1 /*
2  * Copyright 2012 Red Hat Inc.
3  *
4  * Permission is hereby granted, free of charge, to any person obtaining a
5  * copy of this software and associated documentation files (the "Software"),
6  * to deal in the Software without restriction, including without limitation
7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8  * and/or sell copies of the Software, and to permit persons to whom the
9  * Software is furnished to do so, subject to the following conditions:
10  *
11  * The above copyright notice and this permission notice shall be included in
12  * all copies or substantial portions of the Software.
13  *
14  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
17  * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20  * OTHER DEALINGS IN THE SOFTWARE.
21  *
22  * Authors: Ben Skeggs
23  */
24 #include "nv50.h"
25 #include "pll.h"
26 #include "seq.h"
27 
28 #include <subdev/bios.h>
29 #include <subdev/bios/pll.h>
30 
31 static u32
32 read_div(struct nv50_clk *clk)
33 {
34 	struct nvkm_device *device = clk->base.subdev.device;
35 	switch (device->chipset) {
36 	case 0x50: /* it exists, but only has bit 31, not the dividers.. */
37 	case 0x84:
38 	case 0x86:
39 	case 0x98:
40 	case 0xa0:
41 		return nvkm_rd32(device, 0x004700);
42 	case 0x92:
43 	case 0x94:
44 	case 0x96:
45 		return nvkm_rd32(device, 0x004800);
46 	default:
47 		return 0x00000000;
48 	}
49 }
50 
51 static u32
52 read_pll_src(struct nv50_clk *clk, u32 base)
53 {
54 	struct nvkm_device *device = clk->base.subdev.device;
55 	u32 coef, ref = clk->base.read(&clk->base, nv_clk_src_crystal);
56 	u32 rsel = nvkm_rd32(device, 0x00e18c);
57 	int P, N, M, id;
58 
59 	switch (device->chipset) {
60 	case 0x50:
61 	case 0xa0:
62 		switch (base) {
63 		case 0x4020:
64 		case 0x4028: id = !!(rsel & 0x00000004); break;
65 		case 0x4008: id = !!(rsel & 0x00000008); break;
66 		case 0x4030: id = 0; break;
67 		default:
68 			nv_error(clk, "ref: bad pll 0x%06x\n", base);
69 			return 0;
70 		}
71 
72 		coef = nvkm_rd32(device, 0x00e81c + (id * 0x0c));
73 		ref *=  (coef & 0x01000000) ? 2 : 4;
74 		P    =  (coef & 0x00070000) >> 16;
75 		N    = ((coef & 0x0000ff00) >> 8) + 1;
76 		M    = ((coef & 0x000000ff) >> 0) + 1;
77 		break;
78 	case 0x84:
79 	case 0x86:
80 	case 0x92:
81 		coef = nvkm_rd32(device, 0x00e81c);
82 		P    = (coef & 0x00070000) >> 16;
83 		N    = (coef & 0x0000ff00) >> 8;
84 		M    = (coef & 0x000000ff) >> 0;
85 		break;
86 	case 0x94:
87 	case 0x96:
88 	case 0x98:
89 		rsel = nvkm_rd32(device, 0x00c050);
90 		switch (base) {
91 		case 0x4020: rsel = (rsel & 0x00000003) >> 0; break;
92 		case 0x4008: rsel = (rsel & 0x0000000c) >> 2; break;
93 		case 0x4028: rsel = (rsel & 0x00001800) >> 11; break;
94 		case 0x4030: rsel = 3; break;
95 		default:
96 			nv_error(clk, "ref: bad pll 0x%06x\n", base);
97 			return 0;
98 		}
99 
100 		switch (rsel) {
101 		case 0: id = 1; break;
102 		case 1: return clk->base.read(&clk->base, nv_clk_src_crystal);
103 		case 2: return clk->base.read(&clk->base, nv_clk_src_href);
104 		case 3: id = 0; break;
105 		}
106 
107 		coef =  nvkm_rd32(device, 0x00e81c + (id * 0x28));
108 		P    = (nvkm_rd32(device, 0x00e824 + (id * 0x28)) >> 16) & 7;
109 		P   += (coef & 0x00070000) >> 16;
110 		N    = (coef & 0x0000ff00) >> 8;
111 		M    = (coef & 0x000000ff) >> 0;
112 		break;
113 	default:
114 		BUG_ON(1);
115 	}
116 
117 	if (M)
118 		return (ref * N / M) >> P;
119 
120 	return 0;
121 }
122 
123 static u32
124 read_pll_ref(struct nv50_clk *clk, u32 base)
125 {
126 	struct nvkm_device *device = clk->base.subdev.device;
127 	u32 src, mast = nvkm_rd32(device, 0x00c040);
128 
129 	switch (base) {
130 	case 0x004028:
131 		src = !!(mast & 0x00200000);
132 		break;
133 	case 0x004020:
134 		src = !!(mast & 0x00400000);
135 		break;
136 	case 0x004008:
137 		src = !!(mast & 0x00010000);
138 		break;
139 	case 0x004030:
140 		src = !!(mast & 0x02000000);
141 		break;
142 	case 0x00e810:
143 		return clk->base.read(&clk->base, nv_clk_src_crystal);
144 	default:
145 		nv_error(clk, "bad pll 0x%06x\n", base);
146 		return 0;
147 	}
148 
149 	if (src)
150 		return clk->base.read(&clk->base, nv_clk_src_href);
151 
152 	return read_pll_src(clk, base);
153 }
154 
155 static u32
156 read_pll(struct nv50_clk *clk, u32 base)
157 {
158 	struct nvkm_device *device = clk->base.subdev.device;
159 	u32 mast = nvkm_rd32(device, 0x00c040);
160 	u32 ctrl = nvkm_rd32(device, base + 0);
161 	u32 coef = nvkm_rd32(device, base + 4);
162 	u32 ref = read_pll_ref(clk, base);
163 	u32 freq = 0;
164 	int N1, N2, M1, M2;
165 
166 	if (base == 0x004028 && (mast & 0x00100000)) {
167 		/* wtf, appears to only disable post-divider on gt200 */
168 		if (device->chipset != 0xa0)
169 			return clk->base.read(&clk->base, nv_clk_src_dom6);
170 	}
171 
172 	N2 = (coef & 0xff000000) >> 24;
173 	M2 = (coef & 0x00ff0000) >> 16;
174 	N1 = (coef & 0x0000ff00) >> 8;
175 	M1 = (coef & 0x000000ff);
176 	if ((ctrl & 0x80000000) && M1) {
177 		freq = ref * N1 / M1;
178 		if ((ctrl & 0x40000100) == 0x40000000) {
179 			if (M2)
180 				freq = freq * N2 / M2;
181 			else
182 				freq = 0;
183 		}
184 	}
185 
186 	return freq;
187 }
188 
189 static int
190 nv50_clk_read(struct nvkm_clk *obj, enum nv_clk_src src)
191 {
192 	struct nv50_clk *clk = container_of(obj, typeof(*clk), base);
193 	struct nvkm_device *device = clk->base.subdev.device;
194 	u32 mast = nvkm_rd32(device, 0x00c040);
195 	u32 P = 0;
196 
197 	switch (src) {
198 	case nv_clk_src_crystal:
199 		return device->crystal;
200 	case nv_clk_src_href:
201 		return 100000; /* PCIE reference clock */
202 	case nv_clk_src_hclk:
203 		return div_u64((u64)clk->base.read(&clk->base, nv_clk_src_href) * 27778, 10000);
204 	case nv_clk_src_hclkm3:
205 		return clk->base.read(&clk->base, nv_clk_src_hclk) * 3;
206 	case nv_clk_src_hclkm3d2:
207 		return clk->base.read(&clk->base, nv_clk_src_hclk) * 3 / 2;
208 	case nv_clk_src_host:
209 		switch (mast & 0x30000000) {
210 		case 0x00000000: return clk->base.read(&clk->base, nv_clk_src_href);
211 		case 0x10000000: break;
212 		case 0x20000000: /* !0x50 */
213 		case 0x30000000: return clk->base.read(&clk->base, nv_clk_src_hclk);
214 		}
215 		break;
216 	case nv_clk_src_core:
217 		if (!(mast & 0x00100000))
218 			P = (nvkm_rd32(device, 0x004028) & 0x00070000) >> 16;
219 		switch (mast & 0x00000003) {
220 		case 0x00000000: return clk->base.read(&clk->base, nv_clk_src_crystal) >> P;
221 		case 0x00000001: return clk->base.read(&clk->base, nv_clk_src_dom6);
222 		case 0x00000002: return read_pll(clk, 0x004020) >> P;
223 		case 0x00000003: return read_pll(clk, 0x004028) >> P;
224 		}
225 		break;
226 	case nv_clk_src_shader:
227 		P = (nvkm_rd32(device, 0x004020) & 0x00070000) >> 16;
228 		switch (mast & 0x00000030) {
229 		case 0x00000000:
230 			if (mast & 0x00000080)
231 				return clk->base.read(&clk->base, nv_clk_src_host) >> P;
232 			return clk->base.read(&clk->base, nv_clk_src_crystal) >> P;
233 		case 0x00000010: break;
234 		case 0x00000020: return read_pll(clk, 0x004028) >> P;
235 		case 0x00000030: return read_pll(clk, 0x004020) >> P;
236 		}
237 		break;
238 	case nv_clk_src_mem:
239 		P = (nvkm_rd32(device, 0x004008) & 0x00070000) >> 16;
240 		if (nvkm_rd32(device, 0x004008) & 0x00000200) {
241 			switch (mast & 0x0000c000) {
242 			case 0x00000000:
243 				return clk->base.read(&clk->base, nv_clk_src_crystal) >> P;
244 			case 0x00008000:
245 			case 0x0000c000:
246 				return clk->base.read(&clk->base, nv_clk_src_href) >> P;
247 			}
248 		} else {
249 			return read_pll(clk, 0x004008) >> P;
250 		}
251 		break;
252 	case nv_clk_src_vdec:
253 		P = (read_div(clk) & 0x00000700) >> 8;
254 		switch (device->chipset) {
255 		case 0x84:
256 		case 0x86:
257 		case 0x92:
258 		case 0x94:
259 		case 0x96:
260 		case 0xa0:
261 			switch (mast & 0x00000c00) {
262 			case 0x00000000:
263 				if (device->chipset == 0xa0) /* wtf?? */
264 					return clk->base.read(&clk->base, nv_clk_src_core) >> P;
265 				return clk->base.read(&clk->base, nv_clk_src_crystal) >> P;
266 			case 0x00000400:
267 				return 0;
268 			case 0x00000800:
269 				if (mast & 0x01000000)
270 					return read_pll(clk, 0x004028) >> P;
271 				return read_pll(clk, 0x004030) >> P;
272 			case 0x00000c00:
273 				return clk->base.read(&clk->base, nv_clk_src_core) >> P;
274 			}
275 			break;
276 		case 0x98:
277 			switch (mast & 0x00000c00) {
278 			case 0x00000000:
279 				return clk->base.read(&clk->base, nv_clk_src_core) >> P;
280 			case 0x00000400:
281 				return 0;
282 			case 0x00000800:
283 				return clk->base.read(&clk->base, nv_clk_src_hclkm3d2) >> P;
284 			case 0x00000c00:
285 				return clk->base.read(&clk->base, nv_clk_src_mem) >> P;
286 			}
287 			break;
288 		}
289 		break;
290 	case nv_clk_src_dom6:
291 		switch (device->chipset) {
292 		case 0x50:
293 		case 0xa0:
294 			return read_pll(clk, 0x00e810) >> 2;
295 		case 0x84:
296 		case 0x86:
297 		case 0x92:
298 		case 0x94:
299 		case 0x96:
300 		case 0x98:
301 			P = (read_div(clk) & 0x00000007) >> 0;
302 			switch (mast & 0x0c000000) {
303 			case 0x00000000: return clk->base.read(&clk->base, nv_clk_src_href);
304 			case 0x04000000: break;
305 			case 0x08000000: return clk->base.read(&clk->base, nv_clk_src_hclk);
306 			case 0x0c000000:
307 				return clk->base.read(&clk->base, nv_clk_src_hclkm3) >> P;
308 			}
309 			break;
310 		default:
311 			break;
312 		}
313 	default:
314 		break;
315 	}
316 
317 	nv_debug(clk, "unknown clock source %d 0x%08x\n", src, mast);
318 	return -EINVAL;
319 }
320 
321 static u32
322 calc_pll(struct nv50_clk *clk, u32 reg, u32 idx, int *N, int *M, int *P)
323 {
324 	struct nvkm_bios *bios = nvkm_bios(clk);
325 	struct nvbios_pll pll;
326 	int ret;
327 
328 	ret = nvbios_pll_parse(bios, reg, &pll);
329 	if (ret)
330 		return 0;
331 
332 	pll.vco2.max_freq = 0;
333 	pll.refclk = read_pll_ref(clk, reg);
334 	if (!pll.refclk)
335 		return 0;
336 
337 	return nv04_pll_calc(nv_subdev(clk), &pll, idx, N, M, NULL, NULL, P);
338 }
339 
340 static inline u32
341 calc_div(u32 src, u32 target, int *div)
342 {
343 	u32 clk0 = src, clk1 = src;
344 	for (*div = 0; *div <= 7; (*div)++) {
345 		if (clk0 <= target) {
346 			clk1 = clk0 << (*div ? 1 : 0);
347 			break;
348 		}
349 		clk0 >>= 1;
350 	}
351 
352 	if (target - clk0 <= clk1 - target)
353 		return clk0;
354 	(*div)--;
355 	return clk1;
356 }
357 
358 static inline u32
359 clk_same(u32 a, u32 b)
360 {
361 	return ((a / 1000) == (b / 1000));
362 }
363 
364 static int
365 nv50_clk_calc(struct nvkm_clk *obj, struct nvkm_cstate *cstate)
366 {
367 	struct nv50_clk *clk = container_of(obj, typeof(*clk), base);
368 	struct nv50_clk_hwsq *hwsq = &clk->hwsq;
369 	const int shader = cstate->domain[nv_clk_src_shader];
370 	const int core = cstate->domain[nv_clk_src_core];
371 	const int vdec = cstate->domain[nv_clk_src_vdec];
372 	const int dom6 = cstate->domain[nv_clk_src_dom6];
373 	u32 mastm = 0, mastv = 0;
374 	u32 divsm = 0, divsv = 0;
375 	int N, M, P1, P2;
376 	int freq, out;
377 
378 	/* prepare a hwsq script from which we'll perform the reclock */
379 	out = clk_init(hwsq, nv_subdev(clk));
380 	if (out)
381 		return out;
382 
383 	clk_wr32(hwsq, fifo, 0x00000001); /* block fifo */
384 	clk_nsec(hwsq, 8000);
385 	clk_setf(hwsq, 0x10, 0x00); /* disable fb */
386 	clk_wait(hwsq, 0x00, 0x01); /* wait for fb disabled */
387 
388 	/* vdec: avoid modifying xpll until we know exactly how the other
389 	 * clock domains work, i suspect at least some of them can also be
390 	 * tied to xpll...
391 	 */
392 	if (vdec) {
393 		/* see how close we can get using nvclk as a source */
394 		freq = calc_div(core, vdec, &P1);
395 
396 		/* see how close we can get using xpll/hclk as a source */
397 		if (nv_device(clk)->chipset != 0x98)
398 			out = read_pll(clk, 0x004030);
399 		else
400 			out = clk->base.read(&clk->base, nv_clk_src_hclkm3d2);
401 		out = calc_div(out, vdec, &P2);
402 
403 		/* select whichever gets us closest */
404 		if (abs(vdec - freq) <= abs(vdec - out)) {
405 			if (nv_device(clk)->chipset != 0x98)
406 				mastv |= 0x00000c00;
407 			divsv |= P1 << 8;
408 		} else {
409 			mastv |= 0x00000800;
410 			divsv |= P2 << 8;
411 		}
412 
413 		mastm |= 0x00000c00;
414 		divsm |= 0x00000700;
415 	}
416 
417 	/* dom6: nfi what this is, but we're limited to various combinations
418 	 * of the host clock frequency
419 	 */
420 	if (dom6) {
421 		if (clk_same(dom6, clk->base.read(&clk->base, nv_clk_src_href))) {
422 			mastv |= 0x00000000;
423 		} else
424 		if (clk_same(dom6, clk->base.read(&clk->base, nv_clk_src_hclk))) {
425 			mastv |= 0x08000000;
426 		} else {
427 			freq = clk->base.read(&clk->base, nv_clk_src_hclk) * 3;
428 			calc_div(freq, dom6, &P1);
429 
430 			mastv |= 0x0c000000;
431 			divsv |= P1;
432 		}
433 
434 		mastm |= 0x0c000000;
435 		divsm |= 0x00000007;
436 	}
437 
438 	/* vdec/dom6: switch to "safe" clocks temporarily, update dividers
439 	 * and then switch to target clocks
440 	 */
441 	clk_mask(hwsq, mast, mastm, 0x00000000);
442 	clk_mask(hwsq, divs, divsm, divsv);
443 	clk_mask(hwsq, mast, mastm, mastv);
444 
445 	/* core/shader: disconnect nvclk/sclk from their PLLs (nvclk to dom6,
446 	 * sclk to hclk) before reprogramming
447 	 */
448 	if (nv_device(clk)->chipset < 0x92)
449 		clk_mask(hwsq, mast, 0x001000b0, 0x00100080);
450 	else
451 		clk_mask(hwsq, mast, 0x000000b3, 0x00000081);
452 
453 	/* core: for the moment at least, always use nvpll */
454 	freq = calc_pll(clk, 0x4028, core, &N, &M, &P1);
455 	if (freq == 0)
456 		return -ERANGE;
457 
458 	clk_mask(hwsq, nvpll[0], 0xc03f0100,
459 				 0x80000000 | (P1 << 19) | (P1 << 16));
460 	clk_mask(hwsq, nvpll[1], 0x0000ffff, (N << 8) | M);
461 
462 	/* shader: tie to nvclk if possible, otherwise use spll.  have to be
463 	 * very careful that the shader clock is at least twice the core, or
464 	 * some chipsets will be very unhappy.  i expect most or all of these
465 	 * cases will be handled by tying to nvclk, but it's possible there's
466 	 * corners
467 	 */
468 	if (P1-- && shader == (core << 1)) {
469 		clk_mask(hwsq, spll[0], 0xc03f0100, (P1 << 19) | (P1 << 16));
470 		clk_mask(hwsq, mast, 0x00100033, 0x00000023);
471 	} else {
472 		freq = calc_pll(clk, 0x4020, shader, &N, &M, &P1);
473 		if (freq == 0)
474 			return -ERANGE;
475 
476 		clk_mask(hwsq, spll[0], 0xc03f0100,
477 					0x80000000 | (P1 << 19) | (P1 << 16));
478 		clk_mask(hwsq, spll[1], 0x0000ffff, (N << 8) | M);
479 		clk_mask(hwsq, mast, 0x00100033, 0x00000033);
480 	}
481 
482 	/* restore normal operation */
483 	clk_setf(hwsq, 0x10, 0x01); /* enable fb */
484 	clk_wait(hwsq, 0x00, 0x00); /* wait for fb enabled */
485 	clk_wr32(hwsq, fifo, 0x00000000); /* un-block fifo */
486 	return 0;
487 }
488 
489 static int
490 nv50_clk_prog(struct nvkm_clk *obj)
491 {
492 	struct nv50_clk *clk = container_of(obj, typeof(*clk), base);
493 	return clk_exec(&clk->hwsq, true);
494 }
495 
496 static void
497 nv50_clk_tidy(struct nvkm_clk *obj)
498 {
499 	struct nv50_clk *clk = container_of(obj, typeof(*clk), base);
500 	clk_exec(&clk->hwsq, false);
501 }
502 
503 int
504 nv50_clk_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
505 	      struct nvkm_oclass *oclass, void *data, u32 size,
506 	      struct nvkm_object **pobject)
507 {
508 	struct nv50_clk_oclass *pclass = (void *)oclass;
509 	struct nv50_clk *clk;
510 	int ret;
511 
512 	ret = nvkm_clk_create(parent, engine, oclass, pclass->domains,
513 			      NULL, 0, nv_device(parent)->chipset == 0xa0,
514 			      &clk);
515 	*pobject = nv_object(clk);
516 	if (ret)
517 		return ret;
518 
519 	clk->hwsq.r_fifo = hwsq_reg(0x002504);
520 	clk->hwsq.r_spll[0] = hwsq_reg(0x004020);
521 	clk->hwsq.r_spll[1] = hwsq_reg(0x004024);
522 	clk->hwsq.r_nvpll[0] = hwsq_reg(0x004028);
523 	clk->hwsq.r_nvpll[1] = hwsq_reg(0x00402c);
524 	switch (nv_device(clk)->chipset) {
525 	case 0x92:
526 	case 0x94:
527 	case 0x96:
528 		clk->hwsq.r_divs = hwsq_reg(0x004800);
529 		break;
530 	default:
531 		clk->hwsq.r_divs = hwsq_reg(0x004700);
532 		break;
533 	}
534 	clk->hwsq.r_mast = hwsq_reg(0x00c040);
535 
536 	clk->base.read = nv50_clk_read;
537 	clk->base.calc = nv50_clk_calc;
538 	clk->base.prog = nv50_clk_prog;
539 	clk->base.tidy = nv50_clk_tidy;
540 	return 0;
541 }
542 
543 static struct nvkm_domain
544 nv50_domains[] = {
545 	{ nv_clk_src_crystal, 0xff },
546 	{ nv_clk_src_href   , 0xff },
547 	{ nv_clk_src_core   , 0xff, 0, "core", 1000 },
548 	{ nv_clk_src_shader , 0xff, 0, "shader", 1000 },
549 	{ nv_clk_src_mem    , 0xff, 0, "memory", 1000 },
550 	{ nv_clk_src_max }
551 };
552 
553 struct nvkm_oclass *
554 nv50_clk_oclass = &(struct nv50_clk_oclass) {
555 	.base.handle = NV_SUBDEV(CLK, 0x50),
556 	.base.ofuncs = &(struct nvkm_ofuncs) {
557 		.ctor = nv50_clk_ctor,
558 		.dtor = _nvkm_clk_dtor,
559 		.init = _nvkm_clk_init,
560 		.fini = _nvkm_clk_fini,
561 	},
562 	.domains = nv50_domains,
563 }.base;
564