xref: /freebsd/sys/dev/cas/if_cas.c (revision 732a02b4e77866604a120a275c082bb6221bd2ff)
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 static devclass_t cas_devclass;
176 
177 MODULE_DEPEND(cas, ether, 1, 1, 1);
178 MODULE_DEPEND(cas, miibus, 1, 1, 1);
179 
180 #ifdef CAS_DEBUG
181 #include <sys/ktr.h>
182 #define	KTR_CAS		KTR_SPARE2
183 #endif
184 
185 static int
186 cas_attach(struct cas_softc *sc)
187 {
188 	struct cas_txsoft *txs;
189 	struct ifnet *ifp;
190 	int error, i;
191 	uint32_t v;
192 
193 	/* Set up ifnet structure. */
194 	ifp = sc->sc_ifp = if_alloc(IFT_ETHER);
195 	if (ifp == NULL)
196 		return (ENOSPC);
197 	ifp->if_softc = sc;
198 	if_initname(ifp, device_get_name(sc->sc_dev),
199 	    device_get_unit(sc->sc_dev));
200 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
201 	ifp->if_start = cas_start;
202 	ifp->if_ioctl = cas_ioctl;
203 	ifp->if_init = cas_init;
204 	IFQ_SET_MAXLEN(&ifp->if_snd, CAS_TXQUEUELEN);
205 	ifp->if_snd.ifq_drv_maxlen = CAS_TXQUEUELEN;
206 	IFQ_SET_READY(&ifp->if_snd);
207 
208 	callout_init_mtx(&sc->sc_tick_ch, &sc->sc_mtx, 0);
209 	callout_init_mtx(&sc->sc_rx_ch, &sc->sc_mtx, 0);
210 	/* Create local taskq. */
211 	NET_TASK_INIT(&sc->sc_intr_task, 0, cas_intr_task, sc);
212 	TASK_INIT(&sc->sc_tx_task, 1, cas_tx_task, ifp);
213 	sc->sc_tq = taskqueue_create_fast("cas_taskq", M_WAITOK,
214 	    taskqueue_thread_enqueue, &sc->sc_tq);
215 	if (sc->sc_tq == NULL) {
216 		device_printf(sc->sc_dev, "could not create taskqueue\n");
217 		error = ENXIO;
218 		goto fail_ifnet;
219 	}
220 	error = taskqueue_start_threads(&sc->sc_tq, 1, PI_NET, "%s taskq",
221 	    device_get_nameunit(sc->sc_dev));
222 	if (error != 0) {
223 		device_printf(sc->sc_dev, "could not start threads\n");
224 		goto fail_taskq;
225 	}
226 
227 	/* Make sure the chip is stopped. */
228 	cas_reset(sc);
229 
230 	error = bus_dma_tag_create(bus_get_dma_tag(sc->sc_dev), 1, 0,
231 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
232 	    BUS_SPACE_MAXSIZE, 0, BUS_SPACE_MAXSIZE, 0, NULL, NULL,
233 	    &sc->sc_pdmatag);
234 	if (error != 0)
235 		goto fail_taskq;
236 
237 	error = bus_dma_tag_create(sc->sc_pdmatag, 1, 0,
238 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
239 	    CAS_PAGE_SIZE, 1, CAS_PAGE_SIZE, 0, NULL, NULL, &sc->sc_rdmatag);
240 	if (error != 0)
241 		goto fail_ptag;
242 
243 	error = bus_dma_tag_create(sc->sc_pdmatag, 1, 0,
244 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
245 	    MCLBYTES * CAS_NTXSEGS, CAS_NTXSEGS, MCLBYTES,
246 	    BUS_DMA_ALLOCNOW, NULL, NULL, &sc->sc_tdmatag);
247 	if (error != 0)
248 		goto fail_rtag;
249 
250 	error = bus_dma_tag_create(sc->sc_pdmatag, CAS_TX_DESC_ALIGN, 0,
251 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
252 	    sizeof(struct cas_control_data), 1,
253 	    sizeof(struct cas_control_data), 0,
254 	    NULL, NULL, &sc->sc_cdmatag);
255 	if (error != 0)
256 		goto fail_ttag;
257 
258 	/*
259 	 * Allocate the control data structures, create and load the
260 	 * DMA map for it.
261 	 */
262 	if ((error = bus_dmamem_alloc(sc->sc_cdmatag,
263 	    (void **)&sc->sc_control_data,
264 	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
265 	    &sc->sc_cddmamap)) != 0) {
266 		device_printf(sc->sc_dev,
267 		    "unable to allocate control data, error = %d\n", error);
268 		goto fail_ctag;
269 	}
270 
271 	sc->sc_cddma = 0;
272 	if ((error = bus_dmamap_load(sc->sc_cdmatag, sc->sc_cddmamap,
273 	    sc->sc_control_data, sizeof(struct cas_control_data),
274 	    cas_cddma_callback, sc, 0)) != 0 || sc->sc_cddma == 0) {
275 		device_printf(sc->sc_dev,
276 		    "unable to load control data DMA map, error = %d\n",
277 		    error);
278 		goto fail_cmem;
279 	}
280 
281 	/*
282 	 * Initialize the transmit job descriptors.
283 	 */
284 	STAILQ_INIT(&sc->sc_txfreeq);
285 	STAILQ_INIT(&sc->sc_txdirtyq);
286 
287 	/*
288 	 * Create the transmit buffer DMA maps.
289 	 */
290 	error = ENOMEM;
291 	for (i = 0; i < CAS_TXQUEUELEN; i++) {
292 		txs = &sc->sc_txsoft[i];
293 		txs->txs_mbuf = NULL;
294 		txs->txs_ndescs = 0;
295 		if ((error = bus_dmamap_create(sc->sc_tdmatag, 0,
296 		    &txs->txs_dmamap)) != 0) {
297 			device_printf(sc->sc_dev,
298 			    "unable to create TX DMA map %d, error = %d\n",
299 			    i, error);
300 			goto fail_txd;
301 		}
302 		STAILQ_INSERT_TAIL(&sc->sc_txfreeq, txs, txs_q);
303 	}
304 
305 	/*
306 	 * Allocate the receive buffers, create and load the DMA maps
307 	 * for them.
308 	 */
309 	for (i = 0; i < CAS_NRXDESC; i++) {
310 		if ((error = bus_dmamem_alloc(sc->sc_rdmatag,
311 		    &sc->sc_rxdsoft[i].rxds_buf, BUS_DMA_WAITOK,
312 		    &sc->sc_rxdsoft[i].rxds_dmamap)) != 0) {
313 			device_printf(sc->sc_dev,
314 			    "unable to allocate RX buffer %d, error = %d\n",
315 			    i, error);
316 			goto fail_rxmem;
317 		}
318 
319 		sc->sc_rxdptr = i;
320 		sc->sc_rxdsoft[i].rxds_paddr = 0;
321 		if ((error = bus_dmamap_load(sc->sc_rdmatag,
322 		    sc->sc_rxdsoft[i].rxds_dmamap, sc->sc_rxdsoft[i].rxds_buf,
323 		    CAS_PAGE_SIZE, cas_rxdma_callback, sc, 0)) != 0 ||
324 		    sc->sc_rxdsoft[i].rxds_paddr == 0) {
325 			device_printf(sc->sc_dev,
326 			    "unable to load RX DMA map %d, error = %d\n",
327 			    i, error);
328 			goto fail_rxmap;
329 		}
330 	}
331 
332 	if ((sc->sc_flags & CAS_SERDES) == 0) {
333 		CAS_WRITE_4(sc, CAS_PCS_DATAPATH, CAS_PCS_DATAPATH_MII);
334 		CAS_BARRIER(sc, CAS_PCS_DATAPATH, 4,
335 		    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
336 		cas_mifinit(sc);
337 		/*
338 		 * Look for an external PHY.
339 		 */
340 		error = ENXIO;
341 		v = CAS_READ_4(sc, CAS_MIF_CONF);
342 		if ((v & CAS_MIF_CONF_MDI1) != 0) {
343 			v |= CAS_MIF_CONF_PHY_SELECT;
344 			CAS_WRITE_4(sc, CAS_MIF_CONF, v);
345 			CAS_BARRIER(sc, CAS_MIF_CONF, 4,
346 			    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
347 			/* Enable/unfreeze the GMII pins of Saturn. */
348 			if (sc->sc_variant == CAS_SATURN) {
349 				CAS_WRITE_4(sc, CAS_SATURN_PCFG,
350 				    CAS_READ_4(sc, CAS_SATURN_PCFG) &
351 				    ~CAS_SATURN_PCFG_FSI);
352 				CAS_BARRIER(sc, CAS_SATURN_PCFG, 4,
353 				    BUS_SPACE_BARRIER_READ |
354 				    BUS_SPACE_BARRIER_WRITE);
355 				DELAY(10000);
356 			}
357 			error = mii_attach(sc->sc_dev, &sc->sc_miibus, ifp,
358 			    cas_mediachange, cas_mediastatus, BMSR_DEFCAPMASK,
359 			    MII_PHY_ANY, MII_OFFSET_ANY, MIIF_DOPAUSE);
360 		}
361 		/*
362 		 * Fall back on an internal PHY if no external PHY was found.
363 		 */
364 		if (error != 0 && (v & CAS_MIF_CONF_MDI0) != 0) {
365 			v &= ~CAS_MIF_CONF_PHY_SELECT;
366 			CAS_WRITE_4(sc, CAS_MIF_CONF, v);
367 			CAS_BARRIER(sc, CAS_MIF_CONF, 4,
368 			    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
369 			/* Freeze the GMII pins of Saturn for saving power. */
370 			if (sc->sc_variant == CAS_SATURN) {
371 				CAS_WRITE_4(sc, CAS_SATURN_PCFG,
372 				    CAS_READ_4(sc, CAS_SATURN_PCFG) |
373 				    CAS_SATURN_PCFG_FSI);
374 				CAS_BARRIER(sc, CAS_SATURN_PCFG, 4,
375 				    BUS_SPACE_BARRIER_READ |
376 				    BUS_SPACE_BARRIER_WRITE);
377 				DELAY(10000);
378 			}
379 			error = mii_attach(sc->sc_dev, &sc->sc_miibus, ifp,
380 			    cas_mediachange, cas_mediastatus, BMSR_DEFCAPMASK,
381 			    MII_PHY_ANY, MII_OFFSET_ANY, MIIF_DOPAUSE);
382 		}
383 	} else {
384 		/*
385 		 * Use the external PCS SERDES.
386 		 */
387 		CAS_WRITE_4(sc, CAS_PCS_DATAPATH, CAS_PCS_DATAPATH_SERDES);
388 		CAS_BARRIER(sc, CAS_PCS_DATAPATH, 4, BUS_SPACE_BARRIER_WRITE);
389 		/* Enable/unfreeze the SERDES pins of Saturn. */
390 		if (sc->sc_variant == CAS_SATURN) {
391 			CAS_WRITE_4(sc, CAS_SATURN_PCFG, 0);
392 			CAS_BARRIER(sc, CAS_SATURN_PCFG, 4,
393 			    BUS_SPACE_BARRIER_WRITE);
394 		}
395 		CAS_WRITE_4(sc, CAS_PCS_SERDES_CTRL, CAS_PCS_SERDES_CTRL_ESD);
396 		CAS_BARRIER(sc, CAS_PCS_SERDES_CTRL, 4,
397 		    BUS_SPACE_BARRIER_WRITE);
398 		CAS_WRITE_4(sc, CAS_PCS_CONF, CAS_PCS_CONF_EN);
399 		CAS_BARRIER(sc, CAS_PCS_CONF, 4,
400 		    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
401 		error = mii_attach(sc->sc_dev, &sc->sc_miibus, ifp,
402 		    cas_mediachange, cas_mediastatus, BMSR_DEFCAPMASK,
403 		    CAS_PHYAD_EXTERNAL, MII_OFFSET_ANY, MIIF_DOPAUSE);
404 	}
405 	if (error != 0) {
406 		device_printf(sc->sc_dev, "attaching PHYs failed\n");
407 		goto fail_rxmap;
408 	}
409 	sc->sc_mii = device_get_softc(sc->sc_miibus);
410 
411 	/*
412 	 * From this point forward, the attachment cannot fail.  A failure
413 	 * before this point releases all resources that may have been
414 	 * allocated.
415 	 */
416 
417 	/* Announce FIFO sizes. */
418 	v = CAS_READ_4(sc, CAS_TX_FIFO_SIZE);
419 	device_printf(sc->sc_dev, "%ukB RX FIFO, %ukB TX FIFO\n",
420 	    CAS_RX_FIFO_SIZE / 1024, v / 16);
421 
422 	/* Attach the interface. */
423 	ether_ifattach(ifp, sc->sc_enaddr);
424 
425 	/*
426 	 * Tell the upper layer(s) we support long frames/checksum offloads.
427 	 */
428 	ifp->if_hdrlen = sizeof(struct ether_vlan_header);
429 	ifp->if_capabilities = IFCAP_VLAN_MTU;
430 	if ((sc->sc_flags & CAS_NO_CSUM) == 0) {
431 		ifp->if_capabilities |= IFCAP_HWCSUM;
432 		ifp->if_hwassist = CAS_CSUM_FEATURES;
433 	}
434 	ifp->if_capenable = ifp->if_capabilities;
435 
436 	return (0);
437 
438 	/*
439 	 * Free any resources we've allocated during the failed attach
440 	 * attempt.  Do this in reverse order and fall through.
441 	 */
442  fail_rxmap:
443 	for (i = 0; i < CAS_NRXDESC; i++)
444 		if (sc->sc_rxdsoft[i].rxds_paddr != 0)
445 			bus_dmamap_unload(sc->sc_rdmatag,
446 			    sc->sc_rxdsoft[i].rxds_dmamap);
447  fail_rxmem:
448 	for (i = 0; i < CAS_NRXDESC; i++)
449 		if (sc->sc_rxdsoft[i].rxds_buf != NULL)
450 			bus_dmamem_free(sc->sc_rdmatag,
451 			    sc->sc_rxdsoft[i].rxds_buf,
452 			    sc->sc_rxdsoft[i].rxds_dmamap);
453  fail_txd:
454 	for (i = 0; i < CAS_TXQUEUELEN; i++)
455 		if (sc->sc_txsoft[i].txs_dmamap != NULL)
456 			bus_dmamap_destroy(sc->sc_tdmatag,
457 			    sc->sc_txsoft[i].txs_dmamap);
458 	bus_dmamap_unload(sc->sc_cdmatag, sc->sc_cddmamap);
459  fail_cmem:
460 	bus_dmamem_free(sc->sc_cdmatag, sc->sc_control_data,
461 	    sc->sc_cddmamap);
462  fail_ctag:
463 	bus_dma_tag_destroy(sc->sc_cdmatag);
464  fail_ttag:
465 	bus_dma_tag_destroy(sc->sc_tdmatag);
466  fail_rtag:
467 	bus_dma_tag_destroy(sc->sc_rdmatag);
468  fail_ptag:
469 	bus_dma_tag_destroy(sc->sc_pdmatag);
470  fail_taskq:
471 	taskqueue_free(sc->sc_tq);
472  fail_ifnet:
473 	if_free(ifp);
474 	return (error);
475 }
476 
477 static void
478 cas_detach(struct cas_softc *sc)
479 {
480 	struct ifnet *ifp = sc->sc_ifp;
481 	int i;
482 
483 	ether_ifdetach(ifp);
484 	CAS_LOCK(sc);
485 	cas_stop(ifp);
486 	CAS_UNLOCK(sc);
487 	callout_drain(&sc->sc_tick_ch);
488 	callout_drain(&sc->sc_rx_ch);
489 	taskqueue_drain(sc->sc_tq, &sc->sc_intr_task);
490 	taskqueue_drain(sc->sc_tq, &sc->sc_tx_task);
491 	if_free(ifp);
492 	taskqueue_free(sc->sc_tq);
493 	device_delete_child(sc->sc_dev, sc->sc_miibus);
494 
495 	for (i = 0; i < CAS_NRXDESC; i++)
496 		if (sc->sc_rxdsoft[i].rxds_dmamap != NULL)
497 			bus_dmamap_sync(sc->sc_rdmatag,
498 			    sc->sc_rxdsoft[i].rxds_dmamap,
499 			    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
500 	for (i = 0; i < CAS_NRXDESC; i++)
501 		if (sc->sc_rxdsoft[i].rxds_paddr != 0)
502 			bus_dmamap_unload(sc->sc_rdmatag,
503 			    sc->sc_rxdsoft[i].rxds_dmamap);
504 	for (i = 0; i < CAS_NRXDESC; i++)
505 		if (sc->sc_rxdsoft[i].rxds_buf != NULL)
506 			bus_dmamem_free(sc->sc_rdmatag,
507 			    sc->sc_rxdsoft[i].rxds_buf,
508 			    sc->sc_rxdsoft[i].rxds_dmamap);
509 	for (i = 0; i < CAS_TXQUEUELEN; i++)
510 		if (sc->sc_txsoft[i].txs_dmamap != NULL)
511 			bus_dmamap_destroy(sc->sc_tdmatag,
512 			    sc->sc_txsoft[i].txs_dmamap);
513 	CAS_CDSYNC(sc, BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
514 	bus_dmamap_unload(sc->sc_cdmatag, sc->sc_cddmamap);
515 	bus_dmamem_free(sc->sc_cdmatag, sc->sc_control_data,
516 	    sc->sc_cddmamap);
517 	bus_dma_tag_destroy(sc->sc_cdmatag);
518 	bus_dma_tag_destroy(sc->sc_tdmatag);
519 	bus_dma_tag_destroy(sc->sc_rdmatag);
520 	bus_dma_tag_destroy(sc->sc_pdmatag);
521 }
522 
523 static void
524 cas_suspend(struct cas_softc *sc)
525 {
526 	struct ifnet *ifp = sc->sc_ifp;
527 
528 	CAS_LOCK(sc);
529 	cas_stop(ifp);
530 	CAS_UNLOCK(sc);
531 }
532 
533 static void
534 cas_resume(struct cas_softc *sc)
535 {
536 	struct ifnet *ifp = sc->sc_ifp;
537 
538 	CAS_LOCK(sc);
539 	/*
540 	 * On resume all registers have to be initialized again like
541 	 * after power-on.
542 	 */
543 	sc->sc_flags &= ~CAS_INITED;
544 	if (ifp->if_flags & IFF_UP)
545 		cas_init_locked(sc);
546 	CAS_UNLOCK(sc);
547 }
548 
549 static inline void
550 cas_rxcksum(struct mbuf *m, uint16_t cksum)
551 {
552 	struct ether_header *eh;
553 	struct ip *ip;
554 	struct udphdr *uh;
555 	uint16_t *opts;
556 	int32_t hlen, len, pktlen;
557 	uint32_t temp32;
558 
559 	pktlen = m->m_pkthdr.len;
560 	if (pktlen < sizeof(struct ether_header) + sizeof(struct ip))
561 		return;
562 	eh = mtod(m, struct ether_header *);
563 	if (eh->ether_type != htons(ETHERTYPE_IP))
564 		return;
565 	ip = (struct ip *)(eh + 1);
566 	if (ip->ip_v != IPVERSION)
567 		return;
568 
569 	hlen = ip->ip_hl << 2;
570 	pktlen -= sizeof(struct ether_header);
571 	if (hlen < sizeof(struct ip))
572 		return;
573 	if (ntohs(ip->ip_len) < hlen)
574 		return;
575 	if (ntohs(ip->ip_len) != pktlen)
576 		return;
577 	if (ip->ip_off & htons(IP_MF | IP_OFFMASK))
578 		return;	/* Cannot handle fragmented packet. */
579 
580 	switch (ip->ip_p) {
581 	case IPPROTO_TCP:
582 		if (pktlen < (hlen + sizeof(struct tcphdr)))
583 			return;
584 		break;
585 	case IPPROTO_UDP:
586 		if (pktlen < (hlen + sizeof(struct udphdr)))
587 			return;
588 		uh = (struct udphdr *)((uint8_t *)ip + hlen);
589 		if (uh->uh_sum == 0)
590 			return; /* no checksum */
591 		break;
592 	default:
593 		return;
594 	}
595 
596 	cksum = ~cksum;
597 	/* checksum fixup for IP options */
598 	len = hlen - sizeof(struct ip);
599 	if (len > 0) {
600 		opts = (uint16_t *)(ip + 1);
601 		for (; len > 0; len -= sizeof(uint16_t), opts++) {
602 			temp32 = cksum - *opts;
603 			temp32 = (temp32 >> 16) + (temp32 & 65535);
604 			cksum = temp32 & 65535;
605 		}
606 	}
607 	m->m_pkthdr.csum_flags |= CSUM_DATA_VALID;
608 	m->m_pkthdr.csum_data = cksum;
609 }
610 
611 static void
612 cas_cddma_callback(void *xsc, bus_dma_segment_t *segs, int nsegs, int error)
613 {
614 	struct cas_softc *sc = xsc;
615 
616 	if (error != 0)
617 		return;
618 	if (nsegs != 1)
619 		panic("%s: bad control buffer segment count", __func__);
620 	sc->sc_cddma = segs[0].ds_addr;
621 }
622 
623 static void
624 cas_rxdma_callback(void *xsc, bus_dma_segment_t *segs, int nsegs, int error)
625 {
626 	struct cas_softc *sc = xsc;
627 
628 	if (error != 0)
629 		return;
630 	if (nsegs != 1)
631 		panic("%s: bad RX buffer segment count", __func__);
632 	sc->sc_rxdsoft[sc->sc_rxdptr].rxds_paddr = segs[0].ds_addr;
633 }
634 
635 static void
636 cas_tick(void *arg)
637 {
638 	struct cas_softc *sc = arg;
639 	struct ifnet *ifp = sc->sc_ifp;
640 	uint32_t v;
641 
642 	CAS_LOCK_ASSERT(sc, MA_OWNED);
643 
644 	/*
645 	 * Unload collision and error counters.
646 	 */
647 	if_inc_counter(ifp, IFCOUNTER_COLLISIONS,
648 	    CAS_READ_4(sc, CAS_MAC_NORM_COLL_CNT) +
649 	    CAS_READ_4(sc, CAS_MAC_FIRST_COLL_CNT));
650 	v = CAS_READ_4(sc, CAS_MAC_EXCESS_COLL_CNT) +
651 	    CAS_READ_4(sc, CAS_MAC_LATE_COLL_CNT);
652 	if_inc_counter(ifp, IFCOUNTER_COLLISIONS, v);
653 	if_inc_counter(ifp, IFCOUNTER_OERRORS, v);
654 	if_inc_counter(ifp, IFCOUNTER_IERRORS,
655 	    CAS_READ_4(sc, CAS_MAC_RX_LEN_ERR_CNT) +
656 	    CAS_READ_4(sc, CAS_MAC_RX_ALIGN_ERR) +
657 	    CAS_READ_4(sc, CAS_MAC_RX_CRC_ERR_CNT) +
658 	    CAS_READ_4(sc, CAS_MAC_RX_CODE_VIOL));
659 
660 	/*
661 	 * Then clear the hardware counters.
662 	 */
663 	CAS_WRITE_4(sc, CAS_MAC_NORM_COLL_CNT, 0);
664 	CAS_WRITE_4(sc, CAS_MAC_FIRST_COLL_CNT, 0);
665 	CAS_WRITE_4(sc, CAS_MAC_EXCESS_COLL_CNT, 0);
666 	CAS_WRITE_4(sc, CAS_MAC_LATE_COLL_CNT, 0);
667 	CAS_WRITE_4(sc, CAS_MAC_RX_LEN_ERR_CNT, 0);
668 	CAS_WRITE_4(sc, CAS_MAC_RX_ALIGN_ERR, 0);
669 	CAS_WRITE_4(sc, CAS_MAC_RX_CRC_ERR_CNT, 0);
670 	CAS_WRITE_4(sc, CAS_MAC_RX_CODE_VIOL, 0);
671 
672 	mii_tick(sc->sc_mii);
673 
674 	if (sc->sc_txfree != CAS_MAXTXFREE)
675 		cas_tint(sc);
676 
677 	cas_watchdog(sc);
678 
679 	callout_reset(&sc->sc_tick_ch, hz, cas_tick, sc);
680 }
681 
682 static int
683 cas_bitwait(struct cas_softc *sc, bus_addr_t r, uint32_t clr, uint32_t set)
684 {
685 	int i;
686 	uint32_t reg;
687 
688 	for (i = CAS_TRIES; i--; DELAY(100)) {
689 		reg = CAS_READ_4(sc, r);
690 		if ((reg & clr) == 0 && (reg & set) == set)
691 			return (1);
692 	}
693 	return (0);
694 }
695 
696 static void
697 cas_reset(struct cas_softc *sc)
698 {
699 
700 #ifdef CAS_DEBUG
701 	CTR2(KTR_CAS, "%s: %s", device_get_name(sc->sc_dev), __func__);
702 #endif
703 	/* Disable all interrupts in order to avoid spurious ones. */
704 	CAS_WRITE_4(sc, CAS_INTMASK, 0xffffffff);
705 
706 	cas_reset_rx(sc);
707 	cas_reset_tx(sc);
708 
709 	/*
710 	 * Do a full reset modulo the result of the last auto-negotiation
711 	 * when using the SERDES.
712 	 */
713 	CAS_WRITE_4(sc, CAS_RESET, CAS_RESET_RX | CAS_RESET_TX |
714 	    ((sc->sc_flags & CAS_SERDES) != 0 ? CAS_RESET_PCS_DIS : 0));
715 	CAS_BARRIER(sc, CAS_RESET, 4,
716 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
717 	DELAY(3000);
718 	if (!cas_bitwait(sc, CAS_RESET, CAS_RESET_RX | CAS_RESET_TX, 0))
719 		device_printf(sc->sc_dev, "cannot reset device\n");
720 }
721 
722 static void
723 cas_stop(struct ifnet *ifp)
724 {
725 	struct cas_softc *sc = ifp->if_softc;
726 	struct cas_txsoft *txs;
727 
728 #ifdef CAS_DEBUG
729 	CTR2(KTR_CAS, "%s: %s", device_get_name(sc->sc_dev), __func__);
730 #endif
731 
732 	callout_stop(&sc->sc_tick_ch);
733 	callout_stop(&sc->sc_rx_ch);
734 
735 	/* Disable all interrupts in order to avoid spurious ones. */
736 	CAS_WRITE_4(sc, CAS_INTMASK, 0xffffffff);
737 
738 	cas_reset_tx(sc);
739 	cas_reset_rx(sc);
740 
741 	/*
742 	 * Release any queued transmit buffers.
743 	 */
744 	while ((txs = STAILQ_FIRST(&sc->sc_txdirtyq)) != NULL) {
745 		STAILQ_REMOVE_HEAD(&sc->sc_txdirtyq, txs_q);
746 		if (txs->txs_ndescs != 0) {
747 			bus_dmamap_sync(sc->sc_tdmatag, txs->txs_dmamap,
748 			    BUS_DMASYNC_POSTWRITE);
749 			bus_dmamap_unload(sc->sc_tdmatag, txs->txs_dmamap);
750 			if (txs->txs_mbuf != NULL) {
751 				m_freem(txs->txs_mbuf);
752 				txs->txs_mbuf = NULL;
753 			}
754 		}
755 		STAILQ_INSERT_TAIL(&sc->sc_txfreeq, txs, txs_q);
756 	}
757 
758 	/*
759 	 * Mark the interface down and cancel the watchdog timer.
760 	 */
761 	ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
762 	sc->sc_flags &= ~CAS_LINK;
763 	sc->sc_wdog_timer = 0;
764 }
765 
766 static int
767 cas_reset_rx(struct cas_softc *sc)
768 {
769 
770 	/*
771 	 * Resetting while DMA is in progress can cause a bus hang, so we
772 	 * disable DMA first.
773 	 */
774 	(void)cas_disable_rx(sc);
775 	CAS_WRITE_4(sc, CAS_RX_CONF, 0);
776 	CAS_BARRIER(sc, CAS_RX_CONF, 4,
777 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
778 	if (!cas_bitwait(sc, CAS_RX_CONF, CAS_RX_CONF_RXDMA_EN, 0))
779 		device_printf(sc->sc_dev, "cannot disable RX DMA\n");
780 
781 	/* Finally, reset the ERX. */
782 	CAS_WRITE_4(sc, CAS_RESET, CAS_RESET_RX |
783 	    ((sc->sc_flags & CAS_SERDES) != 0 ? CAS_RESET_PCS_DIS : 0));
784 	CAS_BARRIER(sc, CAS_RESET, 4,
785 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
786 	if (!cas_bitwait(sc, CAS_RESET, CAS_RESET_RX, 0)) {
787 		device_printf(sc->sc_dev, "cannot reset receiver\n");
788 		return (1);
789 	}
790 	return (0);
791 }
792 
793 static int
794 cas_reset_tx(struct cas_softc *sc)
795 {
796 
797 	/*
798 	 * Resetting while DMA is in progress can cause a bus hang, so we
799 	 * disable DMA first.
800 	 */
801 	(void)cas_disable_tx(sc);
802 	CAS_WRITE_4(sc, CAS_TX_CONF, 0);
803 	CAS_BARRIER(sc, CAS_TX_CONF, 4,
804 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
805 	if (!cas_bitwait(sc, CAS_TX_CONF, CAS_TX_CONF_TXDMA_EN, 0))
806 		device_printf(sc->sc_dev, "cannot disable TX DMA\n");
807 
808 	/* Finally, reset the ETX. */
809 	CAS_WRITE_4(sc, CAS_RESET, CAS_RESET_TX |
810 	    ((sc->sc_flags & CAS_SERDES) != 0 ? CAS_RESET_PCS_DIS : 0));
811 	CAS_BARRIER(sc, CAS_RESET, 4,
812 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
813 	if (!cas_bitwait(sc, CAS_RESET, CAS_RESET_TX, 0)) {
814 		device_printf(sc->sc_dev, "cannot reset transmitter\n");
815 		return (1);
816 	}
817 	return (0);
818 }
819 
820 static int
821 cas_disable_rx(struct cas_softc *sc)
822 {
823 
824 	CAS_WRITE_4(sc, CAS_MAC_RX_CONF,
825 	    CAS_READ_4(sc, CAS_MAC_RX_CONF) & ~CAS_MAC_RX_CONF_EN);
826 	CAS_BARRIER(sc, CAS_MAC_RX_CONF, 4,
827 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
828 	if (cas_bitwait(sc, CAS_MAC_RX_CONF, CAS_MAC_RX_CONF_EN, 0))
829 		return (1);
830 	if (bootverbose)
831 		device_printf(sc->sc_dev, "cannot disable RX MAC\n");
832 	return (0);
833 }
834 
835 static int
836 cas_disable_tx(struct cas_softc *sc)
837 {
838 
839 	CAS_WRITE_4(sc, CAS_MAC_TX_CONF,
840 	    CAS_READ_4(sc, CAS_MAC_TX_CONF) & ~CAS_MAC_TX_CONF_EN);
841 	CAS_BARRIER(sc, CAS_MAC_TX_CONF, 4,
842 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
843 	if (cas_bitwait(sc, CAS_MAC_TX_CONF, CAS_MAC_TX_CONF_EN, 0))
844 		return (1);
845 	if (bootverbose)
846 		device_printf(sc->sc_dev, "cannot disable TX MAC\n");
847 	return (0);
848 }
849 
850 static inline void
851 cas_rxcompinit(struct cas_rx_comp *rxcomp)
852 {
853 
854 	rxcomp->crc_word1 = 0;
855 	rxcomp->crc_word2 = 0;
856 	rxcomp->crc_word3 =
857 	    htole64(CAS_SET(ETHER_HDR_LEN + sizeof(struct ip), CAS_RC3_CSO));
858 	rxcomp->crc_word4 = htole64(CAS_RC4_ZERO);
859 }
860 
861 static void
862 cas_meminit(struct cas_softc *sc)
863 {
864 	int i;
865 
866 	CAS_LOCK_ASSERT(sc, MA_OWNED);
867 
868 	/*
869 	 * Initialize the transmit descriptor ring.
870 	 */
871 	for (i = 0; i < CAS_NTXDESC; i++) {
872 		sc->sc_txdescs[i].cd_flags = 0;
873 		sc->sc_txdescs[i].cd_buf_ptr = 0;
874 	}
875 	sc->sc_txfree = CAS_MAXTXFREE;
876 	sc->sc_txnext = 0;
877 	sc->sc_txwin = 0;
878 
879 	/*
880 	 * Initialize the receive completion ring.
881 	 */
882 	for (i = 0; i < CAS_NRXCOMP; i++)
883 		cas_rxcompinit(&sc->sc_rxcomps[i]);
884 	sc->sc_rxcptr = 0;
885 
886 	/*
887 	 * Initialize the first receive descriptor ring.  We leave
888 	 * the second one zeroed as we don't actually use it.
889 	 */
890 	for (i = 0; i < CAS_NRXDESC; i++)
891 		CAS_INIT_RXDESC(sc, i, i);
892 	sc->sc_rxdptr = 0;
893 
894 	CAS_CDSYNC(sc, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
895 }
896 
897 static u_int
898 cas_descsize(u_int sz)
899 {
900 
901 	switch (sz) {
902 	case 32:
903 		return (CAS_DESC_32);
904 	case 64:
905 		return (CAS_DESC_64);
906 	case 128:
907 		return (CAS_DESC_128);
908 	case 256:
909 		return (CAS_DESC_256);
910 	case 512:
911 		return (CAS_DESC_512);
912 	case 1024:
913 		return (CAS_DESC_1K);
914 	case 2048:
915 		return (CAS_DESC_2K);
916 	case 4096:
917 		return (CAS_DESC_4K);
918 	case 8192:
919 		return (CAS_DESC_8K);
920 	default:
921 		printf("%s: invalid descriptor ring size %d\n", __func__, sz);
922 		return (CAS_DESC_32);
923 	}
924 }
925 
926 static u_int
927 cas_rxcompsize(u_int sz)
928 {
929 
930 	switch (sz) {
931 	case 128:
932 		return (CAS_RX_CONF_COMP_128);
933 	case 256:
934 		return (CAS_RX_CONF_COMP_256);
935 	case 512:
936 		return (CAS_RX_CONF_COMP_512);
937 	case 1024:
938 		return (CAS_RX_CONF_COMP_1K);
939 	case 2048:
940 		return (CAS_RX_CONF_COMP_2K);
941 	case 4096:
942 		return (CAS_RX_CONF_COMP_4K);
943 	case 8192:
944 		return (CAS_RX_CONF_COMP_8K);
945 	case 16384:
946 		return (CAS_RX_CONF_COMP_16K);
947 	case 32768:
948 		return (CAS_RX_CONF_COMP_32K);
949 	default:
950 		printf("%s: invalid dcompletion ring size %d\n", __func__, sz);
951 		return (CAS_RX_CONF_COMP_128);
952 	}
953 }
954 
955 static void
956 cas_init(void *xsc)
957 {
958 	struct cas_softc *sc = xsc;
959 
960 	CAS_LOCK(sc);
961 	cas_init_locked(sc);
962 	CAS_UNLOCK(sc);
963 }
964 
965 /*
966  * Initialization of interface; set up initialization block
967  * and transmit/receive descriptor rings.
968  */
969 static void
970 cas_init_locked(struct cas_softc *sc)
971 {
972 	struct ifnet *ifp = sc->sc_ifp;
973 	uint32_t v;
974 
975 	CAS_LOCK_ASSERT(sc, MA_OWNED);
976 
977 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
978 		return;
979 
980 #ifdef CAS_DEBUG
981 	CTR2(KTR_CAS, "%s: %s: calling stop", device_get_name(sc->sc_dev),
982 	    __func__);
983 #endif
984 	/*
985 	 * Initialization sequence.  The numbered steps below correspond
986 	 * to the sequence outlined in section 6.3.5.1 in the Ethernet
987 	 * Channel Engine manual (part of the PCIO manual).
988 	 * See also the STP2002-STQ document from Sun Microsystems.
989 	 */
990 
991 	/* step 1 & 2.  Reset the Ethernet Channel. */
992 	cas_stop(ifp);
993 	cas_reset(sc);
994 #ifdef CAS_DEBUG
995 	CTR2(KTR_CAS, "%s: %s: restarting", device_get_name(sc->sc_dev),
996 	    __func__);
997 #endif
998 
999 	if ((sc->sc_flags & CAS_SERDES) == 0)
1000 		/* Re-initialize the MIF. */
1001 		cas_mifinit(sc);
1002 
1003 	/* step 3.  Setup data structures in host memory. */
1004 	cas_meminit(sc);
1005 
1006 	/* step 4.  TX MAC registers & counters */
1007 	cas_init_regs(sc);
1008 
1009 	/* step 5.  RX MAC registers & counters */
1010 
1011 	/* step 6 & 7.  Program Ring Base Addresses. */
1012 	CAS_WRITE_4(sc, CAS_TX_DESC3_BASE_HI,
1013 	    (((uint64_t)CAS_CDTXDADDR(sc, 0)) >> 32));
1014 	CAS_WRITE_4(sc, CAS_TX_DESC3_BASE_LO,
1015 	    CAS_CDTXDADDR(sc, 0) & 0xffffffff);
1016 
1017 	CAS_WRITE_4(sc, CAS_RX_COMP_BASE_HI,
1018 	    (((uint64_t)CAS_CDRXCADDR(sc, 0)) >> 32));
1019 	CAS_WRITE_4(sc, CAS_RX_COMP_BASE_LO,
1020 	    CAS_CDRXCADDR(sc, 0) & 0xffffffff);
1021 
1022 	CAS_WRITE_4(sc, CAS_RX_DESC_BASE_HI,
1023 	    (((uint64_t)CAS_CDRXDADDR(sc, 0)) >> 32));
1024 	CAS_WRITE_4(sc, CAS_RX_DESC_BASE_LO,
1025 	    CAS_CDRXDADDR(sc, 0) & 0xffffffff);
1026 
1027 	if ((sc->sc_flags & CAS_REG_PLUS) != 0) {
1028 		CAS_WRITE_4(sc, CAS_RX_DESC2_BASE_HI,
1029 		    (((uint64_t)CAS_CDRXD2ADDR(sc, 0)) >> 32));
1030 		CAS_WRITE_4(sc, CAS_RX_DESC2_BASE_LO,
1031 		    CAS_CDRXD2ADDR(sc, 0) & 0xffffffff);
1032 	}
1033 
1034 #ifdef CAS_DEBUG
1035 	CTR5(KTR_CAS,
1036 	    "loading TXDR %lx, RXCR %lx, RXDR %lx, RXD2R %lx, cddma %lx",
1037 	    CAS_CDTXDADDR(sc, 0), CAS_CDRXCADDR(sc, 0), CAS_CDRXDADDR(sc, 0),
1038 	    CAS_CDRXD2ADDR(sc, 0), sc->sc_cddma);
1039 #endif
1040 
1041 	/* step 8.  Global Configuration & Interrupt Masks */
1042 
1043 	/* Disable weighted round robin. */
1044 	CAS_WRITE_4(sc, CAS_CAW, CAS_CAW_RR_DIS);
1045 
1046 	/*
1047 	 * Enable infinite bursts for revisions without PCI issues if
1048 	 * applicable.  Doing so greatly improves the TX performance.
1049 	 */
1050 	CAS_WRITE_4(sc, CAS_INF_BURST,
1051 	    (sc->sc_flags & CAS_TABORT) == 0 ? CAS_INF_BURST_EN :
1052 	    0);
1053 
1054 	/* Set up interrupts. */
1055 	CAS_WRITE_4(sc, CAS_INTMASK,
1056 	    ~(CAS_INTR_TX_INT_ME | CAS_INTR_TX_TAG_ERR |
1057 	    CAS_INTR_RX_DONE | CAS_INTR_RX_BUF_NA | CAS_INTR_RX_TAG_ERR |
1058 	    CAS_INTR_RX_COMP_FULL | CAS_INTR_RX_BUF_AEMPTY |
1059 	    CAS_INTR_RX_COMP_AFULL | CAS_INTR_RX_LEN_MMATCH |
1060 	    CAS_INTR_PCI_ERROR_INT
1061 #ifdef CAS_DEBUG
1062 	    | CAS_INTR_PCS_INT | CAS_INTR_MIF
1063 #endif
1064 	    ));
1065 	/* Don't clear top level interrupts when CAS_STATUS_ALIAS is read. */
1066 	CAS_WRITE_4(sc, CAS_CLEAR_ALIAS, 0);
1067 	CAS_WRITE_4(sc, CAS_MAC_RX_MASK, ~CAS_MAC_RX_OVERFLOW);
1068 	CAS_WRITE_4(sc, CAS_MAC_TX_MASK,
1069 	    ~(CAS_MAC_TX_UNDERRUN | CAS_MAC_TX_MAX_PKT_ERR));
1070 #ifdef CAS_DEBUG
1071 	CAS_WRITE_4(sc, CAS_MAC_CTRL_MASK,
1072 	    ~(CAS_MAC_CTRL_PAUSE_RCVD | CAS_MAC_CTRL_PAUSE |
1073 	    CAS_MAC_CTRL_NON_PAUSE));
1074 #else
1075 	CAS_WRITE_4(sc, CAS_MAC_CTRL_MASK,
1076 	    CAS_MAC_CTRL_PAUSE_RCVD | CAS_MAC_CTRL_PAUSE |
1077 	    CAS_MAC_CTRL_NON_PAUSE);
1078 #endif
1079 
1080 	/* Enable PCI error interrupts. */
1081 	CAS_WRITE_4(sc, CAS_ERROR_MASK,
1082 	    ~(CAS_ERROR_DTRTO | CAS_ERROR_OTHER | CAS_ERROR_DMAW_ZERO |
1083 	    CAS_ERROR_DMAR_ZERO | CAS_ERROR_RTRTO));
1084 
1085 	/* Enable PCI error interrupts in BIM configuration. */
1086 	CAS_WRITE_4(sc, CAS_BIM_CONF,
1087 	    CAS_BIM_CONF_DPAR_EN | CAS_BIM_CONF_RMA_EN | CAS_BIM_CONF_RTA_EN);
1088 
1089 	/*
1090 	 * step 9.  ETX Configuration: encode receive descriptor ring size,
1091 	 * enable DMA and disable pre-interrupt writeback completion.
1092 	 */
1093 	v = cas_descsize(CAS_NTXDESC) << CAS_TX_CONF_DESC3_SHFT;
1094 	CAS_WRITE_4(sc, CAS_TX_CONF, v | CAS_TX_CONF_TXDMA_EN |
1095 	    CAS_TX_CONF_RDPP_DIS | CAS_TX_CONF_PICWB_DIS);
1096 
1097 	/* step 10.  ERX Configuration */
1098 
1099 	/*
1100 	 * Encode receive completion and descriptor ring sizes, set the
1101 	 * swivel offset.
1102 	 */
1103 	v = cas_rxcompsize(CAS_NRXCOMP) << CAS_RX_CONF_COMP_SHFT;
1104 	v |= cas_descsize(CAS_NRXDESC) << CAS_RX_CONF_DESC_SHFT;
1105 	if ((sc->sc_flags & CAS_REG_PLUS) != 0)
1106 		v |= cas_descsize(CAS_NRXDESC2) << CAS_RX_CONF_DESC2_SHFT;
1107 	CAS_WRITE_4(sc, CAS_RX_CONF,
1108 	    v | (ETHER_ALIGN << CAS_RX_CONF_SOFF_SHFT));
1109 
1110 	/* Set the PAUSE thresholds.  We use the maximum OFF threshold. */
1111 	CAS_WRITE_4(sc, CAS_RX_PTHRS,
1112 	    (111 << CAS_RX_PTHRS_XOFF_SHFT) | (15 << CAS_RX_PTHRS_XON_SHFT));
1113 
1114 	/* RX blanking */
1115 	CAS_WRITE_4(sc, CAS_RX_BLANK,
1116 	    (15 << CAS_RX_BLANK_TIME_SHFT) | (5 << CAS_RX_BLANK_PKTS_SHFT));
1117 
1118 	/* Set RX_COMP_AFULL threshold to half of the RX completions. */
1119 	CAS_WRITE_4(sc, CAS_RX_AEMPTY_THRS,
1120 	    (CAS_NRXCOMP / 2) << CAS_RX_AEMPTY_COMP_SHFT);
1121 
1122 	/* Initialize the RX page size register as appropriate for 8k. */
1123 	CAS_WRITE_4(sc, CAS_RX_PSZ,
1124 	    (CAS_RX_PSZ_8K << CAS_RX_PSZ_SHFT) |
1125 	    (4 << CAS_RX_PSZ_MB_CNT_SHFT) |
1126 	    (CAS_RX_PSZ_MB_STRD_2K << CAS_RX_PSZ_MB_STRD_SHFT) |
1127 	    (CAS_RX_PSZ_MB_OFF_64 << CAS_RX_PSZ_MB_OFF_SHFT));
1128 
1129 	/* Disable RX random early detection. */
1130 	CAS_WRITE_4(sc,	CAS_RX_RED, 0);
1131 
1132 	/* Zero the RX reassembly DMA table. */
1133 	for (v = 0; v <= CAS_RX_REAS_DMA_ADDR_LC; v++) {
1134 		CAS_WRITE_4(sc,	CAS_RX_REAS_DMA_ADDR, v);
1135 		CAS_WRITE_4(sc,	CAS_RX_REAS_DMA_DATA_LO, 0);
1136 		CAS_WRITE_4(sc,	CAS_RX_REAS_DMA_DATA_MD, 0);
1137 		CAS_WRITE_4(sc,	CAS_RX_REAS_DMA_DATA_HI, 0);
1138 	}
1139 
1140 	/* Ensure the RX control FIFO and RX IPP FIFO addresses are zero. */
1141 	CAS_WRITE_4(sc, CAS_RX_CTRL_FIFO, 0);
1142 	CAS_WRITE_4(sc, CAS_RX_IPP_ADDR, 0);
1143 
1144 	/* Finally, enable RX DMA. */
1145 	CAS_WRITE_4(sc, CAS_RX_CONF,
1146 	    CAS_READ_4(sc, CAS_RX_CONF) | CAS_RX_CONF_RXDMA_EN);
1147 
1148 	/* step 11.  Configure Media. */
1149 
1150 	/* step 12.  RX_MAC Configuration Register */
1151 	v = CAS_READ_4(sc, CAS_MAC_RX_CONF);
1152 	v &= ~(CAS_MAC_RX_CONF_STRPPAD | CAS_MAC_RX_CONF_EN);
1153 	v |= CAS_MAC_RX_CONF_STRPFCS;
1154 	sc->sc_mac_rxcfg = v;
1155 	/*
1156 	 * Clear the RX filter and reprogram it.  This will also set the
1157 	 * current RX MAC configuration and enable it.
1158 	 */
1159 	cas_setladrf(sc);
1160 
1161 	/* step 13.  TX_MAC Configuration Register */
1162 	v = CAS_READ_4(sc, CAS_MAC_TX_CONF);
1163 	v |= CAS_MAC_TX_CONF_EN;
1164 	(void)cas_disable_tx(sc);
1165 	CAS_WRITE_4(sc, CAS_MAC_TX_CONF, v);
1166 
1167 	/* step 14.  Issue Transmit Pending command. */
1168 
1169 	/* step 15.  Give the receiver a swift kick. */
1170 	CAS_WRITE_4(sc, CAS_RX_KICK, CAS_NRXDESC - 4);
1171 	CAS_WRITE_4(sc, CAS_RX_COMP_TAIL, 0);
1172 	if ((sc->sc_flags & CAS_REG_PLUS) != 0)
1173 		CAS_WRITE_4(sc, CAS_RX_KICK2, CAS_NRXDESC2 - 4);
1174 
1175 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1176 	ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1177 
1178 	mii_mediachg(sc->sc_mii);
1179 
1180 	/* Start the one second timer. */
1181 	sc->sc_wdog_timer = 0;
1182 	callout_reset(&sc->sc_tick_ch, hz, cas_tick, sc);
1183 }
1184 
1185 static int
1186 cas_load_txmbuf(struct cas_softc *sc, struct mbuf **m_head)
1187 {
1188 	bus_dma_segment_t txsegs[CAS_NTXSEGS];
1189 	struct cas_txsoft *txs;
1190 	struct ip *ip;
1191 	struct mbuf *m;
1192 	uint64_t cflags;
1193 	int error, nexttx, nsegs, offset, seg;
1194 
1195 	CAS_LOCK_ASSERT(sc, MA_OWNED);
1196 
1197 	/* Get a work queue entry. */
1198 	if ((txs = STAILQ_FIRST(&sc->sc_txfreeq)) == NULL) {
1199 		/* Ran out of descriptors. */
1200 		return (ENOBUFS);
1201 	}
1202 
1203 	cflags = 0;
1204 	if (((*m_head)->m_pkthdr.csum_flags & CAS_CSUM_FEATURES) != 0) {
1205 		if (M_WRITABLE(*m_head) == 0) {
1206 			m = m_dup(*m_head, M_NOWAIT);
1207 			m_freem(*m_head);
1208 			*m_head = m;
1209 			if (m == NULL)
1210 				return (ENOBUFS);
1211 		}
1212 		offset = sizeof(struct ether_header);
1213 		m = m_pullup(*m_head, offset + sizeof(struct ip));
1214 		if (m == NULL) {
1215 			*m_head = NULL;
1216 			return (ENOBUFS);
1217 		}
1218 		ip = (struct ip *)(mtod(m, caddr_t) + offset);
1219 		offset += (ip->ip_hl << 2);
1220 		cflags = (offset << CAS_TD_CKSUM_START_SHFT) |
1221 		    ((offset + m->m_pkthdr.csum_data) <<
1222 		    CAS_TD_CKSUM_STUFF_SHFT) | CAS_TD_CKSUM_EN;
1223 		*m_head = m;
1224 	}
1225 
1226 	error = bus_dmamap_load_mbuf_sg(sc->sc_tdmatag, txs->txs_dmamap,
1227 	    *m_head, txsegs, &nsegs, BUS_DMA_NOWAIT);
1228 	if (error == EFBIG) {
1229 		m = m_collapse(*m_head, M_NOWAIT, CAS_NTXSEGS);
1230 		if (m == NULL) {
1231 			m_freem(*m_head);
1232 			*m_head = NULL;
1233 			return (ENOBUFS);
1234 		}
1235 		*m_head = m;
1236 		error = bus_dmamap_load_mbuf_sg(sc->sc_tdmatag,
1237 		    txs->txs_dmamap, *m_head, txsegs, &nsegs,
1238 		    BUS_DMA_NOWAIT);
1239 		if (error != 0) {
1240 			m_freem(*m_head);
1241 			*m_head = NULL;
1242 			return (error);
1243 		}
1244 	} else if (error != 0)
1245 		return (error);
1246 	/* If nsegs is wrong then the stack is corrupt. */
1247 	KASSERT(nsegs <= CAS_NTXSEGS,
1248 	    ("%s: too many DMA segments (%d)", __func__, nsegs));
1249 	if (nsegs == 0) {
1250 		m_freem(*m_head);
1251 		*m_head = NULL;
1252 		return (EIO);
1253 	}
1254 
1255 	/*
1256 	 * Ensure we have enough descriptors free to describe
1257 	 * the packet.  Note, we always reserve one descriptor
1258 	 * at the end of the ring as a termination point, in
1259 	 * order to prevent wrap-around.
1260 	 */
1261 	if (nsegs > sc->sc_txfree - 1) {
1262 		txs->txs_ndescs = 0;
1263 		bus_dmamap_unload(sc->sc_tdmatag, txs->txs_dmamap);
1264 		return (ENOBUFS);
1265 	}
1266 
1267 	txs->txs_ndescs = nsegs;
1268 	txs->txs_firstdesc = sc->sc_txnext;
1269 	nexttx = txs->txs_firstdesc;
1270 	for (seg = 0; seg < nsegs; seg++, nexttx = CAS_NEXTTX(nexttx)) {
1271 #ifdef CAS_DEBUG
1272 		CTR6(KTR_CAS,
1273 		    "%s: mapping seg %d (txd %d), len %lx, addr %#lx (%#lx)",
1274 		    __func__, seg, nexttx, txsegs[seg].ds_len,
1275 		    txsegs[seg].ds_addr, htole64(txsegs[seg].ds_addr));
1276 #endif
1277 		sc->sc_txdescs[nexttx].cd_buf_ptr =
1278 		    htole64(txsegs[seg].ds_addr);
1279 		KASSERT(txsegs[seg].ds_len <
1280 		    CAS_TD_BUF_LEN_MASK >> CAS_TD_BUF_LEN_SHFT,
1281 		    ("%s: segment size too large!", __func__));
1282 		sc->sc_txdescs[nexttx].cd_flags =
1283 		    htole64(txsegs[seg].ds_len << CAS_TD_BUF_LEN_SHFT);
1284 		txs->txs_lastdesc = nexttx;
1285 	}
1286 
1287 	/* Set EOF on the last descriptor. */
1288 #ifdef CAS_DEBUG
1289 	CTR3(KTR_CAS, "%s: end of frame at segment %d, TX %d",
1290 	    __func__, seg, nexttx);
1291 #endif
1292 	sc->sc_txdescs[txs->txs_lastdesc].cd_flags |=
1293 	    htole64(CAS_TD_END_OF_FRAME);
1294 
1295 	/* Lastly set SOF on the first descriptor. */
1296 #ifdef CAS_DEBUG
1297 	CTR3(KTR_CAS, "%s: start of frame at segment %d, TX %d",
1298 	    __func__, seg, nexttx);
1299 #endif
1300 	if (sc->sc_txwin += nsegs > CAS_MAXTXFREE * 2 / 3) {
1301 		sc->sc_txwin = 0;
1302 		sc->sc_txdescs[txs->txs_firstdesc].cd_flags |=
1303 		    htole64(cflags | CAS_TD_START_OF_FRAME | CAS_TD_INT_ME);
1304 	} else
1305 		sc->sc_txdescs[txs->txs_firstdesc].cd_flags |=
1306 		    htole64(cflags | CAS_TD_START_OF_FRAME);
1307 
1308 	/* Sync the DMA map. */
1309 	bus_dmamap_sync(sc->sc_tdmatag, txs->txs_dmamap,
1310 	    BUS_DMASYNC_PREWRITE);
1311 
1312 #ifdef CAS_DEBUG
1313 	CTR4(KTR_CAS, "%s: setting firstdesc=%d, lastdesc=%d, ndescs=%d",
1314 	    __func__, txs->txs_firstdesc, txs->txs_lastdesc,
1315 	    txs->txs_ndescs);
1316 #endif
1317 	STAILQ_REMOVE_HEAD(&sc->sc_txfreeq, txs_q);
1318 	STAILQ_INSERT_TAIL(&sc->sc_txdirtyq, txs, txs_q);
1319 	txs->txs_mbuf = *m_head;
1320 
1321 	sc->sc_txnext = CAS_NEXTTX(txs->txs_lastdesc);
1322 	sc->sc_txfree -= txs->txs_ndescs;
1323 
1324 	return (0);
1325 }
1326 
1327 static void
1328 cas_init_regs(struct cas_softc *sc)
1329 {
1330 	int i;
1331 	const u_char *laddr = IF_LLADDR(sc->sc_ifp);
1332 
1333 	CAS_LOCK_ASSERT(sc, MA_OWNED);
1334 
1335 	/* These registers are not cleared on reset. */
1336 	if ((sc->sc_flags & CAS_INITED) == 0) {
1337 		/* magic values */
1338 		CAS_WRITE_4(sc, CAS_MAC_IPG0, 0);
1339 		CAS_WRITE_4(sc, CAS_MAC_IPG1, 8);
1340 		CAS_WRITE_4(sc, CAS_MAC_IPG2, 4);
1341 
1342 		/* min frame length */
1343 		CAS_WRITE_4(sc, CAS_MAC_MIN_FRAME, ETHER_MIN_LEN);
1344 		/* max frame length and max burst size */
1345 		CAS_WRITE_4(sc, CAS_MAC_MAX_BF,
1346 		    ((ETHER_MAX_LEN_JUMBO + ETHER_VLAN_ENCAP_LEN) <<
1347 		    CAS_MAC_MAX_BF_FRM_SHFT) |
1348 		    (0x2000 << CAS_MAC_MAX_BF_BST_SHFT));
1349 
1350 		/* more magic values */
1351 		CAS_WRITE_4(sc, CAS_MAC_PREAMBLE_LEN, 0x7);
1352 		CAS_WRITE_4(sc, CAS_MAC_JAM_SIZE, 0x4);
1353 		CAS_WRITE_4(sc, CAS_MAC_ATTEMPT_LIMIT, 0x10);
1354 		CAS_WRITE_4(sc, CAS_MAC_CTRL_TYPE, 0x8808);
1355 
1356 		/* random number seed */
1357 		CAS_WRITE_4(sc, CAS_MAC_RANDOM_SEED,
1358 		    ((laddr[5] << 8) | laddr[4]) & 0x3ff);
1359 
1360 		/* secondary MAC addresses: 0:0:0:0:0:0 */
1361 		for (i = CAS_MAC_ADDR3; i <= CAS_MAC_ADDR41;
1362 		    i += CAS_MAC_ADDR4 - CAS_MAC_ADDR3)
1363 			CAS_WRITE_4(sc, i, 0);
1364 
1365 		/* MAC control address: 01:80:c2:00:00:01 */
1366 		CAS_WRITE_4(sc, CAS_MAC_ADDR42, 0x0001);
1367 		CAS_WRITE_4(sc, CAS_MAC_ADDR43, 0xc200);
1368 		CAS_WRITE_4(sc, CAS_MAC_ADDR44, 0x0180);
1369 
1370 		/* MAC filter address: 0:0:0:0:0:0 */
1371 		CAS_WRITE_4(sc, CAS_MAC_AFILTER0, 0);
1372 		CAS_WRITE_4(sc, CAS_MAC_AFILTER1, 0);
1373 		CAS_WRITE_4(sc, CAS_MAC_AFILTER2, 0);
1374 		CAS_WRITE_4(sc, CAS_MAC_AFILTER_MASK1_2, 0);
1375 		CAS_WRITE_4(sc, CAS_MAC_AFILTER_MASK0, 0);
1376 
1377 		/* Zero the hash table. */
1378 		for (i = CAS_MAC_HASH0; i <= CAS_MAC_HASH15;
1379 		    i += CAS_MAC_HASH1 - CAS_MAC_HASH0)
1380 			CAS_WRITE_4(sc, i, 0);
1381 
1382 		sc->sc_flags |= CAS_INITED;
1383 	}
1384 
1385 	/* Counters need to be zeroed. */
1386 	CAS_WRITE_4(sc, CAS_MAC_NORM_COLL_CNT, 0);
1387 	CAS_WRITE_4(sc, CAS_MAC_FIRST_COLL_CNT, 0);
1388 	CAS_WRITE_4(sc, CAS_MAC_EXCESS_COLL_CNT, 0);
1389 	CAS_WRITE_4(sc, CAS_MAC_LATE_COLL_CNT, 0);
1390 	CAS_WRITE_4(sc, CAS_MAC_DEFER_TMR_CNT, 0);
1391 	CAS_WRITE_4(sc, CAS_MAC_PEAK_ATTEMPTS, 0);
1392 	CAS_WRITE_4(sc, CAS_MAC_RX_FRAME_COUNT, 0);
1393 	CAS_WRITE_4(sc, CAS_MAC_RX_LEN_ERR_CNT, 0);
1394 	CAS_WRITE_4(sc, CAS_MAC_RX_ALIGN_ERR, 0);
1395 	CAS_WRITE_4(sc, CAS_MAC_RX_CRC_ERR_CNT, 0);
1396 	CAS_WRITE_4(sc, CAS_MAC_RX_CODE_VIOL, 0);
1397 
1398 	/* Set XOFF PAUSE time. */
1399 	CAS_WRITE_4(sc, CAS_MAC_SPC, 0x1BF0 << CAS_MAC_SPC_TIME_SHFT);
1400 
1401 	/* Set the station address. */
1402 	CAS_WRITE_4(sc, CAS_MAC_ADDR0, (laddr[4] << 8) | laddr[5]);
1403 	CAS_WRITE_4(sc, CAS_MAC_ADDR1, (laddr[2] << 8) | laddr[3]);
1404 	CAS_WRITE_4(sc, CAS_MAC_ADDR2, (laddr[0] << 8) | laddr[1]);
1405 
1406 	/* Enable MII outputs. */
1407 	CAS_WRITE_4(sc, CAS_MAC_XIF_CONF, CAS_MAC_XIF_CONF_TX_OE);
1408 }
1409 
1410 static void
1411 cas_tx_task(void *arg, int pending __unused)
1412 {
1413 	struct ifnet *ifp;
1414 
1415 	ifp = (struct ifnet *)arg;
1416 	cas_start(ifp);
1417 }
1418 
1419 static inline void
1420 cas_txkick(struct cas_softc *sc)
1421 {
1422 
1423 	/*
1424 	 * Update the TX kick register.  This register has to point to the
1425 	 * descriptor after the last valid one and for optimum performance
1426 	 * should be incremented in multiples of 4 (the DMA engine fetches/
1427 	 * updates descriptors in batches of 4).
1428 	 */
1429 #ifdef CAS_DEBUG
1430 	CTR3(KTR_CAS, "%s: %s: kicking TX %d",
1431 	    device_get_name(sc->sc_dev), __func__, sc->sc_txnext);
1432 #endif
1433 	CAS_CDSYNC(sc, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1434 	CAS_WRITE_4(sc, CAS_TX_KICK3, sc->sc_txnext);
1435 }
1436 
1437 static void
1438 cas_start(struct ifnet *ifp)
1439 {
1440 	struct cas_softc *sc = ifp->if_softc;
1441 	struct mbuf *m;
1442 	int kicked, ntx;
1443 
1444 	CAS_LOCK(sc);
1445 
1446 	if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
1447 	    IFF_DRV_RUNNING || (sc->sc_flags & CAS_LINK) == 0) {
1448 		CAS_UNLOCK(sc);
1449 		return;
1450 	}
1451 
1452 	if (sc->sc_txfree < CAS_MAXTXFREE / 4)
1453 		cas_tint(sc);
1454 
1455 #ifdef CAS_DEBUG
1456 	CTR4(KTR_CAS, "%s: %s: txfree %d, txnext %d",
1457 	    device_get_name(sc->sc_dev), __func__, sc->sc_txfree,
1458 	    sc->sc_txnext);
1459 #endif
1460 	ntx = 0;
1461 	kicked = 0;
1462 	for (; !IFQ_DRV_IS_EMPTY(&ifp->if_snd) && sc->sc_txfree > 1;) {
1463 		IFQ_DRV_DEQUEUE(&ifp->if_snd, m);
1464 		if (m == NULL)
1465 			break;
1466 		if (cas_load_txmbuf(sc, &m) != 0) {
1467 			if (m == NULL)
1468 				break;
1469 			ifp->if_drv_flags |= IFF_DRV_OACTIVE;
1470 			IFQ_DRV_PREPEND(&ifp->if_snd, m);
1471 			break;
1472 		}
1473 		if ((sc->sc_txnext % 4) == 0) {
1474 			cas_txkick(sc);
1475 			kicked = 1;
1476 		} else
1477 			kicked = 0;
1478 		ntx++;
1479 		BPF_MTAP(ifp, m);
1480 	}
1481 
1482 	if (ntx > 0) {
1483 		if (kicked == 0)
1484 			cas_txkick(sc);
1485 #ifdef CAS_DEBUG
1486 		CTR2(KTR_CAS, "%s: packets enqueued, OWN on %d",
1487 		    device_get_name(sc->sc_dev), sc->sc_txnext);
1488 #endif
1489 
1490 		/* Set a watchdog timer in case the chip flakes out. */
1491 		sc->sc_wdog_timer = 5;
1492 #ifdef CAS_DEBUG
1493 		CTR3(KTR_CAS, "%s: %s: watchdog %d",
1494 		    device_get_name(sc->sc_dev), __func__,
1495 		    sc->sc_wdog_timer);
1496 #endif
1497 	}
1498 
1499 	CAS_UNLOCK(sc);
1500 }
1501 
1502 static void
1503 cas_tint(struct cas_softc *sc)
1504 {
1505 	struct ifnet *ifp = sc->sc_ifp;
1506 	struct cas_txsoft *txs;
1507 	int progress;
1508 	uint32_t txlast;
1509 #ifdef CAS_DEBUG
1510 	int i;
1511 
1512 	CAS_LOCK_ASSERT(sc, MA_OWNED);
1513 
1514 	CTR2(KTR_CAS, "%s: %s", device_get_name(sc->sc_dev), __func__);
1515 #endif
1516 
1517 	/*
1518 	 * Go through our TX list and free mbufs for those
1519 	 * frames that have been transmitted.
1520 	 */
1521 	progress = 0;
1522 	CAS_CDSYNC(sc, BUS_DMASYNC_POSTREAD);
1523 	while ((txs = STAILQ_FIRST(&sc->sc_txdirtyq)) != NULL) {
1524 #ifdef CAS_DEBUG
1525 		if ((ifp->if_flags & IFF_DEBUG) != 0) {
1526 			printf("    txsoft %p transmit chain:\n", txs);
1527 			for (i = txs->txs_firstdesc;; i = CAS_NEXTTX(i)) {
1528 				printf("descriptor %d: ", i);
1529 				printf("cd_flags: 0x%016llx\t",
1530 				    (long long)le64toh(
1531 				    sc->sc_txdescs[i].cd_flags));
1532 				printf("cd_buf_ptr: 0x%016llx\n",
1533 				    (long long)le64toh(
1534 				    sc->sc_txdescs[i].cd_buf_ptr));
1535 				if (i == txs->txs_lastdesc)
1536 					break;
1537 			}
1538 		}
1539 #endif
1540 
1541 		/*
1542 		 * In theory, we could harvest some descriptors before
1543 		 * the ring is empty, but that's a bit complicated.
1544 		 *
1545 		 * CAS_TX_COMPn points to the last descriptor
1546 		 * processed + 1.
1547 		 */
1548 		txlast = CAS_READ_4(sc, CAS_TX_COMP3);
1549 #ifdef CAS_DEBUG
1550 		CTR4(KTR_CAS, "%s: txs->txs_firstdesc = %d, "
1551 		    "txs->txs_lastdesc = %d, txlast = %d",
1552 		    __func__, txs->txs_firstdesc, txs->txs_lastdesc, txlast);
1553 #endif
1554 		if (txs->txs_firstdesc <= txs->txs_lastdesc) {
1555 			if ((txlast >= txs->txs_firstdesc) &&
1556 			    (txlast <= txs->txs_lastdesc))
1557 				break;
1558 		} else {
1559 			/* Ick -- this command wraps. */
1560 			if ((txlast >= txs->txs_firstdesc) ||
1561 			    (txlast <= txs->txs_lastdesc))
1562 				break;
1563 		}
1564 
1565 #ifdef CAS_DEBUG
1566 		CTR1(KTR_CAS, "%s: releasing a descriptor", __func__);
1567 #endif
1568 		STAILQ_REMOVE_HEAD(&sc->sc_txdirtyq, txs_q);
1569 
1570 		sc->sc_txfree += txs->txs_ndescs;
1571 
1572 		bus_dmamap_sync(sc->sc_tdmatag, txs->txs_dmamap,
1573 		    BUS_DMASYNC_POSTWRITE);
1574 		bus_dmamap_unload(sc->sc_tdmatag, txs->txs_dmamap);
1575 		if (txs->txs_mbuf != NULL) {
1576 			m_freem(txs->txs_mbuf);
1577 			txs->txs_mbuf = NULL;
1578 		}
1579 
1580 		STAILQ_INSERT_TAIL(&sc->sc_txfreeq, txs, txs_q);
1581 
1582 		if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
1583 		progress = 1;
1584 	}
1585 
1586 #ifdef CAS_DEBUG
1587 	CTR5(KTR_CAS, "%s: CAS_TX_SM1 %x CAS_TX_SM2 %x CAS_TX_DESC_BASE %llx "
1588 	    "CAS_TX_COMP3 %x",
1589 	    __func__, CAS_READ_4(sc, CAS_TX_SM1), CAS_READ_4(sc, CAS_TX_SM2),
1590 	    ((long long)CAS_READ_4(sc, CAS_TX_DESC3_BASE_HI) << 32) |
1591 	    CAS_READ_4(sc, CAS_TX_DESC3_BASE_LO),
1592 	    CAS_READ_4(sc, CAS_TX_COMP3));
1593 #endif
1594 
1595 	if (progress) {
1596 		/* We freed some descriptors, so reset IFF_DRV_OACTIVE. */
1597 		ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1598 		if (STAILQ_EMPTY(&sc->sc_txdirtyq))
1599 			sc->sc_wdog_timer = 0;
1600 	}
1601 
1602 #ifdef CAS_DEBUG
1603 	CTR3(KTR_CAS, "%s: %s: watchdog %d",
1604 	    device_get_name(sc->sc_dev), __func__, sc->sc_wdog_timer);
1605 #endif
1606 }
1607 
1608 static void
1609 cas_rint_timeout(void *arg)
1610 {
1611 	struct epoch_tracker et;
1612 	struct cas_softc *sc = arg;
1613 
1614 	CAS_LOCK_ASSERT(sc, MA_OWNED);
1615 
1616 	NET_EPOCH_ENTER(et);
1617 	cas_rint(sc);
1618 	NET_EPOCH_EXIT(et);
1619 }
1620 
1621 static void
1622 cas_rint(struct cas_softc *sc)
1623 {
1624 	struct cas_rxdsoft *rxds, *rxds2;
1625 	struct ifnet *ifp = sc->sc_ifp;
1626 	struct mbuf *m, *m2;
1627 	uint64_t word1, word2, word3, word4;
1628 	uint32_t rxhead;
1629 	u_int idx, idx2, len, off, skip;
1630 
1631 	CAS_LOCK_ASSERT(sc, MA_OWNED);
1632 
1633 	callout_stop(&sc->sc_rx_ch);
1634 
1635 #ifdef CAS_DEBUG
1636 	CTR2(KTR_CAS, "%s: %s", device_get_name(sc->sc_dev), __func__);
1637 #endif
1638 
1639 #define	PRINTWORD(n, delimiter)						\
1640 	printf("word ## n: 0x%016llx%c", (long long)word ## n, delimiter)
1641 
1642 #define	SKIPASSERT(n)							\
1643 	KASSERT(sc->sc_rxcomps[sc->sc_rxcptr].crc_word ## n == 0,	\
1644 	    ("%s: word ## n not 0", __func__))
1645 
1646 #define	WORDTOH(n)							\
1647 	word ## n = le64toh(sc->sc_rxcomps[sc->sc_rxcptr].crc_word ## n)
1648 
1649 	/*
1650 	 * Read the completion head register once.  This limits
1651 	 * how long the following loop can execute.
1652 	 */
1653 	rxhead = CAS_READ_4(sc, CAS_RX_COMP_HEAD);
1654 #ifdef CAS_DEBUG
1655 	CTR4(KTR_CAS, "%s: sc->sc_rxcptr %d, sc->sc_rxdptr %d, head %d",
1656 	    __func__, sc->sc_rxcptr, sc->sc_rxdptr, rxhead);
1657 #endif
1658 	skip = 0;
1659 	CAS_CDSYNC(sc, BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
1660 	for (; sc->sc_rxcptr != rxhead;
1661 	    sc->sc_rxcptr = CAS_NEXTRXCOMP(sc->sc_rxcptr)) {
1662 		if (skip != 0) {
1663 			SKIPASSERT(1);
1664 			SKIPASSERT(2);
1665 			SKIPASSERT(3);
1666 
1667 			--skip;
1668 			goto skip;
1669 		}
1670 
1671 		WORDTOH(1);
1672 		WORDTOH(2);
1673 		WORDTOH(3);
1674 		WORDTOH(4);
1675 
1676 #ifdef CAS_DEBUG
1677 		if ((ifp->if_flags & IFF_DEBUG) != 0) {
1678 			printf("    completion %d: ", sc->sc_rxcptr);
1679 			PRINTWORD(1, '\t');
1680 			PRINTWORD(2, '\t');
1681 			PRINTWORD(3, '\t');
1682 			PRINTWORD(4, '\n');
1683 		}
1684 #endif
1685 
1686 		if (__predict_false(
1687 		    (word1 & CAS_RC1_TYPE_MASK) == CAS_RC1_TYPE_HW ||
1688 		    (word4 & CAS_RC4_ZERO) != 0)) {
1689 			/*
1690 			 * The descriptor is still marked as owned, although
1691 			 * it is supposed to have completed.  This has been
1692 			 * observed on some machines.  Just exiting here
1693 			 * might leave the packet sitting around until another
1694 			 * one arrives to trigger a new interrupt, which is
1695 			 * generally undesirable, so set up a timeout.
1696 			 */
1697 			callout_reset(&sc->sc_rx_ch, CAS_RXOWN_TICKS,
1698 			    cas_rint_timeout, sc);
1699 			break;
1700 		}
1701 
1702 		if (__predict_false(
1703 		    (word4 & (CAS_RC4_BAD | CAS_RC4_LEN_MMATCH)) != 0)) {
1704 			if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
1705 			device_printf(sc->sc_dev,
1706 			    "receive error: CRC error\n");
1707 			continue;
1708 		}
1709 
1710 		KASSERT(CAS_GET(word1, CAS_RC1_DATA_SIZE) == 0 ||
1711 		    CAS_GET(word2, CAS_RC2_HDR_SIZE) == 0,
1712 		    ("%s: data and header present", __func__));
1713 		KASSERT((word1 & CAS_RC1_SPLIT_PKT) == 0 ||
1714 		    CAS_GET(word2, CAS_RC2_HDR_SIZE) == 0,
1715 		    ("%s: split and header present", __func__));
1716 		KASSERT(CAS_GET(word1, CAS_RC1_DATA_SIZE) == 0 ||
1717 		    (word1 & CAS_RC1_RELEASE_HDR) == 0,
1718 		    ("%s: data present but header release", __func__));
1719 		KASSERT(CAS_GET(word2, CAS_RC2_HDR_SIZE) == 0 ||
1720 		    (word1 & CAS_RC1_RELEASE_DATA) == 0,
1721 		    ("%s: header present but data release", __func__));
1722 
1723 		if ((len = CAS_GET(word2, CAS_RC2_HDR_SIZE)) != 0) {
1724 			idx = CAS_GET(word2, CAS_RC2_HDR_INDEX);
1725 			off = CAS_GET(word2, CAS_RC2_HDR_OFF);
1726 #ifdef CAS_DEBUG
1727 			CTR4(KTR_CAS, "%s: hdr at idx %d, off %d, len %d",
1728 			    __func__, idx, off, len);
1729 #endif
1730 			rxds = &sc->sc_rxdsoft[idx];
1731 			MGETHDR(m, M_NOWAIT, MT_DATA);
1732 			if (m != NULL) {
1733 				refcount_acquire(&rxds->rxds_refcount);
1734 				bus_dmamap_sync(sc->sc_rdmatag,
1735 				    rxds->rxds_dmamap, BUS_DMASYNC_POSTREAD);
1736 				m_extadd(m, (char *)rxds->rxds_buf +
1737 				    off * 256 + ETHER_ALIGN, len, cas_free,
1738 				    sc, (void *)(uintptr_t)idx,
1739 				    M_RDONLY, EXT_NET_DRV);
1740 				if ((m->m_flags & M_EXT) == 0) {
1741 					m_freem(m);
1742 					m = NULL;
1743 				}
1744 			}
1745 			if (m != NULL) {
1746 				m->m_pkthdr.rcvif = ifp;
1747 				m->m_pkthdr.len = m->m_len = len;
1748 				if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
1749 				if ((ifp->if_capenable & IFCAP_RXCSUM) != 0)
1750 					cas_rxcksum(m, CAS_GET(word4,
1751 					    CAS_RC4_TCP_CSUM));
1752 				/* Pass it on. */
1753 				CAS_UNLOCK(sc);
1754 				(*ifp->if_input)(ifp, m);
1755 				CAS_LOCK(sc);
1756 			} else
1757 				if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1758 
1759 			if ((word1 & CAS_RC1_RELEASE_HDR) != 0 &&
1760 			    refcount_release(&rxds->rxds_refcount) != 0)
1761 				cas_add_rxdesc(sc, idx);
1762 		} else if ((len = CAS_GET(word1, CAS_RC1_DATA_SIZE)) != 0) {
1763 			idx = CAS_GET(word1, CAS_RC1_DATA_INDEX);
1764 			off = CAS_GET(word1, CAS_RC1_DATA_OFF);
1765 #ifdef CAS_DEBUG
1766 			CTR4(KTR_CAS, "%s: data at idx %d, off %d, len %d",
1767 			    __func__, idx, off, len);
1768 #endif
1769 			rxds = &sc->sc_rxdsoft[idx];
1770 			MGETHDR(m, M_NOWAIT, MT_DATA);
1771 			if (m != NULL) {
1772 				refcount_acquire(&rxds->rxds_refcount);
1773 				off += ETHER_ALIGN;
1774 				m->m_len = min(CAS_PAGE_SIZE - off, len);
1775 				bus_dmamap_sync(sc->sc_rdmatag,
1776 				    rxds->rxds_dmamap, BUS_DMASYNC_POSTREAD);
1777 				m_extadd(m, (char *)rxds->rxds_buf + off,
1778 				    m->m_len, cas_free, sc,
1779 				    (void *)(uintptr_t)idx, M_RDONLY,
1780 				    EXT_NET_DRV);
1781 				if ((m->m_flags & M_EXT) == 0) {
1782 					m_freem(m);
1783 					m = NULL;
1784 				}
1785 			}
1786 			idx2 = 0;
1787 			m2 = NULL;
1788 			rxds2 = NULL;
1789 			if ((word1 & CAS_RC1_SPLIT_PKT) != 0) {
1790 				KASSERT((word1 & CAS_RC1_RELEASE_NEXT) != 0,
1791 				    ("%s: split but no release next",
1792 				    __func__));
1793 
1794 				idx2 = CAS_GET(word2, CAS_RC2_NEXT_INDEX);
1795 #ifdef CAS_DEBUG
1796 				CTR2(KTR_CAS, "%s: split at idx %d",
1797 				    __func__, idx2);
1798 #endif
1799 				rxds2 = &sc->sc_rxdsoft[idx2];
1800 				if (m != NULL) {
1801 					MGET(m2, M_NOWAIT, MT_DATA);
1802 					if (m2 != NULL) {
1803 						refcount_acquire(
1804 						    &rxds2->rxds_refcount);
1805 						m2->m_len = len - m->m_len;
1806 						bus_dmamap_sync(
1807 						    sc->sc_rdmatag,
1808 						    rxds2->rxds_dmamap,
1809 						    BUS_DMASYNC_POSTREAD);
1810 						m_extadd(m2,
1811 						    (char *)rxds2->rxds_buf,
1812 						    m2->m_len, cas_free, sc,
1813 						    (void *)(uintptr_t)idx2,
1814 						    M_RDONLY, EXT_NET_DRV);
1815 						if ((m2->m_flags & M_EXT) ==
1816 						    0) {
1817 							m_freem(m2);
1818 							m2 = NULL;
1819 						}
1820 					}
1821 				}
1822 				if (m2 != NULL)
1823 					m->m_next = m2;
1824 				else if (m != NULL) {
1825 					m_freem(m);
1826 					m = NULL;
1827 				}
1828 			}
1829 			if (m != NULL) {
1830 				m->m_pkthdr.rcvif = ifp;
1831 				m->m_pkthdr.len = len;
1832 				if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
1833 				if ((ifp->if_capenable & IFCAP_RXCSUM) != 0)
1834 					cas_rxcksum(m, CAS_GET(word4,
1835 					    CAS_RC4_TCP_CSUM));
1836 				/* Pass it on. */
1837 				CAS_UNLOCK(sc);
1838 				(*ifp->if_input)(ifp, m);
1839 				CAS_LOCK(sc);
1840 			} else
1841 				if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1842 
1843 			if ((word1 & CAS_RC1_RELEASE_DATA) != 0 &&
1844 			    refcount_release(&rxds->rxds_refcount) != 0)
1845 				cas_add_rxdesc(sc, idx);
1846 			if ((word1 & CAS_RC1_SPLIT_PKT) != 0 &&
1847 			    refcount_release(&rxds2->rxds_refcount) != 0)
1848 				cas_add_rxdesc(sc, idx2);
1849 		}
1850 
1851 		skip = CAS_GET(word1, CAS_RC1_SKIP);
1852 
1853  skip:
1854 		cas_rxcompinit(&sc->sc_rxcomps[sc->sc_rxcptr]);
1855 		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1856 			break;
1857 	}
1858 	CAS_CDSYNC(sc, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1859 	CAS_WRITE_4(sc, CAS_RX_COMP_TAIL, sc->sc_rxcptr);
1860 
1861 #undef PRINTWORD
1862 #undef SKIPASSERT
1863 #undef WORDTOH
1864 
1865 #ifdef CAS_DEBUG
1866 	CTR4(KTR_CAS, "%s: done sc->sc_rxcptr %d, sc->sc_rxdptr %d, head %d",
1867 	    __func__, sc->sc_rxcptr, sc->sc_rxdptr,
1868 	    CAS_READ_4(sc, CAS_RX_COMP_HEAD));
1869 #endif
1870 }
1871 
1872 static void
1873 cas_free(struct mbuf *m)
1874 {
1875 	struct cas_rxdsoft *rxds;
1876 	struct cas_softc *sc;
1877 	u_int idx, locked;
1878 
1879 	sc = m->m_ext.ext_arg1;
1880 	idx = (uintptr_t)m->m_ext.ext_arg2;
1881 	rxds = &sc->sc_rxdsoft[idx];
1882 	if (refcount_release(&rxds->rxds_refcount) == 0)
1883 		return;
1884 
1885 	/*
1886 	 * NB: this function can be called via m_freem(9) within
1887 	 * this driver!
1888 	 */
1889 	if ((locked = CAS_LOCK_OWNED(sc)) == 0)
1890 		CAS_LOCK(sc);
1891 	cas_add_rxdesc(sc, idx);
1892 	if (locked == 0)
1893 		CAS_UNLOCK(sc);
1894 }
1895 
1896 static inline void
1897 cas_add_rxdesc(struct cas_softc *sc, u_int idx)
1898 {
1899 
1900 	CAS_LOCK_ASSERT(sc, MA_OWNED);
1901 
1902 	bus_dmamap_sync(sc->sc_rdmatag, sc->sc_rxdsoft[idx].rxds_dmamap,
1903 	    BUS_DMASYNC_PREREAD);
1904 	CAS_UPDATE_RXDESC(sc, sc->sc_rxdptr, idx);
1905 	sc->sc_rxdptr = CAS_NEXTRXDESC(sc->sc_rxdptr);
1906 
1907 	/*
1908 	 * Update the RX kick register.  This register has to point to the
1909 	 * descriptor after the last valid one (before the current batch)
1910 	 * and for optimum performance should be incremented in multiples
1911 	 * of 4 (the DMA engine fetches/updates descriptors in batches of 4).
1912 	 */
1913 	if ((sc->sc_rxdptr % 4) == 0) {
1914 		CAS_CDSYNC(sc, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1915 		CAS_WRITE_4(sc, CAS_RX_KICK,
1916 		    (sc->sc_rxdptr + CAS_NRXDESC - 4) & CAS_NRXDESC_MASK);
1917 	}
1918 }
1919 
1920 static void
1921 cas_eint(struct cas_softc *sc, u_int status)
1922 {
1923 	struct ifnet *ifp = sc->sc_ifp;
1924 
1925 	CAS_LOCK_ASSERT(sc, MA_OWNED);
1926 
1927 	if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
1928 
1929 	device_printf(sc->sc_dev, "%s: status 0x%x", __func__, status);
1930 	if ((status & CAS_INTR_PCI_ERROR_INT) != 0) {
1931 		status = CAS_READ_4(sc, CAS_ERROR_STATUS);
1932 		printf(", PCI bus error 0x%x", status);
1933 		if ((status & CAS_ERROR_OTHER) != 0) {
1934 			status = pci_read_config(sc->sc_dev, PCIR_STATUS, 2);
1935 			printf(", PCI status 0x%x", status);
1936 			pci_write_config(sc->sc_dev, PCIR_STATUS, status, 2);
1937 		}
1938 	}
1939 	printf("\n");
1940 
1941 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1942 	cas_init_locked(sc);
1943 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
1944 		taskqueue_enqueue(sc->sc_tq, &sc->sc_tx_task);
1945 }
1946 
1947 static int
1948 cas_intr(void *v)
1949 {
1950 	struct cas_softc *sc = v;
1951 
1952 	if (__predict_false((CAS_READ_4(sc, CAS_STATUS_ALIAS) &
1953 	    CAS_INTR_SUMMARY) == 0))
1954 		return (FILTER_STRAY);
1955 
1956 	/* Disable interrupts. */
1957 	CAS_WRITE_4(sc, CAS_INTMASK, 0xffffffff);
1958 	taskqueue_enqueue(sc->sc_tq, &sc->sc_intr_task);
1959 
1960 	return (FILTER_HANDLED);
1961 }
1962 
1963 static void
1964 cas_intr_task(void *arg, int pending __unused)
1965 {
1966 	struct cas_softc *sc = arg;
1967 	struct ifnet *ifp = sc->sc_ifp;
1968 	uint32_t status, status2;
1969 
1970 	CAS_LOCK_ASSERT(sc, MA_NOTOWNED);
1971 
1972 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1973 		return;
1974 
1975 	status = CAS_READ_4(sc, CAS_STATUS);
1976 	if (__predict_false((status & CAS_INTR_SUMMARY) == 0))
1977 		goto done;
1978 
1979 	CAS_LOCK(sc);
1980 #ifdef CAS_DEBUG
1981 	CTR4(KTR_CAS, "%s: %s: cplt %x, status %x",
1982 	    device_get_name(sc->sc_dev), __func__,
1983 	    (status >> CAS_STATUS_TX_COMP3_SHFT), (u_int)status);
1984 
1985 	/*
1986 	 * PCS interrupts must be cleared, otherwise no traffic is passed!
1987 	 */
1988 	if ((status & CAS_INTR_PCS_INT) != 0) {
1989 		status2 =
1990 		    CAS_READ_4(sc, CAS_PCS_INTR_STATUS) |
1991 		    CAS_READ_4(sc, CAS_PCS_INTR_STATUS);
1992 		if ((status2 & CAS_PCS_INTR_LINK) != 0)
1993 			device_printf(sc->sc_dev,
1994 			    "%s: PCS link status changed\n", __func__);
1995 	}
1996 	if ((status & CAS_MAC_CTRL_STATUS) != 0) {
1997 		status2 = CAS_READ_4(sc, CAS_MAC_CTRL_STATUS);
1998 		if ((status2 & CAS_MAC_CTRL_PAUSE) != 0)
1999 			device_printf(sc->sc_dev,
2000 			    "%s: PAUSE received (PAUSE time %d slots)\n",
2001 			    __func__,
2002 			    (status2 & CAS_MAC_CTRL_STATUS_PT_MASK) >>
2003 			    CAS_MAC_CTRL_STATUS_PT_SHFT);
2004 		if ((status2 & CAS_MAC_CTRL_PAUSE) != 0)
2005 			device_printf(sc->sc_dev,
2006 			    "%s: transited to PAUSE state\n", __func__);
2007 		if ((status2 & CAS_MAC_CTRL_NON_PAUSE) != 0)
2008 			device_printf(sc->sc_dev,
2009 			    "%s: transited to non-PAUSE state\n", __func__);
2010 	}
2011 	if ((status & CAS_INTR_MIF) != 0)
2012 		device_printf(sc->sc_dev, "%s: MIF interrupt\n", __func__);
2013 #endif
2014 
2015 	if (__predict_false((status &
2016 	    (CAS_INTR_TX_TAG_ERR | CAS_INTR_RX_TAG_ERR |
2017 	    CAS_INTR_RX_LEN_MMATCH | CAS_INTR_PCI_ERROR_INT)) != 0)) {
2018 		cas_eint(sc, status);
2019 		CAS_UNLOCK(sc);
2020 		return;
2021 	}
2022 
2023 	if (__predict_false(status & CAS_INTR_TX_MAC_INT)) {
2024 		status2 = CAS_READ_4(sc, CAS_MAC_TX_STATUS);
2025 		if ((status2 &
2026 		    (CAS_MAC_TX_UNDERRUN | CAS_MAC_TX_MAX_PKT_ERR)) != 0)
2027 			if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
2028 		else if ((status2 & ~CAS_MAC_TX_FRAME_XMTD) != 0)
2029 			device_printf(sc->sc_dev,
2030 			    "MAC TX fault, status %x\n", status2);
2031 	}
2032 
2033 	if (__predict_false(status & CAS_INTR_RX_MAC_INT)) {
2034 		status2 = CAS_READ_4(sc, CAS_MAC_RX_STATUS);
2035 		if ((status2 & CAS_MAC_RX_OVERFLOW) != 0)
2036 			if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
2037 		else if ((status2 & ~CAS_MAC_RX_FRAME_RCVD) != 0)
2038 			device_printf(sc->sc_dev,
2039 			    "MAC RX fault, status %x\n", status2);
2040 	}
2041 
2042 	if ((status &
2043 	    (CAS_INTR_RX_DONE | CAS_INTR_RX_BUF_NA | CAS_INTR_RX_COMP_FULL |
2044 	    CAS_INTR_RX_BUF_AEMPTY | CAS_INTR_RX_COMP_AFULL)) != 0) {
2045 		cas_rint(sc);
2046 #ifdef CAS_DEBUG
2047 		if (__predict_false((status &
2048 		    (CAS_INTR_RX_BUF_NA | CAS_INTR_RX_COMP_FULL |
2049 		    CAS_INTR_RX_BUF_AEMPTY | CAS_INTR_RX_COMP_AFULL)) != 0))
2050 			device_printf(sc->sc_dev,
2051 			    "RX fault, status %x\n", status);
2052 #endif
2053 	}
2054 
2055 	if ((status &
2056 	    (CAS_INTR_TX_INT_ME | CAS_INTR_TX_ALL | CAS_INTR_TX_DONE)) != 0)
2057 		cas_tint(sc);
2058 
2059 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2060 		CAS_UNLOCK(sc);
2061 		return;
2062 	} else if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2063 		taskqueue_enqueue(sc->sc_tq, &sc->sc_tx_task);
2064 	CAS_UNLOCK(sc);
2065 
2066 	status = CAS_READ_4(sc, CAS_STATUS_ALIAS);
2067 	if (__predict_false((status & CAS_INTR_SUMMARY) != 0)) {
2068 		taskqueue_enqueue(sc->sc_tq, &sc->sc_intr_task);
2069 		return;
2070 	}
2071 
2072  done:
2073 	/* Re-enable interrupts. */
2074 	CAS_WRITE_4(sc, CAS_INTMASK,
2075 	    ~(CAS_INTR_TX_INT_ME | CAS_INTR_TX_TAG_ERR |
2076 	    CAS_INTR_RX_DONE | CAS_INTR_RX_BUF_NA | CAS_INTR_RX_TAG_ERR |
2077 	    CAS_INTR_RX_COMP_FULL | CAS_INTR_RX_BUF_AEMPTY |
2078 	    CAS_INTR_RX_COMP_AFULL | CAS_INTR_RX_LEN_MMATCH |
2079 	    CAS_INTR_PCI_ERROR_INT
2080 #ifdef CAS_DEBUG
2081 	    | CAS_INTR_PCS_INT | CAS_INTR_MIF
2082 #endif
2083 	));
2084 }
2085 
2086 static void
2087 cas_watchdog(struct cas_softc *sc)
2088 {
2089 	struct ifnet *ifp = sc->sc_ifp;
2090 
2091 	CAS_LOCK_ASSERT(sc, MA_OWNED);
2092 
2093 #ifdef CAS_DEBUG
2094 	CTR4(KTR_CAS,
2095 	    "%s: CAS_RX_CONF %x CAS_MAC_RX_STATUS %x CAS_MAC_RX_CONF %x",
2096 	    __func__, CAS_READ_4(sc, CAS_RX_CONF),
2097 	    CAS_READ_4(sc, CAS_MAC_RX_STATUS),
2098 	    CAS_READ_4(sc, CAS_MAC_RX_CONF));
2099 	CTR4(KTR_CAS,
2100 	    "%s: CAS_TX_CONF %x CAS_MAC_TX_STATUS %x CAS_MAC_TX_CONF %x",
2101 	    __func__, CAS_READ_4(sc, CAS_TX_CONF),
2102 	    CAS_READ_4(sc, CAS_MAC_TX_STATUS),
2103 	    CAS_READ_4(sc, CAS_MAC_TX_CONF));
2104 #endif
2105 
2106 	if (sc->sc_wdog_timer == 0 || --sc->sc_wdog_timer != 0)
2107 		return;
2108 
2109 	if ((sc->sc_flags & CAS_LINK) != 0)
2110 		device_printf(sc->sc_dev, "device timeout\n");
2111 	else if (bootverbose)
2112 		device_printf(sc->sc_dev, "device timeout (no link)\n");
2113 	if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
2114 
2115 	/* Try to get more packets going. */
2116 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2117 	cas_init_locked(sc);
2118 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2119 		taskqueue_enqueue(sc->sc_tq, &sc->sc_tx_task);
2120 }
2121 
2122 static void
2123 cas_mifinit(struct cas_softc *sc)
2124 {
2125 
2126 	/* Configure the MIF in frame mode. */
2127 	CAS_WRITE_4(sc, CAS_MIF_CONF,
2128 	    CAS_READ_4(sc, CAS_MIF_CONF) & ~CAS_MIF_CONF_BB_MODE);
2129 	CAS_BARRIER(sc, CAS_MIF_CONF, 4,
2130 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2131 }
2132 
2133 /*
2134  * MII interface
2135  *
2136  * The MII interface supports at least three different operating modes:
2137  *
2138  * Bitbang mode is implemented using data, clock and output enable registers.
2139  *
2140  * Frame mode is implemented by loading a complete frame into the frame
2141  * register and polling the valid bit for completion.
2142  *
2143  * Polling mode uses the frame register but completion is indicated by
2144  * an interrupt.
2145  *
2146  */
2147 static int
2148 cas_mii_readreg(device_t dev, int phy, int reg)
2149 {
2150 	struct cas_softc *sc;
2151 	int n;
2152 	uint32_t v;
2153 
2154 #ifdef CAS_DEBUG_PHY
2155 	printf("%s: phy %d reg %d\n", __func__, phy, reg);
2156 #endif
2157 
2158 	sc = device_get_softc(dev);
2159 	if ((sc->sc_flags & CAS_SERDES) != 0) {
2160 		switch (reg) {
2161 		case MII_BMCR:
2162 			reg = CAS_PCS_CTRL;
2163 			break;
2164 		case MII_BMSR:
2165 			reg = CAS_PCS_STATUS;
2166 			break;
2167 		case MII_PHYIDR1:
2168 		case MII_PHYIDR2:
2169 			return (0);
2170 		case MII_ANAR:
2171 			reg = CAS_PCS_ANAR;
2172 			break;
2173 		case MII_ANLPAR:
2174 			reg = CAS_PCS_ANLPAR;
2175 			break;
2176 		case MII_EXTSR:
2177 			return (EXTSR_1000XFDX | EXTSR_1000XHDX);
2178 		default:
2179 			device_printf(sc->sc_dev,
2180 			    "%s: unhandled register %d\n", __func__, reg);
2181 			return (0);
2182 		}
2183 		return (CAS_READ_4(sc, reg));
2184 	}
2185 
2186 	/* Construct the frame command. */
2187 	v = CAS_MIF_FRAME_READ |
2188 	    (phy << CAS_MIF_FRAME_PHY_SHFT) |
2189 	    (reg << CAS_MIF_FRAME_REG_SHFT);
2190 
2191 	CAS_WRITE_4(sc, CAS_MIF_FRAME, v);
2192 	CAS_BARRIER(sc, CAS_MIF_FRAME, 4,
2193 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2194 	for (n = 0; n < 100; n++) {
2195 		DELAY(1);
2196 		v = CAS_READ_4(sc, CAS_MIF_FRAME);
2197 		if (v & CAS_MIF_FRAME_TA_LSB)
2198 			return (v & CAS_MIF_FRAME_DATA);
2199 	}
2200 
2201 	device_printf(sc->sc_dev, "%s: timed out\n", __func__);
2202 	return (0);
2203 }
2204 
2205 static int
2206 cas_mii_writereg(device_t dev, int phy, int reg, int val)
2207 {
2208 	struct cas_softc *sc;
2209 	int n;
2210 	uint32_t v;
2211 
2212 #ifdef CAS_DEBUG_PHY
2213 	printf("%s: phy %d reg %d val %x\n", phy, reg, val, __func__);
2214 #endif
2215 
2216 	sc = device_get_softc(dev);
2217 	if ((sc->sc_flags & CAS_SERDES) != 0) {
2218 		switch (reg) {
2219 		case MII_BMSR:
2220 			reg = CAS_PCS_STATUS;
2221 			break;
2222 		case MII_BMCR:
2223 			reg = CAS_PCS_CTRL;
2224 			if ((val & CAS_PCS_CTRL_RESET) == 0)
2225 				break;
2226 			CAS_WRITE_4(sc, CAS_PCS_CTRL, val);
2227 			CAS_BARRIER(sc, CAS_PCS_CTRL, 4,
2228 			    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2229 			if (!cas_bitwait(sc, CAS_PCS_CTRL,
2230 			    CAS_PCS_CTRL_RESET, 0))
2231 				device_printf(sc->sc_dev,
2232 				    "cannot reset PCS\n");
2233 			/* FALLTHROUGH */
2234 		case MII_ANAR:
2235 			CAS_WRITE_4(sc, CAS_PCS_CONF, 0);
2236 			CAS_BARRIER(sc, CAS_PCS_CONF, 4,
2237 			    BUS_SPACE_BARRIER_WRITE);
2238 			CAS_WRITE_4(sc, CAS_PCS_ANAR, val);
2239 			CAS_BARRIER(sc, CAS_PCS_ANAR, 4,
2240 			    BUS_SPACE_BARRIER_WRITE);
2241 			CAS_WRITE_4(sc, CAS_PCS_SERDES_CTRL,
2242 			    CAS_PCS_SERDES_CTRL_ESD);
2243 			CAS_BARRIER(sc, CAS_PCS_CONF, 4,
2244 			    BUS_SPACE_BARRIER_WRITE);
2245 			CAS_WRITE_4(sc, CAS_PCS_CONF,
2246 			    CAS_PCS_CONF_EN);
2247 			CAS_BARRIER(sc, CAS_PCS_CONF, 4,
2248 			    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2249 			return (0);
2250 		case MII_ANLPAR:
2251 			reg = CAS_PCS_ANLPAR;
2252 			break;
2253 		default:
2254 			device_printf(sc->sc_dev,
2255 			    "%s: unhandled register %d\n", __func__, reg);
2256 			return (0);
2257 		}
2258 		CAS_WRITE_4(sc, reg, val);
2259 		CAS_BARRIER(sc, reg, 4,
2260 		    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2261 		return (0);
2262 	}
2263 
2264 	/* Construct the frame command. */
2265 	v = CAS_MIF_FRAME_WRITE |
2266 	    (phy << CAS_MIF_FRAME_PHY_SHFT) |
2267 	    (reg << CAS_MIF_FRAME_REG_SHFT) |
2268 	    (val & CAS_MIF_FRAME_DATA);
2269 
2270 	CAS_WRITE_4(sc, CAS_MIF_FRAME, v);
2271 	CAS_BARRIER(sc, CAS_MIF_FRAME, 4,
2272 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2273 	for (n = 0; n < 100; n++) {
2274 		DELAY(1);
2275 		v = CAS_READ_4(sc, CAS_MIF_FRAME);
2276 		if (v & CAS_MIF_FRAME_TA_LSB)
2277 			return (1);
2278 	}
2279 
2280 	device_printf(sc->sc_dev, "%s: timed out\n", __func__);
2281 	return (0);
2282 }
2283 
2284 static void
2285 cas_mii_statchg(device_t dev)
2286 {
2287 	struct cas_softc *sc;
2288 	struct ifnet *ifp;
2289 	int gigabit;
2290 	uint32_t rxcfg, txcfg, v;
2291 
2292 	sc = device_get_softc(dev);
2293 	ifp = sc->sc_ifp;
2294 
2295 	CAS_LOCK_ASSERT(sc, MA_OWNED);
2296 
2297 #ifdef CAS_DEBUG
2298 	if ((ifp->if_flags & IFF_DEBUG) != 0)
2299 		device_printf(sc->sc_dev, "%s: status changen", __func__);
2300 #endif
2301 
2302 	if ((sc->sc_mii->mii_media_status & IFM_ACTIVE) != 0 &&
2303 	    IFM_SUBTYPE(sc->sc_mii->mii_media_active) != IFM_NONE)
2304 		sc->sc_flags |= CAS_LINK;
2305 	else
2306 		sc->sc_flags &= ~CAS_LINK;
2307 
2308 	switch (IFM_SUBTYPE(sc->sc_mii->mii_media_active)) {
2309 	case IFM_1000_SX:
2310 	case IFM_1000_LX:
2311 	case IFM_1000_CX:
2312 	case IFM_1000_T:
2313 		gigabit = 1;
2314 		break;
2315 	default:
2316 		gigabit = 0;
2317 	}
2318 
2319 	/*
2320 	 * The configuration done here corresponds to the steps F) and
2321 	 * G) and as far as enabling of RX and TX MAC goes also step H)
2322 	 * of the initialization sequence outlined in section 11.2.1 of
2323 	 * the Cassini+ ASIC Specification.
2324 	 */
2325 
2326 	rxcfg = sc->sc_mac_rxcfg;
2327 	rxcfg &= ~CAS_MAC_RX_CONF_CARR;
2328 	txcfg = CAS_MAC_TX_CONF_EN_IPG0 | CAS_MAC_TX_CONF_NGU |
2329 	    CAS_MAC_TX_CONF_NGUL;
2330 	if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) & IFM_FDX) != 0)
2331 		txcfg |= CAS_MAC_TX_CONF_ICARR | CAS_MAC_TX_CONF_ICOLLIS;
2332 	else if (gigabit != 0) {
2333 		rxcfg |= CAS_MAC_RX_CONF_CARR;
2334 		txcfg |= CAS_MAC_TX_CONF_CARR;
2335 	}
2336 	(void)cas_disable_tx(sc);
2337 	CAS_WRITE_4(sc, CAS_MAC_TX_CONF, txcfg);
2338 	(void)cas_disable_rx(sc);
2339 	CAS_WRITE_4(sc, CAS_MAC_RX_CONF, rxcfg);
2340 
2341 	v = CAS_READ_4(sc, CAS_MAC_CTRL_CONF) &
2342 	    ~(CAS_MAC_CTRL_CONF_TXP | CAS_MAC_CTRL_CONF_RXP);
2343 	if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) &
2344 	    IFM_ETH_RXPAUSE) != 0)
2345 		v |= CAS_MAC_CTRL_CONF_RXP;
2346 	if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) &
2347 	    IFM_ETH_TXPAUSE) != 0)
2348 		v |= CAS_MAC_CTRL_CONF_TXP;
2349 	CAS_WRITE_4(sc, CAS_MAC_CTRL_CONF, v);
2350 
2351 	/*
2352 	 * All supported chips have a bug causing incorrect checksum
2353 	 * to be calculated when letting them strip the FCS in half-
2354 	 * duplex mode.  In theory we could disable FCS stripping and
2355 	 * manually adjust the checksum accordingly.  It seems to make
2356 	 * more sense to optimze for the common case and just disable
2357 	 * hardware checksumming in half-duplex mode though.
2358 	 */
2359 	if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) & IFM_FDX) == 0) {
2360 		ifp->if_capenable &= ~IFCAP_HWCSUM;
2361 		ifp->if_hwassist = 0;
2362 	} else if ((sc->sc_flags & CAS_NO_CSUM) == 0) {
2363 		ifp->if_capenable = ifp->if_capabilities;
2364 		ifp->if_hwassist = CAS_CSUM_FEATURES;
2365 	}
2366 
2367 	if (sc->sc_variant == CAS_SATURN) {
2368 		if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) & IFM_FDX) == 0)
2369 			/* silicon bug workaround */
2370 			CAS_WRITE_4(sc, CAS_MAC_PREAMBLE_LEN, 0x41);
2371 		else
2372 			CAS_WRITE_4(sc, CAS_MAC_PREAMBLE_LEN, 0x7);
2373 	}
2374 
2375 	if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) & IFM_FDX) == 0 &&
2376 	    gigabit != 0)
2377 		CAS_WRITE_4(sc, CAS_MAC_SLOT_TIME,
2378 		    CAS_MAC_SLOT_TIME_CARR);
2379 	else
2380 		CAS_WRITE_4(sc, CAS_MAC_SLOT_TIME,
2381 		    CAS_MAC_SLOT_TIME_NORM);
2382 
2383 	/* XIF Configuration */
2384 	v = CAS_MAC_XIF_CONF_TX_OE | CAS_MAC_XIF_CONF_LNKLED;
2385 	if ((sc->sc_flags & CAS_SERDES) == 0) {
2386 		if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) & IFM_FDX) == 0)
2387 			v |= CAS_MAC_XIF_CONF_NOECHO;
2388 		v |= CAS_MAC_XIF_CONF_BUF_OE;
2389 	}
2390 	if (gigabit != 0)
2391 		v |= CAS_MAC_XIF_CONF_GMII;
2392 	if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) & IFM_FDX) != 0)
2393 		v |= CAS_MAC_XIF_CONF_FDXLED;
2394 	CAS_WRITE_4(sc, CAS_MAC_XIF_CONF, v);
2395 
2396 	sc->sc_mac_rxcfg = rxcfg;
2397 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0 &&
2398 	    (sc->sc_flags & CAS_LINK) != 0) {
2399 		CAS_WRITE_4(sc, CAS_MAC_TX_CONF,
2400 		    txcfg | CAS_MAC_TX_CONF_EN);
2401 		CAS_WRITE_4(sc, CAS_MAC_RX_CONF,
2402 		    rxcfg | CAS_MAC_RX_CONF_EN);
2403 	}
2404 }
2405 
2406 static int
2407 cas_mediachange(struct ifnet *ifp)
2408 {
2409 	struct cas_softc *sc = ifp->if_softc;
2410 	int error;
2411 
2412 	/* XXX add support for serial media. */
2413 
2414 	CAS_LOCK(sc);
2415 	error = mii_mediachg(sc->sc_mii);
2416 	CAS_UNLOCK(sc);
2417 	return (error);
2418 }
2419 
2420 static void
2421 cas_mediastatus(struct ifnet *ifp, struct ifmediareq *ifmr)
2422 {
2423 	struct cas_softc *sc = ifp->if_softc;
2424 
2425 	CAS_LOCK(sc);
2426 	if ((ifp->if_flags & IFF_UP) == 0) {
2427 		CAS_UNLOCK(sc);
2428 		return;
2429 	}
2430 
2431 	mii_pollstat(sc->sc_mii);
2432 	ifmr->ifm_active = sc->sc_mii->mii_media_active;
2433 	ifmr->ifm_status = sc->sc_mii->mii_media_status;
2434 	CAS_UNLOCK(sc);
2435 }
2436 
2437 static int
2438 cas_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
2439 {
2440 	struct cas_softc *sc = ifp->if_softc;
2441 	struct ifreq *ifr = (struct ifreq *)data;
2442 	int error;
2443 
2444 	error = 0;
2445 	switch (cmd) {
2446 	case SIOCSIFFLAGS:
2447 		CAS_LOCK(sc);
2448 		if ((ifp->if_flags & IFF_UP) != 0) {
2449 			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0 &&
2450 			    ((ifp->if_flags ^ sc->sc_ifflags) &
2451 			    (IFF_ALLMULTI | IFF_PROMISC)) != 0)
2452 				cas_setladrf(sc);
2453 			else
2454 				cas_init_locked(sc);
2455 		} else if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
2456 			cas_stop(ifp);
2457 		sc->sc_ifflags = ifp->if_flags;
2458 		CAS_UNLOCK(sc);
2459 		break;
2460 	case SIOCSIFCAP:
2461 		CAS_LOCK(sc);
2462 		if ((sc->sc_flags & CAS_NO_CSUM) != 0) {
2463 			error = EINVAL;
2464 			CAS_UNLOCK(sc);
2465 			break;
2466 		}
2467 		ifp->if_capenable = ifr->ifr_reqcap;
2468 		if ((ifp->if_capenable & IFCAP_TXCSUM) != 0)
2469 			ifp->if_hwassist = CAS_CSUM_FEATURES;
2470 		else
2471 			ifp->if_hwassist = 0;
2472 		CAS_UNLOCK(sc);
2473 		break;
2474 	case SIOCADDMULTI:
2475 	case SIOCDELMULTI:
2476 		CAS_LOCK(sc);
2477 		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
2478 			cas_setladrf(sc);
2479 		CAS_UNLOCK(sc);
2480 		break;
2481 	case SIOCSIFMTU:
2482 		if ((ifr->ifr_mtu < ETHERMIN) ||
2483 		    (ifr->ifr_mtu > ETHERMTU_JUMBO))
2484 			error = EINVAL;
2485 		else
2486 			ifp->if_mtu = ifr->ifr_mtu;
2487 		break;
2488 	case SIOCGIFMEDIA:
2489 	case SIOCSIFMEDIA:
2490 		error = ifmedia_ioctl(ifp, ifr, &sc->sc_mii->mii_media, cmd);
2491 		break;
2492 	default:
2493 		error = ether_ioctl(ifp, cmd, data);
2494 		break;
2495 	}
2496 
2497 	return (error);
2498 }
2499 
2500 static u_int
2501 cas_hash_maddr(void *arg, struct sockaddr_dl *sdl, u_int cnt)
2502 {
2503 	uint32_t crc, *hash = arg;
2504 
2505 	crc = ether_crc32_le(LLADDR(sdl), ETHER_ADDR_LEN);
2506 	/* We just want the 8 most significant bits. */
2507 	crc >>= 24;
2508 	/* Set the corresponding bit in the filter. */
2509 	hash[crc >> 4] |= 1 << (15 - (crc & 15));
2510 
2511 	return (1);
2512 }
2513 
2514 static void
2515 cas_setladrf(struct cas_softc *sc)
2516 {
2517 	struct ifnet *ifp = sc->sc_ifp;
2518 	int i;
2519 	uint32_t hash[16];
2520 	uint32_t v;
2521 
2522 	CAS_LOCK_ASSERT(sc, MA_OWNED);
2523 
2524 	/*
2525 	 * Turn off the RX MAC and the hash filter as required by the Sun
2526 	 * Cassini programming restrictions.
2527 	 */
2528 	v = sc->sc_mac_rxcfg & ~(CAS_MAC_RX_CONF_HFILTER |
2529 	    CAS_MAC_RX_CONF_EN);
2530 	CAS_WRITE_4(sc, CAS_MAC_RX_CONF, v);
2531 	CAS_BARRIER(sc, CAS_MAC_RX_CONF, 4,
2532 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2533 	if (!cas_bitwait(sc, CAS_MAC_RX_CONF, CAS_MAC_RX_CONF_HFILTER |
2534 	    CAS_MAC_RX_CONF_EN, 0))
2535 		device_printf(sc->sc_dev,
2536 		    "cannot disable RX MAC or hash filter\n");
2537 
2538 	v &= ~(CAS_MAC_RX_CONF_PROMISC | CAS_MAC_RX_CONF_PGRP);
2539 	if ((ifp->if_flags & IFF_PROMISC) != 0) {
2540 		v |= CAS_MAC_RX_CONF_PROMISC;
2541 		goto chipit;
2542 	}
2543 	if ((ifp->if_flags & IFF_ALLMULTI) != 0) {
2544 		v |= CAS_MAC_RX_CONF_PGRP;
2545 		goto chipit;
2546 	}
2547 
2548 	/*
2549 	 * Set up multicast address filter by passing all multicast
2550 	 * addresses through a crc generator, and then using the high
2551 	 * order 8 bits as an index into the 256 bit logical address
2552 	 * filter.  The high order 4 bits selects the word, while the
2553 	 * other 4 bits select the bit within the word (where bit 0
2554 	 * is the MSB).
2555 	 */
2556 
2557 	memset(hash, 0, sizeof(hash));
2558 	if_foreach_llmaddr(ifp, cas_hash_maddr, &hash);
2559 
2560 	v |= CAS_MAC_RX_CONF_HFILTER;
2561 
2562 	/* Now load the hash table into the chip (if we are using it). */
2563 	for (i = 0; i < 16; i++)
2564 		CAS_WRITE_4(sc,
2565 		    CAS_MAC_HASH0 + i * (CAS_MAC_HASH1 - CAS_MAC_HASH0),
2566 		    hash[i]);
2567 
2568  chipit:
2569 	sc->sc_mac_rxcfg = v;
2570 	CAS_WRITE_4(sc, CAS_MAC_RX_CONF, v | CAS_MAC_RX_CONF_EN);
2571 }
2572 
2573 static int	cas_pci_attach(device_t dev);
2574 static int	cas_pci_detach(device_t dev);
2575 static int	cas_pci_probe(device_t dev);
2576 static int	cas_pci_resume(device_t dev);
2577 static int	cas_pci_suspend(device_t dev);
2578 
2579 static device_method_t cas_pci_methods[] = {
2580 	/* Device interface */
2581 	DEVMETHOD(device_probe,		cas_pci_probe),
2582 	DEVMETHOD(device_attach,	cas_pci_attach),
2583 	DEVMETHOD(device_detach,	cas_pci_detach),
2584 	DEVMETHOD(device_suspend,	cas_pci_suspend),
2585 	DEVMETHOD(device_resume,	cas_pci_resume),
2586 	/* Use the suspend handler here, it is all that is required. */
2587 	DEVMETHOD(device_shutdown,	cas_pci_suspend),
2588 
2589 	/* MII interface */
2590 	DEVMETHOD(miibus_readreg,	cas_mii_readreg),
2591 	DEVMETHOD(miibus_writereg,	cas_mii_writereg),
2592 	DEVMETHOD(miibus_statchg,	cas_mii_statchg),
2593 
2594 	DEVMETHOD_END
2595 };
2596 
2597 static driver_t cas_pci_driver = {
2598 	"cas",
2599 	cas_pci_methods,
2600 	sizeof(struct cas_softc)
2601 };
2602 
2603 static const struct cas_pci_dev {
2604 	uint32_t	cpd_devid;
2605 	uint8_t		cpd_revid;
2606 	int		cpd_variant;
2607 	const char	*cpd_desc;
2608 } cas_pci_devlist[] = {
2609 	{ 0x0035100b, 0x0, CAS_SATURN, "NS DP83065 Saturn Gigabit Ethernet" },
2610 	{ 0xabba108e, 0x10, CAS_CASPLUS, "Sun Cassini+ Gigabit Ethernet" },
2611 	{ 0xabba108e, 0x0, CAS_CAS, "Sun Cassini Gigabit Ethernet" },
2612 	{ 0, 0, 0, NULL }
2613 };
2614 
2615 DRIVER_MODULE(cas, pci, cas_pci_driver, cas_devclass, 0, 0);
2616 MODULE_PNP_INFO("W32:vendor/device", pci, cas, cas_pci_devlist,
2617     nitems(cas_pci_devlist) - 1);
2618 DRIVER_MODULE(miibus, cas, miibus_driver, miibus_devclass, 0, 0);
2619 MODULE_DEPEND(cas, pci, 1, 1, 1);
2620 
2621 static int
2622 cas_pci_probe(device_t dev)
2623 {
2624 	int i;
2625 
2626 	for (i = 0; cas_pci_devlist[i].cpd_desc != NULL; i++) {
2627 		if (pci_get_devid(dev) == cas_pci_devlist[i].cpd_devid &&
2628 		    pci_get_revid(dev) >= cas_pci_devlist[i].cpd_revid) {
2629 			device_set_desc(dev, cas_pci_devlist[i].cpd_desc);
2630 			return (BUS_PROBE_DEFAULT);
2631 		}
2632 	}
2633 
2634 	return (ENXIO);
2635 }
2636 
2637 static struct resource_spec cas_pci_res_spec[] = {
2638 	{ SYS_RES_IRQ, 0, RF_SHAREABLE | RF_ACTIVE },	/* CAS_RES_INTR */
2639 	{ SYS_RES_MEMORY, PCIR_BAR(0), RF_ACTIVE },	/* CAS_RES_MEM */
2640 	{ -1, 0 }
2641 };
2642 
2643 #define	CAS_LOCAL_MAC_ADDRESS	"local-mac-address"
2644 #define	CAS_PHY_INTERFACE	"phy-interface"
2645 #define	CAS_PHY_TYPE		"phy-type"
2646 #define	CAS_PHY_TYPE_PCS	"pcs"
2647 
2648 static int
2649 cas_pci_attach(device_t dev)
2650 {
2651 	char buf[sizeof(CAS_LOCAL_MAC_ADDRESS)];
2652 	struct cas_softc *sc;
2653 	int i;
2654 #if !defined(__powerpc__)
2655 	u_char enaddr[4][ETHER_ADDR_LEN];
2656 	u_int j, k, lma, pcs[4], phy;
2657 #endif
2658 
2659 	sc = device_get_softc(dev);
2660 	sc->sc_variant = CAS_UNKNOWN;
2661 	for (i = 0; cas_pci_devlist[i].cpd_desc != NULL; i++) {
2662 		if (pci_get_devid(dev) == cas_pci_devlist[i].cpd_devid &&
2663 		    pci_get_revid(dev) >= cas_pci_devlist[i].cpd_revid) {
2664 			sc->sc_variant = cas_pci_devlist[i].cpd_variant;
2665 			break;
2666 		}
2667 	}
2668 	if (sc->sc_variant == CAS_UNKNOWN) {
2669 		device_printf(dev, "unknown adaptor\n");
2670 		return (ENXIO);
2671 	}
2672 
2673 	/* PCI configuration */
2674 	pci_write_config(dev, PCIR_COMMAND,
2675 	    pci_read_config(dev, PCIR_COMMAND, 2) | PCIM_CMD_BUSMASTEREN |
2676 	    PCIM_CMD_MWRICEN | PCIM_CMD_PERRESPEN | PCIM_CMD_SERRESPEN, 2);
2677 
2678 	sc->sc_dev = dev;
2679 	if (sc->sc_variant == CAS_CAS && pci_get_devid(dev) < 0x02)
2680 		/* Hardware checksumming may hang TX. */
2681 		sc->sc_flags |= CAS_NO_CSUM;
2682 	if (sc->sc_variant == CAS_CASPLUS || sc->sc_variant == CAS_SATURN)
2683 		sc->sc_flags |= CAS_REG_PLUS;
2684 	if (sc->sc_variant == CAS_CAS ||
2685 	    (sc->sc_variant == CAS_CASPLUS && pci_get_revid(dev) < 0x11))
2686 		sc->sc_flags |= CAS_TABORT;
2687 	if (bootverbose)
2688 		device_printf(dev, "flags=0x%x\n", sc->sc_flags);
2689 
2690 	if (bus_alloc_resources(dev, cas_pci_res_spec, sc->sc_res)) {
2691 		device_printf(dev, "failed to allocate resources\n");
2692 		bus_release_resources(dev, cas_pci_res_spec, sc->sc_res);
2693 		return (ENXIO);
2694 	}
2695 
2696 	CAS_LOCK_INIT(sc, device_get_nameunit(dev));
2697 
2698 #if defined(__powerpc__)
2699 	OF_getetheraddr(dev, sc->sc_enaddr);
2700 	if (OF_getprop(ofw_bus_get_node(dev), CAS_PHY_INTERFACE, buf,
2701 	    sizeof(buf)) > 0 || OF_getprop(ofw_bus_get_node(dev),
2702 	    CAS_PHY_TYPE, buf, sizeof(buf)) > 0) {
2703 		buf[sizeof(buf) - 1] = '\0';
2704 		if (strcmp(buf, CAS_PHY_TYPE_PCS) == 0)
2705 			sc->sc_flags |= CAS_SERDES;
2706 	}
2707 #else
2708 	/*
2709 	 * Dig out VPD (vital product data) and read the MAC address as well
2710 	 * as the PHY type.  The VPD resides in the PCI Expansion ROM (PCI
2711 	 * FCode) and can't be accessed via the PCI capability pointer.
2712 	 * SUNW,pci-ce and SUNW,pci-qge use the Enhanced VPD format described
2713 	 * in the free US Patent 7149820.
2714 	 */
2715 
2716 #define	PCI_ROMHDR_SIZE			0x1c
2717 #define	PCI_ROMHDR_SIG			0x00
2718 #define	PCI_ROMHDR_SIG_MAGIC		0xaa55		/* little endian */
2719 #define	PCI_ROMHDR_PTR_DATA		0x18
2720 #define	PCI_ROM_SIZE			0x18
2721 #define	PCI_ROM_SIG			0x00
2722 #define	PCI_ROM_SIG_MAGIC		0x52494350	/* "PCIR", endian */
2723 							/* reversed */
2724 #define	PCI_ROM_VENDOR			0x04
2725 #define	PCI_ROM_DEVICE			0x06
2726 #define	PCI_ROM_PTR_VPD			0x08
2727 #define	PCI_VPDRES_BYTE0		0x00
2728 #define	PCI_VPDRES_ISLARGE(x)		((x) & 0x80)
2729 #define	PCI_VPDRES_LARGE_NAME(x)	((x) & 0x7f)
2730 #define	PCI_VPDRES_LARGE_LEN_LSB	0x01
2731 #define	PCI_VPDRES_LARGE_LEN_MSB	0x02
2732 #define	PCI_VPDRES_LARGE_SIZE		0x03
2733 #define	PCI_VPDRES_TYPE_ID_STRING	0x02		/* large */
2734 #define	PCI_VPDRES_TYPE_VPD		0x10		/* large */
2735 #define	PCI_VPD_KEY0			0x00
2736 #define	PCI_VPD_KEY1			0x01
2737 #define	PCI_VPD_LEN			0x02
2738 #define	PCI_VPD_SIZE			0x03
2739 
2740 #define	CAS_ROM_READ_1(sc, offs)					\
2741 	CAS_READ_1((sc), CAS_PCI_ROM_OFFSET + (offs))
2742 #define	CAS_ROM_READ_2(sc, offs)					\
2743 	CAS_READ_2((sc), CAS_PCI_ROM_OFFSET + (offs))
2744 #define	CAS_ROM_READ_4(sc, offs)					\
2745 	CAS_READ_4((sc), CAS_PCI_ROM_OFFSET + (offs))
2746 
2747 	lma = phy = 0;
2748 	memset(enaddr, 0, sizeof(enaddr));
2749 	memset(pcs, 0, sizeof(pcs));
2750 
2751 	/* Enable PCI Expansion ROM access. */
2752 	CAS_WRITE_4(sc, CAS_BIM_LDEV_OEN,
2753 	    CAS_BIM_LDEV_OEN_PAD | CAS_BIM_LDEV_OEN_PROM);
2754 
2755 	/* Read PCI Expansion ROM header. */
2756 	if (CAS_ROM_READ_2(sc, PCI_ROMHDR_SIG) != PCI_ROMHDR_SIG_MAGIC ||
2757 	    (i = CAS_ROM_READ_2(sc, PCI_ROMHDR_PTR_DATA)) <
2758 	    PCI_ROMHDR_SIZE) {
2759 		device_printf(dev, "unexpected PCI Expansion ROM header\n");
2760 		goto fail_prom;
2761 	}
2762 
2763 	/* Read PCI Expansion ROM data. */
2764 	if (CAS_ROM_READ_4(sc, i + PCI_ROM_SIG) != PCI_ROM_SIG_MAGIC ||
2765 	    CAS_ROM_READ_2(sc, i + PCI_ROM_VENDOR) != pci_get_vendor(dev) ||
2766 	    CAS_ROM_READ_2(sc, i + PCI_ROM_DEVICE) != pci_get_device(dev) ||
2767 	    (j = CAS_ROM_READ_2(sc, i + PCI_ROM_PTR_VPD)) <
2768 	    i + PCI_ROM_SIZE) {
2769 		device_printf(dev, "unexpected PCI Expansion ROM data\n");
2770 		goto fail_prom;
2771 	}
2772 
2773 	/* Read PCI VPD. */
2774  next:
2775 	if (PCI_VPDRES_ISLARGE(CAS_ROM_READ_1(sc,
2776 	    j + PCI_VPDRES_BYTE0)) == 0) {
2777 		device_printf(dev, "no large PCI VPD\n");
2778 		goto fail_prom;
2779 	}
2780 
2781 	i = (CAS_ROM_READ_1(sc, j + PCI_VPDRES_LARGE_LEN_MSB) << 8) |
2782 	    CAS_ROM_READ_1(sc, j + PCI_VPDRES_LARGE_LEN_LSB);
2783 	switch (PCI_VPDRES_LARGE_NAME(CAS_ROM_READ_1(sc,
2784 	    j + PCI_VPDRES_BYTE0))) {
2785 	case PCI_VPDRES_TYPE_ID_STRING:
2786 		/* Skip identifier string. */
2787 		j += PCI_VPDRES_LARGE_SIZE + i;
2788 		goto next;
2789 	case PCI_VPDRES_TYPE_VPD:
2790 		for (j += PCI_VPDRES_LARGE_SIZE; i > 0;
2791 		    i -= PCI_VPD_SIZE + CAS_ROM_READ_1(sc, j + PCI_VPD_LEN),
2792 		    j += PCI_VPD_SIZE + CAS_ROM_READ_1(sc, j + PCI_VPD_LEN)) {
2793 			if (CAS_ROM_READ_1(sc, j + PCI_VPD_KEY0) != 'Z')
2794 				/* no Enhanced VPD */
2795 				continue;
2796 			if (CAS_ROM_READ_1(sc, j + PCI_VPD_SIZE) != 'I')
2797 				/* no instance property */
2798 				continue;
2799 			if (CAS_ROM_READ_1(sc, j + PCI_VPD_SIZE + 3) == 'B') {
2800 				/* byte array */
2801 				if (CAS_ROM_READ_1(sc,
2802 				    j + PCI_VPD_SIZE + 4) != ETHER_ADDR_LEN)
2803 					continue;
2804 				bus_read_region_1(sc->sc_res[CAS_RES_MEM],
2805 				    CAS_PCI_ROM_OFFSET + j + PCI_VPD_SIZE + 5,
2806 				    buf, sizeof(buf));
2807 				buf[sizeof(buf) - 1] = '\0';
2808 				if (strcmp(buf, CAS_LOCAL_MAC_ADDRESS) != 0)
2809 					continue;
2810 				bus_read_region_1(sc->sc_res[CAS_RES_MEM],
2811 				    CAS_PCI_ROM_OFFSET + j + PCI_VPD_SIZE +
2812 				    5 + sizeof(CAS_LOCAL_MAC_ADDRESS),
2813 				    enaddr[lma], sizeof(enaddr[lma]));
2814 				lma++;
2815 				if (lma == 4 && phy == 4)
2816 					break;
2817 			} else if (CAS_ROM_READ_1(sc, j + PCI_VPD_SIZE + 3) ==
2818 			   'S') {
2819 				/* string */
2820 				if (CAS_ROM_READ_1(sc,
2821 				    j + PCI_VPD_SIZE + 4) !=
2822 				    sizeof(CAS_PHY_TYPE_PCS))
2823 					continue;
2824 				bus_read_region_1(sc->sc_res[CAS_RES_MEM],
2825 				    CAS_PCI_ROM_OFFSET + j + PCI_VPD_SIZE + 5,
2826 				    buf, sizeof(buf));
2827 				buf[sizeof(buf) - 1] = '\0';
2828 				if (strcmp(buf, CAS_PHY_INTERFACE) == 0)
2829 					k = sizeof(CAS_PHY_INTERFACE);
2830 				else if (strcmp(buf, CAS_PHY_TYPE) == 0)
2831 					k = sizeof(CAS_PHY_TYPE);
2832 				else
2833 					continue;
2834 				bus_read_region_1(sc->sc_res[CAS_RES_MEM],
2835 				    CAS_PCI_ROM_OFFSET + j + PCI_VPD_SIZE +
2836 				    5 + k, buf, sizeof(buf));
2837 				buf[sizeof(buf) - 1] = '\0';
2838 				if (strcmp(buf, CAS_PHY_TYPE_PCS) == 0)
2839 					pcs[phy] = 1;
2840 				phy++;
2841 				if (lma == 4 && phy == 4)
2842 					break;
2843 			}
2844 		}
2845 		break;
2846 	default:
2847 		device_printf(dev, "unexpected PCI VPD\n");
2848 		goto fail_prom;
2849 	}
2850 
2851  fail_prom:
2852 	CAS_WRITE_4(sc, CAS_BIM_LDEV_OEN, 0);
2853 
2854 	if (lma == 0) {
2855 		device_printf(dev, "could not determine Ethernet address\n");
2856 		goto fail;
2857 	}
2858 	i = 0;
2859 	if (lma > 1 && pci_get_slot(dev) < nitems(enaddr))
2860 		i = pci_get_slot(dev);
2861 	memcpy(sc->sc_enaddr, enaddr[i], ETHER_ADDR_LEN);
2862 
2863 	if (phy == 0) {
2864 		device_printf(dev, "could not determine PHY type\n");
2865 		goto fail;
2866 	}
2867 	i = 0;
2868 	if (phy > 1 && pci_get_slot(dev) < nitems(pcs))
2869 		i = pci_get_slot(dev);
2870 	if (pcs[i] != 0)
2871 		sc->sc_flags |= CAS_SERDES;
2872 #endif
2873 
2874 	if (cas_attach(sc) != 0) {
2875 		device_printf(dev, "could not be attached\n");
2876 		goto fail;
2877 	}
2878 
2879 	if (bus_setup_intr(dev, sc->sc_res[CAS_RES_INTR], INTR_TYPE_NET |
2880 	    INTR_MPSAFE, cas_intr, NULL, sc, &sc->sc_ih) != 0) {
2881 		device_printf(dev, "failed to set up interrupt\n");
2882 		cas_detach(sc);
2883 		goto fail;
2884 	}
2885 	return (0);
2886 
2887  fail:
2888 	CAS_LOCK_DESTROY(sc);
2889 	bus_release_resources(dev, cas_pci_res_spec, sc->sc_res);
2890 	return (ENXIO);
2891 }
2892 
2893 static int
2894 cas_pci_detach(device_t dev)
2895 {
2896 	struct cas_softc *sc;
2897 
2898 	sc = device_get_softc(dev);
2899 	bus_teardown_intr(dev, sc->sc_res[CAS_RES_INTR], sc->sc_ih);
2900 	cas_detach(sc);
2901 	CAS_LOCK_DESTROY(sc);
2902 	bus_release_resources(dev, cas_pci_res_spec, sc->sc_res);
2903 	return (0);
2904 }
2905 
2906 static int
2907 cas_pci_suspend(device_t dev)
2908 {
2909 
2910 	cas_suspend(device_get_softc(dev));
2911 	return (0);
2912 }
2913 
2914 static int
2915 cas_pci_resume(device_t dev)
2916 {
2917 
2918 	cas_resume(device_get_softc(dev));
2919 	return (0);
2920 }
2921