xref: /freebsd/sys/dev/cas/if_cas.c (revision d5b0e70f7e04d971691517ce1304d86a1e367e2e)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
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 __FBSDID("$FreeBSD$");
36 
37 /*
38  * driver for Sun Cassini/Cassini+ and National Semiconductor DP83065
39  * Saturn Gigabit Ethernet controllers
40  */
41 
42 #if 0
43 #define	CAS_DEBUG
44 #endif
45 
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 #include <sys/bus.h>
49 #include <sys/callout.h>
50 #include <sys/endian.h>
51 #include <sys/mbuf.h>
52 #include <sys/malloc.h>
53 #include <sys/kernel.h>
54 #include <sys/lock.h>
55 #include <sys/module.h>
56 #include <sys/mutex.h>
57 #include <sys/refcount.h>
58 #include <sys/resource.h>
59 #include <sys/rman.h>
60 #include <sys/socket.h>
61 #include <sys/sockio.h>
62 #include <sys/taskqueue.h>
63 
64 #include <net/bpf.h>
65 #include <net/ethernet.h>
66 #include <net/if.h>
67 #include <net/if_var.h>
68 #include <net/if_arp.h>
69 #include <net/if_dl.h>
70 #include <net/if_media.h>
71 #include <net/if_types.h>
72 #include <net/if_vlan_var.h>
73 
74 #include <netinet/in.h>
75 #include <netinet/in_systm.h>
76 #include <netinet/ip.h>
77 #include <netinet/tcp.h>
78 #include <netinet/udp.h>
79 
80 #include <machine/bus.h>
81 #if defined(__powerpc__)
82 #include <dev/ofw/ofw_bus.h>
83 #include <dev/ofw/openfirm.h>
84 #include <machine/ofw_machdep.h>
85 #endif
86 #include <machine/resource.h>
87 
88 #include <dev/mii/mii.h>
89 #include <dev/mii/miivar.h>
90 
91 #include <dev/cas/if_casreg.h>
92 #include <dev/cas/if_casvar.h>
93 
94 #include <dev/pci/pcireg.h>
95 #include <dev/pci/pcivar.h>
96 
97 #include "miibus_if.h"
98 
99 #define RINGASSERT(n , min, max)					\
100 	CTASSERT(powerof2(n) && (n) >= (min) && (n) <= (max))
101 
102 RINGASSERT(CAS_NRXCOMP, 128, 32768);
103 RINGASSERT(CAS_NRXDESC, 32, 8192);
104 RINGASSERT(CAS_NRXDESC2, 32, 8192);
105 RINGASSERT(CAS_NTXDESC, 32, 8192);
106 
107 #undef RINGASSERT
108 
109 #define	CCDASSERT(m, a)							\
110 	CTASSERT((offsetof(struct cas_control_data, m) & ((a) - 1)) == 0)
111 
112 CCDASSERT(ccd_rxcomps, CAS_RX_COMP_ALIGN);
113 CCDASSERT(ccd_rxdescs, CAS_RX_DESC_ALIGN);
114 CCDASSERT(ccd_rxdescs2, CAS_RX_DESC_ALIGN);
115 
116 #undef CCDASSERT
117 
118 #define	CAS_TRIES	10000
119 
120 /*
121  * According to documentation, the hardware has support for basic TCP
122  * checksum offloading only, in practice this can be also used for UDP
123  * however (i.e. the problem of previous Sun NICs that a checksum of 0x0
124  * is not converted to 0xffff no longer exists).
125  */
126 #define	CAS_CSUM_FEATURES	(CSUM_TCP | CSUM_UDP)
127 
128 static inline void cas_add_rxdesc(struct cas_softc *sc, u_int idx);
129 static int	cas_attach(struct cas_softc *sc);
130 static int	cas_bitwait(struct cas_softc *sc, bus_addr_t r, uint32_t clr,
131 		    uint32_t set);
132 static void	cas_cddma_callback(void *xsc, bus_dma_segment_t *segs,
133 		    int nsegs, int error);
134 static void	cas_detach(struct cas_softc *sc);
135 static int	cas_disable_rx(struct cas_softc *sc);
136 static int	cas_disable_tx(struct cas_softc *sc);
137 static void	cas_eint(struct cas_softc *sc, u_int status);
138 static void	cas_free(struct mbuf *m);
139 static void	cas_init(void *xsc);
140 static void	cas_init_locked(struct cas_softc *sc);
141 static void	cas_init_regs(struct cas_softc *sc);
142 static int	cas_intr(void *v);
143 static void	cas_intr_task(void *arg, int pending __unused);
144 static int	cas_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data);
145 static int	cas_load_txmbuf(struct cas_softc *sc, struct mbuf **m_head);
146 static int	cas_mediachange(struct ifnet *ifp);
147 static void	cas_mediastatus(struct ifnet *ifp, struct ifmediareq *ifmr);
148 static void	cas_meminit(struct cas_softc *sc);
149 static void	cas_mifinit(struct cas_softc *sc);
150 static int	cas_mii_readreg(device_t dev, int phy, int reg);
151 static void	cas_mii_statchg(device_t dev);
152 static int	cas_mii_writereg(device_t dev, int phy, int reg, int val);
153 static void	cas_reset(struct cas_softc *sc);
154 static int	cas_reset_rx(struct cas_softc *sc);
155 static int	cas_reset_tx(struct cas_softc *sc);
156 static void	cas_resume(struct cas_softc *sc);
157 static u_int	cas_descsize(u_int sz);
158 static void	cas_rint(struct cas_softc *sc);
159 static void	cas_rint_timeout(void *arg);
160 static inline void cas_rxcksum(struct mbuf *m, uint16_t cksum);
161 static inline void cas_rxcompinit(struct cas_rx_comp *rxcomp);
162 static u_int	cas_rxcompsize(u_int sz);
163 static void	cas_rxdma_callback(void *xsc, bus_dma_segment_t *segs,
164 		    int nsegs, int error);
165 static void	cas_setladrf(struct cas_softc *sc);
166 static void	cas_start(struct ifnet *ifp);
167 static void	cas_stop(struct ifnet *ifp);
168 static void	cas_suspend(struct cas_softc *sc);
169 static void	cas_tick(void *arg);
170 static void	cas_tint(struct cas_softc *sc);
171 static void	cas_tx_task(void *arg, int pending __unused);
172 static inline void cas_txkick(struct cas_softc *sc);
173 static void	cas_watchdog(struct cas_softc *sc);
174 
175 MODULE_DEPEND(cas, ether, 1, 1, 1);
176 MODULE_DEPEND(cas, miibus, 1, 1, 1);
177 
178 #ifdef CAS_DEBUG
179 #include <sys/ktr.h>
180 #define	KTR_CAS		KTR_SPARE2
181 #endif
182 
183 static int
184 cas_attach(struct cas_softc *sc)
185 {
186 	struct cas_txsoft *txs;
187 	struct ifnet *ifp;
188 	int error, i;
189 	uint32_t v;
190 
191 	/* Set up ifnet structure. */
192 	ifp = sc->sc_ifp = if_alloc(IFT_ETHER);
193 	if (ifp == NULL)
194 		return (ENOSPC);
195 	ifp->if_softc = sc;
196 	if_initname(ifp, device_get_name(sc->sc_dev),
197 	    device_get_unit(sc->sc_dev));
198 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
199 	ifp->if_start = cas_start;
200 	ifp->if_ioctl = cas_ioctl;
201 	ifp->if_init = cas_init;
202 	IFQ_SET_MAXLEN(&ifp->if_snd, CAS_TXQUEUELEN);
203 	ifp->if_snd.ifq_drv_maxlen = CAS_TXQUEUELEN;
204 	IFQ_SET_READY(&ifp->if_snd);
205 
206 	callout_init_mtx(&sc->sc_tick_ch, &sc->sc_mtx, 0);
207 	callout_init_mtx(&sc->sc_rx_ch, &sc->sc_mtx, 0);
208 	/* Create local taskq. */
209 	NET_TASK_INIT(&sc->sc_intr_task, 0, cas_intr_task, sc);
210 	TASK_INIT(&sc->sc_tx_task, 1, cas_tx_task, ifp);
211 	sc->sc_tq = taskqueue_create_fast("cas_taskq", M_WAITOK,
212 	    taskqueue_thread_enqueue, &sc->sc_tq);
213 	if (sc->sc_tq == NULL) {
214 		device_printf(sc->sc_dev, "could not create taskqueue\n");
215 		error = ENXIO;
216 		goto fail_ifnet;
217 	}
218 	error = taskqueue_start_threads(&sc->sc_tq, 1, PI_NET, "%s taskq",
219 	    device_get_nameunit(sc->sc_dev));
220 	if (error != 0) {
221 		device_printf(sc->sc_dev, "could not start threads\n");
222 		goto fail_taskq;
223 	}
224 
225 	/* Make sure the chip is stopped. */
226 	cas_reset(sc);
227 
228 	error = bus_dma_tag_create(bus_get_dma_tag(sc->sc_dev), 1, 0,
229 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
230 	    BUS_SPACE_MAXSIZE, 0, BUS_SPACE_MAXSIZE, 0, NULL, NULL,
231 	    &sc->sc_pdmatag);
232 	if (error != 0)
233 		goto fail_taskq;
234 
235 	error = bus_dma_tag_create(sc->sc_pdmatag, 1, 0,
236 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
237 	    CAS_PAGE_SIZE, 1, CAS_PAGE_SIZE, 0, NULL, NULL, &sc->sc_rdmatag);
238 	if (error != 0)
239 		goto fail_ptag;
240 
241 	error = bus_dma_tag_create(sc->sc_pdmatag, 1, 0,
242 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
243 	    MCLBYTES * CAS_NTXSEGS, CAS_NTXSEGS, MCLBYTES,
244 	    BUS_DMA_ALLOCNOW, NULL, NULL, &sc->sc_tdmatag);
245 	if (error != 0)
246 		goto fail_rtag;
247 
248 	error = bus_dma_tag_create(sc->sc_pdmatag, CAS_TX_DESC_ALIGN, 0,
249 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
250 	    sizeof(struct cas_control_data), 1,
251 	    sizeof(struct cas_control_data), 0,
252 	    NULL, NULL, &sc->sc_cdmatag);
253 	if (error != 0)
254 		goto fail_ttag;
255 
256 	/*
257 	 * Allocate the control data structures, create and load the
258 	 * DMA map for it.
259 	 */
260 	if ((error = bus_dmamem_alloc(sc->sc_cdmatag,
261 	    (void **)&sc->sc_control_data,
262 	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
263 	    &sc->sc_cddmamap)) != 0) {
264 		device_printf(sc->sc_dev,
265 		    "unable to allocate control data, error = %d\n", error);
266 		goto fail_ctag;
267 	}
268 
269 	sc->sc_cddma = 0;
270 	if ((error = bus_dmamap_load(sc->sc_cdmatag, sc->sc_cddmamap,
271 	    sc->sc_control_data, sizeof(struct cas_control_data),
272 	    cas_cddma_callback, sc, 0)) != 0 || sc->sc_cddma == 0) {
273 		device_printf(sc->sc_dev,
274 		    "unable to load control data DMA map, error = %d\n",
275 		    error);
276 		goto fail_cmem;
277 	}
278 
279 	/*
280 	 * Initialize the transmit job descriptors.
281 	 */
282 	STAILQ_INIT(&sc->sc_txfreeq);
283 	STAILQ_INIT(&sc->sc_txdirtyq);
284 
285 	/*
286 	 * Create the transmit buffer DMA maps.
287 	 */
288 	error = ENOMEM;
289 	for (i = 0; i < CAS_TXQUEUELEN; i++) {
290 		txs = &sc->sc_txsoft[i];
291 		txs->txs_mbuf = NULL;
292 		txs->txs_ndescs = 0;
293 		if ((error = bus_dmamap_create(sc->sc_tdmatag, 0,
294 		    &txs->txs_dmamap)) != 0) {
295 			device_printf(sc->sc_dev,
296 			    "unable to create TX DMA map %d, error = %d\n",
297 			    i, error);
298 			goto fail_txd;
299 		}
300 		STAILQ_INSERT_TAIL(&sc->sc_txfreeq, txs, txs_q);
301 	}
302 
303 	/*
304 	 * Allocate the receive buffers, create and load the DMA maps
305 	 * for them.
306 	 */
307 	for (i = 0; i < CAS_NRXDESC; i++) {
308 		if ((error = bus_dmamem_alloc(sc->sc_rdmatag,
309 		    &sc->sc_rxdsoft[i].rxds_buf, BUS_DMA_WAITOK,
310 		    &sc->sc_rxdsoft[i].rxds_dmamap)) != 0) {
311 			device_printf(sc->sc_dev,
312 			    "unable to allocate RX buffer %d, error = %d\n",
313 			    i, error);
314 			goto fail_rxmem;
315 		}
316 
317 		sc->sc_rxdptr = i;
318 		sc->sc_rxdsoft[i].rxds_paddr = 0;
319 		if ((error = bus_dmamap_load(sc->sc_rdmatag,
320 		    sc->sc_rxdsoft[i].rxds_dmamap, sc->sc_rxdsoft[i].rxds_buf,
321 		    CAS_PAGE_SIZE, cas_rxdma_callback, sc, 0)) != 0 ||
322 		    sc->sc_rxdsoft[i].rxds_paddr == 0) {
323 			device_printf(sc->sc_dev,
324 			    "unable to load RX DMA map %d, error = %d\n",
325 			    i, error);
326 			goto fail_rxmap;
327 		}
328 	}
329 
330 	if ((sc->sc_flags & CAS_SERDES) == 0) {
331 		CAS_WRITE_4(sc, CAS_PCS_DATAPATH, CAS_PCS_DATAPATH_MII);
332 		CAS_BARRIER(sc, CAS_PCS_DATAPATH, 4,
333 		    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
334 		cas_mifinit(sc);
335 		/*
336 		 * Look for an external PHY.
337 		 */
338 		error = ENXIO;
339 		v = CAS_READ_4(sc, CAS_MIF_CONF);
340 		if ((v & CAS_MIF_CONF_MDI1) != 0) {
341 			v |= CAS_MIF_CONF_PHY_SELECT;
342 			CAS_WRITE_4(sc, CAS_MIF_CONF, v);
343 			CAS_BARRIER(sc, CAS_MIF_CONF, 4,
344 			    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
345 			/* Enable/unfreeze the GMII pins of Saturn. */
346 			if (sc->sc_variant == CAS_SATURN) {
347 				CAS_WRITE_4(sc, CAS_SATURN_PCFG,
348 				    CAS_READ_4(sc, CAS_SATURN_PCFG) &
349 				    ~CAS_SATURN_PCFG_FSI);
350 				CAS_BARRIER(sc, CAS_SATURN_PCFG, 4,
351 				    BUS_SPACE_BARRIER_READ |
352 				    BUS_SPACE_BARRIER_WRITE);
353 				DELAY(10000);
354 			}
355 			error = mii_attach(sc->sc_dev, &sc->sc_miibus, ifp,
356 			    cas_mediachange, cas_mediastatus, BMSR_DEFCAPMASK,
357 			    MII_PHY_ANY, MII_OFFSET_ANY, MIIF_DOPAUSE);
358 		}
359 		/*
360 		 * Fall back on an internal PHY if no external PHY was found.
361 		 */
362 		if (error != 0 && (v & CAS_MIF_CONF_MDI0) != 0) {
363 			v &= ~CAS_MIF_CONF_PHY_SELECT;
364 			CAS_WRITE_4(sc, CAS_MIF_CONF, v);
365 			CAS_BARRIER(sc, CAS_MIF_CONF, 4,
366 			    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
367 			/* Freeze the GMII pins of Saturn for saving power. */
368 			if (sc->sc_variant == CAS_SATURN) {
369 				CAS_WRITE_4(sc, CAS_SATURN_PCFG,
370 				    CAS_READ_4(sc, CAS_SATURN_PCFG) |
371 				    CAS_SATURN_PCFG_FSI);
372 				CAS_BARRIER(sc, CAS_SATURN_PCFG, 4,
373 				    BUS_SPACE_BARRIER_READ |
374 				    BUS_SPACE_BARRIER_WRITE);
375 				DELAY(10000);
376 			}
377 			error = mii_attach(sc->sc_dev, &sc->sc_miibus, ifp,
378 			    cas_mediachange, cas_mediastatus, BMSR_DEFCAPMASK,
379 			    MII_PHY_ANY, MII_OFFSET_ANY, MIIF_DOPAUSE);
380 		}
381 	} else {
382 		/*
383 		 * Use the external PCS SERDES.
384 		 */
385 		CAS_WRITE_4(sc, CAS_PCS_DATAPATH, CAS_PCS_DATAPATH_SERDES);
386 		CAS_BARRIER(sc, CAS_PCS_DATAPATH, 4, BUS_SPACE_BARRIER_WRITE);
387 		/* Enable/unfreeze the SERDES pins of Saturn. */
388 		if (sc->sc_variant == CAS_SATURN) {
389 			CAS_WRITE_4(sc, CAS_SATURN_PCFG, 0);
390 			CAS_BARRIER(sc, CAS_SATURN_PCFG, 4,
391 			    BUS_SPACE_BARRIER_WRITE);
392 		}
393 		CAS_WRITE_4(sc, CAS_PCS_SERDES_CTRL, CAS_PCS_SERDES_CTRL_ESD);
394 		CAS_BARRIER(sc, CAS_PCS_SERDES_CTRL, 4,
395 		    BUS_SPACE_BARRIER_WRITE);
396 		CAS_WRITE_4(sc, CAS_PCS_CONF, CAS_PCS_CONF_EN);
397 		CAS_BARRIER(sc, CAS_PCS_CONF, 4,
398 		    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
399 		error = mii_attach(sc->sc_dev, &sc->sc_miibus, ifp,
400 		    cas_mediachange, cas_mediastatus, BMSR_DEFCAPMASK,
401 		    CAS_PHYAD_EXTERNAL, MII_OFFSET_ANY, MIIF_DOPAUSE);
402 	}
403 	if (error != 0) {
404 		device_printf(sc->sc_dev, "attaching PHYs failed\n");
405 		goto fail_rxmap;
406 	}
407 	sc->sc_mii = device_get_softc(sc->sc_miibus);
408 
409 	/*
410 	 * From this point forward, the attachment cannot fail.  A failure
411 	 * before this point releases all resources that may have been
412 	 * allocated.
413 	 */
414 
415 	/* Announce FIFO sizes. */
416 	v = CAS_READ_4(sc, CAS_TX_FIFO_SIZE);
417 	device_printf(sc->sc_dev, "%ukB RX FIFO, %ukB TX FIFO\n",
418 	    CAS_RX_FIFO_SIZE / 1024, v / 16);
419 
420 	/* Attach the interface. */
421 	ether_ifattach(ifp, sc->sc_enaddr);
422 
423 	/*
424 	 * Tell the upper layer(s) we support long frames/checksum offloads.
425 	 */
426 	ifp->if_hdrlen = sizeof(struct ether_vlan_header);
427 	ifp->if_capabilities = IFCAP_VLAN_MTU;
428 	if ((sc->sc_flags & CAS_NO_CSUM) == 0) {
429 		ifp->if_capabilities |= IFCAP_HWCSUM;
430 		ifp->if_hwassist = CAS_CSUM_FEATURES;
431 	}
432 	ifp->if_capenable = ifp->if_capabilities;
433 
434 	return (0);
435 
436 	/*
437 	 * Free any resources we've allocated during the failed attach
438 	 * attempt.  Do this in reverse order and fall through.
439 	 */
440  fail_rxmap:
441 	for (i = 0; i < CAS_NRXDESC; i++)
442 		if (sc->sc_rxdsoft[i].rxds_paddr != 0)
443 			bus_dmamap_unload(sc->sc_rdmatag,
444 			    sc->sc_rxdsoft[i].rxds_dmamap);
445  fail_rxmem:
446 	for (i = 0; i < CAS_NRXDESC; i++)
447 		if (sc->sc_rxdsoft[i].rxds_buf != NULL)
448 			bus_dmamem_free(sc->sc_rdmatag,
449 			    sc->sc_rxdsoft[i].rxds_buf,
450 			    sc->sc_rxdsoft[i].rxds_dmamap);
451  fail_txd:
452 	for (i = 0; i < CAS_TXQUEUELEN; i++)
453 		if (sc->sc_txsoft[i].txs_dmamap != NULL)
454 			bus_dmamap_destroy(sc->sc_tdmatag,
455 			    sc->sc_txsoft[i].txs_dmamap);
456 	bus_dmamap_unload(sc->sc_cdmatag, sc->sc_cddmamap);
457  fail_cmem:
458 	bus_dmamem_free(sc->sc_cdmatag, sc->sc_control_data,
459 	    sc->sc_cddmamap);
460  fail_ctag:
461 	bus_dma_tag_destroy(sc->sc_cdmatag);
462  fail_ttag:
463 	bus_dma_tag_destroy(sc->sc_tdmatag);
464  fail_rtag:
465 	bus_dma_tag_destroy(sc->sc_rdmatag);
466  fail_ptag:
467 	bus_dma_tag_destroy(sc->sc_pdmatag);
468  fail_taskq:
469 	taskqueue_free(sc->sc_tq);
470  fail_ifnet:
471 	if_free(ifp);
472 	return (error);
473 }
474 
475 static void
476 cas_detach(struct cas_softc *sc)
477 {
478 	struct ifnet *ifp = sc->sc_ifp;
479 	int i;
480 
481 	ether_ifdetach(ifp);
482 	CAS_LOCK(sc);
483 	cas_stop(ifp);
484 	CAS_UNLOCK(sc);
485 	callout_drain(&sc->sc_tick_ch);
486 	callout_drain(&sc->sc_rx_ch);
487 	taskqueue_drain(sc->sc_tq, &sc->sc_intr_task);
488 	taskqueue_drain(sc->sc_tq, &sc->sc_tx_task);
489 	if_free(ifp);
490 	taskqueue_free(sc->sc_tq);
491 	device_delete_child(sc->sc_dev, sc->sc_miibus);
492 
493 	for (i = 0; i < CAS_NRXDESC; i++)
494 		if (sc->sc_rxdsoft[i].rxds_dmamap != NULL)
495 			bus_dmamap_sync(sc->sc_rdmatag,
496 			    sc->sc_rxdsoft[i].rxds_dmamap,
497 			    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
498 	for (i = 0; i < CAS_NRXDESC; i++)
499 		if (sc->sc_rxdsoft[i].rxds_paddr != 0)
500 			bus_dmamap_unload(sc->sc_rdmatag,
501 			    sc->sc_rxdsoft[i].rxds_dmamap);
502 	for (i = 0; i < CAS_NRXDESC; i++)
503 		if (sc->sc_rxdsoft[i].rxds_buf != NULL)
504 			bus_dmamem_free(sc->sc_rdmatag,
505 			    sc->sc_rxdsoft[i].rxds_buf,
506 			    sc->sc_rxdsoft[i].rxds_dmamap);
507 	for (i = 0; i < CAS_TXQUEUELEN; i++)
508 		if (sc->sc_txsoft[i].txs_dmamap != NULL)
509 			bus_dmamap_destroy(sc->sc_tdmatag,
510 			    sc->sc_txsoft[i].txs_dmamap);
511 	CAS_CDSYNC(sc, BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
512 	bus_dmamap_unload(sc->sc_cdmatag, sc->sc_cddmamap);
513 	bus_dmamem_free(sc->sc_cdmatag, sc->sc_control_data,
514 	    sc->sc_cddmamap);
515 	bus_dma_tag_destroy(sc->sc_cdmatag);
516 	bus_dma_tag_destroy(sc->sc_tdmatag);
517 	bus_dma_tag_destroy(sc->sc_rdmatag);
518 	bus_dma_tag_destroy(sc->sc_pdmatag);
519 }
520 
521 static void
522 cas_suspend(struct cas_softc *sc)
523 {
524 	struct ifnet *ifp = sc->sc_ifp;
525 
526 	CAS_LOCK(sc);
527 	cas_stop(ifp);
528 	CAS_UNLOCK(sc);
529 }
530 
531 static void
532 cas_resume(struct cas_softc *sc)
533 {
534 	struct ifnet *ifp = sc->sc_ifp;
535 
536 	CAS_LOCK(sc);
537 	/*
538 	 * On resume all registers have to be initialized again like
539 	 * after power-on.
540 	 */
541 	sc->sc_flags &= ~CAS_INITED;
542 	if (ifp->if_flags & IFF_UP)
543 		cas_init_locked(sc);
544 	CAS_UNLOCK(sc);
545 }
546 
547 static inline void
548 cas_rxcksum(struct mbuf *m, uint16_t cksum)
549 {
550 	struct ether_header *eh;
551 	struct ip *ip;
552 	struct udphdr *uh;
553 	uint16_t *opts;
554 	int32_t hlen, len, pktlen;
555 	uint32_t temp32;
556 
557 	pktlen = m->m_pkthdr.len;
558 	if (pktlen < sizeof(struct ether_header) + sizeof(struct ip))
559 		return;
560 	eh = mtod(m, struct ether_header *);
561 	if (eh->ether_type != htons(ETHERTYPE_IP))
562 		return;
563 	ip = (struct ip *)(eh + 1);
564 	if (ip->ip_v != IPVERSION)
565 		return;
566 
567 	hlen = ip->ip_hl << 2;
568 	pktlen -= sizeof(struct ether_header);
569 	if (hlen < sizeof(struct ip))
570 		return;
571 	if (ntohs(ip->ip_len) < hlen)
572 		return;
573 	if (ntohs(ip->ip_len) != pktlen)
574 		return;
575 	if (ip->ip_off & htons(IP_MF | IP_OFFMASK))
576 		return;	/* Cannot handle fragmented packet. */
577 
578 	switch (ip->ip_p) {
579 	case IPPROTO_TCP:
580 		if (pktlen < (hlen + sizeof(struct tcphdr)))
581 			return;
582 		break;
583 	case IPPROTO_UDP:
584 		if (pktlen < (hlen + sizeof(struct udphdr)))
585 			return;
586 		uh = (struct udphdr *)((uint8_t *)ip + hlen);
587 		if (uh->uh_sum == 0)
588 			return; /* no checksum */
589 		break;
590 	default:
591 		return;
592 	}
593 
594 	cksum = ~cksum;
595 	/* checksum fixup for IP options */
596 	len = hlen - sizeof(struct ip);
597 	if (len > 0) {
598 		opts = (uint16_t *)(ip + 1);
599 		for (; len > 0; len -= sizeof(uint16_t), opts++) {
600 			temp32 = cksum - *opts;
601 			temp32 = (temp32 >> 16) + (temp32 & 65535);
602 			cksum = temp32 & 65535;
603 		}
604 	}
605 	m->m_pkthdr.csum_flags |= CSUM_DATA_VALID;
606 	m->m_pkthdr.csum_data = cksum;
607 }
608 
609 static void
610 cas_cddma_callback(void *xsc, bus_dma_segment_t *segs, int nsegs, int error)
611 {
612 	struct cas_softc *sc = xsc;
613 
614 	if (error != 0)
615 		return;
616 	if (nsegs != 1)
617 		panic("%s: bad control buffer segment count", __func__);
618 	sc->sc_cddma = segs[0].ds_addr;
619 }
620 
621 static void
622 cas_rxdma_callback(void *xsc, bus_dma_segment_t *segs, int nsegs, int error)
623 {
624 	struct cas_softc *sc = xsc;
625 
626 	if (error != 0)
627 		return;
628 	if (nsegs != 1)
629 		panic("%s: bad RX buffer segment count", __func__);
630 	sc->sc_rxdsoft[sc->sc_rxdptr].rxds_paddr = segs[0].ds_addr;
631 }
632 
633 static void
634 cas_tick(void *arg)
635 {
636 	struct cas_softc *sc = arg;
637 	struct ifnet *ifp = sc->sc_ifp;
638 	uint32_t v;
639 
640 	CAS_LOCK_ASSERT(sc, MA_OWNED);
641 
642 	/*
643 	 * Unload collision and error counters.
644 	 */
645 	if_inc_counter(ifp, IFCOUNTER_COLLISIONS,
646 	    CAS_READ_4(sc, CAS_MAC_NORM_COLL_CNT) +
647 	    CAS_READ_4(sc, CAS_MAC_FIRST_COLL_CNT));
648 	v = CAS_READ_4(sc, CAS_MAC_EXCESS_COLL_CNT) +
649 	    CAS_READ_4(sc, CAS_MAC_LATE_COLL_CNT);
650 	if_inc_counter(ifp, IFCOUNTER_COLLISIONS, v);
651 	if_inc_counter(ifp, IFCOUNTER_OERRORS, v);
652 	if_inc_counter(ifp, IFCOUNTER_IERRORS,
653 	    CAS_READ_4(sc, CAS_MAC_RX_LEN_ERR_CNT) +
654 	    CAS_READ_4(sc, CAS_MAC_RX_ALIGN_ERR) +
655 	    CAS_READ_4(sc, CAS_MAC_RX_CRC_ERR_CNT) +
656 	    CAS_READ_4(sc, CAS_MAC_RX_CODE_VIOL));
657 
658 	/*
659 	 * Then clear the hardware counters.
660 	 */
661 	CAS_WRITE_4(sc, CAS_MAC_NORM_COLL_CNT, 0);
662 	CAS_WRITE_4(sc, CAS_MAC_FIRST_COLL_CNT, 0);
663 	CAS_WRITE_4(sc, CAS_MAC_EXCESS_COLL_CNT, 0);
664 	CAS_WRITE_4(sc, CAS_MAC_LATE_COLL_CNT, 0);
665 	CAS_WRITE_4(sc, CAS_MAC_RX_LEN_ERR_CNT, 0);
666 	CAS_WRITE_4(sc, CAS_MAC_RX_ALIGN_ERR, 0);
667 	CAS_WRITE_4(sc, CAS_MAC_RX_CRC_ERR_CNT, 0);
668 	CAS_WRITE_4(sc, CAS_MAC_RX_CODE_VIOL, 0);
669 
670 	mii_tick(sc->sc_mii);
671 
672 	if (sc->sc_txfree != CAS_MAXTXFREE)
673 		cas_tint(sc);
674 
675 	cas_watchdog(sc);
676 
677 	callout_reset(&sc->sc_tick_ch, hz, cas_tick, sc);
678 }
679 
680 static int
681 cas_bitwait(struct cas_softc *sc, bus_addr_t r, uint32_t clr, uint32_t set)
682 {
683 	int i;
684 	uint32_t reg;
685 
686 	for (i = CAS_TRIES; i--; DELAY(100)) {
687 		reg = CAS_READ_4(sc, r);
688 		if ((reg & clr) == 0 && (reg & set) == set)
689 			return (1);
690 	}
691 	return (0);
692 }
693 
694 static void
695 cas_reset(struct cas_softc *sc)
696 {
697 
698 #ifdef CAS_DEBUG
699 	CTR2(KTR_CAS, "%s: %s", device_get_name(sc->sc_dev), __func__);
700 #endif
701 	/* Disable all interrupts in order to avoid spurious ones. */
702 	CAS_WRITE_4(sc, CAS_INTMASK, 0xffffffff);
703 
704 	cas_reset_rx(sc);
705 	cas_reset_tx(sc);
706 
707 	/*
708 	 * Do a full reset modulo the result of the last auto-negotiation
709 	 * when using the SERDES.
710 	 */
711 	CAS_WRITE_4(sc, CAS_RESET, CAS_RESET_RX | CAS_RESET_TX |
712 	    ((sc->sc_flags & CAS_SERDES) != 0 ? CAS_RESET_PCS_DIS : 0));
713 	CAS_BARRIER(sc, CAS_RESET, 4,
714 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
715 	DELAY(3000);
716 	if (!cas_bitwait(sc, CAS_RESET, CAS_RESET_RX | CAS_RESET_TX, 0))
717 		device_printf(sc->sc_dev, "cannot reset device\n");
718 }
719 
720 static void
721 cas_stop(struct ifnet *ifp)
722 {
723 	struct cas_softc *sc = ifp->if_softc;
724 	struct cas_txsoft *txs;
725 
726 #ifdef CAS_DEBUG
727 	CTR2(KTR_CAS, "%s: %s", device_get_name(sc->sc_dev), __func__);
728 #endif
729 
730 	callout_stop(&sc->sc_tick_ch);
731 	callout_stop(&sc->sc_rx_ch);
732 
733 	/* Disable all interrupts in order to avoid spurious ones. */
734 	CAS_WRITE_4(sc, CAS_INTMASK, 0xffffffff);
735 
736 	cas_reset_tx(sc);
737 	cas_reset_rx(sc);
738 
739 	/*
740 	 * Release any queued transmit buffers.
741 	 */
742 	while ((txs = STAILQ_FIRST(&sc->sc_txdirtyq)) != NULL) {
743 		STAILQ_REMOVE_HEAD(&sc->sc_txdirtyq, txs_q);
744 		if (txs->txs_ndescs != 0) {
745 			bus_dmamap_sync(sc->sc_tdmatag, txs->txs_dmamap,
746 			    BUS_DMASYNC_POSTWRITE);
747 			bus_dmamap_unload(sc->sc_tdmatag, txs->txs_dmamap);
748 			if (txs->txs_mbuf != NULL) {
749 				m_freem(txs->txs_mbuf);
750 				txs->txs_mbuf = NULL;
751 			}
752 		}
753 		STAILQ_INSERT_TAIL(&sc->sc_txfreeq, txs, txs_q);
754 	}
755 
756 	/*
757 	 * Mark the interface down and cancel the watchdog timer.
758 	 */
759 	ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
760 	sc->sc_flags &= ~CAS_LINK;
761 	sc->sc_wdog_timer = 0;
762 }
763 
764 static int
765 cas_reset_rx(struct cas_softc *sc)
766 {
767 
768 	/*
769 	 * Resetting while DMA is in progress can cause a bus hang, so we
770 	 * disable DMA first.
771 	 */
772 	(void)cas_disable_rx(sc);
773 	CAS_WRITE_4(sc, CAS_RX_CONF, 0);
774 	CAS_BARRIER(sc, CAS_RX_CONF, 4,
775 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
776 	if (!cas_bitwait(sc, CAS_RX_CONF, CAS_RX_CONF_RXDMA_EN, 0))
777 		device_printf(sc->sc_dev, "cannot disable RX DMA\n");
778 
779 	/* Finally, reset the ERX. */
780 	CAS_WRITE_4(sc, CAS_RESET, CAS_RESET_RX |
781 	    ((sc->sc_flags & CAS_SERDES) != 0 ? CAS_RESET_PCS_DIS : 0));
782 	CAS_BARRIER(sc, CAS_RESET, 4,
783 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
784 	if (!cas_bitwait(sc, CAS_RESET, CAS_RESET_RX, 0)) {
785 		device_printf(sc->sc_dev, "cannot reset receiver\n");
786 		return (1);
787 	}
788 	return (0);
789 }
790 
791 static int
792 cas_reset_tx(struct cas_softc *sc)
793 {
794 
795 	/*
796 	 * Resetting while DMA is in progress can cause a bus hang, so we
797 	 * disable DMA first.
798 	 */
799 	(void)cas_disable_tx(sc);
800 	CAS_WRITE_4(sc, CAS_TX_CONF, 0);
801 	CAS_BARRIER(sc, CAS_TX_CONF, 4,
802 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
803 	if (!cas_bitwait(sc, CAS_TX_CONF, CAS_TX_CONF_TXDMA_EN, 0))
804 		device_printf(sc->sc_dev, "cannot disable TX DMA\n");
805 
806 	/* Finally, reset the ETX. */
807 	CAS_WRITE_4(sc, CAS_RESET, CAS_RESET_TX |
808 	    ((sc->sc_flags & CAS_SERDES) != 0 ? CAS_RESET_PCS_DIS : 0));
809 	CAS_BARRIER(sc, CAS_RESET, 4,
810 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
811 	if (!cas_bitwait(sc, CAS_RESET, CAS_RESET_TX, 0)) {
812 		device_printf(sc->sc_dev, "cannot reset transmitter\n");
813 		return (1);
814 	}
815 	return (0);
816 }
817 
818 static int
819 cas_disable_rx(struct cas_softc *sc)
820 {
821 
822 	CAS_WRITE_4(sc, CAS_MAC_RX_CONF,
823 	    CAS_READ_4(sc, CAS_MAC_RX_CONF) & ~CAS_MAC_RX_CONF_EN);
824 	CAS_BARRIER(sc, CAS_MAC_RX_CONF, 4,
825 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
826 	if (cas_bitwait(sc, CAS_MAC_RX_CONF, CAS_MAC_RX_CONF_EN, 0))
827 		return (1);
828 	if (bootverbose)
829 		device_printf(sc->sc_dev, "cannot disable RX MAC\n");
830 	return (0);
831 }
832 
833 static int
834 cas_disable_tx(struct cas_softc *sc)
835 {
836 
837 	CAS_WRITE_4(sc, CAS_MAC_TX_CONF,
838 	    CAS_READ_4(sc, CAS_MAC_TX_CONF) & ~CAS_MAC_TX_CONF_EN);
839 	CAS_BARRIER(sc, CAS_MAC_TX_CONF, 4,
840 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
841 	if (cas_bitwait(sc, CAS_MAC_TX_CONF, CAS_MAC_TX_CONF_EN, 0))
842 		return (1);
843 	if (bootverbose)
844 		device_printf(sc->sc_dev, "cannot disable TX MAC\n");
845 	return (0);
846 }
847 
848 static inline void
849 cas_rxcompinit(struct cas_rx_comp *rxcomp)
850 {
851 
852 	rxcomp->crc_word1 = 0;
853 	rxcomp->crc_word2 = 0;
854 	rxcomp->crc_word3 =
855 	    htole64(CAS_SET(ETHER_HDR_LEN + sizeof(struct ip), CAS_RC3_CSO));
856 	rxcomp->crc_word4 = htole64(CAS_RC4_ZERO);
857 }
858 
859 static void
860 cas_meminit(struct cas_softc *sc)
861 {
862 	int i;
863 
864 	CAS_LOCK_ASSERT(sc, MA_OWNED);
865 
866 	/*
867 	 * Initialize the transmit descriptor ring.
868 	 */
869 	for (i = 0; i < CAS_NTXDESC; i++) {
870 		sc->sc_txdescs[i].cd_flags = 0;
871 		sc->sc_txdescs[i].cd_buf_ptr = 0;
872 	}
873 	sc->sc_txfree = CAS_MAXTXFREE;
874 	sc->sc_txnext = 0;
875 	sc->sc_txwin = 0;
876 
877 	/*
878 	 * Initialize the receive completion ring.
879 	 */
880 	for (i = 0; i < CAS_NRXCOMP; i++)
881 		cas_rxcompinit(&sc->sc_rxcomps[i]);
882 	sc->sc_rxcptr = 0;
883 
884 	/*
885 	 * Initialize the first receive descriptor ring.  We leave
886 	 * the second one zeroed as we don't actually use it.
887 	 */
888 	for (i = 0; i < CAS_NRXDESC; i++)
889 		CAS_INIT_RXDESC(sc, i, i);
890 	sc->sc_rxdptr = 0;
891 
892 	CAS_CDSYNC(sc, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
893 }
894 
895 static u_int
896 cas_descsize(u_int sz)
897 {
898 
899 	switch (sz) {
900 	case 32:
901 		return (CAS_DESC_32);
902 	case 64:
903 		return (CAS_DESC_64);
904 	case 128:
905 		return (CAS_DESC_128);
906 	case 256:
907 		return (CAS_DESC_256);
908 	case 512:
909 		return (CAS_DESC_512);
910 	case 1024:
911 		return (CAS_DESC_1K);
912 	case 2048:
913 		return (CAS_DESC_2K);
914 	case 4096:
915 		return (CAS_DESC_4K);
916 	case 8192:
917 		return (CAS_DESC_8K);
918 	default:
919 		printf("%s: invalid descriptor ring size %d\n", __func__, sz);
920 		return (CAS_DESC_32);
921 	}
922 }
923 
924 static u_int
925 cas_rxcompsize(u_int sz)
926 {
927 
928 	switch (sz) {
929 	case 128:
930 		return (CAS_RX_CONF_COMP_128);
931 	case 256:
932 		return (CAS_RX_CONF_COMP_256);
933 	case 512:
934 		return (CAS_RX_CONF_COMP_512);
935 	case 1024:
936 		return (CAS_RX_CONF_COMP_1K);
937 	case 2048:
938 		return (CAS_RX_CONF_COMP_2K);
939 	case 4096:
940 		return (CAS_RX_CONF_COMP_4K);
941 	case 8192:
942 		return (CAS_RX_CONF_COMP_8K);
943 	case 16384:
944 		return (CAS_RX_CONF_COMP_16K);
945 	case 32768:
946 		return (CAS_RX_CONF_COMP_32K);
947 	default:
948 		printf("%s: invalid dcompletion ring size %d\n", __func__, sz);
949 		return (CAS_RX_CONF_COMP_128);
950 	}
951 }
952 
953 static void
954 cas_init(void *xsc)
955 {
956 	struct cas_softc *sc = xsc;
957 
958 	CAS_LOCK(sc);
959 	cas_init_locked(sc);
960 	CAS_UNLOCK(sc);
961 }
962 
963 /*
964  * Initialization of interface; set up initialization block
965  * and transmit/receive descriptor rings.
966  */
967 static void
968 cas_init_locked(struct cas_softc *sc)
969 {
970 	struct ifnet *ifp = sc->sc_ifp;
971 	uint32_t v;
972 
973 	CAS_LOCK_ASSERT(sc, MA_OWNED);
974 
975 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
976 		return;
977 
978 #ifdef CAS_DEBUG
979 	CTR2(KTR_CAS, "%s: %s: calling stop", device_get_name(sc->sc_dev),
980 	    __func__);
981 #endif
982 	/*
983 	 * Initialization sequence.  The numbered steps below correspond
984 	 * to the sequence outlined in section 6.3.5.1 in the Ethernet
985 	 * Channel Engine manual (part of the PCIO manual).
986 	 * See also the STP2002-STQ document from Sun Microsystems.
987 	 */
988 
989 	/* step 1 & 2.  Reset the Ethernet Channel. */
990 	cas_stop(ifp);
991 	cas_reset(sc);
992 #ifdef CAS_DEBUG
993 	CTR2(KTR_CAS, "%s: %s: restarting", device_get_name(sc->sc_dev),
994 	    __func__);
995 #endif
996 
997 	if ((sc->sc_flags & CAS_SERDES) == 0)
998 		/* Re-initialize the MIF. */
999 		cas_mifinit(sc);
1000 
1001 	/* step 3.  Setup data structures in host memory. */
1002 	cas_meminit(sc);
1003 
1004 	/* step 4.  TX MAC registers & counters */
1005 	cas_init_regs(sc);
1006 
1007 	/* step 5.  RX MAC registers & counters */
1008 
1009 	/* step 6 & 7.  Program Ring Base Addresses. */
1010 	CAS_WRITE_4(sc, CAS_TX_DESC3_BASE_HI,
1011 	    (((uint64_t)CAS_CDTXDADDR(sc, 0)) >> 32));
1012 	CAS_WRITE_4(sc, CAS_TX_DESC3_BASE_LO,
1013 	    CAS_CDTXDADDR(sc, 0) & 0xffffffff);
1014 
1015 	CAS_WRITE_4(sc, CAS_RX_COMP_BASE_HI,
1016 	    (((uint64_t)CAS_CDRXCADDR(sc, 0)) >> 32));
1017 	CAS_WRITE_4(sc, CAS_RX_COMP_BASE_LO,
1018 	    CAS_CDRXCADDR(sc, 0) & 0xffffffff);
1019 
1020 	CAS_WRITE_4(sc, CAS_RX_DESC_BASE_HI,
1021 	    (((uint64_t)CAS_CDRXDADDR(sc, 0)) >> 32));
1022 	CAS_WRITE_4(sc, CAS_RX_DESC_BASE_LO,
1023 	    CAS_CDRXDADDR(sc, 0) & 0xffffffff);
1024 
1025 	if ((sc->sc_flags & CAS_REG_PLUS) != 0) {
1026 		CAS_WRITE_4(sc, CAS_RX_DESC2_BASE_HI,
1027 		    (((uint64_t)CAS_CDRXD2ADDR(sc, 0)) >> 32));
1028 		CAS_WRITE_4(sc, CAS_RX_DESC2_BASE_LO,
1029 		    CAS_CDRXD2ADDR(sc, 0) & 0xffffffff);
1030 	}
1031 
1032 #ifdef CAS_DEBUG
1033 	CTR5(KTR_CAS,
1034 	    "loading TXDR %lx, RXCR %lx, RXDR %lx, RXD2R %lx, cddma %lx",
1035 	    CAS_CDTXDADDR(sc, 0), CAS_CDRXCADDR(sc, 0), CAS_CDRXDADDR(sc, 0),
1036 	    CAS_CDRXD2ADDR(sc, 0), sc->sc_cddma);
1037 #endif
1038 
1039 	/* step 8.  Global Configuration & Interrupt Masks */
1040 
1041 	/* Disable weighted round robin. */
1042 	CAS_WRITE_4(sc, CAS_CAW, CAS_CAW_RR_DIS);
1043 
1044 	/*
1045 	 * Enable infinite bursts for revisions without PCI issues if
1046 	 * applicable.  Doing so greatly improves the TX performance.
1047 	 */
1048 	CAS_WRITE_4(sc, CAS_INF_BURST,
1049 	    (sc->sc_flags & CAS_TABORT) == 0 ? CAS_INF_BURST_EN :
1050 	    0);
1051 
1052 	/* Set up interrupts. */
1053 	CAS_WRITE_4(sc, CAS_INTMASK,
1054 	    ~(CAS_INTR_TX_INT_ME | CAS_INTR_TX_TAG_ERR |
1055 	    CAS_INTR_RX_DONE | CAS_INTR_RX_BUF_NA | CAS_INTR_RX_TAG_ERR |
1056 	    CAS_INTR_RX_COMP_FULL | CAS_INTR_RX_BUF_AEMPTY |
1057 	    CAS_INTR_RX_COMP_AFULL | CAS_INTR_RX_LEN_MMATCH |
1058 	    CAS_INTR_PCI_ERROR_INT
1059 #ifdef CAS_DEBUG
1060 	    | CAS_INTR_PCS_INT | CAS_INTR_MIF
1061 #endif
1062 	    ));
1063 	/* Don't clear top level interrupts when CAS_STATUS_ALIAS is read. */
1064 	CAS_WRITE_4(sc, CAS_CLEAR_ALIAS, 0);
1065 	CAS_WRITE_4(sc, CAS_MAC_RX_MASK, ~CAS_MAC_RX_OVERFLOW);
1066 	CAS_WRITE_4(sc, CAS_MAC_TX_MASK,
1067 	    ~(CAS_MAC_TX_UNDERRUN | CAS_MAC_TX_MAX_PKT_ERR));
1068 #ifdef CAS_DEBUG
1069 	CAS_WRITE_4(sc, CAS_MAC_CTRL_MASK,
1070 	    ~(CAS_MAC_CTRL_PAUSE_RCVD | CAS_MAC_CTRL_PAUSE |
1071 	    CAS_MAC_CTRL_NON_PAUSE));
1072 #else
1073 	CAS_WRITE_4(sc, CAS_MAC_CTRL_MASK,
1074 	    CAS_MAC_CTRL_PAUSE_RCVD | CAS_MAC_CTRL_PAUSE |
1075 	    CAS_MAC_CTRL_NON_PAUSE);
1076 #endif
1077 
1078 	/* Enable PCI error interrupts. */
1079 	CAS_WRITE_4(sc, CAS_ERROR_MASK,
1080 	    ~(CAS_ERROR_DTRTO | CAS_ERROR_OTHER | CAS_ERROR_DMAW_ZERO |
1081 	    CAS_ERROR_DMAR_ZERO | CAS_ERROR_RTRTO));
1082 
1083 	/* Enable PCI error interrupts in BIM configuration. */
1084 	CAS_WRITE_4(sc, CAS_BIM_CONF,
1085 	    CAS_BIM_CONF_DPAR_EN | CAS_BIM_CONF_RMA_EN | CAS_BIM_CONF_RTA_EN);
1086 
1087 	/*
1088 	 * step 9.  ETX Configuration: encode receive descriptor ring size,
1089 	 * enable DMA and disable pre-interrupt writeback completion.
1090 	 */
1091 	v = cas_descsize(CAS_NTXDESC) << CAS_TX_CONF_DESC3_SHFT;
1092 	CAS_WRITE_4(sc, CAS_TX_CONF, v | CAS_TX_CONF_TXDMA_EN |
1093 	    CAS_TX_CONF_RDPP_DIS | CAS_TX_CONF_PICWB_DIS);
1094 
1095 	/* step 10.  ERX Configuration */
1096 
1097 	/*
1098 	 * Encode receive completion and descriptor ring sizes, set the
1099 	 * swivel offset.
1100 	 */
1101 	v = cas_rxcompsize(CAS_NRXCOMP) << CAS_RX_CONF_COMP_SHFT;
1102 	v |= cas_descsize(CAS_NRXDESC) << CAS_RX_CONF_DESC_SHFT;
1103 	if ((sc->sc_flags & CAS_REG_PLUS) != 0)
1104 		v |= cas_descsize(CAS_NRXDESC2) << CAS_RX_CONF_DESC2_SHFT;
1105 	CAS_WRITE_4(sc, CAS_RX_CONF,
1106 	    v | (ETHER_ALIGN << CAS_RX_CONF_SOFF_SHFT));
1107 
1108 	/* Set the PAUSE thresholds.  We use the maximum OFF threshold. */
1109 	CAS_WRITE_4(sc, CAS_RX_PTHRS,
1110 	    (111 << CAS_RX_PTHRS_XOFF_SHFT) | (15 << CAS_RX_PTHRS_XON_SHFT));
1111 
1112 	/* RX blanking */
1113 	CAS_WRITE_4(sc, CAS_RX_BLANK,
1114 	    (15 << CAS_RX_BLANK_TIME_SHFT) | (5 << CAS_RX_BLANK_PKTS_SHFT));
1115 
1116 	/* Set RX_COMP_AFULL threshold to half of the RX completions. */
1117 	CAS_WRITE_4(sc, CAS_RX_AEMPTY_THRS,
1118 	    (CAS_NRXCOMP / 2) << CAS_RX_AEMPTY_COMP_SHFT);
1119 
1120 	/* Initialize the RX page size register as appropriate for 8k. */
1121 	CAS_WRITE_4(sc, CAS_RX_PSZ,
1122 	    (CAS_RX_PSZ_8K << CAS_RX_PSZ_SHFT) |
1123 	    (4 << CAS_RX_PSZ_MB_CNT_SHFT) |
1124 	    (CAS_RX_PSZ_MB_STRD_2K << CAS_RX_PSZ_MB_STRD_SHFT) |
1125 	    (CAS_RX_PSZ_MB_OFF_64 << CAS_RX_PSZ_MB_OFF_SHFT));
1126 
1127 	/* Disable RX random early detection. */
1128 	CAS_WRITE_4(sc,	CAS_RX_RED, 0);
1129 
1130 	/* Zero the RX reassembly DMA table. */
1131 	for (v = 0; v <= CAS_RX_REAS_DMA_ADDR_LC; v++) {
1132 		CAS_WRITE_4(sc,	CAS_RX_REAS_DMA_ADDR, v);
1133 		CAS_WRITE_4(sc,	CAS_RX_REAS_DMA_DATA_LO, 0);
1134 		CAS_WRITE_4(sc,	CAS_RX_REAS_DMA_DATA_MD, 0);
1135 		CAS_WRITE_4(sc,	CAS_RX_REAS_DMA_DATA_HI, 0);
1136 	}
1137 
1138 	/* Ensure the RX control FIFO and RX IPP FIFO addresses are zero. */
1139 	CAS_WRITE_4(sc, CAS_RX_CTRL_FIFO, 0);
1140 	CAS_WRITE_4(sc, CAS_RX_IPP_ADDR, 0);
1141 
1142 	/* Finally, enable RX DMA. */
1143 	CAS_WRITE_4(sc, CAS_RX_CONF,
1144 	    CAS_READ_4(sc, CAS_RX_CONF) | CAS_RX_CONF_RXDMA_EN);
1145 
1146 	/* step 11.  Configure Media. */
1147 
1148 	/* step 12.  RX_MAC Configuration Register */
1149 	v = CAS_READ_4(sc, CAS_MAC_RX_CONF);
1150 	v &= ~(CAS_MAC_RX_CONF_STRPPAD | CAS_MAC_RX_CONF_EN);
1151 	v |= CAS_MAC_RX_CONF_STRPFCS;
1152 	sc->sc_mac_rxcfg = v;
1153 	/*
1154 	 * Clear the RX filter and reprogram it.  This will also set the
1155 	 * current RX MAC configuration and enable it.
1156 	 */
1157 	cas_setladrf(sc);
1158 
1159 	/* step 13.  TX_MAC Configuration Register */
1160 	v = CAS_READ_4(sc, CAS_MAC_TX_CONF);
1161 	v |= CAS_MAC_TX_CONF_EN;
1162 	(void)cas_disable_tx(sc);
1163 	CAS_WRITE_4(sc, CAS_MAC_TX_CONF, v);
1164 
1165 	/* step 14.  Issue Transmit Pending command. */
1166 
1167 	/* step 15.  Give the receiver a swift kick. */
1168 	CAS_WRITE_4(sc, CAS_RX_KICK, CAS_NRXDESC - 4);
1169 	CAS_WRITE_4(sc, CAS_RX_COMP_TAIL, 0);
1170 	if ((sc->sc_flags & CAS_REG_PLUS) != 0)
1171 		CAS_WRITE_4(sc, CAS_RX_KICK2, CAS_NRXDESC2 - 4);
1172 
1173 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1174 	ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1175 
1176 	mii_mediachg(sc->sc_mii);
1177 
1178 	/* Start the one second timer. */
1179 	sc->sc_wdog_timer = 0;
1180 	callout_reset(&sc->sc_tick_ch, hz, cas_tick, sc);
1181 }
1182 
1183 static int
1184 cas_load_txmbuf(struct cas_softc *sc, struct mbuf **m_head)
1185 {
1186 	bus_dma_segment_t txsegs[CAS_NTXSEGS];
1187 	struct cas_txsoft *txs;
1188 	struct ip *ip;
1189 	struct mbuf *m;
1190 	uint64_t cflags;
1191 	int error, nexttx, nsegs, offset, seg;
1192 
1193 	CAS_LOCK_ASSERT(sc, MA_OWNED);
1194 
1195 	/* Get a work queue entry. */
1196 	if ((txs = STAILQ_FIRST(&sc->sc_txfreeq)) == NULL) {
1197 		/* Ran out of descriptors. */
1198 		return (ENOBUFS);
1199 	}
1200 
1201 	cflags = 0;
1202 	if (((*m_head)->m_pkthdr.csum_flags & CAS_CSUM_FEATURES) != 0) {
1203 		if (M_WRITABLE(*m_head) == 0) {
1204 			m = m_dup(*m_head, M_NOWAIT);
1205 			m_freem(*m_head);
1206 			*m_head = m;
1207 			if (m == NULL)
1208 				return (ENOBUFS);
1209 		}
1210 		offset = sizeof(struct ether_header);
1211 		m = m_pullup(*m_head, offset + sizeof(struct ip));
1212 		if (m == NULL) {
1213 			*m_head = NULL;
1214 			return (ENOBUFS);
1215 		}
1216 		ip = (struct ip *)(mtod(m, caddr_t) + offset);
1217 		offset += (ip->ip_hl << 2);
1218 		cflags = (offset << CAS_TD_CKSUM_START_SHFT) |
1219 		    ((offset + m->m_pkthdr.csum_data) <<
1220 		    CAS_TD_CKSUM_STUFF_SHFT) | CAS_TD_CKSUM_EN;
1221 		*m_head = m;
1222 	}
1223 
1224 	error = bus_dmamap_load_mbuf_sg(sc->sc_tdmatag, txs->txs_dmamap,
1225 	    *m_head, txsegs, &nsegs, BUS_DMA_NOWAIT);
1226 	if (error == EFBIG) {
1227 		m = m_collapse(*m_head, M_NOWAIT, CAS_NTXSEGS);
1228 		if (m == NULL) {
1229 			m_freem(*m_head);
1230 			*m_head = NULL;
1231 			return (ENOBUFS);
1232 		}
1233 		*m_head = m;
1234 		error = bus_dmamap_load_mbuf_sg(sc->sc_tdmatag,
1235 		    txs->txs_dmamap, *m_head, txsegs, &nsegs,
1236 		    BUS_DMA_NOWAIT);
1237 		if (error != 0) {
1238 			m_freem(*m_head);
1239 			*m_head = NULL;
1240 			return (error);
1241 		}
1242 	} else if (error != 0)
1243 		return (error);
1244 	/* If nsegs is wrong then the stack is corrupt. */
1245 	KASSERT(nsegs <= CAS_NTXSEGS,
1246 	    ("%s: too many DMA segments (%d)", __func__, nsegs));
1247 	if (nsegs == 0) {
1248 		m_freem(*m_head);
1249 		*m_head = NULL;
1250 		return (EIO);
1251 	}
1252 
1253 	/*
1254 	 * Ensure we have enough descriptors free to describe
1255 	 * the packet.  Note, we always reserve one descriptor
1256 	 * at the end of the ring as a termination point, in
1257 	 * order to prevent wrap-around.
1258 	 */
1259 	if (nsegs > sc->sc_txfree - 1) {
1260 		txs->txs_ndescs = 0;
1261 		bus_dmamap_unload(sc->sc_tdmatag, txs->txs_dmamap);
1262 		return (ENOBUFS);
1263 	}
1264 
1265 	txs->txs_ndescs = nsegs;
1266 	txs->txs_firstdesc = sc->sc_txnext;
1267 	nexttx = txs->txs_firstdesc;
1268 	for (seg = 0; seg < nsegs; seg++, nexttx = CAS_NEXTTX(nexttx)) {
1269 #ifdef CAS_DEBUG
1270 		CTR6(KTR_CAS,
1271 		    "%s: mapping seg %d (txd %d), len %lx, addr %#lx (%#lx)",
1272 		    __func__, seg, nexttx, txsegs[seg].ds_len,
1273 		    txsegs[seg].ds_addr, htole64(txsegs[seg].ds_addr));
1274 #endif
1275 		sc->sc_txdescs[nexttx].cd_buf_ptr =
1276 		    htole64(txsegs[seg].ds_addr);
1277 		KASSERT(txsegs[seg].ds_len <
1278 		    CAS_TD_BUF_LEN_MASK >> CAS_TD_BUF_LEN_SHFT,
1279 		    ("%s: segment size too large!", __func__));
1280 		sc->sc_txdescs[nexttx].cd_flags =
1281 		    htole64(txsegs[seg].ds_len << CAS_TD_BUF_LEN_SHFT);
1282 		txs->txs_lastdesc = nexttx;
1283 	}
1284 
1285 	/* Set EOF on the last descriptor. */
1286 #ifdef CAS_DEBUG
1287 	CTR3(KTR_CAS, "%s: end of frame at segment %d, TX %d",
1288 	    __func__, seg, nexttx);
1289 #endif
1290 	sc->sc_txdescs[txs->txs_lastdesc].cd_flags |=
1291 	    htole64(CAS_TD_END_OF_FRAME);
1292 
1293 	/* Lastly set SOF on the first descriptor. */
1294 #ifdef CAS_DEBUG
1295 	CTR3(KTR_CAS, "%s: start of frame at segment %d, TX %d",
1296 	    __func__, seg, nexttx);
1297 #endif
1298 	if (sc->sc_txwin += nsegs > CAS_MAXTXFREE * 2 / 3) {
1299 		sc->sc_txwin = 0;
1300 		sc->sc_txdescs[txs->txs_firstdesc].cd_flags |=
1301 		    htole64(cflags | CAS_TD_START_OF_FRAME | CAS_TD_INT_ME);
1302 	} else
1303 		sc->sc_txdescs[txs->txs_firstdesc].cd_flags |=
1304 		    htole64(cflags | CAS_TD_START_OF_FRAME);
1305 
1306 	/* Sync the DMA map. */
1307 	bus_dmamap_sync(sc->sc_tdmatag, txs->txs_dmamap,
1308 	    BUS_DMASYNC_PREWRITE);
1309 
1310 #ifdef CAS_DEBUG
1311 	CTR4(KTR_CAS, "%s: setting firstdesc=%d, lastdesc=%d, ndescs=%d",
1312 	    __func__, txs->txs_firstdesc, txs->txs_lastdesc,
1313 	    txs->txs_ndescs);
1314 #endif
1315 	STAILQ_REMOVE_HEAD(&sc->sc_txfreeq, txs_q);
1316 	STAILQ_INSERT_TAIL(&sc->sc_txdirtyq, txs, txs_q);
1317 	txs->txs_mbuf = *m_head;
1318 
1319 	sc->sc_txnext = CAS_NEXTTX(txs->txs_lastdesc);
1320 	sc->sc_txfree -= txs->txs_ndescs;
1321 
1322 	return (0);
1323 }
1324 
1325 static void
1326 cas_init_regs(struct cas_softc *sc)
1327 {
1328 	int i;
1329 	const u_char *laddr = IF_LLADDR(sc->sc_ifp);
1330 
1331 	CAS_LOCK_ASSERT(sc, MA_OWNED);
1332 
1333 	/* These registers are not cleared on reset. */
1334 	if ((sc->sc_flags & CAS_INITED) == 0) {
1335 		/* magic values */
1336 		CAS_WRITE_4(sc, CAS_MAC_IPG0, 0);
1337 		CAS_WRITE_4(sc, CAS_MAC_IPG1, 8);
1338 		CAS_WRITE_4(sc, CAS_MAC_IPG2, 4);
1339 
1340 		/* min frame length */
1341 		CAS_WRITE_4(sc, CAS_MAC_MIN_FRAME, ETHER_MIN_LEN);
1342 		/* max frame length and max burst size */
1343 		CAS_WRITE_4(sc, CAS_MAC_MAX_BF,
1344 		    ((ETHER_MAX_LEN_JUMBO + ETHER_VLAN_ENCAP_LEN) <<
1345 		    CAS_MAC_MAX_BF_FRM_SHFT) |
1346 		    (0x2000 << CAS_MAC_MAX_BF_BST_SHFT));
1347 
1348 		/* more magic values */
1349 		CAS_WRITE_4(sc, CAS_MAC_PREAMBLE_LEN, 0x7);
1350 		CAS_WRITE_4(sc, CAS_MAC_JAM_SIZE, 0x4);
1351 		CAS_WRITE_4(sc, CAS_MAC_ATTEMPT_LIMIT, 0x10);
1352 		CAS_WRITE_4(sc, CAS_MAC_CTRL_TYPE, 0x8808);
1353 
1354 		/* random number seed */
1355 		CAS_WRITE_4(sc, CAS_MAC_RANDOM_SEED,
1356 		    ((laddr[5] << 8) | laddr[4]) & 0x3ff);
1357 
1358 		/* secondary MAC addresses: 0:0:0:0:0:0 */
1359 		for (i = CAS_MAC_ADDR3; i <= CAS_MAC_ADDR41;
1360 		    i += CAS_MAC_ADDR4 - CAS_MAC_ADDR3)
1361 			CAS_WRITE_4(sc, i, 0);
1362 
1363 		/* MAC control address: 01:80:c2:00:00:01 */
1364 		CAS_WRITE_4(sc, CAS_MAC_ADDR42, 0x0001);
1365 		CAS_WRITE_4(sc, CAS_MAC_ADDR43, 0xc200);
1366 		CAS_WRITE_4(sc, CAS_MAC_ADDR44, 0x0180);
1367 
1368 		/* MAC filter address: 0:0:0:0:0:0 */
1369 		CAS_WRITE_4(sc, CAS_MAC_AFILTER0, 0);
1370 		CAS_WRITE_4(sc, CAS_MAC_AFILTER1, 0);
1371 		CAS_WRITE_4(sc, CAS_MAC_AFILTER2, 0);
1372 		CAS_WRITE_4(sc, CAS_MAC_AFILTER_MASK1_2, 0);
1373 		CAS_WRITE_4(sc, CAS_MAC_AFILTER_MASK0, 0);
1374 
1375 		/* Zero the hash table. */
1376 		for (i = CAS_MAC_HASH0; i <= CAS_MAC_HASH15;
1377 		    i += CAS_MAC_HASH1 - CAS_MAC_HASH0)
1378 			CAS_WRITE_4(sc, i, 0);
1379 
1380 		sc->sc_flags |= CAS_INITED;
1381 	}
1382 
1383 	/* Counters need to be zeroed. */
1384 	CAS_WRITE_4(sc, CAS_MAC_NORM_COLL_CNT, 0);
1385 	CAS_WRITE_4(sc, CAS_MAC_FIRST_COLL_CNT, 0);
1386 	CAS_WRITE_4(sc, CAS_MAC_EXCESS_COLL_CNT, 0);
1387 	CAS_WRITE_4(sc, CAS_MAC_LATE_COLL_CNT, 0);
1388 	CAS_WRITE_4(sc, CAS_MAC_DEFER_TMR_CNT, 0);
1389 	CAS_WRITE_4(sc, CAS_MAC_PEAK_ATTEMPTS, 0);
1390 	CAS_WRITE_4(sc, CAS_MAC_RX_FRAME_COUNT, 0);
1391 	CAS_WRITE_4(sc, CAS_MAC_RX_LEN_ERR_CNT, 0);
1392 	CAS_WRITE_4(sc, CAS_MAC_RX_ALIGN_ERR, 0);
1393 	CAS_WRITE_4(sc, CAS_MAC_RX_CRC_ERR_CNT, 0);
1394 	CAS_WRITE_4(sc, CAS_MAC_RX_CODE_VIOL, 0);
1395 
1396 	/* Set XOFF PAUSE time. */
1397 	CAS_WRITE_4(sc, CAS_MAC_SPC, 0x1BF0 << CAS_MAC_SPC_TIME_SHFT);
1398 
1399 	/* Set the station address. */
1400 	CAS_WRITE_4(sc, CAS_MAC_ADDR0, (laddr[4] << 8) | laddr[5]);
1401 	CAS_WRITE_4(sc, CAS_MAC_ADDR1, (laddr[2] << 8) | laddr[3]);
1402 	CAS_WRITE_4(sc, CAS_MAC_ADDR2, (laddr[0] << 8) | laddr[1]);
1403 
1404 	/* Enable MII outputs. */
1405 	CAS_WRITE_4(sc, CAS_MAC_XIF_CONF, CAS_MAC_XIF_CONF_TX_OE);
1406 }
1407 
1408 static void
1409 cas_tx_task(void *arg, int pending __unused)
1410 {
1411 	struct ifnet *ifp;
1412 
1413 	ifp = (struct ifnet *)arg;
1414 	cas_start(ifp);
1415 }
1416 
1417 static inline void
1418 cas_txkick(struct cas_softc *sc)
1419 {
1420 
1421 	/*
1422 	 * Update the TX kick register.  This register has to point to the
1423 	 * descriptor after the last valid one and for optimum performance
1424 	 * should be incremented in multiples of 4 (the DMA engine fetches/
1425 	 * updates descriptors in batches of 4).
1426 	 */
1427 #ifdef CAS_DEBUG
1428 	CTR3(KTR_CAS, "%s: %s: kicking TX %d",
1429 	    device_get_name(sc->sc_dev), __func__, sc->sc_txnext);
1430 #endif
1431 	CAS_CDSYNC(sc, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1432 	CAS_WRITE_4(sc, CAS_TX_KICK3, sc->sc_txnext);
1433 }
1434 
1435 static void
1436 cas_start(struct ifnet *ifp)
1437 {
1438 	struct cas_softc *sc = ifp->if_softc;
1439 	struct mbuf *m;
1440 	int kicked, ntx;
1441 
1442 	CAS_LOCK(sc);
1443 
1444 	if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
1445 	    IFF_DRV_RUNNING || (sc->sc_flags & CAS_LINK) == 0) {
1446 		CAS_UNLOCK(sc);
1447 		return;
1448 	}
1449 
1450 	if (sc->sc_txfree < CAS_MAXTXFREE / 4)
1451 		cas_tint(sc);
1452 
1453 #ifdef CAS_DEBUG
1454 	CTR4(KTR_CAS, "%s: %s: txfree %d, txnext %d",
1455 	    device_get_name(sc->sc_dev), __func__, sc->sc_txfree,
1456 	    sc->sc_txnext);
1457 #endif
1458 	ntx = 0;
1459 	kicked = 0;
1460 	for (; !IFQ_DRV_IS_EMPTY(&ifp->if_snd) && sc->sc_txfree > 1;) {
1461 		IFQ_DRV_DEQUEUE(&ifp->if_snd, m);
1462 		if (m == NULL)
1463 			break;
1464 		if (cas_load_txmbuf(sc, &m) != 0) {
1465 			if (m == NULL)
1466 				break;
1467 			ifp->if_drv_flags |= IFF_DRV_OACTIVE;
1468 			IFQ_DRV_PREPEND(&ifp->if_snd, m);
1469 			break;
1470 		}
1471 		if ((sc->sc_txnext % 4) == 0) {
1472 			cas_txkick(sc);
1473 			kicked = 1;
1474 		} else
1475 			kicked = 0;
1476 		ntx++;
1477 		BPF_MTAP(ifp, m);
1478 	}
1479 
1480 	if (ntx > 0) {
1481 		if (kicked == 0)
1482 			cas_txkick(sc);
1483 #ifdef CAS_DEBUG
1484 		CTR2(KTR_CAS, "%s: packets enqueued, OWN on %d",
1485 		    device_get_name(sc->sc_dev), sc->sc_txnext);
1486 #endif
1487 
1488 		/* Set a watchdog timer in case the chip flakes out. */
1489 		sc->sc_wdog_timer = 5;
1490 #ifdef CAS_DEBUG
1491 		CTR3(KTR_CAS, "%s: %s: watchdog %d",
1492 		    device_get_name(sc->sc_dev), __func__,
1493 		    sc->sc_wdog_timer);
1494 #endif
1495 	}
1496 
1497 	CAS_UNLOCK(sc);
1498 }
1499 
1500 static void
1501 cas_tint(struct cas_softc *sc)
1502 {
1503 	struct ifnet *ifp = sc->sc_ifp;
1504 	struct cas_txsoft *txs;
1505 	int progress;
1506 	uint32_t txlast;
1507 #ifdef CAS_DEBUG
1508 	int i;
1509 
1510 	CAS_LOCK_ASSERT(sc, MA_OWNED);
1511 
1512 	CTR2(KTR_CAS, "%s: %s", device_get_name(sc->sc_dev), __func__);
1513 #endif
1514 
1515 	/*
1516 	 * Go through our TX list and free mbufs for those
1517 	 * frames that have been transmitted.
1518 	 */
1519 	progress = 0;
1520 	CAS_CDSYNC(sc, BUS_DMASYNC_POSTREAD);
1521 	while ((txs = STAILQ_FIRST(&sc->sc_txdirtyq)) != NULL) {
1522 #ifdef CAS_DEBUG
1523 		if ((ifp->if_flags & IFF_DEBUG) != 0) {
1524 			printf("    txsoft %p transmit chain:\n", txs);
1525 			for (i = txs->txs_firstdesc;; i = CAS_NEXTTX(i)) {
1526 				printf("descriptor %d: ", i);
1527 				printf("cd_flags: 0x%016llx\t",
1528 				    (long long)le64toh(
1529 				    sc->sc_txdescs[i].cd_flags));
1530 				printf("cd_buf_ptr: 0x%016llx\n",
1531 				    (long long)le64toh(
1532 				    sc->sc_txdescs[i].cd_buf_ptr));
1533 				if (i == txs->txs_lastdesc)
1534 					break;
1535 			}
1536 		}
1537 #endif
1538 
1539 		/*
1540 		 * In theory, we could harvest some descriptors before
1541 		 * the ring is empty, but that's a bit complicated.
1542 		 *
1543 		 * CAS_TX_COMPn points to the last descriptor
1544 		 * processed + 1.
1545 		 */
1546 		txlast = CAS_READ_4(sc, CAS_TX_COMP3);
1547 #ifdef CAS_DEBUG
1548 		CTR4(KTR_CAS, "%s: txs->txs_firstdesc = %d, "
1549 		    "txs->txs_lastdesc = %d, txlast = %d",
1550 		    __func__, txs->txs_firstdesc, txs->txs_lastdesc, txlast);
1551 #endif
1552 		if (txs->txs_firstdesc <= txs->txs_lastdesc) {
1553 			if ((txlast >= txs->txs_firstdesc) &&
1554 			    (txlast <= txs->txs_lastdesc))
1555 				break;
1556 		} else {
1557 			/* Ick -- this command wraps. */
1558 			if ((txlast >= txs->txs_firstdesc) ||
1559 			    (txlast <= txs->txs_lastdesc))
1560 				break;
1561 		}
1562 
1563 #ifdef CAS_DEBUG
1564 		CTR1(KTR_CAS, "%s: releasing a descriptor", __func__);
1565 #endif
1566 		STAILQ_REMOVE_HEAD(&sc->sc_txdirtyq, txs_q);
1567 
1568 		sc->sc_txfree += txs->txs_ndescs;
1569 
1570 		bus_dmamap_sync(sc->sc_tdmatag, txs->txs_dmamap,
1571 		    BUS_DMASYNC_POSTWRITE);
1572 		bus_dmamap_unload(sc->sc_tdmatag, txs->txs_dmamap);
1573 		if (txs->txs_mbuf != NULL) {
1574 			m_freem(txs->txs_mbuf);
1575 			txs->txs_mbuf = NULL;
1576 		}
1577 
1578 		STAILQ_INSERT_TAIL(&sc->sc_txfreeq, txs, txs_q);
1579 
1580 		if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
1581 		progress = 1;
1582 	}
1583 
1584 #ifdef CAS_DEBUG
1585 	CTR5(KTR_CAS, "%s: CAS_TX_SM1 %x CAS_TX_SM2 %x CAS_TX_DESC_BASE %llx "
1586 	    "CAS_TX_COMP3 %x",
1587 	    __func__, CAS_READ_4(sc, CAS_TX_SM1), CAS_READ_4(sc, CAS_TX_SM2),
1588 	    ((long long)CAS_READ_4(sc, CAS_TX_DESC3_BASE_HI) << 32) |
1589 	    CAS_READ_4(sc, CAS_TX_DESC3_BASE_LO),
1590 	    CAS_READ_4(sc, CAS_TX_COMP3));
1591 #endif
1592 
1593 	if (progress) {
1594 		/* We freed some descriptors, so reset IFF_DRV_OACTIVE. */
1595 		ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1596 		if (STAILQ_EMPTY(&sc->sc_txdirtyq))
1597 			sc->sc_wdog_timer = 0;
1598 	}
1599 
1600 #ifdef CAS_DEBUG
1601 	CTR3(KTR_CAS, "%s: %s: watchdog %d",
1602 	    device_get_name(sc->sc_dev), __func__, sc->sc_wdog_timer);
1603 #endif
1604 }
1605 
1606 static void
1607 cas_rint_timeout(void *arg)
1608 {
1609 	struct epoch_tracker et;
1610 	struct cas_softc *sc = arg;
1611 
1612 	CAS_LOCK_ASSERT(sc, MA_OWNED);
1613 
1614 	NET_EPOCH_ENTER(et);
1615 	cas_rint(sc);
1616 	NET_EPOCH_EXIT(et);
1617 }
1618 
1619 static void
1620 cas_rint(struct cas_softc *sc)
1621 {
1622 	struct cas_rxdsoft *rxds, *rxds2;
1623 	struct ifnet *ifp = sc->sc_ifp;
1624 	struct mbuf *m, *m2;
1625 	uint64_t word1, word2, word3 __unused, word4;
1626 	uint32_t rxhead;
1627 	u_int idx, idx2, len, off, skip;
1628 
1629 	CAS_LOCK_ASSERT(sc, MA_OWNED);
1630 
1631 	callout_stop(&sc->sc_rx_ch);
1632 
1633 #ifdef CAS_DEBUG
1634 	CTR2(KTR_CAS, "%s: %s", device_get_name(sc->sc_dev), __func__);
1635 #endif
1636 
1637 #define	PRINTWORD(n, delimiter)						\
1638 	printf("word ## n: 0x%016llx%c", (long long)word ## n, delimiter)
1639 
1640 #define	SKIPASSERT(n)							\
1641 	KASSERT(sc->sc_rxcomps[sc->sc_rxcptr].crc_word ## n == 0,	\
1642 	    ("%s: word ## n not 0", __func__))
1643 
1644 #define	WORDTOH(n)							\
1645 	word ## n = le64toh(sc->sc_rxcomps[sc->sc_rxcptr].crc_word ## n)
1646 
1647 	/*
1648 	 * Read the completion head register once.  This limits
1649 	 * how long the following loop can execute.
1650 	 */
1651 	rxhead = CAS_READ_4(sc, CAS_RX_COMP_HEAD);
1652 #ifdef CAS_DEBUG
1653 	CTR4(KTR_CAS, "%s: sc->sc_rxcptr %d, sc->sc_rxdptr %d, head %d",
1654 	    __func__, sc->sc_rxcptr, sc->sc_rxdptr, rxhead);
1655 #endif
1656 	skip = 0;
1657 	CAS_CDSYNC(sc, BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
1658 	for (; sc->sc_rxcptr != rxhead;
1659 	    sc->sc_rxcptr = CAS_NEXTRXCOMP(sc->sc_rxcptr)) {
1660 		if (skip != 0) {
1661 			SKIPASSERT(1);
1662 			SKIPASSERT(2);
1663 			SKIPASSERT(3);
1664 
1665 			--skip;
1666 			goto skip;
1667 		}
1668 
1669 		WORDTOH(1);
1670 		WORDTOH(2);
1671 		WORDTOH(3);
1672 		WORDTOH(4);
1673 
1674 #ifdef CAS_DEBUG
1675 		if ((ifp->if_flags & IFF_DEBUG) != 0) {
1676 			printf("    completion %d: ", sc->sc_rxcptr);
1677 			PRINTWORD(1, '\t');
1678 			PRINTWORD(2, '\t');
1679 			PRINTWORD(3, '\t');
1680 			PRINTWORD(4, '\n');
1681 		}
1682 #endif
1683 
1684 		if (__predict_false(
1685 		    (word1 & CAS_RC1_TYPE_MASK) == CAS_RC1_TYPE_HW ||
1686 		    (word4 & CAS_RC4_ZERO) != 0)) {
1687 			/*
1688 			 * The descriptor is still marked as owned, although
1689 			 * it is supposed to have completed.  This has been
1690 			 * observed on some machines.  Just exiting here
1691 			 * might leave the packet sitting around until another
1692 			 * one arrives to trigger a new interrupt, which is
1693 			 * generally undesirable, so set up a timeout.
1694 			 */
1695 			callout_reset(&sc->sc_rx_ch, CAS_RXOWN_TICKS,
1696 			    cas_rint_timeout, sc);
1697 			break;
1698 		}
1699 
1700 		if (__predict_false(
1701 		    (word4 & (CAS_RC4_BAD | CAS_RC4_LEN_MMATCH)) != 0)) {
1702 			if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
1703 			device_printf(sc->sc_dev,
1704 			    "receive error: CRC error\n");
1705 			continue;
1706 		}
1707 
1708 		KASSERT(CAS_GET(word1, CAS_RC1_DATA_SIZE) == 0 ||
1709 		    CAS_GET(word2, CAS_RC2_HDR_SIZE) == 0,
1710 		    ("%s: data and header present", __func__));
1711 		KASSERT((word1 & CAS_RC1_SPLIT_PKT) == 0 ||
1712 		    CAS_GET(word2, CAS_RC2_HDR_SIZE) == 0,
1713 		    ("%s: split and header present", __func__));
1714 		KASSERT(CAS_GET(word1, CAS_RC1_DATA_SIZE) == 0 ||
1715 		    (word1 & CAS_RC1_RELEASE_HDR) == 0,
1716 		    ("%s: data present but header release", __func__));
1717 		KASSERT(CAS_GET(word2, CAS_RC2_HDR_SIZE) == 0 ||
1718 		    (word1 & CAS_RC1_RELEASE_DATA) == 0,
1719 		    ("%s: header present but data release", __func__));
1720 
1721 		if ((len = CAS_GET(word2, CAS_RC2_HDR_SIZE)) != 0) {
1722 			idx = CAS_GET(word2, CAS_RC2_HDR_INDEX);
1723 			off = CAS_GET(word2, CAS_RC2_HDR_OFF);
1724 #ifdef CAS_DEBUG
1725 			CTR4(KTR_CAS, "%s: hdr at idx %d, off %d, len %d",
1726 			    __func__, idx, off, len);
1727 #endif
1728 			rxds = &sc->sc_rxdsoft[idx];
1729 			MGETHDR(m, M_NOWAIT, MT_DATA);
1730 			if (m != NULL) {
1731 				refcount_acquire(&rxds->rxds_refcount);
1732 				bus_dmamap_sync(sc->sc_rdmatag,
1733 				    rxds->rxds_dmamap, BUS_DMASYNC_POSTREAD);
1734 				m_extadd(m, (char *)rxds->rxds_buf +
1735 				    off * 256 + ETHER_ALIGN, len, cas_free,
1736 				    sc, (void *)(uintptr_t)idx,
1737 				    M_RDONLY, EXT_NET_DRV);
1738 				if ((m->m_flags & M_EXT) == 0) {
1739 					m_freem(m);
1740 					m = NULL;
1741 				}
1742 			}
1743 			if (m != NULL) {
1744 				m->m_pkthdr.rcvif = ifp;
1745 				m->m_pkthdr.len = m->m_len = len;
1746 				if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
1747 				if ((ifp->if_capenable & IFCAP_RXCSUM) != 0)
1748 					cas_rxcksum(m, CAS_GET(word4,
1749 					    CAS_RC4_TCP_CSUM));
1750 				/* Pass it on. */
1751 				CAS_UNLOCK(sc);
1752 				(*ifp->if_input)(ifp, m);
1753 				CAS_LOCK(sc);
1754 			} else
1755 				if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1756 
1757 			if ((word1 & CAS_RC1_RELEASE_HDR) != 0 &&
1758 			    refcount_release(&rxds->rxds_refcount) != 0)
1759 				cas_add_rxdesc(sc, idx);
1760 		} else if ((len = CAS_GET(word1, CAS_RC1_DATA_SIZE)) != 0) {
1761 			idx = CAS_GET(word1, CAS_RC1_DATA_INDEX);
1762 			off = CAS_GET(word1, CAS_RC1_DATA_OFF);
1763 #ifdef CAS_DEBUG
1764 			CTR4(KTR_CAS, "%s: data at idx %d, off %d, len %d",
1765 			    __func__, idx, off, len);
1766 #endif
1767 			rxds = &sc->sc_rxdsoft[idx];
1768 			MGETHDR(m, M_NOWAIT, MT_DATA);
1769 			if (m != NULL) {
1770 				refcount_acquire(&rxds->rxds_refcount);
1771 				off += ETHER_ALIGN;
1772 				m->m_len = min(CAS_PAGE_SIZE - off, len);
1773 				bus_dmamap_sync(sc->sc_rdmatag,
1774 				    rxds->rxds_dmamap, BUS_DMASYNC_POSTREAD);
1775 				m_extadd(m, (char *)rxds->rxds_buf + off,
1776 				    m->m_len, cas_free, sc,
1777 				    (void *)(uintptr_t)idx, M_RDONLY,
1778 				    EXT_NET_DRV);
1779 				if ((m->m_flags & M_EXT) == 0) {
1780 					m_freem(m);
1781 					m = NULL;
1782 				}
1783 			}
1784 			idx2 = 0;
1785 			m2 = NULL;
1786 			rxds2 = NULL;
1787 			if ((word1 & CAS_RC1_SPLIT_PKT) != 0) {
1788 				KASSERT((word1 & CAS_RC1_RELEASE_NEXT) != 0,
1789 				    ("%s: split but no release next",
1790 				    __func__));
1791 
1792 				idx2 = CAS_GET(word2, CAS_RC2_NEXT_INDEX);
1793 #ifdef CAS_DEBUG
1794 				CTR2(KTR_CAS, "%s: split at idx %d",
1795 				    __func__, idx2);
1796 #endif
1797 				rxds2 = &sc->sc_rxdsoft[idx2];
1798 				if (m != NULL) {
1799 					MGET(m2, M_NOWAIT, MT_DATA);
1800 					if (m2 != NULL) {
1801 						refcount_acquire(
1802 						    &rxds2->rxds_refcount);
1803 						m2->m_len = len - m->m_len;
1804 						bus_dmamap_sync(
1805 						    sc->sc_rdmatag,
1806 						    rxds2->rxds_dmamap,
1807 						    BUS_DMASYNC_POSTREAD);
1808 						m_extadd(m2,
1809 						    (char *)rxds2->rxds_buf,
1810 						    m2->m_len, cas_free, sc,
1811 						    (void *)(uintptr_t)idx2,
1812 						    M_RDONLY, EXT_NET_DRV);
1813 						if ((m2->m_flags & M_EXT) ==
1814 						    0) {
1815 							m_freem(m2);
1816 							m2 = NULL;
1817 						}
1818 					}
1819 				}
1820 				if (m2 != NULL)
1821 					m->m_next = m2;
1822 				else if (m != NULL) {
1823 					m_freem(m);
1824 					m = NULL;
1825 				}
1826 			}
1827 			if (m != NULL) {
1828 				m->m_pkthdr.rcvif = ifp;
1829 				m->m_pkthdr.len = len;
1830 				if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
1831 				if ((ifp->if_capenable & IFCAP_RXCSUM) != 0)
1832 					cas_rxcksum(m, CAS_GET(word4,
1833 					    CAS_RC4_TCP_CSUM));
1834 				/* Pass it on. */
1835 				CAS_UNLOCK(sc);
1836 				(*ifp->if_input)(ifp, m);
1837 				CAS_LOCK(sc);
1838 			} else
1839 				if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1840 
1841 			if ((word1 & CAS_RC1_RELEASE_DATA) != 0 &&
1842 			    refcount_release(&rxds->rxds_refcount) != 0)
1843 				cas_add_rxdesc(sc, idx);
1844 			if ((word1 & CAS_RC1_SPLIT_PKT) != 0 &&
1845 			    refcount_release(&rxds2->rxds_refcount) != 0)
1846 				cas_add_rxdesc(sc, idx2);
1847 		}
1848 
1849 		skip = CAS_GET(word1, CAS_RC1_SKIP);
1850 
1851  skip:
1852 		cas_rxcompinit(&sc->sc_rxcomps[sc->sc_rxcptr]);
1853 		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1854 			break;
1855 	}
1856 	CAS_CDSYNC(sc, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1857 	CAS_WRITE_4(sc, CAS_RX_COMP_TAIL, sc->sc_rxcptr);
1858 
1859 #undef PRINTWORD
1860 #undef SKIPASSERT
1861 #undef WORDTOH
1862 
1863 #ifdef CAS_DEBUG
1864 	CTR4(KTR_CAS, "%s: done sc->sc_rxcptr %d, sc->sc_rxdptr %d, head %d",
1865 	    __func__, sc->sc_rxcptr, sc->sc_rxdptr,
1866 	    CAS_READ_4(sc, CAS_RX_COMP_HEAD));
1867 #endif
1868 }
1869 
1870 static void
1871 cas_free(struct mbuf *m)
1872 {
1873 	struct cas_rxdsoft *rxds;
1874 	struct cas_softc *sc;
1875 	u_int idx, locked;
1876 
1877 	sc = m->m_ext.ext_arg1;
1878 	idx = (uintptr_t)m->m_ext.ext_arg2;
1879 	rxds = &sc->sc_rxdsoft[idx];
1880 	if (refcount_release(&rxds->rxds_refcount) == 0)
1881 		return;
1882 
1883 	/*
1884 	 * NB: this function can be called via m_freem(9) within
1885 	 * this driver!
1886 	 */
1887 	if ((locked = CAS_LOCK_OWNED(sc)) == 0)
1888 		CAS_LOCK(sc);
1889 	cas_add_rxdesc(sc, idx);
1890 	if (locked == 0)
1891 		CAS_UNLOCK(sc);
1892 }
1893 
1894 static inline void
1895 cas_add_rxdesc(struct cas_softc *sc, u_int idx)
1896 {
1897 
1898 	CAS_LOCK_ASSERT(sc, MA_OWNED);
1899 
1900 	bus_dmamap_sync(sc->sc_rdmatag, sc->sc_rxdsoft[idx].rxds_dmamap,
1901 	    BUS_DMASYNC_PREREAD);
1902 	CAS_UPDATE_RXDESC(sc, sc->sc_rxdptr, idx);
1903 	sc->sc_rxdptr = CAS_NEXTRXDESC(sc->sc_rxdptr);
1904 
1905 	/*
1906 	 * Update the RX kick register.  This register has to point to the
1907 	 * descriptor after the last valid one (before the current batch)
1908 	 * and for optimum performance should be incremented in multiples
1909 	 * of 4 (the DMA engine fetches/updates descriptors in batches of 4).
1910 	 */
1911 	if ((sc->sc_rxdptr % 4) == 0) {
1912 		CAS_CDSYNC(sc, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1913 		CAS_WRITE_4(sc, CAS_RX_KICK,
1914 		    (sc->sc_rxdptr + CAS_NRXDESC - 4) & CAS_NRXDESC_MASK);
1915 	}
1916 }
1917 
1918 static void
1919 cas_eint(struct cas_softc *sc, u_int status)
1920 {
1921 	struct ifnet *ifp = sc->sc_ifp;
1922 
1923 	CAS_LOCK_ASSERT(sc, MA_OWNED);
1924 
1925 	if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
1926 
1927 	device_printf(sc->sc_dev, "%s: status 0x%x", __func__, status);
1928 	if ((status & CAS_INTR_PCI_ERROR_INT) != 0) {
1929 		status = CAS_READ_4(sc, CAS_ERROR_STATUS);
1930 		printf(", PCI bus error 0x%x", status);
1931 		if ((status & CAS_ERROR_OTHER) != 0) {
1932 			status = pci_read_config(sc->sc_dev, PCIR_STATUS, 2);
1933 			printf(", PCI status 0x%x", status);
1934 			pci_write_config(sc->sc_dev, PCIR_STATUS, status, 2);
1935 		}
1936 	}
1937 	printf("\n");
1938 
1939 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1940 	cas_init_locked(sc);
1941 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
1942 		taskqueue_enqueue(sc->sc_tq, &sc->sc_tx_task);
1943 }
1944 
1945 static int
1946 cas_intr(void *v)
1947 {
1948 	struct cas_softc *sc = v;
1949 
1950 	if (__predict_false((CAS_READ_4(sc, CAS_STATUS_ALIAS) &
1951 	    CAS_INTR_SUMMARY) == 0))
1952 		return (FILTER_STRAY);
1953 
1954 	/* Disable interrupts. */
1955 	CAS_WRITE_4(sc, CAS_INTMASK, 0xffffffff);
1956 	taskqueue_enqueue(sc->sc_tq, &sc->sc_intr_task);
1957 
1958 	return (FILTER_HANDLED);
1959 }
1960 
1961 static void
1962 cas_intr_task(void *arg, int pending __unused)
1963 {
1964 	struct cas_softc *sc = arg;
1965 	struct ifnet *ifp = sc->sc_ifp;
1966 	uint32_t status, status2;
1967 
1968 	CAS_LOCK_ASSERT(sc, MA_NOTOWNED);
1969 
1970 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1971 		return;
1972 
1973 	status = CAS_READ_4(sc, CAS_STATUS);
1974 	if (__predict_false((status & CAS_INTR_SUMMARY) == 0))
1975 		goto done;
1976 
1977 	CAS_LOCK(sc);
1978 #ifdef CAS_DEBUG
1979 	CTR4(KTR_CAS, "%s: %s: cplt %x, status %x",
1980 	    device_get_name(sc->sc_dev), __func__,
1981 	    (status >> CAS_STATUS_TX_COMP3_SHFT), (u_int)status);
1982 
1983 	/*
1984 	 * PCS interrupts must be cleared, otherwise no traffic is passed!
1985 	 */
1986 	if ((status & CAS_INTR_PCS_INT) != 0) {
1987 		status2 =
1988 		    CAS_READ_4(sc, CAS_PCS_INTR_STATUS) |
1989 		    CAS_READ_4(sc, CAS_PCS_INTR_STATUS);
1990 		if ((status2 & CAS_PCS_INTR_LINK) != 0)
1991 			device_printf(sc->sc_dev,
1992 			    "%s: PCS link status changed\n", __func__);
1993 	}
1994 	if ((status & CAS_MAC_CTRL_STATUS) != 0) {
1995 		status2 = CAS_READ_4(sc, CAS_MAC_CTRL_STATUS);
1996 		if ((status2 & CAS_MAC_CTRL_PAUSE) != 0)
1997 			device_printf(sc->sc_dev,
1998 			    "%s: PAUSE received (PAUSE time %d slots)\n",
1999 			    __func__,
2000 			    (status2 & CAS_MAC_CTRL_STATUS_PT_MASK) >>
2001 			    CAS_MAC_CTRL_STATUS_PT_SHFT);
2002 		if ((status2 & CAS_MAC_CTRL_PAUSE) != 0)
2003 			device_printf(sc->sc_dev,
2004 			    "%s: transited to PAUSE state\n", __func__);
2005 		if ((status2 & CAS_MAC_CTRL_NON_PAUSE) != 0)
2006 			device_printf(sc->sc_dev,
2007 			    "%s: transited to non-PAUSE state\n", __func__);
2008 	}
2009 	if ((status & CAS_INTR_MIF) != 0)
2010 		device_printf(sc->sc_dev, "%s: MIF interrupt\n", __func__);
2011 #endif
2012 
2013 	if (__predict_false((status &
2014 	    (CAS_INTR_TX_TAG_ERR | CAS_INTR_RX_TAG_ERR |
2015 	    CAS_INTR_RX_LEN_MMATCH | CAS_INTR_PCI_ERROR_INT)) != 0)) {
2016 		cas_eint(sc, status);
2017 		CAS_UNLOCK(sc);
2018 		return;
2019 	}
2020 
2021 	if (__predict_false(status & CAS_INTR_TX_MAC_INT)) {
2022 		status2 = CAS_READ_4(sc, CAS_MAC_TX_STATUS);
2023 		if ((status2 &
2024 		    (CAS_MAC_TX_UNDERRUN | CAS_MAC_TX_MAX_PKT_ERR)) != 0)
2025 			if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
2026 		else if ((status2 & ~CAS_MAC_TX_FRAME_XMTD) != 0)
2027 			device_printf(sc->sc_dev,
2028 			    "MAC TX fault, status %x\n", status2);
2029 	}
2030 
2031 	if (__predict_false(status & CAS_INTR_RX_MAC_INT)) {
2032 		status2 = CAS_READ_4(sc, CAS_MAC_RX_STATUS);
2033 		if ((status2 & CAS_MAC_RX_OVERFLOW) != 0)
2034 			if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
2035 		else if ((status2 & ~CAS_MAC_RX_FRAME_RCVD) != 0)
2036 			device_printf(sc->sc_dev,
2037 			    "MAC RX fault, status %x\n", status2);
2038 	}
2039 
2040 	if ((status &
2041 	    (CAS_INTR_RX_DONE | CAS_INTR_RX_BUF_NA | CAS_INTR_RX_COMP_FULL |
2042 	    CAS_INTR_RX_BUF_AEMPTY | CAS_INTR_RX_COMP_AFULL)) != 0) {
2043 		cas_rint(sc);
2044 #ifdef CAS_DEBUG
2045 		if (__predict_false((status &
2046 		    (CAS_INTR_RX_BUF_NA | CAS_INTR_RX_COMP_FULL |
2047 		    CAS_INTR_RX_BUF_AEMPTY | CAS_INTR_RX_COMP_AFULL)) != 0))
2048 			device_printf(sc->sc_dev,
2049 			    "RX fault, status %x\n", status);
2050 #endif
2051 	}
2052 
2053 	if ((status &
2054 	    (CAS_INTR_TX_INT_ME | CAS_INTR_TX_ALL | CAS_INTR_TX_DONE)) != 0)
2055 		cas_tint(sc);
2056 
2057 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2058 		CAS_UNLOCK(sc);
2059 		return;
2060 	} else if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2061 		taskqueue_enqueue(sc->sc_tq, &sc->sc_tx_task);
2062 	CAS_UNLOCK(sc);
2063 
2064 	status = CAS_READ_4(sc, CAS_STATUS_ALIAS);
2065 	if (__predict_false((status & CAS_INTR_SUMMARY) != 0)) {
2066 		taskqueue_enqueue(sc->sc_tq, &sc->sc_intr_task);
2067 		return;
2068 	}
2069 
2070  done:
2071 	/* Re-enable interrupts. */
2072 	CAS_WRITE_4(sc, CAS_INTMASK,
2073 	    ~(CAS_INTR_TX_INT_ME | CAS_INTR_TX_TAG_ERR |
2074 	    CAS_INTR_RX_DONE | CAS_INTR_RX_BUF_NA | CAS_INTR_RX_TAG_ERR |
2075 	    CAS_INTR_RX_COMP_FULL | CAS_INTR_RX_BUF_AEMPTY |
2076 	    CAS_INTR_RX_COMP_AFULL | CAS_INTR_RX_LEN_MMATCH |
2077 	    CAS_INTR_PCI_ERROR_INT
2078 #ifdef CAS_DEBUG
2079 	    | CAS_INTR_PCS_INT | CAS_INTR_MIF
2080 #endif
2081 	));
2082 }
2083 
2084 static void
2085 cas_watchdog(struct cas_softc *sc)
2086 {
2087 	struct ifnet *ifp = sc->sc_ifp;
2088 
2089 	CAS_LOCK_ASSERT(sc, MA_OWNED);
2090 
2091 #ifdef CAS_DEBUG
2092 	CTR4(KTR_CAS,
2093 	    "%s: CAS_RX_CONF %x CAS_MAC_RX_STATUS %x CAS_MAC_RX_CONF %x",
2094 	    __func__, CAS_READ_4(sc, CAS_RX_CONF),
2095 	    CAS_READ_4(sc, CAS_MAC_RX_STATUS),
2096 	    CAS_READ_4(sc, CAS_MAC_RX_CONF));
2097 	CTR4(KTR_CAS,
2098 	    "%s: CAS_TX_CONF %x CAS_MAC_TX_STATUS %x CAS_MAC_TX_CONF %x",
2099 	    __func__, CAS_READ_4(sc, CAS_TX_CONF),
2100 	    CAS_READ_4(sc, CAS_MAC_TX_STATUS),
2101 	    CAS_READ_4(sc, CAS_MAC_TX_CONF));
2102 #endif
2103 
2104 	if (sc->sc_wdog_timer == 0 || --sc->sc_wdog_timer != 0)
2105 		return;
2106 
2107 	if ((sc->sc_flags & CAS_LINK) != 0)
2108 		device_printf(sc->sc_dev, "device timeout\n");
2109 	else if (bootverbose)
2110 		device_printf(sc->sc_dev, "device timeout (no link)\n");
2111 	if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
2112 
2113 	/* Try to get more packets going. */
2114 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2115 	cas_init_locked(sc);
2116 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2117 		taskqueue_enqueue(sc->sc_tq, &sc->sc_tx_task);
2118 }
2119 
2120 static void
2121 cas_mifinit(struct cas_softc *sc)
2122 {
2123 
2124 	/* Configure the MIF in frame mode. */
2125 	CAS_WRITE_4(sc, CAS_MIF_CONF,
2126 	    CAS_READ_4(sc, CAS_MIF_CONF) & ~CAS_MIF_CONF_BB_MODE);
2127 	CAS_BARRIER(sc, CAS_MIF_CONF, 4,
2128 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2129 }
2130 
2131 /*
2132  * MII interface
2133  *
2134  * The MII interface supports at least three different operating modes:
2135  *
2136  * Bitbang mode is implemented using data, clock and output enable registers.
2137  *
2138  * Frame mode is implemented by loading a complete frame into the frame
2139  * register and polling the valid bit for completion.
2140  *
2141  * Polling mode uses the frame register but completion is indicated by
2142  * an interrupt.
2143  *
2144  */
2145 static int
2146 cas_mii_readreg(device_t dev, int phy, int reg)
2147 {
2148 	struct cas_softc *sc;
2149 	int n;
2150 	uint32_t v;
2151 
2152 #ifdef CAS_DEBUG_PHY
2153 	printf("%s: phy %d reg %d\n", __func__, phy, reg);
2154 #endif
2155 
2156 	sc = device_get_softc(dev);
2157 	if ((sc->sc_flags & CAS_SERDES) != 0) {
2158 		switch (reg) {
2159 		case MII_BMCR:
2160 			reg = CAS_PCS_CTRL;
2161 			break;
2162 		case MII_BMSR:
2163 			reg = CAS_PCS_STATUS;
2164 			break;
2165 		case MII_PHYIDR1:
2166 		case MII_PHYIDR2:
2167 			return (0);
2168 		case MII_ANAR:
2169 			reg = CAS_PCS_ANAR;
2170 			break;
2171 		case MII_ANLPAR:
2172 			reg = CAS_PCS_ANLPAR;
2173 			break;
2174 		case MII_EXTSR:
2175 			return (EXTSR_1000XFDX | EXTSR_1000XHDX);
2176 		default:
2177 			device_printf(sc->sc_dev,
2178 			    "%s: unhandled register %d\n", __func__, reg);
2179 			return (0);
2180 		}
2181 		return (CAS_READ_4(sc, reg));
2182 	}
2183 
2184 	/* Construct the frame command. */
2185 	v = CAS_MIF_FRAME_READ |
2186 	    (phy << CAS_MIF_FRAME_PHY_SHFT) |
2187 	    (reg << CAS_MIF_FRAME_REG_SHFT);
2188 
2189 	CAS_WRITE_4(sc, CAS_MIF_FRAME, v);
2190 	CAS_BARRIER(sc, CAS_MIF_FRAME, 4,
2191 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2192 	for (n = 0; n < 100; n++) {
2193 		DELAY(1);
2194 		v = CAS_READ_4(sc, CAS_MIF_FRAME);
2195 		if (v & CAS_MIF_FRAME_TA_LSB)
2196 			return (v & CAS_MIF_FRAME_DATA);
2197 	}
2198 
2199 	device_printf(sc->sc_dev, "%s: timed out\n", __func__);
2200 	return (0);
2201 }
2202 
2203 static int
2204 cas_mii_writereg(device_t dev, int phy, int reg, int val)
2205 {
2206 	struct cas_softc *sc;
2207 	int n;
2208 	uint32_t v;
2209 
2210 #ifdef CAS_DEBUG_PHY
2211 	printf("%s: phy %d reg %d val %x\n", phy, reg, val, __func__);
2212 #endif
2213 
2214 	sc = device_get_softc(dev);
2215 	if ((sc->sc_flags & CAS_SERDES) != 0) {
2216 		switch (reg) {
2217 		case MII_BMSR:
2218 			reg = CAS_PCS_STATUS;
2219 			break;
2220 		case MII_BMCR:
2221 			reg = CAS_PCS_CTRL;
2222 			if ((val & CAS_PCS_CTRL_RESET) == 0)
2223 				break;
2224 			CAS_WRITE_4(sc, CAS_PCS_CTRL, val);
2225 			CAS_BARRIER(sc, CAS_PCS_CTRL, 4,
2226 			    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2227 			if (!cas_bitwait(sc, CAS_PCS_CTRL,
2228 			    CAS_PCS_CTRL_RESET, 0))
2229 				device_printf(sc->sc_dev,
2230 				    "cannot reset PCS\n");
2231 			/* FALLTHROUGH */
2232 		case MII_ANAR:
2233 			CAS_WRITE_4(sc, CAS_PCS_CONF, 0);
2234 			CAS_BARRIER(sc, CAS_PCS_CONF, 4,
2235 			    BUS_SPACE_BARRIER_WRITE);
2236 			CAS_WRITE_4(sc, CAS_PCS_ANAR, val);
2237 			CAS_BARRIER(sc, CAS_PCS_ANAR, 4,
2238 			    BUS_SPACE_BARRIER_WRITE);
2239 			CAS_WRITE_4(sc, CAS_PCS_SERDES_CTRL,
2240 			    CAS_PCS_SERDES_CTRL_ESD);
2241 			CAS_BARRIER(sc, CAS_PCS_CONF, 4,
2242 			    BUS_SPACE_BARRIER_WRITE);
2243 			CAS_WRITE_4(sc, CAS_PCS_CONF,
2244 			    CAS_PCS_CONF_EN);
2245 			CAS_BARRIER(sc, CAS_PCS_CONF, 4,
2246 			    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2247 			return (0);
2248 		case MII_ANLPAR:
2249 			reg = CAS_PCS_ANLPAR;
2250 			break;
2251 		default:
2252 			device_printf(sc->sc_dev,
2253 			    "%s: unhandled register %d\n", __func__, reg);
2254 			return (0);
2255 		}
2256 		CAS_WRITE_4(sc, reg, val);
2257 		CAS_BARRIER(sc, reg, 4,
2258 		    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2259 		return (0);
2260 	}
2261 
2262 	/* Construct the frame command. */
2263 	v = CAS_MIF_FRAME_WRITE |
2264 	    (phy << CAS_MIF_FRAME_PHY_SHFT) |
2265 	    (reg << CAS_MIF_FRAME_REG_SHFT) |
2266 	    (val & CAS_MIF_FRAME_DATA);
2267 
2268 	CAS_WRITE_4(sc, CAS_MIF_FRAME, v);
2269 	CAS_BARRIER(sc, CAS_MIF_FRAME, 4,
2270 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2271 	for (n = 0; n < 100; n++) {
2272 		DELAY(1);
2273 		v = CAS_READ_4(sc, CAS_MIF_FRAME);
2274 		if (v & CAS_MIF_FRAME_TA_LSB)
2275 			return (1);
2276 	}
2277 
2278 	device_printf(sc->sc_dev, "%s: timed out\n", __func__);
2279 	return (0);
2280 }
2281 
2282 static void
2283 cas_mii_statchg(device_t dev)
2284 {
2285 	struct cas_softc *sc;
2286 	struct ifnet *ifp;
2287 	int gigabit;
2288 	uint32_t rxcfg, txcfg, v;
2289 
2290 	sc = device_get_softc(dev);
2291 	ifp = sc->sc_ifp;
2292 
2293 	CAS_LOCK_ASSERT(sc, MA_OWNED);
2294 
2295 #ifdef CAS_DEBUG
2296 	if ((ifp->if_flags & IFF_DEBUG) != 0)
2297 		device_printf(sc->sc_dev, "%s: status changen", __func__);
2298 #endif
2299 
2300 	if ((sc->sc_mii->mii_media_status & IFM_ACTIVE) != 0 &&
2301 	    IFM_SUBTYPE(sc->sc_mii->mii_media_active) != IFM_NONE)
2302 		sc->sc_flags |= CAS_LINK;
2303 	else
2304 		sc->sc_flags &= ~CAS_LINK;
2305 
2306 	switch (IFM_SUBTYPE(sc->sc_mii->mii_media_active)) {
2307 	case IFM_1000_SX:
2308 	case IFM_1000_LX:
2309 	case IFM_1000_CX:
2310 	case IFM_1000_T:
2311 		gigabit = 1;
2312 		break;
2313 	default:
2314 		gigabit = 0;
2315 	}
2316 
2317 	/*
2318 	 * The configuration done here corresponds to the steps F) and
2319 	 * G) and as far as enabling of RX and TX MAC goes also step H)
2320 	 * of the initialization sequence outlined in section 11.2.1 of
2321 	 * the Cassini+ ASIC Specification.
2322 	 */
2323 
2324 	rxcfg = sc->sc_mac_rxcfg;
2325 	rxcfg &= ~CAS_MAC_RX_CONF_CARR;
2326 	txcfg = CAS_MAC_TX_CONF_EN_IPG0 | CAS_MAC_TX_CONF_NGU |
2327 	    CAS_MAC_TX_CONF_NGUL;
2328 	if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) & IFM_FDX) != 0)
2329 		txcfg |= CAS_MAC_TX_CONF_ICARR | CAS_MAC_TX_CONF_ICOLLIS;
2330 	else if (gigabit != 0) {
2331 		rxcfg |= CAS_MAC_RX_CONF_CARR;
2332 		txcfg |= CAS_MAC_TX_CONF_CARR;
2333 	}
2334 	(void)cas_disable_tx(sc);
2335 	CAS_WRITE_4(sc, CAS_MAC_TX_CONF, txcfg);
2336 	(void)cas_disable_rx(sc);
2337 	CAS_WRITE_4(sc, CAS_MAC_RX_CONF, rxcfg);
2338 
2339 	v = CAS_READ_4(sc, CAS_MAC_CTRL_CONF) &
2340 	    ~(CAS_MAC_CTRL_CONF_TXP | CAS_MAC_CTRL_CONF_RXP);
2341 	if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) &
2342 	    IFM_ETH_RXPAUSE) != 0)
2343 		v |= CAS_MAC_CTRL_CONF_RXP;
2344 	if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) &
2345 	    IFM_ETH_TXPAUSE) != 0)
2346 		v |= CAS_MAC_CTRL_CONF_TXP;
2347 	CAS_WRITE_4(sc, CAS_MAC_CTRL_CONF, v);
2348 
2349 	/*
2350 	 * All supported chips have a bug causing incorrect checksum
2351 	 * to be calculated when letting them strip the FCS in half-
2352 	 * duplex mode.  In theory we could disable FCS stripping and
2353 	 * manually adjust the checksum accordingly.  It seems to make
2354 	 * more sense to optimze for the common case and just disable
2355 	 * hardware checksumming in half-duplex mode though.
2356 	 */
2357 	if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) & IFM_FDX) == 0) {
2358 		ifp->if_capenable &= ~IFCAP_HWCSUM;
2359 		ifp->if_hwassist = 0;
2360 	} else if ((sc->sc_flags & CAS_NO_CSUM) == 0) {
2361 		ifp->if_capenable = ifp->if_capabilities;
2362 		ifp->if_hwassist = CAS_CSUM_FEATURES;
2363 	}
2364 
2365 	if (sc->sc_variant == CAS_SATURN) {
2366 		if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) & IFM_FDX) == 0)
2367 			/* silicon bug workaround */
2368 			CAS_WRITE_4(sc, CAS_MAC_PREAMBLE_LEN, 0x41);
2369 		else
2370 			CAS_WRITE_4(sc, CAS_MAC_PREAMBLE_LEN, 0x7);
2371 	}
2372 
2373 	if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) & IFM_FDX) == 0 &&
2374 	    gigabit != 0)
2375 		CAS_WRITE_4(sc, CAS_MAC_SLOT_TIME,
2376 		    CAS_MAC_SLOT_TIME_CARR);
2377 	else
2378 		CAS_WRITE_4(sc, CAS_MAC_SLOT_TIME,
2379 		    CAS_MAC_SLOT_TIME_NORM);
2380 
2381 	/* XIF Configuration */
2382 	v = CAS_MAC_XIF_CONF_TX_OE | CAS_MAC_XIF_CONF_LNKLED;
2383 	if ((sc->sc_flags & CAS_SERDES) == 0) {
2384 		if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) & IFM_FDX) == 0)
2385 			v |= CAS_MAC_XIF_CONF_NOECHO;
2386 		v |= CAS_MAC_XIF_CONF_BUF_OE;
2387 	}
2388 	if (gigabit != 0)
2389 		v |= CAS_MAC_XIF_CONF_GMII;
2390 	if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) & IFM_FDX) != 0)
2391 		v |= CAS_MAC_XIF_CONF_FDXLED;
2392 	CAS_WRITE_4(sc, CAS_MAC_XIF_CONF, v);
2393 
2394 	sc->sc_mac_rxcfg = rxcfg;
2395 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0 &&
2396 	    (sc->sc_flags & CAS_LINK) != 0) {
2397 		CAS_WRITE_4(sc, CAS_MAC_TX_CONF,
2398 		    txcfg | CAS_MAC_TX_CONF_EN);
2399 		CAS_WRITE_4(sc, CAS_MAC_RX_CONF,
2400 		    rxcfg | CAS_MAC_RX_CONF_EN);
2401 	}
2402 }
2403 
2404 static int
2405 cas_mediachange(struct ifnet *ifp)
2406 {
2407 	struct cas_softc *sc = ifp->if_softc;
2408 	int error;
2409 
2410 	/* XXX add support for serial media. */
2411 
2412 	CAS_LOCK(sc);
2413 	error = mii_mediachg(sc->sc_mii);
2414 	CAS_UNLOCK(sc);
2415 	return (error);
2416 }
2417 
2418 static void
2419 cas_mediastatus(struct ifnet *ifp, struct ifmediareq *ifmr)
2420 {
2421 	struct cas_softc *sc = ifp->if_softc;
2422 
2423 	CAS_LOCK(sc);
2424 	if ((ifp->if_flags & IFF_UP) == 0) {
2425 		CAS_UNLOCK(sc);
2426 		return;
2427 	}
2428 
2429 	mii_pollstat(sc->sc_mii);
2430 	ifmr->ifm_active = sc->sc_mii->mii_media_active;
2431 	ifmr->ifm_status = sc->sc_mii->mii_media_status;
2432 	CAS_UNLOCK(sc);
2433 }
2434 
2435 static int
2436 cas_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
2437 {
2438 	struct cas_softc *sc = ifp->if_softc;
2439 	struct ifreq *ifr = (struct ifreq *)data;
2440 	int error;
2441 
2442 	error = 0;
2443 	switch (cmd) {
2444 	case SIOCSIFFLAGS:
2445 		CAS_LOCK(sc);
2446 		if ((ifp->if_flags & IFF_UP) != 0) {
2447 			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0 &&
2448 			    ((ifp->if_flags ^ sc->sc_ifflags) &
2449 			    (IFF_ALLMULTI | IFF_PROMISC)) != 0)
2450 				cas_setladrf(sc);
2451 			else
2452 				cas_init_locked(sc);
2453 		} else if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
2454 			cas_stop(ifp);
2455 		sc->sc_ifflags = ifp->if_flags;
2456 		CAS_UNLOCK(sc);
2457 		break;
2458 	case SIOCSIFCAP:
2459 		CAS_LOCK(sc);
2460 		if ((sc->sc_flags & CAS_NO_CSUM) != 0) {
2461 			error = EINVAL;
2462 			CAS_UNLOCK(sc);
2463 			break;
2464 		}
2465 		ifp->if_capenable = ifr->ifr_reqcap;
2466 		if ((ifp->if_capenable & IFCAP_TXCSUM) != 0)
2467 			ifp->if_hwassist = CAS_CSUM_FEATURES;
2468 		else
2469 			ifp->if_hwassist = 0;
2470 		CAS_UNLOCK(sc);
2471 		break;
2472 	case SIOCADDMULTI:
2473 	case SIOCDELMULTI:
2474 		CAS_LOCK(sc);
2475 		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
2476 			cas_setladrf(sc);
2477 		CAS_UNLOCK(sc);
2478 		break;
2479 	case SIOCSIFMTU:
2480 		if ((ifr->ifr_mtu < ETHERMIN) ||
2481 		    (ifr->ifr_mtu > ETHERMTU_JUMBO))
2482 			error = EINVAL;
2483 		else
2484 			ifp->if_mtu = ifr->ifr_mtu;
2485 		break;
2486 	case SIOCGIFMEDIA:
2487 	case SIOCSIFMEDIA:
2488 		error = ifmedia_ioctl(ifp, ifr, &sc->sc_mii->mii_media, cmd);
2489 		break;
2490 	default:
2491 		error = ether_ioctl(ifp, cmd, data);
2492 		break;
2493 	}
2494 
2495 	return (error);
2496 }
2497 
2498 static u_int
2499 cas_hash_maddr(void *arg, struct sockaddr_dl *sdl, u_int cnt)
2500 {
2501 	uint32_t crc, *hash = arg;
2502 
2503 	crc = ether_crc32_le(LLADDR(sdl), ETHER_ADDR_LEN);
2504 	/* We just want the 8 most significant bits. */
2505 	crc >>= 24;
2506 	/* Set the corresponding bit in the filter. */
2507 	hash[crc >> 4] |= 1 << (15 - (crc & 15));
2508 
2509 	return (1);
2510 }
2511 
2512 static void
2513 cas_setladrf(struct cas_softc *sc)
2514 {
2515 	struct ifnet *ifp = sc->sc_ifp;
2516 	int i;
2517 	uint32_t hash[16];
2518 	uint32_t v;
2519 
2520 	CAS_LOCK_ASSERT(sc, MA_OWNED);
2521 
2522 	/*
2523 	 * Turn off the RX MAC and the hash filter as required by the Sun
2524 	 * Cassini programming restrictions.
2525 	 */
2526 	v = sc->sc_mac_rxcfg & ~(CAS_MAC_RX_CONF_HFILTER |
2527 	    CAS_MAC_RX_CONF_EN);
2528 	CAS_WRITE_4(sc, CAS_MAC_RX_CONF, v);
2529 	CAS_BARRIER(sc, CAS_MAC_RX_CONF, 4,
2530 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2531 	if (!cas_bitwait(sc, CAS_MAC_RX_CONF, CAS_MAC_RX_CONF_HFILTER |
2532 	    CAS_MAC_RX_CONF_EN, 0))
2533 		device_printf(sc->sc_dev,
2534 		    "cannot disable RX MAC or hash filter\n");
2535 
2536 	v &= ~(CAS_MAC_RX_CONF_PROMISC | CAS_MAC_RX_CONF_PGRP);
2537 	if ((ifp->if_flags & IFF_PROMISC) != 0) {
2538 		v |= CAS_MAC_RX_CONF_PROMISC;
2539 		goto chipit;
2540 	}
2541 	if ((ifp->if_flags & IFF_ALLMULTI) != 0) {
2542 		v |= CAS_MAC_RX_CONF_PGRP;
2543 		goto chipit;
2544 	}
2545 
2546 	/*
2547 	 * Set up multicast address filter by passing all multicast
2548 	 * addresses through a crc generator, and then using the high
2549 	 * order 8 bits as an index into the 256 bit logical address
2550 	 * filter.  The high order 4 bits selects the word, while the
2551 	 * other 4 bits select the bit within the word (where bit 0
2552 	 * is the MSB).
2553 	 */
2554 
2555 	memset(hash, 0, sizeof(hash));
2556 	if_foreach_llmaddr(ifp, cas_hash_maddr, &hash);
2557 
2558 	v |= CAS_MAC_RX_CONF_HFILTER;
2559 
2560 	/* Now load the hash table into the chip (if we are using it). */
2561 	for (i = 0; i < 16; i++)
2562 		CAS_WRITE_4(sc,
2563 		    CAS_MAC_HASH0 + i * (CAS_MAC_HASH1 - CAS_MAC_HASH0),
2564 		    hash[i]);
2565 
2566  chipit:
2567 	sc->sc_mac_rxcfg = v;
2568 	CAS_WRITE_4(sc, CAS_MAC_RX_CONF, v | CAS_MAC_RX_CONF_EN);
2569 }
2570 
2571 static int	cas_pci_attach(device_t dev);
2572 static int	cas_pci_detach(device_t dev);
2573 static int	cas_pci_probe(device_t dev);
2574 static int	cas_pci_resume(device_t dev);
2575 static int	cas_pci_suspend(device_t dev);
2576 
2577 static device_method_t cas_pci_methods[] = {
2578 	/* Device interface */
2579 	DEVMETHOD(device_probe,		cas_pci_probe),
2580 	DEVMETHOD(device_attach,	cas_pci_attach),
2581 	DEVMETHOD(device_detach,	cas_pci_detach),
2582 	DEVMETHOD(device_suspend,	cas_pci_suspend),
2583 	DEVMETHOD(device_resume,	cas_pci_resume),
2584 	/* Use the suspend handler here, it is all that is required. */
2585 	DEVMETHOD(device_shutdown,	cas_pci_suspend),
2586 
2587 	/* MII interface */
2588 	DEVMETHOD(miibus_readreg,	cas_mii_readreg),
2589 	DEVMETHOD(miibus_writereg,	cas_mii_writereg),
2590 	DEVMETHOD(miibus_statchg,	cas_mii_statchg),
2591 
2592 	DEVMETHOD_END
2593 };
2594 
2595 static driver_t cas_pci_driver = {
2596 	"cas",
2597 	cas_pci_methods,
2598 	sizeof(struct cas_softc)
2599 };
2600 
2601 static const struct cas_pci_dev {
2602 	uint32_t	cpd_devid;
2603 	uint8_t		cpd_revid;
2604 	int		cpd_variant;
2605 	const char	*cpd_desc;
2606 } cas_pci_devlist[] = {
2607 	{ 0x0035100b, 0x0, CAS_SATURN, "NS DP83065 Saturn Gigabit Ethernet" },
2608 	{ 0xabba108e, 0x10, CAS_CASPLUS, "Sun Cassini+ Gigabit Ethernet" },
2609 	{ 0xabba108e, 0x0, CAS_CAS, "Sun Cassini Gigabit Ethernet" },
2610 	{ 0, 0, 0, NULL }
2611 };
2612 
2613 DRIVER_MODULE(cas, pci, cas_pci_driver, 0, 0);
2614 MODULE_PNP_INFO("W32:vendor/device", pci, cas, cas_pci_devlist,
2615     nitems(cas_pci_devlist) - 1);
2616 DRIVER_MODULE(miibus, cas, miibus_driver, 0, 0);
2617 MODULE_DEPEND(cas, pci, 1, 1, 1);
2618 
2619 static int
2620 cas_pci_probe(device_t dev)
2621 {
2622 	int i;
2623 
2624 	for (i = 0; cas_pci_devlist[i].cpd_desc != NULL; i++) {
2625 		if (pci_get_devid(dev) == cas_pci_devlist[i].cpd_devid &&
2626 		    pci_get_revid(dev) >= cas_pci_devlist[i].cpd_revid) {
2627 			device_set_desc(dev, cas_pci_devlist[i].cpd_desc);
2628 			return (BUS_PROBE_DEFAULT);
2629 		}
2630 	}
2631 
2632 	return (ENXIO);
2633 }
2634 
2635 static struct resource_spec cas_pci_res_spec[] = {
2636 	{ SYS_RES_IRQ, 0, RF_SHAREABLE | RF_ACTIVE },	/* CAS_RES_INTR */
2637 	{ SYS_RES_MEMORY, PCIR_BAR(0), RF_ACTIVE },	/* CAS_RES_MEM */
2638 	{ -1, 0 }
2639 };
2640 
2641 #define	CAS_LOCAL_MAC_ADDRESS	"local-mac-address"
2642 #define	CAS_PHY_INTERFACE	"phy-interface"
2643 #define	CAS_PHY_TYPE		"phy-type"
2644 #define	CAS_PHY_TYPE_PCS	"pcs"
2645 
2646 static int
2647 cas_pci_attach(device_t dev)
2648 {
2649 	char buf[sizeof(CAS_LOCAL_MAC_ADDRESS)];
2650 	struct cas_softc *sc;
2651 	int i;
2652 #if !defined(__powerpc__)
2653 	u_char enaddr[4][ETHER_ADDR_LEN];
2654 	u_int j, k, lma, pcs[4], phy;
2655 #endif
2656 
2657 	sc = device_get_softc(dev);
2658 	sc->sc_variant = CAS_UNKNOWN;
2659 	for (i = 0; cas_pci_devlist[i].cpd_desc != NULL; i++) {
2660 		if (pci_get_devid(dev) == cas_pci_devlist[i].cpd_devid &&
2661 		    pci_get_revid(dev) >= cas_pci_devlist[i].cpd_revid) {
2662 			sc->sc_variant = cas_pci_devlist[i].cpd_variant;
2663 			break;
2664 		}
2665 	}
2666 	if (sc->sc_variant == CAS_UNKNOWN) {
2667 		device_printf(dev, "unknown adaptor\n");
2668 		return (ENXIO);
2669 	}
2670 
2671 	/* PCI configuration */
2672 	pci_write_config(dev, PCIR_COMMAND,
2673 	    pci_read_config(dev, PCIR_COMMAND, 2) | PCIM_CMD_BUSMASTEREN |
2674 	    PCIM_CMD_MWRICEN | PCIM_CMD_PERRESPEN | PCIM_CMD_SERRESPEN, 2);
2675 
2676 	sc->sc_dev = dev;
2677 	if (sc->sc_variant == CAS_CAS && pci_get_devid(dev) < 0x02)
2678 		/* Hardware checksumming may hang TX. */
2679 		sc->sc_flags |= CAS_NO_CSUM;
2680 	if (sc->sc_variant == CAS_CASPLUS || sc->sc_variant == CAS_SATURN)
2681 		sc->sc_flags |= CAS_REG_PLUS;
2682 	if (sc->sc_variant == CAS_CAS ||
2683 	    (sc->sc_variant == CAS_CASPLUS && pci_get_revid(dev) < 0x11))
2684 		sc->sc_flags |= CAS_TABORT;
2685 	if (bootverbose)
2686 		device_printf(dev, "flags=0x%x\n", sc->sc_flags);
2687 
2688 	if (bus_alloc_resources(dev, cas_pci_res_spec, sc->sc_res)) {
2689 		device_printf(dev, "failed to allocate resources\n");
2690 		bus_release_resources(dev, cas_pci_res_spec, sc->sc_res);
2691 		return (ENXIO);
2692 	}
2693 
2694 	CAS_LOCK_INIT(sc, device_get_nameunit(dev));
2695 
2696 #if defined(__powerpc__)
2697 	OF_getetheraddr(dev, sc->sc_enaddr);
2698 	if (OF_getprop(ofw_bus_get_node(dev), CAS_PHY_INTERFACE, buf,
2699 	    sizeof(buf)) > 0 || OF_getprop(ofw_bus_get_node(dev),
2700 	    CAS_PHY_TYPE, buf, sizeof(buf)) > 0) {
2701 		buf[sizeof(buf) - 1] = '\0';
2702 		if (strcmp(buf, CAS_PHY_TYPE_PCS) == 0)
2703 			sc->sc_flags |= CAS_SERDES;
2704 	}
2705 #else
2706 	/*
2707 	 * Dig out VPD (vital product data) and read the MAC address as well
2708 	 * as the PHY type.  The VPD resides in the PCI Expansion ROM (PCI
2709 	 * FCode) and can't be accessed via the PCI capability pointer.
2710 	 * SUNW,pci-ce and SUNW,pci-qge use the Enhanced VPD format described
2711 	 * in the free US Patent 7149820.
2712 	 */
2713 
2714 #define	PCI_ROMHDR_SIZE			0x1c
2715 #define	PCI_ROMHDR_SIG			0x00
2716 #define	PCI_ROMHDR_SIG_MAGIC		0xaa55		/* little endian */
2717 #define	PCI_ROMHDR_PTR_DATA		0x18
2718 #define	PCI_ROM_SIZE			0x18
2719 #define	PCI_ROM_SIG			0x00
2720 #define	PCI_ROM_SIG_MAGIC		0x52494350	/* "PCIR", endian */
2721 							/* reversed */
2722 #define	PCI_ROM_VENDOR			0x04
2723 #define	PCI_ROM_DEVICE			0x06
2724 #define	PCI_ROM_PTR_VPD			0x08
2725 #define	PCI_VPDRES_BYTE0		0x00
2726 #define	PCI_VPDRES_ISLARGE(x)		((x) & 0x80)
2727 #define	PCI_VPDRES_LARGE_NAME(x)	((x) & 0x7f)
2728 #define	PCI_VPDRES_LARGE_LEN_LSB	0x01
2729 #define	PCI_VPDRES_LARGE_LEN_MSB	0x02
2730 #define	PCI_VPDRES_LARGE_SIZE		0x03
2731 #define	PCI_VPDRES_TYPE_ID_STRING	0x02		/* large */
2732 #define	PCI_VPDRES_TYPE_VPD		0x10		/* large */
2733 #define	PCI_VPD_KEY0			0x00
2734 #define	PCI_VPD_KEY1			0x01
2735 #define	PCI_VPD_LEN			0x02
2736 #define	PCI_VPD_SIZE			0x03
2737 
2738 #define	CAS_ROM_READ_1(sc, offs)					\
2739 	CAS_READ_1((sc), CAS_PCI_ROM_OFFSET + (offs))
2740 #define	CAS_ROM_READ_2(sc, offs)					\
2741 	CAS_READ_2((sc), CAS_PCI_ROM_OFFSET + (offs))
2742 #define	CAS_ROM_READ_4(sc, offs)					\
2743 	CAS_READ_4((sc), CAS_PCI_ROM_OFFSET + (offs))
2744 
2745 	lma = phy = 0;
2746 	memset(enaddr, 0, sizeof(enaddr));
2747 	memset(pcs, 0, sizeof(pcs));
2748 
2749 	/* Enable PCI Expansion ROM access. */
2750 	CAS_WRITE_4(sc, CAS_BIM_LDEV_OEN,
2751 	    CAS_BIM_LDEV_OEN_PAD | CAS_BIM_LDEV_OEN_PROM);
2752 
2753 	/* Read PCI Expansion ROM header. */
2754 	if (CAS_ROM_READ_2(sc, PCI_ROMHDR_SIG) != PCI_ROMHDR_SIG_MAGIC ||
2755 	    (i = CAS_ROM_READ_2(sc, PCI_ROMHDR_PTR_DATA)) <
2756 	    PCI_ROMHDR_SIZE) {
2757 		device_printf(dev, "unexpected PCI Expansion ROM header\n");
2758 		goto fail_prom;
2759 	}
2760 
2761 	/* Read PCI Expansion ROM data. */
2762 	if (CAS_ROM_READ_4(sc, i + PCI_ROM_SIG) != PCI_ROM_SIG_MAGIC ||
2763 	    CAS_ROM_READ_2(sc, i + PCI_ROM_VENDOR) != pci_get_vendor(dev) ||
2764 	    CAS_ROM_READ_2(sc, i + PCI_ROM_DEVICE) != pci_get_device(dev) ||
2765 	    (j = CAS_ROM_READ_2(sc, i + PCI_ROM_PTR_VPD)) <
2766 	    i + PCI_ROM_SIZE) {
2767 		device_printf(dev, "unexpected PCI Expansion ROM data\n");
2768 		goto fail_prom;
2769 	}
2770 
2771 	/* Read PCI VPD. */
2772  next:
2773 	if (PCI_VPDRES_ISLARGE(CAS_ROM_READ_1(sc,
2774 	    j + PCI_VPDRES_BYTE0)) == 0) {
2775 		device_printf(dev, "no large PCI VPD\n");
2776 		goto fail_prom;
2777 	}
2778 
2779 	i = (CAS_ROM_READ_1(sc, j + PCI_VPDRES_LARGE_LEN_MSB) << 8) |
2780 	    CAS_ROM_READ_1(sc, j + PCI_VPDRES_LARGE_LEN_LSB);
2781 	switch (PCI_VPDRES_LARGE_NAME(CAS_ROM_READ_1(sc,
2782 	    j + PCI_VPDRES_BYTE0))) {
2783 	case PCI_VPDRES_TYPE_ID_STRING:
2784 		/* Skip identifier string. */
2785 		j += PCI_VPDRES_LARGE_SIZE + i;
2786 		goto next;
2787 	case PCI_VPDRES_TYPE_VPD:
2788 		for (j += PCI_VPDRES_LARGE_SIZE; i > 0;
2789 		    i -= PCI_VPD_SIZE + CAS_ROM_READ_1(sc, j + PCI_VPD_LEN),
2790 		    j += PCI_VPD_SIZE + CAS_ROM_READ_1(sc, j + PCI_VPD_LEN)) {
2791 			if (CAS_ROM_READ_1(sc, j + PCI_VPD_KEY0) != 'Z')
2792 				/* no Enhanced VPD */
2793 				continue;
2794 			if (CAS_ROM_READ_1(sc, j + PCI_VPD_SIZE) != 'I')
2795 				/* no instance property */
2796 				continue;
2797 			if (CAS_ROM_READ_1(sc, j + PCI_VPD_SIZE + 3) == 'B') {
2798 				/* byte array */
2799 				if (CAS_ROM_READ_1(sc,
2800 				    j + PCI_VPD_SIZE + 4) != ETHER_ADDR_LEN)
2801 					continue;
2802 				bus_read_region_1(sc->sc_res[CAS_RES_MEM],
2803 				    CAS_PCI_ROM_OFFSET + j + PCI_VPD_SIZE + 5,
2804 				    buf, sizeof(buf));
2805 				buf[sizeof(buf) - 1] = '\0';
2806 				if (strcmp(buf, CAS_LOCAL_MAC_ADDRESS) != 0)
2807 					continue;
2808 				bus_read_region_1(sc->sc_res[CAS_RES_MEM],
2809 				    CAS_PCI_ROM_OFFSET + j + PCI_VPD_SIZE +
2810 				    5 + sizeof(CAS_LOCAL_MAC_ADDRESS),
2811 				    enaddr[lma], sizeof(enaddr[lma]));
2812 				lma++;
2813 				if (lma == 4 && phy == 4)
2814 					break;
2815 			} else if (CAS_ROM_READ_1(sc, j + PCI_VPD_SIZE + 3) ==
2816 			   'S') {
2817 				/* string */
2818 				if (CAS_ROM_READ_1(sc,
2819 				    j + PCI_VPD_SIZE + 4) !=
2820 				    sizeof(CAS_PHY_TYPE_PCS))
2821 					continue;
2822 				bus_read_region_1(sc->sc_res[CAS_RES_MEM],
2823 				    CAS_PCI_ROM_OFFSET + j + PCI_VPD_SIZE + 5,
2824 				    buf, sizeof(buf));
2825 				buf[sizeof(buf) - 1] = '\0';
2826 				if (strcmp(buf, CAS_PHY_INTERFACE) == 0)
2827 					k = sizeof(CAS_PHY_INTERFACE);
2828 				else if (strcmp(buf, CAS_PHY_TYPE) == 0)
2829 					k = sizeof(CAS_PHY_TYPE);
2830 				else
2831 					continue;
2832 				bus_read_region_1(sc->sc_res[CAS_RES_MEM],
2833 				    CAS_PCI_ROM_OFFSET + j + PCI_VPD_SIZE +
2834 				    5 + k, buf, sizeof(buf));
2835 				buf[sizeof(buf) - 1] = '\0';
2836 				if (strcmp(buf, CAS_PHY_TYPE_PCS) == 0)
2837 					pcs[phy] = 1;
2838 				phy++;
2839 				if (lma == 4 && phy == 4)
2840 					break;
2841 			}
2842 		}
2843 		break;
2844 	default:
2845 		device_printf(dev, "unexpected PCI VPD\n");
2846 		goto fail_prom;
2847 	}
2848 
2849  fail_prom:
2850 	CAS_WRITE_4(sc, CAS_BIM_LDEV_OEN, 0);
2851 
2852 	if (lma == 0) {
2853 		device_printf(dev, "could not determine Ethernet address\n");
2854 		goto fail;
2855 	}
2856 	i = 0;
2857 	if (lma > 1 && pci_get_slot(dev) < nitems(enaddr))
2858 		i = pci_get_slot(dev);
2859 	memcpy(sc->sc_enaddr, enaddr[i], ETHER_ADDR_LEN);
2860 
2861 	if (phy == 0) {
2862 		device_printf(dev, "could not determine PHY type\n");
2863 		goto fail;
2864 	}
2865 	i = 0;
2866 	if (phy > 1 && pci_get_slot(dev) < nitems(pcs))
2867 		i = pci_get_slot(dev);
2868 	if (pcs[i] != 0)
2869 		sc->sc_flags |= CAS_SERDES;
2870 #endif
2871 
2872 	if (cas_attach(sc) != 0) {
2873 		device_printf(dev, "could not be attached\n");
2874 		goto fail;
2875 	}
2876 
2877 	if (bus_setup_intr(dev, sc->sc_res[CAS_RES_INTR], INTR_TYPE_NET |
2878 	    INTR_MPSAFE, cas_intr, NULL, sc, &sc->sc_ih) != 0) {
2879 		device_printf(dev, "failed to set up interrupt\n");
2880 		cas_detach(sc);
2881 		goto fail;
2882 	}
2883 	return (0);
2884 
2885  fail:
2886 	CAS_LOCK_DESTROY(sc);
2887 	bus_release_resources(dev, cas_pci_res_spec, sc->sc_res);
2888 	return (ENXIO);
2889 }
2890 
2891 static int
2892 cas_pci_detach(device_t dev)
2893 {
2894 	struct cas_softc *sc;
2895 
2896 	sc = device_get_softc(dev);
2897 	bus_teardown_intr(dev, sc->sc_res[CAS_RES_INTR], sc->sc_ih);
2898 	cas_detach(sc);
2899 	CAS_LOCK_DESTROY(sc);
2900 	bus_release_resources(dev, cas_pci_res_spec, sc->sc_res);
2901 	return (0);
2902 }
2903 
2904 static int
2905 cas_pci_suspend(device_t dev)
2906 {
2907 
2908 	cas_suspend(device_get_softc(dev));
2909 	return (0);
2910 }
2911 
2912 static int
2913 cas_pci_resume(device_t dev)
2914 {
2915 
2916 	cas_resume(device_get_softc(dev));
2917 	return (0);
2918 }
2919