if_cas.c revision 213893
1/*-
2 * Copyright (C) 2001 Eduardo Horvath.
3 * Copyright (c) 2001-2003 Thomas Moestl
4 * Copyright (c) 2007-2009 Marius Strobl <marius@FreeBSD.org>
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 *    notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 *    notice, this list of conditions and the following disclaimer in the
14 *    documentation and/or other materials provided with the distribution.
15 *
16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR  ``AS IS'' AND
17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR  BE LIABLE
20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26 * SUCH DAMAGE.
27 *
28 *	from: NetBSD: gem.c,v 1.21 2002/06/01 23:50:58 lukem Exp
29 *	from: FreeBSD: if_gem.c 182060 2008-08-23 15:03:26Z marius
30 */
31
32#include <sys/cdefs.h>
33__FBSDID("$FreeBSD: head/sys/dev/cas/if_cas.c 213893 2010-10-15 14:52:11Z marius $");
34
35/*
36 * driver for Sun Cassini/Cassini+ and National Semiconductor DP83065
37 * Saturn Gigabit Ethernet controllers
38 */
39
40#if 0
41#define	CAS_DEBUG
42#endif
43
44#include <sys/param.h>
45#include <sys/systm.h>
46#include <sys/bus.h>
47#include <sys/callout.h>
48#include <sys/endian.h>
49#include <sys/mbuf.h>
50#include <sys/malloc.h>
51#include <sys/kernel.h>
52#include <sys/lock.h>
53#include <sys/module.h>
54#include <sys/mutex.h>
55#include <sys/refcount.h>
56#include <sys/resource.h>
57#include <sys/rman.h>
58#include <sys/socket.h>
59#include <sys/sockio.h>
60#include <sys/taskqueue.h>
61
62#include <net/bpf.h>
63#include <net/ethernet.h>
64#include <net/if.h>
65#include <net/if_arp.h>
66#include <net/if_dl.h>
67#include <net/if_media.h>
68#include <net/if_types.h>
69#include <net/if_vlan_var.h>
70
71#include <netinet/in.h>
72#include <netinet/in_systm.h>
73#include <netinet/ip.h>
74#include <netinet/tcp.h>
75#include <netinet/udp.h>
76
77#include <machine/bus.h>
78#if defined(__powerpc__) || defined(__sparc64__)
79#include <dev/ofw/ofw_bus.h>
80#include <dev/ofw/openfirm.h>
81#include <machine/ofw_machdep.h>
82#endif
83#include <machine/resource.h>
84
85#include <dev/mii/mii.h>
86#include <dev/mii/miivar.h>
87
88#include <dev/cas/if_casreg.h>
89#include <dev/cas/if_casvar.h>
90
91#include <dev/pci/pcireg.h>
92#include <dev/pci/pcivar.h>
93
94#include "miibus_if.h"
95
96#define RINGASSERT(n , min, max)					\
97	CTASSERT(powerof2(n) && (n) >= (min) && (n) <= (max))
98
99RINGASSERT(CAS_NRXCOMP, 128, 32768);
100RINGASSERT(CAS_NRXDESC, 32, 8192);
101RINGASSERT(CAS_NRXDESC2, 32, 8192);
102RINGASSERT(CAS_NTXDESC, 32, 8192);
103
104#undef RINGASSERT
105
106#define	CCDASSERT(m, a)							\
107	CTASSERT((offsetof(struct cas_control_data, m) & ((a) - 1)) == 0)
108
109CCDASSERT(ccd_rxcomps, CAS_RX_COMP_ALIGN);
110CCDASSERT(ccd_rxdescs, CAS_RX_DESC_ALIGN);
111CCDASSERT(ccd_rxdescs2, CAS_RX_DESC_ALIGN);
112
113#undef CCDASSERT
114
115#define	CAS_TRIES	10000
116
117/*
118 * According to documentation, the hardware has support for basic TCP
119 * checksum offloading only, in practice this can be also used for UDP
120 * however (i.e. the problem of previous Sun NICs that a checksum of 0x0
121 * is not converted to 0xffff no longer exists).
122 */
123#define	CAS_CSUM_FEATURES	(CSUM_TCP | CSUM_UDP)
124
125static inline void cas_add_rxdesc(struct cas_softc *sc, u_int idx);
126static int	cas_attach(struct cas_softc *sc);
127static int	cas_bitwait(struct cas_softc *sc, bus_addr_t r, uint32_t clr,
128		    uint32_t set);
129static void	cas_cddma_callback(void *xsc, bus_dma_segment_t *segs,
130		    int nsegs, int error);
131static void	cas_detach(struct cas_softc *sc);
132static int	cas_disable_rx(struct cas_softc *sc);
133static int	cas_disable_tx(struct cas_softc *sc);
134static void	cas_eint(struct cas_softc *sc, u_int status);
135static void	cas_free(void *arg1, void* arg2);
136static void	cas_init(void *xsc);
137static void	cas_init_locked(struct cas_softc *sc);
138static void	cas_init_regs(struct cas_softc *sc);
139static int	cas_intr(void *v);
140static void	cas_intr_task(void *arg, int pending __unused);
141static int	cas_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data);
142static int	cas_load_txmbuf(struct cas_softc *sc, struct mbuf **m_head);
143static int	cas_mediachange(struct ifnet *ifp);
144static void	cas_mediastatus(struct ifnet *ifp, struct ifmediareq *ifmr);
145static void	cas_meminit(struct cas_softc *sc);
146static void	cas_mifinit(struct cas_softc *sc);
147static int	cas_mii_readreg(device_t dev, int phy, int reg);
148static void	cas_mii_statchg(device_t dev);
149static int	cas_mii_writereg(device_t dev, int phy, int reg, int val);
150static void	cas_reset(struct cas_softc *sc);
151static int	cas_reset_rx(struct cas_softc *sc);
152static int	cas_reset_tx(struct cas_softc *sc);
153static void	cas_resume(struct cas_softc *sc);
154static u_int	cas_descsize(u_int sz);
155static void	cas_rint(struct cas_softc *sc);
156static void	cas_rint_timeout(void *arg);
157static inline void cas_rxcksum(struct mbuf *m, uint16_t cksum);
158static inline void cas_rxcompinit(struct cas_rx_comp *rxcomp);
159static u_int	cas_rxcompsize(u_int sz);
160static void	cas_rxdma_callback(void *xsc, bus_dma_segment_t *segs,
161		    int nsegs, int error);
162static void	cas_setladrf(struct cas_softc *sc);
163static void	cas_start(struct ifnet *ifp);
164static void	cas_stop(struct ifnet *ifp);
165static void	cas_suspend(struct cas_softc *sc);
166static void	cas_tick(void *arg);
167static void	cas_tint(struct cas_softc *sc);
168static void	cas_tx_task(void *arg, int pending __unused);
169static inline void cas_txkick(struct cas_softc *sc);
170static void	cas_watchdog(struct cas_softc *sc);
171
172static devclass_t cas_devclass;
173
174MODULE_DEPEND(cas, ether, 1, 1, 1);
175MODULE_DEPEND(cas, miibus, 1, 1, 1);
176
177#ifdef CAS_DEBUG
178#include <sys/ktr.h>
179#define	KTR_CAS		KTR_SPARE2
180#endif
181
182static int
183cas_attach(struct cas_softc *sc)
184{
185	struct cas_txsoft *txs;
186	struct ifnet *ifp;
187	int error, i;
188	uint32_t v;
189
190	/* Set up ifnet structure. */
191	ifp = sc->sc_ifp = if_alloc(IFT_ETHER);
192	if (ifp == NULL)
193		return (ENOSPC);
194	ifp->if_softc = sc;
195	if_initname(ifp, device_get_name(sc->sc_dev),
196	    device_get_unit(sc->sc_dev));
197	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
198	ifp->if_start = cas_start;
199	ifp->if_ioctl = cas_ioctl;
200	ifp->if_init = cas_init;
201	IFQ_SET_MAXLEN(&ifp->if_snd, CAS_TXQUEUELEN);
202	ifp->if_snd.ifq_drv_maxlen = CAS_TXQUEUELEN;
203	IFQ_SET_READY(&ifp->if_snd);
204
205	callout_init_mtx(&sc->sc_tick_ch, &sc->sc_mtx, 0);
206	callout_init(&sc->sc_rx_ch, 1);
207	/* Create local taskq. */
208	TASK_INIT(&sc->sc_intr_task, 0, cas_intr_task, sc);
209	TASK_INIT(&sc->sc_tx_task, 1, cas_tx_task, ifp);
210	sc->sc_tq = taskqueue_create_fast("cas_taskq", M_WAITOK,
211	    taskqueue_thread_enqueue, &sc->sc_tq);
212	if (sc->sc_tq == NULL) {
213		device_printf(sc->sc_dev, "could not create taskqueue\n");
214		error = ENXIO;
215		goto fail_ifnet;
216	}
217	taskqueue_start_threads(&sc->sc_tq, 1, PI_NET, "%s taskq",
218	    device_get_nameunit(sc->sc_dev));
219
220	/* Make sure the chip is stopped. */
221	cas_reset(sc);
222
223	error = bus_dma_tag_create(bus_get_dma_tag(sc->sc_dev), 1, 0,
224	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
225	    BUS_SPACE_MAXSIZE, 0, BUS_SPACE_MAXSIZE, 0, NULL, NULL,
226	    &sc->sc_pdmatag);
227	if (error != 0)
228		goto fail_taskq;
229
230	error = bus_dma_tag_create(sc->sc_pdmatag, 1, 0,
231	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
232	    CAS_PAGE_SIZE, 1, CAS_PAGE_SIZE, 0, NULL, NULL, &sc->sc_rdmatag);
233	if (error != 0)
234		goto fail_ptag;
235
236	error = bus_dma_tag_create(sc->sc_pdmatag, 1, 0,
237	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
238	    MCLBYTES * CAS_NTXSEGS, CAS_NTXSEGS, MCLBYTES,
239	    BUS_DMA_ALLOCNOW, NULL, NULL, &sc->sc_tdmatag);
240	if (error != 0)
241		goto fail_rtag;
242
243	error = bus_dma_tag_create(sc->sc_pdmatag, CAS_TX_DESC_ALIGN, 0,
244	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
245	    sizeof(struct cas_control_data), 1,
246	    sizeof(struct cas_control_data), 0,
247	    NULL, NULL, &sc->sc_cdmatag);
248	if (error != 0)
249		goto fail_ttag;
250
251	/*
252	 * Allocate the control data structures, create and load the
253	 * DMA map for it.
254	 */
255	if ((error = bus_dmamem_alloc(sc->sc_cdmatag,
256	    (void **)&sc->sc_control_data,
257	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
258	    &sc->sc_cddmamap)) != 0) {
259		device_printf(sc->sc_dev,
260		    "unable to allocate control data, error = %d\n", error);
261		goto fail_ctag;
262	}
263
264	sc->sc_cddma = 0;
265	if ((error = bus_dmamap_load(sc->sc_cdmatag, sc->sc_cddmamap,
266	    sc->sc_control_data, sizeof(struct cas_control_data),
267	    cas_cddma_callback, sc, 0)) != 0 || sc->sc_cddma == 0) {
268		device_printf(sc->sc_dev,
269		    "unable to load control data DMA map, error = %d\n",
270		    error);
271		goto fail_cmem;
272	}
273
274	/*
275	 * Initialize the transmit job descriptors.
276	 */
277	STAILQ_INIT(&sc->sc_txfreeq);
278	STAILQ_INIT(&sc->sc_txdirtyq);
279
280	/*
281	 * Create the transmit buffer DMA maps.
282	 */
283	error = ENOMEM;
284	for (i = 0; i < CAS_TXQUEUELEN; i++) {
285		txs = &sc->sc_txsoft[i];
286		txs->txs_mbuf = NULL;
287		txs->txs_ndescs = 0;
288		if ((error = bus_dmamap_create(sc->sc_tdmatag, 0,
289		    &txs->txs_dmamap)) != 0) {
290			device_printf(sc->sc_dev,
291			    "unable to create TX DMA map %d, error = %d\n",
292			    i, error);
293			goto fail_txd;
294		}
295		STAILQ_INSERT_TAIL(&sc->sc_txfreeq, txs, txs_q);
296	}
297
298	/*
299	 * Allocate the receive buffers, create and load the DMA maps
300	 * for them.
301	 */
302	for (i = 0; i < CAS_NRXDESC; i++) {
303		if ((error = bus_dmamem_alloc(sc->sc_rdmatag,
304		    &sc->sc_rxdsoft[i].rxds_buf, BUS_DMA_WAITOK,
305		    &sc->sc_rxdsoft[i].rxds_dmamap)) != 0) {
306			device_printf(sc->sc_dev,
307			    "unable to allocate RX buffer %d, error = %d\n",
308			    i, error);
309			goto fail_rxmem;
310		}
311
312		sc->sc_rxdptr = i;
313		sc->sc_rxdsoft[i].rxds_paddr = 0;
314		if ((error = bus_dmamap_load(sc->sc_rdmatag,
315		    sc->sc_rxdsoft[i].rxds_dmamap, sc->sc_rxdsoft[i].rxds_buf,
316		    CAS_PAGE_SIZE, cas_rxdma_callback, sc, 0)) != 0 ||
317		    sc->sc_rxdsoft[i].rxds_paddr == 0) {
318			device_printf(sc->sc_dev,
319			    "unable to load RX DMA map %d, error = %d\n",
320			    i, error);
321			goto fail_rxmap;
322		}
323	}
324
325	if ((sc->sc_flags & CAS_SERDES) == 0) {
326		CAS_WRITE_4(sc, CAS_PCS_DATAPATH, CAS_PCS_DATAPATH_MII);
327		CAS_BARRIER(sc, CAS_PCS_DATAPATH, 4,
328		    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
329		cas_mifinit(sc);
330		/*
331		 * Look for an external PHY.
332		 */
333		error = ENXIO;
334		v = CAS_READ_4(sc, CAS_MIF_CONF);
335		if ((v & CAS_MIF_CONF_MDI1) != 0) {
336			v |= CAS_MIF_CONF_PHY_SELECT;
337			CAS_WRITE_4(sc, CAS_MIF_CONF, v);
338			CAS_BARRIER(sc, CAS_MIF_CONF, 4,
339			    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
340			/* Enable/unfreeze the GMII pins of Saturn. */
341			if (sc->sc_variant == CAS_SATURN) {
342				CAS_WRITE_4(sc, CAS_SATURN_PCFG, 0);
343				CAS_BARRIER(sc, CAS_SATURN_PCFG, 4,
344				    BUS_SPACE_BARRIER_READ |
345				    BUS_SPACE_BARRIER_WRITE);
346			}
347			error = mii_attach(sc->sc_dev, &sc->sc_miibus, ifp,
348			    cas_mediachange, cas_mediastatus, BMSR_DEFCAPMASK,
349			    MII_PHY_ANY, MII_OFFSET_ANY, 0);
350		}
351		/*
352		 * Fall back on an internal PHY if no external PHY was found.
353		 */
354		if (error != 0 && (v & CAS_MIF_CONF_MDI0) != 0) {
355			v &= ~CAS_MIF_CONF_PHY_SELECT;
356			CAS_WRITE_4(sc, CAS_MIF_CONF, v);
357			CAS_BARRIER(sc, CAS_MIF_CONF, 4,
358			    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
359			/* Freeze the GMII pins of Saturn for saving power. */
360			if (sc->sc_variant == CAS_SATURN) {
361				CAS_WRITE_4(sc, CAS_SATURN_PCFG,
362				    CAS_SATURN_PCFG_FSI);
363				CAS_BARRIER(sc, CAS_SATURN_PCFG, 4,
364				    BUS_SPACE_BARRIER_READ |
365				    BUS_SPACE_BARRIER_WRITE);
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, 0);
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, 0);
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	ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
417	ifp->if_capabilities = IFCAP_VLAN_MTU;
418	if ((sc->sc_flags & CAS_NO_CSUM) == 0) {
419		ifp->if_capabilities |= IFCAP_HWCSUM;
420		ifp->if_hwassist = CAS_CSUM_FEATURES;
421	}
422	ifp->if_capenable = ifp->if_capabilities;
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 fail_ifnet:
461	if_free(ifp);
462	return (error);
463}
464
465static void
466cas_detach(struct cas_softc *sc)
467{
468	struct ifnet *ifp = sc->sc_ifp;
469	int i;
470
471	ether_ifdetach(ifp);
472	CAS_LOCK(sc);
473	cas_stop(ifp);
474	CAS_UNLOCK(sc);
475	callout_drain(&sc->sc_tick_ch);
476	callout_drain(&sc->sc_rx_ch);
477	taskqueue_drain(sc->sc_tq, &sc->sc_intr_task);
478	taskqueue_drain(sc->sc_tq, &sc->sc_tx_task);
479	if_free(ifp);
480	taskqueue_free(sc->sc_tq);
481	device_delete_child(sc->sc_dev, sc->sc_miibus);
482
483	for (i = 0; i < CAS_NRXDESC; i++)
484		if (sc->sc_rxdsoft[i].rxds_dmamap != NULL)
485			bus_dmamap_sync(sc->sc_rdmatag,
486			    sc->sc_rxdsoft[i].rxds_dmamap,
487			    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
488	for (i = 0; i < CAS_NRXDESC; i++)
489		if (sc->sc_rxdsoft[i].rxds_paddr != 0)
490			bus_dmamap_unload(sc->sc_rdmatag,
491			    sc->sc_rxdsoft[i].rxds_dmamap);
492	for (i = 0; i < CAS_NRXDESC; i++)
493		if (sc->sc_rxdsoft[i].rxds_buf != NULL)
494			bus_dmamem_free(sc->sc_rdmatag,
495			    sc->sc_rxdsoft[i].rxds_buf,
496			    sc->sc_rxdsoft[i].rxds_dmamap);
497	for (i = 0; i < CAS_TXQUEUELEN; i++)
498		if (sc->sc_txsoft[i].txs_dmamap != NULL)
499			bus_dmamap_destroy(sc->sc_tdmatag,
500			    sc->sc_txsoft[i].txs_dmamap);
501	CAS_CDSYNC(sc, BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
502	bus_dmamap_unload(sc->sc_cdmatag, sc->sc_cddmamap);
503	bus_dmamem_free(sc->sc_cdmatag, sc->sc_control_data,
504	    sc->sc_cddmamap);
505	bus_dma_tag_destroy(sc->sc_cdmatag);
506	bus_dma_tag_destroy(sc->sc_tdmatag);
507	bus_dma_tag_destroy(sc->sc_rdmatag);
508	bus_dma_tag_destroy(sc->sc_pdmatag);
509}
510
511static void
512cas_suspend(struct cas_softc *sc)
513{
514	struct ifnet *ifp = sc->sc_ifp;
515
516	CAS_LOCK(sc);
517	cas_stop(ifp);
518	CAS_UNLOCK(sc);
519}
520
521static void
522cas_resume(struct cas_softc *sc)
523{
524	struct ifnet *ifp = sc->sc_ifp;
525
526	CAS_LOCK(sc);
527	/*
528	 * On resume all registers have to be initialized again like
529	 * after power-on.
530	 */
531	sc->sc_flags &= ~CAS_INITED;
532	if (ifp->if_flags & IFF_UP)
533		cas_init_locked(sc);
534	CAS_UNLOCK(sc);
535}
536
537static inline void
538cas_rxcksum(struct mbuf *m, uint16_t cksum)
539{
540	struct ether_header *eh;
541	struct ip *ip;
542	struct udphdr *uh;
543	uint16_t *opts;
544	int32_t hlen, len, pktlen;
545	uint32_t temp32;
546
547	pktlen = m->m_pkthdr.len;
548	if (pktlen < sizeof(struct ether_header) + sizeof(struct ip))
549		return;
550	eh = mtod(m, struct ether_header *);
551	if (eh->ether_type != htons(ETHERTYPE_IP))
552		return;
553	ip = (struct ip *)(eh + 1);
554	if (ip->ip_v != IPVERSION)
555		return;
556
557	hlen = ip->ip_hl << 2;
558	pktlen -= sizeof(struct ether_header);
559	if (hlen < sizeof(struct ip))
560		return;
561	if (ntohs(ip->ip_len) < hlen)
562		return;
563	if (ntohs(ip->ip_len) != pktlen)
564		return;
565	if (ip->ip_off & htons(IP_MF | IP_OFFMASK))
566		return;	/* Cannot handle fragmented packet. */
567
568	switch (ip->ip_p) {
569	case IPPROTO_TCP:
570		if (pktlen < (hlen + sizeof(struct tcphdr)))
571			return;
572		break;
573	case IPPROTO_UDP:
574		if (pktlen < (hlen + sizeof(struct udphdr)))
575			return;
576		uh = (struct udphdr *)((uint8_t *)ip + hlen);
577		if (uh->uh_sum == 0)
578			return; /* no checksum */
579		break;
580	default:
581		return;
582	}
583
584	cksum = ~cksum;
585	/* checksum fixup for IP options */
586	len = hlen - sizeof(struct ip);
587	if (len > 0) {
588		opts = (uint16_t *)(ip + 1);
589		for (; len > 0; len -= sizeof(uint16_t), opts++) {
590			temp32 = cksum - *opts;
591			temp32 = (temp32 >> 16) + (temp32 & 65535);
592			cksum = temp32 & 65535;
593		}
594	}
595	m->m_pkthdr.csum_flags |= CSUM_DATA_VALID;
596	m->m_pkthdr.csum_data = cksum;
597}
598
599static void
600cas_cddma_callback(void *xsc, bus_dma_segment_t *segs, int nsegs, int error)
601{
602	struct cas_softc *sc = xsc;
603
604	if (error != 0)
605		return;
606	if (nsegs != 1)
607		panic("%s: bad control buffer segment count", __func__);
608	sc->sc_cddma = segs[0].ds_addr;
609}
610
611static void
612cas_rxdma_callback(void *xsc, bus_dma_segment_t *segs, int nsegs, int error)
613{
614	struct cas_softc *sc = xsc;
615
616	if (error != 0)
617		return;
618	if (nsegs != 1)
619		panic("%s: bad RX buffer segment count", __func__);
620	sc->sc_rxdsoft[sc->sc_rxdptr].rxds_paddr = segs[0].ds_addr;
621}
622
623static void
624cas_tick(void *arg)
625{
626	struct cas_softc *sc = arg;
627	struct ifnet *ifp = sc->sc_ifp;
628	uint32_t v;
629
630	CAS_LOCK_ASSERT(sc, MA_OWNED);
631
632	/*
633	 * Unload collision and error counters.
634	 */
635	ifp->if_collisions +=
636	    CAS_READ_4(sc, CAS_MAC_NORM_COLL_CNT) +
637	    CAS_READ_4(sc, CAS_MAC_FIRST_COLL_CNT);
638	v = CAS_READ_4(sc, CAS_MAC_EXCESS_COLL_CNT) +
639	    CAS_READ_4(sc, CAS_MAC_LATE_COLL_CNT);
640	ifp->if_collisions += v;
641	ifp->if_oerrors += v;
642	ifp->if_ierrors +=
643	    CAS_READ_4(sc, CAS_MAC_RX_LEN_ERR_CNT) +
644	    CAS_READ_4(sc, CAS_MAC_RX_ALIGN_ERR) +
645	    CAS_READ_4(sc, CAS_MAC_RX_CRC_ERR_CNT) +
646	    CAS_READ_4(sc, CAS_MAC_RX_CODE_VIOL);
647
648	/*
649	 * Then clear the hardware counters.
650	 */
651	CAS_WRITE_4(sc, CAS_MAC_NORM_COLL_CNT, 0);
652	CAS_WRITE_4(sc, CAS_MAC_FIRST_COLL_CNT, 0);
653	CAS_WRITE_4(sc, CAS_MAC_EXCESS_COLL_CNT, 0);
654	CAS_WRITE_4(sc, CAS_MAC_LATE_COLL_CNT, 0);
655	CAS_WRITE_4(sc, CAS_MAC_RX_LEN_ERR_CNT, 0);
656	CAS_WRITE_4(sc, CAS_MAC_RX_ALIGN_ERR, 0);
657	CAS_WRITE_4(sc, CAS_MAC_RX_CRC_ERR_CNT, 0);
658	CAS_WRITE_4(sc, CAS_MAC_RX_CODE_VIOL, 0);
659
660	mii_tick(sc->sc_mii);
661
662	if (sc->sc_txfree != CAS_MAXTXFREE)
663		cas_tint(sc);
664
665	cas_watchdog(sc);
666
667	callout_reset(&sc->sc_tick_ch, hz, cas_tick, sc);
668}
669
670static int
671cas_bitwait(struct cas_softc *sc, bus_addr_t r, uint32_t clr, uint32_t set)
672{
673	int i;
674	uint32_t reg;
675
676	for (i = CAS_TRIES; i--; DELAY(100)) {
677		reg = CAS_READ_4(sc, r);
678		if ((reg & clr) == 0 && (reg & set) == set)
679			return (1);
680	}
681	return (0);
682}
683
684static void
685cas_reset(struct cas_softc *sc)
686{
687
688#ifdef CAS_DEBUG
689	CTR2(KTR_CAS, "%s: %s", device_get_name(sc->sc_dev), __func__);
690#endif
691	/* Disable all interrupts in order to avoid spurious ones. */
692	CAS_WRITE_4(sc, CAS_INTMASK, 0xffffffff);
693
694	cas_reset_rx(sc);
695	cas_reset_tx(sc);
696
697	/*
698	 * Do a full reset modulo the result of the last auto-negotiation
699	 * when using the SERDES.
700	 */
701	CAS_WRITE_4(sc, CAS_RESET, CAS_RESET_RX | CAS_RESET_TX |
702	    ((sc->sc_flags & CAS_SERDES) != 0 ? CAS_RESET_PCS_DIS : 0));
703	CAS_BARRIER(sc, CAS_RESET, 4,
704	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
705	DELAY(3000);
706	if (!cas_bitwait(sc, CAS_RESET, CAS_RESET_RX | CAS_RESET_TX, 0))
707		device_printf(sc->sc_dev, "cannot reset device\n");
708}
709
710static void
711cas_stop(struct ifnet *ifp)
712{
713	struct cas_softc *sc = ifp->if_softc;
714	struct cas_txsoft *txs;
715
716#ifdef CAS_DEBUG
717	CTR2(KTR_CAS, "%s: %s", device_get_name(sc->sc_dev), __func__);
718#endif
719
720	callout_stop(&sc->sc_tick_ch);
721	callout_stop(&sc->sc_rx_ch);
722
723	/* Disable all interrupts in order to avoid spurious ones. */
724	CAS_WRITE_4(sc, CAS_INTMASK, 0xffffffff);
725
726	cas_reset_tx(sc);
727	cas_reset_rx(sc);
728
729	/*
730	 * Release any queued transmit buffers.
731	 */
732	while ((txs = STAILQ_FIRST(&sc->sc_txdirtyq)) != NULL) {
733		STAILQ_REMOVE_HEAD(&sc->sc_txdirtyq, txs_q);
734		if (txs->txs_ndescs != 0) {
735			bus_dmamap_sync(sc->sc_tdmatag, txs->txs_dmamap,
736			    BUS_DMASYNC_POSTWRITE);
737			bus_dmamap_unload(sc->sc_tdmatag, txs->txs_dmamap);
738			if (txs->txs_mbuf != NULL) {
739				m_freem(txs->txs_mbuf);
740				txs->txs_mbuf = NULL;
741			}
742		}
743		STAILQ_INSERT_TAIL(&sc->sc_txfreeq, txs, txs_q);
744	}
745
746	/*
747	 * Mark the interface down and cancel the watchdog timer.
748	 */
749	ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
750	sc->sc_flags &= ~CAS_LINK;
751	sc->sc_wdog_timer = 0;
752}
753
754static int
755cas_reset_rx(struct cas_softc *sc)
756{
757
758	/*
759	 * Resetting while DMA is in progress can cause a bus hang, so we
760	 * disable DMA first.
761	 */
762	cas_disable_rx(sc);
763	CAS_WRITE_4(sc, CAS_RX_CONF, 0);
764	CAS_BARRIER(sc, CAS_RX_CONF, 4,
765	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
766	if (!cas_bitwait(sc, CAS_RX_CONF, CAS_RX_CONF_RXDMA_EN, 0))
767		device_printf(sc->sc_dev, "cannot disable RX DMA\n");
768
769	/* Finally, reset the ERX. */
770	CAS_WRITE_4(sc, CAS_RESET, CAS_RESET_RX |
771	    ((sc->sc_flags & CAS_SERDES) != 0 ? CAS_RESET_PCS_DIS : 0));
772	CAS_BARRIER(sc, CAS_RESET, 4,
773	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
774	if (!cas_bitwait(sc, CAS_RESET, CAS_RESET_RX | CAS_RESET_TX, 0)) {
775		device_printf(sc->sc_dev, "cannot reset receiver\n");
776		return (1);
777	}
778	return (0);
779}
780
781static int
782cas_reset_tx(struct cas_softc *sc)
783{
784
785	/*
786	 * Resetting while DMA is in progress can cause a bus hang, so we
787	 * disable DMA first.
788	 */
789	cas_disable_tx(sc);
790	CAS_WRITE_4(sc, CAS_TX_CONF, 0);
791	CAS_BARRIER(sc, CAS_TX_CONF, 4,
792	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
793	if (!cas_bitwait(sc, CAS_TX_CONF, CAS_TX_CONF_TXDMA_EN, 0))
794		device_printf(sc->sc_dev, "cannot disable TX DMA\n");
795
796	/* Finally, reset the ETX. */
797	CAS_WRITE_4(sc, CAS_RESET, CAS_RESET_TX |
798	    ((sc->sc_flags & CAS_SERDES) != 0 ? CAS_RESET_PCS_DIS : 0));
799	CAS_BARRIER(sc, CAS_RESET, 4,
800	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
801	if (!cas_bitwait(sc, CAS_RESET, CAS_RESET_RX | CAS_RESET_TX, 0)) {
802		device_printf(sc->sc_dev, "cannot reset transmitter\n");
803		return (1);
804	}
805	return (0);
806}
807
808static int
809cas_disable_rx(struct cas_softc *sc)
810{
811
812	CAS_WRITE_4(sc, CAS_MAC_RX_CONF,
813	    CAS_READ_4(sc, CAS_MAC_RX_CONF) & ~CAS_MAC_RX_CONF_EN);
814	CAS_BARRIER(sc, CAS_MAC_RX_CONF, 4,
815	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
816	return (cas_bitwait(sc, CAS_MAC_RX_CONF, CAS_MAC_RX_CONF_EN, 0));
817}
818
819static int
820cas_disable_tx(struct cas_softc *sc)
821{
822
823	CAS_WRITE_4(sc, CAS_MAC_TX_CONF,
824	    CAS_READ_4(sc, CAS_MAC_TX_CONF) & ~CAS_MAC_TX_CONF_EN);
825	CAS_BARRIER(sc, CAS_MAC_TX_CONF, 4,
826	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
827	return (cas_bitwait(sc, CAS_MAC_TX_CONF, CAS_MAC_TX_CONF_EN, 0));
828}
829
830static inline void
831cas_rxcompinit(struct cas_rx_comp *rxcomp)
832{
833
834	rxcomp->crc_word1 = 0;
835	rxcomp->crc_word2 = 0;
836	rxcomp->crc_word3 =
837	    htole64(CAS_SET(ETHER_HDR_LEN + sizeof(struct ip), CAS_RC3_CSO));
838	rxcomp->crc_word4 = htole64(CAS_RC4_ZERO);
839}
840
841static void
842cas_meminit(struct cas_softc *sc)
843{
844	int i;
845
846	CAS_LOCK_ASSERT(sc, MA_OWNED);
847
848	/*
849	 * Initialize the transmit descriptor ring.
850	 */
851	for (i = 0; i < CAS_NTXDESC; i++) {
852		sc->sc_txdescs[i].cd_flags = 0;
853		sc->sc_txdescs[i].cd_buf_ptr = 0;
854	}
855	sc->sc_txfree = CAS_MAXTXFREE;
856	sc->sc_txnext = 0;
857	sc->sc_txwin = 0;
858
859	/*
860	 * Initialize the receive completion ring.
861	 */
862	for (i = 0; i < CAS_NRXCOMP; i++)
863		cas_rxcompinit(&sc->sc_rxcomps[i]);
864	sc->sc_rxcptr = 0;
865
866	/*
867	 * Initialize the first receive descriptor ring.  We leave
868	 * the second one zeroed as we don't actually use it.
869	 */
870	for (i = 0; i < CAS_NRXDESC; i++)
871		CAS_INIT_RXDESC(sc, i, i);
872	sc->sc_rxdptr = 0;
873
874	CAS_CDSYNC(sc, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
875}
876
877static u_int
878cas_descsize(u_int sz)
879{
880
881	switch (sz) {
882	case 32:
883		return (CAS_DESC_32);
884	case 64:
885		return (CAS_DESC_64);
886	case 128:
887		return (CAS_DESC_128);
888	case 256:
889		return (CAS_DESC_256);
890	case 512:
891		return (CAS_DESC_512);
892	case 1024:
893		return (CAS_DESC_1K);
894	case 2048:
895		return (CAS_DESC_2K);
896	case 4096:
897		return (CAS_DESC_4K);
898	case 8192:
899		return (CAS_DESC_8K);
900	default:
901		printf("%s: invalid descriptor ring size %d\n", __func__, sz);
902		return (CAS_DESC_32);
903	}
904}
905
906static u_int
907cas_rxcompsize(u_int sz)
908{
909
910	switch (sz) {
911	case 128:
912		return (CAS_RX_CONF_COMP_128);
913	case 256:
914		return (CAS_RX_CONF_COMP_256);
915	case 512:
916		return (CAS_RX_CONF_COMP_512);
917	case 1024:
918		return (CAS_RX_CONF_COMP_1K);
919	case 2048:
920		return (CAS_RX_CONF_COMP_2K);
921	case 4096:
922		return (CAS_RX_CONF_COMP_4K);
923	case 8192:
924		return (CAS_RX_CONF_COMP_8K);
925	case 16384:
926		return (CAS_RX_CONF_COMP_16K);
927	case 32768:
928		return (CAS_RX_CONF_COMP_32K);
929	default:
930		printf("%s: invalid dcompletion ring size %d\n", __func__, sz);
931		return (CAS_RX_CONF_COMP_128);
932	}
933}
934
935static void
936cas_init(void *xsc)
937{
938	struct cas_softc *sc = xsc;
939
940	CAS_LOCK(sc);
941	cas_init_locked(sc);
942	CAS_UNLOCK(sc);
943}
944
945/*
946 * Initialization of interface; set up initialization block
947 * and transmit/receive descriptor rings.
948 */
949static void
950cas_init_locked(struct cas_softc *sc)
951{
952	struct ifnet *ifp = sc->sc_ifp;
953	uint32_t v;
954
955	CAS_LOCK_ASSERT(sc, MA_OWNED);
956
957	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
958		return;
959
960#ifdef CAS_DEBUG
961	CTR2(KTR_CAS, "%s: %s: calling stop", device_get_name(sc->sc_dev),
962	    __func__);
963#endif
964	/*
965	 * Initialization sequence.  The numbered steps below correspond
966	 * to the sequence outlined in section 6.3.5.1 in the Ethernet
967	 * Channel Engine manual (part of the PCIO manual).
968	 * See also the STP2002-STQ document from Sun Microsystems.
969	 */
970
971	/* step 1 & 2.  Reset the Ethernet Channel. */
972	cas_stop(ifp);
973	cas_reset(sc);
974#ifdef CAS_DEBUG
975	CTR2(KTR_CAS, "%s: %s: restarting", device_get_name(sc->sc_dev),
976	    __func__);
977#endif
978
979	if ((sc->sc_flags & CAS_SERDES) == 0)
980		/* Re-initialize the MIF. */
981		cas_mifinit(sc);
982
983	/* step 3.  Setup data structures in host memory. */
984	cas_meminit(sc);
985
986	/* step 4.  TX MAC registers & counters */
987	cas_init_regs(sc);
988
989	/* step 5.  RX MAC registers & counters */
990	cas_setladrf(sc);
991
992	/* step 6 & 7.  Program Ring Base Addresses. */
993	CAS_WRITE_4(sc, CAS_TX_DESC3_BASE_HI,
994	    (((uint64_t)CAS_CDTXDADDR(sc, 0)) >> 32));
995	CAS_WRITE_4(sc, CAS_TX_DESC3_BASE_LO,
996	    CAS_CDTXDADDR(sc, 0) & 0xffffffff);
997
998	CAS_WRITE_4(sc, CAS_RX_COMP_BASE_HI,
999	    (((uint64_t)CAS_CDRXCADDR(sc, 0)) >> 32));
1000	CAS_WRITE_4(sc, CAS_RX_COMP_BASE_LO,
1001	    CAS_CDRXCADDR(sc, 0) & 0xffffffff);
1002
1003	CAS_WRITE_4(sc, CAS_RX_DESC_BASE_HI,
1004	    (((uint64_t)CAS_CDRXDADDR(sc, 0)) >> 32));
1005	CAS_WRITE_4(sc, CAS_RX_DESC_BASE_LO,
1006	    CAS_CDRXDADDR(sc, 0) & 0xffffffff);
1007
1008	if ((sc->sc_flags & CAS_REG_PLUS) != 0) {
1009		CAS_WRITE_4(sc, CAS_RX_DESC2_BASE_HI,
1010		    (((uint64_t)CAS_CDRXD2ADDR(sc, 0)) >> 32));
1011		CAS_WRITE_4(sc, CAS_RX_DESC2_BASE_LO,
1012		    CAS_CDRXD2ADDR(sc, 0) & 0xffffffff);
1013	}
1014
1015#ifdef CAS_DEBUG
1016	CTR5(KTR_CAS,
1017	    "loading TXDR %lx, RXCR %lx, RXDR %lx, RXD2R %lx, cddma %lx",
1018	    CAS_CDTXDADDR(sc, 0), CAS_CDRXCADDR(sc, 0), CAS_CDRXDADDR(sc, 0),
1019	    CAS_CDRXD2ADDR(sc, 0), sc->sc_cddma);
1020#endif
1021
1022	/* step 8.  Global Configuration & Interrupt Masks */
1023
1024	/* Disable weighted round robin. */
1025	CAS_WRITE_4(sc, CAS_CAW, CAS_CAW_RR_DIS);
1026
1027	/*
1028	 * Enable infinite bursts for revisions without PCI issues if
1029	 * applicable.  Doing so greatly improves the TX performance on
1030	 * !__sparc64__.
1031	 */
1032	CAS_WRITE_4(sc, CAS_INF_BURST,
1033#if !defined(__sparc64__)
1034	    (sc->sc_flags & CAS_TABORT) == 0 ? CAS_INF_BURST_EN :
1035#endif
1036	    0);
1037
1038	/* Set up interrupts. */
1039	CAS_WRITE_4(sc, CAS_INTMASK,
1040	    ~(CAS_INTR_TX_INT_ME | CAS_INTR_TX_TAG_ERR |
1041	    CAS_INTR_RX_DONE | CAS_INTR_RX_BUF_NA | CAS_INTR_RX_TAG_ERR |
1042	    CAS_INTR_RX_COMP_FULL | CAS_INTR_RX_BUF_AEMPTY |
1043	    CAS_INTR_RX_COMP_AFULL | CAS_INTR_RX_LEN_MMATCH |
1044	    CAS_INTR_PCI_ERROR_INT
1045#ifdef CAS_DEBUG
1046	    | CAS_INTR_PCS_INT | CAS_INTR_MIF
1047#endif
1048	    ));
1049	/* Don't clear top level interrupts when CAS_STATUS_ALIAS is read. */
1050	CAS_WRITE_4(sc, CAS_CLEAR_ALIAS, 0);
1051	CAS_WRITE_4(sc, CAS_MAC_RX_MASK, ~CAS_MAC_RX_OVERFLOW);
1052	CAS_WRITE_4(sc, CAS_MAC_TX_MASK,
1053	    ~(CAS_MAC_TX_UNDERRUN | CAS_MAC_TX_MAX_PKT_ERR));
1054#ifdef CAS_DEBUG
1055	CAS_WRITE_4(sc, CAS_MAC_CTRL_MASK,
1056	    ~(CAS_MAC_CTRL_PAUSE_RCVD | CAS_MAC_CTRL_PAUSE |
1057	    CAS_MAC_CTRL_NON_PAUSE));
1058#else
1059	CAS_WRITE_4(sc, CAS_MAC_CTRL_MASK,
1060	    CAS_MAC_CTRL_PAUSE_RCVD | CAS_MAC_CTRL_PAUSE |
1061	    CAS_MAC_CTRL_NON_PAUSE);
1062#endif
1063
1064	/* Enable PCI error interrupts. */
1065	CAS_WRITE_4(sc, CAS_ERROR_MASK,
1066	    ~(CAS_ERROR_DTRTO | CAS_ERROR_OTHER | CAS_ERROR_DMAW_ZERO |
1067	    CAS_ERROR_DMAR_ZERO | CAS_ERROR_RTRTO));
1068
1069	/* Enable PCI error interrupts in BIM configuration. */
1070	CAS_WRITE_4(sc, CAS_BIM_CONF,
1071	    CAS_BIM_CONF_DPAR_EN | CAS_BIM_CONF_RMA_EN | CAS_BIM_CONF_RTA_EN);
1072
1073	/*
1074	 * step 9.  ETX Configuration: encode receive descriptor ring size,
1075	 * enable DMA and disable pre-interrupt writeback completion.
1076	 */
1077	v = cas_descsize(CAS_NTXDESC) << CAS_TX_CONF_DESC3_SHFT;
1078	CAS_WRITE_4(sc, CAS_TX_CONF, v | CAS_TX_CONF_TXDMA_EN |
1079	    CAS_TX_CONF_RDPP_DIS | CAS_TX_CONF_PICWB_DIS);
1080
1081	/* step 10.  ERX Configuration */
1082
1083	/*
1084	 * Encode receive completion and descriptor ring sizes, set the
1085	 * swivel offset.
1086	 */
1087	v = cas_rxcompsize(CAS_NRXCOMP) << CAS_RX_CONF_COMP_SHFT;
1088	v |= cas_descsize(CAS_NRXDESC) << CAS_RX_CONF_DESC_SHFT;
1089	if ((sc->sc_flags & CAS_REG_PLUS) != 0)
1090		v |= cas_descsize(CAS_NRXDESC2) << CAS_RX_CONF_DESC2_SHFT;
1091	CAS_WRITE_4(sc, CAS_RX_CONF,
1092	    v | (ETHER_ALIGN << CAS_RX_CONF_SOFF_SHFT));
1093
1094	/* Set the PAUSE thresholds.  We use the maximum OFF threshold. */
1095	CAS_WRITE_4(sc, CAS_RX_PTHRS,
1096	    ((111 * 64) << CAS_RX_PTHRS_XOFF_SHFT) |
1097	    ((15 * 64) << CAS_RX_PTHRS_XON_SHFT));
1098
1099	/* RX blanking */
1100	CAS_WRITE_4(sc, CAS_RX_BLANK,
1101	    (15 << CAS_RX_BLANK_TIME_SHFT) | (5 << CAS_RX_BLANK_PKTS_SHFT));
1102
1103	/* Set RX_COMP_AFULL threshold to half of the RX completions. */
1104	CAS_WRITE_4(sc, CAS_RX_AEMPTY_THRS,
1105	    (CAS_NRXCOMP / 2) << CAS_RX_AEMPTY_COMP_SHFT);
1106
1107	/* Initialize the RX page size register as appropriate for 8k. */
1108	CAS_WRITE_4(sc, CAS_RX_PSZ,
1109	    (CAS_RX_PSZ_8K << CAS_RX_PSZ_SHFT) |
1110	    (4 << CAS_RX_PSZ_MB_CNT_SHFT) |
1111	    (CAS_RX_PSZ_MB_STRD_2K << CAS_RX_PSZ_MB_STRD_SHFT) |
1112	    (CAS_RX_PSZ_MB_OFF_64 << CAS_RX_PSZ_MB_OFF_SHFT));
1113
1114	/* Disable RX random early detection. */
1115	CAS_WRITE_4(sc,	CAS_RX_RED, 0);
1116
1117	/* Zero the RX reassembly DMA table. */
1118	for (v = 0; v <= CAS_RX_REAS_DMA_ADDR_LC; v++) {
1119		CAS_WRITE_4(sc,	CAS_RX_REAS_DMA_ADDR, v);
1120		CAS_WRITE_4(sc,	CAS_RX_REAS_DMA_DATA_LO, 0);
1121		CAS_WRITE_4(sc,	CAS_RX_REAS_DMA_DATA_MD, 0);
1122		CAS_WRITE_4(sc,	CAS_RX_REAS_DMA_DATA_HI, 0);
1123	}
1124
1125	/* Ensure the RX control FIFO and RX IPP FIFO addresses are zero. */
1126	CAS_WRITE_4(sc, CAS_RX_CTRL_FIFO, 0);
1127	CAS_WRITE_4(sc, CAS_RX_IPP_ADDR, 0);
1128
1129	/* Finally, enable RX DMA. */
1130	CAS_WRITE_4(sc, CAS_RX_CONF,
1131	    CAS_READ_4(sc, CAS_RX_CONF) | CAS_RX_CONF_RXDMA_EN);
1132
1133	/* step 11.  Configure Media. */
1134
1135	/* step 12.  RX_MAC Configuration Register */
1136	v = CAS_READ_4(sc, CAS_MAC_RX_CONF) & ~CAS_MAC_RX_CONF_STRPPAD;
1137	v |= CAS_MAC_RX_CONF_EN | CAS_MAC_RX_CONF_STRPFCS;
1138	CAS_WRITE_4(sc, CAS_MAC_RX_CONF, 0);
1139	CAS_BARRIER(sc, CAS_MAC_RX_CONF, 4,
1140	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
1141	if (!cas_bitwait(sc, CAS_MAC_RX_CONF, CAS_MAC_RX_CONF_EN, 0))
1142		device_printf(sc->sc_dev, "cannot configure RX MAC\n");
1143	CAS_WRITE_4(sc, CAS_MAC_RX_CONF, v);
1144
1145	/* step 13.  TX_MAC Configuration Register */
1146	v = CAS_READ_4(sc, CAS_MAC_TX_CONF);
1147	v |= CAS_MAC_TX_CONF_EN;
1148	CAS_WRITE_4(sc, CAS_MAC_TX_CONF, 0);
1149	CAS_BARRIER(sc, CAS_MAC_TX_CONF, 4,
1150	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
1151	if (!cas_bitwait(sc, CAS_MAC_TX_CONF, CAS_MAC_TX_CONF_EN, 0))
1152		device_printf(sc->sc_dev, "cannot configure TX MAC\n");
1153	CAS_WRITE_4(sc, CAS_MAC_TX_CONF, v);
1154
1155	/* step 14.  Issue Transmit Pending command. */
1156
1157	/* step 15.  Give the reciever a swift kick. */
1158	CAS_WRITE_4(sc, CAS_RX_KICK, CAS_NRXDESC - 4);
1159	CAS_WRITE_4(sc, CAS_RX_COMP_TAIL, 0);
1160	if ((sc->sc_flags & CAS_REG_PLUS) != 0)
1161		CAS_WRITE_4(sc, CAS_RX_KICK2, CAS_NRXDESC2 - 4);
1162
1163	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1164	ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1165
1166	mii_mediachg(sc->sc_mii);
1167
1168	/* Start the one second timer. */
1169	sc->sc_wdog_timer = 0;
1170	callout_reset(&sc->sc_tick_ch, hz, cas_tick, sc);
1171}
1172
1173static int
1174cas_load_txmbuf(struct cas_softc *sc, struct mbuf **m_head)
1175{
1176	bus_dma_segment_t txsegs[CAS_NTXSEGS];
1177	struct cas_txsoft *txs;
1178	struct ip *ip;
1179	struct mbuf *m;
1180	uint64_t cflags;
1181	int error, nexttx, nsegs, offset, seg;
1182
1183	CAS_LOCK_ASSERT(sc, MA_OWNED);
1184
1185	/* Get a work queue entry. */
1186	if ((txs = STAILQ_FIRST(&sc->sc_txfreeq)) == NULL) {
1187		/* Ran out of descriptors. */
1188		return (ENOBUFS);
1189	}
1190
1191	cflags = 0;
1192	if (((*m_head)->m_pkthdr.csum_flags & CAS_CSUM_FEATURES) != 0) {
1193		if (M_WRITABLE(*m_head) == 0) {
1194			m = m_dup(*m_head, M_DONTWAIT);
1195			m_freem(*m_head);
1196			*m_head = m;
1197			if (m == NULL)
1198				return (ENOBUFS);
1199		}
1200		offset = sizeof(struct ether_header);
1201		m = m_pullup(*m_head, offset + sizeof(struct ip));
1202		if (m == NULL) {
1203			*m_head = NULL;
1204			return (ENOBUFS);
1205		}
1206		ip = (struct ip *)(mtod(m, caddr_t) + offset);
1207		offset += (ip->ip_hl << 2);
1208		cflags = (offset << CAS_TD_CKSUM_START_SHFT) |
1209		    ((offset + m->m_pkthdr.csum_data) <<
1210		    CAS_TD_CKSUM_STUFF_SHFT) | CAS_TD_CKSUM_EN;
1211		*m_head = m;
1212	}
1213
1214	error = bus_dmamap_load_mbuf_sg(sc->sc_tdmatag, txs->txs_dmamap,
1215	    *m_head, txsegs, &nsegs, BUS_DMA_NOWAIT);
1216	if (error == EFBIG) {
1217		m = m_collapse(*m_head, M_DONTWAIT, CAS_NTXSEGS);
1218		if (m == NULL) {
1219			m_freem(*m_head);
1220			*m_head = NULL;
1221			return (ENOBUFS);
1222		}
1223		*m_head = m;
1224		error = bus_dmamap_load_mbuf_sg(sc->sc_tdmatag,
1225		    txs->txs_dmamap, *m_head, txsegs, &nsegs,
1226		    BUS_DMA_NOWAIT);
1227		if (error != 0) {
1228			m_freem(*m_head);
1229			*m_head = NULL;
1230			return (error);
1231		}
1232	} else if (error != 0)
1233		return (error);
1234	/* If nsegs is wrong then the stack is corrupt. */
1235	KASSERT(nsegs <= CAS_NTXSEGS,
1236	    ("%s: too many DMA segments (%d)", __func__, nsegs));
1237	if (nsegs == 0) {
1238		m_freem(*m_head);
1239		*m_head = NULL;
1240		return (EIO);
1241	}
1242
1243	/*
1244	 * Ensure we have enough descriptors free to describe
1245	 * the packet.  Note, we always reserve one descriptor
1246	 * at the end of the ring as a termination point, in
1247	 * order to prevent wrap-around.
1248	 */
1249	if (nsegs > sc->sc_txfree - 1) {
1250		txs->txs_ndescs = 0;
1251		bus_dmamap_unload(sc->sc_tdmatag, txs->txs_dmamap);
1252		return (ENOBUFS);
1253	}
1254
1255	txs->txs_ndescs = nsegs;
1256	txs->txs_firstdesc = sc->sc_txnext;
1257	nexttx = txs->txs_firstdesc;
1258	for (seg = 0; seg < nsegs; seg++, nexttx = CAS_NEXTTX(nexttx)) {
1259#ifdef CAS_DEBUG
1260		CTR6(KTR_CAS,
1261		    "%s: mapping seg %d (txd %d), len %lx, addr %#lx (%#lx)",
1262		    __func__, seg, nexttx, txsegs[seg].ds_len,
1263		    txsegs[seg].ds_addr, htole64(txsegs[seg].ds_addr));
1264#endif
1265		sc->sc_txdescs[nexttx].cd_buf_ptr =
1266		    htole64(txsegs[seg].ds_addr);
1267		KASSERT(txsegs[seg].ds_len <
1268		    CAS_TD_BUF_LEN_MASK >> CAS_TD_BUF_LEN_SHFT,
1269		    ("%s: segment size too large!", __func__));
1270		sc->sc_txdescs[nexttx].cd_flags =
1271		    htole64(txsegs[seg].ds_len << CAS_TD_BUF_LEN_SHFT);
1272		txs->txs_lastdesc = nexttx;
1273	}
1274
1275	/* Set EOF on the last descriptor. */
1276#ifdef CAS_DEBUG
1277	CTR3(KTR_CAS, "%s: end of frame at segment %d, TX %d",
1278	    __func__, seg, nexttx);
1279#endif
1280	sc->sc_txdescs[txs->txs_lastdesc].cd_flags |=
1281	    htole64(CAS_TD_END_OF_FRAME);
1282
1283	/* Lastly set SOF on the first descriptor. */
1284#ifdef CAS_DEBUG
1285	CTR3(KTR_CAS, "%s: start of frame at segment %d, TX %d",
1286	    __func__, seg, nexttx);
1287#endif
1288	if (sc->sc_txwin += nsegs > CAS_MAXTXFREE * 2 / 3) {
1289		sc->sc_txwin = 0;
1290		sc->sc_txdescs[txs->txs_firstdesc].cd_flags |=
1291		    htole64(cflags | CAS_TD_START_OF_FRAME | CAS_TD_INT_ME);
1292	} else
1293		sc->sc_txdescs[txs->txs_firstdesc].cd_flags |=
1294		    htole64(cflags | CAS_TD_START_OF_FRAME);
1295
1296	/* Sync the DMA map. */
1297	bus_dmamap_sync(sc->sc_tdmatag, txs->txs_dmamap,
1298	    BUS_DMASYNC_PREWRITE);
1299
1300#ifdef CAS_DEBUG
1301	CTR4(KTR_CAS, "%s: setting firstdesc=%d, lastdesc=%d, ndescs=%d",
1302	    __func__, txs->txs_firstdesc, txs->txs_lastdesc,
1303	    txs->txs_ndescs);
1304#endif
1305	STAILQ_REMOVE_HEAD(&sc->sc_txfreeq, txs_q);
1306	STAILQ_INSERT_TAIL(&sc->sc_txdirtyq, txs, txs_q);
1307	txs->txs_mbuf = *m_head;
1308
1309	sc->sc_txnext = CAS_NEXTTX(txs->txs_lastdesc);
1310	sc->sc_txfree -= txs->txs_ndescs;
1311
1312	return (0);
1313}
1314
1315static void
1316cas_init_regs(struct cas_softc *sc)
1317{
1318	int i;
1319	const u_char *laddr = IF_LLADDR(sc->sc_ifp);
1320
1321	CAS_LOCK_ASSERT(sc, MA_OWNED);
1322
1323	/* These registers are not cleared on reset. */
1324	if ((sc->sc_flags & CAS_INITED) == 0) {
1325		/* magic values */
1326		CAS_WRITE_4(sc, CAS_MAC_IPG0, 0);
1327		CAS_WRITE_4(sc, CAS_MAC_IPG1, 8);
1328		CAS_WRITE_4(sc, CAS_MAC_IPG2, 4);
1329
1330		/* min frame length */
1331		CAS_WRITE_4(sc, CAS_MAC_MIN_FRAME, ETHER_MIN_LEN);
1332		/* max frame length and max burst size */
1333		CAS_WRITE_4(sc, CAS_MAC_MAX_BF,
1334		    ((ETHER_MAX_LEN_JUMBO + ETHER_VLAN_ENCAP_LEN) <<
1335		    CAS_MAC_MAX_BF_FRM_SHFT) |
1336		    (0x2000 << CAS_MAC_MAX_BF_BST_SHFT));
1337
1338		/* more magic values */
1339		CAS_WRITE_4(sc, CAS_MAC_PREAMBLE_LEN, 0x7);
1340		CAS_WRITE_4(sc, CAS_MAC_JAM_SIZE, 0x4);
1341		CAS_WRITE_4(sc, CAS_MAC_ATTEMPT_LIMIT, 0x10);
1342		CAS_WRITE_4(sc, CAS_MAC_CTRL_TYPE, 0x8088);
1343
1344		/* random number seed */
1345		CAS_WRITE_4(sc, CAS_MAC_RANDOM_SEED,
1346		    ((laddr[5] << 8) | laddr[4]) & 0x3ff);
1347
1348		/* secondary MAC addresses: 0:0:0:0:0:0 */
1349		for (i = CAS_MAC_ADDR3; i <= CAS_MAC_ADDR41;
1350		    i += CAS_MAC_ADDR4 - CAS_MAC_ADDR3)
1351			CAS_WRITE_4(sc, i, 0);
1352
1353		/* MAC control address: 01:80:c2:00:00:01 */
1354		CAS_WRITE_4(sc, CAS_MAC_ADDR42, 0x0001);
1355		CAS_WRITE_4(sc, CAS_MAC_ADDR43, 0xc200);
1356		CAS_WRITE_4(sc, CAS_MAC_ADDR44, 0x0180);
1357
1358		/* MAC filter address: 0:0:0:0:0:0 */
1359		CAS_WRITE_4(sc, CAS_MAC_AFILTER0, 0);
1360		CAS_WRITE_4(sc, CAS_MAC_AFILTER1, 0);
1361		CAS_WRITE_4(sc, CAS_MAC_AFILTER2, 0);
1362		CAS_WRITE_4(sc, CAS_MAC_AFILTER_MASK1_2, 0);
1363		CAS_WRITE_4(sc, CAS_MAC_AFILTER_MASK0, 0);
1364
1365		/* Zero the hash table. */
1366		for (i = CAS_MAC_HASH0; i <= CAS_MAC_HASH15;
1367		    i += CAS_MAC_HASH1 - CAS_MAC_HASH0)
1368			CAS_WRITE_4(sc, i, 0);
1369
1370		sc->sc_flags |= CAS_INITED;
1371	}
1372
1373	/* Counters need to be zeroed. */
1374	CAS_WRITE_4(sc, CAS_MAC_NORM_COLL_CNT, 0);
1375	CAS_WRITE_4(sc, CAS_MAC_FIRST_COLL_CNT, 0);
1376	CAS_WRITE_4(sc, CAS_MAC_EXCESS_COLL_CNT, 0);
1377	CAS_WRITE_4(sc, CAS_MAC_LATE_COLL_CNT, 0);
1378	CAS_WRITE_4(sc, CAS_MAC_DEFER_TMR_CNT, 0);
1379	CAS_WRITE_4(sc, CAS_MAC_PEAK_ATTEMPTS, 0);
1380	CAS_WRITE_4(sc, CAS_MAC_RX_FRAME_COUNT, 0);
1381	CAS_WRITE_4(sc, CAS_MAC_RX_LEN_ERR_CNT, 0);
1382	CAS_WRITE_4(sc, CAS_MAC_RX_ALIGN_ERR, 0);
1383	CAS_WRITE_4(sc, CAS_MAC_RX_CRC_ERR_CNT, 0);
1384	CAS_WRITE_4(sc, CAS_MAC_RX_CODE_VIOL, 0);
1385
1386	/* Set XOFF PAUSE time. */
1387	CAS_WRITE_4(sc, CAS_MAC_SPC, 0x1BF0 << CAS_MAC_SPC_TIME_SHFT);
1388
1389	/* Set the station address. */
1390	CAS_WRITE_4(sc, CAS_MAC_ADDR0, (laddr[4] << 8) | laddr[5]);
1391	CAS_WRITE_4(sc, CAS_MAC_ADDR1, (laddr[2] << 8) | laddr[3]);
1392	CAS_WRITE_4(sc, CAS_MAC_ADDR2, (laddr[0] << 8) | laddr[1]);
1393
1394	/* Enable MII outputs. */
1395	CAS_WRITE_4(sc, CAS_MAC_XIF_CONF, CAS_MAC_XIF_CONF_TX_OE);
1396}
1397
1398static void
1399cas_tx_task(void *arg, int pending __unused)
1400{
1401	struct ifnet *ifp;
1402
1403	ifp = (struct ifnet *)arg;
1404	cas_start(ifp);
1405}
1406
1407static inline void
1408cas_txkick(struct cas_softc *sc)
1409{
1410
1411	/*
1412	 * Update the TX kick register.  This register has to point to the
1413	 * descriptor after the last valid one and for optimum performance
1414	 * should be incremented in multiples of 4 (the DMA engine fetches/
1415	 * updates descriptors in batches of 4).
1416	 */
1417#ifdef CAS_DEBUG
1418	CTR3(KTR_CAS, "%s: %s: kicking TX %d",
1419	    device_get_name(sc->sc_dev), __func__, sc->sc_txnext);
1420#endif
1421	CAS_CDSYNC(sc, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1422	CAS_WRITE_4(sc, CAS_TX_KICK3, sc->sc_txnext);
1423}
1424
1425static void
1426cas_start(struct ifnet *ifp)
1427{
1428	struct cas_softc *sc = ifp->if_softc;
1429	struct mbuf *m;
1430	int kicked, ntx;
1431
1432	CAS_LOCK(sc);
1433
1434	if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
1435	    IFF_DRV_RUNNING || (sc->sc_flags & CAS_LINK) == 0) {
1436		CAS_UNLOCK(sc);
1437		return;
1438	}
1439
1440	if (sc->sc_txfree < CAS_MAXTXFREE / 4)
1441		cas_tint(sc);
1442
1443#ifdef CAS_DEBUG
1444	CTR4(KTR_CAS, "%s: %s: txfree %d, txnext %d",
1445	    device_get_name(sc->sc_dev), __func__, sc->sc_txfree,
1446	    sc->sc_txnext);
1447#endif
1448	ntx = 0;
1449	kicked = 0;
1450	for (; !IFQ_DRV_IS_EMPTY(&ifp->if_snd) && sc->sc_txfree > 1;) {
1451		IFQ_DRV_DEQUEUE(&ifp->if_snd, m);
1452		if (m == NULL)
1453			break;
1454		if (cas_load_txmbuf(sc, &m) != 0) {
1455			if (m == NULL)
1456				break;
1457			ifp->if_drv_flags |= IFF_DRV_OACTIVE;
1458			IFQ_DRV_PREPEND(&ifp->if_snd, m);
1459			break;
1460		}
1461		if ((sc->sc_txnext % 4) == 0) {
1462			cas_txkick(sc);
1463			kicked = 1;
1464		} else
1465			kicked = 0;
1466		ntx++;
1467		BPF_MTAP(ifp, m);
1468	}
1469
1470	if (ntx > 0) {
1471		if (kicked == 0)
1472			cas_txkick(sc);
1473#ifdef CAS_DEBUG
1474		CTR2(KTR_CAS, "%s: packets enqueued, OWN on %d",
1475		    device_get_name(sc->sc_dev), sc->sc_txnext);
1476#endif
1477
1478		/* Set a watchdog timer in case the chip flakes out. */
1479		sc->sc_wdog_timer = 5;
1480#ifdef CAS_DEBUG
1481		CTR3(KTR_CAS, "%s: %s: watchdog %d",
1482		    device_get_name(sc->sc_dev), __func__,
1483		    sc->sc_wdog_timer);
1484#endif
1485	}
1486
1487	CAS_UNLOCK(sc);
1488}
1489
1490static void
1491cas_tint(struct cas_softc *sc)
1492{
1493	struct ifnet *ifp = sc->sc_ifp;
1494	struct cas_txsoft *txs;
1495	int progress;
1496	uint32_t txlast;
1497#ifdef CAS_DEBUG
1498	int i;
1499
1500	CAS_LOCK_ASSERT(sc, MA_OWNED);
1501
1502	CTR2(KTR_CAS, "%s: %s", device_get_name(sc->sc_dev), __func__);
1503#endif
1504
1505	/*
1506	 * Go through our TX list and free mbufs for those
1507	 * frames that have been transmitted.
1508	 */
1509	progress = 0;
1510	CAS_CDSYNC(sc, BUS_DMASYNC_POSTREAD);
1511	while ((txs = STAILQ_FIRST(&sc->sc_txdirtyq)) != NULL) {
1512#ifdef CAS_DEBUG
1513		if ((ifp->if_flags & IFF_DEBUG) != 0) {
1514			printf("    txsoft %p transmit chain:\n", txs);
1515			for (i = txs->txs_firstdesc;; i = CAS_NEXTTX(i)) {
1516				printf("descriptor %d: ", i);
1517				printf("cd_flags: 0x%016llx\t",
1518				    (long long)le64toh(
1519				    sc->sc_txdescs[i].cd_flags));
1520				printf("cd_buf_ptr: 0x%016llx\n",
1521				    (long long)le64toh(
1522				    sc->sc_txdescs[i].cd_buf_ptr));
1523				if (i == txs->txs_lastdesc)
1524					break;
1525			}
1526		}
1527#endif
1528
1529		/*
1530		 * In theory, we could harvest some descriptors before
1531		 * the ring is empty, but that's a bit complicated.
1532		 *
1533		 * CAS_TX_COMPn points to the last descriptor
1534		 * processed + 1.
1535		 */
1536		txlast = CAS_READ_4(sc, CAS_TX_COMP3);
1537#ifdef CAS_DEBUG
1538		CTR4(KTR_CAS, "%s: txs->txs_firstdesc = %d, "
1539		    "txs->txs_lastdesc = %d, txlast = %d",
1540		    __func__, txs->txs_firstdesc, txs->txs_lastdesc, txlast);
1541#endif
1542		if (txs->txs_firstdesc <= txs->txs_lastdesc) {
1543			if ((txlast >= txs->txs_firstdesc) &&
1544			    (txlast <= txs->txs_lastdesc))
1545				break;
1546		} else {
1547			/* Ick -- this command wraps. */
1548			if ((txlast >= txs->txs_firstdesc) ||
1549			    (txlast <= txs->txs_lastdesc))
1550				break;
1551		}
1552
1553#ifdef CAS_DEBUG
1554		CTR1(KTR_CAS, "%s: releasing a descriptor", __func__);
1555#endif
1556		STAILQ_REMOVE_HEAD(&sc->sc_txdirtyq, txs_q);
1557
1558		sc->sc_txfree += txs->txs_ndescs;
1559
1560		bus_dmamap_sync(sc->sc_tdmatag, txs->txs_dmamap,
1561		    BUS_DMASYNC_POSTWRITE);
1562		bus_dmamap_unload(sc->sc_tdmatag, txs->txs_dmamap);
1563		if (txs->txs_mbuf != NULL) {
1564			m_freem(txs->txs_mbuf);
1565			txs->txs_mbuf = NULL;
1566		}
1567
1568		STAILQ_INSERT_TAIL(&sc->sc_txfreeq, txs, txs_q);
1569
1570		ifp->if_opackets++;
1571		progress = 1;
1572	}
1573
1574#ifdef CAS_DEBUG
1575	CTR4(KTR_CAS, "%s: CAS_TX_STATE_MACHINE %x CAS_TX_DESC_BASE %llx "
1576	    "CAS_TX_COMP3 %x",
1577	    __func__, CAS_READ_4(sc, CAS_TX_STATE_MACHINE),
1578	    ((long long)CAS_READ_4(sc, CAS_TX_DESC_BASE_HI3) << 32) |
1579	    CAS_READ_4(sc, CAS_TX_DESC_BASE_LO3),
1580	    CAS_READ_4(sc, CAS_TX_COMP3));
1581#endif
1582
1583	if (progress) {
1584		/* We freed some descriptors, so reset IFF_DRV_OACTIVE. */
1585		ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1586		if (STAILQ_EMPTY(&sc->sc_txdirtyq))
1587			sc->sc_wdog_timer = 0;
1588	}
1589
1590#ifdef CAS_DEBUG
1591	CTR3(KTR_CAS, "%s: %s: watchdog %d",
1592	    device_get_name(sc->sc_dev), __func__, sc->sc_wdog_timer);
1593#endif
1594}
1595
1596static void
1597cas_rint_timeout(void *arg)
1598{
1599	struct cas_softc *sc = arg;
1600
1601	CAS_LOCK_ASSERT(sc, MA_NOTOWNED);
1602
1603	cas_rint(sc);
1604}
1605
1606static void
1607cas_rint(struct cas_softc *sc)
1608{
1609	struct cas_rxdsoft *rxds, *rxds2;
1610	struct ifnet *ifp = sc->sc_ifp;
1611	struct mbuf *m, *m2;
1612	uint64_t word1, word2, word3, word4;
1613	uint32_t rxhead;
1614	u_int idx, idx2, len, off, skip;
1615
1616	CAS_LOCK_ASSERT(sc, MA_NOTOWNED);
1617
1618	callout_stop(&sc->sc_rx_ch);
1619
1620#ifdef CAS_DEBUG
1621	CTR2(KTR_CAS, "%s: %s", device_get_name(sc->sc_dev), __func__);
1622#endif
1623
1624#define	PRINTWORD(n, delimiter)						\
1625	printf("word ## n: 0x%016llx%c", (long long)word ## n, delimiter)
1626
1627#define	SKIPASSERT(n)							\
1628	KASSERT(sc->sc_rxcomps[sc->sc_rxcptr].crc_word ## n == 0,	\
1629	    ("%s: word ## n not 0", __func__))
1630
1631#define	WORDTOH(n)							\
1632	word ## n = le64toh(sc->sc_rxcomps[sc->sc_rxcptr].crc_word ## n)
1633
1634	/*
1635	 * Read the completion head register once.  This limits
1636	 * how long the following loop can execute.
1637	 */
1638	rxhead = CAS_READ_4(sc, CAS_RX_COMP_HEAD);
1639#ifdef CAS_DEBUG
1640	CTR4(KTR_CAS, "%s: sc->sc_rxcptr %d, sc->sc_rxdptr %d, head %d",
1641	    __func__, sc->rxcptr, sc->sc_rxdptr, rxhead);
1642#endif
1643	skip = 0;
1644	CAS_CDSYNC(sc, BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
1645	for (; sc->sc_rxcptr != rxhead;
1646	    sc->sc_rxcptr = CAS_NEXTRXCOMP(sc->sc_rxcptr)) {
1647		if (skip != 0) {
1648			SKIPASSERT(1);
1649			SKIPASSERT(2);
1650			SKIPASSERT(3);
1651
1652			--skip;
1653			goto skip;
1654		}
1655
1656		WORDTOH(1);
1657		WORDTOH(2);
1658		WORDTOH(3);
1659		WORDTOH(4);
1660
1661#ifdef CAS_DEBUG
1662		if ((ifp->if_flags & IFF_DEBUG) != 0) {
1663			printf("    completion %d: ", sc->sc_rxcptr);
1664			PRINTWORD(1, '\t');
1665			PRINTWORD(2, '\t');
1666			PRINTWORD(3, '\t');
1667			PRINTWORD(4, '\n');
1668		}
1669#endif
1670
1671		if (__predict_false(
1672		    (word1 & CAS_RC1_TYPE_MASK) == CAS_RC1_TYPE_HW ||
1673		    (word4 & CAS_RC4_ZERO) != 0)) {
1674			/*
1675			 * The descriptor is still marked as owned, although
1676			 * it is supposed to have completed.  This has been
1677			 * observed on some machines.  Just exiting here
1678			 * might leave the packet sitting around until another
1679			 * one arrives to trigger a new interrupt, which is
1680			 * generally undesirable, so set up a timeout.
1681			 */
1682			callout_reset(&sc->sc_rx_ch, CAS_RXOWN_TICKS,
1683			    cas_rint_timeout, sc);
1684			break;
1685		}
1686
1687		if (__predict_false(
1688		    (word4 & (CAS_RC4_BAD | CAS_RC4_LEN_MMATCH)) != 0)) {
1689			ifp->if_ierrors++;
1690			device_printf(sc->sc_dev,
1691			    "receive error: CRC error\n");
1692			continue;
1693		}
1694
1695		KASSERT(CAS_GET(word1, CAS_RC1_DATA_SIZE) == 0 ||
1696		    CAS_GET(word2, CAS_RC2_HDR_SIZE) == 0,
1697		    ("%s: data and header present", __func__));
1698		KASSERT((word1 & CAS_RC1_SPLIT_PKT) == 0 ||
1699		    CAS_GET(word2, CAS_RC2_HDR_SIZE) == 0,
1700		    ("%s: split and header present", __func__));
1701		KASSERT(CAS_GET(word1, CAS_RC1_DATA_SIZE) == 0 ||
1702		    (word1 & CAS_RC1_RELEASE_HDR) == 0,
1703		    ("%s: data present but header release", __func__));
1704		KASSERT(CAS_GET(word2, CAS_RC2_HDR_SIZE) == 0 ||
1705		    (word1 & CAS_RC1_RELEASE_DATA) == 0,
1706		    ("%s: header present but data release", __func__));
1707
1708		if ((len = CAS_GET(word2, CAS_RC2_HDR_SIZE)) != 0) {
1709			idx = CAS_GET(word2, CAS_RC2_HDR_INDEX);
1710			off = CAS_GET(word2, CAS_RC2_HDR_OFF);
1711#ifdef CAS_DEBUG
1712			CTR4(KTR_CAS, "%s: hdr at idx %d, off %d, len %d",
1713			    __func__, idx, off, len);
1714#endif
1715			rxds = &sc->sc_rxdsoft[idx];
1716			MGETHDR(m, M_DONTWAIT, MT_DATA);
1717			if (m != NULL) {
1718				refcount_acquire(&rxds->rxds_refcount);
1719				bus_dmamap_sync(sc->sc_rdmatag,
1720				    rxds->rxds_dmamap, BUS_DMASYNC_POSTREAD);
1721#if __FreeBSD_version < 800016
1722				MEXTADD(m, (caddr_t)rxds->rxds_buf +
1723				    off * 256 + ETHER_ALIGN, len, cas_free,
1724				    rxds, M_RDONLY, EXT_NET_DRV);
1725#else
1726				MEXTADD(m, (caddr_t)rxds->rxds_buf +
1727				    off * 256 + ETHER_ALIGN, len, cas_free,
1728				    sc, (void *)(uintptr_t)idx,
1729				    M_RDONLY, EXT_NET_DRV);
1730#endif
1731				if ((m->m_flags & M_EXT) == 0) {
1732					m_freem(m);
1733					m = NULL;
1734				}
1735			}
1736			if (m != NULL) {
1737				m->m_pkthdr.rcvif = ifp;
1738				m->m_pkthdr.len = m->m_len = len;
1739				ifp->if_ipackets++;
1740				if ((ifp->if_capenable & IFCAP_RXCSUM) != 0)
1741					cas_rxcksum(m, CAS_GET(word4,
1742					    CAS_RC4_TCP_CSUM));
1743				/* Pass it on. */
1744				(*ifp->if_input)(ifp, m);
1745			} else
1746				ifp->if_ierrors++;
1747
1748			if ((word1 & CAS_RC1_RELEASE_HDR) != 0 &&
1749			    refcount_release(&rxds->rxds_refcount) != 0)
1750				cas_add_rxdesc(sc, idx);
1751		} else if ((len = CAS_GET(word1, CAS_RC1_DATA_SIZE)) != 0) {
1752			idx = CAS_GET(word1, CAS_RC1_DATA_INDEX);
1753			off = CAS_GET(word1, CAS_RC1_DATA_OFF);
1754#ifdef CAS_DEBUG
1755			CTR4(KTR_CAS, "%s: data at idx %d, off %d, len %d",
1756			    __func__, idx, off, len);
1757#endif
1758			rxds = &sc->sc_rxdsoft[idx];
1759			MGETHDR(m, M_DONTWAIT, MT_DATA);
1760			if (m != NULL) {
1761				refcount_acquire(&rxds->rxds_refcount);
1762				off += ETHER_ALIGN;
1763				m->m_len = min(CAS_PAGE_SIZE - off, len);
1764				bus_dmamap_sync(sc->sc_rdmatag,
1765				    rxds->rxds_dmamap, BUS_DMASYNC_POSTREAD);
1766#if __FreeBSD_version < 800016
1767				MEXTADD(m, (caddr_t)rxds->rxds_buf + off,
1768				    m->m_len, cas_free, rxds, M_RDONLY,
1769				    EXT_NET_DRV);
1770#else
1771				MEXTADD(m, (caddr_t)rxds->rxds_buf + off,
1772				    m->m_len, cas_free, sc,
1773				    (void *)(uintptr_t)idx, M_RDONLY,
1774				    EXT_NET_DRV);
1775#endif
1776				if ((m->m_flags & M_EXT) == 0) {
1777					m_freem(m);
1778					m = NULL;
1779				}
1780			}
1781			idx2 = 0;
1782			m2 = NULL;
1783			rxds2 = NULL;
1784			if ((word1 & CAS_RC1_SPLIT_PKT) != 0) {
1785				KASSERT((word1 & CAS_RC1_RELEASE_NEXT) != 0,
1786				    ("%s: split but no release next",
1787				    __func__));
1788
1789				idx2 = CAS_GET(word2, CAS_RC2_NEXT_INDEX);
1790#ifdef CAS_DEBUG
1791				CTR2(KTR_CAS, "%s: split at idx %d",
1792				    __func__, idx2);
1793#endif
1794				rxds2 = &sc->sc_rxdsoft[idx2];
1795				if (m != NULL) {
1796					MGET(m2, M_DONTWAIT, MT_DATA);
1797					if (m2 != NULL) {
1798						refcount_acquire(
1799						    &rxds2->rxds_refcount);
1800						m2->m_len = len - m->m_len;
1801						bus_dmamap_sync(
1802						    sc->sc_rdmatag,
1803						    rxds2->rxds_dmamap,
1804						    BUS_DMASYNC_POSTREAD);
1805#if __FreeBSD_version < 800016
1806						MEXTADD(m2,
1807						    (caddr_t)rxds2->rxds_buf,
1808						    m2->m_len, cas_free,
1809						    rxds2, M_RDONLY,
1810						    EXT_NET_DRV);
1811#else
1812						MEXTADD(m2,
1813						    (caddr_t)rxds2->rxds_buf,
1814						    m2->m_len, cas_free, sc,
1815						    (void *)(uintptr_t)idx2,
1816						    M_RDONLY, EXT_NET_DRV);
1817#endif
1818						if ((m2->m_flags & M_EXT) ==
1819						    0) {
1820							m_freem(m2);
1821							m2 = NULL;
1822						}
1823					}
1824				}
1825				if (m2 != NULL)
1826					m->m_next = m2;
1827				else if (m != NULL) {
1828					m_freem(m);
1829					m = NULL;
1830				}
1831			}
1832			if (m != NULL) {
1833				m->m_pkthdr.rcvif = ifp;
1834				m->m_pkthdr.len = len;
1835				ifp->if_ipackets++;
1836				if ((ifp->if_capenable & IFCAP_RXCSUM) != 0)
1837					cas_rxcksum(m, CAS_GET(word4,
1838					    CAS_RC4_TCP_CSUM));
1839				/* Pass it on. */
1840				(*ifp->if_input)(ifp, m);
1841			} else
1842				ifp->if_ierrors++;
1843
1844			if ((word1 & CAS_RC1_RELEASE_DATA) != 0 &&
1845			    refcount_release(&rxds->rxds_refcount) != 0)
1846				cas_add_rxdesc(sc, idx);
1847			if ((word1 & CAS_RC1_SPLIT_PKT) != 0 &&
1848			    refcount_release(&rxds2->rxds_refcount) != 0)
1849				cas_add_rxdesc(sc, idx2);
1850		}
1851
1852		skip = CAS_GET(word1, CAS_RC1_SKIP);
1853
1854 skip:
1855		cas_rxcompinit(&sc->sc_rxcomps[sc->sc_rxcptr]);
1856		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1857			break;
1858	}
1859	CAS_CDSYNC(sc, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1860	CAS_WRITE_4(sc, CAS_RX_COMP_TAIL, sc->sc_rxcptr);
1861
1862#undef PRINTWORD
1863#undef SKIPASSERT
1864#undef WORDTOH
1865
1866#ifdef CAS_DEBUG
1867	CTR4(KTR_CAS, "%s: done sc->sc_rxcptr %d, sc->sc_rxdptr %d, head %d",
1868	    __func__, sc->rxcptr, sc->sc_rxdptr,
1869	    CAS_READ_4(sc, CAS_RX_COMP_HEAD));
1870#endif
1871}
1872
1873static void
1874cas_free(void *arg1, void *arg2)
1875{
1876	struct cas_rxdsoft *rxds;
1877	struct cas_softc *sc;
1878	u_int idx;
1879
1880#if __FreeBSD_version < 800016
1881	rxds = arg2;
1882	sc = rxds->rxds_sc;
1883	idx = rxds->rxds_idx;
1884#else
1885	sc = arg1;
1886	idx = (uintptr_t)arg2;
1887	rxds = &sc->sc_rxdsoft[idx];
1888#endif
1889	if (refcount_release(&rxds->rxds_refcount) == 0)
1890		return;
1891
1892	/*
1893	 * NB: this function can be called via m_freem(9) within
1894	 * this driver!
1895	 */
1896
1897	cas_add_rxdesc(sc, idx);
1898}
1899
1900static inline void
1901cas_add_rxdesc(struct cas_softc *sc, u_int idx)
1902{
1903	u_int locked;
1904
1905	if ((locked = CAS_LOCK_OWNED(sc)) == 0)
1906		CAS_LOCK(sc);
1907
1908	bus_dmamap_sync(sc->sc_rdmatag, sc->sc_rxdsoft[idx].rxds_dmamap,
1909	    BUS_DMASYNC_PREREAD);
1910	CAS_UPDATE_RXDESC(sc, sc->sc_rxdptr, idx);
1911	sc->sc_rxdptr = CAS_NEXTRXDESC(sc->sc_rxdptr);
1912
1913	/*
1914	 * Update the RX kick register.  This register has to point to the
1915	 * descriptor after the last valid one (before the current batch)
1916	 * and for optimum performance should be incremented in multiples
1917	 * of 4 (the DMA engine fetches/updates descriptors in batches of 4).
1918	 */
1919	if ((sc->sc_rxdptr % 4) == 0) {
1920		CAS_CDSYNC(sc, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1921		CAS_WRITE_4(sc, CAS_RX_KICK,
1922		    (sc->sc_rxdptr + CAS_NRXDESC - 4) & CAS_NRXDESC_MASK);
1923	}
1924
1925	if (locked == 0)
1926		CAS_UNLOCK(sc);
1927}
1928
1929static void
1930cas_eint(struct cas_softc *sc, u_int status)
1931{
1932	struct ifnet *ifp = sc->sc_ifp;
1933
1934	CAS_LOCK_ASSERT(sc, MA_NOTOWNED);
1935
1936	ifp->if_ierrors++;
1937
1938	device_printf(sc->sc_dev, "%s: status 0x%x", __func__, status);
1939	if ((status & CAS_INTR_PCI_ERROR_INT) != 0) {
1940		status = CAS_READ_4(sc, CAS_ERROR_STATUS);
1941		printf(", PCI bus error 0x%x", status);
1942		if ((status & CAS_ERROR_OTHER) != 0) {
1943			status = pci_read_config(sc->sc_dev, PCIR_STATUS, 2);
1944			printf(", PCI status 0x%x", status);
1945			pci_write_config(sc->sc_dev, PCIR_STATUS, status, 2);
1946		}
1947	}
1948	printf("\n");
1949
1950	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1951	cas_init(sc);
1952	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
1953		taskqueue_enqueue(sc->sc_tq, &sc->sc_tx_task);
1954}
1955
1956static int
1957cas_intr(void *v)
1958{
1959	struct cas_softc *sc = v;
1960
1961	if (__predict_false((CAS_READ_4(sc, CAS_STATUS_ALIAS) &
1962	    CAS_INTR_SUMMARY) == 0))
1963		return (FILTER_STRAY);
1964
1965	/* Disable interrupts. */
1966	CAS_WRITE_4(sc, CAS_INTMASK, 0xffffffff);
1967	taskqueue_enqueue(sc->sc_tq, &sc->sc_intr_task);
1968
1969	return (FILTER_HANDLED);
1970}
1971
1972static void
1973cas_intr_task(void *arg, int pending __unused)
1974{
1975	struct cas_softc *sc = arg;
1976	struct ifnet *ifp = sc->sc_ifp;
1977	uint32_t status, status2;
1978
1979	CAS_LOCK_ASSERT(sc, MA_NOTOWNED);
1980
1981	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1982		return;
1983
1984	status = CAS_READ_4(sc, CAS_STATUS);
1985	if (__predict_false((status & CAS_INTR_SUMMARY) == 0))
1986		goto done;
1987
1988#ifdef CAS_DEBUG
1989	CTR4(KTR_CAS, "%s: %s: cplt %x, status %x",
1990	    device_get_name(sc->sc_dev), __func__,
1991	    (status >> CAS_STATUS_TX_COMP3_SHIFT), (u_int)status);
1992
1993	/*
1994	 * PCS interrupts must be cleared, otherwise no traffic is passed!
1995	 */
1996	if ((status & CAS_INTR_PCS_INT) != 0) {
1997		status2 =
1998		    CAS_READ_4(sc, CAS_PCS_INTR_STATUS) |
1999		    CAS_READ_4(sc, CAS_PCS_INTR_STATUS);
2000		if ((status2 & CAS_PCS_INTR_LINK) != 0)
2001			device_printf(sc->sc_dev,
2002			    "%s: PCS link status changed\n", __func__);
2003	}
2004	if ((status & CAS_MAC_CTRL_STATUS) != 0) {
2005		status2 = CAS_READ_4(sc, CAS_MAC_CTRL_STATUS);
2006		if ((status2 & CAS_MAC_CTRL_PAUSE) != 0)
2007			device_printf(sc->sc_dev,
2008			    "%s: PAUSE received (PAUSE time %d slots)\n",
2009			    __func__,
2010			    (status2 & CAS_MAC_CTRL_STATUS_PT_MASK) >>
2011			    CAS_MAC_CTRL_STATUS_PT_SHFT);
2012		if ((status2 & CAS_MAC_CTRL_PAUSE) != 0)
2013			device_printf(sc->sc_dev,
2014			    "%s: transited to PAUSE state\n", __func__);
2015		if ((status2 & CAS_MAC_CTRL_NON_PAUSE) != 0)
2016			device_printf(sc->sc_dev,
2017			    "%s: transited to non-PAUSE state\n", __func__);
2018	}
2019	if ((status & CAS_INTR_MIF) != 0)
2020		device_printf(sc->sc_dev, "%s: MIF interrupt\n", __func__);
2021#endif
2022
2023	if (__predict_false((status &
2024	    (CAS_INTR_TX_TAG_ERR | CAS_INTR_RX_TAG_ERR |
2025	    CAS_INTR_RX_LEN_MMATCH | CAS_INTR_PCI_ERROR_INT)) != 0)) {
2026		cas_eint(sc, status);
2027		return;
2028	}
2029
2030	if (__predict_false(status & CAS_INTR_TX_MAC_INT)) {
2031		status2 = CAS_READ_4(sc, CAS_MAC_TX_STATUS);
2032		if ((status2 &
2033		    (CAS_MAC_TX_UNDERRUN | CAS_MAC_TX_MAX_PKT_ERR)) != 0)
2034			sc->sc_ifp->if_oerrors++;
2035		else if ((status2 & ~CAS_MAC_TX_FRAME_XMTD) != 0)
2036			device_printf(sc->sc_dev,
2037			    "MAC TX fault, status %x\n", status2);
2038	}
2039
2040	if (__predict_false(status & CAS_INTR_RX_MAC_INT)) {
2041		status2 = CAS_READ_4(sc, CAS_MAC_RX_STATUS);
2042		if ((status2 & CAS_MAC_RX_OVERFLOW) != 0)
2043			sc->sc_ifp->if_ierrors++;
2044		else if ((status2 & ~CAS_MAC_RX_FRAME_RCVD) != 0)
2045			device_printf(sc->sc_dev,
2046			    "MAC RX fault, status %x\n", status2);
2047	}
2048
2049	if ((status &
2050	    (CAS_INTR_RX_DONE | CAS_INTR_RX_BUF_NA | CAS_INTR_RX_COMP_FULL |
2051	    CAS_INTR_RX_BUF_AEMPTY | CAS_INTR_RX_COMP_AFULL)) != 0) {
2052		cas_rint(sc);
2053#ifdef CAS_DEBUG
2054		if (__predict_false((status &
2055		    (CAS_INTR_RX_BUF_NA | CAS_INTR_RX_COMP_FULL |
2056		    CAS_INTR_RX_BUF_AEMPTY | CAS_INTR_RX_COMP_AFULL)) != 0))
2057			device_printf(sc->sc_dev,
2058			    "RX fault, status %x\n", status);
2059#endif
2060	}
2061
2062	if ((status &
2063	    (CAS_INTR_TX_INT_ME | CAS_INTR_TX_ALL | CAS_INTR_TX_DONE)) != 0) {
2064		CAS_LOCK(sc);
2065		cas_tint(sc);
2066		CAS_UNLOCK(sc);
2067	}
2068
2069	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
2070		return;
2071	else if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2072		taskqueue_enqueue(sc->sc_tq, &sc->sc_tx_task);
2073
2074	status = CAS_READ_4(sc, CAS_STATUS_ALIAS);
2075	if (__predict_false((status & CAS_INTR_SUMMARY) != 0)) {
2076		taskqueue_enqueue(sc->sc_tq, &sc->sc_intr_task);
2077		return;
2078	}
2079
2080 done:
2081	/* Re-enable interrupts. */
2082	CAS_WRITE_4(sc, CAS_INTMASK,
2083	    ~(CAS_INTR_TX_INT_ME | CAS_INTR_TX_TAG_ERR |
2084	    CAS_INTR_RX_DONE | CAS_INTR_RX_BUF_NA | CAS_INTR_RX_TAG_ERR |
2085	    CAS_INTR_RX_COMP_FULL | CAS_INTR_RX_BUF_AEMPTY |
2086	    CAS_INTR_RX_COMP_AFULL | CAS_INTR_RX_LEN_MMATCH |
2087	    CAS_INTR_PCI_ERROR_INT
2088#ifdef CAS_DEBUG
2089	    | CAS_INTR_PCS_INT | CAS_INTR_MIF
2090#endif
2091	));
2092}
2093
2094static void
2095cas_watchdog(struct cas_softc *sc)
2096{
2097	struct ifnet *ifp = sc->sc_ifp;
2098
2099	CAS_LOCK_ASSERT(sc, MA_OWNED);
2100
2101#ifdef CAS_DEBUG
2102	CTR4(KTR_CAS,
2103	    "%s: CAS_RX_CONFIG %x CAS_MAC_RX_STATUS %x CAS_MAC_RX_CONFIG %x",
2104	    __func__, CAS_READ_4(sc, CAS_RX_CONFIG),
2105	    CAS_READ_4(sc, CAS_MAC_RX_STATUS),
2106	    CAS_READ_4(sc, CAS_MAC_RX_CONFIG));
2107	CTR4(KTR_CAS,
2108	    "%s: CAS_TX_CONFIG %x CAS_MAC_TX_STATUS %x CAS_MAC_TX_CONFIG %x",
2109	    __func__, CAS_READ_4(sc, CAS_TX_CONFIG),
2110	    CAS_READ_4(sc, CAS_MAC_TX_STATUS),
2111	    CAS_READ_4(sc, CAS_MAC_TX_CONFIG));
2112#endif
2113
2114	if (sc->sc_wdog_timer == 0 || --sc->sc_wdog_timer != 0)
2115		return;
2116
2117	if ((sc->sc_flags & CAS_LINK) != 0)
2118		device_printf(sc->sc_dev, "device timeout\n");
2119	else if (bootverbose)
2120		device_printf(sc->sc_dev, "device timeout (no link)\n");
2121	++ifp->if_oerrors;
2122
2123	/* Try to get more packets going. */
2124	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2125	cas_init_locked(sc);
2126	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2127		taskqueue_enqueue(sc->sc_tq, &sc->sc_tx_task);
2128}
2129
2130static void
2131cas_mifinit(struct cas_softc *sc)
2132{
2133
2134	/* Configure the MIF in frame mode. */
2135	CAS_WRITE_4(sc, CAS_MIF_CONF,
2136	    CAS_READ_4(sc, CAS_MIF_CONF) & ~CAS_MIF_CONF_BB_MODE);
2137	CAS_BARRIER(sc, CAS_MIF_CONF, 4,
2138	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2139}
2140
2141/*
2142 * MII interface
2143 *
2144 * The MII interface supports at least three different operating modes:
2145 *
2146 * Bitbang mode is implemented using data, clock and output enable registers.
2147 *
2148 * Frame mode is implemented by loading a complete frame into the frame
2149 * register and polling the valid bit for completion.
2150 *
2151 * Polling mode uses the frame register but completion is indicated by
2152 * an interrupt.
2153 *
2154 */
2155static int
2156cas_mii_readreg(device_t dev, int phy, int reg)
2157{
2158	struct cas_softc *sc;
2159	int n;
2160	uint32_t v;
2161
2162#ifdef CAS_DEBUG_PHY
2163	printf("%s: phy %d reg %d\n", __func__, phy, reg);
2164#endif
2165
2166	sc = device_get_softc(dev);
2167	if ((sc->sc_flags & CAS_SERDES) != 0) {
2168		switch (reg) {
2169		case MII_BMCR:
2170			reg = CAS_PCS_CTRL;
2171			break;
2172		case MII_BMSR:
2173			reg = CAS_PCS_STATUS;
2174			break;
2175		case MII_PHYIDR1:
2176		case MII_PHYIDR2:
2177			return (0);
2178		case MII_ANAR:
2179			reg = CAS_PCS_ANAR;
2180			break;
2181		case MII_ANLPAR:
2182			reg = CAS_PCS_ANLPAR;
2183			break;
2184		case MII_EXTSR:
2185			return (EXTSR_1000XFDX | EXTSR_1000XHDX);
2186		default:
2187			device_printf(sc->sc_dev,
2188			    "%s: unhandled register %d\n", __func__, reg);
2189			return (0);
2190		}
2191		return (CAS_READ_4(sc, reg));
2192	}
2193
2194	/* Construct the frame command. */
2195	v = CAS_MIF_FRAME_READ |
2196	    (phy << CAS_MIF_FRAME_PHY_SHFT) |
2197	    (reg << CAS_MIF_FRAME_REG_SHFT);
2198
2199	CAS_WRITE_4(sc, CAS_MIF_FRAME, v);
2200	CAS_BARRIER(sc, CAS_MIF_FRAME, 4,
2201	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2202	for (n = 0; n < 100; n++) {
2203		DELAY(1);
2204		v = CAS_READ_4(sc, CAS_MIF_FRAME);
2205		if (v & CAS_MIF_FRAME_TA_LSB)
2206			return (v & CAS_MIF_FRAME_DATA);
2207	}
2208
2209	device_printf(sc->sc_dev, "%s: timed out\n", __func__);
2210	return (0);
2211}
2212
2213static int
2214cas_mii_writereg(device_t dev, int phy, int reg, int val)
2215{
2216	struct cas_softc *sc;
2217	int n;
2218	uint32_t v;
2219
2220#ifdef CAS_DEBUG_PHY
2221	printf("%s: phy %d reg %d val %x\n", phy, reg, val, __func__);
2222#endif
2223
2224	sc = device_get_softc(dev);
2225	if ((sc->sc_flags & CAS_SERDES) != 0) {
2226		switch (reg) {
2227		case MII_BMSR:
2228			reg = CAS_PCS_STATUS;
2229			break;
2230		case MII_BMCR:
2231			reg = CAS_PCS_CTRL;
2232			if ((val & CAS_PCS_CTRL_RESET) == 0)
2233				break;
2234			CAS_WRITE_4(sc, CAS_PCS_CTRL, val);
2235			CAS_BARRIER(sc, CAS_PCS_CTRL, 4,
2236			    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2237			if (!cas_bitwait(sc, CAS_PCS_CTRL,
2238			    CAS_PCS_CTRL_RESET, 0))
2239				device_printf(sc->sc_dev,
2240				    "cannot reset PCS\n");
2241			/* FALLTHROUGH */
2242		case MII_ANAR:
2243			CAS_WRITE_4(sc, CAS_PCS_CONF, 0);
2244			CAS_BARRIER(sc, CAS_PCS_CONF, 4,
2245			    BUS_SPACE_BARRIER_WRITE);
2246			CAS_WRITE_4(sc, CAS_PCS_ANAR, val);
2247			CAS_BARRIER(sc, CAS_PCS_ANAR, 4,
2248			    BUS_SPACE_BARRIER_WRITE);
2249			CAS_WRITE_4(sc, CAS_PCS_SERDES_CTRL,
2250			    CAS_PCS_SERDES_CTRL_ESD);
2251			CAS_BARRIER(sc, CAS_PCS_CONF, 4,
2252			    BUS_SPACE_BARRIER_WRITE);
2253			CAS_WRITE_4(sc, CAS_PCS_CONF,
2254			    CAS_PCS_CONF_EN);
2255			CAS_BARRIER(sc, CAS_PCS_CONF, 4,
2256			    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2257			return (0);
2258		case MII_ANLPAR:
2259			reg = CAS_PCS_ANLPAR;
2260			break;
2261		default:
2262			device_printf(sc->sc_dev,
2263			    "%s: unhandled register %d\n", __func__, reg);
2264			return (0);
2265		}
2266		CAS_WRITE_4(sc, reg, val);
2267		CAS_BARRIER(sc, reg, 4,
2268		    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2269		return (0);
2270	}
2271
2272	/* Construct the frame command. */
2273	v = CAS_MIF_FRAME_WRITE |
2274	    (phy << CAS_MIF_FRAME_PHY_SHFT) |
2275	    (reg << CAS_MIF_FRAME_REG_SHFT) |
2276	    (val & CAS_MIF_FRAME_DATA);
2277
2278	CAS_WRITE_4(sc, CAS_MIF_FRAME, v);
2279	CAS_BARRIER(sc, CAS_MIF_FRAME, 4,
2280	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2281	for (n = 0; n < 100; n++) {
2282		DELAY(1);
2283		v = CAS_READ_4(sc, CAS_MIF_FRAME);
2284		if (v & CAS_MIF_FRAME_TA_LSB)
2285			return (1);
2286	}
2287
2288	device_printf(sc->sc_dev, "%s: timed out\n", __func__);
2289	return (0);
2290}
2291
2292static void
2293cas_mii_statchg(device_t dev)
2294{
2295	struct cas_softc *sc;
2296	struct ifnet *ifp;
2297	int gigabit;
2298	uint32_t rxcfg, txcfg, v;
2299
2300	sc = device_get_softc(dev);
2301	ifp = sc->sc_ifp;
2302
2303	CAS_LOCK_ASSERT(sc, MA_OWNED);
2304
2305#ifdef CAS_DEBUG
2306	if ((ifp->if_flags & IFF_DEBUG) != 0)
2307		device_printf(sc->sc_dev, "%s: status changen", __func__);
2308#endif
2309
2310	if ((sc->sc_mii->mii_media_status & IFM_ACTIVE) != 0 &&
2311	    IFM_SUBTYPE(sc->sc_mii->mii_media_active) != IFM_NONE)
2312		sc->sc_flags |= CAS_LINK;
2313	else
2314		sc->sc_flags &= ~CAS_LINK;
2315
2316	switch (IFM_SUBTYPE(sc->sc_mii->mii_media_active)) {
2317	case IFM_1000_SX:
2318	case IFM_1000_LX:
2319	case IFM_1000_CX:
2320	case IFM_1000_T:
2321		gigabit = 1;
2322		break;
2323	default:
2324		gigabit = 0;
2325	}
2326
2327	/*
2328	 * The configuration done here corresponds to the steps F) and
2329	 * G) and as far as enabling of RX and TX MAC goes also step H)
2330	 * of the initialization sequence outlined in section 11.2.1 of
2331	 * the Cassini+ ASIC Specification.
2332	 */
2333
2334	rxcfg = CAS_READ_4(sc, CAS_MAC_RX_CONF);
2335	rxcfg &= ~(CAS_MAC_RX_CONF_EN | CAS_MAC_RX_CONF_CARR);
2336	txcfg = CAS_MAC_TX_CONF_EN_IPG0 | CAS_MAC_TX_CONF_NGU |
2337	    CAS_MAC_TX_CONF_NGUL;
2338	if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) & IFM_FDX) != 0)
2339		txcfg |= CAS_MAC_TX_CONF_ICARR | CAS_MAC_TX_CONF_ICOLLIS;
2340	else if (gigabit != 0) {
2341		rxcfg |= CAS_MAC_RX_CONF_CARR;
2342		txcfg |= CAS_MAC_TX_CONF_CARR;
2343	}
2344	CAS_WRITE_4(sc, CAS_MAC_TX_CONF, 0);
2345	CAS_BARRIER(sc, CAS_MAC_TX_CONF, 4,
2346	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2347	if (!cas_bitwait(sc, CAS_MAC_TX_CONF, CAS_MAC_TX_CONF_EN, 0))
2348		device_printf(sc->sc_dev, "cannot disable TX MAC\n");
2349	CAS_WRITE_4(sc, CAS_MAC_TX_CONF, txcfg);
2350	CAS_WRITE_4(sc, CAS_MAC_RX_CONF, 0);
2351	CAS_BARRIER(sc, CAS_MAC_RX_CONF, 4,
2352	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2353	if (!cas_bitwait(sc, CAS_MAC_RX_CONF, CAS_MAC_RX_CONF_EN, 0))
2354		device_printf(sc->sc_dev, "cannot disable RX MAC\n");
2355	CAS_WRITE_4(sc, CAS_MAC_RX_CONF, rxcfg);
2356
2357	v = CAS_READ_4(sc, CAS_MAC_CTRL_CONF) &
2358	    ~(CAS_MAC_CTRL_CONF_TXP | CAS_MAC_CTRL_CONF_RXP);
2359#ifdef notyet
2360	if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) &
2361	    IFM_ETH_RXPAUSE) != 0)
2362		v |= CAS_MAC_CTRL_CONF_RXP;
2363	if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) &
2364	    IFM_ETH_TXPAUSE) != 0)
2365		v |= CAS_MAC_CTRL_CONF_TXP;
2366#endif
2367	CAS_WRITE_4(sc, CAS_MAC_CTRL_CONF, v);
2368
2369	/*
2370	 * All supported chips have a bug causing incorrect checksum
2371	 * to be calculated when letting them strip the FCS in half-
2372	 * duplex mode.  In theory we could disable FCS stripping and
2373	 * manually adjust the checksum accordingly.  It seems to make
2374	 * more sense to optimze for the common case and just disable
2375	 * hardware checksumming in half-duplex mode though.
2376	 */
2377	if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) & IFM_FDX) == 0) {
2378		ifp->if_capenable &= ~IFCAP_HWCSUM;
2379		ifp->if_hwassist = 0;
2380	} else if ((sc->sc_flags & CAS_NO_CSUM) == 0) {
2381		ifp->if_capenable = ifp->if_capabilities;
2382		ifp->if_hwassist = CAS_CSUM_FEATURES;
2383	}
2384
2385	if (sc->sc_variant == CAS_SATURN) {
2386		if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) & IFM_FDX) == 0)
2387			/* silicon bug workaround */
2388			CAS_WRITE_4(sc, CAS_MAC_PREAMBLE_LEN, 0x41);
2389		else
2390			CAS_WRITE_4(sc, CAS_MAC_PREAMBLE_LEN, 0x7);
2391	}
2392
2393	if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) & IFM_FDX) == 0 &&
2394	    gigabit != 0)
2395		CAS_WRITE_4(sc, CAS_MAC_SLOT_TIME,
2396		    CAS_MAC_SLOT_TIME_CARR);
2397	else
2398		CAS_WRITE_4(sc, CAS_MAC_SLOT_TIME,
2399		    CAS_MAC_SLOT_TIME_NORM);
2400
2401	/* XIF Configuration */
2402	v = CAS_MAC_XIF_CONF_TX_OE | CAS_MAC_XIF_CONF_LNKLED;
2403	if ((sc->sc_flags & CAS_SERDES) == 0) {
2404		if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) & IFM_FDX) == 0)
2405			v |= CAS_MAC_XIF_CONF_NOECHO;
2406		v |= CAS_MAC_XIF_CONF_BUF_OE;
2407	}
2408	if (gigabit != 0)
2409		v |= CAS_MAC_XIF_CONF_GMII;
2410	if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) & IFM_FDX) != 0)
2411		v |= CAS_MAC_XIF_CONF_FDXLED;
2412	CAS_WRITE_4(sc, CAS_MAC_XIF_CONF, v);
2413
2414	if ((sc->sc_ifp->if_drv_flags & IFF_DRV_RUNNING) != 0 &&
2415	    (sc->sc_flags & CAS_LINK) != 0) {
2416		CAS_WRITE_4(sc, CAS_MAC_TX_CONF,
2417		    txcfg | CAS_MAC_TX_CONF_EN);
2418		CAS_WRITE_4(sc, CAS_MAC_RX_CONF,
2419		    rxcfg | CAS_MAC_RX_CONF_EN);
2420	}
2421}
2422
2423static int
2424cas_mediachange(struct ifnet *ifp)
2425{
2426	struct cas_softc *sc = ifp->if_softc;
2427	int error;
2428
2429	/* XXX add support for serial media. */
2430
2431	CAS_LOCK(sc);
2432	error = mii_mediachg(sc->sc_mii);
2433	CAS_UNLOCK(sc);
2434	return (error);
2435}
2436
2437static void
2438cas_mediastatus(struct ifnet *ifp, struct ifmediareq *ifmr)
2439{
2440	struct cas_softc *sc = ifp->if_softc;
2441
2442	CAS_LOCK(sc);
2443	if ((ifp->if_flags & IFF_UP) == 0) {
2444		CAS_UNLOCK(sc);
2445		return;
2446	}
2447
2448	mii_pollstat(sc->sc_mii);
2449	ifmr->ifm_active = sc->sc_mii->mii_media_active;
2450	ifmr->ifm_status = sc->sc_mii->mii_media_status;
2451	CAS_UNLOCK(sc);
2452}
2453
2454static int
2455cas_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
2456{
2457	struct cas_softc *sc = ifp->if_softc;
2458	struct ifreq *ifr = (struct ifreq *)data;
2459	int error;
2460
2461	error = 0;
2462	switch (cmd) {
2463	case SIOCSIFFLAGS:
2464		CAS_LOCK(sc);
2465		if ((ifp->if_flags & IFF_UP) != 0) {
2466			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0 &&
2467			    ((ifp->if_flags ^ sc->sc_ifflags) &
2468			    (IFF_ALLMULTI | IFF_PROMISC)) != 0)
2469				cas_setladrf(sc);
2470			else
2471				cas_init_locked(sc);
2472		} else if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
2473			cas_stop(ifp);
2474		sc->sc_ifflags = ifp->if_flags;
2475		CAS_UNLOCK(sc);
2476		break;
2477	case SIOCSIFCAP:
2478		CAS_LOCK(sc);
2479		if ((sc->sc_flags & CAS_NO_CSUM) != 0) {
2480			error = EINVAL;
2481			CAS_UNLOCK(sc);
2482			break;
2483		}
2484		ifp->if_capenable = ifr->ifr_reqcap;
2485		if ((ifp->if_capenable & IFCAP_TXCSUM) != 0)
2486			ifp->if_hwassist = CAS_CSUM_FEATURES;
2487		else
2488			ifp->if_hwassist = 0;
2489		CAS_UNLOCK(sc);
2490		break;
2491	case SIOCADDMULTI:
2492	case SIOCDELMULTI:
2493		CAS_LOCK(sc);
2494		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
2495			cas_setladrf(sc);
2496		CAS_UNLOCK(sc);
2497		break;
2498	case SIOCSIFMTU:
2499		if ((ifr->ifr_mtu < ETHERMIN) ||
2500		    (ifr->ifr_mtu > ETHERMTU_JUMBO))
2501			error = EINVAL;
2502		else
2503			ifp->if_mtu = ifr->ifr_mtu;
2504		break;
2505	case SIOCGIFMEDIA:
2506	case SIOCSIFMEDIA:
2507		error = ifmedia_ioctl(ifp, ifr, &sc->sc_mii->mii_media, cmd);
2508		break;
2509	default:
2510		error = ether_ioctl(ifp, cmd, data);
2511		break;
2512	}
2513
2514	return (error);
2515}
2516
2517static void
2518cas_setladrf(struct cas_softc *sc)
2519{
2520	struct ifnet *ifp = sc->sc_ifp;
2521	struct ifmultiaddr *inm;
2522	int i;
2523	uint32_t hash[16];
2524	uint32_t crc, v;
2525
2526	CAS_LOCK_ASSERT(sc, MA_OWNED);
2527
2528	/* Get the current RX configuration. */
2529	v = CAS_READ_4(sc, CAS_MAC_RX_CONF);
2530
2531	/*
2532	 * Turn off promiscuous mode, promiscuous group mode (all multicast),
2533	 * and hash filter.  Depending on the case, the right bit will be
2534	 * enabled.
2535	 */
2536	v &= ~(CAS_MAC_RX_CONF_PROMISC | CAS_MAC_RX_CONF_HFILTER |
2537	    CAS_MAC_RX_CONF_PGRP);
2538
2539	CAS_WRITE_4(sc, CAS_MAC_RX_CONF, v);
2540	CAS_BARRIER(sc, CAS_MAC_RX_CONF, 4,
2541	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2542	if (!cas_bitwait(sc, CAS_MAC_RX_CONF, CAS_MAC_RX_CONF_HFILTER, 0))
2543		device_printf(sc->sc_dev, "cannot disable RX hash filter\n");
2544
2545	if ((ifp->if_flags & IFF_PROMISC) != 0) {
2546		v |= CAS_MAC_RX_CONF_PROMISC;
2547		goto chipit;
2548	}
2549	if ((ifp->if_flags & IFF_ALLMULTI) != 0) {
2550		v |= CAS_MAC_RX_CONF_PGRP;
2551		goto chipit;
2552	}
2553
2554	/*
2555	 * Set up multicast address filter by passing all multicast
2556	 * addresses through a crc generator, and then using the high
2557	 * order 8 bits as an index into the 256 bit logical address
2558	 * filter.  The high order 4 bits selects the word, while the
2559	 * other 4 bits select the bit within the word (where bit 0
2560	 * is the MSB).
2561	 */
2562
2563	/* Clear the hash table. */
2564	memset(hash, 0, sizeof(hash));
2565
2566	if_maddr_rlock(ifp);
2567	TAILQ_FOREACH(inm, &ifp->if_multiaddrs, ifma_link) {
2568		if (inm->ifma_addr->sa_family != AF_LINK)
2569			continue;
2570		crc = ether_crc32_le(LLADDR((struct sockaddr_dl *)
2571		    inm->ifma_addr), ETHER_ADDR_LEN);
2572
2573		/* We just want the 8 most significant bits. */
2574		crc >>= 24;
2575
2576		/* Set the corresponding bit in the filter. */
2577		hash[crc >> 4] |= 1 << (15 - (crc & 15));
2578	}
2579	if_maddr_runlock(ifp);
2580
2581	v |= CAS_MAC_RX_CONF_HFILTER;
2582
2583	/* Now load the hash table into the chip (if we are using it). */
2584	for (i = 0; i < 16; i++)
2585		CAS_WRITE_4(sc,
2586		    CAS_MAC_HASH0 + i * (CAS_MAC_HASH1 - CAS_MAC_HASH0),
2587		    hash[i]);
2588
2589 chipit:
2590	CAS_WRITE_4(sc, CAS_MAC_RX_CONF, v);
2591}
2592
2593static int	cas_pci_attach(device_t dev);
2594static int	cas_pci_detach(device_t dev);
2595static int	cas_pci_probe(device_t dev);
2596static int	cas_pci_resume(device_t dev);
2597static int	cas_pci_suspend(device_t dev);
2598
2599static device_method_t cas_pci_methods[] = {
2600	/* Device interface */
2601	DEVMETHOD(device_probe,		cas_pci_probe),
2602	DEVMETHOD(device_attach,	cas_pci_attach),
2603	DEVMETHOD(device_detach,	cas_pci_detach),
2604	DEVMETHOD(device_suspend,	cas_pci_suspend),
2605	DEVMETHOD(device_resume,	cas_pci_resume),
2606	/* Use the suspend handler here, it is all that is required. */
2607	DEVMETHOD(device_shutdown,	cas_pci_suspend),
2608
2609	/* bus interface */
2610	DEVMETHOD(bus_print_child,	bus_generic_print_child),
2611	DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
2612
2613	/* MII interface */
2614	DEVMETHOD(miibus_readreg,	cas_mii_readreg),
2615	DEVMETHOD(miibus_writereg,	cas_mii_writereg),
2616	DEVMETHOD(miibus_statchg,	cas_mii_statchg),
2617
2618	KOBJMETHOD_END
2619};
2620
2621static driver_t cas_pci_driver = {
2622	"cas",
2623	cas_pci_methods,
2624	sizeof(struct cas_softc)
2625};
2626
2627DRIVER_MODULE(cas, pci, cas_pci_driver, cas_devclass, 0, 0);
2628DRIVER_MODULE(miibus, cas, miibus_driver, miibus_devclass, 0, 0);
2629MODULE_DEPEND(cas, pci, 1, 1, 1);
2630
2631static const struct cas_pci_dev {
2632	uint32_t	cpd_devid;
2633	uint8_t		cpd_revid;
2634	int		cpd_variant;
2635	const char	*cpd_desc;
2636} const cas_pci_devlist[] = {
2637	{ 0x0035100b, 0x0, CAS_SATURN, "NS DP83065 Saturn Gigabit Ethernet" },
2638	{ 0xabba108e, 0x10, CAS_CASPLUS, "Sun Cassini+ Gigabit Ethernet" },
2639	{ 0xabba108e, 0x0, CAS_CAS, "Sun Cassini Gigabit Ethernet" },
2640	{ 0, 0, 0, NULL }
2641};
2642
2643static int
2644cas_pci_probe(device_t dev)
2645{
2646	int i;
2647
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			device_set_desc(dev, cas_pci_devlist[i].cpd_desc);
2652			return (BUS_PROBE_DEFAULT);
2653		}
2654	}
2655
2656	return (ENXIO);
2657}
2658
2659static struct resource_spec cas_pci_res_spec[] = {
2660	{ SYS_RES_IRQ, 0, RF_SHAREABLE | RF_ACTIVE },	/* CAS_RES_INTR */
2661	{ SYS_RES_MEMORY, PCIR_BAR(0), RF_ACTIVE },	/* CAS_RES_MEM */
2662	{ -1, 0 }
2663};
2664
2665#define	CAS_LOCAL_MAC_ADDRESS	"local-mac-address"
2666#define	CAS_PHY_INTERFACE	"phy-interface"
2667#define	CAS_PHY_TYPE		"phy-type"
2668#define	CAS_PHY_TYPE_PCS	"pcs"
2669
2670static int
2671cas_pci_attach(device_t dev)
2672{
2673	char buf[sizeof(CAS_LOCAL_MAC_ADDRESS)];
2674	struct cas_softc *sc;
2675	int i;
2676#if !(defined(__powerpc__) || defined(__sparc64__))
2677	u_char enaddr[4][ETHER_ADDR_LEN];
2678	u_int j, k, lma, pcs[4], phy;
2679#endif
2680
2681	sc = device_get_softc(dev);
2682	sc->sc_variant = CAS_UNKNOWN;
2683	for (i = 0; cas_pci_devlist[i].cpd_desc != NULL; i++) {
2684		if (pci_get_devid(dev) == cas_pci_devlist[i].cpd_devid &&
2685		    pci_get_revid(dev) >= cas_pci_devlist[i].cpd_revid) {
2686			sc->sc_variant = cas_pci_devlist[i].cpd_variant;
2687			break;
2688		}
2689	}
2690	if (sc->sc_variant == CAS_UNKNOWN) {
2691		device_printf(dev, "unknown adaptor\n");
2692		return (ENXIO);
2693	}
2694
2695	pci_enable_busmaster(dev);
2696
2697	sc->sc_dev = dev;
2698	if (sc->sc_variant == CAS_CAS && pci_get_devid(dev) < 0x02)
2699		/* Hardware checksumming may hang TX. */
2700		sc->sc_flags |= CAS_NO_CSUM;
2701	if (sc->sc_variant == CAS_CASPLUS || sc->sc_variant == CAS_SATURN)
2702		sc->sc_flags |= CAS_REG_PLUS;
2703	if (sc->sc_variant == CAS_CAS ||
2704	    (sc->sc_variant == CAS_CASPLUS && pci_get_revid(dev) < 0x11))
2705		sc->sc_flags |= CAS_TABORT;
2706	if (bootverbose)
2707		device_printf(dev, "flags=0x%x\n", sc->sc_flags);
2708
2709	if (bus_alloc_resources(dev, cas_pci_res_spec, sc->sc_res)) {
2710		device_printf(dev, "failed to allocate resources\n");
2711		bus_release_resources(dev, cas_pci_res_spec, sc->sc_res);
2712		return (ENXIO);
2713	}
2714
2715	CAS_LOCK_INIT(sc, device_get_nameunit(dev));
2716
2717#if defined(__powerpc__) || defined(__sparc64__)
2718	OF_getetheraddr(dev, sc->sc_enaddr);
2719	if (OF_getprop(ofw_bus_get_node(dev), CAS_PHY_INTERFACE, buf,
2720	    sizeof(buf)) > 0 || OF_getprop(ofw_bus_get_node(dev),
2721	    CAS_PHY_TYPE, buf, sizeof(buf)) > 0) {
2722		buf[sizeof(buf) - 1] = '\0';
2723		if (strcmp(buf, CAS_PHY_TYPE_PCS) == 0)
2724			sc->sc_flags |= CAS_SERDES;
2725	}
2726#else
2727	/*
2728	 * Dig out VPD (vital product data) and read the MAC address as well
2729	 * as the PHY type.  The VPD resides in the PCI Expansion ROM (PCI
2730	 * FCode) and can't be accessed via the PCI capability pointer.
2731	 * SUNW,pci-ce and SUNW,pci-qge use the Enhanced VPD format described
2732	 * in the free US Patent 7149820.
2733	 */
2734
2735#define	PCI_ROMHDR_SIZE			0x1c
2736#define	PCI_ROMHDR_SIG			0x00
2737#define	PCI_ROMHDR_SIG_MAGIC		0xaa55		/* little endian */
2738#define	PCI_ROMHDR_PTR_DATA		0x18
2739#define	PCI_ROM_SIZE			0x18
2740#define	PCI_ROM_SIG			0x00
2741#define	PCI_ROM_SIG_MAGIC		0x52494350	/* "PCIR", endian */
2742							/* reversed */
2743#define	PCI_ROM_VENDOR			0x04
2744#define	PCI_ROM_DEVICE			0x06
2745#define	PCI_ROM_PTR_VPD			0x08
2746#define	PCI_VPDRES_BYTE0		0x00
2747#define	PCI_VPDRES_ISLARGE(x)		((x) & 0x80)
2748#define	PCI_VPDRES_LARGE_NAME(x)	((x) & 0x7f)
2749#define	PCI_VPDRES_LARGE_LEN_LSB	0x01
2750#define	PCI_VPDRES_LARGE_LEN_MSB	0x02
2751#define	PCI_VPDRES_LARGE_SIZE		0x03
2752#define	PCI_VPDRES_TYPE_ID_STRING	0x02		/* large */
2753#define	PCI_VPDRES_TYPE_VPD		0x10		/* large */
2754#define	PCI_VPD_KEY0			0x00
2755#define	PCI_VPD_KEY1			0x01
2756#define	PCI_VPD_LEN			0x02
2757#define	PCI_VPD_SIZE			0x03
2758
2759#define	CAS_ROM_READ_1(sc, offs)					\
2760	CAS_READ_1((sc), CAS_PCI_ROM_OFFSET + (offs))
2761#define	CAS_ROM_READ_2(sc, offs)					\
2762	CAS_READ_2((sc), CAS_PCI_ROM_OFFSET + (offs))
2763#define	CAS_ROM_READ_4(sc, offs)					\
2764	CAS_READ_4((sc), CAS_PCI_ROM_OFFSET + (offs))
2765
2766	lma = phy = 0;
2767	memset(enaddr, 0, sizeof(enaddr));
2768	memset(pcs, 0, sizeof(pcs));
2769
2770	/* Enable PCI Expansion ROM access. */
2771	CAS_WRITE_4(sc, CAS_BIM_LDEV_OEN,
2772	    CAS_BIM_LDEV_OEN_PAD | CAS_BIM_LDEV_OEN_PROM);
2773
2774	/* Read PCI Expansion ROM header. */
2775	if (CAS_ROM_READ_2(sc, PCI_ROMHDR_SIG) != PCI_ROMHDR_SIG_MAGIC ||
2776	    (i = CAS_ROM_READ_2(sc, PCI_ROMHDR_PTR_DATA)) <
2777	    PCI_ROMHDR_SIZE) {
2778		device_printf(dev, "unexpected PCI Expansion ROM header\n");
2779		goto fail_prom;
2780	}
2781
2782	/* Read PCI Expansion ROM data. */
2783	if (CAS_ROM_READ_4(sc, i + PCI_ROM_SIG) != PCI_ROM_SIG_MAGIC ||
2784	    CAS_ROM_READ_2(sc, i + PCI_ROM_VENDOR) != pci_get_vendor(dev) ||
2785	    CAS_ROM_READ_2(sc, i + PCI_ROM_DEVICE) != pci_get_device(dev) ||
2786	    (j = CAS_ROM_READ_2(sc, i + PCI_ROM_PTR_VPD)) <
2787	    i + PCI_ROM_SIZE) {
2788		device_printf(dev, "unexpected PCI Expansion ROM data\n");
2789		goto fail_prom;
2790	}
2791
2792	/* Read PCI VPD. */
2793 next:
2794	if (PCI_VPDRES_ISLARGE(CAS_ROM_READ_1(sc,
2795	    j + PCI_VPDRES_BYTE0)) == 0) {
2796		device_printf(dev, "no large PCI VPD\n");
2797		goto fail_prom;
2798	}
2799
2800	i = (CAS_ROM_READ_1(sc, j + PCI_VPDRES_LARGE_LEN_MSB) << 8) |
2801	    CAS_ROM_READ_1(sc, j + PCI_VPDRES_LARGE_LEN_LSB);
2802	switch (PCI_VPDRES_LARGE_NAME(CAS_ROM_READ_1(sc,
2803	    j + PCI_VPDRES_BYTE0))) {
2804	case PCI_VPDRES_TYPE_ID_STRING:
2805		/* Skip identifier string. */
2806		j += PCI_VPDRES_LARGE_SIZE + i;
2807		goto next;
2808	case PCI_VPDRES_TYPE_VPD:
2809		for (j += PCI_VPDRES_LARGE_SIZE; i > 0;
2810		    i -= PCI_VPD_SIZE + CAS_ROM_READ_1(sc, j + PCI_VPD_LEN),
2811		    j += PCI_VPD_SIZE + CAS_ROM_READ_1(sc, j + PCI_VPD_LEN)) {
2812			if (CAS_ROM_READ_1(sc, j + PCI_VPD_KEY0) != 'Z')
2813				/* no Enhanced VPD */
2814				continue;
2815			if (CAS_ROM_READ_1(sc, j + PCI_VPD_SIZE) != 'I')
2816				/* no instance property */
2817				continue;
2818			if (CAS_ROM_READ_1(sc, j + PCI_VPD_SIZE + 3) == 'B') {
2819				/* byte array */
2820				if (CAS_ROM_READ_1(sc,
2821				    j + PCI_VPD_SIZE + 4) != ETHER_ADDR_LEN)
2822					continue;
2823				bus_read_region_1(sc->sc_res[CAS_RES_MEM],
2824				    CAS_PCI_ROM_OFFSET + j + PCI_VPD_SIZE + 5,
2825				    buf, sizeof(buf));
2826				buf[sizeof(buf) - 1] = '\0';
2827				if (strcmp(buf, CAS_LOCAL_MAC_ADDRESS) != 0)
2828					continue;
2829				bus_read_region_1(sc->sc_res[CAS_RES_MEM],
2830				    CAS_PCI_ROM_OFFSET + j + PCI_VPD_SIZE +
2831				    5 + sizeof(CAS_LOCAL_MAC_ADDRESS),
2832				    enaddr[lma], sizeof(enaddr[lma]));
2833				lma++;
2834				if (lma == 4 && phy == 4)
2835					break;
2836			} else if (CAS_ROM_READ_1(sc, j + PCI_VPD_SIZE + 3) ==
2837			   'S') {
2838				/* string */
2839				if (CAS_ROM_READ_1(sc,
2840				    j + PCI_VPD_SIZE + 4) !=
2841				    sizeof(CAS_PHY_TYPE_PCS))
2842					continue;
2843				bus_read_region_1(sc->sc_res[CAS_RES_MEM],
2844				    CAS_PCI_ROM_OFFSET + j + PCI_VPD_SIZE + 5,
2845				    buf, sizeof(buf));
2846				buf[sizeof(buf) - 1] = '\0';
2847				if (strcmp(buf, CAS_PHY_INTERFACE) == 0)
2848					k = sizeof(CAS_PHY_INTERFACE);
2849				else if (strcmp(buf, CAS_PHY_TYPE) == 0)
2850					k = sizeof(CAS_PHY_TYPE);
2851				else
2852					continue;
2853				bus_read_region_1(sc->sc_res[CAS_RES_MEM],
2854				    CAS_PCI_ROM_OFFSET + j + PCI_VPD_SIZE +
2855				    5 + k, buf, sizeof(buf));
2856				buf[sizeof(buf) - 1] = '\0';
2857				if (strcmp(buf, CAS_PHY_TYPE_PCS) == 0)
2858					pcs[phy] = 1;
2859				phy++;
2860				if (lma == 4 && phy == 4)
2861					break;
2862			}
2863		}
2864		break;
2865	default:
2866		device_printf(dev, "unexpected PCI VPD\n");
2867		goto fail_prom;
2868	}
2869
2870 fail_prom:
2871	CAS_WRITE_4(sc, CAS_BIM_LDEV_OEN, 0);
2872
2873	if (lma == 0) {
2874		device_printf(dev, "could not determine Ethernet address\n");
2875		goto fail;
2876	}
2877	i = 0;
2878	if (lma > 1 && pci_get_slot(dev) < sizeof(enaddr) / sizeof(*enaddr))
2879		i = pci_get_slot(dev);
2880	memcpy(sc->sc_enaddr, enaddr[i], ETHER_ADDR_LEN);
2881
2882	if (phy == 0) {
2883		device_printf(dev, "could not determine PHY type\n");
2884		goto fail;
2885	}
2886	i = 0;
2887	if (phy > 1 && pci_get_slot(dev) < sizeof(pcs) / sizeof(*pcs))
2888		i = pci_get_slot(dev);
2889	if (pcs[i] != 0)
2890		sc->sc_flags |= CAS_SERDES;
2891#endif
2892
2893	if (cas_attach(sc) != 0) {
2894		device_printf(dev, "could not be attached\n");
2895		goto fail;
2896	}
2897
2898	if (bus_setup_intr(dev, sc->sc_res[CAS_RES_INTR], INTR_TYPE_NET |
2899	    INTR_MPSAFE, cas_intr, NULL, sc, &sc->sc_ih) != 0) {
2900		device_printf(dev, "failed to set up interrupt\n");
2901		cas_detach(sc);
2902		goto fail;
2903	}
2904	return (0);
2905
2906 fail:
2907	CAS_LOCK_DESTROY(sc);
2908	bus_release_resources(dev, cas_pci_res_spec, sc->sc_res);
2909	return (ENXIO);
2910}
2911
2912static int
2913cas_pci_detach(device_t dev)
2914{
2915	struct cas_softc *sc;
2916
2917	sc = device_get_softc(dev);
2918	bus_teardown_intr(dev, sc->sc_res[CAS_RES_INTR], sc->sc_ih);
2919	cas_detach(sc);
2920	CAS_LOCK_DESTROY(sc);
2921	bus_release_resources(dev, cas_pci_res_spec, sc->sc_res);
2922	return (0);
2923}
2924
2925static int
2926cas_pci_suspend(device_t dev)
2927{
2928
2929	cas_suspend(device_get_softc(dev));
2930	return (0);
2931}
2932
2933static int
2934cas_pci_resume(device_t dev)
2935{
2936
2937	cas_resume(device_get_softc(dev));
2938	return (0);
2939}
2940