xref: /openbmc/linux/drivers/gpu/drm/i915/gt/intel_sseu.c (revision ed84ef1c)
1 // SPDX-License-Identifier: MIT
2 /*
3  * Copyright © 2019 Intel Corporation
4  */
5 
6 #include "i915_drv.h"
7 #include "intel_lrc_reg.h"
8 #include "intel_sseu.h"
9 
10 void intel_sseu_set_info(struct sseu_dev_info *sseu, u8 max_slices,
11 			 u8 max_subslices, u8 max_eus_per_subslice)
12 {
13 	sseu->max_slices = max_slices;
14 	sseu->max_subslices = max_subslices;
15 	sseu->max_eus_per_subslice = max_eus_per_subslice;
16 
17 	sseu->ss_stride = GEN_SSEU_STRIDE(sseu->max_subslices);
18 	GEM_BUG_ON(sseu->ss_stride > GEN_MAX_SUBSLICE_STRIDE);
19 	sseu->eu_stride = GEN_SSEU_STRIDE(sseu->max_eus_per_subslice);
20 	GEM_BUG_ON(sseu->eu_stride > GEN_MAX_EU_STRIDE);
21 }
22 
23 unsigned int
24 intel_sseu_subslice_total(const struct sseu_dev_info *sseu)
25 {
26 	unsigned int i, total = 0;
27 
28 	for (i = 0; i < ARRAY_SIZE(sseu->subslice_mask); i++)
29 		total += hweight8(sseu->subslice_mask[i]);
30 
31 	return total;
32 }
33 
34 u32 intel_sseu_get_subslices(const struct sseu_dev_info *sseu, u8 slice)
35 {
36 	int i, offset = slice * sseu->ss_stride;
37 	u32 mask = 0;
38 
39 	GEM_BUG_ON(slice >= sseu->max_slices);
40 
41 	for (i = 0; i < sseu->ss_stride; i++)
42 		mask |= (u32)sseu->subslice_mask[offset + i] <<
43 			i * BITS_PER_BYTE;
44 
45 	return mask;
46 }
47 
48 void intel_sseu_set_subslices(struct sseu_dev_info *sseu, int slice,
49 			      u32 ss_mask)
50 {
51 	int offset = slice * sseu->ss_stride;
52 
53 	memcpy(&sseu->subslice_mask[offset], &ss_mask, sseu->ss_stride);
54 }
55 
56 unsigned int
57 intel_sseu_subslices_per_slice(const struct sseu_dev_info *sseu, u8 slice)
58 {
59 	return hweight32(intel_sseu_get_subslices(sseu, slice));
60 }
61 
62 static int sseu_eu_idx(const struct sseu_dev_info *sseu, int slice,
63 		       int subslice)
64 {
65 	int slice_stride = sseu->max_subslices * sseu->eu_stride;
66 
67 	return slice * slice_stride + subslice * sseu->eu_stride;
68 }
69 
70 static u16 sseu_get_eus(const struct sseu_dev_info *sseu, int slice,
71 			int subslice)
72 {
73 	int i, offset = sseu_eu_idx(sseu, slice, subslice);
74 	u16 eu_mask = 0;
75 
76 	for (i = 0; i < sseu->eu_stride; i++)
77 		eu_mask |=
78 			((u16)sseu->eu_mask[offset + i]) << (i * BITS_PER_BYTE);
79 
80 	return eu_mask;
81 }
82 
83 static void sseu_set_eus(struct sseu_dev_info *sseu, int slice, int subslice,
84 			 u16 eu_mask)
85 {
86 	int i, offset = sseu_eu_idx(sseu, slice, subslice);
87 
88 	for (i = 0; i < sseu->eu_stride; i++)
89 		sseu->eu_mask[offset + i] =
90 			(eu_mask >> (BITS_PER_BYTE * i)) & 0xff;
91 }
92 
93 static u16 compute_eu_total(const struct sseu_dev_info *sseu)
94 {
95 	u16 i, total = 0;
96 
97 	for (i = 0; i < ARRAY_SIZE(sseu->eu_mask); i++)
98 		total += hweight8(sseu->eu_mask[i]);
99 
100 	return total;
101 }
102 
103 static void gen11_compute_sseu_info(struct sseu_dev_info *sseu,
104 				    u8 s_en, u32 ss_en, u16 eu_en)
105 {
106 	int s, ss;
107 
108 	/* ss_en represents entire subslice mask across all slices */
109 	GEM_BUG_ON(sseu->max_slices * sseu->max_subslices >
110 		   sizeof(ss_en) * BITS_PER_BYTE);
111 
112 	for (s = 0; s < sseu->max_slices; s++) {
113 		if ((s_en & BIT(s)) == 0)
114 			continue;
115 
116 		sseu->slice_mask |= BIT(s);
117 
118 		intel_sseu_set_subslices(sseu, s, ss_en);
119 
120 		for (ss = 0; ss < sseu->max_subslices; ss++)
121 			if (intel_sseu_has_subslice(sseu, s, ss))
122 				sseu_set_eus(sseu, s, ss, eu_en);
123 	}
124 	sseu->eu_per_subslice = hweight16(eu_en);
125 	sseu->eu_total = compute_eu_total(sseu);
126 }
127 
128 static void gen12_sseu_info_init(struct intel_gt *gt)
129 {
130 	struct sseu_dev_info *sseu = &gt->info.sseu;
131 	struct intel_uncore *uncore = gt->uncore;
132 	u32 dss_en;
133 	u16 eu_en = 0;
134 	u8 eu_en_fuse;
135 	u8 s_en;
136 	int eu;
137 
138 	/*
139 	 * Gen12 has Dual-Subslices, which behave similarly to 2 gen11 SS.
140 	 * Instead of splitting these, provide userspace with an array
141 	 * of DSS to more closely represent the hardware resource.
142 	 *
143 	 * In addition, the concept of slice has been removed in Xe_HP.
144 	 * To be compatible with prior generations, assume a single slice
145 	 * across the entire device. Then calculate out the DSS for each
146 	 * workload type within that software slice.
147 	 */
148 	if (IS_DG2(gt->i915) || IS_XEHPSDV(gt->i915))
149 		intel_sseu_set_info(sseu, 1, 32, 16);
150 	else
151 		intel_sseu_set_info(sseu, 1, 6, 16);
152 
153 	/*
154 	 * As mentioned above, Xe_HP does not have the concept of a slice.
155 	 * Enable one for software backwards compatibility.
156 	 */
157 	if (GRAPHICS_VER_FULL(gt->i915) >= IP_VER(12, 50))
158 		s_en = 0x1;
159 	else
160 		s_en = intel_uncore_read(uncore, GEN11_GT_SLICE_ENABLE) &
161 		       GEN11_GT_S_ENA_MASK;
162 
163 	dss_en = intel_uncore_read(uncore, GEN12_GT_DSS_ENABLE);
164 
165 	/* one bit per pair of EUs */
166 	if (GRAPHICS_VER_FULL(gt->i915) >= IP_VER(12, 50))
167 		eu_en_fuse = intel_uncore_read(uncore, XEHP_EU_ENABLE) & XEHP_EU_ENA_MASK;
168 	else
169 		eu_en_fuse = ~(intel_uncore_read(uncore, GEN11_EU_DISABLE) &
170 			       GEN11_EU_DIS_MASK);
171 
172 	for (eu = 0; eu < sseu->max_eus_per_subslice / 2; eu++)
173 		if (eu_en_fuse & BIT(eu))
174 			eu_en |= BIT(eu * 2) | BIT(eu * 2 + 1);
175 
176 	gen11_compute_sseu_info(sseu, s_en, dss_en, eu_en);
177 
178 	/* TGL only supports slice-level power gating */
179 	sseu->has_slice_pg = 1;
180 }
181 
182 static void gen11_sseu_info_init(struct intel_gt *gt)
183 {
184 	struct sseu_dev_info *sseu = &gt->info.sseu;
185 	struct intel_uncore *uncore = gt->uncore;
186 	u32 ss_en;
187 	u8 eu_en;
188 	u8 s_en;
189 
190 	if (IS_JSL_EHL(gt->i915))
191 		intel_sseu_set_info(sseu, 1, 4, 8);
192 	else
193 		intel_sseu_set_info(sseu, 1, 8, 8);
194 
195 	s_en = intel_uncore_read(uncore, GEN11_GT_SLICE_ENABLE) &
196 		GEN11_GT_S_ENA_MASK;
197 	ss_en = ~intel_uncore_read(uncore, GEN11_GT_SUBSLICE_DISABLE);
198 
199 	eu_en = ~(intel_uncore_read(uncore, GEN11_EU_DISABLE) &
200 		  GEN11_EU_DIS_MASK);
201 
202 	gen11_compute_sseu_info(sseu, s_en, ss_en, eu_en);
203 
204 	/* ICL has no power gating restrictions. */
205 	sseu->has_slice_pg = 1;
206 	sseu->has_subslice_pg = 1;
207 	sseu->has_eu_pg = 1;
208 }
209 
210 static void cherryview_sseu_info_init(struct intel_gt *gt)
211 {
212 	struct sseu_dev_info *sseu = &gt->info.sseu;
213 	u32 fuse;
214 	u8 subslice_mask = 0;
215 
216 	fuse = intel_uncore_read(gt->uncore, CHV_FUSE_GT);
217 
218 	sseu->slice_mask = BIT(0);
219 	intel_sseu_set_info(sseu, 1, 2, 8);
220 
221 	if (!(fuse & CHV_FGT_DISABLE_SS0)) {
222 		u8 disabled_mask =
223 			((fuse & CHV_FGT_EU_DIS_SS0_R0_MASK) >>
224 			 CHV_FGT_EU_DIS_SS0_R0_SHIFT) |
225 			(((fuse & CHV_FGT_EU_DIS_SS0_R1_MASK) >>
226 			  CHV_FGT_EU_DIS_SS0_R1_SHIFT) << 4);
227 
228 		subslice_mask |= BIT(0);
229 		sseu_set_eus(sseu, 0, 0, ~disabled_mask);
230 	}
231 
232 	if (!(fuse & CHV_FGT_DISABLE_SS1)) {
233 		u8 disabled_mask =
234 			((fuse & CHV_FGT_EU_DIS_SS1_R0_MASK) >>
235 			 CHV_FGT_EU_DIS_SS1_R0_SHIFT) |
236 			(((fuse & CHV_FGT_EU_DIS_SS1_R1_MASK) >>
237 			  CHV_FGT_EU_DIS_SS1_R1_SHIFT) << 4);
238 
239 		subslice_mask |= BIT(1);
240 		sseu_set_eus(sseu, 0, 1, ~disabled_mask);
241 	}
242 
243 	intel_sseu_set_subslices(sseu, 0, subslice_mask);
244 
245 	sseu->eu_total = compute_eu_total(sseu);
246 
247 	/*
248 	 * CHV expected to always have a uniform distribution of EU
249 	 * across subslices.
250 	 */
251 	sseu->eu_per_subslice = intel_sseu_subslice_total(sseu) ?
252 		sseu->eu_total /
253 		intel_sseu_subslice_total(sseu) :
254 		0;
255 	/*
256 	 * CHV supports subslice power gating on devices with more than
257 	 * one subslice, and supports EU power gating on devices with
258 	 * more than one EU pair per subslice.
259 	 */
260 	sseu->has_slice_pg = 0;
261 	sseu->has_subslice_pg = intel_sseu_subslice_total(sseu) > 1;
262 	sseu->has_eu_pg = (sseu->eu_per_subslice > 2);
263 }
264 
265 static void gen9_sseu_info_init(struct intel_gt *gt)
266 {
267 	struct drm_i915_private *i915 = gt->i915;
268 	struct intel_device_info *info = mkwrite_device_info(i915);
269 	struct sseu_dev_info *sseu = &gt->info.sseu;
270 	struct intel_uncore *uncore = gt->uncore;
271 	u32 fuse2, eu_disable, subslice_mask;
272 	const u8 eu_mask = 0xff;
273 	int s, ss;
274 
275 	fuse2 = intel_uncore_read(uncore, GEN8_FUSE2);
276 	sseu->slice_mask = (fuse2 & GEN8_F2_S_ENA_MASK) >> GEN8_F2_S_ENA_SHIFT;
277 
278 	/* BXT has a single slice and at most 3 subslices. */
279 	intel_sseu_set_info(sseu, IS_GEN9_LP(i915) ? 1 : 3,
280 			    IS_GEN9_LP(i915) ? 3 : 4, 8);
281 
282 	/*
283 	 * The subslice disable field is global, i.e. it applies
284 	 * to each of the enabled slices.
285 	 */
286 	subslice_mask = (1 << sseu->max_subslices) - 1;
287 	subslice_mask &= ~((fuse2 & GEN9_F2_SS_DIS_MASK) >>
288 			   GEN9_F2_SS_DIS_SHIFT);
289 
290 	/*
291 	 * Iterate through enabled slices and subslices to
292 	 * count the total enabled EU.
293 	 */
294 	for (s = 0; s < sseu->max_slices; s++) {
295 		if (!(sseu->slice_mask & BIT(s)))
296 			/* skip disabled slice */
297 			continue;
298 
299 		intel_sseu_set_subslices(sseu, s, subslice_mask);
300 
301 		eu_disable = intel_uncore_read(uncore, GEN9_EU_DISABLE(s));
302 		for (ss = 0; ss < sseu->max_subslices; ss++) {
303 			int eu_per_ss;
304 			u8 eu_disabled_mask;
305 
306 			if (!intel_sseu_has_subslice(sseu, s, ss))
307 				/* skip disabled subslice */
308 				continue;
309 
310 			eu_disabled_mask = (eu_disable >> (ss * 8)) & eu_mask;
311 
312 			sseu_set_eus(sseu, s, ss, ~eu_disabled_mask);
313 
314 			eu_per_ss = sseu->max_eus_per_subslice -
315 				hweight8(eu_disabled_mask);
316 
317 			/*
318 			 * Record which subslice(s) has(have) 7 EUs. we
319 			 * can tune the hash used to spread work among
320 			 * subslices if they are unbalanced.
321 			 */
322 			if (eu_per_ss == 7)
323 				sseu->subslice_7eu[s] |= BIT(ss);
324 		}
325 	}
326 
327 	sseu->eu_total = compute_eu_total(sseu);
328 
329 	/*
330 	 * SKL is expected to always have a uniform distribution
331 	 * of EU across subslices with the exception that any one
332 	 * EU in any one subslice may be fused off for die
333 	 * recovery. BXT is expected to be perfectly uniform in EU
334 	 * distribution.
335 	 */
336 	sseu->eu_per_subslice =
337 		intel_sseu_subslice_total(sseu) ?
338 		DIV_ROUND_UP(sseu->eu_total, intel_sseu_subslice_total(sseu)) :
339 		0;
340 
341 	/*
342 	 * SKL+ supports slice power gating on devices with more than
343 	 * one slice, and supports EU power gating on devices with
344 	 * more than one EU pair per subslice. BXT+ supports subslice
345 	 * power gating on devices with more than one subslice, and
346 	 * supports EU power gating on devices with more than one EU
347 	 * pair per subslice.
348 	 */
349 	sseu->has_slice_pg =
350 		!IS_GEN9_LP(i915) && hweight8(sseu->slice_mask) > 1;
351 	sseu->has_subslice_pg =
352 		IS_GEN9_LP(i915) && intel_sseu_subslice_total(sseu) > 1;
353 	sseu->has_eu_pg = sseu->eu_per_subslice > 2;
354 
355 	if (IS_GEN9_LP(i915)) {
356 #define IS_SS_DISABLED(ss)	(!(sseu->subslice_mask[0] & BIT(ss)))
357 		info->has_pooled_eu = hweight8(sseu->subslice_mask[0]) == 3;
358 
359 		sseu->min_eu_in_pool = 0;
360 		if (info->has_pooled_eu) {
361 			if (IS_SS_DISABLED(2) || IS_SS_DISABLED(0))
362 				sseu->min_eu_in_pool = 3;
363 			else if (IS_SS_DISABLED(1))
364 				sseu->min_eu_in_pool = 6;
365 			else
366 				sseu->min_eu_in_pool = 9;
367 		}
368 #undef IS_SS_DISABLED
369 	}
370 }
371 
372 static void bdw_sseu_info_init(struct intel_gt *gt)
373 {
374 	struct sseu_dev_info *sseu = &gt->info.sseu;
375 	struct intel_uncore *uncore = gt->uncore;
376 	int s, ss;
377 	u32 fuse2, subslice_mask, eu_disable[3]; /* s_max */
378 	u32 eu_disable0, eu_disable1, eu_disable2;
379 
380 	fuse2 = intel_uncore_read(uncore, GEN8_FUSE2);
381 	sseu->slice_mask = (fuse2 & GEN8_F2_S_ENA_MASK) >> GEN8_F2_S_ENA_SHIFT;
382 	intel_sseu_set_info(sseu, 3, 3, 8);
383 
384 	/*
385 	 * The subslice disable field is global, i.e. it applies
386 	 * to each of the enabled slices.
387 	 */
388 	subslice_mask = GENMASK(sseu->max_subslices - 1, 0);
389 	subslice_mask &= ~((fuse2 & GEN8_F2_SS_DIS_MASK) >>
390 			   GEN8_F2_SS_DIS_SHIFT);
391 	eu_disable0 = intel_uncore_read(uncore, GEN8_EU_DISABLE0);
392 	eu_disable1 = intel_uncore_read(uncore, GEN8_EU_DISABLE1);
393 	eu_disable2 = intel_uncore_read(uncore, GEN8_EU_DISABLE2);
394 	eu_disable[0] = eu_disable0 & GEN8_EU_DIS0_S0_MASK;
395 	eu_disable[1] = (eu_disable0 >> GEN8_EU_DIS0_S1_SHIFT) |
396 		((eu_disable1 & GEN8_EU_DIS1_S1_MASK) <<
397 		 (32 - GEN8_EU_DIS0_S1_SHIFT));
398 	eu_disable[2] = (eu_disable1 >> GEN8_EU_DIS1_S2_SHIFT) |
399 		((eu_disable2 & GEN8_EU_DIS2_S2_MASK) <<
400 		 (32 - GEN8_EU_DIS1_S2_SHIFT));
401 
402 	/*
403 	 * Iterate through enabled slices and subslices to
404 	 * count the total enabled EU.
405 	 */
406 	for (s = 0; s < sseu->max_slices; s++) {
407 		if (!(sseu->slice_mask & BIT(s)))
408 			/* skip disabled slice */
409 			continue;
410 
411 		intel_sseu_set_subslices(sseu, s, subslice_mask);
412 
413 		for (ss = 0; ss < sseu->max_subslices; ss++) {
414 			u8 eu_disabled_mask;
415 			u32 n_disabled;
416 
417 			if (!intel_sseu_has_subslice(sseu, s, ss))
418 				/* skip disabled subslice */
419 				continue;
420 
421 			eu_disabled_mask =
422 				eu_disable[s] >> (ss * sseu->max_eus_per_subslice);
423 
424 			sseu_set_eus(sseu, s, ss, ~eu_disabled_mask);
425 
426 			n_disabled = hweight8(eu_disabled_mask);
427 
428 			/*
429 			 * Record which subslices have 7 EUs.
430 			 */
431 			if (sseu->max_eus_per_subslice - n_disabled == 7)
432 				sseu->subslice_7eu[s] |= 1 << ss;
433 		}
434 	}
435 
436 	sseu->eu_total = compute_eu_total(sseu);
437 
438 	/*
439 	 * BDW is expected to always have a uniform distribution of EU across
440 	 * subslices with the exception that any one EU in any one subslice may
441 	 * be fused off for die recovery.
442 	 */
443 	sseu->eu_per_subslice =
444 		intel_sseu_subslice_total(sseu) ?
445 		DIV_ROUND_UP(sseu->eu_total, intel_sseu_subslice_total(sseu)) :
446 		0;
447 
448 	/*
449 	 * BDW supports slice power gating on devices with more than
450 	 * one slice.
451 	 */
452 	sseu->has_slice_pg = hweight8(sseu->slice_mask) > 1;
453 	sseu->has_subslice_pg = 0;
454 	sseu->has_eu_pg = 0;
455 }
456 
457 static void hsw_sseu_info_init(struct intel_gt *gt)
458 {
459 	struct drm_i915_private *i915 = gt->i915;
460 	struct sseu_dev_info *sseu = &gt->info.sseu;
461 	u32 fuse1;
462 	u8 subslice_mask = 0;
463 	int s, ss;
464 
465 	/*
466 	 * There isn't a register to tell us how many slices/subslices. We
467 	 * work off the PCI-ids here.
468 	 */
469 	switch (INTEL_INFO(i915)->gt) {
470 	default:
471 		MISSING_CASE(INTEL_INFO(i915)->gt);
472 		fallthrough;
473 	case 1:
474 		sseu->slice_mask = BIT(0);
475 		subslice_mask = BIT(0);
476 		break;
477 	case 2:
478 		sseu->slice_mask = BIT(0);
479 		subslice_mask = BIT(0) | BIT(1);
480 		break;
481 	case 3:
482 		sseu->slice_mask = BIT(0) | BIT(1);
483 		subslice_mask = BIT(0) | BIT(1);
484 		break;
485 	}
486 
487 	fuse1 = intel_uncore_read(gt->uncore, HSW_PAVP_FUSE1);
488 	switch ((fuse1 & HSW_F1_EU_DIS_MASK) >> HSW_F1_EU_DIS_SHIFT) {
489 	default:
490 		MISSING_CASE((fuse1 & HSW_F1_EU_DIS_MASK) >>
491 			     HSW_F1_EU_DIS_SHIFT);
492 		fallthrough;
493 	case HSW_F1_EU_DIS_10EUS:
494 		sseu->eu_per_subslice = 10;
495 		break;
496 	case HSW_F1_EU_DIS_8EUS:
497 		sseu->eu_per_subslice = 8;
498 		break;
499 	case HSW_F1_EU_DIS_6EUS:
500 		sseu->eu_per_subslice = 6;
501 		break;
502 	}
503 
504 	intel_sseu_set_info(sseu, hweight8(sseu->slice_mask),
505 			    hweight8(subslice_mask),
506 			    sseu->eu_per_subslice);
507 
508 	for (s = 0; s < sseu->max_slices; s++) {
509 		intel_sseu_set_subslices(sseu, s, subslice_mask);
510 
511 		for (ss = 0; ss < sseu->max_subslices; ss++) {
512 			sseu_set_eus(sseu, s, ss,
513 				     (1UL << sseu->eu_per_subslice) - 1);
514 		}
515 	}
516 
517 	sseu->eu_total = compute_eu_total(sseu);
518 
519 	/* No powergating for you. */
520 	sseu->has_slice_pg = 0;
521 	sseu->has_subslice_pg = 0;
522 	sseu->has_eu_pg = 0;
523 }
524 
525 void intel_sseu_info_init(struct intel_gt *gt)
526 {
527 	struct drm_i915_private *i915 = gt->i915;
528 
529 	if (IS_HASWELL(i915))
530 		hsw_sseu_info_init(gt);
531 	else if (IS_CHERRYVIEW(i915))
532 		cherryview_sseu_info_init(gt);
533 	else if (IS_BROADWELL(i915))
534 		bdw_sseu_info_init(gt);
535 	else if (GRAPHICS_VER(i915) == 9)
536 		gen9_sseu_info_init(gt);
537 	else if (GRAPHICS_VER(i915) == 11)
538 		gen11_sseu_info_init(gt);
539 	else if (GRAPHICS_VER(i915) >= 12)
540 		gen12_sseu_info_init(gt);
541 }
542 
543 u32 intel_sseu_make_rpcs(struct intel_gt *gt,
544 			 const struct intel_sseu *req_sseu)
545 {
546 	struct drm_i915_private *i915 = gt->i915;
547 	const struct sseu_dev_info *sseu = &gt->info.sseu;
548 	bool subslice_pg = sseu->has_subslice_pg;
549 	u8 slices, subslices;
550 	u32 rpcs = 0;
551 
552 	/*
553 	 * No explicit RPCS request is needed to ensure full
554 	 * slice/subslice/EU enablement prior to Gen9.
555 	 */
556 	if (GRAPHICS_VER(i915) < 9)
557 		return 0;
558 
559 	/*
560 	 * If i915/perf is active, we want a stable powergating configuration
561 	 * on the system. Use the configuration pinned by i915/perf.
562 	 */
563 	if (i915->perf.exclusive_stream)
564 		req_sseu = &i915->perf.sseu;
565 
566 	slices = hweight8(req_sseu->slice_mask);
567 	subslices = hweight8(req_sseu->subslice_mask);
568 
569 	/*
570 	 * Since the SScount bitfield in GEN8_R_PWR_CLK_STATE is only three bits
571 	 * wide and Icelake has up to eight subslices, specfial programming is
572 	 * needed in order to correctly enable all subslices.
573 	 *
574 	 * According to documentation software must consider the configuration
575 	 * as 2x4x8 and hardware will translate this to 1x8x8.
576 	 *
577 	 * Furthemore, even though SScount is three bits, maximum documented
578 	 * value for it is four. From this some rules/restrictions follow:
579 	 *
580 	 * 1.
581 	 * If enabled subslice count is greater than four, two whole slices must
582 	 * be enabled instead.
583 	 *
584 	 * 2.
585 	 * When more than one slice is enabled, hardware ignores the subslice
586 	 * count altogether.
587 	 *
588 	 * From these restrictions it follows that it is not possible to enable
589 	 * a count of subslices between the SScount maximum of four restriction,
590 	 * and the maximum available number on a particular SKU. Either all
591 	 * subslices are enabled, or a count between one and four on the first
592 	 * slice.
593 	 */
594 	if (GRAPHICS_VER(i915) == 11 &&
595 	    slices == 1 &&
596 	    subslices > min_t(u8, 4, hweight8(sseu->subslice_mask[0]) / 2)) {
597 		GEM_BUG_ON(subslices & 1);
598 
599 		subslice_pg = false;
600 		slices *= 2;
601 	}
602 
603 	/*
604 	 * Starting in Gen9, render power gating can leave
605 	 * slice/subslice/EU in a partially enabled state. We
606 	 * must make an explicit request through RPCS for full
607 	 * enablement.
608 	 */
609 	if (sseu->has_slice_pg) {
610 		u32 mask, val = slices;
611 
612 		if (GRAPHICS_VER(i915) >= 11) {
613 			mask = GEN11_RPCS_S_CNT_MASK;
614 			val <<= GEN11_RPCS_S_CNT_SHIFT;
615 		} else {
616 			mask = GEN8_RPCS_S_CNT_MASK;
617 			val <<= GEN8_RPCS_S_CNT_SHIFT;
618 		}
619 
620 		GEM_BUG_ON(val & ~mask);
621 		val &= mask;
622 
623 		rpcs |= GEN8_RPCS_ENABLE | GEN8_RPCS_S_CNT_ENABLE | val;
624 	}
625 
626 	if (subslice_pg) {
627 		u32 val = subslices;
628 
629 		val <<= GEN8_RPCS_SS_CNT_SHIFT;
630 
631 		GEM_BUG_ON(val & ~GEN8_RPCS_SS_CNT_MASK);
632 		val &= GEN8_RPCS_SS_CNT_MASK;
633 
634 		rpcs |= GEN8_RPCS_ENABLE | GEN8_RPCS_SS_CNT_ENABLE | val;
635 	}
636 
637 	if (sseu->has_eu_pg) {
638 		u32 val;
639 
640 		val = req_sseu->min_eus_per_subslice << GEN8_RPCS_EU_MIN_SHIFT;
641 		GEM_BUG_ON(val & ~GEN8_RPCS_EU_MIN_MASK);
642 		val &= GEN8_RPCS_EU_MIN_MASK;
643 
644 		rpcs |= val;
645 
646 		val = req_sseu->max_eus_per_subslice << GEN8_RPCS_EU_MAX_SHIFT;
647 		GEM_BUG_ON(val & ~GEN8_RPCS_EU_MAX_MASK);
648 		val &= GEN8_RPCS_EU_MAX_MASK;
649 
650 		rpcs |= val;
651 
652 		rpcs |= GEN8_RPCS_ENABLE;
653 	}
654 
655 	return rpcs;
656 }
657 
658 void intel_sseu_dump(const struct sseu_dev_info *sseu, struct drm_printer *p)
659 {
660 	int s;
661 
662 	drm_printf(p, "slice total: %u, mask=%04x\n",
663 		   hweight8(sseu->slice_mask), sseu->slice_mask);
664 	drm_printf(p, "subslice total: %u\n", intel_sseu_subslice_total(sseu));
665 	for (s = 0; s < sseu->max_slices; s++) {
666 		drm_printf(p, "slice%d: %u subslices, mask=%08x\n",
667 			   s, intel_sseu_subslices_per_slice(sseu, s),
668 			   intel_sseu_get_subslices(sseu, s));
669 	}
670 	drm_printf(p, "EU total: %u\n", sseu->eu_total);
671 	drm_printf(p, "EU per subslice: %u\n", sseu->eu_per_subslice);
672 	drm_printf(p, "has slice power gating: %s\n",
673 		   yesno(sseu->has_slice_pg));
674 	drm_printf(p, "has subslice power gating: %s\n",
675 		   yesno(sseu->has_subslice_pg));
676 	drm_printf(p, "has EU power gating: %s\n", yesno(sseu->has_eu_pg));
677 }
678 
679 void intel_sseu_print_topology(const struct sseu_dev_info *sseu,
680 			       struct drm_printer *p)
681 {
682 	int s, ss;
683 
684 	if (sseu->max_slices == 0) {
685 		drm_printf(p, "Unavailable\n");
686 		return;
687 	}
688 
689 	for (s = 0; s < sseu->max_slices; s++) {
690 		drm_printf(p, "slice%d: %u subslice(s) (0x%08x):\n",
691 			   s, intel_sseu_subslices_per_slice(sseu, s),
692 			   intel_sseu_get_subslices(sseu, s));
693 
694 		for (ss = 0; ss < sseu->max_subslices; ss++) {
695 			u16 enabled_eus = sseu_get_eus(sseu, s, ss);
696 
697 			drm_printf(p, "\tsubslice%d: %u EUs (0x%hx)\n",
698 				   ss, hweight16(enabled_eus), enabled_eus);
699 		}
700 	}
701 }
702 
703 u16 intel_slicemask_from_dssmask(u64 dss_mask, int dss_per_slice)
704 {
705 	u16 slice_mask = 0;
706 	int i;
707 
708 	WARN_ON(sizeof(dss_mask) * 8 / dss_per_slice > 8 * sizeof(slice_mask));
709 
710 	for (i = 0; dss_mask; i++) {
711 		if (dss_mask & GENMASK(dss_per_slice - 1, 0))
712 			slice_mask |= BIT(i);
713 
714 		dss_mask >>= dss_per_slice;
715 	}
716 
717 	return slice_mask;
718 }
719 
720