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_device_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(const struct intel_device_info *info,
117 			    struct drm_printer *p)
118 {
119 	struct drm_i915_private *dev_priv =
120 		container_of(info, struct drm_i915_private, info);
121 
122 	drm_printf(p, "pciid=0x%04x rev=0x%02x platform=%s gen=%i\n",
123 		   INTEL_DEVID(dev_priv),
124 		   INTEL_REVID(dev_priv),
125 		   intel_platform_name(info->platform),
126 		   info->gen);
127 
128 	intel_device_info_dump_flags(info, p);
129 }
130 
131 void intel_device_info_dump_topology(const struct sseu_dev_info *sseu,
132 				     struct drm_printer *p)
133 {
134 	int s, ss;
135 
136 	if (sseu->max_slices == 0) {
137 		drm_printf(p, "Unavailable\n");
138 		return;
139 	}
140 
141 	for (s = 0; s < sseu->max_slices; s++) {
142 		drm_printf(p, "slice%d: %u subslice(s) (0x%hhx):\n",
143 			   s, hweight8(sseu->subslice_mask[s]),
144 			   sseu->subslice_mask[s]);
145 
146 		for (ss = 0; ss < sseu->max_subslices; ss++) {
147 			u16 enabled_eus = sseu_get_eus(sseu, s, ss);
148 
149 			drm_printf(p, "\tsubslice%d: %u EUs (0x%hx)\n",
150 				   ss, hweight16(enabled_eus), enabled_eus);
151 		}
152 	}
153 }
154 
155 static u16 compute_eu_total(const struct sseu_dev_info *sseu)
156 {
157 	u16 i, total = 0;
158 
159 	for (i = 0; i < ARRAY_SIZE(sseu->eu_mask); i++)
160 		total += hweight8(sseu->eu_mask[i]);
161 
162 	return total;
163 }
164 
165 static void gen11_sseu_info_init(struct drm_i915_private *dev_priv)
166 {
167 	struct sseu_dev_info *sseu = &mkwrite_device_info(dev_priv)->sseu;
168 	u8 s_en;
169 	u32 ss_en, ss_en_mask;
170 	u8 eu_en;
171 	int s;
172 
173 	sseu->max_slices = 1;
174 	sseu->max_subslices = 8;
175 	sseu->max_eus_per_subslice = 8;
176 
177 	s_en = I915_READ(GEN11_GT_SLICE_ENABLE) & GEN11_GT_S_ENA_MASK;
178 	ss_en = ~I915_READ(GEN11_GT_SUBSLICE_DISABLE);
179 	ss_en_mask = BIT(sseu->max_subslices) - 1;
180 	eu_en = ~(I915_READ(GEN11_EU_DISABLE) & GEN11_EU_DIS_MASK);
181 
182 	for (s = 0; s < sseu->max_slices; s++) {
183 		if (s_en & BIT(s)) {
184 			int ss_idx = sseu->max_subslices * s;
185 			int ss;
186 
187 			sseu->slice_mask |= BIT(s);
188 			sseu->subslice_mask[s] = (ss_en >> ss_idx) & ss_en_mask;
189 			for (ss = 0; ss < sseu->max_subslices; ss++) {
190 				if (sseu->subslice_mask[s] & BIT(ss))
191 					sseu_set_eus(sseu, s, ss, eu_en);
192 			}
193 		}
194 	}
195 	sseu->eu_per_subslice = hweight8(eu_en);
196 	sseu->eu_total = compute_eu_total(sseu);
197 
198 	/* ICL has no power gating restrictions. */
199 	sseu->has_slice_pg = 1;
200 	sseu->has_subslice_pg = 1;
201 	sseu->has_eu_pg = 1;
202 }
203 
204 static void gen10_sseu_info_init(struct drm_i915_private *dev_priv)
205 {
206 	struct sseu_dev_info *sseu = &mkwrite_device_info(dev_priv)->sseu;
207 	const u32 fuse2 = I915_READ(GEN8_FUSE2);
208 	int s, ss;
209 	const int eu_mask = 0xff;
210 	u32 subslice_mask, eu_en;
211 
212 	sseu->slice_mask = (fuse2 & GEN10_F2_S_ENA_MASK) >>
213 			    GEN10_F2_S_ENA_SHIFT;
214 	sseu->max_slices = 6;
215 	sseu->max_subslices = 4;
216 	sseu->max_eus_per_subslice = 8;
217 
218 	subslice_mask = (1 << 4) - 1;
219 	subslice_mask &= ~((fuse2 & GEN10_F2_SS_DIS_MASK) >>
220 			   GEN10_F2_SS_DIS_SHIFT);
221 
222 	/*
223 	 * Slice0 can have up to 3 subslices, but there are only 2 in
224 	 * slice1/2.
225 	 */
226 	sseu->subslice_mask[0] = subslice_mask;
227 	for (s = 1; s < sseu->max_slices; s++)
228 		sseu->subslice_mask[s] = subslice_mask & 0x3;
229 
230 	/* Slice0 */
231 	eu_en = ~I915_READ(GEN8_EU_DISABLE0);
232 	for (ss = 0; ss < sseu->max_subslices; ss++)
233 		sseu_set_eus(sseu, 0, ss, (eu_en >> (8 * ss)) & eu_mask);
234 	/* Slice1 */
235 	sseu_set_eus(sseu, 1, 0, (eu_en >> 24) & eu_mask);
236 	eu_en = ~I915_READ(GEN8_EU_DISABLE1);
237 	sseu_set_eus(sseu, 1, 1, eu_en & eu_mask);
238 	/* Slice2 */
239 	sseu_set_eus(sseu, 2, 0, (eu_en >> 8) & eu_mask);
240 	sseu_set_eus(sseu, 2, 1, (eu_en >> 16) & eu_mask);
241 	/* Slice3 */
242 	sseu_set_eus(sseu, 3, 0, (eu_en >> 24) & eu_mask);
243 	eu_en = ~I915_READ(GEN8_EU_DISABLE2);
244 	sseu_set_eus(sseu, 3, 1, eu_en & eu_mask);
245 	/* Slice4 */
246 	sseu_set_eus(sseu, 4, 0, (eu_en >> 8) & eu_mask);
247 	sseu_set_eus(sseu, 4, 1, (eu_en >> 16) & eu_mask);
248 	/* Slice5 */
249 	sseu_set_eus(sseu, 5, 0, (eu_en >> 24) & eu_mask);
250 	eu_en = ~I915_READ(GEN10_EU_DISABLE3);
251 	sseu_set_eus(sseu, 5, 1, eu_en & eu_mask);
252 
253 	/* Do a second pass where we mark the subslices disabled if all their
254 	 * eus are off.
255 	 */
256 	for (s = 0; s < sseu->max_slices; s++) {
257 		for (ss = 0; ss < sseu->max_subslices; ss++) {
258 			if (sseu_get_eus(sseu, s, ss) == 0)
259 				sseu->subslice_mask[s] &= ~BIT(ss);
260 		}
261 	}
262 
263 	sseu->eu_total = compute_eu_total(sseu);
264 
265 	/*
266 	 * CNL is expected to always have a uniform distribution
267 	 * of EU across subslices with the exception that any one
268 	 * EU in any one subslice may be fused off for die
269 	 * recovery.
270 	 */
271 	sseu->eu_per_subslice = sseu_subslice_total(sseu) ?
272 				DIV_ROUND_UP(sseu->eu_total,
273 					     sseu_subslice_total(sseu)) : 0;
274 
275 	/* No restrictions on Power Gating */
276 	sseu->has_slice_pg = 1;
277 	sseu->has_subslice_pg = 1;
278 	sseu->has_eu_pg = 1;
279 }
280 
281 static void cherryview_sseu_info_init(struct drm_i915_private *dev_priv)
282 {
283 	struct sseu_dev_info *sseu = &mkwrite_device_info(dev_priv)->sseu;
284 	u32 fuse;
285 
286 	fuse = I915_READ(CHV_FUSE_GT);
287 
288 	sseu->slice_mask = BIT(0);
289 	sseu->max_slices = 1;
290 	sseu->max_subslices = 2;
291 	sseu->max_eus_per_subslice = 8;
292 
293 	if (!(fuse & CHV_FGT_DISABLE_SS0)) {
294 		u8 disabled_mask =
295 			((fuse & CHV_FGT_EU_DIS_SS0_R0_MASK) >>
296 			 CHV_FGT_EU_DIS_SS0_R0_SHIFT) |
297 			(((fuse & CHV_FGT_EU_DIS_SS0_R1_MASK) >>
298 			  CHV_FGT_EU_DIS_SS0_R1_SHIFT) << 4);
299 
300 		sseu->subslice_mask[0] |= BIT(0);
301 		sseu_set_eus(sseu, 0, 0, ~disabled_mask);
302 	}
303 
304 	if (!(fuse & CHV_FGT_DISABLE_SS1)) {
305 		u8 disabled_mask =
306 			((fuse & CHV_FGT_EU_DIS_SS1_R0_MASK) >>
307 			 CHV_FGT_EU_DIS_SS1_R0_SHIFT) |
308 			(((fuse & CHV_FGT_EU_DIS_SS1_R1_MASK) >>
309 			  CHV_FGT_EU_DIS_SS1_R1_SHIFT) << 4);
310 
311 		sseu->subslice_mask[0] |= BIT(1);
312 		sseu_set_eus(sseu, 0, 1, ~disabled_mask);
313 	}
314 
315 	sseu->eu_total = compute_eu_total(sseu);
316 
317 	/*
318 	 * CHV expected to always have a uniform distribution of EU
319 	 * across subslices.
320 	*/
321 	sseu->eu_per_subslice = sseu_subslice_total(sseu) ?
322 				sseu->eu_total / sseu_subslice_total(sseu) :
323 				0;
324 	/*
325 	 * CHV supports subslice power gating on devices with more than
326 	 * one subslice, and supports EU power gating on devices with
327 	 * more than one EU pair per subslice.
328 	*/
329 	sseu->has_slice_pg = 0;
330 	sseu->has_subslice_pg = sseu_subslice_total(sseu) > 1;
331 	sseu->has_eu_pg = (sseu->eu_per_subslice > 2);
332 }
333 
334 static void gen9_sseu_info_init(struct drm_i915_private *dev_priv)
335 {
336 	struct intel_device_info *info = mkwrite_device_info(dev_priv);
337 	struct sseu_dev_info *sseu = &info->sseu;
338 	int s, ss;
339 	u32 fuse2, eu_disable, subslice_mask;
340 	const u8 eu_mask = 0xff;
341 
342 	fuse2 = I915_READ(GEN8_FUSE2);
343 	sseu->slice_mask = (fuse2 & GEN8_F2_S_ENA_MASK) >> GEN8_F2_S_ENA_SHIFT;
344 
345 	/* BXT has a single slice and at most 3 subslices. */
346 	sseu->max_slices = IS_GEN9_LP(dev_priv) ? 1 : 3;
347 	sseu->max_subslices = IS_GEN9_LP(dev_priv) ? 3 : 4;
348 	sseu->max_eus_per_subslice = 8;
349 
350 	/*
351 	 * The subslice disable field is global, i.e. it applies
352 	 * to each of the enabled slices.
353 	*/
354 	subslice_mask = (1 << sseu->max_subslices) - 1;
355 	subslice_mask &= ~((fuse2 & GEN9_F2_SS_DIS_MASK) >>
356 			   GEN9_F2_SS_DIS_SHIFT);
357 
358 	/*
359 	 * Iterate through enabled slices and subslices to
360 	 * count the total enabled EU.
361 	*/
362 	for (s = 0; s < sseu->max_slices; s++) {
363 		if (!(sseu->slice_mask & BIT(s)))
364 			/* skip disabled slice */
365 			continue;
366 
367 		sseu->subslice_mask[s] = subslice_mask;
368 
369 		eu_disable = I915_READ(GEN9_EU_DISABLE(s));
370 		for (ss = 0; ss < sseu->max_subslices; ss++) {
371 			int eu_per_ss;
372 			u8 eu_disabled_mask;
373 
374 			if (!(sseu->subslice_mask[s] & BIT(ss)))
375 				/* skip disabled subslice */
376 				continue;
377 
378 			eu_disabled_mask = (eu_disable >> (ss * 8)) & eu_mask;
379 
380 			sseu_set_eus(sseu, s, ss, ~eu_disabled_mask);
381 
382 			eu_per_ss = sseu->max_eus_per_subslice -
383 				hweight8(eu_disabled_mask);
384 
385 			/*
386 			 * Record which subslice(s) has(have) 7 EUs. we
387 			 * can tune the hash used to spread work among
388 			 * subslices if they are unbalanced.
389 			 */
390 			if (eu_per_ss == 7)
391 				sseu->subslice_7eu[s] |= BIT(ss);
392 		}
393 	}
394 
395 	sseu->eu_total = compute_eu_total(sseu);
396 
397 	/*
398 	 * SKL is expected to always have a uniform distribution
399 	 * of EU across subslices with the exception that any one
400 	 * EU in any one subslice may be fused off for die
401 	 * recovery. BXT is expected to be perfectly uniform in EU
402 	 * distribution.
403 	*/
404 	sseu->eu_per_subslice = sseu_subslice_total(sseu) ?
405 				DIV_ROUND_UP(sseu->eu_total,
406 					     sseu_subslice_total(sseu)) : 0;
407 	/*
408 	 * SKL+ supports slice power gating on devices with more than
409 	 * one slice, and supports EU power gating on devices with
410 	 * more than one EU pair per subslice. BXT+ supports subslice
411 	 * power gating on devices with more than one subslice, and
412 	 * supports EU power gating on devices with more than one EU
413 	 * pair per subslice.
414 	*/
415 	sseu->has_slice_pg =
416 		!IS_GEN9_LP(dev_priv) && hweight8(sseu->slice_mask) > 1;
417 	sseu->has_subslice_pg =
418 		IS_GEN9_LP(dev_priv) && sseu_subslice_total(sseu) > 1;
419 	sseu->has_eu_pg = sseu->eu_per_subslice > 2;
420 
421 	if (IS_GEN9_LP(dev_priv)) {
422 #define IS_SS_DISABLED(ss)	(!(sseu->subslice_mask[0] & BIT(ss)))
423 		info->has_pooled_eu = hweight8(sseu->subslice_mask[0]) == 3;
424 
425 		sseu->min_eu_in_pool = 0;
426 		if (info->has_pooled_eu) {
427 			if (IS_SS_DISABLED(2) || IS_SS_DISABLED(0))
428 				sseu->min_eu_in_pool = 3;
429 			else if (IS_SS_DISABLED(1))
430 				sseu->min_eu_in_pool = 6;
431 			else
432 				sseu->min_eu_in_pool = 9;
433 		}
434 #undef IS_SS_DISABLED
435 	}
436 }
437 
438 static void broadwell_sseu_info_init(struct drm_i915_private *dev_priv)
439 {
440 	struct sseu_dev_info *sseu = &mkwrite_device_info(dev_priv)->sseu;
441 	int s, ss;
442 	u32 fuse2, subslice_mask, eu_disable[3]; /* s_max */
443 
444 	fuse2 = I915_READ(GEN8_FUSE2);
445 	sseu->slice_mask = (fuse2 & GEN8_F2_S_ENA_MASK) >> GEN8_F2_S_ENA_SHIFT;
446 	sseu->max_slices = 3;
447 	sseu->max_subslices = 3;
448 	sseu->max_eus_per_subslice = 8;
449 
450 	/*
451 	 * The subslice disable field is global, i.e. it applies
452 	 * to each of the enabled slices.
453 	 */
454 	subslice_mask = GENMASK(sseu->max_subslices - 1, 0);
455 	subslice_mask &= ~((fuse2 & GEN8_F2_SS_DIS_MASK) >>
456 			   GEN8_F2_SS_DIS_SHIFT);
457 
458 	eu_disable[0] = I915_READ(GEN8_EU_DISABLE0) & GEN8_EU_DIS0_S0_MASK;
459 	eu_disable[1] = (I915_READ(GEN8_EU_DISABLE0) >> GEN8_EU_DIS0_S1_SHIFT) |
460 			((I915_READ(GEN8_EU_DISABLE1) & GEN8_EU_DIS1_S1_MASK) <<
461 			 (32 - GEN8_EU_DIS0_S1_SHIFT));
462 	eu_disable[2] = (I915_READ(GEN8_EU_DISABLE1) >> GEN8_EU_DIS1_S2_SHIFT) |
463 			((I915_READ(GEN8_EU_DISABLE2) & GEN8_EU_DIS2_S2_MASK) <<
464 			 (32 - GEN8_EU_DIS1_S2_SHIFT));
465 
466 	/*
467 	 * Iterate through enabled slices and subslices to
468 	 * count the total enabled EU.
469 	 */
470 	for (s = 0; s < sseu->max_slices; s++) {
471 		if (!(sseu->slice_mask & BIT(s)))
472 			/* skip disabled slice */
473 			continue;
474 
475 		sseu->subslice_mask[s] = subslice_mask;
476 
477 		for (ss = 0; ss < sseu->max_subslices; ss++) {
478 			u8 eu_disabled_mask;
479 			u32 n_disabled;
480 
481 			if (!(sseu->subslice_mask[s] & BIT(ss)))
482 				/* skip disabled subslice */
483 				continue;
484 
485 			eu_disabled_mask =
486 				eu_disable[s] >> (ss * sseu->max_eus_per_subslice);
487 
488 			sseu_set_eus(sseu, s, ss, ~eu_disabled_mask);
489 
490 			n_disabled = hweight8(eu_disabled_mask);
491 
492 			/*
493 			 * Record which subslices have 7 EUs.
494 			 */
495 			if (sseu->max_eus_per_subslice - n_disabled == 7)
496 				sseu->subslice_7eu[s] |= 1 << ss;
497 		}
498 	}
499 
500 	sseu->eu_total = compute_eu_total(sseu);
501 
502 	/*
503 	 * BDW is expected to always have a uniform distribution of EU across
504 	 * subslices with the exception that any one EU in any one subslice may
505 	 * be fused off for die recovery.
506 	 */
507 	sseu->eu_per_subslice = sseu_subslice_total(sseu) ?
508 				DIV_ROUND_UP(sseu->eu_total,
509 					     sseu_subslice_total(sseu)) : 0;
510 
511 	/*
512 	 * BDW supports slice power gating on devices with more than
513 	 * one slice.
514 	 */
515 	sseu->has_slice_pg = hweight8(sseu->slice_mask) > 1;
516 	sseu->has_subslice_pg = 0;
517 	sseu->has_eu_pg = 0;
518 }
519 
520 static void haswell_sseu_info_init(struct drm_i915_private *dev_priv)
521 {
522 	struct intel_device_info *info = mkwrite_device_info(dev_priv);
523 	struct sseu_dev_info *sseu = &info->sseu;
524 	u32 fuse1;
525 	int s, ss;
526 
527 	/*
528 	 * There isn't a register to tell us how many slices/subslices. We
529 	 * work off the PCI-ids here.
530 	 */
531 	switch (info->gt) {
532 	default:
533 		MISSING_CASE(info->gt);
534 		/* fall through */
535 	case 1:
536 		sseu->slice_mask = BIT(0);
537 		sseu->subslice_mask[0] = BIT(0);
538 		break;
539 	case 2:
540 		sseu->slice_mask = BIT(0);
541 		sseu->subslice_mask[0] = BIT(0) | BIT(1);
542 		break;
543 	case 3:
544 		sseu->slice_mask = BIT(0) | BIT(1);
545 		sseu->subslice_mask[0] = BIT(0) | BIT(1);
546 		sseu->subslice_mask[1] = BIT(0) | BIT(1);
547 		break;
548 	}
549 
550 	sseu->max_slices = hweight8(sseu->slice_mask);
551 	sseu->max_subslices = hweight8(sseu->subslice_mask[0]);
552 
553 	fuse1 = I915_READ(HSW_PAVP_FUSE1);
554 	switch ((fuse1 & HSW_F1_EU_DIS_MASK) >> HSW_F1_EU_DIS_SHIFT) {
555 	default:
556 		MISSING_CASE((fuse1 & HSW_F1_EU_DIS_MASK) >>
557 			     HSW_F1_EU_DIS_SHIFT);
558 		/* fall through */
559 	case HSW_F1_EU_DIS_10EUS:
560 		sseu->eu_per_subslice = 10;
561 		break;
562 	case HSW_F1_EU_DIS_8EUS:
563 		sseu->eu_per_subslice = 8;
564 		break;
565 	case HSW_F1_EU_DIS_6EUS:
566 		sseu->eu_per_subslice = 6;
567 		break;
568 	}
569 	sseu->max_eus_per_subslice = sseu->eu_per_subslice;
570 
571 	for (s = 0; s < sseu->max_slices; s++) {
572 		for (ss = 0; ss < sseu->max_subslices; ss++) {
573 			sseu_set_eus(sseu, s, ss,
574 				     (1UL << sseu->eu_per_subslice) - 1);
575 		}
576 	}
577 
578 	sseu->eu_total = compute_eu_total(sseu);
579 
580 	/* No powergating for you. */
581 	sseu->has_slice_pg = 0;
582 	sseu->has_subslice_pg = 0;
583 	sseu->has_eu_pg = 0;
584 }
585 
586 static u32 read_reference_ts_freq(struct drm_i915_private *dev_priv)
587 {
588 	u32 ts_override = I915_READ(GEN9_TIMESTAMP_OVERRIDE);
589 	u32 base_freq, frac_freq;
590 
591 	base_freq = ((ts_override & GEN9_TIMESTAMP_OVERRIDE_US_COUNTER_DIVIDER_MASK) >>
592 		     GEN9_TIMESTAMP_OVERRIDE_US_COUNTER_DIVIDER_SHIFT) + 1;
593 	base_freq *= 1000;
594 
595 	frac_freq = ((ts_override &
596 		      GEN9_TIMESTAMP_OVERRIDE_US_COUNTER_DENOMINATOR_MASK) >>
597 		     GEN9_TIMESTAMP_OVERRIDE_US_COUNTER_DENOMINATOR_SHIFT);
598 	frac_freq = 1000 / (frac_freq + 1);
599 
600 	return base_freq + frac_freq;
601 }
602 
603 static u32 gen10_get_crystal_clock_freq(struct drm_i915_private *dev_priv,
604 					u32 rpm_config_reg)
605 {
606 	u32 f19_2_mhz = 19200;
607 	u32 f24_mhz = 24000;
608 	u32 crystal_clock = (rpm_config_reg &
609 			     GEN9_RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_MASK) >>
610 			    GEN9_RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_SHIFT;
611 
612 	switch (crystal_clock) {
613 	case GEN9_RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_19_2_MHZ:
614 		return f19_2_mhz;
615 	case GEN9_RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_24_MHZ:
616 		return f24_mhz;
617 	default:
618 		MISSING_CASE(crystal_clock);
619 		return 0;
620 	}
621 }
622 
623 static u32 gen11_get_crystal_clock_freq(struct drm_i915_private *dev_priv,
624 					u32 rpm_config_reg)
625 {
626 	u32 f19_2_mhz = 19200;
627 	u32 f24_mhz = 24000;
628 	u32 f25_mhz = 25000;
629 	u32 f38_4_mhz = 38400;
630 	u32 crystal_clock = (rpm_config_reg &
631 			     GEN11_RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_MASK) >>
632 			    GEN11_RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_SHIFT;
633 
634 	switch (crystal_clock) {
635 	case GEN11_RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_24_MHZ:
636 		return f24_mhz;
637 	case GEN11_RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_19_2_MHZ:
638 		return f19_2_mhz;
639 	case GEN11_RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_38_4_MHZ:
640 		return f38_4_mhz;
641 	case GEN11_RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_25_MHZ:
642 		return f25_mhz;
643 	default:
644 		MISSING_CASE(crystal_clock);
645 		return 0;
646 	}
647 }
648 
649 static u32 read_timestamp_frequency(struct drm_i915_private *dev_priv)
650 {
651 	u32 f12_5_mhz = 12500;
652 	u32 f19_2_mhz = 19200;
653 	u32 f24_mhz = 24000;
654 
655 	if (INTEL_GEN(dev_priv) <= 4) {
656 		/* PRMs say:
657 		 *
658 		 *     "The value in this register increments once every 16
659 		 *      hclks." (through the “Clocking Configuration”
660 		 *      (“CLKCFG”) MCHBAR register)
661 		 */
662 		return dev_priv->rawclk_freq / 16;
663 	} else if (INTEL_GEN(dev_priv) <= 8) {
664 		/* PRMs say:
665 		 *
666 		 *     "The PCU TSC counts 10ns increments; this timestamp
667 		 *      reflects bits 38:3 of the TSC (i.e. 80ns granularity,
668 		 *      rolling over every 1.5 hours).
669 		 */
670 		return f12_5_mhz;
671 	} else if (INTEL_GEN(dev_priv) <= 9) {
672 		u32 ctc_reg = I915_READ(CTC_MODE);
673 		u32 freq = 0;
674 
675 		if ((ctc_reg & CTC_SOURCE_PARAMETER_MASK) == CTC_SOURCE_DIVIDE_LOGIC) {
676 			freq = read_reference_ts_freq(dev_priv);
677 		} else {
678 			freq = IS_GEN9_LP(dev_priv) ? f19_2_mhz : f24_mhz;
679 
680 			/* Now figure out how the command stream's timestamp
681 			 * register increments from this frequency (it might
682 			 * increment only every few clock cycle).
683 			 */
684 			freq >>= 3 - ((ctc_reg & CTC_SHIFT_PARAMETER_MASK) >>
685 				      CTC_SHIFT_PARAMETER_SHIFT);
686 		}
687 
688 		return freq;
689 	} else if (INTEL_GEN(dev_priv) <= 11) {
690 		u32 ctc_reg = I915_READ(CTC_MODE);
691 		u32 freq = 0;
692 
693 		/* First figure out the reference frequency. There are 2 ways
694 		 * we can compute the frequency, either through the
695 		 * TIMESTAMP_OVERRIDE register or through RPM_CONFIG. CTC_MODE
696 		 * tells us which one we should use.
697 		 */
698 		if ((ctc_reg & CTC_SOURCE_PARAMETER_MASK) == CTC_SOURCE_DIVIDE_LOGIC) {
699 			freq = read_reference_ts_freq(dev_priv);
700 		} else {
701 			u32 rpm_config_reg = I915_READ(RPM_CONFIG0);
702 
703 			if (INTEL_GEN(dev_priv) <= 10)
704 				freq = gen10_get_crystal_clock_freq(dev_priv,
705 								rpm_config_reg);
706 			else
707 				freq = gen11_get_crystal_clock_freq(dev_priv,
708 								rpm_config_reg);
709 
710 			/* Now figure out how the command stream's timestamp
711 			 * register increments from this frequency (it might
712 			 * increment only every few clock cycle).
713 			 */
714 			freq >>= 3 - ((rpm_config_reg &
715 				       GEN10_RPM_CONFIG0_CTC_SHIFT_PARAMETER_MASK) >>
716 				      GEN10_RPM_CONFIG0_CTC_SHIFT_PARAMETER_SHIFT);
717 		}
718 
719 		return freq;
720 	}
721 
722 	MISSING_CASE("Unknown gen, unable to read command streamer timestamp frequency\n");
723 	return 0;
724 }
725 
726 /**
727  * intel_device_info_runtime_init - initialize runtime info
728  * @info: intel device info struct
729  *
730  * Determine various intel_device_info fields at runtime.
731  *
732  * Use it when either:
733  *   - it's judged too laborious to fill n static structures with the limit
734  *     when a simple if statement does the job,
735  *   - run-time checks (eg read fuse/strap registers) are needed.
736  *
737  * This function needs to be called:
738  *   - after the MMIO has been setup as we are reading registers,
739  *   - after the PCH has been detected,
740  *   - before the first usage of the fields it can tweak.
741  */
742 void intel_device_info_runtime_init(struct intel_device_info *info)
743 {
744 	struct drm_i915_private *dev_priv =
745 		container_of(info, struct drm_i915_private, info);
746 	enum pipe pipe;
747 
748 	if (INTEL_GEN(dev_priv) >= 10) {
749 		for_each_pipe(dev_priv, pipe)
750 			info->num_scalers[pipe] = 2;
751 	} else if (IS_GEN9(dev_priv)) {
752 		info->num_scalers[PIPE_A] = 2;
753 		info->num_scalers[PIPE_B] = 2;
754 		info->num_scalers[PIPE_C] = 1;
755 	}
756 
757 	BUILD_BUG_ON(I915_NUM_ENGINES > BITS_PER_TYPE(intel_ring_mask_t));
758 
759 	if (IS_GEN11(dev_priv))
760 		for_each_pipe(dev_priv, pipe)
761 			info->num_sprites[pipe] = 6;
762 	else if (IS_GEN10(dev_priv) || IS_GEMINILAKE(dev_priv))
763 		for_each_pipe(dev_priv, pipe)
764 			info->num_sprites[pipe] = 3;
765 	else if (IS_BROXTON(dev_priv)) {
766 		/*
767 		 * Skylake and Broxton currently don't expose the topmost plane as its
768 		 * use is exclusive with the legacy cursor and we only want to expose
769 		 * one of those, not both. Until we can safely expose the topmost plane
770 		 * as a DRM_PLANE_TYPE_CURSOR with all the features exposed/supported,
771 		 * we don't expose the topmost plane at all to prevent ABI breakage
772 		 * down the line.
773 		 */
774 
775 		info->num_sprites[PIPE_A] = 2;
776 		info->num_sprites[PIPE_B] = 2;
777 		info->num_sprites[PIPE_C] = 1;
778 	} else if (IS_VALLEYVIEW(dev_priv) || IS_CHERRYVIEW(dev_priv)) {
779 		for_each_pipe(dev_priv, pipe)
780 			info->num_sprites[pipe] = 2;
781 	} else if (INTEL_GEN(dev_priv) >= 5 || IS_G4X(dev_priv)) {
782 		for_each_pipe(dev_priv, pipe)
783 			info->num_sprites[pipe] = 1;
784 	}
785 
786 	if (i915_modparams.disable_display) {
787 		DRM_INFO("Display disabled (module parameter)\n");
788 		info->num_pipes = 0;
789 	} else if (HAS_DISPLAY(dev_priv) &&
790 		   (IS_GEN7(dev_priv) || IS_GEN8(dev_priv)) &&
791 		   HAS_PCH_SPLIT(dev_priv)) {
792 		u32 fuse_strap = I915_READ(FUSE_STRAP);
793 		u32 sfuse_strap = I915_READ(SFUSE_STRAP);
794 
795 		/*
796 		 * SFUSE_STRAP is supposed to have a bit signalling the display
797 		 * is fused off. Unfortunately it seems that, at least in
798 		 * certain cases, fused off display means that PCH display
799 		 * reads don't land anywhere. In that case, we read 0s.
800 		 *
801 		 * On CPT/PPT, we can detect this case as SFUSE_STRAP_FUSE_LOCK
802 		 * should be set when taking over after the firmware.
803 		 */
804 		if (fuse_strap & ILK_INTERNAL_DISPLAY_DISABLE ||
805 		    sfuse_strap & SFUSE_STRAP_DISPLAY_DISABLED ||
806 		    (HAS_PCH_CPT(dev_priv) &&
807 		     !(sfuse_strap & SFUSE_STRAP_FUSE_LOCK))) {
808 			DRM_INFO("Display fused off, disabling\n");
809 			info->num_pipes = 0;
810 		} else if (fuse_strap & IVB_PIPE_C_DISABLE) {
811 			DRM_INFO("PipeC fused off\n");
812 			info->num_pipes -= 1;
813 		}
814 	} else if (HAS_DISPLAY(dev_priv) && IS_GEN9(dev_priv)) {
815 		u32 dfsm = I915_READ(SKL_DFSM);
816 		u8 disabled_mask = 0;
817 		bool invalid;
818 		int num_bits;
819 
820 		if (dfsm & SKL_DFSM_PIPE_A_DISABLE)
821 			disabled_mask |= BIT(PIPE_A);
822 		if (dfsm & SKL_DFSM_PIPE_B_DISABLE)
823 			disabled_mask |= BIT(PIPE_B);
824 		if (dfsm & SKL_DFSM_PIPE_C_DISABLE)
825 			disabled_mask |= BIT(PIPE_C);
826 
827 		num_bits = hweight8(disabled_mask);
828 
829 		switch (disabled_mask) {
830 		case BIT(PIPE_A):
831 		case BIT(PIPE_B):
832 		case BIT(PIPE_A) | BIT(PIPE_B):
833 		case BIT(PIPE_A) | BIT(PIPE_C):
834 			invalid = true;
835 			break;
836 		default:
837 			invalid = false;
838 		}
839 
840 		if (num_bits > info->num_pipes || invalid)
841 			DRM_ERROR("invalid pipe fuse configuration: 0x%x\n",
842 				  disabled_mask);
843 		else
844 			info->num_pipes -= num_bits;
845 	}
846 
847 	/* Initialize slice/subslice/EU info */
848 	if (IS_HASWELL(dev_priv))
849 		haswell_sseu_info_init(dev_priv);
850 	else if (IS_CHERRYVIEW(dev_priv))
851 		cherryview_sseu_info_init(dev_priv);
852 	else if (IS_BROADWELL(dev_priv))
853 		broadwell_sseu_info_init(dev_priv);
854 	else if (IS_GEN9(dev_priv))
855 		gen9_sseu_info_init(dev_priv);
856 	else if (IS_GEN10(dev_priv))
857 		gen10_sseu_info_init(dev_priv);
858 	else if (INTEL_GEN(dev_priv) >= 11)
859 		gen11_sseu_info_init(dev_priv);
860 
861 	if (IS_GEN6(dev_priv) && intel_vtd_active()) {
862 		DRM_INFO("Disabling ppGTT for VT-d support\n");
863 		info->ppgtt = INTEL_PPGTT_NONE;
864 	}
865 
866 	/* Initialize command stream timestamp frequency */
867 	info->cs_timestamp_frequency_khz = read_timestamp_frequency(dev_priv);
868 }
869 
870 void intel_driver_caps_print(const struct intel_driver_caps *caps,
871 			     struct drm_printer *p)
872 {
873 	drm_printf(p, "Has logical contexts? %s\n",
874 		   yesno(caps->has_logical_contexts));
875 	drm_printf(p, "scheduler: %x\n", caps->scheduler);
876 }
877 
878 /*
879  * Determine which engines are fused off in our particular hardware. Since the
880  * fuse register is in the blitter powerwell, we need forcewake to be ready at
881  * this point (but later we need to prune the forcewake domains for engines that
882  * are indeed fused off).
883  */
884 void intel_device_info_init_mmio(struct drm_i915_private *dev_priv)
885 {
886 	struct intel_device_info *info = mkwrite_device_info(dev_priv);
887 	u32 media_fuse;
888 	unsigned int i;
889 
890 	if (INTEL_GEN(dev_priv) < 11)
891 		return;
892 
893 	media_fuse = ~I915_READ(GEN11_GT_VEBOX_VDBOX_DISABLE);
894 
895 	info->vdbox_enable = media_fuse & GEN11_GT_VDBOX_DISABLE_MASK;
896 	info->vebox_enable = (media_fuse & GEN11_GT_VEBOX_DISABLE_MASK) >>
897 			     GEN11_GT_VEBOX_DISABLE_SHIFT;
898 
899 	DRM_DEBUG_DRIVER("vdbox enable: %04x\n", info->vdbox_enable);
900 	for (i = 0; i < I915_MAX_VCS; i++) {
901 		if (!HAS_ENGINE(dev_priv, _VCS(i)))
902 			continue;
903 
904 		if (!(BIT(i) & info->vdbox_enable)) {
905 			info->ring_mask &= ~ENGINE_MASK(_VCS(i));
906 			DRM_DEBUG_DRIVER("vcs%u fused off\n", i);
907 		}
908 	}
909 
910 	DRM_DEBUG_DRIVER("vebox enable: %04x\n", info->vebox_enable);
911 	for (i = 0; i < I915_MAX_VECS; i++) {
912 		if (!HAS_ENGINE(dev_priv, _VECS(i)))
913 			continue;
914 
915 		if (!(BIT(i) & info->vebox_enable)) {
916 			info->ring_mask &= ~ENGINE_MASK(_VECS(i));
917 			DRM_DEBUG_DRIVER("vecs%u fused off\n", i);
918 		}
919 	}
920 }
921