xref: /netbsd/sys/dev/tc/if_le_ioasic.c (revision bf9ec67e)
1 /*	$NetBSD: if_le_ioasic.c,v 1.18 2001/11/13 06:26:10 lukem Exp $	*/
2 
3 /*
4  * Copyright (c) 1996 Carnegie-Mellon University.
5  * All rights reserved.
6  *
7  * Author: Chris G. Demetriou
8  *
9  * Permission to use, copy, modify and distribute this software and
10  * its documentation is hereby granted, provided that both the copyright
11  * notice and this permission notice appear in all copies of the
12  * software, derivative works or modified versions, and any portions
13  * thereof, and that both notices appear in supporting documentation.
14  *
15  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
16  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
17  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
18  *
19  * Carnegie Mellon requests users of this software to return to
20  *
21  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
22  *  School of Computer Science
23  *  Carnegie Mellon University
24  *  Pittsburgh PA 15213-3890
25  *
26  * any improvements or extensions that they make and grant Carnegie the
27  * rights to redistribute these changes.
28  */
29 
30 /*
31  * LANCE on DEC IOCTL ASIC.
32  */
33 
34 #include <sys/cdefs.h>
35 __KERNEL_RCSID(0, "$NetBSD: if_le_ioasic.c,v 1.18 2001/11/13 06:26:10 lukem Exp $");
36 
37 #include "opt_inet.h"
38 
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/mbuf.h>
42 #include <sys/syslog.h>
43 #include <sys/socket.h>
44 #include <sys/device.h>
45 
46 #include <net/if.h>
47 #include <net/if_ether.h>
48 #include <net/if_media.h>
49 
50 #ifdef INET
51 #include <netinet/in.h>
52 #include <netinet/if_inarp.h>
53 #endif
54 
55 #include <dev/ic/lancereg.h>
56 #include <dev/ic/lancevar.h>
57 #include <dev/ic/am7990reg.h>
58 #include <dev/ic/am7990var.h>
59 
60 #include <dev/tc/if_levar.h>
61 #include <dev/tc/tcvar.h>
62 #include <dev/tc/ioasicreg.h>
63 #include <dev/tc/ioasicvar.h>
64 
65 struct le_ioasic_softc {
66 	struct	am7990_softc sc_am7990;	/* glue to MI code */
67 	struct	lereg1 *sc_r1;		/* LANCE registers */
68 	/* XXX must match with le_softc of if_levar.h XXX */
69 
70 	bus_dma_tag_t sc_dmat;		/* bus dma tag */
71 	bus_dmamap_t sc_dmamap;		/* bus dmamap */
72 };
73 
74 static int  le_ioasic_match __P((struct device *, struct cfdata *, void *));
75 static void le_ioasic_attach __P((struct device *, struct device *, void *));
76 
77 struct cfattach le_ioasic_ca = {
78 	sizeof(struct le_softc), le_ioasic_match, le_ioasic_attach
79 };
80 
81 static void le_ioasic_copytobuf_gap2 __P((struct lance_softc *, void *,
82 	    int, int));
83 static void le_ioasic_copyfrombuf_gap2 __P((struct lance_softc *, void *,
84 	    int, int));
85 static void le_ioasic_copytobuf_gap16 __P((struct lance_softc *, void *,
86 	    int, int));
87 static void le_ioasic_copyfrombuf_gap16 __P((struct lance_softc *, void *,
88 	    int, int));
89 static void le_ioasic_zerobuf_gap16 __P((struct lance_softc *, int, int));
90 
91 static int
92 le_ioasic_match(parent, match, aux)
93 	struct device *parent;
94 	struct cfdata *match;
95 	void *aux;
96 {
97 	struct ioasicdev_attach_args *d = aux;
98 
99 	if (strncmp("PMAD-BA ", d->iada_modname, TC_ROM_LLEN) != 0)
100 		return 0;
101 
102 	return 1;
103 }
104 
105 /* IOASIC LANCE DMA needs 128KB boundary aligned 128KB chunk */
106 #define	LE_IOASIC_MEMSIZE	(128*1024)
107 #define	LE_IOASIC_MEMALIGN	(128*1024)
108 
109 static void
110 le_ioasic_attach(parent, self, aux)
111 	struct device *parent, *self;
112 	void *aux;
113 {
114 	struct le_ioasic_softc *sc = (void *)self;
115 	struct ioasicdev_attach_args *d = aux;
116 	struct lance_softc *le = &sc->sc_am7990.lsc;
117 	bus_space_tag_t ioasic_bst;
118 	bus_space_handle_t ioasic_bsh;
119 	bus_dma_tag_t dmat;
120 	bus_dma_segment_t seg;
121 	tc_addr_t tca;
122 	u_int32_t ssr;
123 	int rseg;
124 	caddr_t le_iomem;
125 
126 	ioasic_bst = ((struct ioasic_softc *)parent)->sc_bst;
127 	ioasic_bsh = ((struct ioasic_softc *)parent)->sc_bsh;
128 	dmat = sc->sc_dmat = ((struct ioasic_softc *)parent)->sc_dmat;
129 	/*
130 	 * Allocate a DMA area for the chip.
131 	 */
132 	if (bus_dmamem_alloc(dmat, LE_IOASIC_MEMSIZE, LE_IOASIC_MEMALIGN,
133 	    0, &seg, 1, &rseg, BUS_DMA_NOWAIT)) {
134 		printf("can't allocate DMA area for LANCE\n");
135 		return;
136 	}
137 	if (bus_dmamem_map(dmat, &seg, rseg, LE_IOASIC_MEMSIZE,
138 	    &le_iomem, BUS_DMA_NOWAIT|BUS_DMA_COHERENT)) {
139 		printf("can't map DMA area for LANCE\n");
140 		bus_dmamem_free(dmat, &seg, rseg);
141 		return;
142 	}
143 	/*
144 	 * Create and load the DMA map for the DMA area.
145 	 */
146 	if (bus_dmamap_create(dmat, LE_IOASIC_MEMSIZE, 1,
147 	    LE_IOASIC_MEMSIZE, 0, BUS_DMA_NOWAIT, &sc->sc_dmamap)) {
148 		printf("can't create DMA map\n");
149 		goto bad;
150 	}
151 	if (bus_dmamap_load(dmat, sc->sc_dmamap,
152 	    le_iomem, LE_IOASIC_MEMSIZE, NULL, BUS_DMA_NOWAIT)) {
153 		printf("can't load DMA map\n");
154 		goto bad;
155 	}
156 	/*
157 	 * Bind 128KB buffer with IOASIC DMA.
158 	 */
159 	tca = IOASIC_DMA_ADDR(sc->sc_dmamap->dm_segs[0].ds_addr);
160 	bus_space_write_4(ioasic_bst, ioasic_bsh, IOASIC_LANCE_DMAPTR, tca);
161 	ssr = bus_space_read_4(ioasic_bst, ioasic_bsh, IOASIC_CSR);
162 	ssr |= IOASIC_CSR_DMAEN_LANCE;
163 	bus_space_write_4(ioasic_bst, ioasic_bsh, IOASIC_CSR, ssr);
164 
165 	sc->sc_r1 = (struct lereg1 *)
166 		TC_DENSE_TO_SPARSE(TC_PHYS_TO_UNCACHED(d->iada_addr));
167 	le->sc_mem = (void *)TC_PHYS_TO_UNCACHED(le_iomem);
168 	le->sc_copytodesc = le_ioasic_copytobuf_gap2;
169 	le->sc_copyfromdesc = le_ioasic_copyfrombuf_gap2;
170 	le->sc_copytobuf = le_ioasic_copytobuf_gap16;
171 	le->sc_copyfrombuf = le_ioasic_copyfrombuf_gap16;
172 	le->sc_zerobuf = le_ioasic_zerobuf_gap16;
173 
174 	dec_le_common_attach(&sc->sc_am7990,
175 	    (u_char *)((struct ioasic_softc *)parent)->sc_base
176 	        + IOASIC_SLOT_2_START);
177 
178 	ioasic_intr_establish(parent, d->iada_cookie, TC_IPL_NET,
179 	    am7990_intr, sc);
180 	return;
181 
182  bad:
183 	bus_dmamem_unmap(dmat, le_iomem, LE_IOASIC_MEMSIZE);
184 	bus_dmamem_free(dmat, &seg, rseg);
185 }
186 
187 /*
188  * Special memory access functions needed by ioasic-attached LANCE
189  * chips.
190  */
191 
192 /*
193  * gap2: two bytes of data followed by two bytes of pad.
194  *
195  * Buffers must be 4-byte aligned.  The code doesn't worry about
196  * doing an extra byte.
197  */
198 
199 void
200 le_ioasic_copytobuf_gap2(sc, fromv, boff, len)
201 	struct lance_softc *sc;
202 	void *fromv;
203 	int boff;
204 	int len;
205 {
206 	volatile caddr_t buf = sc->sc_mem;
207 	caddr_t from = fromv;
208 	volatile u_int16_t *bptr;
209 
210 	if (boff & 0x1) {
211 		/* handle unaligned first byte */
212 		bptr = ((volatile u_int16_t *)buf) + (boff - 1);
213 		*bptr = (*from++ << 8) | (*bptr & 0xff);
214 		bptr += 2;
215 		len--;
216 	} else
217 		bptr = ((volatile u_int16_t *)buf) + boff;
218 	while (len > 1) {
219 		*bptr = (from[1] << 8) | (from[0] & 0xff);
220 		bptr += 2;
221 		from += 2;
222 		len -= 2;
223 	}
224 	if (len == 1)
225 		*bptr = (u_int16_t)*from;
226 }
227 
228 void
229 le_ioasic_copyfrombuf_gap2(sc, tov, boff, len)
230 	struct lance_softc *sc;
231 	void *tov;
232 	int boff, len;
233 {
234 	volatile caddr_t buf = sc->sc_mem;
235 	caddr_t to = tov;
236 	volatile u_int16_t *bptr;
237 	u_int16_t tmp;
238 
239 	if (boff & 0x1) {
240 		/* handle unaligned first byte */
241 		bptr = ((volatile u_int16_t *)buf) + (boff - 1);
242 		*to++ = (*bptr >> 8) & 0xff;
243 		bptr += 2;
244 		len--;
245 	} else
246 		bptr = ((volatile u_int16_t *)buf) + boff;
247 	while (len > 1) {
248 		tmp = *bptr;
249 		*to++ = tmp & 0xff;
250 		*to++ = (tmp >> 8) & 0xff;
251 		bptr += 2;
252 		len -= 2;
253 	}
254 	if (len == 1)
255 		*to = *bptr & 0xff;
256 }
257 
258 /*
259  * gap16: 16 bytes of data followed by 16 bytes of pad.
260  *
261  * Buffers must be 32-byte aligned.
262  */
263 
264 void
265 le_ioasic_copytobuf_gap16(sc, fromv, boff, len)
266 	struct lance_softc *sc;
267 	void *fromv;
268 	int boff;
269 	int len;
270 {
271 	volatile caddr_t buf = sc->sc_mem;
272 	caddr_t from = fromv;
273 	caddr_t bptr;
274 
275 	bptr = buf + ((boff << 1) & ~0x1f);
276 	boff &= 0xf;
277 
278 	/*
279 	 * Dispose of boff so destination of subsequent copies is
280 	 * 16-byte aligned.
281 	 */
282 	if (boff) {
283 		int xfer;
284 		xfer = min(len, 16 - boff);
285 		bcopy(from, bptr + boff, xfer);
286 		from += xfer;
287 		bptr += 32;
288 		len -= xfer;
289 	}
290 
291 	/* Destination of  copies is now 16-byte aligned. */
292 	if (len >= 16)
293 		switch ((u_long)from & (sizeof(u_int32_t) -1)) {
294 		case 2:
295 			/*  Ethernet headers make this the dominant case. */
296 		do {
297 			u_int32_t *dst = (u_int32_t*)bptr;
298 			u_int16_t t0;
299 			u_int32_t t1,  t2, t3, t4;
300 
301 			/* read from odd-16-bit-aligned, cached src */
302 			t0 = *(u_int16_t*)from;
303 			t1 = *(u_int32_t*)(from+2);
304 			t2 = *(u_int32_t*)(from+6);
305 			t3 = *(u_int32_t*)(from+10);
306 			t4 = *(u_int16_t*)(from+14);
307 
308 			/* DMA buffer is uncached on mips */
309 			dst[0] =         t0 |  (t1 << 16);
310 			dst[1] = (t1 >> 16) |  (t2 << 16);
311 			dst[2] = (t2 >> 16) |  (t3 << 16);
312 			dst[3] = (t3 >> 16) |  (t4 << 16);
313 
314 			from += 16;
315 			bptr += 32;
316 			len -= 16;
317 		} while (len >= 16);
318 		break;
319 
320 		case 0:
321 		do {
322 			u_int32_t *src = (u_int32_t*)from;
323 			u_int32_t *dst = (u_int32_t*)bptr;
324 			u_int32_t t0, t1, t2, t3;
325 
326 			t0 = src[0]; t1 = src[1]; t2 = src[2]; t3 = src[3];
327 			dst[0] = t0; dst[1] = t1; dst[2] = t2; dst[3] = t3;
328 
329 			from += 16;
330 			bptr += 32;
331 			len -= 16;
332 		} while (len >= 16);
333 		break;
334 
335 		default:
336 		/* Does odd-aligned case ever happen? */
337 		do {
338 			bcopy(from, bptr, 16);
339 			from += 16;
340 			bptr += 32;
341 			len -= 16;
342 		} while (len >= 16);
343 		break;
344 	}
345 	if (len)
346 		bcopy(from, bptr, len);
347 }
348 
349 void
350 le_ioasic_copyfrombuf_gap16(sc, tov, boff, len)
351 	struct lance_softc *sc;
352 	void *tov;
353 	int boff, len;
354 {
355 	volatile caddr_t buf = sc->sc_mem;
356 	caddr_t to = tov;
357 	caddr_t bptr;
358 
359 	bptr = buf + ((boff << 1) & ~0x1f);
360 	boff &= 0xf;
361 
362 	/* Dispose of boff. source of copy is subsequently 16-byte aligned. */
363 	if (boff) {
364 		int xfer;
365 		xfer = min(len, 16 - boff);
366 		bcopy(bptr+boff, to, xfer);
367 		to += xfer;
368 		bptr += 32;
369 		len -= xfer;
370 	}
371 	if (len >= 16)
372 	switch ((u_long)to & (sizeof(u_int32_t) -1)) {
373 	case 2:
374 		/*
375 		 * to is aligned to an odd 16-bit boundary.  Ethernet headers
376 		 * make this the dominant case (98% or more).
377 		 */
378 		do {
379 			u_int32_t *src = (u_int32_t*)bptr;
380 			u_int32_t t0, t1, t2, t3;
381 
382 			/* read from uncached aligned DMA buf */
383 			t0 = src[0]; t1 = src[1]; t2 = src[2]; t3 = src[3];
384 
385 			/* write to odd-16-bit-word aligned dst */
386 			*(u_int16_t *) (to+0)  = (u_short)  t0;
387 			*(u_int32_t *) (to+2)  = (t0 >> 16) |  (t1 << 16);
388 			*(u_int32_t *) (to+6)  = (t1 >> 16) |  (t2 << 16);
389 			*(u_int32_t *) (to+10) = (t2 >> 16) |  (t3 << 16);
390 			*(u_int16_t *) (to+14) = (t3 >> 16);
391 			bptr += 32;
392 			to += 16;
393 			len -= 16;
394 		} while (len > 16);
395 		break;
396 	case 0:
397 		/* 32-bit aligned aligned copy. Rare. */
398 		do {
399 			u_int32_t *src = (u_int32_t*)bptr;
400 			u_int32_t *dst = (u_int32_t*)to;
401 			u_int32_t t0, t1, t2, t3;
402 
403 			t0 = src[0]; t1 = src[1]; t2 = src[2]; t3 = src[3];
404 			dst[0] = t0; dst[1] = t1; dst[2] = t2; dst[3] = t3;
405 			to += 16;
406 			bptr += 32;
407 			len -= 16;
408 		} while (len  > 16);
409 		break;
410 
411 	/* XXX Does odd-byte-aligned case ever happen? */
412 	default:
413 		do {
414 			bcopy(bptr, to, 16);
415 			to += 16;
416 			bptr += 32;
417 			len -= 16;
418 		} while (len  > 16);
419 		break;
420 	}
421 	if (len)
422 		bcopy(bptr, to, len);
423 }
424 
425 void
426 le_ioasic_zerobuf_gap16(sc, boff, len)
427 	struct lance_softc *sc;
428 	int boff, len;
429 {
430 	volatile caddr_t buf = sc->sc_mem;
431 	caddr_t bptr;
432 	int xfer;
433 
434 	bptr = buf + ((boff << 1) & ~0x1f);
435 	boff &= 0xf;
436 	xfer = min(len, 16 - boff);
437 	while (len > 0) {
438 		bzero(bptr + boff, xfer);
439 		bptr += 32;
440 		boff = 0;
441 		len -= xfer;
442 		xfer = min(len, 16);
443 	}
444 }
445