetting rxdiscard -> 0\n",
1695 				    device_xname(sc->sc_dev)));
1696 				sc->sc_rxdiscard = 0;
1697 			}
1698 			continue;
1699 		}
1700 
1701 		bus_dmamap_sync(sc->sc_dmat, rxs->rxs_dmamap, 0,
1702 		    rxs->rxs_dmamap->dm_mapsize, BUS_DMASYNC_POSTREAD);
1703 
1704 		m = rxs->rxs_mbuf;
1705 
1706 		/*
1707 		 * Add a new receive buffer to the ring.
1708 		 */
1709 		if (dge_add_rxbuf(sc, i) != 0) {
1710 			/*
1711 			 * Failed, throw away what we've done so
1712 			 * far, and discard the rest of the packet.
1713 			 */
1714 			ifp->if_ierrors++;
1715 			bus_dmamap_sync(sc->sc_dmat, rxs->rxs_dmamap, 0,
1716 			    rxs->rxs_dmamap->dm_mapsize, BUS_DMASYNC_PREREAD);
1717 			DGE_INIT_RXDESC(sc, i);
1718 			if ((status & RDESC_STS_EOP) == 0)
1719 				sc->sc_rxdiscard = 1;
1720 			if (sc->sc_rxhead != NULL)
1721 				m_freem(sc->sc_rxhead);
1722 			DGE_RXCHAIN_RESET(sc);
1723 			DPRINTF(DGE_DEBUG_RX,
1724 			    ("%s: RX: Rx buffer allocation failed, "
1725 			    "dropping packet%s\n", device_xname(sc->sc_dev),
1726 			    sc->sc_rxdiscard ? " (discard)" : ""));
1727 			continue;
1728 		}
1729 		DGE_INIT_RXDESC(sc, DGE_PREVRX(i)); /* Write the descriptor */
1730 
1731 		DGE_RXCHAIN_LINK(sc, m);
1732 
1733 		m->m_len = len;
1734 
1735 		DPRINTF(DGE_DEBUG_RX,
1736 		    ("%s: RX: buffer at %p len %d\n",
1737 		    device_xname(sc->sc_dev), m->m_data, len));
1738 
1739 		/*
1740 		 * If this is not the end of the packet, keep
1741 		 * looking.
1742 		 */
1743 		if ((status & RDESC_STS_EOP) == 0) {
1744 			sc->sc_rxlen += len;
1745 			DPRINTF(DGE_DEBUG_RX,
1746 			    ("%s: RX: not yet EOP, rxlen -> %d\n",
1747 			    device_xname(sc->sc_dev), sc->sc_rxlen));
1748 			continue;
1749 		}
1750 
1751 		/*
1752 		 * Okay, we have the entire packet now...
1753 		 */
1754 		*sc->sc_rxtailp = NULL;
1755 		m = sc->sc_rxhead;
1756 		len += sc->sc_rxlen;
1757 
1758 		DGE_RXCHAIN_RESET(sc);
1759 
1760 		DPRINTF(DGE_DEBUG_RX,
1761 		    ("%s: RX: have entire packet, len -> %d\n",
1762 		    device_xname(sc->sc_dev), len));
1763 
1764 		/*
1765 		 * If an error occurred, update stats and drop the packet.
1766 		 */
1767 		if (errors &
1768 		     (RDESC_ERR_CE|RDESC_ERR_SE|RDESC_ERR_P|RDESC_ERR_RXE)) {
1769 			ifp->if_ierrors++;
1770 			if (errors & RDESC_ERR_SE)
1771 				printf("%s: symbol error\n",
1772 				    device_xname(sc->sc_dev));
1773 			else if (errors & RDESC_ERR_P)
1774 				printf("%s: parity error\n",
1775 				    device_xname(sc->sc_dev));
1776 			else if (errors & RDESC_ERR_CE)
1777 				printf("%s: CRC error\n",
1778 				    device_xname(sc->sc_dev));
1779 			m_freem(m);
1780 			continue;
1781 		}
1782 
1783 		/*
1784 		 * No errors.  Receive the packet.
1785 		 */
1786 		m_set_rcvif(m, ifp);
1787 		m->m_pkthdr.len = len;
1788 
1789 		/*
1790 		 * Set up checksum info for this packet.
1791 		 */
1792 		if (status & RDESC_STS_IPCS) {
1793 			DGE_EVCNT_INCR(&sc->sc_ev_rxipsum);
1794 			m->m_pkthdr.csum_flags |= M_CSUM_IPv4;
1795 			if (errors & RDESC_ERR_IPE)
1796 				m->m_pkthdr.csum_flags |= M_CSUM_IPv4_BAD;
1797 		}
1798 		if (status & RDESC_STS_TCPCS) {
1799 			/*
1800 			 * Note: we don't know if this was TCP or UDP,
1801 			 * so we just set both bits, and expect the
1802 			 * upper layers to deal.
1803 			 */
1804 			DGE_EVCNT_INCR(&sc->sc_ev_rxtusum);
1805 			m->m_pkthdr.csum_flags |= M_CSUM_TCPv4|M_CSUM_UDPv4;
1806 			if (errors & RDESC_ERR_TCPE)
1807 				m->m_pkthdr.csum_flags |= M_CSUM_TCP_UDP_BAD;
1808 		}
1809 
1810 		ifp->if_ipackets++;
1811 
1812 		/* Pass this up to any BPF listeners. */
1813 		bpf_mtap(ifp, m);
1814 
1815 		/* Pass it on. */
1816 		if_percpuq_enqueue(ifp->if_percpuq, m);
1817 	}
1818 
1819 	/* Update the receive pointer. */
1820 	sc->sc_rxptr = i;
1821 
1822 	DPRINTF(DGE_DEBUG_RX,
1823 	    ("%s: RX: rxptr -> %d\n", device_xname(sc->sc_dev), i));
1824 }
1825 
1826 /*
1827  * dge_linkintr:
1828  *
1829  *	Helper; handle link interrupts.
1830  */
1831 static void
1832 dge_linkintr(struct dge_softc *sc, uint32_t icr)
1833 {
1834 	uint32_t status;
1835 
1836 	if (icr & ICR_LSC) {
1837 		status = CSR_READ(sc, DGE_STATUS);
1838 		if (status & STATUS_LINKUP) {
1839 			DPRINTF(DGE_DEBUG_LINK, ("%s: LINK: LSC -> up\n",
1840 			    device_xname(sc->sc_dev)));
1841 		} else {
1842 			DPRINTF(DGE_DEBUG_LINK, ("%s: LINK: LSC -> down\n",
1843 			    device_xname(sc->sc_dev)));
1844 		}
1845 	} else if (icr & ICR_RXSEQ) {
1846 		DPRINTF(DGE_DEBUG_LINK,
1847 		    ("%s: LINK: Receive sequence error\n",
1848 		    device_xname(sc->sc_dev)));
1849 	}
1850 	/* XXX - fix errata */
1851 }
1852 
1853 /*
1854  * dge_reset:
1855  *
1856  *	Reset the i82597 chip.
1857  */
1858 static void
1859 dge_reset(struct dge_softc *sc)
1860 {
1861 	int i;
1862 
1863 	/*
1864 	 * Do a chip reset.
1865 	 */
1866 	CSR_WRITE(sc, DGE_CTRL0, CTRL0_RST | sc->sc_ctrl0);
1867 
1868 	delay(10000);
1869 
1870 	for (i = 0; i < 1000; i++) {
1871 		if ((CSR_READ(sc, DGE_CTRL0) & CTRL0_RST) == 0)
1872 			break;
1873 		delay(20);
1874 	}
1875 
1876 	if (CSR_READ(sc, DGE_CTRL0) & CTRL0_RST)
1877 		printf("%s: WARNING: reset failed to complete\n",
1878 		    device_xname(sc->sc_dev));
1879         /*
1880          * Reset the EEPROM logic.
1881          * This will cause the chip to reread its default values,
1882 	 * which doesn't happen otherwise (errata).
1883          */
1884         CSR_WRITE(sc, DGE_CTRL1, CTRL1_EE_RST);
1885         delay(10000);
1886 }
1887 
1888 /*
1889  * dge_init:		[ifnet interface function]
1890  *
1891  *	Initialize the interface.  Must be called at splnet().
1892  */
1893 static int
1894 dge_init(struct ifnet *ifp)
1895 {
1896 	struct dge_softc *sc = ifp->if_softc;
1897 	struct dge_rxsoft *rxs;
1898 	int i, error = 0;
1899 	uint32_t reg;
1900 
1901 	/*
1902 	 * *_HDR_ALIGNED_P is constant 1 if __NO_STRICT_ALIGMENT is set.
1903 	 * There is a small but measurable benefit to avoiding the adjusment
1904 	 * of the descriptor so that the headers are aligned, for normal mtu,
1905 	 * on such platforms.  One possibility is that the DMA itself is
1906 	 * slightly more efficient if the front of the entire packet (instead
1907 	 * of the front of the headers) is aligned.
1908 	 *
1909 	 * Note we must always set align_tweak to 0 if we are using
1910 	 * jumbo frames.
1911 	 */
1912 #ifdef __NO_STRICT_ALIGNMENT
1913 	sc->sc_align_tweak = 0;
1914 #else
1915 	if ((ifp->if_mtu + ETHER_HDR_LEN + ETHER_CRC_LEN) > (MCLBYTES - 2))
1916 		sc->sc_align_tweak = 0;
1917 	else
1918 		sc->sc_align_tweak = 2;
1919 #endif /* __NO_STRICT_ALIGNMENT */
1920 
1921 	/* Cancel any pending I/O. */
1922 	dge_stop(ifp, 0);
1923 
1924 	/* Reset the chip to a known state. */
1925 	dge_reset(sc);
1926 
1927 	/* Initialize the transmit descriptor ring. */
1928 	memset(sc->sc_txdescs, 0, sizeof(sc->sc_txdescs));
1929 	DGE_CDTXSYNC(sc, 0, DGE_NTXDESC,
1930 	    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1931 	sc->sc_txfree = DGE_NTXDESC;
1932 	sc->sc_txnext = 0;
1933 
1934 	sc->sc_txctx_ipcs = 0xffffffff;
1935 	sc->sc_txctx_tucs = 0xffffffff;
1936 
1937 	CSR_WRITE(sc, DGE_TDBAH, 0);
1938 	CSR_WRITE(sc, DGE_TDBAL, DGE_CDTXADDR(sc, 0));
1939 	CSR_WRITE(sc, DGE_TDLEN, sizeof(sc->sc_txdescs));
1940 	CSR_WRITE(sc, DGE_TDH, 0);
1941 	CSR_WRITE(sc, DGE_TDT, 0);
1942 	CSR_WRITE(sc, DGE_TIDV, TIDV);
1943 
1944 #if 0
1945 	CSR_WRITE(sc, DGE_TXDCTL, TXDCTL_PTHRESH(0) |
1946 	    TXDCTL_HTHRESH(0) | TXDCTL_WTHRESH(0));
1947 #endif
1948 	CSR_WRITE(sc, DGE_RXDCTL,
1949 	    RXDCTL_PTHRESH(RXDCTL_PTHRESH_VAL) |
1950 	    RXDCTL_HTHRESH(RXDCTL_HTHRESH_VAL) |
1951 	    RXDCTL_WTHRESH(RXDCTL_WTHRESH_VAL));
1952 
1953 	/* Initialize the transmit job descriptors. */
1954 	for (i = 0; i < DGE_TXQUEUELEN; i++)
1955 		sc->sc_txsoft[i].txs_mbuf = NULL;
1956 	sc->sc_txsfree = DGE_TXQUEUELEN;
1957 	sc->sc_txsnext = 0;
1958 	sc->sc_txsdirty = 0;
1959 
1960 	/*
1961 	 * Initialize the receive descriptor and receive job
1962 	 * descriptor rings.
1963 	 */
1964 	CSR_WRITE(sc, DGE_RDBAH, 0);
1965 	CSR_WRITE(sc, DGE_RDBAL, DGE_CDRXADDR(sc, 0));
1966 	CSR_WRITE(sc, DGE_RDLEN, sizeof(sc->sc_rxdescs));
1967 	CSR_WRITE(sc, DGE_RDH, DGE_RXSPACE);
1968 	CSR_WRITE(sc, DGE_RDT, 0);
1969 	CSR_WRITE(sc, DGE_RDTR, RDTR | 0x80000000);
1970 	CSR_WRITE(sc, DGE_FCRTL, FCRTL | FCRTL_XONE);
1971 	CSR_WRITE(sc, DGE_FCRTH, FCRTH);
1972 
1973 	for (i = 0; i < DGE_NRXDESC; i++) {
1974 		rxs = &sc->sc_rxsoft[i];
1975 		if (rxs->rxs_mbuf == NULL) {
1976 			if ((error = dge_add_rxbuf(sc, i)) != 0) {
1977 				printf("%s: unable to allocate or map rx "
1978 				    "buffer %d, error = %d\n",
1979 				    device_xname(sc->sc_dev), i, error);
1980 				/*
1981 				 * XXX Should attempt to run with fewer receive
1982 				 * XXX buffers instead of just failing.
1983 				 */
1984 				dge_rxdrain(sc);
1985 				goto out;
1986 			}
1987 		}
1988 		DGE_INIT_RXDESC(sc, i);
1989 	}
1990 	sc->sc_rxptr = DGE_RXSPACE;
1991 	sc->sc_rxdiscard = 0;
1992 	DGE_RXCHAIN_RESET(sc);
1993 
1994 	if (sc->sc_ethercom.ec_capabilities & ETHERCAP_JUMBO_MTU) {
1995 		sc->sc_ctrl0 |= CTRL0_JFE;
1996 		CSR_WRITE(sc, DGE_MFS, ETHER_MAX_LEN_JUMBO << 16);
1997 	}
1998 
1999 	/* Write the control registers. */
2000 	CSR_WRITE(sc, DGE_CTRL0, sc->sc_ctrl0);
2001 
2002 	/*
2003 	 * Set up checksum offload parameters.
2004 	 */
2005 	reg = CSR_READ(sc, DGE_RXCSUM);
2006 	if (ifp->if_capenable & IFCAP_CSUM_IPv4_Rx)
2007 		reg |= RXCSUM_IPOFL;
2008 	else
2009 		reg &= ~RXCSUM_IPOFL;
2010 	if (ifp->if_capenable & (IFCAP_CSUM_TCPv4_Rx | IFCAP_CSUM_UDPv4_Rx))
2011 		reg |= RXCSUM_IPOFL | RXCSUM_TUOFL;
2012 	else {
2013 		reg &= ~RXCSUM_TUOFL;
2014 		if ((ifp->if_capenable & IFCAP_CSUM_IPv4_Rx) == 0)
2015 			reg &= ~RXCSUM_IPOFL;
2016 	}
2017 	CSR_WRITE(sc, DGE_RXCSUM, reg);
2018 
2019 	/*
2020 	 * Set up the interrupt registers.
2021 	 */
2022 	CSR_WRITE(sc, DGE_IMC, 0xffffffffU);
2023 	sc->sc_icr = ICR_TXDW | ICR_LSC | ICR_RXSEQ | ICR_RXDMT0 |
2024 	    ICR_RXO | ICR_RXT0;
2025 
2026 	CSR_WRITE(sc, DGE_IMS, sc->sc_icr);
2027 
2028 	/*
2029 	 * Set up the transmit control register.
2030 	 */
2031 	sc->sc_tctl = TCTL_TCE|TCTL_TPDE|TCTL_TXEN;
2032 	CSR_WRITE(sc, DGE_TCTL, sc->sc_tctl);
2033 
2034 	/*
2035 	 * Set up the receive control register; we actually program
2036 	 * the register when we set the receive filter.  Use multicast
2037 	 * address offset type 0.
2038 	 */
2039 	sc->sc_mchash_type = 0;
2040 
2041 	sc->sc_rctl = RCTL_RXEN | RCTL_RDMTS_12 | RCTL_RPDA_MC |
2042 	    RCTL_CFF | RCTL_SECRC | RCTL_MO(sc->sc_mchash_type);
2043 
2044 #ifdef DGE_OFFBYONE_RXBUG
2045 	sc->sc_rctl |= RCTL_BSIZE_16k;
2046 #else
2047 	switch(MCLBYTES) {
2048 	case 2048:
2049 		sc->sc_rctl |= RCTL_BSIZE_2k;
2050 		break;
2051 	case 4096:
2052 		sc->sc_rctl |= RCTL_BSIZE_4k;
2053 		break;
2054 	case 8192:
2055 		sc->sc_rctl |= RCTL_BSIZE_8k;
2056 		break;
2057 	case 16384:
2058 		sc->sc_rctl |= RCTL_BSIZE_16k;
2059 		break;
2060 	default:
2061 		panic("dge_init: MCLBYTES %d unsupported", MCLBYTES);
2062 	}
2063 #endif
2064 
2065 	/* Set the receive filter. */
2066 	/* Also sets RCTL */
2067 	dge_set_filter(sc);
2068 
2069 	/* ...all done! */
2070 	ifp->if_flags |= IFF_RUNNING;
2071 	ifp->if_flags &= ~IFF_OACTIVE;
2072 
2073  out:
2074 	if (error)
2075 		printf("%s: interface not running\n", device_xname(sc->sc_dev));
2076 	return (error);
2077 }
2078 
2079 /*
2080  * dge_rxdrain:
2081  *
2082  *	Drain the receive queue.
2083  */
2084 static void
2085 dge_rxdrain(struct dge_softc *sc)
2086 {
2087 	struct dge_rxsoft *rxs;
2088 	int i;
2089 
2090 	for (i = 0; i < DGE_NRXDESC; i++) {
2091 		rxs = &sc->sc_rxsoft[i];
2092 		if (rxs->rxs_mbuf != NULL) {
2093 			bus_dmamap_unload(sc->sc_dmat, rxs->rxs_dmamap);
2094 			m_freem(rxs->rxs_mbuf);
2095 			rxs->rxs_mbuf = NULL;
2096 		}
2097 	}
2098 }
2099 
2100 /*
2101  * dge_stop:		[ifnet interface function]
2102  *
2103  *	Stop transmission on the interface.
2104  */
2105 static void
2106 dge_stop(struct ifnet *ifp, int disable)
2107 {
2108 	struct dge_softc *sc = ifp->if_softc;
2109 	struct dge_txsoft *txs;
2110 	int i;
2111 
2112 	/* Stop the transmit and receive processes. */
2113 	CSR_WRITE(sc, DGE_TCTL, 0);
2114 	CSR_WRITE(sc, DGE_RCTL, 0);
2115 
2116 	/* Release any queued transmit buffers. */
2117 	for (i = 0; i < DGE_TXQUEUELEN; i++) {
2118 		txs = &sc->sc_txsoft[i];
2119 		if (txs->txs_mbuf != NULL) {
2120 			bus_dmamap_unload(sc->sc_dmat, txs->txs_dmamap);
2121 			m_freem(txs->txs_mbuf);
2122 			txs->txs_mbuf = NULL;
2123 		}
2124 	}
2125 
2126 	/* Mark the interface as down and cancel the watchdog timer. */
2127 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
2128 	ifp->if_timer = 0;
2129 
2130 	if (disable)
2131 		dge_rxdrain(sc);
2132 }
2133 
2134 /*
2135  * dge_add_rxbuf:
2136  *
2137  *	Add a receive buffer to the indiciated descriptor.
2138  */
2139 static int
2140 dge_add_rxbuf(struct dge_softc *sc, int idx)
2141 {
2142 	struct dge_rxsoft *rxs = &sc->sc_rxsoft[idx];
2143 	struct mbuf *m;
2144 	int error;
2145 #ifdef DGE_OFFBYONE_RXBUG
2146 	void *buf;
2147 #endif
2148 
2149 	MGETHDR(m, M_DONTWAIT, MT_DATA);
2150 	if (m == NULL)
2151 		return (ENOBUFS);
2152 
2153 #ifdef DGE_OFFBYONE_RXBUG
2154 	if ((buf = dge_getbuf(sc)) == NULL)
2155 		return ENOBUFS;
2156 
2157 	m->m_len = m->m_pkthdr.len = DGE_BUFFER_SIZE;
2158 	MEXTADD(m, buf, DGE_BUFFER_SIZE, M_DEVBUF, dge_freebuf, sc);
2159 	m->m_flags |= M_EXT_RW;
2160 
2161 	if (rxs->rxs_mbuf != NULL)
2162 		bus_dmamap_unload(sc->sc_dmat, rxs->rxs_dmamap);
2163 	rxs->rxs_mbuf = m;
2164 
2165 	error = bus_dmamap_load(sc->sc_dmat, rxs->rxs_dmamap, buf,
2166 	    DGE_BUFFER_SIZE, NULL, BUS_DMA_READ|BUS_DMA_NOWAIT);
2167 #else
2168 	MCLGET(m, M_DONTWAIT);
2169 	if ((m->m_flags & M_EXT) == 0) {
2170 		m_freem(m);
2171 		return (ENOBUFS);
2172 	}
2173 
2174 	if (rxs->rxs_mbuf != NULL)
2175 		bus_dmamap_unload(sc->sc_dmat, rxs->rxs_dmamap);
2176 
2177 	rxs->rxs_mbuf = m;
2178 
2179 	m->m_len = m->m_pkthdr.len = m->m_ext.ext_size;
2180 	error = bus_dmamap_load_mbuf(sc->sc_dmat, rxs->rxs_dmamap, m,
2181 	    BUS_DMA_READ|BUS_DMA_NOWAIT);
2182 #endif
2183 	if (error) {
2184 		printf("%s: unable to load rx DMA map %d, error = %d\n",
2185 		    device_xname(sc->sc_dev), idx, error);
2186 		panic("dge_add_rxbuf");	/* XXX XXX XXX */
2187 	}
2188 	bus_dmamap_sync(sc->sc_dmat, rxs->rxs_dmamap, 0,
2189 	    rxs->rxs_dmamap->dm_mapsize, BUS_DMASYNC_PREREAD);
2190 
2191 	return (0);
2192 }
2193 
2194 /*
2195  * dge_set_ral:
2196  *
2197  *	Set an entry in the receive address list.
2198  */
2199 static void
2200 dge_set_ral(struct dge_softc *sc, const uint8_t *enaddr, int idx)
2201 {
2202 	uint32_t ral_lo, ral_hi;
2203 
2204 	if (enaddr != NULL) {
2205 		ral_lo = enaddr[0] | (enaddr[1] << 8) | (enaddr[2] << 16) |
2206 		    (enaddr[3] << 24);
2207 		ral_hi = enaddr[4] | (enaddr[5] << 8);
2208 		ral_hi |= RAH_AV;
2209 	} else {
2210 		ral_lo = 0;
2211 		ral_hi = 0;
2212 	}
2213 	CSR_WRITE(sc, RA_ADDR(DGE_RAL, idx), ral_lo);
2214 	CSR_WRITE(sc, RA_ADDR(DGE_RAH, idx), ral_hi);
2215 }
2216 
2217 /*
2218  * dge_mchash:
2219  *
2220  *	Compute the hash of the multicast address for the 4096-bit
2221  *	multicast filter.
2222  */
2223 static uint32_t
2224 dge_mchash(struct dge_softc *sc, const uint8_t *enaddr)
2225 {
2226 	static const int lo_shift[4] = { 4, 3, 2, 0 };
2227 	static const int hi_shift[4] = { 4, 5, 6, 8 };
2228 	uint32_t hash;
2229 
2230 	hash = (enaddr[4] >> lo_shift[sc->sc_mchash_type]) |
2231 	    (((uint16_t) enaddr[5]) << hi_shift[sc->sc_mchash_type]);
2232 
2233 	return (hash & 0xfff);
2234 }
2235 
2236 /*
2237  * dge_set_filter:
2238  *
2239  *	Set up the receive filter.
2240  */
2241 static void
2242 dge_set_filter(struct dge_softc *sc)
2243 {
2244 	struct ethercom *ec = &sc->sc_ethercom;
2245 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
2246 	struct ether_multi *enm;
2247 	struct ether_multistep step;
2248 	uint32_t hash, reg, bit;
2249 	int i;
2250 
2251 	sc->sc_rctl &= ~(RCTL_BAM | RCTL_UPE | RCTL_MPE);
2252 
2253 	if (ifp->if_flags & IFF_BROADCAST)
2254 		sc->sc_rctl |= RCTL_BAM;
2255 	if (ifp->if_flags & IFF_PROMISC) {
2256 		sc->sc_rctl |= RCTL_UPE;
2257 		goto allmulti;
2258 	}
2259 
2260 	/*
2261 	 * Set the station address in the first RAL slot, and
2262 	 * clear the remaining slots.
2263 	 */
2264 	dge_set_ral(sc, CLLADDR(ifp->if_sadl), 0);
2265 	for (i = 1; i < RA_TABSIZE; i++)
2266 		dge_set_ral(sc, NULL, i);
2267 
2268 	/* Clear out the multicast table. */
2269 	for (i = 0; i < MC_TABSIZE; i++)
2270 		CSR_WRITE(sc, DGE_MTA + (i << 2), 0);
2271 
2272 	ETHER_FIRST_MULTI(step, ec, enm);
2273 	while (enm != NULL) {
2274 		if (memcmp(enm->enm_addrlo, enm->enm_addrhi, ETHER_ADDR_LEN)) {
2275 			/*
2276 			 * We must listen to a range of multicast addresses.
2277 			 * For now, just accept all multicasts, rather than
2278 			 * trying to set only those filter bits needed to match
2279 			 * the range.  (At this time, the only use of address
2280 			 * ranges is for IP multicast routing, for which the
2281 			 * range is big enough to require all bits set.)
2282 			 */
2283 			goto allmulti;
2284 		}
2285 
2286 		hash = dge_mchash(sc, enm->enm_addrlo);
2287 
2288 		reg = (hash >> 5) & 0x7f;
2289 		bit = hash & 0x1f;
2290 
2291 		hash = CSR_READ(sc, DGE_MTA + (reg << 2));
2292 		hash |= 1U << bit;
2293 
2294 		CSR_WRITE(sc, DGE_MTA + (reg << 2), hash);
2295 
2296 		ETHER_NEXT_MULTI(step, enm);
2297 	}
2298 
2299 	ifp->if_flags &= ~IFF_ALLMULTI;
2300 	goto setit;
2301 
2302  allmulti:
2303 	ifp->if_flags |= IFF_ALLMULTI;
2304 	sc->sc_rctl |= RCTL_MPE;
2305 
2306  setit:
2307 	CSR_WRITE(sc, DGE_RCTL, sc->sc_rctl);
2308 }
2309 
2310 /*
2311  * Read in the EEPROM info and verify checksum.
2312  */
2313 int
2314 dge_read_eeprom(struct dge_softc *sc)
2315 {
2316 	uint16_t cksum;
2317 	int i;
2318 
2319 	cksum = 0;
2320 	for (i = 0; i < EEPROM_SIZE; i++) {
2321 		sc->sc_eeprom[i] = dge_eeprom_word(sc, i);
2322 		cksum += sc->sc_eeprom[i];
2323 	}
2324 	return cksum != EEPROM_CKSUM;
2325 }
2326 
2327 
2328 /*
2329  * Read a 16-bit word from address addr in the serial EEPROM.
2330  */
2331 uint16_t
2332 dge_eeprom_word(struct dge_softc *sc, int addr)
2333 {
2334 	uint32_t reg;
2335 	uint16_t rval = 0;
2336 	int i;
2337 
2338 	reg = CSR_READ(sc, DGE_EECD) & ~(EECD_SK|EECD_DI|EECD_CS);
2339 
2340 	/* Lower clock pulse (and data in to chip) */
2341 	CSR_WRITE(sc, DGE_EECD, reg);
2342 	/* Select chip */
2343 	CSR_WRITE(sc, DGE_EECD, reg|EECD_CS);
2344 
2345 	/* Send read command */
2346 	dge_eeprom_clockout(sc, 1);
2347 	dge_eeprom_clockout(sc, 1);
2348 	dge_eeprom_clockout(sc, 0);
2349 
2350 	/* Send address */
2351 	for (i = 5; i >= 0; i--)
2352 		dge_eeprom_clockout(sc, (addr >> i) & 1);
2353 
2354 	/* Read data */
2355 	for (i = 0; i < 16; i++) {
2356 		rval <<= 1;
2357 		rval |= dge_eeprom_clockin(sc);
2358 	}
2359 
2360 	/* Deselect chip */
2361 	CSR_WRITE(sc, DGE_EECD, reg);
2362 
2363 	return rval;
2364 }
2365 
2366 /*
2367  * Clock out a single bit to the EEPROM.
2368  */
2369 void
2370 dge_eeprom_clockout(struct dge_softc *sc, int bit)
2371 {
2372 	int reg;
2373 
2374 	reg = CSR_READ(sc, DGE_EECD) & ~(EECD_DI|EECD_SK);
2375 	if (bit)
2376 		reg |= EECD_DI;
2377 
2378 	CSR_WRITE(sc, DGE_EECD, reg);
2379 	delay(2);
2380 	CSR_WRITE(sc, DGE_EECD, reg|EECD_SK);
2381 	delay(2);
2382 	CSR_WRITE(sc, DGE_EECD, reg);
2383 	delay(2);
2384 }
2385 
2386 /*
2387  * Clock in a single bit from EEPROM.
2388  */
2389 int
2390 dge_eeprom_clockin(struct dge_softc *sc)
2391 {
2392 	int reg, rv;
2393 
2394 	reg = CSR_READ(sc, DGE_EECD) & ~(EECD_DI|EECD_DO|EECD_SK);
2395 
2396 	CSR_WRITE(sc, DGE_EECD, reg|EECD_SK); /* Raise clock */
2397 	delay(2);
2398 	rv = (CSR_READ(sc, DGE_EECD) & EECD_DO) != 0; /* Get bit */
2399 	CSR_WRITE(sc, DGE_EECD, reg); /* Lower clock */
2400 	delay(2);
2401 
2402 	return rv;
2403 }
2404 
2405 static void
2406 dge_xgmii_mediastatus(struct ifnet *ifp, struct ifmediareq *ifmr)
2407 {
2408 	struct dge_softc *sc = ifp->if_softc;
2409 
2410 	ifmr->ifm_status = IFM_AVALID;
2411 	if (sc->sc_dgep->dgep_flags & DGEP_F_10G_SR ) {
2412 		ifmr->ifm_active = IFM_ETHER|IFM_10G_SR;
2413 	} else {
2414 		ifmr->ifm_active = IFM_ETHER|IFM_10G_LR;
2415 	}
2416 
2417 	if (CSR_READ(sc, DGE_STATUS) & STATUS_LINKUP)
2418 		ifmr->ifm_status |= IFM_ACTIVE;
2419 }
2420 
2421 static inline int
2422 phwait(struct dge_softc *sc, int p, int r, int d, int type)
2423 {
2424         int i, mdic;
2425 
2426         CSR_WRITE(sc, DGE_MDIO,
2427 	    MDIO_PHY(p) | MDIO_REG(r) | MDIO_DEV(d) | type | MDIO_CMD);
2428         for (i = 0; i < 10; i++) {
2429                 delay(10);
2430                 if (((mdic = CSR_READ(sc, DGE_MDIO)) & MDIO_CMD) == 0)
2431                         break;
2432         }
2433         return mdic;
2434 }
2435 
2436 static void
2437 dge_xgmii_writereg(struct dge_softc *sc, int phy, int reg, int val)
2438 {
2439 	int mdic;
2440 
2441 	CSR_WRITE(sc, DGE_MDIRW, val);
2442 	if (((mdic = phwait(sc, phy, reg, 1, MDIO_ADDR)) & MDIO_CMD)) {
2443 		printf("%s: address cycle timeout; phy %d reg %d\n",
2444 		    device_xname(sc->sc_dev), phy, reg);
2445 		return;
2446 	}
2447 	if (((mdic = phwait(sc, phy, reg, 1, MDIO_WRITE)) & MDIO_CMD)) {
2448 		printf("%s: write cycle timeout; phy %d reg %d\n",
2449 		    device_xname(sc->sc_dev), phy, reg);
2450 		return;
2451 	}
2452 }
2453 
2454 static void
2455 dge_xgmii_reset(struct dge_softc *sc)
2456 {
2457 	dge_xgmii_writereg(sc, 0, 0, BMCR_RESET);
2458 }
2459 
2460 static int
2461 dge_xgmii_mediachange(struct ifnet *ifp)
2462 {
2463 	return 0;
2464 }
2465