]> git.ipfire.org Git - thirdparty/kernel/stable.git/blob - drivers/gpu/drm/i915/intel_device_info.c
Merge branch 'drm-next-5.1' of git://people.freedesktop.org/~agd5f/linux into drm...
[thirdparty/kernel/stable.git] / drivers / gpu / drm / i915 / intel_device_info.c
1 /*
2 * Copyright © 2016 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 */
24
25 #include <drm/drm_print.h>
26
27 #include "intel_device_info.h"
28 #include "i915_drv.h"
29
30 #define PLATFORM_NAME(x) [INTEL_##x] = #x
31 static const char * const platform_names[] = {
32 PLATFORM_NAME(I830),
33 PLATFORM_NAME(I845G),
34 PLATFORM_NAME(I85X),
35 PLATFORM_NAME(I865G),
36 PLATFORM_NAME(I915G),
37 PLATFORM_NAME(I915GM),
38 PLATFORM_NAME(I945G),
39 PLATFORM_NAME(I945GM),
40 PLATFORM_NAME(G33),
41 PLATFORM_NAME(PINEVIEW),
42 PLATFORM_NAME(I965G),
43 PLATFORM_NAME(I965GM),
44 PLATFORM_NAME(G45),
45 PLATFORM_NAME(GM45),
46 PLATFORM_NAME(IRONLAKE),
47 PLATFORM_NAME(SANDYBRIDGE),
48 PLATFORM_NAME(IVYBRIDGE),
49 PLATFORM_NAME(VALLEYVIEW),
50 PLATFORM_NAME(HASWELL),
51 PLATFORM_NAME(BROADWELL),
52 PLATFORM_NAME(CHERRYVIEW),
53 PLATFORM_NAME(SKYLAKE),
54 PLATFORM_NAME(BROXTON),
55 PLATFORM_NAME(KABYLAKE),
56 PLATFORM_NAME(GEMINILAKE),
57 PLATFORM_NAME(COFFEELAKE),
58 PLATFORM_NAME(CANNONLAKE),
59 PLATFORM_NAME(ICELAKE),
60 };
61 #undef PLATFORM_NAME
62
63 const char *intel_platform_name(enum intel_platform platform)
64 {
65 BUILD_BUG_ON(ARRAY_SIZE(platform_names) != INTEL_MAX_PLATFORMS);
66
67 if (WARN_ON_ONCE(platform >= ARRAY_SIZE(platform_names) ||
68 platform_names[platform] == NULL))
69 return "<unknown>";
70
71 return platform_names[platform];
72 }
73
74 void intel_device_info_dump_flags(const struct intel_device_info *info,
75 struct drm_printer *p)
76 {
77 #define PRINT_FLAG(name) drm_printf(p, "%s: %s\n", #name, yesno(info->name));
78 DEV_INFO_FOR_EACH_FLAG(PRINT_FLAG);
79 #undef PRINT_FLAG
80
81 #define PRINT_FLAG(name) drm_printf(p, "%s: %s\n", #name, yesno(info->display.name));
82 DEV_INFO_DISPLAY_FOR_EACH_FLAG(PRINT_FLAG);
83 #undef PRINT_FLAG
84 }
85
86 static void sseu_dump(const struct sseu_dev_info *sseu, struct drm_printer *p)
87 {
88 int s;
89
90 drm_printf(p, "slice total: %u, mask=%04x\n",
91 hweight8(sseu->slice_mask), sseu->slice_mask);
92 drm_printf(p, "subslice total: %u\n", sseu_subslice_total(sseu));
93 for (s = 0; s < sseu->max_slices; s++) {
94 drm_printf(p, "slice%d: %u subslices, mask=%04x\n",
95 s, hweight8(sseu->subslice_mask[s]),
96 sseu->subslice_mask[s]);
97 }
98 drm_printf(p, "EU total: %u\n", sseu->eu_total);
99 drm_printf(p, "EU per subslice: %u\n", sseu->eu_per_subslice);
100 drm_printf(p, "has slice power gating: %s\n",
101 yesno(sseu->has_slice_pg));
102 drm_printf(p, "has subslice power gating: %s\n",
103 yesno(sseu->has_subslice_pg));
104 drm_printf(p, "has EU power gating: %s\n", yesno(sseu->has_eu_pg));
105 }
106
107 void intel_device_info_dump_runtime(const struct intel_runtime_info *info,
108 struct drm_printer *p)
109 {
110 sseu_dump(&info->sseu, p);
111
112 drm_printf(p, "CS timestamp frequency: %u kHz\n",
113 info->cs_timestamp_frequency_khz);
114 }
115
116 void intel_device_info_dump_topology(const struct sseu_dev_info *sseu,
117 struct drm_printer *p)
118 {
119 int s, ss;
120
121 if (sseu->max_slices == 0) {
122 drm_printf(p, "Unavailable\n");
123 return;
124 }
125
126 for (s = 0; s < sseu->max_slices; s++) {
127 drm_printf(p, "slice%d: %u subslice(s) (0x%hhx):\n",
128 s, hweight8(sseu->subslice_mask[s]),
129 sseu->subslice_mask[s]);
130
131 for (ss = 0; ss < sseu->max_subslices; ss++) {
132 u16 enabled_eus = sseu_get_eus(sseu, s, ss);
133
134 drm_printf(p, "\tsubslice%d: %u EUs (0x%hx)\n",
135 ss, hweight16(enabled_eus), enabled_eus);
136 }
137 }
138 }
139
140 static u16 compute_eu_total(const struct sseu_dev_info *sseu)
141 {
142 u16 i, total = 0;
143
144 for (i = 0; i < ARRAY_SIZE(sseu->eu_mask); i++)
145 total += hweight8(sseu->eu_mask[i]);
146
147 return total;
148 }
149
150 static void gen11_sseu_info_init(struct drm_i915_private *dev_priv)
151 {
152 struct sseu_dev_info *sseu = &RUNTIME_INFO(dev_priv)->sseu;
153 u8 s_en;
154 u32 ss_en, ss_en_mask;
155 u8 eu_en;
156 int s;
157
158 sseu->max_slices = 1;
159 sseu->max_subslices = 8;
160 sseu->max_eus_per_subslice = 8;
161
162 s_en = I915_READ(GEN11_GT_SLICE_ENABLE) & GEN11_GT_S_ENA_MASK;
163 ss_en = ~I915_READ(GEN11_GT_SUBSLICE_DISABLE);
164 ss_en_mask = BIT(sseu->max_subslices) - 1;
165 eu_en = ~(I915_READ(GEN11_EU_DISABLE) & GEN11_EU_DIS_MASK);
166
167 for (s = 0; s < sseu->max_slices; s++) {
168 if (s_en & BIT(s)) {
169 int ss_idx = sseu->max_subslices * s;
170 int ss;
171
172 sseu->slice_mask |= BIT(s);
173 sseu->subslice_mask[s] = (ss_en >> ss_idx) & ss_en_mask;
174 for (ss = 0; ss < sseu->max_subslices; ss++) {
175 if (sseu->subslice_mask[s] & BIT(ss))
176 sseu_set_eus(sseu, s, ss, eu_en);
177 }
178 }
179 }
180 sseu->eu_per_subslice = hweight8(eu_en);
181 sseu->eu_total = compute_eu_total(sseu);
182
183 /* ICL has no power gating restrictions. */
184 sseu->has_slice_pg = 1;
185 sseu->has_subslice_pg = 1;
186 sseu->has_eu_pg = 1;
187 }
188
189 static void gen10_sseu_info_init(struct drm_i915_private *dev_priv)
190 {
191 struct sseu_dev_info *sseu = &RUNTIME_INFO(dev_priv)->sseu;
192 const u32 fuse2 = I915_READ(GEN8_FUSE2);
193 int s, ss;
194 const int eu_mask = 0xff;
195 u32 subslice_mask, eu_en;
196
197 sseu->slice_mask = (fuse2 & GEN10_F2_S_ENA_MASK) >>
198 GEN10_F2_S_ENA_SHIFT;
199 sseu->max_slices = 6;
200 sseu->max_subslices = 4;
201 sseu->max_eus_per_subslice = 8;
202
203 subslice_mask = (1 << 4) - 1;
204 subslice_mask &= ~((fuse2 & GEN10_F2_SS_DIS_MASK) >>
205 GEN10_F2_SS_DIS_SHIFT);
206
207 /*
208 * Slice0 can have up to 3 subslices, but there are only 2 in
209 * slice1/2.
210 */
211 sseu->subslice_mask[0] = subslice_mask;
212 for (s = 1; s < sseu->max_slices; s++)
213 sseu->subslice_mask[s] = subslice_mask & 0x3;
214
215 /* Slice0 */
216 eu_en = ~I915_READ(GEN8_EU_DISABLE0);
217 for (ss = 0; ss < sseu->max_subslices; ss++)
218 sseu_set_eus(sseu, 0, ss, (eu_en >> (8 * ss)) & eu_mask);
219 /* Slice1 */
220 sseu_set_eus(sseu, 1, 0, (eu_en >> 24) & eu_mask);
221 eu_en = ~I915_READ(GEN8_EU_DISABLE1);
222 sseu_set_eus(sseu, 1, 1, eu_en & eu_mask);
223 /* Slice2 */
224 sseu_set_eus(sseu, 2, 0, (eu_en >> 8) & eu_mask);
225 sseu_set_eus(sseu, 2, 1, (eu_en >> 16) & eu_mask);
226 /* Slice3 */
227 sseu_set_eus(sseu, 3, 0, (eu_en >> 24) & eu_mask);
228 eu_en = ~I915_READ(GEN8_EU_DISABLE2);
229 sseu_set_eus(sseu, 3, 1, eu_en & eu_mask);
230 /* Slice4 */
231 sseu_set_eus(sseu, 4, 0, (eu_en >> 8) & eu_mask);
232 sseu_set_eus(sseu, 4, 1, (eu_en >> 16) & eu_mask);
233 /* Slice5 */
234 sseu_set_eus(sseu, 5, 0, (eu_en >> 24) & eu_mask);
235 eu_en = ~I915_READ(GEN10_EU_DISABLE3);
236 sseu_set_eus(sseu, 5, 1, eu_en & eu_mask);
237
238 /* Do a second pass where we mark the subslices disabled if all their
239 * eus are off.
240 */
241 for (s = 0; s < sseu->max_slices; s++) {
242 for (ss = 0; ss < sseu->max_subslices; ss++) {
243 if (sseu_get_eus(sseu, s, ss) == 0)
244 sseu->subslice_mask[s] &= ~BIT(ss);
245 }
246 }
247
248 sseu->eu_total = compute_eu_total(sseu);
249
250 /*
251 * CNL is expected to always have a uniform distribution
252 * of EU across subslices with the exception that any one
253 * EU in any one subslice may be fused off for die
254 * recovery.
255 */
256 sseu->eu_per_subslice = sseu_subslice_total(sseu) ?
257 DIV_ROUND_UP(sseu->eu_total,
258 sseu_subslice_total(sseu)) : 0;
259
260 /* No restrictions on Power Gating */
261 sseu->has_slice_pg = 1;
262 sseu->has_subslice_pg = 1;
263 sseu->has_eu_pg = 1;
264 }
265
266 static void cherryview_sseu_info_init(struct drm_i915_private *dev_priv)
267 {
268 struct sseu_dev_info *sseu = &RUNTIME_INFO(dev_priv)->sseu;
269 u32 fuse;
270
271 fuse = I915_READ(CHV_FUSE_GT);
272
273 sseu->slice_mask = BIT(0);
274 sseu->max_slices = 1;
275 sseu->max_subslices = 2;
276 sseu->max_eus_per_subslice = 8;
277
278 if (!(fuse & CHV_FGT_DISABLE_SS0)) {
279 u8 disabled_mask =
280 ((fuse & CHV_FGT_EU_DIS_SS0_R0_MASK) >>
281 CHV_FGT_EU_DIS_SS0_R0_SHIFT) |
282 (((fuse & CHV_FGT_EU_DIS_SS0_R1_MASK) >>
283 CHV_FGT_EU_DIS_SS0_R1_SHIFT) << 4);
284
285 sseu->subslice_mask[0] |= BIT(0);
286 sseu_set_eus(sseu, 0, 0, ~disabled_mask);
287 }
288
289 if (!(fuse & CHV_FGT_DISABLE_SS1)) {
290 u8 disabled_mask =
291 ((fuse & CHV_FGT_EU_DIS_SS1_R0_MASK) >>
292 CHV_FGT_EU_DIS_SS1_R0_SHIFT) |
293 (((fuse & CHV_FGT_EU_DIS_SS1_R1_MASK) >>
294 CHV_FGT_EU_DIS_SS1_R1_SHIFT) << 4);
295
296 sseu->subslice_mask[0] |= BIT(1);
297 sseu_set_eus(sseu, 0, 1, ~disabled_mask);
298 }
299
300 sseu->eu_total = compute_eu_total(sseu);
301
302 /*
303 * CHV expected to always have a uniform distribution of EU
304 * across subslices.
305 */
306 sseu->eu_per_subslice = sseu_subslice_total(sseu) ?
307 sseu->eu_total / sseu_subslice_total(sseu) :
308 0;
309 /*
310 * CHV supports subslice power gating on devices with more than
311 * one subslice, and supports EU power gating on devices with
312 * more than one EU pair per subslice.
313 */
314 sseu->has_slice_pg = 0;
315 sseu->has_subslice_pg = sseu_subslice_total(sseu) > 1;
316 sseu->has_eu_pg = (sseu->eu_per_subslice > 2);
317 }
318
319 static void gen9_sseu_info_init(struct drm_i915_private *dev_priv)
320 {
321 struct intel_device_info *info = mkwrite_device_info(dev_priv);
322 struct sseu_dev_info *sseu = &RUNTIME_INFO(dev_priv)->sseu;
323 int s, ss;
324 u32 fuse2, eu_disable, subslice_mask;
325 const u8 eu_mask = 0xff;
326
327 fuse2 = I915_READ(GEN8_FUSE2);
328 sseu->slice_mask = (fuse2 & GEN8_F2_S_ENA_MASK) >> GEN8_F2_S_ENA_SHIFT;
329
330 /* BXT has a single slice and at most 3 subslices. */
331 sseu->max_slices = IS_GEN9_LP(dev_priv) ? 1 : 3;
332 sseu->max_subslices = IS_GEN9_LP(dev_priv) ? 3 : 4;
333 sseu->max_eus_per_subslice = 8;
334
335 /*
336 * The subslice disable field is global, i.e. it applies
337 * to each of the enabled slices.
338 */
339 subslice_mask = (1 << sseu->max_subslices) - 1;
340 subslice_mask &= ~((fuse2 & GEN9_F2_SS_DIS_MASK) >>
341 GEN9_F2_SS_DIS_SHIFT);
342
343 /*
344 * Iterate through enabled slices and subslices to
345 * count the total enabled EU.
346 */
347 for (s = 0; s < sseu->max_slices; s++) {
348 if (!(sseu->slice_mask & BIT(s)))
349 /* skip disabled slice */
350 continue;
351
352 sseu->subslice_mask[s] = subslice_mask;
353
354 eu_disable = I915_READ(GEN9_EU_DISABLE(s));
355 for (ss = 0; ss < sseu->max_subslices; ss++) {
356 int eu_per_ss;
357 u8 eu_disabled_mask;
358
359 if (!(sseu->subslice_mask[s] & BIT(ss)))
360 /* skip disabled subslice */
361 continue;
362
363 eu_disabled_mask = (eu_disable >> (ss * 8)) & eu_mask;
364
365 sseu_set_eus(sseu, s, ss, ~eu_disabled_mask);
366
367 eu_per_ss = sseu->max_eus_per_subslice -
368 hweight8(eu_disabled_mask);
369
370 /*
371 * Record which subslice(s) has(have) 7 EUs. we
372 * can tune the hash used to spread work among
373 * subslices if they are unbalanced.
374 */
375 if (eu_per_ss == 7)
376 sseu->subslice_7eu[s] |= BIT(ss);
377 }
378 }
379
380 sseu->eu_total = compute_eu_total(sseu);
381
382 /*
383 * SKL is expected to always have a uniform distribution
384 * of EU across subslices with the exception that any one
385 * EU in any one subslice may be fused off for die
386 * recovery. BXT is expected to be perfectly uniform in EU
387 * distribution.
388 */
389 sseu->eu_per_subslice = sseu_subslice_total(sseu) ?
390 DIV_ROUND_UP(sseu->eu_total,
391 sseu_subslice_total(sseu)) : 0;
392 /*
393 * SKL+ supports slice power gating on devices with more than
394 * one slice, and supports EU power gating on devices with
395 * more than one EU pair per subslice. BXT+ supports subslice
396 * power gating on devices with more than one subslice, and
397 * supports EU power gating on devices with more than one EU
398 * pair per subslice.
399 */
400 sseu->has_slice_pg =
401 !IS_GEN9_LP(dev_priv) && hweight8(sseu->slice_mask) > 1;
402 sseu->has_subslice_pg =
403 IS_GEN9_LP(dev_priv) && sseu_subslice_total(sseu) > 1;
404 sseu->has_eu_pg = sseu->eu_per_subslice > 2;
405
406 if (IS_GEN9_LP(dev_priv)) {
407 #define IS_SS_DISABLED(ss) (!(sseu->subslice_mask[0] & BIT(ss)))
408 info->has_pooled_eu = hweight8(sseu->subslice_mask[0]) == 3;
409
410 sseu->min_eu_in_pool = 0;
411 if (info->has_pooled_eu) {
412 if (IS_SS_DISABLED(2) || IS_SS_DISABLED(0))
413 sseu->min_eu_in_pool = 3;
414 else if (IS_SS_DISABLED(1))
415 sseu->min_eu_in_pool = 6;
416 else
417 sseu->min_eu_in_pool = 9;
418 }
419 #undef IS_SS_DISABLED
420 }
421 }
422
423 static void broadwell_sseu_info_init(struct drm_i915_private *dev_priv)
424 {
425 struct sseu_dev_info *sseu = &RUNTIME_INFO(dev_priv)->sseu;
426 int s, ss;
427 u32 fuse2, subslice_mask, eu_disable[3]; /* s_max */
428
429 fuse2 = I915_READ(GEN8_FUSE2);
430 sseu->slice_mask = (fuse2 & GEN8_F2_S_ENA_MASK) >> GEN8_F2_S_ENA_SHIFT;
431 sseu->max_slices = 3;
432 sseu->max_subslices = 3;
433 sseu->max_eus_per_subslice = 8;
434
435 /*
436 * The subslice disable field is global, i.e. it applies
437 * to each of the enabled slices.
438 */
439 subslice_mask = GENMASK(sseu->max_subslices - 1, 0);
440 subslice_mask &= ~((fuse2 & GEN8_F2_SS_DIS_MASK) >>
441 GEN8_F2_SS_DIS_SHIFT);
442
443 eu_disable[0] = I915_READ(GEN8_EU_DISABLE0) & GEN8_EU_DIS0_S0_MASK;
444 eu_disable[1] = (I915_READ(GEN8_EU_DISABLE0) >> GEN8_EU_DIS0_S1_SHIFT) |
445 ((I915_READ(GEN8_EU_DISABLE1) & GEN8_EU_DIS1_S1_MASK) <<
446 (32 - GEN8_EU_DIS0_S1_SHIFT));
447 eu_disable[2] = (I915_READ(GEN8_EU_DISABLE1) >> GEN8_EU_DIS1_S2_SHIFT) |
448 ((I915_READ(GEN8_EU_DISABLE2) & GEN8_EU_DIS2_S2_MASK) <<
449 (32 - GEN8_EU_DIS1_S2_SHIFT));
450
451 /*
452 * Iterate through enabled slices and subslices to
453 * count the total enabled EU.
454 */
455 for (s = 0; s < sseu->max_slices; s++) {
456 if (!(sseu->slice_mask & BIT(s)))
457 /* skip disabled slice */
458 continue;
459
460 sseu->subslice_mask[s] = subslice_mask;
461
462 for (ss = 0; ss < sseu->max_subslices; ss++) {
463 u8 eu_disabled_mask;
464 u32 n_disabled;
465
466 if (!(sseu->subslice_mask[s] & BIT(ss)))
467 /* skip disabled subslice */
468 continue;
469
470 eu_disabled_mask =
471 eu_disable[s] >> (ss * sseu->max_eus_per_subslice);
472
473 sseu_set_eus(sseu, s, ss, ~eu_disabled_mask);
474
475 n_disabled = hweight8(eu_disabled_mask);
476
477 /*
478 * Record which subslices have 7 EUs.
479 */
480 if (sseu->max_eus_per_subslice - n_disabled == 7)
481 sseu->subslice_7eu[s] |= 1 << ss;
482 }
483 }
484
485 sseu->eu_total = compute_eu_total(sseu);
486
487 /*
488 * BDW is expected to always have a uniform distribution of EU across
489 * subslices with the exception that any one EU in any one subslice may
490 * be fused off for die recovery.
491 */
492 sseu->eu_per_subslice = sseu_subslice_total(sseu) ?
493 DIV_ROUND_UP(sseu->eu_total,
494 sseu_subslice_total(sseu)) : 0;
495
496 /*
497 * BDW supports slice power gating on devices with more than
498 * one slice.
499 */
500 sseu->has_slice_pg = hweight8(sseu->slice_mask) > 1;
501 sseu->has_subslice_pg = 0;
502 sseu->has_eu_pg = 0;
503 }
504
505 static void haswell_sseu_info_init(struct drm_i915_private *dev_priv)
506 {
507 struct sseu_dev_info *sseu = &RUNTIME_INFO(dev_priv)->sseu;
508 u32 fuse1;
509 int s, ss;
510
511 /*
512 * There isn't a register to tell us how many slices/subslices. We
513 * work off the PCI-ids here.
514 */
515 switch (INTEL_INFO(dev_priv)->gt) {
516 default:
517 MISSING_CASE(INTEL_INFO(dev_priv)->gt);
518 /* fall through */
519 case 1:
520 sseu->slice_mask = BIT(0);
521 sseu->subslice_mask[0] = BIT(0);
522 break;
523 case 2:
524 sseu->slice_mask = BIT(0);
525 sseu->subslice_mask[0] = BIT(0) | BIT(1);
526 break;
527 case 3:
528 sseu->slice_mask = BIT(0) | BIT(1);
529 sseu->subslice_mask[0] = BIT(0) | BIT(1);
530 sseu->subslice_mask[1] = BIT(0) | BIT(1);
531 break;
532 }
533
534 sseu->max_slices = hweight8(sseu->slice_mask);
535 sseu->max_subslices = hweight8(sseu->subslice_mask[0]);
536
537 fuse1 = I915_READ(HSW_PAVP_FUSE1);
538 switch ((fuse1 & HSW_F1_EU_DIS_MASK) >> HSW_F1_EU_DIS_SHIFT) {
539 default:
540 MISSING_CASE((fuse1 & HSW_F1_EU_DIS_MASK) >>
541 HSW_F1_EU_DIS_SHIFT);
542 /* fall through */
543 case HSW_F1_EU_DIS_10EUS:
544 sseu->eu_per_subslice = 10;
545 break;
546 case HSW_F1_EU_DIS_8EUS:
547 sseu->eu_per_subslice = 8;
548 break;
549 case HSW_F1_EU_DIS_6EUS:
550 sseu->eu_per_subslice = 6;
551 break;
552 }
553 sseu->max_eus_per_subslice = sseu->eu_per_subslice;
554
555 for (s = 0; s < sseu->max_slices; s++) {
556 for (ss = 0; ss < sseu->max_subslices; ss++) {
557 sseu_set_eus(sseu, s, ss,
558 (1UL << sseu->eu_per_subslice) - 1);
559 }
560 }
561
562 sseu->eu_total = compute_eu_total(sseu);
563
564 /* No powergating for you. */
565 sseu->has_slice_pg = 0;
566 sseu->has_subslice_pg = 0;
567 sseu->has_eu_pg = 0;
568 }
569
570 static u32 read_reference_ts_freq(struct drm_i915_private *dev_priv)
571 {
572 u32 ts_override = I915_READ(GEN9_TIMESTAMP_OVERRIDE);
573 u32 base_freq, frac_freq;
574
575 base_freq = ((ts_override & GEN9_TIMESTAMP_OVERRIDE_US_COUNTER_DIVIDER_MASK) >>
576 GEN9_TIMESTAMP_OVERRIDE_US_COUNTER_DIVIDER_SHIFT) + 1;
577 base_freq *= 1000;
578
579 frac_freq = ((ts_override &
580 GEN9_TIMESTAMP_OVERRIDE_US_COUNTER_DENOMINATOR_MASK) >>
581 GEN9_TIMESTAMP_OVERRIDE_US_COUNTER_DENOMINATOR_SHIFT);
582 frac_freq = 1000 / (frac_freq + 1);
583
584 return base_freq + frac_freq;
585 }
586
587 static u32 gen10_get_crystal_clock_freq(struct drm_i915_private *dev_priv,
588 u32 rpm_config_reg)
589 {
590 u32 f19_2_mhz = 19200;
591 u32 f24_mhz = 24000;
592 u32 crystal_clock = (rpm_config_reg &
593 GEN9_RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_MASK) >>
594 GEN9_RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_SHIFT;
595
596 switch (crystal_clock) {
597 case GEN9_RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_19_2_MHZ:
598 return f19_2_mhz;
599 case GEN9_RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_24_MHZ:
600 return f24_mhz;
601 default:
602 MISSING_CASE(crystal_clock);
603 return 0;
604 }
605 }
606
607 static u32 gen11_get_crystal_clock_freq(struct drm_i915_private *dev_priv,
608 u32 rpm_config_reg)
609 {
610 u32 f19_2_mhz = 19200;
611 u32 f24_mhz = 24000;
612 u32 f25_mhz = 25000;
613 u32 f38_4_mhz = 38400;
614 u32 crystal_clock = (rpm_config_reg &
615 GEN11_RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_MASK) >>
616 GEN11_RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_SHIFT;
617
618 switch (crystal_clock) {
619 case GEN11_RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_24_MHZ:
620 return f24_mhz;
621 case GEN11_RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_19_2_MHZ:
622 return f19_2_mhz;
623 case GEN11_RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_38_4_MHZ:
624 return f38_4_mhz;
625 case GEN11_RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_25_MHZ:
626 return f25_mhz;
627 default:
628 MISSING_CASE(crystal_clock);
629 return 0;
630 }
631 }
632
633 static u32 read_timestamp_frequency(struct drm_i915_private *dev_priv)
634 {
635 u32 f12_5_mhz = 12500;
636 u32 f19_2_mhz = 19200;
637 u32 f24_mhz = 24000;
638
639 if (INTEL_GEN(dev_priv) <= 4) {
640 /* PRMs say:
641 *
642 * "The value in this register increments once every 16
643 * hclks." (through the “Clocking Configuration”
644 * (“CLKCFG”) MCHBAR register)
645 */
646 return dev_priv->rawclk_freq / 16;
647 } else if (INTEL_GEN(dev_priv) <= 8) {
648 /* PRMs say:
649 *
650 * "The PCU TSC counts 10ns increments; this timestamp
651 * reflects bits 38:3 of the TSC (i.e. 80ns granularity,
652 * rolling over every 1.5 hours).
653 */
654 return f12_5_mhz;
655 } else if (INTEL_GEN(dev_priv) <= 9) {
656 u32 ctc_reg = I915_READ(CTC_MODE);
657 u32 freq = 0;
658
659 if ((ctc_reg & CTC_SOURCE_PARAMETER_MASK) == CTC_SOURCE_DIVIDE_LOGIC) {
660 freq = read_reference_ts_freq(dev_priv);
661 } else {
662 freq = IS_GEN9_LP(dev_priv) ? f19_2_mhz : f24_mhz;
663
664 /* Now figure out how the command stream's timestamp
665 * register increments from this frequency (it might
666 * increment only every few clock cycle).
667 */
668 freq >>= 3 - ((ctc_reg & CTC_SHIFT_PARAMETER_MASK) >>
669 CTC_SHIFT_PARAMETER_SHIFT);
670 }
671
672 return freq;
673 } else if (INTEL_GEN(dev_priv) <= 11) {
674 u32 ctc_reg = I915_READ(CTC_MODE);
675 u32 freq = 0;
676
677 /* First figure out the reference frequency. There are 2 ways
678 * we can compute the frequency, either through the
679 * TIMESTAMP_OVERRIDE register or through RPM_CONFIG. CTC_MODE
680 * tells us which one we should use.
681 */
682 if ((ctc_reg & CTC_SOURCE_PARAMETER_MASK) == CTC_SOURCE_DIVIDE_LOGIC) {
683 freq = read_reference_ts_freq(dev_priv);
684 } else {
685 u32 rpm_config_reg = I915_READ(RPM_CONFIG0);
686
687 if (INTEL_GEN(dev_priv) <= 10)
688 freq = gen10_get_crystal_clock_freq(dev_priv,
689 rpm_config_reg);
690 else
691 freq = gen11_get_crystal_clock_freq(dev_priv,
692 rpm_config_reg);
693
694 /* Now figure out how the command stream's timestamp
695 * register increments from this frequency (it might
696 * increment only every few clock cycle).
697 */
698 freq >>= 3 - ((rpm_config_reg &
699 GEN10_RPM_CONFIG0_CTC_SHIFT_PARAMETER_MASK) >>
700 GEN10_RPM_CONFIG0_CTC_SHIFT_PARAMETER_SHIFT);
701 }
702
703 return freq;
704 }
705
706 MISSING_CASE("Unknown gen, unable to read command streamer timestamp frequency\n");
707 return 0;
708 }
709
710 /**
711 * intel_device_info_runtime_init - initialize runtime info
712 * @dev_priv: the i915 device
713 *
714 * Determine various intel_device_info fields at runtime.
715 *
716 * Use it when either:
717 * - it's judged too laborious to fill n static structures with the limit
718 * when a simple if statement does the job,
719 * - run-time checks (eg read fuse/strap registers) are needed.
720 *
721 * This function needs to be called:
722 * - after the MMIO has been setup as we are reading registers,
723 * - after the PCH has been detected,
724 * - before the first usage of the fields it can tweak.
725 */
726 void intel_device_info_runtime_init(struct drm_i915_private *dev_priv)
727 {
728 struct intel_device_info *info = mkwrite_device_info(dev_priv);
729 struct intel_runtime_info *runtime = RUNTIME_INFO(dev_priv);
730 enum pipe pipe;
731
732 if (INTEL_GEN(dev_priv) >= 10) {
733 for_each_pipe(dev_priv, pipe)
734 runtime->num_scalers[pipe] = 2;
735 } else if (IS_GEN(dev_priv, 9)) {
736 runtime->num_scalers[PIPE_A] = 2;
737 runtime->num_scalers[PIPE_B] = 2;
738 runtime->num_scalers[PIPE_C] = 1;
739 }
740
741 BUILD_BUG_ON(I915_NUM_ENGINES > BITS_PER_TYPE(intel_ring_mask_t));
742
743 if (IS_GEN(dev_priv, 11))
744 for_each_pipe(dev_priv, pipe)
745 runtime->num_sprites[pipe] = 6;
746 else if (IS_GEN(dev_priv, 10) || IS_GEMINILAKE(dev_priv))
747 for_each_pipe(dev_priv, pipe)
748 runtime->num_sprites[pipe] = 3;
749 else if (IS_BROXTON(dev_priv)) {
750 /*
751 * Skylake and Broxton currently don't expose the topmost plane as its
752 * use is exclusive with the legacy cursor and we only want to expose
753 * one of those, not both. Until we can safely expose the topmost plane
754 * as a DRM_PLANE_TYPE_CURSOR with all the features exposed/supported,
755 * we don't expose the topmost plane at all to prevent ABI breakage
756 * down the line.
757 */
758
759 runtime->num_sprites[PIPE_A] = 2;
760 runtime->num_sprites[PIPE_B] = 2;
761 runtime->num_sprites[PIPE_C] = 1;
762 } else if (IS_VALLEYVIEW(dev_priv) || IS_CHERRYVIEW(dev_priv)) {
763 for_each_pipe(dev_priv, pipe)
764 runtime->num_sprites[pipe] = 2;
765 } else if (INTEL_GEN(dev_priv) >= 5 || IS_G4X(dev_priv)) {
766 for_each_pipe(dev_priv, pipe)
767 runtime->num_sprites[pipe] = 1;
768 }
769
770 if (i915_modparams.disable_display) {
771 DRM_INFO("Display disabled (module parameter)\n");
772 info->num_pipes = 0;
773 } else if (HAS_DISPLAY(dev_priv) &&
774 (IS_GEN_RANGE(dev_priv, 7, 8)) &&
775 HAS_PCH_SPLIT(dev_priv)) {
776 u32 fuse_strap = I915_READ(FUSE_STRAP);
777 u32 sfuse_strap = I915_READ(SFUSE_STRAP);
778
779 /*
780 * SFUSE_STRAP is supposed to have a bit signalling the display
781 * is fused off. Unfortunately it seems that, at least in
782 * certain cases, fused off display means that PCH display
783 * reads don't land anywhere. In that case, we read 0s.
784 *
785 * On CPT/PPT, we can detect this case as SFUSE_STRAP_FUSE_LOCK
786 * should be set when taking over after the firmware.
787 */
788 if (fuse_strap & ILK_INTERNAL_DISPLAY_DISABLE ||
789 sfuse_strap & SFUSE_STRAP_DISPLAY_DISABLED ||
790 (HAS_PCH_CPT(dev_priv) &&
791 !(sfuse_strap & SFUSE_STRAP_FUSE_LOCK))) {
792 DRM_INFO("Display fused off, disabling\n");
793 info->num_pipes = 0;
794 } else if (fuse_strap & IVB_PIPE_C_DISABLE) {
795 DRM_INFO("PipeC fused off\n");
796 info->num_pipes -= 1;
797 }
798 } else if (HAS_DISPLAY(dev_priv) && INTEL_GEN(dev_priv) >= 9) {
799 u32 dfsm = I915_READ(SKL_DFSM);
800 u8 disabled_mask = 0;
801 bool invalid;
802 int num_bits;
803
804 if (dfsm & SKL_DFSM_PIPE_A_DISABLE)
805 disabled_mask |= BIT(PIPE_A);
806 if (dfsm & SKL_DFSM_PIPE_B_DISABLE)
807 disabled_mask |= BIT(PIPE_B);
808 if (dfsm & SKL_DFSM_PIPE_C_DISABLE)
809 disabled_mask |= BIT(PIPE_C);
810
811 num_bits = hweight8(disabled_mask);
812
813 switch (disabled_mask) {
814 case BIT(PIPE_A):
815 case BIT(PIPE_B):
816 case BIT(PIPE_A) | BIT(PIPE_B):
817 case BIT(PIPE_A) | BIT(PIPE_C):
818 invalid = true;
819 break;
820 default:
821 invalid = false;
822 }
823
824 if (num_bits > info->num_pipes || invalid)
825 DRM_ERROR("invalid pipe fuse configuration: 0x%x\n",
826 disabled_mask);
827 else
828 info->num_pipes -= num_bits;
829 }
830
831 /* Initialize slice/subslice/EU info */
832 if (IS_HASWELL(dev_priv))
833 haswell_sseu_info_init(dev_priv);
834 else if (IS_CHERRYVIEW(dev_priv))
835 cherryview_sseu_info_init(dev_priv);
836 else if (IS_BROADWELL(dev_priv))
837 broadwell_sseu_info_init(dev_priv);
838 else if (IS_GEN(dev_priv, 9))
839 gen9_sseu_info_init(dev_priv);
840 else if (IS_GEN(dev_priv, 10))
841 gen10_sseu_info_init(dev_priv);
842 else if (INTEL_GEN(dev_priv) >= 11)
843 gen11_sseu_info_init(dev_priv);
844
845 if (IS_GEN(dev_priv, 6) && intel_vtd_active()) {
846 DRM_INFO("Disabling ppGTT for VT-d support\n");
847 info->ppgtt = INTEL_PPGTT_NONE;
848 }
849
850 /* Initialize command stream timestamp frequency */
851 runtime->cs_timestamp_frequency_khz = read_timestamp_frequency(dev_priv);
852 }
853
854 void intel_driver_caps_print(const struct intel_driver_caps *caps,
855 struct drm_printer *p)
856 {
857 drm_printf(p, "Has logical contexts? %s\n",
858 yesno(caps->has_logical_contexts));
859 drm_printf(p, "scheduler: %x\n", caps->scheduler);
860 }
861
862 /*
863 * Determine which engines are fused off in our particular hardware. Since the
864 * fuse register is in the blitter powerwell, we need forcewake to be ready at
865 * this point (but later we need to prune the forcewake domains for engines that
866 * are indeed fused off).
867 */
868 void intel_device_info_init_mmio(struct drm_i915_private *dev_priv)
869 {
870 struct intel_device_info *info = mkwrite_device_info(dev_priv);
871 unsigned int logical_vdbox = 0;
872 unsigned int i;
873 u32 media_fuse;
874
875 if (INTEL_GEN(dev_priv) < 11)
876 return;
877
878 media_fuse = ~I915_READ(GEN11_GT_VEBOX_VDBOX_DISABLE);
879
880 RUNTIME_INFO(dev_priv)->vdbox_enable = media_fuse & GEN11_GT_VDBOX_DISABLE_MASK;
881 RUNTIME_INFO(dev_priv)->vebox_enable = (media_fuse & GEN11_GT_VEBOX_DISABLE_MASK) >>
882 GEN11_GT_VEBOX_DISABLE_SHIFT;
883
884 DRM_DEBUG_DRIVER("vdbox enable: %04x\n", RUNTIME_INFO(dev_priv)->vdbox_enable);
885 for (i = 0; i < I915_MAX_VCS; i++) {
886 if (!HAS_ENGINE(dev_priv, _VCS(i)))
887 continue;
888
889 if (!(BIT(i) & RUNTIME_INFO(dev_priv)->vdbox_enable)) {
890 info->ring_mask &= ~ENGINE_MASK(_VCS(i));
891 DRM_DEBUG_DRIVER("vcs%u fused off\n", i);
892 continue;
893 }
894
895 /*
896 * In Gen11, only even numbered logical VDBOXes are
897 * hooked up to an SFC (Scaler & Format Converter) unit.
898 */
899 if (logical_vdbox++ % 2 == 0)
900 RUNTIME_INFO(dev_priv)->vdbox_sfc_access |= BIT(i);
901 }
902
903 DRM_DEBUG_DRIVER("vebox enable: %04x\n", RUNTIME_INFO(dev_priv)->vebox_enable);
904 for (i = 0; i < I915_MAX_VECS; i++) {
905 if (!HAS_ENGINE(dev_priv, _VECS(i)))
906 continue;
907
908 if (!(BIT(i) & RUNTIME_INFO(dev_priv)->vebox_enable)) {
909 info->ring_mask &= ~ENGINE_MASK(_VECS(i));
910 DRM_DEBUG_DRIVER("vecs%u fused off\n", i);
911 }
912 }
913 }