xref: /netbsd-src/sys/dev/tc/if_le_ioasic.c (revision 181254a7b1bdde6873432bffef2d2decc4b5c22f)
1 /*	$NetBSD: if_le_ioasic.c,v 1.34 2018/09/03 16:29:33 riastradh 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 1.34 2018/09/03 16:29:33 riastradh 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(device_t, cfdata_t, void *);
75 static void le_ioasic_attach(device_t, device_t, void *);
76 
77 CFATTACH_DECL_NEW(le_ioasic, sizeof(struct le_softc),
78     le_ioasic_match, le_ioasic_attach, NULL, NULL);
79 
80 static void le_ioasic_copytobuf_gap2(struct lance_softc *, void *, int, int);
81 static void le_ioasic_copyfrombuf_gap2(struct lance_softc *, void *, int, int);
82 static void le_ioasic_copytobuf_gap16(struct lance_softc *, void *, int, int);
83 static void le_ioasic_copyfrombuf_gap16(struct lance_softc *, void *,
84 	    int, int);
85 static void le_ioasic_zerobuf_gap16(struct lance_softc *, int, int);
86 
87 static int
88 le_ioasic_match(device_t parent, cfdata_t cf, void *aux)
89 {
90 	struct ioasicdev_attach_args *d = aux;
91 
92 	if (strncmp("PMAD-BA ", d->iada_modname, TC_ROM_LLEN) != 0)
93 		return 0;
94 
95 	return 1;
96 }
97 
98 /* IOASIC LANCE DMA needs 128KB boundary aligned 128KB chunk */
99 #define	LE_IOASIC_MEMSIZE	(128 * 1024)
100 #define	LE_IOASIC_MEMALIGN	(128 * 1024)
101 
102 static void
103 le_ioasic_attach(device_t parent, device_t self, void *aux)
104 {
105 	struct le_ioasic_softc *sc = device_private(self);
106 	struct ioasicdev_attach_args *d = aux;
107 	struct lance_softc *le = &sc->sc_am7990.lsc;
108 	struct ioasic_softc *iosc = device_private(parent);
109 	bus_space_tag_t ioasic_bst;
110 	bus_space_handle_t ioasic_bsh;
111 	bus_dma_tag_t dmat;
112 	bus_dma_segment_t seg;
113 	tc_addr_t tca;
114 	uint32_t ssr;
115 	int rseg;
116 	void *le_iomem;
117 
118 	le->sc_dev = self;
119 	ioasic_bst = iosc->sc_bst;
120 	ioasic_bsh = iosc->sc_bsh;
121 	dmat = sc->sc_dmat = iosc->sc_dmat;
122 	/*
123 	 * Allocate a DMA area for the chip.
124 	 */
125 	if (bus_dmamem_alloc(dmat, LE_IOASIC_MEMSIZE, LE_IOASIC_MEMALIGN,
126 	    0, &seg, 1, &rseg, BUS_DMA_NOWAIT)) {
127 		aprint_error(": can't allocate DMA area for LANCE\n");
128 		return;
129 	}
130 	if (bus_dmamem_map(dmat, &seg, rseg, LE_IOASIC_MEMSIZE,
131 	    &le_iomem, BUS_DMA_NOWAIT|BUS_DMA_COHERENT)) {
132 		aprint_error(": can't map DMA area for LANCE\n");
133 		bus_dmamem_free(dmat, &seg, rseg);
134 		return;
135 	}
136 	/*
137 	 * Create and load the DMA map for the DMA area.
138 	 */
139 	if (bus_dmamap_create(dmat, LE_IOASIC_MEMSIZE, 1,
140 	    LE_IOASIC_MEMSIZE, 0, BUS_DMA_NOWAIT, &sc->sc_dmamap)) {
141 		aprint_error(": can't create DMA map\n");
142 		goto bad;
143 	}
144 	if (bus_dmamap_load(dmat, sc->sc_dmamap,
145 	    le_iomem, LE_IOASIC_MEMSIZE, NULL, BUS_DMA_NOWAIT)) {
146 		aprint_error(": can't load DMA map\n");
147 		goto bad;
148 	}
149 	/*
150 	 * Bind 128KB buffer with IOASIC DMA.
151 	 */
152 	tca = IOASIC_DMA_ADDR(sc->sc_dmamap->dm_segs[0].ds_addr);
153 	bus_space_write_4(ioasic_bst, ioasic_bsh, IOASIC_LANCE_DMAPTR, tca);
154 	ssr = bus_space_read_4(ioasic_bst, ioasic_bsh, IOASIC_CSR);
155 	ssr |= IOASIC_CSR_DMAEN_LANCE;
156 	bus_space_write_4(ioasic_bst, ioasic_bsh, IOASIC_CSR, ssr);
157 
158 	sc->sc_r1 = (struct lereg1 *)
159 		TC_DENSE_TO_SPARSE(TC_PHYS_TO_UNCACHED(d->iada_addr));
160 	le->sc_mem = (void *)TC_PHYS_TO_UNCACHED(le_iomem);
161 	le->sc_copytodesc = le_ioasic_copytobuf_gap2;
162 	le->sc_copyfromdesc = le_ioasic_copyfrombuf_gap2;
163 	le->sc_copytobuf = le_ioasic_copytobuf_gap16;
164 	le->sc_copyfrombuf = le_ioasic_copyfrombuf_gap16;
165 	le->sc_zerobuf = le_ioasic_zerobuf_gap16;
166 
167 	dec_le_common_attach(&sc->sc_am7990,
168 	    (uint8_t *)iosc->sc_base + IOASIC_SLOT_2_START);
169 
170 	ioasic_intr_establish(parent, d->iada_cookie, TC_IPL_NET,
171 	    am7990_intr, sc);
172 	return;
173 
174  bad:
175 	bus_dmamem_unmap(dmat, le_iomem, LE_IOASIC_MEMSIZE);
176 	bus_dmamem_free(dmat, &seg, rseg);
177 }
178 
179 /*
180  * Special memory access functions needed by ioasic-attached LANCE
181  * chips.
182  */
183 
184 /*
185  * gap2: two bytes of data followed by two bytes of pad.
186  *
187  * Buffers must be 4-byte aligned.  The code doesn't worry about
188  * doing an extra byte.
189  */
190 
191 void
192 le_ioasic_copytobuf_gap2(struct lance_softc *sc, void *fromv, int boff, int len)
193 {
194 	volatile void *buf = sc->sc_mem;
195 	uint8_t *from = fromv;
196 	volatile uint16_t *bptr;
197 
198 	if (boff & 0x1) {
199 		/* handle unaligned first byte */
200 		bptr = ((volatile uint16_t *)buf) + (boff - 1);
201 		*bptr = (*from++ << 8) | (*bptr & 0xff);
202 		bptr += 2;
203 		len--;
204 	} else
205 		bptr = ((volatile uint16_t *)buf) + boff;
206 	while (len > 1) {
207 		*bptr = (from[1] << 8) | (from[0] & 0xff);
208 		bptr += 2;
209 		from += 2;
210 		len -= 2;
211 	}
212 	if (len == 1)
213 		*bptr = (uint16_t)*from;
214 }
215 
216 void
217 le_ioasic_copyfrombuf_gap2(struct lance_softc *sc, void *tov, int boff, int len)
218 {
219 	volatile void *buf = sc->sc_mem;
220 	uint8_t *to = tov;
221 	volatile uint16_t *bptr;
222 	uint16_t tmp;
223 
224 	if (boff & 0x1) {
225 		/* handle unaligned first byte */
226 		bptr = ((volatile uint16_t *)buf) + (boff - 1);
227 		*to++ = (*bptr >> 8) & 0xff;
228 		bptr += 2;
229 		len--;
230 	} else
231 		bptr = ((volatile uint16_t *)buf) + boff;
232 	while (len > 1) {
233 		tmp = *bptr;
234 		*to++ = tmp & 0xff;
235 		*to++ = (tmp >> 8) & 0xff;
236 		bptr += 2;
237 		len -= 2;
238 	}
239 	if (len == 1)
240 		*to = *bptr & 0xff;
241 }
242 
243 /*
244  * gap16: 16 bytes of data followed by 16 bytes of pad.
245  *
246  * Buffers must be 32-byte aligned.
247  */
248 
249 void
250 le_ioasic_copytobuf_gap16(struct lance_softc *sc, void *fromv, int boff,
251     int len)
252 {
253 	uint8_t *buf = sc->sc_mem;
254 	uint8_t *from = fromv;
255 	uint8_t *bptr;
256 
257 	bptr = buf + ((boff << 1) & ~0x1f);
258 	boff &= 0xf;
259 
260 	/*
261 	 * Dispose of boff so destination of subsequent copies is
262 	 * 16-byte aligned.
263 	 */
264 	if (boff) {
265 		int xfer;
266 		xfer = uimin(len, 16 - boff);
267 		memcpy(bptr + boff, from, xfer);
268 		from += xfer;
269 		bptr += 32;
270 		len -= xfer;
271 	}
272 
273 	/* Destination of  copies is now 16-byte aligned. */
274 	if (len >= 16)
275 		switch ((u_long)from & (sizeof(uint32_t) -1)) {
276 		case 2:
277 			/*  Ethernet headers make this the dominant case. */
278 		do {
279 			uint32_t *dst = (uint32_t *)bptr;
280 			uint16_t t0;
281 			uint32_t t1,  t2, t3, t4;
282 
283 			/* read from odd-16-bit-aligned, cached src */
284 			t0 = *(uint16_t *)(from +  0);
285 			t1 = *(uint32_t *)(from +  2);
286 			t2 = *(uint32_t *)(from +  6);
287 			t3 = *(uint32_t *)(from + 10);
288 			t4 = *(uint16_t *)(from + 14);
289 
290 			/* DMA buffer is uncached on mips */
291 			dst[0] =         t0 |  (t1 << 16);
292 			dst[1] = (t1 >> 16) |  (t2 << 16);
293 			dst[2] = (t2 >> 16) |  (t3 << 16);
294 			dst[3] = (t3 >> 16) |  (t4 << 16);
295 
296 			from += 16;
297 			bptr += 32;
298 			len -= 16;
299 		} while (len >= 16);
300 		break;
301 
302 		case 0:
303 		do {
304 			uint32_t *src = (uint32_t*)from;
305 			uint32_t *dst = (uint32_t*)bptr;
306 			uint32_t t0, t1, t2, t3;
307 
308 			t0 = src[0]; t1 = src[1]; t2 = src[2]; t3 = src[3];
309 			dst[0] = t0; dst[1] = t1; dst[2] = t2; dst[3] = t3;
310 
311 			from += 16;
312 			bptr += 32;
313 			len -= 16;
314 		} while (len >= 16);
315 		break;
316 
317 		default:
318 		/* Does odd-aligned case ever happen? */
319 		do {
320 			memcpy(bptr, from, 16);
321 			from += 16;
322 			bptr += 32;
323 			len -= 16;
324 		} while (len >= 16);
325 		break;
326 	}
327 	if (len)
328 		memcpy(bptr, from, len);
329 }
330 
331 void
332 le_ioasic_copyfrombuf_gap16(struct lance_softc *sc, void *tov, int boff,
333     int len)
334 {
335 	uint8_t *buf = sc->sc_mem;
336 	uint8_t *to = tov;
337 	uint8_t *bptr;
338 
339 	bptr = buf + ((boff << 1) & ~0x1f);
340 	boff &= 0xf;
341 
342 	/* Dispose of boff. source of copy is subsequently 16-byte aligned. */
343 	if (boff) {
344 		int xfer;
345 		xfer = uimin(len, 16 - boff);
346 		memcpy(to, bptr + boff, xfer);
347 		to += xfer;
348 		bptr += 32;
349 		len -= xfer;
350 	}
351 	if (len >= 16)
352 	switch ((u_long)to & (sizeof(uint32_t) -1)) {
353 	case 2:
354 		/*
355 		 * to is aligned to an odd 16-bit boundary.  Ethernet headers
356 		 * make this the dominant case (98% or more).
357 		 */
358 		do {
359 			uint32_t *src = (uint32_t *)bptr;
360 			uint32_t t0, t1, t2, t3;
361 
362 			/* read from uncached aligned DMA buf */
363 			t0 = src[0]; t1 = src[1]; t2 = src[2]; t3 = src[3];
364 
365 			/* write to odd-16-bit-word aligned dst */
366 			*(uint16_t *)(to +  0) = (uint16_t)t0;
367 			*(uint32_t *)(to +  2) = (t0 >> 16) | (t1 << 16);
368 			*(uint32_t *)(to +  6) = (t1 >> 16) | (t2 << 16);
369 			*(uint32_t *)(to + 10) = (t2 >> 16) | (t3 << 16);
370 			*(uint16_t *)(to + 14) = (t3 >> 16);
371 			bptr += 32;
372 			to += 16;
373 			len -= 16;
374 		} while (len > 16);
375 		break;
376 	case 0:
377 		/* 32-bit aligned aligned copy. Rare. */
378 		do {
379 			uint32_t *src = (uint32_t *)bptr;
380 			uint32_t *dst = (uint32_t *)to;
381 			uint32_t t0, t1, t2, t3;
382 
383 			t0 = src[0]; t1 = src[1]; t2 = src[2]; t3 = src[3];
384 			dst[0] = t0; dst[1] = t1; dst[2] = t2; dst[3] = t3;
385 			to += 16;
386 			bptr += 32;
387 			len -= 16;
388 		} while (len  > 16);
389 		break;
390 
391 	/* XXX Does odd-byte-aligned case ever happen? */
392 	default:
393 		do {
394 			memcpy(to, bptr, 16);
395 			to += 16;
396 			bptr += 32;
397 			len -= 16;
398 		} while (len  > 16);
399 		break;
400 	}
401 	if (len)
402 		memcpy(to, bptr, len);
403 }
404 
405 void
406 le_ioasic_zerobuf_gap16(struct lance_softc *sc, int boff, int len)
407 {
408 	uint8_t *buf = sc->sc_mem;
409 	uint8_t *bptr;
410 	int xfer;
411 
412 	bptr = buf + ((boff << 1) & ~0x1f);
413 	boff &= 0xf;
414 	xfer = uimin(len, 16 - boff);
415 	while (len > 0) {
416 		memset(bptr + boff, 0, xfer);
417 		bptr += 32;
418 		boff = 0;
419 		len -= xfer;
420 		xfer = uimin(len, 16);
421 	}
422 }
423