xref: /freebsd/sys/dev/cas/if_cas.c (revision 96190b4fef3b4a0cc3ca0606b0c4e3e69a5e6717)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (C) 2001 Eduardo Horvath.
5  * Copyright (c) 2001-2003 Thomas Moestl
6  * Copyright (c) 2007-2009 Marius Strobl <marius@FreeBSD.org>
7  * All rights reserved.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR  ``AS IS'' AND
19  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR  BE LIABLE
22  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28  * SUCH DAMAGE.
29  *
30  *	from: NetBSD: gem.c,v 1.21 2002/06/01 23:50:58 lukem Exp
31  *	from: FreeBSD: if_gem.c 182060 2008-08-23 15:03:26Z marius
32  */
33 
34 #include <sys/cdefs.h>
35 /*
36  * driver for Sun Cassini/Cassini+ and National Semiconductor DP83065
37  * Saturn Gigabit Ethernet controllers
38  */
39 
40 #if 0
41 #define	CAS_DEBUG
42 #endif
43 
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/bus.h>
47 #include <sys/callout.h>
48 #include <sys/endian.h>
49 #include <sys/mbuf.h>
50 #include <sys/malloc.h>
51 #include <sys/kernel.h>
52 #include <sys/lock.h>
53 #include <sys/module.h>
54 #include <sys/mutex.h>
55 #include <sys/refcount.h>
56 #include <sys/resource.h>
57 #include <sys/rman.h>
58 #include <sys/socket.h>
59 #include <sys/sockio.h>
60 #include <sys/taskqueue.h>
61 
62 #include <net/bpf.h>
63 #include <net/ethernet.h>
64 #include <net/if.h>
65 #include <net/if_var.h>
66 #include <net/if_arp.h>
67 #include <net/if_dl.h>
68 #include <net/if_media.h>
69 #include <net/if_types.h>
70 #include <net/if_vlan_var.h>
71 
72 #include <netinet/in.h>
73 #include <netinet/in_systm.h>
74 #include <netinet/ip.h>
75 #include <netinet/tcp.h>
76 #include <netinet/udp.h>
77 
78 #include <machine/bus.h>
79 #if defined(__powerpc__)
80 #include <dev/ofw/ofw_bus.h>
81 #include <dev/ofw/openfirm.h>
82 #include <machine/ofw_machdep.h>
83 #endif
84 #include <machine/resource.h>
85 
86 #include <dev/mii/mii.h>
87 #include <dev/mii/miivar.h>
88 
89 #include <dev/cas/if_casreg.h>
90 #include <dev/cas/if_casvar.h>
91 
92 #include <dev/pci/pcireg.h>
93 #include <dev/pci/pcivar.h>
94 
95 #include "miibus_if.h"
96 
97 #define RINGASSERT(n , min, max)					\
98 	CTASSERT(powerof2(n) && (n) >= (min) && (n) <= (max))
99 
100 RINGASSERT(CAS_NRXCOMP, 128, 32768);
101 RINGASSERT(CAS_NRXDESC, 32, 8192);
102 RINGASSERT(CAS_NRXDESC2, 32, 8192);
103 RINGASSERT(CAS_NTXDESC, 32, 8192);
104 
105 #undef RINGASSERT
106 
107 #define	CCDASSERT(m, a)							\
108 	CTASSERT((offsetof(struct cas_control_data, m) & ((a) - 1)) == 0)
109 
110 CCDASSERT(ccd_rxcomps, CAS_RX_COMP_ALIGN);
111 CCDASSERT(ccd_rxdescs, CAS_RX_DESC_ALIGN);
112 CCDASSERT(ccd_rxdescs2, CAS_RX_DESC_ALIGN);
113 
114 #undef CCDASSERT
115 
116 #define	CAS_TRIES	10000
117 
118 /*
119  * According to documentation, the hardware has support for basic TCP
120  * checksum offloading only, in practice this can be also used for UDP
121  * however (i.e. the problem of previous Sun NICs that a checksum of 0x0
122  * is not converted to 0xffff no longer exists).
123  */
124 #define	CAS_CSUM_FEATURES	(CSUM_TCP | CSUM_UDP)
125 
126 static inline void cas_add_rxdesc(struct cas_softc *sc, u_int idx);
127 static int	cas_attach(struct cas_softc *sc);
128 static int	cas_bitwait(struct cas_softc *sc, bus_addr_t r, uint32_t clr,
129 		    uint32_t set);
130 static void	cas_cddma_callback(void *xsc, bus_dma_segment_t *segs,
131 		    int nsegs, int error);
132 static void	cas_detach(struct cas_softc *sc);
133 static int	cas_disable_rx(struct cas_softc *sc);
134 static int	cas_disable_tx(struct cas_softc *sc);
135 static void	cas_eint(struct cas_softc *sc, u_int status);
136 static void	cas_free(struct mbuf *m);
137 static void	cas_init(void *xsc);
138 static void	cas_init_locked(struct cas_softc *sc);
139 static void	cas_init_regs(struct cas_softc *sc);
140 static int	cas_intr(void *v);
141 static void	cas_intr_task(void *arg, int pending __unused);
142 static int	cas_ioctl(if_t ifp, u_long cmd, caddr_t data);
143 static int	cas_load_txmbuf(struct cas_softc *sc, struct mbuf **m_head);
144 static int	cas_mediachange(if_t ifp);
145 static void	cas_mediastatus(if_t ifp, struct ifmediareq *ifmr);
146 static void	cas_meminit(struct cas_softc *sc);
147 static void	cas_mifinit(struct cas_softc *sc);
148 static int	cas_mii_readreg(device_t dev, int phy, int reg);
149 static void	cas_mii_statchg(device_t dev);
150 static int	cas_mii_writereg(device_t dev, int phy, int reg, int val);
151 static void	cas_reset(struct cas_softc *sc);
152 static int	cas_reset_rx(struct cas_softc *sc);
153 static int	cas_reset_tx(struct cas_softc *sc);
154 static void	cas_resume(struct cas_softc *sc);
155 static u_int	cas_descsize(u_int sz);
156 static void	cas_rint(struct cas_softc *sc);
157 static void	cas_rint_timeout(void *arg);
158 static inline void cas_rxcksum(struct mbuf *m, uint16_t cksum);
159 static inline void cas_rxcompinit(struct cas_rx_comp *rxcomp);
160 static u_int	cas_rxcompsize(u_int sz);
161 static void	cas_rxdma_callback(void *xsc, bus_dma_segment_t *segs,
162 		    int nsegs, int error);
163 static void	cas_setladrf(struct cas_softc *sc);
164 static void	cas_start(if_t ifp);
165 static void	cas_stop(if_t ifp);
166 static void	cas_suspend(struct cas_softc *sc);
167 static void	cas_tick(void *arg);
168 static void	cas_tint(struct cas_softc *sc);
169 static void	cas_tx_task(void *arg, int pending __unused);
170 static inline void cas_txkick(struct cas_softc *sc);
171 static void	cas_watchdog(struct cas_softc *sc);
172 
173 MODULE_DEPEND(cas, ether, 1, 1, 1);
174 MODULE_DEPEND(cas, miibus, 1, 1, 1);
175 
176 #ifdef CAS_DEBUG
177 #include <sys/ktr.h>
178 #define	KTR_CAS		KTR_SPARE2
179 #endif
180 
181 static int
182 cas_attach(struct cas_softc *sc)
183 {
184 	struct cas_txsoft *txs;
185 	if_t ifp;
186 	int error, i;
187 	uint32_t v;
188 
189 	/* Set up ifnet structure. */
190 	ifp = sc->sc_ifp = if_alloc(IFT_ETHER);
191 	if_setsoftc(ifp, sc);
192 	if_initname(ifp, device_get_name(sc->sc_dev),
193 	    device_get_unit(sc->sc_dev));
194 	if_setflags(ifp, IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST);
195 	if_setstartfn(ifp, cas_start);
196 	if_setioctlfn(ifp, cas_ioctl);
197 	if_setinitfn(ifp, cas_init);
198 	if_setsendqlen(ifp, CAS_TXQUEUELEN);
199 	if_setsendqready(ifp);
200 
201 	callout_init_mtx(&sc->sc_tick_ch, &sc->sc_mtx, 0);
202 	callout_init_mtx(&sc->sc_rx_ch, &sc->sc_mtx, 0);
203 	/* Create local taskq. */
204 	NET_TASK_INIT(&sc->sc_intr_task, 0, cas_intr_task, sc);
205 	TASK_INIT(&sc->sc_tx_task, 1, cas_tx_task, ifp);
206 	sc->sc_tq = taskqueue_create_fast("cas_taskq", M_WAITOK,
207 	    taskqueue_thread_enqueue, &sc->sc_tq);
208 	error = taskqueue_start_threads(&sc->sc_tq, 1, PI_NET, "%s taskq",
209 	    device_get_nameunit(sc->sc_dev));
210 	if (error != 0) {
211 		device_printf(sc->sc_dev, "could not start threads\n");
212 		goto fail_taskq;
213 	}
214 
215 	/* Make sure the chip is stopped. */
216 	cas_reset(sc);
217 
218 	error = bus_dma_tag_create(bus_get_dma_tag(sc->sc_dev), 1, 0,
219 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
220 	    BUS_SPACE_MAXSIZE, 0, BUS_SPACE_MAXSIZE, 0, NULL, NULL,
221 	    &sc->sc_pdmatag);
222 	if (error != 0)
223 		goto fail_taskq;
224 
225 	error = bus_dma_tag_create(sc->sc_pdmatag, 1, 0,
226 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
227 	    CAS_PAGE_SIZE, 1, CAS_PAGE_SIZE, 0, NULL, NULL, &sc->sc_rdmatag);
228 	if (error != 0)
229 		goto fail_ptag;
230 
231 	error = bus_dma_tag_create(sc->sc_pdmatag, 1, 0,
232 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
233 	    MCLBYTES * CAS_NTXSEGS, CAS_NTXSEGS, MCLBYTES,
234 	    BUS_DMA_ALLOCNOW, NULL, NULL, &sc->sc_tdmatag);
235 	if (error != 0)
236 		goto fail_rtag;
237 
238 	error = bus_dma_tag_create(sc->sc_pdmatag, CAS_TX_DESC_ALIGN, 0,
239 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
240 	    sizeof(struct cas_control_data), 1,
241 	    sizeof(struct cas_control_data), 0,
242 	    NULL, NULL, &sc->sc_cdmatag);
243 	if (error != 0)
244 		goto fail_ttag;
245 
246 	/*
247 	 * Allocate the control data structures, create and load the
248 	 * DMA map for it.
249 	 */
250 	if ((error = bus_dmamem_alloc(sc->sc_cdmatag,
251 	    (void **)&sc->sc_control_data,
252 	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
253 	    &sc->sc_cddmamap)) != 0) {
254 		device_printf(sc->sc_dev,
255 		    "unable to allocate control data, error = %d\n", error);
256 		goto fail_ctag;
257 	}
258 
259 	sc->sc_cddma = 0;
260 	if ((error = bus_dmamap_load(sc->sc_cdmatag, sc->sc_cddmamap,
261 	    sc->sc_control_data, sizeof(struct cas_control_data),
262 	    cas_cddma_callback, sc, 0)) != 0 || sc->sc_cddma == 0) {
263 		device_printf(sc->sc_dev,
264 		    "unable to load control data DMA map, error = %d\n",
265 		    error);
266 		goto fail_cmem;
267 	}
268 
269 	/*
270 	 * Initialize the transmit job descriptors.
271 	 */
272 	STAILQ_INIT(&sc->sc_txfreeq);
273 	STAILQ_INIT(&sc->sc_txdirtyq);
274 
275 	/*
276 	 * Create the transmit buffer DMA maps.
277 	 */
278 	error = ENOMEM;
279 	for (i = 0; i < CAS_TXQUEUELEN; i++) {
280 		txs = &sc->sc_txsoft[i];
281 		txs->txs_mbuf = NULL;
282 		txs->txs_ndescs = 0;
283 		if ((error = bus_dmamap_create(sc->sc_tdmatag, 0,
284 		    &txs->txs_dmamap)) != 0) {
285 			device_printf(sc->sc_dev,
286 			    "unable to create TX DMA map %d, error = %d\n",
287 			    i, error);
288 			goto fail_txd;
289 		}
290 		STAILQ_INSERT_TAIL(&sc->sc_txfreeq, txs, txs_q);
291 	}
292 
293 	/*
294 	 * Allocate the receive buffers, create and load the DMA maps
295 	 * for them.
296 	 */
297 	for (i = 0; i < CAS_NRXDESC; i++) {
298 		if ((error = bus_dmamem_alloc(sc->sc_rdmatag,
299 		    &sc->sc_rxdsoft[i].rxds_buf, BUS_DMA_WAITOK,
300 		    &sc->sc_rxdsoft[i].rxds_dmamap)) != 0) {
301 			device_printf(sc->sc_dev,
302 			    "unable to allocate RX buffer %d, error = %d\n",
303 			    i, error);
304 			goto fail_rxmem;
305 		}
306 
307 		sc->sc_rxdptr = i;
308 		sc->sc_rxdsoft[i].rxds_paddr = 0;
309 		if ((error = bus_dmamap_load(sc->sc_rdmatag,
310 		    sc->sc_rxdsoft[i].rxds_dmamap, sc->sc_rxdsoft[i].rxds_buf,
311 		    CAS_PAGE_SIZE, cas_rxdma_callback, sc, 0)) != 0 ||
312 		    sc->sc_rxdsoft[i].rxds_paddr == 0) {
313 			device_printf(sc->sc_dev,
314 			    "unable to load RX DMA map %d, error = %d\n",
315 			    i, error);
316 			goto fail_rxmap;
317 		}
318 	}
319 
320 	if ((sc->sc_flags & CAS_SERDES) == 0) {
321 		CAS_WRITE_4(sc, CAS_PCS_DATAPATH, CAS_PCS_DATAPATH_MII);
322 		CAS_BARRIER(sc, CAS_PCS_DATAPATH, 4,
323 		    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
324 		cas_mifinit(sc);
325 		/*
326 		 * Look for an external PHY.
327 		 */
328 		error = ENXIO;
329 		v = CAS_READ_4(sc, CAS_MIF_CONF);
330 		if ((v & CAS_MIF_CONF_MDI1) != 0) {
331 			v |= CAS_MIF_CONF_PHY_SELECT;
332 			CAS_WRITE_4(sc, CAS_MIF_CONF, v);
333 			CAS_BARRIER(sc, CAS_MIF_CONF, 4,
334 			    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
335 			/* Enable/unfreeze the GMII pins of Saturn. */
336 			if (sc->sc_variant == CAS_SATURN) {
337 				CAS_WRITE_4(sc, CAS_SATURN_PCFG,
338 				    CAS_READ_4(sc, CAS_SATURN_PCFG) &
339 				    ~CAS_SATURN_PCFG_FSI);
340 				CAS_BARRIER(sc, CAS_SATURN_PCFG, 4,
341 				    BUS_SPACE_BARRIER_READ |
342 				    BUS_SPACE_BARRIER_WRITE);
343 				DELAY(10000);
344 			}
345 			error = mii_attach(sc->sc_dev, &sc->sc_miibus, ifp,
346 			    cas_mediachange, cas_mediastatus, BMSR_DEFCAPMASK,
347 			    MII_PHY_ANY, MII_OFFSET_ANY, MIIF_DOPAUSE);
348 		}
349 		/*
350 		 * Fall back on an internal PHY if no external PHY was found.
351 		 */
352 		if (error != 0 && (v & CAS_MIF_CONF_MDI0) != 0) {
353 			v &= ~CAS_MIF_CONF_PHY_SELECT;
354 			CAS_WRITE_4(sc, CAS_MIF_CONF, v);
355 			CAS_BARRIER(sc, CAS_MIF_CONF, 4,
356 			    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
357 			/* Freeze the GMII pins of Saturn for saving power. */
358 			if (sc->sc_variant == CAS_SATURN) {
359 				CAS_WRITE_4(sc, CAS_SATURN_PCFG,
360 				    CAS_READ_4(sc, CAS_SATURN_PCFG) |
361 				    CAS_SATURN_PCFG_FSI);
362 				CAS_BARRIER(sc, CAS_SATURN_PCFG, 4,
363 				    BUS_SPACE_BARRIER_READ |
364 				    BUS_SPACE_BARRIER_WRITE);
365 				DELAY(10000);
366 			}
367 			error = mii_attach(sc->sc_dev, &sc->sc_miibus, ifp,
368 			    cas_mediachange, cas_mediastatus, BMSR_DEFCAPMASK,
369 			    MII_PHY_ANY, MII_OFFSET_ANY, MIIF_DOPAUSE);
370 		}
371 	} else {
372 		/*
373 		 * Use the external PCS SERDES.
374 		 */
375 		CAS_WRITE_4(sc, CAS_PCS_DATAPATH, CAS_PCS_DATAPATH_SERDES);
376 		CAS_BARRIER(sc, CAS_PCS_DATAPATH, 4, BUS_SPACE_BARRIER_WRITE);
377 		/* Enable/unfreeze the SERDES pins of Saturn. */
378 		if (sc->sc_variant == CAS_SATURN) {
379 			CAS_WRITE_4(sc, CAS_SATURN_PCFG, 0);
380 			CAS_BARRIER(sc, CAS_SATURN_PCFG, 4,
381 			    BUS_SPACE_BARRIER_WRITE);
382 		}
383 		CAS_WRITE_4(sc, CAS_PCS_SERDES_CTRL, CAS_PCS_SERDES_CTRL_ESD);
384 		CAS_BARRIER(sc, CAS_PCS_SERDES_CTRL, 4,
385 		    BUS_SPACE_BARRIER_WRITE);
386 		CAS_WRITE_4(sc, CAS_PCS_CONF, CAS_PCS_CONF_EN);
387 		CAS_BARRIER(sc, CAS_PCS_CONF, 4,
388 		    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
389 		error = mii_attach(sc->sc_dev, &sc->sc_miibus, ifp,
390 		    cas_mediachange, cas_mediastatus, BMSR_DEFCAPMASK,
391 		    CAS_PHYAD_EXTERNAL, MII_OFFSET_ANY, MIIF_DOPAUSE);
392 	}
393 	if (error != 0) {
394 		device_printf(sc->sc_dev, "attaching PHYs failed\n");
395 		goto fail_rxmap;
396 	}
397 	sc->sc_mii = device_get_softc(sc->sc_miibus);
398 
399 	/*
400 	 * From this point forward, the attachment cannot fail.  A failure
401 	 * before this point releases all resources that may have been
402 	 * allocated.
403 	 */
404 
405 	/* Announce FIFO sizes. */
406 	v = CAS_READ_4(sc, CAS_TX_FIFO_SIZE);
407 	device_printf(sc->sc_dev, "%ukB RX FIFO, %ukB TX FIFO\n",
408 	    CAS_RX_FIFO_SIZE / 1024, v / 16);
409 
410 	/* Attach the interface. */
411 	ether_ifattach(ifp, sc->sc_enaddr);
412 
413 	/*
414 	 * Tell the upper layer(s) we support long frames/checksum offloads.
415 	 */
416 	if_setifheaderlen(ifp, sizeof(struct ether_vlan_header));
417 	if_setcapabilities(ifp, IFCAP_VLAN_MTU);
418 	if ((sc->sc_flags & CAS_NO_CSUM) == 0) {
419 		if_setcapabilitiesbit(ifp, IFCAP_HWCSUM, 0);
420 		if_sethwassist(ifp, CAS_CSUM_FEATURES);
421 	}
422 	if_setcapenable(ifp, if_getcapabilities(ifp));
423 
424 	return (0);
425 
426 	/*
427 	 * Free any resources we've allocated during the failed attach
428 	 * attempt.  Do this in reverse order and fall through.
429 	 */
430  fail_rxmap:
431 	for (i = 0; i < CAS_NRXDESC; i++)
432 		if (sc->sc_rxdsoft[i].rxds_paddr != 0)
433 			bus_dmamap_unload(sc->sc_rdmatag,
434 			    sc->sc_rxdsoft[i].rxds_dmamap);
435  fail_rxmem:
436 	for (i = 0; i < CAS_NRXDESC; i++)
437 		if (sc->sc_rxdsoft[i].rxds_buf != NULL)
438 			bus_dmamem_free(sc->sc_rdmatag,
439 			    sc->sc_rxdsoft[i].rxds_buf,
440 			    sc->sc_rxdsoft[i].rxds_dmamap);
441  fail_txd:
442 	for (i = 0; i < CAS_TXQUEUELEN; i++)
443 		if (sc->sc_txsoft[i].txs_dmamap != NULL)
444 			bus_dmamap_destroy(sc->sc_tdmatag,
445 			    sc->sc_txsoft[i].txs_dmamap);
446 	bus_dmamap_unload(sc->sc_cdmatag, sc->sc_cddmamap);
447  fail_cmem:
448 	bus_dmamem_free(sc->sc_cdmatag, sc->sc_control_data,
449 	    sc->sc_cddmamap);
450  fail_ctag:
451 	bus_dma_tag_destroy(sc->sc_cdmatag);
452  fail_ttag:
453 	bus_dma_tag_destroy(sc->sc_tdmatag);
454  fail_rtag:
455 	bus_dma_tag_destroy(sc->sc_rdmatag);
456  fail_ptag:
457 	bus_dma_tag_destroy(sc->sc_pdmatag);
458  fail_taskq:
459 	taskqueue_free(sc->sc_tq);
460 	if_free(ifp);
461 	return (error);
462 }
463 
464 static void
465 cas_detach(struct cas_softc *sc)
466 {
467 	if_t ifp = sc->sc_ifp;
468 	int i;
469 
470 	ether_ifdetach(ifp);
471 	CAS_LOCK(sc);
472 	cas_stop(ifp);
473 	CAS_UNLOCK(sc);
474 	callout_drain(&sc->sc_tick_ch);
475 	callout_drain(&sc->sc_rx_ch);
476 	taskqueue_drain(sc->sc_tq, &sc->sc_intr_task);
477 	taskqueue_drain(sc->sc_tq, &sc->sc_tx_task);
478 	if_free(ifp);
479 	taskqueue_free(sc->sc_tq);
480 	device_delete_child(sc->sc_dev, sc->sc_miibus);
481 
482 	for (i = 0; i < CAS_NRXDESC; i++)
483 		if (sc->sc_rxdsoft[i].rxds_dmamap != NULL)
484 			bus_dmamap_sync(sc->sc_rdmatag,
485 			    sc->sc_rxdsoft[i].rxds_dmamap,
486 			    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
487 	for (i = 0; i < CAS_NRXDESC; i++)
488 		if (sc->sc_rxdsoft[i].rxds_paddr != 0)
489 			bus_dmamap_unload(sc->sc_rdmatag,
490 			    sc->sc_rxdsoft[i].rxds_dmamap);
491 	for (i = 0; i < CAS_NRXDESC; i++)
492 		if (sc->sc_rxdsoft[i].rxds_buf != NULL)
493 			bus_dmamem_free(sc->sc_rdmatag,
494 			    sc->sc_rxdsoft[i].rxds_buf,
495 			    sc->sc_rxdsoft[i].rxds_dmamap);
496 	for (i = 0; i < CAS_TXQUEUELEN; i++)
497 		if (sc->sc_txsoft[i].txs_dmamap != NULL)
498 			bus_dmamap_destroy(sc->sc_tdmatag,
499 			    sc->sc_txsoft[i].txs_dmamap);
500 	CAS_CDSYNC(sc, BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
501 	bus_dmamap_unload(sc->sc_cdmatag, sc->sc_cddmamap);
502 	bus_dmamem_free(sc->sc_cdmatag, sc->sc_control_data,
503 	    sc->sc_cddmamap);
504 	bus_dma_tag_destroy(sc->sc_cdmatag);
505 	bus_dma_tag_destroy(sc->sc_tdmatag);
506 	bus_dma_tag_destroy(sc->sc_rdmatag);
507 	bus_dma_tag_destroy(sc->sc_pdmatag);
508 }
509 
510 static void
511 cas_suspend(struct cas_softc *sc)
512 {
513 	if_t ifp = sc->sc_ifp;
514 
515 	CAS_LOCK(sc);
516 	cas_stop(ifp);
517 	CAS_UNLOCK(sc);
518 }
519 
520 static void
521 cas_resume(struct cas_softc *sc)
522 {
523 	if_t ifp = sc->sc_ifp;
524 
525 	CAS_LOCK(sc);
526 	/*
527 	 * On resume all registers have to be initialized again like
528 	 * after power-on.
529 	 */
530 	sc->sc_flags &= ~CAS_INITED;
531 	if (if_getflags(ifp) & IFF_UP)
532 		cas_init_locked(sc);
533 	CAS_UNLOCK(sc);
534 }
535 
536 static inline void
537 cas_rxcksum(struct mbuf *m, uint16_t cksum)
538 {
539 	struct ether_header *eh;
540 	struct ip *ip;
541 	struct udphdr *uh;
542 	uint16_t *opts;
543 	int32_t hlen, len, pktlen;
544 	uint32_t temp32;
545 
546 	pktlen = m->m_pkthdr.len;
547 	if (pktlen < sizeof(struct ether_header) + sizeof(struct ip))
548 		return;
549 	eh = mtod(m, struct ether_header *);
550 	if (eh->ether_type != htons(ETHERTYPE_IP))
551 		return;
552 	ip = (struct ip *)(eh + 1);
553 	if (ip->ip_v != IPVERSION)
554 		return;
555 
556 	hlen = ip->ip_hl << 2;
557 	pktlen -= sizeof(struct ether_header);
558 	if (hlen < sizeof(struct ip))
559 		return;
560 	if (ntohs(ip->ip_len) < hlen)
561 		return;
562 	if (ntohs(ip->ip_len) != pktlen)
563 		return;
564 	if (ip->ip_off & htons(IP_MF | IP_OFFMASK))
565 		return;	/* Cannot handle fragmented packet. */
566 
567 	switch (ip->ip_p) {
568 	case IPPROTO_TCP:
569 		if (pktlen < (hlen + sizeof(struct tcphdr)))
570 			return;
571 		break;
572 	case IPPROTO_UDP:
573 		if (pktlen < (hlen + sizeof(struct udphdr)))
574 			return;
575 		uh = (struct udphdr *)((uint8_t *)ip + hlen);
576 		if (uh->uh_sum == 0)
577 			return; /* no checksum */
578 		break;
579 	default:
580 		return;
581 	}
582 
583 	cksum = ~cksum;
584 	/* checksum fixup for IP options */
585 	len = hlen - sizeof(struct ip);
586 	if (len > 0) {
587 		opts = (uint16_t *)(ip + 1);
588 		for (; len > 0; len -= sizeof(uint16_t), opts++) {
589 			temp32 = cksum - *opts;
590 			temp32 = (temp32 >> 16) + (temp32 & 65535);
591 			cksum = temp32 & 65535;
592 		}
593 	}
594 	m->m_pkthdr.csum_flags |= CSUM_DATA_VALID;
595 	m->m_pkthdr.csum_data = cksum;
596 }
597 
598 static void
599 cas_cddma_callback(void *xsc, bus_dma_segment_t *segs, int nsegs, int error)
600 {
601 	struct cas_softc *sc = xsc;
602 
603 	if (error != 0)
604 		return;
605 	if (nsegs != 1)
606 		panic("%s: bad control buffer segment count", __func__);
607 	sc->sc_cddma = segs[0].ds_addr;
608 }
609 
610 static void
611 cas_rxdma_callback(void *xsc, bus_dma_segment_t *segs, int nsegs, int error)
612 {
613 	struct cas_softc *sc = xsc;
614 
615 	if (error != 0)
616 		return;
617 	if (nsegs != 1)
618 		panic("%s: bad RX buffer segment count", __func__);
619 	sc->sc_rxdsoft[sc->sc_rxdptr].rxds_paddr = segs[0].ds_addr;
620 }
621 
622 static void
623 cas_tick(void *arg)
624 {
625 	struct cas_softc *sc = arg;
626 	if_t ifp = sc->sc_ifp;
627 	uint32_t v;
628 
629 	CAS_LOCK_ASSERT(sc, MA_OWNED);
630 
631 	/*
632 	 * Unload collision and error counters.
633 	 */
634 	if_inc_counter(ifp, IFCOUNTER_COLLISIONS,
635 	    CAS_READ_4(sc, CAS_MAC_NORM_COLL_CNT) +
636 	    CAS_READ_4(sc, CAS_MAC_FIRST_COLL_CNT));
637 	v = CAS_READ_4(sc, CAS_MAC_EXCESS_COLL_CNT) +
638 	    CAS_READ_4(sc, CAS_MAC_LATE_COLL_CNT);
639 	if_inc_counter(ifp, IFCOUNTER_COLLISIONS, v);
640 	if_inc_counter(ifp, IFCOUNTER_OERRORS, v);
641 	if_inc_counter(ifp, IFCOUNTER_IERRORS,
642 	    CAS_READ_4(sc, CAS_MAC_RX_LEN_ERR_CNT) +
643 	    CAS_READ_4(sc, CAS_MAC_RX_ALIGN_ERR) +
644 	    CAS_READ_4(sc, CAS_MAC_RX_CRC_ERR_CNT) +
645 	    CAS_READ_4(sc, CAS_MAC_RX_CODE_VIOL));
646 
647 	/*
648 	 * Then clear the hardware counters.
649 	 */
650 	CAS_WRITE_4(sc, CAS_MAC_NORM_COLL_CNT, 0);
651 	CAS_WRITE_4(sc, CAS_MAC_FIRST_COLL_CNT, 0);
652 	CAS_WRITE_4(sc, CAS_MAC_EXCESS_COLL_CNT, 0);
653 	CAS_WRITE_4(sc, CAS_MAC_LATE_COLL_CNT, 0);
654 	CAS_WRITE_4(sc, CAS_MAC_RX_LEN_ERR_CNT, 0);
655 	CAS_WRITE_4(sc, CAS_MAC_RX_ALIGN_ERR, 0);
656 	CAS_WRITE_4(sc, CAS_MAC_RX_CRC_ERR_CNT, 0);
657 	CAS_WRITE_4(sc, CAS_MAC_RX_CODE_VIOL, 0);
658 
659 	mii_tick(sc->sc_mii);
660 
661 	if (sc->sc_txfree != CAS_MAXTXFREE)
662 		cas_tint(sc);
663 
664 	cas_watchdog(sc);
665 
666 	callout_reset(&sc->sc_tick_ch, hz, cas_tick, sc);
667 }
668 
669 static int
670 cas_bitwait(struct cas_softc *sc, bus_addr_t r, uint32_t clr, uint32_t set)
671 {
672 	int i;
673 	uint32_t reg;
674 
675 	for (i = CAS_TRIES; i--; DELAY(100)) {
676 		reg = CAS_READ_4(sc, r);
677 		if ((reg & clr) == 0 && (reg & set) == set)
678 			return (1);
679 	}
680 	return (0);
681 }
682 
683 static void
684 cas_reset(struct cas_softc *sc)
685 {
686 
687 #ifdef CAS_DEBUG
688 	CTR2(KTR_CAS, "%s: %s", device_get_name(sc->sc_dev), __func__);
689 #endif
690 	/* Disable all interrupts in order to avoid spurious ones. */
691 	CAS_WRITE_4(sc, CAS_INTMASK, 0xffffffff);
692 
693 	cas_reset_rx(sc);
694 	cas_reset_tx(sc);
695 
696 	/*
697 	 * Do a full reset modulo the result of the last auto-negotiation
698 	 * when using the SERDES.
699 	 */
700 	CAS_WRITE_4(sc, CAS_RESET, CAS_RESET_RX | CAS_RESET_TX |
701 	    ((sc->sc_flags & CAS_SERDES) != 0 ? CAS_RESET_PCS_DIS : 0));
702 	CAS_BARRIER(sc, CAS_RESET, 4,
703 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
704 	DELAY(3000);
705 	if (!cas_bitwait(sc, CAS_RESET, CAS_RESET_RX | CAS_RESET_TX, 0))
706 		device_printf(sc->sc_dev, "cannot reset device\n");
707 }
708 
709 static void
710 cas_stop(if_t ifp)
711 {
712 	struct cas_softc *sc = if_getsoftc(ifp);
713 	struct cas_txsoft *txs;
714 
715 #ifdef CAS_DEBUG
716 	CTR2(KTR_CAS, "%s: %s", device_get_name(sc->sc_dev), __func__);
717 #endif
718 
719 	callout_stop(&sc->sc_tick_ch);
720 	callout_stop(&sc->sc_rx_ch);
721 
722 	/* Disable all interrupts in order to avoid spurious ones. */
723 	CAS_WRITE_4(sc, CAS_INTMASK, 0xffffffff);
724 
725 	cas_reset_tx(sc);
726 	cas_reset_rx(sc);
727 
728 	/*
729 	 * Release any queued transmit buffers.
730 	 */
731 	while ((txs = STAILQ_FIRST(&sc->sc_txdirtyq)) != NULL) {
732 		STAILQ_REMOVE_HEAD(&sc->sc_txdirtyq, txs_q);
733 		if (txs->txs_ndescs != 0) {
734 			bus_dmamap_sync(sc->sc_tdmatag, txs->txs_dmamap,
735 			    BUS_DMASYNC_POSTWRITE);
736 			bus_dmamap_unload(sc->sc_tdmatag, txs->txs_dmamap);
737 			if (txs->txs_mbuf != NULL) {
738 				m_freem(txs->txs_mbuf);
739 				txs->txs_mbuf = NULL;
740 			}
741 		}
742 		STAILQ_INSERT_TAIL(&sc->sc_txfreeq, txs, txs_q);
743 	}
744 
745 	/*
746 	 * Mark the interface down and cancel the watchdog timer.
747 	 */
748 	if_setdrvflagbits(ifp, 0, (IFF_DRV_RUNNING | IFF_DRV_OACTIVE));
749 	sc->sc_flags &= ~CAS_LINK;
750 	sc->sc_wdog_timer = 0;
751 }
752 
753 static int
754 cas_reset_rx(struct cas_softc *sc)
755 {
756 
757 	/*
758 	 * Resetting while DMA is in progress can cause a bus hang, so we
759 	 * disable DMA first.
760 	 */
761 	(void)cas_disable_rx(sc);
762 	CAS_WRITE_4(sc, CAS_RX_CONF, 0);
763 	CAS_BARRIER(sc, CAS_RX_CONF, 4,
764 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
765 	if (!cas_bitwait(sc, CAS_RX_CONF, CAS_RX_CONF_RXDMA_EN, 0))
766 		device_printf(sc->sc_dev, "cannot disable RX DMA\n");
767 
768 	/* Finally, reset the ERX. */
769 	CAS_WRITE_4(sc, CAS_RESET, CAS_RESET_RX |
770 	    ((sc->sc_flags & CAS_SERDES) != 0 ? CAS_RESET_PCS_DIS : 0));
771 	CAS_BARRIER(sc, CAS_RESET, 4,
772 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
773 	if (!cas_bitwait(sc, CAS_RESET, CAS_RESET_RX, 0)) {
774 		device_printf(sc->sc_dev, "cannot reset receiver\n");
775 		return (1);
776 	}
777 	return (0);
778 }
779 
780 static int
781 cas_reset_tx(struct cas_softc *sc)
782 {
783 
784 	/*
785 	 * Resetting while DMA is in progress can cause a bus hang, so we
786 	 * disable DMA first.
787 	 */
788 	(void)cas_disable_tx(sc);
789 	CAS_WRITE_4(sc, CAS_TX_CONF, 0);
790 	CAS_BARRIER(sc, CAS_TX_CONF, 4,
791 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
792 	if (!cas_bitwait(sc, CAS_TX_CONF, CAS_TX_CONF_TXDMA_EN, 0))
793 		device_printf(sc->sc_dev, "cannot disable TX DMA\n");
794 
795 	/* Finally, reset the ETX. */
796 	CAS_WRITE_4(sc, CAS_RESET, CAS_RESET_TX |
797 	    ((sc->sc_flags & CAS_SERDES) != 0 ? CAS_RESET_PCS_DIS : 0));
798 	CAS_BARRIER(sc, CAS_RESET, 4,
799 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
800 	if (!cas_bitwait(sc, CAS_RESET, CAS_RESET_TX, 0)) {
801 		device_printf(sc->sc_dev, "cannot reset transmitter\n");
802 		return (1);
803 	}
804 	return (0);
805 }
806 
807 static int
808 cas_disable_rx(struct cas_softc *sc)
809 {
810 
811 	CAS_WRITE_4(sc, CAS_MAC_RX_CONF,
812 	    CAS_READ_4(sc, CAS_MAC_RX_CONF) & ~CAS_MAC_RX_CONF_EN);
813 	CAS_BARRIER(sc, CAS_MAC_RX_CONF, 4,
814 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
815 	if (cas_bitwait(sc, CAS_MAC_RX_CONF, CAS_MAC_RX_CONF_EN, 0))
816 		return (1);
817 	if (bootverbose)
818 		device_printf(sc->sc_dev, "cannot disable RX MAC\n");
819 	return (0);
820 }
821 
822 static int
823 cas_disable_tx(struct cas_softc *sc)
824 {
825 
826 	CAS_WRITE_4(sc, CAS_MAC_TX_CONF,
827 	    CAS_READ_4(sc, CAS_MAC_TX_CONF) & ~CAS_MAC_TX_CONF_EN);
828 	CAS_BARRIER(sc, CAS_MAC_TX_CONF, 4,
829 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
830 	if (cas_bitwait(sc, CAS_MAC_TX_CONF, CAS_MAC_TX_CONF_EN, 0))
831 		return (1);
832 	if (bootverbose)
833 		device_printf(sc->sc_dev, "cannot disable TX MAC\n");
834 	return (0);
835 }
836 
837 static inline void
838 cas_rxcompinit(struct cas_rx_comp *rxcomp)
839 {
840 
841 	rxcomp->crc_word1 = 0;
842 	rxcomp->crc_word2 = 0;
843 	rxcomp->crc_word3 =
844 	    htole64(CAS_SET(ETHER_HDR_LEN + sizeof(struct ip), CAS_RC3_CSO));
845 	rxcomp->crc_word4 = htole64(CAS_RC4_ZERO);
846 }
847 
848 static void
849 cas_meminit(struct cas_softc *sc)
850 {
851 	int i;
852 
853 	CAS_LOCK_ASSERT(sc, MA_OWNED);
854 
855 	/*
856 	 * Initialize the transmit descriptor ring.
857 	 */
858 	for (i = 0; i < CAS_NTXDESC; i++) {
859 		sc->sc_txdescs[i].cd_flags = 0;
860 		sc->sc_txdescs[i].cd_buf_ptr = 0;
861 	}
862 	sc->sc_txfree = CAS_MAXTXFREE;
863 	sc->sc_txnext = 0;
864 	sc->sc_txwin = 0;
865 
866 	/*
867 	 * Initialize the receive completion ring.
868 	 */
869 	for (i = 0; i < CAS_NRXCOMP; i++)
870 		cas_rxcompinit(&sc->sc_rxcomps[i]);
871 	sc->sc_rxcptr = 0;
872 
873 	/*
874 	 * Initialize the first receive descriptor ring.  We leave
875 	 * the second one zeroed as we don't actually use it.
876 	 */
877 	for (i = 0; i < CAS_NRXDESC; i++)
878 		CAS_INIT_RXDESC(sc, i, i);
879 	sc->sc_rxdptr = 0;
880 
881 	CAS_CDSYNC(sc, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
882 }
883 
884 static u_int
885 cas_descsize(u_int sz)
886 {
887 
888 	switch (sz) {
889 	case 32:
890 		return (CAS_DESC_32);
891 	case 64:
892 		return (CAS_DESC_64);
893 	case 128:
894 		return (CAS_DESC_128);
895 	case 256:
896 		return (CAS_DESC_256);
897 	case 512:
898 		return (CAS_DESC_512);
899 	case 1024:
900 		return (CAS_DESC_1K);
901 	case 2048:
902 		return (CAS_DESC_2K);
903 	case 4096:
904 		return (CAS_DESC_4K);
905 	case 8192:
906 		return (CAS_DESC_8K);
907 	default:
908 		printf("%s: invalid descriptor ring size %d\n", __func__, sz);
909 		return (CAS_DESC_32);
910 	}
911 }
912 
913 static u_int
914 cas_rxcompsize(u_int sz)
915 {
916 
917 	switch (sz) {
918 	case 128:
919 		return (CAS_RX_CONF_COMP_128);
920 	case 256:
921 		return (CAS_RX_CONF_COMP_256);
922 	case 512:
923 		return (CAS_RX_CONF_COMP_512);
924 	case 1024:
925 		return (CAS_RX_CONF_COMP_1K);
926 	case 2048:
927 		return (CAS_RX_CONF_COMP_2K);
928 	case 4096:
929 		return (CAS_RX_CONF_COMP_4K);
930 	case 8192:
931 		return (CAS_RX_CONF_COMP_8K);
932 	case 16384:
933 		return (CAS_RX_CONF_COMP_16K);
934 	case 32768:
935 		return (CAS_RX_CONF_COMP_32K);
936 	default:
937 		printf("%s: invalid dcompletion ring size %d\n", __func__, sz);
938 		return (CAS_RX_CONF_COMP_128);
939 	}
940 }
941 
942 static void
943 cas_init(void *xsc)
944 {
945 	struct cas_softc *sc = xsc;
946 
947 	CAS_LOCK(sc);
948 	cas_init_locked(sc);
949 	CAS_UNLOCK(sc);
950 }
951 
952 /*
953  * Initialization of interface; set up initialization block
954  * and transmit/receive descriptor rings.
955  */
956 static void
957 cas_init_locked(struct cas_softc *sc)
958 {
959 	if_t ifp = sc->sc_ifp;
960 	uint32_t v;
961 
962 	CAS_LOCK_ASSERT(sc, MA_OWNED);
963 
964 	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0)
965 		return;
966 
967 #ifdef CAS_DEBUG
968 	CTR2(KTR_CAS, "%s: %s: calling stop", device_get_name(sc->sc_dev),
969 	    __func__);
970 #endif
971 	/*
972 	 * Initialization sequence.  The numbered steps below correspond
973 	 * to the sequence outlined in section 6.3.5.1 in the Ethernet
974 	 * Channel Engine manual (part of the PCIO manual).
975 	 * See also the STP2002-STQ document from Sun Microsystems.
976 	 */
977 
978 	/* step 1 & 2.  Reset the Ethernet Channel. */
979 	cas_stop(ifp);
980 	cas_reset(sc);
981 #ifdef CAS_DEBUG
982 	CTR2(KTR_CAS, "%s: %s: restarting", device_get_name(sc->sc_dev),
983 	    __func__);
984 #endif
985 
986 	if ((sc->sc_flags & CAS_SERDES) == 0)
987 		/* Re-initialize the MIF. */
988 		cas_mifinit(sc);
989 
990 	/* step 3.  Setup data structures in host memory. */
991 	cas_meminit(sc);
992 
993 	/* step 4.  TX MAC registers & counters */
994 	cas_init_regs(sc);
995 
996 	/* step 5.  RX MAC registers & counters */
997 
998 	/* step 6 & 7.  Program Ring Base Addresses. */
999 	CAS_WRITE_4(sc, CAS_TX_DESC3_BASE_HI,
1000 	    (((uint64_t)CAS_CDTXDADDR(sc, 0)) >> 32));
1001 	CAS_WRITE_4(sc, CAS_TX_DESC3_BASE_LO,
1002 	    CAS_CDTXDADDR(sc, 0) & 0xffffffff);
1003 
1004 	CAS_WRITE_4(sc, CAS_RX_COMP_BASE_HI,
1005 	    (((uint64_t)CAS_CDRXCADDR(sc, 0)) >> 32));
1006 	CAS_WRITE_4(sc, CAS_RX_COMP_BASE_LO,
1007 	    CAS_CDRXCADDR(sc, 0) & 0xffffffff);
1008 
1009 	CAS_WRITE_4(sc, CAS_RX_DESC_BASE_HI,
1010 	    (((uint64_t)CAS_CDRXDADDR(sc, 0)) >> 32));
1011 	CAS_WRITE_4(sc, CAS_RX_DESC_BASE_LO,
1012 	    CAS_CDRXDADDR(sc, 0) & 0xffffffff);
1013 
1014 	if ((sc->sc_flags & CAS_REG_PLUS) != 0) {
1015 		CAS_WRITE_4(sc, CAS_RX_DESC2_BASE_HI,
1016 		    (((uint64_t)CAS_CDRXD2ADDR(sc, 0)) >> 32));
1017 		CAS_WRITE_4(sc, CAS_RX_DESC2_BASE_LO,
1018 		    CAS_CDRXD2ADDR(sc, 0) & 0xffffffff);
1019 	}
1020 
1021 #ifdef CAS_DEBUG
1022 	CTR5(KTR_CAS,
1023 	    "loading TXDR %lx, RXCR %lx, RXDR %lx, RXD2R %lx, cddma %lx",
1024 	    CAS_CDTXDADDR(sc, 0), CAS_CDRXCADDR(sc, 0), CAS_CDRXDADDR(sc, 0),
1025 	    CAS_CDRXD2ADDR(sc, 0), sc->sc_cddma);
1026 #endif
1027 
1028 	/* step 8.  Global Configuration & Interrupt Masks */
1029 
1030 	/* Disable weighted round robin. */
1031 	CAS_WRITE_4(sc, CAS_CAW, CAS_CAW_RR_DIS);
1032 
1033 	/*
1034 	 * Enable infinite bursts for revisions without PCI issues if
1035 	 * applicable.  Doing so greatly improves the TX performance.
1036 	 */
1037 	CAS_WRITE_4(sc, CAS_INF_BURST,
1038 	    (sc->sc_flags & CAS_TABORT) == 0 ? CAS_INF_BURST_EN :
1039 	    0);
1040 
1041 	/* Set up interrupts. */
1042 	CAS_WRITE_4(sc, CAS_INTMASK,
1043 	    ~(CAS_INTR_TX_INT_ME | CAS_INTR_TX_TAG_ERR |
1044 	    CAS_INTR_RX_DONE | CAS_INTR_RX_BUF_NA | CAS_INTR_RX_TAG_ERR |
1045 	    CAS_INTR_RX_COMP_FULL | CAS_INTR_RX_BUF_AEMPTY |
1046 	    CAS_INTR_RX_COMP_AFULL | CAS_INTR_RX_LEN_MMATCH |
1047 	    CAS_INTR_PCI_ERROR_INT
1048 #ifdef CAS_DEBUG
1049 	    | CAS_INTR_PCS_INT | CAS_INTR_MIF
1050 #endif
1051 	    ));
1052 	/* Don't clear top level interrupts when CAS_STATUS_ALIAS is read. */
1053 	CAS_WRITE_4(sc, CAS_CLEAR_ALIAS, 0);
1054 	CAS_WRITE_4(sc, CAS_MAC_RX_MASK, ~CAS_MAC_RX_OVERFLOW);
1055 	CAS_WRITE_4(sc, CAS_MAC_TX_MASK,
1056 	    ~(CAS_MAC_TX_UNDERRUN | CAS_MAC_TX_MAX_PKT_ERR));
1057 #ifdef CAS_DEBUG
1058 	CAS_WRITE_4(sc, CAS_MAC_CTRL_MASK,
1059 	    ~(CAS_MAC_CTRL_PAUSE_RCVD | CAS_MAC_CTRL_PAUSE |
1060 	    CAS_MAC_CTRL_NON_PAUSE));
1061 #else
1062 	CAS_WRITE_4(sc, CAS_MAC_CTRL_MASK,
1063 	    CAS_MAC_CTRL_PAUSE_RCVD | CAS_MAC_CTRL_PAUSE |
1064 	    CAS_MAC_CTRL_NON_PAUSE);
1065 #endif
1066 
1067 	/* Enable PCI error interrupts. */
1068 	CAS_WRITE_4(sc, CAS_ERROR_MASK,
1069 	    ~(CAS_ERROR_DTRTO | CAS_ERROR_OTHER | CAS_ERROR_DMAW_ZERO |
1070 	    CAS_ERROR_DMAR_ZERO | CAS_ERROR_RTRTO));
1071 
1072 	/* Enable PCI error interrupts in BIM configuration. */
1073 	CAS_WRITE_4(sc, CAS_BIM_CONF,
1074 	    CAS_BIM_CONF_DPAR_EN | CAS_BIM_CONF_RMA_EN | CAS_BIM_CONF_RTA_EN);
1075 
1076 	/*
1077 	 * step 9.  ETX Configuration: encode receive descriptor ring size,
1078 	 * enable DMA and disable pre-interrupt writeback completion.
1079 	 */
1080 	v = cas_descsize(CAS_NTXDESC) << CAS_TX_CONF_DESC3_SHFT;
1081 	CAS_WRITE_4(sc, CAS_TX_CONF, v | CAS_TX_CONF_TXDMA_EN |
1082 	    CAS_TX_CONF_RDPP_DIS | CAS_TX_CONF_PICWB_DIS);
1083 
1084 	/* step 10.  ERX Configuration */
1085 
1086 	/*
1087 	 * Encode receive completion and descriptor ring sizes, set the
1088 	 * swivel offset.
1089 	 */
1090 	v = cas_rxcompsize(CAS_NRXCOMP) << CAS_RX_CONF_COMP_SHFT;
1091 	v |= cas_descsize(CAS_NRXDESC) << CAS_RX_CONF_DESC_SHFT;
1092 	if ((sc->sc_flags & CAS_REG_PLUS) != 0)
1093 		v |= cas_descsize(CAS_NRXDESC2) << CAS_RX_CONF_DESC2_SHFT;
1094 	CAS_WRITE_4(sc, CAS_RX_CONF,
1095 	    v | (ETHER_ALIGN << CAS_RX_CONF_SOFF_SHFT));
1096 
1097 	/* Set the PAUSE thresholds.  We use the maximum OFF threshold. */
1098 	CAS_WRITE_4(sc, CAS_RX_PTHRS,
1099 	    (111 << CAS_RX_PTHRS_XOFF_SHFT) | (15 << CAS_RX_PTHRS_XON_SHFT));
1100 
1101 	/* RX blanking */
1102 	CAS_WRITE_4(sc, CAS_RX_BLANK,
1103 	    (15 << CAS_RX_BLANK_TIME_SHFT) | (5 << CAS_RX_BLANK_PKTS_SHFT));
1104 
1105 	/* Set RX_COMP_AFULL threshold to half of the RX completions. */
1106 	CAS_WRITE_4(sc, CAS_RX_AEMPTY_THRS,
1107 	    (CAS_NRXCOMP / 2) << CAS_RX_AEMPTY_COMP_SHFT);
1108 
1109 	/* Initialize the RX page size register as appropriate for 8k. */
1110 	CAS_WRITE_4(sc, CAS_RX_PSZ,
1111 	    (CAS_RX_PSZ_8K << CAS_RX_PSZ_SHFT) |
1112 	    (4 << CAS_RX_PSZ_MB_CNT_SHFT) |
1113 	    (CAS_RX_PSZ_MB_STRD_2K << CAS_RX_PSZ_MB_STRD_SHFT) |
1114 	    (CAS_RX_PSZ_MB_OFF_64 << CAS_RX_PSZ_MB_OFF_SHFT));
1115 
1116 	/* Disable RX random early detection. */
1117 	CAS_WRITE_4(sc,	CAS_RX_RED, 0);
1118 
1119 	/* Zero the RX reassembly DMA table. */
1120 	for (v = 0; v <= CAS_RX_REAS_DMA_ADDR_LC; v++) {
1121 		CAS_WRITE_4(sc,	CAS_RX_REAS_DMA_ADDR, v);
1122 		CAS_WRITE_4(sc,	CAS_RX_REAS_DMA_DATA_LO, 0);
1123 		CAS_WRITE_4(sc,	CAS_RX_REAS_DMA_DATA_MD, 0);
1124 		CAS_WRITE_4(sc,	CAS_RX_REAS_DMA_DATA_HI, 0);
1125 	}
1126 
1127 	/* Ensure the RX control FIFO and RX IPP FIFO addresses are zero. */
1128 	CAS_WRITE_4(sc, CAS_RX_CTRL_FIFO, 0);
1129 	CAS_WRITE_4(sc, CAS_RX_IPP_ADDR, 0);
1130 
1131 	/* Finally, enable RX DMA. */
1132 	CAS_WRITE_4(sc, CAS_RX_CONF,
1133 	    CAS_READ_4(sc, CAS_RX_CONF) | CAS_RX_CONF_RXDMA_EN);
1134 
1135 	/* step 11.  Configure Media. */
1136 
1137 	/* step 12.  RX_MAC Configuration Register */
1138 	v = CAS_READ_4(sc, CAS_MAC_RX_CONF);
1139 	v &= ~(CAS_MAC_RX_CONF_STRPPAD | CAS_MAC_RX_CONF_EN);
1140 	v |= CAS_MAC_RX_CONF_STRPFCS;
1141 	sc->sc_mac_rxcfg = v;
1142 	/*
1143 	 * Clear the RX filter and reprogram it.  This will also set the
1144 	 * current RX MAC configuration and enable it.
1145 	 */
1146 	cas_setladrf(sc);
1147 
1148 	/* step 13.  TX_MAC Configuration Register */
1149 	v = CAS_READ_4(sc, CAS_MAC_TX_CONF);
1150 	v |= CAS_MAC_TX_CONF_EN;
1151 	(void)cas_disable_tx(sc);
1152 	CAS_WRITE_4(sc, CAS_MAC_TX_CONF, v);
1153 
1154 	/* step 14.  Issue Transmit Pending command. */
1155 
1156 	/* step 15.  Give the receiver a swift kick. */
1157 	CAS_WRITE_4(sc, CAS_RX_KICK, CAS_NRXDESC - 4);
1158 	CAS_WRITE_4(sc, CAS_RX_COMP_TAIL, 0);
1159 	if ((sc->sc_flags & CAS_REG_PLUS) != 0)
1160 		CAS_WRITE_4(sc, CAS_RX_KICK2, CAS_NRXDESC2 - 4);
1161 
1162 	if_setdrvflagbits(ifp, IFF_DRV_RUNNING, 0);
1163 	if_setdrvflagbits(ifp, 0, IFF_DRV_OACTIVE);
1164 
1165 	mii_mediachg(sc->sc_mii);
1166 
1167 	/* Start the one second timer. */
1168 	sc->sc_wdog_timer = 0;
1169 	callout_reset(&sc->sc_tick_ch, hz, cas_tick, sc);
1170 }
1171 
1172 static int
1173 cas_load_txmbuf(struct cas_softc *sc, struct mbuf **m_head)
1174 {
1175 	bus_dma_segment_t txsegs[CAS_NTXSEGS];
1176 	struct cas_txsoft *txs;
1177 	struct ip *ip;
1178 	struct mbuf *m;
1179 	uint64_t cflags;
1180 	int error, nexttx, nsegs, offset, seg;
1181 
1182 	CAS_LOCK_ASSERT(sc, MA_OWNED);
1183 
1184 	/* Get a work queue entry. */
1185 	if ((txs = STAILQ_FIRST(&sc->sc_txfreeq)) == NULL) {
1186 		/* Ran out of descriptors. */
1187 		return (ENOBUFS);
1188 	}
1189 
1190 	cflags = 0;
1191 	if (((*m_head)->m_pkthdr.csum_flags & CAS_CSUM_FEATURES) != 0) {
1192 		if (M_WRITABLE(*m_head) == 0) {
1193 			m = m_dup(*m_head, M_NOWAIT);
1194 			m_freem(*m_head);
1195 			*m_head = m;
1196 			if (m == NULL)
1197 				return (ENOBUFS);
1198 		}
1199 		offset = sizeof(struct ether_header);
1200 		m = m_pullup(*m_head, offset + sizeof(struct ip));
1201 		if (m == NULL) {
1202 			*m_head = NULL;
1203 			return (ENOBUFS);
1204 		}
1205 		ip = (struct ip *)(mtod(m, caddr_t) + offset);
1206 		offset += (ip->ip_hl << 2);
1207 		cflags = (offset << CAS_TD_CKSUM_START_SHFT) |
1208 		    ((offset + m->m_pkthdr.csum_data) <<
1209 		    CAS_TD_CKSUM_STUFF_SHFT) | CAS_TD_CKSUM_EN;
1210 		*m_head = m;
1211 	}
1212 
1213 	error = bus_dmamap_load_mbuf_sg(sc->sc_tdmatag, txs->txs_dmamap,
1214 	    *m_head, txsegs, &nsegs, BUS_DMA_NOWAIT);
1215 	if (error == EFBIG) {
1216 		m = m_collapse(*m_head, M_NOWAIT, CAS_NTXSEGS);
1217 		if (m == NULL) {
1218 			m_freem(*m_head);
1219 			*m_head = NULL;
1220 			return (ENOBUFS);
1221 		}
1222 		*m_head = m;
1223 		error = bus_dmamap_load_mbuf_sg(sc->sc_tdmatag,
1224 		    txs->txs_dmamap, *m_head, txsegs, &nsegs,
1225 		    BUS_DMA_NOWAIT);
1226 		if (error != 0) {
1227 			m_freem(*m_head);
1228 			*m_head = NULL;
1229 			return (error);
1230 		}
1231 	} else if (error != 0)
1232 		return (error);
1233 	/* If nsegs is wrong then the stack is corrupt. */
1234 	KASSERT(nsegs <= CAS_NTXSEGS,
1235 	    ("%s: too many DMA segments (%d)", __func__, nsegs));
1236 	if (nsegs == 0) {
1237 		m_freem(*m_head);
1238 		*m_head = NULL;
1239 		return (EIO);
1240 	}
1241 
1242 	/*
1243 	 * Ensure we have enough descriptors free to describe
1244 	 * the packet.  Note, we always reserve one descriptor
1245 	 * at the end of the ring as a termination point, in
1246 	 * order to prevent wrap-around.
1247 	 */
1248 	if (nsegs > sc->sc_txfree - 1) {
1249 		txs->txs_ndescs = 0;
1250 		bus_dmamap_unload(sc->sc_tdmatag, txs->txs_dmamap);
1251 		return (ENOBUFS);
1252 	}
1253 
1254 	txs->txs_ndescs = nsegs;
1255 	txs->txs_firstdesc = sc->sc_txnext;
1256 	nexttx = txs->txs_firstdesc;
1257 	for (seg = 0; seg < nsegs; seg++, nexttx = CAS_NEXTTX(nexttx)) {
1258 #ifdef CAS_DEBUG
1259 		CTR6(KTR_CAS,
1260 		    "%s: mapping seg %d (txd %d), len %lx, addr %#lx (%#lx)",
1261 		    __func__, seg, nexttx, txsegs[seg].ds_len,
1262 		    txsegs[seg].ds_addr, htole64(txsegs[seg].ds_addr));
1263 #endif
1264 		sc->sc_txdescs[nexttx].cd_buf_ptr =
1265 		    htole64(txsegs[seg].ds_addr);
1266 		KASSERT(txsegs[seg].ds_len <
1267 		    CAS_TD_BUF_LEN_MASK >> CAS_TD_BUF_LEN_SHFT,
1268 		    ("%s: segment size too large!", __func__));
1269 		sc->sc_txdescs[nexttx].cd_flags =
1270 		    htole64(txsegs[seg].ds_len << CAS_TD_BUF_LEN_SHFT);
1271 		txs->txs_lastdesc = nexttx;
1272 	}
1273 
1274 	/* Set EOF on the last descriptor. */
1275 #ifdef CAS_DEBUG
1276 	CTR3(KTR_CAS, "%s: end of frame at segment %d, TX %d",
1277 	    __func__, seg, nexttx);
1278 #endif
1279 	sc->sc_txdescs[txs->txs_lastdesc].cd_flags |=
1280 	    htole64(CAS_TD_END_OF_FRAME);
1281 
1282 	/* Lastly set SOF on the first descriptor. */
1283 #ifdef CAS_DEBUG
1284 	CTR3(KTR_CAS, "%s: start of frame at segment %d, TX %d",
1285 	    __func__, seg, nexttx);
1286 #endif
1287 	if (sc->sc_txwin += nsegs > CAS_MAXTXFREE * 2 / 3) {
1288 		sc->sc_txwin = 0;
1289 		sc->sc_txdescs[txs->txs_firstdesc].cd_flags |=
1290 		    htole64(cflags | CAS_TD_START_OF_FRAME | CAS_TD_INT_ME);
1291 	} else
1292 		sc->sc_txdescs[txs->txs_firstdesc].cd_flags |=
1293 		    htole64(cflags | CAS_TD_START_OF_FRAME);
1294 
1295 	/* Sync the DMA map. */
1296 	bus_dmamap_sync(sc->sc_tdmatag, txs->txs_dmamap,
1297 	    BUS_DMASYNC_PREWRITE);
1298 
1299 #ifdef CAS_DEBUG
1300 	CTR4(KTR_CAS, "%s: setting firstdesc=%d, lastdesc=%d, ndescs=%d",
1301 	    __func__, txs->txs_firstdesc, txs->txs_lastdesc,
1302 	    txs->txs_ndescs);
1303 #endif
1304 	STAILQ_REMOVE_HEAD(&sc->sc_txfreeq, txs_q);
1305 	STAILQ_INSERT_TAIL(&sc->sc_txdirtyq, txs, txs_q);
1306 	txs->txs_mbuf = *m_head;
1307 
1308 	sc->sc_txnext = CAS_NEXTTX(txs->txs_lastdesc);
1309 	sc->sc_txfree -= txs->txs_ndescs;
1310 
1311 	return (0);
1312 }
1313 
1314 static void
1315 cas_init_regs(struct cas_softc *sc)
1316 {
1317 	int i;
1318 	const u_char *laddr = if_getlladdr(sc->sc_ifp);
1319 
1320 	CAS_LOCK_ASSERT(sc, MA_OWNED);
1321 
1322 	/* These registers are not cleared on reset. */
1323 	if ((sc->sc_flags & CAS_INITED) == 0) {
1324 		/* magic values */
1325 		CAS_WRITE_4(sc, CAS_MAC_IPG0, 0);
1326 		CAS_WRITE_4(sc, CAS_MAC_IPG1, 8);
1327 		CAS_WRITE_4(sc, CAS_MAC_IPG2, 4);
1328 
1329 		/* min frame length */
1330 		CAS_WRITE_4(sc, CAS_MAC_MIN_FRAME, ETHER_MIN_LEN);
1331 		/* max frame length and max burst size */
1332 		CAS_WRITE_4(sc, CAS_MAC_MAX_BF,
1333 		    ((ETHER_MAX_LEN_JUMBO + ETHER_VLAN_ENCAP_LEN) <<
1334 		    CAS_MAC_MAX_BF_FRM_SHFT) |
1335 		    (0x2000 << CAS_MAC_MAX_BF_BST_SHFT));
1336 
1337 		/* more magic values */
1338 		CAS_WRITE_4(sc, CAS_MAC_PREAMBLE_LEN, 0x7);
1339 		CAS_WRITE_4(sc, CAS_MAC_JAM_SIZE, 0x4);
1340 		CAS_WRITE_4(sc, CAS_MAC_ATTEMPT_LIMIT, 0x10);
1341 		CAS_WRITE_4(sc, CAS_MAC_CTRL_TYPE, 0x8808);
1342 
1343 		/* random number seed */
1344 		CAS_WRITE_4(sc, CAS_MAC_RANDOM_SEED,
1345 		    ((laddr[5] << 8) | laddr[4]) & 0x3ff);
1346 
1347 		/* secondary MAC addresses: 0:0:0:0:0:0 */
1348 		for (i = CAS_MAC_ADDR3; i <= CAS_MAC_ADDR41;
1349 		    i += CAS_MAC_ADDR4 - CAS_MAC_ADDR3)
1350 			CAS_WRITE_4(sc, i, 0);
1351 
1352 		/* MAC control address: 01:80:c2:00:00:01 */
1353 		CAS_WRITE_4(sc, CAS_MAC_ADDR42, 0x0001);
1354 		CAS_WRITE_4(sc, CAS_MAC_ADDR43, 0xc200);
1355 		CAS_WRITE_4(sc, CAS_MAC_ADDR44, 0x0180);
1356 
1357 		/* MAC filter address: 0:0:0:0:0:0 */
1358 		CAS_WRITE_4(sc, CAS_MAC_AFILTER0, 0);
1359 		CAS_WRITE_4(sc, CAS_MAC_AFILTER1, 0);
1360 		CAS_WRITE_4(sc, CAS_MAC_AFILTER2, 0);
1361 		CAS_WRITE_4(sc, CAS_MAC_AFILTER_MASK1_2, 0);
1362 		CAS_WRITE_4(sc, CAS_MAC_AFILTER_MASK0, 0);
1363 
1364 		/* Zero the hash table. */
1365 		for (i = CAS_MAC_HASH0; i <= CAS_MAC_HASH15;
1366 		    i += CAS_MAC_HASH1 - CAS_MAC_HASH0)
1367 			CAS_WRITE_4(sc, i, 0);
1368 
1369 		sc->sc_flags |= CAS_INITED;
1370 	}
1371 
1372 	/* Counters need to be zeroed. */
1373 	CAS_WRITE_4(sc, CAS_MAC_NORM_COLL_CNT, 0);
1374 	CAS_WRITE_4(sc, CAS_MAC_FIRST_COLL_CNT, 0);
1375 	CAS_WRITE_4(sc, CAS_MAC_EXCESS_COLL_CNT, 0);
1376 	CAS_WRITE_4(sc, CAS_MAC_LATE_COLL_CNT, 0);
1377 	CAS_WRITE_4(sc, CAS_MAC_DEFER_TMR_CNT, 0);
1378 	CAS_WRITE_4(sc, CAS_MAC_PEAK_ATTEMPTS, 0);
1379 	CAS_WRITE_4(sc, CAS_MAC_RX_FRAME_COUNT, 0);
1380 	CAS_WRITE_4(sc, CAS_MAC_RX_LEN_ERR_CNT, 0);
1381 	CAS_WRITE_4(sc, CAS_MAC_RX_ALIGN_ERR, 0);
1382 	CAS_WRITE_4(sc, CAS_MAC_RX_CRC_ERR_CNT, 0);
1383 	CAS_WRITE_4(sc, CAS_MAC_RX_CODE_VIOL, 0);
1384 
1385 	/* Set XOFF PAUSE time. */
1386 	CAS_WRITE_4(sc, CAS_MAC_SPC, 0x1BF0 << CAS_MAC_SPC_TIME_SHFT);
1387 
1388 	/* Set the station address. */
1389 	CAS_WRITE_4(sc, CAS_MAC_ADDR0, (laddr[4] << 8) | laddr[5]);
1390 	CAS_WRITE_4(sc, CAS_MAC_ADDR1, (laddr[2] << 8) | laddr[3]);
1391 	CAS_WRITE_4(sc, CAS_MAC_ADDR2, (laddr[0] << 8) | laddr[1]);
1392 
1393 	/* Enable MII outputs. */
1394 	CAS_WRITE_4(sc, CAS_MAC_XIF_CONF, CAS_MAC_XIF_CONF_TX_OE);
1395 }
1396 
1397 static void
1398 cas_tx_task(void *arg, int pending __unused)
1399 {
1400 	if_t ifp;
1401 
1402 	ifp = (if_t)arg;
1403 	cas_start(ifp);
1404 }
1405 
1406 static inline void
1407 cas_txkick(struct cas_softc *sc)
1408 {
1409 
1410 	/*
1411 	 * Update the TX kick register.  This register has to point to the
1412 	 * descriptor after the last valid one and for optimum performance
1413 	 * should be incremented in multiples of 4 (the DMA engine fetches/
1414 	 * updates descriptors in batches of 4).
1415 	 */
1416 #ifdef CAS_DEBUG
1417 	CTR3(KTR_CAS, "%s: %s: kicking TX %d",
1418 	    device_get_name(sc->sc_dev), __func__, sc->sc_txnext);
1419 #endif
1420 	CAS_CDSYNC(sc, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1421 	CAS_WRITE_4(sc, CAS_TX_KICK3, sc->sc_txnext);
1422 }
1423 
1424 static void
1425 cas_start(if_t ifp)
1426 {
1427 	struct cas_softc *sc = if_getsoftc(ifp);
1428 	struct mbuf *m;
1429 	int kicked, ntx;
1430 
1431 	CAS_LOCK(sc);
1432 
1433 	if ((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
1434 	    IFF_DRV_RUNNING || (sc->sc_flags & CAS_LINK) == 0) {
1435 		CAS_UNLOCK(sc);
1436 		return;
1437 	}
1438 
1439 	if (sc->sc_txfree < CAS_MAXTXFREE / 4)
1440 		cas_tint(sc);
1441 
1442 #ifdef CAS_DEBUG
1443 	CTR4(KTR_CAS, "%s: %s: txfree %d, txnext %d",
1444 	    device_get_name(sc->sc_dev), __func__, sc->sc_txfree,
1445 	    sc->sc_txnext);
1446 #endif
1447 	ntx = 0;
1448 	kicked = 0;
1449 	for (; !if_sendq_empty(ifp) && sc->sc_txfree > 1;) {
1450 		m = if_dequeue(ifp);
1451 		if (m == NULL)
1452 			break;
1453 		if (cas_load_txmbuf(sc, &m) != 0) {
1454 			if (m == NULL)
1455 				break;
1456 			if_setdrvflagbits(ifp, IFF_DRV_OACTIVE, 0);
1457 			if_sendq_prepend(ifp, m);
1458 			break;
1459 		}
1460 		if ((sc->sc_txnext % 4) == 0) {
1461 			cas_txkick(sc);
1462 			kicked = 1;
1463 		} else
1464 			kicked = 0;
1465 		ntx++;
1466 		BPF_MTAP(ifp, m);
1467 	}
1468 
1469 	if (ntx > 0) {
1470 		if (kicked == 0)
1471 			cas_txkick(sc);
1472 #ifdef CAS_DEBUG
1473 		CTR2(KTR_CAS, "%s: packets enqueued, OWN on %d",
1474 		    device_get_name(sc->sc_dev), sc->sc_txnext);
1475 #endif
1476 
1477 		/* Set a watchdog timer in case the chip flakes out. */
1478 		sc->sc_wdog_timer = 5;
1479 #ifdef CAS_DEBUG
1480 		CTR3(KTR_CAS, "%s: %s: watchdog %d",
1481 		    device_get_name(sc->sc_dev), __func__,
1482 		    sc->sc_wdog_timer);
1483 #endif
1484 	}
1485 
1486 	CAS_UNLOCK(sc);
1487 }
1488 
1489 static void
1490 cas_tint(struct cas_softc *sc)
1491 {
1492 	if_t ifp = sc->sc_ifp;
1493 	struct cas_txsoft *txs;
1494 	int progress;
1495 	uint32_t txlast;
1496 #ifdef CAS_DEBUG
1497 	int i;
1498 
1499 	CAS_LOCK_ASSERT(sc, MA_OWNED);
1500 
1501 	CTR2(KTR_CAS, "%s: %s", device_get_name(sc->sc_dev), __func__);
1502 #endif
1503 
1504 	/*
1505 	 * Go through our TX list and free mbufs for those
1506 	 * frames that have been transmitted.
1507 	 */
1508 	progress = 0;
1509 	CAS_CDSYNC(sc, BUS_DMASYNC_POSTREAD);
1510 	while ((txs = STAILQ_FIRST(&sc->sc_txdirtyq)) != NULL) {
1511 #ifdef CAS_DEBUG
1512 		if ((if_getflags(ifp) & IFF_DEBUG) != 0) {
1513 			printf("    txsoft %p transmit chain:\n", txs);
1514 			for (i = txs->txs_firstdesc;; i = CAS_NEXTTX(i)) {
1515 				printf("descriptor %d: ", i);
1516 				printf("cd_flags: 0x%016llx\t",
1517 				    (long long)le64toh(
1518 				    sc->sc_txdescs[i].cd_flags));
1519 				printf("cd_buf_ptr: 0x%016llx\n",
1520 				    (long long)le64toh(
1521 				    sc->sc_txdescs[i].cd_buf_ptr));
1522 				if (i == txs->txs_lastdesc)
1523 					break;
1524 			}
1525 		}
1526 #endif
1527 
1528 		/*
1529 		 * In theory, we could harvest some descriptors before
1530 		 * the ring is empty, but that's a bit complicated.
1531 		 *
1532 		 * CAS_TX_COMPn points to the last descriptor
1533 		 * processed + 1.
1534 		 */
1535 		txlast = CAS_READ_4(sc, CAS_TX_COMP3);
1536 #ifdef CAS_DEBUG
1537 		CTR4(KTR_CAS, "%s: txs->txs_firstdesc = %d, "
1538 		    "txs->txs_lastdesc = %d, txlast = %d",
1539 		    __func__, txs->txs_firstdesc, txs->txs_lastdesc, txlast);
1540 #endif
1541 		if (txs->txs_firstdesc <= txs->txs_lastdesc) {
1542 			if ((txlast >= txs->txs_firstdesc) &&
1543 			    (txlast <= txs->txs_lastdesc))
1544 				break;
1545 		} else {
1546 			/* Ick -- this command wraps. */
1547 			if ((txlast >= txs->txs_firstdesc) ||
1548 			    (txlast <= txs->txs_lastdesc))
1549 				break;
1550 		}
1551 
1552 #ifdef CAS_DEBUG
1553 		CTR1(KTR_CAS, "%s: releasing a descriptor", __func__);
1554 #endif
1555 		STAILQ_REMOVE_HEAD(&sc->sc_txdirtyq, txs_q);
1556 
1557 		sc->sc_txfree += txs->txs_ndescs;
1558 
1559 		bus_dmamap_sync(sc->sc_tdmatag, txs->txs_dmamap,
1560 		    BUS_DMASYNC_POSTWRITE);
1561 		bus_dmamap_unload(sc->sc_tdmatag, txs->txs_dmamap);
1562 		if (txs->txs_mbuf != NULL) {
1563 			m_freem(txs->txs_mbuf);
1564 			txs->txs_mbuf = NULL;
1565 		}
1566 
1567 		STAILQ_INSERT_TAIL(&sc->sc_txfreeq, txs, txs_q);
1568 
1569 		if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
1570 		progress = 1;
1571 	}
1572 
1573 #ifdef CAS_DEBUG
1574 	CTR5(KTR_CAS, "%s: CAS_TX_SM1 %x CAS_TX_SM2 %x CAS_TX_DESC_BASE %llx "
1575 	    "CAS_TX_COMP3 %x",
1576 	    __func__, CAS_READ_4(sc, CAS_TX_SM1), CAS_READ_4(sc, CAS_TX_SM2),
1577 	    ((long long)CAS_READ_4(sc, CAS_TX_DESC3_BASE_HI) << 32) |
1578 	    CAS_READ_4(sc, CAS_TX_DESC3_BASE_LO),
1579 	    CAS_READ_4(sc, CAS_TX_COMP3));
1580 #endif
1581 
1582 	if (progress) {
1583 		/* We freed some descriptors, so reset IFF_DRV_OACTIVE. */
1584 		if_setdrvflagbits(ifp, 0, IFF_DRV_OACTIVE);
1585 		if (STAILQ_EMPTY(&sc->sc_txdirtyq))
1586 			sc->sc_wdog_timer = 0;
1587 	}
1588 
1589 #ifdef CAS_DEBUG
1590 	CTR3(KTR_CAS, "%s: %s: watchdog %d",
1591 	    device_get_name(sc->sc_dev), __func__, sc->sc_wdog_timer);
1592 #endif
1593 }
1594 
1595 static void
1596 cas_rint_timeout(void *arg)
1597 {
1598 	struct epoch_tracker et;
1599 	struct cas_softc *sc = arg;
1600 
1601 	CAS_LOCK_ASSERT(sc, MA_OWNED);
1602 
1603 	NET_EPOCH_ENTER(et);
1604 	cas_rint(sc);
1605 	NET_EPOCH_EXIT(et);
1606 }
1607 
1608 static void
1609 cas_rint(struct cas_softc *sc)
1610 {
1611 	struct cas_rxdsoft *rxds, *rxds2;
1612 	if_t ifp = sc->sc_ifp;
1613 	struct mbuf *m, *m2;
1614 	uint64_t word1, word2, word3 __unused, word4;
1615 	uint32_t rxhead;
1616 	u_int idx, idx2, len, off, skip;
1617 
1618 	CAS_LOCK_ASSERT(sc, MA_OWNED);
1619 
1620 	callout_stop(&sc->sc_rx_ch);
1621 
1622 #ifdef CAS_DEBUG
1623 	CTR2(KTR_CAS, "%s: %s", device_get_name(sc->sc_dev), __func__);
1624 #endif
1625 
1626 #define	PRINTWORD(n, delimiter)						\
1627 	printf("word ## n: 0x%016llx%c", (long long)word ## n, delimiter)
1628 
1629 #define	SKIPASSERT(n)							\
1630 	KASSERT(sc->sc_rxcomps[sc->sc_rxcptr].crc_word ## n == 0,	\
1631 	    ("%s: word ## n not 0", __func__))
1632 
1633 #define	WORDTOH(n)							\
1634 	word ## n = le64toh(sc->sc_rxcomps[sc->sc_rxcptr].crc_word ## n)
1635 
1636 	/*
1637 	 * Read the completion head register once.  This limits
1638 	 * how long the following loop can execute.
1639 	 */
1640 	rxhead = CAS_READ_4(sc, CAS_RX_COMP_HEAD);
1641 #ifdef CAS_DEBUG
1642 	CTR4(KTR_CAS, "%s: sc->sc_rxcptr %d, sc->sc_rxdptr %d, head %d",
1643 	    __func__, sc->sc_rxcptr, sc->sc_rxdptr, rxhead);
1644 #endif
1645 	skip = 0;
1646 	CAS_CDSYNC(sc, BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
1647 	for (; sc->sc_rxcptr != rxhead;
1648 	    sc->sc_rxcptr = CAS_NEXTRXCOMP(sc->sc_rxcptr)) {
1649 		if (skip != 0) {
1650 			SKIPASSERT(1);
1651 			SKIPASSERT(2);
1652 			SKIPASSERT(3);
1653 
1654 			--skip;
1655 			goto skip;
1656 		}
1657 
1658 		WORDTOH(1);
1659 		WORDTOH(2);
1660 		WORDTOH(3);
1661 		WORDTOH(4);
1662 
1663 #ifdef CAS_DEBUG
1664 		if ((if_getflags(ifp) & IFF_DEBUG) != 0) {
1665 			printf("    completion %d: ", sc->sc_rxcptr);
1666 			PRINTWORD(1, '\t');
1667 			PRINTWORD(2, '\t');
1668 			PRINTWORD(3, '\t');
1669 			PRINTWORD(4, '\n');
1670 		}
1671 #endif
1672 
1673 		if (__predict_false(
1674 		    (word1 & CAS_RC1_TYPE_MASK) == CAS_RC1_TYPE_HW ||
1675 		    (word4 & CAS_RC4_ZERO) != 0)) {
1676 			/*
1677 			 * The descriptor is still marked as owned, although
1678 			 * it is supposed to have completed.  This has been
1679 			 * observed on some machines.  Just exiting here
1680 			 * might leave the packet sitting around until another
1681 			 * one arrives to trigger a new interrupt, which is
1682 			 * generally undesirable, so set up a timeout.
1683 			 */
1684 			callout_reset(&sc->sc_rx_ch, CAS_RXOWN_TICKS,
1685 			    cas_rint_timeout, sc);
1686 			break;
1687 		}
1688 
1689 		if (__predict_false(
1690 		    (word4 & (CAS_RC4_BAD | CAS_RC4_LEN_MMATCH)) != 0)) {
1691 			if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
1692 			device_printf(sc->sc_dev,
1693 			    "receive error: CRC error\n");
1694 			continue;
1695 		}
1696 
1697 		KASSERT(CAS_GET(word1, CAS_RC1_DATA_SIZE) == 0 ||
1698 		    CAS_GET(word2, CAS_RC2_HDR_SIZE) == 0,
1699 		    ("%s: data and header present", __func__));
1700 		KASSERT((word1 & CAS_RC1_SPLIT_PKT) == 0 ||
1701 		    CAS_GET(word2, CAS_RC2_HDR_SIZE) == 0,
1702 		    ("%s: split and header present", __func__));
1703 		KASSERT(CAS_GET(word1, CAS_RC1_DATA_SIZE) == 0 ||
1704 		    (word1 & CAS_RC1_RELEASE_HDR) == 0,
1705 		    ("%s: data present but header release", __func__));
1706 		KASSERT(CAS_GET(word2, CAS_RC2_HDR_SIZE) == 0 ||
1707 		    (word1 & CAS_RC1_RELEASE_DATA) == 0,
1708 		    ("%s: header present but data release", __func__));
1709 
1710 		if ((len = CAS_GET(word2, CAS_RC2_HDR_SIZE)) != 0) {
1711 			idx = CAS_GET(word2, CAS_RC2_HDR_INDEX);
1712 			off = CAS_GET(word2, CAS_RC2_HDR_OFF);
1713 #ifdef CAS_DEBUG
1714 			CTR4(KTR_CAS, "%s: hdr at idx %d, off %d, len %d",
1715 			    __func__, idx, off, len);
1716 #endif
1717 			rxds = &sc->sc_rxdsoft[idx];
1718 			MGETHDR(m, M_NOWAIT, MT_DATA);
1719 			if (m != NULL) {
1720 				refcount_acquire(&rxds->rxds_refcount);
1721 				bus_dmamap_sync(sc->sc_rdmatag,
1722 				    rxds->rxds_dmamap, BUS_DMASYNC_POSTREAD);
1723 				m_extadd(m, (char *)rxds->rxds_buf +
1724 				    off * 256 + ETHER_ALIGN, len, cas_free,
1725 				    sc, (void *)(uintptr_t)idx,
1726 				    M_RDONLY, EXT_NET_DRV);
1727 				if ((m->m_flags & M_EXT) == 0) {
1728 					m_freem(m);
1729 					m = NULL;
1730 				}
1731 			}
1732 			if (m != NULL) {
1733 				m->m_pkthdr.rcvif = ifp;
1734 				m->m_pkthdr.len = m->m_len = len;
1735 				if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
1736 				if ((if_getcapenable(ifp) & IFCAP_RXCSUM) != 0)
1737 					cas_rxcksum(m, CAS_GET(word4,
1738 					    CAS_RC4_TCP_CSUM));
1739 				/* Pass it on. */
1740 				CAS_UNLOCK(sc);
1741 				if_input(ifp, m);
1742 				CAS_LOCK(sc);
1743 			} else
1744 				if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1745 
1746 			if ((word1 & CAS_RC1_RELEASE_HDR) != 0 &&
1747 			    refcount_release(&rxds->rxds_refcount) != 0)
1748 				cas_add_rxdesc(sc, idx);
1749 		} else if ((len = CAS_GET(word1, CAS_RC1_DATA_SIZE)) != 0) {
1750 			idx = CAS_GET(word1, CAS_RC1_DATA_INDEX);
1751 			off = CAS_GET(word1, CAS_RC1_DATA_OFF);
1752 #ifdef CAS_DEBUG
1753 			CTR4(KTR_CAS, "%s: data at idx %d, off %d, len %d",
1754 			    __func__, idx, off, len);
1755 #endif
1756 			rxds = &sc->sc_rxdsoft[idx];
1757 			MGETHDR(m, M_NOWAIT, MT_DATA);
1758 			if (m != NULL) {
1759 				refcount_acquire(&rxds->rxds_refcount);
1760 				off += ETHER_ALIGN;
1761 				m->m_len = min(CAS_PAGE_SIZE - off, len);
1762 				bus_dmamap_sync(sc->sc_rdmatag,
1763 				    rxds->rxds_dmamap, BUS_DMASYNC_POSTREAD);
1764 				m_extadd(m, (char *)rxds->rxds_buf + off,
1765 				    m->m_len, cas_free, sc,
1766 				    (void *)(uintptr_t)idx, M_RDONLY,
1767 				    EXT_NET_DRV);
1768 				if ((m->m_flags & M_EXT) == 0) {
1769 					m_freem(m);
1770 					m = NULL;
1771 				}
1772 			}
1773 			idx2 = 0;
1774 			m2 = NULL;
1775 			rxds2 = NULL;
1776 			if ((word1 & CAS_RC1_SPLIT_PKT) != 0) {
1777 				KASSERT((word1 & CAS_RC1_RELEASE_NEXT) != 0,
1778 				    ("%s: split but no release next",
1779 				    __func__));
1780 
1781 				idx2 = CAS_GET(word2, CAS_RC2_NEXT_INDEX);
1782 #ifdef CAS_DEBUG
1783 				CTR2(KTR_CAS, "%s: split at idx %d",
1784 				    __func__, idx2);
1785 #endif
1786 				rxds2 = &sc->sc_rxdsoft[idx2];
1787 				if (m != NULL) {
1788 					MGET(m2, M_NOWAIT, MT_DATA);
1789 					if (m2 != NULL) {
1790 						refcount_acquire(
1791 						    &rxds2->rxds_refcount);
1792 						m2->m_len = len - m->m_len;
1793 						bus_dmamap_sync(
1794 						    sc->sc_rdmatag,
1795 						    rxds2->rxds_dmamap,
1796 						    BUS_DMASYNC_POSTREAD);
1797 						m_extadd(m2,
1798 						    (char *)rxds2->rxds_buf,
1799 						    m2->m_len, cas_free, sc,
1800 						    (void *)(uintptr_t)idx2,
1801 						    M_RDONLY, EXT_NET_DRV);
1802 						if ((m2->m_flags & M_EXT) ==
1803 						    0) {
1804 							m_freem(m2);
1805 							m2 = NULL;
1806 						}
1807 					}
1808 				}
1809 				if (m2 != NULL)
1810 					m->m_next = m2;
1811 				else if (m != NULL) {
1812 					m_freem(m);
1813 					m = NULL;
1814 				}
1815 			}
1816 			if (m != NULL) {
1817 				m->m_pkthdr.rcvif = ifp;
1818 				m->m_pkthdr.len = len;
1819 				if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
1820 				if ((if_getcapenable(ifp) & IFCAP_RXCSUM) != 0)
1821 					cas_rxcksum(m, CAS_GET(word4,
1822 					    CAS_RC4_TCP_CSUM));
1823 				/* Pass it on. */
1824 				CAS_UNLOCK(sc);
1825 				if_input(ifp, m);
1826 				CAS_LOCK(sc);
1827 			} else
1828 				if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1829 
1830 			if ((word1 & CAS_RC1_RELEASE_DATA) != 0 &&
1831 			    refcount_release(&rxds->rxds_refcount) != 0)
1832 				cas_add_rxdesc(sc, idx);
1833 			if ((word1 & CAS_RC1_SPLIT_PKT) != 0 &&
1834 			    refcount_release(&rxds2->rxds_refcount) != 0)
1835 				cas_add_rxdesc(sc, idx2);
1836 		}
1837 
1838 		skip = CAS_GET(word1, CAS_RC1_SKIP);
1839 
1840  skip:
1841 		cas_rxcompinit(&sc->sc_rxcomps[sc->sc_rxcptr]);
1842 		if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0)
1843 			break;
1844 	}
1845 	CAS_CDSYNC(sc, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1846 	CAS_WRITE_4(sc, CAS_RX_COMP_TAIL, sc->sc_rxcptr);
1847 
1848 #undef PRINTWORD
1849 #undef SKIPASSERT
1850 #undef WORDTOH
1851 
1852 #ifdef CAS_DEBUG
1853 	CTR4(KTR_CAS, "%s: done sc->sc_rxcptr %d, sc->sc_rxdptr %d, head %d",
1854 	    __func__, sc->sc_rxcptr, sc->sc_rxdptr,
1855 	    CAS_READ_4(sc, CAS_RX_COMP_HEAD));
1856 #endif
1857 }
1858 
1859 static void
1860 cas_free(struct mbuf *m)
1861 {
1862 	struct cas_rxdsoft *rxds;
1863 	struct cas_softc *sc;
1864 	u_int idx, locked;
1865 
1866 	sc = m->m_ext.ext_arg1;
1867 	idx = (uintptr_t)m->m_ext.ext_arg2;
1868 	rxds = &sc->sc_rxdsoft[idx];
1869 	if (refcount_release(&rxds->rxds_refcount) == 0)
1870 		return;
1871 
1872 	/*
1873 	 * NB: this function can be called via m_freem(9) within
1874 	 * this driver!
1875 	 */
1876 	if ((locked = CAS_LOCK_OWNED(sc)) == 0)
1877 		CAS_LOCK(sc);
1878 	cas_add_rxdesc(sc, idx);
1879 	if (locked == 0)
1880 		CAS_UNLOCK(sc);
1881 }
1882 
1883 static inline void
1884 cas_add_rxdesc(struct cas_softc *sc, u_int idx)
1885 {
1886 
1887 	CAS_LOCK_ASSERT(sc, MA_OWNED);
1888 
1889 	bus_dmamap_sync(sc->sc_rdmatag, sc->sc_rxdsoft[idx].rxds_dmamap,
1890 	    BUS_DMASYNC_PREREAD);
1891 	CAS_UPDATE_RXDESC(sc, sc->sc_rxdptr, idx);
1892 	sc->sc_rxdptr = CAS_NEXTRXDESC(sc->sc_rxdptr);
1893 
1894 	/*
1895 	 * Update the RX kick register.  This register has to point to the
1896 	 * descriptor after the last valid one (before the current batch)
1897 	 * and for optimum performance should be incremented in multiples
1898 	 * of 4 (the DMA engine fetches/updates descriptors in batches of 4).
1899 	 */
1900 	if ((sc->sc_rxdptr % 4) == 0) {
1901 		CAS_CDSYNC(sc, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1902 		CAS_WRITE_4(sc, CAS_RX_KICK,
1903 		    (sc->sc_rxdptr + CAS_NRXDESC - 4) & CAS_NRXDESC_MASK);
1904 	}
1905 }
1906 
1907 static void
1908 cas_eint(struct cas_softc *sc, u_int status)
1909 {
1910 	if_t ifp = sc->sc_ifp;
1911 
1912 	CAS_LOCK_ASSERT(sc, MA_OWNED);
1913 
1914 	if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
1915 
1916 	device_printf(sc->sc_dev, "%s: status 0x%x", __func__, status);
1917 	if ((status & CAS_INTR_PCI_ERROR_INT) != 0) {
1918 		status = CAS_READ_4(sc, CAS_ERROR_STATUS);
1919 		printf(", PCI bus error 0x%x", status);
1920 		if ((status & CAS_ERROR_OTHER) != 0) {
1921 			status = pci_read_config(sc->sc_dev, PCIR_STATUS, 2);
1922 			printf(", PCI status 0x%x", status);
1923 			pci_write_config(sc->sc_dev, PCIR_STATUS, status, 2);
1924 		}
1925 	}
1926 	printf("\n");
1927 
1928 	if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
1929 	cas_init_locked(sc);
1930 	if (!if_sendq_empty(ifp))
1931 		taskqueue_enqueue(sc->sc_tq, &sc->sc_tx_task);
1932 }
1933 
1934 static int
1935 cas_intr(void *v)
1936 {
1937 	struct cas_softc *sc = v;
1938 
1939 	if (__predict_false((CAS_READ_4(sc, CAS_STATUS_ALIAS) &
1940 	    CAS_INTR_SUMMARY) == 0))
1941 		return (FILTER_STRAY);
1942 
1943 	/* Disable interrupts. */
1944 	CAS_WRITE_4(sc, CAS_INTMASK, 0xffffffff);
1945 	taskqueue_enqueue(sc->sc_tq, &sc->sc_intr_task);
1946 
1947 	return (FILTER_HANDLED);
1948 }
1949 
1950 static void
1951 cas_intr_task(void *arg, int pending __unused)
1952 {
1953 	struct cas_softc *sc = arg;
1954 	if_t ifp = sc->sc_ifp;
1955 	uint32_t status, status2;
1956 
1957 	CAS_LOCK_ASSERT(sc, MA_NOTOWNED);
1958 
1959 	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0)
1960 		return;
1961 
1962 	status = CAS_READ_4(sc, CAS_STATUS);
1963 	if (__predict_false((status & CAS_INTR_SUMMARY) == 0))
1964 		goto done;
1965 
1966 	CAS_LOCK(sc);
1967 #ifdef CAS_DEBUG
1968 	CTR4(KTR_CAS, "%s: %s: cplt %x, status %x",
1969 	    device_get_name(sc->sc_dev), __func__,
1970 	    (status >> CAS_STATUS_TX_COMP3_SHFT), (u_int)status);
1971 
1972 	/*
1973 	 * PCS interrupts must be cleared, otherwise no traffic is passed!
1974 	 */
1975 	if ((status & CAS_INTR_PCS_INT) != 0) {
1976 		status2 =
1977 		    CAS_READ_4(sc, CAS_PCS_INTR_STATUS) |
1978 		    CAS_READ_4(sc, CAS_PCS_INTR_STATUS);
1979 		if ((status2 & CAS_PCS_INTR_LINK) != 0)
1980 			device_printf(sc->sc_dev,
1981 			    "%s: PCS link status changed\n", __func__);
1982 	}
1983 	if ((status & CAS_MAC_CTRL_STATUS) != 0) {
1984 		status2 = CAS_READ_4(sc, CAS_MAC_CTRL_STATUS);
1985 		if ((status2 & CAS_MAC_CTRL_PAUSE) != 0)
1986 			device_printf(sc->sc_dev,
1987 			    "%s: PAUSE received (PAUSE time %d slots)\n",
1988 			    __func__,
1989 			    (status2 & CAS_MAC_CTRL_STATUS_PT_MASK) >>
1990 			    CAS_MAC_CTRL_STATUS_PT_SHFT);
1991 		if ((status2 & CAS_MAC_CTRL_PAUSE) != 0)
1992 			device_printf(sc->sc_dev,
1993 			    "%s: transited to PAUSE state\n", __func__);
1994 		if ((status2 & CAS_MAC_CTRL_NON_PAUSE) != 0)
1995 			device_printf(sc->sc_dev,
1996 			    "%s: transited to non-PAUSE state\n", __func__);
1997 	}
1998 	if ((status & CAS_INTR_MIF) != 0)
1999 		device_printf(sc->sc_dev, "%s: MIF interrupt\n", __func__);
2000 #endif
2001 
2002 	if (__predict_false((status &
2003 	    (CAS_INTR_TX_TAG_ERR | CAS_INTR_RX_TAG_ERR |
2004 	    CAS_INTR_RX_LEN_MMATCH | CAS_INTR_PCI_ERROR_INT)) != 0)) {
2005 		cas_eint(sc, status);
2006 		CAS_UNLOCK(sc);
2007 		return;
2008 	}
2009 
2010 	if (__predict_false(status & CAS_INTR_TX_MAC_INT)) {
2011 		status2 = CAS_READ_4(sc, CAS_MAC_TX_STATUS);
2012 		if ((status2 &
2013 		    (CAS_MAC_TX_UNDERRUN | CAS_MAC_TX_MAX_PKT_ERR)) != 0)
2014 			if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
2015 		else if ((status2 & ~CAS_MAC_TX_FRAME_XMTD) != 0)
2016 			device_printf(sc->sc_dev,
2017 			    "MAC TX fault, status %x\n", status2);
2018 	}
2019 
2020 	if (__predict_false(status & CAS_INTR_RX_MAC_INT)) {
2021 		status2 = CAS_READ_4(sc, CAS_MAC_RX_STATUS);
2022 		if ((status2 & CAS_MAC_RX_OVERFLOW) != 0)
2023 			if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
2024 		else if ((status2 & ~CAS_MAC_RX_FRAME_RCVD) != 0)
2025 			device_printf(sc->sc_dev,
2026 			    "MAC RX fault, status %x\n", status2);
2027 	}
2028 
2029 	if ((status &
2030 	    (CAS_INTR_RX_DONE | CAS_INTR_RX_BUF_NA | CAS_INTR_RX_COMP_FULL |
2031 	    CAS_INTR_RX_BUF_AEMPTY | CAS_INTR_RX_COMP_AFULL)) != 0) {
2032 		cas_rint(sc);
2033 #ifdef CAS_DEBUG
2034 		if (__predict_false((status &
2035 		    (CAS_INTR_RX_BUF_NA | CAS_INTR_RX_COMP_FULL |
2036 		    CAS_INTR_RX_BUF_AEMPTY | CAS_INTR_RX_COMP_AFULL)) != 0))
2037 			device_printf(sc->sc_dev,
2038 			    "RX fault, status %x\n", status);
2039 #endif
2040 	}
2041 
2042 	if ((status &
2043 	    (CAS_INTR_TX_INT_ME | CAS_INTR_TX_ALL | CAS_INTR_TX_DONE)) != 0)
2044 		cas_tint(sc);
2045 
2046 	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0) {
2047 		CAS_UNLOCK(sc);
2048 		return;
2049 	} else if (!if_sendq_empty(ifp))
2050 		taskqueue_enqueue(sc->sc_tq, &sc->sc_tx_task);
2051 	CAS_UNLOCK(sc);
2052 
2053 	status = CAS_READ_4(sc, CAS_STATUS_ALIAS);
2054 	if (__predict_false((status & CAS_INTR_SUMMARY) != 0)) {
2055 		taskqueue_enqueue(sc->sc_tq, &sc->sc_intr_task);
2056 		return;
2057 	}
2058 
2059  done:
2060 	/* Re-enable interrupts. */
2061 	CAS_WRITE_4(sc, CAS_INTMASK,
2062 	    ~(CAS_INTR_TX_INT_ME | CAS_INTR_TX_TAG_ERR |
2063 	    CAS_INTR_RX_DONE | CAS_INTR_RX_BUF_NA | CAS_INTR_RX_TAG_ERR |
2064 	    CAS_INTR_RX_COMP_FULL | CAS_INTR_RX_BUF_AEMPTY |
2065 	    CAS_INTR_RX_COMP_AFULL | CAS_INTR_RX_LEN_MMATCH |
2066 	    CAS_INTR_PCI_ERROR_INT
2067 #ifdef CAS_DEBUG
2068 	    | CAS_INTR_PCS_INT | CAS_INTR_MIF
2069 #endif
2070 	));
2071 }
2072 
2073 static void
2074 cas_watchdog(struct cas_softc *sc)
2075 {
2076 	if_t ifp = sc->sc_ifp;
2077 
2078 	CAS_LOCK_ASSERT(sc, MA_OWNED);
2079 
2080 #ifdef CAS_DEBUG
2081 	CTR4(KTR_CAS,
2082 	    "%s: CAS_RX_CONF %x CAS_MAC_RX_STATUS %x CAS_MAC_RX_CONF %x",
2083 	    __func__, CAS_READ_4(sc, CAS_RX_CONF),
2084 	    CAS_READ_4(sc, CAS_MAC_RX_STATUS),
2085 	    CAS_READ_4(sc, CAS_MAC_RX_CONF));
2086 	CTR4(KTR_CAS,
2087 	    "%s: CAS_TX_CONF %x CAS_MAC_TX_STATUS %x CAS_MAC_TX_CONF %x",
2088 	    __func__, CAS_READ_4(sc, CAS_TX_CONF),
2089 	    CAS_READ_4(sc, CAS_MAC_TX_STATUS),
2090 	    CAS_READ_4(sc, CAS_MAC_TX_CONF));
2091 #endif
2092 
2093 	if (sc->sc_wdog_timer == 0 || --sc->sc_wdog_timer != 0)
2094 		return;
2095 
2096 	if ((sc->sc_flags & CAS_LINK) != 0)
2097 		device_printf(sc->sc_dev, "device timeout\n");
2098 	else if (bootverbose)
2099 		device_printf(sc->sc_dev, "device timeout (no link)\n");
2100 	if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
2101 
2102 	/* Try to get more packets going. */
2103 	if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
2104 	cas_init_locked(sc);
2105 	if (!if_sendq_empty(ifp))
2106 		taskqueue_enqueue(sc->sc_tq, &sc->sc_tx_task);
2107 }
2108 
2109 static void
2110 cas_mifinit(struct cas_softc *sc)
2111 {
2112 
2113 	/* Configure the MIF in frame mode. */
2114 	CAS_WRITE_4(sc, CAS_MIF_CONF,
2115 	    CAS_READ_4(sc, CAS_MIF_CONF) & ~CAS_MIF_CONF_BB_MODE);
2116 	CAS_BARRIER(sc, CAS_MIF_CONF, 4,
2117 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2118 }
2119 
2120 /*
2121  * MII interface
2122  *
2123  * The MII interface supports at least three different operating modes:
2124  *
2125  * Bitbang mode is implemented using data, clock and output enable registers.
2126  *
2127  * Frame mode is implemented by loading a complete frame into the frame
2128  * register and polling the valid bit for completion.
2129  *
2130  * Polling mode uses the frame register but completion is indicated by
2131  * an interrupt.
2132  *
2133  */
2134 static int
2135 cas_mii_readreg(device_t dev, int phy, int reg)
2136 {
2137 	struct cas_softc *sc;
2138 	int n;
2139 	uint32_t v;
2140 
2141 #ifdef CAS_DEBUG_PHY
2142 	printf("%s: phy %d reg %d\n", __func__, phy, reg);
2143 #endif
2144 
2145 	sc = device_get_softc(dev);
2146 	if ((sc->sc_flags & CAS_SERDES) != 0) {
2147 		switch (reg) {
2148 		case MII_BMCR:
2149 			reg = CAS_PCS_CTRL;
2150 			break;
2151 		case MII_BMSR:
2152 			reg = CAS_PCS_STATUS;
2153 			break;
2154 		case MII_PHYIDR1:
2155 		case MII_PHYIDR2:
2156 			return (0);
2157 		case MII_ANAR:
2158 			reg = CAS_PCS_ANAR;
2159 			break;
2160 		case MII_ANLPAR:
2161 			reg = CAS_PCS_ANLPAR;
2162 			break;
2163 		case MII_EXTSR:
2164 			return (EXTSR_1000XFDX | EXTSR_1000XHDX);
2165 		default:
2166 			device_printf(sc->sc_dev,
2167 			    "%s: unhandled register %d\n", __func__, reg);
2168 			return (0);
2169 		}
2170 		return (CAS_READ_4(sc, reg));
2171 	}
2172 
2173 	/* Construct the frame command. */
2174 	v = CAS_MIF_FRAME_READ |
2175 	    (phy << CAS_MIF_FRAME_PHY_SHFT) |
2176 	    (reg << CAS_MIF_FRAME_REG_SHFT);
2177 
2178 	CAS_WRITE_4(sc, CAS_MIF_FRAME, v);
2179 	CAS_BARRIER(sc, CAS_MIF_FRAME, 4,
2180 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2181 	for (n = 0; n < 100; n++) {
2182 		DELAY(1);
2183 		v = CAS_READ_4(sc, CAS_MIF_FRAME);
2184 		if (v & CAS_MIF_FRAME_TA_LSB)
2185 			return (v & CAS_MIF_FRAME_DATA);
2186 	}
2187 
2188 	device_printf(sc->sc_dev, "%s: timed out\n", __func__);
2189 	return (0);
2190 }
2191 
2192 static int
2193 cas_mii_writereg(device_t dev, int phy, int reg, int val)
2194 {
2195 	struct cas_softc *sc;
2196 	int n;
2197 	uint32_t v;
2198 
2199 #ifdef CAS_DEBUG_PHY
2200 	printf("%s: phy %d reg %d val %x\n", phy, reg, val, __func__);
2201 #endif
2202 
2203 	sc = device_get_softc(dev);
2204 	if ((sc->sc_flags & CAS_SERDES) != 0) {
2205 		switch (reg) {
2206 		case MII_BMSR:
2207 			reg = CAS_PCS_STATUS;
2208 			break;
2209 		case MII_BMCR:
2210 			reg = CAS_PCS_CTRL;
2211 			if ((val & CAS_PCS_CTRL_RESET) == 0)
2212 				break;
2213 			CAS_WRITE_4(sc, CAS_PCS_CTRL, val);
2214 			CAS_BARRIER(sc, CAS_PCS_CTRL, 4,
2215 			    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2216 			if (!cas_bitwait(sc, CAS_PCS_CTRL,
2217 			    CAS_PCS_CTRL_RESET, 0))
2218 				device_printf(sc->sc_dev,
2219 				    "cannot reset PCS\n");
2220 			/* FALLTHROUGH */
2221 		case MII_ANAR:
2222 			CAS_WRITE_4(sc, CAS_PCS_CONF, 0);
2223 			CAS_BARRIER(sc, CAS_PCS_CONF, 4,
2224 			    BUS_SPACE_BARRIER_WRITE);
2225 			CAS_WRITE_4(sc, CAS_PCS_ANAR, val);
2226 			CAS_BARRIER(sc, CAS_PCS_ANAR, 4,
2227 			    BUS_SPACE_BARRIER_WRITE);
2228 			CAS_WRITE_4(sc, CAS_PCS_SERDES_CTRL,
2229 			    CAS_PCS_SERDES_CTRL_ESD);
2230 			CAS_BARRIER(sc, CAS_PCS_CONF, 4,
2231 			    BUS_SPACE_BARRIER_WRITE);
2232 			CAS_WRITE_4(sc, CAS_PCS_CONF,
2233 			    CAS_PCS_CONF_EN);
2234 			CAS_BARRIER(sc, CAS_PCS_CONF, 4,
2235 			    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2236 			return (0);
2237 		case MII_ANLPAR:
2238 			reg = CAS_PCS_ANLPAR;
2239 			break;
2240 		default:
2241 			device_printf(sc->sc_dev,
2242 			    "%s: unhandled register %d\n", __func__, reg);
2243 			return (0);
2244 		}
2245 		CAS_WRITE_4(sc, reg, val);
2246 		CAS_BARRIER(sc, reg, 4,
2247 		    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2248 		return (0);
2249 	}
2250 
2251 	/* Construct the frame command. */
2252 	v = CAS_MIF_FRAME_WRITE |
2253 	    (phy << CAS_MIF_FRAME_PHY_SHFT) |
2254 	    (reg << CAS_MIF_FRAME_REG_SHFT) |
2255 	    (val & CAS_MIF_FRAME_DATA);
2256 
2257 	CAS_WRITE_4(sc, CAS_MIF_FRAME, v);
2258 	CAS_BARRIER(sc, CAS_MIF_FRAME, 4,
2259 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2260 	for (n = 0; n < 100; n++) {
2261 		DELAY(1);
2262 		v = CAS_READ_4(sc, CAS_MIF_FRAME);
2263 		if (v & CAS_MIF_FRAME_TA_LSB)
2264 			return (1);
2265 	}
2266 
2267 	device_printf(sc->sc_dev, "%s: timed out\n", __func__);
2268 	return (0);
2269 }
2270 
2271 static void
2272 cas_mii_statchg(device_t dev)
2273 {
2274 	struct cas_softc *sc;
2275 	if_t ifp;
2276 	int gigabit;
2277 	uint32_t rxcfg, txcfg, v;
2278 
2279 	sc = device_get_softc(dev);
2280 	ifp = sc->sc_ifp;
2281 
2282 	CAS_LOCK_ASSERT(sc, MA_OWNED);
2283 
2284 #ifdef CAS_DEBUG
2285 	if ((if_getflags(ifp) & IFF_DEBUG) != 0)
2286 		device_printf(sc->sc_dev, "%s: status changen", __func__);
2287 #endif
2288 
2289 	if ((sc->sc_mii->mii_media_status & IFM_ACTIVE) != 0 &&
2290 	    IFM_SUBTYPE(sc->sc_mii->mii_media_active) != IFM_NONE)
2291 		sc->sc_flags |= CAS_LINK;
2292 	else
2293 		sc->sc_flags &= ~CAS_LINK;
2294 
2295 	switch (IFM_SUBTYPE(sc->sc_mii->mii_media_active)) {
2296 	case IFM_1000_SX:
2297 	case IFM_1000_LX:
2298 	case IFM_1000_CX:
2299 	case IFM_1000_T:
2300 		gigabit = 1;
2301 		break;
2302 	default:
2303 		gigabit = 0;
2304 	}
2305 
2306 	/*
2307 	 * The configuration done here corresponds to the steps F) and
2308 	 * G) and as far as enabling of RX and TX MAC goes also step H)
2309 	 * of the initialization sequence outlined in section 11.2.1 of
2310 	 * the Cassini+ ASIC Specification.
2311 	 */
2312 
2313 	rxcfg = sc->sc_mac_rxcfg;
2314 	rxcfg &= ~CAS_MAC_RX_CONF_CARR;
2315 	txcfg = CAS_MAC_TX_CONF_EN_IPG0 | CAS_MAC_TX_CONF_NGU |
2316 	    CAS_MAC_TX_CONF_NGUL;
2317 	if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) & IFM_FDX) != 0)
2318 		txcfg |= CAS_MAC_TX_CONF_ICARR | CAS_MAC_TX_CONF_ICOLLIS;
2319 	else if (gigabit != 0) {
2320 		rxcfg |= CAS_MAC_RX_CONF_CARR;
2321 		txcfg |= CAS_MAC_TX_CONF_CARR;
2322 	}
2323 	(void)cas_disable_tx(sc);
2324 	CAS_WRITE_4(sc, CAS_MAC_TX_CONF, txcfg);
2325 	(void)cas_disable_rx(sc);
2326 	CAS_WRITE_4(sc, CAS_MAC_RX_CONF, rxcfg);
2327 
2328 	v = CAS_READ_4(sc, CAS_MAC_CTRL_CONF) &
2329 	    ~(CAS_MAC_CTRL_CONF_TXP | CAS_MAC_CTRL_CONF_RXP);
2330 	if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) &
2331 	    IFM_ETH_RXPAUSE) != 0)
2332 		v |= CAS_MAC_CTRL_CONF_RXP;
2333 	if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) &
2334 	    IFM_ETH_TXPAUSE) != 0)
2335 		v |= CAS_MAC_CTRL_CONF_TXP;
2336 	CAS_WRITE_4(sc, CAS_MAC_CTRL_CONF, v);
2337 
2338 	/*
2339 	 * All supported chips have a bug causing incorrect checksum
2340 	 * to be calculated when letting them strip the FCS in half-
2341 	 * duplex mode.  In theory we could disable FCS stripping and
2342 	 * manually adjust the checksum accordingly.  It seems to make
2343 	 * more sense to optimze for the common case and just disable
2344 	 * hardware checksumming in half-duplex mode though.
2345 	 */
2346 	if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) & IFM_FDX) == 0) {
2347 		if_setcapenablebit(ifp, 0, IFCAP_HWCSUM);
2348 		if_sethwassist(ifp, 0);
2349 	} else if ((sc->sc_flags & CAS_NO_CSUM) == 0) {
2350 		if_setcapenable(ifp, if_getcapabilities(ifp));
2351 		if_sethwassist(ifp, CAS_CSUM_FEATURES);
2352 	}
2353 
2354 	if (sc->sc_variant == CAS_SATURN) {
2355 		if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) & IFM_FDX) == 0)
2356 			/* silicon bug workaround */
2357 			CAS_WRITE_4(sc, CAS_MAC_PREAMBLE_LEN, 0x41);
2358 		else
2359 			CAS_WRITE_4(sc, CAS_MAC_PREAMBLE_LEN, 0x7);
2360 	}
2361 
2362 	if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) & IFM_FDX) == 0 &&
2363 	    gigabit != 0)
2364 		CAS_WRITE_4(sc, CAS_MAC_SLOT_TIME,
2365 		    CAS_MAC_SLOT_TIME_CARR);
2366 	else
2367 		CAS_WRITE_4(sc, CAS_MAC_SLOT_TIME,
2368 		    CAS_MAC_SLOT_TIME_NORM);
2369 
2370 	/* XIF Configuration */
2371 	v = CAS_MAC_XIF_CONF_TX_OE | CAS_MAC_XIF_CONF_LNKLED;
2372 	if ((sc->sc_flags & CAS_SERDES) == 0) {
2373 		if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) & IFM_FDX) == 0)
2374 			v |= CAS_MAC_XIF_CONF_NOECHO;
2375 		v |= CAS_MAC_XIF_CONF_BUF_OE;
2376 	}
2377 	if (gigabit != 0)
2378 		v |= CAS_MAC_XIF_CONF_GMII;
2379 	if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) & IFM_FDX) != 0)
2380 		v |= CAS_MAC_XIF_CONF_FDXLED;
2381 	CAS_WRITE_4(sc, CAS_MAC_XIF_CONF, v);
2382 
2383 	sc->sc_mac_rxcfg = rxcfg;
2384 	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0 &&
2385 	    (sc->sc_flags & CAS_LINK) != 0) {
2386 		CAS_WRITE_4(sc, CAS_MAC_TX_CONF,
2387 		    txcfg | CAS_MAC_TX_CONF_EN);
2388 		CAS_WRITE_4(sc, CAS_MAC_RX_CONF,
2389 		    rxcfg | CAS_MAC_RX_CONF_EN);
2390 	}
2391 }
2392 
2393 static int
2394 cas_mediachange(if_t ifp)
2395 {
2396 	struct cas_softc *sc = if_getsoftc(ifp);
2397 	int error;
2398 
2399 	/* XXX add support for serial media. */
2400 
2401 	CAS_LOCK(sc);
2402 	error = mii_mediachg(sc->sc_mii);
2403 	CAS_UNLOCK(sc);
2404 	return (error);
2405 }
2406 
2407 static void
2408 cas_mediastatus(if_t ifp, struct ifmediareq *ifmr)
2409 {
2410 	struct cas_softc *sc = if_getsoftc(ifp);
2411 
2412 	CAS_LOCK(sc);
2413 	if ((if_getflags(ifp) & IFF_UP) == 0) {
2414 		CAS_UNLOCK(sc);
2415 		return;
2416 	}
2417 
2418 	mii_pollstat(sc->sc_mii);
2419 	ifmr->ifm_active = sc->sc_mii->mii_media_active;
2420 	ifmr->ifm_status = sc->sc_mii->mii_media_status;
2421 	CAS_UNLOCK(sc);
2422 }
2423 
2424 static int
2425 cas_ioctl(if_t ifp, u_long cmd, caddr_t data)
2426 {
2427 	struct cas_softc *sc = if_getsoftc(ifp);
2428 	struct ifreq *ifr = (struct ifreq *)data;
2429 	int error;
2430 
2431 	error = 0;
2432 	switch (cmd) {
2433 	case SIOCSIFFLAGS:
2434 		CAS_LOCK(sc);
2435 		if ((if_getflags(ifp) & IFF_UP) != 0) {
2436 			if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0 &&
2437 			    ((if_getflags(ifp) ^ sc->sc_ifflags) &
2438 			    (IFF_ALLMULTI | IFF_PROMISC)) != 0)
2439 				cas_setladrf(sc);
2440 			else
2441 				cas_init_locked(sc);
2442 		} else if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0)
2443 			cas_stop(ifp);
2444 		sc->sc_ifflags = if_getflags(ifp);
2445 		CAS_UNLOCK(sc);
2446 		break;
2447 	case SIOCSIFCAP:
2448 		CAS_LOCK(sc);
2449 		if ((sc->sc_flags & CAS_NO_CSUM) != 0) {
2450 			error = EINVAL;
2451 			CAS_UNLOCK(sc);
2452 			break;
2453 		}
2454 		if_setcapenable(ifp, ifr->ifr_reqcap);
2455 		if ((if_getcapenable(ifp) & IFCAP_TXCSUM) != 0)
2456 			if_sethwassist(ifp, CAS_CSUM_FEATURES);
2457 		else
2458 			if_sethwassist(ifp, 0);
2459 		CAS_UNLOCK(sc);
2460 		break;
2461 	case SIOCADDMULTI:
2462 	case SIOCDELMULTI:
2463 		CAS_LOCK(sc);
2464 		if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0)
2465 			cas_setladrf(sc);
2466 		CAS_UNLOCK(sc);
2467 		break;
2468 	case SIOCSIFMTU:
2469 		if ((ifr->ifr_mtu < ETHERMIN) ||
2470 		    (ifr->ifr_mtu > ETHERMTU_JUMBO))
2471 			error = EINVAL;
2472 		else
2473 			if_setmtu(ifp, ifr->ifr_mtu);
2474 		break;
2475 	case SIOCGIFMEDIA:
2476 	case SIOCSIFMEDIA:
2477 		error = ifmedia_ioctl(ifp, ifr, &sc->sc_mii->mii_media, cmd);
2478 		break;
2479 	default:
2480 		error = ether_ioctl(ifp, cmd, data);
2481 		break;
2482 	}
2483 
2484 	return (error);
2485 }
2486 
2487 static u_int
2488 cas_hash_maddr(void *arg, struct sockaddr_dl *sdl, u_int cnt)
2489 {
2490 	uint32_t crc, *hash = arg;
2491 
2492 	crc = ether_crc32_le(LLADDR(sdl), ETHER_ADDR_LEN);
2493 	/* We just want the 8 most significant bits. */
2494 	crc >>= 24;
2495 	/* Set the corresponding bit in the filter. */
2496 	hash[crc >> 4] |= 1 << (15 - (crc & 15));
2497 
2498 	return (1);
2499 }
2500 
2501 static void
2502 cas_setladrf(struct cas_softc *sc)
2503 {
2504 	if_t ifp = sc->sc_ifp;
2505 	int i;
2506 	uint32_t hash[16];
2507 	uint32_t v;
2508 
2509 	CAS_LOCK_ASSERT(sc, MA_OWNED);
2510 
2511 	/*
2512 	 * Turn off the RX MAC and the hash filter as required by the Sun
2513 	 * Cassini programming restrictions.
2514 	 */
2515 	v = sc->sc_mac_rxcfg & ~(CAS_MAC_RX_CONF_HFILTER |
2516 	    CAS_MAC_RX_CONF_EN);
2517 	CAS_WRITE_4(sc, CAS_MAC_RX_CONF, v);
2518 	CAS_BARRIER(sc, CAS_MAC_RX_CONF, 4,
2519 	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2520 	if (!cas_bitwait(sc, CAS_MAC_RX_CONF, CAS_MAC_RX_CONF_HFILTER |
2521 	    CAS_MAC_RX_CONF_EN, 0))
2522 		device_printf(sc->sc_dev,
2523 		    "cannot disable RX MAC or hash filter\n");
2524 
2525 	v &= ~(CAS_MAC_RX_CONF_PROMISC | CAS_MAC_RX_CONF_PGRP);
2526 	if ((if_getflags(ifp) & IFF_PROMISC) != 0) {
2527 		v |= CAS_MAC_RX_CONF_PROMISC;
2528 		goto chipit;
2529 	}
2530 	if ((if_getflags(ifp) & IFF_ALLMULTI) != 0) {
2531 		v |= CAS_MAC_RX_CONF_PGRP;
2532 		goto chipit;
2533 	}
2534 
2535 	/*
2536 	 * Set up multicast address filter by passing all multicast
2537 	 * addresses through a crc generator, and then using the high
2538 	 * order 8 bits as an index into the 256 bit logical address
2539 	 * filter.  The high order 4 bits selects the word, while the
2540 	 * other 4 bits select the bit within the word (where bit 0
2541 	 * is the MSB).
2542 	 */
2543 
2544 	memset(hash, 0, sizeof(hash));
2545 	if_foreach_llmaddr(ifp, cas_hash_maddr, &hash);
2546 
2547 	v |= CAS_MAC_RX_CONF_HFILTER;
2548 
2549 	/* Now load the hash table into the chip (if we are using it). */
2550 	for (i = 0; i < 16; i++)
2551 		CAS_WRITE_4(sc,
2552 		    CAS_MAC_HASH0 + i * (CAS_MAC_HASH1 - CAS_MAC_HASH0),
2553 		    hash[i]);
2554 
2555  chipit:
2556 	sc->sc_mac_rxcfg = v;
2557 	CAS_WRITE_4(sc, CAS_MAC_RX_CONF, v | CAS_MAC_RX_CONF_EN);
2558 }
2559 
2560 static int	cas_pci_attach(device_t dev);
2561 static int	cas_pci_detach(device_t dev);
2562 static int	cas_pci_probe(device_t dev);
2563 static int	cas_pci_resume(device_t dev);
2564 static int	cas_pci_suspend(device_t dev);
2565 
2566 static device_method_t cas_pci_methods[] = {
2567 	/* Device interface */
2568 	DEVMETHOD(device_probe,		cas_pci_probe),
2569 	DEVMETHOD(device_attach,	cas_pci_attach),
2570 	DEVMETHOD(device_detach,	cas_pci_detach),
2571 	DEVMETHOD(device_suspend,	cas_pci_suspend),
2572 	DEVMETHOD(device_resume,	cas_pci_resume),
2573 	/* Use the suspend handler here, it is all that is required. */
2574 	DEVMETHOD(device_shutdown,	cas_pci_suspend),
2575 
2576 	/* MII interface */
2577 	DEVMETHOD(miibus_readreg,	cas_mii_readreg),
2578 	DEVMETHOD(miibus_writereg,	cas_mii_writereg),
2579 	DEVMETHOD(miibus_statchg,	cas_mii_statchg),
2580 
2581 	DEVMETHOD_END
2582 };
2583 
2584 static driver_t cas_pci_driver = {
2585 	"cas",
2586 	cas_pci_methods,
2587 	sizeof(struct cas_softc)
2588 };
2589 
2590 static const struct cas_pci_dev {
2591 	uint32_t	cpd_devid;
2592 	uint8_t		cpd_revid;
2593 	int		cpd_variant;
2594 	const char	*cpd_desc;
2595 } cas_pci_devlist[] = {
2596 	{ 0x0035100b, 0x0, CAS_SATURN, "NS DP83065 Saturn Gigabit Ethernet" },
2597 	{ 0xabba108e, 0x10, CAS_CASPLUS, "Sun Cassini+ Gigabit Ethernet" },
2598 	{ 0xabba108e, 0x0, CAS_CAS, "Sun Cassini Gigabit Ethernet" },
2599 	{ 0, 0, 0, NULL }
2600 };
2601 
2602 DRIVER_MODULE(cas, pci, cas_pci_driver, 0, 0);
2603 MODULE_PNP_INFO("W32:vendor/device", pci, cas, cas_pci_devlist,
2604     nitems(cas_pci_devlist) - 1);
2605 DRIVER_MODULE(miibus, cas, miibus_driver, 0, 0);
2606 MODULE_DEPEND(cas, pci, 1, 1, 1);
2607 
2608 static int
2609 cas_pci_probe(device_t dev)
2610 {
2611 	int i;
2612 
2613 	for (i = 0; cas_pci_devlist[i].cpd_desc != NULL; i++) {
2614 		if (pci_get_devid(dev) == cas_pci_devlist[i].cpd_devid &&
2615 		    pci_get_revid(dev) >= cas_pci_devlist[i].cpd_revid) {
2616 			device_set_desc(dev, cas_pci_devlist[i].cpd_desc);
2617 			return (BUS_PROBE_DEFAULT);
2618 		}
2619 	}
2620 
2621 	return (ENXIO);
2622 }
2623 
2624 static struct resource_spec cas_pci_res_spec[] = {
2625 	{ SYS_RES_IRQ, 0, RF_SHAREABLE | RF_ACTIVE },	/* CAS_RES_INTR */
2626 	{ SYS_RES_MEMORY, PCIR_BAR(0), RF_ACTIVE },	/* CAS_RES_MEM */
2627 	{ -1, 0 }
2628 };
2629 
2630 #define	CAS_LOCAL_MAC_ADDRESS	"local-mac-address"
2631 #define	CAS_PHY_INTERFACE	"phy-interface"
2632 #define	CAS_PHY_TYPE		"phy-type"
2633 #define	CAS_PHY_TYPE_PCS	"pcs"
2634 
2635 static int
2636 cas_pci_attach(device_t dev)
2637 {
2638 	char buf[sizeof(CAS_LOCAL_MAC_ADDRESS)];
2639 	struct cas_softc *sc;
2640 	int i;
2641 #if !defined(__powerpc__)
2642 	u_char enaddr[4][ETHER_ADDR_LEN];
2643 	u_int j, k, lma, pcs[4], phy;
2644 #endif
2645 
2646 	sc = device_get_softc(dev);
2647 	sc->sc_variant = CAS_UNKNOWN;
2648 	for (i = 0; cas_pci_devlist[i].cpd_desc != NULL; i++) {
2649 		if (pci_get_devid(dev) == cas_pci_devlist[i].cpd_devid &&
2650 		    pci_get_revid(dev) >= cas_pci_devlist[i].cpd_revid) {
2651 			sc->sc_variant = cas_pci_devlist[i].cpd_variant;
2652 			break;
2653 		}
2654 	}
2655 	if (sc->sc_variant == CAS_UNKNOWN) {
2656 		device_printf(dev, "unknown adaptor\n");
2657 		return (ENXIO);
2658 	}
2659 
2660 	/* PCI configuration */
2661 	pci_write_config(dev, PCIR_COMMAND,
2662 	    pci_read_config(dev, PCIR_COMMAND, 2) | PCIM_CMD_BUSMASTEREN |
2663 	    PCIM_CMD_MWRICEN | PCIM_CMD_PERRESPEN | PCIM_CMD_SERRESPEN, 2);
2664 
2665 	sc->sc_dev = dev;
2666 	if (sc->sc_variant == CAS_CAS && pci_get_devid(dev) < 0x02)
2667 		/* Hardware checksumming may hang TX. */
2668 		sc->sc_flags |= CAS_NO_CSUM;
2669 	if (sc->sc_variant == CAS_CASPLUS || sc->sc_variant == CAS_SATURN)
2670 		sc->sc_flags |= CAS_REG_PLUS;
2671 	if (sc->sc_variant == CAS_CAS ||
2672 	    (sc->sc_variant == CAS_CASPLUS && pci_get_revid(dev) < 0x11))
2673 		sc->sc_flags |= CAS_TABORT;
2674 	if (bootverbose)
2675 		device_printf(dev, "flags=0x%x\n", sc->sc_flags);
2676 
2677 	if (bus_alloc_resources(dev, cas_pci_res_spec, sc->sc_res)) {
2678 		device_printf(dev, "failed to allocate resources\n");
2679 		bus_release_resources(dev, cas_pci_res_spec, sc->sc_res);
2680 		return (ENXIO);
2681 	}
2682 
2683 	CAS_LOCK_INIT(sc, device_get_nameunit(dev));
2684 
2685 #if defined(__powerpc__)
2686 	OF_getetheraddr(dev, sc->sc_enaddr);
2687 	if (OF_getprop(ofw_bus_get_node(dev), CAS_PHY_INTERFACE, buf,
2688 	    sizeof(buf)) > 0 || OF_getprop(ofw_bus_get_node(dev),
2689 	    CAS_PHY_TYPE, buf, sizeof(buf)) > 0) {
2690 		buf[sizeof(buf) - 1] = '\0';
2691 		if (strcmp(buf, CAS_PHY_TYPE_PCS) == 0)
2692 			sc->sc_flags |= CAS_SERDES;
2693 	}
2694 #else
2695 	/*
2696 	 * Dig out VPD (vital product data) and read the MAC address as well
2697 	 * as the PHY type.  The VPD resides in the PCI Expansion ROM (PCI
2698 	 * FCode) and can't be accessed via the PCI capability pointer.
2699 	 * SUNW,pci-ce and SUNW,pci-qge use the Enhanced VPD format described
2700 	 * in the free US Patent 7149820.
2701 	 */
2702 
2703 #define	PCI_ROMHDR_SIZE			0x1c
2704 #define	PCI_ROMHDR_SIG			0x00
2705 #define	PCI_ROMHDR_SIG_MAGIC		0xaa55		/* little endian */
2706 #define	PCI_ROMHDR_PTR_DATA		0x18
2707 #define	PCI_ROM_SIZE			0x18
2708 #define	PCI_ROM_SIG			0x00
2709 #define	PCI_ROM_SIG_MAGIC		0x52494350	/* "PCIR", endian */
2710 							/* reversed */
2711 #define	PCI_ROM_VENDOR			0x04
2712 #define	PCI_ROM_DEVICE			0x06
2713 #define	PCI_ROM_PTR_VPD			0x08
2714 #define	PCI_VPDRES_BYTE0		0x00
2715 #define	PCI_VPDRES_ISLARGE(x)		((x) & 0x80)
2716 #define	PCI_VPDRES_LARGE_NAME(x)	((x) & 0x7f)
2717 #define	PCI_VPDRES_LARGE_LEN_LSB	0x01
2718 #define	PCI_VPDRES_LARGE_LEN_MSB	0x02
2719 #define	PCI_VPDRES_LARGE_SIZE		0x03
2720 #define	PCI_VPDRES_TYPE_ID_STRING	0x02		/* large */
2721 #define	PCI_VPDRES_TYPE_VPD		0x10		/* large */
2722 #define	PCI_VPD_KEY0			0x00
2723 #define	PCI_VPD_KEY1			0x01
2724 #define	PCI_VPD_LEN			0x02
2725 #define	PCI_VPD_SIZE			0x03
2726 
2727 #define	CAS_ROM_READ_1(sc, offs)					\
2728 	CAS_READ_1((sc), CAS_PCI_ROM_OFFSET + (offs))
2729 #define	CAS_ROM_READ_2(sc, offs)					\
2730 	CAS_READ_2((sc), CAS_PCI_ROM_OFFSET + (offs))
2731 #define	CAS_ROM_READ_4(sc, offs)					\
2732 	CAS_READ_4((sc), CAS_PCI_ROM_OFFSET + (offs))
2733 
2734 	lma = phy = 0;
2735 	memset(enaddr, 0, sizeof(enaddr));
2736 	memset(pcs, 0, sizeof(pcs));
2737 
2738 	/* Enable PCI Expansion ROM access. */
2739 	CAS_WRITE_4(sc, CAS_BIM_LDEV_OEN,
2740 	    CAS_BIM_LDEV_OEN_PAD | CAS_BIM_LDEV_OEN_PROM);
2741 
2742 	/* Read PCI Expansion ROM header. */
2743 	if (CAS_ROM_READ_2(sc, PCI_ROMHDR_SIG) != PCI_ROMHDR_SIG_MAGIC ||
2744 	    (i = CAS_ROM_READ_2(sc, PCI_ROMHDR_PTR_DATA)) <
2745 	    PCI_ROMHDR_SIZE) {
2746 		device_printf(dev, "unexpected PCI Expansion ROM header\n");
2747 		goto fail_prom;
2748 	}
2749 
2750 	/* Read PCI Expansion ROM data. */
2751 	if (CAS_ROM_READ_4(sc, i + PCI_ROM_SIG) != PCI_ROM_SIG_MAGIC ||
2752 	    CAS_ROM_READ_2(sc, i + PCI_ROM_VENDOR) != pci_get_vendor(dev) ||
2753 	    CAS_ROM_READ_2(sc, i + PCI_ROM_DEVICE) != pci_get_device(dev) ||
2754 	    (j = CAS_ROM_READ_2(sc, i + PCI_ROM_PTR_VPD)) <
2755 	    i + PCI_ROM_SIZE) {
2756 		device_printf(dev, "unexpected PCI Expansion ROM data\n");
2757 		goto fail_prom;
2758 	}
2759 
2760 	/* Read PCI VPD. */
2761  next:
2762 	if (PCI_VPDRES_ISLARGE(CAS_ROM_READ_1(sc,
2763 	    j + PCI_VPDRES_BYTE0)) == 0) {
2764 		device_printf(dev, "no large PCI VPD\n");
2765 		goto fail_prom;
2766 	}
2767 
2768 	i = (CAS_ROM_READ_1(sc, j + PCI_VPDRES_LARGE_LEN_MSB) << 8) |
2769 	    CAS_ROM_READ_1(sc, j + PCI_VPDRES_LARGE_LEN_LSB);
2770 	switch (PCI_VPDRES_LARGE_NAME(CAS_ROM_READ_1(sc,
2771 	    j + PCI_VPDRES_BYTE0))) {
2772 	case PCI_VPDRES_TYPE_ID_STRING:
2773 		/* Skip identifier string. */
2774 		j += PCI_VPDRES_LARGE_SIZE + i;
2775 		goto next;
2776 	case PCI_VPDRES_TYPE_VPD:
2777 		for (j += PCI_VPDRES_LARGE_SIZE; i > 0;
2778 		    i -= PCI_VPD_SIZE + CAS_ROM_READ_1(sc, j + PCI_VPD_LEN),
2779 		    j += PCI_VPD_SIZE + CAS_ROM_READ_1(sc, j + PCI_VPD_LEN)) {
2780 			if (CAS_ROM_READ_1(sc, j + PCI_VPD_KEY0) != 'Z')
2781 				/* no Enhanced VPD */
2782 				continue;
2783 			if (CAS_ROM_READ_1(sc, j + PCI_VPD_SIZE) != 'I')
2784 				/* no instance property */
2785 				continue;
2786 			if (CAS_ROM_READ_1(sc, j + PCI_VPD_SIZE + 3) == 'B') {
2787 				/* byte array */
2788 				if (CAS_ROM_READ_1(sc,
2789 				    j + PCI_VPD_SIZE + 4) != ETHER_ADDR_LEN)
2790 					continue;
2791 				bus_read_region_1(sc->sc_res[CAS_RES_MEM],
2792 				    CAS_PCI_ROM_OFFSET + j + PCI_VPD_SIZE + 5,
2793 				    buf, sizeof(buf));
2794 				buf[sizeof(buf) - 1] = '\0';
2795 				if (strcmp(buf, CAS_LOCAL_MAC_ADDRESS) != 0)
2796 					continue;
2797 				bus_read_region_1(sc->sc_res[CAS_RES_MEM],
2798 				    CAS_PCI_ROM_OFFSET + j + PCI_VPD_SIZE +
2799 				    5 + sizeof(CAS_LOCAL_MAC_ADDRESS),
2800 				    enaddr[lma], sizeof(enaddr[lma]));
2801 				lma++;
2802 				if (lma == 4 && phy == 4)
2803 					break;
2804 			} else if (CAS_ROM_READ_1(sc, j + PCI_VPD_SIZE + 3) ==
2805 			   'S') {
2806 				/* string */
2807 				if (CAS_ROM_READ_1(sc,
2808 				    j + PCI_VPD_SIZE + 4) !=
2809 				    sizeof(CAS_PHY_TYPE_PCS))
2810 					continue;
2811 				bus_read_region_1(sc->sc_res[CAS_RES_MEM],
2812 				    CAS_PCI_ROM_OFFSET + j + PCI_VPD_SIZE + 5,
2813 				    buf, sizeof(buf));
2814 				buf[sizeof(buf) - 1] = '\0';
2815 				if (strcmp(buf, CAS_PHY_INTERFACE) == 0)
2816 					k = sizeof(CAS_PHY_INTERFACE);
2817 				else if (strcmp(buf, CAS_PHY_TYPE) == 0)
2818 					k = sizeof(CAS_PHY_TYPE);
2819 				else
2820 					continue;
2821 				bus_read_region_1(sc->sc_res[CAS_RES_MEM],
2822 				    CAS_PCI_ROM_OFFSET + j + PCI_VPD_SIZE +
2823 				    5 + k, buf, sizeof(buf));
2824 				buf[sizeof(buf) - 1] = '\0';
2825 				if (strcmp(buf, CAS_PHY_TYPE_PCS) == 0)
2826 					pcs[phy] = 1;
2827 				phy++;
2828 				if (lma == 4 && phy == 4)
2829 					break;
2830 			}
2831 		}
2832 		break;
2833 	default:
2834 		device_printf(dev, "unexpected PCI VPD\n");
2835 		goto fail_prom;
2836 	}
2837 
2838  fail_prom:
2839 	CAS_WRITE_4(sc, CAS_BIM_LDEV_OEN, 0);
2840 
2841 	if (lma == 0) {
2842 		device_printf(dev, "could not determine Ethernet address\n");
2843 		goto fail;
2844 	}
2845 	i = 0;
2846 	if (lma > 1 && pci_get_slot(dev) < nitems(enaddr))
2847 		i = pci_get_slot(dev);
2848 	memcpy(sc->sc_enaddr, enaddr[i], ETHER_ADDR_LEN);
2849 
2850 	if (phy == 0) {
2851 		device_printf(dev, "could not determine PHY type\n");
2852 		goto fail;
2853 	}
2854 	i = 0;
2855 	if (phy > 1 && pci_get_slot(dev) < nitems(pcs))
2856 		i = pci_get_slot(dev);
2857 	if (pcs[i] != 0)
2858 		sc->sc_flags |= CAS_SERDES;
2859 #endif
2860 
2861 	if (cas_attach(sc) != 0) {
2862 		device_printf(dev, "could not be attached\n");
2863 		goto fail;
2864 	}
2865 
2866 	if (bus_setup_intr(dev, sc->sc_res[CAS_RES_INTR], INTR_TYPE_NET |
2867 	    INTR_MPSAFE, cas_intr, NULL, sc, &sc->sc_ih) != 0) {
2868 		device_printf(dev, "failed to set up interrupt\n");
2869 		cas_detach(sc);
2870 		goto fail;
2871 	}
2872 	return (0);
2873 
2874  fail:
2875 	CAS_LOCK_DESTROY(sc);
2876 	bus_release_resources(dev, cas_pci_res_spec, sc->sc_res);
2877 	return (ENXIO);
2878 }
2879 
2880 static int
2881 cas_pci_detach(device_t dev)
2882 {
2883 	struct cas_softc *sc;
2884 
2885 	sc = device_get_softc(dev);
2886 	bus_teardown_intr(dev, sc->sc_res[CAS_RES_INTR], sc->sc_ih);
2887 	cas_detach(sc);
2888 	CAS_LOCK_DESTROY(sc);
2889 	bus_release_resources(dev, cas_pci_res_spec, sc->sc_res);
2890 	return (0);
2891 }
2892 
2893 static int
2894 cas_pci_suspend(device_t dev)
2895 {
2896 
2897 	cas_suspend(device_get_softc(dev));
2898 	return (0);
2899 }
2900 
2901 static int
2902 cas_pci_resume(device_t dev)
2903 {
2904 
2905 	cas_resume(device_get_softc(dev));
2906 	return (0);
2907 }
2908