drivers/gpu/drm/i915/intel_pm.c

   1 /*
   2  * Copyright © 2012 Intel Corporation
   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 (including the next
  12  * paragraph) shall be included in all copies or substantial portions of the
  13  * Software.
  14  *
  15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  17  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  18  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  19  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  20  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
  21  * IN THE SOFTWARE.
  22  *
  23  * Authors:
  24  *    Eugeni Dodonov <eugeni.dodonov@intel.com>
  25  *
  26  */
  27
  28 #include <linux/cpufreq.h>
  29 #include "i915_drv.h"
  30 #include "intel_drv.h"
  31 #include "../../../platform/x86/intel_ips.h"
  32 #include <linux/module.h>
  33 #include <drm/i915_powerwell.h>
  34
  35 /* FBC, or Frame Buffer Compression, is a technique employed to compress the
  36  * framebuffer contents in-memory, aiming at reducing the required bandwidth
  37  * during in-memory transfers and, therefore, reduce the power packet.
  38  *
  39  * The benefits of FBC are mostly visible with solid backgrounds and
  40  * variation-less patterns.
  41  *
  42  * FBC-related functionality can be enabled by the means of the
  43  * i915.i915_enable_fbc parameter
  44  */
  45
  46 static void i8xx_disable_fbc(struct drm_device *dev)
  47 {
  48         struct drm_i915_private *dev_priv = dev->dev_private;
  49         u32 fbc_ctl;
  50
  51         /* Disable compression */
  52         fbc_ctl = I915_READ(FBC_CONTROL);
  53         if ((fbc_ctl & FBC_CTL_EN) == 0)
  54                 return;
  55
  56         fbc_ctl &= ~FBC_CTL_EN;
  57         I915_WRITE(FBC_CONTROL, fbc_ctl);
  58
  59         /* Wait for compressing bit to clear */
  60         if (wait_for((I915_READ(FBC_STATUS) & FBC_STAT_COMPRESSING) == 0, 10)) {
  61                 DRM_DEBUG_KMS("FBC idle timed out\n");
  62                 return;
  63         }
  64
  65         DRM_DEBUG_KMS("disabled FBC\n");
  66 }
  67
  68 static void i8xx_enable_fbc(struct drm_crtc *crtc, unsigned long interval)
  69 {
  70         struct drm_device *dev = crtc->dev;
  71         struct drm_i915_private *dev_priv = dev->dev_private;
  72         struct drm_framebuffer *fb = crtc->fb;
  73         struct intel_framebuffer *intel_fb = to_intel_framebuffer(fb);
  74         struct drm_i915_gem_object *obj = intel_fb->obj;
  75         struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
  76         int cfb_pitch;
  77         int plane, i;
  78         u32 fbc_ctl, fbc_ctl2;
  79
  80         cfb_pitch = dev_priv->fbc.size / FBC_LL_SIZE;
  81         if (fb->pitches[0] < cfb_pitch)
  82                 cfb_pitch = fb->pitches[0];
  83
  84         /* FBC_CTL wants 64B units */
  85         cfb_pitch = (cfb_pitch / 64) - 1;
  86         plane = intel_crtc->plane == 0 ? FBC_CTL_PLANEA : FBC_CTL_PLANEB;
  87
  88         /* Clear old tags */
  89         for (i = 0; i < (FBC_LL_SIZE / 32) + 1; i++)
  90                 I915_WRITE(FBC_TAG + (i * 4), 0);
  91
  92         /* Set it up... */
  93         fbc_ctl2 = FBC_CTL_FENCE_DBL | FBC_CTL_IDLE_IMM | FBC_CTL_CPU_FENCE;
  94         fbc_ctl2 |= plane;
  95         I915_WRITE(FBC_CONTROL2, fbc_ctl2);
  96         I915_WRITE(FBC_FENCE_OFF, crtc->y);
  97
  98         /* enable it... */
  99         fbc_ctl = FBC_CTL_EN | FBC_CTL_PERIODIC;
 100         if (IS_I945GM(dev))
 101                 fbc_ctl |= FBC_CTL_C3_IDLE; /* 945 needs special SR handling */
 102         fbc_ctl |= (cfb_pitch & 0xff) << FBC_CTL_STRIDE_SHIFT;
 103         fbc_ctl |= (interval & 0x2fff) << FBC_CTL_INTERVAL_SHIFT;
 104         fbc_ctl |= obj->fence_reg;
 105         I915_WRITE(FBC_CONTROL, fbc_ctl);
 106
 107         DRM_DEBUG_KMS("enabled FBC, pitch %d, yoff %d, plane %c, ",
 108                       cfb_pitch, crtc->y, plane_name(intel_crtc->plane));
 109 }
 110
 111 static bool i8xx_fbc_enabled(struct drm_device *dev)
 112 {
 113         struct drm_i915_private *dev_priv = dev->dev_private;
 114
 115         return I915_READ(FBC_CONTROL) & FBC_CTL_EN;
 116 }
 117
 118 static void g4x_enable_fbc(struct drm_crtc *crtc, unsigned long interval)
 119 {
 120         struct drm_device *dev = crtc->dev;
 121         struct drm_i915_private *dev_priv = dev->dev_private;
 122         struct drm_framebuffer *fb = crtc->fb;
 123         struct intel_framebuffer *intel_fb = to_intel_framebuffer(fb);
 124         struct drm_i915_gem_object *obj = intel_fb->obj;
 125         struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
 126         int plane = intel_crtc->plane == 0 ? DPFC_CTL_PLANEA : DPFC_CTL_PLANEB;
 127         unsigned long stall_watermark = 200;
 128         u32 dpfc_ctl;
 129
 130         dpfc_ctl = plane | DPFC_SR_EN | DPFC_CTL_LIMIT_1X;
 131         dpfc_ctl |= DPFC_CTL_FENCE_EN | obj->fence_reg;
 132         I915_WRITE(DPFC_CHICKEN, DPFC_HT_MODIFY);
 133
 134         I915_WRITE(DPFC_RECOMP_CTL, DPFC_RECOMP_STALL_EN |
 135                    (stall_watermark << DPFC_RECOMP_STALL_WM_SHIFT) |
 136                    (interval << DPFC_RECOMP_TIMER_COUNT_SHIFT));
 137         I915_WRITE(DPFC_FENCE_YOFF, crtc->y);
 138
 139         /* enable it... */
 140         I915_WRITE(DPFC_CONTROL, I915_READ(DPFC_CONTROL) | DPFC_CTL_EN);
 141
 142         DRM_DEBUG_KMS("enabled fbc on plane %c\n", plane_name(intel_crtc->plane));
 143 }
 144
 145 static void g4x_disable_fbc(struct drm_device *dev)
 146 {
 147         struct drm_i915_private *dev_priv = dev->dev_private;
 148         u32 dpfc_ctl;
 149
 150         /* Disable compression */
 151         dpfc_ctl = I915_READ(DPFC_CONTROL);
 152         if (dpfc_ctl & DPFC_CTL_EN) {
 153                 dpfc_ctl &= ~DPFC_CTL_EN;
 154                 I915_WRITE(DPFC_CONTROL, dpfc_ctl);
 155
 156                 DRM_DEBUG_KMS("disabled FBC\n");
 157         }
 158 }
 159
 160 static bool g4x_fbc_enabled(struct drm_device *dev)
 161 {
 162         struct drm_i915_private *dev_priv = dev->dev_private;
 163
 164         return I915_READ(DPFC_CONTROL) & DPFC_CTL_EN;
 165 }
 166
 167 static void sandybridge_blit_fbc_update(struct drm_device *dev)
 168 {
 169         struct drm_i915_private *dev_priv = dev->dev_private;
 170         u32 blt_ecoskpd;
 171
 172         /* Make sure blitter notifies FBC of writes */
 173         gen6_gt_force_wake_get(dev_priv);
 174         blt_ecoskpd = I915_READ(GEN6_BLITTER_ECOSKPD);
 175         blt_ecoskpd |= GEN6_BLITTER_FBC_NOTIFY <<
 176                 GEN6_BLITTER_LOCK_SHIFT;
 177         I915_WRITE(GEN6_BLITTER_ECOSKPD, blt_ecoskpd);
 178         blt_ecoskpd |= GEN6_BLITTER_FBC_NOTIFY;
 179         I915_WRITE(GEN6_BLITTER_ECOSKPD, blt_ecoskpd);
 180         blt_ecoskpd &= ~(GEN6_BLITTER_FBC_NOTIFY <<
 181                          GEN6_BLITTER_LOCK_SHIFT);
 182         I915_WRITE(GEN6_BLITTER_ECOSKPD, blt_ecoskpd);
 183         POSTING_READ(GEN6_BLITTER_ECOSKPD);
 184         gen6_gt_force_wake_put(dev_priv);
 185 }
 186
 187 static void ironlake_enable_fbc(struct drm_crtc *crtc, unsigned long interval)
 188 {
 189         struct drm_device *dev = crtc->dev;
 190         struct drm_i915_private *dev_priv = dev->dev_private;
 191         struct drm_framebuffer *fb = crtc->fb;
 192         struct intel_framebuffer *intel_fb = to_intel_framebuffer(fb);
 193         struct drm_i915_gem_object *obj = intel_fb->obj;
 194         struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
 195         int plane = intel_crtc->plane == 0 ? DPFC_CTL_PLANEA : DPFC_CTL_PLANEB;
 196         unsigned long stall_watermark = 200;
 197         u32 dpfc_ctl;
 198
 199         dpfc_ctl = I915_READ(ILK_DPFC_CONTROL);
 200         dpfc_ctl &= DPFC_RESERVED;
 201         dpfc_ctl |= (plane | DPFC_CTL_LIMIT_1X);
 202         /* Set persistent mode for front-buffer rendering, ala X. */
 203         dpfc_ctl |= DPFC_CTL_PERSISTENT_MODE;
 204         dpfc_ctl |= (DPFC_CTL_FENCE_EN | obj->fence_reg);
 205         I915_WRITE(ILK_DPFC_CHICKEN, DPFC_HT_MODIFY);
 206
 207         I915_WRITE(ILK_DPFC_RECOMP_CTL, DPFC_RECOMP_STALL_EN |
 208                    (stall_watermark << DPFC_RECOMP_STALL_WM_SHIFT) |
 209                    (interval << DPFC_RECOMP_TIMER_COUNT_SHIFT));
 210         I915_WRITE(ILK_DPFC_FENCE_YOFF, crtc->y);
 211         I915_WRITE(ILK_FBC_RT_BASE, i915_gem_obj_ggtt_offset(obj) | ILK_FBC_RT_VALID);
 212         /* enable it... */
 213         I915_WRITE(ILK_DPFC_CONTROL, dpfc_ctl | DPFC_CTL_EN);
 214
 215         if (IS_GEN6(dev)) {
 216                 I915_WRITE(SNB_DPFC_CTL_SA,
 217                            SNB_CPU_FENCE_ENABLE | obj->fence_reg);
 218                 I915_WRITE(DPFC_CPU_FENCE_OFFSET, crtc->y);
 219                 sandybridge_blit_fbc_update(dev);
 220         }
 221
 222         DRM_DEBUG_KMS("enabled fbc on plane %c\n", plane_name(intel_crtc->plane));
 223 }
 224
 225 static void ironlake_disable_fbc(struct drm_device *dev)
 226 {
 227         struct drm_i915_private *dev_priv = dev->dev_private;
 228         u32 dpfc_ctl;
 229
 230         /* Disable compression */
 231         dpfc_ctl = I915_READ(ILK_DPFC_CONTROL);
 232         if (dpfc_ctl & DPFC_CTL_EN) {
 233                 dpfc_ctl &= ~DPFC_CTL_EN;
 234                 I915_WRITE(ILK_DPFC_CONTROL, dpfc_ctl);
 235
 236                 if (IS_IVYBRIDGE(dev))
 237                         /* WaFbcDisableDpfcClockGating:ivb */
 238                         I915_WRITE(ILK_DSPCLK_GATE_D,
 239                                    I915_READ(ILK_DSPCLK_GATE_D) &
 240                                    ~ILK_DPFCUNIT_CLOCK_GATE_DISABLE);
 241
 242                 if (IS_HASWELL(dev))
 243                         /* WaFbcDisableDpfcClockGating:hsw */
 244                         I915_WRITE(HSW_CLKGATE_DISABLE_PART_1,
 245                                    I915_READ(HSW_CLKGATE_DISABLE_PART_1) &
 246                                    ~HSW_DPFC_GATING_DISABLE);
 247
 248                 DRM_DEBUG_KMS("disabled FBC\n");
 249         }
 250 }
 251
 252 static bool ironlake_fbc_enabled(struct drm_device *dev)
 253 {
 254         struct drm_i915_private *dev_priv = dev->dev_private;
 255
 256         return I915_READ(ILK_DPFC_CONTROL) & DPFC_CTL_EN;
 257 }
 258
 259 static void gen7_enable_fbc(struct drm_crtc *crtc, unsigned long interval)
 260 {
 261         struct drm_device *dev = crtc->dev;
 262         struct drm_i915_private *dev_priv = dev->dev_private;
 263         struct drm_framebuffer *fb = crtc->fb;
 264         struct intel_framebuffer *intel_fb = to_intel_framebuffer(fb);
 265         struct drm_i915_gem_object *obj = intel_fb->obj;
 266         struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
 267
 268         I915_WRITE(IVB_FBC_RT_BASE, i915_gem_obj_ggtt_offset(obj));
 269
 270         I915_WRITE(ILK_DPFC_CONTROL, DPFC_CTL_EN | DPFC_CTL_LIMIT_1X |
 271                    IVB_DPFC_CTL_FENCE_EN |
 272                    intel_crtc->plane << IVB_DPFC_CTL_PLANE_SHIFT);
 273
 274         if (IS_IVYBRIDGE(dev)) {
 275                 /* WaFbcAsynchFlipDisableFbcQueue:ivb */
 276                 I915_WRITE(ILK_DISPLAY_CHICKEN1, ILK_FBCQ_DIS);
 277                 /* WaFbcDisableDpfcClockGating:ivb */
 278                 I915_WRITE(ILK_DSPCLK_GATE_D,
 279                            I915_READ(ILK_DSPCLK_GATE_D) |
 280                            ILK_DPFCUNIT_CLOCK_GATE_DISABLE);
 281         } else {
 282                 /* WaFbcAsynchFlipDisableFbcQueue:hsw */
 283                 I915_WRITE(HSW_PIPE_SLICE_CHICKEN_1(intel_crtc->pipe),
 284                            HSW_BYPASS_FBC_QUEUE);
 285                 /* WaFbcDisableDpfcClockGating:hsw */
 286                 I915_WRITE(HSW_CLKGATE_DISABLE_PART_1,
 287                            I915_READ(HSW_CLKGATE_DISABLE_PART_1) |
 288                            HSW_DPFC_GATING_DISABLE);
 289         }
 290
 291         I915_WRITE(SNB_DPFC_CTL_SA,
 292                    SNB_CPU_FENCE_ENABLE | obj->fence_reg);
 293         I915_WRITE(DPFC_CPU_FENCE_OFFSET, crtc->y);
 294
 295         sandybridge_blit_fbc_update(dev);
 296
 297         DRM_DEBUG_KMS("enabled fbc on plane %d\n", intel_crtc->plane);
 298 }
 299
 300 bool intel_fbc_enabled(struct drm_device *dev)
 301 {
 302         struct drm_i915_private *dev_priv = dev->dev_private;
 303
 304         if (!dev_priv->display.fbc_enabled)
 305                 return false;
 306
 307         return dev_priv->display.fbc_enabled(dev);
 308 }
 309
 310 static void intel_fbc_work_fn(struct work_struct *__work)
 311 {
 312         struct intel_fbc_work *work =
 313                 container_of(to_delayed_work(__work),
 314                              struct intel_fbc_work, work);
 315         struct drm_device *dev = work->crtc->dev;
 316         struct drm_i915_private *dev_priv = dev->dev_private;
 317
 318         mutex_lock(&dev->struct_mutex);
 319         if (work == dev_priv->fbc.fbc_work) {
 320                 /* Double check that we haven't switched fb without cancelling
 321                  * the prior work.
 322                  */
 323                 if (work->crtc->fb == work->fb) {
 324                         dev_priv->display.enable_fbc(work->crtc,
 325                                                      work->interval);
 326
 327                         dev_priv->fbc.plane = to_intel_crtc(work->crtc)->plane;
 328                         dev_priv->fbc.fb_id = work->crtc->fb->base.id;
 329                         dev_priv->fbc.y = work->crtc->y;
 330                 }
 331
 332                 dev_priv->fbc.fbc_work = NULL;
 333         }
 334         mutex_unlock(&dev->struct_mutex);
 335
 336         kfree(work);
 337 }
 338
 339 static void intel_cancel_fbc_work(struct drm_i915_private *dev_priv)
 340 {
 341         if (dev_priv->fbc.fbc_work == NULL)
 342                 return;
 343
 344         DRM_DEBUG_KMS("cancelling pending FBC enable\n");
 345
 346         /* Synchronisation is provided by struct_mutex and checking of
 347          * dev_priv->fbc.fbc_work, so we can perform the cancellation
 348          * entirely asynchronously.
 349          */
 350         if (cancel_delayed_work(&dev_priv->fbc.fbc_work->work))
 351                 /* tasklet was killed before being run, clean up */
 352                 kfree(dev_priv->fbc.fbc_work);
 353
 354         /* Mark the work as no longer wanted so that if it does
 355          * wake-up (because the work was already running and waiting
 356          * for our mutex), it will discover that is no longer
 357          * necessary to run.
 358          */
 359         dev_priv->fbc.fbc_work = NULL;
 360 }
 361
 362 static void intel_enable_fbc(struct drm_crtc *crtc, unsigned long interval)
 363 {
 364         struct intel_fbc_work *work;
 365         struct drm_device *dev = crtc->dev;
 366         struct drm_i915_private *dev_priv = dev->dev_private;
 367
 368         if (!dev_priv->display.enable_fbc)
 369                 return;
 370
 371         intel_cancel_fbc_work(dev_priv);
 372
 373         work = kzalloc(sizeof(*work), GFP_KERNEL);
 374         if (work == NULL) {
 375                 DRM_ERROR("Failed to allocate FBC work structure\n");
 376                 dev_priv->display.enable_fbc(crtc, interval);
 377                 return;
 378         }
 379
 380         work->crtc = crtc;
 381         work->fb = crtc->fb;
 382         work->interval = interval;
 383         INIT_DELAYED_WORK(&work->work, intel_fbc_work_fn);
 384
 385         dev_priv->fbc.fbc_work = work;
 386
 387         /* Delay the actual enabling to let pageflipping cease and the
 388          * display to settle before starting the compression. Note that
 389          * this delay also serves a second purpose: it allows for a
 390          * vblank to pass after disabling the FBC before we attempt
 391          * to modify the control registers.
 392          *
 393          * A more complicated solution would involve tracking vblanks
 394          * following the termination of the page-flipping sequence
 395          * and indeed performing the enable as a co-routine and not
 396          * waiting synchronously upon the vblank.
 397          *
 398          * WaFbcWaitForVBlankBeforeEnable:ilk,snb
 399          */
 400         schedule_delayed_work(&work->work, msecs_to_jiffies(50));
 401 }
 402
 403 void intel_disable_fbc(struct drm_device *dev)
 404 {
 405         struct drm_i915_private *dev_priv = dev->dev_private;
 406
 407         intel_cancel_fbc_work(dev_priv);
 408
 409         if (!dev_priv->display.disable_fbc)
 410                 return;
 411
 412         dev_priv->display.disable_fbc(dev);
 413         dev_priv->fbc.plane = -1;
 414 }
 415
 416 static bool set_no_fbc_reason(struct drm_i915_private *dev_priv,
 417                               enum no_fbc_reason reason)
 418 {
 419         if (dev_priv->fbc.no_fbc_reason == reason)
 420                 return false;
 421
 422         dev_priv->fbc.no_fbc_reason = reason;
 423         return true;
 424 }
 425
 426 /**
 427  * intel_update_fbc - enable/disable FBC as needed
 428  * @dev: the drm_device
 429  *
 430  * Set up the framebuffer compression hardware at mode set time.  We
 431  * enable it if possible:
 432  *   - plane A only (on pre-965)
 433  *   - no pixel mulitply/line duplication
 434  *   - no alpha buffer discard
 435  *   - no dual wide
 436  *   - framebuffer <= max_hdisplay in width, max_vdisplay in height
 437  *
 438  * We can't assume that any compression will take place (worst case),
 439  * so the compressed buffer has to be the same size as the uncompressed
 440  * one.  It also must reside (along with the line length buffer) in
 441  * stolen memory.
 442  *
 443  * We need to enable/disable FBC on a global basis.
 444  */
 445 void intel_update_fbc(struct drm_device *dev)
 446 {
 447         struct drm_i915_private *dev_priv = dev->dev_private;
 448         struct drm_crtc *crtc = NULL, *tmp_crtc;
 449         struct intel_crtc *intel_crtc;
 450         struct drm_framebuffer *fb;
 451         struct intel_framebuffer *intel_fb;
 452         struct drm_i915_gem_object *obj;
 453         const struct drm_display_mode *adjusted_mode;
 454         unsigned int max_width, max_height;
 455
 456         if (!I915_HAS_FBC(dev)) {
 457                 set_no_fbc_reason(dev_priv, FBC_UNSUPPORTED);
 458                 return;
 459         }
 460
 461         if (!i915_powersave) {
 462                 if (set_no_fbc_reason(dev_priv, FBC_MODULE_PARAM))
 463                         DRM_DEBUG_KMS("fbc disabled per module param\n");
 464                 return;
 465         }
 466
 467         /*
 468          * If FBC is already on, we just have to verify that we can
 469          * keep it that way...
 470          * Need to disable if:
 471          *   - more than one pipe is active
 472          *   - changing FBC params (stride, fence, mode)
 473          *   - new fb is too large to fit in compressed buffer
 474          *   - going to an unsupported config (interlace, pixel multiply, etc.)
 475          */
 476         list_for_each_entry(tmp_crtc, &dev->mode_config.crtc_list, head) {
 477                 if (intel_crtc_active(tmp_crtc) &&
 478                     to_intel_crtc(tmp_crtc)->primary_enabled) {
 479                         if (crtc) {
 480                                 if (set_no_fbc_reason(dev_priv, FBC_MULTIPLE_PIPES))
 481                                         DRM_DEBUG_KMS("more than one pipe active, disabling compression\n");
 482                                 goto out_disable;
 483                         }
 484                         crtc = tmp_crtc;
 485                 }
 486         }
 487
 488         if (!crtc || crtc->fb == NULL) {
 489                 if (set_no_fbc_reason(dev_priv, FBC_NO_OUTPUT))
 490                         DRM_DEBUG_KMS("no output, disabling\n");
 491                 goto out_disable;
 492         }
 493
 494         intel_crtc = to_intel_crtc(crtc);
 495         fb = crtc->fb;
 496         intel_fb = to_intel_framebuffer(fb);
 497         obj = intel_fb->obj;
 498         adjusted_mode = &intel_crtc->config.adjusted_mode;
 499
 500         if (i915_enable_fbc < 0 &&
 501             INTEL_INFO(dev)->gen <= 7 && !IS_HASWELL(dev)) {
 502                 if (set_no_fbc_reason(dev_priv, FBC_CHIP_DEFAULT))
 503                         DRM_DEBUG_KMS("disabled per chip default\n");
 504                 goto out_disable;
 505         }
 506         if (!i915_enable_fbc) {
 507                 if (set_no_fbc_reason(dev_priv, FBC_MODULE_PARAM))
 508                         DRM_DEBUG_KMS("fbc disabled per module param\n");
 509                 goto out_disable;
 510         }
 511         if ((adjusted_mode->flags & DRM_MODE_FLAG_INTERLACE) ||
 512             (adjusted_mode->flags & DRM_MODE_FLAG_DBLSCAN)) {
 513                 if (set_no_fbc_reason(dev_priv, FBC_UNSUPPORTED_MODE))
 514                         DRM_DEBUG_KMS("mode incompatible with compression, "
 515                                       "disabling\n");
 516                 goto out_disable;
 517         }
 518
 519         if (IS_G4X(dev) || INTEL_INFO(dev)->gen >= 5) {
 520                 max_width = 4096;
 521                 max_height = 2048;
 522         } else {
 523                 max_width = 2048;
 524                 max_height = 1536;
 525         }
 526         if (intel_crtc->config.pipe_src_w > max_width ||
 527             intel_crtc->config.pipe_src_h > max_height) {
 528                 if (set_no_fbc_reason(dev_priv, FBC_MODE_TOO_LARGE))
 529                         DRM_DEBUG_KMS("mode too large for compression, disabling\n");
 530                 goto out_disable;
 531         }
 532         if ((IS_I915GM(dev) || IS_I945GM(dev) || IS_HASWELL(dev)) &&
 533             intel_crtc->plane != 0) {
 534                 if (set_no_fbc_reason(dev_priv, FBC_BAD_PLANE))
 535                         DRM_DEBUG_KMS("plane not 0, disabling compression\n");
 536                 goto out_disable;
 537         }
 538
 539         /* The use of a CPU fence is mandatory in order to detect writes
 540          * by the CPU to the scanout and trigger updates to the FBC.
 541          */
 542         if (obj->tiling_mode != I915_TILING_X ||
 543             obj->fence_reg == I915_FENCE_REG_NONE) {
 544                 if (set_no_fbc_reason(dev_priv, FBC_NOT_TILED))
 545                         DRM_DEBUG_KMS("framebuffer not tiled or fenced, disabling compression\n");
 546                 goto out_disable;
 547         }
 548
 549         /* If the kernel debugger is active, always disable compression */
 550         if (in_dbg_master())
 551                 goto out_disable;
 552
 553         if (i915_gem_stolen_setup_compression(dev, intel_fb->obj->base.size)) {
 554                 if (set_no_fbc_reason(dev_priv, FBC_STOLEN_TOO_SMALL))
 555                         DRM_DEBUG_KMS("framebuffer too large, disabling compression\n");
 556                 goto out_disable;
 557         }
 558
 559         /* If the scanout has not changed, don't modify the FBC settings.
 560          * Note that we make the fundamental assumption that the fb->obj
 561          * cannot be unpinned (and have its GTT offset and fence revoked)
 562          * without first being decoupled from the scanout and FBC disabled.
 563          */
 564         if (dev_priv->fbc.plane == intel_crtc->plane &&
 565             dev_priv->fbc.fb_id == fb->base.id &&
 566             dev_priv->fbc.y == crtc->y)
 567                 return;
 568
 569         if (intel_fbc_enabled(dev)) {
 570                 /* We update FBC along two paths, after changing fb/crtc
 571                  * configuration (modeswitching) and after page-flipping
 572                  * finishes. For the latter, we know that not only did
 573                  * we disable the FBC at the start of the page-flip
 574                  * sequence, but also more than one vblank has passed.
 575                  *
 576                  * For the former case of modeswitching, it is possible
 577                  * to switch between two FBC valid configurations
 578                  * instantaneously so we do need to disable the FBC
 579                  * before we can modify its control registers. We also
 580                  * have to wait for the next vblank for that to take
 581                  * effect. However, since we delay enabling FBC we can
 582                  * assume that a vblank has passed since disabling and
 583                  * that we can safely alter the registers in the deferred
 584                  * callback.
 585                  *
 586                  * In the scenario that we go from a valid to invalid
 587                  * and then back to valid FBC configuration we have
 588                  * no strict enforcement that a vblank occurred since
 589                  * disabling the FBC. However, along all current pipe
 590                  * disabling paths we do need to wait for a vblank at
 591                  * some point. And we wait before enabling FBC anyway.
 592                  */
 593                 DRM_DEBUG_KMS("disabling active FBC for update\n");
 594                 intel_disable_fbc(dev);
 595         }
 596
 597         intel_enable_fbc(crtc, 500);
 598         dev_priv->fbc.no_fbc_reason = FBC_OK;
 599         return;
 600
 601 out_disable:
 602         /* Multiple disables should be harmless */
 603         if (intel_fbc_enabled(dev)) {
 604                 DRM_DEBUG_KMS("unsupported config, disabling FBC\n");
 605                 intel_disable_fbc(dev);
 606         }
 607         i915_gem_stolen_cleanup_compression(dev);
 608 }
 609
 610 static void i915_pineview_get_mem_freq(struct drm_device *dev)
 611 {
 612         drm_i915_private_t *dev_priv = dev->dev_private;
 613         u32 tmp;
 614
 615         tmp = I915_READ(CLKCFG);
 616
 617         switch (tmp & CLKCFG_FSB_MASK) {
 618         case CLKCFG_FSB_533:
 619                 dev_priv->fsb_freq = 533; /* 133*4 */
 620                 break;
 621         case CLKCFG_FSB_800:
 622                 dev_priv->fsb_freq = 800; /* 200*4 */
 623                 break;
 624         case CLKCFG_FSB_667:
 625                 dev_priv->fsb_freq =  667; /* 167*4 */
 626                 break;
 627         case CLKCFG_FSB_400:
 628                 dev_priv->fsb_freq = 400; /* 100*4 */
 629                 break;
 630         }
 631
 632         switch (tmp & CLKCFG_MEM_MASK) {
 633         case CLKCFG_MEM_533:
 634                 dev_priv->mem_freq = 533;
 635                 break;
 636         case CLKCFG_MEM_667:
 637                 dev_priv->mem_freq = 667;
 638                 break;
 639         case CLKCFG_MEM_800:
 640                 dev_priv->mem_freq = 800;
 641                 break;
 642         }
 643
 644         /* detect pineview DDR3 setting */
 645         tmp = I915_READ(CSHRDDR3CTL);
 646         dev_priv->is_ddr3 = (tmp & CSHRDDR3CTL_DDR3) ? 1 : 0;
 647 }
 648
 649 static void i915_ironlake_get_mem_freq(struct drm_device *dev)
 650 {
 651         drm_i915_private_t *dev_priv = dev->dev_private;
 652         u16 ddrpll, csipll;
 653
 654         ddrpll = I915_READ16(DDRMPLL1);
 655         csipll = I915_READ16(CSIPLL0);
 656
 657         switch (ddrpll & 0xff) {
 658         case 0xc:
 659                 dev_priv->mem_freq = 800;
 660                 break;
 661         case 0x10:
 662                 dev_priv->mem_freq = 1066;
 663                 break;
 664         case 0x14:
 665                 dev_priv->mem_freq = 1333;
 666                 break;
 667         case 0x18:
 668                 dev_priv->mem_freq = 1600;
 669                 break;
 670         default:
 671                 DRM_DEBUG_DRIVER("unknown memory frequency 0x%02x\n",
 672                                  ddrpll & 0xff);
 673                 dev_priv->mem_freq = 0;
 674                 break;
 675         }
 676
 677         dev_priv->ips.r_t = dev_priv->mem_freq;
 678
 679         switch (csipll & 0x3ff) {
 680         case 0x00c:
 681                 dev_priv->fsb_freq = 3200;
 682                 break;
 683         case 0x00e:
 684                 dev_priv->fsb_freq = 3733;
 685                 break;
 686         case 0x010:
 687                 dev_priv->fsb_freq = 4266;
 688                 break;
 689         case 0x012:
 690                 dev_priv->fsb_freq = 4800;
 691                 break;
 692         case 0x014:
 693                 dev_priv->fsb_freq = 5333;
 694                 break;
 695         case 0x016:
 696                 dev_priv->fsb_freq = 5866;
 697                 break;
 698         case 0x018:
 699                 dev_priv->fsb_freq = 6400;
 700                 break;
 701         default:
 702                 DRM_DEBUG_DRIVER("unknown fsb frequency 0x%04x\n",
 703                                  csipll & 0x3ff);
 704                 dev_priv->fsb_freq = 0;
 705                 break;
 706         }
 707
 708         if (dev_priv->fsb_freq == 3200) {
 709                 dev_priv->ips.c_m = 0;
 710         } else if (dev_priv->fsb_freq > 3200 && dev_priv->fsb_freq <= 4800) {
 711                 dev_priv->ips.c_m = 1;
 712         } else {
 713                 dev_priv->ips.c_m = 2;
 714         }
 715 }
 716
 717 static const struct cxsr_latency cxsr_latency_table[] = {
 718         {1, 0, 800, 400, 3382, 33382, 3983, 33983},    /* DDR2-400 SC */
 719         {1, 0, 800, 667, 3354, 33354, 3807, 33807},    /* DDR2-667 SC */
 720         {1, 0, 800, 800, 3347, 33347, 3763, 33763},    /* DDR2-800 SC */
 721         {1, 1, 800, 667, 6420, 36420, 6873, 36873},    /* DDR3-667 SC */
 722         {1, 1, 800, 800, 5902, 35902, 6318, 36318},    /* DDR3-800 SC */
 723
 724         {1, 0, 667, 400, 3400, 33400, 4021, 34021},    /* DDR2-400 SC */
 725         {1, 0, 667, 667, 3372, 33372, 3845, 33845},    /* DDR2-667 SC */
 726         {1, 0, 667, 800, 3386, 33386, 3822, 33822},    /* DDR2-800 SC */
 727         {1, 1, 667, 667, 6438, 36438, 6911, 36911},    /* DDR3-667 SC */
 728         {1, 1, 667, 800, 5941, 35941, 6377, 36377},    /* DDR3-800 SC */
 729
 730         {1, 0, 400, 400, 3472, 33472, 4173, 34173},    /* DDR2-400 SC */
 731         {1, 0, 400, 667, 3443, 33443, 3996, 33996},    /* DDR2-667 SC */
 732         {1, 0, 400, 800, 3430, 33430, 3946, 33946},    /* DDR2-800 SC */
 733         {1, 1, 400, 667, 6509, 36509, 7062, 37062},    /* DDR3-667 SC */
 734         {1, 1, 400, 800, 5985, 35985, 6501, 36501},    /* DDR3-800 SC */
 735
 736         {0, 0, 800, 400, 3438, 33438, 4065, 34065},    /* DDR2-400 SC */
 737         {0, 0, 800, 667, 3410, 33410, 3889, 33889},    /* DDR2-667 SC */
 738         {0, 0, 800, 800, 3403, 33403, 3845, 33845},    /* DDR2-800 SC */
 739         {0, 1, 800, 667, 6476, 36476, 6955, 36955},    /* DDR3-667 SC */
 740         {0, 1, 800, 800, 5958, 35958, 6400, 36400},    /* DDR3-800 SC */
 741
 742         {0, 0, 667, 400, 3456, 33456, 4103, 34106},    /* DDR2-400 SC */
 743         {0, 0, 667, 667, 3428, 33428, 3927, 33927},    /* DDR2-667 SC */
 744         {0, 0, 667, 800, 3443, 33443, 3905, 33905},    /* DDR2-800 SC */
 745         {0, 1, 667, 667, 6494, 36494, 6993, 36993},    /* DDR3-667 SC */
 746         {0, 1, 667, 800, 5998, 35998, 6460, 36460},    /* DDR3-800 SC */
 747
 748         {0, 0, 400, 400, 3528, 33528, 4255, 34255},    /* DDR2-400 SC */
 749         {0, 0, 400, 667, 3500, 33500, 4079, 34079},    /* DDR2-667 SC */
 750         {0, 0, 400, 800, 3487, 33487, 4029, 34029},    /* DDR2-800 SC */
 751         {0, 1, 400, 667, 6566, 36566, 7145, 37145},    /* DDR3-667 SC */
 752         {0, 1, 400, 800, 6042, 36042, 6584, 36584},    /* DDR3-800 SC */
 753 };
 754
 755 static const struct cxsr_latency *intel_get_cxsr_latency(int is_desktop,
 756                                                          int is_ddr3,
 757                                                          int fsb,
 758                                                          int mem)
 759 {
 760         const struct cxsr_latency *latency;
 761         int i;
 762
 763         if (fsb == 0 || mem == 0)
 764                 return NULL;
 765
 766         for (i = 0; i < ARRAY_SIZE(cxsr_latency_table); i++) {
 767                 latency = &cxsr_latency_table[i];
 768                 if (is_desktop == latency->is_desktop &&
 769                     is_ddr3 == latency->is_ddr3 &&
 770                     fsb == latency->fsb_freq && mem == latency->mem_freq)
 771                         return latency;
 772         }
 773
 774         DRM_DEBUG_KMS("Unknown FSB/MEM found, disable CxSR\n");
 775
 776         return NULL;
 777 }
 778
 779 static void pineview_disable_cxsr(struct drm_device *dev)
 780 {
 781         struct drm_i915_private *dev_priv = dev->dev_private;
 782
 783         /* deactivate cxsr */
 784         I915_WRITE(DSPFW3, I915_READ(DSPFW3) & ~PINEVIEW_SELF_REFRESH_EN);
 785 }
 786
 787 /*
 788  * Latency for FIFO fetches is dependent on several factors:
 789  *   - memory configuration (speed, channels)
 790  *   - chipset
 791  *   - current MCH state
 792  * It can be fairly high in some situations, so here we assume a fairly
 793  * pessimal value.  It's a tradeoff between extra memory fetches (if we
 794  * set this value too high, the FIFO will fetch frequently to stay full)
 795  * and power consumption (set it too low to save power and we might see
 796  * FIFO underruns and display "flicker").
 797  *
 798  * A value of 5us seems to be a good balance; safe for very low end
 799  * platforms but not overly aggressive on lower latency configs.
 800  */
 801 static const int latency_ns = 5000;
 802
 803 static int i9xx_get_fifo_size(struct drm_device *dev, int plane)
 804 {
 805         struct drm_i915_private *dev_priv = dev->dev_private;
 806         uint32_t dsparb = I915_READ(DSPARB);
 807         int size;
 808
 809         size = dsparb & 0x7f;
 810         if (plane)
 811                 size = ((dsparb >> DSPARB_CSTART_SHIFT) & 0x7f) - size;
 812
 813         DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb,
 814                       plane ? "B" : "A", size);
 815
 816         return size;
 817 }
 818
 819 static int i85x_get_fifo_size(struct drm_device *dev, int plane)
 820 {
 821         struct drm_i915_private *dev_priv = dev->dev_private;
 822         uint32_t dsparb = I915_READ(DSPARB);
 823         int size;
 824
 825         size = dsparb & 0x1ff;
 826         if (plane)
 827                 size = ((dsparb >> DSPARB_BEND_SHIFT) & 0x1ff) - size;
 828         size >>= 1; /* Convert to cachelines */
 829
 830         DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb,
 831                       plane ? "B" : "A", size);
 832
 833         return size;
 834 }
 835
 836 static int i845_get_fifo_size(struct drm_device *dev, int plane)
 837 {
 838         struct drm_i915_private *dev_priv = dev->dev_private;
 839         uint32_t dsparb = I915_READ(DSPARB);
 840         int size;
 841
 842         size = dsparb & 0x7f;
 843         size >>= 2; /* Convert to cachelines */
 844
 845         DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb,
 846                       plane ? "B" : "A",
 847                       size);
 848
 849         return size;
 850 }
 851
 852 static int i830_get_fifo_size(struct drm_device *dev, int plane)
 853 {
 854         struct drm_i915_private *dev_priv = dev->dev_private;
 855         uint32_t dsparb = I915_READ(DSPARB);
 856         int size;
 857
 858         size = dsparb & 0x7f;
 859         size >>= 1; /* Convert to cachelines */
 860
 861         DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb,
 862                       plane ? "B" : "A", size);
 863
 864         return size;
 865 }
 866
 867 /* Pineview has different values for various configs */
 868 static const struct intel_watermark_params pineview_display_wm = {
 869         PINEVIEW_DISPLAY_FIFO,
 870         PINEVIEW_MAX_WM,
 871         PINEVIEW_DFT_WM,
 872         PINEVIEW_GUARD_WM,
 873         PINEVIEW_FIFO_LINE_SIZE
 874 };
 875 static const struct intel_watermark_params pineview_display_hplloff_wm = {
 876         PINEVIEW_DISPLAY_FIFO,
 877         PINEVIEW_MAX_WM,
 878         PINEVIEW_DFT_HPLLOFF_WM,
 879         PINEVIEW_GUARD_WM,
 880         PINEVIEW_FIFO_LINE_SIZE
 881 };
 882 static const struct intel_watermark_params pineview_cursor_wm = {
 883         PINEVIEW_CURSOR_FIFO,
 884         PINEVIEW_CURSOR_MAX_WM,
 885         PINEVIEW_CURSOR_DFT_WM,
 886         PINEVIEW_CURSOR_GUARD_WM,
 887         PINEVIEW_FIFO_LINE_SIZE,
 888 };
 889 static const struct intel_watermark_params pineview_cursor_hplloff_wm = {
 890         PINEVIEW_CURSOR_FIFO,
 891         PINEVIEW_CURSOR_MAX_WM,
 892         PINEVIEW_CURSOR_DFT_WM,
 893         PINEVIEW_CURSOR_GUARD_WM,
 894         PINEVIEW_FIFO_LINE_SIZE
 895 };
 896 static const struct intel_watermark_params g4x_wm_info = {
 897         G4X_FIFO_SIZE,
 898         G4X_MAX_WM,
 899         G4X_MAX_WM,
 900         2,
 901         G4X_FIFO_LINE_SIZE,
 902 };
 903 static const struct intel_watermark_params g4x_cursor_wm_info = {
 904         I965_CURSOR_FIFO,
 905         I965_CURSOR_MAX_WM,
 906         I965_CURSOR_DFT_WM,
 907         2,
 908         G4X_FIFO_LINE_SIZE,
 909 };
 910 static const struct intel_watermark_params valleyview_wm_info = {
 911         VALLEYVIEW_FIFO_SIZE,
 912         VALLEYVIEW_MAX_WM,
 913         VALLEYVIEW_MAX_WM,
 914         2,
 915         G4X_FIFO_LINE_SIZE,
 916 };
 917 static const struct intel_watermark_params valleyview_cursor_wm_info = {
 918         I965_CURSOR_FIFO,
 919         VALLEYVIEW_CURSOR_MAX_WM,
 920         I965_CURSOR_DFT_WM,
 921         2,
 922         G4X_FIFO_LINE_SIZE,
 923 };
 924 static const struct intel_watermark_params i965_cursor_wm_info = {
 925         I965_CURSOR_FIFO,
 926         I965_CURSOR_MAX_WM,
 927         I965_CURSOR_DFT_WM,
 928         2,
 929         I915_FIFO_LINE_SIZE,
 930 };
 931 static const struct intel_watermark_params i945_wm_info = {
 932         I945_FIFO_SIZE,
 933         I915_MAX_WM,
 934         1,
 935         2,
 936         I915_FIFO_LINE_SIZE
 937 };
 938 static const struct intel_watermark_params i915_wm_info = {
 939         I915_FIFO_SIZE,
 940         I915_MAX_WM,
 941         1,
 942         2,
 943         I915_FIFO_LINE_SIZE
 944 };
 945 static const struct intel_watermark_params i855_wm_info = {
 946         I855GM_FIFO_SIZE,
 947         I915_MAX_WM,
 948         1,
 949         2,
 950         I830_FIFO_LINE_SIZE
 951 };
 952 static const struct intel_watermark_params i830_wm_info = {
 953         I830_FIFO_SIZE,
 954         I915_MAX_WM,
 955         1,
 956         2,
 957         I830_FIFO_LINE_SIZE
 958 };
 959
 960 static const struct intel_watermark_params ironlake_display_wm_info = {
 961         ILK_DISPLAY_FIFO,
 962         ILK_DISPLAY_MAXWM,
 963         ILK_DISPLAY_DFTWM,
 964         2,
 965         ILK_FIFO_LINE_SIZE
 966 };
 967 static const struct intel_watermark_params ironlake_cursor_wm_info = {
 968         ILK_CURSOR_FIFO,
 969         ILK_CURSOR_MAXWM,
 970         ILK_CURSOR_DFTWM,
 971         2,
 972         ILK_FIFO_LINE_SIZE
 973 };
 974 static const struct intel_watermark_params ironlake_display_srwm_info = {
 975         ILK_DISPLAY_SR_FIFO,
 976         ILK_DISPLAY_MAX_SRWM,
 977         ILK_DISPLAY_DFT_SRWM,
 978         2,
 979         ILK_FIFO_LINE_SIZE
 980 };
 981 static const struct intel_watermark_params ironlake_cursor_srwm_info = {
 982         ILK_CURSOR_SR_FIFO,
 983         ILK_CURSOR_MAX_SRWM,
 984         ILK_CURSOR_DFT_SRWM,
 985         2,
 986         ILK_FIFO_LINE_SIZE
 987 };
 988
 989 static const struct intel_watermark_params sandybridge_display_wm_info = {
 990         SNB_DISPLAY_FIFO,
 991         SNB_DISPLAY_MAXWM,
 992         SNB_DISPLAY_DFTWM,
 993         2,
 994         SNB_FIFO_LINE_SIZE
 995 };
 996 static const struct intel_watermark_params sandybridge_cursor_wm_info = {
 997         SNB_CURSOR_FIFO,
 998         SNB_CURSOR_MAXWM,
 999         SNB_CURSOR_DFTWM,
1000         2,
1001         SNB_FIFO_LINE_SIZE
1002 };
1003 static const struct intel_watermark_params sandybridge_display_srwm_info = {
1004         SNB_DISPLAY_SR_FIFO,
1005         SNB_DISPLAY_MAX_SRWM,
1006         SNB_DISPLAY_DFT_SRWM,
1007         2,
1008         SNB_FIFO_LINE_SIZE
1009 };
1010 static const struct intel_watermark_params sandybridge_cursor_srwm_info = {
1011         SNB_CURSOR_SR_FIFO,
1012         SNB_CURSOR_MAX_SRWM,
1013         SNB_CURSOR_DFT_SRWM,
1014         2,
1015         SNB_FIFO_LINE_SIZE
1016 };
1017
1018
1019 /**
1020  * intel_calculate_wm - calculate watermark level
1021  * @clock_in_khz: pixel clock
1022  * @wm: chip FIFO params
1023  * @pixel_size: display pixel size
1024  * @latency_ns: memory latency for the platform
1025  *
1026  * Calculate the watermark level (the level at which the display plane will
1027  * start fetching from memory again).  Each chip has a different display
1028  * FIFO size and allocation, so the caller needs to figure that out and pass
1029  * in the correct intel_watermark_params structure.
1030  *
1031  * As the pixel clock runs, the FIFO will be drained at a rate that depends
1032  * on the pixel size.  When it reaches the watermark level, it'll start
1033  * fetching FIFO line sized based chunks from memory until the FIFO fills
1034  * past the watermark point.  If the FIFO drains completely, a FIFO underrun
1035  * will occur, and a display engine hang could result.
1036  */
1037 static unsigned long intel_calculate_wm(unsigned long clock_in_khz,
1038                                         const struct intel_watermark_params *wm,
1039                                         int fifo_size,
1040                                         int pixel_size,
1041                                         unsigned long latency_ns)
1042 {
1043         long entries_required, wm_size;
1044
1045         /*
1046          * Note: we need to make sure we don't overflow for various clock &
1047          * latency values.
1048          * clocks go from a few thousand to several hundred thousand.
1049          * latency is usually a few thousand
1050          */
1051         entries_required = ((clock_in_khz / 1000) * pixel_size * latency_ns) /
1052                 1000;
1053         entries_required = DIV_ROUND_UP(entries_required, wm->cacheline_size);
1054
1055         DRM_DEBUG_KMS("FIFO entries required for mode: %ld\n", entries_required);
1056
1057         wm_size = fifo_size - (entries_required + wm->guard_size);
1058
1059         DRM_DEBUG_KMS("FIFO watermark level: %ld\n", wm_size);
1060
1061         /* Don't promote wm_size to unsigned... */
1062         if (wm_size > (long)wm->max_wm)
1063                 wm_size = wm->max_wm;
1064         if (wm_size <= 0)
1065                 wm_size = wm->default_wm;
1066         return wm_size;
1067 }
1068
1069 static struct drm_crtc *single_enabled_crtc(struct drm_device *dev)
1070 {
1071         struct drm_crtc *crtc, *enabled = NULL;
1072
1073         list_for_each_entry(crtc, &dev->mode_config.crtc_list, head) {
1074                 if (intel_crtc_active(crtc)) {
1075                         if (enabled)
1076                                 return NULL;
1077                         enabled = crtc;
1078                 }
1079         }
1080
1081         return enabled;
1082 }
1083
1084 static void pineview_update_wm(struct drm_crtc *unused_crtc)
1085 {
1086         struct drm_device *dev = unused_crtc->dev;
1087         struct drm_i915_private *dev_priv = dev->dev_private;
1088         struct drm_crtc *crtc;
1089         const struct cxsr_latency *latency;
1090         u32 reg;
1091         unsigned long wm;
1092
1093         latency = intel_get_cxsr_latency(IS_PINEVIEW_G(dev), dev_priv->is_ddr3,
1094                                          dev_priv->fsb_freq, dev_priv->mem_freq);
1095         if (!latency) {
1096                 DRM_DEBUG_KMS("Unknown FSB/MEM found, disable CxSR\n");
1097                 pineview_disable_cxsr(dev);
1098                 return;
1099         }
1100
1101         crtc = single_enabled_crtc(dev);
1102         if (crtc) {
1103                 const struct drm_display_mode *adjusted_mode;
1104                 int pixel_size = crtc->fb->bits_per_pixel / 8;
1105                 int clock;
1106
1107                 adjusted_mode = &to_intel_crtc(crtc)->config.adjusted_mode;
1108                 clock = adjusted_mode->crtc_clock;
1109
1110                 /* Display SR */
1111                 wm = intel_calculate_wm(clock, &pineview_display_wm,
1112                                         pineview_display_wm.fifo_size,
1113                                         pixel_size, latency->display_sr);
1114                 reg = I915_READ(DSPFW1);
1115                 reg &= ~DSPFW_SR_MASK;
1116                 reg |= wm << DSPFW_SR_SHIFT;
1117                 I915_WRITE(DSPFW1, reg);
1118                 DRM_DEBUG_KMS("DSPFW1 register is %x\n", reg);
1119
1120                 /* cursor SR */
1121                 wm = intel_calculate_wm(clock, &pineview_cursor_wm,
1122                                         pineview_display_wm.fifo_size,
1123                                         pixel_size, latency->cursor_sr);
1124                 reg = I915_READ(DSPFW3);
1125                 reg &= ~DSPFW_CURSOR_SR_MASK;
1126                 reg |= (wm & 0x3f) << DSPFW_CURSOR_SR_SHIFT;
1127                 I915_WRITE(DSPFW3, reg);
1128
1129                 /* Display HPLL off SR */
1130                 wm = intel_calculate_wm(clock, &pineview_display_hplloff_wm,
1131                                         pineview_display_hplloff_wm.fifo_size,
1132                                         pixel_size, latency->display_hpll_disable);
1133                 reg = I915_READ(DSPFW3);
1134                 reg &= ~DSPFW_HPLL_SR_MASK;
1135                 reg |= wm & DSPFW_HPLL_SR_MASK;
1136                 I915_WRITE(DSPFW3, reg);
1137
1138                 /* cursor HPLL off SR */
1139                 wm = intel_calculate_wm(clock, &pineview_cursor_hplloff_wm,
1140                                         pineview_display_hplloff_wm.fifo_size,
1141                                         pixel_size, latency->cursor_hpll_disable);
1142                 reg = I915_READ(DSPFW3);
1143                 reg &= ~DSPFW_HPLL_CURSOR_MASK;
1144                 reg |= (wm & 0x3f) << DSPFW_HPLL_CURSOR_SHIFT;
1145                 I915_WRITE(DSPFW3, reg);
1146                 DRM_DEBUG_KMS("DSPFW3 register is %x\n", reg);
1147
1148                 /* activate cxsr */
1149                 I915_WRITE(DSPFW3,
1150                            I915_READ(DSPFW3) | PINEVIEW_SELF_REFRESH_EN);
1151                 DRM_DEBUG_KMS("Self-refresh is enabled\n");
1152         } else {
1153                 pineview_disable_cxsr(dev);
1154                 DRM_DEBUG_KMS("Self-refresh is disabled\n");
1155         }
1156 }
1157
1158 static bool g4x_compute_wm0(struct drm_device *dev,
1159                             int plane,
1160                             const struct intel_watermark_params *display,
1161                             int display_latency_ns,
1162                             const struct intel_watermark_params *cursor,
1163                             int cursor_latency_ns,
1164                             int *plane_wm,
1165                             int *cursor_wm)
1166 {
1167         struct drm_crtc *crtc;
1168         const struct drm_display_mode *adjusted_mode;
1169         int htotal, hdisplay, clock, pixel_size;
1170         int line_time_us, line_count;
1171         int entries, tlb_miss;
1172
1173         crtc = intel_get_crtc_for_plane(dev, plane);
1174         if (!intel_crtc_active(crtc)) {
1175                 *cursor_wm = cursor->guard_size;
1176                 *plane_wm = display->guard_size;
1177                 return false;
1178         }
1179
1180         adjusted_mode = &to_intel_crtc(crtc)->config.adjusted_mode;
1181         clock = adjusted_mode->crtc_clock;
1182         htotal = adjusted_mode->htotal;
1183         hdisplay = to_intel_crtc(crtc)->config.pipe_src_w;
1184         pixel_size = crtc->fb->bits_per_pixel / 8;
1185
1186         /* Use the small buffer method to calculate plane watermark */
1187         entries = ((clock * pixel_size / 1000) * display_latency_ns) / 1000;
1188         tlb_miss = display->fifo_size*display->cacheline_size - hdisplay * 8;
1189         if (tlb_miss > 0)
1190                 entries += tlb_miss;
1191         entries = DIV_ROUND_UP(entries, display->cacheline_size);
1192         *plane_wm = entries + display->guard_size;
1193         if (*plane_wm > (int)display->max_wm)
1194                 *plane_wm = display->max_wm;
1195
1196         /* Use the large buffer method to calculate cursor watermark */
1197         line_time_us = ((htotal * 1000) / clock);
1198         line_count = (cursor_latency_ns / line_time_us + 1000) / 1000;
1199         entries = line_count * 64 * pixel_size;
1200         tlb_miss = cursor->fifo_size*cursor->cacheline_size - hdisplay * 8;
1201         if (tlb_miss > 0)
1202                 entries += tlb_miss;
1203         entries = DIV_ROUND_UP(entries, cursor->cacheline_size);
1204         *cursor_wm = entries + cursor->guard_size;
1205         if (*cursor_wm > (int)cursor->max_wm)
1206                 *cursor_wm = (int)cursor->max_wm;
1207
1208         return true;
1209 }
1210
1211 /*
1212  * Check the wm result.
1213  *
1214  * If any calculated watermark values is larger than the maximum value that
1215  * can be programmed into the associated watermark register, that watermark
1216  * must be disabled.
1217  */
1218 static bool g4x_check_srwm(struct drm_device *dev,
1219                            int display_wm, int cursor_wm,
1220                            const struct intel_watermark_params *display,
1221                            const struct intel_watermark_params *cursor)
1222 {
1223         DRM_DEBUG_KMS("SR watermark: display plane %d, cursor %d\n",
1224                       display_wm, cursor_wm);
1225
1226         if (display_wm > display->max_wm) {
1227                 DRM_DEBUG_KMS("display watermark is too large(%d/%ld), disabling\n",
1228                               display_wm, display->max_wm);
1229                 return false;
1230         }
1231
1232         if (cursor_wm > cursor->max_wm) {
1233                 DRM_DEBUG_KMS("cursor watermark is too large(%d/%ld), disabling\n",
1234                               cursor_wm, cursor->max_wm);
1235                 return false;
1236         }
1237
1238         if (!(display_wm || cursor_wm)) {
1239                 DRM_DEBUG_KMS("SR latency is 0, disabling\n");
1240                 return false;
1241         }
1242
1243         return true;
1244 }
1245
1246 static bool g4x_compute_srwm(struct drm_device *dev,
1247                              int plane,
1248                              int latency_ns,
1249                              const struct intel_watermark_params *display,
1250                              const struct intel_watermark_params *cursor,
1251                              int *display_wm, int *cursor_wm)
1252 {
1253         struct drm_crtc *crtc;
1254         const struct drm_display_mode *adjusted_mode;
1255         int hdisplay, htotal, pixel_size, clock;
1256         unsigned long line_time_us;
1257         int line_count, line_size;
1258         int small, large;
1259         int entries;
1260
1261         if (!latency_ns) {
1262                 *display_wm = *cursor_wm = 0;
1263                 return false;
1264         }
1265
1266         crtc = intel_get_crtc_for_plane(dev, plane);
1267         adjusted_mode = &to_intel_crtc(crtc)->config.adjusted_mode;
1268         clock = adjusted_mode->crtc_clock;
1269         htotal = adjusted_mode->htotal;
1270         hdisplay = to_intel_crtc(crtc)->config.pipe_src_w;
1271         pixel_size = crtc->fb->bits_per_pixel / 8;
1272
1273         line_time_us = (htotal * 1000) / clock;
1274         line_count = (latency_ns / line_time_us + 1000) / 1000;
1275         line_size = hdisplay * pixel_size;
1276
1277         /* Use the minimum of the small and large buffer method for primary */
1278         small = ((clock * pixel_size / 1000) * latency_ns) / 1000;
1279         large = line_count * line_size;
1280
1281         entries = DIV_ROUND_UP(min(small, large), display->cacheline_size);
1282         *display_wm = entries + display->guard_size;
1283
1284         /* calculate the self-refresh watermark for display cursor */
1285         entries = line_count * pixel_size * 64;
1286         entries = DIV_ROUND_UP(entries, cursor->cacheline_size);
1287         *cursor_wm = entries + cursor->guard_size;
1288
1289         return g4x_check_srwm(dev,
1290                               *display_wm, *cursor_wm,
1291                               display, cursor);
1292 }
1293
1294 static bool vlv_compute_drain_latency(struct drm_device *dev,
1295                                      int plane,
1296                                      int *plane_prec_mult,
1297                                      int *plane_dl,
1298                                      int *cursor_prec_mult,
1299                                      int *cursor_dl)
1300 {
1301         struct drm_crtc *crtc;
1302         int clock, pixel_size;
1303         int entries;
1304
1305         crtc = intel_get_crtc_for_plane(dev, plane);
1306         if (!intel_crtc_active(crtc))
1307                 return false;
1308
1309         clock = to_intel_crtc(crtc)->config.adjusted_mode.crtc_clock;
1310         pixel_size = crtc->fb->bits_per_pixel / 8;      /* BPP */
1311
1312         entries = (clock / 1000) * pixel_size;
1313         *plane_prec_mult = (entries > 256) ?
1314                 DRAIN_LATENCY_PRECISION_32 : DRAIN_LATENCY_PRECISION_16;
1315         *plane_dl = (64 * (*plane_prec_mult) * 4) / ((clock / 1000) *
1316                                                      pixel_size);
1317
1318         entries = (clock / 1000) * 4;   /* BPP is always 4 for cursor */
1319         *cursor_prec_mult = (entries > 256) ?
1320                 DRAIN_LATENCY_PRECISION_32 : DRAIN_LATENCY_PRECISION_16;
1321         *cursor_dl = (64 * (*cursor_prec_mult) * 4) / ((clock / 1000) * 4);
1322
1323         return true;
1324 }
1325
1326 /*
1327  * Update drain latency registers of memory arbiter
1328  *
1329  * Valleyview SoC has a new memory arbiter and needs drain latency registers
1330  * to be programmed. Each plane has a drain latency multiplier and a drain
1331  * latency value.
1332  */
1333
1334 static void vlv_update_drain_latency(struct drm_device *dev)
1335 {
1336         struct drm_i915_private *dev_priv = dev->dev_private;
1337         int planea_prec, planea_dl, planeb_prec, planeb_dl;
1338         int cursora_prec, cursora_dl, cursorb_prec, cursorb_dl;
1339         int plane_prec_mult, cursor_prec_mult; /* Precision multiplier is
1340                                                         either 16 or 32 */
1341
1342         /* For plane A, Cursor A */
1343         if (vlv_compute_drain_latency(dev, 0, &plane_prec_mult, &planea_dl,
1344                                       &cursor_prec_mult, &cursora_dl)) {
1345                 cursora_prec = (cursor_prec_mult == DRAIN_LATENCY_PRECISION_32) ?
1346                         DDL_CURSORA_PRECISION_32 : DDL_CURSORA_PRECISION_16;
1347                 planea_prec = (plane_prec_mult == DRAIN_LATENCY_PRECISION_32) ?
1348                         DDL_PLANEA_PRECISION_32 : DDL_PLANEA_PRECISION_16;
1349
1350                 I915_WRITE(VLV_DDL1, cursora_prec |
1351                                 (cursora_dl << DDL_CURSORA_SHIFT) |
1352                                 planea_prec | planea_dl);
1353         }
1354
1355         /* For plane B, Cursor B */
1356         if (vlv_compute_drain_latency(dev, 1, &plane_prec_mult, &planeb_dl,
1357                                       &cursor_prec_mult, &cursorb_dl)) {
1358                 cursorb_prec = (cursor_prec_mult == DRAIN_LATENCY_PRECISION_32) ?
1359                         DDL_CURSORB_PRECISION_32 : DDL_CURSORB_PRECISION_16;
1360                 planeb_prec = (plane_prec_mult == DRAIN_LATENCY_PRECISION_32) ?
1361                         DDL_PLANEB_PRECISION_32 : DDL_PLANEB_PRECISION_16;
1362
1363                 I915_WRITE(VLV_DDL2, cursorb_prec |
1364                                 (cursorb_dl << DDL_CURSORB_SHIFT) |
1365                                 planeb_prec | planeb_dl);
1366         }
1367 }
1368
1369 #define single_plane_enabled(mask) is_power_of_2(mask)
1370
1371 static void valleyview_update_wm(struct drm_crtc *crtc)
1372 {
1373         struct drm_device *dev = crtc->dev;
1374         static const int sr_latency_ns = 12000;
1375         struct drm_i915_private *dev_priv = dev->dev_private;
1376         int planea_wm, planeb_wm, cursora_wm, cursorb_wm;
1377         int plane_sr, cursor_sr;
1378         int ignore_plane_sr, ignore_cursor_sr;
1379         unsigned int enabled = 0;
1380
1381         vlv_update_drain_latency(dev);
1382
1383         if (g4x_compute_wm0(dev, PIPE_A,
1384                             &valleyview_wm_info, latency_ns,
1385                             &valleyview_cursor_wm_info, latency_ns,
1386                             &planea_wm, &cursora_wm))
1387                 enabled |= 1 << PIPE_A;
1388
1389         if (g4x_compute_wm0(dev, PIPE_B,
1390                             &valleyview_wm_info, latency_ns,
1391                             &valleyview_cursor_wm_info, latency_ns,
1392                             &planeb_wm, &cursorb_wm))
1393                 enabled |= 1 << PIPE_B;
1394
1395         if (single_plane_enabled(enabled) &&
1396             g4x_compute_srwm(dev, ffs(enabled) - 1,
1397                              sr_latency_ns,
1398                              &valleyview_wm_info,
1399                              &valleyview_cursor_wm_info,
1400                              &plane_sr, &ignore_cursor_sr) &&
1401             g4x_compute_srwm(dev, ffs(enabled) - 1,
1402                              2*sr_latency_ns,
1403                              &valleyview_wm_info,
1404                              &valleyview_cursor_wm_info,
1405                              &ignore_plane_sr, &cursor_sr)) {
1406                 I915_WRITE(FW_BLC_SELF_VLV, FW_CSPWRDWNEN);
1407         } else {
1408                 I915_WRITE(FW_BLC_SELF_VLV,
1409                            I915_READ(FW_BLC_SELF_VLV) & ~FW_CSPWRDWNEN);
1410                 plane_sr = cursor_sr = 0;
1411         }
1412
1413         DRM_DEBUG_KMS("Setting FIFO watermarks - A: plane=%d, cursor=%d, B: plane=%d, cursor=%d, SR: plane=%d, cursor=%d\n",
1414                       planea_wm, cursora_wm,
1415                       planeb_wm, cursorb_wm,
1416                       plane_sr, cursor_sr);
1417
1418         I915_WRITE(DSPFW1,
1419                    (plane_sr << DSPFW_SR_SHIFT) |
1420                    (cursorb_wm << DSPFW_CURSORB_SHIFT) |
1421                    (planeb_wm << DSPFW_PLANEB_SHIFT) |
1422                    planea_wm);
1423         I915_WRITE(DSPFW2,
1424                    (I915_READ(DSPFW2) & ~DSPFW_CURSORA_MASK) |
1425                    (cursora_wm << DSPFW_CURSORA_SHIFT));
1426         I915_WRITE(DSPFW3,
1427                    (I915_READ(DSPFW3) & ~DSPFW_CURSOR_SR_MASK) |
1428                    (cursor_sr << DSPFW_CURSOR_SR_SHIFT));
1429 }
1430
1431 static void g4x_update_wm(struct drm_crtc *crtc)
1432 {
1433         struct drm_device *dev = crtc->dev;
1434         static const int sr_latency_ns = 12000;
1435         struct drm_i915_private *dev_priv = dev->dev_private;
1436         int planea_wm, planeb_wm, cursora_wm, cursorb_wm;
1437         int plane_sr, cursor_sr;
1438         unsigned int enabled = 0;
1439
1440         if (g4x_compute_wm0(dev, PIPE_A,
1441                             &g4x_wm_info, latency_ns,
1442                             &g4x_cursor_wm_info, latency_ns,
1443                             &planea_wm, &cursora_wm))
1444                 enabled |= 1 << PIPE_A;
1445
1446         if (g4x_compute_wm0(dev, PIPE_B,
1447                             &g4x_wm_info, latency_ns,
1448                             &g4x_cursor_wm_info, latency_ns,
1449                             &planeb_wm, &cursorb_wm))
1450                 enabled |= 1 << PIPE_B;
1451
1452         if (single_plane_enabled(enabled) &&
1453             g4x_compute_srwm(dev, ffs(enabled) - 1,
1454                              sr_latency_ns,
1455                              &g4x_wm_info,
1456                              &g4x_cursor_wm_info,
1457                              &plane_sr, &cursor_sr)) {
1458                 I915_WRITE(FW_BLC_SELF, FW_BLC_SELF_EN);
1459         } else {
1460                 I915_WRITE(FW_BLC_SELF,
1461                            I915_READ(FW_BLC_SELF) & ~FW_BLC_SELF_EN);
1462                 plane_sr = cursor_sr = 0;
1463         }
1464
1465         DRM_DEBUG_KMS("Setting FIFO watermarks - A: plane=%d, cursor=%d, B: plane=%d, cursor=%d, SR: plane=%d, cursor=%d\n",
1466                       planea_wm, cursora_wm,
1467                       planeb_wm, cursorb_wm,
1468                       plane_sr, cursor_sr);
1469
1470         I915_WRITE(DSPFW1,
1471                    (plane_sr << DSPFW_SR_SHIFT) |
1472                    (cursorb_wm << DSPFW_CURSORB_SHIFT) |
1473                    (planeb_wm << DSPFW_PLANEB_SHIFT) |
1474                    planea_wm);
1475         I915_WRITE(DSPFW2,
1476                    (I915_READ(DSPFW2) & ~DSPFW_CURSORA_MASK) |
1477                    (cursora_wm << DSPFW_CURSORA_SHIFT));
1478         /* HPLL off in SR has some issues on G4x... disable it */
1479         I915_WRITE(DSPFW3,
1480                    (I915_READ(DSPFW3) & ~(DSPFW_HPLL_SR_EN | DSPFW_CURSOR_SR_MASK)) |
1481                    (cursor_sr << DSPFW_CURSOR_SR_SHIFT));
1482 }
1483
1484 static void i965_update_wm(struct drm_crtc *unused_crtc)
1485 {
1486         struct drm_device *dev = unused_crtc->dev;
1487         struct drm_i915_private *dev_priv = dev->dev_private;
1488         struct drm_crtc *crtc;
1489         int srwm = 1;
1490         int cursor_sr = 16;
1491
1492         /* Calc sr entries for one plane configs */
1493         crtc = single_enabled_crtc(dev);
1494         if (crtc) {
1495                 /* self-refresh has much higher latency */
1496                 static const int sr_latency_ns = 12000;
1497                 const struct drm_display_mode *adjusted_mode =
1498                         &to_intel_crtc(crtc)->config.adjusted_mode;
1499                 int clock = adjusted_mode->crtc_clock;
1500                 int htotal = adjusted_mode->htotal;
1501                 int hdisplay = to_intel_crtc(crtc)->config.pipe_src_w;
1502                 int pixel_size = crtc->fb->bits_per_pixel / 8;
1503                 unsigned long line_time_us;
1504                 int entries;
1505
1506                 line_time_us = ((htotal * 1000) / clock);
1507
1508                 /* Use ns/us then divide to preserve precision */
1509                 entries = (((sr_latency_ns / line_time_us) + 1000) / 1000) *
1510                         pixel_size * hdisplay;
1511                 entries = DIV_ROUND_UP(entries, I915_FIFO_LINE_SIZE);
1512                 srwm = I965_FIFO_SIZE - entries;
1513                 if (srwm < 0)
1514                         srwm = 1;
1515                 srwm &= 0x1ff;
1516                 DRM_DEBUG_KMS("self-refresh entries: %d, wm: %d\n",
1517                               entries, srwm);
1518
1519                 entries = (((sr_latency_ns / line_time_us) + 1000) / 1000) *
1520                         pixel_size * 64;
1521                 entries = DIV_ROUND_UP(entries,
1522                                           i965_cursor_wm_info.cacheline_size);
1523                 cursor_sr = i965_cursor_wm_info.fifo_size -
1524                         (entries + i965_cursor_wm_info.guard_size);
1525
1526                 if (cursor_sr > i965_cursor_wm_info.max_wm)
1527                         cursor_sr = i965_cursor_wm_info.max_wm;
1528
1529                 DRM_DEBUG_KMS("self-refresh watermark: display plane %d "
1530                               "cursor %d\n", srwm, cursor_sr);
1531
1532                 if (IS_CRESTLINE(dev))
1533                         I915_WRITE(FW_BLC_SELF, FW_BLC_SELF_EN);
1534         } else {
1535                 /* Turn off self refresh if both pipes are enabled */
1536                 if (IS_CRESTLINE(dev))
1537                         I915_WRITE(FW_BLC_SELF, I915_READ(FW_BLC_SELF)
1538                                    & ~FW_BLC_SELF_EN);
1539         }
1540
1541         DRM_DEBUG_KMS("Setting FIFO watermarks - A: 8, B: 8, C: 8, SR %d\n",
1542                       srwm);
1543
1544         /* 965 has limitations... */
1545         I915_WRITE(DSPFW1, (srwm << DSPFW_SR_SHIFT) |
1546                    (8 << 16) | (8 << 8) | (8 << 0));
1547         I915_WRITE(DSPFW2, (8 << 8) | (8 << 0));
1548         /* update cursor SR watermark */
1549         I915_WRITE(DSPFW3, (cursor_sr << DSPFW_CURSOR_SR_SHIFT));
1550 }
1551
1552 static void i9xx_update_wm(struct drm_crtc *unused_crtc)
1553 {
1554         struct drm_device *dev = unused_crtc->dev;
1555         struct drm_i915_private *dev_priv = dev->dev_private;
1556         const struct intel_watermark_params *wm_info;
1557         uint32_t fwater_lo;
1558         uint32_t fwater_hi;
1559         int cwm, srwm = 1;
1560         int fifo_size;
1561         int planea_wm, planeb_wm;
1562         struct drm_crtc *crtc, *enabled = NULL;
1563
1564         if (IS_I945GM(dev))
1565                 wm_info = &i945_wm_info;
1566         else if (!IS_GEN2(dev))
1567                 wm_info = &i915_wm_info;
1568         else
1569                 wm_info = &i855_wm_info;
1570
1571         fifo_size = dev_priv->display.get_fifo_size(dev, 0);
1572         crtc = intel_get_crtc_for_plane(dev, 0);
1573         if (intel_crtc_active(crtc)) {
1574                 const struct drm_display_mode *adjusted_mode;
1575                 int cpp = crtc->fb->bits_per_pixel / 8;
1576                 if (IS_GEN2(dev))
1577                         cpp = 4;
1578
1579                 adjusted_mode = &to_intel_crtc(crtc)->config.adjusted_mode;
1580                 planea_wm = intel_calculate_wm(adjusted_mode->crtc_clock,
1581                                                wm_info, fifo_size, cpp,
1582                                                latency_ns);
1583                 enabled = crtc;
1584         } else
1585                 planea_wm = fifo_size - wm_info->guard_size;
1586
1587         fifo_size = dev_priv->display.get_fifo_size(dev, 1);
1588         crtc = intel_get_crtc_for_plane(dev, 1);
1589         if (intel_crtc_active(crtc)) {
1590                 const struct drm_display_mode *adjusted_mode;
1591                 int cpp = crtc->fb->bits_per_pixel / 8;
1592                 if (IS_GEN2(dev))
1593                         cpp = 4;
1594
1595                 adjusted_mode = &to_intel_crtc(crtc)->config.adjusted_mode;
1596                 planeb_wm = intel_calculate_wm(adjusted_mode->crtc_clock,
1597                                                wm_info, fifo_size, cpp,
1598                                                latency_ns);
1599                 if (enabled == NULL)
1600                         enabled = crtc;
1601                 else
1602                         enabled = NULL;
1603         } else
1604                 planeb_wm = fifo_size - wm_info->guard_size;
1605
1606         DRM_DEBUG_KMS("FIFO watermarks - A: %d, B: %d\n", planea_wm, planeb_wm);
1607
1608         /*
1609          * Overlay gets an aggressive default since video jitter is bad.
1610          */
1611         cwm = 2;
1612
1613         /* Play safe and disable self-refresh before adjusting watermarks. */
1614         if (IS_I945G(dev) || IS_I945GM(dev))
1615                 I915_WRITE(FW_BLC_SELF, FW_BLC_SELF_EN_MASK | 0);
1616         else if (IS_I915GM(dev))
1617                 I915_WRITE(INSTPM, I915_READ(INSTPM) & ~INSTPM_SELF_EN);
1618
1619         /* Calc sr entries for one plane configs */
1620         if (HAS_FW_BLC(dev) && enabled) {
1621                 /* self-refresh has much higher latency */
1622                 static const int sr_latency_ns = 6000;
1623                 const struct drm_display_mode *adjusted_mode =
1624                         &to_intel_crtc(enabled)->config.adjusted_mode;
1625                 int clock = adjusted_mode->crtc_clock;
1626                 int htotal = adjusted_mode->htotal;
1627                 int hdisplay = to_intel_crtc(crtc)->config.pipe_src_w;
1628                 int pixel_size = enabled->fb->bits_per_pixel / 8;
1629                 unsigned long line_time_us;
1630                 int entries;
1631
1632                 line_time_us = (htotal * 1000) / clock;
1633
1634                 /* Use ns/us then divide to preserve precision */
1635                 entries = (((sr_latency_ns / line_time_us) + 1000) / 1000) *
1636                         pixel_size * hdisplay;
1637                 entries = DIV_ROUND_UP(entries, wm_info->cacheline_size);
1638                 DRM_DEBUG_KMS("self-refresh entries: %d\n", entries);
1639                 srwm = wm_info->fifo_size - entries;
1640                 if (srwm < 0)
1641                         srwm = 1;
1642
1643                 if (IS_I945G(dev) || IS_I945GM(dev))
1644                         I915_WRITE(FW_BLC_SELF,
1645                                    FW_BLC_SELF_FIFO_MASK | (srwm & 0xff));
1646                 else if (IS_I915GM(dev))
1647                         I915_WRITE(FW_BLC_SELF, srwm & 0x3f);
1648         }
1649
1650         DRM_DEBUG_KMS("Setting FIFO watermarks - A: %d, B: %d, C: %d, SR %d\n",
1651                       planea_wm, planeb_wm, cwm, srwm);
1652
1653         fwater_lo = ((planeb_wm & 0x3f) << 16) | (planea_wm & 0x3f);
1654         fwater_hi = (cwm & 0x1f);
1655
1656         /* Set request length to 8 cachelines per fetch */
1657         fwater_lo = fwater_lo | (1 << 24) | (1 << 8);
1658         fwater_hi = fwater_hi | (1 << 8);
1659
1660         I915_WRITE(FW_BLC, fwater_lo);
1661         I915_WRITE(FW_BLC2, fwater_hi);
1662
1663         if (HAS_FW_BLC(dev)) {
1664                 if (enabled) {
1665                         if (IS_I945G(dev) || IS_I945GM(dev))
1666                                 I915_WRITE(FW_BLC_SELF,
1667                                            FW_BLC_SELF_EN_MASK | FW_BLC_SELF_EN);
1668                         else if (IS_I915GM(dev))
1669                                 I915_WRITE(INSTPM, I915_READ(INSTPM) | INSTPM_SELF_EN);
1670                         DRM_DEBUG_KMS("memory self refresh enabled\n");
1671                 } else
1672                         DRM_DEBUG_KMS("memory self refresh disabled\n");
1673         }
1674 }
1675
1676 static void i830_update_wm(struct drm_crtc *unused_crtc)
1677 {
1678         struct drm_device *dev = unused_crtc->dev;
1679         struct drm_i915_private *dev_priv = dev->dev_private;
1680         struct drm_crtc *crtc;
1681         const struct drm_display_mode *adjusted_mode;
1682         uint32_t fwater_lo;
1683         int planea_wm;
1684
1685         crtc = single_enabled_crtc(dev);
1686         if (crtc == NULL)
1687                 return;
1688
1689         adjusted_mode = &to_intel_crtc(crtc)->config.adjusted_mode;
1690         planea_wm = intel_calculate_wm(adjusted_mode->crtc_clock,
1691                                        &i830_wm_info,
1692                                        dev_priv->display.get_fifo_size(dev, 0),
1693                                        4, latency_ns);
1694         fwater_lo = I915_READ(FW_BLC) & ~0xfff;
1695         fwater_lo |= (3<<8) | planea_wm;
1696
1697         DRM_DEBUG_KMS("Setting FIFO watermarks - A: %d\n", planea_wm);
1698
1699         I915_WRITE(FW_BLC, fwater_lo);
1700 }
1701
1702 /*
1703  * Check the wm result.
1704  *
1705  * If any calculated watermark values is larger than the maximum value that
1706  * can be programmed into the associated watermark register, that watermark
1707  * must be disabled.
1708  */
1709 static bool ironlake_check_srwm(struct drm_device *dev, int level,
1710                                 int fbc_wm, int display_wm, int cursor_wm,
1711                                 const struct intel_watermark_params *display,
1712                                 const struct intel_watermark_params *cursor)
1713 {
1714         struct drm_i915_private *dev_priv = dev->dev_private;
1715
1716         DRM_DEBUG_KMS("watermark %d: display plane %d, fbc lines %d,"
1717                       " cursor %d\n", level, display_wm, fbc_wm, cursor_wm);
1718
1719         if (fbc_wm > SNB_FBC_MAX_SRWM) {
1720                 DRM_DEBUG_KMS("fbc watermark(%d) is too large(%d), disabling wm%d+\n",
1721                               fbc_wm, SNB_FBC_MAX_SRWM, level);
1722
1723                 /* fbc has it's own way to disable FBC WM */
1724                 I915_WRITE(DISP_ARB_CTL,
1725                            I915_READ(DISP_ARB_CTL) | DISP_FBC_WM_DIS);
1726                 return false;
1727         } else if (INTEL_INFO(dev)->gen >= 6) {
1728                 /* enable FBC WM (except on ILK, where it must remain off) */
1729                 I915_WRITE(DISP_ARB_CTL,
1730                            I915_READ(DISP_ARB_CTL) & ~DISP_FBC_WM_DIS);
1731         }
1732
1733         if (display_wm > display->max_wm) {
1734                 DRM_DEBUG_KMS("display watermark(%d) is too large(%d), disabling wm%d+\n",
1735                               display_wm, SNB_DISPLAY_MAX_SRWM, level);
1736                 return false;
1737         }
1738
1739         if (cursor_wm > cursor->max_wm) {
1740                 DRM_DEBUG_KMS("cursor watermark(%d) is too large(%d), disabling wm%d+\n",
1741                               cursor_wm, SNB_CURSOR_MAX_SRWM, level);
1742                 return false;
1743         }
1744
1745         if (!(fbc_wm || display_wm || cursor_wm)) {
1746                 DRM_DEBUG_KMS("latency %d is 0, disabling wm%d+\n", level, level);
1747                 return false;
1748         }
1749
1750         return true;
1751 }
1752
1753 /*
1754  * Compute watermark values of WM[1-3],
1755  */
1756 static bool ironlake_compute_srwm(struct drm_device *dev, int level, int plane,
1757                                   int latency_ns,
1758                                   const struct intel_watermark_params *display,
1759                                   const struct intel_watermark_params *cursor,
1760                                   int *fbc_wm, int *display_wm, int *cursor_wm)
1761 {
1762         struct drm_crtc *crtc;
1763         const struct drm_display_mode *adjusted_mode;
1764         unsigned long line_time_us;
1765         int hdisplay, htotal, pixel_size, clock;
1766         int line_count, line_size;
1767         int small, large;
1768         int entries;
1769
1770         if (!latency_ns) {
1771                 *fbc_wm = *display_wm = *cursor_wm = 0;
1772                 return false;
1773         }
1774
1775         crtc = intel_get_crtc_for_plane(dev, plane);
1776         adjusted_mode = &to_intel_crtc(crtc)->config.adjusted_mode;
1777         clock = adjusted_mode->crtc_clock;
1778         htotal = adjusted_mode->htotal;
1779         hdisplay = to_intel_crtc(crtc)->config.pipe_src_w;
1780         pixel_size = crtc->fb->bits_per_pixel / 8;
1781
1782         line_time_us = (htotal * 1000) / clock;
1783         line_count = (latency_ns / line_time_us + 1000) / 1000;
1784         line_size = hdisplay * pixel_size;
1785
1786         /* Use the minimum of the small and large buffer method for primary */
1787         small = ((clock * pixel_size / 1000) * latency_ns) / 1000;
1788         large = line_count * line_size;
1789
1790         entries = DIV_ROUND_UP(min(small, large), display->cacheline_size);
1791         *display_wm = entries + display->guard_size;
1792
1793         /*
1794          * Spec says:
1795          * FBC WM = ((Final Primary WM * 64) / number of bytes per line) + 2
1796          */
1797         *fbc_wm = DIV_ROUND_UP(*display_wm * 64, line_size) + 2;
1798
1799         /* calculate the self-refresh watermark for display cursor */
1800         entries = line_count * pixel_size * 64;
1801         entries = DIV_ROUND_UP(entries, cursor->cacheline_size);
1802         *cursor_wm = entries + cursor->guard_size;
1803
1804         return ironlake_check_srwm(dev, level,
1805                                    *fbc_wm, *display_wm, *cursor_wm,
1806                                    display, cursor);
1807 }
1808
1809 static void ironlake_update_wm(struct drm_crtc *crtc)
1810 {
1811         struct drm_device *dev = crtc->dev;
1812         struct drm_i915_private *dev_priv = dev->dev_private;
1813         int fbc_wm, plane_wm, cursor_wm;
1814         unsigned int enabled;
1815
1816         enabled = 0;
1817         if (g4x_compute_wm0(dev, PIPE_A,
1818                             &ironlake_display_wm_info,
1819                             dev_priv->wm.pri_latency[0] * 100,
1820                             &ironlake_cursor_wm_info,
1821                             dev_priv->wm.cur_latency[0] * 100,
1822                             &plane_wm, &cursor_wm)) {
1823                 I915_WRITE(WM0_PIPEA_ILK,
1824                            (plane_wm << WM0_PIPE_PLANE_SHIFT) | cursor_wm);
1825                 DRM_DEBUG_KMS("FIFO watermarks For pipe A -"
1826                               " plane %d, " "cursor: %d\n",
1827                               plane_wm, cursor_wm);
1828                 enabled |= 1 << PIPE_A;
1829         }
1830
1831         if (g4x_compute_wm0(dev, PIPE_B,
1832                             &ironlake_display_wm_info,
1833                             dev_priv->wm.pri_latency[0] * 100,
1834                             &ironlake_cursor_wm_info,
1835                             dev_priv->wm.cur_latency[0] * 100,
1836                             &plane_wm, &cursor_wm)) {
1837                 I915_WRITE(WM0_PIPEB_ILK,
1838                            (plane_wm << WM0_PIPE_PLANE_SHIFT) | cursor_wm);
1839                 DRM_DEBUG_KMS("FIFO watermarks For pipe B -"
1840                               " plane %d, cursor: %d\n",
1841                               plane_wm, cursor_wm);
1842                 enabled |= 1 << PIPE_B;
1843         }
1844
1845         /*
1846          * Calculate and update the self-refresh watermark only when one
1847          * display plane is used.
1848          */
1849         I915_WRITE(WM3_LP_ILK, 0);
1850         I915_WRITE(WM2_LP_ILK, 0);
1851         I915_WRITE(WM1_LP_ILK, 0);
1852
1853         if (!single_plane_enabled(enabled))
1854                 return;
1855         enabled = ffs(enabled) - 1;
1856
1857         /* WM1 */
1858         if (!ironlake_compute_srwm(dev, 1, enabled,
1859                                    dev_priv->wm.pri_latency[1] * 500,
1860                                    &ironlake_display_srwm_info,
1861                                    &ironlake_cursor_srwm_info,
1862                                    &fbc_wm, &plane_wm, &cursor_wm))
1863                 return;
1864
1865         I915_WRITE(WM1_LP_ILK,
1866                    WM1_LP_SR_EN |
1867                    (dev_priv->wm.pri_latency[1] << WM1_LP_LATENCY_SHIFT) |
1868                    (fbc_wm << WM1_LP_FBC_SHIFT) |
1869                    (plane_wm << WM1_LP_SR_SHIFT) |
1870                    cursor_wm);
1871
1872         /* WM2 */
1873         if (!ironlake_compute_srwm(dev, 2, enabled,
1874                                    dev_priv->wm.pri_latency[2] * 500,
1875                                    &ironlake_display_srwm_info,
1876                                    &ironlake_cursor_srwm_info,
1877                                    &fbc_wm, &plane_wm, &cursor_wm))
1878                 return;
1879
1880         I915_WRITE(WM2_LP_ILK,
1881                    WM2_LP_EN |
1882                    (dev_priv->wm.pri_latency[2] << WM1_LP_LATENCY_SHIFT) |
1883                    (fbc_wm << WM1_LP_FBC_SHIFT) |
1884                    (plane_wm << WM1_LP_SR_SHIFT) |
1885                    cursor_wm);
1886
1887         /*
1888          * WM3 is unsupported on ILK, probably because we don't have latency
1889          * data for that power state
1890          */
1891 }
1892
1893 static void sandybridge_update_wm(struct drm_crtc *crtc)
1894 {
1895         struct drm_device *dev = crtc->dev;
1896         struct drm_i915_private *dev_priv = dev->dev_private;
1897         int latency = dev_priv->wm.pri_latency[0] * 100;        /* In unit 0.1us */
1898         u32 val;
1899         int fbc_wm, plane_wm, cursor_wm;
1900         unsigned int enabled;
1901
1902         enabled = 0;
1903         if (g4x_compute_wm0(dev, PIPE_A,
1904                             &sandybridge_display_wm_info, latency,
1905                             &sandybridge_cursor_wm_info, latency,
1906                             &plane_wm, &cursor_wm)) {
1907                 val = I915_READ(WM0_PIPEA_ILK);
1908                 val &= ~(WM0_PIPE_PLANE_MASK | WM0_PIPE_CURSOR_MASK);
1909                 I915_WRITE(WM0_PIPEA_ILK, val |
1910                            ((plane_wm << WM0_PIPE_PLANE_SHIFT) | cursor_wm));
1911                 DRM_DEBUG_KMS("FIFO watermarks For pipe A -"
1912                               " plane %d, " "cursor: %d\n",
1913                               plane_wm, cursor_wm);
1914                 enabled |= 1 << PIPE_A;
1915         }
1916
1917         if (g4x_compute_wm0(dev, PIPE_B,
1918                             &sandybridge_display_wm_info, latency,
1919                             &sandybridge_cursor_wm_info, latency,
1920                             &plane_wm, &cursor_wm)) {
1921                 val = I915_READ(WM0_PIPEB_ILK);
1922                 val &= ~(WM0_PIPE_PLANE_MASK | WM0_PIPE_CURSOR_MASK);
1923                 I915_WRITE(WM0_PIPEB_ILK, val |
1924                            ((plane_wm << WM0_PIPE_PLANE_SHIFT) | cursor_wm));
1925                 DRM_DEBUG_KMS("FIFO watermarks For pipe B -"
1926                               " plane %d, cursor: %d\n",
1927                               plane_wm, cursor_wm);
1928                 enabled |= 1 << PIPE_B;
1929         }
1930
1931         /*
1932          * Calculate and update the self-refresh watermark only when one
1933          * display plane is used.
1934          *
1935          * SNB support 3 levels of watermark.
1936          *
1937          * WM1/WM2/WM2 watermarks have to be enabled in the ascending order,
1938          * and disabled in the descending order
1939          *
1940          */
1941         I915_WRITE(WM3_LP_ILK, 0);
1942         I915_WRITE(WM2_LP_ILK, 0);
1943         I915_WRITE(WM1_LP_ILK, 0);
1944
1945         if (!single_plane_enabled(enabled) ||
1946             dev_priv->sprite_scaling_enabled)
1947                 return;
1948         enabled = ffs(enabled) - 1;
1949
1950         /* WM1 */
1951         if (!ironlake_compute_srwm(dev, 1, enabled,
1952                                    dev_priv->wm.pri_latency[1] * 500,
1953                                    &sandybridge_display_srwm_info,
1954                                    &sandybridge_cursor_srwm_info,
1955                                    &fbc_wm, &plane_wm, &cursor_wm))
1956                 return;
1957
1958         I915_WRITE(WM1_LP_ILK,
1959                    WM1_LP_SR_EN |
1960                    (dev_priv->wm.pri_latency[1] << WM1_LP_LATENCY_SHIFT) |
1961                    (fbc_wm << WM1_LP_FBC_SHIFT) |
1962                    (plane_wm << WM1_LP_SR_SHIFT) |
1963                    cursor_wm);
1964
1965         /* WM2 */
1966         if (!ironlake_compute_srwm(dev, 2, enabled,
1967                                    dev_priv->wm.pri_latency[2] * 500,
1968                                    &sandybridge_display_srwm_info,
1969                                    &sandybridge_cursor_srwm_info,
1970                                    &fbc_wm, &plane_wm, &cursor_wm))
1971                 return;
1972
1973         I915_WRITE(WM2_LP_ILK,
1974                    WM2_LP_EN |
1975                    (dev_priv->wm.pri_latency[2] << WM1_LP_LATENCY_SHIFT) |
1976                    (fbc_wm << WM1_LP_FBC_SHIFT) |
1977                    (plane_wm << WM1_LP_SR_SHIFT) |
1978                    cursor_wm);
1979
1980         /* WM3 */
1981         if (!ironlake_compute_srwm(dev, 3, enabled,
1982                                    dev_priv->wm.pri_latency[3] * 500,
1983                                    &sandybridge_display_srwm_info,
1984                                    &sandybridge_cursor_srwm_info,
1985                                    &fbc_wm, &plane_wm, &cursor_wm))
1986                 return;
1987
1988         I915_WRITE(WM3_LP_ILK,
1989                    WM3_LP_EN |
1990                    (dev_priv->wm.pri_latency[3] << WM1_LP_LATENCY_SHIFT) |
1991                    (fbc_wm << WM1_LP_FBC_SHIFT) |
1992                    (plane_wm << WM1_LP_SR_SHIFT) |
1993                    cursor_wm);
1994 }
1995
1996 static void ivybridge_update_wm(struct drm_crtc *crtc)
1997 {
1998         struct drm_device *dev = crtc->dev;
1999         struct drm_i915_private *dev_priv = dev->dev_private;
2000         int latency = dev_priv->wm.pri_latency[0] * 100;        /* In unit 0.1us */
2001         u32 val;
2002         int fbc_wm, plane_wm, cursor_wm;
2003         int ignore_fbc_wm, ignore_plane_wm, ignore_cursor_wm;
2004         unsigned int enabled;
2005
2006         enabled = 0;
2007         if (g4x_compute_wm0(dev, PIPE_A,
2008                             &sandybridge_display_wm_info, latency,
2009                             &sandybridge_cursor_wm_info, latency,
2010                             &plane_wm, &cursor_wm)) {
2011                 val = I915_READ(WM0_PIPEA_ILK);
2012                 val &= ~(WM0_PIPE_PLANE_MASK | WM0_PIPE_CURSOR_MASK);
2013                 I915_WRITE(WM0_PIPEA_ILK, val |
2014                            ((plane_wm << WM0_PIPE_PLANE_SHIFT) | cursor_wm));
2015                 DRM_DEBUG_KMS("FIFO watermarks For pipe A -"
2016                               " plane %d, " "cursor: %d\n",
2017                               plane_wm, cursor_wm);
2018                 enabled |= 1 << PIPE_A;
2019         }
2020
2021         if (g4x_compute_wm0(dev, PIPE_B,
2022                             &sandybridge_display_wm_info, latency,
2023                             &sandybridge_cursor_wm_info, latency,
2024                             &plane_wm, &cursor_wm)) {
2025                 val = I915_READ(WM0_PIPEB_ILK);
2026                 val &= ~(WM0_PIPE_PLANE_MASK | WM0_PIPE_CURSOR_MASK);
2027                 I915_WRITE(WM0_PIPEB_ILK, val |
2028                            ((plane_wm << WM0_PIPE_PLANE_SHIFT) | cursor_wm));
2029                 DRM_DEBUG_KMS("FIFO watermarks For pipe B -"
2030                               " plane %d, cursor: %d\n",
2031                               plane_wm, cursor_wm);
2032                 enabled |= 1 << PIPE_B;
2033         }
2034
2035         if (g4x_compute_wm0(dev, PIPE_C,
2036                             &sandybridge_display_wm_info, latency,
2037                             &sandybridge_cursor_wm_info, latency,
2038                             &plane_wm, &cursor_wm)) {
2039                 val = I915_READ(WM0_PIPEC_IVB);
2040                 val &= ~(WM0_PIPE_PLANE_MASK | WM0_PIPE_CURSOR_MASK);
2041                 I915_WRITE(WM0_PIPEC_IVB, val |
2042                            ((plane_wm << WM0_PIPE_PLANE_SHIFT) | cursor_wm));
2043                 DRM_DEBUG_KMS("FIFO watermarks For pipe C -"
2044                               " plane %d, cursor: %d\n",
2045                               plane_wm, cursor_wm);
2046                 enabled |= 1 << PIPE_C;
2047         }
2048
2049         /*
2050          * Calculate and update the self-refresh watermark only when one
2051          * display plane is used.
2052          *
2053          * SNB support 3 levels of watermark.
2054          *
2055          * WM1/WM2/WM2 watermarks have to be enabled in the ascending order,
2056          * and disabled in the descending order
2057          *
2058          */
2059         I915_WRITE(WM3_LP_ILK, 0);
2060         I915_WRITE(WM2_LP_ILK, 0);
2061         I915_WRITE(WM1_LP_ILK, 0);
2062
2063         if (!single_plane_enabled(enabled) ||
2064             dev_priv->sprite_scaling_enabled)
2065                 return;
2066         enabled = ffs(enabled) - 1;
2067
2068         /* WM1 */
2069         if (!ironlake_compute_srwm(dev, 1, enabled,
2070                                    dev_priv->wm.pri_latency[1] * 500,
2071                                    &sandybridge_display_srwm_info,
2072                                    &sandybridge_cursor_srwm_info,
2073                                    &fbc_wm, &plane_wm, &cursor_wm))
2074                 return;
2075
2076         I915_WRITE(WM1_LP_ILK,
2077                    WM1_LP_SR_EN |
2078                    (dev_priv->wm.pri_latency[1] << WM1_LP_LATENCY_SHIFT) |
2079                    (fbc_wm << WM1_LP_FBC_SHIFT) |
2080                    (plane_wm << WM1_LP_SR_SHIFT) |
2081                    cursor_wm);
2082
2083         /* WM2 */
2084         if (!ironlake_compute_srwm(dev, 2, enabled,
2085                                    dev_priv->wm.pri_latency[2] * 500,
2086                                    &sandybridge_display_srwm_info,
2087                                    &sandybridge_cursor_srwm_info,
2088                                    &fbc_wm, &plane_wm, &cursor_wm))
2089                 return;
2090
2091         I915_WRITE(WM2_LP_ILK,
2092                    WM2_LP_EN |
2093                    (dev_priv->wm.pri_latency[2] << WM1_LP_LATENCY_SHIFT) |
2094                    (fbc_wm << WM1_LP_FBC_SHIFT) |
2095                    (plane_wm << WM1_LP_SR_SHIFT) |
2096                    cursor_wm);
2097
2098         /* WM3, note we have to correct the cursor latency */
2099         if (!ironlake_compute_srwm(dev, 3, enabled,
2100                                    dev_priv->wm.pri_latency[3] * 500,
2101                                    &sandybridge_display_srwm_info,
2102                                    &sandybridge_cursor_srwm_info,
2103                                    &fbc_wm, &plane_wm, &ignore_cursor_wm) ||
2104             !ironlake_compute_srwm(dev, 3, enabled,
2105                                    dev_priv->wm.cur_latency[3] * 500,
2106                                    &sandybridge_display_srwm_info,
2107                                    &sandybridge_cursor_srwm_info,
2108                                    &ignore_fbc_wm, &ignore_plane_wm, &cursor_wm))
2109                 return;
2110
2111         I915_WRITE(WM3_LP_ILK,
2112                    WM3_LP_EN |
2113                    (dev_priv->wm.pri_latency[3] << WM1_LP_LATENCY_SHIFT) |
2114                    (fbc_wm << WM1_LP_FBC_SHIFT) |
2115                    (plane_wm << WM1_LP_SR_SHIFT) |
2116                    cursor_wm);
2117 }
2118
2119 static uint32_t ilk_pipe_pixel_rate(struct drm_device *dev,
2120                                     struct drm_crtc *crtc)
2121 {
2122         struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
2123         uint32_t pixel_rate;
2124
2125         pixel_rate = intel_crtc->config.adjusted_mode.crtc_clock;
2126
2127         /* We only use IF-ID interlacing. If we ever use PF-ID we'll need to
2128          * adjust the pixel_rate here. */
2129
2130         if (intel_crtc->config.pch_pfit.enabled) {
2131                 uint64_t pipe_w, pipe_h, pfit_w, pfit_h;
2132                 uint32_t pfit_size = intel_crtc->config.pch_pfit.size;
2133
2134                 pipe_w = intel_crtc->config.pipe_src_w;
2135                 pipe_h = intel_crtc->config.pipe_src_h;
2136                 pfit_w = (pfit_size >> 16) & 0xFFFF;
2137                 pfit_h = pfit_size & 0xFFFF;
2138                 if (pipe_w < pfit_w)
2139                         pipe_w = pfit_w;
2140                 if (pipe_h < pfit_h)
2141                         pipe_h = pfit_h;
2142
2143                 pixel_rate = div_u64((uint64_t) pixel_rate * pipe_w * pipe_h,
2144                                      pfit_w * pfit_h);
2145         }
2146
2147         return pixel_rate;
2148 }
2149
2150 /* latency must be in 0.1us units. */
2151 static uint32_t ilk_wm_method1(uint32_t pixel_rate, uint8_t bytes_per_pixel,
2152                                uint32_t latency)
2153 {
2154         uint64_t ret;
2155
2156         if (WARN(latency == 0, "Latency value missing\n"))
2157                 return UINT_MAX;
2158
2159         ret = (uint64_t) pixel_rate * bytes_per_pixel * latency;
2160         ret = DIV_ROUND_UP_ULL(ret, 64 * 10000) + 2;
2161
2162         return ret;
2163 }
2164
2165 /* latency must be in 0.1us units. */
2166 static uint32_t ilk_wm_method2(uint32_t pixel_rate, uint32_t pipe_htotal,
2167                                uint32_t horiz_pixels, uint8_t bytes_per_pixel,
2168                                uint32_t latency)
2169 {
2170         uint32_t ret;
2171
2172         if (WARN(latency == 0, "Latency value missing\n"))
2173                 return UINT_MAX;
2174
2175         ret = (latency * pixel_rate) / (pipe_htotal * 10000);
2176         ret = (ret + 1) * horiz_pixels * bytes_per_pixel;
2177         ret = DIV_ROUND_UP(ret, 64) + 2;
2178         return ret;
2179 }
2180
2181 static uint32_t ilk_wm_fbc(uint32_t pri_val, uint32_t horiz_pixels,
2182                            uint8_t bytes_per_pixel)
2183 {
2184         return DIV_ROUND_UP(pri_val * 64, horiz_pixels * bytes_per_pixel) + 2;
2185 }
2186
2187 struct hsw_pipe_wm_parameters {
2188         bool active;
2189         uint32_t pipe_htotal;
2190         uint32_t pixel_rate;
2191         struct intel_plane_wm_parameters pri;
2192         struct intel_plane_wm_parameters spr;
2193         struct intel_plane_wm_parameters cur;
2194 };
2195
2196 struct hsw_wm_maximums {
2197         uint16_t pri;
2198         uint16_t spr;
2199         uint16_t cur;
2200         uint16_t fbc;
2201 };
2202
2203 struct hsw_wm_values {
2204         uint32_t wm_pipe[3];
2205         uint32_t wm_lp[3];
2206         uint32_t wm_lp_spr[3];
2207         uint32_t wm_linetime[3];
2208         bool enable_fbc_wm;
2209 };
2210
2211 /* used in computing the new watermarks state */
2212 struct intel_wm_config {
2213         unsigned int num_pipes_active;
2214         bool sprites_enabled;
2215         bool sprites_scaled;
2216 };
2217
2218 /*
2219  * For both WM_PIPE and WM_LP.
2220  * mem_value must be in 0.1us units.
2221  */
2222 static uint32_t ilk_compute_pri_wm(const struct hsw_pipe_wm_parameters *params,
2223                                    uint32_t mem_value,
2224                                    bool is_lp)
2225 {
2226         uint32_t method1, method2;
2227
2228         if (!params->active || !params->pri.enabled)
2229                 return 0;
2230
2231         method1 = ilk_wm_method1(params->pixel_rate,
2232                                  params->pri.bytes_per_pixel,
2233                                  mem_value);
2234
2235         if (!is_lp)
2236                 return method1;
2237
2238         method2 = ilk_wm_method2(params->pixel_rate,
2239                                  params->pipe_htotal,
2240                                  params->pri.horiz_pixels,
2241                                  params->pri.bytes_per_pixel,
2242                                  mem_value);
2243
2244         return min(method1, method2);
2245 }
2246
2247 /*
2248  * For both WM_PIPE and WM_LP.
2249  * mem_value must be in 0.1us units.
2250  */
2251 static uint32_t ilk_compute_spr_wm(const struct hsw_pipe_wm_parameters *params,
2252                                    uint32_t mem_value)
2253 {
2254         uint32_t method1, method2;
2255
2256         if (!params->active || !params->spr.enabled)
2257                 return 0;
2258
2259         method1 = ilk_wm_method1(params->pixel_rate,
2260                                  params->spr.bytes_per_pixel,
2261                                  mem_value);
2262         method2 = ilk_wm_method2(params->pixel_rate,
2263                                  params->pipe_htotal,
2264                                  params->spr.horiz_pixels,
2265                                  params->spr.bytes_per_pixel,
2266                                  mem_value);
2267         return min(method1, method2);
2268 }
2269
2270 /*
2271  * For both WM_PIPE and WM_LP.
2272  * mem_value must be in 0.1us units.
2273  */
2274 static uint32_t ilk_compute_cur_wm(const struct hsw_pipe_wm_parameters *params,
2275                                    uint32_t mem_value)
2276 {
2277         if (!params->active || !params->cur.enabled)
2278                 return 0;
2279
2280         return ilk_wm_method2(params->pixel_rate,
2281                               params->pipe_htotal,
2282                               params->cur.horiz_pixels,
2283                               params->cur.bytes_per_pixel,
2284                               mem_value);
2285 }
2286
2287 /* Only for WM_LP. */
2288 static uint32_t ilk_compute_fbc_wm(const struct hsw_pipe_wm_parameters *params,
2289                                    uint32_t pri_val)
2290 {
2291         if (!params->active || !params->pri.enabled)
2292                 return 0;
2293
2294         return ilk_wm_fbc(pri_val,
2295                           params->pri.horiz_pixels,
2296                           params->pri.bytes_per_pixel);
2297 }
2298
2299 static unsigned int ilk_display_fifo_size(const struct drm_device *dev)
2300 {
2301         if (INTEL_INFO(dev)->gen >= 7)
2302                 return 768;
2303         else
2304                 return 512;
2305 }
2306
2307 /* Calculate the maximum primary/sprite plane watermark */
2308 static unsigned int ilk_plane_wm_max(const struct drm_device *dev,
2309                                      int level,
2310                                      const struct intel_wm_config *config,
2311                                      enum intel_ddb_partitioning ddb_partitioning,
2312                                      bool is_sprite)
2313 {
2314         unsigned int fifo_size = ilk_display_fifo_size(dev);
2315         unsigned int max;
2316
2317         /* if sprites aren't enabled, sprites get nothing */
2318         if (is_sprite && !config->sprites_enabled)
2319                 return 0;
2320
2321         /* HSW allows LP1+ watermarks even with multiple pipes */
2322         if (level == 0 || config->num_pipes_active > 1) {
2323                 fifo_size /= INTEL_INFO(dev)->num_pipes;
2324
2325                 /*
2326                  * For some reason the non self refresh
2327                  * FIFO size is only half of the self
2328                  * refresh FIFO size on ILK/SNB.
2329                  */
2330                 if (INTEL_INFO(dev)->gen <= 6)
2331                         fifo_size /= 2;
2332         }
2333
2334         if (config->sprites_enabled) {
2335                 /* level 0 is always calculated with 1:1 split */
2336                 if (level > 0 && ddb_partitioning == INTEL_DDB_PART_5_6) {
2337                         if (is_sprite)
2338                                 fifo_size *= 5;
2339                         fifo_size /= 6;
2340                 } else {
2341                         fifo_size /= 2;
2342                 }
2343         }
2344
2345         /* clamp to max that the registers can hold */
2346         if (INTEL_INFO(dev)->gen >= 7)
2347                 /* IVB/HSW primary/sprite plane watermarks */
2348                 max = level == 0 ? 127 : 1023;
2349         else if (!is_sprite)
2350                 /* ILK/SNB primary plane watermarks */
2351                 max = level == 0 ? 127 : 511;
2352         else
2353                 /* ILK/SNB sprite plane watermarks */
2354                 max = level == 0 ? 63 : 255;
2355
2356         return min(fifo_size, max);
2357 }
2358
2359 /* Calculate the maximum cursor plane watermark */
2360 static unsigned int ilk_cursor_wm_max(const struct drm_device *dev,
2361                                       int level,
2362                                       const struct intel_wm_config *config)
2363 {
2364         /* HSW LP1+ watermarks w/ multiple pipes */
2365         if (level > 0 && config->num_pipes_active > 1)
2366                 return 64;
2367
2368         /* otherwise just report max that registers can hold */
2369         if (INTEL_INFO(dev)->gen >= 7)
2370                 return level == 0 ? 63 : 255;
2371         else
2372                 return level == 0 ? 31 : 63;
2373 }
2374
2375 /* Calculate the maximum FBC watermark */
2376 static unsigned int ilk_fbc_wm_max(void)
2377 {
2378         /* max that registers can hold */
2379         return 15;
2380 }
2381
2382 static void ilk_wm_max(struct drm_device *dev,
2383                        int level,
2384                        const struct intel_wm_config *config,
2385                        enum intel_ddb_partitioning ddb_partitioning,
2386                        struct hsw_wm_maximums *max)
2387 {
2388         max->pri = ilk_plane_wm_max(dev, level, config, ddb_partitioning, false);
2389         max->spr = ilk_plane_wm_max(dev, level, config, ddb_partitioning, true);
2390         max->cur = ilk_cursor_wm_max(dev, level, config);
2391         max->fbc = ilk_fbc_wm_max();
2392 }
2393
2394 static bool ilk_check_wm(int level,
2395                          const struct hsw_wm_maximums *max,
2396                          struct intel_wm_level *result)
2397 {
2398         bool ret;
2399
2400         /* already determined to be invalid? */
2401         if (!result->enable)
2402                 return false;
2403
2404         result->enable = result->pri_val <= max->pri &&
2405                          result->spr_val <= max->spr &&
2406                          result->cur_val <= max->cur;
2407
2408         ret = result->enable;
2409
2410         /*
2411          * HACK until we can pre-compute everything,
2412          * and thus fail gracefully if LP0 watermarks
2413          * are exceeded...
2414          */
2415         if (level == 0 && !result->enable) {
2416                 if (result->pri_val > max->pri)
2417                         DRM_DEBUG_KMS("Primary WM%d too large %u (max %u)\n",
2418                                       level, result->pri_val, max->pri);
2419                 if (result->spr_val > max->spr)
2420                         DRM_DEBUG_KMS("Sprite WM%d too large %u (max %u)\n",
2421                                       level, result->spr_val, max->spr);
2422                 if (result->cur_val > max->cur)
2423                         DRM_DEBUG_KMS("Cursor WM%d too large %u (max %u)\n",
2424                                       level, result->cur_val, max->cur);
2425
2426                 result->pri_val = min_t(uint32_t, result->pri_val, max->pri);
2427                 result->spr_val = min_t(uint32_t, result->spr_val, max->spr);
2428                 result->cur_val = min_t(uint32_t, result->cur_val, max->cur);
2429                 result->enable = true;
2430         }
2431
2432         DRM_DEBUG_KMS("WM%d: %sabled\n", level, result->enable ? "en" : "dis");
2433
2434         return ret;
2435 }
2436
2437 static void ilk_compute_wm_level(struct drm_i915_private *dev_priv,
2438                                  int level,
2439                                  const struct hsw_pipe_wm_parameters *p,
2440                                  struct intel_wm_level *result)
2441 {
2442         uint16_t pri_latency = dev_priv->wm.pri_latency[level];
2443         uint16_t spr_latency = dev_priv->wm.spr_latency[level];
2444         uint16_t cur_latency = dev_priv->wm.cur_latency[level];
2445
2446         /* WM1+ latency values stored in 0.5us units */
2447         if (level > 0) {
2448                 pri_latency *= 5;
2449                 spr_latency *= 5;
2450                 cur_latency *= 5;
2451         }
2452
2453         result->pri_val = ilk_compute_pri_wm(p, pri_latency, level);
2454         result->spr_val = ilk_compute_spr_wm(p, spr_latency);
2455         result->cur_val = ilk_compute_cur_wm(p, cur_latency);
2456         result->fbc_val = ilk_compute_fbc_wm(p, result->pri_val);
2457         result->enable = true;
2458 }
2459
2460 static uint32_t
2461 hsw_compute_linetime_wm(struct drm_device *dev, struct drm_crtc *crtc)
2462 {
2463         struct drm_i915_private *dev_priv = dev->dev_private;
2464         struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
2465         struct drm_display_mode *mode = &intel_crtc->config.adjusted_mode;
2466         u32 linetime, ips_linetime;
2467
2468         if (!intel_crtc_active(crtc))
2469                 return 0;
2470
2471         /* The WM are computed with base on how long it takes to fill a single
2472          * row at the given clock rate, multiplied by 8.
2473          * */
2474         linetime = DIV_ROUND_CLOSEST(mode->htotal * 1000 * 8, mode->clock);
2475         ips_linetime = DIV_ROUND_CLOSEST(mode->htotal * 1000 * 8,
2476                                          intel_ddi_get_cdclk_freq(dev_priv));
2477
2478         return PIPE_WM_LINETIME_IPS_LINETIME(ips_linetime) |
2479                PIPE_WM_LINETIME_TIME(linetime);
2480 }
2481
2482 static void intel_read_wm_latency(struct drm_device *dev, uint16_t wm[5])
2483 {
2484         struct drm_i915_private *dev_priv = dev->dev_private;
2485
2486         if (IS_HASWELL(dev)) {
2487                 uint64_t sskpd = I915_READ64(MCH_SSKPD);
2488
2489                 wm[0] = (sskpd >> 56) & 0xFF;
2490                 if (wm[0] == 0)
2491                         wm[0] = sskpd & 0xF;
2492                 wm[1] = (sskpd >> 4) & 0xFF;
2493                 wm[2] = (sskpd >> 12) & 0xFF;
2494                 wm[3] = (sskpd >> 20) & 0x1FF;
2495                 wm[4] = (sskpd >> 32) & 0x1FF;
2496         } else if (INTEL_INFO(dev)->gen >= 6) {
2497                 uint32_t sskpd = I915_READ(MCH_SSKPD);
2498
2499                 wm[0] = (sskpd >> SSKPD_WM0_SHIFT) & SSKPD_WM_MASK;
2500                 wm[1] = (sskpd >> SSKPD_WM1_SHIFT) & SSKPD_WM_MASK;
2501                 wm[2] = (sskpd >> SSKPD_WM2_SHIFT) & SSKPD_WM_MASK;
2502                 wm[3] = (sskpd >> SSKPD_WM3_SHIFT) & SSKPD_WM_MASK;
2503         } else if (INTEL_INFO(dev)->gen >= 5) {
2504                 uint32_t mltr = I915_READ(MLTR_ILK);
2505
2506                 /* ILK primary LP0 latency is 700 ns */
2507                 wm[0] = 7;
2508                 wm[1] = (mltr >> MLTR_WM1_SHIFT) & ILK_SRLT_MASK;
2509                 wm[2] = (mltr >> MLTR_WM2_SHIFT) & ILK_SRLT_MASK;
2510         }
2511 }
2512
2513 static void intel_fixup_spr_wm_latency(struct drm_device *dev, uint16_t wm[5])
2514 {
2515         /* ILK sprite LP0 latency is 1300 ns */
2516         if (INTEL_INFO(dev)->gen == 5)
2517                 wm[0] = 13;
2518 }
2519
2520 static void intel_fixup_cur_wm_latency(struct drm_device *dev, uint16_t wm[5])
2521 {
2522         /* ILK cursor LP0 latency is 1300 ns */
2523         if (INTEL_INFO(dev)->gen == 5)
2524                 wm[0] = 13;
2525
2526         /* WaDoubleCursorLP3Latency:ivb */
2527         if (IS_IVYBRIDGE(dev))
2528                 wm[3] *= 2;
2529 }
2530
2531 static int ilk_wm_max_level(const struct drm_device *dev)
2532 {
2533         /* how many WM levels are we expecting */
2534         if (IS_HASWELL(dev))
2535                 return 4;
2536         else if (INTEL_INFO(dev)->gen >= 6)
2537                 return 3;
2538         else
2539                 return 2;
2540 }
2541
2542 static void intel_print_wm_latency(struct drm_device *dev,
2543                                    const char *name,
2544                                    const uint16_t wm[5])
2545 {
2546         int level, max_level = ilk_wm_max_level(dev);
2547
2548         for (level = 0; level <= max_level; level++) {
2549                 unsigned int latency = wm[level];
2550
2551                 if (latency == 0) {
2552                         DRM_ERROR("%s WM%d latency not provided\n",
2553                                   name, level);
2554                         continue;
2555                 }
2556
2557                 /* WM1+ latency values in 0.5us units */
2558                 if (level > 0)
2559                         latency *= 5;
2560
2561                 DRM_DEBUG_KMS("%s WM%d latency %u (%u.%u usec)\n",
2562                               name, level, wm[level],
2563                               latency / 10, latency % 10);
2564         }
2565 }
2566
2567 static void intel_setup_wm_latency(struct drm_device *dev)
2568 {
2569         struct drm_i915_private *dev_priv = dev->dev_private;
2570
2571         intel_read_wm_latency(dev, dev_priv->wm.pri_latency);
2572
2573         memcpy(dev_priv->wm.spr_latency, dev_priv->wm.pri_latency,
2574                sizeof(dev_priv->wm.pri_latency));
2575         memcpy(dev_priv->wm.cur_latency, dev_priv->wm.pri_latency,
2576                sizeof(dev_priv->wm.pri_latency));
2577
2578         intel_fixup_spr_wm_latency(dev, dev_priv->wm.spr_latency);
2579         intel_fixup_cur_wm_latency(dev, dev_priv->wm.cur_latency);
2580
2581         intel_print_wm_latency(dev, "Primary", dev_priv->wm.pri_latency);
2582         intel_print_wm_latency(dev, "Sprite", dev_priv->wm.spr_latency);
2583         intel_print_wm_latency(dev, "Cursor", dev_priv->wm.cur_latency);
2584 }
2585
2586 static void hsw_compute_wm_parameters(struct drm_crtc *crtc,
2587                                       struct hsw_pipe_wm_parameters *p,
2588                                       struct intel_wm_config *config)
2589 {
2590         struct drm_device *dev = crtc->dev;
2591         struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
2592         enum pipe pipe = intel_crtc->pipe;
2593         struct drm_plane *plane;
2594
2595         p->active = intel_crtc_active(crtc);
2596         if (p->active) {
2597                 p->pipe_htotal = intel_crtc->config.adjusted_mode.htotal;
2598                 p->pixel_rate = ilk_pipe_pixel_rate(dev, crtc);
2599                 p->pri.bytes_per_pixel = crtc->fb->bits_per_pixel / 8;
2600                 p->cur.bytes_per_pixel = 4;
2601                 p->pri.horiz_pixels = intel_crtc->config.pipe_src_w;
2602                 p->cur.horiz_pixels = 64;
2603                 /* TODO: for now, assume primary and cursor planes are always enabled. */
2604                 p->pri.enabled = true;
2605                 p->cur.enabled = true;
2606         }
2607
2608         list_for_each_entry(crtc, &dev->mode_config.crtc_list, head)
2609                 config->num_pipes_active += intel_crtc_active(crtc);
2610
2611         list_for_each_entry(plane, &dev->mode_config.plane_list, head) {
2612                 struct intel_plane *intel_plane = to_intel_plane(plane);
2613
2614                 if (intel_plane->pipe == pipe)
2615                         p->spr = intel_plane->wm;
2616
2617                 config->sprites_enabled |= intel_plane->wm.enabled;
2618                 config->sprites_scaled |= intel_plane->wm.scaled;
2619         }
2620 }
2621
2622 /* Compute new watermarks for the pipe */
2623 static bool intel_compute_pipe_wm(struct drm_crtc *crtc,
2624                                   const struct hsw_pipe_wm_parameters *params,
2625                                   struct intel_pipe_wm *pipe_wm)
2626 {
2627         struct drm_device *dev = crtc->dev;
2628         struct drm_i915_private *dev_priv = dev->dev_private;
2629         int level, max_level = ilk_wm_max_level(dev);
2630         /* LP0 watermark maximums depend on this pipe alone */
2631         struct intel_wm_config config = {
2632                 .num_pipes_active = 1,
2633                 .sprites_enabled = params->spr.enabled,
2634                 .sprites_scaled = params->spr.scaled,
2635         };
2636         struct hsw_wm_maximums max;
2637
2638         /* LP0 watermarks always use 1/2 DDB partitioning */
2639         ilk_wm_max(dev, 0, &config, INTEL_DDB_PART_1_2, &max);
2640
2641         for (level = 0; level <= max_level; level++)
2642                 ilk_compute_wm_level(dev_priv, level, params,
2643                                      &pipe_wm->wm[level]);
2644
2645         pipe_wm->linetime = hsw_compute_linetime_wm(dev, crtc);
2646
2647         /* At least LP0 must be valid */
2648         return ilk_check_wm(0, &max, &pipe_wm->wm[0]);
2649 }
2650
2651 /*
2652  * Merge the watermarks from all active pipes for a specific level.
2653  */
2654 static void ilk_merge_wm_level(struct drm_device *dev,
2655                                int level,
2656                                struct intel_wm_level *ret_wm)
2657 {
2658         const struct intel_crtc *intel_crtc;
2659
2660         list_for_each_entry(intel_crtc, &dev->mode_config.crtc_list, base.head) {
2661                 const struct intel_wm_level *wm =
2662                         &intel_crtc->wm.active.wm[level];
2663
2664                 if (!wm->enable)
2665                         return;
2666
2667                 ret_wm->pri_val = max(ret_wm->pri_val, wm->pri_val);
2668                 ret_wm->spr_val = max(ret_wm->spr_val, wm->spr_val);
2669                 ret_wm->cur_val = max(ret_wm->cur_val, wm->cur_val);
2670                 ret_wm->fbc_val = max(ret_wm->fbc_val, wm->fbc_val);
2671         }
2672
2673         ret_wm->enable = true;
2674 }
2675
2676 /*
2677  * Merge all low power watermarks for all active pipes.
2678  */
2679 static void ilk_wm_merge(struct drm_device *dev,
2680                          const struct hsw_wm_maximums *max,
2681                          struct intel_pipe_wm *merged)
2682 {
2683         int level, max_level = ilk_wm_max_level(dev);
2684
2685         merged->fbc_wm_enabled = true;
2686
2687         /* merge each WM1+ level */
2688         for (level = 1; level <= max_level; level++) {
2689                 struct intel_wm_level *wm = &merged->wm[level];
2690
2691                 ilk_merge_wm_level(dev, level, wm);
2692
2693                 if (!ilk_check_wm(level, max, wm))
2694                         break;
2695
2696                 /*
2697                  * The spec says it is preferred to disable
2698                  * FBC WMs instead of disabling a WM level.
2699                  */
2700                 if (wm->fbc_val > max->fbc) {
2701                         merged->fbc_wm_enabled = false;
2702                         wm->fbc_val = 0;
2703                 }
2704         }
2705 }
2706
2707 static int ilk_wm_lp_to_level(int wm_lp, const struct intel_pipe_wm *pipe_wm)
2708 {
2709         /* LP1,LP2,LP3 levels are either 1,2,3 or 1,3,4 */
2710         return wm_lp + (wm_lp >= 2 && pipe_wm->wm[4].enable);
2711 }
2712
2713 static void hsw_compute_wm_results(struct drm_device *dev,
2714                                    const struct intel_pipe_wm *merged,
2715                                    struct hsw_wm_values *results)
2716 {
2717         struct intel_crtc *intel_crtc;
2718         int level, wm_lp;
2719
2720         results->enable_fbc_wm = merged->fbc_wm_enabled;
2721
2722         /* LP1+ register values */
2723         for (wm_lp = 1; wm_lp <= 3; wm_lp++) {
2724                 const struct intel_wm_level *r;
2725
2726                 level = ilk_wm_lp_to_level(wm_lp, merged);
2727
2728                 r = &merged->wm[level];
2729                 if (!r->enable)
2730                         break;
2731
2732                 results->wm_lp[wm_lp - 1] = HSW_WM_LP_VAL(level * 2,
2733                                                           r->fbc_val,
2734                                                           r->pri_val,
2735                                                           r->cur_val);
2736                 results->wm_lp_spr[wm_lp - 1] = r->spr_val;
2737         }
2738
2739         /* LP0 register values */
2740         list_for_each_entry(intel_crtc, &dev->mode_config.crtc_list, base.head) {
2741                 enum pipe pipe = intel_crtc->pipe;
2742                 const struct intel_wm_level *r =
2743                         &intel_crtc->wm.active.wm[0];
2744
2745                 if (WARN_ON(!r->enable))
2746                         continue;
2747
2748                 results->wm_linetime[pipe] = intel_crtc->wm.active.linetime;
2749
2750                 results->wm_pipe[pipe] =
2751                         (r->pri_val << WM0_PIPE_PLANE_SHIFT) |
2752                         (r->spr_val << WM0_PIPE_SPRITE_SHIFT) |
2753                         r->cur_val;
2754         }
2755 }
2756
2757 /* Find the result with the highest level enabled. Check for enable_fbc_wm in
2758  * case both are at the same level. Prefer r1 in case they're the same. */
2759 static struct intel_pipe_wm *hsw_find_best_result(struct drm_device *dev,
2760                                                   struct intel_pipe_wm *r1,
2761                                                   struct intel_pipe_wm *r2)
2762 {
2763         int level, max_level = ilk_wm_max_level(dev);
2764         int level1 = 0, level2 = 0;
2765
2766         for (level = 1; level <= max_level; level++) {
2767                 if (r1->wm[level].enable)
2768                         level1 = level;
2769                 if (r2->wm[level].enable)
2770                         level2 = level;
2771         }
2772
2773         if (level1 == level2) {
2774                 if (r2->fbc_wm_enabled && !r1->fbc_wm_enabled)
2775                         return r2;
2776                 else
2777                         return r1;
2778         } else if (level1 > level2) {
2779                 return r1;
2780         } else {
2781                 return r2;
2782         }
2783 }
2784
2785 /*
2786  * The spec says we shouldn't write when we don't need, because every write
2787  * causes WMs to be re-evaluated, expending some power.
2788  */
2789 static void hsw_write_wm_values(struct drm_i915_private *dev_priv,
2790                                 struct hsw_wm_values *results,
2791                                 enum intel_ddb_partitioning partitioning)
2792 {
2793         struct hsw_wm_values previous;
2794         uint32_t val;
2795         enum intel_ddb_partitioning prev_partitioning;
2796         bool prev_enable_fbc_wm;
2797
2798         previous.wm_pipe[0] = I915_READ(WM0_PIPEA_ILK);
2799         previous.wm_pipe[1] = I915_READ(WM0_PIPEB_ILK);
2800         previous.wm_pipe[2] = I915_READ(WM0_PIPEC_IVB);
2801         previous.wm_lp[0] = I915_READ(WM1_LP_ILK);
2802         previous.wm_lp[1] = I915_READ(WM2_LP_ILK);
2803         previous.wm_lp[2] = I915_READ(WM3_LP_ILK);
2804         previous.wm_lp_spr[0] = I915_READ(WM1S_LP_ILK);
2805         previous.wm_lp_spr[1] = I915_READ(WM2S_LP_IVB);
2806         previous.wm_lp_spr[2] = I915_READ(WM3S_LP_IVB);
2807         previous.wm_linetime[0] = I915_READ(PIPE_WM_LINETIME(PIPE_A));
2808         previous.wm_linetime[1] = I915_READ(PIPE_WM_LINETIME(PIPE_B));
2809         previous.wm_linetime[2] = I915_READ(PIPE_WM_LINETIME(PIPE_C));
2810
2811         prev_partitioning = (I915_READ(WM_MISC) & WM_MISC_DATA_PARTITION_5_6) ?
2812                                 INTEL_DDB_PART_5_6 : INTEL_DDB_PART_1_2;
2813
2814         prev_enable_fbc_wm = !(I915_READ(DISP_ARB_CTL) & DISP_FBC_WM_DIS);
2815
2816         if (memcmp(results->wm_pipe, previous.wm_pipe,
2817                    sizeof(results->wm_pipe)) == 0 &&
2818             memcmp(results->wm_lp, previous.wm_lp,
2819                    sizeof(results->wm_lp)) == 0 &&
2820             memcmp(results->wm_lp_spr, previous.wm_lp_spr,
2821                    sizeof(results->wm_lp_spr)) == 0 &&
2822             memcmp(results->wm_linetime, previous.wm_linetime,
2823                    sizeof(results->wm_linetime)) == 0 &&
2824             partitioning == prev_partitioning &&
2825             results->enable_fbc_wm == prev_enable_fbc_wm)
2826                 return;
2827
2828         if (previous.wm_lp[2] != 0)
2829                 I915_WRITE(WM3_LP_ILK, 0);
2830         if (previous.wm_lp[1] != 0)
2831                 I915_WRITE(WM2_LP_ILK, 0);
2832         if (previous.wm_lp[0] != 0)
2833                 I915_WRITE(WM1_LP_ILK, 0);
2834
2835         if (previous.wm_pipe[0] != results->wm_pipe[0])
2836                 I915_WRITE(WM0_PIPEA_ILK, results->wm_pipe[0]);
2837         if (previous.wm_pipe[1] != results->wm_pipe[1])
2838                 I915_WRITE(WM0_PIPEB_ILK, results->wm_pipe[1]);
2839         if (previous.wm_pipe[2] != results->wm_pipe[2])
2840                 I915_WRITE(WM0_PIPEC_IVB, results->wm_pipe[2]);
2841
2842         if (previous.wm_linetime[0] != results->wm_linetime[0])
2843                 I915_WRITE(PIPE_WM_LINETIME(PIPE_A), results->wm_linetime[0]);
2844         if (previous.wm_linetime[1] != results->wm_linetime[1])
2845                 I915_WRITE(PIPE_WM_LINETIME(PIPE_B), results->wm_linetime[1]);
2846         if (previous.wm_linetime[2] != results->wm_linetime[2])
2847                 I915_WRITE(PIPE_WM_LINETIME(PIPE_C), results->wm_linetime[2]);
2848
2849         if (prev_partitioning != partitioning) {
2850                 val = I915_READ(WM_MISC);
2851                 if (partitioning == INTEL_DDB_PART_1_2)
2852                         val &= ~WM_MISC_DATA_PARTITION_5_6;
2853                 else
2854                         val |= WM_MISC_DATA_PARTITION_5_6;
2855                 I915_WRITE(WM_MISC, val);
2856         }
2857
2858         if (prev_enable_fbc_wm != results->enable_fbc_wm) {
2859                 val = I915_READ(DISP_ARB_CTL);
2860                 if (results->enable_fbc_wm)
2861                         val &= ~DISP_FBC_WM_DIS;
2862                 else
2863                         val |= DISP_FBC_WM_DIS;
2864                 I915_WRITE(DISP_ARB_CTL, val);
2865         }
2866
2867         if (previous.wm_lp_spr[0] != results->wm_lp_spr[0])
2868                 I915_WRITE(WM1S_LP_ILK, results->wm_lp_spr[0]);
2869         if (previous.wm_lp_spr[1] != results->wm_lp_spr[1])
2870                 I915_WRITE(WM2S_LP_IVB, results->wm_lp_spr[1]);
2871         if (previous.wm_lp_spr[2] != results->wm_lp_spr[2])
2872                 I915_WRITE(WM3S_LP_IVB, results->wm_lp_spr[2]);
2873
2874         if (results->wm_lp[0] != 0)
2875                 I915_WRITE(WM1_LP_ILK, results->wm_lp[0]);
2876         if (results->wm_lp[1] != 0)
2877                 I915_WRITE(WM2_LP_ILK, results->wm_lp[1]);
2878         if (results->wm_lp[2] != 0)
2879                 I915_WRITE(WM3_LP_ILK, results->wm_lp[2]);
2880 }
2881
2882 static void haswell_update_wm(struct drm_crtc *crtc)
2883 {
2884         struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
2885         struct drm_device *dev = crtc->dev;
2886         struct drm_i915_private *dev_priv = dev->dev_private;
2887         struct hsw_wm_maximums max;
2888         struct hsw_pipe_wm_parameters params = {};
2889         struct hsw_wm_values results = {};
2890         enum intel_ddb_partitioning partitioning;
2891         struct intel_pipe_wm pipe_wm = {};
2892         struct intel_pipe_wm lp_wm_1_2 = {}, lp_wm_5_6 = {}, *best_lp_wm;
2893         struct intel_wm_config config = {};
2894
2895         hsw_compute_wm_parameters(crtc, &params, &config);
2896
2897         intel_compute_pipe_wm(crtc, &params, &pipe_wm);
2898
2899         if (!memcmp(&intel_crtc->wm.active, &pipe_wm, sizeof(pipe_wm)))
2900                 return;
2901
2902         intel_crtc->wm.active = pipe_wm;
2903
2904         ilk_wm_max(dev, 1, &config, INTEL_DDB_PART_1_2, &max);
2905         ilk_wm_merge(dev, &max, &lp_wm_1_2);
2906
2907         /* 5/6 split only in single pipe config on IVB+ */
2908         if (INTEL_INFO(dev)->gen >= 7 && config.num_pipes_active == 1) {
2909                 ilk_wm_max(dev, 1, &config, INTEL_DDB_PART_5_6, &max);
2910                 ilk_wm_merge(dev, &max, &lp_wm_5_6);
2911
2912                 best_lp_wm = hsw_find_best_result(dev, &lp_wm_1_2, &lp_wm_5_6);
2913         } else {
2914                 best_lp_wm = &lp_wm_1_2;
2915         }
2916
2917         hsw_compute_wm_results(dev, best_lp_wm, &results);
2918
2919         partitioning = (best_lp_wm == &lp_wm_1_2) ?
2920                        INTEL_DDB_PART_1_2 : INTEL_DDB_PART_5_6;
2921
2922         hsw_write_wm_values(dev_priv, &results, partitioning);
2923 }
2924
2925 static void haswell_update_sprite_wm(struct drm_plane *plane,
2926                                      struct drm_crtc *crtc,
2927                                      uint32_t sprite_width, int pixel_size,
2928                                      bool enabled, bool scaled)
2929 {
2930         struct intel_plane *intel_plane = to_intel_plane(plane);
2931
2932         intel_plane->wm.enabled = enabled;
2933         intel_plane->wm.scaled = scaled;
2934         intel_plane->wm.horiz_pixels = sprite_width;
2935         intel_plane->wm.bytes_per_pixel = pixel_size;
2936
2937         haswell_update_wm(crtc);
2938 }
2939
2940 static bool
2941 sandybridge_compute_sprite_wm(struct drm_device *dev, int plane,
2942                               uint32_t sprite_width, int pixel_size,
2943                               const struct intel_watermark_params *display,
2944                               int display_latency_ns, int *sprite_wm)
2945 {
2946         struct drm_crtc *crtc;
2947         int clock;
2948         int entries, tlb_miss;
2949
2950         crtc = intel_get_crtc_for_plane(dev, plane);
2951         if (!intel_crtc_active(crtc)) {
2952                 *sprite_wm = display->guard_size;
2953                 return false;
2954         }
2955
2956         clock = to_intel_crtc(crtc)->config.adjusted_mode.crtc_clock;
2957
2958         /* Use the small buffer method to calculate the sprite watermark */
2959         entries = ((clock * pixel_size / 1000) * display_latency_ns) / 1000;
2960         tlb_miss = display->fifo_size*display->cacheline_size -
2961                 sprite_width * 8;
2962         if (tlb_miss > 0)
2963                 entries += tlb_miss;
2964         entries = DIV_ROUND_UP(entries, display->cacheline_size);
2965         *sprite_wm = entries + display->guard_size;
2966         if (*sprite_wm > (int)display->max_wm)
2967                 *sprite_wm = display->max_wm;
2968
2969         return true;
2970 }
2971
2972 static bool
2973 sandybridge_compute_sprite_srwm(struct drm_device *dev, int plane,
2974                                 uint32_t sprite_width, int pixel_size,
2975                                 const struct intel_watermark_params *display,
2976                                 int latency_ns, int *sprite_wm)
2977 {
2978         struct drm_crtc *crtc;
2979         unsigned long line_time_us;
2980         int clock;
2981         int line_count, line_size;
2982         int small, large;
2983         int entries;
2984
2985         if (!latency_ns) {
2986                 *sprite_wm = 0;
2987                 return false;
2988         }
2989
2990         crtc = intel_get_crtc_for_plane(dev, plane);
2991         clock = to_intel_crtc(crtc)->config.adjusted_mode.crtc_clock;
2992         if (!clock) {
2993                 *sprite_wm = 0;
2994                 return false;
2995         }
2996
2997         line_time_us = (sprite_width * 1000) / clock;
2998         if (!line_time_us) {
2999                 *sprite_wm = 0;
3000                 return false;
3001         }
3002
3003         line_count = (latency_ns / line_time_us + 1000) / 1000;
3004         line_size = sprite_width * pixel_size;
3005
3006         /* Use the minimum of the small and large buffer method for primary */
3007         small = ((clock * pixel_size / 1000) * latency_ns) / 1000;
3008         large = line_count * line_size;
3009
3010         entries = DIV_ROUND_UP(min(small, large), display->cacheline_size);
3011         *sprite_wm = entries + display->guard_size;
3012
3013         return *sprite_wm > 0x3ff ? false : true;
3014 }
3015
3016 static void sandybridge_update_sprite_wm(struct drm_plane *plane,
3017                                          struct drm_crtc *crtc,
3018                                          uint32_t sprite_width, int pixel_size,
3019                                          bool enabled, bool scaled)
3020 {
3021         struct drm_device *dev = plane->dev;
3022         struct drm_i915_private *dev_priv = dev->dev_private;
3023         int pipe = to_intel_plane(plane)->pipe;
3024         int latency = dev_priv->wm.spr_latency[0] * 100;        /* In unit 0.1us */
3025         u32 val;
3026         int sprite_wm, reg;
3027         int ret;
3028
3029         if (!enabled)
3030                 return;
3031
3032         switch (pipe) {
3033         case 0:
3034                 reg = WM0_PIPEA_ILK;
3035                 break;
3036         case 1:
3037                 reg = WM0_PIPEB_ILK;
3038                 break;
3039         case 2:
3040                 reg = WM0_PIPEC_IVB;
3041                 break;
3042         default:
3043                 return; /* bad pipe */
3044         }
3045
3046         ret = sandybridge_compute_sprite_wm(dev, pipe, sprite_width, pixel_size,
3047                                             &sandybridge_display_wm_info,
3048                                             latency, &sprite_wm);
3049         if (!ret) {
3050                 DRM_DEBUG_KMS("failed to compute sprite wm for pipe %c\n",
3051                               pipe_name(pipe));
3052                 return;
3053         }
3054
3055         val = I915_READ(reg);
3056         val &= ~WM0_PIPE_SPRITE_MASK;
3057         I915_WRITE(reg, val | (sprite_wm << WM0_PIPE_SPRITE_SHIFT));
3058         DRM_DEBUG_KMS("sprite watermarks For pipe %c - %d\n", pipe_name(pipe), sprite_wm);
3059
3060
3061         ret = sandybridge_compute_sprite_srwm(dev, pipe, sprite_width,
3062                                               pixel_size,
3063                                               &sandybridge_display_srwm_info,
3064                                               dev_priv->wm.spr_latency[1] * 500,
3065                                               &sprite_wm);
3066         if (!ret) {
3067                 DRM_DEBUG_KMS("failed to compute sprite lp1 wm on pipe %c\n",
3068                               pipe_name(pipe));
3069                 return;
3070         }
3071         I915_WRITE(WM1S_LP_ILK, sprite_wm);
3072
3073         /* Only IVB has two more LP watermarks for sprite */
3074         if (!IS_IVYBRIDGE(dev))
3075                 return;
3076
3077         ret = sandybridge_compute_sprite_srwm(dev, pipe, sprite_width,
3078                                               pixel_size,
3079                                               &sandybridge_display_srwm_info,
3080                                               dev_priv->wm.spr_latency[2] * 500,
3081                                               &sprite_wm);
3082         if (!ret) {
3083                 DRM_DEBUG_KMS("failed to compute sprite lp2 wm on pipe %c\n",
3084                               pipe_name(pipe));
3085                 return;
3086         }
3087         I915_WRITE(WM2S_LP_IVB, sprite_wm);
3088
3089         ret = sandybridge_compute_sprite_srwm(dev, pipe, sprite_width,
3090                                               pixel_size,
3091                                               &sandybridge_display_srwm_info,
3092                                               dev_priv->wm.spr_latency[3] * 500,
3093                                               &sprite_wm);
3094         if (!ret) {
3095                 DRM_DEBUG_KMS("failed to compute sprite lp3 wm on pipe %c\n",
3096                               pipe_name(pipe));
3097                 return;
3098         }
3099         I915_WRITE(WM3S_LP_IVB, sprite_wm);
3100 }
3101
3102 /**
3103  * intel_update_watermarks - update FIFO watermark values based on current modes
3104  *
3105  * Calculate watermark values for the various WM regs based on current mode
3106  * and plane configuration.
3107  *
3108  * There are several cases to deal with here:
3109  *   - normal (i.e. non-self-refresh)
3110  *   - self-refresh (SR) mode
3111  *   - lines are large relative to FIFO size (buffer can hold up to 2)
3112  *   - lines are small relative to FIFO size (buffer can hold more than 2
3113  *     lines), so need to account for TLB latency
3114  *
3115  *   The normal calculation is:
3116  *     watermark = dotclock * bytes per pixel * latency
3117  *   where latency is platform & configuration dependent (we assume pessimal
3118  *   values here).
3119  *
3120  *   The SR calculation is:
3121  *     watermark = (trunc(latency/line time)+1) * surface width *
3122  *       bytes per pixel
3123  *   where
3124  *     line time = htotal / dotclock
3125  *     surface width = hdisplay for normal plane and 64 for cursor
3126  *   and latency is assumed to be high, as above.
3127  *
3128  * The final value programmed to the register should always be rounded up,
3129  * and include an extra 2 entries to account for clock crossings.
3130  *
3131  * We don't use the sprite, so we can ignore that.  And on Crestline we have
3132  * to set the non-SR watermarks to 8.
3133  */
3134 void intel_update_watermarks(struct drm_crtc *crtc)
3135 {
3136         struct drm_i915_private *dev_priv = crtc->dev->dev_private;
3137
3138         if (dev_priv->display.update_wm)
3139                 dev_priv->display.update_wm(crtc);
3140 }
3141
3142 void intel_update_sprite_watermarks(struct drm_plane *plane,
3143                                     struct drm_crtc *crtc,
3144                                     uint32_t sprite_width, int pixel_size,
3145                                     bool enabled, bool scaled)
3146 {
3147         struct drm_i915_private *dev_priv = plane->dev->dev_private;
3148
3149         if (dev_priv->display.update_sprite_wm)
3150                 dev_priv->display.update_sprite_wm(plane, crtc, sprite_width,
3151                                                    pixel_size, enabled, scaled);
3152 }
3153
3154 static struct drm_i915_gem_object *
3155 intel_alloc_context_page(struct drm_device *dev)
3156 {
3157         struct drm_i915_gem_object *ctx;
3158         int ret;
3159
3160         WARN_ON(!mutex_is_locked(&dev->struct_mutex));
3161
3162         ctx = i915_gem_alloc_object(dev, 4096);
3163         if (!ctx) {
3164                 DRM_DEBUG("failed to alloc power context, RC6 disabled\n");
3165                 return NULL;
3166         }
3167
3168         ret = i915_gem_obj_ggtt_pin(ctx, 4096, true, false);
3169         if (ret) {
3170                 DRM_ERROR("failed to pin power context: %d\n", ret);
3171                 goto err_unref;
3172         }
3173
3174         ret = i915_gem_object_set_to_gtt_domain(ctx, 1);
3175         if (ret) {
3176                 DRM_ERROR("failed to set-domain on power context: %d\n", ret);
3177                 goto err_unpin;
3178         }
3179
3180         return ctx;
3181
3182 err_unpin:
3183         i915_gem_object_unpin(ctx);
3184 err_unref:
3185         drm_gem_object_unreference(&ctx->base);
3186         return NULL;
3187 }
3188
3189 /**
3190  * Lock protecting IPS related data structures
3191  */
3192 DEFINE_SPINLOCK(mchdev_lock);
3193
3194 /* Global for IPS driver to get at the current i915 device. Protected by
3195  * mchdev_lock. */
3196 static struct drm_i915_private *i915_mch_dev;
3197
3198 bool ironlake_set_drps(struct drm_device *dev, u8 val)
3199 {
3200         struct drm_i915_private *dev_priv = dev->dev_private;
3201         u16 rgvswctl;
3202
3203         assert_spin_locked(&mchdev_lock);
3204
3205         rgvswctl = I915_READ16(MEMSWCTL);
3206         if (rgvswctl & MEMCTL_CMD_STS) {
3207                 DRM_DEBUG("gpu busy, RCS change rejected\n");
3208                 return false; /* still busy with another command */
3209         }
3210
3211         rgvswctl = (MEMCTL_CMD_CHFREQ << MEMCTL_CMD_SHIFT) |
3212                 (val << MEMCTL_FREQ_SHIFT) | MEMCTL_SFCAVM;
3213         I915_WRITE16(MEMSWCTL, rgvswctl);
3214         POSTING_READ16(MEMSWCTL);
3215
3216         rgvswctl |= MEMCTL_CMD_STS;
3217         I915_WRITE16(MEMSWCTL, rgvswctl);
3218
3219         return true;
3220 }
3221
3222 static void ironlake_enable_drps(struct drm_device *dev)
3223 {
3224         struct drm_i915_private *dev_priv = dev->dev_private;
3225         u32 rgvmodectl = I915_READ(MEMMODECTL);
3226         u8 fmax, fmin, fstart, vstart;
3227
3228         spin_lock_irq(&mchdev_lock);
3229
3230         /* Enable temp reporting */
3231         I915_WRITE16(PMMISC, I915_READ(PMMISC) | MCPPCE_EN);
3232         I915_WRITE16(TSC1, I915_READ(TSC1) | TSE);
3233
3234         /* 100ms RC evaluation intervals */
3235         I915_WRITE(RCUPEI, 100000);
3236         I915_WRITE(RCDNEI, 100000);
3237
3238         /* Set max/min thresholds to 90ms and 80ms respectively */
3239         I915_WRITE(RCBMAXAVG, 90000);
3240         I915_WRITE(RCBMINAVG, 80000);
3241
3242         I915_WRITE(MEMIHYST, 1);
3243
3244         /* Set up min, max, and cur for interrupt handling */
3245         fmax = (rgvmodectl & MEMMODE_FMAX_MASK) >> MEMMODE_FMAX_SHIFT;
3246         fmin = (rgvmodectl & MEMMODE_FMIN_MASK);
3247         fstart = (rgvmodectl & MEMMODE_FSTART_MASK) >>
3248                 MEMMODE_FSTART_SHIFT;
3249
3250         vstart = (I915_READ(PXVFREQ_BASE + (fstart * 4)) & PXVFREQ_PX_MASK) >>
3251                 PXVFREQ_PX_SHIFT;
3252
3253         dev_priv->ips.fmax = fmax; /* IPS callback will increase this */
3254         dev_priv->ips.fstart = fstart;
3255
3256         dev_priv->ips.max_delay = fstart;
3257         dev_priv->ips.min_delay = fmin;
3258         dev_priv->ips.cur_delay = fstart;
3259
3260         DRM_DEBUG_DRIVER("fmax: %d, fmin: %d, fstart: %d\n",
3261                          fmax, fmin, fstart);
3262
3263         I915_WRITE(MEMINTREN, MEMINT_CX_SUPR_EN | MEMINT_EVAL_CHG_EN);
3264
3265         /*
3266          * Interrupts will be enabled in ironlake_irq_postinstall
3267          */
3268
3269         I915_WRITE(VIDSTART, vstart);
3270         POSTING_READ(VIDSTART);
3271
3272         rgvmodectl |= MEMMODE_SWMODE_EN;
3273         I915_WRITE(MEMMODECTL, rgvmodectl);
3274
3275         if (wait_for_atomic((I915_READ(MEMSWCTL) & MEMCTL_CMD_STS) == 0, 10))
3276                 DRM_ERROR("stuck trying to change perf mode\n");
3277         mdelay(1);
3278
3279         ironlake_set_drps(dev, fstart);
3280
3281         dev_priv->ips.last_count1 = I915_READ(0x112e4) + I915_READ(0x112e8) +
3282                 I915_READ(0x112e0);
3283         dev_priv->ips.last_time1 = jiffies_to_msecs(jiffies);
3284         dev_priv->ips.last_count2 = I915_READ(0x112f4);
3285         getrawmonotonic(&dev_priv->ips.last_time2);
3286
3287         spin_unlock_irq(&mchdev_lock);
3288 }
3289
3290 static void ironlake_disable_drps(struct drm_device *dev)
3291 {
3292         struct drm_i915_private *dev_priv = dev->dev_private;
3293         u16 rgvswctl;
3294
3295         spin_lock_irq(&mchdev_lock);
3296
3297         rgvswctl = I915_READ16(MEMSWCTL);
3298
3299         /* Ack interrupts, disable EFC interrupt */
3300         I915_WRITE(MEMINTREN, I915_READ(MEMINTREN) & ~MEMINT_EVAL_CHG_EN);
3301         I915_WRITE(MEMINTRSTS, MEMINT_EVAL_CHG);
3302         I915_WRITE(DEIER, I915_READ(DEIER) & ~DE_PCU_EVENT);
3303         I915_WRITE(DEIIR, DE_PCU_EVENT);
3304         I915_WRITE(DEIMR, I915_READ(DEIMR) | DE_PCU_EVENT);
3305
3306         /* Go back to the starting frequency */
3307         ironlake_set_drps(dev, dev_priv->ips.fstart);
3308         mdelay(1);
3309         rgvswctl |= MEMCTL_CMD_STS;
3310         I915_WRITE(MEMSWCTL, rgvswctl);
3311         mdelay(1);
3312
3313         spin_unlock_irq(&mchdev_lock);
3314 }
3315
3316 /* There's a funny hw issue where the hw returns all 0 when reading from
3317  * GEN6_RP_INTERRUPT_LIMITS. Hence we always need to compute the desired value
3318  * ourselves, instead of doing a rmw cycle (which might result in us clearing
3319  * all limits and the gpu stuck at whatever frequency it is at atm).
3320  */
3321 static u32 gen6_rps_limits(struct drm_i915_private *dev_priv, u8 *val)
3322 {
3323         u32 limits;
3324
3325         limits = 0;
3326
3327         if (*val >= dev_priv->rps.max_delay)
3328                 *val = dev_priv->rps.max_delay;
3329         limits |= dev_priv->rps.max_delay << 24;
3330
3331         /* Only set the down limit when we've reached the lowest level to avoid
3332          * getting more interrupts, otherwise leave this clear. This prevents a
3333          * race in the hw when coming out of rc6: There's a tiny window where
3334          * the hw runs at the minimal clock before selecting the desired
3335          * frequency, if the down threshold expires in that window we will not
3336          * receive a down interrupt. */
3337         if (*val <= dev_priv->rps.min_delay) {
3338                 *val = dev_priv->rps.min_delay;
3339                 limits |= dev_priv->rps.min_delay << 16;
3340         }
3341
3342         return limits;
3343 }
3344
3345 static void gen6_set_rps_thresholds(struct drm_i915_private *dev_priv, u8 val)
3346 {
3347         int new_power;
3348
3349         new_power = dev_priv->rps.power;
3350         switch (dev_priv->rps.power) {
3351         case LOW_POWER:
3352                 if (val > dev_priv->rps.rpe_delay + 1 && val > dev_priv->rps.cur_delay)
3353                         new_power = BETWEEN;
3354                 break;
3355
3356         case BETWEEN:
3357                 if (val <= dev_priv->rps.rpe_delay && val < dev_priv->rps.cur_delay)
3358                         new_power = LOW_POWER;
3359                 else if (val >= dev_priv->rps.rp0_delay && val > dev_priv->rps.cur_delay)
3360                         new_power = HIGH_POWER;
3361                 break;
3362
3363         case HIGH_POWER:
3364                 if (val < (dev_priv->rps.rp1_delay + dev_priv->rps.rp0_delay) >> 1 && val < dev_priv->rps.cur_delay)
3365                         new_power = BETWEEN;
3366                 break;
3367         }
3368         /* Max/min bins are special */
3369         if (val == dev_priv->rps.min_delay)
3370                 new_power = LOW_POWER;
3371         if (val == dev_priv->rps.max_delay)
3372                 new_power = HIGH_POWER;
3373         if (new_power == dev_priv->rps.power)
3374                 return;
3375
3376         /* Note the units here are not exactly 1us, but 1280ns. */
3377         switch (new_power) {
3378         case LOW_POWER:
3379                 /* Upclock if more than 95% busy over 16ms */
3380                 I915_WRITE(GEN6_RP_UP_EI, 12500);
3381                 I915_WRITE(GEN6_RP_UP_THRESHOLD, 11800);
3382
3383                 /* Downclock if less than 85% busy over 32ms */
3384                 I915_WRITE(GEN6_RP_DOWN_EI, 25000);
3385                 I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 21250);
3386
3387                 I915_WRITE(GEN6_RP_CONTROL,
3388                            GEN6_RP_MEDIA_TURBO |
3389                            GEN6_RP_MEDIA_HW_NORMAL_MODE |
3390                            GEN6_RP_MEDIA_IS_GFX |
3391                            GEN6_RP_ENABLE |
3392                            GEN6_RP_UP_BUSY_AVG |
3393                            GEN6_RP_DOWN_IDLE_AVG);
3394                 break;
3395
3396         case BETWEEN:
3397                 /* Upclock if more than 90% busy over 13ms */
3398                 I915_WRITE(GEN6_RP_UP_EI, 10250);
3399                 I915_WRITE(GEN6_RP_UP_THRESHOLD, 9225);
3400
3401                 /* Downclock if less than 75% busy over 32ms */
3402                 I915_WRITE(GEN6_RP_DOWN_EI, 25000);
3403                 I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 18750);
3404
3405                 I915_WRITE(GEN6_RP_CONTROL,
3406                            GEN6_RP_MEDIA_TURBO |
3407                            GEN6_RP_MEDIA_HW_NORMAL_MODE |
3408                            GEN6_RP_MEDIA_IS_GFX |
3409                            GEN6_RP_ENABLE |
3410                            GEN6_RP_UP_BUSY_AVG |
3411                            GEN6_RP_DOWN_IDLE_AVG);
3412                 break;
3413
3414         case HIGH_POWER:
3415                 /* Upclock if more than 85% busy over 10ms */
3416                 I915_WRITE(GEN6_RP_UP_EI, 8000);
3417                 I915_WRITE(GEN6_RP_UP_THRESHOLD, 6800);
3418
3419                 /* Downclock if less than 60% busy over 32ms */
3420                 I915_WRITE(GEN6_RP_DOWN_EI, 25000);
3421                 I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 15000);
3422
3423                 I915_WRITE(GEN6_RP_CONTROL,
3424                            GEN6_RP_MEDIA_TURBO |
3425                            GEN6_RP_MEDIA_HW_NORMAL_MODE |
3426                            GEN6_RP_MEDIA_IS_GFX |
3427                            GEN6_RP_ENABLE |
3428                            GEN6_RP_UP_BUSY_AVG |
3429                            GEN6_RP_DOWN_IDLE_AVG);
3430                 break;
3431         }
3432
3433         dev_priv->rps.power = new_power;
3434         dev_priv->rps.last_adj = 0;
3435 }
3436
3437 void gen6_set_rps(struct drm_device *dev, u8 val)
3438 {
3439         struct drm_i915_private *dev_priv = dev->dev_private;
3440         u32 limits = gen6_rps_limits(dev_priv, &val);
3441
3442         WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
3443         WARN_ON(val > dev_priv->rps.max_delay);
3444         WARN_ON(val < dev_priv->rps.min_delay);
3445
3446         if (val == dev_priv->rps.cur_delay)
3447                 return;
3448
3449         gen6_set_rps_thresholds(dev_priv, val);
3450
3451         if (IS_HASWELL(dev))
3452                 I915_WRITE(GEN6_RPNSWREQ,
3453                            HSW_FREQUENCY(val));
3454         else
3455                 I915_WRITE(GEN6_RPNSWREQ,
3456                            GEN6_FREQUENCY(val) |
3457                            GEN6_OFFSET(0) |
3458                            GEN6_AGGRESSIVE_TURBO);
3459
3460         /* Make sure we continue to get interrupts
3461          * until we hit the minimum or maximum frequencies.
3462          */
3463         I915_WRITE(GEN6_RP_INTERRUPT_LIMITS, limits);
3464
3465         POSTING_READ(GEN6_RPNSWREQ);
3466
3467         dev_priv->rps.cur_delay = val;
3468
3469         trace_intel_gpu_freq_change(val * 50);
3470 }
3471
3472 void gen6_rps_idle(struct drm_i915_private *dev_priv)
3473 {
3474         mutex_lock(&dev_priv->rps.hw_lock);
3475         if (dev_priv->rps.enabled) {
3476                 if (dev_priv->info->is_valleyview)
3477                         valleyview_set_rps(dev_priv->dev, dev_priv->rps.min_delay);
3478                 else
3479                         gen6_set_rps(dev_priv->dev, dev_priv->rps.min_delay);
3480                 dev_priv->rps.last_adj = 0;
3481         }
3482         mutex_unlock(&dev_priv->rps.hw_lock);
3483 }
3484
3485 void gen6_rps_boost(struct drm_i915_private *dev_priv)
3486 {
3487         mutex_lock(&dev_priv->rps.hw_lock);
3488         if (dev_priv->rps.enabled) {
3489                 if (dev_priv->info->is_valleyview)
3490                         valleyview_set_rps(dev_priv->dev, dev_priv->rps.max_delay);
3491                 else
3492                         gen6_set_rps(dev_priv->dev, dev_priv->rps.max_delay);
3493                 dev_priv->rps.last_adj = 0;
3494         }
3495         mutex_unlock(&dev_priv->rps.hw_lock);
3496 }
3497
3498 /*
3499  * Wait until the previous freq change has completed,
3500  * or the timeout elapsed, and then update our notion
3501  * of the current GPU frequency.
3502  */
3503 static void vlv_update_rps_cur_delay(struct drm_i915_private *dev_priv)
3504 {
3505         u32 pval;
3506
3507         WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
3508
3509         if (wait_for(((pval = vlv_punit_read(dev_priv, PUNIT_REG_GPU_FREQ_STS)) & GENFREQSTATUS) == 0, 10))
3510                 DRM_DEBUG_DRIVER("timed out waiting for Punit\n");
3511
3512         pval >>= 8;
3513
3514         if (pval != dev_priv->rps.cur_delay)
3515                 DRM_DEBUG_DRIVER("Punit overrode GPU freq: %d MHz (%u) requested, but got %d Mhz (%u)\n",
3516                                  vlv_gpu_freq(dev_priv->mem_freq, dev_priv->rps.cur_delay),
3517                                  dev_priv->rps.cur_delay,
3518                                  vlv_gpu_freq(dev_priv->mem_freq, pval), pval);
3519
3520         dev_priv->rps.cur_delay = pval;
3521 }
3522
3523 void valleyview_set_rps(struct drm_device *dev, u8 val)
3524 {
3525         struct drm_i915_private *dev_priv = dev->dev_private;
3526
3527         gen6_rps_limits(dev_priv, &val);
3528
3529         WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
3530         WARN_ON(val > dev_priv->rps.max_delay);
3531         WARN_ON(val < dev_priv->rps.min_delay);
3532
3533         vlv_update_rps_cur_delay(dev_priv);
3534
3535         DRM_DEBUG_DRIVER("GPU freq request from %d MHz (%u) to %d MHz (%u)\n",
3536                          vlv_gpu_freq(dev_priv->mem_freq,
3537                                       dev_priv->rps.cur_delay),
3538                          dev_priv->rps.cur_delay,
3539                          vlv_gpu_freq(dev_priv->mem_freq, val), val);
3540
3541         if (val == dev_priv->rps.cur_delay)
3542                 return;
3543
3544         vlv_punit_write(dev_priv, PUNIT_REG_GPU_FREQ_REQ, val);
3545
3546         dev_priv->rps.cur_delay = val;
3547
3548         trace_intel_gpu_freq_change(vlv_gpu_freq(dev_priv->mem_freq, val));
3549 }
3550
3551 static void gen6_disable_rps_interrupts(struct drm_device *dev)
3552 {
3553         struct drm_i915_private *dev_priv = dev->dev_private;
3554
3555         I915_WRITE(GEN6_PMINTRMSK, 0xffffffff);
3556         I915_WRITE(GEN6_PMIER, I915_READ(GEN6_PMIER) & ~GEN6_PM_RPS_EVENTS);
3557         /* Complete PM interrupt masking here doesn't race with the rps work
3558          * item again unmasking PM interrupts because that is using a different
3559          * register (PMIMR) to mask PM interrupts. The only risk is in leaving
3560          * stale bits in PMIIR and PMIMR which gen6_enable_rps will clean up. */
3561
3562         spin_lock_irq(&dev_priv->irq_lock);
3563         dev_priv->rps.pm_iir = 0;
3564         spin_unlock_irq(&dev_priv->irq_lock);
3565
3566         I915_WRITE(GEN6_PMIIR, GEN6_PM_RPS_EVENTS);
3567 }
3568
3569 static void gen6_disable_rps(struct drm_device *dev)
3570 {
3571         struct drm_i915_private *dev_priv = dev->dev_private;
3572
3573         I915_WRITE(GEN6_RC_CONTROL, 0);
3574         I915_WRITE(GEN6_RPNSWREQ, 1 << 31);
3575
3576         gen6_disable_rps_interrupts(dev);
3577 }
3578
3579 static void valleyview_disable_rps(struct drm_device *dev)
3580 {
3581         struct drm_i915_private *dev_priv = dev->dev_private;
3582
3583         I915_WRITE(GEN6_RC_CONTROL, 0);
3584
3585         gen6_disable_rps_interrupts(dev);
3586
3587         if (dev_priv->vlv_pctx) {
3588                 drm_gem_object_unreference(&dev_priv->vlv_pctx->base);
3589                 dev_priv->vlv_pctx = NULL;
3590         }
3591 }
3592
3593 int intel_enable_rc6(const struct drm_device *dev)
3594 {
3595         /* No RC6 before Ironlake */
3596         if (INTEL_INFO(dev)->gen < 5)
3597                 return 0;
3598
3599         /* Respect the kernel parameter if it is set */
3600         if (i915_enable_rc6 >= 0)
3601                 return i915_enable_rc6;
3602
3603         /* Disable RC6 on Ironlake */
3604         if (INTEL_INFO(dev)->gen == 5)
3605                 return 0;
3606
3607         if (IS_HASWELL(dev)) {
3608                 DRM_DEBUG_DRIVER("Haswell: only RC6 available\n");
3609                 return INTEL_RC6_ENABLE;
3610         }
3611
3612         /* snb/ivb have more than one rc6 state. */
3613         if (INTEL_INFO(dev)->gen == 6) {
3614                 DRM_DEBUG_DRIVER("Sandybridge: deep RC6 disabled\n");
3615                 return INTEL_RC6_ENABLE;
3616         }
3617
3618         DRM_DEBUG_DRIVER("RC6 and deep RC6 enabled\n");
3619         return (INTEL_RC6_ENABLE | INTEL_RC6p_ENABLE);
3620 }
3621
3622 static void gen6_enable_rps_interrupts(struct drm_device *dev)
3623 {
3624         struct drm_i915_private *dev_priv = dev->dev_private;
3625         u32 enabled_intrs;
3626
3627         spin_lock_irq(&dev_priv->irq_lock);
3628         WARN_ON(dev_priv->rps.pm_iir);
3629         snb_enable_pm_irq(dev_priv, GEN6_PM_RPS_EVENTS);
3630         I915_WRITE(GEN6_PMIIR, GEN6_PM_RPS_EVENTS);
3631         spin_unlock_irq(&dev_priv->irq_lock);
3632
3633         /* only unmask PM interrupts we need. Mask all others. */
3634         enabled_intrs = GEN6_PM_RPS_EVENTS;
3635
3636         /* IVB and SNB hard hangs on looping batchbuffer
3637          * if GEN6_PM_UP_EI_EXPIRED is masked.
3638          */
3639         if (INTEL_INFO(dev)->gen <= 7 && !IS_HASWELL(dev))
3640                 enabled_intrs |= GEN6_PM_RP_UP_EI_EXPIRED;
3641
3642         I915_WRITE(GEN6_PMINTRMSK, ~enabled_intrs);
3643 }
3644
3645 static void gen6_enable_rps(struct drm_device *dev)
3646 {
3647         struct drm_i915_private *dev_priv = dev->dev_private;
3648         struct intel_ring_buffer *ring;
3649         u32 rp_state_cap;
3650         u32 gt_perf_status;
3651         u32 rc6vids, pcu_mbox, rc6_mask = 0;
3652         u32 gtfifodbg;
3653         int rc6_mode;
3654         int i, ret;
3655
3656         WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
3657
3658         /* Here begins a magic sequence of register writes to enable
3659          * auto-downclocking.
3660          *
3661          * Perhaps there might be some value in exposing these to
3662          * userspace...
3663          */
3664         I915_WRITE(GEN6_RC_STATE, 0);
3665
3666         /* Clear the DBG now so we don't confuse earlier errors */
3667         if ((gtfifodbg = I915_READ(GTFIFODBG))) {
3668                 DRM_ERROR("GT fifo had a previous error %x\n", gtfifodbg);
3669                 I915_WRITE(GTFIFODBG, gtfifodbg);
3670         }
3671
3672         gen6_gt_force_wake_get(dev_priv);
3673
3674         rp_state_cap = I915_READ(GEN6_RP_STATE_CAP);
3675         gt_perf_status = I915_READ(GEN6_GT_PERF_STATUS);
3676
3677         /* In units of 50MHz */
3678         dev_priv->rps.hw_max = dev_priv->rps.max_delay = rp_state_cap & 0xff;
3679         dev_priv->rps.min_delay = (rp_state_cap >> 16) & 0xff;
3680         dev_priv->rps.rp1_delay = (rp_state_cap >>  8) & 0xff;
3681         dev_priv->rps.rp0_delay = (rp_state_cap >>  0) & 0xff;
3682         dev_priv->rps.rpe_delay = dev_priv->rps.rp1_delay;
3683         dev_priv->rps.cur_delay = 0;
3684
3685         /* disable the counters and set deterministic thresholds */
3686         I915_WRITE(GEN6_RC_CONTROL, 0);
3687
3688         I915_WRITE(GEN6_RC1_WAKE_RATE_LIMIT, 1000 << 16);
3689         I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 40 << 16 | 30);
3690         I915_WRITE(GEN6_RC6pp_WAKE_RATE_LIMIT, 30);
3691         I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000);
3692         I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25);
3693
3694         for_each_ring(ring, dev_priv, i)
3695                 I915_WRITE(RING_MAX_IDLE(ring->mmio_base), 10);
3696
3697         I915_WRITE(GEN6_RC_SLEEP, 0);
3698         I915_WRITE(GEN6_RC1e_THRESHOLD, 1000);
3699         if (INTEL_INFO(dev)->gen <= 6 || IS_IVYBRIDGE(dev))
3700                 I915_WRITE(GEN6_RC6_THRESHOLD, 125000);
3701         else
3702                 I915_WRITE(GEN6_RC6_THRESHOLD, 50000);
3703         I915_WRITE(GEN6_RC6p_THRESHOLD, 150000);
3704         I915_WRITE(GEN6_RC6pp_THRESHOLD, 64000); /* unused */
3705
3706         /* Check if we are enabling RC6 */
3707         rc6_mode = intel_enable_rc6(dev_priv->dev);
3708         if (rc6_mode & INTEL_RC6_ENABLE)
3709                 rc6_mask |= GEN6_RC_CTL_RC6_ENABLE;
3710
3711         /* We don't use those on Haswell */
3712         if (!IS_HASWELL(dev)) {
3713                 if (rc6_mode & INTEL_RC6p_ENABLE)
3714                         rc6_mask |= GEN6_RC_CTL_RC6p_ENABLE;
3715
3716                 if (rc6_mode & INTEL_RC6pp_ENABLE)
3717                         rc6_mask |= GEN6_RC_CTL_RC6pp_ENABLE;
3718         }
3719
3720         DRM_INFO("Enabling RC6 states: RC6 %s, RC6p %s, RC6pp %s\n",
3721                         (rc6_mask & GEN6_RC_CTL_RC6_ENABLE) ? "on" : "off",
3722                         (rc6_mask & GEN6_RC_CTL_RC6p_ENABLE) ? "on" : "off",
3723                         (rc6_mask & GEN6_RC_CTL_RC6pp_ENABLE) ? "on" : "off");
3724
3725         I915_WRITE(GEN6_RC_CONTROL,
3726                    rc6_mask |
3727                    GEN6_RC_CTL_EI_MODE(1) |
3728                    GEN6_RC_CTL_HW_ENABLE);
3729
3730         /* Power down if completely idle for over 50ms */
3731         I915_WRITE(GEN6_RP_DOWN_TIMEOUT, 50000);
3732         I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 10);
3733
3734         ret = sandybridge_pcode_write(dev_priv, GEN6_PCODE_WRITE_MIN_FREQ_TABLE, 0);
3735         if (!ret) {
3736                 pcu_mbox = 0;
3737                 ret = sandybridge_pcode_read(dev_priv, GEN6_READ_OC_PARAMS, &pcu_mbox);
3738                 if (!ret && (pcu_mbox & (1<<31))) { /* OC supported */
3739                         DRM_DEBUG_DRIVER("Overclocking supported. Max: %dMHz, Overclock max: %dMHz\n",
3740                                          (dev_priv->rps.max_delay & 0xff) * 50,
3741                                          (pcu_mbox & 0xff) * 50);
3742                         dev_priv->rps.hw_max = pcu_mbox & 0xff;
3743                 }
3744         } else {
3745                 DRM_DEBUG_DRIVER("Failed to set the min frequency\n");
3746         }
3747
3748         dev_priv->rps.power = HIGH_POWER; /* force a reset */
3749         gen6_set_rps(dev_priv->dev, dev_priv->rps.min_delay);
3750
3751         gen6_enable_rps_interrupts(dev);
3752
3753         rc6vids = 0;
3754         ret = sandybridge_pcode_read(dev_priv, GEN6_PCODE_READ_RC6VIDS, &rc6vids);
3755         if (IS_GEN6(dev) && ret) {
3756                 DRM_DEBUG_DRIVER("Couldn't check for BIOS workaround\n");
3757         } else if (IS_GEN6(dev) && (GEN6_DECODE_RC6_VID(rc6vids & 0xff) < 450)) {
3758                 DRM_DEBUG_DRIVER("You should update your BIOS. Correcting minimum rc6 voltage (%dmV->%dmV)\n",
3759                           GEN6_DECODE_RC6_VID(rc6vids & 0xff), 450);
3760                 rc6vids &= 0xffff00;
3761                 rc6vids |= GEN6_ENCODE_RC6_VID(450);
3762                 ret = sandybridge_pcode_write(dev_priv, GEN6_PCODE_WRITE_RC6VIDS, rc6vids);
3763                 if (ret)
3764                         DRM_ERROR("Couldn't fix incorrect rc6 voltage\n");
3765         }
3766
3767         gen6_gt_force_wake_put(dev_priv);
3768 }
3769
3770 void gen6_update_ring_freq(struct drm_device *dev)
3771 {
3772         struct drm_i915_private *dev_priv = dev->dev_private;
3773         int min_freq = 15;
3774         unsigned int gpu_freq;
3775         unsigned int max_ia_freq, min_ring_freq;
3776         int scaling_factor = 180;
3777         struct cpufreq_policy *policy;
3778
3779         WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
3780
3781         policy = cpufreq_cpu_get(0);
3782         if (policy) {
3783                 max_ia_freq = policy->cpuinfo.max_freq;
3784                 cpufreq_cpu_put(policy);
3785         } else {
3786                 /*
3787                  * Default to measured freq if none found, PCU will ensure we
3788                  * don't go over
3789                  */
3790                 max_ia_freq = tsc_khz;
3791         }
3792
3793         /* Convert from kHz to MHz */
3794         max_ia_freq /= 1000;
3795
3796         min_ring_freq = I915_READ(MCHBAR_MIRROR_BASE_SNB + DCLK) & 0xf;
3797         /* convert DDR frequency from units of 266.6MHz to bandwidth */
3798         min_ring_freq = mult_frac(min_ring_freq, 8, 3);
3799
3800         /*
3801          * For each potential GPU frequency, load a ring frequency we'd like
3802          * to use for memory access.  We do this by specifying the IA frequency
3803          * the PCU should use as a reference to determine the ring frequency.
3804          */
3805         for (gpu_freq = dev_priv->rps.max_delay; gpu_freq >= dev_priv->rps.min_delay;
3806              gpu_freq--) {
3807                 int diff = dev_priv->rps.max_delay - gpu_freq;
3808                 unsigned int ia_freq = 0, ring_freq = 0;
3809
3810                 if (IS_HASWELL(dev)) {
3811                         ring_freq = mult_frac(gpu_freq, 5, 4);
3812                         ring_freq = max(min_ring_freq, ring_freq);
3813                         /* leave ia_freq as the default, chosen by cpufreq */
3814                 } else {
3815                         /* On older processors, there is no separate ring
3816                          * clock domain, so in order to boost the bandwidth
3817                          * of the ring, we need to upclock the CPU (ia_freq).
3818                          *
3819                          * For GPU frequencies less than 750MHz,
3820                          * just use the lowest ring freq.
3821                          */
3822                         if (gpu_freq < min_freq)
3823                                 ia_freq = 800;
3824                         else
3825                                 ia_freq = max_ia_freq - ((diff * scaling_factor) / 2);
3826                         ia_freq = DIV_ROUND_CLOSEST(ia_freq, 100);
3827                 }
3828
3829                 sandybridge_pcode_write(dev_priv,
3830                                         GEN6_PCODE_WRITE_MIN_FREQ_TABLE,
3831                                         ia_freq << GEN6_PCODE_FREQ_IA_RATIO_SHIFT |
3832                                         ring_freq << GEN6_PCODE_FREQ_RING_RATIO_SHIFT |
3833                                         gpu_freq);
3834         }
3835 }
3836
3837 int valleyview_rps_max_freq(struct drm_i915_private *dev_priv)
3838 {
3839         u32 val, rp0;
3840
3841         val = vlv_nc_read(dev_priv, IOSF_NC_FB_GFX_FREQ_FUSE);
3842
3843         rp0 = (val & FB_GFX_MAX_FREQ_FUSE_MASK) >> FB_GFX_MAX_FREQ_FUSE_SHIFT;
3844         /* Clamp to max */
3845         rp0 = min_t(u32, rp0, 0xea);
3846
3847         return rp0;
3848 }
3849
3850 static int valleyview_rps_rpe_freq(struct drm_i915_private *dev_priv)
3851 {
3852         u32 val, rpe;
3853
3854         val = vlv_nc_read(dev_priv, IOSF_NC_FB_GFX_FMAX_FUSE_LO);
3855         rpe = (val & FB_FMAX_VMIN_FREQ_LO_MASK) >> FB_FMAX_VMIN_FREQ_LO_SHIFT;
3856         val = vlv_nc_read(dev_priv, IOSF_NC_FB_GFX_FMAX_FUSE_HI);
3857         rpe |= (val & FB_FMAX_VMIN_FREQ_HI_MASK) << 5;
3858
3859         return rpe;
3860 }
3861
3862 int valleyview_rps_min_freq(struct drm_i915_private *dev_priv)
3863 {
3864         return vlv_punit_read(dev_priv, PUNIT_REG_GPU_LFM) & 0xff;
3865 }
3866
3867 static void valleyview_setup_pctx(struct drm_device *dev)
3868 {
3869         struct drm_i915_private *dev_priv = dev->dev_private;
3870         struct drm_i915_gem_object *pctx;
3871         unsigned long pctx_paddr;
3872         u32 pcbr;
3873         int pctx_size = 24*1024;
3874
3875         pcbr = I915_READ(VLV_PCBR);
3876         if (pcbr) {
3877                 /* BIOS set it up already, grab the pre-alloc'd space */
3878                 int pcbr_offset;
3879
3880                 pcbr_offset = (pcbr & (~4095)) - dev_priv->mm.stolen_base;
3881                 pctx = i915_gem_object_create_stolen_for_preallocated(dev_priv->dev,
3882                                                                       pcbr_offset,
3883                                                                       I915_GTT_OFFSET_NONE,
3884                                                                       pctx_size);
3885                 goto out;
3886         }
3887
3888         /*
3889          * From the Gunit register HAS:
3890          * The Gfx driver is expected to program this register and ensure
3891          * proper allocation within Gfx stolen memory.  For example, this
3892          * register should be programmed such than the PCBR range does not
3893          * overlap with other ranges, such as the frame buffer, protected
3894          * memory, or any other relevant ranges.
3895          */
3896         pctx = i915_gem_object_create_stolen(dev, pctx_size);
3897         if (!pctx) {
3898                 DRM_DEBUG("not enough stolen space for PCTX, disabling\n");
3899                 return;
3900         }
3901
3902         pctx_paddr = dev_priv->mm.stolen_base + pctx->stolen->start;
3903         I915_WRITE(VLV_PCBR, pctx_paddr);
3904
3905 out:
3906         dev_priv->vlv_pctx = pctx;
3907 }
3908
3909 static void valleyview_enable_rps(struct drm_device *dev)
3910 {
3911         struct drm_i915_private *dev_priv = dev->dev_private;
3912         struct intel_ring_buffer *ring;
3913         u32 gtfifodbg, val, rc6_mode = 0;
3914         int i;
3915
3916         WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
3917
3918         if ((gtfifodbg = I915_READ(GTFIFODBG))) {
3919                 DRM_DEBUG_DRIVER("GT fifo had a previous error %x\n",
3920                                  gtfifodbg);
3921                 I915_WRITE(GTFIFODBG, gtfifodbg);
3922         }
3923
3924         valleyview_setup_pctx(dev);
3925
3926         gen6_gt_force_wake_get(dev_priv);
3927
3928         I915_WRITE(GEN6_RP_UP_THRESHOLD, 59400);
3929         I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 245000);
3930         I915_WRITE(GEN6_RP_UP_EI, 66000);
3931         I915_WRITE(GEN6_RP_DOWN_EI, 350000);
3932
3933         I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 10);
3934
3935         I915_WRITE(GEN6_RP_CONTROL,
3936                    GEN6_RP_MEDIA_TURBO |
3937                    GEN6_RP_MEDIA_HW_NORMAL_MODE |
3938                    GEN6_RP_MEDIA_IS_GFX |
3939                    GEN6_RP_ENABLE |
3940                    GEN6_RP_UP_BUSY_AVG |
3941                    GEN6_RP_DOWN_IDLE_CONT);
3942
3943         I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 0x00280000);
3944         I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000);
3945         I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25);
3946
3947         for_each_ring(ring, dev_priv, i)
3948                 I915_WRITE(RING_MAX_IDLE(ring->mmio_base), 10);
3949
3950         I915_WRITE(GEN6_RC6_THRESHOLD, 0xc350);
3951
3952         /* allows RC6 residency counter to work */
3953         I915_WRITE(VLV_COUNTER_CONTROL,
3954                    _MASKED_BIT_ENABLE(VLV_COUNT_RANGE_HIGH |
3955                                       VLV_MEDIA_RC6_COUNT_EN |
3956                                       VLV_RENDER_RC6_COUNT_EN));
3957         if (intel_enable_rc6(dev) & INTEL_RC6_ENABLE)
3958                 rc6_mode = GEN7_RC_CTL_TO_MODE;
3959         I915_WRITE(GEN6_RC_CONTROL, rc6_mode);
3960
3961         val = vlv_punit_read(dev_priv, PUNIT_REG_GPU_FREQ_STS);
3962         switch ((val >> 6) & 3) {
3963         case 0:
3964         case 1:
3965                 dev_priv->mem_freq = 800;
3966                 break;
3967         case 2:
3968                 dev_priv->mem_freq = 1066;
3969                 break;
3970         case 3:
3971                 dev_priv->mem_freq = 1333;
3972                 break;
3973         }
3974         DRM_DEBUG_DRIVER("DDR speed: %d MHz", dev_priv->mem_freq);
3975
3976         DRM_DEBUG_DRIVER("GPLL enabled? %s\n", val & 0x10 ? "yes" : "no");
3977         DRM_DEBUG_DRIVER("GPU status: 0x%08x\n", val);
3978
3979         dev_priv->rps.cur_delay = (val >> 8) & 0xff;
3980         DRM_DEBUG_DRIVER("current GPU freq: %d MHz (%u)\n",
3981                          vlv_gpu_freq(dev_priv->mem_freq,
3982                                       dev_priv->rps.cur_delay),
3983                          dev_priv->rps.cur_delay);
3984
3985         dev_priv->rps.max_delay = valleyview_rps_max_freq(dev_priv);
3986         dev_priv->rps.hw_max = dev_priv->rps.max_delay;
3987         DRM_DEBUG_DRIVER("max GPU freq: %d MHz (%u)\n",
3988                          vlv_gpu_freq(dev_priv->mem_freq,
3989                                       dev_priv->rps.max_delay),
3990                          dev_priv->rps.max_delay);
3991
3992         dev_priv->rps.rpe_delay = valleyview_rps_rpe_freq(dev_priv);
3993         DRM_DEBUG_DRIVER("RPe GPU freq: %d MHz (%u)\n",
3994                          vlv_gpu_freq(dev_priv->mem_freq,
3995                                       dev_priv->rps.rpe_delay),
3996                          dev_priv->rps.rpe_delay);
3997
3998         dev_priv->rps.min_delay = valleyview_rps_min_freq(dev_priv);
3999         DRM_DEBUG_DRIVER("min GPU freq: %d MHz (%u)\n",
4000                          vlv_gpu_freq(dev_priv->mem_freq,
4001                                       dev_priv->rps.min_delay),
4002                          dev_priv->rps.min_delay);
4003
4004         DRM_DEBUG_DRIVER("setting GPU freq to %d MHz (%u)\n",
4005                          vlv_gpu_freq(dev_priv->mem_freq,
4006                                       dev_priv->rps.rpe_delay),
4007                          dev_priv->rps.rpe_delay);
4008
4009         valleyview_set_rps(dev_priv->dev, dev_priv->rps.rpe_delay);
4010
4011         gen6_enable_rps_interrupts(dev);
4012
4013         gen6_gt_force_wake_put(dev_priv);
4014 }
4015
4016 void ironlake_teardown_rc6(struct drm_device *dev)
4017 {
4018         struct drm_i915_private *dev_priv = dev->dev_private;
4019
4020         if (dev_priv->ips.renderctx) {
4021                 i915_gem_object_unpin(dev_priv->ips.renderctx);
4022                 drm_gem_object_unreference(&dev_priv->ips.renderctx->base);
4023                 dev_priv->ips.renderctx = NULL;
4024         }
4025
4026         if (dev_priv->ips.pwrctx) {
4027                 i915_gem_object_unpin(dev_priv->ips.pwrctx);
4028                 drm_gem_object_unreference(&dev_priv->ips.pwrctx->base);
4029                 dev_priv->ips.pwrctx = NULL;
4030         }
4031 }
4032
4033 static void ironlake_disable_rc6(struct drm_device *dev)
4034 {
4035         struct drm_i915_private *dev_priv = dev->dev_private;
4036
4037         if (I915_READ(PWRCTXA)) {
4038                 /* Wake the GPU, prevent RC6, then restore RSTDBYCTL */
4039                 I915_WRITE(RSTDBYCTL, I915_READ(RSTDBYCTL) | RCX_SW_EXIT);
4040                 wait_for(((I915_READ(RSTDBYCTL) & RSX_STATUS_MASK) == RSX_STATUS_ON),
4041                          50);
4042
4043                 I915_WRITE(PWRCTXA, 0);
4044                 POSTING_READ(PWRCTXA);
4045
4046                 I915_WRITE(RSTDBYCTL, I915_READ(RSTDBYCTL) & ~RCX_SW_EXIT);
4047                 POSTING_READ(RSTDBYCTL);
4048         }
4049 }
4050
4051 static int ironlake_setup_rc6(struct drm_device *dev)
4052 {
4053         struct drm_i915_private *dev_priv = dev->dev_private;
4054
4055         if (dev_priv->ips.renderctx == NULL)
4056                 dev_priv->ips.renderctx = intel_alloc_context_page(dev);
4057         if (!dev_priv->ips.renderctx)
4058                 return -ENOMEM;
4059
4060         if (dev_priv->ips.pwrctx == NULL)
4061                 dev_priv->ips.pwrctx = intel_alloc_context_page(dev);
4062         if (!dev_priv->ips.pwrctx) {
4063                 ironlake_teardown_rc6(dev);
4064                 return -ENOMEM;
4065         }
4066
4067         return 0;
4068 }
4069
4070 static void ironlake_enable_rc6(struct drm_device *dev)
4071 {
4072         struct drm_i915_private *dev_priv = dev->dev_private;
4073         struct intel_ring_buffer *ring = &dev_priv->ring[RCS];
4074         bool was_interruptible;
4075         int ret;
4076
4077         /* rc6 disabled by default due to repeated reports of hanging during
4078          * boot and resume.
4079          */
4080         if (!intel_enable_rc6(dev))
4081                 return;
4082
4083         WARN_ON(!mutex_is_locked(&dev->struct_mutex));
4084
4085         ret = ironlake_setup_rc6(dev);
4086         if (ret)
4087                 return;
4088
4089         was_interruptible = dev_priv->mm.interruptible;
4090         dev_priv->mm.interruptible = false;
4091
4092         /*
4093          * GPU can automatically power down the render unit if given a page
4094          * to save state.
4095          */
4096         ret = intel_ring_begin(ring, 6);
4097         if (ret) {
4098                 ironlake_teardown_rc6(dev);
4099                 dev_priv->mm.interruptible = was_interruptible;
4100                 return;
4101         }
4102
4103         intel_ring_emit(ring, MI_SUSPEND_FLUSH | MI_SUSPEND_FLUSH_EN);
4104         intel_ring_emit(ring, MI_SET_CONTEXT);
4105         intel_ring_emit(ring, i915_gem_obj_ggtt_offset(dev_priv->ips.renderctx) |
4106                         MI_MM_SPACE_GTT |
4107                         MI_SAVE_EXT_STATE_EN |
4108                         MI_RESTORE_EXT_STATE_EN |
4109                         MI_RESTORE_INHIBIT);
4110         intel_ring_emit(ring, MI_SUSPEND_FLUSH);
4111         intel_ring_emit(ring, MI_NOOP);
4112         intel_ring_emit(ring, MI_FLUSH);
4113         intel_ring_advance(ring);
4114
4115         /*
4116          * Wait for the command parser to advance past MI_SET_CONTEXT. The HW
4117          * does an implicit flush, combined with MI_FLUSH above, it should be
4118          * safe to assume that renderctx is valid
4119          */
4120         ret = intel_ring_idle(ring);
4121         dev_priv->mm.interruptible = was_interruptible;
4122         if (ret) {
4123                 DRM_ERROR("failed to enable ironlake power savings\n");
4124                 ironlake_teardown_rc6(dev);
4125                 return;
4126         }
4127
4128         I915_WRITE(PWRCTXA, i915_gem_obj_ggtt_offset(dev_priv->ips.pwrctx) | PWRCTX_EN);
4129         I915_WRITE(RSTDBYCTL, I915_READ(RSTDBYCTL) & ~RCX_SW_EXIT);
4130 }
4131
4132 static unsigned long intel_pxfreq(u32 vidfreq)
4133 {
4134         unsigned long freq;
4135         int div = (vidfreq & 0x3f0000) >> 16;
4136         int post = (vidfreq & 0x3000) >> 12;
4137         int pre = (vidfreq & 0x7);
4138
4139         if (!pre)
4140                 return 0;
4141
4142         freq = ((div * 133333) / ((1<<post) * pre));
4143
4144         return freq;
4145 }
4146
4147 static const struct cparams {
4148         u16 i;
4149         u16 t;
4150         u16 m;
4151         u16 c;
4152 } cparams[] = {
4153         { 1, 1333, 301, 28664 },
4154         { 1, 1066, 294, 24460 },
4155         { 1, 800, 294, 25192 },
4156         { 0, 1333, 276, 27605 },
4157         { 0, 1066, 276, 27605 },
4158         { 0, 800, 231, 23784 },
4159 };
4160
4161 static unsigned long __i915_chipset_val(struct drm_i915_private *dev_priv)
4162 {
4163         u64 total_count, diff, ret;
4164         u32 count1, count2, count3, m = 0, c = 0;
4165         unsigned long now = jiffies_to_msecs(jiffies), diff1;
4166         int i;
4167
4168         assert_spin_locked(&mchdev_lock);
4169
4170         diff1 = now - dev_priv->ips.last_time1;
4171
4172         /* Prevent division-by-zero if we are asking too fast.
4173          * Also, we don't get interesting results if we are polling
4174          * faster than once in 10ms, so just return the saved value
4175          * in such cases.
4176          */
4177         if (diff1 <= 10)
4178                 return dev_priv->ips.chipset_power;
4179
4180         count1 = I915_READ(DMIEC);
4181         count2 = I915_READ(DDREC);
4182         count3 = I915_READ(CSIEC);
4183
4184         total_count = count1 + count2 + count3;
4185
4186         /* FIXME: handle per-counter overflow */
4187         if (total_count < dev_priv->ips.last_count1) {
4188                 diff = ~0UL - dev_priv->ips.last_count1;
4189                 diff += total_count;
4190         } else {
4191                 diff = total_count - dev_priv->ips.last_count1;
4192         }
4193
4194         for (i = 0; i < ARRAY_SIZE(cparams); i++) {
4195                 if (cparams[i].i == dev_priv->ips.c_m &&
4196                     cparams[i].t == dev_priv->ips.r_t) {
4197                         m = cparams[i].m;
4198                         c = cparams[i].c;
4199                         break;
4200                 }
4201         }
4202
4203         diff = div_u64(diff, diff1);
4204         ret = ((m * diff) + c);
4205         ret = div_u64(ret, 10);
4206
4207         dev_priv->ips.last_count1 = total_count;
4208         dev_priv->ips.last_time1 = now;
4209
4210         dev_priv->ips.chipset_power = ret;
4211
4212         return ret;
4213 }
4214
4215 unsigned long i915_chipset_val(struct drm_i915_private *dev_priv)
4216 {
4217         unsigned long val;
4218
4219         if (dev_priv->info->gen != 5)
4220                 return 0;
4221
4222         spin_lock_irq(&mchdev_lock);
4223
4224         val = __i915_chipset_val(dev_priv);
4225
4226         spin_unlock_irq(&mchdev_lock);
4227
4228         return val;
4229 }
4230
4231 unsigned long i915_mch_val(struct drm_i915_private *dev_priv)
4232 {
4233         unsigned long m, x, b;
4234         u32 tsfs;
4235
4236         tsfs = I915_READ(TSFS);
4237
4238         m = ((tsfs & TSFS_SLOPE_MASK) >> TSFS_SLOPE_SHIFT);
4239         x = I915_READ8(TR1);
4240
4241         b = tsfs & TSFS_INTR_MASK;
4242
4243         return ((m * x) / 127) - b;
4244 }
4245
4246 static u16 pvid_to_extvid(struct drm_i915_private *dev_priv, u8 pxvid)
4247 {
4248         static const struct v_table {
4249                 u16 vd; /* in .1 mil */
4250                 u16 vm; /* in .1 mil */
4251         } v_table[] = {
4252                 { 0, 0, },
4253                 { 375, 0, },
4254                 { 500, 0, },
4255                 { 625, 0, },
4256                 { 750, 0, },
4257                 { 875, 0, },
4258                 { 1000, 0, },
4259                 { 1125, 0, },
4260                 { 4125, 3000, },
4261                 { 4125, 3000, },
4262                 { 4125, 3000, },
4263                 { 4125, 3000, },
4264                 { 4125, 3000, },
4265                 { 4125, 3000, },
4266                 { 4125, 3000, },
4267                 { 4125, 3000, },
4268                 { 4125, 3000, },
4269                 { 4125, 3000, },
4270                 { 4125, 3000, },
4271                 { 4125, 3000, },
4272                 { 4125, 3000, },
4273                 { 4125, 3000, },
4274                 { 4125, 3000, },
4275                 { 4125, 3000, },
4276                 { 4125, 3000, },
4277                 { 4125, 3000, },
4278                 { 4125, 3000, },
4279                 { 4125, 3000, },
4280                 { 4125, 3000, },
4281                 { 4125, 3000, },
4282                 { 4125, 3000, },
4283                 { 4125, 3000, },
4284                 { 4250, 3125, },
4285                 { 4375, 3250, },
4286                 { 4500, 3375, },
4287                 { 4625, 3500, },
4288                 { 4750, 3625, },
4289                 { 4875, 3750, },
4290                 { 5000, 3875, },
4291                 { 5125, 4000, },
4292                 { 5250, 4125, },
4293                 { 5375, 4250, },
4294                 { 5500, 4375, },
4295                 { 5625, 4500, },
4296                 { 5750, 4625, },
4297                 { 5875, 4750, },
4298                 { 6000, 4875, },
4299                 { 6125, 5000, },
4300                 { 6250, 5125, },
4301                 { 6375, 5250, },
4302                 { 6500, 5375, },
4303                 { 6625, 5500, },
4304                 { 6750, 5625, },
4305                 { 6875, 5750, },
4306                 { 7000, 5875, },
4307                 { 7125, 6000, },
4308                 { 7250, 6125, },
4309                 { 7375, 6250, },
4310                 { 7500, 6375, },
4311                 { 7625, 6500, },
4312                 { 7750, 6625, },
4313                 { 7875, 6750, },
4314                 { 8000, 6875, },
4315                 { 8125, 7000, },
4316                 { 8250, 7125, },
4317                 { 8375, 7250, },
4318                 { 8500, 7375, },
4319                 { 8625, 7500, },
4320                 { 8750, 7625, },
4321                 { 8875, 7750, },
4322                 { 9000, 7875, },
4323                 { 9125, 8000, },
4324                 { 9250, 8125, },
4325                 { 9375, 8250, },
4326                 { 9500, 8375, },
4327                 { 9625, 8500, },
4328                 { 9750, 8625, },
4329                 { 9875, 8750, },
4330                 { 10000, 8875, },
4331                 { 10125, 9000, },
4332                 { 10250, 9125, },
4333                 { 10375, 9250, },
4334                 { 10500, 9375, },
4335                 { 10625, 9500, },
4336                 { 10750, 9625, },
4337                 { 10875, 9750, },
4338                 { 11000, 9875, },
4339                 { 11125, 10000, },
4340                 { 11250, 10125, },
4341                 { 11375, 10250, },
4342                 { 11500, 10375, },
4343                 { 11625, 10500, },
4344                 { 11750, 10625, },
4345                 { 11875, 10750, },
4346                 { 12000, 10875, },
4347                 { 12125, 11000, },
4348                 { 12250, 11125, },
4349                 { 12375, 11250, },
4350                 { 12500, 11375, },
4351                 { 12625, 11500, },
4352                 { 12750, 11625, },
4353                 { 12875, 11750, },
4354                 { 13000, 11875, },
4355                 { 13125, 12000, },
4356                 { 13250, 12125, },
4357                 { 13375, 12250, },
4358                 { 13500, 12375, },
4359                 { 13625, 12500, },
4360                 { 13750, 12625, },
4361                 { 13875, 12750, },
4362                 { 14000, 12875, },
4363                 { 14125, 13000, },
4364                 { 14250, 13125, },
4365                 { 14375, 13250, },
4366                 { 14500, 13375, },
4367                 { 14625, 13500, },
4368                 { 14750, 13625, },
4369                 { 14875, 13750, },
4370                 { 15000, 13875, },
4371                 { 15125, 14000, },
4372                 { 15250, 14125, },
4373                 { 15375, 14250, },
4374                 { 15500, 14375, },
4375                 { 15625, 14500, },
4376                 { 15750, 14625, },
4377                 { 15875, 14750, },
4378                 { 16000, 14875, },
4379                 { 16125, 15000, },
4380         };
4381         if (dev_priv->info->is_mobile)
4382                 return v_table[pxvid].vm;
4383         else
4384                 return v_table[pxvid].vd;
4385 }
4386
4387 static void __i915_update_gfx_val(struct drm_i915_private *dev_priv)
4388 {
4389         struct timespec now, diff1;
4390         u64 diff;
4391         unsigned long diffms;
4392         u32 count;
4393
4394         assert_spin_locked(&mchdev_lock);
4395
4396         getrawmonotonic(&now);
4397         diff1 = timespec_sub(now, dev_priv->ips.last_time2);
4398
4399         /* Don't divide by 0 */
4400         diffms = diff1.tv_sec * 1000 + diff1.tv_nsec / 1000000;
4401         if (!diffms)
4402                 return;
4403
4404         count = I915_READ(GFXEC);
4405
4406         if (count < dev_priv->ips.last_count2) {
4407                 diff = ~0UL - dev_priv->ips.last_count2;
4408                 diff += count;
4409         } else {
4410                 diff = count - dev_priv->ips.last_count2;
4411         }
4412
4413         dev_priv->ips.last_count2 = count;
4414         dev_priv->ips.last_time2 = now;
4415
4416         /* More magic constants... */
4417         diff = diff * 1181;
4418         diff = div_u64(diff, diffms * 10);
4419         dev_priv->ips.gfx_power = diff;
4420 }
4421
4422 void i915_update_gfx_val(struct drm_i915_private *dev_priv)
4423 {
4424         if (dev_priv->info->gen != 5)
4425                 return;
4426
4427         spin_lock_irq(&mchdev_lock);
4428
4429         __i915_update_gfx_val(dev_priv);
4430
4431         spin_unlock_irq(&mchdev_lock);
4432 }
4433
4434 static unsigned long __i915_gfx_val(struct drm_i915_private *dev_priv)
4435 {
4436         unsigned long t, corr, state1, corr2, state2;
4437         u32 pxvid, ext_v;
4438
4439         assert_spin_locked(&mchdev_lock);
4440
4441         pxvid = I915_READ(PXVFREQ_BASE + (dev_priv->rps.cur_delay * 4));
4442         pxvid = (pxvid >> 24) & 0x7f;
4443         ext_v = pvid_to_extvid(dev_priv, pxvid);
4444
4445         state1 = ext_v;
4446
4447         t = i915_mch_val(dev_priv);
4448
4449         /* Revel in the empirically derived constants */
4450
4451         /* Correction factor in 1/100000 units */
4452         if (t > 80)
4453                 corr = ((t * 2349) + 135940);
4454         else if (t >= 50)
4455                 corr = ((t * 964) + 29317);
4456         else /* < 50 */
4457                 corr = ((t * 301) + 1004);
4458
4459         corr = corr * ((150142 * state1) / 10000 - 78642);
4460         corr /= 100000;
4461         corr2 = (corr * dev_priv->ips.corr);
4462
4463         state2 = (corr2 * state1) / 10000;
4464         state2 /= 100; /* convert to mW */
4465
4466         __i915_update_gfx_val(dev_priv);
4467
4468         return dev_priv->ips.gfx_power + state2;
4469 }
4470
4471 unsigned long i915_gfx_val(struct drm_i915_private *dev_priv)
4472 {
4473         unsigned long val;
4474
4475         if (dev_priv->info->gen != 5)
4476                 return 0;
4477
4478         spin_lock_irq(&mchdev_lock);
4479
4480         val = __i915_gfx_val(dev_priv);
4481
4482         spin_unlock_irq(&mchdev_lock);
4483
4484         return val;
4485 }
4486
4487 /**
4488  * i915_read_mch_val - return value for IPS use
4489  *
4490  * Calculate and return a value for the IPS driver to use when deciding whether
4491  * we have thermal and power headroom to increase CPU or GPU power budget.
4492  */
4493 unsigned long i915_read_mch_val(void)
4494 {
4495         struct drm_i915_private *dev_priv;
4496         unsigned long chipset_val, graphics_val, ret = 0;
4497
4498         spin_lock_irq(&mchdev_lock);
4499         if (!i915_mch_dev)
4500                 goto out_unlock;
4501         dev_priv = i915_mch_dev;
4502
4503         chipset_val = __i915_chipset_val(dev_priv);
4504         graphics_val = __i915_gfx_val(dev_priv);
4505
4506         ret = chipset_val + graphics_val;
4507
4508 out_unlock:
4509         spin_unlock_irq(&mchdev_lock);
4510
4511         return ret;
4512 }
4513 EXPORT_SYMBOL_GPL(i915_read_mch_val);
4514
4515 /**
4516  * i915_gpu_raise - raise GPU frequency limit
4517  *
4518  * Raise the limit; IPS indicates we have thermal headroom.
4519  */
4520 bool i915_gpu_raise(void)
4521 {
4522         struct drm_i915_private *dev_priv;
4523         bool ret = true;
4524
4525         spin_lock_irq(&mchdev_lock);
4526         if (!i915_mch_dev) {
4527                 ret = false;
4528                 goto out_unlock;
4529         }
4530         dev_priv = i915_mch_dev;
4531
4532         if (dev_priv->ips.max_delay > dev_priv->ips.fmax)
4533                 dev_priv->ips.max_delay--;
4534
4535 out_unlock:
4536         spin_unlock_irq(&mchdev_lock);
4537
4538         return ret;
4539 }
4540 EXPORT_SYMBOL_GPL(i915_gpu_raise);
4541
4542 /**
4543  * i915_gpu_lower - lower GPU frequency limit
4544  *
4545  * IPS indicates we're close to a thermal limit, so throttle back the GPU
4546  * frequency maximum.
4547  */
4548 bool i915_gpu_lower(void)
4549 {
4550         struct drm_i915_private *dev_priv;
4551         bool ret = true;
4552
4553         spin_lock_irq(&mchdev_lock);
4554         if (!i915_mch_dev) {
4555                 ret = false;
4556                 goto out_unlock;
4557         }
4558         dev_priv = i915_mch_dev;
4559
4560         if (dev_priv->ips.max_delay < dev_priv->ips.min_delay)
4561                 dev_priv->ips.max_delay++;
4562
4563 out_unlock:
4564         spin_unlock_irq(&mchdev_lock);
4565
4566         return ret;
4567 }
4568 EXPORT_SYMBOL_GPL(i915_gpu_lower);
4569
4570 /**
4571  * i915_gpu_busy - indicate GPU business to IPS
4572  *
4573  * Tell the IPS driver whether or not the GPU is busy.
4574  */
4575 bool i915_gpu_busy(void)
4576 {
4577         struct drm_i915_private *dev_priv;
4578         struct intel_ring_buffer *ring;
4579         bool ret = false;
4580         int i;
4581
4582         spin_lock_irq(&mchdev_lock);
4583         if (!i915_mch_dev)
4584                 goto out_unlock;
4585         dev_priv = i915_mch_dev;
4586
4587         for_each_ring(ring, dev_priv, i)
4588                 ret |= !list_empty(&ring->request_list);
4589
4590 out_unlock:
4591         spin_unlock_irq(&mchdev_lock);
4592
4593         return ret;
4594 }
4595 EXPORT_SYMBOL_GPL(i915_gpu_busy);
4596
4597 /**
4598  * i915_gpu_turbo_disable - disable graphics turbo
4599  *
4600  * Disable graphics turbo by resetting the max frequency and setting the
4601  * current frequency to the default.
4602  */
4603 bool i915_gpu_turbo_disable(void)
4604 {
4605         struct drm_i915_private *dev_priv;
4606         bool ret = true;
4607
4608         spin_lock_irq(&mchdev_lock);
4609         if (!i915_mch_dev) {
4610                 ret = false;
4611                 goto out_unlock;
4612         }
4613         dev_priv = i915_mch_dev;
4614
4615         dev_priv->ips.max_delay = dev_priv->ips.fstart;
4616
4617         if (!ironlake_set_drps(dev_priv->dev, dev_priv->ips.fstart))
4618                 ret = false;
4619
4620 out_unlock:
4621         spin_unlock_irq(&mchdev_lock);
4622
4623         return ret;
4624 }
4625 EXPORT_SYMBOL_GPL(i915_gpu_turbo_disable);
4626
4627 /**
4628  * Tells the intel_ips driver that the i915 driver is now loaded, if
4629  * IPS got loaded first.
4630  *
4631  * This awkward dance is so that neither module has to depend on the
4632  * other in order for IPS to do the appropriate communication of
4633  * GPU turbo limits to i915.
4634  */
4635 static void
4636 ips_ping_for_i915_load(void)
4637 {
4638         void (*link)(void);
4639
4640         link = symbol_get(ips_link_to_i915_driver);
4641         if (link) {
4642                 link();
4643                 symbol_put(ips_link_to_i915_driver);
4644         }
4645 }
4646
4647 void intel_gpu_ips_init(struct drm_i915_private *dev_priv)
4648 {
4649         /* We only register the i915 ips part with intel-ips once everything is
4650          * set up, to avoid intel-ips sneaking in and reading bogus values. */
4651         spin_lock_irq(&mchdev_lock);
4652         i915_mch_dev = dev_priv;
4653         spin_unlock_irq(&mchdev_lock);
4654
4655         ips_ping_for_i915_load();
4656 }
4657
4658 void intel_gpu_ips_teardown(void)
4659 {
4660         spin_lock_irq(&mchdev_lock);
4661         i915_mch_dev = NULL;
4662         spin_unlock_irq(&mchdev_lock);
4663 }
4664 static void intel_init_emon(struct drm_device *dev)
4665 {
4666         struct drm_i915_private *dev_priv = dev->dev_private;
4667         u32 lcfuse;
4668         u8 pxw[16];
4669         int i;
4670
4671         /* Disable to program */
4672         I915_WRITE(ECR, 0);
4673         POSTING_READ(ECR);
4674
4675         /* Program energy weights for various events */
4676         I915_WRITE(SDEW, 0x15040d00);
4677         I915_WRITE(CSIEW0, 0x007f0000);
4678         I915_WRITE(CSIEW1, 0x1e220004);
4679         I915_WRITE(CSIEW2, 0x04000004);
4680
4681         for (i = 0; i < 5; i++)
4682                 I915_WRITE(PEW + (i * 4), 0);
4683         for (i = 0; i < 3; i++)
4684                 I915_WRITE(DEW + (i * 4), 0);
4685
4686         /* Program P-state weights to account for frequency power adjustment */
4687         for (i = 0; i < 16; i++) {
4688                 u32 pxvidfreq = I915_READ(PXVFREQ_BASE + (i * 4));
4689                 unsigned long freq = intel_pxfreq(pxvidfreq);
4690                 unsigned long vid = (pxvidfreq & PXVFREQ_PX_MASK) >>
4691                         PXVFREQ_PX_SHIFT;
4692                 unsigned long val;
4693
4694                 val = vid * vid;
4695                 val *= (freq / 1000);
4696                 val *= 255;
4697                 val /= (127*127*900);
4698                 if (val > 0xff)
4699                         DRM_ERROR("bad pxval: %ld\n", val);
4700                 pxw[i] = val;
4701         }
4702         /* Render standby states get 0 weight */
4703         pxw[14] = 0;
4704         pxw[15] = 0;
4705
4706         for (i = 0; i < 4; i++) {
4707                 u32 val = (pxw[i*4] << 24) | (pxw[(i*4)+1] << 16) |
4708                         (pxw[(i*4)+2] << 8) | (pxw[(i*4)+3]);
4709                 I915_WRITE(PXW + (i * 4), val);
4710         }
4711
4712         /* Adjust magic regs to magic values (more experimental results) */
4713         I915_WRITE(OGW0, 0);
4714         I915_WRITE(OGW1, 0);
4715         I915_WRITE(EG0, 0x00007f00);
4716         I915_WRITE(EG1, 0x0000000e);
4717         I915_WRITE(EG2, 0x000e0000);
4718         I915_WRITE(EG3, 0x68000300);
4719         I915_WRITE(EG4, 0x42000000);
4720         I915_WRITE(EG5, 0x00140031);
4721         I915_WRITE(EG6, 0);
4722         I915_WRITE(EG7, 0);
4723
4724         for (i = 0; i < 8; i++)
4725                 I915_WRITE(PXWL + (i * 4), 0);
4726
4727         /* Enable PMON + select events */
4728         I915_WRITE(ECR, 0x80000019);
4729
4730         lcfuse = I915_READ(LCFUSE02);
4731
4732         dev_priv->ips.corr = (lcfuse & LCFUSE_HIV_MASK);
4733 }
4734
4735 void intel_disable_gt_powersave(struct drm_device *dev)
4736 {
4737         struct drm_i915_private *dev_priv = dev->dev_private;
4738
4739         /* Interrupts should be disabled already to avoid re-arming. */
4740         WARN_ON(dev->irq_enabled);
4741
4742         if (IS_IRONLAKE_M(dev)) {
4743                 ironlake_disable_drps(dev);
4744                 ironlake_disable_rc6(dev);
4745         } else if (INTEL_INFO(dev)->gen >= 6) {
4746                 cancel_delayed_work_sync(&dev_priv->rps.delayed_resume_work);
4747                 cancel_work_sync(&dev_priv->rps.work);
4748                 mutex_lock(&dev_priv->rps.hw_lock);
4749                 if (IS_VALLEYVIEW(dev))
4750                         valleyview_disable_rps(dev);
4751                 else
4752                         gen6_disable_rps(dev);
4753                 dev_priv->rps.enabled = false;
4754                 mutex_unlock(&dev_priv->rps.hw_lock);
4755         }
4756 }
4757
4758 static void intel_gen6_powersave_work(struct work_struct *work)
4759 {
4760         struct drm_i915_private *dev_priv =
4761                 container_of(work, struct drm_i915_private,
4762                              rps.delayed_resume_work.work);
4763         struct drm_device *dev = dev_priv->dev;
4764
4765         mutex_lock(&dev_priv->rps.hw_lock);
4766
4767         if (IS_VALLEYVIEW(dev)) {
4768                 valleyview_enable_rps(dev);
4769         } else {
4770                 gen6_enable_rps(dev);
4771                 gen6_update_ring_freq(dev);
4772         }
4773         dev_priv->rps.enabled = true;
4774         mutex_unlock(&dev_priv->rps.hw_lock);
4775 }
4776
4777 void intel_enable_gt_powersave(struct drm_device *dev)
4778 {
4779         struct drm_i915_private *dev_priv = dev->dev_private;
4780
4781         if (IS_IRONLAKE_M(dev)) {
4782                 ironlake_enable_drps(dev);
4783                 ironlake_enable_rc6(dev);
4784                 intel_init_emon(dev);
4785         } else if (IS_GEN6(dev) || IS_GEN7(dev)) {
4786                 /*
4787                  * PCU communication is slow and this doesn't need to be
4788                  * done at any specific time, so do this out of our fast path
4789                  * to make resume and init faster.
4790                  */
4791                 schedule_delayed_work(&dev_priv->rps.delayed_resume_work,
4792                                       round_jiffies_up_relative(HZ));
4793         }
4794 }
4795
4796 static void ibx_init_clock_gating(struct drm_device *dev)
4797 {
4798         struct drm_i915_private *dev_priv = dev->dev_private;
4799
4800         /*
4801          * On Ibex Peak and Cougar Point, we need to disable clock
4802          * gating for the panel power sequencer or it will fail to
4803          * start up when no ports are active.
4804          */
4805         I915_WRITE(SOUTH_DSPCLK_GATE_D, PCH_DPLSUNIT_CLOCK_GATE_DISABLE);
4806 }
4807
4808 static void g4x_disable_trickle_feed(struct drm_device *dev)
4809 {
4810         struct drm_i915_private *dev_priv = dev->dev_private;
4811         int pipe;
4812
4813         for_each_pipe(pipe) {
4814                 I915_WRITE(DSPCNTR(pipe),
4815                            I915_READ(DSPCNTR(pipe)) |
4816                            DISPPLANE_TRICKLE_FEED_DISABLE);
4817                 intel_flush_primary_plane(dev_priv, pipe);
4818         }
4819 }
4820
4821 static void ironlake_init_clock_gating(struct drm_device *dev)
4822 {
4823         struct drm_i915_private *dev_priv = dev->dev_private;
4824         uint32_t dspclk_gate = ILK_VRHUNIT_CLOCK_GATE_DISABLE;
4825
4826         /*
4827          * Required for FBC
4828          * WaFbcDisableDpfcClockGating:ilk
4829          */
4830         dspclk_gate |= ILK_DPFCRUNIT_CLOCK_GATE_DISABLE |
4831                    ILK_DPFCUNIT_CLOCK_GATE_DISABLE |
4832                    ILK_DPFDUNIT_CLOCK_GATE_ENABLE;
4833
4834         I915_WRITE(PCH_3DCGDIS0,
4835                    MARIUNIT_CLOCK_GATE_DISABLE |
4836                    SVSMUNIT_CLOCK_GATE_DISABLE);
4837         I915_WRITE(PCH_3DCGDIS1,
4838                    VFMUNIT_CLOCK_GATE_DISABLE);
4839
4840         /*
4841          * According to the spec the following bits should be set in
4842          * order to enable memory self-refresh
4843          * The bit 22/21 of 0x42004
4844          * The bit 5 of 0x42020
4845          * The bit 15 of 0x45000
4846          */
4847         I915_WRITE(ILK_DISPLAY_CHICKEN2,
4848                    (I915_READ(ILK_DISPLAY_CHICKEN2) |
4849                     ILK_DPARB_GATE | ILK_VSDPFD_FULL));
4850         dspclk_gate |= ILK_DPARBUNIT_CLOCK_GATE_ENABLE;
4851         I915_WRITE(DISP_ARB_CTL,
4852                    (I915_READ(DISP_ARB_CTL) |
4853                     DISP_FBC_WM_DIS));
4854         I915_WRITE(WM3_LP_ILK, 0);
4855         I915_WRITE(WM2_LP_ILK, 0);
4856         I915_WRITE(WM1_LP_ILK, 0);
4857
4858         /*
4859          * Based on the document from hardware guys the following bits
4860          * should be set unconditionally in order to enable FBC.
4861          * The bit 22 of 0x42000
4862          * The bit 22 of 0x42004
4863          * The bit 7,8,9 of 0x42020.
4864          */
4865         if (IS_IRONLAKE_M(dev)) {
4866                 /* WaFbcAsynchFlipDisableFbcQueue:ilk */
4867                 I915_WRITE(ILK_DISPLAY_CHICKEN1,
4868                            I915_READ(ILK_DISPLAY_CHICKEN1) |
4869                            ILK_FBCQ_DIS);
4870                 I915_WRITE(ILK_DISPLAY_CHICKEN2,
4871                            I915_READ(ILK_DISPLAY_CHICKEN2) |
4872                            ILK_DPARB_GATE);
4873         }
4874
4875         I915_WRITE(ILK_DSPCLK_GATE_D, dspclk_gate);
4876
4877         I915_WRITE(ILK_DISPLAY_CHICKEN2,
4878                    I915_READ(ILK_DISPLAY_CHICKEN2) |
4879                    ILK_ELPIN_409_SELECT);
4880         I915_WRITE(_3D_CHICKEN2,
4881                    _3D_CHICKEN2_WM_READ_PIPELINED << 16 |
4882                    _3D_CHICKEN2_WM_READ_PIPELINED);
4883
4884         /* WaDisableRenderCachePipelinedFlush:ilk */
4885         I915_WRITE(CACHE_MODE_0,
4886                    _MASKED_BIT_ENABLE(CM0_PIPELINED_RENDER_FLUSH_DISABLE));
4887
4888         g4x_disable_trickle_feed(dev);
4889
4890         ibx_init_clock_gating(dev);
4891 }
4892
4893 static void cpt_init_clock_gating(struct drm_device *dev)
4894 {
4895         struct drm_i915_private *dev_priv = dev->dev_private;
4896         int pipe;
4897         uint32_t val;
4898
4899         /*
4900          * On Ibex Peak and Cougar Point, we need to disable clock
4901          * gating for the panel power sequencer or it will fail to
4902          * start up when no ports are active.
4903          */
4904         I915_WRITE(SOUTH_DSPCLK_GATE_D, PCH_DPLSUNIT_CLOCK_GATE_DISABLE);
4905         I915_WRITE(SOUTH_CHICKEN2, I915_READ(SOUTH_CHICKEN2) |
4906                    DPLS_EDP_PPS_FIX_DIS);
4907         /* The below fixes the weird display corruption, a few pixels shifted
4908          * downward, on (only) LVDS of some HP laptops with IVY.
4909          */
4910         for_each_pipe(pipe) {
4911                 val = I915_READ(TRANS_CHICKEN2(pipe));
4912                 val |= TRANS_CHICKEN2_TIMING_OVERRIDE;
4913                 val &= ~TRANS_CHICKEN2_FDI_POLARITY_REVERSED;
4914                 if (dev_priv->vbt.fdi_rx_polarity_inverted)
4915                         val |= TRANS_CHICKEN2_FDI_POLARITY_REVERSED;
4916                 val &= ~TRANS_CHICKEN2_FRAME_START_DELAY_MASK;
4917                 val &= ~TRANS_CHICKEN2_DISABLE_DEEP_COLOR_COUNTER;
4918                 val &= ~TRANS_CHICKEN2_DISABLE_DEEP_COLOR_MODESWITCH;
4919                 I915_WRITE(TRANS_CHICKEN2(pipe), val);
4920         }
4921         /* WADP0ClockGatingDisable */
4922         for_each_pipe(pipe) {
4923                 I915_WRITE(TRANS_CHICKEN1(pipe),
4924                            TRANS_CHICKEN1_DP0UNIT_GC_DISABLE);
4925         }
4926 }
4927
4928 static void gen6_check_mch_setup(struct drm_device *dev)
4929 {
4930         struct drm_i915_private *dev_priv = dev->dev_private;
4931         uint32_t tmp;
4932
4933         tmp = I915_READ(MCH_SSKPD);
4934         if ((tmp & MCH_SSKPD_WM0_MASK) != MCH_SSKPD_WM0_VAL) {
4935                 DRM_INFO("Wrong MCH_SSKPD value: 0x%08x\n", tmp);
4936                 DRM_INFO("This can cause pipe underruns and display issues.\n");
4937                 DRM_INFO("Please upgrade your BIOS to fix this.\n");
4938         }
4939 }
4940
4941 static void gen6_init_clock_gating(struct drm_device *dev)
4942 {
4943         struct drm_i915_private *dev_priv = dev->dev_private;
4944         uint32_t dspclk_gate = ILK_VRHUNIT_CLOCK_GATE_DISABLE;
4945
4946         I915_WRITE(ILK_DSPCLK_GATE_D, dspclk_gate);
4947
4948         I915_WRITE(ILK_DISPLAY_CHICKEN2,
4949                    I915_READ(ILK_DISPLAY_CHICKEN2) |
4950                    ILK_ELPIN_409_SELECT);
4951
4952         /* WaDisableHiZPlanesWhenMSAAEnabled:snb */
4953         I915_WRITE(_3D_CHICKEN,
4954                    _MASKED_BIT_ENABLE(_3D_CHICKEN_HIZ_PLANE_DISABLE_MSAA_4X_SNB));
4955
4956         /* WaSetupGtModeTdRowDispatch:snb */
4957         if (IS_SNB_GT1(dev))
4958                 I915_WRITE(GEN6_GT_MODE,
4959                            _MASKED_BIT_ENABLE(GEN6_TD_FOUR_ROW_DISPATCH_DISABLE));
4960
4961         I915_WRITE(WM3_LP_ILK, 0);
4962         I915_WRITE(WM2_LP_ILK, 0);
4963         I915_WRITE(WM1_LP_ILK, 0);
4964
4965         I915_WRITE(CACHE_MODE_0,
4966                    _MASKED_BIT_DISABLE(CM0_STC_EVICT_DISABLE_LRA_SNB));
4967
4968         I915_WRITE(GEN6_UCGCTL1,
4969                    I915_READ(GEN6_UCGCTL1) |
4970                    GEN6_BLBUNIT_CLOCK_GATE_DISABLE |
4971                    GEN6_CSUNIT_CLOCK_GATE_DISABLE);
4972
4973         /* According to the BSpec vol1g, bit 12 (RCPBUNIT) clock
4974          * gating disable must be set.  Failure to set it results in
4975          * flickering pixels due to Z write ordering failures after
4976          * some amount of runtime in the Mesa "fire" demo, and Unigine
4977          * Sanctuary and Tropics, and apparently anything else with
4978          * alpha test or pixel discard.
4979          *
4980          * According to the spec, bit 11 (RCCUNIT) must also be set,
4981          * but we didn't debug actual testcases to find it out.
4982          *
4983          * Also apply WaDisableVDSUnitClockGating:snb and
4984          * WaDisableRCPBUnitClockGating:snb.
4985          */
4986         I915_WRITE(GEN6_UCGCTL2,
4987                    GEN7_VDSUNIT_CLOCK_GATE_DISABLE |
4988                    GEN6_RCPBUNIT_CLOCK_GATE_DISABLE |
4989                    GEN6_RCCUNIT_CLOCK_GATE_DISABLE);
4990
4991         /* Bspec says we need to always set all mask bits. */
4992         I915_WRITE(_3D_CHICKEN3, (0xFFFF << 16) |
4993                    _3D_CHICKEN3_SF_DISABLE_FASTCLIP_CULL);
4994
4995         /*
4996          * According to the spec the following bits should be
4997          * set in order to enable memory self-refresh and fbc:
4998          * The bit21 and bit22 of 0x42000
4999          * The bit21 and bit22 of 0x42004
5000          * The bit5 and bit7 of 0x42020
5001          * The bit14 of 0x70180
5002          * The bit14 of 0x71180
5003          *
5004          * WaFbcAsynchFlipDisableFbcQueue:snb
5005          */
5006         I915_WRITE(ILK_DISPLAY_CHICKEN1,
5007                    I915_READ(ILK_DISPLAY_CHICKEN1) |
5008                    ILK_FBCQ_DIS | ILK_PABSTRETCH_DIS);
5009         I915_WRITE(ILK_DISPLAY_CHICKEN2,
5010                    I915_READ(ILK_DISPLAY_CHICKEN2) |
5011                    ILK_DPARB_GATE | ILK_VSDPFD_FULL);
5012         I915_WRITE(ILK_DSPCLK_GATE_D,
5013                    I915_READ(ILK_DSPCLK_GATE_D) |
5014                    ILK_DPARBUNIT_CLOCK_GATE_ENABLE  |
5015                    ILK_DPFDUNIT_CLOCK_GATE_ENABLE);
5016
5017         g4x_disable_trickle_feed(dev);
5018
5019         /* The default value should be 0x200 according to docs, but the two
5020          * platforms I checked have a 0 for this. (Maybe BIOS overrides?) */
5021         I915_WRITE(GEN6_GT_MODE, _MASKED_BIT_DISABLE(0xffff));
5022         I915_WRITE(GEN6_GT_MODE, _MASKED_BIT_ENABLE(GEN6_GT_MODE_HI));
5023
5024         cpt_init_clock_gating(dev);
5025
5026         gen6_check_mch_setup(dev);
5027 }
5028
5029 static void gen7_setup_fixed_func_scheduler(struct drm_i915_private *dev_priv)
5030 {
5031         uint32_t reg = I915_READ(GEN7_FF_THREAD_MODE);
5032
5033         reg &= ~GEN7_FF_SCHED_MASK;
5034         reg |= GEN7_FF_TS_SCHED_HW;
5035         reg |= GEN7_FF_VS_SCHED_HW;
5036         reg |= GEN7_FF_DS_SCHED_HW;
5037
5038         if (IS_HASWELL(dev_priv->dev))
5039                 reg &= ~GEN7_FF_VS_REF_CNT_FFME;
5040
5041         I915_WRITE(GEN7_FF_THREAD_MODE, reg);
5042 }
5043
5044 static void lpt_init_clock_gating(struct drm_device *dev)
5045 {
5046         struct drm_i915_private *dev_priv = dev->dev_private;
5047
5048         /*
5049          * TODO: this bit should only be enabled when really needed, then
5050          * disabled when not needed anymore in order to save power.
5051          */
5052         if (dev_priv->pch_id == INTEL_PCH_LPT_LP_DEVICE_ID_TYPE)
5053                 I915_WRITE(SOUTH_DSPCLK_GATE_D,
5054                            I915_READ(SOUTH_DSPCLK_GATE_D) |
5055                            PCH_LP_PARTITION_LEVEL_DISABLE);
5056
5057         /* WADPOClockGatingDisable:hsw */
5058         I915_WRITE(_TRANSA_CHICKEN1,
5059                    I915_READ(_TRANSA_CHICKEN1) |
5060                    TRANS_CHICKEN1_DP0UNIT_GC_DISABLE);
5061 }
5062
5063 static void lpt_suspend_hw(struct drm_device *dev)
5064 {
5065         struct drm_i915_private *dev_priv = dev->dev_private;
5066
5067         if (dev_priv->pch_id == INTEL_PCH_LPT_LP_DEVICE_ID_TYPE) {
5068                 uint32_t val = I915_READ(SOUTH_DSPCLK_GATE_D);
5069
5070                 val &= ~PCH_LP_PARTITION_LEVEL_DISABLE;
5071                 I915_WRITE(SOUTH_DSPCLK_GATE_D, val);
5072         }
5073 }
5074
5075 static void haswell_init_clock_gating(struct drm_device *dev)
5076 {
5077         struct drm_i915_private *dev_priv = dev->dev_private;
5078
5079         I915_WRITE(WM3_LP_ILK, 0);
5080         I915_WRITE(WM2_LP_ILK, 0);
5081         I915_WRITE(WM1_LP_ILK, 0);
5082
5083         /* According to the spec, bit 13 (RCZUNIT) must be set on IVB.
5084          * This implements the WaDisableRCZUnitClockGating:hsw workaround.
5085          */
5086         I915_WRITE(GEN6_UCGCTL2, GEN6_RCZUNIT_CLOCK_GATE_DISABLE);
5087
5088         /* Apply the WaDisableRHWOOptimizationForRenderHang:hsw workaround. */
5089         I915_WRITE(GEN7_COMMON_SLICE_CHICKEN1,
5090                    GEN7_CSC1_RHWO_OPT_DISABLE_IN_RCC);
5091
5092         /* WaApplyL3ControlAndL3ChickenMode:hsw */
5093         I915_WRITE(GEN7_L3CNTLREG1,
5094                         GEN7_WA_FOR_GEN7_L3_CONTROL);
5095         I915_WRITE(GEN7_L3_CHICKEN_MODE_REGISTER,
5096                         GEN7_WA_L3_CHICKEN_MODE);
5097
5098         /* This is required by WaCatErrorRejectionIssue:hsw */
5099         I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG,
5100                         I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG) |
5101                         GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB);
5102
5103         /* WaVSRefCountFullforceMissDisable:hsw */
5104         gen7_setup_fixed_func_scheduler(dev_priv);
5105
5106         /* WaDisable4x2SubspanOptimization:hsw */
5107         I915_WRITE(CACHE_MODE_1,
5108                    _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE));
5109
5110         /* WaSwitchSolVfFArbitrationPriority:hsw */
5111         I915_WRITE(GAM_ECOCHK, I915_READ(GAM_ECOCHK) | HSW_ECOCHK_ARB_PRIO_SOL);
5112
5113         /* WaRsPkgCStateDisplayPMReq:hsw */
5114         I915_WRITE(CHICKEN_PAR1_1,
5115                    I915_READ(CHICKEN_PAR1_1) | FORCE_ARB_IDLE_PLANES);
5116
5117         lpt_init_clock_gating(dev);
5118 }
5119
5120 static void ivybridge_init_clock_gating(struct drm_device *dev)
5121 {
5122         struct drm_i915_private *dev_priv = dev->dev_private;
5123         uint32_t snpcr;
5124
5125         I915_WRITE(WM3_LP_ILK, 0);
5126         I915_WRITE(WM2_LP_ILK, 0);
5127         I915_WRITE(WM1_LP_ILK, 0);
5128
5129         I915_WRITE(ILK_DSPCLK_GATE_D, ILK_VRHUNIT_CLOCK_GATE_DISABLE);
5130
5131         /* WaDisableEarlyCull:ivb */
5132         I915_WRITE(_3D_CHICKEN3,
5133                    _MASKED_BIT_ENABLE(_3D_CHICKEN_SF_DISABLE_OBJEND_CULL));
5134
5135         /* WaDisableBackToBackFlipFix:ivb */
5136         I915_WRITE(IVB_CHICKEN3,
5137                    CHICKEN3_DGMG_REQ_OUT_FIX_DISABLE |
5138                    CHICKEN3_DGMG_DONE_FIX_DISABLE);
5139
5140         /* WaDisablePSDDualDispatchEnable:ivb */
5141         if (IS_IVB_GT1(dev))
5142                 I915_WRITE(GEN7_HALF_SLICE_CHICKEN1,
5143                            _MASKED_BIT_ENABLE(GEN7_PSD_SINGLE_PORT_DISPATCH_ENABLE));
5144         else
5145                 I915_WRITE(GEN7_HALF_SLICE_CHICKEN1_GT2,
5146                            _MASKED_BIT_ENABLE(GEN7_PSD_SINGLE_PORT_DISPATCH_ENABLE));
5147
5148         /* Apply the WaDisableRHWOOptimizationForRenderHang:ivb workaround. */
5149         I915_WRITE(GEN7_COMMON_SLICE_CHICKEN1,
5150                    GEN7_CSC1_RHWO_OPT_DISABLE_IN_RCC);
5151
5152         /* WaApplyL3ControlAndL3ChickenMode:ivb */
5153         I915_WRITE(GEN7_L3CNTLREG1,
5154                         GEN7_WA_FOR_GEN7_L3_CONTROL);
5155         I915_WRITE(GEN7_L3_CHICKEN_MODE_REGISTER,
5156                    GEN7_WA_L3_CHICKEN_MODE);
5157         if (IS_IVB_GT1(dev))
5158                 I915_WRITE(GEN7_ROW_CHICKEN2,
5159                            _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE));
5160         else
5161                 I915_WRITE(GEN7_ROW_CHICKEN2_GT2,
5162                            _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE));
5163
5164
5165         /* WaForceL3Serialization:ivb */
5166         I915_WRITE(GEN7_L3SQCREG4, I915_READ(GEN7_L3SQCREG4) &
5167                    ~L3SQ_URB_READ_CAM_MATCH_DISABLE);
5168
5169         /* According to the BSpec vol1g, bit 12 (RCPBUNIT) clock
5170          * gating disable must be set.  Failure to set it results in
5171          * flickering pixels due to Z write ordering failures after
5172          * some amount of runtime in the Mesa "fire" demo, and Unigine
5173          * Sanctuary and Tropics, and apparently anything else with
5174          * alpha test or pixel discard.
5175          *
5176          * According to the spec, bit 11 (RCCUNIT) must also be set,
5177          * but we didn't debug actual testcases to find it out.
5178          *
5179          * According to the spec, bit 13 (RCZUNIT) must be set on IVB.
5180          * This implements the WaDisableRCZUnitClockGating:ivb workaround.
5181          */
5182         I915_WRITE(GEN6_UCGCTL2,
5183                    GEN6_RCZUNIT_CLOCK_GATE_DISABLE |
5184                    GEN6_RCCUNIT_CLOCK_GATE_DISABLE);
5185
5186         /* This is required by WaCatErrorRejectionIssue:ivb */
5187         I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG,
5188                         I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG) |
5189                         GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB);
5190
5191         g4x_disable_trickle_feed(dev);
5192
5193         /* WaVSRefCountFullforceMissDisable:ivb */
5194         gen7_setup_fixed_func_scheduler(dev_priv);
5195
5196         /* WaDisable4x2SubspanOptimization:ivb */
5197         I915_WRITE(CACHE_MODE_1,
5198                    _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE));
5199
5200         snpcr = I915_READ(GEN6_MBCUNIT_SNPCR);
5201         snpcr &= ~GEN6_MBC_SNPCR_MASK;
5202         snpcr |= GEN6_MBC_SNPCR_MED;
5203         I915_WRITE(GEN6_MBCUNIT_SNPCR, snpcr);
5204
5205         if (!HAS_PCH_NOP(dev))
5206                 cpt_init_clock_gating(dev);
5207
5208         gen6_check_mch_setup(dev);
5209 }
5210
5211 static void valleyview_init_clock_gating(struct drm_device *dev)
5212 {
5213         struct drm_i915_private *dev_priv = dev->dev_private;
5214
5215         I915_WRITE(DSPCLK_GATE_D, VRHUNIT_CLOCK_GATE_DISABLE);
5216
5217         /* WaDisableEarlyCull:vlv */
5218         I915_WRITE(_3D_CHICKEN3,
5219                    _MASKED_BIT_ENABLE(_3D_CHICKEN_SF_DISABLE_OBJEND_CULL));
5220
5221         /* WaDisableBackToBackFlipFix:vlv */
5222         I915_WRITE(IVB_CHICKEN3,
5223                    CHICKEN3_DGMG_REQ_OUT_FIX_DISABLE |
5224                    CHICKEN3_DGMG_DONE_FIX_DISABLE);
5225
5226         /* WaDisablePSDDualDispatchEnable:vlv */
5227         I915_WRITE(GEN7_HALF_SLICE_CHICKEN1,
5228                    _MASKED_BIT_ENABLE(GEN7_MAX_PS_THREAD_DEP |
5229                                       GEN7_PSD_SINGLE_PORT_DISPATCH_ENABLE));
5230
5231         /* Apply the WaDisableRHWOOptimizationForRenderHang:vlv workaround. */
5232         I915_WRITE(GEN7_COMMON_SLICE_CHICKEN1,
5233                    GEN7_CSC1_RHWO_OPT_DISABLE_IN_RCC);
5234
5235         /* WaApplyL3ControlAndL3ChickenMode:vlv */
5236         I915_WRITE(GEN7_L3CNTLREG1, I915_READ(GEN7_L3CNTLREG1) | GEN7_L3AGDIS);
5237         I915_WRITE(GEN7_L3_CHICKEN_MODE_REGISTER, GEN7_WA_L3_CHICKEN_MODE);
5238
5239         /* WaForceL3Serialization:vlv */
5240         I915_WRITE(GEN7_L3SQCREG4, I915_READ(GEN7_L3SQCREG4) &
5241                    ~L3SQ_URB_READ_CAM_MATCH_DISABLE);
5242
5243         /* WaDisableDopClockGating:vlv */
5244         I915_WRITE(GEN7_ROW_CHICKEN2,
5245                    _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE));
5246
5247         /* This is required by WaCatErrorRejectionIssue:vlv */
5248         I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG,
5249                    I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG) |
5250                    GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB);
5251
5252         /* According to the BSpec vol1g, bit 12 (RCPBUNIT) clock
5253          * gating disable must be set.  Failure to set it results in
5254          * flickering pixels due to Z write ordering failures after
5255          * some amount of runtime in the Mesa "fire" demo, and Unigine
5256          * Sanctuary and Tropics, and apparently anything else with
5257          * alpha test or pixel discard.
5258          *
5259          * According to the spec, bit 11 (RCCUNIT) must also be set,
5260          * but we didn't debug actual testcases to find it out.
5261          *
5262          * According to the spec, bit 13 (RCZUNIT) must be set on IVB.
5263          * This implements the WaDisableRCZUnitClockGating:vlv workaround.
5264          *
5265          * Also apply WaDisableVDSUnitClockGating:vlv and
5266          * WaDisableRCPBUnitClockGating:vlv.
5267          */
5268         I915_WRITE(GEN6_UCGCTL2,
5269                    GEN7_VDSUNIT_CLOCK_GATE_DISABLE |
5270                    GEN7_TDLUNIT_CLOCK_GATE_DISABLE |
5271                    GEN6_RCZUNIT_CLOCK_GATE_DISABLE |
5272                    GEN6_RCPBUNIT_CLOCK_GATE_DISABLE |
5273                    GEN6_RCCUNIT_CLOCK_GATE_DISABLE);
5274
5275         I915_WRITE(GEN7_UCGCTL4, GEN7_L3BANK2X_CLOCK_GATE_DISABLE);
5276
5277         I915_WRITE(MI_ARB_VLV, MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE);
5278
5279         I915_WRITE(CACHE_MODE_1,
5280                    _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE));
5281
5282         /*
5283          * WaDisableVLVClockGating_VBIIssue:vlv
5284          * Disable clock gating on th GCFG unit to prevent a delay
5285          * in the reporting of vblank events.
5286          */
5287         I915_WRITE(VLV_GUNIT_CLOCK_GATE, 0xffffffff);
5288
5289         /* Conservative clock gating settings for now */
5290         I915_WRITE(0x9400, 0xffffffff);
5291         I915_WRITE(0x9404, 0xffffffff);
5292         I915_WRITE(0x9408, 0xffffffff);
5293         I915_WRITE(0x940c, 0xffffffff);
5294         I915_WRITE(0x9410, 0xffffffff);
5295         I915_WRITE(0x9414, 0xffffffff);
5296         I915_WRITE(0x9418, 0xffffffff);
5297 }
5298
5299 static void g4x_init_clock_gating(struct drm_device *dev)
5300 {
5301         struct drm_i915_private *dev_priv = dev->dev_private;
5302         uint32_t dspclk_gate;
5303
5304         I915_WRITE(RENCLK_GATE_D1, 0);
5305         I915_WRITE(RENCLK_GATE_D2, VF_UNIT_CLOCK_GATE_DISABLE |
5306                    GS_UNIT_CLOCK_GATE_DISABLE |
5307                    CL_UNIT_CLOCK_GATE_DISABLE);
5308         I915_WRITE(RAMCLK_GATE_D, 0);
5309         dspclk_gate = VRHUNIT_CLOCK_GATE_DISABLE |
5310                 OVRUNIT_CLOCK_GATE_DISABLE |
5311                 OVCUNIT_CLOCK_GATE_DISABLE;
5312         if (IS_GM45(dev))
5313                 dspclk_gate |= DSSUNIT_CLOCK_GATE_DISABLE;
5314         I915_WRITE(DSPCLK_GATE_D, dspclk_gate);
5315
5316         /* WaDisableRenderCachePipelinedFlush */
5317         I915_WRITE(CACHE_MODE_0,
5318                    _MASKED_BIT_ENABLE(CM0_PIPELINED_RENDER_FLUSH_DISABLE));
5319
5320         g4x_disable_trickle_feed(dev);
5321 }
5322
5323 static void crestline_init_clock_gating(struct drm_device *dev)
5324 {
5325         struct drm_i915_private *dev_priv = dev->dev_private;
5326
5327         I915_WRITE(RENCLK_GATE_D1, I965_RCC_CLOCK_GATE_DISABLE);
5328         I915_WRITE(RENCLK_GATE_D2, 0);
5329         I915_WRITE(DSPCLK_GATE_D, 0);
5330         I915_WRITE(RAMCLK_GATE_D, 0);
5331         I915_WRITE16(DEUC, 0);
5332         I915_WRITE(MI_ARB_STATE,
5333                    _MASKED_BIT_ENABLE(MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE));
5334 }
5335
5336 static void broadwater_init_clock_gating(struct drm_device *dev)
5337 {
5338         struct drm_i915_private *dev_priv = dev->dev_private;
5339
5340         I915_WRITE(RENCLK_GATE_D1, I965_RCZ_CLOCK_GATE_DISABLE |
5341                    I965_RCC_CLOCK_GATE_DISABLE |
5342                    I965_RCPB_CLOCK_GATE_DISABLE |
5343                    I965_ISC_CLOCK_GATE_DISABLE |
5344                    I965_FBC_CLOCK_GATE_DISABLE);
5345         I915_WRITE(RENCLK_GATE_D2, 0);
5346         I915_WRITE(MI_ARB_STATE,
5347                    _MASKED_BIT_ENABLE(MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE));
5348 }
5349
5350 static void gen3_init_clock_gating(struct drm_device *dev)
5351 {
5352         struct drm_i915_private *dev_priv = dev->dev_private;
5353         u32 dstate = I915_READ(D_STATE);
5354
5355         dstate |= DSTATE_PLL_D3_OFF | DSTATE_GFX_CLOCK_GATING |
5356                 DSTATE_DOT_CLOCK_GATING;
5357         I915_WRITE(D_STATE, dstate);
5358
5359         if (IS_PINEVIEW(dev))
5360                 I915_WRITE(ECOSKPD, _MASKED_BIT_ENABLE(ECO_GATING_CX_ONLY));
5361
5362         /* IIR "flip pending" means done if this bit is set */
5363         I915_WRITE(ECOSKPD, _MASKED_BIT_DISABLE(ECO_FLIP_DONE));
5364 }
5365
5366 static void i85x_init_clock_gating(struct drm_device *dev)
5367 {
5368         struct drm_i915_private *dev_priv = dev->dev_private;
5369
5370         I915_WRITE(RENCLK_GATE_D1, SV_CLOCK_GATE_DISABLE);
5371 }
5372
5373 static void i830_init_clock_gating(struct drm_device *dev)
5374 {
5375         struct drm_i915_private *dev_priv = dev->dev_private;
5376
5377         I915_WRITE(DSPCLK_GATE_D, OVRUNIT_CLOCK_GATE_DISABLE);
5378 }
5379
5380 void intel_init_clock_gating(struct drm_device *dev)
5381 {
5382         struct drm_i915_private *dev_priv = dev->dev_private;
5383
5384         dev_priv->display.init_clock_gating(dev);
5385 }
5386
5387 void intel_suspend_hw(struct drm_device *dev)
5388 {
5389         if (HAS_PCH_LPT(dev))
5390                 lpt_suspend_hw(dev);
5391 }
5392
5393 /**
5394  * We should only use the power well if we explicitly asked the hardware to
5395  * enable it, so check if it's enabled and also check if we've requested it to
5396  * be enabled.
5397  */
5398 bool intel_display_power_enabled(struct drm_device *dev,
5399                                  enum intel_display_power_domain domain)
5400 {
5401         struct drm_i915_private *dev_priv = dev->dev_private;
5402
5403         if (!HAS_POWER_WELL(dev))
5404                 return true;
5405
5406         switch (domain) {
5407         case POWER_DOMAIN_PIPE_A:
5408         case POWER_DOMAIN_TRANSCODER_EDP:
5409                 return true;
5410         case POWER_DOMAIN_VGA:
5411         case POWER_DOMAIN_PIPE_B:
5412         case POWER_DOMAIN_PIPE_C:
5413         case POWER_DOMAIN_PIPE_A_PANEL_FITTER:
5414         case POWER_DOMAIN_PIPE_B_PANEL_FITTER:
5415         case POWER_DOMAIN_PIPE_C_PANEL_FITTER:
5416         case POWER_DOMAIN_TRANSCODER_A:
5417         case POWER_DOMAIN_TRANSCODER_B:
5418         case POWER_DOMAIN_TRANSCODER_C:
5419                 return I915_READ(HSW_PWR_WELL_DRIVER) ==
5420                      (HSW_PWR_WELL_ENABLE_REQUEST | HSW_PWR_WELL_STATE_ENABLED);
5421         default:
5422                 BUG();
5423         }
5424 }
5425
5426 static void __intel_set_power_well(struct drm_device *dev, bool enable)
5427 {
5428         struct drm_i915_private *dev_priv = dev->dev_private;
5429         bool is_enabled, enable_requested;
5430         uint32_t tmp;
5431
5432         tmp = I915_READ(HSW_PWR_WELL_DRIVER);
5433         is_enabled = tmp & HSW_PWR_WELL_STATE_ENABLED;
5434         enable_requested = tmp & HSW_PWR_WELL_ENABLE_REQUEST;
5435
5436         if (enable) {
5437                 if (!enable_requested)
5438                         I915_WRITE(HSW_PWR_WELL_DRIVER,
5439                                    HSW_PWR_WELL_ENABLE_REQUEST);
5440
5441                 if (!is_enabled) {
5442                         DRM_DEBUG_KMS("Enabling power well\n");
5443                         if (wait_for((I915_READ(HSW_PWR_WELL_DRIVER) &
5444                                       HSW_PWR_WELL_STATE_ENABLED), 20))
5445                                 DRM_ERROR("Timeout enabling power well\n");
5446                 }
5447         } else {
5448                 if (enable_requested) {
5449                         unsigned long irqflags;
5450                         enum pipe p;
5451
5452                         I915_WRITE(HSW_PWR_WELL_DRIVER, 0);
5453                         POSTING_READ(HSW_PWR_WELL_DRIVER);
5454                         DRM_DEBUG_KMS("Requesting to disable the power well\n");
5455
5456                         /*
5457                          * After this, the registers on the pipes that are part
5458                          * of the power well will become zero, so we have to
5459                          * adjust our counters according to that.
5460                          *
5461                          * FIXME: Should we do this in general in
5462                          * drm_vblank_post_modeset?
5463                          */
5464                         spin_lock_irqsave(&dev->vbl_lock, irqflags);
5465                         for_each_pipe(p)
5466                                 if (p != PIPE_A)
5467                                         dev->vblank[p].last = 0;
5468                         spin_unlock_irqrestore(&dev->vbl_lock, irqflags);
5469                 }
5470         }
5471 }
5472
5473 static void __intel_power_well_get(struct i915_power_well *power_well)
5474 {
5475         if (!power_well->count++)
5476                 __intel_set_power_well(power_well->device, true);
5477 }
5478
5479 static void __intel_power_well_put(struct i915_power_well *power_well)
5480 {
5481         WARN_ON(!power_well->count);
5482         if (!--power_well->count)
5483                 __intel_set_power_well(power_well->device, false);
5484 }
5485
5486 void intel_display_power_get(struct drm_device *dev,
5487                              enum intel_display_power_domain domain)
5488 {
5489         struct drm_i915_private *dev_priv = dev->dev_private;
5490         struct i915_power_well *power_well = &dev_priv->power_well;
5491
5492         if (!HAS_POWER_WELL(dev))
5493                 return;
5494
5495         switch (domain) {
5496         case POWER_DOMAIN_PIPE_A:
5497         case POWER_DOMAIN_TRANSCODER_EDP:
5498                 return;
5499         case POWER_DOMAIN_VGA:
5500         case POWER_DOMAIN_PIPE_B:
5501         case POWER_DOMAIN_PIPE_C:
5502         case POWER_DOMAIN_PIPE_A_PANEL_FITTER:
5503         case POWER_DOMAIN_PIPE_B_PANEL_FITTER:
5504         case POWER_DOMAIN_PIPE_C_PANEL_FITTER:
5505         case POWER_DOMAIN_TRANSCODER_A:
5506         case POWER_DOMAIN_TRANSCODER_B:
5507         case POWER_DOMAIN_TRANSCODER_C:
5508                 spin_lock_irq(&power_well->lock);
5509                 __intel_power_well_get(power_well);
5510                 spin_unlock_irq(&power_well->lock);
5511                 return;
5512         default:
5513                 BUG();
5514         }
5515 }
5516
5517 void intel_display_power_put(struct drm_device *dev,
5518                              enum intel_display_power_domain domain)
5519 {
5520         struct drm_i915_private *dev_priv = dev->dev_private;
5521         struct i915_power_well *power_well = &dev_priv->power_well;
5522
5523         if (!HAS_POWER_WELL(dev))
5524                 return;
5525
5526         switch (domain) {
5527         case POWER_DOMAIN_PIPE_A:
5528         case POWER_DOMAIN_TRANSCODER_EDP:
5529                 return;
5530         case POWER_DOMAIN_VGA:
5531         case POWER_DOMAIN_PIPE_B:
5532         case POWER_DOMAIN_PIPE_C:
5533         case POWER_DOMAIN_PIPE_A_PANEL_FITTER:
5534         case POWER_DOMAIN_PIPE_B_PANEL_FITTER:
5535         case POWER_DOMAIN_PIPE_C_PANEL_FITTER:
5536         case POWER_DOMAIN_TRANSCODER_A:
5537         case POWER_DOMAIN_TRANSCODER_B:
5538         case POWER_DOMAIN_TRANSCODER_C:
5539                 spin_lock_irq(&power_well->lock);
5540                 __intel_power_well_put(power_well);
5541                 spin_unlock_irq(&power_well->lock);
5542                 return;
5543         default:
5544                 BUG();
5545         }
5546 }
5547
5548 static struct i915_power_well *hsw_pwr;
5549
5550 /* Display audio driver power well request */
5551 void i915_request_power_well(void)
5552 {
5553         if (WARN_ON(!hsw_pwr))
5554                 return;
5555
5556         spin_lock_irq(&hsw_pwr->lock);
5557         __intel_power_well_get(hsw_pwr);
5558         spin_unlock_irq(&hsw_pwr->lock);
5559 }
5560 EXPORT_SYMBOL_GPL(i915_request_power_well);
5561
5562 /* Display audio driver power well release */
5563 void i915_release_power_well(void)
5564 {
5565         if (WARN_ON(!hsw_pwr))
5566                 return;
5567
5568         spin_lock_irq(&hsw_pwr->lock);
5569         __intel_power_well_put(hsw_pwr);
5570         spin_unlock_irq(&hsw_pwr->lock);
5571 }
5572 EXPORT_SYMBOL_GPL(i915_release_power_well);
5573
5574 int i915_init_power_well(struct drm_device *dev)
5575 {
5576         struct drm_i915_private *dev_priv = dev->dev_private;
5577
5578         hsw_pwr = &dev_priv->power_well;
5579
5580         hsw_pwr->device = dev;
5581         spin_lock_init(&hsw_pwr->lock);
5582         hsw_pwr->count = 0;
5583
5584         return 0;
5585 }
5586
5587 void i915_remove_power_well(struct drm_device *dev)
5588 {
5589         hsw_pwr = NULL;
5590 }
5591
5592 void intel_set_power_well(struct drm_device *dev, bool enable)
5593 {
5594         struct drm_i915_private *dev_priv = dev->dev_private;
5595         struct i915_power_well *power_well = &dev_priv->power_well;
5596
5597         if (!HAS_POWER_WELL(dev))
5598                 return;
5599
5600         if (!i915_disable_power_well && !enable)
5601                 return;
5602
5603         spin_lock_irq(&power_well->lock);
5604
5605         /*
5606          * This function will only ever contribute one
5607          * to the power well reference count. i915_request
5608          * is what tracks whether we have or have not
5609          * added the one to the reference count.
5610          */
5611         if (power_well->i915_request == enable)
5612                 goto out;
5613
5614         power_well->i915_request = enable;
5615
5616         if (enable)
5617                 __intel_power_well_get(power_well);
5618         else
5619                 __intel_power_well_put(power_well);
5620
5621  out:
5622         spin_unlock_irq(&power_well->lock);
5623 }
5624
5625 static void intel_resume_power_well(struct drm_device *dev)
5626 {
5627         struct drm_i915_private *dev_priv = dev->dev_private;
5628         struct i915_power_well *power_well = &dev_priv->power_well;
5629
5630         if (!HAS_POWER_WELL(dev))
5631                 return;
5632
5633         spin_lock_irq(&power_well->lock);
5634         __intel_set_power_well(dev, power_well->count > 0);
5635         spin_unlock_irq(&power_well->lock);
5636 }
5637
5638 /*
5639  * Starting with Haswell, we have a "Power Down Well" that can be turned off
5640  * when not needed anymore. We have 4 registers that can request the power well
5641  * to be enabled, and it will only be disabled if none of the registers is
5642  * requesting it to be enabled.
5643  */
5644 void intel_init_power_well(struct drm_device *dev)
5645 {
5646         struct drm_i915_private *dev_priv = dev->dev_private;
5647
5648         if (!HAS_POWER_WELL(dev))
5649                 return;
5650
5651         /* For now, we need the power well to be always enabled. */
5652         intel_set_power_well(dev, true);
5653         intel_resume_power_well(dev);
5654
5655         /* We're taking over the BIOS, so clear any requests made by it since
5656          * the driver is in charge now. */
5657         if (I915_READ(HSW_PWR_WELL_BIOS) & HSW_PWR_WELL_ENABLE_REQUEST)
5658                 I915_WRITE(HSW_PWR_WELL_BIOS, 0);
5659 }
5660
5661 /* Disables PC8 so we can use the GMBUS and DP AUX interrupts. */
5662 void intel_aux_display_runtime_get(struct drm_i915_private *dev_priv)
5663 {
5664         hsw_disable_package_c8(dev_priv);
5665 }
5666
5667 void intel_aux_display_runtime_put(struct drm_i915_private *dev_priv)
5668 {
5669         hsw_enable_package_c8(dev_priv);
5670 }
5671
5672 /* Set up chip specific power management-related functions */
5673 void intel_init_pm(struct drm_device *dev)
5674 {
5675         struct drm_i915_private *dev_priv = dev->dev_private;
5676
5677         if (I915_HAS_FBC(dev)) {
5678                 if (HAS_PCH_SPLIT(dev)) {
5679                         dev_priv->display.fbc_enabled = ironlake_fbc_enabled;
5680                         if (IS_IVYBRIDGE(dev) || IS_HASWELL(dev))
5681                                 dev_priv->display.enable_fbc =
5682                                         gen7_enable_fbc;
5683                         else
5684                                 dev_priv->display.enable_fbc =
5685                                         ironlake_enable_fbc;
5686                         dev_priv->display.disable_fbc = ironlake_disable_fbc;
5687                 } else if (IS_GM45(dev)) {
5688                         dev_priv->display.fbc_enabled = g4x_fbc_enabled;
5689                         dev_priv->display.enable_fbc = g4x_enable_fbc;
5690                         dev_priv->display.disable_fbc = g4x_disable_fbc;
5691                 } else if (IS_CRESTLINE(dev)) {
5692                         dev_priv->display.fbc_enabled = i8xx_fbc_enabled;
5693                         dev_priv->display.enable_fbc = i8xx_enable_fbc;
5694                         dev_priv->display.disable_fbc = i8xx_disable_fbc;
5695                 }
5696                 /* 855GM needs testing */
5697         }
5698
5699         /* For cxsr */
5700         if (IS_PINEVIEW(dev))
5701                 i915_pineview_get_mem_freq(dev);
5702         else if (IS_GEN5(dev))
5703                 i915_ironlake_get_mem_freq(dev);
5704
5705         /* For FIFO watermark updates */
5706         if (HAS_PCH_SPLIT(dev)) {
5707                 intel_setup_wm_latency(dev);
5708
5709                 if (IS_GEN5(dev)) {
5710                         if (dev_priv->wm.pri_latency[1] &&
5711                             dev_priv->wm.spr_latency[1] &&
5712                             dev_priv->wm.cur_latency[1])
5713                                 dev_priv->display.update_wm = ironlake_update_wm;
5714                         else {
5715                                 DRM_DEBUG_KMS("Failed to get proper latency. "
5716                                               "Disable CxSR\n");
5717                                 dev_priv->display.update_wm = NULL;
5718                         }
5719                         dev_priv->display.init_clock_gating = ironlake_init_clock_gating;
5720                 } else if (IS_GEN6(dev)) {
5721                         if (dev_priv->wm.pri_latency[0] &&
5722                             dev_priv->wm.spr_latency[0] &&
5723                             dev_priv->wm.cur_latency[0]) {
5724                                 dev_priv->display.update_wm = sandybridge_update_wm;
5725                                 dev_priv->display.update_sprite_wm = sandybridge_update_sprite_wm;
5726                         } else {
5727                                 DRM_DEBUG_KMS("Failed to read display plane latency. "
5728                                               "Disable CxSR\n");
5729                                 dev_priv->display.update_wm = NULL;
5730                         }
5731                         dev_priv->display.init_clock_gating = gen6_init_clock_gating;
5732                 } else if (IS_IVYBRIDGE(dev)) {
5733                         if (dev_priv->wm.pri_latency[0] &&
5734                             dev_priv->wm.spr_latency[0] &&
5735                             dev_priv->wm.cur_latency[0]) {
5736                                 dev_priv->display.update_wm = ivybridge_update_wm;
5737                                 dev_priv->display.update_sprite_wm = sandybridge_update_sprite_wm;
5738                         } else {
5739                                 DRM_DEBUG_KMS("Failed to read display plane latency. "
5740                                               "Disable CxSR\n");
5741                                 dev_priv->display.update_wm = NULL;
5742                         }
5743                         dev_priv->display.init_clock_gating = ivybridge_init_clock_gating;
5744                 } else if (IS_HASWELL(dev)) {
5745                         if (dev_priv->wm.pri_latency[0] &&
5746                             dev_priv->wm.spr_latency[0] &&
5747                             dev_priv->wm.cur_latency[0]) {
5748                                 dev_priv->display.update_wm = haswell_update_wm;
5749                                 dev_priv->display.update_sprite_wm =
5750                                         haswell_update_sprite_wm;
5751                         } else {
5752                                 DRM_DEBUG_KMS("Failed to read display plane latency. "
5753                                               "Disable CxSR\n");
5754                                 dev_priv->display.update_wm = NULL;
5755                         }
5756                         dev_priv->display.init_clock_gating = haswell_init_clock_gating;
5757                 } else
5758                         dev_priv->display.update_wm = NULL;
5759         } else if (IS_VALLEYVIEW(dev)) {
5760                 dev_priv->display.update_wm = valleyview_update_wm;
5761                 dev_priv->display.init_clock_gating =
5762                         valleyview_init_clock_gating;
5763         } else if (IS_PINEVIEW(dev)) {
5764                 if (!intel_get_cxsr_latency(IS_PINEVIEW_G(dev),
5765                                             dev_priv->is_ddr3,
5766                                             dev_priv->fsb_freq,
5767                                             dev_priv->mem_freq)) {
5768                         DRM_INFO("failed to find known CxSR latency "
5769                                  "(found ddr%s fsb freq %d, mem freq %d), "
5770                                  "disabling CxSR\n",
5771                                  (dev_priv->is_ddr3 == 1) ? "3" : "2",
5772                                  dev_priv->fsb_freq, dev_priv->mem_freq);
5773                         /* Disable CxSR and never update its watermark again */
5774                         pineview_disable_cxsr(dev);
5775                         dev_priv->display.update_wm = NULL;
5776                 } else
5777                         dev_priv->display.update_wm = pineview_update_wm;
5778                 dev_priv->display.init_clock_gating = gen3_init_clock_gating;
5779         } else if (IS_G4X(dev)) {
5780                 dev_priv->display.update_wm = g4x_update_wm;
5781                 dev_priv->display.init_clock_gating = g4x_init_clock_gating;
5782         } else if (IS_GEN4(dev)) {
5783                 dev_priv->display.update_wm = i965_update_wm;
5784                 if (IS_CRESTLINE(dev))
5785                         dev_priv->display.init_clock_gating = crestline_init_clock_gating;
5786                 else if (IS_BROADWATER(dev))
5787                         dev_priv->display.init_clock_gating = broadwater_init_clock_gating;
5788         } else if (IS_GEN3(dev)) {
5789                 dev_priv->display.update_wm = i9xx_update_wm;
5790                 dev_priv->display.get_fifo_size = i9xx_get_fifo_size;
5791                 dev_priv->display.init_clock_gating = gen3_init_clock_gating;
5792         } else if (IS_I865G(dev)) {
5793                 dev_priv->display.update_wm = i830_update_wm;
5794                 dev_priv->display.init_clock_gating = i85x_init_clock_gating;
5795                 dev_priv->display.get_fifo_size = i830_get_fifo_size;
5796         } else if (IS_I85X(dev)) {
5797                 dev_priv->display.update_wm = i9xx_update_wm;
5798                 dev_priv->display.get_fifo_size = i85x_get_fifo_size;
5799                 dev_priv->display.init_clock_gating = i85x_init_clock_gating;
5800         } else {
5801                 dev_priv->display.update_wm = i830_update_wm;
5802                 dev_priv->display.init_clock_gating = i830_init_clock_gating;
5803                 if (IS_845G(dev))
5804                         dev_priv->display.get_fifo_size = i845_get_fifo_size;
5805                 else
5806                         dev_priv->display.get_fifo_size = i830_get_fifo_size;
5807         }
5808 }
5809
5810 int sandybridge_pcode_read(struct drm_i915_private *dev_priv, u8 mbox, u32 *val)
5811 {
5812         WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
5813
5814         if (I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) {
5815                 DRM_DEBUG_DRIVER("warning: pcode (read) mailbox access failed\n");
5816                 return -EAGAIN;
5817         }
5818
5819         I915_WRITE(GEN6_PCODE_DATA, *val);
5820         I915_WRITE(GEN6_PCODE_MAILBOX, GEN6_PCODE_READY | mbox);
5821
5822         if (wait_for((I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) == 0,
5823                      500)) {
5824                 DRM_ERROR("timeout waiting for pcode read (%d) to finish\n", mbox);
5825                 return -ETIMEDOUT;
5826         }
5827
5828         *val = I915_READ(GEN6_PCODE_DATA);
5829         I915_WRITE(GEN6_PCODE_DATA, 0);
5830
5831         return 0;
5832 }
5833
5834 int sandybridge_pcode_write(struct drm_i915_private *dev_priv, u8 mbox, u32 val)
5835 {
5836         WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
5837
5838         if (I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) {
5839                 DRM_DEBUG_DRIVER("warning: pcode (write) mailbox access failed\n");
5840                 return -EAGAIN;
5841         }
5842
5843         I915_WRITE(GEN6_PCODE_DATA, val);
5844         I915_WRITE(GEN6_PCODE_MAILBOX, GEN6_PCODE_READY | mbox);
5845
5846         if (wait_for((I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) == 0,
5847                      500)) {
5848                 DRM_ERROR("timeout waiting for pcode write (%d) to finish\n", mbox);
5849                 return -ETIMEDOUT;
5850         }
5851
5852         I915_WRITE(GEN6_PCODE_DATA, 0);
5853
5854         return 0;
5855 }
5856
5857 int vlv_gpu_freq(int ddr_freq, int val)
5858 {
5859         int mult, base;
5860
5861         switch (ddr_freq) {
5862         case 800:
5863                 mult = 20;
5864                 base = 120;
5865                 break;
5866         case 1066:
5867                 mult = 22;
5868                 base = 133;
5869                 break;
5870         case 1333:
5871                 mult = 21;
5872                 base = 125;
5873                 break;
5874         default:
5875                 return -1;
5876         }
5877
5878         return ((val - 0xbd) * mult) + base;
5879 }
5880
5881 int vlv_freq_opcode(int ddr_freq, int val)
5882 {
5883         int mult, base;
5884
5885         switch (ddr_freq) {
5886         case 800:
5887                 mult = 20;
5888                 base = 120;
5889                 break;
5890         case 1066:
5891                 mult = 22;
5892                 base = 133;
5893                 break;
5894         case 1333:
5895                 mult = 21;
5896                 base = 125;
5897                 break;
5898         default:
5899                 return -1;
5900         }
5901
5902         val /= mult;
5903         val -= base / mult;
5904         val += 0xbd;
5905
5906         if (val > 0xea)
5907                 val = 0xea;
5908
5909         return val;
5910 }
5911
5912 void intel_pm_init(struct drm_device *dev)
5913 {
5914         struct drm_i915_private *dev_priv = dev->dev_private;
5915
5916         INIT_DELAYED_WORK(&dev_priv->rps.delayed_resume_work,
5917                           intel_gen6_powersave_work);
5918 }
5919