xref: /freebsd/sys/dev/cas/if_cas.c (revision a970610a3af63b3f4df5b69d91c6b4093a00ed8f)
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 
100 RINGASSERT(CAS_NRXCOMP, 128, 32768);
101 RINGASSERT(CAS_NRXDESC, 32, 8192);
102 RINGASSERT(CAS_NRXDESC2, 32, 8192);
103 RINGASSERT(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 
110 CCDASSERT(ccd_rxcomps, CAS_RX_COMP_ALIGN);
111 CCDASSERT(ccd_rxdescs, CAS_RX_DESC_ALIGN);
112 CCDASSERT(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 
126 static inline void cas_add_rxdesc(struct cas_softc *sc, u_int idx);
127 static int	cas_attach(struct cas_softc *sc);
128 static int	cas_bitwait(struct cas_softc *sc, bus_addr_t r, uint32_t clr,
129 		    uint32_t set);
130 static void	cas_cddma_callback(void *xsc, bus_dma_segment_t *segs,
131 		    int nsegs, int error);
132 static void	cas_detach(struct cas_softc *sc);
133 static int	cas_disable_rx(struct cas_softc *sc);
134 static int	cas_disable_tx(struct cas_softc *sc);
135 static void	cas_eint(struct cas_softc *sc, u_int status);
136 static void	cas_free(struct mbuf *m);
137 static void	cas_init(void *xsc);
138 static void	cas_init_locked(struct cas_softc *sc);
139 static void	cas_init_regs(struct cas_softc *sc);
140 static int	cas_intr(void *v);
141 static void	cas_intr_task(void *arg, int pending __unused);
142 static int	cas_ioctl(if_t ifp, u_long cmd, caddr_t data);
143 static int	cas_load_txmbuf(struct cas_softc *sc, struct mbuf **m_head);
144 static int	cas_mediachange(if_t ifp);
145 static void	cas_mediastatus(if_t ifp, struct ifmediareq *ifmr);
146 static void	cas_meminit(struct cas_softc *sc);
147 static void	cas_mifinit(struct cas_softc *sc);
148 static int	cas_mii_readreg(device_t dev, int phy, int reg);
149 static void	cas_mii_statchg(device_t dev);
150 static int	cas_mii_writereg(device_t dev, int phy, int reg, int val);
151 static void	cas_reset(struct cas_softc *sc);
152 static int	cas_reset_rx(struct cas_softc *sc);
153 static int	cas_reset_tx(struct cas_softc *sc);
154 static void	cas_resume(struct cas_softc *sc);
155 static u_int	cas_descsize(u_int sz);
156 static void	cas_rint(struct cas_softc *sc);
157 static void	cas_rint_timeout(void *arg);
158 static inline void cas_rxcksum(struct mbuf *m, uint16_t cksum);
159 static inline void cas_rxcompinit(struct cas_rx_comp *rxcomp);
160 static u_int	cas_rxcompsize(u_int sz);
161 static void	cas_rxdma_callback(void *xsc, bus_dma_segment_t *segs,
162 		    int nsegs, int error);
163 static void	cas_setladrf(struct cas_softc *sc);
164 static void	cas_start(if_t ifp);
165 static void	cas_stop(if_t ifp);
166 static void	cas_suspend(struct cas_softc *sc);
167 static void	cas_tick(void *arg);
168 static void	cas_tint(struct cas_softc *sc);
169 static void	cas_tx_task(void *arg, int pending __unused);
170 static inline void cas_txkick(struct cas_softc *sc);
171 static void	cas_watchdog(struct cas_softc *sc);
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_SPARE2
179 #endif
180 
181 static int
182 cas_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_setsoftc(ifp, sc);
192 	if_initname(ifp, device_get_name(sc->sc_dev),
193 	    device_get_unit(sc->sc_dev));
194 	if_setflags(ifp, IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST);
195 	if_setstartfn(ifp, cas_start);
196 	if_setioctlfn(ifp, cas_ioctl);
197 	if_setinitfn(ifp, cas_init);
198 	if_setsendqlen(ifp, CAS_TXQUEUELEN);
199 	if_setsendqready(ifp);
200 
201 	callout_init_mtx(&sc->sc_tick_ch, &sc->sc_mtx, 0);
202 	callout_init_mtx(&sc->sc_rx_ch, &sc->sc_mtx, 0);
203 	/* Create local taskq. */
204 	NET_TASK_INIT(&sc->sc_intr_task, 0, cas_intr_task, sc);
205 	TASK_INIT(&sc->sc_tx_task, 1, cas_tx_task, ifp);
206 	sc->sc_tq = taskqueue_create_fast("cas_taskq", M_WAITOK,
207 	    taskqueue_thread_enqueue, &sc->sc_tq);
208 	if (sc->sc_tq == NULL) {
209 		device_printf(sc->sc_dev, "could not create taskqueue\n");
210 		error = ENXIO;
211 		goto fail_ifnet;
212 	}
213 	error = taskqueue_start_threads(&sc->sc_tq, 1, PI_NET, "%s taskq",
214 	    device_get_nameunit(sc->sc_dev));
215 	if (error != 0) {
216 		device_printf(sc->sc_dev, "could not start threads\n");
217 		goto fail_taskq;
218 	}
219 
220 	/* Make sure the chip is stopped. */
221 	cas_reset(sc);
222 
223 	error = bus_dma_tag_create(bus_get_dma_tag(sc->sc_dev), 1, 0,
224 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
225 	    BUS_SPACE_MAXSIZE, 0, BUS_SPACE_MAXSIZE, 0, NULL, NULL,
226 	    &sc->sc_pdmatag);
227 	if (error != 0)
228 		goto fail_taskq;
229 
230 	error = bus_dma_tag_create(sc->sc_pdmatag, 1, 0,
231 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
232 	    CAS_PAGE_SIZE, 1, CAS_PAGE_SIZE, 0, NULL, NULL, &sc->sc_rdmatag);
233 	if (error != 0)
234 		goto fail_ptag;
235 
236 	error = bus_dma_tag_create(sc->sc_pdmatag, 1, 0,
237 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
238 	    MCLBYTES * CAS_NTXSEGS, CAS_NTXSEGS, MCLBYTES,
239 	    BUS_DMA_ALLOCNOW, NULL, NULL, &sc->sc_tdmatag);
240 	if (error != 0)
241 		goto fail_rtag;
242 
243 	error = bus_dma_tag_create(sc->sc_pdmatag, CAS_TX_DESC_ALIGN, 0,
244 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
245 	    sizeof(struct cas_control_data), 1,
246 	    sizeof(struct cas_control_data), 0,
247 	    NULL, NULL, &sc->sc_cdmatag);
248 	if (error != 0)
249 		goto fail_ttag;
250 
251 	/*
252 	 * Allocate the control data structures, create and load the
253 	 * DMA map for it.
254 	 */
255 	if ((error = bus_dmamem_alloc(sc->sc_cdmatag,
256 	    (void **)&sc->sc_control_data,
257 	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
258 	    &sc->sc_cddmamap)) != 0) {
259 		device_printf(sc->sc_dev,
260 		    "unable to allocate control data, error = %d\n", error);
261 		goto fail_ctag;
262 	}
263 
264 	sc->sc_cddma = 0;
265 	if ((error = bus_dmamap_load(sc->sc_cdmatag, sc->sc_cddmamap,
266 	    sc->sc_control_data, sizeof(struct cas_control_data),
267 	    cas_cddma_callback, sc, 0)) != 0 || sc->sc_cddma == 0) {
268 		device_printf(sc->sc_dev,
269 		    "unable to load control data DMA map, error = %d\n",
270 		    error);
271 		goto fail_cmem;
272 	}
273 
274 	/*
275 	 * Initialize the transmit job descriptors.
276 	 */
277 	STAILQ_INIT(&sc->sc_txfreeq);
278 	STAILQ_INIT(&sc->sc_txdirtyq);
279 
280 	/*
281 	 * Create the transmit buffer DMA maps.
282 	 */
283 	error = ENOMEM;
284 	for (i = 0; i < CAS_TXQUEUELEN; i++) {
285 		txs = &sc->sc_txsoft[i];
286 		txs->txs_mbuf = NULL;
287 		txs->txs_ndescs = 0;
288 		if ((error = bus_dmamap_create(sc->sc_tdmatag, 0,
289 		    &txs->txs_dmamap)) != 0) {
290 			device_printf(sc->sc_dev,
291 			    "unable to create TX DMA map %d, error = %d\n",
292 			    i, error);
293 			goto fail_txd;
294 		}
295 		STAILQ_INSERT_TAIL(&sc->sc_txfreeq, txs, txs_q);
296 	}
297 
298 	/*
299 	 * Allocate the receive buffers, create and load the DMA maps
300 	 * for them.
301 	 */
302 	for (i = 0; i < CAS_NRXDESC; i++) {
303 		if ((error = bus_dmamem_alloc(sc->sc_rdmatag,
304 		    &sc->sc_rxdsoft[i].rxds_buf, BUS_DMA_WAITOK,
305 		    &sc->sc_rxdsoft[i].rxds_dmamap)) != 0) {
306 			device_printf(sc->sc_dev,
307 			    "unable to allocate RX buffer %d, error = %d\n",
308 			    i, error);
309 			goto fail_rxmem;
310 		}
311 
312 		sc->sc_rxdptr = i;
313 		sc->sc_rxdsoft[i].rxds_paddr = 0;
314 		if ((error = bus_dmamap_load(sc->sc_rdmatag,
315 		    sc->sc_rxdsoft[i].rxds_dmamap, sc->sc_rxdsoft[i].rxds_buf,
316 		    CAS_PAGE_SIZE, cas_rxdma_callback, sc, 0)) != 0 ||
317 		    sc->sc_rxdsoft[i].rxds_paddr == 0) {
318 			device_printf(sc->sc_dev,
319 			    "unable to load RX DMA map %d, error = %d\n",
320 			    i, error);
321 			goto fail_rxmap;
322 		}
323 	}
324 
325 	if ((sc->sc_flags & CAS_SERDES) == 0) {
326 		CAS_WRITE_4(sc, CAS_PCS_DATAPATH, CAS_PCS_DATAPATH_MII);
327 		CAS_BARRIER(sc, CAS_PCS_DATAPATH, 4,
328 		    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
329 		cas_mifinit(sc);
330 		/*
331 		 * Look for an external PHY.
332 		 */
333 		error = ENXIO;
334 		v = CAS_READ_4(sc, CAS_MIF_CONF);
335 		if ((v & CAS_MIF_CONF_MDI1) != 0) {
336 			v |= CAS_MIF_CONF_PHY_SELECT;
337 			CAS_WRITE_4(sc, CAS_MIF_CONF, v);
338 			CAS_BARRIER(sc, CAS_MIF_CONF, 4,
339 			    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
340 			/* Enable/unfreeze the GMII pins of Saturn. */
341 			if (sc->sc_variant == CAS_SATURN) {
342 				CAS_WRITE_4(sc, CAS_SATURN_PCFG,
343 				    CAS_READ_4(sc, CAS_SATURN_PCFG) &
344 				    ~CAS_SATURN_PCFG_FSI);
345 				CAS_BARRIER(sc, CAS_SATURN_PCFG, 4,
346 				    BUS_SPACE_BARRIER_READ |
347 				    BUS_SPACE_BARRIER_WRITE);
348 				DELAY(10000);
349 			}
350 			error = mii_attach(sc->sc_dev, &sc->sc_miibus, ifp,
351 			    cas_mediachange, cas_mediastatus, BMSR_DEFCAPMASK,
352 			    MII_PHY_ANY, MII_OFFSET_ANY, MIIF_DOPAUSE);
353 		}
354 		/*
355 		 * Fall back on an internal PHY if no external PHY was found.
356 		 */
357 		if (error != 0 && (v & CAS_MIF_CONF_MDI0) != 0) {
358 			v &= ~CAS_MIF_CONF_PHY_SELECT;
359 			CAS_WRITE_4(sc, CAS_MIF_CONF, v);
360 			CAS_BARRIER(sc, CAS_MIF_CONF, 4,
361 			    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
362 			/* Freeze the GMII pins of Saturn for saving power. */
363 			if (sc->sc_variant == CAS_SATURN) {
364 				CAS_WRITE_4(sc, CAS_SATURN_PCFG,
365 				    CAS_READ_4(sc, CAS_SATURN_PCFG) |
366 				    CAS_SATURN_PCFG_FSI);
367 				CAS_BARRIER(sc, CAS_SATURN_PCFG, 4,
368 				    BUS_SPACE_BARRIER_READ |
369 				    BUS_SPACE_BARRIER_WRITE);
370 				DELAY(10000);
371 			}
372 			error = mii_attach(sc->sc_dev, &sc->sc_miibus, ifp,
373 			    cas_mediachange, cas_mediastatus, BMSR_DEFCAPMASK,
374 			    MII_PHY_ANY, MII_OFFSET_ANY, MIIF_DOPAUSE);
375 		}
376 	} else {
377 		/*
378 		 * Use the external PCS SERDES.
379 		 */
380 		CAS_WRITE_4(sc, CAS_PCS_DATAPATH, CAS_PCS_DATAPATH_SERDES);
381 		CAS_BARRIER(sc, CAS_PCS_DATAPATH, 4, BUS_SPACE_BARRIER_WRITE);
382 		/* Enable/unfreeze the SERDES pins of Saturn. */
383 		if (sc->sc_variant == CAS_SATURN) {
384 			CAS_WRITE_4(sc, CAS_SATURN_PCFG, 0);
385 			CAS_BARRIER(sc, CAS_SATURN_PCFG, 4,
386 			    BUS_SPACE_BARRIER_WRITE);
387 		}
388 		CAS_WRITE_4(sc, CAS_PCS_SERDES_CTRL, CAS_PCS_SERDES_CTRL_ESD);
389 		CAS_BARRIER(sc, CAS_PCS_SERDES_CTRL, 4,
390 		    BUS_SPACE_BARRIER_WRITE);
391 		CAS_WRITE_4(sc, CAS_PCS_CONF, CAS_PCS_CONF_EN);
392 		CAS_BARRIER(sc, CAS_PCS_CONF, 4,
393 		    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
394 		error = mii_attach(sc->sc_dev, &sc->sc_miibus, ifp,
395 		    cas_mediachange, cas_mediastatus, BMSR_DEFCAPMASK,
396 		    CAS_PHYAD_EXTERNAL, MII_OFFSET_ANY, MIIF_DOPAUSE);
397 	}
398 	if (error != 0) {
399 		device_printf(sc->sc_dev, "attaching PHYs failed\n");
400 		goto fail_rxmap;
401 	}
402 	sc->sc_mii = device_get_softc(sc->sc_miibus);
403 
404 	/*
405 	 * From this point forward, the attachment cannot fail.  A failure
406 	 * before this point releases all resources that may have been
407 	 * allocated.
408 	 */
409 
410 	/* Announce FIFO sizes. */
411 	v = CAS_READ_4(sc, CAS_TX_FIFO_SIZE);
412 	device_printf(sc->sc_dev, "%ukB RX FIFO, %ukB TX FIFO\n",
413 	    CAS_RX_FIFO_SIZE / 1024, v / 16);
414 
415 	/* Attach the interface. */
416 	ether_ifattach(ifp, sc->sc_enaddr);
417 
418 	/*
419 	 * Tell the upper layer(s) we support long frames/checksum offloads.
420 	 */
421 	if_setifheaderlen(ifp, sizeof(struct ether_vlan_header));
422 	if_setcapabilities(ifp, IFCAP_VLAN_MTU);
423 	if ((sc->sc_flags & CAS_NO_CSUM) == 0) {
424 		if_setcapabilitiesbit(ifp, IFCAP_HWCSUM, 0);
425 		if_sethwassist(ifp, CAS_CSUM_FEATURES);
426 	}
427 	if_setcapenable(ifp, if_getcapabilities(ifp));
428 
429 	return (0);
430 
431 	/*
432 	 * Free any resources we've allocated during the failed attach
433 	 * attempt.  Do this in reverse order and fall through.
434 	 */
435  fail_rxmap:
436 	for (i = 0; i < CAS_NRXDESC; i++)
437 		if (sc->sc_rxdsoft[i].rxds_paddr != 0)
438 			bus_dmamap_unload(sc->sc_rdmatag,
439 			    sc->sc_rxdsoft[i].rxds_dmamap);
440  fail_rxmem:
441 	for (i = 0; i < CAS_NRXDESC; i++)
442 		if (sc->sc_rxdsoft[i].rxds_buf != NULL)
443 			bus_dmamem_free(sc->sc_rdmatag,
444 			    sc->sc_rxdsoft[i].rxds_buf,
445 			    sc->sc_rxdsoft[i].rxds_dmamap);
446  fail_txd:
447 	for (i = 0; i < CAS_TXQUEUELEN; i++)
448 		if (sc->sc_txsoft[i].txs_dmamap != NULL)
449 			bus_dmamap_destroy(sc->sc_tdmatag,
450 			    sc->sc_txsoft[i].txs_dmamap);
451 	bus_dmamap_unload(sc->sc_cdmatag, sc->sc_cddmamap);
452  fail_cmem:
453 	bus_dmamem_free(sc->sc_cdmatag, sc->sc_control_data,
454 	    sc->sc_cddmamap);
455  fail_ctag:
456 	bus_dma_tag_destroy(sc->sc_cdmatag);
457  fail_ttag:
458 	bus_dma_tag_destroy(sc->sc_tdmatag);
459  fail_rtag:
460 	bus_dma_tag_destroy(sc->sc_rdmatag);
461  fail_ptag:
462 	bus_dma_tag_destroy(sc->sc_pdmatag);
463  fail_taskq:
464 	taskqueue_free(sc->sc_tq);
465  fail_ifnet:
466 	if_free(ifp);
467 	return (error);
468 }
469 
470 static void
471 cas_detach(struct cas_softc *sc)
472 {
473 	if_t ifp = sc->sc_ifp;
474 	int i;
475 
476 	ether_ifdetach(ifp);
477 	CAS_LOCK(sc);
478 	cas_stop(ifp);
479 	CAS_UNLOCK(sc);
480 	callout_drain(&sc->sc_tick_ch);
481 	callout_drain(&sc->sc_rx_ch);
482 	taskqueue_drain(sc->sc_tq, &sc->sc_intr_task);
483 	taskqueue_drain(sc->sc_tq, &sc->sc_tx_task);
484 	if_free(ifp);
485 	taskqueue_free(sc->sc_tq);
486 	device_delete_child(sc->sc_dev, sc->sc_miibus);
487 
488 	for (i = 0; i < CAS_NRXDESC; i++)
489 		if (sc->sc_rxdsoft[i].rxds_dmamap != NULL)
490 			bus_dmamap_sync(sc->sc_rdmatag,
491 			    sc->sc_rxdsoft[i].rxds_dmamap,
492 			    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
493 	for (i = 0; i < CAS_NRXDESC; i++)
494 		if (sc->sc_rxdsoft[i].rxds_paddr != 0)
495 			bus_dmamap_unload(sc->sc_rdmatag,
496 			    sc->sc_rxdsoft[i].rxds_dmamap);
497 	for (i = 0; i < CAS_NRXDESC; i++)
498 		if (sc->sc_rxdsoft[i].rxds_buf != NULL)
499 			bus_dmamem_free(sc->sc_rdmatag,
500 			    sc->sc_rxdsoft[i].rxds_buf,
501 			    sc->sc_rxdsoft[i].rxds_dmamap);
502 	for (i = 0; i < CAS_TXQUEUELEN; i++)
503 		if (sc->sc_txsoft[i].txs_dmamap != NULL)
504 			bus_dmamap_destroy(sc->sc_tdmatag,
505 			    sc->sc_txsoft[i].txs_dmamap);
506 	CAS_CDSYNC(sc, BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
507 	bus_dmamap_unload(sc->sc_cdmatag, sc->sc_cddmamap);
508 	bus_dmamem_free(sc->sc_cdmatag, sc->sc_control_data,
509 	    sc->sc_cddmamap);
510 	bus_dma_tag_destroy(sc->sc_cdmatag);
511 	bus_dma_tag_destroy(sc->sc_tdmatag);
512 	bus_dma_tag_destroy(sc->sc_rdmatag);
513 	bus_dma_tag_destroy(sc->sc_pdmatag);
514 }
515 
516 static void
517 cas_suspend(struct cas_softc *sc)
518 {
519 	if_t ifp = sc->sc_ifp;
520 
521 	CAS_LOCK(sc);
522 	cas_stop(ifp);
523 	CAS_UNLOCK(sc);
524 }
525 
526 static void
527 cas_resume(struct cas_softc *sc)
528 {
529 	if_t ifp = sc->sc_ifp;
530 
531 	CAS_LOCK(sc);
532 	/*
533 	 * On resume all registers have to be initialized again like
534 	 * after power-on.
535 	 */
536 	sc->sc_flags &= ~CAS_INITED;
537 	if (if_getflags(ifp) & IFF_UP)
538 		cas_init_locked(sc);
539 	CAS_UNLOCK(sc);
540 }
541 
542 static inline void
543 cas_rxcksum(struct mbuf *m, uint16_t cksum)
544 {
545 	struct ether_header *eh;
546 	struct ip *ip;
547 	struct udphdr *uh;
548 	uint16_t *opts;
549 	int32_t hlen, len, pktlen;
550 	uint32_t temp32;
551 
552 	pktlen = m->m_pkthdr.len;
553 	if (pktlen < sizeof(struct ether_header) + sizeof(struct ip))
554 		return;
555 	eh = mtod(m, struct ether_header *);
556 	if (eh->ether_type != htons(ETHERTYPE_IP))
557 		return;
558 	ip = (struct ip *)(eh + 1);
559 	if (ip->ip_v != IPVERSION)
560 		return;
561 
562 	hlen = ip->ip_hl << 2;
563 	pktlen -= sizeof(struct ether_header);
564 	if (hlen < sizeof(struct ip))
565 		return;
566 	if (ntohs(ip->ip_len) < hlen)
567 		return;
568 	if (ntohs(ip->ip_len) != pktlen)
569 		return;
570 	if (ip->ip_off & htons(IP_MF | IP_OFFMASK))
571 		return;	/* Cannot handle fragmented packet. */
572 
573 	switch (ip->ip_p) {
574 	case IPPROTO_TCP:
575 		if (pktlen < (hlen + sizeof(struct tcphdr)))
576 			return;
577 		break;
578 	case IPPROTO_UDP:
579 		if (pktlen < (hlen + sizeof(struct udphdr)))
580 			return;
581 		uh = (struct udphdr *)((uint8_t *)ip + hlen);
582 		if (uh->uh_sum == 0)
583 			return; /* no checksum */
584 		break;
585 	default:
586 		return;
587 	}
588 
589 	cksum = ~cksum;
590 	/* checksum fixup for IP options */
591 	len = hlen - sizeof(struct ip);
592 	if (len > 0) {
593 		opts = (uint16_t *)(ip + 1);
594 		for (; len > 0; len -= sizeof(uint16_t), opts++) {
595 			temp32 = cksum - *opts;
596 			temp32 = (temp32 >> 16) + (temp32 & 65535);
597 			cksum = temp32 & 65535;
598 		}
599 	}
600 	m->m_pkthdr.csum_flags |= CSUM_DATA_VALID;
601 	m->m_pkthdr.csum_data = cksum;
602 }
603 
604 static void
605 cas_cddma_callback(void *xsc, bus_dma_segment_t *segs, int nsegs, int error)
606 {
607 	struct cas_softc *sc = xsc;
608 
609 	if (error != 0)
610 		return;
611 	if (nsegs != 1)
612 		panic("%s: bad control buffer segment count", __func__);
613 	sc->sc_cddma = segs[0].ds_addr;
614 }
615 
616 static void
617 cas_rxdma_callback(void *xsc, bus_dma_segment_t *segs, int nsegs, int error)
618 {
619 	struct cas_softc *sc = xsc;
620 
621 	if (error != 0)
622 		return;
623 	if (nsegs != 1)
624 		panic("%s: bad RX buffer segment count", __func__);
625 	sc->sc_rxdsoft[sc->sc_rxdptr].rxds_paddr = segs[0].ds_addr;
626 }
627 
628 static void
629 cas_tick(void *arg)
630 {
631 	struct cas_softc *sc = arg;
632 	if_t ifp = sc->sc_ifp;
633 	uint32_t v;
634 
635 	CAS_LOCK_ASSERT(sc, MA_OWNED);
636 
637 	/*
638 	 * Unload collision and error counters.
639 	 */
640 	if_inc_counter(ifp, IFCOUNTER_COLLISIONS,
641 	    CAS_READ_4(sc, CAS_MAC_NORM_COLL_CNT) +
642 	    CAS_READ_4(sc, CAS_MAC_FIRST_COLL_CNT));
643 	v = CAS_READ_4(sc, CAS_MAC_EXCESS_COLL_CNT) +
644 	    CAS_READ_4(sc, CAS_MAC_LATE_COLL_CNT);
645 	if_inc_counter(ifp, IFCOUNTER_COLLISIONS, v);
646 	if_inc_counter(ifp, IFCOUNTER_OERRORS, v);
647 	if_inc_counter(ifp, IFCOUNTER_IERRORS,
648 	    CAS_READ_4(sc, CAS_MAC_RX_LEN_ERR_CNT) +
649 	    CAS_READ_4(sc, CAS_MAC_RX_ALIGN_ERR) +
650 	    CAS_READ_4(sc, CAS_MAC_RX_CRC_ERR_CNT) +
651 	    CAS_READ_4(sc, CAS_MAC_RX_CODE_VIOL));
652 
653 	/*
654 	 * Then clear the hardware counters.
655 	 */
656 	CAS_WRITE_4(sc, CAS_MAC_NORM_COLL_CNT, 0);
657 	CAS_WRITE_4(sc, CAS_MAC_FIRST_COLL_CNT, 0);
658 	CAS_WRITE_4(sc, CAS_MAC_EXCESS_COLL_CNT, 0);
659 	CAS_WRITE_4(sc, CAS_MAC_LATE_COLL_CNT, 0);
660 	CAS_WRITE_4(sc, CAS_MAC_RX_LEN_ERR_CNT, 0);
661 	CAS_WRITE_4(sc, CAS_MAC_RX_ALIGN_ERR, 0);
662 	CAS_WRITE_4(sc, CAS_MAC_RX_CRC_ERR_CNT, 0);
663 	CAS_WRITE_4(sc, CAS_MAC_RX_CODE_VIOL, 0);
664 
665 	mii_tick(sc->sc_mii);
666 
667 	if (sc->sc_txfree != CAS_MAXTXFREE)
668 		cas_tint(sc);
669 
670 	cas_watchdog(sc);
671 
672 	callout_reset(&sc->sc_tick_ch, hz, cas_tick, sc);
673 }
674 
675 static int
676 cas_bitwait(struct cas_softc *sc, bus_addr_t r, uint32_t clr, uint32_t set)
677 {
678 	int i;
679 	uint32_t reg;
680 
681 	for (i = CAS_TRIES; i--; DELAY(100)) {
682 		reg = CAS_READ_4(sc, r);
683 		if ((reg & clr) == 0 && (reg & set) == set)
684 			return (1);
685 	}
686 	return (0);
687 }
688 
689 static void
690 cas_reset(struct cas_softc *sc)
691 {
692 
693 #ifdef CAS_DEBUG
694 	CTR2(KTR_CAS, "%s: %s", device_get_name(sc->sc_dev), __func__);
695 #endif
696 	/* Disable all interrupts in order to avoid spurious ones. */
697 	CAS_WRITE_4(sc, CAS_INTMASK, 0xffffffff);
698 
699 	cas_reset_rx(sc);
700 	cas_reset_tx(sc);
701 
702 	/*
703 	 * Do a full reset modulo the result of the last auto-negotiation
704 	 * when using the SERDES.
705 	 */
706 	CAS_WRITE_4(sc, CAS_RESET, CAS_RESET_RX | CAS_RESET_TX |
707 	    ((sc->sc_flags & CAS_SERDES) != 0 ? CAS_RESET_PCS_DIS : 0));
708 	CAS_BARRIER(sc, CAS_RESET, 4,
709 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
710 	DELAY(3000);
711 	if (!cas_bitwait(sc, CAS_RESET, CAS_RESET_RX | CAS_RESET_TX, 0))
712 		device_printf(sc->sc_dev, "cannot reset device\n");
713 }
714 
715 static void
716 cas_stop(if_t ifp)
717 {
718 	struct cas_softc *sc = if_getsoftc(ifp);
719 	struct cas_txsoft *txs;
720 
721 #ifdef CAS_DEBUG
722 	CTR2(KTR_CAS, "%s: %s", device_get_name(sc->sc_dev), __func__);
723 #endif
724 
725 	callout_stop(&sc->sc_tick_ch);
726 	callout_stop(&sc->sc_rx_ch);
727 
728 	/* Disable all interrupts in order to avoid spurious ones. */
729 	CAS_WRITE_4(sc, CAS_INTMASK, 0xffffffff);
730 
731 	cas_reset_tx(sc);
732 	cas_reset_rx(sc);
733 
734 	/*
735 	 * Release any queued transmit buffers.
736 	 */
737 	while ((txs = STAILQ_FIRST(&sc->sc_txdirtyq)) != NULL) {
738 		STAILQ_REMOVE_HEAD(&sc->sc_txdirtyq, txs_q);
739 		if (txs->txs_ndescs != 0) {
740 			bus_dmamap_sync(sc->sc_tdmatag, txs->txs_dmamap,
741 			    BUS_DMASYNC_POSTWRITE);
742 			bus_dmamap_unload(sc->sc_tdmatag, txs->txs_dmamap);
743 			if (txs->txs_mbuf != NULL) {
744 				m_freem(txs->txs_mbuf);
745 				txs->txs_mbuf = NULL;
746 			}
747 		}
748 		STAILQ_INSERT_TAIL(&sc->sc_txfreeq, txs, txs_q);
749 	}
750 
751 	/*
752 	 * Mark the interface down and cancel the watchdog timer.
753 	 */
754 	if_setdrvflagbits(ifp, 0, (IFF_DRV_RUNNING | IFF_DRV_OACTIVE));
755 	sc->sc_flags &= ~CAS_LINK;
756 	sc->sc_wdog_timer = 0;
757 }
758 
759 static int
760 cas_reset_rx(struct cas_softc *sc)
761 {
762 
763 	/*
764 	 * Resetting while DMA is in progress can cause a bus hang, so we
765 	 * disable DMA first.
766 	 */
767 	(void)cas_disable_rx(sc);
768 	CAS_WRITE_4(sc, CAS_RX_CONF, 0);
769 	CAS_BARRIER(sc, CAS_RX_CONF, 4,
770 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
771 	if (!cas_bitwait(sc, CAS_RX_CONF, CAS_RX_CONF_RXDMA_EN, 0))
772 		device_printf(sc->sc_dev, "cannot disable RX DMA\n");
773 
774 	/* Finally, reset the ERX. */
775 	CAS_WRITE_4(sc, CAS_RESET, CAS_RESET_RX |
776 	    ((sc->sc_flags & CAS_SERDES) != 0 ? CAS_RESET_PCS_DIS : 0));
777 	CAS_BARRIER(sc, CAS_RESET, 4,
778 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
779 	if (!cas_bitwait(sc, CAS_RESET, CAS_RESET_RX, 0)) {
780 		device_printf(sc->sc_dev, "cannot reset receiver\n");
781 		return (1);
782 	}
783 	return (0);
784 }
785 
786 static int
787 cas_reset_tx(struct cas_softc *sc)
788 {
789 
790 	/*
791 	 * Resetting while DMA is in progress can cause a bus hang, so we
792 	 * disable DMA first.
793 	 */
794 	(void)cas_disable_tx(sc);
795 	CAS_WRITE_4(sc, CAS_TX_CONF, 0);
796 	CAS_BARRIER(sc, CAS_TX_CONF, 4,
797 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
798 	if (!cas_bitwait(sc, CAS_TX_CONF, CAS_TX_CONF_TXDMA_EN, 0))
799 		device_printf(sc->sc_dev, "cannot disable TX DMA\n");
800 
801 	/* Finally, reset the ETX. */
802 	CAS_WRITE_4(sc, CAS_RESET, CAS_RESET_TX |
803 	    ((sc->sc_flags & CAS_SERDES) != 0 ? CAS_RESET_PCS_DIS : 0));
804 	CAS_BARRIER(sc, CAS_RESET, 4,
805 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
806 	if (!cas_bitwait(sc, CAS_RESET, CAS_RESET_TX, 0)) {
807 		device_printf(sc->sc_dev, "cannot reset transmitter\n");
808 		return (1);
809 	}
810 	return (0);
811 }
812 
813 static int
814 cas_disable_rx(struct cas_softc *sc)
815 {
816 
817 	CAS_WRITE_4(sc, CAS_MAC_RX_CONF,
818 	    CAS_READ_4(sc, CAS_MAC_RX_CONF) & ~CAS_MAC_RX_CONF_EN);
819 	CAS_BARRIER(sc, CAS_MAC_RX_CONF, 4,
820 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
821 	if (cas_bitwait(sc, CAS_MAC_RX_CONF, CAS_MAC_RX_CONF_EN, 0))
822 		return (1);
823 	if (bootverbose)
824 		device_printf(sc->sc_dev, "cannot disable RX MAC\n");
825 	return (0);
826 }
827 
828 static int
829 cas_disable_tx(struct cas_softc *sc)
830 {
831 
832 	CAS_WRITE_4(sc, CAS_MAC_TX_CONF,
833 	    CAS_READ_4(sc, CAS_MAC_TX_CONF) & ~CAS_MAC_TX_CONF_EN);
834 	CAS_BARRIER(sc, CAS_MAC_TX_CONF, 4,
835 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
836 	if (cas_bitwait(sc, CAS_MAC_TX_CONF, CAS_MAC_TX_CONF_EN, 0))
837 		return (1);
838 	if (bootverbose)
839 		device_printf(sc->sc_dev, "cannot disable TX MAC\n");
840 	return (0);
841 }
842 
843 static inline void
844 cas_rxcompinit(struct cas_rx_comp *rxcomp)
845 {
846 
847 	rxcomp->crc_word1 = 0;
848 	rxcomp->crc_word2 = 0;
849 	rxcomp->crc_word3 =
850 	    htole64(CAS_SET(ETHER_HDR_LEN + sizeof(struct ip), CAS_RC3_CSO));
851 	rxcomp->crc_word4 = htole64(CAS_RC4_ZERO);
852 }
853 
854 static void
855 cas_meminit(struct cas_softc *sc)
856 {
857 	int i;
858 
859 	CAS_LOCK_ASSERT(sc, MA_OWNED);
860 
861 	/*
862 	 * Initialize the transmit descriptor ring.
863 	 */
864 	for (i = 0; i < CAS_NTXDESC; i++) {
865 		sc->sc_txdescs[i].cd_flags = 0;
866 		sc->sc_txdescs[i].cd_buf_ptr = 0;
867 	}
868 	sc->sc_txfree = CAS_MAXTXFREE;
869 	sc->sc_txnext = 0;
870 	sc->sc_txwin = 0;
871 
872 	/*
873 	 * Initialize the receive completion ring.
874 	 */
875 	for (i = 0; i < CAS_NRXCOMP; i++)
876 		cas_rxcompinit(&sc->sc_rxcomps[i]);
877 	sc->sc_rxcptr = 0;
878 
879 	/*
880 	 * Initialize the first receive descriptor ring.  We leave
881 	 * the second one zeroed as we don't actually use it.
882 	 */
883 	for (i = 0; i < CAS_NRXDESC; i++)
884 		CAS_INIT_RXDESC(sc, i, i);
885 	sc->sc_rxdptr = 0;
886 
887 	CAS_CDSYNC(sc, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
888 }
889 
890 static u_int
891 cas_descsize(u_int sz)
892 {
893 
894 	switch (sz) {
895 	case 32:
896 		return (CAS_DESC_32);
897 	case 64:
898 		return (CAS_DESC_64);
899 	case 128:
900 		return (CAS_DESC_128);
901 	case 256:
902 		return (CAS_DESC_256);
903 	case 512:
904 		return (CAS_DESC_512);
905 	case 1024:
906 		return (CAS_DESC_1K);
907 	case 2048:
908 		return (CAS_DESC_2K);
909 	case 4096:
910 		return (CAS_DESC_4K);
911 	case 8192:
912 		return (CAS_DESC_8K);
913 	default:
914 		printf("%s: invalid descriptor ring size %d\n", __func__, sz);
915 		return (CAS_DESC_32);
916 	}
917 }
918 
919 static u_int
920 cas_rxcompsize(u_int sz)
921 {
922 
923 	switch (sz) {
924 	case 128:
925 		return (CAS_RX_CONF_COMP_128);
926 	case 256:
927 		return (CAS_RX_CONF_COMP_256);
928 	case 512:
929 		return (CAS_RX_CONF_COMP_512);
930 	case 1024:
931 		return (CAS_RX_CONF_COMP_1K);
932 	case 2048:
933 		return (CAS_RX_CONF_COMP_2K);
934 	case 4096:
935 		return (CAS_RX_CONF_COMP_4K);
936 	case 8192:
937 		return (CAS_RX_CONF_COMP_8K);
938 	case 16384:
939 		return (CAS_RX_CONF_COMP_16K);
940 	case 32768:
941 		return (CAS_RX_CONF_COMP_32K);
942 	default:
943 		printf("%s: invalid dcompletion ring size %d\n", __func__, sz);
944 		return (CAS_RX_CONF_COMP_128);
945 	}
946 }
947 
948 static void
949 cas_init(void *xsc)
950 {
951 	struct cas_softc *sc = xsc;
952 
953 	CAS_LOCK(sc);
954 	cas_init_locked(sc);
955 	CAS_UNLOCK(sc);
956 }
957 
958 /*
959  * Initialization of interface; set up initialization block
960  * and transmit/receive descriptor rings.
961  */
962 static void
963 cas_init_locked(struct cas_softc *sc)
964 {
965 	if_t ifp = sc->sc_ifp;
966 	uint32_t v;
967 
968 	CAS_LOCK_ASSERT(sc, MA_OWNED);
969 
970 	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0)
971 		return;
972 
973 #ifdef CAS_DEBUG
974 	CTR2(KTR_CAS, "%s: %s: calling stop", device_get_name(sc->sc_dev),
975 	    __func__);
976 #endif
977 	/*
978 	 * Initialization sequence.  The numbered steps below correspond
979 	 * to the sequence outlined in section 6.3.5.1 in the Ethernet
980 	 * Channel Engine manual (part of the PCIO manual).
981 	 * See also the STP2002-STQ document from Sun Microsystems.
982 	 */
983 
984 	/* step 1 & 2.  Reset the Ethernet Channel. */
985 	cas_stop(ifp);
986 	cas_reset(sc);
987 #ifdef CAS_DEBUG
988 	CTR2(KTR_CAS, "%s: %s: restarting", device_get_name(sc->sc_dev),
989 	    __func__);
990 #endif
991 
992 	if ((sc->sc_flags & CAS_SERDES) == 0)
993 		/* Re-initialize the MIF. */
994 		cas_mifinit(sc);
995 
996 	/* step 3.  Setup data structures in host memory. */
997 	cas_meminit(sc);
998 
999 	/* step 4.  TX MAC registers & counters */
1000 	cas_init_regs(sc);
1001 
1002 	/* step 5.  RX MAC registers & counters */
1003 
1004 	/* step 6 & 7.  Program Ring Base Addresses. */
1005 	CAS_WRITE_4(sc, CAS_TX_DESC3_BASE_HI,
1006 	    (((uint64_t)CAS_CDTXDADDR(sc, 0)) >> 32));
1007 	CAS_WRITE_4(sc, CAS_TX_DESC3_BASE_LO,
1008 	    CAS_CDTXDADDR(sc, 0) & 0xffffffff);
1009 
1010 	CAS_WRITE_4(sc, CAS_RX_COMP_BASE_HI,
1011 	    (((uint64_t)CAS_CDRXCADDR(sc, 0)) >> 32));
1012 	CAS_WRITE_4(sc, CAS_RX_COMP_BASE_LO,
1013 	    CAS_CDRXCADDR(sc, 0) & 0xffffffff);
1014 
1015 	CAS_WRITE_4(sc, CAS_RX_DESC_BASE_HI,
1016 	    (((uint64_t)CAS_CDRXDADDR(sc, 0)) >> 32));
1017 	CAS_WRITE_4(sc, CAS_RX_DESC_BASE_LO,
1018 	    CAS_CDRXDADDR(sc, 0) & 0xffffffff);
1019 
1020 	if ((sc->sc_flags & CAS_REG_PLUS) != 0) {
1021 		CAS_WRITE_4(sc, CAS_RX_DESC2_BASE_HI,
1022 		    (((uint64_t)CAS_CDRXD2ADDR(sc, 0)) >> 32));
1023 		CAS_WRITE_4(sc, CAS_RX_DESC2_BASE_LO,
1024 		    CAS_CDRXD2ADDR(sc, 0) & 0xffffffff);
1025 	}
1026 
1027 #ifdef CAS_DEBUG
1028 	CTR5(KTR_CAS,
1029 	    "loading TXDR %lx, RXCR %lx, RXDR %lx, RXD2R %lx, cddma %lx",
1030 	    CAS_CDTXDADDR(sc, 0), CAS_CDRXCADDR(sc, 0), CAS_CDRXDADDR(sc, 0),
1031 	    CAS_CDRXD2ADDR(sc, 0), sc->sc_cddma);
1032 #endif
1033 
1034 	/* step 8.  Global Configuration & Interrupt Masks */
1035 
1036 	/* Disable weighted round robin. */
1037 	CAS_WRITE_4(sc, CAS_CAW, CAS_CAW_RR_DIS);
1038 
1039 	/*
1040 	 * Enable infinite bursts for revisions without PCI issues if
1041 	 * applicable.  Doing so greatly improves the TX performance.
1042 	 */
1043 	CAS_WRITE_4(sc, CAS_INF_BURST,
1044 	    (sc->sc_flags & CAS_TABORT) == 0 ? CAS_INF_BURST_EN :
1045 	    0);
1046 
1047 	/* Set up interrupts. */
1048 	CAS_WRITE_4(sc, CAS_INTMASK,
1049 	    ~(CAS_INTR_TX_INT_ME | CAS_INTR_TX_TAG_ERR |
1050 	    CAS_INTR_RX_DONE | CAS_INTR_RX_BUF_NA | CAS_INTR_RX_TAG_ERR |
1051 	    CAS_INTR_RX_COMP_FULL | CAS_INTR_RX_BUF_AEMPTY |
1052 	    CAS_INTR_RX_COMP_AFULL | CAS_INTR_RX_LEN_MMATCH |
1053 	    CAS_INTR_PCI_ERROR_INT
1054 #ifdef CAS_DEBUG
1055 	    | CAS_INTR_PCS_INT | CAS_INTR_MIF
1056 #endif
1057 	    ));
1058 	/* Don't clear top level interrupts when CAS_STATUS_ALIAS is read. */
1059 	CAS_WRITE_4(sc, CAS_CLEAR_ALIAS, 0);
1060 	CAS_WRITE_4(sc, CAS_MAC_RX_MASK, ~CAS_MAC_RX_OVERFLOW);
1061 	CAS_WRITE_4(sc, CAS_MAC_TX_MASK,
1062 	    ~(CAS_MAC_TX_UNDERRUN | CAS_MAC_TX_MAX_PKT_ERR));
1063 #ifdef CAS_DEBUG
1064 	CAS_WRITE_4(sc, CAS_MAC_CTRL_MASK,
1065 	    ~(CAS_MAC_CTRL_PAUSE_RCVD | CAS_MAC_CTRL_PAUSE |
1066 	    CAS_MAC_CTRL_NON_PAUSE));
1067 #else
1068 	CAS_WRITE_4(sc, CAS_MAC_CTRL_MASK,
1069 	    CAS_MAC_CTRL_PAUSE_RCVD | CAS_MAC_CTRL_PAUSE |
1070 	    CAS_MAC_CTRL_NON_PAUSE);
1071 #endif
1072 
1073 	/* Enable PCI error interrupts. */
1074 	CAS_WRITE_4(sc, CAS_ERROR_MASK,
1075 	    ~(CAS_ERROR_DTRTO | CAS_ERROR_OTHER | CAS_ERROR_DMAW_ZERO |
1076 	    CAS_ERROR_DMAR_ZERO | CAS_ERROR_RTRTO));
1077 
1078 	/* Enable PCI error interrupts in BIM configuration. */
1079 	CAS_WRITE_4(sc, CAS_BIM_CONF,
1080 	    CAS_BIM_CONF_DPAR_EN | CAS_BIM_CONF_RMA_EN | CAS_BIM_CONF_RTA_EN);
1081 
1082 	/*
1083 	 * step 9.  ETX Configuration: encode receive descriptor ring size,
1084 	 * enable DMA and disable pre-interrupt writeback completion.
1085 	 */
1086 	v = cas_descsize(CAS_NTXDESC) << CAS_TX_CONF_DESC3_SHFT;
1087 	CAS_WRITE_4(sc, CAS_TX_CONF, v | CAS_TX_CONF_TXDMA_EN |
1088 	    CAS_TX_CONF_RDPP_DIS | CAS_TX_CONF_PICWB_DIS);
1089 
1090 	/* step 10.  ERX Configuration */
1091 
1092 	/*
1093 	 * Encode receive completion and descriptor ring sizes, set the
1094 	 * swivel offset.
1095 	 */
1096 	v = cas_rxcompsize(CAS_NRXCOMP) << CAS_RX_CONF_COMP_SHFT;
1097 	v |= cas_descsize(CAS_NRXDESC) << CAS_RX_CONF_DESC_SHFT;
1098 	if ((sc->sc_flags & CAS_REG_PLUS) != 0)
1099 		v |= cas_descsize(CAS_NRXDESC2) << CAS_RX_CONF_DESC2_SHFT;
1100 	CAS_WRITE_4(sc, CAS_RX_CONF,
1101 	    v | (ETHER_ALIGN << CAS_RX_CONF_SOFF_SHFT));
1102 
1103 	/* Set the PAUSE thresholds.  We use the maximum OFF threshold. */
1104 	CAS_WRITE_4(sc, CAS_RX_PTHRS,
1105 	    (111 << CAS_RX_PTHRS_XOFF_SHFT) | (15 << CAS_RX_PTHRS_XON_SHFT));
1106 
1107 	/* RX blanking */
1108 	CAS_WRITE_4(sc, CAS_RX_BLANK,
1109 	    (15 << CAS_RX_BLANK_TIME_SHFT) | (5 << CAS_RX_BLANK_PKTS_SHFT));
1110 
1111 	/* Set RX_COMP_AFULL threshold to half of the RX completions. */
1112 	CAS_WRITE_4(sc, CAS_RX_AEMPTY_THRS,
1113 	    (CAS_NRXCOMP / 2) << CAS_RX_AEMPTY_COMP_SHFT);
1114 
1115 	/* Initialize the RX page size register as appropriate for 8k. */
1116 	CAS_WRITE_4(sc, CAS_RX_PSZ,
1117 	    (CAS_RX_PSZ_8K << CAS_RX_PSZ_SHFT) |
1118 	    (4 << CAS_RX_PSZ_MB_CNT_SHFT) |
1119 	    (CAS_RX_PSZ_MB_STRD_2K << CAS_RX_PSZ_MB_STRD_SHFT) |
1120 	    (CAS_RX_PSZ_MB_OFF_64 << CAS_RX_PSZ_MB_OFF_SHFT));
1121 
1122 	/* Disable RX random early detection. */
1123 	CAS_WRITE_4(sc,	CAS_RX_RED, 0);
1124 
1125 	/* Zero the RX reassembly DMA table. */
1126 	for (v = 0; v <= CAS_RX_REAS_DMA_ADDR_LC; v++) {
1127 		CAS_WRITE_4(sc,	CAS_RX_REAS_DMA_ADDR, v);
1128 		CAS_WRITE_4(sc,	CAS_RX_REAS_DMA_DATA_LO, 0);
1129 		CAS_WRITE_4(sc,	CAS_RX_REAS_DMA_DATA_MD, 0);
1130 		CAS_WRITE_4(sc,	CAS_RX_REAS_DMA_DATA_HI, 0);
1131 	}
1132 
1133 	/* Ensure the RX control FIFO and RX IPP FIFO addresses are zero. */
1134 	CAS_WRITE_4(sc, CAS_RX_CTRL_FIFO, 0);
1135 	CAS_WRITE_4(sc, CAS_RX_IPP_ADDR, 0);
1136 
1137 	/* Finally, enable RX DMA. */
1138 	CAS_WRITE_4(sc, CAS_RX_CONF,
1139 	    CAS_READ_4(sc, CAS_RX_CONF) | CAS_RX_CONF_RXDMA_EN);
1140 
1141 	/* step 11.  Configure Media. */
1142 
1143 	/* step 12.  RX_MAC Configuration Register */
1144 	v = CAS_READ_4(sc, CAS_MAC_RX_CONF);
1145 	v &= ~(CAS_MAC_RX_CONF_STRPPAD | CAS_MAC_RX_CONF_EN);
1146 	v |= CAS_MAC_RX_CONF_STRPFCS;
1147 	sc->sc_mac_rxcfg = v;
1148 	/*
1149 	 * Clear the RX filter and reprogram it.  This will also set the
1150 	 * current RX MAC configuration and enable it.
1151 	 */
1152 	cas_setladrf(sc);
1153 
1154 	/* step 13.  TX_MAC Configuration Register */
1155 	v = CAS_READ_4(sc, CAS_MAC_TX_CONF);
1156 	v |= CAS_MAC_TX_CONF_EN;
1157 	(void)cas_disable_tx(sc);
1158 	CAS_WRITE_4(sc, CAS_MAC_TX_CONF, v);
1159 
1160 	/* step 14.  Issue Transmit Pending command. */
1161 
1162 	/* step 15.  Give the receiver a swift kick. */
1163 	CAS_WRITE_4(sc, CAS_RX_KICK, CAS_NRXDESC - 4);
1164 	CAS_WRITE_4(sc, CAS_RX_COMP_TAIL, 0);
1165 	if ((sc->sc_flags & CAS_REG_PLUS) != 0)
1166 		CAS_WRITE_4(sc, CAS_RX_KICK2, CAS_NRXDESC2 - 4);
1167 
1168 	if_setdrvflagbits(ifp, IFF_DRV_RUNNING, 0);
1169 	if_setdrvflagbits(ifp, 0, IFF_DRV_OACTIVE);
1170 
1171 	mii_mediachg(sc->sc_mii);
1172 
1173 	/* Start the one second timer. */
1174 	sc->sc_wdog_timer = 0;
1175 	callout_reset(&sc->sc_tick_ch, hz, cas_tick, sc);
1176 }
1177 
1178 static int
1179 cas_load_txmbuf(struct cas_softc *sc, struct mbuf **m_head)
1180 {
1181 	bus_dma_segment_t txsegs[CAS_NTXSEGS];
1182 	struct cas_txsoft *txs;
1183 	struct ip *ip;
1184 	struct mbuf *m;
1185 	uint64_t cflags;
1186 	int error, nexttx, nsegs, offset, seg;
1187 
1188 	CAS_LOCK_ASSERT(sc, MA_OWNED);
1189 
1190 	/* Get a work queue entry. */
1191 	if ((txs = STAILQ_FIRST(&sc->sc_txfreeq)) == NULL) {
1192 		/* Ran out of descriptors. */
1193 		return (ENOBUFS);
1194 	}
1195 
1196 	cflags = 0;
1197 	if (((*m_head)->m_pkthdr.csum_flags & CAS_CSUM_FEATURES) != 0) {
1198 		if (M_WRITABLE(*m_head) == 0) {
1199 			m = m_dup(*m_head, M_NOWAIT);
1200 			m_freem(*m_head);
1201 			*m_head = m;
1202 			if (m == NULL)
1203 				return (ENOBUFS);
1204 		}
1205 		offset = sizeof(struct ether_header);
1206 		m = m_pullup(*m_head, offset + sizeof(struct ip));
1207 		if (m == NULL) {
1208 			*m_head = NULL;
1209 			return (ENOBUFS);
1210 		}
1211 		ip = (struct ip *)(mtod(m, caddr_t) + offset);
1212 		offset += (ip->ip_hl << 2);
1213 		cflags = (offset << CAS_TD_CKSUM_START_SHFT) |
1214 		    ((offset + m->m_pkthdr.csum_data) <<
1215 		    CAS_TD_CKSUM_STUFF_SHFT) | CAS_TD_CKSUM_EN;
1216 		*m_head = m;
1217 	}
1218 
1219 	error = bus_dmamap_load_mbuf_sg(sc->sc_tdmatag, txs->txs_dmamap,
1220 	    *m_head, txsegs, &nsegs, BUS_DMA_NOWAIT);
1221 	if (error == EFBIG) {
1222 		m = m_collapse(*m_head, M_NOWAIT, CAS_NTXSEGS);
1223 		if (m == NULL) {
1224 			m_freem(*m_head);
1225 			*m_head = NULL;
1226 			return (ENOBUFS);
1227 		}
1228 		*m_head = m;
1229 		error = bus_dmamap_load_mbuf_sg(sc->sc_tdmatag,
1230 		    txs->txs_dmamap, *m_head, txsegs, &nsegs,
1231 		    BUS_DMA_NOWAIT);
1232 		if (error != 0) {
1233 			m_freem(*m_head);
1234 			*m_head = NULL;
1235 			return (error);
1236 		}
1237 	} else if (error != 0)
1238 		return (error);
1239 	/* If nsegs is wrong then the stack is corrupt. */
1240 	KASSERT(nsegs <= CAS_NTXSEGS,
1241 	    ("%s: too many DMA segments (%d)", __func__, nsegs));
1242 	if (nsegs == 0) {
1243 		m_freem(*m_head);
1244 		*m_head = NULL;
1245 		return (EIO);
1246 	}
1247 
1248 	/*
1249 	 * Ensure we have enough descriptors free to describe
1250 	 * the packet.  Note, we always reserve one descriptor
1251 	 * at the end of the ring as a termination point, in
1252 	 * order to prevent wrap-around.
1253 	 */
1254 	if (nsegs > sc->sc_txfree - 1) {
1255 		txs->txs_ndescs = 0;
1256 		bus_dmamap_unload(sc->sc_tdmatag, txs->txs_dmamap);
1257 		return (ENOBUFS);
1258 	}
1259 
1260 	txs->txs_ndescs = nsegs;
1261 	txs->txs_firstdesc = sc->sc_txnext;
1262 	nexttx = txs->txs_firstdesc;
1263 	for (seg = 0; seg < nsegs; seg++, nexttx = CAS_NEXTTX(nexttx)) {
1264 #ifdef CAS_DEBUG
1265 		CTR6(KTR_CAS,
1266 		    "%s: mapping seg %d (txd %d), len %lx, addr %#lx (%#lx)",
1267 		    __func__, seg, nexttx, txsegs[seg].ds_len,
1268 		    txsegs[seg].ds_addr, htole64(txsegs[seg].ds_addr));
1269 #endif
1270 		sc->sc_txdescs[nexttx].cd_buf_ptr =
1271 		    htole64(txsegs[seg].ds_addr);
1272 		KASSERT(txsegs[seg].ds_len <
1273 		    CAS_TD_BUF_LEN_MASK >> CAS_TD_BUF_LEN_SHFT,
1274 		    ("%s: segment size too large!", __func__));
1275 		sc->sc_txdescs[nexttx].cd_flags =
1276 		    htole64(txsegs[seg].ds_len << CAS_TD_BUF_LEN_SHFT);
1277 		txs->txs_lastdesc = nexttx;
1278 	}
1279 
1280 	/* Set EOF on the last descriptor. */
1281 #ifdef CAS_DEBUG
1282 	CTR3(KTR_CAS, "%s: end of frame at segment %d, TX %d",
1283 	    __func__, seg, nexttx);
1284 #endif
1285 	sc->sc_txdescs[txs->txs_lastdesc].cd_flags |=
1286 	    htole64(CAS_TD_END_OF_FRAME);
1287 
1288 	/* Lastly set SOF on the first descriptor. */
1289 #ifdef CAS_DEBUG
1290 	CTR3(KTR_CAS, "%s: start of frame at segment %d, TX %d",
1291 	    __func__, seg, nexttx);
1292 #endif
1293 	if (sc->sc_txwin += nsegs > CAS_MAXTXFREE * 2 / 3) {
1294 		sc->sc_txwin = 0;
1295 		sc->sc_txdescs[txs->txs_firstdesc].cd_flags |=
1296 		    htole64(cflags | CAS_TD_START_OF_FRAME | CAS_TD_INT_ME);
1297 	} else
1298 		sc->sc_txdescs[txs->txs_firstdesc].cd_flags |=
1299 		    htole64(cflags | CAS_TD_START_OF_FRAME);
1300 
1301 	/* Sync the DMA map. */
1302 	bus_dmamap_sync(sc->sc_tdmatag, txs->txs_dmamap,
1303 	    BUS_DMASYNC_PREWRITE);
1304 
1305 #ifdef CAS_DEBUG
1306 	CTR4(KTR_CAS, "%s: setting firstdesc=%d, lastdesc=%d, ndescs=%d",
1307 	    __func__, txs->txs_firstdesc, txs->txs_lastdesc,
1308 	    txs->txs_ndescs);
1309 #endif
1310 	STAILQ_REMOVE_HEAD(&sc->sc_txfreeq, txs_q);
1311 	STAILQ_INSERT_TAIL(&sc->sc_txdirtyq, txs, txs_q);
1312 	txs->txs_mbuf = *m_head;
1313 
1314 	sc->sc_txnext = CAS_NEXTTX(txs->txs_lastdesc);
1315 	sc->sc_txfree -= txs->txs_ndescs;
1316 
1317 	return (0);
1318 }
1319 
1320 static void
1321 cas_init_regs(struct cas_softc *sc)
1322 {
1323 	int i;
1324 	const u_char *laddr = if_getlladdr(sc->sc_ifp);
1325 
1326 	CAS_LOCK_ASSERT(sc, MA_OWNED);
1327 
1328 	/* These registers are not cleared on reset. */
1329 	if ((sc->sc_flags & CAS_INITED) == 0) {
1330 		/* magic values */
1331 		CAS_WRITE_4(sc, CAS_MAC_IPG0, 0);
1332 		CAS_WRITE_4(sc, CAS_MAC_IPG1, 8);
1333 		CAS_WRITE_4(sc, CAS_MAC_IPG2, 4);
1334 
1335 		/* min frame length */
1336 		CAS_WRITE_4(sc, CAS_MAC_MIN_FRAME, ETHER_MIN_LEN);
1337 		/* max frame length and max burst size */
1338 		CAS_WRITE_4(sc, CAS_MAC_MAX_BF,
1339 		    ((ETHER_MAX_LEN_JUMBO + ETHER_VLAN_ENCAP_LEN) <<
1340 		    CAS_MAC_MAX_BF_FRM_SHFT) |
1341 		    (0x2000 << CAS_MAC_MAX_BF_BST_SHFT));
1342 
1343 		/* more magic values */
1344 		CAS_WRITE_4(sc, CAS_MAC_PREAMBLE_LEN, 0x7);
1345 		CAS_WRITE_4(sc, CAS_MAC_JAM_SIZE, 0x4);
1346 		CAS_WRITE_4(sc, CAS_MAC_ATTEMPT_LIMIT, 0x10);
1347 		CAS_WRITE_4(sc, CAS_MAC_CTRL_TYPE, 0x8808);
1348 
1349 		/* random number seed */
1350 		CAS_WRITE_4(sc, CAS_MAC_RANDOM_SEED,
1351 		    ((laddr[5] << 8) | laddr[4]) & 0x3ff);
1352 
1353 		/* secondary MAC addresses: 0:0:0:0:0:0 */
1354 		for (i = CAS_MAC_ADDR3; i <= CAS_MAC_ADDR41;
1355 		    i += CAS_MAC_ADDR4 - CAS_MAC_ADDR3)
1356 			CAS_WRITE_4(sc, i, 0);
1357 
1358 		/* MAC control address: 01:80:c2:00:00:01 */
1359 		CAS_WRITE_4(sc, CAS_MAC_ADDR42, 0x0001);
1360 		CAS_WRITE_4(sc, CAS_MAC_ADDR43, 0xc200);
1361 		CAS_WRITE_4(sc, CAS_MAC_ADDR44, 0x0180);
1362 
1363 		/* MAC filter address: 0:0:0:0:0:0 */
1364 		CAS_WRITE_4(sc, CAS_MAC_AFILTER0, 0);
1365 		CAS_WRITE_4(sc, CAS_MAC_AFILTER1, 0);
1366 		CAS_WRITE_4(sc, CAS_MAC_AFILTER2, 0);
1367 		CAS_WRITE_4(sc, CAS_MAC_AFILTER_MASK1_2, 0);
1368 		CAS_WRITE_4(sc, CAS_MAC_AFILTER_MASK0, 0);
1369 
1370 		/* Zero the hash table. */
1371 		for (i = CAS_MAC_HASH0; i <= CAS_MAC_HASH15;
1372 		    i += CAS_MAC_HASH1 - CAS_MAC_HASH0)
1373 			CAS_WRITE_4(sc, i, 0);
1374 
1375 		sc->sc_flags |= CAS_INITED;
1376 	}
1377 
1378 	/* Counters need to be zeroed. */
1379 	CAS_WRITE_4(sc, CAS_MAC_NORM_COLL_CNT, 0);
1380 	CAS_WRITE_4(sc, CAS_MAC_FIRST_COLL_CNT, 0);
1381 	CAS_WRITE_4(sc, CAS_MAC_EXCESS_COLL_CNT, 0);
1382 	CAS_WRITE_4(sc, CAS_MAC_LATE_COLL_CNT, 0);
1383 	CAS_WRITE_4(sc, CAS_MAC_DEFER_TMR_CNT, 0);
1384 	CAS_WRITE_4(sc, CAS_MAC_PEAK_ATTEMPTS, 0);
1385 	CAS_WRITE_4(sc, CAS_MAC_RX_FRAME_COUNT, 0);
1386 	CAS_WRITE_4(sc, CAS_MAC_RX_LEN_ERR_CNT, 0);
1387 	CAS_WRITE_4(sc, CAS_MAC_RX_ALIGN_ERR, 0);
1388 	CAS_WRITE_4(sc, CAS_MAC_RX_CRC_ERR_CNT, 0);
1389 	CAS_WRITE_4(sc, CAS_MAC_RX_CODE_VIOL, 0);
1390 
1391 	/* Set XOFF PAUSE time. */
1392 	CAS_WRITE_4(sc, CAS_MAC_SPC, 0x1BF0 << CAS_MAC_SPC_TIME_SHFT);
1393 
1394 	/* Set the station address. */
1395 	CAS_WRITE_4(sc, CAS_MAC_ADDR0, (laddr[4] << 8) | laddr[5]);
1396 	CAS_WRITE_4(sc, CAS_MAC_ADDR1, (laddr[2] << 8) | laddr[3]);
1397 	CAS_WRITE_4(sc, CAS_MAC_ADDR2, (laddr[0] << 8) | laddr[1]);
1398 
1399 	/* Enable MII outputs. */
1400 	CAS_WRITE_4(sc, CAS_MAC_XIF_CONF, CAS_MAC_XIF_CONF_TX_OE);
1401 }
1402 
1403 static void
1404 cas_tx_task(void *arg, int pending __unused)
1405 {
1406 	if_t ifp;
1407 
1408 	ifp = (if_t)arg;
1409 	cas_start(ifp);
1410 }
1411 
1412 static inline void
1413 cas_txkick(struct cas_softc *sc)
1414 {
1415 
1416 	/*
1417 	 * Update the TX kick register.  This register has to point to the
1418 	 * descriptor after the last valid one and for optimum performance
1419 	 * should be incremented in multiples of 4 (the DMA engine fetches/
1420 	 * updates descriptors in batches of 4).
1421 	 */
1422 #ifdef CAS_DEBUG
1423 	CTR3(KTR_CAS, "%s: %s: kicking TX %d",
1424 	    device_get_name(sc->sc_dev), __func__, sc->sc_txnext);
1425 #endif
1426 	CAS_CDSYNC(sc, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1427 	CAS_WRITE_4(sc, CAS_TX_KICK3, sc->sc_txnext);
1428 }
1429 
1430 static void
1431 cas_start(if_t ifp)
1432 {
1433 	struct cas_softc *sc = if_getsoftc(ifp);
1434 	struct mbuf *m;
1435 	int kicked, ntx;
1436 
1437 	CAS_LOCK(sc);
1438 
1439 	if ((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
1440 	    IFF_DRV_RUNNING || (sc->sc_flags & CAS_LINK) == 0) {
1441 		CAS_UNLOCK(sc);
1442 		return;
1443 	}
1444 
1445 	if (sc->sc_txfree < CAS_MAXTXFREE / 4)
1446 		cas_tint(sc);
1447 
1448 #ifdef CAS_DEBUG
1449 	CTR4(KTR_CAS, "%s: %s: txfree %d, txnext %d",
1450 	    device_get_name(sc->sc_dev), __func__, sc->sc_txfree,
1451 	    sc->sc_txnext);
1452 #endif
1453 	ntx = 0;
1454 	kicked = 0;
1455 	for (; !if_sendq_empty(ifp) && sc->sc_txfree > 1;) {
1456 		m = if_dequeue(ifp);
1457 		if (m == NULL)
1458 			break;
1459 		if (cas_load_txmbuf(sc, &m) != 0) {
1460 			if (m == NULL)
1461 				break;
1462 			if_setdrvflagbits(ifp, IFF_DRV_OACTIVE, 0);
1463 			if_sendq_prepend(ifp, m);
1464 			break;
1465 		}
1466 		if ((sc->sc_txnext % 4) == 0) {
1467 			cas_txkick(sc);
1468 			kicked = 1;
1469 		} else
1470 			kicked = 0;
1471 		ntx++;
1472 		BPF_MTAP(ifp, m);
1473 	}
1474 
1475 	if (ntx > 0) {
1476 		if (kicked == 0)
1477 			cas_txkick(sc);
1478 #ifdef CAS_DEBUG
1479 		CTR2(KTR_CAS, "%s: packets enqueued, OWN on %d",
1480 		    device_get_name(sc->sc_dev), sc->sc_txnext);
1481 #endif
1482 
1483 		/* Set a watchdog timer in case the chip flakes out. */
1484 		sc->sc_wdog_timer = 5;
1485 #ifdef CAS_DEBUG
1486 		CTR3(KTR_CAS, "%s: %s: watchdog %d",
1487 		    device_get_name(sc->sc_dev), __func__,
1488 		    sc->sc_wdog_timer);
1489 #endif
1490 	}
1491 
1492 	CAS_UNLOCK(sc);
1493 }
1494 
1495 static void
1496 cas_tint(struct cas_softc *sc)
1497 {
1498 	if_t ifp = sc->sc_ifp;
1499 	struct cas_txsoft *txs;
1500 	int progress;
1501 	uint32_t txlast;
1502 #ifdef CAS_DEBUG
1503 	int i;
1504 
1505 	CAS_LOCK_ASSERT(sc, MA_OWNED);
1506 
1507 	CTR2(KTR_CAS, "%s: %s", device_get_name(sc->sc_dev), __func__);
1508 #endif
1509 
1510 	/*
1511 	 * Go through our TX list and free mbufs for those
1512 	 * frames that have been transmitted.
1513 	 */
1514 	progress = 0;
1515 	CAS_CDSYNC(sc, BUS_DMASYNC_POSTREAD);
1516 	while ((txs = STAILQ_FIRST(&sc->sc_txdirtyq)) != NULL) {
1517 #ifdef CAS_DEBUG
1518 		if ((if_getflags(ifp) & IFF_DEBUG) != 0) {
1519 			printf("    txsoft %p transmit chain:\n", txs);
1520 			for (i = txs->txs_firstdesc;; i = CAS_NEXTTX(i)) {
1521 				printf("descriptor %d: ", i);
1522 				printf("cd_flags: 0x%016llx\t",
1523 				    (long long)le64toh(
1524 				    sc->sc_txdescs[i].cd_flags));
1525 				printf("cd_buf_ptr: 0x%016llx\n",
1526 				    (long long)le64toh(
1527 				    sc->sc_txdescs[i].cd_buf_ptr));
1528 				if (i == txs->txs_lastdesc)
1529 					break;
1530 			}
1531 		}
1532 #endif
1533 
1534 		/*
1535 		 * In theory, we could harvest some descriptors before
1536 		 * the ring is empty, but that's a bit complicated.
1537 		 *
1538 		 * CAS_TX_COMPn points to the last descriptor
1539 		 * processed + 1.
1540 		 */
1541 		txlast = CAS_READ_4(sc, CAS_TX_COMP3);
1542 #ifdef CAS_DEBUG
1543 		CTR4(KTR_CAS, "%s: txs->txs_firstdesc = %d, "
1544 		    "txs->txs_lastdesc = %d, txlast = %d",
1545 		    __func__, txs->txs_firstdesc, txs->txs_lastdesc, txlast);
1546 #endif
1547 		if (txs->txs_firstdesc <= txs->txs_lastdesc) {
1548 			if ((txlast >= txs->txs_firstdesc) &&
1549 			    (txlast <= txs->txs_lastdesc))
1550 				break;
1551 		} else {
1552 			/* Ick -- this command wraps. */
1553 			if ((txlast >= txs->txs_firstdesc) ||
1554 			    (txlast <= txs->txs_lastdesc))
1555 				break;
1556 		}
1557 
1558 #ifdef CAS_DEBUG
1559 		CTR1(KTR_CAS, "%s: releasing a descriptor", __func__);
1560 #endif
1561 		STAILQ_REMOVE_HEAD(&sc->sc_txdirtyq, txs_q);
1562 
1563 		sc->sc_txfree += txs->txs_ndescs;
1564 
1565 		bus_dmamap_sync(sc->sc_tdmatag, txs->txs_dmamap,
1566 		    BUS_DMASYNC_POSTWRITE);
1567 		bus_dmamap_unload(sc->sc_tdmatag, txs->txs_dmamap);
1568 		if (txs->txs_mbuf != NULL) {
1569 			m_freem(txs->txs_mbuf);
1570 			txs->txs_mbuf = NULL;
1571 		}
1572 
1573 		STAILQ_INSERT_TAIL(&sc->sc_txfreeq, txs, txs_q);
1574 
1575 		if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
1576 		progress = 1;
1577 	}
1578 
1579 #ifdef CAS_DEBUG
1580 	CTR5(KTR_CAS, "%s: CAS_TX_SM1 %x CAS_TX_SM2 %x CAS_TX_DESC_BASE %llx "
1581 	    "CAS_TX_COMP3 %x",
1582 	    __func__, CAS_READ_4(sc, CAS_TX_SM1), CAS_READ_4(sc, CAS_TX_SM2),
1583 	    ((long long)CAS_READ_4(sc, CAS_TX_DESC3_BASE_HI) << 32) |
1584 	    CAS_READ_4(sc, CAS_TX_DESC3_BASE_LO),
1585 	    CAS_READ_4(sc, CAS_TX_COMP3));
1586 #endif
1587 
1588 	if (progress) {
1589 		/* We freed some descriptors, so reset IFF_DRV_OACTIVE. */
1590 		if_setdrvflagbits(ifp, 0, IFF_DRV_OACTIVE);
1591 		if (STAILQ_EMPTY(&sc->sc_txdirtyq))
1592 			sc->sc_wdog_timer = 0;
1593 	}
1594 
1595 #ifdef CAS_DEBUG
1596 	CTR3(KTR_CAS, "%s: %s: watchdog %d",
1597 	    device_get_name(sc->sc_dev), __func__, sc->sc_wdog_timer);
1598 #endif
1599 }
1600 
1601 static void
1602 cas_rint_timeout(void *arg)
1603 {
1604 	struct epoch_tracker et;
1605 	struct cas_softc *sc = arg;
1606 
1607 	CAS_LOCK_ASSERT(sc, MA_OWNED);
1608 
1609 	NET_EPOCH_ENTER(et);
1610 	cas_rint(sc);
1611 	NET_EPOCH_EXIT(et);
1612 }
1613 
1614 static void
1615 cas_rint(struct cas_softc *sc)
1616 {
1617 	struct cas_rxdsoft *rxds, *rxds2;
1618 	if_t ifp = sc->sc_ifp;
1619 	struct mbuf *m, *m2;
1620 	uint64_t word1, word2, word3 __unused, word4;
1621 	uint32_t rxhead;
1622 	u_int idx, idx2, len, off, skip;
1623 
1624 	CAS_LOCK_ASSERT(sc, MA_OWNED);
1625 
1626 	callout_stop(&sc->sc_rx_ch);
1627 
1628 #ifdef CAS_DEBUG
1629 	CTR2(KTR_CAS, "%s: %s", device_get_name(sc->sc_dev), __func__);
1630 #endif
1631 
1632 #define	PRINTWORD(n, delimiter)						\
1633 	printf("word ## n: 0x%016llx%c", (long long)word ## n, delimiter)
1634 
1635 #define	SKIPASSERT(n)							\
1636 	KASSERT(sc->sc_rxcomps[sc->sc_rxcptr].crc_word ## n == 0,	\
1637 	    ("%s: word ## n not 0", __func__))
1638 
1639 #define	WORDTOH(n)							\
1640 	word ## n = le64toh(sc->sc_rxcomps[sc->sc_rxcptr].crc_word ## n)
1641 
1642 	/*
1643 	 * Read the completion head register once.  This limits
1644 	 * how long the following loop can execute.
1645 	 */
1646 	rxhead = CAS_READ_4(sc, CAS_RX_COMP_HEAD);
1647 #ifdef CAS_DEBUG
1648 	CTR4(KTR_CAS, "%s: sc->sc_rxcptr %d, sc->sc_rxdptr %d, head %d",
1649 	    __func__, sc->sc_rxcptr, sc->sc_rxdptr, rxhead);
1650 #endif
1651 	skip = 0;
1652 	CAS_CDSYNC(sc, BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
1653 	for (; sc->sc_rxcptr != rxhead;
1654 	    sc->sc_rxcptr = CAS_NEXTRXCOMP(sc->sc_rxcptr)) {
1655 		if (skip != 0) {
1656 			SKIPASSERT(1);
1657 			SKIPASSERT(2);
1658 			SKIPASSERT(3);
1659 
1660 			--skip;
1661 			goto skip;
1662 		}
1663 
1664 		WORDTOH(1);
1665 		WORDTOH(2);
1666 		WORDTOH(3);
1667 		WORDTOH(4);
1668 
1669 #ifdef CAS_DEBUG
1670 		if ((if_getflags(ifp) & IFF_DEBUG) != 0) {
1671 			printf("    completion %d: ", sc->sc_rxcptr);
1672 			PRINTWORD(1, '\t');
1673 			PRINTWORD(2, '\t');
1674 			PRINTWORD(3, '\t');
1675 			PRINTWORD(4, '\n');
1676 		}
1677 #endif
1678 
1679 		if (__predict_false(
1680 		    (word1 & CAS_RC1_TYPE_MASK) == CAS_RC1_TYPE_HW ||
1681 		    (word4 & CAS_RC4_ZERO) != 0)) {
1682 			/*
1683 			 * The descriptor is still marked as owned, although
1684 			 * it is supposed to have completed.  This has been
1685 			 * observed on some machines.  Just exiting here
1686 			 * might leave the packet sitting around until another
1687 			 * one arrives to trigger a new interrupt, which is
1688 			 * generally undesirable, so set up a timeout.
1689 			 */
1690 			callout_reset(&sc->sc_rx_ch, CAS_RXOWN_TICKS,
1691 			    cas_rint_timeout, sc);
1692 			break;
1693 		}
1694 
1695 		if (__predict_false(
1696 		    (word4 & (CAS_RC4_BAD | CAS_RC4_LEN_MMATCH)) != 0)) {
1697 			if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
1698 			device_printf(sc->sc_dev,
1699 			    "receive error: CRC error\n");
1700 			continue;
1701 		}
1702 
1703 		KASSERT(CAS_GET(word1, CAS_RC1_DATA_SIZE) == 0 ||
1704 		    CAS_GET(word2, CAS_RC2_HDR_SIZE) == 0,
1705 		    ("%s: data and header present", __func__));
1706 		KASSERT((word1 & CAS_RC1_SPLIT_PKT) == 0 ||
1707 		    CAS_GET(word2, CAS_RC2_HDR_SIZE) == 0,
1708 		    ("%s: split and header present", __func__));
1709 		KASSERT(CAS_GET(word1, CAS_RC1_DATA_SIZE) == 0 ||
1710 		    (word1 & CAS_RC1_RELEASE_HDR) == 0,
1711 		    ("%s: data present but header release", __func__));
1712 		KASSERT(CAS_GET(word2, CAS_RC2_HDR_SIZE) == 0 ||
1713 		    (word1 & CAS_RC1_RELEASE_DATA) == 0,
1714 		    ("%s: header present but data release", __func__));
1715 
1716 		if ((len = CAS_GET(word2, CAS_RC2_HDR_SIZE)) != 0) {
1717 			idx = CAS_GET(word2, CAS_RC2_HDR_INDEX);
1718 			off = CAS_GET(word2, CAS_RC2_HDR_OFF);
1719 #ifdef CAS_DEBUG
1720 			CTR4(KTR_CAS, "%s: hdr at idx %d, off %d, len %d",
1721 			    __func__, idx, off, len);
1722 #endif
1723 			rxds = &sc->sc_rxdsoft[idx];
1724 			MGETHDR(m, M_NOWAIT, MT_DATA);
1725 			if (m != NULL) {
1726 				refcount_acquire(&rxds->rxds_refcount);
1727 				bus_dmamap_sync(sc->sc_rdmatag,
1728 				    rxds->rxds_dmamap, BUS_DMASYNC_POSTREAD);
1729 				m_extadd(m, (char *)rxds->rxds_buf +
1730 				    off * 256 + ETHER_ALIGN, len, cas_free,
1731 				    sc, (void *)(uintptr_t)idx,
1732 				    M_RDONLY, EXT_NET_DRV);
1733 				if ((m->m_flags & M_EXT) == 0) {
1734 					m_freem(m);
1735 					m = NULL;
1736 				}
1737 			}
1738 			if (m != NULL) {
1739 				m->m_pkthdr.rcvif = ifp;
1740 				m->m_pkthdr.len = m->m_len = len;
1741 				if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
1742 				if ((if_getcapenable(ifp) & IFCAP_RXCSUM) != 0)
1743 					cas_rxcksum(m, CAS_GET(word4,
1744 					    CAS_RC4_TCP_CSUM));
1745 				/* Pass it on. */
1746 				CAS_UNLOCK(sc);
1747 				if_input(ifp, m);
1748 				CAS_LOCK(sc);
1749 			} else
1750 				if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1751 
1752 			if ((word1 & CAS_RC1_RELEASE_HDR) != 0 &&
1753 			    refcount_release(&rxds->rxds_refcount) != 0)
1754 				cas_add_rxdesc(sc, idx);
1755 		} else if ((len = CAS_GET(word1, CAS_RC1_DATA_SIZE)) != 0) {
1756 			idx = CAS_GET(word1, CAS_RC1_DATA_INDEX);
1757 			off = CAS_GET(word1, CAS_RC1_DATA_OFF);
1758 #ifdef CAS_DEBUG
1759 			CTR4(KTR_CAS, "%s: data at idx %d, off %d, len %d",
1760 			    __func__, idx, off, len);
1761 #endif
1762 			rxds = &sc->sc_rxdsoft[idx];
1763 			MGETHDR(m, M_NOWAIT, MT_DATA);
1764 			if (m != NULL) {
1765 				refcount_acquire(&rxds->rxds_refcount);
1766 				off += ETHER_ALIGN;
1767 				m->m_len = min(CAS_PAGE_SIZE - off, len);
1768 				bus_dmamap_sync(sc->sc_rdmatag,
1769 				    rxds->rxds_dmamap, BUS_DMASYNC_POSTREAD);
1770 				m_extadd(m, (char *)rxds->rxds_buf + off,
1771 				    m->m_len, cas_free, sc,
1772 				    (void *)(uintptr_t)idx, M_RDONLY,
1773 				    EXT_NET_DRV);
1774 				if ((m->m_flags & M_EXT) == 0) {
1775 					m_freem(m);
1776 					m = NULL;
1777 				}
1778 			}
1779 			idx2 = 0;
1780 			m2 = NULL;
1781 			rxds2 = NULL;
1782 			if ((word1 & CAS_RC1_SPLIT_PKT) != 0) {
1783 				KASSERT((word1 & CAS_RC1_RELEASE_NEXT) != 0,
1784 				    ("%s: split but no release next",
1785 				    __func__));
1786 
1787 				idx2 = CAS_GET(word2, CAS_RC2_NEXT_INDEX);
1788 #ifdef CAS_DEBUG
1789 				CTR2(KTR_CAS, "%s: split at idx %d",
1790 				    __func__, idx2);
1791 #endif
1792 				rxds2 = &sc->sc_rxdsoft[idx2];
1793 				if (m != NULL) {
1794 					MGET(m2, M_NOWAIT, MT_DATA);
1795 					if (m2 != NULL) {
1796 						refcount_acquire(
1797 						    &rxds2->rxds_refcount);
1798 						m2->m_len = len - m->m_len;
1799 						bus_dmamap_sync(
1800 						    sc->sc_rdmatag,
1801 						    rxds2->rxds_dmamap,
1802 						    BUS_DMASYNC_POSTREAD);
1803 						m_extadd(m2,
1804 						    (char *)rxds2->rxds_buf,
1805 						    m2->m_len, cas_free, sc,
1806 						    (void *)(uintptr_t)idx2,
1807 						    M_RDONLY, EXT_NET_DRV);
1808 						if ((m2->m_flags & M_EXT) ==
1809 						    0) {
1810 							m_freem(m2);
1811 							m2 = NULL;
1812 						}
1813 					}
1814 				}
1815 				if (m2 != NULL)
1816 					m->m_next = m2;
1817 				else if (m != NULL) {
1818 					m_freem(m);
1819 					m = NULL;
1820 				}
1821 			}
1822 			if (m != NULL) {
1823 				m->m_pkthdr.rcvif = ifp;
1824 				m->m_pkthdr.len = len;
1825 				if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
1826 				if ((if_getcapenable(ifp) & IFCAP_RXCSUM) != 0)
1827 					cas_rxcksum(m, CAS_GET(word4,
1828 					    CAS_RC4_TCP_CSUM));
1829 				/* Pass it on. */
1830 				CAS_UNLOCK(sc);
1831 				if_input(ifp, m);
1832 				CAS_LOCK(sc);
1833 			} else
1834 				if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1835 
1836 			if ((word1 & CAS_RC1_RELEASE_DATA) != 0 &&
1837 			    refcount_release(&rxds->rxds_refcount) != 0)
1838 				cas_add_rxdesc(sc, idx);
1839 			if ((word1 & CAS_RC1_SPLIT_PKT) != 0 &&
1840 			    refcount_release(&rxds2->rxds_refcount) != 0)
1841 				cas_add_rxdesc(sc, idx2);
1842 		}
1843 
1844 		skip = CAS_GET(word1, CAS_RC1_SKIP);
1845 
1846  skip:
1847 		cas_rxcompinit(&sc->sc_rxcomps[sc->sc_rxcptr]);
1848 		if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0)
1849 			break;
1850 	}
1851 	CAS_CDSYNC(sc, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1852 	CAS_WRITE_4(sc, CAS_RX_COMP_TAIL, sc->sc_rxcptr);
1853 
1854 #undef PRINTWORD
1855 #undef SKIPASSERT
1856 #undef WORDTOH
1857 
1858 #ifdef CAS_DEBUG
1859 	CTR4(KTR_CAS, "%s: done sc->sc_rxcptr %d, sc->sc_rxdptr %d, head %d",
1860 	    __func__, sc->sc_rxcptr, sc->sc_rxdptr,
1861 	    CAS_READ_4(sc, CAS_RX_COMP_HEAD));
1862 #endif
1863 }
1864 
1865 static void
1866 cas_free(struct mbuf *m)
1867 {
1868 	struct cas_rxdsoft *rxds;
1869 	struct cas_softc *sc;
1870 	u_int idx, locked;
1871 
1872 	sc = m->m_ext.ext_arg1;
1873 	idx = (uintptr_t)m->m_ext.ext_arg2;
1874 	rxds = &sc->sc_rxdsoft[idx];
1875 	if (refcount_release(&rxds->rxds_refcount) == 0)
1876 		return;
1877 
1878 	/*
1879 	 * NB: this function can be called via m_freem(9) within
1880 	 * this driver!
1881 	 */
1882 	if ((locked = CAS_LOCK_OWNED(sc)) == 0)
1883 		CAS_LOCK(sc);
1884 	cas_add_rxdesc(sc, idx);
1885 	if (locked == 0)
1886 		CAS_UNLOCK(sc);
1887 }
1888 
1889 static inline void
1890 cas_add_rxdesc(struct cas_softc *sc, u_int idx)
1891 {
1892 
1893 	CAS_LOCK_ASSERT(sc, MA_OWNED);
1894 
1895 	bus_dmamap_sync(sc->sc_rdmatag, sc->sc_rxdsoft[idx].rxds_dmamap,
1896 	    BUS_DMASYNC_PREREAD);
1897 	CAS_UPDATE_RXDESC(sc, sc->sc_rxdptr, idx);
1898 	sc->sc_rxdptr = CAS_NEXTRXDESC(sc->sc_rxdptr);
1899 
1900 	/*
1901 	 * Update the RX kick register.  This register has to point to the
1902 	 * descriptor after the last valid one (before the current batch)
1903 	 * and for optimum performance should be incremented in multiples
1904 	 * of 4 (the DMA engine fetches/updates descriptors in batches of 4).
1905 	 */
1906 	if ((sc->sc_rxdptr % 4) == 0) {
1907 		CAS_CDSYNC(sc, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1908 		CAS_WRITE_4(sc, CAS_RX_KICK,
1909 		    (sc->sc_rxdptr + CAS_NRXDESC - 4) & CAS_NRXDESC_MASK);
1910 	}
1911 }
1912 
1913 static void
1914 cas_eint(struct cas_softc *sc, u_int status)
1915 {
1916 	if_t ifp = sc->sc_ifp;
1917 
1918 	CAS_LOCK_ASSERT(sc, MA_OWNED);
1919 
1920 	if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
1921 
1922 	device_printf(sc->sc_dev, "%s: status 0x%x", __func__, status);
1923 	if ((status & CAS_INTR_PCI_ERROR_INT) != 0) {
1924 		status = CAS_READ_4(sc, CAS_ERROR_STATUS);
1925 		printf(", PCI bus error 0x%x", status);
1926 		if ((status & CAS_ERROR_OTHER) != 0) {
1927 			status = pci_read_config(sc->sc_dev, PCIR_STATUS, 2);
1928 			printf(", PCI status 0x%x", status);
1929 			pci_write_config(sc->sc_dev, PCIR_STATUS, status, 2);
1930 		}
1931 	}
1932 	printf("\n");
1933 
1934 	if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
1935 	cas_init_locked(sc);
1936 	if (!if_sendq_empty(ifp))
1937 		taskqueue_enqueue(sc->sc_tq, &sc->sc_tx_task);
1938 }
1939 
1940 static int
1941 cas_intr(void *v)
1942 {
1943 	struct cas_softc *sc = v;
1944 
1945 	if (__predict_false((CAS_READ_4(sc, CAS_STATUS_ALIAS) &
1946 	    CAS_INTR_SUMMARY) == 0))
1947 		return (FILTER_STRAY);
1948 
1949 	/* Disable interrupts. */
1950 	CAS_WRITE_4(sc, CAS_INTMASK, 0xffffffff);
1951 	taskqueue_enqueue(sc->sc_tq, &sc->sc_intr_task);
1952 
1953 	return (FILTER_HANDLED);
1954 }
1955 
1956 static void
1957 cas_intr_task(void *arg, int pending __unused)
1958 {
1959 	struct cas_softc *sc = arg;
1960 	if_t ifp = sc->sc_ifp;
1961 	uint32_t status, status2;
1962 
1963 	CAS_LOCK_ASSERT(sc, MA_NOTOWNED);
1964 
1965 	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0)
1966 		return;
1967 
1968 	status = CAS_READ_4(sc, CAS_STATUS);
1969 	if (__predict_false((status & CAS_INTR_SUMMARY) == 0))
1970 		goto done;
1971 
1972 	CAS_LOCK(sc);
1973 #ifdef CAS_DEBUG
1974 	CTR4(KTR_CAS, "%s: %s: cplt %x, status %x",
1975 	    device_get_name(sc->sc_dev), __func__,
1976 	    (status >> CAS_STATUS_TX_COMP3_SHFT), (u_int)status);
1977 
1978 	/*
1979 	 * PCS interrupts must be cleared, otherwise no traffic is passed!
1980 	 */
1981 	if ((status & CAS_INTR_PCS_INT) != 0) {
1982 		status2 =
1983 		    CAS_READ_4(sc, CAS_PCS_INTR_STATUS) |
1984 		    CAS_READ_4(sc, CAS_PCS_INTR_STATUS);
1985 		if ((status2 & CAS_PCS_INTR_LINK) != 0)
1986 			device_printf(sc->sc_dev,
1987 			    "%s: PCS link status changed\n", __func__);
1988 	}
1989 	if ((status & CAS_MAC_CTRL_STATUS) != 0) {
1990 		status2 = CAS_READ_4(sc, CAS_MAC_CTRL_STATUS);
1991 		if ((status2 & CAS_MAC_CTRL_PAUSE) != 0)
1992 			device_printf(sc->sc_dev,
1993 			    "%s: PAUSE received (PAUSE time %d slots)\n",
1994 			    __func__,
1995 			    (status2 & CAS_MAC_CTRL_STATUS_PT_MASK) >>
1996 			    CAS_MAC_CTRL_STATUS_PT_SHFT);
1997 		if ((status2 & CAS_MAC_CTRL_PAUSE) != 0)
1998 			device_printf(sc->sc_dev,
1999 			    "%s: transited to PAUSE state\n", __func__);
2000 		if ((status2 & CAS_MAC_CTRL_NON_PAUSE) != 0)
2001 			device_printf(sc->sc_dev,
2002 			    "%s: transited to non-PAUSE state\n", __func__);
2003 	}
2004 	if ((status & CAS_INTR_MIF) != 0)
2005 		device_printf(sc->sc_dev, "%s: MIF interrupt\n", __func__);
2006 #endif
2007 
2008 	if (__predict_false((status &
2009 	    (CAS_INTR_TX_TAG_ERR | CAS_INTR_RX_TAG_ERR |
2010 	    CAS_INTR_RX_LEN_MMATCH | CAS_INTR_PCI_ERROR_INT)) != 0)) {
2011 		cas_eint(sc, status);
2012 		CAS_UNLOCK(sc);
2013 		return;
2014 	}
2015 
2016 	if (__predict_false(status & CAS_INTR_TX_MAC_INT)) {
2017 		status2 = CAS_READ_4(sc, CAS_MAC_TX_STATUS);
2018 		if ((status2 &
2019 		    (CAS_MAC_TX_UNDERRUN | CAS_MAC_TX_MAX_PKT_ERR)) != 0)
2020 			if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
2021 		else if ((status2 & ~CAS_MAC_TX_FRAME_XMTD) != 0)
2022 			device_printf(sc->sc_dev,
2023 			    "MAC TX fault, status %x\n", status2);
2024 	}
2025 
2026 	if (__predict_false(status & CAS_INTR_RX_MAC_INT)) {
2027 		status2 = CAS_READ_4(sc, CAS_MAC_RX_STATUS);
2028 		if ((status2 & CAS_MAC_RX_OVERFLOW) != 0)
2029 			if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
2030 		else if ((status2 & ~CAS_MAC_RX_FRAME_RCVD) != 0)
2031 			device_printf(sc->sc_dev,
2032 			    "MAC RX fault, status %x\n", status2);
2033 	}
2034 
2035 	if ((status &
2036 	    (CAS_INTR_RX_DONE | CAS_INTR_RX_BUF_NA | CAS_INTR_RX_COMP_FULL |
2037 	    CAS_INTR_RX_BUF_AEMPTY | CAS_INTR_RX_COMP_AFULL)) != 0) {
2038 		cas_rint(sc);
2039 #ifdef CAS_DEBUG
2040 		if (__predict_false((status &
2041 		    (CAS_INTR_RX_BUF_NA | CAS_INTR_RX_COMP_FULL |
2042 		    CAS_INTR_RX_BUF_AEMPTY | CAS_INTR_RX_COMP_AFULL)) != 0))
2043 			device_printf(sc->sc_dev,
2044 			    "RX fault, status %x\n", status);
2045 #endif
2046 	}
2047 
2048 	if ((status &
2049 	    (CAS_INTR_TX_INT_ME | CAS_INTR_TX_ALL | CAS_INTR_TX_DONE)) != 0)
2050 		cas_tint(sc);
2051 
2052 	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0) {
2053 		CAS_UNLOCK(sc);
2054 		return;
2055 	} else if (!if_sendq_empty(ifp))
2056 		taskqueue_enqueue(sc->sc_tq, &sc->sc_tx_task);
2057 	CAS_UNLOCK(sc);
2058 
2059 	status = CAS_READ_4(sc, CAS_STATUS_ALIAS);
2060 	if (__predict_false((status & CAS_INTR_SUMMARY) != 0)) {
2061 		taskqueue_enqueue(sc->sc_tq, &sc->sc_intr_task);
2062 		return;
2063 	}
2064 
2065  done:
2066 	/* Re-enable interrupts. */
2067 	CAS_WRITE_4(sc, CAS_INTMASK,
2068 	    ~(CAS_INTR_TX_INT_ME | CAS_INTR_TX_TAG_ERR |
2069 	    CAS_INTR_RX_DONE | CAS_INTR_RX_BUF_NA | CAS_INTR_RX_TAG_ERR |
2070 	    CAS_INTR_RX_COMP_FULL | CAS_INTR_RX_BUF_AEMPTY |
2071 	    CAS_INTR_RX_COMP_AFULL | CAS_INTR_RX_LEN_MMATCH |
2072 	    CAS_INTR_PCI_ERROR_INT
2073 #ifdef CAS_DEBUG
2074 	    | CAS_INTR_PCS_INT | CAS_INTR_MIF
2075 #endif
2076 	));
2077 }
2078 
2079 static void
2080 cas_watchdog(struct cas_softc *sc)
2081 {
2082 	if_t ifp = sc->sc_ifp;
2083 
2084 	CAS_LOCK_ASSERT(sc, MA_OWNED);
2085 
2086 #ifdef CAS_DEBUG
2087 	CTR4(KTR_CAS,
2088 	    "%s: CAS_RX_CONF %x CAS_MAC_RX_STATUS %x CAS_MAC_RX_CONF %x",
2089 	    __func__, CAS_READ_4(sc, CAS_RX_CONF),
2090 	    CAS_READ_4(sc, CAS_MAC_RX_STATUS),
2091 	    CAS_READ_4(sc, CAS_MAC_RX_CONF));
2092 	CTR4(KTR_CAS,
2093 	    "%s: CAS_TX_CONF %x CAS_MAC_TX_STATUS %x CAS_MAC_TX_CONF %x",
2094 	    __func__, CAS_READ_4(sc, CAS_TX_CONF),
2095 	    CAS_READ_4(sc, CAS_MAC_TX_STATUS),
2096 	    CAS_READ_4(sc, CAS_MAC_TX_CONF));
2097 #endif
2098 
2099 	if (sc->sc_wdog_timer == 0 || --sc->sc_wdog_timer != 0)
2100 		return;
2101 
2102 	if ((sc->sc_flags & CAS_LINK) != 0)
2103 		device_printf(sc->sc_dev, "device timeout\n");
2104 	else if (bootverbose)
2105 		device_printf(sc->sc_dev, "device timeout (no link)\n");
2106 	if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
2107 
2108 	/* Try to get more packets going. */
2109 	if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
2110 	cas_init_locked(sc);
2111 	if (!if_sendq_empty(ifp))
2112 		taskqueue_enqueue(sc->sc_tq, &sc->sc_tx_task);
2113 }
2114 
2115 static void
2116 cas_mifinit(struct cas_softc *sc)
2117 {
2118 
2119 	/* Configure the MIF in frame mode. */
2120 	CAS_WRITE_4(sc, CAS_MIF_CONF,
2121 	    CAS_READ_4(sc, CAS_MIF_CONF) & ~CAS_MIF_CONF_BB_MODE);
2122 	CAS_BARRIER(sc, CAS_MIF_CONF, 4,
2123 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2124 }
2125 
2126 /*
2127  * MII interface
2128  *
2129  * The MII interface supports at least three different operating modes:
2130  *
2131  * Bitbang mode is implemented using data, clock and output enable registers.
2132  *
2133  * Frame mode is implemented by loading a complete frame into the frame
2134  * register and polling the valid bit for completion.
2135  *
2136  * Polling mode uses the frame register but completion is indicated by
2137  * an interrupt.
2138  *
2139  */
2140 static int
2141 cas_mii_readreg(device_t dev, int phy, int reg)
2142 {
2143 	struct cas_softc *sc;
2144 	int n;
2145 	uint32_t v;
2146 
2147 #ifdef CAS_DEBUG_PHY
2148 	printf("%s: phy %d reg %d\n", __func__, phy, reg);
2149 #endif
2150 
2151 	sc = device_get_softc(dev);
2152 	if ((sc->sc_flags & CAS_SERDES) != 0) {
2153 		switch (reg) {
2154 		case MII_BMCR:
2155 			reg = CAS_PCS_CTRL;
2156 			break;
2157 		case MII_BMSR:
2158 			reg = CAS_PCS_STATUS;
2159 			break;
2160 		case MII_PHYIDR1:
2161 		case MII_PHYIDR2:
2162 			return (0);
2163 		case MII_ANAR:
2164 			reg = CAS_PCS_ANAR;
2165 			break;
2166 		case MII_ANLPAR:
2167 			reg = CAS_PCS_ANLPAR;
2168 			break;
2169 		case MII_EXTSR:
2170 			return (EXTSR_1000XFDX | EXTSR_1000XHDX);
2171 		default:
2172 			device_printf(sc->sc_dev,
2173 			    "%s: unhandled register %d\n", __func__, reg);
2174 			return (0);
2175 		}
2176 		return (CAS_READ_4(sc, reg));
2177 	}
2178 
2179 	/* Construct the frame command. */
2180 	v = CAS_MIF_FRAME_READ |
2181 	    (phy << CAS_MIF_FRAME_PHY_SHFT) |
2182 	    (reg << CAS_MIF_FRAME_REG_SHFT);
2183 
2184 	CAS_WRITE_4(sc, CAS_MIF_FRAME, v);
2185 	CAS_BARRIER(sc, CAS_MIF_FRAME, 4,
2186 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2187 	for (n = 0; n < 100; n++) {
2188 		DELAY(1);
2189 		v = CAS_READ_4(sc, CAS_MIF_FRAME);
2190 		if (v & CAS_MIF_FRAME_TA_LSB)
2191 			return (v & CAS_MIF_FRAME_DATA);
2192 	}
2193 
2194 	device_printf(sc->sc_dev, "%s: timed out\n", __func__);
2195 	return (0);
2196 }
2197 
2198 static int
2199 cas_mii_writereg(device_t dev, int phy, int reg, int val)
2200 {
2201 	struct cas_softc *sc;
2202 	int n;
2203 	uint32_t v;
2204 
2205 #ifdef CAS_DEBUG_PHY
2206 	printf("%s: phy %d reg %d val %x\n", phy, reg, val, __func__);
2207 #endif
2208 
2209 	sc = device_get_softc(dev);
2210 	if ((sc->sc_flags & CAS_SERDES) != 0) {
2211 		switch (reg) {
2212 		case MII_BMSR:
2213 			reg = CAS_PCS_STATUS;
2214 			break;
2215 		case MII_BMCR:
2216 			reg = CAS_PCS_CTRL;
2217 			if ((val & CAS_PCS_CTRL_RESET) == 0)
2218 				break;
2219 			CAS_WRITE_4(sc, CAS_PCS_CTRL, val);
2220 			CAS_BARRIER(sc, CAS_PCS_CTRL, 4,
2221 			    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2222 			if (!cas_bitwait(sc, CAS_PCS_CTRL,
2223 			    CAS_PCS_CTRL_RESET, 0))
2224 				device_printf(sc->sc_dev,
2225 				    "cannot reset PCS\n");
2226 			/* FALLTHROUGH */
2227 		case MII_ANAR:
2228 			CAS_WRITE_4(sc, CAS_PCS_CONF, 0);
2229 			CAS_BARRIER(sc, CAS_PCS_CONF, 4,
2230 			    BUS_SPACE_BARRIER_WRITE);
2231 			CAS_WRITE_4(sc, CAS_PCS_ANAR, val);
2232 			CAS_BARRIER(sc, CAS_PCS_ANAR, 4,
2233 			    BUS_SPACE_BARRIER_WRITE);
2234 			CAS_WRITE_4(sc, CAS_PCS_SERDES_CTRL,
2235 			    CAS_PCS_SERDES_CTRL_ESD);
2236 			CAS_BARRIER(sc, CAS_PCS_CONF, 4,
2237 			    BUS_SPACE_BARRIER_WRITE);
2238 			CAS_WRITE_4(sc, CAS_PCS_CONF,
2239 			    CAS_PCS_CONF_EN);
2240 			CAS_BARRIER(sc, CAS_PCS_CONF, 4,
2241 			    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2242 			return (0);
2243 		case MII_ANLPAR:
2244 			reg = CAS_PCS_ANLPAR;
2245 			break;
2246 		default:
2247 			device_printf(sc->sc_dev,
2248 			    "%s: unhandled register %d\n", __func__, reg);
2249 			return (0);
2250 		}
2251 		CAS_WRITE_4(sc, reg, val);
2252 		CAS_BARRIER(sc, reg, 4,
2253 		    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2254 		return (0);
2255 	}
2256 
2257 	/* Construct the frame command. */
2258 	v = CAS_MIF_FRAME_WRITE |
2259 	    (phy << CAS_MIF_FRAME_PHY_SHFT) |
2260 	    (reg << CAS_MIF_FRAME_REG_SHFT) |
2261 	    (val & CAS_MIF_FRAME_DATA);
2262 
2263 	CAS_WRITE_4(sc, CAS_MIF_FRAME, v);
2264 	CAS_BARRIER(sc, CAS_MIF_FRAME, 4,
2265 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2266 	for (n = 0; n < 100; n++) {
2267 		DELAY(1);
2268 		v = CAS_READ_4(sc, CAS_MIF_FRAME);
2269 		if (v & CAS_MIF_FRAME_TA_LSB)
2270 			return (1);
2271 	}
2272 
2273 	device_printf(sc->sc_dev, "%s: timed out\n", __func__);
2274 	return (0);
2275 }
2276 
2277 static void
2278 cas_mii_statchg(device_t dev)
2279 {
2280 	struct cas_softc *sc;
2281 	if_t ifp;
2282 	int gigabit;
2283 	uint32_t rxcfg, txcfg, v;
2284 
2285 	sc = device_get_softc(dev);
2286 	ifp = sc->sc_ifp;
2287 
2288 	CAS_LOCK_ASSERT(sc, MA_OWNED);
2289 
2290 #ifdef CAS_DEBUG
2291 	if ((if_getflags(ifp) & IFF_DEBUG) != 0)
2292 		device_printf(sc->sc_dev, "%s: status changen", __func__);
2293 #endif
2294 
2295 	if ((sc->sc_mii->mii_media_status & IFM_ACTIVE) != 0 &&
2296 	    IFM_SUBTYPE(sc->sc_mii->mii_media_active) != IFM_NONE)
2297 		sc->sc_flags |= CAS_LINK;
2298 	else
2299 		sc->sc_flags &= ~CAS_LINK;
2300 
2301 	switch (IFM_SUBTYPE(sc->sc_mii->mii_media_active)) {
2302 	case IFM_1000_SX:
2303 	case IFM_1000_LX:
2304 	case IFM_1000_CX:
2305 	case IFM_1000_T:
2306 		gigabit = 1;
2307 		break;
2308 	default:
2309 		gigabit = 0;
2310 	}
2311 
2312 	/*
2313 	 * The configuration done here corresponds to the steps F) and
2314 	 * G) and as far as enabling of RX and TX MAC goes also step H)
2315 	 * of the initialization sequence outlined in section 11.2.1 of
2316 	 * the Cassini+ ASIC Specification.
2317 	 */
2318 
2319 	rxcfg = sc->sc_mac_rxcfg;
2320 	rxcfg &= ~CAS_MAC_RX_CONF_CARR;
2321 	txcfg = CAS_MAC_TX_CONF_EN_IPG0 | CAS_MAC_TX_CONF_NGU |
2322 	    CAS_MAC_TX_CONF_NGUL;
2323 	if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) & IFM_FDX) != 0)
2324 		txcfg |= CAS_MAC_TX_CONF_ICARR | CAS_MAC_TX_CONF_ICOLLIS;
2325 	else if (gigabit != 0) {
2326 		rxcfg |= CAS_MAC_RX_CONF_CARR;
2327 		txcfg |= CAS_MAC_TX_CONF_CARR;
2328 	}
2329 	(void)cas_disable_tx(sc);
2330 	CAS_WRITE_4(sc, CAS_MAC_TX_CONF, txcfg);
2331 	(void)cas_disable_rx(sc);
2332 	CAS_WRITE_4(sc, CAS_MAC_RX_CONF, rxcfg);
2333 
2334 	v = CAS_READ_4(sc, CAS_MAC_CTRL_CONF) &
2335 	    ~(CAS_MAC_CTRL_CONF_TXP | CAS_MAC_CTRL_CONF_RXP);
2336 	if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) &
2337 	    IFM_ETH_RXPAUSE) != 0)
2338 		v |= CAS_MAC_CTRL_CONF_RXP;
2339 	if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) &
2340 	    IFM_ETH_TXPAUSE) != 0)
2341 		v |= CAS_MAC_CTRL_CONF_TXP;
2342 	CAS_WRITE_4(sc, CAS_MAC_CTRL_CONF, v);
2343 
2344 	/*
2345 	 * All supported chips have a bug causing incorrect checksum
2346 	 * to be calculated when letting them strip the FCS in half-
2347 	 * duplex mode.  In theory we could disable FCS stripping and
2348 	 * manually adjust the checksum accordingly.  It seems to make
2349 	 * more sense to optimze for the common case and just disable
2350 	 * hardware checksumming in half-duplex mode though.
2351 	 */
2352 	if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) & IFM_FDX) == 0) {
2353 		if_setcapenablebit(ifp, 0, IFCAP_HWCSUM);
2354 		if_sethwassist(ifp, 0);
2355 	} else if ((sc->sc_flags & CAS_NO_CSUM) == 0) {
2356 		if_setcapenable(ifp, if_getcapabilities(ifp));
2357 		if_sethwassist(ifp, CAS_CSUM_FEATURES);
2358 	}
2359 
2360 	if (sc->sc_variant == CAS_SATURN) {
2361 		if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) & IFM_FDX) == 0)
2362 			/* silicon bug workaround */
2363 			CAS_WRITE_4(sc, CAS_MAC_PREAMBLE_LEN, 0x41);
2364 		else
2365 			CAS_WRITE_4(sc, CAS_MAC_PREAMBLE_LEN, 0x7);
2366 	}
2367 
2368 	if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) & IFM_FDX) == 0 &&
2369 	    gigabit != 0)
2370 		CAS_WRITE_4(sc, CAS_MAC_SLOT_TIME,
2371 		    CAS_MAC_SLOT_TIME_CARR);
2372 	else
2373 		CAS_WRITE_4(sc, CAS_MAC_SLOT_TIME,
2374 		    CAS_MAC_SLOT_TIME_NORM);
2375 
2376 	/* XIF Configuration */
2377 	v = CAS_MAC_XIF_CONF_TX_OE | CAS_MAC_XIF_CONF_LNKLED;
2378 	if ((sc->sc_flags & CAS_SERDES) == 0) {
2379 		if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) & IFM_FDX) == 0)
2380 			v |= CAS_MAC_XIF_CONF_NOECHO;
2381 		v |= CAS_MAC_XIF_CONF_BUF_OE;
2382 	}
2383 	if (gigabit != 0)
2384 		v |= CAS_MAC_XIF_CONF_GMII;
2385 	if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) & IFM_FDX) != 0)
2386 		v |= CAS_MAC_XIF_CONF_FDXLED;
2387 	CAS_WRITE_4(sc, CAS_MAC_XIF_CONF, v);
2388 
2389 	sc->sc_mac_rxcfg = rxcfg;
2390 	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0 &&
2391 	    (sc->sc_flags & CAS_LINK) != 0) {
2392 		CAS_WRITE_4(sc, CAS_MAC_TX_CONF,
2393 		    txcfg | CAS_MAC_TX_CONF_EN);
2394 		CAS_WRITE_4(sc, CAS_MAC_RX_CONF,
2395 		    rxcfg | CAS_MAC_RX_CONF_EN);
2396 	}
2397 }
2398 
2399 static int
2400 cas_mediachange(if_t ifp)
2401 {
2402 	struct cas_softc *sc = if_getsoftc(ifp);
2403 	int error;
2404 
2405 	/* XXX add support for serial media. */
2406 
2407 	CAS_LOCK(sc);
2408 	error = mii_mediachg(sc->sc_mii);
2409 	CAS_UNLOCK(sc);
2410 	return (error);
2411 }
2412 
2413 static void
2414 cas_mediastatus(if_t ifp, struct ifmediareq *ifmr)
2415 {
2416 	struct cas_softc *sc = if_getsoftc(ifp);
2417 
2418 	CAS_LOCK(sc);
2419 	if ((if_getflags(ifp) & IFF_UP) == 0) {
2420 		CAS_UNLOCK(sc);
2421 		return;
2422 	}
2423 
2424 	mii_pollstat(sc->sc_mii);
2425 	ifmr->ifm_active = sc->sc_mii->mii_media_active;
2426 	ifmr->ifm_status = sc->sc_mii->mii_media_status;
2427 	CAS_UNLOCK(sc);
2428 }
2429 
2430 static int
2431 cas_ioctl(if_t ifp, u_long cmd, caddr_t data)
2432 {
2433 	struct cas_softc *sc = if_getsoftc(ifp);
2434 	struct ifreq *ifr = (struct ifreq *)data;
2435 	int error;
2436 
2437 	error = 0;
2438 	switch (cmd) {
2439 	case SIOCSIFFLAGS:
2440 		CAS_LOCK(sc);
2441 		if ((if_getflags(ifp) & IFF_UP) != 0) {
2442 			if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0 &&
2443 			    ((if_getflags(ifp) ^ sc->sc_ifflags) &
2444 			    (IFF_ALLMULTI | IFF_PROMISC)) != 0)
2445 				cas_setladrf(sc);
2446 			else
2447 				cas_init_locked(sc);
2448 		} else if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0)
2449 			cas_stop(ifp);
2450 		sc->sc_ifflags = if_getflags(ifp);
2451 		CAS_UNLOCK(sc);
2452 		break;
2453 	case SIOCSIFCAP:
2454 		CAS_LOCK(sc);
2455 		if ((sc->sc_flags & CAS_NO_CSUM) != 0) {
2456 			error = EINVAL;
2457 			CAS_UNLOCK(sc);
2458 			break;
2459 		}
2460 		if_setcapenable(ifp, ifr->ifr_reqcap);
2461 		if ((if_getcapenable(ifp) & IFCAP_TXCSUM) != 0)
2462 			if_sethwassist(ifp, CAS_CSUM_FEATURES);
2463 		else
2464 			if_sethwassist(ifp, 0);
2465 		CAS_UNLOCK(sc);
2466 		break;
2467 	case SIOCADDMULTI:
2468 	case SIOCDELMULTI:
2469 		CAS_LOCK(sc);
2470 		if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0)
2471 			cas_setladrf(sc);
2472 		CAS_UNLOCK(sc);
2473 		break;
2474 	case SIOCSIFMTU:
2475 		if ((ifr->ifr_mtu < ETHERMIN) ||
2476 		    (ifr->ifr_mtu > ETHERMTU_JUMBO))
2477 			error = EINVAL;
2478 		else
2479 			if_setmtu(ifp, ifr->ifr_mtu);
2480 		break;
2481 	case SIOCGIFMEDIA:
2482 	case SIOCSIFMEDIA:
2483 		error = ifmedia_ioctl(ifp, ifr, &sc->sc_mii->mii_media, cmd);
2484 		break;
2485 	default:
2486 		error = ether_ioctl(ifp, cmd, data);
2487 		break;
2488 	}
2489 
2490 	return (error);
2491 }
2492 
2493 static u_int
2494 cas_hash_maddr(void *arg, struct sockaddr_dl *sdl, u_int cnt)
2495 {
2496 	uint32_t crc, *hash = arg;
2497 
2498 	crc = ether_crc32_le(LLADDR(sdl), ETHER_ADDR_LEN);
2499 	/* We just want the 8 most significant bits. */
2500 	crc >>= 24;
2501 	/* Set the corresponding bit in the filter. */
2502 	hash[crc >> 4] |= 1 << (15 - (crc & 15));
2503 
2504 	return (1);
2505 }
2506 
2507 static void
2508 cas_setladrf(struct cas_softc *sc)
2509 {
2510 	if_t ifp = sc->sc_ifp;
2511 	int i;
2512 	uint32_t hash[16];
2513 	uint32_t v;
2514 
2515 	CAS_LOCK_ASSERT(sc, MA_OWNED);
2516 
2517 	/*
2518 	 * Turn off the RX MAC and the hash filter as required by the Sun
2519 	 * Cassini programming restrictions.
2520 	 */
2521 	v = sc->sc_mac_rxcfg & ~(CAS_MAC_RX_CONF_HFILTER |
2522 	    CAS_MAC_RX_CONF_EN);
2523 	CAS_WRITE_4(sc, CAS_MAC_RX_CONF, v);
2524 	CAS_BARRIER(sc, CAS_MAC_RX_CONF, 4,
2525 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2526 	if (!cas_bitwait(sc, CAS_MAC_RX_CONF, CAS_MAC_RX_CONF_HFILTER |
2527 	    CAS_MAC_RX_CONF_EN, 0))
2528 		device_printf(sc->sc_dev,
2529 		    "cannot disable RX MAC or hash filter\n");
2530 
2531 	v &= ~(CAS_MAC_RX_CONF_PROMISC | CAS_MAC_RX_CONF_PGRP);
2532 	if ((if_getflags(ifp) & IFF_PROMISC) != 0) {
2533 		v |= CAS_MAC_RX_CONF_PROMISC;
2534 		goto chipit;
2535 	}
2536 	if ((if_getflags(ifp) & IFF_ALLMULTI) != 0) {
2537 		v |= CAS_MAC_RX_CONF_PGRP;
2538 		goto chipit;
2539 	}
2540 
2541 	/*
2542 	 * Set up multicast address filter by passing all multicast
2543 	 * addresses through a crc generator, and then using the high
2544 	 * order 8 bits as an index into the 256 bit logical address
2545 	 * filter.  The high order 4 bits selects the word, while the
2546 	 * other 4 bits select the bit within the word (where bit 0
2547 	 * is the MSB).
2548 	 */
2549 
2550 	memset(hash, 0, sizeof(hash));
2551 	if_foreach_llmaddr(ifp, cas_hash_maddr, &hash);
2552 
2553 	v |= CAS_MAC_RX_CONF_HFILTER;
2554 
2555 	/* Now load the hash table into the chip (if we are using it). */
2556 	for (i = 0; i < 16; i++)
2557 		CAS_WRITE_4(sc,
2558 		    CAS_MAC_HASH0 + i * (CAS_MAC_HASH1 - CAS_MAC_HASH0),
2559 		    hash[i]);
2560 
2561  chipit:
2562 	sc->sc_mac_rxcfg = v;
2563 	CAS_WRITE_4(sc, CAS_MAC_RX_CONF, v | CAS_MAC_RX_CONF_EN);
2564 }
2565 
2566 static int	cas_pci_attach(device_t dev);
2567 static int	cas_pci_detach(device_t dev);
2568 static int	cas_pci_probe(device_t dev);
2569 static int	cas_pci_resume(device_t dev);
2570 static int	cas_pci_suspend(device_t dev);
2571 
2572 static device_method_t cas_pci_methods[] = {
2573 	/* Device interface */
2574 	DEVMETHOD(device_probe,		cas_pci_probe),
2575 	DEVMETHOD(device_attach,	cas_pci_attach),
2576 	DEVMETHOD(device_detach,	cas_pci_detach),
2577 	DEVMETHOD(device_suspend,	cas_pci_suspend),
2578 	DEVMETHOD(device_resume,	cas_pci_resume),
2579 	/* Use the suspend handler here, it is all that is required. */
2580 	DEVMETHOD(device_shutdown,	cas_pci_suspend),
2581 
2582 	/* MII interface */
2583 	DEVMETHOD(miibus_readreg,	cas_mii_readreg),
2584 	DEVMETHOD(miibus_writereg,	cas_mii_writereg),
2585 	DEVMETHOD(miibus_statchg,	cas_mii_statchg),
2586 
2587 	DEVMETHOD_END
2588 };
2589 
2590 static driver_t cas_pci_driver = {
2591 	"cas",
2592 	cas_pci_methods,
2593 	sizeof(struct cas_softc)
2594 };
2595 
2596 static const struct cas_pci_dev {
2597 	uint32_t	cpd_devid;
2598 	uint8_t		cpd_revid;
2599 	int		cpd_variant;
2600 	const char	*cpd_desc;
2601 } cas_pci_devlist[] = {
2602 	{ 0x0035100b, 0x0, CAS_SATURN, "NS DP83065 Saturn Gigabit Ethernet" },
2603 	{ 0xabba108e, 0x10, CAS_CASPLUS, "Sun Cassini+ Gigabit Ethernet" },
2604 	{ 0xabba108e, 0x0, CAS_CAS, "Sun Cassini Gigabit Ethernet" },
2605 	{ 0, 0, 0, NULL }
2606 };
2607 
2608 DRIVER_MODULE(cas, pci, cas_pci_driver, 0, 0);
2609 MODULE_PNP_INFO("W32:vendor/device", pci, cas, cas_pci_devlist,
2610     nitems(cas_pci_devlist) - 1);
2611 DRIVER_MODULE(miibus, cas, miibus_driver, 0, 0);
2612 MODULE_DEPEND(cas, pci, 1, 1, 1);
2613 
2614 static int
2615 cas_pci_probe(device_t dev)
2616 {
2617 	int i;
2618 
2619 	for (i = 0; cas_pci_devlist[i].cpd_desc != NULL; i++) {
2620 		if (pci_get_devid(dev) == cas_pci_devlist[i].cpd_devid &&
2621 		    pci_get_revid(dev) >= cas_pci_devlist[i].cpd_revid) {
2622 			device_set_desc(dev, cas_pci_devlist[i].cpd_desc);
2623 			return (BUS_PROBE_DEFAULT);
2624 		}
2625 	}
2626 
2627 	return (ENXIO);
2628 }
2629 
2630 static struct resource_spec cas_pci_res_spec[] = {
2631 	{ SYS_RES_IRQ, 0, RF_SHAREABLE | RF_ACTIVE },	/* CAS_RES_INTR */
2632 	{ SYS_RES_MEMORY, PCIR_BAR(0), RF_ACTIVE },	/* CAS_RES_MEM */
2633 	{ -1, 0 }
2634 };
2635 
2636 #define	CAS_LOCAL_MAC_ADDRESS	"local-mac-address"
2637 #define	CAS_PHY_INTERFACE	"phy-interface"
2638 #define	CAS_PHY_TYPE		"phy-type"
2639 #define	CAS_PHY_TYPE_PCS	"pcs"
2640 
2641 static int
2642 cas_pci_attach(device_t dev)
2643 {
2644 	char buf[sizeof(CAS_LOCAL_MAC_ADDRESS)];
2645 	struct cas_softc *sc;
2646 	int i;
2647 #if !defined(__powerpc__)
2648 	u_char enaddr[4][ETHER_ADDR_LEN];
2649 	u_int j, k, lma, pcs[4], phy;
2650 #endif
2651 
2652 	sc = device_get_softc(dev);
2653 	sc->sc_variant = CAS_UNKNOWN;
2654 	for (i = 0; cas_pci_devlist[i].cpd_desc != NULL; i++) {
2655 		if (pci_get_devid(dev) == cas_pci_devlist[i].cpd_devid &&
2656 		    pci_get_revid(dev) >= cas_pci_devlist[i].cpd_revid) {
2657 			sc->sc_variant = cas_pci_devlist[i].cpd_variant;
2658 			break;
2659 		}
2660 	}
2661 	if (sc->sc_variant == CAS_UNKNOWN) {
2662 		device_printf(dev, "unknown adaptor\n");
2663 		return (ENXIO);
2664 	}
2665 
2666 	/* PCI configuration */
2667 	pci_write_config(dev, PCIR_COMMAND,
2668 	    pci_read_config(dev, PCIR_COMMAND, 2) | PCIM_CMD_BUSMASTEREN |
2669 	    PCIM_CMD_MWRICEN | PCIM_CMD_PERRESPEN | PCIM_CMD_SERRESPEN, 2);
2670 
2671 	sc->sc_dev = dev;
2672 	if (sc->sc_variant == CAS_CAS && pci_get_devid(dev) < 0x02)
2673 		/* Hardware checksumming may hang TX. */
2674 		sc->sc_flags |= CAS_NO_CSUM;
2675 	if (sc->sc_variant == CAS_CASPLUS || sc->sc_variant == CAS_SATURN)
2676 		sc->sc_flags |= CAS_REG_PLUS;
2677 	if (sc->sc_variant == CAS_CAS ||
2678 	    (sc->sc_variant == CAS_CASPLUS && pci_get_revid(dev) < 0x11))
2679 		sc->sc_flags |= CAS_TABORT;
2680 	if (bootverbose)
2681 		device_printf(dev, "flags=0x%x\n", sc->sc_flags);
2682 
2683 	if (bus_alloc_resources(dev, cas_pci_res_spec, sc->sc_res)) {
2684 		device_printf(dev, "failed to allocate resources\n");
2685 		bus_release_resources(dev, cas_pci_res_spec, sc->sc_res);
2686 		return (ENXIO);
2687 	}
2688 
2689 	CAS_LOCK_INIT(sc, device_get_nameunit(dev));
2690 
2691 #if defined(__powerpc__)
2692 	OF_getetheraddr(dev, sc->sc_enaddr);
2693 	if (OF_getprop(ofw_bus_get_node(dev), CAS_PHY_INTERFACE, buf,
2694 	    sizeof(buf)) > 0 || OF_getprop(ofw_bus_get_node(dev),
2695 	    CAS_PHY_TYPE, buf, sizeof(buf)) > 0) {
2696 		buf[sizeof(buf) - 1] = '\0';
2697 		if (strcmp(buf, CAS_PHY_TYPE_PCS) == 0)
2698 			sc->sc_flags |= CAS_SERDES;
2699 	}
2700 #else
2701 	/*
2702 	 * Dig out VPD (vital product data) and read the MAC address as well
2703 	 * as the PHY type.  The VPD resides in the PCI Expansion ROM (PCI
2704 	 * FCode) and can't be accessed via the PCI capability pointer.
2705 	 * SUNW,pci-ce and SUNW,pci-qge use the Enhanced VPD format described
2706 	 * in the free US Patent 7149820.
2707 	 */
2708 
2709 #define	PCI_ROMHDR_SIZE			0x1c
2710 #define	PCI_ROMHDR_SIG			0x00
2711 #define	PCI_ROMHDR_SIG_MAGIC		0xaa55		/* little endian */
2712 #define	PCI_ROMHDR_PTR_DATA		0x18
2713 #define	PCI_ROM_SIZE			0x18
2714 #define	PCI_ROM_SIG			0x00
2715 #define	PCI_ROM_SIG_MAGIC		0x52494350	/* "PCIR", endian */
2716 							/* reversed */
2717 #define	PCI_ROM_VENDOR			0x04
2718 #define	PCI_ROM_DEVICE			0x06
2719 #define	PCI_ROM_PTR_VPD			0x08
2720 #define	PCI_VPDRES_BYTE0		0x00
2721 #define	PCI_VPDRES_ISLARGE(x)		((x) & 0x80)
2722 #define	PCI_VPDRES_LARGE_NAME(x)	((x) & 0x7f)
2723 #define	PCI_VPDRES_LARGE_LEN_LSB	0x01
2724 #define	PCI_VPDRES_LARGE_LEN_MSB	0x02
2725 #define	PCI_VPDRES_LARGE_SIZE		0x03
2726 #define	PCI_VPDRES_TYPE_ID_STRING	0x02		/* large */
2727 #define	PCI_VPDRES_TYPE_VPD		0x10		/* large */
2728 #define	PCI_VPD_KEY0			0x00
2729 #define	PCI_VPD_KEY1			0x01
2730 #define	PCI_VPD_LEN			0x02
2731 #define	PCI_VPD_SIZE			0x03
2732 
2733 #define	CAS_ROM_READ_1(sc, offs)					\
2734 	CAS_READ_1((sc), CAS_PCI_ROM_OFFSET + (offs))
2735 #define	CAS_ROM_READ_2(sc, offs)					\
2736 	CAS_READ_2((sc), CAS_PCI_ROM_OFFSET + (offs))
2737 #define	CAS_ROM_READ_4(sc, offs)					\
2738 	CAS_READ_4((sc), CAS_PCI_ROM_OFFSET + (offs))
2739 
2740 	lma = phy = 0;
2741 	memset(enaddr, 0, sizeof(enaddr));
2742 	memset(pcs, 0, sizeof(pcs));
2743 
2744 	/* Enable PCI Expansion ROM access. */
2745 	CAS_WRITE_4(sc, CAS_BIM_LDEV_OEN,
2746 	    CAS_BIM_LDEV_OEN_PAD | CAS_BIM_LDEV_OEN_PROM);
2747 
2748 	/* Read PCI Expansion ROM header. */
2749 	if (CAS_ROM_READ_2(sc, PCI_ROMHDR_SIG) != PCI_ROMHDR_SIG_MAGIC ||
2750 	    (i = CAS_ROM_READ_2(sc, PCI_ROMHDR_PTR_DATA)) <
2751 	    PCI_ROMHDR_SIZE) {
2752 		device_printf(dev, "unexpected PCI Expansion ROM header\n");
2753 		goto fail_prom;
2754 	}
2755 
2756 	/* Read PCI Expansion ROM data. */
2757 	if (CAS_ROM_READ_4(sc, i + PCI_ROM_SIG) != PCI_ROM_SIG_MAGIC ||
2758 	    CAS_ROM_READ_2(sc, i + PCI_ROM_VENDOR) != pci_get_vendor(dev) ||
2759 	    CAS_ROM_READ_2(sc, i + PCI_ROM_DEVICE) != pci_get_device(dev) ||
2760 	    (j = CAS_ROM_READ_2(sc, i + PCI_ROM_PTR_VPD)) <
2761 	    i + PCI_ROM_SIZE) {
2762 		device_printf(dev, "unexpected PCI Expansion ROM data\n");
2763 		goto fail_prom;
2764 	}
2765 
2766 	/* Read PCI VPD. */
2767  next:
2768 	if (PCI_VPDRES_ISLARGE(CAS_ROM_READ_1(sc,
2769 	    j + PCI_VPDRES_BYTE0)) == 0) {
2770 		device_printf(dev, "no large PCI VPD\n");
2771 		goto fail_prom;
2772 	}
2773 
2774 	i = (CAS_ROM_READ_1(sc, j + PCI_VPDRES_LARGE_LEN_MSB) << 8) |
2775 	    CAS_ROM_READ_1(sc, j + PCI_VPDRES_LARGE_LEN_LSB);
2776 	switch (PCI_VPDRES_LARGE_NAME(CAS_ROM_READ_1(sc,
2777 	    j + PCI_VPDRES_BYTE0))) {
2778 	case PCI_VPDRES_TYPE_ID_STRING:
2779 		/* Skip identifier string. */
2780 		j += PCI_VPDRES_LARGE_SIZE + i;
2781 		goto next;
2782 	case PCI_VPDRES_TYPE_VPD:
2783 		for (j += PCI_VPDRES_LARGE_SIZE; i > 0;
2784 		    i -= PCI_VPD_SIZE + CAS_ROM_READ_1(sc, j + PCI_VPD_LEN),
2785 		    j += PCI_VPD_SIZE + CAS_ROM_READ_1(sc, j + PCI_VPD_LEN)) {
2786 			if (CAS_ROM_READ_1(sc, j + PCI_VPD_KEY0) != 'Z')
2787 				/* no Enhanced VPD */
2788 				continue;
2789 			if (CAS_ROM_READ_1(sc, j + PCI_VPD_SIZE) != 'I')
2790 				/* no instance property */
2791 				continue;
2792 			if (CAS_ROM_READ_1(sc, j + PCI_VPD_SIZE + 3) == 'B') {
2793 				/* byte array */
2794 				if (CAS_ROM_READ_1(sc,
2795 				    j + PCI_VPD_SIZE + 4) != ETHER_ADDR_LEN)
2796 					continue;
2797 				bus_read_region_1(sc->sc_res[CAS_RES_MEM],
2798 				    CAS_PCI_ROM_OFFSET + j + PCI_VPD_SIZE + 5,
2799 				    buf, sizeof(buf));
2800 				buf[sizeof(buf) - 1] = '\0';
2801 				if (strcmp(buf, CAS_LOCAL_MAC_ADDRESS) != 0)
2802 					continue;
2803 				bus_read_region_1(sc->sc_res[CAS_RES_MEM],
2804 				    CAS_PCI_ROM_OFFSET + j + PCI_VPD_SIZE +
2805 				    5 + sizeof(CAS_LOCAL_MAC_ADDRESS),
2806 				    enaddr[lma], sizeof(enaddr[lma]));
2807 				lma++;
2808 				if (lma == 4 && phy == 4)
2809 					break;
2810 			} else if (CAS_ROM_READ_1(sc, j + PCI_VPD_SIZE + 3) ==
2811 			   'S') {
2812 				/* string */
2813 				if (CAS_ROM_READ_1(sc,
2814 				    j + PCI_VPD_SIZE + 4) !=
2815 				    sizeof(CAS_PHY_TYPE_PCS))
2816 					continue;
2817 				bus_read_region_1(sc->sc_res[CAS_RES_MEM],
2818 				    CAS_PCI_ROM_OFFSET + j + PCI_VPD_SIZE + 5,
2819 				    buf, sizeof(buf));
2820 				buf[sizeof(buf) - 1] = '\0';
2821 				if (strcmp(buf, CAS_PHY_INTERFACE) == 0)
2822 					k = sizeof(CAS_PHY_INTERFACE);
2823 				else if (strcmp(buf, CAS_PHY_TYPE) == 0)
2824 					k = sizeof(CAS_PHY_TYPE);
2825 				else
2826 					continue;
2827 				bus_read_region_1(sc->sc_res[CAS_RES_MEM],
2828 				    CAS_PCI_ROM_OFFSET + j + PCI_VPD_SIZE +
2829 				    5 + k, buf, sizeof(buf));
2830 				buf[sizeof(buf) - 1] = '\0';
2831 				if (strcmp(buf, CAS_PHY_TYPE_PCS) == 0)
2832 					pcs[phy] = 1;
2833 				phy++;
2834 				if (lma == 4 && phy == 4)
2835 					break;
2836 			}
2837 		}
2838 		break;
2839 	default:
2840 		device_printf(dev, "unexpected PCI VPD\n");
2841 		goto fail_prom;
2842 	}
2843 
2844  fail_prom:
2845 	CAS_WRITE_4(sc, CAS_BIM_LDEV_OEN, 0);
2846 
2847 	if (lma == 0) {
2848 		device_printf(dev, "could not determine Ethernet address\n");
2849 		goto fail;
2850 	}
2851 	i = 0;
2852 	if (lma > 1 && pci_get_slot(dev) < nitems(enaddr))
2853 		i = pci_get_slot(dev);
2854 	memcpy(sc->sc_enaddr, enaddr[i], ETHER_ADDR_LEN);
2855 
2856 	if (phy == 0) {
2857 		device_printf(dev, "could not determine PHY type\n");
2858 		goto fail;
2859 	}
2860 	i = 0;
2861 	if (phy > 1 && pci_get_slot(dev) < nitems(pcs))
2862 		i = pci_get_slot(dev);
2863 	if (pcs[i] != 0)
2864 		sc->sc_flags |= CAS_SERDES;
2865 #endif
2866 
2867 	if (cas_attach(sc) != 0) {
2868 		device_printf(dev, "could not be attached\n");
2869 		goto fail;
2870 	}
2871 
2872 	if (bus_setup_intr(dev, sc->sc_res[CAS_RES_INTR], INTR_TYPE_NET |
2873 	    INTR_MPSAFE, cas_intr, NULL, sc, &sc->sc_ih) != 0) {
2874 		device_printf(dev, "failed to set up interrupt\n");
2875 		cas_detach(sc);
2876 		goto fail;
2877 	}
2878 	return (0);
2879 
2880  fail:
2881 	CAS_LOCK_DESTROY(sc);
2882 	bus_release_resources(dev, cas_pci_res_spec, sc->sc_res);
2883 	return (ENXIO);
2884 }
2885 
2886 static int
2887 cas_pci_detach(device_t dev)
2888 {
2889 	struct cas_softc *sc;
2890 
2891 	sc = device_get_softc(dev);
2892 	bus_teardown_intr(dev, sc->sc_res[CAS_RES_INTR], sc->sc_ih);
2893 	cas_detach(sc);
2894 	CAS_LOCK_DESTROY(sc);
2895 	bus_release_resources(dev, cas_pci_res_spec, sc->sc_res);
2896 	return (0);
2897 }
2898 
2899 static int
2900 cas_pci_suspend(device_t dev)
2901 {
2902 
2903 	cas_suspend(device_get_softc(dev));
2904 	return (0);
2905 }
2906 
2907 static int
2908 cas_pci_resume(device_t dev)
2909 {
2910 
2911 	cas_resume(device_get_softc(dev));
2912 	return (0);
2913 }
2914