drm/i915: Fix unsigned overflow when calculating total data rate, v2.
[linux-2.6-block.git] / drivers / gpu / drm / i915 / intel_device_info.c
CommitLineData
94b4f3ba
CW
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
a8c9b849
MW
25#include <drm/drm_print.h>
26
b978520d 27#include "intel_device_info.h"
94b4f3ba
CW
28#include "i915_drv.h"
29
2e0d26f8
JN
30#define PLATFORM_NAME(x) [INTEL_##x] = #x
31static 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),
c0f86832
JN
42 PLATFORM_NAME(I965G),
43 PLATFORM_NAME(I965GM),
f69c11ae
JN
44 PLATFORM_NAME(G45),
45 PLATFORM_NAME(GM45),
2e0d26f8
JN
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),
71851fa8 57 PLATFORM_NAME(COFFEELAKE),
413f3c19 58 PLATFORM_NAME(CANNONLAKE),
41231001 59 PLATFORM_NAME(ICELAKE),
2e0d26f8
JN
60};
61#undef PLATFORM_NAME
62
63const char *intel_platform_name(enum intel_platform platform)
64{
9160095c
JN
65 BUILD_BUG_ON(ARRAY_SIZE(platform_names) != INTEL_MAX_PLATFORMS);
66
2e0d26f8
JN
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
a8c9b849
MW
74void 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
5fbbe8d4
MW
82static void sseu_dump(const struct sseu_dev_info *sseu, struct drm_printer *p)
83{
8cc76693
LL
84 int s;
85
0ef904bb
TU
86 drm_printf(p, "slice total: %u, mask=%04x\n",
87 hweight8(sseu->slice_mask), sseu->slice_mask);
5fbbe8d4 88 drm_printf(p, "subslice total: %u\n", sseu_subslice_total(sseu));
0ef904bb
TU
89 for (s = 0; s < sseu->max_slices; s++) {
90 drm_printf(p, "slice%d: %u subslices, mask=%04x\n",
8cc76693
LL
91 s, hweight8(sseu->subslice_mask[s]),
92 sseu->subslice_mask[s]);
93 }
5fbbe8d4
MW
94 drm_printf(p, "EU total: %u\n", sseu->eu_total);
95 drm_printf(p, "EU per subslice: %u\n", sseu->eu_per_subslice);
96 drm_printf(p, "has slice power gating: %s\n",
97 yesno(sseu->has_slice_pg));
98 drm_printf(p, "has subslice power gating: %s\n",
99 yesno(sseu->has_subslice_pg));
100 drm_printf(p, "has EU power gating: %s\n", yesno(sseu->has_eu_pg));
101}
102
103void intel_device_info_dump_runtime(const struct intel_device_info *info,
104 struct drm_printer *p)
105{
106 sseu_dump(&info->sseu, p);
107
108 drm_printf(p, "CS timestamp frequency: %u kHz\n",
109 info->cs_timestamp_frequency_khz);
110}
111
eb10ed9a
MW
112void intel_device_info_dump(const struct intel_device_info *info,
113 struct drm_printer *p)
94b4f3ba 114{
eb10ed9a
MW
115 struct drm_i915_private *dev_priv =
116 container_of(info, struct drm_i915_private, info);
94b4f3ba 117
eb10ed9a
MW
118 drm_printf(p, "pciid=0x%04x rev=0x%02x platform=%s gen=%i\n",
119 INTEL_DEVID(dev_priv),
120 INTEL_REVID(dev_priv),
121 intel_platform_name(info->platform),
122 info->gen);
a8c9b849 123
eb10ed9a 124 intel_device_info_dump_flags(info, p);
94b4f3ba
CW
125}
126
79e9cd5f
LL
127void intel_device_info_dump_topology(const struct sseu_dev_info *sseu,
128 struct drm_printer *p)
129{
130 int s, ss;
131
132 if (sseu->max_slices == 0) {
133 drm_printf(p, "Unavailable\n");
134 return;
135 }
136
137 for (s = 0; s < sseu->max_slices; s++) {
138 drm_printf(p, "slice%d: %u subslice(s) (0x%hhx):\n",
139 s, hweight8(sseu->subslice_mask[s]),
140 sseu->subslice_mask[s]);
141
142 for (ss = 0; ss < sseu->max_subslices; ss++) {
143 u16 enabled_eus = sseu_get_eus(sseu, s, ss);
144
145 drm_printf(p, "\tsubslice%d: %u EUs (0x%hx)\n",
146 ss, hweight16(enabled_eus), enabled_eus);
147 }
148 }
149}
150
8cc76693
LL
151static u16 compute_eu_total(const struct sseu_dev_info *sseu)
152{
153 u16 i, total = 0;
154
155 for (i = 0; i < ARRAY_SIZE(sseu->eu_mask); i++)
156 total += hweight8(sseu->eu_mask[i]);
157
158 return total;
159}
160
8b5eb5e2
KG
161static void gen11_sseu_info_init(struct drm_i915_private *dev_priv)
162{
163 struct sseu_dev_info *sseu = &mkwrite_device_info(dev_priv)->sseu;
164 u8 s_en;
165 u32 ss_en, ss_en_mask;
166 u8 eu_en;
167 int s;
168
169 sseu->max_slices = 1;
170 sseu->max_subslices = 8;
171 sseu->max_eus_per_subslice = 8;
172
173 s_en = I915_READ(GEN11_GT_SLICE_ENABLE) & GEN11_GT_S_ENA_MASK;
174 ss_en = ~I915_READ(GEN11_GT_SUBSLICE_DISABLE);
175 ss_en_mask = BIT(sseu->max_subslices) - 1;
176 eu_en = ~(I915_READ(GEN11_EU_DISABLE) & GEN11_EU_DIS_MASK);
177
178 for (s = 0; s < sseu->max_slices; s++) {
179 if (s_en & BIT(s)) {
180 int ss_idx = sseu->max_subslices * s;
181 int ss;
182
183 sseu->slice_mask |= BIT(s);
184 sseu->subslice_mask[s] = (ss_en >> ss_idx) & ss_en_mask;
185 for (ss = 0; ss < sseu->max_subslices; ss++) {
186 if (sseu->subslice_mask[s] & BIT(ss))
187 sseu_set_eus(sseu, s, ss, eu_en);
188 }
189 }
190 }
191 sseu->eu_per_subslice = hweight8(eu_en);
192 sseu->eu_total = compute_eu_total(sseu);
193
194 /* ICL has no power gating restrictions. */
195 sseu->has_slice_pg = 1;
196 sseu->has_subslice_pg = 1;
197 sseu->has_eu_pg = 1;
198}
199
4e9767bc
BW
200static void gen10_sseu_info_init(struct drm_i915_private *dev_priv)
201{
202 struct sseu_dev_info *sseu = &mkwrite_device_info(dev_priv)->sseu;
203 const u32 fuse2 = I915_READ(GEN8_FUSE2);
8cc76693
LL
204 int s, ss;
205 const int eu_mask = 0xff;
206 u32 subslice_mask, eu_en;
4e9767bc
BW
207
208 sseu->slice_mask = (fuse2 & GEN10_F2_S_ENA_MASK) >>
209 GEN10_F2_S_ENA_SHIFT;
8cc76693
LL
210 sseu->max_slices = 6;
211 sseu->max_subslices = 4;
212 sseu->max_eus_per_subslice = 8;
4e9767bc 213
8cc76693
LL
214 subslice_mask = (1 << 4) - 1;
215 subslice_mask &= ~((fuse2 & GEN10_F2_SS_DIS_MASK) >>
216 GEN10_F2_SS_DIS_SHIFT);
217
218 /*
219 * Slice0 can have up to 3 subslices, but there are only 2 in
220 * slice1/2.
221 */
222 sseu->subslice_mask[0] = subslice_mask;
223 for (s = 1; s < sseu->max_slices; s++)
224 sseu->subslice_mask[s] = subslice_mask & 0x3;
225
226 /* Slice0 */
227 eu_en = ~I915_READ(GEN8_EU_DISABLE0);
228 for (ss = 0; ss < sseu->max_subslices; ss++)
229 sseu_set_eus(sseu, 0, ss, (eu_en >> (8 * ss)) & eu_mask);
230 /* Slice1 */
231 sseu_set_eus(sseu, 1, 0, (eu_en >> 24) & eu_mask);
232 eu_en = ~I915_READ(GEN8_EU_DISABLE1);
233 sseu_set_eus(sseu, 1, 1, eu_en & eu_mask);
234 /* Slice2 */
235 sseu_set_eus(sseu, 2, 0, (eu_en >> 8) & eu_mask);
236 sseu_set_eus(sseu, 2, 1, (eu_en >> 16) & eu_mask);
237 /* Slice3 */
238 sseu_set_eus(sseu, 3, 0, (eu_en >> 24) & eu_mask);
239 eu_en = ~I915_READ(GEN8_EU_DISABLE2);
240 sseu_set_eus(sseu, 3, 1, eu_en & eu_mask);
241 /* Slice4 */
242 sseu_set_eus(sseu, 4, 0, (eu_en >> 8) & eu_mask);
243 sseu_set_eus(sseu, 4, 1, (eu_en >> 16) & eu_mask);
244 /* Slice5 */
245 sseu_set_eus(sseu, 5, 0, (eu_en >> 24) & eu_mask);
246 eu_en = ~I915_READ(GEN10_EU_DISABLE3);
247 sseu_set_eus(sseu, 5, 1, eu_en & eu_mask);
248
249 /* Do a second pass where we mark the subslices disabled if all their
250 * eus are off.
251 */
252 for (s = 0; s < sseu->max_slices; s++) {
253 for (ss = 0; ss < sseu->max_subslices; ss++) {
254 if (sseu_get_eus(sseu, s, ss) == 0)
255 sseu->subslice_mask[s] &= ~BIT(ss);
256 }
257 }
258
259 sseu->eu_total = compute_eu_total(sseu);
4e9767bc
BW
260
261 /*
262 * CNL is expected to always have a uniform distribution
263 * of EU across subslices with the exception that any one
264 * EU in any one subslice may be fused off for die
265 * recovery.
266 */
267 sseu->eu_per_subslice = sseu_subslice_total(sseu) ?
268 DIV_ROUND_UP(sseu->eu_total,
269 sseu_subslice_total(sseu)) : 0;
270
271 /* No restrictions on Power Gating */
272 sseu->has_slice_pg = 1;
273 sseu->has_subslice_pg = 1;
274 sseu->has_eu_pg = 1;
275}
276
94b4f3ba
CW
277static void cherryview_sseu_info_init(struct drm_i915_private *dev_priv)
278{
43b67998 279 struct sseu_dev_info *sseu = &mkwrite_device_info(dev_priv)->sseu;
8cc76693 280 u32 fuse;
94b4f3ba
CW
281
282 fuse = I915_READ(CHV_FUSE_GT);
283
f08a0c92 284 sseu->slice_mask = BIT(0);
8cc76693
LL
285 sseu->max_slices = 1;
286 sseu->max_subslices = 2;
287 sseu->max_eus_per_subslice = 8;
94b4f3ba
CW
288
289 if (!(fuse & CHV_FGT_DISABLE_SS0)) {
8cc76693
LL
290 u8 disabled_mask =
291 ((fuse & CHV_FGT_EU_DIS_SS0_R0_MASK) >>
292 CHV_FGT_EU_DIS_SS0_R0_SHIFT) |
293 (((fuse & CHV_FGT_EU_DIS_SS0_R1_MASK) >>
294 CHV_FGT_EU_DIS_SS0_R1_SHIFT) << 4);
295
296 sseu->subslice_mask[0] |= BIT(0);
297 sseu_set_eus(sseu, 0, 0, ~disabled_mask);
94b4f3ba
CW
298 }
299
300 if (!(fuse & CHV_FGT_DISABLE_SS1)) {
8cc76693
LL
301 u8 disabled_mask =
302 ((fuse & CHV_FGT_EU_DIS_SS1_R0_MASK) >>
303 CHV_FGT_EU_DIS_SS1_R0_SHIFT) |
304 (((fuse & CHV_FGT_EU_DIS_SS1_R1_MASK) >>
305 CHV_FGT_EU_DIS_SS1_R1_SHIFT) << 4);
306
307 sseu->subslice_mask[0] |= BIT(1);
308 sseu_set_eus(sseu, 0, 1, ~disabled_mask);
94b4f3ba
CW
309 }
310
8cc76693
LL
311 sseu->eu_total = compute_eu_total(sseu);
312
94b4f3ba
CW
313 /*
314 * CHV expected to always have a uniform distribution of EU
315 * across subslices.
316 */
57ec171e
ID
317 sseu->eu_per_subslice = sseu_subslice_total(sseu) ?
318 sseu->eu_total / sseu_subslice_total(sseu) :
94b4f3ba
CW
319 0;
320 /*
321 * CHV supports subslice power gating on devices with more than
322 * one subslice, and supports EU power gating on devices with
323 * more than one EU pair per subslice.
324 */
43b67998 325 sseu->has_slice_pg = 0;
57ec171e 326 sseu->has_subslice_pg = sseu_subslice_total(sseu) > 1;
43b67998 327 sseu->has_eu_pg = (sseu->eu_per_subslice > 2);
94b4f3ba
CW
328}
329
330static void gen9_sseu_info_init(struct drm_i915_private *dev_priv)
331{
332 struct intel_device_info *info = mkwrite_device_info(dev_priv);
43b67998 333 struct sseu_dev_info *sseu = &info->sseu;
94b4f3ba 334 int s, ss;
8cc76693
LL
335 u32 fuse2, eu_disable, subslice_mask;
336 const u8 eu_mask = 0xff;
94b4f3ba
CW
337
338 fuse2 = I915_READ(GEN8_FUSE2);
f08a0c92 339 sseu->slice_mask = (fuse2 & GEN8_F2_S_ENA_MASK) >> GEN8_F2_S_ENA_SHIFT;
94b4f3ba 340
8cc76693
LL
341 /* BXT has a single slice and at most 3 subslices. */
342 sseu->max_slices = IS_GEN9_LP(dev_priv) ? 1 : 3;
343 sseu->max_subslices = IS_GEN9_LP(dev_priv) ? 3 : 4;
344 sseu->max_eus_per_subslice = 8;
345
94b4f3ba
CW
346 /*
347 * The subslice disable field is global, i.e. it applies
348 * to each of the enabled slices.
349 */
8cc76693
LL
350 subslice_mask = (1 << sseu->max_subslices) - 1;
351 subslice_mask &= ~((fuse2 & GEN9_F2_SS_DIS_MASK) >>
352 GEN9_F2_SS_DIS_SHIFT);
94b4f3ba
CW
353
354 /*
355 * Iterate through enabled slices and subslices to
356 * count the total enabled EU.
357 */
8cc76693 358 for (s = 0; s < sseu->max_slices; s++) {
f08a0c92 359 if (!(sseu->slice_mask & BIT(s)))
94b4f3ba
CW
360 /* skip disabled slice */
361 continue;
362
8cc76693
LL
363 sseu->subslice_mask[s] = subslice_mask;
364
94b4f3ba 365 eu_disable = I915_READ(GEN9_EU_DISABLE(s));
8cc76693 366 for (ss = 0; ss < sseu->max_subslices; ss++) {
94b4f3ba 367 int eu_per_ss;
8cc76693 368 u8 eu_disabled_mask;
94b4f3ba 369
8cc76693 370 if (!(sseu->subslice_mask[s] & BIT(ss)))
94b4f3ba
CW
371 /* skip disabled subslice */
372 continue;
373
b3e7f866 374 eu_disabled_mask = (eu_disable >> (ss * 8)) & eu_mask;
8cc76693
LL
375
376 sseu_set_eus(sseu, s, ss, ~eu_disabled_mask);
377
378 eu_per_ss = sseu->max_eus_per_subslice -
379 hweight8(eu_disabled_mask);
94b4f3ba
CW
380
381 /*
382 * Record which subslice(s) has(have) 7 EUs. we
383 * can tune the hash used to spread work among
384 * subslices if they are unbalanced.
385 */
386 if (eu_per_ss == 7)
43b67998 387 sseu->subslice_7eu[s] |= BIT(ss);
94b4f3ba
CW
388 }
389 }
390
8cc76693
LL
391 sseu->eu_total = compute_eu_total(sseu);
392
94b4f3ba
CW
393 /*
394 * SKL is expected to always have a uniform distribution
395 * of EU across subslices with the exception that any one
396 * EU in any one subslice may be fused off for die
397 * recovery. BXT is expected to be perfectly uniform in EU
398 * distribution.
399 */
57ec171e 400 sseu->eu_per_subslice = sseu_subslice_total(sseu) ?
43b67998 401 DIV_ROUND_UP(sseu->eu_total,
57ec171e 402 sseu_subslice_total(sseu)) : 0;
94b4f3ba 403 /*
c7ae7e9a 404 * SKL+ supports slice power gating on devices with more than
94b4f3ba 405 * one slice, and supports EU power gating on devices with
c7ae7e9a 406 * more than one EU pair per subslice. BXT+ supports subslice
94b4f3ba
CW
407 * power gating on devices with more than one subslice, and
408 * supports EU power gating on devices with more than one EU
409 * pair per subslice.
410 */
43b67998 411 sseu->has_slice_pg =
c7ae7e9a 412 !IS_GEN9_LP(dev_priv) && hweight8(sseu->slice_mask) > 1;
43b67998 413 sseu->has_subslice_pg =
254e0931 414 IS_GEN9_LP(dev_priv) && sseu_subslice_total(sseu) > 1;
43b67998 415 sseu->has_eu_pg = sseu->eu_per_subslice > 2;
94b4f3ba 416
234516af 417 if (IS_GEN9_LP(dev_priv)) {
8cc76693
LL
418#define IS_SS_DISABLED(ss) (!(sseu->subslice_mask[0] & BIT(ss)))
419 info->has_pooled_eu = hweight8(sseu->subslice_mask[0]) == 3;
234516af 420
43b67998 421 sseu->min_eu_in_pool = 0;
94b4f3ba 422 if (info->has_pooled_eu) {
57ec171e 423 if (IS_SS_DISABLED(2) || IS_SS_DISABLED(0))
43b67998 424 sseu->min_eu_in_pool = 3;
57ec171e 425 else if (IS_SS_DISABLED(1))
43b67998 426 sseu->min_eu_in_pool = 6;
94b4f3ba 427 else
43b67998 428 sseu->min_eu_in_pool = 9;
94b4f3ba
CW
429 }
430#undef IS_SS_DISABLED
431 }
432}
433
434static void broadwell_sseu_info_init(struct drm_i915_private *dev_priv)
435{
43b67998 436 struct sseu_dev_info *sseu = &mkwrite_device_info(dev_priv)->sseu;
94b4f3ba 437 int s, ss;
8cc76693 438 u32 fuse2, subslice_mask, eu_disable[3]; /* s_max */
94b4f3ba
CW
439
440 fuse2 = I915_READ(GEN8_FUSE2);
f08a0c92 441 sseu->slice_mask = (fuse2 & GEN8_F2_S_ENA_MASK) >> GEN8_F2_S_ENA_SHIFT;
8cc76693
LL
442 sseu->max_slices = 3;
443 sseu->max_subslices = 3;
444 sseu->max_eus_per_subslice = 8;
445
57ec171e
ID
446 /*
447 * The subslice disable field is global, i.e. it applies
448 * to each of the enabled slices.
449 */
8cc76693
LL
450 subslice_mask = GENMASK(sseu->max_subslices - 1, 0);
451 subslice_mask &= ~((fuse2 & GEN8_F2_SS_DIS_MASK) >>
452 GEN8_F2_SS_DIS_SHIFT);
94b4f3ba
CW
453
454 eu_disable[0] = I915_READ(GEN8_EU_DISABLE0) & GEN8_EU_DIS0_S0_MASK;
455 eu_disable[1] = (I915_READ(GEN8_EU_DISABLE0) >> GEN8_EU_DIS0_S1_SHIFT) |
456 ((I915_READ(GEN8_EU_DISABLE1) & GEN8_EU_DIS1_S1_MASK) <<
457 (32 - GEN8_EU_DIS0_S1_SHIFT));
458 eu_disable[2] = (I915_READ(GEN8_EU_DISABLE1) >> GEN8_EU_DIS1_S2_SHIFT) |
459 ((I915_READ(GEN8_EU_DISABLE2) & GEN8_EU_DIS2_S2_MASK) <<
460 (32 - GEN8_EU_DIS1_S2_SHIFT));
461
94b4f3ba
CW
462 /*
463 * Iterate through enabled slices and subslices to
464 * count the total enabled EU.
465 */
8cc76693 466 for (s = 0; s < sseu->max_slices; s++) {
f08a0c92 467 if (!(sseu->slice_mask & BIT(s)))
94b4f3ba
CW
468 /* skip disabled slice */
469 continue;
470
8cc76693
LL
471 sseu->subslice_mask[s] = subslice_mask;
472
473 for (ss = 0; ss < sseu->max_subslices; ss++) {
474 u8 eu_disabled_mask;
94b4f3ba
CW
475 u32 n_disabled;
476
8cc76693 477 if (!(sseu->subslice_mask[ss] & BIT(ss)))
94b4f3ba
CW
478 /* skip disabled subslice */
479 continue;
480
8cc76693
LL
481 eu_disabled_mask =
482 eu_disable[s] >> (ss * sseu->max_eus_per_subslice);
483
484 sseu_set_eus(sseu, s, ss, ~eu_disabled_mask);
485
486 n_disabled = hweight8(eu_disabled_mask);
94b4f3ba
CW
487
488 /*
489 * Record which subslices have 7 EUs.
490 */
8cc76693 491 if (sseu->max_eus_per_subslice - n_disabled == 7)
43b67998 492 sseu->subslice_7eu[s] |= 1 << ss;
94b4f3ba
CW
493 }
494 }
495
8cc76693
LL
496 sseu->eu_total = compute_eu_total(sseu);
497
94b4f3ba
CW
498 /*
499 * BDW is expected to always have a uniform distribution of EU across
500 * subslices with the exception that any one EU in any one subslice may
501 * be fused off for die recovery.
502 */
57ec171e
ID
503 sseu->eu_per_subslice = sseu_subslice_total(sseu) ?
504 DIV_ROUND_UP(sseu->eu_total,
505 sseu_subslice_total(sseu)) : 0;
94b4f3ba
CW
506
507 /*
508 * BDW supports slice power gating on devices with more than
509 * one slice.
510 */
f08a0c92 511 sseu->has_slice_pg = hweight8(sseu->slice_mask) > 1;
43b67998
ID
512 sseu->has_subslice_pg = 0;
513 sseu->has_eu_pg = 0;
94b4f3ba
CW
514}
515
b8ec759e
LL
516static void haswell_sseu_info_init(struct drm_i915_private *dev_priv)
517{
518 struct intel_device_info *info = mkwrite_device_info(dev_priv);
519 struct sseu_dev_info *sseu = &info->sseu;
520 u32 fuse1;
8cc76693 521 int s, ss;
b8ec759e
LL
522
523 /*
524 * There isn't a register to tell us how many slices/subslices. We
525 * work off the PCI-ids here.
526 */
527 switch (info->gt) {
528 default:
529 MISSING_CASE(info->gt);
530 /* fall through */
531 case 1:
532 sseu->slice_mask = BIT(0);
8cc76693 533 sseu->subslice_mask[0] = BIT(0);
b8ec759e
LL
534 break;
535 case 2:
536 sseu->slice_mask = BIT(0);
8cc76693 537 sseu->subslice_mask[0] = BIT(0) | BIT(1);
b8ec759e
LL
538 break;
539 case 3:
540 sseu->slice_mask = BIT(0) | BIT(1);
8cc76693
LL
541 sseu->subslice_mask[0] = BIT(0) | BIT(1);
542 sseu->subslice_mask[1] = BIT(0) | BIT(1);
b8ec759e
LL
543 break;
544 }
545
8cc76693
LL
546 sseu->max_slices = hweight8(sseu->slice_mask);
547 sseu->max_subslices = hweight8(sseu->subslice_mask[0]);
548
b8ec759e
LL
549 fuse1 = I915_READ(HSW_PAVP_FUSE1);
550 switch ((fuse1 & HSW_F1_EU_DIS_MASK) >> HSW_F1_EU_DIS_SHIFT) {
551 default:
552 MISSING_CASE((fuse1 & HSW_F1_EU_DIS_MASK) >>
553 HSW_F1_EU_DIS_SHIFT);
554 /* fall through */
555 case HSW_F1_EU_DIS_10EUS:
556 sseu->eu_per_subslice = 10;
557 break;
558 case HSW_F1_EU_DIS_8EUS:
559 sseu->eu_per_subslice = 8;
560 break;
561 case HSW_F1_EU_DIS_6EUS:
562 sseu->eu_per_subslice = 6;
563 break;
564 }
8cc76693
LL
565 sseu->max_eus_per_subslice = sseu->eu_per_subslice;
566
567 for (s = 0; s < sseu->max_slices; s++) {
568 for (ss = 0; ss < sseu->max_subslices; ss++) {
569 sseu_set_eus(sseu, s, ss,
570 (1UL << sseu->eu_per_subslice) - 1);
571 }
572 }
b8ec759e 573
8cc76693 574 sseu->eu_total = compute_eu_total(sseu);
b8ec759e
LL
575
576 /* No powergating for you. */
577 sseu->has_slice_pg = 0;
578 sseu->has_subslice_pg = 0;
579 sseu->has_eu_pg = 0;
580}
581
f577a03b 582static u32 read_reference_ts_freq(struct drm_i915_private *dev_priv)
dab91783
LL
583{
584 u32 ts_override = I915_READ(GEN9_TIMESTAMP_OVERRIDE);
f577a03b 585 u32 base_freq, frac_freq;
dab91783
LL
586
587 base_freq = ((ts_override & GEN9_TIMESTAMP_OVERRIDE_US_COUNTER_DIVIDER_MASK) >>
588 GEN9_TIMESTAMP_OVERRIDE_US_COUNTER_DIVIDER_SHIFT) + 1;
f577a03b 589 base_freq *= 1000;
dab91783
LL
590
591 frac_freq = ((ts_override &
592 GEN9_TIMESTAMP_OVERRIDE_US_COUNTER_DENOMINATOR_MASK) >>
593 GEN9_TIMESTAMP_OVERRIDE_US_COUNTER_DENOMINATOR_SHIFT);
f577a03b 594 frac_freq = 1000 / (frac_freq + 1);
dab91783
LL
595
596 return base_freq + frac_freq;
597}
598
d775a7b1
PZ
599static u32 gen10_get_crystal_clock_freq(struct drm_i915_private *dev_priv,
600 u32 rpm_config_reg)
601{
602 u32 f19_2_mhz = 19200;
603 u32 f24_mhz = 24000;
604 u32 crystal_clock = (rpm_config_reg &
605 GEN9_RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_MASK) >>
606 GEN9_RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_SHIFT;
607
608 switch (crystal_clock) {
609 case GEN9_RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_19_2_MHZ:
610 return f19_2_mhz;
611 case GEN9_RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_24_MHZ:
612 return f24_mhz;
613 default:
614 MISSING_CASE(crystal_clock);
615 return 0;
616 }
617}
618
619static u32 gen11_get_crystal_clock_freq(struct drm_i915_private *dev_priv,
620 u32 rpm_config_reg)
621{
622 u32 f19_2_mhz = 19200;
623 u32 f24_mhz = 24000;
624 u32 f25_mhz = 25000;
625 u32 f38_4_mhz = 38400;
626 u32 crystal_clock = (rpm_config_reg &
627 GEN11_RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_MASK) >>
628 GEN11_RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_SHIFT;
629
630 switch (crystal_clock) {
631 case GEN11_RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_24_MHZ:
632 return f24_mhz;
633 case GEN11_RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_19_2_MHZ:
634 return f19_2_mhz;
635 case GEN11_RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_38_4_MHZ:
636 return f38_4_mhz;
637 case GEN11_RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_25_MHZ:
638 return f25_mhz;
639 default:
640 MISSING_CASE(crystal_clock);
641 return 0;
642 }
643}
644
f577a03b 645static u32 read_timestamp_frequency(struct drm_i915_private *dev_priv)
dab91783 646{
f577a03b
LL
647 u32 f12_5_mhz = 12500;
648 u32 f19_2_mhz = 19200;
649 u32 f24_mhz = 24000;
dab91783
LL
650
651 if (INTEL_GEN(dev_priv) <= 4) {
652 /* PRMs say:
653 *
654 * "The value in this register increments once every 16
655 * hclks." (through the “Clocking Configuration”
656 * (“CLKCFG”) MCHBAR register)
657 */
f577a03b 658 return dev_priv->rawclk_freq / 16;
dab91783
LL
659 } else if (INTEL_GEN(dev_priv) <= 8) {
660 /* PRMs say:
661 *
662 * "The PCU TSC counts 10ns increments; this timestamp
663 * reflects bits 38:3 of the TSC (i.e. 80ns granularity,
664 * rolling over every 1.5 hours).
665 */
666 return f12_5_mhz;
667 } else if (INTEL_GEN(dev_priv) <= 9) {
668 u32 ctc_reg = I915_READ(CTC_MODE);
f577a03b 669 u32 freq = 0;
dab91783
LL
670
671 if ((ctc_reg & CTC_SOURCE_PARAMETER_MASK) == CTC_SOURCE_DIVIDE_LOGIC) {
672 freq = read_reference_ts_freq(dev_priv);
673 } else {
674 freq = IS_GEN9_LP(dev_priv) ? f19_2_mhz : f24_mhz;
675
676 /* Now figure out how the command stream's timestamp
677 * register increments from this frequency (it might
678 * increment only every few clock cycle).
679 */
680 freq >>= 3 - ((ctc_reg & CTC_SHIFT_PARAMETER_MASK) >>
681 CTC_SHIFT_PARAMETER_SHIFT);
682 }
683
684 return freq;
d775a7b1 685 } else if (INTEL_GEN(dev_priv) <= 11) {
dab91783 686 u32 ctc_reg = I915_READ(CTC_MODE);
f577a03b 687 u32 freq = 0;
dab91783
LL
688
689 /* First figure out the reference frequency. There are 2 ways
690 * we can compute the frequency, either through the
691 * TIMESTAMP_OVERRIDE register or through RPM_CONFIG. CTC_MODE
692 * tells us which one we should use.
693 */
694 if ((ctc_reg & CTC_SOURCE_PARAMETER_MASK) == CTC_SOURCE_DIVIDE_LOGIC) {
695 freq = read_reference_ts_freq(dev_priv);
696 } else {
d775a7b1
PZ
697 u32 rpm_config_reg = I915_READ(RPM_CONFIG0);
698
699 if (INTEL_GEN(dev_priv) <= 10)
700 freq = gen10_get_crystal_clock_freq(dev_priv,
701 rpm_config_reg);
702 else
703 freq = gen11_get_crystal_clock_freq(dev_priv,
704 rpm_config_reg);
dab91783 705
53ff2641
LL
706 /* Now figure out how the command stream's timestamp
707 * register increments from this frequency (it might
708 * increment only every few clock cycle).
709 */
710 freq >>= 3 - ((rpm_config_reg &
711 GEN10_RPM_CONFIG0_CTC_SHIFT_PARAMETER_MASK) >>
712 GEN10_RPM_CONFIG0_CTC_SHIFT_PARAMETER_SHIFT);
713 }
dab91783
LL
714
715 return freq;
716 }
717
fe66e928 718 MISSING_CASE("Unknown gen, unable to read command streamer timestamp frequency\n");
dab91783
LL
719 return 0;
720}
721
6a7e51f3
MW
722/**
723 * intel_device_info_runtime_init - initialize runtime info
724 * @info: intel device info struct
725 *
94b4f3ba
CW
726 * Determine various intel_device_info fields at runtime.
727 *
728 * Use it when either:
729 * - it's judged too laborious to fill n static structures with the limit
730 * when a simple if statement does the job,
731 * - run-time checks (eg read fuse/strap registers) are needed.
732 *
733 * This function needs to be called:
734 * - after the MMIO has been setup as we are reading registers,
735 * - after the PCH has been detected,
736 * - before the first usage of the fields it can tweak.
737 */
6a7e51f3 738void intel_device_info_runtime_init(struct intel_device_info *info)
94b4f3ba 739{
6a7e51f3
MW
740 struct drm_i915_private *dev_priv =
741 container_of(info, struct drm_i915_private, info);
94b4f3ba
CW
742 enum pipe pipe;
743
6e7406db
MK
744 if (INTEL_GEN(dev_priv) >= 10) {
745 for_each_pipe(dev_priv, pipe)
746 info->num_scalers[pipe] = 2;
747 } else if (INTEL_GEN(dev_priv) == 9) {
0bf0230e
ACO
748 info->num_scalers[PIPE_A] = 2;
749 info->num_scalers[PIPE_B] = 2;
750 info->num_scalers[PIPE_C] = 1;
751 }
752
74f6e183 753 BUILD_BUG_ON(I915_NUM_ENGINES > BITS_PER_TYPE(intel_ring_mask_t));
022d3093 754
94b4f3ba
CW
755 /*
756 * Skylake and Broxton currently don't expose the topmost plane as its
757 * use is exclusive with the legacy cursor and we only want to expose
758 * one of those, not both. Until we can safely expose the topmost plane
759 * as a DRM_PLANE_TYPE_CURSOR with all the features exposed/supported,
760 * we don't expose the topmost plane at all to prevent ABI breakage
761 * down the line.
762 */
8366be98 763 if (IS_GEN10(dev_priv) || IS_GEMINILAKE(dev_priv))
e9c98825
ACO
764 for_each_pipe(dev_priv, pipe)
765 info->num_sprites[pipe] = 3;
766 else if (IS_BROXTON(dev_priv)) {
94b4f3ba
CW
767 info->num_sprites[PIPE_A] = 2;
768 info->num_sprites[PIPE_B] = 2;
769 info->num_sprites[PIPE_C] = 1;
33edc24d 770 } else if (IS_VALLEYVIEW(dev_priv) || IS_CHERRYVIEW(dev_priv)) {
94b4f3ba
CW
771 for_each_pipe(dev_priv, pipe)
772 info->num_sprites[pipe] = 2;
ab33081a 773 } else if (INTEL_GEN(dev_priv) >= 5 || IS_G4X(dev_priv)) {
94b4f3ba
CW
774 for_each_pipe(dev_priv, pipe)
775 info->num_sprites[pipe] = 1;
33edc24d 776 }
94b4f3ba 777
4f044a88 778 if (i915_modparams.disable_display) {
94b4f3ba
CW
779 DRM_INFO("Display disabled (module parameter)\n");
780 info->num_pipes = 0;
781 } else if (info->num_pipes > 0 &&
782 (IS_GEN7(dev_priv) || IS_GEN8(dev_priv)) &&
783 HAS_PCH_SPLIT(dev_priv)) {
784 u32 fuse_strap = I915_READ(FUSE_STRAP);
785 u32 sfuse_strap = I915_READ(SFUSE_STRAP);
786
787 /*
788 * SFUSE_STRAP is supposed to have a bit signalling the display
789 * is fused off. Unfortunately it seems that, at least in
790 * certain cases, fused off display means that PCH display
791 * reads don't land anywhere. In that case, we read 0s.
792 *
793 * On CPT/PPT, we can detect this case as SFUSE_STRAP_FUSE_LOCK
794 * should be set when taking over after the firmware.
795 */
796 if (fuse_strap & ILK_INTERNAL_DISPLAY_DISABLE ||
797 sfuse_strap & SFUSE_STRAP_DISPLAY_DISABLED ||
b9eb89b2 798 (HAS_PCH_CPT(dev_priv) &&
94b4f3ba
CW
799 !(sfuse_strap & SFUSE_STRAP_FUSE_LOCK))) {
800 DRM_INFO("Display fused off, disabling\n");
801 info->num_pipes = 0;
802 } else if (fuse_strap & IVB_PIPE_C_DISABLE) {
803 DRM_INFO("PipeC fused off\n");
804 info->num_pipes -= 1;
805 }
806 } else if (info->num_pipes > 0 && IS_GEN9(dev_priv)) {
807 u32 dfsm = I915_READ(SKL_DFSM);
808 u8 disabled_mask = 0;
809 bool invalid;
810 int num_bits;
811
812 if (dfsm & SKL_DFSM_PIPE_A_DISABLE)
813 disabled_mask |= BIT(PIPE_A);
814 if (dfsm & SKL_DFSM_PIPE_B_DISABLE)
815 disabled_mask |= BIT(PIPE_B);
816 if (dfsm & SKL_DFSM_PIPE_C_DISABLE)
817 disabled_mask |= BIT(PIPE_C);
818
819 num_bits = hweight8(disabled_mask);
820
821 switch (disabled_mask) {
822 case BIT(PIPE_A):
823 case BIT(PIPE_B):
824 case BIT(PIPE_A) | BIT(PIPE_B):
825 case BIT(PIPE_A) | BIT(PIPE_C):
826 invalid = true;
827 break;
828 default:
829 invalid = false;
830 }
831
832 if (num_bits > info->num_pipes || invalid)
833 DRM_ERROR("invalid pipe fuse configuration: 0x%x\n",
834 disabled_mask);
835 else
836 info->num_pipes -= num_bits;
837 }
838
839 /* Initialize slice/subslice/EU info */
b8ec759e
LL
840 if (IS_HASWELL(dev_priv))
841 haswell_sseu_info_init(dev_priv);
842 else if (IS_CHERRYVIEW(dev_priv))
94b4f3ba
CW
843 cherryview_sseu_info_init(dev_priv);
844 else if (IS_BROADWELL(dev_priv))
845 broadwell_sseu_info_init(dev_priv);
4e9767bc 846 else if (INTEL_GEN(dev_priv) == 9)
94b4f3ba 847 gen9_sseu_info_init(dev_priv);
8b5eb5e2 848 else if (INTEL_GEN(dev_priv) == 10)
4e9767bc 849 gen10_sseu_info_init(dev_priv);
f60fa408 850 else if (INTEL_GEN(dev_priv) >= 11)
8b5eb5e2 851 gen11_sseu_info_init(dev_priv);
94b4f3ba 852
4bdafb9d
CW
853 if (IS_GEN6(dev_priv) && intel_vtd_active()) {
854 DRM_INFO("Disabling ppGTT for VT-d support\n");
855 info->ppgtt = INTEL_PPGTT_NONE;
856 }
857
dab91783 858 /* Initialize command stream timestamp frequency */
f577a03b 859 info->cs_timestamp_frequency_khz = read_timestamp_frequency(dev_priv);
94b4f3ba 860}
3fed1808
CW
861
862void intel_driver_caps_print(const struct intel_driver_caps *caps,
863 struct drm_printer *p)
864{
481827b4
CW
865 drm_printf(p, "Has logical contexts? %s\n",
866 yesno(caps->has_logical_contexts));
3fed1808
CW
867 drm_printf(p, "scheduler: %x\n", caps->scheduler);
868}
26376a7e
OM
869
870/*
871 * Determine which engines are fused off in our particular hardware. Since the
872 * fuse register is in the blitter powerwell, we need forcewake to be ready at
873 * this point (but later we need to prune the forcewake domains for engines that
874 * are indeed fused off).
875 */
876void intel_device_info_init_mmio(struct drm_i915_private *dev_priv)
877{
878 struct intel_device_info *info = mkwrite_device_info(dev_priv);
879 u8 vdbox_disable, vebox_disable;
880 u32 media_fuse;
881 int i;
882
883 if (INTEL_GEN(dev_priv) < 11)
884 return;
885
886 media_fuse = I915_READ(GEN11_GT_VEBOX_VDBOX_DISABLE);
887
888 vdbox_disable = media_fuse & GEN11_GT_VDBOX_DISABLE_MASK;
889 vebox_disable = (media_fuse & GEN11_GT_VEBOX_DISABLE_MASK) >>
890 GEN11_GT_VEBOX_DISABLE_SHIFT;
891
892 DRM_DEBUG_DRIVER("vdbox disable: %04x\n", vdbox_disable);
893 for (i = 0; i < I915_MAX_VCS; i++) {
894 if (!HAS_ENGINE(dev_priv, _VCS(i)))
895 continue;
896
897 if (!(BIT(i) & vdbox_disable))
898 continue;
899
900 info->ring_mask &= ~ENGINE_MASK(_VCS(i));
901 DRM_DEBUG_DRIVER("vcs%u fused off\n", i);
902 }
903
904 DRM_DEBUG_DRIVER("vebox disable: %04x\n", vebox_disable);
905 for (i = 0; i < I915_MAX_VECS; i++) {
906 if (!HAS_ENGINE(dev_priv, _VECS(i)))
907 continue;
908
909 if (!(BIT(i) & vebox_disable))
910 continue;
911
912 info->ring_mask &= ~ENGINE_MASK(_VECS(i));
913 DRM_DEBUG_DRIVER("vecs%u fused off\n", i);
914 }
915}