1/*-
2 * SPDX-License-Identifier: BSD-2-Clause
3 *
4 * Copyright (C) 2001 Eduardo Horvath.
5 * Copyright (c) 2001-2003 Thomas Moestl
6 * Copyright (c) 2007-2009 Marius Strobl <marius@FreeBSD.org>
7 * All rights reserved.
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 *
18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR  ``AS IS'' AND
19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR  BE LIABLE
22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28 * SUCH DAMAGE.
29 *
30 *	from: NetBSD: gem.c,v 1.21 2002/06/01 23:50:58 lukem Exp
31 *	from: FreeBSD: if_gem.c 182060 2008-08-23 15:03:26Z marius
32 */
33
34#include <sys/cdefs.h>
35/*
36 * driver for Sun Cassini/Cassini+ and National Semiconductor DP83065
37 * Saturn Gigabit Ethernet controllers
38 */
39
40#if 0
41#define	CAS_DEBUG
42#endif
43
44#include <sys/param.h>
45#include <sys/systm.h>
46#include <sys/bus.h>
47#include <sys/callout.h>
48#include <sys/endian.h>
49#include <sys/mbuf.h>
50#include <sys/malloc.h>
51#include <sys/kernel.h>
52#include <sys/lock.h>
53#include <sys/module.h>
54#include <sys/mutex.h>
55#include <sys/refcount.h>
56#include <sys/resource.h>
57#include <sys/rman.h>
58#include <sys/socket.h>
59#include <sys/sockio.h>
60#include <sys/taskqueue.h>
61
62#include <net/bpf.h>
63#include <net/ethernet.h>
64#include <net/if.h>
65#include <net/if_var.h>
66#include <net/if_arp.h>
67#include <net/if_dl.h>
68#include <net/if_media.h>
69#include <net/if_types.h>
70#include <net/if_vlan_var.h>
71
72#include <netinet/in.h>
73#include <netinet/in_systm.h>
74#include <netinet/ip.h>
75#include <netinet/tcp.h>
76#include <netinet/udp.h>
77
78#include <machine/bus.h>
79#if defined(__powerpc__)
80#include <dev/ofw/ofw_bus.h>
81#include <dev/ofw/openfirm.h>
82#include <machine/ofw_machdep.h>
83#endif
84#include <machine/resource.h>
85
86#include <dev/mii/mii.h>
87#include <dev/mii/miivar.h>
88
89#include <dev/cas/if_casreg.h>
90#include <dev/cas/if_casvar.h>
91
92#include <dev/pci/pcireg.h>
93#include <dev/pci/pcivar.h>
94
95#include "miibus_if.h"
96
97#define RINGASSERT(n , min, max)					\
98	CTASSERT(powerof2(n) && (n) >= (min) && (n) <= (max))
99
100RINGASSERT(CAS_NRXCOMP, 128, 32768);
101RINGASSERT(CAS_NRXDESC, 32, 8192);
102RINGASSERT(CAS_NRXDESC2, 32, 8192);
103RINGASSERT(CAS_NTXDESC, 32, 8192);
104
105#undef RINGASSERT
106
107#define	CCDASSERT(m, a)							\
108	CTASSERT((offsetof(struct cas_control_data, m) & ((a) - 1)) == 0)
109
110CCDASSERT(ccd_rxcomps, CAS_RX_COMP_ALIGN);
111CCDASSERT(ccd_rxdescs, CAS_RX_DESC_ALIGN);
112CCDASSERT(ccd_rxdescs2, CAS_RX_DESC_ALIGN);
113
114#undef CCDASSERT
115
116#define	CAS_TRIES	10000
117
118/*
119 * According to documentation, the hardware has support for basic TCP
120 * checksum offloading only, in practice this can be also used for UDP
121 * however (i.e. the problem of previous Sun NICs that a checksum of 0x0
122 * is not converted to 0xffff no longer exists).
123 */
124#define	CAS_CSUM_FEATURES	(CSUM_TCP | CSUM_UDP)
125
126static inline void cas_add_rxdesc(struct cas_softc *sc, u_int idx);
127static int	cas_attach(struct cas_softc *sc);
128static int	cas_bitwait(struct cas_softc *sc, bus_addr_t r, uint32_t clr,
129		    uint32_t set);
130static void	cas_cddma_callback(void *xsc, bus_dma_segment_t *segs,
131		    int nsegs, int error);
132static void	cas_detach(struct cas_softc *sc);
133static int	cas_disable_rx(struct cas_softc *sc);
134static int	cas_disable_tx(struct cas_softc *sc);
135static void	cas_eint(struct cas_softc *sc, u_int status);
136static void	cas_free(struct mbuf *m);
137static void	cas_init(void *xsc);
138static void	cas_init_locked(struct cas_softc *sc);
139static void	cas_init_regs(struct cas_softc *sc);
140static int	cas_intr(void *v);
141static void	cas_intr_task(void *arg, int pending __unused);
142static int	cas_ioctl(if_t ifp, u_long cmd, caddr_t data);
143static int	cas_load_txmbuf(struct cas_softc *sc, struct mbuf **m_head);
144static int	cas_mediachange(if_t ifp);
145static void	cas_mediastatus(if_t ifp, struct ifmediareq *ifmr);
146static void	cas_meminit(struct cas_softc *sc);
147static void	cas_mifinit(struct cas_softc *sc);
148static int	cas_mii_readreg(device_t dev, int phy, int reg);
149static void	cas_mii_statchg(device_t dev);
150static int	cas_mii_writereg(device_t dev, int phy, int reg, int val);
151static void	cas_reset(struct cas_softc *sc);
152static int	cas_reset_rx(struct cas_softc *sc);
153static int	cas_reset_tx(struct cas_softc *sc);
154static void	cas_resume(struct cas_softc *sc);
155static u_int	cas_descsize(u_int sz);
156static void	cas_rint(struct cas_softc *sc);
157static void	cas_rint_timeout(void *arg);
158static inline void cas_rxcksum(struct mbuf *m, uint16_t cksum);
159static inline void cas_rxcompinit(struct cas_rx_comp *rxcomp);
160static u_int	cas_rxcompsize(u_int sz);
161static void	cas_rxdma_callback(void *xsc, bus_dma_segment_t *segs,
162		    int nsegs, int error);
163static void	cas_setladrf(struct cas_softc *sc);
164static void	cas_start(if_t ifp);
165static void	cas_stop(if_t ifp);
166static void	cas_suspend(struct cas_softc *sc);
167static void	cas_tick(void *arg);
168static void	cas_tint(struct cas_softc *sc);
169static void	cas_tx_task(void *arg, int pending __unused);
170static inline void cas_txkick(struct cas_softc *sc);
171static void	cas_watchdog(struct cas_softc *sc);
172
173MODULE_DEPEND(cas, ether, 1, 1, 1);
174MODULE_DEPEND(cas, miibus, 1, 1, 1);
175
176#ifdef CAS_DEBUG
177#include <sys/ktr.h>
178#define	KTR_CAS		KTR_SPARE2
179#endif
180
181static int
182cas_attach(struct cas_softc *sc)
183{
184	struct cas_txsoft *txs;
185	if_t ifp;
186	int error, i;
187	uint32_t v;
188
189	/* Set up ifnet structure. */
190	ifp = sc->sc_ifp = if_alloc(IFT_ETHER);
191	if (ifp == NULL)
192		return (ENOSPC);
193	if_setsoftc(ifp, sc);
194	if_initname(ifp, device_get_name(sc->sc_dev),
195	    device_get_unit(sc->sc_dev));
196	if_setflags(ifp, IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST);
197	if_setstartfn(ifp, cas_start);
198	if_setioctlfn(ifp, cas_ioctl);
199	if_setinitfn(ifp, cas_init);
200	if_setsendqlen(ifp, CAS_TXQUEUELEN);
201	if_setsendqready(ifp);
202
203	callout_init_mtx(&sc->sc_tick_ch, &sc->sc_mtx, 0);
204	callout_init_mtx(&sc->sc_rx_ch, &sc->sc_mtx, 0);
205	/* Create local taskq. */
206	NET_TASK_INIT(&sc->sc_intr_task, 0, cas_intr_task, sc);
207	TASK_INIT(&sc->sc_tx_task, 1, cas_tx_task, ifp);
208	sc->sc_tq = taskqueue_create_fast("cas_taskq", M_WAITOK,
209	    taskqueue_thread_enqueue, &sc->sc_tq);
210	if (sc->sc_tq == NULL) {
211		device_printf(sc->sc_dev, "could not create taskqueue\n");
212		error = ENXIO;
213		goto fail_ifnet;
214	}
215	error = taskqueue_start_threads(&sc->sc_tq, 1, PI_NET, "%s taskq",
216	    device_get_nameunit(sc->sc_dev));
217	if (error != 0) {
218		device_printf(sc->sc_dev, "could not start threads\n");
219		goto fail_taskq;
220	}
221
222	/* Make sure the chip is stopped. */
223	cas_reset(sc);
224
225	error = bus_dma_tag_create(bus_get_dma_tag(sc->sc_dev), 1, 0,
226	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
227	    BUS_SPACE_MAXSIZE, 0, BUS_SPACE_MAXSIZE, 0, NULL, NULL,
228	    &sc->sc_pdmatag);
229	if (error != 0)
230		goto fail_taskq;
231
232	error = bus_dma_tag_create(sc->sc_pdmatag, 1, 0,
233	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
234	    CAS_PAGE_SIZE, 1, CAS_PAGE_SIZE, 0, NULL, NULL, &sc->sc_rdmatag);
235	if (error != 0)
236		goto fail_ptag;
237
238	error = bus_dma_tag_create(sc->sc_pdmatag, 1, 0,
239	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
240	    MCLBYTES * CAS_NTXSEGS, CAS_NTXSEGS, MCLBYTES,
241	    BUS_DMA_ALLOCNOW, NULL, NULL, &sc->sc_tdmatag);
242	if (error != 0)
243		goto fail_rtag;
244
245	error = bus_dma_tag_create(sc->sc_pdmatag, CAS_TX_DESC_ALIGN, 0,
246	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
247	    sizeof(struct cas_control_data), 1,
248	    sizeof(struct cas_control_data), 0,
249	    NULL, NULL, &sc->sc_cdmatag);
250	if (error != 0)
251		goto fail_ttag;
252
253	/*
254	 * Allocate the control data structures, create and load the
255	 * DMA map for it.
256	 */
257	if ((error = bus_dmamem_alloc(sc->sc_cdmatag,
258	    (void **)&sc->sc_control_data,
259	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
260	    &sc->sc_cddmamap)) != 0) {
261		device_printf(sc->sc_dev,
262		    "unable to allocate control data, error = %d\n", error);
263		goto fail_ctag;
264	}
265
266	sc->sc_cddma = 0;
267	if ((error = bus_dmamap_load(sc->sc_cdmatag, sc->sc_cddmamap,
268	    sc->sc_control_data, sizeof(struct cas_control_data),
269	    cas_cddma_callback, sc, 0)) != 0 || sc->sc_cddma == 0) {
270		device_printf(sc->sc_dev,
271		    "unable to load control data DMA map, error = %d\n",
272		    error);
273		goto fail_cmem;
274	}
275
276	/*
277	 * Initialize the transmit job descriptors.
278	 */
279	STAILQ_INIT(&sc->sc_txfreeq);
280	STAILQ_INIT(&sc->sc_txdirtyq);
281
282	/*
283	 * Create the transmit buffer DMA maps.
284	 */
285	error = ENOMEM;
286	for (i = 0; i < CAS_TXQUEUELEN; i++) {
287		txs = &sc->sc_txsoft[i];
288		txs->txs_mbuf = NULL;
289		txs->txs_ndescs = 0;
290		if ((error = bus_dmamap_create(sc->sc_tdmatag, 0,
291		    &txs->txs_dmamap)) != 0) {
292			device_printf(sc->sc_dev,
293			    "unable to create TX DMA map %d, error = %d\n",
294			    i, error);
295			goto fail_txd;
296		}
297		STAILQ_INSERT_TAIL(&sc->sc_txfreeq, txs, txs_q);
298	}
299
300	/*
301	 * Allocate the receive buffers, create and load the DMA maps
302	 * for them.
303	 */
304	for (i = 0; i < CAS_NRXDESC; i++) {
305		if ((error = bus_dmamem_alloc(sc->sc_rdmatag,
306		    &sc->sc_rxdsoft[i].rxds_buf, BUS_DMA_WAITOK,
307		    &sc->sc_rxdsoft[i].rxds_dmamap)) != 0) {
308			device_printf(sc->sc_dev,
309			    "unable to allocate RX buffer %d, error = %d\n",
310			    i, error);
311			goto fail_rxmem;
312		}
313
314		sc->sc_rxdptr = i;
315		sc->sc_rxdsoft[i].rxds_paddr = 0;
316		if ((error = bus_dmamap_load(sc->sc_rdmatag,
317		    sc->sc_rxdsoft[i].rxds_dmamap, sc->sc_rxdsoft[i].rxds_buf,
318		    CAS_PAGE_SIZE, cas_rxdma_callback, sc, 0)) != 0 ||
319		    sc->sc_rxdsoft[i].rxds_paddr == 0) {
320			device_printf(sc->sc_dev,
321			    "unable to load RX DMA map %d, error = %d\n",
322			    i, error);
323			goto fail_rxmap;
324		}
325	}
326
327	if ((sc->sc_flags & CAS_SERDES) == 0) {
328		CAS_WRITE_4(sc, CAS_PCS_DATAPATH, CAS_PCS_DATAPATH_MII);
329		CAS_BARRIER(sc, CAS_PCS_DATAPATH, 4,
330		    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
331		cas_mifinit(sc);
332		/*
333		 * Look for an external PHY.
334		 */
335		error = ENXIO;
336		v = CAS_READ_4(sc, CAS_MIF_CONF);
337		if ((v & CAS_MIF_CONF_MDI1) != 0) {
338			v |= CAS_MIF_CONF_PHY_SELECT;
339			CAS_WRITE_4(sc, CAS_MIF_CONF, v);
340			CAS_BARRIER(sc, CAS_MIF_CONF, 4,
341			    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
342			/* Enable/unfreeze the GMII pins of Saturn. */
343			if (sc->sc_variant == CAS_SATURN) {
344				CAS_WRITE_4(sc, CAS_SATURN_PCFG,
345				    CAS_READ_4(sc, CAS_SATURN_PCFG) &
346				    ~CAS_SATURN_PCFG_FSI);
347				CAS_BARRIER(sc, CAS_SATURN_PCFG, 4,
348				    BUS_SPACE_BARRIER_READ |
349				    BUS_SPACE_BARRIER_WRITE);
350				DELAY(10000);
351			}
352			error = mii_attach(sc->sc_dev, &sc->sc_miibus, ifp,
353			    cas_mediachange, cas_mediastatus, BMSR_DEFCAPMASK,
354			    MII_PHY_ANY, MII_OFFSET_ANY, MIIF_DOPAUSE);
355		}
356		/*
357		 * Fall back on an internal PHY if no external PHY was found.
358		 */
359		if (error != 0 && (v & CAS_MIF_CONF_MDI0) != 0) {
360			v &= ~CAS_MIF_CONF_PHY_SELECT;
361			CAS_WRITE_4(sc, CAS_MIF_CONF, v);
362			CAS_BARRIER(sc, CAS_MIF_CONF, 4,
363			    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
364			/* Freeze the GMII pins of Saturn for saving power. */
365			if (sc->sc_variant == CAS_SATURN) {
366				CAS_WRITE_4(sc, CAS_SATURN_PCFG,
367				    CAS_READ_4(sc, CAS_SATURN_PCFG) |
368				    CAS_SATURN_PCFG_FSI);
369				CAS_BARRIER(sc, CAS_SATURN_PCFG, 4,
370				    BUS_SPACE_BARRIER_READ |
371				    BUS_SPACE_BARRIER_WRITE);
372				DELAY(10000);
373			}
374			error = mii_attach(sc->sc_dev, &sc->sc_miibus, ifp,
375			    cas_mediachange, cas_mediastatus, BMSR_DEFCAPMASK,
376			    MII_PHY_ANY, MII_OFFSET_ANY, MIIF_DOPAUSE);
377		}
378	} else {
379		/*
380		 * Use the external PCS SERDES.
381		 */
382		CAS_WRITE_4(sc, CAS_PCS_DATAPATH, CAS_PCS_DATAPATH_SERDES);
383		CAS_BARRIER(sc, CAS_PCS_DATAPATH, 4, BUS_SPACE_BARRIER_WRITE);
384		/* Enable/unfreeze the SERDES pins of Saturn. */
385		if (sc->sc_variant == CAS_SATURN) {
386			CAS_WRITE_4(sc, CAS_SATURN_PCFG, 0);
387			CAS_BARRIER(sc, CAS_SATURN_PCFG, 4,
388			    BUS_SPACE_BARRIER_WRITE);
389		}
390		CAS_WRITE_4(sc, CAS_PCS_SERDES_CTRL, CAS_PCS_SERDES_CTRL_ESD);
391		CAS_BARRIER(sc, CAS_PCS_SERDES_CTRL, 4,
392		    BUS_SPACE_BARRIER_WRITE);
393		CAS_WRITE_4(sc, CAS_PCS_CONF, CAS_PCS_CONF_EN);
394		CAS_BARRIER(sc, CAS_PCS_CONF, 4,
395		    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
396		error = mii_attach(sc->sc_dev, &sc->sc_miibus, ifp,
397		    cas_mediachange, cas_mediastatus, BMSR_DEFCAPMASK,
398		    CAS_PHYAD_EXTERNAL, MII_OFFSET_ANY, MIIF_DOPAUSE);
399	}
400	if (error != 0) {
401		device_printf(sc->sc_dev, "attaching PHYs failed\n");
402		goto fail_rxmap;
403	}
404	sc->sc_mii = device_get_softc(sc->sc_miibus);
405
406	/*
407	 * From this point forward, the attachment cannot fail.  A failure
408	 * before this point releases all resources that may have been
409	 * allocated.
410	 */
411
412	/* Announce FIFO sizes. */
413	v = CAS_READ_4(sc, CAS_TX_FIFO_SIZE);
414	device_printf(sc->sc_dev, "%ukB RX FIFO, %ukB TX FIFO\n",
415	    CAS_RX_FIFO_SIZE / 1024, v / 16);
416
417	/* Attach the interface. */
418	ether_ifattach(ifp, sc->sc_enaddr);
419
420	/*
421	 * Tell the upper layer(s) we support long frames/checksum offloads.
422	 */
423	if_setifheaderlen(ifp, sizeof(struct ether_vlan_header));
424	if_setcapabilities(ifp, IFCAP_VLAN_MTU);
425	if ((sc->sc_flags & CAS_NO_CSUM) == 0) {
426		if_setcapabilitiesbit(ifp, IFCAP_HWCSUM, 0);
427		if_sethwassist(ifp, CAS_CSUM_FEATURES);
428	}
429	if_setcapenable(ifp, if_getcapabilities(ifp));
430
431	return (0);
432
433	/*
434	 * Free any resources we've allocated during the failed attach
435	 * attempt.  Do this in reverse order and fall through.
436	 */
437 fail_rxmap:
438	for (i = 0; i < CAS_NRXDESC; i++)
439		if (sc->sc_rxdsoft[i].rxds_paddr != 0)
440			bus_dmamap_unload(sc->sc_rdmatag,
441			    sc->sc_rxdsoft[i].rxds_dmamap);
442 fail_rxmem:
443	for (i = 0; i < CAS_NRXDESC; i++)
444		if (sc->sc_rxdsoft[i].rxds_buf != NULL)
445			bus_dmamem_free(sc->sc_rdmatag,
446			    sc->sc_rxdsoft[i].rxds_buf,
447			    sc->sc_rxdsoft[i].rxds_dmamap);
448 fail_txd:
449	for (i = 0; i < CAS_TXQUEUELEN; i++)
450		if (sc->sc_txsoft[i].txs_dmamap != NULL)
451			bus_dmamap_destroy(sc->sc_tdmatag,
452			    sc->sc_txsoft[i].txs_dmamap);
453	bus_dmamap_unload(sc->sc_cdmatag, sc->sc_cddmamap);
454 fail_cmem:
455	bus_dmamem_free(sc->sc_cdmatag, sc->sc_control_data,
456	    sc->sc_cddmamap);
457 fail_ctag:
458	bus_dma_tag_destroy(sc->sc_cdmatag);
459 fail_ttag:
460	bus_dma_tag_destroy(sc->sc_tdmatag);
461 fail_rtag:
462	bus_dma_tag_destroy(sc->sc_rdmatag);
463 fail_ptag:
464	bus_dma_tag_destroy(sc->sc_pdmatag);
465 fail_taskq:
466	taskqueue_free(sc->sc_tq);
467 fail_ifnet:
468	if_free(ifp);
469	return (error);
470}
471
472static void
473cas_detach(struct cas_softc *sc)
474{
475	if_t ifp = sc->sc_ifp;
476	int i;
477
478	ether_ifdetach(ifp);
479	CAS_LOCK(sc);
480	cas_stop(ifp);
481	CAS_UNLOCK(sc);
482	callout_drain(&sc->sc_tick_ch);
483	callout_drain(&sc->sc_rx_ch);
484	taskqueue_drain(sc->sc_tq, &sc->sc_intr_task);
485	taskqueue_drain(sc->sc_tq, &sc->sc_tx_task);
486	if_free(ifp);
487	taskqueue_free(sc->sc_tq);
488	device_delete_child(sc->sc_dev, sc->sc_miibus);
489
490	for (i = 0; i < CAS_NRXDESC; i++)
491		if (sc->sc_rxdsoft[i].rxds_dmamap != NULL)
492			bus_dmamap_sync(sc->sc_rdmatag,
493			    sc->sc_rxdsoft[i].rxds_dmamap,
494			    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
495	for (i = 0; i < CAS_NRXDESC; i++)
496		if (sc->sc_rxdsoft[i].rxds_paddr != 0)
497			bus_dmamap_unload(sc->sc_rdmatag,
498			    sc->sc_rxdsoft[i].rxds_dmamap);
499	for (i = 0; i < CAS_NRXDESC; i++)
500		if (sc->sc_rxdsoft[i].rxds_buf != NULL)
501			bus_dmamem_free(sc->sc_rdmatag,
502			    sc->sc_rxdsoft[i].rxds_buf,
503			    sc->sc_rxdsoft[i].rxds_dmamap);
504	for (i = 0; i < CAS_TXQUEUELEN; i++)
505		if (sc->sc_txsoft[i].txs_dmamap != NULL)
506			bus_dmamap_destroy(sc->sc_tdmatag,
507			    sc->sc_txsoft[i].txs_dmamap);
508	CAS_CDSYNC(sc, BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
509	bus_dmamap_unload(sc->sc_cdmatag, sc->sc_cddmamap);
510	bus_dmamem_free(sc->sc_cdmatag, sc->sc_control_data,
511	    sc->sc_cddmamap);
512	bus_dma_tag_destroy(sc->sc_cdmatag);
513	bus_dma_tag_destroy(sc->sc_tdmatag);
514	bus_dma_tag_destroy(sc->sc_rdmatag);
515	bus_dma_tag_destroy(sc->sc_pdmatag);
516}
517
518static void
519cas_suspend(struct cas_softc *sc)
520{
521	if_t ifp = sc->sc_ifp;
522
523	CAS_LOCK(sc);
524	cas_stop(ifp);
525	CAS_UNLOCK(sc);
526}
527
528static void
529cas_resume(struct cas_softc *sc)
530{
531	if_t ifp = sc->sc_ifp;
532
533	CAS_LOCK(sc);
534	/*
535	 * On resume all registers have to be initialized again like
536	 * after power-on.
537	 */
538	sc->sc_flags &= ~CAS_INITED;
539	if (if_getflags(ifp) & IFF_UP)
540		cas_init_locked(sc);
541	CAS_UNLOCK(sc);
542}
543
544static inline void
545cas_rxcksum(struct mbuf *m, uint16_t cksum)
546{
547	struct ether_header *eh;
548	struct ip *ip;
549	struct udphdr *uh;
550	uint16_t *opts;
551	int32_t hlen, len, pktlen;
552	uint32_t temp32;
553
554	pktlen = m->m_pkthdr.len;
555	if (pktlen < sizeof(struct ether_header) + sizeof(struct ip))
556		return;
557	eh = mtod(m, struct ether_header *);
558	if (eh->ether_type != htons(ETHERTYPE_IP))
559		return;
560	ip = (struct ip *)(eh + 1);
561	if (ip->ip_v != IPVERSION)
562		return;
563
564	hlen = ip->ip_hl << 2;
565	pktlen -= sizeof(struct ether_header);
566	if (hlen < sizeof(struct ip))
567		return;
568	if (ntohs(ip->ip_len) < hlen)
569		return;
570	if (ntohs(ip->ip_len) != pktlen)
571		return;
572	if (ip->ip_off & htons(IP_MF | IP_OFFMASK))
573		return;	/* Cannot handle fragmented packet. */
574
575	switch (ip->ip_p) {
576	case IPPROTO_TCP:
577		if (pktlen < (hlen + sizeof(struct tcphdr)))
578			return;
579		break;
580	case IPPROTO_UDP:
581		if (pktlen < (hlen + sizeof(struct udphdr)))
582			return;
583		uh = (struct udphdr *)((uint8_t *)ip + hlen);
584		if (uh->uh_sum == 0)
585			return; /* no checksum */
586		break;
587	default:
588		return;
589	}
590
591	cksum = ~cksum;
592	/* checksum fixup for IP options */
593	len = hlen - sizeof(struct ip);
594	if (len > 0) {
595		opts = (uint16_t *)(ip + 1);
596		for (; len > 0; len -= sizeof(uint16_t), opts++) {
597			temp32 = cksum - *opts;
598			temp32 = (temp32 >> 16) + (temp32 & 65535);
599			cksum = temp32 & 65535;
600		}
601	}
602	m->m_pkthdr.csum_flags |= CSUM_DATA_VALID;
603	m->m_pkthdr.csum_data = cksum;
604}
605
606static void
607cas_cddma_callback(void *xsc, bus_dma_segment_t *segs, int nsegs, int error)
608{
609	struct cas_softc *sc = xsc;
610
611	if (error != 0)
612		return;
613	if (nsegs != 1)
614		panic("%s: bad control buffer segment count", __func__);
615	sc->sc_cddma = segs[0].ds_addr;
616}
617
618static void
619cas_rxdma_callback(void *xsc, bus_dma_segment_t *segs, int nsegs, int error)
620{
621	struct cas_softc *sc = xsc;
622
623	if (error != 0)
624		return;
625	if (nsegs != 1)
626		panic("%s: bad RX buffer segment count", __func__);
627	sc->sc_rxdsoft[sc->sc_rxdptr].rxds_paddr = segs[0].ds_addr;
628}
629
630static void
631cas_tick(void *arg)
632{
633	struct cas_softc *sc = arg;
634	if_t ifp = sc->sc_ifp;
635	uint32_t v;
636
637	CAS_LOCK_ASSERT(sc, MA_OWNED);
638
639	/*
640	 * Unload collision and error counters.
641	 */
642	if_inc_counter(ifp, IFCOUNTER_COLLISIONS,
643	    CAS_READ_4(sc, CAS_MAC_NORM_COLL_CNT) +
644	    CAS_READ_4(sc, CAS_MAC_FIRST_COLL_CNT));
645	v = CAS_READ_4(sc, CAS_MAC_EXCESS_COLL_CNT) +
646	    CAS_READ_4(sc, CAS_MAC_LATE_COLL_CNT);
647	if_inc_counter(ifp, IFCOUNTER_COLLISIONS, v);
648	if_inc_counter(ifp, IFCOUNTER_OERRORS, v);
649	if_inc_counter(ifp, IFCOUNTER_IERRORS,
650	    CAS_READ_4(sc, CAS_MAC_RX_LEN_ERR_CNT) +
651	    CAS_READ_4(sc, CAS_MAC_RX_ALIGN_ERR) +
652	    CAS_READ_4(sc, CAS_MAC_RX_CRC_ERR_CNT) +
653	    CAS_READ_4(sc, CAS_MAC_RX_CODE_VIOL));
654
655	/*
656	 * Then clear the hardware counters.
657	 */
658	CAS_WRITE_4(sc, CAS_MAC_NORM_COLL_CNT, 0);
659	CAS_WRITE_4(sc, CAS_MAC_FIRST_COLL_CNT, 0);
660	CAS_WRITE_4(sc, CAS_MAC_EXCESS_COLL_CNT, 0);
661	CAS_WRITE_4(sc, CAS_MAC_LATE_COLL_CNT, 0);
662	CAS_WRITE_4(sc, CAS_MAC_RX_LEN_ERR_CNT, 0);
663	CAS_WRITE_4(sc, CAS_MAC_RX_ALIGN_ERR, 0);
664	CAS_WRITE_4(sc, CAS_MAC_RX_CRC_ERR_CNT, 0);
665	CAS_WRITE_4(sc, CAS_MAC_RX_CODE_VIOL, 0);
666
667	mii_tick(sc->sc_mii);
668
669	if (sc->sc_txfree != CAS_MAXTXFREE)
670		cas_tint(sc);
671
672	cas_watchdog(sc);
673
674	callout_reset(&sc->sc_tick_ch, hz, cas_tick, sc);
675}
676
677static int
678cas_bitwait(struct cas_softc *sc, bus_addr_t r, uint32_t clr, uint32_t set)
679{
680	int i;
681	uint32_t reg;
682
683	for (i = CAS_TRIES; i--; DELAY(100)) {
684		reg = CAS_READ_4(sc, r);
685		if ((reg & clr) == 0 && (reg & set) == set)
686			return (1);
687	}
688	return (0);
689}
690
691static void
692cas_reset(struct cas_softc *sc)
693{
694
695#ifdef CAS_DEBUG
696	CTR2(KTR_CAS, "%s: %s", device_get_name(sc->sc_dev), __func__);
697#endif
698	/* Disable all interrupts in order to avoid spurious ones. */
699	CAS_WRITE_4(sc, CAS_INTMASK, 0xffffffff);
700
701	cas_reset_rx(sc);
702	cas_reset_tx(sc);
703
704	/*
705	 * Do a full reset modulo the result of the last auto-negotiation
706	 * when using the SERDES.
707	 */
708	CAS_WRITE_4(sc, CAS_RESET, CAS_RESET_RX | CAS_RESET_TX |
709	    ((sc->sc_flags & CAS_SERDES) != 0 ? CAS_RESET_PCS_DIS : 0));
710	CAS_BARRIER(sc, CAS_RESET, 4,
711	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
712	DELAY(3000);
713	if (!cas_bitwait(sc, CAS_RESET, CAS_RESET_RX | CAS_RESET_TX, 0))
714		device_printf(sc->sc_dev, "cannot reset device\n");
715}
716
717static void
718cas_stop(if_t ifp)
719{
720	struct cas_softc *sc = if_getsoftc(ifp);
721	struct cas_txsoft *txs;
722
723#ifdef CAS_DEBUG
724	CTR2(KTR_CAS, "%s: %s", device_get_name(sc->sc_dev), __func__);
725#endif
726
727	callout_stop(&sc->sc_tick_ch);
728	callout_stop(&sc->sc_rx_ch);
729
730	/* Disable all interrupts in order to avoid spurious ones. */
731	CAS_WRITE_4(sc, CAS_INTMASK, 0xffffffff);
732
733	cas_reset_tx(sc);
734	cas_reset_rx(sc);
735
736	/*
737	 * Release any queued transmit buffers.
738	 */
739	while ((txs = STAILQ_FIRST(&sc->sc_txdirtyq)) != NULL) {
740		STAILQ_REMOVE_HEAD(&sc->sc_txdirtyq, txs_q);
741		if (txs->txs_ndescs != 0) {
742			bus_dmamap_sync(sc->sc_tdmatag, txs->txs_dmamap,
743			    BUS_DMASYNC_POSTWRITE);
744			bus_dmamap_unload(sc->sc_tdmatag, txs->txs_dmamap);
745			if (txs->txs_mbuf != NULL) {
746				m_freem(txs->txs_mbuf);
747				txs->txs_mbuf = NULL;
748			}
749		}
750		STAILQ_INSERT_TAIL(&sc->sc_txfreeq, txs, txs_q);
751	}
752
753	/*
754	 * Mark the interface down and cancel the watchdog timer.
755	 */
756	if_setdrvflagbits(ifp, 0, (IFF_DRV_RUNNING | IFF_DRV_OACTIVE));
757	sc->sc_flags &= ~CAS_LINK;
758	sc->sc_wdog_timer = 0;
759}
760
761static int
762cas_reset_rx(struct cas_softc *sc)
763{
764
765	/*
766	 * Resetting while DMA is in progress can cause a bus hang, so we
767	 * disable DMA first.
768	 */
769	(void)cas_disable_rx(sc);
770	CAS_WRITE_4(sc, CAS_RX_CONF, 0);
771	CAS_BARRIER(sc, CAS_RX_CONF, 4,
772	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
773	if (!cas_bitwait(sc, CAS_RX_CONF, CAS_RX_CONF_RXDMA_EN, 0))
774		device_printf(sc->sc_dev, "cannot disable RX DMA\n");
775
776	/* Finally, reset the ERX. */
777	CAS_WRITE_4(sc, CAS_RESET, CAS_RESET_RX |
778	    ((sc->sc_flags & CAS_SERDES) != 0 ? CAS_RESET_PCS_DIS : 0));
779	CAS_BARRIER(sc, CAS_RESET, 4,
780	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
781	if (!cas_bitwait(sc, CAS_RESET, CAS_RESET_RX, 0)) {
782		device_printf(sc->sc_dev, "cannot reset receiver\n");
783		return (1);
784	}
785	return (0);
786}
787
788static int
789cas_reset_tx(struct cas_softc *sc)
790{
791
792	/*
793	 * Resetting while DMA is in progress can cause a bus hang, so we
794	 * disable DMA first.
795	 */
796	(void)cas_disable_tx(sc);
797	CAS_WRITE_4(sc, CAS_TX_CONF, 0);
798	CAS_BARRIER(sc, CAS_TX_CONF, 4,
799	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
800	if (!cas_bitwait(sc, CAS_TX_CONF, CAS_TX_CONF_TXDMA_EN, 0))
801		device_printf(sc->sc_dev, "cannot disable TX DMA\n");
802
803	/* Finally, reset the ETX. */
804	CAS_WRITE_4(sc, CAS_RESET, CAS_RESET_TX |
805	    ((sc->sc_flags & CAS_SERDES) != 0 ? CAS_RESET_PCS_DIS : 0));
806	CAS_BARRIER(sc, CAS_RESET, 4,
807	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
808	if (!cas_bitwait(sc, CAS_RESET, CAS_RESET_TX, 0)) {
809		device_printf(sc->sc_dev, "cannot reset transmitter\n");
810		return (1);
811	}
812	return (0);
813}
814
815static int
816cas_disable_rx(struct cas_softc *sc)
817{
818
819	CAS_WRITE_4(sc, CAS_MAC_RX_CONF,
820	    CAS_READ_4(sc, CAS_MAC_RX_CONF) & ~CAS_MAC_RX_CONF_EN);
821	CAS_BARRIER(sc, CAS_MAC_RX_CONF, 4,
822	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
823	if (cas_bitwait(sc, CAS_MAC_RX_CONF, CAS_MAC_RX_CONF_EN, 0))
824		return (1);
825	if (bootverbose)
826		device_printf(sc->sc_dev, "cannot disable RX MAC\n");
827	return (0);
828}
829
830static int
831cas_disable_tx(struct cas_softc *sc)
832{
833
834	CAS_WRITE_4(sc, CAS_MAC_TX_CONF,
835	    CAS_READ_4(sc, CAS_MAC_TX_CONF) & ~CAS_MAC_TX_CONF_EN);
836	CAS_BARRIER(sc, CAS_MAC_TX_CONF, 4,
837	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
838	if (cas_bitwait(sc, CAS_MAC_TX_CONF, CAS_MAC_TX_CONF_EN, 0))
839		return (1);
840	if (bootverbose)
841		device_printf(sc->sc_dev, "cannot disable TX MAC\n");
842	return (0);
843}
844
845static inline void
846cas_rxcompinit(struct cas_rx_comp *rxcomp)
847{
848
849	rxcomp->crc_word1 = 0;
850	rxcomp->crc_word2 = 0;
851	rxcomp->crc_word3 =
852	    htole64(CAS_SET(ETHER_HDR_LEN + sizeof(struct ip), CAS_RC3_CSO));
853	rxcomp->crc_word4 = htole64(CAS_RC4_ZERO);
854}
855
856static void
857cas_meminit(struct cas_softc *sc)
858{
859	int i;
860
861	CAS_LOCK_ASSERT(sc, MA_OWNED);
862
863	/*
864	 * Initialize the transmit descriptor ring.
865	 */
866	for (i = 0; i < CAS_NTXDESC; i++) {
867		sc->sc_txdescs[i].cd_flags = 0;
868		sc->sc_txdescs[i].cd_buf_ptr = 0;
869	}
870	sc->sc_txfree = CAS_MAXTXFREE;
871	sc->sc_txnext = 0;
872	sc->sc_txwin = 0;
873
874	/*
875	 * Initialize the receive completion ring.
876	 */
877	for (i = 0; i < CAS_NRXCOMP; i++)
878		cas_rxcompinit(&sc->sc_rxcomps[i]);
879	sc->sc_rxcptr = 0;
880
881	/*
882	 * Initialize the first receive descriptor ring.  We leave
883	 * the second one zeroed as we don't actually use it.
884	 */
885	for (i = 0; i < CAS_NRXDESC; i++)
886		CAS_INIT_RXDESC(sc, i, i);
887	sc->sc_rxdptr = 0;
888
889	CAS_CDSYNC(sc, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
890}
891
892static u_int
893cas_descsize(u_int sz)
894{
895
896	switch (sz) {
897	case 32:
898		return (CAS_DESC_32);
899	case 64:
900		return (CAS_DESC_64);
901	case 128:
902		return (CAS_DESC_128);
903	case 256:
904		return (CAS_DESC_256);
905	case 512:
906		return (CAS_DESC_512);
907	case 1024:
908		return (CAS_DESC_1K);
909	case 2048:
910		return (CAS_DESC_2K);
911	case 4096:
912		return (CAS_DESC_4K);
913	case 8192:
914		return (CAS_DESC_8K);
915	default:
916		printf("%s: invalid descriptor ring size %d\n", __func__, sz);
917		return (CAS_DESC_32);
918	}
919}
920
921static u_int
922cas_rxcompsize(u_int sz)
923{
924
925	switch (sz) {
926	case 128:
927		return (CAS_RX_CONF_COMP_128);
928	case 256:
929		return (CAS_RX_CONF_COMP_256);
930	case 512:
931		return (CAS_RX_CONF_COMP_512);
932	case 1024:
933		return (CAS_RX_CONF_COMP_1K);
934	case 2048:
935		return (CAS_RX_CONF_COMP_2K);
936	case 4096:
937		return (CAS_RX_CONF_COMP_4K);
938	case 8192:
939		return (CAS_RX_CONF_COMP_8K);
940	case 16384:
941		return (CAS_RX_CONF_COMP_16K);
942	case 32768:
943		return (CAS_RX_CONF_COMP_32K);
944	default:
945		printf("%s: invalid dcompletion ring size %d\n", __func__, sz);
946		return (CAS_RX_CONF_COMP_128);
947	}
948}
949
950static void
951cas_init(void *xsc)
952{
953	struct cas_softc *sc = xsc;
954
955	CAS_LOCK(sc);
956	cas_init_locked(sc);
957	CAS_UNLOCK(sc);
958}
959
960/*
961 * Initialization of interface; set up initialization block
962 * and transmit/receive descriptor rings.
963 */
964static void
965cas_init_locked(struct cas_softc *sc)
966{
967	if_t ifp = sc->sc_ifp;
968	uint32_t v;
969
970	CAS_LOCK_ASSERT(sc, MA_OWNED);
971
972	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0)
973		return;
974
975#ifdef CAS_DEBUG
976	CTR2(KTR_CAS, "%s: %s: calling stop", device_get_name(sc->sc_dev),
977	    __func__);
978#endif
979	/*
980	 * Initialization sequence.  The numbered steps below correspond
981	 * to the sequence outlined in section 6.3.5.1 in the Ethernet
982	 * Channel Engine manual (part of the PCIO manual).
983	 * See also the STP2002-STQ document from Sun Microsystems.
984	 */
985
986	/* step 1 & 2.  Reset the Ethernet Channel. */
987	cas_stop(ifp);
988	cas_reset(sc);
989#ifdef CAS_DEBUG
990	CTR2(KTR_CAS, "%s: %s: restarting", device_get_name(sc->sc_dev),
991	    __func__);
992#endif
993
994	if ((sc->sc_flags & CAS_SERDES) == 0)
995		/* Re-initialize the MIF. */
996		cas_mifinit(sc);
997
998	/* step 3.  Setup data structures in host memory. */
999	cas_meminit(sc);
1000
1001	/* step 4.  TX MAC registers & counters */
1002	cas_init_regs(sc);
1003
1004	/* step 5.  RX MAC registers & counters */
1005
1006	/* step 6 & 7.  Program Ring Base Addresses. */
1007	CAS_WRITE_4(sc, CAS_TX_DESC3_BASE_HI,
1008	    (((uint64_t)CAS_CDTXDADDR(sc, 0)) >> 32));
1009	CAS_WRITE_4(sc, CAS_TX_DESC3_BASE_LO,
1010	    CAS_CDTXDADDR(sc, 0) & 0xffffffff);
1011
1012	CAS_WRITE_4(sc, CAS_RX_COMP_BASE_HI,
1013	    (((uint64_t)CAS_CDRXCADDR(sc, 0)) >> 32));
1014	CAS_WRITE_4(sc, CAS_RX_COMP_BASE_LO,
1015	    CAS_CDRXCADDR(sc, 0) & 0xffffffff);
1016
1017	CAS_WRITE_4(sc, CAS_RX_DESC_BASE_HI,
1018	    (((uint64_t)CAS_CDRXDADDR(sc, 0)) >> 32));
1019	CAS_WRITE_4(sc, CAS_RX_DESC_BASE_LO,
1020	    CAS_CDRXDADDR(sc, 0) & 0xffffffff);
1021
1022	if ((sc->sc_flags & CAS_REG_PLUS) != 0) {
1023		CAS_WRITE_4(sc, CAS_RX_DESC2_BASE_HI,
1024		    (((uint64_t)CAS_CDRXD2ADDR(sc, 0)) >> 32));
1025		CAS_WRITE_4(sc, CAS_RX_DESC2_BASE_LO,
1026		    CAS_CDRXD2ADDR(sc, 0) & 0xffffffff);
1027	}
1028
1029#ifdef CAS_DEBUG
1030	CTR5(KTR_CAS,
1031	    "loading TXDR %lx, RXCR %lx, RXDR %lx, RXD2R %lx, cddma %lx",
1032	    CAS_CDTXDADDR(sc, 0), CAS_CDRXCADDR(sc, 0), CAS_CDRXDADDR(sc, 0),
1033	    CAS_CDRXD2ADDR(sc, 0), sc->sc_cddma);
1034#endif
1035
1036	/* step 8.  Global Configuration & Interrupt Masks */
1037
1038	/* Disable weighted round robin. */
1039	CAS_WRITE_4(sc, CAS_CAW, CAS_CAW_RR_DIS);
1040
1041	/*
1042	 * Enable infinite bursts for revisions without PCI issues if
1043	 * applicable.  Doing so greatly improves the TX performance.
1044	 */
1045	CAS_WRITE_4(sc, CAS_INF_BURST,
1046	    (sc->sc_flags & CAS_TABORT) == 0 ? CAS_INF_BURST_EN :
1047	    0);
1048
1049	/* Set up interrupts. */
1050	CAS_WRITE_4(sc, CAS_INTMASK,
1051	    ~(CAS_INTR_TX_INT_ME | CAS_INTR_TX_TAG_ERR |
1052	    CAS_INTR_RX_DONE | CAS_INTR_RX_BUF_NA | CAS_INTR_RX_TAG_ERR |
1053	    CAS_INTR_RX_COMP_FULL | CAS_INTR_RX_BUF_AEMPTY |
1054	    CAS_INTR_RX_COMP_AFULL | CAS_INTR_RX_LEN_MMATCH |
1055	    CAS_INTR_PCI_ERROR_INT
1056#ifdef CAS_DEBUG
1057	    | CAS_INTR_PCS_INT | CAS_INTR_MIF
1058#endif
1059	    ));
1060	/* Don't clear top level interrupts when CAS_STATUS_ALIAS is read. */
1061	CAS_WRITE_4(sc, CAS_CLEAR_ALIAS, 0);
1062	CAS_WRITE_4(sc, CAS_MAC_RX_MASK, ~CAS_MAC_RX_OVERFLOW);
1063	CAS_WRITE_4(sc, CAS_MAC_TX_MASK,
1064	    ~(CAS_MAC_TX_UNDERRUN | CAS_MAC_TX_MAX_PKT_ERR));
1065#ifdef CAS_DEBUG
1066	CAS_WRITE_4(sc, CAS_MAC_CTRL_MASK,
1067	    ~(CAS_MAC_CTRL_PAUSE_RCVD | CAS_MAC_CTRL_PAUSE |
1068	    CAS_MAC_CTRL_NON_PAUSE));
1069#else
1070	CAS_WRITE_4(sc, CAS_MAC_CTRL_MASK,
1071	    CAS_MAC_CTRL_PAUSE_RCVD | CAS_MAC_CTRL_PAUSE |
1072	    CAS_MAC_CTRL_NON_PAUSE);
1073#endif
1074
1075	/* Enable PCI error interrupts. */
1076	CAS_WRITE_4(sc, CAS_ERROR_MASK,
1077	    ~(CAS_ERROR_DTRTO | CAS_ERROR_OTHER | CAS_ERROR_DMAW_ZERO |
1078	    CAS_ERROR_DMAR_ZERO | CAS_ERROR_RTRTO));
1079
1080	/* Enable PCI error interrupts in BIM configuration. */
1081	CAS_WRITE_4(sc, CAS_BIM_CONF,
1082	    CAS_BIM_CONF_DPAR_EN | CAS_BIM_CONF_RMA_EN | CAS_BIM_CONF_RTA_EN);
1083
1084	/*
1085	 * step 9.  ETX Configuration: encode receive descriptor ring size,
1086	 * enable DMA and disable pre-interrupt writeback completion.
1087	 */
1088	v = cas_descsize(CAS_NTXDESC) << CAS_TX_CONF_DESC3_SHFT;
1089	CAS_WRITE_4(sc, CAS_TX_CONF, v | CAS_TX_CONF_TXDMA_EN |
1090	    CAS_TX_CONF_RDPP_DIS | CAS_TX_CONF_PICWB_DIS);
1091
1092	/* step 10.  ERX Configuration */
1093
1094	/*
1095	 * Encode receive completion and descriptor ring sizes, set the
1096	 * swivel offset.
1097	 */
1098	v = cas_rxcompsize(CAS_NRXCOMP) << CAS_RX_CONF_COMP_SHFT;
1099	v |= cas_descsize(CAS_NRXDESC) << CAS_RX_CONF_DESC_SHFT;
1100	if ((sc->sc_flags & CAS_REG_PLUS) != 0)
1101		v |= cas_descsize(CAS_NRXDESC2) << CAS_RX_CONF_DESC2_SHFT;
1102	CAS_WRITE_4(sc, CAS_RX_CONF,
1103	    v | (ETHER_ALIGN << CAS_RX_CONF_SOFF_SHFT));
1104
1105	/* Set the PAUSE thresholds.  We use the maximum OFF threshold. */
1106	CAS_WRITE_4(sc, CAS_RX_PTHRS,
1107	    (111 << CAS_RX_PTHRS_XOFF_SHFT) | (15 << CAS_RX_PTHRS_XON_SHFT));
1108
1109	/* RX blanking */
1110	CAS_WRITE_4(sc, CAS_RX_BLANK,
1111	    (15 << CAS_RX_BLANK_TIME_SHFT) | (5 << CAS_RX_BLANK_PKTS_SHFT));
1112
1113	/* Set RX_COMP_AFULL threshold to half of the RX completions. */
1114	CAS_WRITE_4(sc, CAS_RX_AEMPTY_THRS,
1115	    (CAS_NRXCOMP / 2) << CAS_RX_AEMPTY_COMP_SHFT);
1116
1117	/* Initialize the RX page size register as appropriate for 8k. */
1118	CAS_WRITE_4(sc, CAS_RX_PSZ,
1119	    (CAS_RX_PSZ_8K << CAS_RX_PSZ_SHFT) |
1120	    (4 << CAS_RX_PSZ_MB_CNT_SHFT) |
1121	    (CAS_RX_PSZ_MB_STRD_2K << CAS_RX_PSZ_MB_STRD_SHFT) |
1122	    (CAS_RX_PSZ_MB_OFF_64 << CAS_RX_PSZ_MB_OFF_SHFT));
1123
1124	/* Disable RX random early detection. */
1125	CAS_WRITE_4(sc,	CAS_RX_RED, 0);
1126
1127	/* Zero the RX reassembly DMA table. */
1128	for (v = 0; v <= CAS_RX_REAS_DMA_ADDR_LC; v++) {
1129		CAS_WRITE_4(sc,	CAS_RX_REAS_DMA_ADDR, v);
1130		CAS_WRITE_4(sc,	CAS_RX_REAS_DMA_DATA_LO, 0);
1131		CAS_WRITE_4(sc,	CAS_RX_REAS_DMA_DATA_MD, 0);
1132		CAS_WRITE_4(sc,	CAS_RX_REAS_DMA_DATA_HI, 0);
1133	}
1134
1135	/* Ensure the RX control FIFO and RX IPP FIFO addresses are zero. */
1136	CAS_WRITE_4(sc, CAS_RX_CTRL_FIFO, 0);
1137	CAS_WRITE_4(sc, CAS_RX_IPP_ADDR, 0);
1138
1139	/* Finally, enable RX DMA. */
1140	CAS_WRITE_4(sc, CAS_RX_CONF,
1141	    CAS_READ_4(sc, CAS_RX_CONF) | CAS_RX_CONF_RXDMA_EN);
1142
1143	/* step 11.  Configure Media. */
1144
1145	/* step 12.  RX_MAC Configuration Register */
1146	v = CAS_READ_4(sc, CAS_MAC_RX_CONF);
1147	v &= ~(CAS_MAC_RX_CONF_STRPPAD | CAS_MAC_RX_CONF_EN);
1148	v |= CAS_MAC_RX_CONF_STRPFCS;
1149	sc->sc_mac_rxcfg = v;
1150	/*
1151	 * Clear the RX filter and reprogram it.  This will also set the
1152	 * current RX MAC configuration and enable it.
1153	 */
1154	cas_setladrf(sc);
1155
1156	/* step 13.  TX_MAC Configuration Register */
1157	v = CAS_READ_4(sc, CAS_MAC_TX_CONF);
1158	v |= CAS_MAC_TX_CONF_EN;
1159	(void)cas_disable_tx(sc);
1160	CAS_WRITE_4(sc, CAS_MAC_TX_CONF, v);
1161
1162	/* step 14.  Issue Transmit Pending command. */
1163
1164	/* step 15.  Give the receiver a swift kick. */
1165	CAS_WRITE_4(sc, CAS_RX_KICK, CAS_NRXDESC - 4);
1166	CAS_WRITE_4(sc, CAS_RX_COMP_TAIL, 0);
1167	if ((sc->sc_flags & CAS_REG_PLUS) != 0)
1168		CAS_WRITE_4(sc, CAS_RX_KICK2, CAS_NRXDESC2 - 4);
1169
1170	if_setdrvflagbits(ifp, IFF_DRV_RUNNING, 0);
1171	if_setdrvflagbits(ifp, 0, IFF_DRV_OACTIVE);
1172
1173	mii_mediachg(sc->sc_mii);
1174
1175	/* Start the one second timer. */
1176	sc->sc_wdog_timer = 0;
1177	callout_reset(&sc->sc_tick_ch, hz, cas_tick, sc);
1178}
1179
1180static int
1181cas_load_txmbuf(struct cas_softc *sc, struct mbuf **m_head)
1182{
1183	bus_dma_segment_t txsegs[CAS_NTXSEGS];
1184	struct cas_txsoft *txs;
1185	struct ip *ip;
1186	struct mbuf *m;
1187	uint64_t cflags;
1188	int error, nexttx, nsegs, offset, seg;
1189
1190	CAS_LOCK_ASSERT(sc, MA_OWNED);
1191
1192	/* Get a work queue entry. */
1193	if ((txs = STAILQ_FIRST(&sc->sc_txfreeq)) == NULL) {
1194		/* Ran out of descriptors. */
1195		return (ENOBUFS);
1196	}
1197
1198	cflags = 0;
1199	if (((*m_head)->m_pkthdr.csum_flags & CAS_CSUM_FEATURES) != 0) {
1200		if (M_WRITABLE(*m_head) == 0) {
1201			m = m_dup(*m_head, M_NOWAIT);
1202			m_freem(*m_head);
1203			*m_head = m;
1204			if (m == NULL)
1205				return (ENOBUFS);
1206		}
1207		offset = sizeof(struct ether_header);
1208		m = m_pullup(*m_head, offset + sizeof(struct ip));
1209		if (m == NULL) {
1210			*m_head = NULL;
1211			return (ENOBUFS);
1212		}
1213		ip = (struct ip *)(mtod(m, caddr_t) + offset);
1214		offset += (ip->ip_hl << 2);
1215		cflags = (offset << CAS_TD_CKSUM_START_SHFT) |
1216		    ((offset + m->m_pkthdr.csum_data) <<
1217		    CAS_TD_CKSUM_STUFF_SHFT) | CAS_TD_CKSUM_EN;
1218		*m_head = m;
1219	}
1220
1221	error = bus_dmamap_load_mbuf_sg(sc->sc_tdmatag, txs->txs_dmamap,
1222	    *m_head, txsegs, &nsegs, BUS_DMA_NOWAIT);
1223	if (error == EFBIG) {
1224		m = m_collapse(*m_head, M_NOWAIT, CAS_NTXSEGS);
1225		if (m == NULL) {
1226			m_freem(*m_head);
1227			*m_head = NULL;
1228			return (ENOBUFS);
1229		}
1230		*m_head = m;
1231		error = bus_dmamap_load_mbuf_sg(sc->sc_tdmatag,
1232		    txs->txs_dmamap, *m_head, txsegs, &nsegs,
1233		    BUS_DMA_NOWAIT);
1234		if (error != 0) {
1235			m_freem(*m_head);
1236			*m_head = NULL;
1237			return (error);
1238		}
1239	} else if (error != 0)
1240		return (error);
1241	/* If nsegs is wrong then the stack is corrupt. */
1242	KASSERT(nsegs <= CAS_NTXSEGS,
1243	    ("%s: too many DMA segments (%d)", __func__, nsegs));
1244	if (nsegs == 0) {
1245		m_freem(*m_head);
1246		*m_head = NULL;
1247		return (EIO);
1248	}
1249
1250	/*
1251	 * Ensure we have enough descriptors free to describe
1252	 * the packet.  Note, we always reserve one descriptor
1253	 * at the end of the ring as a termination point, in
1254	 * order to prevent wrap-around.
1255	 */
1256	if (nsegs > sc->sc_txfree - 1) {
1257		txs->txs_ndescs = 0;
1258		bus_dmamap_unload(sc->sc_tdmatag, txs->txs_dmamap);
1259		return (ENOBUFS);
1260	}
1261
1262	txs->txs_ndescs = nsegs;
1263	txs->txs_firstdesc = sc->sc_txnext;
1264	nexttx = txs->txs_firstdesc;
1265	for (seg = 0; seg < nsegs; seg++, nexttx = CAS_NEXTTX(nexttx)) {
1266#ifdef CAS_DEBUG
1267		CTR6(KTR_CAS,
1268		    "%s: mapping seg %d (txd %d), len %lx, addr %#lx (%#lx)",
1269		    __func__, seg, nexttx, txsegs[seg].ds_len,
1270		    txsegs[seg].ds_addr, htole64(txsegs[seg].ds_addr));
1271#endif
1272		sc->sc_txdescs[nexttx].cd_buf_ptr =
1273		    htole64(txsegs[seg].ds_addr);
1274		KASSERT(txsegs[seg].ds_len <
1275		    CAS_TD_BUF_LEN_MASK >> CAS_TD_BUF_LEN_SHFT,
1276		    ("%s: segment size too large!", __func__));
1277		sc->sc_txdescs[nexttx].cd_flags =
1278		    htole64(txsegs[seg].ds_len << CAS_TD_BUF_LEN_SHFT);
1279		txs->txs_lastdesc = nexttx;
1280	}
1281
1282	/* Set EOF on the last descriptor. */
1283#ifdef CAS_DEBUG
1284	CTR3(KTR_CAS, "%s: end of frame at segment %d, TX %d",
1285	    __func__, seg, nexttx);
1286#endif
1287	sc->sc_txdescs[txs->txs_lastdesc].cd_flags |=
1288	    htole64(CAS_TD_END_OF_FRAME);
1289
1290	/* Lastly set SOF on the first descriptor. */
1291#ifdef CAS_DEBUG
1292	CTR3(KTR_CAS, "%s: start of frame at segment %d, TX %d",
1293	    __func__, seg, nexttx);
1294#endif
1295	if (sc->sc_txwin += nsegs > CAS_MAXTXFREE * 2 / 3) {
1296		sc->sc_txwin = 0;
1297		sc->sc_txdescs[txs->txs_firstdesc].cd_flags |=
1298		    htole64(cflags | CAS_TD_START_OF_FRAME | CAS_TD_INT_ME);
1299	} else
1300		sc->sc_txdescs[txs->txs_firstdesc].cd_flags |=
1301		    htole64(cflags | CAS_TD_START_OF_FRAME);
1302
1303	/* Sync the DMA map. */
1304	bus_dmamap_sync(sc->sc_tdmatag, txs->txs_dmamap,
1305	    BUS_DMASYNC_PREWRITE);
1306
1307#ifdef CAS_DEBUG
1308	CTR4(KTR_CAS, "%s: setting firstdesc=%d, lastdesc=%d, ndescs=%d",
1309	    __func__, txs->txs_firstdesc, txs->txs_lastdesc,
1310	    txs->txs_ndescs);
1311#endif
1312	STAILQ_REMOVE_HEAD(&sc->sc_txfreeq, txs_q);
1313	STAILQ_INSERT_TAIL(&sc->sc_txdirtyq, txs, txs_q);
1314	txs->txs_mbuf = *m_head;
1315
1316	sc->sc_txnext = CAS_NEXTTX(txs->txs_lastdesc);
1317	sc->sc_txfree -= txs->txs_ndescs;
1318
1319	return (0);
1320}
1321
1322static void
1323cas_init_regs(struct cas_softc *sc)
1324{
1325	int i;
1326	const u_char *laddr = if_getlladdr(sc->sc_ifp);
1327
1328	CAS_LOCK_ASSERT(sc, MA_OWNED);
1329
1330	/* These registers are not cleared on reset. */
1331	if ((sc->sc_flags & CAS_INITED) == 0) {
1332		/* magic values */
1333		CAS_WRITE_4(sc, CAS_MAC_IPG0, 0);
1334		CAS_WRITE_4(sc, CAS_MAC_IPG1, 8);
1335		CAS_WRITE_4(sc, CAS_MAC_IPG2, 4);
1336
1337		/* min frame length */
1338		CAS_WRITE_4(sc, CAS_MAC_MIN_FRAME, ETHER_MIN_LEN);
1339		/* max frame length and max burst size */
1340		CAS_WRITE_4(sc, CAS_MAC_MAX_BF,
1341		    ((ETHER_MAX_LEN_JUMBO + ETHER_VLAN_ENCAP_LEN) <<
1342		    CAS_MAC_MAX_BF_FRM_SHFT) |
1343		    (0x2000 << CAS_MAC_MAX_BF_BST_SHFT));
1344
1345		/* more magic values */
1346		CAS_WRITE_4(sc, CAS_MAC_PREAMBLE_LEN, 0x7);
1347		CAS_WRITE_4(sc, CAS_MAC_JAM_SIZE, 0x4);
1348		CAS_WRITE_4(sc, CAS_MAC_ATTEMPT_LIMIT, 0x10);
1349		CAS_WRITE_4(sc, CAS_MAC_CTRL_TYPE, 0x8808);
1350
1351		/* random number seed */
1352		CAS_WRITE_4(sc, CAS_MAC_RANDOM_SEED,
1353		    ((laddr[5] << 8) | laddr[4]) & 0x3ff);
1354
1355		/* secondary MAC addresses: 0:0:0:0:0:0 */
1356		for (i = CAS_MAC_ADDR3; i <= CAS_MAC_ADDR41;
1357		    i += CAS_MAC_ADDR4 - CAS_MAC_ADDR3)
1358			CAS_WRITE_4(sc, i, 0);
1359
1360		/* MAC control address: 01:80:c2:00:00:01 */
1361		CAS_WRITE_4(sc, CAS_MAC_ADDR42, 0x0001);
1362		CAS_WRITE_4(sc, CAS_MAC_ADDR43, 0xc200);
1363		CAS_WRITE_4(sc, CAS_MAC_ADDR44, 0x0180);
1364
1365		/* MAC filter address: 0:0:0:0:0:0 */
1366		CAS_WRITE_4(sc, CAS_MAC_AFILTER0, 0);
1367		CAS_WRITE_4(sc, CAS_MAC_AFILTER1, 0);
1368		CAS_WRITE_4(sc, CAS_MAC_AFILTER2, 0);
1369		CAS_WRITE_4(sc, CAS_MAC_AFILTER_MASK1_2, 0);
1370		CAS_WRITE_4(sc, CAS_MAC_AFILTER_MASK0, 0);
1371
1372		/* Zero the hash table. */
1373		for (i = CAS_MAC_HASH0; i <= CAS_MAC_HASH15;
1374		    i += CAS_MAC_HASH1 - CAS_MAC_HASH0)
1375			CAS_WRITE_4(sc, i, 0);
1376
1377		sc->sc_flags |= CAS_INITED;
1378	}
1379
1380	/* Counters need to be zeroed. */
1381	CAS_WRITE_4(sc, CAS_MAC_NORM_COLL_CNT, 0);
1382	CAS_WRITE_4(sc, CAS_MAC_FIRST_COLL_CNT, 0);
1383	CAS_WRITE_4(sc, CAS_MAC_EXCESS_COLL_CNT, 0);
1384	CAS_WRITE_4(sc, CAS_MAC_LATE_COLL_CNT, 0);
1385	CAS_WRITE_4(sc, CAS_MAC_DEFER_TMR_CNT, 0);
1386	CAS_WRITE_4(sc, CAS_MAC_PEAK_ATTEMPTS, 0);
1387	CAS_WRITE_4(sc, CAS_MAC_RX_FRAME_COUNT, 0);
1388	CAS_WRITE_4(sc, CAS_MAC_RX_LEN_ERR_CNT, 0);
1389	CAS_WRITE_4(sc, CAS_MAC_RX_ALIGN_ERR, 0);
1390	CAS_WRITE_4(sc, CAS_MAC_RX_CRC_ERR_CNT, 0);
1391	CAS_WRITE_4(sc, CAS_MAC_RX_CODE_VIOL, 0);
1392
1393	/* Set XOFF PAUSE time. */
1394	CAS_WRITE_4(sc, CAS_MAC_SPC, 0x1BF0 << CAS_MAC_SPC_TIME_SHFT);
1395
1396	/* Set the station address. */
1397	CAS_WRITE_4(sc, CAS_MAC_ADDR0, (laddr[4] << 8) | laddr[5]);
1398	CAS_WRITE_4(sc, CAS_MAC_ADDR1, (laddr[2] << 8) | laddr[3]);
1399	CAS_WRITE_4(sc, CAS_MAC_ADDR2, (laddr[0] << 8) | laddr[1]);
1400
1401	/* Enable MII outputs. */
1402	CAS_WRITE_4(sc, CAS_MAC_XIF_CONF, CAS_MAC_XIF_CONF_TX_OE);
1403}
1404
1405static void
1406cas_tx_task(void *arg, int pending __unused)
1407{
1408	if_t ifp;
1409
1410	ifp = (if_t)arg;
1411	cas_start(ifp);
1412}
1413
1414static inline void
1415cas_txkick(struct cas_softc *sc)
1416{
1417
1418	/*
1419	 * Update the TX kick register.  This register has to point to the
1420	 * descriptor after the last valid one and for optimum performance
1421	 * should be incremented in multiples of 4 (the DMA engine fetches/
1422	 * updates descriptors in batches of 4).
1423	 */
1424#ifdef CAS_DEBUG
1425	CTR3(KTR_CAS, "%s: %s: kicking TX %d",
1426	    device_get_name(sc->sc_dev), __func__, sc->sc_txnext);
1427#endif
1428	CAS_CDSYNC(sc, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1429	CAS_WRITE_4(sc, CAS_TX_KICK3, sc->sc_txnext);
1430}
1431
1432static void
1433cas_start(if_t ifp)
1434{
1435	struct cas_softc *sc = if_getsoftc(ifp);
1436	struct mbuf *m;
1437	int kicked, ntx;
1438
1439	CAS_LOCK(sc);
1440
1441	if ((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
1442	    IFF_DRV_RUNNING || (sc->sc_flags & CAS_LINK) == 0) {
1443		CAS_UNLOCK(sc);
1444		return;
1445	}
1446
1447	if (sc->sc_txfree < CAS_MAXTXFREE / 4)
1448		cas_tint(sc);
1449
1450#ifdef CAS_DEBUG
1451	CTR4(KTR_CAS, "%s: %s: txfree %d, txnext %d",
1452	    device_get_name(sc->sc_dev), __func__, sc->sc_txfree,
1453	    sc->sc_txnext);
1454#endif
1455	ntx = 0;
1456	kicked = 0;
1457	for (; !if_sendq_empty(ifp) && sc->sc_txfree > 1;) {
1458		m = if_dequeue(ifp);
1459		if (m == NULL)
1460			break;
1461		if (cas_load_txmbuf(sc, &m) != 0) {
1462			if (m == NULL)
1463				break;
1464			if_setdrvflagbits(ifp, IFF_DRV_OACTIVE, 0);
1465			if_sendq_prepend(ifp, m);
1466			break;
1467		}
1468		if ((sc->sc_txnext % 4) == 0) {
1469			cas_txkick(sc);
1470			kicked = 1;
1471		} else
1472			kicked = 0;
1473		ntx++;
1474		BPF_MTAP(ifp, m);
1475	}
1476
1477	if (ntx > 0) {
1478		if (kicked == 0)
1479			cas_txkick(sc);
1480#ifdef CAS_DEBUG
1481		CTR2(KTR_CAS, "%s: packets enqueued, OWN on %d",
1482		    device_get_name(sc->sc_dev), sc->sc_txnext);
1483#endif
1484
1485		/* Set a watchdog timer in case the chip flakes out. */
1486		sc->sc_wdog_timer = 5;
1487#ifdef CAS_DEBUG
1488		CTR3(KTR_CAS, "%s: %s: watchdog %d",
1489		    device_get_name(sc->sc_dev), __func__,
1490		    sc->sc_wdog_timer);
1491#endif
1492	}
1493
1494	CAS_UNLOCK(sc);
1495}
1496
1497static void
1498cas_tint(struct cas_softc *sc)
1499{
1500	if_t ifp = sc->sc_ifp;
1501	struct cas_txsoft *txs;
1502	int progress;
1503	uint32_t txlast;
1504#ifdef CAS_DEBUG
1505	int i;
1506
1507	CAS_LOCK_ASSERT(sc, MA_OWNED);
1508
1509	CTR2(KTR_CAS, "%s: %s", device_get_name(sc->sc_dev), __func__);
1510#endif
1511
1512	/*
1513	 * Go through our TX list and free mbufs for those
1514	 * frames that have been transmitted.
1515	 */
1516	progress = 0;
1517	CAS_CDSYNC(sc, BUS_DMASYNC_POSTREAD);
1518	while ((txs = STAILQ_FIRST(&sc->sc_txdirtyq)) != NULL) {
1519#ifdef CAS_DEBUG
1520		if ((if_getflags(ifp) & IFF_DEBUG) != 0) {
1521			printf("    txsoft %p transmit chain:\n", txs);
1522			for (i = txs->txs_firstdesc;; i = CAS_NEXTTX(i)) {
1523				printf("descriptor %d: ", i);
1524				printf("cd_flags: 0x%016llx\t",
1525				    (long long)le64toh(
1526				    sc->sc_txdescs[i].cd_flags));
1527				printf("cd_buf_ptr: 0x%016llx\n",
1528				    (long long)le64toh(
1529				    sc->sc_txdescs[i].cd_buf_ptr));
1530				if (i == txs->txs_lastdesc)
1531					break;
1532			}
1533		}
1534#endif
1535
1536		/*
1537		 * In theory, we could harvest some descriptors before
1538		 * the ring is empty, but that's a bit complicated.
1539		 *
1540		 * CAS_TX_COMPn points to the last descriptor
1541		 * processed + 1.
1542		 */
1543		txlast = CAS_READ_4(sc, CAS_TX_COMP3);
1544#ifdef CAS_DEBUG
1545		CTR4(KTR_CAS, "%s: txs->txs_firstdesc = %d, "
1546		    "txs->txs_lastdesc = %d, txlast = %d",
1547		    __func__, txs->txs_firstdesc, txs->txs_lastdesc, txlast);
1548#endif
1549		if (txs->txs_firstdesc <= txs->txs_lastdesc) {
1550			if ((txlast >= txs->txs_firstdesc) &&
1551			    (txlast <= txs->txs_lastdesc))
1552				break;
1553		} else {
1554			/* Ick -- this command wraps. */
1555			if ((txlast >= txs->txs_firstdesc) ||
1556			    (txlast <= txs->txs_lastdesc))
1557				break;
1558		}
1559
1560#ifdef CAS_DEBUG
1561		CTR1(KTR_CAS, "%s: releasing a descriptor", __func__);
1562#endif
1563		STAILQ_REMOVE_HEAD(&sc->sc_txdirtyq, txs_q);
1564
1565		sc->sc_txfree += txs->txs_ndescs;
1566
1567		bus_dmamap_sync(sc->sc_tdmatag, txs->txs_dmamap,
1568		    BUS_DMASYNC_POSTWRITE);
1569		bus_dmamap_unload(sc->sc_tdmatag, txs->txs_dmamap);
1570		if (txs->txs_mbuf != NULL) {
1571			m_freem(txs->txs_mbuf);
1572			txs->txs_mbuf = NULL;
1573		}
1574
1575		STAILQ_INSERT_TAIL(&sc->sc_txfreeq, txs, txs_q);
1576
1577		if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
1578		progress = 1;
1579	}
1580
1581#ifdef CAS_DEBUG
1582	CTR5(KTR_CAS, "%s: CAS_TX_SM1 %x CAS_TX_SM2 %x CAS_TX_DESC_BASE %llx "
1583	    "CAS_TX_COMP3 %x",
1584	    __func__, CAS_READ_4(sc, CAS_TX_SM1), CAS_READ_4(sc, CAS_TX_SM2),
1585	    ((long long)CAS_READ_4(sc, CAS_TX_DESC3_BASE_HI) << 32) |
1586	    CAS_READ_4(sc, CAS_TX_DESC3_BASE_LO),
1587	    CAS_READ_4(sc, CAS_TX_COMP3));
1588#endif
1589
1590	if (progress) {
1591		/* We freed some descriptors, so reset IFF_DRV_OACTIVE. */
1592		if_setdrvflagbits(ifp, 0, IFF_DRV_OACTIVE);
1593		if (STAILQ_EMPTY(&sc->sc_txdirtyq))
1594			sc->sc_wdog_timer = 0;
1595	}
1596
1597#ifdef CAS_DEBUG
1598	CTR3(KTR_CAS, "%s: %s: watchdog %d",
1599	    device_get_name(sc->sc_dev), __func__, sc->sc_wdog_timer);
1600#endif
1601}
1602
1603static void
1604cas_rint_timeout(void *arg)
1605{
1606	struct epoch_tracker et;
1607	struct cas_softc *sc = arg;
1608
1609	CAS_LOCK_ASSERT(sc, MA_OWNED);
1610
1611	NET_EPOCH_ENTER(et);
1612	cas_rint(sc);
1613	NET_EPOCH_EXIT(et);
1614}
1615
1616static void
1617cas_rint(struct cas_softc *sc)
1618{
1619	struct cas_rxdsoft *rxds, *rxds2;
1620	if_t ifp = sc->sc_ifp;
1621	struct mbuf *m, *m2;
1622	uint64_t word1, word2, word3 __unused, word4;
1623	uint32_t rxhead;
1624	u_int idx, idx2, len, off, skip;
1625
1626	CAS_LOCK_ASSERT(sc, MA_OWNED);
1627
1628	callout_stop(&sc->sc_rx_ch);
1629
1630#ifdef CAS_DEBUG
1631	CTR2(KTR_CAS, "%s: %s", device_get_name(sc->sc_dev), __func__);
1632#endif
1633
1634#define	PRINTWORD(n, delimiter)						\
1635	printf("word ## n: 0x%016llx%c", (long long)word ## n, delimiter)
1636
1637#define	SKIPASSERT(n)							\
1638	KASSERT(sc->sc_rxcomps[sc->sc_rxcptr].crc_word ## n == 0,	\
1639	    ("%s: word ## n not 0", __func__))
1640
1641#define	WORDTOH(n)							\
1642	word ## n = le64toh(sc->sc_rxcomps[sc->sc_rxcptr].crc_word ## n)
1643
1644	/*
1645	 * Read the completion head register once.  This limits
1646	 * how long the following loop can execute.
1647	 */
1648	rxhead = CAS_READ_4(sc, CAS_RX_COMP_HEAD);
1649#ifdef CAS_DEBUG
1650	CTR4(KTR_CAS, "%s: sc->sc_rxcptr %d, sc->sc_rxdptr %d, head %d",
1651	    __func__, sc->sc_rxcptr, sc->sc_rxdptr, rxhead);
1652#endif
1653	skip = 0;
1654	CAS_CDSYNC(sc, BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
1655	for (; sc->sc_rxcptr != rxhead;
1656	    sc->sc_rxcptr = CAS_NEXTRXCOMP(sc->sc_rxcptr)) {
1657		if (skip != 0) {
1658			SKIPASSERT(1);
1659			SKIPASSERT(2);
1660			SKIPASSERT(3);
1661
1662			--skip;
1663			goto skip;
1664		}
1665
1666		WORDTOH(1);
1667		WORDTOH(2);
1668		WORDTOH(3);
1669		WORDTOH(4);
1670
1671#ifdef CAS_DEBUG
1672		if ((if_getflags(ifp) & IFF_DEBUG) != 0) {
1673			printf("    completion %d: ", sc->sc_rxcptr);
1674			PRINTWORD(1, '\t');
1675			PRINTWORD(2, '\t');
1676			PRINTWORD(3, '\t');
1677			PRINTWORD(4, '\n');
1678		}
1679#endif
1680
1681		if (__predict_false(
1682		    (word1 & CAS_RC1_TYPE_MASK) == CAS_RC1_TYPE_HW ||
1683		    (word4 & CAS_RC4_ZERO) != 0)) {
1684			/*
1685			 * The descriptor is still marked as owned, although
1686			 * it is supposed to have completed.  This has been
1687			 * observed on some machines.  Just exiting here
1688			 * might leave the packet sitting around until another
1689			 * one arrives to trigger a new interrupt, which is
1690			 * generally undesirable, so set up a timeout.
1691			 */
1692			callout_reset(&sc->sc_rx_ch, CAS_RXOWN_TICKS,
1693			    cas_rint_timeout, sc);
1694			break;
1695		}
1696
1697		if (__predict_false(
1698		    (word4 & (CAS_RC4_BAD | CAS_RC4_LEN_MMATCH)) != 0)) {
1699			if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
1700			device_printf(sc->sc_dev,
1701			    "receive error: CRC error\n");
1702			continue;
1703		}
1704
1705		KASSERT(CAS_GET(word1, CAS_RC1_DATA_SIZE) == 0 ||
1706		    CAS_GET(word2, CAS_RC2_HDR_SIZE) == 0,
1707		    ("%s: data and header present", __func__));
1708		KASSERT((word1 & CAS_RC1_SPLIT_PKT) == 0 ||
1709		    CAS_GET(word2, CAS_RC2_HDR_SIZE) == 0,
1710		    ("%s: split and header present", __func__));
1711		KASSERT(CAS_GET(word1, CAS_RC1_DATA_SIZE) == 0 ||
1712		    (word1 & CAS_RC1_RELEASE_HDR) == 0,
1713		    ("%s: data present but header release", __func__));
1714		KASSERT(CAS_GET(word2, CAS_RC2_HDR_SIZE) == 0 ||
1715		    (word1 & CAS_RC1_RELEASE_DATA) == 0,
1716		    ("%s: header present but data release", __func__));
1717
1718		if ((len = CAS_GET(word2, CAS_RC2_HDR_SIZE)) != 0) {
1719			idx = CAS_GET(word2, CAS_RC2_HDR_INDEX);
1720			off = CAS_GET(word2, CAS_RC2_HDR_OFF);
1721#ifdef CAS_DEBUG
1722			CTR4(KTR_CAS, "%s: hdr at idx %d, off %d, len %d",
1723			    __func__, idx, off, len);
1724#endif
1725			rxds = &sc->sc_rxdsoft[idx];
1726			MGETHDR(m, M_NOWAIT, MT_DATA);
1727			if (m != NULL) {
1728				refcount_acquire(&rxds->rxds_refcount);
1729				bus_dmamap_sync(sc->sc_rdmatag,
1730				    rxds->rxds_dmamap, BUS_DMASYNC_POSTREAD);
1731				m_extadd(m, (char *)rxds->rxds_buf +
1732				    off * 256 + ETHER_ALIGN, len, cas_free,
1733				    sc, (void *)(uintptr_t)idx,
1734				    M_RDONLY, EXT_NET_DRV);
1735				if ((m->m_flags & M_EXT) == 0) {
1736					m_freem(m);
1737					m = NULL;
1738				}
1739			}
1740			if (m != NULL) {
1741				m->m_pkthdr.rcvif = ifp;
1742				m->m_pkthdr.len = m->m_len = len;
1743				if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
1744				if ((if_getcapenable(ifp) & IFCAP_RXCSUM) != 0)
1745					cas_rxcksum(m, CAS_GET(word4,
1746					    CAS_RC4_TCP_CSUM));
1747				/* Pass it on. */
1748				CAS_UNLOCK(sc);
1749				if_input(ifp, m);
1750				CAS_LOCK(sc);
1751			} else
1752				if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1753
1754			if ((word1 & CAS_RC1_RELEASE_HDR) != 0 &&
1755			    refcount_release(&rxds->rxds_refcount) != 0)
1756				cas_add_rxdesc(sc, idx);
1757		} else if ((len = CAS_GET(word1, CAS_RC1_DATA_SIZE)) != 0) {
1758			idx = CAS_GET(word1, CAS_RC1_DATA_INDEX);
1759			off = CAS_GET(word1, CAS_RC1_DATA_OFF);
1760#ifdef CAS_DEBUG
1761			CTR4(KTR_CAS, "%s: data at idx %d, off %d, len %d",
1762			    __func__, idx, off, len);
1763#endif
1764			rxds = &sc->sc_rxdsoft[idx];
1765			MGETHDR(m, M_NOWAIT, MT_DATA);
1766			if (m != NULL) {
1767				refcount_acquire(&rxds->rxds_refcount);
1768				off += ETHER_ALIGN;
1769				m->m_len = min(CAS_PAGE_SIZE - off, len);
1770				bus_dmamap_sync(sc->sc_rdmatag,
1771				    rxds->rxds_dmamap, BUS_DMASYNC_POSTREAD);
1772				m_extadd(m, (char *)rxds->rxds_buf + off,
1773				    m->m_len, cas_free, sc,
1774				    (void *)(uintptr_t)idx, M_RDONLY,
1775				    EXT_NET_DRV);
1776				if ((m->m_flags & M_EXT) == 0) {
1777					m_freem(m);
1778					m = NULL;
1779				}
1780			}
1781			idx2 = 0;
1782			m2 = NULL;
1783			rxds2 = NULL;
1784			if ((word1 & CAS_RC1_SPLIT_PKT) != 0) {
1785				KASSERT((word1 & CAS_RC1_RELEASE_NEXT) != 0,
1786				    ("%s: split but no release next",
1787				    __func__));
1788
1789				idx2 = CAS_GET(word2, CAS_RC2_NEXT_INDEX);
1790#ifdef CAS_DEBUG
1791				CTR2(KTR_CAS, "%s: split at idx %d",
1792				    __func__, idx2);
1793#endif
1794				rxds2 = &sc->sc_rxdsoft[idx2];
1795				if (m != NULL) {
1796					MGET(m2, M_NOWAIT, MT_DATA);
1797					if (m2 != NULL) {
1798						refcount_acquire(
1799						    &rxds2->rxds_refcount);
1800						m2->m_len = len - m->m_len;
1801						bus_dmamap_sync(
1802						    sc->sc_rdmatag,
1803						    rxds2->rxds_dmamap,
1804						    BUS_DMASYNC_POSTREAD);
1805						m_extadd(m2,
1806						    (char *)rxds2->rxds_buf,
1807						    m2->m_len, cas_free, sc,
1808						    (void *)(uintptr_t)idx2,
1809						    M_RDONLY, EXT_NET_DRV);
1810						if ((m2->m_flags & M_EXT) ==
1811						    0) {
1812							m_freem(m2);
1813							m2 = NULL;
1814						}
1815					}
1816				}
1817				if (m2 != NULL)
1818					m->m_next = m2;
1819				else if (m != NULL) {
1820					m_freem(m);
1821					m = NULL;
1822				}
1823			}
1824			if (m != NULL) {
1825				m->m_pkthdr.rcvif = ifp;
1826				m->m_pkthdr.len = len;
1827				if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
1828				if ((if_getcapenable(ifp) & IFCAP_RXCSUM) != 0)
1829					cas_rxcksum(m, CAS_GET(word4,
1830					    CAS_RC4_TCP_CSUM));
1831				/* Pass it on. */
1832				CAS_UNLOCK(sc);
1833				if_input(ifp, m);
1834				CAS_LOCK(sc);
1835			} else
1836				if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1837
1838			if ((word1 & CAS_RC1_RELEASE_DATA) != 0 &&
1839			    refcount_release(&rxds->rxds_refcount) != 0)
1840				cas_add_rxdesc(sc, idx);
1841			if ((word1 & CAS_RC1_SPLIT_PKT) != 0 &&
1842			    refcount_release(&rxds2->rxds_refcount) != 0)
1843				cas_add_rxdesc(sc, idx2);
1844		}
1845
1846		skip = CAS_GET(word1, CAS_RC1_SKIP);
1847
1848 skip:
1849		cas_rxcompinit(&sc->sc_rxcomps[sc->sc_rxcptr]);
1850		if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0)
1851			break;
1852	}
1853	CAS_CDSYNC(sc, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1854	CAS_WRITE_4(sc, CAS_RX_COMP_TAIL, sc->sc_rxcptr);
1855
1856#undef PRINTWORD
1857#undef SKIPASSERT
1858#undef WORDTOH
1859
1860#ifdef CAS_DEBUG
1861	CTR4(KTR_CAS, "%s: done sc->sc_rxcptr %d, sc->sc_rxdptr %d, head %d",
1862	    __func__, sc->sc_rxcptr, sc->sc_rxdptr,
1863	    CAS_READ_4(sc, CAS_RX_COMP_HEAD));
1864#endif
1865}
1866
1867static void
1868cas_free(struct mbuf *m)
1869{
1870	struct cas_rxdsoft *rxds;
1871	struct cas_softc *sc;
1872	u_int idx, locked;
1873
1874	sc = m->m_ext.ext_arg1;
1875	idx = (uintptr_t)m->m_ext.ext_arg2;
1876	rxds = &sc->sc_rxdsoft[idx];
1877	if (refcount_release(&rxds->rxds_refcount) == 0)
1878		return;
1879
1880	/*
1881	 * NB: this function can be called via m_freem(9) within
1882	 * this driver!
1883	 */
1884	if ((locked = CAS_LOCK_OWNED(sc)) == 0)
1885		CAS_LOCK(sc);
1886	cas_add_rxdesc(sc, idx);
1887	if (locked == 0)
1888		CAS_UNLOCK(sc);
1889}
1890
1891static inline void
1892cas_add_rxdesc(struct cas_softc *sc, u_int idx)
1893{
1894
1895	CAS_LOCK_ASSERT(sc, MA_OWNED);
1896
1897	bus_dmamap_sync(sc->sc_rdmatag, sc->sc_rxdsoft[idx].rxds_dmamap,
1898	    BUS_DMASYNC_PREREAD);
1899	CAS_UPDATE_RXDESC(sc, sc->sc_rxdptr, idx);
1900	sc->sc_rxdptr = CAS_NEXTRXDESC(sc->sc_rxdptr);
1901
1902	/*
1903	 * Update the RX kick register.  This register has to point to the
1904	 * descriptor after the last valid one (before the current batch)
1905	 * and for optimum performance should be incremented in multiples
1906	 * of 4 (the DMA engine fetches/updates descriptors in batches of 4).
1907	 */
1908	if ((sc->sc_rxdptr % 4) == 0) {
1909		CAS_CDSYNC(sc, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1910		CAS_WRITE_4(sc, CAS_RX_KICK,
1911		    (sc->sc_rxdptr + CAS_NRXDESC - 4) & CAS_NRXDESC_MASK);
1912	}
1913}
1914
1915static void
1916cas_eint(struct cas_softc *sc, u_int status)
1917{
1918	if_t ifp = sc->sc_ifp;
1919
1920	CAS_LOCK_ASSERT(sc, MA_OWNED);
1921
1922	if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
1923
1924	device_printf(sc->sc_dev, "%s: status 0x%x", __func__, status);
1925	if ((status & CAS_INTR_PCI_ERROR_INT) != 0) {
1926		status = CAS_READ_4(sc, CAS_ERROR_STATUS);
1927		printf(", PCI bus error 0x%x", status);
1928		if ((status & CAS_ERROR_OTHER) != 0) {
1929			status = pci_read_config(sc->sc_dev, PCIR_STATUS, 2);
1930			printf(", PCI status 0x%x", status);
1931			pci_write_config(sc->sc_dev, PCIR_STATUS, status, 2);
1932		}
1933	}
1934	printf("\n");
1935
1936	if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
1937	cas_init_locked(sc);
1938	if (!if_sendq_empty(ifp))
1939		taskqueue_enqueue(sc->sc_tq, &sc->sc_tx_task);
1940}
1941
1942static int
1943cas_intr(void *v)
1944{
1945	struct cas_softc *sc = v;
1946
1947	if (__predict_false((CAS_READ_4(sc, CAS_STATUS_ALIAS) &
1948	    CAS_INTR_SUMMARY) == 0))
1949		return (FILTER_STRAY);
1950
1951	/* Disable interrupts. */
1952	CAS_WRITE_4(sc, CAS_INTMASK, 0xffffffff);
1953	taskqueue_enqueue(sc->sc_tq, &sc->sc_intr_task);
1954
1955	return (FILTER_HANDLED);
1956}
1957
1958static void
1959cas_intr_task(void *arg, int pending __unused)
1960{
1961	struct cas_softc *sc = arg;
1962	if_t ifp = sc->sc_ifp;
1963	uint32_t status, status2;
1964
1965	CAS_LOCK_ASSERT(sc, MA_NOTOWNED);
1966
1967	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0)
1968		return;
1969
1970	status = CAS_READ_4(sc, CAS_STATUS);
1971	if (__predict_false((status & CAS_INTR_SUMMARY) == 0))
1972		goto done;
1973
1974	CAS_LOCK(sc);
1975#ifdef CAS_DEBUG
1976	CTR4(KTR_CAS, "%s: %s: cplt %x, status %x",
1977	    device_get_name(sc->sc_dev), __func__,
1978	    (status >> CAS_STATUS_TX_COMP3_SHFT), (u_int)status);
1979
1980	/*
1981	 * PCS interrupts must be cleared, otherwise no traffic is passed!
1982	 */
1983	if ((status & CAS_INTR_PCS_INT) != 0) {
1984		status2 =
1985		    CAS_READ_4(sc, CAS_PCS_INTR_STATUS) |
1986		    CAS_READ_4(sc, CAS_PCS_INTR_STATUS);
1987		if ((status2 & CAS_PCS_INTR_LINK) != 0)
1988			device_printf(sc->sc_dev,
1989			    "%s: PCS link status changed\n", __func__);
1990	}
1991	if ((status & CAS_MAC_CTRL_STATUS) != 0) {
1992		status2 = CAS_READ_4(sc, CAS_MAC_CTRL_STATUS);
1993		if ((status2 & CAS_MAC_CTRL_PAUSE) != 0)
1994			device_printf(sc->sc_dev,
1995			    "%s: PAUSE received (PAUSE time %d slots)\n",
1996			    __func__,
1997			    (status2 & CAS_MAC_CTRL_STATUS_PT_MASK) >>
1998			    CAS_MAC_CTRL_STATUS_PT_SHFT);
1999		if ((status2 & CAS_MAC_CTRL_PAUSE) != 0)
2000			device_printf(sc->sc_dev,
2001			    "%s: transited to PAUSE state\n", __func__);
2002		if ((status2 & CAS_MAC_CTRL_NON_PAUSE) != 0)
2003			device_printf(sc->sc_dev,
2004			    "%s: transited to non-PAUSE state\n", __func__);
2005	}
2006	if ((status & CAS_INTR_MIF) != 0)
2007		device_printf(sc->sc_dev, "%s: MIF interrupt\n", __func__);
2008#endif
2009
2010	if (__predict_false((status &
2011	    (CAS_INTR_TX_TAG_ERR | CAS_INTR_RX_TAG_ERR |
2012	    CAS_INTR_RX_LEN_MMATCH | CAS_INTR_PCI_ERROR_INT)) != 0)) {
2013		cas_eint(sc, status);
2014		CAS_UNLOCK(sc);
2015		return;
2016	}
2017
2018	if (__predict_false(status & CAS_INTR_TX_MAC_INT)) {
2019		status2 = CAS_READ_4(sc, CAS_MAC_TX_STATUS);
2020		if ((status2 &
2021		    (CAS_MAC_TX_UNDERRUN | CAS_MAC_TX_MAX_PKT_ERR)) != 0)
2022			if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
2023		else if ((status2 & ~CAS_MAC_TX_FRAME_XMTD) != 0)
2024			device_printf(sc->sc_dev,
2025			    "MAC TX fault, status %x\n", status2);
2026	}
2027
2028	if (__predict_false(status & CAS_INTR_RX_MAC_INT)) {
2029		status2 = CAS_READ_4(sc, CAS_MAC_RX_STATUS);
2030		if ((status2 & CAS_MAC_RX_OVERFLOW) != 0)
2031			if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
2032		else if ((status2 & ~CAS_MAC_RX_FRAME_RCVD) != 0)
2033			device_printf(sc->sc_dev,
2034			    "MAC RX fault, status %x\n", status2);
2035	}
2036
2037	if ((status &
2038	    (CAS_INTR_RX_DONE | CAS_INTR_RX_BUF_NA | CAS_INTR_RX_COMP_FULL |
2039	    CAS_INTR_RX_BUF_AEMPTY | CAS_INTR_RX_COMP_AFULL)) != 0) {
2040		cas_rint(sc);
2041#ifdef CAS_DEBUG
2042		if (__predict_false((status &
2043		    (CAS_INTR_RX_BUF_NA | CAS_INTR_RX_COMP_FULL |
2044		    CAS_INTR_RX_BUF_AEMPTY | CAS_INTR_RX_COMP_AFULL)) != 0))
2045			device_printf(sc->sc_dev,
2046			    "RX fault, status %x\n", status);
2047#endif
2048	}
2049
2050	if ((status &
2051	    (CAS_INTR_TX_INT_ME | CAS_INTR_TX_ALL | CAS_INTR_TX_DONE)) != 0)
2052		cas_tint(sc);
2053
2054	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0) {
2055		CAS_UNLOCK(sc);
2056		return;
2057	} else if (!if_sendq_empty(ifp))
2058		taskqueue_enqueue(sc->sc_tq, &sc->sc_tx_task);
2059	CAS_UNLOCK(sc);
2060
2061	status = CAS_READ_4(sc, CAS_STATUS_ALIAS);
2062	if (__predict_false((status & CAS_INTR_SUMMARY) != 0)) {
2063		taskqueue_enqueue(sc->sc_tq, &sc->sc_intr_task);
2064		return;
2065	}
2066
2067 done:
2068	/* Re-enable interrupts. */
2069	CAS_WRITE_4(sc, CAS_INTMASK,
2070	    ~(CAS_INTR_TX_INT_ME | CAS_INTR_TX_TAG_ERR |
2071	    CAS_INTR_RX_DONE | CAS_INTR_RX_BUF_NA | CAS_INTR_RX_TAG_ERR |
2072	    CAS_INTR_RX_COMP_FULL | CAS_INTR_RX_BUF_AEMPTY |
2073	    CAS_INTR_RX_COMP_AFULL | CAS_INTR_RX_LEN_MMATCH |
2074	    CAS_INTR_PCI_ERROR_INT
2075#ifdef CAS_DEBUG
2076	    | CAS_INTR_PCS_INT | CAS_INTR_MIF
2077#endif
2078	));
2079}
2080
2081static void
2082cas_watchdog(struct cas_softc *sc)
2083{
2084	if_t ifp = sc->sc_ifp;
2085
2086	CAS_LOCK_ASSERT(sc, MA_OWNED);
2087
2088#ifdef CAS_DEBUG
2089	CTR4(KTR_CAS,
2090	    "%s: CAS_RX_CONF %x CAS_MAC_RX_STATUS %x CAS_MAC_RX_CONF %x",
2091	    __func__, CAS_READ_4(sc, CAS_RX_CONF),
2092	    CAS_READ_4(sc, CAS_MAC_RX_STATUS),
2093	    CAS_READ_4(sc, CAS_MAC_RX_CONF));
2094	CTR4(KTR_CAS,
2095	    "%s: CAS_TX_CONF %x CAS_MAC_TX_STATUS %x CAS_MAC_TX_CONF %x",
2096	    __func__, CAS_READ_4(sc, CAS_TX_CONF),
2097	    CAS_READ_4(sc, CAS_MAC_TX_STATUS),
2098	    CAS_READ_4(sc, CAS_MAC_TX_CONF));
2099#endif
2100
2101	if (sc->sc_wdog_timer == 0 || --sc->sc_wdog_timer != 0)
2102		return;
2103
2104	if ((sc->sc_flags & CAS_LINK) != 0)
2105		device_printf(sc->sc_dev, "device timeout\n");
2106	else if (bootverbose)
2107		device_printf(sc->sc_dev, "device timeout (no link)\n");
2108	if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
2109
2110	/* Try to get more packets going. */
2111	if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
2112	cas_init_locked(sc);
2113	if (!if_sendq_empty(ifp))
2114		taskqueue_enqueue(sc->sc_tq, &sc->sc_tx_task);
2115}
2116
2117static void
2118cas_mifinit(struct cas_softc *sc)
2119{
2120
2121	/* Configure the MIF in frame mode. */
2122	CAS_WRITE_4(sc, CAS_MIF_CONF,
2123	    CAS_READ_4(sc, CAS_MIF_CONF) & ~CAS_MIF_CONF_BB_MODE);
2124	CAS_BARRIER(sc, CAS_MIF_CONF, 4,
2125	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2126}
2127
2128/*
2129 * MII interface
2130 *
2131 * The MII interface supports at least three different operating modes:
2132 *
2133 * Bitbang mode is implemented using data, clock and output enable registers.
2134 *
2135 * Frame mode is implemented by loading a complete frame into the frame
2136 * register and polling the valid bit for completion.
2137 *
2138 * Polling mode uses the frame register but completion is indicated by
2139 * an interrupt.
2140 *
2141 */
2142static int
2143cas_mii_readreg(device_t dev, int phy, int reg)
2144{
2145	struct cas_softc *sc;
2146	int n;
2147	uint32_t v;
2148
2149#ifdef CAS_DEBUG_PHY
2150	printf("%s: phy %d reg %d\n", __func__, phy, reg);
2151#endif
2152
2153	sc = device_get_softc(dev);
2154	if ((sc->sc_flags & CAS_SERDES) != 0) {
2155		switch (reg) {
2156		case MII_BMCR:
2157			reg = CAS_PCS_CTRL;
2158			break;
2159		case MII_BMSR:
2160			reg = CAS_PCS_STATUS;
2161			break;
2162		case MII_PHYIDR1:
2163		case MII_PHYIDR2:
2164			return (0);
2165		case MII_ANAR:
2166			reg = CAS_PCS_ANAR;
2167			break;
2168		case MII_ANLPAR:
2169			reg = CAS_PCS_ANLPAR;
2170			break;
2171		case MII_EXTSR:
2172			return (EXTSR_1000XFDX | EXTSR_1000XHDX);
2173		default:
2174			device_printf(sc->sc_dev,
2175			    "%s: unhandled register %d\n", __func__, reg);
2176			return (0);
2177		}
2178		return (CAS_READ_4(sc, reg));
2179	}
2180
2181	/* Construct the frame command. */
2182	v = CAS_MIF_FRAME_READ |
2183	    (phy << CAS_MIF_FRAME_PHY_SHFT) |
2184	    (reg << CAS_MIF_FRAME_REG_SHFT);
2185
2186	CAS_WRITE_4(sc, CAS_MIF_FRAME, v);
2187	CAS_BARRIER(sc, CAS_MIF_FRAME, 4,
2188	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2189	for (n = 0; n < 100; n++) {
2190		DELAY(1);
2191		v = CAS_READ_4(sc, CAS_MIF_FRAME);
2192		if (v & CAS_MIF_FRAME_TA_LSB)
2193			return (v & CAS_MIF_FRAME_DATA);
2194	}
2195
2196	device_printf(sc->sc_dev, "%s: timed out\n", __func__);
2197	return (0);
2198}
2199
2200static int
2201cas_mii_writereg(device_t dev, int phy, int reg, int val)
2202{
2203	struct cas_softc *sc;
2204	int n;
2205	uint32_t v;
2206
2207#ifdef CAS_DEBUG_PHY
2208	printf("%s: phy %d reg %d val %x\n", phy, reg, val, __func__);
2209#endif
2210
2211	sc = device_get_softc(dev);
2212	if ((sc->sc_flags & CAS_SERDES) != 0) {
2213		switch (reg) {
2214		case MII_BMSR:
2215			reg = CAS_PCS_STATUS;
2216			break;
2217		case MII_BMCR:
2218			reg = CAS_PCS_CTRL;
2219			if ((val & CAS_PCS_CTRL_RESET) == 0)
2220				break;
2221			CAS_WRITE_4(sc, CAS_PCS_CTRL, val);
2222			CAS_BARRIER(sc, CAS_PCS_CTRL, 4,
2223			    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2224			if (!cas_bitwait(sc, CAS_PCS_CTRL,
2225			    CAS_PCS_CTRL_RESET, 0))
2226				device_printf(sc->sc_dev,
2227				    "cannot reset PCS\n");
2228			/* FALLTHROUGH */
2229		case MII_ANAR:
2230			CAS_WRITE_4(sc, CAS_PCS_CONF, 0);
2231			CAS_BARRIER(sc, CAS_PCS_CONF, 4,
2232			    BUS_SPACE_BARRIER_WRITE);
2233			CAS_WRITE_4(sc, CAS_PCS_ANAR, val);
2234			CAS_BARRIER(sc, CAS_PCS_ANAR, 4,
2235			    BUS_SPACE_BARRIER_WRITE);
2236			CAS_WRITE_4(sc, CAS_PCS_SERDES_CTRL,
2237			    CAS_PCS_SERDES_CTRL_ESD);
2238			CAS_BARRIER(sc, CAS_PCS_CONF, 4,
2239			    BUS_SPACE_BARRIER_WRITE);
2240			CAS_WRITE_4(sc, CAS_PCS_CONF,
2241			    CAS_PCS_CONF_EN);
2242			CAS_BARRIER(sc, CAS_PCS_CONF, 4,
2243			    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2244			return (0);
2245		case MII_ANLPAR:
2246			reg = CAS_PCS_ANLPAR;
2247			break;
2248		default:
2249			device_printf(sc->sc_dev,
2250			    "%s: unhandled register %d\n", __func__, reg);
2251			return (0);
2252		}
2253		CAS_WRITE_4(sc, reg, val);
2254		CAS_BARRIER(sc, reg, 4,
2255		    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2256		return (0);
2257	}
2258
2259	/* Construct the frame command. */
2260	v = CAS_MIF_FRAME_WRITE |
2261	    (phy << CAS_MIF_FRAME_PHY_SHFT) |
2262	    (reg << CAS_MIF_FRAME_REG_SHFT) |
2263	    (val & CAS_MIF_FRAME_DATA);
2264
2265	CAS_WRITE_4(sc, CAS_MIF_FRAME, v);
2266	CAS_BARRIER(sc, CAS_MIF_FRAME, 4,
2267	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2268	for (n = 0; n < 100; n++) {
2269		DELAY(1);
2270		v = CAS_READ_4(sc, CAS_MIF_FRAME);
2271		if (v & CAS_MIF_FRAME_TA_LSB)
2272			return (1);
2273	}
2274
2275	device_printf(sc->sc_dev, "%s: timed out\n", __func__);
2276	return (0);
2277}
2278
2279static void
2280cas_mii_statchg(device_t dev)
2281{
2282	struct cas_softc *sc;
2283	if_t ifp;
2284	int gigabit;
2285	uint32_t rxcfg, txcfg, v;
2286
2287	sc = device_get_softc(dev);
2288	ifp = sc->sc_ifp;
2289
2290	CAS_LOCK_ASSERT(sc, MA_OWNED);
2291
2292#ifdef CAS_DEBUG
2293	if ((if_getflags(ifp) & IFF_DEBUG) != 0)
2294		device_printf(sc->sc_dev, "%s: status changen", __func__);
2295#endif
2296
2297	if ((sc->sc_mii->mii_media_status & IFM_ACTIVE) != 0 &&
2298	    IFM_SUBTYPE(sc->sc_mii->mii_media_active) != IFM_NONE)
2299		sc->sc_flags |= CAS_LINK;
2300	else
2301		sc->sc_flags &= ~CAS_LINK;
2302
2303	switch (IFM_SUBTYPE(sc->sc_mii->mii_media_active)) {
2304	case IFM_1000_SX:
2305	case IFM_1000_LX:
2306	case IFM_1000_CX:
2307	case IFM_1000_T:
2308		gigabit = 1;
2309		break;
2310	default:
2311		gigabit = 0;
2312	}
2313
2314	/*
2315	 * The configuration done here corresponds to the steps F) and
2316	 * G) and as far as enabling of RX and TX MAC goes also step H)
2317	 * of the initialization sequence outlined in section 11.2.1 of
2318	 * the Cassini+ ASIC Specification.
2319	 */
2320
2321	rxcfg = sc->sc_mac_rxcfg;
2322	rxcfg &= ~CAS_MAC_RX_CONF_CARR;
2323	txcfg = CAS_MAC_TX_CONF_EN_IPG0 | CAS_MAC_TX_CONF_NGU |
2324	    CAS_MAC_TX_CONF_NGUL;
2325	if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) & IFM_FDX) != 0)
2326		txcfg |= CAS_MAC_TX_CONF_ICARR | CAS_MAC_TX_CONF_ICOLLIS;
2327	else if (gigabit != 0) {
2328		rxcfg |= CAS_MAC_RX_CONF_CARR;
2329		txcfg |= CAS_MAC_TX_CONF_CARR;
2330	}
2331	(void)cas_disable_tx(sc);
2332	CAS_WRITE_4(sc, CAS_MAC_TX_CONF, txcfg);
2333	(void)cas_disable_rx(sc);
2334	CAS_WRITE_4(sc, CAS_MAC_RX_CONF, rxcfg);
2335
2336	v = CAS_READ_4(sc, CAS_MAC_CTRL_CONF) &
2337	    ~(CAS_MAC_CTRL_CONF_TXP | CAS_MAC_CTRL_CONF_RXP);
2338	if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) &
2339	    IFM_ETH_RXPAUSE) != 0)
2340		v |= CAS_MAC_CTRL_CONF_RXP;
2341	if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) &
2342	    IFM_ETH_TXPAUSE) != 0)
2343		v |= CAS_MAC_CTRL_CONF_TXP;
2344	CAS_WRITE_4(sc, CAS_MAC_CTRL_CONF, v);
2345
2346	/*
2347	 * All supported chips have a bug causing incorrect checksum
2348	 * to be calculated when letting them strip the FCS in half-
2349	 * duplex mode.  In theory we could disable FCS stripping and
2350	 * manually adjust the checksum accordingly.  It seems to make
2351	 * more sense to optimze for the common case and just disable
2352	 * hardware checksumming in half-duplex mode though.
2353	 */
2354	if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) & IFM_FDX) == 0) {
2355		if_setcapenablebit(ifp, 0, IFCAP_HWCSUM);
2356		if_sethwassist(ifp, 0);
2357	} else if ((sc->sc_flags & CAS_NO_CSUM) == 0) {
2358		if_setcapenable(ifp, if_getcapabilities(ifp));
2359		if_sethwassist(ifp, CAS_CSUM_FEATURES);
2360	}
2361
2362	if (sc->sc_variant == CAS_SATURN) {
2363		if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) & IFM_FDX) == 0)
2364			/* silicon bug workaround */
2365			CAS_WRITE_4(sc, CAS_MAC_PREAMBLE_LEN, 0x41);
2366		else
2367			CAS_WRITE_4(sc, CAS_MAC_PREAMBLE_LEN, 0x7);
2368	}
2369
2370	if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) & IFM_FDX) == 0 &&
2371	    gigabit != 0)
2372		CAS_WRITE_4(sc, CAS_MAC_SLOT_TIME,
2373		    CAS_MAC_SLOT_TIME_CARR);
2374	else
2375		CAS_WRITE_4(sc, CAS_MAC_SLOT_TIME,
2376		    CAS_MAC_SLOT_TIME_NORM);
2377
2378	/* XIF Configuration */
2379	v = CAS_MAC_XIF_CONF_TX_OE | CAS_MAC_XIF_CONF_LNKLED;
2380	if ((sc->sc_flags & CAS_SERDES) == 0) {
2381		if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) & IFM_FDX) == 0)
2382			v |= CAS_MAC_XIF_CONF_NOECHO;
2383		v |= CAS_MAC_XIF_CONF_BUF_OE;
2384	}
2385	if (gigabit != 0)
2386		v |= CAS_MAC_XIF_CONF_GMII;
2387	if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) & IFM_FDX) != 0)
2388		v |= CAS_MAC_XIF_CONF_FDXLED;
2389	CAS_WRITE_4(sc, CAS_MAC_XIF_CONF, v);
2390
2391	sc->sc_mac_rxcfg = rxcfg;
2392	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0 &&
2393	    (sc->sc_flags & CAS_LINK) != 0) {
2394		CAS_WRITE_4(sc, CAS_MAC_TX_CONF,
2395		    txcfg | CAS_MAC_TX_CONF_EN);
2396		CAS_WRITE_4(sc, CAS_MAC_RX_CONF,
2397		    rxcfg | CAS_MAC_RX_CONF_EN);
2398	}
2399}
2400
2401static int
2402cas_mediachange(if_t ifp)
2403{
2404	struct cas_softc *sc = if_getsoftc(ifp);
2405	int error;
2406
2407	/* XXX add support for serial media. */
2408
2409	CAS_LOCK(sc);
2410	error = mii_mediachg(sc->sc_mii);
2411	CAS_UNLOCK(sc);
2412	return (error);
2413}
2414
2415static void
2416cas_mediastatus(if_t ifp, struct ifmediareq *ifmr)
2417{
2418	struct cas_softc *sc = if_getsoftc(ifp);
2419
2420	CAS_LOCK(sc);
2421	if ((if_getflags(ifp) & IFF_UP) == 0) {
2422		CAS_UNLOCK(sc);
2423		return;
2424	}
2425
2426	mii_pollstat(sc->sc_mii);
2427	ifmr->ifm_active = sc->sc_mii->mii_media_active;
2428	ifmr->ifm_status = sc->sc_mii->mii_media_status;
2429	CAS_UNLOCK(sc);
2430}
2431
2432static int
2433cas_ioctl(if_t ifp, u_long cmd, caddr_t data)
2434{
2435	struct cas_softc *sc = if_getsoftc(ifp);
2436	struct ifreq *ifr = (struct ifreq *)data;
2437	int error;
2438
2439	error = 0;
2440	switch (cmd) {
2441	case SIOCSIFFLAGS:
2442		CAS_LOCK(sc);
2443		if ((if_getflags(ifp) & IFF_UP) != 0) {
2444			if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0 &&
2445			    ((if_getflags(ifp) ^ sc->sc_ifflags) &
2446			    (IFF_ALLMULTI | IFF_PROMISC)) != 0)
2447				cas_setladrf(sc);
2448			else
2449				cas_init_locked(sc);
2450		} else if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0)
2451			cas_stop(ifp);
2452		sc->sc_ifflags = if_getflags(ifp);
2453		CAS_UNLOCK(sc);
2454		break;
2455	case SIOCSIFCAP:
2456		CAS_LOCK(sc);
2457		if ((sc->sc_flags & CAS_NO_CSUM) != 0) {
2458			error = EINVAL;
2459			CAS_UNLOCK(sc);
2460			break;
2461		}
2462		if_setcapenable(ifp, ifr->ifr_reqcap);
2463		if ((if_getcapenable(ifp) & IFCAP_TXCSUM) != 0)
2464			if_sethwassist(ifp, CAS_CSUM_FEATURES);
2465		else
2466			if_sethwassist(ifp, 0);
2467		CAS_UNLOCK(sc);
2468		break;
2469	case SIOCADDMULTI:
2470	case SIOCDELMULTI:
2471		CAS_LOCK(sc);
2472		if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0)
2473			cas_setladrf(sc);
2474		CAS_UNLOCK(sc);
2475		break;
2476	case SIOCSIFMTU:
2477		if ((ifr->ifr_mtu < ETHERMIN) ||
2478		    (ifr->ifr_mtu > ETHERMTU_JUMBO))
2479			error = EINVAL;
2480		else
2481			if_setmtu(ifp, ifr->ifr_mtu);
2482		break;
2483	case SIOCGIFMEDIA:
2484	case SIOCSIFMEDIA:
2485		error = ifmedia_ioctl(ifp, ifr, &sc->sc_mii->mii_media, cmd);
2486		break;
2487	default:
2488		error = ether_ioctl(ifp, cmd, data);
2489		break;
2490	}
2491
2492	return (error);
2493}
2494
2495static u_int
2496cas_hash_maddr(void *arg, struct sockaddr_dl *sdl, u_int cnt)
2497{
2498	uint32_t crc, *hash = arg;
2499
2500	crc = ether_crc32_le(LLADDR(sdl), ETHER_ADDR_LEN);
2501	/* We just want the 8 most significant bits. */
2502	crc >>= 24;
2503	/* Set the corresponding bit in the filter. */
2504	hash[crc >> 4] |= 1 << (15 - (crc & 15));
2505
2506	return (1);
2507}
2508
2509static void
2510cas_setladrf(struct cas_softc *sc)
2511{
2512	if_t ifp = sc->sc_ifp;
2513	int i;
2514	uint32_t hash[16];
2515	uint32_t v;
2516
2517	CAS_LOCK_ASSERT(sc, MA_OWNED);
2518
2519	/*
2520	 * Turn off the RX MAC and the hash filter as required by the Sun
2521	 * Cassini programming restrictions.
2522	 */
2523	v = sc->sc_mac_rxcfg & ~(CAS_MAC_RX_CONF_HFILTER |
2524	    CAS_MAC_RX_CONF_EN);
2525	CAS_WRITE_4(sc, CAS_MAC_RX_CONF, v);
2526	CAS_BARRIER(sc, CAS_MAC_RX_CONF, 4,
2527	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2528	if (!cas_bitwait(sc, CAS_MAC_RX_CONF, CAS_MAC_RX_CONF_HFILTER |
2529	    CAS_MAC_RX_CONF_EN, 0))
2530		device_printf(sc->sc_dev,
2531		    "cannot disable RX MAC or hash filter\n");
2532
2533	v &= ~(CAS_MAC_RX_CONF_PROMISC | CAS_MAC_RX_CONF_PGRP);
2534	if ((if_getflags(ifp) & IFF_PROMISC) != 0) {
2535		v |= CAS_MAC_RX_CONF_PROMISC;
2536		goto chipit;
2537	}
2538	if ((if_getflags(ifp) & IFF_ALLMULTI) != 0) {
2539		v |= CAS_MAC_RX_CONF_PGRP;
2540		goto chipit;
2541	}
2542
2543	/*
2544	 * Set up multicast address filter by passing all multicast
2545	 * addresses through a crc generator, and then using the high
2546	 * order 8 bits as an index into the 256 bit logical address
2547	 * filter.  The high order 4 bits selects the word, while the
2548	 * other 4 bits select the bit within the word (where bit 0
2549	 * is the MSB).
2550	 */
2551
2552	memset(hash, 0, sizeof(hash));
2553	if_foreach_llmaddr(ifp, cas_hash_maddr, &hash);
2554
2555	v |= CAS_MAC_RX_CONF_HFILTER;
2556
2557	/* Now load the hash table into the chip (if we are using it). */
2558	for (i = 0; i < 16; i++)
2559		CAS_WRITE_4(sc,
2560		    CAS_MAC_HASH0 + i * (CAS_MAC_HASH1 - CAS_MAC_HASH0),
2561		    hash[i]);
2562
2563 chipit:
2564	sc->sc_mac_rxcfg = v;
2565	CAS_WRITE_4(sc, CAS_MAC_RX_CONF, v | CAS_MAC_RX_CONF_EN);
2566}
2567
2568static int	cas_pci_attach(device_t dev);
2569static int	cas_pci_detach(device_t dev);
2570static int	cas_pci_probe(device_t dev);
2571static int	cas_pci_resume(device_t dev);
2572static int	cas_pci_suspend(device_t dev);
2573
2574static device_method_t cas_pci_methods[] = {
2575	/* Device interface */
2576	DEVMETHOD(device_probe,		cas_pci_probe),
2577	DEVMETHOD(device_attach,	cas_pci_attach),
2578	DEVMETHOD(device_detach,	cas_pci_detach),
2579	DEVMETHOD(device_suspend,	cas_pci_suspend),
2580	DEVMETHOD(device_resume,	cas_pci_resume),
2581	/* Use the suspend handler here, it is all that is required. */
2582	DEVMETHOD(device_shutdown,	cas_pci_suspend),
2583
2584	/* MII interface */
2585	DEVMETHOD(miibus_readreg,	cas_mii_readreg),
2586	DEVMETHOD(miibus_writereg,	cas_mii_writereg),
2587	DEVMETHOD(miibus_statchg,	cas_mii_statchg),
2588
2589	DEVMETHOD_END
2590};
2591
2592static driver_t cas_pci_driver = {
2593	"cas",
2594	cas_pci_methods,
2595	sizeof(struct cas_softc)
2596};
2597
2598static const struct cas_pci_dev {
2599	uint32_t	cpd_devid;
2600	uint8_t		cpd_revid;
2601	int		cpd_variant;
2602	const char	*cpd_desc;
2603} cas_pci_devlist[] = {
2604	{ 0x0035100b, 0x0, CAS_SATURN, "NS DP83065 Saturn Gigabit Ethernet" },
2605	{ 0xabba108e, 0x10, CAS_CASPLUS, "Sun Cassini+ Gigabit Ethernet" },
2606	{ 0xabba108e, 0x0, CAS_CAS, "Sun Cassini Gigabit Ethernet" },
2607	{ 0, 0, 0, NULL }
2608};
2609
2610DRIVER_MODULE(cas, pci, cas_pci_driver, 0, 0);
2611MODULE_PNP_INFO("W32:vendor/device", pci, cas, cas_pci_devlist,
2612    nitems(cas_pci_devlist) - 1);
2613DRIVER_MODULE(miibus, cas, miibus_driver, 0, 0);
2614MODULE_DEPEND(cas, pci, 1, 1, 1);
2615
2616static int
2617cas_pci_probe(device_t dev)
2618{
2619	int i;
2620
2621	for (i = 0; cas_pci_devlist[i].cpd_desc != NULL; i++) {
2622		if (pci_get_devid(dev) == cas_pci_devlist[i].cpd_devid &&
2623		    pci_get_revid(dev) >= cas_pci_devlist[i].cpd_revid) {
2624			device_set_desc(dev, cas_pci_devlist[i].cpd_desc);
2625			return (BUS_PROBE_DEFAULT);
2626		}
2627	}
2628
2629	return (ENXIO);
2630}
2631
2632static struct resource_spec cas_pci_res_spec[] = {
2633	{ SYS_RES_IRQ, 0, RF_SHAREABLE | RF_ACTIVE },	/* CAS_RES_INTR */
2634	{ SYS_RES_MEMORY, PCIR_BAR(0), RF_ACTIVE },	/* CAS_RES_MEM */
2635	{ -1, 0 }
2636};
2637
2638#define	CAS_LOCAL_MAC_ADDRESS	"local-mac-address"
2639#define	CAS_PHY_INTERFACE	"phy-interface"
2640#define	CAS_PHY_TYPE		"phy-type"
2641#define	CAS_PHY_TYPE_PCS	"pcs"
2642
2643static int
2644cas_pci_attach(device_t dev)
2645{
2646	char buf[sizeof(CAS_LOCAL_MAC_ADDRESS)];
2647	struct cas_softc *sc;
2648	int i;
2649#if !defined(__powerpc__)
2650	u_char enaddr[4][ETHER_ADDR_LEN];
2651	u_int j, k, lma, pcs[4], phy;
2652#endif
2653
2654	sc = device_get_softc(dev);
2655	sc->sc_variant = CAS_UNKNOWN;
2656	for (i = 0; cas_pci_devlist[i].cpd_desc != NULL; i++) {
2657		if (pci_get_devid(dev) == cas_pci_devlist[i].cpd_devid &&
2658		    pci_get_revid(dev) >= cas_pci_devlist[i].cpd_revid) {
2659			sc->sc_variant = cas_pci_devlist[i].cpd_variant;
2660			break;
2661		}
2662	}
2663	if (sc->sc_variant == CAS_UNKNOWN) {
2664		device_printf(dev, "unknown adaptor\n");
2665		return (ENXIO);
2666	}
2667
2668	/* PCI configuration */
2669	pci_write_config(dev, PCIR_COMMAND,
2670	    pci_read_config(dev, PCIR_COMMAND, 2) | PCIM_CMD_BUSMASTEREN |
2671	    PCIM_CMD_MWRICEN | PCIM_CMD_PERRESPEN | PCIM_CMD_SERRESPEN, 2);
2672
2673	sc->sc_dev = dev;
2674	if (sc->sc_variant == CAS_CAS && pci_get_devid(dev) < 0x02)
2675		/* Hardware checksumming may hang TX. */
2676		sc->sc_flags |= CAS_NO_CSUM;
2677	if (sc->sc_variant == CAS_CASPLUS || sc->sc_variant == CAS_SATURN)
2678		sc->sc_flags |= CAS_REG_PLUS;
2679	if (sc->sc_variant == CAS_CAS ||
2680	    (sc->sc_variant == CAS_CASPLUS && pci_get_revid(dev) < 0x11))
2681		sc->sc_flags |= CAS_TABORT;
2682	if (bootverbose)
2683		device_printf(dev, "flags=0x%x\n", sc->sc_flags);
2684
2685	if (bus_alloc_resources(dev, cas_pci_res_spec, sc->sc_res)) {
2686		device_printf(dev, "failed to allocate resources\n");
2687		bus_release_resources(dev, cas_pci_res_spec, sc->sc_res);
2688		return (ENXIO);
2689	}
2690
2691	CAS_LOCK_INIT(sc, device_get_nameunit(dev));
2692
2693#if defined(__powerpc__)
2694	OF_getetheraddr(dev, sc->sc_enaddr);
2695	if (OF_getprop(ofw_bus_get_node(dev), CAS_PHY_INTERFACE, buf,
2696	    sizeof(buf)) > 0 || OF_getprop(ofw_bus_get_node(dev),
2697	    CAS_PHY_TYPE, buf, sizeof(buf)) > 0) {
2698		buf[sizeof(buf) - 1] = '\0';
2699		if (strcmp(buf, CAS_PHY_TYPE_PCS) == 0)
2700			sc->sc_flags |= CAS_SERDES;
2701	}
2702#else
2703	/*
2704	 * Dig out VPD (vital product data) and read the MAC address as well
2705	 * as the PHY type.  The VPD resides in the PCI Expansion ROM (PCI
2706	 * FCode) and can't be accessed via the PCI capability pointer.
2707	 * SUNW,pci-ce and SUNW,pci-qge use the Enhanced VPD format described
2708	 * in the free US Patent 7149820.
2709	 */
2710
2711#define	PCI_ROMHDR_SIZE			0x1c
2712#define	PCI_ROMHDR_SIG			0x00
2713#define	PCI_ROMHDR_SIG_MAGIC		0xaa55		/* little endian */
2714#define	PCI_ROMHDR_PTR_DATA		0x18
2715#define	PCI_ROM_SIZE			0x18
2716#define	PCI_ROM_SIG			0x00
2717#define	PCI_ROM_SIG_MAGIC		0x52494350	/* "PCIR", endian */
2718							/* reversed */
2719#define	PCI_ROM_VENDOR			0x04
2720#define	PCI_ROM_DEVICE			0x06
2721#define	PCI_ROM_PTR_VPD			0x08
2722#define	PCI_VPDRES_BYTE0		0x00
2723#define	PCI_VPDRES_ISLARGE(x)		((x) & 0x80)
2724#define	PCI_VPDRES_LARGE_NAME(x)	((x) & 0x7f)
2725#define	PCI_VPDRES_LARGE_LEN_LSB	0x01
2726#define	PCI_VPDRES_LARGE_LEN_MSB	0x02
2727#define	PCI_VPDRES_LARGE_SIZE		0x03
2728#define	PCI_VPDRES_TYPE_ID_STRING	0x02		/* large */
2729#define	PCI_VPDRES_TYPE_VPD		0x10		/* large */
2730#define	PCI_VPD_KEY0			0x00
2731#define	PCI_VPD_KEY1			0x01
2732#define	PCI_VPD_LEN			0x02
2733#define	PCI_VPD_SIZE			0x03
2734
2735#define	CAS_ROM_READ_1(sc, offs)					\
2736	CAS_READ_1((sc), CAS_PCI_ROM_OFFSET + (offs))
2737#define	CAS_ROM_READ_2(sc, offs)					\
2738	CAS_READ_2((sc), CAS_PCI_ROM_OFFSET + (offs))
2739#define	CAS_ROM_READ_4(sc, offs)					\
2740	CAS_READ_4((sc), CAS_PCI_ROM_OFFSET + (offs))
2741
2742	lma = phy = 0;
2743	memset(enaddr, 0, sizeof(enaddr));
2744	memset(pcs, 0, sizeof(pcs));
2745
2746	/* Enable PCI Expansion ROM access. */
2747	CAS_WRITE_4(sc, CAS_BIM_LDEV_OEN,
2748	    CAS_BIM_LDEV_OEN_PAD | CAS_BIM_LDEV_OEN_PROM);
2749
2750	/* Read PCI Expansion ROM header. */
2751	if (CAS_ROM_READ_2(sc, PCI_ROMHDR_SIG) != PCI_ROMHDR_SIG_MAGIC ||
2752	    (i = CAS_ROM_READ_2(sc, PCI_ROMHDR_PTR_DATA)) <
2753	    PCI_ROMHDR_SIZE) {
2754		device_printf(dev, "unexpected PCI Expansion ROM header\n");
2755		goto fail_prom;
2756	}
2757
2758	/* Read PCI Expansion ROM data. */
2759	if (CAS_ROM_READ_4(sc, i + PCI_ROM_SIG) != PCI_ROM_SIG_MAGIC ||
2760	    CAS_ROM_READ_2(sc, i + PCI_ROM_VENDOR) != pci_get_vendor(dev) ||
2761	    CAS_ROM_READ_2(sc, i + PCI_ROM_DEVICE) != pci_get_device(dev) ||
2762	    (j = CAS_ROM_READ_2(sc, i + PCI_ROM_PTR_VPD)) <
2763	    i + PCI_ROM_SIZE) {
2764		device_printf(dev, "unexpected PCI Expansion ROM data\n");
2765		goto fail_prom;
2766	}
2767
2768	/* Read PCI VPD. */
2769 next:
2770	if (PCI_VPDRES_ISLARGE(CAS_ROM_READ_1(sc,
2771	    j + PCI_VPDRES_BYTE0)) == 0) {
2772		device_printf(dev, "no large PCI VPD\n");
2773		goto fail_prom;
2774	}
2775
2776	i = (CAS_ROM_READ_1(sc, j + PCI_VPDRES_LARGE_LEN_MSB) << 8) |
2777	    CAS_ROM_READ_1(sc, j + PCI_VPDRES_LARGE_LEN_LSB);
2778	switch (PCI_VPDRES_LARGE_NAME(CAS_ROM_READ_1(sc,
2779	    j + PCI_VPDRES_BYTE0))) {
2780	case PCI_VPDRES_TYPE_ID_STRING:
2781		/* Skip identifier string. */
2782		j += PCI_VPDRES_LARGE_SIZE + i;
2783		goto next;
2784	case PCI_VPDRES_TYPE_VPD:
2785		for (j += PCI_VPDRES_LARGE_SIZE; i > 0;
2786		    i -= PCI_VPD_SIZE + CAS_ROM_READ_1(sc, j + PCI_VPD_LEN),
2787		    j += PCI_VPD_SIZE + CAS_ROM_READ_1(sc, j + PCI_VPD_LEN)) {
2788			if (CAS_ROM_READ_1(sc, j + PCI_VPD_KEY0) != 'Z')
2789				/* no Enhanced VPD */
2790				continue;
2791			if (CAS_ROM_READ_1(sc, j + PCI_VPD_SIZE) != 'I')
2792				/* no instance property */
2793				continue;
2794			if (CAS_ROM_READ_1(sc, j + PCI_VPD_SIZE + 3) == 'B') {
2795				/* byte array */
2796				if (CAS_ROM_READ_1(sc,
2797				    j + PCI_VPD_SIZE + 4) != ETHER_ADDR_LEN)
2798					continue;
2799				bus_read_region_1(sc->sc_res[CAS_RES_MEM],
2800				    CAS_PCI_ROM_OFFSET + j + PCI_VPD_SIZE + 5,
2801				    buf, sizeof(buf));
2802				buf[sizeof(buf) - 1] = '\0';
2803				if (strcmp(buf, CAS_LOCAL_MAC_ADDRESS) != 0)
2804					continue;
2805				bus_read_region_1(sc->sc_res[CAS_RES_MEM],
2806				    CAS_PCI_ROM_OFFSET + j + PCI_VPD_SIZE +
2807				    5 + sizeof(CAS_LOCAL_MAC_ADDRESS),
2808				    enaddr[lma], sizeof(enaddr[lma]));
2809				lma++;
2810				if (lma == 4 && phy == 4)
2811					break;
2812			} else if (CAS_ROM_READ_1(sc, j + PCI_VPD_SIZE + 3) ==
2813			   'S') {
2814				/* string */
2815				if (CAS_ROM_READ_1(sc,
2816				    j + PCI_VPD_SIZE + 4) !=
2817				    sizeof(CAS_PHY_TYPE_PCS))
2818					continue;
2819				bus_read_region_1(sc->sc_res[CAS_RES_MEM],
2820				    CAS_PCI_ROM_OFFSET + j + PCI_VPD_SIZE + 5,
2821				    buf, sizeof(buf));
2822				buf[sizeof(buf) - 1] = '\0';
2823				if (strcmp(buf, CAS_PHY_INTERFACE) == 0)
2824					k = sizeof(CAS_PHY_INTERFACE);
2825				else if (strcmp(buf, CAS_PHY_TYPE) == 0)
2826					k = sizeof(CAS_PHY_TYPE);
2827				else
2828					continue;
2829				bus_read_region_1(sc->sc_res[CAS_RES_MEM],
2830				    CAS_PCI_ROM_OFFSET + j + PCI_VPD_SIZE +
2831				    5 + k, buf, sizeof(buf));
2832				buf[sizeof(buf) - 1] = '\0';
2833				if (strcmp(buf, CAS_PHY_TYPE_PCS) == 0)
2834					pcs[phy] = 1;
2835				phy++;
2836				if (lma == 4 && phy == 4)
2837					break;
2838			}
2839		}
2840		break;
2841	default:
2842		device_printf(dev, "unexpected PCI VPD\n");
2843		goto fail_prom;
2844	}
2845
2846 fail_prom:
2847	CAS_WRITE_4(sc, CAS_BIM_LDEV_OEN, 0);
2848
2849	if (lma == 0) {
2850		device_printf(dev, "could not determine Ethernet address\n");
2851		goto fail;
2852	}
2853	i = 0;
2854	if (lma > 1 && pci_get_slot(dev) < nitems(enaddr))
2855		i = pci_get_slot(dev);
2856	memcpy(sc->sc_enaddr, enaddr[i], ETHER_ADDR_LEN);
2857
2858	if (phy == 0) {
2859		device_printf(dev, "could not determine PHY type\n");
2860		goto fail;
2861	}
2862	i = 0;
2863	if (phy > 1 && pci_get_slot(dev) < nitems(pcs))
2864		i = pci_get_slot(dev);
2865	if (pcs[i] != 0)
2866		sc->sc_flags |= CAS_SERDES;
2867#endif
2868
2869	if (cas_attach(sc) != 0) {
2870		device_printf(dev, "could not be attached\n");
2871		goto fail;
2872	}
2873
2874	if (bus_setup_intr(dev, sc->sc_res[CAS_RES_INTR], INTR_TYPE_NET |
2875	    INTR_MPSAFE, cas_intr, NULL, sc, &sc->sc_ih) != 0) {
2876		device_printf(dev, "failed to set up interrupt\n");
2877		cas_detach(sc);
2878		goto fail;
2879	}
2880	return (0);
2881
2882 fail:
2883	CAS_LOCK_DESTROY(sc);
2884	bus_release_resources(dev, cas_pci_res_spec, sc->sc_res);
2885	return (ENXIO);
2886}
2887
2888static int
2889cas_pci_detach(device_t dev)
2890{
2891	struct cas_softc *sc;
2892
2893	sc = device_get_softc(dev);
2894	bus_teardown_intr(dev, sc->sc_res[CAS_RES_INTR], sc->sc_ih);
2895	cas_detach(sc);
2896	CAS_LOCK_DESTROY(sc);
2897	bus_release_resources(dev, cas_pci_res_spec, sc->sc_res);
2898	return (0);
2899}
2900
2901static int
2902cas_pci_suspend(device_t dev)
2903{
2904
2905	cas_suspend(device_get_softc(dev));
2906	return (0);
2907}
2908
2909static int
2910cas_pci_resume(device_t dev)
2911{
2912
2913	cas_resume(device_get_softc(dev));
2914	return (0);
2915}
2916