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 #define gt215_clk(p) container_of((p), struct gt215_clk, base)
26 #include "gt215.h"
27 #include "pll.h"
28
29 #include <engine/fifo.h>
30 #include <subdev/bios.h>
31 #include <subdev/bios/pll.h>
32 #include <subdev/timer.h>
33
34 struct gt215_clk {
35 struct nvkm_clk base;
36 struct gt215_clk_info eng[nv_clk_src_max];
37 };
38
39 static u32 read_clk(struct gt215_clk *, int, bool);
40 static u32 read_pll(struct gt215_clk *, int, u32);
41
42 static u32
read_vco(struct gt215_clk * clk,int idx)43 read_vco(struct gt215_clk *clk, int idx)
44 {
45 struct nvkm_device *device = clk->base.subdev.device;
46 u32 sctl = nvkm_rd32(device, 0x4120 + (idx * 4));
47
48 switch (sctl & 0x00000030) {
49 case 0x00000000:
50 return device->crystal;
51 case 0x00000020:
52 return read_pll(clk, 0x41, 0x00e820);
53 case 0x00000030:
54 return read_pll(clk, 0x42, 0x00e8a0);
55 default:
56 return 0;
57 }
58 }
59
60 static u32
read_clk(struct gt215_clk * clk,int idx,bool ignore_en)61 read_clk(struct gt215_clk *clk, int idx, bool ignore_en)
62 {
63 struct nvkm_device *device = clk->base.subdev.device;
64 u32 sctl, sdiv, sclk;
65
66 /* refclk for the 0xe8xx plls is a fixed frequency */
67 if (idx >= 0x40) {
68 if (device->chipset == 0xaf) {
69 /* no joke.. seriously.. sigh.. */
70 return nvkm_rd32(device, 0x00471c) * 1000;
71 }
72
73 return device->crystal;
74 }
75
76 sctl = nvkm_rd32(device, 0x4120 + (idx * 4));
77 if (!ignore_en && !(sctl & 0x00000100))
78 return 0;
79
80 /* out_alt */
81 if (sctl & 0x00000400)
82 return 108000;
83
84 /* vco_out */
85 switch (sctl & 0x00003000) {
86 case 0x00000000:
87 if (!(sctl & 0x00000200))
88 return device->crystal;
89 return 0;
90 case 0x00002000:
91 if (sctl & 0x00000040)
92 return 108000;
93 return 100000;
94 case 0x00003000:
95 /* vco_enable */
96 if (!(sctl & 0x00000001))
97 return 0;
98
99 sclk = read_vco(clk, idx);
100 sdiv = ((sctl & 0x003f0000) >> 16) + 2;
101 return (sclk * 2) / sdiv;
102 default:
103 return 0;
104 }
105 }
106
107 static u32
read_pll(struct gt215_clk * clk,int idx,u32 pll)108 read_pll(struct gt215_clk *clk, int idx, u32 pll)
109 {
110 struct nvkm_device *device = clk->base.subdev.device;
111 u32 ctrl = nvkm_rd32(device, pll + 0);
112 u32 sclk = 0, P = 1, N = 1, M = 1;
113 u32 MP;
114
115 if (!(ctrl & 0x00000008)) {
116 if (ctrl & 0x00000001) {
117 u32 coef = nvkm_rd32(device, pll + 4);
118 M = (coef & 0x000000ff) >> 0;
119 N = (coef & 0x0000ff00) >> 8;
120 P = (coef & 0x003f0000) >> 16;
121
122 /* no post-divider on these..
123 * XXX: it looks more like two post-"dividers" that
124 * cross each other out in the default RPLL config */
125 if ((pll & 0x00ff00) == 0x00e800)
126 P = 1;
127
128 sclk = read_clk(clk, 0x00 + idx, false);
129 }
130 } else {
131 sclk = read_clk(clk, 0x10 + idx, false);
132 }
133
134 MP = M * P;
135
136 if (!MP)
137 return 0;
138
139 return sclk * N / MP;
140 }
141
142 static int
gt215_clk_read(struct nvkm_clk * base,enum nv_clk_src src)143 gt215_clk_read(struct nvkm_clk *base, enum nv_clk_src src)
144 {
145 struct gt215_clk *clk = gt215_clk(base);
146 struct nvkm_subdev *subdev = &clk->base.subdev;
147 struct nvkm_device *device = subdev->device;
148 u32 hsrc;
149
150 switch (src) {
151 case nv_clk_src_crystal:
152 return device->crystal;
153 case nv_clk_src_core:
154 case nv_clk_src_core_intm:
155 return read_pll(clk, 0x00, 0x4200);
156 case nv_clk_src_shader:
157 return read_pll(clk, 0x01, 0x4220);
158 case nv_clk_src_mem:
159 return read_pll(clk, 0x02, 0x4000);
160 case nv_clk_src_disp:
161 return read_clk(clk, 0x20, false);
162 case nv_clk_src_vdec:
163 return read_clk(clk, 0x21, false);
164 case nv_clk_src_pmu:
165 return read_clk(clk, 0x25, false);
166 case nv_clk_src_host:
167 hsrc = (nvkm_rd32(device, 0xc040) & 0x30000000) >> 28;
168 switch (hsrc) {
169 case 0:
170 return read_clk(clk, 0x1d, false);
171 case 2:
172 case 3:
173 return 277000;
174 default:
175 nvkm_error(subdev, "unknown HOST clock source %d\n", hsrc);
176 return -EINVAL;
177 }
178 default:
179 nvkm_error(subdev, "invalid clock source %d\n", src);
180 return -EINVAL;
181 }
182
183 return 0;
184 }
185
186 static int
gt215_clk_info(struct nvkm_clk * base,int idx,u32 khz,struct gt215_clk_info * info)187 gt215_clk_info(struct nvkm_clk *base, int idx, u32 khz,
188 struct gt215_clk_info *info)
189 {
190 struct gt215_clk *clk = gt215_clk(base);
191 u32 oclk, sclk, sdiv;
192 s32 diff;
193
194 info->clk = 0;
195
196 switch (khz) {
197 case 27000:
198 info->clk = 0x00000100;
199 return khz;
200 case 100000:
201 info->clk = 0x00002100;
202 return khz;
203 case 108000:
204 info->clk = 0x00002140;
205 return khz;
206 default:
207 sclk = read_vco(clk, idx);
208 sdiv = min((sclk * 2) / khz, (u32)65);
209 oclk = (sclk * 2) / sdiv;
210 diff = ((khz + 3000) - oclk);
211
212 /* When imprecise, play it safe and aim for a clock lower than
213 * desired rather than higher */
214 if (diff < 0) {
215 sdiv++;
216 oclk = (sclk * 2) / sdiv;
217 }
218
219 /* divider can go as low as 2, limited here because NVIDIA
220 * and the VBIOS on my NVA8 seem to prefer using the PLL
221 * for 810MHz - is there a good reason?
222 * XXX: PLLs with refclk 810MHz? */
223 if (sdiv > 4) {
224 info->clk = (((sdiv - 2) << 16) | 0x00003100);
225 return oclk;
226 }
227
228 break;
229 }
230
231 return -ERANGE;
232 }
233
234 int
gt215_pll_info(struct nvkm_clk * base,int idx,u32 pll,u32 khz,struct gt215_clk_info * info)235 gt215_pll_info(struct nvkm_clk *base, int idx, u32 pll, u32 khz,
236 struct gt215_clk_info *info)
237 {
238 struct gt215_clk *clk = gt215_clk(base);
239 struct nvkm_subdev *subdev = &clk->base.subdev;
240 struct nvbios_pll limits;
241 int P, N, M, diff;
242 int ret;
243
244 info->pll = 0;
245
246 /* If we can get a within [-2, 3) MHz of a divider, we'll disable the
247 * PLL and use the divider instead. */
248 ret = gt215_clk_info(&clk->base, idx, khz, info);
249 diff = khz - ret;
250 if (!pll || (diff >= -2000 && diff < 3000)) {
251 goto out;
252 }
253
254 /* Try with PLL */
255 ret = nvbios_pll_parse(subdev->device->bios, pll, &limits);
256 if (ret)
257 return ret;
258
259 ret = gt215_clk_info(&clk->base, idx - 0x10, limits.refclk, info);
260 if (ret != limits.refclk)
261 return -EINVAL;
262
263 ret = gt215_pll_calc(subdev, &limits, khz, &N, NULL, &M, &P);
264 if (ret >= 0) {
265 info->pll = (P << 16) | (N << 8) | M;
266 }
267
268 out:
269 info->fb_delay = max(((khz + 7566) / 15133), (u32) 18);
270 return ret ? ret : -ERANGE;
271 }
272
273 static int
calc_clk(struct gt215_clk * clk,struct nvkm_cstate * cstate,int idx,u32 pll,int dom)274 calc_clk(struct gt215_clk *clk, struct nvkm_cstate *cstate,
275 int idx, u32 pll, int dom)
276 {
277 int ret = gt215_pll_info(&clk->base, idx, pll, cstate->domain[dom],
278 &clk->eng[dom]);
279 if (ret >= 0)
280 return 0;
281 return ret;
282 }
283
284 static int
calc_host(struct gt215_clk * clk,struct nvkm_cstate * cstate)285 calc_host(struct gt215_clk *clk, struct nvkm_cstate *cstate)
286 {
287 int ret = 0;
288 u32 kHz = cstate->domain[nv_clk_src_host];
289 struct gt215_clk_info *info = &clk->eng[nv_clk_src_host];
290
291 if (kHz == 277000) {
292 info->clk = 0;
293 info->host_out = NVA3_HOST_277;
294 return 0;
295 }
296
297 info->host_out = NVA3_HOST_CLK;
298
299 ret = gt215_clk_info(&clk->base, 0x1d, kHz, info);
300 if (ret >= 0)
301 return 0;
302
303 return ret;
304 }
305
306 int
gt215_clk_pre(struct nvkm_clk * clk,unsigned long * flags)307 gt215_clk_pre(struct nvkm_clk *clk, unsigned long *flags)
308 {
309 struct nvkm_device *device = clk->subdev.device;
310 struct nvkm_fifo *fifo = device->fifo;
311
312 /* halt and idle execution engines */
313 nvkm_mask(device, 0x020060, 0x00070000, 0x00000000);
314 nvkm_mask(device, 0x002504, 0x00000001, 0x00000001);
315 /* Wait until the interrupt handler is finished */
316 if (nvkm_msec(device, 2000,
317 if (!nvkm_rd32(device, 0x000100))
318 break;
319 ) < 0)
320 return -EBUSY;
321
322 if (fifo)
323 nvkm_fifo_pause(fifo, flags);
324
325 if (nvkm_msec(device, 2000,
326 if (nvkm_rd32(device, 0x002504) & 0x00000010)
327 break;
328 ) < 0)
329 return -EIO;
330
331 if (nvkm_msec(device, 2000,
332 u32 tmp = nvkm_rd32(device, 0x00251c) & 0x0000003f;
333 if (tmp == 0x0000003f)
334 break;
335 ) < 0)
336 return -EIO;
337
338 return 0;
339 }
340
341 void
gt215_clk_post(struct nvkm_clk * clk,unsigned long * flags)342 gt215_clk_post(struct nvkm_clk *clk, unsigned long *flags)
343 {
344 struct nvkm_device *device = clk->subdev.device;
345 struct nvkm_fifo *fifo = device->fifo;
346
347 if (fifo && flags)
348 nvkm_fifo_start(fifo, flags);
349
350 nvkm_mask(device, 0x002504, 0x00000001, 0x00000000);
351 nvkm_mask(device, 0x020060, 0x00070000, 0x00040000);
352 }
353
354 static void
disable_clk_src(struct gt215_clk * clk,u32 src)355 disable_clk_src(struct gt215_clk *clk, u32 src)
356 {
357 struct nvkm_device *device = clk->base.subdev.device;
358 nvkm_mask(device, src, 0x00000100, 0x00000000);
359 nvkm_mask(device, src, 0x00000001, 0x00000000);
360 }
361
362 static void
prog_pll(struct gt215_clk * clk,int idx,u32 pll,int dom)363 prog_pll(struct gt215_clk *clk, int idx, u32 pll, int dom)
364 {
365 struct gt215_clk_info *info = &clk->eng[dom];
366 struct nvkm_device *device = clk->base.subdev.device;
367 const u32 src0 = 0x004120 + (idx * 4);
368 const u32 src1 = 0x004160 + (idx * 4);
369 const u32 ctrl = pll + 0;
370 const u32 coef = pll + 4;
371 u32 bypass;
372
373 if (info->pll) {
374 /* Always start from a non-PLL clock */
375 bypass = nvkm_rd32(device, ctrl) & 0x00000008;
376 if (!bypass) {
377 nvkm_mask(device, src1, 0x00000101, 0x00000101);
378 nvkm_mask(device, ctrl, 0x00000008, 0x00000008);
379 udelay(20);
380 }
381
382 nvkm_mask(device, src0, 0x003f3141, 0x00000101 | info->clk);
383 nvkm_wr32(device, coef, info->pll);
384 nvkm_mask(device, ctrl, 0x00000015, 0x00000015);
385 nvkm_mask(device, ctrl, 0x00000010, 0x00000000);
386 if (nvkm_msec(device, 2000,
387 if (nvkm_rd32(device, ctrl) & 0x00020000)
388 break;
389 ) < 0) {
390 nvkm_mask(device, ctrl, 0x00000010, 0x00000010);
391 nvkm_mask(device, src0, 0x00000101, 0x00000000);
392 return;
393 }
394 nvkm_mask(device, ctrl, 0x00000010, 0x00000010);
395 nvkm_mask(device, ctrl, 0x00000008, 0x00000000);
396 disable_clk_src(clk, src1);
397 } else {
398 nvkm_mask(device, src1, 0x003f3141, 0x00000101 | info->clk);
399 nvkm_mask(device, ctrl, 0x00000018, 0x00000018);
400 udelay(20);
401 nvkm_mask(device, ctrl, 0x00000001, 0x00000000);
402 disable_clk_src(clk, src0);
403 }
404 }
405
406 static void
prog_clk(struct gt215_clk * clk,int idx,int dom)407 prog_clk(struct gt215_clk *clk, int idx, int dom)
408 {
409 struct gt215_clk_info *info = &clk->eng[dom];
410 struct nvkm_device *device = clk->base.subdev.device;
411 nvkm_mask(device, 0x004120 + (idx * 4), 0x003f3141, 0x00000101 | info->clk);
412 }
413
414 static void
prog_host(struct gt215_clk * clk)415 prog_host(struct gt215_clk *clk)
416 {
417 struct gt215_clk_info *info = &clk->eng[nv_clk_src_host];
418 struct nvkm_device *device = clk->base.subdev.device;
419 u32 hsrc = (nvkm_rd32(device, 0xc040));
420
421 switch (info->host_out) {
422 case NVA3_HOST_277:
423 if ((hsrc & 0x30000000) == 0) {
424 nvkm_wr32(device, 0xc040, hsrc | 0x20000000);
425 disable_clk_src(clk, 0x4194);
426 }
427 break;
428 case NVA3_HOST_CLK:
429 prog_clk(clk, 0x1d, nv_clk_src_host);
430 if ((hsrc & 0x30000000) >= 0x20000000) {
431 nvkm_wr32(device, 0xc040, hsrc & ~0x30000000);
432 }
433 break;
434 default:
435 break;
436 }
437
438 /* This seems to be a clock gating factor on idle, always set to 64 */
439 nvkm_wr32(device, 0xc044, 0x3e);
440 }
441
442 static void
prog_core(struct gt215_clk * clk,int dom)443 prog_core(struct gt215_clk *clk, int dom)
444 {
445 struct gt215_clk_info *info = &clk->eng[dom];
446 struct nvkm_device *device = clk->base.subdev.device;
447 u32 fb_delay = nvkm_rd32(device, 0x10002c);
448
449 if (fb_delay < info->fb_delay)
450 nvkm_wr32(device, 0x10002c, info->fb_delay);
451
452 prog_pll(clk, 0x00, 0x004200, dom);
453
454 if (fb_delay > info->fb_delay)
455 nvkm_wr32(device, 0x10002c, info->fb_delay);
456 }
457
458 static int
gt215_clk_calc(struct nvkm_clk * base,struct nvkm_cstate * cstate)459 gt215_clk_calc(struct nvkm_clk *base, struct nvkm_cstate *cstate)
460 {
461 struct gt215_clk *clk = gt215_clk(base);
462 struct gt215_clk_info *core = &clk->eng[nv_clk_src_core];
463 int ret;
464
465 if ((ret = calc_clk(clk, cstate, 0x10, 0x4200, nv_clk_src_core)) ||
466 (ret = calc_clk(clk, cstate, 0x11, 0x4220, nv_clk_src_shader)) ||
467 (ret = calc_clk(clk, cstate, 0x20, 0x0000, nv_clk_src_disp)) ||
468 (ret = calc_clk(clk, cstate, 0x21, 0x0000, nv_clk_src_vdec)) ||
469 (ret = calc_host(clk, cstate)))
470 return ret;
471
472 /* XXX: Should be reading the highest bit in the VBIOS clock to decide
473 * whether to use a PLL or not... but using a PLL defeats the purpose */
474 if (core->pll) {
475 ret = gt215_clk_info(&clk->base, 0x10,
476 cstate->domain[nv_clk_src_core_intm],
477 &clk->eng[nv_clk_src_core_intm]);
478 if (ret < 0)
479 return ret;
480 }
481
482 return 0;
483 }
484
485 static int
gt215_clk_prog(struct nvkm_clk * base)486 gt215_clk_prog(struct nvkm_clk *base)
487 {
488 struct gt215_clk *clk = gt215_clk(base);
489 struct gt215_clk_info *core = &clk->eng[nv_clk_src_core];
490 int ret = 0;
491 unsigned long flags;
492 unsigned long *f = &flags;
493
494 ret = gt215_clk_pre(&clk->base, f);
495 if (ret)
496 goto out;
497
498 if (core->pll)
499 prog_core(clk, nv_clk_src_core_intm);
500
501 prog_core(clk, nv_clk_src_core);
502 prog_pll(clk, 0x01, 0x004220, nv_clk_src_shader);
503 prog_clk(clk, 0x20, nv_clk_src_disp);
504 prog_clk(clk, 0x21, nv_clk_src_vdec);
505 prog_host(clk);
506
507 out:
508 if (ret == -EBUSY)
509 f = NULL;
510
511 gt215_clk_post(&clk->base, f);
512 return ret;
513 }
514
515 static void
gt215_clk_tidy(struct nvkm_clk * base)516 gt215_clk_tidy(struct nvkm_clk *base)
517 {
518 }
519
520 static const struct nvkm_clk_func
521 gt215_clk = {
522 .read = gt215_clk_read,
523 .calc = gt215_clk_calc,
524 .prog = gt215_clk_prog,
525 .tidy = gt215_clk_tidy,
526 .domains = {
527 { nv_clk_src_crystal , 0xff },
528 { nv_clk_src_core , 0x00, 0, "core", 1000 },
529 { nv_clk_src_shader , 0x01, 0, "shader", 1000 },
530 { nv_clk_src_mem , 0x02, 0, "memory", 1000 },
531 { nv_clk_src_vdec , 0x03 },
532 { nv_clk_src_disp , 0x04 },
533 { nv_clk_src_host , 0x05 },
534 { nv_clk_src_core_intm, 0x06 },
535 { nv_clk_src_max }
536 }
537 };
538
539 int
gt215_clk_new(struct nvkm_device * device,int index,struct nvkm_clk ** pclk)540 gt215_clk_new(struct nvkm_device *device, int index, struct nvkm_clk **pclk)
541 {
542 struct gt215_clk *clk;
543
544 if (!(clk = kzalloc(sizeof(*clk), GFP_KERNEL)))
545 return -ENOMEM;
546 *pclk = &clk->base;
547
548 return nvkm_clk_ctor(>215_clk, device, index, true, &clk->base);
549 }
550