if_dge.c revision 1.57
1/*	$NetBSD: if_dge.c,v 1.57 2020/01/30 05:24:53 thorpej 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 1.57 2020/01/30 05:24:53 thorpej 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#include <sys/rndsource.h>
99
100#include <net/if.h>
101#include <net/if_dl.h>
102#include <net/if_media.h>
103#include <net/if_ether.h>
104#include <net/bpf.h>
105
106#include <netinet/in.h>			/* XXX for struct ip */
107#include <netinet/in_systm.h>		/* XXX for struct ip */
108#include <netinet/ip.h>			/* XXX for struct ip */
109#include <netinet/tcp.h>		/* XXX for struct tcphdr */
110
111#include <sys/bus.h>
112#include <sys/intr.h>
113#include <machine/endian.h>
114
115#include <dev/mii/mii.h>
116#include <dev/mii/miivar.h>
117#include <dev/mii/mii_bitbang.h>
118
119#include <dev/pci/pcireg.h>
120#include <dev/pci/pcivar.h>
121#include <dev/pci/pcidevs.h>
122
123#include <dev/pci/if_dgereg.h>
124
125/*
126 * The receive engine may sometimes become off-by-one when writing back
127 * chained descriptors.	 Avoid this by allocating a large chunk of
128 * memory and use if instead (to avoid chained descriptors).
129 * This only happens with chained descriptors under heavy load.
130 */
131#define DGE_OFFBYONE_RXBUG
132
133#define DGE_EVENT_COUNTERS
134#define DGE_DEBUG
135
136#ifdef DGE_DEBUG
137#define DGE_DEBUG_LINK		0x01
138#define DGE_DEBUG_TX		0x02
139#define DGE_DEBUG_RX		0x04
140#define DGE_DEBUG_CKSUM		0x08
141int	dge_debug = 0;
142
143#define DPRINTF(x, y)	if (dge_debug & (x)) printf y
144#else
145#define DPRINTF(x, y)	/* nothing */
146#endif /* DGE_DEBUG */
147
148/*
149 * Transmit descriptor list size. We allow up to 100 DMA segments per
150 * packet (Intel reports of jumbo frame packets with as
151 * many as 80 DMA segments when using 16k buffers).
152 */
153#define DGE_NTXSEGS		100
154#define DGE_IFQUEUELEN		20000
155#define DGE_TXQUEUELEN		2048
156#define DGE_TXQUEUELEN_MASK	(DGE_TXQUEUELEN - 1)
157#define DGE_TXQUEUE_GC		(DGE_TXQUEUELEN / 8)
158#define DGE_NTXDESC		1024
159#define DGE_NTXDESC_MASK		(DGE_NTXDESC - 1)
160#define DGE_NEXTTX(x)		(((x) + 1) & DGE_NTXDESC_MASK)
161#define DGE_NEXTTXS(x)		(((x) + 1) & DGE_TXQUEUELEN_MASK)
162
163/*
164 * Receive descriptor list size.
165 * Packet is of size MCLBYTES, and for jumbo packets buffers may
166 * be chained.	Due to the nature of the card (high-speed), keep this
167 * ring large. With 2k buffers the ring can store 400 jumbo packets,
168 * which at full speed will be received in just under 3ms.
169 */
170#define DGE_NRXDESC		2048
171#define DGE_NRXDESC_MASK	(DGE_NRXDESC - 1)
172#define DGE_NEXTRX(x)		(((x) + 1) & DGE_NRXDESC_MASK)
173/*
174 * # of descriptors between head and written descriptors.
175 * This is to work-around two erratas.
176 */
177#define DGE_RXSPACE		10
178#define DGE_PREVRX(x)		(((x) - DGE_RXSPACE) & DGE_NRXDESC_MASK)
179/*
180 * Receive descriptor fetch threshholds. These are values recommended
181 * by Intel, do not touch them unless you know what you are doing.
182 */
183#define RXDCTL_PTHRESH_VAL	128
184#define RXDCTL_HTHRESH_VAL	16
185#define RXDCTL_WTHRESH_VAL	16
186
187
188/*
189 * Tweakable parameters; default values.
190 */
191#define FCRTH	0x30000 /* Send XOFF water mark */
192#define FCRTL	0x28000 /* Send XON water mark */
193#define RDTR	0x20	/* Interrupt delay after receive, .8192us units */
194#define TIDV	0x20	/* Interrupt delay after send, .8192us units */
195
196/*
197 * Control structures are DMA'd to the i82597 chip.  We allocate them in
198 * a single clump that maps to a single DMA segment to make serveral things
199 * easier.
200 */
201struct dge_control_data {
202	/*
203	 * The transmit descriptors.
204	 */
205	struct dge_tdes wcd_txdescs[DGE_NTXDESC];
206
207	/*
208	 * The receive descriptors.
209	 */
210	struct dge_rdes wcd_rxdescs[DGE_NRXDESC];
211};
212
213#define DGE_CDOFF(x)	offsetof(struct dge_control_data, x)
214#define DGE_CDTXOFF(x)	DGE_CDOFF(wcd_txdescs[(x)])
215#define DGE_CDRXOFF(x)	DGE_CDOFF(wcd_rxdescs[(x)])
216
217/*
218 * The DGE interface have a higher max MTU size than normal jumbo frames.
219 */
220#define DGE_MAX_MTU	16288	/* Max MTU size for this interface */
221
222/*
223 * Software state for transmit jobs.
224 */
225struct dge_txsoft {
226	struct mbuf *txs_mbuf;		/* head of our mbuf chain */
227	bus_dmamap_t txs_dmamap;	/* our DMA map */
228	int txs_firstdesc;		/* first descriptor in packet */
229	int txs_lastdesc;		/* last descriptor in packet */
230	int txs_ndesc;			/* # of descriptors used */
231};
232
233/*
234 * Software state for receive buffers.	Each descriptor gets a
235 * 2k (MCLBYTES) buffer and a DMA map.	For packets which fill
236 * more than one buffer, we chain them together.
237 */
238struct dge_rxsoft {
239	struct mbuf *rxs_mbuf;		/* head of our mbuf chain */
240	bus_dmamap_t rxs_dmamap;	/* our DMA map */
241};
242
243/*
244 * Software state per device.
245 */
246struct dge_softc {
247	device_t sc_dev;		/* generic device information */
248	bus_space_tag_t sc_st;		/* bus space tag */
249	bus_space_handle_t sc_sh;	/* bus space handle */
250	bus_dma_tag_t sc_dmat;		/* bus DMA tag */
251	struct ethercom sc_ethercom;	/* ethernet common data */
252
253	int sc_flags;			/* flags; see below */
254	int sc_bus_speed;		/* PCI/PCIX bus speed */
255	int sc_pcix_offset;		/* PCIX capability register offset */
256
257	const struct dge_product *sc_dgep; /* Pointer to the dge_product entry */
258	pci_chipset_tag_t sc_pc;
259	pcitag_t sc_pt;
260	int sc_mmrbc;			/* Max PCIX memory read byte count */
261
262	void *sc_ih;			/* interrupt cookie */
263
264	struct ifmedia sc_media;
265
266	bus_dmamap_t sc_cddmamap;	/* control data DMA map */
267#define sc_cddma	sc_cddmamap->dm_segs[0].ds_addr
268
269	int		sc_align_tweak;
270
271	/*
272	 * Software state for the transmit and receive descriptors.
273	 */
274	struct dge_txsoft sc_txsoft[DGE_TXQUEUELEN];
275	struct dge_rxsoft sc_rxsoft[DGE_NRXDESC];
276
277	/*
278	 * Control data structures.
279	 */
280	struct dge_control_data *sc_control_data;
281#define sc_txdescs	sc_control_data->wcd_txdescs
282#define sc_rxdescs	sc_control_data->wcd_rxdescs
283
284#ifdef DGE_EVENT_COUNTERS
285	/* Event counters. */
286	struct evcnt sc_ev_txsstall;	/* Tx stalled due to no txs */
287	struct evcnt sc_ev_txdstall;	/* Tx stalled due to no txd */
288	struct evcnt sc_ev_txforceintr; /* Tx interrupts forced */
289	struct evcnt sc_ev_txdw;	/* Tx descriptor interrupts */
290	struct evcnt sc_ev_txqe;	/* Tx queue empty interrupts */
291	struct evcnt sc_ev_rxintr;	/* Rx interrupts */
292	struct evcnt sc_ev_linkintr;	/* Link interrupts */
293
294	struct evcnt sc_ev_rxipsum;	/* IP checksums checked in-bound */
295	struct evcnt sc_ev_rxtusum;	/* TCP/UDP cksums checked in-bound */
296	struct evcnt sc_ev_txipsum;	/* IP checksums comp. out-bound */
297	struct evcnt sc_ev_txtusum;	/* TCP/UDP cksums comp. out-bound */
298
299	struct evcnt sc_ev_txctx_init;	/* Tx cksum context cache initialized */
300	struct evcnt sc_ev_txctx_hit;	/* Tx cksum context cache hit */
301	struct evcnt sc_ev_txctx_miss;	/* Tx cksum context cache miss */
302
303	struct evcnt sc_ev_txseg[DGE_NTXSEGS]; /* Tx packets w/ N segments */
304	struct evcnt sc_ev_txdrop;	/* Tx packets dropped (too many segs) */
305#endif /* DGE_EVENT_COUNTERS */
306
307	int	sc_txfree;		/* number of free Tx descriptors */
308	int	sc_txnext;		/* next ready Tx descriptor */
309
310	int	sc_txsfree;		/* number of free Tx jobs */
311	int	sc_txsnext;		/* next free Tx job */
312	int	sc_txsdirty;		/* dirty Tx jobs */
313
314	uint32_t sc_txctx_ipcs;		/* cached Tx IP cksum ctx */
315	uint32_t sc_txctx_tucs;		/* cached Tx TCP/UDP cksum ctx */
316
317	int	sc_rxptr;		/* next ready Rx descriptor/queue ent */
318	int	sc_rxdiscard;
319	int	sc_rxlen;
320	struct mbuf *sc_rxhead;
321	struct mbuf *sc_rxtail;
322	struct mbuf **sc_rxtailp;
323
324	uint32_t sc_ctrl0;		/* prototype CTRL0 register */
325	uint32_t sc_icr;		/* prototype interrupt bits */
326	uint32_t sc_tctl;		/* prototype TCTL register */
327	uint32_t sc_rctl;		/* prototype RCTL register */
328
329	int sc_mchash_type;		/* multicast filter offset */
330
331	uint16_t sc_eeprom[EEPROM_SIZE];
332
333	krndsource_t rnd_source; /* random source */
334#ifdef DGE_OFFBYONE_RXBUG
335	void *sc_bugbuf;
336	SLIST_HEAD(, rxbugentry) sc_buglist;
337	bus_dmamap_t sc_bugmap;
338	struct rxbugentry *sc_entry;
339#endif
340};
341
342#define DGE_RXCHAIN_RESET(sc)						\
343do {									\
344	(sc)->sc_rxtailp = &(sc)->sc_rxhead;				\
345	*(sc)->sc_rxtailp = NULL;					\
346	(sc)->sc_rxlen = 0;						\
347} while (/*CONSTCOND*/0)
348
349#define DGE_RXCHAIN_LINK(sc, m)						\
350do {									\
351	*(sc)->sc_rxtailp = (sc)->sc_rxtail = (m);			\
352	(sc)->sc_rxtailp = &(m)->m_next;				\
353} while (/*CONSTCOND*/0)
354
355/* sc_flags */
356#define DGE_F_BUS64		0x20	/* bus is 64-bit */
357#define DGE_F_PCIX		0x40	/* bus is PCI-X */
358
359#ifdef DGE_EVENT_COUNTERS
360#define DGE_EVCNT_INCR(ev)	(ev)->ev_count++
361#else
362#define DGE_EVCNT_INCR(ev)	/* nothing */
363#endif
364
365#define CSR_READ(sc, reg)						\
366	bus_space_read_4((sc)->sc_st, (sc)->sc_sh, (reg))
367#define CSR_WRITE(sc, reg, val)						\
368	bus_space_write_4((sc)->sc_st, (sc)->sc_sh, (reg), (val))
369
370#define DGE_CDTXADDR(sc, x)	((sc)->sc_cddma + DGE_CDTXOFF((x)))
371#define DGE_CDRXADDR(sc, x)	((sc)->sc_cddma + DGE_CDRXOFF((x)))
372
373#define DGE_CDTXSYNC(sc, x, n, ops)					\
374do {									\
375	int __x, __n;							\
376									\
377	__x = (x);							\
378	__n = (n);							\
379									\
380	/* If it will wrap around, sync to the end of the ring. */	\
381	if ((__x + __n) > DGE_NTXDESC) {				\
382		bus_dmamap_sync((sc)->sc_dmat, (sc)->sc_cddmamap,	\
383		    DGE_CDTXOFF(__x), sizeof(struct dge_tdes) *		\
384		    (DGE_NTXDESC - __x), (ops));			\
385		__n -= (DGE_NTXDESC - __x);				\
386		__x = 0;						\
387	}								\
388									\
389	/* Now sync whatever is left. */				\
390	bus_dmamap_sync((sc)->sc_dmat, (sc)->sc_cddmamap,		\
391	    DGE_CDTXOFF(__x), sizeof(struct dge_tdes) * __n, (ops));	\
392} while (/*CONSTCOND*/0)
393
394#define DGE_CDRXSYNC(sc, x, ops)						\
395do {									\
396	bus_dmamap_sync((sc)->sc_dmat, (sc)->sc_cddmamap,		\
397	   DGE_CDRXOFF((x)), sizeof(struct dge_rdes), (ops));		\
398} while (/*CONSTCOND*/0)
399
400#ifdef DGE_OFFBYONE_RXBUG
401#define DGE_INIT_RXDESC(sc, x)						\
402do {									\
403	struct dge_rxsoft *__rxs = &(sc)->sc_rxsoft[(x)];		\
404	struct dge_rdes *__rxd = &(sc)->sc_rxdescs[(x)];		\
405	struct mbuf *__m = __rxs->rxs_mbuf;				\
406									\
407	__rxd->dr_baddrl = htole32(sc->sc_bugmap->dm_segs[0].ds_addr +	\
408	    (mtod((__m), char *) - (char *)sc->sc_bugbuf));		\
409	__rxd->dr_baddrh = 0;						\
410	__rxd->dr_len = 0;						\
411	__rxd->dr_cksum = 0;						\
412	__rxd->dr_status = 0;						\
413	__rxd->dr_errors = 0;						\
414	__rxd->dr_special = 0;						\
415	DGE_CDRXSYNC((sc), (x), BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); \
416									\
417	CSR_WRITE((sc), DGE_RDT, (x));					\
418} while (/*CONSTCOND*/0)
419#else
420#define DGE_INIT_RXDESC(sc, x)						\
421do {									\
422	struct dge_rxsoft *__rxs = &(sc)->sc_rxsoft[(x)];		\
423	struct dge_rdes *__rxd = &(sc)->sc_rxdescs[(x)];		\
424	struct mbuf *__m = __rxs->rxs_mbuf;				\
425									\
426	/*								\
427	 * Note: We scoot the packet forward 2 bytes in the buffer	\
428	 * so that the payload after the Ethernet header is aligned	\
429	 * to a 4-byte boundary.					\
430	 *								\
431	 * XXX BRAINDAMAGE ALERT!					\
432	 * The stupid chip uses the same size for every buffer, which	\
433	 * is set in the Receive Control register.  We are using the 2K \
434	 * size option, but what we REALLY want is (2K - 2)!  For this	\
435	 * reason, we can't "scoot" packets longer than the standard	\
436	 * Ethernet MTU.  On strict-alignment platforms, if the total	\
437	 * size exceeds (2K - 2) we set align_tweak to 0 and let	\
438	 * the upper layer copy the headers.				\
439	 */								\
440	__m->m_data = __m->m_ext.ext_buf + (sc)->sc_align_tweak;	\
441									\
442	__rxd->dr_baddrl =						\
443	    htole32(__rxs->rxs_dmamap->dm_segs[0].ds_addr +		\
444		(sc)->sc_align_tweak);					\
445	__rxd->dr_baddrh = 0;						\
446	__rxd->dr_len = 0;						\
447	__rxd->dr_cksum = 0;						\
448	__rxd->dr_status = 0;						\
449	__rxd->dr_errors = 0;						\
450	__rxd->dr_special = 0;						\
451	DGE_CDRXSYNC((sc), (x), BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); \
452									\
453	CSR_WRITE((sc), DGE_RDT, (x));					\
454} while (/*CONSTCOND*/0)
455#endif
456
457#ifdef DGE_OFFBYONE_RXBUG
458/*
459 * Allocation constants.  Much memory may be used for this.
460 */
461#ifndef DGE_BUFFER_SIZE
462#define DGE_BUFFER_SIZE DGE_MAX_MTU
463#endif
464#define DGE_NBUFFERS	(4*DGE_NRXDESC)
465#define DGE_RXMEM	(DGE_NBUFFERS*DGE_BUFFER_SIZE)
466
467struct rxbugentry {
468	SLIST_ENTRY(rxbugentry) rb_entry;
469	int rb_slot;
470};
471
472static int
473dge_alloc_rcvmem(struct dge_softc *sc)
474{
475	char *kva;
476	bus_dma_segment_t seg;
477	int i, rseg, state, error;
478	struct rxbugentry *entry;
479
480	state = error = 0;
481
482	if (bus_dmamem_alloc(sc->sc_dmat, DGE_RXMEM, PAGE_SIZE, 0,
483	     &seg, 1, &rseg, BUS_DMA_NOWAIT)) {
484		aprint_error_dev(sc->sc_dev, "can't alloc rx buffers\n");
485		return ENOBUFS;
486	}
487
488	state = 1;
489	if (bus_dmamem_map(sc->sc_dmat, &seg, rseg, DGE_RXMEM, (void **)&kva,
490	    BUS_DMA_NOWAIT)) {
491		aprint_error_dev(sc->sc_dev,
492		    "can't map DMA buffers (%d bytes)\n", (int)DGE_RXMEM);
493		error = ENOBUFS;
494		goto out;
495	}
496
497	state = 2;
498	if (bus_dmamap_create(sc->sc_dmat, DGE_RXMEM, 1, DGE_RXMEM, 0,
499	    BUS_DMA_NOWAIT, &sc->sc_bugmap)) {
500		aprint_error_dev(sc->sc_dev, "can't create DMA map\n");
501		error = ENOBUFS;
502		goto out;
503	}
504
505	state = 3;
506	if (bus_dmamap_load(sc->sc_dmat, sc->sc_bugmap,
507	    kva, DGE_RXMEM, NULL, BUS_DMA_NOWAIT)) {
508		aprint_error_dev(sc->sc_dev, "can't load DMA map\n");
509		error = ENOBUFS;
510		goto out;
511	}
512
513	state = 4;
514	sc->sc_bugbuf = (void *)kva;
515	SLIST_INIT(&sc->sc_buglist);
516
517	/*
518	 * Now divide it up into DGE_BUFFER_SIZE pieces and save the addresses
519	 * in an array.
520	 */
521	entry = malloc(sizeof(*entry) * DGE_NBUFFERS, M_DEVBUF, M_WAITOK);
522	sc->sc_entry = entry;
523	for (i = 0; i < DGE_NBUFFERS; i++) {
524		entry[i].rb_slot = i;
525		SLIST_INSERT_HEAD(&sc->sc_buglist, &entry[i], rb_entry);
526	}
527out:
528	if (error != 0) {
529		switch (state) {
530		case 4:
531			bus_dmamap_unload(sc->sc_dmat, sc->sc_bugmap);
532			/* FALLTHROUGH */
533		case 3:
534			bus_dmamap_destroy(sc->sc_dmat, sc->sc_bugmap);
535			/* FALLTHROUGH */
536		case 2:
537			bus_dmamem_unmap(sc->sc_dmat, kva, DGE_RXMEM);
538			/* FALLTHROUGH */
539		case 1:
540			bus_dmamem_free(sc->sc_dmat, &seg, rseg);
541			break;
542		default:
543			break;
544		}
545	}
546
547	return error;
548}
549
550/*
551 * Allocate a jumbo buffer.
552 */
553static void *
554dge_getbuf(struct dge_softc *sc)
555{
556	struct rxbugentry *entry;
557
558	entry = SLIST_FIRST(&sc->sc_buglist);
559
560	if (entry == NULL) {
561		printf("%s: no free RX buffers\n", device_xname(sc->sc_dev));
562		return NULL;
563	}
564
565	SLIST_REMOVE_HEAD(&sc->sc_buglist, rb_entry);
566	return (char *)sc->sc_bugbuf + entry->rb_slot * DGE_BUFFER_SIZE;
567}
568
569/*
570 * Release a jumbo buffer.
571 */
572static void
573dge_freebuf(struct mbuf *m, void *buf, size_t size, void *arg)
574{
575	struct rxbugentry *entry;
576	struct dge_softc *sc;
577	int i, s;
578
579	/* Extract the softc struct pointer. */
580	sc = (struct dge_softc *)arg;
581
582	if (sc == NULL)
583		panic("dge_freebuf: can't find softc pointer!");
584
585	/* calculate the slot this buffer belongs to */
586
587	i = ((char *)buf - (char *)sc->sc_bugbuf) / DGE_BUFFER_SIZE;
588
589	if ((i < 0) || (i >= DGE_NBUFFERS))
590		panic("dge_freebuf: asked to free buffer %d!", i);
591
592	s = splvm();
593	entry = sc->sc_entry + i;
594	SLIST_INSERT_HEAD(&sc->sc_buglist, entry, rb_entry);
595
596	if (__predict_true(m != NULL))
597		pool_cache_put(mb_cache, m);
598	splx(s);
599}
600#endif
601
602static void	dge_start(struct ifnet *);
603static void	dge_watchdog(struct ifnet *);
604static int	dge_ioctl(struct ifnet *, u_long, void *);
605static int	dge_init(struct ifnet *);
606static void	dge_stop(struct ifnet *, int);
607
608static bool	dge_shutdown(device_t, int);
609
610static void	dge_reset(struct dge_softc *);
611static void	dge_rxdrain(struct dge_softc *);
612static int	dge_add_rxbuf(struct dge_softc *, int);
613
614static void	dge_set_filter(struct dge_softc *);
615
616static int	dge_intr(void *);
617static void	dge_txintr(struct dge_softc *);
618static void	dge_rxintr(struct dge_softc *);
619static void	dge_linkintr(struct dge_softc *, uint32_t);
620
621static int	dge_match(device_t, cfdata_t, void *);
622static void	dge_attach(device_t, device_t, void *);
623
624static int	dge_read_eeprom(struct dge_softc *sc);
625static int	dge_eeprom_clockin(struct dge_softc *sc);
626static void	dge_eeprom_clockout(struct dge_softc *sc, int bit);
627static uint16_t	dge_eeprom_word(struct dge_softc *sc, int addr);
628static int	dge_xgmii_mediachange(struct ifnet *);
629static void	dge_xgmii_mediastatus(struct ifnet *, struct ifmediareq *);
630static void	dge_xgmii_reset(struct dge_softc *);
631static void	dge_xgmii_writereg(struct dge_softc *, int, int, int);
632
633
634CFATTACH_DECL_NEW(dge, sizeof(struct dge_softc),
635    dge_match, dge_attach, NULL, NULL);
636
637#ifdef DGE_EVENT_COUNTERS
638#if DGE_NTXSEGS > 100
639#error Update dge_txseg_evcnt_names
640#endif
641static char (*dge_txseg_evcnt_names)[DGE_NTXSEGS][8 /* "txseg00" + \0 */];
642#endif /* DGE_EVENT_COUNTERS */
643
644/*
645 * Devices supported by this driver.
646 */
647static const struct dge_product {
648	pci_vendor_id_t dgep_vendor;
649	pci_product_id_t dgep_product;
650	const char *dgep_name;
651	int dgep_flags;
652#define DGEP_F_10G_LR	  0x01
653#define DGEP_F_10G_SR	  0x02
654} dge_products[] = {
655	{ PCI_VENDOR_INTEL,  PCI_PRODUCT_INTEL_82597EX,
656	  "Intel i82597EX 10GbE-LR Ethernet",
657	  DGEP_F_10G_LR },
658
659	{ PCI_VENDOR_INTEL,  PCI_PRODUCT_INTEL_82597EX_SR,
660	  "Intel i82597EX 10GbE-SR Ethernet",
661	  DGEP_F_10G_SR },
662
663	{ 0,	    0,
664	  NULL,
665	  0 },
666};
667
668static const struct dge_product *
669dge_lookup(const struct pci_attach_args *pa)
670{
671	const struct dge_product *dgep;
672
673	for (dgep = dge_products; dgep->dgep_name != NULL; dgep++) {
674		if (PCI_VENDOR(pa->pa_id) == dgep->dgep_vendor &&
675		    PCI_PRODUCT(pa->pa_id) == dgep->dgep_product)
676			return dgep;
677		}
678	return NULL;
679}
680
681static int
682dge_match(device_t parent, cfdata_t cf, void *aux)
683{
684	struct pci_attach_args *pa = aux;
685
686	if (dge_lookup(pa) != NULL)
687		return 1;
688
689	return 0;
690}
691
692static void
693dge_attach(device_t parent, device_t self, void *aux)
694{
695	struct dge_softc *sc = device_private(self);
696	struct pci_attach_args *pa = aux;
697	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
698	pci_chipset_tag_t pc = pa->pa_pc;
699	pci_intr_handle_t ih;
700	const char *intrstr = NULL;
701	bus_dma_segment_t seg;
702	int i, rseg, error;
703	uint8_t enaddr[ETHER_ADDR_LEN];
704	pcireg_t preg, memtype;
705	uint32_t reg;
706	char intrbuf[PCI_INTRSTR_LEN];
707	const struct dge_product *dgep;
708
709	sc->sc_dgep = dgep = dge_lookup(pa);
710	if (dgep == NULL) {
711		printf("\n");
712		panic("dge_attach: impossible");
713	}
714
715	sc->sc_dev = self;
716	sc->sc_dmat = pa->pa_dmat;
717	sc->sc_pc = pa->pa_pc;
718	sc->sc_pt = pa->pa_tag;
719
720	pci_aprint_devinfo_fancy(pa, "Ethernet controller",
721		dgep->dgep_name, 1);
722
723	memtype = pci_mapreg_type(pa->pa_pc, pa->pa_tag, DGE_PCI_BAR);
724	if (pci_mapreg_map(pa, DGE_PCI_BAR, memtype, 0,
725	    &sc->sc_st, &sc->sc_sh, NULL, NULL)) {
726		aprint_error_dev(sc->sc_dev,
727		    "unable to map device registers\n");
728		return;
729	}
730
731	/* Enable bus mastering */
732	preg = pci_conf_read(pc, pa->pa_tag, PCI_COMMAND_STATUS_REG);
733	preg |= PCI_COMMAND_MASTER_ENABLE;
734	pci_conf_write(pc, pa->pa_tag, PCI_COMMAND_STATUS_REG, preg);
735
736	/*
737	 * Map and establish our interrupt.
738	 */
739	if (pci_intr_map(pa, &ih)) {
740		aprint_error_dev(sc->sc_dev, "unable to map interrupt\n");
741		return;
742	}
743	intrstr = pci_intr_string(pc, ih, intrbuf, sizeof(intrbuf));
744	sc->sc_ih = pci_intr_establish_xname(pc, ih, IPL_NET, dge_intr, sc,
745	    device_xname(self));
746	if (sc->sc_ih == NULL) {
747		aprint_error_dev(sc->sc_dev, "unable to establish interrupt");
748		if (intrstr != NULL)
749			aprint_error(" at %s", intrstr);
750		aprint_error("\n");
751		return;
752	}
753	aprint_normal_dev(sc->sc_dev, "interrupting at %s\n", intrstr);
754
755	/*
756	 * Determine a few things about the bus we're connected to.
757	 */
758	reg = CSR_READ(sc, DGE_STATUS);
759	if (reg & STATUS_BUS64)
760		sc->sc_flags |= DGE_F_BUS64;
761
762	sc->sc_flags |= DGE_F_PCIX;
763	if (pci_get_capability(pa->pa_pc, pa->pa_tag,
764			       PCI_CAP_PCIX,
765			       &sc->sc_pcix_offset, NULL) == 0)
766		aprint_error_dev(sc->sc_dev, "unable to find PCIX "
767		    "capability\n");
768
769	if (sc->sc_flags & DGE_F_PCIX) {
770		switch (reg & STATUS_PCIX_MSK) {
771		case STATUS_PCIX_66:
772			sc->sc_bus_speed = 66;
773			break;
774		case STATUS_PCIX_100:
775			sc->sc_bus_speed = 100;
776			break;
777		case STATUS_PCIX_133:
778			sc->sc_bus_speed = 133;
779			break;
780		default:
781			aprint_error_dev(sc->sc_dev,
782			    "unknown PCIXSPD %d; assuming 66MHz\n",
783			    reg & STATUS_PCIX_MSK);
784			sc->sc_bus_speed = 66;
785		}
786	} else
787		sc->sc_bus_speed = (reg & STATUS_BUS64) ? 66 : 33;
788	aprint_verbose_dev(sc->sc_dev, "%d-bit %dMHz %s bus\n",
789	    (sc->sc_flags & DGE_F_BUS64) ? 64 : 32, sc->sc_bus_speed,
790	    (sc->sc_flags & DGE_F_PCIX) ? "PCIX" : "PCI");
791
792	/*
793	 * Allocate the control data structures, and create and load the
794	 * DMA map for it.
795	 */
796	if ((error = bus_dmamem_alloc(sc->sc_dmat,
797	    sizeof(struct dge_control_data), PAGE_SIZE, 0, &seg, 1, &rseg,
798	    0)) != 0) {
799		aprint_error_dev(sc->sc_dev,
800		    "unable to allocate control data, error = %d\n",
801		    error);
802		goto fail_0;
803	}
804
805	if ((error = bus_dmamem_map(sc->sc_dmat, &seg, rseg,
806	    sizeof(struct dge_control_data), (void **)&sc->sc_control_data,
807	    0)) != 0) {
808		aprint_error_dev(sc->sc_dev,
809		    "unable to map control data, error = %d\n", error);
810		goto fail_1;
811	}
812
813	if ((error = bus_dmamap_create(sc->sc_dmat,
814	    sizeof(struct dge_control_data), 1,
815	    sizeof(struct dge_control_data), 0, 0, &sc->sc_cddmamap)) != 0) {
816		aprint_error_dev(sc->sc_dev, "unable to create control data "
817		    "DMA map, error = %d\n", error);
818		goto fail_2;
819	}
820
821	if ((error = bus_dmamap_load(sc->sc_dmat, sc->sc_cddmamap,
822	    sc->sc_control_data, sizeof(struct dge_control_data), NULL,
823	    0)) != 0) {
824		aprint_error_dev(sc->sc_dev,
825		    "unable to load control data DMA map, error = %d\n",
826		    error);
827		goto fail_3;
828	}
829
830#ifdef DGE_OFFBYONE_RXBUG
831	if (dge_alloc_rcvmem(sc) != 0)
832		return; /* Already complained */
833#endif
834	/*
835	 * Create the transmit buffer DMA maps.
836	 */
837	for (i = 0; i < DGE_TXQUEUELEN; i++) {
838		if ((error = bus_dmamap_create(sc->sc_dmat, DGE_MAX_MTU,
839		    DGE_NTXSEGS, MCLBYTES, 0, 0,
840		    &sc->sc_txsoft[i].txs_dmamap)) != 0) {
841			aprint_error_dev(sc->sc_dev, "unable to create Tx DMA map %d, "
842			    "error = %d\n", i, error);
843			goto fail_4;
844		}
845	}
846
847	/*
848	 * Create the receive buffer DMA maps.
849	 */
850	for (i = 0; i < DGE_NRXDESC; i++) {
851#ifdef DGE_OFFBYONE_RXBUG
852		if ((error = bus_dmamap_create(sc->sc_dmat, DGE_BUFFER_SIZE, 1,
853		    DGE_BUFFER_SIZE, 0, 0, &sc->sc_rxsoft[i].rxs_dmamap)) != 0) {
854#else
855		if ((error = bus_dmamap_create(sc->sc_dmat, MCLBYTES, 1,
856		    MCLBYTES, 0, 0, &sc->sc_rxsoft[i].rxs_dmamap)) != 0) {
857#endif
858			aprint_error_dev(sc->sc_dev, "unable to create Rx DMA "
859			    "map %d, error = %d\n", i, error);
860			goto fail_5;
861		}
862		sc->sc_rxsoft[i].rxs_mbuf = NULL;
863	}
864
865	/*
866	 * Set bits in ctrl0 register.
867	 * Should get the software defined pins out of EEPROM?
868	 */
869	sc->sc_ctrl0 |= CTRL0_RPE | CTRL0_TPE; /* XON/XOFF */
870	sc->sc_ctrl0 |= CTRL0_SDP3_DIR | CTRL0_SDP2_DIR | CTRL0_SDP1_DIR |
871	    CTRL0_SDP0_DIR | CTRL0_SDP3 | CTRL0_SDP2 | CTRL0_SDP0;
872
873	/*
874	 * Reset the chip to a known state.
875	 */
876	dge_reset(sc);
877
878	/*
879	 * Reset the PHY.
880	 */
881	dge_xgmii_reset(sc);
882
883	/*
884	 * Read in EEPROM data.
885	 */
886	if (dge_read_eeprom(sc)) {
887		aprint_error_dev(sc->sc_dev, "couldn't read EEPROM\n");
888		return;
889	}
890
891	/*
892	 * Get the ethernet address.
893	 */
894	enaddr[0] = sc->sc_eeprom[EE_ADDR01] & 0377;
895	enaddr[1] = sc->sc_eeprom[EE_ADDR01] >> 8;
896	enaddr[2] = sc->sc_eeprom[EE_ADDR23] & 0377;
897	enaddr[3] = sc->sc_eeprom[EE_ADDR23] >> 8;
898	enaddr[4] = sc->sc_eeprom[EE_ADDR45] & 0377;
899	enaddr[5] = sc->sc_eeprom[EE_ADDR45] >> 8;
900
901	aprint_normal_dev(sc->sc_dev, "Ethernet address %s\n",
902	    ether_sprintf(enaddr));
903
904	/*
905	 * Setup media stuff.
906	 */
907	sc->sc_ethercom.ec_ifmedia = &sc->sc_media;
908	ifmedia_init(&sc->sc_media, IFM_IMASK, dge_xgmii_mediachange,
909	    dge_xgmii_mediastatus);
910	if (dgep->dgep_flags & DGEP_F_10G_SR) {
911		ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_10G_SR, 0, NULL);
912		ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_10G_SR);
913	} else { /* XXX default is LR */
914		ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_10G_LR, 0, NULL);
915		ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_10G_LR);
916	}
917
918	ifp = &sc->sc_ethercom.ec_if;
919	strlcpy(ifp->if_xname, device_xname(sc->sc_dev), IFNAMSIZ);
920	ifp->if_softc = sc;
921	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
922	ifp->if_ioctl = dge_ioctl;
923	ifp->if_start = dge_start;
924	ifp->if_watchdog = dge_watchdog;
925	ifp->if_init = dge_init;
926	ifp->if_stop = dge_stop;
927	IFQ_SET_MAXLEN(&ifp->if_snd, uimax(DGE_IFQUEUELEN, IFQ_MAXLEN));
928	IFQ_SET_READY(&ifp->if_snd);
929
930	sc->sc_ethercom.ec_capabilities |=
931	    ETHERCAP_JUMBO_MTU | ETHERCAP_VLAN_MTU;
932
933	/*
934	 * We can perform TCPv4 and UDPv4 checkums in-bound.
935	 */
936	ifp->if_capabilities |=
937	    IFCAP_CSUM_IPv4_Tx | IFCAP_CSUM_IPv4_Rx |
938	    IFCAP_CSUM_TCPv4_Tx | IFCAP_CSUM_TCPv4_Rx |
939	    IFCAP_CSUM_UDPv4_Tx | IFCAP_CSUM_UDPv4_Rx;
940
941	/*
942	 * Attach the interface.
943	 */
944	if_attach(ifp);
945	if_deferred_start_init(ifp, NULL);
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 */
1044static bool
1045dge_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 */
1061static int
1062dge_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 */
1180static void
1181dge_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, BPF_D_OUT);
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 */
1413static void
1414dge_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		if_statinc(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 */
1443static int
1444dge_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 SIOCSIFMTU:
1455		if (ifr->ifr_mtu < ETHERMIN || ifr->ifr_mtu > DGE_MAX_MTU)
1456			error = EINVAL;
1457		else if ((error = ifioctl_common(ifp, cmd, data)) != ENETRESET)
1458			break;
1459		else if (ifp->if_flags & IFF_UP)
1460			error = (*ifp->if_init)(ifp);
1461		else
1462			error = 0;
1463		break;
1464
1465	case SIOCSIFFLAGS:
1466		if ((error = ifioctl_common(ifp, cmd, data)) != 0)
1467			break;
1468		/* extract link flags */
1469		if ((ifp->if_flags & IFF_LINK0) == 0 &&
1470		    (ifp->if_flags & IFF_LINK1) == 0)
1471			mmrbc = PCIX_MMRBC_512;
1472		else if ((ifp->if_flags & IFF_LINK0) == 0 &&
1473		    (ifp->if_flags & IFF_LINK1) != 0)
1474			mmrbc = PCIX_MMRBC_1024;
1475		else if ((ifp->if_flags & IFF_LINK0) != 0 &&
1476		    (ifp->if_flags & IFF_LINK1) == 0)
1477			mmrbc = PCIX_MMRBC_2048;
1478		else
1479			mmrbc = PCIX_MMRBC_4096;
1480		if (mmrbc != sc->sc_mmrbc) {
1481			preg = pci_conf_read(sc->sc_pc, sc->sc_pt,DGE_PCIX_CMD);
1482			preg &= ~PCIX_MMRBC_MSK;
1483			preg |= mmrbc;
1484			pci_conf_write(sc->sc_pc, sc->sc_pt,DGE_PCIX_CMD, preg);
1485			sc->sc_mmrbc = mmrbc;
1486		}
1487		/* FALLTHROUGH */
1488	default:
1489		if ((error = ether_ioctl(ifp, cmd, data)) != ENETRESET)
1490			break;
1491
1492		error = 0;
1493
1494		if (cmd == SIOCSIFCAP)
1495			error = (*ifp->if_init)(ifp);
1496		else if (cmd != SIOCADDMULTI && cmd != SIOCDELMULTI)
1497			;
1498		else if (ifp->if_flags & IFF_RUNNING) {
1499			/*
1500			 * Multicast list has changed; set the hardware filter
1501			 * accordingly.
1502			 */
1503			dge_set_filter(sc);
1504		}
1505		break;
1506	}
1507
1508	/* Try to get more packets going. */
1509	dge_start(ifp);
1510
1511	splx(s);
1512	return error;
1513}
1514
1515/*
1516 * dge_intr:
1517 *
1518 *	Interrupt service routine.
1519 */
1520static int
1521dge_intr(void *arg)
1522{
1523	struct dge_softc *sc = arg;
1524	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
1525	uint32_t icr;
1526	int wantinit, handled = 0;
1527
1528	for (wantinit = 0; wantinit == 0;) {
1529		icr = CSR_READ(sc, DGE_ICR);
1530		if ((icr & sc->sc_icr) == 0)
1531			break;
1532
1533		rnd_add_uint32(&sc->rnd_source, icr);
1534
1535		handled = 1;
1536
1537#if defined(DGE_DEBUG) || defined(DGE_EVENT_COUNTERS)
1538		if (icr & (ICR_RXDMT0 | ICR_RXT0)) {
1539			DPRINTF(DGE_DEBUG_RX,
1540			    ("%s: RX: got Rx intr 0x%08x\n",
1541			    device_xname(sc->sc_dev),
1542			    icr & (ICR_RXDMT0 | ICR_RXT0)));
1543			DGE_EVCNT_INCR(&sc->sc_ev_rxintr);
1544		}
1545#endif
1546		dge_rxintr(sc);
1547
1548#if defined(DGE_DEBUG) || defined(DGE_EVENT_COUNTERS)
1549		if (icr & ICR_TXDW) {
1550			DPRINTF(DGE_DEBUG_TX,
1551			    ("%s: TX: got TXDW interrupt\n",
1552			    device_xname(sc->sc_dev)));
1553			DGE_EVCNT_INCR(&sc->sc_ev_txdw);
1554		}
1555		if (icr & ICR_TXQE)
1556			DGE_EVCNT_INCR(&sc->sc_ev_txqe);
1557#endif
1558		dge_txintr(sc);
1559
1560		if (icr & (ICR_LSC | ICR_RXSEQ)) {
1561			DGE_EVCNT_INCR(&sc->sc_ev_linkintr);
1562			dge_linkintr(sc, icr);
1563		}
1564
1565		if (icr & ICR_RXO) {
1566			printf("%s: Receive overrun\n",
1567			    device_xname(sc->sc_dev));
1568			wantinit = 1;
1569		}
1570	}
1571
1572	if (handled) {
1573		if (wantinit)
1574			dge_init(ifp);
1575
1576		/* Try to get more packets going. */
1577		if_schedule_deferred_start(ifp);
1578	}
1579
1580	return handled;
1581}
1582
1583/*
1584 * dge_txintr:
1585 *
1586 *	Helper; handle transmit interrupts.
1587 */
1588static void
1589dge_txintr(struct dge_softc *sc)
1590{
1591	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
1592	struct dge_txsoft *txs;
1593	uint8_t status;
1594	int i;
1595
1596	ifp->if_flags &= ~IFF_OACTIVE;
1597
1598	/*
1599	 * Go through the Tx list and free mbufs for those
1600	 * frames which have been transmitted.
1601	 */
1602	for (i = sc->sc_txsdirty; sc->sc_txsfree != DGE_TXQUEUELEN;
1603	     i = DGE_NEXTTXS(i), sc->sc_txsfree++) {
1604		txs = &sc->sc_txsoft[i];
1605
1606		DPRINTF(DGE_DEBUG_TX,
1607		    ("%s: TX: checking job %d\n", device_xname(sc->sc_dev), i));
1608
1609		DGE_CDTXSYNC(sc, txs->txs_firstdesc, txs->txs_dmamap->dm_nsegs,
1610		    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
1611
1612		status =
1613		    sc->sc_txdescs[txs->txs_lastdesc].dt_status;
1614		if ((status & TDESC_STA_DD) == 0) {
1615			DGE_CDTXSYNC(sc, txs->txs_lastdesc, 1,
1616			    BUS_DMASYNC_PREREAD);
1617			break;
1618		}
1619
1620		DPRINTF(DGE_DEBUG_TX,
1621		    ("%s: TX: job %d done: descs %d..%d\n",
1622		    device_xname(sc->sc_dev), i, txs->txs_firstdesc,
1623		    txs->txs_lastdesc));
1624
1625		if_statinc(ifp, if_opackets);
1626		sc->sc_txfree += txs->txs_ndesc;
1627		bus_dmamap_sync(sc->sc_dmat, txs->txs_dmamap,
1628		    0, txs->txs_dmamap->dm_mapsize, BUS_DMASYNC_POSTWRITE);
1629		bus_dmamap_unload(sc->sc_dmat, txs->txs_dmamap);
1630		m_freem(txs->txs_mbuf);
1631		txs->txs_mbuf = NULL;
1632	}
1633
1634	/* Update the dirty transmit buffer pointer. */
1635	sc->sc_txsdirty = i;
1636	DPRINTF(DGE_DEBUG_TX,
1637	    ("%s: TX: txsdirty -> %d\n", device_xname(sc->sc_dev), i));
1638
1639	/*
1640	 * If there are no more pending transmissions, cancel the watchdog
1641	 * timer.
1642	 */
1643	if (sc->sc_txsfree == DGE_TXQUEUELEN)
1644		ifp->if_timer = 0;
1645}
1646
1647/*
1648 * dge_rxintr:
1649 *
1650 *	Helper; handle receive interrupts.
1651 */
1652static void
1653dge_rxintr(struct dge_softc *sc)
1654{
1655	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
1656	struct dge_rxsoft *rxs;
1657	struct mbuf *m;
1658	int i, len;
1659	uint8_t status, errors;
1660
1661	for (i = sc->sc_rxptr;; i = DGE_NEXTRX(i)) {
1662		rxs = &sc->sc_rxsoft[i];
1663
1664		DPRINTF(DGE_DEBUG_RX,
1665		    ("%s: RX: checking descriptor %d\n",
1666		    device_xname(sc->sc_dev), i));
1667
1668		DGE_CDRXSYNC(sc, i,
1669		    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
1670
1671		status = sc->sc_rxdescs[i].dr_status;
1672		errors = sc->sc_rxdescs[i].dr_errors;
1673		len = le16toh(sc->sc_rxdescs[i].dr_len);
1674
1675		if ((status & RDESC_STS_DD) == 0) {
1676			/* We have processed all of the receive descriptors. */
1677			DGE_CDRXSYNC(sc, i, BUS_DMASYNC_PREREAD);
1678			break;
1679		}
1680
1681		if (__predict_false(sc->sc_rxdiscard)) {
1682			DPRINTF(DGE_DEBUG_RX,
1683			    ("%s: RX: discarding contents of descriptor %d\n",
1684			    device_xname(sc->sc_dev), i));
1685			DGE_INIT_RXDESC(sc, i);
1686			if (status & RDESC_STS_EOP) {
1687				/* Reset our state. */
1688				DPRINTF(DGE_DEBUG_RX,
1689				    ("%s: RX: resetting rxdiscard -> 0\n",
1690				    device_xname(sc->sc_dev)));
1691				sc->sc_rxdiscard = 0;
1692			}
1693			continue;
1694		}
1695
1696		bus_dmamap_sync(sc->sc_dmat, rxs->rxs_dmamap, 0,
1697		    rxs->rxs_dmamap->dm_mapsize, BUS_DMASYNC_POSTREAD);
1698
1699		m = rxs->rxs_mbuf;
1700
1701		/*
1702		 * Add a new receive buffer to the ring.
1703		 */
1704		if (dge_add_rxbuf(sc, i) != 0) {
1705			/*
1706			 * Failed, throw away what we've done so
1707			 * far, and discard the rest of the packet.
1708			 */
1709			if_statinc(ifp, if_ierrors);
1710			bus_dmamap_sync(sc->sc_dmat, rxs->rxs_dmamap, 0,
1711			    rxs->rxs_dmamap->dm_mapsize, BUS_DMASYNC_PREREAD);
1712			DGE_INIT_RXDESC(sc, i);
1713			if ((status & RDESC_STS_EOP) == 0)
1714				sc->sc_rxdiscard = 1;
1715			if (sc->sc_rxhead != NULL)
1716				m_freem(sc->sc_rxhead);
1717			DGE_RXCHAIN_RESET(sc);
1718			DPRINTF(DGE_DEBUG_RX,
1719			    ("%s: RX: Rx buffer allocation failed, "
1720			    "dropping packet%s\n", device_xname(sc->sc_dev),
1721			    sc->sc_rxdiscard ? " (discard)" : ""));
1722			continue;
1723		}
1724		DGE_INIT_RXDESC(sc, DGE_PREVRX(i)); /* Write the descriptor */
1725
1726		DGE_RXCHAIN_LINK(sc, m);
1727
1728		m->m_len = len;
1729
1730		DPRINTF(DGE_DEBUG_RX,
1731		    ("%s: RX: buffer at %p len %d\n",
1732		    device_xname(sc->sc_dev), m->m_data, len));
1733
1734		/*
1735		 * If this is not the end of the packet, keep
1736		 * looking.
1737		 */
1738		if ((status & RDESC_STS_EOP) == 0) {
1739			sc->sc_rxlen += len;
1740			DPRINTF(DGE_DEBUG_RX,
1741			    ("%s: RX: not yet EOP, rxlen -> %d\n",
1742			    device_xname(sc->sc_dev), sc->sc_rxlen));
1743			continue;
1744		}
1745
1746		/*
1747		 * Okay, we have the entire packet now...
1748		 */
1749		*sc->sc_rxtailp = NULL;
1750		m = sc->sc_rxhead;
1751		len += sc->sc_rxlen;
1752
1753		DGE_RXCHAIN_RESET(sc);
1754
1755		DPRINTF(DGE_DEBUG_RX,
1756		    ("%s: RX: have entire packet, len -> %d\n",
1757		    device_xname(sc->sc_dev), len));
1758
1759		/*
1760		 * If an error occurred, update stats and drop the packet.
1761		 */
1762		if (errors & (RDESC_ERR_CE | RDESC_ERR_SE | RDESC_ERR_P |
1763		    RDESC_ERR_RXE)) {
1764			if_statinc(ifp, if_ierrors);
1765			if (errors & RDESC_ERR_SE)
1766				printf("%s: symbol error\n",
1767				    device_xname(sc->sc_dev));
1768			else if (errors & RDESC_ERR_P)
1769				printf("%s: parity error\n",
1770				    device_xname(sc->sc_dev));
1771			else if (errors & RDESC_ERR_CE)
1772				printf("%s: CRC error\n",
1773				    device_xname(sc->sc_dev));
1774			m_freem(m);
1775			continue;
1776		}
1777
1778		/*
1779		 * No errors.  Receive the packet.
1780		 */
1781		m_set_rcvif(m, ifp);
1782		m->m_pkthdr.len = len;
1783
1784		/*
1785		 * Set up checksum info for this packet.
1786		 */
1787		if (status & RDESC_STS_IPCS) {
1788			DGE_EVCNT_INCR(&sc->sc_ev_rxipsum);
1789			m->m_pkthdr.csum_flags |= M_CSUM_IPv4;
1790			if (errors & RDESC_ERR_IPE)
1791				m->m_pkthdr.csum_flags |= M_CSUM_IPv4_BAD;
1792		}
1793		if (status & RDESC_STS_TCPCS) {
1794			/*
1795			 * Note: we don't know if this was TCP or UDP,
1796			 * so we just set both bits, and expect the
1797			 * upper layers to deal.
1798			 */
1799			DGE_EVCNT_INCR(&sc->sc_ev_rxtusum);
1800			m->m_pkthdr.csum_flags |= M_CSUM_TCPv4 | M_CSUM_UDPv4;
1801			if (errors & RDESC_ERR_TCPE)
1802				m->m_pkthdr.csum_flags |= M_CSUM_TCP_UDP_BAD;
1803		}
1804
1805		/* Pass it on. */
1806		if_percpuq_enqueue(ifp->if_percpuq, m);
1807	}
1808
1809	/* Update the receive pointer. */
1810	sc->sc_rxptr = i;
1811
1812	DPRINTF(DGE_DEBUG_RX,
1813	    ("%s: RX: rxptr -> %d\n", device_xname(sc->sc_dev), i));
1814}
1815
1816/*
1817 * dge_linkintr:
1818 *
1819 *	Helper; handle link interrupts.
1820 */
1821static void
1822dge_linkintr(struct dge_softc *sc, uint32_t icr)
1823{
1824	uint32_t status;
1825
1826	if (icr & ICR_LSC) {
1827		status = CSR_READ(sc, DGE_STATUS);
1828		if (status & STATUS_LINKUP) {
1829			DPRINTF(DGE_DEBUG_LINK, ("%s: LINK: LSC -> up\n",
1830			    device_xname(sc->sc_dev)));
1831		} else {
1832			DPRINTF(DGE_DEBUG_LINK, ("%s: LINK: LSC -> down\n",
1833			    device_xname(sc->sc_dev)));
1834		}
1835	} else if (icr & ICR_RXSEQ) {
1836		DPRINTF(DGE_DEBUG_LINK,
1837		    ("%s: LINK: Receive sequence error\n",
1838		    device_xname(sc->sc_dev)));
1839	}
1840	/* XXX - fix errata */
1841}
1842
1843/*
1844 * dge_reset:
1845 *
1846 *	Reset the i82597 chip.
1847 */
1848static void
1849dge_reset(struct dge_softc *sc)
1850{
1851	int i;
1852
1853	/*
1854	 * Do a chip reset.
1855	 */
1856	CSR_WRITE(sc, DGE_CTRL0, CTRL0_RST | sc->sc_ctrl0);
1857
1858	delay(10000);
1859
1860	for (i = 0; i < 1000; i++) {
1861		if ((CSR_READ(sc, DGE_CTRL0) & CTRL0_RST) == 0)
1862			break;
1863		delay(20);
1864	}
1865
1866	if (CSR_READ(sc, DGE_CTRL0) & CTRL0_RST)
1867		printf("%s: WARNING: reset failed to complete\n",
1868		    device_xname(sc->sc_dev));
1869	/*
1870	 * Reset the EEPROM logic.
1871	 * This will cause the chip to reread its default values,
1872	 * which doesn't happen otherwise (errata).
1873	 */
1874	CSR_WRITE(sc, DGE_CTRL1, CTRL1_EE_RST);
1875	delay(10000);
1876}
1877
1878/*
1879 * dge_init:		[ifnet interface function]
1880 *
1881 *	Initialize the interface.  Must be called at splnet().
1882 */
1883static int
1884dge_init(struct ifnet *ifp)
1885{
1886	struct dge_softc *sc = ifp->if_softc;
1887	struct dge_rxsoft *rxs;
1888	int i, error = 0;
1889	uint32_t reg;
1890
1891	/*
1892	 * *_HDR_ALIGNED_P is constant 1 if __NO_STRICT_ALIGMENT is set.
1893	 * There is a small but measurable benefit to avoiding the adjusment
1894	 * of the descriptor so that the headers are aligned, for normal mtu,
1895	 * on such platforms.  One possibility is that the DMA itself is
1896	 * slightly more efficient if the front of the entire packet (instead
1897	 * of the front of the headers) is aligned.
1898	 *
1899	 * Note we must always set align_tweak to 0 if we are using
1900	 * jumbo frames.
1901	 */
1902#ifdef __NO_STRICT_ALIGNMENT
1903	sc->sc_align_tweak = 0;
1904#else
1905	if ((ifp->if_mtu + ETHER_HDR_LEN + ETHER_CRC_LEN) > (MCLBYTES - 2))
1906		sc->sc_align_tweak = 0;
1907	else
1908		sc->sc_align_tweak = 2;
1909#endif /* __NO_STRICT_ALIGNMENT */
1910
1911	/* Cancel any pending I/O. */
1912	dge_stop(ifp, 0);
1913
1914	/* Reset the chip to a known state. */
1915	dge_reset(sc);
1916
1917	/* Initialize the transmit descriptor ring. */
1918	memset(sc->sc_txdescs, 0, sizeof(sc->sc_txdescs));
1919	DGE_CDTXSYNC(sc, 0, DGE_NTXDESC,
1920	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1921	sc->sc_txfree = DGE_NTXDESC;
1922	sc->sc_txnext = 0;
1923
1924	sc->sc_txctx_ipcs = 0xffffffff;
1925	sc->sc_txctx_tucs = 0xffffffff;
1926
1927	CSR_WRITE(sc, DGE_TDBAH, 0);
1928	CSR_WRITE(sc, DGE_TDBAL, DGE_CDTXADDR(sc, 0));
1929	CSR_WRITE(sc, DGE_TDLEN, sizeof(sc->sc_txdescs));
1930	CSR_WRITE(sc, DGE_TDH, 0);
1931	CSR_WRITE(sc, DGE_TDT, 0);
1932	CSR_WRITE(sc, DGE_TIDV, TIDV);
1933
1934#if 0
1935	CSR_WRITE(sc, DGE_TXDCTL, TXDCTL_PTHRESH(0) |
1936	    TXDCTL_HTHRESH(0) | TXDCTL_WTHRESH(0));
1937#endif
1938	CSR_WRITE(sc, DGE_RXDCTL,
1939	    RXDCTL_PTHRESH(RXDCTL_PTHRESH_VAL) |
1940	    RXDCTL_HTHRESH(RXDCTL_HTHRESH_VAL) |
1941	    RXDCTL_WTHRESH(RXDCTL_WTHRESH_VAL));
1942
1943	/* Initialize the transmit job descriptors. */
1944	for (i = 0; i < DGE_TXQUEUELEN; i++)
1945		sc->sc_txsoft[i].txs_mbuf = NULL;
1946	sc->sc_txsfree = DGE_TXQUEUELEN;
1947	sc->sc_txsnext = 0;
1948	sc->sc_txsdirty = 0;
1949
1950	/*
1951	 * Initialize the receive descriptor and receive job
1952	 * descriptor rings.
1953	 */
1954	CSR_WRITE(sc, DGE_RDBAH, 0);
1955	CSR_WRITE(sc, DGE_RDBAL, DGE_CDRXADDR(sc, 0));
1956	CSR_WRITE(sc, DGE_RDLEN, sizeof(sc->sc_rxdescs));
1957	CSR_WRITE(sc, DGE_RDH, DGE_RXSPACE);
1958	CSR_WRITE(sc, DGE_RDT, 0);
1959	CSR_WRITE(sc, DGE_RDTR, RDTR | 0x80000000);
1960	CSR_WRITE(sc, DGE_FCRTL, FCRTL | FCRTL_XONE);
1961	CSR_WRITE(sc, DGE_FCRTH, FCRTH);
1962
1963	for (i = 0; i < DGE_NRXDESC; i++) {
1964		rxs = &sc->sc_rxsoft[i];
1965		if (rxs->rxs_mbuf == NULL) {
1966			if ((error = dge_add_rxbuf(sc, i)) != 0) {
1967				printf("%s: unable to allocate or map rx "
1968				    "buffer %d, error = %d\n",
1969				    device_xname(sc->sc_dev), i, error);
1970				/*
1971				 * XXX Should attempt to run with fewer receive
1972				 * XXX buffers instead of just failing.
1973				 */
1974				dge_rxdrain(sc);
1975				goto out;
1976			}
1977		}
1978		DGE_INIT_RXDESC(sc, i);
1979	}
1980	sc->sc_rxptr = DGE_RXSPACE;
1981	sc->sc_rxdiscard = 0;
1982	DGE_RXCHAIN_RESET(sc);
1983
1984	if (sc->sc_ethercom.ec_capabilities & ETHERCAP_JUMBO_MTU) {
1985		sc->sc_ctrl0 |= CTRL0_JFE;
1986		CSR_WRITE(sc, DGE_MFS, ETHER_MAX_LEN_JUMBO << 16);
1987	}
1988
1989	/* Write the control registers. */
1990	CSR_WRITE(sc, DGE_CTRL0, sc->sc_ctrl0);
1991
1992	/*
1993	 * Set up checksum offload parameters.
1994	 */
1995	reg = CSR_READ(sc, DGE_RXCSUM);
1996	if (ifp->if_capenable & IFCAP_CSUM_IPv4_Rx)
1997		reg |= RXCSUM_IPOFL;
1998	else
1999		reg &= ~RXCSUM_IPOFL;
2000	if (ifp->if_capenable & (IFCAP_CSUM_TCPv4_Rx | IFCAP_CSUM_UDPv4_Rx))
2001		reg |= RXCSUM_IPOFL | RXCSUM_TUOFL;
2002	else {
2003		reg &= ~RXCSUM_TUOFL;
2004		if ((ifp->if_capenable & IFCAP_CSUM_IPv4_Rx) == 0)
2005			reg &= ~RXCSUM_IPOFL;
2006	}
2007	CSR_WRITE(sc, DGE_RXCSUM, reg);
2008
2009	/*
2010	 * Set up the interrupt registers.
2011	 */
2012	CSR_WRITE(sc, DGE_IMC, 0xffffffffU);
2013	sc->sc_icr = ICR_TXDW | ICR_LSC | ICR_RXSEQ | ICR_RXDMT0 |
2014	    ICR_RXO | ICR_RXT0;
2015
2016	CSR_WRITE(sc, DGE_IMS, sc->sc_icr);
2017
2018	/*
2019	 * Set up the transmit control register.
2020	 */
2021	sc->sc_tctl = TCTL_TCE | TCTL_TPDE | TCTL_TXEN;
2022	CSR_WRITE(sc, DGE_TCTL, sc->sc_tctl);
2023
2024	/*
2025	 * Set up the receive control register; we actually program
2026	 * the register when we set the receive filter.  Use multicast
2027	 * address offset type 0.
2028	 */
2029	sc->sc_mchash_type = 0;
2030
2031	sc->sc_rctl = RCTL_RXEN | RCTL_RDMTS_12 | RCTL_RPDA_MC |
2032	    RCTL_CFF | RCTL_SECRC | RCTL_MO(sc->sc_mchash_type);
2033
2034#ifdef DGE_OFFBYONE_RXBUG
2035	sc->sc_rctl |= RCTL_BSIZE_16k;
2036#else
2037	switch (MCLBYTES) {
2038	case 2048:
2039		sc->sc_rctl |= RCTL_BSIZE_2k;
2040		break;
2041	case 4096:
2042		sc->sc_rctl |= RCTL_BSIZE_4k;
2043		break;
2044	case 8192:
2045		sc->sc_rctl |= RCTL_BSIZE_8k;
2046		break;
2047	case 16384:
2048		sc->sc_rctl |= RCTL_BSIZE_16k;
2049		break;
2050	default:
2051		panic("dge_init: MCLBYTES %d unsupported", MCLBYTES);
2052	}
2053#endif
2054
2055	/* Set the receive filter. */
2056	/* Also sets RCTL */
2057	dge_set_filter(sc);
2058
2059	/* ...all done! */
2060	ifp->if_flags |= IFF_RUNNING;
2061	ifp->if_flags &= ~IFF_OACTIVE;
2062
2063 out:
2064	if (error)
2065		printf("%s: interface not running\n", device_xname(sc->sc_dev));
2066	return error;
2067}
2068
2069/*
2070 * dge_rxdrain:
2071 *
2072 *	Drain the receive queue.
2073 */
2074static void
2075dge_rxdrain(struct dge_softc *sc)
2076{
2077	struct dge_rxsoft *rxs;
2078	int i;
2079
2080	for (i = 0; i < DGE_NRXDESC; i++) {
2081		rxs = &sc->sc_rxsoft[i];
2082		if (rxs->rxs_mbuf != NULL) {
2083			bus_dmamap_unload(sc->sc_dmat, rxs->rxs_dmamap);
2084			m_freem(rxs->rxs_mbuf);
2085			rxs->rxs_mbuf = NULL;
2086		}
2087	}
2088}
2089
2090/*
2091 * dge_stop:		[ifnet interface function]
2092 *
2093 *	Stop transmission on the interface.
2094 */
2095static void
2096dge_stop(struct ifnet *ifp, int disable)
2097{
2098	struct dge_softc *sc = ifp->if_softc;
2099	struct dge_txsoft *txs;
2100	int i;
2101
2102	/* Stop the transmit and receive processes. */
2103	CSR_WRITE(sc, DGE_TCTL, 0);
2104	CSR_WRITE(sc, DGE_RCTL, 0);
2105
2106	/* Release any queued transmit buffers. */
2107	for (i = 0; i < DGE_TXQUEUELEN; i++) {
2108		txs = &sc->sc_txsoft[i];
2109		if (txs->txs_mbuf != NULL) {
2110			bus_dmamap_unload(sc->sc_dmat, txs->txs_dmamap);
2111			m_freem(txs->txs_mbuf);
2112			txs->txs_mbuf = NULL;
2113		}
2114	}
2115
2116	/* Mark the interface as down and cancel the watchdog timer. */
2117	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
2118	ifp->if_timer = 0;
2119
2120	if (disable)
2121		dge_rxdrain(sc);
2122}
2123
2124/*
2125 * dge_add_rxbuf:
2126 *
2127 *	Add a receive buffer to the indiciated descriptor.
2128 */
2129static int
2130dge_add_rxbuf(struct dge_softc *sc, int idx)
2131{
2132	struct dge_rxsoft *rxs = &sc->sc_rxsoft[idx];
2133	struct mbuf *m;
2134	int error;
2135#ifdef DGE_OFFBYONE_RXBUG
2136	void *buf;
2137#endif
2138
2139	MGETHDR(m, M_DONTWAIT, MT_DATA);
2140	if (m == NULL)
2141		return ENOBUFS;
2142
2143#ifdef DGE_OFFBYONE_RXBUG
2144	if ((buf = dge_getbuf(sc)) == NULL)
2145		return ENOBUFS;
2146
2147	m->m_len = m->m_pkthdr.len = DGE_BUFFER_SIZE;
2148	MEXTADD(m, buf, DGE_BUFFER_SIZE, M_DEVBUF, dge_freebuf, sc);
2149	m->m_flags |= M_EXT_RW;
2150
2151	if (rxs->rxs_mbuf != NULL)
2152		bus_dmamap_unload(sc->sc_dmat, rxs->rxs_dmamap);
2153	rxs->rxs_mbuf = m;
2154
2155	error = bus_dmamap_load(sc->sc_dmat, rxs->rxs_dmamap, buf,
2156	    DGE_BUFFER_SIZE, NULL, BUS_DMA_READ | BUS_DMA_NOWAIT);
2157#else
2158	MCLGET(m, M_DONTWAIT);
2159	if ((m->m_flags & M_EXT) == 0) {
2160		m_freem(m);
2161		return ENOBUFS;
2162	}
2163
2164	if (rxs->rxs_mbuf != NULL)
2165		bus_dmamap_unload(sc->sc_dmat, rxs->rxs_dmamap);
2166
2167	rxs->rxs_mbuf = m;
2168
2169	m->m_len = m->m_pkthdr.len = m->m_ext.ext_size;
2170	error = bus_dmamap_load_mbuf(sc->sc_dmat, rxs->rxs_dmamap, m,
2171	    BUS_DMA_READ | BUS_DMA_NOWAIT);
2172#endif
2173	if (error) {
2174		printf("%s: unable to load rx DMA map %d, error = %d\n",
2175		    device_xname(sc->sc_dev), idx, error);
2176		panic("dge_add_rxbuf");	/* XXX XXX XXX */
2177	}
2178	bus_dmamap_sync(sc->sc_dmat, rxs->rxs_dmamap, 0,
2179	    rxs->rxs_dmamap->dm_mapsize, BUS_DMASYNC_PREREAD);
2180
2181	return 0;
2182}
2183
2184/*
2185 * dge_set_ral:
2186 *
2187 *	Set an entry in the receive address list.
2188 */
2189static void
2190dge_set_ral(struct dge_softc *sc, const uint8_t *enaddr, int idx)
2191{
2192	uint32_t ral_lo, ral_hi;
2193
2194	if (enaddr != NULL) {
2195		ral_lo = enaddr[0] | (enaddr[1] << 8) | (enaddr[2] << 16) |
2196		    (enaddr[3] << 24);
2197		ral_hi = enaddr[4] | (enaddr[5] << 8);
2198		ral_hi |= RAH_AV;
2199	} else {
2200		ral_lo = 0;
2201		ral_hi = 0;
2202	}
2203	CSR_WRITE(sc, RA_ADDR(DGE_RAL, idx), ral_lo);
2204	CSR_WRITE(sc, RA_ADDR(DGE_RAH, idx), ral_hi);
2205}
2206
2207/*
2208 * dge_mchash:
2209 *
2210 *	Compute the hash of the multicast address for the 4096-bit
2211 *	multicast filter.
2212 */
2213static uint32_t
2214dge_mchash(struct dge_softc *sc, const uint8_t *enaddr)
2215{
2216	static const int lo_shift[4] = { 4, 3, 2, 0 };
2217	static const int hi_shift[4] = { 4, 5, 6, 8 };
2218	uint32_t hash;
2219
2220	hash = (enaddr[4] >> lo_shift[sc->sc_mchash_type]) |
2221	    (((uint16_t) enaddr[5]) << hi_shift[sc->sc_mchash_type]);
2222
2223	return (hash & 0xfff);
2224}
2225
2226/*
2227 * dge_set_filter:
2228 *
2229 *	Set up the receive filter.
2230 */
2231static void
2232dge_set_filter(struct dge_softc *sc)
2233{
2234	struct ethercom *ec = &sc->sc_ethercom;
2235	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
2236	struct ether_multi *enm;
2237	struct ether_multistep step;
2238	uint32_t hash, reg, bit;
2239	int i;
2240
2241	sc->sc_rctl &= ~(RCTL_BAM | RCTL_UPE | RCTL_MPE);
2242
2243	if (ifp->if_flags & IFF_BROADCAST)
2244		sc->sc_rctl |= RCTL_BAM;
2245	if (ifp->if_flags & IFF_PROMISC) {
2246		sc->sc_rctl |= RCTL_UPE;
2247		goto allmulti;
2248	}
2249
2250	/*
2251	 * Set the station address in the first RAL slot, and
2252	 * clear the remaining slots.
2253	 */
2254	dge_set_ral(sc, CLLADDR(ifp->if_sadl), 0);
2255	for (i = 1; i < RA_TABSIZE; i++)
2256		dge_set_ral(sc, NULL, i);
2257
2258	/* Clear out the multicast table. */
2259	for (i = 0; i < MC_TABSIZE; i++)
2260		CSR_WRITE(sc, DGE_MTA + (i << 2), 0);
2261
2262	ETHER_LOCK(ec);
2263	ETHER_FIRST_MULTI(step, ec, enm);
2264	while (enm != NULL) {
2265		if (memcmp(enm->enm_addrlo, enm->enm_addrhi, ETHER_ADDR_LEN)) {
2266			/*
2267			 * We must listen to a range of multicast addresses.
2268			 * For now, just accept all multicasts, rather than
2269			 * trying to set only those filter bits needed to match
2270			 * the range.  (At this time, the only use of address
2271			 * ranges is for IP multicast routing, for which the
2272			 * range is big enough to require all bits set.)
2273			 */
2274			ETHER_UNLOCK(ec);
2275			goto allmulti;
2276		}
2277
2278		hash = dge_mchash(sc, enm->enm_addrlo);
2279
2280		reg = (hash >> 5) & 0x7f;
2281		bit = hash & 0x1f;
2282
2283		hash = CSR_READ(sc, DGE_MTA + (reg << 2));
2284		hash |= 1U << bit;
2285
2286		CSR_WRITE(sc, DGE_MTA + (reg << 2), hash);
2287
2288		ETHER_NEXT_MULTI(step, enm);
2289	}
2290	ETHER_UNLOCK(ec);
2291
2292	ifp->if_flags &= ~IFF_ALLMULTI;
2293	goto setit;
2294
2295 allmulti:
2296	ifp->if_flags |= IFF_ALLMULTI;
2297	sc->sc_rctl |= RCTL_MPE;
2298
2299 setit:
2300	CSR_WRITE(sc, DGE_RCTL, sc->sc_rctl);
2301}
2302
2303/*
2304 * Read in the EEPROM info and verify checksum.
2305 */
2306int
2307dge_read_eeprom(struct dge_softc *sc)
2308{
2309	uint16_t cksum;
2310	int i;
2311
2312	cksum = 0;
2313	for (i = 0; i < EEPROM_SIZE; i++) {
2314		sc->sc_eeprom[i] = dge_eeprom_word(sc, i);
2315		cksum += sc->sc_eeprom[i];
2316	}
2317	return cksum != EEPROM_CKSUM;
2318}
2319
2320
2321/*
2322 * Read a 16-bit word from address addr in the serial EEPROM.
2323 */
2324uint16_t
2325dge_eeprom_word(struct dge_softc *sc, int addr)
2326{
2327	uint32_t reg;
2328	uint16_t rval = 0;
2329	int i;
2330
2331	reg = CSR_READ(sc, DGE_EECD) & ~(EECD_SK | EECD_DI | EECD_CS);
2332
2333	/* Lower clock pulse (and data in to chip) */
2334	CSR_WRITE(sc, DGE_EECD, reg);
2335	/* Select chip */
2336	CSR_WRITE(sc, DGE_EECD, reg | EECD_CS);
2337
2338	/* Send read command */
2339	dge_eeprom_clockout(sc, 1);
2340	dge_eeprom_clockout(sc, 1);
2341	dge_eeprom_clockout(sc, 0);
2342
2343	/* Send address */
2344	for (i = 5; i >= 0; i--)
2345		dge_eeprom_clockout(sc, (addr >> i) & 1);
2346
2347	/* Read data */
2348	for (i = 0; i < 16; i++) {
2349		rval <<= 1;
2350		rval |= dge_eeprom_clockin(sc);
2351	}
2352
2353	/* Deselect chip */
2354	CSR_WRITE(sc, DGE_EECD, reg);
2355
2356	return rval;
2357}
2358
2359/*
2360 * Clock out a single bit to the EEPROM.
2361 */
2362void
2363dge_eeprom_clockout(struct dge_softc *sc, int bit)
2364{
2365	int reg;
2366
2367	reg = CSR_READ(sc, DGE_EECD) & ~(EECD_DI | EECD_SK);
2368	if (bit)
2369		reg |= EECD_DI;
2370
2371	CSR_WRITE(sc, DGE_EECD, reg);
2372	delay(2);
2373	CSR_WRITE(sc, DGE_EECD, reg | EECD_SK);
2374	delay(2);
2375	CSR_WRITE(sc, DGE_EECD, reg);
2376	delay(2);
2377}
2378
2379/*
2380 * Clock in a single bit from EEPROM.
2381 */
2382int
2383dge_eeprom_clockin(struct dge_softc *sc)
2384{
2385	int reg, rv;
2386
2387	reg = CSR_READ(sc, DGE_EECD) & ~(EECD_DI | EECD_DO | EECD_SK);
2388
2389	CSR_WRITE(sc, DGE_EECD, reg | EECD_SK); /* Raise clock */
2390	delay(2);
2391	rv = (CSR_READ(sc, DGE_EECD) & EECD_DO) != 0; /* Get bit */
2392	CSR_WRITE(sc, DGE_EECD, reg); /* Lower clock */
2393	delay(2);
2394
2395	return rv;
2396}
2397
2398static void
2399dge_xgmii_mediastatus(struct ifnet *ifp, struct ifmediareq *ifmr)
2400{
2401	struct dge_softc *sc = ifp->if_softc;
2402
2403	ifmr->ifm_status = IFM_AVALID;
2404	if (sc->sc_dgep->dgep_flags & DGEP_F_10G_SR ) {
2405		ifmr->ifm_active = IFM_ETHER | IFM_10G_SR;
2406	} else {
2407		ifmr->ifm_active = IFM_ETHER | IFM_10G_LR;
2408	}
2409
2410	if (CSR_READ(sc, DGE_STATUS) & STATUS_LINKUP)
2411		ifmr->ifm_status |= IFM_ACTIVE;
2412}
2413
2414static inline int
2415phwait(struct dge_softc *sc, int p, int r, int d, int type)
2416{
2417	int i, mdic;
2418
2419	CSR_WRITE(sc, DGE_MDIO,
2420	    MDIO_PHY(p) | MDIO_REG(r) | MDIO_DEV(d) | type | MDIO_CMD);
2421	for (i = 0; i < 10; i++) {
2422		delay(10);
2423		if (((mdic = CSR_READ(sc, DGE_MDIO)) & MDIO_CMD) == 0)
2424			break;
2425	}
2426	return mdic;
2427}
2428
2429static void
2430dge_xgmii_writereg(struct dge_softc *sc, int phy, int reg, int val)
2431{
2432	int mdic;
2433
2434	CSR_WRITE(sc, DGE_MDIRW, val);
2435	if (((mdic = phwait(sc, phy, reg, 1, MDIO_ADDR)) & MDIO_CMD)) {
2436		printf("%s: address cycle timeout; phy %d reg %d\n",
2437		    device_xname(sc->sc_dev), phy, reg);
2438		return;
2439	}
2440	if (((mdic = phwait(sc, phy, reg, 1, MDIO_WRITE)) & MDIO_CMD)) {
2441		printf("%s: write cycle timeout; phy %d reg %d\n",
2442		    device_xname(sc->sc_dev), phy, reg);
2443		return;
2444	}
2445}
2446
2447static void
2448dge_xgmii_reset(struct dge_softc *sc)
2449{
2450	dge_xgmii_writereg(sc, 0, 0, BMCR_RESET);
2451}
2452
2453static int
2454dge_xgmii_mediachange(struct ifnet *ifp)
2455{
2456	return 0;
2457}
2458