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 */
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
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[] = {
100 [I915_MOCS_UNCACHED] = {
101 /* 0x00000009 */
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 },
110 [I915_MOCS_PTE] = {
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 },
119 [I915_MOCS_CACHED] = {
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 },
128 };
129
130 /* NOTE: the LE_TGT_CACHE is not used on Broxton */
131 static const struct drm_i915_mocs_entry broxton_mocs_table[] = {
132 [I915_MOCS_UNCACHED] = {
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 },
142 [I915_MOCS_PTE] = {
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 },
152 [I915_MOCS_CACHED] = {
153 /* 0x00000039 */
154 .control_value = LE_CACHEABILITY(LE_UC) |
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 },
162 };
163
164 /**
165 * get_mocs_settings()
166 * @dev_priv: i915 device.
167 * @table: Output table that will be made to point at appropriate
168 * MOCS values for the device.
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 */
get_mocs_settings(struct drm_i915_private * dev_priv,struct drm_i915_mocs_table * table)176 static bool get_mocs_settings(struct drm_i915_private *dev_priv,
177 struct drm_i915_mocs_table *table)
178 {
179 bool result = false;
180
181 if (IS_GEN9_BC(dev_priv) || IS_CANNONLAKE(dev_priv)) {
182 table->size = ARRAY_SIZE(skylake_mocs_table);
183 table->table = skylake_mocs_table;
184 result = true;
185 } else if (IS_GEN9_LP(dev_priv)) {
186 table->size = ARRAY_SIZE(broxton_mocs_table);
187 table->table = broxton_mocs_table;
188 result = true;
189 } else {
190 WARN_ONCE(INTEL_INFO(dev_priv)->gen >= 9,
191 "Platform that should have a MOCS table does not.\n");
192 }
193
194 /* WaDisableSkipCaching:skl,bxt,kbl,glk */
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 &
200 (L3_ESC(1) | L3_SCC(0x7))))
201 return false;
202 }
203
204 return result;
205 }
206
mocs_register(enum intel_engine_id engine_id,int index)207 static i915_reg_t mocs_register(enum intel_engine_id engine_id, int index)
208 {
209 switch (engine_id) {
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:
221 MISSING_CASE(engine_id);
222 return INVALID_MMIO_REG;
223 }
224 }
225
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 */
intel_mocs_init_engine(struct intel_engine_cs * engine)235 int intel_mocs_init_engine(struct intel_engine_cs *engine)
236 {
237 struct drm_i915_private *dev_priv = engine->i915;
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
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.
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 */
emit_mocs_control_table(struct drm_i915_gem_request * req,const struct drm_i915_mocs_table * table)276 static int emit_mocs_control_table(struct drm_i915_gem_request *req,
277 const struct drm_i915_mocs_table *table)
278 {
279 enum intel_engine_id engine = req->engine->id;
280 unsigned int index;
281 u32 *cs;
282
283 if (WARN_ON(table->size > GEN9_NUM_MOCS_ENTRIES))
284 return -ENODEV;
285
286 cs = intel_ring_begin(req, 2 + 2 * GEN9_NUM_MOCS_ENTRIES);
287 if (IS_ERR(cs))
288 return PTR_ERR(cs);
289
290 *cs++ = MI_LOAD_REGISTER_IMM(GEN9_NUM_MOCS_ENTRIES);
291
292 for (index = 0; index < table->size; index++) {
293 *cs++ = i915_mmio_reg_offset(mocs_register(engine, index));
294 *cs++ = table->table[index].control_value;
295 }
296
297 /*
298 * Ok, now set the unused entries to uncached. These entries
299 * are officially undefined and no contract for the contents
300 * and settings is given for these entries.
301 *
302 * Entry 0 in the table is uncached - so we are just writing
303 * that value to all the used entries.
304 */
305 for (; index < GEN9_NUM_MOCS_ENTRIES; index++) {
306 *cs++ = i915_mmio_reg_offset(mocs_register(engine, index));
307 *cs++ = table->table[0].control_value;
308 }
309
310 *cs++ = MI_NOOP;
311 intel_ring_advance(req, cs);
312
313 return 0;
314 }
315
l3cc_combine(const struct drm_i915_mocs_table * table,u16 low,u16 high)316 static inline u32 l3cc_combine(const struct drm_i915_mocs_table *table,
317 u16 low,
318 u16 high)
319 {
320 return table->table[low].l3cc_value |
321 table->table[high].l3cc_value << 16;
322 }
323
324 /**
325 * emit_mocs_l3cc_table() - emit the mocs control table
326 * @req: Request to set up the MOCS table for.
327 * @table: The values to program into the control regs.
328 *
329 * This function simply emits a MI_LOAD_REGISTER_IMM command for the
330 * given table starting at the given address. This register set is
331 * programmed in pairs.
332 *
333 * Return: 0 on success, otherwise the error status.
334 */
emit_mocs_l3cc_table(struct drm_i915_gem_request * req,const struct drm_i915_mocs_table * table)335 static int emit_mocs_l3cc_table(struct drm_i915_gem_request *req,
336 const struct drm_i915_mocs_table *table)
337 {
338 unsigned int i;
339 u32 *cs;
340
341 if (WARN_ON(table->size > GEN9_NUM_MOCS_ENTRIES))
342 return -ENODEV;
343
344 cs = intel_ring_begin(req, 2 + GEN9_NUM_MOCS_ENTRIES);
345 if (IS_ERR(cs))
346 return PTR_ERR(cs);
347
348 *cs++ = MI_LOAD_REGISTER_IMM(GEN9_NUM_MOCS_ENTRIES / 2);
349
350 for (i = 0; i < table->size/2; i++) {
351 *cs++ = i915_mmio_reg_offset(GEN9_LNCFCMOCS(i));
352 *cs++ = l3cc_combine(table, 2 * i, 2 * i + 1);
353 }
354
355 if (table->size & 0x01) {
356 /* Odd table size - 1 left over */
357 *cs++ = i915_mmio_reg_offset(GEN9_LNCFCMOCS(i));
358 *cs++ = l3cc_combine(table, 2 * i, 0);
359 i++;
360 }
361
362 /*
363 * Now set the rest of the table to uncached - use entry 0 as
364 * this will be uncached. Leave the last pair uninitialised as
365 * they are reserved by the hardware.
366 */
367 for (; i < GEN9_NUM_MOCS_ENTRIES / 2; i++) {
368 *cs++ = i915_mmio_reg_offset(GEN9_LNCFCMOCS(i));
369 *cs++ = l3cc_combine(table, 0, 0);
370 }
371
372 *cs++ = MI_NOOP;
373 intel_ring_advance(req, cs);
374
375 return 0;
376 }
377
378 /**
379 * intel_mocs_init_l3cc_table() - program the mocs control table
380 * @dev_priv: i915 device private
381 *
382 * This function simply programs the mocs registers for the given table
383 * starting at the given address. This register set is programmed in pairs.
384 *
385 * These registers may get programmed more than once, it is simpler to
386 * re-program 32 registers than maintain the state of when they were programmed.
387 * We are always reprogramming with the same values and this only on context
388 * start.
389 *
390 * Return: Nothing.
391 */
intel_mocs_init_l3cc_table(struct drm_i915_private * dev_priv)392 void intel_mocs_init_l3cc_table(struct drm_i915_private *dev_priv)
393 {
394 struct drm_i915_mocs_table table;
395 unsigned int i;
396
397 if (!get_mocs_settings(dev_priv, &table))
398 return;
399
400 for (i = 0; i < table.size/2; i++)
401 I915_WRITE(GEN9_LNCFCMOCS(i), l3cc_combine(&table, 2*i, 2*i+1));
402
403 /* Odd table size - 1 left over */
404 if (table.size & 0x01) {
405 I915_WRITE(GEN9_LNCFCMOCS(i), l3cc_combine(&table, 2*i, 0));
406 i++;
407 }
408
409 /*
410 * Now set the rest of the table to uncached - use entry 0 as
411 * this will be uncached. Leave the last pair as initialised as
412 * they are reserved by the hardware.
413 */
414 for (; i < (GEN9_NUM_MOCS_ENTRIES / 2); i++)
415 I915_WRITE(GEN9_LNCFCMOCS(i), l3cc_combine(&table, 0, 0));
416 }
417
418 /**
419 * intel_rcs_context_init_mocs() - program the MOCS register.
420 * @req: Request to set up the MOCS tables for.
421 *
422 * This function will emit a batch buffer with the values required for
423 * programming the MOCS register values for all the currently supported
424 * rings.
425 *
426 * These registers are partially stored in the RCS context, so they are
427 * emitted at the same time so that when a context is created these registers
428 * are set up. These registers have to be emitted into the start of the
429 * context as setting the ELSP will re-init some of these registers back
430 * to the hw values.
431 *
432 * Return: 0 on success, otherwise the error status.
433 */
intel_rcs_context_init_mocs(struct drm_i915_gem_request * req)434 int intel_rcs_context_init_mocs(struct drm_i915_gem_request *req)
435 {
436 struct drm_i915_mocs_table t;
437 int ret;
438
439 if (get_mocs_settings(req->i915, &t)) {
440 /* Program the RCS control registers */
441 ret = emit_mocs_control_table(req, &t);
442 if (ret)
443 return ret;
444
445 /* Now program the l3cc registers */
446 ret = emit_mocs_l3cc_table(req, &t);
447 if (ret)
448 return ret;
449 }
450
451 return 0;
452 }
453