1 /*
2 * Copyright 2010 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
25 #include "drmP.h"
26 #include "nouveau_drv.h"
27 #include "nouveau_bios.h"
28 #include "nouveau_hw.h"
29 #include "nouveau_pm.h"
30 #include "nouveau_hwsq.h"
31 #include "nv50_display.h"
32
33 enum clk_src {
34 clk_src_crystal,
35 clk_src_href,
36 clk_src_hclk,
37 clk_src_hclkm3,
38 clk_src_hclkm3d2,
39 clk_src_host,
40 clk_src_nvclk,
41 clk_src_sclk,
42 clk_src_mclk,
43 clk_src_vdec,
44 clk_src_dom6
45 };
46
47 static u32 read_clk(struct drm_device *, enum clk_src);
48
49 static u32
read_div(struct drm_device * dev)50 read_div(struct drm_device *dev)
51 {
52 struct drm_nouveau_private *dev_priv = dev->dev_private;
53
54 switch (dev_priv->chipset) {
55 case 0x50: /* it exists, but only has bit 31, not the dividers.. */
56 case 0x84:
57 case 0x86:
58 case 0x98:
59 case 0xa0:
60 return nv_rd32(dev, 0x004700);
61 case 0x92:
62 case 0x94:
63 case 0x96:
64 return nv_rd32(dev, 0x004800);
65 default:
66 return 0x00000000;
67 }
68 }
69
70 static u32
read_pll_src(struct drm_device * dev,u32 base)71 read_pll_src(struct drm_device *dev, u32 base)
72 {
73 struct drm_nouveau_private *dev_priv = dev->dev_private;
74 u32 coef, ref = read_clk(dev, clk_src_crystal);
75 u32 rsel = nv_rd32(dev, 0x00e18c);
76 int P, N, M, id;
77
78 switch (dev_priv->chipset) {
79 case 0x50:
80 case 0xa0:
81 switch (base) {
82 case 0x4020:
83 case 0x4028: id = !!(rsel & 0x00000004); break;
84 case 0x4008: id = !!(rsel & 0x00000008); break;
85 case 0x4030: id = 0; break;
86 default:
87 NV_ERROR(dev, "ref: bad pll 0x%06x\n", base);
88 return 0;
89 }
90
91 coef = nv_rd32(dev, 0x00e81c + (id * 0x0c));
92 ref *= (coef & 0x01000000) ? 2 : 4;
93 P = (coef & 0x00070000) >> 16;
94 N = ((coef & 0x0000ff00) >> 8) + 1;
95 M = ((coef & 0x000000ff) >> 0) + 1;
96 break;
97 case 0x84:
98 case 0x86:
99 case 0x92:
100 coef = nv_rd32(dev, 0x00e81c);
101 P = (coef & 0x00070000) >> 16;
102 N = (coef & 0x0000ff00) >> 8;
103 M = (coef & 0x000000ff) >> 0;
104 break;
105 case 0x94:
106 case 0x96:
107 case 0x98:
108 rsel = nv_rd32(dev, 0x00c050);
109 switch (base) {
110 case 0x4020: rsel = (rsel & 0x00000003) >> 0; break;
111 case 0x4008: rsel = (rsel & 0x0000000c) >> 2; break;
112 case 0x4028: rsel = (rsel & 0x00001800) >> 11; break;
113 case 0x4030: rsel = 3; break;
114 default:
115 NV_ERROR(dev, "ref: bad pll 0x%06x\n", base);
116 return 0;
117 }
118
119 switch (rsel) {
120 case 0: id = 1; break;
121 case 1: return read_clk(dev, clk_src_crystal);
122 case 2: return read_clk(dev, clk_src_href);
123 case 3: id = 0; break;
124 }
125
126 coef = nv_rd32(dev, 0x00e81c + (id * 0x28));
127 P = (nv_rd32(dev, 0x00e824 + (id * 0x28)) >> 16) & 7;
128 P += (coef & 0x00070000) >> 16;
129 N = (coef & 0x0000ff00) >> 8;
130 M = (coef & 0x000000ff) >> 0;
131 break;
132 default:
133 BUG_ON(1);
134 }
135
136 if (M)
137 return (ref * N / M) >> P;
138 return 0;
139 }
140
141 static u32
read_pll_ref(struct drm_device * dev,u32 base)142 read_pll_ref(struct drm_device *dev, u32 base)
143 {
144 u32 src, mast = nv_rd32(dev, 0x00c040);
145
146 switch (base) {
147 case 0x004028:
148 src = !!(mast & 0x00200000);
149 break;
150 case 0x004020:
151 src = !!(mast & 0x00400000);
152 break;
153 case 0x004008:
154 src = !!(mast & 0x00010000);
155 break;
156 case 0x004030:
157 src = !!(mast & 0x02000000);
158 break;
159 case 0x00e810:
160 return read_clk(dev, clk_src_crystal);
161 default:
162 NV_ERROR(dev, "bad pll 0x%06x\n", base);
163 return 0;
164 }
165
166 if (src)
167 return read_clk(dev, clk_src_href);
168 return read_pll_src(dev, base);
169 }
170
171 static u32
read_pll(struct drm_device * dev,u32 base)172 read_pll(struct drm_device *dev, u32 base)
173 {
174 struct drm_nouveau_private *dev_priv = dev->dev_private;
175 u32 mast = nv_rd32(dev, 0x00c040);
176 u32 ctrl = nv_rd32(dev, base + 0);
177 u32 coef = nv_rd32(dev, base + 4);
178 u32 ref = read_pll_ref(dev, base);
179 u32 clk = 0;
180 int N1, N2, M1, M2;
181
182 if (base == 0x004028 && (mast & 0x00100000)) {
183 /* wtf, appears to only disable post-divider on nva0 */
184 if (dev_priv->chipset != 0xa0)
185 return read_clk(dev, clk_src_dom6);
186 }
187
188 N2 = (coef & 0xff000000) >> 24;
189 M2 = (coef & 0x00ff0000) >> 16;
190 N1 = (coef & 0x0000ff00) >> 8;
191 M1 = (coef & 0x000000ff);
192 if ((ctrl & 0x80000000) && M1) {
193 clk = ref * N1 / M1;
194 if ((ctrl & 0x40000100) == 0x40000000) {
195 if (M2)
196 clk = clk * N2 / M2;
197 else
198 clk = 0;
199 }
200 }
201
202 return clk;
203 }
204
205 static u32
read_clk(struct drm_device * dev,enum clk_src src)206 read_clk(struct drm_device *dev, enum clk_src src)
207 {
208 struct drm_nouveau_private *dev_priv = dev->dev_private;
209 u32 mast = nv_rd32(dev, 0x00c040);
210 u32 P = 0;
211
212 switch (src) {
213 case clk_src_crystal:
214 return dev_priv->crystal;
215 case clk_src_href:
216 return 100000; /* PCIE reference clock */
217 case clk_src_hclk:
218 return read_clk(dev, clk_src_href) * 27778 / 10000;
219 case clk_src_hclkm3:
220 return read_clk(dev, clk_src_hclk) * 3;
221 case clk_src_hclkm3d2:
222 return read_clk(dev, clk_src_hclk) * 3 / 2;
223 case clk_src_host:
224 switch (mast & 0x30000000) {
225 case 0x00000000: return read_clk(dev, clk_src_href);
226 case 0x10000000: break;
227 case 0x20000000: /* !0x50 */
228 case 0x30000000: return read_clk(dev, clk_src_hclk);
229 }
230 break;
231 case clk_src_nvclk:
232 if (!(mast & 0x00100000))
233 P = (nv_rd32(dev, 0x004028) & 0x00070000) >> 16;
234 switch (mast & 0x00000003) {
235 case 0x00000000: return read_clk(dev, clk_src_crystal) >> P;
236 case 0x00000001: return read_clk(dev, clk_src_dom6);
237 case 0x00000002: return read_pll(dev, 0x004020) >> P;
238 case 0x00000003: return read_pll(dev, 0x004028) >> P;
239 }
240 break;
241 case clk_src_sclk:
242 P = (nv_rd32(dev, 0x004020) & 0x00070000) >> 16;
243 switch (mast & 0x00000030) {
244 case 0x00000000:
245 if (mast & 0x00000080)
246 return read_clk(dev, clk_src_host) >> P;
247 return read_clk(dev, clk_src_crystal) >> P;
248 case 0x00000010: break;
249 case 0x00000020: return read_pll(dev, 0x004028) >> P;
250 case 0x00000030: return read_pll(dev, 0x004020) >> P;
251 }
252 break;
253 case clk_src_mclk:
254 P = (nv_rd32(dev, 0x004008) & 0x00070000) >> 16;
255 if (nv_rd32(dev, 0x004008) & 0x00000200) {
256 switch (mast & 0x0000c000) {
257 case 0x00000000:
258 return read_clk(dev, clk_src_crystal) >> P;
259 case 0x00008000:
260 case 0x0000c000:
261 return read_clk(dev, clk_src_href) >> P;
262 }
263 } else {
264 return read_pll(dev, 0x004008) >> P;
265 }
266 break;
267 case clk_src_vdec:
268 P = (read_div(dev) & 0x00000700) >> 8;
269 switch (dev_priv->chipset) {
270 case 0x84:
271 case 0x86:
272 case 0x92:
273 case 0x94:
274 case 0x96:
275 case 0xa0:
276 switch (mast & 0x00000c00) {
277 case 0x00000000:
278 if (dev_priv->chipset == 0xa0) /* wtf?? */
279 return read_clk(dev, clk_src_nvclk) >> P;
280 return read_clk(dev, clk_src_crystal) >> P;
281 case 0x00000400:
282 return 0;
283 case 0x00000800:
284 if (mast & 0x01000000)
285 return read_pll(dev, 0x004028) >> P;
286 return read_pll(dev, 0x004030) >> P;
287 case 0x00000c00:
288 return read_clk(dev, clk_src_nvclk) >> P;
289 }
290 break;
291 case 0x98:
292 switch (mast & 0x00000c00) {
293 case 0x00000000:
294 return read_clk(dev, clk_src_nvclk) >> P;
295 case 0x00000400:
296 return 0;
297 case 0x00000800:
298 return read_clk(dev, clk_src_hclkm3d2) >> P;
299 case 0x00000c00:
300 return read_clk(dev, clk_src_mclk) >> P;
301 }
302 break;
303 }
304 break;
305 case clk_src_dom6:
306 switch (dev_priv->chipset) {
307 case 0x50:
308 case 0xa0:
309 return read_pll(dev, 0x00e810) >> 2;
310 case 0x84:
311 case 0x86:
312 case 0x92:
313 case 0x94:
314 case 0x96:
315 case 0x98:
316 P = (read_div(dev) & 0x00000007) >> 0;
317 switch (mast & 0x0c000000) {
318 case 0x00000000: return read_clk(dev, clk_src_href);
319 case 0x04000000: break;
320 case 0x08000000: return read_clk(dev, clk_src_hclk);
321 case 0x0c000000:
322 return read_clk(dev, clk_src_hclkm3) >> P;
323 }
324 break;
325 default:
326 break;
327 }
328 default:
329 break;
330 }
331
332 NV_DEBUG(dev, "unknown clock source %d 0x%08x\n", src, mast);
333 return 0;
334 }
335
336 int
nv50_pm_clocks_get(struct drm_device * dev,struct nouveau_pm_level * perflvl)337 nv50_pm_clocks_get(struct drm_device *dev, struct nouveau_pm_level *perflvl)
338 {
339 struct drm_nouveau_private *dev_priv = dev->dev_private;
340 if (dev_priv->chipset == 0xaa ||
341 dev_priv->chipset == 0xac)
342 return 0;
343
344 perflvl->core = read_clk(dev, clk_src_nvclk);
345 perflvl->shader = read_clk(dev, clk_src_sclk);
346 perflvl->memory = read_clk(dev, clk_src_mclk);
347 if (dev_priv->chipset != 0x50) {
348 perflvl->vdec = read_clk(dev, clk_src_vdec);
349 perflvl->dom6 = read_clk(dev, clk_src_dom6);
350 }
351
352 return 0;
353 }
354
355 struct nv50_pm_state {
356 struct nouveau_pm_level *perflvl;
357 struct hwsq_ucode eclk_hwsq;
358 struct hwsq_ucode mclk_hwsq;
359 u32 mscript;
360 u32 mmast;
361 u32 mctrl;
362 u32 mcoef;
363 };
364
365 static u32
calc_pll(struct drm_device * dev,u32 reg,struct pll_lims * pll,u32 clk,int * N1,int * M1,int * log2P)366 calc_pll(struct drm_device *dev, u32 reg, struct pll_lims *pll,
367 u32 clk, int *N1, int *M1, int *log2P)
368 {
369 struct nouveau_pll_vals coef;
370 int ret;
371
372 ret = get_pll_limits(dev, reg, pll);
373 if (ret)
374 return 0;
375
376 pll->vco2.maxfreq = 0;
377 pll->refclk = read_pll_ref(dev, reg);
378 if (!pll->refclk)
379 return 0;
380
381 ret = nouveau_calc_pll_mnp(dev, pll, clk, &coef);
382 if (ret == 0)
383 return 0;
384
385 *N1 = coef.N1;
386 *M1 = coef.M1;
387 *log2P = coef.log2P;
388 return ret;
389 }
390
391 static inline u32
calc_div(u32 src,u32 target,int * div)392 calc_div(u32 src, u32 target, int *div)
393 {
394 u32 clk0 = src, clk1 = src;
395 for (*div = 0; *div <= 7; (*div)++) {
396 if (clk0 <= target) {
397 clk1 = clk0 << (*div ? 1 : 0);
398 break;
399 }
400 clk0 >>= 1;
401 }
402
403 if (target - clk0 <= clk1 - target)
404 return clk0;
405 (*div)--;
406 return clk1;
407 }
408
409 static inline u32
clk_same(u32 a,u32 b)410 clk_same(u32 a, u32 b)
411 {
412 return ((a / 1000) == (b / 1000));
413 }
414
415 static void
mclk_precharge(struct nouveau_mem_exec_func * exec)416 mclk_precharge(struct nouveau_mem_exec_func *exec)
417 {
418 struct nv50_pm_state *info = exec->priv;
419 struct hwsq_ucode *hwsq = &info->mclk_hwsq;
420
421 hwsq_wr32(hwsq, 0x1002d4, 0x00000001);
422 }
423
424 static void
mclk_refresh(struct nouveau_mem_exec_func * exec)425 mclk_refresh(struct nouveau_mem_exec_func *exec)
426 {
427 struct nv50_pm_state *info = exec->priv;
428 struct hwsq_ucode *hwsq = &info->mclk_hwsq;
429
430 hwsq_wr32(hwsq, 0x1002d0, 0x00000001);
431 }
432
433 static void
mclk_refresh_auto(struct nouveau_mem_exec_func * exec,bool enable)434 mclk_refresh_auto(struct nouveau_mem_exec_func *exec, bool enable)
435 {
436 struct nv50_pm_state *info = exec->priv;
437 struct hwsq_ucode *hwsq = &info->mclk_hwsq;
438
439 hwsq_wr32(hwsq, 0x100210, enable ? 0x80000000 : 0x00000000);
440 }
441
442 static void
mclk_refresh_self(struct nouveau_mem_exec_func * exec,bool enable)443 mclk_refresh_self(struct nouveau_mem_exec_func *exec, bool enable)
444 {
445 struct nv50_pm_state *info = exec->priv;
446 struct hwsq_ucode *hwsq = &info->mclk_hwsq;
447
448 hwsq_wr32(hwsq, 0x1002dc, enable ? 0x00000001 : 0x00000000);
449 }
450
451 static void
mclk_wait(struct nouveau_mem_exec_func * exec,u32 nsec)452 mclk_wait(struct nouveau_mem_exec_func *exec, u32 nsec)
453 {
454 struct nv50_pm_state *info = exec->priv;
455 struct hwsq_ucode *hwsq = &info->mclk_hwsq;
456
457 if (nsec > 1000)
458 hwsq_usec(hwsq, (nsec + 500) / 1000);
459 }
460
461 static u32
mclk_mrg(struct nouveau_mem_exec_func * exec,int mr)462 mclk_mrg(struct nouveau_mem_exec_func *exec, int mr)
463 {
464 if (mr <= 1)
465 return nv_rd32(exec->dev, 0x1002c0 + ((mr - 0) * 4));
466 if (mr <= 3)
467 return nv_rd32(exec->dev, 0x1002e0 + ((mr - 2) * 4));
468 return 0;
469 }
470
471 static void
mclk_mrs(struct nouveau_mem_exec_func * exec,int mr,u32 data)472 mclk_mrs(struct nouveau_mem_exec_func *exec, int mr, u32 data)
473 {
474 struct drm_nouveau_private *dev_priv = exec->dev->dev_private;
475 struct nv50_pm_state *info = exec->priv;
476 struct hwsq_ucode *hwsq = &info->mclk_hwsq;
477
478 if (mr <= 1) {
479 if (dev_priv->vram_rank_B)
480 hwsq_wr32(hwsq, 0x1002c8 + ((mr - 0) * 4), data);
481 hwsq_wr32(hwsq, 0x1002c0 + ((mr - 0) * 4), data);
482 } else
483 if (mr <= 3) {
484 if (dev_priv->vram_rank_B)
485 hwsq_wr32(hwsq, 0x1002e8 + ((mr - 2) * 4), data);
486 hwsq_wr32(hwsq, 0x1002e0 + ((mr - 2) * 4), data);
487 }
488 }
489
490 static void
mclk_clock_set(struct nouveau_mem_exec_func * exec)491 mclk_clock_set(struct nouveau_mem_exec_func *exec)
492 {
493 struct nv50_pm_state *info = exec->priv;
494 struct hwsq_ucode *hwsq = &info->mclk_hwsq;
495 u32 ctrl = nv_rd32(exec->dev, 0x004008);
496
497 info->mmast = nv_rd32(exec->dev, 0x00c040);
498 info->mmast &= ~0xc0000000; /* get MCLK_2 from HREF */
499 info->mmast |= 0x0000c000; /* use MCLK_2 as MPLL_BYPASS clock */
500
501 hwsq_wr32(hwsq, 0xc040, info->mmast);
502 hwsq_wr32(hwsq, 0x4008, ctrl | 0x00000200); /* bypass MPLL */
503 if (info->mctrl & 0x80000000)
504 hwsq_wr32(hwsq, 0x400c, info->mcoef);
505 hwsq_wr32(hwsq, 0x4008, info->mctrl);
506 }
507
508 static void
mclk_timing_set(struct nouveau_mem_exec_func * exec)509 mclk_timing_set(struct nouveau_mem_exec_func *exec)
510 {
511 struct drm_device *dev = exec->dev;
512 struct nv50_pm_state *info = exec->priv;
513 struct nouveau_pm_level *perflvl = info->perflvl;
514 struct hwsq_ucode *hwsq = &info->mclk_hwsq;
515 int i;
516
517 for (i = 0; i < 9; i++) {
518 u32 reg = 0x100220 + (i * 4);
519 u32 val = nv_rd32(dev, reg);
520 if (val != perflvl->timing.reg[i])
521 hwsq_wr32(hwsq, reg, perflvl->timing.reg[i]);
522 }
523 }
524
525 static int
calc_mclk(struct drm_device * dev,struct nouveau_pm_level * perflvl,struct nv50_pm_state * info)526 calc_mclk(struct drm_device *dev, struct nouveau_pm_level *perflvl,
527 struct nv50_pm_state *info)
528 {
529 struct drm_nouveau_private *dev_priv = dev->dev_private;
530 u32 crtc_mask = nv50_display_active_crtcs(dev);
531 struct nouveau_mem_exec_func exec = {
532 .dev = dev,
533 .precharge = mclk_precharge,
534 .refresh = mclk_refresh,
535 .refresh_auto = mclk_refresh_auto,
536 .refresh_self = mclk_refresh_self,
537 .wait = mclk_wait,
538 .mrg = mclk_mrg,
539 .mrs = mclk_mrs,
540 .clock_set = mclk_clock_set,
541 .timing_set = mclk_timing_set,
542 .priv = info
543 };
544 struct hwsq_ucode *hwsq = &info->mclk_hwsq;
545 struct pll_lims pll;
546 int N, M, P;
547 int ret;
548
549 /* use pcie refclock if possible, otherwise use mpll */
550 info->mctrl = nv_rd32(dev, 0x004008);
551 info->mctrl &= ~0x81ff0200;
552 if (clk_same(perflvl->memory, read_clk(dev, clk_src_href))) {
553 info->mctrl |= 0x00000200 | (pll.log2p_bias << 19);
554 } else {
555 ret = calc_pll(dev, 0x4008, &pll, perflvl->memory, &N, &M, &P);
556 if (ret == 0)
557 return -EINVAL;
558
559 info->mctrl |= 0x80000000 | (P << 22) | (P << 16);
560 info->mctrl |= pll.log2p_bias << 19;
561 info->mcoef = (N << 8) | M;
562 }
563
564 /* build the ucode which will reclock the memory for us */
565 hwsq_init(hwsq);
566 if (crtc_mask) {
567 hwsq_op5f(hwsq, crtc_mask, 0x00); /* wait for scanout */
568 hwsq_op5f(hwsq, crtc_mask, 0x01); /* wait for vblank */
569 }
570 if (dev_priv->chipset >= 0x92)
571 hwsq_wr32(hwsq, 0x611200, 0x00003300); /* disable scanout */
572 hwsq_setf(hwsq, 0x10, 0); /* disable bus access */
573 hwsq_op5f(hwsq, 0x00, 0x01); /* no idea :s */
574
575 ret = nouveau_mem_exec(&exec, perflvl);
576 if (ret)
577 return ret;
578
579 hwsq_setf(hwsq, 0x10, 1); /* enable bus access */
580 hwsq_op5f(hwsq, 0x00, 0x00); /* no idea, reverse of 0x00, 0x01? */
581 if (dev_priv->chipset >= 0x92)
582 hwsq_wr32(hwsq, 0x611200, 0x00003330); /* enable scanout */
583 hwsq_fini(hwsq);
584 return 0;
585 }
586
587 void *
nv50_pm_clocks_pre(struct drm_device * dev,struct nouveau_pm_level * perflvl)588 nv50_pm_clocks_pre(struct drm_device *dev, struct nouveau_pm_level *perflvl)
589 {
590 struct drm_nouveau_private *dev_priv = dev->dev_private;
591 struct nv50_pm_state *info;
592 struct hwsq_ucode *hwsq;
593 struct pll_lims pll;
594 u32 out, mast, divs, ctrl;
595 int clk, ret = -EINVAL;
596 int N, M, P1, P2;
597
598 if (dev_priv->chipset == 0xaa ||
599 dev_priv->chipset == 0xac)
600 return ERR_PTR(-ENODEV);
601
602 info = kmalloc(sizeof(*info), GFP_KERNEL);
603 if (!info)
604 return ERR_PTR(-ENOMEM);
605 info->perflvl = perflvl;
606
607 /* memory: build hwsq ucode which we'll use to reclock memory.
608 * use pcie refclock if possible, otherwise use mpll */
609 info->mclk_hwsq.len = 0;
610 if (perflvl->memory) {
611 ret = calc_mclk(dev, perflvl, info);
612 if (ret)
613 goto error;
614 info->mscript = perflvl->memscript;
615 }
616
617 divs = read_div(dev);
618 mast = info->mmast;
619
620 /* start building HWSQ script for engine reclocking */
621 hwsq = &info->eclk_hwsq;
622 hwsq_init(hwsq);
623 hwsq_setf(hwsq, 0x10, 0); /* disable bus access */
624 hwsq_op5f(hwsq, 0x00, 0x01); /* wait for access disabled? */
625
626 /* vdec/dom6: switch to "safe" clocks temporarily */
627 if (perflvl->vdec) {
628 mast &= ~0x00000c00;
629 divs &= ~0x00000700;
630 }
631
632 if (perflvl->dom6) {
633 mast &= ~0x0c000000;
634 divs &= ~0x00000007;
635 }
636
637 hwsq_wr32(hwsq, 0x00c040, mast);
638
639 /* vdec: avoid modifying xpll until we know exactly how the other
640 * clock domains work, i suspect at least some of them can also be
641 * tied to xpll...
642 */
643 if (perflvl->vdec) {
644 /* see how close we can get using nvclk as a source */
645 clk = calc_div(perflvl->core, perflvl->vdec, &P1);
646
647 /* see how close we can get using xpll/hclk as a source */
648 if (dev_priv->chipset != 0x98)
649 out = read_pll(dev, 0x004030);
650 else
651 out = read_clk(dev, clk_src_hclkm3d2);
652 out = calc_div(out, perflvl->vdec, &P2);
653
654 /* select whichever gets us closest */
655 if (abs((int)perflvl->vdec - clk) <=
656 abs((int)perflvl->vdec - out)) {
657 if (dev_priv->chipset != 0x98)
658 mast |= 0x00000c00;
659 divs |= P1 << 8;
660 } else {
661 mast |= 0x00000800;
662 divs |= P2 << 8;
663 }
664 }
665
666 /* dom6: nfi what this is, but we're limited to various combinations
667 * of the host clock frequency
668 */
669 if (perflvl->dom6) {
670 if (clk_same(perflvl->dom6, read_clk(dev, clk_src_href))) {
671 mast |= 0x00000000;
672 } else
673 if (clk_same(perflvl->dom6, read_clk(dev, clk_src_hclk))) {
674 mast |= 0x08000000;
675 } else {
676 clk = read_clk(dev, clk_src_hclk) * 3;
677 clk = calc_div(clk, perflvl->dom6, &P1);
678
679 mast |= 0x0c000000;
680 divs |= P1;
681 }
682 }
683
684 /* vdec/dom6: complete switch to new clocks */
685 switch (dev_priv->chipset) {
686 case 0x92:
687 case 0x94:
688 case 0x96:
689 hwsq_wr32(hwsq, 0x004800, divs);
690 break;
691 default:
692 hwsq_wr32(hwsq, 0x004700, divs);
693 break;
694 }
695
696 hwsq_wr32(hwsq, 0x00c040, mast);
697
698 /* core/shader: make sure sclk/nvclk are disconnected from their
699 * PLLs (nvclk to dom6, sclk to hclk)
700 */
701 if (dev_priv->chipset < 0x92)
702 mast = (mast & ~0x001000b0) | 0x00100080;
703 else
704 mast = (mast & ~0x000000b3) | 0x00000081;
705
706 hwsq_wr32(hwsq, 0x00c040, mast);
707
708 /* core: for the moment at least, always use nvpll */
709 clk = calc_pll(dev, 0x4028, &pll, perflvl->core, &N, &M, &P1);
710 if (clk == 0)
711 goto error;
712
713 ctrl = nv_rd32(dev, 0x004028) & ~0xc03f0100;
714 mast &= ~0x00100000;
715 mast |= 3;
716
717 hwsq_wr32(hwsq, 0x004028, 0x80000000 | (P1 << 19) | (P1 << 16) | ctrl);
718 hwsq_wr32(hwsq, 0x00402c, (N << 8) | M);
719
720 /* shader: tie to nvclk if possible, otherwise use spll. have to be
721 * very careful that the shader clock is at least twice the core, or
722 * some chipsets will be very unhappy. i expect most or all of these
723 * cases will be handled by tying to nvclk, but it's possible there's
724 * corners
725 */
726 ctrl = nv_rd32(dev, 0x004020) & ~0xc03f0100;
727
728 if (P1-- && perflvl->shader == (perflvl->core << 1)) {
729 hwsq_wr32(hwsq, 0x004020, (P1 << 19) | (P1 << 16) | ctrl);
730 hwsq_wr32(hwsq, 0x00c040, 0x00000020 | mast);
731 } else {
732 clk = calc_pll(dev, 0x4020, &pll, perflvl->shader, &N, &M, &P1);
733 if (clk == 0)
734 goto error;
735 ctrl |= 0x80000000;
736
737 hwsq_wr32(hwsq, 0x004020, (P1 << 19) | (P1 << 16) | ctrl);
738 hwsq_wr32(hwsq, 0x004024, (N << 8) | M);
739 hwsq_wr32(hwsq, 0x00c040, 0x00000030 | mast);
740 }
741
742 hwsq_setf(hwsq, 0x10, 1); /* enable bus access */
743 hwsq_op5f(hwsq, 0x00, 0x00); /* wait for access enabled? */
744 hwsq_fini(hwsq);
745
746 return info;
747 error:
748 kfree(info);
749 return ERR_PTR(ret);
750 }
751
752 static int
prog_hwsq(struct drm_device * dev,struct hwsq_ucode * hwsq)753 prog_hwsq(struct drm_device *dev, struct hwsq_ucode *hwsq)
754 {
755 struct drm_nouveau_private *dev_priv = dev->dev_private;
756 u32 hwsq_data, hwsq_kick;
757 int i;
758
759 if (dev_priv->chipset < 0x94) {
760 hwsq_data = 0x001400;
761 hwsq_kick = 0x00000003;
762 } else {
763 hwsq_data = 0x080000;
764 hwsq_kick = 0x00000001;
765 }
766 /* upload hwsq ucode */
767 nv_mask(dev, 0x001098, 0x00000008, 0x00000000);
768 nv_wr32(dev, 0x001304, 0x00000000);
769 if (dev_priv->chipset >= 0x92)
770 nv_wr32(dev, 0x001318, 0x00000000);
771 for (i = 0; i < hwsq->len / 4; i++)
772 nv_wr32(dev, hwsq_data + (i * 4), hwsq->ptr.u32[i]);
773 nv_mask(dev, 0x001098, 0x00000018, 0x00000018);
774
775 /* launch, and wait for completion */
776 nv_wr32(dev, 0x00130c, hwsq_kick);
777 if (!nv_wait(dev, 0x001308, 0x00000100, 0x00000000)) {
778 NV_ERROR(dev, "hwsq ucode exec timed out\n");
779 NV_ERROR(dev, "0x001308: 0x%08x\n", nv_rd32(dev, 0x001308));
780 for (i = 0; i < hwsq->len / 4; i++) {
781 NV_ERROR(dev, "0x%06x: 0x%08x\n", 0x1400 + (i * 4),
782 nv_rd32(dev, 0x001400 + (i * 4)));
783 }
784
785 return -EIO;
786 }
787
788 return 0;
789 }
790
791 int
nv50_pm_clocks_set(struct drm_device * dev,void * data)792 nv50_pm_clocks_set(struct drm_device *dev, void *data)
793 {
794 struct nv50_pm_state *info = data;
795 struct bit_entry M;
796 int ret = -EBUSY;
797
798 /* halt and idle execution engines */
799 nv_mask(dev, 0x002504, 0x00000001, 0x00000001);
800 if (!nv_wait(dev, 0x002504, 0x00000010, 0x00000010))
801 goto resume;
802 if (!nv_wait(dev, 0x00251c, 0x0000003f, 0x0000003f))
803 goto resume;
804
805 /* program memory clock, if necessary - must come before engine clock
806 * reprogramming due to how we construct the hwsq scripts in pre()
807 */
808 if (info->mclk_hwsq.len) {
809 /* execute some scripts that do ??? from the vbios.. */
810 if (!bit_table(dev, 'M', &M) && M.version == 1) {
811 if (M.length >= 6)
812 nouveau_bios_init_exec(dev, ROM16(M.data[5]));
813 if (M.length >= 8)
814 nouveau_bios_init_exec(dev, ROM16(M.data[7]));
815 if (M.length >= 10)
816 nouveau_bios_init_exec(dev, ROM16(M.data[9]));
817 nouveau_bios_init_exec(dev, info->mscript);
818 }
819
820 ret = prog_hwsq(dev, &info->mclk_hwsq);
821 if (ret)
822 goto resume;
823 }
824
825 /* program engine clocks */
826 ret = prog_hwsq(dev, &info->eclk_hwsq);
827
828 resume:
829 nv_mask(dev, 0x002504, 0x00000001, 0x00000000);
830 kfree(info);
831 return ret;
832 }
833
834 static int
pwm_info(struct drm_device * dev,int * line,int * ctrl,int * indx)835 pwm_info(struct drm_device *dev, int *line, int *ctrl, int *indx)
836 {
837 if (*line == 0x04) {
838 *ctrl = 0x00e100;
839 *line = 4;
840 *indx = 0;
841 } else
842 if (*line == 0x09) {
843 *ctrl = 0x00e100;
844 *line = 9;
845 *indx = 1;
846 } else
847 if (*line == 0x10) {
848 *ctrl = 0x00e28c;
849 *line = 0;
850 *indx = 0;
851 } else {
852 NV_ERROR(dev, "unknown pwm ctrl for gpio %d\n", *line);
853 return -ENODEV;
854 }
855
856 return 0;
857 }
858
859 int
nv50_pm_pwm_get(struct drm_device * dev,int line,u32 * divs,u32 * duty)860 nv50_pm_pwm_get(struct drm_device *dev, int line, u32 *divs, u32 *duty)
861 {
862 int ctrl, id, ret = pwm_info(dev, &line, &ctrl, &id);
863 if (ret)
864 return ret;
865
866 if (nv_rd32(dev, ctrl) & (1 << line)) {
867 *divs = nv_rd32(dev, 0x00e114 + (id * 8));
868 *duty = nv_rd32(dev, 0x00e118 + (id * 8));
869 return 0;
870 }
871
872 return -EINVAL;
873 }
874
875 int
nv50_pm_pwm_set(struct drm_device * dev,int line,u32 divs,u32 duty)876 nv50_pm_pwm_set(struct drm_device *dev, int line, u32 divs, u32 duty)
877 {
878 int ctrl, id, ret = pwm_info(dev, &line, &ctrl, &id);
879 if (ret)
880 return ret;
881
882 nv_mask(dev, ctrl, 0x00010001 << line, 0x00000001 << line);
883 nv_wr32(dev, 0x00e114 + (id * 8), divs);
884 nv_wr32(dev, 0x00e118 + (id * 8), duty | 0x80000000);
885 return 0;
886 }
887