2 * Copyright © 2016 Intel Corporation
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:
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
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
25 #include <drm/drm_print.h>
27 #include "intel_device_info.h"
30 #define PLATFORM_NAME(x) [INTEL_##x] = #x
31 static const char * const platform_names
[] = {
37 PLATFORM_NAME(I915GM
),
39 PLATFORM_NAME(I945GM
),
41 PLATFORM_NAME(PINEVIEW
),
43 PLATFORM_NAME(I965GM
),
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 PLATFORM_NAME(ELKHARTLAKE
),
61 PLATFORM_NAME(TIGERLAKE
),
65 const char *intel_platform_name(enum intel_platform platform
)
67 BUILD_BUG_ON(ARRAY_SIZE(platform_names
) != INTEL_MAX_PLATFORMS
);
69 if (WARN_ON_ONCE(platform
>= ARRAY_SIZE(platform_names
) ||
70 platform_names
[platform
] == NULL
))
73 return platform_names
[platform
];
76 static const char *iommu_name(void)
78 const char *msg
= "n/a";
80 #ifdef CONFIG_INTEL_IOMMU
81 msg
= enableddisabled(intel_iommu_gfx_mapped
);
87 void intel_device_info_print_static(const struct intel_device_info
*info
,
88 struct drm_printer
*p
)
90 drm_printf(p
, "engines: %x\n", info
->engine_mask
);
91 drm_printf(p
, "gen: %d\n", info
->gen
);
92 drm_printf(p
, "gt: %d\n", info
->gt
);
93 drm_printf(p
, "iommu: %s\n", iommu_name());
94 drm_printf(p
, "memory-regions: %x\n", info
->memory_regions
);
95 drm_printf(p
, "page-sizes: %x\n", info
->page_sizes
);
96 drm_printf(p
, "platform: %s\n", intel_platform_name(info
->platform
));
97 drm_printf(p
, "ppgtt-size: %d\n", info
->ppgtt_size
);
98 drm_printf(p
, "ppgtt-type: %d\n", info
->ppgtt_type
);
100 #define PRINT_FLAG(name) drm_printf(p, "%s: %s\n", #name, yesno(info->name));
101 DEV_INFO_FOR_EACH_FLAG(PRINT_FLAG
);
104 #define PRINT_FLAG(name) drm_printf(p, "%s: %s\n", #name, yesno(info->display.name));
105 DEV_INFO_DISPLAY_FOR_EACH_FLAG(PRINT_FLAG
);
109 static void sseu_dump(const struct sseu_dev_info
*sseu
, struct drm_printer
*p
)
113 drm_printf(p
, "slice total: %u, mask=%04x\n",
114 hweight8(sseu
->slice_mask
), sseu
->slice_mask
);
115 drm_printf(p
, "subslice total: %u\n", intel_sseu_subslice_total(sseu
));
116 for (s
= 0; s
< sseu
->max_slices
; s
++) {
117 drm_printf(p
, "slice%d: %u subslices, mask=%08x\n",
118 s
, intel_sseu_subslices_per_slice(sseu
, s
),
119 intel_sseu_get_subslices(sseu
, s
));
121 drm_printf(p
, "EU total: %u\n", sseu
->eu_total
);
122 drm_printf(p
, "EU per subslice: %u\n", sseu
->eu_per_subslice
);
123 drm_printf(p
, "has slice power gating: %s\n",
124 yesno(sseu
->has_slice_pg
));
125 drm_printf(p
, "has subslice power gating: %s\n",
126 yesno(sseu
->has_subslice_pg
));
127 drm_printf(p
, "has EU power gating: %s\n", yesno(sseu
->has_eu_pg
));
130 void intel_device_info_print_runtime(const struct intel_runtime_info
*info
,
131 struct drm_printer
*p
)
133 sseu_dump(&info
->sseu
, p
);
135 drm_printf(p
, "CS timestamp frequency: %u kHz\n",
136 info
->cs_timestamp_frequency_khz
);
139 static int sseu_eu_idx(const struct sseu_dev_info
*sseu
, int slice
,
142 int slice_stride
= sseu
->max_subslices
* sseu
->eu_stride
;
144 return slice
* slice_stride
+ subslice
* sseu
->eu_stride
;
147 static u16
sseu_get_eus(const struct sseu_dev_info
*sseu
, int slice
,
150 int i
, offset
= sseu_eu_idx(sseu
, slice
, subslice
);
153 for (i
= 0; i
< sseu
->eu_stride
; i
++) {
154 eu_mask
|= ((u16
)sseu
->eu_mask
[offset
+ i
]) <<
161 static void sseu_set_eus(struct sseu_dev_info
*sseu
, int slice
, int subslice
,
164 int i
, offset
= sseu_eu_idx(sseu
, slice
, subslice
);
166 for (i
= 0; i
< sseu
->eu_stride
; i
++) {
167 sseu
->eu_mask
[offset
+ i
] =
168 (eu_mask
>> (BITS_PER_BYTE
* i
)) & 0xff;
172 void intel_device_info_print_topology(const struct sseu_dev_info
*sseu
,
173 struct drm_printer
*p
)
177 if (sseu
->max_slices
== 0) {
178 drm_printf(p
, "Unavailable\n");
182 for (s
= 0; s
< sseu
->max_slices
; s
++) {
183 drm_printf(p
, "slice%d: %u subslice(s) (0x%08x):\n",
184 s
, intel_sseu_subslices_per_slice(sseu
, s
),
185 intel_sseu_get_subslices(sseu
, s
));
187 for (ss
= 0; ss
< sseu
->max_subslices
; ss
++) {
188 u16 enabled_eus
= sseu_get_eus(sseu
, s
, ss
);
190 drm_printf(p
, "\tsubslice%d: %u EUs (0x%hx)\n",
191 ss
, hweight16(enabled_eus
), enabled_eus
);
196 static u16
compute_eu_total(const struct sseu_dev_info
*sseu
)
200 for (i
= 0; i
< ARRAY_SIZE(sseu
->eu_mask
); i
++)
201 total
+= hweight8(sseu
->eu_mask
[i
]);
206 static void gen11_compute_sseu_info(struct sseu_dev_info
*sseu
,
207 u8 s_en
, u32 ss_en
, u16 eu_en
)
211 /* ss_en represents entire subslice mask across all slices */
212 GEM_BUG_ON(sseu
->max_slices
* sseu
->max_subslices
>
213 sizeof(ss_en
) * BITS_PER_BYTE
);
215 for (s
= 0; s
< sseu
->max_slices
; s
++) {
216 if ((s_en
& BIT(s
)) == 0)
219 sseu
->slice_mask
|= BIT(s
);
221 intel_sseu_set_subslices(sseu
, s
, ss_en
);
223 for (ss
= 0; ss
< sseu
->max_subslices
; ss
++)
224 if (intel_sseu_has_subslice(sseu
, s
, ss
))
225 sseu_set_eus(sseu
, s
, ss
, eu_en
);
227 sseu
->eu_per_subslice
= hweight16(eu_en
);
228 sseu
->eu_total
= compute_eu_total(sseu
);
231 static void gen12_sseu_info_init(struct drm_i915_private
*dev_priv
)
233 struct sseu_dev_info
*sseu
= &RUNTIME_INFO(dev_priv
)->sseu
;
241 * Gen12 has Dual-Subslices, which behave similarly to 2 gen11 SS.
242 * Instead of splitting these, provide userspace with an array
243 * of DSS to more closely represent the hardware resource.
245 intel_sseu_set_info(sseu
, 1, 6, 16);
247 s_en
= I915_READ(GEN11_GT_SLICE_ENABLE
) & GEN11_GT_S_ENA_MASK
;
249 dss_en
= I915_READ(GEN12_GT_DSS_ENABLE
);
251 /* one bit per pair of EUs */
252 eu_en_fuse
= ~(I915_READ(GEN11_EU_DISABLE
) & GEN11_EU_DIS_MASK
);
253 for (eu
= 0; eu
< sseu
->max_eus_per_subslice
/ 2; eu
++)
254 if (eu_en_fuse
& BIT(eu
))
255 eu_en
|= BIT(eu
* 2) | BIT(eu
* 2 + 1);
257 gen11_compute_sseu_info(sseu
, s_en
, dss_en
, eu_en
);
259 /* TGL only supports slice-level power gating */
260 sseu
->has_slice_pg
= 1;
263 static void gen11_sseu_info_init(struct drm_i915_private
*dev_priv
)
265 struct sseu_dev_info
*sseu
= &RUNTIME_INFO(dev_priv
)->sseu
;
270 if (IS_ELKHARTLAKE(dev_priv
))
271 intel_sseu_set_info(sseu
, 1, 4, 8);
273 intel_sseu_set_info(sseu
, 1, 8, 8);
275 s_en
= I915_READ(GEN11_GT_SLICE_ENABLE
) & GEN11_GT_S_ENA_MASK
;
276 ss_en
= ~I915_READ(GEN11_GT_SUBSLICE_DISABLE
);
277 eu_en
= ~(I915_READ(GEN11_EU_DISABLE
) & GEN11_EU_DIS_MASK
);
279 gen11_compute_sseu_info(sseu
, s_en
, ss_en
, eu_en
);
281 /* ICL has no power gating restrictions. */
282 sseu
->has_slice_pg
= 1;
283 sseu
->has_subslice_pg
= 1;
287 static void gen10_sseu_info_init(struct drm_i915_private
*dev_priv
)
289 struct sseu_dev_info
*sseu
= &RUNTIME_INFO(dev_priv
)->sseu
;
290 const u32 fuse2
= I915_READ(GEN8_FUSE2
);
292 const int eu_mask
= 0xff;
293 u32 subslice_mask
, eu_en
;
295 intel_sseu_set_info(sseu
, 6, 4, 8);
297 sseu
->slice_mask
= (fuse2
& GEN10_F2_S_ENA_MASK
) >>
298 GEN10_F2_S_ENA_SHIFT
;
301 eu_en
= ~I915_READ(GEN8_EU_DISABLE0
);
302 for (ss
= 0; ss
< sseu
->max_subslices
; ss
++)
303 sseu_set_eus(sseu
, 0, ss
, (eu_en
>> (8 * ss
)) & eu_mask
);
305 sseu_set_eus(sseu
, 1, 0, (eu_en
>> 24) & eu_mask
);
306 eu_en
= ~I915_READ(GEN8_EU_DISABLE1
);
307 sseu_set_eus(sseu
, 1, 1, eu_en
& eu_mask
);
309 sseu_set_eus(sseu
, 2, 0, (eu_en
>> 8) & eu_mask
);
310 sseu_set_eus(sseu
, 2, 1, (eu_en
>> 16) & eu_mask
);
312 sseu_set_eus(sseu
, 3, 0, (eu_en
>> 24) & eu_mask
);
313 eu_en
= ~I915_READ(GEN8_EU_DISABLE2
);
314 sseu_set_eus(sseu
, 3, 1, eu_en
& eu_mask
);
316 sseu_set_eus(sseu
, 4, 0, (eu_en
>> 8) & eu_mask
);
317 sseu_set_eus(sseu
, 4, 1, (eu_en
>> 16) & eu_mask
);
319 sseu_set_eus(sseu
, 5, 0, (eu_en
>> 24) & eu_mask
);
320 eu_en
= ~I915_READ(GEN10_EU_DISABLE3
);
321 sseu_set_eus(sseu
, 5, 1, eu_en
& eu_mask
);
323 subslice_mask
= (1 << 4) - 1;
324 subslice_mask
&= ~((fuse2
& GEN10_F2_SS_DIS_MASK
) >>
325 GEN10_F2_SS_DIS_SHIFT
);
327 for (s
= 0; s
< sseu
->max_slices
; s
++) {
328 u32 subslice_mask_with_eus
= subslice_mask
;
330 for (ss
= 0; ss
< sseu
->max_subslices
; ss
++) {
331 if (sseu_get_eus(sseu
, s
, ss
) == 0)
332 subslice_mask_with_eus
&= ~BIT(ss
);
336 * Slice0 can have up to 3 subslices, but there are only 2 in
339 intel_sseu_set_subslices(sseu
, s
, s
== 0 ?
340 subslice_mask_with_eus
:
341 subslice_mask_with_eus
& 0x3);
344 sseu
->eu_total
= compute_eu_total(sseu
);
347 * CNL is expected to always have a uniform distribution
348 * of EU across subslices with the exception that any one
349 * EU in any one subslice may be fused off for die
352 sseu
->eu_per_subslice
= intel_sseu_subslice_total(sseu
) ?
353 DIV_ROUND_UP(sseu
->eu_total
,
354 intel_sseu_subslice_total(sseu
)) :
357 /* No restrictions on Power Gating */
358 sseu
->has_slice_pg
= 1;
359 sseu
->has_subslice_pg
= 1;
363 static void cherryview_sseu_info_init(struct drm_i915_private
*dev_priv
)
365 struct sseu_dev_info
*sseu
= &RUNTIME_INFO(dev_priv
)->sseu
;
367 u8 subslice_mask
= 0;
369 fuse
= I915_READ(CHV_FUSE_GT
);
371 sseu
->slice_mask
= BIT(0);
372 intel_sseu_set_info(sseu
, 1, 2, 8);
374 if (!(fuse
& CHV_FGT_DISABLE_SS0
)) {
376 ((fuse
& CHV_FGT_EU_DIS_SS0_R0_MASK
) >>
377 CHV_FGT_EU_DIS_SS0_R0_SHIFT
) |
378 (((fuse
& CHV_FGT_EU_DIS_SS0_R1_MASK
) >>
379 CHV_FGT_EU_DIS_SS0_R1_SHIFT
) << 4);
381 subslice_mask
|= BIT(0);
382 sseu_set_eus(sseu
, 0, 0, ~disabled_mask
);
385 if (!(fuse
& CHV_FGT_DISABLE_SS1
)) {
387 ((fuse
& CHV_FGT_EU_DIS_SS1_R0_MASK
) >>
388 CHV_FGT_EU_DIS_SS1_R0_SHIFT
) |
389 (((fuse
& CHV_FGT_EU_DIS_SS1_R1_MASK
) >>
390 CHV_FGT_EU_DIS_SS1_R1_SHIFT
) << 4);
392 subslice_mask
|= BIT(1);
393 sseu_set_eus(sseu
, 0, 1, ~disabled_mask
);
396 intel_sseu_set_subslices(sseu
, 0, subslice_mask
);
398 sseu
->eu_total
= compute_eu_total(sseu
);
401 * CHV expected to always have a uniform distribution of EU
404 sseu
->eu_per_subslice
= intel_sseu_subslice_total(sseu
) ?
406 intel_sseu_subslice_total(sseu
) :
409 * CHV supports subslice power gating on devices with more than
410 * one subslice, and supports EU power gating on devices with
411 * more than one EU pair per subslice.
413 sseu
->has_slice_pg
= 0;
414 sseu
->has_subslice_pg
= intel_sseu_subslice_total(sseu
) > 1;
415 sseu
->has_eu_pg
= (sseu
->eu_per_subslice
> 2);
418 static void gen9_sseu_info_init(struct drm_i915_private
*dev_priv
)
420 struct intel_device_info
*info
= mkwrite_device_info(dev_priv
);
421 struct sseu_dev_info
*sseu
= &RUNTIME_INFO(dev_priv
)->sseu
;
423 u32 fuse2
, eu_disable
, subslice_mask
;
424 const u8 eu_mask
= 0xff;
426 fuse2
= I915_READ(GEN8_FUSE2
);
427 sseu
->slice_mask
= (fuse2
& GEN8_F2_S_ENA_MASK
) >> GEN8_F2_S_ENA_SHIFT
;
429 /* BXT has a single slice and at most 3 subslices. */
430 intel_sseu_set_info(sseu
, IS_GEN9_LP(dev_priv
) ? 1 : 3,
431 IS_GEN9_LP(dev_priv
) ? 3 : 4, 8);
434 * The subslice disable field is global, i.e. it applies
435 * to each of the enabled slices.
437 subslice_mask
= (1 << sseu
->max_subslices
) - 1;
438 subslice_mask
&= ~((fuse2
& GEN9_F2_SS_DIS_MASK
) >>
439 GEN9_F2_SS_DIS_SHIFT
);
442 * Iterate through enabled slices and subslices to
443 * count the total enabled EU.
445 for (s
= 0; s
< sseu
->max_slices
; s
++) {
446 if (!(sseu
->slice_mask
& BIT(s
)))
447 /* skip disabled slice */
450 intel_sseu_set_subslices(sseu
, s
, subslice_mask
);
452 eu_disable
= I915_READ(GEN9_EU_DISABLE(s
));
453 for (ss
= 0; ss
< sseu
->max_subslices
; ss
++) {
457 if (!intel_sseu_has_subslice(sseu
, s
, ss
))
458 /* skip disabled subslice */
461 eu_disabled_mask
= (eu_disable
>> (ss
* 8)) & eu_mask
;
463 sseu_set_eus(sseu
, s
, ss
, ~eu_disabled_mask
);
465 eu_per_ss
= sseu
->max_eus_per_subslice
-
466 hweight8(eu_disabled_mask
);
469 * Record which subslice(s) has(have) 7 EUs. we
470 * can tune the hash used to spread work among
471 * subslices if they are unbalanced.
474 sseu
->subslice_7eu
[s
] |= BIT(ss
);
478 sseu
->eu_total
= compute_eu_total(sseu
);
481 * SKL is expected to always have a uniform distribution
482 * of EU across subslices with the exception that any one
483 * EU in any one subslice may be fused off for die
484 * recovery. BXT is expected to be perfectly uniform in EU
487 sseu
->eu_per_subslice
= intel_sseu_subslice_total(sseu
) ?
488 DIV_ROUND_UP(sseu
->eu_total
,
489 intel_sseu_subslice_total(sseu
)) :
492 * SKL+ supports slice power gating on devices with more than
493 * one slice, and supports EU power gating on devices with
494 * more than one EU pair per subslice. BXT+ supports subslice
495 * power gating on devices with more than one subslice, and
496 * supports EU power gating on devices with more than one EU
500 !IS_GEN9_LP(dev_priv
) && hweight8(sseu
->slice_mask
) > 1;
501 sseu
->has_subslice_pg
=
502 IS_GEN9_LP(dev_priv
) && intel_sseu_subslice_total(sseu
) > 1;
503 sseu
->has_eu_pg
= sseu
->eu_per_subslice
> 2;
505 if (IS_GEN9_LP(dev_priv
)) {
506 #define IS_SS_DISABLED(ss) (!(sseu->subslice_mask[0] & BIT(ss)))
507 info
->has_pooled_eu
= hweight8(sseu
->subslice_mask
[0]) == 3;
509 sseu
->min_eu_in_pool
= 0;
510 if (info
->has_pooled_eu
) {
511 if (IS_SS_DISABLED(2) || IS_SS_DISABLED(0))
512 sseu
->min_eu_in_pool
= 3;
513 else if (IS_SS_DISABLED(1))
514 sseu
->min_eu_in_pool
= 6;
516 sseu
->min_eu_in_pool
= 9;
518 #undef IS_SS_DISABLED
522 static void bdw_sseu_info_init(struct drm_i915_private
*dev_priv
)
524 struct sseu_dev_info
*sseu
= &RUNTIME_INFO(dev_priv
)->sseu
;
526 u32 fuse2
, subslice_mask
, eu_disable
[3]; /* s_max */
528 fuse2
= I915_READ(GEN8_FUSE2
);
529 sseu
->slice_mask
= (fuse2
& GEN8_F2_S_ENA_MASK
) >> GEN8_F2_S_ENA_SHIFT
;
530 intel_sseu_set_info(sseu
, 3, 3, 8);
533 * The subslice disable field is global, i.e. it applies
534 * to each of the enabled slices.
536 subslice_mask
= GENMASK(sseu
->max_subslices
- 1, 0);
537 subslice_mask
&= ~((fuse2
& GEN8_F2_SS_DIS_MASK
) >>
538 GEN8_F2_SS_DIS_SHIFT
);
540 eu_disable
[0] = I915_READ(GEN8_EU_DISABLE0
) & GEN8_EU_DIS0_S0_MASK
;
541 eu_disable
[1] = (I915_READ(GEN8_EU_DISABLE0
) >> GEN8_EU_DIS0_S1_SHIFT
) |
542 ((I915_READ(GEN8_EU_DISABLE1
) & GEN8_EU_DIS1_S1_MASK
) <<
543 (32 - GEN8_EU_DIS0_S1_SHIFT
));
544 eu_disable
[2] = (I915_READ(GEN8_EU_DISABLE1
) >> GEN8_EU_DIS1_S2_SHIFT
) |
545 ((I915_READ(GEN8_EU_DISABLE2
) & GEN8_EU_DIS2_S2_MASK
) <<
546 (32 - GEN8_EU_DIS1_S2_SHIFT
));
549 * Iterate through enabled slices and subslices to
550 * count the total enabled EU.
552 for (s
= 0; s
< sseu
->max_slices
; s
++) {
553 if (!(sseu
->slice_mask
& BIT(s
)))
554 /* skip disabled slice */
557 intel_sseu_set_subslices(sseu
, s
, subslice_mask
);
559 for (ss
= 0; ss
< sseu
->max_subslices
; ss
++) {
563 if (!intel_sseu_has_subslice(sseu
, s
, ss
))
564 /* skip disabled subslice */
568 eu_disable
[s
] >> (ss
* sseu
->max_eus_per_subslice
);
570 sseu_set_eus(sseu
, s
, ss
, ~eu_disabled_mask
);
572 n_disabled
= hweight8(eu_disabled_mask
);
575 * Record which subslices have 7 EUs.
577 if (sseu
->max_eus_per_subslice
- n_disabled
== 7)
578 sseu
->subslice_7eu
[s
] |= 1 << ss
;
582 sseu
->eu_total
= compute_eu_total(sseu
);
585 * BDW is expected to always have a uniform distribution of EU across
586 * subslices with the exception that any one EU in any one subslice may
587 * be fused off for die recovery.
589 sseu
->eu_per_subslice
= intel_sseu_subslice_total(sseu
) ?
590 DIV_ROUND_UP(sseu
->eu_total
,
591 intel_sseu_subslice_total(sseu
)) :
595 * BDW supports slice power gating on devices with more than
598 sseu
->has_slice_pg
= hweight8(sseu
->slice_mask
) > 1;
599 sseu
->has_subslice_pg
= 0;
603 static void hsw_sseu_info_init(struct drm_i915_private
*dev_priv
)
605 struct sseu_dev_info
*sseu
= &RUNTIME_INFO(dev_priv
)->sseu
;
607 u8 subslice_mask
= 0;
611 * There isn't a register to tell us how many slices/subslices. We
612 * work off the PCI-ids here.
614 switch (INTEL_INFO(dev_priv
)->gt
) {
616 MISSING_CASE(INTEL_INFO(dev_priv
)->gt
);
619 sseu
->slice_mask
= BIT(0);
620 subslice_mask
= BIT(0);
623 sseu
->slice_mask
= BIT(0);
624 subslice_mask
= BIT(0) | BIT(1);
627 sseu
->slice_mask
= BIT(0) | BIT(1);
628 subslice_mask
= BIT(0) | BIT(1);
632 fuse1
= I915_READ(HSW_PAVP_FUSE1
);
633 switch ((fuse1
& HSW_F1_EU_DIS_MASK
) >> HSW_F1_EU_DIS_SHIFT
) {
635 MISSING_CASE((fuse1
& HSW_F1_EU_DIS_MASK
) >>
636 HSW_F1_EU_DIS_SHIFT
);
638 case HSW_F1_EU_DIS_10EUS
:
639 sseu
->eu_per_subslice
= 10;
641 case HSW_F1_EU_DIS_8EUS
:
642 sseu
->eu_per_subslice
= 8;
644 case HSW_F1_EU_DIS_6EUS
:
645 sseu
->eu_per_subslice
= 6;
649 intel_sseu_set_info(sseu
, hweight8(sseu
->slice_mask
),
650 hweight8(subslice_mask
),
651 sseu
->eu_per_subslice
);
653 for (s
= 0; s
< sseu
->max_slices
; s
++) {
654 intel_sseu_set_subslices(sseu
, s
, subslice_mask
);
656 for (ss
= 0; ss
< sseu
->max_subslices
; ss
++) {
657 sseu_set_eus(sseu
, s
, ss
,
658 (1UL << sseu
->eu_per_subslice
) - 1);
662 sseu
->eu_total
= compute_eu_total(sseu
);
664 /* No powergating for you. */
665 sseu
->has_slice_pg
= 0;
666 sseu
->has_subslice_pg
= 0;
670 static u32
read_reference_ts_freq(struct drm_i915_private
*dev_priv
)
672 u32 ts_override
= I915_READ(GEN9_TIMESTAMP_OVERRIDE
);
673 u32 base_freq
, frac_freq
;
675 base_freq
= ((ts_override
& GEN9_TIMESTAMP_OVERRIDE_US_COUNTER_DIVIDER_MASK
) >>
676 GEN9_TIMESTAMP_OVERRIDE_US_COUNTER_DIVIDER_SHIFT
) + 1;
679 frac_freq
= ((ts_override
&
680 GEN9_TIMESTAMP_OVERRIDE_US_COUNTER_DENOMINATOR_MASK
) >>
681 GEN9_TIMESTAMP_OVERRIDE_US_COUNTER_DENOMINATOR_SHIFT
);
682 frac_freq
= 1000 / (frac_freq
+ 1);
684 return base_freq
+ frac_freq
;
687 static u32
gen10_get_crystal_clock_freq(struct drm_i915_private
*dev_priv
,
690 u32 f19_2_mhz
= 19200;
692 u32 crystal_clock
= (rpm_config_reg
&
693 GEN9_RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_MASK
) >>
694 GEN9_RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_SHIFT
;
696 switch (crystal_clock
) {
697 case GEN9_RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_19_2_MHZ
:
699 case GEN9_RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_24_MHZ
:
702 MISSING_CASE(crystal_clock
);
707 static u32
gen11_get_crystal_clock_freq(struct drm_i915_private
*dev_priv
,
710 u32 f19_2_mhz
= 19200;
713 u32 f38_4_mhz
= 38400;
714 u32 crystal_clock
= (rpm_config_reg
&
715 GEN11_RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_MASK
) >>
716 GEN11_RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_SHIFT
;
718 switch (crystal_clock
) {
719 case GEN11_RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_24_MHZ
:
721 case GEN11_RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_19_2_MHZ
:
723 case GEN11_RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_38_4_MHZ
:
725 case GEN11_RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_25_MHZ
:
728 MISSING_CASE(crystal_clock
);
733 static u32
read_timestamp_frequency(struct drm_i915_private
*dev_priv
)
735 u32 f12_5_mhz
= 12500;
736 u32 f19_2_mhz
= 19200;
739 if (INTEL_GEN(dev_priv
) <= 4) {
742 * "The value in this register increments once every 16
743 * hclks." (through the “Clocking Configuration”
744 * (“CLKCFG”) MCHBAR register)
746 return dev_priv
->rawclk_freq
/ 16;
747 } else if (INTEL_GEN(dev_priv
) <= 8) {
750 * "The PCU TSC counts 10ns increments; this timestamp
751 * reflects bits 38:3 of the TSC (i.e. 80ns granularity,
752 * rolling over every 1.5 hours).
755 } else if (INTEL_GEN(dev_priv
) <= 9) {
756 u32 ctc_reg
= I915_READ(CTC_MODE
);
759 if ((ctc_reg
& CTC_SOURCE_PARAMETER_MASK
) == CTC_SOURCE_DIVIDE_LOGIC
) {
760 freq
= read_reference_ts_freq(dev_priv
);
762 freq
= IS_GEN9_LP(dev_priv
) ? f19_2_mhz
: f24_mhz
;
764 /* Now figure out how the command stream's timestamp
765 * register increments from this frequency (it might
766 * increment only every few clock cycle).
768 freq
>>= 3 - ((ctc_reg
& CTC_SHIFT_PARAMETER_MASK
) >>
769 CTC_SHIFT_PARAMETER_SHIFT
);
773 } else if (INTEL_GEN(dev_priv
) <= 12) {
774 u32 ctc_reg
= I915_READ(CTC_MODE
);
777 /* First figure out the reference frequency. There are 2 ways
778 * we can compute the frequency, either through the
779 * TIMESTAMP_OVERRIDE register or through RPM_CONFIG. CTC_MODE
780 * tells us which one we should use.
782 if ((ctc_reg
& CTC_SOURCE_PARAMETER_MASK
) == CTC_SOURCE_DIVIDE_LOGIC
) {
783 freq
= read_reference_ts_freq(dev_priv
);
785 u32 rpm_config_reg
= I915_READ(RPM_CONFIG0
);
787 if (INTEL_GEN(dev_priv
) <= 10)
788 freq
= gen10_get_crystal_clock_freq(dev_priv
,
791 freq
= gen11_get_crystal_clock_freq(dev_priv
,
794 /* Now figure out how the command stream's timestamp
795 * register increments from this frequency (it might
796 * increment only every few clock cycle).
798 freq
>>= 3 - ((rpm_config_reg
&
799 GEN10_RPM_CONFIG0_CTC_SHIFT_PARAMETER_MASK
) >>
800 GEN10_RPM_CONFIG0_CTC_SHIFT_PARAMETER_SHIFT
);
806 MISSING_CASE("Unknown gen, unable to read command streamer timestamp frequency\n");
810 #undef INTEL_VGA_DEVICE
811 #define INTEL_VGA_DEVICE(id, info) (id)
813 static const u16 subplatform_ult_ids
[] = {
814 INTEL_HSW_ULT_GT1_IDS(0),
815 INTEL_HSW_ULT_GT2_IDS(0),
816 INTEL_HSW_ULT_GT3_IDS(0),
817 INTEL_BDW_ULT_GT1_IDS(0),
818 INTEL_BDW_ULT_GT2_IDS(0),
819 INTEL_BDW_ULT_GT3_IDS(0),
820 INTEL_BDW_ULT_RSVD_IDS(0),
821 INTEL_SKL_ULT_GT1_IDS(0),
822 INTEL_SKL_ULT_GT2_IDS(0),
823 INTEL_SKL_ULT_GT3_IDS(0),
824 INTEL_KBL_ULT_GT1_IDS(0),
825 INTEL_KBL_ULT_GT2_IDS(0),
826 INTEL_KBL_ULT_GT3_IDS(0),
827 INTEL_CFL_U_GT2_IDS(0),
828 INTEL_CFL_U_GT3_IDS(0),
829 INTEL_WHL_U_GT1_IDS(0),
830 INTEL_WHL_U_GT2_IDS(0),
831 INTEL_WHL_U_GT3_IDS(0),
832 INTEL_CML_U_GT1_IDS(0),
833 INTEL_CML_U_GT2_IDS(0),
836 static const u16 subplatform_ulx_ids
[] = {
837 INTEL_HSW_ULX_GT1_IDS(0),
838 INTEL_HSW_ULX_GT2_IDS(0),
839 INTEL_BDW_ULX_GT1_IDS(0),
840 INTEL_BDW_ULX_GT2_IDS(0),
841 INTEL_BDW_ULX_GT3_IDS(0),
842 INTEL_BDW_ULX_RSVD_IDS(0),
843 INTEL_SKL_ULX_GT1_IDS(0),
844 INTEL_SKL_ULX_GT2_IDS(0),
845 INTEL_KBL_ULX_GT1_IDS(0),
846 INTEL_KBL_ULX_GT2_IDS(0),
847 INTEL_AML_KBL_GT2_IDS(0),
848 INTEL_AML_CFL_GT2_IDS(0),
851 static const u16 subplatform_portf_ids
[] = {
852 INTEL_CNL_PORT_F_IDS(0),
853 INTEL_ICL_PORT_F_IDS(0),
856 static bool find_devid(u16 id
, const u16
*p
, unsigned int num
)
858 for (; num
; num
--, p
++) {
866 void intel_device_info_subplatform_init(struct drm_i915_private
*i915
)
868 const struct intel_device_info
*info
= INTEL_INFO(i915
);
869 const struct intel_runtime_info
*rinfo
= RUNTIME_INFO(i915
);
870 const unsigned int pi
= __platform_mask_index(rinfo
, info
->platform
);
871 const unsigned int pb
= __platform_mask_bit(rinfo
, info
->platform
);
872 u16 devid
= INTEL_DEVID(i915
);
875 /* Make sure IS_<platform> checks are working. */
876 RUNTIME_INFO(i915
)->platform_mask
[pi
] = BIT(pb
);
878 /* Find and mark subplatform bits based on the PCI device id. */
879 if (find_devid(devid
, subplatform_ult_ids
,
880 ARRAY_SIZE(subplatform_ult_ids
))) {
881 mask
= BIT(INTEL_SUBPLATFORM_ULT
);
882 } else if (find_devid(devid
, subplatform_ulx_ids
,
883 ARRAY_SIZE(subplatform_ulx_ids
))) {
884 mask
= BIT(INTEL_SUBPLATFORM_ULX
);
885 if (IS_HASWELL(i915
) || IS_BROADWELL(i915
)) {
886 /* ULX machines are also considered ULT. */
887 mask
|= BIT(INTEL_SUBPLATFORM_ULT
);
889 } else if (find_devid(devid
, subplatform_portf_ids
,
890 ARRAY_SIZE(subplatform_portf_ids
))) {
891 mask
= BIT(INTEL_SUBPLATFORM_PORTF
);
894 GEM_BUG_ON(mask
& ~INTEL_SUBPLATFORM_BITS
);
896 RUNTIME_INFO(i915
)->platform_mask
[pi
] |= mask
;
900 * intel_device_info_runtime_init - initialize runtime info
901 * @dev_priv: the i915 device
903 * Determine various intel_device_info fields at runtime.
905 * Use it when either:
906 * - it's judged too laborious to fill n static structures with the limit
907 * when a simple if statement does the job,
908 * - run-time checks (eg read fuse/strap registers) are needed.
910 * This function needs to be called:
911 * - after the MMIO has been setup as we are reading registers,
912 * - after the PCH has been detected,
913 * - before the first usage of the fields it can tweak.
915 void intel_device_info_runtime_init(struct drm_i915_private
*dev_priv
)
917 struct intel_device_info
*info
= mkwrite_device_info(dev_priv
);
918 struct intel_runtime_info
*runtime
= RUNTIME_INFO(dev_priv
);
921 if (INTEL_GEN(dev_priv
) >= 10) {
922 for_each_pipe(dev_priv
, pipe
)
923 runtime
->num_scalers
[pipe
] = 2;
924 } else if (IS_GEN(dev_priv
, 9)) {
925 runtime
->num_scalers
[PIPE_A
] = 2;
926 runtime
->num_scalers
[PIPE_B
] = 2;
927 runtime
->num_scalers
[PIPE_C
] = 1;
930 BUILD_BUG_ON(BITS_PER_TYPE(intel_engine_mask_t
) < I915_NUM_ENGINES
);
932 if (INTEL_GEN(dev_priv
) >= 11)
933 for_each_pipe(dev_priv
, pipe
)
934 runtime
->num_sprites
[pipe
] = 6;
935 else if (IS_GEN(dev_priv
, 10) || IS_GEMINILAKE(dev_priv
))
936 for_each_pipe(dev_priv
, pipe
)
937 runtime
->num_sprites
[pipe
] = 3;
938 else if (IS_BROXTON(dev_priv
)) {
940 * Skylake and Broxton currently don't expose the topmost plane as its
941 * use is exclusive with the legacy cursor and we only want to expose
942 * one of those, not both. Until we can safely expose the topmost plane
943 * as a DRM_PLANE_TYPE_CURSOR with all the features exposed/supported,
944 * we don't expose the topmost plane at all to prevent ABI breakage
948 runtime
->num_sprites
[PIPE_A
] = 2;
949 runtime
->num_sprites
[PIPE_B
] = 2;
950 runtime
->num_sprites
[PIPE_C
] = 1;
951 } else if (IS_VALLEYVIEW(dev_priv
) || IS_CHERRYVIEW(dev_priv
)) {
952 for_each_pipe(dev_priv
, pipe
)
953 runtime
->num_sprites
[pipe
] = 2;
954 } else if (INTEL_GEN(dev_priv
) >= 5 || IS_G4X(dev_priv
)) {
955 for_each_pipe(dev_priv
, pipe
)
956 runtime
->num_sprites
[pipe
] = 1;
959 if (HAS_DISPLAY(dev_priv
) && IS_GEN_RANGE(dev_priv
, 7, 8) &&
960 HAS_PCH_SPLIT(dev_priv
)) {
961 u32 fuse_strap
= I915_READ(FUSE_STRAP
);
962 u32 sfuse_strap
= I915_READ(SFUSE_STRAP
);
965 * SFUSE_STRAP is supposed to have a bit signalling the display
966 * is fused off. Unfortunately it seems that, at least in
967 * certain cases, fused off display means that PCH display
968 * reads don't land anywhere. In that case, we read 0s.
970 * On CPT/PPT, we can detect this case as SFUSE_STRAP_FUSE_LOCK
971 * should be set when taking over after the firmware.
973 if (fuse_strap
& ILK_INTERNAL_DISPLAY_DISABLE
||
974 sfuse_strap
& SFUSE_STRAP_DISPLAY_DISABLED
||
975 (HAS_PCH_CPT(dev_priv
) &&
976 !(sfuse_strap
& SFUSE_STRAP_FUSE_LOCK
))) {
977 DRM_INFO("Display fused off, disabling\n");
979 } else if (fuse_strap
& IVB_PIPE_C_DISABLE
) {
980 DRM_INFO("PipeC fused off\n");
981 info
->pipe_mask
&= ~BIT(PIPE_C
);
983 } else if (HAS_DISPLAY(dev_priv
) && INTEL_GEN(dev_priv
) >= 9) {
984 u32 dfsm
= I915_READ(SKL_DFSM
);
985 u8 enabled_mask
= info
->pipe_mask
;
987 if (dfsm
& SKL_DFSM_PIPE_A_DISABLE
)
988 enabled_mask
&= ~BIT(PIPE_A
);
989 if (dfsm
& SKL_DFSM_PIPE_B_DISABLE
)
990 enabled_mask
&= ~BIT(PIPE_B
);
991 if (dfsm
& SKL_DFSM_PIPE_C_DISABLE
)
992 enabled_mask
&= ~BIT(PIPE_C
);
993 if (INTEL_GEN(dev_priv
) >= 12 &&
994 (dfsm
& TGL_DFSM_PIPE_D_DISABLE
))
995 enabled_mask
&= ~BIT(PIPE_D
);
998 * At least one pipe should be enabled and if there are
999 * disabled pipes, they should be the last ones, with no holes
1002 if (enabled_mask
== 0 || !is_power_of_2(enabled_mask
+ 1))
1003 DRM_ERROR("invalid pipe fuse configuration: enabled_mask=0x%x\n",
1006 info
->pipe_mask
= enabled_mask
;
1008 if (dfsm
& SKL_DFSM_DISPLAY_HDCP_DISABLE
)
1009 info
->display
.has_hdcp
= 0;
1011 if (dfsm
& SKL_DFSM_DISPLAY_PM_DISABLE
)
1012 info
->display
.has_fbc
= 0;
1014 if (INTEL_GEN(dev_priv
) >= 11 && (dfsm
& ICL_DFSM_DMC_DISABLE
))
1015 info
->display
.has_csr
= 0;
1017 if (INTEL_GEN(dev_priv
) >= 10 &&
1018 (dfsm
& CNL_DFSM_DISPLAY_DSC_DISABLE
))
1019 info
->display
.has_dsc
= 0;
1022 /* Initialize slice/subslice/EU info */
1023 if (IS_HASWELL(dev_priv
))
1024 hsw_sseu_info_init(dev_priv
);
1025 else if (IS_CHERRYVIEW(dev_priv
))
1026 cherryview_sseu_info_init(dev_priv
);
1027 else if (IS_BROADWELL(dev_priv
))
1028 bdw_sseu_info_init(dev_priv
);
1029 else if (IS_GEN(dev_priv
, 9))
1030 gen9_sseu_info_init(dev_priv
);
1031 else if (IS_GEN(dev_priv
, 10))
1032 gen10_sseu_info_init(dev_priv
);
1033 else if (IS_GEN(dev_priv
, 11))
1034 gen11_sseu_info_init(dev_priv
);
1035 else if (INTEL_GEN(dev_priv
) >= 12)
1036 gen12_sseu_info_init(dev_priv
);
1038 if (IS_GEN(dev_priv
, 6) && intel_vtd_active()) {
1039 DRM_INFO("Disabling ppGTT for VT-d support\n");
1040 info
->ppgtt_type
= INTEL_PPGTT_NONE
;
1043 /* Initialize command stream timestamp frequency */
1044 runtime
->cs_timestamp_frequency_khz
= read_timestamp_frequency(dev_priv
);
1047 void intel_driver_caps_print(const struct intel_driver_caps
*caps
,
1048 struct drm_printer
*p
)
1050 drm_printf(p
, "Has logical contexts? %s\n",
1051 yesno(caps
->has_logical_contexts
));
1052 drm_printf(p
, "scheduler: %x\n", caps
->scheduler
);
1056 * Determine which engines are fused off in our particular hardware. Since the
1057 * fuse register is in the blitter powerwell, we need forcewake to be ready at
1058 * this point (but later we need to prune the forcewake domains for engines that
1059 * are indeed fused off).
1061 void intel_device_info_init_mmio(struct drm_i915_private
*dev_priv
)
1063 struct intel_device_info
*info
= mkwrite_device_info(dev_priv
);
1064 unsigned int logical_vdbox
= 0;
1070 if (INTEL_GEN(dev_priv
) < 11)
1073 media_fuse
= ~I915_READ(GEN11_GT_VEBOX_VDBOX_DISABLE
);
1075 vdbox_mask
= media_fuse
& GEN11_GT_VDBOX_DISABLE_MASK
;
1076 vebox_mask
= (media_fuse
& GEN11_GT_VEBOX_DISABLE_MASK
) >>
1077 GEN11_GT_VEBOX_DISABLE_SHIFT
;
1079 for (i
= 0; i
< I915_MAX_VCS
; i
++) {
1080 if (!HAS_ENGINE(dev_priv
, _VCS(i
))) {
1081 vdbox_mask
&= ~BIT(i
);
1085 if (!(BIT(i
) & vdbox_mask
)) {
1086 info
->engine_mask
&= ~BIT(_VCS(i
));
1087 DRM_DEBUG_DRIVER("vcs%u fused off\n", i
);
1092 * In Gen11, only even numbered logical VDBOXes are
1093 * hooked up to an SFC (Scaler & Format Converter) unit.
1094 * In TGL each VDBOX has access to an SFC.
1096 if (INTEL_GEN(dev_priv
) >= 12 || logical_vdbox
++ % 2 == 0)
1097 RUNTIME_INFO(dev_priv
)->vdbox_sfc_access
|= BIT(i
);
1099 DRM_DEBUG_DRIVER("vdbox enable: %04x, instances: %04lx\n",
1100 vdbox_mask
, VDBOX_MASK(dev_priv
));
1101 GEM_BUG_ON(vdbox_mask
!= VDBOX_MASK(dev_priv
));
1103 for (i
= 0; i
< I915_MAX_VECS
; i
++) {
1104 if (!HAS_ENGINE(dev_priv
, _VECS(i
))) {
1105 vebox_mask
&= ~BIT(i
);
1109 if (!(BIT(i
) & vebox_mask
)) {
1110 info
->engine_mask
&= ~BIT(_VECS(i
));
1111 DRM_DEBUG_DRIVER("vecs%u fused off\n", i
);
1114 DRM_DEBUG_DRIVER("vebox enable: %04x, instances: %04lx\n",
1115 vebox_mask
, VEBOX_MASK(dev_priv
));
1116 GEM_BUG_ON(vebox_mask
!= VEBOX_MASK(dev_priv
));