xref: /freebsd/sys/dev/cas/if_cas.c (revision 830940567b49bb0c08dfaed40418999e76616909)
1 /*-
2  * Copyright (C) 2001 Eduardo Horvath.
3  * Copyright (c) 2001-2003 Thomas Moestl
4  * Copyright (c) 2007-2009 Marius Strobl <marius@FreeBSD.org>
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR  ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR  BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  *
28  *	from: NetBSD: gem.c,v 1.21 2002/06/01 23:50:58 lukem Exp
29  *	from: FreeBSD: if_gem.c 182060 2008-08-23 15:03:26Z marius
30  */
31 
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34 
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_arp.h>
66 #include <net/if_dl.h>
67 #include <net/if_media.h>
68 #include <net/if_types.h>
69 #include <net/if_vlan_var.h>
70 
71 #include <netinet/in.h>
72 #include <netinet/in_systm.h>
73 #include <netinet/ip.h>
74 #include <netinet/tcp.h>
75 #include <netinet/udp.h>
76 
77 #include <machine/bus.h>
78 #if defined(__powerpc__) || defined(__sparc64__)
79 #include <dev/ofw/openfirm.h>
80 #include <machine/ofw_machdep.h>
81 #endif
82 #include <machine/resource.h>
83 
84 #include <dev/mii/mii.h>
85 #include <dev/mii/miivar.h>
86 
87 #include <dev/cas/if_casreg.h>
88 #include <dev/cas/if_casvar.h>
89 
90 #include <dev/pci/pcireg.h>
91 #include <dev/pci/pcivar.h>
92 
93 #include "miibus_if.h"
94 
95 #define RINGASSERT(n , min, max)					\
96 	CTASSERT(powerof2(n) && (n) >= (min) && (n) <= (max))
97 
98 RINGASSERT(CAS_NRXCOMP, 128, 32768);
99 RINGASSERT(CAS_NRXDESC, 32, 8192);
100 RINGASSERT(CAS_NRXDESC2, 32, 8192);
101 RINGASSERT(CAS_NTXDESC, 32, 8192);
102 
103 #undef RINGASSERT
104 
105 #define	CCDASSERT(m, a)							\
106 	CTASSERT((offsetof(struct cas_control_data, m) & ((a) - 1)) == 0)
107 
108 CCDASSERT(ccd_rxcomps, CAS_RX_COMP_ALIGN);
109 CCDASSERT(ccd_rxdescs, CAS_RX_DESC_ALIGN);
110 CCDASSERT(ccd_rxdescs2, CAS_RX_DESC_ALIGN);
111 
112 #undef CCDASSERT
113 
114 #define	CAS_TRIES	10000
115 
116 /*
117  * According to documentation, the hardware has support for basic TCP
118  * checksum offloading only, in practice this can be also used for UDP
119  * however (i.e. the problem of previous Sun NICs that a checksum of 0x0
120  * is not converted to 0xffff no longer exists).
121  */
122 #define	CAS_CSUM_FEATURES	(CSUM_TCP | CSUM_UDP)
123 
124 static inline void cas_add_rxdesc(struct cas_softc *sc, u_int idx);
125 static int	cas_attach(struct cas_softc *sc);
126 static int	cas_bitwait(struct cas_softc *sc, bus_addr_t r, uint32_t clr,
127 		    uint32_t set);
128 static void	cas_cddma_callback(void *xsc, bus_dma_segment_t *segs,
129 		    int nsegs, int error);
130 static void	cas_detach(struct cas_softc *sc);
131 static int	cas_disable_rx(struct cas_softc *sc);
132 static int	cas_disable_tx(struct cas_softc *sc);
133 static void	cas_eint(struct cas_softc *sc, u_int status);
134 static void	cas_free(void *arg1, void* arg2);
135 static void	cas_init(void *xsc);
136 static void	cas_init_locked(struct cas_softc *sc);
137 static void	cas_init_regs(struct cas_softc *sc);
138 static int	cas_intr(void *v);
139 static void	cas_intr_task(void *arg, int pending __unused);
140 static int	cas_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data);
141 static int	cas_load_txmbuf(struct cas_softc *sc, struct mbuf **m_head);
142 static int	cas_mediachange(struct ifnet *ifp);
143 static void	cas_mediastatus(struct ifnet *ifp, struct ifmediareq *ifmr);
144 static void	cas_meminit(struct cas_softc *sc);
145 static void	cas_mifinit(struct cas_softc *sc);
146 static int	cas_mii_readreg(device_t dev, int phy, int reg);
147 static void	cas_mii_statchg(device_t dev);
148 static int	cas_mii_writereg(device_t dev, int phy, int reg, int val);
149 static void	cas_reset(struct cas_softc *sc);
150 static int	cas_reset_rx(struct cas_softc *sc);
151 static int	cas_reset_tx(struct cas_softc *sc);
152 static void	cas_resume(struct cas_softc *sc);
153 static u_int	cas_descsize(u_int sz);
154 static void	cas_rint(struct cas_softc *sc);
155 static void	cas_rint_timeout(void *arg);
156 static inline void cas_rxcksum(struct mbuf *m, uint16_t cksum);
157 static inline void cas_rxcompinit(struct cas_rx_comp *rxcomp);
158 static u_int	cas_rxcompsize(u_int sz);
159 static void	cas_rxdma_callback(void *xsc, bus_dma_segment_t *segs,
160 		    int nsegs, int error);
161 static void	cas_setladrf(struct cas_softc *sc);
162 static void	cas_start(struct ifnet *ifp);
163 static void	cas_stop(struct ifnet *ifp);
164 static void	cas_suspend(struct cas_softc *sc);
165 static void	cas_tick(void *arg);
166 static void	cas_tint(struct cas_softc *sc);
167 static void	cas_tx_task(void *arg, int pending __unused);
168 static inline void cas_txkick(struct cas_softc *sc);
169 static void	cas_watchdog(struct cas_softc *sc);
170 
171 static devclass_t cas_devclass;
172 
173 MODULE_DEPEND(cas, ether, 1, 1, 1);
174 MODULE_DEPEND(cas, miibus, 1, 1, 1);
175 
176 #ifdef CAS_DEBUG
177 #include <sys/ktr.h>
178 #define	KTR_CAS		KTR_CT2
179 #endif
180 
181 static int
182 cas_attach(struct cas_softc *sc)
183 {
184 	struct cas_txsoft *txs;
185 	struct ifnet *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 	ifp->if_softc = sc;
194 	if_initname(ifp, device_get_name(sc->sc_dev),
195 	    device_get_unit(sc->sc_dev));
196 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
197 	ifp->if_start = cas_start;
198 	ifp->if_ioctl = cas_ioctl;
199 	ifp->if_init = cas_init;
200 	IFQ_SET_MAXLEN(&ifp->if_snd, CAS_TXQUEUELEN);
201 	ifp->if_snd.ifq_drv_maxlen = CAS_TXQUEUELEN;
202 	IFQ_SET_READY(&ifp->if_snd);
203 
204 	callout_init_mtx(&sc->sc_tick_ch, &sc->sc_mtx, 0);
205 	callout_init(&sc->sc_rx_ch, 1);
206 	/* Create local taskq. */
207 	TASK_INIT(&sc->sc_intr_task, 0, cas_intr_task, sc);
208 	TASK_INIT(&sc->sc_tx_task, 1, cas_tx_task, ifp);
209 	sc->sc_tq = taskqueue_create_fast("cas_taskq", M_WAITOK,
210 	    taskqueue_thread_enqueue, &sc->sc_tq);
211 	if (sc->sc_tq == NULL) {
212 		device_printf(sc->sc_dev, "could not create taskqueue\n");
213 		error = ENXIO;
214 		goto fail_ifnet;
215 	}
216 	taskqueue_start_threads(&sc->sc_tq, 1, PI_NET, "%s taskq",
217 	    device_get_nameunit(sc->sc_dev));
218 
219 	/* Make sure the chip is stopped. */
220 	cas_reset(sc);
221 
222 	error = bus_dma_tag_create(bus_get_dma_tag(sc->sc_dev), 1, 0,
223 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
224 	    BUS_SPACE_MAXSIZE, 0, BUS_SPACE_MAXSIZE, 0, NULL, NULL,
225 	    &sc->sc_pdmatag);
226 	if (error != 0)
227 		goto fail_taskq;
228 
229 	error = bus_dma_tag_create(sc->sc_pdmatag, 1, 0,
230 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
231 	    CAS_PAGE_SIZE, 1, CAS_PAGE_SIZE, 0, NULL, NULL, &sc->sc_rdmatag);
232 	if (error != 0)
233 		goto fail_ptag;
234 
235 	error = bus_dma_tag_create(sc->sc_pdmatag, 1, 0,
236 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
237 	    MCLBYTES * CAS_NTXSEGS, CAS_NTXSEGS, MCLBYTES,
238 	    BUS_DMA_ALLOCNOW, NULL, NULL, &sc->sc_tdmatag);
239 	if (error != 0)
240 		goto fail_rtag;
241 
242 	error = bus_dma_tag_create(sc->sc_pdmatag, CAS_TX_DESC_ALIGN, 0,
243 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
244 	    sizeof(struct cas_control_data), 1,
245 	    sizeof(struct cas_control_data), 0,
246 	    NULL, NULL, &sc->sc_cdmatag);
247 	if (error != 0)
248 		goto fail_ttag;
249 
250 	/*
251 	 * Allocate the control data structures, create and load the
252 	 * DMA map for it.
253 	 */
254 	if ((error = bus_dmamem_alloc(sc->sc_cdmatag,
255 	    (void **)&sc->sc_control_data,
256 	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
257 	    &sc->sc_cddmamap)) != 0) {
258 		device_printf(sc->sc_dev,
259 		    "unable to allocate control data, error = %d\n", error);
260 		goto fail_ctag;
261 	}
262 
263 	sc->sc_cddma = 0;
264 	if ((error = bus_dmamap_load(sc->sc_cdmatag, sc->sc_cddmamap,
265 	    sc->sc_control_data, sizeof(struct cas_control_data),
266 	    cas_cddma_callback, sc, 0)) != 0 || sc->sc_cddma == 0) {
267 		device_printf(sc->sc_dev,
268 		    "unable to load control data DMA map, error = %d\n",
269 		    error);
270 		goto fail_cmem;
271 	}
272 
273 	/*
274 	 * Initialize the transmit job descriptors.
275 	 */
276 	STAILQ_INIT(&sc->sc_txfreeq);
277 	STAILQ_INIT(&sc->sc_txdirtyq);
278 
279 	/*
280 	 * Create the transmit buffer DMA maps.
281 	 */
282 	error = ENOMEM;
283 	for (i = 0; i < CAS_TXQUEUELEN; i++) {
284 		txs = &sc->sc_txsoft[i];
285 		txs->txs_mbuf = NULL;
286 		txs->txs_ndescs = 0;
287 		if ((error = bus_dmamap_create(sc->sc_tdmatag, 0,
288 		    &txs->txs_dmamap)) != 0) {
289 			device_printf(sc->sc_dev,
290 			    "unable to create TX DMA map %d, error = %d\n",
291 			    i, error);
292 			goto fail_txd;
293 		}
294 		STAILQ_INSERT_TAIL(&sc->sc_txfreeq, txs, txs_q);
295 	}
296 
297 	/*
298 	 * Allocate the receive buffers, create and load the DMA maps
299 	 * for them.
300 	 */
301 	for (i = 0; i < CAS_NRXDESC; i++) {
302 		if ((error = bus_dmamem_alloc(sc->sc_rdmatag,
303 		    &sc->sc_rxdsoft[i].rxds_buf, BUS_DMA_WAITOK,
304 		    &sc->sc_rxdsoft[i].rxds_dmamap)) != 0) {
305 			device_printf(sc->sc_dev,
306 			    "unable to allocate RX buffer %d, error = %d\n",
307 			    i, error);
308 			goto fail_rxmem;
309 		}
310 
311 		sc->sc_rxdptr = i;
312 		sc->sc_rxdsoft[i].rxds_paddr = 0;
313 		if ((error = bus_dmamap_load(sc->sc_rdmatag,
314 		    sc->sc_rxdsoft[i].rxds_dmamap, sc->sc_rxdsoft[i].rxds_buf,
315 		    CAS_PAGE_SIZE, cas_rxdma_callback, sc, 0)) != 0 ||
316 		    sc->sc_rxdsoft[i].rxds_paddr == 0) {
317 			device_printf(sc->sc_dev,
318 			    "unable to load RX DMA map %d, error = %d\n",
319 			    i, error);
320 			goto fail_rxmap;
321 		}
322 	}
323 
324 	CAS_WRITE_4(sc, CAS_PCS_DATAPATH, CAS_PCS_DATAPATH_MII);
325 
326 	cas_mifinit(sc);
327 
328 	/*
329 	 * Look for an external PHY.
330 	 */
331 	error = ENXIO;
332 	v = CAS_READ_4(sc, CAS_MIF_CONF);
333 	if ((v & CAS_MIF_CONF_MDI1) != 0) {
334 		v |= CAS_MIF_CONF_PHY_SELECT;
335 		CAS_WRITE_4(sc, CAS_MIF_CONF, v);
336 		switch (sc->sc_variant) {
337 		default:
338 			sc->sc_phyad = -1;
339 			break;
340 		}
341 		error = mii_phy_probe(sc->sc_dev, &sc->sc_miibus,
342 		    cas_mediachange, cas_mediastatus);
343 	}
344 
345 	/*
346 	 * Fall back on an internal PHY if no external PHY was found.
347 	 */
348 	if (error != 0 && (v & CAS_MIF_CONF_MDI0) != 0) {
349 		v &= ~CAS_MIF_CONF_PHY_SELECT;
350 		CAS_WRITE_4(sc, CAS_MIF_CONF, v);
351 		switch (sc->sc_variant) {
352 		default:
353 			sc->sc_phyad = -1;
354 			break;
355 		}
356 		error = mii_phy_probe(sc->sc_dev, &sc->sc_miibus,
357 		    cas_mediachange, cas_mediastatus);
358 	}
359 
360 	/*
361 	 * Try the external PCS SERDES if we didn't find any PHYs.
362 	 */
363 	if (error != 0) {
364 		CAS_WRITE_4(sc, CAS_PCS_DATAPATH, CAS_PCS_DATAPATH_SERDES);
365 		CAS_WRITE_4(sc, CAS_PCS_SERDES_CTRL, CAS_PCS_SERDES_CTRL_ESD);
366 		CAS_WRITE_4(sc, CAS_PCS_CONF_EN, CAS_PCS_CONF_EN);
367 		sc->sc_flags |= CAS_SERDES;
368 		sc->sc_phyad = CAS_PHYAD_EXTERNAL;
369 		error = mii_phy_probe(sc->sc_dev, &sc->sc_miibus,
370 		    cas_mediachange, cas_mediastatus);
371 	}
372 
373 	if (error != 0) {
374 		device_printf(sc->sc_dev, "PHY probe failed: %d\n", error);
375 		goto fail_rxmap;
376 	}
377 	sc->sc_mii = device_get_softc(sc->sc_miibus);
378 
379 	/*
380 	 * From this point forward, the attachment cannot fail.  A failure
381 	 * before this point releases all resources that may have been
382 	 * allocated.
383 	 */
384 
385 	/* Announce FIFO sizes. */
386 	v = CAS_READ_4(sc, CAS_TX_FIFO_SIZE);
387 	device_printf(sc->sc_dev, "%ukB RX FIFO, %ukB TX FIFO\n",
388 	    CAS_RX_FIFO_SIZE / 1024, v / 16);
389 
390 	/* Attach the interface. */
391 	ether_ifattach(ifp, sc->sc_enaddr);
392 
393 	/*
394 	 * Tell the upper layer(s) we support long frames/checksum offloads.
395 	 */
396 	ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
397 	ifp->if_capabilities = IFCAP_VLAN_MTU;
398 	if ((sc->sc_flags & CAS_NO_CSUM) == 0) {
399 		ifp->if_capabilities |= IFCAP_HWCSUM;
400 		ifp->if_hwassist = CAS_CSUM_FEATURES;
401 	}
402 	ifp->if_capenable = ifp->if_capabilities;
403 
404 	return (0);
405 
406 	/*
407 	 * Free any resources we've allocated during the failed attach
408 	 * attempt.  Do this in reverse order and fall through.
409 	 */
410  fail_rxmap:
411 	for (i = 0; i < CAS_NRXDESC; i++)
412 		if (sc->sc_rxdsoft[i].rxds_paddr != 0)
413 			bus_dmamap_unload(sc->sc_rdmatag,
414 			    sc->sc_rxdsoft[i].rxds_dmamap);
415  fail_rxmem:
416 	for (i = 0; i < CAS_NRXDESC; i++)
417 		if (sc->sc_rxdsoft[i].rxds_buf != NULL)
418 			bus_dmamem_free(sc->sc_rdmatag,
419 			    sc->sc_rxdsoft[i].rxds_buf,
420 			    sc->sc_rxdsoft[i].rxds_dmamap);
421  fail_txd:
422 	for (i = 0; i < CAS_TXQUEUELEN; i++)
423 		if (sc->sc_txsoft[i].txs_dmamap != NULL)
424 			bus_dmamap_destroy(sc->sc_tdmatag,
425 			    sc->sc_txsoft[i].txs_dmamap);
426 	bus_dmamap_unload(sc->sc_cdmatag, sc->sc_cddmamap);
427  fail_cmem:
428 	bus_dmamem_free(sc->sc_cdmatag, sc->sc_control_data,
429 	    sc->sc_cddmamap);
430  fail_ctag:
431 	bus_dma_tag_destroy(sc->sc_cdmatag);
432  fail_ttag:
433 	bus_dma_tag_destroy(sc->sc_tdmatag);
434  fail_rtag:
435 	bus_dma_tag_destroy(sc->sc_rdmatag);
436  fail_ptag:
437 	bus_dma_tag_destroy(sc->sc_pdmatag);
438  fail_taskq:
439 	taskqueue_free(sc->sc_tq);
440  fail_ifnet:
441 	if_free(ifp);
442 	return (error);
443 }
444 
445 static void
446 cas_detach(struct cas_softc *sc)
447 {
448 	struct ifnet *ifp = sc->sc_ifp;
449 	int i;
450 
451 	ether_ifdetach(ifp);
452 	CAS_LOCK(sc);
453 	cas_stop(ifp);
454 	CAS_UNLOCK(sc);
455 	callout_drain(&sc->sc_tick_ch);
456 	callout_drain(&sc->sc_rx_ch);
457 	taskqueue_drain(sc->sc_tq, &sc->sc_intr_task);
458 	taskqueue_drain(sc->sc_tq, &sc->sc_tx_task);
459 	if_free(ifp);
460 	taskqueue_free(sc->sc_tq);
461 	device_delete_child(sc->sc_dev, sc->sc_miibus);
462 
463 	for (i = 0; i < CAS_NRXDESC; i++)
464 		if (sc->sc_rxdsoft[i].rxds_dmamap != NULL)
465 			bus_dmamap_sync(sc->sc_rdmatag,
466 			    sc->sc_rxdsoft[i].rxds_dmamap,
467 			    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
468 	for (i = 0; i < CAS_NRXDESC; i++)
469 		if (sc->sc_rxdsoft[i].rxds_paddr != 0)
470 			bus_dmamap_unload(sc->sc_rdmatag,
471 			    sc->sc_rxdsoft[i].rxds_dmamap);
472 	for (i = 0; i < CAS_NRXDESC; i++)
473 		if (sc->sc_rxdsoft[i].rxds_buf != NULL)
474 			bus_dmamem_free(sc->sc_rdmatag,
475 			    sc->sc_rxdsoft[i].rxds_buf,
476 			    sc->sc_rxdsoft[i].rxds_dmamap);
477 	for (i = 0; i < CAS_TXQUEUELEN; i++)
478 		if (sc->sc_txsoft[i].txs_dmamap != NULL)
479 			bus_dmamap_destroy(sc->sc_tdmatag,
480 			    sc->sc_txsoft[i].txs_dmamap);
481 	CAS_CDSYNC(sc, BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
482 	bus_dmamap_unload(sc->sc_cdmatag, sc->sc_cddmamap);
483 	bus_dmamem_free(sc->sc_cdmatag, sc->sc_control_data,
484 	    sc->sc_cddmamap);
485 	bus_dma_tag_destroy(sc->sc_cdmatag);
486 	bus_dma_tag_destroy(sc->sc_tdmatag);
487 	bus_dma_tag_destroy(sc->sc_rdmatag);
488 	bus_dma_tag_destroy(sc->sc_pdmatag);
489 }
490 
491 static void
492 cas_suspend(struct cas_softc *sc)
493 {
494 	struct ifnet *ifp = sc->sc_ifp;
495 
496 	CAS_LOCK(sc);
497 	cas_stop(ifp);
498 	CAS_UNLOCK(sc);
499 }
500 
501 static void
502 cas_resume(struct cas_softc *sc)
503 {
504 	struct ifnet *ifp = sc->sc_ifp;
505 
506 	CAS_LOCK(sc);
507 	/*
508 	 * On resume all registers have to be initialized again like
509 	 * after power-on.
510 	 */
511 	sc->sc_flags &= ~CAS_INITED;
512 	if (ifp->if_flags & IFF_UP)
513 		cas_init_locked(sc);
514 	CAS_UNLOCK(sc);
515 }
516 
517 static inline void
518 cas_rxcksum(struct mbuf *m, uint16_t cksum)
519 {
520 	struct ether_header *eh;
521 	struct ip *ip;
522 	struct udphdr *uh;
523 	uint16_t *opts;
524 	int32_t hlen, len, pktlen;
525 	uint32_t temp32;
526 
527 	pktlen = m->m_pkthdr.len;
528 	if (pktlen < sizeof(struct ether_header) + sizeof(struct ip))
529 		return;
530 	eh = mtod(m, struct ether_header *);
531 	if (eh->ether_type != htons(ETHERTYPE_IP))
532 		return;
533 	ip = (struct ip *)(eh + 1);
534 	if (ip->ip_v != IPVERSION)
535 		return;
536 
537 	hlen = ip->ip_hl << 2;
538 	pktlen -= sizeof(struct ether_header);
539 	if (hlen < sizeof(struct ip))
540 		return;
541 	if (ntohs(ip->ip_len) < hlen)
542 		return;
543 	if (ntohs(ip->ip_len) != pktlen)
544 		return;
545 	if (ip->ip_off & htons(IP_MF | IP_OFFMASK))
546 		return;	/* Cannot handle fragmented packet. */
547 
548 	switch (ip->ip_p) {
549 	case IPPROTO_TCP:
550 		if (pktlen < (hlen + sizeof(struct tcphdr)))
551 			return;
552 		break;
553 	case IPPROTO_UDP:
554 		if (pktlen < (hlen + sizeof(struct udphdr)))
555 			return;
556 		uh = (struct udphdr *)((uint8_t *)ip + hlen);
557 		if (uh->uh_sum == 0)
558 			return; /* no checksum */
559 		break;
560 	default:
561 		return;
562 	}
563 
564 	cksum = ~cksum;
565 	/* checksum fixup for IP options */
566 	len = hlen - sizeof(struct ip);
567 	if (len > 0) {
568 		opts = (uint16_t *)(ip + 1);
569 		for (; len > 0; len -= sizeof(uint16_t), opts++) {
570 			temp32 = cksum - *opts;
571 			temp32 = (temp32 >> 16) + (temp32 & 65535);
572 			cksum = temp32 & 65535;
573 		}
574 	}
575 	m->m_pkthdr.csum_flags |= CSUM_DATA_VALID;
576 	m->m_pkthdr.csum_data = cksum;
577 }
578 
579 static void
580 cas_cddma_callback(void *xsc, bus_dma_segment_t *segs, int nsegs, int error)
581 {
582 	struct cas_softc *sc = xsc;
583 
584 	if (error != 0)
585 		return;
586 	if (nsegs != 1)
587 		panic("%s: bad control buffer segment count", __func__);
588 	sc->sc_cddma = segs[0].ds_addr;
589 }
590 
591 static void
592 cas_rxdma_callback(void *xsc, bus_dma_segment_t *segs, int nsegs, int error)
593 {
594 	struct cas_softc *sc = xsc;
595 
596 	if (error != 0)
597 		return;
598 	if (nsegs != 1)
599 		panic("%s: bad RX buffer segment count", __func__);
600 	sc->sc_rxdsoft[sc->sc_rxdptr].rxds_paddr = segs[0].ds_addr;
601 }
602 
603 static void
604 cas_tick(void *arg)
605 {
606 	struct cas_softc *sc = arg;
607 	struct ifnet *ifp = sc->sc_ifp;
608 	uint32_t v;
609 
610 	CAS_LOCK_ASSERT(sc, MA_OWNED);
611 
612 	/*
613 	 * Unload collision and error counters.
614 	 */
615 	ifp->if_collisions +=
616 	    CAS_READ_4(sc, CAS_MAC_NORM_COLL_CNT) +
617 	    CAS_READ_4(sc, CAS_MAC_FIRST_COLL_CNT);
618 	v = CAS_READ_4(sc, CAS_MAC_EXCESS_COLL_CNT) +
619 	    CAS_READ_4(sc, CAS_MAC_LATE_COLL_CNT);
620 	ifp->if_collisions += v;
621 	ifp->if_oerrors += v;
622 	ifp->if_ierrors +=
623 	    CAS_READ_4(sc, CAS_MAC_RX_LEN_ERR_CNT) +
624 	    CAS_READ_4(sc, CAS_MAC_RX_ALIGN_ERR) +
625 	    CAS_READ_4(sc, CAS_MAC_RX_CRC_ERR_CNT) +
626 	    CAS_READ_4(sc, CAS_MAC_RX_CODE_VIOL);
627 
628 	/*
629 	 * Then clear the hardware counters.
630 	 */
631 	CAS_WRITE_4(sc, CAS_MAC_NORM_COLL_CNT, 0);
632 	CAS_WRITE_4(sc, CAS_MAC_FIRST_COLL_CNT, 0);
633 	CAS_WRITE_4(sc, CAS_MAC_EXCESS_COLL_CNT, 0);
634 	CAS_WRITE_4(sc, CAS_MAC_LATE_COLL_CNT, 0);
635 	CAS_WRITE_4(sc, CAS_MAC_RX_LEN_ERR_CNT, 0);
636 	CAS_WRITE_4(sc, CAS_MAC_RX_ALIGN_ERR, 0);
637 	CAS_WRITE_4(sc, CAS_MAC_RX_CRC_ERR_CNT, 0);
638 	CAS_WRITE_4(sc, CAS_MAC_RX_CODE_VIOL, 0);
639 
640 	mii_tick(sc->sc_mii);
641 
642 	if (sc->sc_txfree != CAS_MAXTXFREE)
643 		cas_tint(sc);
644 
645 	cas_watchdog(sc);
646 
647 	callout_reset(&sc->sc_tick_ch, hz, cas_tick, sc);
648 }
649 
650 static int
651 cas_bitwait(struct cas_softc *sc, bus_addr_t r, uint32_t clr, uint32_t set)
652 {
653 	int i;
654 	uint32_t reg;
655 
656 	for (i = CAS_TRIES; i--; DELAY(100)) {
657 		reg = CAS_READ_4(sc, r);
658 		if ((reg & clr) == 0 && (reg & set) == set)
659 			return (1);
660 	}
661 	return (0);
662 }
663 
664 static void
665 cas_reset(struct cas_softc *sc)
666 {
667 
668 #ifdef CAS_DEBUG
669 	CTR2(KTR_CAS, "%s: %s", device_get_name(sc->sc_dev), __func__);
670 #endif
671 	/* Disable all interrupts in order to avoid spurious ones. */
672 	CAS_WRITE_4(sc, CAS_INTMASK, 0xffffffff);
673 
674 	cas_reset_rx(sc);
675 	cas_reset_tx(sc);
676 
677 	/*
678 	 * Do a full reset modulo the result of the last auto-negotiation
679 	 * when using the SERDES.
680 	 */
681 	CAS_WRITE_4(sc, CAS_RESET, CAS_RESET_RX | CAS_RESET_TX |
682 	    ((sc->sc_flags & CAS_SERDES) != 0 ? CAS_RESET_PCS_DIS : 0));
683 	CAS_BARRIER(sc, CAS_RESET, 4,
684 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
685 	DELAY(3000);
686 	if (!cas_bitwait(sc, CAS_RESET, CAS_RESET_RX | CAS_RESET_TX, 0))
687 		device_printf(sc->sc_dev, "cannot reset device\n");
688 }
689 
690 static void
691 cas_stop(struct ifnet *ifp)
692 {
693 	struct cas_softc *sc = ifp->if_softc;
694 	struct cas_txsoft *txs;
695 
696 #ifdef CAS_DEBUG
697 	CTR2(KTR_CAS, "%s: %s", device_get_name(sc->sc_dev), __func__);
698 #endif
699 
700 	callout_stop(&sc->sc_tick_ch);
701 	callout_stop(&sc->sc_rx_ch);
702 
703 	/* Disable all interrupts in order to avoid spurious ones. */
704 	CAS_WRITE_4(sc, CAS_INTMASK, 0xffffffff);
705 
706 	cas_reset_tx(sc);
707 	cas_reset_rx(sc);
708 
709 	/*
710 	 * Release any queued transmit buffers.
711 	 */
712 	while ((txs = STAILQ_FIRST(&sc->sc_txdirtyq)) != NULL) {
713 		STAILQ_REMOVE_HEAD(&sc->sc_txdirtyq, txs_q);
714 		if (txs->txs_ndescs != 0) {
715 			bus_dmamap_sync(sc->sc_tdmatag, txs->txs_dmamap,
716 			    BUS_DMASYNC_POSTWRITE);
717 			bus_dmamap_unload(sc->sc_tdmatag, txs->txs_dmamap);
718 			if (txs->txs_mbuf != NULL) {
719 				m_freem(txs->txs_mbuf);
720 				txs->txs_mbuf = NULL;
721 			}
722 		}
723 		STAILQ_INSERT_TAIL(&sc->sc_txfreeq, txs, txs_q);
724 	}
725 
726 	/*
727 	 * Mark the interface down and cancel the watchdog timer.
728 	 */
729 	ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
730 	sc->sc_flags &= ~CAS_LINK;
731 	sc->sc_wdog_timer = 0;
732 }
733 
734 static int
735 cas_reset_rx(struct cas_softc *sc)
736 {
737 
738 	/*
739 	 * Resetting while DMA is in progress can cause a bus hang, so we
740 	 * disable DMA first.
741 	 */
742 	cas_disable_rx(sc);
743 	CAS_WRITE_4(sc, CAS_RX_CONF, 0);
744 	CAS_BARRIER(sc, CAS_RX_CONF, 4,
745 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
746 	if (!cas_bitwait(sc, CAS_RX_CONF, CAS_RX_CONF_RXDMA_EN, 0))
747 		device_printf(sc->sc_dev, "cannot disable RX DMA\n");
748 
749 	/* Finally, reset the ERX. */
750 	CAS_WRITE_4(sc, CAS_RESET, CAS_RESET_RX |
751 	    ((sc->sc_flags & CAS_SERDES) != 0 ? CAS_RESET_PCS_DIS : 0));
752 	CAS_BARRIER(sc, CAS_RESET, 4,
753 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
754 	if (!cas_bitwait(sc, CAS_RESET, CAS_RESET_RX | CAS_RESET_TX, 0)) {
755 		device_printf(sc->sc_dev, "cannot reset receiver\n");
756 		return (1);
757 	}
758 	return (0);
759 }
760 
761 static int
762 cas_reset_tx(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 	cas_disable_tx(sc);
770 	CAS_WRITE_4(sc, CAS_TX_CONF, 0);
771 	CAS_BARRIER(sc, CAS_TX_CONF, 4,
772 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
773 	if (!cas_bitwait(sc, CAS_TX_CONF, CAS_TX_CONF_TXDMA_EN, 0))
774 		device_printf(sc->sc_dev, "cannot disable TX DMA\n");
775 
776 	/* Finally, reset the ETX. */
777 	CAS_WRITE_4(sc, CAS_RESET, CAS_RESET_TX |
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 | CAS_RESET_TX, 0)) {
782 		device_printf(sc->sc_dev, "cannot reset transmitter\n");
783 		return (1);
784 	}
785 	return (0);
786 }
787 
788 static int
789 cas_disable_rx(struct cas_softc *sc)
790 {
791 
792 	CAS_WRITE_4(sc, CAS_MAC_RX_CONF,
793 	    CAS_READ_4(sc, CAS_MAC_RX_CONF) & ~CAS_MAC_RX_CONF_EN);
794 	CAS_BARRIER(sc, CAS_MAC_RX_CONF, 4,
795 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
796 	return (cas_bitwait(sc, CAS_MAC_RX_CONF, CAS_MAC_RX_CONF_EN, 0));
797 }
798 
799 static int
800 cas_disable_tx(struct cas_softc *sc)
801 {
802 
803 	CAS_WRITE_4(sc, CAS_MAC_TX_CONF,
804 	    CAS_READ_4(sc, CAS_MAC_TX_CONF) & ~CAS_MAC_TX_CONF_EN);
805 	CAS_BARRIER(sc, CAS_MAC_TX_CONF, 4,
806 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
807 	return (cas_bitwait(sc, CAS_MAC_TX_CONF, CAS_MAC_TX_CONF_EN, 0));
808 }
809 
810 static inline void
811 cas_rxcompinit(struct cas_rx_comp *rxcomp)
812 {
813 
814 	rxcomp->crc_word1 = 0;
815 	rxcomp->crc_word2 = 0;
816 	rxcomp->crc_word3 =
817 	    htole64(CAS_SET(ETHER_HDR_LEN + sizeof(struct ip), CAS_RC3_CSO));
818 	rxcomp->crc_word4 = htole64(CAS_RC4_ZERO);
819 }
820 
821 static void
822 cas_meminit(struct cas_softc *sc)
823 {
824 	int i;
825 
826 	CAS_LOCK_ASSERT(sc, MA_OWNED);
827 
828 	/*
829 	 * Initialize the transmit descriptor ring.
830 	 */
831 	for (i = 0; i < CAS_NTXDESC; i++) {
832 		sc->sc_txdescs[i].cd_flags = 0;
833 		sc->sc_txdescs[i].cd_buf_ptr = 0;
834 	}
835 	sc->sc_txfree = CAS_MAXTXFREE;
836 	sc->sc_txnext = 0;
837 	sc->sc_txwin = 0;
838 
839 	/*
840 	 * Initialize the receive completion ring.
841 	 */
842 	for (i = 0; i < CAS_NRXCOMP; i++)
843 		cas_rxcompinit(&sc->sc_rxcomps[i]);
844 	sc->sc_rxcptr = 0;
845 
846 	/*
847 	 * Initialize the first receive descriptor ring.  We leave
848 	 * the second one zeroed as we don't actually use it.
849 	 */
850 	for (i = 0; i < CAS_NRXDESC; i++)
851 		CAS_INIT_RXDESC(sc, i, i);
852 	sc->sc_rxdptr = 0;
853 
854 	CAS_CDSYNC(sc, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
855 }
856 
857 static u_int
858 cas_descsize(u_int sz)
859 {
860 
861 	switch (sz) {
862 	case 32:
863 		return (CAS_DESC_32);
864 	case 64:
865 		return (CAS_DESC_64);
866 	case 128:
867 		return (CAS_DESC_128);
868 	case 256:
869 		return (CAS_DESC_256);
870 	case 512:
871 		return (CAS_DESC_512);
872 	case 1024:
873 		return (CAS_DESC_1K);
874 	case 2048:
875 		return (CAS_DESC_2K);
876 	case 4096:
877 		return (CAS_DESC_4K);
878 	case 8192:
879 		return (CAS_DESC_8K);
880 	default:
881 		printf("%s: invalid descriptor ring size %d\n", __func__, sz);
882 		return (CAS_DESC_32);
883 	}
884 }
885 
886 static u_int
887 cas_rxcompsize(u_int sz)
888 {
889 
890 	switch (sz) {
891 	case 128:
892 		return (CAS_RX_CONF_COMP_128);
893 	case 256:
894 		return (CAS_RX_CONF_COMP_256);
895 	case 512:
896 		return (CAS_RX_CONF_COMP_512);
897 	case 1024:
898 		return (CAS_RX_CONF_COMP_1K);
899 	case 2048:
900 		return (CAS_RX_CONF_COMP_2K);
901 	case 4096:
902 		return (CAS_RX_CONF_COMP_4K);
903 	case 8192:
904 		return (CAS_RX_CONF_COMP_8K);
905 	case 16384:
906 		return (CAS_RX_CONF_COMP_16K);
907 	case 32768:
908 		return (CAS_RX_CONF_COMP_32K);
909 	default:
910 		printf("%s: invalid dcompletion ring size %d\n", __func__, sz);
911 		return (CAS_RX_CONF_COMP_128);
912 	}
913 }
914 
915 static void
916 cas_init(void *xsc)
917 {
918 	struct cas_softc *sc = xsc;
919 
920 	CAS_LOCK(sc);
921 	cas_init_locked(sc);
922 	CAS_UNLOCK(sc);
923 }
924 
925 /*
926  * Initialization of interface; set up initialization block
927  * and transmit/receive descriptor rings.
928  */
929 static void
930 cas_init_locked(struct cas_softc *sc)
931 {
932 	struct ifnet *ifp = sc->sc_ifp;
933 	uint32_t v;
934 
935 	CAS_LOCK_ASSERT(sc, MA_OWNED);
936 
937 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
938 		return;
939 
940 #ifdef CAS_DEBUG
941 	CTR2(KTR_CAS, "%s: %s: calling stop", device_get_name(sc->sc_dev),
942 	    __func__);
943 #endif
944 	/*
945 	 * Initialization sequence.  The numbered steps below correspond
946 	 * to the sequence outlined in section 6.3.5.1 in the Ethernet
947 	 * Channel Engine manual (part of the PCIO manual).
948 	 * See also the STP2002-STQ document from Sun Microsystems.
949 	 */
950 
951 	/* step 1 & 2.  Reset the Ethernet Channel. */
952 	cas_stop(ifp);
953 	cas_reset(sc);
954 #ifdef CAS_DEBUG
955 	CTR2(KTR_CAS, "%s: %s: restarting", device_get_name(sc->sc_dev),
956 	    __func__);
957 #endif
958 
959 	/* Re-initialize the MIF. */
960 	cas_mifinit(sc);
961 
962 	/* step 3.  Setup data structures in host memory. */
963 	cas_meminit(sc);
964 
965 	/* step 4.  TX MAC registers & counters */
966 	cas_init_regs(sc);
967 
968 	/* step 5.  RX MAC registers & counters */
969 	cas_setladrf(sc);
970 
971 	/* step 6 & 7.  Program Ring Base Addresses. */
972 	CAS_WRITE_4(sc, CAS_TX_DESC3_BASE_HI,
973 	    (((uint64_t)CAS_CDTXDADDR(sc, 0)) >> 32));
974 	CAS_WRITE_4(sc, CAS_TX_DESC3_BASE_LO,
975 	    CAS_CDTXDADDR(sc, 0) & 0xffffffff);
976 
977 	CAS_WRITE_4(sc, CAS_RX_COMP_BASE_HI,
978 	    (((uint64_t)CAS_CDRXCADDR(sc, 0)) >> 32));
979 	CAS_WRITE_4(sc, CAS_RX_COMP_BASE_LO,
980 	    CAS_CDRXCADDR(sc, 0) & 0xffffffff);
981 
982 	CAS_WRITE_4(sc, CAS_RX_DESC_BASE_HI,
983 	    (((uint64_t)CAS_CDRXDADDR(sc, 0)) >> 32));
984 	CAS_WRITE_4(sc, CAS_RX_DESC_BASE_LO,
985 	    CAS_CDRXDADDR(sc, 0) & 0xffffffff);
986 
987 	if ((sc->sc_flags & CAS_REG_PLUS) != 0) {
988 		CAS_WRITE_4(sc, CAS_RX_DESC2_BASE_HI,
989 		    (((uint64_t)CAS_CDRXD2ADDR(sc, 0)) >> 32));
990 		CAS_WRITE_4(sc, CAS_RX_DESC2_BASE_LO,
991 		    CAS_CDRXD2ADDR(sc, 0) & 0xffffffff);
992 	}
993 
994 #ifdef CAS_DEBUG
995 	CTR5(KTR_CAS,
996 	    "loading TXDR %lx, RXCR %lx, RXDR %lx, RXD2R %lx, cddma %lx",
997 	    CAS_CDTXDADDR(sc, 0), CAS_CDRXCADDR(sc, 0), CAS_CDRXDADDR(sc, 0),
998 	    CAS_CDRXD2ADDR(sc, 0), sc->sc_cddma);
999 #endif
1000 
1001 	/* step 8.  Global Configuration & Interrupt Masks */
1002 
1003 	/* Disable weighted round robin. */
1004 	CAS_WRITE_4(sc, CAS_CAW, CAS_CAW_RR_DIS);
1005 
1006 	/*
1007 	 * Enable infinite bursts for revisions without PCI issues if
1008 	 * applicable.  Doing so greatly improves the TX performance on
1009 	 * !__sparc64__.
1010 	 */
1011 	CAS_WRITE_4(sc, CAS_INF_BURST,
1012 #if !defined(__sparc64__)
1013 	    (sc->sc_flags & CAS_TABORT) == 0 ? CAS_INF_BURST_EN :
1014 #endif
1015 	    0);
1016 
1017 	/* Set up interrupts. */
1018 	CAS_WRITE_4(sc, CAS_INTMASK,
1019 	    ~(CAS_INTR_TX_INT_ME | CAS_INTR_TX_TAG_ERR |
1020 	    CAS_INTR_RX_DONE | CAS_INTR_RX_BUF_NA | CAS_INTR_RX_TAG_ERR |
1021 	    CAS_INTR_RX_COMP_FULL | CAS_INTR_RX_BUF_AEMPTY |
1022 	    CAS_INTR_RX_COMP_AFULL | CAS_INTR_RX_LEN_MMATCH |
1023 	    CAS_INTR_PCI_ERROR_INT
1024 #ifdef CAS_DEBUG
1025 	    | CAS_INTR_PCS_INT | CAS_INTR_MIF
1026 #endif
1027 	    ));
1028 	/* Don't clear top level interrupts when CAS_STATUS_ALIAS is read. */
1029 	CAS_WRITE_4(sc, CAS_CLEAR_ALIAS, 0);
1030 	CAS_WRITE_4(sc, CAS_MAC_RX_MASK, ~CAS_MAC_RX_OVERFLOW);
1031 	CAS_WRITE_4(sc, CAS_MAC_TX_MASK,
1032 	    ~(CAS_MAC_TX_UNDERRUN | CAS_MAC_TX_MAX_PKT_ERR));
1033 #ifdef CAS_DEBUG
1034 	CAS_WRITE_4(sc, CAS_MAC_CTRL_MASK,
1035 	    ~(CAS_MAC_CTRL_PAUSE_RCVD | CAS_MAC_CTRL_PAUSE |
1036 	    CAS_MAC_CTRL_NON_PAUSE));
1037 #else
1038 	CAS_WRITE_4(sc, CAS_MAC_CTRL_MASK,
1039 	    CAS_MAC_CTRL_PAUSE_RCVD | CAS_MAC_CTRL_PAUSE |
1040 	    CAS_MAC_CTRL_NON_PAUSE);
1041 #endif
1042 
1043 	/* Enable PCI error interrupts. */
1044 	CAS_WRITE_4(sc, CAS_ERROR_MASK,
1045 	    ~(CAS_ERROR_DTRTO | CAS_ERROR_OTHER | CAS_ERROR_DMAW_ZERO |
1046 	    CAS_ERROR_DMAR_ZERO | CAS_ERROR_RTRTO));
1047 
1048 	/* Enable PCI error interrupts in BIM configuration. */
1049 	CAS_WRITE_4(sc, CAS_BIM_CONF,
1050 	    CAS_BIM_CONF_DPAR_EN | CAS_BIM_CONF_RMA_EN | CAS_BIM_CONF_RTA_EN);
1051 
1052 	/*
1053 	 * step 9.  ETX Configuration: encode receive descriptor ring size,
1054 	 * enable DMA and disable pre-interrupt writeback completion.
1055 	 */
1056 	v = cas_descsize(CAS_NTXDESC) << CAS_TX_CONF_DESC3_SHFT;
1057 	CAS_WRITE_4(sc, CAS_TX_CONF, v | CAS_TX_CONF_TXDMA_EN |
1058 	    CAS_TX_CONF_RDPP_DIS | CAS_TX_CONF_PICWB_DIS);
1059 
1060 	/* step 10.  ERX Configuration */
1061 
1062 	/*
1063 	 * Encode receive completion and descriptor ring sizes, set the
1064 	 * swivel offset.
1065 	 */
1066 	v = cas_rxcompsize(CAS_NRXCOMP) << CAS_RX_CONF_COMP_SHFT;
1067 	v |= cas_descsize(CAS_NRXDESC) << CAS_RX_CONF_DESC_SHFT;
1068 	if ((sc->sc_flags & CAS_REG_PLUS) != 0)
1069 		v |= cas_descsize(CAS_NRXDESC2) << CAS_RX_CONF_DESC2_SHFT;
1070 	CAS_WRITE_4(sc, CAS_RX_CONF,
1071 	    v | (ETHER_ALIGN << CAS_RX_CONF_SOFF_SHFT));
1072 
1073 	/* Set the PAUSE thresholds.  We use the maximum OFF threshold. */
1074 	CAS_WRITE_4(sc, CAS_RX_PTHRS,
1075 	    ((111 * 64) << CAS_RX_PTHRS_XOFF_SHFT) |
1076 	    ((15 * 64) << CAS_RX_PTHRS_XON_SHFT));
1077 
1078 	/* RX blanking */
1079 	CAS_WRITE_4(sc, CAS_RX_BLANK,
1080 	    (15 << CAS_RX_BLANK_TIME_SHFT) | (5 << CAS_RX_BLANK_PKTS_SHFT));
1081 
1082 	/* Set RX_COMP_AFULL threshold to half of the RX completions. */
1083 	CAS_WRITE_4(sc, CAS_RX_AEMPTY_THRS,
1084 	    (CAS_NRXCOMP / 2) << CAS_RX_AEMPTY_COMP_SHFT);
1085 
1086 	/* Initialize the RX page size register as appropriate for 8k. */
1087 	CAS_WRITE_4(sc, CAS_RX_PSZ,
1088 	    (CAS_RX_PSZ_8K << CAS_RX_PSZ_SHFT) |
1089 	    (4 << CAS_RX_PSZ_MB_CNT_SHFT) |
1090 	    (CAS_RX_PSZ_MB_STRD_2K << CAS_RX_PSZ_MB_STRD_SHFT) |
1091 	    (CAS_RX_PSZ_MB_OFF_64 << CAS_RX_PSZ_MB_OFF_SHFT));
1092 
1093 	/* Disable RX random early detection. */
1094 	CAS_WRITE_4(sc,	CAS_RX_RED, 0);
1095 
1096 	/* Zero the RX reassembly DMA table. */
1097 	for (v = 0; v <= CAS_RX_REAS_DMA_ADDR_LC; v++) {
1098 		CAS_WRITE_4(sc,	CAS_RX_REAS_DMA_ADDR, v);
1099 		CAS_WRITE_4(sc,	CAS_RX_REAS_DMA_DATA_LO, 0);
1100 		CAS_WRITE_4(sc,	CAS_RX_REAS_DMA_DATA_MD, 0);
1101 		CAS_WRITE_4(sc,	CAS_RX_REAS_DMA_DATA_HI, 0);
1102 	}
1103 
1104 	/* Ensure the RX control FIFO and RX IPP FIFO addresses are zero. */
1105 	CAS_WRITE_4(sc, CAS_RX_CTRL_FIFO, 0);
1106 	CAS_WRITE_4(sc, CAS_RX_IPP_ADDR, 0);
1107 
1108 	/* Finally, enable RX DMA. */
1109 	CAS_WRITE_4(sc, CAS_RX_CONF,
1110 	    CAS_READ_4(sc, CAS_RX_CONF) | CAS_RX_CONF_RXDMA_EN);
1111 
1112 	/* step 11.  Configure Media. */
1113 
1114 	/* step 12.  RX_MAC Configuration Register */
1115 	v = CAS_READ_4(sc, CAS_MAC_RX_CONF) & ~CAS_MAC_RX_CONF_STRPPAD;
1116 	v |= CAS_MAC_RX_CONF_EN | CAS_MAC_RX_CONF_STRPFCS;
1117 	CAS_WRITE_4(sc, CAS_MAC_RX_CONF, 0);
1118 	CAS_BARRIER(sc, CAS_MAC_RX_CONF, 4,
1119 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
1120 	if (!cas_bitwait(sc, CAS_MAC_RX_CONF, CAS_MAC_RX_CONF_EN, 0))
1121 		device_printf(sc->sc_dev, "cannot configure RX MAC\n");
1122 	CAS_WRITE_4(sc, CAS_MAC_RX_CONF, v);
1123 
1124 	/* step 13.  TX_MAC Configuration Register */
1125 	v = CAS_READ_4(sc, CAS_MAC_TX_CONF);
1126 	v |= CAS_MAC_TX_CONF_EN;
1127 	CAS_WRITE_4(sc, CAS_MAC_TX_CONF, 0);
1128 	CAS_BARRIER(sc, CAS_MAC_TX_CONF, 4,
1129 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
1130 	if (!cas_bitwait(sc, CAS_MAC_TX_CONF, CAS_MAC_TX_CONF_EN, 0))
1131 		device_printf(sc->sc_dev, "cannot configure TX MAC\n");
1132 	CAS_WRITE_4(sc, CAS_MAC_TX_CONF, v);
1133 
1134 	/* step 14.  Issue Transmit Pending command. */
1135 
1136 	/* step 15.  Give the reciever a swift kick. */
1137 	CAS_WRITE_4(sc, CAS_RX_KICK, CAS_NRXDESC - 4);
1138 	CAS_WRITE_4(sc, CAS_RX_COMP_TAIL, 0);
1139 	if ((sc->sc_flags & CAS_REG_PLUS) != 0)
1140 		CAS_WRITE_4(sc, CAS_RX_KICK2, CAS_NRXDESC2 - 4);
1141 
1142 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1143 	ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1144 
1145 	mii_mediachg(sc->sc_mii);
1146 
1147 	/* Start the one second timer. */
1148 	sc->sc_wdog_timer = 0;
1149 	callout_reset(&sc->sc_tick_ch, hz, cas_tick, sc);
1150 }
1151 
1152 static int
1153 cas_load_txmbuf(struct cas_softc *sc, struct mbuf **m_head)
1154 {
1155 	bus_dma_segment_t txsegs[CAS_NTXSEGS];
1156 	struct cas_txsoft *txs;
1157 	struct ip *ip;
1158 	struct mbuf *m;
1159 	uint64_t cflags;
1160 	int error, nexttx, nsegs, offset, seg;
1161 
1162 	CAS_LOCK_ASSERT(sc, MA_OWNED);
1163 
1164 	/* Get a work queue entry. */
1165 	if ((txs = STAILQ_FIRST(&sc->sc_txfreeq)) == NULL) {
1166 		/* Ran out of descriptors. */
1167 		return (ENOBUFS);
1168 	}
1169 
1170 	cflags = 0;
1171 	if (((*m_head)->m_pkthdr.csum_flags & CAS_CSUM_FEATURES) != 0) {
1172 		if (M_WRITABLE(*m_head) == 0) {
1173 			m = m_dup(*m_head, M_DONTWAIT);
1174 			m_freem(*m_head);
1175 			*m_head = m;
1176 			if (m == NULL)
1177 				return (ENOBUFS);
1178 		}
1179 		offset = sizeof(struct ether_header);
1180 		m = m_pullup(*m_head, offset + sizeof(struct ip));
1181 		if (m == NULL) {
1182 			*m_head = NULL;
1183 			return (ENOBUFS);
1184 		}
1185 		ip = (struct ip *)(mtod(m, caddr_t) + offset);
1186 		offset += (ip->ip_hl << 2);
1187 		cflags = (offset << CAS_TD_CKSUM_START_SHFT) |
1188 		    ((offset + m->m_pkthdr.csum_data) <<
1189 		    CAS_TD_CKSUM_STUFF_SHFT) | CAS_TD_CKSUM_EN;
1190 		*m_head = m;
1191 	}
1192 
1193 	error = bus_dmamap_load_mbuf_sg(sc->sc_tdmatag, txs->txs_dmamap,
1194 	    *m_head, txsegs, &nsegs, BUS_DMA_NOWAIT);
1195 	if (error == EFBIG) {
1196 		m = m_collapse(*m_head, M_DONTWAIT, CAS_NTXSEGS);
1197 		if (m == NULL) {
1198 			m_freem(*m_head);
1199 			*m_head = NULL;
1200 			return (ENOBUFS);
1201 		}
1202 		*m_head = m;
1203 		error = bus_dmamap_load_mbuf_sg(sc->sc_tdmatag,
1204 		    txs->txs_dmamap, *m_head, txsegs, &nsegs,
1205 		    BUS_DMA_NOWAIT);
1206 		if (error != 0) {
1207 			m_freem(*m_head);
1208 			*m_head = NULL;
1209 			return (error);
1210 		}
1211 	} else if (error != 0)
1212 		return (error);
1213 	/* If nsegs is wrong then the stack is corrupt. */
1214 	KASSERT(nsegs <= CAS_NTXSEGS,
1215 	    ("%s: too many DMA segments (%d)", __func__, nsegs));
1216 	if (nsegs == 0) {
1217 		m_freem(*m_head);
1218 		*m_head = NULL;
1219 		return (EIO);
1220 	}
1221 
1222 	/*
1223 	 * Ensure we have enough descriptors free to describe
1224 	 * the packet.  Note, we always reserve one descriptor
1225 	 * at the end of the ring as a termination point, in
1226 	 * order to prevent wrap-around.
1227 	 */
1228 	if (nsegs > sc->sc_txfree - 1) {
1229 		txs->txs_ndescs = 0;
1230 		bus_dmamap_unload(sc->sc_tdmatag, txs->txs_dmamap);
1231 		return (ENOBUFS);
1232 	}
1233 
1234 	txs->txs_ndescs = nsegs;
1235 	txs->txs_firstdesc = sc->sc_txnext;
1236 	nexttx = txs->txs_firstdesc;
1237 	for (seg = 0; seg < nsegs; seg++, nexttx = CAS_NEXTTX(nexttx)) {
1238 #ifdef CAS_DEBUG
1239 		CTR6(KTR_CAS,
1240 		    "%s: mapping seg %d (txd %d), len %lx, addr %#lx (%#lx)",
1241 		    __func__, seg, nexttx, txsegs[seg].ds_len,
1242 		    txsegs[seg].ds_addr, htole64(txsegs[seg].ds_addr));
1243 #endif
1244 		sc->sc_txdescs[nexttx].cd_buf_ptr =
1245 		    htole64(txsegs[seg].ds_addr);
1246 		KASSERT(txsegs[seg].ds_len <
1247 		    CAS_TD_BUF_LEN_MASK >> CAS_TD_BUF_LEN_SHFT,
1248 		    ("%s: segment size too large!", __func__));
1249 		sc->sc_txdescs[nexttx].cd_flags =
1250 		    htole64(txsegs[seg].ds_len << CAS_TD_BUF_LEN_SHFT);
1251 		txs->txs_lastdesc = nexttx;
1252 	}
1253 
1254 	/* Set EOF on the last descriptor. */
1255 #ifdef CAS_DEBUG
1256 	CTR3(KTR_CAS, "%s: end of frame at segment %d, TX %d",
1257 	    __func__, seg, nexttx);
1258 #endif
1259 	sc->sc_txdescs[txs->txs_lastdesc].cd_flags |=
1260 	    htole64(CAS_TD_END_OF_FRAME);
1261 
1262 	/* Lastly set SOF on the first descriptor. */
1263 #ifdef CAS_DEBUG
1264 	CTR3(KTR_CAS, "%s: start of frame at segment %d, TX %d",
1265 	    __func__, seg, nexttx);
1266 #endif
1267 	if (sc->sc_txwin += nsegs > CAS_MAXTXFREE * 2 / 3) {
1268 		sc->sc_txwin = 0;
1269 		sc->sc_txdescs[txs->txs_firstdesc].cd_flags |=
1270 		    htole64(cflags | CAS_TD_START_OF_FRAME | CAS_TD_INT_ME);
1271 	} else
1272 		sc->sc_txdescs[txs->txs_firstdesc].cd_flags |=
1273 		    htole64(cflags | CAS_TD_START_OF_FRAME);
1274 
1275 	/* Sync the DMA map. */
1276 	bus_dmamap_sync(sc->sc_tdmatag, txs->txs_dmamap,
1277 	    BUS_DMASYNC_PREWRITE);
1278 
1279 #ifdef CAS_DEBUG
1280 	CTR4(KTR_CAS, "%s: setting firstdesc=%d, lastdesc=%d, ndescs=%d",
1281 	    __func__, txs->txs_firstdesc, txs->txs_lastdesc,
1282 	    txs->txs_ndescs);
1283 #endif
1284 	STAILQ_REMOVE_HEAD(&sc->sc_txfreeq, txs_q);
1285 	STAILQ_INSERT_TAIL(&sc->sc_txdirtyq, txs, txs_q);
1286 	txs->txs_mbuf = *m_head;
1287 
1288 	sc->sc_txnext = CAS_NEXTTX(txs->txs_lastdesc);
1289 	sc->sc_txfree -= txs->txs_ndescs;
1290 
1291 	return (0);
1292 }
1293 
1294 static void
1295 cas_init_regs(struct cas_softc *sc)
1296 {
1297 	int i;
1298 	const u_char *laddr = IF_LLADDR(sc->sc_ifp);
1299 
1300 	CAS_LOCK_ASSERT(sc, MA_OWNED);
1301 
1302 	/* These registers are not cleared on reset. */
1303 	if ((sc->sc_flags & CAS_INITED) == 0) {
1304 		/* magic values */
1305 		CAS_WRITE_4(sc, CAS_MAC_IPG0, 0);
1306 		CAS_WRITE_4(sc, CAS_MAC_IPG1, 8);
1307 		CAS_WRITE_4(sc, CAS_MAC_IPG2, 4);
1308 
1309 		/* min frame length */
1310 		CAS_WRITE_4(sc, CAS_MAC_MIN_FRAME, ETHER_MIN_LEN);
1311 		/* max frame length and max burst size */
1312 		CAS_WRITE_4(sc, CAS_MAC_MAX_BF,
1313 		    ((ETHER_MAX_LEN_JUMBO + ETHER_VLAN_ENCAP_LEN) <<
1314 		    CAS_MAC_MAX_BF_FRM_SHFT) |
1315 		    (0x2000 << CAS_MAC_MAX_BF_BST_SHFT));
1316 
1317 		/* more magic values */
1318 		CAS_WRITE_4(sc, CAS_MAC_PREAMBLE_LEN, 0x7);
1319 		CAS_WRITE_4(sc, CAS_MAC_JAM_SIZE, 0x4);
1320 		CAS_WRITE_4(sc, CAS_MAC_ATTEMPT_LIMIT, 0x10);
1321 		CAS_WRITE_4(sc, CAS_MAC_CTRL_TYPE, 0x8088);
1322 
1323 		/* random number seed */
1324 		CAS_WRITE_4(sc, CAS_MAC_RANDOM_SEED,
1325 		    ((laddr[5] << 8) | laddr[4]) & 0x3ff);
1326 
1327 		/* secondary MAC addresses: 0:0:0:0:0:0 */
1328 		for (i = CAS_MAC_ADDR3; i <= CAS_MAC_ADDR41;
1329 		    i += CAS_MAC_ADDR4 - CAS_MAC_ADDR3)
1330 			CAS_WRITE_4(sc, i, 0);
1331 
1332 		/* MAC control address: 01:80:c2:00:00:01 */
1333 		CAS_WRITE_4(sc, CAS_MAC_ADDR42, 0x0001);
1334 		CAS_WRITE_4(sc, CAS_MAC_ADDR43, 0xc200);
1335 		CAS_WRITE_4(sc, CAS_MAC_ADDR44, 0x0180);
1336 
1337 		/* MAC filter address: 0:0:0:0:0:0 */
1338 		CAS_WRITE_4(sc, CAS_MAC_AFILTER0, 0);
1339 		CAS_WRITE_4(sc, CAS_MAC_AFILTER1, 0);
1340 		CAS_WRITE_4(sc, CAS_MAC_AFILTER2, 0);
1341 		CAS_WRITE_4(sc, CAS_MAC_AFILTER_MASK1_2, 0);
1342 		CAS_WRITE_4(sc, CAS_MAC_AFILTER_MASK0, 0);
1343 
1344 		/* Zero the hash table. */
1345 		for (i = CAS_MAC_HASH0; i <= CAS_MAC_HASH15;
1346 		    i += CAS_MAC_HASH1 - CAS_MAC_HASH0)
1347 			CAS_WRITE_4(sc, i, 0);
1348 
1349 		sc->sc_flags |= CAS_INITED;
1350 	}
1351 
1352 	/* Counters need to be zeroed. */
1353 	CAS_WRITE_4(sc, CAS_MAC_NORM_COLL_CNT, 0);
1354 	CAS_WRITE_4(sc, CAS_MAC_FIRST_COLL_CNT, 0);
1355 	CAS_WRITE_4(sc, CAS_MAC_EXCESS_COLL_CNT, 0);
1356 	CAS_WRITE_4(sc, CAS_MAC_LATE_COLL_CNT, 0);
1357 	CAS_WRITE_4(sc, CAS_MAC_DEFER_TMR_CNT, 0);
1358 	CAS_WRITE_4(sc, CAS_MAC_PEAK_ATTEMPTS, 0);
1359 	CAS_WRITE_4(sc, CAS_MAC_RX_FRAME_COUNT, 0);
1360 	CAS_WRITE_4(sc, CAS_MAC_RX_LEN_ERR_CNT, 0);
1361 	CAS_WRITE_4(sc, CAS_MAC_RX_ALIGN_ERR, 0);
1362 	CAS_WRITE_4(sc, CAS_MAC_RX_CRC_ERR_CNT, 0);
1363 	CAS_WRITE_4(sc, CAS_MAC_RX_CODE_VIOL, 0);
1364 
1365 	/* Set XOFF PAUSE time. */
1366 	CAS_WRITE_4(sc, CAS_MAC_SPC, 0x1BF0 << CAS_MAC_SPC_TIME_SHFT);
1367 
1368 	/* Set the station address. */
1369 	CAS_WRITE_4(sc, CAS_MAC_ADDR0, (laddr[4] << 8) | laddr[5]);
1370 	CAS_WRITE_4(sc, CAS_MAC_ADDR1, (laddr[2] << 8) | laddr[3]);
1371 	CAS_WRITE_4(sc, CAS_MAC_ADDR2, (laddr[0] << 8) | laddr[1]);
1372 
1373 	/* Enable MII outputs. */
1374 	CAS_WRITE_4(sc, CAS_MAC_XIF_CONF, CAS_MAC_XIF_CONF_TX_OE);
1375 }
1376 
1377 static void
1378 cas_tx_task(void *arg, int pending __unused)
1379 {
1380 	struct ifnet *ifp;
1381 
1382 	ifp = (struct ifnet *)arg;
1383 	cas_start(ifp);
1384 }
1385 
1386 static inline void
1387 cas_txkick(struct cas_softc *sc)
1388 {
1389 
1390 	/*
1391 	 * Update the TX kick register.  This register has to point to the
1392 	 * descriptor after the last valid one and for optimum performance
1393 	 * should be incremented in multiples of 4 (the DMA engine fetches/
1394 	 * updates descriptors in batches of 4).
1395 	 */
1396 #ifdef CAS_DEBUG
1397 	CTR3(KTR_CAS, "%s: %s: kicking TX %d",
1398 	    device_get_name(sc->sc_dev), __func__, sc->sc_txnext);
1399 #endif
1400 	CAS_CDSYNC(sc, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1401 	CAS_WRITE_4(sc, CAS_TX_KICK3, sc->sc_txnext);
1402 }
1403 
1404 static void
1405 cas_start(struct ifnet *ifp)
1406 {
1407 	struct cas_softc *sc = ifp->if_softc;
1408 	struct mbuf *m;
1409 	int kicked, ntx;
1410 
1411 	CAS_LOCK(sc);
1412 
1413 	if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
1414 	    IFF_DRV_RUNNING || (sc->sc_flags & CAS_LINK) == 0) {
1415 		CAS_UNLOCK(sc);
1416 		return;
1417 	}
1418 
1419 	if (sc->sc_txfree < CAS_MAXTXFREE / 4)
1420 		cas_tint(sc);
1421 
1422 #ifdef CAS_DEBUG
1423 	CTR4(KTR_CAS, "%s: %s: txfree %d, txnext %d",
1424 	    device_get_name(sc->sc_dev), __func__, sc->sc_txfree,
1425 	    sc->sc_txnext);
1426 #endif
1427 	ntx = 0;
1428 	kicked = 0;
1429 	for (; !IFQ_DRV_IS_EMPTY(&ifp->if_snd) && sc->sc_txfree > 1;) {
1430 		IFQ_DRV_DEQUEUE(&ifp->if_snd, m);
1431 		if (m == NULL)
1432 			break;
1433 		if (cas_load_txmbuf(sc, &m) != 0) {
1434 			if (m == NULL)
1435 				break;
1436 			ifp->if_drv_flags |= IFF_DRV_OACTIVE;
1437 			IFQ_DRV_PREPEND(&ifp->if_snd, m);
1438 			break;
1439 		}
1440 		if ((sc->sc_txnext % 4) == 0) {
1441 			cas_txkick(sc);
1442 			kicked = 1;
1443 		} else
1444 			kicked = 0;
1445 		ntx++;
1446 		BPF_MTAP(ifp, m);
1447 	}
1448 
1449 	if (ntx > 0) {
1450 		if (kicked == 0)
1451 			cas_txkick(sc);
1452 #ifdef CAS_DEBUG
1453 		CTR2(KTR_CAS, "%s: packets enqueued, OWN on %d",
1454 		    device_get_name(sc->sc_dev), sc->sc_txnext);
1455 #endif
1456 
1457 		/* Set a watchdog timer in case the chip flakes out. */
1458 		sc->sc_wdog_timer = 5;
1459 #ifdef CAS_DEBUG
1460 		CTR3(KTR_CAS, "%s: %s: watchdog %d",
1461 		    device_get_name(sc->sc_dev), __func__,
1462 		    sc->sc_wdog_timer);
1463 #endif
1464 	}
1465 
1466 	CAS_UNLOCK(sc);
1467 }
1468 
1469 static void
1470 cas_tint(struct cas_softc *sc)
1471 {
1472 	struct ifnet *ifp = sc->sc_ifp;
1473 	struct cas_txsoft *txs;
1474 	int progress;
1475 	uint32_t txlast;
1476 #ifdef CAS_DEBUG
1477 	int i;
1478 
1479 	CAS_LOCK_ASSERT(sc, MA_OWNED);
1480 
1481 	CTR2(KTR_CAS, "%s: %s", device_get_name(sc->sc_dev), __func__);
1482 #endif
1483 
1484 	/*
1485 	 * Go through our TX list and free mbufs for those
1486 	 * frames that have been transmitted.
1487 	 */
1488 	progress = 0;
1489 	CAS_CDSYNC(sc, BUS_DMASYNC_POSTREAD);
1490 	while ((txs = STAILQ_FIRST(&sc->sc_txdirtyq)) != NULL) {
1491 #ifdef CAS_DEBUG
1492 		if ((ifp->if_flags & IFF_DEBUG) != 0) {
1493 			printf("    txsoft %p transmit chain:\n", txs);
1494 			for (i = txs->txs_firstdesc;; i = CAS_NEXTTX(i)) {
1495 				printf("descriptor %d: ", i);
1496 				printf("cd_flags: 0x%016llx\t",
1497 				    (long long)le64toh(
1498 				    sc->sc_txdescs[i].cd_flags));
1499 				printf("cd_buf_ptr: 0x%016llx\n",
1500 				    (long long)le64toh(
1501 				    sc->sc_txdescs[i].cd_buf_ptr));
1502 				if (i == txs->txs_lastdesc)
1503 					break;
1504 			}
1505 		}
1506 #endif
1507 
1508 		/*
1509 		 * In theory, we could harvest some descriptors before
1510 		 * the ring is empty, but that's a bit complicated.
1511 		 *
1512 		 * CAS_TX_COMPn points to the last descriptor
1513 		 * processed + 1.
1514 		 */
1515 		txlast = CAS_READ_4(sc, CAS_TX_COMP3);
1516 #ifdef CAS_DEBUG
1517 		CTR4(KTR_CAS, "%s: txs->txs_firstdesc = %d, "
1518 		    "txs->txs_lastdesc = %d, txlast = %d",
1519 		    __func__, txs->txs_firstdesc, txs->txs_lastdesc, txlast);
1520 #endif
1521 		if (txs->txs_firstdesc <= txs->txs_lastdesc) {
1522 			if ((txlast >= txs->txs_firstdesc) &&
1523 			    (txlast <= txs->txs_lastdesc))
1524 				break;
1525 		} else {
1526 			/* Ick -- this command wraps. */
1527 			if ((txlast >= txs->txs_firstdesc) ||
1528 			    (txlast <= txs->txs_lastdesc))
1529 				break;
1530 		}
1531 
1532 #ifdef CAS_DEBUG
1533 		CTR1(KTR_CAS, "%s: releasing a descriptor", __func__);
1534 #endif
1535 		STAILQ_REMOVE_HEAD(&sc->sc_txdirtyq, txs_q);
1536 
1537 		sc->sc_txfree += txs->txs_ndescs;
1538 
1539 		bus_dmamap_sync(sc->sc_tdmatag, txs->txs_dmamap,
1540 		    BUS_DMASYNC_POSTWRITE);
1541 		bus_dmamap_unload(sc->sc_tdmatag, txs->txs_dmamap);
1542 		if (txs->txs_mbuf != NULL) {
1543 			m_freem(txs->txs_mbuf);
1544 			txs->txs_mbuf = NULL;
1545 		}
1546 
1547 		STAILQ_INSERT_TAIL(&sc->sc_txfreeq, txs, txs_q);
1548 
1549 		ifp->if_opackets++;
1550 		progress = 1;
1551 	}
1552 
1553 #ifdef CAS_DEBUG
1554 	CTR4(KTR_CAS, "%s: CAS_TX_STATE_MACHINE %x CAS_TX_DESC_BASE %llx "
1555 	    "CAS_TX_COMP3 %x",
1556 	    __func__, CAS_READ_4(sc, CAS_TX_STATE_MACHINE),
1557 	    ((long long)CAS_READ_4(sc, CAS_TX_DESC_BASE_HI3) << 32) |
1558 	    CAS_READ_4(sc, CAS_TX_DESC_BASE_LO3),
1559 	    CAS_READ_4(sc, CAS_TX_COMP3));
1560 #endif
1561 
1562 	if (progress) {
1563 		/* We freed some descriptors, so reset IFF_DRV_OACTIVE. */
1564 		ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1565 		if (STAILQ_EMPTY(&sc->sc_txdirtyq))
1566 			sc->sc_wdog_timer = 0;
1567 	}
1568 
1569 #ifdef CAS_DEBUG
1570 	CTR3(KTR_CAS, "%s: %s: watchdog %d",
1571 	    device_get_name(sc->sc_dev), __func__, sc->sc_wdog_timer);
1572 #endif
1573 }
1574 
1575 static void
1576 cas_rint_timeout(void *arg)
1577 {
1578 	struct cas_softc *sc = arg;
1579 
1580 	CAS_LOCK_ASSERT(sc, MA_NOTOWNED);
1581 
1582 	cas_rint(sc);
1583 }
1584 
1585 static void
1586 cas_rint(struct cas_softc *sc)
1587 {
1588 	struct cas_rxdsoft *rxds, *rxds2;
1589 	struct ifnet *ifp = sc->sc_ifp;
1590 	struct mbuf *m, *m2;
1591 	uint64_t word1, word2, word3, word4;
1592 	uint32_t rxhead;
1593 	u_int idx, idx2, len, off, skip;
1594 
1595 	CAS_LOCK_ASSERT(sc, MA_NOTOWNED);
1596 
1597 	callout_stop(&sc->sc_rx_ch);
1598 
1599 #ifdef CAS_DEBUG
1600 	CTR2(KTR_CAS, "%s: %s", device_get_name(sc->sc_dev), __func__);
1601 #endif
1602 
1603 #define	PRINTWORD(n, delimiter)						\
1604 	printf("word ## n: 0x%016llx%c", (long long)word ## n, delimiter)
1605 
1606 #define	SKIPASSERT(n)							\
1607 	KASSERT(sc->sc_rxcomps[sc->sc_rxcptr].crc_word ## n == 0,	\
1608 	    ("%s: word ## n not 0", __func__))
1609 
1610 #define	WORDTOH(n)							\
1611 	word ## n = le64toh(sc->sc_rxcomps[sc->sc_rxcptr].crc_word ## n)
1612 
1613 	/*
1614 	 * Read the completion head register once.  This limits
1615 	 * how long the following loop can execute.
1616 	 */
1617 	rxhead = CAS_READ_4(sc, CAS_RX_COMP_HEAD);
1618 #ifdef CAS_DEBUG
1619 	CTR4(KTR_CAS, "%s: sc->sc_rxcptr %d, sc->sc_rxdptr %d, head %d",
1620 	    __func__, sc->rxcptr, sc->sc_rxdptr, rxhead);
1621 #endif
1622 	skip = 0;
1623 	CAS_CDSYNC(sc, BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
1624 	for (; sc->sc_rxcptr != rxhead;
1625 	    sc->sc_rxcptr = CAS_NEXTRXCOMP(sc->sc_rxcptr)) {
1626 		if (skip != 0) {
1627 			SKIPASSERT(1);
1628 			SKIPASSERT(2);
1629 			SKIPASSERT(3);
1630 
1631 			--skip;
1632 			goto skip;
1633 		}
1634 
1635 		WORDTOH(1);
1636 		WORDTOH(2);
1637 		WORDTOH(3);
1638 		WORDTOH(4);
1639 
1640 #ifdef CAS_DEBUG
1641 		if ((ifp->if_flags & IFF_DEBUG) != 0) {
1642 			printf("    completion %d: ", sc->sc_rxcptr);
1643 			PRINTWORD(1, '\t');
1644 			PRINTWORD(2, '\t');
1645 			PRINTWORD(3, '\t');
1646 			PRINTWORD(4, '\n');
1647 		}
1648 #endif
1649 
1650 		if (__predict_false(
1651 		    (word1 & CAS_RC1_TYPE_MASK) == CAS_RC1_TYPE_HW ||
1652 		    (word4 & CAS_RC4_ZERO) != 0)) {
1653 			/*
1654 			 * The descriptor is still marked as owned, although
1655 			 * it is supposed to have completed.  This has been
1656 			 * observed on some machines.  Just exiting here
1657 			 * might leave the packet sitting around until another
1658 			 * one arrives to trigger a new interrupt, which is
1659 			 * generally undesirable, so set up a timeout.
1660 			 */
1661 			callout_reset(&sc->sc_rx_ch, CAS_RXOWN_TICKS,
1662 			    cas_rint_timeout, sc);
1663 			break;
1664 		}
1665 
1666 		if (__predict_false(
1667 		    (word4 & (CAS_RC4_BAD | CAS_RC4_LEN_MMATCH)) != 0)) {
1668 			ifp->if_ierrors++;
1669 			device_printf(sc->sc_dev,
1670 			    "receive error: CRC error\n");
1671 			continue;
1672 		}
1673 
1674 		KASSERT(CAS_GET(word1, CAS_RC1_DATA_SIZE) == 0 ||
1675 		    CAS_GET(word2, CAS_RC2_HDR_SIZE) == 0,
1676 		    ("%s: data and header present", __func__));
1677 		KASSERT((word1 & CAS_RC1_SPLIT_PKT) == 0 ||
1678 		    CAS_GET(word2, CAS_RC2_HDR_SIZE) == 0,
1679 		    ("%s: split and header present", __func__));
1680 		KASSERT(CAS_GET(word1, CAS_RC1_DATA_SIZE) == 0 ||
1681 		    (word1 & CAS_RC1_RELEASE_HDR) == 0,
1682 		    ("%s: data present but header release", __func__));
1683 		KASSERT(CAS_GET(word2, CAS_RC2_HDR_SIZE) == 0 ||
1684 		    (word1 & CAS_RC1_RELEASE_DATA) == 0,
1685 		    ("%s: header present but data release", __func__));
1686 
1687 		if ((len = CAS_GET(word2, CAS_RC2_HDR_SIZE)) != 0) {
1688 			idx = CAS_GET(word2, CAS_RC2_HDR_INDEX);
1689 			off = CAS_GET(word2, CAS_RC2_HDR_OFF);
1690 #ifdef CAS_DEBUG
1691 			CTR4(KTR_CAS, "%s: hdr at idx %d, off %d, len %d",
1692 			    __func__, idx, off, len);
1693 #endif
1694 			rxds = &sc->sc_rxdsoft[idx];
1695 			MGETHDR(m, M_DONTWAIT, MT_DATA);
1696 			if (m != NULL) {
1697 				refcount_acquire(&rxds->rxds_refcount);
1698 				bus_dmamap_sync(sc->sc_rdmatag,
1699 				    rxds->rxds_dmamap, BUS_DMASYNC_POSTREAD);
1700 #if __FreeBSD_version < 800016
1701 				MEXTADD(m, (caddr_t)rxds->rxds_buf +
1702 				    off * 256 + ETHER_ALIGN, len, cas_free,
1703 				    rxds, M_RDONLY, EXT_NET_DRV);
1704 #else
1705 				MEXTADD(m, (caddr_t)rxds->rxds_buf +
1706 				    off * 256 + ETHER_ALIGN, len, cas_free,
1707 				    sc, (void *)(uintptr_t)idx,
1708 				    M_RDONLY, EXT_NET_DRV);
1709 #endif
1710 				if ((m->m_flags & M_EXT) == 0) {
1711 					m_freem(m);
1712 					m = NULL;
1713 				}
1714 			}
1715 			if (m != NULL) {
1716 				m->m_pkthdr.rcvif = ifp;
1717 				m->m_pkthdr.len = m->m_len = len;
1718 				ifp->if_ipackets++;
1719 				if ((ifp->if_capenable & IFCAP_RXCSUM) != 0)
1720 					cas_rxcksum(m, CAS_GET(word4,
1721 					    CAS_RC4_TCP_CSUM));
1722 				/* Pass it on. */
1723 				(*ifp->if_input)(ifp, m);
1724 			} else
1725 				ifp->if_ierrors++;
1726 
1727 			if ((word1 & CAS_RC1_RELEASE_HDR) != 0 &&
1728 			    refcount_release(&rxds->rxds_refcount) != 0)
1729 				cas_add_rxdesc(sc, idx);
1730 		} else if ((len = CAS_GET(word1, CAS_RC1_DATA_SIZE)) != 0) {
1731 			idx = CAS_GET(word1, CAS_RC1_DATA_INDEX);
1732 			off = CAS_GET(word1, CAS_RC1_DATA_OFF);
1733 #ifdef CAS_DEBUG
1734 			CTR4(KTR_CAS, "%s: data at idx %d, off %d, len %d",
1735 			    __func__, idx, off, len);
1736 #endif
1737 			rxds = &sc->sc_rxdsoft[idx];
1738 			MGETHDR(m, M_DONTWAIT, MT_DATA);
1739 			if (m != NULL) {
1740 				refcount_acquire(&rxds->rxds_refcount);
1741 				off += ETHER_ALIGN;
1742 				m->m_len = min(CAS_PAGE_SIZE - off, len);
1743 				bus_dmamap_sync(sc->sc_rdmatag,
1744 				    rxds->rxds_dmamap, BUS_DMASYNC_POSTREAD);
1745 #if __FreeBSD_version < 800016
1746 				MEXTADD(m, (caddr_t)rxds->rxds_buf + off,
1747 				    m->m_len, cas_free, rxds, M_RDONLY,
1748 				    EXT_NET_DRV);
1749 #else
1750 				MEXTADD(m, (caddr_t)rxds->rxds_buf + off,
1751 				    m->m_len, cas_free, sc,
1752 				    (void *)(uintptr_t)idx, M_RDONLY,
1753 				    EXT_NET_DRV);
1754 #endif
1755 				if ((m->m_flags & M_EXT) == 0) {
1756 					m_freem(m);
1757 					m = NULL;
1758 				}
1759 			}
1760 			idx2 = 0;
1761 			rxds2 = NULL;
1762 			if ((word1 & CAS_RC1_SPLIT_PKT) != 0) {
1763 				KASSERT((word1 & CAS_RC1_RELEASE_NEXT) != 0,
1764 				    ("%s: split but no release next",
1765 				    __func__));
1766 
1767 				idx2 = CAS_GET(word2, CAS_RC2_NEXT_INDEX);
1768 #ifdef CAS_DEBUG
1769 				CTR2(KTR_CAS, "%s: split at idx %d",
1770 				    __func__, idx2);
1771 #endif
1772 				rxds2 = &sc->sc_rxdsoft[idx2];
1773 				MGET(m2, M_DONTWAIT, MT_DATA);
1774 				if (m2 != NULL) {
1775 					refcount_acquire(
1776 					    &rxds2->rxds_refcount);
1777 					m2->m_len = len - m->m_len;
1778 					bus_dmamap_sync(sc->sc_rdmatag,
1779 					    rxds2->rxds_dmamap,
1780 					    BUS_DMASYNC_POSTREAD);
1781 #if __FreeBSD_version < 800016
1782 					MEXTADD(m2, (caddr_t)rxds2->rxds_buf,
1783 					    m2->m_len, cas_free, rxds2,
1784 					    M_RDONLY, EXT_NET_DRV);
1785 #else
1786 					MEXTADD(m2, (caddr_t)rxds2->rxds_buf,
1787 					    m2->m_len, cas_free,
1788 					    sc, (void *)(uintptr_t)idx2,
1789 					    M_RDONLY, EXT_NET_DRV);
1790 #endif
1791 					if ((m2->m_flags & M_EXT) == 0) {
1792 						m_freem(m2);
1793 						m2 = NULL;
1794 					}
1795 				}
1796 				if (m2 != NULL)
1797 					m->m_next = m2;
1798 				else {
1799 					m_freem(m);
1800 					m = NULL;
1801 				}
1802 			}
1803 			if (m != NULL) {
1804 				m->m_pkthdr.rcvif = ifp;
1805 				m->m_pkthdr.len = len;
1806 				ifp->if_ipackets++;
1807 				if ((ifp->if_capenable & IFCAP_RXCSUM) != 0)
1808 					cas_rxcksum(m, CAS_GET(word4,
1809 					    CAS_RC4_TCP_CSUM));
1810 				/* Pass it on. */
1811 				(*ifp->if_input)(ifp, m);
1812 			} else
1813 				ifp->if_ierrors++;
1814 
1815 			if ((word1 & CAS_RC1_RELEASE_DATA) != 0 &&
1816 			    refcount_release(&rxds->rxds_refcount) != 0)
1817 				cas_add_rxdesc(sc, idx);
1818 			if ((word1 & CAS_RC1_SPLIT_PKT) != 0 &&
1819 			    refcount_release(&rxds2->rxds_refcount) != 0)
1820 				cas_add_rxdesc(sc, idx2);
1821 		}
1822 
1823 		skip = CAS_GET(word1, CAS_RC1_SKIP);
1824 
1825  skip:
1826 		cas_rxcompinit(&sc->sc_rxcomps[sc->sc_rxcptr]);
1827 		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1828 			break;
1829 	}
1830 	CAS_CDSYNC(sc, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1831 	CAS_WRITE_4(sc, CAS_RX_COMP_TAIL, sc->sc_rxcptr);
1832 
1833 #undef PRINTWORD
1834 #undef SKIPASSERT
1835 #undef WORDTOH
1836 
1837 #ifdef CAS_DEBUG
1838 	CTR4(KTR_CAS, "%s: done sc->sc_rxcptr %d, sc->sc_rxdptr %d, head %d",
1839 	    __func__, sc->rxcptr, sc->sc_rxdptr,
1840 	    CAS_READ_4(sc, CAS_RX_COMP_HEAD));
1841 #endif
1842 }
1843 
1844 static void
1845 cas_free(void *arg1, void *arg2)
1846 {
1847 	struct cas_rxdsoft *rxds;
1848 	struct cas_softc *sc;
1849 	u_int idx;
1850 
1851 #if __FreeBSD_version < 800016
1852 	rxds = arg2;
1853 	sc = rxds->rxds_sc;
1854 	idx = rxds->rxds_idx;
1855 #else
1856 	sc = arg1;
1857 	idx = (uintptr_t)arg2;
1858 	rxds = &sc->sc_rxdsoft[idx];
1859 #endif
1860 	if (refcount_release(&rxds->rxds_refcount) == 0)
1861 		return;
1862 
1863 	/*
1864 	 * NB: this function can be called via m_freem(9) within
1865 	 * this driver!
1866 	 */
1867 
1868 	cas_add_rxdesc(sc, idx);
1869 }
1870 
1871 static inline void
1872 cas_add_rxdesc(struct cas_softc *sc, u_int idx)
1873 {
1874 	u_int locked;
1875 
1876 	if ((locked = CAS_LOCK_OWNED(sc)) == 0)
1877 		CAS_LOCK(sc);
1878 
1879 	bus_dmamap_sync(sc->sc_rdmatag, sc->sc_rxdsoft[idx].rxds_dmamap,
1880 	    BUS_DMASYNC_PREREAD);
1881 	CAS_UPDATE_RXDESC(sc, sc->sc_rxdptr, idx);
1882 	sc->sc_rxdptr = CAS_NEXTRXDESC(sc->sc_rxdptr);
1883 
1884 	/*
1885 	 * Update the RX kick register.  This register has to point to the
1886 	 * descriptor after the last valid one (before the current batch)
1887 	 * and for optimum performance should be incremented in multiples
1888 	 * of 4 (the DMA engine fetches/updates descriptors in batches of 4).
1889 	 */
1890 	if ((sc->sc_rxdptr % 4) == 0) {
1891 		CAS_CDSYNC(sc, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1892 		CAS_WRITE_4(sc, CAS_RX_KICK,
1893 		    (sc->sc_rxdptr + CAS_NRXDESC - 4) & CAS_NRXDESC_MASK);
1894 	}
1895 
1896 	if (locked == 0)
1897 		CAS_UNLOCK(sc);
1898 }
1899 
1900 static void
1901 cas_eint(struct cas_softc *sc, u_int status)
1902 {
1903 	struct ifnet *ifp = sc->sc_ifp;
1904 
1905 	CAS_LOCK_ASSERT(sc, MA_NOTOWNED);
1906 
1907 	ifp->if_ierrors++;
1908 
1909 	device_printf(sc->sc_dev, "%s: status 0x%x", __func__, status);
1910 	if ((status & CAS_INTR_PCI_ERROR_INT) != 0) {
1911 		status = CAS_READ_4(sc, CAS_ERROR_STATUS);
1912 		printf(", PCI bus error 0x%x", status);
1913 		if ((status & CAS_ERROR_OTHER) != 0) {
1914 			status = pci_read_config(sc->sc_dev, PCIR_STATUS, 2);
1915 			printf(", PCI status 0x%x", status);
1916 			pci_write_config(sc->sc_dev, PCIR_STATUS, status, 2);
1917 		}
1918 	}
1919 	printf("\n");
1920 
1921 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1922 	cas_init(sc);
1923 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
1924 		taskqueue_enqueue(sc->sc_tq, &sc->sc_tx_task);
1925 }
1926 
1927 static int
1928 cas_intr(void *v)
1929 {
1930 	struct cas_softc *sc = v;
1931 
1932 	if (__predict_false((CAS_READ_4(sc, CAS_STATUS_ALIAS) &
1933 	    CAS_INTR_SUMMARY) == 0))
1934 		return (FILTER_STRAY);
1935 
1936 	/* Disable interrupts. */
1937 	CAS_WRITE_4(sc, CAS_INTMASK, 0xffffffff);
1938 	taskqueue_enqueue(sc->sc_tq, &sc->sc_intr_task);
1939 
1940 	return (FILTER_HANDLED);
1941 }
1942 
1943 static void
1944 cas_intr_task(void *arg, int pending __unused)
1945 {
1946 	struct cas_softc *sc = arg;
1947 	struct ifnet *ifp = sc->sc_ifp;
1948 	uint32_t status, status2;
1949 
1950 	CAS_LOCK_ASSERT(sc, MA_NOTOWNED);
1951 
1952 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1953 		return;
1954 
1955 	status = CAS_READ_4(sc, CAS_STATUS);
1956 	if (__predict_false((status & CAS_INTR_SUMMARY) == 0))
1957 		goto done;
1958 
1959 #ifdef CAS_DEBUG
1960 	CTR4(KTR_CAS, "%s: %s: cplt %x, status %x",
1961 	    device_get_name(sc->sc_dev), __func__,
1962 	    (status >> CAS_STATUS_TX_COMP3_SHIFT), (u_int)status);
1963 
1964 	/*
1965 	 * PCS interrupts must be cleared, otherwise no traffic is passed!
1966 	 */
1967 	if ((status & CAS_INTR_PCS_INT) != 0) {
1968 		status2 =
1969 		    CAS_READ_4(sc, CAS_PCS_INTR_STATUS) |
1970 		    CAS_READ_4(sc, CAS_PCS_INTR_STATUS);
1971 		if ((status2 & CAS_PCS_INTR_LINK) != 0)
1972 			device_printf(sc->sc_dev,
1973 			    "%s: PCS link status changed\n", __func__);
1974 	}
1975 	if ((status & CAS_MAC_CTRL_STATUS) != 0) {
1976 		status2 = CAS_READ_4(sc, CAS_MAC_CTRL_STATUS);
1977 		if ((status2 & CAS_MAC_CTRL_PAUSE) != 0)
1978 			device_printf(sc->sc_dev,
1979 			    "%s: PAUSE received (PAUSE time %d slots)\n",
1980 			    __func__,
1981 			    (status2 & CAS_MAC_CTRL_STATUS_PT_MASK) >>
1982 			    CAS_MAC_CTRL_STATUS_PT_SHFT);
1983 		if ((status2 & CAS_MAC_CTRL_PAUSE) != 0)
1984 			device_printf(sc->sc_dev,
1985 			    "%s: transited to PAUSE state\n", __func__);
1986 		if ((status2 & CAS_MAC_CTRL_NON_PAUSE) != 0)
1987 			device_printf(sc->sc_dev,
1988 			    "%s: transited to non-PAUSE state\n", __func__);
1989 	}
1990 	if ((status & CAS_INTR_MIF) != 0)
1991 		device_printf(sc->sc_dev, "%s: MIF interrupt\n", __func__);
1992 #endif
1993 
1994 	if (__predict_false((status &
1995 	    (CAS_INTR_TX_TAG_ERR | CAS_INTR_RX_TAG_ERR |
1996 	    CAS_INTR_RX_LEN_MMATCH | CAS_INTR_PCI_ERROR_INT)) != 0)) {
1997 		cas_eint(sc, status);
1998 		return;
1999 	}
2000 
2001 	if (__predict_false(status & CAS_INTR_TX_MAC_INT)) {
2002 		status2 = CAS_READ_4(sc, CAS_MAC_TX_STATUS);
2003 		if ((status2 &
2004 		    (CAS_MAC_TX_UNDERRUN | CAS_MAC_TX_MAX_PKT_ERR)) != 0)
2005 			sc->sc_ifp->if_oerrors++;
2006 		else if ((status2 & ~CAS_MAC_TX_FRAME_XMTD) != 0)
2007 			device_printf(sc->sc_dev,
2008 			    "MAC TX fault, status %x\n", status2);
2009 	}
2010 
2011 	if (__predict_false(status & CAS_INTR_RX_MAC_INT)) {
2012 		status2 = CAS_READ_4(sc, CAS_MAC_RX_STATUS);
2013 		if ((status2 & CAS_MAC_RX_OVERFLOW) != 0)
2014 			sc->sc_ifp->if_ierrors++;
2015 		else if ((status2 & ~CAS_MAC_RX_FRAME_RCVD) != 0)
2016 			device_printf(sc->sc_dev,
2017 			    "MAC RX fault, status %x\n", status2);
2018 	}
2019 
2020 	if ((status &
2021 	    (CAS_INTR_RX_DONE | CAS_INTR_RX_BUF_NA | CAS_INTR_RX_COMP_FULL |
2022 	    CAS_INTR_RX_BUF_AEMPTY | CAS_INTR_RX_COMP_AFULL)) != 0) {
2023 		cas_rint(sc);
2024 #ifdef CAS_DEBUG
2025 		if (__predict_false((status &
2026 		    (CAS_INTR_RX_BUF_NA | CAS_INTR_RX_COMP_FULL |
2027 		    CAS_INTR_RX_BUF_AEMPTY | CAS_INTR_RX_COMP_AFULL)) != 0))
2028 			device_printf(sc->sc_dev,
2029 			    "RX fault, status %x\n", status);
2030 #endif
2031 	}
2032 
2033 	if ((status &
2034 	    (CAS_INTR_TX_INT_ME | CAS_INTR_TX_ALL | CAS_INTR_TX_DONE)) != 0) {
2035 		CAS_LOCK(sc);
2036 		cas_tint(sc);
2037 		CAS_UNLOCK(sc);
2038 	}
2039 
2040 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
2041 		return;
2042 	else if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2043 		taskqueue_enqueue(sc->sc_tq, &sc->sc_tx_task);
2044 
2045 	status = CAS_READ_4(sc, CAS_STATUS_ALIAS);
2046 	if (__predict_false((status & CAS_INTR_SUMMARY) != 0)) {
2047 		taskqueue_enqueue(sc->sc_tq, &sc->sc_intr_task);
2048 		return;
2049 	}
2050 
2051  done:
2052 	/* Re-enable interrupts. */
2053 	CAS_WRITE_4(sc, CAS_INTMASK,
2054 	    ~(CAS_INTR_TX_INT_ME | CAS_INTR_TX_TAG_ERR |
2055 	    CAS_INTR_RX_DONE | CAS_INTR_RX_BUF_NA | CAS_INTR_RX_TAG_ERR |
2056 	    CAS_INTR_RX_COMP_FULL | CAS_INTR_RX_BUF_AEMPTY |
2057 	    CAS_INTR_RX_COMP_AFULL | CAS_INTR_RX_LEN_MMATCH |
2058 	    CAS_INTR_PCI_ERROR_INT
2059 #ifdef CAS_DEBUG
2060 	    | CAS_INTR_PCS_INT | CAS_INTR_MIF
2061 #endif
2062 	));
2063 }
2064 
2065 static void
2066 cas_watchdog(struct cas_softc *sc)
2067 {
2068 	struct ifnet *ifp = sc->sc_ifp;
2069 
2070 	CAS_LOCK_ASSERT(sc, MA_OWNED);
2071 
2072 #ifdef CAS_DEBUG
2073 	CTR4(KTR_CAS,
2074 	    "%s: CAS_RX_CONFIG %x CAS_MAC_RX_STATUS %x CAS_MAC_RX_CONFIG %x",
2075 	    __func__, CAS_READ_4(sc, CAS_RX_CONFIG),
2076 	    CAS_READ_4(sc, CAS_MAC_RX_STATUS),
2077 	    CAS_READ_4(sc, CAS_MAC_RX_CONFIG));
2078 	CTR4(KTR_CAS,
2079 	    "%s: CAS_TX_CONFIG %x CAS_MAC_TX_STATUS %x CAS_MAC_TX_CONFIG %x",
2080 	    __func__, CAS_READ_4(sc, CAS_TX_CONFIG),
2081 	    CAS_READ_4(sc, CAS_MAC_TX_STATUS),
2082 	    CAS_READ_4(sc, CAS_MAC_TX_CONFIG));
2083 #endif
2084 
2085 	if (sc->sc_wdog_timer == 0 || --sc->sc_wdog_timer != 0)
2086 		return;
2087 
2088 	if ((sc->sc_flags & CAS_LINK) != 0)
2089 		device_printf(sc->sc_dev, "device timeout\n");
2090 	else if (bootverbose)
2091 		device_printf(sc->sc_dev, "device timeout (no link)\n");
2092 	++ifp->if_oerrors;
2093 
2094 	/* Try to get more packets going. */
2095 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2096 	cas_init_locked(sc);
2097 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2098 		taskqueue_enqueue(sc->sc_tq, &sc->sc_tx_task);
2099 }
2100 
2101 static void
2102 cas_mifinit(struct cas_softc *sc)
2103 {
2104 
2105 	/* Configure the MIF in frame mode. */
2106 	CAS_WRITE_4(sc, CAS_MIF_CONF,
2107 	    CAS_READ_4(sc, CAS_MIF_CONF) & ~CAS_MIF_CONF_BB_MODE);
2108 }
2109 
2110 /*
2111  * MII interface
2112  *
2113  * The MII interface supports at least three different operating modes:
2114  *
2115  * Bitbang mode is implemented using data, clock and output enable registers.
2116  *
2117  * Frame mode is implemented by loading a complete frame into the frame
2118  * register and polling the valid bit for completion.
2119  *
2120  * Polling mode uses the frame register but completion is indicated by
2121  * an interrupt.
2122  *
2123  */
2124 static int
2125 cas_mii_readreg(device_t dev, int phy, int reg)
2126 {
2127 	struct cas_softc *sc;
2128 	int n;
2129 	uint32_t v;
2130 
2131 #ifdef CAS_DEBUG_PHY
2132 	printf("%s: phy %d reg %d\n", __func__, phy, reg);
2133 #endif
2134 
2135 	sc = device_get_softc(dev);
2136 	if (sc->sc_phyad != -1 && phy != sc->sc_phyad)
2137 		return (0);
2138 
2139 	if ((sc->sc_flags & CAS_SERDES) != 0) {
2140 		switch (reg) {
2141 		case MII_BMCR:
2142 			reg = CAS_PCS_CTRL;
2143 			break;
2144 		case MII_BMSR:
2145 			reg = CAS_PCS_STATUS;
2146 			break;
2147 		case MII_PHYIDR1:
2148 		case MII_PHYIDR2:
2149 			return (0);
2150 		case MII_ANAR:
2151 			reg = CAS_PCS_ANAR;
2152 			break;
2153 		case MII_ANLPAR:
2154 			reg = CAS_PCS_ANLPAR;
2155 			break;
2156 		case MII_EXTSR:
2157 			return (EXTSR_1000XFDX | EXTSR_1000XHDX);
2158 		default:
2159 			device_printf(sc->sc_dev,
2160 			    "%s: unhandled register %d\n", __func__, reg);
2161 			return (0);
2162 		}
2163 		return (CAS_READ_4(sc, reg));
2164 	}
2165 
2166 	/* Construct the frame command. */
2167 	v = CAS_MIF_FRAME_READ |
2168 	    (phy << CAS_MIF_FRAME_PHY_SHFT) |
2169 	    (reg << CAS_MIF_FRAME_REG_SHFT);
2170 
2171 	CAS_WRITE_4(sc, CAS_MIF_FRAME, v);
2172 	CAS_BARRIER(sc, CAS_MIF_FRAME, 4,
2173 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2174 	for (n = 0; n < 100; n++) {
2175 		DELAY(1);
2176 		v = CAS_READ_4(sc, CAS_MIF_FRAME);
2177 		if (v & CAS_MIF_FRAME_TA_LSB)
2178 			return (v & CAS_MIF_FRAME_DATA);
2179 	}
2180 
2181 	device_printf(sc->sc_dev, "%s: timed out\n", __func__);
2182 	return (0);
2183 }
2184 
2185 static int
2186 cas_mii_writereg(device_t dev, int phy, int reg, int val)
2187 {
2188 	struct cas_softc *sc;
2189 	int n;
2190 	uint32_t v;
2191 
2192 #ifdef CAS_DEBUG_PHY
2193 	printf("%s: phy %d reg %d val %x\n", phy, reg, val, __func__);
2194 #endif
2195 
2196 	sc = device_get_softc(dev);
2197 	if (sc->sc_phyad != -1 && phy != sc->sc_phyad)
2198 		return (0);
2199 
2200 	if ((sc->sc_flags & CAS_SERDES) != 0) {
2201 		switch (reg) {
2202 		case MII_BMSR:
2203 			reg = CAS_PCS_STATUS;
2204 			break;
2205 		case MII_BMCR:
2206 			reg = CAS_PCS_CTRL;
2207 			if ((val & CAS_PCS_CTRL_RESET) == 0)
2208 				break;
2209 			CAS_WRITE_4(sc, CAS_PCS_CTRL, val);
2210 			CAS_BARRIER(sc, CAS_PCS_CTRL, 4,
2211 			    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2212 			if (!cas_bitwait(sc, CAS_PCS_CTRL,
2213 			    CAS_PCS_CTRL_RESET, 0))
2214 				device_printf(sc->sc_dev,
2215 				    "cannot reset PCS\n");
2216 			/* FALLTHROUGH */
2217 		case MII_ANAR:
2218 			CAS_WRITE_4(sc, CAS_PCS_CONF, 0);
2219 			CAS_BARRIER(sc, CAS_PCS_CONF, 4,
2220 			    BUS_SPACE_BARRIER_WRITE);
2221 			CAS_WRITE_4(sc, CAS_PCS_ANAR, val);
2222 			CAS_WRITE_4(sc, CAS_PCS_SERDES_CTRL,
2223 			    CAS_PCS_SERDES_CTRL_ESD);
2224 			CAS_WRITE_4(sc, CAS_PCS_CONF,
2225 			    CAS_PCS_CONF_EN);
2226 			return (0);
2227 		case MII_ANLPAR:
2228 			reg = CAS_PCS_ANLPAR;
2229 			break;
2230 		default:
2231 			device_printf(sc->sc_dev,
2232 			    "%s: unhandled register %d\n", __func__, reg);
2233 			return (0);
2234 		}
2235 		CAS_WRITE_4(sc, reg, val);
2236 		return (0);
2237 	}
2238 
2239 	/* Construct the frame command. */
2240 	v = CAS_MIF_FRAME_WRITE |
2241 	    (phy << CAS_MIF_FRAME_PHY_SHFT) |
2242 	    (reg << CAS_MIF_FRAME_REG_SHFT) |
2243 	    (val & CAS_MIF_FRAME_DATA);
2244 
2245 	CAS_WRITE_4(sc, CAS_MIF_FRAME, v);
2246 	CAS_BARRIER(sc, CAS_MIF_FRAME, 4,
2247 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2248 	for (n = 0; n < 100; n++) {
2249 		DELAY(1);
2250 		v = CAS_READ_4(sc, CAS_MIF_FRAME);
2251 		if (v & CAS_MIF_FRAME_TA_LSB)
2252 			return (1);
2253 	}
2254 
2255 	device_printf(sc->sc_dev, "%s: timed out\n", __func__);
2256 	return (0);
2257 }
2258 
2259 static void
2260 cas_mii_statchg(device_t dev)
2261 {
2262 	struct cas_softc *sc;
2263 	struct ifnet *ifp;
2264 	int gigabit;
2265 	uint32_t rxcfg, txcfg, v;
2266 
2267 	sc = device_get_softc(dev);
2268 	ifp = sc->sc_ifp;
2269 
2270 	CAS_LOCK_ASSERT(sc, MA_OWNED);
2271 
2272 #ifdef CAS_DEBUG
2273 	if ((ifp->if_flags & IFF_DEBUG) != 0)
2274 		device_printf(sc->sc_dev, "%s: status change: PHY = %d\n",
2275 		    __func__, sc->sc_phyad);
2276 #endif
2277 
2278 	if ((sc->sc_mii->mii_media_status & IFM_ACTIVE) != 0 &&
2279 	    IFM_SUBTYPE(sc->sc_mii->mii_media_active) != IFM_NONE)
2280 		sc->sc_flags |= CAS_LINK;
2281 	else
2282 		sc->sc_flags &= ~CAS_LINK;
2283 
2284 	switch (IFM_SUBTYPE(sc->sc_mii->mii_media_active)) {
2285 	case IFM_1000_SX:
2286 	case IFM_1000_LX:
2287 	case IFM_1000_CX:
2288 	case IFM_1000_T:
2289 		gigabit = 1;
2290 		break;
2291 	default:
2292 		gigabit = 0;
2293 	}
2294 
2295 	/*
2296 	 * The configuration done here corresponds to the steps F) and
2297 	 * G) and as far as enabling of RX and TX MAC goes also step H)
2298 	 * of the initialization sequence outlined in section 11.2.1 of
2299 	 * the Cassini+ ASIC Specification.
2300 	 */
2301 
2302 	rxcfg = CAS_READ_4(sc, CAS_MAC_RX_CONF);
2303 	rxcfg &= ~(CAS_MAC_RX_CONF_EN | CAS_MAC_RX_CONF_CARR);
2304 	txcfg = CAS_MAC_TX_CONF_EN_IPG0 | CAS_MAC_TX_CONF_NGU |
2305 	    CAS_MAC_TX_CONF_NGUL;
2306 	if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) & IFM_FDX) != 0)
2307 		txcfg |= CAS_MAC_TX_CONF_ICARR | CAS_MAC_TX_CONF_ICOLLIS;
2308 	else if (gigabit != 0) {
2309 		rxcfg |= CAS_MAC_RX_CONF_CARR;
2310 		txcfg |= CAS_MAC_TX_CONF_CARR;
2311 	}
2312 	CAS_WRITE_4(sc, CAS_MAC_TX_CONF, 0);
2313 	CAS_BARRIER(sc, CAS_MAC_TX_CONF, 4,
2314 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2315 	if (!cas_bitwait(sc, CAS_MAC_TX_CONF, CAS_MAC_TX_CONF_EN, 0))
2316 		device_printf(sc->sc_dev, "cannot disable TX MAC\n");
2317 	CAS_WRITE_4(sc, CAS_MAC_TX_CONF, txcfg);
2318 	CAS_WRITE_4(sc, CAS_MAC_RX_CONF, 0);
2319 	CAS_BARRIER(sc, CAS_MAC_RX_CONF, 4,
2320 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2321 	if (!cas_bitwait(sc, CAS_MAC_RX_CONF, CAS_MAC_RX_CONF_EN, 0))
2322 		device_printf(sc->sc_dev, "cannot disable RX MAC\n");
2323 	CAS_WRITE_4(sc, CAS_MAC_RX_CONF, rxcfg);
2324 
2325 	v = CAS_READ_4(sc, CAS_MAC_CTRL_CONF) &
2326 	    ~(CAS_MAC_CTRL_CONF_TXP | CAS_MAC_CTRL_CONF_RXP);
2327 #ifdef notyet
2328 	if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) &
2329 	    IFM_ETH_RXPAUSE) != 0)
2330 		v |= CAS_MAC_CTRL_CONF_RXP;
2331 	if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) &
2332 	    IFM_ETH_TXPAUSE) != 0)
2333 		v |= CAS_MAC_CTRL_CONF_TXP;
2334 #endif
2335 	CAS_WRITE_4(sc, CAS_MAC_CTRL_CONF, v);
2336 
2337 	/*
2338 	 * All supported chips have a bug causing incorrect checksum
2339 	 * to be calculated when letting them strip the FCS in half-
2340 	 * duplex mode.  In theory we could disable FCS stripping and
2341 	 * manually adjust the checksum accordingly.  It seems to make
2342 	 * more sense to optimze for the common case and just disable
2343 	 * hardware checksumming in half-duplex mode though.
2344 	 */
2345 	if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) & IFM_FDX) == 0) {
2346 		ifp->if_capenable &= ~IFCAP_HWCSUM;
2347 		ifp->if_hwassist = 0;
2348 	} else if ((sc->sc_flags & CAS_NO_CSUM) == 0) {
2349 		ifp->if_capenable = ifp->if_capabilities;
2350 		ifp->if_hwassist = CAS_CSUM_FEATURES;
2351 	}
2352 
2353 	if (sc->sc_variant == CAS_SATURN) {
2354 		if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) & IFM_FDX) == 0)
2355 			/* silicon bug workaround */
2356 			CAS_WRITE_4(sc, CAS_MAC_PREAMBLE_LEN, 0x41);
2357 		else
2358 			CAS_WRITE_4(sc, CAS_MAC_PREAMBLE_LEN, 0x7);
2359 	}
2360 
2361 	if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) & IFM_FDX) == 0 &&
2362 	    gigabit != 0)
2363 		CAS_WRITE_4(sc, CAS_MAC_SLOT_TIME,
2364 		    CAS_MAC_SLOT_TIME_CARR);
2365 	else
2366 		CAS_WRITE_4(sc, CAS_MAC_SLOT_TIME,
2367 		    CAS_MAC_SLOT_TIME_NORM);
2368 
2369 	/* XIF Configuration */
2370 	v = CAS_MAC_XIF_CONF_TX_OE | CAS_MAC_XIF_CONF_LNKLED;
2371 	if ((sc->sc_flags & CAS_SERDES) == 0) {
2372 		if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) & IFM_FDX) == 0)
2373 			v |= CAS_MAC_XIF_CONF_NOECHO;
2374 		v |= CAS_MAC_XIF_CONF_BUF_OE;
2375 	}
2376 	if (gigabit != 0)
2377 		v |= CAS_MAC_XIF_CONF_GMII;
2378 	if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) & IFM_FDX) != 0)
2379 		v |= CAS_MAC_XIF_CONF_FDXLED;
2380 	CAS_WRITE_4(sc, CAS_MAC_XIF_CONF, v);
2381 
2382 	if ((sc->sc_ifp->if_drv_flags & IFF_DRV_RUNNING) != 0 &&
2383 	    (sc->sc_flags & CAS_LINK) != 0) {
2384 		CAS_WRITE_4(sc, CAS_MAC_TX_CONF,
2385 		    txcfg | CAS_MAC_TX_CONF_EN);
2386 		CAS_WRITE_4(sc, CAS_MAC_RX_CONF,
2387 		    rxcfg | CAS_MAC_RX_CONF_EN);
2388 	}
2389 }
2390 
2391 static int
2392 cas_mediachange(struct ifnet *ifp)
2393 {
2394 	struct cas_softc *sc = ifp->if_softc;
2395 	int error;
2396 
2397 	/* XXX add support for serial media. */
2398 
2399 	CAS_LOCK(sc);
2400 	error = mii_mediachg(sc->sc_mii);
2401 	CAS_UNLOCK(sc);
2402 	return (error);
2403 }
2404 
2405 static void
2406 cas_mediastatus(struct ifnet *ifp, struct ifmediareq *ifmr)
2407 {
2408 	struct cas_softc *sc = ifp->if_softc;
2409 
2410 	CAS_LOCK(sc);
2411 	if ((ifp->if_flags & IFF_UP) == 0) {
2412 		CAS_UNLOCK(sc);
2413 		return;
2414 	}
2415 
2416 	mii_pollstat(sc->sc_mii);
2417 	ifmr->ifm_active = sc->sc_mii->mii_media_active;
2418 	ifmr->ifm_status = sc->sc_mii->mii_media_status;
2419 	CAS_UNLOCK(sc);
2420 }
2421 
2422 static int
2423 cas_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
2424 {
2425 	struct cas_softc *sc = ifp->if_softc;
2426 	struct ifreq *ifr = (struct ifreq *)data;
2427 	int error;
2428 
2429 	error = 0;
2430 	switch (cmd) {
2431 	case SIOCSIFFLAGS:
2432 		CAS_LOCK(sc);
2433 		if ((ifp->if_flags & IFF_UP) != 0) {
2434 			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0 &&
2435 			    ((ifp->if_flags ^ sc->sc_ifflags) &
2436 			    (IFF_ALLMULTI | IFF_PROMISC)) != 0)
2437 				cas_setladrf(sc);
2438 			else
2439 				cas_init_locked(sc);
2440 		} else if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
2441 			cas_stop(ifp);
2442 		sc->sc_ifflags = ifp->if_flags;
2443 		CAS_UNLOCK(sc);
2444 		break;
2445 	case SIOCSIFCAP:
2446 		CAS_LOCK(sc);
2447 		if ((sc->sc_flags & CAS_NO_CSUM) != 0) {
2448 			error = EINVAL;
2449 			CAS_UNLOCK(sc);
2450 			break;
2451 		}
2452 		ifp->if_capenable = ifr->ifr_reqcap;
2453 		if ((ifp->if_capenable & IFCAP_TXCSUM) != 0)
2454 			ifp->if_hwassist = CAS_CSUM_FEATURES;
2455 		else
2456 			ifp->if_hwassist = 0;
2457 		CAS_UNLOCK(sc);
2458 		break;
2459 	case SIOCADDMULTI:
2460 	case SIOCDELMULTI:
2461 		CAS_LOCK(sc);
2462 		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
2463 			cas_setladrf(sc);
2464 		CAS_UNLOCK(sc);
2465 		break;
2466 	case SIOCSIFMTU:
2467 		if ((ifr->ifr_mtu < ETHERMIN) ||
2468 		    (ifr->ifr_mtu > ETHERMTU_JUMBO))
2469 			error = EINVAL;
2470 		else
2471 			ifp->if_mtu = ifr->ifr_mtu;
2472 		break;
2473 	case SIOCGIFMEDIA:
2474 	case SIOCSIFMEDIA:
2475 		error = ifmedia_ioctl(ifp, ifr, &sc->sc_mii->mii_media, cmd);
2476 		break;
2477 	default:
2478 		error = ether_ioctl(ifp, cmd, data);
2479 		break;
2480 	}
2481 
2482 	return (error);
2483 }
2484 
2485 static void
2486 cas_setladrf(struct cas_softc *sc)
2487 {
2488 	struct ifnet *ifp = sc->sc_ifp;
2489 	struct ifmultiaddr *inm;
2490 	int i;
2491 	uint32_t hash[16];
2492 	uint32_t crc, v;
2493 
2494 	CAS_LOCK_ASSERT(sc, MA_OWNED);
2495 
2496 	/* Get the current RX configuration. */
2497 	v = CAS_READ_4(sc, CAS_MAC_RX_CONF);
2498 
2499 	/*
2500 	 * Turn off promiscuous mode, promiscuous group mode (all multicast),
2501 	 * and hash filter.  Depending on the case, the right bit will be
2502 	 * enabled.
2503 	 */
2504 	v &= ~(CAS_MAC_RX_CONF_PROMISC | CAS_MAC_RX_CONF_HFILTER |
2505 	    CAS_MAC_RX_CONF_PGRP);
2506 
2507 	CAS_WRITE_4(sc, CAS_MAC_RX_CONF, v);
2508 	CAS_BARRIER(sc, CAS_MAC_RX_CONF, 4,
2509 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2510 	if (!cas_bitwait(sc, CAS_MAC_RX_CONF, CAS_MAC_RX_CONF_HFILTER, 0))
2511 		device_printf(sc->sc_dev, "cannot disable RX hash filter\n");
2512 
2513 	if ((ifp->if_flags & IFF_PROMISC) != 0) {
2514 		v |= CAS_MAC_RX_CONF_PROMISC;
2515 		goto chipit;
2516 	}
2517 	if ((ifp->if_flags & IFF_ALLMULTI) != 0) {
2518 		v |= CAS_MAC_RX_CONF_PGRP;
2519 		goto chipit;
2520 	}
2521 
2522 	/*
2523 	 * Set up multicast address filter by passing all multicast
2524 	 * addresses through a crc generator, and then using the high
2525 	 * order 8 bits as an index into the 256 bit logical address
2526 	 * filter.  The high order 4 bits selects the word, while the
2527 	 * other 4 bits select the bit within the word (where bit 0
2528 	 * is the MSB).
2529 	 */
2530 
2531 	/* Clear the hash table. */
2532 	memset(hash, 0, sizeof(hash));
2533 
2534 	if_maddr_rlock(ifp);
2535 	TAILQ_FOREACH(inm, &ifp->if_multiaddrs, ifma_link) {
2536 		if (inm->ifma_addr->sa_family != AF_LINK)
2537 			continue;
2538 		crc = ether_crc32_le(LLADDR((struct sockaddr_dl *)
2539 		    inm->ifma_addr), ETHER_ADDR_LEN);
2540 
2541 		/* We just want the 8 most significant bits. */
2542 		crc >>= 24;
2543 
2544 		/* Set the corresponding bit in the filter. */
2545 		hash[crc >> 4] |= 1 << (15 - (crc & 15));
2546 	}
2547 	if_maddr_runlock(ifp);
2548 
2549 	v |= CAS_MAC_RX_CONF_HFILTER;
2550 
2551 	/* Now load the hash table into the chip (if we are using it). */
2552 	for (i = 0; i < 16; i++)
2553 		CAS_WRITE_4(sc,
2554 		    CAS_MAC_HASH0 + i * (CAS_MAC_HASH1 - CAS_MAC_HASH0),
2555 		    hash[i]);
2556 
2557  chipit:
2558 	CAS_WRITE_4(sc, CAS_MAC_RX_CONF, v);
2559 }
2560 
2561 static int	cas_pci_attach(device_t dev);
2562 static int	cas_pci_detach(device_t dev);
2563 static int	cas_pci_probe(device_t dev);
2564 static int	cas_pci_resume(device_t dev);
2565 static int	cas_pci_suspend(device_t dev);
2566 
2567 static device_method_t cas_pci_methods[] = {
2568 	/* Device interface */
2569 	DEVMETHOD(device_probe,		cas_pci_probe),
2570 	DEVMETHOD(device_attach,	cas_pci_attach),
2571 	DEVMETHOD(device_detach,	cas_pci_detach),
2572 	DEVMETHOD(device_suspend,	cas_pci_suspend),
2573 	DEVMETHOD(device_resume,	cas_pci_resume),
2574 	/* Use the suspend handler here, it is all that is required. */
2575 	DEVMETHOD(device_shutdown,	cas_pci_suspend),
2576 
2577 	/* bus interface */
2578 	DEVMETHOD(bus_print_child,	bus_generic_print_child),
2579 	DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
2580 
2581 	/* MII interface */
2582 	DEVMETHOD(miibus_readreg,	cas_mii_readreg),
2583 	DEVMETHOD(miibus_writereg,	cas_mii_writereg),
2584 	DEVMETHOD(miibus_statchg,	cas_mii_statchg),
2585 
2586 	KOBJMETHOD_END
2587 };
2588 
2589 static driver_t cas_pci_driver = {
2590 	"cas",
2591 	cas_pci_methods,
2592 	sizeof(struct cas_softc)
2593 };
2594 
2595 DRIVER_MODULE(cas, pci, cas_pci_driver, cas_devclass, 0, 0);
2596 DRIVER_MODULE(miibus, cas, miibus_driver, miibus_devclass, 0, 0);
2597 MODULE_DEPEND(cas, pci, 1, 1, 1);
2598 
2599 static const struct cas_pci_dev {
2600 	uint32_t	cpd_devid;
2601 	uint8_t		cpd_revid;
2602 	int		cpd_variant;
2603 	const char	*cpd_desc;
2604 } const cas_pci_devlist[] = {
2605 	{ 0x0035100b, 0x0, CAS_SATURN, "NS DP83065 Saturn Gigabit Ethernet" },
2606 	{ 0xabba108e, 0x10, CAS_CASPLUS, "Sun Cassini+ Gigabit Ethernet" },
2607 	{ 0xabba108e, 0x0, CAS_CAS, "Sun Cassini Gigabit Ethernet" },
2608 	{ 0, 0, 0, NULL }
2609 };
2610 
2611 static int
2612 cas_pci_probe(device_t dev)
2613 {
2614 	int i;
2615 
2616 	for (i = 0; cas_pci_devlist[i].cpd_desc != NULL; i++) {
2617 		if (pci_get_devid(dev) == cas_pci_devlist[i].cpd_devid &&
2618 		    pci_get_revid(dev) >= cas_pci_devlist[i].cpd_revid) {
2619 			device_set_desc(dev, cas_pci_devlist[i].cpd_desc);
2620 			return (BUS_PROBE_DEFAULT);
2621 		}
2622 	}
2623 
2624 	return (ENXIO);
2625 }
2626 
2627 static struct resource_spec cas_pci_res_spec[] = {
2628 	{ SYS_RES_IRQ, 0, RF_SHAREABLE | RF_ACTIVE },	/* CAS_RES_INTR */
2629 	{ SYS_RES_MEMORY, PCIR_BAR(0), RF_ACTIVE },	/* CAS_RES_MEM */
2630 	{ -1, 0 }
2631 };
2632 
2633 static int
2634 cas_pci_attach(device_t dev)
2635 {
2636 	struct cas_softc *sc;
2637 	int i;
2638 #if !(defined(__powerpc__) || defined(__sparc64__))
2639 	u_char enaddr[4][ETHER_ADDR_LEN];
2640 	char lma[sizeof("local-mac-address")];
2641 	int found, j;
2642 #endif
2643 
2644 	sc = device_get_softc(dev);
2645 	sc->sc_variant = CAS_UNKNOWN;
2646 	for (i = 0; cas_pci_devlist[i].cpd_desc != NULL; i++) {
2647 		if (pci_get_devid(dev) == cas_pci_devlist[i].cpd_devid &&
2648 		    pci_get_revid(dev) >= cas_pci_devlist[i].cpd_revid) {
2649 			sc->sc_variant = cas_pci_devlist[i].cpd_variant;
2650 			break;
2651 		}
2652 	}
2653 	if (sc->sc_variant == CAS_UNKNOWN) {
2654 		device_printf(dev, "unknown adaptor\n");
2655 		return (ENXIO);
2656 	}
2657 
2658 	pci_enable_busmaster(dev);
2659 
2660 	sc->sc_dev = dev;
2661 	if (sc->sc_variant == CAS_CAS && pci_get_devid(dev) < 0x02)
2662 		/* Hardware checksumming may hang TX. */
2663 		sc->sc_flags |= CAS_NO_CSUM;
2664 	if (sc->sc_variant == CAS_CASPLUS || sc->sc_variant == CAS_SATURN)
2665 		sc->sc_flags |= CAS_REG_PLUS;
2666 	if (sc->sc_variant == CAS_CAS ||
2667 	    (sc->sc_variant == CAS_CASPLUS && pci_get_revid(dev) < 0x11))
2668 		sc->sc_flags |= CAS_TABORT;
2669 	if (bootverbose)
2670 		device_printf(dev, "flags=0x%x\n", sc->sc_flags);
2671 
2672 	if (bus_alloc_resources(dev, cas_pci_res_spec, sc->sc_res)) {
2673 		device_printf(dev, "failed to allocate resources\n");
2674 		bus_release_resources(dev, cas_pci_res_spec, sc->sc_res);
2675 		return (ENXIO);
2676 	}
2677 
2678 	CAS_LOCK_INIT(sc, device_get_nameunit(dev));
2679 
2680 #if defined(__powerpc__) || defined(__sparc64__)
2681 	OF_getetheraddr(dev, sc->sc_enaddr);
2682 #else
2683 	/*
2684 	 * Dig out VPD (vital product data) and read the MAX address.
2685 	 * The VPD resides in the PCI Expansion ROM (PCI FCode) and
2686 	 * can't be accessed via the PCI capability pointer.
2687 	 * SUNW,pci-ce and SUNW,pci-qge use the Enhanced VPD format
2688 	 * described in US Patent 7149820.
2689 	 */
2690 
2691 #define	PCI_ROMHDR_SIZE			0x1c
2692 #define	PCI_ROMHDR_SIG			0x00
2693 #define	PCI_ROMHDR_SIG_MAGIC		0xaa55		/* little endian */
2694 #define	PCI_ROMHDR_PTR_DATA		0x18
2695 #define	PCI_ROM_SIZE			0x18
2696 #define	PCI_ROM_SIG			0x00
2697 #define	PCI_ROM_SIG_MAGIC		0x52494350	/* "PCIR", endian */
2698 							/* reversed */
2699 #define	PCI_ROM_VENDOR			0x04
2700 #define	PCI_ROM_DEVICE			0x06
2701 #define	PCI_ROM_PTR_VPD			0x08
2702 #define	PCI_VPDRES_BYTE0		0x00
2703 #define	PCI_VPDRES_ISLARGE(x)		((x) & 0x80)
2704 #define	PCI_VPDRES_LARGE_NAME(x)	((x) & 0x7f)
2705 #define	PCI_VPDRES_LARGE_LEN_LSB	0x01
2706 #define	PCI_VPDRES_LARGE_LEN_MSB	0x02
2707 #define	PCI_VPDRES_LARGE_SIZE		0x03
2708 #define	PCI_VPDRES_TYPE_ID_STRING	0x02		/* large */
2709 #define	PCI_VPDRES_TYPE_VPD		0x10		/* large */
2710 #define	PCI_VPD_KEY0			0x00
2711 #define	PCI_VPD_KEY1			0x01
2712 #define	PCI_VPD_LEN			0x02
2713 #define	PCI_VPD_SIZE			0x03
2714 
2715 #define	CAS_ROM_READ_1(sc, offs)					\
2716 	CAS_READ_1((sc), CAS_PCI_ROM_OFFSET + (offs))
2717 #define	CAS_ROM_READ_2(sc, offs)					\
2718 	CAS_READ_2((sc), CAS_PCI_ROM_OFFSET + (offs))
2719 #define	CAS_ROM_READ_4(sc, offs)					\
2720 	CAS_READ_4((sc), CAS_PCI_ROM_OFFSET + (offs))
2721 
2722 	found = 0;
2723 	/* Enable PCI Expansion ROM access. */
2724 	CAS_WRITE_4(sc, CAS_BIM_LDEV_OEN,
2725 	    CAS_BIM_LDEV_OEN_PAD | CAS_BIM_LDEV_OEN_PROM);
2726 
2727 	/* Read PCI Expansion ROM header. */
2728 	if (CAS_ROM_READ_2(sc, PCI_ROMHDR_SIG) != PCI_ROMHDR_SIG_MAGIC ||
2729 	    (i = CAS_ROM_READ_2(sc, PCI_ROMHDR_PTR_DATA)) <
2730 	    PCI_ROMHDR_SIZE) {
2731 		device_printf(dev, "unexpected PCI Expansion ROM header\n");
2732 		goto fail_prom;
2733 	}
2734 
2735 	/* Read PCI Expansion ROM data. */
2736 	if (CAS_ROM_READ_4(sc, i + PCI_ROM_SIG) != PCI_ROM_SIG_MAGIC ||
2737 	    CAS_ROM_READ_2(sc, i + PCI_ROM_VENDOR) != pci_get_vendor(dev) ||
2738 	    CAS_ROM_READ_2(sc, i + PCI_ROM_DEVICE) != pci_get_device(dev) ||
2739 	    (j = CAS_ROM_READ_2(sc, i + PCI_ROM_PTR_VPD)) <
2740 	    i + PCI_ROM_SIZE) {
2741 		device_printf(dev, "unexpected PCI Expansion ROM data\n");
2742 		goto fail_prom;
2743 	}
2744 
2745 	/* Read PCI VPD. */
2746  next:
2747 	if (PCI_VPDRES_ISLARGE(CAS_ROM_READ_1(sc,
2748 	    j + PCI_VPDRES_BYTE0)) == 0) {
2749 		device_printf(dev, "no large PCI VPD\n");
2750 		goto fail_prom;
2751 	}
2752 
2753 	i = (CAS_ROM_READ_1(sc, j + PCI_VPDRES_LARGE_LEN_MSB) << 8) |
2754 	    CAS_ROM_READ_1(sc, j + PCI_VPDRES_LARGE_LEN_LSB);
2755 	switch (PCI_VPDRES_LARGE_NAME(CAS_ROM_READ_1(sc,
2756 	    j + PCI_VPDRES_BYTE0))) {
2757 	case PCI_VPDRES_TYPE_ID_STRING:
2758 		/* Skip identifier string. */
2759 		j += PCI_VPDRES_LARGE_SIZE + i;
2760 		goto next;
2761 	case PCI_VPDRES_TYPE_VPD:
2762 		for (j += PCI_VPDRES_LARGE_SIZE; i > 0;
2763 		    i -= PCI_VPD_SIZE + CAS_ROM_READ_1(sc, j + PCI_VPD_LEN),
2764 		    j += PCI_VPD_SIZE + CAS_ROM_READ_1(sc, j + PCI_VPD_LEN)) {
2765 			if (CAS_ROM_READ_1(sc, j + PCI_VPD_KEY0) != 'Z')
2766 				/* no Enhanced VPD */
2767 				continue;
2768 			if (CAS_ROM_READ_1(sc, j + PCI_VPD_SIZE) != 'I')
2769 				/* no instance property */
2770 				continue;
2771 			if (CAS_ROM_READ_1(sc, j + PCI_VPD_SIZE + 3) != 'B')
2772 				/* no byte array */
2773 				continue;
2774 			if (CAS_ROM_READ_1(sc, j + PCI_VPD_SIZE + 4) !=
2775 			    ETHER_ADDR_LEN)
2776 				continue;
2777 			bus_read_region_1(sc->sc_res[CAS_RES_MEM],
2778 			    CAS_PCI_ROM_OFFSET + j + PCI_VPD_SIZE + 5,
2779 			    lma, sizeof(lma));
2780 			if (strcmp(lma, "local-mac-address") != 0)
2781 				continue;
2782 			bus_read_region_1(sc->sc_res[CAS_RES_MEM],
2783 			    CAS_PCI_ROM_OFFSET + j + PCI_VPD_SIZE + 5 +
2784 			    sizeof(lma), enaddr[found],
2785 			    sizeof(enaddr[found]));
2786 			if (found++ == 4)
2787 				break;
2788 		}
2789 		break;
2790 	default:
2791 		device_printf(dev, "unexpected PCI VPD\n");
2792 		goto fail_prom;
2793 	}
2794 
2795  fail_prom:
2796 	CAS_WRITE_4(sc, CAS_BIM_LDEV_OEN, 0);
2797 
2798 	if (found == 0) {
2799 		device_printf(dev, "could not determine Ethernet address\n");
2800 		goto fail;
2801 	}
2802 	i = 0;
2803 	if (found > 1 && pci_get_slot(dev) < sizeof(enaddr) / sizeof(*enaddr))
2804 		i = pci_get_slot(dev);
2805 	memcpy(sc->sc_enaddr, enaddr[i], ETHER_ADDR_LEN);
2806 #endif
2807 
2808 	if (cas_attach(sc) != 0) {
2809 		device_printf(dev, "could not be attached\n");
2810 		goto fail;
2811 	}
2812 
2813 	if (bus_setup_intr(dev, sc->sc_res[CAS_RES_INTR], INTR_TYPE_NET |
2814 	    INTR_MPSAFE, cas_intr, NULL, sc, &sc->sc_ih) != 0) {
2815 		device_printf(dev, "failed to set up interrupt\n");
2816 		cas_detach(sc);
2817 		goto fail;
2818 	}
2819 	return (0);
2820 
2821  fail:
2822 	CAS_LOCK_DESTROY(sc);
2823 	bus_release_resources(dev, cas_pci_res_spec, sc->sc_res);
2824 	return (ENXIO);
2825 }
2826 
2827 static int
2828 cas_pci_detach(device_t dev)
2829 {
2830 	struct cas_softc *sc;
2831 
2832 	sc = device_get_softc(dev);
2833 	bus_teardown_intr(dev, sc->sc_res[CAS_RES_INTR], sc->sc_ih);
2834 	cas_detach(sc);
2835 	CAS_LOCK_DESTROY(sc);
2836 	bus_release_resources(dev, cas_pci_res_spec, sc->sc_res);
2837 	return (0);
2838 }
2839 
2840 static int
2841 cas_pci_suspend(device_t dev)
2842 {
2843 
2844 	cas_suspend(device_get_softc(dev));
2845 	return (0);
2846 }
2847 
2848 static int
2849 cas_pci_resume(device_t dev)
2850 {
2851 
2852 	cas_resume(device_get_softc(dev));
2853 	return (0);
2854 }
2855