1/*
2 * Copyright 2010 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 <drm/drmP.h>
28#include "radeon.h"
29#include "radeon_asic.h"
30#include "radeon_audio.h"
31#include <drm/radeon_drm.h>
32#include "nid.h"
33#include "atom.h"
34#include "ni_reg.h"
35#include "cayman_blit_shaders.h"
36#include "radeon_ucode.h"
37#include "clearstate_cayman.h"
38
39/*
40 * Indirect registers accessor
41 */
42u32 tn_smc_rreg(struct radeon_device *rdev, u32 reg)
43{
44 unsigned long flags;
45 u32 r;
46
47 spin_lock_irqsave(&rdev->smc_idx_lock, flags);
48 WREG32(TN_SMC_IND_INDEX_0, (reg));
49 r = RREG32(TN_SMC_IND_DATA_0);
50 spin_unlock_irqrestore(&rdev->smc_idx_lock, flags);
51 return r;
52}
53
54void tn_smc_wreg(struct radeon_device *rdev, u32 reg, u32 v)
55{
56 unsigned long flags;
57
58 spin_lock_irqsave(&rdev->smc_idx_lock, flags);
59 WREG32(TN_SMC_IND_INDEX_0, (reg));
60 WREG32(TN_SMC_IND_DATA_0, (v));
61 spin_unlock_irqrestore(&rdev->smc_idx_lock, flags);
62}
63
64static const u32 tn_rlc_save_restore_register_list[] =
65{
66 0x98fc,
67 0x98f0,
68 0x9834,
69 0x9838,
70 0x9870,
71 0x9874,
72 0x8a14,
73 0x8b24,
74 0x8bcc,
75 0x8b10,
76 0x8c30,
77 0x8d00,
78 0x8d04,
79 0x8c00,
80 0x8c04,
81 0x8c10,
82 0x8c14,
83 0x8d8c,
84 0x8cf0,
85 0x8e38,
86 0x9508,
87 0x9688,
88 0x9608,
89 0x960c,
90 0x9610,
91 0x9614,
92 0x88c4,
93 0x8978,
94 0x88d4,
95 0x900c,
96 0x9100,
97 0x913c,
98 0x90e8,
99 0x9354,
100 0xa008,
101 0x98f8,
102 0x9148,
103 0x914c,
104 0x3f94,
105 0x98f4,
106 0x9b7c,
107 0x3f8c,
108 0x8950,
109 0x8954,
110 0x8a18,
111 0x8b28,
112 0x9144,
113 0x3f90,
114 0x915c,
115 0x9160,
116 0x9178,
117 0x917c,
118 0x9180,
119 0x918c,
120 0x9190,
121 0x9194,
122 0x9198,
123 0x919c,
124 0x91a8,
125 0x91ac,
126 0x91b0,
127 0x91b4,
128 0x91b8,
129 0x91c4,
130 0x91c8,
131 0x91cc,
132 0x91d0,
133 0x91d4,
134 0x91e0,
135 0x91e4,
136 0x91ec,
137 0x91f0,
138 0x91f4,
139 0x9200,
140 0x9204,
141 0x929c,
142 0x8030,
143 0x9150,
144 0x9a60,
145 0x920c,
146 0x9210,
147 0x9228,
148 0x922c,
149 0x9244,
150 0x9248,
151 0x91e8,
152 0x9294,
153 0x9208,
154 0x9224,
155 0x9240,
156 0x9220,
157 0x923c,
158 0x9258,
159 0x9744,
160 0xa200,
161 0xa204,
162 0xa208,
163 0xa20c,
164 0x8d58,
165 0x9030,
166 0x9034,
167 0x9038,
168 0x903c,
169 0x9040,
170 0x9654,
171 0x897c,
172 0xa210,
173 0xa214,
174 0x9868,
175 0xa02c,
176 0x9664,
177 0x9698,
178 0x949c,
179 0x8e10,
180 0x8e18,
181 0x8c50,
182 0x8c58,
183 0x8c60,
184 0x8c68,
185 0x89b4,
186 0x9830,
187 0x802c,
188};
189
190extern bool evergreen_is_display_hung(struct radeon_device *rdev);
191extern void evergreen_print_gpu_status_regs(struct radeon_device *rdev);
192extern void evergreen_mc_stop(struct radeon_device *rdev, struct evergreen_mc_save *save);
193extern void evergreen_mc_resume(struct radeon_device *rdev, struct evergreen_mc_save *save);
194extern int evergreen_mc_wait_for_idle(struct radeon_device *rdev);
195extern void evergreen_mc_program(struct radeon_device *rdev);
196extern void evergreen_irq_suspend(struct radeon_device *rdev);
197extern int evergreen_mc_init(struct radeon_device *rdev);
198extern void evergreen_fix_pci_max_read_req_size(struct radeon_device *rdev);
199extern void evergreen_pcie_gen2_enable(struct radeon_device *rdev);
200extern void evergreen_program_aspm(struct radeon_device *rdev);
201extern void sumo_rlc_fini(struct radeon_device *rdev);
202extern int sumo_rlc_init(struct radeon_device *rdev);
203extern void evergreen_gpu_pci_config_reset(struct radeon_device *rdev);
204
205/* Firmware Names */
206MODULE_FIRMWARE("radeon/BARTS_pfp.bin");
207MODULE_FIRMWARE("radeon/BARTS_me.bin");
208MODULE_FIRMWARE("radeon/BARTS_mc.bin");
209MODULE_FIRMWARE("radeon/BARTS_smc.bin");
210MODULE_FIRMWARE("radeon/BTC_rlc.bin");
211MODULE_FIRMWARE("radeon/TURKS_pfp.bin");
212MODULE_FIRMWARE("radeon/TURKS_me.bin");
213MODULE_FIRMWARE("radeon/TURKS_mc.bin");
214MODULE_FIRMWARE("radeon/TURKS_smc.bin");
215MODULE_FIRMWARE("radeon/CAICOS_pfp.bin");
216MODULE_FIRMWARE("radeon/CAICOS_me.bin");
217MODULE_FIRMWARE("radeon/CAICOS_mc.bin");
218MODULE_FIRMWARE("radeon/CAICOS_smc.bin");
219MODULE_FIRMWARE("radeon/CAYMAN_pfp.bin");
220MODULE_FIRMWARE("radeon/CAYMAN_me.bin");
221MODULE_FIRMWARE("radeon/CAYMAN_mc.bin");
222MODULE_FIRMWARE("radeon/CAYMAN_rlc.bin");
223MODULE_FIRMWARE("radeon/CAYMAN_smc.bin");
224MODULE_FIRMWARE("radeon/ARUBA_pfp.bin");
225MODULE_FIRMWARE("radeon/ARUBA_me.bin");
226MODULE_FIRMWARE("radeon/ARUBA_rlc.bin");
227
228
229static const u32 cayman_golden_registers2[] =
230{
231 0x3e5c, 0xffffffff, 0x00000000,
232 0x3e48, 0xffffffff, 0x00000000,
233 0x3e4c, 0xffffffff, 0x00000000,
234 0x3e64, 0xffffffff, 0x00000000,
235 0x3e50, 0xffffffff, 0x00000000,
236 0x3e60, 0xffffffff, 0x00000000
237};
238
239static const u32 cayman_golden_registers[] =
240{
241 0x5eb4, 0xffffffff, 0x00000002,
242 0x5e78, 0x8f311ff1, 0x001000f0,
243 0x3f90, 0xffff0000, 0xff000000,
244 0x9148, 0xffff0000, 0xff000000,
245 0x3f94, 0xffff0000, 0xff000000,
246 0x914c, 0xffff0000, 0xff000000,
247 0xc78, 0x00000080, 0x00000080,
248 0xbd4, 0x70073777, 0x00011003,
249 0xd02c, 0xbfffff1f, 0x08421000,
250 0xd0b8, 0x73773777, 0x02011003,
251 0x5bc0, 0x00200000, 0x50100000,
252 0x98f8, 0x33773777, 0x02011003,
253 0x98fc, 0xffffffff, 0x76541032,
254 0x7030, 0x31000311, 0x00000011,
255 0x2f48, 0x33773777, 0x42010001,
256 0x6b28, 0x00000010, 0x00000012,
257 0x7728, 0x00000010, 0x00000012,
258 0x10328, 0x00000010, 0x00000012,
259 0x10f28, 0x00000010, 0x00000012,
260 0x11b28, 0x00000010, 0x00000012,
261 0x12728, 0x00000010, 0x00000012,
262 0x240c, 0x000007ff, 0x00000000,
263 0x8a14, 0xf000001f, 0x00000007,
264 0x8b24, 0x3fff3fff, 0x00ff0fff,
265 0x8b10, 0x0000ff0f, 0x00000000,
266 0x28a4c, 0x07ffffff, 0x06000000,
267 0x10c, 0x00000001, 0x00010003,
268 0xa02c, 0xffffffff, 0x0000009b,
269 0x913c, 0x0000010f, 0x01000100,
270 0x8c04, 0xf8ff00ff, 0x40600060,
271 0x28350, 0x00000f01, 0x00000000,
272 0x9508, 0x3700001f, 0x00000002,
273 0x960c, 0xffffffff, 0x54763210,
274 0x88c4, 0x001f3ae3, 0x00000082,
275 0x88d0, 0xffffffff, 0x0f40df40,
276 0x88d4, 0x0000001f, 0x00000010,
277 0x8974, 0xffffffff, 0x00000000
278};
279
280static const u32 dvst_golden_registers2[] =
281{
282 0x8f8, 0xffffffff, 0,
283 0x8fc, 0x00380000, 0,
284 0x8f8, 0xffffffff, 1,
285 0x8fc, 0x0e000000, 0
286};
287
288static const u32 dvst_golden_registers[] =
289{
290 0x690, 0x3fff3fff, 0x20c00033,
291 0x918c, 0x0fff0fff, 0x00010006,
292 0x91a8, 0x0fff0fff, 0x00010006,
293 0x9150, 0xffffdfff, 0x6e944040,
294 0x917c, 0x0fff0fff, 0x00030002,
295 0x9198, 0x0fff0fff, 0x00030002,
296 0x915c, 0x0fff0fff, 0x00010000,
297 0x3f90, 0xffff0001, 0xff000000,
298 0x9178, 0x0fff0fff, 0x00070000,
299 0x9194, 0x0fff0fff, 0x00070000,
300 0x9148, 0xffff0001, 0xff000000,
301 0x9190, 0x0fff0fff, 0x00090008,
302 0x91ac, 0x0fff0fff, 0x00090008,
303 0x3f94, 0xffff0000, 0xff000000,
304 0x914c, 0xffff0000, 0xff000000,
305 0x929c, 0x00000fff, 0x00000001,
306 0x55e4, 0xff607fff, 0xfc000100,
307 0x8a18, 0xff000fff, 0x00000100,
308 0x8b28, 0xff000fff, 0x00000100,
309 0x9144, 0xfffc0fff, 0x00000100,
310 0x6ed8, 0x00010101, 0x00010000,
311 0x9830, 0xffffffff, 0x00000000,
312 0x9834, 0xf00fffff, 0x00000400,
313 0x9838, 0xfffffffe, 0x00000000,
314 0xd0c0, 0xff000fff, 0x00000100,
315 0xd02c, 0xbfffff1f, 0x08421000,
316 0xd0b8, 0x73773777, 0x12010001,
317 0x5bb0, 0x000000f0, 0x00000070,
318 0x98f8, 0x73773777, 0x12010001,
319 0x98fc, 0xffffffff, 0x00000010,
320 0x9b7c, 0x00ff0000, 0x00fc0000,
321 0x8030, 0x00001f0f, 0x0000100a,
322 0x2f48, 0x73773777, 0x12010001,
323 0x2408, 0x00030000, 0x000c007f,
324 0x8a14, 0xf000003f, 0x00000007,
325 0x8b24, 0x3fff3fff, 0x00ff0fff,
326 0x8b10, 0x0000ff0f, 0x00000000,
327 0x28a4c, 0x07ffffff, 0x06000000,
328 0x4d8, 0x00000fff, 0x00000100,
329 0xa008, 0xffffffff, 0x00010000,
330 0x913c, 0xffff03ff, 0x01000100,
331 0x8c00, 0x000000ff, 0x00000003,
332 0x8c04, 0xf8ff00ff, 0x40600060,
333 0x8cf0, 0x1fff1fff, 0x08e00410,
334 0x28350, 0x00000f01, 0x00000000,
335 0x9508, 0xf700071f, 0x00000002,
336 0x960c, 0xffffffff, 0x54763210,
337 0x20ef8, 0x01ff01ff, 0x00000002,
338 0x20e98, 0xfffffbff, 0x00200000,
339 0x2015c, 0xffffffff, 0x00000f40,
340 0x88c4, 0x001f3ae3, 0x00000082,
341 0x8978, 0x3fffffff, 0x04050140,
342 0x88d4, 0x0000001f, 0x00000010,
343 0x8974, 0xffffffff, 0x00000000
344};
345
346static const u32 scrapper_golden_registers[] =
347{
348 0x690, 0x3fff3fff, 0x20c00033,
349 0x918c, 0x0fff0fff, 0x00010006,
350 0x918c, 0x0fff0fff, 0x00010006,
351 0x91a8, 0x0fff0fff, 0x00010006,
352 0x91a8, 0x0fff0fff, 0x00010006,
353 0x9150, 0xffffdfff, 0x6e944040,
354 0x9150, 0xffffdfff, 0x6e944040,
355 0x917c, 0x0fff0fff, 0x00030002,
356 0x917c, 0x0fff0fff, 0x00030002,
357 0x9198, 0x0fff0fff, 0x00030002,
358 0x9198, 0x0fff0fff, 0x00030002,
359 0x915c, 0x0fff0fff, 0x00010000,
360 0x915c, 0x0fff0fff, 0x00010000,
361 0x3f90, 0xffff0001, 0xff000000,
362 0x3f90, 0xffff0001, 0xff000000,
363 0x9178, 0x0fff0fff, 0x00070000,
364 0x9178, 0x0fff0fff, 0x00070000,
365 0x9194, 0x0fff0fff, 0x00070000,
366 0x9194, 0x0fff0fff, 0x00070000,
367 0x9148, 0xffff0001, 0xff000000,
368 0x9148, 0xffff0001, 0xff000000,
369 0x9190, 0x0fff0fff, 0x00090008,
370 0x9190, 0x0fff0fff, 0x00090008,
371 0x91ac, 0x0fff0fff, 0x00090008,
372 0x91ac, 0x0fff0fff, 0x00090008,
373 0x3f94, 0xffff0000, 0xff000000,
374 0x3f94, 0xffff0000, 0xff000000,
375 0x914c, 0xffff0000, 0xff000000,
376 0x914c, 0xffff0000, 0xff000000,
377 0x929c, 0x00000fff, 0x00000001,
378 0x929c, 0x00000fff, 0x00000001,
379 0x55e4, 0xff607fff, 0xfc000100,
380 0x8a18, 0xff000fff, 0x00000100,
381 0x8a18, 0xff000fff, 0x00000100,
382 0x8b28, 0xff000fff, 0x00000100,
383 0x8b28, 0xff000fff, 0x00000100,
384 0x9144, 0xfffc0fff, 0x00000100,
385 0x9144, 0xfffc0fff, 0x00000100,
386 0x6ed8, 0x00010101, 0x00010000,
387 0x9830, 0xffffffff, 0x00000000,
388 0x9830, 0xffffffff, 0x00000000,
389 0x9834, 0xf00fffff, 0x00000400,
390 0x9834, 0xf00fffff, 0x00000400,
391 0x9838, 0xfffffffe, 0x00000000,
392 0x9838, 0xfffffffe, 0x00000000,
393 0xd0c0, 0xff000fff, 0x00000100,
394 0xd02c, 0xbfffff1f, 0x08421000,
395 0xd02c, 0xbfffff1f, 0x08421000,
396 0xd0b8, 0x73773777, 0x12010001,
397 0xd0b8, 0x73773777, 0x12010001,
398 0x5bb0, 0x000000f0, 0x00000070,
399 0x98f8, 0x73773777, 0x12010001,
400 0x98f8, 0x73773777, 0x12010001,
401 0x98fc, 0xffffffff, 0x00000010,
402 0x98fc, 0xffffffff, 0x00000010,
403 0x9b7c, 0x00ff0000, 0x00fc0000,
404 0x9b7c, 0x00ff0000, 0x00fc0000,
405 0x8030, 0x00001f0f, 0x0000100a,
406 0x8030, 0x00001f0f, 0x0000100a,
407 0x2f48, 0x73773777, 0x12010001,
408 0x2f48, 0x73773777, 0x12010001,
409 0x2408, 0x00030000, 0x000c007f,
410 0x8a14, 0xf000003f, 0x00000007,
411 0x8a14, 0xf000003f, 0x00000007,
412 0x8b24, 0x3fff3fff, 0x00ff0fff,
413 0x8b24, 0x3fff3fff, 0x00ff0fff,
414 0x8b10, 0x0000ff0f, 0x00000000,
415 0x8b10, 0x0000ff0f, 0x00000000,
416 0x28a4c, 0x07ffffff, 0x06000000,
417 0x28a4c, 0x07ffffff, 0x06000000,
418 0x4d8, 0x00000fff, 0x00000100,
419 0x4d8, 0x00000fff, 0x00000100,
420 0xa008, 0xffffffff, 0x00010000,
421 0xa008, 0xffffffff, 0x00010000,
422 0x913c, 0xffff03ff, 0x01000100,
423 0x913c, 0xffff03ff, 0x01000100,
424 0x90e8, 0x001fffff, 0x010400c0,
425 0x8c00, 0x000000ff, 0x00000003,
426 0x8c00, 0x000000ff, 0x00000003,
427 0x8c04, 0xf8ff00ff, 0x40600060,
428 0x8c04, 0xf8ff00ff, 0x40600060,
429 0x8c30, 0x0000000f, 0x00040005,
430 0x8cf0, 0x1fff1fff, 0x08e00410,
431 0x8cf0, 0x1fff1fff, 0x08e00410,
432 0x900c, 0x00ffffff, 0x0017071f,
433 0x28350, 0x00000f01, 0x00000000,
434 0x28350, 0x00000f01, 0x00000000,
435 0x9508, 0xf700071f, 0x00000002,
436 0x9508, 0xf700071f, 0x00000002,
437 0x9688, 0x00300000, 0x0017000f,
438 0x960c, 0xffffffff, 0x54763210,
439 0x960c, 0xffffffff, 0x54763210,
440 0x20ef8, 0x01ff01ff, 0x00000002,
441 0x20e98, 0xfffffbff, 0x00200000,
442 0x2015c, 0xffffffff, 0x00000f40,
443 0x88c4, 0x001f3ae3, 0x00000082,
444 0x88c4, 0x001f3ae3, 0x00000082,
445 0x8978, 0x3fffffff, 0x04050140,
446 0x8978, 0x3fffffff, 0x04050140,
447 0x88d4, 0x0000001f, 0x00000010,
448 0x88d4, 0x0000001f, 0x00000010,
449 0x8974, 0xffffffff, 0x00000000,
450 0x8974, 0xffffffff, 0x00000000
451};
452
453static void ni_init_golden_registers(struct radeon_device *rdev)
454{
455 switch (rdev->family) {
456 case CHIP_CAYMAN:
457 radeon_program_register_sequence(rdev,
458 cayman_golden_registers,
459 (const u32)ARRAY_SIZE(cayman_golden_registers));
460 radeon_program_register_sequence(rdev,
461 cayman_golden_registers2,
462 (const u32)ARRAY_SIZE(cayman_golden_registers2));
463 break;
464 case CHIP_ARUBA:
465 if ((rdev->pdev->device == 0x9900) ||
466 (rdev->pdev->device == 0x9901) ||
467 (rdev->pdev->device == 0x9903) ||
468 (rdev->pdev->device == 0x9904) ||
469 (rdev->pdev->device == 0x9905) ||
470 (rdev->pdev->device == 0x9906) ||
471 (rdev->pdev->device == 0x9907) ||
472 (rdev->pdev->device == 0x9908) ||
473 (rdev->pdev->device == 0x9909) ||
474 (rdev->pdev->device == 0x990A) ||
475 (rdev->pdev->device == 0x990B) ||
476 (rdev->pdev->device == 0x990C) ||
477 (rdev->pdev->device == 0x990D) ||
478 (rdev->pdev->device == 0x990E) ||
479 (rdev->pdev->device == 0x990F) ||
480 (rdev->pdev->device == 0x9910) ||
481 (rdev->pdev->device == 0x9913) ||
482 (rdev->pdev->device == 0x9917) ||
483 (rdev->pdev->device == 0x9918)) {
484 radeon_program_register_sequence(rdev,
485 dvst_golden_registers,
486 (const u32)ARRAY_SIZE(dvst_golden_registers));
487 radeon_program_register_sequence(rdev,
488 dvst_golden_registers2,
489 (const u32)ARRAY_SIZE(dvst_golden_registers2));
490 } else {
491 radeon_program_register_sequence(rdev,
492 scrapper_golden_registers,
493 (const u32)ARRAY_SIZE(scrapper_golden_registers));
494 radeon_program_register_sequence(rdev,
495 dvst_golden_registers2,
496 (const u32)ARRAY_SIZE(dvst_golden_registers2));
497 }
498 break;
499 default:
500 break;
501 }
502}
503
504#define BTC_IO_MC_REGS_SIZE 29
505
506static const u32 barts_io_mc_regs[BTC_IO_MC_REGS_SIZE][2] = {
507 {0x00000077, 0xff010100},
508 {0x00000078, 0x00000000},
509 {0x00000079, 0x00001434},
510 {0x0000007a, 0xcc08ec08},
511 {0x0000007b, 0x00040000},
512 {0x0000007c, 0x000080c0},
513 {0x0000007d, 0x09000000},
514 {0x0000007e, 0x00210404},
515 {0x00000081, 0x08a8e800},
516 {0x00000082, 0x00030444},
517 {0x00000083, 0x00000000},
518 {0x00000085, 0x00000001},
519 {0x00000086, 0x00000002},
520 {0x00000087, 0x48490000},
521 {0x00000088, 0x20244647},
522 {0x00000089, 0x00000005},
523 {0x0000008b, 0x66030000},
524 {0x0000008c, 0x00006603},
525 {0x0000008d, 0x00000100},
526 {0x0000008f, 0x00001c0a},
527 {0x00000090, 0xff000001},
528 {0x00000094, 0x00101101},
529 {0x00000095, 0x00000fff},
530 {0x00000096, 0x00116fff},
531 {0x00000097, 0x60010000},
532 {0x00000098, 0x10010000},
533 {0x00000099, 0x00006000},
534 {0x0000009a, 0x00001000},
535 {0x0000009f, 0x00946a00}
536};
537
538static const u32 turks_io_mc_regs[BTC_IO_MC_REGS_SIZE][2] = {
539 {0x00000077, 0xff010100},
540 {0x00000078, 0x00000000},
541 {0x00000079, 0x00001434},
542 {0x0000007a, 0xcc08ec08},
543 {0x0000007b, 0x00040000},
544 {0x0000007c, 0x000080c0},
545 {0x0000007d, 0x09000000},
546 {0x0000007e, 0x00210404},
547 {0x00000081, 0x08a8e800},
548 {0x00000082, 0x00030444},
549 {0x00000083, 0x00000000},
550 {0x00000085, 0x00000001},
551 {0x00000086, 0x00000002},
552 {0x00000087, 0x48490000},
553 {0x00000088, 0x20244647},
554 {0x00000089, 0x00000005},
555 {0x0000008b, 0x66030000},
556 {0x0000008c, 0x00006603},
557 {0x0000008d, 0x00000100},
558 {0x0000008f, 0x00001c0a},
559 {0x00000090, 0xff000001},
560 {0x00000094, 0x00101101},
561 {0x00000095, 0x00000fff},
562 {0x00000096, 0x00116fff},
563 {0x00000097, 0x60010000},
564 {0x00000098, 0x10010000},
565 {0x00000099, 0x00006000},
566 {0x0000009a, 0x00001000},
567 {0x0000009f, 0x00936a00}
568};
569
570static const u32 caicos_io_mc_regs[BTC_IO_MC_REGS_SIZE][2] = {
571 {0x00000077, 0xff010100},
572 {0x00000078, 0x00000000},
573 {0x00000079, 0x00001434},
574 {0x0000007a, 0xcc08ec08},
575 {0x0000007b, 0x00040000},
576 {0x0000007c, 0x000080c0},
577 {0x0000007d, 0x09000000},
578 {0x0000007e, 0x00210404},
579 {0x00000081, 0x08a8e800},
580 {0x00000082, 0x00030444},
581 {0x00000083, 0x00000000},
582 {0x00000085, 0x00000001},
583 {0x00000086, 0x00000002},
584 {0x00000087, 0x48490000},
585 {0x00000088, 0x20244647},
586 {0x00000089, 0x00000005},
587 {0x0000008b, 0x66030000},
588 {0x0000008c, 0x00006603},
589 {0x0000008d, 0x00000100},
590 {0x0000008f, 0x00001c0a},
591 {0x00000090, 0xff000001},
592 {0x00000094, 0x00101101},
593 {0x00000095, 0x00000fff},
594 {0x00000096, 0x00116fff},
595 {0x00000097, 0x60010000},
596 {0x00000098, 0x10010000},
597 {0x00000099, 0x00006000},
598 {0x0000009a, 0x00001000},
599 {0x0000009f, 0x00916a00}
600};
601
602static const u32 cayman_io_mc_regs[BTC_IO_MC_REGS_SIZE][2] = {
603 {0x00000077, 0xff010100},
604 {0x00000078, 0x00000000},
605 {0x00000079, 0x00001434},
606 {0x0000007a, 0xcc08ec08},
607 {0x0000007b, 0x00040000},
608 {0x0000007c, 0x000080c0},
609 {0x0000007d, 0x09000000},
610 {0x0000007e, 0x00210404},
611 {0x00000081, 0x08a8e800},
612 {0x00000082, 0x00030444},
613 {0x00000083, 0x00000000},
614 {0x00000085, 0x00000001},
615 {0x00000086, 0x00000002},
616 {0x00000087, 0x48490000},
617 {0x00000088, 0x20244647},
618 {0x00000089, 0x00000005},
619 {0x0000008b, 0x66030000},
620 {0x0000008c, 0x00006603},
621 {0x0000008d, 0x00000100},
622 {0x0000008f, 0x00001c0a},
623 {0x00000090, 0xff000001},
624 {0x00000094, 0x00101101},
625 {0x00000095, 0x00000fff},
626 {0x00000096, 0x00116fff},
627 {0x00000097, 0x60010000},
628 {0x00000098, 0x10010000},
629 {0x00000099, 0x00006000},
630 {0x0000009a, 0x00001000},
631 {0x0000009f, 0x00976b00}
632};
633
634int ni_mc_load_microcode(struct radeon_device *rdev)
635{
636 const __be32 *fw_data;
637 u32 mem_type, running, blackout = 0;
638 u32 *io_mc_regs;
639 int i, ucode_size, regs_size;
640
641 if (!rdev->mc_fw)
642 return -EINVAL;
643
644 switch (rdev->family) {
645 case CHIP_BARTS:
646 io_mc_regs = (u32 *)&barts_io_mc_regs;
647 ucode_size = BTC_MC_UCODE_SIZE;
648 regs_size = BTC_IO_MC_REGS_SIZE;
649 break;
650 case CHIP_TURKS:
651 io_mc_regs = (u32 *)&turks_io_mc_regs;
652 ucode_size = BTC_MC_UCODE_SIZE;
653 regs_size = BTC_IO_MC_REGS_SIZE;
654 break;
655 case CHIP_CAICOS:
656 default:
657 io_mc_regs = (u32 *)&caicos_io_mc_regs;
658 ucode_size = BTC_MC_UCODE_SIZE;
659 regs_size = BTC_IO_MC_REGS_SIZE;
660 break;
661 case CHIP_CAYMAN:
662 io_mc_regs = (u32 *)&cayman_io_mc_regs;
663 ucode_size = CAYMAN_MC_UCODE_SIZE;
664 regs_size = BTC_IO_MC_REGS_SIZE;
665 break;
666 }
667
668 mem_type = (RREG32(MC_SEQ_MISC0) & MC_SEQ_MISC0_GDDR5_MASK) >> MC_SEQ_MISC0_GDDR5_SHIFT;
669 running = RREG32(MC_SEQ_SUP_CNTL) & RUN_MASK;
670
671 if ((mem_type == MC_SEQ_MISC0_GDDR5_VALUE) && (running == 0)) {
672 if (running) {
673 blackout = RREG32(MC_SHARED_BLACKOUT_CNTL);
674 WREG32(MC_SHARED_BLACKOUT_CNTL, 1);
675 }
676
677 /* reset the engine and set to writable */
678 WREG32(MC_SEQ_SUP_CNTL, 0x00000008);
679 WREG32(MC_SEQ_SUP_CNTL, 0x00000010);
680
681 /* load mc io regs */
682 for (i = 0; i < regs_size; i++) {
683 WREG32(MC_SEQ_IO_DEBUG_INDEX, io_mc_regs[(i << 1)]);
684 WREG32(MC_SEQ_IO_DEBUG_DATA, io_mc_regs[(i << 1) + 1]);
685 }
686 /* load the MC ucode */
687 fw_data = (const __be32 *)rdev->mc_fw->data;
688 for (i = 0; i < ucode_size; i++)
689 WREG32(MC_SEQ_SUP_PGM, be32_to_cpup(fw_data++));
690
691 /* put the engine back into the active state */
692 WREG32(MC_SEQ_SUP_CNTL, 0x00000008);
693 WREG32(MC_SEQ_SUP_CNTL, 0x00000004);
694 WREG32(MC_SEQ_SUP_CNTL, 0x00000001);
695
696 /* wait for training to complete */
697 for (i = 0; i < rdev->usec_timeout; i++) {
698 if (RREG32(MC_IO_PAD_CNTL_D0) & MEM_FALL_OUT_CMD)
699 break;
700 udelay(1);
701 }
702
703 if (running)
704 WREG32(MC_SHARED_BLACKOUT_CNTL, blackout);
705 }
706
707 return 0;
708}
709
710int ni_init_microcode(struct radeon_device *rdev)
711{
712 const char *chip_name;
713 const char *rlc_chip_name;
714 size_t pfp_req_size, me_req_size, rlc_req_size, mc_req_size;
715 size_t smc_req_size = 0;
716 char fw_name[30];
717 int err;
718
719 DRM_DEBUG("\n");
720
721 switch (rdev->family) {
722 case CHIP_BARTS:
723 chip_name = "BARTS";
724 rlc_chip_name = "BTC";
725 pfp_req_size = EVERGREEN_PFP_UCODE_SIZE * 4;
726 me_req_size = EVERGREEN_PM4_UCODE_SIZE * 4;
727 rlc_req_size = EVERGREEN_RLC_UCODE_SIZE * 4;
728 mc_req_size = BTC_MC_UCODE_SIZE * 4;
729 smc_req_size = ALIGN(BARTS_SMC_UCODE_SIZE, 4);
730 break;
731 case CHIP_TURKS:
732 chip_name = "TURKS";
733 rlc_chip_name = "BTC";
734 pfp_req_size = EVERGREEN_PFP_UCODE_SIZE * 4;
735 me_req_size = EVERGREEN_PM4_UCODE_SIZE * 4;
736 rlc_req_size = EVERGREEN_RLC_UCODE_SIZE * 4;
737 mc_req_size = BTC_MC_UCODE_SIZE * 4;
738 smc_req_size = ALIGN(TURKS_SMC_UCODE_SIZE, 4);
739 break;
740 case CHIP_CAICOS:
741 chip_name = "CAICOS";
742 rlc_chip_name = "BTC";
743 pfp_req_size = EVERGREEN_PFP_UCODE_SIZE * 4;
744 me_req_size = EVERGREEN_PM4_UCODE_SIZE * 4;
745 rlc_req_size = EVERGREEN_RLC_UCODE_SIZE * 4;
746 mc_req_size = BTC_MC_UCODE_SIZE * 4;
747 smc_req_size = ALIGN(CAICOS_SMC_UCODE_SIZE, 4);
748 break;
749 case CHIP_CAYMAN:
750 chip_name = "CAYMAN";
751 rlc_chip_name = "CAYMAN";
752 pfp_req_size = CAYMAN_PFP_UCODE_SIZE * 4;
753 me_req_size = CAYMAN_PM4_UCODE_SIZE * 4;
754 rlc_req_size = CAYMAN_RLC_UCODE_SIZE * 4;
755 mc_req_size = CAYMAN_MC_UCODE_SIZE * 4;
756 smc_req_size = ALIGN(CAYMAN_SMC_UCODE_SIZE, 4);
757 break;
758 case CHIP_ARUBA:
759 chip_name = "ARUBA";
760 rlc_chip_name = "ARUBA";
761 /* pfp/me same size as CAYMAN */
762 pfp_req_size = CAYMAN_PFP_UCODE_SIZE * 4;
763 me_req_size = CAYMAN_PM4_UCODE_SIZE * 4;
764 rlc_req_size = ARUBA_RLC_UCODE_SIZE * 4;
765 mc_req_size = 0;
766 break;
767 default: BUG();
768 }
769
770 DRM_INFO("Loading %s Microcode\n", chip_name);
771
772 snprintf(fw_name, sizeof(fw_name), "radeon/%s_pfp.bin", chip_name);
773 err = request_firmware(&rdev->pfp_fw, fw_name, rdev->dev);
774 if (err)
775 goto out;
776 if (rdev->pfp_fw->size != pfp_req_size) {
777 pr_err("ni_cp: Bogus length %zu in firmware \"%s\"\n",
778 rdev->pfp_fw->size, fw_name);
779 err = -EINVAL;
780 goto out;
781 }
782
783 snprintf(fw_name, sizeof(fw_name), "radeon/%s_me.bin", chip_name);
784 err = request_firmware(&rdev->me_fw, fw_name, rdev->dev);
785 if (err)
786 goto out;
787 if (rdev->me_fw->size != me_req_size) {
788 pr_err("ni_cp: Bogus length %zu in firmware \"%s\"\n",
789 rdev->me_fw->size, fw_name);
790 err = -EINVAL;
791 }
792
793 snprintf(fw_name, sizeof(fw_name), "radeon/%s_rlc.bin", rlc_chip_name);
794 err = request_firmware(&rdev->rlc_fw, fw_name, rdev->dev);
795 if (err)
796 goto out;
797 if (rdev->rlc_fw->size != rlc_req_size) {
798 pr_err("ni_rlc: Bogus length %zu in firmware \"%s\"\n",
799 rdev->rlc_fw->size, fw_name);
800 err = -EINVAL;
801 }
802
803 /* no MC ucode on TN */
804 if (!(rdev->flags & RADEON_IS_IGP)) {
805 snprintf(fw_name, sizeof(fw_name), "radeon/%s_mc.bin", chip_name);
806 err = request_firmware(&rdev->mc_fw, fw_name, rdev->dev);
807 if (err)
808 goto out;
809 if (rdev->mc_fw->size != mc_req_size) {
810 pr_err("ni_mc: Bogus length %zu in firmware \"%s\"\n",
811 rdev->mc_fw->size, fw_name);
812 err = -EINVAL;
813 }
814 }
815
816 if ((rdev->family >= CHIP_BARTS) && (rdev->family <= CHIP_CAYMAN)) {
817 snprintf(fw_name, sizeof(fw_name), "radeon/%s_smc.bin", chip_name);
818 err = request_firmware(&rdev->smc_fw, fw_name, rdev->dev);
819 if (err) {
820 pr_err("smc: error loading firmware \"%s\"\n", fw_name);
821 release_firmware(rdev->smc_fw);
822 rdev->smc_fw = NULL;
823 err = 0;
824 } else if (rdev->smc_fw->size != smc_req_size) {
825 pr_err("ni_mc: Bogus length %zu in firmware \"%s\"\n",
826 rdev->mc_fw->size, fw_name);
827 err = -EINVAL;
828 }
829 }
830
831out:
832 if (err) {
833 if (err != -EINVAL)
834 pr_err("ni_cp: Failed to load firmware \"%s\"\n",
835 fw_name);
836 release_firmware(rdev->pfp_fw);
837 rdev->pfp_fw = NULL;
838 release_firmware(rdev->me_fw);
839 rdev->me_fw = NULL;
840 release_firmware(rdev->rlc_fw);
841 rdev->rlc_fw = NULL;
842 release_firmware(rdev->mc_fw);
843 rdev->mc_fw = NULL;
844 }
845 return err;
846}
847
848/**
849 * cayman_get_allowed_info_register - fetch the register for the info ioctl
850 *
851 * @rdev: radeon_device pointer
852 * @reg: register offset in bytes
853 * @val: register value
854 *
855 * Returns 0 for success or -EINVAL for an invalid register
856 *
857 */
858int cayman_get_allowed_info_register(struct radeon_device *rdev,
859 u32 reg, u32 *val)
860{
861 switch (reg) {
862 case GRBM_STATUS:
863 case GRBM_STATUS_SE0:
864 case GRBM_STATUS_SE1:
865 case SRBM_STATUS:
866 case SRBM_STATUS2:
867 case (DMA_STATUS_REG + DMA0_REGISTER_OFFSET):
868 case (DMA_STATUS_REG + DMA1_REGISTER_OFFSET):
869 case UVD_STATUS:
870 *val = RREG32(reg);
871 return 0;
872 default:
873 return -EINVAL;
874 }
875}
876
877int tn_get_temp(struct radeon_device *rdev)
878{
879 u32 temp = RREG32_SMC(TN_CURRENT_GNB_TEMP) & 0x7ff;
880 int actual_temp = (temp / 8) - 49;
881
882 return actual_temp * 1000;
883}
884
885/*
886 * Core functions
887 */
888static void cayman_gpu_init(struct radeon_device *rdev)
889{
890 u32 gb_addr_config = 0;
891 u32 mc_shared_chmap, mc_arb_ramcfg;
892 u32 cgts_tcc_disable;
893 u32 sx_debug_1;
894 u32 smx_dc_ctl0;
895 u32 cgts_sm_ctrl_reg;
896 u32 hdp_host_path_cntl;
897 u32 tmp;
898 u32 disabled_rb_mask;
899 int i, j;
900
901 switch (rdev->family) {
902 case CHIP_CAYMAN:
903 rdev->config.cayman.max_shader_engines = 2;
904 rdev->config.cayman.max_pipes_per_simd = 4;
905 rdev->config.cayman.max_tile_pipes = 8;
906 rdev->config.cayman.max_simds_per_se = 12;
907 rdev->config.cayman.max_backends_per_se = 4;
908 rdev->config.cayman.max_texture_channel_caches = 8;
909 rdev->config.cayman.max_gprs = 256;
910 rdev->config.cayman.max_threads = 256;
911 rdev->config.cayman.max_gs_threads = 32;
912 rdev->config.cayman.max_stack_entries = 512;
913 rdev->config.cayman.sx_num_of_sets = 8;
914 rdev->config.cayman.sx_max_export_size = 256;
915 rdev->config.cayman.sx_max_export_pos_size = 64;
916 rdev->config.cayman.sx_max_export_smx_size = 192;
917 rdev->config.cayman.max_hw_contexts = 8;
918 rdev->config.cayman.sq_num_cf_insts = 2;
919
920 rdev->config.cayman.sc_prim_fifo_size = 0x100;
921 rdev->config.cayman.sc_hiz_tile_fifo_size = 0x30;
922 rdev->config.cayman.sc_earlyz_tile_fifo_size = 0x130;
923 gb_addr_config = CAYMAN_GB_ADDR_CONFIG_GOLDEN;
924 break;
925 case CHIP_ARUBA:
926 default:
927 rdev->config.cayman.max_shader_engines = 1;
928 rdev->config.cayman.max_pipes_per_simd = 4;
929 rdev->config.cayman.max_tile_pipes = 2;
930 if ((rdev->pdev->device == 0x9900) ||
931 (rdev->pdev->device == 0x9901) ||
932 (rdev->pdev->device == 0x9905) ||
933 (rdev->pdev->device == 0x9906) ||
934 (rdev->pdev->device == 0x9907) ||
935 (rdev->pdev->device == 0x9908) ||
936 (rdev->pdev->device == 0x9909) ||
937 (rdev->pdev->device == 0x990B) ||
938 (rdev->pdev->device == 0x990C) ||
939 (rdev->pdev->device == 0x990F) ||
940 (rdev->pdev->device == 0x9910) ||
941 (rdev->pdev->device == 0x9917) ||
942 (rdev->pdev->device == 0x9999) ||
943 (rdev->pdev->device == 0x999C)) {
944 rdev->config.cayman.max_simds_per_se = 6;
945 rdev->config.cayman.max_backends_per_se = 2;
946 rdev->config.cayman.max_hw_contexts = 8;
947 rdev->config.cayman.sx_max_export_size = 256;
948 rdev->config.cayman.sx_max_export_pos_size = 64;
949 rdev->config.cayman.sx_max_export_smx_size = 192;
950 } else if ((rdev->pdev->device == 0x9903) ||
951 (rdev->pdev->device == 0x9904) ||
952 (rdev->pdev->device == 0x990A) ||
953 (rdev->pdev->device == 0x990D) ||
954 (rdev->pdev->device == 0x990E) ||
955 (rdev->pdev->device == 0x9913) ||
956 (rdev->pdev->device == 0x9918) ||
957 (rdev->pdev->device == 0x999D)) {
958 rdev->config.cayman.max_simds_per_se = 4;
959 rdev->config.cayman.max_backends_per_se = 2;
960 rdev->config.cayman.max_hw_contexts = 8;
961 rdev->config.cayman.sx_max_export_size = 256;
962 rdev->config.cayman.sx_max_export_pos_size = 64;
963 rdev->config.cayman.sx_max_export_smx_size = 192;
964 } else if ((rdev->pdev->device == 0x9919) ||
965 (rdev->pdev->device == 0x9990) ||
966 (rdev->pdev->device == 0x9991) ||
967 (rdev->pdev->device == 0x9994) ||
968 (rdev->pdev->device == 0x9995) ||
969 (rdev->pdev->device == 0x9996) ||
970 (rdev->pdev->device == 0x999A) ||
971 (rdev->pdev->device == 0x99A0)) {
972 rdev->config.cayman.max_simds_per_se = 3;
973 rdev->config.cayman.max_backends_per_se = 1;
974 rdev->config.cayman.max_hw_contexts = 4;
975 rdev->config.cayman.sx_max_export_size = 128;
976 rdev->config.cayman.sx_max_export_pos_size = 32;
977 rdev->config.cayman.sx_max_export_smx_size = 96;
978 } else {
979 rdev->config.cayman.max_simds_per_se = 2;
980 rdev->config.cayman.max_backends_per_se = 1;
981 rdev->config.cayman.max_hw_contexts = 4;
982 rdev->config.cayman.sx_max_export_size = 128;
983 rdev->config.cayman.sx_max_export_pos_size = 32;
984 rdev->config.cayman.sx_max_export_smx_size = 96;
985 }
986 rdev->config.cayman.max_texture_channel_caches = 2;
987 rdev->config.cayman.max_gprs = 256;
988 rdev->config.cayman.max_threads = 256;
989 rdev->config.cayman.max_gs_threads = 32;
990 rdev->config.cayman.max_stack_entries = 512;
991 rdev->config.cayman.sx_num_of_sets = 8;
992 rdev->config.cayman.sq_num_cf_insts = 2;
993
994 rdev->config.cayman.sc_prim_fifo_size = 0x40;
995 rdev->config.cayman.sc_hiz_tile_fifo_size = 0x30;
996 rdev->config.cayman.sc_earlyz_tile_fifo_size = 0x130;
997 gb_addr_config = ARUBA_GB_ADDR_CONFIG_GOLDEN;
998 break;
999 }
1000
1001 /* Initialize HDP */
1002 for (i = 0, j = 0; i < 32; i++, j += 0x18) {
1003 WREG32((0x2c14 + j), 0x00000000);
1004 WREG32((0x2c18 + j), 0x00000000);
1005 WREG32((0x2c1c + j), 0x00000000);
1006 WREG32((0x2c20 + j), 0x00000000);
1007 WREG32((0x2c24 + j), 0x00000000);
1008 }
1009
1010 WREG32(GRBM_CNTL, GRBM_READ_TIMEOUT(0xff));
1011 WREG32(SRBM_INT_CNTL, 0x1);
1012 WREG32(SRBM_INT_ACK, 0x1);
1013
1014 evergreen_fix_pci_max_read_req_size(rdev);
1015
1016 mc_shared_chmap = RREG32(MC_SHARED_CHMAP);
1017 mc_arb_ramcfg = RREG32(MC_ARB_RAMCFG);
1018
1019 tmp = (mc_arb_ramcfg & NOOFCOLS_MASK) >> NOOFCOLS_SHIFT;
1020 rdev->config.cayman.mem_row_size_in_kb = (4 * (1 << (8 + tmp))) / 1024;
1021 if (rdev->config.cayman.mem_row_size_in_kb > 4)
1022 rdev->config.cayman.mem_row_size_in_kb = 4;
1023 /* XXX use MC settings? */
1024 rdev->config.cayman.shader_engine_tile_size = 32;
1025 rdev->config.cayman.num_gpus = 1;
1026 rdev->config.cayman.multi_gpu_tile_size = 64;
1027
1028 tmp = (gb_addr_config & NUM_PIPES_MASK) >> NUM_PIPES_SHIFT;
1029 rdev->config.cayman.num_tile_pipes = (1 << tmp);
1030 tmp = (gb_addr_config & PIPE_INTERLEAVE_SIZE_MASK) >> PIPE_INTERLEAVE_SIZE_SHIFT;
1031 rdev->config.cayman.mem_max_burst_length_bytes = (tmp + 1) * 256;
1032 tmp = (gb_addr_config & NUM_SHADER_ENGINES_MASK) >> NUM_SHADER_ENGINES_SHIFT;
1033 rdev->config.cayman.num_shader_engines = tmp + 1;
1034 tmp = (gb_addr_config & NUM_GPUS_MASK) >> NUM_GPUS_SHIFT;
1035 rdev->config.cayman.num_gpus = tmp + 1;
1036 tmp = (gb_addr_config & MULTI_GPU_TILE_SIZE_MASK) >> MULTI_GPU_TILE_SIZE_SHIFT;
1037 rdev->config.cayman.multi_gpu_tile_size = 1 << tmp;
1038 tmp = (gb_addr_config & ROW_SIZE_MASK) >> ROW_SIZE_SHIFT;
1039 rdev->config.cayman.mem_row_size_in_kb = 1 << tmp;
1040
1041
1042 /* setup tiling info dword. gb_addr_config is not adequate since it does
1043 * not have bank info, so create a custom tiling dword.
1044 * bits 3:0 num_pipes
1045 * bits 7:4 num_banks
1046 * bits 11:8 group_size
1047 * bits 15:12 row_size
1048 */
1049 rdev->config.cayman.tile_config = 0;
1050 switch (rdev->config.cayman.num_tile_pipes) {
1051 case 1:
1052 default:
1053 rdev->config.cayman.tile_config |= (0 << 0);
1054 break;
1055 case 2:
1056 rdev->config.cayman.tile_config |= (1 << 0);
1057 break;
1058 case 4:
1059 rdev->config.cayman.tile_config |= (2 << 0);
1060 break;
1061 case 8:
1062 rdev->config.cayman.tile_config |= (3 << 0);
1063 break;
1064 }
1065
1066 /* num banks is 8 on all fusion asics. 0 = 4, 1 = 8, 2 = 16 */
1067 if (rdev->flags & RADEON_IS_IGP)
1068 rdev->config.cayman.tile_config |= 1 << 4;
1069 else {
1070 switch ((mc_arb_ramcfg & NOOFBANK_MASK) >> NOOFBANK_SHIFT) {
1071 case 0: /* four banks */
1072 rdev->config.cayman.tile_config |= 0 << 4;
1073 break;
1074 case 1: /* eight banks */
1075 rdev->config.cayman.tile_config |= 1 << 4;
1076 break;
1077 case 2: /* sixteen banks */
1078 default:
1079 rdev->config.cayman.tile_config |= 2 << 4;
1080 break;
1081 }
1082 }
1083 rdev->config.cayman.tile_config |=
1084 ((gb_addr_config & PIPE_INTERLEAVE_SIZE_MASK) >> PIPE_INTERLEAVE_SIZE_SHIFT) << 8;
1085 rdev->config.cayman.tile_config |=
1086 ((gb_addr_config & ROW_SIZE_MASK) >> ROW_SIZE_SHIFT) << 12;
1087
1088 tmp = 0;
1089 for (i = (rdev->config.cayman.max_shader_engines - 1); i >= 0; i--) {
1090 u32 rb_disable_bitmap;
1091
1092 WREG32(GRBM_GFX_INDEX, INSTANCE_BROADCAST_WRITES | SE_INDEX(i));
1093 WREG32(RLC_GFX_INDEX, INSTANCE_BROADCAST_WRITES | SE_INDEX(i));
1094 rb_disable_bitmap = (RREG32(CC_RB_BACKEND_DISABLE) & 0x00ff0000) >> 16;
1095 tmp <<= 4;
1096 tmp |= rb_disable_bitmap;
1097 }
1098 /* enabled rb are just the one not disabled :) */
1099 disabled_rb_mask = tmp;
1100 tmp = 0;
1101 for (i = 0; i < (rdev->config.cayman.max_backends_per_se * rdev->config.cayman.max_shader_engines); i++)
1102 tmp |= (1 << i);
1103 /* if all the backends are disabled, fix it up here */
1104 if ((disabled_rb_mask & tmp) == tmp) {
1105 for (i = 0; i < (rdev->config.cayman.max_backends_per_se * rdev->config.cayman.max_shader_engines); i++)
1106 disabled_rb_mask &= ~(1 << i);
1107 }
1108
1109 for (i = 0; i < rdev->config.cayman.max_shader_engines; i++) {
1110 u32 simd_disable_bitmap;
1111
1112 WREG32(GRBM_GFX_INDEX, INSTANCE_BROADCAST_WRITES | SE_INDEX(i));
1113 WREG32(RLC_GFX_INDEX, INSTANCE_BROADCAST_WRITES | SE_INDEX(i));
1114 simd_disable_bitmap = (RREG32(CC_GC_SHADER_PIPE_CONFIG) & 0xffff0000) >> 16;
1115 simd_disable_bitmap |= 0xffffffff << rdev->config.cayman.max_simds_per_se;
1116 tmp <<= 16;
1117 tmp |= simd_disable_bitmap;
1118 }
1119 rdev->config.cayman.active_simds = hweight32(~tmp);
1120
1121 WREG32(GRBM_GFX_INDEX, INSTANCE_BROADCAST_WRITES | SE_BROADCAST_WRITES);
1122 WREG32(RLC_GFX_INDEX, INSTANCE_BROADCAST_WRITES | SE_BROADCAST_WRITES);
1123
1124 WREG32(GB_ADDR_CONFIG, gb_addr_config);
1125 WREG32(DMIF_ADDR_CONFIG, gb_addr_config);
1126 if (ASIC_IS_DCE6(rdev))
1127 WREG32(DMIF_ADDR_CALC, gb_addr_config);
1128 WREG32(HDP_ADDR_CONFIG, gb_addr_config);
1129 WREG32(DMA_TILING_CONFIG + DMA0_REGISTER_OFFSET, gb_addr_config);
1130 WREG32(DMA_TILING_CONFIG + DMA1_REGISTER_OFFSET, gb_addr_config);
1131 WREG32(UVD_UDEC_ADDR_CONFIG, gb_addr_config);
1132 WREG32(UVD_UDEC_DB_ADDR_CONFIG, gb_addr_config);
1133 WREG32(UVD_UDEC_DBW_ADDR_CONFIG, gb_addr_config);
1134
1135 if ((rdev->config.cayman.max_backends_per_se == 1) &&
1136 (rdev->flags & RADEON_IS_IGP)) {
1137 if ((disabled_rb_mask & 3) == 2) {
1138 /* RB1 disabled, RB0 enabled */
1139 tmp = 0x00000000;
1140 } else {
1141 /* RB0 disabled, RB1 enabled */
1142 tmp = 0x11111111;
1143 }
1144 } else {
1145 tmp = gb_addr_config & NUM_PIPES_MASK;
1146 tmp = r6xx_remap_render_backend(rdev, tmp,
1147 rdev->config.cayman.max_backends_per_se *
1148 rdev->config.cayman.max_shader_engines,
1149 CAYMAN_MAX_BACKENDS, disabled_rb_mask);
1150 }
1151 rdev->config.cayman.backend_map = tmp;
1152 WREG32(GB_BACKEND_MAP, tmp);
1153
1154 cgts_tcc_disable = 0xffff0000;
1155 for (i = 0; i < rdev->config.cayman.max_texture_channel_caches; i++)
1156 cgts_tcc_disable &= ~(1 << (16 + i));
1157 WREG32(CGTS_TCC_DISABLE, cgts_tcc_disable);
1158 WREG32(CGTS_SYS_TCC_DISABLE, cgts_tcc_disable);
1159 WREG32(CGTS_USER_SYS_TCC_DISABLE, cgts_tcc_disable);
1160 WREG32(CGTS_USER_TCC_DISABLE, cgts_tcc_disable);
1161
1162 /* reprogram the shader complex */
1163 cgts_sm_ctrl_reg = RREG32(CGTS_SM_CTRL_REG);
1164 for (i = 0; i < 16; i++)
1165 WREG32(CGTS_SM_CTRL_REG, OVERRIDE);
1166 WREG32(CGTS_SM_CTRL_REG, cgts_sm_ctrl_reg);
1167
1168 /* set HW defaults for 3D engine */
1169 WREG32(CP_MEQ_THRESHOLDS, MEQ1_START(0x30) | MEQ2_START(0x60));
1170
1171 sx_debug_1 = RREG32(SX_DEBUG_1);
1172 sx_debug_1 |= ENABLE_NEW_SMX_ADDRESS;
1173 WREG32(SX_DEBUG_1, sx_debug_1);
1174
1175 smx_dc_ctl0 = RREG32(SMX_DC_CTL0);
1176 smx_dc_ctl0 &= ~NUMBER_OF_SETS(0x1ff);
1177 smx_dc_ctl0 |= NUMBER_OF_SETS(rdev->config.cayman.sx_num_of_sets);
1178 WREG32(SMX_DC_CTL0, smx_dc_ctl0);
1179
1180 WREG32(SPI_CONFIG_CNTL_1, VTX_DONE_DELAY(4) | CRC_SIMD_ID_WADDR_DISABLE);
1181
1182 /* need to be explicitly zero-ed */
1183 WREG32(VGT_OFFCHIP_LDS_BASE, 0);
1184 WREG32(SQ_LSTMP_RING_BASE, 0);
1185 WREG32(SQ_HSTMP_RING_BASE, 0);
1186 WREG32(SQ_ESTMP_RING_BASE, 0);
1187 WREG32(SQ_GSTMP_RING_BASE, 0);
1188 WREG32(SQ_VSTMP_RING_BASE, 0);
1189 WREG32(SQ_PSTMP_RING_BASE, 0);
1190
1191 WREG32(TA_CNTL_AUX, DISABLE_CUBE_ANISO);
1192
1193 WREG32(SX_EXPORT_BUFFER_SIZES, (COLOR_BUFFER_SIZE((rdev->config.cayman.sx_max_export_size / 4) - 1) |
1194 POSITION_BUFFER_SIZE((rdev->config.cayman.sx_max_export_pos_size / 4) - 1) |
1195 SMX_BUFFER_SIZE((rdev->config.cayman.sx_max_export_smx_size / 4) - 1)));
1196
1197 WREG32(PA_SC_FIFO_SIZE, (SC_PRIM_FIFO_SIZE(rdev->config.cayman.sc_prim_fifo_size) |
1198 SC_HIZ_TILE_FIFO_SIZE(rdev->config.cayman.sc_hiz_tile_fifo_size) |
1199 SC_EARLYZ_TILE_FIFO_SIZE(rdev->config.cayman.sc_earlyz_tile_fifo_size)));
1200
1201
1202 WREG32(VGT_NUM_INSTANCES, 1);
1203
1204 WREG32(CP_PERFMON_CNTL, 0);
1205
1206 WREG32(SQ_MS_FIFO_SIZES, (CACHE_FIFO_SIZE(16 * rdev->config.cayman.sq_num_cf_insts) |
1207 FETCH_FIFO_HIWATER(0x4) |
1208 DONE_FIFO_HIWATER(0xe0) |
1209 ALU_UPDATE_FIFO_HIWATER(0x8)));
1210
1211 WREG32(SQ_GPR_RESOURCE_MGMT_1, NUM_CLAUSE_TEMP_GPRS(4));
1212 WREG32(SQ_CONFIG, (VC_ENABLE |
1213 EXPORT_SRC_C |
1214 GFX_PRIO(0) |
1215 CS1_PRIO(0) |
1216 CS2_PRIO(1)));
1217 WREG32(SQ_DYN_GPR_CNTL_PS_FLUSH_REQ, DYN_GPR_ENABLE);
1218
1219 WREG32(PA_SC_FORCE_EOV_MAX_CNTS, (FORCE_EOV_MAX_CLK_CNT(4095) |
1220 FORCE_EOV_MAX_REZ_CNT(255)));
1221
1222 WREG32(VGT_CACHE_INVALIDATION, CACHE_INVALIDATION(VC_AND_TC) |
1223 AUTO_INVLD_EN(ES_AND_GS_AUTO));
1224
1225 WREG32(VGT_GS_VERTEX_REUSE, 16);
1226 WREG32(PA_SC_LINE_STIPPLE_STATE, 0);
1227
1228 WREG32(CB_PERF_CTR0_SEL_0, 0);
1229 WREG32(CB_PERF_CTR0_SEL_1, 0);
1230 WREG32(CB_PERF_CTR1_SEL_0, 0);
1231 WREG32(CB_PERF_CTR1_SEL_1, 0);
1232 WREG32(CB_PERF_CTR2_SEL_0, 0);
1233 WREG32(CB_PERF_CTR2_SEL_1, 0);
1234 WREG32(CB_PERF_CTR3_SEL_0, 0);
1235 WREG32(CB_PERF_CTR3_SEL_1, 0);
1236
1237 tmp = RREG32(HDP_MISC_CNTL);
1238 tmp |= HDP_FLUSH_INVALIDATE_CACHE;
1239 WREG32(HDP_MISC_CNTL, tmp);
1240
1241 hdp_host_path_cntl = RREG32(HDP_HOST_PATH_CNTL);
1242 WREG32(HDP_HOST_PATH_CNTL, hdp_host_path_cntl);
1243
1244 WREG32(PA_CL_ENHANCE, CLIP_VTX_REORDER_ENA | NUM_CLIP_SEQ(3));
1245
1246 udelay(50);
1247
1248 /* set clockgating golden values on TN */
1249 if (rdev->family == CHIP_ARUBA) {
1250 tmp = RREG32_CG(CG_CGTT_LOCAL_0);
1251 tmp &= ~0x00380000;
1252 WREG32_CG(CG_CGTT_LOCAL_0, tmp);
1253 tmp = RREG32_CG(CG_CGTT_LOCAL_1);
1254 tmp &= ~0x0e000000;
1255 WREG32_CG(CG_CGTT_LOCAL_1, tmp);
1256 }
1257}
1258
1259/*
1260 * GART
1261 */
1262void cayman_pcie_gart_tlb_flush(struct radeon_device *rdev)
1263{
1264 /* flush hdp cache */
1265 WREG32(HDP_MEM_COHERENCY_FLUSH_CNTL, 0x1);
1266
1267 /* bits 0-7 are the VM contexts0-7 */
1268 WREG32(VM_INVALIDATE_REQUEST, 1);
1269}
1270
1271static int cayman_pcie_gart_enable(struct radeon_device *rdev)
1272{
1273 int i, r;
1274
1275 if (rdev->gart.robj == NULL) {
1276 dev_err(rdev->dev, "No VRAM object for PCIE GART.\n");
1277 return -EINVAL;
1278 }
1279 r = radeon_gart_table_vram_pin(rdev);
1280 if (r)
1281 return r;
1282 /* Setup TLB control */
1283 WREG32(MC_VM_MX_L1_TLB_CNTL,
1284 (0xA << 7) |
1285 ENABLE_L1_TLB |
1286 ENABLE_L1_FRAGMENT_PROCESSING |
1287 SYSTEM_ACCESS_MODE_NOT_IN_SYS |
1288 ENABLE_ADVANCED_DRIVER_MODEL |
1289 SYSTEM_APERTURE_UNMAPPED_ACCESS_PASS_THRU);
1290 /* Setup L2 cache */
1291 WREG32(VM_L2_CNTL, ENABLE_L2_CACHE |
1292 ENABLE_L2_FRAGMENT_PROCESSING |
1293 ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE |
1294 ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE |
1295 EFFECTIVE_L2_QUEUE_SIZE(7) |
1296 CONTEXT1_IDENTITY_ACCESS_MODE(1));
1297 WREG32(VM_L2_CNTL2, INVALIDATE_ALL_L1_TLBS | INVALIDATE_L2_CACHE);
1298 WREG32(VM_L2_CNTL3, L2_CACHE_BIGK_ASSOCIATIVITY |
1299 BANK_SELECT(6) |
1300 L2_CACHE_BIGK_FRAGMENT_SIZE(6));
1301 /* setup context0 */
1302 WREG32(VM_CONTEXT0_PAGE_TABLE_START_ADDR, rdev->mc.gtt_start >> 12);
1303 WREG32(VM_CONTEXT0_PAGE_TABLE_END_ADDR, rdev->mc.gtt_end >> 12);
1304 WREG32(VM_CONTEXT0_PAGE_TABLE_BASE_ADDR, rdev->gart.table_addr >> 12);
1305 WREG32(VM_CONTEXT0_PROTECTION_FAULT_DEFAULT_ADDR,
1306 (u32)(rdev->dummy_page.addr >> 12));
1307 WREG32(VM_CONTEXT0_CNTL2, 0);
1308 WREG32(VM_CONTEXT0_CNTL, ENABLE_CONTEXT | PAGE_TABLE_DEPTH(0) |
1309 RANGE_PROTECTION_FAULT_ENABLE_DEFAULT);
1310
1311 WREG32(0x15D4, 0);
1312 WREG32(0x15D8, 0);
1313 WREG32(0x15DC, 0);
1314
1315 /* empty context1-7 */
1316 /* Assign the pt base to something valid for now; the pts used for
1317 * the VMs are determined by the application and setup and assigned
1318 * on the fly in the vm part of radeon_gart.c
1319 */
1320 for (i = 1; i < 8; i++) {
1321 WREG32(VM_CONTEXT0_PAGE_TABLE_START_ADDR + (i << 2), 0);
1322 WREG32(VM_CONTEXT0_PAGE_TABLE_END_ADDR + (i << 2),
1323 rdev->vm_manager.max_pfn - 1);
1324 WREG32(VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + (i << 2),
1325 rdev->vm_manager.saved_table_addr[i]);
1326 }
1327
1328 /* enable context1-7 */
1329 WREG32(VM_CONTEXT1_PROTECTION_FAULT_DEFAULT_ADDR,
1330 (u32)(rdev->dummy_page.addr >> 12));
1331 WREG32(VM_CONTEXT1_CNTL2, 4);
1332 WREG32(VM_CONTEXT1_CNTL, ENABLE_CONTEXT | PAGE_TABLE_DEPTH(1) |
1333 PAGE_TABLE_BLOCK_SIZE(radeon_vm_block_size - 9) |
1334 RANGE_PROTECTION_FAULT_ENABLE_INTERRUPT |
1335 RANGE_PROTECTION_FAULT_ENABLE_DEFAULT |
1336 DUMMY_PAGE_PROTECTION_FAULT_ENABLE_INTERRUPT |
1337 DUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT |
1338 PDE0_PROTECTION_FAULT_ENABLE_INTERRUPT |
1339 PDE0_PROTECTION_FAULT_ENABLE_DEFAULT |
1340 VALID_PROTECTION_FAULT_ENABLE_INTERRUPT |
1341 VALID_PROTECTION_FAULT_ENABLE_DEFAULT |
1342 READ_PROTECTION_FAULT_ENABLE_INTERRUPT |
1343 READ_PROTECTION_FAULT_ENABLE_DEFAULT |
1344 WRITE_PROTECTION_FAULT_ENABLE_INTERRUPT |
1345 WRITE_PROTECTION_FAULT_ENABLE_DEFAULT);
1346
1347 cayman_pcie_gart_tlb_flush(rdev);
1348 DRM_INFO("PCIE GART of %uM enabled (table at 0x%016llX).\n",
1349 (unsigned)(rdev->mc.gtt_size >> 20),
1350 (unsigned long long)rdev->gart.table_addr);
1351 rdev->gart.ready = true;
1352 return 0;
1353}
1354
1355static void cayman_pcie_gart_disable(struct radeon_device *rdev)
1356{
1357 unsigned i;
1358
1359 for (i = 1; i < 8; ++i) {
1360 rdev->vm_manager.saved_table_addr[i] = RREG32(
1361 VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + (i << 2));
1362 }
1363
1364 /* Disable all tables */
1365 WREG32(VM_CONTEXT0_CNTL, 0);
1366 WREG32(VM_CONTEXT1_CNTL, 0);
1367 /* Setup TLB control */
1368 WREG32(MC_VM_MX_L1_TLB_CNTL, ENABLE_L1_FRAGMENT_PROCESSING |
1369 SYSTEM_ACCESS_MODE_NOT_IN_SYS |
1370 SYSTEM_APERTURE_UNMAPPED_ACCESS_PASS_THRU);
1371 /* Setup L2 cache */
1372 WREG32(VM_L2_CNTL, ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE |
1373 ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE |
1374 EFFECTIVE_L2_QUEUE_SIZE(7) |
1375 CONTEXT1_IDENTITY_ACCESS_MODE(1));
1376 WREG32(VM_L2_CNTL2, 0);
1377 WREG32(VM_L2_CNTL3, L2_CACHE_BIGK_ASSOCIATIVITY |
1378 L2_CACHE_BIGK_FRAGMENT_SIZE(6));
1379 radeon_gart_table_vram_unpin(rdev);
1380}
1381
1382static void cayman_pcie_gart_fini(struct radeon_device *rdev)
1383{
1384 cayman_pcie_gart_disable(rdev);
1385 radeon_gart_table_vram_free(rdev);
1386 radeon_gart_fini(rdev);
1387}
1388
1389void cayman_cp_int_cntl_setup(struct radeon_device *rdev,
1390 int ring, u32 cp_int_cntl)
1391{
1392 WREG32(SRBM_GFX_CNTL, RINGID(ring));
1393 WREG32(CP_INT_CNTL, cp_int_cntl);
1394}
1395
1396/*
1397 * CP.
1398 */
1399void cayman_fence_ring_emit(struct radeon_device *rdev,
1400 struct radeon_fence *fence)
1401{
1402 struct radeon_ring *ring = &rdev->ring[fence->ring];
1403 u64 addr = rdev->fence_drv[fence->ring].gpu_addr;
1404 u32 cp_coher_cntl = PACKET3_FULL_CACHE_ENA | PACKET3_TC_ACTION_ENA |
1405 PACKET3_SH_ACTION_ENA;
1406
1407 /* flush read cache over gart for this vmid */
1408 radeon_ring_write(ring, PACKET3(PACKET3_SURFACE_SYNC, 3));
1409 radeon_ring_write(ring, PACKET3_ENGINE_ME | cp_coher_cntl);
1410 radeon_ring_write(ring, 0xFFFFFFFF);
1411 radeon_ring_write(ring, 0);
1412 radeon_ring_write(ring, 10); /* poll interval */
1413 /* EVENT_WRITE_EOP - flush caches, send int */
1414 radeon_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE_EOP, 4));
1415 radeon_ring_write(ring, EVENT_TYPE(CACHE_FLUSH_AND_INV_EVENT_TS) | EVENT_INDEX(5));
1416 radeon_ring_write(ring, lower_32_bits(addr));
1417 radeon_ring_write(ring, (upper_32_bits(addr) & 0xff) | DATA_SEL(1) | INT_SEL(2));
1418 radeon_ring_write(ring, fence->seq);
1419 radeon_ring_write(ring, 0);
1420}
1421
1422void cayman_ring_ib_execute(struct radeon_device *rdev, struct radeon_ib *ib)
1423{
1424 struct radeon_ring *ring = &rdev->ring[ib->ring];
1425 unsigned vm_id = ib->vm ? ib->vm->ids[ib->ring].id : 0;
1426 u32 cp_coher_cntl = PACKET3_FULL_CACHE_ENA | PACKET3_TC_ACTION_ENA |
1427 PACKET3_SH_ACTION_ENA;
1428
1429 /* set to DX10/11 mode */
1430 radeon_ring_write(ring, PACKET3(PACKET3_MODE_CONTROL, 0));
1431 radeon_ring_write(ring, 1);
1432
1433 if (ring->rptr_save_reg) {
1434 uint32_t next_rptr = ring->wptr + 3 + 4 + 8;
1435 radeon_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1));
1436 radeon_ring_write(ring, ((ring->rptr_save_reg -
1437 PACKET3_SET_CONFIG_REG_START) >> 2));
1438 radeon_ring_write(ring, next_rptr);
1439 }
1440
1441 radeon_ring_write(ring, PACKET3(PACKET3_INDIRECT_BUFFER, 2));
1442 radeon_ring_write(ring,
1443#ifdef __BIG_ENDIAN
1444 (2 << 0) |
1445#endif
1446 (ib->gpu_addr & 0xFFFFFFFC));
1447 radeon_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xFF);
1448 radeon_ring_write(ring, ib->length_dw | (vm_id << 24));
1449
1450 /* flush read cache over gart for this vmid */
1451 radeon_ring_write(ring, PACKET3(PACKET3_SURFACE_SYNC, 3));
1452 radeon_ring_write(ring, PACKET3_ENGINE_ME | cp_coher_cntl);
1453 radeon_ring_write(ring, 0xFFFFFFFF);
1454 radeon_ring_write(ring, 0);
1455 radeon_ring_write(ring, (vm_id << 24) | 10); /* poll interval */
1456}
1457
1458static void cayman_cp_enable(struct radeon_device *rdev, bool enable)
1459{
1460 if (enable)
1461 WREG32(CP_ME_CNTL, 0);
1462 else {
1463 if (rdev->asic->copy.copy_ring_index == RADEON_RING_TYPE_GFX_INDEX)
1464 radeon_ttm_set_active_vram_size(rdev, rdev->mc.visible_vram_size);
1465 WREG32(CP_ME_CNTL, (CP_ME_HALT | CP_PFP_HALT));
1466 WREG32(SCRATCH_UMSK, 0);
1467 rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = false;
1468 }
1469}
1470
1471u32 cayman_gfx_get_rptr(struct radeon_device *rdev,
1472 struct radeon_ring *ring)
1473{
1474 u32 rptr;
1475
1476 if (rdev->wb.enabled)
1477 rptr = rdev->wb.wb[ring->rptr_offs/4];
1478 else {
1479 if (ring->idx == RADEON_RING_TYPE_GFX_INDEX)
1480 rptr = RREG32(CP_RB0_RPTR);
1481 else if (ring->idx == CAYMAN_RING_TYPE_CP1_INDEX)
1482 rptr = RREG32(CP_RB1_RPTR);
1483 else
1484 rptr = RREG32(CP_RB2_RPTR);
1485 }
1486
1487 return rptr;
1488}
1489
1490u32 cayman_gfx_get_wptr(struct radeon_device *rdev,
1491 struct radeon_ring *ring)
1492{
1493 u32 wptr;
1494
1495 if (ring->idx == RADEON_RING_TYPE_GFX_INDEX)
1496 wptr = RREG32(CP_RB0_WPTR);
1497 else if (ring->idx == CAYMAN_RING_TYPE_CP1_INDEX)
1498 wptr = RREG32(CP_RB1_WPTR);
1499 else
1500 wptr = RREG32(CP_RB2_WPTR);
1501
1502 return wptr;
1503}
1504
1505void cayman_gfx_set_wptr(struct radeon_device *rdev,
1506 struct radeon_ring *ring)
1507{
1508 if (ring->idx == RADEON_RING_TYPE_GFX_INDEX) {
1509 WREG32(CP_RB0_WPTR, ring->wptr);
1510 (void)RREG32(CP_RB0_WPTR);
1511 } else if (ring->idx == CAYMAN_RING_TYPE_CP1_INDEX) {
1512 WREG32(CP_RB1_WPTR, ring->wptr);
1513 (void)RREG32(CP_RB1_WPTR);
1514 } else {
1515 WREG32(CP_RB2_WPTR, ring->wptr);
1516 (void)RREG32(CP_RB2_WPTR);
1517 }
1518}
1519
1520static int cayman_cp_load_microcode(struct radeon_device *rdev)
1521{
1522 const __be32 *fw_data;
1523 int i;
1524
1525 if (!rdev->me_fw || !rdev->pfp_fw)
1526 return -EINVAL;
1527
1528 cayman_cp_enable(rdev, false);
1529
1530 fw_data = (const __be32 *)rdev->pfp_fw->data;
1531 WREG32(CP_PFP_UCODE_ADDR, 0);
1532 for (i = 0; i < CAYMAN_PFP_UCODE_SIZE; i++)
1533 WREG32(CP_PFP_UCODE_DATA, be32_to_cpup(fw_data++));
1534 WREG32(CP_PFP_UCODE_ADDR, 0);
1535
1536 fw_data = (const __be32 *)rdev->me_fw->data;
1537 WREG32(CP_ME_RAM_WADDR, 0);
1538 for (i = 0; i < CAYMAN_PM4_UCODE_SIZE; i++)
1539 WREG32(CP_ME_RAM_DATA, be32_to_cpup(fw_data++));
1540
1541 WREG32(CP_PFP_UCODE_ADDR, 0);
1542 WREG32(CP_ME_RAM_WADDR, 0);
1543 WREG32(CP_ME_RAM_RADDR, 0);
1544 return 0;
1545}
1546
1547static int cayman_cp_start(struct radeon_device *rdev)
1548{
1549 struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
1550 int r, i;
1551
1552 r = radeon_ring_lock(rdev, ring, 7);
1553 if (r) {
1554 DRM_ERROR("radeon: cp failed to lock ring (%d).\n", r);
1555 return r;
1556 }
1557 radeon_ring_write(ring, PACKET3(PACKET3_ME_INITIALIZE, 5));
1558 radeon_ring_write(ring, 0x1);
1559 radeon_ring_write(ring, 0x0);
1560 radeon_ring_write(ring, rdev->config.cayman.max_hw_contexts - 1);
1561 radeon_ring_write(ring, PACKET3_ME_INITIALIZE_DEVICE_ID(1));
1562 radeon_ring_write(ring, 0);
1563 radeon_ring_write(ring, 0);
1564 radeon_ring_unlock_commit(rdev, ring, false);
1565
1566 cayman_cp_enable(rdev, true);
1567
1568 r = radeon_ring_lock(rdev, ring, cayman_default_size + 19);
1569 if (r) {
1570 DRM_ERROR("radeon: cp failed to lock ring (%d).\n", r);
1571 return r;
1572 }
1573
1574 /* setup clear context state */
1575 radeon_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
1576 radeon_ring_write(ring, PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
1577
1578 for (i = 0; i < cayman_default_size; i++)
1579 radeon_ring_write(ring, cayman_default_state[i]);
1580
1581 radeon_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
1582 radeon_ring_write(ring, PACKET3_PREAMBLE_END_CLEAR_STATE);
1583
1584 /* set clear context state */
1585 radeon_ring_write(ring, PACKET3(PACKET3_CLEAR_STATE, 0));
1586 radeon_ring_write(ring, 0);
1587
1588 /* SQ_VTX_BASE_VTX_LOC */
1589 radeon_ring_write(ring, 0xc0026f00);
1590 radeon_ring_write(ring, 0x00000000);
1591 radeon_ring_write(ring, 0x00000000);
1592 radeon_ring_write(ring, 0x00000000);
1593
1594 /* Clear consts */
1595 radeon_ring_write(ring, 0xc0036f00);
1596 radeon_ring_write(ring, 0x00000bc4);
1597 radeon_ring_write(ring, 0xffffffff);
1598 radeon_ring_write(ring, 0xffffffff);
1599 radeon_ring_write(ring, 0xffffffff);
1600
1601 radeon_ring_write(ring, 0xc0026900);
1602 radeon_ring_write(ring, 0x00000316);
1603 radeon_ring_write(ring, 0x0000000e); /* VGT_VERTEX_REUSE_BLOCK_CNTL */
1604 radeon_ring_write(ring, 0x00000010); /* */
1605
1606 radeon_ring_unlock_commit(rdev, ring, false);
1607
1608 /* XXX init other rings */
1609
1610 return 0;
1611}
1612
1613static void cayman_cp_fini(struct radeon_device *rdev)
1614{
1615 struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
1616 cayman_cp_enable(rdev, false);
1617 radeon_ring_fini(rdev, ring);
1618 radeon_scratch_free(rdev, ring->rptr_save_reg);
1619}
1620
1621static int cayman_cp_resume(struct radeon_device *rdev)
1622{
1623 static const int ridx[] = {
1624 RADEON_RING_TYPE_GFX_INDEX,
1625 CAYMAN_RING_TYPE_CP1_INDEX,
1626 CAYMAN_RING_TYPE_CP2_INDEX
1627 };
1628 static const unsigned cp_rb_cntl[] = {
1629 CP_RB0_CNTL,
1630 CP_RB1_CNTL,
1631 CP_RB2_CNTL,
1632 };
1633 static const unsigned cp_rb_rptr_addr[] = {
1634 CP_RB0_RPTR_ADDR,
1635 CP_RB1_RPTR_ADDR,
1636 CP_RB2_RPTR_ADDR
1637 };
1638 static const unsigned cp_rb_rptr_addr_hi[] = {
1639 CP_RB0_RPTR_ADDR_HI,
1640 CP_RB1_RPTR_ADDR_HI,
1641 CP_RB2_RPTR_ADDR_HI
1642 };
1643 static const unsigned cp_rb_base[] = {
1644 CP_RB0_BASE,
1645 CP_RB1_BASE,
1646 CP_RB2_BASE
1647 };
1648 static const unsigned cp_rb_rptr[] = {
1649 CP_RB0_RPTR,
1650 CP_RB1_RPTR,
1651 CP_RB2_RPTR
1652 };
1653 static const unsigned cp_rb_wptr[] = {
1654 CP_RB0_WPTR,
1655 CP_RB1_WPTR,
1656 CP_RB2_WPTR
1657 };
1658 struct radeon_ring *ring;
1659 int i, r;
1660
1661 /* Reset cp; if cp is reset, then PA, SH, VGT also need to be reset */
1662 WREG32(GRBM_SOFT_RESET, (SOFT_RESET_CP |
1663 SOFT_RESET_PA |
1664 SOFT_RESET_SH |
1665 SOFT_RESET_VGT |
1666 SOFT_RESET_SPI |
1667 SOFT_RESET_SX));
1668 RREG32(GRBM_SOFT_RESET);
1669 mdelay(15);
1670 WREG32(GRBM_SOFT_RESET, 0);
1671 RREG32(GRBM_SOFT_RESET);
1672
1673 WREG32(CP_SEM_WAIT_TIMER, 0x0);
1674 WREG32(CP_SEM_INCOMPLETE_TIMER_CNTL, 0x0);
1675
1676 /* Set the write pointer delay */
1677 WREG32(CP_RB_WPTR_DELAY, 0);
1678
1679 WREG32(CP_DEBUG, (1 << 27));
1680
1681 /* set the wb address whether it's enabled or not */
1682 WREG32(SCRATCH_ADDR, ((rdev->wb.gpu_addr + RADEON_WB_SCRATCH_OFFSET) >> 8) & 0xFFFFFFFF);
1683 WREG32(SCRATCH_UMSK, 0xff);
1684
1685 for (i = 0; i < 3; ++i) {
1686 uint32_t rb_cntl;
1687 uint64_t addr;
1688
1689 /* Set ring buffer size */
1690 ring = &rdev->ring[ridx[i]];
1691 rb_cntl = order_base_2(ring->ring_size / 8);
1692 rb_cntl |= order_base_2(RADEON_GPU_PAGE_SIZE/8) << 8;
1693#ifdef __BIG_ENDIAN
1694 rb_cntl |= BUF_SWAP_32BIT;
1695#endif
1696 WREG32(cp_rb_cntl[i], rb_cntl);
1697
1698 /* set the wb address whether it's enabled or not */
1699 addr = rdev->wb.gpu_addr + RADEON_WB_CP_RPTR_OFFSET;
1700 WREG32(cp_rb_rptr_addr[i], addr & 0xFFFFFFFC);
1701 WREG32(cp_rb_rptr_addr_hi[i], upper_32_bits(addr) & 0xFF);
1702 }
1703
1704 /* set the rb base addr, this causes an internal reset of ALL rings */
1705 for (i = 0; i < 3; ++i) {
1706 ring = &rdev->ring[ridx[i]];
1707 WREG32(cp_rb_base[i], ring->gpu_addr >> 8);
1708 }
1709
1710 for (i = 0; i < 3; ++i) {
1711 /* Initialize the ring buffer's read and write pointers */
1712 ring = &rdev->ring[ridx[i]];
1713 WREG32_P(cp_rb_cntl[i], RB_RPTR_WR_ENA, ~RB_RPTR_WR_ENA);
1714
1715 ring->wptr = 0;
1716 WREG32(cp_rb_rptr[i], 0);
1717 WREG32(cp_rb_wptr[i], ring->wptr);
1718
1719 mdelay(1);
1720 WREG32_P(cp_rb_cntl[i], 0, ~RB_RPTR_WR_ENA);
1721 }
1722
1723 /* start the rings */
1724 cayman_cp_start(rdev);
1725 rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = true;
1726 rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX].ready = false;
1727 rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX].ready = false;
1728 /* this only test cp0 */
1729 r = radeon_ring_test(rdev, RADEON_RING_TYPE_GFX_INDEX, &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]);
1730 if (r) {
1731 rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = false;
1732 rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX].ready = false;
1733 rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX].ready = false;
1734 return r;
1735 }
1736
1737 if (rdev->asic->copy.copy_ring_index == RADEON_RING_TYPE_GFX_INDEX)
1738 radeon_ttm_set_active_vram_size(rdev, rdev->mc.real_vram_size);
1739
1740 return 0;
1741}
1742
1743u32 cayman_gpu_check_soft_reset(struct radeon_device *rdev)
1744{
1745 u32 reset_mask = 0;
1746 u32 tmp;
1747
1748 /* GRBM_STATUS */
1749 tmp = RREG32(GRBM_STATUS);
1750 if (tmp & (PA_BUSY | SC_BUSY |
1751 SH_BUSY | SX_BUSY |
1752 TA_BUSY | VGT_BUSY |
1753 DB_BUSY | CB_BUSY |
1754 GDS_BUSY | SPI_BUSY |
1755 IA_BUSY | IA_BUSY_NO_DMA))
1756 reset_mask |= RADEON_RESET_GFX;
1757
1758 if (tmp & (CF_RQ_PENDING | PF_RQ_PENDING |
1759 CP_BUSY | CP_COHERENCY_BUSY))
1760 reset_mask |= RADEON_RESET_CP;
1761
1762 if (tmp & GRBM_EE_BUSY)
1763 reset_mask |= RADEON_RESET_GRBM | RADEON_RESET_GFX | RADEON_RESET_CP;
1764
1765 /* DMA_STATUS_REG 0 */
1766 tmp = RREG32(DMA_STATUS_REG + DMA0_REGISTER_OFFSET);
1767 if (!(tmp & DMA_IDLE))
1768 reset_mask |= RADEON_RESET_DMA;
1769
1770 /* DMA_STATUS_REG 1 */
1771 tmp = RREG32(DMA_STATUS_REG + DMA1_REGISTER_OFFSET);
1772 if (!(tmp & DMA_IDLE))
1773 reset_mask |= RADEON_RESET_DMA1;
1774
1775 /* SRBM_STATUS2 */
1776 tmp = RREG32(SRBM_STATUS2);
1777 if (tmp & DMA_BUSY)
1778 reset_mask |= RADEON_RESET_DMA;
1779
1780 if (tmp & DMA1_BUSY)
1781 reset_mask |= RADEON_RESET_DMA1;
1782
1783 /* SRBM_STATUS */
1784 tmp = RREG32(SRBM_STATUS);
1785 if (tmp & (RLC_RQ_PENDING | RLC_BUSY))
1786 reset_mask |= RADEON_RESET_RLC;
1787
1788 if (tmp & IH_BUSY)
1789 reset_mask |= RADEON_RESET_IH;
1790
1791 if (tmp & SEM_BUSY)
1792 reset_mask |= RADEON_RESET_SEM;
1793
1794 if (tmp & GRBM_RQ_PENDING)
1795 reset_mask |= RADEON_RESET_GRBM;
1796
1797 if (tmp & VMC_BUSY)
1798 reset_mask |= RADEON_RESET_VMC;
1799
1800 if (tmp & (MCB_BUSY | MCB_NON_DISPLAY_BUSY |
1801 MCC_BUSY | MCD_BUSY))
1802 reset_mask |= RADEON_RESET_MC;
1803
1804 if (evergreen_is_display_hung(rdev))
1805 reset_mask |= RADEON_RESET_DISPLAY;
1806
1807 /* VM_L2_STATUS */
1808 tmp = RREG32(VM_L2_STATUS);
1809 if (tmp & L2_BUSY)
1810 reset_mask |= RADEON_RESET_VMC;
1811
1812 /* Skip MC reset as it's mostly likely not hung, just busy */
1813 if (reset_mask & RADEON_RESET_MC) {
1814 DRM_DEBUG("MC busy: 0x%08X, clearing.\n", reset_mask);
1815 reset_mask &= ~RADEON_RESET_MC;
1816 }
1817
1818 return reset_mask;
1819}
1820
1821static void cayman_gpu_soft_reset(struct radeon_device *rdev, u32 reset_mask)
1822{
1823 struct evergreen_mc_save save;
1824 u32 grbm_soft_reset = 0, srbm_soft_reset = 0;
1825 u32 tmp;
1826
1827 if (reset_mask == 0)
1828 return;
1829
1830 dev_info(rdev->dev, "GPU softreset: 0x%08X\n", reset_mask);
1831
1832 evergreen_print_gpu_status_regs(rdev);
1833 dev_info(rdev->dev, " VM_CONTEXT0_PROTECTION_FAULT_ADDR 0x%08X\n",
1834 RREG32(0x14F8));
1835 dev_info(rdev->dev, " VM_CONTEXT0_PROTECTION_FAULT_STATUS 0x%08X\n",
1836 RREG32(0x14D8));
1837 dev_info(rdev->dev, " VM_CONTEXT1_PROTECTION_FAULT_ADDR 0x%08X\n",
1838 RREG32(0x14FC));
1839 dev_info(rdev->dev, " VM_CONTEXT1_PROTECTION_FAULT_STATUS 0x%08X\n",
1840 RREG32(0x14DC));
1841
1842 /* Disable CP parsing/prefetching */
1843 WREG32(CP_ME_CNTL, CP_ME_HALT | CP_PFP_HALT);
1844
1845 if (reset_mask & RADEON_RESET_DMA) {
1846 /* dma0 */
1847 tmp = RREG32(DMA_RB_CNTL + DMA0_REGISTER_OFFSET);
1848 tmp &= ~DMA_RB_ENABLE;
1849 WREG32(DMA_RB_CNTL + DMA0_REGISTER_OFFSET, tmp);
1850 }
1851
1852 if (reset_mask & RADEON_RESET_DMA1) {
1853 /* dma1 */
1854 tmp = RREG32(DMA_RB_CNTL + DMA1_REGISTER_OFFSET);
1855 tmp &= ~DMA_RB_ENABLE;
1856 WREG32(DMA_RB_CNTL + DMA1_REGISTER_OFFSET, tmp);
1857 }
1858
1859 udelay(50);
1860
1861 evergreen_mc_stop(rdev, &save);
1862 if (evergreen_mc_wait_for_idle(rdev)) {
1863 dev_warn(rdev->dev, "Wait for MC idle timedout !\n");
1864 }
1865
1866 if (reset_mask & (RADEON_RESET_GFX | RADEON_RESET_COMPUTE)) {
1867 grbm_soft_reset = SOFT_RESET_CB |
1868 SOFT_RESET_DB |
1869 SOFT_RESET_GDS |
1870 SOFT_RESET_PA |
1871 SOFT_RESET_SC |
1872 SOFT_RESET_SPI |
1873 SOFT_RESET_SH |
1874 SOFT_RESET_SX |
1875 SOFT_RESET_TC |
1876 SOFT_RESET_TA |
1877 SOFT_RESET_VGT |
1878 SOFT_RESET_IA;
1879 }
1880
1881 if (reset_mask & RADEON_RESET_CP) {
1882 grbm_soft_reset |= SOFT_RESET_CP | SOFT_RESET_VGT;
1883
1884 srbm_soft_reset |= SOFT_RESET_GRBM;
1885 }
1886
1887 if (reset_mask & RADEON_RESET_DMA)
1888 srbm_soft_reset |= SOFT_RESET_DMA;
1889
1890 if (reset_mask & RADEON_RESET_DMA1)
1891 srbm_soft_reset |= SOFT_RESET_DMA1;
1892
1893 if (reset_mask & RADEON_RESET_DISPLAY)
1894 srbm_soft_reset |= SOFT_RESET_DC;
1895
1896 if (reset_mask & RADEON_RESET_RLC)
1897 srbm_soft_reset |= SOFT_RESET_RLC;
1898
1899 if (reset_mask & RADEON_RESET_SEM)
1900 srbm_soft_reset |= SOFT_RESET_SEM;
1901
1902 if (reset_mask & RADEON_RESET_IH)
1903 srbm_soft_reset |= SOFT_RESET_IH;
1904
1905 if (reset_mask & RADEON_RESET_GRBM)
1906 srbm_soft_reset |= SOFT_RESET_GRBM;
1907
1908 if (reset_mask & RADEON_RESET_VMC)
1909 srbm_soft_reset |= SOFT_RESET_VMC;
1910
1911 if (!(rdev->flags & RADEON_IS_IGP)) {
1912 if (reset_mask & RADEON_RESET_MC)
1913 srbm_soft_reset |= SOFT_RESET_MC;
1914 }
1915
1916 if (grbm_soft_reset) {
1917 tmp = RREG32(GRBM_SOFT_RESET);
1918 tmp |= grbm_soft_reset;
1919 dev_info(rdev->dev, "GRBM_SOFT_RESET=0x%08X\n", tmp);
1920 WREG32(GRBM_SOFT_RESET, tmp);
1921 tmp = RREG32(GRBM_SOFT_RESET);
1922
1923 udelay(50);
1924
1925 tmp &= ~grbm_soft_reset;
1926 WREG32(GRBM_SOFT_RESET, tmp);
1927 tmp = RREG32(GRBM_SOFT_RESET);
1928 }
1929
1930 if (srbm_soft_reset) {
1931 tmp = RREG32(SRBM_SOFT_RESET);
1932 tmp |= srbm_soft_reset;
1933 dev_info(rdev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp);
1934 WREG32(SRBM_SOFT_RESET, tmp);
1935 tmp = RREG32(SRBM_SOFT_RESET);
1936
1937 udelay(50);
1938
1939 tmp &= ~srbm_soft_reset;
1940 WREG32(SRBM_SOFT_RESET, tmp);
1941 tmp = RREG32(SRBM_SOFT_RESET);
1942 }
1943
1944 /* Wait a little for things to settle down */
1945 udelay(50);
1946
1947 evergreen_mc_resume(rdev, &save);
1948 udelay(50);
1949
1950 evergreen_print_gpu_status_regs(rdev);
1951}
1952
1953int cayman_asic_reset(struct radeon_device *rdev, bool hard)
1954{
1955 u32 reset_mask;
1956
1957 if (hard) {
1958 evergreen_gpu_pci_config_reset(rdev);
1959 return 0;
1960 }
1961
1962 reset_mask = cayman_gpu_check_soft_reset(rdev);
1963
1964 if (reset_mask)
1965 r600_set_bios_scratch_engine_hung(rdev, true);
1966
1967 cayman_gpu_soft_reset(rdev, reset_mask);
1968
1969 reset_mask = cayman_gpu_check_soft_reset(rdev);
1970
1971 if (reset_mask)
1972 evergreen_gpu_pci_config_reset(rdev);
1973
1974 r600_set_bios_scratch_engine_hung(rdev, false);
1975
1976 return 0;
1977}
1978
1979/**
1980 * cayman_gfx_is_lockup - Check if the GFX engine is locked up
1981 *
1982 * @rdev: radeon_device pointer
1983 * @ring: radeon_ring structure holding ring information
1984 *
1985 * Check if the GFX engine is locked up.
1986 * Returns true if the engine appears to be locked up, false if not.
1987 */
1988bool cayman_gfx_is_lockup(struct radeon_device *rdev, struct radeon_ring *ring)
1989{
1990 u32 reset_mask = cayman_gpu_check_soft_reset(rdev);
1991
1992 if (!(reset_mask & (RADEON_RESET_GFX |
1993 RADEON_RESET_COMPUTE |
1994 RADEON_RESET_CP))) {
1995 radeon_ring_lockup_update(rdev, ring);
1996 return false;
1997 }
1998 return radeon_ring_test_lockup(rdev, ring);
1999}
2000
2001static void cayman_uvd_init(struct radeon_device *rdev)
2002{
2003 int r;
2004
2005 if (!rdev->has_uvd)
2006 return;
2007
2008 r = radeon_uvd_init(rdev);
2009 if (r) {
2010 dev_err(rdev->dev, "failed UVD (%d) init.\n", r);
2011 /*
2012 * At this point rdev->uvd.vcpu_bo is NULL which trickles down
2013 * to early fails uvd_v2_2_resume() and thus nothing happens
2014 * there. So it is pointless to try to go through that code
2015 * hence why we disable uvd here.
2016 */
2017 rdev->has_uvd = 0;
2018 return;
2019 }
2020 rdev->ring[R600_RING_TYPE_UVD_INDEX].ring_obj = NULL;
2021 r600_ring_init(rdev, &rdev->ring[R600_RING_TYPE_UVD_INDEX], 4096);
2022}
2023
2024static void cayman_uvd_start(struct radeon_device *rdev)
2025{
2026 int r;
2027
2028 if (!rdev->has_uvd)
2029 return;
2030
2031 r = uvd_v2_2_resume(rdev);
2032 if (r) {
2033 dev_err(rdev->dev, "failed UVD resume (%d).\n", r);
2034 goto error;
2035 }
2036 r = radeon_fence_driver_start_ring(rdev, R600_RING_TYPE_UVD_INDEX);
2037 if (r) {
2038 dev_err(rdev->dev, "failed initializing UVD fences (%d).\n", r);
2039 goto error;
2040 }
2041 return;
2042
2043error:
2044 rdev->ring[R600_RING_TYPE_UVD_INDEX].ring_size = 0;
2045}
2046
2047static void cayman_uvd_resume(struct radeon_device *rdev)
2048{
2049 struct radeon_ring *ring;
2050 int r;
2051
2052 if (!rdev->has_uvd || !rdev->ring[R600_RING_TYPE_UVD_INDEX].ring_size)
2053 return;
2054
2055 ring = &rdev->ring[R600_RING_TYPE_UVD_INDEX];
2056 r = radeon_ring_init(rdev, ring, ring->ring_size, 0, PACKET0(UVD_NO_OP, 0));
2057 if (r) {
2058 dev_err(rdev->dev, "failed initializing UVD ring (%d).\n", r);
2059 return;
2060 }
2061 r = uvd_v1_0_init(rdev);
2062 if (r) {
2063 dev_err(rdev->dev, "failed initializing UVD (%d).\n", r);
2064 return;
2065 }
2066}
2067
2068static void cayman_vce_init(struct radeon_device *rdev)
2069{
2070 int r;
2071
2072 /* Only set for CHIP_ARUBA */
2073 if (!rdev->has_vce)
2074 return;
2075
2076 r = radeon_vce_init(rdev);
2077 if (r) {
2078 dev_err(rdev->dev, "failed VCE (%d) init.\n", r);
2079 /*
2080 * At this point rdev->vce.vcpu_bo is NULL which trickles down
2081 * to early fails cayman_vce_start() and thus nothing happens
2082 * there. So it is pointless to try to go through that code
2083 * hence why we disable vce here.
2084 */
2085 rdev->has_vce = 0;
2086 return;
2087 }
2088 rdev->ring[TN_RING_TYPE_VCE1_INDEX].ring_obj = NULL;
2089 r600_ring_init(rdev, &rdev->ring[TN_RING_TYPE_VCE1_INDEX], 4096);
2090 rdev->ring[TN_RING_TYPE_VCE2_INDEX].ring_obj = NULL;
2091 r600_ring_init(rdev, &rdev->ring[TN_RING_TYPE_VCE2_INDEX], 4096);
2092}
2093
2094static void cayman_vce_start(struct radeon_device *rdev)
2095{
2096 int r;
2097
2098 if (!rdev->has_vce)
2099 return;
2100
2101 r = radeon_vce_resume(rdev);
2102 if (r) {
2103 dev_err(rdev->dev, "failed VCE resume (%d).\n", r);
2104 goto error;
2105 }
2106 r = vce_v1_0_resume(rdev);
2107 if (r) {
2108 dev_err(rdev->dev, "failed VCE resume (%d).\n", r);
2109 goto error;
2110 }
2111 r = radeon_fence_driver_start_ring(rdev, TN_RING_TYPE_VCE1_INDEX);
2112 if (r) {
2113 dev_err(rdev->dev, "failed initializing VCE1 fences (%d).\n", r);
2114 goto error;
2115 }
2116 r = radeon_fence_driver_start_ring(rdev, TN_RING_TYPE_VCE2_INDEX);
2117 if (r) {
2118 dev_err(rdev->dev, "failed initializing VCE2 fences (%d).\n", r);
2119 goto error;
2120 }
2121 return;
2122
2123error:
2124 rdev->ring[TN_RING_TYPE_VCE1_INDEX].ring_size = 0;
2125 rdev->ring[TN_RING_TYPE_VCE2_INDEX].ring_size = 0;
2126}
2127
2128static void cayman_vce_resume(struct radeon_device *rdev)
2129{
2130 struct radeon_ring *ring;
2131 int r;
2132
2133 if (!rdev->has_vce || !rdev->ring[TN_RING_TYPE_VCE1_INDEX].ring_size)
2134 return;
2135
2136 ring = &rdev->ring[TN_RING_TYPE_VCE1_INDEX];
2137 r = radeon_ring_init(rdev, ring, ring->ring_size, 0, 0x0);
2138 if (r) {
2139 dev_err(rdev->dev, "failed initializing VCE1 ring (%d).\n", r);
2140 return;
2141 }
2142 ring = &rdev->ring[TN_RING_TYPE_VCE2_INDEX];
2143 r = radeon_ring_init(rdev, ring, ring->ring_size, 0, 0x0);
2144 if (r) {
2145 dev_err(rdev->dev, "failed initializing VCE1 ring (%d).\n", r);
2146 return;
2147 }
2148 r = vce_v1_0_init(rdev);
2149 if (r) {
2150 dev_err(rdev->dev, "failed initializing VCE (%d).\n", r);
2151 return;
2152 }
2153}
2154
2155static int cayman_startup(struct radeon_device *rdev)
2156{
2157 struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
2158 int r;
2159
2160 /* enable pcie gen2 link */
2161 evergreen_pcie_gen2_enable(rdev);
2162 /* enable aspm */
2163 evergreen_program_aspm(rdev);
2164
2165 /* scratch needs to be initialized before MC */
2166 r = r600_vram_scratch_init(rdev);
2167 if (r)
2168 return r;
2169
2170 evergreen_mc_program(rdev);
2171
2172 if (!(rdev->flags & RADEON_IS_IGP) && !rdev->pm.dpm_enabled) {
2173 r = ni_mc_load_microcode(rdev);
2174 if (r) {
2175 DRM_ERROR("Failed to load MC firmware!\n");
2176 return r;
2177 }
2178 }
2179
2180 r = cayman_pcie_gart_enable(rdev);
2181 if (r)
2182 return r;
2183 cayman_gpu_init(rdev);
2184
2185 /* allocate rlc buffers */
2186 if (rdev->flags & RADEON_IS_IGP) {
2187 rdev->rlc.reg_list = tn_rlc_save_restore_register_list;
2188 rdev->rlc.reg_list_size =
2189 (u32)ARRAY_SIZE(tn_rlc_save_restore_register_list);
2190 rdev->rlc.cs_data = cayman_cs_data;
2191 r = sumo_rlc_init(rdev);
2192 if (r) {
2193 DRM_ERROR("Failed to init rlc BOs!\n");
2194 return r;
2195 }
2196 }
2197
2198 /* allocate wb buffer */
2199 r = radeon_wb_init(rdev);
2200 if (r)
2201 return r;
2202
2203 r = radeon_fence_driver_start_ring(rdev, RADEON_RING_TYPE_GFX_INDEX);
2204 if (r) {
2205 dev_err(rdev->dev, "failed initializing CP fences (%d).\n", r);
2206 return r;
2207 }
2208
2209 cayman_uvd_start(rdev);
2210 cayman_vce_start(rdev);
2211
2212 r = radeon_fence_driver_start_ring(rdev, CAYMAN_RING_TYPE_CP1_INDEX);
2213 if (r) {
2214 dev_err(rdev->dev, "failed initializing CP fences (%d).\n", r);
2215 return r;
2216 }
2217
2218 r = radeon_fence_driver_start_ring(rdev, CAYMAN_RING_TYPE_CP2_INDEX);
2219 if (r) {
2220 dev_err(rdev->dev, "failed initializing CP fences (%d).\n", r);
2221 return r;
2222 }
2223
2224 r = radeon_fence_driver_start_ring(rdev, R600_RING_TYPE_DMA_INDEX);
2225 if (r) {
2226 dev_err(rdev->dev, "failed initializing DMA fences (%d).\n", r);
2227 return r;
2228 }
2229
2230 r = radeon_fence_driver_start_ring(rdev, CAYMAN_RING_TYPE_DMA1_INDEX);
2231 if (r) {
2232 dev_err(rdev->dev, "failed initializing DMA fences (%d).\n", r);
2233 return r;
2234 }
2235
2236 /* Enable IRQ */
2237 if (!rdev->irq.installed) {
2238 r = radeon_irq_kms_init(rdev);
2239 if (r)
2240 return r;
2241 }
2242
2243 r = r600_irq_init(rdev);
2244 if (r) {
2245 DRM_ERROR("radeon: IH init failed (%d).\n", r);
2246 radeon_irq_kms_fini(rdev);
2247 return r;
2248 }
2249 evergreen_irq_set(rdev);
2250
2251 r = radeon_ring_init(rdev, ring, ring->ring_size, RADEON_WB_CP_RPTR_OFFSET,
2252 RADEON_CP_PACKET2);
2253 if (r)
2254 return r;
2255
2256 ring = &rdev->ring[R600_RING_TYPE_DMA_INDEX];
2257 r = radeon_ring_init(rdev, ring, ring->ring_size, R600_WB_DMA_RPTR_OFFSET,
2258 DMA_PACKET(DMA_PACKET_NOP, 0, 0, 0));
2259 if (r)
2260 return r;
2261
2262 ring = &rdev->ring[CAYMAN_RING_TYPE_DMA1_INDEX];
2263 r = radeon_ring_init(rdev, ring, ring->ring_size, CAYMAN_WB_DMA1_RPTR_OFFSET,
2264 DMA_PACKET(DMA_PACKET_NOP, 0, 0, 0));
2265 if (r)
2266 return r;
2267
2268 r = cayman_cp_load_microcode(rdev);
2269 if (r)
2270 return r;
2271 r = cayman_cp_resume(rdev);
2272 if (r)
2273 return r;
2274
2275 r = cayman_dma_resume(rdev);
2276 if (r)
2277 return r;
2278
2279 cayman_uvd_resume(rdev);
2280 cayman_vce_resume(rdev);
2281
2282 r = radeon_ib_pool_init(rdev);
2283 if (r) {
2284 dev_err(rdev->dev, "IB initialization failed (%d).\n", r);
2285 return r;
2286 }
2287
2288 r = radeon_vm_manager_init(rdev);
2289 if (r) {
2290 dev_err(rdev->dev, "vm manager initialization failed (%d).\n", r);
2291 return r;
2292 }
2293
2294 r = radeon_audio_init(rdev);
2295 if (r)
2296 return r;
2297
2298 return 0;
2299}
2300
2301int cayman_resume(struct radeon_device *rdev)
2302{
2303 int r;
2304
2305 /* Do not reset GPU before posting, on rv770 hw unlike on r500 hw,
2306 * posting will perform necessary task to bring back GPU into good
2307 * shape.
2308 */
2309 /* post card */
2310 atom_asic_init(rdev->mode_info.atom_context);
2311
2312 /* init golden registers */
2313 ni_init_golden_registers(rdev);
2314
2315 if (rdev->pm.pm_method == PM_METHOD_DPM)
2316 radeon_pm_resume(rdev);
2317
2318 rdev->accel_working = true;
2319 r = cayman_startup(rdev);
2320 if (r) {
2321 DRM_ERROR("cayman startup failed on resume\n");
2322 rdev->accel_working = false;
2323 return r;
2324 }
2325 return r;
2326}
2327
2328int cayman_suspend(struct radeon_device *rdev)
2329{
2330 radeon_pm_suspend(rdev);
2331 radeon_audio_fini(rdev);
2332 radeon_vm_manager_fini(rdev);
2333 cayman_cp_enable(rdev, false);
2334 cayman_dma_stop(rdev);
2335 if (rdev->has_uvd) {
2336 uvd_v1_0_fini(rdev);
2337 radeon_uvd_suspend(rdev);
2338 }
2339 evergreen_irq_suspend(rdev);
2340 radeon_wb_disable(rdev);
2341 cayman_pcie_gart_disable(rdev);
2342 return 0;
2343}
2344
2345/* Plan is to move initialization in that function and use
2346 * helper function so that radeon_device_init pretty much
2347 * do nothing more than calling asic specific function. This
2348 * should also allow to remove a bunch of callback function
2349 * like vram_info.
2350 */
2351int cayman_init(struct radeon_device *rdev)
2352{
2353 struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
2354 int r;
2355
2356 /* Read BIOS */
2357 if (!radeon_get_bios(rdev)) {
2358 if (ASIC_IS_AVIVO(rdev))
2359 return -EINVAL;
2360 }
2361 /* Must be an ATOMBIOS */
2362 if (!rdev->is_atom_bios) {
2363 dev_err(rdev->dev, "Expecting atombios for cayman GPU\n");
2364 return -EINVAL;
2365 }
2366 r = radeon_atombios_init(rdev);
2367 if (r)
2368 return r;
2369
2370 /* Post card if necessary */
2371 if (!radeon_card_posted(rdev)) {
2372 if (!rdev->bios) {
2373 dev_err(rdev->dev, "Card not posted and no BIOS - ignoring\n");
2374 return -EINVAL;
2375 }
2376 DRM_INFO("GPU not posted. posting now...\n");
2377 atom_asic_init(rdev->mode_info.atom_context);
2378 }
2379 /* init golden registers */
2380 ni_init_golden_registers(rdev);
2381 /* Initialize scratch registers */
2382 r600_scratch_init(rdev);
2383 /* Initialize surface registers */
2384 radeon_surface_init(rdev);
2385 /* Initialize clocks */
2386 radeon_get_clock_info(rdev->ddev);
2387 /* Fence driver */
2388 r = radeon_fence_driver_init(rdev);
2389 if (r)
2390 return r;
2391 /* initialize memory controller */
2392 r = evergreen_mc_init(rdev);
2393 if (r)
2394 return r;
2395 /* Memory manager */
2396 r = radeon_bo_init(rdev);
2397 if (r)
2398 return r;
2399
2400 if (rdev->flags & RADEON_IS_IGP) {
2401 if (!rdev->me_fw || !rdev->pfp_fw || !rdev->rlc_fw) {
2402 r = ni_init_microcode(rdev);
2403 if (r) {
2404 DRM_ERROR("Failed to load firmware!\n");
2405 return r;
2406 }
2407 }
2408 } else {
2409 if (!rdev->me_fw || !rdev->pfp_fw || !rdev->rlc_fw || !rdev->mc_fw) {
2410 r = ni_init_microcode(rdev);
2411 if (r) {
2412 DRM_ERROR("Failed to load firmware!\n");
2413 return r;
2414 }
2415 }
2416 }
2417
2418 /* Initialize power management */
2419 radeon_pm_init(rdev);
2420
2421 ring->ring_obj = NULL;
2422 r600_ring_init(rdev, ring, 1024 * 1024);
2423
2424 ring = &rdev->ring[R600_RING_TYPE_DMA_INDEX];
2425 ring->ring_obj = NULL;
2426 r600_ring_init(rdev, ring, 64 * 1024);
2427
2428 ring = &rdev->ring[CAYMAN_RING_TYPE_DMA1_INDEX];
2429 ring->ring_obj = NULL;
2430 r600_ring_init(rdev, ring, 64 * 1024);
2431
2432 cayman_uvd_init(rdev);
2433 cayman_vce_init(rdev);
2434
2435 rdev->ih.ring_obj = NULL;
2436 r600_ih_ring_init(rdev, 64 * 1024);
2437
2438 r = r600_pcie_gart_init(rdev);
2439 if (r)
2440 return r;
2441
2442 rdev->accel_working = true;
2443 r = cayman_startup(rdev);
2444 if (r) {
2445 dev_err(rdev->dev, "disabling GPU acceleration\n");
2446 cayman_cp_fini(rdev);
2447 cayman_dma_fini(rdev);
2448 r600_irq_fini(rdev);
2449 if (rdev->flags & RADEON_IS_IGP)
2450 sumo_rlc_fini(rdev);
2451 radeon_wb_fini(rdev);
2452 radeon_ib_pool_fini(rdev);
2453 radeon_vm_manager_fini(rdev);
2454 radeon_irq_kms_fini(rdev);
2455 cayman_pcie_gart_fini(rdev);
2456 rdev->accel_working = false;
2457 }
2458
2459 /* Don't start up if the MC ucode is missing.
2460 * The default clocks and voltages before the MC ucode
2461 * is loaded are not suffient for advanced operations.
2462 *
2463 * We can skip this check for TN, because there is no MC
2464 * ucode.
2465 */
2466 if (!rdev->mc_fw && !(rdev->flags & RADEON_IS_IGP)) {
2467 DRM_ERROR("radeon: MC ucode required for NI+.\n");
2468 return -EINVAL;
2469 }
2470
2471 return 0;
2472}
2473
2474void cayman_fini(struct radeon_device *rdev)
2475{
2476 radeon_pm_fini(rdev);
2477 cayman_cp_fini(rdev);
2478 cayman_dma_fini(rdev);
2479 r600_irq_fini(rdev);
2480 if (rdev->flags & RADEON_IS_IGP)
2481 sumo_rlc_fini(rdev);
2482 radeon_wb_fini(rdev);
2483 radeon_vm_manager_fini(rdev);
2484 radeon_ib_pool_fini(rdev);
2485 radeon_irq_kms_fini(rdev);
2486 uvd_v1_0_fini(rdev);
2487 radeon_uvd_fini(rdev);
2488 if (rdev->has_vce)
2489 radeon_vce_fini(rdev);
2490 cayman_pcie_gart_fini(rdev);
2491 r600_vram_scratch_fini(rdev);
2492 radeon_gem_fini(rdev);
2493 radeon_fence_driver_fini(rdev);
2494 radeon_bo_fini(rdev);
2495 radeon_atombios_fini(rdev);
2496 kfree(rdev->bios);
2497 rdev->bios = NULL;
2498}
2499
2500/*
2501 * vm
2502 */
2503int cayman_vm_init(struct radeon_device *rdev)
2504{
2505 /* number of VMs */
2506 rdev->vm_manager.nvm = 8;
2507 /* base offset of vram pages */
2508 if (rdev->flags & RADEON_IS_IGP) {
2509 u64 tmp = RREG32(FUS_MC_VM_FB_OFFSET);
2510 tmp <<= 22;
2511 rdev->vm_manager.vram_base_offset = tmp;
2512 } else
2513 rdev->vm_manager.vram_base_offset = 0;
2514 return 0;
2515}
2516
2517void cayman_vm_fini(struct radeon_device *rdev)
2518{
2519}
2520
2521/**
2522 * cayman_vm_decode_fault - print human readable fault info
2523 *
2524 * @rdev: radeon_device pointer
2525 * @status: VM_CONTEXT1_PROTECTION_FAULT_STATUS register value
2526 * @addr: VM_CONTEXT1_PROTECTION_FAULT_ADDR register value
2527 *
2528 * Print human readable fault information (cayman/TN).
2529 */
2530void cayman_vm_decode_fault(struct radeon_device *rdev,
2531 u32 status, u32 addr)
2532{
2533 u32 mc_id = (status & MEMORY_CLIENT_ID_MASK) >> MEMORY_CLIENT_ID_SHIFT;
2534 u32 vmid = (status & FAULT_VMID_MASK) >> FAULT_VMID_SHIFT;
2535 u32 protections = (status & PROTECTIONS_MASK) >> PROTECTIONS_SHIFT;
2536 char *block;
2537
2538 switch (mc_id) {
2539 case 32:
2540 case 16:
2541 case 96:
2542 case 80:
2543 case 160:
2544 case 144:
2545 case 224:
2546 case 208:
2547 block = "CB";
2548 break;
2549 case 33:
2550 case 17:
2551 case 97:
2552 case 81:
2553 case 161:
2554 case 145:
2555 case 225:
2556 case 209:
2557 block = "CB_FMASK";
2558 break;
2559 case 34:
2560 case 18:
2561 case 98:
2562 case 82:
2563 case 162:
2564 case 146:
2565 case 226:
2566 case 210:
2567 block = "CB_CMASK";
2568 break;
2569 case 35:
2570 case 19:
2571 case 99:
2572 case 83:
2573 case 163:
2574 case 147:
2575 case 227:
2576 case 211:
2577 block = "CB_IMMED";
2578 break;
2579 case 36:
2580 case 20:
2581 case 100:
2582 case 84:
2583 case 164:
2584 case 148:
2585 case 228:
2586 case 212:
2587 block = "DB";
2588 break;
2589 case 37:
2590 case 21:
2591 case 101:
2592 case 85:
2593 case 165:
2594 case 149:
2595 case 229:
2596 case 213:
2597 block = "DB_HTILE";
2598 break;
2599 case 38:
2600 case 22:
2601 case 102:
2602 case 86:
2603 case 166:
2604 case 150:
2605 case 230:
2606 case 214:
2607 block = "SX";
2608 break;
2609 case 39:
2610 case 23:
2611 case 103:
2612 case 87:
2613 case 167:
2614 case 151:
2615 case 231:
2616 case 215:
2617 block = "DB_STEN";
2618 break;
2619 case 40:
2620 case 24:
2621 case 104:
2622 case 88:
2623 case 232:
2624 case 216:
2625 case 168:
2626 case 152:
2627 block = "TC_TFETCH";
2628 break;
2629 case 41:
2630 case 25:
2631 case 105:
2632 case 89:
2633 case 233:
2634 case 217:
2635 case 169:
2636 case 153:
2637 block = "TC_VFETCH";
2638 break;
2639 case 42:
2640 case 26:
2641 case 106:
2642 case 90:
2643 case 234:
2644 case 218:
2645 case 170:
2646 case 154:
2647 block = "VC";
2648 break;
2649 case 112:
2650 block = "CP";
2651 break;
2652 case 113:
2653 case 114:
2654 block = "SH";
2655 break;
2656 case 115:
2657 block = "VGT";
2658 break;
2659 case 178:
2660 block = "IH";
2661 break;
2662 case 51:
2663 block = "RLC";
2664 break;
2665 case 55:
2666 block = "DMA";
2667 break;
2668 case 56:
2669 block = "HDP";
2670 break;
2671 default:
2672 block = "unknown";
2673 break;
2674 }
2675
2676 printk("VM fault (0x%02x, vmid %d) at page %u, %s from %s (%d)\n",
2677 protections, vmid, addr,
2678 (status & MEMORY_CLIENT_RW_MASK) ? "write" : "read",
2679 block, mc_id);
2680}
2681
2682/**
2683 * cayman_vm_flush - vm flush using the CP
2684 *
2685 * @rdev: radeon_device pointer
2686 *
2687 * Update the page table base and flush the VM TLB
2688 * using the CP (cayman-si).
2689 */
2690void cayman_vm_flush(struct radeon_device *rdev, struct radeon_ring *ring,
2691 unsigned vm_id, uint64_t pd_addr)
2692{
2693 radeon_ring_write(ring, PACKET0(VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + (vm_id << 2), 0));
2694 radeon_ring_write(ring, pd_addr >> 12);
2695
2696 /* flush hdp cache */
2697 radeon_ring_write(ring, PACKET0(HDP_MEM_COHERENCY_FLUSH_CNTL, 0));
2698 radeon_ring_write(ring, 0x1);
2699
2700 /* bits 0-7 are the VM contexts0-7 */
2701 radeon_ring_write(ring, PACKET0(VM_INVALIDATE_REQUEST, 0));
2702 radeon_ring_write(ring, 1 << vm_id);
2703
2704 /* wait for the invalidate to complete */
2705 radeon_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
2706 radeon_ring_write(ring, (WAIT_REG_MEM_FUNCTION(0) | /* always */
2707 WAIT_REG_MEM_ENGINE(0))); /* me */
2708 radeon_ring_write(ring, VM_INVALIDATE_REQUEST >> 2);
2709 radeon_ring_write(ring, 0);
2710 radeon_ring_write(ring, 0); /* ref */
2711 radeon_ring_write(ring, 0); /* mask */
2712 radeon_ring_write(ring, 0x20); /* poll interval */
2713
2714 /* sync PFP to ME, otherwise we might get invalid PFP reads */
2715 radeon_ring_write(ring, PACKET3(PACKET3_PFP_SYNC_ME, 0));
2716 radeon_ring_write(ring, 0x0);
2717}
2718
2719int tn_set_vce_clocks(struct radeon_device *rdev, u32 evclk, u32 ecclk)
2720{
2721 struct atom_clock_dividers dividers;
2722 int r, i;
2723
2724 r = radeon_atom_get_clock_dividers(rdev, COMPUTE_ENGINE_PLL_PARAM,
2725 ecclk, false, &dividers);
2726 if (r)
2727 return r;
2728
2729 for (i = 0; i < 100; i++) {
2730 if (RREG32(CG_ECLK_STATUS) & ECLK_STATUS)
2731 break;
2732 mdelay(10);
2733 }
2734 if (i == 100)
2735 return -ETIMEDOUT;
2736
2737 WREG32_P(CG_ECLK_CNTL, dividers.post_div, ~(ECLK_DIR_CNTL_EN|ECLK_DIVIDER_MASK));
2738
2739 for (i = 0; i < 100; i++) {
2740 if (RREG32(CG_ECLK_STATUS) & ECLK_STATUS)
2741 break;
2742 mdelay(10);
2743 }
2744 if (i == 100)
2745 return -ETIMEDOUT;
2746
2747 return 0;
2748}
2749