xref: /freebsd/sys/dev/cas/if_cas.c (revision 2a3e3873a1e4cd958f2b0f85d3b10cfa40575d30)
1 /*-
2  * Copyright (C) 2001 Eduardo Horvath.
3  * Copyright (c) 2001-2003 Thomas Moestl
4  * Copyright (c) 2007-2009 Marius Strobl <marius@FreeBSD.org>
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR  ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR  BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  *
28  *	from: NetBSD: gem.c,v 1.21 2002/06/01 23:50:58 lukem Exp
29  *	from: FreeBSD: if_gem.c 182060 2008-08-23 15:03:26Z marius
30  */
31 
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34 
35 /*
36  * driver for Sun Cassini/Cassini+ and National Semiconductor DP83065
37  * Saturn Gigabit Ethernet controllers
38  */
39 
40 #if 0
41 #define	CAS_DEBUG
42 #endif
43 
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/bus.h>
47 #include <sys/callout.h>
48 #include <sys/endian.h>
49 #include <sys/mbuf.h>
50 #include <sys/malloc.h>
51 #include <sys/kernel.h>
52 #include <sys/lock.h>
53 #include <sys/module.h>
54 #include <sys/mutex.h>
55 #include <sys/refcount.h>
56 #include <sys/resource.h>
57 #include <sys/rman.h>
58 #include <sys/socket.h>
59 #include <sys/sockio.h>
60 #include <sys/taskqueue.h>
61 
62 #include <net/bpf.h>
63 #include <net/ethernet.h>
64 #include <net/if.h>
65 #include <net/if_arp.h>
66 #include <net/if_dl.h>
67 #include <net/if_media.h>
68 #include <net/if_types.h>
69 #include <net/if_vlan_var.h>
70 
71 #include <netinet/in.h>
72 #include <netinet/in_systm.h>
73 #include <netinet/ip.h>
74 #include <netinet/tcp.h>
75 #include <netinet/udp.h>
76 
77 #include <machine/bus.h>
78 #if defined(__powerpc__) || defined(__sparc64__)
79 #include <dev/ofw/ofw_bus.h>
80 #include <dev/ofw/openfirm.h>
81 #include <machine/ofw_machdep.h>
82 #endif
83 #include <machine/resource.h>
84 
85 #include <dev/mii/mii.h>
86 #include <dev/mii/miivar.h>
87 
88 #include <dev/cas/if_casreg.h>
89 #include <dev/cas/if_casvar.h>
90 
91 #include <dev/pci/pcireg.h>
92 #include <dev/pci/pcivar.h>
93 
94 #include "miibus_if.h"
95 
96 #define RINGASSERT(n , min, max)					\
97 	CTASSERT(powerof2(n) && (n) >= (min) && (n) <= (max))
98 
99 RINGASSERT(CAS_NRXCOMP, 128, 32768);
100 RINGASSERT(CAS_NRXDESC, 32, 8192);
101 RINGASSERT(CAS_NRXDESC2, 32, 8192);
102 RINGASSERT(CAS_NTXDESC, 32, 8192);
103 
104 #undef RINGASSERT
105 
106 #define	CCDASSERT(m, a)							\
107 	CTASSERT((offsetof(struct cas_control_data, m) & ((a) - 1)) == 0)
108 
109 CCDASSERT(ccd_rxcomps, CAS_RX_COMP_ALIGN);
110 CCDASSERT(ccd_rxdescs, CAS_RX_DESC_ALIGN);
111 CCDASSERT(ccd_rxdescs2, CAS_RX_DESC_ALIGN);
112 
113 #undef CCDASSERT
114 
115 #define	CAS_TRIES	10000
116 
117 /*
118  * According to documentation, the hardware has support for basic TCP
119  * checksum offloading only, in practice this can be also used for UDP
120  * however (i.e. the problem of previous Sun NICs that a checksum of 0x0
121  * is not converted to 0xffff no longer exists).
122  */
123 #define	CAS_CSUM_FEATURES	(CSUM_TCP | CSUM_UDP)
124 
125 static inline void cas_add_rxdesc(struct cas_softc *sc, u_int idx);
126 static int	cas_attach(struct cas_softc *sc);
127 static int	cas_bitwait(struct cas_softc *sc, bus_addr_t r, uint32_t clr,
128 		    uint32_t set);
129 static void	cas_cddma_callback(void *xsc, bus_dma_segment_t *segs,
130 		    int nsegs, int error);
131 static void	cas_detach(struct cas_softc *sc);
132 static int	cas_disable_rx(struct cas_softc *sc);
133 static int	cas_disable_tx(struct cas_softc *sc);
134 static void	cas_eint(struct cas_softc *sc, u_int status);
135 static void	cas_free(void *arg1, void* arg2);
136 static void	cas_init(void *xsc);
137 static void	cas_init_locked(struct cas_softc *sc);
138 static void	cas_init_regs(struct cas_softc *sc);
139 static int	cas_intr(void *v);
140 static void	cas_intr_task(void *arg, int pending __unused);
141 static int	cas_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data);
142 static int	cas_load_txmbuf(struct cas_softc *sc, struct mbuf **m_head);
143 static int	cas_mediachange(struct ifnet *ifp);
144 static void	cas_mediastatus(struct ifnet *ifp, struct ifmediareq *ifmr);
145 static void	cas_meminit(struct cas_softc *sc);
146 static void	cas_mifinit(struct cas_softc *sc);
147 static int	cas_mii_readreg(device_t dev, int phy, int reg);
148 static void	cas_mii_statchg(device_t dev);
149 static int	cas_mii_writereg(device_t dev, int phy, int reg, int val);
150 static void	cas_reset(struct cas_softc *sc);
151 static int	cas_reset_rx(struct cas_softc *sc);
152 static int	cas_reset_tx(struct cas_softc *sc);
153 static void	cas_resume(struct cas_softc *sc);
154 static u_int	cas_descsize(u_int sz);
155 static void	cas_rint(struct cas_softc *sc);
156 static void	cas_rint_timeout(void *arg);
157 static inline void cas_rxcksum(struct mbuf *m, uint16_t cksum);
158 static inline void cas_rxcompinit(struct cas_rx_comp *rxcomp);
159 static u_int	cas_rxcompsize(u_int sz);
160 static void	cas_rxdma_callback(void *xsc, bus_dma_segment_t *segs,
161 		    int nsegs, int error);
162 static void	cas_setladrf(struct cas_softc *sc);
163 static void	cas_start(struct ifnet *ifp);
164 static void	cas_stop(struct ifnet *ifp);
165 static void	cas_suspend(struct cas_softc *sc);
166 static void	cas_tick(void *arg);
167 static void	cas_tint(struct cas_softc *sc);
168 static void	cas_tx_task(void *arg, int pending __unused);
169 static inline void cas_txkick(struct cas_softc *sc);
170 static void	cas_watchdog(struct cas_softc *sc);
171 
172 static devclass_t cas_devclass;
173 
174 MODULE_DEPEND(cas, ether, 1, 1, 1);
175 MODULE_DEPEND(cas, miibus, 1, 1, 1);
176 
177 #ifdef CAS_DEBUG
178 #include <sys/ktr.h>
179 #define	KTR_CAS		KTR_SPARE2
180 #endif
181 
182 static int
183 cas_attach(struct cas_softc *sc)
184 {
185 	struct cas_txsoft *txs;
186 	struct ifnet *ifp;
187 	int error, i;
188 	uint32_t v;
189 
190 	/* Set up ifnet structure. */
191 	ifp = sc->sc_ifp = if_alloc(IFT_ETHER);
192 	if (ifp == NULL)
193 		return (ENOSPC);
194 	ifp->if_softc = sc;
195 	if_initname(ifp, device_get_name(sc->sc_dev),
196 	    device_get_unit(sc->sc_dev));
197 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
198 	ifp->if_start = cas_start;
199 	ifp->if_ioctl = cas_ioctl;
200 	ifp->if_init = cas_init;
201 	IFQ_SET_MAXLEN(&ifp->if_snd, CAS_TXQUEUELEN);
202 	ifp->if_snd.ifq_drv_maxlen = CAS_TXQUEUELEN;
203 	IFQ_SET_READY(&ifp->if_snd);
204 
205 	callout_init_mtx(&sc->sc_tick_ch, &sc->sc_mtx, 0);
206 	callout_init_mtx(&sc->sc_rx_ch, &sc->sc_mtx, 0);
207 	/* Create local taskq. */
208 	TASK_INIT(&sc->sc_intr_task, 0, cas_intr_task, sc);
209 	TASK_INIT(&sc->sc_tx_task, 1, cas_tx_task, ifp);
210 	sc->sc_tq = taskqueue_create_fast("cas_taskq", M_WAITOK,
211 	    taskqueue_thread_enqueue, &sc->sc_tq);
212 	if (sc->sc_tq == NULL) {
213 		device_printf(sc->sc_dev, "could not create taskqueue\n");
214 		error = ENXIO;
215 		goto fail_ifnet;
216 	}
217 	error = taskqueue_start_threads(&sc->sc_tq, 1, PI_NET, "%s taskq",
218 	    device_get_nameunit(sc->sc_dev));
219 	if (error != 0) {
220 		device_printf(sc->sc_dev, "could not start threads\n");
221 		goto fail_taskq;
222 	}
223 
224 	/* Make sure the chip is stopped. */
225 	cas_reset(sc);
226 
227 	error = bus_dma_tag_create(bus_get_dma_tag(sc->sc_dev), 1, 0,
228 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
229 	    BUS_SPACE_MAXSIZE, 0, BUS_SPACE_MAXSIZE, 0, NULL, NULL,
230 	    &sc->sc_pdmatag);
231 	if (error != 0)
232 		goto fail_taskq;
233 
234 	error = bus_dma_tag_create(sc->sc_pdmatag, 1, 0,
235 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
236 	    CAS_PAGE_SIZE, 1, CAS_PAGE_SIZE, 0, NULL, NULL, &sc->sc_rdmatag);
237 	if (error != 0)
238 		goto fail_ptag;
239 
240 	error = bus_dma_tag_create(sc->sc_pdmatag, 1, 0,
241 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
242 	    MCLBYTES * CAS_NTXSEGS, CAS_NTXSEGS, MCLBYTES,
243 	    BUS_DMA_ALLOCNOW, NULL, NULL, &sc->sc_tdmatag);
244 	if (error != 0)
245 		goto fail_rtag;
246 
247 	error = bus_dma_tag_create(sc->sc_pdmatag, CAS_TX_DESC_ALIGN, 0,
248 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
249 	    sizeof(struct cas_control_data), 1,
250 	    sizeof(struct cas_control_data), 0,
251 	    NULL, NULL, &sc->sc_cdmatag);
252 	if (error != 0)
253 		goto fail_ttag;
254 
255 	/*
256 	 * Allocate the control data structures, create and load the
257 	 * DMA map for it.
258 	 */
259 	if ((error = bus_dmamem_alloc(sc->sc_cdmatag,
260 	    (void **)&sc->sc_control_data,
261 	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
262 	    &sc->sc_cddmamap)) != 0) {
263 		device_printf(sc->sc_dev,
264 		    "unable to allocate control data, error = %d\n", error);
265 		goto fail_ctag;
266 	}
267 
268 	sc->sc_cddma = 0;
269 	if ((error = bus_dmamap_load(sc->sc_cdmatag, sc->sc_cddmamap,
270 	    sc->sc_control_data, sizeof(struct cas_control_data),
271 	    cas_cddma_callback, sc, 0)) != 0 || sc->sc_cddma == 0) {
272 		device_printf(sc->sc_dev,
273 		    "unable to load control data DMA map, error = %d\n",
274 		    error);
275 		goto fail_cmem;
276 	}
277 
278 	/*
279 	 * Initialize the transmit job descriptors.
280 	 */
281 	STAILQ_INIT(&sc->sc_txfreeq);
282 	STAILQ_INIT(&sc->sc_txdirtyq);
283 
284 	/*
285 	 * Create the transmit buffer DMA maps.
286 	 */
287 	error = ENOMEM;
288 	for (i = 0; i < CAS_TXQUEUELEN; i++) {
289 		txs = &sc->sc_txsoft[i];
290 		txs->txs_mbuf = NULL;
291 		txs->txs_ndescs = 0;
292 		if ((error = bus_dmamap_create(sc->sc_tdmatag, 0,
293 		    &txs->txs_dmamap)) != 0) {
294 			device_printf(sc->sc_dev,
295 			    "unable to create TX DMA map %d, error = %d\n",
296 			    i, error);
297 			goto fail_txd;
298 		}
299 		STAILQ_INSERT_TAIL(&sc->sc_txfreeq, txs, txs_q);
300 	}
301 
302 	/*
303 	 * Allocate the receive buffers, create and load the DMA maps
304 	 * for them.
305 	 */
306 	for (i = 0; i < CAS_NRXDESC; i++) {
307 		if ((error = bus_dmamem_alloc(sc->sc_rdmatag,
308 		    &sc->sc_rxdsoft[i].rxds_buf, BUS_DMA_WAITOK,
309 		    &sc->sc_rxdsoft[i].rxds_dmamap)) != 0) {
310 			device_printf(sc->sc_dev,
311 			    "unable to allocate RX buffer %d, error = %d\n",
312 			    i, error);
313 			goto fail_rxmem;
314 		}
315 
316 		sc->sc_rxdptr = i;
317 		sc->sc_rxdsoft[i].rxds_paddr = 0;
318 		if ((error = bus_dmamap_load(sc->sc_rdmatag,
319 		    sc->sc_rxdsoft[i].rxds_dmamap, sc->sc_rxdsoft[i].rxds_buf,
320 		    CAS_PAGE_SIZE, cas_rxdma_callback, sc, 0)) != 0 ||
321 		    sc->sc_rxdsoft[i].rxds_paddr == 0) {
322 			device_printf(sc->sc_dev,
323 			    "unable to load RX DMA map %d, error = %d\n",
324 			    i, error);
325 			goto fail_rxmap;
326 		}
327 	}
328 
329 	if ((sc->sc_flags & CAS_SERDES) == 0) {
330 		CAS_WRITE_4(sc, CAS_PCS_DATAPATH, CAS_PCS_DATAPATH_MII);
331 		CAS_BARRIER(sc, CAS_PCS_DATAPATH, 4,
332 		    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
333 		cas_mifinit(sc);
334 		/*
335 		 * Look for an external PHY.
336 		 */
337 		error = ENXIO;
338 		v = CAS_READ_4(sc, CAS_MIF_CONF);
339 		if ((v & CAS_MIF_CONF_MDI1) != 0) {
340 			v |= CAS_MIF_CONF_PHY_SELECT;
341 			CAS_WRITE_4(sc, CAS_MIF_CONF, v);
342 			CAS_BARRIER(sc, CAS_MIF_CONF, 4,
343 			    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
344 			/* Enable/unfreeze the GMII pins of Saturn. */
345 			if (sc->sc_variant == CAS_SATURN) {
346 				CAS_WRITE_4(sc, CAS_SATURN_PCFG,
347 				    CAS_READ_4(sc, CAS_SATURN_PCFG) &
348 				    ~CAS_SATURN_PCFG_FSI);
349 				CAS_BARRIER(sc, CAS_SATURN_PCFG, 4,
350 				    BUS_SPACE_BARRIER_READ |
351 				    BUS_SPACE_BARRIER_WRITE);
352 				DELAY(10000);
353 			}
354 			error = mii_attach(sc->sc_dev, &sc->sc_miibus, ifp,
355 			    cas_mediachange, cas_mediastatus, BMSR_DEFCAPMASK,
356 			    MII_PHY_ANY, MII_OFFSET_ANY, MIIF_DOPAUSE);
357 		}
358 		/*
359 		 * Fall back on an internal PHY if no external PHY was found.
360 		 */
361 		if (error != 0 && (v & CAS_MIF_CONF_MDI0) != 0) {
362 			v &= ~CAS_MIF_CONF_PHY_SELECT;
363 			CAS_WRITE_4(sc, CAS_MIF_CONF, v);
364 			CAS_BARRIER(sc, CAS_MIF_CONF, 4,
365 			    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
366 			/* Freeze the GMII pins of Saturn for saving power. */
367 			if (sc->sc_variant == CAS_SATURN) {
368 				CAS_WRITE_4(sc, CAS_SATURN_PCFG,
369 				    CAS_READ_4(sc, CAS_SATURN_PCFG) |
370 				    CAS_SATURN_PCFG_FSI);
371 				CAS_BARRIER(sc, CAS_SATURN_PCFG, 4,
372 				    BUS_SPACE_BARRIER_READ |
373 				    BUS_SPACE_BARRIER_WRITE);
374 				DELAY(10000);
375 			}
376 			error = mii_attach(sc->sc_dev, &sc->sc_miibus, ifp,
377 			    cas_mediachange, cas_mediastatus, BMSR_DEFCAPMASK,
378 			    MII_PHY_ANY, MII_OFFSET_ANY, MIIF_DOPAUSE);
379 		}
380 	} else {
381 		/*
382 		 * Use the external PCS SERDES.
383 		 */
384 		CAS_WRITE_4(sc, CAS_PCS_DATAPATH, CAS_PCS_DATAPATH_SERDES);
385 		CAS_BARRIER(sc, CAS_PCS_DATAPATH, 4, BUS_SPACE_BARRIER_WRITE);
386 		/* Enable/unfreeze the SERDES pins of Saturn. */
387 		if (sc->sc_variant == CAS_SATURN) {
388 			CAS_WRITE_4(sc, CAS_SATURN_PCFG, 0);
389 			CAS_BARRIER(sc, CAS_SATURN_PCFG, 4,
390 			    BUS_SPACE_BARRIER_WRITE);
391 		}
392 		CAS_WRITE_4(sc, CAS_PCS_SERDES_CTRL, CAS_PCS_SERDES_CTRL_ESD);
393 		CAS_BARRIER(sc, CAS_PCS_SERDES_CTRL, 4,
394 		    BUS_SPACE_BARRIER_WRITE);
395 		CAS_WRITE_4(sc, CAS_PCS_CONF, CAS_PCS_CONF_EN);
396 		CAS_BARRIER(sc, CAS_PCS_CONF, 4,
397 		    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
398 		error = mii_attach(sc->sc_dev, &sc->sc_miibus, ifp,
399 		    cas_mediachange, cas_mediastatus, BMSR_DEFCAPMASK,
400 		    CAS_PHYAD_EXTERNAL, MII_OFFSET_ANY, MIIF_DOPAUSE);
401 	}
402 	if (error != 0) {
403 		device_printf(sc->sc_dev, "attaching PHYs failed\n");
404 		goto fail_rxmap;
405 	}
406 	sc->sc_mii = device_get_softc(sc->sc_miibus);
407 
408 	/*
409 	 * From this point forward, the attachment cannot fail.  A failure
410 	 * before this point releases all resources that may have been
411 	 * allocated.
412 	 */
413 
414 	/* Announce FIFO sizes. */
415 	v = CAS_READ_4(sc, CAS_TX_FIFO_SIZE);
416 	device_printf(sc->sc_dev, "%ukB RX FIFO, %ukB TX FIFO\n",
417 	    CAS_RX_FIFO_SIZE / 1024, v / 16);
418 
419 	/* Attach the interface. */
420 	ether_ifattach(ifp, sc->sc_enaddr);
421 
422 	/*
423 	 * Tell the upper layer(s) we support long frames/checksum offloads.
424 	 */
425 	ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
426 	ifp->if_capabilities = IFCAP_VLAN_MTU;
427 	if ((sc->sc_flags & CAS_NO_CSUM) == 0) {
428 		ifp->if_capabilities |= IFCAP_HWCSUM;
429 		ifp->if_hwassist = CAS_CSUM_FEATURES;
430 	}
431 	ifp->if_capenable = ifp->if_capabilities;
432 
433 	return (0);
434 
435 	/*
436 	 * Free any resources we've allocated during the failed attach
437 	 * attempt.  Do this in reverse order and fall through.
438 	 */
439  fail_rxmap:
440 	for (i = 0; i < CAS_NRXDESC; i++)
441 		if (sc->sc_rxdsoft[i].rxds_paddr != 0)
442 			bus_dmamap_unload(sc->sc_rdmatag,
443 			    sc->sc_rxdsoft[i].rxds_dmamap);
444  fail_rxmem:
445 	for (i = 0; i < CAS_NRXDESC; i++)
446 		if (sc->sc_rxdsoft[i].rxds_buf != NULL)
447 			bus_dmamem_free(sc->sc_rdmatag,
448 			    sc->sc_rxdsoft[i].rxds_buf,
449 			    sc->sc_rxdsoft[i].rxds_dmamap);
450  fail_txd:
451 	for (i = 0; i < CAS_TXQUEUELEN; i++)
452 		if (sc->sc_txsoft[i].txs_dmamap != NULL)
453 			bus_dmamap_destroy(sc->sc_tdmatag,
454 			    sc->sc_txsoft[i].txs_dmamap);
455 	bus_dmamap_unload(sc->sc_cdmatag, sc->sc_cddmamap);
456  fail_cmem:
457 	bus_dmamem_free(sc->sc_cdmatag, sc->sc_control_data,
458 	    sc->sc_cddmamap);
459  fail_ctag:
460 	bus_dma_tag_destroy(sc->sc_cdmatag);
461  fail_ttag:
462 	bus_dma_tag_destroy(sc->sc_tdmatag);
463  fail_rtag:
464 	bus_dma_tag_destroy(sc->sc_rdmatag);
465  fail_ptag:
466 	bus_dma_tag_destroy(sc->sc_pdmatag);
467  fail_taskq:
468 	taskqueue_free(sc->sc_tq);
469  fail_ifnet:
470 	if_free(ifp);
471 	return (error);
472 }
473 
474 static void
475 cas_detach(struct cas_softc *sc)
476 {
477 	struct ifnet *ifp = sc->sc_ifp;
478 	int i;
479 
480 	ether_ifdetach(ifp);
481 	CAS_LOCK(sc);
482 	cas_stop(ifp);
483 	CAS_UNLOCK(sc);
484 	callout_drain(&sc->sc_tick_ch);
485 	callout_drain(&sc->sc_rx_ch);
486 	taskqueue_drain(sc->sc_tq, &sc->sc_intr_task);
487 	taskqueue_drain(sc->sc_tq, &sc->sc_tx_task);
488 	if_free(ifp);
489 	taskqueue_free(sc->sc_tq);
490 	device_delete_child(sc->sc_dev, sc->sc_miibus);
491 
492 	for (i = 0; i < CAS_NRXDESC; i++)
493 		if (sc->sc_rxdsoft[i].rxds_dmamap != NULL)
494 			bus_dmamap_sync(sc->sc_rdmatag,
495 			    sc->sc_rxdsoft[i].rxds_dmamap,
496 			    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
497 	for (i = 0; i < CAS_NRXDESC; i++)
498 		if (sc->sc_rxdsoft[i].rxds_paddr != 0)
499 			bus_dmamap_unload(sc->sc_rdmatag,
500 			    sc->sc_rxdsoft[i].rxds_dmamap);
501 	for (i = 0; i < CAS_NRXDESC; i++)
502 		if (sc->sc_rxdsoft[i].rxds_buf != NULL)
503 			bus_dmamem_free(sc->sc_rdmatag,
504 			    sc->sc_rxdsoft[i].rxds_buf,
505 			    sc->sc_rxdsoft[i].rxds_dmamap);
506 	for (i = 0; i < CAS_TXQUEUELEN; i++)
507 		if (sc->sc_txsoft[i].txs_dmamap != NULL)
508 			bus_dmamap_destroy(sc->sc_tdmatag,
509 			    sc->sc_txsoft[i].txs_dmamap);
510 	CAS_CDSYNC(sc, BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
511 	bus_dmamap_unload(sc->sc_cdmatag, sc->sc_cddmamap);
512 	bus_dmamem_free(sc->sc_cdmatag, sc->sc_control_data,
513 	    sc->sc_cddmamap);
514 	bus_dma_tag_destroy(sc->sc_cdmatag);
515 	bus_dma_tag_destroy(sc->sc_tdmatag);
516 	bus_dma_tag_destroy(sc->sc_rdmatag);
517 	bus_dma_tag_destroy(sc->sc_pdmatag);
518 }
519 
520 static void
521 cas_suspend(struct cas_softc *sc)
522 {
523 	struct ifnet *ifp = sc->sc_ifp;
524 
525 	CAS_LOCK(sc);
526 	cas_stop(ifp);
527 	CAS_UNLOCK(sc);
528 }
529 
530 static void
531 cas_resume(struct cas_softc *sc)
532 {
533 	struct ifnet *ifp = sc->sc_ifp;
534 
535 	CAS_LOCK(sc);
536 	/*
537 	 * On resume all registers have to be initialized again like
538 	 * after power-on.
539 	 */
540 	sc->sc_flags &= ~CAS_INITED;
541 	if (ifp->if_flags & IFF_UP)
542 		cas_init_locked(sc);
543 	CAS_UNLOCK(sc);
544 }
545 
546 static inline void
547 cas_rxcksum(struct mbuf *m, uint16_t cksum)
548 {
549 	struct ether_header *eh;
550 	struct ip *ip;
551 	struct udphdr *uh;
552 	uint16_t *opts;
553 	int32_t hlen, len, pktlen;
554 	uint32_t temp32;
555 
556 	pktlen = m->m_pkthdr.len;
557 	if (pktlen < sizeof(struct ether_header) + sizeof(struct ip))
558 		return;
559 	eh = mtod(m, struct ether_header *);
560 	if (eh->ether_type != htons(ETHERTYPE_IP))
561 		return;
562 	ip = (struct ip *)(eh + 1);
563 	if (ip->ip_v != IPVERSION)
564 		return;
565 
566 	hlen = ip->ip_hl << 2;
567 	pktlen -= sizeof(struct ether_header);
568 	if (hlen < sizeof(struct ip))
569 		return;
570 	if (ntohs(ip->ip_len) < hlen)
571 		return;
572 	if (ntohs(ip->ip_len) != pktlen)
573 		return;
574 	if (ip->ip_off & htons(IP_MF | IP_OFFMASK))
575 		return;	/* Cannot handle fragmented packet. */
576 
577 	switch (ip->ip_p) {
578 	case IPPROTO_TCP:
579 		if (pktlen < (hlen + sizeof(struct tcphdr)))
580 			return;
581 		break;
582 	case IPPROTO_UDP:
583 		if (pktlen < (hlen + sizeof(struct udphdr)))
584 			return;
585 		uh = (struct udphdr *)((uint8_t *)ip + hlen);
586 		if (uh->uh_sum == 0)
587 			return; /* no checksum */
588 		break;
589 	default:
590 		return;
591 	}
592 
593 	cksum = ~cksum;
594 	/* checksum fixup for IP options */
595 	len = hlen - sizeof(struct ip);
596 	if (len > 0) {
597 		opts = (uint16_t *)(ip + 1);
598 		for (; len > 0; len -= sizeof(uint16_t), opts++) {
599 			temp32 = cksum - *opts;
600 			temp32 = (temp32 >> 16) + (temp32 & 65535);
601 			cksum = temp32 & 65535;
602 		}
603 	}
604 	m->m_pkthdr.csum_flags |= CSUM_DATA_VALID;
605 	m->m_pkthdr.csum_data = cksum;
606 }
607 
608 static void
609 cas_cddma_callback(void *xsc, bus_dma_segment_t *segs, int nsegs, int error)
610 {
611 	struct cas_softc *sc = xsc;
612 
613 	if (error != 0)
614 		return;
615 	if (nsegs != 1)
616 		panic("%s: bad control buffer segment count", __func__);
617 	sc->sc_cddma = segs[0].ds_addr;
618 }
619 
620 static void
621 cas_rxdma_callback(void *xsc, bus_dma_segment_t *segs, int nsegs, int error)
622 {
623 	struct cas_softc *sc = xsc;
624 
625 	if (error != 0)
626 		return;
627 	if (nsegs != 1)
628 		panic("%s: bad RX buffer segment count", __func__);
629 	sc->sc_rxdsoft[sc->sc_rxdptr].rxds_paddr = segs[0].ds_addr;
630 }
631 
632 static void
633 cas_tick(void *arg)
634 {
635 	struct cas_softc *sc = arg;
636 	struct ifnet *ifp = sc->sc_ifp;
637 	uint32_t v;
638 
639 	CAS_LOCK_ASSERT(sc, MA_OWNED);
640 
641 	/*
642 	 * Unload collision and error counters.
643 	 */
644 	ifp->if_collisions +=
645 	    CAS_READ_4(sc, CAS_MAC_NORM_COLL_CNT) +
646 	    CAS_READ_4(sc, CAS_MAC_FIRST_COLL_CNT);
647 	v = CAS_READ_4(sc, CAS_MAC_EXCESS_COLL_CNT) +
648 	    CAS_READ_4(sc, CAS_MAC_LATE_COLL_CNT);
649 	ifp->if_collisions += v;
650 	ifp->if_oerrors += v;
651 	ifp->if_ierrors +=
652 	    CAS_READ_4(sc, CAS_MAC_RX_LEN_ERR_CNT) +
653 	    CAS_READ_4(sc, CAS_MAC_RX_ALIGN_ERR) +
654 	    CAS_READ_4(sc, CAS_MAC_RX_CRC_ERR_CNT) +
655 	    CAS_READ_4(sc, CAS_MAC_RX_CODE_VIOL);
656 
657 	/*
658 	 * Then clear the hardware counters.
659 	 */
660 	CAS_WRITE_4(sc, CAS_MAC_NORM_COLL_CNT, 0);
661 	CAS_WRITE_4(sc, CAS_MAC_FIRST_COLL_CNT, 0);
662 	CAS_WRITE_4(sc, CAS_MAC_EXCESS_COLL_CNT, 0);
663 	CAS_WRITE_4(sc, CAS_MAC_LATE_COLL_CNT, 0);
664 	CAS_WRITE_4(sc, CAS_MAC_RX_LEN_ERR_CNT, 0);
665 	CAS_WRITE_4(sc, CAS_MAC_RX_ALIGN_ERR, 0);
666 	CAS_WRITE_4(sc, CAS_MAC_RX_CRC_ERR_CNT, 0);
667 	CAS_WRITE_4(sc, CAS_MAC_RX_CODE_VIOL, 0);
668 
669 	mii_tick(sc->sc_mii);
670 
671 	if (sc->sc_txfree != CAS_MAXTXFREE)
672 		cas_tint(sc);
673 
674 	cas_watchdog(sc);
675 
676 	callout_reset(&sc->sc_tick_ch, hz, cas_tick, sc);
677 }
678 
679 static int
680 cas_bitwait(struct cas_softc *sc, bus_addr_t r, uint32_t clr, uint32_t set)
681 {
682 	int i;
683 	uint32_t reg;
684 
685 	for (i = CAS_TRIES; i--; DELAY(100)) {
686 		reg = CAS_READ_4(sc, r);
687 		if ((reg & clr) == 0 && (reg & set) == set)
688 			return (1);
689 	}
690 	return (0);
691 }
692 
693 static void
694 cas_reset(struct cas_softc *sc)
695 {
696 
697 #ifdef CAS_DEBUG
698 	CTR2(KTR_CAS, "%s: %s", device_get_name(sc->sc_dev), __func__);
699 #endif
700 	/* Disable all interrupts in order to avoid spurious ones. */
701 	CAS_WRITE_4(sc, CAS_INTMASK, 0xffffffff);
702 
703 	cas_reset_rx(sc);
704 	cas_reset_tx(sc);
705 
706 	/*
707 	 * Do a full reset modulo the result of the last auto-negotiation
708 	 * when using the SERDES.
709 	 */
710 	CAS_WRITE_4(sc, CAS_RESET, CAS_RESET_RX | CAS_RESET_TX |
711 	    ((sc->sc_flags & CAS_SERDES) != 0 ? CAS_RESET_PCS_DIS : 0));
712 	CAS_BARRIER(sc, CAS_RESET, 4,
713 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
714 	DELAY(3000);
715 	if (!cas_bitwait(sc, CAS_RESET, CAS_RESET_RX | CAS_RESET_TX, 0))
716 		device_printf(sc->sc_dev, "cannot reset device\n");
717 }
718 
719 static void
720 cas_stop(struct ifnet *ifp)
721 {
722 	struct cas_softc *sc = ifp->if_softc;
723 	struct cas_txsoft *txs;
724 
725 #ifdef CAS_DEBUG
726 	CTR2(KTR_CAS, "%s: %s", device_get_name(sc->sc_dev), __func__);
727 #endif
728 
729 	callout_stop(&sc->sc_tick_ch);
730 	callout_stop(&sc->sc_rx_ch);
731 
732 	/* Disable all interrupts in order to avoid spurious ones. */
733 	CAS_WRITE_4(sc, CAS_INTMASK, 0xffffffff);
734 
735 	cas_reset_tx(sc);
736 	cas_reset_rx(sc);
737 
738 	/*
739 	 * Release any queued transmit buffers.
740 	 */
741 	while ((txs = STAILQ_FIRST(&sc->sc_txdirtyq)) != NULL) {
742 		STAILQ_REMOVE_HEAD(&sc->sc_txdirtyq, txs_q);
743 		if (txs->txs_ndescs != 0) {
744 			bus_dmamap_sync(sc->sc_tdmatag, txs->txs_dmamap,
745 			    BUS_DMASYNC_POSTWRITE);
746 			bus_dmamap_unload(sc->sc_tdmatag, txs->txs_dmamap);
747 			if (txs->txs_mbuf != NULL) {
748 				m_freem(txs->txs_mbuf);
749 				txs->txs_mbuf = NULL;
750 			}
751 		}
752 		STAILQ_INSERT_TAIL(&sc->sc_txfreeq, txs, txs_q);
753 	}
754 
755 	/*
756 	 * Mark the interface down and cancel the watchdog timer.
757 	 */
758 	ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
759 	sc->sc_flags &= ~CAS_LINK;
760 	sc->sc_wdog_timer = 0;
761 }
762 
763 static int
764 cas_reset_rx(struct cas_softc *sc)
765 {
766 
767 	/*
768 	 * Resetting while DMA is in progress can cause a bus hang, so we
769 	 * disable DMA first.
770 	 */
771 	(void)cas_disable_rx(sc);
772 	CAS_WRITE_4(sc, CAS_RX_CONF, 0);
773 	CAS_BARRIER(sc, CAS_RX_CONF, 4,
774 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
775 	if (!cas_bitwait(sc, CAS_RX_CONF, CAS_RX_CONF_RXDMA_EN, 0))
776 		device_printf(sc->sc_dev, "cannot disable RX DMA\n");
777 
778 	/* Finally, reset the ERX. */
779 	CAS_WRITE_4(sc, CAS_RESET, CAS_RESET_RX |
780 	    ((sc->sc_flags & CAS_SERDES) != 0 ? CAS_RESET_PCS_DIS : 0));
781 	CAS_BARRIER(sc, CAS_RESET, 4,
782 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
783 	if (!cas_bitwait(sc, CAS_RESET, CAS_RESET_RX, 0)) {
784 		device_printf(sc->sc_dev, "cannot reset receiver\n");
785 		return (1);
786 	}
787 	return (0);
788 }
789 
790 static int
791 cas_reset_tx(struct cas_softc *sc)
792 {
793 
794 	/*
795 	 * Resetting while DMA is in progress can cause a bus hang, so we
796 	 * disable DMA first.
797 	 */
798 	(void)cas_disable_tx(sc);
799 	CAS_WRITE_4(sc, CAS_TX_CONF, 0);
800 	CAS_BARRIER(sc, CAS_TX_CONF, 4,
801 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
802 	if (!cas_bitwait(sc, CAS_TX_CONF, CAS_TX_CONF_TXDMA_EN, 0))
803 		device_printf(sc->sc_dev, "cannot disable TX DMA\n");
804 
805 	/* Finally, reset the ETX. */
806 	CAS_WRITE_4(sc, CAS_RESET, CAS_RESET_TX |
807 	    ((sc->sc_flags & CAS_SERDES) != 0 ? CAS_RESET_PCS_DIS : 0));
808 	CAS_BARRIER(sc, CAS_RESET, 4,
809 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
810 	if (!cas_bitwait(sc, CAS_RESET, CAS_RESET_TX, 0)) {
811 		device_printf(sc->sc_dev, "cannot reset transmitter\n");
812 		return (1);
813 	}
814 	return (0);
815 }
816 
817 static int
818 cas_disable_rx(struct cas_softc *sc)
819 {
820 
821 	CAS_WRITE_4(sc, CAS_MAC_RX_CONF,
822 	    CAS_READ_4(sc, CAS_MAC_RX_CONF) & ~CAS_MAC_RX_CONF_EN);
823 	CAS_BARRIER(sc, CAS_MAC_RX_CONF, 4,
824 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
825 	if (cas_bitwait(sc, CAS_MAC_RX_CONF, CAS_MAC_RX_CONF_EN, 0))
826 		return (1);
827 	if (bootverbose)
828 		device_printf(sc->sc_dev, "cannot disable RX MAC\n");
829 	return (0);
830 }
831 
832 static int
833 cas_disable_tx(struct cas_softc *sc)
834 {
835 
836 	CAS_WRITE_4(sc, CAS_MAC_TX_CONF,
837 	    CAS_READ_4(sc, CAS_MAC_TX_CONF) & ~CAS_MAC_TX_CONF_EN);
838 	CAS_BARRIER(sc, CAS_MAC_TX_CONF, 4,
839 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
840 	if (cas_bitwait(sc, CAS_MAC_TX_CONF, CAS_MAC_TX_CONF_EN, 0))
841 		return (1);
842 	if (bootverbose)
843 		device_printf(sc->sc_dev, "cannot disable TX MAC\n");
844 	return (0);
845 }
846 
847 static inline void
848 cas_rxcompinit(struct cas_rx_comp *rxcomp)
849 {
850 
851 	rxcomp->crc_word1 = 0;
852 	rxcomp->crc_word2 = 0;
853 	rxcomp->crc_word3 =
854 	    htole64(CAS_SET(ETHER_HDR_LEN + sizeof(struct ip), CAS_RC3_CSO));
855 	rxcomp->crc_word4 = htole64(CAS_RC4_ZERO);
856 }
857 
858 static void
859 cas_meminit(struct cas_softc *sc)
860 {
861 	int i;
862 
863 	CAS_LOCK_ASSERT(sc, MA_OWNED);
864 
865 	/*
866 	 * Initialize the transmit descriptor ring.
867 	 */
868 	for (i = 0; i < CAS_NTXDESC; i++) {
869 		sc->sc_txdescs[i].cd_flags = 0;
870 		sc->sc_txdescs[i].cd_buf_ptr = 0;
871 	}
872 	sc->sc_txfree = CAS_MAXTXFREE;
873 	sc->sc_txnext = 0;
874 	sc->sc_txwin = 0;
875 
876 	/*
877 	 * Initialize the receive completion ring.
878 	 */
879 	for (i = 0; i < CAS_NRXCOMP; i++)
880 		cas_rxcompinit(&sc->sc_rxcomps[i]);
881 	sc->sc_rxcptr = 0;
882 
883 	/*
884 	 * Initialize the first receive descriptor ring.  We leave
885 	 * the second one zeroed as we don't actually use it.
886 	 */
887 	for (i = 0; i < CAS_NRXDESC; i++)
888 		CAS_INIT_RXDESC(sc, i, i);
889 	sc->sc_rxdptr = 0;
890 
891 	CAS_CDSYNC(sc, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
892 }
893 
894 static u_int
895 cas_descsize(u_int sz)
896 {
897 
898 	switch (sz) {
899 	case 32:
900 		return (CAS_DESC_32);
901 	case 64:
902 		return (CAS_DESC_64);
903 	case 128:
904 		return (CAS_DESC_128);
905 	case 256:
906 		return (CAS_DESC_256);
907 	case 512:
908 		return (CAS_DESC_512);
909 	case 1024:
910 		return (CAS_DESC_1K);
911 	case 2048:
912 		return (CAS_DESC_2K);
913 	case 4096:
914 		return (CAS_DESC_4K);
915 	case 8192:
916 		return (CAS_DESC_8K);
917 	default:
918 		printf("%s: invalid descriptor ring size %d\n", __func__, sz);
919 		return (CAS_DESC_32);
920 	}
921 }
922 
923 static u_int
924 cas_rxcompsize(u_int sz)
925 {
926 
927 	switch (sz) {
928 	case 128:
929 		return (CAS_RX_CONF_COMP_128);
930 	case 256:
931 		return (CAS_RX_CONF_COMP_256);
932 	case 512:
933 		return (CAS_RX_CONF_COMP_512);
934 	case 1024:
935 		return (CAS_RX_CONF_COMP_1K);
936 	case 2048:
937 		return (CAS_RX_CONF_COMP_2K);
938 	case 4096:
939 		return (CAS_RX_CONF_COMP_4K);
940 	case 8192:
941 		return (CAS_RX_CONF_COMP_8K);
942 	case 16384:
943 		return (CAS_RX_CONF_COMP_16K);
944 	case 32768:
945 		return (CAS_RX_CONF_COMP_32K);
946 	default:
947 		printf("%s: invalid dcompletion ring size %d\n", __func__, sz);
948 		return (CAS_RX_CONF_COMP_128);
949 	}
950 }
951 
952 static void
953 cas_init(void *xsc)
954 {
955 	struct cas_softc *sc = xsc;
956 
957 	CAS_LOCK(sc);
958 	cas_init_locked(sc);
959 	CAS_UNLOCK(sc);
960 }
961 
962 /*
963  * Initialization of interface; set up initialization block
964  * and transmit/receive descriptor rings.
965  */
966 static void
967 cas_init_locked(struct cas_softc *sc)
968 {
969 	struct ifnet *ifp = sc->sc_ifp;
970 	uint32_t v;
971 
972 	CAS_LOCK_ASSERT(sc, MA_OWNED);
973 
974 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
975 		return;
976 
977 #ifdef CAS_DEBUG
978 	CTR2(KTR_CAS, "%s: %s: calling stop", device_get_name(sc->sc_dev),
979 	    __func__);
980 #endif
981 	/*
982 	 * Initialization sequence.  The numbered steps below correspond
983 	 * to the sequence outlined in section 6.3.5.1 in the Ethernet
984 	 * Channel Engine manual (part of the PCIO manual).
985 	 * See also the STP2002-STQ document from Sun Microsystems.
986 	 */
987 
988 	/* step 1 & 2.  Reset the Ethernet Channel. */
989 	cas_stop(ifp);
990 	cas_reset(sc);
991 #ifdef CAS_DEBUG
992 	CTR2(KTR_CAS, "%s: %s: restarting", device_get_name(sc->sc_dev),
993 	    __func__);
994 #endif
995 
996 	if ((sc->sc_flags & CAS_SERDES) == 0)
997 		/* Re-initialize the MIF. */
998 		cas_mifinit(sc);
999 
1000 	/* step 3.  Setup data structures in host memory. */
1001 	cas_meminit(sc);
1002 
1003 	/* step 4.  TX MAC registers & counters */
1004 	cas_init_regs(sc);
1005 
1006 	/* step 5.  RX MAC registers & counters */
1007 
1008 	/* step 6 & 7.  Program Ring Base Addresses. */
1009 	CAS_WRITE_4(sc, CAS_TX_DESC3_BASE_HI,
1010 	    (((uint64_t)CAS_CDTXDADDR(sc, 0)) >> 32));
1011 	CAS_WRITE_4(sc, CAS_TX_DESC3_BASE_LO,
1012 	    CAS_CDTXDADDR(sc, 0) & 0xffffffff);
1013 
1014 	CAS_WRITE_4(sc, CAS_RX_COMP_BASE_HI,
1015 	    (((uint64_t)CAS_CDRXCADDR(sc, 0)) >> 32));
1016 	CAS_WRITE_4(sc, CAS_RX_COMP_BASE_LO,
1017 	    CAS_CDRXCADDR(sc, 0) & 0xffffffff);
1018 
1019 	CAS_WRITE_4(sc, CAS_RX_DESC_BASE_HI,
1020 	    (((uint64_t)CAS_CDRXDADDR(sc, 0)) >> 32));
1021 	CAS_WRITE_4(sc, CAS_RX_DESC_BASE_LO,
1022 	    CAS_CDRXDADDR(sc, 0) & 0xffffffff);
1023 
1024 	if ((sc->sc_flags & CAS_REG_PLUS) != 0) {
1025 		CAS_WRITE_4(sc, CAS_RX_DESC2_BASE_HI,
1026 		    (((uint64_t)CAS_CDRXD2ADDR(sc, 0)) >> 32));
1027 		CAS_WRITE_4(sc, CAS_RX_DESC2_BASE_LO,
1028 		    CAS_CDRXD2ADDR(sc, 0) & 0xffffffff);
1029 	}
1030 
1031 #ifdef CAS_DEBUG
1032 	CTR5(KTR_CAS,
1033 	    "loading TXDR %lx, RXCR %lx, RXDR %lx, RXD2R %lx, cddma %lx",
1034 	    CAS_CDTXDADDR(sc, 0), CAS_CDRXCADDR(sc, 0), CAS_CDRXDADDR(sc, 0),
1035 	    CAS_CDRXD2ADDR(sc, 0), sc->sc_cddma);
1036 #endif
1037 
1038 	/* step 8.  Global Configuration & Interrupt Masks */
1039 
1040 	/* Disable weighted round robin. */
1041 	CAS_WRITE_4(sc, CAS_CAW, CAS_CAW_RR_DIS);
1042 
1043 	/*
1044 	 * Enable infinite bursts for revisions without PCI issues if
1045 	 * applicable.  Doing so greatly improves the TX performance on
1046 	 * !__sparc64__ (on sparc64, setting CAS_INF_BURST improves TX
1047 	 * performance only marginally but hurts RX throughput quite a bit).
1048 	 */
1049 	CAS_WRITE_4(sc, CAS_INF_BURST,
1050 #if !defined(__sparc64__)
1051 	    (sc->sc_flags & CAS_TABORT) == 0 ? CAS_INF_BURST_EN :
1052 #endif
1053 	    0);
1054 
1055 	/* Set up interrupts. */
1056 	CAS_WRITE_4(sc, CAS_INTMASK,
1057 	    ~(CAS_INTR_TX_INT_ME | CAS_INTR_TX_TAG_ERR |
1058 	    CAS_INTR_RX_DONE | CAS_INTR_RX_BUF_NA | CAS_INTR_RX_TAG_ERR |
1059 	    CAS_INTR_RX_COMP_FULL | CAS_INTR_RX_BUF_AEMPTY |
1060 	    CAS_INTR_RX_COMP_AFULL | CAS_INTR_RX_LEN_MMATCH |
1061 	    CAS_INTR_PCI_ERROR_INT
1062 #ifdef CAS_DEBUG
1063 	    | CAS_INTR_PCS_INT | CAS_INTR_MIF
1064 #endif
1065 	    ));
1066 	/* Don't clear top level interrupts when CAS_STATUS_ALIAS is read. */
1067 	CAS_WRITE_4(sc, CAS_CLEAR_ALIAS, 0);
1068 	CAS_WRITE_4(sc, CAS_MAC_RX_MASK, ~CAS_MAC_RX_OVERFLOW);
1069 	CAS_WRITE_4(sc, CAS_MAC_TX_MASK,
1070 	    ~(CAS_MAC_TX_UNDERRUN | CAS_MAC_TX_MAX_PKT_ERR));
1071 #ifdef CAS_DEBUG
1072 	CAS_WRITE_4(sc, CAS_MAC_CTRL_MASK,
1073 	    ~(CAS_MAC_CTRL_PAUSE_RCVD | CAS_MAC_CTRL_PAUSE |
1074 	    CAS_MAC_CTRL_NON_PAUSE));
1075 #else
1076 	CAS_WRITE_4(sc, CAS_MAC_CTRL_MASK,
1077 	    CAS_MAC_CTRL_PAUSE_RCVD | CAS_MAC_CTRL_PAUSE |
1078 	    CAS_MAC_CTRL_NON_PAUSE);
1079 #endif
1080 
1081 	/* Enable PCI error interrupts. */
1082 	CAS_WRITE_4(sc, CAS_ERROR_MASK,
1083 	    ~(CAS_ERROR_DTRTO | CAS_ERROR_OTHER | CAS_ERROR_DMAW_ZERO |
1084 	    CAS_ERROR_DMAR_ZERO | CAS_ERROR_RTRTO));
1085 
1086 	/* Enable PCI error interrupts in BIM configuration. */
1087 	CAS_WRITE_4(sc, CAS_BIM_CONF,
1088 	    CAS_BIM_CONF_DPAR_EN | CAS_BIM_CONF_RMA_EN | CAS_BIM_CONF_RTA_EN);
1089 
1090 	/*
1091 	 * step 9.  ETX Configuration: encode receive descriptor ring size,
1092 	 * enable DMA and disable pre-interrupt writeback completion.
1093 	 */
1094 	v = cas_descsize(CAS_NTXDESC) << CAS_TX_CONF_DESC3_SHFT;
1095 	CAS_WRITE_4(sc, CAS_TX_CONF, v | CAS_TX_CONF_TXDMA_EN |
1096 	    CAS_TX_CONF_RDPP_DIS | CAS_TX_CONF_PICWB_DIS);
1097 
1098 	/* step 10.  ERX Configuration */
1099 
1100 	/*
1101 	 * Encode receive completion and descriptor ring sizes, set the
1102 	 * swivel offset.
1103 	 */
1104 	v = cas_rxcompsize(CAS_NRXCOMP) << CAS_RX_CONF_COMP_SHFT;
1105 	v |= cas_descsize(CAS_NRXDESC) << CAS_RX_CONF_DESC_SHFT;
1106 	if ((sc->sc_flags & CAS_REG_PLUS) != 0)
1107 		v |= cas_descsize(CAS_NRXDESC2) << CAS_RX_CONF_DESC2_SHFT;
1108 	CAS_WRITE_4(sc, CAS_RX_CONF,
1109 	    v | (ETHER_ALIGN << CAS_RX_CONF_SOFF_SHFT));
1110 
1111 	/* Set the PAUSE thresholds.  We use the maximum OFF threshold. */
1112 	CAS_WRITE_4(sc, CAS_RX_PTHRS,
1113 	    (111 << CAS_RX_PTHRS_XOFF_SHFT) | (15 << CAS_RX_PTHRS_XON_SHFT));
1114 
1115 	/* RX blanking */
1116 	CAS_WRITE_4(sc, CAS_RX_BLANK,
1117 	    (15 << CAS_RX_BLANK_TIME_SHFT) | (5 << CAS_RX_BLANK_PKTS_SHFT));
1118 
1119 	/* Set RX_COMP_AFULL threshold to half of the RX completions. */
1120 	CAS_WRITE_4(sc, CAS_RX_AEMPTY_THRS,
1121 	    (CAS_NRXCOMP / 2) << CAS_RX_AEMPTY_COMP_SHFT);
1122 
1123 	/* Initialize the RX page size register as appropriate for 8k. */
1124 	CAS_WRITE_4(sc, CAS_RX_PSZ,
1125 	    (CAS_RX_PSZ_8K << CAS_RX_PSZ_SHFT) |
1126 	    (4 << CAS_RX_PSZ_MB_CNT_SHFT) |
1127 	    (CAS_RX_PSZ_MB_STRD_2K << CAS_RX_PSZ_MB_STRD_SHFT) |
1128 	    (CAS_RX_PSZ_MB_OFF_64 << CAS_RX_PSZ_MB_OFF_SHFT));
1129 
1130 	/* Disable RX random early detection. */
1131 	CAS_WRITE_4(sc,	CAS_RX_RED, 0);
1132 
1133 	/* Zero the RX reassembly DMA table. */
1134 	for (v = 0; v <= CAS_RX_REAS_DMA_ADDR_LC; v++) {
1135 		CAS_WRITE_4(sc,	CAS_RX_REAS_DMA_ADDR, v);
1136 		CAS_WRITE_4(sc,	CAS_RX_REAS_DMA_DATA_LO, 0);
1137 		CAS_WRITE_4(sc,	CAS_RX_REAS_DMA_DATA_MD, 0);
1138 		CAS_WRITE_4(sc,	CAS_RX_REAS_DMA_DATA_HI, 0);
1139 	}
1140 
1141 	/* Ensure the RX control FIFO and RX IPP FIFO addresses are zero. */
1142 	CAS_WRITE_4(sc, CAS_RX_CTRL_FIFO, 0);
1143 	CAS_WRITE_4(sc, CAS_RX_IPP_ADDR, 0);
1144 
1145 	/* Finally, enable RX DMA. */
1146 	CAS_WRITE_4(sc, CAS_RX_CONF,
1147 	    CAS_READ_4(sc, CAS_RX_CONF) | CAS_RX_CONF_RXDMA_EN);
1148 
1149 	/* step 11.  Configure Media. */
1150 
1151 	/* step 12.  RX_MAC Configuration Register */
1152 	v = CAS_READ_4(sc, CAS_MAC_RX_CONF);
1153 	v &= ~(CAS_MAC_RX_CONF_STRPPAD | CAS_MAC_RX_CONF_EN);
1154 	v |= CAS_MAC_RX_CONF_STRPFCS;
1155 	sc->sc_mac_rxcfg = v;
1156 	/*
1157 	 * Clear the RX filter and reprogram it.  This will also set the
1158 	 * current RX MAC configuration and enable it.
1159 	 */
1160 	cas_setladrf(sc);
1161 
1162 	/* step 13.  TX_MAC Configuration Register */
1163 	v = CAS_READ_4(sc, CAS_MAC_TX_CONF);
1164 	v |= CAS_MAC_TX_CONF_EN;
1165 	(void)cas_disable_tx(sc);
1166 	CAS_WRITE_4(sc, CAS_MAC_TX_CONF, v);
1167 
1168 	/* step 14.  Issue Transmit Pending command. */
1169 
1170 	/* step 15.  Give the receiver a swift kick. */
1171 	CAS_WRITE_4(sc, CAS_RX_KICK, CAS_NRXDESC - 4);
1172 	CAS_WRITE_4(sc, CAS_RX_COMP_TAIL, 0);
1173 	if ((sc->sc_flags & CAS_REG_PLUS) != 0)
1174 		CAS_WRITE_4(sc, CAS_RX_KICK2, CAS_NRXDESC2 - 4);
1175 
1176 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1177 	ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1178 
1179 	mii_mediachg(sc->sc_mii);
1180 
1181 	/* Start the one second timer. */
1182 	sc->sc_wdog_timer = 0;
1183 	callout_reset(&sc->sc_tick_ch, hz, cas_tick, sc);
1184 }
1185 
1186 static int
1187 cas_load_txmbuf(struct cas_softc *sc, struct mbuf **m_head)
1188 {
1189 	bus_dma_segment_t txsegs[CAS_NTXSEGS];
1190 	struct cas_txsoft *txs;
1191 	struct ip *ip;
1192 	struct mbuf *m;
1193 	uint64_t cflags;
1194 	int error, nexttx, nsegs, offset, seg;
1195 
1196 	CAS_LOCK_ASSERT(sc, MA_OWNED);
1197 
1198 	/* Get a work queue entry. */
1199 	if ((txs = STAILQ_FIRST(&sc->sc_txfreeq)) == NULL) {
1200 		/* Ran out of descriptors. */
1201 		return (ENOBUFS);
1202 	}
1203 
1204 	cflags = 0;
1205 	if (((*m_head)->m_pkthdr.csum_flags & CAS_CSUM_FEATURES) != 0) {
1206 		if (M_WRITABLE(*m_head) == 0) {
1207 			m = m_dup(*m_head, M_NOWAIT);
1208 			m_freem(*m_head);
1209 			*m_head = m;
1210 			if (m == NULL)
1211 				return (ENOBUFS);
1212 		}
1213 		offset = sizeof(struct ether_header);
1214 		m = m_pullup(*m_head, offset + sizeof(struct ip));
1215 		if (m == NULL) {
1216 			*m_head = NULL;
1217 			return (ENOBUFS);
1218 		}
1219 		ip = (struct ip *)(mtod(m, caddr_t) + offset);
1220 		offset += (ip->ip_hl << 2);
1221 		cflags = (offset << CAS_TD_CKSUM_START_SHFT) |
1222 		    ((offset + m->m_pkthdr.csum_data) <<
1223 		    CAS_TD_CKSUM_STUFF_SHFT) | CAS_TD_CKSUM_EN;
1224 		*m_head = m;
1225 	}
1226 
1227 	error = bus_dmamap_load_mbuf_sg(sc->sc_tdmatag, txs->txs_dmamap,
1228 	    *m_head, txsegs, &nsegs, BUS_DMA_NOWAIT);
1229 	if (error == EFBIG) {
1230 		m = m_collapse(*m_head, M_NOWAIT, CAS_NTXSEGS);
1231 		if (m == NULL) {
1232 			m_freem(*m_head);
1233 			*m_head = NULL;
1234 			return (ENOBUFS);
1235 		}
1236 		*m_head = m;
1237 		error = bus_dmamap_load_mbuf_sg(sc->sc_tdmatag,
1238 		    txs->txs_dmamap, *m_head, txsegs, &nsegs,
1239 		    BUS_DMA_NOWAIT);
1240 		if (error != 0) {
1241 			m_freem(*m_head);
1242 			*m_head = NULL;
1243 			return (error);
1244 		}
1245 	} else if (error != 0)
1246 		return (error);
1247 	/* If nsegs is wrong then the stack is corrupt. */
1248 	KASSERT(nsegs <= CAS_NTXSEGS,
1249 	    ("%s: too many DMA segments (%d)", __func__, nsegs));
1250 	if (nsegs == 0) {
1251 		m_freem(*m_head);
1252 		*m_head = NULL;
1253 		return (EIO);
1254 	}
1255 
1256 	/*
1257 	 * Ensure we have enough descriptors free to describe
1258 	 * the packet.  Note, we always reserve one descriptor
1259 	 * at the end of the ring as a termination point, in
1260 	 * order to prevent wrap-around.
1261 	 */
1262 	if (nsegs > sc->sc_txfree - 1) {
1263 		txs->txs_ndescs = 0;
1264 		bus_dmamap_unload(sc->sc_tdmatag, txs->txs_dmamap);
1265 		return (ENOBUFS);
1266 	}
1267 
1268 	txs->txs_ndescs = nsegs;
1269 	txs->txs_firstdesc = sc->sc_txnext;
1270 	nexttx = txs->txs_firstdesc;
1271 	for (seg = 0; seg < nsegs; seg++, nexttx = CAS_NEXTTX(nexttx)) {
1272 #ifdef CAS_DEBUG
1273 		CTR6(KTR_CAS,
1274 		    "%s: mapping seg %d (txd %d), len %lx, addr %#lx (%#lx)",
1275 		    __func__, seg, nexttx, txsegs[seg].ds_len,
1276 		    txsegs[seg].ds_addr, htole64(txsegs[seg].ds_addr));
1277 #endif
1278 		sc->sc_txdescs[nexttx].cd_buf_ptr =
1279 		    htole64(txsegs[seg].ds_addr);
1280 		KASSERT(txsegs[seg].ds_len <
1281 		    CAS_TD_BUF_LEN_MASK >> CAS_TD_BUF_LEN_SHFT,
1282 		    ("%s: segment size too large!", __func__));
1283 		sc->sc_txdescs[nexttx].cd_flags =
1284 		    htole64(txsegs[seg].ds_len << CAS_TD_BUF_LEN_SHFT);
1285 		txs->txs_lastdesc = nexttx;
1286 	}
1287 
1288 	/* Set EOF on the last descriptor. */
1289 #ifdef CAS_DEBUG
1290 	CTR3(KTR_CAS, "%s: end of frame at segment %d, TX %d",
1291 	    __func__, seg, nexttx);
1292 #endif
1293 	sc->sc_txdescs[txs->txs_lastdesc].cd_flags |=
1294 	    htole64(CAS_TD_END_OF_FRAME);
1295 
1296 	/* Lastly set SOF on the first descriptor. */
1297 #ifdef CAS_DEBUG
1298 	CTR3(KTR_CAS, "%s: start of frame at segment %d, TX %d",
1299 	    __func__, seg, nexttx);
1300 #endif
1301 	if (sc->sc_txwin += nsegs > CAS_MAXTXFREE * 2 / 3) {
1302 		sc->sc_txwin = 0;
1303 		sc->sc_txdescs[txs->txs_firstdesc].cd_flags |=
1304 		    htole64(cflags | CAS_TD_START_OF_FRAME | CAS_TD_INT_ME);
1305 	} else
1306 		sc->sc_txdescs[txs->txs_firstdesc].cd_flags |=
1307 		    htole64(cflags | CAS_TD_START_OF_FRAME);
1308 
1309 	/* Sync the DMA map. */
1310 	bus_dmamap_sync(sc->sc_tdmatag, txs->txs_dmamap,
1311 	    BUS_DMASYNC_PREWRITE);
1312 
1313 #ifdef CAS_DEBUG
1314 	CTR4(KTR_CAS, "%s: setting firstdesc=%d, lastdesc=%d, ndescs=%d",
1315 	    __func__, txs->txs_firstdesc, txs->txs_lastdesc,
1316 	    txs->txs_ndescs);
1317 #endif
1318 	STAILQ_REMOVE_HEAD(&sc->sc_txfreeq, txs_q);
1319 	STAILQ_INSERT_TAIL(&sc->sc_txdirtyq, txs, txs_q);
1320 	txs->txs_mbuf = *m_head;
1321 
1322 	sc->sc_txnext = CAS_NEXTTX(txs->txs_lastdesc);
1323 	sc->sc_txfree -= txs->txs_ndescs;
1324 
1325 	return (0);
1326 }
1327 
1328 static void
1329 cas_init_regs(struct cas_softc *sc)
1330 {
1331 	int i;
1332 	const u_char *laddr = IF_LLADDR(sc->sc_ifp);
1333 
1334 	CAS_LOCK_ASSERT(sc, MA_OWNED);
1335 
1336 	/* These registers are not cleared on reset. */
1337 	if ((sc->sc_flags & CAS_INITED) == 0) {
1338 		/* magic values */
1339 		CAS_WRITE_4(sc, CAS_MAC_IPG0, 0);
1340 		CAS_WRITE_4(sc, CAS_MAC_IPG1, 8);
1341 		CAS_WRITE_4(sc, CAS_MAC_IPG2, 4);
1342 
1343 		/* min frame length */
1344 		CAS_WRITE_4(sc, CAS_MAC_MIN_FRAME, ETHER_MIN_LEN);
1345 		/* max frame length and max burst size */
1346 		CAS_WRITE_4(sc, CAS_MAC_MAX_BF,
1347 		    ((ETHER_MAX_LEN_JUMBO + ETHER_VLAN_ENCAP_LEN) <<
1348 		    CAS_MAC_MAX_BF_FRM_SHFT) |
1349 		    (0x2000 << CAS_MAC_MAX_BF_BST_SHFT));
1350 
1351 		/* more magic values */
1352 		CAS_WRITE_4(sc, CAS_MAC_PREAMBLE_LEN, 0x7);
1353 		CAS_WRITE_4(sc, CAS_MAC_JAM_SIZE, 0x4);
1354 		CAS_WRITE_4(sc, CAS_MAC_ATTEMPT_LIMIT, 0x10);
1355 		CAS_WRITE_4(sc, CAS_MAC_CTRL_TYPE, 0x8808);
1356 
1357 		/* random number seed */
1358 		CAS_WRITE_4(sc, CAS_MAC_RANDOM_SEED,
1359 		    ((laddr[5] << 8) | laddr[4]) & 0x3ff);
1360 
1361 		/* secondary MAC addresses: 0:0:0:0:0:0 */
1362 		for (i = CAS_MAC_ADDR3; i <= CAS_MAC_ADDR41;
1363 		    i += CAS_MAC_ADDR4 - CAS_MAC_ADDR3)
1364 			CAS_WRITE_4(sc, i, 0);
1365 
1366 		/* MAC control address: 01:80:c2:00:00:01 */
1367 		CAS_WRITE_4(sc, CAS_MAC_ADDR42, 0x0001);
1368 		CAS_WRITE_4(sc, CAS_MAC_ADDR43, 0xc200);
1369 		CAS_WRITE_4(sc, CAS_MAC_ADDR44, 0x0180);
1370 
1371 		/* MAC filter address: 0:0:0:0:0:0 */
1372 		CAS_WRITE_4(sc, CAS_MAC_AFILTER0, 0);
1373 		CAS_WRITE_4(sc, CAS_MAC_AFILTER1, 0);
1374 		CAS_WRITE_4(sc, CAS_MAC_AFILTER2, 0);
1375 		CAS_WRITE_4(sc, CAS_MAC_AFILTER_MASK1_2, 0);
1376 		CAS_WRITE_4(sc, CAS_MAC_AFILTER_MASK0, 0);
1377 
1378 		/* Zero the hash table. */
1379 		for (i = CAS_MAC_HASH0; i <= CAS_MAC_HASH15;
1380 		    i += CAS_MAC_HASH1 - CAS_MAC_HASH0)
1381 			CAS_WRITE_4(sc, i, 0);
1382 
1383 		sc->sc_flags |= CAS_INITED;
1384 	}
1385 
1386 	/* Counters need to be zeroed. */
1387 	CAS_WRITE_4(sc, CAS_MAC_NORM_COLL_CNT, 0);
1388 	CAS_WRITE_4(sc, CAS_MAC_FIRST_COLL_CNT, 0);
1389 	CAS_WRITE_4(sc, CAS_MAC_EXCESS_COLL_CNT, 0);
1390 	CAS_WRITE_4(sc, CAS_MAC_LATE_COLL_CNT, 0);
1391 	CAS_WRITE_4(sc, CAS_MAC_DEFER_TMR_CNT, 0);
1392 	CAS_WRITE_4(sc, CAS_MAC_PEAK_ATTEMPTS, 0);
1393 	CAS_WRITE_4(sc, CAS_MAC_RX_FRAME_COUNT, 0);
1394 	CAS_WRITE_4(sc, CAS_MAC_RX_LEN_ERR_CNT, 0);
1395 	CAS_WRITE_4(sc, CAS_MAC_RX_ALIGN_ERR, 0);
1396 	CAS_WRITE_4(sc, CAS_MAC_RX_CRC_ERR_CNT, 0);
1397 	CAS_WRITE_4(sc, CAS_MAC_RX_CODE_VIOL, 0);
1398 
1399 	/* Set XOFF PAUSE time. */
1400 	CAS_WRITE_4(sc, CAS_MAC_SPC, 0x1BF0 << CAS_MAC_SPC_TIME_SHFT);
1401 
1402 	/* Set the station address. */
1403 	CAS_WRITE_4(sc, CAS_MAC_ADDR0, (laddr[4] << 8) | laddr[5]);
1404 	CAS_WRITE_4(sc, CAS_MAC_ADDR1, (laddr[2] << 8) | laddr[3]);
1405 	CAS_WRITE_4(sc, CAS_MAC_ADDR2, (laddr[0] << 8) | laddr[1]);
1406 
1407 	/* Enable MII outputs. */
1408 	CAS_WRITE_4(sc, CAS_MAC_XIF_CONF, CAS_MAC_XIF_CONF_TX_OE);
1409 }
1410 
1411 static void
1412 cas_tx_task(void *arg, int pending __unused)
1413 {
1414 	struct ifnet *ifp;
1415 
1416 	ifp = (struct ifnet *)arg;
1417 	cas_start(ifp);
1418 }
1419 
1420 static inline void
1421 cas_txkick(struct cas_softc *sc)
1422 {
1423 
1424 	/*
1425 	 * Update the TX kick register.  This register has to point to the
1426 	 * descriptor after the last valid one and for optimum performance
1427 	 * should be incremented in multiples of 4 (the DMA engine fetches/
1428 	 * updates descriptors in batches of 4).
1429 	 */
1430 #ifdef CAS_DEBUG
1431 	CTR3(KTR_CAS, "%s: %s: kicking TX %d",
1432 	    device_get_name(sc->sc_dev), __func__, sc->sc_txnext);
1433 #endif
1434 	CAS_CDSYNC(sc, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1435 	CAS_WRITE_4(sc, CAS_TX_KICK3, sc->sc_txnext);
1436 }
1437 
1438 static void
1439 cas_start(struct ifnet *ifp)
1440 {
1441 	struct cas_softc *sc = ifp->if_softc;
1442 	struct mbuf *m;
1443 	int kicked, ntx;
1444 
1445 	CAS_LOCK(sc);
1446 
1447 	if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
1448 	    IFF_DRV_RUNNING || (sc->sc_flags & CAS_LINK) == 0) {
1449 		CAS_UNLOCK(sc);
1450 		return;
1451 	}
1452 
1453 	if (sc->sc_txfree < CAS_MAXTXFREE / 4)
1454 		cas_tint(sc);
1455 
1456 #ifdef CAS_DEBUG
1457 	CTR4(KTR_CAS, "%s: %s: txfree %d, txnext %d",
1458 	    device_get_name(sc->sc_dev), __func__, sc->sc_txfree,
1459 	    sc->sc_txnext);
1460 #endif
1461 	ntx = 0;
1462 	kicked = 0;
1463 	for (; !IFQ_DRV_IS_EMPTY(&ifp->if_snd) && sc->sc_txfree > 1;) {
1464 		IFQ_DRV_DEQUEUE(&ifp->if_snd, m);
1465 		if (m == NULL)
1466 			break;
1467 		if (cas_load_txmbuf(sc, &m) != 0) {
1468 			if (m == NULL)
1469 				break;
1470 			ifp->if_drv_flags |= IFF_DRV_OACTIVE;
1471 			IFQ_DRV_PREPEND(&ifp->if_snd, m);
1472 			break;
1473 		}
1474 		if ((sc->sc_txnext % 4) == 0) {
1475 			cas_txkick(sc);
1476 			kicked = 1;
1477 		} else
1478 			kicked = 0;
1479 		ntx++;
1480 		BPF_MTAP(ifp, m);
1481 	}
1482 
1483 	if (ntx > 0) {
1484 		if (kicked == 0)
1485 			cas_txkick(sc);
1486 #ifdef CAS_DEBUG
1487 		CTR2(KTR_CAS, "%s: packets enqueued, OWN on %d",
1488 		    device_get_name(sc->sc_dev), sc->sc_txnext);
1489 #endif
1490 
1491 		/* Set a watchdog timer in case the chip flakes out. */
1492 		sc->sc_wdog_timer = 5;
1493 #ifdef CAS_DEBUG
1494 		CTR3(KTR_CAS, "%s: %s: watchdog %d",
1495 		    device_get_name(sc->sc_dev), __func__,
1496 		    sc->sc_wdog_timer);
1497 #endif
1498 	}
1499 
1500 	CAS_UNLOCK(sc);
1501 }
1502 
1503 static void
1504 cas_tint(struct cas_softc *sc)
1505 {
1506 	struct ifnet *ifp = sc->sc_ifp;
1507 	struct cas_txsoft *txs;
1508 	int progress;
1509 	uint32_t txlast;
1510 #ifdef CAS_DEBUG
1511 	int i;
1512 
1513 	CAS_LOCK_ASSERT(sc, MA_OWNED);
1514 
1515 	CTR2(KTR_CAS, "%s: %s", device_get_name(sc->sc_dev), __func__);
1516 #endif
1517 
1518 	/*
1519 	 * Go through our TX list and free mbufs for those
1520 	 * frames that have been transmitted.
1521 	 */
1522 	progress = 0;
1523 	CAS_CDSYNC(sc, BUS_DMASYNC_POSTREAD);
1524 	while ((txs = STAILQ_FIRST(&sc->sc_txdirtyq)) != NULL) {
1525 #ifdef CAS_DEBUG
1526 		if ((ifp->if_flags & IFF_DEBUG) != 0) {
1527 			printf("    txsoft %p transmit chain:\n", txs);
1528 			for (i = txs->txs_firstdesc;; i = CAS_NEXTTX(i)) {
1529 				printf("descriptor %d: ", i);
1530 				printf("cd_flags: 0x%016llx\t",
1531 				    (long long)le64toh(
1532 				    sc->sc_txdescs[i].cd_flags));
1533 				printf("cd_buf_ptr: 0x%016llx\n",
1534 				    (long long)le64toh(
1535 				    sc->sc_txdescs[i].cd_buf_ptr));
1536 				if (i == txs->txs_lastdesc)
1537 					break;
1538 			}
1539 		}
1540 #endif
1541 
1542 		/*
1543 		 * In theory, we could harvest some descriptors before
1544 		 * the ring is empty, but that's a bit complicated.
1545 		 *
1546 		 * CAS_TX_COMPn points to the last descriptor
1547 		 * processed + 1.
1548 		 */
1549 		txlast = CAS_READ_4(sc, CAS_TX_COMP3);
1550 #ifdef CAS_DEBUG
1551 		CTR4(KTR_CAS, "%s: txs->txs_firstdesc = %d, "
1552 		    "txs->txs_lastdesc = %d, txlast = %d",
1553 		    __func__, txs->txs_firstdesc, txs->txs_lastdesc, txlast);
1554 #endif
1555 		if (txs->txs_firstdesc <= txs->txs_lastdesc) {
1556 			if ((txlast >= txs->txs_firstdesc) &&
1557 			    (txlast <= txs->txs_lastdesc))
1558 				break;
1559 		} else {
1560 			/* Ick -- this command wraps. */
1561 			if ((txlast >= txs->txs_firstdesc) ||
1562 			    (txlast <= txs->txs_lastdesc))
1563 				break;
1564 		}
1565 
1566 #ifdef CAS_DEBUG
1567 		CTR1(KTR_CAS, "%s: releasing a descriptor", __func__);
1568 #endif
1569 		STAILQ_REMOVE_HEAD(&sc->sc_txdirtyq, txs_q);
1570 
1571 		sc->sc_txfree += txs->txs_ndescs;
1572 
1573 		bus_dmamap_sync(sc->sc_tdmatag, txs->txs_dmamap,
1574 		    BUS_DMASYNC_POSTWRITE);
1575 		bus_dmamap_unload(sc->sc_tdmatag, txs->txs_dmamap);
1576 		if (txs->txs_mbuf != NULL) {
1577 			m_freem(txs->txs_mbuf);
1578 			txs->txs_mbuf = NULL;
1579 		}
1580 
1581 		STAILQ_INSERT_TAIL(&sc->sc_txfreeq, txs, txs_q);
1582 
1583 		ifp->if_opackets++;
1584 		progress = 1;
1585 	}
1586 
1587 #ifdef CAS_DEBUG
1588 	CTR5(KTR_CAS, "%s: CAS_TX_SM1 %x CAS_TX_SM2 %x CAS_TX_DESC_BASE %llx "
1589 	    "CAS_TX_COMP3 %x",
1590 	    __func__, CAS_READ_4(sc, CAS_TX_SM1), CAS_READ_4(sc, CAS_TX_SM2),
1591 	    ((long long)CAS_READ_4(sc, CAS_TX_DESC3_BASE_HI) << 32) |
1592 	    CAS_READ_4(sc, CAS_TX_DESC3_BASE_LO),
1593 	    CAS_READ_4(sc, CAS_TX_COMP3));
1594 #endif
1595 
1596 	if (progress) {
1597 		/* We freed some descriptors, so reset IFF_DRV_OACTIVE. */
1598 		ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1599 		if (STAILQ_EMPTY(&sc->sc_txdirtyq))
1600 			sc->sc_wdog_timer = 0;
1601 	}
1602 
1603 #ifdef CAS_DEBUG
1604 	CTR3(KTR_CAS, "%s: %s: watchdog %d",
1605 	    device_get_name(sc->sc_dev), __func__, sc->sc_wdog_timer);
1606 #endif
1607 }
1608 
1609 static void
1610 cas_rint_timeout(void *arg)
1611 {
1612 	struct cas_softc *sc = arg;
1613 
1614 	CAS_LOCK_ASSERT(sc, MA_OWNED);
1615 
1616 	cas_rint(sc);
1617 }
1618 
1619 static void
1620 cas_rint(struct cas_softc *sc)
1621 {
1622 	struct cas_rxdsoft *rxds, *rxds2;
1623 	struct ifnet *ifp = sc->sc_ifp;
1624 	struct mbuf *m, *m2;
1625 	uint64_t word1, word2, word3, word4;
1626 	uint32_t rxhead;
1627 	u_int idx, idx2, len, off, skip;
1628 
1629 	CAS_LOCK_ASSERT(sc, MA_OWNED);
1630 
1631 	callout_stop(&sc->sc_rx_ch);
1632 
1633 #ifdef CAS_DEBUG
1634 	CTR2(KTR_CAS, "%s: %s", device_get_name(sc->sc_dev), __func__);
1635 #endif
1636 
1637 #define	PRINTWORD(n, delimiter)						\
1638 	printf("word ## n: 0x%016llx%c", (long long)word ## n, delimiter)
1639 
1640 #define	SKIPASSERT(n)							\
1641 	KASSERT(sc->sc_rxcomps[sc->sc_rxcptr].crc_word ## n == 0,	\
1642 	    ("%s: word ## n not 0", __func__))
1643 
1644 #define	WORDTOH(n)							\
1645 	word ## n = le64toh(sc->sc_rxcomps[sc->sc_rxcptr].crc_word ## n)
1646 
1647 	/*
1648 	 * Read the completion head register once.  This limits
1649 	 * how long the following loop can execute.
1650 	 */
1651 	rxhead = CAS_READ_4(sc, CAS_RX_COMP_HEAD);
1652 #ifdef CAS_DEBUG
1653 	CTR4(KTR_CAS, "%s: sc->sc_rxcptr %d, sc->sc_rxdptr %d, head %d",
1654 	    __func__, sc->sc_rxcptr, sc->sc_rxdptr, rxhead);
1655 #endif
1656 	skip = 0;
1657 	CAS_CDSYNC(sc, BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
1658 	for (; sc->sc_rxcptr != rxhead;
1659 	    sc->sc_rxcptr = CAS_NEXTRXCOMP(sc->sc_rxcptr)) {
1660 		if (skip != 0) {
1661 			SKIPASSERT(1);
1662 			SKIPASSERT(2);
1663 			SKIPASSERT(3);
1664 
1665 			--skip;
1666 			goto skip;
1667 		}
1668 
1669 		WORDTOH(1);
1670 		WORDTOH(2);
1671 		WORDTOH(3);
1672 		WORDTOH(4);
1673 
1674 #ifdef CAS_DEBUG
1675 		if ((ifp->if_flags & IFF_DEBUG) != 0) {
1676 			printf("    completion %d: ", sc->sc_rxcptr);
1677 			PRINTWORD(1, '\t');
1678 			PRINTWORD(2, '\t');
1679 			PRINTWORD(3, '\t');
1680 			PRINTWORD(4, '\n');
1681 		}
1682 #endif
1683 
1684 		if (__predict_false(
1685 		    (word1 & CAS_RC1_TYPE_MASK) == CAS_RC1_TYPE_HW ||
1686 		    (word4 & CAS_RC4_ZERO) != 0)) {
1687 			/*
1688 			 * The descriptor is still marked as owned, although
1689 			 * it is supposed to have completed.  This has been
1690 			 * observed on some machines.  Just exiting here
1691 			 * might leave the packet sitting around until another
1692 			 * one arrives to trigger a new interrupt, which is
1693 			 * generally undesirable, so set up a timeout.
1694 			 */
1695 			callout_reset(&sc->sc_rx_ch, CAS_RXOWN_TICKS,
1696 			    cas_rint_timeout, sc);
1697 			break;
1698 		}
1699 
1700 		if (__predict_false(
1701 		    (word4 & (CAS_RC4_BAD | CAS_RC4_LEN_MMATCH)) != 0)) {
1702 			ifp->if_ierrors++;
1703 			device_printf(sc->sc_dev,
1704 			    "receive error: CRC error\n");
1705 			continue;
1706 		}
1707 
1708 		KASSERT(CAS_GET(word1, CAS_RC1_DATA_SIZE) == 0 ||
1709 		    CAS_GET(word2, CAS_RC2_HDR_SIZE) == 0,
1710 		    ("%s: data and header present", __func__));
1711 		KASSERT((word1 & CAS_RC1_SPLIT_PKT) == 0 ||
1712 		    CAS_GET(word2, CAS_RC2_HDR_SIZE) == 0,
1713 		    ("%s: split and header present", __func__));
1714 		KASSERT(CAS_GET(word1, CAS_RC1_DATA_SIZE) == 0 ||
1715 		    (word1 & CAS_RC1_RELEASE_HDR) == 0,
1716 		    ("%s: data present but header release", __func__));
1717 		KASSERT(CAS_GET(word2, CAS_RC2_HDR_SIZE) == 0 ||
1718 		    (word1 & CAS_RC1_RELEASE_DATA) == 0,
1719 		    ("%s: header present but data release", __func__));
1720 
1721 		if ((len = CAS_GET(word2, CAS_RC2_HDR_SIZE)) != 0) {
1722 			idx = CAS_GET(word2, CAS_RC2_HDR_INDEX);
1723 			off = CAS_GET(word2, CAS_RC2_HDR_OFF);
1724 #ifdef CAS_DEBUG
1725 			CTR4(KTR_CAS, "%s: hdr at idx %d, off %d, len %d",
1726 			    __func__, idx, off, len);
1727 #endif
1728 			rxds = &sc->sc_rxdsoft[idx];
1729 			MGETHDR(m, M_NOWAIT, MT_DATA);
1730 			if (m != NULL) {
1731 				refcount_acquire(&rxds->rxds_refcount);
1732 				bus_dmamap_sync(sc->sc_rdmatag,
1733 				    rxds->rxds_dmamap, BUS_DMASYNC_POSTREAD);
1734 #if __FreeBSD_version < 800016
1735 				MEXTADD(m, (caddr_t)rxds->rxds_buf +
1736 				    off * 256 + ETHER_ALIGN, len, cas_free,
1737 				    rxds, M_RDONLY, EXT_NET_DRV);
1738 #else
1739 				MEXTADD(m, (caddr_t)rxds->rxds_buf +
1740 				    off * 256 + ETHER_ALIGN, len, cas_free,
1741 				    sc, (void *)(uintptr_t)idx,
1742 				    M_RDONLY, EXT_NET_DRV);
1743 #endif
1744 				if ((m->m_flags & M_EXT) == 0) {
1745 					m_freem(m);
1746 					m = NULL;
1747 				}
1748 			}
1749 			if (m != NULL) {
1750 				m->m_pkthdr.rcvif = ifp;
1751 				m->m_pkthdr.len = m->m_len = len;
1752 				ifp->if_ipackets++;
1753 				if ((ifp->if_capenable & IFCAP_RXCSUM) != 0)
1754 					cas_rxcksum(m, CAS_GET(word4,
1755 					    CAS_RC4_TCP_CSUM));
1756 				/* Pass it on. */
1757 				CAS_UNLOCK(sc);
1758 				(*ifp->if_input)(ifp, m);
1759 				CAS_LOCK(sc);
1760 			} else
1761 				ifp->if_iqdrops++;
1762 
1763 			if ((word1 & CAS_RC1_RELEASE_HDR) != 0 &&
1764 			    refcount_release(&rxds->rxds_refcount) != 0)
1765 				cas_add_rxdesc(sc, idx);
1766 		} else if ((len = CAS_GET(word1, CAS_RC1_DATA_SIZE)) != 0) {
1767 			idx = CAS_GET(word1, CAS_RC1_DATA_INDEX);
1768 			off = CAS_GET(word1, CAS_RC1_DATA_OFF);
1769 #ifdef CAS_DEBUG
1770 			CTR4(KTR_CAS, "%s: data at idx %d, off %d, len %d",
1771 			    __func__, idx, off, len);
1772 #endif
1773 			rxds = &sc->sc_rxdsoft[idx];
1774 			MGETHDR(m, M_NOWAIT, MT_DATA);
1775 			if (m != NULL) {
1776 				refcount_acquire(&rxds->rxds_refcount);
1777 				off += ETHER_ALIGN;
1778 				m->m_len = min(CAS_PAGE_SIZE - off, len);
1779 				bus_dmamap_sync(sc->sc_rdmatag,
1780 				    rxds->rxds_dmamap, BUS_DMASYNC_POSTREAD);
1781 #if __FreeBSD_version < 800016
1782 				MEXTADD(m, (caddr_t)rxds->rxds_buf + off,
1783 				    m->m_len, cas_free, rxds, M_RDONLY,
1784 				    EXT_NET_DRV);
1785 #else
1786 				MEXTADD(m, (caddr_t)rxds->rxds_buf + off,
1787 				    m->m_len, cas_free, sc,
1788 				    (void *)(uintptr_t)idx, M_RDONLY,
1789 				    EXT_NET_DRV);
1790 #endif
1791 				if ((m->m_flags & M_EXT) == 0) {
1792 					m_freem(m);
1793 					m = NULL;
1794 				}
1795 			}
1796 			idx2 = 0;
1797 			m2 = NULL;
1798 			rxds2 = NULL;
1799 			if ((word1 & CAS_RC1_SPLIT_PKT) != 0) {
1800 				KASSERT((word1 & CAS_RC1_RELEASE_NEXT) != 0,
1801 				    ("%s: split but no release next",
1802 				    __func__));
1803 
1804 				idx2 = CAS_GET(word2, CAS_RC2_NEXT_INDEX);
1805 #ifdef CAS_DEBUG
1806 				CTR2(KTR_CAS, "%s: split at idx %d",
1807 				    __func__, idx2);
1808 #endif
1809 				rxds2 = &sc->sc_rxdsoft[idx2];
1810 				if (m != NULL) {
1811 					MGET(m2, M_NOWAIT, MT_DATA);
1812 					if (m2 != NULL) {
1813 						refcount_acquire(
1814 						    &rxds2->rxds_refcount);
1815 						m2->m_len = len - m->m_len;
1816 						bus_dmamap_sync(
1817 						    sc->sc_rdmatag,
1818 						    rxds2->rxds_dmamap,
1819 						    BUS_DMASYNC_POSTREAD);
1820 #if __FreeBSD_version < 800016
1821 						MEXTADD(m2,
1822 						    (caddr_t)rxds2->rxds_buf,
1823 						    m2->m_len, cas_free,
1824 						    rxds2, M_RDONLY,
1825 						    EXT_NET_DRV);
1826 #else
1827 						MEXTADD(m2,
1828 						    (caddr_t)rxds2->rxds_buf,
1829 						    m2->m_len, cas_free, sc,
1830 						    (void *)(uintptr_t)idx2,
1831 						    M_RDONLY, EXT_NET_DRV);
1832 #endif
1833 						if ((m2->m_flags & M_EXT) ==
1834 						    0) {
1835 							m_freem(m2);
1836 							m2 = NULL;
1837 						}
1838 					}
1839 				}
1840 				if (m2 != NULL)
1841 					m->m_next = m2;
1842 				else if (m != NULL) {
1843 					m_freem(m);
1844 					m = NULL;
1845 				}
1846 			}
1847 			if (m != NULL) {
1848 				m->m_pkthdr.rcvif = ifp;
1849 				m->m_pkthdr.len = len;
1850 				ifp->if_ipackets++;
1851 				if ((ifp->if_capenable & IFCAP_RXCSUM) != 0)
1852 					cas_rxcksum(m, CAS_GET(word4,
1853 					    CAS_RC4_TCP_CSUM));
1854 				/* Pass it on. */
1855 				CAS_UNLOCK(sc);
1856 				(*ifp->if_input)(ifp, m);
1857 				CAS_LOCK(sc);
1858 			} else
1859 				ifp->if_iqdrops++;
1860 
1861 			if ((word1 & CAS_RC1_RELEASE_DATA) != 0 &&
1862 			    refcount_release(&rxds->rxds_refcount) != 0)
1863 				cas_add_rxdesc(sc, idx);
1864 			if ((word1 & CAS_RC1_SPLIT_PKT) != 0 &&
1865 			    refcount_release(&rxds2->rxds_refcount) != 0)
1866 				cas_add_rxdesc(sc, idx2);
1867 		}
1868 
1869 		skip = CAS_GET(word1, CAS_RC1_SKIP);
1870 
1871  skip:
1872 		cas_rxcompinit(&sc->sc_rxcomps[sc->sc_rxcptr]);
1873 		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1874 			break;
1875 	}
1876 	CAS_CDSYNC(sc, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1877 	CAS_WRITE_4(sc, CAS_RX_COMP_TAIL, sc->sc_rxcptr);
1878 
1879 #undef PRINTWORD
1880 #undef SKIPASSERT
1881 #undef WORDTOH
1882 
1883 #ifdef CAS_DEBUG
1884 	CTR4(KTR_CAS, "%s: done sc->sc_rxcptr %d, sc->sc_rxdptr %d, head %d",
1885 	    __func__, sc->sc_rxcptr, sc->sc_rxdptr,
1886 	    CAS_READ_4(sc, CAS_RX_COMP_HEAD));
1887 #endif
1888 }
1889 
1890 static void
1891 cas_free(void *arg1, void *arg2)
1892 {
1893 	struct cas_rxdsoft *rxds;
1894 	struct cas_softc *sc;
1895 	u_int idx, locked;
1896 
1897 #if __FreeBSD_version < 800016
1898 	rxds = arg2;
1899 	sc = rxds->rxds_sc;
1900 	idx = rxds->rxds_idx;
1901 #else
1902 	sc = arg1;
1903 	idx = (uintptr_t)arg2;
1904 	rxds = &sc->sc_rxdsoft[idx];
1905 #endif
1906 	if (refcount_release(&rxds->rxds_refcount) == 0)
1907 		return;
1908 
1909 	/*
1910 	 * NB: this function can be called via m_freem(9) within
1911 	 * this driver!
1912 	 */
1913 	if ((locked = CAS_LOCK_OWNED(sc)) == 0)
1914 		CAS_LOCK(sc);
1915 	cas_add_rxdesc(sc, idx);
1916 	if (locked == 0)
1917 		CAS_UNLOCK(sc);
1918 }
1919 
1920 static inline void
1921 cas_add_rxdesc(struct cas_softc *sc, u_int idx)
1922 {
1923 
1924 	CAS_LOCK_ASSERT(sc, MA_OWNED);
1925 
1926 	bus_dmamap_sync(sc->sc_rdmatag, sc->sc_rxdsoft[idx].rxds_dmamap,
1927 	    BUS_DMASYNC_PREREAD);
1928 	CAS_UPDATE_RXDESC(sc, sc->sc_rxdptr, idx);
1929 	sc->sc_rxdptr = CAS_NEXTRXDESC(sc->sc_rxdptr);
1930 
1931 	/*
1932 	 * Update the RX kick register.  This register has to point to the
1933 	 * descriptor after the last valid one (before the current batch)
1934 	 * and for optimum performance should be incremented in multiples
1935 	 * of 4 (the DMA engine fetches/updates descriptors in batches of 4).
1936 	 */
1937 	if ((sc->sc_rxdptr % 4) == 0) {
1938 		CAS_CDSYNC(sc, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1939 		CAS_WRITE_4(sc, CAS_RX_KICK,
1940 		    (sc->sc_rxdptr + CAS_NRXDESC - 4) & CAS_NRXDESC_MASK);
1941 	}
1942 }
1943 
1944 static void
1945 cas_eint(struct cas_softc *sc, u_int status)
1946 {
1947 	struct ifnet *ifp = sc->sc_ifp;
1948 
1949 	CAS_LOCK_ASSERT(sc, MA_OWNED);
1950 
1951 	ifp->if_ierrors++;
1952 
1953 	device_printf(sc->sc_dev, "%s: status 0x%x", __func__, status);
1954 	if ((status & CAS_INTR_PCI_ERROR_INT) != 0) {
1955 		status = CAS_READ_4(sc, CAS_ERROR_STATUS);
1956 		printf(", PCI bus error 0x%x", status);
1957 		if ((status & CAS_ERROR_OTHER) != 0) {
1958 			status = pci_read_config(sc->sc_dev, PCIR_STATUS, 2);
1959 			printf(", PCI status 0x%x", status);
1960 			pci_write_config(sc->sc_dev, PCIR_STATUS, status, 2);
1961 		}
1962 	}
1963 	printf("\n");
1964 
1965 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1966 	cas_init_locked(sc);
1967 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
1968 		taskqueue_enqueue(sc->sc_tq, &sc->sc_tx_task);
1969 }
1970 
1971 static int
1972 cas_intr(void *v)
1973 {
1974 	struct cas_softc *sc = v;
1975 
1976 	if (__predict_false((CAS_READ_4(sc, CAS_STATUS_ALIAS) &
1977 	    CAS_INTR_SUMMARY) == 0))
1978 		return (FILTER_STRAY);
1979 
1980 	/* Disable interrupts. */
1981 	CAS_WRITE_4(sc, CAS_INTMASK, 0xffffffff);
1982 	taskqueue_enqueue(sc->sc_tq, &sc->sc_intr_task);
1983 
1984 	return (FILTER_HANDLED);
1985 }
1986 
1987 static void
1988 cas_intr_task(void *arg, int pending __unused)
1989 {
1990 	struct cas_softc *sc = arg;
1991 	struct ifnet *ifp = sc->sc_ifp;
1992 	uint32_t status, status2;
1993 
1994 	CAS_LOCK_ASSERT(sc, MA_NOTOWNED);
1995 
1996 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1997 		return;
1998 
1999 	status = CAS_READ_4(sc, CAS_STATUS);
2000 	if (__predict_false((status & CAS_INTR_SUMMARY) == 0))
2001 		goto done;
2002 
2003 	CAS_LOCK(sc);
2004 #ifdef CAS_DEBUG
2005 	CTR4(KTR_CAS, "%s: %s: cplt %x, status %x",
2006 	    device_get_name(sc->sc_dev), __func__,
2007 	    (status >> CAS_STATUS_TX_COMP3_SHFT), (u_int)status);
2008 
2009 	/*
2010 	 * PCS interrupts must be cleared, otherwise no traffic is passed!
2011 	 */
2012 	if ((status & CAS_INTR_PCS_INT) != 0) {
2013 		status2 =
2014 		    CAS_READ_4(sc, CAS_PCS_INTR_STATUS) |
2015 		    CAS_READ_4(sc, CAS_PCS_INTR_STATUS);
2016 		if ((status2 & CAS_PCS_INTR_LINK) != 0)
2017 			device_printf(sc->sc_dev,
2018 			    "%s: PCS link status changed\n", __func__);
2019 	}
2020 	if ((status & CAS_MAC_CTRL_STATUS) != 0) {
2021 		status2 = CAS_READ_4(sc, CAS_MAC_CTRL_STATUS);
2022 		if ((status2 & CAS_MAC_CTRL_PAUSE) != 0)
2023 			device_printf(sc->sc_dev,
2024 			    "%s: PAUSE received (PAUSE time %d slots)\n",
2025 			    __func__,
2026 			    (status2 & CAS_MAC_CTRL_STATUS_PT_MASK) >>
2027 			    CAS_MAC_CTRL_STATUS_PT_SHFT);
2028 		if ((status2 & CAS_MAC_CTRL_PAUSE) != 0)
2029 			device_printf(sc->sc_dev,
2030 			    "%s: transited to PAUSE state\n", __func__);
2031 		if ((status2 & CAS_MAC_CTRL_NON_PAUSE) != 0)
2032 			device_printf(sc->sc_dev,
2033 			    "%s: transited to non-PAUSE state\n", __func__);
2034 	}
2035 	if ((status & CAS_INTR_MIF) != 0)
2036 		device_printf(sc->sc_dev, "%s: MIF interrupt\n", __func__);
2037 #endif
2038 
2039 	if (__predict_false((status &
2040 	    (CAS_INTR_TX_TAG_ERR | CAS_INTR_RX_TAG_ERR |
2041 	    CAS_INTR_RX_LEN_MMATCH | CAS_INTR_PCI_ERROR_INT)) != 0)) {
2042 		cas_eint(sc, status);
2043 		CAS_UNLOCK(sc);
2044 		return;
2045 	}
2046 
2047 	if (__predict_false(status & CAS_INTR_TX_MAC_INT)) {
2048 		status2 = CAS_READ_4(sc, CAS_MAC_TX_STATUS);
2049 		if ((status2 &
2050 		    (CAS_MAC_TX_UNDERRUN | CAS_MAC_TX_MAX_PKT_ERR)) != 0)
2051 			ifp->if_oerrors++;
2052 		else if ((status2 & ~CAS_MAC_TX_FRAME_XMTD) != 0)
2053 			device_printf(sc->sc_dev,
2054 			    "MAC TX fault, status %x\n", status2);
2055 	}
2056 
2057 	if (__predict_false(status & CAS_INTR_RX_MAC_INT)) {
2058 		status2 = CAS_READ_4(sc, CAS_MAC_RX_STATUS);
2059 		if ((status2 & CAS_MAC_RX_OVERFLOW) != 0)
2060 			ifp->if_ierrors++;
2061 		else if ((status2 & ~CAS_MAC_RX_FRAME_RCVD) != 0)
2062 			device_printf(sc->sc_dev,
2063 			    "MAC RX fault, status %x\n", status2);
2064 	}
2065 
2066 	if ((status &
2067 	    (CAS_INTR_RX_DONE | CAS_INTR_RX_BUF_NA | CAS_INTR_RX_COMP_FULL |
2068 	    CAS_INTR_RX_BUF_AEMPTY | CAS_INTR_RX_COMP_AFULL)) != 0) {
2069 		cas_rint(sc);
2070 #ifdef CAS_DEBUG
2071 		if (__predict_false((status &
2072 		    (CAS_INTR_RX_BUF_NA | CAS_INTR_RX_COMP_FULL |
2073 		    CAS_INTR_RX_BUF_AEMPTY | CAS_INTR_RX_COMP_AFULL)) != 0))
2074 			device_printf(sc->sc_dev,
2075 			    "RX fault, status %x\n", status);
2076 #endif
2077 	}
2078 
2079 	if ((status &
2080 	    (CAS_INTR_TX_INT_ME | CAS_INTR_TX_ALL | CAS_INTR_TX_DONE)) != 0)
2081 		cas_tint(sc);
2082 
2083 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2084 		CAS_UNLOCK(sc);
2085 		return;
2086 	} else if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2087 		taskqueue_enqueue(sc->sc_tq, &sc->sc_tx_task);
2088 	CAS_UNLOCK(sc);
2089 
2090 	status = CAS_READ_4(sc, CAS_STATUS_ALIAS);
2091 	if (__predict_false((status & CAS_INTR_SUMMARY) != 0)) {
2092 		taskqueue_enqueue(sc->sc_tq, &sc->sc_intr_task);
2093 		return;
2094 	}
2095 
2096  done:
2097 	/* Re-enable interrupts. */
2098 	CAS_WRITE_4(sc, CAS_INTMASK,
2099 	    ~(CAS_INTR_TX_INT_ME | CAS_INTR_TX_TAG_ERR |
2100 	    CAS_INTR_RX_DONE | CAS_INTR_RX_BUF_NA | CAS_INTR_RX_TAG_ERR |
2101 	    CAS_INTR_RX_COMP_FULL | CAS_INTR_RX_BUF_AEMPTY |
2102 	    CAS_INTR_RX_COMP_AFULL | CAS_INTR_RX_LEN_MMATCH |
2103 	    CAS_INTR_PCI_ERROR_INT
2104 #ifdef CAS_DEBUG
2105 	    | CAS_INTR_PCS_INT | CAS_INTR_MIF
2106 #endif
2107 	));
2108 }
2109 
2110 static void
2111 cas_watchdog(struct cas_softc *sc)
2112 {
2113 	struct ifnet *ifp = sc->sc_ifp;
2114 
2115 	CAS_LOCK_ASSERT(sc, MA_OWNED);
2116 
2117 #ifdef CAS_DEBUG
2118 	CTR4(KTR_CAS,
2119 	    "%s: CAS_RX_CONF %x CAS_MAC_RX_STATUS %x CAS_MAC_RX_CONF %x",
2120 	    __func__, CAS_READ_4(sc, CAS_RX_CONF),
2121 	    CAS_READ_4(sc, CAS_MAC_RX_STATUS),
2122 	    CAS_READ_4(sc, CAS_MAC_RX_CONF));
2123 	CTR4(KTR_CAS,
2124 	    "%s: CAS_TX_CONF %x CAS_MAC_TX_STATUS %x CAS_MAC_TX_CONF %x",
2125 	    __func__, CAS_READ_4(sc, CAS_TX_CONF),
2126 	    CAS_READ_4(sc, CAS_MAC_TX_STATUS),
2127 	    CAS_READ_4(sc, CAS_MAC_TX_CONF));
2128 #endif
2129 
2130 	if (sc->sc_wdog_timer == 0 || --sc->sc_wdog_timer != 0)
2131 		return;
2132 
2133 	if ((sc->sc_flags & CAS_LINK) != 0)
2134 		device_printf(sc->sc_dev, "device timeout\n");
2135 	else if (bootverbose)
2136 		device_printf(sc->sc_dev, "device timeout (no link)\n");
2137 	++ifp->if_oerrors;
2138 
2139 	/* Try to get more packets going. */
2140 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2141 	cas_init_locked(sc);
2142 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2143 		taskqueue_enqueue(sc->sc_tq, &sc->sc_tx_task);
2144 }
2145 
2146 static void
2147 cas_mifinit(struct cas_softc *sc)
2148 {
2149 
2150 	/* Configure the MIF in frame mode. */
2151 	CAS_WRITE_4(sc, CAS_MIF_CONF,
2152 	    CAS_READ_4(sc, CAS_MIF_CONF) & ~CAS_MIF_CONF_BB_MODE);
2153 	CAS_BARRIER(sc, CAS_MIF_CONF, 4,
2154 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2155 }
2156 
2157 /*
2158  * MII interface
2159  *
2160  * The MII interface supports at least three different operating modes:
2161  *
2162  * Bitbang mode is implemented using data, clock and output enable registers.
2163  *
2164  * Frame mode is implemented by loading a complete frame into the frame
2165  * register and polling the valid bit for completion.
2166  *
2167  * Polling mode uses the frame register but completion is indicated by
2168  * an interrupt.
2169  *
2170  */
2171 static int
2172 cas_mii_readreg(device_t dev, int phy, int reg)
2173 {
2174 	struct cas_softc *sc;
2175 	int n;
2176 	uint32_t v;
2177 
2178 #ifdef CAS_DEBUG_PHY
2179 	printf("%s: phy %d reg %d\n", __func__, phy, reg);
2180 #endif
2181 
2182 	sc = device_get_softc(dev);
2183 	if ((sc->sc_flags & CAS_SERDES) != 0) {
2184 		switch (reg) {
2185 		case MII_BMCR:
2186 			reg = CAS_PCS_CTRL;
2187 			break;
2188 		case MII_BMSR:
2189 			reg = CAS_PCS_STATUS;
2190 			break;
2191 		case MII_PHYIDR1:
2192 		case MII_PHYIDR2:
2193 			return (0);
2194 		case MII_ANAR:
2195 			reg = CAS_PCS_ANAR;
2196 			break;
2197 		case MII_ANLPAR:
2198 			reg = CAS_PCS_ANLPAR;
2199 			break;
2200 		case MII_EXTSR:
2201 			return (EXTSR_1000XFDX | EXTSR_1000XHDX);
2202 		default:
2203 			device_printf(sc->sc_dev,
2204 			    "%s: unhandled register %d\n", __func__, reg);
2205 			return (0);
2206 		}
2207 		return (CAS_READ_4(sc, reg));
2208 	}
2209 
2210 	/* Construct the frame command. */
2211 	v = CAS_MIF_FRAME_READ |
2212 	    (phy << CAS_MIF_FRAME_PHY_SHFT) |
2213 	    (reg << CAS_MIF_FRAME_REG_SHFT);
2214 
2215 	CAS_WRITE_4(sc, CAS_MIF_FRAME, v);
2216 	CAS_BARRIER(sc, CAS_MIF_FRAME, 4,
2217 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2218 	for (n = 0; n < 100; n++) {
2219 		DELAY(1);
2220 		v = CAS_READ_4(sc, CAS_MIF_FRAME);
2221 		if (v & CAS_MIF_FRAME_TA_LSB)
2222 			return (v & CAS_MIF_FRAME_DATA);
2223 	}
2224 
2225 	device_printf(sc->sc_dev, "%s: timed out\n", __func__);
2226 	return (0);
2227 }
2228 
2229 static int
2230 cas_mii_writereg(device_t dev, int phy, int reg, int val)
2231 {
2232 	struct cas_softc *sc;
2233 	int n;
2234 	uint32_t v;
2235 
2236 #ifdef CAS_DEBUG_PHY
2237 	printf("%s: phy %d reg %d val %x\n", phy, reg, val, __func__);
2238 #endif
2239 
2240 	sc = device_get_softc(dev);
2241 	if ((sc->sc_flags & CAS_SERDES) != 0) {
2242 		switch (reg) {
2243 		case MII_BMSR:
2244 			reg = CAS_PCS_STATUS;
2245 			break;
2246 		case MII_BMCR:
2247 			reg = CAS_PCS_CTRL;
2248 			if ((val & CAS_PCS_CTRL_RESET) == 0)
2249 				break;
2250 			CAS_WRITE_4(sc, CAS_PCS_CTRL, val);
2251 			CAS_BARRIER(sc, CAS_PCS_CTRL, 4,
2252 			    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2253 			if (!cas_bitwait(sc, CAS_PCS_CTRL,
2254 			    CAS_PCS_CTRL_RESET, 0))
2255 				device_printf(sc->sc_dev,
2256 				    "cannot reset PCS\n");
2257 			/* FALLTHROUGH */
2258 		case MII_ANAR:
2259 			CAS_WRITE_4(sc, CAS_PCS_CONF, 0);
2260 			CAS_BARRIER(sc, CAS_PCS_CONF, 4,
2261 			    BUS_SPACE_BARRIER_WRITE);
2262 			CAS_WRITE_4(sc, CAS_PCS_ANAR, val);
2263 			CAS_BARRIER(sc, CAS_PCS_ANAR, 4,
2264 			    BUS_SPACE_BARRIER_WRITE);
2265 			CAS_WRITE_4(sc, CAS_PCS_SERDES_CTRL,
2266 			    CAS_PCS_SERDES_CTRL_ESD);
2267 			CAS_BARRIER(sc, CAS_PCS_CONF, 4,
2268 			    BUS_SPACE_BARRIER_WRITE);
2269 			CAS_WRITE_4(sc, CAS_PCS_CONF,
2270 			    CAS_PCS_CONF_EN);
2271 			CAS_BARRIER(sc, CAS_PCS_CONF, 4,
2272 			    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2273 			return (0);
2274 		case MII_ANLPAR:
2275 			reg = CAS_PCS_ANLPAR;
2276 			break;
2277 		default:
2278 			device_printf(sc->sc_dev,
2279 			    "%s: unhandled register %d\n", __func__, reg);
2280 			return (0);
2281 		}
2282 		CAS_WRITE_4(sc, reg, val);
2283 		CAS_BARRIER(sc, reg, 4,
2284 		    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2285 		return (0);
2286 	}
2287 
2288 	/* Construct the frame command. */
2289 	v = CAS_MIF_FRAME_WRITE |
2290 	    (phy << CAS_MIF_FRAME_PHY_SHFT) |
2291 	    (reg << CAS_MIF_FRAME_REG_SHFT) |
2292 	    (val & CAS_MIF_FRAME_DATA);
2293 
2294 	CAS_WRITE_4(sc, CAS_MIF_FRAME, v);
2295 	CAS_BARRIER(sc, CAS_MIF_FRAME, 4,
2296 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2297 	for (n = 0; n < 100; n++) {
2298 		DELAY(1);
2299 		v = CAS_READ_4(sc, CAS_MIF_FRAME);
2300 		if (v & CAS_MIF_FRAME_TA_LSB)
2301 			return (1);
2302 	}
2303 
2304 	device_printf(sc->sc_dev, "%s: timed out\n", __func__);
2305 	return (0);
2306 }
2307 
2308 static void
2309 cas_mii_statchg(device_t dev)
2310 {
2311 	struct cas_softc *sc;
2312 	struct ifnet *ifp;
2313 	int gigabit;
2314 	uint32_t rxcfg, txcfg, v;
2315 
2316 	sc = device_get_softc(dev);
2317 	ifp = sc->sc_ifp;
2318 
2319 	CAS_LOCK_ASSERT(sc, MA_OWNED);
2320 
2321 #ifdef CAS_DEBUG
2322 	if ((ifp->if_flags & IFF_DEBUG) != 0)
2323 		device_printf(sc->sc_dev, "%s: status changen", __func__);
2324 #endif
2325 
2326 	if ((sc->sc_mii->mii_media_status & IFM_ACTIVE) != 0 &&
2327 	    IFM_SUBTYPE(sc->sc_mii->mii_media_active) != IFM_NONE)
2328 		sc->sc_flags |= CAS_LINK;
2329 	else
2330 		sc->sc_flags &= ~CAS_LINK;
2331 
2332 	switch (IFM_SUBTYPE(sc->sc_mii->mii_media_active)) {
2333 	case IFM_1000_SX:
2334 	case IFM_1000_LX:
2335 	case IFM_1000_CX:
2336 	case IFM_1000_T:
2337 		gigabit = 1;
2338 		break;
2339 	default:
2340 		gigabit = 0;
2341 	}
2342 
2343 	/*
2344 	 * The configuration done here corresponds to the steps F) and
2345 	 * G) and as far as enabling of RX and TX MAC goes also step H)
2346 	 * of the initialization sequence outlined in section 11.2.1 of
2347 	 * the Cassini+ ASIC Specification.
2348 	 */
2349 
2350 	rxcfg = sc->sc_mac_rxcfg;
2351 	rxcfg &= ~CAS_MAC_RX_CONF_CARR;
2352 	txcfg = CAS_MAC_TX_CONF_EN_IPG0 | CAS_MAC_TX_CONF_NGU |
2353 	    CAS_MAC_TX_CONF_NGUL;
2354 	if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) & IFM_FDX) != 0)
2355 		txcfg |= CAS_MAC_TX_CONF_ICARR | CAS_MAC_TX_CONF_ICOLLIS;
2356 	else if (gigabit != 0) {
2357 		rxcfg |= CAS_MAC_RX_CONF_CARR;
2358 		txcfg |= CAS_MAC_TX_CONF_CARR;
2359 	}
2360 	(void)cas_disable_tx(sc);
2361 	CAS_WRITE_4(sc, CAS_MAC_TX_CONF, txcfg);
2362 	(void)cas_disable_rx(sc);
2363 	CAS_WRITE_4(sc, CAS_MAC_RX_CONF, rxcfg);
2364 
2365 	v = CAS_READ_4(sc, CAS_MAC_CTRL_CONF) &
2366 	    ~(CAS_MAC_CTRL_CONF_TXP | CAS_MAC_CTRL_CONF_RXP);
2367 	if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) &
2368 	    IFM_ETH_RXPAUSE) != 0)
2369 		v |= CAS_MAC_CTRL_CONF_RXP;
2370 	if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) &
2371 	    IFM_ETH_TXPAUSE) != 0)
2372 		v |= CAS_MAC_CTRL_CONF_TXP;
2373 	CAS_WRITE_4(sc, CAS_MAC_CTRL_CONF, v);
2374 
2375 	/*
2376 	 * All supported chips have a bug causing incorrect checksum
2377 	 * to be calculated when letting them strip the FCS in half-
2378 	 * duplex mode.  In theory we could disable FCS stripping and
2379 	 * manually adjust the checksum accordingly.  It seems to make
2380 	 * more sense to optimze for the common case and just disable
2381 	 * hardware checksumming in half-duplex mode though.
2382 	 */
2383 	if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) & IFM_FDX) == 0) {
2384 		ifp->if_capenable &= ~IFCAP_HWCSUM;
2385 		ifp->if_hwassist = 0;
2386 	} else if ((sc->sc_flags & CAS_NO_CSUM) == 0) {
2387 		ifp->if_capenable = ifp->if_capabilities;
2388 		ifp->if_hwassist = CAS_CSUM_FEATURES;
2389 	}
2390 
2391 	if (sc->sc_variant == CAS_SATURN) {
2392 		if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) & IFM_FDX) == 0)
2393 			/* silicon bug workaround */
2394 			CAS_WRITE_4(sc, CAS_MAC_PREAMBLE_LEN, 0x41);
2395 		else
2396 			CAS_WRITE_4(sc, CAS_MAC_PREAMBLE_LEN, 0x7);
2397 	}
2398 
2399 	if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) & IFM_FDX) == 0 &&
2400 	    gigabit != 0)
2401 		CAS_WRITE_4(sc, CAS_MAC_SLOT_TIME,
2402 		    CAS_MAC_SLOT_TIME_CARR);
2403 	else
2404 		CAS_WRITE_4(sc, CAS_MAC_SLOT_TIME,
2405 		    CAS_MAC_SLOT_TIME_NORM);
2406 
2407 	/* XIF Configuration */
2408 	v = CAS_MAC_XIF_CONF_TX_OE | CAS_MAC_XIF_CONF_LNKLED;
2409 	if ((sc->sc_flags & CAS_SERDES) == 0) {
2410 		if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) & IFM_FDX) == 0)
2411 			v |= CAS_MAC_XIF_CONF_NOECHO;
2412 		v |= CAS_MAC_XIF_CONF_BUF_OE;
2413 	}
2414 	if (gigabit != 0)
2415 		v |= CAS_MAC_XIF_CONF_GMII;
2416 	if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) & IFM_FDX) != 0)
2417 		v |= CAS_MAC_XIF_CONF_FDXLED;
2418 	CAS_WRITE_4(sc, CAS_MAC_XIF_CONF, v);
2419 
2420 	sc->sc_mac_rxcfg = rxcfg;
2421 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0 &&
2422 	    (sc->sc_flags & CAS_LINK) != 0) {
2423 		CAS_WRITE_4(sc, CAS_MAC_TX_CONF,
2424 		    txcfg | CAS_MAC_TX_CONF_EN);
2425 		CAS_WRITE_4(sc, CAS_MAC_RX_CONF,
2426 		    rxcfg | CAS_MAC_RX_CONF_EN);
2427 	}
2428 }
2429 
2430 static int
2431 cas_mediachange(struct ifnet *ifp)
2432 {
2433 	struct cas_softc *sc = ifp->if_softc;
2434 	int error;
2435 
2436 	/* XXX add support for serial media. */
2437 
2438 	CAS_LOCK(sc);
2439 	error = mii_mediachg(sc->sc_mii);
2440 	CAS_UNLOCK(sc);
2441 	return (error);
2442 }
2443 
2444 static void
2445 cas_mediastatus(struct ifnet *ifp, struct ifmediareq *ifmr)
2446 {
2447 	struct cas_softc *sc = ifp->if_softc;
2448 
2449 	CAS_LOCK(sc);
2450 	if ((ifp->if_flags & IFF_UP) == 0) {
2451 		CAS_UNLOCK(sc);
2452 		return;
2453 	}
2454 
2455 	mii_pollstat(sc->sc_mii);
2456 	ifmr->ifm_active = sc->sc_mii->mii_media_active;
2457 	ifmr->ifm_status = sc->sc_mii->mii_media_status;
2458 	CAS_UNLOCK(sc);
2459 }
2460 
2461 static int
2462 cas_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
2463 {
2464 	struct cas_softc *sc = ifp->if_softc;
2465 	struct ifreq *ifr = (struct ifreq *)data;
2466 	int error;
2467 
2468 	error = 0;
2469 	switch (cmd) {
2470 	case SIOCSIFFLAGS:
2471 		CAS_LOCK(sc);
2472 		if ((ifp->if_flags & IFF_UP) != 0) {
2473 			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0 &&
2474 			    ((ifp->if_flags ^ sc->sc_ifflags) &
2475 			    (IFF_ALLMULTI | IFF_PROMISC)) != 0)
2476 				cas_setladrf(sc);
2477 			else
2478 				cas_init_locked(sc);
2479 		} else if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
2480 			cas_stop(ifp);
2481 		sc->sc_ifflags = ifp->if_flags;
2482 		CAS_UNLOCK(sc);
2483 		break;
2484 	case SIOCSIFCAP:
2485 		CAS_LOCK(sc);
2486 		if ((sc->sc_flags & CAS_NO_CSUM) != 0) {
2487 			error = EINVAL;
2488 			CAS_UNLOCK(sc);
2489 			break;
2490 		}
2491 		ifp->if_capenable = ifr->ifr_reqcap;
2492 		if ((ifp->if_capenable & IFCAP_TXCSUM) != 0)
2493 			ifp->if_hwassist = CAS_CSUM_FEATURES;
2494 		else
2495 			ifp->if_hwassist = 0;
2496 		CAS_UNLOCK(sc);
2497 		break;
2498 	case SIOCADDMULTI:
2499 	case SIOCDELMULTI:
2500 		CAS_LOCK(sc);
2501 		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
2502 			cas_setladrf(sc);
2503 		CAS_UNLOCK(sc);
2504 		break;
2505 	case SIOCSIFMTU:
2506 		if ((ifr->ifr_mtu < ETHERMIN) ||
2507 		    (ifr->ifr_mtu > ETHERMTU_JUMBO))
2508 			error = EINVAL;
2509 		else
2510 			ifp->if_mtu = ifr->ifr_mtu;
2511 		break;
2512 	case SIOCGIFMEDIA:
2513 	case SIOCSIFMEDIA:
2514 		error = ifmedia_ioctl(ifp, ifr, &sc->sc_mii->mii_media, cmd);
2515 		break;
2516 	default:
2517 		error = ether_ioctl(ifp, cmd, data);
2518 		break;
2519 	}
2520 
2521 	return (error);
2522 }
2523 
2524 static void
2525 cas_setladrf(struct cas_softc *sc)
2526 {
2527 	struct ifnet *ifp = sc->sc_ifp;
2528 	struct ifmultiaddr *inm;
2529 	int i;
2530 	uint32_t hash[16];
2531 	uint32_t crc, v;
2532 
2533 	CAS_LOCK_ASSERT(sc, MA_OWNED);
2534 
2535 	/*
2536 	 * Turn off the RX MAC and the hash filter as required by the Sun
2537 	 * Cassini programming restrictions.
2538 	 */
2539 	v = sc->sc_mac_rxcfg & ~(CAS_MAC_RX_CONF_HFILTER |
2540 	    CAS_MAC_RX_CONF_EN);
2541 	CAS_WRITE_4(sc, CAS_MAC_RX_CONF, v);
2542 	CAS_BARRIER(sc, CAS_MAC_RX_CONF, 4,
2543 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2544 	if (!cas_bitwait(sc, CAS_MAC_RX_CONF, CAS_MAC_RX_CONF_HFILTER |
2545 	    CAS_MAC_RX_CONF_EN, 0))
2546 		device_printf(sc->sc_dev,
2547 		    "cannot disable RX MAC or hash filter\n");
2548 
2549 	v &= ~(CAS_MAC_RX_CONF_PROMISC | CAS_MAC_RX_CONF_PGRP);
2550 	if ((ifp->if_flags & IFF_PROMISC) != 0) {
2551 		v |= CAS_MAC_RX_CONF_PROMISC;
2552 		goto chipit;
2553 	}
2554 	if ((ifp->if_flags & IFF_ALLMULTI) != 0) {
2555 		v |= CAS_MAC_RX_CONF_PGRP;
2556 		goto chipit;
2557 	}
2558 
2559 	/*
2560 	 * Set up multicast address filter by passing all multicast
2561 	 * addresses through a crc generator, and then using the high
2562 	 * order 8 bits as an index into the 256 bit logical address
2563 	 * filter.  The high order 4 bits selects the word, while the
2564 	 * other 4 bits select the bit within the word (where bit 0
2565 	 * is the MSB).
2566 	 */
2567 
2568 	/* Clear the hash table. */
2569 	memset(hash, 0, sizeof(hash));
2570 
2571 	if_maddr_rlock(ifp);
2572 	TAILQ_FOREACH(inm, &ifp->if_multiaddrs, ifma_link) {
2573 		if (inm->ifma_addr->sa_family != AF_LINK)
2574 			continue;
2575 		crc = ether_crc32_le(LLADDR((struct sockaddr_dl *)
2576 		    inm->ifma_addr), ETHER_ADDR_LEN);
2577 
2578 		/* We just want the 8 most significant bits. */
2579 		crc >>= 24;
2580 
2581 		/* Set the corresponding bit in the filter. */
2582 		hash[crc >> 4] |= 1 << (15 - (crc & 15));
2583 	}
2584 	if_maddr_runlock(ifp);
2585 
2586 	v |= CAS_MAC_RX_CONF_HFILTER;
2587 
2588 	/* Now load the hash table into the chip (if we are using it). */
2589 	for (i = 0; i < 16; i++)
2590 		CAS_WRITE_4(sc,
2591 		    CAS_MAC_HASH0 + i * (CAS_MAC_HASH1 - CAS_MAC_HASH0),
2592 		    hash[i]);
2593 
2594  chipit:
2595 	sc->sc_mac_rxcfg = v;
2596 	CAS_WRITE_4(sc, CAS_MAC_RX_CONF, v | CAS_MAC_RX_CONF_EN);
2597 }
2598 
2599 static int	cas_pci_attach(device_t dev);
2600 static int	cas_pci_detach(device_t dev);
2601 static int	cas_pci_probe(device_t dev);
2602 static int	cas_pci_resume(device_t dev);
2603 static int	cas_pci_suspend(device_t dev);
2604 
2605 static device_method_t cas_pci_methods[] = {
2606 	/* Device interface */
2607 	DEVMETHOD(device_probe,		cas_pci_probe),
2608 	DEVMETHOD(device_attach,	cas_pci_attach),
2609 	DEVMETHOD(device_detach,	cas_pci_detach),
2610 	DEVMETHOD(device_suspend,	cas_pci_suspend),
2611 	DEVMETHOD(device_resume,	cas_pci_resume),
2612 	/* Use the suspend handler here, it is all that is required. */
2613 	DEVMETHOD(device_shutdown,	cas_pci_suspend),
2614 
2615 	/* MII interface */
2616 	DEVMETHOD(miibus_readreg,	cas_mii_readreg),
2617 	DEVMETHOD(miibus_writereg,	cas_mii_writereg),
2618 	DEVMETHOD(miibus_statchg,	cas_mii_statchg),
2619 
2620 	DEVMETHOD_END
2621 };
2622 
2623 static driver_t cas_pci_driver = {
2624 	"cas",
2625 	cas_pci_methods,
2626 	sizeof(struct cas_softc)
2627 };
2628 
2629 DRIVER_MODULE(cas, pci, cas_pci_driver, cas_devclass, 0, 0);
2630 DRIVER_MODULE(miibus, cas, miibus_driver, miibus_devclass, 0, 0);
2631 MODULE_DEPEND(cas, pci, 1, 1, 1);
2632 
2633 static const struct cas_pci_dev {
2634 	uint32_t	cpd_devid;
2635 	uint8_t		cpd_revid;
2636 	int		cpd_variant;
2637 	const char	*cpd_desc;
2638 } cas_pci_devlist[] = {
2639 	{ 0x0035100b, 0x0, CAS_SATURN, "NS DP83065 Saturn Gigabit Ethernet" },
2640 	{ 0xabba108e, 0x10, CAS_CASPLUS, "Sun Cassini+ Gigabit Ethernet" },
2641 	{ 0xabba108e, 0x0, CAS_CAS, "Sun Cassini Gigabit Ethernet" },
2642 	{ 0, 0, 0, NULL }
2643 };
2644 
2645 static int
2646 cas_pci_probe(device_t dev)
2647 {
2648 	int i;
2649 
2650 	for (i = 0; cas_pci_devlist[i].cpd_desc != NULL; i++) {
2651 		if (pci_get_devid(dev) == cas_pci_devlist[i].cpd_devid &&
2652 		    pci_get_revid(dev) >= cas_pci_devlist[i].cpd_revid) {
2653 			device_set_desc(dev, cas_pci_devlist[i].cpd_desc);
2654 			return (BUS_PROBE_DEFAULT);
2655 		}
2656 	}
2657 
2658 	return (ENXIO);
2659 }
2660 
2661 static struct resource_spec cas_pci_res_spec[] = {
2662 	{ SYS_RES_IRQ, 0, RF_SHAREABLE | RF_ACTIVE },	/* CAS_RES_INTR */
2663 	{ SYS_RES_MEMORY, PCIR_BAR(0), RF_ACTIVE },	/* CAS_RES_MEM */
2664 	{ -1, 0 }
2665 };
2666 
2667 #define	CAS_LOCAL_MAC_ADDRESS	"local-mac-address"
2668 #define	CAS_PHY_INTERFACE	"phy-interface"
2669 #define	CAS_PHY_TYPE		"phy-type"
2670 #define	CAS_PHY_TYPE_PCS	"pcs"
2671 
2672 static int
2673 cas_pci_attach(device_t dev)
2674 {
2675 	char buf[sizeof(CAS_LOCAL_MAC_ADDRESS)];
2676 	struct cas_softc *sc;
2677 	int i;
2678 #if !(defined(__powerpc__) || defined(__sparc64__))
2679 	u_char enaddr[4][ETHER_ADDR_LEN];
2680 	u_int j, k, lma, pcs[4], phy;
2681 #endif
2682 
2683 	sc = device_get_softc(dev);
2684 	sc->sc_variant = CAS_UNKNOWN;
2685 	for (i = 0; cas_pci_devlist[i].cpd_desc != NULL; i++) {
2686 		if (pci_get_devid(dev) == cas_pci_devlist[i].cpd_devid &&
2687 		    pci_get_revid(dev) >= cas_pci_devlist[i].cpd_revid) {
2688 			sc->sc_variant = cas_pci_devlist[i].cpd_variant;
2689 			break;
2690 		}
2691 	}
2692 	if (sc->sc_variant == CAS_UNKNOWN) {
2693 		device_printf(dev, "unknown adaptor\n");
2694 		return (ENXIO);
2695 	}
2696 
2697 	/* PCI configuration */
2698 	pci_write_config(dev, PCIR_COMMAND,
2699 	    pci_read_config(dev, PCIR_COMMAND, 2) | PCIM_CMD_BUSMASTEREN |
2700 	    PCIM_CMD_MWRICEN | PCIM_CMD_PERRESPEN | PCIM_CMD_SERRESPEN, 2);
2701 
2702 	sc->sc_dev = dev;
2703 	if (sc->sc_variant == CAS_CAS && pci_get_devid(dev) < 0x02)
2704 		/* Hardware checksumming may hang TX. */
2705 		sc->sc_flags |= CAS_NO_CSUM;
2706 	if (sc->sc_variant == CAS_CASPLUS || sc->sc_variant == CAS_SATURN)
2707 		sc->sc_flags |= CAS_REG_PLUS;
2708 	if (sc->sc_variant == CAS_CAS ||
2709 	    (sc->sc_variant == CAS_CASPLUS && pci_get_revid(dev) < 0x11))
2710 		sc->sc_flags |= CAS_TABORT;
2711 	if (bootverbose)
2712 		device_printf(dev, "flags=0x%x\n", sc->sc_flags);
2713 
2714 	if (bus_alloc_resources(dev, cas_pci_res_spec, sc->sc_res)) {
2715 		device_printf(dev, "failed to allocate resources\n");
2716 		bus_release_resources(dev, cas_pci_res_spec, sc->sc_res);
2717 		return (ENXIO);
2718 	}
2719 
2720 	CAS_LOCK_INIT(sc, device_get_nameunit(dev));
2721 
2722 #if defined(__powerpc__) || defined(__sparc64__)
2723 	OF_getetheraddr(dev, sc->sc_enaddr);
2724 	if (OF_getprop(ofw_bus_get_node(dev), CAS_PHY_INTERFACE, buf,
2725 	    sizeof(buf)) > 0 || OF_getprop(ofw_bus_get_node(dev),
2726 	    CAS_PHY_TYPE, buf, sizeof(buf)) > 0) {
2727 		buf[sizeof(buf) - 1] = '\0';
2728 		if (strcmp(buf, CAS_PHY_TYPE_PCS) == 0)
2729 			sc->sc_flags |= CAS_SERDES;
2730 	}
2731 #else
2732 	/*
2733 	 * Dig out VPD (vital product data) and read the MAC address as well
2734 	 * as the PHY type.  The VPD resides in the PCI Expansion ROM (PCI
2735 	 * FCode) and can't be accessed via the PCI capability pointer.
2736 	 * SUNW,pci-ce and SUNW,pci-qge use the Enhanced VPD format described
2737 	 * in the free US Patent 7149820.
2738 	 */
2739 
2740 #define	PCI_ROMHDR_SIZE			0x1c
2741 #define	PCI_ROMHDR_SIG			0x00
2742 #define	PCI_ROMHDR_SIG_MAGIC		0xaa55		/* little endian */
2743 #define	PCI_ROMHDR_PTR_DATA		0x18
2744 #define	PCI_ROM_SIZE			0x18
2745 #define	PCI_ROM_SIG			0x00
2746 #define	PCI_ROM_SIG_MAGIC		0x52494350	/* "PCIR", endian */
2747 							/* reversed */
2748 #define	PCI_ROM_VENDOR			0x04
2749 #define	PCI_ROM_DEVICE			0x06
2750 #define	PCI_ROM_PTR_VPD			0x08
2751 #define	PCI_VPDRES_BYTE0		0x00
2752 #define	PCI_VPDRES_ISLARGE(x)		((x) & 0x80)
2753 #define	PCI_VPDRES_LARGE_NAME(x)	((x) & 0x7f)
2754 #define	PCI_VPDRES_LARGE_LEN_LSB	0x01
2755 #define	PCI_VPDRES_LARGE_LEN_MSB	0x02
2756 #define	PCI_VPDRES_LARGE_SIZE		0x03
2757 #define	PCI_VPDRES_TYPE_ID_STRING	0x02		/* large */
2758 #define	PCI_VPDRES_TYPE_VPD		0x10		/* large */
2759 #define	PCI_VPD_KEY0			0x00
2760 #define	PCI_VPD_KEY1			0x01
2761 #define	PCI_VPD_LEN			0x02
2762 #define	PCI_VPD_SIZE			0x03
2763 
2764 #define	CAS_ROM_READ_1(sc, offs)					\
2765 	CAS_READ_1((sc), CAS_PCI_ROM_OFFSET + (offs))
2766 #define	CAS_ROM_READ_2(sc, offs)					\
2767 	CAS_READ_2((sc), CAS_PCI_ROM_OFFSET + (offs))
2768 #define	CAS_ROM_READ_4(sc, offs)					\
2769 	CAS_READ_4((sc), CAS_PCI_ROM_OFFSET + (offs))
2770 
2771 	lma = phy = 0;
2772 	memset(enaddr, 0, sizeof(enaddr));
2773 	memset(pcs, 0, sizeof(pcs));
2774 
2775 	/* Enable PCI Expansion ROM access. */
2776 	CAS_WRITE_4(sc, CAS_BIM_LDEV_OEN,
2777 	    CAS_BIM_LDEV_OEN_PAD | CAS_BIM_LDEV_OEN_PROM);
2778 
2779 	/* Read PCI Expansion ROM header. */
2780 	if (CAS_ROM_READ_2(sc, PCI_ROMHDR_SIG) != PCI_ROMHDR_SIG_MAGIC ||
2781 	    (i = CAS_ROM_READ_2(sc, PCI_ROMHDR_PTR_DATA)) <
2782 	    PCI_ROMHDR_SIZE) {
2783 		device_printf(dev, "unexpected PCI Expansion ROM header\n");
2784 		goto fail_prom;
2785 	}
2786 
2787 	/* Read PCI Expansion ROM data. */
2788 	if (CAS_ROM_READ_4(sc, i + PCI_ROM_SIG) != PCI_ROM_SIG_MAGIC ||
2789 	    CAS_ROM_READ_2(sc, i + PCI_ROM_VENDOR) != pci_get_vendor(dev) ||
2790 	    CAS_ROM_READ_2(sc, i + PCI_ROM_DEVICE) != pci_get_device(dev) ||
2791 	    (j = CAS_ROM_READ_2(sc, i + PCI_ROM_PTR_VPD)) <
2792 	    i + PCI_ROM_SIZE) {
2793 		device_printf(dev, "unexpected PCI Expansion ROM data\n");
2794 		goto fail_prom;
2795 	}
2796 
2797 	/* Read PCI VPD. */
2798  next:
2799 	if (PCI_VPDRES_ISLARGE(CAS_ROM_READ_1(sc,
2800 	    j + PCI_VPDRES_BYTE0)) == 0) {
2801 		device_printf(dev, "no large PCI VPD\n");
2802 		goto fail_prom;
2803 	}
2804 
2805 	i = (CAS_ROM_READ_1(sc, j + PCI_VPDRES_LARGE_LEN_MSB) << 8) |
2806 	    CAS_ROM_READ_1(sc, j + PCI_VPDRES_LARGE_LEN_LSB);
2807 	switch (PCI_VPDRES_LARGE_NAME(CAS_ROM_READ_1(sc,
2808 	    j + PCI_VPDRES_BYTE0))) {
2809 	case PCI_VPDRES_TYPE_ID_STRING:
2810 		/* Skip identifier string. */
2811 		j += PCI_VPDRES_LARGE_SIZE + i;
2812 		goto next;
2813 	case PCI_VPDRES_TYPE_VPD:
2814 		for (j += PCI_VPDRES_LARGE_SIZE; i > 0;
2815 		    i -= PCI_VPD_SIZE + CAS_ROM_READ_1(sc, j + PCI_VPD_LEN),
2816 		    j += PCI_VPD_SIZE + CAS_ROM_READ_1(sc, j + PCI_VPD_LEN)) {
2817 			if (CAS_ROM_READ_1(sc, j + PCI_VPD_KEY0) != 'Z')
2818 				/* no Enhanced VPD */
2819 				continue;
2820 			if (CAS_ROM_READ_1(sc, j + PCI_VPD_SIZE) != 'I')
2821 				/* no instance property */
2822 				continue;
2823 			if (CAS_ROM_READ_1(sc, j + PCI_VPD_SIZE + 3) == 'B') {
2824 				/* byte array */
2825 				if (CAS_ROM_READ_1(sc,
2826 				    j + PCI_VPD_SIZE + 4) != ETHER_ADDR_LEN)
2827 					continue;
2828 				bus_read_region_1(sc->sc_res[CAS_RES_MEM],
2829 				    CAS_PCI_ROM_OFFSET + j + PCI_VPD_SIZE + 5,
2830 				    buf, sizeof(buf));
2831 				buf[sizeof(buf) - 1] = '\0';
2832 				if (strcmp(buf, CAS_LOCAL_MAC_ADDRESS) != 0)
2833 					continue;
2834 				bus_read_region_1(sc->sc_res[CAS_RES_MEM],
2835 				    CAS_PCI_ROM_OFFSET + j + PCI_VPD_SIZE +
2836 				    5 + sizeof(CAS_LOCAL_MAC_ADDRESS),
2837 				    enaddr[lma], sizeof(enaddr[lma]));
2838 				lma++;
2839 				if (lma == 4 && phy == 4)
2840 					break;
2841 			} else if (CAS_ROM_READ_1(sc, j + PCI_VPD_SIZE + 3) ==
2842 			   'S') {
2843 				/* string */
2844 				if (CAS_ROM_READ_1(sc,
2845 				    j + PCI_VPD_SIZE + 4) !=
2846 				    sizeof(CAS_PHY_TYPE_PCS))
2847 					continue;
2848 				bus_read_region_1(sc->sc_res[CAS_RES_MEM],
2849 				    CAS_PCI_ROM_OFFSET + j + PCI_VPD_SIZE + 5,
2850 				    buf, sizeof(buf));
2851 				buf[sizeof(buf) - 1] = '\0';
2852 				if (strcmp(buf, CAS_PHY_INTERFACE) == 0)
2853 					k = sizeof(CAS_PHY_INTERFACE);
2854 				else if (strcmp(buf, CAS_PHY_TYPE) == 0)
2855 					k = sizeof(CAS_PHY_TYPE);
2856 				else
2857 					continue;
2858 				bus_read_region_1(sc->sc_res[CAS_RES_MEM],
2859 				    CAS_PCI_ROM_OFFSET + j + PCI_VPD_SIZE +
2860 				    5 + k, buf, sizeof(buf));
2861 				buf[sizeof(buf) - 1] = '\0';
2862 				if (strcmp(buf, CAS_PHY_TYPE_PCS) == 0)
2863 					pcs[phy] = 1;
2864 				phy++;
2865 				if (lma == 4 && phy == 4)
2866 					break;
2867 			}
2868 		}
2869 		break;
2870 	default:
2871 		device_printf(dev, "unexpected PCI VPD\n");
2872 		goto fail_prom;
2873 	}
2874 
2875  fail_prom:
2876 	CAS_WRITE_4(sc, CAS_BIM_LDEV_OEN, 0);
2877 
2878 	if (lma == 0) {
2879 		device_printf(dev, "could not determine Ethernet address\n");
2880 		goto fail;
2881 	}
2882 	i = 0;
2883 	if (lma > 1 && pci_get_slot(dev) < nitems(enaddr))
2884 		i = pci_get_slot(dev);
2885 	memcpy(sc->sc_enaddr, enaddr[i], ETHER_ADDR_LEN);
2886 
2887 	if (phy == 0) {
2888 		device_printf(dev, "could not determine PHY type\n");
2889 		goto fail;
2890 	}
2891 	i = 0;
2892 	if (phy > 1 && pci_get_slot(dev) < nitems(pcs))
2893 		i = pci_get_slot(dev);
2894 	if (pcs[i] != 0)
2895 		sc->sc_flags |= CAS_SERDES;
2896 #endif
2897 
2898 	if (cas_attach(sc) != 0) {
2899 		device_printf(dev, "could not be attached\n");
2900 		goto fail;
2901 	}
2902 
2903 	if (bus_setup_intr(dev, sc->sc_res[CAS_RES_INTR], INTR_TYPE_NET |
2904 	    INTR_MPSAFE, cas_intr, NULL, sc, &sc->sc_ih) != 0) {
2905 		device_printf(dev, "failed to set up interrupt\n");
2906 		cas_detach(sc);
2907 		goto fail;
2908 	}
2909 	return (0);
2910 
2911  fail:
2912 	CAS_LOCK_DESTROY(sc);
2913 	bus_release_resources(dev, cas_pci_res_spec, sc->sc_res);
2914 	return (ENXIO);
2915 }
2916 
2917 static int
2918 cas_pci_detach(device_t dev)
2919 {
2920 	struct cas_softc *sc;
2921 
2922 	sc = device_get_softc(dev);
2923 	bus_teardown_intr(dev, sc->sc_res[CAS_RES_INTR], sc->sc_ih);
2924 	cas_detach(sc);
2925 	CAS_LOCK_DESTROY(sc);
2926 	bus_release_resources(dev, cas_pci_res_spec, sc->sc_res);
2927 	return (0);
2928 }
2929 
2930 static int
2931 cas_pci_suspend(device_t dev)
2932 {
2933 
2934 	cas_suspend(device_get_softc(dev));
2935 	return (0);
2936 }
2937 
2938 static int
2939 cas_pci_resume(device_t dev)
2940 {
2941 
2942 	cas_resume(device_get_softc(dev));
2943 	return (0);
2944 }
2945