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  *          Roy Spliet
24  */
25 #include "gt215.h"
26 #include "pll.h"
27 
28 #include <engine/fifo.h>
29 #include <subdev/bios.h>
30 #include <subdev/bios/pll.h>
31 #include <subdev/timer.h>
32 
33 struct gt215_clk {
34 	struct nvkm_clk base;
35 	struct gt215_clk_info eng[nv_clk_src_max];
36 };
37 
38 static u32 read_clk(struct gt215_clk *, int, bool);
39 static u32 read_pll(struct gt215_clk *, int, u32);
40 
41 static u32
42 read_vco(struct gt215_clk *clk, int idx)
43 {
44 	struct nvkm_device *device = clk->base.subdev.device;
45 	u32 sctl = nvkm_rd32(device, 0x4120 + (idx * 4));
46 
47 	switch (sctl & 0x00000030) {
48 	case 0x00000000:
49 		return device->crystal;
50 	case 0x00000020:
51 		return read_pll(clk, 0x41, 0x00e820);
52 	case 0x00000030:
53 		return read_pll(clk, 0x42, 0x00e8a0);
54 	default:
55 		return 0;
56 	}
57 }
58 
59 static u32
60 read_clk(struct gt215_clk *clk, int idx, bool ignore_en)
61 {
62 	struct nvkm_device *device = clk->base.subdev.device;
63 	u32 sctl, sdiv, sclk;
64 
65 	/* refclk for the 0xe8xx plls is a fixed frequency */
66 	if (idx >= 0x40) {
67 		if (device->chipset == 0xaf) {
68 			/* no joke.. seriously.. sigh.. */
69 			return nvkm_rd32(device, 0x00471c) * 1000;
70 		}
71 
72 		return device->crystal;
73 	}
74 
75 	sctl = nvkm_rd32(device, 0x4120 + (idx * 4));
76 	if (!ignore_en && !(sctl & 0x00000100))
77 		return 0;
78 
79 	/* out_alt */
80 	if (sctl & 0x00000400)
81 		return 108000;
82 
83 	/* vco_out */
84 	switch (sctl & 0x00003000) {
85 	case 0x00000000:
86 		if (!(sctl & 0x00000200))
87 			return device->crystal;
88 		return 0;
89 	case 0x00002000:
90 		if (sctl & 0x00000040)
91 			return 108000;
92 		return 100000;
93 	case 0x00003000:
94 		/* vco_enable */
95 		if (!(sctl & 0x00000001))
96 			return 0;
97 
98 		sclk = read_vco(clk, idx);
99 		sdiv = ((sctl & 0x003f0000) >> 16) + 2;
100 		return (sclk * 2) / sdiv;
101 	default:
102 		return 0;
103 	}
104 }
105 
106 static u32
107 read_pll(struct gt215_clk *clk, int idx, u32 pll)
108 {
109 	struct nvkm_device *device = clk->base.subdev.device;
110 	u32 ctrl = nvkm_rd32(device, pll + 0);
111 	u32 sclk = 0, P = 1, N = 1, M = 1;
112 
113 	if (!(ctrl & 0x00000008)) {
114 		if (ctrl & 0x00000001) {
115 			u32 coef = nvkm_rd32(device, pll + 4);
116 			M = (coef & 0x000000ff) >> 0;
117 			N = (coef & 0x0000ff00) >> 8;
118 			P = (coef & 0x003f0000) >> 16;
119 
120 			/* no post-divider on these..
121 			 * XXX: it looks more like two post-"dividers" that
122 			 * cross each other out in the default RPLL config */
123 			if ((pll & 0x00ff00) == 0x00e800)
124 				P = 1;
125 
126 			sclk = read_clk(clk, 0x00 + idx, false);
127 		}
128 	} else {
129 		sclk = read_clk(clk, 0x10 + idx, false);
130 	}
131 
132 	if (M * P)
133 		return sclk * N / (M * P);
134 
135 	return 0;
136 }
137 
138 static int
139 gt215_clk_read(struct nvkm_clk *obj, enum nv_clk_src src)
140 {
141 	struct gt215_clk *clk = container_of(obj, typeof(*clk), base);
142 	struct nvkm_device *device = clk->base.subdev.device;
143 	u32 hsrc;
144 
145 	switch (src) {
146 	case nv_clk_src_crystal:
147 		return device->crystal;
148 	case nv_clk_src_core:
149 	case nv_clk_src_core_intm:
150 		return read_pll(clk, 0x00, 0x4200);
151 	case nv_clk_src_shader:
152 		return read_pll(clk, 0x01, 0x4220);
153 	case nv_clk_src_mem:
154 		return read_pll(clk, 0x02, 0x4000);
155 	case nv_clk_src_disp:
156 		return read_clk(clk, 0x20, false);
157 	case nv_clk_src_vdec:
158 		return read_clk(clk, 0x21, false);
159 	case nv_clk_src_daemon:
160 		return read_clk(clk, 0x25, false);
161 	case nv_clk_src_host:
162 		hsrc = (nvkm_rd32(device, 0xc040) & 0x30000000) >> 28;
163 		switch (hsrc) {
164 		case 0:
165 			return read_clk(clk, 0x1d, false);
166 		case 2:
167 		case 3:
168 			return 277000;
169 		default:
170 			nv_error(clk, "unknown HOST clock source %d\n", hsrc);
171 			return -EINVAL;
172 		}
173 	default:
174 		nv_error(clk, "invalid clock source %d\n", src);
175 		return -EINVAL;
176 	}
177 
178 	return 0;
179 }
180 
181 int
182 gt215_clk_info(struct nvkm_clk *obj, int idx, u32 khz,
183 	       struct gt215_clk_info *info)
184 {
185 	struct gt215_clk *clk = container_of(obj, typeof(*clk), base);
186 	u32 oclk, sclk, sdiv;
187 	s32 diff;
188 
189 	info->clk = 0;
190 
191 	switch (khz) {
192 	case 27000:
193 		info->clk = 0x00000100;
194 		return khz;
195 	case 100000:
196 		info->clk = 0x00002100;
197 		return khz;
198 	case 108000:
199 		info->clk = 0x00002140;
200 		return khz;
201 	default:
202 		sclk = read_vco(clk, idx);
203 		sdiv = min((sclk * 2) / khz, (u32)65);
204 		oclk = (sclk * 2) / sdiv;
205 		diff = ((khz + 3000) - oclk);
206 
207 		/* When imprecise, play it safe and aim for a clock lower than
208 		 * desired rather than higher */
209 		if (diff < 0) {
210 			sdiv++;
211 			oclk = (sclk * 2) / sdiv;
212 		}
213 
214 		/* divider can go as low as 2, limited here because NVIDIA
215 		 * and the VBIOS on my NVA8 seem to prefer using the PLL
216 		 * for 810MHz - is there a good reason?
217 		 * XXX: PLLs with refclk 810MHz?  */
218 		if (sdiv > 4) {
219 			info->clk = (((sdiv - 2) << 16) | 0x00003100);
220 			return oclk;
221 		}
222 
223 		break;
224 	}
225 
226 	return -ERANGE;
227 }
228 
229 int
230 gt215_pll_info(struct nvkm_clk *clock, int idx, u32 pll, u32 khz,
231 	       struct gt215_clk_info *info)
232 {
233 	struct nvkm_bios *bios = nvkm_bios(clock);
234 	struct gt215_clk *clk = (void *)clock;
235 	struct nvbios_pll limits;
236 	int P, N, M, diff;
237 	int ret;
238 
239 	info->pll = 0;
240 
241 	/* If we can get a within [-2, 3) MHz of a divider, we'll disable the
242 	 * PLL and use the divider instead. */
243 	ret = gt215_clk_info(clock, idx, khz, info);
244 	diff = khz - ret;
245 	if (!pll || (diff >= -2000 && diff < 3000)) {
246 		goto out;
247 	}
248 
249 	/* Try with PLL */
250 	ret = nvbios_pll_parse(bios, pll, &limits);
251 	if (ret)
252 		return ret;
253 
254 	ret = gt215_clk_info(clock, idx - 0x10, limits.refclk, info);
255 	if (ret != limits.refclk)
256 		return -EINVAL;
257 
258 	ret = gt215_pll_calc(nv_subdev(clk), &limits, khz, &N, NULL, &M, &P);
259 	if (ret >= 0) {
260 		info->pll = (P << 16) | (N << 8) | M;
261 	}
262 
263 out:
264 	info->fb_delay = max(((khz + 7566) / 15133), (u32) 18);
265 	return ret ? ret : -ERANGE;
266 }
267 
268 static int
269 calc_clk(struct gt215_clk *clk, struct nvkm_cstate *cstate,
270 	 int idx, u32 pll, int dom)
271 {
272 	int ret = gt215_pll_info(&clk->base, idx, pll, cstate->domain[dom],
273 				 &clk->eng[dom]);
274 	if (ret >= 0)
275 		return 0;
276 	return ret;
277 }
278 
279 static int
280 calc_host(struct gt215_clk *clk, struct nvkm_cstate *cstate)
281 {
282 	int ret = 0;
283 	u32 kHz = cstate->domain[nv_clk_src_host];
284 	struct gt215_clk_info *info = &clk->eng[nv_clk_src_host];
285 
286 	if (kHz == 277000) {
287 		info->clk = 0;
288 		info->host_out = NVA3_HOST_277;
289 		return 0;
290 	}
291 
292 	info->host_out = NVA3_HOST_CLK;
293 
294 	ret = gt215_clk_info(&clk->base, 0x1d, kHz, info);
295 	if (ret >= 0)
296 		return 0;
297 
298 	return ret;
299 }
300 
301 int
302 gt215_clk_pre(struct nvkm_clk *clk, unsigned long *flags)
303 {
304 	struct nvkm_device *device = clk->subdev.device;
305 	struct nvkm_fifo *fifo = nvkm_fifo(clk);
306 
307 	/* halt and idle execution engines */
308 	nvkm_mask(device, 0x020060, 0x00070000, 0x00000000);
309 	nvkm_mask(device, 0x002504, 0x00000001, 0x00000001);
310 	/* Wait until the interrupt handler is finished */
311 	if (!nv_wait(clk, 0x000100, 0xffffffff, 0x00000000))
312 		return -EBUSY;
313 
314 	if (fifo)
315 		fifo->pause(fifo, flags);
316 
317 	if (!nv_wait(clk, 0x002504, 0x00000010, 0x00000010))
318 		return -EIO;
319 	if (!nv_wait(clk, 0x00251c, 0x0000003f, 0x0000003f))
320 		return -EIO;
321 
322 	return 0;
323 }
324 
325 void
326 gt215_clk_post(struct nvkm_clk *clk, unsigned long *flags)
327 {
328 	struct nvkm_device *device = clk->subdev.device;
329 	struct nvkm_fifo *fifo = nvkm_fifo(clk);
330 
331 	if (fifo && flags)
332 		fifo->start(fifo, flags);
333 
334 	nvkm_mask(device, 0x002504, 0x00000001, 0x00000000);
335 	nvkm_mask(device, 0x020060, 0x00070000, 0x00040000);
336 }
337 
338 static void
339 disable_clk_src(struct gt215_clk *clk, u32 src)
340 {
341 	struct nvkm_device *device = clk->base.subdev.device;
342 	nvkm_mask(device, src, 0x00000100, 0x00000000);
343 	nvkm_mask(device, src, 0x00000001, 0x00000000);
344 }
345 
346 static void
347 prog_pll(struct gt215_clk *clk, int idx, u32 pll, int dom)
348 {
349 	struct gt215_clk_info *info = &clk->eng[dom];
350 	struct nvkm_device *device = clk->base.subdev.device;
351 	const u32 src0 = 0x004120 + (idx * 4);
352 	const u32 src1 = 0x004160 + (idx * 4);
353 	const u32 ctrl = pll + 0;
354 	const u32 coef = pll + 4;
355 	u32 bypass;
356 
357 	if (info->pll) {
358 		/* Always start from a non-PLL clock */
359 		bypass = nvkm_rd32(device, ctrl)  & 0x00000008;
360 		if (!bypass) {
361 			nvkm_mask(device, src1, 0x00000101, 0x00000101);
362 			nvkm_mask(device, ctrl, 0x00000008, 0x00000008);
363 			udelay(20);
364 		}
365 
366 		nvkm_mask(device, src0, 0x003f3141, 0x00000101 | info->clk);
367 		nvkm_wr32(device, coef, info->pll);
368 		nvkm_mask(device, ctrl, 0x00000015, 0x00000015);
369 		nvkm_mask(device, ctrl, 0x00000010, 0x00000000);
370 		if (!nv_wait(clk, ctrl, 0x00020000, 0x00020000)) {
371 			nvkm_mask(device, ctrl, 0x00000010, 0x00000010);
372 			nvkm_mask(device, src0, 0x00000101, 0x00000000);
373 			return;
374 		}
375 		nvkm_mask(device, ctrl, 0x00000010, 0x00000010);
376 		nvkm_mask(device, ctrl, 0x00000008, 0x00000000);
377 		disable_clk_src(clk, src1);
378 	} else {
379 		nvkm_mask(device, src1, 0x003f3141, 0x00000101 | info->clk);
380 		nvkm_mask(device, ctrl, 0x00000018, 0x00000018);
381 		udelay(20);
382 		nvkm_mask(device, ctrl, 0x00000001, 0x00000000);
383 		disable_clk_src(clk, src0);
384 	}
385 }
386 
387 static void
388 prog_clk(struct gt215_clk *clk, int idx, int dom)
389 {
390 	struct gt215_clk_info *info = &clk->eng[dom];
391 	struct nvkm_device *device = clk->base.subdev.device;
392 	nvkm_mask(device, 0x004120 + (idx * 4), 0x003f3141, 0x00000101 | info->clk);
393 }
394 
395 static void
396 prog_host(struct gt215_clk *clk)
397 {
398 	struct gt215_clk_info *info = &clk->eng[nv_clk_src_host];
399 	struct nvkm_device *device = clk->base.subdev.device;
400 	u32 hsrc = (nvkm_rd32(device, 0xc040));
401 
402 	switch (info->host_out) {
403 	case NVA3_HOST_277:
404 		if ((hsrc & 0x30000000) == 0) {
405 			nvkm_wr32(device, 0xc040, hsrc | 0x20000000);
406 			disable_clk_src(clk, 0x4194);
407 		}
408 		break;
409 	case NVA3_HOST_CLK:
410 		prog_clk(clk, 0x1d, nv_clk_src_host);
411 		if ((hsrc & 0x30000000) >= 0x20000000) {
412 			nvkm_wr32(device, 0xc040, hsrc & ~0x30000000);
413 		}
414 		break;
415 	default:
416 		break;
417 	}
418 
419 	/* This seems to be a clock gating factor on idle, always set to 64 */
420 	nvkm_wr32(device, 0xc044, 0x3e);
421 }
422 
423 static void
424 prog_core(struct gt215_clk *clk, int dom)
425 {
426 	struct gt215_clk_info *info = &clk->eng[dom];
427 	struct nvkm_device *device = clk->base.subdev.device;
428 	u32 fb_delay = nvkm_rd32(device, 0x10002c);
429 
430 	if (fb_delay < info->fb_delay)
431 		nvkm_wr32(device, 0x10002c, info->fb_delay);
432 
433 	prog_pll(clk, 0x00, 0x004200, dom);
434 
435 	if (fb_delay > info->fb_delay)
436 		nvkm_wr32(device, 0x10002c, info->fb_delay);
437 }
438 
439 static int
440 gt215_clk_calc(struct nvkm_clk *obj, struct nvkm_cstate *cstate)
441 {
442 	struct gt215_clk *clk = container_of(obj, typeof(*clk), base);
443 	struct gt215_clk_info *core = &clk->eng[nv_clk_src_core];
444 	int ret;
445 
446 	if ((ret = calc_clk(clk, cstate, 0x10, 0x4200, nv_clk_src_core)) ||
447 	    (ret = calc_clk(clk, cstate, 0x11, 0x4220, nv_clk_src_shader)) ||
448 	    (ret = calc_clk(clk, cstate, 0x20, 0x0000, nv_clk_src_disp)) ||
449 	    (ret = calc_clk(clk, cstate, 0x21, 0x0000, nv_clk_src_vdec)) ||
450 	    (ret = calc_host(clk, cstate)))
451 		return ret;
452 
453 	/* XXX: Should be reading the highest bit in the VBIOS clock to decide
454 	 * whether to use a PLL or not... but using a PLL defeats the purpose */
455 	if (core->pll) {
456 		ret = gt215_clk_info(&clk->base, 0x10,
457 				     cstate->domain[nv_clk_src_core_intm],
458 				     &clk->eng[nv_clk_src_core_intm]);
459 		if (ret < 0)
460 			return ret;
461 	}
462 
463 	return 0;
464 }
465 
466 static int
467 gt215_clk_prog(struct nvkm_clk *obj)
468 {
469 	struct gt215_clk *clk = container_of(obj, typeof(*clk), base);
470 	struct gt215_clk_info *core = &clk->eng[nv_clk_src_core];
471 	int ret = 0;
472 	unsigned long flags;
473 	unsigned long *f = &flags;
474 
475 	ret = gt215_clk_pre(&clk->base, f);
476 	if (ret)
477 		goto out;
478 
479 	if (core->pll)
480 		prog_core(clk, nv_clk_src_core_intm);
481 
482 	prog_core(clk,  nv_clk_src_core);
483 	prog_pll(clk, 0x01, 0x004220, nv_clk_src_shader);
484 	prog_clk(clk, 0x20, nv_clk_src_disp);
485 	prog_clk(clk, 0x21, nv_clk_src_vdec);
486 	prog_host(clk);
487 
488 out:
489 	if (ret == -EBUSY)
490 		f = NULL;
491 
492 	gt215_clk_post(&clk->base, f);
493 	return ret;
494 }
495 
496 static void
497 gt215_clk_tidy(struct nvkm_clk *obj)
498 {
499 }
500 
501 static struct nvkm_domain
502 gt215_domain[] = {
503 	{ nv_clk_src_crystal  , 0xff },
504 	{ nv_clk_src_core     , 0x00, 0, "core", 1000 },
505 	{ nv_clk_src_shader   , 0x01, 0, "shader", 1000 },
506 	{ nv_clk_src_mem      , 0x02, 0, "memory", 1000 },
507 	{ nv_clk_src_vdec     , 0x03 },
508 	{ nv_clk_src_disp     , 0x04 },
509 	{ nv_clk_src_host     , 0x05 },
510 	{ nv_clk_src_core_intm, 0x06 },
511 	{ nv_clk_src_max }
512 };
513 
514 static int
515 gt215_clk_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
516 	       struct nvkm_oclass *oclass, void *data, u32 size,
517 	       struct nvkm_object **pobject)
518 {
519 	struct gt215_clk *clk;
520 	int ret;
521 
522 	ret = nvkm_clk_create(parent, engine, oclass, gt215_domain,
523 			      NULL, 0, true, &clk);
524 	*pobject = nv_object(clk);
525 	if (ret)
526 		return ret;
527 
528 	clk->base.read = gt215_clk_read;
529 	clk->base.calc = gt215_clk_calc;
530 	clk->base.prog = gt215_clk_prog;
531 	clk->base.tidy = gt215_clk_tidy;
532 	return 0;
533 }
534 
535 struct nvkm_oclass
536 gt215_clk_oclass = {
537 	.handle = NV_SUBDEV(CLK, 0xa3),
538 	.ofuncs = &(struct nvkm_ofuncs) {
539 		.ctor = gt215_clk_ctor,
540 		.dtor = _nvkm_clk_dtor,
541 		.init = _nvkm_clk_init,
542 		.fini = _nvkm_clk_fini,
543 	},
544 };
545