drm/radeon/dpm: implement UVD powergating for KB/KV
[deliverable/linux.git] / drivers / gpu / drm / radeon / cik.c
1 /*
2 * Copyright 2012 Advanced Micro Devices, Inc.
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 shall be included in
12 * all copies or substantial portions of the 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 COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20 * OTHER DEALINGS IN THE SOFTWARE.
21 *
22 * Authors: Alex Deucher
23 */
24 #include <linux/firmware.h>
25 #include <linux/slab.h>
26 #include <linux/module.h>
27 #include "drmP.h"
28 #include "radeon.h"
29 #include "radeon_asic.h"
30 #include "cikd.h"
31 #include "atom.h"
32 #include "cik_blit_shaders.h"
33 #include "radeon_ucode.h"
34 #include "clearstate_ci.h"
35
36 MODULE_FIRMWARE("radeon/BONAIRE_pfp.bin");
37 MODULE_FIRMWARE("radeon/BONAIRE_me.bin");
38 MODULE_FIRMWARE("radeon/BONAIRE_ce.bin");
39 MODULE_FIRMWARE("radeon/BONAIRE_mec.bin");
40 MODULE_FIRMWARE("radeon/BONAIRE_mc.bin");
41 MODULE_FIRMWARE("radeon/BONAIRE_rlc.bin");
42 MODULE_FIRMWARE("radeon/BONAIRE_sdma.bin");
43 MODULE_FIRMWARE("radeon/BONAIRE_smc.bin");
44 MODULE_FIRMWARE("radeon/KAVERI_pfp.bin");
45 MODULE_FIRMWARE("radeon/KAVERI_me.bin");
46 MODULE_FIRMWARE("radeon/KAVERI_ce.bin");
47 MODULE_FIRMWARE("radeon/KAVERI_mec.bin");
48 MODULE_FIRMWARE("radeon/KAVERI_rlc.bin");
49 MODULE_FIRMWARE("radeon/KAVERI_sdma.bin");
50 MODULE_FIRMWARE("radeon/KABINI_pfp.bin");
51 MODULE_FIRMWARE("radeon/KABINI_me.bin");
52 MODULE_FIRMWARE("radeon/KABINI_ce.bin");
53 MODULE_FIRMWARE("radeon/KABINI_mec.bin");
54 MODULE_FIRMWARE("radeon/KABINI_rlc.bin");
55 MODULE_FIRMWARE("radeon/KABINI_sdma.bin");
56
57 extern int r600_ih_ring_alloc(struct radeon_device *rdev);
58 extern void r600_ih_ring_fini(struct radeon_device *rdev);
59 extern void evergreen_mc_stop(struct radeon_device *rdev, struct evergreen_mc_save *save);
60 extern void evergreen_mc_resume(struct radeon_device *rdev, struct evergreen_mc_save *save);
61 extern bool evergreen_is_display_hung(struct radeon_device *rdev);
62 extern void sumo_rlc_fini(struct radeon_device *rdev);
63 extern int sumo_rlc_init(struct radeon_device *rdev);
64 extern void si_vram_gtt_location(struct radeon_device *rdev, struct radeon_mc *mc);
65 extern void si_rlc_reset(struct radeon_device *rdev);
66 extern void si_init_uvd_internal_cg(struct radeon_device *rdev);
67 static void cik_rlc_stop(struct radeon_device *rdev);
68 static void cik_pcie_gen3_enable(struct radeon_device *rdev);
69 static void cik_program_aspm(struct radeon_device *rdev);
70 static void cik_init_pg(struct radeon_device *rdev);
71 static void cik_init_cg(struct radeon_device *rdev);
72 void cik_uvd_resume(struct radeon_device *rdev);
73
74 /* get temperature in millidegrees */
75 int ci_get_temp(struct radeon_device *rdev)
76 {
77 u32 temp;
78 int actual_temp = 0;
79
80 temp = (RREG32_SMC(CG_MULT_THERMAL_STATUS) & CTF_TEMP_MASK) >>
81 CTF_TEMP_SHIFT;
82
83 if (temp & 0x200)
84 actual_temp = 255;
85 else
86 actual_temp = temp & 0x1ff;
87
88 actual_temp = actual_temp * 1000;
89
90 return actual_temp;
91 }
92
93 /* get temperature in millidegrees */
94 int kv_get_temp(struct radeon_device *rdev)
95 {
96 u32 temp;
97 int actual_temp = 0;
98
99 temp = RREG32_SMC(0xC0300E0C);
100
101 if (temp)
102 actual_temp = (temp / 8) - 49;
103 else
104 actual_temp = 0;
105
106 actual_temp = actual_temp * 1000;
107
108 return actual_temp;
109 }
110
111 /*
112 * Indirect registers accessor
113 */
114 u32 cik_pciep_rreg(struct radeon_device *rdev, u32 reg)
115 {
116 u32 r;
117
118 WREG32(PCIE_INDEX, reg);
119 (void)RREG32(PCIE_INDEX);
120 r = RREG32(PCIE_DATA);
121 return r;
122 }
123
124 void cik_pciep_wreg(struct radeon_device *rdev, u32 reg, u32 v)
125 {
126 WREG32(PCIE_INDEX, reg);
127 (void)RREG32(PCIE_INDEX);
128 WREG32(PCIE_DATA, v);
129 (void)RREG32(PCIE_DATA);
130 }
131
132 static const u32 spectre_rlc_save_restore_register_list[] =
133 {
134 (0x0e00 << 16) | (0xc12c >> 2),
135 0x00000000,
136 (0x0e00 << 16) | (0xc140 >> 2),
137 0x00000000,
138 (0x0e00 << 16) | (0xc150 >> 2),
139 0x00000000,
140 (0x0e00 << 16) | (0xc15c >> 2),
141 0x00000000,
142 (0x0e00 << 16) | (0xc168 >> 2),
143 0x00000000,
144 (0x0e00 << 16) | (0xc170 >> 2),
145 0x00000000,
146 (0x0e00 << 16) | (0xc178 >> 2),
147 0x00000000,
148 (0x0e00 << 16) | (0xc204 >> 2),
149 0x00000000,
150 (0x0e00 << 16) | (0xc2b4 >> 2),
151 0x00000000,
152 (0x0e00 << 16) | (0xc2b8 >> 2),
153 0x00000000,
154 (0x0e00 << 16) | (0xc2bc >> 2),
155 0x00000000,
156 (0x0e00 << 16) | (0xc2c0 >> 2),
157 0x00000000,
158 (0x0e00 << 16) | (0x8228 >> 2),
159 0x00000000,
160 (0x0e00 << 16) | (0x829c >> 2),
161 0x00000000,
162 (0x0e00 << 16) | (0x869c >> 2),
163 0x00000000,
164 (0x0600 << 16) | (0x98f4 >> 2),
165 0x00000000,
166 (0x0e00 << 16) | (0x98f8 >> 2),
167 0x00000000,
168 (0x0e00 << 16) | (0x9900 >> 2),
169 0x00000000,
170 (0x0e00 << 16) | (0xc260 >> 2),
171 0x00000000,
172 (0x0e00 << 16) | (0x90e8 >> 2),
173 0x00000000,
174 (0x0e00 << 16) | (0x3c000 >> 2),
175 0x00000000,
176 (0x0e00 << 16) | (0x3c00c >> 2),
177 0x00000000,
178 (0x0e00 << 16) | (0x8c1c >> 2),
179 0x00000000,
180 (0x0e00 << 16) | (0x9700 >> 2),
181 0x00000000,
182 (0x0e00 << 16) | (0xcd20 >> 2),
183 0x00000000,
184 (0x4e00 << 16) | (0xcd20 >> 2),
185 0x00000000,
186 (0x5e00 << 16) | (0xcd20 >> 2),
187 0x00000000,
188 (0x6e00 << 16) | (0xcd20 >> 2),
189 0x00000000,
190 (0x7e00 << 16) | (0xcd20 >> 2),
191 0x00000000,
192 (0x8e00 << 16) | (0xcd20 >> 2),
193 0x00000000,
194 (0x9e00 << 16) | (0xcd20 >> 2),
195 0x00000000,
196 (0xae00 << 16) | (0xcd20 >> 2),
197 0x00000000,
198 (0xbe00 << 16) | (0xcd20 >> 2),
199 0x00000000,
200 (0x0e00 << 16) | (0x89bc >> 2),
201 0x00000000,
202 (0x0e00 << 16) | (0x8900 >> 2),
203 0x00000000,
204 0x3,
205 (0x0e00 << 16) | (0xc130 >> 2),
206 0x00000000,
207 (0x0e00 << 16) | (0xc134 >> 2),
208 0x00000000,
209 (0x0e00 << 16) | (0xc1fc >> 2),
210 0x00000000,
211 (0x0e00 << 16) | (0xc208 >> 2),
212 0x00000000,
213 (0x0e00 << 16) | (0xc264 >> 2),
214 0x00000000,
215 (0x0e00 << 16) | (0xc268 >> 2),
216 0x00000000,
217 (0x0e00 << 16) | (0xc26c >> 2),
218 0x00000000,
219 (0x0e00 << 16) | (0xc270 >> 2),
220 0x00000000,
221 (0x0e00 << 16) | (0xc274 >> 2),
222 0x00000000,
223 (0x0e00 << 16) | (0xc278 >> 2),
224 0x00000000,
225 (0x0e00 << 16) | (0xc27c >> 2),
226 0x00000000,
227 (0x0e00 << 16) | (0xc280 >> 2),
228 0x00000000,
229 (0x0e00 << 16) | (0xc284 >> 2),
230 0x00000000,
231 (0x0e00 << 16) | (0xc288 >> 2),
232 0x00000000,
233 (0x0e00 << 16) | (0xc28c >> 2),
234 0x00000000,
235 (0x0e00 << 16) | (0xc290 >> 2),
236 0x00000000,
237 (0x0e00 << 16) | (0xc294 >> 2),
238 0x00000000,
239 (0x0e00 << 16) | (0xc298 >> 2),
240 0x00000000,
241 (0x0e00 << 16) | (0xc29c >> 2),
242 0x00000000,
243 (0x0e00 << 16) | (0xc2a0 >> 2),
244 0x00000000,
245 (0x0e00 << 16) | (0xc2a4 >> 2),
246 0x00000000,
247 (0x0e00 << 16) | (0xc2a8 >> 2),
248 0x00000000,
249 (0x0e00 << 16) | (0xc2ac >> 2),
250 0x00000000,
251 (0x0e00 << 16) | (0xc2b0 >> 2),
252 0x00000000,
253 (0x0e00 << 16) | (0x301d0 >> 2),
254 0x00000000,
255 (0x0e00 << 16) | (0x30238 >> 2),
256 0x00000000,
257 (0x0e00 << 16) | (0x30250 >> 2),
258 0x00000000,
259 (0x0e00 << 16) | (0x30254 >> 2),
260 0x00000000,
261 (0x0e00 << 16) | (0x30258 >> 2),
262 0x00000000,
263 (0x0e00 << 16) | (0x3025c >> 2),
264 0x00000000,
265 (0x4e00 << 16) | (0xc900 >> 2),
266 0x00000000,
267 (0x5e00 << 16) | (0xc900 >> 2),
268 0x00000000,
269 (0x6e00 << 16) | (0xc900 >> 2),
270 0x00000000,
271 (0x7e00 << 16) | (0xc900 >> 2),
272 0x00000000,
273 (0x8e00 << 16) | (0xc900 >> 2),
274 0x00000000,
275 (0x9e00 << 16) | (0xc900 >> 2),
276 0x00000000,
277 (0xae00 << 16) | (0xc900 >> 2),
278 0x00000000,
279 (0xbe00 << 16) | (0xc900 >> 2),
280 0x00000000,
281 (0x4e00 << 16) | (0xc904 >> 2),
282 0x00000000,
283 (0x5e00 << 16) | (0xc904 >> 2),
284 0x00000000,
285 (0x6e00 << 16) | (0xc904 >> 2),
286 0x00000000,
287 (0x7e00 << 16) | (0xc904 >> 2),
288 0x00000000,
289 (0x8e00 << 16) | (0xc904 >> 2),
290 0x00000000,
291 (0x9e00 << 16) | (0xc904 >> 2),
292 0x00000000,
293 (0xae00 << 16) | (0xc904 >> 2),
294 0x00000000,
295 (0xbe00 << 16) | (0xc904 >> 2),
296 0x00000000,
297 (0x4e00 << 16) | (0xc908 >> 2),
298 0x00000000,
299 (0x5e00 << 16) | (0xc908 >> 2),
300 0x00000000,
301 (0x6e00 << 16) | (0xc908 >> 2),
302 0x00000000,
303 (0x7e00 << 16) | (0xc908 >> 2),
304 0x00000000,
305 (0x8e00 << 16) | (0xc908 >> 2),
306 0x00000000,
307 (0x9e00 << 16) | (0xc908 >> 2),
308 0x00000000,
309 (0xae00 << 16) | (0xc908 >> 2),
310 0x00000000,
311 (0xbe00 << 16) | (0xc908 >> 2),
312 0x00000000,
313 (0x4e00 << 16) | (0xc90c >> 2),
314 0x00000000,
315 (0x5e00 << 16) | (0xc90c >> 2),
316 0x00000000,
317 (0x6e00 << 16) | (0xc90c >> 2),
318 0x00000000,
319 (0x7e00 << 16) | (0xc90c >> 2),
320 0x00000000,
321 (0x8e00 << 16) | (0xc90c >> 2),
322 0x00000000,
323 (0x9e00 << 16) | (0xc90c >> 2),
324 0x00000000,
325 (0xae00 << 16) | (0xc90c >> 2),
326 0x00000000,
327 (0xbe00 << 16) | (0xc90c >> 2),
328 0x00000000,
329 (0x4e00 << 16) | (0xc910 >> 2),
330 0x00000000,
331 (0x5e00 << 16) | (0xc910 >> 2),
332 0x00000000,
333 (0x6e00 << 16) | (0xc910 >> 2),
334 0x00000000,
335 (0x7e00 << 16) | (0xc910 >> 2),
336 0x00000000,
337 (0x8e00 << 16) | (0xc910 >> 2),
338 0x00000000,
339 (0x9e00 << 16) | (0xc910 >> 2),
340 0x00000000,
341 (0xae00 << 16) | (0xc910 >> 2),
342 0x00000000,
343 (0xbe00 << 16) | (0xc910 >> 2),
344 0x00000000,
345 (0x0e00 << 16) | (0xc99c >> 2),
346 0x00000000,
347 (0x0e00 << 16) | (0x9834 >> 2),
348 0x00000000,
349 (0x0000 << 16) | (0x30f00 >> 2),
350 0x00000000,
351 (0x0001 << 16) | (0x30f00 >> 2),
352 0x00000000,
353 (0x0000 << 16) | (0x30f04 >> 2),
354 0x00000000,
355 (0x0001 << 16) | (0x30f04 >> 2),
356 0x00000000,
357 (0x0000 << 16) | (0x30f08 >> 2),
358 0x00000000,
359 (0x0001 << 16) | (0x30f08 >> 2),
360 0x00000000,
361 (0x0000 << 16) | (0x30f0c >> 2),
362 0x00000000,
363 (0x0001 << 16) | (0x30f0c >> 2),
364 0x00000000,
365 (0x0600 << 16) | (0x9b7c >> 2),
366 0x00000000,
367 (0x0e00 << 16) | (0x8a14 >> 2),
368 0x00000000,
369 (0x0e00 << 16) | (0x8a18 >> 2),
370 0x00000000,
371 (0x0600 << 16) | (0x30a00 >> 2),
372 0x00000000,
373 (0x0e00 << 16) | (0x8bf0 >> 2),
374 0x00000000,
375 (0x0e00 << 16) | (0x8bcc >> 2),
376 0x00000000,
377 (0x0e00 << 16) | (0x8b24 >> 2),
378 0x00000000,
379 (0x0e00 << 16) | (0x30a04 >> 2),
380 0x00000000,
381 (0x0600 << 16) | (0x30a10 >> 2),
382 0x00000000,
383 (0x0600 << 16) | (0x30a14 >> 2),
384 0x00000000,
385 (0x0600 << 16) | (0x30a18 >> 2),
386 0x00000000,
387 (0x0600 << 16) | (0x30a2c >> 2),
388 0x00000000,
389 (0x0e00 << 16) | (0xc700 >> 2),
390 0x00000000,
391 (0x0e00 << 16) | (0xc704 >> 2),
392 0x00000000,
393 (0x0e00 << 16) | (0xc708 >> 2),
394 0x00000000,
395 (0x0e00 << 16) | (0xc768 >> 2),
396 0x00000000,
397 (0x0400 << 16) | (0xc770 >> 2),
398 0x00000000,
399 (0x0400 << 16) | (0xc774 >> 2),
400 0x00000000,
401 (0x0400 << 16) | (0xc778 >> 2),
402 0x00000000,
403 (0x0400 << 16) | (0xc77c >> 2),
404 0x00000000,
405 (0x0400 << 16) | (0xc780 >> 2),
406 0x00000000,
407 (0x0400 << 16) | (0xc784 >> 2),
408 0x00000000,
409 (0x0400 << 16) | (0xc788 >> 2),
410 0x00000000,
411 (0x0400 << 16) | (0xc78c >> 2),
412 0x00000000,
413 (0x0400 << 16) | (0xc798 >> 2),
414 0x00000000,
415 (0x0400 << 16) | (0xc79c >> 2),
416 0x00000000,
417 (0x0400 << 16) | (0xc7a0 >> 2),
418 0x00000000,
419 (0x0400 << 16) | (0xc7a4 >> 2),
420 0x00000000,
421 (0x0400 << 16) | (0xc7a8 >> 2),
422 0x00000000,
423 (0x0400 << 16) | (0xc7ac >> 2),
424 0x00000000,
425 (0x0400 << 16) | (0xc7b0 >> 2),
426 0x00000000,
427 (0x0400 << 16) | (0xc7b4 >> 2),
428 0x00000000,
429 (0x0e00 << 16) | (0x9100 >> 2),
430 0x00000000,
431 (0x0e00 << 16) | (0x3c010 >> 2),
432 0x00000000,
433 (0x0e00 << 16) | (0x92a8 >> 2),
434 0x00000000,
435 (0x0e00 << 16) | (0x92ac >> 2),
436 0x00000000,
437 (0x0e00 << 16) | (0x92b4 >> 2),
438 0x00000000,
439 (0x0e00 << 16) | (0x92b8 >> 2),
440 0x00000000,
441 (0x0e00 << 16) | (0x92bc >> 2),
442 0x00000000,
443 (0x0e00 << 16) | (0x92c0 >> 2),
444 0x00000000,
445 (0x0e00 << 16) | (0x92c4 >> 2),
446 0x00000000,
447 (0x0e00 << 16) | (0x92c8 >> 2),
448 0x00000000,
449 (0x0e00 << 16) | (0x92cc >> 2),
450 0x00000000,
451 (0x0e00 << 16) | (0x92d0 >> 2),
452 0x00000000,
453 (0x0e00 << 16) | (0x8c00 >> 2),
454 0x00000000,
455 (0x0e00 << 16) | (0x8c04 >> 2),
456 0x00000000,
457 (0x0e00 << 16) | (0x8c20 >> 2),
458 0x00000000,
459 (0x0e00 << 16) | (0x8c38 >> 2),
460 0x00000000,
461 (0x0e00 << 16) | (0x8c3c >> 2),
462 0x00000000,
463 (0x0e00 << 16) | (0xae00 >> 2),
464 0x00000000,
465 (0x0e00 << 16) | (0x9604 >> 2),
466 0x00000000,
467 (0x0e00 << 16) | (0xac08 >> 2),
468 0x00000000,
469 (0x0e00 << 16) | (0xac0c >> 2),
470 0x00000000,
471 (0x0e00 << 16) | (0xac10 >> 2),
472 0x00000000,
473 (0x0e00 << 16) | (0xac14 >> 2),
474 0x00000000,
475 (0x0e00 << 16) | (0xac58 >> 2),
476 0x00000000,
477 (0x0e00 << 16) | (0xac68 >> 2),
478 0x00000000,
479 (0x0e00 << 16) | (0xac6c >> 2),
480 0x00000000,
481 (0x0e00 << 16) | (0xac70 >> 2),
482 0x00000000,
483 (0x0e00 << 16) | (0xac74 >> 2),
484 0x00000000,
485 (0x0e00 << 16) | (0xac78 >> 2),
486 0x00000000,
487 (0x0e00 << 16) | (0xac7c >> 2),
488 0x00000000,
489 (0x0e00 << 16) | (0xac80 >> 2),
490 0x00000000,
491 (0x0e00 << 16) | (0xac84 >> 2),
492 0x00000000,
493 (0x0e00 << 16) | (0xac88 >> 2),
494 0x00000000,
495 (0x0e00 << 16) | (0xac8c >> 2),
496 0x00000000,
497 (0x0e00 << 16) | (0x970c >> 2),
498 0x00000000,
499 (0x0e00 << 16) | (0x9714 >> 2),
500 0x00000000,
501 (0x0e00 << 16) | (0x9718 >> 2),
502 0x00000000,
503 (0x0e00 << 16) | (0x971c >> 2),
504 0x00000000,
505 (0x0e00 << 16) | (0x31068 >> 2),
506 0x00000000,
507 (0x4e00 << 16) | (0x31068 >> 2),
508 0x00000000,
509 (0x5e00 << 16) | (0x31068 >> 2),
510 0x00000000,
511 (0x6e00 << 16) | (0x31068 >> 2),
512 0x00000000,
513 (0x7e00 << 16) | (0x31068 >> 2),
514 0x00000000,
515 (0x8e00 << 16) | (0x31068 >> 2),
516 0x00000000,
517 (0x9e00 << 16) | (0x31068 >> 2),
518 0x00000000,
519 (0xae00 << 16) | (0x31068 >> 2),
520 0x00000000,
521 (0xbe00 << 16) | (0x31068 >> 2),
522 0x00000000,
523 (0x0e00 << 16) | (0xcd10 >> 2),
524 0x00000000,
525 (0x0e00 << 16) | (0xcd14 >> 2),
526 0x00000000,
527 (0x0e00 << 16) | (0x88b0 >> 2),
528 0x00000000,
529 (0x0e00 << 16) | (0x88b4 >> 2),
530 0x00000000,
531 (0x0e00 << 16) | (0x88b8 >> 2),
532 0x00000000,
533 (0x0e00 << 16) | (0x88bc >> 2),
534 0x00000000,
535 (0x0400 << 16) | (0x89c0 >> 2),
536 0x00000000,
537 (0x0e00 << 16) | (0x88c4 >> 2),
538 0x00000000,
539 (0x0e00 << 16) | (0x88c8 >> 2),
540 0x00000000,
541 (0x0e00 << 16) | (0x88d0 >> 2),
542 0x00000000,
543 (0x0e00 << 16) | (0x88d4 >> 2),
544 0x00000000,
545 (0x0e00 << 16) | (0x88d8 >> 2),
546 0x00000000,
547 (0x0e00 << 16) | (0x8980 >> 2),
548 0x00000000,
549 (0x0e00 << 16) | (0x30938 >> 2),
550 0x00000000,
551 (0x0e00 << 16) | (0x3093c >> 2),
552 0x00000000,
553 (0x0e00 << 16) | (0x30940 >> 2),
554 0x00000000,
555 (0x0e00 << 16) | (0x89a0 >> 2),
556 0x00000000,
557 (0x0e00 << 16) | (0x30900 >> 2),
558 0x00000000,
559 (0x0e00 << 16) | (0x30904 >> 2),
560 0x00000000,
561 (0x0e00 << 16) | (0x89b4 >> 2),
562 0x00000000,
563 (0x0e00 << 16) | (0x3c210 >> 2),
564 0x00000000,
565 (0x0e00 << 16) | (0x3c214 >> 2),
566 0x00000000,
567 (0x0e00 << 16) | (0x3c218 >> 2),
568 0x00000000,
569 (0x0e00 << 16) | (0x8904 >> 2),
570 0x00000000,
571 0x5,
572 (0x0e00 << 16) | (0x8c28 >> 2),
573 (0x0e00 << 16) | (0x8c2c >> 2),
574 (0x0e00 << 16) | (0x8c30 >> 2),
575 (0x0e00 << 16) | (0x8c34 >> 2),
576 (0x0e00 << 16) | (0x9600 >> 2),
577 };
578
579 static const u32 kalindi_rlc_save_restore_register_list[] =
580 {
581 (0x0e00 << 16) | (0xc12c >> 2),
582 0x00000000,
583 (0x0e00 << 16) | (0xc140 >> 2),
584 0x00000000,
585 (0x0e00 << 16) | (0xc150 >> 2),
586 0x00000000,
587 (0x0e00 << 16) | (0xc15c >> 2),
588 0x00000000,
589 (0x0e00 << 16) | (0xc168 >> 2),
590 0x00000000,
591 (0x0e00 << 16) | (0xc170 >> 2),
592 0x00000000,
593 (0x0e00 << 16) | (0xc204 >> 2),
594 0x00000000,
595 (0x0e00 << 16) | (0xc2b4 >> 2),
596 0x00000000,
597 (0x0e00 << 16) | (0xc2b8 >> 2),
598 0x00000000,
599 (0x0e00 << 16) | (0xc2bc >> 2),
600 0x00000000,
601 (0x0e00 << 16) | (0xc2c0 >> 2),
602 0x00000000,
603 (0x0e00 << 16) | (0x8228 >> 2),
604 0x00000000,
605 (0x0e00 << 16) | (0x829c >> 2),
606 0x00000000,
607 (0x0e00 << 16) | (0x869c >> 2),
608 0x00000000,
609 (0x0600 << 16) | (0x98f4 >> 2),
610 0x00000000,
611 (0x0e00 << 16) | (0x98f8 >> 2),
612 0x00000000,
613 (0x0e00 << 16) | (0x9900 >> 2),
614 0x00000000,
615 (0x0e00 << 16) | (0xc260 >> 2),
616 0x00000000,
617 (0x0e00 << 16) | (0x90e8 >> 2),
618 0x00000000,
619 (0x0e00 << 16) | (0x3c000 >> 2),
620 0x00000000,
621 (0x0e00 << 16) | (0x3c00c >> 2),
622 0x00000000,
623 (0x0e00 << 16) | (0x8c1c >> 2),
624 0x00000000,
625 (0x0e00 << 16) | (0x9700 >> 2),
626 0x00000000,
627 (0x0e00 << 16) | (0xcd20 >> 2),
628 0x00000000,
629 (0x4e00 << 16) | (0xcd20 >> 2),
630 0x00000000,
631 (0x5e00 << 16) | (0xcd20 >> 2),
632 0x00000000,
633 (0x6e00 << 16) | (0xcd20 >> 2),
634 0x00000000,
635 (0x7e00 << 16) | (0xcd20 >> 2),
636 0x00000000,
637 (0x0e00 << 16) | (0x89bc >> 2),
638 0x00000000,
639 (0x0e00 << 16) | (0x8900 >> 2),
640 0x00000000,
641 0x3,
642 (0x0e00 << 16) | (0xc130 >> 2),
643 0x00000000,
644 (0x0e00 << 16) | (0xc134 >> 2),
645 0x00000000,
646 (0x0e00 << 16) | (0xc1fc >> 2),
647 0x00000000,
648 (0x0e00 << 16) | (0xc208 >> 2),
649 0x00000000,
650 (0x0e00 << 16) | (0xc264 >> 2),
651 0x00000000,
652 (0x0e00 << 16) | (0xc268 >> 2),
653 0x00000000,
654 (0x0e00 << 16) | (0xc26c >> 2),
655 0x00000000,
656 (0x0e00 << 16) | (0xc270 >> 2),
657 0x00000000,
658 (0x0e00 << 16) | (0xc274 >> 2),
659 0x00000000,
660 (0x0e00 << 16) | (0xc28c >> 2),
661 0x00000000,
662 (0x0e00 << 16) | (0xc290 >> 2),
663 0x00000000,
664 (0x0e00 << 16) | (0xc294 >> 2),
665 0x00000000,
666 (0x0e00 << 16) | (0xc298 >> 2),
667 0x00000000,
668 (0x0e00 << 16) | (0xc2a0 >> 2),
669 0x00000000,
670 (0x0e00 << 16) | (0xc2a4 >> 2),
671 0x00000000,
672 (0x0e00 << 16) | (0xc2a8 >> 2),
673 0x00000000,
674 (0x0e00 << 16) | (0xc2ac >> 2),
675 0x00000000,
676 (0x0e00 << 16) | (0x301d0 >> 2),
677 0x00000000,
678 (0x0e00 << 16) | (0x30238 >> 2),
679 0x00000000,
680 (0x0e00 << 16) | (0x30250 >> 2),
681 0x00000000,
682 (0x0e00 << 16) | (0x30254 >> 2),
683 0x00000000,
684 (0x0e00 << 16) | (0x30258 >> 2),
685 0x00000000,
686 (0x0e00 << 16) | (0x3025c >> 2),
687 0x00000000,
688 (0x4e00 << 16) | (0xc900 >> 2),
689 0x00000000,
690 (0x5e00 << 16) | (0xc900 >> 2),
691 0x00000000,
692 (0x6e00 << 16) | (0xc900 >> 2),
693 0x00000000,
694 (0x7e00 << 16) | (0xc900 >> 2),
695 0x00000000,
696 (0x4e00 << 16) | (0xc904 >> 2),
697 0x00000000,
698 (0x5e00 << 16) | (0xc904 >> 2),
699 0x00000000,
700 (0x6e00 << 16) | (0xc904 >> 2),
701 0x00000000,
702 (0x7e00 << 16) | (0xc904 >> 2),
703 0x00000000,
704 (0x4e00 << 16) | (0xc908 >> 2),
705 0x00000000,
706 (0x5e00 << 16) | (0xc908 >> 2),
707 0x00000000,
708 (0x6e00 << 16) | (0xc908 >> 2),
709 0x00000000,
710 (0x7e00 << 16) | (0xc908 >> 2),
711 0x00000000,
712 (0x4e00 << 16) | (0xc90c >> 2),
713 0x00000000,
714 (0x5e00 << 16) | (0xc90c >> 2),
715 0x00000000,
716 (0x6e00 << 16) | (0xc90c >> 2),
717 0x00000000,
718 (0x7e00 << 16) | (0xc90c >> 2),
719 0x00000000,
720 (0x4e00 << 16) | (0xc910 >> 2),
721 0x00000000,
722 (0x5e00 << 16) | (0xc910 >> 2),
723 0x00000000,
724 (0x6e00 << 16) | (0xc910 >> 2),
725 0x00000000,
726 (0x7e00 << 16) | (0xc910 >> 2),
727 0x00000000,
728 (0x0e00 << 16) | (0xc99c >> 2),
729 0x00000000,
730 (0x0e00 << 16) | (0x9834 >> 2),
731 0x00000000,
732 (0x0000 << 16) | (0x30f00 >> 2),
733 0x00000000,
734 (0x0000 << 16) | (0x30f04 >> 2),
735 0x00000000,
736 (0x0000 << 16) | (0x30f08 >> 2),
737 0x00000000,
738 (0x0000 << 16) | (0x30f0c >> 2),
739 0x00000000,
740 (0x0600 << 16) | (0x9b7c >> 2),
741 0x00000000,
742 (0x0e00 << 16) | (0x8a14 >> 2),
743 0x00000000,
744 (0x0e00 << 16) | (0x8a18 >> 2),
745 0x00000000,
746 (0x0600 << 16) | (0x30a00 >> 2),
747 0x00000000,
748 (0x0e00 << 16) | (0x8bf0 >> 2),
749 0x00000000,
750 (0x0e00 << 16) | (0x8bcc >> 2),
751 0x00000000,
752 (0x0e00 << 16) | (0x8b24 >> 2),
753 0x00000000,
754 (0x0e00 << 16) | (0x30a04 >> 2),
755 0x00000000,
756 (0x0600 << 16) | (0x30a10 >> 2),
757 0x00000000,
758 (0x0600 << 16) | (0x30a14 >> 2),
759 0x00000000,
760 (0x0600 << 16) | (0x30a18 >> 2),
761 0x00000000,
762 (0x0600 << 16) | (0x30a2c >> 2),
763 0x00000000,
764 (0x0e00 << 16) | (0xc700 >> 2),
765 0x00000000,
766 (0x0e00 << 16) | (0xc704 >> 2),
767 0x00000000,
768 (0x0e00 << 16) | (0xc708 >> 2),
769 0x00000000,
770 (0x0e00 << 16) | (0xc768 >> 2),
771 0x00000000,
772 (0x0400 << 16) | (0xc770 >> 2),
773 0x00000000,
774 (0x0400 << 16) | (0xc774 >> 2),
775 0x00000000,
776 (0x0400 << 16) | (0xc798 >> 2),
777 0x00000000,
778 (0x0400 << 16) | (0xc79c >> 2),
779 0x00000000,
780 (0x0e00 << 16) | (0x9100 >> 2),
781 0x00000000,
782 (0x0e00 << 16) | (0x3c010 >> 2),
783 0x00000000,
784 (0x0e00 << 16) | (0x8c00 >> 2),
785 0x00000000,
786 (0x0e00 << 16) | (0x8c04 >> 2),
787 0x00000000,
788 (0x0e00 << 16) | (0x8c20 >> 2),
789 0x00000000,
790 (0x0e00 << 16) | (0x8c38 >> 2),
791 0x00000000,
792 (0x0e00 << 16) | (0x8c3c >> 2),
793 0x00000000,
794 (0x0e00 << 16) | (0xae00 >> 2),
795 0x00000000,
796 (0x0e00 << 16) | (0x9604 >> 2),
797 0x00000000,
798 (0x0e00 << 16) | (0xac08 >> 2),
799 0x00000000,
800 (0x0e00 << 16) | (0xac0c >> 2),
801 0x00000000,
802 (0x0e00 << 16) | (0xac10 >> 2),
803 0x00000000,
804 (0x0e00 << 16) | (0xac14 >> 2),
805 0x00000000,
806 (0x0e00 << 16) | (0xac58 >> 2),
807 0x00000000,
808 (0x0e00 << 16) | (0xac68 >> 2),
809 0x00000000,
810 (0x0e00 << 16) | (0xac6c >> 2),
811 0x00000000,
812 (0x0e00 << 16) | (0xac70 >> 2),
813 0x00000000,
814 (0x0e00 << 16) | (0xac74 >> 2),
815 0x00000000,
816 (0x0e00 << 16) | (0xac78 >> 2),
817 0x00000000,
818 (0x0e00 << 16) | (0xac7c >> 2),
819 0x00000000,
820 (0x0e00 << 16) | (0xac80 >> 2),
821 0x00000000,
822 (0x0e00 << 16) | (0xac84 >> 2),
823 0x00000000,
824 (0x0e00 << 16) | (0xac88 >> 2),
825 0x00000000,
826 (0x0e00 << 16) | (0xac8c >> 2),
827 0x00000000,
828 (0x0e00 << 16) | (0x970c >> 2),
829 0x00000000,
830 (0x0e00 << 16) | (0x9714 >> 2),
831 0x00000000,
832 (0x0e00 << 16) | (0x9718 >> 2),
833 0x00000000,
834 (0x0e00 << 16) | (0x971c >> 2),
835 0x00000000,
836 (0x0e00 << 16) | (0x31068 >> 2),
837 0x00000000,
838 (0x4e00 << 16) | (0x31068 >> 2),
839 0x00000000,
840 (0x5e00 << 16) | (0x31068 >> 2),
841 0x00000000,
842 (0x6e00 << 16) | (0x31068 >> 2),
843 0x00000000,
844 (0x7e00 << 16) | (0x31068 >> 2),
845 0x00000000,
846 (0x0e00 << 16) | (0xcd10 >> 2),
847 0x00000000,
848 (0x0e00 << 16) | (0xcd14 >> 2),
849 0x00000000,
850 (0x0e00 << 16) | (0x88b0 >> 2),
851 0x00000000,
852 (0x0e00 << 16) | (0x88b4 >> 2),
853 0x00000000,
854 (0x0e00 << 16) | (0x88b8 >> 2),
855 0x00000000,
856 (0x0e00 << 16) | (0x88bc >> 2),
857 0x00000000,
858 (0x0400 << 16) | (0x89c0 >> 2),
859 0x00000000,
860 (0x0e00 << 16) | (0x88c4 >> 2),
861 0x00000000,
862 (0x0e00 << 16) | (0x88c8 >> 2),
863 0x00000000,
864 (0x0e00 << 16) | (0x88d0 >> 2),
865 0x00000000,
866 (0x0e00 << 16) | (0x88d4 >> 2),
867 0x00000000,
868 (0x0e00 << 16) | (0x88d8 >> 2),
869 0x00000000,
870 (0x0e00 << 16) | (0x8980 >> 2),
871 0x00000000,
872 (0x0e00 << 16) | (0x30938 >> 2),
873 0x00000000,
874 (0x0e00 << 16) | (0x3093c >> 2),
875 0x00000000,
876 (0x0e00 << 16) | (0x30940 >> 2),
877 0x00000000,
878 (0x0e00 << 16) | (0x89a0 >> 2),
879 0x00000000,
880 (0x0e00 << 16) | (0x30900 >> 2),
881 0x00000000,
882 (0x0e00 << 16) | (0x30904 >> 2),
883 0x00000000,
884 (0x0e00 << 16) | (0x89b4 >> 2),
885 0x00000000,
886 (0x0e00 << 16) | (0x3e1fc >> 2),
887 0x00000000,
888 (0x0e00 << 16) | (0x3c210 >> 2),
889 0x00000000,
890 (0x0e00 << 16) | (0x3c214 >> 2),
891 0x00000000,
892 (0x0e00 << 16) | (0x3c218 >> 2),
893 0x00000000,
894 (0x0e00 << 16) | (0x8904 >> 2),
895 0x00000000,
896 0x5,
897 (0x0e00 << 16) | (0x8c28 >> 2),
898 (0x0e00 << 16) | (0x8c2c >> 2),
899 (0x0e00 << 16) | (0x8c30 >> 2),
900 (0x0e00 << 16) | (0x8c34 >> 2),
901 (0x0e00 << 16) | (0x9600 >> 2),
902 };
903
904 static const u32 bonaire_golden_spm_registers[] =
905 {
906 0x30800, 0xe0ffffff, 0xe0000000
907 };
908
909 static const u32 bonaire_golden_common_registers[] =
910 {
911 0xc770, 0xffffffff, 0x00000800,
912 0xc774, 0xffffffff, 0x00000800,
913 0xc798, 0xffffffff, 0x00007fbf,
914 0xc79c, 0xffffffff, 0x00007faf
915 };
916
917 static const u32 bonaire_golden_registers[] =
918 {
919 0x3354, 0x00000333, 0x00000333,
920 0x3350, 0x000c0fc0, 0x00040200,
921 0x9a10, 0x00010000, 0x00058208,
922 0x3c000, 0xffff1fff, 0x00140000,
923 0x3c200, 0xfdfc0fff, 0x00000100,
924 0x3c234, 0x40000000, 0x40000200,
925 0x9830, 0xffffffff, 0x00000000,
926 0x9834, 0xf00fffff, 0x00000400,
927 0x9838, 0x0002021c, 0x00020200,
928 0xc78, 0x00000080, 0x00000000,
929 0x5bb0, 0x000000f0, 0x00000070,
930 0x5bc0, 0xf0311fff, 0x80300000,
931 0x98f8, 0x73773777, 0x12010001,
932 0x350c, 0x00810000, 0x408af000,
933 0x7030, 0x31000111, 0x00000011,
934 0x2f48, 0x73773777, 0x12010001,
935 0x220c, 0x00007fb6, 0x0021a1b1,
936 0x2210, 0x00007fb6, 0x002021b1,
937 0x2180, 0x00007fb6, 0x00002191,
938 0x2218, 0x00007fb6, 0x002121b1,
939 0x221c, 0x00007fb6, 0x002021b1,
940 0x21dc, 0x00007fb6, 0x00002191,
941 0x21e0, 0x00007fb6, 0x00002191,
942 0x3628, 0x0000003f, 0x0000000a,
943 0x362c, 0x0000003f, 0x0000000a,
944 0x2ae4, 0x00073ffe, 0x000022a2,
945 0x240c, 0x000007ff, 0x00000000,
946 0x8a14, 0xf000003f, 0x00000007,
947 0x8bf0, 0x00002001, 0x00000001,
948 0x8b24, 0xffffffff, 0x00ffffff,
949 0x30a04, 0x0000ff0f, 0x00000000,
950 0x28a4c, 0x07ffffff, 0x06000000,
951 0x4d8, 0x00000fff, 0x00000100,
952 0x3e78, 0x00000001, 0x00000002,
953 0x9100, 0x03000000, 0x0362c688,
954 0x8c00, 0x000000ff, 0x00000001,
955 0xe40, 0x00001fff, 0x00001fff,
956 0x9060, 0x0000007f, 0x00000020,
957 0x9508, 0x00010000, 0x00010000,
958 0xac14, 0x000003ff, 0x000000f3,
959 0xac0c, 0xffffffff, 0x00001032
960 };
961
962 static const u32 bonaire_mgcg_cgcg_init[] =
963 {
964 0xc420, 0xffffffff, 0xfffffffc,
965 0x30800, 0xffffffff, 0xe0000000,
966 0x3c2a0, 0xffffffff, 0x00000100,
967 0x3c208, 0xffffffff, 0x00000100,
968 0x3c2c0, 0xffffffff, 0xc0000100,
969 0x3c2c8, 0xffffffff, 0xc0000100,
970 0x3c2c4, 0xffffffff, 0xc0000100,
971 0x55e4, 0xffffffff, 0x00600100,
972 0x3c280, 0xffffffff, 0x00000100,
973 0x3c214, 0xffffffff, 0x06000100,
974 0x3c220, 0xffffffff, 0x00000100,
975 0x3c218, 0xffffffff, 0x06000100,
976 0x3c204, 0xffffffff, 0x00000100,
977 0x3c2e0, 0xffffffff, 0x00000100,
978 0x3c224, 0xffffffff, 0x00000100,
979 0x3c200, 0xffffffff, 0x00000100,
980 0x3c230, 0xffffffff, 0x00000100,
981 0x3c234, 0xffffffff, 0x00000100,
982 0x3c250, 0xffffffff, 0x00000100,
983 0x3c254, 0xffffffff, 0x00000100,
984 0x3c258, 0xffffffff, 0x00000100,
985 0x3c25c, 0xffffffff, 0x00000100,
986 0x3c260, 0xffffffff, 0x00000100,
987 0x3c27c, 0xffffffff, 0x00000100,
988 0x3c278, 0xffffffff, 0x00000100,
989 0x3c210, 0xffffffff, 0x06000100,
990 0x3c290, 0xffffffff, 0x00000100,
991 0x3c274, 0xffffffff, 0x00000100,
992 0x3c2b4, 0xffffffff, 0x00000100,
993 0x3c2b0, 0xffffffff, 0x00000100,
994 0x3c270, 0xffffffff, 0x00000100,
995 0x30800, 0xffffffff, 0xe0000000,
996 0x3c020, 0xffffffff, 0x00010000,
997 0x3c024, 0xffffffff, 0x00030002,
998 0x3c028, 0xffffffff, 0x00040007,
999 0x3c02c, 0xffffffff, 0x00060005,
1000 0x3c030, 0xffffffff, 0x00090008,
1001 0x3c034, 0xffffffff, 0x00010000,
1002 0x3c038, 0xffffffff, 0x00030002,
1003 0x3c03c, 0xffffffff, 0x00040007,
1004 0x3c040, 0xffffffff, 0x00060005,
1005 0x3c044, 0xffffffff, 0x00090008,
1006 0x3c048, 0xffffffff, 0x00010000,
1007 0x3c04c, 0xffffffff, 0x00030002,
1008 0x3c050, 0xffffffff, 0x00040007,
1009 0x3c054, 0xffffffff, 0x00060005,
1010 0x3c058, 0xffffffff, 0x00090008,
1011 0x3c05c, 0xffffffff, 0x00010000,
1012 0x3c060, 0xffffffff, 0x00030002,
1013 0x3c064, 0xffffffff, 0x00040007,
1014 0x3c068, 0xffffffff, 0x00060005,
1015 0x3c06c, 0xffffffff, 0x00090008,
1016 0x3c070, 0xffffffff, 0x00010000,
1017 0x3c074, 0xffffffff, 0x00030002,
1018 0x3c078, 0xffffffff, 0x00040007,
1019 0x3c07c, 0xffffffff, 0x00060005,
1020 0x3c080, 0xffffffff, 0x00090008,
1021 0x3c084, 0xffffffff, 0x00010000,
1022 0x3c088, 0xffffffff, 0x00030002,
1023 0x3c08c, 0xffffffff, 0x00040007,
1024 0x3c090, 0xffffffff, 0x00060005,
1025 0x3c094, 0xffffffff, 0x00090008,
1026 0x3c098, 0xffffffff, 0x00010000,
1027 0x3c09c, 0xffffffff, 0x00030002,
1028 0x3c0a0, 0xffffffff, 0x00040007,
1029 0x3c0a4, 0xffffffff, 0x00060005,
1030 0x3c0a8, 0xffffffff, 0x00090008,
1031 0x3c000, 0xffffffff, 0x96e00200,
1032 0x8708, 0xffffffff, 0x00900100,
1033 0xc424, 0xffffffff, 0x0020003f,
1034 0x38, 0xffffffff, 0x0140001c,
1035 0x3c, 0x000f0000, 0x000f0000,
1036 0x220, 0xffffffff, 0xC060000C,
1037 0x224, 0xc0000fff, 0x00000100,
1038 0xf90, 0xffffffff, 0x00000100,
1039 0xf98, 0x00000101, 0x00000000,
1040 0x20a8, 0xffffffff, 0x00000104,
1041 0x55e4, 0xff000fff, 0x00000100,
1042 0x30cc, 0xc0000fff, 0x00000104,
1043 0xc1e4, 0x00000001, 0x00000001,
1044 0xd00c, 0xff000ff0, 0x00000100,
1045 0xd80c, 0xff000ff0, 0x00000100
1046 };
1047
1048 static const u32 spectre_golden_spm_registers[] =
1049 {
1050 0x30800, 0xe0ffffff, 0xe0000000
1051 };
1052
1053 static const u32 spectre_golden_common_registers[] =
1054 {
1055 0xc770, 0xffffffff, 0x00000800,
1056 0xc774, 0xffffffff, 0x00000800,
1057 0xc798, 0xffffffff, 0x00007fbf,
1058 0xc79c, 0xffffffff, 0x00007faf
1059 };
1060
1061 static const u32 spectre_golden_registers[] =
1062 {
1063 0x3c000, 0xffff1fff, 0x96940200,
1064 0x3c00c, 0xffff0001, 0xff000000,
1065 0x3c200, 0xfffc0fff, 0x00000100,
1066 0x6ed8, 0x00010101, 0x00010000,
1067 0x9834, 0xf00fffff, 0x00000400,
1068 0x9838, 0xfffffffc, 0x00020200,
1069 0x5bb0, 0x000000f0, 0x00000070,
1070 0x5bc0, 0xf0311fff, 0x80300000,
1071 0x98f8, 0x73773777, 0x12010001,
1072 0x9b7c, 0x00ff0000, 0x00fc0000,
1073 0x2f48, 0x73773777, 0x12010001,
1074 0x8a14, 0xf000003f, 0x00000007,
1075 0x8b24, 0xffffffff, 0x00ffffff,
1076 0x28350, 0x3f3f3fff, 0x00000082,
1077 0x28355, 0x0000003f, 0x00000000,
1078 0x3e78, 0x00000001, 0x00000002,
1079 0x913c, 0xffff03df, 0x00000004,
1080 0xc768, 0x00000008, 0x00000008,
1081 0x8c00, 0x000008ff, 0x00000800,
1082 0x9508, 0x00010000, 0x00010000,
1083 0xac0c, 0xffffffff, 0x54763210,
1084 0x214f8, 0x01ff01ff, 0x00000002,
1085 0x21498, 0x007ff800, 0x00200000,
1086 0x2015c, 0xffffffff, 0x00000f40,
1087 0x30934, 0xffffffff, 0x00000001
1088 };
1089
1090 static const u32 spectre_mgcg_cgcg_init[] =
1091 {
1092 0xc420, 0xffffffff, 0xfffffffc,
1093 0x30800, 0xffffffff, 0xe0000000,
1094 0x3c2a0, 0xffffffff, 0x00000100,
1095 0x3c208, 0xffffffff, 0x00000100,
1096 0x3c2c0, 0xffffffff, 0x00000100,
1097 0x3c2c8, 0xffffffff, 0x00000100,
1098 0x3c2c4, 0xffffffff, 0x00000100,
1099 0x55e4, 0xffffffff, 0x00600100,
1100 0x3c280, 0xffffffff, 0x00000100,
1101 0x3c214, 0xffffffff, 0x06000100,
1102 0x3c220, 0xffffffff, 0x00000100,
1103 0x3c218, 0xffffffff, 0x06000100,
1104 0x3c204, 0xffffffff, 0x00000100,
1105 0x3c2e0, 0xffffffff, 0x00000100,
1106 0x3c224, 0xffffffff, 0x00000100,
1107 0x3c200, 0xffffffff, 0x00000100,
1108 0x3c230, 0xffffffff, 0x00000100,
1109 0x3c234, 0xffffffff, 0x00000100,
1110 0x3c250, 0xffffffff, 0x00000100,
1111 0x3c254, 0xffffffff, 0x00000100,
1112 0x3c258, 0xffffffff, 0x00000100,
1113 0x3c25c, 0xffffffff, 0x00000100,
1114 0x3c260, 0xffffffff, 0x00000100,
1115 0x3c27c, 0xffffffff, 0x00000100,
1116 0x3c278, 0xffffffff, 0x00000100,
1117 0x3c210, 0xffffffff, 0x06000100,
1118 0x3c290, 0xffffffff, 0x00000100,
1119 0x3c274, 0xffffffff, 0x00000100,
1120 0x3c2b4, 0xffffffff, 0x00000100,
1121 0x3c2b0, 0xffffffff, 0x00000100,
1122 0x3c270, 0xffffffff, 0x00000100,
1123 0x30800, 0xffffffff, 0xe0000000,
1124 0x3c020, 0xffffffff, 0x00010000,
1125 0x3c024, 0xffffffff, 0x00030002,
1126 0x3c028, 0xffffffff, 0x00040007,
1127 0x3c02c, 0xffffffff, 0x00060005,
1128 0x3c030, 0xffffffff, 0x00090008,
1129 0x3c034, 0xffffffff, 0x00010000,
1130 0x3c038, 0xffffffff, 0x00030002,
1131 0x3c03c, 0xffffffff, 0x00040007,
1132 0x3c040, 0xffffffff, 0x00060005,
1133 0x3c044, 0xffffffff, 0x00090008,
1134 0x3c048, 0xffffffff, 0x00010000,
1135 0x3c04c, 0xffffffff, 0x00030002,
1136 0x3c050, 0xffffffff, 0x00040007,
1137 0x3c054, 0xffffffff, 0x00060005,
1138 0x3c058, 0xffffffff, 0x00090008,
1139 0x3c05c, 0xffffffff, 0x00010000,
1140 0x3c060, 0xffffffff, 0x00030002,
1141 0x3c064, 0xffffffff, 0x00040007,
1142 0x3c068, 0xffffffff, 0x00060005,
1143 0x3c06c, 0xffffffff, 0x00090008,
1144 0x3c070, 0xffffffff, 0x00010000,
1145 0x3c074, 0xffffffff, 0x00030002,
1146 0x3c078, 0xffffffff, 0x00040007,
1147 0x3c07c, 0xffffffff, 0x00060005,
1148 0x3c080, 0xffffffff, 0x00090008,
1149 0x3c084, 0xffffffff, 0x00010000,
1150 0x3c088, 0xffffffff, 0x00030002,
1151 0x3c08c, 0xffffffff, 0x00040007,
1152 0x3c090, 0xffffffff, 0x00060005,
1153 0x3c094, 0xffffffff, 0x00090008,
1154 0x3c098, 0xffffffff, 0x00010000,
1155 0x3c09c, 0xffffffff, 0x00030002,
1156 0x3c0a0, 0xffffffff, 0x00040007,
1157 0x3c0a4, 0xffffffff, 0x00060005,
1158 0x3c0a8, 0xffffffff, 0x00090008,
1159 0x3c0ac, 0xffffffff, 0x00010000,
1160 0x3c0b0, 0xffffffff, 0x00030002,
1161 0x3c0b4, 0xffffffff, 0x00040007,
1162 0x3c0b8, 0xffffffff, 0x00060005,
1163 0x3c0bc, 0xffffffff, 0x00090008,
1164 0x3c000, 0xffffffff, 0x96e00200,
1165 0x8708, 0xffffffff, 0x00900100,
1166 0xc424, 0xffffffff, 0x0020003f,
1167 0x38, 0xffffffff, 0x0140001c,
1168 0x3c, 0x000f0000, 0x000f0000,
1169 0x220, 0xffffffff, 0xC060000C,
1170 0x224, 0xc0000fff, 0x00000100,
1171 0xf90, 0xffffffff, 0x00000100,
1172 0xf98, 0x00000101, 0x00000000,
1173 0x20a8, 0xffffffff, 0x00000104,
1174 0x55e4, 0xff000fff, 0x00000100,
1175 0x30cc, 0xc0000fff, 0x00000104,
1176 0xc1e4, 0x00000001, 0x00000001,
1177 0xd00c, 0xff000ff0, 0x00000100,
1178 0xd80c, 0xff000ff0, 0x00000100
1179 };
1180
1181 static const u32 kalindi_golden_spm_registers[] =
1182 {
1183 0x30800, 0xe0ffffff, 0xe0000000
1184 };
1185
1186 static const u32 kalindi_golden_common_registers[] =
1187 {
1188 0xc770, 0xffffffff, 0x00000800,
1189 0xc774, 0xffffffff, 0x00000800,
1190 0xc798, 0xffffffff, 0x00007fbf,
1191 0xc79c, 0xffffffff, 0x00007faf
1192 };
1193
1194 static const u32 kalindi_golden_registers[] =
1195 {
1196 0x3c000, 0xffffdfff, 0x6e944040,
1197 0x55e4, 0xff607fff, 0xfc000100,
1198 0x3c220, 0xff000fff, 0x00000100,
1199 0x3c224, 0xff000fff, 0x00000100,
1200 0x3c200, 0xfffc0fff, 0x00000100,
1201 0x6ed8, 0x00010101, 0x00010000,
1202 0x9830, 0xffffffff, 0x00000000,
1203 0x9834, 0xf00fffff, 0x00000400,
1204 0x5bb0, 0x000000f0, 0x00000070,
1205 0x5bc0, 0xf0311fff, 0x80300000,
1206 0x98f8, 0x73773777, 0x12010001,
1207 0x98fc, 0xffffffff, 0x00000010,
1208 0x9b7c, 0x00ff0000, 0x00fc0000,
1209 0x8030, 0x00001f0f, 0x0000100a,
1210 0x2f48, 0x73773777, 0x12010001,
1211 0x2408, 0x000fffff, 0x000c007f,
1212 0x8a14, 0xf000003f, 0x00000007,
1213 0x8b24, 0x3fff3fff, 0x00ffcfff,
1214 0x30a04, 0x0000ff0f, 0x00000000,
1215 0x28a4c, 0x07ffffff, 0x06000000,
1216 0x4d8, 0x00000fff, 0x00000100,
1217 0x3e78, 0x00000001, 0x00000002,
1218 0xc768, 0x00000008, 0x00000008,
1219 0x8c00, 0x000000ff, 0x00000003,
1220 0x214f8, 0x01ff01ff, 0x00000002,
1221 0x21498, 0x007ff800, 0x00200000,
1222 0x2015c, 0xffffffff, 0x00000f40,
1223 0x88c4, 0x001f3ae3, 0x00000082,
1224 0x88d4, 0x0000001f, 0x00000010,
1225 0x30934, 0xffffffff, 0x00000000
1226 };
1227
1228 static const u32 kalindi_mgcg_cgcg_init[] =
1229 {
1230 0xc420, 0xffffffff, 0xfffffffc,
1231 0x30800, 0xffffffff, 0xe0000000,
1232 0x3c2a0, 0xffffffff, 0x00000100,
1233 0x3c208, 0xffffffff, 0x00000100,
1234 0x3c2c0, 0xffffffff, 0x00000100,
1235 0x3c2c8, 0xffffffff, 0x00000100,
1236 0x3c2c4, 0xffffffff, 0x00000100,
1237 0x55e4, 0xffffffff, 0x00600100,
1238 0x3c280, 0xffffffff, 0x00000100,
1239 0x3c214, 0xffffffff, 0x06000100,
1240 0x3c220, 0xffffffff, 0x00000100,
1241 0x3c218, 0xffffffff, 0x06000100,
1242 0x3c204, 0xffffffff, 0x00000100,
1243 0x3c2e0, 0xffffffff, 0x00000100,
1244 0x3c224, 0xffffffff, 0x00000100,
1245 0x3c200, 0xffffffff, 0x00000100,
1246 0x3c230, 0xffffffff, 0x00000100,
1247 0x3c234, 0xffffffff, 0x00000100,
1248 0x3c250, 0xffffffff, 0x00000100,
1249 0x3c254, 0xffffffff, 0x00000100,
1250 0x3c258, 0xffffffff, 0x00000100,
1251 0x3c25c, 0xffffffff, 0x00000100,
1252 0x3c260, 0xffffffff, 0x00000100,
1253 0x3c27c, 0xffffffff, 0x00000100,
1254 0x3c278, 0xffffffff, 0x00000100,
1255 0x3c210, 0xffffffff, 0x06000100,
1256 0x3c290, 0xffffffff, 0x00000100,
1257 0x3c274, 0xffffffff, 0x00000100,
1258 0x3c2b4, 0xffffffff, 0x00000100,
1259 0x3c2b0, 0xffffffff, 0x00000100,
1260 0x3c270, 0xffffffff, 0x00000100,
1261 0x30800, 0xffffffff, 0xe0000000,
1262 0x3c020, 0xffffffff, 0x00010000,
1263 0x3c024, 0xffffffff, 0x00030002,
1264 0x3c028, 0xffffffff, 0x00040007,
1265 0x3c02c, 0xffffffff, 0x00060005,
1266 0x3c030, 0xffffffff, 0x00090008,
1267 0x3c034, 0xffffffff, 0x00010000,
1268 0x3c038, 0xffffffff, 0x00030002,
1269 0x3c03c, 0xffffffff, 0x00040007,
1270 0x3c040, 0xffffffff, 0x00060005,
1271 0x3c044, 0xffffffff, 0x00090008,
1272 0x3c000, 0xffffffff, 0x96e00200,
1273 0x8708, 0xffffffff, 0x00900100,
1274 0xc424, 0xffffffff, 0x0020003f,
1275 0x38, 0xffffffff, 0x0140001c,
1276 0x3c, 0x000f0000, 0x000f0000,
1277 0x220, 0xffffffff, 0xC060000C,
1278 0x224, 0xc0000fff, 0x00000100,
1279 0x20a8, 0xffffffff, 0x00000104,
1280 0x55e4, 0xff000fff, 0x00000100,
1281 0x30cc, 0xc0000fff, 0x00000104,
1282 0xc1e4, 0x00000001, 0x00000001,
1283 0xd00c, 0xff000ff0, 0x00000100,
1284 0xd80c, 0xff000ff0, 0x00000100
1285 };
1286
1287 static void cik_init_golden_registers(struct radeon_device *rdev)
1288 {
1289 switch (rdev->family) {
1290 case CHIP_BONAIRE:
1291 radeon_program_register_sequence(rdev,
1292 bonaire_mgcg_cgcg_init,
1293 (const u32)ARRAY_SIZE(bonaire_mgcg_cgcg_init));
1294 radeon_program_register_sequence(rdev,
1295 bonaire_golden_registers,
1296 (const u32)ARRAY_SIZE(bonaire_golden_registers));
1297 radeon_program_register_sequence(rdev,
1298 bonaire_golden_common_registers,
1299 (const u32)ARRAY_SIZE(bonaire_golden_common_registers));
1300 radeon_program_register_sequence(rdev,
1301 bonaire_golden_spm_registers,
1302 (const u32)ARRAY_SIZE(bonaire_golden_spm_registers));
1303 break;
1304 case CHIP_KABINI:
1305 radeon_program_register_sequence(rdev,
1306 kalindi_mgcg_cgcg_init,
1307 (const u32)ARRAY_SIZE(kalindi_mgcg_cgcg_init));
1308 radeon_program_register_sequence(rdev,
1309 kalindi_golden_registers,
1310 (const u32)ARRAY_SIZE(kalindi_golden_registers));
1311 radeon_program_register_sequence(rdev,
1312 kalindi_golden_common_registers,
1313 (const u32)ARRAY_SIZE(kalindi_golden_common_registers));
1314 radeon_program_register_sequence(rdev,
1315 kalindi_golden_spm_registers,
1316 (const u32)ARRAY_SIZE(kalindi_golden_spm_registers));
1317 break;
1318 case CHIP_KAVERI:
1319 radeon_program_register_sequence(rdev,
1320 spectre_mgcg_cgcg_init,
1321 (const u32)ARRAY_SIZE(spectre_mgcg_cgcg_init));
1322 radeon_program_register_sequence(rdev,
1323 spectre_golden_registers,
1324 (const u32)ARRAY_SIZE(spectre_golden_registers));
1325 radeon_program_register_sequence(rdev,
1326 spectre_golden_common_registers,
1327 (const u32)ARRAY_SIZE(spectre_golden_common_registers));
1328 radeon_program_register_sequence(rdev,
1329 spectre_golden_spm_registers,
1330 (const u32)ARRAY_SIZE(spectre_golden_spm_registers));
1331 break;
1332 default:
1333 break;
1334 }
1335 }
1336
1337 /**
1338 * cik_get_xclk - get the xclk
1339 *
1340 * @rdev: radeon_device pointer
1341 *
1342 * Returns the reference clock used by the gfx engine
1343 * (CIK).
1344 */
1345 u32 cik_get_xclk(struct radeon_device *rdev)
1346 {
1347 u32 reference_clock = rdev->clock.spll.reference_freq;
1348
1349 if (rdev->flags & RADEON_IS_IGP) {
1350 if (RREG32_SMC(GENERAL_PWRMGT) & GPU_COUNTER_CLK)
1351 return reference_clock / 2;
1352 } else {
1353 if (RREG32_SMC(CG_CLKPIN_CNTL) & XTALIN_DIVIDE)
1354 return reference_clock / 4;
1355 }
1356 return reference_clock;
1357 }
1358
1359 /**
1360 * cik_mm_rdoorbell - read a doorbell dword
1361 *
1362 * @rdev: radeon_device pointer
1363 * @offset: byte offset into the aperture
1364 *
1365 * Returns the value in the doorbell aperture at the
1366 * requested offset (CIK).
1367 */
1368 u32 cik_mm_rdoorbell(struct radeon_device *rdev, u32 offset)
1369 {
1370 if (offset < rdev->doorbell.size) {
1371 return readl(((void __iomem *)rdev->doorbell.ptr) + offset);
1372 } else {
1373 DRM_ERROR("reading beyond doorbell aperture: 0x%08x!\n", offset);
1374 return 0;
1375 }
1376 }
1377
1378 /**
1379 * cik_mm_wdoorbell - write a doorbell dword
1380 *
1381 * @rdev: radeon_device pointer
1382 * @offset: byte offset into the aperture
1383 * @v: value to write
1384 *
1385 * Writes @v to the doorbell aperture at the
1386 * requested offset (CIK).
1387 */
1388 void cik_mm_wdoorbell(struct radeon_device *rdev, u32 offset, u32 v)
1389 {
1390 if (offset < rdev->doorbell.size) {
1391 writel(v, ((void __iomem *)rdev->doorbell.ptr) + offset);
1392 } else {
1393 DRM_ERROR("writing beyond doorbell aperture: 0x%08x!\n", offset);
1394 }
1395 }
1396
1397 #define BONAIRE_IO_MC_REGS_SIZE 36
1398
1399 static const u32 bonaire_io_mc_regs[BONAIRE_IO_MC_REGS_SIZE][2] =
1400 {
1401 {0x00000070, 0x04400000},
1402 {0x00000071, 0x80c01803},
1403 {0x00000072, 0x00004004},
1404 {0x00000073, 0x00000100},
1405 {0x00000074, 0x00ff0000},
1406 {0x00000075, 0x34000000},
1407 {0x00000076, 0x08000014},
1408 {0x00000077, 0x00cc08ec},
1409 {0x00000078, 0x00000400},
1410 {0x00000079, 0x00000000},
1411 {0x0000007a, 0x04090000},
1412 {0x0000007c, 0x00000000},
1413 {0x0000007e, 0x4408a8e8},
1414 {0x0000007f, 0x00000304},
1415 {0x00000080, 0x00000000},
1416 {0x00000082, 0x00000001},
1417 {0x00000083, 0x00000002},
1418 {0x00000084, 0xf3e4f400},
1419 {0x00000085, 0x052024e3},
1420 {0x00000087, 0x00000000},
1421 {0x00000088, 0x01000000},
1422 {0x0000008a, 0x1c0a0000},
1423 {0x0000008b, 0xff010000},
1424 {0x0000008d, 0xffffefff},
1425 {0x0000008e, 0xfff3efff},
1426 {0x0000008f, 0xfff3efbf},
1427 {0x00000092, 0xf7ffffff},
1428 {0x00000093, 0xffffff7f},
1429 {0x00000095, 0x00101101},
1430 {0x00000096, 0x00000fff},
1431 {0x00000097, 0x00116fff},
1432 {0x00000098, 0x60010000},
1433 {0x00000099, 0x10010000},
1434 {0x0000009a, 0x00006000},
1435 {0x0000009b, 0x00001000},
1436 {0x0000009f, 0x00b48000}
1437 };
1438
1439 /**
1440 * cik_srbm_select - select specific register instances
1441 *
1442 * @rdev: radeon_device pointer
1443 * @me: selected ME (micro engine)
1444 * @pipe: pipe
1445 * @queue: queue
1446 * @vmid: VMID
1447 *
1448 * Switches the currently active registers instances. Some
1449 * registers are instanced per VMID, others are instanced per
1450 * me/pipe/queue combination.
1451 */
1452 static void cik_srbm_select(struct radeon_device *rdev,
1453 u32 me, u32 pipe, u32 queue, u32 vmid)
1454 {
1455 u32 srbm_gfx_cntl = (PIPEID(pipe & 0x3) |
1456 MEID(me & 0x3) |
1457 VMID(vmid & 0xf) |
1458 QUEUEID(queue & 0x7));
1459 WREG32(SRBM_GFX_CNTL, srbm_gfx_cntl);
1460 }
1461
1462 /* ucode loading */
1463 /**
1464 * ci_mc_load_microcode - load MC ucode into the hw
1465 *
1466 * @rdev: radeon_device pointer
1467 *
1468 * Load the GDDR MC ucode into the hw (CIK).
1469 * Returns 0 on success, error on failure.
1470 */
1471 static int ci_mc_load_microcode(struct radeon_device *rdev)
1472 {
1473 const __be32 *fw_data;
1474 u32 running, blackout = 0;
1475 u32 *io_mc_regs;
1476 int i, ucode_size, regs_size;
1477
1478 if (!rdev->mc_fw)
1479 return -EINVAL;
1480
1481 switch (rdev->family) {
1482 case CHIP_BONAIRE:
1483 default:
1484 io_mc_regs = (u32 *)&bonaire_io_mc_regs;
1485 ucode_size = CIK_MC_UCODE_SIZE;
1486 regs_size = BONAIRE_IO_MC_REGS_SIZE;
1487 break;
1488 }
1489
1490 running = RREG32(MC_SEQ_SUP_CNTL) & RUN_MASK;
1491
1492 if (running == 0) {
1493 if (running) {
1494 blackout = RREG32(MC_SHARED_BLACKOUT_CNTL);
1495 WREG32(MC_SHARED_BLACKOUT_CNTL, blackout | 1);
1496 }
1497
1498 /* reset the engine and set to writable */
1499 WREG32(MC_SEQ_SUP_CNTL, 0x00000008);
1500 WREG32(MC_SEQ_SUP_CNTL, 0x00000010);
1501
1502 /* load mc io regs */
1503 for (i = 0; i < regs_size; i++) {
1504 WREG32(MC_SEQ_IO_DEBUG_INDEX, io_mc_regs[(i << 1)]);
1505 WREG32(MC_SEQ_IO_DEBUG_DATA, io_mc_regs[(i << 1) + 1]);
1506 }
1507 /* load the MC ucode */
1508 fw_data = (const __be32 *)rdev->mc_fw->data;
1509 for (i = 0; i < ucode_size; i++)
1510 WREG32(MC_SEQ_SUP_PGM, be32_to_cpup(fw_data++));
1511
1512 /* put the engine back into the active state */
1513 WREG32(MC_SEQ_SUP_CNTL, 0x00000008);
1514 WREG32(MC_SEQ_SUP_CNTL, 0x00000004);
1515 WREG32(MC_SEQ_SUP_CNTL, 0x00000001);
1516
1517 /* wait for training to complete */
1518 for (i = 0; i < rdev->usec_timeout; i++) {
1519 if (RREG32(MC_SEQ_TRAIN_WAKEUP_CNTL) & TRAIN_DONE_D0)
1520 break;
1521 udelay(1);
1522 }
1523 for (i = 0; i < rdev->usec_timeout; i++) {
1524 if (RREG32(MC_SEQ_TRAIN_WAKEUP_CNTL) & TRAIN_DONE_D1)
1525 break;
1526 udelay(1);
1527 }
1528
1529 if (running)
1530 WREG32(MC_SHARED_BLACKOUT_CNTL, blackout);
1531 }
1532
1533 return 0;
1534 }
1535
1536 /**
1537 * cik_init_microcode - load ucode images from disk
1538 *
1539 * @rdev: radeon_device pointer
1540 *
1541 * Use the firmware interface to load the ucode images into
1542 * the driver (not loaded into hw).
1543 * Returns 0 on success, error on failure.
1544 */
1545 static int cik_init_microcode(struct radeon_device *rdev)
1546 {
1547 const char *chip_name;
1548 size_t pfp_req_size, me_req_size, ce_req_size,
1549 mec_req_size, rlc_req_size, mc_req_size,
1550 sdma_req_size, smc_req_size;
1551 char fw_name[30];
1552 int err;
1553
1554 DRM_DEBUG("\n");
1555
1556 switch (rdev->family) {
1557 case CHIP_BONAIRE:
1558 chip_name = "BONAIRE";
1559 pfp_req_size = CIK_PFP_UCODE_SIZE * 4;
1560 me_req_size = CIK_ME_UCODE_SIZE * 4;
1561 ce_req_size = CIK_CE_UCODE_SIZE * 4;
1562 mec_req_size = CIK_MEC_UCODE_SIZE * 4;
1563 rlc_req_size = BONAIRE_RLC_UCODE_SIZE * 4;
1564 mc_req_size = CIK_MC_UCODE_SIZE * 4;
1565 sdma_req_size = CIK_SDMA_UCODE_SIZE * 4;
1566 smc_req_size = ALIGN(BONAIRE_SMC_UCODE_SIZE, 4);
1567 break;
1568 case CHIP_KAVERI:
1569 chip_name = "KAVERI";
1570 pfp_req_size = CIK_PFP_UCODE_SIZE * 4;
1571 me_req_size = CIK_ME_UCODE_SIZE * 4;
1572 ce_req_size = CIK_CE_UCODE_SIZE * 4;
1573 mec_req_size = CIK_MEC_UCODE_SIZE * 4;
1574 rlc_req_size = KV_RLC_UCODE_SIZE * 4;
1575 sdma_req_size = CIK_SDMA_UCODE_SIZE * 4;
1576 break;
1577 case CHIP_KABINI:
1578 chip_name = "KABINI";
1579 pfp_req_size = CIK_PFP_UCODE_SIZE * 4;
1580 me_req_size = CIK_ME_UCODE_SIZE * 4;
1581 ce_req_size = CIK_CE_UCODE_SIZE * 4;
1582 mec_req_size = CIK_MEC_UCODE_SIZE * 4;
1583 rlc_req_size = KB_RLC_UCODE_SIZE * 4;
1584 sdma_req_size = CIK_SDMA_UCODE_SIZE * 4;
1585 break;
1586 default: BUG();
1587 }
1588
1589 DRM_INFO("Loading %s Microcode\n", chip_name);
1590
1591 snprintf(fw_name, sizeof(fw_name), "radeon/%s_pfp.bin", chip_name);
1592 err = request_firmware(&rdev->pfp_fw, fw_name, rdev->dev);
1593 if (err)
1594 goto out;
1595 if (rdev->pfp_fw->size != pfp_req_size) {
1596 printk(KERN_ERR
1597 "cik_cp: Bogus length %zu in firmware \"%s\"\n",
1598 rdev->pfp_fw->size, fw_name);
1599 err = -EINVAL;
1600 goto out;
1601 }
1602
1603 snprintf(fw_name, sizeof(fw_name), "radeon/%s_me.bin", chip_name);
1604 err = request_firmware(&rdev->me_fw, fw_name, rdev->dev);
1605 if (err)
1606 goto out;
1607 if (rdev->me_fw->size != me_req_size) {
1608 printk(KERN_ERR
1609 "cik_cp: Bogus length %zu in firmware \"%s\"\n",
1610 rdev->me_fw->size, fw_name);
1611 err = -EINVAL;
1612 }
1613
1614 snprintf(fw_name, sizeof(fw_name), "radeon/%s_ce.bin", chip_name);
1615 err = request_firmware(&rdev->ce_fw, fw_name, rdev->dev);
1616 if (err)
1617 goto out;
1618 if (rdev->ce_fw->size != ce_req_size) {
1619 printk(KERN_ERR
1620 "cik_cp: Bogus length %zu in firmware \"%s\"\n",
1621 rdev->ce_fw->size, fw_name);
1622 err = -EINVAL;
1623 }
1624
1625 snprintf(fw_name, sizeof(fw_name), "radeon/%s_mec.bin", chip_name);
1626 err = request_firmware(&rdev->mec_fw, fw_name, rdev->dev);
1627 if (err)
1628 goto out;
1629 if (rdev->mec_fw->size != mec_req_size) {
1630 printk(KERN_ERR
1631 "cik_cp: Bogus length %zu in firmware \"%s\"\n",
1632 rdev->mec_fw->size, fw_name);
1633 err = -EINVAL;
1634 }
1635
1636 snprintf(fw_name, sizeof(fw_name), "radeon/%s_rlc.bin", chip_name);
1637 err = request_firmware(&rdev->rlc_fw, fw_name, rdev->dev);
1638 if (err)
1639 goto out;
1640 if (rdev->rlc_fw->size != rlc_req_size) {
1641 printk(KERN_ERR
1642 "cik_rlc: Bogus length %zu in firmware \"%s\"\n",
1643 rdev->rlc_fw->size, fw_name);
1644 err = -EINVAL;
1645 }
1646
1647 snprintf(fw_name, sizeof(fw_name), "radeon/%s_sdma.bin", chip_name);
1648 err = request_firmware(&rdev->sdma_fw, fw_name, rdev->dev);
1649 if (err)
1650 goto out;
1651 if (rdev->sdma_fw->size != sdma_req_size) {
1652 printk(KERN_ERR
1653 "cik_sdma: Bogus length %zu in firmware \"%s\"\n",
1654 rdev->sdma_fw->size, fw_name);
1655 err = -EINVAL;
1656 }
1657
1658 /* No SMC, MC ucode on APUs */
1659 if (!(rdev->flags & RADEON_IS_IGP)) {
1660 snprintf(fw_name, sizeof(fw_name), "radeon/%s_mc.bin", chip_name);
1661 err = request_firmware(&rdev->mc_fw, fw_name, rdev->dev);
1662 if (err)
1663 goto out;
1664 if (rdev->mc_fw->size != mc_req_size) {
1665 printk(KERN_ERR
1666 "cik_mc: Bogus length %zu in firmware \"%s\"\n",
1667 rdev->mc_fw->size, fw_name);
1668 err = -EINVAL;
1669 }
1670
1671 snprintf(fw_name, sizeof(fw_name), "radeon/%s_smc.bin", chip_name);
1672 err = request_firmware(&rdev->smc_fw, fw_name, rdev->dev);
1673 if (err) {
1674 printk(KERN_ERR
1675 "smc: error loading firmware \"%s\"\n",
1676 fw_name);
1677 release_firmware(rdev->smc_fw);
1678 rdev->smc_fw = NULL;
1679 } else if (rdev->smc_fw->size != smc_req_size) {
1680 printk(KERN_ERR
1681 "cik_smc: Bogus length %zu in firmware \"%s\"\n",
1682 rdev->smc_fw->size, fw_name);
1683 err = -EINVAL;
1684 }
1685 }
1686
1687 out:
1688 if (err) {
1689 if (err != -EINVAL)
1690 printk(KERN_ERR
1691 "cik_cp: Failed to load firmware \"%s\"\n",
1692 fw_name);
1693 release_firmware(rdev->pfp_fw);
1694 rdev->pfp_fw = NULL;
1695 release_firmware(rdev->me_fw);
1696 rdev->me_fw = NULL;
1697 release_firmware(rdev->ce_fw);
1698 rdev->ce_fw = NULL;
1699 release_firmware(rdev->rlc_fw);
1700 rdev->rlc_fw = NULL;
1701 release_firmware(rdev->mc_fw);
1702 rdev->mc_fw = NULL;
1703 release_firmware(rdev->smc_fw);
1704 rdev->smc_fw = NULL;
1705 }
1706 return err;
1707 }
1708
1709 /*
1710 * Core functions
1711 */
1712 /**
1713 * cik_tiling_mode_table_init - init the hw tiling table
1714 *
1715 * @rdev: radeon_device pointer
1716 *
1717 * Starting with SI, the tiling setup is done globally in a
1718 * set of 32 tiling modes. Rather than selecting each set of
1719 * parameters per surface as on older asics, we just select
1720 * which index in the tiling table we want to use, and the
1721 * surface uses those parameters (CIK).
1722 */
1723 static void cik_tiling_mode_table_init(struct radeon_device *rdev)
1724 {
1725 const u32 num_tile_mode_states = 32;
1726 const u32 num_secondary_tile_mode_states = 16;
1727 u32 reg_offset, gb_tile_moden, split_equal_to_row_size;
1728 u32 num_pipe_configs;
1729 u32 num_rbs = rdev->config.cik.max_backends_per_se *
1730 rdev->config.cik.max_shader_engines;
1731
1732 switch (rdev->config.cik.mem_row_size_in_kb) {
1733 case 1:
1734 split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_1KB;
1735 break;
1736 case 2:
1737 default:
1738 split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_2KB;
1739 break;
1740 case 4:
1741 split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_4KB;
1742 break;
1743 }
1744
1745 num_pipe_configs = rdev->config.cik.max_tile_pipes;
1746 if (num_pipe_configs > 8)
1747 num_pipe_configs = 8; /* ??? */
1748
1749 if (num_pipe_configs == 8) {
1750 for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) {
1751 switch (reg_offset) {
1752 case 0:
1753 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1754 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1755 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
1756 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B));
1757 break;
1758 case 1:
1759 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1760 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1761 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
1762 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B));
1763 break;
1764 case 2:
1765 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1766 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1767 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
1768 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B));
1769 break;
1770 case 3:
1771 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1772 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1773 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
1774 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B));
1775 break;
1776 case 4:
1777 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1778 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1779 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
1780 TILE_SPLIT(split_equal_to_row_size));
1781 break;
1782 case 5:
1783 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1784 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1785 break;
1786 case 6:
1787 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
1788 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1789 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
1790 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B));
1791 break;
1792 case 7:
1793 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
1794 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1795 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
1796 TILE_SPLIT(split_equal_to_row_size));
1797 break;
1798 case 8:
1799 gb_tile_moden = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
1800 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16));
1801 break;
1802 case 9:
1803 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1804 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING));
1805 break;
1806 case 10:
1807 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1808 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
1809 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
1810 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1811 break;
1812 case 11:
1813 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1814 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
1815 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1816 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1817 break;
1818 case 12:
1819 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
1820 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
1821 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
1822 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1823 break;
1824 case 13:
1825 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1826 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING));
1827 break;
1828 case 14:
1829 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1830 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1831 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
1832 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1833 break;
1834 case 16:
1835 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1836 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1837 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1838 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1839 break;
1840 case 17:
1841 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
1842 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1843 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
1844 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1845 break;
1846 case 27:
1847 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1848 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING));
1849 break;
1850 case 28:
1851 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1852 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
1853 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
1854 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1855 break;
1856 case 29:
1857 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1858 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
1859 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1860 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1861 break;
1862 case 30:
1863 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
1864 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
1865 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
1866 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1867 break;
1868 default:
1869 gb_tile_moden = 0;
1870 break;
1871 }
1872 rdev->config.cik.tile_mode_array[reg_offset] = gb_tile_moden;
1873 WREG32(GB_TILE_MODE0 + (reg_offset * 4), gb_tile_moden);
1874 }
1875 for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++) {
1876 switch (reg_offset) {
1877 case 0:
1878 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1879 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1880 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1881 NUM_BANKS(ADDR_SURF_16_BANK));
1882 break;
1883 case 1:
1884 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1885 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1886 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1887 NUM_BANKS(ADDR_SURF_16_BANK));
1888 break;
1889 case 2:
1890 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1891 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1892 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1893 NUM_BANKS(ADDR_SURF_16_BANK));
1894 break;
1895 case 3:
1896 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1897 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1898 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1899 NUM_BANKS(ADDR_SURF_16_BANK));
1900 break;
1901 case 4:
1902 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1903 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1904 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1905 NUM_BANKS(ADDR_SURF_8_BANK));
1906 break;
1907 case 5:
1908 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1909 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1910 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1911 NUM_BANKS(ADDR_SURF_4_BANK));
1912 break;
1913 case 6:
1914 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1915 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1916 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1917 NUM_BANKS(ADDR_SURF_2_BANK));
1918 break;
1919 case 8:
1920 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1921 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
1922 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1923 NUM_BANKS(ADDR_SURF_16_BANK));
1924 break;
1925 case 9:
1926 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1927 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1928 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1929 NUM_BANKS(ADDR_SURF_16_BANK));
1930 break;
1931 case 10:
1932 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1933 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1934 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1935 NUM_BANKS(ADDR_SURF_16_BANK));
1936 break;
1937 case 11:
1938 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1939 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1940 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1941 NUM_BANKS(ADDR_SURF_16_BANK));
1942 break;
1943 case 12:
1944 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1945 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1946 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1947 NUM_BANKS(ADDR_SURF_8_BANK));
1948 break;
1949 case 13:
1950 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1951 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1952 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1953 NUM_BANKS(ADDR_SURF_4_BANK));
1954 break;
1955 case 14:
1956 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1957 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1958 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1959 NUM_BANKS(ADDR_SURF_2_BANK));
1960 break;
1961 default:
1962 gb_tile_moden = 0;
1963 break;
1964 }
1965 WREG32(GB_MACROTILE_MODE0 + (reg_offset * 4), gb_tile_moden);
1966 }
1967 } else if (num_pipe_configs == 4) {
1968 if (num_rbs == 4) {
1969 for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) {
1970 switch (reg_offset) {
1971 case 0:
1972 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1973 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1974 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1975 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B));
1976 break;
1977 case 1:
1978 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1979 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1980 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1981 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B));
1982 break;
1983 case 2:
1984 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1985 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1986 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1987 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B));
1988 break;
1989 case 3:
1990 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1991 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1992 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1993 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B));
1994 break;
1995 case 4:
1996 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1997 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1998 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1999 TILE_SPLIT(split_equal_to_row_size));
2000 break;
2001 case 5:
2002 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2003 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2004 break;
2005 case 6:
2006 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2007 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2008 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2009 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B));
2010 break;
2011 case 7:
2012 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2013 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2014 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2015 TILE_SPLIT(split_equal_to_row_size));
2016 break;
2017 case 8:
2018 gb_tile_moden = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
2019 PIPE_CONFIG(ADDR_SURF_P4_16x16));
2020 break;
2021 case 9:
2022 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2023 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING));
2024 break;
2025 case 10:
2026 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2027 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2028 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2029 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2030 break;
2031 case 11:
2032 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2033 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2034 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
2035 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2036 break;
2037 case 12:
2038 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2039 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2040 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2041 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2042 break;
2043 case 13:
2044 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2045 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING));
2046 break;
2047 case 14:
2048 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2049 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2050 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2051 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2052 break;
2053 case 16:
2054 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2055 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2056 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
2057 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2058 break;
2059 case 17:
2060 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2061 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2062 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2063 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2064 break;
2065 case 27:
2066 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2067 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING));
2068 break;
2069 case 28:
2070 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2071 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2072 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2073 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2074 break;
2075 case 29:
2076 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2077 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2078 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
2079 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2080 break;
2081 case 30:
2082 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2083 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2084 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2085 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2086 break;
2087 default:
2088 gb_tile_moden = 0;
2089 break;
2090 }
2091 rdev->config.cik.tile_mode_array[reg_offset] = gb_tile_moden;
2092 WREG32(GB_TILE_MODE0 + (reg_offset * 4), gb_tile_moden);
2093 }
2094 } else if (num_rbs < 4) {
2095 for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) {
2096 switch (reg_offset) {
2097 case 0:
2098 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2099 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2100 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
2101 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B));
2102 break;
2103 case 1:
2104 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2105 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2106 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
2107 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B));
2108 break;
2109 case 2:
2110 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2111 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2112 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
2113 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B));
2114 break;
2115 case 3:
2116 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2117 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2118 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
2119 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B));
2120 break;
2121 case 4:
2122 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2123 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2124 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
2125 TILE_SPLIT(split_equal_to_row_size));
2126 break;
2127 case 5:
2128 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2129 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2130 break;
2131 case 6:
2132 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2133 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2134 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
2135 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B));
2136 break;
2137 case 7:
2138 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2139 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2140 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
2141 TILE_SPLIT(split_equal_to_row_size));
2142 break;
2143 case 8:
2144 gb_tile_moden = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
2145 PIPE_CONFIG(ADDR_SURF_P4_8x16));
2146 break;
2147 case 9:
2148 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2149 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING));
2150 break;
2151 case 10:
2152 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2153 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2154 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
2155 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2156 break;
2157 case 11:
2158 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2159 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2160 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
2161 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2162 break;
2163 case 12:
2164 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2165 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2166 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
2167 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2168 break;
2169 case 13:
2170 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2171 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING));
2172 break;
2173 case 14:
2174 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2175 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2176 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
2177 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2178 break;
2179 case 16:
2180 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2181 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2182 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
2183 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2184 break;
2185 case 17:
2186 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2187 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2188 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
2189 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2190 break;
2191 case 27:
2192 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2193 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING));
2194 break;
2195 case 28:
2196 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2197 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2198 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
2199 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2200 break;
2201 case 29:
2202 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2203 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2204 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
2205 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2206 break;
2207 case 30:
2208 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2209 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2210 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
2211 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2212 break;
2213 default:
2214 gb_tile_moden = 0;
2215 break;
2216 }
2217 rdev->config.cik.tile_mode_array[reg_offset] = gb_tile_moden;
2218 WREG32(GB_TILE_MODE0 + (reg_offset * 4), gb_tile_moden);
2219 }
2220 }
2221 for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++) {
2222 switch (reg_offset) {
2223 case 0:
2224 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2225 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2226 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2227 NUM_BANKS(ADDR_SURF_16_BANK));
2228 break;
2229 case 1:
2230 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2231 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2232 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2233 NUM_BANKS(ADDR_SURF_16_BANK));
2234 break;
2235 case 2:
2236 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2237 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2238 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2239 NUM_BANKS(ADDR_SURF_16_BANK));
2240 break;
2241 case 3:
2242 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2243 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2244 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2245 NUM_BANKS(ADDR_SURF_16_BANK));
2246 break;
2247 case 4:
2248 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2249 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2250 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2251 NUM_BANKS(ADDR_SURF_16_BANK));
2252 break;
2253 case 5:
2254 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2255 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2256 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2257 NUM_BANKS(ADDR_SURF_8_BANK));
2258 break;
2259 case 6:
2260 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2261 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2262 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2263 NUM_BANKS(ADDR_SURF_4_BANK));
2264 break;
2265 case 8:
2266 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
2267 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
2268 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2269 NUM_BANKS(ADDR_SURF_16_BANK));
2270 break;
2271 case 9:
2272 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
2273 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2274 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2275 NUM_BANKS(ADDR_SURF_16_BANK));
2276 break;
2277 case 10:
2278 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2279 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2280 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2281 NUM_BANKS(ADDR_SURF_16_BANK));
2282 break;
2283 case 11:
2284 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2285 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2286 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2287 NUM_BANKS(ADDR_SURF_16_BANK));
2288 break;
2289 case 12:
2290 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2291 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2292 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2293 NUM_BANKS(ADDR_SURF_16_BANK));
2294 break;
2295 case 13:
2296 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2297 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2298 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2299 NUM_BANKS(ADDR_SURF_8_BANK));
2300 break;
2301 case 14:
2302 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2303 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2304 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2305 NUM_BANKS(ADDR_SURF_4_BANK));
2306 break;
2307 default:
2308 gb_tile_moden = 0;
2309 break;
2310 }
2311 WREG32(GB_MACROTILE_MODE0 + (reg_offset * 4), gb_tile_moden);
2312 }
2313 } else if (num_pipe_configs == 2) {
2314 for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) {
2315 switch (reg_offset) {
2316 case 0:
2317 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2318 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2319 PIPE_CONFIG(ADDR_SURF_P2) |
2320 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B));
2321 break;
2322 case 1:
2323 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2324 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2325 PIPE_CONFIG(ADDR_SURF_P2) |
2326 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B));
2327 break;
2328 case 2:
2329 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2330 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2331 PIPE_CONFIG(ADDR_SURF_P2) |
2332 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B));
2333 break;
2334 case 3:
2335 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2336 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2337 PIPE_CONFIG(ADDR_SURF_P2) |
2338 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B));
2339 break;
2340 case 4:
2341 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2342 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2343 PIPE_CONFIG(ADDR_SURF_P2) |
2344 TILE_SPLIT(split_equal_to_row_size));
2345 break;
2346 case 5:
2347 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2348 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2349 break;
2350 case 6:
2351 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2352 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2353 PIPE_CONFIG(ADDR_SURF_P2) |
2354 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B));
2355 break;
2356 case 7:
2357 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2358 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2359 PIPE_CONFIG(ADDR_SURF_P2) |
2360 TILE_SPLIT(split_equal_to_row_size));
2361 break;
2362 case 8:
2363 gb_tile_moden = ARRAY_MODE(ARRAY_LINEAR_ALIGNED);
2364 break;
2365 case 9:
2366 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2367 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING));
2368 break;
2369 case 10:
2370 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2371 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2372 PIPE_CONFIG(ADDR_SURF_P2) |
2373 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2374 break;
2375 case 11:
2376 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2377 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2378 PIPE_CONFIG(ADDR_SURF_P2) |
2379 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2380 break;
2381 case 12:
2382 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2383 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2384 PIPE_CONFIG(ADDR_SURF_P2) |
2385 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2386 break;
2387 case 13:
2388 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2389 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING));
2390 break;
2391 case 14:
2392 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2393 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2394 PIPE_CONFIG(ADDR_SURF_P2) |
2395 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2396 break;
2397 case 16:
2398 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2399 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2400 PIPE_CONFIG(ADDR_SURF_P2) |
2401 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2402 break;
2403 case 17:
2404 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2405 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2406 PIPE_CONFIG(ADDR_SURF_P2) |
2407 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2408 break;
2409 case 27:
2410 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2411 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING));
2412 break;
2413 case 28:
2414 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2415 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2416 PIPE_CONFIG(ADDR_SURF_P2) |
2417 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2418 break;
2419 case 29:
2420 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2421 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2422 PIPE_CONFIG(ADDR_SURF_P2) |
2423 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2424 break;
2425 case 30:
2426 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2427 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2428 PIPE_CONFIG(ADDR_SURF_P2) |
2429 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2430 break;
2431 default:
2432 gb_tile_moden = 0;
2433 break;
2434 }
2435 rdev->config.cik.tile_mode_array[reg_offset] = gb_tile_moden;
2436 WREG32(GB_TILE_MODE0 + (reg_offset * 4), gb_tile_moden);
2437 }
2438 for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++) {
2439 switch (reg_offset) {
2440 case 0:
2441 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
2442 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2443 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2444 NUM_BANKS(ADDR_SURF_16_BANK));
2445 break;
2446 case 1:
2447 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
2448 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2449 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2450 NUM_BANKS(ADDR_SURF_16_BANK));
2451 break;
2452 case 2:
2453 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2454 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2455 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2456 NUM_BANKS(ADDR_SURF_16_BANK));
2457 break;
2458 case 3:
2459 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2460 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2461 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2462 NUM_BANKS(ADDR_SURF_16_BANK));
2463 break;
2464 case 4:
2465 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2466 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2467 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2468 NUM_BANKS(ADDR_SURF_16_BANK));
2469 break;
2470 case 5:
2471 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2472 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2473 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2474 NUM_BANKS(ADDR_SURF_16_BANK));
2475 break;
2476 case 6:
2477 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2478 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2479 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2480 NUM_BANKS(ADDR_SURF_8_BANK));
2481 break;
2482 case 8:
2483 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
2484 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
2485 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2486 NUM_BANKS(ADDR_SURF_16_BANK));
2487 break;
2488 case 9:
2489 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
2490 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2491 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2492 NUM_BANKS(ADDR_SURF_16_BANK));
2493 break;
2494 case 10:
2495 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
2496 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2497 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2498 NUM_BANKS(ADDR_SURF_16_BANK));
2499 break;
2500 case 11:
2501 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
2502 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2503 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2504 NUM_BANKS(ADDR_SURF_16_BANK));
2505 break;
2506 case 12:
2507 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2508 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2509 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2510 NUM_BANKS(ADDR_SURF_16_BANK));
2511 break;
2512 case 13:
2513 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2514 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2515 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2516 NUM_BANKS(ADDR_SURF_16_BANK));
2517 break;
2518 case 14:
2519 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2520 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2521 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2522 NUM_BANKS(ADDR_SURF_8_BANK));
2523 break;
2524 default:
2525 gb_tile_moden = 0;
2526 break;
2527 }
2528 WREG32(GB_MACROTILE_MODE0 + (reg_offset * 4), gb_tile_moden);
2529 }
2530 } else
2531 DRM_ERROR("unknown num pipe config: 0x%x\n", num_pipe_configs);
2532 }
2533
2534 /**
2535 * cik_select_se_sh - select which SE, SH to address
2536 *
2537 * @rdev: radeon_device pointer
2538 * @se_num: shader engine to address
2539 * @sh_num: sh block to address
2540 *
2541 * Select which SE, SH combinations to address. Certain
2542 * registers are instanced per SE or SH. 0xffffffff means
2543 * broadcast to all SEs or SHs (CIK).
2544 */
2545 static void cik_select_se_sh(struct radeon_device *rdev,
2546 u32 se_num, u32 sh_num)
2547 {
2548 u32 data = INSTANCE_BROADCAST_WRITES;
2549
2550 if ((se_num == 0xffffffff) && (sh_num == 0xffffffff))
2551 data |= SH_BROADCAST_WRITES | SE_BROADCAST_WRITES;
2552 else if (se_num == 0xffffffff)
2553 data |= SE_BROADCAST_WRITES | SH_INDEX(sh_num);
2554 else if (sh_num == 0xffffffff)
2555 data |= SH_BROADCAST_WRITES | SE_INDEX(se_num);
2556 else
2557 data |= SH_INDEX(sh_num) | SE_INDEX(se_num);
2558 WREG32(GRBM_GFX_INDEX, data);
2559 }
2560
2561 /**
2562 * cik_create_bitmask - create a bitmask
2563 *
2564 * @bit_width: length of the mask
2565 *
2566 * create a variable length bit mask (CIK).
2567 * Returns the bitmask.
2568 */
2569 static u32 cik_create_bitmask(u32 bit_width)
2570 {
2571 u32 i, mask = 0;
2572
2573 for (i = 0; i < bit_width; i++) {
2574 mask <<= 1;
2575 mask |= 1;
2576 }
2577 return mask;
2578 }
2579
2580 /**
2581 * cik_select_se_sh - select which SE, SH to address
2582 *
2583 * @rdev: radeon_device pointer
2584 * @max_rb_num: max RBs (render backends) for the asic
2585 * @se_num: number of SEs (shader engines) for the asic
2586 * @sh_per_se: number of SH blocks per SE for the asic
2587 *
2588 * Calculates the bitmask of disabled RBs (CIK).
2589 * Returns the disabled RB bitmask.
2590 */
2591 static u32 cik_get_rb_disabled(struct radeon_device *rdev,
2592 u32 max_rb_num, u32 se_num,
2593 u32 sh_per_se)
2594 {
2595 u32 data, mask;
2596
2597 data = RREG32(CC_RB_BACKEND_DISABLE);
2598 if (data & 1)
2599 data &= BACKEND_DISABLE_MASK;
2600 else
2601 data = 0;
2602 data |= RREG32(GC_USER_RB_BACKEND_DISABLE);
2603
2604 data >>= BACKEND_DISABLE_SHIFT;
2605
2606 mask = cik_create_bitmask(max_rb_num / se_num / sh_per_se);
2607
2608 return data & mask;
2609 }
2610
2611 /**
2612 * cik_setup_rb - setup the RBs on the asic
2613 *
2614 * @rdev: radeon_device pointer
2615 * @se_num: number of SEs (shader engines) for the asic
2616 * @sh_per_se: number of SH blocks per SE for the asic
2617 * @max_rb_num: max RBs (render backends) for the asic
2618 *
2619 * Configures per-SE/SH RB registers (CIK).
2620 */
2621 static void cik_setup_rb(struct radeon_device *rdev,
2622 u32 se_num, u32 sh_per_se,
2623 u32 max_rb_num)
2624 {
2625 int i, j;
2626 u32 data, mask;
2627 u32 disabled_rbs = 0;
2628 u32 enabled_rbs = 0;
2629
2630 for (i = 0; i < se_num; i++) {
2631 for (j = 0; j < sh_per_se; j++) {
2632 cik_select_se_sh(rdev, i, j);
2633 data = cik_get_rb_disabled(rdev, max_rb_num, se_num, sh_per_se);
2634 disabled_rbs |= data << ((i * sh_per_se + j) * CIK_RB_BITMAP_WIDTH_PER_SH);
2635 }
2636 }
2637 cik_select_se_sh(rdev, 0xffffffff, 0xffffffff);
2638
2639 mask = 1;
2640 for (i = 0; i < max_rb_num; i++) {
2641 if (!(disabled_rbs & mask))
2642 enabled_rbs |= mask;
2643 mask <<= 1;
2644 }
2645
2646 for (i = 0; i < se_num; i++) {
2647 cik_select_se_sh(rdev, i, 0xffffffff);
2648 data = 0;
2649 for (j = 0; j < sh_per_se; j++) {
2650 switch (enabled_rbs & 3) {
2651 case 1:
2652 data |= (RASTER_CONFIG_RB_MAP_0 << (i * sh_per_se + j) * 2);
2653 break;
2654 case 2:
2655 data |= (RASTER_CONFIG_RB_MAP_3 << (i * sh_per_se + j) * 2);
2656 break;
2657 case 3:
2658 default:
2659 data |= (RASTER_CONFIG_RB_MAP_2 << (i * sh_per_se + j) * 2);
2660 break;
2661 }
2662 enabled_rbs >>= 2;
2663 }
2664 WREG32(PA_SC_RASTER_CONFIG, data);
2665 }
2666 cik_select_se_sh(rdev, 0xffffffff, 0xffffffff);
2667 }
2668
2669 /**
2670 * cik_gpu_init - setup the 3D engine
2671 *
2672 * @rdev: radeon_device pointer
2673 *
2674 * Configures the 3D engine and tiling configuration
2675 * registers so that the 3D engine is usable.
2676 */
2677 static void cik_gpu_init(struct radeon_device *rdev)
2678 {
2679 u32 gb_addr_config = RREG32(GB_ADDR_CONFIG);
2680 u32 mc_shared_chmap, mc_arb_ramcfg;
2681 u32 hdp_host_path_cntl;
2682 u32 tmp;
2683 int i, j;
2684
2685 switch (rdev->family) {
2686 case CHIP_BONAIRE:
2687 rdev->config.cik.max_shader_engines = 2;
2688 rdev->config.cik.max_tile_pipes = 4;
2689 rdev->config.cik.max_cu_per_sh = 7;
2690 rdev->config.cik.max_sh_per_se = 1;
2691 rdev->config.cik.max_backends_per_se = 2;
2692 rdev->config.cik.max_texture_channel_caches = 4;
2693 rdev->config.cik.max_gprs = 256;
2694 rdev->config.cik.max_gs_threads = 32;
2695 rdev->config.cik.max_hw_contexts = 8;
2696
2697 rdev->config.cik.sc_prim_fifo_size_frontend = 0x20;
2698 rdev->config.cik.sc_prim_fifo_size_backend = 0x100;
2699 rdev->config.cik.sc_hiz_tile_fifo_size = 0x30;
2700 rdev->config.cik.sc_earlyz_tile_fifo_size = 0x130;
2701 gb_addr_config = BONAIRE_GB_ADDR_CONFIG_GOLDEN;
2702 break;
2703 case CHIP_KAVERI:
2704 /* TODO */
2705 break;
2706 case CHIP_KABINI:
2707 default:
2708 rdev->config.cik.max_shader_engines = 1;
2709 rdev->config.cik.max_tile_pipes = 2;
2710 rdev->config.cik.max_cu_per_sh = 2;
2711 rdev->config.cik.max_sh_per_se = 1;
2712 rdev->config.cik.max_backends_per_se = 1;
2713 rdev->config.cik.max_texture_channel_caches = 2;
2714 rdev->config.cik.max_gprs = 256;
2715 rdev->config.cik.max_gs_threads = 16;
2716 rdev->config.cik.max_hw_contexts = 8;
2717
2718 rdev->config.cik.sc_prim_fifo_size_frontend = 0x20;
2719 rdev->config.cik.sc_prim_fifo_size_backend = 0x100;
2720 rdev->config.cik.sc_hiz_tile_fifo_size = 0x30;
2721 rdev->config.cik.sc_earlyz_tile_fifo_size = 0x130;
2722 gb_addr_config = BONAIRE_GB_ADDR_CONFIG_GOLDEN;
2723 break;
2724 }
2725
2726 /* Initialize HDP */
2727 for (i = 0, j = 0; i < 32; i++, j += 0x18) {
2728 WREG32((0x2c14 + j), 0x00000000);
2729 WREG32((0x2c18 + j), 0x00000000);
2730 WREG32((0x2c1c + j), 0x00000000);
2731 WREG32((0x2c20 + j), 0x00000000);
2732 WREG32((0x2c24 + j), 0x00000000);
2733 }
2734
2735 WREG32(GRBM_CNTL, GRBM_READ_TIMEOUT(0xff));
2736
2737 WREG32(BIF_FB_EN, FB_READ_EN | FB_WRITE_EN);
2738
2739 mc_shared_chmap = RREG32(MC_SHARED_CHMAP);
2740 mc_arb_ramcfg = RREG32(MC_ARB_RAMCFG);
2741
2742 rdev->config.cik.num_tile_pipes = rdev->config.cik.max_tile_pipes;
2743 rdev->config.cik.mem_max_burst_length_bytes = 256;
2744 tmp = (mc_arb_ramcfg & NOOFCOLS_MASK) >> NOOFCOLS_SHIFT;
2745 rdev->config.cik.mem_row_size_in_kb = (4 * (1 << (8 + tmp))) / 1024;
2746 if (rdev->config.cik.mem_row_size_in_kb > 4)
2747 rdev->config.cik.mem_row_size_in_kb = 4;
2748 /* XXX use MC settings? */
2749 rdev->config.cik.shader_engine_tile_size = 32;
2750 rdev->config.cik.num_gpus = 1;
2751 rdev->config.cik.multi_gpu_tile_size = 64;
2752
2753 /* fix up row size */
2754 gb_addr_config &= ~ROW_SIZE_MASK;
2755 switch (rdev->config.cik.mem_row_size_in_kb) {
2756 case 1:
2757 default:
2758 gb_addr_config |= ROW_SIZE(0);
2759 break;
2760 case 2:
2761 gb_addr_config |= ROW_SIZE(1);
2762 break;
2763 case 4:
2764 gb_addr_config |= ROW_SIZE(2);
2765 break;
2766 }
2767
2768 /* setup tiling info dword. gb_addr_config is not adequate since it does
2769 * not have bank info, so create a custom tiling dword.
2770 * bits 3:0 num_pipes
2771 * bits 7:4 num_banks
2772 * bits 11:8 group_size
2773 * bits 15:12 row_size
2774 */
2775 rdev->config.cik.tile_config = 0;
2776 switch (rdev->config.cik.num_tile_pipes) {
2777 case 1:
2778 rdev->config.cik.tile_config |= (0 << 0);
2779 break;
2780 case 2:
2781 rdev->config.cik.tile_config |= (1 << 0);
2782 break;
2783 case 4:
2784 rdev->config.cik.tile_config |= (2 << 0);
2785 break;
2786 case 8:
2787 default:
2788 /* XXX what about 12? */
2789 rdev->config.cik.tile_config |= (3 << 0);
2790 break;
2791 }
2792 if ((mc_arb_ramcfg & NOOFBANK_MASK) >> NOOFBANK_SHIFT)
2793 rdev->config.cik.tile_config |= 1 << 4;
2794 else
2795 rdev->config.cik.tile_config |= 0 << 4;
2796 rdev->config.cik.tile_config |=
2797 ((gb_addr_config & PIPE_INTERLEAVE_SIZE_MASK) >> PIPE_INTERLEAVE_SIZE_SHIFT) << 8;
2798 rdev->config.cik.tile_config |=
2799 ((gb_addr_config & ROW_SIZE_MASK) >> ROW_SIZE_SHIFT) << 12;
2800
2801 WREG32(GB_ADDR_CONFIG, gb_addr_config);
2802 WREG32(HDP_ADDR_CONFIG, gb_addr_config);
2803 WREG32(DMIF_ADDR_CALC, gb_addr_config);
2804 WREG32(SDMA0_TILING_CONFIG + SDMA0_REGISTER_OFFSET, gb_addr_config & 0x70);
2805 WREG32(SDMA0_TILING_CONFIG + SDMA1_REGISTER_OFFSET, gb_addr_config & 0x70);
2806 WREG32(UVD_UDEC_ADDR_CONFIG, gb_addr_config);
2807 WREG32(UVD_UDEC_DB_ADDR_CONFIG, gb_addr_config);
2808 WREG32(UVD_UDEC_DBW_ADDR_CONFIG, gb_addr_config);
2809
2810 cik_tiling_mode_table_init(rdev);
2811
2812 cik_setup_rb(rdev, rdev->config.cik.max_shader_engines,
2813 rdev->config.cik.max_sh_per_se,
2814 rdev->config.cik.max_backends_per_se);
2815
2816 /* set HW defaults for 3D engine */
2817 WREG32(CP_MEQ_THRESHOLDS, MEQ1_START(0x30) | MEQ2_START(0x60));
2818
2819 WREG32(SX_DEBUG_1, 0x20);
2820
2821 WREG32(TA_CNTL_AUX, 0x00010000);
2822
2823 tmp = RREG32(SPI_CONFIG_CNTL);
2824 tmp |= 0x03000000;
2825 WREG32(SPI_CONFIG_CNTL, tmp);
2826
2827 WREG32(SQ_CONFIG, 1);
2828
2829 WREG32(DB_DEBUG, 0);
2830
2831 tmp = RREG32(DB_DEBUG2) & ~0xf00fffff;
2832 tmp |= 0x00000400;
2833 WREG32(DB_DEBUG2, tmp);
2834
2835 tmp = RREG32(DB_DEBUG3) & ~0x0002021c;
2836 tmp |= 0x00020200;
2837 WREG32(DB_DEBUG3, tmp);
2838
2839 tmp = RREG32(CB_HW_CONTROL) & ~0x00010000;
2840 tmp |= 0x00018208;
2841 WREG32(CB_HW_CONTROL, tmp);
2842
2843 WREG32(SPI_CONFIG_CNTL_1, VTX_DONE_DELAY(4));
2844
2845 WREG32(PA_SC_FIFO_SIZE, (SC_FRONTEND_PRIM_FIFO_SIZE(rdev->config.cik.sc_prim_fifo_size_frontend) |
2846 SC_BACKEND_PRIM_FIFO_SIZE(rdev->config.cik.sc_prim_fifo_size_backend) |
2847 SC_HIZ_TILE_FIFO_SIZE(rdev->config.cik.sc_hiz_tile_fifo_size) |
2848 SC_EARLYZ_TILE_FIFO_SIZE(rdev->config.cik.sc_earlyz_tile_fifo_size)));
2849
2850 WREG32(VGT_NUM_INSTANCES, 1);
2851
2852 WREG32(CP_PERFMON_CNTL, 0);
2853
2854 WREG32(SQ_CONFIG, 0);
2855
2856 WREG32(PA_SC_FORCE_EOV_MAX_CNTS, (FORCE_EOV_MAX_CLK_CNT(4095) |
2857 FORCE_EOV_MAX_REZ_CNT(255)));
2858
2859 WREG32(VGT_CACHE_INVALIDATION, CACHE_INVALIDATION(VC_AND_TC) |
2860 AUTO_INVLD_EN(ES_AND_GS_AUTO));
2861
2862 WREG32(VGT_GS_VERTEX_REUSE, 16);
2863 WREG32(PA_SC_LINE_STIPPLE_STATE, 0);
2864
2865 tmp = RREG32(HDP_MISC_CNTL);
2866 tmp |= HDP_FLUSH_INVALIDATE_CACHE;
2867 WREG32(HDP_MISC_CNTL, tmp);
2868
2869 hdp_host_path_cntl = RREG32(HDP_HOST_PATH_CNTL);
2870 WREG32(HDP_HOST_PATH_CNTL, hdp_host_path_cntl);
2871
2872 WREG32(PA_CL_ENHANCE, CLIP_VTX_REORDER_ENA | NUM_CLIP_SEQ(3));
2873 WREG32(PA_SC_ENHANCE, ENABLE_PA_SC_OUT_OF_ORDER);
2874
2875 udelay(50);
2876 }
2877
2878 /*
2879 * GPU scratch registers helpers function.
2880 */
2881 /**
2882 * cik_scratch_init - setup driver info for CP scratch regs
2883 *
2884 * @rdev: radeon_device pointer
2885 *
2886 * Set up the number and offset of the CP scratch registers.
2887 * NOTE: use of CP scratch registers is a legacy inferface and
2888 * is not used by default on newer asics (r6xx+). On newer asics,
2889 * memory buffers are used for fences rather than scratch regs.
2890 */
2891 static void cik_scratch_init(struct radeon_device *rdev)
2892 {
2893 int i;
2894
2895 rdev->scratch.num_reg = 7;
2896 rdev->scratch.reg_base = SCRATCH_REG0;
2897 for (i = 0; i < rdev->scratch.num_reg; i++) {
2898 rdev->scratch.free[i] = true;
2899 rdev->scratch.reg[i] = rdev->scratch.reg_base + (i * 4);
2900 }
2901 }
2902
2903 /**
2904 * cik_ring_test - basic gfx ring test
2905 *
2906 * @rdev: radeon_device pointer
2907 * @ring: radeon_ring structure holding ring information
2908 *
2909 * Allocate a scratch register and write to it using the gfx ring (CIK).
2910 * Provides a basic gfx ring test to verify that the ring is working.
2911 * Used by cik_cp_gfx_resume();
2912 * Returns 0 on success, error on failure.
2913 */
2914 int cik_ring_test(struct radeon_device *rdev, struct radeon_ring *ring)
2915 {
2916 uint32_t scratch;
2917 uint32_t tmp = 0;
2918 unsigned i;
2919 int r;
2920
2921 r = radeon_scratch_get(rdev, &scratch);
2922 if (r) {
2923 DRM_ERROR("radeon: cp failed to get scratch reg (%d).\n", r);
2924 return r;
2925 }
2926 WREG32(scratch, 0xCAFEDEAD);
2927 r = radeon_ring_lock(rdev, ring, 3);
2928 if (r) {
2929 DRM_ERROR("radeon: cp failed to lock ring %d (%d).\n", ring->idx, r);
2930 radeon_scratch_free(rdev, scratch);
2931 return r;
2932 }
2933 radeon_ring_write(ring, PACKET3(PACKET3_SET_UCONFIG_REG, 1));
2934 radeon_ring_write(ring, ((scratch - PACKET3_SET_UCONFIG_REG_START) >> 2));
2935 radeon_ring_write(ring, 0xDEADBEEF);
2936 radeon_ring_unlock_commit(rdev, ring);
2937
2938 for (i = 0; i < rdev->usec_timeout; i++) {
2939 tmp = RREG32(scratch);
2940 if (tmp == 0xDEADBEEF)
2941 break;
2942 DRM_UDELAY(1);
2943 }
2944 if (i < rdev->usec_timeout) {
2945 DRM_INFO("ring test on %d succeeded in %d usecs\n", ring->idx, i);
2946 } else {
2947 DRM_ERROR("radeon: ring %d test failed (scratch(0x%04X)=0x%08X)\n",
2948 ring->idx, scratch, tmp);
2949 r = -EINVAL;
2950 }
2951 radeon_scratch_free(rdev, scratch);
2952 return r;
2953 }
2954
2955 /**
2956 * cik_fence_gfx_ring_emit - emit a fence on the gfx ring
2957 *
2958 * @rdev: radeon_device pointer
2959 * @fence: radeon fence object
2960 *
2961 * Emits a fence sequnce number on the gfx ring and flushes
2962 * GPU caches.
2963 */
2964 void cik_fence_gfx_ring_emit(struct radeon_device *rdev,
2965 struct radeon_fence *fence)
2966 {
2967 struct radeon_ring *ring = &rdev->ring[fence->ring];
2968 u64 addr = rdev->fence_drv[fence->ring].gpu_addr;
2969
2970 /* EVENT_WRITE_EOP - flush caches, send int */
2971 radeon_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE_EOP, 4));
2972 radeon_ring_write(ring, (EOP_TCL1_ACTION_EN |
2973 EOP_TC_ACTION_EN |
2974 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
2975 EVENT_INDEX(5)));
2976 radeon_ring_write(ring, addr & 0xfffffffc);
2977 radeon_ring_write(ring, (upper_32_bits(addr) & 0xffff) | DATA_SEL(1) | INT_SEL(2));
2978 radeon_ring_write(ring, fence->seq);
2979 radeon_ring_write(ring, 0);
2980 /* HDP flush */
2981 /* We should be using the new WAIT_REG_MEM special op packet here
2982 * but it causes the CP to hang
2983 */
2984 radeon_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
2985 radeon_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
2986 WRITE_DATA_DST_SEL(0)));
2987 radeon_ring_write(ring, HDP_MEM_COHERENCY_FLUSH_CNTL >> 2);
2988 radeon_ring_write(ring, 0);
2989 radeon_ring_write(ring, 0);
2990 }
2991
2992 /**
2993 * cik_fence_compute_ring_emit - emit a fence on the compute ring
2994 *
2995 * @rdev: radeon_device pointer
2996 * @fence: radeon fence object
2997 *
2998 * Emits a fence sequnce number on the compute ring and flushes
2999 * GPU caches.
3000 */
3001 void cik_fence_compute_ring_emit(struct radeon_device *rdev,
3002 struct radeon_fence *fence)
3003 {
3004 struct radeon_ring *ring = &rdev->ring[fence->ring];
3005 u64 addr = rdev->fence_drv[fence->ring].gpu_addr;
3006
3007 /* RELEASE_MEM - flush caches, send int */
3008 radeon_ring_write(ring, PACKET3(PACKET3_RELEASE_MEM, 5));
3009 radeon_ring_write(ring, (EOP_TCL1_ACTION_EN |
3010 EOP_TC_ACTION_EN |
3011 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
3012 EVENT_INDEX(5)));
3013 radeon_ring_write(ring, DATA_SEL(1) | INT_SEL(2));
3014 radeon_ring_write(ring, addr & 0xfffffffc);
3015 radeon_ring_write(ring, upper_32_bits(addr));
3016 radeon_ring_write(ring, fence->seq);
3017 radeon_ring_write(ring, 0);
3018 /* HDP flush */
3019 /* We should be using the new WAIT_REG_MEM special op packet here
3020 * but it causes the CP to hang
3021 */
3022 radeon_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
3023 radeon_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
3024 WRITE_DATA_DST_SEL(0)));
3025 radeon_ring_write(ring, HDP_MEM_COHERENCY_FLUSH_CNTL >> 2);
3026 radeon_ring_write(ring, 0);
3027 radeon_ring_write(ring, 0);
3028 }
3029
3030 void cik_semaphore_ring_emit(struct radeon_device *rdev,
3031 struct radeon_ring *ring,
3032 struct radeon_semaphore *semaphore,
3033 bool emit_wait)
3034 {
3035 uint64_t addr = semaphore->gpu_addr;
3036 unsigned sel = emit_wait ? PACKET3_SEM_SEL_WAIT : PACKET3_SEM_SEL_SIGNAL;
3037
3038 radeon_ring_write(ring, PACKET3(PACKET3_MEM_SEMAPHORE, 1));
3039 radeon_ring_write(ring, addr & 0xffffffff);
3040 radeon_ring_write(ring, (upper_32_bits(addr) & 0xffff) | sel);
3041 }
3042
3043 /*
3044 * IB stuff
3045 */
3046 /**
3047 * cik_ring_ib_execute - emit an IB (Indirect Buffer) on the gfx ring
3048 *
3049 * @rdev: radeon_device pointer
3050 * @ib: radeon indirect buffer object
3051 *
3052 * Emits an DE (drawing engine) or CE (constant engine) IB
3053 * on the gfx ring. IBs are usually generated by userspace
3054 * acceleration drivers and submitted to the kernel for
3055 * sheduling on the ring. This function schedules the IB
3056 * on the gfx ring for execution by the GPU.
3057 */
3058 void cik_ring_ib_execute(struct radeon_device *rdev, struct radeon_ib *ib)
3059 {
3060 struct radeon_ring *ring = &rdev->ring[ib->ring];
3061 u32 header, control = INDIRECT_BUFFER_VALID;
3062
3063 if (ib->is_const_ib) {
3064 /* set switch buffer packet before const IB */
3065 radeon_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
3066 radeon_ring_write(ring, 0);
3067
3068 header = PACKET3(PACKET3_INDIRECT_BUFFER_CONST, 2);
3069 } else {
3070 u32 next_rptr;
3071 if (ring->rptr_save_reg) {
3072 next_rptr = ring->wptr + 3 + 4;
3073 radeon_ring_write(ring, PACKET3(PACKET3_SET_UCONFIG_REG, 1));
3074 radeon_ring_write(ring, ((ring->rptr_save_reg -
3075 PACKET3_SET_UCONFIG_REG_START) >> 2));
3076 radeon_ring_write(ring, next_rptr);
3077 } else if (rdev->wb.enabled) {
3078 next_rptr = ring->wptr + 5 + 4;
3079 radeon_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
3080 radeon_ring_write(ring, WRITE_DATA_DST_SEL(1));
3081 radeon_ring_write(ring, ring->next_rptr_gpu_addr & 0xfffffffc);
3082 radeon_ring_write(ring, upper_32_bits(ring->next_rptr_gpu_addr) & 0xffffffff);
3083 radeon_ring_write(ring, next_rptr);
3084 }
3085
3086 header = PACKET3(PACKET3_INDIRECT_BUFFER, 2);
3087 }
3088
3089 control |= ib->length_dw |
3090 (ib->vm ? (ib->vm->id << 24) : 0);
3091
3092 radeon_ring_write(ring, header);
3093 radeon_ring_write(ring,
3094 #ifdef __BIG_ENDIAN
3095 (2 << 0) |
3096 #endif
3097 (ib->gpu_addr & 0xFFFFFFFC));
3098 radeon_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xFFFF);
3099 radeon_ring_write(ring, control);
3100 }
3101
3102 /**
3103 * cik_ib_test - basic gfx ring IB test
3104 *
3105 * @rdev: radeon_device pointer
3106 * @ring: radeon_ring structure holding ring information
3107 *
3108 * Allocate an IB and execute it on the gfx ring (CIK).
3109 * Provides a basic gfx ring test to verify that IBs are working.
3110 * Returns 0 on success, error on failure.
3111 */
3112 int cik_ib_test(struct radeon_device *rdev, struct radeon_ring *ring)
3113 {
3114 struct radeon_ib ib;
3115 uint32_t scratch;
3116 uint32_t tmp = 0;
3117 unsigned i;
3118 int r;
3119
3120 r = radeon_scratch_get(rdev, &scratch);
3121 if (r) {
3122 DRM_ERROR("radeon: failed to get scratch reg (%d).\n", r);
3123 return r;
3124 }
3125 WREG32(scratch, 0xCAFEDEAD);
3126 r = radeon_ib_get(rdev, ring->idx, &ib, NULL, 256);
3127 if (r) {
3128 DRM_ERROR("radeon: failed to get ib (%d).\n", r);
3129 return r;
3130 }
3131 ib.ptr[0] = PACKET3(PACKET3_SET_UCONFIG_REG, 1);
3132 ib.ptr[1] = ((scratch - PACKET3_SET_UCONFIG_REG_START) >> 2);
3133 ib.ptr[2] = 0xDEADBEEF;
3134 ib.length_dw = 3;
3135 r = radeon_ib_schedule(rdev, &ib, NULL);
3136 if (r) {
3137 radeon_scratch_free(rdev, scratch);
3138 radeon_ib_free(rdev, &ib);
3139 DRM_ERROR("radeon: failed to schedule ib (%d).\n", r);
3140 return r;
3141 }
3142 r = radeon_fence_wait(ib.fence, false);
3143 if (r) {
3144 DRM_ERROR("radeon: fence wait failed (%d).\n", r);
3145 return r;
3146 }
3147 for (i = 0; i < rdev->usec_timeout; i++) {
3148 tmp = RREG32(scratch);
3149 if (tmp == 0xDEADBEEF)
3150 break;
3151 DRM_UDELAY(1);
3152 }
3153 if (i < rdev->usec_timeout) {
3154 DRM_INFO("ib test on ring %d succeeded in %u usecs\n", ib.fence->ring, i);
3155 } else {
3156 DRM_ERROR("radeon: ib test failed (scratch(0x%04X)=0x%08X)\n",
3157 scratch, tmp);
3158 r = -EINVAL;
3159 }
3160 radeon_scratch_free(rdev, scratch);
3161 radeon_ib_free(rdev, &ib);
3162 return r;
3163 }
3164
3165 /*
3166 * CP.
3167 * On CIK, gfx and compute now have independant command processors.
3168 *
3169 * GFX
3170 * Gfx consists of a single ring and can process both gfx jobs and
3171 * compute jobs. The gfx CP consists of three microengines (ME):
3172 * PFP - Pre-Fetch Parser
3173 * ME - Micro Engine
3174 * CE - Constant Engine
3175 * The PFP and ME make up what is considered the Drawing Engine (DE).
3176 * The CE is an asynchronous engine used for updating buffer desciptors
3177 * used by the DE so that they can be loaded into cache in parallel
3178 * while the DE is processing state update packets.
3179 *
3180 * Compute
3181 * The compute CP consists of two microengines (ME):
3182 * MEC1 - Compute MicroEngine 1
3183 * MEC2 - Compute MicroEngine 2
3184 * Each MEC supports 4 compute pipes and each pipe supports 8 queues.
3185 * The queues are exposed to userspace and are programmed directly
3186 * by the compute runtime.
3187 */
3188 /**
3189 * cik_cp_gfx_enable - enable/disable the gfx CP MEs
3190 *
3191 * @rdev: radeon_device pointer
3192 * @enable: enable or disable the MEs
3193 *
3194 * Halts or unhalts the gfx MEs.
3195 */
3196 static void cik_cp_gfx_enable(struct radeon_device *rdev, bool enable)
3197 {
3198 if (enable)
3199 WREG32(CP_ME_CNTL, 0);
3200 else {
3201 WREG32(CP_ME_CNTL, (CP_ME_HALT | CP_PFP_HALT | CP_CE_HALT));
3202 rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = false;
3203 }
3204 udelay(50);
3205 }
3206
3207 /**
3208 * cik_cp_gfx_load_microcode - load the gfx CP ME ucode
3209 *
3210 * @rdev: radeon_device pointer
3211 *
3212 * Loads the gfx PFP, ME, and CE ucode.
3213 * Returns 0 for success, -EINVAL if the ucode is not available.
3214 */
3215 static int cik_cp_gfx_load_microcode(struct radeon_device *rdev)
3216 {
3217 const __be32 *fw_data;
3218 int i;
3219
3220 if (!rdev->me_fw || !rdev->pfp_fw || !rdev->ce_fw)
3221 return -EINVAL;
3222
3223 cik_cp_gfx_enable(rdev, false);
3224
3225 /* PFP */
3226 fw_data = (const __be32 *)rdev->pfp_fw->data;
3227 WREG32(CP_PFP_UCODE_ADDR, 0);
3228 for (i = 0; i < CIK_PFP_UCODE_SIZE; i++)
3229 WREG32(CP_PFP_UCODE_DATA, be32_to_cpup(fw_data++));
3230 WREG32(CP_PFP_UCODE_ADDR, 0);
3231
3232 /* CE */
3233 fw_data = (const __be32 *)rdev->ce_fw->data;
3234 WREG32(CP_CE_UCODE_ADDR, 0);
3235 for (i = 0; i < CIK_CE_UCODE_SIZE; i++)
3236 WREG32(CP_CE_UCODE_DATA, be32_to_cpup(fw_data++));
3237 WREG32(CP_CE_UCODE_ADDR, 0);
3238
3239 /* ME */
3240 fw_data = (const __be32 *)rdev->me_fw->data;
3241 WREG32(CP_ME_RAM_WADDR, 0);
3242 for (i = 0; i < CIK_ME_UCODE_SIZE; i++)
3243 WREG32(CP_ME_RAM_DATA, be32_to_cpup(fw_data++));
3244 WREG32(CP_ME_RAM_WADDR, 0);
3245
3246 WREG32(CP_PFP_UCODE_ADDR, 0);
3247 WREG32(CP_CE_UCODE_ADDR, 0);
3248 WREG32(CP_ME_RAM_WADDR, 0);
3249 WREG32(CP_ME_RAM_RADDR, 0);
3250 return 0;
3251 }
3252
3253 /**
3254 * cik_cp_gfx_start - start the gfx ring
3255 *
3256 * @rdev: radeon_device pointer
3257 *
3258 * Enables the ring and loads the clear state context and other
3259 * packets required to init the ring.
3260 * Returns 0 for success, error for failure.
3261 */
3262 static int cik_cp_gfx_start(struct radeon_device *rdev)
3263 {
3264 struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
3265 int r, i;
3266
3267 /* init the CP */
3268 WREG32(CP_MAX_CONTEXT, rdev->config.cik.max_hw_contexts - 1);
3269 WREG32(CP_ENDIAN_SWAP, 0);
3270 WREG32(CP_DEVICE_ID, 1);
3271
3272 cik_cp_gfx_enable(rdev, true);
3273
3274 r = radeon_ring_lock(rdev, ring, cik_default_size + 17);
3275 if (r) {
3276 DRM_ERROR("radeon: cp failed to lock ring (%d).\n", r);
3277 return r;
3278 }
3279
3280 /* init the CE partitions. CE only used for gfx on CIK */
3281 radeon_ring_write(ring, PACKET3(PACKET3_SET_BASE, 2));
3282 radeon_ring_write(ring, PACKET3_BASE_INDEX(CE_PARTITION_BASE));
3283 radeon_ring_write(ring, 0xc000);
3284 radeon_ring_write(ring, 0xc000);
3285
3286 /* setup clear context state */
3287 radeon_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
3288 radeon_ring_write(ring, PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
3289
3290 radeon_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1));
3291 radeon_ring_write(ring, 0x80000000);
3292 radeon_ring_write(ring, 0x80000000);
3293
3294 for (i = 0; i < cik_default_size; i++)
3295 radeon_ring_write(ring, cik_default_state[i]);
3296
3297 radeon_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
3298 radeon_ring_write(ring, PACKET3_PREAMBLE_END_CLEAR_STATE);
3299
3300 /* set clear context state */
3301 radeon_ring_write(ring, PACKET3(PACKET3_CLEAR_STATE, 0));
3302 radeon_ring_write(ring, 0);
3303
3304 radeon_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 2));
3305 radeon_ring_write(ring, 0x00000316);
3306 radeon_ring_write(ring, 0x0000000e); /* VGT_VERTEX_REUSE_BLOCK_CNTL */
3307 radeon_ring_write(ring, 0x00000010); /* VGT_OUT_DEALLOC_CNTL */
3308
3309 radeon_ring_unlock_commit(rdev, ring);
3310
3311 return 0;
3312 }
3313
3314 /**
3315 * cik_cp_gfx_fini - stop the gfx ring
3316 *
3317 * @rdev: radeon_device pointer
3318 *
3319 * Stop the gfx ring and tear down the driver ring
3320 * info.
3321 */
3322 static void cik_cp_gfx_fini(struct radeon_device *rdev)
3323 {
3324 cik_cp_gfx_enable(rdev, false);
3325 radeon_ring_fini(rdev, &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]);
3326 }
3327
3328 /**
3329 * cik_cp_gfx_resume - setup the gfx ring buffer registers
3330 *
3331 * @rdev: radeon_device pointer
3332 *
3333 * Program the location and size of the gfx ring buffer
3334 * and test it to make sure it's working.
3335 * Returns 0 for success, error for failure.
3336 */
3337 static int cik_cp_gfx_resume(struct radeon_device *rdev)
3338 {
3339 struct radeon_ring *ring;
3340 u32 tmp;
3341 u32 rb_bufsz;
3342 u64 rb_addr;
3343 int r;
3344
3345 WREG32(CP_SEM_WAIT_TIMER, 0x0);
3346 WREG32(CP_SEM_INCOMPLETE_TIMER_CNTL, 0x0);
3347
3348 /* Set the write pointer delay */
3349 WREG32(CP_RB_WPTR_DELAY, 0);
3350
3351 /* set the RB to use vmid 0 */
3352 WREG32(CP_RB_VMID, 0);
3353
3354 WREG32(SCRATCH_ADDR, ((rdev->wb.gpu_addr + RADEON_WB_SCRATCH_OFFSET) >> 8) & 0xFFFFFFFF);
3355
3356 /* ring 0 - compute and gfx */
3357 /* Set ring buffer size */
3358 ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
3359 rb_bufsz = drm_order(ring->ring_size / 8);
3360 tmp = (drm_order(RADEON_GPU_PAGE_SIZE/8) << 8) | rb_bufsz;
3361 #ifdef __BIG_ENDIAN
3362 tmp |= BUF_SWAP_32BIT;
3363 #endif
3364 WREG32(CP_RB0_CNTL, tmp);
3365
3366 /* Initialize the ring buffer's read and write pointers */
3367 WREG32(CP_RB0_CNTL, tmp | RB_RPTR_WR_ENA);
3368 ring->wptr = 0;
3369 WREG32(CP_RB0_WPTR, ring->wptr);
3370
3371 /* set the wb address wether it's enabled or not */
3372 WREG32(CP_RB0_RPTR_ADDR, (rdev->wb.gpu_addr + RADEON_WB_CP_RPTR_OFFSET) & 0xFFFFFFFC);
3373 WREG32(CP_RB0_RPTR_ADDR_HI, upper_32_bits(rdev->wb.gpu_addr + RADEON_WB_CP_RPTR_OFFSET) & 0xFF);
3374
3375 /* scratch register shadowing is no longer supported */
3376 WREG32(SCRATCH_UMSK, 0);
3377
3378 if (!rdev->wb.enabled)
3379 tmp |= RB_NO_UPDATE;
3380
3381 mdelay(1);
3382 WREG32(CP_RB0_CNTL, tmp);
3383
3384 rb_addr = ring->gpu_addr >> 8;
3385 WREG32(CP_RB0_BASE, rb_addr);
3386 WREG32(CP_RB0_BASE_HI, upper_32_bits(rb_addr));
3387
3388 ring->rptr = RREG32(CP_RB0_RPTR);
3389
3390 /* start the ring */
3391 cik_cp_gfx_start(rdev);
3392 rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = true;
3393 r = radeon_ring_test(rdev, RADEON_RING_TYPE_GFX_INDEX, &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]);
3394 if (r) {
3395 rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = false;
3396 return r;
3397 }
3398 return 0;
3399 }
3400
3401 u32 cik_compute_ring_get_rptr(struct radeon_device *rdev,
3402 struct radeon_ring *ring)
3403 {
3404 u32 rptr;
3405
3406
3407
3408 if (rdev->wb.enabled) {
3409 rptr = le32_to_cpu(rdev->wb.wb[ring->rptr_offs/4]);
3410 } else {
3411 mutex_lock(&rdev->srbm_mutex);
3412 cik_srbm_select(rdev, ring->me, ring->pipe, ring->queue, 0);
3413 rptr = RREG32(CP_HQD_PQ_RPTR);
3414 cik_srbm_select(rdev, 0, 0, 0, 0);
3415 mutex_unlock(&rdev->srbm_mutex);
3416 }
3417 rptr = (rptr & ring->ptr_reg_mask) >> ring->ptr_reg_shift;
3418
3419 return rptr;
3420 }
3421
3422 u32 cik_compute_ring_get_wptr(struct radeon_device *rdev,
3423 struct radeon_ring *ring)
3424 {
3425 u32 wptr;
3426
3427 if (rdev->wb.enabled) {
3428 wptr = le32_to_cpu(rdev->wb.wb[ring->wptr_offs/4]);
3429 } else {
3430 mutex_lock(&rdev->srbm_mutex);
3431 cik_srbm_select(rdev, ring->me, ring->pipe, ring->queue, 0);
3432 wptr = RREG32(CP_HQD_PQ_WPTR);
3433 cik_srbm_select(rdev, 0, 0, 0, 0);
3434 mutex_unlock(&rdev->srbm_mutex);
3435 }
3436 wptr = (wptr & ring->ptr_reg_mask) >> ring->ptr_reg_shift;
3437
3438 return wptr;
3439 }
3440
3441 void cik_compute_ring_set_wptr(struct radeon_device *rdev,
3442 struct radeon_ring *ring)
3443 {
3444 u32 wptr = (ring->wptr << ring->ptr_reg_shift) & ring->ptr_reg_mask;
3445
3446 rdev->wb.wb[ring->wptr_offs/4] = cpu_to_le32(wptr);
3447 WDOORBELL32(ring->doorbell_offset, wptr);
3448 }
3449
3450 /**
3451 * cik_cp_compute_enable - enable/disable the compute CP MEs
3452 *
3453 * @rdev: radeon_device pointer
3454 * @enable: enable or disable the MEs
3455 *
3456 * Halts or unhalts the compute MEs.
3457 */
3458 static void cik_cp_compute_enable(struct radeon_device *rdev, bool enable)
3459 {
3460 if (enable)
3461 WREG32(CP_MEC_CNTL, 0);
3462 else
3463 WREG32(CP_MEC_CNTL, (MEC_ME1_HALT | MEC_ME2_HALT));
3464 udelay(50);
3465 }
3466
3467 /**
3468 * cik_cp_compute_load_microcode - load the compute CP ME ucode
3469 *
3470 * @rdev: radeon_device pointer
3471 *
3472 * Loads the compute MEC1&2 ucode.
3473 * Returns 0 for success, -EINVAL if the ucode is not available.
3474 */
3475 static int cik_cp_compute_load_microcode(struct radeon_device *rdev)
3476 {
3477 const __be32 *fw_data;
3478 int i;
3479
3480 if (!rdev->mec_fw)
3481 return -EINVAL;
3482
3483 cik_cp_compute_enable(rdev, false);
3484
3485 /* MEC1 */
3486 fw_data = (const __be32 *)rdev->mec_fw->data;
3487 WREG32(CP_MEC_ME1_UCODE_ADDR, 0);
3488 for (i = 0; i < CIK_MEC_UCODE_SIZE; i++)
3489 WREG32(CP_MEC_ME1_UCODE_DATA, be32_to_cpup(fw_data++));
3490 WREG32(CP_MEC_ME1_UCODE_ADDR, 0);
3491
3492 if (rdev->family == CHIP_KAVERI) {
3493 /* MEC2 */
3494 fw_data = (const __be32 *)rdev->mec_fw->data;
3495 WREG32(CP_MEC_ME2_UCODE_ADDR, 0);
3496 for (i = 0; i < CIK_MEC_UCODE_SIZE; i++)
3497 WREG32(CP_MEC_ME2_UCODE_DATA, be32_to_cpup(fw_data++));
3498 WREG32(CP_MEC_ME2_UCODE_ADDR, 0);
3499 }
3500
3501 return 0;
3502 }
3503
3504 /**
3505 * cik_cp_compute_start - start the compute queues
3506 *
3507 * @rdev: radeon_device pointer
3508 *
3509 * Enable the compute queues.
3510 * Returns 0 for success, error for failure.
3511 */
3512 static int cik_cp_compute_start(struct radeon_device *rdev)
3513 {
3514 cik_cp_compute_enable(rdev, true);
3515
3516 return 0;
3517 }
3518
3519 /**
3520 * cik_cp_compute_fini - stop the compute queues
3521 *
3522 * @rdev: radeon_device pointer
3523 *
3524 * Stop the compute queues and tear down the driver queue
3525 * info.
3526 */
3527 static void cik_cp_compute_fini(struct radeon_device *rdev)
3528 {
3529 int i, idx, r;
3530
3531 cik_cp_compute_enable(rdev, false);
3532
3533 for (i = 0; i < 2; i++) {
3534 if (i == 0)
3535 idx = CAYMAN_RING_TYPE_CP1_INDEX;
3536 else
3537 idx = CAYMAN_RING_TYPE_CP2_INDEX;
3538
3539 if (rdev->ring[idx].mqd_obj) {
3540 r = radeon_bo_reserve(rdev->ring[idx].mqd_obj, false);
3541 if (unlikely(r != 0))
3542 dev_warn(rdev->dev, "(%d) reserve MQD bo failed\n", r);
3543
3544 radeon_bo_unpin(rdev->ring[idx].mqd_obj);
3545 radeon_bo_unreserve(rdev->ring[idx].mqd_obj);
3546
3547 radeon_bo_unref(&rdev->ring[idx].mqd_obj);
3548 rdev->ring[idx].mqd_obj = NULL;
3549 }
3550 }
3551 }
3552
3553 static void cik_mec_fini(struct radeon_device *rdev)
3554 {
3555 int r;
3556
3557 if (rdev->mec.hpd_eop_obj) {
3558 r = radeon_bo_reserve(rdev->mec.hpd_eop_obj, false);
3559 if (unlikely(r != 0))
3560 dev_warn(rdev->dev, "(%d) reserve HPD EOP bo failed\n", r);
3561 radeon_bo_unpin(rdev->mec.hpd_eop_obj);
3562 radeon_bo_unreserve(rdev->mec.hpd_eop_obj);
3563
3564 radeon_bo_unref(&rdev->mec.hpd_eop_obj);
3565 rdev->mec.hpd_eop_obj = NULL;
3566 }
3567 }
3568
3569 #define MEC_HPD_SIZE 2048
3570
3571 static int cik_mec_init(struct radeon_device *rdev)
3572 {
3573 int r;
3574 u32 *hpd;
3575
3576 /*
3577 * KV: 2 MEC, 4 Pipes/MEC, 8 Queues/Pipe - 64 Queues total
3578 * CI/KB: 1 MEC, 4 Pipes/MEC, 8 Queues/Pipe - 32 Queues total
3579 */
3580 if (rdev->family == CHIP_KAVERI)
3581 rdev->mec.num_mec = 2;
3582 else
3583 rdev->mec.num_mec = 1;
3584 rdev->mec.num_pipe = 4;
3585 rdev->mec.num_queue = rdev->mec.num_mec * rdev->mec.num_pipe * 8;
3586
3587 if (rdev->mec.hpd_eop_obj == NULL) {
3588 r = radeon_bo_create(rdev,
3589 rdev->mec.num_mec *rdev->mec.num_pipe * MEC_HPD_SIZE * 2,
3590 PAGE_SIZE, true,
3591 RADEON_GEM_DOMAIN_GTT, NULL,
3592 &rdev->mec.hpd_eop_obj);
3593 if (r) {
3594 dev_warn(rdev->dev, "(%d) create HDP EOP bo failed\n", r);
3595 return r;
3596 }
3597 }
3598
3599 r = radeon_bo_reserve(rdev->mec.hpd_eop_obj, false);
3600 if (unlikely(r != 0)) {
3601 cik_mec_fini(rdev);
3602 return r;
3603 }
3604 r = radeon_bo_pin(rdev->mec.hpd_eop_obj, RADEON_GEM_DOMAIN_GTT,
3605 &rdev->mec.hpd_eop_gpu_addr);
3606 if (r) {
3607 dev_warn(rdev->dev, "(%d) pin HDP EOP bo failed\n", r);
3608 cik_mec_fini(rdev);
3609 return r;
3610 }
3611 r = radeon_bo_kmap(rdev->mec.hpd_eop_obj, (void **)&hpd);
3612 if (r) {
3613 dev_warn(rdev->dev, "(%d) map HDP EOP bo failed\n", r);
3614 cik_mec_fini(rdev);
3615 return r;
3616 }
3617
3618 /* clear memory. Not sure if this is required or not */
3619 memset(hpd, 0, rdev->mec.num_mec *rdev->mec.num_pipe * MEC_HPD_SIZE * 2);
3620
3621 radeon_bo_kunmap(rdev->mec.hpd_eop_obj);
3622 radeon_bo_unreserve(rdev->mec.hpd_eop_obj);
3623
3624 return 0;
3625 }
3626
3627 struct hqd_registers
3628 {
3629 u32 cp_mqd_base_addr;
3630 u32 cp_mqd_base_addr_hi;
3631 u32 cp_hqd_active;
3632 u32 cp_hqd_vmid;
3633 u32 cp_hqd_persistent_state;
3634 u32 cp_hqd_pipe_priority;
3635 u32 cp_hqd_queue_priority;
3636 u32 cp_hqd_quantum;
3637 u32 cp_hqd_pq_base;
3638 u32 cp_hqd_pq_base_hi;
3639 u32 cp_hqd_pq_rptr;
3640 u32 cp_hqd_pq_rptr_report_addr;
3641 u32 cp_hqd_pq_rptr_report_addr_hi;
3642 u32 cp_hqd_pq_wptr_poll_addr;
3643 u32 cp_hqd_pq_wptr_poll_addr_hi;
3644 u32 cp_hqd_pq_doorbell_control;
3645 u32 cp_hqd_pq_wptr;
3646 u32 cp_hqd_pq_control;
3647 u32 cp_hqd_ib_base_addr;
3648 u32 cp_hqd_ib_base_addr_hi;
3649 u32 cp_hqd_ib_rptr;
3650 u32 cp_hqd_ib_control;
3651 u32 cp_hqd_iq_timer;
3652 u32 cp_hqd_iq_rptr;
3653 u32 cp_hqd_dequeue_request;
3654 u32 cp_hqd_dma_offload;
3655 u32 cp_hqd_sema_cmd;
3656 u32 cp_hqd_msg_type;
3657 u32 cp_hqd_atomic0_preop_lo;
3658 u32 cp_hqd_atomic0_preop_hi;
3659 u32 cp_hqd_atomic1_preop_lo;
3660 u32 cp_hqd_atomic1_preop_hi;
3661 u32 cp_hqd_hq_scheduler0;
3662 u32 cp_hqd_hq_scheduler1;
3663 u32 cp_mqd_control;
3664 };
3665
3666 struct bonaire_mqd
3667 {
3668 u32 header;
3669 u32 dispatch_initiator;
3670 u32 dimensions[3];
3671 u32 start_idx[3];
3672 u32 num_threads[3];
3673 u32 pipeline_stat_enable;
3674 u32 perf_counter_enable;
3675 u32 pgm[2];
3676 u32 tba[2];
3677 u32 tma[2];
3678 u32 pgm_rsrc[2];
3679 u32 vmid;
3680 u32 resource_limits;
3681 u32 static_thread_mgmt01[2];
3682 u32 tmp_ring_size;
3683 u32 static_thread_mgmt23[2];
3684 u32 restart[3];
3685 u32 thread_trace_enable;
3686 u32 reserved1;
3687 u32 user_data[16];
3688 u32 vgtcs_invoke_count[2];
3689 struct hqd_registers queue_state;
3690 u32 dequeue_cntr;
3691 u32 interrupt_queue[64];
3692 };
3693
3694 /**
3695 * cik_cp_compute_resume - setup the compute queue registers
3696 *
3697 * @rdev: radeon_device pointer
3698 *
3699 * Program the compute queues and test them to make sure they
3700 * are working.
3701 * Returns 0 for success, error for failure.
3702 */
3703 static int cik_cp_compute_resume(struct radeon_device *rdev)
3704 {
3705 int r, i, idx;
3706 u32 tmp;
3707 bool use_doorbell = true;
3708 u64 hqd_gpu_addr;
3709 u64 mqd_gpu_addr;
3710 u64 eop_gpu_addr;
3711 u64 wb_gpu_addr;
3712 u32 *buf;
3713 struct bonaire_mqd *mqd;
3714
3715 r = cik_cp_compute_start(rdev);
3716 if (r)
3717 return r;
3718
3719 /* fix up chicken bits */
3720 tmp = RREG32(CP_CPF_DEBUG);
3721 tmp |= (1 << 23);
3722 WREG32(CP_CPF_DEBUG, tmp);
3723
3724 /* init the pipes */
3725 mutex_lock(&rdev->srbm_mutex);
3726 for (i = 0; i < (rdev->mec.num_pipe * rdev->mec.num_mec); i++) {
3727 int me = (i < 4) ? 1 : 2;
3728 int pipe = (i < 4) ? i : (i - 4);
3729
3730 eop_gpu_addr = rdev->mec.hpd_eop_gpu_addr + (i * MEC_HPD_SIZE * 2);
3731
3732 cik_srbm_select(rdev, me, pipe, 0, 0);
3733
3734 /* write the EOP addr */
3735 WREG32(CP_HPD_EOP_BASE_ADDR, eop_gpu_addr >> 8);
3736 WREG32(CP_HPD_EOP_BASE_ADDR_HI, upper_32_bits(eop_gpu_addr) >> 8);
3737
3738 /* set the VMID assigned */
3739 WREG32(CP_HPD_EOP_VMID, 0);
3740
3741 /* set the EOP size, register value is 2^(EOP_SIZE+1) dwords */
3742 tmp = RREG32(CP_HPD_EOP_CONTROL);
3743 tmp &= ~EOP_SIZE_MASK;
3744 tmp |= drm_order(MEC_HPD_SIZE / 8);
3745 WREG32(CP_HPD_EOP_CONTROL, tmp);
3746 }
3747 cik_srbm_select(rdev, 0, 0, 0, 0);
3748 mutex_unlock(&rdev->srbm_mutex);
3749
3750 /* init the queues. Just two for now. */
3751 for (i = 0; i < 2; i++) {
3752 if (i == 0)
3753 idx = CAYMAN_RING_TYPE_CP1_INDEX;
3754 else
3755 idx = CAYMAN_RING_TYPE_CP2_INDEX;
3756
3757 if (rdev->ring[idx].mqd_obj == NULL) {
3758 r = radeon_bo_create(rdev,
3759 sizeof(struct bonaire_mqd),
3760 PAGE_SIZE, true,
3761 RADEON_GEM_DOMAIN_GTT, NULL,
3762 &rdev->ring[idx].mqd_obj);
3763 if (r) {
3764 dev_warn(rdev->dev, "(%d) create MQD bo failed\n", r);
3765 return r;
3766 }
3767 }
3768
3769 r = radeon_bo_reserve(rdev->ring[idx].mqd_obj, false);
3770 if (unlikely(r != 0)) {
3771 cik_cp_compute_fini(rdev);
3772 return r;
3773 }
3774 r = radeon_bo_pin(rdev->ring[idx].mqd_obj, RADEON_GEM_DOMAIN_GTT,
3775 &mqd_gpu_addr);
3776 if (r) {
3777 dev_warn(rdev->dev, "(%d) pin MQD bo failed\n", r);
3778 cik_cp_compute_fini(rdev);
3779 return r;
3780 }
3781 r = radeon_bo_kmap(rdev->ring[idx].mqd_obj, (void **)&buf);
3782 if (r) {
3783 dev_warn(rdev->dev, "(%d) map MQD bo failed\n", r);
3784 cik_cp_compute_fini(rdev);
3785 return r;
3786 }
3787
3788 /* doorbell offset */
3789 rdev->ring[idx].doorbell_offset =
3790 (rdev->ring[idx].doorbell_page_num * PAGE_SIZE) + 0;
3791
3792 /* init the mqd struct */
3793 memset(buf, 0, sizeof(struct bonaire_mqd));
3794
3795 mqd = (struct bonaire_mqd *)buf;
3796 mqd->header = 0xC0310800;
3797 mqd->static_thread_mgmt01[0] = 0xffffffff;
3798 mqd->static_thread_mgmt01[1] = 0xffffffff;
3799 mqd->static_thread_mgmt23[0] = 0xffffffff;
3800 mqd->static_thread_mgmt23[1] = 0xffffffff;
3801
3802 mutex_lock(&rdev->srbm_mutex);
3803 cik_srbm_select(rdev, rdev->ring[idx].me,
3804 rdev->ring[idx].pipe,
3805 rdev->ring[idx].queue, 0);
3806
3807 /* disable wptr polling */
3808 tmp = RREG32(CP_PQ_WPTR_POLL_CNTL);
3809 tmp &= ~WPTR_POLL_EN;
3810 WREG32(CP_PQ_WPTR_POLL_CNTL, tmp);
3811
3812 /* enable doorbell? */
3813 mqd->queue_state.cp_hqd_pq_doorbell_control =
3814 RREG32(CP_HQD_PQ_DOORBELL_CONTROL);
3815 if (use_doorbell)
3816 mqd->queue_state.cp_hqd_pq_doorbell_control |= DOORBELL_EN;
3817 else
3818 mqd->queue_state.cp_hqd_pq_doorbell_control &= ~DOORBELL_EN;
3819 WREG32(CP_HQD_PQ_DOORBELL_CONTROL,
3820 mqd->queue_state.cp_hqd_pq_doorbell_control);
3821
3822 /* disable the queue if it's active */
3823 mqd->queue_state.cp_hqd_dequeue_request = 0;
3824 mqd->queue_state.cp_hqd_pq_rptr = 0;
3825 mqd->queue_state.cp_hqd_pq_wptr= 0;
3826 if (RREG32(CP_HQD_ACTIVE) & 1) {
3827 WREG32(CP_HQD_DEQUEUE_REQUEST, 1);
3828 for (i = 0; i < rdev->usec_timeout; i++) {
3829 if (!(RREG32(CP_HQD_ACTIVE) & 1))
3830 break;
3831 udelay(1);
3832 }
3833 WREG32(CP_HQD_DEQUEUE_REQUEST, mqd->queue_state.cp_hqd_dequeue_request);
3834 WREG32(CP_HQD_PQ_RPTR, mqd->queue_state.cp_hqd_pq_rptr);
3835 WREG32(CP_HQD_PQ_WPTR, mqd->queue_state.cp_hqd_pq_wptr);
3836 }
3837
3838 /* set the pointer to the MQD */
3839 mqd->queue_state.cp_mqd_base_addr = mqd_gpu_addr & 0xfffffffc;
3840 mqd->queue_state.cp_mqd_base_addr_hi = upper_32_bits(mqd_gpu_addr);
3841 WREG32(CP_MQD_BASE_ADDR, mqd->queue_state.cp_mqd_base_addr);
3842 WREG32(CP_MQD_BASE_ADDR_HI, mqd->queue_state.cp_mqd_base_addr_hi);
3843 /* set MQD vmid to 0 */
3844 mqd->queue_state.cp_mqd_control = RREG32(CP_MQD_CONTROL);
3845 mqd->queue_state.cp_mqd_control &= ~MQD_VMID_MASK;
3846 WREG32(CP_MQD_CONTROL, mqd->queue_state.cp_mqd_control);
3847
3848 /* set the pointer to the HQD, this is similar CP_RB0_BASE/_HI */
3849 hqd_gpu_addr = rdev->ring[idx].gpu_addr >> 8;
3850 mqd->queue_state.cp_hqd_pq_base = hqd_gpu_addr;
3851 mqd->queue_state.cp_hqd_pq_base_hi = upper_32_bits(hqd_gpu_addr);
3852 WREG32(CP_HQD_PQ_BASE, mqd->queue_state.cp_hqd_pq_base);
3853 WREG32(CP_HQD_PQ_BASE_HI, mqd->queue_state.cp_hqd_pq_base_hi);
3854
3855 /* set up the HQD, this is similar to CP_RB0_CNTL */
3856 mqd->queue_state.cp_hqd_pq_control = RREG32(CP_HQD_PQ_CONTROL);
3857 mqd->queue_state.cp_hqd_pq_control &=
3858 ~(QUEUE_SIZE_MASK | RPTR_BLOCK_SIZE_MASK);
3859
3860 mqd->queue_state.cp_hqd_pq_control |=
3861 drm_order(rdev->ring[idx].ring_size / 8);
3862 mqd->queue_state.cp_hqd_pq_control |=
3863 (drm_order(RADEON_GPU_PAGE_SIZE/8) << 8);
3864 #ifdef __BIG_ENDIAN
3865 mqd->queue_state.cp_hqd_pq_control |= BUF_SWAP_32BIT;
3866 #endif
3867 mqd->queue_state.cp_hqd_pq_control &=
3868 ~(UNORD_DISPATCH | ROQ_PQ_IB_FLIP | PQ_VOLATILE);
3869 mqd->queue_state.cp_hqd_pq_control |=
3870 PRIV_STATE | KMD_QUEUE; /* assuming kernel queue control */
3871 WREG32(CP_HQD_PQ_CONTROL, mqd->queue_state.cp_hqd_pq_control);
3872
3873 /* only used if CP_PQ_WPTR_POLL_CNTL.WPTR_POLL_EN=1 */
3874 if (i == 0)
3875 wb_gpu_addr = rdev->wb.gpu_addr + CIK_WB_CP1_WPTR_OFFSET;
3876 else
3877 wb_gpu_addr = rdev->wb.gpu_addr + CIK_WB_CP2_WPTR_OFFSET;
3878 mqd->queue_state.cp_hqd_pq_wptr_poll_addr = wb_gpu_addr & 0xfffffffc;
3879 mqd->queue_state.cp_hqd_pq_wptr_poll_addr_hi = upper_32_bits(wb_gpu_addr) & 0xffff;
3880 WREG32(CP_HQD_PQ_WPTR_POLL_ADDR, mqd->queue_state.cp_hqd_pq_wptr_poll_addr);
3881 WREG32(CP_HQD_PQ_WPTR_POLL_ADDR_HI,
3882 mqd->queue_state.cp_hqd_pq_wptr_poll_addr_hi);
3883
3884 /* set the wb address wether it's enabled or not */
3885 if (i == 0)
3886 wb_gpu_addr = rdev->wb.gpu_addr + RADEON_WB_CP1_RPTR_OFFSET;
3887 else
3888 wb_gpu_addr = rdev->wb.gpu_addr + RADEON_WB_CP2_RPTR_OFFSET;
3889 mqd->queue_state.cp_hqd_pq_rptr_report_addr = wb_gpu_addr & 0xfffffffc;
3890 mqd->queue_state.cp_hqd_pq_rptr_report_addr_hi =
3891 upper_32_bits(wb_gpu_addr) & 0xffff;
3892 WREG32(CP_HQD_PQ_RPTR_REPORT_ADDR,
3893 mqd->queue_state.cp_hqd_pq_rptr_report_addr);
3894 WREG32(CP_HQD_PQ_RPTR_REPORT_ADDR_HI,
3895 mqd->queue_state.cp_hqd_pq_rptr_report_addr_hi);
3896
3897 /* enable the doorbell if requested */
3898 if (use_doorbell) {
3899 mqd->queue_state.cp_hqd_pq_doorbell_control =
3900 RREG32(CP_HQD_PQ_DOORBELL_CONTROL);
3901 mqd->queue_state.cp_hqd_pq_doorbell_control &= ~DOORBELL_OFFSET_MASK;
3902 mqd->queue_state.cp_hqd_pq_doorbell_control |=
3903 DOORBELL_OFFSET(rdev->ring[idx].doorbell_offset / 4);
3904 mqd->queue_state.cp_hqd_pq_doorbell_control |= DOORBELL_EN;
3905 mqd->queue_state.cp_hqd_pq_doorbell_control &=
3906 ~(DOORBELL_SOURCE | DOORBELL_HIT);
3907
3908 } else {
3909 mqd->queue_state.cp_hqd_pq_doorbell_control = 0;
3910 }
3911 WREG32(CP_HQD_PQ_DOORBELL_CONTROL,
3912 mqd->queue_state.cp_hqd_pq_doorbell_control);
3913
3914 /* read and write pointers, similar to CP_RB0_WPTR/_RPTR */
3915 rdev->ring[idx].wptr = 0;
3916 mqd->queue_state.cp_hqd_pq_wptr = rdev->ring[idx].wptr;
3917 WREG32(CP_HQD_PQ_WPTR, mqd->queue_state.cp_hqd_pq_wptr);
3918 rdev->ring[idx].rptr = RREG32(CP_HQD_PQ_RPTR);
3919 mqd->queue_state.cp_hqd_pq_rptr = rdev->ring[idx].rptr;
3920
3921 /* set the vmid for the queue */
3922 mqd->queue_state.cp_hqd_vmid = 0;
3923 WREG32(CP_HQD_VMID, mqd->queue_state.cp_hqd_vmid);
3924
3925 /* activate the queue */
3926 mqd->queue_state.cp_hqd_active = 1;
3927 WREG32(CP_HQD_ACTIVE, mqd->queue_state.cp_hqd_active);
3928
3929 cik_srbm_select(rdev, 0, 0, 0, 0);
3930 mutex_unlock(&rdev->srbm_mutex);
3931
3932 radeon_bo_kunmap(rdev->ring[idx].mqd_obj);
3933 radeon_bo_unreserve(rdev->ring[idx].mqd_obj);
3934
3935 rdev->ring[idx].ready = true;
3936 r = radeon_ring_test(rdev, idx, &rdev->ring[idx]);
3937 if (r)
3938 rdev->ring[idx].ready = false;
3939 }
3940
3941 return 0;
3942 }
3943
3944 static void cik_cp_enable(struct radeon_device *rdev, bool enable)
3945 {
3946 cik_cp_gfx_enable(rdev, enable);
3947 cik_cp_compute_enable(rdev, enable);
3948 }
3949
3950 static int cik_cp_load_microcode(struct radeon_device *rdev)
3951 {
3952 int r;
3953
3954 r = cik_cp_gfx_load_microcode(rdev);
3955 if (r)
3956 return r;
3957 r = cik_cp_compute_load_microcode(rdev);
3958 if (r)
3959 return r;
3960
3961 return 0;
3962 }
3963
3964 static void cik_cp_fini(struct radeon_device *rdev)
3965 {
3966 cik_cp_gfx_fini(rdev);
3967 cik_cp_compute_fini(rdev);
3968 }
3969
3970 static int cik_cp_resume(struct radeon_device *rdev)
3971 {
3972 int r;
3973
3974 /* Reset all cp blocks */
3975 WREG32(GRBM_SOFT_RESET, SOFT_RESET_CP);
3976 RREG32(GRBM_SOFT_RESET);
3977 mdelay(15);
3978 WREG32(GRBM_SOFT_RESET, 0);
3979 RREG32(GRBM_SOFT_RESET);
3980
3981 r = cik_cp_load_microcode(rdev);
3982 if (r)
3983 return r;
3984
3985 r = cik_cp_gfx_resume(rdev);
3986 if (r)
3987 return r;
3988 r = cik_cp_compute_resume(rdev);
3989 if (r)
3990 return r;
3991
3992 return 0;
3993 }
3994
3995 /*
3996 * sDMA - System DMA
3997 * Starting with CIK, the GPU has new asynchronous
3998 * DMA engines. These engines are used for compute
3999 * and gfx. There are two DMA engines (SDMA0, SDMA1)
4000 * and each one supports 1 ring buffer used for gfx
4001 * and 2 queues used for compute.
4002 *
4003 * The programming model is very similar to the CP
4004 * (ring buffer, IBs, etc.), but sDMA has it's own
4005 * packet format that is different from the PM4 format
4006 * used by the CP. sDMA supports copying data, writing
4007 * embedded data, solid fills, and a number of other
4008 * things. It also has support for tiling/detiling of
4009 * buffers.
4010 */
4011 /**
4012 * cik_sdma_ring_ib_execute - Schedule an IB on the DMA engine
4013 *
4014 * @rdev: radeon_device pointer
4015 * @ib: IB object to schedule
4016 *
4017 * Schedule an IB in the DMA ring (CIK).
4018 */
4019 void cik_sdma_ring_ib_execute(struct radeon_device *rdev,
4020 struct radeon_ib *ib)
4021 {
4022 struct radeon_ring *ring = &rdev->ring[ib->ring];
4023 u32 extra_bits = (ib->vm ? ib->vm->id : 0) & 0xf;
4024
4025 if (rdev->wb.enabled) {
4026 u32 next_rptr = ring->wptr + 5;
4027 while ((next_rptr & 7) != 4)
4028 next_rptr++;
4029 next_rptr += 4;
4030 radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_WRITE, SDMA_WRITE_SUB_OPCODE_LINEAR, 0));
4031 radeon_ring_write(ring, ring->next_rptr_gpu_addr & 0xfffffffc);
4032 radeon_ring_write(ring, upper_32_bits(ring->next_rptr_gpu_addr) & 0xffffffff);
4033 radeon_ring_write(ring, 1); /* number of DWs to follow */
4034 radeon_ring_write(ring, next_rptr);
4035 }
4036
4037 /* IB packet must end on a 8 DW boundary */
4038 while ((ring->wptr & 7) != 4)
4039 radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_NOP, 0, 0));
4040 radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_INDIRECT_BUFFER, 0, extra_bits));
4041 radeon_ring_write(ring, ib->gpu_addr & 0xffffffe0); /* base must be 32 byte aligned */
4042 radeon_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xffffffff);
4043 radeon_ring_write(ring, ib->length_dw);
4044
4045 }
4046
4047 /**
4048 * cik_sdma_fence_ring_emit - emit a fence on the DMA ring
4049 *
4050 * @rdev: radeon_device pointer
4051 * @fence: radeon fence object
4052 *
4053 * Add a DMA fence packet to the ring to write
4054 * the fence seq number and DMA trap packet to generate
4055 * an interrupt if needed (CIK).
4056 */
4057 void cik_sdma_fence_ring_emit(struct radeon_device *rdev,
4058 struct radeon_fence *fence)
4059 {
4060 struct radeon_ring *ring = &rdev->ring[fence->ring];
4061 u64 addr = rdev->fence_drv[fence->ring].gpu_addr;
4062 u32 extra_bits = (SDMA_POLL_REG_MEM_EXTRA_OP(1) |
4063 SDMA_POLL_REG_MEM_EXTRA_FUNC(3)); /* == */
4064 u32 ref_and_mask;
4065
4066 if (fence->ring == R600_RING_TYPE_DMA_INDEX)
4067 ref_and_mask = SDMA0;
4068 else
4069 ref_and_mask = SDMA1;
4070
4071 /* write the fence */
4072 radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_FENCE, 0, 0));
4073 radeon_ring_write(ring, addr & 0xffffffff);
4074 radeon_ring_write(ring, upper_32_bits(addr) & 0xffffffff);
4075 radeon_ring_write(ring, fence->seq);
4076 /* generate an interrupt */
4077 radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_TRAP, 0, 0));
4078 /* flush HDP */
4079 radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_POLL_REG_MEM, 0, extra_bits));
4080 radeon_ring_write(ring, GPU_HDP_FLUSH_DONE);
4081 radeon_ring_write(ring, GPU_HDP_FLUSH_REQ);
4082 radeon_ring_write(ring, ref_and_mask); /* REFERENCE */
4083 radeon_ring_write(ring, ref_and_mask); /* MASK */
4084 radeon_ring_write(ring, (4 << 16) | 10); /* RETRY_COUNT, POLL_INTERVAL */
4085 }
4086
4087 /**
4088 * cik_sdma_semaphore_ring_emit - emit a semaphore on the dma ring
4089 *
4090 * @rdev: radeon_device pointer
4091 * @ring: radeon_ring structure holding ring information
4092 * @semaphore: radeon semaphore object
4093 * @emit_wait: wait or signal semaphore
4094 *
4095 * Add a DMA semaphore packet to the ring wait on or signal
4096 * other rings (CIK).
4097 */
4098 void cik_sdma_semaphore_ring_emit(struct radeon_device *rdev,
4099 struct radeon_ring *ring,
4100 struct radeon_semaphore *semaphore,
4101 bool emit_wait)
4102 {
4103 u64 addr = semaphore->gpu_addr;
4104 u32 extra_bits = emit_wait ? 0 : SDMA_SEMAPHORE_EXTRA_S;
4105
4106 radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_SEMAPHORE, 0, extra_bits));
4107 radeon_ring_write(ring, addr & 0xfffffff8);
4108 radeon_ring_write(ring, upper_32_bits(addr) & 0xffffffff);
4109 }
4110
4111 /**
4112 * cik_sdma_gfx_stop - stop the gfx async dma engines
4113 *
4114 * @rdev: radeon_device pointer
4115 *
4116 * Stop the gfx async dma ring buffers (CIK).
4117 */
4118 static void cik_sdma_gfx_stop(struct radeon_device *rdev)
4119 {
4120 u32 rb_cntl, reg_offset;
4121 int i;
4122
4123 radeon_ttm_set_active_vram_size(rdev, rdev->mc.visible_vram_size);
4124
4125 for (i = 0; i < 2; i++) {
4126 if (i == 0)
4127 reg_offset = SDMA0_REGISTER_OFFSET;
4128 else
4129 reg_offset = SDMA1_REGISTER_OFFSET;
4130 rb_cntl = RREG32(SDMA0_GFX_RB_CNTL + reg_offset);
4131 rb_cntl &= ~SDMA_RB_ENABLE;
4132 WREG32(SDMA0_GFX_RB_CNTL + reg_offset, rb_cntl);
4133 WREG32(SDMA0_GFX_IB_CNTL + reg_offset, 0);
4134 }
4135 }
4136
4137 /**
4138 * cik_sdma_rlc_stop - stop the compute async dma engines
4139 *
4140 * @rdev: radeon_device pointer
4141 *
4142 * Stop the compute async dma queues (CIK).
4143 */
4144 static void cik_sdma_rlc_stop(struct radeon_device *rdev)
4145 {
4146 /* XXX todo */
4147 }
4148
4149 /**
4150 * cik_sdma_enable - stop the async dma engines
4151 *
4152 * @rdev: radeon_device pointer
4153 * @enable: enable/disable the DMA MEs.
4154 *
4155 * Halt or unhalt the async dma engines (CIK).
4156 */
4157 static void cik_sdma_enable(struct radeon_device *rdev, bool enable)
4158 {
4159 u32 me_cntl, reg_offset;
4160 int i;
4161
4162 for (i = 0; i < 2; i++) {
4163 if (i == 0)
4164 reg_offset = SDMA0_REGISTER_OFFSET;
4165 else
4166 reg_offset = SDMA1_REGISTER_OFFSET;
4167 me_cntl = RREG32(SDMA0_ME_CNTL + reg_offset);
4168 if (enable)
4169 me_cntl &= ~SDMA_HALT;
4170 else
4171 me_cntl |= SDMA_HALT;
4172 WREG32(SDMA0_ME_CNTL + reg_offset, me_cntl);
4173 }
4174 }
4175
4176 /**
4177 * cik_sdma_gfx_resume - setup and start the async dma engines
4178 *
4179 * @rdev: radeon_device pointer
4180 *
4181 * Set up the gfx DMA ring buffers and enable them (CIK).
4182 * Returns 0 for success, error for failure.
4183 */
4184 static int cik_sdma_gfx_resume(struct radeon_device *rdev)
4185 {
4186 struct radeon_ring *ring;
4187 u32 rb_cntl, ib_cntl;
4188 u32 rb_bufsz;
4189 u32 reg_offset, wb_offset;
4190 int i, r;
4191
4192 for (i = 0; i < 2; i++) {
4193 if (i == 0) {
4194 ring = &rdev->ring[R600_RING_TYPE_DMA_INDEX];
4195 reg_offset = SDMA0_REGISTER_OFFSET;
4196 wb_offset = R600_WB_DMA_RPTR_OFFSET;
4197 } else {
4198 ring = &rdev->ring[CAYMAN_RING_TYPE_DMA1_INDEX];
4199 reg_offset = SDMA1_REGISTER_OFFSET;
4200 wb_offset = CAYMAN_WB_DMA1_RPTR_OFFSET;
4201 }
4202
4203 WREG32(SDMA0_SEM_INCOMPLETE_TIMER_CNTL + reg_offset, 0);
4204 WREG32(SDMA0_SEM_WAIT_FAIL_TIMER_CNTL + reg_offset, 0);
4205
4206 /* Set ring buffer size in dwords */
4207 rb_bufsz = drm_order(ring->ring_size / 4);
4208 rb_cntl = rb_bufsz << 1;
4209 #ifdef __BIG_ENDIAN
4210 rb_cntl |= SDMA_RB_SWAP_ENABLE | SDMA_RPTR_WRITEBACK_SWAP_ENABLE;
4211 #endif
4212 WREG32(SDMA0_GFX_RB_CNTL + reg_offset, rb_cntl);
4213
4214 /* Initialize the ring buffer's read and write pointers */
4215 WREG32(SDMA0_GFX_RB_RPTR + reg_offset, 0);
4216 WREG32(SDMA0_GFX_RB_WPTR + reg_offset, 0);
4217
4218 /* set the wb address whether it's enabled or not */
4219 WREG32(SDMA0_GFX_RB_RPTR_ADDR_HI + reg_offset,
4220 upper_32_bits(rdev->wb.gpu_addr + wb_offset) & 0xFFFFFFFF);
4221 WREG32(SDMA0_GFX_RB_RPTR_ADDR_LO + reg_offset,
4222 ((rdev->wb.gpu_addr + wb_offset) & 0xFFFFFFFC));
4223
4224 if (rdev->wb.enabled)
4225 rb_cntl |= SDMA_RPTR_WRITEBACK_ENABLE;
4226
4227 WREG32(SDMA0_GFX_RB_BASE + reg_offset, ring->gpu_addr >> 8);
4228 WREG32(SDMA0_GFX_RB_BASE_HI + reg_offset, ring->gpu_addr >> 40);
4229
4230 ring->wptr = 0;
4231 WREG32(SDMA0_GFX_RB_WPTR + reg_offset, ring->wptr << 2);
4232
4233 ring->rptr = RREG32(SDMA0_GFX_RB_RPTR + reg_offset) >> 2;
4234
4235 /* enable DMA RB */
4236 WREG32(SDMA0_GFX_RB_CNTL + reg_offset, rb_cntl | SDMA_RB_ENABLE);
4237
4238 ib_cntl = SDMA_IB_ENABLE;
4239 #ifdef __BIG_ENDIAN
4240 ib_cntl |= SDMA_IB_SWAP_ENABLE;
4241 #endif
4242 /* enable DMA IBs */
4243 WREG32(SDMA0_GFX_IB_CNTL + reg_offset, ib_cntl);
4244
4245 ring->ready = true;
4246
4247 r = radeon_ring_test(rdev, ring->idx, ring);
4248 if (r) {
4249 ring->ready = false;
4250 return r;
4251 }
4252 }
4253
4254 radeon_ttm_set_active_vram_size(rdev, rdev->mc.real_vram_size);
4255
4256 return 0;
4257 }
4258
4259 /**
4260 * cik_sdma_rlc_resume - setup and start the async dma engines
4261 *
4262 * @rdev: radeon_device pointer
4263 *
4264 * Set up the compute DMA queues and enable them (CIK).
4265 * Returns 0 for success, error for failure.
4266 */
4267 static int cik_sdma_rlc_resume(struct radeon_device *rdev)
4268 {
4269 /* XXX todo */
4270 return 0;
4271 }
4272
4273 /**
4274 * cik_sdma_load_microcode - load the sDMA ME ucode
4275 *
4276 * @rdev: radeon_device pointer
4277 *
4278 * Loads the sDMA0/1 ucode.
4279 * Returns 0 for success, -EINVAL if the ucode is not available.
4280 */
4281 static int cik_sdma_load_microcode(struct radeon_device *rdev)
4282 {
4283 const __be32 *fw_data;
4284 int i;
4285
4286 if (!rdev->sdma_fw)
4287 return -EINVAL;
4288
4289 /* stop the gfx rings and rlc compute queues */
4290 cik_sdma_gfx_stop(rdev);
4291 cik_sdma_rlc_stop(rdev);
4292
4293 /* halt the MEs */
4294 cik_sdma_enable(rdev, false);
4295
4296 /* sdma0 */
4297 fw_data = (const __be32 *)rdev->sdma_fw->data;
4298 WREG32(SDMA0_UCODE_ADDR + SDMA0_REGISTER_OFFSET, 0);
4299 for (i = 0; i < CIK_SDMA_UCODE_SIZE; i++)
4300 WREG32(SDMA0_UCODE_DATA + SDMA0_REGISTER_OFFSET, be32_to_cpup(fw_data++));
4301 WREG32(SDMA0_UCODE_DATA + SDMA0_REGISTER_OFFSET, CIK_SDMA_UCODE_VERSION);
4302
4303 /* sdma1 */
4304 fw_data = (const __be32 *)rdev->sdma_fw->data;
4305 WREG32(SDMA0_UCODE_ADDR + SDMA1_REGISTER_OFFSET, 0);
4306 for (i = 0; i < CIK_SDMA_UCODE_SIZE; i++)
4307 WREG32(SDMA0_UCODE_DATA + SDMA1_REGISTER_OFFSET, be32_to_cpup(fw_data++));
4308 WREG32(SDMA0_UCODE_DATA + SDMA1_REGISTER_OFFSET, CIK_SDMA_UCODE_VERSION);
4309
4310 WREG32(SDMA0_UCODE_ADDR + SDMA0_REGISTER_OFFSET, 0);
4311 WREG32(SDMA0_UCODE_ADDR + SDMA1_REGISTER_OFFSET, 0);
4312 return 0;
4313 }
4314
4315 /**
4316 * cik_sdma_resume - setup and start the async dma engines
4317 *
4318 * @rdev: radeon_device pointer
4319 *
4320 * Set up the DMA engines and enable them (CIK).
4321 * Returns 0 for success, error for failure.
4322 */
4323 static int cik_sdma_resume(struct radeon_device *rdev)
4324 {
4325 int r;
4326
4327 /* Reset dma */
4328 WREG32(SRBM_SOFT_RESET, SOFT_RESET_SDMA | SOFT_RESET_SDMA1);
4329 RREG32(SRBM_SOFT_RESET);
4330 udelay(50);
4331 WREG32(SRBM_SOFT_RESET, 0);
4332 RREG32(SRBM_SOFT_RESET);
4333
4334 r = cik_sdma_load_microcode(rdev);
4335 if (r)
4336 return r;
4337
4338 /* unhalt the MEs */
4339 cik_sdma_enable(rdev, true);
4340
4341 /* start the gfx rings and rlc compute queues */
4342 r = cik_sdma_gfx_resume(rdev);
4343 if (r)
4344 return r;
4345 r = cik_sdma_rlc_resume(rdev);
4346 if (r)
4347 return r;
4348
4349 return 0;
4350 }
4351
4352 /**
4353 * cik_sdma_fini - tear down the async dma engines
4354 *
4355 * @rdev: radeon_device pointer
4356 *
4357 * Stop the async dma engines and free the rings (CIK).
4358 */
4359 static void cik_sdma_fini(struct radeon_device *rdev)
4360 {
4361 /* stop the gfx rings and rlc compute queues */
4362 cik_sdma_gfx_stop(rdev);
4363 cik_sdma_rlc_stop(rdev);
4364 /* halt the MEs */
4365 cik_sdma_enable(rdev, false);
4366 radeon_ring_fini(rdev, &rdev->ring[R600_RING_TYPE_DMA_INDEX]);
4367 radeon_ring_fini(rdev, &rdev->ring[CAYMAN_RING_TYPE_DMA1_INDEX]);
4368 /* XXX - compute dma queue tear down */
4369 }
4370
4371 /**
4372 * cik_copy_dma - copy pages using the DMA engine
4373 *
4374 * @rdev: radeon_device pointer
4375 * @src_offset: src GPU address
4376 * @dst_offset: dst GPU address
4377 * @num_gpu_pages: number of GPU pages to xfer
4378 * @fence: radeon fence object
4379 *
4380 * Copy GPU paging using the DMA engine (CIK).
4381 * Used by the radeon ttm implementation to move pages if
4382 * registered as the asic copy callback.
4383 */
4384 int cik_copy_dma(struct radeon_device *rdev,
4385 uint64_t src_offset, uint64_t dst_offset,
4386 unsigned num_gpu_pages,
4387 struct radeon_fence **fence)
4388 {
4389 struct radeon_semaphore *sem = NULL;
4390 int ring_index = rdev->asic->copy.dma_ring_index;
4391 struct radeon_ring *ring = &rdev->ring[ring_index];
4392 u32 size_in_bytes, cur_size_in_bytes;
4393 int i, num_loops;
4394 int r = 0;
4395
4396 r = radeon_semaphore_create(rdev, &sem);
4397 if (r) {
4398 DRM_ERROR("radeon: moving bo (%d).\n", r);
4399 return r;
4400 }
4401
4402 size_in_bytes = (num_gpu_pages << RADEON_GPU_PAGE_SHIFT);
4403 num_loops = DIV_ROUND_UP(size_in_bytes, 0x1fffff);
4404 r = radeon_ring_lock(rdev, ring, num_loops * 7 + 14);
4405 if (r) {
4406 DRM_ERROR("radeon: moving bo (%d).\n", r);
4407 radeon_semaphore_free(rdev, &sem, NULL);
4408 return r;
4409 }
4410
4411 if (radeon_fence_need_sync(*fence, ring->idx)) {
4412 radeon_semaphore_sync_rings(rdev, sem, (*fence)->ring,
4413 ring->idx);
4414 radeon_fence_note_sync(*fence, ring->idx);
4415 } else {
4416 radeon_semaphore_free(rdev, &sem, NULL);
4417 }
4418
4419 for (i = 0; i < num_loops; i++) {
4420 cur_size_in_bytes = size_in_bytes;
4421 if (cur_size_in_bytes > 0x1fffff)
4422 cur_size_in_bytes = 0x1fffff;
4423 size_in_bytes -= cur_size_in_bytes;
4424 radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_COPY, SDMA_COPY_SUB_OPCODE_LINEAR, 0));
4425 radeon_ring_write(ring, cur_size_in_bytes);
4426 radeon_ring_write(ring, 0); /* src/dst endian swap */
4427 radeon_ring_write(ring, src_offset & 0xffffffff);
4428 radeon_ring_write(ring, upper_32_bits(src_offset) & 0xffffffff);
4429 radeon_ring_write(ring, dst_offset & 0xfffffffc);
4430 radeon_ring_write(ring, upper_32_bits(dst_offset) & 0xffffffff);
4431 src_offset += cur_size_in_bytes;
4432 dst_offset += cur_size_in_bytes;
4433 }
4434
4435 r = radeon_fence_emit(rdev, fence, ring->idx);
4436 if (r) {
4437 radeon_ring_unlock_undo(rdev, ring);
4438 return r;
4439 }
4440
4441 radeon_ring_unlock_commit(rdev, ring);
4442 radeon_semaphore_free(rdev, &sem, *fence);
4443
4444 return r;
4445 }
4446
4447 /**
4448 * cik_sdma_ring_test - simple async dma engine test
4449 *
4450 * @rdev: radeon_device pointer
4451 * @ring: radeon_ring structure holding ring information
4452 *
4453 * Test the DMA engine by writing using it to write an
4454 * value to memory. (CIK).
4455 * Returns 0 for success, error for failure.
4456 */
4457 int cik_sdma_ring_test(struct radeon_device *rdev,
4458 struct radeon_ring *ring)
4459 {
4460 unsigned i;
4461 int r;
4462 void __iomem *ptr = (void *)rdev->vram_scratch.ptr;
4463 u32 tmp;
4464
4465 if (!ptr) {
4466 DRM_ERROR("invalid vram scratch pointer\n");
4467 return -EINVAL;
4468 }
4469
4470 tmp = 0xCAFEDEAD;
4471 writel(tmp, ptr);
4472
4473 r = radeon_ring_lock(rdev, ring, 4);
4474 if (r) {
4475 DRM_ERROR("radeon: dma failed to lock ring %d (%d).\n", ring->idx, r);
4476 return r;
4477 }
4478 radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_WRITE, SDMA_WRITE_SUB_OPCODE_LINEAR, 0));
4479 radeon_ring_write(ring, rdev->vram_scratch.gpu_addr & 0xfffffffc);
4480 radeon_ring_write(ring, upper_32_bits(rdev->vram_scratch.gpu_addr) & 0xffffffff);
4481 radeon_ring_write(ring, 1); /* number of DWs to follow */
4482 radeon_ring_write(ring, 0xDEADBEEF);
4483 radeon_ring_unlock_commit(rdev, ring);
4484
4485 for (i = 0; i < rdev->usec_timeout; i++) {
4486 tmp = readl(ptr);
4487 if (tmp == 0xDEADBEEF)
4488 break;
4489 DRM_UDELAY(1);
4490 }
4491
4492 if (i < rdev->usec_timeout) {
4493 DRM_INFO("ring test on %d succeeded in %d usecs\n", ring->idx, i);
4494 } else {
4495 DRM_ERROR("radeon: ring %d test failed (0x%08X)\n",
4496 ring->idx, tmp);
4497 r = -EINVAL;
4498 }
4499 return r;
4500 }
4501
4502 /**
4503 * cik_sdma_ib_test - test an IB on the DMA engine
4504 *
4505 * @rdev: radeon_device pointer
4506 * @ring: radeon_ring structure holding ring information
4507 *
4508 * Test a simple IB in the DMA ring (CIK).
4509 * Returns 0 on success, error on failure.
4510 */
4511 int cik_sdma_ib_test(struct radeon_device *rdev, struct radeon_ring *ring)
4512 {
4513 struct radeon_ib ib;
4514 unsigned i;
4515 int r;
4516 void __iomem *ptr = (void *)rdev->vram_scratch.ptr;
4517 u32 tmp = 0;
4518
4519 if (!ptr) {
4520 DRM_ERROR("invalid vram scratch pointer\n");
4521 return -EINVAL;
4522 }
4523
4524 tmp = 0xCAFEDEAD;
4525 writel(tmp, ptr);
4526
4527 r = radeon_ib_get(rdev, ring->idx, &ib, NULL, 256);
4528 if (r) {
4529 DRM_ERROR("radeon: failed to get ib (%d).\n", r);
4530 return r;
4531 }
4532
4533 ib.ptr[0] = SDMA_PACKET(SDMA_OPCODE_WRITE, SDMA_WRITE_SUB_OPCODE_LINEAR, 0);
4534 ib.ptr[1] = rdev->vram_scratch.gpu_addr & 0xfffffffc;
4535 ib.ptr[2] = upper_32_bits(rdev->vram_scratch.gpu_addr) & 0xffffffff;
4536 ib.ptr[3] = 1;
4537 ib.ptr[4] = 0xDEADBEEF;
4538 ib.length_dw = 5;
4539
4540 r = radeon_ib_schedule(rdev, &ib, NULL);
4541 if (r) {
4542 radeon_ib_free(rdev, &ib);
4543 DRM_ERROR("radeon: failed to schedule ib (%d).\n", r);
4544 return r;
4545 }
4546 r = radeon_fence_wait(ib.fence, false);
4547 if (r) {
4548 DRM_ERROR("radeon: fence wait failed (%d).\n", r);
4549 return r;
4550 }
4551 for (i = 0; i < rdev->usec_timeout; i++) {
4552 tmp = readl(ptr);
4553 if (tmp == 0xDEADBEEF)
4554 break;
4555 DRM_UDELAY(1);
4556 }
4557 if (i < rdev->usec_timeout) {
4558 DRM_INFO("ib test on ring %d succeeded in %u usecs\n", ib.fence->ring, i);
4559 } else {
4560 DRM_ERROR("radeon: ib test failed (0x%08X)\n", tmp);
4561 r = -EINVAL;
4562 }
4563 radeon_ib_free(rdev, &ib);
4564 return r;
4565 }
4566
4567
4568 static void cik_print_gpu_status_regs(struct radeon_device *rdev)
4569 {
4570 dev_info(rdev->dev, " GRBM_STATUS=0x%08X\n",
4571 RREG32(GRBM_STATUS));
4572 dev_info(rdev->dev, " GRBM_STATUS2=0x%08X\n",
4573 RREG32(GRBM_STATUS2));
4574 dev_info(rdev->dev, " GRBM_STATUS_SE0=0x%08X\n",
4575 RREG32(GRBM_STATUS_SE0));
4576 dev_info(rdev->dev, " GRBM_STATUS_SE1=0x%08X\n",
4577 RREG32(GRBM_STATUS_SE1));
4578 dev_info(rdev->dev, " GRBM_STATUS_SE2=0x%08X\n",
4579 RREG32(GRBM_STATUS_SE2));
4580 dev_info(rdev->dev, " GRBM_STATUS_SE3=0x%08X\n",
4581 RREG32(GRBM_STATUS_SE3));
4582 dev_info(rdev->dev, " SRBM_STATUS=0x%08X\n",
4583 RREG32(SRBM_STATUS));
4584 dev_info(rdev->dev, " SRBM_STATUS2=0x%08X\n",
4585 RREG32(SRBM_STATUS2));
4586 dev_info(rdev->dev, " SDMA0_STATUS_REG = 0x%08X\n",
4587 RREG32(SDMA0_STATUS_REG + SDMA0_REGISTER_OFFSET));
4588 dev_info(rdev->dev, " SDMA1_STATUS_REG = 0x%08X\n",
4589 RREG32(SDMA0_STATUS_REG + SDMA1_REGISTER_OFFSET));
4590 dev_info(rdev->dev, " CP_STAT = 0x%08x\n", RREG32(CP_STAT));
4591 dev_info(rdev->dev, " CP_STALLED_STAT1 = 0x%08x\n",
4592 RREG32(CP_STALLED_STAT1));
4593 dev_info(rdev->dev, " CP_STALLED_STAT2 = 0x%08x\n",
4594 RREG32(CP_STALLED_STAT2));
4595 dev_info(rdev->dev, " CP_STALLED_STAT3 = 0x%08x\n",
4596 RREG32(CP_STALLED_STAT3));
4597 dev_info(rdev->dev, " CP_CPF_BUSY_STAT = 0x%08x\n",
4598 RREG32(CP_CPF_BUSY_STAT));
4599 dev_info(rdev->dev, " CP_CPF_STALLED_STAT1 = 0x%08x\n",
4600 RREG32(CP_CPF_STALLED_STAT1));
4601 dev_info(rdev->dev, " CP_CPF_STATUS = 0x%08x\n", RREG32(CP_CPF_STATUS));
4602 dev_info(rdev->dev, " CP_CPC_BUSY_STAT = 0x%08x\n", RREG32(CP_CPC_BUSY_STAT));
4603 dev_info(rdev->dev, " CP_CPC_STALLED_STAT1 = 0x%08x\n",
4604 RREG32(CP_CPC_STALLED_STAT1));
4605 dev_info(rdev->dev, " CP_CPC_STATUS = 0x%08x\n", RREG32(CP_CPC_STATUS));
4606 }
4607
4608 /**
4609 * cik_gpu_check_soft_reset - check which blocks are busy
4610 *
4611 * @rdev: radeon_device pointer
4612 *
4613 * Check which blocks are busy and return the relevant reset
4614 * mask to be used by cik_gpu_soft_reset().
4615 * Returns a mask of the blocks to be reset.
4616 */
4617 static u32 cik_gpu_check_soft_reset(struct radeon_device *rdev)
4618 {
4619 u32 reset_mask = 0;
4620 u32 tmp;
4621
4622 /* GRBM_STATUS */
4623 tmp = RREG32(GRBM_STATUS);
4624 if (tmp & (PA_BUSY | SC_BUSY |
4625 BCI_BUSY | SX_BUSY |
4626 TA_BUSY | VGT_BUSY |
4627 DB_BUSY | CB_BUSY |
4628 GDS_BUSY | SPI_BUSY |
4629 IA_BUSY | IA_BUSY_NO_DMA))
4630 reset_mask |= RADEON_RESET_GFX;
4631
4632 if (tmp & (CP_BUSY | CP_COHERENCY_BUSY))
4633 reset_mask |= RADEON_RESET_CP;
4634
4635 /* GRBM_STATUS2 */
4636 tmp = RREG32(GRBM_STATUS2);
4637 if (tmp & RLC_BUSY)
4638 reset_mask |= RADEON_RESET_RLC;
4639
4640 /* SDMA0_STATUS_REG */
4641 tmp = RREG32(SDMA0_STATUS_REG + SDMA0_REGISTER_OFFSET);
4642 if (!(tmp & SDMA_IDLE))
4643 reset_mask |= RADEON_RESET_DMA;
4644
4645 /* SDMA1_STATUS_REG */
4646 tmp = RREG32(SDMA0_STATUS_REG + SDMA1_REGISTER_OFFSET);
4647 if (!(tmp & SDMA_IDLE))
4648 reset_mask |= RADEON_RESET_DMA1;
4649
4650 /* SRBM_STATUS2 */
4651 tmp = RREG32(SRBM_STATUS2);
4652 if (tmp & SDMA_BUSY)
4653 reset_mask |= RADEON_RESET_DMA;
4654
4655 if (tmp & SDMA1_BUSY)
4656 reset_mask |= RADEON_RESET_DMA1;
4657
4658 /* SRBM_STATUS */
4659 tmp = RREG32(SRBM_STATUS);
4660
4661 if (tmp & IH_BUSY)
4662 reset_mask |= RADEON_RESET_IH;
4663
4664 if (tmp & SEM_BUSY)
4665 reset_mask |= RADEON_RESET_SEM;
4666
4667 if (tmp & GRBM_RQ_PENDING)
4668 reset_mask |= RADEON_RESET_GRBM;
4669
4670 if (tmp & VMC_BUSY)
4671 reset_mask |= RADEON_RESET_VMC;
4672
4673 if (tmp & (MCB_BUSY | MCB_NON_DISPLAY_BUSY |
4674 MCC_BUSY | MCD_BUSY))
4675 reset_mask |= RADEON_RESET_MC;
4676
4677 if (evergreen_is_display_hung(rdev))
4678 reset_mask |= RADEON_RESET_DISPLAY;
4679
4680 /* Skip MC reset as it's mostly likely not hung, just busy */
4681 if (reset_mask & RADEON_RESET_MC) {
4682 DRM_DEBUG("MC busy: 0x%08X, clearing.\n", reset_mask);
4683 reset_mask &= ~RADEON_RESET_MC;
4684 }
4685
4686 return reset_mask;
4687 }
4688
4689 /**
4690 * cik_gpu_soft_reset - soft reset GPU
4691 *
4692 * @rdev: radeon_device pointer
4693 * @reset_mask: mask of which blocks to reset
4694 *
4695 * Soft reset the blocks specified in @reset_mask.
4696 */
4697 static void cik_gpu_soft_reset(struct radeon_device *rdev, u32 reset_mask)
4698 {
4699 struct evergreen_mc_save save;
4700 u32 grbm_soft_reset = 0, srbm_soft_reset = 0;
4701 u32 tmp;
4702
4703 if (reset_mask == 0)
4704 return;
4705
4706 dev_info(rdev->dev, "GPU softreset: 0x%08X\n", reset_mask);
4707
4708 cik_print_gpu_status_regs(rdev);
4709 dev_info(rdev->dev, " VM_CONTEXT1_PROTECTION_FAULT_ADDR 0x%08X\n",
4710 RREG32(VM_CONTEXT1_PROTECTION_FAULT_ADDR));
4711 dev_info(rdev->dev, " VM_CONTEXT1_PROTECTION_FAULT_STATUS 0x%08X\n",
4712 RREG32(VM_CONTEXT1_PROTECTION_FAULT_STATUS));
4713
4714 /* stop the rlc */
4715 cik_rlc_stop(rdev);
4716
4717 /* Disable GFX parsing/prefetching */
4718 WREG32(CP_ME_CNTL, CP_ME_HALT | CP_PFP_HALT | CP_CE_HALT);
4719
4720 /* Disable MEC parsing/prefetching */
4721 WREG32(CP_MEC_CNTL, MEC_ME1_HALT | MEC_ME2_HALT);
4722
4723 if (reset_mask & RADEON_RESET_DMA) {
4724 /* sdma0 */
4725 tmp = RREG32(SDMA0_ME_CNTL + SDMA0_REGISTER_OFFSET);
4726 tmp |= SDMA_HALT;
4727 WREG32(SDMA0_ME_CNTL + SDMA0_REGISTER_OFFSET, tmp);
4728 }
4729 if (reset_mask & RADEON_RESET_DMA1) {
4730 /* sdma1 */
4731 tmp = RREG32(SDMA0_ME_CNTL + SDMA1_REGISTER_OFFSET);
4732 tmp |= SDMA_HALT;
4733 WREG32(SDMA0_ME_CNTL + SDMA1_REGISTER_OFFSET, tmp);
4734 }
4735
4736 evergreen_mc_stop(rdev, &save);
4737 if (evergreen_mc_wait_for_idle(rdev)) {
4738 dev_warn(rdev->dev, "Wait for MC idle timedout !\n");
4739 }
4740
4741 if (reset_mask & (RADEON_RESET_GFX | RADEON_RESET_COMPUTE | RADEON_RESET_CP))
4742 grbm_soft_reset = SOFT_RESET_CP | SOFT_RESET_GFX;
4743
4744 if (reset_mask & RADEON_RESET_CP) {
4745 grbm_soft_reset |= SOFT_RESET_CP;
4746
4747 srbm_soft_reset |= SOFT_RESET_GRBM;
4748 }
4749
4750 if (reset_mask & RADEON_RESET_DMA)
4751 srbm_soft_reset |= SOFT_RESET_SDMA;
4752
4753 if (reset_mask & RADEON_RESET_DMA1)
4754 srbm_soft_reset |= SOFT_RESET_SDMA1;
4755
4756 if (reset_mask & RADEON_RESET_DISPLAY)
4757 srbm_soft_reset |= SOFT_RESET_DC;
4758
4759 if (reset_mask & RADEON_RESET_RLC)
4760 grbm_soft_reset |= SOFT_RESET_RLC;
4761
4762 if (reset_mask & RADEON_RESET_SEM)
4763 srbm_soft_reset |= SOFT_RESET_SEM;
4764
4765 if (reset_mask & RADEON_RESET_IH)
4766 srbm_soft_reset |= SOFT_RESET_IH;
4767
4768 if (reset_mask & RADEON_RESET_GRBM)
4769 srbm_soft_reset |= SOFT_RESET_GRBM;
4770
4771 if (reset_mask & RADEON_RESET_VMC)
4772 srbm_soft_reset |= SOFT_RESET_VMC;
4773
4774 if (!(rdev->flags & RADEON_IS_IGP)) {
4775 if (reset_mask & RADEON_RESET_MC)
4776 srbm_soft_reset |= SOFT_RESET_MC;
4777 }
4778
4779 if (grbm_soft_reset) {
4780 tmp = RREG32(GRBM_SOFT_RESET);
4781 tmp |= grbm_soft_reset;
4782 dev_info(rdev->dev, "GRBM_SOFT_RESET=0x%08X\n", tmp);
4783 WREG32(GRBM_SOFT_RESET, tmp);
4784 tmp = RREG32(GRBM_SOFT_RESET);
4785
4786 udelay(50);
4787
4788 tmp &= ~grbm_soft_reset;
4789 WREG32(GRBM_SOFT_RESET, tmp);
4790 tmp = RREG32(GRBM_SOFT_RESET);
4791 }
4792
4793 if (srbm_soft_reset) {
4794 tmp = RREG32(SRBM_SOFT_RESET);
4795 tmp |= srbm_soft_reset;
4796 dev_info(rdev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp);
4797 WREG32(SRBM_SOFT_RESET, tmp);
4798 tmp = RREG32(SRBM_SOFT_RESET);
4799
4800 udelay(50);
4801
4802 tmp &= ~srbm_soft_reset;
4803 WREG32(SRBM_SOFT_RESET, tmp);
4804 tmp = RREG32(SRBM_SOFT_RESET);
4805 }
4806
4807 /* Wait a little for things to settle down */
4808 udelay(50);
4809
4810 evergreen_mc_resume(rdev, &save);
4811 udelay(50);
4812
4813 cik_print_gpu_status_regs(rdev);
4814 }
4815
4816 /**
4817 * cik_asic_reset - soft reset GPU
4818 *
4819 * @rdev: radeon_device pointer
4820 *
4821 * Look up which blocks are hung and attempt
4822 * to reset them.
4823 * Returns 0 for success.
4824 */
4825 int cik_asic_reset(struct radeon_device *rdev)
4826 {
4827 u32 reset_mask;
4828
4829 reset_mask = cik_gpu_check_soft_reset(rdev);
4830
4831 if (reset_mask)
4832 r600_set_bios_scratch_engine_hung(rdev, true);
4833
4834 cik_gpu_soft_reset(rdev, reset_mask);
4835
4836 reset_mask = cik_gpu_check_soft_reset(rdev);
4837
4838 if (!reset_mask)
4839 r600_set_bios_scratch_engine_hung(rdev, false);
4840
4841 return 0;
4842 }
4843
4844 /**
4845 * cik_gfx_is_lockup - check if the 3D engine is locked up
4846 *
4847 * @rdev: radeon_device pointer
4848 * @ring: radeon_ring structure holding ring information
4849 *
4850 * Check if the 3D engine is locked up (CIK).
4851 * Returns true if the engine is locked, false if not.
4852 */
4853 bool cik_gfx_is_lockup(struct radeon_device *rdev, struct radeon_ring *ring)
4854 {
4855 u32 reset_mask = cik_gpu_check_soft_reset(rdev);
4856
4857 if (!(reset_mask & (RADEON_RESET_GFX |
4858 RADEON_RESET_COMPUTE |
4859 RADEON_RESET_CP))) {
4860 radeon_ring_lockup_update(ring);
4861 return false;
4862 }
4863 /* force CP activities */
4864 radeon_ring_force_activity(rdev, ring);
4865 return radeon_ring_test_lockup(rdev, ring);
4866 }
4867
4868 /**
4869 * cik_sdma_is_lockup - Check if the DMA engine is locked up
4870 *
4871 * @rdev: radeon_device pointer
4872 * @ring: radeon_ring structure holding ring information
4873 *
4874 * Check if the async DMA engine is locked up (CIK).
4875 * Returns true if the engine appears to be locked up, false if not.
4876 */
4877 bool cik_sdma_is_lockup(struct radeon_device *rdev, struct radeon_ring *ring)
4878 {
4879 u32 reset_mask = cik_gpu_check_soft_reset(rdev);
4880 u32 mask;
4881
4882 if (ring->idx == R600_RING_TYPE_DMA_INDEX)
4883 mask = RADEON_RESET_DMA;
4884 else
4885 mask = RADEON_RESET_DMA1;
4886
4887 if (!(reset_mask & mask)) {
4888 radeon_ring_lockup_update(ring);
4889 return false;
4890 }
4891 /* force ring activities */
4892 radeon_ring_force_activity(rdev, ring);
4893 return radeon_ring_test_lockup(rdev, ring);
4894 }
4895
4896 /* MC */
4897 /**
4898 * cik_mc_program - program the GPU memory controller
4899 *
4900 * @rdev: radeon_device pointer
4901 *
4902 * Set the location of vram, gart, and AGP in the GPU's
4903 * physical address space (CIK).
4904 */
4905 static void cik_mc_program(struct radeon_device *rdev)
4906 {
4907 struct evergreen_mc_save save;
4908 u32 tmp;
4909 int i, j;
4910
4911 /* Initialize HDP */
4912 for (i = 0, j = 0; i < 32; i++, j += 0x18) {
4913 WREG32((0x2c14 + j), 0x00000000);
4914 WREG32((0x2c18 + j), 0x00000000);
4915 WREG32((0x2c1c + j), 0x00000000);
4916 WREG32((0x2c20 + j), 0x00000000);
4917 WREG32((0x2c24 + j), 0x00000000);
4918 }
4919 WREG32(HDP_REG_COHERENCY_FLUSH_CNTL, 0);
4920
4921 evergreen_mc_stop(rdev, &save);
4922 if (radeon_mc_wait_for_idle(rdev)) {
4923 dev_warn(rdev->dev, "Wait for MC idle timedout !\n");
4924 }
4925 /* Lockout access through VGA aperture*/
4926 WREG32(VGA_HDP_CONTROL, VGA_MEMORY_DISABLE);
4927 /* Update configuration */
4928 WREG32(MC_VM_SYSTEM_APERTURE_LOW_ADDR,
4929 rdev->mc.vram_start >> 12);
4930 WREG32(MC_VM_SYSTEM_APERTURE_HIGH_ADDR,
4931 rdev->mc.vram_end >> 12);
4932 WREG32(MC_VM_SYSTEM_APERTURE_DEFAULT_ADDR,
4933 rdev->vram_scratch.gpu_addr >> 12);
4934 tmp = ((rdev->mc.vram_end >> 24) & 0xFFFF) << 16;
4935 tmp |= ((rdev->mc.vram_start >> 24) & 0xFFFF);
4936 WREG32(MC_VM_FB_LOCATION, tmp);
4937 /* XXX double check these! */
4938 WREG32(HDP_NONSURFACE_BASE, (rdev->mc.vram_start >> 8));
4939 WREG32(HDP_NONSURFACE_INFO, (2 << 7) | (1 << 30));
4940 WREG32(HDP_NONSURFACE_SIZE, 0x3FFFFFFF);
4941 WREG32(MC_VM_AGP_BASE, 0);
4942 WREG32(MC_VM_AGP_TOP, 0x0FFFFFFF);
4943 WREG32(MC_VM_AGP_BOT, 0x0FFFFFFF);
4944 if (radeon_mc_wait_for_idle(rdev)) {
4945 dev_warn(rdev->dev, "Wait for MC idle timedout !\n");
4946 }
4947 evergreen_mc_resume(rdev, &save);
4948 /* we need to own VRAM, so turn off the VGA renderer here
4949 * to stop it overwriting our objects */
4950 rv515_vga_render_disable(rdev);
4951 }
4952
4953 /**
4954 * cik_mc_init - initialize the memory controller driver params
4955 *
4956 * @rdev: radeon_device pointer
4957 *
4958 * Look up the amount of vram, vram width, and decide how to place
4959 * vram and gart within the GPU's physical address space (CIK).
4960 * Returns 0 for success.
4961 */
4962 static int cik_mc_init(struct radeon_device *rdev)
4963 {
4964 u32 tmp;
4965 int chansize, numchan;
4966
4967 /* Get VRAM informations */
4968 rdev->mc.vram_is_ddr = true;
4969 tmp = RREG32(MC_ARB_RAMCFG);
4970 if (tmp & CHANSIZE_MASK) {
4971 chansize = 64;
4972 } else {
4973 chansize = 32;
4974 }
4975 tmp = RREG32(MC_SHARED_CHMAP);
4976 switch ((tmp & NOOFCHAN_MASK) >> NOOFCHAN_SHIFT) {
4977 case 0:
4978 default:
4979 numchan = 1;
4980 break;
4981 case 1:
4982 numchan = 2;
4983 break;
4984 case 2:
4985 numchan = 4;
4986 break;
4987 case 3:
4988 numchan = 8;
4989 break;
4990 case 4:
4991 numchan = 3;
4992 break;
4993 case 5:
4994 numchan = 6;
4995 break;
4996 case 6:
4997 numchan = 10;
4998 break;
4999 case 7:
5000 numchan = 12;
5001 break;
5002 case 8:
5003 numchan = 16;
5004 break;
5005 }
5006 rdev->mc.vram_width = numchan * chansize;
5007 /* Could aper size report 0 ? */
5008 rdev->mc.aper_base = pci_resource_start(rdev->pdev, 0);
5009 rdev->mc.aper_size = pci_resource_len(rdev->pdev, 0);
5010 /* size in MB on si */
5011 rdev->mc.mc_vram_size = RREG32(CONFIG_MEMSIZE) * 1024 * 1024;
5012 rdev->mc.real_vram_size = RREG32(CONFIG_MEMSIZE) * 1024 * 1024;
5013 rdev->mc.visible_vram_size = rdev->mc.aper_size;
5014 si_vram_gtt_location(rdev, &rdev->mc);
5015 radeon_update_bandwidth_info(rdev);
5016
5017 return 0;
5018 }
5019
5020 /*
5021 * GART
5022 * VMID 0 is the physical GPU addresses as used by the kernel.
5023 * VMIDs 1-15 are used for userspace clients and are handled
5024 * by the radeon vm/hsa code.
5025 */
5026 /**
5027 * cik_pcie_gart_tlb_flush - gart tlb flush callback
5028 *
5029 * @rdev: radeon_device pointer
5030 *
5031 * Flush the TLB for the VMID 0 page table (CIK).
5032 */
5033 void cik_pcie_gart_tlb_flush(struct radeon_device *rdev)
5034 {
5035 /* flush hdp cache */
5036 WREG32(HDP_MEM_COHERENCY_FLUSH_CNTL, 0);
5037
5038 /* bits 0-15 are the VM contexts0-15 */
5039 WREG32(VM_INVALIDATE_REQUEST, 0x1);
5040 }
5041
5042 /**
5043 * cik_pcie_gart_enable - gart enable
5044 *
5045 * @rdev: radeon_device pointer
5046 *
5047 * This sets up the TLBs, programs the page tables for VMID0,
5048 * sets up the hw for VMIDs 1-15 which are allocated on
5049 * demand, and sets up the global locations for the LDS, GDS,
5050 * and GPUVM for FSA64 clients (CIK).
5051 * Returns 0 for success, errors for failure.
5052 */
5053 static int cik_pcie_gart_enable(struct radeon_device *rdev)
5054 {
5055 int r, i;
5056
5057 if (rdev->gart.robj == NULL) {
5058 dev_err(rdev->dev, "No VRAM object for PCIE GART.\n");
5059 return -EINVAL;
5060 }
5061 r = radeon_gart_table_vram_pin(rdev);
5062 if (r)
5063 return r;
5064 radeon_gart_restore(rdev);
5065 /* Setup TLB control */
5066 WREG32(MC_VM_MX_L1_TLB_CNTL,
5067 (0xA << 7) |
5068 ENABLE_L1_TLB |
5069 SYSTEM_ACCESS_MODE_NOT_IN_SYS |
5070 ENABLE_ADVANCED_DRIVER_MODEL |
5071 SYSTEM_APERTURE_UNMAPPED_ACCESS_PASS_THRU);
5072 /* Setup L2 cache */
5073 WREG32(VM_L2_CNTL, ENABLE_L2_CACHE |
5074 ENABLE_L2_FRAGMENT_PROCESSING |
5075 ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE |
5076 ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE |
5077 EFFECTIVE_L2_QUEUE_SIZE(7) |
5078 CONTEXT1_IDENTITY_ACCESS_MODE(1));
5079 WREG32(VM_L2_CNTL2, INVALIDATE_ALL_L1_TLBS | INVALIDATE_L2_CACHE);
5080 WREG32(VM_L2_CNTL3, L2_CACHE_BIGK_ASSOCIATIVITY |
5081 L2_CACHE_BIGK_FRAGMENT_SIZE(6));
5082 /* setup context0 */
5083 WREG32(VM_CONTEXT0_PAGE_TABLE_START_ADDR, rdev->mc.gtt_start >> 12);
5084 WREG32(VM_CONTEXT0_PAGE_TABLE_END_ADDR, rdev->mc.gtt_end >> 12);
5085 WREG32(VM_CONTEXT0_PAGE_TABLE_BASE_ADDR, rdev->gart.table_addr >> 12);
5086 WREG32(VM_CONTEXT0_PROTECTION_FAULT_DEFAULT_ADDR,
5087 (u32)(rdev->dummy_page.addr >> 12));
5088 WREG32(VM_CONTEXT0_CNTL2, 0);
5089 WREG32(VM_CONTEXT0_CNTL, (ENABLE_CONTEXT | PAGE_TABLE_DEPTH(0) |
5090 RANGE_PROTECTION_FAULT_ENABLE_DEFAULT));
5091
5092 WREG32(0x15D4, 0);
5093 WREG32(0x15D8, 0);
5094 WREG32(0x15DC, 0);
5095
5096 /* empty context1-15 */
5097 /* FIXME start with 4G, once using 2 level pt switch to full
5098 * vm size space
5099 */
5100 /* set vm size, must be a multiple of 4 */
5101 WREG32(VM_CONTEXT1_PAGE_TABLE_START_ADDR, 0);
5102 WREG32(VM_CONTEXT1_PAGE_TABLE_END_ADDR, rdev->vm_manager.max_pfn);
5103 for (i = 1; i < 16; i++) {
5104 if (i < 8)
5105 WREG32(VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + (i << 2),
5106 rdev->gart.table_addr >> 12);
5107 else
5108 WREG32(VM_CONTEXT8_PAGE_TABLE_BASE_ADDR + ((i - 8) << 2),
5109 rdev->gart.table_addr >> 12);
5110 }
5111
5112 /* enable context1-15 */
5113 WREG32(VM_CONTEXT1_PROTECTION_FAULT_DEFAULT_ADDR,
5114 (u32)(rdev->dummy_page.addr >> 12));
5115 WREG32(VM_CONTEXT1_CNTL2, 4);
5116 WREG32(VM_CONTEXT1_CNTL, ENABLE_CONTEXT | PAGE_TABLE_DEPTH(1) |
5117 RANGE_PROTECTION_FAULT_ENABLE_INTERRUPT |
5118 RANGE_PROTECTION_FAULT_ENABLE_DEFAULT |
5119 DUMMY_PAGE_PROTECTION_FAULT_ENABLE_INTERRUPT |
5120 DUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT |
5121 PDE0_PROTECTION_FAULT_ENABLE_INTERRUPT |
5122 PDE0_PROTECTION_FAULT_ENABLE_DEFAULT |
5123 VALID_PROTECTION_FAULT_ENABLE_INTERRUPT |
5124 VALID_PROTECTION_FAULT_ENABLE_DEFAULT |
5125 READ_PROTECTION_FAULT_ENABLE_INTERRUPT |
5126 READ_PROTECTION_FAULT_ENABLE_DEFAULT |
5127 WRITE_PROTECTION_FAULT_ENABLE_INTERRUPT |
5128 WRITE_PROTECTION_FAULT_ENABLE_DEFAULT);
5129
5130 /* TC cache setup ??? */
5131 WREG32(TC_CFG_L1_LOAD_POLICY0, 0);
5132 WREG32(TC_CFG_L1_LOAD_POLICY1, 0);
5133 WREG32(TC_CFG_L1_STORE_POLICY, 0);
5134
5135 WREG32(TC_CFG_L2_LOAD_POLICY0, 0);
5136 WREG32(TC_CFG_L2_LOAD_POLICY1, 0);
5137 WREG32(TC_CFG_L2_STORE_POLICY0, 0);
5138 WREG32(TC_CFG_L2_STORE_POLICY1, 0);
5139 WREG32(TC_CFG_L2_ATOMIC_POLICY, 0);
5140
5141 WREG32(TC_CFG_L1_VOLATILE, 0);
5142 WREG32(TC_CFG_L2_VOLATILE, 0);
5143
5144 if (rdev->family == CHIP_KAVERI) {
5145 u32 tmp = RREG32(CHUB_CONTROL);
5146 tmp &= ~BYPASS_VM;
5147 WREG32(CHUB_CONTROL, tmp);
5148 }
5149
5150 /* XXX SH_MEM regs */
5151 /* where to put LDS, scratch, GPUVM in FSA64 space */
5152 mutex_lock(&rdev->srbm_mutex);
5153 for (i = 0; i < 16; i++) {
5154 cik_srbm_select(rdev, 0, 0, 0, i);
5155 /* CP and shaders */
5156 WREG32(SH_MEM_CONFIG, 0);
5157 WREG32(SH_MEM_APE1_BASE, 1);
5158 WREG32(SH_MEM_APE1_LIMIT, 0);
5159 WREG32(SH_MEM_BASES, 0);
5160 /* SDMA GFX */
5161 WREG32(SDMA0_GFX_VIRTUAL_ADDR + SDMA0_REGISTER_OFFSET, 0);
5162 WREG32(SDMA0_GFX_APE1_CNTL + SDMA0_REGISTER_OFFSET, 0);
5163 WREG32(SDMA0_GFX_VIRTUAL_ADDR + SDMA1_REGISTER_OFFSET, 0);
5164 WREG32(SDMA0_GFX_APE1_CNTL + SDMA1_REGISTER_OFFSET, 0);
5165 /* XXX SDMA RLC - todo */
5166 }
5167 cik_srbm_select(rdev, 0, 0, 0, 0);
5168 mutex_unlock(&rdev->srbm_mutex);
5169
5170 cik_pcie_gart_tlb_flush(rdev);
5171 DRM_INFO("PCIE GART of %uM enabled (table at 0x%016llX).\n",
5172 (unsigned)(rdev->mc.gtt_size >> 20),
5173 (unsigned long long)rdev->gart.table_addr);
5174 rdev->gart.ready = true;
5175 return 0;
5176 }
5177
5178 /**
5179 * cik_pcie_gart_disable - gart disable
5180 *
5181 * @rdev: radeon_device pointer
5182 *
5183 * This disables all VM page table (CIK).
5184 */
5185 static void cik_pcie_gart_disable(struct radeon_device *rdev)
5186 {
5187 /* Disable all tables */
5188 WREG32(VM_CONTEXT0_CNTL, 0);
5189 WREG32(VM_CONTEXT1_CNTL, 0);
5190 /* Setup TLB control */
5191 WREG32(MC_VM_MX_L1_TLB_CNTL, SYSTEM_ACCESS_MODE_NOT_IN_SYS |
5192 SYSTEM_APERTURE_UNMAPPED_ACCESS_PASS_THRU);
5193 /* Setup L2 cache */
5194 WREG32(VM_L2_CNTL,
5195 ENABLE_L2_FRAGMENT_PROCESSING |
5196 ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE |
5197 ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE |
5198 EFFECTIVE_L2_QUEUE_SIZE(7) |
5199 CONTEXT1_IDENTITY_ACCESS_MODE(1));
5200 WREG32(VM_L2_CNTL2, 0);
5201 WREG32(VM_L2_CNTL3, L2_CACHE_BIGK_ASSOCIATIVITY |
5202 L2_CACHE_BIGK_FRAGMENT_SIZE(6));
5203 radeon_gart_table_vram_unpin(rdev);
5204 }
5205
5206 /**
5207 * cik_pcie_gart_fini - vm fini callback
5208 *
5209 * @rdev: radeon_device pointer
5210 *
5211 * Tears down the driver GART/VM setup (CIK).
5212 */
5213 static void cik_pcie_gart_fini(struct radeon_device *rdev)
5214 {
5215 cik_pcie_gart_disable(rdev);
5216 radeon_gart_table_vram_free(rdev);
5217 radeon_gart_fini(rdev);
5218 }
5219
5220 /* vm parser */
5221 /**
5222 * cik_ib_parse - vm ib_parse callback
5223 *
5224 * @rdev: radeon_device pointer
5225 * @ib: indirect buffer pointer
5226 *
5227 * CIK uses hw IB checking so this is a nop (CIK).
5228 */
5229 int cik_ib_parse(struct radeon_device *rdev, struct radeon_ib *ib)
5230 {
5231 return 0;
5232 }
5233
5234 /*
5235 * vm
5236 * VMID 0 is the physical GPU addresses as used by the kernel.
5237 * VMIDs 1-15 are used for userspace clients and are handled
5238 * by the radeon vm/hsa code.
5239 */
5240 /**
5241 * cik_vm_init - cik vm init callback
5242 *
5243 * @rdev: radeon_device pointer
5244 *
5245 * Inits cik specific vm parameters (number of VMs, base of vram for
5246 * VMIDs 1-15) (CIK).
5247 * Returns 0 for success.
5248 */
5249 int cik_vm_init(struct radeon_device *rdev)
5250 {
5251 /* number of VMs */
5252 rdev->vm_manager.nvm = 16;
5253 /* base offset of vram pages */
5254 if (rdev->flags & RADEON_IS_IGP) {
5255 u64 tmp = RREG32(MC_VM_FB_OFFSET);
5256 tmp <<= 22;
5257 rdev->vm_manager.vram_base_offset = tmp;
5258 } else
5259 rdev->vm_manager.vram_base_offset = 0;
5260
5261 return 0;
5262 }
5263
5264 /**
5265 * cik_vm_fini - cik vm fini callback
5266 *
5267 * @rdev: radeon_device pointer
5268 *
5269 * Tear down any asic specific VM setup (CIK).
5270 */
5271 void cik_vm_fini(struct radeon_device *rdev)
5272 {
5273 }
5274
5275 /**
5276 * cik_vm_decode_fault - print human readable fault info
5277 *
5278 * @rdev: radeon_device pointer
5279 * @status: VM_CONTEXT1_PROTECTION_FAULT_STATUS register value
5280 * @addr: VM_CONTEXT1_PROTECTION_FAULT_ADDR register value
5281 *
5282 * Print human readable fault information (CIK).
5283 */
5284 static void cik_vm_decode_fault(struct radeon_device *rdev,
5285 u32 status, u32 addr, u32 mc_client)
5286 {
5287 u32 mc_id = (status & MEMORY_CLIENT_ID_MASK) >> MEMORY_CLIENT_ID_SHIFT;
5288 u32 vmid = (status & FAULT_VMID_MASK) >> FAULT_VMID_SHIFT;
5289 u32 protections = (status & PROTECTIONS_MASK) >> PROTECTIONS_SHIFT;
5290 char *block = (char *)&mc_client;
5291
5292 printk("VM fault (0x%02x, vmid %d) at page %u, %s from %s (%d)\n",
5293 protections, vmid, addr,
5294 (status & MEMORY_CLIENT_RW_MASK) ? "write" : "read",
5295 block, mc_id);
5296 }
5297
5298 /**
5299 * cik_vm_flush - cik vm flush using the CP
5300 *
5301 * @rdev: radeon_device pointer
5302 *
5303 * Update the page table base and flush the VM TLB
5304 * using the CP (CIK).
5305 */
5306 void cik_vm_flush(struct radeon_device *rdev, int ridx, struct radeon_vm *vm)
5307 {
5308 struct radeon_ring *ring = &rdev->ring[ridx];
5309
5310 if (vm == NULL)
5311 return;
5312
5313 radeon_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5314 radeon_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
5315 WRITE_DATA_DST_SEL(0)));
5316 if (vm->id < 8) {
5317 radeon_ring_write(ring,
5318 (VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + (vm->id << 2)) >> 2);
5319 } else {
5320 radeon_ring_write(ring,
5321 (VM_CONTEXT8_PAGE_TABLE_BASE_ADDR + ((vm->id - 8) << 2)) >> 2);
5322 }
5323 radeon_ring_write(ring, 0);
5324 radeon_ring_write(ring, vm->pd_gpu_addr >> 12);
5325
5326 /* update SH_MEM_* regs */
5327 radeon_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5328 radeon_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
5329 WRITE_DATA_DST_SEL(0)));
5330 radeon_ring_write(ring, SRBM_GFX_CNTL >> 2);
5331 radeon_ring_write(ring, 0);
5332 radeon_ring_write(ring, VMID(vm->id));
5333
5334 radeon_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 6));
5335 radeon_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
5336 WRITE_DATA_DST_SEL(0)));
5337 radeon_ring_write(ring, SH_MEM_BASES >> 2);
5338 radeon_ring_write(ring, 0);
5339
5340 radeon_ring_write(ring, 0); /* SH_MEM_BASES */
5341 radeon_ring_write(ring, 0); /* SH_MEM_CONFIG */
5342 radeon_ring_write(ring, 1); /* SH_MEM_APE1_BASE */
5343 radeon_ring_write(ring, 0); /* SH_MEM_APE1_LIMIT */
5344
5345 radeon_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5346 radeon_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
5347 WRITE_DATA_DST_SEL(0)));
5348 radeon_ring_write(ring, SRBM_GFX_CNTL >> 2);
5349 radeon_ring_write(ring, 0);
5350 radeon_ring_write(ring, VMID(0));
5351
5352 /* HDP flush */
5353 /* We should be using the WAIT_REG_MEM packet here like in
5354 * cik_fence_ring_emit(), but it causes the CP to hang in this
5355 * context...
5356 */
5357 radeon_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5358 radeon_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
5359 WRITE_DATA_DST_SEL(0)));
5360 radeon_ring_write(ring, HDP_MEM_COHERENCY_FLUSH_CNTL >> 2);
5361 radeon_ring_write(ring, 0);
5362 radeon_ring_write(ring, 0);
5363
5364 /* bits 0-15 are the VM contexts0-15 */
5365 radeon_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5366 radeon_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
5367 WRITE_DATA_DST_SEL(0)));
5368 radeon_ring_write(ring, VM_INVALIDATE_REQUEST >> 2);
5369 radeon_ring_write(ring, 0);
5370 radeon_ring_write(ring, 1 << vm->id);
5371
5372 /* compute doesn't have PFP */
5373 if (ridx == RADEON_RING_TYPE_GFX_INDEX) {
5374 /* sync PFP to ME, otherwise we might get invalid PFP reads */
5375 radeon_ring_write(ring, PACKET3(PACKET3_PFP_SYNC_ME, 0));
5376 radeon_ring_write(ring, 0x0);
5377 }
5378 }
5379
5380 /**
5381 * cik_vm_set_page - update the page tables using sDMA
5382 *
5383 * @rdev: radeon_device pointer
5384 * @ib: indirect buffer to fill with commands
5385 * @pe: addr of the page entry
5386 * @addr: dst addr to write into pe
5387 * @count: number of page entries to update
5388 * @incr: increase next addr by incr bytes
5389 * @flags: access flags
5390 *
5391 * Update the page tables using CP or sDMA (CIK).
5392 */
5393 void cik_vm_set_page(struct radeon_device *rdev,
5394 struct radeon_ib *ib,
5395 uint64_t pe,
5396 uint64_t addr, unsigned count,
5397 uint32_t incr, uint32_t flags)
5398 {
5399 uint32_t r600_flags = cayman_vm_page_flags(rdev, flags);
5400 uint64_t value;
5401 unsigned ndw;
5402
5403 if (rdev->asic->vm.pt_ring_index == RADEON_RING_TYPE_GFX_INDEX) {
5404 /* CP */
5405 while (count) {
5406 ndw = 2 + count * 2;
5407 if (ndw > 0x3FFE)
5408 ndw = 0x3FFE;
5409
5410 ib->ptr[ib->length_dw++] = PACKET3(PACKET3_WRITE_DATA, ndw);
5411 ib->ptr[ib->length_dw++] = (WRITE_DATA_ENGINE_SEL(0) |
5412 WRITE_DATA_DST_SEL(1));
5413 ib->ptr[ib->length_dw++] = pe;
5414 ib->ptr[ib->length_dw++] = upper_32_bits(pe);
5415 for (; ndw > 2; ndw -= 2, --count, pe += 8) {
5416 if (flags & RADEON_VM_PAGE_SYSTEM) {
5417 value = radeon_vm_map_gart(rdev, addr);
5418 value &= 0xFFFFFFFFFFFFF000ULL;
5419 } else if (flags & RADEON_VM_PAGE_VALID) {
5420 value = addr;
5421 } else {
5422 value = 0;
5423 }
5424 addr += incr;
5425 value |= r600_flags;
5426 ib->ptr[ib->length_dw++] = value;
5427 ib->ptr[ib->length_dw++] = upper_32_bits(value);
5428 }
5429 }
5430 } else {
5431 /* DMA */
5432 if (flags & RADEON_VM_PAGE_SYSTEM) {
5433 while (count) {
5434 ndw = count * 2;
5435 if (ndw > 0xFFFFE)
5436 ndw = 0xFFFFE;
5437
5438 /* for non-physically contiguous pages (system) */
5439 ib->ptr[ib->length_dw++] = SDMA_PACKET(SDMA_OPCODE_WRITE, SDMA_WRITE_SUB_OPCODE_LINEAR, 0);
5440 ib->ptr[ib->length_dw++] = pe;
5441 ib->ptr[ib->length_dw++] = upper_32_bits(pe);
5442 ib->ptr[ib->length_dw++] = ndw;
5443 for (; ndw > 0; ndw -= 2, --count, pe += 8) {
5444 if (flags & RADEON_VM_PAGE_SYSTEM) {
5445 value = radeon_vm_map_gart(rdev, addr);
5446 value &= 0xFFFFFFFFFFFFF000ULL;
5447 } else if (flags & RADEON_VM_PAGE_VALID) {
5448 value = addr;
5449 } else {
5450 value = 0;
5451 }
5452 addr += incr;
5453 value |= r600_flags;
5454 ib->ptr[ib->length_dw++] = value;
5455 ib->ptr[ib->length_dw++] = upper_32_bits(value);
5456 }
5457 }
5458 } else {
5459 while (count) {
5460 ndw = count;
5461 if (ndw > 0x7FFFF)
5462 ndw = 0x7FFFF;
5463
5464 if (flags & RADEON_VM_PAGE_VALID)
5465 value = addr;
5466 else
5467 value = 0;
5468 /* for physically contiguous pages (vram) */
5469 ib->ptr[ib->length_dw++] = SDMA_PACKET(SDMA_OPCODE_GENERATE_PTE_PDE, 0, 0);
5470 ib->ptr[ib->length_dw++] = pe; /* dst addr */
5471 ib->ptr[ib->length_dw++] = upper_32_bits(pe);
5472 ib->ptr[ib->length_dw++] = r600_flags; /* mask */
5473 ib->ptr[ib->length_dw++] = 0;
5474 ib->ptr[ib->length_dw++] = value; /* value */
5475 ib->ptr[ib->length_dw++] = upper_32_bits(value);
5476 ib->ptr[ib->length_dw++] = incr; /* increment size */
5477 ib->ptr[ib->length_dw++] = 0;
5478 ib->ptr[ib->length_dw++] = ndw; /* number of entries */
5479 pe += ndw * 8;
5480 addr += ndw * incr;
5481 count -= ndw;
5482 }
5483 }
5484 while (ib->length_dw & 0x7)
5485 ib->ptr[ib->length_dw++] = SDMA_PACKET(SDMA_OPCODE_NOP, 0, 0);
5486 }
5487 }
5488
5489 /**
5490 * cik_dma_vm_flush - cik vm flush using sDMA
5491 *
5492 * @rdev: radeon_device pointer
5493 *
5494 * Update the page table base and flush the VM TLB
5495 * using sDMA (CIK).
5496 */
5497 void cik_dma_vm_flush(struct radeon_device *rdev, int ridx, struct radeon_vm *vm)
5498 {
5499 struct radeon_ring *ring = &rdev->ring[ridx];
5500 u32 extra_bits = (SDMA_POLL_REG_MEM_EXTRA_OP(1) |
5501 SDMA_POLL_REG_MEM_EXTRA_FUNC(3)); /* == */
5502 u32 ref_and_mask;
5503
5504 if (vm == NULL)
5505 return;
5506
5507 if (ridx == R600_RING_TYPE_DMA_INDEX)
5508 ref_and_mask = SDMA0;
5509 else
5510 ref_and_mask = SDMA1;
5511
5512 radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_SRBM_WRITE, 0, 0xf000));
5513 if (vm->id < 8) {
5514 radeon_ring_write(ring, (VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + (vm->id << 2)) >> 2);
5515 } else {
5516 radeon_ring_write(ring, (VM_CONTEXT8_PAGE_TABLE_BASE_ADDR + ((vm->id - 8) << 2)) >> 2);
5517 }
5518 radeon_ring_write(ring, vm->pd_gpu_addr >> 12);
5519
5520 /* update SH_MEM_* regs */
5521 radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_SRBM_WRITE, 0, 0xf000));
5522 radeon_ring_write(ring, SRBM_GFX_CNTL >> 2);
5523 radeon_ring_write(ring, VMID(vm->id));
5524
5525 radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_SRBM_WRITE, 0, 0xf000));
5526 radeon_ring_write(ring, SH_MEM_BASES >> 2);
5527 radeon_ring_write(ring, 0);
5528
5529 radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_SRBM_WRITE, 0, 0xf000));
5530 radeon_ring_write(ring, SH_MEM_CONFIG >> 2);
5531 radeon_ring_write(ring, 0);
5532
5533 radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_SRBM_WRITE, 0, 0xf000));
5534 radeon_ring_write(ring, SH_MEM_APE1_BASE >> 2);
5535 radeon_ring_write(ring, 1);
5536
5537 radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_SRBM_WRITE, 0, 0xf000));
5538 radeon_ring_write(ring, SH_MEM_APE1_LIMIT >> 2);
5539 radeon_ring_write(ring, 0);
5540
5541 radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_SRBM_WRITE, 0, 0xf000));
5542 radeon_ring_write(ring, SRBM_GFX_CNTL >> 2);
5543 radeon_ring_write(ring, VMID(0));
5544
5545 /* flush HDP */
5546 radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_POLL_REG_MEM, 0, extra_bits));
5547 radeon_ring_write(ring, GPU_HDP_FLUSH_DONE);
5548 radeon_ring_write(ring, GPU_HDP_FLUSH_REQ);
5549 radeon_ring_write(ring, ref_and_mask); /* REFERENCE */
5550 radeon_ring_write(ring, ref_and_mask); /* MASK */
5551 radeon_ring_write(ring, (4 << 16) | 10); /* RETRY_COUNT, POLL_INTERVAL */
5552
5553 /* flush TLB */
5554 radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_SRBM_WRITE, 0, 0xf000));
5555 radeon_ring_write(ring, VM_INVALIDATE_REQUEST >> 2);
5556 radeon_ring_write(ring, 1 << vm->id);
5557 }
5558
5559 /*
5560 * RLC
5561 * The RLC is a multi-purpose microengine that handles a
5562 * variety of functions, the most important of which is
5563 * the interrupt controller.
5564 */
5565 static void cik_enable_gui_idle_interrupt(struct radeon_device *rdev,
5566 bool enable)
5567 {
5568 u32 tmp = RREG32(CP_INT_CNTL_RING0);
5569
5570 if (enable)
5571 tmp |= (CNTX_BUSY_INT_ENABLE | CNTX_EMPTY_INT_ENABLE);
5572 else
5573 tmp &= ~(CNTX_BUSY_INT_ENABLE | CNTX_EMPTY_INT_ENABLE);
5574 WREG32(CP_INT_CNTL_RING0, tmp);
5575 }
5576
5577 static void cik_enable_lbpw(struct radeon_device *rdev, bool enable)
5578 {
5579 u32 tmp;
5580
5581 tmp = RREG32(RLC_LB_CNTL);
5582 if (enable)
5583 tmp |= LOAD_BALANCE_ENABLE;
5584 else
5585 tmp &= ~LOAD_BALANCE_ENABLE;
5586 WREG32(RLC_LB_CNTL, tmp);
5587 }
5588
5589 static void cik_wait_for_rlc_serdes(struct radeon_device *rdev)
5590 {
5591 u32 i, j, k;
5592 u32 mask;
5593
5594 for (i = 0; i < rdev->config.cik.max_shader_engines; i++) {
5595 for (j = 0; j < rdev->config.cik.max_sh_per_se; j++) {
5596 cik_select_se_sh(rdev, i, j);
5597 for (k = 0; k < rdev->usec_timeout; k++) {
5598 if (RREG32(RLC_SERDES_CU_MASTER_BUSY) == 0)
5599 break;
5600 udelay(1);
5601 }
5602 }
5603 }
5604 cik_select_se_sh(rdev, 0xffffffff, 0xffffffff);
5605
5606 mask = SE_MASTER_BUSY_MASK | GC_MASTER_BUSY | TC0_MASTER_BUSY | TC1_MASTER_BUSY;
5607 for (k = 0; k < rdev->usec_timeout; k++) {
5608 if ((RREG32(RLC_SERDES_NONCU_MASTER_BUSY) & mask) == 0)
5609 break;
5610 udelay(1);
5611 }
5612 }
5613
5614 static void cik_update_rlc(struct radeon_device *rdev, u32 rlc)
5615 {
5616 u32 tmp;
5617
5618 tmp = RREG32(RLC_CNTL);
5619 if (tmp != rlc)
5620 WREG32(RLC_CNTL, rlc);
5621 }
5622
5623 static u32 cik_halt_rlc(struct radeon_device *rdev)
5624 {
5625 u32 data, orig;
5626
5627 orig = data = RREG32(RLC_CNTL);
5628
5629 if (data & RLC_ENABLE) {
5630 u32 i;
5631
5632 data &= ~RLC_ENABLE;
5633 WREG32(RLC_CNTL, data);
5634
5635 for (i = 0; i < rdev->usec_timeout; i++) {
5636 if ((RREG32(RLC_GPM_STAT) & RLC_GPM_BUSY) == 0)
5637 break;
5638 udelay(1);
5639 }
5640
5641 cik_wait_for_rlc_serdes(rdev);
5642 }
5643
5644 return orig;
5645 }
5646
5647 void cik_enter_rlc_safe_mode(struct radeon_device *rdev)
5648 {
5649 u32 tmp, i, mask;
5650
5651 tmp = REQ | MESSAGE(MSG_ENTER_RLC_SAFE_MODE);
5652 WREG32(RLC_GPR_REG2, tmp);
5653
5654 mask = GFX_POWER_STATUS | GFX_CLOCK_STATUS;
5655 for (i = 0; i < rdev->usec_timeout; i++) {
5656 if ((RREG32(RLC_GPM_STAT) & mask) == mask)
5657 break;
5658 udelay(1);
5659 }
5660
5661 for (i = 0; i < rdev->usec_timeout; i++) {
5662 if ((RREG32(RLC_GPR_REG2) & REQ) == 0)
5663 break;
5664 udelay(1);
5665 }
5666 }
5667
5668 void cik_exit_rlc_safe_mode(struct radeon_device *rdev)
5669 {
5670 u32 tmp;
5671
5672 tmp = REQ | MESSAGE(MSG_EXIT_RLC_SAFE_MODE);
5673 WREG32(RLC_GPR_REG2, tmp);
5674 }
5675
5676 /**
5677 * cik_rlc_stop - stop the RLC ME
5678 *
5679 * @rdev: radeon_device pointer
5680 *
5681 * Halt the RLC ME (MicroEngine) (CIK).
5682 */
5683 static void cik_rlc_stop(struct radeon_device *rdev)
5684 {
5685 WREG32(RLC_CNTL, 0);
5686
5687 cik_enable_gui_idle_interrupt(rdev, false);
5688
5689 cik_wait_for_rlc_serdes(rdev);
5690 }
5691
5692 /**
5693 * cik_rlc_start - start the RLC ME
5694 *
5695 * @rdev: radeon_device pointer
5696 *
5697 * Unhalt the RLC ME (MicroEngine) (CIK).
5698 */
5699 static void cik_rlc_start(struct radeon_device *rdev)
5700 {
5701 WREG32(RLC_CNTL, RLC_ENABLE);
5702
5703 cik_enable_gui_idle_interrupt(rdev, true);
5704
5705 udelay(50);
5706 }
5707
5708 /**
5709 * cik_rlc_resume - setup the RLC hw
5710 *
5711 * @rdev: radeon_device pointer
5712 *
5713 * Initialize the RLC registers, load the ucode,
5714 * and start the RLC (CIK).
5715 * Returns 0 for success, -EINVAL if the ucode is not available.
5716 */
5717 static int cik_rlc_resume(struct radeon_device *rdev)
5718 {
5719 u32 i, size, tmp;
5720 const __be32 *fw_data;
5721
5722 if (!rdev->rlc_fw)
5723 return -EINVAL;
5724
5725 switch (rdev->family) {
5726 case CHIP_BONAIRE:
5727 default:
5728 size = BONAIRE_RLC_UCODE_SIZE;
5729 break;
5730 case CHIP_KAVERI:
5731 size = KV_RLC_UCODE_SIZE;
5732 break;
5733 case CHIP_KABINI:
5734 size = KB_RLC_UCODE_SIZE;
5735 break;
5736 }
5737
5738 cik_rlc_stop(rdev);
5739
5740 /* disable CG */
5741 tmp = RREG32(RLC_CGCG_CGLS_CTRL) & 0xfffffffc;
5742 WREG32(RLC_CGCG_CGLS_CTRL, tmp);
5743
5744 si_rlc_reset(rdev);
5745
5746 cik_init_pg(rdev);
5747
5748 cik_init_cg(rdev);
5749
5750 WREG32(RLC_LB_CNTR_INIT, 0);
5751 WREG32(RLC_LB_CNTR_MAX, 0x00008000);
5752
5753 cik_select_se_sh(rdev, 0xffffffff, 0xffffffff);
5754 WREG32(RLC_LB_INIT_CU_MASK, 0xffffffff);
5755 WREG32(RLC_LB_PARAMS, 0x00600408);
5756 WREG32(RLC_LB_CNTL, 0x80000004);
5757
5758 WREG32(RLC_MC_CNTL, 0);
5759 WREG32(RLC_UCODE_CNTL, 0);
5760
5761 fw_data = (const __be32 *)rdev->rlc_fw->data;
5762 WREG32(RLC_GPM_UCODE_ADDR, 0);
5763 for (i = 0; i < size; i++)
5764 WREG32(RLC_GPM_UCODE_DATA, be32_to_cpup(fw_data++));
5765 WREG32(RLC_GPM_UCODE_ADDR, 0);
5766
5767 /* XXX - find out what chips support lbpw */
5768 cik_enable_lbpw(rdev, false);
5769
5770 if (rdev->family == CHIP_BONAIRE)
5771 WREG32(RLC_DRIVER_DMA_STATUS, 0);
5772
5773 cik_rlc_start(rdev);
5774
5775 return 0;
5776 }
5777
5778 static void cik_enable_cgcg(struct radeon_device *rdev, bool enable)
5779 {
5780 u32 data, orig, tmp, tmp2;
5781
5782 orig = data = RREG32(RLC_CGCG_CGLS_CTRL);
5783
5784 cik_enable_gui_idle_interrupt(rdev, enable);
5785
5786 if (enable) {
5787 tmp = cik_halt_rlc(rdev);
5788
5789 cik_select_se_sh(rdev, 0xffffffff, 0xffffffff);
5790 WREG32(RLC_SERDES_WR_CU_MASTER_MASK, 0xffffffff);
5791 WREG32(RLC_SERDES_WR_NONCU_MASTER_MASK, 0xffffffff);
5792 tmp2 = BPM_ADDR_MASK | CGCG_OVERRIDE_0 | CGLS_ENABLE;
5793 WREG32(RLC_SERDES_WR_CTRL, tmp2);
5794
5795 cik_update_rlc(rdev, tmp);
5796
5797 data |= CGCG_EN | CGLS_EN;
5798 } else {
5799 RREG32(CB_CGTT_SCLK_CTRL);
5800 RREG32(CB_CGTT_SCLK_CTRL);
5801 RREG32(CB_CGTT_SCLK_CTRL);
5802 RREG32(CB_CGTT_SCLK_CTRL);
5803
5804 data &= ~(CGCG_EN | CGLS_EN);
5805 }
5806
5807 if (orig != data)
5808 WREG32(RLC_CGCG_CGLS_CTRL, data);
5809
5810 }
5811
5812 static void cik_enable_mgcg(struct radeon_device *rdev, bool enable)
5813 {
5814 u32 data, orig, tmp = 0;
5815
5816 if (enable) {
5817 orig = data = RREG32(CP_MEM_SLP_CNTL);
5818 data |= CP_MEM_LS_EN;
5819 if (orig != data)
5820 WREG32(CP_MEM_SLP_CNTL, data);
5821
5822 orig = data = RREG32(RLC_CGTT_MGCG_OVERRIDE);
5823 data &= 0xfffffffd;
5824 if (orig != data)
5825 WREG32(RLC_CGTT_MGCG_OVERRIDE, data);
5826
5827 tmp = cik_halt_rlc(rdev);
5828
5829 cik_select_se_sh(rdev, 0xffffffff, 0xffffffff);
5830 WREG32(RLC_SERDES_WR_CU_MASTER_MASK, 0xffffffff);
5831 WREG32(RLC_SERDES_WR_NONCU_MASTER_MASK, 0xffffffff);
5832 data = BPM_ADDR_MASK | MGCG_OVERRIDE_0;
5833 WREG32(RLC_SERDES_WR_CTRL, data);
5834
5835 cik_update_rlc(rdev, tmp);
5836
5837 orig = data = RREG32(CGTS_SM_CTRL_REG);
5838 data &= ~SM_MODE_MASK;
5839 data |= SM_MODE(0x2);
5840 data |= SM_MODE_ENABLE;
5841 data &= ~CGTS_OVERRIDE;
5842 data &= ~CGTS_LS_OVERRIDE;
5843 data &= ~ON_MONITOR_ADD_MASK;
5844 data |= ON_MONITOR_ADD_EN;
5845 data |= ON_MONITOR_ADD(0x96);
5846 if (orig != data)
5847 WREG32(CGTS_SM_CTRL_REG, data);
5848 } else {
5849 orig = data = RREG32(RLC_CGTT_MGCG_OVERRIDE);
5850 data |= 0x00000002;
5851 if (orig != data)
5852 WREG32(RLC_CGTT_MGCG_OVERRIDE, data);
5853
5854 data = RREG32(RLC_MEM_SLP_CNTL);
5855 if (data & RLC_MEM_LS_EN) {
5856 data &= ~RLC_MEM_LS_EN;
5857 WREG32(RLC_MEM_SLP_CNTL, data);
5858 }
5859
5860 data = RREG32(CP_MEM_SLP_CNTL);
5861 if (data & CP_MEM_LS_EN) {
5862 data &= ~CP_MEM_LS_EN;
5863 WREG32(CP_MEM_SLP_CNTL, data);
5864 }
5865
5866 orig = data = RREG32(CGTS_SM_CTRL_REG);
5867 data |= CGTS_OVERRIDE | CGTS_LS_OVERRIDE;
5868 if (orig != data)
5869 WREG32(CGTS_SM_CTRL_REG, data);
5870
5871 tmp = cik_halt_rlc(rdev);
5872
5873 cik_select_se_sh(rdev, 0xffffffff, 0xffffffff);
5874 WREG32(RLC_SERDES_WR_CU_MASTER_MASK, 0xffffffff);
5875 WREG32(RLC_SERDES_WR_NONCU_MASTER_MASK, 0xffffffff);
5876 data = BPM_ADDR_MASK | MGCG_OVERRIDE_1;
5877 WREG32(RLC_SERDES_WR_CTRL, data);
5878
5879 cik_update_rlc(rdev, tmp);
5880 }
5881 }
5882
5883 static const u32 mc_cg_registers[] =
5884 {
5885 MC_HUB_MISC_HUB_CG,
5886 MC_HUB_MISC_SIP_CG,
5887 MC_HUB_MISC_VM_CG,
5888 MC_XPB_CLK_GAT,
5889 ATC_MISC_CG,
5890 MC_CITF_MISC_WR_CG,
5891 MC_CITF_MISC_RD_CG,
5892 MC_CITF_MISC_VM_CG,
5893 VM_L2_CG,
5894 };
5895
5896 static void cik_enable_mc_ls(struct radeon_device *rdev,
5897 bool enable)
5898 {
5899 int i;
5900 u32 orig, data;
5901
5902 for (i = 0; i < ARRAY_SIZE(mc_cg_registers); i++) {
5903 orig = data = RREG32(mc_cg_registers[i]);
5904 if (enable)
5905 data |= MC_LS_ENABLE;
5906 else
5907 data &= ~MC_LS_ENABLE;
5908 if (data != orig)
5909 WREG32(mc_cg_registers[i], data);
5910 }
5911 }
5912
5913 static void cik_enable_mc_mgcg(struct radeon_device *rdev,
5914 bool enable)
5915 {
5916 int i;
5917 u32 orig, data;
5918
5919 for (i = 0; i < ARRAY_SIZE(mc_cg_registers); i++) {
5920 orig = data = RREG32(mc_cg_registers[i]);
5921 if (enable)
5922 data |= MC_CG_ENABLE;
5923 else
5924 data &= ~MC_CG_ENABLE;
5925 if (data != orig)
5926 WREG32(mc_cg_registers[i], data);
5927 }
5928 }
5929
5930 static void cik_enable_sdma_mgcg(struct radeon_device *rdev,
5931 bool enable)
5932 {
5933 u32 orig, data;
5934
5935 if (enable) {
5936 WREG32(SDMA0_CLK_CTRL + SDMA0_REGISTER_OFFSET, 0x00000100);
5937 WREG32(SDMA0_CLK_CTRL + SDMA1_REGISTER_OFFSET, 0x00000100);
5938 } else {
5939 orig = data = RREG32(SDMA0_CLK_CTRL + SDMA0_REGISTER_OFFSET);
5940 data |= 0xff000000;
5941 if (data != orig)
5942 WREG32(SDMA0_CLK_CTRL + SDMA0_REGISTER_OFFSET, data);
5943
5944 orig = data = RREG32(SDMA0_CLK_CTRL + SDMA1_REGISTER_OFFSET);
5945 data |= 0xff000000;
5946 if (data != orig)
5947 WREG32(SDMA0_CLK_CTRL + SDMA1_REGISTER_OFFSET, data);
5948 }
5949 }
5950
5951 static void cik_enable_sdma_mgls(struct radeon_device *rdev,
5952 bool enable)
5953 {
5954 u32 orig, data;
5955
5956 if (enable) {
5957 orig = data = RREG32(SDMA0_POWER_CNTL + SDMA0_REGISTER_OFFSET);
5958 data |= 0x100;
5959 if (orig != data)
5960 WREG32(SDMA0_POWER_CNTL + SDMA0_REGISTER_OFFSET, data);
5961
5962 orig = data = RREG32(SDMA0_POWER_CNTL + SDMA1_REGISTER_OFFSET);
5963 data |= 0x100;
5964 if (orig != data)
5965 WREG32(SDMA0_POWER_CNTL + SDMA1_REGISTER_OFFSET, data);
5966 } else {
5967 orig = data = RREG32(SDMA0_POWER_CNTL + SDMA0_REGISTER_OFFSET);
5968 data &= ~0x100;
5969 if (orig != data)
5970 WREG32(SDMA0_POWER_CNTL + SDMA0_REGISTER_OFFSET, data);
5971
5972 orig = data = RREG32(SDMA0_POWER_CNTL + SDMA1_REGISTER_OFFSET);
5973 data &= ~0x100;
5974 if (orig != data)
5975 WREG32(SDMA0_POWER_CNTL + SDMA1_REGISTER_OFFSET, data);
5976 }
5977 }
5978
5979 static void cik_enable_uvd_mgcg(struct radeon_device *rdev,
5980 bool enable)
5981 {
5982 u32 orig, data;
5983
5984 if (enable) {
5985 data = RREG32_UVD_CTX(UVD_CGC_MEM_CTRL);
5986 data = 0xfff;
5987 WREG32_UVD_CTX(UVD_CGC_MEM_CTRL, data);
5988
5989 orig = data = RREG32(UVD_CGC_CTRL);
5990 data |= DCM;
5991 if (orig != data)
5992 WREG32(UVD_CGC_CTRL, data);
5993 } else {
5994 data = RREG32_UVD_CTX(UVD_CGC_MEM_CTRL);
5995 data &= ~0xfff;
5996 WREG32_UVD_CTX(UVD_CGC_MEM_CTRL, data);
5997
5998 orig = data = RREG32(UVD_CGC_CTRL);
5999 data &= ~DCM;
6000 if (orig != data)
6001 WREG32(UVD_CGC_CTRL, data);
6002 }
6003 }
6004
6005 static void cik_enable_hdp_mgcg(struct radeon_device *rdev,
6006 bool enable)
6007 {
6008 u32 orig, data;
6009
6010 orig = data = RREG32(HDP_HOST_PATH_CNTL);
6011
6012 if (enable)
6013 data &= ~CLOCK_GATING_DIS;
6014 else
6015 data |= CLOCK_GATING_DIS;
6016
6017 if (orig != data)
6018 WREG32(HDP_HOST_PATH_CNTL, data);
6019 }
6020
6021 static void cik_enable_hdp_ls(struct radeon_device *rdev,
6022 bool enable)
6023 {
6024 u32 orig, data;
6025
6026 orig = data = RREG32(HDP_MEM_POWER_LS);
6027
6028 if (enable)
6029 data |= HDP_LS_ENABLE;
6030 else
6031 data &= ~HDP_LS_ENABLE;
6032
6033 if (orig != data)
6034 WREG32(HDP_MEM_POWER_LS, data);
6035 }
6036
6037 void cik_update_cg(struct radeon_device *rdev,
6038 u32 block, bool enable)
6039 {
6040 if (block & RADEON_CG_BLOCK_GFX) {
6041 /* order matters! */
6042 if (enable) {
6043 cik_enable_mgcg(rdev, true);
6044 cik_enable_cgcg(rdev, true);
6045 } else {
6046 cik_enable_cgcg(rdev, false);
6047 cik_enable_mgcg(rdev, false);
6048 }
6049 }
6050
6051 if (block & RADEON_CG_BLOCK_MC) {
6052 if (!(rdev->flags & RADEON_IS_IGP)) {
6053 cik_enable_mc_mgcg(rdev, enable);
6054 cik_enable_mc_ls(rdev, enable);
6055 }
6056 }
6057
6058 if (block & RADEON_CG_BLOCK_SDMA) {
6059 cik_enable_sdma_mgcg(rdev, enable);
6060 cik_enable_sdma_mgls(rdev, enable);
6061 }
6062
6063 if (block & RADEON_CG_BLOCK_UVD) {
6064 if (rdev->has_uvd)
6065 cik_enable_uvd_mgcg(rdev, enable);
6066 }
6067
6068 if (block & RADEON_CG_BLOCK_HDP) {
6069 cik_enable_hdp_mgcg(rdev, enable);
6070 cik_enable_hdp_ls(rdev, enable);
6071 }
6072 }
6073
6074 static void cik_init_cg(struct radeon_device *rdev)
6075 {
6076
6077 cik_update_cg(rdev, RADEON_CG_BLOCK_GFX, false); /* XXX true */
6078
6079 if (rdev->has_uvd)
6080 si_init_uvd_internal_cg(rdev);
6081
6082 cik_update_cg(rdev, (RADEON_CG_BLOCK_MC |
6083 RADEON_CG_BLOCK_SDMA |
6084 RADEON_CG_BLOCK_UVD |
6085 RADEON_CG_BLOCK_HDP), true);
6086 }
6087
6088 static void cik_enable_sck_slowdown_on_pu(struct radeon_device *rdev,
6089 bool enable)
6090 {
6091 u32 data, orig;
6092
6093 orig = data = RREG32(RLC_PG_CNTL);
6094 if (enable)
6095 data |= SMU_CLK_SLOWDOWN_ON_PU_ENABLE;
6096 else
6097 data &= ~SMU_CLK_SLOWDOWN_ON_PU_ENABLE;
6098 if (orig != data)
6099 WREG32(RLC_PG_CNTL, data);
6100 }
6101
6102 static void cik_enable_sck_slowdown_on_pd(struct radeon_device *rdev,
6103 bool enable)
6104 {
6105 u32 data, orig;
6106
6107 orig = data = RREG32(RLC_PG_CNTL);
6108 if (enable)
6109 data |= SMU_CLK_SLOWDOWN_ON_PD_ENABLE;
6110 else
6111 data &= ~SMU_CLK_SLOWDOWN_ON_PD_ENABLE;
6112 if (orig != data)
6113 WREG32(RLC_PG_CNTL, data);
6114 }
6115
6116 static void cik_enable_cp_pg(struct radeon_device *rdev, bool enable)
6117 {
6118 u32 data, orig;
6119
6120 orig = data = RREG32(RLC_PG_CNTL);
6121 if (enable)
6122 data &= ~DISABLE_CP_PG;
6123 else
6124 data |= DISABLE_CP_PG;
6125 if (orig != data)
6126 WREG32(RLC_PG_CNTL, data);
6127 }
6128
6129 static void cik_enable_gds_pg(struct radeon_device *rdev, bool enable)
6130 {
6131 u32 data, orig;
6132
6133 orig = data = RREG32(RLC_PG_CNTL);
6134 if (enable)
6135 data &= ~DISABLE_GDS_PG;
6136 else
6137 data |= DISABLE_GDS_PG;
6138 if (orig != data)
6139 WREG32(RLC_PG_CNTL, data);
6140 }
6141
6142 #define CP_ME_TABLE_SIZE 96
6143 #define CP_ME_TABLE_OFFSET 2048
6144 #define CP_MEC_TABLE_OFFSET 4096
6145
6146 void cik_init_cp_pg_table(struct radeon_device *rdev)
6147 {
6148 const __be32 *fw_data;
6149 volatile u32 *dst_ptr;
6150 int me, i, max_me = 4;
6151 u32 bo_offset = 0;
6152 u32 table_offset;
6153
6154 if (rdev->family == CHIP_KAVERI)
6155 max_me = 5;
6156
6157 if (rdev->rlc.cp_table_ptr == NULL)
6158 return;
6159
6160 /* write the cp table buffer */
6161 dst_ptr = rdev->rlc.cp_table_ptr;
6162 for (me = 0; me < max_me; me++) {
6163 if (me == 0) {
6164 fw_data = (const __be32 *)rdev->ce_fw->data;
6165 table_offset = CP_ME_TABLE_OFFSET;
6166 } else if (me == 1) {
6167 fw_data = (const __be32 *)rdev->pfp_fw->data;
6168 table_offset = CP_ME_TABLE_OFFSET;
6169 } else if (me == 2) {
6170 fw_data = (const __be32 *)rdev->me_fw->data;
6171 table_offset = CP_ME_TABLE_OFFSET;
6172 } else {
6173 fw_data = (const __be32 *)rdev->mec_fw->data;
6174 table_offset = CP_MEC_TABLE_OFFSET;
6175 }
6176
6177 for (i = 0; i < CP_ME_TABLE_SIZE; i ++) {
6178 dst_ptr[bo_offset + i] = be32_to_cpu(fw_data[table_offset + i]);
6179 }
6180 bo_offset += CP_ME_TABLE_SIZE;
6181 }
6182 }
6183
6184 static void cik_enable_gfx_cgpg(struct radeon_device *rdev,
6185 bool enable)
6186 {
6187 u32 data, orig;
6188
6189 if (enable) {
6190 orig = data = RREG32(RLC_PG_CNTL);
6191 data |= GFX_PG_ENABLE;
6192 if (orig != data)
6193 WREG32(RLC_PG_CNTL, data);
6194
6195 orig = data = RREG32(RLC_AUTO_PG_CTRL);
6196 data |= AUTO_PG_EN;
6197 if (orig != data)
6198 WREG32(RLC_AUTO_PG_CTRL, data);
6199 } else {
6200 orig = data = RREG32(RLC_PG_CNTL);
6201 data &= ~GFX_PG_ENABLE;
6202 if (orig != data)
6203 WREG32(RLC_PG_CNTL, data);
6204
6205 orig = data = RREG32(RLC_AUTO_PG_CTRL);
6206 data &= ~AUTO_PG_EN;
6207 if (orig != data)
6208 WREG32(RLC_AUTO_PG_CTRL, data);
6209
6210 data = RREG32(DB_RENDER_CONTROL);
6211 }
6212 }
6213
6214 static u32 cik_get_cu_active_bitmap(struct radeon_device *rdev, u32 se, u32 sh)
6215 {
6216 u32 mask = 0, tmp, tmp1;
6217 int i;
6218
6219 cik_select_se_sh(rdev, se, sh);
6220 tmp = RREG32(CC_GC_SHADER_ARRAY_CONFIG);
6221 tmp1 = RREG32(GC_USER_SHADER_ARRAY_CONFIG);
6222 cik_select_se_sh(rdev, 0xffffffff, 0xffffffff);
6223
6224 tmp &= 0xffff0000;
6225
6226 tmp |= tmp1;
6227 tmp >>= 16;
6228
6229 for (i = 0; i < rdev->config.cik.max_cu_per_sh; i ++) {
6230 mask <<= 1;
6231 mask |= 1;
6232 }
6233
6234 return (~tmp) & mask;
6235 }
6236
6237 static void cik_init_ao_cu_mask(struct radeon_device *rdev)
6238 {
6239 u32 i, j, k, active_cu_number = 0;
6240 u32 mask, counter, cu_bitmap;
6241 u32 tmp = 0;
6242
6243 for (i = 0; i < rdev->config.cik.max_shader_engines; i++) {
6244 for (j = 0; j < rdev->config.cik.max_sh_per_se; j++) {
6245 mask = 1;
6246 cu_bitmap = 0;
6247 counter = 0;
6248 for (k = 0; k < rdev->config.cik.max_cu_per_sh; k ++) {
6249 if (cik_get_cu_active_bitmap(rdev, i, j) & mask) {
6250 if (counter < 2)
6251 cu_bitmap |= mask;
6252 counter ++;
6253 }
6254 mask <<= 1;
6255 }
6256
6257 active_cu_number += counter;
6258 tmp |= (cu_bitmap << (i * 16 + j * 8));
6259 }
6260 }
6261
6262 WREG32(RLC_PG_AO_CU_MASK, tmp);
6263
6264 tmp = RREG32(RLC_MAX_PG_CU);
6265 tmp &= ~MAX_PU_CU_MASK;
6266 tmp |= MAX_PU_CU(active_cu_number);
6267 WREG32(RLC_MAX_PG_CU, tmp);
6268 }
6269
6270 static void cik_enable_gfx_static_mgpg(struct radeon_device *rdev,
6271 bool enable)
6272 {
6273 u32 data, orig;
6274
6275 orig = data = RREG32(RLC_PG_CNTL);
6276 if (enable)
6277 data |= STATIC_PER_CU_PG_ENABLE;
6278 else
6279 data &= ~STATIC_PER_CU_PG_ENABLE;
6280 if (orig != data)
6281 WREG32(RLC_PG_CNTL, data);
6282 }
6283
6284 static void cik_enable_gfx_dynamic_mgpg(struct radeon_device *rdev,
6285 bool enable)
6286 {
6287 u32 data, orig;
6288
6289 orig = data = RREG32(RLC_PG_CNTL);
6290 if (enable)
6291 data |= DYN_PER_CU_PG_ENABLE;
6292 else
6293 data &= ~DYN_PER_CU_PG_ENABLE;
6294 if (orig != data)
6295 WREG32(RLC_PG_CNTL, data);
6296 }
6297
6298 #define RLC_SAVE_AND_RESTORE_STARTING_OFFSET 0x90
6299 #define RLC_CLEAR_STATE_DESCRIPTOR_OFFSET 0x3D
6300
6301 static void cik_init_gfx_cgpg(struct radeon_device *rdev)
6302 {
6303 u32 data, orig;
6304 u32 i;
6305
6306 if (rdev->rlc.cs_data) {
6307 WREG32(RLC_GPM_SCRATCH_ADDR, RLC_CLEAR_STATE_DESCRIPTOR_OFFSET);
6308 WREG32(RLC_GPM_SCRATCH_DATA, upper_32_bits(rdev->rlc.clear_state_gpu_addr));
6309 WREG32(RLC_GPM_SCRATCH_DATA, rdev->rlc.clear_state_gpu_addr);
6310 WREG32(RLC_GPM_SCRATCH_DATA, rdev->rlc.clear_state_size);
6311 } else {
6312 WREG32(RLC_GPM_SCRATCH_ADDR, RLC_CLEAR_STATE_DESCRIPTOR_OFFSET);
6313 for (i = 0; i < 3; i++)
6314 WREG32(RLC_GPM_SCRATCH_DATA, 0);
6315 }
6316 if (rdev->rlc.reg_list) {
6317 WREG32(RLC_GPM_SCRATCH_ADDR, RLC_SAVE_AND_RESTORE_STARTING_OFFSET);
6318 for (i = 0; i < rdev->rlc.reg_list_size; i++)
6319 WREG32(RLC_GPM_SCRATCH_DATA, rdev->rlc.reg_list[i]);
6320 }
6321
6322 orig = data = RREG32(RLC_PG_CNTL);
6323 data |= GFX_PG_SRC;
6324 if (orig != data)
6325 WREG32(RLC_PG_CNTL, data);
6326
6327 WREG32(RLC_SAVE_AND_RESTORE_BASE, rdev->rlc.save_restore_gpu_addr >> 8);
6328 WREG32(RLC_CP_TABLE_RESTORE, rdev->rlc.cp_table_gpu_addr >> 8);
6329
6330 data = RREG32(CP_RB_WPTR_POLL_CNTL);
6331 data &= ~IDLE_POLL_COUNT_MASK;
6332 data |= IDLE_POLL_COUNT(0x60);
6333 WREG32(CP_RB_WPTR_POLL_CNTL, data);
6334
6335 data = 0x10101010;
6336 WREG32(RLC_PG_DELAY, data);
6337
6338 data = RREG32(RLC_PG_DELAY_2);
6339 data &= ~0xff;
6340 data |= 0x3;
6341 WREG32(RLC_PG_DELAY_2, data);
6342
6343 data = RREG32(RLC_AUTO_PG_CTRL);
6344 data &= ~GRBM_REG_SGIT_MASK;
6345 data |= GRBM_REG_SGIT(0x700);
6346 WREG32(RLC_AUTO_PG_CTRL, data);
6347
6348 }
6349
6350 static void cik_update_gfx_pg(struct radeon_device *rdev, bool enable)
6351 {
6352 bool has_pg = false;
6353 bool has_dyn_mgpg = false;
6354 bool has_static_mgpg = false;
6355
6356 /* only APUs have PG */
6357 if (rdev->flags & RADEON_IS_IGP) {
6358 has_pg = true;
6359 has_static_mgpg = true;
6360 if (rdev->family == CHIP_KAVERI)
6361 has_dyn_mgpg = true;
6362 }
6363
6364 if (has_pg) {
6365 cik_enable_gfx_cgpg(rdev, enable);
6366 if (enable) {
6367 cik_enable_gfx_static_mgpg(rdev, has_static_mgpg);
6368 cik_enable_gfx_dynamic_mgpg(rdev, has_dyn_mgpg);
6369 } else {
6370 cik_enable_gfx_static_mgpg(rdev, false);
6371 cik_enable_gfx_dynamic_mgpg(rdev, false);
6372 }
6373 }
6374
6375 }
6376
6377 void cik_init_pg(struct radeon_device *rdev)
6378 {
6379 bool has_pg = false;
6380
6381 /* only APUs have PG */
6382 if (rdev->flags & RADEON_IS_IGP) {
6383 /* XXX disable this for now */
6384 /* has_pg = true; */
6385 }
6386
6387 if (has_pg) {
6388 cik_enable_sck_slowdown_on_pu(rdev, true);
6389 cik_enable_sck_slowdown_on_pd(rdev, true);
6390 cik_init_gfx_cgpg(rdev);
6391 cik_enable_cp_pg(rdev, true);
6392 cik_enable_gds_pg(rdev, true);
6393 cik_init_ao_cu_mask(rdev);
6394 cik_update_gfx_pg(rdev, true);
6395 }
6396 }
6397
6398 /*
6399 * Interrupts
6400 * Starting with r6xx, interrupts are handled via a ring buffer.
6401 * Ring buffers are areas of GPU accessible memory that the GPU
6402 * writes interrupt vectors into and the host reads vectors out of.
6403 * There is a rptr (read pointer) that determines where the
6404 * host is currently reading, and a wptr (write pointer)
6405 * which determines where the GPU has written. When the
6406 * pointers are equal, the ring is idle. When the GPU
6407 * writes vectors to the ring buffer, it increments the
6408 * wptr. When there is an interrupt, the host then starts
6409 * fetching commands and processing them until the pointers are
6410 * equal again at which point it updates the rptr.
6411 */
6412
6413 /**
6414 * cik_enable_interrupts - Enable the interrupt ring buffer
6415 *
6416 * @rdev: radeon_device pointer
6417 *
6418 * Enable the interrupt ring buffer (CIK).
6419 */
6420 static void cik_enable_interrupts(struct radeon_device *rdev)
6421 {
6422 u32 ih_cntl = RREG32(IH_CNTL);
6423 u32 ih_rb_cntl = RREG32(IH_RB_CNTL);
6424
6425 ih_cntl |= ENABLE_INTR;
6426 ih_rb_cntl |= IH_RB_ENABLE;
6427 WREG32(IH_CNTL, ih_cntl);
6428 WREG32(IH_RB_CNTL, ih_rb_cntl);
6429 rdev->ih.enabled = true;
6430 }
6431
6432 /**
6433 * cik_disable_interrupts - Disable the interrupt ring buffer
6434 *
6435 * @rdev: radeon_device pointer
6436 *
6437 * Disable the interrupt ring buffer (CIK).
6438 */
6439 static void cik_disable_interrupts(struct radeon_device *rdev)
6440 {
6441 u32 ih_rb_cntl = RREG32(IH_RB_CNTL);
6442 u32 ih_cntl = RREG32(IH_CNTL);
6443
6444 ih_rb_cntl &= ~IH_RB_ENABLE;
6445 ih_cntl &= ~ENABLE_INTR;
6446 WREG32(IH_RB_CNTL, ih_rb_cntl);
6447 WREG32(IH_CNTL, ih_cntl);
6448 /* set rptr, wptr to 0 */
6449 WREG32(IH_RB_RPTR, 0);
6450 WREG32(IH_RB_WPTR, 0);
6451 rdev->ih.enabled = false;
6452 rdev->ih.rptr = 0;
6453 }
6454
6455 /**
6456 * cik_disable_interrupt_state - Disable all interrupt sources
6457 *
6458 * @rdev: radeon_device pointer
6459 *
6460 * Clear all interrupt enable bits used by the driver (CIK).
6461 */
6462 static void cik_disable_interrupt_state(struct radeon_device *rdev)
6463 {
6464 u32 tmp;
6465
6466 /* gfx ring */
6467 WREG32(CP_INT_CNTL_RING0, CNTX_BUSY_INT_ENABLE | CNTX_EMPTY_INT_ENABLE);
6468 /* sdma */
6469 tmp = RREG32(SDMA0_CNTL + SDMA0_REGISTER_OFFSET) & ~TRAP_ENABLE;
6470 WREG32(SDMA0_CNTL + SDMA0_REGISTER_OFFSET, tmp);
6471 tmp = RREG32(SDMA0_CNTL + SDMA1_REGISTER_OFFSET) & ~TRAP_ENABLE;
6472 WREG32(SDMA0_CNTL + SDMA1_REGISTER_OFFSET, tmp);
6473 /* compute queues */
6474 WREG32(CP_ME1_PIPE0_INT_CNTL, 0);
6475 WREG32(CP_ME1_PIPE1_INT_CNTL, 0);
6476 WREG32(CP_ME1_PIPE2_INT_CNTL, 0);
6477 WREG32(CP_ME1_PIPE3_INT_CNTL, 0);
6478 WREG32(CP_ME2_PIPE0_INT_CNTL, 0);
6479 WREG32(CP_ME2_PIPE1_INT_CNTL, 0);
6480 WREG32(CP_ME2_PIPE2_INT_CNTL, 0);
6481 WREG32(CP_ME2_PIPE3_INT_CNTL, 0);
6482 /* grbm */
6483 WREG32(GRBM_INT_CNTL, 0);
6484 /* vline/vblank, etc. */
6485 WREG32(LB_INTERRUPT_MASK + EVERGREEN_CRTC0_REGISTER_OFFSET, 0);
6486 WREG32(LB_INTERRUPT_MASK + EVERGREEN_CRTC1_REGISTER_OFFSET, 0);
6487 if (rdev->num_crtc >= 4) {
6488 WREG32(LB_INTERRUPT_MASK + EVERGREEN_CRTC2_REGISTER_OFFSET, 0);
6489 WREG32(LB_INTERRUPT_MASK + EVERGREEN_CRTC3_REGISTER_OFFSET, 0);
6490 }
6491 if (rdev->num_crtc >= 6) {
6492 WREG32(LB_INTERRUPT_MASK + EVERGREEN_CRTC4_REGISTER_OFFSET, 0);
6493 WREG32(LB_INTERRUPT_MASK + EVERGREEN_CRTC5_REGISTER_OFFSET, 0);
6494 }
6495
6496 /* dac hotplug */
6497 WREG32(DAC_AUTODETECT_INT_CONTROL, 0);
6498
6499 /* digital hotplug */
6500 tmp = RREG32(DC_HPD1_INT_CONTROL) & DC_HPDx_INT_POLARITY;
6501 WREG32(DC_HPD1_INT_CONTROL, tmp);
6502 tmp = RREG32(DC_HPD2_INT_CONTROL) & DC_HPDx_INT_POLARITY;
6503 WREG32(DC_HPD2_INT_CONTROL, tmp);
6504 tmp = RREG32(DC_HPD3_INT_CONTROL) & DC_HPDx_INT_POLARITY;
6505 WREG32(DC_HPD3_INT_CONTROL, tmp);
6506 tmp = RREG32(DC_HPD4_INT_CONTROL) & DC_HPDx_INT_POLARITY;
6507 WREG32(DC_HPD4_INT_CONTROL, tmp);
6508 tmp = RREG32(DC_HPD5_INT_CONTROL) & DC_HPDx_INT_POLARITY;
6509 WREG32(DC_HPD5_INT_CONTROL, tmp);
6510 tmp = RREG32(DC_HPD6_INT_CONTROL) & DC_HPDx_INT_POLARITY;
6511 WREG32(DC_HPD6_INT_CONTROL, tmp);
6512
6513 }
6514
6515 /**
6516 * cik_irq_init - init and enable the interrupt ring
6517 *
6518 * @rdev: radeon_device pointer
6519 *
6520 * Allocate a ring buffer for the interrupt controller,
6521 * enable the RLC, disable interrupts, enable the IH
6522 * ring buffer and enable it (CIK).
6523 * Called at device load and reume.
6524 * Returns 0 for success, errors for failure.
6525 */
6526 static int cik_irq_init(struct radeon_device *rdev)
6527 {
6528 int ret = 0;
6529 int rb_bufsz;
6530 u32 interrupt_cntl, ih_cntl, ih_rb_cntl;
6531
6532 /* allocate ring */
6533 ret = r600_ih_ring_alloc(rdev);
6534 if (ret)
6535 return ret;
6536
6537 /* disable irqs */
6538 cik_disable_interrupts(rdev);
6539
6540 /* init rlc */
6541 ret = cik_rlc_resume(rdev);
6542 if (ret) {
6543 r600_ih_ring_fini(rdev);
6544 return ret;
6545 }
6546
6547 /* setup interrupt control */
6548 /* XXX this should actually be a bus address, not an MC address. same on older asics */
6549 WREG32(INTERRUPT_CNTL2, rdev->ih.gpu_addr >> 8);
6550 interrupt_cntl = RREG32(INTERRUPT_CNTL);
6551 /* IH_DUMMY_RD_OVERRIDE=0 - dummy read disabled with msi, enabled without msi
6552 * IH_DUMMY_RD_OVERRIDE=1 - dummy read controlled by IH_DUMMY_RD_EN
6553 */
6554 interrupt_cntl &= ~IH_DUMMY_RD_OVERRIDE;
6555 /* IH_REQ_NONSNOOP_EN=1 if ring is in non-cacheable memory, e.g., vram */
6556 interrupt_cntl &= ~IH_REQ_NONSNOOP_EN;
6557 WREG32(INTERRUPT_CNTL, interrupt_cntl);
6558
6559 WREG32(IH_RB_BASE, rdev->ih.gpu_addr >> 8);
6560 rb_bufsz = drm_order(rdev->ih.ring_size / 4);
6561
6562 ih_rb_cntl = (IH_WPTR_OVERFLOW_ENABLE |
6563 IH_WPTR_OVERFLOW_CLEAR |
6564 (rb_bufsz << 1));
6565
6566 if (rdev->wb.enabled)
6567 ih_rb_cntl |= IH_WPTR_WRITEBACK_ENABLE;
6568
6569 /* set the writeback address whether it's enabled or not */
6570 WREG32(IH_RB_WPTR_ADDR_LO, (rdev->wb.gpu_addr + R600_WB_IH_WPTR_OFFSET) & 0xFFFFFFFC);
6571 WREG32(IH_RB_WPTR_ADDR_HI, upper_32_bits(rdev->wb.gpu_addr + R600_WB_IH_WPTR_OFFSET) & 0xFF);
6572
6573 WREG32(IH_RB_CNTL, ih_rb_cntl);
6574
6575 /* set rptr, wptr to 0 */
6576 WREG32(IH_RB_RPTR, 0);
6577 WREG32(IH_RB_WPTR, 0);
6578
6579 /* Default settings for IH_CNTL (disabled at first) */
6580 ih_cntl = MC_WRREQ_CREDIT(0x10) | MC_WR_CLEAN_CNT(0x10) | MC_VMID(0);
6581 /* RPTR_REARM only works if msi's are enabled */
6582 if (rdev->msi_enabled)
6583 ih_cntl |= RPTR_REARM;
6584 WREG32(IH_CNTL, ih_cntl);
6585
6586 /* force the active interrupt state to all disabled */
6587 cik_disable_interrupt_state(rdev);
6588
6589 pci_set_master(rdev->pdev);
6590
6591 /* enable irqs */
6592 cik_enable_interrupts(rdev);
6593
6594 return ret;
6595 }
6596
6597 /**
6598 * cik_irq_set - enable/disable interrupt sources
6599 *
6600 * @rdev: radeon_device pointer
6601 *
6602 * Enable interrupt sources on the GPU (vblanks, hpd,
6603 * etc.) (CIK).
6604 * Returns 0 for success, errors for failure.
6605 */
6606 int cik_irq_set(struct radeon_device *rdev)
6607 {
6608 u32 cp_int_cntl = CNTX_BUSY_INT_ENABLE | CNTX_EMPTY_INT_ENABLE |
6609 PRIV_INSTR_INT_ENABLE | PRIV_REG_INT_ENABLE;
6610 u32 cp_m1p0, cp_m1p1, cp_m1p2, cp_m1p3;
6611 u32 cp_m2p0, cp_m2p1, cp_m2p2, cp_m2p3;
6612 u32 crtc1 = 0, crtc2 = 0, crtc3 = 0, crtc4 = 0, crtc5 = 0, crtc6 = 0;
6613 u32 hpd1, hpd2, hpd3, hpd4, hpd5, hpd6;
6614 u32 grbm_int_cntl = 0;
6615 u32 dma_cntl, dma_cntl1;
6616 u32 thermal_int;
6617
6618 if (!rdev->irq.installed) {
6619 WARN(1, "Can't enable IRQ/MSI because no handler is installed\n");
6620 return -EINVAL;
6621 }
6622 /* don't enable anything if the ih is disabled */
6623 if (!rdev->ih.enabled) {
6624 cik_disable_interrupts(rdev);
6625 /* force the active interrupt state to all disabled */
6626 cik_disable_interrupt_state(rdev);
6627 return 0;
6628 }
6629
6630 hpd1 = RREG32(DC_HPD1_INT_CONTROL) & ~DC_HPDx_INT_EN;
6631 hpd2 = RREG32(DC_HPD2_INT_CONTROL) & ~DC_HPDx_INT_EN;
6632 hpd3 = RREG32(DC_HPD3_INT_CONTROL) & ~DC_HPDx_INT_EN;
6633 hpd4 = RREG32(DC_HPD4_INT_CONTROL) & ~DC_HPDx_INT_EN;
6634 hpd5 = RREG32(DC_HPD5_INT_CONTROL) & ~DC_HPDx_INT_EN;
6635 hpd6 = RREG32(DC_HPD6_INT_CONTROL) & ~DC_HPDx_INT_EN;
6636
6637 dma_cntl = RREG32(SDMA0_CNTL + SDMA0_REGISTER_OFFSET) & ~TRAP_ENABLE;
6638 dma_cntl1 = RREG32(SDMA0_CNTL + SDMA1_REGISTER_OFFSET) & ~TRAP_ENABLE;
6639
6640 cp_m1p0 = RREG32(CP_ME1_PIPE0_INT_CNTL) & ~TIME_STAMP_INT_ENABLE;
6641 cp_m1p1 = RREG32(CP_ME1_PIPE1_INT_CNTL) & ~TIME_STAMP_INT_ENABLE;
6642 cp_m1p2 = RREG32(CP_ME1_PIPE2_INT_CNTL) & ~TIME_STAMP_INT_ENABLE;
6643 cp_m1p3 = RREG32(CP_ME1_PIPE3_INT_CNTL) & ~TIME_STAMP_INT_ENABLE;
6644 cp_m2p0 = RREG32(CP_ME2_PIPE0_INT_CNTL) & ~TIME_STAMP_INT_ENABLE;
6645 cp_m2p1 = RREG32(CP_ME2_PIPE1_INT_CNTL) & ~TIME_STAMP_INT_ENABLE;
6646 cp_m2p2 = RREG32(CP_ME2_PIPE2_INT_CNTL) & ~TIME_STAMP_INT_ENABLE;
6647 cp_m2p3 = RREG32(CP_ME2_PIPE3_INT_CNTL) & ~TIME_STAMP_INT_ENABLE;
6648
6649 if (rdev->flags & RADEON_IS_IGP)
6650 thermal_int = RREG32_SMC(CG_THERMAL_INT_CTRL) &
6651 ~(THERM_INTH_MASK | THERM_INTL_MASK);
6652 else
6653 thermal_int = RREG32_SMC(CG_THERMAL_INT) &
6654 ~(THERM_INT_MASK_HIGH | THERM_INT_MASK_LOW);
6655
6656 /* enable CP interrupts on all rings */
6657 if (atomic_read(&rdev->irq.ring_int[RADEON_RING_TYPE_GFX_INDEX])) {
6658 DRM_DEBUG("cik_irq_set: sw int gfx\n");
6659 cp_int_cntl |= TIME_STAMP_INT_ENABLE;
6660 }
6661 if (atomic_read(&rdev->irq.ring_int[CAYMAN_RING_TYPE_CP1_INDEX])) {
6662 struct radeon_ring *ring = &rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX];
6663 DRM_DEBUG("si_irq_set: sw int cp1\n");
6664 if (ring->me == 1) {
6665 switch (ring->pipe) {
6666 case 0:
6667 cp_m1p0 |= TIME_STAMP_INT_ENABLE;
6668 break;
6669 case 1:
6670 cp_m1p1 |= TIME_STAMP_INT_ENABLE;
6671 break;
6672 case 2:
6673 cp_m1p2 |= TIME_STAMP_INT_ENABLE;
6674 break;
6675 case 3:
6676 cp_m1p2 |= TIME_STAMP_INT_ENABLE;
6677 break;
6678 default:
6679 DRM_DEBUG("si_irq_set: sw int cp1 invalid pipe %d\n", ring->pipe);
6680 break;
6681 }
6682 } else if (ring->me == 2) {
6683 switch (ring->pipe) {
6684 case 0:
6685 cp_m2p0 |= TIME_STAMP_INT_ENABLE;
6686 break;
6687 case 1:
6688 cp_m2p1 |= TIME_STAMP_INT_ENABLE;
6689 break;
6690 case 2:
6691 cp_m2p2 |= TIME_STAMP_INT_ENABLE;
6692 break;
6693 case 3:
6694 cp_m2p2 |= TIME_STAMP_INT_ENABLE;
6695 break;
6696 default:
6697 DRM_DEBUG("si_irq_set: sw int cp1 invalid pipe %d\n", ring->pipe);
6698 break;
6699 }
6700 } else {
6701 DRM_DEBUG("si_irq_set: sw int cp1 invalid me %d\n", ring->me);
6702 }
6703 }
6704 if (atomic_read(&rdev->irq.ring_int[CAYMAN_RING_TYPE_CP2_INDEX])) {
6705 struct radeon_ring *ring = &rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX];
6706 DRM_DEBUG("si_irq_set: sw int cp2\n");
6707 if (ring->me == 1) {
6708 switch (ring->pipe) {
6709 case 0:
6710 cp_m1p0 |= TIME_STAMP_INT_ENABLE;
6711 break;
6712 case 1:
6713 cp_m1p1 |= TIME_STAMP_INT_ENABLE;
6714 break;
6715 case 2:
6716 cp_m1p2 |= TIME_STAMP_INT_ENABLE;
6717 break;
6718 case 3:
6719 cp_m1p2 |= TIME_STAMP_INT_ENABLE;
6720 break;
6721 default:
6722 DRM_DEBUG("si_irq_set: sw int cp2 invalid pipe %d\n", ring->pipe);
6723 break;
6724 }
6725 } else if (ring->me == 2) {
6726 switch (ring->pipe) {
6727 case 0:
6728 cp_m2p0 |= TIME_STAMP_INT_ENABLE;
6729 break;
6730 case 1:
6731 cp_m2p1 |= TIME_STAMP_INT_ENABLE;
6732 break;
6733 case 2:
6734 cp_m2p2 |= TIME_STAMP_INT_ENABLE;
6735 break;
6736 case 3:
6737 cp_m2p2 |= TIME_STAMP_INT_ENABLE;
6738 break;
6739 default:
6740 DRM_DEBUG("si_irq_set: sw int cp2 invalid pipe %d\n", ring->pipe);
6741 break;
6742 }
6743 } else {
6744 DRM_DEBUG("si_irq_set: sw int cp2 invalid me %d\n", ring->me);
6745 }
6746 }
6747
6748 if (atomic_read(&rdev->irq.ring_int[R600_RING_TYPE_DMA_INDEX])) {
6749 DRM_DEBUG("cik_irq_set: sw int dma\n");
6750 dma_cntl |= TRAP_ENABLE;
6751 }
6752
6753 if (atomic_read(&rdev->irq.ring_int[CAYMAN_RING_TYPE_DMA1_INDEX])) {
6754 DRM_DEBUG("cik_irq_set: sw int dma1\n");
6755 dma_cntl1 |= TRAP_ENABLE;
6756 }
6757
6758 if (rdev->irq.crtc_vblank_int[0] ||
6759 atomic_read(&rdev->irq.pflip[0])) {
6760 DRM_DEBUG("cik_irq_set: vblank 0\n");
6761 crtc1 |= VBLANK_INTERRUPT_MASK;
6762 }
6763 if (rdev->irq.crtc_vblank_int[1] ||
6764 atomic_read(&rdev->irq.pflip[1])) {
6765 DRM_DEBUG("cik_irq_set: vblank 1\n");
6766 crtc2 |= VBLANK_INTERRUPT_MASK;
6767 }
6768 if (rdev->irq.crtc_vblank_int[2] ||
6769 atomic_read(&rdev->irq.pflip[2])) {
6770 DRM_DEBUG("cik_irq_set: vblank 2\n");
6771 crtc3 |= VBLANK_INTERRUPT_MASK;
6772 }
6773 if (rdev->irq.crtc_vblank_int[3] ||
6774 atomic_read(&rdev->irq.pflip[3])) {
6775 DRM_DEBUG("cik_irq_set: vblank 3\n");
6776 crtc4 |= VBLANK_INTERRUPT_MASK;
6777 }
6778 if (rdev->irq.crtc_vblank_int[4] ||
6779 atomic_read(&rdev->irq.pflip[4])) {
6780 DRM_DEBUG("cik_irq_set: vblank 4\n");
6781 crtc5 |= VBLANK_INTERRUPT_MASK;
6782 }
6783 if (rdev->irq.crtc_vblank_int[5] ||
6784 atomic_read(&rdev->irq.pflip[5])) {
6785 DRM_DEBUG("cik_irq_set: vblank 5\n");
6786 crtc6 |= VBLANK_INTERRUPT_MASK;
6787 }
6788 if (rdev->irq.hpd[0]) {
6789 DRM_DEBUG("cik_irq_set: hpd 1\n");
6790 hpd1 |= DC_HPDx_INT_EN;
6791 }
6792 if (rdev->irq.hpd[1]) {
6793 DRM_DEBUG("cik_irq_set: hpd 2\n");
6794 hpd2 |= DC_HPDx_INT_EN;
6795 }
6796 if (rdev->irq.hpd[2]) {
6797 DRM_DEBUG("cik_irq_set: hpd 3\n");
6798 hpd3 |= DC_HPDx_INT_EN;
6799 }
6800 if (rdev->irq.hpd[3]) {
6801 DRM_DEBUG("cik_irq_set: hpd 4\n");
6802 hpd4 |= DC_HPDx_INT_EN;
6803 }
6804 if (rdev->irq.hpd[4]) {
6805 DRM_DEBUG("cik_irq_set: hpd 5\n");
6806 hpd5 |= DC_HPDx_INT_EN;
6807 }
6808 if (rdev->irq.hpd[5]) {
6809 DRM_DEBUG("cik_irq_set: hpd 6\n");
6810 hpd6 |= DC_HPDx_INT_EN;
6811 }
6812
6813 if (rdev->irq.dpm_thermal) {
6814 DRM_DEBUG("dpm thermal\n");
6815 if (rdev->flags & RADEON_IS_IGP)
6816 thermal_int |= THERM_INTH_MASK | THERM_INTL_MASK;
6817 else
6818 thermal_int |= THERM_INT_MASK_HIGH | THERM_INT_MASK_LOW;
6819 }
6820
6821 WREG32(CP_INT_CNTL_RING0, cp_int_cntl);
6822
6823 WREG32(SDMA0_CNTL + SDMA0_REGISTER_OFFSET, dma_cntl);
6824 WREG32(SDMA0_CNTL + SDMA1_REGISTER_OFFSET, dma_cntl1);
6825
6826 WREG32(CP_ME1_PIPE0_INT_CNTL, cp_m1p0);
6827 WREG32(CP_ME1_PIPE1_INT_CNTL, cp_m1p1);
6828 WREG32(CP_ME1_PIPE2_INT_CNTL, cp_m1p2);
6829 WREG32(CP_ME1_PIPE3_INT_CNTL, cp_m1p3);
6830 WREG32(CP_ME2_PIPE0_INT_CNTL, cp_m2p0);
6831 WREG32(CP_ME2_PIPE1_INT_CNTL, cp_m2p1);
6832 WREG32(CP_ME2_PIPE2_INT_CNTL, cp_m2p2);
6833 WREG32(CP_ME2_PIPE3_INT_CNTL, cp_m2p3);
6834
6835 WREG32(GRBM_INT_CNTL, grbm_int_cntl);
6836
6837 WREG32(LB_INTERRUPT_MASK + EVERGREEN_CRTC0_REGISTER_OFFSET, crtc1);
6838 WREG32(LB_INTERRUPT_MASK + EVERGREEN_CRTC1_REGISTER_OFFSET, crtc2);
6839 if (rdev->num_crtc >= 4) {
6840 WREG32(LB_INTERRUPT_MASK + EVERGREEN_CRTC2_REGISTER_OFFSET, crtc3);
6841 WREG32(LB_INTERRUPT_MASK + EVERGREEN_CRTC3_REGISTER_OFFSET, crtc4);
6842 }
6843 if (rdev->num_crtc >= 6) {
6844 WREG32(LB_INTERRUPT_MASK + EVERGREEN_CRTC4_REGISTER_OFFSET, crtc5);
6845 WREG32(LB_INTERRUPT_MASK + EVERGREEN_CRTC5_REGISTER_OFFSET, crtc6);
6846 }
6847
6848 WREG32(DC_HPD1_INT_CONTROL, hpd1);
6849 WREG32(DC_HPD2_INT_CONTROL, hpd2);
6850 WREG32(DC_HPD3_INT_CONTROL, hpd3);
6851 WREG32(DC_HPD4_INT_CONTROL, hpd4);
6852 WREG32(DC_HPD5_INT_CONTROL, hpd5);
6853 WREG32(DC_HPD6_INT_CONTROL, hpd6);
6854
6855 if (rdev->flags & RADEON_IS_IGP)
6856 WREG32_SMC(CG_THERMAL_INT_CTRL, thermal_int);
6857 else
6858 WREG32_SMC(CG_THERMAL_INT, thermal_int);
6859
6860 return 0;
6861 }
6862
6863 /**
6864 * cik_irq_ack - ack interrupt sources
6865 *
6866 * @rdev: radeon_device pointer
6867 *
6868 * Ack interrupt sources on the GPU (vblanks, hpd,
6869 * etc.) (CIK). Certain interrupts sources are sw
6870 * generated and do not require an explicit ack.
6871 */
6872 static inline void cik_irq_ack(struct radeon_device *rdev)
6873 {
6874 u32 tmp;
6875
6876 rdev->irq.stat_regs.cik.disp_int = RREG32(DISP_INTERRUPT_STATUS);
6877 rdev->irq.stat_regs.cik.disp_int_cont = RREG32(DISP_INTERRUPT_STATUS_CONTINUE);
6878 rdev->irq.stat_regs.cik.disp_int_cont2 = RREG32(DISP_INTERRUPT_STATUS_CONTINUE2);
6879 rdev->irq.stat_regs.cik.disp_int_cont3 = RREG32(DISP_INTERRUPT_STATUS_CONTINUE3);
6880 rdev->irq.stat_regs.cik.disp_int_cont4 = RREG32(DISP_INTERRUPT_STATUS_CONTINUE4);
6881 rdev->irq.stat_regs.cik.disp_int_cont5 = RREG32(DISP_INTERRUPT_STATUS_CONTINUE5);
6882 rdev->irq.stat_regs.cik.disp_int_cont6 = RREG32(DISP_INTERRUPT_STATUS_CONTINUE6);
6883
6884 if (rdev->irq.stat_regs.cik.disp_int & LB_D1_VBLANK_INTERRUPT)
6885 WREG32(LB_VBLANK_STATUS + EVERGREEN_CRTC0_REGISTER_OFFSET, VBLANK_ACK);
6886 if (rdev->irq.stat_regs.cik.disp_int & LB_D1_VLINE_INTERRUPT)
6887 WREG32(LB_VLINE_STATUS + EVERGREEN_CRTC0_REGISTER_OFFSET, VLINE_ACK);
6888 if (rdev->irq.stat_regs.cik.disp_int_cont & LB_D2_VBLANK_INTERRUPT)
6889 WREG32(LB_VBLANK_STATUS + EVERGREEN_CRTC1_REGISTER_OFFSET, VBLANK_ACK);
6890 if (rdev->irq.stat_regs.cik.disp_int_cont & LB_D2_VLINE_INTERRUPT)
6891 WREG32(LB_VLINE_STATUS + EVERGREEN_CRTC1_REGISTER_OFFSET, VLINE_ACK);
6892
6893 if (rdev->num_crtc >= 4) {
6894 if (rdev->irq.stat_regs.cik.disp_int_cont2 & LB_D3_VBLANK_INTERRUPT)
6895 WREG32(LB_VBLANK_STATUS + EVERGREEN_CRTC2_REGISTER_OFFSET, VBLANK_ACK);
6896 if (rdev->irq.stat_regs.cik.disp_int_cont2 & LB_D3_VLINE_INTERRUPT)
6897 WREG32(LB_VLINE_STATUS + EVERGREEN_CRTC2_REGISTER_OFFSET, VLINE_ACK);
6898 if (rdev->irq.stat_regs.cik.disp_int_cont3 & LB_D4_VBLANK_INTERRUPT)
6899 WREG32(LB_VBLANK_STATUS + EVERGREEN_CRTC3_REGISTER_OFFSET, VBLANK_ACK);
6900 if (rdev->irq.stat_regs.cik.disp_int_cont3 & LB_D4_VLINE_INTERRUPT)
6901 WREG32(LB_VLINE_STATUS + EVERGREEN_CRTC3_REGISTER_OFFSET, VLINE_ACK);
6902 }
6903
6904 if (rdev->num_crtc >= 6) {
6905 if (rdev->irq.stat_regs.cik.disp_int_cont4 & LB_D5_VBLANK_INTERRUPT)
6906 WREG32(LB_VBLANK_STATUS + EVERGREEN_CRTC4_REGISTER_OFFSET, VBLANK_ACK);
6907 if (rdev->irq.stat_regs.cik.disp_int_cont4 & LB_D5_VLINE_INTERRUPT)
6908 WREG32(LB_VLINE_STATUS + EVERGREEN_CRTC4_REGISTER_OFFSET, VLINE_ACK);
6909 if (rdev->irq.stat_regs.cik.disp_int_cont5 & LB_D6_VBLANK_INTERRUPT)
6910 WREG32(LB_VBLANK_STATUS + EVERGREEN_CRTC5_REGISTER_OFFSET, VBLANK_ACK);
6911 if (rdev->irq.stat_regs.cik.disp_int_cont5 & LB_D6_VLINE_INTERRUPT)
6912 WREG32(LB_VLINE_STATUS + EVERGREEN_CRTC5_REGISTER_OFFSET, VLINE_ACK);
6913 }
6914
6915 if (rdev->irq.stat_regs.cik.disp_int & DC_HPD1_INTERRUPT) {
6916 tmp = RREG32(DC_HPD1_INT_CONTROL);
6917 tmp |= DC_HPDx_INT_ACK;
6918 WREG32(DC_HPD1_INT_CONTROL, tmp);
6919 }
6920 if (rdev->irq.stat_regs.cik.disp_int_cont & DC_HPD2_INTERRUPT) {
6921 tmp = RREG32(DC_HPD2_INT_CONTROL);
6922 tmp |= DC_HPDx_INT_ACK;
6923 WREG32(DC_HPD2_INT_CONTROL, tmp);
6924 }
6925 if (rdev->irq.stat_regs.cik.disp_int_cont2 & DC_HPD3_INTERRUPT) {
6926 tmp = RREG32(DC_HPD3_INT_CONTROL);
6927 tmp |= DC_HPDx_INT_ACK;
6928 WREG32(DC_HPD3_INT_CONTROL, tmp);
6929 }
6930 if (rdev->irq.stat_regs.cik.disp_int_cont3 & DC_HPD4_INTERRUPT) {
6931 tmp = RREG32(DC_HPD4_INT_CONTROL);
6932 tmp |= DC_HPDx_INT_ACK;
6933 WREG32(DC_HPD4_INT_CONTROL, tmp);
6934 }
6935 if (rdev->irq.stat_regs.cik.disp_int_cont4 & DC_HPD5_INTERRUPT) {
6936 tmp = RREG32(DC_HPD5_INT_CONTROL);
6937 tmp |= DC_HPDx_INT_ACK;
6938 WREG32(DC_HPD5_INT_CONTROL, tmp);
6939 }
6940 if (rdev->irq.stat_regs.cik.disp_int_cont5 & DC_HPD6_INTERRUPT) {
6941 tmp = RREG32(DC_HPD5_INT_CONTROL);
6942 tmp |= DC_HPDx_INT_ACK;
6943 WREG32(DC_HPD6_INT_CONTROL, tmp);
6944 }
6945 }
6946
6947 /**
6948 * cik_irq_disable - disable interrupts
6949 *
6950 * @rdev: radeon_device pointer
6951 *
6952 * Disable interrupts on the hw (CIK).
6953 */
6954 static void cik_irq_disable(struct radeon_device *rdev)
6955 {
6956 cik_disable_interrupts(rdev);
6957 /* Wait and acknowledge irq */
6958 mdelay(1);
6959 cik_irq_ack(rdev);
6960 cik_disable_interrupt_state(rdev);
6961 }
6962
6963 /**
6964 * cik_irq_disable - disable interrupts for suspend
6965 *
6966 * @rdev: radeon_device pointer
6967 *
6968 * Disable interrupts and stop the RLC (CIK).
6969 * Used for suspend.
6970 */
6971 static void cik_irq_suspend(struct radeon_device *rdev)
6972 {
6973 cik_irq_disable(rdev);
6974 cik_rlc_stop(rdev);
6975 }
6976
6977 /**
6978 * cik_irq_fini - tear down interrupt support
6979 *
6980 * @rdev: radeon_device pointer
6981 *
6982 * Disable interrupts on the hw and free the IH ring
6983 * buffer (CIK).
6984 * Used for driver unload.
6985 */
6986 static void cik_irq_fini(struct radeon_device *rdev)
6987 {
6988 cik_irq_suspend(rdev);
6989 r600_ih_ring_fini(rdev);
6990 }
6991
6992 /**
6993 * cik_get_ih_wptr - get the IH ring buffer wptr
6994 *
6995 * @rdev: radeon_device pointer
6996 *
6997 * Get the IH ring buffer wptr from either the register
6998 * or the writeback memory buffer (CIK). Also check for
6999 * ring buffer overflow and deal with it.
7000 * Used by cik_irq_process().
7001 * Returns the value of the wptr.
7002 */
7003 static inline u32 cik_get_ih_wptr(struct radeon_device *rdev)
7004 {
7005 u32 wptr, tmp;
7006
7007 if (rdev->wb.enabled)
7008 wptr = le32_to_cpu(rdev->wb.wb[R600_WB_IH_WPTR_OFFSET/4]);
7009 else
7010 wptr = RREG32(IH_RB_WPTR);
7011
7012 if (wptr & RB_OVERFLOW) {
7013 /* When a ring buffer overflow happen start parsing interrupt
7014 * from the last not overwritten vector (wptr + 16). Hopefully
7015 * this should allow us to catchup.
7016 */
7017 dev_warn(rdev->dev, "IH ring buffer overflow (0x%08X, %d, %d)\n",
7018 wptr, rdev->ih.rptr, (wptr + 16) + rdev->ih.ptr_mask);
7019 rdev->ih.rptr = (wptr + 16) & rdev->ih.ptr_mask;
7020 tmp = RREG32(IH_RB_CNTL);
7021 tmp |= IH_WPTR_OVERFLOW_CLEAR;
7022 WREG32(IH_RB_CNTL, tmp);
7023 }
7024 return (wptr & rdev->ih.ptr_mask);
7025 }
7026
7027 /* CIK IV Ring
7028 * Each IV ring entry is 128 bits:
7029 * [7:0] - interrupt source id
7030 * [31:8] - reserved
7031 * [59:32] - interrupt source data
7032 * [63:60] - reserved
7033 * [71:64] - RINGID
7034 * CP:
7035 * ME_ID [1:0], PIPE_ID[1:0], QUEUE_ID[2:0]
7036 * QUEUE_ID - for compute, which of the 8 queues owned by the dispatcher
7037 * - for gfx, hw shader state (0=PS...5=LS, 6=CS)
7038 * ME_ID - 0 = gfx, 1 = first 4 CS pipes, 2 = second 4 CS pipes
7039 * PIPE_ID - ME0 0=3D
7040 * - ME1&2 compute dispatcher (4 pipes each)
7041 * SDMA:
7042 * INSTANCE_ID [1:0], QUEUE_ID[1:0]
7043 * INSTANCE_ID - 0 = sdma0, 1 = sdma1
7044 * QUEUE_ID - 0 = gfx, 1 = rlc0, 2 = rlc1
7045 * [79:72] - VMID
7046 * [95:80] - PASID
7047 * [127:96] - reserved
7048 */
7049 /**
7050 * cik_irq_process - interrupt handler
7051 *
7052 * @rdev: radeon_device pointer
7053 *
7054 * Interrupt hander (CIK). Walk the IH ring,
7055 * ack interrupts and schedule work to handle
7056 * interrupt events.
7057 * Returns irq process return code.
7058 */
7059 int cik_irq_process(struct radeon_device *rdev)
7060 {
7061 struct radeon_ring *cp1_ring = &rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX];
7062 struct radeon_ring *cp2_ring = &rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX];
7063 u32 wptr;
7064 u32 rptr;
7065 u32 src_id, src_data, ring_id;
7066 u8 me_id, pipe_id, queue_id;
7067 u32 ring_index;
7068 bool queue_hotplug = false;
7069 bool queue_reset = false;
7070 u32 addr, status, mc_client;
7071 bool queue_thermal = false;
7072
7073 if (!rdev->ih.enabled || rdev->shutdown)
7074 return IRQ_NONE;
7075
7076 wptr = cik_get_ih_wptr(rdev);
7077
7078 restart_ih:
7079 /* is somebody else already processing irqs? */
7080 if (atomic_xchg(&rdev->ih.lock, 1))
7081 return IRQ_NONE;
7082
7083 rptr = rdev->ih.rptr;
7084 DRM_DEBUG("cik_irq_process start: rptr %d, wptr %d\n", rptr, wptr);
7085
7086 /* Order reading of wptr vs. reading of IH ring data */
7087 rmb();
7088
7089 /* display interrupts */
7090 cik_irq_ack(rdev);
7091
7092 while (rptr != wptr) {
7093 /* wptr/rptr are in bytes! */
7094 ring_index = rptr / 4;
7095 src_id = le32_to_cpu(rdev->ih.ring[ring_index]) & 0xff;
7096 src_data = le32_to_cpu(rdev->ih.ring[ring_index + 1]) & 0xfffffff;
7097 ring_id = le32_to_cpu(rdev->ih.ring[ring_index + 2]) & 0xff;
7098
7099 switch (src_id) {
7100 case 1: /* D1 vblank/vline */
7101 switch (src_data) {
7102 case 0: /* D1 vblank */
7103 if (rdev->irq.stat_regs.cik.disp_int & LB_D1_VBLANK_INTERRUPT) {
7104 if (rdev->irq.crtc_vblank_int[0]) {
7105 drm_handle_vblank(rdev->ddev, 0);
7106 rdev->pm.vblank_sync = true;
7107 wake_up(&rdev->irq.vblank_queue);
7108 }
7109 if (atomic_read(&rdev->irq.pflip[0]))
7110 radeon_crtc_handle_flip(rdev, 0);
7111 rdev->irq.stat_regs.cik.disp_int &= ~LB_D1_VBLANK_INTERRUPT;
7112 DRM_DEBUG("IH: D1 vblank\n");
7113 }
7114 break;
7115 case 1: /* D1 vline */
7116 if (rdev->irq.stat_regs.cik.disp_int & LB_D1_VLINE_INTERRUPT) {
7117 rdev->irq.stat_regs.cik.disp_int &= ~LB_D1_VLINE_INTERRUPT;
7118 DRM_DEBUG("IH: D1 vline\n");
7119 }
7120 break;
7121 default:
7122 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
7123 break;
7124 }
7125 break;
7126 case 2: /* D2 vblank/vline */
7127 switch (src_data) {
7128 case 0: /* D2 vblank */
7129 if (rdev->irq.stat_regs.cik.disp_int_cont & LB_D2_VBLANK_INTERRUPT) {
7130 if (rdev->irq.crtc_vblank_int[1]) {
7131 drm_handle_vblank(rdev->ddev, 1);
7132 rdev->pm.vblank_sync = true;
7133 wake_up(&rdev->irq.vblank_queue);
7134 }
7135 if (atomic_read(&rdev->irq.pflip[1]))
7136 radeon_crtc_handle_flip(rdev, 1);
7137 rdev->irq.stat_regs.cik.disp_int_cont &= ~LB_D2_VBLANK_INTERRUPT;
7138 DRM_DEBUG("IH: D2 vblank\n");
7139 }
7140 break;
7141 case 1: /* D2 vline */
7142 if (rdev->irq.stat_regs.cik.disp_int_cont & LB_D2_VLINE_INTERRUPT) {
7143 rdev->irq.stat_regs.cik.disp_int_cont &= ~LB_D2_VLINE_INTERRUPT;
7144 DRM_DEBUG("IH: D2 vline\n");
7145 }
7146 break;
7147 default:
7148 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
7149 break;
7150 }
7151 break;
7152 case 3: /* D3 vblank/vline */
7153 switch (src_data) {
7154 case 0: /* D3 vblank */
7155 if (rdev->irq.stat_regs.cik.disp_int_cont2 & LB_D3_VBLANK_INTERRUPT) {
7156 if (rdev->irq.crtc_vblank_int[2]) {
7157 drm_handle_vblank(rdev->ddev, 2);
7158 rdev->pm.vblank_sync = true;
7159 wake_up(&rdev->irq.vblank_queue);
7160 }
7161 if (atomic_read(&rdev->irq.pflip[2]))
7162 radeon_crtc_handle_flip(rdev, 2);
7163 rdev->irq.stat_regs.cik.disp_int_cont2 &= ~LB_D3_VBLANK_INTERRUPT;
7164 DRM_DEBUG("IH: D3 vblank\n");
7165 }
7166 break;
7167 case 1: /* D3 vline */
7168 if (rdev->irq.stat_regs.cik.disp_int_cont2 & LB_D3_VLINE_INTERRUPT) {
7169 rdev->irq.stat_regs.cik.disp_int_cont2 &= ~LB_D3_VLINE_INTERRUPT;
7170 DRM_DEBUG("IH: D3 vline\n");
7171 }
7172 break;
7173 default:
7174 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
7175 break;
7176 }
7177 break;
7178 case 4: /* D4 vblank/vline */
7179 switch (src_data) {
7180 case 0: /* D4 vblank */
7181 if (rdev->irq.stat_regs.cik.disp_int_cont3 & LB_D4_VBLANK_INTERRUPT) {
7182 if (rdev->irq.crtc_vblank_int[3]) {
7183 drm_handle_vblank(rdev->ddev, 3);
7184 rdev->pm.vblank_sync = true;
7185 wake_up(&rdev->irq.vblank_queue);
7186 }
7187 if (atomic_read(&rdev->irq.pflip[3]))
7188 radeon_crtc_handle_flip(rdev, 3);
7189 rdev->irq.stat_regs.cik.disp_int_cont3 &= ~LB_D4_VBLANK_INTERRUPT;
7190 DRM_DEBUG("IH: D4 vblank\n");
7191 }
7192 break;
7193 case 1: /* D4 vline */
7194 if (rdev->irq.stat_regs.cik.disp_int_cont3 & LB_D4_VLINE_INTERRUPT) {
7195 rdev->irq.stat_regs.cik.disp_int_cont3 &= ~LB_D4_VLINE_INTERRUPT;
7196 DRM_DEBUG("IH: D4 vline\n");
7197 }
7198 break;
7199 default:
7200 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
7201 break;
7202 }
7203 break;
7204 case 5: /* D5 vblank/vline */
7205 switch (src_data) {
7206 case 0: /* D5 vblank */
7207 if (rdev->irq.stat_regs.cik.disp_int_cont4 & LB_D5_VBLANK_INTERRUPT) {
7208 if (rdev->irq.crtc_vblank_int[4]) {
7209 drm_handle_vblank(rdev->ddev, 4);
7210 rdev->pm.vblank_sync = true;
7211 wake_up(&rdev->irq.vblank_queue);
7212 }
7213 if (atomic_read(&rdev->irq.pflip[4]))
7214 radeon_crtc_handle_flip(rdev, 4);
7215 rdev->irq.stat_regs.cik.disp_int_cont4 &= ~LB_D5_VBLANK_INTERRUPT;
7216 DRM_DEBUG("IH: D5 vblank\n");
7217 }
7218 break;
7219 case 1: /* D5 vline */
7220 if (rdev->irq.stat_regs.cik.disp_int_cont4 & LB_D5_VLINE_INTERRUPT) {
7221 rdev->irq.stat_regs.cik.disp_int_cont4 &= ~LB_D5_VLINE_INTERRUPT;
7222 DRM_DEBUG("IH: D5 vline\n");
7223 }
7224 break;
7225 default:
7226 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
7227 break;
7228 }
7229 break;
7230 case 6: /* D6 vblank/vline */
7231 switch (src_data) {
7232 case 0: /* D6 vblank */
7233 if (rdev->irq.stat_regs.cik.disp_int_cont5 & LB_D6_VBLANK_INTERRUPT) {
7234 if (rdev->irq.crtc_vblank_int[5]) {
7235 drm_handle_vblank(rdev->ddev, 5);
7236 rdev->pm.vblank_sync = true;
7237 wake_up(&rdev->irq.vblank_queue);
7238 }
7239 if (atomic_read(&rdev->irq.pflip[5]))
7240 radeon_crtc_handle_flip(rdev, 5);
7241 rdev->irq.stat_regs.cik.disp_int_cont5 &= ~LB_D6_VBLANK_INTERRUPT;
7242 DRM_DEBUG("IH: D6 vblank\n");
7243 }
7244 break;
7245 case 1: /* D6 vline */
7246 if (rdev->irq.stat_regs.cik.disp_int_cont5 & LB_D6_VLINE_INTERRUPT) {
7247 rdev->irq.stat_regs.cik.disp_int_cont5 &= ~LB_D6_VLINE_INTERRUPT;
7248 DRM_DEBUG("IH: D6 vline\n");
7249 }
7250 break;
7251 default:
7252 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
7253 break;
7254 }
7255 break;
7256 case 42: /* HPD hotplug */
7257 switch (src_data) {
7258 case 0:
7259 if (rdev->irq.stat_regs.cik.disp_int & DC_HPD1_INTERRUPT) {
7260 rdev->irq.stat_regs.cik.disp_int &= ~DC_HPD1_INTERRUPT;
7261 queue_hotplug = true;
7262 DRM_DEBUG("IH: HPD1\n");
7263 }
7264 break;
7265 case 1:
7266 if (rdev->irq.stat_regs.cik.disp_int_cont & DC_HPD2_INTERRUPT) {
7267 rdev->irq.stat_regs.cik.disp_int_cont &= ~DC_HPD2_INTERRUPT;
7268 queue_hotplug = true;
7269 DRM_DEBUG("IH: HPD2\n");
7270 }
7271 break;
7272 case 2:
7273 if (rdev->irq.stat_regs.cik.disp_int_cont2 & DC_HPD3_INTERRUPT) {
7274 rdev->irq.stat_regs.cik.disp_int_cont2 &= ~DC_HPD3_INTERRUPT;
7275 queue_hotplug = true;
7276 DRM_DEBUG("IH: HPD3\n");
7277 }
7278 break;
7279 case 3:
7280 if (rdev->irq.stat_regs.cik.disp_int_cont3 & DC_HPD4_INTERRUPT) {
7281 rdev->irq.stat_regs.cik.disp_int_cont3 &= ~DC_HPD4_INTERRUPT;
7282 queue_hotplug = true;
7283 DRM_DEBUG("IH: HPD4\n");
7284 }
7285 break;
7286 case 4:
7287 if (rdev->irq.stat_regs.cik.disp_int_cont4 & DC_HPD5_INTERRUPT) {
7288 rdev->irq.stat_regs.cik.disp_int_cont4 &= ~DC_HPD5_INTERRUPT;
7289 queue_hotplug = true;
7290 DRM_DEBUG("IH: HPD5\n");
7291 }
7292 break;
7293 case 5:
7294 if (rdev->irq.stat_regs.cik.disp_int_cont5 & DC_HPD6_INTERRUPT) {
7295 rdev->irq.stat_regs.cik.disp_int_cont5 &= ~DC_HPD6_INTERRUPT;
7296 queue_hotplug = true;
7297 DRM_DEBUG("IH: HPD6\n");
7298 }
7299 break;
7300 default:
7301 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
7302 break;
7303 }
7304 break;
7305 case 146:
7306 case 147:
7307 addr = RREG32(VM_CONTEXT1_PROTECTION_FAULT_ADDR);
7308 status = RREG32(VM_CONTEXT1_PROTECTION_FAULT_STATUS);
7309 mc_client = RREG32(VM_CONTEXT1_PROTECTION_FAULT_MCCLIENT);
7310 dev_err(rdev->dev, "GPU fault detected: %d 0x%08x\n", src_id, src_data);
7311 dev_err(rdev->dev, " VM_CONTEXT1_PROTECTION_FAULT_ADDR 0x%08X\n",
7312 addr);
7313 dev_err(rdev->dev, " VM_CONTEXT1_PROTECTION_FAULT_STATUS 0x%08X\n",
7314 status);
7315 cik_vm_decode_fault(rdev, status, addr, mc_client);
7316 /* reset addr and status */
7317 WREG32_P(VM_CONTEXT1_CNTL2, 1, ~1);
7318 break;
7319 case 176: /* GFX RB CP_INT */
7320 case 177: /* GFX IB CP_INT */
7321 radeon_fence_process(rdev, RADEON_RING_TYPE_GFX_INDEX);
7322 break;
7323 case 181: /* CP EOP event */
7324 DRM_DEBUG("IH: CP EOP\n");
7325 /* XXX check the bitfield order! */
7326 me_id = (ring_id & 0x60) >> 5;
7327 pipe_id = (ring_id & 0x18) >> 3;
7328 queue_id = (ring_id & 0x7) >> 0;
7329 switch (me_id) {
7330 case 0:
7331 radeon_fence_process(rdev, RADEON_RING_TYPE_GFX_INDEX);
7332 break;
7333 case 1:
7334 case 2:
7335 if ((cp1_ring->me == me_id) & (cp1_ring->pipe == pipe_id))
7336 radeon_fence_process(rdev, CAYMAN_RING_TYPE_CP1_INDEX);
7337 if ((cp2_ring->me == me_id) & (cp2_ring->pipe == pipe_id))
7338 radeon_fence_process(rdev, CAYMAN_RING_TYPE_CP2_INDEX);
7339 break;
7340 }
7341 break;
7342 case 184: /* CP Privileged reg access */
7343 DRM_ERROR("Illegal register access in command stream\n");
7344 /* XXX check the bitfield order! */
7345 me_id = (ring_id & 0x60) >> 5;
7346 pipe_id = (ring_id & 0x18) >> 3;
7347 queue_id = (ring_id & 0x7) >> 0;
7348 switch (me_id) {
7349 case 0:
7350 /* This results in a full GPU reset, but all we need to do is soft
7351 * reset the CP for gfx
7352 */
7353 queue_reset = true;
7354 break;
7355 case 1:
7356 /* XXX compute */
7357 queue_reset = true;
7358 break;
7359 case 2:
7360 /* XXX compute */
7361 queue_reset = true;
7362 break;
7363 }
7364 break;
7365 case 185: /* CP Privileged inst */
7366 DRM_ERROR("Illegal instruction in command stream\n");
7367 /* XXX check the bitfield order! */
7368 me_id = (ring_id & 0x60) >> 5;
7369 pipe_id = (ring_id & 0x18) >> 3;
7370 queue_id = (ring_id & 0x7) >> 0;
7371 switch (me_id) {
7372 case 0:
7373 /* This results in a full GPU reset, but all we need to do is soft
7374 * reset the CP for gfx
7375 */
7376 queue_reset = true;
7377 break;
7378 case 1:
7379 /* XXX compute */
7380 queue_reset = true;
7381 break;
7382 case 2:
7383 /* XXX compute */
7384 queue_reset = true;
7385 break;
7386 }
7387 break;
7388 case 224: /* SDMA trap event */
7389 /* XXX check the bitfield order! */
7390 me_id = (ring_id & 0x3) >> 0;
7391 queue_id = (ring_id & 0xc) >> 2;
7392 DRM_DEBUG("IH: SDMA trap\n");
7393 switch (me_id) {
7394 case 0:
7395 switch (queue_id) {
7396 case 0:
7397 radeon_fence_process(rdev, R600_RING_TYPE_DMA_INDEX);
7398 break;
7399 case 1:
7400 /* XXX compute */
7401 break;
7402 case 2:
7403 /* XXX compute */
7404 break;
7405 }
7406 break;
7407 case 1:
7408 switch (queue_id) {
7409 case 0:
7410 radeon_fence_process(rdev, CAYMAN_RING_TYPE_DMA1_INDEX);
7411 break;
7412 case 1:
7413 /* XXX compute */
7414 break;
7415 case 2:
7416 /* XXX compute */
7417 break;
7418 }
7419 break;
7420 }
7421 break;
7422 case 230: /* thermal low to high */
7423 DRM_DEBUG("IH: thermal low to high\n");
7424 rdev->pm.dpm.thermal.high_to_low = false;
7425 queue_thermal = true;
7426 break;
7427 case 231: /* thermal high to low */
7428 DRM_DEBUG("IH: thermal high to low\n");
7429 rdev->pm.dpm.thermal.high_to_low = true;
7430 queue_thermal = true;
7431 break;
7432 case 233: /* GUI IDLE */
7433 DRM_DEBUG("IH: GUI idle\n");
7434 break;
7435 case 241: /* SDMA Privileged inst */
7436 case 247: /* SDMA Privileged inst */
7437 DRM_ERROR("Illegal instruction in SDMA command stream\n");
7438 /* XXX check the bitfield order! */
7439 me_id = (ring_id & 0x3) >> 0;
7440 queue_id = (ring_id & 0xc) >> 2;
7441 switch (me_id) {
7442 case 0:
7443 switch (queue_id) {
7444 case 0:
7445 queue_reset = true;
7446 break;
7447 case 1:
7448 /* XXX compute */
7449 queue_reset = true;
7450 break;
7451 case 2:
7452 /* XXX compute */
7453 queue_reset = true;
7454 break;
7455 }
7456 break;
7457 case 1:
7458 switch (queue_id) {
7459 case 0:
7460 queue_reset = true;
7461 break;
7462 case 1:
7463 /* XXX compute */
7464 queue_reset = true;
7465 break;
7466 case 2:
7467 /* XXX compute */
7468 queue_reset = true;
7469 break;
7470 }
7471 break;
7472 }
7473 break;
7474 default:
7475 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
7476 break;
7477 }
7478
7479 /* wptr/rptr are in bytes! */
7480 rptr += 16;
7481 rptr &= rdev->ih.ptr_mask;
7482 }
7483 if (queue_hotplug)
7484 schedule_work(&rdev->hotplug_work);
7485 if (queue_reset)
7486 schedule_work(&rdev->reset_work);
7487 if (queue_thermal)
7488 schedule_work(&rdev->pm.dpm.thermal.work);
7489 rdev->ih.rptr = rptr;
7490 WREG32(IH_RB_RPTR, rdev->ih.rptr);
7491 atomic_set(&rdev->ih.lock, 0);
7492
7493 /* make sure wptr hasn't changed while processing */
7494 wptr = cik_get_ih_wptr(rdev);
7495 if (wptr != rptr)
7496 goto restart_ih;
7497
7498 return IRQ_HANDLED;
7499 }
7500
7501 /*
7502 * startup/shutdown callbacks
7503 */
7504 /**
7505 * cik_startup - program the asic to a functional state
7506 *
7507 * @rdev: radeon_device pointer
7508 *
7509 * Programs the asic to a functional state (CIK).
7510 * Called by cik_init() and cik_resume().
7511 * Returns 0 for success, error for failure.
7512 */
7513 static int cik_startup(struct radeon_device *rdev)
7514 {
7515 struct radeon_ring *ring;
7516 int r;
7517
7518 /* enable pcie gen2/3 link */
7519 cik_pcie_gen3_enable(rdev);
7520 /* enable aspm */
7521 cik_program_aspm(rdev);
7522
7523 cik_mc_program(rdev);
7524
7525 if (rdev->flags & RADEON_IS_IGP) {
7526 if (!rdev->me_fw || !rdev->pfp_fw || !rdev->ce_fw ||
7527 !rdev->mec_fw || !rdev->sdma_fw || !rdev->rlc_fw) {
7528 r = cik_init_microcode(rdev);
7529 if (r) {
7530 DRM_ERROR("Failed to load firmware!\n");
7531 return r;
7532 }
7533 }
7534 } else {
7535 if (!rdev->me_fw || !rdev->pfp_fw || !rdev->ce_fw ||
7536 !rdev->mec_fw || !rdev->sdma_fw || !rdev->rlc_fw ||
7537 !rdev->mc_fw) {
7538 r = cik_init_microcode(rdev);
7539 if (r) {
7540 DRM_ERROR("Failed to load firmware!\n");
7541 return r;
7542 }
7543 }
7544
7545 r = ci_mc_load_microcode(rdev);
7546 if (r) {
7547 DRM_ERROR("Failed to load MC firmware!\n");
7548 return r;
7549 }
7550 }
7551
7552 r = r600_vram_scratch_init(rdev);
7553 if (r)
7554 return r;
7555
7556 r = cik_pcie_gart_enable(rdev);
7557 if (r)
7558 return r;
7559 cik_gpu_init(rdev);
7560
7561 /* allocate rlc buffers */
7562 if (rdev->flags & RADEON_IS_IGP) {
7563 if (rdev->family == CHIP_KAVERI) {
7564 rdev->rlc.reg_list = spectre_rlc_save_restore_register_list;
7565 rdev->rlc.reg_list_size =
7566 (u32)ARRAY_SIZE(spectre_rlc_save_restore_register_list);
7567 } else {
7568 rdev->rlc.reg_list = kalindi_rlc_save_restore_register_list;
7569 rdev->rlc.reg_list_size =
7570 (u32)ARRAY_SIZE(kalindi_rlc_save_restore_register_list);
7571 }
7572 }
7573 rdev->rlc.cs_data = ci_cs_data;
7574 rdev->rlc.cp_table_size = CP_ME_TABLE_SIZE * 5 * 4;
7575 r = sumo_rlc_init(rdev);
7576 if (r) {
7577 DRM_ERROR("Failed to init rlc BOs!\n");
7578 return r;
7579 }
7580
7581 /* allocate wb buffer */
7582 r = radeon_wb_init(rdev);
7583 if (r)
7584 return r;
7585
7586 /* allocate mec buffers */
7587 r = cik_mec_init(rdev);
7588 if (r) {
7589 DRM_ERROR("Failed to init MEC BOs!\n");
7590 return r;
7591 }
7592
7593 r = radeon_fence_driver_start_ring(rdev, RADEON_RING_TYPE_GFX_INDEX);
7594 if (r) {
7595 dev_err(rdev->dev, "failed initializing CP fences (%d).\n", r);
7596 return r;
7597 }
7598
7599 r = radeon_fence_driver_start_ring(rdev, CAYMAN_RING_TYPE_CP1_INDEX);
7600 if (r) {
7601 dev_err(rdev->dev, "failed initializing CP fences (%d).\n", r);
7602 return r;
7603 }
7604
7605 r = radeon_fence_driver_start_ring(rdev, CAYMAN_RING_TYPE_CP2_INDEX);
7606 if (r) {
7607 dev_err(rdev->dev, "failed initializing CP fences (%d).\n", r);
7608 return r;
7609 }
7610
7611 r = radeon_fence_driver_start_ring(rdev, R600_RING_TYPE_DMA_INDEX);
7612 if (r) {
7613 dev_err(rdev->dev, "failed initializing DMA fences (%d).\n", r);
7614 return r;
7615 }
7616
7617 r = radeon_fence_driver_start_ring(rdev, CAYMAN_RING_TYPE_DMA1_INDEX);
7618 if (r) {
7619 dev_err(rdev->dev, "failed initializing DMA fences (%d).\n", r);
7620 return r;
7621 }
7622
7623 r = radeon_uvd_resume(rdev);
7624 if (!r) {
7625 cik_uvd_resume(rdev);
7626 r = radeon_fence_driver_start_ring(rdev,
7627 R600_RING_TYPE_UVD_INDEX);
7628 if (r)
7629 dev_err(rdev->dev, "UVD fences init error (%d).\n", r);
7630 }
7631 if (r)
7632 rdev->ring[R600_RING_TYPE_UVD_INDEX].ring_size = 0;
7633
7634 /* Enable IRQ */
7635 if (!rdev->irq.installed) {
7636 r = radeon_irq_kms_init(rdev);
7637 if (r)
7638 return r;
7639 }
7640
7641 r = cik_irq_init(rdev);
7642 if (r) {
7643 DRM_ERROR("radeon: IH init failed (%d).\n", r);
7644 radeon_irq_kms_fini(rdev);
7645 return r;
7646 }
7647 cik_irq_set(rdev);
7648
7649 ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
7650 r = radeon_ring_init(rdev, ring, ring->ring_size, RADEON_WB_CP_RPTR_OFFSET,
7651 CP_RB0_RPTR, CP_RB0_WPTR,
7652 0, 0xfffff, RADEON_CP_PACKET2);
7653 if (r)
7654 return r;
7655
7656 /* set up the compute queues */
7657 /* type-2 packets are deprecated on MEC, use type-3 instead */
7658 ring = &rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX];
7659 r = radeon_ring_init(rdev, ring, ring->ring_size, RADEON_WB_CP1_RPTR_OFFSET,
7660 CP_HQD_PQ_RPTR, CP_HQD_PQ_WPTR,
7661 0, 0xfffff, PACKET3(PACKET3_NOP, 0x3FFF));
7662 if (r)
7663 return r;
7664 ring->me = 1; /* first MEC */
7665 ring->pipe = 0; /* first pipe */
7666 ring->queue = 0; /* first queue */
7667 ring->wptr_offs = CIK_WB_CP1_WPTR_OFFSET;
7668
7669 /* type-2 packets are deprecated on MEC, use type-3 instead */
7670 ring = &rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX];
7671 r = radeon_ring_init(rdev, ring, ring->ring_size, RADEON_WB_CP2_RPTR_OFFSET,
7672 CP_HQD_PQ_RPTR, CP_HQD_PQ_WPTR,
7673 0, 0xffffffff, PACKET3(PACKET3_NOP, 0x3FFF));
7674 if (r)
7675 return r;
7676 /* dGPU only have 1 MEC */
7677 ring->me = 1; /* first MEC */
7678 ring->pipe = 0; /* first pipe */
7679 ring->queue = 1; /* second queue */
7680 ring->wptr_offs = CIK_WB_CP2_WPTR_OFFSET;
7681
7682 ring = &rdev->ring[R600_RING_TYPE_DMA_INDEX];
7683 r = radeon_ring_init(rdev, ring, ring->ring_size, R600_WB_DMA_RPTR_OFFSET,
7684 SDMA0_GFX_RB_RPTR + SDMA0_REGISTER_OFFSET,
7685 SDMA0_GFX_RB_WPTR + SDMA0_REGISTER_OFFSET,
7686 2, 0xfffffffc, SDMA_PACKET(SDMA_OPCODE_NOP, 0, 0));
7687 if (r)
7688 return r;
7689
7690 ring = &rdev->ring[CAYMAN_RING_TYPE_DMA1_INDEX];
7691 r = radeon_ring_init(rdev, ring, ring->ring_size, CAYMAN_WB_DMA1_RPTR_OFFSET,
7692 SDMA0_GFX_RB_RPTR + SDMA1_REGISTER_OFFSET,
7693 SDMA0_GFX_RB_WPTR + SDMA1_REGISTER_OFFSET,
7694 2, 0xfffffffc, SDMA_PACKET(SDMA_OPCODE_NOP, 0, 0));
7695 if (r)
7696 return r;
7697
7698 r = cik_cp_resume(rdev);
7699 if (r)
7700 return r;
7701
7702 r = cik_sdma_resume(rdev);
7703 if (r)
7704 return r;
7705
7706 ring = &rdev->ring[R600_RING_TYPE_UVD_INDEX];
7707 if (ring->ring_size) {
7708 r = radeon_ring_init(rdev, ring, ring->ring_size,
7709 R600_WB_UVD_RPTR_OFFSET,
7710 UVD_RBC_RB_RPTR, UVD_RBC_RB_WPTR,
7711 0, 0xfffff, RADEON_CP_PACKET2);
7712 if (!r)
7713 r = r600_uvd_init(rdev, true);
7714 if (r)
7715 DRM_ERROR("radeon: failed initializing UVD (%d).\n", r);
7716 }
7717
7718 r = radeon_ib_pool_init(rdev);
7719 if (r) {
7720 dev_err(rdev->dev, "IB initialization failed (%d).\n", r);
7721 return r;
7722 }
7723
7724 r = radeon_vm_manager_init(rdev);
7725 if (r) {
7726 dev_err(rdev->dev, "vm manager initialization failed (%d).\n", r);
7727 return r;
7728 }
7729
7730 return 0;
7731 }
7732
7733 /**
7734 * cik_resume - resume the asic to a functional state
7735 *
7736 * @rdev: radeon_device pointer
7737 *
7738 * Programs the asic to a functional state (CIK).
7739 * Called at resume.
7740 * Returns 0 for success, error for failure.
7741 */
7742 int cik_resume(struct radeon_device *rdev)
7743 {
7744 int r;
7745
7746 /* post card */
7747 atom_asic_init(rdev->mode_info.atom_context);
7748
7749 /* init golden registers */
7750 cik_init_golden_registers(rdev);
7751
7752 rdev->accel_working = true;
7753 r = cik_startup(rdev);
7754 if (r) {
7755 DRM_ERROR("cik startup failed on resume\n");
7756 rdev->accel_working = false;
7757 return r;
7758 }
7759
7760 return r;
7761
7762 }
7763
7764 /**
7765 * cik_suspend - suspend the asic
7766 *
7767 * @rdev: radeon_device pointer
7768 *
7769 * Bring the chip into a state suitable for suspend (CIK).
7770 * Called at suspend.
7771 * Returns 0 for success.
7772 */
7773 int cik_suspend(struct radeon_device *rdev)
7774 {
7775 radeon_vm_manager_fini(rdev);
7776 cik_cp_enable(rdev, false);
7777 cik_sdma_enable(rdev, false);
7778 r600_uvd_stop(rdev);
7779 radeon_uvd_suspend(rdev);
7780 cik_irq_suspend(rdev);
7781 radeon_wb_disable(rdev);
7782 cik_pcie_gart_disable(rdev);
7783 return 0;
7784 }
7785
7786 /* Plan is to move initialization in that function and use
7787 * helper function so that radeon_device_init pretty much
7788 * do nothing more than calling asic specific function. This
7789 * should also allow to remove a bunch of callback function
7790 * like vram_info.
7791 */
7792 /**
7793 * cik_init - asic specific driver and hw init
7794 *
7795 * @rdev: radeon_device pointer
7796 *
7797 * Setup asic specific driver variables and program the hw
7798 * to a functional state (CIK).
7799 * Called at driver startup.
7800 * Returns 0 for success, errors for failure.
7801 */
7802 int cik_init(struct radeon_device *rdev)
7803 {
7804 struct radeon_ring *ring;
7805 int r;
7806
7807 /* Read BIOS */
7808 if (!radeon_get_bios(rdev)) {
7809 if (ASIC_IS_AVIVO(rdev))
7810 return -EINVAL;
7811 }
7812 /* Must be an ATOMBIOS */
7813 if (!rdev->is_atom_bios) {
7814 dev_err(rdev->dev, "Expecting atombios for cayman GPU\n");
7815 return -EINVAL;
7816 }
7817 r = radeon_atombios_init(rdev);
7818 if (r)
7819 return r;
7820
7821 /* Post card if necessary */
7822 if (!radeon_card_posted(rdev)) {
7823 if (!rdev->bios) {
7824 dev_err(rdev->dev, "Card not posted and no BIOS - ignoring\n");
7825 return -EINVAL;
7826 }
7827 DRM_INFO("GPU not posted. posting now...\n");
7828 atom_asic_init(rdev->mode_info.atom_context);
7829 }
7830 /* init golden registers */
7831 cik_init_golden_registers(rdev);
7832 /* Initialize scratch registers */
7833 cik_scratch_init(rdev);
7834 /* Initialize surface registers */
7835 radeon_surface_init(rdev);
7836 /* Initialize clocks */
7837 radeon_get_clock_info(rdev->ddev);
7838
7839 /* Fence driver */
7840 r = radeon_fence_driver_init(rdev);
7841 if (r)
7842 return r;
7843
7844 /* initialize memory controller */
7845 r = cik_mc_init(rdev);
7846 if (r)
7847 return r;
7848 /* Memory manager */
7849 r = radeon_bo_init(rdev);
7850 if (r)
7851 return r;
7852
7853 ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
7854 ring->ring_obj = NULL;
7855 r600_ring_init(rdev, ring, 1024 * 1024);
7856
7857 ring = &rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX];
7858 ring->ring_obj = NULL;
7859 r600_ring_init(rdev, ring, 1024 * 1024);
7860 r = radeon_doorbell_get(rdev, &ring->doorbell_page_num);
7861 if (r)
7862 return r;
7863
7864 ring = &rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX];
7865 ring->ring_obj = NULL;
7866 r600_ring_init(rdev, ring, 1024 * 1024);
7867 r = radeon_doorbell_get(rdev, &ring->doorbell_page_num);
7868 if (r)
7869 return r;
7870
7871 ring = &rdev->ring[R600_RING_TYPE_DMA_INDEX];
7872 ring->ring_obj = NULL;
7873 r600_ring_init(rdev, ring, 256 * 1024);
7874
7875 ring = &rdev->ring[CAYMAN_RING_TYPE_DMA1_INDEX];
7876 ring->ring_obj = NULL;
7877 r600_ring_init(rdev, ring, 256 * 1024);
7878
7879 r = radeon_uvd_init(rdev);
7880 if (!r) {
7881 ring = &rdev->ring[R600_RING_TYPE_UVD_INDEX];
7882 ring->ring_obj = NULL;
7883 r600_ring_init(rdev, ring, 4096);
7884 }
7885
7886 rdev->ih.ring_obj = NULL;
7887 r600_ih_ring_init(rdev, 64 * 1024);
7888
7889 r = r600_pcie_gart_init(rdev);
7890 if (r)
7891 return r;
7892
7893 rdev->accel_working = true;
7894 r = cik_startup(rdev);
7895 if (r) {
7896 dev_err(rdev->dev, "disabling GPU acceleration\n");
7897 cik_cp_fini(rdev);
7898 cik_sdma_fini(rdev);
7899 cik_irq_fini(rdev);
7900 sumo_rlc_fini(rdev);
7901 cik_mec_fini(rdev);
7902 radeon_wb_fini(rdev);
7903 radeon_ib_pool_fini(rdev);
7904 radeon_vm_manager_fini(rdev);
7905 radeon_irq_kms_fini(rdev);
7906 cik_pcie_gart_fini(rdev);
7907 rdev->accel_working = false;
7908 }
7909
7910 /* Don't start up if the MC ucode is missing.
7911 * The default clocks and voltages before the MC ucode
7912 * is loaded are not suffient for advanced operations.
7913 */
7914 if (!rdev->mc_fw && !(rdev->flags & RADEON_IS_IGP)) {
7915 DRM_ERROR("radeon: MC ucode required for NI+.\n");
7916 return -EINVAL;
7917 }
7918
7919 return 0;
7920 }
7921
7922 /**
7923 * cik_fini - asic specific driver and hw fini
7924 *
7925 * @rdev: radeon_device pointer
7926 *
7927 * Tear down the asic specific driver variables and program the hw
7928 * to an idle state (CIK).
7929 * Called at driver unload.
7930 */
7931 void cik_fini(struct radeon_device *rdev)
7932 {
7933 cik_cp_fini(rdev);
7934 cik_sdma_fini(rdev);
7935 cik_irq_fini(rdev);
7936 sumo_rlc_fini(rdev);
7937 cik_mec_fini(rdev);
7938 radeon_wb_fini(rdev);
7939 radeon_vm_manager_fini(rdev);
7940 radeon_ib_pool_fini(rdev);
7941 radeon_irq_kms_fini(rdev);
7942 r600_uvd_stop(rdev);
7943 radeon_uvd_fini(rdev);
7944 cik_pcie_gart_fini(rdev);
7945 r600_vram_scratch_fini(rdev);
7946 radeon_gem_fini(rdev);
7947 radeon_fence_driver_fini(rdev);
7948 radeon_bo_fini(rdev);
7949 radeon_atombios_fini(rdev);
7950 kfree(rdev->bios);
7951 rdev->bios = NULL;
7952 }
7953
7954 /* display watermark setup */
7955 /**
7956 * dce8_line_buffer_adjust - Set up the line buffer
7957 *
7958 * @rdev: radeon_device pointer
7959 * @radeon_crtc: the selected display controller
7960 * @mode: the current display mode on the selected display
7961 * controller
7962 *
7963 * Setup up the line buffer allocation for
7964 * the selected display controller (CIK).
7965 * Returns the line buffer size in pixels.
7966 */
7967 static u32 dce8_line_buffer_adjust(struct radeon_device *rdev,
7968 struct radeon_crtc *radeon_crtc,
7969 struct drm_display_mode *mode)
7970 {
7971 u32 tmp;
7972
7973 /*
7974 * Line Buffer Setup
7975 * There are 6 line buffers, one for each display controllers.
7976 * There are 3 partitions per LB. Select the number of partitions
7977 * to enable based on the display width. For display widths larger
7978 * than 4096, you need use to use 2 display controllers and combine
7979 * them using the stereo blender.
7980 */
7981 if (radeon_crtc->base.enabled && mode) {
7982 if (mode->crtc_hdisplay < 1920)
7983 tmp = 1;
7984 else if (mode->crtc_hdisplay < 2560)
7985 tmp = 2;
7986 else if (mode->crtc_hdisplay < 4096)
7987 tmp = 0;
7988 else {
7989 DRM_DEBUG_KMS("Mode too big for LB!\n");
7990 tmp = 0;
7991 }
7992 } else
7993 tmp = 1;
7994
7995 WREG32(LB_MEMORY_CTRL + radeon_crtc->crtc_offset,
7996 LB_MEMORY_CONFIG(tmp) | LB_MEMORY_SIZE(0x6B0));
7997
7998 if (radeon_crtc->base.enabled && mode) {
7999 switch (tmp) {
8000 case 0:
8001 default:
8002 return 4096 * 2;
8003 case 1:
8004 return 1920 * 2;
8005 case 2:
8006 return 2560 * 2;
8007 }
8008 }
8009
8010 /* controller not enabled, so no lb used */
8011 return 0;
8012 }
8013
8014 /**
8015 * cik_get_number_of_dram_channels - get the number of dram channels
8016 *
8017 * @rdev: radeon_device pointer
8018 *
8019 * Look up the number of video ram channels (CIK).
8020 * Used for display watermark bandwidth calculations
8021 * Returns the number of dram channels
8022 */
8023 static u32 cik_get_number_of_dram_channels(struct radeon_device *rdev)
8024 {
8025 u32 tmp = RREG32(MC_SHARED_CHMAP);
8026
8027 switch ((tmp & NOOFCHAN_MASK) >> NOOFCHAN_SHIFT) {
8028 case 0:
8029 default:
8030 return 1;
8031 case 1:
8032 return 2;
8033 case 2:
8034 return 4;
8035 case 3:
8036 return 8;
8037 case 4:
8038 return 3;
8039 case 5:
8040 return 6;
8041 case 6:
8042 return 10;
8043 case 7:
8044 return 12;
8045 case 8:
8046 return 16;
8047 }
8048 }
8049
8050 struct dce8_wm_params {
8051 u32 dram_channels; /* number of dram channels */
8052 u32 yclk; /* bandwidth per dram data pin in kHz */
8053 u32 sclk; /* engine clock in kHz */
8054 u32 disp_clk; /* display clock in kHz */
8055 u32 src_width; /* viewport width */
8056 u32 active_time; /* active display time in ns */
8057 u32 blank_time; /* blank time in ns */
8058 bool interlaced; /* mode is interlaced */
8059 fixed20_12 vsc; /* vertical scale ratio */
8060 u32 num_heads; /* number of active crtcs */
8061 u32 bytes_per_pixel; /* bytes per pixel display + overlay */
8062 u32 lb_size; /* line buffer allocated to pipe */
8063 u32 vtaps; /* vertical scaler taps */
8064 };
8065
8066 /**
8067 * dce8_dram_bandwidth - get the dram bandwidth
8068 *
8069 * @wm: watermark calculation data
8070 *
8071 * Calculate the raw dram bandwidth (CIK).
8072 * Used for display watermark bandwidth calculations
8073 * Returns the dram bandwidth in MBytes/s
8074 */
8075 static u32 dce8_dram_bandwidth(struct dce8_wm_params *wm)
8076 {
8077 /* Calculate raw DRAM Bandwidth */
8078 fixed20_12 dram_efficiency; /* 0.7 */
8079 fixed20_12 yclk, dram_channels, bandwidth;
8080 fixed20_12 a;
8081
8082 a.full = dfixed_const(1000);
8083 yclk.full = dfixed_const(wm->yclk);
8084 yclk.full = dfixed_div(yclk, a);
8085 dram_channels.full = dfixed_const(wm->dram_channels * 4);
8086 a.full = dfixed_const(10);
8087 dram_efficiency.full = dfixed_const(7);
8088 dram_efficiency.full = dfixed_div(dram_efficiency, a);
8089 bandwidth.full = dfixed_mul(dram_channels, yclk);
8090 bandwidth.full = dfixed_mul(bandwidth, dram_efficiency);
8091
8092 return dfixed_trunc(bandwidth);
8093 }
8094
8095 /**
8096 * dce8_dram_bandwidth_for_display - get the dram bandwidth for display
8097 *
8098 * @wm: watermark calculation data
8099 *
8100 * Calculate the dram bandwidth used for display (CIK).
8101 * Used for display watermark bandwidth calculations
8102 * Returns the dram bandwidth for display in MBytes/s
8103 */
8104 static u32 dce8_dram_bandwidth_for_display(struct dce8_wm_params *wm)
8105 {
8106 /* Calculate DRAM Bandwidth and the part allocated to display. */
8107 fixed20_12 disp_dram_allocation; /* 0.3 to 0.7 */
8108 fixed20_12 yclk, dram_channels, bandwidth;
8109 fixed20_12 a;
8110
8111 a.full = dfixed_const(1000);
8112 yclk.full = dfixed_const(wm->yclk);
8113 yclk.full = dfixed_div(yclk, a);
8114 dram_channels.full = dfixed_const(wm->dram_channels * 4);
8115 a.full = dfixed_const(10);
8116 disp_dram_allocation.full = dfixed_const(3); /* XXX worse case value 0.3 */
8117 disp_dram_allocation.full = dfixed_div(disp_dram_allocation, a);
8118 bandwidth.full = dfixed_mul(dram_channels, yclk);
8119 bandwidth.full = dfixed_mul(bandwidth, disp_dram_allocation);
8120
8121 return dfixed_trunc(bandwidth);
8122 }
8123
8124 /**
8125 * dce8_data_return_bandwidth - get the data return bandwidth
8126 *
8127 * @wm: watermark calculation data
8128 *
8129 * Calculate the data return bandwidth used for display (CIK).
8130 * Used for display watermark bandwidth calculations
8131 * Returns the data return bandwidth in MBytes/s
8132 */
8133 static u32 dce8_data_return_bandwidth(struct dce8_wm_params *wm)
8134 {
8135 /* Calculate the display Data return Bandwidth */
8136 fixed20_12 return_efficiency; /* 0.8 */
8137 fixed20_12 sclk, bandwidth;
8138 fixed20_12 a;
8139
8140 a.full = dfixed_const(1000);
8141 sclk.full = dfixed_const(wm->sclk);
8142 sclk.full = dfixed_div(sclk, a);
8143 a.full = dfixed_const(10);
8144 return_efficiency.full = dfixed_const(8);
8145 return_efficiency.full = dfixed_div(return_efficiency, a);
8146 a.full = dfixed_const(32);
8147 bandwidth.full = dfixed_mul(a, sclk);
8148 bandwidth.full = dfixed_mul(bandwidth, return_efficiency);
8149
8150 return dfixed_trunc(bandwidth);
8151 }
8152
8153 /**
8154 * dce8_dmif_request_bandwidth - get the dmif bandwidth
8155 *
8156 * @wm: watermark calculation data
8157 *
8158 * Calculate the dmif bandwidth used for display (CIK).
8159 * Used for display watermark bandwidth calculations
8160 * Returns the dmif bandwidth in MBytes/s
8161 */
8162 static u32 dce8_dmif_request_bandwidth(struct dce8_wm_params *wm)
8163 {
8164 /* Calculate the DMIF Request Bandwidth */
8165 fixed20_12 disp_clk_request_efficiency; /* 0.8 */
8166 fixed20_12 disp_clk, bandwidth;
8167 fixed20_12 a, b;
8168
8169 a.full = dfixed_const(1000);
8170 disp_clk.full = dfixed_const(wm->disp_clk);
8171 disp_clk.full = dfixed_div(disp_clk, a);
8172 a.full = dfixed_const(32);
8173 b.full = dfixed_mul(a, disp_clk);
8174
8175 a.full = dfixed_const(10);
8176 disp_clk_request_efficiency.full = dfixed_const(8);
8177 disp_clk_request_efficiency.full = dfixed_div(disp_clk_request_efficiency, a);
8178
8179 bandwidth.full = dfixed_mul(b, disp_clk_request_efficiency);
8180
8181 return dfixed_trunc(bandwidth);
8182 }
8183
8184 /**
8185 * dce8_available_bandwidth - get the min available bandwidth
8186 *
8187 * @wm: watermark calculation data
8188 *
8189 * Calculate the min available bandwidth used for display (CIK).
8190 * Used for display watermark bandwidth calculations
8191 * Returns the min available bandwidth in MBytes/s
8192 */
8193 static u32 dce8_available_bandwidth(struct dce8_wm_params *wm)
8194 {
8195 /* Calculate the Available bandwidth. Display can use this temporarily but not in average. */
8196 u32 dram_bandwidth = dce8_dram_bandwidth(wm);
8197 u32 data_return_bandwidth = dce8_data_return_bandwidth(wm);
8198 u32 dmif_req_bandwidth = dce8_dmif_request_bandwidth(wm);
8199
8200 return min(dram_bandwidth, min(data_return_bandwidth, dmif_req_bandwidth));
8201 }
8202
8203 /**
8204 * dce8_average_bandwidth - get the average available bandwidth
8205 *
8206 * @wm: watermark calculation data
8207 *
8208 * Calculate the average available bandwidth used for display (CIK).
8209 * Used for display watermark bandwidth calculations
8210 * Returns the average available bandwidth in MBytes/s
8211 */
8212 static u32 dce8_average_bandwidth(struct dce8_wm_params *wm)
8213 {
8214 /* Calculate the display mode Average Bandwidth
8215 * DisplayMode should contain the source and destination dimensions,
8216 * timing, etc.
8217 */
8218 fixed20_12 bpp;
8219 fixed20_12 line_time;
8220 fixed20_12 src_width;
8221 fixed20_12 bandwidth;
8222 fixed20_12 a;
8223
8224 a.full = dfixed_const(1000);
8225 line_time.full = dfixed_const(wm->active_time + wm->blank_time);
8226 line_time.full = dfixed_div(line_time, a);
8227 bpp.full = dfixed_const(wm->bytes_per_pixel);
8228 src_width.full = dfixed_const(wm->src_width);
8229 bandwidth.full = dfixed_mul(src_width, bpp);
8230 bandwidth.full = dfixed_mul(bandwidth, wm->vsc);
8231 bandwidth.full = dfixed_div(bandwidth, line_time);
8232
8233 return dfixed_trunc(bandwidth);
8234 }
8235
8236 /**
8237 * dce8_latency_watermark - get the latency watermark
8238 *
8239 * @wm: watermark calculation data
8240 *
8241 * Calculate the latency watermark (CIK).
8242 * Used for display watermark bandwidth calculations
8243 * Returns the latency watermark in ns
8244 */
8245 static u32 dce8_latency_watermark(struct dce8_wm_params *wm)
8246 {
8247 /* First calculate the latency in ns */
8248 u32 mc_latency = 2000; /* 2000 ns. */
8249 u32 available_bandwidth = dce8_available_bandwidth(wm);
8250 u32 worst_chunk_return_time = (512 * 8 * 1000) / available_bandwidth;
8251 u32 cursor_line_pair_return_time = (128 * 4 * 1000) / available_bandwidth;
8252 u32 dc_latency = 40000000 / wm->disp_clk; /* dc pipe latency */
8253 u32 other_heads_data_return_time = ((wm->num_heads + 1) * worst_chunk_return_time) +
8254 (wm->num_heads * cursor_line_pair_return_time);
8255 u32 latency = mc_latency + other_heads_data_return_time + dc_latency;
8256 u32 max_src_lines_per_dst_line, lb_fill_bw, line_fill_time;
8257 u32 tmp, dmif_size = 12288;
8258 fixed20_12 a, b, c;
8259
8260 if (wm->num_heads == 0)
8261 return 0;
8262
8263 a.full = dfixed_const(2);
8264 b.full = dfixed_const(1);
8265 if ((wm->vsc.full > a.full) ||
8266 ((wm->vsc.full > b.full) && (wm->vtaps >= 3)) ||
8267 (wm->vtaps >= 5) ||
8268 ((wm->vsc.full >= a.full) && wm->interlaced))
8269 max_src_lines_per_dst_line = 4;
8270 else
8271 max_src_lines_per_dst_line = 2;
8272
8273 a.full = dfixed_const(available_bandwidth);
8274 b.full = dfixed_const(wm->num_heads);
8275 a.full = dfixed_div(a, b);
8276
8277 b.full = dfixed_const(mc_latency + 512);
8278 c.full = dfixed_const(wm->disp_clk);
8279 b.full = dfixed_div(b, c);
8280
8281 c.full = dfixed_const(dmif_size);
8282 b.full = dfixed_div(c, b);
8283
8284 tmp = min(dfixed_trunc(a), dfixed_trunc(b));
8285
8286 b.full = dfixed_const(1000);
8287 c.full = dfixed_const(wm->disp_clk);
8288 b.full = dfixed_div(c, b);
8289 c.full = dfixed_const(wm->bytes_per_pixel);
8290 b.full = dfixed_mul(b, c);
8291
8292 lb_fill_bw = min(tmp, dfixed_trunc(b));
8293
8294 a.full = dfixed_const(max_src_lines_per_dst_line * wm->src_width * wm->bytes_per_pixel);
8295 b.full = dfixed_const(1000);
8296 c.full = dfixed_const(lb_fill_bw);
8297 b.full = dfixed_div(c, b);
8298 a.full = dfixed_div(a, b);
8299 line_fill_time = dfixed_trunc(a);
8300
8301 if (line_fill_time < wm->active_time)
8302 return latency;
8303 else
8304 return latency + (line_fill_time - wm->active_time);
8305
8306 }
8307
8308 /**
8309 * dce8_average_bandwidth_vs_dram_bandwidth_for_display - check
8310 * average and available dram bandwidth
8311 *
8312 * @wm: watermark calculation data
8313 *
8314 * Check if the display average bandwidth fits in the display
8315 * dram bandwidth (CIK).
8316 * Used for display watermark bandwidth calculations
8317 * Returns true if the display fits, false if not.
8318 */
8319 static bool dce8_average_bandwidth_vs_dram_bandwidth_for_display(struct dce8_wm_params *wm)
8320 {
8321 if (dce8_average_bandwidth(wm) <=
8322 (dce8_dram_bandwidth_for_display(wm) / wm->num_heads))
8323 return true;
8324 else
8325 return false;
8326 }
8327
8328 /**
8329 * dce8_average_bandwidth_vs_available_bandwidth - check
8330 * average and available bandwidth
8331 *
8332 * @wm: watermark calculation data
8333 *
8334 * Check if the display average bandwidth fits in the display
8335 * available bandwidth (CIK).
8336 * Used for display watermark bandwidth calculations
8337 * Returns true if the display fits, false if not.
8338 */
8339 static bool dce8_average_bandwidth_vs_available_bandwidth(struct dce8_wm_params *wm)
8340 {
8341 if (dce8_average_bandwidth(wm) <=
8342 (dce8_available_bandwidth(wm) / wm->num_heads))
8343 return true;
8344 else
8345 return false;
8346 }
8347
8348 /**
8349 * dce8_check_latency_hiding - check latency hiding
8350 *
8351 * @wm: watermark calculation data
8352 *
8353 * Check latency hiding (CIK).
8354 * Used for display watermark bandwidth calculations
8355 * Returns true if the display fits, false if not.
8356 */
8357 static bool dce8_check_latency_hiding(struct dce8_wm_params *wm)
8358 {
8359 u32 lb_partitions = wm->lb_size / wm->src_width;
8360 u32 line_time = wm->active_time + wm->blank_time;
8361 u32 latency_tolerant_lines;
8362 u32 latency_hiding;
8363 fixed20_12 a;
8364
8365 a.full = dfixed_const(1);
8366 if (wm->vsc.full > a.full)
8367 latency_tolerant_lines = 1;
8368 else {
8369 if (lb_partitions <= (wm->vtaps + 1))
8370 latency_tolerant_lines = 1;
8371 else
8372 latency_tolerant_lines = 2;
8373 }
8374
8375 latency_hiding = (latency_tolerant_lines * line_time + wm->blank_time);
8376
8377 if (dce8_latency_watermark(wm) <= latency_hiding)
8378 return true;
8379 else
8380 return false;
8381 }
8382
8383 /**
8384 * dce8_program_watermarks - program display watermarks
8385 *
8386 * @rdev: radeon_device pointer
8387 * @radeon_crtc: the selected display controller
8388 * @lb_size: line buffer size
8389 * @num_heads: number of display controllers in use
8390 *
8391 * Calculate and program the display watermarks for the
8392 * selected display controller (CIK).
8393 */
8394 static void dce8_program_watermarks(struct radeon_device *rdev,
8395 struct radeon_crtc *radeon_crtc,
8396 u32 lb_size, u32 num_heads)
8397 {
8398 struct drm_display_mode *mode = &radeon_crtc->base.mode;
8399 struct dce8_wm_params wm_low, wm_high;
8400 u32 pixel_period;
8401 u32 line_time = 0;
8402 u32 latency_watermark_a = 0, latency_watermark_b = 0;
8403 u32 tmp, wm_mask;
8404
8405 if (radeon_crtc->base.enabled && num_heads && mode) {
8406 pixel_period = 1000000 / (u32)mode->clock;
8407 line_time = min((u32)mode->crtc_htotal * pixel_period, (u32)65535);
8408
8409 /* watermark for high clocks */
8410 if ((rdev->pm.pm_method == PM_METHOD_DPM) &&
8411 rdev->pm.dpm_enabled) {
8412 wm_high.yclk =
8413 radeon_dpm_get_mclk(rdev, false) * 10;
8414 wm_high.sclk =
8415 radeon_dpm_get_sclk(rdev, false) * 10;
8416 } else {
8417 wm_high.yclk = rdev->pm.current_mclk * 10;
8418 wm_high.sclk = rdev->pm.current_sclk * 10;
8419 }
8420
8421 wm_high.disp_clk = mode->clock;
8422 wm_high.src_width = mode->crtc_hdisplay;
8423 wm_high.active_time = mode->crtc_hdisplay * pixel_period;
8424 wm_high.blank_time = line_time - wm_high.active_time;
8425 wm_high.interlaced = false;
8426 if (mode->flags & DRM_MODE_FLAG_INTERLACE)
8427 wm_high.interlaced = true;
8428 wm_high.vsc = radeon_crtc->vsc;
8429 wm_high.vtaps = 1;
8430 if (radeon_crtc->rmx_type != RMX_OFF)
8431 wm_high.vtaps = 2;
8432 wm_high.bytes_per_pixel = 4; /* XXX: get this from fb config */
8433 wm_high.lb_size = lb_size;
8434 wm_high.dram_channels = cik_get_number_of_dram_channels(rdev);
8435 wm_high.num_heads = num_heads;
8436
8437 /* set for high clocks */
8438 latency_watermark_a = min(dce8_latency_watermark(&wm_high), (u32)65535);
8439
8440 /* possibly force display priority to high */
8441 /* should really do this at mode validation time... */
8442 if (!dce8_average_bandwidth_vs_dram_bandwidth_for_display(&wm_high) ||
8443 !dce8_average_bandwidth_vs_available_bandwidth(&wm_high) ||
8444 !dce8_check_latency_hiding(&wm_high) ||
8445 (rdev->disp_priority == 2)) {
8446 DRM_DEBUG_KMS("force priority to high\n");
8447 }
8448
8449 /* watermark for low clocks */
8450 if ((rdev->pm.pm_method == PM_METHOD_DPM) &&
8451 rdev->pm.dpm_enabled) {
8452 wm_low.yclk =
8453 radeon_dpm_get_mclk(rdev, true) * 10;
8454 wm_low.sclk =
8455 radeon_dpm_get_sclk(rdev, true) * 10;
8456 } else {
8457 wm_low.yclk = rdev->pm.current_mclk * 10;
8458 wm_low.sclk = rdev->pm.current_sclk * 10;
8459 }
8460
8461 wm_low.disp_clk = mode->clock;
8462 wm_low.src_width = mode->crtc_hdisplay;
8463 wm_low.active_time = mode->crtc_hdisplay * pixel_period;
8464 wm_low.blank_time = line_time - wm_low.active_time;
8465 wm_low.interlaced = false;
8466 if (mode->flags & DRM_MODE_FLAG_INTERLACE)
8467 wm_low.interlaced = true;
8468 wm_low.vsc = radeon_crtc->vsc;
8469 wm_low.vtaps = 1;
8470 if (radeon_crtc->rmx_type != RMX_OFF)
8471 wm_low.vtaps = 2;
8472 wm_low.bytes_per_pixel = 4; /* XXX: get this from fb config */
8473 wm_low.lb_size = lb_size;
8474 wm_low.dram_channels = cik_get_number_of_dram_channels(rdev);
8475 wm_low.num_heads = num_heads;
8476
8477 /* set for low clocks */
8478 latency_watermark_b = min(dce8_latency_watermark(&wm_low), (u32)65535);
8479
8480 /* possibly force display priority to high */
8481 /* should really do this at mode validation time... */
8482 if (!dce8_average_bandwidth_vs_dram_bandwidth_for_display(&wm_low) ||
8483 !dce8_average_bandwidth_vs_available_bandwidth(&wm_low) ||
8484 !dce8_check_latency_hiding(&wm_low) ||
8485 (rdev->disp_priority == 2)) {
8486 DRM_DEBUG_KMS("force priority to high\n");
8487 }
8488 }
8489
8490 /* select wm A */
8491 wm_mask = RREG32(DPG_WATERMARK_MASK_CONTROL + radeon_crtc->crtc_offset);
8492 tmp = wm_mask;
8493 tmp &= ~LATENCY_WATERMARK_MASK(3);
8494 tmp |= LATENCY_WATERMARK_MASK(1);
8495 WREG32(DPG_WATERMARK_MASK_CONTROL + radeon_crtc->crtc_offset, tmp);
8496 WREG32(DPG_PIPE_LATENCY_CONTROL + radeon_crtc->crtc_offset,
8497 (LATENCY_LOW_WATERMARK(latency_watermark_a) |
8498 LATENCY_HIGH_WATERMARK(line_time)));
8499 /* select wm B */
8500 tmp = RREG32(DPG_WATERMARK_MASK_CONTROL + radeon_crtc->crtc_offset);
8501 tmp &= ~LATENCY_WATERMARK_MASK(3);
8502 tmp |= LATENCY_WATERMARK_MASK(2);
8503 WREG32(DPG_WATERMARK_MASK_CONTROL + radeon_crtc->crtc_offset, tmp);
8504 WREG32(DPG_PIPE_LATENCY_CONTROL + radeon_crtc->crtc_offset,
8505 (LATENCY_LOW_WATERMARK(latency_watermark_b) |
8506 LATENCY_HIGH_WATERMARK(line_time)));
8507 /* restore original selection */
8508 WREG32(DPG_WATERMARK_MASK_CONTROL + radeon_crtc->crtc_offset, wm_mask);
8509
8510 /* save values for DPM */
8511 radeon_crtc->line_time = line_time;
8512 radeon_crtc->wm_high = latency_watermark_a;
8513 radeon_crtc->wm_low = latency_watermark_b;
8514 }
8515
8516 /**
8517 * dce8_bandwidth_update - program display watermarks
8518 *
8519 * @rdev: radeon_device pointer
8520 *
8521 * Calculate and program the display watermarks and line
8522 * buffer allocation (CIK).
8523 */
8524 void dce8_bandwidth_update(struct radeon_device *rdev)
8525 {
8526 struct drm_display_mode *mode = NULL;
8527 u32 num_heads = 0, lb_size;
8528 int i;
8529
8530 radeon_update_display_priority(rdev);
8531
8532 for (i = 0; i < rdev->num_crtc; i++) {
8533 if (rdev->mode_info.crtcs[i]->base.enabled)
8534 num_heads++;
8535 }
8536 for (i = 0; i < rdev->num_crtc; i++) {
8537 mode = &rdev->mode_info.crtcs[i]->base.mode;
8538 lb_size = dce8_line_buffer_adjust(rdev, rdev->mode_info.crtcs[i], mode);
8539 dce8_program_watermarks(rdev, rdev->mode_info.crtcs[i], lb_size, num_heads);
8540 }
8541 }
8542
8543 /**
8544 * cik_get_gpu_clock_counter - return GPU clock counter snapshot
8545 *
8546 * @rdev: radeon_device pointer
8547 *
8548 * Fetches a GPU clock counter snapshot (SI).
8549 * Returns the 64 bit clock counter snapshot.
8550 */
8551 uint64_t cik_get_gpu_clock_counter(struct radeon_device *rdev)
8552 {
8553 uint64_t clock;
8554
8555 mutex_lock(&rdev->gpu_clock_mutex);
8556 WREG32(RLC_CAPTURE_GPU_CLOCK_COUNT, 1);
8557 clock = (uint64_t)RREG32(RLC_GPU_CLOCK_COUNT_LSB) |
8558 ((uint64_t)RREG32(RLC_GPU_CLOCK_COUNT_MSB) << 32ULL);
8559 mutex_unlock(&rdev->gpu_clock_mutex);
8560 return clock;
8561 }
8562
8563 static int cik_set_uvd_clock(struct radeon_device *rdev, u32 clock,
8564 u32 cntl_reg, u32 status_reg)
8565 {
8566 int r, i;
8567 struct atom_clock_dividers dividers;
8568 uint32_t tmp;
8569
8570 r = radeon_atom_get_clock_dividers(rdev, COMPUTE_GPUCLK_INPUT_FLAG_DEFAULT_GPUCLK,
8571 clock, false, &dividers);
8572 if (r)
8573 return r;
8574
8575 tmp = RREG32_SMC(cntl_reg);
8576 tmp &= ~(DCLK_DIR_CNTL_EN|DCLK_DIVIDER_MASK);
8577 tmp |= dividers.post_divider;
8578 WREG32_SMC(cntl_reg, tmp);
8579
8580 for (i = 0; i < 100; i++) {
8581 if (RREG32_SMC(status_reg) & DCLK_STATUS)
8582 break;
8583 mdelay(10);
8584 }
8585 if (i == 100)
8586 return -ETIMEDOUT;
8587
8588 return 0;
8589 }
8590
8591 int cik_set_uvd_clocks(struct radeon_device *rdev, u32 vclk, u32 dclk)
8592 {
8593 int r = 0;
8594
8595 r = cik_set_uvd_clock(rdev, vclk, CG_VCLK_CNTL, CG_VCLK_STATUS);
8596 if (r)
8597 return r;
8598
8599 r = cik_set_uvd_clock(rdev, dclk, CG_DCLK_CNTL, CG_DCLK_STATUS);
8600 return r;
8601 }
8602
8603 void cik_uvd_resume(struct radeon_device *rdev)
8604 {
8605 uint64_t addr;
8606 uint32_t size;
8607
8608 /* programm the VCPU memory controller bits 0-27 */
8609 addr = rdev->uvd.gpu_addr >> 3;
8610 size = RADEON_GPU_PAGE_ALIGN(rdev->uvd_fw->size + 4) >> 3;
8611 WREG32(UVD_VCPU_CACHE_OFFSET0, addr);
8612 WREG32(UVD_VCPU_CACHE_SIZE0, size);
8613
8614 addr += size;
8615 size = RADEON_UVD_STACK_SIZE >> 3;
8616 WREG32(UVD_VCPU_CACHE_OFFSET1, addr);
8617 WREG32(UVD_VCPU_CACHE_SIZE1, size);
8618
8619 addr += size;
8620 size = RADEON_UVD_HEAP_SIZE >> 3;
8621 WREG32(UVD_VCPU_CACHE_OFFSET2, addr);
8622 WREG32(UVD_VCPU_CACHE_SIZE2, size);
8623
8624 /* bits 28-31 */
8625 addr = (rdev->uvd.gpu_addr >> 28) & 0xF;
8626 WREG32(UVD_LMI_ADDR_EXT, (addr << 12) | (addr << 0));
8627
8628 /* bits 32-39 */
8629 addr = (rdev->uvd.gpu_addr >> 32) & 0xFF;
8630 WREG32(UVD_LMI_EXT40_ADDR, addr | (0x9 << 16) | (0x1 << 31));
8631
8632 }
8633
8634 static void cik_pcie_gen3_enable(struct radeon_device *rdev)
8635 {
8636 struct pci_dev *root = rdev->pdev->bus->self;
8637 int bridge_pos, gpu_pos;
8638 u32 speed_cntl, mask, current_data_rate;
8639 int ret, i;
8640 u16 tmp16;
8641
8642 if (radeon_pcie_gen2 == 0)
8643 return;
8644
8645 if (rdev->flags & RADEON_IS_IGP)
8646 return;
8647
8648 if (!(rdev->flags & RADEON_IS_PCIE))
8649 return;
8650
8651 ret = drm_pcie_get_speed_cap_mask(rdev->ddev, &mask);
8652 if (ret != 0)
8653 return;
8654
8655 if (!(mask & (DRM_PCIE_SPEED_50 | DRM_PCIE_SPEED_80)))
8656 return;
8657
8658 speed_cntl = RREG32_PCIE_PORT(PCIE_LC_SPEED_CNTL);
8659 current_data_rate = (speed_cntl & LC_CURRENT_DATA_RATE_MASK) >>
8660 LC_CURRENT_DATA_RATE_SHIFT;
8661 if (mask & DRM_PCIE_SPEED_80) {
8662 if (current_data_rate == 2) {
8663 DRM_INFO("PCIE gen 3 link speeds already enabled\n");
8664 return;
8665 }
8666 DRM_INFO("enabling PCIE gen 3 link speeds, disable with radeon.pcie_gen2=0\n");
8667 } else if (mask & DRM_PCIE_SPEED_50) {
8668 if (current_data_rate == 1) {
8669 DRM_INFO("PCIE gen 2 link speeds already enabled\n");
8670 return;
8671 }
8672 DRM_INFO("enabling PCIE gen 2 link speeds, disable with radeon.pcie_gen2=0\n");
8673 }
8674
8675 bridge_pos = pci_pcie_cap(root);
8676 if (!bridge_pos)
8677 return;
8678
8679 gpu_pos = pci_pcie_cap(rdev->pdev);
8680 if (!gpu_pos)
8681 return;
8682
8683 if (mask & DRM_PCIE_SPEED_80) {
8684 /* re-try equalization if gen3 is not already enabled */
8685 if (current_data_rate != 2) {
8686 u16 bridge_cfg, gpu_cfg;
8687 u16 bridge_cfg2, gpu_cfg2;
8688 u32 max_lw, current_lw, tmp;
8689
8690 pci_read_config_word(root, bridge_pos + PCI_EXP_LNKCTL, &bridge_cfg);
8691 pci_read_config_word(rdev->pdev, gpu_pos + PCI_EXP_LNKCTL, &gpu_cfg);
8692
8693 tmp16 = bridge_cfg | PCI_EXP_LNKCTL_HAWD;
8694 pci_write_config_word(root, bridge_pos + PCI_EXP_LNKCTL, tmp16);
8695
8696 tmp16 = gpu_cfg | PCI_EXP_LNKCTL_HAWD;
8697 pci_write_config_word(rdev->pdev, gpu_pos + PCI_EXP_LNKCTL, tmp16);
8698
8699 tmp = RREG32_PCIE_PORT(PCIE_LC_STATUS1);
8700 max_lw = (tmp & LC_DETECTED_LINK_WIDTH_MASK) >> LC_DETECTED_LINK_WIDTH_SHIFT;
8701 current_lw = (tmp & LC_OPERATING_LINK_WIDTH_MASK) >> LC_OPERATING_LINK_WIDTH_SHIFT;
8702
8703 if (current_lw < max_lw) {
8704 tmp = RREG32_PCIE_PORT(PCIE_LC_LINK_WIDTH_CNTL);
8705 if (tmp & LC_RENEGOTIATION_SUPPORT) {
8706 tmp &= ~(LC_LINK_WIDTH_MASK | LC_UPCONFIGURE_DIS);
8707 tmp |= (max_lw << LC_LINK_WIDTH_SHIFT);
8708 tmp |= LC_UPCONFIGURE_SUPPORT | LC_RENEGOTIATE_EN | LC_RECONFIG_NOW;
8709 WREG32_PCIE_PORT(PCIE_LC_LINK_WIDTH_CNTL, tmp);
8710 }
8711 }
8712
8713 for (i = 0; i < 10; i++) {
8714 /* check status */
8715 pci_read_config_word(rdev->pdev, gpu_pos + PCI_EXP_DEVSTA, &tmp16);
8716 if (tmp16 & PCI_EXP_DEVSTA_TRPND)
8717 break;
8718
8719 pci_read_config_word(root, bridge_pos + PCI_EXP_LNKCTL, &bridge_cfg);
8720 pci_read_config_word(rdev->pdev, gpu_pos + PCI_EXP_LNKCTL, &gpu_cfg);
8721
8722 pci_read_config_word(root, bridge_pos + PCI_EXP_LNKCTL2, &bridge_cfg2);
8723 pci_read_config_word(rdev->pdev, gpu_pos + PCI_EXP_LNKCTL2, &gpu_cfg2);
8724
8725 tmp = RREG32_PCIE_PORT(PCIE_LC_CNTL4);
8726 tmp |= LC_SET_QUIESCE;
8727 WREG32_PCIE_PORT(PCIE_LC_CNTL4, tmp);
8728
8729 tmp = RREG32_PCIE_PORT(PCIE_LC_CNTL4);
8730 tmp |= LC_REDO_EQ;
8731 WREG32_PCIE_PORT(PCIE_LC_CNTL4, tmp);
8732
8733 mdelay(100);
8734
8735 /* linkctl */
8736 pci_read_config_word(root, bridge_pos + PCI_EXP_LNKCTL, &tmp16);
8737 tmp16 &= ~PCI_EXP_LNKCTL_HAWD;
8738 tmp16 |= (bridge_cfg & PCI_EXP_LNKCTL_HAWD);
8739 pci_write_config_word(root, bridge_pos + PCI_EXP_LNKCTL, tmp16);
8740
8741 pci_read_config_word(rdev->pdev, gpu_pos + PCI_EXP_LNKCTL, &tmp16);
8742 tmp16 &= ~PCI_EXP_LNKCTL_HAWD;
8743 tmp16 |= (gpu_cfg & PCI_EXP_LNKCTL_HAWD);
8744 pci_write_config_word(rdev->pdev, gpu_pos + PCI_EXP_LNKCTL, tmp16);
8745
8746 /* linkctl2 */
8747 pci_read_config_word(root, bridge_pos + PCI_EXP_LNKCTL2, &tmp16);
8748 tmp16 &= ~((1 << 4) | (7 << 9));
8749 tmp16 |= (bridge_cfg2 & ((1 << 4) | (7 << 9)));
8750 pci_write_config_word(root, bridge_pos + PCI_EXP_LNKCTL2, tmp16);
8751
8752 pci_read_config_word(rdev->pdev, gpu_pos + PCI_EXP_LNKCTL2, &tmp16);
8753 tmp16 &= ~((1 << 4) | (7 << 9));
8754 tmp16 |= (gpu_cfg2 & ((1 << 4) | (7 << 9)));
8755 pci_write_config_word(rdev->pdev, gpu_pos + PCI_EXP_LNKCTL2, tmp16);
8756
8757 tmp = RREG32_PCIE_PORT(PCIE_LC_CNTL4);
8758 tmp &= ~LC_SET_QUIESCE;
8759 WREG32_PCIE_PORT(PCIE_LC_CNTL4, tmp);
8760 }
8761 }
8762 }
8763
8764 /* set the link speed */
8765 speed_cntl |= LC_FORCE_EN_SW_SPEED_CHANGE | LC_FORCE_DIS_HW_SPEED_CHANGE;
8766 speed_cntl &= ~LC_FORCE_DIS_SW_SPEED_CHANGE;
8767 WREG32_PCIE_PORT(PCIE_LC_SPEED_CNTL, speed_cntl);
8768
8769 pci_read_config_word(rdev->pdev, gpu_pos + PCI_EXP_LNKCTL2, &tmp16);
8770 tmp16 &= ~0xf;
8771 if (mask & DRM_PCIE_SPEED_80)
8772 tmp16 |= 3; /* gen3 */
8773 else if (mask & DRM_PCIE_SPEED_50)
8774 tmp16 |= 2; /* gen2 */
8775 else
8776 tmp16 |= 1; /* gen1 */
8777 pci_write_config_word(rdev->pdev, gpu_pos + PCI_EXP_LNKCTL2, tmp16);
8778
8779 speed_cntl = RREG32_PCIE_PORT(PCIE_LC_SPEED_CNTL);
8780 speed_cntl |= LC_INITIATE_LINK_SPEED_CHANGE;
8781 WREG32_PCIE_PORT(PCIE_LC_SPEED_CNTL, speed_cntl);
8782
8783 for (i = 0; i < rdev->usec_timeout; i++) {
8784 speed_cntl = RREG32_PCIE_PORT(PCIE_LC_SPEED_CNTL);
8785 if ((speed_cntl & LC_INITIATE_LINK_SPEED_CHANGE) == 0)
8786 break;
8787 udelay(1);
8788 }
8789 }
8790
8791 static void cik_program_aspm(struct radeon_device *rdev)
8792 {
8793 u32 data, orig;
8794 bool disable_l0s = false, disable_l1 = false, disable_plloff_in_l1 = false;
8795 bool disable_clkreq = false;
8796
8797 if (radeon_aspm == 0)
8798 return;
8799
8800 /* XXX double check IGPs */
8801 if (rdev->flags & RADEON_IS_IGP)
8802 return;
8803
8804 if (!(rdev->flags & RADEON_IS_PCIE))
8805 return;
8806
8807 orig = data = RREG32_PCIE_PORT(PCIE_LC_N_FTS_CNTL);
8808 data &= ~LC_XMIT_N_FTS_MASK;
8809 data |= LC_XMIT_N_FTS(0x24) | LC_XMIT_N_FTS_OVERRIDE_EN;
8810 if (orig != data)
8811 WREG32_PCIE_PORT(PCIE_LC_N_FTS_CNTL, data);
8812
8813 orig = data = RREG32_PCIE_PORT(PCIE_LC_CNTL3);
8814 data |= LC_GO_TO_RECOVERY;
8815 if (orig != data)
8816 WREG32_PCIE_PORT(PCIE_LC_CNTL3, data);
8817
8818 orig = data = RREG32_PCIE_PORT(PCIE_P_CNTL);
8819 data |= P_IGNORE_EDB_ERR;
8820 if (orig != data)
8821 WREG32_PCIE_PORT(PCIE_P_CNTL, data);
8822
8823 orig = data = RREG32_PCIE_PORT(PCIE_LC_CNTL);
8824 data &= ~(LC_L0S_INACTIVITY_MASK | LC_L1_INACTIVITY_MASK);
8825 data |= LC_PMI_TO_L1_DIS;
8826 if (!disable_l0s)
8827 data |= LC_L0S_INACTIVITY(7);
8828
8829 if (!disable_l1) {
8830 data |= LC_L1_INACTIVITY(7);
8831 data &= ~LC_PMI_TO_L1_DIS;
8832 if (orig != data)
8833 WREG32_PCIE_PORT(PCIE_LC_CNTL, data);
8834
8835 if (!disable_plloff_in_l1) {
8836 bool clk_req_support;
8837
8838 orig = data = RREG32_PCIE_PORT(PB0_PIF_PWRDOWN_0);
8839 data &= ~(PLL_POWER_STATE_IN_OFF_0_MASK | PLL_POWER_STATE_IN_TXS2_0_MASK);
8840 data |= PLL_POWER_STATE_IN_OFF_0(7) | PLL_POWER_STATE_IN_TXS2_0(7);
8841 if (orig != data)
8842 WREG32_PCIE_PORT(PB0_PIF_PWRDOWN_0, data);
8843
8844 orig = data = RREG32_PCIE_PORT(PB0_PIF_PWRDOWN_1);
8845 data &= ~(PLL_POWER_STATE_IN_OFF_1_MASK | PLL_POWER_STATE_IN_TXS2_1_MASK);
8846 data |= PLL_POWER_STATE_IN_OFF_1(7) | PLL_POWER_STATE_IN_TXS2_1(7);
8847 if (orig != data)
8848 WREG32_PCIE_PORT(PB0_PIF_PWRDOWN_1, data);
8849
8850 orig = data = RREG32_PCIE_PORT(PB1_PIF_PWRDOWN_0);
8851 data &= ~(PLL_POWER_STATE_IN_OFF_0_MASK | PLL_POWER_STATE_IN_TXS2_0_MASK);
8852 data |= PLL_POWER_STATE_IN_OFF_0(7) | PLL_POWER_STATE_IN_TXS2_0(7);
8853 if (orig != data)
8854 WREG32_PCIE_PORT(PB1_PIF_PWRDOWN_0, data);
8855
8856 orig = data = RREG32_PCIE_PORT(PB1_PIF_PWRDOWN_1);
8857 data &= ~(PLL_POWER_STATE_IN_OFF_1_MASK | PLL_POWER_STATE_IN_TXS2_1_MASK);
8858 data |= PLL_POWER_STATE_IN_OFF_1(7) | PLL_POWER_STATE_IN_TXS2_1(7);
8859 if (orig != data)
8860 WREG32_PCIE_PORT(PB1_PIF_PWRDOWN_1, data);
8861
8862 orig = data = RREG32_PCIE_PORT(PCIE_LC_LINK_WIDTH_CNTL);
8863 data &= ~LC_DYN_LANES_PWR_STATE_MASK;
8864 data |= LC_DYN_LANES_PWR_STATE(3);
8865 if (orig != data)
8866 WREG32_PCIE_PORT(PCIE_LC_LINK_WIDTH_CNTL, data);
8867
8868 if (!disable_clkreq) {
8869 struct pci_dev *root = rdev->pdev->bus->self;
8870 u32 lnkcap;
8871
8872 clk_req_support = false;
8873 pcie_capability_read_dword(root, PCI_EXP_LNKCAP, &lnkcap);
8874 if (lnkcap & PCI_EXP_LNKCAP_CLKPM)
8875 clk_req_support = true;
8876 } else {
8877 clk_req_support = false;
8878 }
8879
8880 if (clk_req_support) {
8881 orig = data = RREG32_PCIE_PORT(PCIE_LC_CNTL2);
8882 data |= LC_ALLOW_PDWN_IN_L1 | LC_ALLOW_PDWN_IN_L23;
8883 if (orig != data)
8884 WREG32_PCIE_PORT(PCIE_LC_CNTL2, data);
8885
8886 orig = data = RREG32_SMC(THM_CLK_CNTL);
8887 data &= ~(CMON_CLK_SEL_MASK | TMON_CLK_SEL_MASK);
8888 data |= CMON_CLK_SEL(1) | TMON_CLK_SEL(1);
8889 if (orig != data)
8890 WREG32_SMC(THM_CLK_CNTL, data);
8891
8892 orig = data = RREG32_SMC(MISC_CLK_CTRL);
8893 data &= ~(DEEP_SLEEP_CLK_SEL_MASK | ZCLK_SEL_MASK);
8894 data |= DEEP_SLEEP_CLK_SEL(1) | ZCLK_SEL(1);
8895 if (orig != data)
8896 WREG32_SMC(MISC_CLK_CTRL, data);
8897
8898 orig = data = RREG32_SMC(CG_CLKPIN_CNTL);
8899 data &= ~BCLK_AS_XCLK;
8900 if (orig != data)
8901 WREG32_SMC(CG_CLKPIN_CNTL, data);
8902
8903 orig = data = RREG32_SMC(CG_CLKPIN_CNTL_2);
8904 data &= ~FORCE_BIF_REFCLK_EN;
8905 if (orig != data)
8906 WREG32_SMC(CG_CLKPIN_CNTL_2, data);
8907
8908 orig = data = RREG32_SMC(MPLL_BYPASSCLK_SEL);
8909 data &= ~MPLL_CLKOUT_SEL_MASK;
8910 data |= MPLL_CLKOUT_SEL(4);
8911 if (orig != data)
8912 WREG32_SMC(MPLL_BYPASSCLK_SEL, data);
8913 }
8914 }
8915 } else {
8916 if (orig != data)
8917 WREG32_PCIE_PORT(PCIE_LC_CNTL, data);
8918 }
8919
8920 orig = data = RREG32_PCIE_PORT(PCIE_CNTL2);
8921 data |= SLV_MEM_LS_EN | MST_MEM_LS_EN | REPLAY_MEM_LS_EN;
8922 if (orig != data)
8923 WREG32_PCIE_PORT(PCIE_CNTL2, data);
8924
8925 if (!disable_l0s) {
8926 data = RREG32_PCIE_PORT(PCIE_LC_N_FTS_CNTL);
8927 if((data & LC_N_FTS_MASK) == LC_N_FTS_MASK) {
8928 data = RREG32_PCIE_PORT(PCIE_LC_STATUS1);
8929 if ((data & LC_REVERSE_XMIT) && (data & LC_REVERSE_RCVR)) {
8930 orig = data = RREG32_PCIE_PORT(PCIE_LC_CNTL);
8931 data &= ~LC_L0S_INACTIVITY_MASK;
8932 if (orig != data)
8933 WREG32_PCIE_PORT(PCIE_LC_CNTL, data);
8934 }
8935 }
8936 }
8937 }
This page took 0.298625 seconds and 6 git commands to generate.