drivers/gpu/drm/nouveau/nvc0_pm.c

   1 /*
   2  * Copyright 2011 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_pm.h"
  29
  30 static u32 read_div(struct drm_device *, int, u32, u32);
  31 static u32 read_pll(struct drm_device *, u32);
  32
  33 static u32
  34 read_vco(struct drm_device *dev, u32 dsrc)
  35 {
  36         u32 ssrc = nv_rd32(dev, dsrc);
  37         if (!(ssrc & 0x00000100))
  38                 return read_pll(dev, 0x00e800);
  39         return read_pll(dev, 0x00e820);
  40 }
  41
  42 static u32
  43 read_pll(struct drm_device *dev, u32 pll)
  44 {
  45         u32 ctrl = nv_rd32(dev, pll + 0);
  46         u32 coef = nv_rd32(dev, pll + 4);
  47         u32 P = (coef & 0x003f0000) >> 16;
  48         u32 N = (coef & 0x0000ff00) >> 8;
  49         u32 M = (coef & 0x000000ff) >> 0;
  50         u32 sclk, doff;
  51
  52         if (!(ctrl & 0x00000001))
  53                 return 0;
  54
  55         switch (pll & 0xfff000) {
  56         case 0x00e000:
  57                 sclk = 27000;
  58                 P = 1;
  59                 break;
  60         case 0x137000:
  61                 doff = (pll - 0x137000) / 0x20;
  62                 sclk = read_div(dev, doff, 0x137120, 0x137140);
  63                 break;
  64         case 0x132000:
  65                 switch (pll) {
  66                 case 0x132000:
  67                         sclk = read_pll(dev, 0x132020);
  68                         break;
  69                 case 0x132020:
  70                         sclk = read_div(dev, 0, 0x137320, 0x137330);
  71                         break;
  72                 default:
  73                         return 0;
  74                 }
  75                 break;
  76         default:
  77                 return 0;
  78         }
  79
  80         return sclk * N / M / P;
  81 }
  82
  83 static u32
  84 read_div(struct drm_device *dev, int doff, u32 dsrc, u32 dctl)
  85 {
  86         u32 ssrc = nv_rd32(dev, dsrc + (doff * 4));
  87         u32 sctl = nv_rd32(dev, dctl + (doff * 4));
  88
  89         switch (ssrc & 0x00000003) {
  90         case 0:
  91                 if ((ssrc & 0x00030000) != 0x00030000)
  92                         return 27000;
  93                 return 108000;
  94         case 2:
  95                 return 100000;
  96         case 3:
  97                 if (sctl & 0x80000000) {
  98                         u32 sclk = read_vco(dev, dsrc + (doff * 4));
  99                         u32 sdiv = (sctl & 0x0000003f) + 2;
 100                         return (sclk * 2) / sdiv;
 101                 }
 102
 103                 return read_vco(dev, dsrc + (doff * 4));
 104         default:
 105                 return 0;
 106         }
 107 }
 108
 109 static u32
 110 read_mem(struct drm_device *dev)
 111 {
 112         u32 ssel = nv_rd32(dev, 0x1373f0);
 113         if (ssel & 0x00000001)
 114                 return read_div(dev, 0, 0x137300, 0x137310);
 115         return read_pll(dev, 0x132000);
 116 }
 117
 118 static u32
 119 read_clk(struct drm_device *dev, int clk)
 120 {
 121         u32 sctl = nv_rd32(dev, 0x137250 + (clk * 4));
 122         u32 ssel = nv_rd32(dev, 0x137100);
 123         u32 sclk, sdiv;
 124
 125         if (ssel & (1 << clk)) {
 126                 if (clk < 7)
 127                         sclk = read_pll(dev, 0x137000 + (clk * 0x20));
 128                 else
 129                         sclk = read_pll(dev, 0x1370e0);
 130                 sdiv = ((sctl & 0x00003f00) >> 8) + 2;
 131         } else {
 132                 sclk = read_div(dev, clk, 0x137160, 0x1371d0);
 133                 sdiv = ((sctl & 0x0000003f) >> 0) + 2;
 134         }
 135
 136         if (sctl & 0x80000000)
 137                 return (sclk * 2) / sdiv;
 138         return sclk;
 139 }
 140
 141 int
 142 nvc0_pm_clocks_get(struct drm_device *dev, struct nouveau_pm_level *perflvl)
 143 {
 144         perflvl->shader = read_clk(dev, 0x00);
 145         perflvl->core   = perflvl->shader / 2;
 146         perflvl->memory = read_mem(dev);
 147         perflvl->rop    = read_clk(dev, 0x01);
 148         perflvl->hub07  = read_clk(dev, 0x02);
 149         perflvl->hub06  = read_clk(dev, 0x07);
 150         perflvl->hub01  = read_clk(dev, 0x08);
 151         perflvl->copy   = read_clk(dev, 0x09);
 152         perflvl->daemon = read_clk(dev, 0x0c);
 153         perflvl->vdec   = read_clk(dev, 0x0e);
 154         return 0;
 155 }
 156
 157 struct nvc0_pm_clock {
 158         u32 freq;
 159         u32 ssel;
 160         u32 mdiv;
 161         u32 dsrc;
 162         u32 ddiv;
 163         u32 coef;
 164 };
 165
 166 struct nvc0_pm_state {
 167         struct nouveau_pm_level *perflvl;
 168         struct nvc0_pm_clock eng[16];
 169         struct nvc0_pm_clock mem;
 170 };
 171
 172 static u32
 173 calc_div(struct drm_device *dev, int clk, u32 ref, u32 freq, u32 *ddiv)
 174 {
 175         u32 div = min((ref * 2) / freq, (u32)65);
 176         if (div < 2)
 177                 div = 2;
 178
 179         *ddiv = div - 2;
 180         return (ref * 2) / div;
 181 }
 182
 183 static u32
 184 calc_src(struct drm_device *dev, int clk, u32 freq, u32 *dsrc, u32 *ddiv)
 185 {
 186         u32 sclk;
 187
 188         /* use one of the fixed frequencies if possible */
 189         *ddiv = 0x00000000;
 190         switch (freq) {
 191         case  27000:
 192         case 108000:
 193                 *dsrc = 0x00000000;
 194                 if (freq == 108000)
 195                         *dsrc |= 0x00030000;
 196                 return freq;
 197         case 100000:
 198                 *dsrc = 0x00000002;
 199                 return freq;
 200         default:
 201                 *dsrc = 0x00000003;
 202                 break;
 203         }
 204
 205         /* otherwise, calculate the closest divider */
 206         sclk = read_vco(dev, clk);
 207         if (clk < 7)
 208                 sclk = calc_div(dev, clk, sclk, freq, ddiv);
 209         return sclk;
 210 }
 211
 212 static u32
 213 calc_pll(struct drm_device *dev, int clk, u32 freq, u32 *coef)
 214 {
 215         struct pll_lims limits;
 216         int N, M, P, ret;
 217
 218         ret = get_pll_limits(dev, 0x137000 + (clk * 0x20), &limits);
 219         if (ret)
 220                 return 0;
 221
 222         limits.refclk = read_div(dev, clk, 0x137120, 0x137140);
 223         if (!limits.refclk)
 224                 return 0;
 225
 226         ret = nva3_calc_pll(dev, &limits, freq, &N, NULL, &M, &P);
 227         if (ret <= 0)
 228                 return 0;
 229
 230         *coef = (P << 16) | (N << 8) | M;
 231         return ret;
 232 }
 233
 234 /* A (likely rather simplified and incomplete) view of the clock tree
 235  *
 236  * Key:
 237  *
 238  * S: source select
 239  * D: divider
 240  * P: pll
 241  * F: switch
 242  *
 243  * Engine clocks:
 244  *
 245  * 137250(D) ---- 137100(F0) ---- 137160(S)/1371d0(D) ------------------- ref
 246  *                      (F1) ---- 1370X0(P) ---- 137120(S)/137140(D) ---- ref
 247  *
 248  * Not all registers exist for all clocks.  For example: clocks >= 8 don't
 249  * have their own PLL (all tied to clock 7's PLL when in PLL mode), nor do
 250  * they have the divider at 1371d0, though the source selection at 137160
 251  * still exists.  You must use the divider at 137250 for these instead.
 252  *
 253  * Memory clock:
 254  *
 255  * TBD, read_mem() above is likely very wrong...
 256  *
 257  */
 258
 259 static int
 260 calc_clk(struct drm_device *dev, int clk, struct nvc0_pm_clock *info, u32 freq)
 261 {
 262         u32 src0, div0, div1D, div1P = 0;
 263         u32 clk0, clk1 = 0;
 264
 265         /* invalid clock domain */
 266         if (!freq)
 267                 return 0;
 268
 269         /* first possible path, using only dividers */
 270         clk0 = calc_src(dev, clk, freq, &src0, &div0);
 271         clk0 = calc_div(dev, clk, clk0, freq, &div1D);
 272
 273         /* see if we can get any closer using PLLs */
 274         if (clk0 != freq && (0x00004387 & (1 << clk))) {
 275                 if (clk < 7)
 276                         clk1 = calc_pll(dev, clk, freq, &info->coef);
 277                 else
 278                         clk1 = read_pll(dev, 0x1370e0);
 279                 clk1 = calc_div(dev, clk, clk1, freq, &div1P);
 280         }
 281
 282         /* select the method which gets closest to target freq */
 283         if (abs((int)freq - clk0) <= abs((int)freq - clk1)) {
 284                 info->dsrc = src0;
 285                 if (div0) {
 286                         info->ddiv |= 0x80000000;
 287                         info->ddiv |= div0 << 8;
 288                         info->ddiv |= div0;
 289                 }
 290                 if (div1D) {
 291                         info->mdiv |= 0x80000000;
 292                         info->mdiv |= div1D;
 293                 }
 294                 info->ssel = 0;
 295                 info->freq = clk0;
 296         } else {
 297                 if (div1P) {
 298                         info->mdiv |= 0x80000000;
 299                         info->mdiv |= div1P << 8;
 300                 }
 301                 info->ssel = (1 << clk);
 302                 info->freq = clk1;
 303         }
 304
 305         return 0;
 306 }
 307
 308 static int
 309 calc_mem(struct drm_device *dev, struct nvc0_pm_clock *info, u32 freq)
 310 {
 311         struct pll_lims pll;
 312         int N, M, P, ret;
 313         u32 ctrl;
 314
 315         /* mclk pll input freq comes from another pll, make sure it's on */
 316         ctrl = nv_rd32(dev, 0x132020);
 317         if (!(ctrl & 0x00000001)) {
 318                 /* if not, program it to 567MHz.  nfi where this value comes
 319                  * from - it looks like it's in the pll limits table for
 320                  * 132000 but the binary driver ignores all my attempts to
 321                  * change this value.
 322                  */
 323                 nv_wr32(dev, 0x137320, 0x00000103);
 324                 nv_wr32(dev, 0x137330, 0x81200606);
 325                 nv_wait(dev, 0x132020, 0x00010000, 0x00010000);
 326                 nv_wr32(dev, 0x132024, 0x0001150f);
 327                 nv_mask(dev, 0x132020, 0x00000001, 0x00000001);
 328                 nv_wait(dev, 0x137390, 0x00020000, 0x00020000);
 329                 nv_mask(dev, 0x132020, 0x00000004, 0x00000004);
 330         }
 331
 332         /* for the moment, until the clock tree is better understood, use
 333          * pll mode for all clock frequencies
 334          */
 335         ret = get_pll_limits(dev, 0x132000, &pll);
 336         if (ret == 0) {
 337                 pll.refclk = read_pll(dev, 0x132020);
 338                 if (pll.refclk) {
 339                         ret = nva3_calc_pll(dev, &pll, freq, &N, NULL, &M, &P);
 340                         if (ret > 0) {
 341                                 info->coef = (P << 16) | (N << 8) | M;
 342                                 return 0;
 343                         }
 344                 }
 345         }
 346
 347         return -EINVAL;
 348 }
 349
 350 void *
 351 nvc0_pm_clocks_pre(struct drm_device *dev, struct nouveau_pm_level *perflvl)
 352 {
 353         struct drm_nouveau_private *dev_priv = dev->dev_private;
 354         struct nvc0_pm_state *info;
 355         int ret;
 356
 357         info = kzalloc(sizeof(*info), GFP_KERNEL);
 358         if (!info)
 359                 return ERR_PTR(-ENOMEM);
 360
 361         /* NFI why this is still in the performance table, the ROPCs appear
 362          * to get their clock from clock 2 ("hub07", actually hub05 on this
 363          * chip, but, anyway...) as well.  nvatiming confirms hub05 and ROP
 364          * are always the same freq with the binary driver even when the
 365          * performance table says they should differ.
 366          */
 367         if (dev_priv->chipset == 0xd9)
 368                 perflvl->rop = 0;
 369
 370         if ((ret = calc_clk(dev, 0x00, &info->eng[0x00], perflvl->shader)) ||
 371             (ret = calc_clk(dev, 0x01, &info->eng[0x01], perflvl->rop)) ||
 372             (ret = calc_clk(dev, 0x02, &info->eng[0x02], perflvl->hub07)) ||
 373             (ret = calc_clk(dev, 0x07, &info->eng[0x07], perflvl->hub06)) ||
 374             (ret = calc_clk(dev, 0x08, &info->eng[0x08], perflvl->hub01)) ||
 375             (ret = calc_clk(dev, 0x09, &info->eng[0x09], perflvl->copy)) ||
 376             (ret = calc_clk(dev, 0x0c, &info->eng[0x0c], perflvl->daemon)) ||
 377             (ret = calc_clk(dev, 0x0e, &info->eng[0x0e], perflvl->vdec))) {
 378                 kfree(info);
 379                 return ERR_PTR(ret);
 380         }
 381
 382         if (perflvl->memory) {
 383                 ret = calc_mem(dev, &info->mem, perflvl->memory);
 384                 if (ret) {
 385                         kfree(info);
 386                         return ERR_PTR(ret);
 387                 }
 388         }
 389
 390         info->perflvl = perflvl;
 391         return info;
 392 }
 393
 394 static void
 395 prog_clk(struct drm_device *dev, int clk, struct nvc0_pm_clock *info)
 396 {
 397         /* program dividers at 137160/1371d0 first */
 398         if (clk < 7 && !info->ssel) {
 399                 nv_mask(dev, 0x1371d0 + (clk * 0x04), 0x80003f3f, info->ddiv);
 400                 nv_wr32(dev, 0x137160 + (clk * 0x04), info->dsrc);
 401         }
 402
 403         /* switch clock to non-pll mode */
 404         nv_mask(dev, 0x137100, (1 << clk), 0x00000000);
 405         nv_wait(dev, 0x137100, (1 << clk), 0x00000000);
 406
 407         /* reprogram pll */
 408         if (clk < 7) {
 409                 /* make sure it's disabled first... */
 410                 u32 base = 0x137000 + (clk * 0x20);
 411                 u32 ctrl = nv_rd32(dev, base + 0x00);
 412                 if (ctrl & 0x00000001) {
 413                         nv_mask(dev, base + 0x00, 0x00000004, 0x00000000);
 414                         nv_mask(dev, base + 0x00, 0x00000001, 0x00000000);
 415                 }
 416                 /* program it to new values, if necessary */
 417                 if (info->ssel) {
 418                         nv_wr32(dev, base + 0x04, info->coef);
 419                         nv_mask(dev, base + 0x00, 0x00000001, 0x00000001);
 420                         nv_wait(dev, base + 0x00, 0x00020000, 0x00020000);
 421                         nv_mask(dev, base + 0x00, 0x00020004, 0x00000004);
 422                 }
 423         }
 424
 425         /* select pll/non-pll mode, and program final clock divider */
 426         nv_mask(dev, 0x137100, (1 << clk), info->ssel);
 427         nv_wait(dev, 0x137100, (1 << clk), info->ssel);
 428         nv_mask(dev, 0x137250 + (clk * 0x04), 0x00003f3f, info->mdiv);
 429 }
 430
 431 static void
 432 mclk_precharge(struct nouveau_mem_exec_func *exec)
 433 {
 434 }
 435
 436 static void
 437 mclk_refresh(struct nouveau_mem_exec_func *exec)
 438 {
 439 }
 440
 441 static void
 442 mclk_refresh_auto(struct nouveau_mem_exec_func *exec, bool enable)
 443 {
 444         nv_wr32(exec->dev, 0x10f210, enable ? 0x80000000 : 0x00000000);
 445 }
 446
 447 static void
 448 mclk_refresh_self(struct nouveau_mem_exec_func *exec, bool enable)
 449 {
 450 }
 451
 452 static void
 453 mclk_wait(struct nouveau_mem_exec_func *exec, u32 nsec)
 454 {
 455         udelay((nsec + 500) / 1000);
 456 }
 457
 458 static u32
 459 mclk_mrg(struct nouveau_mem_exec_func *exec, int mr)
 460 {
 461         struct drm_device *dev = exec->dev;
 462         struct drm_nouveau_private *dev_priv = dev->dev_private;
 463         if (dev_priv->vram_type != NV_MEM_TYPE_GDDR5) {
 464                 if (mr <= 1)
 465                         return nv_rd32(dev, 0x10f300 + ((mr - 0) * 4));
 466                 return nv_rd32(dev, 0x10f320 + ((mr - 2) * 4));
 467         } else {
 468                 if (mr == 0)
 469                         return nv_rd32(dev, 0x10f300 + (mr * 4));
 470                 else
 471                 if (mr <= 7)
 472                         return nv_rd32(dev, 0x10f32c + (mr * 4));
 473                 return nv_rd32(dev, 0x10f34c);
 474         }
 475 }
 476
 477 static void
 478 mclk_mrs(struct nouveau_mem_exec_func *exec, int mr, u32 data)
 479 {
 480         struct drm_device *dev = exec->dev;
 481         struct drm_nouveau_private *dev_priv = dev->dev_private;
 482         if (dev_priv->vram_type != NV_MEM_TYPE_GDDR5) {
 483                 if (mr <= 1) {
 484                         nv_wr32(dev, 0x10f300 + ((mr - 0) * 4), data);
 485                         if (dev_priv->vram_rank_B)
 486                                 nv_wr32(dev, 0x10f308 + ((mr - 0) * 4), data);
 487                 } else
 488                 if (mr <= 3) {
 489                         nv_wr32(dev, 0x10f320 + ((mr - 2) * 4), data);
 490                         if (dev_priv->vram_rank_B)
 491                                 nv_wr32(dev, 0x10f328 + ((mr - 2) * 4), data);
 492                 }
 493         } else {
 494                 if      (mr ==  0) nv_wr32(dev, 0x10f300 + (mr * 4), data);
 495                 else if (mr <=  7) nv_wr32(dev, 0x10f32c + (mr * 4), data);
 496                 else if (mr == 15) nv_wr32(dev, 0x10f34c, data);
 497         }
 498 }
 499
 500 static void
 501 mclk_clock_set(struct nouveau_mem_exec_func *exec)
 502 {
 503         struct nvc0_pm_state *info = exec->priv;
 504         struct drm_device *dev = exec->dev;
 505         u32 ctrl = nv_rd32(dev, 0x132000);
 506
 507         nv_wr32(dev, 0x137360, 0x00000001);
 508         nv_wr32(dev, 0x137370, 0x00000000);
 509         nv_wr32(dev, 0x137380, 0x00000000);
 510         if (ctrl & 0x00000001)
 511                 nv_wr32(dev, 0x132000, (ctrl &= ~0x00000001));
 512
 513         nv_wr32(dev, 0x132004, info->mem.coef);
 514         nv_wr32(dev, 0x132000, (ctrl |= 0x00000001));
 515         nv_wait(dev, 0x137390, 0x00000002, 0x00000002);
 516         nv_wr32(dev, 0x132018, 0x00005000);
 517
 518         nv_wr32(dev, 0x137370, 0x00000001);
 519         nv_wr32(dev, 0x137380, 0x00000001);
 520         nv_wr32(dev, 0x137360, 0x00000000);
 521 }
 522
 523 static void
 524 mclk_timing_set(struct nouveau_mem_exec_func *exec)
 525 {
 526         struct nvc0_pm_state *info = exec->priv;
 527         struct nouveau_pm_level *perflvl = info->perflvl;
 528         int i;
 529
 530         for (i = 0; i < 5; i++)
 531                 nv_wr32(exec->dev, 0x10f290 + (i * 4), perflvl->timing.reg[i]);
 532 }
 533
 534 static void
 535 prog_mem(struct drm_device *dev, struct nvc0_pm_state *info)
 536 {
 537         struct drm_nouveau_private *dev_priv = dev->dev_private;
 538         struct nouveau_mem_exec_func exec = {
 539                 .dev = dev,
 540                 .precharge = mclk_precharge,
 541                 .refresh = mclk_refresh,
 542                 .refresh_auto = mclk_refresh_auto,
 543                 .refresh_self = mclk_refresh_self,
 544                 .wait = mclk_wait,
 545                 .mrg = mclk_mrg,
 546                 .mrs = mclk_mrs,
 547                 .clock_set = mclk_clock_set,
 548                 .timing_set = mclk_timing_set,
 549                 .priv = info
 550         };
 551
 552         if (dev_priv->chipset < 0xd0)
 553                 nv_wr32(dev, 0x611200, 0x00003300);
 554         else
 555                 nv_wr32(dev, 0x62c000, 0x03030000);
 556
 557         nouveau_mem_exec(&exec, info->perflvl);
 558
 559         if (dev_priv->chipset < 0xd0)
 560                 nv_wr32(dev, 0x611200, 0x00003330);
 561         else
 562                 nv_wr32(dev, 0x62c000, 0x03030300);
 563 }
 564 int
 565 nvc0_pm_clocks_set(struct drm_device *dev, void *data)
 566 {
 567         struct nvc0_pm_state *info = data;
 568         int i;
 569
 570         if (info->mem.coef)
 571                 prog_mem(dev, info);
 572
 573         for (i = 0; i < 16; i++) {
 574                 if (!info->eng[i].freq)
 575                         continue;
 576                 prog_clk(dev, i, &info->eng[i]);
 577         }
 578
 579         kfree(info);
 580         return 0;
 581 }