]>
Commit | Line | Data |
---|---|---|
85208be0 ED |
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 | ||
2b4e57bd | 28 | #include <linux/cpufreq.h> |
85208be0 ED |
29 | #include "i915_drv.h" |
30 | #include "intel_drv.h" | |
eb48eb00 DV |
31 | #include "../../../platform/x86/intel_ips.h" |
32 | #include <linux/module.h> | |
f4db9321 | 33 | #include <drm/i915_powerwell.h> |
85208be0 | 34 | |
057d3860 | 35 | #define FORCEWAKE_ACK_TIMEOUT_MS 2 |
b67a4376 | 36 | |
f6750b3c ED |
37 | /* FBC, or Frame Buffer Compression, is a technique employed to compress the |
38 | * framebuffer contents in-memory, aiming at reducing the required bandwidth | |
39 | * during in-memory transfers and, therefore, reduce the power packet. | |
85208be0 | 40 | * |
f6750b3c ED |
41 | * The benefits of FBC are mostly visible with solid backgrounds and |
42 | * variation-less patterns. | |
85208be0 | 43 | * |
f6750b3c ED |
44 | * FBC-related functionality can be enabled by the means of the |
45 | * i915.i915_enable_fbc parameter | |
85208be0 ED |
46 | */ |
47 | ||
3490ea5d CW |
48 | static bool intel_crtc_active(struct drm_crtc *crtc) |
49 | { | |
50 | /* Be paranoid as we can arrive here with only partial | |
51 | * state retrieved from the hardware during setup. | |
52 | */ | |
53 | return to_intel_crtc(crtc)->active && crtc->fb && crtc->mode.clock; | |
54 | } | |
55 | ||
1fa61106 | 56 | static void i8xx_disable_fbc(struct drm_device *dev) |
85208be0 ED |
57 | { |
58 | struct drm_i915_private *dev_priv = dev->dev_private; | |
59 | u32 fbc_ctl; | |
60 | ||
61 | /* Disable compression */ | |
62 | fbc_ctl = I915_READ(FBC_CONTROL); | |
63 | if ((fbc_ctl & FBC_CTL_EN) == 0) | |
64 | return; | |
65 | ||
66 | fbc_ctl &= ~FBC_CTL_EN; | |
67 | I915_WRITE(FBC_CONTROL, fbc_ctl); | |
68 | ||
69 | /* Wait for compressing bit to clear */ | |
70 | if (wait_for((I915_READ(FBC_STATUS) & FBC_STAT_COMPRESSING) == 0, 10)) { | |
71 | DRM_DEBUG_KMS("FBC idle timed out\n"); | |
72 | return; | |
73 | } | |
74 | ||
75 | DRM_DEBUG_KMS("disabled FBC\n"); | |
76 | } | |
77 | ||
1fa61106 | 78 | static void i8xx_enable_fbc(struct drm_crtc *crtc, unsigned long interval) |
85208be0 ED |
79 | { |
80 | struct drm_device *dev = crtc->dev; | |
81 | struct drm_i915_private *dev_priv = dev->dev_private; | |
82 | struct drm_framebuffer *fb = crtc->fb; | |
83 | struct intel_framebuffer *intel_fb = to_intel_framebuffer(fb); | |
84 | struct drm_i915_gem_object *obj = intel_fb->obj; | |
85 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc); | |
86 | int cfb_pitch; | |
87 | int plane, i; | |
88 | u32 fbc_ctl, fbc_ctl2; | |
89 | ||
5c3fe8b0 | 90 | cfb_pitch = dev_priv->fbc.size / FBC_LL_SIZE; |
85208be0 ED |
91 | if (fb->pitches[0] < cfb_pitch) |
92 | cfb_pitch = fb->pitches[0]; | |
93 | ||
94 | /* FBC_CTL wants 64B units */ | |
95 | cfb_pitch = (cfb_pitch / 64) - 1; | |
96 | plane = intel_crtc->plane == 0 ? FBC_CTL_PLANEA : FBC_CTL_PLANEB; | |
97 | ||
98 | /* Clear old tags */ | |
99 | for (i = 0; i < (FBC_LL_SIZE / 32) + 1; i++) | |
100 | I915_WRITE(FBC_TAG + (i * 4), 0); | |
101 | ||
102 | /* Set it up... */ | |
103 | fbc_ctl2 = FBC_CTL_FENCE_DBL | FBC_CTL_IDLE_IMM | FBC_CTL_CPU_FENCE; | |
104 | fbc_ctl2 |= plane; | |
105 | I915_WRITE(FBC_CONTROL2, fbc_ctl2); | |
106 | I915_WRITE(FBC_FENCE_OFF, crtc->y); | |
107 | ||
108 | /* enable it... */ | |
109 | fbc_ctl = FBC_CTL_EN | FBC_CTL_PERIODIC; | |
110 | if (IS_I945GM(dev)) | |
111 | fbc_ctl |= FBC_CTL_C3_IDLE; /* 945 needs special SR handling */ | |
112 | fbc_ctl |= (cfb_pitch & 0xff) << FBC_CTL_STRIDE_SHIFT; | |
113 | fbc_ctl |= (interval & 0x2fff) << FBC_CTL_INTERVAL_SHIFT; | |
114 | fbc_ctl |= obj->fence_reg; | |
115 | I915_WRITE(FBC_CONTROL, fbc_ctl); | |
116 | ||
84f44ce7 VS |
117 | DRM_DEBUG_KMS("enabled FBC, pitch %d, yoff %d, plane %c, ", |
118 | cfb_pitch, crtc->y, plane_name(intel_crtc->plane)); | |
85208be0 ED |
119 | } |
120 | ||
1fa61106 | 121 | static bool i8xx_fbc_enabled(struct drm_device *dev) |
85208be0 ED |
122 | { |
123 | struct drm_i915_private *dev_priv = dev->dev_private; | |
124 | ||
125 | return I915_READ(FBC_CONTROL) & FBC_CTL_EN; | |
126 | } | |
127 | ||
1fa61106 | 128 | static void g4x_enable_fbc(struct drm_crtc *crtc, unsigned long interval) |
85208be0 ED |
129 | { |
130 | struct drm_device *dev = crtc->dev; | |
131 | struct drm_i915_private *dev_priv = dev->dev_private; | |
132 | struct drm_framebuffer *fb = crtc->fb; | |
133 | struct intel_framebuffer *intel_fb = to_intel_framebuffer(fb); | |
134 | struct drm_i915_gem_object *obj = intel_fb->obj; | |
135 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc); | |
136 | int plane = intel_crtc->plane == 0 ? DPFC_CTL_PLANEA : DPFC_CTL_PLANEB; | |
137 | unsigned long stall_watermark = 200; | |
138 | u32 dpfc_ctl; | |
139 | ||
140 | dpfc_ctl = plane | DPFC_SR_EN | DPFC_CTL_LIMIT_1X; | |
141 | dpfc_ctl |= DPFC_CTL_FENCE_EN | obj->fence_reg; | |
142 | I915_WRITE(DPFC_CHICKEN, DPFC_HT_MODIFY); | |
143 | ||
144 | I915_WRITE(DPFC_RECOMP_CTL, DPFC_RECOMP_STALL_EN | | |
145 | (stall_watermark << DPFC_RECOMP_STALL_WM_SHIFT) | | |
146 | (interval << DPFC_RECOMP_TIMER_COUNT_SHIFT)); | |
147 | I915_WRITE(DPFC_FENCE_YOFF, crtc->y); | |
148 | ||
149 | /* enable it... */ | |
150 | I915_WRITE(DPFC_CONTROL, I915_READ(DPFC_CONTROL) | DPFC_CTL_EN); | |
151 | ||
84f44ce7 | 152 | DRM_DEBUG_KMS("enabled fbc on plane %c\n", plane_name(intel_crtc->plane)); |
85208be0 ED |
153 | } |
154 | ||
1fa61106 | 155 | static void g4x_disable_fbc(struct drm_device *dev) |
85208be0 ED |
156 | { |
157 | struct drm_i915_private *dev_priv = dev->dev_private; | |
158 | u32 dpfc_ctl; | |
159 | ||
160 | /* Disable compression */ | |
161 | dpfc_ctl = I915_READ(DPFC_CONTROL); | |
162 | if (dpfc_ctl & DPFC_CTL_EN) { | |
163 | dpfc_ctl &= ~DPFC_CTL_EN; | |
164 | I915_WRITE(DPFC_CONTROL, dpfc_ctl); | |
165 | ||
166 | DRM_DEBUG_KMS("disabled FBC\n"); | |
167 | } | |
168 | } | |
169 | ||
1fa61106 | 170 | static bool g4x_fbc_enabled(struct drm_device *dev) |
85208be0 ED |
171 | { |
172 | struct drm_i915_private *dev_priv = dev->dev_private; | |
173 | ||
174 | return I915_READ(DPFC_CONTROL) & DPFC_CTL_EN; | |
175 | } | |
176 | ||
177 | static void sandybridge_blit_fbc_update(struct drm_device *dev) | |
178 | { | |
179 | struct drm_i915_private *dev_priv = dev->dev_private; | |
180 | u32 blt_ecoskpd; | |
181 | ||
182 | /* Make sure blitter notifies FBC of writes */ | |
183 | gen6_gt_force_wake_get(dev_priv); | |
184 | blt_ecoskpd = I915_READ(GEN6_BLITTER_ECOSKPD); | |
185 | blt_ecoskpd |= GEN6_BLITTER_FBC_NOTIFY << | |
186 | GEN6_BLITTER_LOCK_SHIFT; | |
187 | I915_WRITE(GEN6_BLITTER_ECOSKPD, blt_ecoskpd); | |
188 | blt_ecoskpd |= GEN6_BLITTER_FBC_NOTIFY; | |
189 | I915_WRITE(GEN6_BLITTER_ECOSKPD, blt_ecoskpd); | |
190 | blt_ecoskpd &= ~(GEN6_BLITTER_FBC_NOTIFY << | |
191 | GEN6_BLITTER_LOCK_SHIFT); | |
192 | I915_WRITE(GEN6_BLITTER_ECOSKPD, blt_ecoskpd); | |
193 | POSTING_READ(GEN6_BLITTER_ECOSKPD); | |
194 | gen6_gt_force_wake_put(dev_priv); | |
195 | } | |
196 | ||
1fa61106 | 197 | static void ironlake_enable_fbc(struct drm_crtc *crtc, unsigned long interval) |
85208be0 ED |
198 | { |
199 | struct drm_device *dev = crtc->dev; | |
200 | struct drm_i915_private *dev_priv = dev->dev_private; | |
201 | struct drm_framebuffer *fb = crtc->fb; | |
202 | struct intel_framebuffer *intel_fb = to_intel_framebuffer(fb); | |
203 | struct drm_i915_gem_object *obj = intel_fb->obj; | |
204 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc); | |
205 | int plane = intel_crtc->plane == 0 ? DPFC_CTL_PLANEA : DPFC_CTL_PLANEB; | |
206 | unsigned long stall_watermark = 200; | |
207 | u32 dpfc_ctl; | |
208 | ||
209 | dpfc_ctl = I915_READ(ILK_DPFC_CONTROL); | |
210 | dpfc_ctl &= DPFC_RESERVED; | |
211 | dpfc_ctl |= (plane | DPFC_CTL_LIMIT_1X); | |
212 | /* Set persistent mode for front-buffer rendering, ala X. */ | |
213 | dpfc_ctl |= DPFC_CTL_PERSISTENT_MODE; | |
214 | dpfc_ctl |= (DPFC_CTL_FENCE_EN | obj->fence_reg); | |
215 | I915_WRITE(ILK_DPFC_CHICKEN, DPFC_HT_MODIFY); | |
216 | ||
217 | I915_WRITE(ILK_DPFC_RECOMP_CTL, DPFC_RECOMP_STALL_EN | | |
218 | (stall_watermark << DPFC_RECOMP_STALL_WM_SHIFT) | | |
219 | (interval << DPFC_RECOMP_TIMER_COUNT_SHIFT)); | |
220 | I915_WRITE(ILK_DPFC_FENCE_YOFF, crtc->y); | |
f343c5f6 | 221 | I915_WRITE(ILK_FBC_RT_BASE, i915_gem_obj_ggtt_offset(obj) | ILK_FBC_RT_VALID); |
85208be0 ED |
222 | /* enable it... */ |
223 | I915_WRITE(ILK_DPFC_CONTROL, dpfc_ctl | DPFC_CTL_EN); | |
224 | ||
225 | if (IS_GEN6(dev)) { | |
226 | I915_WRITE(SNB_DPFC_CTL_SA, | |
227 | SNB_CPU_FENCE_ENABLE | obj->fence_reg); | |
228 | I915_WRITE(DPFC_CPU_FENCE_OFFSET, crtc->y); | |
229 | sandybridge_blit_fbc_update(dev); | |
230 | } | |
231 | ||
84f44ce7 | 232 | DRM_DEBUG_KMS("enabled fbc on plane %c\n", plane_name(intel_crtc->plane)); |
85208be0 ED |
233 | } |
234 | ||
1fa61106 | 235 | static void ironlake_disable_fbc(struct drm_device *dev) |
85208be0 ED |
236 | { |
237 | struct drm_i915_private *dev_priv = dev->dev_private; | |
238 | u32 dpfc_ctl; | |
239 | ||
240 | /* Disable compression */ | |
241 | dpfc_ctl = I915_READ(ILK_DPFC_CONTROL); | |
242 | if (dpfc_ctl & DPFC_CTL_EN) { | |
243 | dpfc_ctl &= ~DPFC_CTL_EN; | |
244 | I915_WRITE(ILK_DPFC_CONTROL, dpfc_ctl); | |
245 | ||
b74ea102 | 246 | if (IS_IVYBRIDGE(dev)) |
7dd23ba0 | 247 | /* WaFbcDisableDpfcClockGating:ivb */ |
b74ea102 RV |
248 | I915_WRITE(ILK_DSPCLK_GATE_D, |
249 | I915_READ(ILK_DSPCLK_GATE_D) & | |
250 | ~ILK_DPFCUNIT_CLOCK_GATE_DISABLE); | |
251 | ||
d89f2071 | 252 | if (IS_HASWELL(dev)) |
7dd23ba0 | 253 | /* WaFbcDisableDpfcClockGating:hsw */ |
d89f2071 RV |
254 | I915_WRITE(HSW_CLKGATE_DISABLE_PART_1, |
255 | I915_READ(HSW_CLKGATE_DISABLE_PART_1) & | |
256 | ~HSW_DPFC_GATING_DISABLE); | |
257 | ||
85208be0 ED |
258 | DRM_DEBUG_KMS("disabled FBC\n"); |
259 | } | |
260 | } | |
261 | ||
1fa61106 | 262 | static bool ironlake_fbc_enabled(struct drm_device *dev) |
85208be0 ED |
263 | { |
264 | struct drm_i915_private *dev_priv = dev->dev_private; | |
265 | ||
266 | return I915_READ(ILK_DPFC_CONTROL) & DPFC_CTL_EN; | |
267 | } | |
268 | ||
abe959c7 RV |
269 | static void gen7_enable_fbc(struct drm_crtc *crtc, unsigned long interval) |
270 | { | |
271 | struct drm_device *dev = crtc->dev; | |
272 | struct drm_i915_private *dev_priv = dev->dev_private; | |
273 | struct drm_framebuffer *fb = crtc->fb; | |
274 | struct intel_framebuffer *intel_fb = to_intel_framebuffer(fb); | |
275 | struct drm_i915_gem_object *obj = intel_fb->obj; | |
276 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc); | |
277 | ||
f343c5f6 | 278 | I915_WRITE(IVB_FBC_RT_BASE, i915_gem_obj_ggtt_offset(obj)); |
abe959c7 RV |
279 | |
280 | I915_WRITE(ILK_DPFC_CONTROL, DPFC_CTL_EN | DPFC_CTL_LIMIT_1X | | |
281 | IVB_DPFC_CTL_FENCE_EN | | |
282 | intel_crtc->plane << IVB_DPFC_CTL_PLANE_SHIFT); | |
283 | ||
891348b2 | 284 | if (IS_IVYBRIDGE(dev)) { |
7dd23ba0 | 285 | /* WaFbcAsynchFlipDisableFbcQueue:ivb */ |
891348b2 | 286 | I915_WRITE(ILK_DISPLAY_CHICKEN1, ILK_FBCQ_DIS); |
7dd23ba0 | 287 | /* WaFbcDisableDpfcClockGating:ivb */ |
891348b2 RV |
288 | I915_WRITE(ILK_DSPCLK_GATE_D, |
289 | I915_READ(ILK_DSPCLK_GATE_D) | | |
290 | ILK_DPFCUNIT_CLOCK_GATE_DISABLE); | |
28554164 | 291 | } else { |
7dd23ba0 | 292 | /* WaFbcAsynchFlipDisableFbcQueue:hsw */ |
28554164 RV |
293 | I915_WRITE(HSW_PIPE_SLICE_CHICKEN_1(intel_crtc->pipe), |
294 | HSW_BYPASS_FBC_QUEUE); | |
7dd23ba0 | 295 | /* WaFbcDisableDpfcClockGating:hsw */ |
d89f2071 RV |
296 | I915_WRITE(HSW_CLKGATE_DISABLE_PART_1, |
297 | I915_READ(HSW_CLKGATE_DISABLE_PART_1) | | |
298 | HSW_DPFC_GATING_DISABLE); | |
891348b2 | 299 | } |
b74ea102 | 300 | |
abe959c7 RV |
301 | I915_WRITE(SNB_DPFC_CTL_SA, |
302 | SNB_CPU_FENCE_ENABLE | obj->fence_reg); | |
303 | I915_WRITE(DPFC_CPU_FENCE_OFFSET, crtc->y); | |
304 | ||
305 | sandybridge_blit_fbc_update(dev); | |
306 | ||
307 | DRM_DEBUG_KMS("enabled fbc on plane %d\n", intel_crtc->plane); | |
308 | } | |
309 | ||
85208be0 ED |
310 | bool intel_fbc_enabled(struct drm_device *dev) |
311 | { | |
312 | struct drm_i915_private *dev_priv = dev->dev_private; | |
313 | ||
314 | if (!dev_priv->display.fbc_enabled) | |
315 | return false; | |
316 | ||
317 | return dev_priv->display.fbc_enabled(dev); | |
318 | } | |
319 | ||
320 | static void intel_fbc_work_fn(struct work_struct *__work) | |
321 | { | |
322 | struct intel_fbc_work *work = | |
323 | container_of(to_delayed_work(__work), | |
324 | struct intel_fbc_work, work); | |
325 | struct drm_device *dev = work->crtc->dev; | |
326 | struct drm_i915_private *dev_priv = dev->dev_private; | |
327 | ||
328 | mutex_lock(&dev->struct_mutex); | |
5c3fe8b0 | 329 | if (work == dev_priv->fbc.fbc_work) { |
85208be0 ED |
330 | /* Double check that we haven't switched fb without cancelling |
331 | * the prior work. | |
332 | */ | |
333 | if (work->crtc->fb == work->fb) { | |
334 | dev_priv->display.enable_fbc(work->crtc, | |
335 | work->interval); | |
336 | ||
5c3fe8b0 BW |
337 | dev_priv->fbc.plane = to_intel_crtc(work->crtc)->plane; |
338 | dev_priv->fbc.fb_id = work->crtc->fb->base.id; | |
339 | dev_priv->fbc.y = work->crtc->y; | |
85208be0 ED |
340 | } |
341 | ||
5c3fe8b0 | 342 | dev_priv->fbc.fbc_work = NULL; |
85208be0 ED |
343 | } |
344 | mutex_unlock(&dev->struct_mutex); | |
345 | ||
346 | kfree(work); | |
347 | } | |
348 | ||
349 | static void intel_cancel_fbc_work(struct drm_i915_private *dev_priv) | |
350 | { | |
5c3fe8b0 | 351 | if (dev_priv->fbc.fbc_work == NULL) |
85208be0 ED |
352 | return; |
353 | ||
354 | DRM_DEBUG_KMS("cancelling pending FBC enable\n"); | |
355 | ||
356 | /* Synchronisation is provided by struct_mutex and checking of | |
5c3fe8b0 | 357 | * dev_priv->fbc.fbc_work, so we can perform the cancellation |
85208be0 ED |
358 | * entirely asynchronously. |
359 | */ | |
5c3fe8b0 | 360 | if (cancel_delayed_work(&dev_priv->fbc.fbc_work->work)) |
85208be0 | 361 | /* tasklet was killed before being run, clean up */ |
5c3fe8b0 | 362 | kfree(dev_priv->fbc.fbc_work); |
85208be0 ED |
363 | |
364 | /* Mark the work as no longer wanted so that if it does | |
365 | * wake-up (because the work was already running and waiting | |
366 | * for our mutex), it will discover that is no longer | |
367 | * necessary to run. | |
368 | */ | |
5c3fe8b0 | 369 | dev_priv->fbc.fbc_work = NULL; |
85208be0 ED |
370 | } |
371 | ||
b63fb44c | 372 | static void intel_enable_fbc(struct drm_crtc *crtc, unsigned long interval) |
85208be0 ED |
373 | { |
374 | struct intel_fbc_work *work; | |
375 | struct drm_device *dev = crtc->dev; | |
376 | struct drm_i915_private *dev_priv = dev->dev_private; | |
377 | ||
378 | if (!dev_priv->display.enable_fbc) | |
379 | return; | |
380 | ||
381 | intel_cancel_fbc_work(dev_priv); | |
382 | ||
383 | work = kzalloc(sizeof *work, GFP_KERNEL); | |
384 | if (work == NULL) { | |
6cdcb5e7 | 385 | DRM_ERROR("Failed to allocate FBC work structure\n"); |
85208be0 ED |
386 | dev_priv->display.enable_fbc(crtc, interval); |
387 | return; | |
388 | } | |
389 | ||
390 | work->crtc = crtc; | |
391 | work->fb = crtc->fb; | |
392 | work->interval = interval; | |
393 | INIT_DELAYED_WORK(&work->work, intel_fbc_work_fn); | |
394 | ||
5c3fe8b0 | 395 | dev_priv->fbc.fbc_work = work; |
85208be0 | 396 | |
85208be0 ED |
397 | /* Delay the actual enabling to let pageflipping cease and the |
398 | * display to settle before starting the compression. Note that | |
399 | * this delay also serves a second purpose: it allows for a | |
400 | * vblank to pass after disabling the FBC before we attempt | |
401 | * to modify the control registers. | |
402 | * | |
403 | * A more complicated solution would involve tracking vblanks | |
404 | * following the termination of the page-flipping sequence | |
405 | * and indeed performing the enable as a co-routine and not | |
406 | * waiting synchronously upon the vblank. | |
7457d617 DL |
407 | * |
408 | * WaFbcWaitForVBlankBeforeEnable:ilk,snb | |
85208be0 ED |
409 | */ |
410 | schedule_delayed_work(&work->work, msecs_to_jiffies(50)); | |
411 | } | |
412 | ||
413 | void intel_disable_fbc(struct drm_device *dev) | |
414 | { | |
415 | struct drm_i915_private *dev_priv = dev->dev_private; | |
416 | ||
417 | intel_cancel_fbc_work(dev_priv); | |
418 | ||
419 | if (!dev_priv->display.disable_fbc) | |
420 | return; | |
421 | ||
422 | dev_priv->display.disable_fbc(dev); | |
5c3fe8b0 | 423 | dev_priv->fbc.plane = -1; |
85208be0 ED |
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 | |
f85da868 | 436 | * - framebuffer <= max_hdisplay in width, max_vdisplay in height |
85208be0 ED |
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; | |
f85da868 | 453 | unsigned int max_hdisplay, max_vdisplay; |
85208be0 | 454 | |
85208be0 ED |
455 | if (!i915_powersave) |
456 | return; | |
457 | ||
458 | if (!I915_HAS_FBC(dev)) | |
459 | return; | |
460 | ||
461 | /* | |
462 | * If FBC is already on, we just have to verify that we can | |
463 | * keep it that way... | |
464 | * Need to disable if: | |
465 | * - more than one pipe is active | |
466 | * - changing FBC params (stride, fence, mode) | |
467 | * - new fb is too large to fit in compressed buffer | |
468 | * - going to an unsupported config (interlace, pixel multiply, etc.) | |
469 | */ | |
470 | list_for_each_entry(tmp_crtc, &dev->mode_config.crtc_list, head) { | |
3490ea5d CW |
471 | if (intel_crtc_active(tmp_crtc) && |
472 | !to_intel_crtc(tmp_crtc)->primary_disabled) { | |
85208be0 ED |
473 | if (crtc) { |
474 | DRM_DEBUG_KMS("more than one pipe active, disabling compression\n"); | |
5c3fe8b0 BW |
475 | dev_priv->fbc.no_fbc_reason = |
476 | FBC_MULTIPLE_PIPES; | |
85208be0 ED |
477 | goto out_disable; |
478 | } | |
479 | crtc = tmp_crtc; | |
480 | } | |
481 | } | |
482 | ||
483 | if (!crtc || crtc->fb == NULL) { | |
484 | DRM_DEBUG_KMS("no output, disabling\n"); | |
5c3fe8b0 | 485 | dev_priv->fbc.no_fbc_reason = FBC_NO_OUTPUT; |
85208be0 ED |
486 | goto out_disable; |
487 | } | |
488 | ||
489 | intel_crtc = to_intel_crtc(crtc); | |
490 | fb = crtc->fb; | |
491 | intel_fb = to_intel_framebuffer(fb); | |
492 | obj = intel_fb->obj; | |
493 | ||
8a5729a3 DL |
494 | if (i915_enable_fbc < 0 && |
495 | INTEL_INFO(dev)->gen <= 7 && !IS_HASWELL(dev)) { | |
496 | DRM_DEBUG_KMS("disabled per chip default\n"); | |
5c3fe8b0 | 497 | dev_priv->fbc.no_fbc_reason = FBC_CHIP_DEFAULT; |
8a5729a3 | 498 | goto out_disable; |
85208be0 | 499 | } |
8a5729a3 | 500 | if (!i915_enable_fbc) { |
85208be0 | 501 | DRM_DEBUG_KMS("fbc disabled per module param\n"); |
5c3fe8b0 | 502 | dev_priv->fbc.no_fbc_reason = FBC_MODULE_PARAM; |
85208be0 ED |
503 | goto out_disable; |
504 | } | |
85208be0 ED |
505 | if ((crtc->mode.flags & DRM_MODE_FLAG_INTERLACE) || |
506 | (crtc->mode.flags & DRM_MODE_FLAG_DBLSCAN)) { | |
507 | DRM_DEBUG_KMS("mode incompatible with compression, " | |
508 | "disabling\n"); | |
5c3fe8b0 | 509 | dev_priv->fbc.no_fbc_reason = FBC_UNSUPPORTED_MODE; |
85208be0 ED |
510 | goto out_disable; |
511 | } | |
f85da868 PZ |
512 | |
513 | if (IS_G4X(dev) || INTEL_INFO(dev)->gen >= 5) { | |
514 | max_hdisplay = 4096; | |
515 | max_vdisplay = 2048; | |
516 | } else { | |
517 | max_hdisplay = 2048; | |
518 | max_vdisplay = 1536; | |
519 | } | |
520 | if ((crtc->mode.hdisplay > max_hdisplay) || | |
521 | (crtc->mode.vdisplay > max_vdisplay)) { | |
85208be0 | 522 | DRM_DEBUG_KMS("mode too large for compression, disabling\n"); |
5c3fe8b0 | 523 | dev_priv->fbc.no_fbc_reason = FBC_MODE_TOO_LARGE; |
85208be0 ED |
524 | goto out_disable; |
525 | } | |
891348b2 RV |
526 | if ((IS_I915GM(dev) || IS_I945GM(dev) || IS_HASWELL(dev)) && |
527 | intel_crtc->plane != 0) { | |
85208be0 | 528 | DRM_DEBUG_KMS("plane not 0, disabling compression\n"); |
5c3fe8b0 | 529 | dev_priv->fbc.no_fbc_reason = FBC_BAD_PLANE; |
85208be0 ED |
530 | goto out_disable; |
531 | } | |
532 | ||
533 | /* The use of a CPU fence is mandatory in order to detect writes | |
534 | * by the CPU to the scanout and trigger updates to the FBC. | |
535 | */ | |
536 | if (obj->tiling_mode != I915_TILING_X || | |
537 | obj->fence_reg == I915_FENCE_REG_NONE) { | |
538 | DRM_DEBUG_KMS("framebuffer not tiled or fenced, disabling compression\n"); | |
5c3fe8b0 | 539 | dev_priv->fbc.no_fbc_reason = FBC_NOT_TILED; |
85208be0 ED |
540 | goto out_disable; |
541 | } | |
542 | ||
543 | /* If the kernel debugger is active, always disable compression */ | |
544 | if (in_dbg_master()) | |
545 | goto out_disable; | |
546 | ||
11be49eb | 547 | if (i915_gem_stolen_setup_compression(dev, intel_fb->obj->base.size)) { |
11be49eb | 548 | DRM_DEBUG_KMS("framebuffer too large, disabling compression\n"); |
5c3fe8b0 | 549 | dev_priv->fbc.no_fbc_reason = FBC_STOLEN_TOO_SMALL; |
11be49eb CW |
550 | goto out_disable; |
551 | } | |
552 | ||
85208be0 ED |
553 | /* If the scanout has not changed, don't modify the FBC settings. |
554 | * Note that we make the fundamental assumption that the fb->obj | |
555 | * cannot be unpinned (and have its GTT offset and fence revoked) | |
556 | * without first being decoupled from the scanout and FBC disabled. | |
557 | */ | |
5c3fe8b0 BW |
558 | if (dev_priv->fbc.plane == intel_crtc->plane && |
559 | dev_priv->fbc.fb_id == fb->base.id && | |
560 | dev_priv->fbc.y == crtc->y) | |
85208be0 ED |
561 | return; |
562 | ||
563 | if (intel_fbc_enabled(dev)) { | |
564 | /* We update FBC along two paths, after changing fb/crtc | |
565 | * configuration (modeswitching) and after page-flipping | |
566 | * finishes. For the latter, we know that not only did | |
567 | * we disable the FBC at the start of the page-flip | |
568 | * sequence, but also more than one vblank has passed. | |
569 | * | |
570 | * For the former case of modeswitching, it is possible | |
571 | * to switch between two FBC valid configurations | |
572 | * instantaneously so we do need to disable the FBC | |
573 | * before we can modify its control registers. We also | |
574 | * have to wait for the next vblank for that to take | |
575 | * effect. However, since we delay enabling FBC we can | |
576 | * assume that a vblank has passed since disabling and | |
577 | * that we can safely alter the registers in the deferred | |
578 | * callback. | |
579 | * | |
580 | * In the scenario that we go from a valid to invalid | |
581 | * and then back to valid FBC configuration we have | |
582 | * no strict enforcement that a vblank occurred since | |
583 | * disabling the FBC. However, along all current pipe | |
584 | * disabling paths we do need to wait for a vblank at | |
585 | * some point. And we wait before enabling FBC anyway. | |
586 | */ | |
587 | DRM_DEBUG_KMS("disabling active FBC for update\n"); | |
588 | intel_disable_fbc(dev); | |
589 | } | |
590 | ||
591 | intel_enable_fbc(crtc, 500); | |
592 | return; | |
593 | ||
594 | out_disable: | |
595 | /* Multiple disables should be harmless */ | |
596 | if (intel_fbc_enabled(dev)) { | |
597 | DRM_DEBUG_KMS("unsupported config, disabling FBC\n"); | |
598 | intel_disable_fbc(dev); | |
599 | } | |
11be49eb | 600 | i915_gem_stolen_cleanup_compression(dev); |
85208be0 ED |
601 | } |
602 | ||
c921aba8 DV |
603 | static void i915_pineview_get_mem_freq(struct drm_device *dev) |
604 | { | |
605 | drm_i915_private_t *dev_priv = dev->dev_private; | |
606 | u32 tmp; | |
607 | ||
608 | tmp = I915_READ(CLKCFG); | |
609 | ||
610 | switch (tmp & CLKCFG_FSB_MASK) { | |
611 | case CLKCFG_FSB_533: | |
612 | dev_priv->fsb_freq = 533; /* 133*4 */ | |
613 | break; | |
614 | case CLKCFG_FSB_800: | |
615 | dev_priv->fsb_freq = 800; /* 200*4 */ | |
616 | break; | |
617 | case CLKCFG_FSB_667: | |
618 | dev_priv->fsb_freq = 667; /* 167*4 */ | |
619 | break; | |
620 | case CLKCFG_FSB_400: | |
621 | dev_priv->fsb_freq = 400; /* 100*4 */ | |
622 | break; | |
623 | } | |
624 | ||
625 | switch (tmp & CLKCFG_MEM_MASK) { | |
626 | case CLKCFG_MEM_533: | |
627 | dev_priv->mem_freq = 533; | |
628 | break; | |
629 | case CLKCFG_MEM_667: | |
630 | dev_priv->mem_freq = 667; | |
631 | break; | |
632 | case CLKCFG_MEM_800: | |
633 | dev_priv->mem_freq = 800; | |
634 | break; | |
635 | } | |
636 | ||
637 | /* detect pineview DDR3 setting */ | |
638 | tmp = I915_READ(CSHRDDR3CTL); | |
639 | dev_priv->is_ddr3 = (tmp & CSHRDDR3CTL_DDR3) ? 1 : 0; | |
640 | } | |
641 | ||
642 | static void i915_ironlake_get_mem_freq(struct drm_device *dev) | |
643 | { | |
644 | drm_i915_private_t *dev_priv = dev->dev_private; | |
645 | u16 ddrpll, csipll; | |
646 | ||
647 | ddrpll = I915_READ16(DDRMPLL1); | |
648 | csipll = I915_READ16(CSIPLL0); | |
649 | ||
650 | switch (ddrpll & 0xff) { | |
651 | case 0xc: | |
652 | dev_priv->mem_freq = 800; | |
653 | break; | |
654 | case 0x10: | |
655 | dev_priv->mem_freq = 1066; | |
656 | break; | |
657 | case 0x14: | |
658 | dev_priv->mem_freq = 1333; | |
659 | break; | |
660 | case 0x18: | |
661 | dev_priv->mem_freq = 1600; | |
662 | break; | |
663 | default: | |
664 | DRM_DEBUG_DRIVER("unknown memory frequency 0x%02x\n", | |
665 | ddrpll & 0xff); | |
666 | dev_priv->mem_freq = 0; | |
667 | break; | |
668 | } | |
669 | ||
20e4d407 | 670 | dev_priv->ips.r_t = dev_priv->mem_freq; |
c921aba8 DV |
671 | |
672 | switch (csipll & 0x3ff) { | |
673 | case 0x00c: | |
674 | dev_priv->fsb_freq = 3200; | |
675 | break; | |
676 | case 0x00e: | |
677 | dev_priv->fsb_freq = 3733; | |
678 | break; | |
679 | case 0x010: | |
680 | dev_priv->fsb_freq = 4266; | |
681 | break; | |
682 | case 0x012: | |
683 | dev_priv->fsb_freq = 4800; | |
684 | break; | |
685 | case 0x014: | |
686 | dev_priv->fsb_freq = 5333; | |
687 | break; | |
688 | case 0x016: | |
689 | dev_priv->fsb_freq = 5866; | |
690 | break; | |
691 | case 0x018: | |
692 | dev_priv->fsb_freq = 6400; | |
693 | break; | |
694 | default: | |
695 | DRM_DEBUG_DRIVER("unknown fsb frequency 0x%04x\n", | |
696 | csipll & 0x3ff); | |
697 | dev_priv->fsb_freq = 0; | |
698 | break; | |
699 | } | |
700 | ||
701 | if (dev_priv->fsb_freq == 3200) { | |
20e4d407 | 702 | dev_priv->ips.c_m = 0; |
c921aba8 | 703 | } else if (dev_priv->fsb_freq > 3200 && dev_priv->fsb_freq <= 4800) { |
20e4d407 | 704 | dev_priv->ips.c_m = 1; |
c921aba8 | 705 | } else { |
20e4d407 | 706 | dev_priv->ips.c_m = 2; |
c921aba8 DV |
707 | } |
708 | } | |
709 | ||
b445e3b0 ED |
710 | static const struct cxsr_latency cxsr_latency_table[] = { |
711 | {1, 0, 800, 400, 3382, 33382, 3983, 33983}, /* DDR2-400 SC */ | |
712 | {1, 0, 800, 667, 3354, 33354, 3807, 33807}, /* DDR2-667 SC */ | |
713 | {1, 0, 800, 800, 3347, 33347, 3763, 33763}, /* DDR2-800 SC */ | |
714 | {1, 1, 800, 667, 6420, 36420, 6873, 36873}, /* DDR3-667 SC */ | |
715 | {1, 1, 800, 800, 5902, 35902, 6318, 36318}, /* DDR3-800 SC */ | |
716 | ||
717 | {1, 0, 667, 400, 3400, 33400, 4021, 34021}, /* DDR2-400 SC */ | |
718 | {1, 0, 667, 667, 3372, 33372, 3845, 33845}, /* DDR2-667 SC */ | |
719 | {1, 0, 667, 800, 3386, 33386, 3822, 33822}, /* DDR2-800 SC */ | |
720 | {1, 1, 667, 667, 6438, 36438, 6911, 36911}, /* DDR3-667 SC */ | |
721 | {1, 1, 667, 800, 5941, 35941, 6377, 36377}, /* DDR3-800 SC */ | |
722 | ||
723 | {1, 0, 400, 400, 3472, 33472, 4173, 34173}, /* DDR2-400 SC */ | |
724 | {1, 0, 400, 667, 3443, 33443, 3996, 33996}, /* DDR2-667 SC */ | |
725 | {1, 0, 400, 800, 3430, 33430, 3946, 33946}, /* DDR2-800 SC */ | |
726 | {1, 1, 400, 667, 6509, 36509, 7062, 37062}, /* DDR3-667 SC */ | |
727 | {1, 1, 400, 800, 5985, 35985, 6501, 36501}, /* DDR3-800 SC */ | |
728 | ||
729 | {0, 0, 800, 400, 3438, 33438, 4065, 34065}, /* DDR2-400 SC */ | |
730 | {0, 0, 800, 667, 3410, 33410, 3889, 33889}, /* DDR2-667 SC */ | |
731 | {0, 0, 800, 800, 3403, 33403, 3845, 33845}, /* DDR2-800 SC */ | |
732 | {0, 1, 800, 667, 6476, 36476, 6955, 36955}, /* DDR3-667 SC */ | |
733 | {0, 1, 800, 800, 5958, 35958, 6400, 36400}, /* DDR3-800 SC */ | |
734 | ||
735 | {0, 0, 667, 400, 3456, 33456, 4103, 34106}, /* DDR2-400 SC */ | |
736 | {0, 0, 667, 667, 3428, 33428, 3927, 33927}, /* DDR2-667 SC */ | |
737 | {0, 0, 667, 800, 3443, 33443, 3905, 33905}, /* DDR2-800 SC */ | |
738 | {0, 1, 667, 667, 6494, 36494, 6993, 36993}, /* DDR3-667 SC */ | |
739 | {0, 1, 667, 800, 5998, 35998, 6460, 36460}, /* DDR3-800 SC */ | |
740 | ||
741 | {0, 0, 400, 400, 3528, 33528, 4255, 34255}, /* DDR2-400 SC */ | |
742 | {0, 0, 400, 667, 3500, 33500, 4079, 34079}, /* DDR2-667 SC */ | |
743 | {0, 0, 400, 800, 3487, 33487, 4029, 34029}, /* DDR2-800 SC */ | |
744 | {0, 1, 400, 667, 6566, 36566, 7145, 37145}, /* DDR3-667 SC */ | |
745 | {0, 1, 400, 800, 6042, 36042, 6584, 36584}, /* DDR3-800 SC */ | |
746 | }; | |
747 | ||
63c62275 | 748 | static const struct cxsr_latency *intel_get_cxsr_latency(int is_desktop, |
b445e3b0 ED |
749 | int is_ddr3, |
750 | int fsb, | |
751 | int mem) | |
752 | { | |
753 | const struct cxsr_latency *latency; | |
754 | int i; | |
755 | ||
756 | if (fsb == 0 || mem == 0) | |
757 | return NULL; | |
758 | ||
759 | for (i = 0; i < ARRAY_SIZE(cxsr_latency_table); i++) { | |
760 | latency = &cxsr_latency_table[i]; | |
761 | if (is_desktop == latency->is_desktop && | |
762 | is_ddr3 == latency->is_ddr3 && | |
763 | fsb == latency->fsb_freq && mem == latency->mem_freq) | |
764 | return latency; | |
765 | } | |
766 | ||
767 | DRM_DEBUG_KMS("Unknown FSB/MEM found, disable CxSR\n"); | |
768 | ||
769 | return NULL; | |
770 | } | |
771 | ||
1fa61106 | 772 | static void pineview_disable_cxsr(struct drm_device *dev) |
b445e3b0 ED |
773 | { |
774 | struct drm_i915_private *dev_priv = dev->dev_private; | |
775 | ||
776 | /* deactivate cxsr */ | |
777 | I915_WRITE(DSPFW3, I915_READ(DSPFW3) & ~PINEVIEW_SELF_REFRESH_EN); | |
778 | } | |
779 | ||
780 | /* | |
781 | * Latency for FIFO fetches is dependent on several factors: | |
782 | * - memory configuration (speed, channels) | |
783 | * - chipset | |
784 | * - current MCH state | |
785 | * It can be fairly high in some situations, so here we assume a fairly | |
786 | * pessimal value. It's a tradeoff between extra memory fetches (if we | |
787 | * set this value too high, the FIFO will fetch frequently to stay full) | |
788 | * and power consumption (set it too low to save power and we might see | |
789 | * FIFO underruns and display "flicker"). | |
790 | * | |
791 | * A value of 5us seems to be a good balance; safe for very low end | |
792 | * platforms but not overly aggressive on lower latency configs. | |
793 | */ | |
794 | static const int latency_ns = 5000; | |
795 | ||
1fa61106 | 796 | static int i9xx_get_fifo_size(struct drm_device *dev, int plane) |
b445e3b0 ED |
797 | { |
798 | struct drm_i915_private *dev_priv = dev->dev_private; | |
799 | uint32_t dsparb = I915_READ(DSPARB); | |
800 | int size; | |
801 | ||
802 | size = dsparb & 0x7f; | |
803 | if (plane) | |
804 | size = ((dsparb >> DSPARB_CSTART_SHIFT) & 0x7f) - size; | |
805 | ||
806 | DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb, | |
807 | plane ? "B" : "A", size); | |
808 | ||
809 | return size; | |
810 | } | |
811 | ||
1fa61106 | 812 | static int i85x_get_fifo_size(struct drm_device *dev, int plane) |
b445e3b0 ED |
813 | { |
814 | struct drm_i915_private *dev_priv = dev->dev_private; | |
815 | uint32_t dsparb = I915_READ(DSPARB); | |
816 | int size; | |
817 | ||
818 | size = dsparb & 0x1ff; | |
819 | if (plane) | |
820 | size = ((dsparb >> DSPARB_BEND_SHIFT) & 0x1ff) - size; | |
821 | size >>= 1; /* Convert to cachelines */ | |
822 | ||
823 | DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb, | |
824 | plane ? "B" : "A", size); | |
825 | ||
826 | return size; | |
827 | } | |
828 | ||
1fa61106 | 829 | static int i845_get_fifo_size(struct drm_device *dev, int plane) |
b445e3b0 ED |
830 | { |
831 | struct drm_i915_private *dev_priv = dev->dev_private; | |
832 | uint32_t dsparb = I915_READ(DSPARB); | |
833 | int size; | |
834 | ||
835 | size = dsparb & 0x7f; | |
836 | size >>= 2; /* Convert to cachelines */ | |
837 | ||
838 | DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb, | |
839 | plane ? "B" : "A", | |
840 | size); | |
841 | ||
842 | return size; | |
843 | } | |
844 | ||
1fa61106 | 845 | static int i830_get_fifo_size(struct drm_device *dev, int plane) |
b445e3b0 ED |
846 | { |
847 | struct drm_i915_private *dev_priv = dev->dev_private; | |
848 | uint32_t dsparb = I915_READ(DSPARB); | |
849 | int size; | |
850 | ||
851 | size = dsparb & 0x7f; | |
852 | size >>= 1; /* Convert to cachelines */ | |
853 | ||
854 | DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb, | |
855 | plane ? "B" : "A", size); | |
856 | ||
857 | return size; | |
858 | } | |
859 | ||
860 | /* Pineview has different values for various configs */ | |
861 | static const struct intel_watermark_params pineview_display_wm = { | |
862 | PINEVIEW_DISPLAY_FIFO, | |
863 | PINEVIEW_MAX_WM, | |
864 | PINEVIEW_DFT_WM, | |
865 | PINEVIEW_GUARD_WM, | |
866 | PINEVIEW_FIFO_LINE_SIZE | |
867 | }; | |
868 | static const struct intel_watermark_params pineview_display_hplloff_wm = { | |
869 | PINEVIEW_DISPLAY_FIFO, | |
870 | PINEVIEW_MAX_WM, | |
871 | PINEVIEW_DFT_HPLLOFF_WM, | |
872 | PINEVIEW_GUARD_WM, | |
873 | PINEVIEW_FIFO_LINE_SIZE | |
874 | }; | |
875 | static const struct intel_watermark_params pineview_cursor_wm = { | |
876 | PINEVIEW_CURSOR_FIFO, | |
877 | PINEVIEW_CURSOR_MAX_WM, | |
878 | PINEVIEW_CURSOR_DFT_WM, | |
879 | PINEVIEW_CURSOR_GUARD_WM, | |
880 | PINEVIEW_FIFO_LINE_SIZE, | |
881 | }; | |
882 | static const struct intel_watermark_params pineview_cursor_hplloff_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 g4x_wm_info = { | |
890 | G4X_FIFO_SIZE, | |
891 | G4X_MAX_WM, | |
892 | G4X_MAX_WM, | |
893 | 2, | |
894 | G4X_FIFO_LINE_SIZE, | |
895 | }; | |
896 | static const struct intel_watermark_params g4x_cursor_wm_info = { | |
897 | I965_CURSOR_FIFO, | |
898 | I965_CURSOR_MAX_WM, | |
899 | I965_CURSOR_DFT_WM, | |
900 | 2, | |
901 | G4X_FIFO_LINE_SIZE, | |
902 | }; | |
903 | static const struct intel_watermark_params valleyview_wm_info = { | |
904 | VALLEYVIEW_FIFO_SIZE, | |
905 | VALLEYVIEW_MAX_WM, | |
906 | VALLEYVIEW_MAX_WM, | |
907 | 2, | |
908 | G4X_FIFO_LINE_SIZE, | |
909 | }; | |
910 | static const struct intel_watermark_params valleyview_cursor_wm_info = { | |
911 | I965_CURSOR_FIFO, | |
912 | VALLEYVIEW_CURSOR_MAX_WM, | |
913 | I965_CURSOR_DFT_WM, | |
914 | 2, | |
915 | G4X_FIFO_LINE_SIZE, | |
916 | }; | |
917 | static const struct intel_watermark_params i965_cursor_wm_info = { | |
918 | I965_CURSOR_FIFO, | |
919 | I965_CURSOR_MAX_WM, | |
920 | I965_CURSOR_DFT_WM, | |
921 | 2, | |
922 | I915_FIFO_LINE_SIZE, | |
923 | }; | |
924 | static const struct intel_watermark_params i945_wm_info = { | |
925 | I945_FIFO_SIZE, | |
926 | I915_MAX_WM, | |
927 | 1, | |
928 | 2, | |
929 | I915_FIFO_LINE_SIZE | |
930 | }; | |
931 | static const struct intel_watermark_params i915_wm_info = { | |
932 | I915_FIFO_SIZE, | |
933 | I915_MAX_WM, | |
934 | 1, | |
935 | 2, | |
936 | I915_FIFO_LINE_SIZE | |
937 | }; | |
938 | static const struct intel_watermark_params i855_wm_info = { | |
939 | I855GM_FIFO_SIZE, | |
940 | I915_MAX_WM, | |
941 | 1, | |
942 | 2, | |
943 | I830_FIFO_LINE_SIZE | |
944 | }; | |
945 | static const struct intel_watermark_params i830_wm_info = { | |
946 | I830_FIFO_SIZE, | |
947 | I915_MAX_WM, | |
948 | 1, | |
949 | 2, | |
950 | I830_FIFO_LINE_SIZE | |
951 | }; | |
952 | ||
953 | static const struct intel_watermark_params ironlake_display_wm_info = { | |
954 | ILK_DISPLAY_FIFO, | |
955 | ILK_DISPLAY_MAXWM, | |
956 | ILK_DISPLAY_DFTWM, | |
957 | 2, | |
958 | ILK_FIFO_LINE_SIZE | |
959 | }; | |
960 | static const struct intel_watermark_params ironlake_cursor_wm_info = { | |
961 | ILK_CURSOR_FIFO, | |
962 | ILK_CURSOR_MAXWM, | |
963 | ILK_CURSOR_DFTWM, | |
964 | 2, | |
965 | ILK_FIFO_LINE_SIZE | |
966 | }; | |
967 | static const struct intel_watermark_params ironlake_display_srwm_info = { | |
968 | ILK_DISPLAY_SR_FIFO, | |
969 | ILK_DISPLAY_MAX_SRWM, | |
970 | ILK_DISPLAY_DFT_SRWM, | |
971 | 2, | |
972 | ILK_FIFO_LINE_SIZE | |
973 | }; | |
974 | static const struct intel_watermark_params ironlake_cursor_srwm_info = { | |
975 | ILK_CURSOR_SR_FIFO, | |
976 | ILK_CURSOR_MAX_SRWM, | |
977 | ILK_CURSOR_DFT_SRWM, | |
978 | 2, | |
979 | ILK_FIFO_LINE_SIZE | |
980 | }; | |
981 | ||
982 | static const struct intel_watermark_params sandybridge_display_wm_info = { | |
983 | SNB_DISPLAY_FIFO, | |
984 | SNB_DISPLAY_MAXWM, | |
985 | SNB_DISPLAY_DFTWM, | |
986 | 2, | |
987 | SNB_FIFO_LINE_SIZE | |
988 | }; | |
989 | static const struct intel_watermark_params sandybridge_cursor_wm_info = { | |
990 | SNB_CURSOR_FIFO, | |
991 | SNB_CURSOR_MAXWM, | |
992 | SNB_CURSOR_DFTWM, | |
993 | 2, | |
994 | SNB_FIFO_LINE_SIZE | |
995 | }; | |
996 | static const struct intel_watermark_params sandybridge_display_srwm_info = { | |
997 | SNB_DISPLAY_SR_FIFO, | |
998 | SNB_DISPLAY_MAX_SRWM, | |
999 | SNB_DISPLAY_DFT_SRWM, | |
1000 | 2, | |
1001 | SNB_FIFO_LINE_SIZE | |
1002 | }; | |
1003 | static const struct intel_watermark_params sandybridge_cursor_srwm_info = { | |
1004 | SNB_CURSOR_SR_FIFO, | |
1005 | SNB_CURSOR_MAX_SRWM, | |
1006 | SNB_CURSOR_DFT_SRWM, | |
1007 | 2, | |
1008 | SNB_FIFO_LINE_SIZE | |
1009 | }; | |
1010 | ||
1011 | ||
1012 | /** | |
1013 | * intel_calculate_wm - calculate watermark level | |
1014 | * @clock_in_khz: pixel clock | |
1015 | * @wm: chip FIFO params | |
1016 | * @pixel_size: display pixel size | |
1017 | * @latency_ns: memory latency for the platform | |
1018 | * | |
1019 | * Calculate the watermark level (the level at which the display plane will | |
1020 | * start fetching from memory again). Each chip has a different display | |
1021 | * FIFO size and allocation, so the caller needs to figure that out and pass | |
1022 | * in the correct intel_watermark_params structure. | |
1023 | * | |
1024 | * As the pixel clock runs, the FIFO will be drained at a rate that depends | |
1025 | * on the pixel size. When it reaches the watermark level, it'll start | |
1026 | * fetching FIFO line sized based chunks from memory until the FIFO fills | |
1027 | * past the watermark point. If the FIFO drains completely, a FIFO underrun | |
1028 | * will occur, and a display engine hang could result. | |
1029 | */ | |
1030 | static unsigned long intel_calculate_wm(unsigned long clock_in_khz, | |
1031 | const struct intel_watermark_params *wm, | |
1032 | int fifo_size, | |
1033 | int pixel_size, | |
1034 | unsigned long latency_ns) | |
1035 | { | |
1036 | long entries_required, wm_size; | |
1037 | ||
1038 | /* | |
1039 | * Note: we need to make sure we don't overflow for various clock & | |
1040 | * latency values. | |
1041 | * clocks go from a few thousand to several hundred thousand. | |
1042 | * latency is usually a few thousand | |
1043 | */ | |
1044 | entries_required = ((clock_in_khz / 1000) * pixel_size * latency_ns) / | |
1045 | 1000; | |
1046 | entries_required = DIV_ROUND_UP(entries_required, wm->cacheline_size); | |
1047 | ||
1048 | DRM_DEBUG_KMS("FIFO entries required for mode: %ld\n", entries_required); | |
1049 | ||
1050 | wm_size = fifo_size - (entries_required + wm->guard_size); | |
1051 | ||
1052 | DRM_DEBUG_KMS("FIFO watermark level: %ld\n", wm_size); | |
1053 | ||
1054 | /* Don't promote wm_size to unsigned... */ | |
1055 | if (wm_size > (long)wm->max_wm) | |
1056 | wm_size = wm->max_wm; | |
1057 | if (wm_size <= 0) | |
1058 | wm_size = wm->default_wm; | |
1059 | return wm_size; | |
1060 | } | |
1061 | ||
1062 | static struct drm_crtc *single_enabled_crtc(struct drm_device *dev) | |
1063 | { | |
1064 | struct drm_crtc *crtc, *enabled = NULL; | |
1065 | ||
1066 | list_for_each_entry(crtc, &dev->mode_config.crtc_list, head) { | |
3490ea5d | 1067 | if (intel_crtc_active(crtc)) { |
b445e3b0 ED |
1068 | if (enabled) |
1069 | return NULL; | |
1070 | enabled = crtc; | |
1071 | } | |
1072 | } | |
1073 | ||
1074 | return enabled; | |
1075 | } | |
1076 | ||
1fa61106 | 1077 | static void pineview_update_wm(struct drm_device *dev) |
b445e3b0 ED |
1078 | { |
1079 | struct drm_i915_private *dev_priv = dev->dev_private; | |
1080 | struct drm_crtc *crtc; | |
1081 | const struct cxsr_latency *latency; | |
1082 | u32 reg; | |
1083 | unsigned long wm; | |
1084 | ||
1085 | latency = intel_get_cxsr_latency(IS_PINEVIEW_G(dev), dev_priv->is_ddr3, | |
1086 | dev_priv->fsb_freq, dev_priv->mem_freq); | |
1087 | if (!latency) { | |
1088 | DRM_DEBUG_KMS("Unknown FSB/MEM found, disable CxSR\n"); | |
1089 | pineview_disable_cxsr(dev); | |
1090 | return; | |
1091 | } | |
1092 | ||
1093 | crtc = single_enabled_crtc(dev); | |
1094 | if (crtc) { | |
1095 | int clock = crtc->mode.clock; | |
1096 | int pixel_size = crtc->fb->bits_per_pixel / 8; | |
1097 | ||
1098 | /* Display SR */ | |
1099 | wm = intel_calculate_wm(clock, &pineview_display_wm, | |
1100 | pineview_display_wm.fifo_size, | |
1101 | pixel_size, latency->display_sr); | |
1102 | reg = I915_READ(DSPFW1); | |
1103 | reg &= ~DSPFW_SR_MASK; | |
1104 | reg |= wm << DSPFW_SR_SHIFT; | |
1105 | I915_WRITE(DSPFW1, reg); | |
1106 | DRM_DEBUG_KMS("DSPFW1 register is %x\n", reg); | |
1107 | ||
1108 | /* cursor SR */ | |
1109 | wm = intel_calculate_wm(clock, &pineview_cursor_wm, | |
1110 | pineview_display_wm.fifo_size, | |
1111 | pixel_size, latency->cursor_sr); | |
1112 | reg = I915_READ(DSPFW3); | |
1113 | reg &= ~DSPFW_CURSOR_SR_MASK; | |
1114 | reg |= (wm & 0x3f) << DSPFW_CURSOR_SR_SHIFT; | |
1115 | I915_WRITE(DSPFW3, reg); | |
1116 | ||
1117 | /* Display HPLL off SR */ | |
1118 | wm = intel_calculate_wm(clock, &pineview_display_hplloff_wm, | |
1119 | pineview_display_hplloff_wm.fifo_size, | |
1120 | pixel_size, latency->display_hpll_disable); | |
1121 | reg = I915_READ(DSPFW3); | |
1122 | reg &= ~DSPFW_HPLL_SR_MASK; | |
1123 | reg |= wm & DSPFW_HPLL_SR_MASK; | |
1124 | I915_WRITE(DSPFW3, reg); | |
1125 | ||
1126 | /* cursor HPLL off SR */ | |
1127 | wm = intel_calculate_wm(clock, &pineview_cursor_hplloff_wm, | |
1128 | pineview_display_hplloff_wm.fifo_size, | |
1129 | pixel_size, latency->cursor_hpll_disable); | |
1130 | reg = I915_READ(DSPFW3); | |
1131 | reg &= ~DSPFW_HPLL_CURSOR_MASK; | |
1132 | reg |= (wm & 0x3f) << DSPFW_HPLL_CURSOR_SHIFT; | |
1133 | I915_WRITE(DSPFW3, reg); | |
1134 | DRM_DEBUG_KMS("DSPFW3 register is %x\n", reg); | |
1135 | ||
1136 | /* activate cxsr */ | |
1137 | I915_WRITE(DSPFW3, | |
1138 | I915_READ(DSPFW3) | PINEVIEW_SELF_REFRESH_EN); | |
1139 | DRM_DEBUG_KMS("Self-refresh is enabled\n"); | |
1140 | } else { | |
1141 | pineview_disable_cxsr(dev); | |
1142 | DRM_DEBUG_KMS("Self-refresh is disabled\n"); | |
1143 | } | |
1144 | } | |
1145 | ||
1146 | static bool g4x_compute_wm0(struct drm_device *dev, | |
1147 | int plane, | |
1148 | const struct intel_watermark_params *display, | |
1149 | int display_latency_ns, | |
1150 | const struct intel_watermark_params *cursor, | |
1151 | int cursor_latency_ns, | |
1152 | int *plane_wm, | |
1153 | int *cursor_wm) | |
1154 | { | |
1155 | struct drm_crtc *crtc; | |
1156 | int htotal, hdisplay, clock, pixel_size; | |
1157 | int line_time_us, line_count; | |
1158 | int entries, tlb_miss; | |
1159 | ||
1160 | crtc = intel_get_crtc_for_plane(dev, plane); | |
3490ea5d | 1161 | if (!intel_crtc_active(crtc)) { |
b445e3b0 ED |
1162 | *cursor_wm = cursor->guard_size; |
1163 | *plane_wm = display->guard_size; | |
1164 | return false; | |
1165 | } | |
1166 | ||
1167 | htotal = crtc->mode.htotal; | |
1168 | hdisplay = crtc->mode.hdisplay; | |
1169 | clock = crtc->mode.clock; | |
1170 | pixel_size = crtc->fb->bits_per_pixel / 8; | |
1171 | ||
1172 | /* Use the small buffer method to calculate plane watermark */ | |
1173 | entries = ((clock * pixel_size / 1000) * display_latency_ns) / 1000; | |
1174 | tlb_miss = display->fifo_size*display->cacheline_size - hdisplay * 8; | |
1175 | if (tlb_miss > 0) | |
1176 | entries += tlb_miss; | |
1177 | entries = DIV_ROUND_UP(entries, display->cacheline_size); | |
1178 | *plane_wm = entries + display->guard_size; | |
1179 | if (*plane_wm > (int)display->max_wm) | |
1180 | *plane_wm = display->max_wm; | |
1181 | ||
1182 | /* Use the large buffer method to calculate cursor watermark */ | |
1183 | line_time_us = ((htotal * 1000) / clock); | |
1184 | line_count = (cursor_latency_ns / line_time_us + 1000) / 1000; | |
1185 | entries = line_count * 64 * pixel_size; | |
1186 | tlb_miss = cursor->fifo_size*cursor->cacheline_size - hdisplay * 8; | |
1187 | if (tlb_miss > 0) | |
1188 | entries += tlb_miss; | |
1189 | entries = DIV_ROUND_UP(entries, cursor->cacheline_size); | |
1190 | *cursor_wm = entries + cursor->guard_size; | |
1191 | if (*cursor_wm > (int)cursor->max_wm) | |
1192 | *cursor_wm = (int)cursor->max_wm; | |
1193 | ||
1194 | return true; | |
1195 | } | |
1196 | ||
1197 | /* | |
1198 | * Check the wm result. | |
1199 | * | |
1200 | * If any calculated watermark values is larger than the maximum value that | |
1201 | * can be programmed into the associated watermark register, that watermark | |
1202 | * must be disabled. | |
1203 | */ | |
1204 | static bool g4x_check_srwm(struct drm_device *dev, | |
1205 | int display_wm, int cursor_wm, | |
1206 | const struct intel_watermark_params *display, | |
1207 | const struct intel_watermark_params *cursor) | |
1208 | { | |
1209 | DRM_DEBUG_KMS("SR watermark: display plane %d, cursor %d\n", | |
1210 | display_wm, cursor_wm); | |
1211 | ||
1212 | if (display_wm > display->max_wm) { | |
1213 | DRM_DEBUG_KMS("display watermark is too large(%d/%ld), disabling\n", | |
1214 | display_wm, display->max_wm); | |
1215 | return false; | |
1216 | } | |
1217 | ||
1218 | if (cursor_wm > cursor->max_wm) { | |
1219 | DRM_DEBUG_KMS("cursor watermark is too large(%d/%ld), disabling\n", | |
1220 | cursor_wm, cursor->max_wm); | |
1221 | return false; | |
1222 | } | |
1223 | ||
1224 | if (!(display_wm || cursor_wm)) { | |
1225 | DRM_DEBUG_KMS("SR latency is 0, disabling\n"); | |
1226 | return false; | |
1227 | } | |
1228 | ||
1229 | return true; | |
1230 | } | |
1231 | ||
1232 | static bool g4x_compute_srwm(struct drm_device *dev, | |
1233 | int plane, | |
1234 | int latency_ns, | |
1235 | const struct intel_watermark_params *display, | |
1236 | const struct intel_watermark_params *cursor, | |
1237 | int *display_wm, int *cursor_wm) | |
1238 | { | |
1239 | struct drm_crtc *crtc; | |
1240 | int hdisplay, htotal, pixel_size, clock; | |
1241 | unsigned long line_time_us; | |
1242 | int line_count, line_size; | |
1243 | int small, large; | |
1244 | int entries; | |
1245 | ||
1246 | if (!latency_ns) { | |
1247 | *display_wm = *cursor_wm = 0; | |
1248 | return false; | |
1249 | } | |
1250 | ||
1251 | crtc = intel_get_crtc_for_plane(dev, plane); | |
1252 | hdisplay = crtc->mode.hdisplay; | |
1253 | htotal = crtc->mode.htotal; | |
1254 | clock = crtc->mode.clock; | |
1255 | pixel_size = crtc->fb->bits_per_pixel / 8; | |
1256 | ||
1257 | line_time_us = (htotal * 1000) / clock; | |
1258 | line_count = (latency_ns / line_time_us + 1000) / 1000; | |
1259 | line_size = hdisplay * pixel_size; | |
1260 | ||
1261 | /* Use the minimum of the small and large buffer method for primary */ | |
1262 | small = ((clock * pixel_size / 1000) * latency_ns) / 1000; | |
1263 | large = line_count * line_size; | |
1264 | ||
1265 | entries = DIV_ROUND_UP(min(small, large), display->cacheline_size); | |
1266 | *display_wm = entries + display->guard_size; | |
1267 | ||
1268 | /* calculate the self-refresh watermark for display cursor */ | |
1269 | entries = line_count * pixel_size * 64; | |
1270 | entries = DIV_ROUND_UP(entries, cursor->cacheline_size); | |
1271 | *cursor_wm = entries + cursor->guard_size; | |
1272 | ||
1273 | return g4x_check_srwm(dev, | |
1274 | *display_wm, *cursor_wm, | |
1275 | display, cursor); | |
1276 | } | |
1277 | ||
1278 | static bool vlv_compute_drain_latency(struct drm_device *dev, | |
1279 | int plane, | |
1280 | int *plane_prec_mult, | |
1281 | int *plane_dl, | |
1282 | int *cursor_prec_mult, | |
1283 | int *cursor_dl) | |
1284 | { | |
1285 | struct drm_crtc *crtc; | |
1286 | int clock, pixel_size; | |
1287 | int entries; | |
1288 | ||
1289 | crtc = intel_get_crtc_for_plane(dev, plane); | |
3490ea5d | 1290 | if (!intel_crtc_active(crtc)) |
b445e3b0 ED |
1291 | return false; |
1292 | ||
1293 | clock = crtc->mode.clock; /* VESA DOT Clock */ | |
1294 | pixel_size = crtc->fb->bits_per_pixel / 8; /* BPP */ | |
1295 | ||
1296 | entries = (clock / 1000) * pixel_size; | |
1297 | *plane_prec_mult = (entries > 256) ? | |
1298 | DRAIN_LATENCY_PRECISION_32 : DRAIN_LATENCY_PRECISION_16; | |
1299 | *plane_dl = (64 * (*plane_prec_mult) * 4) / ((clock / 1000) * | |
1300 | pixel_size); | |
1301 | ||
1302 | entries = (clock / 1000) * 4; /* BPP is always 4 for cursor */ | |
1303 | *cursor_prec_mult = (entries > 256) ? | |
1304 | DRAIN_LATENCY_PRECISION_32 : DRAIN_LATENCY_PRECISION_16; | |
1305 | *cursor_dl = (64 * (*cursor_prec_mult) * 4) / ((clock / 1000) * 4); | |
1306 | ||
1307 | return true; | |
1308 | } | |
1309 | ||
1310 | /* | |
1311 | * Update drain latency registers of memory arbiter | |
1312 | * | |
1313 | * Valleyview SoC has a new memory arbiter and needs drain latency registers | |
1314 | * to be programmed. Each plane has a drain latency multiplier and a drain | |
1315 | * latency value. | |
1316 | */ | |
1317 | ||
1318 | static void vlv_update_drain_latency(struct drm_device *dev) | |
1319 | { | |
1320 | struct drm_i915_private *dev_priv = dev->dev_private; | |
1321 | int planea_prec, planea_dl, planeb_prec, planeb_dl; | |
1322 | int cursora_prec, cursora_dl, cursorb_prec, cursorb_dl; | |
1323 | int plane_prec_mult, cursor_prec_mult; /* Precision multiplier is | |
1324 | either 16 or 32 */ | |
1325 | ||
1326 | /* For plane A, Cursor A */ | |
1327 | if (vlv_compute_drain_latency(dev, 0, &plane_prec_mult, &planea_dl, | |
1328 | &cursor_prec_mult, &cursora_dl)) { | |
1329 | cursora_prec = (cursor_prec_mult == DRAIN_LATENCY_PRECISION_32) ? | |
1330 | DDL_CURSORA_PRECISION_32 : DDL_CURSORA_PRECISION_16; | |
1331 | planea_prec = (plane_prec_mult == DRAIN_LATENCY_PRECISION_32) ? | |
1332 | DDL_PLANEA_PRECISION_32 : DDL_PLANEA_PRECISION_16; | |
1333 | ||
1334 | I915_WRITE(VLV_DDL1, cursora_prec | | |
1335 | (cursora_dl << DDL_CURSORA_SHIFT) | | |
1336 | planea_prec | planea_dl); | |
1337 | } | |
1338 | ||
1339 | /* For plane B, Cursor B */ | |
1340 | if (vlv_compute_drain_latency(dev, 1, &plane_prec_mult, &planeb_dl, | |
1341 | &cursor_prec_mult, &cursorb_dl)) { | |
1342 | cursorb_prec = (cursor_prec_mult == DRAIN_LATENCY_PRECISION_32) ? | |
1343 | DDL_CURSORB_PRECISION_32 : DDL_CURSORB_PRECISION_16; | |
1344 | planeb_prec = (plane_prec_mult == DRAIN_LATENCY_PRECISION_32) ? | |
1345 | DDL_PLANEB_PRECISION_32 : DDL_PLANEB_PRECISION_16; | |
1346 | ||
1347 | I915_WRITE(VLV_DDL2, cursorb_prec | | |
1348 | (cursorb_dl << DDL_CURSORB_SHIFT) | | |
1349 | planeb_prec | planeb_dl); | |
1350 | } | |
1351 | } | |
1352 | ||
1353 | #define single_plane_enabled(mask) is_power_of_2(mask) | |
1354 | ||
1fa61106 | 1355 | static void valleyview_update_wm(struct drm_device *dev) |
b445e3b0 ED |
1356 | { |
1357 | static const int sr_latency_ns = 12000; | |
1358 | struct drm_i915_private *dev_priv = dev->dev_private; | |
1359 | int planea_wm, planeb_wm, cursora_wm, cursorb_wm; | |
1360 | int plane_sr, cursor_sr; | |
af6c4575 | 1361 | int ignore_plane_sr, ignore_cursor_sr; |
b445e3b0 ED |
1362 | unsigned int enabled = 0; |
1363 | ||
1364 | vlv_update_drain_latency(dev); | |
1365 | ||
51cea1f4 | 1366 | if (g4x_compute_wm0(dev, PIPE_A, |
b445e3b0 ED |
1367 | &valleyview_wm_info, latency_ns, |
1368 | &valleyview_cursor_wm_info, latency_ns, | |
1369 | &planea_wm, &cursora_wm)) | |
51cea1f4 | 1370 | enabled |= 1 << PIPE_A; |
b445e3b0 | 1371 | |
51cea1f4 | 1372 | if (g4x_compute_wm0(dev, PIPE_B, |
b445e3b0 ED |
1373 | &valleyview_wm_info, latency_ns, |
1374 | &valleyview_cursor_wm_info, latency_ns, | |
1375 | &planeb_wm, &cursorb_wm)) | |
51cea1f4 | 1376 | enabled |= 1 << PIPE_B; |
b445e3b0 | 1377 | |
b445e3b0 ED |
1378 | if (single_plane_enabled(enabled) && |
1379 | g4x_compute_srwm(dev, ffs(enabled) - 1, | |
1380 | sr_latency_ns, | |
1381 | &valleyview_wm_info, | |
1382 | &valleyview_cursor_wm_info, | |
af6c4575 CW |
1383 | &plane_sr, &ignore_cursor_sr) && |
1384 | g4x_compute_srwm(dev, ffs(enabled) - 1, | |
1385 | 2*sr_latency_ns, | |
1386 | &valleyview_wm_info, | |
1387 | &valleyview_cursor_wm_info, | |
52bd02d8 | 1388 | &ignore_plane_sr, &cursor_sr)) { |
b445e3b0 | 1389 | I915_WRITE(FW_BLC_SELF_VLV, FW_CSPWRDWNEN); |
52bd02d8 | 1390 | } else { |
b445e3b0 ED |
1391 | I915_WRITE(FW_BLC_SELF_VLV, |
1392 | I915_READ(FW_BLC_SELF_VLV) & ~FW_CSPWRDWNEN); | |
52bd02d8 CW |
1393 | plane_sr = cursor_sr = 0; |
1394 | } | |
b445e3b0 ED |
1395 | |
1396 | DRM_DEBUG_KMS("Setting FIFO watermarks - A: plane=%d, cursor=%d, B: plane=%d, cursor=%d, SR: plane=%d, cursor=%d\n", | |
1397 | planea_wm, cursora_wm, | |
1398 | planeb_wm, cursorb_wm, | |
1399 | plane_sr, cursor_sr); | |
1400 | ||
1401 | I915_WRITE(DSPFW1, | |
1402 | (plane_sr << DSPFW_SR_SHIFT) | | |
1403 | (cursorb_wm << DSPFW_CURSORB_SHIFT) | | |
1404 | (planeb_wm << DSPFW_PLANEB_SHIFT) | | |
1405 | planea_wm); | |
1406 | I915_WRITE(DSPFW2, | |
8c919b28 | 1407 | (I915_READ(DSPFW2) & ~DSPFW_CURSORA_MASK) | |
b445e3b0 ED |
1408 | (cursora_wm << DSPFW_CURSORA_SHIFT)); |
1409 | I915_WRITE(DSPFW3, | |
8c919b28 CW |
1410 | (I915_READ(DSPFW3) & ~DSPFW_CURSOR_SR_MASK) | |
1411 | (cursor_sr << DSPFW_CURSOR_SR_SHIFT)); | |
b445e3b0 ED |
1412 | } |
1413 | ||
1fa61106 | 1414 | static void g4x_update_wm(struct drm_device *dev) |
b445e3b0 ED |
1415 | { |
1416 | static const int sr_latency_ns = 12000; | |
1417 | struct drm_i915_private *dev_priv = dev->dev_private; | |
1418 | int planea_wm, planeb_wm, cursora_wm, cursorb_wm; | |
1419 | int plane_sr, cursor_sr; | |
1420 | unsigned int enabled = 0; | |
1421 | ||
51cea1f4 | 1422 | if (g4x_compute_wm0(dev, PIPE_A, |
b445e3b0 ED |
1423 | &g4x_wm_info, latency_ns, |
1424 | &g4x_cursor_wm_info, latency_ns, | |
1425 | &planea_wm, &cursora_wm)) | |
51cea1f4 | 1426 | enabled |= 1 << PIPE_A; |
b445e3b0 | 1427 | |
51cea1f4 | 1428 | if (g4x_compute_wm0(dev, PIPE_B, |
b445e3b0 ED |
1429 | &g4x_wm_info, latency_ns, |
1430 | &g4x_cursor_wm_info, latency_ns, | |
1431 | &planeb_wm, &cursorb_wm)) | |
51cea1f4 | 1432 | enabled |= 1 << PIPE_B; |
b445e3b0 | 1433 | |
b445e3b0 ED |
1434 | if (single_plane_enabled(enabled) && |
1435 | g4x_compute_srwm(dev, ffs(enabled) - 1, | |
1436 | sr_latency_ns, | |
1437 | &g4x_wm_info, | |
1438 | &g4x_cursor_wm_info, | |
52bd02d8 | 1439 | &plane_sr, &cursor_sr)) { |
b445e3b0 | 1440 | I915_WRITE(FW_BLC_SELF, FW_BLC_SELF_EN); |
52bd02d8 | 1441 | } else { |
b445e3b0 ED |
1442 | I915_WRITE(FW_BLC_SELF, |
1443 | I915_READ(FW_BLC_SELF) & ~FW_BLC_SELF_EN); | |
52bd02d8 CW |
1444 | plane_sr = cursor_sr = 0; |
1445 | } | |
b445e3b0 ED |
1446 | |
1447 | DRM_DEBUG_KMS("Setting FIFO watermarks - A: plane=%d, cursor=%d, B: plane=%d, cursor=%d, SR: plane=%d, cursor=%d\n", | |
1448 | planea_wm, cursora_wm, | |
1449 | planeb_wm, cursorb_wm, | |
1450 | plane_sr, cursor_sr); | |
1451 | ||
1452 | I915_WRITE(DSPFW1, | |
1453 | (plane_sr << DSPFW_SR_SHIFT) | | |
1454 | (cursorb_wm << DSPFW_CURSORB_SHIFT) | | |
1455 | (planeb_wm << DSPFW_PLANEB_SHIFT) | | |
1456 | planea_wm); | |
1457 | I915_WRITE(DSPFW2, | |
8c919b28 | 1458 | (I915_READ(DSPFW2) & ~DSPFW_CURSORA_MASK) | |
b445e3b0 ED |
1459 | (cursora_wm << DSPFW_CURSORA_SHIFT)); |
1460 | /* HPLL off in SR has some issues on G4x... disable it */ | |
1461 | I915_WRITE(DSPFW3, | |
8c919b28 | 1462 | (I915_READ(DSPFW3) & ~(DSPFW_HPLL_SR_EN | DSPFW_CURSOR_SR_MASK)) | |
b445e3b0 ED |
1463 | (cursor_sr << DSPFW_CURSOR_SR_SHIFT)); |
1464 | } | |
1465 | ||
1fa61106 | 1466 | static void i965_update_wm(struct drm_device *dev) |
b445e3b0 ED |
1467 | { |
1468 | struct drm_i915_private *dev_priv = dev->dev_private; | |
1469 | struct drm_crtc *crtc; | |
1470 | int srwm = 1; | |
1471 | int cursor_sr = 16; | |
1472 | ||
1473 | /* Calc sr entries for one plane configs */ | |
1474 | crtc = single_enabled_crtc(dev); | |
1475 | if (crtc) { | |
1476 | /* self-refresh has much higher latency */ | |
1477 | static const int sr_latency_ns = 12000; | |
1478 | int clock = crtc->mode.clock; | |
1479 | int htotal = crtc->mode.htotal; | |
1480 | int hdisplay = crtc->mode.hdisplay; | |
1481 | int pixel_size = crtc->fb->bits_per_pixel / 8; | |
1482 | unsigned long line_time_us; | |
1483 | int entries; | |
1484 | ||
1485 | line_time_us = ((htotal * 1000) / clock); | |
1486 | ||
1487 | /* Use ns/us then divide to preserve precision */ | |
1488 | entries = (((sr_latency_ns / line_time_us) + 1000) / 1000) * | |
1489 | pixel_size * hdisplay; | |
1490 | entries = DIV_ROUND_UP(entries, I915_FIFO_LINE_SIZE); | |
1491 | srwm = I965_FIFO_SIZE - entries; | |
1492 | if (srwm < 0) | |
1493 | srwm = 1; | |
1494 | srwm &= 0x1ff; | |
1495 | DRM_DEBUG_KMS("self-refresh entries: %d, wm: %d\n", | |
1496 | entries, srwm); | |
1497 | ||
1498 | entries = (((sr_latency_ns / line_time_us) + 1000) / 1000) * | |
1499 | pixel_size * 64; | |
1500 | entries = DIV_ROUND_UP(entries, | |
1501 | i965_cursor_wm_info.cacheline_size); | |
1502 | cursor_sr = i965_cursor_wm_info.fifo_size - | |
1503 | (entries + i965_cursor_wm_info.guard_size); | |
1504 | ||
1505 | if (cursor_sr > i965_cursor_wm_info.max_wm) | |
1506 | cursor_sr = i965_cursor_wm_info.max_wm; | |
1507 | ||
1508 | DRM_DEBUG_KMS("self-refresh watermark: display plane %d " | |
1509 | "cursor %d\n", srwm, cursor_sr); | |
1510 | ||
1511 | if (IS_CRESTLINE(dev)) | |
1512 | I915_WRITE(FW_BLC_SELF, FW_BLC_SELF_EN); | |
1513 | } else { | |
1514 | /* Turn off self refresh if both pipes are enabled */ | |
1515 | if (IS_CRESTLINE(dev)) | |
1516 | I915_WRITE(FW_BLC_SELF, I915_READ(FW_BLC_SELF) | |
1517 | & ~FW_BLC_SELF_EN); | |
1518 | } | |
1519 | ||
1520 | DRM_DEBUG_KMS("Setting FIFO watermarks - A: 8, B: 8, C: 8, SR %d\n", | |
1521 | srwm); | |
1522 | ||
1523 | /* 965 has limitations... */ | |
1524 | I915_WRITE(DSPFW1, (srwm << DSPFW_SR_SHIFT) | | |
1525 | (8 << 16) | (8 << 8) | (8 << 0)); | |
1526 | I915_WRITE(DSPFW2, (8 << 8) | (8 << 0)); | |
1527 | /* update cursor SR watermark */ | |
1528 | I915_WRITE(DSPFW3, (cursor_sr << DSPFW_CURSOR_SR_SHIFT)); | |
1529 | } | |
1530 | ||
1fa61106 | 1531 | static void i9xx_update_wm(struct drm_device *dev) |
b445e3b0 ED |
1532 | { |
1533 | struct drm_i915_private *dev_priv = dev->dev_private; | |
1534 | const struct intel_watermark_params *wm_info; | |
1535 | uint32_t fwater_lo; | |
1536 | uint32_t fwater_hi; | |
1537 | int cwm, srwm = 1; | |
1538 | int fifo_size; | |
1539 | int planea_wm, planeb_wm; | |
1540 | struct drm_crtc *crtc, *enabled = NULL; | |
1541 | ||
1542 | if (IS_I945GM(dev)) | |
1543 | wm_info = &i945_wm_info; | |
1544 | else if (!IS_GEN2(dev)) | |
1545 | wm_info = &i915_wm_info; | |
1546 | else | |
1547 | wm_info = &i855_wm_info; | |
1548 | ||
1549 | fifo_size = dev_priv->display.get_fifo_size(dev, 0); | |
1550 | crtc = intel_get_crtc_for_plane(dev, 0); | |
3490ea5d | 1551 | if (intel_crtc_active(crtc)) { |
b9e0bda3 CW |
1552 | int cpp = crtc->fb->bits_per_pixel / 8; |
1553 | if (IS_GEN2(dev)) | |
1554 | cpp = 4; | |
1555 | ||
b445e3b0 | 1556 | planea_wm = intel_calculate_wm(crtc->mode.clock, |
b9e0bda3 | 1557 | wm_info, fifo_size, cpp, |
b445e3b0 ED |
1558 | latency_ns); |
1559 | enabled = crtc; | |
1560 | } else | |
1561 | planea_wm = fifo_size - wm_info->guard_size; | |
1562 | ||
1563 | fifo_size = dev_priv->display.get_fifo_size(dev, 1); | |
1564 | crtc = intel_get_crtc_for_plane(dev, 1); | |
3490ea5d | 1565 | if (intel_crtc_active(crtc)) { |
b9e0bda3 CW |
1566 | int cpp = crtc->fb->bits_per_pixel / 8; |
1567 | if (IS_GEN2(dev)) | |
1568 | cpp = 4; | |
1569 | ||
b445e3b0 | 1570 | planeb_wm = intel_calculate_wm(crtc->mode.clock, |
b9e0bda3 | 1571 | wm_info, fifo_size, cpp, |
b445e3b0 ED |
1572 | latency_ns); |
1573 | if (enabled == NULL) | |
1574 | enabled = crtc; | |
1575 | else | |
1576 | enabled = NULL; | |
1577 | } else | |
1578 | planeb_wm = fifo_size - wm_info->guard_size; | |
1579 | ||
1580 | DRM_DEBUG_KMS("FIFO watermarks - A: %d, B: %d\n", planea_wm, planeb_wm); | |
1581 | ||
1582 | /* | |
1583 | * Overlay gets an aggressive default since video jitter is bad. | |
1584 | */ | |
1585 | cwm = 2; | |
1586 | ||
1587 | /* Play safe and disable self-refresh before adjusting watermarks. */ | |
1588 | if (IS_I945G(dev) || IS_I945GM(dev)) | |
1589 | I915_WRITE(FW_BLC_SELF, FW_BLC_SELF_EN_MASK | 0); | |
1590 | else if (IS_I915GM(dev)) | |
1591 | I915_WRITE(INSTPM, I915_READ(INSTPM) & ~INSTPM_SELF_EN); | |
1592 | ||
1593 | /* Calc sr entries for one plane configs */ | |
1594 | if (HAS_FW_BLC(dev) && enabled) { | |
1595 | /* self-refresh has much higher latency */ | |
1596 | static const int sr_latency_ns = 6000; | |
1597 | int clock = enabled->mode.clock; | |
1598 | int htotal = enabled->mode.htotal; | |
1599 | int hdisplay = enabled->mode.hdisplay; | |
1600 | int pixel_size = enabled->fb->bits_per_pixel / 8; | |
1601 | unsigned long line_time_us; | |
1602 | int entries; | |
1603 | ||
1604 | line_time_us = (htotal * 1000) / clock; | |
1605 | ||
1606 | /* Use ns/us then divide to preserve precision */ | |
1607 | entries = (((sr_latency_ns / line_time_us) + 1000) / 1000) * | |
1608 | pixel_size * hdisplay; | |
1609 | entries = DIV_ROUND_UP(entries, wm_info->cacheline_size); | |
1610 | DRM_DEBUG_KMS("self-refresh entries: %d\n", entries); | |
1611 | srwm = wm_info->fifo_size - entries; | |
1612 | if (srwm < 0) | |
1613 | srwm = 1; | |
1614 | ||
1615 | if (IS_I945G(dev) || IS_I945GM(dev)) | |
1616 | I915_WRITE(FW_BLC_SELF, | |
1617 | FW_BLC_SELF_FIFO_MASK | (srwm & 0xff)); | |
1618 | else if (IS_I915GM(dev)) | |
1619 | I915_WRITE(FW_BLC_SELF, srwm & 0x3f); | |
1620 | } | |
1621 | ||
1622 | DRM_DEBUG_KMS("Setting FIFO watermarks - A: %d, B: %d, C: %d, SR %d\n", | |
1623 | planea_wm, planeb_wm, cwm, srwm); | |
1624 | ||
1625 | fwater_lo = ((planeb_wm & 0x3f) << 16) | (planea_wm & 0x3f); | |
1626 | fwater_hi = (cwm & 0x1f); | |
1627 | ||
1628 | /* Set request length to 8 cachelines per fetch */ | |
1629 | fwater_lo = fwater_lo | (1 << 24) | (1 << 8); | |
1630 | fwater_hi = fwater_hi | (1 << 8); | |
1631 | ||
1632 | I915_WRITE(FW_BLC, fwater_lo); | |
1633 | I915_WRITE(FW_BLC2, fwater_hi); | |
1634 | ||
1635 | if (HAS_FW_BLC(dev)) { | |
1636 | if (enabled) { | |
1637 | if (IS_I945G(dev) || IS_I945GM(dev)) | |
1638 | I915_WRITE(FW_BLC_SELF, | |
1639 | FW_BLC_SELF_EN_MASK | FW_BLC_SELF_EN); | |
1640 | else if (IS_I915GM(dev)) | |
1641 | I915_WRITE(INSTPM, I915_READ(INSTPM) | INSTPM_SELF_EN); | |
1642 | DRM_DEBUG_KMS("memory self refresh enabled\n"); | |
1643 | } else | |
1644 | DRM_DEBUG_KMS("memory self refresh disabled\n"); | |
1645 | } | |
1646 | } | |
1647 | ||
1fa61106 | 1648 | static void i830_update_wm(struct drm_device *dev) |
b445e3b0 ED |
1649 | { |
1650 | struct drm_i915_private *dev_priv = dev->dev_private; | |
1651 | struct drm_crtc *crtc; | |
1652 | uint32_t fwater_lo; | |
1653 | int planea_wm; | |
1654 | ||
1655 | crtc = single_enabled_crtc(dev); | |
1656 | if (crtc == NULL) | |
1657 | return; | |
1658 | ||
1659 | planea_wm = intel_calculate_wm(crtc->mode.clock, &i830_wm_info, | |
1660 | dev_priv->display.get_fifo_size(dev, 0), | |
b9e0bda3 | 1661 | 4, latency_ns); |
b445e3b0 ED |
1662 | fwater_lo = I915_READ(FW_BLC) & ~0xfff; |
1663 | fwater_lo |= (3<<8) | planea_wm; | |
1664 | ||
1665 | DRM_DEBUG_KMS("Setting FIFO watermarks - A: %d\n", planea_wm); | |
1666 | ||
1667 | I915_WRITE(FW_BLC, fwater_lo); | |
1668 | } | |
1669 | ||
1670 | #define ILK_LP0_PLANE_LATENCY 700 | |
1671 | #define ILK_LP0_CURSOR_LATENCY 1300 | |
1672 | ||
1673 | /* | |
1674 | * Check the wm result. | |
1675 | * | |
1676 | * If any calculated watermark values is larger than the maximum value that | |
1677 | * can be programmed into the associated watermark register, that watermark | |
1678 | * must be disabled. | |
1679 | */ | |
1680 | static bool ironlake_check_srwm(struct drm_device *dev, int level, | |
1681 | int fbc_wm, int display_wm, int cursor_wm, | |
1682 | const struct intel_watermark_params *display, | |
1683 | const struct intel_watermark_params *cursor) | |
1684 | { | |
1685 | struct drm_i915_private *dev_priv = dev->dev_private; | |
1686 | ||
1687 | DRM_DEBUG_KMS("watermark %d: display plane %d, fbc lines %d," | |
1688 | " cursor %d\n", level, display_wm, fbc_wm, cursor_wm); | |
1689 | ||
1690 | if (fbc_wm > SNB_FBC_MAX_SRWM) { | |
1691 | DRM_DEBUG_KMS("fbc watermark(%d) is too large(%d), disabling wm%d+\n", | |
1692 | fbc_wm, SNB_FBC_MAX_SRWM, level); | |
1693 | ||
1694 | /* fbc has it's own way to disable FBC WM */ | |
1695 | I915_WRITE(DISP_ARB_CTL, | |
1696 | I915_READ(DISP_ARB_CTL) | DISP_FBC_WM_DIS); | |
1697 | return false; | |
615aaa5f VS |
1698 | } else if (INTEL_INFO(dev)->gen >= 6) { |
1699 | /* enable FBC WM (except on ILK, where it must remain off) */ | |
1700 | I915_WRITE(DISP_ARB_CTL, | |
1701 | I915_READ(DISP_ARB_CTL) & ~DISP_FBC_WM_DIS); | |
b445e3b0 ED |
1702 | } |
1703 | ||
1704 | if (display_wm > display->max_wm) { | |
1705 | DRM_DEBUG_KMS("display watermark(%d) is too large(%d), disabling wm%d+\n", | |
1706 | display_wm, SNB_DISPLAY_MAX_SRWM, level); | |
1707 | return false; | |
1708 | } | |
1709 | ||
1710 | if (cursor_wm > cursor->max_wm) { | |
1711 | DRM_DEBUG_KMS("cursor watermark(%d) is too large(%d), disabling wm%d+\n", | |
1712 | cursor_wm, SNB_CURSOR_MAX_SRWM, level); | |
1713 | return false; | |
1714 | } | |
1715 | ||
1716 | if (!(fbc_wm || display_wm || cursor_wm)) { | |
1717 | DRM_DEBUG_KMS("latency %d is 0, disabling wm%d+\n", level, level); | |
1718 | return false; | |
1719 | } | |
1720 | ||
1721 | return true; | |
1722 | } | |
1723 | ||
1724 | /* | |
1725 | * Compute watermark values of WM[1-3], | |
1726 | */ | |
1727 | static bool ironlake_compute_srwm(struct drm_device *dev, int level, int plane, | |
1728 | int latency_ns, | |
1729 | const struct intel_watermark_params *display, | |
1730 | const struct intel_watermark_params *cursor, | |
1731 | int *fbc_wm, int *display_wm, int *cursor_wm) | |
1732 | { | |
1733 | struct drm_crtc *crtc; | |
1734 | unsigned long line_time_us; | |
1735 | int hdisplay, htotal, pixel_size, clock; | |
1736 | int line_count, line_size; | |
1737 | int small, large; | |
1738 | int entries; | |
1739 | ||
1740 | if (!latency_ns) { | |
1741 | *fbc_wm = *display_wm = *cursor_wm = 0; | |
1742 | return false; | |
1743 | } | |
1744 | ||
1745 | crtc = intel_get_crtc_for_plane(dev, plane); | |
1746 | hdisplay = crtc->mode.hdisplay; | |
1747 | htotal = crtc->mode.htotal; | |
1748 | clock = crtc->mode.clock; | |
1749 | pixel_size = crtc->fb->bits_per_pixel / 8; | |
1750 | ||
1751 | line_time_us = (htotal * 1000) / clock; | |
1752 | line_count = (latency_ns / line_time_us + 1000) / 1000; | |
1753 | line_size = hdisplay * pixel_size; | |
1754 | ||
1755 | /* Use the minimum of the small and large buffer method for primary */ | |
1756 | small = ((clock * pixel_size / 1000) * latency_ns) / 1000; | |
1757 | large = line_count * line_size; | |
1758 | ||
1759 | entries = DIV_ROUND_UP(min(small, large), display->cacheline_size); | |
1760 | *display_wm = entries + display->guard_size; | |
1761 | ||
1762 | /* | |
1763 | * Spec says: | |
1764 | * FBC WM = ((Final Primary WM * 64) / number of bytes per line) + 2 | |
1765 | */ | |
1766 | *fbc_wm = DIV_ROUND_UP(*display_wm * 64, line_size) + 2; | |
1767 | ||
1768 | /* calculate the self-refresh watermark for display cursor */ | |
1769 | entries = line_count * pixel_size * 64; | |
1770 | entries = DIV_ROUND_UP(entries, cursor->cacheline_size); | |
1771 | *cursor_wm = entries + cursor->guard_size; | |
1772 | ||
1773 | return ironlake_check_srwm(dev, level, | |
1774 | *fbc_wm, *display_wm, *cursor_wm, | |
1775 | display, cursor); | |
1776 | } | |
1777 | ||
1fa61106 | 1778 | static void ironlake_update_wm(struct drm_device *dev) |
b445e3b0 ED |
1779 | { |
1780 | struct drm_i915_private *dev_priv = dev->dev_private; | |
1781 | int fbc_wm, plane_wm, cursor_wm; | |
1782 | unsigned int enabled; | |
1783 | ||
1784 | enabled = 0; | |
51cea1f4 | 1785 | if (g4x_compute_wm0(dev, PIPE_A, |
b445e3b0 ED |
1786 | &ironlake_display_wm_info, |
1787 | ILK_LP0_PLANE_LATENCY, | |
1788 | &ironlake_cursor_wm_info, | |
1789 | ILK_LP0_CURSOR_LATENCY, | |
1790 | &plane_wm, &cursor_wm)) { | |
1791 | I915_WRITE(WM0_PIPEA_ILK, | |
1792 | (plane_wm << WM0_PIPE_PLANE_SHIFT) | cursor_wm); | |
1793 | DRM_DEBUG_KMS("FIFO watermarks For pipe A -" | |
1794 | " plane %d, " "cursor: %d\n", | |
1795 | plane_wm, cursor_wm); | |
51cea1f4 | 1796 | enabled |= 1 << PIPE_A; |
b445e3b0 ED |
1797 | } |
1798 | ||
51cea1f4 | 1799 | if (g4x_compute_wm0(dev, PIPE_B, |
b445e3b0 ED |
1800 | &ironlake_display_wm_info, |
1801 | ILK_LP0_PLANE_LATENCY, | |
1802 | &ironlake_cursor_wm_info, | |
1803 | ILK_LP0_CURSOR_LATENCY, | |
1804 | &plane_wm, &cursor_wm)) { | |
1805 | I915_WRITE(WM0_PIPEB_ILK, | |
1806 | (plane_wm << WM0_PIPE_PLANE_SHIFT) | cursor_wm); | |
1807 | DRM_DEBUG_KMS("FIFO watermarks For pipe B -" | |
1808 | " plane %d, cursor: %d\n", | |
1809 | plane_wm, cursor_wm); | |
51cea1f4 | 1810 | enabled |= 1 << PIPE_B; |
b445e3b0 ED |
1811 | } |
1812 | ||
1813 | /* | |
1814 | * Calculate and update the self-refresh watermark only when one | |
1815 | * display plane is used. | |
1816 | */ | |
1817 | I915_WRITE(WM3_LP_ILK, 0); | |
1818 | I915_WRITE(WM2_LP_ILK, 0); | |
1819 | I915_WRITE(WM1_LP_ILK, 0); | |
1820 | ||
1821 | if (!single_plane_enabled(enabled)) | |
1822 | return; | |
1823 | enabled = ffs(enabled) - 1; | |
1824 | ||
1825 | /* WM1 */ | |
1826 | if (!ironlake_compute_srwm(dev, 1, enabled, | |
1827 | ILK_READ_WM1_LATENCY() * 500, | |
1828 | &ironlake_display_srwm_info, | |
1829 | &ironlake_cursor_srwm_info, | |
1830 | &fbc_wm, &plane_wm, &cursor_wm)) | |
1831 | return; | |
1832 | ||
1833 | I915_WRITE(WM1_LP_ILK, | |
1834 | WM1_LP_SR_EN | | |
1835 | (ILK_READ_WM1_LATENCY() << WM1_LP_LATENCY_SHIFT) | | |
1836 | (fbc_wm << WM1_LP_FBC_SHIFT) | | |
1837 | (plane_wm << WM1_LP_SR_SHIFT) | | |
1838 | cursor_wm); | |
1839 | ||
1840 | /* WM2 */ | |
1841 | if (!ironlake_compute_srwm(dev, 2, enabled, | |
1842 | ILK_READ_WM2_LATENCY() * 500, | |
1843 | &ironlake_display_srwm_info, | |
1844 | &ironlake_cursor_srwm_info, | |
1845 | &fbc_wm, &plane_wm, &cursor_wm)) | |
1846 | return; | |
1847 | ||
1848 | I915_WRITE(WM2_LP_ILK, | |
1849 | WM2_LP_EN | | |
1850 | (ILK_READ_WM2_LATENCY() << WM1_LP_LATENCY_SHIFT) | | |
1851 | (fbc_wm << WM1_LP_FBC_SHIFT) | | |
1852 | (plane_wm << WM1_LP_SR_SHIFT) | | |
1853 | cursor_wm); | |
1854 | ||
1855 | /* | |
1856 | * WM3 is unsupported on ILK, probably because we don't have latency | |
1857 | * data for that power state | |
1858 | */ | |
1859 | } | |
1860 | ||
1fa61106 | 1861 | static void sandybridge_update_wm(struct drm_device *dev) |
b445e3b0 ED |
1862 | { |
1863 | struct drm_i915_private *dev_priv = dev->dev_private; | |
1864 | int latency = SNB_READ_WM0_LATENCY() * 100; /* In unit 0.1us */ | |
1865 | u32 val; | |
1866 | int fbc_wm, plane_wm, cursor_wm; | |
1867 | unsigned int enabled; | |
1868 | ||
1869 | enabled = 0; | |
51cea1f4 | 1870 | if (g4x_compute_wm0(dev, PIPE_A, |
b445e3b0 ED |
1871 | &sandybridge_display_wm_info, latency, |
1872 | &sandybridge_cursor_wm_info, latency, | |
1873 | &plane_wm, &cursor_wm)) { | |
1874 | val = I915_READ(WM0_PIPEA_ILK); | |
1875 | val &= ~(WM0_PIPE_PLANE_MASK | WM0_PIPE_CURSOR_MASK); | |
1876 | I915_WRITE(WM0_PIPEA_ILK, val | | |
1877 | ((plane_wm << WM0_PIPE_PLANE_SHIFT) | cursor_wm)); | |
1878 | DRM_DEBUG_KMS("FIFO watermarks For pipe A -" | |
1879 | " plane %d, " "cursor: %d\n", | |
1880 | plane_wm, cursor_wm); | |
51cea1f4 | 1881 | enabled |= 1 << PIPE_A; |
b445e3b0 ED |
1882 | } |
1883 | ||
51cea1f4 | 1884 | if (g4x_compute_wm0(dev, PIPE_B, |
b445e3b0 ED |
1885 | &sandybridge_display_wm_info, latency, |
1886 | &sandybridge_cursor_wm_info, latency, | |
1887 | &plane_wm, &cursor_wm)) { | |
1888 | val = I915_READ(WM0_PIPEB_ILK); | |
1889 | val &= ~(WM0_PIPE_PLANE_MASK | WM0_PIPE_CURSOR_MASK); | |
1890 | I915_WRITE(WM0_PIPEB_ILK, val | | |
1891 | ((plane_wm << WM0_PIPE_PLANE_SHIFT) | cursor_wm)); | |
1892 | DRM_DEBUG_KMS("FIFO watermarks For pipe B -" | |
1893 | " plane %d, cursor: %d\n", | |
1894 | plane_wm, cursor_wm); | |
51cea1f4 | 1895 | enabled |= 1 << PIPE_B; |
b445e3b0 ED |
1896 | } |
1897 | ||
c43d0188 CW |
1898 | /* |
1899 | * Calculate and update the self-refresh watermark only when one | |
1900 | * display plane is used. | |
1901 | * | |
1902 | * SNB support 3 levels of watermark. | |
1903 | * | |
1904 | * WM1/WM2/WM2 watermarks have to be enabled in the ascending order, | |
1905 | * and disabled in the descending order | |
1906 | * | |
1907 | */ | |
1908 | I915_WRITE(WM3_LP_ILK, 0); | |
1909 | I915_WRITE(WM2_LP_ILK, 0); | |
1910 | I915_WRITE(WM1_LP_ILK, 0); | |
1911 | ||
1912 | if (!single_plane_enabled(enabled) || | |
1913 | dev_priv->sprite_scaling_enabled) | |
1914 | return; | |
1915 | enabled = ffs(enabled) - 1; | |
1916 | ||
1917 | /* WM1 */ | |
1918 | if (!ironlake_compute_srwm(dev, 1, enabled, | |
1919 | SNB_READ_WM1_LATENCY() * 500, | |
1920 | &sandybridge_display_srwm_info, | |
1921 | &sandybridge_cursor_srwm_info, | |
1922 | &fbc_wm, &plane_wm, &cursor_wm)) | |
1923 | return; | |
1924 | ||
1925 | I915_WRITE(WM1_LP_ILK, | |
1926 | WM1_LP_SR_EN | | |
1927 | (SNB_READ_WM1_LATENCY() << WM1_LP_LATENCY_SHIFT) | | |
1928 | (fbc_wm << WM1_LP_FBC_SHIFT) | | |
1929 | (plane_wm << WM1_LP_SR_SHIFT) | | |
1930 | cursor_wm); | |
1931 | ||
1932 | /* WM2 */ | |
1933 | if (!ironlake_compute_srwm(dev, 2, enabled, | |
1934 | SNB_READ_WM2_LATENCY() * 500, | |
1935 | &sandybridge_display_srwm_info, | |
1936 | &sandybridge_cursor_srwm_info, | |
1937 | &fbc_wm, &plane_wm, &cursor_wm)) | |
1938 | return; | |
1939 | ||
1940 | I915_WRITE(WM2_LP_ILK, | |
1941 | WM2_LP_EN | | |
1942 | (SNB_READ_WM2_LATENCY() << WM1_LP_LATENCY_SHIFT) | | |
1943 | (fbc_wm << WM1_LP_FBC_SHIFT) | | |
1944 | (plane_wm << WM1_LP_SR_SHIFT) | | |
1945 | cursor_wm); | |
1946 | ||
1947 | /* WM3 */ | |
1948 | if (!ironlake_compute_srwm(dev, 3, enabled, | |
1949 | SNB_READ_WM3_LATENCY() * 500, | |
1950 | &sandybridge_display_srwm_info, | |
1951 | &sandybridge_cursor_srwm_info, | |
1952 | &fbc_wm, &plane_wm, &cursor_wm)) | |
1953 | return; | |
1954 | ||
1955 | I915_WRITE(WM3_LP_ILK, | |
1956 | WM3_LP_EN | | |
1957 | (SNB_READ_WM3_LATENCY() << WM1_LP_LATENCY_SHIFT) | | |
1958 | (fbc_wm << WM1_LP_FBC_SHIFT) | | |
1959 | (plane_wm << WM1_LP_SR_SHIFT) | | |
1960 | cursor_wm); | |
1961 | } | |
1962 | ||
1963 | static void ivybridge_update_wm(struct drm_device *dev) | |
1964 | { | |
1965 | struct drm_i915_private *dev_priv = dev->dev_private; | |
1966 | int latency = SNB_READ_WM0_LATENCY() * 100; /* In unit 0.1us */ | |
1967 | u32 val; | |
1968 | int fbc_wm, plane_wm, cursor_wm; | |
1969 | int ignore_fbc_wm, ignore_plane_wm, ignore_cursor_wm; | |
1970 | unsigned int enabled; | |
1971 | ||
1972 | enabled = 0; | |
51cea1f4 | 1973 | if (g4x_compute_wm0(dev, PIPE_A, |
c43d0188 CW |
1974 | &sandybridge_display_wm_info, latency, |
1975 | &sandybridge_cursor_wm_info, latency, | |
1976 | &plane_wm, &cursor_wm)) { | |
1977 | val = I915_READ(WM0_PIPEA_ILK); | |
1978 | val &= ~(WM0_PIPE_PLANE_MASK | WM0_PIPE_CURSOR_MASK); | |
1979 | I915_WRITE(WM0_PIPEA_ILK, val | | |
1980 | ((plane_wm << WM0_PIPE_PLANE_SHIFT) | cursor_wm)); | |
1981 | DRM_DEBUG_KMS("FIFO watermarks For pipe A -" | |
1982 | " plane %d, " "cursor: %d\n", | |
1983 | plane_wm, cursor_wm); | |
51cea1f4 | 1984 | enabled |= 1 << PIPE_A; |
c43d0188 CW |
1985 | } |
1986 | ||
51cea1f4 | 1987 | if (g4x_compute_wm0(dev, PIPE_B, |
c43d0188 CW |
1988 | &sandybridge_display_wm_info, latency, |
1989 | &sandybridge_cursor_wm_info, latency, | |
1990 | &plane_wm, &cursor_wm)) { | |
1991 | val = I915_READ(WM0_PIPEB_ILK); | |
1992 | val &= ~(WM0_PIPE_PLANE_MASK | WM0_PIPE_CURSOR_MASK); | |
1993 | I915_WRITE(WM0_PIPEB_ILK, val | | |
1994 | ((plane_wm << WM0_PIPE_PLANE_SHIFT) | cursor_wm)); | |
1995 | DRM_DEBUG_KMS("FIFO watermarks For pipe B -" | |
1996 | " plane %d, cursor: %d\n", | |
1997 | plane_wm, cursor_wm); | |
51cea1f4 | 1998 | enabled |= 1 << PIPE_B; |
c43d0188 CW |
1999 | } |
2000 | ||
51cea1f4 | 2001 | if (g4x_compute_wm0(dev, PIPE_C, |
b445e3b0 ED |
2002 | &sandybridge_display_wm_info, latency, |
2003 | &sandybridge_cursor_wm_info, latency, | |
2004 | &plane_wm, &cursor_wm)) { | |
2005 | val = I915_READ(WM0_PIPEC_IVB); | |
2006 | val &= ~(WM0_PIPE_PLANE_MASK | WM0_PIPE_CURSOR_MASK); | |
2007 | I915_WRITE(WM0_PIPEC_IVB, val | | |
2008 | ((plane_wm << WM0_PIPE_PLANE_SHIFT) | cursor_wm)); | |
2009 | DRM_DEBUG_KMS("FIFO watermarks For pipe C -" | |
2010 | " plane %d, cursor: %d\n", | |
2011 | plane_wm, cursor_wm); | |
51cea1f4 | 2012 | enabled |= 1 << PIPE_C; |
b445e3b0 ED |
2013 | } |
2014 | ||
2015 | /* | |
2016 | * Calculate and update the self-refresh watermark only when one | |
2017 | * display plane is used. | |
2018 | * | |
2019 | * SNB support 3 levels of watermark. | |
2020 | * | |
2021 | * WM1/WM2/WM2 watermarks have to be enabled in the ascending order, | |
2022 | * and disabled in the descending order | |
2023 | * | |
2024 | */ | |
2025 | I915_WRITE(WM3_LP_ILK, 0); | |
2026 | I915_WRITE(WM2_LP_ILK, 0); | |
2027 | I915_WRITE(WM1_LP_ILK, 0); | |
2028 | ||
2029 | if (!single_plane_enabled(enabled) || | |
2030 | dev_priv->sprite_scaling_enabled) | |
2031 | return; | |
2032 | enabled = ffs(enabled) - 1; | |
2033 | ||
2034 | /* WM1 */ | |
2035 | if (!ironlake_compute_srwm(dev, 1, enabled, | |
2036 | SNB_READ_WM1_LATENCY() * 500, | |
2037 | &sandybridge_display_srwm_info, | |
2038 | &sandybridge_cursor_srwm_info, | |
2039 | &fbc_wm, &plane_wm, &cursor_wm)) | |
2040 | return; | |
2041 | ||
2042 | I915_WRITE(WM1_LP_ILK, | |
2043 | WM1_LP_SR_EN | | |
2044 | (SNB_READ_WM1_LATENCY() << WM1_LP_LATENCY_SHIFT) | | |
2045 | (fbc_wm << WM1_LP_FBC_SHIFT) | | |
2046 | (plane_wm << WM1_LP_SR_SHIFT) | | |
2047 | cursor_wm); | |
2048 | ||
2049 | /* WM2 */ | |
2050 | if (!ironlake_compute_srwm(dev, 2, enabled, | |
2051 | SNB_READ_WM2_LATENCY() * 500, | |
2052 | &sandybridge_display_srwm_info, | |
2053 | &sandybridge_cursor_srwm_info, | |
2054 | &fbc_wm, &plane_wm, &cursor_wm)) | |
2055 | return; | |
2056 | ||
2057 | I915_WRITE(WM2_LP_ILK, | |
2058 | WM2_LP_EN | | |
2059 | (SNB_READ_WM2_LATENCY() << WM1_LP_LATENCY_SHIFT) | | |
2060 | (fbc_wm << WM1_LP_FBC_SHIFT) | | |
2061 | (plane_wm << WM1_LP_SR_SHIFT) | | |
2062 | cursor_wm); | |
2063 | ||
c43d0188 | 2064 | /* WM3, note we have to correct the cursor latency */ |
b445e3b0 ED |
2065 | if (!ironlake_compute_srwm(dev, 3, enabled, |
2066 | SNB_READ_WM3_LATENCY() * 500, | |
2067 | &sandybridge_display_srwm_info, | |
2068 | &sandybridge_cursor_srwm_info, | |
c43d0188 CW |
2069 | &fbc_wm, &plane_wm, &ignore_cursor_wm) || |
2070 | !ironlake_compute_srwm(dev, 3, enabled, | |
2071 | 2 * SNB_READ_WM3_LATENCY() * 500, | |
2072 | &sandybridge_display_srwm_info, | |
2073 | &sandybridge_cursor_srwm_info, | |
2074 | &ignore_fbc_wm, &ignore_plane_wm, &cursor_wm)) | |
b445e3b0 ED |
2075 | return; |
2076 | ||
2077 | I915_WRITE(WM3_LP_ILK, | |
2078 | WM3_LP_EN | | |
2079 | (SNB_READ_WM3_LATENCY() << WM1_LP_LATENCY_SHIFT) | | |
2080 | (fbc_wm << WM1_LP_FBC_SHIFT) | | |
2081 | (plane_wm << WM1_LP_SR_SHIFT) | | |
2082 | cursor_wm); | |
2083 | } | |
2084 | ||
801bcfff PZ |
2085 | static uint32_t hsw_wm_get_pixel_rate(struct drm_device *dev, |
2086 | struct drm_crtc *crtc) | |
2087 | { | |
2088 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc); | |
2089 | uint32_t pixel_rate, pfit_size; | |
2090 | ||
ff9a6750 | 2091 | pixel_rate = intel_crtc->config.adjusted_mode.clock; |
801bcfff PZ |
2092 | |
2093 | /* We only use IF-ID interlacing. If we ever use PF-ID we'll need to | |
2094 | * adjust the pixel_rate here. */ | |
2095 | ||
2096 | pfit_size = intel_crtc->config.pch_pfit.size; | |
2097 | if (pfit_size) { | |
2098 | uint64_t pipe_w, pipe_h, pfit_w, pfit_h; | |
2099 | ||
2100 | pipe_w = intel_crtc->config.requested_mode.hdisplay; | |
2101 | pipe_h = intel_crtc->config.requested_mode.vdisplay; | |
2102 | pfit_w = (pfit_size >> 16) & 0xFFFF; | |
2103 | pfit_h = pfit_size & 0xFFFF; | |
2104 | if (pipe_w < pfit_w) | |
2105 | pipe_w = pfit_w; | |
2106 | if (pipe_h < pfit_h) | |
2107 | pipe_h = pfit_h; | |
2108 | ||
2109 | pixel_rate = div_u64((uint64_t) pixel_rate * pipe_w * pipe_h, | |
2110 | pfit_w * pfit_h); | |
2111 | } | |
2112 | ||
2113 | return pixel_rate; | |
2114 | } | |
2115 | ||
2116 | static uint32_t hsw_wm_method1(uint32_t pixel_rate, uint8_t bytes_per_pixel, | |
2117 | uint32_t latency) | |
2118 | { | |
2119 | uint64_t ret; | |
2120 | ||
2121 | ret = (uint64_t) pixel_rate * bytes_per_pixel * latency; | |
2122 | ret = DIV_ROUND_UP_ULL(ret, 64 * 10000) + 2; | |
2123 | ||
2124 | return ret; | |
2125 | } | |
2126 | ||
2127 | static uint32_t hsw_wm_method2(uint32_t pixel_rate, uint32_t pipe_htotal, | |
2128 | uint32_t horiz_pixels, uint8_t bytes_per_pixel, | |
2129 | uint32_t latency) | |
2130 | { | |
2131 | uint32_t ret; | |
2132 | ||
2133 | ret = (latency * pixel_rate) / (pipe_htotal * 10000); | |
2134 | ret = (ret + 1) * horiz_pixels * bytes_per_pixel; | |
2135 | ret = DIV_ROUND_UP(ret, 64) + 2; | |
2136 | return ret; | |
2137 | } | |
2138 | ||
cca32e9a PZ |
2139 | static uint32_t hsw_wm_fbc(uint32_t pri_val, uint32_t horiz_pixels, |
2140 | uint8_t bytes_per_pixel) | |
2141 | { | |
2142 | return DIV_ROUND_UP(pri_val * 64, horiz_pixels * bytes_per_pixel) + 2; | |
2143 | } | |
2144 | ||
801bcfff PZ |
2145 | struct hsw_pipe_wm_parameters { |
2146 | bool active; | |
2147 | bool sprite_enabled; | |
2148 | uint8_t pri_bytes_per_pixel; | |
2149 | uint8_t spr_bytes_per_pixel; | |
2150 | uint8_t cur_bytes_per_pixel; | |
2151 | uint32_t pri_horiz_pixels; | |
2152 | uint32_t spr_horiz_pixels; | |
2153 | uint32_t cur_horiz_pixels; | |
2154 | uint32_t pipe_htotal; | |
2155 | uint32_t pixel_rate; | |
2156 | }; | |
2157 | ||
cca32e9a PZ |
2158 | struct hsw_wm_maximums { |
2159 | uint16_t pri; | |
2160 | uint16_t spr; | |
2161 | uint16_t cur; | |
2162 | uint16_t fbc; | |
2163 | }; | |
2164 | ||
2165 | struct hsw_lp_wm_result { | |
2166 | bool enable; | |
2167 | bool fbc_enable; | |
2168 | uint32_t pri_val; | |
2169 | uint32_t spr_val; | |
2170 | uint32_t cur_val; | |
2171 | uint32_t fbc_val; | |
2172 | }; | |
2173 | ||
801bcfff PZ |
2174 | struct hsw_wm_values { |
2175 | uint32_t wm_pipe[3]; | |
2176 | uint32_t wm_lp[3]; | |
2177 | uint32_t wm_lp_spr[3]; | |
2178 | uint32_t wm_linetime[3]; | |
cca32e9a | 2179 | bool enable_fbc_wm; |
801bcfff PZ |
2180 | }; |
2181 | ||
2182 | enum hsw_data_buf_partitioning { | |
2183 | HSW_DATA_BUF_PART_1_2, | |
2184 | HSW_DATA_BUF_PART_5_6, | |
2185 | }; | |
2186 | ||
cca32e9a PZ |
2187 | /* For both WM_PIPE and WM_LP. */ |
2188 | static uint32_t hsw_compute_pri_wm(struct hsw_pipe_wm_parameters *params, | |
2189 | uint32_t mem_value, | |
2190 | bool is_lp) | |
801bcfff | 2191 | { |
cca32e9a PZ |
2192 | uint32_t method1, method2; |
2193 | ||
801bcfff PZ |
2194 | /* TODO: for now, assume the primary plane is always enabled. */ |
2195 | if (!params->active) | |
2196 | return 0; | |
2197 | ||
cca32e9a PZ |
2198 | method1 = hsw_wm_method1(params->pixel_rate, |
2199 | params->pri_bytes_per_pixel, | |
2200 | mem_value); | |
2201 | ||
2202 | if (!is_lp) | |
2203 | return method1; | |
2204 | ||
2205 | method2 = hsw_wm_method2(params->pixel_rate, | |
2206 | params->pipe_htotal, | |
2207 | params->pri_horiz_pixels, | |
2208 | params->pri_bytes_per_pixel, | |
2209 | mem_value); | |
2210 | ||
2211 | return min(method1, method2); | |
801bcfff PZ |
2212 | } |
2213 | ||
2214 | /* For both WM_PIPE and WM_LP. */ | |
2215 | static uint32_t hsw_compute_spr_wm(struct hsw_pipe_wm_parameters *params, | |
2216 | uint32_t mem_value) | |
2217 | { | |
2218 | uint32_t method1, method2; | |
2219 | ||
2220 | if (!params->active || !params->sprite_enabled) | |
2221 | return 0; | |
2222 | ||
2223 | method1 = hsw_wm_method1(params->pixel_rate, | |
2224 | params->spr_bytes_per_pixel, | |
2225 | mem_value); | |
2226 | method2 = hsw_wm_method2(params->pixel_rate, | |
2227 | params->pipe_htotal, | |
2228 | params->spr_horiz_pixels, | |
2229 | params->spr_bytes_per_pixel, | |
2230 | mem_value); | |
2231 | return min(method1, method2); | |
2232 | } | |
2233 | ||
2234 | /* For both WM_PIPE and WM_LP. */ | |
2235 | static uint32_t hsw_compute_cur_wm(struct hsw_pipe_wm_parameters *params, | |
2236 | uint32_t mem_value) | |
2237 | { | |
2238 | if (!params->active) | |
2239 | return 0; | |
2240 | ||
2241 | return hsw_wm_method2(params->pixel_rate, | |
2242 | params->pipe_htotal, | |
2243 | params->cur_horiz_pixels, | |
2244 | params->cur_bytes_per_pixel, | |
2245 | mem_value); | |
2246 | } | |
2247 | ||
cca32e9a PZ |
2248 | /* Only for WM_LP. */ |
2249 | static uint32_t hsw_compute_fbc_wm(struct hsw_pipe_wm_parameters *params, | |
2250 | uint32_t pri_val, | |
2251 | uint32_t mem_value) | |
2252 | { | |
2253 | if (!params->active) | |
2254 | return 0; | |
2255 | ||
2256 | return hsw_wm_fbc(pri_val, | |
2257 | params->pri_horiz_pixels, | |
2258 | params->pri_bytes_per_pixel); | |
2259 | } | |
2260 | ||
2261 | static bool hsw_compute_lp_wm(uint32_t mem_value, struct hsw_wm_maximums *max, | |
2262 | struct hsw_pipe_wm_parameters *params, | |
2263 | struct hsw_lp_wm_result *result) | |
2264 | { | |
2265 | enum pipe pipe; | |
2266 | uint32_t pri_val[3], spr_val[3], cur_val[3], fbc_val[3]; | |
2267 | ||
2268 | for (pipe = PIPE_A; pipe <= PIPE_C; pipe++) { | |
2269 | struct hsw_pipe_wm_parameters *p = ¶ms[pipe]; | |
2270 | ||
2271 | pri_val[pipe] = hsw_compute_pri_wm(p, mem_value, true); | |
2272 | spr_val[pipe] = hsw_compute_spr_wm(p, mem_value); | |
2273 | cur_val[pipe] = hsw_compute_cur_wm(p, mem_value); | |
2274 | fbc_val[pipe] = hsw_compute_fbc_wm(p, pri_val[pipe], mem_value); | |
2275 | } | |
2276 | ||
2277 | result->pri_val = max3(pri_val[0], pri_val[1], pri_val[2]); | |
2278 | result->spr_val = max3(spr_val[0], spr_val[1], spr_val[2]); | |
2279 | result->cur_val = max3(cur_val[0], cur_val[1], cur_val[2]); | |
2280 | result->fbc_val = max3(fbc_val[0], fbc_val[1], fbc_val[2]); | |
2281 | ||
2282 | if (result->fbc_val > max->fbc) { | |
2283 | result->fbc_enable = false; | |
2284 | result->fbc_val = 0; | |
2285 | } else { | |
2286 | result->fbc_enable = true; | |
2287 | } | |
2288 | ||
2289 | result->enable = result->pri_val <= max->pri && | |
2290 | result->spr_val <= max->spr && | |
2291 | result->cur_val <= max->cur; | |
2292 | return result->enable; | |
2293 | } | |
2294 | ||
801bcfff PZ |
2295 | static uint32_t hsw_compute_wm_pipe(struct drm_i915_private *dev_priv, |
2296 | uint32_t mem_value, enum pipe pipe, | |
2297 | struct hsw_pipe_wm_parameters *params) | |
2298 | { | |
2299 | uint32_t pri_val, cur_val, spr_val; | |
2300 | ||
cca32e9a | 2301 | pri_val = hsw_compute_pri_wm(params, mem_value, false); |
801bcfff PZ |
2302 | spr_val = hsw_compute_spr_wm(params, mem_value); |
2303 | cur_val = hsw_compute_cur_wm(params, mem_value); | |
2304 | ||
2305 | WARN(pri_val > 127, | |
2306 | "Primary WM error, mode not supported for pipe %c\n", | |
2307 | pipe_name(pipe)); | |
2308 | WARN(spr_val > 127, | |
2309 | "Sprite WM error, mode not supported for pipe %c\n", | |
2310 | pipe_name(pipe)); | |
2311 | WARN(cur_val > 63, | |
2312 | "Cursor WM error, mode not supported for pipe %c\n", | |
2313 | pipe_name(pipe)); | |
2314 | ||
2315 | return (pri_val << WM0_PIPE_PLANE_SHIFT) | | |
2316 | (spr_val << WM0_PIPE_SPRITE_SHIFT) | | |
2317 | cur_val; | |
2318 | } | |
2319 | ||
2320 | static uint32_t | |
2321 | hsw_compute_linetime_wm(struct drm_device *dev, struct drm_crtc *crtc) | |
1f8eeabf ED |
2322 | { |
2323 | struct drm_i915_private *dev_priv = dev->dev_private; | |
1011d8c4 | 2324 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc); |
1011d8c4 | 2325 | struct drm_display_mode *mode = &intel_crtc->config.adjusted_mode; |
85a02deb | 2326 | u32 linetime, ips_linetime; |
1f8eeabf | 2327 | |
801bcfff PZ |
2328 | if (!intel_crtc_active(crtc)) |
2329 | return 0; | |
1011d8c4 | 2330 | |
1f8eeabf ED |
2331 | /* The WM are computed with base on how long it takes to fill a single |
2332 | * row at the given clock rate, multiplied by 8. | |
2333 | * */ | |
85a02deb PZ |
2334 | linetime = DIV_ROUND_CLOSEST(mode->htotal * 1000 * 8, mode->clock); |
2335 | ips_linetime = DIV_ROUND_CLOSEST(mode->htotal * 1000 * 8, | |
2336 | intel_ddi_get_cdclk_freq(dev_priv)); | |
1f8eeabf | 2337 | |
801bcfff PZ |
2338 | return PIPE_WM_LINETIME_IPS_LINETIME(ips_linetime) | |
2339 | PIPE_WM_LINETIME_TIME(linetime); | |
1f8eeabf ED |
2340 | } |
2341 | ||
801bcfff PZ |
2342 | static void hsw_compute_wm_parameters(struct drm_device *dev, |
2343 | struct hsw_pipe_wm_parameters *params, | |
cca32e9a | 2344 | uint32_t *wm, |
861f3389 PZ |
2345 | struct hsw_wm_maximums *lp_max_1_2, |
2346 | struct hsw_wm_maximums *lp_max_5_6) | |
1011d8c4 PZ |
2347 | { |
2348 | struct drm_i915_private *dev_priv = dev->dev_private; | |
2349 | struct drm_crtc *crtc; | |
801bcfff PZ |
2350 | struct drm_plane *plane; |
2351 | uint64_t sskpd = I915_READ64(MCH_SSKPD); | |
1011d8c4 | 2352 | enum pipe pipe; |
cca32e9a | 2353 | int pipes_active = 0, sprites_enabled = 0; |
1011d8c4 | 2354 | |
801bcfff PZ |
2355 | if ((sskpd >> 56) & 0xFF) |
2356 | wm[0] = (sskpd >> 56) & 0xFF; | |
2357 | else | |
2358 | wm[0] = sskpd & 0xF; | |
2359 | wm[1] = ((sskpd >> 4) & 0xFF) * 5; | |
2360 | wm[2] = ((sskpd >> 12) & 0xFF) * 5; | |
2361 | wm[3] = ((sskpd >> 20) & 0x1FF) * 5; | |
2362 | wm[4] = ((sskpd >> 32) & 0x1FF) * 5; | |
2363 | ||
2364 | list_for_each_entry(crtc, &dev->mode_config.crtc_list, head) { | |
2365 | struct intel_crtc *intel_crtc = to_intel_crtc(crtc); | |
2366 | struct hsw_pipe_wm_parameters *p; | |
2367 | ||
2368 | pipe = intel_crtc->pipe; | |
2369 | p = ¶ms[pipe]; | |
2370 | ||
2371 | p->active = intel_crtc_active(crtc); | |
2372 | if (!p->active) | |
2373 | continue; | |
2374 | ||
cca32e9a PZ |
2375 | pipes_active++; |
2376 | ||
801bcfff PZ |
2377 | p->pipe_htotal = intel_crtc->config.adjusted_mode.htotal; |
2378 | p->pixel_rate = hsw_wm_get_pixel_rate(dev, crtc); | |
2379 | p->pri_bytes_per_pixel = crtc->fb->bits_per_pixel / 8; | |
2380 | p->cur_bytes_per_pixel = 4; | |
2381 | p->pri_horiz_pixels = | |
2382 | intel_crtc->config.requested_mode.hdisplay; | |
2383 | p->cur_horiz_pixels = 64; | |
2384 | } | |
2385 | ||
2386 | list_for_each_entry(plane, &dev->mode_config.plane_list, head) { | |
2387 | struct intel_plane *intel_plane = to_intel_plane(plane); | |
2388 | struct hsw_pipe_wm_parameters *p; | |
2389 | ||
2390 | pipe = intel_plane->pipe; | |
2391 | p = ¶ms[pipe]; | |
2392 | ||
2393 | p->sprite_enabled = intel_plane->wm.enable; | |
2394 | p->spr_bytes_per_pixel = intel_plane->wm.bytes_per_pixel; | |
2395 | p->spr_horiz_pixels = intel_plane->wm.horiz_pixels; | |
cca32e9a PZ |
2396 | |
2397 | if (p->sprite_enabled) | |
2398 | sprites_enabled++; | |
2399 | } | |
2400 | ||
2401 | if (pipes_active > 1) { | |
861f3389 PZ |
2402 | lp_max_1_2->pri = lp_max_5_6->pri = sprites_enabled ? 128 : 256; |
2403 | lp_max_1_2->spr = lp_max_5_6->spr = 128; | |
2404 | lp_max_1_2->cur = lp_max_5_6->cur = 64; | |
cca32e9a PZ |
2405 | } else { |
2406 | lp_max_1_2->pri = sprites_enabled ? 384 : 768; | |
861f3389 | 2407 | lp_max_5_6->pri = sprites_enabled ? 128 : 768; |
cca32e9a | 2408 | lp_max_1_2->spr = 384; |
861f3389 PZ |
2409 | lp_max_5_6->spr = 640; |
2410 | lp_max_1_2->cur = lp_max_5_6->cur = 255; | |
801bcfff | 2411 | } |
861f3389 | 2412 | lp_max_1_2->fbc = lp_max_5_6->fbc = 15; |
801bcfff PZ |
2413 | } |
2414 | ||
2415 | static void hsw_compute_wm_results(struct drm_device *dev, | |
2416 | struct hsw_pipe_wm_parameters *params, | |
2417 | uint32_t *wm, | |
cca32e9a | 2418 | struct hsw_wm_maximums *lp_maximums, |
801bcfff PZ |
2419 | struct hsw_wm_values *results) |
2420 | { | |
2421 | struct drm_i915_private *dev_priv = dev->dev_private; | |
2422 | struct drm_crtc *crtc; | |
cca32e9a | 2423 | struct hsw_lp_wm_result lp_results[4] = {}; |
801bcfff | 2424 | enum pipe pipe; |
cca32e9a PZ |
2425 | int level, max_level, wm_lp; |
2426 | ||
2427 | for (level = 1; level <= 4; level++) | |
2428 | if (!hsw_compute_lp_wm(wm[level], lp_maximums, params, | |
2429 | &lp_results[level - 1])) | |
2430 | break; | |
2431 | max_level = level - 1; | |
2432 | ||
2433 | /* The spec says it is preferred to disable FBC WMs instead of disabling | |
2434 | * a WM level. */ | |
2435 | results->enable_fbc_wm = true; | |
2436 | for (level = 1; level <= max_level; level++) { | |
2437 | if (!lp_results[level - 1].fbc_enable) { | |
2438 | results->enable_fbc_wm = false; | |
2439 | break; | |
2440 | } | |
2441 | } | |
2442 | ||
2443 | memset(results, 0, sizeof(*results)); | |
2444 | for (wm_lp = 1; wm_lp <= 3; wm_lp++) { | |
2445 | const struct hsw_lp_wm_result *r; | |
801bcfff | 2446 | |
cca32e9a PZ |
2447 | level = (max_level == 4 && wm_lp > 1) ? wm_lp + 1 : wm_lp; |
2448 | if (level > max_level) | |
2449 | break; | |
2450 | ||
2451 | r = &lp_results[level - 1]; | |
2452 | results->wm_lp[wm_lp - 1] = HSW_WM_LP_VAL(level * 2, | |
2453 | r->fbc_val, | |
2454 | r->pri_val, | |
2455 | r->cur_val); | |
2456 | results->wm_lp_spr[wm_lp - 1] = r->spr_val; | |
2457 | } | |
801bcfff PZ |
2458 | |
2459 | for_each_pipe(pipe) | |
2460 | results->wm_pipe[pipe] = hsw_compute_wm_pipe(dev_priv, wm[0], | |
2461 | pipe, | |
2462 | ¶ms[pipe]); | |
1011d8c4 PZ |
2463 | |
2464 | for_each_pipe(pipe) { | |
2465 | crtc = dev_priv->pipe_to_crtc_mapping[pipe]; | |
801bcfff PZ |
2466 | results->wm_linetime[pipe] = hsw_compute_linetime_wm(dev, crtc); |
2467 | } | |
2468 | } | |
2469 | ||
861f3389 PZ |
2470 | /* Find the result with the highest level enabled. Check for enable_fbc_wm in |
2471 | * case both are at the same level. Prefer r1 in case they're the same. */ | |
f4db9321 DL |
2472 | static struct hsw_wm_values *hsw_find_best_result(struct hsw_wm_values *r1, |
2473 | struct hsw_wm_values *r2) | |
861f3389 PZ |
2474 | { |
2475 | int i, val_r1 = 0, val_r2 = 0; | |
2476 | ||
2477 | for (i = 0; i < 3; i++) { | |
2478 | if (r1->wm_lp[i] & WM3_LP_EN) | |
2479 | val_r1 = r1->wm_lp[i] & WM1_LP_LATENCY_MASK; | |
2480 | if (r2->wm_lp[i] & WM3_LP_EN) | |
2481 | val_r2 = r2->wm_lp[i] & WM1_LP_LATENCY_MASK; | |
2482 | } | |
2483 | ||
2484 | if (val_r1 == val_r2) { | |
2485 | if (r2->enable_fbc_wm && !r1->enable_fbc_wm) | |
2486 | return r2; | |
2487 | else | |
2488 | return r1; | |
2489 | } else if (val_r1 > val_r2) { | |
2490 | return r1; | |
2491 | } else { | |
2492 | return r2; | |
2493 | } | |
2494 | } | |
2495 | ||
801bcfff PZ |
2496 | /* |
2497 | * The spec says we shouldn't write when we don't need, because every write | |
2498 | * causes WMs to be re-evaluated, expending some power. | |
2499 | */ | |
2500 | static void hsw_write_wm_values(struct drm_i915_private *dev_priv, | |
2501 | struct hsw_wm_values *results, | |
2502 | enum hsw_data_buf_partitioning partitioning) | |
2503 | { | |
2504 | struct hsw_wm_values previous; | |
2505 | uint32_t val; | |
2506 | enum hsw_data_buf_partitioning prev_partitioning; | |
cca32e9a | 2507 | bool prev_enable_fbc_wm; |
801bcfff PZ |
2508 | |
2509 | previous.wm_pipe[0] = I915_READ(WM0_PIPEA_ILK); | |
2510 | previous.wm_pipe[1] = I915_READ(WM0_PIPEB_ILK); | |
2511 | previous.wm_pipe[2] = I915_READ(WM0_PIPEC_IVB); | |
2512 | previous.wm_lp[0] = I915_READ(WM1_LP_ILK); | |
2513 | previous.wm_lp[1] = I915_READ(WM2_LP_ILK); | |
2514 | previous.wm_lp[2] = I915_READ(WM3_LP_ILK); | |
2515 | previous.wm_lp_spr[0] = I915_READ(WM1S_LP_ILK); | |
2516 | previous.wm_lp_spr[1] = I915_READ(WM2S_LP_IVB); | |
2517 | previous.wm_lp_spr[2] = I915_READ(WM3S_LP_IVB); | |
2518 | previous.wm_linetime[0] = I915_READ(PIPE_WM_LINETIME(PIPE_A)); | |
2519 | previous.wm_linetime[1] = I915_READ(PIPE_WM_LINETIME(PIPE_B)); | |
2520 | previous.wm_linetime[2] = I915_READ(PIPE_WM_LINETIME(PIPE_C)); | |
2521 | ||
2522 | prev_partitioning = (I915_READ(WM_MISC) & WM_MISC_DATA_PARTITION_5_6) ? | |
2523 | HSW_DATA_BUF_PART_5_6 : HSW_DATA_BUF_PART_1_2; | |
2524 | ||
cca32e9a PZ |
2525 | prev_enable_fbc_wm = !(I915_READ(DISP_ARB_CTL) & DISP_FBC_WM_DIS); |
2526 | ||
801bcfff PZ |
2527 | if (memcmp(results->wm_pipe, previous.wm_pipe, |
2528 | sizeof(results->wm_pipe)) == 0 && | |
2529 | memcmp(results->wm_lp, previous.wm_lp, | |
2530 | sizeof(results->wm_lp)) == 0 && | |
2531 | memcmp(results->wm_lp_spr, previous.wm_lp_spr, | |
2532 | sizeof(results->wm_lp_spr)) == 0 && | |
2533 | memcmp(results->wm_linetime, previous.wm_linetime, | |
2534 | sizeof(results->wm_linetime)) == 0 && | |
cca32e9a PZ |
2535 | partitioning == prev_partitioning && |
2536 | results->enable_fbc_wm == prev_enable_fbc_wm) | |
801bcfff PZ |
2537 | return; |
2538 | ||
2539 | if (previous.wm_lp[2] != 0) | |
2540 | I915_WRITE(WM3_LP_ILK, 0); | |
2541 | if (previous.wm_lp[1] != 0) | |
2542 | I915_WRITE(WM2_LP_ILK, 0); | |
2543 | if (previous.wm_lp[0] != 0) | |
2544 | I915_WRITE(WM1_LP_ILK, 0); | |
2545 | ||
2546 | if (previous.wm_pipe[0] != results->wm_pipe[0]) | |
2547 | I915_WRITE(WM0_PIPEA_ILK, results->wm_pipe[0]); | |
2548 | if (previous.wm_pipe[1] != results->wm_pipe[1]) | |
2549 | I915_WRITE(WM0_PIPEB_ILK, results->wm_pipe[1]); | |
2550 | if (previous.wm_pipe[2] != results->wm_pipe[2]) | |
2551 | I915_WRITE(WM0_PIPEC_IVB, results->wm_pipe[2]); | |
2552 | ||
2553 | if (previous.wm_linetime[0] != results->wm_linetime[0]) | |
2554 | I915_WRITE(PIPE_WM_LINETIME(PIPE_A), results->wm_linetime[0]); | |
2555 | if (previous.wm_linetime[1] != results->wm_linetime[1]) | |
2556 | I915_WRITE(PIPE_WM_LINETIME(PIPE_B), results->wm_linetime[1]); | |
2557 | if (previous.wm_linetime[2] != results->wm_linetime[2]) | |
2558 | I915_WRITE(PIPE_WM_LINETIME(PIPE_C), results->wm_linetime[2]); | |
2559 | ||
2560 | if (prev_partitioning != partitioning) { | |
2561 | val = I915_READ(WM_MISC); | |
2562 | if (partitioning == HSW_DATA_BUF_PART_1_2) | |
2563 | val &= ~WM_MISC_DATA_PARTITION_5_6; | |
2564 | else | |
2565 | val |= WM_MISC_DATA_PARTITION_5_6; | |
2566 | I915_WRITE(WM_MISC, val); | |
1011d8c4 PZ |
2567 | } |
2568 | ||
cca32e9a PZ |
2569 | if (prev_enable_fbc_wm != results->enable_fbc_wm) { |
2570 | val = I915_READ(DISP_ARB_CTL); | |
2571 | if (results->enable_fbc_wm) | |
2572 | val &= ~DISP_FBC_WM_DIS; | |
2573 | else | |
2574 | val |= DISP_FBC_WM_DIS; | |
2575 | I915_WRITE(DISP_ARB_CTL, val); | |
2576 | } | |
2577 | ||
801bcfff PZ |
2578 | if (previous.wm_lp_spr[0] != results->wm_lp_spr[0]) |
2579 | I915_WRITE(WM1S_LP_ILK, results->wm_lp_spr[0]); | |
2580 | if (previous.wm_lp_spr[1] != results->wm_lp_spr[1]) | |
2581 | I915_WRITE(WM2S_LP_IVB, results->wm_lp_spr[1]); | |
2582 | if (previous.wm_lp_spr[2] != results->wm_lp_spr[2]) | |
2583 | I915_WRITE(WM3S_LP_IVB, results->wm_lp_spr[2]); | |
2584 | ||
2585 | if (results->wm_lp[0] != 0) | |
2586 | I915_WRITE(WM1_LP_ILK, results->wm_lp[0]); | |
2587 | if (results->wm_lp[1] != 0) | |
2588 | I915_WRITE(WM2_LP_ILK, results->wm_lp[1]); | |
2589 | if (results->wm_lp[2] != 0) | |
2590 | I915_WRITE(WM3_LP_ILK, results->wm_lp[2]); | |
2591 | } | |
2592 | ||
2593 | static void haswell_update_wm(struct drm_device *dev) | |
2594 | { | |
2595 | struct drm_i915_private *dev_priv = dev->dev_private; | |
861f3389 | 2596 | struct hsw_wm_maximums lp_max_1_2, lp_max_5_6; |
801bcfff | 2597 | struct hsw_pipe_wm_parameters params[3]; |
861f3389 | 2598 | struct hsw_wm_values results_1_2, results_5_6, *best_results; |
801bcfff | 2599 | uint32_t wm[5]; |
861f3389 PZ |
2600 | enum hsw_data_buf_partitioning partitioning; |
2601 | ||
2602 | hsw_compute_wm_parameters(dev, params, wm, &lp_max_1_2, &lp_max_5_6); | |
2603 | ||
2604 | hsw_compute_wm_results(dev, params, wm, &lp_max_1_2, &results_1_2); | |
2605 | if (lp_max_1_2.pri != lp_max_5_6.pri) { | |
2606 | hsw_compute_wm_results(dev, params, wm, &lp_max_5_6, | |
2607 | &results_5_6); | |
2608 | best_results = hsw_find_best_result(&results_1_2, &results_5_6); | |
2609 | } else { | |
2610 | best_results = &results_1_2; | |
2611 | } | |
2612 | ||
2613 | partitioning = (best_results == &results_1_2) ? | |
2614 | HSW_DATA_BUF_PART_1_2 : HSW_DATA_BUF_PART_5_6; | |
801bcfff | 2615 | |
861f3389 | 2616 | hsw_write_wm_values(dev_priv, best_results, partitioning); |
1011d8c4 PZ |
2617 | } |
2618 | ||
526682e9 PZ |
2619 | static void haswell_update_sprite_wm(struct drm_device *dev, int pipe, |
2620 | uint32_t sprite_width, int pixel_size, | |
2621 | bool enable) | |
2622 | { | |
2623 | struct drm_plane *plane; | |
2624 | ||
2625 | list_for_each_entry(plane, &dev->mode_config.plane_list, head) { | |
2626 | struct intel_plane *intel_plane = to_intel_plane(plane); | |
2627 | ||
2628 | if (intel_plane->pipe == pipe) { | |
2629 | intel_plane->wm.enable = enable; | |
2630 | intel_plane->wm.horiz_pixels = sprite_width + 1; | |
2631 | intel_plane->wm.bytes_per_pixel = pixel_size; | |
2632 | break; | |
2633 | } | |
2634 | } | |
2635 | ||
2636 | haswell_update_wm(dev); | |
2637 | } | |
2638 | ||
b445e3b0 ED |
2639 | static bool |
2640 | sandybridge_compute_sprite_wm(struct drm_device *dev, int plane, | |
2641 | uint32_t sprite_width, int pixel_size, | |
2642 | const struct intel_watermark_params *display, | |
2643 | int display_latency_ns, int *sprite_wm) | |
2644 | { | |
2645 | struct drm_crtc *crtc; | |
2646 | int clock; | |
2647 | int entries, tlb_miss; | |
2648 | ||
2649 | crtc = intel_get_crtc_for_plane(dev, plane); | |
3490ea5d | 2650 | if (!intel_crtc_active(crtc)) { |
b445e3b0 ED |
2651 | *sprite_wm = display->guard_size; |
2652 | return false; | |
2653 | } | |
2654 | ||
2655 | clock = crtc->mode.clock; | |
2656 | ||
2657 | /* Use the small buffer method to calculate the sprite watermark */ | |
2658 | entries = ((clock * pixel_size / 1000) * display_latency_ns) / 1000; | |
2659 | tlb_miss = display->fifo_size*display->cacheline_size - | |
2660 | sprite_width * 8; | |
2661 | if (tlb_miss > 0) | |
2662 | entries += tlb_miss; | |
2663 | entries = DIV_ROUND_UP(entries, display->cacheline_size); | |
2664 | *sprite_wm = entries + display->guard_size; | |
2665 | if (*sprite_wm > (int)display->max_wm) | |
2666 | *sprite_wm = display->max_wm; | |
2667 | ||
2668 | return true; | |
2669 | } | |
2670 | ||
2671 | static bool | |
2672 | sandybridge_compute_sprite_srwm(struct drm_device *dev, int plane, | |
2673 | uint32_t sprite_width, int pixel_size, | |
2674 | const struct intel_watermark_params *display, | |
2675 | int latency_ns, int *sprite_wm) | |
2676 | { | |
2677 | struct drm_crtc *crtc; | |
2678 | unsigned long line_time_us; | |
2679 | int clock; | |
2680 | int line_count, line_size; | |
2681 | int small, large; | |
2682 | int entries; | |
2683 | ||
2684 | if (!latency_ns) { | |
2685 | *sprite_wm = 0; | |
2686 | return false; | |
2687 | } | |
2688 | ||
2689 | crtc = intel_get_crtc_for_plane(dev, plane); | |
2690 | clock = crtc->mode.clock; | |
2691 | if (!clock) { | |
2692 | *sprite_wm = 0; | |
2693 | return false; | |
2694 | } | |
2695 | ||
2696 | line_time_us = (sprite_width * 1000) / clock; | |
2697 | if (!line_time_us) { | |
2698 | *sprite_wm = 0; | |
2699 | return false; | |
2700 | } | |
2701 | ||
2702 | line_count = (latency_ns / line_time_us + 1000) / 1000; | |
2703 | line_size = sprite_width * pixel_size; | |
2704 | ||
2705 | /* Use the minimum of the small and large buffer method for primary */ | |
2706 | small = ((clock * pixel_size / 1000) * latency_ns) / 1000; | |
2707 | large = line_count * line_size; | |
2708 | ||
2709 | entries = DIV_ROUND_UP(min(small, large), display->cacheline_size); | |
2710 | *sprite_wm = entries + display->guard_size; | |
2711 | ||
2712 | return *sprite_wm > 0x3ff ? false : true; | |
2713 | } | |
2714 | ||
1fa61106 | 2715 | static void sandybridge_update_sprite_wm(struct drm_device *dev, int pipe, |
4c4ff43a PZ |
2716 | uint32_t sprite_width, int pixel_size, |
2717 | bool enable) | |
b445e3b0 ED |
2718 | { |
2719 | struct drm_i915_private *dev_priv = dev->dev_private; | |
2720 | int latency = SNB_READ_WM0_LATENCY() * 100; /* In unit 0.1us */ | |
2721 | u32 val; | |
2722 | int sprite_wm, reg; | |
2723 | int ret; | |
2724 | ||
4c4ff43a PZ |
2725 | if (!enable) |
2726 | return; | |
2727 | ||
b445e3b0 ED |
2728 | switch (pipe) { |
2729 | case 0: | |
2730 | reg = WM0_PIPEA_ILK; | |
2731 | break; | |
2732 | case 1: | |
2733 | reg = WM0_PIPEB_ILK; | |
2734 | break; | |
2735 | case 2: | |
2736 | reg = WM0_PIPEC_IVB; | |
2737 | break; | |
2738 | default: | |
2739 | return; /* bad pipe */ | |
2740 | } | |
2741 | ||
2742 | ret = sandybridge_compute_sprite_wm(dev, pipe, sprite_width, pixel_size, | |
2743 | &sandybridge_display_wm_info, | |
2744 | latency, &sprite_wm); | |
2745 | if (!ret) { | |
84f44ce7 VS |
2746 | DRM_DEBUG_KMS("failed to compute sprite wm for pipe %c\n", |
2747 | pipe_name(pipe)); | |
b445e3b0 ED |
2748 | return; |
2749 | } | |
2750 | ||
2751 | val = I915_READ(reg); | |
2752 | val &= ~WM0_PIPE_SPRITE_MASK; | |
2753 | I915_WRITE(reg, val | (sprite_wm << WM0_PIPE_SPRITE_SHIFT)); | |
84f44ce7 | 2754 | DRM_DEBUG_KMS("sprite watermarks For pipe %c - %d\n", pipe_name(pipe), sprite_wm); |
b445e3b0 ED |
2755 | |
2756 | ||
2757 | ret = sandybridge_compute_sprite_srwm(dev, pipe, sprite_width, | |
2758 | pixel_size, | |
2759 | &sandybridge_display_srwm_info, | |
2760 | SNB_READ_WM1_LATENCY() * 500, | |
2761 | &sprite_wm); | |
2762 | if (!ret) { | |
84f44ce7 VS |
2763 | DRM_DEBUG_KMS("failed to compute sprite lp1 wm on pipe %c\n", |
2764 | pipe_name(pipe)); | |
b445e3b0 ED |
2765 | return; |
2766 | } | |
2767 | I915_WRITE(WM1S_LP_ILK, sprite_wm); | |
2768 | ||
2769 | /* Only IVB has two more LP watermarks for sprite */ | |
2770 | if (!IS_IVYBRIDGE(dev)) | |
2771 | return; | |
2772 | ||
2773 | ret = sandybridge_compute_sprite_srwm(dev, pipe, sprite_width, | |
2774 | pixel_size, | |
2775 | &sandybridge_display_srwm_info, | |
2776 | SNB_READ_WM2_LATENCY() * 500, | |
2777 | &sprite_wm); | |
2778 | if (!ret) { | |
84f44ce7 VS |
2779 | DRM_DEBUG_KMS("failed to compute sprite lp2 wm on pipe %c\n", |
2780 | pipe_name(pipe)); | |
b445e3b0 ED |
2781 | return; |
2782 | } | |
2783 | I915_WRITE(WM2S_LP_IVB, sprite_wm); | |
2784 | ||
2785 | ret = sandybridge_compute_sprite_srwm(dev, pipe, sprite_width, | |
2786 | pixel_size, | |
2787 | &sandybridge_display_srwm_info, | |
2788 | SNB_READ_WM3_LATENCY() * 500, | |
2789 | &sprite_wm); | |
2790 | if (!ret) { | |
84f44ce7 VS |
2791 | DRM_DEBUG_KMS("failed to compute sprite lp3 wm on pipe %c\n", |
2792 | pipe_name(pipe)); | |
b445e3b0 ED |
2793 | return; |
2794 | } | |
2795 | I915_WRITE(WM3S_LP_IVB, sprite_wm); | |
2796 | } | |
2797 | ||
2798 | /** | |
2799 | * intel_update_watermarks - update FIFO watermark values based on current modes | |
2800 | * | |
2801 | * Calculate watermark values for the various WM regs based on current mode | |
2802 | * and plane configuration. | |
2803 | * | |
2804 | * There are several cases to deal with here: | |
2805 | * - normal (i.e. non-self-refresh) | |
2806 | * - self-refresh (SR) mode | |
2807 | * - lines are large relative to FIFO size (buffer can hold up to 2) | |
2808 | * - lines are small relative to FIFO size (buffer can hold more than 2 | |
2809 | * lines), so need to account for TLB latency | |
2810 | * | |
2811 | * The normal calculation is: | |
2812 | * watermark = dotclock * bytes per pixel * latency | |
2813 | * where latency is platform & configuration dependent (we assume pessimal | |
2814 | * values here). | |
2815 | * | |
2816 | * The SR calculation is: | |
2817 | * watermark = (trunc(latency/line time)+1) * surface width * | |
2818 | * bytes per pixel | |
2819 | * where | |
2820 | * line time = htotal / dotclock | |
2821 | * surface width = hdisplay for normal plane and 64 for cursor | |
2822 | * and latency is assumed to be high, as above. | |
2823 | * | |
2824 | * The final value programmed to the register should always be rounded up, | |
2825 | * and include an extra 2 entries to account for clock crossings. | |
2826 | * | |
2827 | * We don't use the sprite, so we can ignore that. And on Crestline we have | |
2828 | * to set the non-SR watermarks to 8. | |
2829 | */ | |
2830 | void intel_update_watermarks(struct drm_device *dev) | |
2831 | { | |
2832 | struct drm_i915_private *dev_priv = dev->dev_private; | |
2833 | ||
2834 | if (dev_priv->display.update_wm) | |
2835 | dev_priv->display.update_wm(dev); | |
2836 | } | |
2837 | ||
2838 | void intel_update_sprite_watermarks(struct drm_device *dev, int pipe, | |
4c4ff43a PZ |
2839 | uint32_t sprite_width, int pixel_size, |
2840 | bool enable) | |
b445e3b0 ED |
2841 | { |
2842 | struct drm_i915_private *dev_priv = dev->dev_private; | |
2843 | ||
2844 | if (dev_priv->display.update_sprite_wm) | |
2845 | dev_priv->display.update_sprite_wm(dev, pipe, sprite_width, | |
4c4ff43a | 2846 | pixel_size, enable); |
b445e3b0 ED |
2847 | } |
2848 | ||
2b4e57bd ED |
2849 | static struct drm_i915_gem_object * |
2850 | intel_alloc_context_page(struct drm_device *dev) | |
2851 | { | |
2852 | struct drm_i915_gem_object *ctx; | |
2853 | int ret; | |
2854 | ||
2855 | WARN_ON(!mutex_is_locked(&dev->struct_mutex)); | |
2856 | ||
2857 | ctx = i915_gem_alloc_object(dev, 4096); | |
2858 | if (!ctx) { | |
2859 | DRM_DEBUG("failed to alloc power context, RC6 disabled\n"); | |
2860 | return NULL; | |
2861 | } | |
2862 | ||
86a1ee26 | 2863 | ret = i915_gem_object_pin(ctx, 4096, true, false); |
2b4e57bd ED |
2864 | if (ret) { |
2865 | DRM_ERROR("failed to pin power context: %d\n", ret); | |
2866 | goto err_unref; | |
2867 | } | |
2868 | ||
2869 | ret = i915_gem_object_set_to_gtt_domain(ctx, 1); | |
2870 | if (ret) { | |
2871 | DRM_ERROR("failed to set-domain on power context: %d\n", ret); | |
2872 | goto err_unpin; | |
2873 | } | |
2874 | ||
2875 | return ctx; | |
2876 | ||
2877 | err_unpin: | |
2878 | i915_gem_object_unpin(ctx); | |
2879 | err_unref: | |
2880 | drm_gem_object_unreference(&ctx->base); | |
2b4e57bd ED |
2881 | return NULL; |
2882 | } | |
2883 | ||
9270388e DV |
2884 | /** |
2885 | * Lock protecting IPS related data structures | |
9270388e DV |
2886 | */ |
2887 | DEFINE_SPINLOCK(mchdev_lock); | |
2888 | ||
2889 | /* Global for IPS driver to get at the current i915 device. Protected by | |
2890 | * mchdev_lock. */ | |
2891 | static struct drm_i915_private *i915_mch_dev; | |
2892 | ||
2b4e57bd ED |
2893 | bool ironlake_set_drps(struct drm_device *dev, u8 val) |
2894 | { | |
2895 | struct drm_i915_private *dev_priv = dev->dev_private; | |
2896 | u16 rgvswctl; | |
2897 | ||
9270388e DV |
2898 | assert_spin_locked(&mchdev_lock); |
2899 | ||
2b4e57bd ED |
2900 | rgvswctl = I915_READ16(MEMSWCTL); |
2901 | if (rgvswctl & MEMCTL_CMD_STS) { | |
2902 | DRM_DEBUG("gpu busy, RCS change rejected\n"); | |
2903 | return false; /* still busy with another command */ | |
2904 | } | |
2905 | ||
2906 | rgvswctl = (MEMCTL_CMD_CHFREQ << MEMCTL_CMD_SHIFT) | | |
2907 | (val << MEMCTL_FREQ_SHIFT) | MEMCTL_SFCAVM; | |
2908 | I915_WRITE16(MEMSWCTL, rgvswctl); | |
2909 | POSTING_READ16(MEMSWCTL); | |
2910 | ||
2911 | rgvswctl |= MEMCTL_CMD_STS; | |
2912 | I915_WRITE16(MEMSWCTL, rgvswctl); | |
2913 | ||
2914 | return true; | |
2915 | } | |
2916 | ||
8090c6b9 | 2917 | static void ironlake_enable_drps(struct drm_device *dev) |
2b4e57bd ED |
2918 | { |
2919 | struct drm_i915_private *dev_priv = dev->dev_private; | |
2920 | u32 rgvmodectl = I915_READ(MEMMODECTL); | |
2921 | u8 fmax, fmin, fstart, vstart; | |
2922 | ||
9270388e DV |
2923 | spin_lock_irq(&mchdev_lock); |
2924 | ||
2b4e57bd ED |
2925 | /* Enable temp reporting */ |
2926 | I915_WRITE16(PMMISC, I915_READ(PMMISC) | MCPPCE_EN); | |
2927 | I915_WRITE16(TSC1, I915_READ(TSC1) | TSE); | |
2928 | ||
2929 | /* 100ms RC evaluation intervals */ | |
2930 | I915_WRITE(RCUPEI, 100000); | |
2931 | I915_WRITE(RCDNEI, 100000); | |
2932 | ||
2933 | /* Set max/min thresholds to 90ms and 80ms respectively */ | |
2934 | I915_WRITE(RCBMAXAVG, 90000); | |
2935 | I915_WRITE(RCBMINAVG, 80000); | |
2936 | ||
2937 | I915_WRITE(MEMIHYST, 1); | |
2938 | ||
2939 | /* Set up min, max, and cur for interrupt handling */ | |
2940 | fmax = (rgvmodectl & MEMMODE_FMAX_MASK) >> MEMMODE_FMAX_SHIFT; | |
2941 | fmin = (rgvmodectl & MEMMODE_FMIN_MASK); | |
2942 | fstart = (rgvmodectl & MEMMODE_FSTART_MASK) >> | |
2943 | MEMMODE_FSTART_SHIFT; | |
2944 | ||
2945 | vstart = (I915_READ(PXVFREQ_BASE + (fstart * 4)) & PXVFREQ_PX_MASK) >> | |
2946 | PXVFREQ_PX_SHIFT; | |
2947 | ||
20e4d407 DV |
2948 | dev_priv->ips.fmax = fmax; /* IPS callback will increase this */ |
2949 | dev_priv->ips.fstart = fstart; | |
2b4e57bd | 2950 | |
20e4d407 DV |
2951 | dev_priv->ips.max_delay = fstart; |
2952 | dev_priv->ips.min_delay = fmin; | |
2953 | dev_priv->ips.cur_delay = fstart; | |
2b4e57bd ED |
2954 | |
2955 | DRM_DEBUG_DRIVER("fmax: %d, fmin: %d, fstart: %d\n", | |
2956 | fmax, fmin, fstart); | |
2957 | ||
2958 | I915_WRITE(MEMINTREN, MEMINT_CX_SUPR_EN | MEMINT_EVAL_CHG_EN); | |
2959 | ||
2960 | /* | |
2961 | * Interrupts will be enabled in ironlake_irq_postinstall | |
2962 | */ | |
2963 | ||
2964 | I915_WRITE(VIDSTART, vstart); | |
2965 | POSTING_READ(VIDSTART); | |
2966 | ||
2967 | rgvmodectl |= MEMMODE_SWMODE_EN; | |
2968 | I915_WRITE(MEMMODECTL, rgvmodectl); | |
2969 | ||
9270388e | 2970 | if (wait_for_atomic((I915_READ(MEMSWCTL) & MEMCTL_CMD_STS) == 0, 10)) |
2b4e57bd | 2971 | DRM_ERROR("stuck trying to change perf mode\n"); |
9270388e | 2972 | mdelay(1); |
2b4e57bd ED |
2973 | |
2974 | ironlake_set_drps(dev, fstart); | |
2975 | ||
20e4d407 | 2976 | dev_priv->ips.last_count1 = I915_READ(0x112e4) + I915_READ(0x112e8) + |
2b4e57bd | 2977 | I915_READ(0x112e0); |
20e4d407 DV |
2978 | dev_priv->ips.last_time1 = jiffies_to_msecs(jiffies); |
2979 | dev_priv->ips.last_count2 = I915_READ(0x112f4); | |
2980 | getrawmonotonic(&dev_priv->ips.last_time2); | |
9270388e DV |
2981 | |
2982 | spin_unlock_irq(&mchdev_lock); | |
2b4e57bd ED |
2983 | } |
2984 | ||
8090c6b9 | 2985 | static void ironlake_disable_drps(struct drm_device *dev) |
2b4e57bd ED |
2986 | { |
2987 | struct drm_i915_private *dev_priv = dev->dev_private; | |
9270388e DV |
2988 | u16 rgvswctl; |
2989 | ||
2990 | spin_lock_irq(&mchdev_lock); | |
2991 | ||
2992 | rgvswctl = I915_READ16(MEMSWCTL); | |
2b4e57bd ED |
2993 | |
2994 | /* Ack interrupts, disable EFC interrupt */ | |
2995 | I915_WRITE(MEMINTREN, I915_READ(MEMINTREN) & ~MEMINT_EVAL_CHG_EN); | |
2996 | I915_WRITE(MEMINTRSTS, MEMINT_EVAL_CHG); | |
2997 | I915_WRITE(DEIER, I915_READ(DEIER) & ~DE_PCU_EVENT); | |
2998 | I915_WRITE(DEIIR, DE_PCU_EVENT); | |
2999 | I915_WRITE(DEIMR, I915_READ(DEIMR) | DE_PCU_EVENT); | |
3000 | ||
3001 | /* Go back to the starting frequency */ | |
20e4d407 | 3002 | ironlake_set_drps(dev, dev_priv->ips.fstart); |
9270388e | 3003 | mdelay(1); |
2b4e57bd ED |
3004 | rgvswctl |= MEMCTL_CMD_STS; |
3005 | I915_WRITE(MEMSWCTL, rgvswctl); | |
9270388e | 3006 | mdelay(1); |
2b4e57bd | 3007 | |
9270388e | 3008 | spin_unlock_irq(&mchdev_lock); |
2b4e57bd ED |
3009 | } |
3010 | ||
acbe9475 DV |
3011 | /* There's a funny hw issue where the hw returns all 0 when reading from |
3012 | * GEN6_RP_INTERRUPT_LIMITS. Hence we always need to compute the desired value | |
3013 | * ourselves, instead of doing a rmw cycle (which might result in us clearing | |
3014 | * all limits and the gpu stuck at whatever frequency it is at atm). | |
3015 | */ | |
65bccb5c | 3016 | static u32 gen6_rps_limits(struct drm_i915_private *dev_priv, u8 *val) |
2b4e57bd | 3017 | { |
7b9e0ae6 | 3018 | u32 limits; |
2b4e57bd | 3019 | |
7b9e0ae6 | 3020 | limits = 0; |
c6a828d3 DV |
3021 | |
3022 | if (*val >= dev_priv->rps.max_delay) | |
3023 | *val = dev_priv->rps.max_delay; | |
3024 | limits |= dev_priv->rps.max_delay << 24; | |
20b46e59 DV |
3025 | |
3026 | /* Only set the down limit when we've reached the lowest level to avoid | |
3027 | * getting more interrupts, otherwise leave this clear. This prevents a | |
3028 | * race in the hw when coming out of rc6: There's a tiny window where | |
3029 | * the hw runs at the minimal clock before selecting the desired | |
3030 | * frequency, if the down threshold expires in that window we will not | |
3031 | * receive a down interrupt. */ | |
c6a828d3 DV |
3032 | if (*val <= dev_priv->rps.min_delay) { |
3033 | *val = dev_priv->rps.min_delay; | |
3034 | limits |= dev_priv->rps.min_delay << 16; | |
20b46e59 DV |
3035 | } |
3036 | ||
3037 | return limits; | |
3038 | } | |
3039 | ||
3040 | void gen6_set_rps(struct drm_device *dev, u8 val) | |
3041 | { | |
3042 | struct drm_i915_private *dev_priv = dev->dev_private; | |
65bccb5c | 3043 | u32 limits = gen6_rps_limits(dev_priv, &val); |
7b9e0ae6 | 3044 | |
4fc688ce | 3045 | WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock)); |
79249636 BW |
3046 | WARN_ON(val > dev_priv->rps.max_delay); |
3047 | WARN_ON(val < dev_priv->rps.min_delay); | |
004777cb | 3048 | |
c6a828d3 | 3049 | if (val == dev_priv->rps.cur_delay) |
7b9e0ae6 CW |
3050 | return; |
3051 | ||
92bd1bf0 RV |
3052 | if (IS_HASWELL(dev)) |
3053 | I915_WRITE(GEN6_RPNSWREQ, | |
3054 | HSW_FREQUENCY(val)); | |
3055 | else | |
3056 | I915_WRITE(GEN6_RPNSWREQ, | |
3057 | GEN6_FREQUENCY(val) | | |
3058 | GEN6_OFFSET(0) | | |
3059 | GEN6_AGGRESSIVE_TURBO); | |
7b9e0ae6 CW |
3060 | |
3061 | /* Make sure we continue to get interrupts | |
3062 | * until we hit the minimum or maximum frequencies. | |
3063 | */ | |
3064 | I915_WRITE(GEN6_RP_INTERRUPT_LIMITS, limits); | |
3065 | ||
d5570a72 BW |
3066 | POSTING_READ(GEN6_RPNSWREQ); |
3067 | ||
c6a828d3 | 3068 | dev_priv->rps.cur_delay = val; |
be2cde9a DV |
3069 | |
3070 | trace_intel_gpu_freq_change(val * 50); | |
2b4e57bd ED |
3071 | } |
3072 | ||
80814ae4 VS |
3073 | /* |
3074 | * Wait until the previous freq change has completed, | |
3075 | * or the timeout elapsed, and then update our notion | |
3076 | * of the current GPU frequency. | |
3077 | */ | |
3078 | static void vlv_update_rps_cur_delay(struct drm_i915_private *dev_priv) | |
3079 | { | |
80814ae4 VS |
3080 | u32 pval; |
3081 | ||
3082 | WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock)); | |
3083 | ||
e8474409 VS |
3084 | if (wait_for(((pval = vlv_punit_read(dev_priv, PUNIT_REG_GPU_FREQ_STS)) & GENFREQSTATUS) == 0, 10)) |
3085 | DRM_DEBUG_DRIVER("timed out waiting for Punit\n"); | |
80814ae4 VS |
3086 | |
3087 | pval >>= 8; | |
3088 | ||
3089 | if (pval != dev_priv->rps.cur_delay) | |
3090 | DRM_DEBUG_DRIVER("Punit overrode GPU freq: %d MHz (%u) requested, but got %d Mhz (%u)\n", | |
3091 | vlv_gpu_freq(dev_priv->mem_freq, dev_priv->rps.cur_delay), | |
3092 | dev_priv->rps.cur_delay, | |
3093 | vlv_gpu_freq(dev_priv->mem_freq, pval), pval); | |
3094 | ||
3095 | dev_priv->rps.cur_delay = pval; | |
3096 | } | |
3097 | ||
0a073b84 JB |
3098 | void valleyview_set_rps(struct drm_device *dev, u8 val) |
3099 | { | |
3100 | struct drm_i915_private *dev_priv = dev->dev_private; | |
7a67092a VS |
3101 | |
3102 | gen6_rps_limits(dev_priv, &val); | |
0a073b84 JB |
3103 | |
3104 | WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock)); | |
3105 | WARN_ON(val > dev_priv->rps.max_delay); | |
3106 | WARN_ON(val < dev_priv->rps.min_delay); | |
3107 | ||
80814ae4 VS |
3108 | vlv_update_rps_cur_delay(dev_priv); |
3109 | ||
73008b98 | 3110 | DRM_DEBUG_DRIVER("GPU freq request from %d MHz (%u) to %d MHz (%u)\n", |
0a073b84 JB |
3111 | vlv_gpu_freq(dev_priv->mem_freq, |
3112 | dev_priv->rps.cur_delay), | |
73008b98 VS |
3113 | dev_priv->rps.cur_delay, |
3114 | vlv_gpu_freq(dev_priv->mem_freq, val), val); | |
0a073b84 JB |
3115 | |
3116 | if (val == dev_priv->rps.cur_delay) | |
3117 | return; | |
3118 | ||
ae99258f | 3119 | vlv_punit_write(dev_priv, PUNIT_REG_GPU_FREQ_REQ, val); |
0a073b84 | 3120 | |
80814ae4 | 3121 | dev_priv->rps.cur_delay = val; |
0a073b84 JB |
3122 | |
3123 | trace_intel_gpu_freq_change(vlv_gpu_freq(dev_priv->mem_freq, val)); | |
3124 | } | |
3125 | ||
44fc7d5c | 3126 | static void gen6_disable_rps_interrupts(struct drm_device *dev) |
2b4e57bd ED |
3127 | { |
3128 | struct drm_i915_private *dev_priv = dev->dev_private; | |
3129 | ||
2b4e57bd | 3130 | I915_WRITE(GEN6_PMINTRMSK, 0xffffffff); |
4848405c | 3131 | I915_WRITE(GEN6_PMIER, I915_READ(GEN6_PMIER) & ~GEN6_PM_RPS_EVENTS); |
2b4e57bd ED |
3132 | /* Complete PM interrupt masking here doesn't race with the rps work |
3133 | * item again unmasking PM interrupts because that is using a different | |
3134 | * register (PMIMR) to mask PM interrupts. The only risk is in leaving | |
3135 | * stale bits in PMIIR and PMIMR which gen6_enable_rps will clean up. */ | |
3136 | ||
59cdb63d | 3137 | spin_lock_irq(&dev_priv->irq_lock); |
c6a828d3 | 3138 | dev_priv->rps.pm_iir = 0; |
59cdb63d | 3139 | spin_unlock_irq(&dev_priv->irq_lock); |
2b4e57bd | 3140 | |
4848405c | 3141 | I915_WRITE(GEN6_PMIIR, GEN6_PM_RPS_EVENTS); |
2b4e57bd ED |
3142 | } |
3143 | ||
44fc7d5c | 3144 | static void gen6_disable_rps(struct drm_device *dev) |
d20d4f0c JB |
3145 | { |
3146 | struct drm_i915_private *dev_priv = dev->dev_private; | |
3147 | ||
3148 | I915_WRITE(GEN6_RC_CONTROL, 0); | |
44fc7d5c | 3149 | I915_WRITE(GEN6_RPNSWREQ, 1 << 31); |
d20d4f0c | 3150 | |
44fc7d5c DV |
3151 | gen6_disable_rps_interrupts(dev); |
3152 | } | |
3153 | ||
3154 | static void valleyview_disable_rps(struct drm_device *dev) | |
3155 | { | |
3156 | struct drm_i915_private *dev_priv = dev->dev_private; | |
3157 | ||
3158 | I915_WRITE(GEN6_RC_CONTROL, 0); | |
d20d4f0c | 3159 | |
44fc7d5c | 3160 | gen6_disable_rps_interrupts(dev); |
c9cddffc JB |
3161 | |
3162 | if (dev_priv->vlv_pctx) { | |
3163 | drm_gem_object_unreference(&dev_priv->vlv_pctx->base); | |
3164 | dev_priv->vlv_pctx = NULL; | |
3165 | } | |
d20d4f0c JB |
3166 | } |
3167 | ||
2b4e57bd ED |
3168 | int intel_enable_rc6(const struct drm_device *dev) |
3169 | { | |
eb4926e4 DL |
3170 | /* No RC6 before Ironlake */ |
3171 | if (INTEL_INFO(dev)->gen < 5) | |
3172 | return 0; | |
3173 | ||
456470eb | 3174 | /* Respect the kernel parameter if it is set */ |
2b4e57bd ED |
3175 | if (i915_enable_rc6 >= 0) |
3176 | return i915_enable_rc6; | |
3177 | ||
6567d748 CW |
3178 | /* Disable RC6 on Ironlake */ |
3179 | if (INTEL_INFO(dev)->gen == 5) | |
3180 | return 0; | |
2b4e57bd | 3181 | |
456470eb DV |
3182 | if (IS_HASWELL(dev)) { |
3183 | DRM_DEBUG_DRIVER("Haswell: only RC6 available\n"); | |
4a637c2c | 3184 | return INTEL_RC6_ENABLE; |
456470eb | 3185 | } |
2b4e57bd | 3186 | |
456470eb | 3187 | /* snb/ivb have more than one rc6 state. */ |
2b4e57bd ED |
3188 | if (INTEL_INFO(dev)->gen == 6) { |
3189 | DRM_DEBUG_DRIVER("Sandybridge: deep RC6 disabled\n"); | |
3190 | return INTEL_RC6_ENABLE; | |
3191 | } | |
456470eb | 3192 | |
2b4e57bd ED |
3193 | DRM_DEBUG_DRIVER("RC6 and deep RC6 enabled\n"); |
3194 | return (INTEL_RC6_ENABLE | INTEL_RC6p_ENABLE); | |
3195 | } | |
3196 | ||
44fc7d5c DV |
3197 | static void gen6_enable_rps_interrupts(struct drm_device *dev) |
3198 | { | |
3199 | struct drm_i915_private *dev_priv = dev->dev_private; | |
3200 | ||
3201 | spin_lock_irq(&dev_priv->irq_lock); | |
a0b3335a | 3202 | WARN_ON(dev_priv->rps.pm_iir); |
44fc7d5c DV |
3203 | I915_WRITE(GEN6_PMIMR, I915_READ(GEN6_PMIMR) & ~GEN6_PM_RPS_EVENTS); |
3204 | I915_WRITE(GEN6_PMIIR, GEN6_PM_RPS_EVENTS); | |
3205 | spin_unlock_irq(&dev_priv->irq_lock); | |
3206 | /* unmask all PM interrupts */ | |
3207 | I915_WRITE(GEN6_PMINTRMSK, 0); | |
3208 | } | |
3209 | ||
79f5b2c7 | 3210 | static void gen6_enable_rps(struct drm_device *dev) |
2b4e57bd | 3211 | { |
79f5b2c7 | 3212 | struct drm_i915_private *dev_priv = dev->dev_private; |
b4519513 | 3213 | struct intel_ring_buffer *ring; |
7b9e0ae6 CW |
3214 | u32 rp_state_cap; |
3215 | u32 gt_perf_status; | |
31643d54 | 3216 | u32 rc6vids, pcu_mbox, rc6_mask = 0; |
2b4e57bd | 3217 | u32 gtfifodbg; |
2b4e57bd | 3218 | int rc6_mode; |
42c0526c | 3219 | int i, ret; |
2b4e57bd | 3220 | |
4fc688ce | 3221 | WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock)); |
79f5b2c7 | 3222 | |
2b4e57bd ED |
3223 | /* Here begins a magic sequence of register writes to enable |
3224 | * auto-downclocking. | |
3225 | * | |
3226 | * Perhaps there might be some value in exposing these to | |
3227 | * userspace... | |
3228 | */ | |
3229 | I915_WRITE(GEN6_RC_STATE, 0); | |
2b4e57bd ED |
3230 | |
3231 | /* Clear the DBG now so we don't confuse earlier errors */ | |
3232 | if ((gtfifodbg = I915_READ(GTFIFODBG))) { | |
3233 | DRM_ERROR("GT fifo had a previous error %x\n", gtfifodbg); | |
3234 | I915_WRITE(GTFIFODBG, gtfifodbg); | |
3235 | } | |
3236 | ||
3237 | gen6_gt_force_wake_get(dev_priv); | |
3238 | ||
7b9e0ae6 CW |
3239 | rp_state_cap = I915_READ(GEN6_RP_STATE_CAP); |
3240 | gt_perf_status = I915_READ(GEN6_GT_PERF_STATUS); | |
3241 | ||
31c77388 BW |
3242 | /* In units of 50MHz */ |
3243 | dev_priv->rps.hw_max = dev_priv->rps.max_delay = rp_state_cap & 0xff; | |
c6a828d3 DV |
3244 | dev_priv->rps.min_delay = (rp_state_cap & 0xff0000) >> 16; |
3245 | dev_priv->rps.cur_delay = 0; | |
7b9e0ae6 | 3246 | |
2b4e57bd ED |
3247 | /* disable the counters and set deterministic thresholds */ |
3248 | I915_WRITE(GEN6_RC_CONTROL, 0); | |
3249 | ||
3250 | I915_WRITE(GEN6_RC1_WAKE_RATE_LIMIT, 1000 << 16); | |
3251 | I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 40 << 16 | 30); | |
3252 | I915_WRITE(GEN6_RC6pp_WAKE_RATE_LIMIT, 30); | |
3253 | I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000); | |
3254 | I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25); | |
3255 | ||
b4519513 CW |
3256 | for_each_ring(ring, dev_priv, i) |
3257 | I915_WRITE(RING_MAX_IDLE(ring->mmio_base), 10); | |
2b4e57bd ED |
3258 | |
3259 | I915_WRITE(GEN6_RC_SLEEP, 0); | |
3260 | I915_WRITE(GEN6_RC1e_THRESHOLD, 1000); | |
3261 | I915_WRITE(GEN6_RC6_THRESHOLD, 50000); | |
0920a487 | 3262 | I915_WRITE(GEN6_RC6p_THRESHOLD, 150000); |
2b4e57bd ED |
3263 | I915_WRITE(GEN6_RC6pp_THRESHOLD, 64000); /* unused */ |
3264 | ||
5a7dc92a | 3265 | /* Check if we are enabling RC6 */ |
2b4e57bd ED |
3266 | rc6_mode = intel_enable_rc6(dev_priv->dev); |
3267 | if (rc6_mode & INTEL_RC6_ENABLE) | |
3268 | rc6_mask |= GEN6_RC_CTL_RC6_ENABLE; | |
3269 | ||
5a7dc92a ED |
3270 | /* We don't use those on Haswell */ |
3271 | if (!IS_HASWELL(dev)) { | |
3272 | if (rc6_mode & INTEL_RC6p_ENABLE) | |
3273 | rc6_mask |= GEN6_RC_CTL_RC6p_ENABLE; | |
2b4e57bd | 3274 | |
5a7dc92a ED |
3275 | if (rc6_mode & INTEL_RC6pp_ENABLE) |
3276 | rc6_mask |= GEN6_RC_CTL_RC6pp_ENABLE; | |
3277 | } | |
2b4e57bd ED |
3278 | |
3279 | DRM_INFO("Enabling RC6 states: RC6 %s, RC6p %s, RC6pp %s\n", | |
5a7dc92a ED |
3280 | (rc6_mask & GEN6_RC_CTL_RC6_ENABLE) ? "on" : "off", |
3281 | (rc6_mask & GEN6_RC_CTL_RC6p_ENABLE) ? "on" : "off", | |
3282 | (rc6_mask & GEN6_RC_CTL_RC6pp_ENABLE) ? "on" : "off"); | |
2b4e57bd ED |
3283 | |
3284 | I915_WRITE(GEN6_RC_CONTROL, | |
3285 | rc6_mask | | |
3286 | GEN6_RC_CTL_EI_MODE(1) | | |
3287 | GEN6_RC_CTL_HW_ENABLE); | |
3288 | ||
92bd1bf0 RV |
3289 | if (IS_HASWELL(dev)) { |
3290 | I915_WRITE(GEN6_RPNSWREQ, | |
3291 | HSW_FREQUENCY(10)); | |
3292 | I915_WRITE(GEN6_RC_VIDEO_FREQ, | |
3293 | HSW_FREQUENCY(12)); | |
3294 | } else { | |
3295 | I915_WRITE(GEN6_RPNSWREQ, | |
3296 | GEN6_FREQUENCY(10) | | |
3297 | GEN6_OFFSET(0) | | |
3298 | GEN6_AGGRESSIVE_TURBO); | |
3299 | I915_WRITE(GEN6_RC_VIDEO_FREQ, | |
3300 | GEN6_FREQUENCY(12)); | |
3301 | } | |
2b4e57bd ED |
3302 | |
3303 | I915_WRITE(GEN6_RP_DOWN_TIMEOUT, 1000000); | |
3304 | I915_WRITE(GEN6_RP_INTERRUPT_LIMITS, | |
c6a828d3 DV |
3305 | dev_priv->rps.max_delay << 24 | |
3306 | dev_priv->rps.min_delay << 16); | |
5a7dc92a | 3307 | |
1ee9ae32 DV |
3308 | I915_WRITE(GEN6_RP_UP_THRESHOLD, 59400); |
3309 | I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 245000); | |
3310 | I915_WRITE(GEN6_RP_UP_EI, 66000); | |
3311 | I915_WRITE(GEN6_RP_DOWN_EI, 350000); | |
5a7dc92a | 3312 | |
2b4e57bd ED |
3313 | I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 10); |
3314 | I915_WRITE(GEN6_RP_CONTROL, | |
3315 | GEN6_RP_MEDIA_TURBO | | |
89ba829e | 3316 | GEN6_RP_MEDIA_HW_NORMAL_MODE | |
2b4e57bd ED |
3317 | GEN6_RP_MEDIA_IS_GFX | |
3318 | GEN6_RP_ENABLE | | |
3319 | GEN6_RP_UP_BUSY_AVG | | |
5a7dc92a | 3320 | (IS_HASWELL(dev) ? GEN7_RP_DOWN_IDLE_AVG : GEN6_RP_DOWN_IDLE_CONT)); |
2b4e57bd | 3321 | |
42c0526c | 3322 | ret = sandybridge_pcode_write(dev_priv, GEN6_PCODE_WRITE_MIN_FREQ_TABLE, 0); |
988b36e5 | 3323 | if (!ret) { |
42c0526c BW |
3324 | pcu_mbox = 0; |
3325 | ret = sandybridge_pcode_read(dev_priv, GEN6_READ_OC_PARAMS, &pcu_mbox); | |
a2b3fc01 | 3326 | if (!ret && (pcu_mbox & (1<<31))) { /* OC supported */ |
10e08497 | 3327 | DRM_DEBUG_DRIVER("Overclocking supported. Max: %dMHz, Overclock max: %dMHz\n", |
a2b3fc01 BW |
3328 | (dev_priv->rps.max_delay & 0xff) * 50, |
3329 | (pcu_mbox & 0xff) * 50); | |
31c77388 | 3330 | dev_priv->rps.hw_max = pcu_mbox & 0xff; |
42c0526c BW |
3331 | } |
3332 | } else { | |
3333 | DRM_DEBUG_DRIVER("Failed to set the min frequency\n"); | |
2b4e57bd ED |
3334 | } |
3335 | ||
7b9e0ae6 | 3336 | gen6_set_rps(dev_priv->dev, (gt_perf_status & 0xff00) >> 8); |
2b4e57bd | 3337 | |
44fc7d5c | 3338 | gen6_enable_rps_interrupts(dev); |
2b4e57bd | 3339 | |
31643d54 BW |
3340 | rc6vids = 0; |
3341 | ret = sandybridge_pcode_read(dev_priv, GEN6_PCODE_READ_RC6VIDS, &rc6vids); | |
3342 | if (IS_GEN6(dev) && ret) { | |
3343 | DRM_DEBUG_DRIVER("Couldn't check for BIOS workaround\n"); | |
3344 | } else if (IS_GEN6(dev) && (GEN6_DECODE_RC6_VID(rc6vids & 0xff) < 450)) { | |
3345 | DRM_DEBUG_DRIVER("You should update your BIOS. Correcting minimum rc6 voltage (%dmV->%dmV)\n", | |
3346 | GEN6_DECODE_RC6_VID(rc6vids & 0xff), 450); | |
3347 | rc6vids &= 0xffff00; | |
3348 | rc6vids |= GEN6_ENCODE_RC6_VID(450); | |
3349 | ret = sandybridge_pcode_write(dev_priv, GEN6_PCODE_WRITE_RC6VIDS, rc6vids); | |
3350 | if (ret) | |
3351 | DRM_ERROR("Couldn't fix incorrect rc6 voltage\n"); | |
3352 | } | |
3353 | ||
2b4e57bd | 3354 | gen6_gt_force_wake_put(dev_priv); |
2b4e57bd ED |
3355 | } |
3356 | ||
79f5b2c7 | 3357 | static void gen6_update_ring_freq(struct drm_device *dev) |
2b4e57bd | 3358 | { |
79f5b2c7 | 3359 | struct drm_i915_private *dev_priv = dev->dev_private; |
2b4e57bd | 3360 | int min_freq = 15; |
3ebecd07 CW |
3361 | unsigned int gpu_freq; |
3362 | unsigned int max_ia_freq, min_ring_freq; | |
2b4e57bd ED |
3363 | int scaling_factor = 180; |
3364 | ||
4fc688ce | 3365 | WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock)); |
79f5b2c7 | 3366 | |
2b4e57bd ED |
3367 | max_ia_freq = cpufreq_quick_get_max(0); |
3368 | /* | |
3369 | * Default to measured freq if none found, PCU will ensure we don't go | |
3370 | * over | |
3371 | */ | |
3372 | if (!max_ia_freq) | |
3373 | max_ia_freq = tsc_khz; | |
3374 | ||
3375 | /* Convert from kHz to MHz */ | |
3376 | max_ia_freq /= 1000; | |
3377 | ||
3ebecd07 CW |
3378 | min_ring_freq = I915_READ(MCHBAR_MIRROR_BASE_SNB + DCLK); |
3379 | /* convert DDR frequency from units of 133.3MHz to bandwidth */ | |
3380 | min_ring_freq = (2 * 4 * min_ring_freq + 2) / 3; | |
3381 | ||
2b4e57bd ED |
3382 | /* |
3383 | * For each potential GPU frequency, load a ring frequency we'd like | |
3384 | * to use for memory access. We do this by specifying the IA frequency | |
3385 | * the PCU should use as a reference to determine the ring frequency. | |
3386 | */ | |
c6a828d3 | 3387 | for (gpu_freq = dev_priv->rps.max_delay; gpu_freq >= dev_priv->rps.min_delay; |
2b4e57bd | 3388 | gpu_freq--) { |
c6a828d3 | 3389 | int diff = dev_priv->rps.max_delay - gpu_freq; |
3ebecd07 CW |
3390 | unsigned int ia_freq = 0, ring_freq = 0; |
3391 | ||
3392 | if (IS_HASWELL(dev)) { | |
3393 | ring_freq = (gpu_freq * 5 + 3) / 4; | |
3394 | ring_freq = max(min_ring_freq, ring_freq); | |
3395 | /* leave ia_freq as the default, chosen by cpufreq */ | |
3396 | } else { | |
3397 | /* On older processors, there is no separate ring | |
3398 | * clock domain, so in order to boost the bandwidth | |
3399 | * of the ring, we need to upclock the CPU (ia_freq). | |
3400 | * | |
3401 | * For GPU frequencies less than 750MHz, | |
3402 | * just use the lowest ring freq. | |
3403 | */ | |
3404 | if (gpu_freq < min_freq) | |
3405 | ia_freq = 800; | |
3406 | else | |
3407 | ia_freq = max_ia_freq - ((diff * scaling_factor) / 2); | |
3408 | ia_freq = DIV_ROUND_CLOSEST(ia_freq, 100); | |
3409 | } | |
2b4e57bd | 3410 | |
42c0526c BW |
3411 | sandybridge_pcode_write(dev_priv, |
3412 | GEN6_PCODE_WRITE_MIN_FREQ_TABLE, | |
3ebecd07 CW |
3413 | ia_freq << GEN6_PCODE_FREQ_IA_RATIO_SHIFT | |
3414 | ring_freq << GEN6_PCODE_FREQ_RING_RATIO_SHIFT | | |
3415 | gpu_freq); | |
2b4e57bd | 3416 | } |
2b4e57bd ED |
3417 | } |
3418 | ||
0a073b84 JB |
3419 | int valleyview_rps_max_freq(struct drm_i915_private *dev_priv) |
3420 | { | |
3421 | u32 val, rp0; | |
3422 | ||
64936258 | 3423 | val = vlv_nc_read(dev_priv, IOSF_NC_FB_GFX_FREQ_FUSE); |
0a073b84 JB |
3424 | |
3425 | rp0 = (val & FB_GFX_MAX_FREQ_FUSE_MASK) >> FB_GFX_MAX_FREQ_FUSE_SHIFT; | |
3426 | /* Clamp to max */ | |
3427 | rp0 = min_t(u32, rp0, 0xea); | |
3428 | ||
3429 | return rp0; | |
3430 | } | |
3431 | ||
3432 | static int valleyview_rps_rpe_freq(struct drm_i915_private *dev_priv) | |
3433 | { | |
3434 | u32 val, rpe; | |
3435 | ||
64936258 | 3436 | val = vlv_nc_read(dev_priv, IOSF_NC_FB_GFX_FMAX_FUSE_LO); |
0a073b84 | 3437 | rpe = (val & FB_FMAX_VMIN_FREQ_LO_MASK) >> FB_FMAX_VMIN_FREQ_LO_SHIFT; |
64936258 | 3438 | val = vlv_nc_read(dev_priv, IOSF_NC_FB_GFX_FMAX_FUSE_HI); |
0a073b84 JB |
3439 | rpe |= (val & FB_FMAX_VMIN_FREQ_HI_MASK) << 5; |
3440 | ||
3441 | return rpe; | |
3442 | } | |
3443 | ||
3444 | int valleyview_rps_min_freq(struct drm_i915_private *dev_priv) | |
3445 | { | |
64936258 | 3446 | return vlv_punit_read(dev_priv, PUNIT_REG_GPU_LFM) & 0xff; |
0a073b84 JB |
3447 | } |
3448 | ||
52ceb908 JB |
3449 | static void vlv_rps_timer_work(struct work_struct *work) |
3450 | { | |
3451 | drm_i915_private_t *dev_priv = container_of(work, drm_i915_private_t, | |
3452 | rps.vlv_work.work); | |
3453 | ||
3454 | /* | |
3455 | * Timer fired, we must be idle. Drop to min voltage state. | |
3456 | * Note: we use RPe here since it should match the | |
3457 | * Vmin we were shooting for. That should give us better | |
3458 | * perf when we come back out of RC6 than if we used the | |
3459 | * min freq available. | |
3460 | */ | |
3461 | mutex_lock(&dev_priv->rps.hw_lock); | |
6dc58488 VS |
3462 | if (dev_priv->rps.cur_delay > dev_priv->rps.rpe_delay) |
3463 | valleyview_set_rps(dev_priv->dev, dev_priv->rps.rpe_delay); | |
52ceb908 JB |
3464 | mutex_unlock(&dev_priv->rps.hw_lock); |
3465 | } | |
3466 | ||
c9cddffc JB |
3467 | static void valleyview_setup_pctx(struct drm_device *dev) |
3468 | { | |
3469 | struct drm_i915_private *dev_priv = dev->dev_private; | |
3470 | struct drm_i915_gem_object *pctx; | |
3471 | unsigned long pctx_paddr; | |
3472 | u32 pcbr; | |
3473 | int pctx_size = 24*1024; | |
3474 | ||
3475 | pcbr = I915_READ(VLV_PCBR); | |
3476 | if (pcbr) { | |
3477 | /* BIOS set it up already, grab the pre-alloc'd space */ | |
3478 | int pcbr_offset; | |
3479 | ||
3480 | pcbr_offset = (pcbr & (~4095)) - dev_priv->mm.stolen_base; | |
3481 | pctx = i915_gem_object_create_stolen_for_preallocated(dev_priv->dev, | |
3482 | pcbr_offset, | |
190d6cd5 | 3483 | I915_GTT_OFFSET_NONE, |
c9cddffc JB |
3484 | pctx_size); |
3485 | goto out; | |
3486 | } | |
3487 | ||
3488 | /* | |
3489 | * From the Gunit register HAS: | |
3490 | * The Gfx driver is expected to program this register and ensure | |
3491 | * proper allocation within Gfx stolen memory. For example, this | |
3492 | * register should be programmed such than the PCBR range does not | |
3493 | * overlap with other ranges, such as the frame buffer, protected | |
3494 | * memory, or any other relevant ranges. | |
3495 | */ | |
3496 | pctx = i915_gem_object_create_stolen(dev, pctx_size); | |
3497 | if (!pctx) { | |
3498 | DRM_DEBUG("not enough stolen space for PCTX, disabling\n"); | |
3499 | return; | |
3500 | } | |
3501 | ||
3502 | pctx_paddr = dev_priv->mm.stolen_base + pctx->stolen->start; | |
3503 | I915_WRITE(VLV_PCBR, pctx_paddr); | |
3504 | ||
3505 | out: | |
3506 | dev_priv->vlv_pctx = pctx; | |
3507 | } | |
3508 | ||
0a073b84 JB |
3509 | static void valleyview_enable_rps(struct drm_device *dev) |
3510 | { | |
3511 | struct drm_i915_private *dev_priv = dev->dev_private; | |
3512 | struct intel_ring_buffer *ring; | |
73008b98 | 3513 | u32 gtfifodbg, val; |
0a073b84 JB |
3514 | int i; |
3515 | ||
3516 | WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock)); | |
3517 | ||
3518 | if ((gtfifodbg = I915_READ(GTFIFODBG))) { | |
3519 | DRM_ERROR("GT fifo had a previous error %x\n", gtfifodbg); | |
3520 | I915_WRITE(GTFIFODBG, gtfifodbg); | |
3521 | } | |
3522 | ||
c9cddffc JB |
3523 | valleyview_setup_pctx(dev); |
3524 | ||
0a073b84 JB |
3525 | gen6_gt_force_wake_get(dev_priv); |
3526 | ||
3527 | I915_WRITE(GEN6_RP_UP_THRESHOLD, 59400); | |
3528 | I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 245000); | |
3529 | I915_WRITE(GEN6_RP_UP_EI, 66000); | |
3530 | I915_WRITE(GEN6_RP_DOWN_EI, 350000); | |
3531 | ||
3532 | I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 10); | |
3533 | ||
3534 | I915_WRITE(GEN6_RP_CONTROL, | |
3535 | GEN6_RP_MEDIA_TURBO | | |
3536 | GEN6_RP_MEDIA_HW_NORMAL_MODE | | |
3537 | GEN6_RP_MEDIA_IS_GFX | | |
3538 | GEN6_RP_ENABLE | | |
3539 | GEN6_RP_UP_BUSY_AVG | | |
3540 | GEN6_RP_DOWN_IDLE_CONT); | |
3541 | ||
3542 | I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 0x00280000); | |
3543 | I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000); | |
3544 | I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25); | |
3545 | ||
3546 | for_each_ring(ring, dev_priv, i) | |
3547 | I915_WRITE(RING_MAX_IDLE(ring->mmio_base), 10); | |
3548 | ||
3549 | I915_WRITE(GEN6_RC6_THRESHOLD, 0xc350); | |
3550 | ||
3551 | /* allows RC6 residency counter to work */ | |
3552 | I915_WRITE(0x138104, _MASKED_BIT_ENABLE(0x3)); | |
3553 | I915_WRITE(GEN6_RC_CONTROL, | |
3554 | GEN7_RC_CTL_TO_MODE); | |
3555 | ||
64936258 | 3556 | val = vlv_punit_read(dev_priv, PUNIT_REG_GPU_FREQ_STS); |
2445966e JB |
3557 | switch ((val >> 6) & 3) { |
3558 | case 0: | |
3559 | case 1: | |
3560 | dev_priv->mem_freq = 800; | |
3561 | break; | |
3562 | case 2: | |
3563 | dev_priv->mem_freq = 1066; | |
3564 | break; | |
3565 | case 3: | |
3566 | dev_priv->mem_freq = 1333; | |
3567 | break; | |
3568 | } | |
0a073b84 JB |
3569 | DRM_DEBUG_DRIVER("DDR speed: %d MHz", dev_priv->mem_freq); |
3570 | ||
3571 | DRM_DEBUG_DRIVER("GPLL enabled? %s\n", val & 0x10 ? "yes" : "no"); | |
3572 | DRM_DEBUG_DRIVER("GPU status: 0x%08x\n", val); | |
3573 | ||
0a073b84 | 3574 | dev_priv->rps.cur_delay = (val >> 8) & 0xff; |
73008b98 VS |
3575 | DRM_DEBUG_DRIVER("current GPU freq: %d MHz (%u)\n", |
3576 | vlv_gpu_freq(dev_priv->mem_freq, | |
3577 | dev_priv->rps.cur_delay), | |
3578 | dev_priv->rps.cur_delay); | |
0a073b84 JB |
3579 | |
3580 | dev_priv->rps.max_delay = valleyview_rps_max_freq(dev_priv); | |
3581 | dev_priv->rps.hw_max = dev_priv->rps.max_delay; | |
73008b98 VS |
3582 | DRM_DEBUG_DRIVER("max GPU freq: %d MHz (%u)\n", |
3583 | vlv_gpu_freq(dev_priv->mem_freq, | |
3584 | dev_priv->rps.max_delay), | |
3585 | dev_priv->rps.max_delay); | |
0a073b84 | 3586 | |
73008b98 VS |
3587 | dev_priv->rps.rpe_delay = valleyview_rps_rpe_freq(dev_priv); |
3588 | DRM_DEBUG_DRIVER("RPe GPU freq: %d MHz (%u)\n", | |
3589 | vlv_gpu_freq(dev_priv->mem_freq, | |
3590 | dev_priv->rps.rpe_delay), | |
3591 | dev_priv->rps.rpe_delay); | |
0a073b84 | 3592 | |
73008b98 VS |
3593 | dev_priv->rps.min_delay = valleyview_rps_min_freq(dev_priv); |
3594 | DRM_DEBUG_DRIVER("min GPU freq: %d MHz (%u)\n", | |
3595 | vlv_gpu_freq(dev_priv->mem_freq, | |
3596 | dev_priv->rps.min_delay), | |
3597 | dev_priv->rps.min_delay); | |
0a073b84 | 3598 | |
73008b98 VS |
3599 | DRM_DEBUG_DRIVER("setting GPU freq to %d MHz (%u)\n", |
3600 | vlv_gpu_freq(dev_priv->mem_freq, | |
3601 | dev_priv->rps.rpe_delay), | |
3602 | dev_priv->rps.rpe_delay); | |
0a073b84 | 3603 | |
52ceb908 JB |
3604 | INIT_DELAYED_WORK(&dev_priv->rps.vlv_work, vlv_rps_timer_work); |
3605 | ||
73008b98 | 3606 | valleyview_set_rps(dev_priv->dev, dev_priv->rps.rpe_delay); |
0a073b84 | 3607 | |
44fc7d5c | 3608 | gen6_enable_rps_interrupts(dev); |
0a073b84 JB |
3609 | |
3610 | gen6_gt_force_wake_put(dev_priv); | |
3611 | } | |
3612 | ||
930ebb46 | 3613 | void ironlake_teardown_rc6(struct drm_device *dev) |
2b4e57bd ED |
3614 | { |
3615 | struct drm_i915_private *dev_priv = dev->dev_private; | |
3616 | ||
3e373948 DV |
3617 | if (dev_priv->ips.renderctx) { |
3618 | i915_gem_object_unpin(dev_priv->ips.renderctx); | |
3619 | drm_gem_object_unreference(&dev_priv->ips.renderctx->base); | |
3620 | dev_priv->ips.renderctx = NULL; | |
2b4e57bd ED |
3621 | } |
3622 | ||
3e373948 DV |
3623 | if (dev_priv->ips.pwrctx) { |
3624 | i915_gem_object_unpin(dev_priv->ips.pwrctx); | |
3625 | drm_gem_object_unreference(&dev_priv->ips.pwrctx->base); | |
3626 | dev_priv->ips.pwrctx = NULL; | |
2b4e57bd ED |
3627 | } |
3628 | } | |
3629 | ||
930ebb46 | 3630 | static void ironlake_disable_rc6(struct drm_device *dev) |
2b4e57bd ED |
3631 | { |
3632 | struct drm_i915_private *dev_priv = dev->dev_private; | |
3633 | ||
3634 | if (I915_READ(PWRCTXA)) { | |
3635 | /* Wake the GPU, prevent RC6, then restore RSTDBYCTL */ | |
3636 | I915_WRITE(RSTDBYCTL, I915_READ(RSTDBYCTL) | RCX_SW_EXIT); | |
3637 | wait_for(((I915_READ(RSTDBYCTL) & RSX_STATUS_MASK) == RSX_STATUS_ON), | |
3638 | 50); | |
3639 | ||
3640 | I915_WRITE(PWRCTXA, 0); | |
3641 | POSTING_READ(PWRCTXA); | |
3642 | ||
3643 | I915_WRITE(RSTDBYCTL, I915_READ(RSTDBYCTL) & ~RCX_SW_EXIT); | |
3644 | POSTING_READ(RSTDBYCTL); | |
3645 | } | |
2b4e57bd ED |
3646 | } |
3647 | ||
3648 | static int ironlake_setup_rc6(struct drm_device *dev) | |
3649 | { | |
3650 | struct drm_i915_private *dev_priv = dev->dev_private; | |
3651 | ||
3e373948 DV |
3652 | if (dev_priv->ips.renderctx == NULL) |
3653 | dev_priv->ips.renderctx = intel_alloc_context_page(dev); | |
3654 | if (!dev_priv->ips.renderctx) | |
2b4e57bd ED |
3655 | return -ENOMEM; |
3656 | ||
3e373948 DV |
3657 | if (dev_priv->ips.pwrctx == NULL) |
3658 | dev_priv->ips.pwrctx = intel_alloc_context_page(dev); | |
3659 | if (!dev_priv->ips.pwrctx) { | |
2b4e57bd ED |
3660 | ironlake_teardown_rc6(dev); |
3661 | return -ENOMEM; | |
3662 | } | |
3663 | ||
3664 | return 0; | |
3665 | } | |
3666 | ||
930ebb46 | 3667 | static void ironlake_enable_rc6(struct drm_device *dev) |
2b4e57bd ED |
3668 | { |
3669 | struct drm_i915_private *dev_priv = dev->dev_private; | |
6d90c952 | 3670 | struct intel_ring_buffer *ring = &dev_priv->ring[RCS]; |
3e960501 | 3671 | bool was_interruptible; |
2b4e57bd ED |
3672 | int ret; |
3673 | ||
3674 | /* rc6 disabled by default due to repeated reports of hanging during | |
3675 | * boot and resume. | |
3676 | */ | |
3677 | if (!intel_enable_rc6(dev)) | |
3678 | return; | |
3679 | ||
79f5b2c7 DV |
3680 | WARN_ON(!mutex_is_locked(&dev->struct_mutex)); |
3681 | ||
2b4e57bd | 3682 | ret = ironlake_setup_rc6(dev); |
79f5b2c7 | 3683 | if (ret) |
2b4e57bd | 3684 | return; |
2b4e57bd | 3685 | |
3e960501 CW |
3686 | was_interruptible = dev_priv->mm.interruptible; |
3687 | dev_priv->mm.interruptible = false; | |
3688 | ||
2b4e57bd ED |
3689 | /* |
3690 | * GPU can automatically power down the render unit if given a page | |
3691 | * to save state. | |
3692 | */ | |
6d90c952 | 3693 | ret = intel_ring_begin(ring, 6); |
2b4e57bd ED |
3694 | if (ret) { |
3695 | ironlake_teardown_rc6(dev); | |
3e960501 | 3696 | dev_priv->mm.interruptible = was_interruptible; |
2b4e57bd ED |
3697 | return; |
3698 | } | |
3699 | ||
6d90c952 DV |
3700 | intel_ring_emit(ring, MI_SUSPEND_FLUSH | MI_SUSPEND_FLUSH_EN); |
3701 | intel_ring_emit(ring, MI_SET_CONTEXT); | |
f343c5f6 | 3702 | intel_ring_emit(ring, i915_gem_obj_ggtt_offset(dev_priv->ips.renderctx) | |
6d90c952 DV |
3703 | MI_MM_SPACE_GTT | |
3704 | MI_SAVE_EXT_STATE_EN | | |
3705 | MI_RESTORE_EXT_STATE_EN | | |
3706 | MI_RESTORE_INHIBIT); | |
3707 | intel_ring_emit(ring, MI_SUSPEND_FLUSH); | |
3708 | intel_ring_emit(ring, MI_NOOP); | |
3709 | intel_ring_emit(ring, MI_FLUSH); | |
3710 | intel_ring_advance(ring); | |
2b4e57bd ED |
3711 | |
3712 | /* | |
3713 | * Wait for the command parser to advance past MI_SET_CONTEXT. The HW | |
3714 | * does an implicit flush, combined with MI_FLUSH above, it should be | |
3715 | * safe to assume that renderctx is valid | |
3716 | */ | |
3e960501 CW |
3717 | ret = intel_ring_idle(ring); |
3718 | dev_priv->mm.interruptible = was_interruptible; | |
2b4e57bd | 3719 | if (ret) { |
def27a58 | 3720 | DRM_ERROR("failed to enable ironlake power savings\n"); |
2b4e57bd | 3721 | ironlake_teardown_rc6(dev); |
2b4e57bd ED |
3722 | return; |
3723 | } | |
3724 | ||
f343c5f6 | 3725 | I915_WRITE(PWRCTXA, i915_gem_obj_ggtt_offset(dev_priv->ips.pwrctx) | PWRCTX_EN); |
2b4e57bd | 3726 | I915_WRITE(RSTDBYCTL, I915_READ(RSTDBYCTL) & ~RCX_SW_EXIT); |
2b4e57bd ED |
3727 | } |
3728 | ||
dde18883 ED |
3729 | static unsigned long intel_pxfreq(u32 vidfreq) |
3730 | { | |
3731 | unsigned long freq; | |
3732 | int div = (vidfreq & 0x3f0000) >> 16; | |
3733 | int post = (vidfreq & 0x3000) >> 12; | |
3734 | int pre = (vidfreq & 0x7); | |
3735 | ||
3736 | if (!pre) | |
3737 | return 0; | |
3738 | ||
3739 | freq = ((div * 133333) / ((1<<post) * pre)); | |
3740 | ||
3741 | return freq; | |
3742 | } | |
3743 | ||
eb48eb00 DV |
3744 | static const struct cparams { |
3745 | u16 i; | |
3746 | u16 t; | |
3747 | u16 m; | |
3748 | u16 c; | |
3749 | } cparams[] = { | |
3750 | { 1, 1333, 301, 28664 }, | |
3751 | { 1, 1066, 294, 24460 }, | |
3752 | { 1, 800, 294, 25192 }, | |
3753 | { 0, 1333, 276, 27605 }, | |
3754 | { 0, 1066, 276, 27605 }, | |
3755 | { 0, 800, 231, 23784 }, | |
3756 | }; | |
3757 | ||
f531dcb2 | 3758 | static unsigned long __i915_chipset_val(struct drm_i915_private *dev_priv) |
eb48eb00 DV |
3759 | { |
3760 | u64 total_count, diff, ret; | |
3761 | u32 count1, count2, count3, m = 0, c = 0; | |
3762 | unsigned long now = jiffies_to_msecs(jiffies), diff1; | |
3763 | int i; | |
3764 | ||
02d71956 DV |
3765 | assert_spin_locked(&mchdev_lock); |
3766 | ||
20e4d407 | 3767 | diff1 = now - dev_priv->ips.last_time1; |
eb48eb00 DV |
3768 | |
3769 | /* Prevent division-by-zero if we are asking too fast. | |
3770 | * Also, we don't get interesting results if we are polling | |
3771 | * faster than once in 10ms, so just return the saved value | |
3772 | * in such cases. | |
3773 | */ | |
3774 | if (diff1 <= 10) | |
20e4d407 | 3775 | return dev_priv->ips.chipset_power; |
eb48eb00 DV |
3776 | |
3777 | count1 = I915_READ(DMIEC); | |
3778 | count2 = I915_READ(DDREC); | |
3779 | count3 = I915_READ(CSIEC); | |
3780 | ||
3781 | total_count = count1 + count2 + count3; | |
3782 | ||
3783 | /* FIXME: handle per-counter overflow */ | |
20e4d407 DV |
3784 | if (total_count < dev_priv->ips.last_count1) { |
3785 | diff = ~0UL - dev_priv->ips.last_count1; | |
eb48eb00 DV |
3786 | diff += total_count; |
3787 | } else { | |
20e4d407 | 3788 | diff = total_count - dev_priv->ips.last_count1; |
eb48eb00 DV |
3789 | } |
3790 | ||
3791 | for (i = 0; i < ARRAY_SIZE(cparams); i++) { | |
20e4d407 DV |
3792 | if (cparams[i].i == dev_priv->ips.c_m && |
3793 | cparams[i].t == dev_priv->ips.r_t) { | |
eb48eb00 DV |
3794 | m = cparams[i].m; |
3795 | c = cparams[i].c; | |
3796 | break; | |
3797 | } | |
3798 | } | |
3799 | ||
3800 | diff = div_u64(diff, diff1); | |
3801 | ret = ((m * diff) + c); | |
3802 | ret = div_u64(ret, 10); | |
3803 | ||
20e4d407 DV |
3804 | dev_priv->ips.last_count1 = total_count; |
3805 | dev_priv->ips.last_time1 = now; | |
eb48eb00 | 3806 | |
20e4d407 | 3807 | dev_priv->ips.chipset_power = ret; |
eb48eb00 DV |
3808 | |
3809 | return ret; | |
3810 | } | |
3811 | ||
f531dcb2 CW |
3812 | unsigned long i915_chipset_val(struct drm_i915_private *dev_priv) |
3813 | { | |
3814 | unsigned long val; | |
3815 | ||
3816 | if (dev_priv->info->gen != 5) | |
3817 | return 0; | |
3818 | ||
3819 | spin_lock_irq(&mchdev_lock); | |
3820 | ||
3821 | val = __i915_chipset_val(dev_priv); | |
3822 | ||
3823 | spin_unlock_irq(&mchdev_lock); | |
3824 | ||
3825 | return val; | |
3826 | } | |
3827 | ||
eb48eb00 DV |
3828 | unsigned long i915_mch_val(struct drm_i915_private *dev_priv) |
3829 | { | |
3830 | unsigned long m, x, b; | |
3831 | u32 tsfs; | |
3832 | ||
3833 | tsfs = I915_READ(TSFS); | |
3834 | ||
3835 | m = ((tsfs & TSFS_SLOPE_MASK) >> TSFS_SLOPE_SHIFT); | |
3836 | x = I915_READ8(TR1); | |
3837 | ||
3838 | b = tsfs & TSFS_INTR_MASK; | |
3839 | ||
3840 | return ((m * x) / 127) - b; | |
3841 | } | |
3842 | ||
3843 | static u16 pvid_to_extvid(struct drm_i915_private *dev_priv, u8 pxvid) | |
3844 | { | |
3845 | static const struct v_table { | |
3846 | u16 vd; /* in .1 mil */ | |
3847 | u16 vm; /* in .1 mil */ | |
3848 | } v_table[] = { | |
3849 | { 0, 0, }, | |
3850 | { 375, 0, }, | |
3851 | { 500, 0, }, | |
3852 | { 625, 0, }, | |
3853 | { 750, 0, }, | |
3854 | { 875, 0, }, | |
3855 | { 1000, 0, }, | |
3856 | { 1125, 0, }, | |
3857 | { 4125, 3000, }, | |
3858 | { 4125, 3000, }, | |
3859 | { 4125, 3000, }, | |
3860 | { 4125, 3000, }, | |
3861 | { 4125, 3000, }, | |
3862 | { 4125, 3000, }, | |
3863 | { 4125, 3000, }, | |
3864 | { 4125, 3000, }, | |
3865 | { 4125, 3000, }, | |
3866 | { 4125, 3000, }, | |
3867 | { 4125, 3000, }, | |
3868 | { 4125, 3000, }, | |
3869 | { 4125, 3000, }, | |
3870 | { 4125, 3000, }, | |
3871 | { 4125, 3000, }, | |
3872 | { 4125, 3000, }, | |
3873 | { 4125, 3000, }, | |
3874 | { 4125, 3000, }, | |
3875 | { 4125, 3000, }, | |
3876 | { 4125, 3000, }, | |
3877 | { 4125, 3000, }, | |
3878 | { 4125, 3000, }, | |
3879 | { 4125, 3000, }, | |
3880 | { 4125, 3000, }, | |
3881 | { 4250, 3125, }, | |
3882 | { 4375, 3250, }, | |
3883 | { 4500, 3375, }, | |
3884 | { 4625, 3500, }, | |
3885 | { 4750, 3625, }, | |
3886 | { 4875, 3750, }, | |
3887 | { 5000, 3875, }, | |
3888 | { 5125, 4000, }, | |
3889 | { 5250, 4125, }, | |
3890 | { 5375, 4250, }, | |
3891 | { 5500, 4375, }, | |
3892 | { 5625, 4500, }, | |
3893 | { 5750, 4625, }, | |
3894 | { 5875, 4750, }, | |
3895 | { 6000, 4875, }, | |
3896 | { 6125, 5000, }, | |
3897 | { 6250, 5125, }, | |
3898 | { 6375, 5250, }, | |
3899 | { 6500, 5375, }, | |
3900 | { 6625, 5500, }, | |
3901 | { 6750, 5625, }, | |
3902 | { 6875, 5750, }, | |
3903 | { 7000, 5875, }, | |
3904 | { 7125, 6000, }, | |
3905 | { 7250, 6125, }, | |
3906 | { 7375, 6250, }, | |
3907 | { 7500, 6375, }, | |
3908 | { 7625, 6500, }, | |
3909 | { 7750, 6625, }, | |
3910 | { 7875, 6750, }, | |
3911 | { 8000, 6875, }, | |
3912 | { 8125, 7000, }, | |
3913 | { 8250, 7125, }, | |
3914 | { 8375, 7250, }, | |
3915 | { 8500, 7375, }, | |
3916 | { 8625, 7500, }, | |
3917 | { 8750, 7625, }, | |
3918 | { 8875, 7750, }, | |
3919 | { 9000, 7875, }, | |
3920 | { 9125, 8000, }, | |
3921 | { 9250, 8125, }, | |
3922 | { 9375, 8250, }, | |
3923 | { 9500, 8375, }, | |
3924 | { 9625, 8500, }, | |
3925 | { 9750, 8625, }, | |
3926 | { 9875, 8750, }, | |
3927 | { 10000, 8875, }, | |
3928 | { 10125, 9000, }, | |
3929 | { 10250, 9125, }, | |
3930 | { 10375, 9250, }, | |
3931 | { 10500, 9375, }, | |
3932 | { 10625, 9500, }, | |
3933 | { 10750, 9625, }, | |
3934 | { 10875, 9750, }, | |
3935 | { 11000, 9875, }, | |
3936 | { 11125, 10000, }, | |
3937 | { 11250, 10125, }, | |
3938 | { 11375, 10250, }, | |
3939 | { 11500, 10375, }, | |
3940 | { 11625, 10500, }, | |
3941 | { 11750, 10625, }, | |
3942 | { 11875, 10750, }, | |
3943 | { 12000, 10875, }, | |
3944 | { 12125, 11000, }, | |
3945 | { 12250, 11125, }, | |
3946 | { 12375, 11250, }, | |
3947 | { 12500, 11375, }, | |
3948 | { 12625, 11500, }, | |
3949 | { 12750, 11625, }, | |
3950 | { 12875, 11750, }, | |
3951 | { 13000, 11875, }, | |
3952 | { 13125, 12000, }, | |
3953 | { 13250, 12125, }, | |
3954 | { 13375, 12250, }, | |
3955 | { 13500, 12375, }, | |
3956 | { 13625, 12500, }, | |
3957 | { 13750, 12625, }, | |
3958 | { 13875, 12750, }, | |
3959 | { 14000, 12875, }, | |
3960 | { 14125, 13000, }, | |
3961 | { 14250, 13125, }, | |
3962 | { 14375, 13250, }, | |
3963 | { 14500, 13375, }, | |
3964 | { 14625, 13500, }, | |
3965 | { 14750, 13625, }, | |
3966 | { 14875, 13750, }, | |
3967 | { 15000, 13875, }, | |
3968 | { 15125, 14000, }, | |
3969 | { 15250, 14125, }, | |
3970 | { 15375, 14250, }, | |
3971 | { 15500, 14375, }, | |
3972 | { 15625, 14500, }, | |
3973 | { 15750, 14625, }, | |
3974 | { 15875, 14750, }, | |
3975 | { 16000, 14875, }, | |
3976 | { 16125, 15000, }, | |
3977 | }; | |
3978 | if (dev_priv->info->is_mobile) | |
3979 | return v_table[pxvid].vm; | |
3980 | else | |
3981 | return v_table[pxvid].vd; | |
3982 | } | |
3983 | ||
02d71956 | 3984 | static void __i915_update_gfx_val(struct drm_i915_private *dev_priv) |
eb48eb00 DV |
3985 | { |
3986 | struct timespec now, diff1; | |
3987 | u64 diff; | |
3988 | unsigned long diffms; | |
3989 | u32 count; | |
3990 | ||
02d71956 | 3991 | assert_spin_locked(&mchdev_lock); |
eb48eb00 DV |
3992 | |
3993 | getrawmonotonic(&now); | |
20e4d407 | 3994 | diff1 = timespec_sub(now, dev_priv->ips.last_time2); |
eb48eb00 DV |
3995 | |
3996 | /* Don't divide by 0 */ | |
3997 | diffms = diff1.tv_sec * 1000 + diff1.tv_nsec / 1000000; | |
3998 | if (!diffms) | |
3999 | return; | |
4000 | ||
4001 | count = I915_READ(GFXEC); | |
4002 | ||
20e4d407 DV |
4003 | if (count < dev_priv->ips.last_count2) { |
4004 | diff = ~0UL - dev_priv->ips.last_count2; | |
eb48eb00 DV |
4005 | diff += count; |
4006 | } else { | |
20e4d407 | 4007 | diff = count - dev_priv->ips.last_count2; |
eb48eb00 DV |
4008 | } |
4009 | ||
20e4d407 DV |
4010 | dev_priv->ips.last_count2 = count; |
4011 | dev_priv->ips.last_time2 = now; | |
eb48eb00 DV |
4012 | |
4013 | /* More magic constants... */ | |
4014 | diff = diff * 1181; | |
4015 | diff = div_u64(diff, diffms * 10); | |
20e4d407 | 4016 | dev_priv->ips.gfx_power = diff; |
eb48eb00 DV |
4017 | } |
4018 | ||
02d71956 DV |
4019 | void i915_update_gfx_val(struct drm_i915_private *dev_priv) |
4020 | { | |
4021 | if (dev_priv->info->gen != 5) | |
4022 | return; | |
4023 | ||
9270388e | 4024 | spin_lock_irq(&mchdev_lock); |
02d71956 DV |
4025 | |
4026 | __i915_update_gfx_val(dev_priv); | |
4027 | ||
9270388e | 4028 | spin_unlock_irq(&mchdev_lock); |
02d71956 DV |
4029 | } |
4030 | ||
f531dcb2 | 4031 | static unsigned long __i915_gfx_val(struct drm_i915_private *dev_priv) |
eb48eb00 DV |
4032 | { |
4033 | unsigned long t, corr, state1, corr2, state2; | |
4034 | u32 pxvid, ext_v; | |
4035 | ||
02d71956 DV |
4036 | assert_spin_locked(&mchdev_lock); |
4037 | ||
c6a828d3 | 4038 | pxvid = I915_READ(PXVFREQ_BASE + (dev_priv->rps.cur_delay * 4)); |
eb48eb00 DV |
4039 | pxvid = (pxvid >> 24) & 0x7f; |
4040 | ext_v = pvid_to_extvid(dev_priv, pxvid); | |
4041 | ||
4042 | state1 = ext_v; | |
4043 | ||
4044 | t = i915_mch_val(dev_priv); | |
4045 | ||
4046 | /* Revel in the empirically derived constants */ | |
4047 | ||
4048 | /* Correction factor in 1/100000 units */ | |
4049 | if (t > 80) | |
4050 | corr = ((t * 2349) + 135940); | |
4051 | else if (t >= 50) | |
4052 | corr = ((t * 964) + 29317); | |
4053 | else /* < 50 */ | |
4054 | corr = ((t * 301) + 1004); | |
4055 | ||
4056 | corr = corr * ((150142 * state1) / 10000 - 78642); | |
4057 | corr /= 100000; | |
20e4d407 | 4058 | corr2 = (corr * dev_priv->ips.corr); |
eb48eb00 DV |
4059 | |
4060 | state2 = (corr2 * state1) / 10000; | |
4061 | state2 /= 100; /* convert to mW */ | |
4062 | ||
02d71956 | 4063 | __i915_update_gfx_val(dev_priv); |
eb48eb00 | 4064 | |
20e4d407 | 4065 | return dev_priv->ips.gfx_power + state2; |
eb48eb00 DV |
4066 | } |
4067 | ||
f531dcb2 CW |
4068 | unsigned long i915_gfx_val(struct drm_i915_private *dev_priv) |
4069 | { | |
4070 | unsigned long val; | |
4071 | ||
4072 | if (dev_priv->info->gen != 5) | |
4073 | return 0; | |
4074 | ||
4075 | spin_lock_irq(&mchdev_lock); | |
4076 | ||
4077 | val = __i915_gfx_val(dev_priv); | |
4078 | ||
4079 | spin_unlock_irq(&mchdev_lock); | |
4080 | ||
4081 | return val; | |
4082 | } | |
4083 | ||
eb48eb00 DV |
4084 | /** |
4085 | * i915_read_mch_val - return value for IPS use | |
4086 | * | |
4087 | * Calculate and return a value for the IPS driver to use when deciding whether | |
4088 | * we have thermal and power headroom to increase CPU or GPU power budget. | |
4089 | */ | |
4090 | unsigned long i915_read_mch_val(void) | |
4091 | { | |
4092 | struct drm_i915_private *dev_priv; | |
4093 | unsigned long chipset_val, graphics_val, ret = 0; | |
4094 | ||
9270388e | 4095 | spin_lock_irq(&mchdev_lock); |
eb48eb00 DV |
4096 | if (!i915_mch_dev) |
4097 | goto out_unlock; | |
4098 | dev_priv = i915_mch_dev; | |
4099 | ||
f531dcb2 CW |
4100 | chipset_val = __i915_chipset_val(dev_priv); |
4101 | graphics_val = __i915_gfx_val(dev_priv); | |
eb48eb00 DV |
4102 | |
4103 | ret = chipset_val + graphics_val; | |
4104 | ||
4105 | out_unlock: | |
9270388e | 4106 | spin_unlock_irq(&mchdev_lock); |
eb48eb00 DV |
4107 | |
4108 | return ret; | |
4109 | } | |
4110 | EXPORT_SYMBOL_GPL(i915_read_mch_val); | |
4111 | ||
4112 | /** | |
4113 | * i915_gpu_raise - raise GPU frequency limit | |
4114 | * | |
4115 | * Raise the limit; IPS indicates we have thermal headroom. | |
4116 | */ | |
4117 | bool i915_gpu_raise(void) | |
4118 | { | |
4119 | struct drm_i915_private *dev_priv; | |
4120 | bool ret = true; | |
4121 | ||
9270388e | 4122 | spin_lock_irq(&mchdev_lock); |
eb48eb00 DV |
4123 | if (!i915_mch_dev) { |
4124 | ret = false; | |
4125 | goto out_unlock; | |
4126 | } | |
4127 | dev_priv = i915_mch_dev; | |
4128 | ||
20e4d407 DV |
4129 | if (dev_priv->ips.max_delay > dev_priv->ips.fmax) |
4130 | dev_priv->ips.max_delay--; | |
eb48eb00 DV |
4131 | |
4132 | out_unlock: | |
9270388e | 4133 | spin_unlock_irq(&mchdev_lock); |
eb48eb00 DV |
4134 | |
4135 | return ret; | |
4136 | } | |
4137 | EXPORT_SYMBOL_GPL(i915_gpu_raise); | |
4138 | ||
4139 | /** | |
4140 | * i915_gpu_lower - lower GPU frequency limit | |
4141 | * | |
4142 | * IPS indicates we're close to a thermal limit, so throttle back the GPU | |
4143 | * frequency maximum. | |
4144 | */ | |
4145 | bool i915_gpu_lower(void) | |
4146 | { | |
4147 | struct drm_i915_private *dev_priv; | |
4148 | bool ret = true; | |
4149 | ||
9270388e | 4150 | spin_lock_irq(&mchdev_lock); |
eb48eb00 DV |
4151 | if (!i915_mch_dev) { |
4152 | ret = false; | |
4153 | goto out_unlock; | |
4154 | } | |
4155 | dev_priv = i915_mch_dev; | |
4156 | ||
20e4d407 DV |
4157 | if (dev_priv->ips.max_delay < dev_priv->ips.min_delay) |
4158 | dev_priv->ips.max_delay++; | |
eb48eb00 DV |
4159 | |
4160 | out_unlock: | |
9270388e | 4161 | spin_unlock_irq(&mchdev_lock); |
eb48eb00 DV |
4162 | |
4163 | return ret; | |
4164 | } | |
4165 | EXPORT_SYMBOL_GPL(i915_gpu_lower); | |
4166 | ||
4167 | /** | |
4168 | * i915_gpu_busy - indicate GPU business to IPS | |
4169 | * | |
4170 | * Tell the IPS driver whether or not the GPU is busy. | |
4171 | */ | |
4172 | bool i915_gpu_busy(void) | |
4173 | { | |
4174 | struct drm_i915_private *dev_priv; | |
f047e395 | 4175 | struct intel_ring_buffer *ring; |
eb48eb00 | 4176 | bool ret = false; |
f047e395 | 4177 | int i; |
eb48eb00 | 4178 | |
9270388e | 4179 | spin_lock_irq(&mchdev_lock); |
eb48eb00 DV |
4180 | if (!i915_mch_dev) |
4181 | goto out_unlock; | |
4182 | dev_priv = i915_mch_dev; | |
4183 | ||
f047e395 CW |
4184 | for_each_ring(ring, dev_priv, i) |
4185 | ret |= !list_empty(&ring->request_list); | |
eb48eb00 DV |
4186 | |
4187 | out_unlock: | |
9270388e | 4188 | spin_unlock_irq(&mchdev_lock); |
eb48eb00 DV |
4189 | |
4190 | return ret; | |
4191 | } | |
4192 | EXPORT_SYMBOL_GPL(i915_gpu_busy); | |
4193 | ||
4194 | /** | |
4195 | * i915_gpu_turbo_disable - disable graphics turbo | |
4196 | * | |
4197 | * Disable graphics turbo by resetting the max frequency and setting the | |
4198 | * current frequency to the default. | |
4199 | */ | |
4200 | bool i915_gpu_turbo_disable(void) | |
4201 | { | |
4202 | struct drm_i915_private *dev_priv; | |
4203 | bool ret = true; | |
4204 | ||
9270388e | 4205 | spin_lock_irq(&mchdev_lock); |
eb48eb00 DV |
4206 | if (!i915_mch_dev) { |
4207 | ret = false; | |
4208 | goto out_unlock; | |
4209 | } | |
4210 | dev_priv = i915_mch_dev; | |
4211 | ||
20e4d407 | 4212 | dev_priv->ips.max_delay = dev_priv->ips.fstart; |
eb48eb00 | 4213 | |
20e4d407 | 4214 | if (!ironlake_set_drps(dev_priv->dev, dev_priv->ips.fstart)) |
eb48eb00 DV |
4215 | ret = false; |
4216 | ||
4217 | out_unlock: | |
9270388e | 4218 | spin_unlock_irq(&mchdev_lock); |
eb48eb00 DV |
4219 | |
4220 | return ret; | |
4221 | } | |
4222 | EXPORT_SYMBOL_GPL(i915_gpu_turbo_disable); | |
4223 | ||
4224 | /** | |
4225 | * Tells the intel_ips driver that the i915 driver is now loaded, if | |
4226 | * IPS got loaded first. | |
4227 | * | |
4228 | * This awkward dance is so that neither module has to depend on the | |
4229 | * other in order for IPS to do the appropriate communication of | |
4230 | * GPU turbo limits to i915. | |
4231 | */ | |
4232 | static void | |
4233 | ips_ping_for_i915_load(void) | |
4234 | { | |
4235 | void (*link)(void); | |
4236 | ||
4237 | link = symbol_get(ips_link_to_i915_driver); | |
4238 | if (link) { | |
4239 | link(); | |
4240 | symbol_put(ips_link_to_i915_driver); | |
4241 | } | |
4242 | } | |
4243 | ||
4244 | void intel_gpu_ips_init(struct drm_i915_private *dev_priv) | |
4245 | { | |
02d71956 DV |
4246 | /* We only register the i915 ips part with intel-ips once everything is |
4247 | * set up, to avoid intel-ips sneaking in and reading bogus values. */ | |
9270388e | 4248 | spin_lock_irq(&mchdev_lock); |
eb48eb00 | 4249 | i915_mch_dev = dev_priv; |
9270388e | 4250 | spin_unlock_irq(&mchdev_lock); |
eb48eb00 DV |
4251 | |
4252 | ips_ping_for_i915_load(); | |
4253 | } | |
4254 | ||
4255 | void intel_gpu_ips_teardown(void) | |
4256 | { | |
9270388e | 4257 | spin_lock_irq(&mchdev_lock); |
eb48eb00 | 4258 | i915_mch_dev = NULL; |
9270388e | 4259 | spin_unlock_irq(&mchdev_lock); |
eb48eb00 | 4260 | } |
8090c6b9 | 4261 | static void intel_init_emon(struct drm_device *dev) |
dde18883 ED |
4262 | { |
4263 | struct drm_i915_private *dev_priv = dev->dev_private; | |
4264 | u32 lcfuse; | |
4265 | u8 pxw[16]; | |
4266 | int i; | |
4267 | ||
4268 | /* Disable to program */ | |
4269 | I915_WRITE(ECR, 0); | |
4270 | POSTING_READ(ECR); | |
4271 | ||
4272 | /* Program energy weights for various events */ | |
4273 | I915_WRITE(SDEW, 0x15040d00); | |
4274 | I915_WRITE(CSIEW0, 0x007f0000); | |
4275 | I915_WRITE(CSIEW1, 0x1e220004); | |
4276 | I915_WRITE(CSIEW2, 0x04000004); | |
4277 | ||
4278 | for (i = 0; i < 5; i++) | |
4279 | I915_WRITE(PEW + (i * 4), 0); | |
4280 | for (i = 0; i < 3; i++) | |
4281 | I915_WRITE(DEW + (i * 4), 0); | |
4282 | ||
4283 | /* Program P-state weights to account for frequency power adjustment */ | |
4284 | for (i = 0; i < 16; i++) { | |
4285 | u32 pxvidfreq = I915_READ(PXVFREQ_BASE + (i * 4)); | |
4286 | unsigned long freq = intel_pxfreq(pxvidfreq); | |
4287 | unsigned long vid = (pxvidfreq & PXVFREQ_PX_MASK) >> | |
4288 | PXVFREQ_PX_SHIFT; | |
4289 | unsigned long val; | |
4290 | ||
4291 | val = vid * vid; | |
4292 | val *= (freq / 1000); | |
4293 | val *= 255; | |
4294 | val /= (127*127*900); | |
4295 | if (val > 0xff) | |
4296 | DRM_ERROR("bad pxval: %ld\n", val); | |
4297 | pxw[i] = val; | |
4298 | } | |
4299 | /* Render standby states get 0 weight */ | |
4300 | pxw[14] = 0; | |
4301 | pxw[15] = 0; | |
4302 | ||
4303 | for (i = 0; i < 4; i++) { | |
4304 | u32 val = (pxw[i*4] << 24) | (pxw[(i*4)+1] << 16) | | |
4305 | (pxw[(i*4)+2] << 8) | (pxw[(i*4)+3]); | |
4306 | I915_WRITE(PXW + (i * 4), val); | |
4307 | } | |
4308 | ||
4309 | /* Adjust magic regs to magic values (more experimental results) */ | |
4310 | I915_WRITE(OGW0, 0); | |
4311 | I915_WRITE(OGW1, 0); | |
4312 | I915_WRITE(EG0, 0x00007f00); | |
4313 | I915_WRITE(EG1, 0x0000000e); | |
4314 | I915_WRITE(EG2, 0x000e0000); | |
4315 | I915_WRITE(EG3, 0x68000300); | |
4316 | I915_WRITE(EG4, 0x42000000); | |
4317 | I915_WRITE(EG5, 0x00140031); | |
4318 | I915_WRITE(EG6, 0); | |
4319 | I915_WRITE(EG7, 0); | |
4320 | ||
4321 | for (i = 0; i < 8; i++) | |
4322 | I915_WRITE(PXWL + (i * 4), 0); | |
4323 | ||
4324 | /* Enable PMON + select events */ | |
4325 | I915_WRITE(ECR, 0x80000019); | |
4326 | ||
4327 | lcfuse = I915_READ(LCFUSE02); | |
4328 | ||
20e4d407 | 4329 | dev_priv->ips.corr = (lcfuse & LCFUSE_HIV_MASK); |
dde18883 ED |
4330 | } |
4331 | ||
8090c6b9 DV |
4332 | void intel_disable_gt_powersave(struct drm_device *dev) |
4333 | { | |
1a01ab3b JB |
4334 | struct drm_i915_private *dev_priv = dev->dev_private; |
4335 | ||
fd0c0642 DV |
4336 | /* Interrupts should be disabled already to avoid re-arming. */ |
4337 | WARN_ON(dev->irq_enabled); | |
4338 | ||
930ebb46 | 4339 | if (IS_IRONLAKE_M(dev)) { |
8090c6b9 | 4340 | ironlake_disable_drps(dev); |
930ebb46 | 4341 | ironlake_disable_rc6(dev); |
0a073b84 | 4342 | } else if (INTEL_INFO(dev)->gen >= 6) { |
1a01ab3b | 4343 | cancel_delayed_work_sync(&dev_priv->rps.delayed_resume_work); |
250848ca | 4344 | cancel_work_sync(&dev_priv->rps.work); |
52ceb908 JB |
4345 | if (IS_VALLEYVIEW(dev)) |
4346 | cancel_delayed_work_sync(&dev_priv->rps.vlv_work); | |
4fc688ce | 4347 | mutex_lock(&dev_priv->rps.hw_lock); |
d20d4f0c JB |
4348 | if (IS_VALLEYVIEW(dev)) |
4349 | valleyview_disable_rps(dev); | |
4350 | else | |
4351 | gen6_disable_rps(dev); | |
4fc688ce | 4352 | mutex_unlock(&dev_priv->rps.hw_lock); |
930ebb46 | 4353 | } |
8090c6b9 DV |
4354 | } |
4355 | ||
1a01ab3b JB |
4356 | static void intel_gen6_powersave_work(struct work_struct *work) |
4357 | { | |
4358 | struct drm_i915_private *dev_priv = | |
4359 | container_of(work, struct drm_i915_private, | |
4360 | rps.delayed_resume_work.work); | |
4361 | struct drm_device *dev = dev_priv->dev; | |
4362 | ||
4fc688ce | 4363 | mutex_lock(&dev_priv->rps.hw_lock); |
0a073b84 JB |
4364 | |
4365 | if (IS_VALLEYVIEW(dev)) { | |
4366 | valleyview_enable_rps(dev); | |
4367 | } else { | |
4368 | gen6_enable_rps(dev); | |
4369 | gen6_update_ring_freq(dev); | |
4370 | } | |
4fc688ce | 4371 | mutex_unlock(&dev_priv->rps.hw_lock); |
1a01ab3b JB |
4372 | } |
4373 | ||
8090c6b9 DV |
4374 | void intel_enable_gt_powersave(struct drm_device *dev) |
4375 | { | |
1a01ab3b JB |
4376 | struct drm_i915_private *dev_priv = dev->dev_private; |
4377 | ||
8090c6b9 DV |
4378 | if (IS_IRONLAKE_M(dev)) { |
4379 | ironlake_enable_drps(dev); | |
4380 | ironlake_enable_rc6(dev); | |
4381 | intel_init_emon(dev); | |
0a073b84 | 4382 | } else if (IS_GEN6(dev) || IS_GEN7(dev)) { |
1a01ab3b JB |
4383 | /* |
4384 | * PCU communication is slow and this doesn't need to be | |
4385 | * done at any specific time, so do this out of our fast path | |
4386 | * to make resume and init faster. | |
4387 | */ | |
4388 | schedule_delayed_work(&dev_priv->rps.delayed_resume_work, | |
4389 | round_jiffies_up_relative(HZ)); | |
8090c6b9 DV |
4390 | } |
4391 | } | |
4392 | ||
3107bd48 DV |
4393 | static void ibx_init_clock_gating(struct drm_device *dev) |
4394 | { | |
4395 | struct drm_i915_private *dev_priv = dev->dev_private; | |
4396 | ||
4397 | /* | |
4398 | * On Ibex Peak and Cougar Point, we need to disable clock | |
4399 | * gating for the panel power sequencer or it will fail to | |
4400 | * start up when no ports are active. | |
4401 | */ | |
4402 | I915_WRITE(SOUTH_DSPCLK_GATE_D, PCH_DPLSUNIT_CLOCK_GATE_DISABLE); | |
4403 | } | |
4404 | ||
0e088b8f VS |
4405 | static void g4x_disable_trickle_feed(struct drm_device *dev) |
4406 | { | |
4407 | struct drm_i915_private *dev_priv = dev->dev_private; | |
4408 | int pipe; | |
4409 | ||
4410 | for_each_pipe(pipe) { | |
4411 | I915_WRITE(DSPCNTR(pipe), | |
4412 | I915_READ(DSPCNTR(pipe)) | | |
4413 | DISPPLANE_TRICKLE_FEED_DISABLE); | |
4414 | intel_flush_display_plane(dev_priv, pipe); | |
4415 | } | |
4416 | } | |
4417 | ||
1fa61106 | 4418 | static void ironlake_init_clock_gating(struct drm_device *dev) |
6f1d69b0 ED |
4419 | { |
4420 | struct drm_i915_private *dev_priv = dev->dev_private; | |
231e54f6 | 4421 | uint32_t dspclk_gate = ILK_VRHUNIT_CLOCK_GATE_DISABLE; |
6f1d69b0 | 4422 | |
f1e8fa56 DL |
4423 | /* |
4424 | * Required for FBC | |
4425 | * WaFbcDisableDpfcClockGating:ilk | |
4426 | */ | |
4d47e4f5 DL |
4427 | dspclk_gate |= ILK_DPFCRUNIT_CLOCK_GATE_DISABLE | |
4428 | ILK_DPFCUNIT_CLOCK_GATE_DISABLE | | |
4429 | ILK_DPFDUNIT_CLOCK_GATE_ENABLE; | |
6f1d69b0 ED |
4430 | |
4431 | I915_WRITE(PCH_3DCGDIS0, | |
4432 | MARIUNIT_CLOCK_GATE_DISABLE | | |
4433 | SVSMUNIT_CLOCK_GATE_DISABLE); | |
4434 | I915_WRITE(PCH_3DCGDIS1, | |
4435 | VFMUNIT_CLOCK_GATE_DISABLE); | |
4436 | ||
6f1d69b0 ED |
4437 | /* |
4438 | * According to the spec the following bits should be set in | |
4439 | * order to enable memory self-refresh | |
4440 | * The bit 22/21 of 0x42004 | |
4441 | * The bit 5 of 0x42020 | |
4442 | * The bit 15 of 0x45000 | |
4443 | */ | |
4444 | I915_WRITE(ILK_DISPLAY_CHICKEN2, | |
4445 | (I915_READ(ILK_DISPLAY_CHICKEN2) | | |
4446 | ILK_DPARB_GATE | ILK_VSDPFD_FULL)); | |
4d47e4f5 | 4447 | dspclk_gate |= ILK_DPARBUNIT_CLOCK_GATE_ENABLE; |
6f1d69b0 ED |
4448 | I915_WRITE(DISP_ARB_CTL, |
4449 | (I915_READ(DISP_ARB_CTL) | | |
4450 | DISP_FBC_WM_DIS)); | |
4451 | I915_WRITE(WM3_LP_ILK, 0); | |
4452 | I915_WRITE(WM2_LP_ILK, 0); | |
4453 | I915_WRITE(WM1_LP_ILK, 0); | |
4454 | ||
4455 | /* | |
4456 | * Based on the document from hardware guys the following bits | |
4457 | * should be set unconditionally in order to enable FBC. | |
4458 | * The bit 22 of 0x42000 | |
4459 | * The bit 22 of 0x42004 | |
4460 | * The bit 7,8,9 of 0x42020. | |
4461 | */ | |
4462 | if (IS_IRONLAKE_M(dev)) { | |
4bb35334 | 4463 | /* WaFbcAsynchFlipDisableFbcQueue:ilk */ |
6f1d69b0 ED |
4464 | I915_WRITE(ILK_DISPLAY_CHICKEN1, |
4465 | I915_READ(ILK_DISPLAY_CHICKEN1) | | |
4466 | ILK_FBCQ_DIS); | |
4467 | I915_WRITE(ILK_DISPLAY_CHICKEN2, | |
4468 | I915_READ(ILK_DISPLAY_CHICKEN2) | | |
4469 | ILK_DPARB_GATE); | |
6f1d69b0 ED |
4470 | } |
4471 | ||
4d47e4f5 DL |
4472 | I915_WRITE(ILK_DSPCLK_GATE_D, dspclk_gate); |
4473 | ||
6f1d69b0 ED |
4474 | I915_WRITE(ILK_DISPLAY_CHICKEN2, |
4475 | I915_READ(ILK_DISPLAY_CHICKEN2) | | |
4476 | ILK_ELPIN_409_SELECT); | |
4477 | I915_WRITE(_3D_CHICKEN2, | |
4478 | _3D_CHICKEN2_WM_READ_PIPELINED << 16 | | |
4479 | _3D_CHICKEN2_WM_READ_PIPELINED); | |
4358a374 | 4480 | |
ecdb4eb7 | 4481 | /* WaDisableRenderCachePipelinedFlush:ilk */ |
4358a374 DV |
4482 | I915_WRITE(CACHE_MODE_0, |
4483 | _MASKED_BIT_ENABLE(CM0_PIPELINED_RENDER_FLUSH_DISABLE)); | |
3107bd48 | 4484 | |
0e088b8f | 4485 | g4x_disable_trickle_feed(dev); |
bdad2b2f | 4486 | |
3107bd48 DV |
4487 | ibx_init_clock_gating(dev); |
4488 | } | |
4489 | ||
4490 | static void cpt_init_clock_gating(struct drm_device *dev) | |
4491 | { | |
4492 | struct drm_i915_private *dev_priv = dev->dev_private; | |
4493 | int pipe; | |
3f704fa2 | 4494 | uint32_t val; |
3107bd48 DV |
4495 | |
4496 | /* | |
4497 | * On Ibex Peak and Cougar Point, we need to disable clock | |
4498 | * gating for the panel power sequencer or it will fail to | |
4499 | * start up when no ports are active. | |
4500 | */ | |
4501 | I915_WRITE(SOUTH_DSPCLK_GATE_D, PCH_DPLSUNIT_CLOCK_GATE_DISABLE); | |
4502 | I915_WRITE(SOUTH_CHICKEN2, I915_READ(SOUTH_CHICKEN2) | | |
4503 | DPLS_EDP_PPS_FIX_DIS); | |
335c07b7 TI |
4504 | /* The below fixes the weird display corruption, a few pixels shifted |
4505 | * downward, on (only) LVDS of some HP laptops with IVY. | |
4506 | */ | |
3f704fa2 | 4507 | for_each_pipe(pipe) { |
dc4bd2d1 PZ |
4508 | val = I915_READ(TRANS_CHICKEN2(pipe)); |
4509 | val |= TRANS_CHICKEN2_TIMING_OVERRIDE; | |
4510 | val &= ~TRANS_CHICKEN2_FDI_POLARITY_REVERSED; | |
41aa3448 | 4511 | if (dev_priv->vbt.fdi_rx_polarity_inverted) |
3f704fa2 | 4512 | val |= TRANS_CHICKEN2_FDI_POLARITY_REVERSED; |
dc4bd2d1 PZ |
4513 | val &= ~TRANS_CHICKEN2_FRAME_START_DELAY_MASK; |
4514 | val &= ~TRANS_CHICKEN2_DISABLE_DEEP_COLOR_COUNTER; | |
4515 | val &= ~TRANS_CHICKEN2_DISABLE_DEEP_COLOR_MODESWITCH; | |
3f704fa2 PZ |
4516 | I915_WRITE(TRANS_CHICKEN2(pipe), val); |
4517 | } | |
3107bd48 DV |
4518 | /* WADP0ClockGatingDisable */ |
4519 | for_each_pipe(pipe) { | |
4520 | I915_WRITE(TRANS_CHICKEN1(pipe), | |
4521 | TRANS_CHICKEN1_DP0UNIT_GC_DISABLE); | |
4522 | } | |
6f1d69b0 ED |
4523 | } |
4524 | ||
1d7aaa0c DV |
4525 | static void gen6_check_mch_setup(struct drm_device *dev) |
4526 | { | |
4527 | struct drm_i915_private *dev_priv = dev->dev_private; | |
4528 | uint32_t tmp; | |
4529 | ||
4530 | tmp = I915_READ(MCH_SSKPD); | |
4531 | if ((tmp & MCH_SSKPD_WM0_MASK) != MCH_SSKPD_WM0_VAL) { | |
4532 | DRM_INFO("Wrong MCH_SSKPD value: 0x%08x\n", tmp); | |
4533 | DRM_INFO("This can cause pipe underruns and display issues.\n"); | |
4534 | DRM_INFO("Please upgrade your BIOS to fix this.\n"); | |
4535 | } | |
4536 | } | |
4537 | ||
1fa61106 | 4538 | static void gen6_init_clock_gating(struct drm_device *dev) |
6f1d69b0 ED |
4539 | { |
4540 | struct drm_i915_private *dev_priv = dev->dev_private; | |
231e54f6 | 4541 | uint32_t dspclk_gate = ILK_VRHUNIT_CLOCK_GATE_DISABLE; |
6f1d69b0 | 4542 | |
231e54f6 | 4543 | I915_WRITE(ILK_DSPCLK_GATE_D, dspclk_gate); |
6f1d69b0 ED |
4544 | |
4545 | I915_WRITE(ILK_DISPLAY_CHICKEN2, | |
4546 | I915_READ(ILK_DISPLAY_CHICKEN2) | | |
4547 | ILK_ELPIN_409_SELECT); | |
4548 | ||
ecdb4eb7 | 4549 | /* WaDisableHiZPlanesWhenMSAAEnabled:snb */ |
4283908e DV |
4550 | I915_WRITE(_3D_CHICKEN, |
4551 | _MASKED_BIT_ENABLE(_3D_CHICKEN_HIZ_PLANE_DISABLE_MSAA_4X_SNB)); | |
4552 | ||
ecdb4eb7 | 4553 | /* WaSetupGtModeTdRowDispatch:snb */ |
6547fbdb DV |
4554 | if (IS_SNB_GT1(dev)) |
4555 | I915_WRITE(GEN6_GT_MODE, | |
4556 | _MASKED_BIT_ENABLE(GEN6_TD_FOUR_ROW_DISPATCH_DISABLE)); | |
4557 | ||
6f1d69b0 ED |
4558 | I915_WRITE(WM3_LP_ILK, 0); |
4559 | I915_WRITE(WM2_LP_ILK, 0); | |
4560 | I915_WRITE(WM1_LP_ILK, 0); | |
4561 | ||
6f1d69b0 | 4562 | I915_WRITE(CACHE_MODE_0, |
50743298 | 4563 | _MASKED_BIT_DISABLE(CM0_STC_EVICT_DISABLE_LRA_SNB)); |
6f1d69b0 ED |
4564 | |
4565 | I915_WRITE(GEN6_UCGCTL1, | |
4566 | I915_READ(GEN6_UCGCTL1) | | |
4567 | GEN6_BLBUNIT_CLOCK_GATE_DISABLE | | |
4568 | GEN6_CSUNIT_CLOCK_GATE_DISABLE); | |
4569 | ||
4570 | /* According to the BSpec vol1g, bit 12 (RCPBUNIT) clock | |
4571 | * gating disable must be set. Failure to set it results in | |
4572 | * flickering pixels due to Z write ordering failures after | |
4573 | * some amount of runtime in the Mesa "fire" demo, and Unigine | |
4574 | * Sanctuary and Tropics, and apparently anything else with | |
4575 | * alpha test or pixel discard. | |
4576 | * | |
4577 | * According to the spec, bit 11 (RCCUNIT) must also be set, | |
4578 | * but we didn't debug actual testcases to find it out. | |
0f846f81 | 4579 | * |
ecdb4eb7 DL |
4580 | * Also apply WaDisableVDSUnitClockGating:snb and |
4581 | * WaDisableRCPBUnitClockGating:snb. | |
6f1d69b0 ED |
4582 | */ |
4583 | I915_WRITE(GEN6_UCGCTL2, | |
0f846f81 | 4584 | GEN7_VDSUNIT_CLOCK_GATE_DISABLE | |
6f1d69b0 ED |
4585 | GEN6_RCPBUNIT_CLOCK_GATE_DISABLE | |
4586 | GEN6_RCCUNIT_CLOCK_GATE_DISABLE); | |
4587 | ||
4588 | /* Bspec says we need to always set all mask bits. */ | |
26b6e44a KG |
4589 | I915_WRITE(_3D_CHICKEN3, (0xFFFF << 16) | |
4590 | _3D_CHICKEN3_SF_DISABLE_FASTCLIP_CULL); | |
6f1d69b0 ED |
4591 | |
4592 | /* | |
4593 | * According to the spec the following bits should be | |
4594 | * set in order to enable memory self-refresh and fbc: | |
4595 | * The bit21 and bit22 of 0x42000 | |
4596 | * The bit21 and bit22 of 0x42004 | |
4597 | * The bit5 and bit7 of 0x42020 | |
4598 | * The bit14 of 0x70180 | |
4599 | * The bit14 of 0x71180 | |
4bb35334 DL |
4600 | * |
4601 | * WaFbcAsynchFlipDisableFbcQueue:snb | |
6f1d69b0 ED |
4602 | */ |
4603 | I915_WRITE(ILK_DISPLAY_CHICKEN1, | |
4604 | I915_READ(ILK_DISPLAY_CHICKEN1) | | |
4605 | ILK_FBCQ_DIS | ILK_PABSTRETCH_DIS); | |
4606 | I915_WRITE(ILK_DISPLAY_CHICKEN2, | |
4607 | I915_READ(ILK_DISPLAY_CHICKEN2) | | |
4608 | ILK_DPARB_GATE | ILK_VSDPFD_FULL); | |
231e54f6 DL |
4609 | I915_WRITE(ILK_DSPCLK_GATE_D, |
4610 | I915_READ(ILK_DSPCLK_GATE_D) | | |
4611 | ILK_DPARBUNIT_CLOCK_GATE_ENABLE | | |
4612 | ILK_DPFDUNIT_CLOCK_GATE_ENABLE); | |
6f1d69b0 | 4613 | |
ecdb4eb7 | 4614 | /* WaMbcDriverBootEnable:snb */ |
b4ae3f22 JB |
4615 | I915_WRITE(GEN6_MBCTL, I915_READ(GEN6_MBCTL) | |
4616 | GEN6_MBCTL_ENABLE_BOOT_FETCH); | |
4617 | ||
0e088b8f | 4618 | g4x_disable_trickle_feed(dev); |
f8f2ac9a BW |
4619 | |
4620 | /* The default value should be 0x200 according to docs, but the two | |
4621 | * platforms I checked have a 0 for this. (Maybe BIOS overrides?) */ | |
4622 | I915_WRITE(GEN6_GT_MODE, _MASKED_BIT_DISABLE(0xffff)); | |
4623 | I915_WRITE(GEN6_GT_MODE, _MASKED_BIT_ENABLE(GEN6_GT_MODE_HI)); | |
3107bd48 DV |
4624 | |
4625 | cpt_init_clock_gating(dev); | |
1d7aaa0c DV |
4626 | |
4627 | gen6_check_mch_setup(dev); | |
6f1d69b0 ED |
4628 | } |
4629 | ||
4630 | static void gen7_setup_fixed_func_scheduler(struct drm_i915_private *dev_priv) | |
4631 | { | |
4632 | uint32_t reg = I915_READ(GEN7_FF_THREAD_MODE); | |
4633 | ||
4634 | reg &= ~GEN7_FF_SCHED_MASK; | |
4635 | reg |= GEN7_FF_TS_SCHED_HW; | |
4636 | reg |= GEN7_FF_VS_SCHED_HW; | |
4637 | reg |= GEN7_FF_DS_SCHED_HW; | |
4638 | ||
41c0b3a8 BW |
4639 | if (IS_HASWELL(dev_priv->dev)) |
4640 | reg &= ~GEN7_FF_VS_REF_CNT_FFME; | |
4641 | ||
6f1d69b0 ED |
4642 | I915_WRITE(GEN7_FF_THREAD_MODE, reg); |
4643 | } | |
4644 | ||
17a303ec PZ |
4645 | static void lpt_init_clock_gating(struct drm_device *dev) |
4646 | { | |
4647 | struct drm_i915_private *dev_priv = dev->dev_private; | |
4648 | ||
4649 | /* | |
4650 | * TODO: this bit should only be enabled when really needed, then | |
4651 | * disabled when not needed anymore in order to save power. | |
4652 | */ | |
4653 | if (dev_priv->pch_id == INTEL_PCH_LPT_LP_DEVICE_ID_TYPE) | |
4654 | I915_WRITE(SOUTH_DSPCLK_GATE_D, | |
4655 | I915_READ(SOUTH_DSPCLK_GATE_D) | | |
4656 | PCH_LP_PARTITION_LEVEL_DISABLE); | |
0a790cdb PZ |
4657 | |
4658 | /* WADPOClockGatingDisable:hsw */ | |
4659 | I915_WRITE(_TRANSA_CHICKEN1, | |
4660 | I915_READ(_TRANSA_CHICKEN1) | | |
4661 | TRANS_CHICKEN1_DP0UNIT_GC_DISABLE); | |
17a303ec PZ |
4662 | } |
4663 | ||
7d708ee4 ID |
4664 | static void lpt_suspend_hw(struct drm_device *dev) |
4665 | { | |
4666 | struct drm_i915_private *dev_priv = dev->dev_private; | |
4667 | ||
4668 | if (dev_priv->pch_id == INTEL_PCH_LPT_LP_DEVICE_ID_TYPE) { | |
4669 | uint32_t val = I915_READ(SOUTH_DSPCLK_GATE_D); | |
4670 | ||
4671 | val &= ~PCH_LP_PARTITION_LEVEL_DISABLE; | |
4672 | I915_WRITE(SOUTH_DSPCLK_GATE_D, val); | |
4673 | } | |
4674 | } | |
4675 | ||
cad2a2d7 ED |
4676 | static void haswell_init_clock_gating(struct drm_device *dev) |
4677 | { | |
4678 | struct drm_i915_private *dev_priv = dev->dev_private; | |
cad2a2d7 ED |
4679 | |
4680 | I915_WRITE(WM3_LP_ILK, 0); | |
4681 | I915_WRITE(WM2_LP_ILK, 0); | |
4682 | I915_WRITE(WM1_LP_ILK, 0); | |
4683 | ||
4684 | /* According to the spec, bit 13 (RCZUNIT) must be set on IVB. | |
ecdb4eb7 | 4685 | * This implements the WaDisableRCZUnitClockGating:hsw workaround. |
cad2a2d7 ED |
4686 | */ |
4687 | I915_WRITE(GEN6_UCGCTL2, GEN6_RCZUNIT_CLOCK_GATE_DISABLE); | |
4688 | ||
ecdb4eb7 | 4689 | /* Apply the WaDisableRHWOOptimizationForRenderHang:hsw workaround. */ |
cad2a2d7 ED |
4690 | I915_WRITE(GEN7_COMMON_SLICE_CHICKEN1, |
4691 | GEN7_CSC1_RHWO_OPT_DISABLE_IN_RCC); | |
4692 | ||
ecdb4eb7 | 4693 | /* WaApplyL3ControlAndL3ChickenMode:hsw */ |
cad2a2d7 ED |
4694 | I915_WRITE(GEN7_L3CNTLREG1, |
4695 | GEN7_WA_FOR_GEN7_L3_CONTROL); | |
4696 | I915_WRITE(GEN7_L3_CHICKEN_MODE_REGISTER, | |
4697 | GEN7_WA_L3_CHICKEN_MODE); | |
4698 | ||
ecdb4eb7 | 4699 | /* This is required by WaCatErrorRejectionIssue:hsw */ |
cad2a2d7 ED |
4700 | I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG, |
4701 | I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG) | | |
4702 | GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB); | |
4703 | ||
0e088b8f | 4704 | g4x_disable_trickle_feed(dev); |
cad2a2d7 | 4705 | |
ecdb4eb7 | 4706 | /* WaVSRefCountFullforceMissDisable:hsw */ |
cad2a2d7 ED |
4707 | gen7_setup_fixed_func_scheduler(dev_priv); |
4708 | ||
ecdb4eb7 | 4709 | /* WaDisable4x2SubspanOptimization:hsw */ |
cad2a2d7 ED |
4710 | I915_WRITE(CACHE_MODE_1, |
4711 | _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE)); | |
1544d9d5 | 4712 | |
ecdb4eb7 | 4713 | /* WaMbcDriverBootEnable:hsw */ |
b3bf0766 PZ |
4714 | I915_WRITE(GEN6_MBCTL, I915_READ(GEN6_MBCTL) | |
4715 | GEN6_MBCTL_ENABLE_BOOT_FETCH); | |
4716 | ||
ecdb4eb7 | 4717 | /* WaSwitchSolVfFArbitrationPriority:hsw */ |
e3dff585 BW |
4718 | I915_WRITE(GAM_ECOCHK, I915_READ(GAM_ECOCHK) | HSW_ECOCHK_ARB_PRIO_SOL); |
4719 | ||
90a88643 PZ |
4720 | /* WaRsPkgCStateDisplayPMReq:hsw */ |
4721 | I915_WRITE(CHICKEN_PAR1_1, | |
4722 | I915_READ(CHICKEN_PAR1_1) | FORCE_ARB_IDLE_PLANES); | |
1544d9d5 | 4723 | |
17a303ec | 4724 | lpt_init_clock_gating(dev); |
cad2a2d7 ED |
4725 | } |
4726 | ||
1fa61106 | 4727 | static void ivybridge_init_clock_gating(struct drm_device *dev) |
6f1d69b0 ED |
4728 | { |
4729 | struct drm_i915_private *dev_priv = dev->dev_private; | |
20848223 | 4730 | uint32_t snpcr; |
6f1d69b0 | 4731 | |
6f1d69b0 ED |
4732 | I915_WRITE(WM3_LP_ILK, 0); |
4733 | I915_WRITE(WM2_LP_ILK, 0); | |
4734 | I915_WRITE(WM1_LP_ILK, 0); | |
4735 | ||
231e54f6 | 4736 | I915_WRITE(ILK_DSPCLK_GATE_D, ILK_VRHUNIT_CLOCK_GATE_DISABLE); |
6f1d69b0 | 4737 | |
ecdb4eb7 | 4738 | /* WaDisableEarlyCull:ivb */ |
87f8020e JB |
4739 | I915_WRITE(_3D_CHICKEN3, |
4740 | _MASKED_BIT_ENABLE(_3D_CHICKEN_SF_DISABLE_OBJEND_CULL)); | |
4741 | ||
ecdb4eb7 | 4742 | /* WaDisableBackToBackFlipFix:ivb */ |
6f1d69b0 ED |
4743 | I915_WRITE(IVB_CHICKEN3, |
4744 | CHICKEN3_DGMG_REQ_OUT_FIX_DISABLE | | |
4745 | CHICKEN3_DGMG_DONE_FIX_DISABLE); | |
4746 | ||
ecdb4eb7 | 4747 | /* WaDisablePSDDualDispatchEnable:ivb */ |
12f3382b JB |
4748 | if (IS_IVB_GT1(dev)) |
4749 | I915_WRITE(GEN7_HALF_SLICE_CHICKEN1, | |
4750 | _MASKED_BIT_ENABLE(GEN7_PSD_SINGLE_PORT_DISPATCH_ENABLE)); | |
4751 | else | |
4752 | I915_WRITE(GEN7_HALF_SLICE_CHICKEN1_GT2, | |
4753 | _MASKED_BIT_ENABLE(GEN7_PSD_SINGLE_PORT_DISPATCH_ENABLE)); | |
4754 | ||
ecdb4eb7 | 4755 | /* Apply the WaDisableRHWOOptimizationForRenderHang:ivb workaround. */ |
6f1d69b0 ED |
4756 | I915_WRITE(GEN7_COMMON_SLICE_CHICKEN1, |
4757 | GEN7_CSC1_RHWO_OPT_DISABLE_IN_RCC); | |
4758 | ||
ecdb4eb7 | 4759 | /* WaApplyL3ControlAndL3ChickenMode:ivb */ |
6f1d69b0 ED |
4760 | I915_WRITE(GEN7_L3CNTLREG1, |
4761 | GEN7_WA_FOR_GEN7_L3_CONTROL); | |
4762 | I915_WRITE(GEN7_L3_CHICKEN_MODE_REGISTER, | |
8ab43976 JB |
4763 | GEN7_WA_L3_CHICKEN_MODE); |
4764 | if (IS_IVB_GT1(dev)) | |
4765 | I915_WRITE(GEN7_ROW_CHICKEN2, | |
4766 | _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE)); | |
4767 | else | |
4768 | I915_WRITE(GEN7_ROW_CHICKEN2_GT2, | |
4769 | _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE)); | |
4770 | ||
6f1d69b0 | 4771 | |
ecdb4eb7 | 4772 | /* WaForceL3Serialization:ivb */ |
61939d97 JB |
4773 | I915_WRITE(GEN7_L3SQCREG4, I915_READ(GEN7_L3SQCREG4) & |
4774 | ~L3SQ_URB_READ_CAM_MATCH_DISABLE); | |
4775 | ||
0f846f81 JB |
4776 | /* According to the BSpec vol1g, bit 12 (RCPBUNIT) clock |
4777 | * gating disable must be set. Failure to set it results in | |
4778 | * flickering pixels due to Z write ordering failures after | |
4779 | * some amount of runtime in the Mesa "fire" demo, and Unigine | |
4780 | * Sanctuary and Tropics, and apparently anything else with | |
4781 | * alpha test or pixel discard. | |
4782 | * | |
4783 | * According to the spec, bit 11 (RCCUNIT) must also be set, | |
4784 | * but we didn't debug actual testcases to find it out. | |
4785 | * | |
4786 | * According to the spec, bit 13 (RCZUNIT) must be set on IVB. | |
ecdb4eb7 | 4787 | * This implements the WaDisableRCZUnitClockGating:ivb workaround. |
0f846f81 JB |
4788 | */ |
4789 | I915_WRITE(GEN6_UCGCTL2, | |
4790 | GEN6_RCZUNIT_CLOCK_GATE_DISABLE | | |
4791 | GEN6_RCCUNIT_CLOCK_GATE_DISABLE); | |
4792 | ||
ecdb4eb7 | 4793 | /* This is required by WaCatErrorRejectionIssue:ivb */ |
6f1d69b0 ED |
4794 | I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG, |
4795 | I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG) | | |
4796 | GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB); | |
4797 | ||
0e088b8f | 4798 | g4x_disable_trickle_feed(dev); |
6f1d69b0 | 4799 | |
ecdb4eb7 | 4800 | /* WaMbcDriverBootEnable:ivb */ |
b4ae3f22 JB |
4801 | I915_WRITE(GEN6_MBCTL, I915_READ(GEN6_MBCTL) | |
4802 | GEN6_MBCTL_ENABLE_BOOT_FETCH); | |
4803 | ||
ecdb4eb7 | 4804 | /* WaVSRefCountFullforceMissDisable:ivb */ |
6f1d69b0 | 4805 | gen7_setup_fixed_func_scheduler(dev_priv); |
97e1930f | 4806 | |
ecdb4eb7 | 4807 | /* WaDisable4x2SubspanOptimization:ivb */ |
97e1930f DV |
4808 | I915_WRITE(CACHE_MODE_1, |
4809 | _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE)); | |
20848223 BW |
4810 | |
4811 | snpcr = I915_READ(GEN6_MBCUNIT_SNPCR); | |
4812 | snpcr &= ~GEN6_MBC_SNPCR_MASK; | |
4813 | snpcr |= GEN6_MBC_SNPCR_MED; | |
4814 | I915_WRITE(GEN6_MBCUNIT_SNPCR, snpcr); | |
3107bd48 | 4815 | |
ab5c608b BW |
4816 | if (!HAS_PCH_NOP(dev)) |
4817 | cpt_init_clock_gating(dev); | |
1d7aaa0c DV |
4818 | |
4819 | gen6_check_mch_setup(dev); | |
6f1d69b0 ED |
4820 | } |
4821 | ||
1fa61106 | 4822 | static void valleyview_init_clock_gating(struct drm_device *dev) |
6f1d69b0 ED |
4823 | { |
4824 | struct drm_i915_private *dev_priv = dev->dev_private; | |
6f1d69b0 | 4825 | |
d7fe0cc0 | 4826 | I915_WRITE(DSPCLK_GATE_D, VRHUNIT_CLOCK_GATE_DISABLE); |
6f1d69b0 | 4827 | |
ecdb4eb7 | 4828 | /* WaDisableEarlyCull:vlv */ |
87f8020e JB |
4829 | I915_WRITE(_3D_CHICKEN3, |
4830 | _MASKED_BIT_ENABLE(_3D_CHICKEN_SF_DISABLE_OBJEND_CULL)); | |
4831 | ||
ecdb4eb7 | 4832 | /* WaDisableBackToBackFlipFix:vlv */ |
6f1d69b0 ED |
4833 | I915_WRITE(IVB_CHICKEN3, |
4834 | CHICKEN3_DGMG_REQ_OUT_FIX_DISABLE | | |
4835 | CHICKEN3_DGMG_DONE_FIX_DISABLE); | |
4836 | ||
ecdb4eb7 | 4837 | /* WaDisablePSDDualDispatchEnable:vlv */ |
12f3382b | 4838 | I915_WRITE(GEN7_HALF_SLICE_CHICKEN1, |
d3bc0303 JB |
4839 | _MASKED_BIT_ENABLE(GEN7_MAX_PS_THREAD_DEP | |
4840 | GEN7_PSD_SINGLE_PORT_DISPATCH_ENABLE)); | |
12f3382b | 4841 | |
ecdb4eb7 | 4842 | /* Apply the WaDisableRHWOOptimizationForRenderHang:vlv workaround. */ |
6f1d69b0 ED |
4843 | I915_WRITE(GEN7_COMMON_SLICE_CHICKEN1, |
4844 | GEN7_CSC1_RHWO_OPT_DISABLE_IN_RCC); | |
4845 | ||
ecdb4eb7 | 4846 | /* WaApplyL3ControlAndL3ChickenMode:vlv */ |
d0cf5ead | 4847 | I915_WRITE(GEN7_L3CNTLREG1, I915_READ(GEN7_L3CNTLREG1) | GEN7_L3AGDIS); |
6f1d69b0 ED |
4848 | I915_WRITE(GEN7_L3_CHICKEN_MODE_REGISTER, GEN7_WA_L3_CHICKEN_MODE); |
4849 | ||
ecdb4eb7 | 4850 | /* WaForceL3Serialization:vlv */ |
61939d97 JB |
4851 | I915_WRITE(GEN7_L3SQCREG4, I915_READ(GEN7_L3SQCREG4) & |
4852 | ~L3SQ_URB_READ_CAM_MATCH_DISABLE); | |
4853 | ||
ecdb4eb7 | 4854 | /* WaDisableDopClockGating:vlv */ |
8ab43976 JB |
4855 | I915_WRITE(GEN7_ROW_CHICKEN2, |
4856 | _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE)); | |
4857 | ||
ecdb4eb7 | 4858 | /* This is required by WaCatErrorRejectionIssue:vlv */ |
6f1d69b0 ED |
4859 | I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG, |
4860 | I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG) | | |
4861 | GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB); | |
4862 | ||
ecdb4eb7 | 4863 | /* WaMbcDriverBootEnable:vlv */ |
b4ae3f22 JB |
4864 | I915_WRITE(GEN6_MBCTL, I915_READ(GEN6_MBCTL) | |
4865 | GEN6_MBCTL_ENABLE_BOOT_FETCH); | |
4866 | ||
0f846f81 JB |
4867 | |
4868 | /* According to the BSpec vol1g, bit 12 (RCPBUNIT) clock | |
4869 | * gating disable must be set. Failure to set it results in | |
4870 | * flickering pixels due to Z write ordering failures after | |
4871 | * some amount of runtime in the Mesa "fire" demo, and Unigine | |
4872 | * Sanctuary and Tropics, and apparently anything else with | |
4873 | * alpha test or pixel discard. | |
4874 | * | |
4875 | * According to the spec, bit 11 (RCCUNIT) must also be set, | |
4876 | * but we didn't debug actual testcases to find it out. | |
4877 | * | |
4878 | * According to the spec, bit 13 (RCZUNIT) must be set on IVB. | |
ecdb4eb7 | 4879 | * This implements the WaDisableRCZUnitClockGating:vlv workaround. |
0f846f81 | 4880 | * |
ecdb4eb7 DL |
4881 | * Also apply WaDisableVDSUnitClockGating:vlv and |
4882 | * WaDisableRCPBUnitClockGating:vlv. | |
0f846f81 JB |
4883 | */ |
4884 | I915_WRITE(GEN6_UCGCTL2, | |
4885 | GEN7_VDSUNIT_CLOCK_GATE_DISABLE | | |
6edaa7fc | 4886 | GEN7_TDLUNIT_CLOCK_GATE_DISABLE | |
0f846f81 JB |
4887 | GEN6_RCZUNIT_CLOCK_GATE_DISABLE | |
4888 | GEN6_RCPBUNIT_CLOCK_GATE_DISABLE | | |
4889 | GEN6_RCCUNIT_CLOCK_GATE_DISABLE); | |
4890 | ||
e3f33d46 JB |
4891 | I915_WRITE(GEN7_UCGCTL4, GEN7_L3BANK2X_CLOCK_GATE_DISABLE); |
4892 | ||
e0d8d59b | 4893 | I915_WRITE(MI_ARB_VLV, MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE); |
6f1d69b0 | 4894 | |
6b26c86d DV |
4895 | I915_WRITE(CACHE_MODE_1, |
4896 | _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE)); | |
7983117f | 4897 | |
2d809570 | 4898 | /* |
ecdb4eb7 | 4899 | * WaDisableVLVClockGating_VBIIssue:vlv |
2d809570 JB |
4900 | * Disable clock gating on th GCFG unit to prevent a delay |
4901 | * in the reporting of vblank events. | |
4902 | */ | |
4e8c84a5 JB |
4903 | I915_WRITE(VLV_GUNIT_CLOCK_GATE, 0xffffffff); |
4904 | ||
4905 | /* Conservative clock gating settings for now */ | |
4906 | I915_WRITE(0x9400, 0xffffffff); | |
4907 | I915_WRITE(0x9404, 0xffffffff); | |
4908 | I915_WRITE(0x9408, 0xffffffff); | |
4909 | I915_WRITE(0x940c, 0xffffffff); | |
4910 | I915_WRITE(0x9410, 0xffffffff); | |
4911 | I915_WRITE(0x9414, 0xffffffff); | |
4912 | I915_WRITE(0x9418, 0xffffffff); | |
6f1d69b0 ED |
4913 | } |
4914 | ||
1fa61106 | 4915 | static void g4x_init_clock_gating(struct drm_device *dev) |
6f1d69b0 ED |
4916 | { |
4917 | struct drm_i915_private *dev_priv = dev->dev_private; | |
4918 | uint32_t dspclk_gate; | |
4919 | ||
4920 | I915_WRITE(RENCLK_GATE_D1, 0); | |
4921 | I915_WRITE(RENCLK_GATE_D2, VF_UNIT_CLOCK_GATE_DISABLE | | |
4922 | GS_UNIT_CLOCK_GATE_DISABLE | | |
4923 | CL_UNIT_CLOCK_GATE_DISABLE); | |
4924 | I915_WRITE(RAMCLK_GATE_D, 0); | |
4925 | dspclk_gate = VRHUNIT_CLOCK_GATE_DISABLE | | |
4926 | OVRUNIT_CLOCK_GATE_DISABLE | | |
4927 | OVCUNIT_CLOCK_GATE_DISABLE; | |
4928 | if (IS_GM45(dev)) | |
4929 | dspclk_gate |= DSSUNIT_CLOCK_GATE_DISABLE; | |
4930 | I915_WRITE(DSPCLK_GATE_D, dspclk_gate); | |
4358a374 DV |
4931 | |
4932 | /* WaDisableRenderCachePipelinedFlush */ | |
4933 | I915_WRITE(CACHE_MODE_0, | |
4934 | _MASKED_BIT_ENABLE(CM0_PIPELINED_RENDER_FLUSH_DISABLE)); | |
de1aa629 | 4935 | |
0e088b8f | 4936 | g4x_disable_trickle_feed(dev); |
6f1d69b0 ED |
4937 | } |
4938 | ||
1fa61106 | 4939 | static void crestline_init_clock_gating(struct drm_device *dev) |
6f1d69b0 ED |
4940 | { |
4941 | struct drm_i915_private *dev_priv = dev->dev_private; | |
4942 | ||
4943 | I915_WRITE(RENCLK_GATE_D1, I965_RCC_CLOCK_GATE_DISABLE); | |
4944 | I915_WRITE(RENCLK_GATE_D2, 0); | |
4945 | I915_WRITE(DSPCLK_GATE_D, 0); | |
4946 | I915_WRITE(RAMCLK_GATE_D, 0); | |
4947 | I915_WRITE16(DEUC, 0); | |
20f94967 VS |
4948 | I915_WRITE(MI_ARB_STATE, |
4949 | _MASKED_BIT_ENABLE(MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE)); | |
6f1d69b0 ED |
4950 | } |
4951 | ||
1fa61106 | 4952 | static void broadwater_init_clock_gating(struct drm_device *dev) |
6f1d69b0 ED |
4953 | { |
4954 | struct drm_i915_private *dev_priv = dev->dev_private; | |
4955 | ||
4956 | I915_WRITE(RENCLK_GATE_D1, I965_RCZ_CLOCK_GATE_DISABLE | | |
4957 | I965_RCC_CLOCK_GATE_DISABLE | | |
4958 | I965_RCPB_CLOCK_GATE_DISABLE | | |
4959 | I965_ISC_CLOCK_GATE_DISABLE | | |
4960 | I965_FBC_CLOCK_GATE_DISABLE); | |
4961 | I915_WRITE(RENCLK_GATE_D2, 0); | |
20f94967 VS |
4962 | I915_WRITE(MI_ARB_STATE, |
4963 | _MASKED_BIT_ENABLE(MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE)); | |
6f1d69b0 ED |
4964 | } |
4965 | ||
1fa61106 | 4966 | static void gen3_init_clock_gating(struct drm_device *dev) |
6f1d69b0 ED |
4967 | { |
4968 | struct drm_i915_private *dev_priv = dev->dev_private; | |
4969 | u32 dstate = I915_READ(D_STATE); | |
4970 | ||
4971 | dstate |= DSTATE_PLL_D3_OFF | DSTATE_GFX_CLOCK_GATING | | |
4972 | DSTATE_DOT_CLOCK_GATING; | |
4973 | I915_WRITE(D_STATE, dstate); | |
13a86b85 CW |
4974 | |
4975 | if (IS_PINEVIEW(dev)) | |
4976 | I915_WRITE(ECOSKPD, _MASKED_BIT_ENABLE(ECO_GATING_CX_ONLY)); | |
974a3b0f DV |
4977 | |
4978 | /* IIR "flip pending" means done if this bit is set */ | |
4979 | I915_WRITE(ECOSKPD, _MASKED_BIT_DISABLE(ECO_FLIP_DONE)); | |
6f1d69b0 ED |
4980 | } |
4981 | ||
1fa61106 | 4982 | static void i85x_init_clock_gating(struct drm_device *dev) |
6f1d69b0 ED |
4983 | { |
4984 | struct drm_i915_private *dev_priv = dev->dev_private; | |
4985 | ||
4986 | I915_WRITE(RENCLK_GATE_D1, SV_CLOCK_GATE_DISABLE); | |
4987 | } | |
4988 | ||
1fa61106 | 4989 | static void i830_init_clock_gating(struct drm_device *dev) |
6f1d69b0 ED |
4990 | { |
4991 | struct drm_i915_private *dev_priv = dev->dev_private; | |
4992 | ||
4993 | I915_WRITE(DSPCLK_GATE_D, OVRUNIT_CLOCK_GATE_DISABLE); | |
4994 | } | |
4995 | ||
6f1d69b0 ED |
4996 | void intel_init_clock_gating(struct drm_device *dev) |
4997 | { | |
4998 | struct drm_i915_private *dev_priv = dev->dev_private; | |
4999 | ||
5000 | dev_priv->display.init_clock_gating(dev); | |
6f1d69b0 ED |
5001 | } |
5002 | ||
7d708ee4 ID |
5003 | void intel_suspend_hw(struct drm_device *dev) |
5004 | { | |
5005 | if (HAS_PCH_LPT(dev)) | |
5006 | lpt_suspend_hw(dev); | |
5007 | } | |
5008 | ||
15d199ea PZ |
5009 | /** |
5010 | * We should only use the power well if we explicitly asked the hardware to | |
5011 | * enable it, so check if it's enabled and also check if we've requested it to | |
5012 | * be enabled. | |
5013 | */ | |
b97186f0 PZ |
5014 | bool intel_display_power_enabled(struct drm_device *dev, |
5015 | enum intel_display_power_domain domain) | |
15d199ea PZ |
5016 | { |
5017 | struct drm_i915_private *dev_priv = dev->dev_private; | |
5018 | ||
b97186f0 PZ |
5019 | if (!HAS_POWER_WELL(dev)) |
5020 | return true; | |
5021 | ||
5022 | switch (domain) { | |
5023 | case POWER_DOMAIN_PIPE_A: | |
5024 | case POWER_DOMAIN_TRANSCODER_EDP: | |
5025 | return true; | |
5026 | case POWER_DOMAIN_PIPE_B: | |
5027 | case POWER_DOMAIN_PIPE_C: | |
5028 | case POWER_DOMAIN_PIPE_A_PANEL_FITTER: | |
5029 | case POWER_DOMAIN_PIPE_B_PANEL_FITTER: | |
5030 | case POWER_DOMAIN_PIPE_C_PANEL_FITTER: | |
5031 | case POWER_DOMAIN_TRANSCODER_A: | |
5032 | case POWER_DOMAIN_TRANSCODER_B: | |
5033 | case POWER_DOMAIN_TRANSCODER_C: | |
15d199ea PZ |
5034 | return I915_READ(HSW_PWR_WELL_DRIVER) == |
5035 | (HSW_PWR_WELL_ENABLE | HSW_PWR_WELL_STATE); | |
b97186f0 PZ |
5036 | default: |
5037 | BUG(); | |
5038 | } | |
15d199ea PZ |
5039 | } |
5040 | ||
a38911a3 | 5041 | static void __intel_set_power_well(struct drm_device *dev, bool enable) |
d0d3e513 ED |
5042 | { |
5043 | struct drm_i915_private *dev_priv = dev->dev_private; | |
fa42e23c PZ |
5044 | bool is_enabled, enable_requested; |
5045 | uint32_t tmp; | |
d0d3e513 | 5046 | |
fa42e23c PZ |
5047 | tmp = I915_READ(HSW_PWR_WELL_DRIVER); |
5048 | is_enabled = tmp & HSW_PWR_WELL_STATE; | |
5049 | enable_requested = tmp & HSW_PWR_WELL_ENABLE; | |
d0d3e513 | 5050 | |
fa42e23c PZ |
5051 | if (enable) { |
5052 | if (!enable_requested) | |
5053 | I915_WRITE(HSW_PWR_WELL_DRIVER, HSW_PWR_WELL_ENABLE); | |
d0d3e513 | 5054 | |
fa42e23c PZ |
5055 | if (!is_enabled) { |
5056 | DRM_DEBUG_KMS("Enabling power well\n"); | |
5057 | if (wait_for((I915_READ(HSW_PWR_WELL_DRIVER) & | |
5058 | HSW_PWR_WELL_STATE), 20)) | |
5059 | DRM_ERROR("Timeout enabling power well\n"); | |
5060 | } | |
5061 | } else { | |
5062 | if (enable_requested) { | |
5063 | I915_WRITE(HSW_PWR_WELL_DRIVER, 0); | |
5064 | DRM_DEBUG_KMS("Requesting to disable the power well\n"); | |
d0d3e513 ED |
5065 | } |
5066 | } | |
fa42e23c | 5067 | } |
d0d3e513 | 5068 | |
a38911a3 WX |
5069 | static struct i915_power_well *hsw_pwr; |
5070 | ||
5071 | /* Display audio driver power well request */ | |
5072 | void i915_request_power_well(void) | |
5073 | { | |
5074 | if (WARN_ON(!hsw_pwr)) | |
5075 | return; | |
5076 | ||
5077 | spin_lock_irq(&hsw_pwr->lock); | |
5078 | if (!hsw_pwr->count++ && | |
5079 | !hsw_pwr->i915_request) | |
5080 | __intel_set_power_well(hsw_pwr->device, true); | |
5081 | spin_unlock_irq(&hsw_pwr->lock); | |
5082 | } | |
5083 | EXPORT_SYMBOL_GPL(i915_request_power_well); | |
5084 | ||
5085 | /* Display audio driver power well release */ | |
5086 | void i915_release_power_well(void) | |
5087 | { | |
5088 | if (WARN_ON(!hsw_pwr)) | |
5089 | return; | |
5090 | ||
5091 | spin_lock_irq(&hsw_pwr->lock); | |
5092 | WARN_ON(!hsw_pwr->count); | |
5093 | if (!--hsw_pwr->count && | |
5094 | !hsw_pwr->i915_request) | |
5095 | __intel_set_power_well(hsw_pwr->device, false); | |
5096 | spin_unlock_irq(&hsw_pwr->lock); | |
5097 | } | |
5098 | EXPORT_SYMBOL_GPL(i915_release_power_well); | |
5099 | ||
5100 | int i915_init_power_well(struct drm_device *dev) | |
5101 | { | |
5102 | struct drm_i915_private *dev_priv = dev->dev_private; | |
5103 | ||
5104 | hsw_pwr = &dev_priv->power_well; | |
5105 | ||
5106 | hsw_pwr->device = dev; | |
5107 | spin_lock_init(&hsw_pwr->lock); | |
5108 | hsw_pwr->count = 0; | |
5109 | ||
5110 | return 0; | |
5111 | } | |
5112 | ||
5113 | void i915_remove_power_well(struct drm_device *dev) | |
5114 | { | |
5115 | hsw_pwr = NULL; | |
5116 | } | |
5117 | ||
5118 | void intel_set_power_well(struct drm_device *dev, bool enable) | |
5119 | { | |
5120 | struct drm_i915_private *dev_priv = dev->dev_private; | |
5121 | struct i915_power_well *power_well = &dev_priv->power_well; | |
5122 | ||
5123 | if (!HAS_POWER_WELL(dev)) | |
5124 | return; | |
5125 | ||
5126 | if (!i915_disable_power_well && !enable) | |
5127 | return; | |
5128 | ||
5129 | spin_lock_irq(&power_well->lock); | |
5130 | power_well->i915_request = enable; | |
5131 | ||
5132 | /* only reject "disable" power well request */ | |
5133 | if (power_well->count && !enable) { | |
5134 | spin_unlock_irq(&power_well->lock); | |
5135 | return; | |
5136 | } | |
5137 | ||
5138 | __intel_set_power_well(dev, enable); | |
5139 | spin_unlock_irq(&power_well->lock); | |
5140 | } | |
5141 | ||
fa42e23c PZ |
5142 | /* |
5143 | * Starting with Haswell, we have a "Power Down Well" that can be turned off | |
5144 | * when not needed anymore. We have 4 registers that can request the power well | |
5145 | * to be enabled, and it will only be disabled if none of the registers is | |
5146 | * requesting it to be enabled. | |
d0d3e513 | 5147 | */ |
fa42e23c | 5148 | void intel_init_power_well(struct drm_device *dev) |
d0d3e513 ED |
5149 | { |
5150 | struct drm_i915_private *dev_priv = dev->dev_private; | |
d0d3e513 | 5151 | |
86d52df6 | 5152 | if (!HAS_POWER_WELL(dev)) |
d0d3e513 ED |
5153 | return; |
5154 | ||
fa42e23c PZ |
5155 | /* For now, we need the power well to be always enabled. */ |
5156 | intel_set_power_well(dev, true); | |
d0d3e513 | 5157 | |
fa42e23c PZ |
5158 | /* We're taking over the BIOS, so clear any requests made by it since |
5159 | * the driver is in charge now. */ | |
5160 | if (I915_READ(HSW_PWR_WELL_BIOS) & HSW_PWR_WELL_ENABLE) | |
5161 | I915_WRITE(HSW_PWR_WELL_BIOS, 0); | |
d0d3e513 ED |
5162 | } |
5163 | ||
1fa61106 ED |
5164 | /* Set up chip specific power management-related functions */ |
5165 | void intel_init_pm(struct drm_device *dev) | |
5166 | { | |
5167 | struct drm_i915_private *dev_priv = dev->dev_private; | |
5168 | ||
5169 | if (I915_HAS_FBC(dev)) { | |
5170 | if (HAS_PCH_SPLIT(dev)) { | |
5171 | dev_priv->display.fbc_enabled = ironlake_fbc_enabled; | |
891348b2 | 5172 | if (IS_IVYBRIDGE(dev) || IS_HASWELL(dev)) |
abe959c7 RV |
5173 | dev_priv->display.enable_fbc = |
5174 | gen7_enable_fbc; | |
5175 | else | |
5176 | dev_priv->display.enable_fbc = | |
5177 | ironlake_enable_fbc; | |
1fa61106 ED |
5178 | dev_priv->display.disable_fbc = ironlake_disable_fbc; |
5179 | } else if (IS_GM45(dev)) { | |
5180 | dev_priv->display.fbc_enabled = g4x_fbc_enabled; | |
5181 | dev_priv->display.enable_fbc = g4x_enable_fbc; | |
5182 | dev_priv->display.disable_fbc = g4x_disable_fbc; | |
5183 | } else if (IS_CRESTLINE(dev)) { | |
5184 | dev_priv->display.fbc_enabled = i8xx_fbc_enabled; | |
5185 | dev_priv->display.enable_fbc = i8xx_enable_fbc; | |
5186 | dev_priv->display.disable_fbc = i8xx_disable_fbc; | |
5187 | } | |
5188 | /* 855GM needs testing */ | |
5189 | } | |
5190 | ||
c921aba8 DV |
5191 | /* For cxsr */ |
5192 | if (IS_PINEVIEW(dev)) | |
5193 | i915_pineview_get_mem_freq(dev); | |
5194 | else if (IS_GEN5(dev)) | |
5195 | i915_ironlake_get_mem_freq(dev); | |
5196 | ||
1fa61106 ED |
5197 | /* For FIFO watermark updates */ |
5198 | if (HAS_PCH_SPLIT(dev)) { | |
1fa61106 ED |
5199 | if (IS_GEN5(dev)) { |
5200 | if (I915_READ(MLTR_ILK) & ILK_SRLT_MASK) | |
5201 | dev_priv->display.update_wm = ironlake_update_wm; | |
5202 | else { | |
5203 | DRM_DEBUG_KMS("Failed to get proper latency. " | |
5204 | "Disable CxSR\n"); | |
5205 | dev_priv->display.update_wm = NULL; | |
5206 | } | |
5207 | dev_priv->display.init_clock_gating = ironlake_init_clock_gating; | |
5208 | } else if (IS_GEN6(dev)) { | |
5209 | if (SNB_READ_WM0_LATENCY()) { | |
5210 | dev_priv->display.update_wm = sandybridge_update_wm; | |
5211 | dev_priv->display.update_sprite_wm = sandybridge_update_sprite_wm; | |
5212 | } else { | |
5213 | DRM_DEBUG_KMS("Failed to read display plane latency. " | |
5214 | "Disable CxSR\n"); | |
5215 | dev_priv->display.update_wm = NULL; | |
5216 | } | |
5217 | dev_priv->display.init_clock_gating = gen6_init_clock_gating; | |
5218 | } else if (IS_IVYBRIDGE(dev)) { | |
1fa61106 | 5219 | if (SNB_READ_WM0_LATENCY()) { |
c43d0188 | 5220 | dev_priv->display.update_wm = ivybridge_update_wm; |
1fa61106 ED |
5221 | dev_priv->display.update_sprite_wm = sandybridge_update_sprite_wm; |
5222 | } else { | |
5223 | DRM_DEBUG_KMS("Failed to read display plane latency. " | |
5224 | "Disable CxSR\n"); | |
5225 | dev_priv->display.update_wm = NULL; | |
5226 | } | |
5227 | dev_priv->display.init_clock_gating = ivybridge_init_clock_gating; | |
6b8a5eeb | 5228 | } else if (IS_HASWELL(dev)) { |
3e1f7266 | 5229 | if (I915_READ64(MCH_SSKPD)) { |
1011d8c4 | 5230 | dev_priv->display.update_wm = haswell_update_wm; |
526682e9 PZ |
5231 | dev_priv->display.update_sprite_wm = |
5232 | haswell_update_sprite_wm; | |
6b8a5eeb ED |
5233 | } else { |
5234 | DRM_DEBUG_KMS("Failed to read display plane latency. " | |
5235 | "Disable CxSR\n"); | |
5236 | dev_priv->display.update_wm = NULL; | |
5237 | } | |
cad2a2d7 | 5238 | dev_priv->display.init_clock_gating = haswell_init_clock_gating; |
1fa61106 ED |
5239 | } else |
5240 | dev_priv->display.update_wm = NULL; | |
5241 | } else if (IS_VALLEYVIEW(dev)) { | |
5242 | dev_priv->display.update_wm = valleyview_update_wm; | |
5243 | dev_priv->display.init_clock_gating = | |
5244 | valleyview_init_clock_gating; | |
1fa61106 ED |
5245 | } else if (IS_PINEVIEW(dev)) { |
5246 | if (!intel_get_cxsr_latency(IS_PINEVIEW_G(dev), | |
5247 | dev_priv->is_ddr3, | |
5248 | dev_priv->fsb_freq, | |
5249 | dev_priv->mem_freq)) { | |
5250 | DRM_INFO("failed to find known CxSR latency " | |
5251 | "(found ddr%s fsb freq %d, mem freq %d), " | |
5252 | "disabling CxSR\n", | |
5253 | (dev_priv->is_ddr3 == 1) ? "3" : "2", | |
5254 | dev_priv->fsb_freq, dev_priv->mem_freq); | |
5255 | /* Disable CxSR and never update its watermark again */ | |
5256 | pineview_disable_cxsr(dev); | |
5257 | dev_priv->display.update_wm = NULL; | |
5258 | } else | |
5259 | dev_priv->display.update_wm = pineview_update_wm; | |
5260 | dev_priv->display.init_clock_gating = gen3_init_clock_gating; | |
5261 | } else if (IS_G4X(dev)) { | |
5262 | dev_priv->display.update_wm = g4x_update_wm; | |
5263 | dev_priv->display.init_clock_gating = g4x_init_clock_gating; | |
5264 | } else if (IS_GEN4(dev)) { | |
5265 | dev_priv->display.update_wm = i965_update_wm; | |
5266 | if (IS_CRESTLINE(dev)) | |
5267 | dev_priv->display.init_clock_gating = crestline_init_clock_gating; | |
5268 | else if (IS_BROADWATER(dev)) | |
5269 | dev_priv->display.init_clock_gating = broadwater_init_clock_gating; | |
5270 | } else if (IS_GEN3(dev)) { | |
5271 | dev_priv->display.update_wm = i9xx_update_wm; | |
5272 | dev_priv->display.get_fifo_size = i9xx_get_fifo_size; | |
5273 | dev_priv->display.init_clock_gating = gen3_init_clock_gating; | |
5274 | } else if (IS_I865G(dev)) { | |
5275 | dev_priv->display.update_wm = i830_update_wm; | |
5276 | dev_priv->display.init_clock_gating = i85x_init_clock_gating; | |
5277 | dev_priv->display.get_fifo_size = i830_get_fifo_size; | |
5278 | } else if (IS_I85X(dev)) { | |
5279 | dev_priv->display.update_wm = i9xx_update_wm; | |
5280 | dev_priv->display.get_fifo_size = i85x_get_fifo_size; | |
5281 | dev_priv->display.init_clock_gating = i85x_init_clock_gating; | |
5282 | } else { | |
5283 | dev_priv->display.update_wm = i830_update_wm; | |
5284 | dev_priv->display.init_clock_gating = i830_init_clock_gating; | |
5285 | if (IS_845G(dev)) | |
5286 | dev_priv->display.get_fifo_size = i845_get_fifo_size; | |
5287 | else | |
5288 | dev_priv->display.get_fifo_size = i830_get_fifo_size; | |
5289 | } | |
5290 | } | |
5291 | ||
6590190d ED |
5292 | static void __gen6_gt_wait_for_thread_c0(struct drm_i915_private *dev_priv) |
5293 | { | |
5294 | u32 gt_thread_status_mask; | |
5295 | ||
5296 | if (IS_HASWELL(dev_priv->dev)) | |
5297 | gt_thread_status_mask = GEN6_GT_THREAD_STATUS_CORE_MASK_HSW; | |
5298 | else | |
5299 | gt_thread_status_mask = GEN6_GT_THREAD_STATUS_CORE_MASK; | |
5300 | ||
5301 | /* w/a for a sporadic read returning 0 by waiting for the GT | |
5302 | * thread to wake up. | |
5303 | */ | |
5304 | if (wait_for_atomic_us((I915_READ_NOTRACE(GEN6_GT_THREAD_STATUS_REG) & gt_thread_status_mask) == 0, 500)) | |
5305 | DRM_ERROR("GT thread status wait timed out\n"); | |
5306 | } | |
5307 | ||
16995a9f CW |
5308 | static void __gen6_gt_force_wake_reset(struct drm_i915_private *dev_priv) |
5309 | { | |
5310 | I915_WRITE_NOTRACE(FORCEWAKE, 0); | |
5311 | POSTING_READ(ECOBUS); /* something from same cacheline, but !FORCEWAKE */ | |
5312 | } | |
5313 | ||
6590190d ED |
5314 | static void __gen6_gt_force_wake_get(struct drm_i915_private *dev_priv) |
5315 | { | |
ebd37ce1 | 5316 | if (wait_for_atomic((I915_READ_NOTRACE(FORCEWAKE_ACK) & 1) == 0, |
057d3860 | 5317 | FORCEWAKE_ACK_TIMEOUT_MS)) |
8a038fd6 | 5318 | DRM_ERROR("Timed out waiting for forcewake old ack to clear.\n"); |
6590190d | 5319 | |
30771e16 | 5320 | I915_WRITE_NOTRACE(FORCEWAKE, 1); |
8dee3eea | 5321 | POSTING_READ(ECOBUS); /* something from same cacheline, but !FORCEWAKE */ |
6590190d | 5322 | |
ebd37ce1 | 5323 | if (wait_for_atomic((I915_READ_NOTRACE(FORCEWAKE_ACK) & 1), |
057d3860 | 5324 | FORCEWAKE_ACK_TIMEOUT_MS)) |
8a038fd6 | 5325 | DRM_ERROR("Timed out waiting for forcewake to ack request.\n"); |
6590190d | 5326 | |
8693a824 | 5327 | /* WaRsForcewakeWaitTC0:snb */ |
6590190d ED |
5328 | __gen6_gt_wait_for_thread_c0(dev_priv); |
5329 | } | |
5330 | ||
16995a9f CW |
5331 | static void __gen6_gt_force_wake_mt_reset(struct drm_i915_private *dev_priv) |
5332 | { | |
5333 | I915_WRITE_NOTRACE(FORCEWAKE_MT, _MASKED_BIT_DISABLE(0xffff)); | |
b5144075 JN |
5334 | /* something from same cacheline, but !FORCEWAKE_MT */ |
5335 | POSTING_READ(ECOBUS); | |
16995a9f CW |
5336 | } |
5337 | ||
6590190d ED |
5338 | static void __gen6_gt_force_wake_mt_get(struct drm_i915_private *dev_priv) |
5339 | { | |
5340 | u32 forcewake_ack; | |
5341 | ||
5342 | if (IS_HASWELL(dev_priv->dev)) | |
5343 | forcewake_ack = FORCEWAKE_ACK_HSW; | |
5344 | else | |
5345 | forcewake_ack = FORCEWAKE_MT_ACK; | |
5346 | ||
83983c8b | 5347 | if (wait_for_atomic((I915_READ_NOTRACE(forcewake_ack) & FORCEWAKE_KERNEL) == 0, |
057d3860 | 5348 | FORCEWAKE_ACK_TIMEOUT_MS)) |
8a038fd6 | 5349 | DRM_ERROR("Timed out waiting for forcewake old ack to clear.\n"); |
6590190d | 5350 | |
c5836c27 | 5351 | I915_WRITE_NOTRACE(FORCEWAKE_MT, _MASKED_BIT_ENABLE(FORCEWAKE_KERNEL)); |
b5144075 JN |
5352 | /* something from same cacheline, but !FORCEWAKE_MT */ |
5353 | POSTING_READ(ECOBUS); | |
6590190d | 5354 | |
83983c8b | 5355 | if (wait_for_atomic((I915_READ_NOTRACE(forcewake_ack) & FORCEWAKE_KERNEL), |
057d3860 | 5356 | FORCEWAKE_ACK_TIMEOUT_MS)) |
8a038fd6 | 5357 | DRM_ERROR("Timed out waiting for forcewake to ack request.\n"); |
6590190d | 5358 | |
8693a824 | 5359 | /* WaRsForcewakeWaitTC0:ivb,hsw */ |
6590190d ED |
5360 | __gen6_gt_wait_for_thread_c0(dev_priv); |
5361 | } | |
5362 | ||
5363 | /* | |
5364 | * Generally this is called implicitly by the register read function. However, | |
5365 | * if some sequence requires the GT to not power down then this function should | |
5366 | * be called at the beginning of the sequence followed by a call to | |
5367 | * gen6_gt_force_wake_put() at the end of the sequence. | |
5368 | */ | |
5369 | void gen6_gt_force_wake_get(struct drm_i915_private *dev_priv) | |
5370 | { | |
5371 | unsigned long irqflags; | |
5372 | ||
5373 | spin_lock_irqsave(&dev_priv->gt_lock, irqflags); | |
5374 | if (dev_priv->forcewake_count++ == 0) | |
5375 | dev_priv->gt.force_wake_get(dev_priv); | |
5376 | spin_unlock_irqrestore(&dev_priv->gt_lock, irqflags); | |
5377 | } | |
5378 | ||
5379 | void gen6_gt_check_fifodbg(struct drm_i915_private *dev_priv) | |
5380 | { | |
5381 | u32 gtfifodbg; | |
5382 | gtfifodbg = I915_READ_NOTRACE(GTFIFODBG); | |
5383 | if (WARN(gtfifodbg & GT_FIFO_CPU_ERROR_MASK, | |
5384 | "MMIO read or write has been dropped %x\n", gtfifodbg)) | |
5385 | I915_WRITE_NOTRACE(GTFIFODBG, GT_FIFO_CPU_ERROR_MASK); | |
5386 | } | |
5387 | ||
5388 | static void __gen6_gt_force_wake_put(struct drm_i915_private *dev_priv) | |
5389 | { | |
5390 | I915_WRITE_NOTRACE(FORCEWAKE, 0); | |
b5144075 JN |
5391 | /* something from same cacheline, but !FORCEWAKE */ |
5392 | POSTING_READ(ECOBUS); | |
6590190d ED |
5393 | gen6_gt_check_fifodbg(dev_priv); |
5394 | } | |
5395 | ||
5396 | static void __gen6_gt_force_wake_mt_put(struct drm_i915_private *dev_priv) | |
5397 | { | |
c5836c27 | 5398 | I915_WRITE_NOTRACE(FORCEWAKE_MT, _MASKED_BIT_DISABLE(FORCEWAKE_KERNEL)); |
b5144075 JN |
5399 | /* something from same cacheline, but !FORCEWAKE_MT */ |
5400 | POSTING_READ(ECOBUS); | |
6590190d ED |
5401 | gen6_gt_check_fifodbg(dev_priv); |
5402 | } | |
5403 | ||
5404 | /* | |
5405 | * see gen6_gt_force_wake_get() | |
5406 | */ | |
5407 | void gen6_gt_force_wake_put(struct drm_i915_private *dev_priv) | |
5408 | { | |
5409 | unsigned long irqflags; | |
5410 | ||
5411 | spin_lock_irqsave(&dev_priv->gt_lock, irqflags); | |
5412 | if (--dev_priv->forcewake_count == 0) | |
5413 | dev_priv->gt.force_wake_put(dev_priv); | |
5414 | spin_unlock_irqrestore(&dev_priv->gt_lock, irqflags); | |
5415 | } | |
5416 | ||
5417 | int __gen6_gt_wait_for_fifo(struct drm_i915_private *dev_priv) | |
5418 | { | |
5419 | int ret = 0; | |
5420 | ||
5421 | if (dev_priv->gt_fifo_count < GT_FIFO_NUM_RESERVED_ENTRIES) { | |
5422 | int loop = 500; | |
5423 | u32 fifo = I915_READ_NOTRACE(GT_FIFO_FREE_ENTRIES); | |
5424 | while (fifo <= GT_FIFO_NUM_RESERVED_ENTRIES && loop--) { | |
5425 | udelay(10); | |
5426 | fifo = I915_READ_NOTRACE(GT_FIFO_FREE_ENTRIES); | |
5427 | } | |
5428 | if (WARN_ON(loop < 0 && fifo <= GT_FIFO_NUM_RESERVED_ENTRIES)) | |
5429 | ++ret; | |
5430 | dev_priv->gt_fifo_count = fifo; | |
5431 | } | |
5432 | dev_priv->gt_fifo_count--; | |
5433 | ||
5434 | return ret; | |
5435 | } | |
5436 | ||
16995a9f CW |
5437 | static void vlv_force_wake_reset(struct drm_i915_private *dev_priv) |
5438 | { | |
5439 | I915_WRITE_NOTRACE(FORCEWAKE_VLV, _MASKED_BIT_DISABLE(0xffff)); | |
b5144075 JN |
5440 | /* something from same cacheline, but !FORCEWAKE_VLV */ |
5441 | POSTING_READ(FORCEWAKE_ACK_VLV); | |
16995a9f CW |
5442 | } |
5443 | ||
6590190d ED |
5444 | static void vlv_force_wake_get(struct drm_i915_private *dev_priv) |
5445 | { | |
83983c8b | 5446 | if (wait_for_atomic((I915_READ_NOTRACE(FORCEWAKE_ACK_VLV) & FORCEWAKE_KERNEL) == 0, |
057d3860 | 5447 | FORCEWAKE_ACK_TIMEOUT_MS)) |
8a038fd6 | 5448 | DRM_ERROR("Timed out waiting for forcewake old ack to clear.\n"); |
6590190d | 5449 | |
c5836c27 | 5450 | I915_WRITE_NOTRACE(FORCEWAKE_VLV, _MASKED_BIT_ENABLE(FORCEWAKE_KERNEL)); |
ed5de399 JB |
5451 | I915_WRITE_NOTRACE(FORCEWAKE_MEDIA_VLV, |
5452 | _MASKED_BIT_ENABLE(FORCEWAKE_KERNEL)); | |
6590190d | 5453 | |
83983c8b | 5454 | if (wait_for_atomic((I915_READ_NOTRACE(FORCEWAKE_ACK_VLV) & FORCEWAKE_KERNEL), |
057d3860 | 5455 | FORCEWAKE_ACK_TIMEOUT_MS)) |
ed5de399 JB |
5456 | DRM_ERROR("Timed out waiting for GT to ack forcewake request.\n"); |
5457 | ||
5458 | if (wait_for_atomic((I915_READ_NOTRACE(FORCEWAKE_ACK_MEDIA_VLV) & | |
5459 | FORCEWAKE_KERNEL), | |
5460 | FORCEWAKE_ACK_TIMEOUT_MS)) | |
5461 | DRM_ERROR("Timed out waiting for media to ack forcewake request.\n"); | |
6590190d | 5462 | |
8693a824 | 5463 | /* WaRsForcewakeWaitTC0:vlv */ |
6590190d ED |
5464 | __gen6_gt_wait_for_thread_c0(dev_priv); |
5465 | } | |
5466 | ||
5467 | static void vlv_force_wake_put(struct drm_i915_private *dev_priv) | |
5468 | { | |
c5836c27 | 5469 | I915_WRITE_NOTRACE(FORCEWAKE_VLV, _MASKED_BIT_DISABLE(FORCEWAKE_KERNEL)); |
ed5de399 JB |
5470 | I915_WRITE_NOTRACE(FORCEWAKE_MEDIA_VLV, |
5471 | _MASKED_BIT_DISABLE(FORCEWAKE_KERNEL)); | |
5472 | /* The below doubles as a POSTING_READ */ | |
5ab140a4 | 5473 | gen6_gt_check_fifodbg(dev_priv); |
6590190d ED |
5474 | } |
5475 | ||
181d1b9e | 5476 | void intel_gt_sanitize(struct drm_device *dev) |
16995a9f CW |
5477 | { |
5478 | struct drm_i915_private *dev_priv = dev->dev_private; | |
5479 | ||
5480 | if (IS_VALLEYVIEW(dev)) { | |
5481 | vlv_force_wake_reset(dev_priv); | |
5482 | } else if (INTEL_INFO(dev)->gen >= 6) { | |
5483 | __gen6_gt_force_wake_reset(dev_priv); | |
5484 | if (IS_IVYBRIDGE(dev) || IS_HASWELL(dev)) | |
5485 | __gen6_gt_force_wake_mt_reset(dev_priv); | |
5486 | } | |
7dcd2677 KK |
5487 | |
5488 | /* BIOS often leaves RC6 enabled, but disable it for hw init */ | |
181d1b9e DV |
5489 | if (INTEL_INFO(dev)->gen >= 6) |
5490 | intel_disable_gt_powersave(dev); | |
16995a9f CW |
5491 | } |
5492 | ||
6590190d ED |
5493 | void intel_gt_init(struct drm_device *dev) |
5494 | { | |
5495 | struct drm_i915_private *dev_priv = dev->dev_private; | |
5496 | ||
6590190d ED |
5497 | if (IS_VALLEYVIEW(dev)) { |
5498 | dev_priv->gt.force_wake_get = vlv_force_wake_get; | |
5499 | dev_priv->gt.force_wake_put = vlv_force_wake_put; | |
c11e5f35 | 5500 | } else if (IS_HASWELL(dev)) { |
36ec8f87 DV |
5501 | dev_priv->gt.force_wake_get = __gen6_gt_force_wake_mt_get; |
5502 | dev_priv->gt.force_wake_put = __gen6_gt_force_wake_mt_put; | |
c11e5f35 GR |
5503 | } else if (IS_IVYBRIDGE(dev)) { |
5504 | u32 ecobus; | |
5505 | ||
5506 | /* IVB configs may use multi-threaded forcewake */ | |
5507 | ||
5508 | /* A small trick here - if the bios hasn't configured | |
5509 | * MT forcewake, and if the device is in RC6, then | |
5510 | * force_wake_mt_get will not wake the device and the | |
5511 | * ECOBUS read will return zero. Which will be | |
5512 | * (correctly) interpreted by the test below as MT | |
5513 | * forcewake being disabled. | |
5514 | */ | |
5515 | mutex_lock(&dev->struct_mutex); | |
5516 | __gen6_gt_force_wake_mt_get(dev_priv); | |
5517 | ecobus = I915_READ_NOTRACE(ECOBUS); | |
5518 | __gen6_gt_force_wake_mt_put(dev_priv); | |
5519 | mutex_unlock(&dev->struct_mutex); | |
5520 | ||
5521 | if (ecobus & FORCEWAKE_MT_ENABLE) { | |
5522 | dev_priv->gt.force_wake_get = | |
5523 | __gen6_gt_force_wake_mt_get; | |
5524 | dev_priv->gt.force_wake_put = | |
5525 | __gen6_gt_force_wake_mt_put; | |
5526 | } else { | |
5527 | DRM_INFO("No MT forcewake available on Ivybridge, this can result in issues\n"); | |
5528 | DRM_INFO("when using vblank-synced partial screen updates.\n"); | |
5529 | dev_priv->gt.force_wake_get = __gen6_gt_force_wake_get; | |
5530 | dev_priv->gt.force_wake_put = __gen6_gt_force_wake_put; | |
5531 | } | |
36ec8f87 | 5532 | } else if (IS_GEN6(dev)) { |
6590190d ED |
5533 | dev_priv->gt.force_wake_get = __gen6_gt_force_wake_get; |
5534 | dev_priv->gt.force_wake_put = __gen6_gt_force_wake_put; | |
6590190d | 5535 | } |
1a01ab3b JB |
5536 | INIT_DELAYED_WORK(&dev_priv->rps.delayed_resume_work, |
5537 | intel_gen6_powersave_work); | |
6590190d ED |
5538 | } |
5539 | ||
42c0526c BW |
5540 | int sandybridge_pcode_read(struct drm_i915_private *dev_priv, u8 mbox, u32 *val) |
5541 | { | |
4fc688ce | 5542 | WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock)); |
42c0526c BW |
5543 | |
5544 | if (I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) { | |
5545 | DRM_DEBUG_DRIVER("warning: pcode (read) mailbox access failed\n"); | |
5546 | return -EAGAIN; | |
5547 | } | |
5548 | ||
5549 | I915_WRITE(GEN6_PCODE_DATA, *val); | |
5550 | I915_WRITE(GEN6_PCODE_MAILBOX, GEN6_PCODE_READY | mbox); | |
5551 | ||
5552 | if (wait_for((I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) == 0, | |
5553 | 500)) { | |
5554 | DRM_ERROR("timeout waiting for pcode read (%d) to finish\n", mbox); | |
5555 | return -ETIMEDOUT; | |
5556 | } | |
5557 | ||
5558 | *val = I915_READ(GEN6_PCODE_DATA); | |
5559 | I915_WRITE(GEN6_PCODE_DATA, 0); | |
5560 | ||
5561 | return 0; | |
5562 | } | |
5563 | ||
5564 | int sandybridge_pcode_write(struct drm_i915_private *dev_priv, u8 mbox, u32 val) | |
5565 | { | |
4fc688ce | 5566 | WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock)); |
42c0526c BW |
5567 | |
5568 | if (I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) { | |
5569 | DRM_DEBUG_DRIVER("warning: pcode (write) mailbox access failed\n"); | |
5570 | return -EAGAIN; | |
5571 | } | |
5572 | ||
5573 | I915_WRITE(GEN6_PCODE_DATA, val); | |
5574 | I915_WRITE(GEN6_PCODE_MAILBOX, GEN6_PCODE_READY | mbox); | |
5575 | ||
5576 | if (wait_for((I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) == 0, | |
5577 | 500)) { | |
5578 | DRM_ERROR("timeout waiting for pcode write (%d) to finish\n", mbox); | |
5579 | return -ETIMEDOUT; | |
5580 | } | |
5581 | ||
5582 | I915_WRITE(GEN6_PCODE_DATA, 0); | |
5583 | ||
5584 | return 0; | |
5585 | } | |
a0e4e199 | 5586 | |
855ba3be JB |
5587 | int vlv_gpu_freq(int ddr_freq, int val) |
5588 | { | |
5589 | int mult, base; | |
5590 | ||
5591 | switch (ddr_freq) { | |
5592 | case 800: | |
5593 | mult = 20; | |
5594 | base = 120; | |
5595 | break; | |
5596 | case 1066: | |
5597 | mult = 22; | |
5598 | base = 133; | |
5599 | break; | |
5600 | case 1333: | |
5601 | mult = 21; | |
5602 | base = 125; | |
5603 | break; | |
5604 | default: | |
5605 | return -1; | |
5606 | } | |
5607 | ||
5608 | return ((val - 0xbd) * mult) + base; | |
5609 | } | |
5610 | ||
5611 | int vlv_freq_opcode(int ddr_freq, int val) | |
5612 | { | |
5613 | int mult, base; | |
5614 | ||
5615 | switch (ddr_freq) { | |
5616 | case 800: | |
5617 | mult = 20; | |
5618 | base = 120; | |
5619 | break; | |
5620 | case 1066: | |
5621 | mult = 22; | |
5622 | base = 133; | |
5623 | break; | |
5624 | case 1333: | |
5625 | mult = 21; | |
5626 | base = 125; | |
5627 | break; | |
5628 | default: | |
5629 | return -1; | |
5630 | } | |
5631 | ||
5632 | val /= mult; | |
5633 | val -= base / mult; | |
5634 | val += 0xbd; | |
5635 | ||
5636 | if (val > 0xea) | |
5637 | val = 0xea; | |
5638 | ||
5639 | return val; | |
5640 | } | |
5641 |