1/*-
2 * Copyright (C) 2001 Eduardo Horvath.
3 * Copyright (c) 2001-2003 Thomas Moestl
4 * Copyright (c) 2007-2009 Marius Strobl <marius@FreeBSD.org>
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 *    notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 *    notice, this list of conditions and the following disclaimer in the
14 *    documentation and/or other materials provided with the distribution.
15 *
16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR  ``AS IS'' AND
17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR  BE LIABLE
20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26 * SUCH DAMAGE.
27 *
28 *	from: NetBSD: gem.c,v 1.21 2002/06/01 23:50:58 lukem Exp
29 *	from: FreeBSD: if_gem.c 182060 2008-08-23 15:03:26Z marius
30 */
31
32#include <sys/cdefs.h>
33__FBSDID("$FreeBSD$");
34
35/*
36 * driver for Sun Cassini/Cassini+ and National Semiconductor DP83065
37 * Saturn Gigabit Ethernet controllers
38 */
39
40#if 0
41#define	CAS_DEBUG
42#endif
43
44#include <sys/param.h>
45#include <sys/systm.h>
46#include <sys/bus.h>
47#include <sys/callout.h>
48#include <sys/endian.h>
49#include <sys/mbuf.h>
50#include <sys/malloc.h>
51#include <sys/kernel.h>
52#include <sys/lock.h>
53#include <sys/module.h>
54#include <sys/mutex.h>
55#include <sys/refcount.h>
56#include <sys/resource.h>
57#include <sys/rman.h>
58#include <sys/socket.h>
59#include <sys/sockio.h>
60#include <sys/taskqueue.h>
61
62#include <net/bpf.h>
63#include <net/ethernet.h>
64#include <net/if.h>
65#include <net/if_var.h>
66#include <net/if_arp.h>
67#include <net/if_dl.h>
68#include <net/if_media.h>
69#include <net/if_types.h>
70#include <net/if_vlan_var.h>
71
72#include <netinet/in.h>
73#include <netinet/in_systm.h>
74#include <netinet/ip.h>
75#include <netinet/tcp.h>
76#include <netinet/udp.h>
77
78#include <machine/bus.h>
79#if defined(__powerpc__) || defined(__sparc64__)
80#include <dev/ofw/ofw_bus.h>
81#include <dev/ofw/openfirm.h>
82#include <machine/ofw_machdep.h>
83#endif
84#include <machine/resource.h>
85
86#include <dev/mii/mii.h>
87#include <dev/mii/miivar.h>
88
89#include <dev/cas/if_casreg.h>
90#include <dev/cas/if_casvar.h>
91
92#include <dev/pci/pcireg.h>
93#include <dev/pci/pcivar.h>
94
95#include "miibus_if.h"
96
97#define RINGASSERT(n , min, max)					\
98	CTASSERT(powerof2(n) && (n) >= (min) && (n) <= (max))
99
100RINGASSERT(CAS_NRXCOMP, 128, 32768);
101RINGASSERT(CAS_NRXDESC, 32, 8192);
102RINGASSERT(CAS_NRXDESC2, 32, 8192);
103RINGASSERT(CAS_NTXDESC, 32, 8192);
104
105#undef RINGASSERT
106
107#define	CCDASSERT(m, a)							\
108	CTASSERT((offsetof(struct cas_control_data, m) & ((a) - 1)) == 0)
109
110CCDASSERT(ccd_rxcomps, CAS_RX_COMP_ALIGN);
111CCDASSERT(ccd_rxdescs, CAS_RX_DESC_ALIGN);
112CCDASSERT(ccd_rxdescs2, CAS_RX_DESC_ALIGN);
113
114#undef CCDASSERT
115
116#define	CAS_TRIES	10000
117
118/*
119 * According to documentation, the hardware has support for basic TCP
120 * checksum offloading only, in practice this can be also used for UDP
121 * however (i.e. the problem of previous Sun NICs that a checksum of 0x0
122 * is not converted to 0xffff no longer exists).
123 */
124#define	CAS_CSUM_FEATURES	(CSUM_TCP | CSUM_UDP)
125
126static inline void cas_add_rxdesc(struct cas_softc *sc, u_int idx);
127static int	cas_attach(struct cas_softc *sc);
128static int	cas_bitwait(struct cas_softc *sc, bus_addr_t r, uint32_t clr,
129		    uint32_t set);
130static void	cas_cddma_callback(void *xsc, bus_dma_segment_t *segs,
131		    int nsegs, int error);
132static void	cas_detach(struct cas_softc *sc);
133static int	cas_disable_rx(struct cas_softc *sc);
134static int	cas_disable_tx(struct cas_softc *sc);
135static void	cas_eint(struct cas_softc *sc, u_int status);
136static void	cas_free(struct mbuf *m, void *arg1, void* arg2);
137static void	cas_init(void *xsc);
138static void	cas_init_locked(struct cas_softc *sc);
139static void	cas_init_regs(struct cas_softc *sc);
140static int	cas_intr(void *v);
141static void	cas_intr_task(void *arg, int pending __unused);
142static int	cas_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data);
143static int	cas_load_txmbuf(struct cas_softc *sc, struct mbuf **m_head);
144static int	cas_mediachange(struct ifnet *ifp);
145static void	cas_mediastatus(struct ifnet *ifp, struct ifmediareq *ifmr);
146static void	cas_meminit(struct cas_softc *sc);
147static void	cas_mifinit(struct cas_softc *sc);
148static int	cas_mii_readreg(device_t dev, int phy, int reg);
149static void	cas_mii_statchg(device_t dev);
150static int	cas_mii_writereg(device_t dev, int phy, int reg, int val);
151static void	cas_reset(struct cas_softc *sc);
152static int	cas_reset_rx(struct cas_softc *sc);
153static int	cas_reset_tx(struct cas_softc *sc);
154static void	cas_resume(struct cas_softc *sc);
155static u_int	cas_descsize(u_int sz);
156static void	cas_rint(struct cas_softc *sc);
157static void	cas_rint_timeout(void *arg);
158static inline void cas_rxcksum(struct mbuf *m, uint16_t cksum);
159static inline void cas_rxcompinit(struct cas_rx_comp *rxcomp);
160static u_int	cas_rxcompsize(u_int sz);
161static void	cas_rxdma_callback(void *xsc, bus_dma_segment_t *segs,
162		    int nsegs, int error);
163static void	cas_setladrf(struct cas_softc *sc);
164static void	cas_start(struct ifnet *ifp);
165static void	cas_stop(struct ifnet *ifp);
166static void	cas_suspend(struct cas_softc *sc);
167static void	cas_tick(void *arg);
168static void	cas_tint(struct cas_softc *sc);
169static void	cas_tx_task(void *arg, int pending __unused);
170static inline void cas_txkick(struct cas_softc *sc);
171static void	cas_watchdog(struct cas_softc *sc);
172
173static devclass_t cas_devclass;
174
175MODULE_DEPEND(cas, ether, 1, 1, 1);
176MODULE_DEPEND(cas, miibus, 1, 1, 1);
177
178#ifdef CAS_DEBUG
179#include <sys/ktr.h>
180#define	KTR_CAS		KTR_SPARE2
181#endif
182
183static int
184cas_attach(struct cas_softc *sc)
185{
186	struct cas_txsoft *txs;
187	struct ifnet *ifp;
188	int error, i;
189	uint32_t v;
190
191	/* Set up ifnet structure. */
192	ifp = sc->sc_ifp = if_alloc(IFT_ETHER);
193	if (ifp == NULL)
194		return (ENOSPC);
195	ifp->if_softc = sc;
196	if_initname(ifp, device_get_name(sc->sc_dev),
197	    device_get_unit(sc->sc_dev));
198	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
199	ifp->if_start = cas_start;
200	ifp->if_ioctl = cas_ioctl;
201	ifp->if_init = cas_init;
202	IFQ_SET_MAXLEN(&ifp->if_snd, CAS_TXQUEUELEN);
203	ifp->if_snd.ifq_drv_maxlen = CAS_TXQUEUELEN;
204	IFQ_SET_READY(&ifp->if_snd);
205
206	callout_init_mtx(&sc->sc_tick_ch, &sc->sc_mtx, 0);
207	callout_init_mtx(&sc->sc_rx_ch, &sc->sc_mtx, 0);
208	/* Create local taskq. */
209	TASK_INIT(&sc->sc_intr_task, 0, cas_intr_task, sc);
210	TASK_INIT(&sc->sc_tx_task, 1, cas_tx_task, ifp);
211	sc->sc_tq = taskqueue_create_fast("cas_taskq", M_WAITOK,
212	    taskqueue_thread_enqueue, &sc->sc_tq);
213	if (sc->sc_tq == NULL) {
214		device_printf(sc->sc_dev, "could not create taskqueue\n");
215		error = ENXIO;
216		goto fail_ifnet;
217	}
218	error = taskqueue_start_threads(&sc->sc_tq, 1, PI_NET, "%s taskq",
219	    device_get_nameunit(sc->sc_dev));
220	if (error != 0) {
221		device_printf(sc->sc_dev, "could not start threads\n");
222		goto fail_taskq;
223	}
224
225	/* Make sure the chip is stopped. */
226	cas_reset(sc);
227
228	error = bus_dma_tag_create(bus_get_dma_tag(sc->sc_dev), 1, 0,
229	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
230	    BUS_SPACE_MAXSIZE, 0, BUS_SPACE_MAXSIZE, 0, NULL, NULL,
231	    &sc->sc_pdmatag);
232	if (error != 0)
233		goto fail_taskq;
234
235	error = bus_dma_tag_create(sc->sc_pdmatag, 1, 0,
236	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
237	    CAS_PAGE_SIZE, 1, CAS_PAGE_SIZE, 0, NULL, NULL, &sc->sc_rdmatag);
238	if (error != 0)
239		goto fail_ptag;
240
241	error = bus_dma_tag_create(sc->sc_pdmatag, 1, 0,
242	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
243	    MCLBYTES * CAS_NTXSEGS, CAS_NTXSEGS, MCLBYTES,
244	    BUS_DMA_ALLOCNOW, NULL, NULL, &sc->sc_tdmatag);
245	if (error != 0)
246		goto fail_rtag;
247
248	error = bus_dma_tag_create(sc->sc_pdmatag, CAS_TX_DESC_ALIGN, 0,
249	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
250	    sizeof(struct cas_control_data), 1,
251	    sizeof(struct cas_control_data), 0,
252	    NULL, NULL, &sc->sc_cdmatag);
253	if (error != 0)
254		goto fail_ttag;
255
256	/*
257	 * Allocate the control data structures, create and load the
258	 * DMA map for it.
259	 */
260	if ((error = bus_dmamem_alloc(sc->sc_cdmatag,
261	    (void **)&sc->sc_control_data,
262	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
263	    &sc->sc_cddmamap)) != 0) {
264		device_printf(sc->sc_dev,
265		    "unable to allocate control data, error = %d\n", error);
266		goto fail_ctag;
267	}
268
269	sc->sc_cddma = 0;
270	if ((error = bus_dmamap_load(sc->sc_cdmatag, sc->sc_cddmamap,
271	    sc->sc_control_data, sizeof(struct cas_control_data),
272	    cas_cddma_callback, sc, 0)) != 0 || sc->sc_cddma == 0) {
273		device_printf(sc->sc_dev,
274		    "unable to load control data DMA map, error = %d\n",
275		    error);
276		goto fail_cmem;
277	}
278
279	/*
280	 * Initialize the transmit job descriptors.
281	 */
282	STAILQ_INIT(&sc->sc_txfreeq);
283	STAILQ_INIT(&sc->sc_txdirtyq);
284
285	/*
286	 * Create the transmit buffer DMA maps.
287	 */
288	error = ENOMEM;
289	for (i = 0; i < CAS_TXQUEUELEN; i++) {
290		txs = &sc->sc_txsoft[i];
291		txs->txs_mbuf = NULL;
292		txs->txs_ndescs = 0;
293		if ((error = bus_dmamap_create(sc->sc_tdmatag, 0,
294		    &txs->txs_dmamap)) != 0) {
295			device_printf(sc->sc_dev,
296			    "unable to create TX DMA map %d, error = %d\n",
297			    i, error);
298			goto fail_txd;
299		}
300		STAILQ_INSERT_TAIL(&sc->sc_txfreeq, txs, txs_q);
301	}
302
303	/*
304	 * Allocate the receive buffers, create and load the DMA maps
305	 * for them.
306	 */
307	for (i = 0; i < CAS_NRXDESC; i++) {
308		if ((error = bus_dmamem_alloc(sc->sc_rdmatag,
309		    &sc->sc_rxdsoft[i].rxds_buf, BUS_DMA_WAITOK,
310		    &sc->sc_rxdsoft[i].rxds_dmamap)) != 0) {
311			device_printf(sc->sc_dev,
312			    "unable to allocate RX buffer %d, error = %d\n",
313			    i, error);
314			goto fail_rxmem;
315		}
316
317		sc->sc_rxdptr = i;
318		sc->sc_rxdsoft[i].rxds_paddr = 0;
319		if ((error = bus_dmamap_load(sc->sc_rdmatag,
320		    sc->sc_rxdsoft[i].rxds_dmamap, sc->sc_rxdsoft[i].rxds_buf,
321		    CAS_PAGE_SIZE, cas_rxdma_callback, sc, 0)) != 0 ||
322		    sc->sc_rxdsoft[i].rxds_paddr == 0) {
323			device_printf(sc->sc_dev,
324			    "unable to load RX DMA map %d, error = %d\n",
325			    i, error);
326			goto fail_rxmap;
327		}
328	}
329
330	if ((sc->sc_flags & CAS_SERDES) == 0) {
331		CAS_WRITE_4(sc, CAS_PCS_DATAPATH, CAS_PCS_DATAPATH_MII);
332		CAS_BARRIER(sc, CAS_PCS_DATAPATH, 4,
333		    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
334		cas_mifinit(sc);
335		/*
336		 * Look for an external PHY.
337		 */
338		error = ENXIO;
339		v = CAS_READ_4(sc, CAS_MIF_CONF);
340		if ((v & CAS_MIF_CONF_MDI1) != 0) {
341			v |= CAS_MIF_CONF_PHY_SELECT;
342			CAS_WRITE_4(sc, CAS_MIF_CONF, v);
343			CAS_BARRIER(sc, CAS_MIF_CONF, 4,
344			    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
345			/* Enable/unfreeze the GMII pins of Saturn. */
346			if (sc->sc_variant == CAS_SATURN) {
347				CAS_WRITE_4(sc, CAS_SATURN_PCFG,
348				    CAS_READ_4(sc, CAS_SATURN_PCFG) &
349				    ~CAS_SATURN_PCFG_FSI);
350				CAS_BARRIER(sc, CAS_SATURN_PCFG, 4,
351				    BUS_SPACE_BARRIER_READ |
352				    BUS_SPACE_BARRIER_WRITE);
353				DELAY(10000);
354			}
355			error = mii_attach(sc->sc_dev, &sc->sc_miibus, ifp,
356			    cas_mediachange, cas_mediastatus, BMSR_DEFCAPMASK,
357			    MII_PHY_ANY, MII_OFFSET_ANY, MIIF_DOPAUSE);
358		}
359		/*
360		 * Fall back on an internal PHY if no external PHY was found.
361		 */
362		if (error != 0 && (v & CAS_MIF_CONF_MDI0) != 0) {
363			v &= ~CAS_MIF_CONF_PHY_SELECT;
364			CAS_WRITE_4(sc, CAS_MIF_CONF, v);
365			CAS_BARRIER(sc, CAS_MIF_CONF, 4,
366			    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
367			/* Freeze the GMII pins of Saturn for saving power. */
368			if (sc->sc_variant == CAS_SATURN) {
369				CAS_WRITE_4(sc, CAS_SATURN_PCFG,
370				    CAS_READ_4(sc, CAS_SATURN_PCFG) |
371				    CAS_SATURN_PCFG_FSI);
372				CAS_BARRIER(sc, CAS_SATURN_PCFG, 4,
373				    BUS_SPACE_BARRIER_READ |
374				    BUS_SPACE_BARRIER_WRITE);
375				DELAY(10000);
376			}
377			error = mii_attach(sc->sc_dev, &sc->sc_miibus, ifp,
378			    cas_mediachange, cas_mediastatus, BMSR_DEFCAPMASK,
379			    MII_PHY_ANY, MII_OFFSET_ANY, MIIF_DOPAUSE);
380		}
381	} else {
382		/*
383		 * Use the external PCS SERDES.
384		 */
385		CAS_WRITE_4(sc, CAS_PCS_DATAPATH, CAS_PCS_DATAPATH_SERDES);
386		CAS_BARRIER(sc, CAS_PCS_DATAPATH, 4, BUS_SPACE_BARRIER_WRITE);
387		/* Enable/unfreeze the SERDES pins of Saturn. */
388		if (sc->sc_variant == CAS_SATURN) {
389			CAS_WRITE_4(sc, CAS_SATURN_PCFG, 0);
390			CAS_BARRIER(sc, CAS_SATURN_PCFG, 4,
391			    BUS_SPACE_BARRIER_WRITE);
392		}
393		CAS_WRITE_4(sc, CAS_PCS_SERDES_CTRL, CAS_PCS_SERDES_CTRL_ESD);
394		CAS_BARRIER(sc, CAS_PCS_SERDES_CTRL, 4,
395		    BUS_SPACE_BARRIER_WRITE);
396		CAS_WRITE_4(sc, CAS_PCS_CONF, CAS_PCS_CONF_EN);
397		CAS_BARRIER(sc, CAS_PCS_CONF, 4,
398		    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
399		error = mii_attach(sc->sc_dev, &sc->sc_miibus, ifp,
400		    cas_mediachange, cas_mediastatus, BMSR_DEFCAPMASK,
401		    CAS_PHYAD_EXTERNAL, MII_OFFSET_ANY, MIIF_DOPAUSE);
402	}
403	if (error != 0) {
404		device_printf(sc->sc_dev, "attaching PHYs failed\n");
405		goto fail_rxmap;
406	}
407	sc->sc_mii = device_get_softc(sc->sc_miibus);
408
409	/*
410	 * From this point forward, the attachment cannot fail.  A failure
411	 * before this point releases all resources that may have been
412	 * allocated.
413	 */
414
415	/* Announce FIFO sizes. */
416	v = CAS_READ_4(sc, CAS_TX_FIFO_SIZE);
417	device_printf(sc->sc_dev, "%ukB RX FIFO, %ukB TX FIFO\n",
418	    CAS_RX_FIFO_SIZE / 1024, v / 16);
419
420	/* Attach the interface. */
421	ether_ifattach(ifp, sc->sc_enaddr);
422
423	/*
424	 * Tell the upper layer(s) we support long frames/checksum offloads.
425	 */
426	ifp->if_hdrlen = sizeof(struct ether_vlan_header);
427	ifp->if_capabilities = IFCAP_VLAN_MTU;
428	if ((sc->sc_flags & CAS_NO_CSUM) == 0) {
429		ifp->if_capabilities |= IFCAP_HWCSUM;
430		ifp->if_hwassist = CAS_CSUM_FEATURES;
431	}
432	ifp->if_capenable = ifp->if_capabilities;
433
434	return (0);
435
436	/*
437	 * Free any resources we've allocated during the failed attach
438	 * attempt.  Do this in reverse order and fall through.
439	 */
440 fail_rxmap:
441	for (i = 0; i < CAS_NRXDESC; i++)
442		if (sc->sc_rxdsoft[i].rxds_paddr != 0)
443			bus_dmamap_unload(sc->sc_rdmatag,
444			    sc->sc_rxdsoft[i].rxds_dmamap);
445 fail_rxmem:
446	for (i = 0; i < CAS_NRXDESC; i++)
447		if (sc->sc_rxdsoft[i].rxds_buf != NULL)
448			bus_dmamem_free(sc->sc_rdmatag,
449			    sc->sc_rxdsoft[i].rxds_buf,
450			    sc->sc_rxdsoft[i].rxds_dmamap);
451 fail_txd:
452	for (i = 0; i < CAS_TXQUEUELEN; i++)
453		if (sc->sc_txsoft[i].txs_dmamap != NULL)
454			bus_dmamap_destroy(sc->sc_tdmatag,
455			    sc->sc_txsoft[i].txs_dmamap);
456	bus_dmamap_unload(sc->sc_cdmatag, sc->sc_cddmamap);
457 fail_cmem:
458	bus_dmamem_free(sc->sc_cdmatag, sc->sc_control_data,
459	    sc->sc_cddmamap);
460 fail_ctag:
461	bus_dma_tag_destroy(sc->sc_cdmatag);
462 fail_ttag:
463	bus_dma_tag_destroy(sc->sc_tdmatag);
464 fail_rtag:
465	bus_dma_tag_destroy(sc->sc_rdmatag);
466 fail_ptag:
467	bus_dma_tag_destroy(sc->sc_pdmatag);
468 fail_taskq:
469	taskqueue_free(sc->sc_tq);
470 fail_ifnet:
471	if_free(ifp);
472	return (error);
473}
474
475static void
476cas_detach(struct cas_softc *sc)
477{
478	struct ifnet *ifp = sc->sc_ifp;
479	int i;
480
481	ether_ifdetach(ifp);
482	CAS_LOCK(sc);
483	cas_stop(ifp);
484	CAS_UNLOCK(sc);
485	callout_drain(&sc->sc_tick_ch);
486	callout_drain(&sc->sc_rx_ch);
487	taskqueue_drain(sc->sc_tq, &sc->sc_intr_task);
488	taskqueue_drain(sc->sc_tq, &sc->sc_tx_task);
489	if_free(ifp);
490	taskqueue_free(sc->sc_tq);
491	device_delete_child(sc->sc_dev, sc->sc_miibus);
492
493	for (i = 0; i < CAS_NRXDESC; i++)
494		if (sc->sc_rxdsoft[i].rxds_dmamap != NULL)
495			bus_dmamap_sync(sc->sc_rdmatag,
496			    sc->sc_rxdsoft[i].rxds_dmamap,
497			    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
498	for (i = 0; i < CAS_NRXDESC; i++)
499		if (sc->sc_rxdsoft[i].rxds_paddr != 0)
500			bus_dmamap_unload(sc->sc_rdmatag,
501			    sc->sc_rxdsoft[i].rxds_dmamap);
502	for (i = 0; i < CAS_NRXDESC; i++)
503		if (sc->sc_rxdsoft[i].rxds_buf != NULL)
504			bus_dmamem_free(sc->sc_rdmatag,
505			    sc->sc_rxdsoft[i].rxds_buf,
506			    sc->sc_rxdsoft[i].rxds_dmamap);
507	for (i = 0; i < CAS_TXQUEUELEN; i++)
508		if (sc->sc_txsoft[i].txs_dmamap != NULL)
509			bus_dmamap_destroy(sc->sc_tdmatag,
510			    sc->sc_txsoft[i].txs_dmamap);
511	CAS_CDSYNC(sc, BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
512	bus_dmamap_unload(sc->sc_cdmatag, sc->sc_cddmamap);
513	bus_dmamem_free(sc->sc_cdmatag, sc->sc_control_data,
514	    sc->sc_cddmamap);
515	bus_dma_tag_destroy(sc->sc_cdmatag);
516	bus_dma_tag_destroy(sc->sc_tdmatag);
517	bus_dma_tag_destroy(sc->sc_rdmatag);
518	bus_dma_tag_destroy(sc->sc_pdmatag);
519}
520
521static void
522cas_suspend(struct cas_softc *sc)
523{
524	struct ifnet *ifp = sc->sc_ifp;
525
526	CAS_LOCK(sc);
527	cas_stop(ifp);
528	CAS_UNLOCK(sc);
529}
530
531static void
532cas_resume(struct cas_softc *sc)
533{
534	struct ifnet *ifp = sc->sc_ifp;
535
536	CAS_LOCK(sc);
537	/*
538	 * On resume all registers have to be initialized again like
539	 * after power-on.
540	 */
541	sc->sc_flags &= ~CAS_INITED;
542	if (ifp->if_flags & IFF_UP)
543		cas_init_locked(sc);
544	CAS_UNLOCK(sc);
545}
546
547static inline void
548cas_rxcksum(struct mbuf *m, uint16_t cksum)
549{
550	struct ether_header *eh;
551	struct ip *ip;
552	struct udphdr *uh;
553	uint16_t *opts;
554	int32_t hlen, len, pktlen;
555	uint32_t temp32;
556
557	pktlen = m->m_pkthdr.len;
558	if (pktlen < sizeof(struct ether_header) + sizeof(struct ip))
559		return;
560	eh = mtod(m, struct ether_header *);
561	if (eh->ether_type != htons(ETHERTYPE_IP))
562		return;
563	ip = (struct ip *)(eh + 1);
564	if (ip->ip_v != IPVERSION)
565		return;
566
567	hlen = ip->ip_hl << 2;
568	pktlen -= sizeof(struct ether_header);
569	if (hlen < sizeof(struct ip))
570		return;
571	if (ntohs(ip->ip_len) < hlen)
572		return;
573	if (ntohs(ip->ip_len) != pktlen)
574		return;
575	if (ip->ip_off & htons(IP_MF | IP_OFFMASK))
576		return;	/* Cannot handle fragmented packet. */
577
578	switch (ip->ip_p) {
579	case IPPROTO_TCP:
580		if (pktlen < (hlen + sizeof(struct tcphdr)))
581			return;
582		break;
583	case IPPROTO_UDP:
584		if (pktlen < (hlen + sizeof(struct udphdr)))
585			return;
586		uh = (struct udphdr *)((uint8_t *)ip + hlen);
587		if (uh->uh_sum == 0)
588			return; /* no checksum */
589		break;
590	default:
591		return;
592	}
593
594	cksum = ~cksum;
595	/* checksum fixup for IP options */
596	len = hlen - sizeof(struct ip);
597	if (len > 0) {
598		opts = (uint16_t *)(ip + 1);
599		for (; len > 0; len -= sizeof(uint16_t), opts++) {
600			temp32 = cksum - *opts;
601			temp32 = (temp32 >> 16) + (temp32 & 65535);
602			cksum = temp32 & 65535;
603		}
604	}
605	m->m_pkthdr.csum_flags |= CSUM_DATA_VALID;
606	m->m_pkthdr.csum_data = cksum;
607}
608
609static void
610cas_cddma_callback(void *xsc, bus_dma_segment_t *segs, int nsegs, int error)
611{
612	struct cas_softc *sc = xsc;
613
614	if (error != 0)
615		return;
616	if (nsegs != 1)
617		panic("%s: bad control buffer segment count", __func__);
618	sc->sc_cddma = segs[0].ds_addr;
619}
620
621static void
622cas_rxdma_callback(void *xsc, bus_dma_segment_t *segs, int nsegs, int error)
623{
624	struct cas_softc *sc = xsc;
625
626	if (error != 0)
627		return;
628	if (nsegs != 1)
629		panic("%s: bad RX buffer segment count", __func__);
630	sc->sc_rxdsoft[sc->sc_rxdptr].rxds_paddr = segs[0].ds_addr;
631}
632
633static void
634cas_tick(void *arg)
635{
636	struct cas_softc *sc = arg;
637	struct ifnet *ifp = sc->sc_ifp;
638	uint32_t v;
639
640	CAS_LOCK_ASSERT(sc, MA_OWNED);
641
642	/*
643	 * Unload collision and error counters.
644	 */
645	if_inc_counter(ifp, IFCOUNTER_COLLISIONS,
646	    CAS_READ_4(sc, CAS_MAC_NORM_COLL_CNT) +
647	    CAS_READ_4(sc, CAS_MAC_FIRST_COLL_CNT));
648	v = CAS_READ_4(sc, CAS_MAC_EXCESS_COLL_CNT) +
649	    CAS_READ_4(sc, CAS_MAC_LATE_COLL_CNT);
650	if_inc_counter(ifp, IFCOUNTER_COLLISIONS, v);
651	if_inc_counter(ifp, IFCOUNTER_OERRORS, v);
652	if_inc_counter(ifp, IFCOUNTER_IERRORS,
653	    CAS_READ_4(sc, CAS_MAC_RX_LEN_ERR_CNT) +
654	    CAS_READ_4(sc, CAS_MAC_RX_ALIGN_ERR) +
655	    CAS_READ_4(sc, CAS_MAC_RX_CRC_ERR_CNT) +
656	    CAS_READ_4(sc, CAS_MAC_RX_CODE_VIOL));
657
658	/*
659	 * Then clear the hardware counters.
660	 */
661	CAS_WRITE_4(sc, CAS_MAC_NORM_COLL_CNT, 0);
662	CAS_WRITE_4(sc, CAS_MAC_FIRST_COLL_CNT, 0);
663	CAS_WRITE_4(sc, CAS_MAC_EXCESS_COLL_CNT, 0);
664	CAS_WRITE_4(sc, CAS_MAC_LATE_COLL_CNT, 0);
665	CAS_WRITE_4(sc, CAS_MAC_RX_LEN_ERR_CNT, 0);
666	CAS_WRITE_4(sc, CAS_MAC_RX_ALIGN_ERR, 0);
667	CAS_WRITE_4(sc, CAS_MAC_RX_CRC_ERR_CNT, 0);
668	CAS_WRITE_4(sc, CAS_MAC_RX_CODE_VIOL, 0);
669
670	mii_tick(sc->sc_mii);
671
672	if (sc->sc_txfree != CAS_MAXTXFREE)
673		cas_tint(sc);
674
675	cas_watchdog(sc);
676
677	callout_reset(&sc->sc_tick_ch, hz, cas_tick, sc);
678}
679
680static int
681cas_bitwait(struct cas_softc *sc, bus_addr_t r, uint32_t clr, uint32_t set)
682{
683	int i;
684	uint32_t reg;
685
686	for (i = CAS_TRIES; i--; DELAY(100)) {
687		reg = CAS_READ_4(sc, r);
688		if ((reg & clr) == 0 && (reg & set) == set)
689			return (1);
690	}
691	return (0);
692}
693
694static void
695cas_reset(struct cas_softc *sc)
696{
697
698#ifdef CAS_DEBUG
699	CTR2(KTR_CAS, "%s: %s", device_get_name(sc->sc_dev), __func__);
700#endif
701	/* Disable all interrupts in order to avoid spurious ones. */
702	CAS_WRITE_4(sc, CAS_INTMASK, 0xffffffff);
703
704	cas_reset_rx(sc);
705	cas_reset_tx(sc);
706
707	/*
708	 * Do a full reset modulo the result of the last auto-negotiation
709	 * when using the SERDES.
710	 */
711	CAS_WRITE_4(sc, CAS_RESET, CAS_RESET_RX | CAS_RESET_TX |
712	    ((sc->sc_flags & CAS_SERDES) != 0 ? CAS_RESET_PCS_DIS : 0));
713	CAS_BARRIER(sc, CAS_RESET, 4,
714	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
715	DELAY(3000);
716	if (!cas_bitwait(sc, CAS_RESET, CAS_RESET_RX | CAS_RESET_TX, 0))
717		device_printf(sc->sc_dev, "cannot reset device\n");
718}
719
720static void
721cas_stop(struct ifnet *ifp)
722{
723	struct cas_softc *sc = ifp->if_softc;
724	struct cas_txsoft *txs;
725
726#ifdef CAS_DEBUG
727	CTR2(KTR_CAS, "%s: %s", device_get_name(sc->sc_dev), __func__);
728#endif
729
730	callout_stop(&sc->sc_tick_ch);
731	callout_stop(&sc->sc_rx_ch);
732
733	/* Disable all interrupts in order to avoid spurious ones. */
734	CAS_WRITE_4(sc, CAS_INTMASK, 0xffffffff);
735
736	cas_reset_tx(sc);
737	cas_reset_rx(sc);
738
739	/*
740	 * Release any queued transmit buffers.
741	 */
742	while ((txs = STAILQ_FIRST(&sc->sc_txdirtyq)) != NULL) {
743		STAILQ_REMOVE_HEAD(&sc->sc_txdirtyq, txs_q);
744		if (txs->txs_ndescs != 0) {
745			bus_dmamap_sync(sc->sc_tdmatag, txs->txs_dmamap,
746			    BUS_DMASYNC_POSTWRITE);
747			bus_dmamap_unload(sc->sc_tdmatag, txs->txs_dmamap);
748			if (txs->txs_mbuf != NULL) {
749				m_freem(txs->txs_mbuf);
750				txs->txs_mbuf = NULL;
751			}
752		}
753		STAILQ_INSERT_TAIL(&sc->sc_txfreeq, txs, txs_q);
754	}
755
756	/*
757	 * Mark the interface down and cancel the watchdog timer.
758	 */
759	ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
760	sc->sc_flags &= ~CAS_LINK;
761	sc->sc_wdog_timer = 0;
762}
763
764static int
765cas_reset_rx(struct cas_softc *sc)
766{
767
768	/*
769	 * Resetting while DMA is in progress can cause a bus hang, so we
770	 * disable DMA first.
771	 */
772	(void)cas_disable_rx(sc);
773	CAS_WRITE_4(sc, CAS_RX_CONF, 0);
774	CAS_BARRIER(sc, CAS_RX_CONF, 4,
775	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
776	if (!cas_bitwait(sc, CAS_RX_CONF, CAS_RX_CONF_RXDMA_EN, 0))
777		device_printf(sc->sc_dev, "cannot disable RX DMA\n");
778
779	/* Finally, reset the ERX. */
780	CAS_WRITE_4(sc, CAS_RESET, CAS_RESET_RX |
781	    ((sc->sc_flags & CAS_SERDES) != 0 ? CAS_RESET_PCS_DIS : 0));
782	CAS_BARRIER(sc, CAS_RESET, 4,
783	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
784	if (!cas_bitwait(sc, CAS_RESET, CAS_RESET_RX, 0)) {
785		device_printf(sc->sc_dev, "cannot reset receiver\n");
786		return (1);
787	}
788	return (0);
789}
790
791static int
792cas_reset_tx(struct cas_softc *sc)
793{
794
795	/*
796	 * Resetting while DMA is in progress can cause a bus hang, so we
797	 * disable DMA first.
798	 */
799	(void)cas_disable_tx(sc);
800	CAS_WRITE_4(sc, CAS_TX_CONF, 0);
801	CAS_BARRIER(sc, CAS_TX_CONF, 4,
802	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
803	if (!cas_bitwait(sc, CAS_TX_CONF, CAS_TX_CONF_TXDMA_EN, 0))
804		device_printf(sc->sc_dev, "cannot disable TX DMA\n");
805
806	/* Finally, reset the ETX. */
807	CAS_WRITE_4(sc, CAS_RESET, CAS_RESET_TX |
808	    ((sc->sc_flags & CAS_SERDES) != 0 ? CAS_RESET_PCS_DIS : 0));
809	CAS_BARRIER(sc, CAS_RESET, 4,
810	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
811	if (!cas_bitwait(sc, CAS_RESET, CAS_RESET_TX, 0)) {
812		device_printf(sc->sc_dev, "cannot reset transmitter\n");
813		return (1);
814	}
815	return (0);
816}
817
818static int
819cas_disable_rx(struct cas_softc *sc)
820{
821
822	CAS_WRITE_4(sc, CAS_MAC_RX_CONF,
823	    CAS_READ_4(sc, CAS_MAC_RX_CONF) & ~CAS_MAC_RX_CONF_EN);
824	CAS_BARRIER(sc, CAS_MAC_RX_CONF, 4,
825	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
826	if (cas_bitwait(sc, CAS_MAC_RX_CONF, CAS_MAC_RX_CONF_EN, 0))
827		return (1);
828	if (bootverbose)
829		device_printf(sc->sc_dev, "cannot disable RX MAC\n");
830	return (0);
831}
832
833static int
834cas_disable_tx(struct cas_softc *sc)
835{
836
837	CAS_WRITE_4(sc, CAS_MAC_TX_CONF,
838	    CAS_READ_4(sc, CAS_MAC_TX_CONF) & ~CAS_MAC_TX_CONF_EN);
839	CAS_BARRIER(sc, CAS_MAC_TX_CONF, 4,
840	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
841	if (cas_bitwait(sc, CAS_MAC_TX_CONF, CAS_MAC_TX_CONF_EN, 0))
842		return (1);
843	if (bootverbose)
844		device_printf(sc->sc_dev, "cannot disable TX MAC\n");
845	return (0);
846}
847
848static inline void
849cas_rxcompinit(struct cas_rx_comp *rxcomp)
850{
851
852	rxcomp->crc_word1 = 0;
853	rxcomp->crc_word2 = 0;
854	rxcomp->crc_word3 =
855	    htole64(CAS_SET(ETHER_HDR_LEN + sizeof(struct ip), CAS_RC3_CSO));
856	rxcomp->crc_word4 = htole64(CAS_RC4_ZERO);
857}
858
859static void
860cas_meminit(struct cas_softc *sc)
861{
862	int i;
863
864	CAS_LOCK_ASSERT(sc, MA_OWNED);
865
866	/*
867	 * Initialize the transmit descriptor ring.
868	 */
869	for (i = 0; i < CAS_NTXDESC; i++) {
870		sc->sc_txdescs[i].cd_flags = 0;
871		sc->sc_txdescs[i].cd_buf_ptr = 0;
872	}
873	sc->sc_txfree = CAS_MAXTXFREE;
874	sc->sc_txnext = 0;
875	sc->sc_txwin = 0;
876
877	/*
878	 * Initialize the receive completion ring.
879	 */
880	for (i = 0; i < CAS_NRXCOMP; i++)
881		cas_rxcompinit(&sc->sc_rxcomps[i]);
882	sc->sc_rxcptr = 0;
883
884	/*
885	 * Initialize the first receive descriptor ring.  We leave
886	 * the second one zeroed as we don't actually use it.
887	 */
888	for (i = 0; i < CAS_NRXDESC; i++)
889		CAS_INIT_RXDESC(sc, i, i);
890	sc->sc_rxdptr = 0;
891
892	CAS_CDSYNC(sc, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
893}
894
895static u_int
896cas_descsize(u_int sz)
897{
898
899	switch (sz) {
900	case 32:
901		return (CAS_DESC_32);
902	case 64:
903		return (CAS_DESC_64);
904	case 128:
905		return (CAS_DESC_128);
906	case 256:
907		return (CAS_DESC_256);
908	case 512:
909		return (CAS_DESC_512);
910	case 1024:
911		return (CAS_DESC_1K);
912	case 2048:
913		return (CAS_DESC_2K);
914	case 4096:
915		return (CAS_DESC_4K);
916	case 8192:
917		return (CAS_DESC_8K);
918	default:
919		printf("%s: invalid descriptor ring size %d\n", __func__, sz);
920		return (CAS_DESC_32);
921	}
922}
923
924static u_int
925cas_rxcompsize(u_int sz)
926{
927
928	switch (sz) {
929	case 128:
930		return (CAS_RX_CONF_COMP_128);
931	case 256:
932		return (CAS_RX_CONF_COMP_256);
933	case 512:
934		return (CAS_RX_CONF_COMP_512);
935	case 1024:
936		return (CAS_RX_CONF_COMP_1K);
937	case 2048:
938		return (CAS_RX_CONF_COMP_2K);
939	case 4096:
940		return (CAS_RX_CONF_COMP_4K);
941	case 8192:
942		return (CAS_RX_CONF_COMP_8K);
943	case 16384:
944		return (CAS_RX_CONF_COMP_16K);
945	case 32768:
946		return (CAS_RX_CONF_COMP_32K);
947	default:
948		printf("%s: invalid dcompletion ring size %d\n", __func__, sz);
949		return (CAS_RX_CONF_COMP_128);
950	}
951}
952
953static void
954cas_init(void *xsc)
955{
956	struct cas_softc *sc = xsc;
957
958	CAS_LOCK(sc);
959	cas_init_locked(sc);
960	CAS_UNLOCK(sc);
961}
962
963/*
964 * Initialization of interface; set up initialization block
965 * and transmit/receive descriptor rings.
966 */
967static void
968cas_init_locked(struct cas_softc *sc)
969{
970	struct ifnet *ifp = sc->sc_ifp;
971	uint32_t v;
972
973	CAS_LOCK_ASSERT(sc, MA_OWNED);
974
975	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
976		return;
977
978#ifdef CAS_DEBUG
979	CTR2(KTR_CAS, "%s: %s: calling stop", device_get_name(sc->sc_dev),
980	    __func__);
981#endif
982	/*
983	 * Initialization sequence.  The numbered steps below correspond
984	 * to the sequence outlined in section 6.3.5.1 in the Ethernet
985	 * Channel Engine manual (part of the PCIO manual).
986	 * See also the STP2002-STQ document from Sun Microsystems.
987	 */
988
989	/* step 1 & 2.  Reset the Ethernet Channel. */
990	cas_stop(ifp);
991	cas_reset(sc);
992#ifdef CAS_DEBUG
993	CTR2(KTR_CAS, "%s: %s: restarting", device_get_name(sc->sc_dev),
994	    __func__);
995#endif
996
997	if ((sc->sc_flags & CAS_SERDES) == 0)
998		/* Re-initialize the MIF. */
999		cas_mifinit(sc);
1000
1001	/* step 3.  Setup data structures in host memory. */
1002	cas_meminit(sc);
1003
1004	/* step 4.  TX MAC registers & counters */
1005	cas_init_regs(sc);
1006
1007	/* step 5.  RX MAC registers & counters */
1008
1009	/* step 6 & 7.  Program Ring Base Addresses. */
1010	CAS_WRITE_4(sc, CAS_TX_DESC3_BASE_HI,
1011	    (((uint64_t)CAS_CDTXDADDR(sc, 0)) >> 32));
1012	CAS_WRITE_4(sc, CAS_TX_DESC3_BASE_LO,
1013	    CAS_CDTXDADDR(sc, 0) & 0xffffffff);
1014
1015	CAS_WRITE_4(sc, CAS_RX_COMP_BASE_HI,
1016	    (((uint64_t)CAS_CDRXCADDR(sc, 0)) >> 32));
1017	CAS_WRITE_4(sc, CAS_RX_COMP_BASE_LO,
1018	    CAS_CDRXCADDR(sc, 0) & 0xffffffff);
1019
1020	CAS_WRITE_4(sc, CAS_RX_DESC_BASE_HI,
1021	    (((uint64_t)CAS_CDRXDADDR(sc, 0)) >> 32));
1022	CAS_WRITE_4(sc, CAS_RX_DESC_BASE_LO,
1023	    CAS_CDRXDADDR(sc, 0) & 0xffffffff);
1024
1025	if ((sc->sc_flags & CAS_REG_PLUS) != 0) {
1026		CAS_WRITE_4(sc, CAS_RX_DESC2_BASE_HI,
1027		    (((uint64_t)CAS_CDRXD2ADDR(sc, 0)) >> 32));
1028		CAS_WRITE_4(sc, CAS_RX_DESC2_BASE_LO,
1029		    CAS_CDRXD2ADDR(sc, 0) & 0xffffffff);
1030	}
1031
1032#ifdef CAS_DEBUG
1033	CTR5(KTR_CAS,
1034	    "loading TXDR %lx, RXCR %lx, RXDR %lx, RXD2R %lx, cddma %lx",
1035	    CAS_CDTXDADDR(sc, 0), CAS_CDRXCADDR(sc, 0), CAS_CDRXDADDR(sc, 0),
1036	    CAS_CDRXD2ADDR(sc, 0), sc->sc_cddma);
1037#endif
1038
1039	/* step 8.  Global Configuration & Interrupt Masks */
1040
1041	/* Disable weighted round robin. */
1042	CAS_WRITE_4(sc, CAS_CAW, CAS_CAW_RR_DIS);
1043
1044	/*
1045	 * Enable infinite bursts for revisions without PCI issues if
1046	 * applicable.  Doing so greatly improves the TX performance on
1047	 * !__sparc64__ (on sparc64, setting CAS_INF_BURST improves TX
1048	 * performance only marginally but hurts RX throughput quite a bit).
1049	 */
1050	CAS_WRITE_4(sc, CAS_INF_BURST,
1051#if !defined(__sparc64__)
1052	    (sc->sc_flags & CAS_TABORT) == 0 ? CAS_INF_BURST_EN :
1053#endif
1054	    0);
1055
1056	/* Set up interrupts. */
1057	CAS_WRITE_4(sc, CAS_INTMASK,
1058	    ~(CAS_INTR_TX_INT_ME | CAS_INTR_TX_TAG_ERR |
1059	    CAS_INTR_RX_DONE | CAS_INTR_RX_BUF_NA | CAS_INTR_RX_TAG_ERR |
1060	    CAS_INTR_RX_COMP_FULL | CAS_INTR_RX_BUF_AEMPTY |
1061	    CAS_INTR_RX_COMP_AFULL | CAS_INTR_RX_LEN_MMATCH |
1062	    CAS_INTR_PCI_ERROR_INT
1063#ifdef CAS_DEBUG
1064	    | CAS_INTR_PCS_INT | CAS_INTR_MIF
1065#endif
1066	    ));
1067	/* Don't clear top level interrupts when CAS_STATUS_ALIAS is read. */
1068	CAS_WRITE_4(sc, CAS_CLEAR_ALIAS, 0);
1069	CAS_WRITE_4(sc, CAS_MAC_RX_MASK, ~CAS_MAC_RX_OVERFLOW);
1070	CAS_WRITE_4(sc, CAS_MAC_TX_MASK,
1071	    ~(CAS_MAC_TX_UNDERRUN | CAS_MAC_TX_MAX_PKT_ERR));
1072#ifdef CAS_DEBUG
1073	CAS_WRITE_4(sc, CAS_MAC_CTRL_MASK,
1074	    ~(CAS_MAC_CTRL_PAUSE_RCVD | CAS_MAC_CTRL_PAUSE |
1075	    CAS_MAC_CTRL_NON_PAUSE));
1076#else
1077	CAS_WRITE_4(sc, CAS_MAC_CTRL_MASK,
1078	    CAS_MAC_CTRL_PAUSE_RCVD | CAS_MAC_CTRL_PAUSE |
1079	    CAS_MAC_CTRL_NON_PAUSE);
1080#endif
1081
1082	/* Enable PCI error interrupts. */
1083	CAS_WRITE_4(sc, CAS_ERROR_MASK,
1084	    ~(CAS_ERROR_DTRTO | CAS_ERROR_OTHER | CAS_ERROR_DMAW_ZERO |
1085	    CAS_ERROR_DMAR_ZERO | CAS_ERROR_RTRTO));
1086
1087	/* Enable PCI error interrupts in BIM configuration. */
1088	CAS_WRITE_4(sc, CAS_BIM_CONF,
1089	    CAS_BIM_CONF_DPAR_EN | CAS_BIM_CONF_RMA_EN | CAS_BIM_CONF_RTA_EN);
1090
1091	/*
1092	 * step 9.  ETX Configuration: encode receive descriptor ring size,
1093	 * enable DMA and disable pre-interrupt writeback completion.
1094	 */
1095	v = cas_descsize(CAS_NTXDESC) << CAS_TX_CONF_DESC3_SHFT;
1096	CAS_WRITE_4(sc, CAS_TX_CONF, v | CAS_TX_CONF_TXDMA_EN |
1097	    CAS_TX_CONF_RDPP_DIS | CAS_TX_CONF_PICWB_DIS);
1098
1099	/* step 10.  ERX Configuration */
1100
1101	/*
1102	 * Encode receive completion and descriptor ring sizes, set the
1103	 * swivel offset.
1104	 */
1105	v = cas_rxcompsize(CAS_NRXCOMP) << CAS_RX_CONF_COMP_SHFT;
1106	v |= cas_descsize(CAS_NRXDESC) << CAS_RX_CONF_DESC_SHFT;
1107	if ((sc->sc_flags & CAS_REG_PLUS) != 0)
1108		v |= cas_descsize(CAS_NRXDESC2) << CAS_RX_CONF_DESC2_SHFT;
1109	CAS_WRITE_4(sc, CAS_RX_CONF,
1110	    v | (ETHER_ALIGN << CAS_RX_CONF_SOFF_SHFT));
1111
1112	/* Set the PAUSE thresholds.  We use the maximum OFF threshold. */
1113	CAS_WRITE_4(sc, CAS_RX_PTHRS,
1114	    (111 << CAS_RX_PTHRS_XOFF_SHFT) | (15 << CAS_RX_PTHRS_XON_SHFT));
1115
1116	/* RX blanking */
1117	CAS_WRITE_4(sc, CAS_RX_BLANK,
1118	    (15 << CAS_RX_BLANK_TIME_SHFT) | (5 << CAS_RX_BLANK_PKTS_SHFT));
1119
1120	/* Set RX_COMP_AFULL threshold to half of the RX completions. */
1121	CAS_WRITE_4(sc, CAS_RX_AEMPTY_THRS,
1122	    (CAS_NRXCOMP / 2) << CAS_RX_AEMPTY_COMP_SHFT);
1123
1124	/* Initialize the RX page size register as appropriate for 8k. */
1125	CAS_WRITE_4(sc, CAS_RX_PSZ,
1126	    (CAS_RX_PSZ_8K << CAS_RX_PSZ_SHFT) |
1127	    (4 << CAS_RX_PSZ_MB_CNT_SHFT) |
1128	    (CAS_RX_PSZ_MB_STRD_2K << CAS_RX_PSZ_MB_STRD_SHFT) |
1129	    (CAS_RX_PSZ_MB_OFF_64 << CAS_RX_PSZ_MB_OFF_SHFT));
1130
1131	/* Disable RX random early detection. */
1132	CAS_WRITE_4(sc,	CAS_RX_RED, 0);
1133
1134	/* Zero the RX reassembly DMA table. */
1135	for (v = 0; v <= CAS_RX_REAS_DMA_ADDR_LC; v++) {
1136		CAS_WRITE_4(sc,	CAS_RX_REAS_DMA_ADDR, v);
1137		CAS_WRITE_4(sc,	CAS_RX_REAS_DMA_DATA_LO, 0);
1138		CAS_WRITE_4(sc,	CAS_RX_REAS_DMA_DATA_MD, 0);
1139		CAS_WRITE_4(sc,	CAS_RX_REAS_DMA_DATA_HI, 0);
1140	}
1141
1142	/* Ensure the RX control FIFO and RX IPP FIFO addresses are zero. */
1143	CAS_WRITE_4(sc, CAS_RX_CTRL_FIFO, 0);
1144	CAS_WRITE_4(sc, CAS_RX_IPP_ADDR, 0);
1145
1146	/* Finally, enable RX DMA. */
1147	CAS_WRITE_4(sc, CAS_RX_CONF,
1148	    CAS_READ_4(sc, CAS_RX_CONF) | CAS_RX_CONF_RXDMA_EN);
1149
1150	/* step 11.  Configure Media. */
1151
1152	/* step 12.  RX_MAC Configuration Register */
1153	v = CAS_READ_4(sc, CAS_MAC_RX_CONF);
1154	v &= ~(CAS_MAC_RX_CONF_STRPPAD | CAS_MAC_RX_CONF_EN);
1155	v |= CAS_MAC_RX_CONF_STRPFCS;
1156	sc->sc_mac_rxcfg = v;
1157	/*
1158	 * Clear the RX filter and reprogram it.  This will also set the
1159	 * current RX MAC configuration and enable it.
1160	 */
1161	cas_setladrf(sc);
1162
1163	/* step 13.  TX_MAC Configuration Register */
1164	v = CAS_READ_4(sc, CAS_MAC_TX_CONF);
1165	v |= CAS_MAC_TX_CONF_EN;
1166	(void)cas_disable_tx(sc);
1167	CAS_WRITE_4(sc, CAS_MAC_TX_CONF, v);
1168
1169	/* step 14.  Issue Transmit Pending command. */
1170
1171	/* step 15.  Give the receiver a swift kick. */
1172	CAS_WRITE_4(sc, CAS_RX_KICK, CAS_NRXDESC - 4);
1173	CAS_WRITE_4(sc, CAS_RX_COMP_TAIL, 0);
1174	if ((sc->sc_flags & CAS_REG_PLUS) != 0)
1175		CAS_WRITE_4(sc, CAS_RX_KICK2, CAS_NRXDESC2 - 4);
1176
1177	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1178	ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1179
1180	mii_mediachg(sc->sc_mii);
1181
1182	/* Start the one second timer. */
1183	sc->sc_wdog_timer = 0;
1184	callout_reset(&sc->sc_tick_ch, hz, cas_tick, sc);
1185}
1186
1187static int
1188cas_load_txmbuf(struct cas_softc *sc, struct mbuf **m_head)
1189{
1190	bus_dma_segment_t txsegs[CAS_NTXSEGS];
1191	struct cas_txsoft *txs;
1192	struct ip *ip;
1193	struct mbuf *m;
1194	uint64_t cflags;
1195	int error, nexttx, nsegs, offset, seg;
1196
1197	CAS_LOCK_ASSERT(sc, MA_OWNED);
1198
1199	/* Get a work queue entry. */
1200	if ((txs = STAILQ_FIRST(&sc->sc_txfreeq)) == NULL) {
1201		/* Ran out of descriptors. */
1202		return (ENOBUFS);
1203	}
1204
1205	cflags = 0;
1206	if (((*m_head)->m_pkthdr.csum_flags & CAS_CSUM_FEATURES) != 0) {
1207		if (M_WRITABLE(*m_head) == 0) {
1208			m = m_dup(*m_head, M_NOWAIT);
1209			m_freem(*m_head);
1210			*m_head = m;
1211			if (m == NULL)
1212				return (ENOBUFS);
1213		}
1214		offset = sizeof(struct ether_header);
1215		m = m_pullup(*m_head, offset + sizeof(struct ip));
1216		if (m == NULL) {
1217			*m_head = NULL;
1218			return (ENOBUFS);
1219		}
1220		ip = (struct ip *)(mtod(m, caddr_t) + offset);
1221		offset += (ip->ip_hl << 2);
1222		cflags = (offset << CAS_TD_CKSUM_START_SHFT) |
1223		    ((offset + m->m_pkthdr.csum_data) <<
1224		    CAS_TD_CKSUM_STUFF_SHFT) | CAS_TD_CKSUM_EN;
1225		*m_head = m;
1226	}
1227
1228	error = bus_dmamap_load_mbuf_sg(sc->sc_tdmatag, txs->txs_dmamap,
1229	    *m_head, txsegs, &nsegs, BUS_DMA_NOWAIT);
1230	if (error == EFBIG) {
1231		m = m_collapse(*m_head, M_NOWAIT, CAS_NTXSEGS);
1232		if (m == NULL) {
1233			m_freem(*m_head);
1234			*m_head = NULL;
1235			return (ENOBUFS);
1236		}
1237		*m_head = m;
1238		error = bus_dmamap_load_mbuf_sg(sc->sc_tdmatag,
1239		    txs->txs_dmamap, *m_head, txsegs, &nsegs,
1240		    BUS_DMA_NOWAIT);
1241		if (error != 0) {
1242			m_freem(*m_head);
1243			*m_head = NULL;
1244			return (error);
1245		}
1246	} else if (error != 0)
1247		return (error);
1248	/* If nsegs is wrong then the stack is corrupt. */
1249	KASSERT(nsegs <= CAS_NTXSEGS,
1250	    ("%s: too many DMA segments (%d)", __func__, nsegs));
1251	if (nsegs == 0) {
1252		m_freem(*m_head);
1253		*m_head = NULL;
1254		return (EIO);
1255	}
1256
1257	/*
1258	 * Ensure we have enough descriptors free to describe
1259	 * the packet.  Note, we always reserve one descriptor
1260	 * at the end of the ring as a termination point, in
1261	 * order to prevent wrap-around.
1262	 */
1263	if (nsegs > sc->sc_txfree - 1) {
1264		txs->txs_ndescs = 0;
1265		bus_dmamap_unload(sc->sc_tdmatag, txs->txs_dmamap);
1266		return (ENOBUFS);
1267	}
1268
1269	txs->txs_ndescs = nsegs;
1270	txs->txs_firstdesc = sc->sc_txnext;
1271	nexttx = txs->txs_firstdesc;
1272	for (seg = 0; seg < nsegs; seg++, nexttx = CAS_NEXTTX(nexttx)) {
1273#ifdef CAS_DEBUG
1274		CTR6(KTR_CAS,
1275		    "%s: mapping seg %d (txd %d), len %lx, addr %#lx (%#lx)",
1276		    __func__, seg, nexttx, txsegs[seg].ds_len,
1277		    txsegs[seg].ds_addr, htole64(txsegs[seg].ds_addr));
1278#endif
1279		sc->sc_txdescs[nexttx].cd_buf_ptr =
1280		    htole64(txsegs[seg].ds_addr);
1281		KASSERT(txsegs[seg].ds_len <
1282		    CAS_TD_BUF_LEN_MASK >> CAS_TD_BUF_LEN_SHFT,
1283		    ("%s: segment size too large!", __func__));
1284		sc->sc_txdescs[nexttx].cd_flags =
1285		    htole64(txsegs[seg].ds_len << CAS_TD_BUF_LEN_SHFT);
1286		txs->txs_lastdesc = nexttx;
1287	}
1288
1289	/* Set EOF on the last descriptor. */
1290#ifdef CAS_DEBUG
1291	CTR3(KTR_CAS, "%s: end of frame at segment %d, TX %d",
1292	    __func__, seg, nexttx);
1293#endif
1294	sc->sc_txdescs[txs->txs_lastdesc].cd_flags |=
1295	    htole64(CAS_TD_END_OF_FRAME);
1296
1297	/* Lastly set SOF on the first descriptor. */
1298#ifdef CAS_DEBUG
1299	CTR3(KTR_CAS, "%s: start of frame at segment %d, TX %d",
1300	    __func__, seg, nexttx);
1301#endif
1302	if (sc->sc_txwin += nsegs > CAS_MAXTXFREE * 2 / 3) {
1303		sc->sc_txwin = 0;
1304		sc->sc_txdescs[txs->txs_firstdesc].cd_flags |=
1305		    htole64(cflags | CAS_TD_START_OF_FRAME | CAS_TD_INT_ME);
1306	} else
1307		sc->sc_txdescs[txs->txs_firstdesc].cd_flags |=
1308		    htole64(cflags | CAS_TD_START_OF_FRAME);
1309
1310	/* Sync the DMA map. */
1311	bus_dmamap_sync(sc->sc_tdmatag, txs->txs_dmamap,
1312	    BUS_DMASYNC_PREWRITE);
1313
1314#ifdef CAS_DEBUG
1315	CTR4(KTR_CAS, "%s: setting firstdesc=%d, lastdesc=%d, ndescs=%d",
1316	    __func__, txs->txs_firstdesc, txs->txs_lastdesc,
1317	    txs->txs_ndescs);
1318#endif
1319	STAILQ_REMOVE_HEAD(&sc->sc_txfreeq, txs_q);
1320	STAILQ_INSERT_TAIL(&sc->sc_txdirtyq, txs, txs_q);
1321	txs->txs_mbuf = *m_head;
1322
1323	sc->sc_txnext = CAS_NEXTTX(txs->txs_lastdesc);
1324	sc->sc_txfree -= txs->txs_ndescs;
1325
1326	return (0);
1327}
1328
1329static void
1330cas_init_regs(struct cas_softc *sc)
1331{
1332	int i;
1333	const u_char *laddr = IF_LLADDR(sc->sc_ifp);
1334
1335	CAS_LOCK_ASSERT(sc, MA_OWNED);
1336
1337	/* These registers are not cleared on reset. */
1338	if ((sc->sc_flags & CAS_INITED) == 0) {
1339		/* magic values */
1340		CAS_WRITE_4(sc, CAS_MAC_IPG0, 0);
1341		CAS_WRITE_4(sc, CAS_MAC_IPG1, 8);
1342		CAS_WRITE_4(sc, CAS_MAC_IPG2, 4);
1343
1344		/* min frame length */
1345		CAS_WRITE_4(sc, CAS_MAC_MIN_FRAME, ETHER_MIN_LEN);
1346		/* max frame length and max burst size */
1347		CAS_WRITE_4(sc, CAS_MAC_MAX_BF,
1348		    ((ETHER_MAX_LEN_JUMBO + ETHER_VLAN_ENCAP_LEN) <<
1349		    CAS_MAC_MAX_BF_FRM_SHFT) |
1350		    (0x2000 << CAS_MAC_MAX_BF_BST_SHFT));
1351
1352		/* more magic values */
1353		CAS_WRITE_4(sc, CAS_MAC_PREAMBLE_LEN, 0x7);
1354		CAS_WRITE_4(sc, CAS_MAC_JAM_SIZE, 0x4);
1355		CAS_WRITE_4(sc, CAS_MAC_ATTEMPT_LIMIT, 0x10);
1356		CAS_WRITE_4(sc, CAS_MAC_CTRL_TYPE, 0x8808);
1357
1358		/* random number seed */
1359		CAS_WRITE_4(sc, CAS_MAC_RANDOM_SEED,
1360		    ((laddr[5] << 8) | laddr[4]) & 0x3ff);
1361
1362		/* secondary MAC addresses: 0:0:0:0:0:0 */
1363		for (i = CAS_MAC_ADDR3; i <= CAS_MAC_ADDR41;
1364		    i += CAS_MAC_ADDR4 - CAS_MAC_ADDR3)
1365			CAS_WRITE_4(sc, i, 0);
1366
1367		/* MAC control address: 01:80:c2:00:00:01 */
1368		CAS_WRITE_4(sc, CAS_MAC_ADDR42, 0x0001);
1369		CAS_WRITE_4(sc, CAS_MAC_ADDR43, 0xc200);
1370		CAS_WRITE_4(sc, CAS_MAC_ADDR44, 0x0180);
1371
1372		/* MAC filter address: 0:0:0:0:0:0 */
1373		CAS_WRITE_4(sc, CAS_MAC_AFILTER0, 0);
1374		CAS_WRITE_4(sc, CAS_MAC_AFILTER1, 0);
1375		CAS_WRITE_4(sc, CAS_MAC_AFILTER2, 0);
1376		CAS_WRITE_4(sc, CAS_MAC_AFILTER_MASK1_2, 0);
1377		CAS_WRITE_4(sc, CAS_MAC_AFILTER_MASK0, 0);
1378
1379		/* Zero the hash table. */
1380		for (i = CAS_MAC_HASH0; i <= CAS_MAC_HASH15;
1381		    i += CAS_MAC_HASH1 - CAS_MAC_HASH0)
1382			CAS_WRITE_4(sc, i, 0);
1383
1384		sc->sc_flags |= CAS_INITED;
1385	}
1386
1387	/* Counters need to be zeroed. */
1388	CAS_WRITE_4(sc, CAS_MAC_NORM_COLL_CNT, 0);
1389	CAS_WRITE_4(sc, CAS_MAC_FIRST_COLL_CNT, 0);
1390	CAS_WRITE_4(sc, CAS_MAC_EXCESS_COLL_CNT, 0);
1391	CAS_WRITE_4(sc, CAS_MAC_LATE_COLL_CNT, 0);
1392	CAS_WRITE_4(sc, CAS_MAC_DEFER_TMR_CNT, 0);
1393	CAS_WRITE_4(sc, CAS_MAC_PEAK_ATTEMPTS, 0);
1394	CAS_WRITE_4(sc, CAS_MAC_RX_FRAME_COUNT, 0);
1395	CAS_WRITE_4(sc, CAS_MAC_RX_LEN_ERR_CNT, 0);
1396	CAS_WRITE_4(sc, CAS_MAC_RX_ALIGN_ERR, 0);
1397	CAS_WRITE_4(sc, CAS_MAC_RX_CRC_ERR_CNT, 0);
1398	CAS_WRITE_4(sc, CAS_MAC_RX_CODE_VIOL, 0);
1399
1400	/* Set XOFF PAUSE time. */
1401	CAS_WRITE_4(sc, CAS_MAC_SPC, 0x1BF0 << CAS_MAC_SPC_TIME_SHFT);
1402
1403	/* Set the station address. */
1404	CAS_WRITE_4(sc, CAS_MAC_ADDR0, (laddr[4] << 8) | laddr[5]);
1405	CAS_WRITE_4(sc, CAS_MAC_ADDR1, (laddr[2] << 8) | laddr[3]);
1406	CAS_WRITE_4(sc, CAS_MAC_ADDR2, (laddr[0] << 8) | laddr[1]);
1407
1408	/* Enable MII outputs. */
1409	CAS_WRITE_4(sc, CAS_MAC_XIF_CONF, CAS_MAC_XIF_CONF_TX_OE);
1410}
1411
1412static void
1413cas_tx_task(void *arg, int pending __unused)
1414{
1415	struct ifnet *ifp;
1416
1417	ifp = (struct ifnet *)arg;
1418	cas_start(ifp);
1419}
1420
1421static inline void
1422cas_txkick(struct cas_softc *sc)
1423{
1424
1425	/*
1426	 * Update the TX kick register.  This register has to point to the
1427	 * descriptor after the last valid one and for optimum performance
1428	 * should be incremented in multiples of 4 (the DMA engine fetches/
1429	 * updates descriptors in batches of 4).
1430	 */
1431#ifdef CAS_DEBUG
1432	CTR3(KTR_CAS, "%s: %s: kicking TX %d",
1433	    device_get_name(sc->sc_dev), __func__, sc->sc_txnext);
1434#endif
1435	CAS_CDSYNC(sc, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1436	CAS_WRITE_4(sc, CAS_TX_KICK3, sc->sc_txnext);
1437}
1438
1439static void
1440cas_start(struct ifnet *ifp)
1441{
1442	struct cas_softc *sc = ifp->if_softc;
1443	struct mbuf *m;
1444	int kicked, ntx;
1445
1446	CAS_LOCK(sc);
1447
1448	if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
1449	    IFF_DRV_RUNNING || (sc->sc_flags & CAS_LINK) == 0) {
1450		CAS_UNLOCK(sc);
1451		return;
1452	}
1453
1454	if (sc->sc_txfree < CAS_MAXTXFREE / 4)
1455		cas_tint(sc);
1456
1457#ifdef CAS_DEBUG
1458	CTR4(KTR_CAS, "%s: %s: txfree %d, txnext %d",
1459	    device_get_name(sc->sc_dev), __func__, sc->sc_txfree,
1460	    sc->sc_txnext);
1461#endif
1462	ntx = 0;
1463	kicked = 0;
1464	for (; !IFQ_DRV_IS_EMPTY(&ifp->if_snd) && sc->sc_txfree > 1;) {
1465		IFQ_DRV_DEQUEUE(&ifp->if_snd, m);
1466		if (m == NULL)
1467			break;
1468		if (cas_load_txmbuf(sc, &m) != 0) {
1469			if (m == NULL)
1470				break;
1471			ifp->if_drv_flags |= IFF_DRV_OACTIVE;
1472			IFQ_DRV_PREPEND(&ifp->if_snd, m);
1473			break;
1474		}
1475		if ((sc->sc_txnext % 4) == 0) {
1476			cas_txkick(sc);
1477			kicked = 1;
1478		} else
1479			kicked = 0;
1480		ntx++;
1481		BPF_MTAP(ifp, m);
1482	}
1483
1484	if (ntx > 0) {
1485		if (kicked == 0)
1486			cas_txkick(sc);
1487#ifdef CAS_DEBUG
1488		CTR2(KTR_CAS, "%s: packets enqueued, OWN on %d",
1489		    device_get_name(sc->sc_dev), sc->sc_txnext);
1490#endif
1491
1492		/* Set a watchdog timer in case the chip flakes out. */
1493		sc->sc_wdog_timer = 5;
1494#ifdef CAS_DEBUG
1495		CTR3(KTR_CAS, "%s: %s: watchdog %d",
1496		    device_get_name(sc->sc_dev), __func__,
1497		    sc->sc_wdog_timer);
1498#endif
1499	}
1500
1501	CAS_UNLOCK(sc);
1502}
1503
1504static void
1505cas_tint(struct cas_softc *sc)
1506{
1507	struct ifnet *ifp = sc->sc_ifp;
1508	struct cas_txsoft *txs;
1509	int progress;
1510	uint32_t txlast;
1511#ifdef CAS_DEBUG
1512	int i;
1513
1514	CAS_LOCK_ASSERT(sc, MA_OWNED);
1515
1516	CTR2(KTR_CAS, "%s: %s", device_get_name(sc->sc_dev), __func__);
1517#endif
1518
1519	/*
1520	 * Go through our TX list and free mbufs for those
1521	 * frames that have been transmitted.
1522	 */
1523	progress = 0;
1524	CAS_CDSYNC(sc, BUS_DMASYNC_POSTREAD);
1525	while ((txs = STAILQ_FIRST(&sc->sc_txdirtyq)) != NULL) {
1526#ifdef CAS_DEBUG
1527		if ((ifp->if_flags & IFF_DEBUG) != 0) {
1528			printf("    txsoft %p transmit chain:\n", txs);
1529			for (i = txs->txs_firstdesc;; i = CAS_NEXTTX(i)) {
1530				printf("descriptor %d: ", i);
1531				printf("cd_flags: 0x%016llx\t",
1532				    (long long)le64toh(
1533				    sc->sc_txdescs[i].cd_flags));
1534				printf("cd_buf_ptr: 0x%016llx\n",
1535				    (long long)le64toh(
1536				    sc->sc_txdescs[i].cd_buf_ptr));
1537				if (i == txs->txs_lastdesc)
1538					break;
1539			}
1540		}
1541#endif
1542
1543		/*
1544		 * In theory, we could harvest some descriptors before
1545		 * the ring is empty, but that's a bit complicated.
1546		 *
1547		 * CAS_TX_COMPn points to the last descriptor
1548		 * processed + 1.
1549		 */
1550		txlast = CAS_READ_4(sc, CAS_TX_COMP3);
1551#ifdef CAS_DEBUG
1552		CTR4(KTR_CAS, "%s: txs->txs_firstdesc = %d, "
1553		    "txs->txs_lastdesc = %d, txlast = %d",
1554		    __func__, txs->txs_firstdesc, txs->txs_lastdesc, txlast);
1555#endif
1556		if (txs->txs_firstdesc <= txs->txs_lastdesc) {
1557			if ((txlast >= txs->txs_firstdesc) &&
1558			    (txlast <= txs->txs_lastdesc))
1559				break;
1560		} else {
1561			/* Ick -- this command wraps. */
1562			if ((txlast >= txs->txs_firstdesc) ||
1563			    (txlast <= txs->txs_lastdesc))
1564				break;
1565		}
1566
1567#ifdef CAS_DEBUG
1568		CTR1(KTR_CAS, "%s: releasing a descriptor", __func__);
1569#endif
1570		STAILQ_REMOVE_HEAD(&sc->sc_txdirtyq, txs_q);
1571
1572		sc->sc_txfree += txs->txs_ndescs;
1573
1574		bus_dmamap_sync(sc->sc_tdmatag, txs->txs_dmamap,
1575		    BUS_DMASYNC_POSTWRITE);
1576		bus_dmamap_unload(sc->sc_tdmatag, txs->txs_dmamap);
1577		if (txs->txs_mbuf != NULL) {
1578			m_freem(txs->txs_mbuf);
1579			txs->txs_mbuf = NULL;
1580		}
1581
1582		STAILQ_INSERT_TAIL(&sc->sc_txfreeq, txs, txs_q);
1583
1584		if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
1585		progress = 1;
1586	}
1587
1588#ifdef CAS_DEBUG
1589	CTR5(KTR_CAS, "%s: CAS_TX_SM1 %x CAS_TX_SM2 %x CAS_TX_DESC_BASE %llx "
1590	    "CAS_TX_COMP3 %x",
1591	    __func__, CAS_READ_4(sc, CAS_TX_SM1), CAS_READ_4(sc, CAS_TX_SM2),
1592	    ((long long)CAS_READ_4(sc, CAS_TX_DESC3_BASE_HI) << 32) |
1593	    CAS_READ_4(sc, CAS_TX_DESC3_BASE_LO),
1594	    CAS_READ_4(sc, CAS_TX_COMP3));
1595#endif
1596
1597	if (progress) {
1598		/* We freed some descriptors, so reset IFF_DRV_OACTIVE. */
1599		ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1600		if (STAILQ_EMPTY(&sc->sc_txdirtyq))
1601			sc->sc_wdog_timer = 0;
1602	}
1603
1604#ifdef CAS_DEBUG
1605	CTR3(KTR_CAS, "%s: %s: watchdog %d",
1606	    device_get_name(sc->sc_dev), __func__, sc->sc_wdog_timer);
1607#endif
1608}
1609
1610static void
1611cas_rint_timeout(void *arg)
1612{
1613	struct cas_softc *sc = arg;
1614
1615	CAS_LOCK_ASSERT(sc, MA_OWNED);
1616
1617	cas_rint(sc);
1618}
1619
1620static void
1621cas_rint(struct cas_softc *sc)
1622{
1623	struct cas_rxdsoft *rxds, *rxds2;
1624	struct ifnet *ifp = sc->sc_ifp;
1625	struct mbuf *m, *m2;
1626	uint64_t word1, word2, word3, word4;
1627	uint32_t rxhead;
1628	u_int idx, idx2, len, off, skip;
1629
1630	CAS_LOCK_ASSERT(sc, MA_OWNED);
1631
1632	callout_stop(&sc->sc_rx_ch);
1633
1634#ifdef CAS_DEBUG
1635	CTR2(KTR_CAS, "%s: %s", device_get_name(sc->sc_dev), __func__);
1636#endif
1637
1638#define	PRINTWORD(n, delimiter)						\
1639	printf("word ## n: 0x%016llx%c", (long long)word ## n, delimiter)
1640
1641#define	SKIPASSERT(n)							\
1642	KASSERT(sc->sc_rxcomps[sc->sc_rxcptr].crc_word ## n == 0,	\
1643	    ("%s: word ## n not 0", __func__))
1644
1645#define	WORDTOH(n)							\
1646	word ## n = le64toh(sc->sc_rxcomps[sc->sc_rxcptr].crc_word ## n)
1647
1648	/*
1649	 * Read the completion head register once.  This limits
1650	 * how long the following loop can execute.
1651	 */
1652	rxhead = CAS_READ_4(sc, CAS_RX_COMP_HEAD);
1653#ifdef CAS_DEBUG
1654	CTR4(KTR_CAS, "%s: sc->sc_rxcptr %d, sc->sc_rxdptr %d, head %d",
1655	    __func__, sc->sc_rxcptr, sc->sc_rxdptr, rxhead);
1656#endif
1657	skip = 0;
1658	CAS_CDSYNC(sc, BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
1659	for (; sc->sc_rxcptr != rxhead;
1660	    sc->sc_rxcptr = CAS_NEXTRXCOMP(sc->sc_rxcptr)) {
1661		if (skip != 0) {
1662			SKIPASSERT(1);
1663			SKIPASSERT(2);
1664			SKIPASSERT(3);
1665
1666			--skip;
1667			goto skip;
1668		}
1669
1670		WORDTOH(1);
1671		WORDTOH(2);
1672		WORDTOH(3);
1673		WORDTOH(4);
1674
1675#ifdef CAS_DEBUG
1676		if ((ifp->if_flags & IFF_DEBUG) != 0) {
1677			printf("    completion %d: ", sc->sc_rxcptr);
1678			PRINTWORD(1, '\t');
1679			PRINTWORD(2, '\t');
1680			PRINTWORD(3, '\t');
1681			PRINTWORD(4, '\n');
1682		}
1683#endif
1684
1685		if (__predict_false(
1686		    (word1 & CAS_RC1_TYPE_MASK) == CAS_RC1_TYPE_HW ||
1687		    (word4 & CAS_RC4_ZERO) != 0)) {
1688			/*
1689			 * The descriptor is still marked as owned, although
1690			 * it is supposed to have completed.  This has been
1691			 * observed on some machines.  Just exiting here
1692			 * might leave the packet sitting around until another
1693			 * one arrives to trigger a new interrupt, which is
1694			 * generally undesirable, so set up a timeout.
1695			 */
1696			callout_reset(&sc->sc_rx_ch, CAS_RXOWN_TICKS,
1697			    cas_rint_timeout, sc);
1698			break;
1699		}
1700
1701		if (__predict_false(
1702		    (word4 & (CAS_RC4_BAD | CAS_RC4_LEN_MMATCH)) != 0)) {
1703			if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
1704			device_printf(sc->sc_dev,
1705			    "receive error: CRC error\n");
1706			continue;
1707		}
1708
1709		KASSERT(CAS_GET(word1, CAS_RC1_DATA_SIZE) == 0 ||
1710		    CAS_GET(word2, CAS_RC2_HDR_SIZE) == 0,
1711		    ("%s: data and header present", __func__));
1712		KASSERT((word1 & CAS_RC1_SPLIT_PKT) == 0 ||
1713		    CAS_GET(word2, CAS_RC2_HDR_SIZE) == 0,
1714		    ("%s: split and header present", __func__));
1715		KASSERT(CAS_GET(word1, CAS_RC1_DATA_SIZE) == 0 ||
1716		    (word1 & CAS_RC1_RELEASE_HDR) == 0,
1717		    ("%s: data present but header release", __func__));
1718		KASSERT(CAS_GET(word2, CAS_RC2_HDR_SIZE) == 0 ||
1719		    (word1 & CAS_RC1_RELEASE_DATA) == 0,
1720		    ("%s: header present but data release", __func__));
1721
1722		if ((len = CAS_GET(word2, CAS_RC2_HDR_SIZE)) != 0) {
1723			idx = CAS_GET(word2, CAS_RC2_HDR_INDEX);
1724			off = CAS_GET(word2, CAS_RC2_HDR_OFF);
1725#ifdef CAS_DEBUG
1726			CTR4(KTR_CAS, "%s: hdr at idx %d, off %d, len %d",
1727			    __func__, idx, off, len);
1728#endif
1729			rxds = &sc->sc_rxdsoft[idx];
1730			MGETHDR(m, M_NOWAIT, MT_DATA);
1731			if (m != NULL) {
1732				refcount_acquire(&rxds->rxds_refcount);
1733				bus_dmamap_sync(sc->sc_rdmatag,
1734				    rxds->rxds_dmamap, BUS_DMASYNC_POSTREAD);
1735#if __FreeBSD_version < 800016
1736				MEXTADD(m, (caddr_t)rxds->rxds_buf +
1737				    off * 256 + ETHER_ALIGN, len, cas_free,
1738				    rxds, M_RDONLY, EXT_NET_DRV);
1739#else
1740				MEXTADD(m, (caddr_t)rxds->rxds_buf +
1741				    off * 256 + ETHER_ALIGN, len, cas_free,
1742				    sc, (void *)(uintptr_t)idx,
1743				    M_RDONLY, EXT_NET_DRV);
1744#endif
1745				if ((m->m_flags & M_EXT) == 0) {
1746					m_freem(m);
1747					m = NULL;
1748				}
1749			}
1750			if (m != NULL) {
1751				m->m_pkthdr.rcvif = ifp;
1752				m->m_pkthdr.len = m->m_len = len;
1753				if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
1754				if ((ifp->if_capenable & IFCAP_RXCSUM) != 0)
1755					cas_rxcksum(m, CAS_GET(word4,
1756					    CAS_RC4_TCP_CSUM));
1757				/* Pass it on. */
1758				CAS_UNLOCK(sc);
1759				(*ifp->if_input)(ifp, m);
1760				CAS_LOCK(sc);
1761			} else
1762				if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1763
1764			if ((word1 & CAS_RC1_RELEASE_HDR) != 0 &&
1765			    refcount_release(&rxds->rxds_refcount) != 0)
1766				cas_add_rxdesc(sc, idx);
1767		} else if ((len = CAS_GET(word1, CAS_RC1_DATA_SIZE)) != 0) {
1768			idx = CAS_GET(word1, CAS_RC1_DATA_INDEX);
1769			off = CAS_GET(word1, CAS_RC1_DATA_OFF);
1770#ifdef CAS_DEBUG
1771			CTR4(KTR_CAS, "%s: data at idx %d, off %d, len %d",
1772			    __func__, idx, off, len);
1773#endif
1774			rxds = &sc->sc_rxdsoft[idx];
1775			MGETHDR(m, M_NOWAIT, MT_DATA);
1776			if (m != NULL) {
1777				refcount_acquire(&rxds->rxds_refcount);
1778				off += ETHER_ALIGN;
1779				m->m_len = min(CAS_PAGE_SIZE - off, len);
1780				bus_dmamap_sync(sc->sc_rdmatag,
1781				    rxds->rxds_dmamap, BUS_DMASYNC_POSTREAD);
1782#if __FreeBSD_version < 800016
1783				MEXTADD(m, (caddr_t)rxds->rxds_buf + off,
1784				    m->m_len, cas_free, rxds, M_RDONLY,
1785				    EXT_NET_DRV);
1786#else
1787				MEXTADD(m, (caddr_t)rxds->rxds_buf + off,
1788				    m->m_len, cas_free, sc,
1789				    (void *)(uintptr_t)idx, M_RDONLY,
1790				    EXT_NET_DRV);
1791#endif
1792				if ((m->m_flags & M_EXT) == 0) {
1793					m_freem(m);
1794					m = NULL;
1795				}
1796			}
1797			idx2 = 0;
1798			m2 = NULL;
1799			rxds2 = NULL;
1800			if ((word1 & CAS_RC1_SPLIT_PKT) != 0) {
1801				KASSERT((word1 & CAS_RC1_RELEASE_NEXT) != 0,
1802				    ("%s: split but no release next",
1803				    __func__));
1804
1805				idx2 = CAS_GET(word2, CAS_RC2_NEXT_INDEX);
1806#ifdef CAS_DEBUG
1807				CTR2(KTR_CAS, "%s: split at idx %d",
1808				    __func__, idx2);
1809#endif
1810				rxds2 = &sc->sc_rxdsoft[idx2];
1811				if (m != NULL) {
1812					MGET(m2, M_NOWAIT, MT_DATA);
1813					if (m2 != NULL) {
1814						refcount_acquire(
1815						    &rxds2->rxds_refcount);
1816						m2->m_len = len - m->m_len;
1817						bus_dmamap_sync(
1818						    sc->sc_rdmatag,
1819						    rxds2->rxds_dmamap,
1820						    BUS_DMASYNC_POSTREAD);
1821#if __FreeBSD_version < 800016
1822						MEXTADD(m2,
1823						    (caddr_t)rxds2->rxds_buf,
1824						    m2->m_len, cas_free,
1825						    rxds2, M_RDONLY,
1826						    EXT_NET_DRV);
1827#else
1828						MEXTADD(m2,
1829						    (caddr_t)rxds2->rxds_buf,
1830						    m2->m_len, cas_free, sc,
1831						    (void *)(uintptr_t)idx2,
1832						    M_RDONLY, EXT_NET_DRV);
1833#endif
1834						if ((m2->m_flags & M_EXT) ==
1835						    0) {
1836							m_freem(m2);
1837							m2 = NULL;
1838						}
1839					}
1840				}
1841				if (m2 != NULL)
1842					m->m_next = m2;
1843				else if (m != NULL) {
1844					m_freem(m);
1845					m = NULL;
1846				}
1847			}
1848			if (m != NULL) {
1849				m->m_pkthdr.rcvif = ifp;
1850				m->m_pkthdr.len = len;
1851				if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
1852				if ((ifp->if_capenable & IFCAP_RXCSUM) != 0)
1853					cas_rxcksum(m, CAS_GET(word4,
1854					    CAS_RC4_TCP_CSUM));
1855				/* Pass it on. */
1856				CAS_UNLOCK(sc);
1857				(*ifp->if_input)(ifp, m);
1858				CAS_LOCK(sc);
1859			} else
1860				if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1861
1862			if ((word1 & CAS_RC1_RELEASE_DATA) != 0 &&
1863			    refcount_release(&rxds->rxds_refcount) != 0)
1864				cas_add_rxdesc(sc, idx);
1865			if ((word1 & CAS_RC1_SPLIT_PKT) != 0 &&
1866			    refcount_release(&rxds2->rxds_refcount) != 0)
1867				cas_add_rxdesc(sc, idx2);
1868		}
1869
1870		skip = CAS_GET(word1, CAS_RC1_SKIP);
1871
1872 skip:
1873		cas_rxcompinit(&sc->sc_rxcomps[sc->sc_rxcptr]);
1874		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1875			break;
1876	}
1877	CAS_CDSYNC(sc, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1878	CAS_WRITE_4(sc, CAS_RX_COMP_TAIL, sc->sc_rxcptr);
1879
1880#undef PRINTWORD
1881#undef SKIPASSERT
1882#undef WORDTOH
1883
1884#ifdef CAS_DEBUG
1885	CTR4(KTR_CAS, "%s: done sc->sc_rxcptr %d, sc->sc_rxdptr %d, head %d",
1886	    __func__, sc->sc_rxcptr, sc->sc_rxdptr,
1887	    CAS_READ_4(sc, CAS_RX_COMP_HEAD));
1888#endif
1889}
1890
1891static void
1892cas_free(struct mbuf *m, void *arg1, void *arg2)
1893{
1894	struct cas_rxdsoft *rxds;
1895	struct cas_softc *sc;
1896	u_int idx, locked;
1897
1898#if __FreeBSD_version < 800016
1899	rxds = arg2;
1900	sc = rxds->rxds_sc;
1901	idx = rxds->rxds_idx;
1902#else
1903	sc = arg1;
1904	idx = (uintptr_t)arg2;
1905	rxds = &sc->sc_rxdsoft[idx];
1906#endif
1907	if (refcount_release(&rxds->rxds_refcount) == 0)
1908		return;
1909
1910	/*
1911	 * NB: this function can be called via m_freem(9) within
1912	 * this driver!
1913	 */
1914	if ((locked = CAS_LOCK_OWNED(sc)) == 0)
1915		CAS_LOCK(sc);
1916	cas_add_rxdesc(sc, idx);
1917	if (locked == 0)
1918		CAS_UNLOCK(sc);
1919}
1920
1921static inline void
1922cas_add_rxdesc(struct cas_softc *sc, u_int idx)
1923{
1924
1925	CAS_LOCK_ASSERT(sc, MA_OWNED);
1926
1927	bus_dmamap_sync(sc->sc_rdmatag, sc->sc_rxdsoft[idx].rxds_dmamap,
1928	    BUS_DMASYNC_PREREAD);
1929	CAS_UPDATE_RXDESC(sc, sc->sc_rxdptr, idx);
1930	sc->sc_rxdptr = CAS_NEXTRXDESC(sc->sc_rxdptr);
1931
1932	/*
1933	 * Update the RX kick register.  This register has to point to the
1934	 * descriptor after the last valid one (before the current batch)
1935	 * and for optimum performance should be incremented in multiples
1936	 * of 4 (the DMA engine fetches/updates descriptors in batches of 4).
1937	 */
1938	if ((sc->sc_rxdptr % 4) == 0) {
1939		CAS_CDSYNC(sc, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1940		CAS_WRITE_4(sc, CAS_RX_KICK,
1941		    (sc->sc_rxdptr + CAS_NRXDESC - 4) & CAS_NRXDESC_MASK);
1942	}
1943}
1944
1945static void
1946cas_eint(struct cas_softc *sc, u_int status)
1947{
1948	struct ifnet *ifp = sc->sc_ifp;
1949
1950	CAS_LOCK_ASSERT(sc, MA_OWNED);
1951
1952	if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
1953
1954	device_printf(sc->sc_dev, "%s: status 0x%x", __func__, status);
1955	if ((status & CAS_INTR_PCI_ERROR_INT) != 0) {
1956		status = CAS_READ_4(sc, CAS_ERROR_STATUS);
1957		printf(", PCI bus error 0x%x", status);
1958		if ((status & CAS_ERROR_OTHER) != 0) {
1959			status = pci_read_config(sc->sc_dev, PCIR_STATUS, 2);
1960			printf(", PCI status 0x%x", status);
1961			pci_write_config(sc->sc_dev, PCIR_STATUS, status, 2);
1962		}
1963	}
1964	printf("\n");
1965
1966	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1967	cas_init_locked(sc);
1968	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
1969		taskqueue_enqueue(sc->sc_tq, &sc->sc_tx_task);
1970}
1971
1972static int
1973cas_intr(void *v)
1974{
1975	struct cas_softc *sc = v;
1976
1977	if (__predict_false((CAS_READ_4(sc, CAS_STATUS_ALIAS) &
1978	    CAS_INTR_SUMMARY) == 0))
1979		return (FILTER_STRAY);
1980
1981	/* Disable interrupts. */
1982	CAS_WRITE_4(sc, CAS_INTMASK, 0xffffffff);
1983	taskqueue_enqueue(sc->sc_tq, &sc->sc_intr_task);
1984
1985	return (FILTER_HANDLED);
1986}
1987
1988static void
1989cas_intr_task(void *arg, int pending __unused)
1990{
1991	struct cas_softc *sc = arg;
1992	struct ifnet *ifp = sc->sc_ifp;
1993	uint32_t status, status2;
1994
1995	CAS_LOCK_ASSERT(sc, MA_NOTOWNED);
1996
1997	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1998		return;
1999
2000	status = CAS_READ_4(sc, CAS_STATUS);
2001	if (__predict_false((status & CAS_INTR_SUMMARY) == 0))
2002		goto done;
2003
2004	CAS_LOCK(sc);
2005#ifdef CAS_DEBUG
2006	CTR4(KTR_CAS, "%s: %s: cplt %x, status %x",
2007	    device_get_name(sc->sc_dev), __func__,
2008	    (status >> CAS_STATUS_TX_COMP3_SHFT), (u_int)status);
2009
2010	/*
2011	 * PCS interrupts must be cleared, otherwise no traffic is passed!
2012	 */
2013	if ((status & CAS_INTR_PCS_INT) != 0) {
2014		status2 =
2015		    CAS_READ_4(sc, CAS_PCS_INTR_STATUS) |
2016		    CAS_READ_4(sc, CAS_PCS_INTR_STATUS);
2017		if ((status2 & CAS_PCS_INTR_LINK) != 0)
2018			device_printf(sc->sc_dev,
2019			    "%s: PCS link status changed\n", __func__);
2020	}
2021	if ((status & CAS_MAC_CTRL_STATUS) != 0) {
2022		status2 = CAS_READ_4(sc, CAS_MAC_CTRL_STATUS);
2023		if ((status2 & CAS_MAC_CTRL_PAUSE) != 0)
2024			device_printf(sc->sc_dev,
2025			    "%s: PAUSE received (PAUSE time %d slots)\n",
2026			    __func__,
2027			    (status2 & CAS_MAC_CTRL_STATUS_PT_MASK) >>
2028			    CAS_MAC_CTRL_STATUS_PT_SHFT);
2029		if ((status2 & CAS_MAC_CTRL_PAUSE) != 0)
2030			device_printf(sc->sc_dev,
2031			    "%s: transited to PAUSE state\n", __func__);
2032		if ((status2 & CAS_MAC_CTRL_NON_PAUSE) != 0)
2033			device_printf(sc->sc_dev,
2034			    "%s: transited to non-PAUSE state\n", __func__);
2035	}
2036	if ((status & CAS_INTR_MIF) != 0)
2037		device_printf(sc->sc_dev, "%s: MIF interrupt\n", __func__);
2038#endif
2039
2040	if (__predict_false((status &
2041	    (CAS_INTR_TX_TAG_ERR | CAS_INTR_RX_TAG_ERR |
2042	    CAS_INTR_RX_LEN_MMATCH | CAS_INTR_PCI_ERROR_INT)) != 0)) {
2043		cas_eint(sc, status);
2044		CAS_UNLOCK(sc);
2045		return;
2046	}
2047
2048	if (__predict_false(status & CAS_INTR_TX_MAC_INT)) {
2049		status2 = CAS_READ_4(sc, CAS_MAC_TX_STATUS);
2050		if ((status2 &
2051		    (CAS_MAC_TX_UNDERRUN | CAS_MAC_TX_MAX_PKT_ERR)) != 0)
2052			if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
2053		else if ((status2 & ~CAS_MAC_TX_FRAME_XMTD) != 0)
2054			device_printf(sc->sc_dev,
2055			    "MAC TX fault, status %x\n", status2);
2056	}
2057
2058	if (__predict_false(status & CAS_INTR_RX_MAC_INT)) {
2059		status2 = CAS_READ_4(sc, CAS_MAC_RX_STATUS);
2060		if ((status2 & CAS_MAC_RX_OVERFLOW) != 0)
2061			if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
2062		else if ((status2 & ~CAS_MAC_RX_FRAME_RCVD) != 0)
2063			device_printf(sc->sc_dev,
2064			    "MAC RX fault, status %x\n", status2);
2065	}
2066
2067	if ((status &
2068	    (CAS_INTR_RX_DONE | CAS_INTR_RX_BUF_NA | CAS_INTR_RX_COMP_FULL |
2069	    CAS_INTR_RX_BUF_AEMPTY | CAS_INTR_RX_COMP_AFULL)) != 0) {
2070		cas_rint(sc);
2071#ifdef CAS_DEBUG
2072		if (__predict_false((status &
2073		    (CAS_INTR_RX_BUF_NA | CAS_INTR_RX_COMP_FULL |
2074		    CAS_INTR_RX_BUF_AEMPTY | CAS_INTR_RX_COMP_AFULL)) != 0))
2075			device_printf(sc->sc_dev,
2076			    "RX fault, status %x\n", status);
2077#endif
2078	}
2079
2080	if ((status &
2081	    (CAS_INTR_TX_INT_ME | CAS_INTR_TX_ALL | CAS_INTR_TX_DONE)) != 0)
2082		cas_tint(sc);
2083
2084	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2085		CAS_UNLOCK(sc);
2086		return;
2087	} else if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2088		taskqueue_enqueue(sc->sc_tq, &sc->sc_tx_task);
2089	CAS_UNLOCK(sc);
2090
2091	status = CAS_READ_4(sc, CAS_STATUS_ALIAS);
2092	if (__predict_false((status & CAS_INTR_SUMMARY) != 0)) {
2093		taskqueue_enqueue(sc->sc_tq, &sc->sc_intr_task);
2094		return;
2095	}
2096
2097 done:
2098	/* Re-enable interrupts. */
2099	CAS_WRITE_4(sc, CAS_INTMASK,
2100	    ~(CAS_INTR_TX_INT_ME | CAS_INTR_TX_TAG_ERR |
2101	    CAS_INTR_RX_DONE | CAS_INTR_RX_BUF_NA | CAS_INTR_RX_TAG_ERR |
2102	    CAS_INTR_RX_COMP_FULL | CAS_INTR_RX_BUF_AEMPTY |
2103	    CAS_INTR_RX_COMP_AFULL | CAS_INTR_RX_LEN_MMATCH |
2104	    CAS_INTR_PCI_ERROR_INT
2105#ifdef CAS_DEBUG
2106	    | CAS_INTR_PCS_INT | CAS_INTR_MIF
2107#endif
2108	));
2109}
2110
2111static void
2112cas_watchdog(struct cas_softc *sc)
2113{
2114	struct ifnet *ifp = sc->sc_ifp;
2115
2116	CAS_LOCK_ASSERT(sc, MA_OWNED);
2117
2118#ifdef CAS_DEBUG
2119	CTR4(KTR_CAS,
2120	    "%s: CAS_RX_CONF %x CAS_MAC_RX_STATUS %x CAS_MAC_RX_CONF %x",
2121	    __func__, CAS_READ_4(sc, CAS_RX_CONF),
2122	    CAS_READ_4(sc, CAS_MAC_RX_STATUS),
2123	    CAS_READ_4(sc, CAS_MAC_RX_CONF));
2124	CTR4(KTR_CAS,
2125	    "%s: CAS_TX_CONF %x CAS_MAC_TX_STATUS %x CAS_MAC_TX_CONF %x",
2126	    __func__, CAS_READ_4(sc, CAS_TX_CONF),
2127	    CAS_READ_4(sc, CAS_MAC_TX_STATUS),
2128	    CAS_READ_4(sc, CAS_MAC_TX_CONF));
2129#endif
2130
2131	if (sc->sc_wdog_timer == 0 || --sc->sc_wdog_timer != 0)
2132		return;
2133
2134	if ((sc->sc_flags & CAS_LINK) != 0)
2135		device_printf(sc->sc_dev, "device timeout\n");
2136	else if (bootverbose)
2137		device_printf(sc->sc_dev, "device timeout (no link)\n");
2138	if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
2139
2140	/* Try to get more packets going. */
2141	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2142	cas_init_locked(sc);
2143	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2144		taskqueue_enqueue(sc->sc_tq, &sc->sc_tx_task);
2145}
2146
2147static void
2148cas_mifinit(struct cas_softc *sc)
2149{
2150
2151	/* Configure the MIF in frame mode. */
2152	CAS_WRITE_4(sc, CAS_MIF_CONF,
2153	    CAS_READ_4(sc, CAS_MIF_CONF) & ~CAS_MIF_CONF_BB_MODE);
2154	CAS_BARRIER(sc, CAS_MIF_CONF, 4,
2155	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2156}
2157
2158/*
2159 * MII interface
2160 *
2161 * The MII interface supports at least three different operating modes:
2162 *
2163 * Bitbang mode is implemented using data, clock and output enable registers.
2164 *
2165 * Frame mode is implemented by loading a complete frame into the frame
2166 * register and polling the valid bit for completion.
2167 *
2168 * Polling mode uses the frame register but completion is indicated by
2169 * an interrupt.
2170 *
2171 */
2172static int
2173cas_mii_readreg(device_t dev, int phy, int reg)
2174{
2175	struct cas_softc *sc;
2176	int n;
2177	uint32_t v;
2178
2179#ifdef CAS_DEBUG_PHY
2180	printf("%s: phy %d reg %d\n", __func__, phy, reg);
2181#endif
2182
2183	sc = device_get_softc(dev);
2184	if ((sc->sc_flags & CAS_SERDES) != 0) {
2185		switch (reg) {
2186		case MII_BMCR:
2187			reg = CAS_PCS_CTRL;
2188			break;
2189		case MII_BMSR:
2190			reg = CAS_PCS_STATUS;
2191			break;
2192		case MII_PHYIDR1:
2193		case MII_PHYIDR2:
2194			return (0);
2195		case MII_ANAR:
2196			reg = CAS_PCS_ANAR;
2197			break;
2198		case MII_ANLPAR:
2199			reg = CAS_PCS_ANLPAR;
2200			break;
2201		case MII_EXTSR:
2202			return (EXTSR_1000XFDX | EXTSR_1000XHDX);
2203		default:
2204			device_printf(sc->sc_dev,
2205			    "%s: unhandled register %d\n", __func__, reg);
2206			return (0);
2207		}
2208		return (CAS_READ_4(sc, reg));
2209	}
2210
2211	/* Construct the frame command. */
2212	v = CAS_MIF_FRAME_READ |
2213	    (phy << CAS_MIF_FRAME_PHY_SHFT) |
2214	    (reg << CAS_MIF_FRAME_REG_SHFT);
2215
2216	CAS_WRITE_4(sc, CAS_MIF_FRAME, v);
2217	CAS_BARRIER(sc, CAS_MIF_FRAME, 4,
2218	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2219	for (n = 0; n < 100; n++) {
2220		DELAY(1);
2221		v = CAS_READ_4(sc, CAS_MIF_FRAME);
2222		if (v & CAS_MIF_FRAME_TA_LSB)
2223			return (v & CAS_MIF_FRAME_DATA);
2224	}
2225
2226	device_printf(sc->sc_dev, "%s: timed out\n", __func__);
2227	return (0);
2228}
2229
2230static int
2231cas_mii_writereg(device_t dev, int phy, int reg, int val)
2232{
2233	struct cas_softc *sc;
2234	int n;
2235	uint32_t v;
2236
2237#ifdef CAS_DEBUG_PHY
2238	printf("%s: phy %d reg %d val %x\n", phy, reg, val, __func__);
2239#endif
2240
2241	sc = device_get_softc(dev);
2242	if ((sc->sc_flags & CAS_SERDES) != 0) {
2243		switch (reg) {
2244		case MII_BMSR:
2245			reg = CAS_PCS_STATUS;
2246			break;
2247		case MII_BMCR:
2248			reg = CAS_PCS_CTRL;
2249			if ((val & CAS_PCS_CTRL_RESET) == 0)
2250				break;
2251			CAS_WRITE_4(sc, CAS_PCS_CTRL, val);
2252			CAS_BARRIER(sc, CAS_PCS_CTRL, 4,
2253			    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2254			if (!cas_bitwait(sc, CAS_PCS_CTRL,
2255			    CAS_PCS_CTRL_RESET, 0))
2256				device_printf(sc->sc_dev,
2257				    "cannot reset PCS\n");
2258			/* FALLTHROUGH */
2259		case MII_ANAR:
2260			CAS_WRITE_4(sc, CAS_PCS_CONF, 0);
2261			CAS_BARRIER(sc, CAS_PCS_CONF, 4,
2262			    BUS_SPACE_BARRIER_WRITE);
2263			CAS_WRITE_4(sc, CAS_PCS_ANAR, val);
2264			CAS_BARRIER(sc, CAS_PCS_ANAR, 4,
2265			    BUS_SPACE_BARRIER_WRITE);
2266			CAS_WRITE_4(sc, CAS_PCS_SERDES_CTRL,
2267			    CAS_PCS_SERDES_CTRL_ESD);
2268			CAS_BARRIER(sc, CAS_PCS_CONF, 4,
2269			    BUS_SPACE_BARRIER_WRITE);
2270			CAS_WRITE_4(sc, CAS_PCS_CONF,
2271			    CAS_PCS_CONF_EN);
2272			CAS_BARRIER(sc, CAS_PCS_CONF, 4,
2273			    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2274			return (0);
2275		case MII_ANLPAR:
2276			reg = CAS_PCS_ANLPAR;
2277			break;
2278		default:
2279			device_printf(sc->sc_dev,
2280			    "%s: unhandled register %d\n", __func__, reg);
2281			return (0);
2282		}
2283		CAS_WRITE_4(sc, reg, val);
2284		CAS_BARRIER(sc, reg, 4,
2285		    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2286		return (0);
2287	}
2288
2289	/* Construct the frame command. */
2290	v = CAS_MIF_FRAME_WRITE |
2291	    (phy << CAS_MIF_FRAME_PHY_SHFT) |
2292	    (reg << CAS_MIF_FRAME_REG_SHFT) |
2293	    (val & CAS_MIF_FRAME_DATA);
2294
2295	CAS_WRITE_4(sc, CAS_MIF_FRAME, v);
2296	CAS_BARRIER(sc, CAS_MIF_FRAME, 4,
2297	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2298	for (n = 0; n < 100; n++) {
2299		DELAY(1);
2300		v = CAS_READ_4(sc, CAS_MIF_FRAME);
2301		if (v & CAS_MIF_FRAME_TA_LSB)
2302			return (1);
2303	}
2304
2305	device_printf(sc->sc_dev, "%s: timed out\n", __func__);
2306	return (0);
2307}
2308
2309static void
2310cas_mii_statchg(device_t dev)
2311{
2312	struct cas_softc *sc;
2313	struct ifnet *ifp;
2314	int gigabit;
2315	uint32_t rxcfg, txcfg, v;
2316
2317	sc = device_get_softc(dev);
2318	ifp = sc->sc_ifp;
2319
2320	CAS_LOCK_ASSERT(sc, MA_OWNED);
2321
2322#ifdef CAS_DEBUG
2323	if ((ifp->if_flags & IFF_DEBUG) != 0)
2324		device_printf(sc->sc_dev, "%s: status changen", __func__);
2325#endif
2326
2327	if ((sc->sc_mii->mii_media_status & IFM_ACTIVE) != 0 &&
2328	    IFM_SUBTYPE(sc->sc_mii->mii_media_active) != IFM_NONE)
2329		sc->sc_flags |= CAS_LINK;
2330	else
2331		sc->sc_flags &= ~CAS_LINK;
2332
2333	switch (IFM_SUBTYPE(sc->sc_mii->mii_media_active)) {
2334	case IFM_1000_SX:
2335	case IFM_1000_LX:
2336	case IFM_1000_CX:
2337	case IFM_1000_T:
2338		gigabit = 1;
2339		break;
2340	default:
2341		gigabit = 0;
2342	}
2343
2344	/*
2345	 * The configuration done here corresponds to the steps F) and
2346	 * G) and as far as enabling of RX and TX MAC goes also step H)
2347	 * of the initialization sequence outlined in section 11.2.1 of
2348	 * the Cassini+ ASIC Specification.
2349	 */
2350
2351	rxcfg = sc->sc_mac_rxcfg;
2352	rxcfg &= ~CAS_MAC_RX_CONF_CARR;
2353	txcfg = CAS_MAC_TX_CONF_EN_IPG0 | CAS_MAC_TX_CONF_NGU |
2354	    CAS_MAC_TX_CONF_NGUL;
2355	if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) & IFM_FDX) != 0)
2356		txcfg |= CAS_MAC_TX_CONF_ICARR | CAS_MAC_TX_CONF_ICOLLIS;
2357	else if (gigabit != 0) {
2358		rxcfg |= CAS_MAC_RX_CONF_CARR;
2359		txcfg |= CAS_MAC_TX_CONF_CARR;
2360	}
2361	(void)cas_disable_tx(sc);
2362	CAS_WRITE_4(sc, CAS_MAC_TX_CONF, txcfg);
2363	(void)cas_disable_rx(sc);
2364	CAS_WRITE_4(sc, CAS_MAC_RX_CONF, rxcfg);
2365
2366	v = CAS_READ_4(sc, CAS_MAC_CTRL_CONF) &
2367	    ~(CAS_MAC_CTRL_CONF_TXP | CAS_MAC_CTRL_CONF_RXP);
2368	if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) &
2369	    IFM_ETH_RXPAUSE) != 0)
2370		v |= CAS_MAC_CTRL_CONF_RXP;
2371	if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) &
2372	    IFM_ETH_TXPAUSE) != 0)
2373		v |= CAS_MAC_CTRL_CONF_TXP;
2374	CAS_WRITE_4(sc, CAS_MAC_CTRL_CONF, v);
2375
2376	/*
2377	 * All supported chips have a bug causing incorrect checksum
2378	 * to be calculated when letting them strip the FCS in half-
2379	 * duplex mode.  In theory we could disable FCS stripping and
2380	 * manually adjust the checksum accordingly.  It seems to make
2381	 * more sense to optimze for the common case and just disable
2382	 * hardware checksumming in half-duplex mode though.
2383	 */
2384	if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) & IFM_FDX) == 0) {
2385		ifp->if_capenable &= ~IFCAP_HWCSUM;
2386		ifp->if_hwassist = 0;
2387	} else if ((sc->sc_flags & CAS_NO_CSUM) == 0) {
2388		ifp->if_capenable = ifp->if_capabilities;
2389		ifp->if_hwassist = CAS_CSUM_FEATURES;
2390	}
2391
2392	if (sc->sc_variant == CAS_SATURN) {
2393		if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) & IFM_FDX) == 0)
2394			/* silicon bug workaround */
2395			CAS_WRITE_4(sc, CAS_MAC_PREAMBLE_LEN, 0x41);
2396		else
2397			CAS_WRITE_4(sc, CAS_MAC_PREAMBLE_LEN, 0x7);
2398	}
2399
2400	if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) & IFM_FDX) == 0 &&
2401	    gigabit != 0)
2402		CAS_WRITE_4(sc, CAS_MAC_SLOT_TIME,
2403		    CAS_MAC_SLOT_TIME_CARR);
2404	else
2405		CAS_WRITE_4(sc, CAS_MAC_SLOT_TIME,
2406		    CAS_MAC_SLOT_TIME_NORM);
2407
2408	/* XIF Configuration */
2409	v = CAS_MAC_XIF_CONF_TX_OE | CAS_MAC_XIF_CONF_LNKLED;
2410	if ((sc->sc_flags & CAS_SERDES) == 0) {
2411		if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) & IFM_FDX) == 0)
2412			v |= CAS_MAC_XIF_CONF_NOECHO;
2413		v |= CAS_MAC_XIF_CONF_BUF_OE;
2414	}
2415	if (gigabit != 0)
2416		v |= CAS_MAC_XIF_CONF_GMII;
2417	if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) & IFM_FDX) != 0)
2418		v |= CAS_MAC_XIF_CONF_FDXLED;
2419	CAS_WRITE_4(sc, CAS_MAC_XIF_CONF, v);
2420
2421	sc->sc_mac_rxcfg = rxcfg;
2422	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0 &&
2423	    (sc->sc_flags & CAS_LINK) != 0) {
2424		CAS_WRITE_4(sc, CAS_MAC_TX_CONF,
2425		    txcfg | CAS_MAC_TX_CONF_EN);
2426		CAS_WRITE_4(sc, CAS_MAC_RX_CONF,
2427		    rxcfg | CAS_MAC_RX_CONF_EN);
2428	}
2429}
2430
2431static int
2432cas_mediachange(struct ifnet *ifp)
2433{
2434	struct cas_softc *sc = ifp->if_softc;
2435	int error;
2436
2437	/* XXX add support for serial media. */
2438
2439	CAS_LOCK(sc);
2440	error = mii_mediachg(sc->sc_mii);
2441	CAS_UNLOCK(sc);
2442	return (error);
2443}
2444
2445static void
2446cas_mediastatus(struct ifnet *ifp, struct ifmediareq *ifmr)
2447{
2448	struct cas_softc *sc = ifp->if_softc;
2449
2450	CAS_LOCK(sc);
2451	if ((ifp->if_flags & IFF_UP) == 0) {
2452		CAS_UNLOCK(sc);
2453		return;
2454	}
2455
2456	mii_pollstat(sc->sc_mii);
2457	ifmr->ifm_active = sc->sc_mii->mii_media_active;
2458	ifmr->ifm_status = sc->sc_mii->mii_media_status;
2459	CAS_UNLOCK(sc);
2460}
2461
2462static int
2463cas_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
2464{
2465	struct cas_softc *sc = ifp->if_softc;
2466	struct ifreq *ifr = (struct ifreq *)data;
2467	int error;
2468
2469	error = 0;
2470	switch (cmd) {
2471	case SIOCSIFFLAGS:
2472		CAS_LOCK(sc);
2473		if ((ifp->if_flags & IFF_UP) != 0) {
2474			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0 &&
2475			    ((ifp->if_flags ^ sc->sc_ifflags) &
2476			    (IFF_ALLMULTI | IFF_PROMISC)) != 0)
2477				cas_setladrf(sc);
2478			else
2479				cas_init_locked(sc);
2480		} else if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
2481			cas_stop(ifp);
2482		sc->sc_ifflags = ifp->if_flags;
2483		CAS_UNLOCK(sc);
2484		break;
2485	case SIOCSIFCAP:
2486		CAS_LOCK(sc);
2487		if ((sc->sc_flags & CAS_NO_CSUM) != 0) {
2488			error = EINVAL;
2489			CAS_UNLOCK(sc);
2490			break;
2491		}
2492		ifp->if_capenable = ifr->ifr_reqcap;
2493		if ((ifp->if_capenable & IFCAP_TXCSUM) != 0)
2494			ifp->if_hwassist = CAS_CSUM_FEATURES;
2495		else
2496			ifp->if_hwassist = 0;
2497		CAS_UNLOCK(sc);
2498		break;
2499	case SIOCADDMULTI:
2500	case SIOCDELMULTI:
2501		CAS_LOCK(sc);
2502		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
2503			cas_setladrf(sc);
2504		CAS_UNLOCK(sc);
2505		break;
2506	case SIOCSIFMTU:
2507		if ((ifr->ifr_mtu < ETHERMIN) ||
2508		    (ifr->ifr_mtu > ETHERMTU_JUMBO))
2509			error = EINVAL;
2510		else
2511			ifp->if_mtu = ifr->ifr_mtu;
2512		break;
2513	case SIOCGIFMEDIA:
2514	case SIOCSIFMEDIA:
2515		error = ifmedia_ioctl(ifp, ifr, &sc->sc_mii->mii_media, cmd);
2516		break;
2517	default:
2518		error = ether_ioctl(ifp, cmd, data);
2519		break;
2520	}
2521
2522	return (error);
2523}
2524
2525static void
2526cas_setladrf(struct cas_softc *sc)
2527{
2528	struct ifnet *ifp = sc->sc_ifp;
2529	struct ifmultiaddr *inm;
2530	int i;
2531	uint32_t hash[16];
2532	uint32_t crc, v;
2533
2534	CAS_LOCK_ASSERT(sc, MA_OWNED);
2535
2536	/*
2537	 * Turn off the RX MAC and the hash filter as required by the Sun
2538	 * Cassini programming restrictions.
2539	 */
2540	v = sc->sc_mac_rxcfg & ~(CAS_MAC_RX_CONF_HFILTER |
2541	    CAS_MAC_RX_CONF_EN);
2542	CAS_WRITE_4(sc, CAS_MAC_RX_CONF, v);
2543	CAS_BARRIER(sc, CAS_MAC_RX_CONF, 4,
2544	    BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
2545	if (!cas_bitwait(sc, CAS_MAC_RX_CONF, CAS_MAC_RX_CONF_HFILTER |
2546	    CAS_MAC_RX_CONF_EN, 0))
2547		device_printf(sc->sc_dev,
2548		    "cannot disable RX MAC or hash filter\n");
2549
2550	v &= ~(CAS_MAC_RX_CONF_PROMISC | CAS_MAC_RX_CONF_PGRP);
2551	if ((ifp->if_flags & IFF_PROMISC) != 0) {
2552		v |= CAS_MAC_RX_CONF_PROMISC;
2553		goto chipit;
2554	}
2555	if ((ifp->if_flags & IFF_ALLMULTI) != 0) {
2556		v |= CAS_MAC_RX_CONF_PGRP;
2557		goto chipit;
2558	}
2559
2560	/*
2561	 * Set up multicast address filter by passing all multicast
2562	 * addresses through a crc generator, and then using the high
2563	 * order 8 bits as an index into the 256 bit logical address
2564	 * filter.  The high order 4 bits selects the word, while the
2565	 * other 4 bits select the bit within the word (where bit 0
2566	 * is the MSB).
2567	 */
2568
2569	/* Clear the hash table. */
2570	memset(hash, 0, sizeof(hash));
2571
2572	if_maddr_rlock(ifp);
2573	TAILQ_FOREACH(inm, &ifp->if_multiaddrs, ifma_link) {
2574		if (inm->ifma_addr->sa_family != AF_LINK)
2575			continue;
2576		crc = ether_crc32_le(LLADDR((struct sockaddr_dl *)
2577		    inm->ifma_addr), ETHER_ADDR_LEN);
2578
2579		/* We just want the 8 most significant bits. */
2580		crc >>= 24;
2581
2582		/* Set the corresponding bit in the filter. */
2583		hash[crc >> 4] |= 1 << (15 - (crc & 15));
2584	}
2585	if_maddr_runlock(ifp);
2586
2587	v |= CAS_MAC_RX_CONF_HFILTER;
2588
2589	/* Now load the hash table into the chip (if we are using it). */
2590	for (i = 0; i < 16; i++)
2591		CAS_WRITE_4(sc,
2592		    CAS_MAC_HASH0 + i * (CAS_MAC_HASH1 - CAS_MAC_HASH0),
2593		    hash[i]);
2594
2595 chipit:
2596	sc->sc_mac_rxcfg = v;
2597	CAS_WRITE_4(sc, CAS_MAC_RX_CONF, v | CAS_MAC_RX_CONF_EN);
2598}
2599
2600static int	cas_pci_attach(device_t dev);
2601static int	cas_pci_detach(device_t dev);
2602static int	cas_pci_probe(device_t dev);
2603static int	cas_pci_resume(device_t dev);
2604static int	cas_pci_suspend(device_t dev);
2605
2606static device_method_t cas_pci_methods[] = {
2607	/* Device interface */
2608	DEVMETHOD(device_probe,		cas_pci_probe),
2609	DEVMETHOD(device_attach,	cas_pci_attach),
2610	DEVMETHOD(device_detach,	cas_pci_detach),
2611	DEVMETHOD(device_suspend,	cas_pci_suspend),
2612	DEVMETHOD(device_resume,	cas_pci_resume),
2613	/* Use the suspend handler here, it is all that is required. */
2614	DEVMETHOD(device_shutdown,	cas_pci_suspend),
2615
2616	/* MII interface */
2617	DEVMETHOD(miibus_readreg,	cas_mii_readreg),
2618	DEVMETHOD(miibus_writereg,	cas_mii_writereg),
2619	DEVMETHOD(miibus_statchg,	cas_mii_statchg),
2620
2621	DEVMETHOD_END
2622};
2623
2624static driver_t cas_pci_driver = {
2625	"cas",
2626	cas_pci_methods,
2627	sizeof(struct cas_softc)
2628};
2629
2630DRIVER_MODULE(cas, pci, cas_pci_driver, cas_devclass, 0, 0);
2631DRIVER_MODULE(miibus, cas, miibus_driver, miibus_devclass, 0, 0);
2632MODULE_DEPEND(cas, pci, 1, 1, 1);
2633
2634static const struct cas_pci_dev {
2635	uint32_t	cpd_devid;
2636	uint8_t		cpd_revid;
2637	int		cpd_variant;
2638	const char	*cpd_desc;
2639} cas_pci_devlist[] = {
2640	{ 0x0035100b, 0x0, CAS_SATURN, "NS DP83065 Saturn Gigabit Ethernet" },
2641	{ 0xabba108e, 0x10, CAS_CASPLUS, "Sun Cassini+ Gigabit Ethernet" },
2642	{ 0xabba108e, 0x0, CAS_CAS, "Sun Cassini Gigabit Ethernet" },
2643	{ 0, 0, 0, NULL }
2644};
2645
2646static int
2647cas_pci_probe(device_t dev)
2648{
2649	int i;
2650
2651	for (i = 0; cas_pci_devlist[i].cpd_desc != NULL; i++) {
2652		if (pci_get_devid(dev) == cas_pci_devlist[i].cpd_devid &&
2653		    pci_get_revid(dev) >= cas_pci_devlist[i].cpd_revid) {
2654			device_set_desc(dev, cas_pci_devlist[i].cpd_desc);
2655			return (BUS_PROBE_DEFAULT);
2656		}
2657	}
2658
2659	return (ENXIO);
2660}
2661
2662static struct resource_spec cas_pci_res_spec[] = {
2663	{ SYS_RES_IRQ, 0, RF_SHAREABLE | RF_ACTIVE },	/* CAS_RES_INTR */
2664	{ SYS_RES_MEMORY, PCIR_BAR(0), RF_ACTIVE },	/* CAS_RES_MEM */
2665	{ -1, 0 }
2666};
2667
2668#define	CAS_LOCAL_MAC_ADDRESS	"local-mac-address"
2669#define	CAS_PHY_INTERFACE	"phy-interface"
2670#define	CAS_PHY_TYPE		"phy-type"
2671#define	CAS_PHY_TYPE_PCS	"pcs"
2672
2673static int
2674cas_pci_attach(device_t dev)
2675{
2676	char buf[sizeof(CAS_LOCAL_MAC_ADDRESS)];
2677	struct cas_softc *sc;
2678	int i;
2679#if !(defined(__powerpc__) || defined(__sparc64__))
2680	u_char enaddr[4][ETHER_ADDR_LEN];
2681	u_int j, k, lma, pcs[4], phy;
2682#endif
2683
2684	sc = device_get_softc(dev);
2685	sc->sc_variant = CAS_UNKNOWN;
2686	for (i = 0; cas_pci_devlist[i].cpd_desc != NULL; i++) {
2687		if (pci_get_devid(dev) == cas_pci_devlist[i].cpd_devid &&
2688		    pci_get_revid(dev) >= cas_pci_devlist[i].cpd_revid) {
2689			sc->sc_variant = cas_pci_devlist[i].cpd_variant;
2690			break;
2691		}
2692	}
2693	if (sc->sc_variant == CAS_UNKNOWN) {
2694		device_printf(dev, "unknown adaptor\n");
2695		return (ENXIO);
2696	}
2697
2698	/* PCI configuration */
2699	pci_write_config(dev, PCIR_COMMAND,
2700	    pci_read_config(dev, PCIR_COMMAND, 2) | PCIM_CMD_BUSMASTEREN |
2701	    PCIM_CMD_MWRICEN | PCIM_CMD_PERRESPEN | PCIM_CMD_SERRESPEN, 2);
2702
2703	sc->sc_dev = dev;
2704	if (sc->sc_variant == CAS_CAS && pci_get_devid(dev) < 0x02)
2705		/* Hardware checksumming may hang TX. */
2706		sc->sc_flags |= CAS_NO_CSUM;
2707	if (sc->sc_variant == CAS_CASPLUS || sc->sc_variant == CAS_SATURN)
2708		sc->sc_flags |= CAS_REG_PLUS;
2709	if (sc->sc_variant == CAS_CAS ||
2710	    (sc->sc_variant == CAS_CASPLUS && pci_get_revid(dev) < 0x11))
2711		sc->sc_flags |= CAS_TABORT;
2712	if (bootverbose)
2713		device_printf(dev, "flags=0x%x\n", sc->sc_flags);
2714
2715	if (bus_alloc_resources(dev, cas_pci_res_spec, sc->sc_res)) {
2716		device_printf(dev, "failed to allocate resources\n");
2717		bus_release_resources(dev, cas_pci_res_spec, sc->sc_res);
2718		return (ENXIO);
2719	}
2720
2721	CAS_LOCK_INIT(sc, device_get_nameunit(dev));
2722
2723#if defined(__powerpc__) || defined(__sparc64__)
2724	OF_getetheraddr(dev, sc->sc_enaddr);
2725	if (OF_getprop(ofw_bus_get_node(dev), CAS_PHY_INTERFACE, buf,
2726	    sizeof(buf)) > 0 || OF_getprop(ofw_bus_get_node(dev),
2727	    CAS_PHY_TYPE, buf, sizeof(buf)) > 0) {
2728		buf[sizeof(buf) - 1] = '\0';
2729		if (strcmp(buf, CAS_PHY_TYPE_PCS) == 0)
2730			sc->sc_flags |= CAS_SERDES;
2731	}
2732#else
2733	/*
2734	 * Dig out VPD (vital product data) and read the MAC address as well
2735	 * as the PHY type.  The VPD resides in the PCI Expansion ROM (PCI
2736	 * FCode) and can't be accessed via the PCI capability pointer.
2737	 * SUNW,pci-ce and SUNW,pci-qge use the Enhanced VPD format described
2738	 * in the free US Patent 7149820.
2739	 */
2740
2741#define	PCI_ROMHDR_SIZE			0x1c
2742#define	PCI_ROMHDR_SIG			0x00
2743#define	PCI_ROMHDR_SIG_MAGIC		0xaa55		/* little endian */
2744#define	PCI_ROMHDR_PTR_DATA		0x18
2745#define	PCI_ROM_SIZE			0x18
2746#define	PCI_ROM_SIG			0x00
2747#define	PCI_ROM_SIG_MAGIC		0x52494350	/* "PCIR", endian */
2748							/* reversed */
2749#define	PCI_ROM_VENDOR			0x04
2750#define	PCI_ROM_DEVICE			0x06
2751#define	PCI_ROM_PTR_VPD			0x08
2752#define	PCI_VPDRES_BYTE0		0x00
2753#define	PCI_VPDRES_ISLARGE(x)		((x) & 0x80)
2754#define	PCI_VPDRES_LARGE_NAME(x)	((x) & 0x7f)
2755#define	PCI_VPDRES_LARGE_LEN_LSB	0x01
2756#define	PCI_VPDRES_LARGE_LEN_MSB	0x02
2757#define	PCI_VPDRES_LARGE_SIZE		0x03
2758#define	PCI_VPDRES_TYPE_ID_STRING	0x02		/* large */
2759#define	PCI_VPDRES_TYPE_VPD		0x10		/* large */
2760#define	PCI_VPD_KEY0			0x00
2761#define	PCI_VPD_KEY1			0x01
2762#define	PCI_VPD_LEN			0x02
2763#define	PCI_VPD_SIZE			0x03
2764
2765#define	CAS_ROM_READ_1(sc, offs)					\
2766	CAS_READ_1((sc), CAS_PCI_ROM_OFFSET + (offs))
2767#define	CAS_ROM_READ_2(sc, offs)					\
2768	CAS_READ_2((sc), CAS_PCI_ROM_OFFSET + (offs))
2769#define	CAS_ROM_READ_4(sc, offs)					\
2770	CAS_READ_4((sc), CAS_PCI_ROM_OFFSET + (offs))
2771
2772	lma = phy = 0;
2773	memset(enaddr, 0, sizeof(enaddr));
2774	memset(pcs, 0, sizeof(pcs));
2775
2776	/* Enable PCI Expansion ROM access. */
2777	CAS_WRITE_4(sc, CAS_BIM_LDEV_OEN,
2778	    CAS_BIM_LDEV_OEN_PAD | CAS_BIM_LDEV_OEN_PROM);
2779
2780	/* Read PCI Expansion ROM header. */
2781	if (CAS_ROM_READ_2(sc, PCI_ROMHDR_SIG) != PCI_ROMHDR_SIG_MAGIC ||
2782	    (i = CAS_ROM_READ_2(sc, PCI_ROMHDR_PTR_DATA)) <
2783	    PCI_ROMHDR_SIZE) {
2784		device_printf(dev, "unexpected PCI Expansion ROM header\n");
2785		goto fail_prom;
2786	}
2787
2788	/* Read PCI Expansion ROM data. */
2789	if (CAS_ROM_READ_4(sc, i + PCI_ROM_SIG) != PCI_ROM_SIG_MAGIC ||
2790	    CAS_ROM_READ_2(sc, i + PCI_ROM_VENDOR) != pci_get_vendor(dev) ||
2791	    CAS_ROM_READ_2(sc, i + PCI_ROM_DEVICE) != pci_get_device(dev) ||
2792	    (j = CAS_ROM_READ_2(sc, i + PCI_ROM_PTR_VPD)) <
2793	    i + PCI_ROM_SIZE) {
2794		device_printf(dev, "unexpected PCI Expansion ROM data\n");
2795		goto fail_prom;
2796	}
2797
2798	/* Read PCI VPD. */
2799 next:
2800	if (PCI_VPDRES_ISLARGE(CAS_ROM_READ_1(sc,
2801	    j + PCI_VPDRES_BYTE0)) == 0) {
2802		device_printf(dev, "no large PCI VPD\n");
2803		goto fail_prom;
2804	}
2805
2806	i = (CAS_ROM_READ_1(sc, j + PCI_VPDRES_LARGE_LEN_MSB) << 8) |
2807	    CAS_ROM_READ_1(sc, j + PCI_VPDRES_LARGE_LEN_LSB);
2808	switch (PCI_VPDRES_LARGE_NAME(CAS_ROM_READ_1(sc,
2809	    j + PCI_VPDRES_BYTE0))) {
2810	case PCI_VPDRES_TYPE_ID_STRING:
2811		/* Skip identifier string. */
2812		j += PCI_VPDRES_LARGE_SIZE + i;
2813		goto next;
2814	case PCI_VPDRES_TYPE_VPD:
2815		for (j += PCI_VPDRES_LARGE_SIZE; i > 0;
2816		    i -= PCI_VPD_SIZE + CAS_ROM_READ_1(sc, j + PCI_VPD_LEN),
2817		    j += PCI_VPD_SIZE + CAS_ROM_READ_1(sc, j + PCI_VPD_LEN)) {
2818			if (CAS_ROM_READ_1(sc, j + PCI_VPD_KEY0) != 'Z')
2819				/* no Enhanced VPD */
2820				continue;
2821			if (CAS_ROM_READ_1(sc, j + PCI_VPD_SIZE) != 'I')
2822				/* no instance property */
2823				continue;
2824			if (CAS_ROM_READ_1(sc, j + PCI_VPD_SIZE + 3) == 'B') {
2825				/* byte array */
2826				if (CAS_ROM_READ_1(sc,
2827				    j + PCI_VPD_SIZE + 4) != ETHER_ADDR_LEN)
2828					continue;
2829				bus_read_region_1(sc->sc_res[CAS_RES_MEM],
2830				    CAS_PCI_ROM_OFFSET + j + PCI_VPD_SIZE + 5,
2831				    buf, sizeof(buf));
2832				buf[sizeof(buf) - 1] = '\0';
2833				if (strcmp(buf, CAS_LOCAL_MAC_ADDRESS) != 0)
2834					continue;
2835				bus_read_region_1(sc->sc_res[CAS_RES_MEM],
2836				    CAS_PCI_ROM_OFFSET + j + PCI_VPD_SIZE +
2837				    5 + sizeof(CAS_LOCAL_MAC_ADDRESS),
2838				    enaddr[lma], sizeof(enaddr[lma]));
2839				lma++;
2840				if (lma == 4 && phy == 4)
2841					break;
2842			} else if (CAS_ROM_READ_1(sc, j + PCI_VPD_SIZE + 3) ==
2843			   'S') {
2844				/* string */
2845				if (CAS_ROM_READ_1(sc,
2846				    j + PCI_VPD_SIZE + 4) !=
2847				    sizeof(CAS_PHY_TYPE_PCS))
2848					continue;
2849				bus_read_region_1(sc->sc_res[CAS_RES_MEM],
2850				    CAS_PCI_ROM_OFFSET + j + PCI_VPD_SIZE + 5,
2851				    buf, sizeof(buf));
2852				buf[sizeof(buf) - 1] = '\0';
2853				if (strcmp(buf, CAS_PHY_INTERFACE) == 0)
2854					k = sizeof(CAS_PHY_INTERFACE);
2855				else if (strcmp(buf, CAS_PHY_TYPE) == 0)
2856					k = sizeof(CAS_PHY_TYPE);
2857				else
2858					continue;
2859				bus_read_region_1(sc->sc_res[CAS_RES_MEM],
2860				    CAS_PCI_ROM_OFFSET + j + PCI_VPD_SIZE +
2861				    5 + k, buf, sizeof(buf));
2862				buf[sizeof(buf) - 1] = '\0';
2863				if (strcmp(buf, CAS_PHY_TYPE_PCS) == 0)
2864					pcs[phy] = 1;
2865				phy++;
2866				if (lma == 4 && phy == 4)
2867					break;
2868			}
2869		}
2870		break;
2871	default:
2872		device_printf(dev, "unexpected PCI VPD\n");
2873		goto fail_prom;
2874	}
2875
2876 fail_prom:
2877	CAS_WRITE_4(sc, CAS_BIM_LDEV_OEN, 0);
2878
2879	if (lma == 0) {
2880		device_printf(dev, "could not determine Ethernet address\n");
2881		goto fail;
2882	}
2883	i = 0;
2884	if (lma > 1 && pci_get_slot(dev) < nitems(enaddr))
2885		i = pci_get_slot(dev);
2886	memcpy(sc->sc_enaddr, enaddr[i], ETHER_ADDR_LEN);
2887
2888	if (phy == 0) {
2889		device_printf(dev, "could not determine PHY type\n");
2890		goto fail;
2891	}
2892	i = 0;
2893	if (phy > 1 && pci_get_slot(dev) < nitems(pcs))
2894		i = pci_get_slot(dev);
2895	if (pcs[i] != 0)
2896		sc->sc_flags |= CAS_SERDES;
2897#endif
2898
2899	if (cas_attach(sc) != 0) {
2900		device_printf(dev, "could not be attached\n");
2901		goto fail;
2902	}
2903
2904	if (bus_setup_intr(dev, sc->sc_res[CAS_RES_INTR], INTR_TYPE_NET |
2905	    INTR_MPSAFE, cas_intr, NULL, sc, &sc->sc_ih) != 0) {
2906		device_printf(dev, "failed to set up interrupt\n");
2907		cas_detach(sc);
2908		goto fail;
2909	}
2910	return (0);
2911
2912 fail:
2913	CAS_LOCK_DESTROY(sc);
2914	bus_release_resources(dev, cas_pci_res_spec, sc->sc_res);
2915	return (ENXIO);
2916}
2917
2918static int
2919cas_pci_detach(device_t dev)
2920{
2921	struct cas_softc *sc;
2922
2923	sc = device_get_softc(dev);
2924	bus_teardown_intr(dev, sc->sc_res[CAS_RES_INTR], sc->sc_ih);
2925	cas_detach(sc);
2926	CAS_LOCK_DESTROY(sc);
2927	bus_release_resources(dev, cas_pci_res_spec, sc->sc_res);
2928	return (0);
2929}
2930
2931static int
2932cas_pci_suspend(device_t dev)
2933{
2934
2935	cas_suspend(device_get_softc(dev));
2936	return (0);
2937}
2938
2939static int
2940cas_pci_resume(device_t dev)
2941{
2942
2943	cas_resume(device_get_softc(dev));
2944	return (0);
2945}
2946