2 * Copyright 2010 Advanced Micro Devices, Inc.
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice shall be included in
12 * all copies or substantial portions of the Software.
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.
22 * Authors: Alex Deucher
25 #include <linux/firmware.h>
26 #include <linux/module.h>
27 #include <linux/pci.h>
28 #include <linux/slab.h>
30 #include <drm/radeon_drm.h>
33 #include "cayman_blit_shaders.h"
34 #include "clearstate_cayman.h"
35 #include "evergreen.h"
40 #include "radeon_asic.h"
41 #include "radeon_audio.h"
42 #include "radeon_ucode.h"
45 * Indirect registers accessor
47 u32 tn_smc_rreg(struct radeon_device *rdev, u32 reg)
52 spin_lock_irqsave(&rdev->smc_idx_lock, flags);
53 WREG32(TN_SMC_IND_INDEX_0, (reg));
54 r = RREG32(TN_SMC_IND_DATA_0);
55 spin_unlock_irqrestore(&rdev->smc_idx_lock, flags);
59 void tn_smc_wreg(struct radeon_device *rdev, u32 reg, u32 v)
63 spin_lock_irqsave(&rdev->smc_idx_lock, flags);
64 WREG32(TN_SMC_IND_INDEX_0, (reg));
65 WREG32(TN_SMC_IND_DATA_0, (v));
66 spin_unlock_irqrestore(&rdev->smc_idx_lock, flags);
69 static const u32 tn_rlc_save_restore_register_list[] =
199 static const u32 cayman_golden_registers2[] =
201 0x3e5c, 0xffffffff, 0x00000000,
202 0x3e48, 0xffffffff, 0x00000000,
203 0x3e4c, 0xffffffff, 0x00000000,
204 0x3e64, 0xffffffff, 0x00000000,
205 0x3e50, 0xffffffff, 0x00000000,
206 0x3e60, 0xffffffff, 0x00000000
209 static const u32 cayman_golden_registers[] =
211 0x5eb4, 0xffffffff, 0x00000002,
212 0x5e78, 0x8f311ff1, 0x001000f0,
213 0x3f90, 0xffff0000, 0xff000000,
214 0x9148, 0xffff0000, 0xff000000,
215 0x3f94, 0xffff0000, 0xff000000,
216 0x914c, 0xffff0000, 0xff000000,
217 0xc78, 0x00000080, 0x00000080,
218 0xbd4, 0x70073777, 0x00011003,
219 0xd02c, 0xbfffff1f, 0x08421000,
220 0xd0b8, 0x73773777, 0x02011003,
221 0x5bc0, 0x00200000, 0x50100000,
222 0x98f8, 0x33773777, 0x02011003,
223 0x98fc, 0xffffffff, 0x76541032,
224 0x7030, 0x31000311, 0x00000011,
225 0x2f48, 0x33773777, 0x42010001,
226 0x6b28, 0x00000010, 0x00000012,
227 0x7728, 0x00000010, 0x00000012,
228 0x10328, 0x00000010, 0x00000012,
229 0x10f28, 0x00000010, 0x00000012,
230 0x11b28, 0x00000010, 0x00000012,
231 0x12728, 0x00000010, 0x00000012,
232 0x240c, 0x000007ff, 0x00000000,
233 0x8a14, 0xf000001f, 0x00000007,
234 0x8b24, 0x3fff3fff, 0x00ff0fff,
235 0x8b10, 0x0000ff0f, 0x00000000,
236 0x28a4c, 0x07ffffff, 0x06000000,
237 0x10c, 0x00000001, 0x00010003,
238 0xa02c, 0xffffffff, 0x0000009b,
239 0x913c, 0x0000010f, 0x01000100,
240 0x8c04, 0xf8ff00ff, 0x40600060,
241 0x28350, 0x00000f01, 0x00000000,
242 0x9508, 0x3700001f, 0x00000002,
243 0x960c, 0xffffffff, 0x54763210,
244 0x88c4, 0x001f3ae3, 0x00000082,
245 0x88d0, 0xffffffff, 0x0f40df40,
246 0x88d4, 0x0000001f, 0x00000010,
247 0x8974, 0xffffffff, 0x00000000
250 static const u32 dvst_golden_registers2[] =
252 0x8f8, 0xffffffff, 0,
253 0x8fc, 0x00380000, 0,
254 0x8f8, 0xffffffff, 1,
258 static const u32 dvst_golden_registers[] =
260 0x690, 0x3fff3fff, 0x20c00033,
261 0x918c, 0x0fff0fff, 0x00010006,
262 0x91a8, 0x0fff0fff, 0x00010006,
263 0x9150, 0xffffdfff, 0x6e944040,
264 0x917c, 0x0fff0fff, 0x00030002,
265 0x9198, 0x0fff0fff, 0x00030002,
266 0x915c, 0x0fff0fff, 0x00010000,
267 0x3f90, 0xffff0001, 0xff000000,
268 0x9178, 0x0fff0fff, 0x00070000,
269 0x9194, 0x0fff0fff, 0x00070000,
270 0x9148, 0xffff0001, 0xff000000,
271 0x9190, 0x0fff0fff, 0x00090008,
272 0x91ac, 0x0fff0fff, 0x00090008,
273 0x3f94, 0xffff0000, 0xff000000,
274 0x914c, 0xffff0000, 0xff000000,
275 0x929c, 0x00000fff, 0x00000001,
276 0x55e4, 0xff607fff, 0xfc000100,
277 0x8a18, 0xff000fff, 0x00000100,
278 0x8b28, 0xff000fff, 0x00000100,
279 0x9144, 0xfffc0fff, 0x00000100,
280 0x6ed8, 0x00010101, 0x00010000,
281 0x9830, 0xffffffff, 0x00000000,
282 0x9834, 0xf00fffff, 0x00000400,
283 0x9838, 0xfffffffe, 0x00000000,
284 0xd0c0, 0xff000fff, 0x00000100,
285 0xd02c, 0xbfffff1f, 0x08421000,
286 0xd0b8, 0x73773777, 0x12010001,
287 0x5bb0, 0x000000f0, 0x00000070,
288 0x98f8, 0x73773777, 0x12010001,
289 0x98fc, 0xffffffff, 0x00000010,
290 0x9b7c, 0x00ff0000, 0x00fc0000,
291 0x8030, 0x00001f0f, 0x0000100a,
292 0x2f48, 0x73773777, 0x12010001,
293 0x2408, 0x00030000, 0x000c007f,
294 0x8a14, 0xf000003f, 0x00000007,
295 0x8b24, 0x3fff3fff, 0x00ff0fff,
296 0x8b10, 0x0000ff0f, 0x00000000,
297 0x28a4c, 0x07ffffff, 0x06000000,
298 0x4d8, 0x00000fff, 0x00000100,
299 0xa008, 0xffffffff, 0x00010000,
300 0x913c, 0xffff03ff, 0x01000100,
301 0x8c00, 0x000000ff, 0x00000003,
302 0x8c04, 0xf8ff00ff, 0x40600060,
303 0x8cf0, 0x1fff1fff, 0x08e00410,
304 0x28350, 0x00000f01, 0x00000000,
305 0x9508, 0xf700071f, 0x00000002,
306 0x960c, 0xffffffff, 0x54763210,
307 0x20ef8, 0x01ff01ff, 0x00000002,
308 0x20e98, 0xfffffbff, 0x00200000,
309 0x2015c, 0xffffffff, 0x00000f40,
310 0x88c4, 0x001f3ae3, 0x00000082,
311 0x8978, 0x3fffffff, 0x04050140,
312 0x88d4, 0x0000001f, 0x00000010,
313 0x8974, 0xffffffff, 0x00000000
316 static const u32 scrapper_golden_registers[] =
318 0x690, 0x3fff3fff, 0x20c00033,
319 0x918c, 0x0fff0fff, 0x00010006,
320 0x918c, 0x0fff0fff, 0x00010006,
321 0x91a8, 0x0fff0fff, 0x00010006,
322 0x91a8, 0x0fff0fff, 0x00010006,
323 0x9150, 0xffffdfff, 0x6e944040,
324 0x9150, 0xffffdfff, 0x6e944040,
325 0x917c, 0x0fff0fff, 0x00030002,
326 0x917c, 0x0fff0fff, 0x00030002,
327 0x9198, 0x0fff0fff, 0x00030002,
328 0x9198, 0x0fff0fff, 0x00030002,
329 0x915c, 0x0fff0fff, 0x00010000,
330 0x915c, 0x0fff0fff, 0x00010000,
331 0x3f90, 0xffff0001, 0xff000000,
332 0x3f90, 0xffff0001, 0xff000000,
333 0x9178, 0x0fff0fff, 0x00070000,
334 0x9178, 0x0fff0fff, 0x00070000,
335 0x9194, 0x0fff0fff, 0x00070000,
336 0x9194, 0x0fff0fff, 0x00070000,
337 0x9148, 0xffff0001, 0xff000000,
338 0x9148, 0xffff0001, 0xff000000,
339 0x9190, 0x0fff0fff, 0x00090008,
340 0x9190, 0x0fff0fff, 0x00090008,
341 0x91ac, 0x0fff0fff, 0x00090008,
342 0x91ac, 0x0fff0fff, 0x00090008,
343 0x3f94, 0xffff0000, 0xff000000,
344 0x3f94, 0xffff0000, 0xff000000,
345 0x914c, 0xffff0000, 0xff000000,
346 0x914c, 0xffff0000, 0xff000000,
347 0x929c, 0x00000fff, 0x00000001,
348 0x929c, 0x00000fff, 0x00000001,
349 0x55e4, 0xff607fff, 0xfc000100,
350 0x8a18, 0xff000fff, 0x00000100,
351 0x8a18, 0xff000fff, 0x00000100,
352 0x8b28, 0xff000fff, 0x00000100,
353 0x8b28, 0xff000fff, 0x00000100,
354 0x9144, 0xfffc0fff, 0x00000100,
355 0x9144, 0xfffc0fff, 0x00000100,
356 0x6ed8, 0x00010101, 0x00010000,
357 0x9830, 0xffffffff, 0x00000000,
358 0x9830, 0xffffffff, 0x00000000,
359 0x9834, 0xf00fffff, 0x00000400,
360 0x9834, 0xf00fffff, 0x00000400,
361 0x9838, 0xfffffffe, 0x00000000,
362 0x9838, 0xfffffffe, 0x00000000,
363 0xd0c0, 0xff000fff, 0x00000100,
364 0xd02c, 0xbfffff1f, 0x08421000,
365 0xd02c, 0xbfffff1f, 0x08421000,
366 0xd0b8, 0x73773777, 0x12010001,
367 0xd0b8, 0x73773777, 0x12010001,
368 0x5bb0, 0x000000f0, 0x00000070,
369 0x98f8, 0x73773777, 0x12010001,
370 0x98f8, 0x73773777, 0x12010001,
371 0x98fc, 0xffffffff, 0x00000010,
372 0x98fc, 0xffffffff, 0x00000010,
373 0x9b7c, 0x00ff0000, 0x00fc0000,
374 0x9b7c, 0x00ff0000, 0x00fc0000,
375 0x8030, 0x00001f0f, 0x0000100a,
376 0x8030, 0x00001f0f, 0x0000100a,
377 0x2f48, 0x73773777, 0x12010001,
378 0x2f48, 0x73773777, 0x12010001,
379 0x2408, 0x00030000, 0x000c007f,
380 0x8a14, 0xf000003f, 0x00000007,
381 0x8a14, 0xf000003f, 0x00000007,
382 0x8b24, 0x3fff3fff, 0x00ff0fff,
383 0x8b24, 0x3fff3fff, 0x00ff0fff,
384 0x8b10, 0x0000ff0f, 0x00000000,
385 0x8b10, 0x0000ff0f, 0x00000000,
386 0x28a4c, 0x07ffffff, 0x06000000,
387 0x28a4c, 0x07ffffff, 0x06000000,
388 0x4d8, 0x00000fff, 0x00000100,
389 0x4d8, 0x00000fff, 0x00000100,
390 0xa008, 0xffffffff, 0x00010000,
391 0xa008, 0xffffffff, 0x00010000,
392 0x913c, 0xffff03ff, 0x01000100,
393 0x913c, 0xffff03ff, 0x01000100,
394 0x90e8, 0x001fffff, 0x010400c0,
395 0x8c00, 0x000000ff, 0x00000003,
396 0x8c00, 0x000000ff, 0x00000003,
397 0x8c04, 0xf8ff00ff, 0x40600060,
398 0x8c04, 0xf8ff00ff, 0x40600060,
399 0x8c30, 0x0000000f, 0x00040005,
400 0x8cf0, 0x1fff1fff, 0x08e00410,
401 0x8cf0, 0x1fff1fff, 0x08e00410,
402 0x900c, 0x00ffffff, 0x0017071f,
403 0x28350, 0x00000f01, 0x00000000,
404 0x28350, 0x00000f01, 0x00000000,
405 0x9508, 0xf700071f, 0x00000002,
406 0x9508, 0xf700071f, 0x00000002,
407 0x9688, 0x00300000, 0x0017000f,
408 0x960c, 0xffffffff, 0x54763210,
409 0x960c, 0xffffffff, 0x54763210,
410 0x20ef8, 0x01ff01ff, 0x00000002,
411 0x20e98, 0xfffffbff, 0x00200000,
412 0x2015c, 0xffffffff, 0x00000f40,
413 0x88c4, 0x001f3ae3, 0x00000082,
414 0x88c4, 0x001f3ae3, 0x00000082,
415 0x8978, 0x3fffffff, 0x04050140,
416 0x8978, 0x3fffffff, 0x04050140,
417 0x88d4, 0x0000001f, 0x00000010,
418 0x88d4, 0x0000001f, 0x00000010,
419 0x8974, 0xffffffff, 0x00000000,
420 0x8974, 0xffffffff, 0x00000000
423 static void ni_init_golden_registers(struct radeon_device *rdev)
425 switch (rdev->family) {
427 radeon_program_register_sequence(rdev,
428 cayman_golden_registers,
429 (const u32)ARRAY_SIZE(cayman_golden_registers));
430 radeon_program_register_sequence(rdev,
431 cayman_golden_registers2,
432 (const u32)ARRAY_SIZE(cayman_golden_registers2));
435 if ((rdev->pdev->device == 0x9900) ||
436 (rdev->pdev->device == 0x9901) ||
437 (rdev->pdev->device == 0x9903) ||
438 (rdev->pdev->device == 0x9904) ||
439 (rdev->pdev->device == 0x9905) ||
440 (rdev->pdev->device == 0x9906) ||
441 (rdev->pdev->device == 0x9907) ||
442 (rdev->pdev->device == 0x9908) ||
443 (rdev->pdev->device == 0x9909) ||
444 (rdev->pdev->device == 0x990A) ||
445 (rdev->pdev->device == 0x990B) ||
446 (rdev->pdev->device == 0x990C) ||
447 (rdev->pdev->device == 0x990D) ||
448 (rdev->pdev->device == 0x990E) ||
449 (rdev->pdev->device == 0x990F) ||
450 (rdev->pdev->device == 0x9910) ||
451 (rdev->pdev->device == 0x9913) ||
452 (rdev->pdev->device == 0x9917) ||
453 (rdev->pdev->device == 0x9918)) {
454 radeon_program_register_sequence(rdev,
455 dvst_golden_registers,
456 (const u32)ARRAY_SIZE(dvst_golden_registers));
457 radeon_program_register_sequence(rdev,
458 dvst_golden_registers2,
459 (const u32)ARRAY_SIZE(dvst_golden_registers2));
461 radeon_program_register_sequence(rdev,
462 scrapper_golden_registers,
463 (const u32)ARRAY_SIZE(scrapper_golden_registers));
464 radeon_program_register_sequence(rdev,
465 dvst_golden_registers2,
466 (const u32)ARRAY_SIZE(dvst_golden_registers2));
474 #define BTC_IO_MC_REGS_SIZE 29
476 static const u32 barts_io_mc_regs[BTC_IO_MC_REGS_SIZE][2] = {
477 {0x00000077, 0xff010100},
478 {0x00000078, 0x00000000},
479 {0x00000079, 0x00001434},
480 {0x0000007a, 0xcc08ec08},
481 {0x0000007b, 0x00040000},
482 {0x0000007c, 0x000080c0},
483 {0x0000007d, 0x09000000},
484 {0x0000007e, 0x00210404},
485 {0x00000081, 0x08a8e800},
486 {0x00000082, 0x00030444},
487 {0x00000083, 0x00000000},
488 {0x00000085, 0x00000001},
489 {0x00000086, 0x00000002},
490 {0x00000087, 0x48490000},
491 {0x00000088, 0x20244647},
492 {0x00000089, 0x00000005},
493 {0x0000008b, 0x66030000},
494 {0x0000008c, 0x00006603},
495 {0x0000008d, 0x00000100},
496 {0x0000008f, 0x00001c0a},
497 {0x00000090, 0xff000001},
498 {0x00000094, 0x00101101},
499 {0x00000095, 0x00000fff},
500 {0x00000096, 0x00116fff},
501 {0x00000097, 0x60010000},
502 {0x00000098, 0x10010000},
503 {0x00000099, 0x00006000},
504 {0x0000009a, 0x00001000},
505 {0x0000009f, 0x00946a00}
508 static const u32 turks_io_mc_regs[BTC_IO_MC_REGS_SIZE][2] = {
509 {0x00000077, 0xff010100},
510 {0x00000078, 0x00000000},
511 {0x00000079, 0x00001434},
512 {0x0000007a, 0xcc08ec08},
513 {0x0000007b, 0x00040000},
514 {0x0000007c, 0x000080c0},
515 {0x0000007d, 0x09000000},
516 {0x0000007e, 0x00210404},
517 {0x00000081, 0x08a8e800},
518 {0x00000082, 0x00030444},
519 {0x00000083, 0x00000000},
520 {0x00000085, 0x00000001},
521 {0x00000086, 0x00000002},
522 {0x00000087, 0x48490000},
523 {0x00000088, 0x20244647},
524 {0x00000089, 0x00000005},
525 {0x0000008b, 0x66030000},
526 {0x0000008c, 0x00006603},
527 {0x0000008d, 0x00000100},
528 {0x0000008f, 0x00001c0a},
529 {0x00000090, 0xff000001},
530 {0x00000094, 0x00101101},
531 {0x00000095, 0x00000fff},
532 {0x00000096, 0x00116fff},
533 {0x00000097, 0x60010000},
534 {0x00000098, 0x10010000},
535 {0x00000099, 0x00006000},
536 {0x0000009a, 0x00001000},
537 {0x0000009f, 0x00936a00}
540 static const u32 caicos_io_mc_regs[BTC_IO_MC_REGS_SIZE][2] = {
541 {0x00000077, 0xff010100},
542 {0x00000078, 0x00000000},
543 {0x00000079, 0x00001434},
544 {0x0000007a, 0xcc08ec08},
545 {0x0000007b, 0x00040000},
546 {0x0000007c, 0x000080c0},
547 {0x0000007d, 0x09000000},
548 {0x0000007e, 0x00210404},
549 {0x00000081, 0x08a8e800},
550 {0x00000082, 0x00030444},
551 {0x00000083, 0x00000000},
552 {0x00000085, 0x00000001},
553 {0x00000086, 0x00000002},
554 {0x00000087, 0x48490000},
555 {0x00000088, 0x20244647},
556 {0x00000089, 0x00000005},
557 {0x0000008b, 0x66030000},
558 {0x0000008c, 0x00006603},
559 {0x0000008d, 0x00000100},
560 {0x0000008f, 0x00001c0a},
561 {0x00000090, 0xff000001},
562 {0x00000094, 0x00101101},
563 {0x00000095, 0x00000fff},
564 {0x00000096, 0x00116fff},
565 {0x00000097, 0x60010000},
566 {0x00000098, 0x10010000},
567 {0x00000099, 0x00006000},
568 {0x0000009a, 0x00001000},
569 {0x0000009f, 0x00916a00}
572 static const u32 cayman_io_mc_regs[BTC_IO_MC_REGS_SIZE][2] = {
573 {0x00000077, 0xff010100},
574 {0x00000078, 0x00000000},
575 {0x00000079, 0x00001434},
576 {0x0000007a, 0xcc08ec08},
577 {0x0000007b, 0x00040000},
578 {0x0000007c, 0x000080c0},
579 {0x0000007d, 0x09000000},
580 {0x0000007e, 0x00210404},
581 {0x00000081, 0x08a8e800},
582 {0x00000082, 0x00030444},
583 {0x00000083, 0x00000000},
584 {0x00000085, 0x00000001},
585 {0x00000086, 0x00000002},
586 {0x00000087, 0x48490000},
587 {0x00000088, 0x20244647},
588 {0x00000089, 0x00000005},
589 {0x0000008b, 0x66030000},
590 {0x0000008c, 0x00006603},
591 {0x0000008d, 0x00000100},
592 {0x0000008f, 0x00001c0a},
593 {0x00000090, 0xff000001},
594 {0x00000094, 0x00101101},
595 {0x00000095, 0x00000fff},
596 {0x00000096, 0x00116fff},
597 {0x00000097, 0x60010000},
598 {0x00000098, 0x10010000},
599 {0x00000099, 0x00006000},
600 {0x0000009a, 0x00001000},
601 {0x0000009f, 0x00976b00}
604 int ni_mc_load_microcode(struct radeon_device *rdev)
606 const __be32 *fw_data;
607 u32 mem_type, running, blackout = 0;
609 int i, ucode_size, regs_size;
614 switch (rdev->family) {
616 io_mc_regs = (u32 *)&barts_io_mc_regs;
617 ucode_size = BTC_MC_UCODE_SIZE;
618 regs_size = BTC_IO_MC_REGS_SIZE;
621 io_mc_regs = (u32 *)&turks_io_mc_regs;
622 ucode_size = BTC_MC_UCODE_SIZE;
623 regs_size = BTC_IO_MC_REGS_SIZE;
627 io_mc_regs = (u32 *)&caicos_io_mc_regs;
628 ucode_size = BTC_MC_UCODE_SIZE;
629 regs_size = BTC_IO_MC_REGS_SIZE;
632 io_mc_regs = (u32 *)&cayman_io_mc_regs;
633 ucode_size = CAYMAN_MC_UCODE_SIZE;
634 regs_size = BTC_IO_MC_REGS_SIZE;
638 mem_type = (RREG32(MC_SEQ_MISC0) & MC_SEQ_MISC0_GDDR5_MASK) >> MC_SEQ_MISC0_GDDR5_SHIFT;
639 running = RREG32(MC_SEQ_SUP_CNTL) & RUN_MASK;
641 if ((mem_type == MC_SEQ_MISC0_GDDR5_VALUE) && (running == 0)) {
643 blackout = RREG32(MC_SHARED_BLACKOUT_CNTL);
644 WREG32(MC_SHARED_BLACKOUT_CNTL, 1);
647 /* reset the engine and set to writable */
648 WREG32(MC_SEQ_SUP_CNTL, 0x00000008);
649 WREG32(MC_SEQ_SUP_CNTL, 0x00000010);
651 /* load mc io regs */
652 for (i = 0; i < regs_size; i++) {
653 WREG32(MC_SEQ_IO_DEBUG_INDEX, io_mc_regs[(i << 1)]);
654 WREG32(MC_SEQ_IO_DEBUG_DATA, io_mc_regs[(i << 1) + 1]);
656 /* load the MC ucode */
657 fw_data = (const __be32 *)rdev->mc_fw->data;
658 for (i = 0; i < ucode_size; i++)
659 WREG32(MC_SEQ_SUP_PGM, be32_to_cpup(fw_data++));
661 /* put the engine back into the active state */
662 WREG32(MC_SEQ_SUP_CNTL, 0x00000008);
663 WREG32(MC_SEQ_SUP_CNTL, 0x00000004);
664 WREG32(MC_SEQ_SUP_CNTL, 0x00000001);
666 /* wait for training to complete */
667 for (i = 0; i < rdev->usec_timeout; i++) {
668 if (RREG32(MC_IO_PAD_CNTL_D0) & MEM_FALL_OUT_CMD)
674 WREG32(MC_SHARED_BLACKOUT_CNTL, blackout);
680 int ni_init_microcode(struct radeon_device *rdev)
682 const char *chip_name;
683 const char *rlc_chip_name;
684 size_t pfp_req_size, me_req_size, rlc_req_size, mc_req_size;
685 size_t smc_req_size = 0;
691 switch (rdev->family) {
694 rlc_chip_name = "BTC";
695 pfp_req_size = EVERGREEN_PFP_UCODE_SIZE * 4;
696 me_req_size = EVERGREEN_PM4_UCODE_SIZE * 4;
697 rlc_req_size = EVERGREEN_RLC_UCODE_SIZE * 4;
698 mc_req_size = BTC_MC_UCODE_SIZE * 4;
699 smc_req_size = ALIGN(BARTS_SMC_UCODE_SIZE, 4);
703 rlc_chip_name = "BTC";
704 pfp_req_size = EVERGREEN_PFP_UCODE_SIZE * 4;
705 me_req_size = EVERGREEN_PM4_UCODE_SIZE * 4;
706 rlc_req_size = EVERGREEN_RLC_UCODE_SIZE * 4;
707 mc_req_size = BTC_MC_UCODE_SIZE * 4;
708 smc_req_size = ALIGN(TURKS_SMC_UCODE_SIZE, 4);
711 chip_name = "CAICOS";
712 rlc_chip_name = "BTC";
713 pfp_req_size = EVERGREEN_PFP_UCODE_SIZE * 4;
714 me_req_size = EVERGREEN_PM4_UCODE_SIZE * 4;
715 rlc_req_size = EVERGREEN_RLC_UCODE_SIZE * 4;
716 mc_req_size = BTC_MC_UCODE_SIZE * 4;
717 smc_req_size = ALIGN(CAICOS_SMC_UCODE_SIZE, 4);
720 chip_name = "CAYMAN";
721 rlc_chip_name = "CAYMAN";
722 pfp_req_size = CAYMAN_PFP_UCODE_SIZE * 4;
723 me_req_size = CAYMAN_PM4_UCODE_SIZE * 4;
724 rlc_req_size = CAYMAN_RLC_UCODE_SIZE * 4;
725 mc_req_size = CAYMAN_MC_UCODE_SIZE * 4;
726 smc_req_size = ALIGN(CAYMAN_SMC_UCODE_SIZE, 4);
730 rlc_chip_name = "ARUBA";
731 /* pfp/me same size as CAYMAN */
732 pfp_req_size = CAYMAN_PFP_UCODE_SIZE * 4;
733 me_req_size = CAYMAN_PM4_UCODE_SIZE * 4;
734 rlc_req_size = ARUBA_RLC_UCODE_SIZE * 4;
740 DRM_INFO("Loading %s Microcode\n", chip_name);
742 snprintf(fw_name, sizeof(fw_name), "/*(DEBLOBBED)*/", chip_name);
743 err = reject_firmware(&rdev->pfp_fw, fw_name, rdev->dev);
746 if (rdev->pfp_fw->size != pfp_req_size) {
747 pr_err("ni_cp: Bogus length %zu in firmware \"%s\"\n",
748 rdev->pfp_fw->size, fw_name);
753 snprintf(fw_name, sizeof(fw_name), "/*(DEBLOBBED)*/", chip_name);
754 err = reject_firmware(&rdev->me_fw, fw_name, rdev->dev);
757 if (rdev->me_fw->size != me_req_size) {
758 pr_err("ni_cp: Bogus length %zu in firmware \"%s\"\n",
759 rdev->me_fw->size, fw_name);
763 snprintf(fw_name, sizeof(fw_name), "/*(DEBLOBBED)*/", rlc_chip_name);
764 err = reject_firmware(&rdev->rlc_fw, fw_name, rdev->dev);
767 if (rdev->rlc_fw->size != rlc_req_size) {
768 pr_err("ni_rlc: Bogus length %zu in firmware \"%s\"\n",
769 rdev->rlc_fw->size, fw_name);
773 /* no MC ucode on TN */
774 if (!(rdev->flags & RADEON_IS_IGP)) {
775 snprintf(fw_name, sizeof(fw_name), "/*(DEBLOBBED)*/", chip_name);
776 err = reject_firmware(&rdev->mc_fw, fw_name, rdev->dev);
779 if (rdev->mc_fw->size != mc_req_size) {
780 pr_err("ni_mc: Bogus length %zu in firmware \"%s\"\n",
781 rdev->mc_fw->size, fw_name);
786 if ((rdev->family >= CHIP_BARTS) && (rdev->family <= CHIP_CAYMAN)) {
787 snprintf(fw_name, sizeof(fw_name), "/*(DEBLOBBED)*/", chip_name);
788 err = reject_firmware(&rdev->smc_fw, fw_name, rdev->dev);
790 pr_err("smc: error loading firmware \"%s\"\n", fw_name);
791 release_firmware(rdev->smc_fw);
794 } else if (rdev->smc_fw->size != smc_req_size) {
795 pr_err("ni_mc: Bogus length %zu in firmware \"%s\"\n",
796 rdev->mc_fw->size, fw_name);
804 pr_err("ni_cp: Failed to load firmware \"%s\"\n",
806 release_firmware(rdev->pfp_fw);
808 release_firmware(rdev->me_fw);
810 release_firmware(rdev->rlc_fw);
812 release_firmware(rdev->mc_fw);
819 * cayman_get_allowed_info_register - fetch the register for the info ioctl
821 * @rdev: radeon_device pointer
822 * @reg: register offset in bytes
823 * @val: register value
825 * Returns 0 for success or -EINVAL for an invalid register
828 int cayman_get_allowed_info_register(struct radeon_device *rdev,
833 case GRBM_STATUS_SE0:
834 case GRBM_STATUS_SE1:
837 case (DMA_STATUS_REG + DMA0_REGISTER_OFFSET):
838 case (DMA_STATUS_REG + DMA1_REGISTER_OFFSET):
847 int tn_get_temp(struct radeon_device *rdev)
849 u32 temp = RREG32_SMC(TN_CURRENT_GNB_TEMP) & 0x7ff;
850 int actual_temp = (temp / 8) - 49;
852 return actual_temp * 1000;
858 static void cayman_gpu_init(struct radeon_device *rdev)
860 u32 gb_addr_config = 0;
862 u32 cgts_tcc_disable;
865 u32 cgts_sm_ctrl_reg;
866 u32 hdp_host_path_cntl;
868 u32 disabled_rb_mask;
871 switch (rdev->family) {
873 rdev->config.cayman.max_shader_engines = 2;
874 rdev->config.cayman.max_pipes_per_simd = 4;
875 rdev->config.cayman.max_tile_pipes = 8;
876 rdev->config.cayman.max_simds_per_se = 12;
877 rdev->config.cayman.max_backends_per_se = 4;
878 rdev->config.cayman.max_texture_channel_caches = 8;
879 rdev->config.cayman.max_gprs = 256;
880 rdev->config.cayman.max_threads = 256;
881 rdev->config.cayman.max_gs_threads = 32;
882 rdev->config.cayman.max_stack_entries = 512;
883 rdev->config.cayman.sx_num_of_sets = 8;
884 rdev->config.cayman.sx_max_export_size = 256;
885 rdev->config.cayman.sx_max_export_pos_size = 64;
886 rdev->config.cayman.sx_max_export_smx_size = 192;
887 rdev->config.cayman.max_hw_contexts = 8;
888 rdev->config.cayman.sq_num_cf_insts = 2;
890 rdev->config.cayman.sc_prim_fifo_size = 0x100;
891 rdev->config.cayman.sc_hiz_tile_fifo_size = 0x30;
892 rdev->config.cayman.sc_earlyz_tile_fifo_size = 0x130;
893 gb_addr_config = CAYMAN_GB_ADDR_CONFIG_GOLDEN;
897 rdev->config.cayman.max_shader_engines = 1;
898 rdev->config.cayman.max_pipes_per_simd = 4;
899 rdev->config.cayman.max_tile_pipes = 2;
900 if ((rdev->pdev->device == 0x9900) ||
901 (rdev->pdev->device == 0x9901) ||
902 (rdev->pdev->device == 0x9905) ||
903 (rdev->pdev->device == 0x9906) ||
904 (rdev->pdev->device == 0x9907) ||
905 (rdev->pdev->device == 0x9908) ||
906 (rdev->pdev->device == 0x9909) ||
907 (rdev->pdev->device == 0x990B) ||
908 (rdev->pdev->device == 0x990C) ||
909 (rdev->pdev->device == 0x990F) ||
910 (rdev->pdev->device == 0x9910) ||
911 (rdev->pdev->device == 0x9917) ||
912 (rdev->pdev->device == 0x9999) ||
913 (rdev->pdev->device == 0x999C)) {
914 rdev->config.cayman.max_simds_per_se = 6;
915 rdev->config.cayman.max_backends_per_se = 2;
916 rdev->config.cayman.max_hw_contexts = 8;
917 rdev->config.cayman.sx_max_export_size = 256;
918 rdev->config.cayman.sx_max_export_pos_size = 64;
919 rdev->config.cayman.sx_max_export_smx_size = 192;
920 } else if ((rdev->pdev->device == 0x9903) ||
921 (rdev->pdev->device == 0x9904) ||
922 (rdev->pdev->device == 0x990A) ||
923 (rdev->pdev->device == 0x990D) ||
924 (rdev->pdev->device == 0x990E) ||
925 (rdev->pdev->device == 0x9913) ||
926 (rdev->pdev->device == 0x9918) ||
927 (rdev->pdev->device == 0x999D)) {
928 rdev->config.cayman.max_simds_per_se = 4;
929 rdev->config.cayman.max_backends_per_se = 2;
930 rdev->config.cayman.max_hw_contexts = 8;
931 rdev->config.cayman.sx_max_export_size = 256;
932 rdev->config.cayman.sx_max_export_pos_size = 64;
933 rdev->config.cayman.sx_max_export_smx_size = 192;
934 } else if ((rdev->pdev->device == 0x9919) ||
935 (rdev->pdev->device == 0x9990) ||
936 (rdev->pdev->device == 0x9991) ||
937 (rdev->pdev->device == 0x9994) ||
938 (rdev->pdev->device == 0x9995) ||
939 (rdev->pdev->device == 0x9996) ||
940 (rdev->pdev->device == 0x999A) ||
941 (rdev->pdev->device == 0x99A0)) {
942 rdev->config.cayman.max_simds_per_se = 3;
943 rdev->config.cayman.max_backends_per_se = 1;
944 rdev->config.cayman.max_hw_contexts = 4;
945 rdev->config.cayman.sx_max_export_size = 128;
946 rdev->config.cayman.sx_max_export_pos_size = 32;
947 rdev->config.cayman.sx_max_export_smx_size = 96;
949 rdev->config.cayman.max_simds_per_se = 2;
950 rdev->config.cayman.max_backends_per_se = 1;
951 rdev->config.cayman.max_hw_contexts = 4;
952 rdev->config.cayman.sx_max_export_size = 128;
953 rdev->config.cayman.sx_max_export_pos_size = 32;
954 rdev->config.cayman.sx_max_export_smx_size = 96;
956 rdev->config.cayman.max_texture_channel_caches = 2;
957 rdev->config.cayman.max_gprs = 256;
958 rdev->config.cayman.max_threads = 256;
959 rdev->config.cayman.max_gs_threads = 32;
960 rdev->config.cayman.max_stack_entries = 512;
961 rdev->config.cayman.sx_num_of_sets = 8;
962 rdev->config.cayman.sq_num_cf_insts = 2;
964 rdev->config.cayman.sc_prim_fifo_size = 0x40;
965 rdev->config.cayman.sc_hiz_tile_fifo_size = 0x30;
966 rdev->config.cayman.sc_earlyz_tile_fifo_size = 0x130;
967 gb_addr_config = ARUBA_GB_ADDR_CONFIG_GOLDEN;
972 for (i = 0, j = 0; i < 32; i++, j += 0x18) {
973 WREG32((0x2c14 + j), 0x00000000);
974 WREG32((0x2c18 + j), 0x00000000);
975 WREG32((0x2c1c + j), 0x00000000);
976 WREG32((0x2c20 + j), 0x00000000);
977 WREG32((0x2c24 + j), 0x00000000);
980 WREG32(GRBM_CNTL, GRBM_READ_TIMEOUT(0xff));
981 WREG32(SRBM_INT_CNTL, 0x1);
982 WREG32(SRBM_INT_ACK, 0x1);
984 evergreen_fix_pci_max_read_req_size(rdev);
986 RREG32(MC_SHARED_CHMAP);
987 mc_arb_ramcfg = RREG32(MC_ARB_RAMCFG);
989 tmp = (mc_arb_ramcfg & NOOFCOLS_MASK) >> NOOFCOLS_SHIFT;
990 rdev->config.cayman.mem_row_size_in_kb = (4 * (1 << (8 + tmp))) / 1024;
991 if (rdev->config.cayman.mem_row_size_in_kb > 4)
992 rdev->config.cayman.mem_row_size_in_kb = 4;
993 /* XXX use MC settings? */
994 rdev->config.cayman.shader_engine_tile_size = 32;
995 rdev->config.cayman.num_gpus = 1;
996 rdev->config.cayman.multi_gpu_tile_size = 64;
998 tmp = (gb_addr_config & NUM_PIPES_MASK) >> NUM_PIPES_SHIFT;
999 rdev->config.cayman.num_tile_pipes = (1 << tmp);
1000 tmp = (gb_addr_config & PIPE_INTERLEAVE_SIZE_MASK) >> PIPE_INTERLEAVE_SIZE_SHIFT;
1001 rdev->config.cayman.mem_max_burst_length_bytes = (tmp + 1) * 256;
1002 tmp = (gb_addr_config & NUM_SHADER_ENGINES_MASK) >> NUM_SHADER_ENGINES_SHIFT;
1003 rdev->config.cayman.num_shader_engines = tmp + 1;
1004 tmp = (gb_addr_config & NUM_GPUS_MASK) >> NUM_GPUS_SHIFT;
1005 rdev->config.cayman.num_gpus = tmp + 1;
1006 tmp = (gb_addr_config & MULTI_GPU_TILE_SIZE_MASK) >> MULTI_GPU_TILE_SIZE_SHIFT;
1007 rdev->config.cayman.multi_gpu_tile_size = 1 << tmp;
1008 tmp = (gb_addr_config & ROW_SIZE_MASK) >> ROW_SIZE_SHIFT;
1009 rdev->config.cayman.mem_row_size_in_kb = 1 << tmp;
1012 /* setup tiling info dword. gb_addr_config is not adequate since it does
1013 * not have bank info, so create a custom tiling dword.
1014 * bits 3:0 num_pipes
1015 * bits 7:4 num_banks
1016 * bits 11:8 group_size
1017 * bits 15:12 row_size
1019 rdev->config.cayman.tile_config = 0;
1020 switch (rdev->config.cayman.num_tile_pipes) {
1023 rdev->config.cayman.tile_config |= (0 << 0);
1026 rdev->config.cayman.tile_config |= (1 << 0);
1029 rdev->config.cayman.tile_config |= (2 << 0);
1032 rdev->config.cayman.tile_config |= (3 << 0);
1036 /* num banks is 8 on all fusion asics. 0 = 4, 1 = 8, 2 = 16 */
1037 if (rdev->flags & RADEON_IS_IGP)
1038 rdev->config.cayman.tile_config |= 1 << 4;
1040 switch ((mc_arb_ramcfg & NOOFBANK_MASK) >> NOOFBANK_SHIFT) {
1041 case 0: /* four banks */
1042 rdev->config.cayman.tile_config |= 0 << 4;
1044 case 1: /* eight banks */
1045 rdev->config.cayman.tile_config |= 1 << 4;
1047 case 2: /* sixteen banks */
1049 rdev->config.cayman.tile_config |= 2 << 4;
1053 rdev->config.cayman.tile_config |=
1054 ((gb_addr_config & PIPE_INTERLEAVE_SIZE_MASK) >> PIPE_INTERLEAVE_SIZE_SHIFT) << 8;
1055 rdev->config.cayman.tile_config |=
1056 ((gb_addr_config & ROW_SIZE_MASK) >> ROW_SIZE_SHIFT) << 12;
1059 for (i = (rdev->config.cayman.max_shader_engines - 1); i >= 0; i--) {
1060 u32 rb_disable_bitmap;
1062 WREG32(GRBM_GFX_INDEX, INSTANCE_BROADCAST_WRITES | SE_INDEX(i));
1063 WREG32(RLC_GFX_INDEX, INSTANCE_BROADCAST_WRITES | SE_INDEX(i));
1064 rb_disable_bitmap = (RREG32(CC_RB_BACKEND_DISABLE) & 0x00ff0000) >> 16;
1066 tmp |= rb_disable_bitmap;
1068 /* enabled rb are just the one not disabled :) */
1069 disabled_rb_mask = tmp;
1071 for (i = 0; i < (rdev->config.cayman.max_backends_per_se * rdev->config.cayman.max_shader_engines); i++)
1073 /* if all the backends are disabled, fix it up here */
1074 if ((disabled_rb_mask & tmp) == tmp) {
1075 for (i = 0; i < (rdev->config.cayman.max_backends_per_se * rdev->config.cayman.max_shader_engines); i++)
1076 disabled_rb_mask &= ~(1 << i);
1079 for (i = 0; i < rdev->config.cayman.max_shader_engines; i++) {
1080 u32 simd_disable_bitmap;
1082 WREG32(GRBM_GFX_INDEX, INSTANCE_BROADCAST_WRITES | SE_INDEX(i));
1083 WREG32(RLC_GFX_INDEX, INSTANCE_BROADCAST_WRITES | SE_INDEX(i));
1084 simd_disable_bitmap = (RREG32(CC_GC_SHADER_PIPE_CONFIG) & 0xffff0000) >> 16;
1085 simd_disable_bitmap |= 0xffffffff << rdev->config.cayman.max_simds_per_se;
1087 tmp |= simd_disable_bitmap;
1089 rdev->config.cayman.active_simds = hweight32(~tmp);
1091 WREG32(GRBM_GFX_INDEX, INSTANCE_BROADCAST_WRITES | SE_BROADCAST_WRITES);
1092 WREG32(RLC_GFX_INDEX, INSTANCE_BROADCAST_WRITES | SE_BROADCAST_WRITES);
1094 WREG32(GB_ADDR_CONFIG, gb_addr_config);
1095 WREG32(DMIF_ADDR_CONFIG, gb_addr_config);
1096 if (ASIC_IS_DCE6(rdev))
1097 WREG32(DMIF_ADDR_CALC, gb_addr_config);
1098 WREG32(HDP_ADDR_CONFIG, gb_addr_config);
1099 WREG32(DMA_TILING_CONFIG + DMA0_REGISTER_OFFSET, gb_addr_config);
1100 WREG32(DMA_TILING_CONFIG + DMA1_REGISTER_OFFSET, gb_addr_config);
1101 WREG32(UVD_UDEC_ADDR_CONFIG, gb_addr_config);
1102 WREG32(UVD_UDEC_DB_ADDR_CONFIG, gb_addr_config);
1103 WREG32(UVD_UDEC_DBW_ADDR_CONFIG, gb_addr_config);
1105 if ((rdev->config.cayman.max_backends_per_se == 1) &&
1106 (rdev->flags & RADEON_IS_IGP)) {
1107 if ((disabled_rb_mask & 3) == 2) {
1108 /* RB1 disabled, RB0 enabled */
1111 /* RB0 disabled, RB1 enabled */
1115 tmp = gb_addr_config & NUM_PIPES_MASK;
1116 tmp = r6xx_remap_render_backend(rdev, tmp,
1117 rdev->config.cayman.max_backends_per_se *
1118 rdev->config.cayman.max_shader_engines,
1119 CAYMAN_MAX_BACKENDS, disabled_rb_mask);
1121 rdev->config.cayman.backend_map = tmp;
1122 WREG32(GB_BACKEND_MAP, tmp);
1124 cgts_tcc_disable = 0xffff0000;
1125 for (i = 0; i < rdev->config.cayman.max_texture_channel_caches; i++)
1126 cgts_tcc_disable &= ~(1 << (16 + i));
1127 WREG32(CGTS_TCC_DISABLE, cgts_tcc_disable);
1128 WREG32(CGTS_SYS_TCC_DISABLE, cgts_tcc_disable);
1129 WREG32(CGTS_USER_SYS_TCC_DISABLE, cgts_tcc_disable);
1130 WREG32(CGTS_USER_TCC_DISABLE, cgts_tcc_disable);
1132 /* reprogram the shader complex */
1133 cgts_sm_ctrl_reg = RREG32(CGTS_SM_CTRL_REG);
1134 for (i = 0; i < 16; i++)
1135 WREG32(CGTS_SM_CTRL_REG, OVERRIDE);
1136 WREG32(CGTS_SM_CTRL_REG, cgts_sm_ctrl_reg);
1138 /* set HW defaults for 3D engine */
1139 WREG32(CP_MEQ_THRESHOLDS, MEQ1_START(0x30) | MEQ2_START(0x60));
1141 sx_debug_1 = RREG32(SX_DEBUG_1);
1142 sx_debug_1 |= ENABLE_NEW_SMX_ADDRESS;
1143 WREG32(SX_DEBUG_1, sx_debug_1);
1145 smx_dc_ctl0 = RREG32(SMX_DC_CTL0);
1146 smx_dc_ctl0 &= ~NUMBER_OF_SETS(0x1ff);
1147 smx_dc_ctl0 |= NUMBER_OF_SETS(rdev->config.cayman.sx_num_of_sets);
1148 WREG32(SMX_DC_CTL0, smx_dc_ctl0);
1150 WREG32(SPI_CONFIG_CNTL_1, VTX_DONE_DELAY(4) | CRC_SIMD_ID_WADDR_DISABLE);
1152 /* need to be explicitly zero-ed */
1153 WREG32(VGT_OFFCHIP_LDS_BASE, 0);
1154 WREG32(SQ_LSTMP_RING_BASE, 0);
1155 WREG32(SQ_HSTMP_RING_BASE, 0);
1156 WREG32(SQ_ESTMP_RING_BASE, 0);
1157 WREG32(SQ_GSTMP_RING_BASE, 0);
1158 WREG32(SQ_VSTMP_RING_BASE, 0);
1159 WREG32(SQ_PSTMP_RING_BASE, 0);
1161 WREG32(TA_CNTL_AUX, DISABLE_CUBE_ANISO);
1163 WREG32(SX_EXPORT_BUFFER_SIZES, (COLOR_BUFFER_SIZE((rdev->config.cayman.sx_max_export_size / 4) - 1) |
1164 POSITION_BUFFER_SIZE((rdev->config.cayman.sx_max_export_pos_size / 4) - 1) |
1165 SMX_BUFFER_SIZE((rdev->config.cayman.sx_max_export_smx_size / 4) - 1)));
1167 WREG32(PA_SC_FIFO_SIZE, (SC_PRIM_FIFO_SIZE(rdev->config.cayman.sc_prim_fifo_size) |
1168 SC_HIZ_TILE_FIFO_SIZE(rdev->config.cayman.sc_hiz_tile_fifo_size) |
1169 SC_EARLYZ_TILE_FIFO_SIZE(rdev->config.cayman.sc_earlyz_tile_fifo_size)));
1172 WREG32(VGT_NUM_INSTANCES, 1);
1174 WREG32(CP_PERFMON_CNTL, 0);
1176 WREG32(SQ_MS_FIFO_SIZES, (CACHE_FIFO_SIZE(16 * rdev->config.cayman.sq_num_cf_insts) |
1177 FETCH_FIFO_HIWATER(0x4) |
1178 DONE_FIFO_HIWATER(0xe0) |
1179 ALU_UPDATE_FIFO_HIWATER(0x8)));
1181 WREG32(SQ_GPR_RESOURCE_MGMT_1, NUM_CLAUSE_TEMP_GPRS(4));
1182 WREG32(SQ_CONFIG, (VC_ENABLE |
1187 WREG32(SQ_DYN_GPR_CNTL_PS_FLUSH_REQ, DYN_GPR_ENABLE);
1189 WREG32(PA_SC_FORCE_EOV_MAX_CNTS, (FORCE_EOV_MAX_CLK_CNT(4095) |
1190 FORCE_EOV_MAX_REZ_CNT(255)));
1192 WREG32(VGT_CACHE_INVALIDATION, CACHE_INVALIDATION(VC_AND_TC) |
1193 AUTO_INVLD_EN(ES_AND_GS_AUTO));
1195 WREG32(VGT_GS_VERTEX_REUSE, 16);
1196 WREG32(PA_SC_LINE_STIPPLE_STATE, 0);
1198 WREG32(CB_PERF_CTR0_SEL_0, 0);
1199 WREG32(CB_PERF_CTR0_SEL_1, 0);
1200 WREG32(CB_PERF_CTR1_SEL_0, 0);
1201 WREG32(CB_PERF_CTR1_SEL_1, 0);
1202 WREG32(CB_PERF_CTR2_SEL_0, 0);
1203 WREG32(CB_PERF_CTR2_SEL_1, 0);
1204 WREG32(CB_PERF_CTR3_SEL_0, 0);
1205 WREG32(CB_PERF_CTR3_SEL_1, 0);
1207 tmp = RREG32(HDP_MISC_CNTL);
1208 tmp |= HDP_FLUSH_INVALIDATE_CACHE;
1209 WREG32(HDP_MISC_CNTL, tmp);
1211 hdp_host_path_cntl = RREG32(HDP_HOST_PATH_CNTL);
1212 WREG32(HDP_HOST_PATH_CNTL, hdp_host_path_cntl);
1214 WREG32(PA_CL_ENHANCE, CLIP_VTX_REORDER_ENA | NUM_CLIP_SEQ(3));
1218 /* set clockgating golden values on TN */
1219 if (rdev->family == CHIP_ARUBA) {
1220 tmp = RREG32_CG(CG_CGTT_LOCAL_0);
1222 WREG32_CG(CG_CGTT_LOCAL_0, tmp);
1223 tmp = RREG32_CG(CG_CGTT_LOCAL_1);
1225 WREG32_CG(CG_CGTT_LOCAL_1, tmp);
1232 void cayman_pcie_gart_tlb_flush(struct radeon_device *rdev)
1234 /* flush hdp cache */
1235 WREG32(HDP_MEM_COHERENCY_FLUSH_CNTL, 0x1);
1237 /* bits 0-7 are the VM contexts0-7 */
1238 WREG32(VM_INVALIDATE_REQUEST, 1);
1241 static int cayman_pcie_gart_enable(struct radeon_device *rdev)
1245 if (rdev->gart.robj == NULL) {
1246 dev_err(rdev->dev, "No VRAM object for PCIE GART.\n");
1249 r = radeon_gart_table_vram_pin(rdev);
1252 /* Setup TLB control */
1253 WREG32(MC_VM_MX_L1_TLB_CNTL,
1256 ENABLE_L1_FRAGMENT_PROCESSING |
1257 SYSTEM_ACCESS_MODE_NOT_IN_SYS |
1258 ENABLE_ADVANCED_DRIVER_MODEL |
1259 SYSTEM_APERTURE_UNMAPPED_ACCESS_PASS_THRU);
1260 /* Setup L2 cache */
1261 WREG32(VM_L2_CNTL, ENABLE_L2_CACHE |
1262 ENABLE_L2_FRAGMENT_PROCESSING |
1263 ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE |
1264 ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE |
1265 EFFECTIVE_L2_QUEUE_SIZE(7) |
1266 CONTEXT1_IDENTITY_ACCESS_MODE(1));
1267 WREG32(VM_L2_CNTL2, INVALIDATE_ALL_L1_TLBS | INVALIDATE_L2_CACHE);
1268 WREG32(VM_L2_CNTL3, L2_CACHE_BIGK_ASSOCIATIVITY |
1270 L2_CACHE_BIGK_FRAGMENT_SIZE(6));
1271 /* setup context0 */
1272 WREG32(VM_CONTEXT0_PAGE_TABLE_START_ADDR, rdev->mc.gtt_start >> 12);
1273 WREG32(VM_CONTEXT0_PAGE_TABLE_END_ADDR, rdev->mc.gtt_end >> 12);
1274 WREG32(VM_CONTEXT0_PAGE_TABLE_BASE_ADDR, rdev->gart.table_addr >> 12);
1275 WREG32(VM_CONTEXT0_PROTECTION_FAULT_DEFAULT_ADDR,
1276 (u32)(rdev->dummy_page.addr >> 12));
1277 WREG32(VM_CONTEXT0_CNTL2, 0);
1278 WREG32(VM_CONTEXT0_CNTL, ENABLE_CONTEXT | PAGE_TABLE_DEPTH(0) |
1279 RANGE_PROTECTION_FAULT_ENABLE_DEFAULT);
1285 /* empty context1-7 */
1286 /* Assign the pt base to something valid for now; the pts used for
1287 * the VMs are determined by the application and setup and assigned
1288 * on the fly in the vm part of radeon_gart.c
1290 for (i = 1; i < 8; i++) {
1291 WREG32(VM_CONTEXT0_PAGE_TABLE_START_ADDR + (i << 2), 0);
1292 WREG32(VM_CONTEXT0_PAGE_TABLE_END_ADDR + (i << 2),
1293 rdev->vm_manager.max_pfn - 1);
1294 WREG32(VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + (i << 2),
1295 rdev->vm_manager.saved_table_addr[i]);
1298 /* enable context1-7 */
1299 WREG32(VM_CONTEXT1_PROTECTION_FAULT_DEFAULT_ADDR,
1300 (u32)(rdev->dummy_page.addr >> 12));
1301 WREG32(VM_CONTEXT1_CNTL2, 4);
1302 WREG32(VM_CONTEXT1_CNTL, ENABLE_CONTEXT | PAGE_TABLE_DEPTH(1) |
1303 PAGE_TABLE_BLOCK_SIZE(radeon_vm_block_size - 9) |
1304 RANGE_PROTECTION_FAULT_ENABLE_INTERRUPT |
1305 RANGE_PROTECTION_FAULT_ENABLE_DEFAULT |
1306 DUMMY_PAGE_PROTECTION_FAULT_ENABLE_INTERRUPT |
1307 DUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT |
1308 PDE0_PROTECTION_FAULT_ENABLE_INTERRUPT |
1309 PDE0_PROTECTION_FAULT_ENABLE_DEFAULT |
1310 VALID_PROTECTION_FAULT_ENABLE_INTERRUPT |
1311 VALID_PROTECTION_FAULT_ENABLE_DEFAULT |
1312 READ_PROTECTION_FAULT_ENABLE_INTERRUPT |
1313 READ_PROTECTION_FAULT_ENABLE_DEFAULT |
1314 WRITE_PROTECTION_FAULT_ENABLE_INTERRUPT |
1315 WRITE_PROTECTION_FAULT_ENABLE_DEFAULT);
1317 cayman_pcie_gart_tlb_flush(rdev);
1318 DRM_INFO("PCIE GART of %uM enabled (table at 0x%016llX).\n",
1319 (unsigned)(rdev->mc.gtt_size >> 20),
1320 (unsigned long long)rdev->gart.table_addr);
1321 rdev->gart.ready = true;
1325 static void cayman_pcie_gart_disable(struct radeon_device *rdev)
1329 for (i = 1; i < 8; ++i) {
1330 rdev->vm_manager.saved_table_addr[i] = RREG32(
1331 VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + (i << 2));
1334 /* Disable all tables */
1335 WREG32(VM_CONTEXT0_CNTL, 0);
1336 WREG32(VM_CONTEXT1_CNTL, 0);
1337 /* Setup TLB control */
1338 WREG32(MC_VM_MX_L1_TLB_CNTL, ENABLE_L1_FRAGMENT_PROCESSING |
1339 SYSTEM_ACCESS_MODE_NOT_IN_SYS |
1340 SYSTEM_APERTURE_UNMAPPED_ACCESS_PASS_THRU);
1341 /* Setup L2 cache */
1342 WREG32(VM_L2_CNTL, ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE |
1343 ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE |
1344 EFFECTIVE_L2_QUEUE_SIZE(7) |
1345 CONTEXT1_IDENTITY_ACCESS_MODE(1));
1346 WREG32(VM_L2_CNTL2, 0);
1347 WREG32(VM_L2_CNTL3, L2_CACHE_BIGK_ASSOCIATIVITY |
1348 L2_CACHE_BIGK_FRAGMENT_SIZE(6));
1349 radeon_gart_table_vram_unpin(rdev);
1352 static void cayman_pcie_gart_fini(struct radeon_device *rdev)
1354 cayman_pcie_gart_disable(rdev);
1355 radeon_gart_table_vram_free(rdev);
1356 radeon_gart_fini(rdev);
1359 void cayman_cp_int_cntl_setup(struct radeon_device *rdev,
1360 int ring, u32 cp_int_cntl)
1362 WREG32(SRBM_GFX_CNTL, RINGID(ring));
1363 WREG32(CP_INT_CNTL, cp_int_cntl);
1369 void cayman_fence_ring_emit(struct radeon_device *rdev,
1370 struct radeon_fence *fence)
1372 struct radeon_ring *ring = &rdev->ring[fence->ring];
1373 u64 addr = rdev->fence_drv[fence->ring].gpu_addr;
1374 u32 cp_coher_cntl = PACKET3_FULL_CACHE_ENA | PACKET3_TC_ACTION_ENA |
1375 PACKET3_SH_ACTION_ENA;
1377 /* flush read cache over gart for this vmid */
1378 radeon_ring_write(ring, PACKET3(PACKET3_SURFACE_SYNC, 3));
1379 radeon_ring_write(ring, PACKET3_ENGINE_ME | cp_coher_cntl);
1380 radeon_ring_write(ring, 0xFFFFFFFF);
1381 radeon_ring_write(ring, 0);
1382 radeon_ring_write(ring, 10); /* poll interval */
1383 /* EVENT_WRITE_EOP - flush caches, send int */
1384 radeon_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE_EOP, 4));
1385 radeon_ring_write(ring, EVENT_TYPE(CACHE_FLUSH_AND_INV_EVENT_TS) | EVENT_INDEX(5));
1386 radeon_ring_write(ring, lower_32_bits(addr));
1387 radeon_ring_write(ring, (upper_32_bits(addr) & 0xff) | DATA_SEL(1) | INT_SEL(2));
1388 radeon_ring_write(ring, fence->seq);
1389 radeon_ring_write(ring, 0);
1392 void cayman_ring_ib_execute(struct radeon_device *rdev, struct radeon_ib *ib)
1394 struct radeon_ring *ring = &rdev->ring[ib->ring];
1395 unsigned vm_id = ib->vm ? ib->vm->ids[ib->ring].id : 0;
1396 u32 cp_coher_cntl = PACKET3_FULL_CACHE_ENA | PACKET3_TC_ACTION_ENA |
1397 PACKET3_SH_ACTION_ENA;
1399 /* set to DX10/11 mode */
1400 radeon_ring_write(ring, PACKET3(PACKET3_MODE_CONTROL, 0));
1401 radeon_ring_write(ring, 1);
1403 if (ring->rptr_save_reg) {
1404 uint32_t next_rptr = ring->wptr + 3 + 4 + 8;
1405 radeon_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1));
1406 radeon_ring_write(ring, ((ring->rptr_save_reg -
1407 PACKET3_SET_CONFIG_REG_START) >> 2));
1408 radeon_ring_write(ring, next_rptr);
1411 radeon_ring_write(ring, PACKET3(PACKET3_INDIRECT_BUFFER, 2));
1412 radeon_ring_write(ring,
1416 (ib->gpu_addr & 0xFFFFFFFC));
1417 radeon_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xFF);
1418 radeon_ring_write(ring, ib->length_dw | (vm_id << 24));
1420 /* flush read cache over gart for this vmid */
1421 radeon_ring_write(ring, PACKET3(PACKET3_SURFACE_SYNC, 3));
1422 radeon_ring_write(ring, PACKET3_ENGINE_ME | cp_coher_cntl);
1423 radeon_ring_write(ring, 0xFFFFFFFF);
1424 radeon_ring_write(ring, 0);
1425 radeon_ring_write(ring, (vm_id << 24) | 10); /* poll interval */
1428 static void cayman_cp_enable(struct radeon_device *rdev, bool enable)
1431 WREG32(CP_ME_CNTL, 0);
1433 if (rdev->asic->copy.copy_ring_index == RADEON_RING_TYPE_GFX_INDEX)
1434 radeon_ttm_set_active_vram_size(rdev, rdev->mc.visible_vram_size);
1435 WREG32(CP_ME_CNTL, (CP_ME_HALT | CP_PFP_HALT));
1436 WREG32(SCRATCH_UMSK, 0);
1437 rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = false;
1441 u32 cayman_gfx_get_rptr(struct radeon_device *rdev,
1442 struct radeon_ring *ring)
1446 if (rdev->wb.enabled)
1447 rptr = rdev->wb.wb[ring->rptr_offs/4];
1449 if (ring->idx == RADEON_RING_TYPE_GFX_INDEX)
1450 rptr = RREG32(CP_RB0_RPTR);
1451 else if (ring->idx == CAYMAN_RING_TYPE_CP1_INDEX)
1452 rptr = RREG32(CP_RB1_RPTR);
1454 rptr = RREG32(CP_RB2_RPTR);
1460 u32 cayman_gfx_get_wptr(struct radeon_device *rdev,
1461 struct radeon_ring *ring)
1465 if (ring->idx == RADEON_RING_TYPE_GFX_INDEX)
1466 wptr = RREG32(CP_RB0_WPTR);
1467 else if (ring->idx == CAYMAN_RING_TYPE_CP1_INDEX)
1468 wptr = RREG32(CP_RB1_WPTR);
1470 wptr = RREG32(CP_RB2_WPTR);
1475 void cayman_gfx_set_wptr(struct radeon_device *rdev,
1476 struct radeon_ring *ring)
1478 if (ring->idx == RADEON_RING_TYPE_GFX_INDEX) {
1479 WREG32(CP_RB0_WPTR, ring->wptr);
1480 (void)RREG32(CP_RB0_WPTR);
1481 } else if (ring->idx == CAYMAN_RING_TYPE_CP1_INDEX) {
1482 WREG32(CP_RB1_WPTR, ring->wptr);
1483 (void)RREG32(CP_RB1_WPTR);
1485 WREG32(CP_RB2_WPTR, ring->wptr);
1486 (void)RREG32(CP_RB2_WPTR);
1490 static int cayman_cp_load_microcode(struct radeon_device *rdev)
1492 const __be32 *fw_data;
1495 if (!rdev->me_fw || !rdev->pfp_fw)
1498 cayman_cp_enable(rdev, false);
1500 fw_data = (const __be32 *)rdev->pfp_fw->data;
1501 WREG32(CP_PFP_UCODE_ADDR, 0);
1502 for (i = 0; i < CAYMAN_PFP_UCODE_SIZE; i++)
1503 WREG32(CP_PFP_UCODE_DATA, be32_to_cpup(fw_data++));
1504 WREG32(CP_PFP_UCODE_ADDR, 0);
1506 fw_data = (const __be32 *)rdev->me_fw->data;
1507 WREG32(CP_ME_RAM_WADDR, 0);
1508 for (i = 0; i < CAYMAN_PM4_UCODE_SIZE; i++)
1509 WREG32(CP_ME_RAM_DATA, be32_to_cpup(fw_data++));
1511 WREG32(CP_PFP_UCODE_ADDR, 0);
1512 WREG32(CP_ME_RAM_WADDR, 0);
1513 WREG32(CP_ME_RAM_RADDR, 0);
1517 static int cayman_cp_start(struct radeon_device *rdev)
1519 struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
1522 r = radeon_ring_lock(rdev, ring, 7);
1524 DRM_ERROR("radeon: cp failed to lock ring (%d).\n", r);
1527 radeon_ring_write(ring, PACKET3(PACKET3_ME_INITIALIZE, 5));
1528 radeon_ring_write(ring, 0x1);
1529 radeon_ring_write(ring, 0x0);
1530 radeon_ring_write(ring, rdev->config.cayman.max_hw_contexts - 1);
1531 radeon_ring_write(ring, PACKET3_ME_INITIALIZE_DEVICE_ID(1));
1532 radeon_ring_write(ring, 0);
1533 radeon_ring_write(ring, 0);
1534 radeon_ring_unlock_commit(rdev, ring, false);
1536 cayman_cp_enable(rdev, true);
1538 r = radeon_ring_lock(rdev, ring, cayman_default_size + 19);
1540 DRM_ERROR("radeon: cp failed to lock ring (%d).\n", r);
1544 /* setup clear context state */
1545 radeon_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
1546 radeon_ring_write(ring, PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
1548 for (i = 0; i < cayman_default_size; i++)
1549 radeon_ring_write(ring, cayman_default_state[i]);
1551 radeon_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
1552 radeon_ring_write(ring, PACKET3_PREAMBLE_END_CLEAR_STATE);
1554 /* set clear context state */
1555 radeon_ring_write(ring, PACKET3(PACKET3_CLEAR_STATE, 0));
1556 radeon_ring_write(ring, 0);
1558 /* SQ_VTX_BASE_VTX_LOC */
1559 radeon_ring_write(ring, 0xc0026f00);
1560 radeon_ring_write(ring, 0x00000000);
1561 radeon_ring_write(ring, 0x00000000);
1562 radeon_ring_write(ring, 0x00000000);
1565 radeon_ring_write(ring, 0xc0036f00);
1566 radeon_ring_write(ring, 0x00000bc4);
1567 radeon_ring_write(ring, 0xffffffff);
1568 radeon_ring_write(ring, 0xffffffff);
1569 radeon_ring_write(ring, 0xffffffff);
1571 radeon_ring_write(ring, 0xc0026900);
1572 radeon_ring_write(ring, 0x00000316);
1573 radeon_ring_write(ring, 0x0000000e); /* VGT_VERTEX_REUSE_BLOCK_CNTL */
1574 radeon_ring_write(ring, 0x00000010); /* */
1576 radeon_ring_unlock_commit(rdev, ring, false);
1578 /* XXX init other rings */
1583 static void cayman_cp_fini(struct radeon_device *rdev)
1585 struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
1586 cayman_cp_enable(rdev, false);
1587 radeon_ring_fini(rdev, ring);
1588 radeon_scratch_free(rdev, ring->rptr_save_reg);
1591 static int cayman_cp_resume(struct radeon_device *rdev)
1593 static const int ridx[] = {
1594 RADEON_RING_TYPE_GFX_INDEX,
1595 CAYMAN_RING_TYPE_CP1_INDEX,
1596 CAYMAN_RING_TYPE_CP2_INDEX
1598 static const unsigned cp_rb_cntl[] = {
1603 static const unsigned cp_rb_rptr_addr[] = {
1608 static const unsigned cp_rb_rptr_addr_hi[] = {
1609 CP_RB0_RPTR_ADDR_HI,
1610 CP_RB1_RPTR_ADDR_HI,
1613 static const unsigned cp_rb_base[] = {
1618 static const unsigned cp_rb_rptr[] = {
1623 static const unsigned cp_rb_wptr[] = {
1628 struct radeon_ring *ring;
1631 /* Reset cp; if cp is reset, then PA, SH, VGT also need to be reset */
1632 WREG32(GRBM_SOFT_RESET, (SOFT_RESET_CP |
1638 RREG32(GRBM_SOFT_RESET);
1640 WREG32(GRBM_SOFT_RESET, 0);
1641 RREG32(GRBM_SOFT_RESET);
1643 WREG32(CP_SEM_WAIT_TIMER, 0x0);
1644 WREG32(CP_SEM_INCOMPLETE_TIMER_CNTL, 0x0);
1646 /* Set the write pointer delay */
1647 WREG32(CP_RB_WPTR_DELAY, 0);
1649 WREG32(CP_DEBUG, (1 << 27));
1651 /* set the wb address whether it's enabled or not */
1652 WREG32(SCRATCH_ADDR, ((rdev->wb.gpu_addr + RADEON_WB_SCRATCH_OFFSET) >> 8) & 0xFFFFFFFF);
1653 WREG32(SCRATCH_UMSK, 0xff);
1655 for (i = 0; i < 3; ++i) {
1659 /* Set ring buffer size */
1660 ring = &rdev->ring[ridx[i]];
1661 rb_cntl = order_base_2(ring->ring_size / 8);
1662 rb_cntl |= order_base_2(RADEON_GPU_PAGE_SIZE/8) << 8;
1664 rb_cntl |= BUF_SWAP_32BIT;
1666 WREG32(cp_rb_cntl[i], rb_cntl);
1668 /* set the wb address whether it's enabled or not */
1669 addr = rdev->wb.gpu_addr + RADEON_WB_CP_RPTR_OFFSET;
1670 WREG32(cp_rb_rptr_addr[i], addr & 0xFFFFFFFC);
1671 WREG32(cp_rb_rptr_addr_hi[i], upper_32_bits(addr) & 0xFF);
1674 /* set the rb base addr, this causes an internal reset of ALL rings */
1675 for (i = 0; i < 3; ++i) {
1676 ring = &rdev->ring[ridx[i]];
1677 WREG32(cp_rb_base[i], ring->gpu_addr >> 8);
1680 for (i = 0; i < 3; ++i) {
1681 /* Initialize the ring buffer's read and write pointers */
1682 ring = &rdev->ring[ridx[i]];
1683 WREG32_P(cp_rb_cntl[i], RB_RPTR_WR_ENA, ~RB_RPTR_WR_ENA);
1686 WREG32(cp_rb_rptr[i], 0);
1687 WREG32(cp_rb_wptr[i], ring->wptr);
1690 WREG32_P(cp_rb_cntl[i], 0, ~RB_RPTR_WR_ENA);
1693 /* start the rings */
1694 cayman_cp_start(rdev);
1695 rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = true;
1696 rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX].ready = false;
1697 rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX].ready = false;
1698 /* this only test cp0 */
1699 r = radeon_ring_test(rdev, RADEON_RING_TYPE_GFX_INDEX, &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]);
1701 rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = false;
1702 rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX].ready = false;
1703 rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX].ready = false;
1707 if (rdev->asic->copy.copy_ring_index == RADEON_RING_TYPE_GFX_INDEX)
1708 radeon_ttm_set_active_vram_size(rdev, rdev->mc.real_vram_size);
1713 u32 cayman_gpu_check_soft_reset(struct radeon_device *rdev)
1719 tmp = RREG32(GRBM_STATUS);
1720 if (tmp & (PA_BUSY | SC_BUSY |
1722 TA_BUSY | VGT_BUSY |
1724 GDS_BUSY | SPI_BUSY |
1725 IA_BUSY | IA_BUSY_NO_DMA))
1726 reset_mask |= RADEON_RESET_GFX;
1728 if (tmp & (CF_RQ_PENDING | PF_RQ_PENDING |
1729 CP_BUSY | CP_COHERENCY_BUSY))
1730 reset_mask |= RADEON_RESET_CP;
1732 if (tmp & GRBM_EE_BUSY)
1733 reset_mask |= RADEON_RESET_GRBM | RADEON_RESET_GFX | RADEON_RESET_CP;
1735 /* DMA_STATUS_REG 0 */
1736 tmp = RREG32(DMA_STATUS_REG + DMA0_REGISTER_OFFSET);
1737 if (!(tmp & DMA_IDLE))
1738 reset_mask |= RADEON_RESET_DMA;
1740 /* DMA_STATUS_REG 1 */
1741 tmp = RREG32(DMA_STATUS_REG + DMA1_REGISTER_OFFSET);
1742 if (!(tmp & DMA_IDLE))
1743 reset_mask |= RADEON_RESET_DMA1;
1746 tmp = RREG32(SRBM_STATUS2);
1748 reset_mask |= RADEON_RESET_DMA;
1750 if (tmp & DMA1_BUSY)
1751 reset_mask |= RADEON_RESET_DMA1;
1754 tmp = RREG32(SRBM_STATUS);
1755 if (tmp & (RLC_RQ_PENDING | RLC_BUSY))
1756 reset_mask |= RADEON_RESET_RLC;
1759 reset_mask |= RADEON_RESET_IH;
1762 reset_mask |= RADEON_RESET_SEM;
1764 if (tmp & GRBM_RQ_PENDING)
1765 reset_mask |= RADEON_RESET_GRBM;
1768 reset_mask |= RADEON_RESET_VMC;
1770 if (tmp & (MCB_BUSY | MCB_NON_DISPLAY_BUSY |
1771 MCC_BUSY | MCD_BUSY))
1772 reset_mask |= RADEON_RESET_MC;
1774 if (evergreen_is_display_hung(rdev))
1775 reset_mask |= RADEON_RESET_DISPLAY;
1778 tmp = RREG32(VM_L2_STATUS);
1780 reset_mask |= RADEON_RESET_VMC;
1782 /* Skip MC reset as it's mostly likely not hung, just busy */
1783 if (reset_mask & RADEON_RESET_MC) {
1784 DRM_DEBUG("MC busy: 0x%08X, clearing.\n", reset_mask);
1785 reset_mask &= ~RADEON_RESET_MC;
1791 static void cayman_gpu_soft_reset(struct radeon_device *rdev, u32 reset_mask)
1793 struct evergreen_mc_save save;
1794 u32 grbm_soft_reset = 0, srbm_soft_reset = 0;
1797 if (reset_mask == 0)
1800 dev_info(rdev->dev, "GPU softreset: 0x%08X\n", reset_mask);
1802 evergreen_print_gpu_status_regs(rdev);
1803 dev_info(rdev->dev, " VM_CONTEXT0_PROTECTION_FAULT_ADDR 0x%08X\n",
1805 dev_info(rdev->dev, " VM_CONTEXT0_PROTECTION_FAULT_STATUS 0x%08X\n",
1807 dev_info(rdev->dev, " VM_CONTEXT1_PROTECTION_FAULT_ADDR 0x%08X\n",
1809 dev_info(rdev->dev, " VM_CONTEXT1_PROTECTION_FAULT_STATUS 0x%08X\n",
1812 /* Disable CP parsing/prefetching */
1813 WREG32(CP_ME_CNTL, CP_ME_HALT | CP_PFP_HALT);
1815 if (reset_mask & RADEON_RESET_DMA) {
1817 tmp = RREG32(DMA_RB_CNTL + DMA0_REGISTER_OFFSET);
1818 tmp &= ~DMA_RB_ENABLE;
1819 WREG32(DMA_RB_CNTL + DMA0_REGISTER_OFFSET, tmp);
1822 if (reset_mask & RADEON_RESET_DMA1) {
1824 tmp = RREG32(DMA_RB_CNTL + DMA1_REGISTER_OFFSET);
1825 tmp &= ~DMA_RB_ENABLE;
1826 WREG32(DMA_RB_CNTL + DMA1_REGISTER_OFFSET, tmp);
1831 evergreen_mc_stop(rdev, &save);
1832 if (evergreen_mc_wait_for_idle(rdev)) {
1833 dev_warn(rdev->dev, "Wait for MC idle timedout !\n");
1836 if (reset_mask & (RADEON_RESET_GFX | RADEON_RESET_COMPUTE)) {
1837 grbm_soft_reset = SOFT_RESET_CB |
1851 if (reset_mask & RADEON_RESET_CP) {
1852 grbm_soft_reset |= SOFT_RESET_CP | SOFT_RESET_VGT;
1854 srbm_soft_reset |= SOFT_RESET_GRBM;
1857 if (reset_mask & RADEON_RESET_DMA)
1858 srbm_soft_reset |= SOFT_RESET_DMA;
1860 if (reset_mask & RADEON_RESET_DMA1)
1861 srbm_soft_reset |= SOFT_RESET_DMA1;
1863 if (reset_mask & RADEON_RESET_DISPLAY)
1864 srbm_soft_reset |= SOFT_RESET_DC;
1866 if (reset_mask & RADEON_RESET_RLC)
1867 srbm_soft_reset |= SOFT_RESET_RLC;
1869 if (reset_mask & RADEON_RESET_SEM)
1870 srbm_soft_reset |= SOFT_RESET_SEM;
1872 if (reset_mask & RADEON_RESET_IH)
1873 srbm_soft_reset |= SOFT_RESET_IH;
1875 if (reset_mask & RADEON_RESET_GRBM)
1876 srbm_soft_reset |= SOFT_RESET_GRBM;
1878 if (reset_mask & RADEON_RESET_VMC)
1879 srbm_soft_reset |= SOFT_RESET_VMC;
1881 if (!(rdev->flags & RADEON_IS_IGP)) {
1882 if (reset_mask & RADEON_RESET_MC)
1883 srbm_soft_reset |= SOFT_RESET_MC;
1886 if (grbm_soft_reset) {
1887 tmp = RREG32(GRBM_SOFT_RESET);
1888 tmp |= grbm_soft_reset;
1889 dev_info(rdev->dev, "GRBM_SOFT_RESET=0x%08X\n", tmp);
1890 WREG32(GRBM_SOFT_RESET, tmp);
1891 tmp = RREG32(GRBM_SOFT_RESET);
1895 tmp &= ~grbm_soft_reset;
1896 WREG32(GRBM_SOFT_RESET, tmp);
1897 tmp = RREG32(GRBM_SOFT_RESET);
1900 if (srbm_soft_reset) {
1901 tmp = RREG32(SRBM_SOFT_RESET);
1902 tmp |= srbm_soft_reset;
1903 dev_info(rdev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp);
1904 WREG32(SRBM_SOFT_RESET, tmp);
1905 tmp = RREG32(SRBM_SOFT_RESET);
1909 tmp &= ~srbm_soft_reset;
1910 WREG32(SRBM_SOFT_RESET, tmp);
1911 tmp = RREG32(SRBM_SOFT_RESET);
1914 /* Wait a little for things to settle down */
1917 evergreen_mc_resume(rdev, &save);
1920 evergreen_print_gpu_status_regs(rdev);
1923 int cayman_asic_reset(struct radeon_device *rdev, bool hard)
1928 evergreen_gpu_pci_config_reset(rdev);
1932 reset_mask = cayman_gpu_check_soft_reset(rdev);
1935 r600_set_bios_scratch_engine_hung(rdev, true);
1937 cayman_gpu_soft_reset(rdev, reset_mask);
1939 reset_mask = cayman_gpu_check_soft_reset(rdev);
1942 evergreen_gpu_pci_config_reset(rdev);
1944 r600_set_bios_scratch_engine_hung(rdev, false);
1950 * cayman_gfx_is_lockup - Check if the GFX engine is locked up
1952 * @rdev: radeon_device pointer
1953 * @ring: radeon_ring structure holding ring information
1955 * Check if the GFX engine is locked up.
1956 * Returns true if the engine appears to be locked up, false if not.
1958 bool cayman_gfx_is_lockup(struct radeon_device *rdev, struct radeon_ring *ring)
1960 u32 reset_mask = cayman_gpu_check_soft_reset(rdev);
1962 if (!(reset_mask & (RADEON_RESET_GFX |
1963 RADEON_RESET_COMPUTE |
1964 RADEON_RESET_CP))) {
1965 radeon_ring_lockup_update(rdev, ring);
1968 return radeon_ring_test_lockup(rdev, ring);
1971 static void cayman_uvd_init(struct radeon_device *rdev)
1978 r = radeon_uvd_init(rdev);
1980 dev_err(rdev->dev, "failed UVD (%d) init.\n", r);
1982 * At this point rdev->uvd.vcpu_bo is NULL which trickles down
1983 * to early fails uvd_v2_2_resume() and thus nothing happens
1984 * there. So it is pointless to try to go through that code
1985 * hence why we disable uvd here.
1987 rdev->has_uvd = false;
1990 rdev->ring[R600_RING_TYPE_UVD_INDEX].ring_obj = NULL;
1991 r600_ring_init(rdev, &rdev->ring[R600_RING_TYPE_UVD_INDEX], 4096);
1994 static void cayman_uvd_start(struct radeon_device *rdev)
2001 r = uvd_v2_2_resume(rdev);
2003 dev_err(rdev->dev, "failed UVD resume (%d).\n", r);
2006 r = radeon_fence_driver_start_ring(rdev, R600_RING_TYPE_UVD_INDEX);
2008 dev_err(rdev->dev, "failed initializing UVD fences (%d).\n", r);
2014 rdev->ring[R600_RING_TYPE_UVD_INDEX].ring_size = 0;
2017 static void cayman_uvd_resume(struct radeon_device *rdev)
2019 struct radeon_ring *ring;
2022 if (!rdev->has_uvd || !rdev->ring[R600_RING_TYPE_UVD_INDEX].ring_size)
2025 ring = &rdev->ring[R600_RING_TYPE_UVD_INDEX];
2026 r = radeon_ring_init(rdev, ring, ring->ring_size, 0, PACKET0(UVD_NO_OP, 0));
2028 dev_err(rdev->dev, "failed initializing UVD ring (%d).\n", r);
2031 r = uvd_v1_0_init(rdev);
2033 dev_err(rdev->dev, "failed initializing UVD (%d).\n", r);
2038 static void cayman_vce_init(struct radeon_device *rdev)
2042 /* Only set for CHIP_ARUBA */
2046 r = radeon_vce_init(rdev);
2048 dev_err(rdev->dev, "failed VCE (%d) init.\n", r);
2050 * At this point rdev->vce.vcpu_bo is NULL which trickles down
2051 * to early fails cayman_vce_start() and thus nothing happens
2052 * there. So it is pointless to try to go through that code
2053 * hence why we disable vce here.
2055 rdev->has_vce = false;
2058 rdev->ring[TN_RING_TYPE_VCE1_INDEX].ring_obj = NULL;
2059 r600_ring_init(rdev, &rdev->ring[TN_RING_TYPE_VCE1_INDEX], 4096);
2060 rdev->ring[TN_RING_TYPE_VCE2_INDEX].ring_obj = NULL;
2061 r600_ring_init(rdev, &rdev->ring[TN_RING_TYPE_VCE2_INDEX], 4096);
2064 static void cayman_vce_start(struct radeon_device *rdev)
2071 r = radeon_vce_resume(rdev);
2073 dev_err(rdev->dev, "failed VCE resume (%d).\n", r);
2076 r = vce_v1_0_resume(rdev);
2078 dev_err(rdev->dev, "failed VCE resume (%d).\n", r);
2081 r = radeon_fence_driver_start_ring(rdev, TN_RING_TYPE_VCE1_INDEX);
2083 dev_err(rdev->dev, "failed initializing VCE1 fences (%d).\n", r);
2086 r = radeon_fence_driver_start_ring(rdev, TN_RING_TYPE_VCE2_INDEX);
2088 dev_err(rdev->dev, "failed initializing VCE2 fences (%d).\n", r);
2094 rdev->ring[TN_RING_TYPE_VCE1_INDEX].ring_size = 0;
2095 rdev->ring[TN_RING_TYPE_VCE2_INDEX].ring_size = 0;
2098 static void cayman_vce_resume(struct radeon_device *rdev)
2100 struct radeon_ring *ring;
2103 if (!rdev->has_vce || !rdev->ring[TN_RING_TYPE_VCE1_INDEX].ring_size)
2106 ring = &rdev->ring[TN_RING_TYPE_VCE1_INDEX];
2107 r = radeon_ring_init(rdev, ring, ring->ring_size, 0, 0x0);
2109 dev_err(rdev->dev, "failed initializing VCE1 ring (%d).\n", r);
2112 ring = &rdev->ring[TN_RING_TYPE_VCE2_INDEX];
2113 r = radeon_ring_init(rdev, ring, ring->ring_size, 0, 0x0);
2115 dev_err(rdev->dev, "failed initializing VCE1 ring (%d).\n", r);
2118 r = vce_v1_0_init(rdev);
2120 dev_err(rdev->dev, "failed initializing VCE (%d).\n", r);
2125 static int cayman_startup(struct radeon_device *rdev)
2127 struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
2130 /* enable pcie gen2 link */
2131 evergreen_pcie_gen2_enable(rdev);
2133 evergreen_program_aspm(rdev);
2135 /* scratch needs to be initialized before MC */
2136 r = r600_vram_scratch_init(rdev);
2140 evergreen_mc_program(rdev);
2142 if (!(rdev->flags & RADEON_IS_IGP) && !rdev->pm.dpm_enabled) {
2143 r = ni_mc_load_microcode(rdev);
2145 DRM_ERROR("Failed to load MC firmware!\n");
2150 r = cayman_pcie_gart_enable(rdev);
2153 cayman_gpu_init(rdev);
2155 /* allocate rlc buffers */
2156 if (rdev->flags & RADEON_IS_IGP) {
2157 rdev->rlc.reg_list = tn_rlc_save_restore_register_list;
2158 rdev->rlc.reg_list_size =
2159 (u32)ARRAY_SIZE(tn_rlc_save_restore_register_list);
2160 rdev->rlc.cs_data = cayman_cs_data;
2161 r = sumo_rlc_init(rdev);
2163 DRM_ERROR("Failed to init rlc BOs!\n");
2168 /* allocate wb buffer */
2169 r = radeon_wb_init(rdev);
2173 r = radeon_fence_driver_start_ring(rdev, RADEON_RING_TYPE_GFX_INDEX);
2175 dev_err(rdev->dev, "failed initializing CP fences (%d).\n", r);
2179 cayman_uvd_start(rdev);
2180 cayman_vce_start(rdev);
2182 r = radeon_fence_driver_start_ring(rdev, CAYMAN_RING_TYPE_CP1_INDEX);
2184 dev_err(rdev->dev, "failed initializing CP fences (%d).\n", r);
2188 r = radeon_fence_driver_start_ring(rdev, CAYMAN_RING_TYPE_CP2_INDEX);
2190 dev_err(rdev->dev, "failed initializing CP fences (%d).\n", r);
2194 r = radeon_fence_driver_start_ring(rdev, R600_RING_TYPE_DMA_INDEX);
2196 dev_err(rdev->dev, "failed initializing DMA fences (%d).\n", r);
2200 r = radeon_fence_driver_start_ring(rdev, CAYMAN_RING_TYPE_DMA1_INDEX);
2202 dev_err(rdev->dev, "failed initializing DMA fences (%d).\n", r);
2207 if (!rdev->irq.installed) {
2208 r = radeon_irq_kms_init(rdev);
2213 r = r600_irq_init(rdev);
2215 DRM_ERROR("radeon: IH init failed (%d).\n", r);
2216 radeon_irq_kms_fini(rdev);
2219 evergreen_irq_set(rdev);
2221 r = radeon_ring_init(rdev, ring, ring->ring_size, RADEON_WB_CP_RPTR_OFFSET,
2226 ring = &rdev->ring[R600_RING_TYPE_DMA_INDEX];
2227 r = radeon_ring_init(rdev, ring, ring->ring_size, R600_WB_DMA_RPTR_OFFSET,
2228 DMA_PACKET(DMA_PACKET_NOP, 0, 0, 0));
2232 ring = &rdev->ring[CAYMAN_RING_TYPE_DMA1_INDEX];
2233 r = radeon_ring_init(rdev, ring, ring->ring_size, CAYMAN_WB_DMA1_RPTR_OFFSET,
2234 DMA_PACKET(DMA_PACKET_NOP, 0, 0, 0));
2238 r = cayman_cp_load_microcode(rdev);
2241 r = cayman_cp_resume(rdev);
2245 r = cayman_dma_resume(rdev);
2249 cayman_uvd_resume(rdev);
2250 cayman_vce_resume(rdev);
2252 r = radeon_ib_pool_init(rdev);
2254 dev_err(rdev->dev, "IB initialization failed (%d).\n", r);
2258 r = radeon_vm_manager_init(rdev);
2260 dev_err(rdev->dev, "vm manager initialization failed (%d).\n", r);
2264 r = radeon_audio_init(rdev);
2271 int cayman_resume(struct radeon_device *rdev)
2275 /* Do not reset GPU before posting, on rv770 hw unlike on r500 hw,
2276 * posting will perform necessary task to bring back GPU into good
2280 atom_asic_init(rdev->mode_info.atom_context);
2282 /* init golden registers */
2283 ni_init_golden_registers(rdev);
2285 if (rdev->pm.pm_method == PM_METHOD_DPM)
2286 radeon_pm_resume(rdev);
2288 rdev->accel_working = true;
2289 r = cayman_startup(rdev);
2291 DRM_ERROR("cayman startup failed on resume\n");
2292 rdev->accel_working = false;
2298 int cayman_suspend(struct radeon_device *rdev)
2300 radeon_pm_suspend(rdev);
2301 radeon_audio_fini(rdev);
2302 radeon_vm_manager_fini(rdev);
2303 cayman_cp_enable(rdev, false);
2304 cayman_dma_stop(rdev);
2305 if (rdev->has_uvd) {
2306 radeon_uvd_suspend(rdev);
2307 uvd_v1_0_fini(rdev);
2309 evergreen_irq_suspend(rdev);
2310 radeon_wb_disable(rdev);
2311 cayman_pcie_gart_disable(rdev);
2315 /* Plan is to move initialization in that function and use
2316 * helper function so that radeon_device_init pretty much
2317 * do nothing more than calling asic specific function. This
2318 * should also allow to remove a bunch of callback function
2321 int cayman_init(struct radeon_device *rdev)
2323 struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
2327 if (!radeon_get_bios(rdev)) {
2328 if (ASIC_IS_AVIVO(rdev))
2331 /* Must be an ATOMBIOS */
2332 if (!rdev->is_atom_bios) {
2333 dev_err(rdev->dev, "Expecting atombios for cayman GPU\n");
2336 r = radeon_atombios_init(rdev);
2340 /* Post card if necessary */
2341 if (!radeon_card_posted(rdev)) {
2343 dev_err(rdev->dev, "Card not posted and no BIOS - ignoring\n");
2346 DRM_INFO("GPU not posted. posting now...\n");
2347 atom_asic_init(rdev->mode_info.atom_context);
2349 /* init golden registers */
2350 ni_init_golden_registers(rdev);
2351 /* Initialize scratch registers */
2352 r600_scratch_init(rdev);
2353 /* Initialize surface registers */
2354 radeon_surface_init(rdev);
2355 /* Initialize clocks */
2356 radeon_get_clock_info(rdev->ddev);
2358 radeon_fence_driver_init(rdev);
2359 /* initialize memory controller */
2360 r = evergreen_mc_init(rdev);
2363 /* Memory manager */
2364 r = radeon_bo_init(rdev);
2368 if (rdev->flags & RADEON_IS_IGP) {
2369 if (!rdev->me_fw || !rdev->pfp_fw || !rdev->rlc_fw) {
2370 r = ni_init_microcode(rdev);
2372 DRM_ERROR("Failed to load firmware!\n");
2377 if (!rdev->me_fw || !rdev->pfp_fw || !rdev->rlc_fw || !rdev->mc_fw) {
2378 r = ni_init_microcode(rdev);
2380 DRM_ERROR("Failed to load firmware!\n");
2386 /* Initialize power management */
2387 radeon_pm_init(rdev);
2389 ring->ring_obj = NULL;
2390 r600_ring_init(rdev, ring, 1024 * 1024);
2392 ring = &rdev->ring[R600_RING_TYPE_DMA_INDEX];
2393 ring->ring_obj = NULL;
2394 r600_ring_init(rdev, ring, 64 * 1024);
2396 ring = &rdev->ring[CAYMAN_RING_TYPE_DMA1_INDEX];
2397 ring->ring_obj = NULL;
2398 r600_ring_init(rdev, ring, 64 * 1024);
2400 cayman_uvd_init(rdev);
2401 cayman_vce_init(rdev);
2403 rdev->ih.ring_obj = NULL;
2404 r600_ih_ring_init(rdev, 64 * 1024);
2406 r = r600_pcie_gart_init(rdev);
2410 rdev->accel_working = true;
2411 r = cayman_startup(rdev);
2413 dev_err(rdev->dev, "disabling GPU acceleration\n");
2414 cayman_cp_fini(rdev);
2415 cayman_dma_fini(rdev);
2416 r600_irq_fini(rdev);
2417 if (rdev->flags & RADEON_IS_IGP)
2418 sumo_rlc_fini(rdev);
2419 radeon_wb_fini(rdev);
2420 radeon_ib_pool_fini(rdev);
2421 radeon_vm_manager_fini(rdev);
2422 radeon_irq_kms_fini(rdev);
2423 cayman_pcie_gart_fini(rdev);
2424 rdev->accel_working = false;
2427 /* Don't start up if the MC ucode is missing.
2428 * The default clocks and voltages before the MC ucode
2429 * is loaded are not suffient for advanced operations.
2431 * We can skip this check for TN, because there is no MC
2434 if (!rdev->mc_fw && !(rdev->flags & RADEON_IS_IGP)) {
2435 DRM_ERROR("radeon: MC ucode required for NI+.\n");
2442 void cayman_fini(struct radeon_device *rdev)
2444 radeon_pm_fini(rdev);
2445 cayman_cp_fini(rdev);
2446 cayman_dma_fini(rdev);
2447 r600_irq_fini(rdev);
2448 if (rdev->flags & RADEON_IS_IGP)
2449 sumo_rlc_fini(rdev);
2450 radeon_wb_fini(rdev);
2451 radeon_vm_manager_fini(rdev);
2452 radeon_ib_pool_fini(rdev);
2453 radeon_irq_kms_fini(rdev);
2454 uvd_v1_0_fini(rdev);
2455 radeon_uvd_fini(rdev);
2457 radeon_vce_fini(rdev);
2458 cayman_pcie_gart_fini(rdev);
2459 r600_vram_scratch_fini(rdev);
2460 radeon_gem_fini(rdev);
2461 radeon_fence_driver_fini(rdev);
2462 radeon_bo_fini(rdev);
2463 radeon_atombios_fini(rdev);
2471 int cayman_vm_init(struct radeon_device *rdev)
2474 rdev->vm_manager.nvm = 8;
2475 /* base offset of vram pages */
2476 if (rdev->flags & RADEON_IS_IGP) {
2477 u64 tmp = RREG32(FUS_MC_VM_FB_OFFSET);
2479 rdev->vm_manager.vram_base_offset = tmp;
2481 rdev->vm_manager.vram_base_offset = 0;
2485 void cayman_vm_fini(struct radeon_device *rdev)
2490 * cayman_vm_decode_fault - print human readable fault info
2492 * @rdev: radeon_device pointer
2493 * @status: VM_CONTEXT1_PROTECTION_FAULT_STATUS register value
2494 * @addr: VM_CONTEXT1_PROTECTION_FAULT_ADDR register value
2496 * Print human readable fault information (cayman/TN).
2498 void cayman_vm_decode_fault(struct radeon_device *rdev,
2499 u32 status, u32 addr)
2501 u32 mc_id = (status & MEMORY_CLIENT_ID_MASK) >> MEMORY_CLIENT_ID_SHIFT;
2502 u32 vmid = (status & FAULT_VMID_MASK) >> FAULT_VMID_SHIFT;
2503 u32 protections = (status & PROTECTIONS_MASK) >> PROTECTIONS_SHIFT;
2595 block = "TC_TFETCH";
2605 block = "TC_VFETCH";
2644 printk("VM fault (0x%02x, vmid %d) at page %u, %s from %s (%d)\n",
2645 protections, vmid, addr,
2646 (status & MEMORY_CLIENT_RW_MASK) ? "write" : "read",
2651 * cayman_vm_flush - vm flush using the CP
2653 * Update the page table base and flush the VM TLB
2654 * using the CP (cayman-si).
2656 void cayman_vm_flush(struct radeon_device *rdev, struct radeon_ring *ring,
2657 unsigned vm_id, uint64_t pd_addr)
2659 radeon_ring_write(ring, PACKET0(VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + (vm_id << 2), 0));
2660 radeon_ring_write(ring, pd_addr >> 12);
2662 /* flush hdp cache */
2663 radeon_ring_write(ring, PACKET0(HDP_MEM_COHERENCY_FLUSH_CNTL, 0));
2664 radeon_ring_write(ring, 0x1);
2666 /* bits 0-7 are the VM contexts0-7 */
2667 radeon_ring_write(ring, PACKET0(VM_INVALIDATE_REQUEST, 0));
2668 radeon_ring_write(ring, 1 << vm_id);
2670 /* wait for the invalidate to complete */
2671 radeon_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
2672 radeon_ring_write(ring, (WAIT_REG_MEM_FUNCTION(0) | /* always */
2673 WAIT_REG_MEM_ENGINE(0))); /* me */
2674 radeon_ring_write(ring, VM_INVALIDATE_REQUEST >> 2);
2675 radeon_ring_write(ring, 0);
2676 radeon_ring_write(ring, 0); /* ref */
2677 radeon_ring_write(ring, 0); /* mask */
2678 radeon_ring_write(ring, 0x20); /* poll interval */
2680 /* sync PFP to ME, otherwise we might get invalid PFP reads */
2681 radeon_ring_write(ring, PACKET3(PACKET3_PFP_SYNC_ME, 0));
2682 radeon_ring_write(ring, 0x0);
2685 int tn_set_vce_clocks(struct radeon_device *rdev, u32 evclk, u32 ecclk)
2687 struct atom_clock_dividers dividers;
2690 r = radeon_atom_get_clock_dividers(rdev, COMPUTE_ENGINE_PLL_PARAM,
2691 ecclk, false, ÷rs);
2695 for (i = 0; i < 100; i++) {
2696 if (RREG32(CG_ECLK_STATUS) & ECLK_STATUS)
2703 WREG32_P(CG_ECLK_CNTL, dividers.post_div, ~(ECLK_DIR_CNTL_EN|ECLK_DIVIDER_MASK));
2705 for (i = 0; i < 100; i++) {
2706 if (RREG32(CG_ECLK_STATUS) & ECLK_STATUS)