Commit | Line | Data |
---|---|---|
3bbaba0c PA |
1 | /* |
2 | * Copyright (c) 2015 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 | * The above copyright notice and this permission notice (including the next | |
11 | * paragraph) shall be included in all copies or substantial portions of the | |
12 | * Software. | |
13 | * | |
14 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR | |
15 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | |
16 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL | |
17 | * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER | |
18 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, | |
19 | * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE | |
20 | * SOFTWARE. | |
21 | */ | |
22 | ||
23 | #include "intel_mocs.h" | |
24 | #include "intel_lrc.h" | |
25 | #include "intel_ringbuffer.h" | |
26 | ||
27 | /* structures required */ | |
28 | struct drm_i915_mocs_entry { | |
29 | u32 control_value; | |
30 | u16 l3cc_value; | |
31 | }; | |
32 | ||
33 | struct drm_i915_mocs_table { | |
34 | u32 size; | |
35 | const struct drm_i915_mocs_entry *table; | |
36 | }; | |
37 | ||
38 | /* Defines for the tables (XXX_MOCS_0 - XXX_MOCS_63) */ | |
39 | #define LE_CACHEABILITY(value) ((value) << 0) | |
40 | #define LE_TGT_CACHE(value) ((value) << 2) | |
41 | #define LE_LRUM(value) ((value) << 4) | |
42 | #define LE_AOM(value) ((value) << 6) | |
43 | #define LE_RSC(value) ((value) << 7) | |
44 | #define LE_SCC(value) ((value) << 8) | |
45 | #define LE_PFM(value) ((value) << 11) | |
46 | #define LE_SCF(value) ((value) << 14) | |
47 | ||
48 | /* Defines for the tables (LNCFMOCS0 - LNCFMOCS31) - two entries per word */ | |
49 | #define L3_ESC(value) ((value) << 0) | |
50 | #define L3_SCC(value) ((value) << 1) | |
51 | #define L3_CACHEABILITY(value) ((value) << 4) | |
52 | ||
53 | /* Helper defines */ | |
54 | #define GEN9_NUM_MOCS_ENTRIES 62 /* 62 out of 64 - 63 & 64 are reserved. */ | |
55 | ||
56 | /* (e)LLC caching options */ | |
57 | #define LE_PAGETABLE 0 | |
58 | #define LE_UC 1 | |
59 | #define LE_WT 2 | |
60 | #define LE_WB 3 | |
61 | ||
62 | /* L3 caching options */ | |
63 | #define L3_DIRECT 0 | |
64 | #define L3_UC 1 | |
65 | #define L3_RESERVED 2 | |
66 | #define L3_WB 3 | |
67 | ||
68 | /* Target cache */ | |
e419899b ID |
69 | #define LE_TC_PAGETABLE 0 |
70 | #define LE_TC_LLC 1 | |
71 | #define LE_TC_LLC_ELLC 2 | |
72 | #define LE_TC_LLC_ELLC_ALT 3 | |
3bbaba0c PA |
73 | |
74 | /* | |
75 | * MOCS tables | |
76 | * | |
77 | * These are the MOCS tables that are programmed across all the rings. | |
78 | * The control value is programmed to all the rings that support the | |
79 | * MOCS registers. While the l3cc_values are only programmed to the | |
80 | * LNCFCMOCS0 - LNCFCMOCS32 registers. | |
81 | * | |
82 | * These tables are intended to be kept reasonably consistent across | |
83 | * platforms. However some of the fields are not applicable to all of | |
84 | * them. | |
85 | * | |
86 | * Entries not part of the following tables are undefined as far as | |
87 | * userspace is concerned and shouldn't be relied upon. For the time | |
88 | * being they will be implicitly initialized to the strictest caching | |
89 | * configuration (uncached) to guarantee forwards compatibility with | |
90 | * userspace programs written against more recent kernels providing | |
91 | * additional MOCS entries. | |
92 | * | |
93 | * NOTE: These tables MUST start with being uncached and the length | |
94 | * MUST be less than 63 as the last two registers are reserved | |
95 | * by the hardware. These tables are part of the kernel ABI and | |
96 | * may only be updated incrementally by adding entries at the | |
97 | * end. | |
98 | */ | |
99 | static const struct drm_i915_mocs_entry skylake_mocs_table[] = { | |
3373ce2e ID |
100 | [I915_MOCS_UNCACHED] = { |
101 | /* 0x00000009 */ | |
e419899b ID |
102 | .control_value = LE_CACHEABILITY(LE_UC) | |
103 | LE_TGT_CACHE(LE_TC_LLC_ELLC) | | |
104 | LE_LRUM(0) | LE_AOM(0) | LE_RSC(0) | LE_SCC(0) | | |
105 | LE_PFM(0) | LE_SCF(0), | |
106 | ||
107 | /* 0x0010 */ | |
108 | .l3cc_value = L3_ESC(0) | L3_SCC(0) | L3_CACHEABILITY(L3_UC), | |
109 | }, | |
3373ce2e | 110 | [I915_MOCS_PTE] = { |
e419899b ID |
111 | /* 0x00000038 */ |
112 | .control_value = LE_CACHEABILITY(LE_PAGETABLE) | | |
113 | LE_TGT_CACHE(LE_TC_LLC_ELLC) | | |
114 | LE_LRUM(3) | LE_AOM(0) | LE_RSC(0) | LE_SCC(0) | | |
115 | LE_PFM(0) | LE_SCF(0), | |
116 | /* 0x0030 */ | |
117 | .l3cc_value = L3_ESC(0) | L3_SCC(0) | L3_CACHEABILITY(L3_WB), | |
118 | }, | |
3373ce2e | 119 | [I915_MOCS_CACHED] = { |
e419899b ID |
120 | /* 0x0000003b */ |
121 | .control_value = LE_CACHEABILITY(LE_WB) | | |
122 | LE_TGT_CACHE(LE_TC_LLC_ELLC) | | |
123 | LE_LRUM(3) | LE_AOM(0) | LE_RSC(0) | LE_SCC(0) | | |
124 | LE_PFM(0) | LE_SCF(0), | |
125 | /* 0x0030 */ | |
126 | .l3cc_value = L3_ESC(0) | L3_SCC(0) | L3_CACHEABILITY(L3_WB), | |
127 | }, | |
3bbaba0c PA |
128 | }; |
129 | ||
130 | /* NOTE: the LE_TGT_CACHE is not used on Broxton */ | |
131 | static const struct drm_i915_mocs_entry broxton_mocs_table[] = { | |
3373ce2e | 132 | [I915_MOCS_UNCACHED] = { |
e419899b ID |
133 | /* 0x00000009 */ |
134 | .control_value = LE_CACHEABILITY(LE_UC) | | |
135 | LE_TGT_CACHE(LE_TC_LLC_ELLC) | | |
136 | LE_LRUM(0) | LE_AOM(0) | LE_RSC(0) | LE_SCC(0) | | |
137 | LE_PFM(0) | LE_SCF(0), | |
138 | ||
139 | /* 0x0010 */ | |
140 | .l3cc_value = L3_ESC(0) | L3_SCC(0) | L3_CACHEABILITY(L3_UC), | |
141 | }, | |
3373ce2e | 142 | [I915_MOCS_PTE] = { |
e419899b ID |
143 | /* 0x00000038 */ |
144 | .control_value = LE_CACHEABILITY(LE_PAGETABLE) | | |
145 | LE_TGT_CACHE(LE_TC_LLC_ELLC) | | |
146 | LE_LRUM(3) | LE_AOM(0) | LE_RSC(0) | LE_SCC(0) | | |
147 | LE_PFM(0) | LE_SCF(0), | |
148 | ||
149 | /* 0x0030 */ | |
150 | .l3cc_value = L3_ESC(0) | L3_SCC(0) | L3_CACHEABILITY(L3_WB), | |
151 | }, | |
3373ce2e | 152 | [I915_MOCS_CACHED] = { |
6bee14ed ID |
153 | /* 0x00000039 */ |
154 | .control_value = LE_CACHEABILITY(LE_UC) | | |
e419899b ID |
155 | LE_TGT_CACHE(LE_TC_LLC_ELLC) | |
156 | LE_LRUM(3) | LE_AOM(0) | LE_RSC(0) | LE_SCC(0) | | |
157 | LE_PFM(0) | LE_SCF(0), | |
158 | ||
159 | /* 0x0030 */ | |
160 | .l3cc_value = L3_ESC(0) | L3_SCC(0) | L3_CACHEABILITY(L3_WB), | |
161 | }, | |
3bbaba0c PA |
162 | }; |
163 | ||
164 | /** | |
165 | * get_mocs_settings() | |
0ccdacf6 | 166 | * @dev_priv: i915 device. |
3bbaba0c | 167 | * @table: Output table that will be made to point at appropriate |
0ccdacf6 | 168 | * MOCS values for the device. |
3bbaba0c PA |
169 | * |
170 | * This function will return the values of the MOCS table that needs to | |
171 | * be programmed for the platform. It will return the values that need | |
172 | * to be programmed and if they need to be programmed. | |
173 | * | |
174 | * Return: true if there are applicable MOCS settings for the device. | |
175 | */ | |
0ccdacf6 | 176 | static bool get_mocs_settings(struct drm_i915_private *dev_priv, |
3bbaba0c PA |
177 | struct drm_i915_mocs_table *table) |
178 | { | |
179 | bool result = false; | |
180 | ||
0ccdacf6 | 181 | if (IS_SKYLAKE(dev_priv) || IS_KABYLAKE(dev_priv)) { |
3bbaba0c PA |
182 | table->size = ARRAY_SIZE(skylake_mocs_table); |
183 | table->table = skylake_mocs_table; | |
184 | result = true; | |
0ccdacf6 | 185 | } else if (IS_BROXTON(dev_priv)) { |
3bbaba0c PA |
186 | table->size = ARRAY_SIZE(broxton_mocs_table); |
187 | table->table = broxton_mocs_table; | |
188 | result = true; | |
189 | } else { | |
0ccdacf6 | 190 | WARN_ONCE(INTEL_INFO(dev_priv)->gen >= 9, |
3bbaba0c PA |
191 | "Platform that should have a MOCS table does not.\n"); |
192 | } | |
193 | ||
6fc29133 MK |
194 | /* WaDisableSkipCaching:skl,bxt,kbl */ |
195 | if (IS_GEN9(dev_priv)) { | |
196 | int i; | |
197 | ||
198 | for (i = 0; i < table->size; i++) | |
199 | if (WARN_ON(table->table[i].l3cc_value & | |
030daa0f | 200 | (L3_ESC(1) | L3_SCC(0x7)))) |
6fc29133 MK |
201 | return false; |
202 | } | |
203 | ||
3bbaba0c PA |
204 | return result; |
205 | } | |
206 | ||
38a0f2db | 207 | static i915_reg_t mocs_register(enum intel_engine_id engine_id, int index) |
e6c4c763 | 208 | { |
38a0f2db | 209 | switch (engine_id) { |
e6c4c763 VS |
210 | case RCS: |
211 | return GEN9_GFX_MOCS(index); | |
212 | case VCS: | |
213 | return GEN9_MFX0_MOCS(index); | |
214 | case BCS: | |
215 | return GEN9_BLT_MOCS(index); | |
216 | case VECS: | |
217 | return GEN9_VEBOX_MOCS(index); | |
218 | case VCS2: | |
219 | return GEN9_MFX1_MOCS(index); | |
220 | default: | |
38a0f2db | 221 | MISSING_CASE(engine_id); |
f0f59a00 | 222 | return INVALID_MMIO_REG; |
e6c4c763 VS |
223 | } |
224 | } | |
225 | ||
0ccdacf6 PA |
226 | /** |
227 | * intel_mocs_init_engine() - emit the mocs control table | |
228 | * @engine: The engine for whom to emit the registers. | |
229 | * | |
230 | * This function simply emits a MI_LOAD_REGISTER_IMM command for the | |
231 | * given table starting at the given address. | |
232 | * | |
233 | * Return: 0 on success, otherwise the error status. | |
234 | */ | |
235 | int intel_mocs_init_engine(struct intel_engine_cs *engine) | |
236 | { | |
c033666a | 237 | struct drm_i915_private *dev_priv = engine->i915; |
0ccdacf6 PA |
238 | struct drm_i915_mocs_table table; |
239 | unsigned int index; | |
240 | ||
241 | if (!get_mocs_settings(dev_priv, &table)) | |
242 | return 0; | |
243 | ||
244 | if (WARN_ON(table.size > GEN9_NUM_MOCS_ENTRIES)) | |
245 | return -ENODEV; | |
246 | ||
247 | for (index = 0; index < table.size; index++) | |
248 | I915_WRITE(mocs_register(engine->id, index), | |
249 | table.table[index].control_value); | |
250 | ||
251 | /* | |
252 | * Ok, now set the unused entries to uncached. These entries | |
253 | * are officially undefined and no contract for the contents | |
254 | * and settings is given for these entries. | |
255 | * | |
256 | * Entry 0 in the table is uncached - so we are just writing | |
257 | * that value to all the used entries. | |
258 | */ | |
259 | for (; index < GEN9_NUM_MOCS_ENTRIES; index++) | |
260 | I915_WRITE(mocs_register(engine->id, index), | |
261 | table.table[0].control_value); | |
262 | ||
263 | return 0; | |
264 | } | |
265 | ||
3bbaba0c PA |
266 | /** |
267 | * emit_mocs_control_table() - emit the mocs control table | |
268 | * @req: Request to set up the MOCS table for. | |
269 | * @table: The values to program into the control regs. | |
3bbaba0c PA |
270 | * |
271 | * This function simply emits a MI_LOAD_REGISTER_IMM command for the | |
272 | * given table starting at the given address. | |
273 | * | |
274 | * Return: 0 on success, otherwise the error status. | |
275 | */ | |
276 | static int emit_mocs_control_table(struct drm_i915_gem_request *req, | |
0ccdacf6 | 277 | const struct drm_i915_mocs_table *table) |
3bbaba0c | 278 | { |
7e37f889 | 279 | struct intel_ring *ring = req->ring; |
0ccdacf6 | 280 | enum intel_engine_id engine = req->engine->id; |
3bbaba0c PA |
281 | unsigned int index; |
282 | int ret; | |
283 | ||
284 | if (WARN_ON(table->size > GEN9_NUM_MOCS_ENTRIES)) | |
285 | return -ENODEV; | |
286 | ||
987046ad CW |
287 | ret = intel_ring_begin(req, 2 + 2 * GEN9_NUM_MOCS_ENTRIES); |
288 | if (ret) | |
3bbaba0c | 289 | return ret; |
3bbaba0c | 290 | |
1dae2dfb | 291 | intel_ring_emit(ring, MI_LOAD_REGISTER_IMM(GEN9_NUM_MOCS_ENTRIES)); |
3bbaba0c PA |
292 | |
293 | for (index = 0; index < table->size; index++) { | |
1dae2dfb CW |
294 | intel_ring_emit_reg(ring, mocs_register(engine, index)); |
295 | intel_ring_emit(ring, table->table[index].control_value); | |
3bbaba0c PA |
296 | } |
297 | ||
298 | /* | |
299 | * Ok, now set the unused entries to uncached. These entries | |
300 | * are officially undefined and no contract for the contents | |
301 | * and settings is given for these entries. | |
302 | * | |
303 | * Entry 0 in the table is uncached - so we are just writing | |
304 | * that value to all the used entries. | |
305 | */ | |
306 | for (; index < GEN9_NUM_MOCS_ENTRIES; index++) { | |
1dae2dfb CW |
307 | intel_ring_emit_reg(ring, mocs_register(engine, index)); |
308 | intel_ring_emit(ring, table->table[0].control_value); | |
3bbaba0c PA |
309 | } |
310 | ||
1dae2dfb CW |
311 | intel_ring_emit(ring, MI_NOOP); |
312 | intel_ring_advance(ring); | |
3bbaba0c PA |
313 | |
314 | return 0; | |
315 | } | |
316 | ||
0ccdacf6 PA |
317 | static inline u32 l3cc_combine(const struct drm_i915_mocs_table *table, |
318 | u16 low, | |
319 | u16 high) | |
320 | { | |
321 | return table->table[low].l3cc_value | | |
322 | table->table[high].l3cc_value << 16; | |
323 | } | |
324 | ||
3bbaba0c PA |
325 | /** |
326 | * emit_mocs_l3cc_table() - emit the mocs control table | |
327 | * @req: Request to set up the MOCS table for. | |
328 | * @table: The values to program into the control regs. | |
329 | * | |
330 | * This function simply emits a MI_LOAD_REGISTER_IMM command for the | |
331 | * given table starting at the given address. This register set is | |
332 | * programmed in pairs. | |
333 | * | |
334 | * Return: 0 on success, otherwise the error status. | |
335 | */ | |
336 | static int emit_mocs_l3cc_table(struct drm_i915_gem_request *req, | |
337 | const struct drm_i915_mocs_table *table) | |
338 | { | |
7e37f889 | 339 | struct intel_ring *ring = req->ring; |
3bbaba0c | 340 | unsigned int i; |
3bbaba0c PA |
341 | int ret; |
342 | ||
343 | if (WARN_ON(table->size > GEN9_NUM_MOCS_ENTRIES)) | |
344 | return -ENODEV; | |
345 | ||
987046ad CW |
346 | ret = intel_ring_begin(req, 2 + GEN9_NUM_MOCS_ENTRIES); |
347 | if (ret) | |
3bbaba0c | 348 | return ret; |
3bbaba0c | 349 | |
1dae2dfb | 350 | intel_ring_emit(ring, |
3bbaba0c PA |
351 | MI_LOAD_REGISTER_IMM(GEN9_NUM_MOCS_ENTRIES / 2)); |
352 | ||
0ccdacf6 | 353 | for (i = 0; i < table->size/2; i++) { |
1dae2dfb CW |
354 | intel_ring_emit_reg(ring, GEN9_LNCFCMOCS(i)); |
355 | intel_ring_emit(ring, l3cc_combine(table, 2*i, 2*i+1)); | |
3bbaba0c PA |
356 | } |
357 | ||
358 | if (table->size & 0x01) { | |
359 | /* Odd table size - 1 left over */ | |
1dae2dfb CW |
360 | intel_ring_emit_reg(ring, GEN9_LNCFCMOCS(i)); |
361 | intel_ring_emit(ring, l3cc_combine(table, 2*i, 0)); | |
0ccdacf6 PA |
362 | i++; |
363 | } | |
3bbaba0c PA |
364 | |
365 | /* | |
366 | * Now set the rest of the table to uncached - use entry 0 as | |
367 | * this will be uncached. Leave the last pair uninitialised as | |
368 | * they are reserved by the hardware. | |
369 | */ | |
370 | for (; i < GEN9_NUM_MOCS_ENTRIES / 2; i++) { | |
1dae2dfb CW |
371 | intel_ring_emit_reg(ring, GEN9_LNCFCMOCS(i)); |
372 | intel_ring_emit(ring, l3cc_combine(table, 0, 0)); | |
3bbaba0c PA |
373 | } |
374 | ||
1dae2dfb CW |
375 | intel_ring_emit(ring, MI_NOOP); |
376 | intel_ring_advance(ring); | |
3bbaba0c PA |
377 | |
378 | return 0; | |
379 | } | |
380 | ||
0ccdacf6 PA |
381 | /** |
382 | * intel_mocs_init_l3cc_table() - program the mocs control table | |
383 | * @dev: The the device to be programmed. | |
384 | * | |
385 | * This function simply programs the mocs registers for the given table | |
386 | * starting at the given address. This register set is programmed in pairs. | |
387 | * | |
388 | * These registers may get programmed more than once, it is simpler to | |
389 | * re-program 32 registers than maintain the state of when they were programmed. | |
390 | * We are always reprogramming with the same values and this only on context | |
391 | * start. | |
392 | * | |
393 | * Return: Nothing. | |
394 | */ | |
395 | void intel_mocs_init_l3cc_table(struct drm_device *dev) | |
396 | { | |
397 | struct drm_i915_private *dev_priv = to_i915(dev); | |
398 | struct drm_i915_mocs_table table; | |
399 | unsigned int i; | |
400 | ||
401 | if (!get_mocs_settings(dev_priv, &table)) | |
402 | return; | |
403 | ||
404 | for (i = 0; i < table.size/2; i++) | |
405 | I915_WRITE(GEN9_LNCFCMOCS(i), l3cc_combine(&table, 2*i, 2*i+1)); | |
406 | ||
407 | /* Odd table size - 1 left over */ | |
408 | if (table.size & 0x01) { | |
409 | I915_WRITE(GEN9_LNCFCMOCS(i), l3cc_combine(&table, 2*i, 0)); | |
410 | i++; | |
411 | } | |
412 | ||
413 | /* | |
414 | * Now set the rest of the table to uncached - use entry 0 as | |
415 | * this will be uncached. Leave the last pair as initialised as | |
416 | * they are reserved by the hardware. | |
417 | */ | |
418 | for (; i < (GEN9_NUM_MOCS_ENTRIES / 2); i++) | |
419 | I915_WRITE(GEN9_LNCFCMOCS(i), l3cc_combine(&table, 0, 0)); | |
420 | } | |
421 | ||
3bbaba0c PA |
422 | /** |
423 | * intel_rcs_context_init_mocs() - program the MOCS register. | |
424 | * @req: Request to set up the MOCS tables for. | |
425 | * | |
426 | * This function will emit a batch buffer with the values required for | |
427 | * programming the MOCS register values for all the currently supported | |
428 | * rings. | |
429 | * | |
430 | * These registers are partially stored in the RCS context, so they are | |
431 | * emitted at the same time so that when a context is created these registers | |
432 | * are set up. These registers have to be emitted into the start of the | |
433 | * context as setting the ELSP will re-init some of these registers back | |
434 | * to the hw values. | |
435 | * | |
436 | * Return: 0 on success, otherwise the error status. | |
437 | */ | |
438 | int intel_rcs_context_init_mocs(struct drm_i915_gem_request *req) | |
439 | { | |
440 | struct drm_i915_mocs_table t; | |
441 | int ret; | |
442 | ||
0ccdacf6 PA |
443 | if (get_mocs_settings(req->i915, &t)) { |
444 | /* Program the RCS control registers */ | |
445 | ret = emit_mocs_control_table(req, &t); | |
446 | if (ret) | |
447 | return ret; | |
3bbaba0c PA |
448 | |
449 | /* Now program the l3cc registers */ | |
450 | ret = emit_mocs_l3cc_table(req, &t); | |
451 | if (ret) | |
452 | return ret; | |
453 | } | |
454 | ||
455 | return 0; | |
456 | } |