2 * Copyright 2012 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
24 #include <linux/firmware.h>
25 #include <linux/slab.h>
26 #include <linux/module.h>
29 #include "radeon_asic.h"
32 #include "cik_blit_shaders.h"
33 #include "radeon_ucode.h"
34 #include "clearstate_ci.h"
36 MODULE_FIRMWARE("radeon/BONAIRE_pfp.bin");
37 MODULE_FIRMWARE("radeon/BONAIRE_me.bin");
38 MODULE_FIRMWARE("radeon/BONAIRE_ce.bin");
39 MODULE_FIRMWARE("radeon/BONAIRE_mec.bin");
40 MODULE_FIRMWARE("radeon/BONAIRE_mc.bin");
41 MODULE_FIRMWARE("radeon/BONAIRE_rlc.bin");
42 MODULE_FIRMWARE("radeon/BONAIRE_sdma.bin");
43 MODULE_FIRMWARE("radeon/BONAIRE_smc.bin");
44 MODULE_FIRMWARE("radeon/HAWAII_pfp.bin");
45 MODULE_FIRMWARE("radeon/HAWAII_me.bin");
46 MODULE_FIRMWARE("radeon/HAWAII_ce.bin");
47 MODULE_FIRMWARE("radeon/HAWAII_mec.bin");
48 MODULE_FIRMWARE("radeon/HAWAII_mc.bin");
49 MODULE_FIRMWARE("radeon/HAWAII_rlc.bin");
50 MODULE_FIRMWARE("radeon/HAWAII_sdma.bin");
51 MODULE_FIRMWARE("radeon/HAWAII_smc.bin");
52 MODULE_FIRMWARE("radeon/KAVERI_pfp.bin");
53 MODULE_FIRMWARE("radeon/KAVERI_me.bin");
54 MODULE_FIRMWARE("radeon/KAVERI_ce.bin");
55 MODULE_FIRMWARE("radeon/KAVERI_mec.bin");
56 MODULE_FIRMWARE("radeon/KAVERI_rlc.bin");
57 MODULE_FIRMWARE("radeon/KAVERI_sdma.bin");
58 MODULE_FIRMWARE("radeon/KABINI_pfp.bin");
59 MODULE_FIRMWARE("radeon/KABINI_me.bin");
60 MODULE_FIRMWARE("radeon/KABINI_ce.bin");
61 MODULE_FIRMWARE("radeon/KABINI_mec.bin");
62 MODULE_FIRMWARE("radeon/KABINI_rlc.bin");
63 MODULE_FIRMWARE("radeon/KABINI_sdma.bin");
65 extern int r600_ih_ring_alloc(struct radeon_device *rdev);
66 extern void r600_ih_ring_fini(struct radeon_device *rdev);
67 extern void evergreen_mc_stop(struct radeon_device *rdev, struct evergreen_mc_save *save);
68 extern void evergreen_mc_resume(struct radeon_device *rdev, struct evergreen_mc_save *save);
69 extern bool evergreen_is_display_hung(struct radeon_device *rdev);
70 extern void sumo_rlc_fini(struct radeon_device *rdev);
71 extern int sumo_rlc_init(struct radeon_device *rdev);
72 extern void si_vram_gtt_location(struct radeon_device *rdev, struct radeon_mc *mc);
73 extern void si_rlc_reset(struct radeon_device *rdev);
74 extern void si_init_uvd_internal_cg(struct radeon_device *rdev);
75 extern int cik_sdma_resume(struct radeon_device *rdev);
76 extern void cik_sdma_enable(struct radeon_device *rdev, bool enable);
77 extern void cik_sdma_fini(struct radeon_device *rdev);
78 static void cik_rlc_stop(struct radeon_device *rdev);
79 static void cik_pcie_gen3_enable(struct radeon_device *rdev);
80 static void cik_program_aspm(struct radeon_device *rdev);
81 static void cik_init_pg(struct radeon_device *rdev);
82 static void cik_init_cg(struct radeon_device *rdev);
83 static void cik_fini_pg(struct radeon_device *rdev);
84 static void cik_fini_cg(struct radeon_device *rdev);
85 static void cik_enable_gui_idle_interrupt(struct radeon_device *rdev,
88 /* get temperature in millidegrees */
89 int ci_get_temp(struct radeon_device *rdev)
94 temp = (RREG32_SMC(CG_MULT_THERMAL_STATUS) & CTF_TEMP_MASK) >>
100 actual_temp = temp & 0x1ff;
102 actual_temp = actual_temp * 1000;
107 /* get temperature in millidegrees */
108 int kv_get_temp(struct radeon_device *rdev)
113 temp = RREG32_SMC(0xC0300E0C);
116 actual_temp = (temp / 8) - 49;
120 actual_temp = actual_temp * 1000;
126 * Indirect registers accessor
128 u32 cik_pciep_rreg(struct radeon_device *rdev, u32 reg)
133 spin_lock_irqsave(&rdev->pciep_idx_lock, flags);
134 WREG32(PCIE_INDEX, reg);
135 (void)RREG32(PCIE_INDEX);
136 r = RREG32(PCIE_DATA);
137 spin_unlock_irqrestore(&rdev->pciep_idx_lock, flags);
141 void cik_pciep_wreg(struct radeon_device *rdev, u32 reg, u32 v)
145 spin_lock_irqsave(&rdev->pciep_idx_lock, flags);
146 WREG32(PCIE_INDEX, reg);
147 (void)RREG32(PCIE_INDEX);
148 WREG32(PCIE_DATA, v);
149 (void)RREG32(PCIE_DATA);
150 spin_unlock_irqrestore(&rdev->pciep_idx_lock, flags);
153 static const u32 spectre_rlc_save_restore_register_list[] =
155 (0x0e00 << 16) | (0xc12c >> 2),
157 (0x0e00 << 16) | (0xc140 >> 2),
159 (0x0e00 << 16) | (0xc150 >> 2),
161 (0x0e00 << 16) | (0xc15c >> 2),
163 (0x0e00 << 16) | (0xc168 >> 2),
165 (0x0e00 << 16) | (0xc170 >> 2),
167 (0x0e00 << 16) | (0xc178 >> 2),
169 (0x0e00 << 16) | (0xc204 >> 2),
171 (0x0e00 << 16) | (0xc2b4 >> 2),
173 (0x0e00 << 16) | (0xc2b8 >> 2),
175 (0x0e00 << 16) | (0xc2bc >> 2),
177 (0x0e00 << 16) | (0xc2c0 >> 2),
179 (0x0e00 << 16) | (0x8228 >> 2),
181 (0x0e00 << 16) | (0x829c >> 2),
183 (0x0e00 << 16) | (0x869c >> 2),
185 (0x0600 << 16) | (0x98f4 >> 2),
187 (0x0e00 << 16) | (0x98f8 >> 2),
189 (0x0e00 << 16) | (0x9900 >> 2),
191 (0x0e00 << 16) | (0xc260 >> 2),
193 (0x0e00 << 16) | (0x90e8 >> 2),
195 (0x0e00 << 16) | (0x3c000 >> 2),
197 (0x0e00 << 16) | (0x3c00c >> 2),
199 (0x0e00 << 16) | (0x8c1c >> 2),
201 (0x0e00 << 16) | (0x9700 >> 2),
203 (0x0e00 << 16) | (0xcd20 >> 2),
205 (0x4e00 << 16) | (0xcd20 >> 2),
207 (0x5e00 << 16) | (0xcd20 >> 2),
209 (0x6e00 << 16) | (0xcd20 >> 2),
211 (0x7e00 << 16) | (0xcd20 >> 2),
213 (0x8e00 << 16) | (0xcd20 >> 2),
215 (0x9e00 << 16) | (0xcd20 >> 2),
217 (0xae00 << 16) | (0xcd20 >> 2),
219 (0xbe00 << 16) | (0xcd20 >> 2),
221 (0x0e00 << 16) | (0x89bc >> 2),
223 (0x0e00 << 16) | (0x8900 >> 2),
226 (0x0e00 << 16) | (0xc130 >> 2),
228 (0x0e00 << 16) | (0xc134 >> 2),
230 (0x0e00 << 16) | (0xc1fc >> 2),
232 (0x0e00 << 16) | (0xc208 >> 2),
234 (0x0e00 << 16) | (0xc264 >> 2),
236 (0x0e00 << 16) | (0xc268 >> 2),
238 (0x0e00 << 16) | (0xc26c >> 2),
240 (0x0e00 << 16) | (0xc270 >> 2),
242 (0x0e00 << 16) | (0xc274 >> 2),
244 (0x0e00 << 16) | (0xc278 >> 2),
246 (0x0e00 << 16) | (0xc27c >> 2),
248 (0x0e00 << 16) | (0xc280 >> 2),
250 (0x0e00 << 16) | (0xc284 >> 2),
252 (0x0e00 << 16) | (0xc288 >> 2),
254 (0x0e00 << 16) | (0xc28c >> 2),
256 (0x0e00 << 16) | (0xc290 >> 2),
258 (0x0e00 << 16) | (0xc294 >> 2),
260 (0x0e00 << 16) | (0xc298 >> 2),
262 (0x0e00 << 16) | (0xc29c >> 2),
264 (0x0e00 << 16) | (0xc2a0 >> 2),
266 (0x0e00 << 16) | (0xc2a4 >> 2),
268 (0x0e00 << 16) | (0xc2a8 >> 2),
270 (0x0e00 << 16) | (0xc2ac >> 2),
272 (0x0e00 << 16) | (0xc2b0 >> 2),
274 (0x0e00 << 16) | (0x301d0 >> 2),
276 (0x0e00 << 16) | (0x30238 >> 2),
278 (0x0e00 << 16) | (0x30250 >> 2),
280 (0x0e00 << 16) | (0x30254 >> 2),
282 (0x0e00 << 16) | (0x30258 >> 2),
284 (0x0e00 << 16) | (0x3025c >> 2),
286 (0x4e00 << 16) | (0xc900 >> 2),
288 (0x5e00 << 16) | (0xc900 >> 2),
290 (0x6e00 << 16) | (0xc900 >> 2),
292 (0x7e00 << 16) | (0xc900 >> 2),
294 (0x8e00 << 16) | (0xc900 >> 2),
296 (0x9e00 << 16) | (0xc900 >> 2),
298 (0xae00 << 16) | (0xc900 >> 2),
300 (0xbe00 << 16) | (0xc900 >> 2),
302 (0x4e00 << 16) | (0xc904 >> 2),
304 (0x5e00 << 16) | (0xc904 >> 2),
306 (0x6e00 << 16) | (0xc904 >> 2),
308 (0x7e00 << 16) | (0xc904 >> 2),
310 (0x8e00 << 16) | (0xc904 >> 2),
312 (0x9e00 << 16) | (0xc904 >> 2),
314 (0xae00 << 16) | (0xc904 >> 2),
316 (0xbe00 << 16) | (0xc904 >> 2),
318 (0x4e00 << 16) | (0xc908 >> 2),
320 (0x5e00 << 16) | (0xc908 >> 2),
322 (0x6e00 << 16) | (0xc908 >> 2),
324 (0x7e00 << 16) | (0xc908 >> 2),
326 (0x8e00 << 16) | (0xc908 >> 2),
328 (0x9e00 << 16) | (0xc908 >> 2),
330 (0xae00 << 16) | (0xc908 >> 2),
332 (0xbe00 << 16) | (0xc908 >> 2),
334 (0x4e00 << 16) | (0xc90c >> 2),
336 (0x5e00 << 16) | (0xc90c >> 2),
338 (0x6e00 << 16) | (0xc90c >> 2),
340 (0x7e00 << 16) | (0xc90c >> 2),
342 (0x8e00 << 16) | (0xc90c >> 2),
344 (0x9e00 << 16) | (0xc90c >> 2),
346 (0xae00 << 16) | (0xc90c >> 2),
348 (0xbe00 << 16) | (0xc90c >> 2),
350 (0x4e00 << 16) | (0xc910 >> 2),
352 (0x5e00 << 16) | (0xc910 >> 2),
354 (0x6e00 << 16) | (0xc910 >> 2),
356 (0x7e00 << 16) | (0xc910 >> 2),
358 (0x8e00 << 16) | (0xc910 >> 2),
360 (0x9e00 << 16) | (0xc910 >> 2),
362 (0xae00 << 16) | (0xc910 >> 2),
364 (0xbe00 << 16) | (0xc910 >> 2),
366 (0x0e00 << 16) | (0xc99c >> 2),
368 (0x0e00 << 16) | (0x9834 >> 2),
370 (0x0000 << 16) | (0x30f00 >> 2),
372 (0x0001 << 16) | (0x30f00 >> 2),
374 (0x0000 << 16) | (0x30f04 >> 2),
376 (0x0001 << 16) | (0x30f04 >> 2),
378 (0x0000 << 16) | (0x30f08 >> 2),
380 (0x0001 << 16) | (0x30f08 >> 2),
382 (0x0000 << 16) | (0x30f0c >> 2),
384 (0x0001 << 16) | (0x30f0c >> 2),
386 (0x0600 << 16) | (0x9b7c >> 2),
388 (0x0e00 << 16) | (0x8a14 >> 2),
390 (0x0e00 << 16) | (0x8a18 >> 2),
392 (0x0600 << 16) | (0x30a00 >> 2),
394 (0x0e00 << 16) | (0x8bf0 >> 2),
396 (0x0e00 << 16) | (0x8bcc >> 2),
398 (0x0e00 << 16) | (0x8b24 >> 2),
400 (0x0e00 << 16) | (0x30a04 >> 2),
402 (0x0600 << 16) | (0x30a10 >> 2),
404 (0x0600 << 16) | (0x30a14 >> 2),
406 (0x0600 << 16) | (0x30a18 >> 2),
408 (0x0600 << 16) | (0x30a2c >> 2),
410 (0x0e00 << 16) | (0xc700 >> 2),
412 (0x0e00 << 16) | (0xc704 >> 2),
414 (0x0e00 << 16) | (0xc708 >> 2),
416 (0x0e00 << 16) | (0xc768 >> 2),
418 (0x0400 << 16) | (0xc770 >> 2),
420 (0x0400 << 16) | (0xc774 >> 2),
422 (0x0400 << 16) | (0xc778 >> 2),
424 (0x0400 << 16) | (0xc77c >> 2),
426 (0x0400 << 16) | (0xc780 >> 2),
428 (0x0400 << 16) | (0xc784 >> 2),
430 (0x0400 << 16) | (0xc788 >> 2),
432 (0x0400 << 16) | (0xc78c >> 2),
434 (0x0400 << 16) | (0xc798 >> 2),
436 (0x0400 << 16) | (0xc79c >> 2),
438 (0x0400 << 16) | (0xc7a0 >> 2),
440 (0x0400 << 16) | (0xc7a4 >> 2),
442 (0x0400 << 16) | (0xc7a8 >> 2),
444 (0x0400 << 16) | (0xc7ac >> 2),
446 (0x0400 << 16) | (0xc7b0 >> 2),
448 (0x0400 << 16) | (0xc7b4 >> 2),
450 (0x0e00 << 16) | (0x9100 >> 2),
452 (0x0e00 << 16) | (0x3c010 >> 2),
454 (0x0e00 << 16) | (0x92a8 >> 2),
456 (0x0e00 << 16) | (0x92ac >> 2),
458 (0x0e00 << 16) | (0x92b4 >> 2),
460 (0x0e00 << 16) | (0x92b8 >> 2),
462 (0x0e00 << 16) | (0x92bc >> 2),
464 (0x0e00 << 16) | (0x92c0 >> 2),
466 (0x0e00 << 16) | (0x92c4 >> 2),
468 (0x0e00 << 16) | (0x92c8 >> 2),
470 (0x0e00 << 16) | (0x92cc >> 2),
472 (0x0e00 << 16) | (0x92d0 >> 2),
474 (0x0e00 << 16) | (0x8c00 >> 2),
476 (0x0e00 << 16) | (0x8c04 >> 2),
478 (0x0e00 << 16) | (0x8c20 >> 2),
480 (0x0e00 << 16) | (0x8c38 >> 2),
482 (0x0e00 << 16) | (0x8c3c >> 2),
484 (0x0e00 << 16) | (0xae00 >> 2),
486 (0x0e00 << 16) | (0x9604 >> 2),
488 (0x0e00 << 16) | (0xac08 >> 2),
490 (0x0e00 << 16) | (0xac0c >> 2),
492 (0x0e00 << 16) | (0xac10 >> 2),
494 (0x0e00 << 16) | (0xac14 >> 2),
496 (0x0e00 << 16) | (0xac58 >> 2),
498 (0x0e00 << 16) | (0xac68 >> 2),
500 (0x0e00 << 16) | (0xac6c >> 2),
502 (0x0e00 << 16) | (0xac70 >> 2),
504 (0x0e00 << 16) | (0xac74 >> 2),
506 (0x0e00 << 16) | (0xac78 >> 2),
508 (0x0e00 << 16) | (0xac7c >> 2),
510 (0x0e00 << 16) | (0xac80 >> 2),
512 (0x0e00 << 16) | (0xac84 >> 2),
514 (0x0e00 << 16) | (0xac88 >> 2),
516 (0x0e00 << 16) | (0xac8c >> 2),
518 (0x0e00 << 16) | (0x970c >> 2),
520 (0x0e00 << 16) | (0x9714 >> 2),
522 (0x0e00 << 16) | (0x9718 >> 2),
524 (0x0e00 << 16) | (0x971c >> 2),
526 (0x0e00 << 16) | (0x31068 >> 2),
528 (0x4e00 << 16) | (0x31068 >> 2),
530 (0x5e00 << 16) | (0x31068 >> 2),
532 (0x6e00 << 16) | (0x31068 >> 2),
534 (0x7e00 << 16) | (0x31068 >> 2),
536 (0x8e00 << 16) | (0x31068 >> 2),
538 (0x9e00 << 16) | (0x31068 >> 2),
540 (0xae00 << 16) | (0x31068 >> 2),
542 (0xbe00 << 16) | (0x31068 >> 2),
544 (0x0e00 << 16) | (0xcd10 >> 2),
546 (0x0e00 << 16) | (0xcd14 >> 2),
548 (0x0e00 << 16) | (0x88b0 >> 2),
550 (0x0e00 << 16) | (0x88b4 >> 2),
552 (0x0e00 << 16) | (0x88b8 >> 2),
554 (0x0e00 << 16) | (0x88bc >> 2),
556 (0x0400 << 16) | (0x89c0 >> 2),
558 (0x0e00 << 16) | (0x88c4 >> 2),
560 (0x0e00 << 16) | (0x88c8 >> 2),
562 (0x0e00 << 16) | (0x88d0 >> 2),
564 (0x0e00 << 16) | (0x88d4 >> 2),
566 (0x0e00 << 16) | (0x88d8 >> 2),
568 (0x0e00 << 16) | (0x8980 >> 2),
570 (0x0e00 << 16) | (0x30938 >> 2),
572 (0x0e00 << 16) | (0x3093c >> 2),
574 (0x0e00 << 16) | (0x30940 >> 2),
576 (0x0e00 << 16) | (0x89a0 >> 2),
578 (0x0e00 << 16) | (0x30900 >> 2),
580 (0x0e00 << 16) | (0x30904 >> 2),
582 (0x0e00 << 16) | (0x89b4 >> 2),
584 (0x0e00 << 16) | (0x3c210 >> 2),
586 (0x0e00 << 16) | (0x3c214 >> 2),
588 (0x0e00 << 16) | (0x3c218 >> 2),
590 (0x0e00 << 16) | (0x8904 >> 2),
593 (0x0e00 << 16) | (0x8c28 >> 2),
594 (0x0e00 << 16) | (0x8c2c >> 2),
595 (0x0e00 << 16) | (0x8c30 >> 2),
596 (0x0e00 << 16) | (0x8c34 >> 2),
597 (0x0e00 << 16) | (0x9600 >> 2),
600 static const u32 kalindi_rlc_save_restore_register_list[] =
602 (0x0e00 << 16) | (0xc12c >> 2),
604 (0x0e00 << 16) | (0xc140 >> 2),
606 (0x0e00 << 16) | (0xc150 >> 2),
608 (0x0e00 << 16) | (0xc15c >> 2),
610 (0x0e00 << 16) | (0xc168 >> 2),
612 (0x0e00 << 16) | (0xc170 >> 2),
614 (0x0e00 << 16) | (0xc204 >> 2),
616 (0x0e00 << 16) | (0xc2b4 >> 2),
618 (0x0e00 << 16) | (0xc2b8 >> 2),
620 (0x0e00 << 16) | (0xc2bc >> 2),
622 (0x0e00 << 16) | (0xc2c0 >> 2),
624 (0x0e00 << 16) | (0x8228 >> 2),
626 (0x0e00 << 16) | (0x829c >> 2),
628 (0x0e00 << 16) | (0x869c >> 2),
630 (0x0600 << 16) | (0x98f4 >> 2),
632 (0x0e00 << 16) | (0x98f8 >> 2),
634 (0x0e00 << 16) | (0x9900 >> 2),
636 (0x0e00 << 16) | (0xc260 >> 2),
638 (0x0e00 << 16) | (0x90e8 >> 2),
640 (0x0e00 << 16) | (0x3c000 >> 2),
642 (0x0e00 << 16) | (0x3c00c >> 2),
644 (0x0e00 << 16) | (0x8c1c >> 2),
646 (0x0e00 << 16) | (0x9700 >> 2),
648 (0x0e00 << 16) | (0xcd20 >> 2),
650 (0x4e00 << 16) | (0xcd20 >> 2),
652 (0x5e00 << 16) | (0xcd20 >> 2),
654 (0x6e00 << 16) | (0xcd20 >> 2),
656 (0x7e00 << 16) | (0xcd20 >> 2),
658 (0x0e00 << 16) | (0x89bc >> 2),
660 (0x0e00 << 16) | (0x8900 >> 2),
663 (0x0e00 << 16) | (0xc130 >> 2),
665 (0x0e00 << 16) | (0xc134 >> 2),
667 (0x0e00 << 16) | (0xc1fc >> 2),
669 (0x0e00 << 16) | (0xc208 >> 2),
671 (0x0e00 << 16) | (0xc264 >> 2),
673 (0x0e00 << 16) | (0xc268 >> 2),
675 (0x0e00 << 16) | (0xc26c >> 2),
677 (0x0e00 << 16) | (0xc270 >> 2),
679 (0x0e00 << 16) | (0xc274 >> 2),
681 (0x0e00 << 16) | (0xc28c >> 2),
683 (0x0e00 << 16) | (0xc290 >> 2),
685 (0x0e00 << 16) | (0xc294 >> 2),
687 (0x0e00 << 16) | (0xc298 >> 2),
689 (0x0e00 << 16) | (0xc2a0 >> 2),
691 (0x0e00 << 16) | (0xc2a4 >> 2),
693 (0x0e00 << 16) | (0xc2a8 >> 2),
695 (0x0e00 << 16) | (0xc2ac >> 2),
697 (0x0e00 << 16) | (0x301d0 >> 2),
699 (0x0e00 << 16) | (0x30238 >> 2),
701 (0x0e00 << 16) | (0x30250 >> 2),
703 (0x0e00 << 16) | (0x30254 >> 2),
705 (0x0e00 << 16) | (0x30258 >> 2),
707 (0x0e00 << 16) | (0x3025c >> 2),
709 (0x4e00 << 16) | (0xc900 >> 2),
711 (0x5e00 << 16) | (0xc900 >> 2),
713 (0x6e00 << 16) | (0xc900 >> 2),
715 (0x7e00 << 16) | (0xc900 >> 2),
717 (0x4e00 << 16) | (0xc904 >> 2),
719 (0x5e00 << 16) | (0xc904 >> 2),
721 (0x6e00 << 16) | (0xc904 >> 2),
723 (0x7e00 << 16) | (0xc904 >> 2),
725 (0x4e00 << 16) | (0xc908 >> 2),
727 (0x5e00 << 16) | (0xc908 >> 2),
729 (0x6e00 << 16) | (0xc908 >> 2),
731 (0x7e00 << 16) | (0xc908 >> 2),
733 (0x4e00 << 16) | (0xc90c >> 2),
735 (0x5e00 << 16) | (0xc90c >> 2),
737 (0x6e00 << 16) | (0xc90c >> 2),
739 (0x7e00 << 16) | (0xc90c >> 2),
741 (0x4e00 << 16) | (0xc910 >> 2),
743 (0x5e00 << 16) | (0xc910 >> 2),
745 (0x6e00 << 16) | (0xc910 >> 2),
747 (0x7e00 << 16) | (0xc910 >> 2),
749 (0x0e00 << 16) | (0xc99c >> 2),
751 (0x0e00 << 16) | (0x9834 >> 2),
753 (0x0000 << 16) | (0x30f00 >> 2),
755 (0x0000 << 16) | (0x30f04 >> 2),
757 (0x0000 << 16) | (0x30f08 >> 2),
759 (0x0000 << 16) | (0x30f0c >> 2),
761 (0x0600 << 16) | (0x9b7c >> 2),
763 (0x0e00 << 16) | (0x8a14 >> 2),
765 (0x0e00 << 16) | (0x8a18 >> 2),
767 (0x0600 << 16) | (0x30a00 >> 2),
769 (0x0e00 << 16) | (0x8bf0 >> 2),
771 (0x0e00 << 16) | (0x8bcc >> 2),
773 (0x0e00 << 16) | (0x8b24 >> 2),
775 (0x0e00 << 16) | (0x30a04 >> 2),
777 (0x0600 << 16) | (0x30a10 >> 2),
779 (0x0600 << 16) | (0x30a14 >> 2),
781 (0x0600 << 16) | (0x30a18 >> 2),
783 (0x0600 << 16) | (0x30a2c >> 2),
785 (0x0e00 << 16) | (0xc700 >> 2),
787 (0x0e00 << 16) | (0xc704 >> 2),
789 (0x0e00 << 16) | (0xc708 >> 2),
791 (0x0e00 << 16) | (0xc768 >> 2),
793 (0x0400 << 16) | (0xc770 >> 2),
795 (0x0400 << 16) | (0xc774 >> 2),
797 (0x0400 << 16) | (0xc798 >> 2),
799 (0x0400 << 16) | (0xc79c >> 2),
801 (0x0e00 << 16) | (0x9100 >> 2),
803 (0x0e00 << 16) | (0x3c010 >> 2),
805 (0x0e00 << 16) | (0x8c00 >> 2),
807 (0x0e00 << 16) | (0x8c04 >> 2),
809 (0x0e00 << 16) | (0x8c20 >> 2),
811 (0x0e00 << 16) | (0x8c38 >> 2),
813 (0x0e00 << 16) | (0x8c3c >> 2),
815 (0x0e00 << 16) | (0xae00 >> 2),
817 (0x0e00 << 16) | (0x9604 >> 2),
819 (0x0e00 << 16) | (0xac08 >> 2),
821 (0x0e00 << 16) | (0xac0c >> 2),
823 (0x0e00 << 16) | (0xac10 >> 2),
825 (0x0e00 << 16) | (0xac14 >> 2),
827 (0x0e00 << 16) | (0xac58 >> 2),
829 (0x0e00 << 16) | (0xac68 >> 2),
831 (0x0e00 << 16) | (0xac6c >> 2),
833 (0x0e00 << 16) | (0xac70 >> 2),
835 (0x0e00 << 16) | (0xac74 >> 2),
837 (0x0e00 << 16) | (0xac78 >> 2),
839 (0x0e00 << 16) | (0xac7c >> 2),
841 (0x0e00 << 16) | (0xac80 >> 2),
843 (0x0e00 << 16) | (0xac84 >> 2),
845 (0x0e00 << 16) | (0xac88 >> 2),
847 (0x0e00 << 16) | (0xac8c >> 2),
849 (0x0e00 << 16) | (0x970c >> 2),
851 (0x0e00 << 16) | (0x9714 >> 2),
853 (0x0e00 << 16) | (0x9718 >> 2),
855 (0x0e00 << 16) | (0x971c >> 2),
857 (0x0e00 << 16) | (0x31068 >> 2),
859 (0x4e00 << 16) | (0x31068 >> 2),
861 (0x5e00 << 16) | (0x31068 >> 2),
863 (0x6e00 << 16) | (0x31068 >> 2),
865 (0x7e00 << 16) | (0x31068 >> 2),
867 (0x0e00 << 16) | (0xcd10 >> 2),
869 (0x0e00 << 16) | (0xcd14 >> 2),
871 (0x0e00 << 16) | (0x88b0 >> 2),
873 (0x0e00 << 16) | (0x88b4 >> 2),
875 (0x0e00 << 16) | (0x88b8 >> 2),
877 (0x0e00 << 16) | (0x88bc >> 2),
879 (0x0400 << 16) | (0x89c0 >> 2),
881 (0x0e00 << 16) | (0x88c4 >> 2),
883 (0x0e00 << 16) | (0x88c8 >> 2),
885 (0x0e00 << 16) | (0x88d0 >> 2),
887 (0x0e00 << 16) | (0x88d4 >> 2),
889 (0x0e00 << 16) | (0x88d8 >> 2),
891 (0x0e00 << 16) | (0x8980 >> 2),
893 (0x0e00 << 16) | (0x30938 >> 2),
895 (0x0e00 << 16) | (0x3093c >> 2),
897 (0x0e00 << 16) | (0x30940 >> 2),
899 (0x0e00 << 16) | (0x89a0 >> 2),
901 (0x0e00 << 16) | (0x30900 >> 2),
903 (0x0e00 << 16) | (0x30904 >> 2),
905 (0x0e00 << 16) | (0x89b4 >> 2),
907 (0x0e00 << 16) | (0x3e1fc >> 2),
909 (0x0e00 << 16) | (0x3c210 >> 2),
911 (0x0e00 << 16) | (0x3c214 >> 2),
913 (0x0e00 << 16) | (0x3c218 >> 2),
915 (0x0e00 << 16) | (0x8904 >> 2),
918 (0x0e00 << 16) | (0x8c28 >> 2),
919 (0x0e00 << 16) | (0x8c2c >> 2),
920 (0x0e00 << 16) | (0x8c30 >> 2),
921 (0x0e00 << 16) | (0x8c34 >> 2),
922 (0x0e00 << 16) | (0x9600 >> 2),
925 static const u32 bonaire_golden_spm_registers[] =
927 0x30800, 0xe0ffffff, 0xe0000000
930 static const u32 bonaire_golden_common_registers[] =
932 0xc770, 0xffffffff, 0x00000800,
933 0xc774, 0xffffffff, 0x00000800,
934 0xc798, 0xffffffff, 0x00007fbf,
935 0xc79c, 0xffffffff, 0x00007faf
938 static const u32 bonaire_golden_registers[] =
940 0x3354, 0x00000333, 0x00000333,
941 0x3350, 0x000c0fc0, 0x00040200,
942 0x9a10, 0x00010000, 0x00058208,
943 0x3c000, 0xffff1fff, 0x00140000,
944 0x3c200, 0xfdfc0fff, 0x00000100,
945 0x3c234, 0x40000000, 0x40000200,
946 0x9830, 0xffffffff, 0x00000000,
947 0x9834, 0xf00fffff, 0x00000400,
948 0x9838, 0x0002021c, 0x00020200,
949 0xc78, 0x00000080, 0x00000000,
950 0x5bb0, 0x000000f0, 0x00000070,
951 0x5bc0, 0xf0311fff, 0x80300000,
952 0x98f8, 0x73773777, 0x12010001,
953 0x350c, 0x00810000, 0x408af000,
954 0x7030, 0x31000111, 0x00000011,
955 0x2f48, 0x73773777, 0x12010001,
956 0x220c, 0x00007fb6, 0x0021a1b1,
957 0x2210, 0x00007fb6, 0x002021b1,
958 0x2180, 0x00007fb6, 0x00002191,
959 0x2218, 0x00007fb6, 0x002121b1,
960 0x221c, 0x00007fb6, 0x002021b1,
961 0x21dc, 0x00007fb6, 0x00002191,
962 0x21e0, 0x00007fb6, 0x00002191,
963 0x3628, 0x0000003f, 0x0000000a,
964 0x362c, 0x0000003f, 0x0000000a,
965 0x2ae4, 0x00073ffe, 0x000022a2,
966 0x240c, 0x000007ff, 0x00000000,
967 0x8a14, 0xf000003f, 0x00000007,
968 0x8bf0, 0x00002001, 0x00000001,
969 0x8b24, 0xffffffff, 0x00ffffff,
970 0x30a04, 0x0000ff0f, 0x00000000,
971 0x28a4c, 0x07ffffff, 0x06000000,
972 0x4d8, 0x00000fff, 0x00000100,
973 0x3e78, 0x00000001, 0x00000002,
974 0x9100, 0x03000000, 0x0362c688,
975 0x8c00, 0x000000ff, 0x00000001,
976 0xe40, 0x00001fff, 0x00001fff,
977 0x9060, 0x0000007f, 0x00000020,
978 0x9508, 0x00010000, 0x00010000,
979 0xac14, 0x000003ff, 0x000000f3,
980 0xac0c, 0xffffffff, 0x00001032
983 static const u32 bonaire_mgcg_cgcg_init[] =
985 0xc420, 0xffffffff, 0xfffffffc,
986 0x30800, 0xffffffff, 0xe0000000,
987 0x3c2a0, 0xffffffff, 0x00000100,
988 0x3c208, 0xffffffff, 0x00000100,
989 0x3c2c0, 0xffffffff, 0xc0000100,
990 0x3c2c8, 0xffffffff, 0xc0000100,
991 0x3c2c4, 0xffffffff, 0xc0000100,
992 0x55e4, 0xffffffff, 0x00600100,
993 0x3c280, 0xffffffff, 0x00000100,
994 0x3c214, 0xffffffff, 0x06000100,
995 0x3c220, 0xffffffff, 0x00000100,
996 0x3c218, 0xffffffff, 0x06000100,
997 0x3c204, 0xffffffff, 0x00000100,
998 0x3c2e0, 0xffffffff, 0x00000100,
999 0x3c224, 0xffffffff, 0x00000100,
1000 0x3c200, 0xffffffff, 0x00000100,
1001 0x3c230, 0xffffffff, 0x00000100,
1002 0x3c234, 0xffffffff, 0x00000100,
1003 0x3c250, 0xffffffff, 0x00000100,
1004 0x3c254, 0xffffffff, 0x00000100,
1005 0x3c258, 0xffffffff, 0x00000100,
1006 0x3c25c, 0xffffffff, 0x00000100,
1007 0x3c260, 0xffffffff, 0x00000100,
1008 0x3c27c, 0xffffffff, 0x00000100,
1009 0x3c278, 0xffffffff, 0x00000100,
1010 0x3c210, 0xffffffff, 0x06000100,
1011 0x3c290, 0xffffffff, 0x00000100,
1012 0x3c274, 0xffffffff, 0x00000100,
1013 0x3c2b4, 0xffffffff, 0x00000100,
1014 0x3c2b0, 0xffffffff, 0x00000100,
1015 0x3c270, 0xffffffff, 0x00000100,
1016 0x30800, 0xffffffff, 0xe0000000,
1017 0x3c020, 0xffffffff, 0x00010000,
1018 0x3c024, 0xffffffff, 0x00030002,
1019 0x3c028, 0xffffffff, 0x00040007,
1020 0x3c02c, 0xffffffff, 0x00060005,
1021 0x3c030, 0xffffffff, 0x00090008,
1022 0x3c034, 0xffffffff, 0x00010000,
1023 0x3c038, 0xffffffff, 0x00030002,
1024 0x3c03c, 0xffffffff, 0x00040007,
1025 0x3c040, 0xffffffff, 0x00060005,
1026 0x3c044, 0xffffffff, 0x00090008,
1027 0x3c048, 0xffffffff, 0x00010000,
1028 0x3c04c, 0xffffffff, 0x00030002,
1029 0x3c050, 0xffffffff, 0x00040007,
1030 0x3c054, 0xffffffff, 0x00060005,
1031 0x3c058, 0xffffffff, 0x00090008,
1032 0x3c05c, 0xffffffff, 0x00010000,
1033 0x3c060, 0xffffffff, 0x00030002,
1034 0x3c064, 0xffffffff, 0x00040007,
1035 0x3c068, 0xffffffff, 0x00060005,
1036 0x3c06c, 0xffffffff, 0x00090008,
1037 0x3c070, 0xffffffff, 0x00010000,
1038 0x3c074, 0xffffffff, 0x00030002,
1039 0x3c078, 0xffffffff, 0x00040007,
1040 0x3c07c, 0xffffffff, 0x00060005,
1041 0x3c080, 0xffffffff, 0x00090008,
1042 0x3c084, 0xffffffff, 0x00010000,
1043 0x3c088, 0xffffffff, 0x00030002,
1044 0x3c08c, 0xffffffff, 0x00040007,
1045 0x3c090, 0xffffffff, 0x00060005,
1046 0x3c094, 0xffffffff, 0x00090008,
1047 0x3c098, 0xffffffff, 0x00010000,
1048 0x3c09c, 0xffffffff, 0x00030002,
1049 0x3c0a0, 0xffffffff, 0x00040007,
1050 0x3c0a4, 0xffffffff, 0x00060005,
1051 0x3c0a8, 0xffffffff, 0x00090008,
1052 0x3c000, 0xffffffff, 0x96e00200,
1053 0x8708, 0xffffffff, 0x00900100,
1054 0xc424, 0xffffffff, 0x0020003f,
1055 0x38, 0xffffffff, 0x0140001c,
1056 0x3c, 0x000f0000, 0x000f0000,
1057 0x220, 0xffffffff, 0xC060000C,
1058 0x224, 0xc0000fff, 0x00000100,
1059 0xf90, 0xffffffff, 0x00000100,
1060 0xf98, 0x00000101, 0x00000000,
1061 0x20a8, 0xffffffff, 0x00000104,
1062 0x55e4, 0xff000fff, 0x00000100,
1063 0x30cc, 0xc0000fff, 0x00000104,
1064 0xc1e4, 0x00000001, 0x00000001,
1065 0xd00c, 0xff000ff0, 0x00000100,
1066 0xd80c, 0xff000ff0, 0x00000100
1069 static const u32 spectre_golden_spm_registers[] =
1071 0x30800, 0xe0ffffff, 0xe0000000
1074 static const u32 spectre_golden_common_registers[] =
1076 0xc770, 0xffffffff, 0x00000800,
1077 0xc774, 0xffffffff, 0x00000800,
1078 0xc798, 0xffffffff, 0x00007fbf,
1079 0xc79c, 0xffffffff, 0x00007faf
1082 static const u32 spectre_golden_registers[] =
1084 0x3c000, 0xffff1fff, 0x96940200,
1085 0x3c00c, 0xffff0001, 0xff000000,
1086 0x3c200, 0xfffc0fff, 0x00000100,
1087 0x6ed8, 0x00010101, 0x00010000,
1088 0x9834, 0xf00fffff, 0x00000400,
1089 0x9838, 0xfffffffc, 0x00020200,
1090 0x5bb0, 0x000000f0, 0x00000070,
1091 0x5bc0, 0xf0311fff, 0x80300000,
1092 0x98f8, 0x73773777, 0x12010001,
1093 0x9b7c, 0x00ff0000, 0x00fc0000,
1094 0x2f48, 0x73773777, 0x12010001,
1095 0x8a14, 0xf000003f, 0x00000007,
1096 0x8b24, 0xffffffff, 0x00ffffff,
1097 0x28350, 0x3f3f3fff, 0x00000082,
1098 0x28355, 0x0000003f, 0x00000000,
1099 0x3e78, 0x00000001, 0x00000002,
1100 0x913c, 0xffff03df, 0x00000004,
1101 0xc768, 0x00000008, 0x00000008,
1102 0x8c00, 0x000008ff, 0x00000800,
1103 0x9508, 0x00010000, 0x00010000,
1104 0xac0c, 0xffffffff, 0x54763210,
1105 0x214f8, 0x01ff01ff, 0x00000002,
1106 0x21498, 0x007ff800, 0x00200000,
1107 0x2015c, 0xffffffff, 0x00000f40,
1108 0x30934, 0xffffffff, 0x00000001
1111 static const u32 spectre_mgcg_cgcg_init[] =
1113 0xc420, 0xffffffff, 0xfffffffc,
1114 0x30800, 0xffffffff, 0xe0000000,
1115 0x3c2a0, 0xffffffff, 0x00000100,
1116 0x3c208, 0xffffffff, 0x00000100,
1117 0x3c2c0, 0xffffffff, 0x00000100,
1118 0x3c2c8, 0xffffffff, 0x00000100,
1119 0x3c2c4, 0xffffffff, 0x00000100,
1120 0x55e4, 0xffffffff, 0x00600100,
1121 0x3c280, 0xffffffff, 0x00000100,
1122 0x3c214, 0xffffffff, 0x06000100,
1123 0x3c220, 0xffffffff, 0x00000100,
1124 0x3c218, 0xffffffff, 0x06000100,
1125 0x3c204, 0xffffffff, 0x00000100,
1126 0x3c2e0, 0xffffffff, 0x00000100,
1127 0x3c224, 0xffffffff, 0x00000100,
1128 0x3c200, 0xffffffff, 0x00000100,
1129 0x3c230, 0xffffffff, 0x00000100,
1130 0x3c234, 0xffffffff, 0x00000100,
1131 0x3c250, 0xffffffff, 0x00000100,
1132 0x3c254, 0xffffffff, 0x00000100,
1133 0x3c258, 0xffffffff, 0x00000100,
1134 0x3c25c, 0xffffffff, 0x00000100,
1135 0x3c260, 0xffffffff, 0x00000100,
1136 0x3c27c, 0xffffffff, 0x00000100,
1137 0x3c278, 0xffffffff, 0x00000100,
1138 0x3c210, 0xffffffff, 0x06000100,
1139 0x3c290, 0xffffffff, 0x00000100,
1140 0x3c274, 0xffffffff, 0x00000100,
1141 0x3c2b4, 0xffffffff, 0x00000100,
1142 0x3c2b0, 0xffffffff, 0x00000100,
1143 0x3c270, 0xffffffff, 0x00000100,
1144 0x30800, 0xffffffff, 0xe0000000,
1145 0x3c020, 0xffffffff, 0x00010000,
1146 0x3c024, 0xffffffff, 0x00030002,
1147 0x3c028, 0xffffffff, 0x00040007,
1148 0x3c02c, 0xffffffff, 0x00060005,
1149 0x3c030, 0xffffffff, 0x00090008,
1150 0x3c034, 0xffffffff, 0x00010000,
1151 0x3c038, 0xffffffff, 0x00030002,
1152 0x3c03c, 0xffffffff, 0x00040007,
1153 0x3c040, 0xffffffff, 0x00060005,
1154 0x3c044, 0xffffffff, 0x00090008,
1155 0x3c048, 0xffffffff, 0x00010000,
1156 0x3c04c, 0xffffffff, 0x00030002,
1157 0x3c050, 0xffffffff, 0x00040007,
1158 0x3c054, 0xffffffff, 0x00060005,
1159 0x3c058, 0xffffffff, 0x00090008,
1160 0x3c05c, 0xffffffff, 0x00010000,
1161 0x3c060, 0xffffffff, 0x00030002,
1162 0x3c064, 0xffffffff, 0x00040007,
1163 0x3c068, 0xffffffff, 0x00060005,
1164 0x3c06c, 0xffffffff, 0x00090008,
1165 0x3c070, 0xffffffff, 0x00010000,
1166 0x3c074, 0xffffffff, 0x00030002,
1167 0x3c078, 0xffffffff, 0x00040007,
1168 0x3c07c, 0xffffffff, 0x00060005,
1169 0x3c080, 0xffffffff, 0x00090008,
1170 0x3c084, 0xffffffff, 0x00010000,
1171 0x3c088, 0xffffffff, 0x00030002,
1172 0x3c08c, 0xffffffff, 0x00040007,
1173 0x3c090, 0xffffffff, 0x00060005,
1174 0x3c094, 0xffffffff, 0x00090008,
1175 0x3c098, 0xffffffff, 0x00010000,
1176 0x3c09c, 0xffffffff, 0x00030002,
1177 0x3c0a0, 0xffffffff, 0x00040007,
1178 0x3c0a4, 0xffffffff, 0x00060005,
1179 0x3c0a8, 0xffffffff, 0x00090008,
1180 0x3c0ac, 0xffffffff, 0x00010000,
1181 0x3c0b0, 0xffffffff, 0x00030002,
1182 0x3c0b4, 0xffffffff, 0x00040007,
1183 0x3c0b8, 0xffffffff, 0x00060005,
1184 0x3c0bc, 0xffffffff, 0x00090008,
1185 0x3c000, 0xffffffff, 0x96e00200,
1186 0x8708, 0xffffffff, 0x00900100,
1187 0xc424, 0xffffffff, 0x0020003f,
1188 0x38, 0xffffffff, 0x0140001c,
1189 0x3c, 0x000f0000, 0x000f0000,
1190 0x220, 0xffffffff, 0xC060000C,
1191 0x224, 0xc0000fff, 0x00000100,
1192 0xf90, 0xffffffff, 0x00000100,
1193 0xf98, 0x00000101, 0x00000000,
1194 0x20a8, 0xffffffff, 0x00000104,
1195 0x55e4, 0xff000fff, 0x00000100,
1196 0x30cc, 0xc0000fff, 0x00000104,
1197 0xc1e4, 0x00000001, 0x00000001,
1198 0xd00c, 0xff000ff0, 0x00000100,
1199 0xd80c, 0xff000ff0, 0x00000100
1202 static const u32 kalindi_golden_spm_registers[] =
1204 0x30800, 0xe0ffffff, 0xe0000000
1207 static const u32 kalindi_golden_common_registers[] =
1209 0xc770, 0xffffffff, 0x00000800,
1210 0xc774, 0xffffffff, 0x00000800,
1211 0xc798, 0xffffffff, 0x00007fbf,
1212 0xc79c, 0xffffffff, 0x00007faf
1215 static const u32 kalindi_golden_registers[] =
1217 0x3c000, 0xffffdfff, 0x6e944040,
1218 0x55e4, 0xff607fff, 0xfc000100,
1219 0x3c220, 0xff000fff, 0x00000100,
1220 0x3c224, 0xff000fff, 0x00000100,
1221 0x3c200, 0xfffc0fff, 0x00000100,
1222 0x6ed8, 0x00010101, 0x00010000,
1223 0x9830, 0xffffffff, 0x00000000,
1224 0x9834, 0xf00fffff, 0x00000400,
1225 0x5bb0, 0x000000f0, 0x00000070,
1226 0x5bc0, 0xf0311fff, 0x80300000,
1227 0x98f8, 0x73773777, 0x12010001,
1228 0x98fc, 0xffffffff, 0x00000010,
1229 0x9b7c, 0x00ff0000, 0x00fc0000,
1230 0x8030, 0x00001f0f, 0x0000100a,
1231 0x2f48, 0x73773777, 0x12010001,
1232 0x2408, 0x000fffff, 0x000c007f,
1233 0x8a14, 0xf000003f, 0x00000007,
1234 0x8b24, 0x3fff3fff, 0x00ffcfff,
1235 0x30a04, 0x0000ff0f, 0x00000000,
1236 0x28a4c, 0x07ffffff, 0x06000000,
1237 0x4d8, 0x00000fff, 0x00000100,
1238 0x3e78, 0x00000001, 0x00000002,
1239 0xc768, 0x00000008, 0x00000008,
1240 0x8c00, 0x000000ff, 0x00000003,
1241 0x214f8, 0x01ff01ff, 0x00000002,
1242 0x21498, 0x007ff800, 0x00200000,
1243 0x2015c, 0xffffffff, 0x00000f40,
1244 0x88c4, 0x001f3ae3, 0x00000082,
1245 0x88d4, 0x0000001f, 0x00000010,
1246 0x30934, 0xffffffff, 0x00000000
1249 static const u32 kalindi_mgcg_cgcg_init[] =
1251 0xc420, 0xffffffff, 0xfffffffc,
1252 0x30800, 0xffffffff, 0xe0000000,
1253 0x3c2a0, 0xffffffff, 0x00000100,
1254 0x3c208, 0xffffffff, 0x00000100,
1255 0x3c2c0, 0xffffffff, 0x00000100,
1256 0x3c2c8, 0xffffffff, 0x00000100,
1257 0x3c2c4, 0xffffffff, 0x00000100,
1258 0x55e4, 0xffffffff, 0x00600100,
1259 0x3c280, 0xffffffff, 0x00000100,
1260 0x3c214, 0xffffffff, 0x06000100,
1261 0x3c220, 0xffffffff, 0x00000100,
1262 0x3c218, 0xffffffff, 0x06000100,
1263 0x3c204, 0xffffffff, 0x00000100,
1264 0x3c2e0, 0xffffffff, 0x00000100,
1265 0x3c224, 0xffffffff, 0x00000100,
1266 0x3c200, 0xffffffff, 0x00000100,
1267 0x3c230, 0xffffffff, 0x00000100,
1268 0x3c234, 0xffffffff, 0x00000100,
1269 0x3c250, 0xffffffff, 0x00000100,
1270 0x3c254, 0xffffffff, 0x00000100,
1271 0x3c258, 0xffffffff, 0x00000100,
1272 0x3c25c, 0xffffffff, 0x00000100,
1273 0x3c260, 0xffffffff, 0x00000100,
1274 0x3c27c, 0xffffffff, 0x00000100,
1275 0x3c278, 0xffffffff, 0x00000100,
1276 0x3c210, 0xffffffff, 0x06000100,
1277 0x3c290, 0xffffffff, 0x00000100,
1278 0x3c274, 0xffffffff, 0x00000100,
1279 0x3c2b4, 0xffffffff, 0x00000100,
1280 0x3c2b0, 0xffffffff, 0x00000100,
1281 0x3c270, 0xffffffff, 0x00000100,
1282 0x30800, 0xffffffff, 0xe0000000,
1283 0x3c020, 0xffffffff, 0x00010000,
1284 0x3c024, 0xffffffff, 0x00030002,
1285 0x3c028, 0xffffffff, 0x00040007,
1286 0x3c02c, 0xffffffff, 0x00060005,
1287 0x3c030, 0xffffffff, 0x00090008,
1288 0x3c034, 0xffffffff, 0x00010000,
1289 0x3c038, 0xffffffff, 0x00030002,
1290 0x3c03c, 0xffffffff, 0x00040007,
1291 0x3c040, 0xffffffff, 0x00060005,
1292 0x3c044, 0xffffffff, 0x00090008,
1293 0x3c000, 0xffffffff, 0x96e00200,
1294 0x8708, 0xffffffff, 0x00900100,
1295 0xc424, 0xffffffff, 0x0020003f,
1296 0x38, 0xffffffff, 0x0140001c,
1297 0x3c, 0x000f0000, 0x000f0000,
1298 0x220, 0xffffffff, 0xC060000C,
1299 0x224, 0xc0000fff, 0x00000100,
1300 0x20a8, 0xffffffff, 0x00000104,
1301 0x55e4, 0xff000fff, 0x00000100,
1302 0x30cc, 0xc0000fff, 0x00000104,
1303 0xc1e4, 0x00000001, 0x00000001,
1304 0xd00c, 0xff000ff0, 0x00000100,
1305 0xd80c, 0xff000ff0, 0x00000100
1308 static const u32 hawaii_golden_spm_registers[] =
1310 0x30800, 0xe0ffffff, 0xe0000000
1313 static const u32 hawaii_golden_common_registers[] =
1315 0x30800, 0xffffffff, 0xe0000000,
1316 0x28350, 0xffffffff, 0x3a00161a,
1317 0x28354, 0xffffffff, 0x0000002e,
1318 0x9a10, 0xffffffff, 0x00018208,
1319 0x98f8, 0xffffffff, 0x12011003
1322 static const u32 hawaii_golden_registers[] =
1324 0x3354, 0x00000333, 0x00000333,
1325 0x9a10, 0x00010000, 0x00058208,
1326 0x9830, 0xffffffff, 0x00000000,
1327 0x9834, 0xf00fffff, 0x00000400,
1328 0x9838, 0x0002021c, 0x00020200,
1329 0xc78, 0x00000080, 0x00000000,
1330 0x5bb0, 0x000000f0, 0x00000070,
1331 0x5bc0, 0xf0311fff, 0x80300000,
1332 0x350c, 0x00810000, 0x408af000,
1333 0x7030, 0x31000111, 0x00000011,
1334 0x2f48, 0x73773777, 0x12010001,
1335 0x2120, 0x0000007f, 0x0000001b,
1336 0x21dc, 0x00007fb6, 0x00002191,
1337 0x3628, 0x0000003f, 0x0000000a,
1338 0x362c, 0x0000003f, 0x0000000a,
1339 0x2ae4, 0x00073ffe, 0x000022a2,
1340 0x240c, 0x000007ff, 0x00000000,
1341 0x8bf0, 0x00002001, 0x00000001,
1342 0x8b24, 0xffffffff, 0x00ffffff,
1343 0x30a04, 0x0000ff0f, 0x00000000,
1344 0x28a4c, 0x07ffffff, 0x06000000,
1345 0x3e78, 0x00000001, 0x00000002,
1346 0xc768, 0x00000008, 0x00000008,
1347 0xc770, 0x00000f00, 0x00000800,
1348 0xc774, 0x00000f00, 0x00000800,
1349 0xc798, 0x00ffffff, 0x00ff7fbf,
1350 0xc79c, 0x00ffffff, 0x00ff7faf,
1351 0x8c00, 0x000000ff, 0x00000800,
1352 0xe40, 0x00001fff, 0x00001fff,
1353 0x9060, 0x0000007f, 0x00000020,
1354 0x9508, 0x00010000, 0x00010000,
1355 0xae00, 0x00100000, 0x000ff07c,
1356 0xac14, 0x000003ff, 0x0000000f,
1357 0xac10, 0xffffffff, 0x7564fdec,
1358 0xac0c, 0xffffffff, 0x3120b9a8,
1359 0xac08, 0x20000000, 0x0f9c0000
1362 static const u32 hawaii_mgcg_cgcg_init[] =
1364 0xc420, 0xffffffff, 0xfffffffd,
1365 0x30800, 0xffffffff, 0xe0000000,
1366 0x3c2a0, 0xffffffff, 0x00000100,
1367 0x3c208, 0xffffffff, 0x00000100,
1368 0x3c2c0, 0xffffffff, 0x00000100,
1369 0x3c2c8, 0xffffffff, 0x00000100,
1370 0x3c2c4, 0xffffffff, 0x00000100,
1371 0x55e4, 0xffffffff, 0x00200100,
1372 0x3c280, 0xffffffff, 0x00000100,
1373 0x3c214, 0xffffffff, 0x06000100,
1374 0x3c220, 0xffffffff, 0x00000100,
1375 0x3c218, 0xffffffff, 0x06000100,
1376 0x3c204, 0xffffffff, 0x00000100,
1377 0x3c2e0, 0xffffffff, 0x00000100,
1378 0x3c224, 0xffffffff, 0x00000100,
1379 0x3c200, 0xffffffff, 0x00000100,
1380 0x3c230, 0xffffffff, 0x00000100,
1381 0x3c234, 0xffffffff, 0x00000100,
1382 0x3c250, 0xffffffff, 0x00000100,
1383 0x3c254, 0xffffffff, 0x00000100,
1384 0x3c258, 0xffffffff, 0x00000100,
1385 0x3c25c, 0xffffffff, 0x00000100,
1386 0x3c260, 0xffffffff, 0x00000100,
1387 0x3c27c, 0xffffffff, 0x00000100,
1388 0x3c278, 0xffffffff, 0x00000100,
1389 0x3c210, 0xffffffff, 0x06000100,
1390 0x3c290, 0xffffffff, 0x00000100,
1391 0x3c274, 0xffffffff, 0x00000100,
1392 0x3c2b4, 0xffffffff, 0x00000100,
1393 0x3c2b0, 0xffffffff, 0x00000100,
1394 0x3c270, 0xffffffff, 0x00000100,
1395 0x30800, 0xffffffff, 0xe0000000,
1396 0x3c020, 0xffffffff, 0x00010000,
1397 0x3c024, 0xffffffff, 0x00030002,
1398 0x3c028, 0xffffffff, 0x00040007,
1399 0x3c02c, 0xffffffff, 0x00060005,
1400 0x3c030, 0xffffffff, 0x00090008,
1401 0x3c034, 0xffffffff, 0x00010000,
1402 0x3c038, 0xffffffff, 0x00030002,
1403 0x3c03c, 0xffffffff, 0x00040007,
1404 0x3c040, 0xffffffff, 0x00060005,
1405 0x3c044, 0xffffffff, 0x00090008,
1406 0x3c048, 0xffffffff, 0x00010000,
1407 0x3c04c, 0xffffffff, 0x00030002,
1408 0x3c050, 0xffffffff, 0x00040007,
1409 0x3c054, 0xffffffff, 0x00060005,
1410 0x3c058, 0xffffffff, 0x00090008,
1411 0x3c05c, 0xffffffff, 0x00010000,
1412 0x3c060, 0xffffffff, 0x00030002,
1413 0x3c064, 0xffffffff, 0x00040007,
1414 0x3c068, 0xffffffff, 0x00060005,
1415 0x3c06c, 0xffffffff, 0x00090008,
1416 0x3c070, 0xffffffff, 0x00010000,
1417 0x3c074, 0xffffffff, 0x00030002,
1418 0x3c078, 0xffffffff, 0x00040007,
1419 0x3c07c, 0xffffffff, 0x00060005,
1420 0x3c080, 0xffffffff, 0x00090008,
1421 0x3c084, 0xffffffff, 0x00010000,
1422 0x3c088, 0xffffffff, 0x00030002,
1423 0x3c08c, 0xffffffff, 0x00040007,
1424 0x3c090, 0xffffffff, 0x00060005,
1425 0x3c094, 0xffffffff, 0x00090008,
1426 0x3c098, 0xffffffff, 0x00010000,
1427 0x3c09c, 0xffffffff, 0x00030002,
1428 0x3c0a0, 0xffffffff, 0x00040007,
1429 0x3c0a4, 0xffffffff, 0x00060005,
1430 0x3c0a8, 0xffffffff, 0x00090008,
1431 0x3c0ac, 0xffffffff, 0x00010000,
1432 0x3c0b0, 0xffffffff, 0x00030002,
1433 0x3c0b4, 0xffffffff, 0x00040007,
1434 0x3c0b8, 0xffffffff, 0x00060005,
1435 0x3c0bc, 0xffffffff, 0x00090008,
1436 0x3c0c0, 0xffffffff, 0x00010000,
1437 0x3c0c4, 0xffffffff, 0x00030002,
1438 0x3c0c8, 0xffffffff, 0x00040007,
1439 0x3c0cc, 0xffffffff, 0x00060005,
1440 0x3c0d0, 0xffffffff, 0x00090008,
1441 0x3c0d4, 0xffffffff, 0x00010000,
1442 0x3c0d8, 0xffffffff, 0x00030002,
1443 0x3c0dc, 0xffffffff, 0x00040007,
1444 0x3c0e0, 0xffffffff, 0x00060005,
1445 0x3c0e4, 0xffffffff, 0x00090008,
1446 0x3c0e8, 0xffffffff, 0x00010000,
1447 0x3c0ec, 0xffffffff, 0x00030002,
1448 0x3c0f0, 0xffffffff, 0x00040007,
1449 0x3c0f4, 0xffffffff, 0x00060005,
1450 0x3c0f8, 0xffffffff, 0x00090008,
1451 0xc318, 0xffffffff, 0x00020200,
1452 0x3350, 0xffffffff, 0x00000200,
1453 0x15c0, 0xffffffff, 0x00000400,
1454 0x55e8, 0xffffffff, 0x00000000,
1455 0x2f50, 0xffffffff, 0x00000902,
1456 0x3c000, 0xffffffff, 0x96940200,
1457 0x8708, 0xffffffff, 0x00900100,
1458 0xc424, 0xffffffff, 0x0020003f,
1459 0x38, 0xffffffff, 0x0140001c,
1460 0x3c, 0x000f0000, 0x000f0000,
1461 0x220, 0xffffffff, 0xc060000c,
1462 0x224, 0xc0000fff, 0x00000100,
1463 0xf90, 0xffffffff, 0x00000100,
1464 0xf98, 0x00000101, 0x00000000,
1465 0x20a8, 0xffffffff, 0x00000104,
1466 0x55e4, 0xff000fff, 0x00000100,
1467 0x30cc, 0xc0000fff, 0x00000104,
1468 0xc1e4, 0x00000001, 0x00000001,
1469 0xd00c, 0xff000ff0, 0x00000100,
1470 0xd80c, 0xff000ff0, 0x00000100
1473 static void cik_init_golden_registers(struct radeon_device *rdev)
1475 switch (rdev->family) {
1477 radeon_program_register_sequence(rdev,
1478 bonaire_mgcg_cgcg_init,
1479 (const u32)ARRAY_SIZE(bonaire_mgcg_cgcg_init));
1480 radeon_program_register_sequence(rdev,
1481 bonaire_golden_registers,
1482 (const u32)ARRAY_SIZE(bonaire_golden_registers));
1483 radeon_program_register_sequence(rdev,
1484 bonaire_golden_common_registers,
1485 (const u32)ARRAY_SIZE(bonaire_golden_common_registers));
1486 radeon_program_register_sequence(rdev,
1487 bonaire_golden_spm_registers,
1488 (const u32)ARRAY_SIZE(bonaire_golden_spm_registers));
1491 radeon_program_register_sequence(rdev,
1492 kalindi_mgcg_cgcg_init,
1493 (const u32)ARRAY_SIZE(kalindi_mgcg_cgcg_init));
1494 radeon_program_register_sequence(rdev,
1495 kalindi_golden_registers,
1496 (const u32)ARRAY_SIZE(kalindi_golden_registers));
1497 radeon_program_register_sequence(rdev,
1498 kalindi_golden_common_registers,
1499 (const u32)ARRAY_SIZE(kalindi_golden_common_registers));
1500 radeon_program_register_sequence(rdev,
1501 kalindi_golden_spm_registers,
1502 (const u32)ARRAY_SIZE(kalindi_golden_spm_registers));
1505 radeon_program_register_sequence(rdev,
1506 spectre_mgcg_cgcg_init,
1507 (const u32)ARRAY_SIZE(spectre_mgcg_cgcg_init));
1508 radeon_program_register_sequence(rdev,
1509 spectre_golden_registers,
1510 (const u32)ARRAY_SIZE(spectre_golden_registers));
1511 radeon_program_register_sequence(rdev,
1512 spectre_golden_common_registers,
1513 (const u32)ARRAY_SIZE(spectre_golden_common_registers));
1514 radeon_program_register_sequence(rdev,
1515 spectre_golden_spm_registers,
1516 (const u32)ARRAY_SIZE(spectre_golden_spm_registers));
1519 radeon_program_register_sequence(rdev,
1520 hawaii_mgcg_cgcg_init,
1521 (const u32)ARRAY_SIZE(hawaii_mgcg_cgcg_init));
1522 radeon_program_register_sequence(rdev,
1523 hawaii_golden_registers,
1524 (const u32)ARRAY_SIZE(hawaii_golden_registers));
1525 radeon_program_register_sequence(rdev,
1526 hawaii_golden_common_registers,
1527 (const u32)ARRAY_SIZE(hawaii_golden_common_registers));
1528 radeon_program_register_sequence(rdev,
1529 hawaii_golden_spm_registers,
1530 (const u32)ARRAY_SIZE(hawaii_golden_spm_registers));
1538 * cik_get_xclk - get the xclk
1540 * @rdev: radeon_device pointer
1542 * Returns the reference clock used by the gfx engine
1545 u32 cik_get_xclk(struct radeon_device *rdev)
1547 u32 reference_clock = rdev->clock.spll.reference_freq;
1549 if (rdev->flags & RADEON_IS_IGP) {
1550 if (RREG32_SMC(GENERAL_PWRMGT) & GPU_COUNTER_CLK)
1551 return reference_clock / 2;
1553 if (RREG32_SMC(CG_CLKPIN_CNTL) & XTALIN_DIVIDE)
1554 return reference_clock / 4;
1556 return reference_clock;
1560 * cik_mm_rdoorbell - read a doorbell dword
1562 * @rdev: radeon_device pointer
1563 * @index: doorbell index
1565 * Returns the value in the doorbell aperture at the
1566 * requested doorbell index (CIK).
1568 u32 cik_mm_rdoorbell(struct radeon_device *rdev, u32 index)
1570 if (index < rdev->doorbell.num_doorbells) {
1571 return readl(rdev->doorbell.ptr + index);
1573 DRM_ERROR("reading beyond doorbell aperture: 0x%08x!\n", index);
1579 * cik_mm_wdoorbell - write a doorbell dword
1581 * @rdev: radeon_device pointer
1582 * @index: doorbell index
1583 * @v: value to write
1585 * Writes @v to the doorbell aperture at the
1586 * requested doorbell index (CIK).
1588 void cik_mm_wdoorbell(struct radeon_device *rdev, u32 index, u32 v)
1590 if (index < rdev->doorbell.num_doorbells) {
1591 writel(v, rdev->doorbell.ptr + index);
1593 DRM_ERROR("writing beyond doorbell aperture: 0x%08x!\n", index);
1597 #define BONAIRE_IO_MC_REGS_SIZE 36
1599 static const u32 bonaire_io_mc_regs[BONAIRE_IO_MC_REGS_SIZE][2] =
1601 {0x00000070, 0x04400000},
1602 {0x00000071, 0x80c01803},
1603 {0x00000072, 0x00004004},
1604 {0x00000073, 0x00000100},
1605 {0x00000074, 0x00ff0000},
1606 {0x00000075, 0x34000000},
1607 {0x00000076, 0x08000014},
1608 {0x00000077, 0x00cc08ec},
1609 {0x00000078, 0x00000400},
1610 {0x00000079, 0x00000000},
1611 {0x0000007a, 0x04090000},
1612 {0x0000007c, 0x00000000},
1613 {0x0000007e, 0x4408a8e8},
1614 {0x0000007f, 0x00000304},
1615 {0x00000080, 0x00000000},
1616 {0x00000082, 0x00000001},
1617 {0x00000083, 0x00000002},
1618 {0x00000084, 0xf3e4f400},
1619 {0x00000085, 0x052024e3},
1620 {0x00000087, 0x00000000},
1621 {0x00000088, 0x01000000},
1622 {0x0000008a, 0x1c0a0000},
1623 {0x0000008b, 0xff010000},
1624 {0x0000008d, 0xffffefff},
1625 {0x0000008e, 0xfff3efff},
1626 {0x0000008f, 0xfff3efbf},
1627 {0x00000092, 0xf7ffffff},
1628 {0x00000093, 0xffffff7f},
1629 {0x00000095, 0x00101101},
1630 {0x00000096, 0x00000fff},
1631 {0x00000097, 0x00116fff},
1632 {0x00000098, 0x60010000},
1633 {0x00000099, 0x10010000},
1634 {0x0000009a, 0x00006000},
1635 {0x0000009b, 0x00001000},
1636 {0x0000009f, 0x00b48000}
1639 #define HAWAII_IO_MC_REGS_SIZE 22
1641 static const u32 hawaii_io_mc_regs[HAWAII_IO_MC_REGS_SIZE][2] =
1643 {0x0000007d, 0x40000000},
1644 {0x0000007e, 0x40180304},
1645 {0x0000007f, 0x0000ff00},
1646 {0x00000081, 0x00000000},
1647 {0x00000083, 0x00000800},
1648 {0x00000086, 0x00000000},
1649 {0x00000087, 0x00000100},
1650 {0x00000088, 0x00020100},
1651 {0x00000089, 0x00000000},
1652 {0x0000008b, 0x00040000},
1653 {0x0000008c, 0x00000100},
1654 {0x0000008e, 0xff010000},
1655 {0x00000090, 0xffffefff},
1656 {0x00000091, 0xfff3efff},
1657 {0x00000092, 0xfff3efbf},
1658 {0x00000093, 0xf7ffffff},
1659 {0x00000094, 0xffffff7f},
1660 {0x00000095, 0x00000fff},
1661 {0x00000096, 0x00116fff},
1662 {0x00000097, 0x60010000},
1663 {0x00000098, 0x10010000},
1664 {0x0000009f, 0x00c79000}
1669 * cik_srbm_select - select specific register instances
1671 * @rdev: radeon_device pointer
1672 * @me: selected ME (micro engine)
1677 * Switches the currently active registers instances. Some
1678 * registers are instanced per VMID, others are instanced per
1679 * me/pipe/queue combination.
1681 static void cik_srbm_select(struct radeon_device *rdev,
1682 u32 me, u32 pipe, u32 queue, u32 vmid)
1684 u32 srbm_gfx_cntl = (PIPEID(pipe & 0x3) |
1687 QUEUEID(queue & 0x7));
1688 WREG32(SRBM_GFX_CNTL, srbm_gfx_cntl);
1693 * ci_mc_load_microcode - load MC ucode into the hw
1695 * @rdev: radeon_device pointer
1697 * Load the GDDR MC ucode into the hw (CIK).
1698 * Returns 0 on success, error on failure.
1700 static int ci_mc_load_microcode(struct radeon_device *rdev)
1702 const __be32 *fw_data;
1703 u32 running, blackout = 0;
1705 int i, ucode_size, regs_size;
1710 switch (rdev->family) {
1712 io_mc_regs = (u32 *)&bonaire_io_mc_regs;
1713 ucode_size = CIK_MC_UCODE_SIZE;
1714 regs_size = BONAIRE_IO_MC_REGS_SIZE;
1717 io_mc_regs = (u32 *)&hawaii_io_mc_regs;
1718 ucode_size = HAWAII_MC_UCODE_SIZE;
1719 regs_size = HAWAII_IO_MC_REGS_SIZE;
1725 running = RREG32(MC_SEQ_SUP_CNTL) & RUN_MASK;
1729 blackout = RREG32(MC_SHARED_BLACKOUT_CNTL);
1730 WREG32(MC_SHARED_BLACKOUT_CNTL, blackout | 1);
1733 /* reset the engine and set to writable */
1734 WREG32(MC_SEQ_SUP_CNTL, 0x00000008);
1735 WREG32(MC_SEQ_SUP_CNTL, 0x00000010);
1737 /* load mc io regs */
1738 for (i = 0; i < regs_size; i++) {
1739 WREG32(MC_SEQ_IO_DEBUG_INDEX, io_mc_regs[(i << 1)]);
1740 WREG32(MC_SEQ_IO_DEBUG_DATA, io_mc_regs[(i << 1) + 1]);
1742 /* load the MC ucode */
1743 fw_data = (const __be32 *)rdev->mc_fw->data;
1744 for (i = 0; i < ucode_size; i++)
1745 WREG32(MC_SEQ_SUP_PGM, be32_to_cpup(fw_data++));
1747 /* put the engine back into the active state */
1748 WREG32(MC_SEQ_SUP_CNTL, 0x00000008);
1749 WREG32(MC_SEQ_SUP_CNTL, 0x00000004);
1750 WREG32(MC_SEQ_SUP_CNTL, 0x00000001);
1752 /* wait for training to complete */
1753 for (i = 0; i < rdev->usec_timeout; i++) {
1754 if (RREG32(MC_SEQ_TRAIN_WAKEUP_CNTL) & TRAIN_DONE_D0)
1758 for (i = 0; i < rdev->usec_timeout; i++) {
1759 if (RREG32(MC_SEQ_TRAIN_WAKEUP_CNTL) & TRAIN_DONE_D1)
1765 WREG32(MC_SHARED_BLACKOUT_CNTL, blackout);
1772 * cik_init_microcode - load ucode images from disk
1774 * @rdev: radeon_device pointer
1776 * Use the firmware interface to load the ucode images into
1777 * the driver (not loaded into hw).
1778 * Returns 0 on success, error on failure.
1780 static int cik_init_microcode(struct radeon_device *rdev)
1782 const char *chip_name;
1783 size_t pfp_req_size, me_req_size, ce_req_size,
1784 mec_req_size, rlc_req_size, mc_req_size = 0,
1785 sdma_req_size, smc_req_size = 0;
1791 switch (rdev->family) {
1793 chip_name = "BONAIRE";
1794 pfp_req_size = CIK_PFP_UCODE_SIZE * 4;
1795 me_req_size = CIK_ME_UCODE_SIZE * 4;
1796 ce_req_size = CIK_CE_UCODE_SIZE * 4;
1797 mec_req_size = CIK_MEC_UCODE_SIZE * 4;
1798 rlc_req_size = BONAIRE_RLC_UCODE_SIZE * 4;
1799 mc_req_size = CIK_MC_UCODE_SIZE * 4;
1800 sdma_req_size = CIK_SDMA_UCODE_SIZE * 4;
1801 smc_req_size = ALIGN(BONAIRE_SMC_UCODE_SIZE, 4);
1804 chip_name = "HAWAII";
1805 pfp_req_size = CIK_PFP_UCODE_SIZE * 4;
1806 me_req_size = CIK_ME_UCODE_SIZE * 4;
1807 ce_req_size = CIK_CE_UCODE_SIZE * 4;
1808 mec_req_size = CIK_MEC_UCODE_SIZE * 4;
1809 rlc_req_size = BONAIRE_RLC_UCODE_SIZE * 4;
1810 mc_req_size = HAWAII_MC_UCODE_SIZE * 4;
1811 sdma_req_size = CIK_SDMA_UCODE_SIZE * 4;
1812 smc_req_size = ALIGN(HAWAII_SMC_UCODE_SIZE, 4);
1815 chip_name = "KAVERI";
1816 pfp_req_size = CIK_PFP_UCODE_SIZE * 4;
1817 me_req_size = CIK_ME_UCODE_SIZE * 4;
1818 ce_req_size = CIK_CE_UCODE_SIZE * 4;
1819 mec_req_size = CIK_MEC_UCODE_SIZE * 4;
1820 rlc_req_size = KV_RLC_UCODE_SIZE * 4;
1821 sdma_req_size = CIK_SDMA_UCODE_SIZE * 4;
1824 chip_name = "KABINI";
1825 pfp_req_size = CIK_PFP_UCODE_SIZE * 4;
1826 me_req_size = CIK_ME_UCODE_SIZE * 4;
1827 ce_req_size = CIK_CE_UCODE_SIZE * 4;
1828 mec_req_size = CIK_MEC_UCODE_SIZE * 4;
1829 rlc_req_size = KB_RLC_UCODE_SIZE * 4;
1830 sdma_req_size = CIK_SDMA_UCODE_SIZE * 4;
1835 DRM_INFO("Loading %s Microcode\n", chip_name);
1837 snprintf(fw_name, sizeof(fw_name), "radeon/%s_pfp.bin", chip_name);
1838 err = request_firmware(&rdev->pfp_fw, fw_name, rdev->dev);
1841 if (rdev->pfp_fw->size != pfp_req_size) {
1843 "cik_cp: Bogus length %zu in firmware \"%s\"\n",
1844 rdev->pfp_fw->size, fw_name);
1849 snprintf(fw_name, sizeof(fw_name), "radeon/%s_me.bin", chip_name);
1850 err = request_firmware(&rdev->me_fw, fw_name, rdev->dev);
1853 if (rdev->me_fw->size != me_req_size) {
1855 "cik_cp: Bogus length %zu in firmware \"%s\"\n",
1856 rdev->me_fw->size, fw_name);
1860 snprintf(fw_name, sizeof(fw_name), "radeon/%s_ce.bin", chip_name);
1861 err = request_firmware(&rdev->ce_fw, fw_name, rdev->dev);
1864 if (rdev->ce_fw->size != ce_req_size) {
1866 "cik_cp: Bogus length %zu in firmware \"%s\"\n",
1867 rdev->ce_fw->size, fw_name);
1871 snprintf(fw_name, sizeof(fw_name), "radeon/%s_mec.bin", chip_name);
1872 err = request_firmware(&rdev->mec_fw, fw_name, rdev->dev);
1875 if (rdev->mec_fw->size != mec_req_size) {
1877 "cik_cp: Bogus length %zu in firmware \"%s\"\n",
1878 rdev->mec_fw->size, fw_name);
1882 snprintf(fw_name, sizeof(fw_name), "radeon/%s_rlc.bin", chip_name);
1883 err = request_firmware(&rdev->rlc_fw, fw_name, rdev->dev);
1886 if (rdev->rlc_fw->size != rlc_req_size) {
1888 "cik_rlc: Bogus length %zu in firmware \"%s\"\n",
1889 rdev->rlc_fw->size, fw_name);
1893 snprintf(fw_name, sizeof(fw_name), "radeon/%s_sdma.bin", chip_name);
1894 err = request_firmware(&rdev->sdma_fw, fw_name, rdev->dev);
1897 if (rdev->sdma_fw->size != sdma_req_size) {
1899 "cik_sdma: Bogus length %zu in firmware \"%s\"\n",
1900 rdev->sdma_fw->size, fw_name);
1904 /* No SMC, MC ucode on APUs */
1905 if (!(rdev->flags & RADEON_IS_IGP)) {
1906 snprintf(fw_name, sizeof(fw_name), "radeon/%s_mc.bin", chip_name);
1907 err = request_firmware(&rdev->mc_fw, fw_name, rdev->dev);
1910 if (rdev->mc_fw->size != mc_req_size) {
1912 "cik_mc: Bogus length %zu in firmware \"%s\"\n",
1913 rdev->mc_fw->size, fw_name);
1917 snprintf(fw_name, sizeof(fw_name), "radeon/%s_smc.bin", chip_name);
1918 err = request_firmware(&rdev->smc_fw, fw_name, rdev->dev);
1921 "smc: error loading firmware \"%s\"\n",
1923 release_firmware(rdev->smc_fw);
1924 rdev->smc_fw = NULL;
1926 } else if (rdev->smc_fw->size != smc_req_size) {
1928 "cik_smc: Bogus length %zu in firmware \"%s\"\n",
1929 rdev->smc_fw->size, fw_name);
1938 "cik_cp: Failed to load firmware \"%s\"\n",
1940 release_firmware(rdev->pfp_fw);
1941 rdev->pfp_fw = NULL;
1942 release_firmware(rdev->me_fw);
1944 release_firmware(rdev->ce_fw);
1946 release_firmware(rdev->rlc_fw);
1947 rdev->rlc_fw = NULL;
1948 release_firmware(rdev->mc_fw);
1950 release_firmware(rdev->smc_fw);
1951 rdev->smc_fw = NULL;
1960 * cik_tiling_mode_table_init - init the hw tiling table
1962 * @rdev: radeon_device pointer
1964 * Starting with SI, the tiling setup is done globally in a
1965 * set of 32 tiling modes. Rather than selecting each set of
1966 * parameters per surface as on older asics, we just select
1967 * which index in the tiling table we want to use, and the
1968 * surface uses those parameters (CIK).
1970 static void cik_tiling_mode_table_init(struct radeon_device *rdev)
1972 const u32 num_tile_mode_states = 32;
1973 const u32 num_secondary_tile_mode_states = 16;
1974 u32 reg_offset, gb_tile_moden, split_equal_to_row_size;
1975 u32 num_pipe_configs;
1976 u32 num_rbs = rdev->config.cik.max_backends_per_se *
1977 rdev->config.cik.max_shader_engines;
1979 switch (rdev->config.cik.mem_row_size_in_kb) {
1981 split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_1KB;
1985 split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_2KB;
1988 split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_4KB;
1992 num_pipe_configs = rdev->config.cik.max_tile_pipes;
1993 if (num_pipe_configs > 8)
1994 num_pipe_configs = 16;
1996 if (num_pipe_configs == 16) {
1997 for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) {
1998 switch (reg_offset) {
2000 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2001 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2002 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2003 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B));
2006 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2007 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2008 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2009 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B));
2012 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2013 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2014 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2015 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B));
2018 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2019 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2020 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2021 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B));
2024 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2025 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2026 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2027 TILE_SPLIT(split_equal_to_row_size));
2030 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2031 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2034 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2035 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2036 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2037 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B));
2040 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2041 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2042 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2043 TILE_SPLIT(split_equal_to_row_size));
2046 gb_tile_moden = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
2047 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16));
2050 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2051 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING));
2054 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2055 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2056 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2057 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2060 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2061 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2062 PIPE_CONFIG(ADDR_SURF_P16_32x32_8x16) |
2063 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2066 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2067 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2068 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2069 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2072 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2073 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING));
2076 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2077 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2078 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2079 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2082 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2083 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2084 PIPE_CONFIG(ADDR_SURF_P16_32x32_8x16) |
2085 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2088 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2089 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2090 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2091 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2094 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2095 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING));
2098 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2099 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2100 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2101 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2104 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2105 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2106 PIPE_CONFIG(ADDR_SURF_P16_32x32_8x16) |
2107 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2110 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2111 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2112 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2113 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2119 rdev->config.cik.tile_mode_array[reg_offset] = gb_tile_moden;
2120 WREG32(GB_TILE_MODE0 + (reg_offset * 4), gb_tile_moden);
2122 for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++) {
2123 switch (reg_offset) {
2125 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2126 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2127 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2128 NUM_BANKS(ADDR_SURF_16_BANK));
2131 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2132 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2133 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2134 NUM_BANKS(ADDR_SURF_16_BANK));
2137 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2138 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2139 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2140 NUM_BANKS(ADDR_SURF_16_BANK));
2143 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2144 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2145 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2146 NUM_BANKS(ADDR_SURF_16_BANK));
2149 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2150 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2151 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2152 NUM_BANKS(ADDR_SURF_8_BANK));
2155 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2156 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2157 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2158 NUM_BANKS(ADDR_SURF_4_BANK));
2161 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2162 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2163 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2164 NUM_BANKS(ADDR_SURF_2_BANK));
2167 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2168 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2169 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2170 NUM_BANKS(ADDR_SURF_16_BANK));
2173 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2174 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2175 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2176 NUM_BANKS(ADDR_SURF_16_BANK));
2179 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2180 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2181 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2182 NUM_BANKS(ADDR_SURF_16_BANK));
2185 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2186 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2187 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2188 NUM_BANKS(ADDR_SURF_8_BANK));
2191 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2192 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2193 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2194 NUM_BANKS(ADDR_SURF_4_BANK));
2197 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2198 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2199 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2200 NUM_BANKS(ADDR_SURF_2_BANK));
2203 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2204 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2205 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2206 NUM_BANKS(ADDR_SURF_2_BANK));
2212 WREG32(GB_MACROTILE_MODE0 + (reg_offset * 4), gb_tile_moden);
2214 } else if (num_pipe_configs == 8) {
2215 for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) {
2216 switch (reg_offset) {
2218 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2219 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2220 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2221 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B));
2224 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2225 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2226 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2227 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B));
2230 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2231 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2232 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2233 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B));
2236 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2237 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2238 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2239 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B));
2242 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2243 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2244 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2245 TILE_SPLIT(split_equal_to_row_size));
2248 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2249 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2252 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2253 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2254 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2255 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B));
2258 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2259 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2260 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2261 TILE_SPLIT(split_equal_to_row_size));
2264 gb_tile_moden = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
2265 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16));
2268 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2269 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING));
2272 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2273 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2274 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2275 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2278 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2279 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2280 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
2281 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2284 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2285 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2286 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2287 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2290 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2291 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING));
2294 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2295 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2296 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2297 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2300 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2301 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2302 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
2303 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2306 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2307 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2308 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2309 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2312 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2313 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING));
2316 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2317 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2318 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2319 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2322 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2323 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2324 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
2325 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2328 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2329 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2330 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2331 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2337 rdev->config.cik.tile_mode_array[reg_offset] = gb_tile_moden;
2338 WREG32(GB_TILE_MODE0 + (reg_offset * 4), gb_tile_moden);
2340 for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++) {
2341 switch (reg_offset) {
2343 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2344 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2345 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2346 NUM_BANKS(ADDR_SURF_16_BANK));
2349 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2350 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2351 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2352 NUM_BANKS(ADDR_SURF_16_BANK));
2355 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2356 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2357 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2358 NUM_BANKS(ADDR_SURF_16_BANK));
2361 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2362 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2363 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2364 NUM_BANKS(ADDR_SURF_16_BANK));
2367 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2368 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2369 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2370 NUM_BANKS(ADDR_SURF_8_BANK));
2373 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2374 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2375 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2376 NUM_BANKS(ADDR_SURF_4_BANK));
2379 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2380 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2381 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2382 NUM_BANKS(ADDR_SURF_2_BANK));
2385 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2386 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
2387 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2388 NUM_BANKS(ADDR_SURF_16_BANK));
2391 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2392 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2393 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2394 NUM_BANKS(ADDR_SURF_16_BANK));
2397 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2398 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2399 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2400 NUM_BANKS(ADDR_SURF_16_BANK));
2403 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2404 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2405 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2406 NUM_BANKS(ADDR_SURF_16_BANK));
2409 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2410 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2411 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2412 NUM_BANKS(ADDR_SURF_8_BANK));
2415 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2416 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2417 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2418 NUM_BANKS(ADDR_SURF_4_BANK));
2421 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2422 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2423 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2424 NUM_BANKS(ADDR_SURF_2_BANK));
2430 WREG32(GB_MACROTILE_MODE0 + (reg_offset * 4), gb_tile_moden);
2432 } else if (num_pipe_configs == 4) {
2434 for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) {
2435 switch (reg_offset) {
2437 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2438 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2439 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2440 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B));
2443 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2444 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2445 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2446 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B));
2449 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2450 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2451 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2452 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B));
2455 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2456 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2457 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2458 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B));
2461 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2462 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2463 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2464 TILE_SPLIT(split_equal_to_row_size));
2467 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2468 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2471 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2472 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2473 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2474 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B));
2477 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2478 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2479 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2480 TILE_SPLIT(split_equal_to_row_size));
2483 gb_tile_moden = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
2484 PIPE_CONFIG(ADDR_SURF_P4_16x16));
2487 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2488 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING));
2491 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2492 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2493 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2494 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2497 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2498 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2499 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
2500 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2503 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2504 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2505 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2506 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2509 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2510 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING));
2513 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2514 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2515 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2516 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2519 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2520 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2521 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
2522 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2525 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2526 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2527 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2528 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2531 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2532 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING));
2535 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2536 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2537 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2538 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2541 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2542 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2543 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
2544 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2547 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2548 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2549 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2550 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2556 rdev->config.cik.tile_mode_array[reg_offset] = gb_tile_moden;
2557 WREG32(GB_TILE_MODE0 + (reg_offset * 4), gb_tile_moden);
2559 } else if (num_rbs < 4) {
2560 for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) {
2561 switch (reg_offset) {
2563 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2564 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2565 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
2566 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B));
2569 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2570 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2571 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
2572 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B));
2575 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2576 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2577 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
2578 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B));
2581 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2582 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2583 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
2584 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B));
2587 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2588 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2589 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
2590 TILE_SPLIT(split_equal_to_row_size));
2593 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2594 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2597 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2598 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2599 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
2600 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B));
2603 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2604 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2605 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
2606 TILE_SPLIT(split_equal_to_row_size));
2609 gb_tile_moden = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
2610 PIPE_CONFIG(ADDR_SURF_P4_8x16));
2613 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2614 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING));
2617 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2618 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2619 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
2620 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2623 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2624 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2625 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
2626 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2629 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2630 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2631 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
2632 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2635 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2636 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING));
2639 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2640 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2641 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
2642 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2645 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2646 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2647 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
2648 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2651 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2652 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2653 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
2654 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2657 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2658 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING));
2661 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2662 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2663 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
2664 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2667 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2668 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2669 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
2670 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2673 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2674 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2675 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
2676 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2682 rdev->config.cik.tile_mode_array[reg_offset] = gb_tile_moden;
2683 WREG32(GB_TILE_MODE0 + (reg_offset * 4), gb_tile_moden);
2686 for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++) {
2687 switch (reg_offset) {
2689 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2690 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2691 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2692 NUM_BANKS(ADDR_SURF_16_BANK));
2695 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2696 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2697 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2698 NUM_BANKS(ADDR_SURF_16_BANK));
2701 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2702 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2703 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2704 NUM_BANKS(ADDR_SURF_16_BANK));
2707 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2708 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2709 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2710 NUM_BANKS(ADDR_SURF_16_BANK));
2713 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2714 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2715 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2716 NUM_BANKS(ADDR_SURF_16_BANK));
2719 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2720 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2721 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2722 NUM_BANKS(ADDR_SURF_8_BANK));
2725 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2726 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2727 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2728 NUM_BANKS(ADDR_SURF_4_BANK));
2731 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
2732 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
2733 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2734 NUM_BANKS(ADDR_SURF_16_BANK));
2737 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
2738 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2739 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2740 NUM_BANKS(ADDR_SURF_16_BANK));
2743 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2744 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2745 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2746 NUM_BANKS(ADDR_SURF_16_BANK));
2749 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2750 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2751 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2752 NUM_BANKS(ADDR_SURF_16_BANK));
2755 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2756 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2757 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2758 NUM_BANKS(ADDR_SURF_16_BANK));
2761 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2762 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2763 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2764 NUM_BANKS(ADDR_SURF_8_BANK));
2767 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2768 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2769 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2770 NUM_BANKS(ADDR_SURF_4_BANK));
2776 WREG32(GB_MACROTILE_MODE0 + (reg_offset * 4), gb_tile_moden);
2778 } else if (num_pipe_configs == 2) {
2779 for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) {
2780 switch (reg_offset) {
2782 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2783 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2784 PIPE_CONFIG(ADDR_SURF_P2) |
2785 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B));
2788 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2789 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2790 PIPE_CONFIG(ADDR_SURF_P2) |
2791 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B));
2794 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2795 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2796 PIPE_CONFIG(ADDR_SURF_P2) |
2797 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B));
2800 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2801 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2802 PIPE_CONFIG(ADDR_SURF_P2) |
2803 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B));
2806 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2807 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2808 PIPE_CONFIG(ADDR_SURF_P2) |
2809 TILE_SPLIT(split_equal_to_row_size));
2812 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2813 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2816 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2817 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2818 PIPE_CONFIG(ADDR_SURF_P2) |
2819 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B));
2822 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2823 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2824 PIPE_CONFIG(ADDR_SURF_P2) |
2825 TILE_SPLIT(split_equal_to_row_size));
2828 gb_tile_moden = ARRAY_MODE(ARRAY_LINEAR_ALIGNED);
2831 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2832 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING));
2835 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2836 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2837 PIPE_CONFIG(ADDR_SURF_P2) |
2838 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2841 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2842 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2843 PIPE_CONFIG(ADDR_SURF_P2) |
2844 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2847 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2848 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2849 PIPE_CONFIG(ADDR_SURF_P2) |
2850 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2853 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2854 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING));
2857 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2858 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2859 PIPE_CONFIG(ADDR_SURF_P2) |
2860 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2863 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2864 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2865 PIPE_CONFIG(ADDR_SURF_P2) |
2866 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2869 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2870 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2871 PIPE_CONFIG(ADDR_SURF_P2) |
2872 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2875 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2876 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING));
2879 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2880 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2881 PIPE_CONFIG(ADDR_SURF_P2) |
2882 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2885 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2886 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2887 PIPE_CONFIG(ADDR_SURF_P2) |
2888 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2891 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2892 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2893 PIPE_CONFIG(ADDR_SURF_P2) |
2894 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2900 rdev->config.cik.tile_mode_array[reg_offset] = gb_tile_moden;
2901 WREG32(GB_TILE_MODE0 + (reg_offset * 4), gb_tile_moden);
2903 for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++) {
2904 switch (reg_offset) {
2906 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
2907 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2908 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2909 NUM_BANKS(ADDR_SURF_16_BANK));
2912 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
2913 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2914 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2915 NUM_BANKS(ADDR_SURF_16_BANK));
2918 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2919 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2920 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2921 NUM_BANKS(ADDR_SURF_16_BANK));
2924 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2925 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2926 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2927 NUM_BANKS(ADDR_SURF_16_BANK));
2930 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2931 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2932 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2933 NUM_BANKS(ADDR_SURF_16_BANK));
2936 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2937 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2938 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2939 NUM_BANKS(ADDR_SURF_16_BANK));
2942 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2943 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2944 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2945 NUM_BANKS(ADDR_SURF_8_BANK));
2948 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
2949 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
2950 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2951 NUM_BANKS(ADDR_SURF_16_BANK));
2954 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
2955 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2956 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2957 NUM_BANKS(ADDR_SURF_16_BANK));
2960 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
2961 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2962 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2963 NUM_BANKS(ADDR_SURF_16_BANK));
2966 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
2967 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2968 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2969 NUM_BANKS(ADDR_SURF_16_BANK));
2972 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2973 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2974 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2975 NUM_BANKS(ADDR_SURF_16_BANK));
2978 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2979 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2980 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2981 NUM_BANKS(ADDR_SURF_16_BANK));
2984 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2985 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2986 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2987 NUM_BANKS(ADDR_SURF_8_BANK));
2993 WREG32(GB_MACROTILE_MODE0 + (reg_offset * 4), gb_tile_moden);
2996 DRM_ERROR("unknown num pipe config: 0x%x\n", num_pipe_configs);
3000 * cik_select_se_sh - select which SE, SH to address
3002 * @rdev: radeon_device pointer
3003 * @se_num: shader engine to address
3004 * @sh_num: sh block to address
3006 * Select which SE, SH combinations to address. Certain
3007 * registers are instanced per SE or SH. 0xffffffff means
3008 * broadcast to all SEs or SHs (CIK).
3010 static void cik_select_se_sh(struct radeon_device *rdev,
3011 u32 se_num, u32 sh_num)
3013 u32 data = INSTANCE_BROADCAST_WRITES;
3015 if ((se_num == 0xffffffff) && (sh_num == 0xffffffff))
3016 data |= SH_BROADCAST_WRITES | SE_BROADCAST_WRITES;
3017 else if (se_num == 0xffffffff)
3018 data |= SE_BROADCAST_WRITES | SH_INDEX(sh_num);
3019 else if (sh_num == 0xffffffff)
3020 data |= SH_BROADCAST_WRITES | SE_INDEX(se_num);
3022 data |= SH_INDEX(sh_num) | SE_INDEX(se_num);
3023 WREG32(GRBM_GFX_INDEX, data);
3027 * cik_create_bitmask - create a bitmask
3029 * @bit_width: length of the mask
3031 * create a variable length bit mask (CIK).
3032 * Returns the bitmask.
3034 static u32 cik_create_bitmask(u32 bit_width)
3038 for (i = 0; i < bit_width; i++) {
3046 * cik_select_se_sh - select which SE, SH to address
3048 * @rdev: radeon_device pointer
3049 * @max_rb_num: max RBs (render backends) for the asic
3050 * @se_num: number of SEs (shader engines) for the asic
3051 * @sh_per_se: number of SH blocks per SE for the asic
3053 * Calculates the bitmask of disabled RBs (CIK).
3054 * Returns the disabled RB bitmask.
3056 static u32 cik_get_rb_disabled(struct radeon_device *rdev,
3057 u32 max_rb_num, u32 se_num,
3062 data = RREG32(CC_RB_BACKEND_DISABLE);
3064 data &= BACKEND_DISABLE_MASK;
3067 data |= RREG32(GC_USER_RB_BACKEND_DISABLE);
3069 data >>= BACKEND_DISABLE_SHIFT;
3071 mask = cik_create_bitmask(max_rb_num / se_num / sh_per_se);
3077 * cik_setup_rb - setup the RBs on the asic
3079 * @rdev: radeon_device pointer
3080 * @se_num: number of SEs (shader engines) for the asic
3081 * @sh_per_se: number of SH blocks per SE for the asic
3082 * @max_rb_num: max RBs (render backends) for the asic
3084 * Configures per-SE/SH RB registers (CIK).
3086 static void cik_setup_rb(struct radeon_device *rdev,
3087 u32 se_num, u32 sh_per_se,
3092 u32 disabled_rbs = 0;
3093 u32 enabled_rbs = 0;
3095 for (i = 0; i < se_num; i++) {
3096 for (j = 0; j < sh_per_se; j++) {
3097 cik_select_se_sh(rdev, i, j);
3098 data = cik_get_rb_disabled(rdev, max_rb_num, se_num, sh_per_se);
3099 if (rdev->family == CHIP_HAWAII)
3100 disabled_rbs |= data << ((i * sh_per_se + j) * HAWAII_RB_BITMAP_WIDTH_PER_SH);
3102 disabled_rbs |= data << ((i * sh_per_se + j) * CIK_RB_BITMAP_WIDTH_PER_SH);
3105 cik_select_se_sh(rdev, 0xffffffff, 0xffffffff);
3108 for (i = 0; i < max_rb_num; i++) {
3109 if (!(disabled_rbs & mask))
3110 enabled_rbs |= mask;
3114 for (i = 0; i < se_num; i++) {
3115 cik_select_se_sh(rdev, i, 0xffffffff);
3117 for (j = 0; j < sh_per_se; j++) {
3118 switch (enabled_rbs & 3) {
3121 data |= PKR_MAP(RASTER_CONFIG_RB_MAP_3);
3123 data |= PKR_MAP(RASTER_CONFIG_RB_MAP_0);
3126 data |= (RASTER_CONFIG_RB_MAP_0 << (i * sh_per_se + j) * 2);
3129 data |= (RASTER_CONFIG_RB_MAP_3 << (i * sh_per_se + j) * 2);
3133 data |= (RASTER_CONFIG_RB_MAP_2 << (i * sh_per_se + j) * 2);
3138 WREG32(PA_SC_RASTER_CONFIG, data);
3140 cik_select_se_sh(rdev, 0xffffffff, 0xffffffff);
3144 * cik_gpu_init - setup the 3D engine
3146 * @rdev: radeon_device pointer
3148 * Configures the 3D engine and tiling configuration
3149 * registers so that the 3D engine is usable.
3151 static void cik_gpu_init(struct radeon_device *rdev)
3153 u32 gb_addr_config = RREG32(GB_ADDR_CONFIG);
3154 u32 mc_shared_chmap, mc_arb_ramcfg;
3155 u32 hdp_host_path_cntl;
3159 switch (rdev->family) {
3161 rdev->config.cik.max_shader_engines = 2;
3162 rdev->config.cik.max_tile_pipes = 4;
3163 rdev->config.cik.max_cu_per_sh = 7;
3164 rdev->config.cik.max_sh_per_se = 1;
3165 rdev->config.cik.max_backends_per_se = 2;
3166 rdev->config.cik.max_texture_channel_caches = 4;
3167 rdev->config.cik.max_gprs = 256;
3168 rdev->config.cik.max_gs_threads = 32;
3169 rdev->config.cik.max_hw_contexts = 8;
3171 rdev->config.cik.sc_prim_fifo_size_frontend = 0x20;
3172 rdev->config.cik.sc_prim_fifo_size_backend = 0x100;
3173 rdev->config.cik.sc_hiz_tile_fifo_size = 0x30;
3174 rdev->config.cik.sc_earlyz_tile_fifo_size = 0x130;
3175 gb_addr_config = BONAIRE_GB_ADDR_CONFIG_GOLDEN;
3178 rdev->config.cik.max_shader_engines = 4;
3179 rdev->config.cik.max_tile_pipes = 16;
3180 rdev->config.cik.max_cu_per_sh = 11;
3181 rdev->config.cik.max_sh_per_se = 1;
3182 rdev->config.cik.max_backends_per_se = 4;
3183 rdev->config.cik.max_texture_channel_caches = 16;
3184 rdev->config.cik.max_gprs = 256;
3185 rdev->config.cik.max_gs_threads = 32;
3186 rdev->config.cik.max_hw_contexts = 8;
3188 rdev->config.cik.sc_prim_fifo_size_frontend = 0x20;
3189 rdev->config.cik.sc_prim_fifo_size_backend = 0x100;
3190 rdev->config.cik.sc_hiz_tile_fifo_size = 0x30;
3191 rdev->config.cik.sc_earlyz_tile_fifo_size = 0x130;
3192 gb_addr_config = HAWAII_GB_ADDR_CONFIG_GOLDEN;
3195 rdev->config.cik.max_shader_engines = 1;
3196 rdev->config.cik.max_tile_pipes = 4;
3197 if ((rdev->pdev->device == 0x1304) ||
3198 (rdev->pdev->device == 0x1305) ||
3199 (rdev->pdev->device == 0x130C) ||
3200 (rdev->pdev->device == 0x130F) ||
3201 (rdev->pdev->device == 0x1310) ||
3202 (rdev->pdev->device == 0x1311) ||
3203 (rdev->pdev->device == 0x131C)) {
3204 rdev->config.cik.max_cu_per_sh = 8;
3205 rdev->config.cik.max_backends_per_se = 2;
3206 } else if ((rdev->pdev->device == 0x1309) ||
3207 (rdev->pdev->device == 0x130A) ||
3208 (rdev->pdev->device == 0x130D) ||
3209 (rdev->pdev->device == 0x1313) ||
3210 (rdev->pdev->device == 0x131D)) {
3211 rdev->config.cik.max_cu_per_sh = 6;
3212 rdev->config.cik.max_backends_per_se = 2;
3213 } else if ((rdev->pdev->device == 0x1306) ||
3214 (rdev->pdev->device == 0x1307) ||
3215 (rdev->pdev->device == 0x130B) ||
3216 (rdev->pdev->device == 0x130E) ||
3217 (rdev->pdev->device == 0x1315) ||
3218 (rdev->pdev->device == 0x131B)) {
3219 rdev->config.cik.max_cu_per_sh = 4;
3220 rdev->config.cik.max_backends_per_se = 1;
3222 rdev->config.cik.max_cu_per_sh = 3;
3223 rdev->config.cik.max_backends_per_se = 1;
3225 rdev->config.cik.max_sh_per_se = 1;
3226 rdev->config.cik.max_texture_channel_caches = 4;
3227 rdev->config.cik.max_gprs = 256;
3228 rdev->config.cik.max_gs_threads = 16;
3229 rdev->config.cik.max_hw_contexts = 8;
3231 rdev->config.cik.sc_prim_fifo_size_frontend = 0x20;
3232 rdev->config.cik.sc_prim_fifo_size_backend = 0x100;
3233 rdev->config.cik.sc_hiz_tile_fifo_size = 0x30;
3234 rdev->config.cik.sc_earlyz_tile_fifo_size = 0x130;
3235 gb_addr_config = BONAIRE_GB_ADDR_CONFIG_GOLDEN;
3239 rdev->config.cik.max_shader_engines = 1;
3240 rdev->config.cik.max_tile_pipes = 2;
3241 rdev->config.cik.max_cu_per_sh = 2;
3242 rdev->config.cik.max_sh_per_se = 1;
3243 rdev->config.cik.max_backends_per_se = 1;
3244 rdev->config.cik.max_texture_channel_caches = 2;
3245 rdev->config.cik.max_gprs = 256;
3246 rdev->config.cik.max_gs_threads = 16;
3247 rdev->config.cik.max_hw_contexts = 8;
3249 rdev->config.cik.sc_prim_fifo_size_frontend = 0x20;
3250 rdev->config.cik.sc_prim_fifo_size_backend = 0x100;
3251 rdev->config.cik.sc_hiz_tile_fifo_size = 0x30;
3252 rdev->config.cik.sc_earlyz_tile_fifo_size = 0x130;
3253 gb_addr_config = BONAIRE_GB_ADDR_CONFIG_GOLDEN;
3257 /* Initialize HDP */
3258 for (i = 0, j = 0; i < 32; i++, j += 0x18) {
3259 WREG32((0x2c14 + j), 0x00000000);
3260 WREG32((0x2c18 + j), 0x00000000);
3261 WREG32((0x2c1c + j), 0x00000000);
3262 WREG32((0x2c20 + j), 0x00000000);
3263 WREG32((0x2c24 + j), 0x00000000);
3266 WREG32(GRBM_CNTL, GRBM_READ_TIMEOUT(0xff));
3268 WREG32(BIF_FB_EN, FB_READ_EN | FB_WRITE_EN);
3270 mc_shared_chmap = RREG32(MC_SHARED_CHMAP);
3271 mc_arb_ramcfg = RREG32(MC_ARB_RAMCFG);
3273 rdev->config.cik.num_tile_pipes = rdev->config.cik.max_tile_pipes;
3274 rdev->config.cik.mem_max_burst_length_bytes = 256;
3275 tmp = (mc_arb_ramcfg & NOOFCOLS_MASK) >> NOOFCOLS_SHIFT;
3276 rdev->config.cik.mem_row_size_in_kb = (4 * (1 << (8 + tmp))) / 1024;
3277 if (rdev->config.cik.mem_row_size_in_kb > 4)
3278 rdev->config.cik.mem_row_size_in_kb = 4;
3279 /* XXX use MC settings? */
3280 rdev->config.cik.shader_engine_tile_size = 32;
3281 rdev->config.cik.num_gpus = 1;
3282 rdev->config.cik.multi_gpu_tile_size = 64;
3284 /* fix up row size */
3285 gb_addr_config &= ~ROW_SIZE_MASK;
3286 switch (rdev->config.cik.mem_row_size_in_kb) {
3289 gb_addr_config |= ROW_SIZE(0);
3292 gb_addr_config |= ROW_SIZE(1);
3295 gb_addr_config |= ROW_SIZE(2);
3299 /* setup tiling info dword. gb_addr_config is not adequate since it does
3300 * not have bank info, so create a custom tiling dword.
3301 * bits 3:0 num_pipes
3302 * bits 7:4 num_banks
3303 * bits 11:8 group_size
3304 * bits 15:12 row_size
3306 rdev->config.cik.tile_config = 0;
3307 switch (rdev->config.cik.num_tile_pipes) {
3309 rdev->config.cik.tile_config |= (0 << 0);
3312 rdev->config.cik.tile_config |= (1 << 0);
3315 rdev->config.cik.tile_config |= (2 << 0);
3319 /* XXX what about 12? */
3320 rdev->config.cik.tile_config |= (3 << 0);
3323 rdev->config.cik.tile_config |=
3324 ((mc_arb_ramcfg & NOOFBANK_MASK) >> NOOFBANK_SHIFT) << 4;
3325 rdev->config.cik.tile_config |=
3326 ((gb_addr_config & PIPE_INTERLEAVE_SIZE_MASK) >> PIPE_INTERLEAVE_SIZE_SHIFT) << 8;
3327 rdev->config.cik.tile_config |=
3328 ((gb_addr_config & ROW_SIZE_MASK) >> ROW_SIZE_SHIFT) << 12;
3330 WREG32(GB_ADDR_CONFIG, gb_addr_config);
3331 WREG32(HDP_ADDR_CONFIG, gb_addr_config);
3332 WREG32(DMIF_ADDR_CALC, gb_addr_config);
3333 WREG32(SDMA0_TILING_CONFIG + SDMA0_REGISTER_OFFSET, gb_addr_config & 0x70);
3334 WREG32(SDMA0_TILING_CONFIG + SDMA1_REGISTER_OFFSET, gb_addr_config & 0x70);
3335 WREG32(UVD_UDEC_ADDR_CONFIG, gb_addr_config);
3336 WREG32(UVD_UDEC_DB_ADDR_CONFIG, gb_addr_config);
3337 WREG32(UVD_UDEC_DBW_ADDR_CONFIG, gb_addr_config);
3339 cik_tiling_mode_table_init(rdev);
3341 cik_setup_rb(rdev, rdev->config.cik.max_shader_engines,
3342 rdev->config.cik.max_sh_per_se,
3343 rdev->config.cik.max_backends_per_se);
3345 /* set HW defaults for 3D engine */
3346 WREG32(CP_MEQ_THRESHOLDS, MEQ1_START(0x30) | MEQ2_START(0x60));
3348 WREG32(SX_DEBUG_1, 0x20);
3350 WREG32(TA_CNTL_AUX, 0x00010000);
3352 tmp = RREG32(SPI_CONFIG_CNTL);
3354 WREG32(SPI_CONFIG_CNTL, tmp);
3356 WREG32(SQ_CONFIG, 1);
3358 WREG32(DB_DEBUG, 0);
3360 tmp = RREG32(DB_DEBUG2) & ~0xf00fffff;
3362 WREG32(DB_DEBUG2, tmp);
3364 tmp = RREG32(DB_DEBUG3) & ~0x0002021c;
3366 WREG32(DB_DEBUG3, tmp);
3368 tmp = RREG32(CB_HW_CONTROL) & ~0x00010000;
3370 WREG32(CB_HW_CONTROL, tmp);
3372 WREG32(SPI_CONFIG_CNTL_1, VTX_DONE_DELAY(4));
3374 WREG32(PA_SC_FIFO_SIZE, (SC_FRONTEND_PRIM_FIFO_SIZE(rdev->config.cik.sc_prim_fifo_size_frontend) |
3375 SC_BACKEND_PRIM_FIFO_SIZE(rdev->config.cik.sc_prim_fifo_size_backend) |
3376 SC_HIZ_TILE_FIFO_SIZE(rdev->config.cik.sc_hiz_tile_fifo_size) |
3377 SC_EARLYZ_TILE_FIFO_SIZE(rdev->config.cik.sc_earlyz_tile_fifo_size)));
3379 WREG32(VGT_NUM_INSTANCES, 1);
3381 WREG32(CP_PERFMON_CNTL, 0);
3383 WREG32(SQ_CONFIG, 0);
3385 WREG32(PA_SC_FORCE_EOV_MAX_CNTS, (FORCE_EOV_MAX_CLK_CNT(4095) |
3386 FORCE_EOV_MAX_REZ_CNT(255)));
3388 WREG32(VGT_CACHE_INVALIDATION, CACHE_INVALIDATION(VC_AND_TC) |
3389 AUTO_INVLD_EN(ES_AND_GS_AUTO));
3391 WREG32(VGT_GS_VERTEX_REUSE, 16);
3392 WREG32(PA_SC_LINE_STIPPLE_STATE, 0);
3394 tmp = RREG32(HDP_MISC_CNTL);
3395 tmp |= HDP_FLUSH_INVALIDATE_CACHE;
3396 WREG32(HDP_MISC_CNTL, tmp);
3398 hdp_host_path_cntl = RREG32(HDP_HOST_PATH_CNTL);
3399 WREG32(HDP_HOST_PATH_CNTL, hdp_host_path_cntl);
3401 WREG32(PA_CL_ENHANCE, CLIP_VTX_REORDER_ENA | NUM_CLIP_SEQ(3));
3402 WREG32(PA_SC_ENHANCE, ENABLE_PA_SC_OUT_OF_ORDER);
3408 * GPU scratch registers helpers function.
3411 * cik_scratch_init - setup driver info for CP scratch regs
3413 * @rdev: radeon_device pointer
3415 * Set up the number and offset of the CP scratch registers.
3416 * NOTE: use of CP scratch registers is a legacy inferface and
3417 * is not used by default on newer asics (r6xx+). On newer asics,
3418 * memory buffers are used for fences rather than scratch regs.
3420 static void cik_scratch_init(struct radeon_device *rdev)
3424 rdev->scratch.num_reg = 7;
3425 rdev->scratch.reg_base = SCRATCH_REG0;
3426 for (i = 0; i < rdev->scratch.num_reg; i++) {
3427 rdev->scratch.free[i] = true;
3428 rdev->scratch.reg[i] = rdev->scratch.reg_base + (i * 4);
3433 * cik_ring_test - basic gfx ring test
3435 * @rdev: radeon_device pointer
3436 * @ring: radeon_ring structure holding ring information
3438 * Allocate a scratch register and write to it using the gfx ring (CIK).
3439 * Provides a basic gfx ring test to verify that the ring is working.
3440 * Used by cik_cp_gfx_resume();
3441 * Returns 0 on success, error on failure.
3443 int cik_ring_test(struct radeon_device *rdev, struct radeon_ring *ring)
3450 r = radeon_scratch_get(rdev, &scratch);
3452 DRM_ERROR("radeon: cp failed to get scratch reg (%d).\n", r);
3455 WREG32(scratch, 0xCAFEDEAD);
3456 r = radeon_ring_lock(rdev, ring, 3);
3458 DRM_ERROR("radeon: cp failed to lock ring %d (%d).\n", ring->idx, r);
3459 radeon_scratch_free(rdev, scratch);
3462 radeon_ring_write(ring, PACKET3(PACKET3_SET_UCONFIG_REG, 1));
3463 radeon_ring_write(ring, ((scratch - PACKET3_SET_UCONFIG_REG_START) >> 2));
3464 radeon_ring_write(ring, 0xDEADBEEF);
3465 radeon_ring_unlock_commit(rdev, ring);
3467 for (i = 0; i < rdev->usec_timeout; i++) {
3468 tmp = RREG32(scratch);
3469 if (tmp == 0xDEADBEEF)
3473 if (i < rdev->usec_timeout) {
3474 DRM_INFO("ring test on %d succeeded in %d usecs\n", ring->idx, i);
3476 DRM_ERROR("radeon: ring %d test failed (scratch(0x%04X)=0x%08X)\n",
3477 ring->idx, scratch, tmp);
3480 radeon_scratch_free(rdev, scratch);
3485 * cik_fence_gfx_ring_emit - emit a fence on the gfx ring
3487 * @rdev: radeon_device pointer
3488 * @fence: radeon fence object
3490 * Emits a fence sequnce number on the gfx ring and flushes
3493 void cik_fence_gfx_ring_emit(struct radeon_device *rdev,
3494 struct radeon_fence *fence)
3496 struct radeon_ring *ring = &rdev->ring[fence->ring];
3497 u64 addr = rdev->fence_drv[fence->ring].gpu_addr;
3499 /* EVENT_WRITE_EOP - flush caches, send int */
3500 radeon_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE_EOP, 4));
3501 radeon_ring_write(ring, (EOP_TCL1_ACTION_EN |
3503 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
3505 radeon_ring_write(ring, addr & 0xfffffffc);
3506 radeon_ring_write(ring, (upper_32_bits(addr) & 0xffff) | DATA_SEL(1) | INT_SEL(2));
3507 radeon_ring_write(ring, fence->seq);
3508 radeon_ring_write(ring, 0);
3510 /* We should be using the new WAIT_REG_MEM special op packet here
3511 * but it causes the CP to hang
3513 radeon_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
3514 radeon_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
3515 WRITE_DATA_DST_SEL(0)));
3516 radeon_ring_write(ring, HDP_MEM_COHERENCY_FLUSH_CNTL >> 2);
3517 radeon_ring_write(ring, 0);
3518 radeon_ring_write(ring, 0);
3522 * cik_fence_compute_ring_emit - emit a fence on the compute ring
3524 * @rdev: radeon_device pointer
3525 * @fence: radeon fence object
3527 * Emits a fence sequnce number on the compute ring and flushes
3530 void cik_fence_compute_ring_emit(struct radeon_device *rdev,
3531 struct radeon_fence *fence)
3533 struct radeon_ring *ring = &rdev->ring[fence->ring];
3534 u64 addr = rdev->fence_drv[fence->ring].gpu_addr;
3536 /* RELEASE_MEM - flush caches, send int */
3537 radeon_ring_write(ring, PACKET3(PACKET3_RELEASE_MEM, 5));
3538 radeon_ring_write(ring, (EOP_TCL1_ACTION_EN |
3540 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
3542 radeon_ring_write(ring, DATA_SEL(1) | INT_SEL(2));
3543 radeon_ring_write(ring, addr & 0xfffffffc);
3544 radeon_ring_write(ring, upper_32_bits(addr));
3545 radeon_ring_write(ring, fence->seq);
3546 radeon_ring_write(ring, 0);
3548 /* We should be using the new WAIT_REG_MEM special op packet here
3549 * but it causes the CP to hang
3551 radeon_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
3552 radeon_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
3553 WRITE_DATA_DST_SEL(0)));
3554 radeon_ring_write(ring, HDP_MEM_COHERENCY_FLUSH_CNTL >> 2);
3555 radeon_ring_write(ring, 0);
3556 radeon_ring_write(ring, 0);
3559 bool cik_semaphore_ring_emit(struct radeon_device *rdev,
3560 struct radeon_ring *ring,
3561 struct radeon_semaphore *semaphore,
3564 /* TODO: figure out why semaphore cause lockups */
3566 uint64_t addr = semaphore->gpu_addr;
3567 unsigned sel = emit_wait ? PACKET3_SEM_SEL_WAIT : PACKET3_SEM_SEL_SIGNAL;
3569 radeon_ring_write(ring, PACKET3(PACKET3_MEM_SEMAPHORE, 1));
3570 radeon_ring_write(ring, addr & 0xffffffff);
3571 radeon_ring_write(ring, (upper_32_bits(addr) & 0xffff) | sel);
3580 * cik_copy_cpdma - copy pages using the CP DMA engine
3582 * @rdev: radeon_device pointer
3583 * @src_offset: src GPU address
3584 * @dst_offset: dst GPU address
3585 * @num_gpu_pages: number of GPU pages to xfer
3586 * @fence: radeon fence object
3588 * Copy GPU paging using the CP DMA engine (CIK+).
3589 * Used by the radeon ttm implementation to move pages if
3590 * registered as the asic copy callback.
3592 int cik_copy_cpdma(struct radeon_device *rdev,
3593 uint64_t src_offset, uint64_t dst_offset,
3594 unsigned num_gpu_pages,
3595 struct radeon_fence **fence)
3597 struct radeon_semaphore *sem = NULL;
3598 int ring_index = rdev->asic->copy.blit_ring_index;
3599 struct radeon_ring *ring = &rdev->ring[ring_index];
3600 u32 size_in_bytes, cur_size_in_bytes, control;
3604 r = radeon_semaphore_create(rdev, &sem);
3606 DRM_ERROR("radeon: moving bo (%d).\n", r);
3610 size_in_bytes = (num_gpu_pages << RADEON_GPU_PAGE_SHIFT);
3611 num_loops = DIV_ROUND_UP(size_in_bytes, 0x1fffff);
3612 r = radeon_ring_lock(rdev, ring, num_loops * 7 + 18);
3614 DRM_ERROR("radeon: moving bo (%d).\n", r);
3615 radeon_semaphore_free(rdev, &sem, NULL);
3619 radeon_semaphore_sync_to(sem, *fence);
3620 radeon_semaphore_sync_rings(rdev, sem, ring->idx);
3622 for (i = 0; i < num_loops; i++) {
3623 cur_size_in_bytes = size_in_bytes;
3624 if (cur_size_in_bytes > 0x1fffff)
3625 cur_size_in_bytes = 0x1fffff;
3626 size_in_bytes -= cur_size_in_bytes;
3628 if (size_in_bytes == 0)
3629 control |= PACKET3_DMA_DATA_CP_SYNC;
3630 radeon_ring_write(ring, PACKET3(PACKET3_DMA_DATA, 5));
3631 radeon_ring_write(ring, control);
3632 radeon_ring_write(ring, lower_32_bits(src_offset));
3633 radeon_ring_write(ring, upper_32_bits(src_offset));
3634 radeon_ring_write(ring, lower_32_bits(dst_offset));
3635 radeon_ring_write(ring, upper_32_bits(dst_offset));
3636 radeon_ring_write(ring, cur_size_in_bytes);
3637 src_offset += cur_size_in_bytes;
3638 dst_offset += cur_size_in_bytes;
3641 r = radeon_fence_emit(rdev, fence, ring->idx);
3643 radeon_ring_unlock_undo(rdev, ring);
3647 radeon_ring_unlock_commit(rdev, ring);
3648 radeon_semaphore_free(rdev, &sem, *fence);
3657 * cik_ring_ib_execute - emit an IB (Indirect Buffer) on the gfx ring
3659 * @rdev: radeon_device pointer
3660 * @ib: radeon indirect buffer object
3662 * Emits an DE (drawing engine) or CE (constant engine) IB
3663 * on the gfx ring. IBs are usually generated by userspace
3664 * acceleration drivers and submitted to the kernel for
3665 * sheduling on the ring. This function schedules the IB
3666 * on the gfx ring for execution by the GPU.
3668 void cik_ring_ib_execute(struct radeon_device *rdev, struct radeon_ib *ib)
3670 struct radeon_ring *ring = &rdev->ring[ib->ring];
3671 u32 header, control = INDIRECT_BUFFER_VALID;
3673 if (ib->is_const_ib) {
3674 /* set switch buffer packet before const IB */
3675 radeon_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
3676 radeon_ring_write(ring, 0);
3678 header = PACKET3(PACKET3_INDIRECT_BUFFER_CONST, 2);
3681 if (ring->rptr_save_reg) {
3682 next_rptr = ring->wptr + 3 + 4;
3683 radeon_ring_write(ring, PACKET3(PACKET3_SET_UCONFIG_REG, 1));
3684 radeon_ring_write(ring, ((ring->rptr_save_reg -
3685 PACKET3_SET_UCONFIG_REG_START) >> 2));
3686 radeon_ring_write(ring, next_rptr);
3687 } else if (rdev->wb.enabled) {
3688 next_rptr = ring->wptr + 5 + 4;
3689 radeon_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
3690 radeon_ring_write(ring, WRITE_DATA_DST_SEL(1));
3691 radeon_ring_write(ring, ring->next_rptr_gpu_addr & 0xfffffffc);
3692 radeon_ring_write(ring, upper_32_bits(ring->next_rptr_gpu_addr) & 0xffffffff);
3693 radeon_ring_write(ring, next_rptr);
3696 header = PACKET3(PACKET3_INDIRECT_BUFFER, 2);
3699 control |= ib->length_dw |
3700 (ib->vm ? (ib->vm->id << 24) : 0);
3702 radeon_ring_write(ring, header);
3703 radeon_ring_write(ring,
3707 (ib->gpu_addr & 0xFFFFFFFC));
3708 radeon_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xFFFF);
3709 radeon_ring_write(ring, control);
3713 * cik_ib_test - basic gfx ring IB test
3715 * @rdev: radeon_device pointer
3716 * @ring: radeon_ring structure holding ring information
3718 * Allocate an IB and execute it on the gfx ring (CIK).
3719 * Provides a basic gfx ring test to verify that IBs are working.
3720 * Returns 0 on success, error on failure.
3722 int cik_ib_test(struct radeon_device *rdev, struct radeon_ring *ring)
3724 struct radeon_ib ib;
3730 r = radeon_scratch_get(rdev, &scratch);
3732 DRM_ERROR("radeon: failed to get scratch reg (%d).\n", r);
3735 WREG32(scratch, 0xCAFEDEAD);
3736 r = radeon_ib_get(rdev, ring->idx, &ib, NULL, 256);
3738 DRM_ERROR("radeon: failed to get ib (%d).\n", r);
3739 radeon_scratch_free(rdev, scratch);
3742 ib.ptr[0] = PACKET3(PACKET3_SET_UCONFIG_REG, 1);
3743 ib.ptr[1] = ((scratch - PACKET3_SET_UCONFIG_REG_START) >> 2);
3744 ib.ptr[2] = 0xDEADBEEF;
3746 r = radeon_ib_schedule(rdev, &ib, NULL);
3748 radeon_scratch_free(rdev, scratch);
3749 radeon_ib_free(rdev, &ib);
3750 DRM_ERROR("radeon: failed to schedule ib (%d).\n", r);
3753 r = radeon_fence_wait(ib.fence, false);
3755 DRM_ERROR("radeon: fence wait failed (%d).\n", r);
3756 radeon_scratch_free(rdev, scratch);
3757 radeon_ib_free(rdev, &ib);
3760 for (i = 0; i < rdev->usec_timeout; i++) {
3761 tmp = RREG32(scratch);
3762 if (tmp == 0xDEADBEEF)
3766 if (i < rdev->usec_timeout) {
3767 DRM_INFO("ib test on ring %d succeeded in %u usecs\n", ib.fence->ring, i);
3769 DRM_ERROR("radeon: ib test failed (scratch(0x%04X)=0x%08X)\n",
3773 radeon_scratch_free(rdev, scratch);
3774 radeon_ib_free(rdev, &ib);
3780 * On CIK, gfx and compute now have independant command processors.
3783 * Gfx consists of a single ring and can process both gfx jobs and
3784 * compute jobs. The gfx CP consists of three microengines (ME):
3785 * PFP - Pre-Fetch Parser
3787 * CE - Constant Engine
3788 * The PFP and ME make up what is considered the Drawing Engine (DE).
3789 * The CE is an asynchronous engine used for updating buffer desciptors
3790 * used by the DE so that they can be loaded into cache in parallel
3791 * while the DE is processing state update packets.
3794 * The compute CP consists of two microengines (ME):
3795 * MEC1 - Compute MicroEngine 1
3796 * MEC2 - Compute MicroEngine 2
3797 * Each MEC supports 4 compute pipes and each pipe supports 8 queues.
3798 * The queues are exposed to userspace and are programmed directly
3799 * by the compute runtime.
3802 * cik_cp_gfx_enable - enable/disable the gfx CP MEs
3804 * @rdev: radeon_device pointer
3805 * @enable: enable or disable the MEs
3807 * Halts or unhalts the gfx MEs.
3809 static void cik_cp_gfx_enable(struct radeon_device *rdev, bool enable)
3812 WREG32(CP_ME_CNTL, 0);
3814 WREG32(CP_ME_CNTL, (CP_ME_HALT | CP_PFP_HALT | CP_CE_HALT));
3815 rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = false;
3821 * cik_cp_gfx_load_microcode - load the gfx CP ME ucode
3823 * @rdev: radeon_device pointer
3825 * Loads the gfx PFP, ME, and CE ucode.
3826 * Returns 0 for success, -EINVAL if the ucode is not available.
3828 static int cik_cp_gfx_load_microcode(struct radeon_device *rdev)
3830 const __be32 *fw_data;
3833 if (!rdev->me_fw || !rdev->pfp_fw || !rdev->ce_fw)
3836 cik_cp_gfx_enable(rdev, false);
3839 fw_data = (const __be32 *)rdev->pfp_fw->data;
3840 WREG32(CP_PFP_UCODE_ADDR, 0);
3841 for (i = 0; i < CIK_PFP_UCODE_SIZE; i++)
3842 WREG32(CP_PFP_UCODE_DATA, be32_to_cpup(fw_data++));
3843 WREG32(CP_PFP_UCODE_ADDR, 0);
3846 fw_data = (const __be32 *)rdev->ce_fw->data;
3847 WREG32(CP_CE_UCODE_ADDR, 0);
3848 for (i = 0; i < CIK_CE_UCODE_SIZE; i++)
3849 WREG32(CP_CE_UCODE_DATA, be32_to_cpup(fw_data++));
3850 WREG32(CP_CE_UCODE_ADDR, 0);
3853 fw_data = (const __be32 *)rdev->me_fw->data;
3854 WREG32(CP_ME_RAM_WADDR, 0);
3855 for (i = 0; i < CIK_ME_UCODE_SIZE; i++)
3856 WREG32(CP_ME_RAM_DATA, be32_to_cpup(fw_data++));
3857 WREG32(CP_ME_RAM_WADDR, 0);
3859 WREG32(CP_PFP_UCODE_ADDR, 0);
3860 WREG32(CP_CE_UCODE_ADDR, 0);
3861 WREG32(CP_ME_RAM_WADDR, 0);
3862 WREG32(CP_ME_RAM_RADDR, 0);
3867 * cik_cp_gfx_start - start the gfx ring
3869 * @rdev: radeon_device pointer
3871 * Enables the ring and loads the clear state context and other
3872 * packets required to init the ring.
3873 * Returns 0 for success, error for failure.
3875 static int cik_cp_gfx_start(struct radeon_device *rdev)
3877 struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
3881 WREG32(CP_MAX_CONTEXT, rdev->config.cik.max_hw_contexts - 1);
3882 WREG32(CP_ENDIAN_SWAP, 0);
3883 WREG32(CP_DEVICE_ID, 1);
3885 cik_cp_gfx_enable(rdev, true);
3887 r = radeon_ring_lock(rdev, ring, cik_default_size + 17);
3889 DRM_ERROR("radeon: cp failed to lock ring (%d).\n", r);
3893 /* init the CE partitions. CE only used for gfx on CIK */
3894 radeon_ring_write(ring, PACKET3(PACKET3_SET_BASE, 2));
3895 radeon_ring_write(ring, PACKET3_BASE_INDEX(CE_PARTITION_BASE));
3896 radeon_ring_write(ring, 0xc000);
3897 radeon_ring_write(ring, 0xc000);
3899 /* setup clear context state */
3900 radeon_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
3901 radeon_ring_write(ring, PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
3903 radeon_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1));
3904 radeon_ring_write(ring, 0x80000000);
3905 radeon_ring_write(ring, 0x80000000);
3907 for (i = 0; i < cik_default_size; i++)
3908 radeon_ring_write(ring, cik_default_state[i]);
3910 radeon_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
3911 radeon_ring_write(ring, PACKET3_PREAMBLE_END_CLEAR_STATE);
3913 /* set clear context state */
3914 radeon_ring_write(ring, PACKET3(PACKET3_CLEAR_STATE, 0));
3915 radeon_ring_write(ring, 0);
3917 radeon_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 2));
3918 radeon_ring_write(ring, 0x00000316);
3919 radeon_ring_write(ring, 0x0000000e); /* VGT_VERTEX_REUSE_BLOCK_CNTL */
3920 radeon_ring_write(ring, 0x00000010); /* VGT_OUT_DEALLOC_CNTL */
3922 radeon_ring_unlock_commit(rdev, ring);
3928 * cik_cp_gfx_fini - stop the gfx ring
3930 * @rdev: radeon_device pointer
3932 * Stop the gfx ring and tear down the driver ring
3935 static void cik_cp_gfx_fini(struct radeon_device *rdev)
3937 cik_cp_gfx_enable(rdev, false);
3938 radeon_ring_fini(rdev, &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]);
3942 * cik_cp_gfx_resume - setup the gfx ring buffer registers
3944 * @rdev: radeon_device pointer
3946 * Program the location and size of the gfx ring buffer
3947 * and test it to make sure it's working.
3948 * Returns 0 for success, error for failure.
3950 static int cik_cp_gfx_resume(struct radeon_device *rdev)
3952 struct radeon_ring *ring;
3958 WREG32(CP_SEM_WAIT_TIMER, 0x0);
3959 if (rdev->family != CHIP_HAWAII)
3960 WREG32(CP_SEM_INCOMPLETE_TIMER_CNTL, 0x0);
3962 /* Set the write pointer delay */
3963 WREG32(CP_RB_WPTR_DELAY, 0);
3965 /* set the RB to use vmid 0 */
3966 WREG32(CP_RB_VMID, 0);
3968 WREG32(SCRATCH_ADDR, ((rdev->wb.gpu_addr + RADEON_WB_SCRATCH_OFFSET) >> 8) & 0xFFFFFFFF);
3970 /* ring 0 - compute and gfx */
3971 /* Set ring buffer size */
3972 ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
3973 rb_bufsz = order_base_2(ring->ring_size / 8);
3974 tmp = (order_base_2(RADEON_GPU_PAGE_SIZE/8) << 8) | rb_bufsz;
3976 tmp |= BUF_SWAP_32BIT;
3978 WREG32(CP_RB0_CNTL, tmp);
3980 /* Initialize the ring buffer's read and write pointers */
3981 WREG32(CP_RB0_CNTL, tmp | RB_RPTR_WR_ENA);
3983 WREG32(CP_RB0_WPTR, ring->wptr);
3985 /* set the wb address wether it's enabled or not */
3986 WREG32(CP_RB0_RPTR_ADDR, (rdev->wb.gpu_addr + RADEON_WB_CP_RPTR_OFFSET) & 0xFFFFFFFC);
3987 WREG32(CP_RB0_RPTR_ADDR_HI, upper_32_bits(rdev->wb.gpu_addr + RADEON_WB_CP_RPTR_OFFSET) & 0xFF);
3989 /* scratch register shadowing is no longer supported */
3990 WREG32(SCRATCH_UMSK, 0);
3992 if (!rdev->wb.enabled)
3993 tmp |= RB_NO_UPDATE;
3996 WREG32(CP_RB0_CNTL, tmp);
3998 rb_addr = ring->gpu_addr >> 8;
3999 WREG32(CP_RB0_BASE, rb_addr);
4000 WREG32(CP_RB0_BASE_HI, upper_32_bits(rb_addr));
4002 ring->rptr = RREG32(CP_RB0_RPTR);
4004 /* start the ring */
4005 cik_cp_gfx_start(rdev);
4006 rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = true;
4007 r = radeon_ring_test(rdev, RADEON_RING_TYPE_GFX_INDEX, &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]);
4009 rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = false;
4015 u32 cik_compute_ring_get_rptr(struct radeon_device *rdev,
4016 struct radeon_ring *ring)
4022 if (rdev->wb.enabled) {
4023 rptr = le32_to_cpu(rdev->wb.wb[ring->rptr_offs/4]);
4025 mutex_lock(&rdev->srbm_mutex);
4026 cik_srbm_select(rdev, ring->me, ring->pipe, ring->queue, 0);
4027 rptr = RREG32(CP_HQD_PQ_RPTR);
4028 cik_srbm_select(rdev, 0, 0, 0, 0);
4029 mutex_unlock(&rdev->srbm_mutex);
4035 u32 cik_compute_ring_get_wptr(struct radeon_device *rdev,
4036 struct radeon_ring *ring)
4040 if (rdev->wb.enabled) {
4041 wptr = le32_to_cpu(rdev->wb.wb[ring->wptr_offs/4]);
4043 mutex_lock(&rdev->srbm_mutex);
4044 cik_srbm_select(rdev, ring->me, ring->pipe, ring->queue, 0);
4045 wptr = RREG32(CP_HQD_PQ_WPTR);
4046 cik_srbm_select(rdev, 0, 0, 0, 0);
4047 mutex_unlock(&rdev->srbm_mutex);
4053 void cik_compute_ring_set_wptr(struct radeon_device *rdev,
4054 struct radeon_ring *ring)
4056 rdev->wb.wb[ring->wptr_offs/4] = cpu_to_le32(ring->wptr);
4057 WDOORBELL32(ring->doorbell_index, ring->wptr);
4061 * cik_cp_compute_enable - enable/disable the compute CP MEs
4063 * @rdev: radeon_device pointer
4064 * @enable: enable or disable the MEs
4066 * Halts or unhalts the compute MEs.
4068 static void cik_cp_compute_enable(struct radeon_device *rdev, bool enable)
4071 WREG32(CP_MEC_CNTL, 0);
4073 WREG32(CP_MEC_CNTL, (MEC_ME1_HALT | MEC_ME2_HALT));
4078 * cik_cp_compute_load_microcode - load the compute CP ME ucode
4080 * @rdev: radeon_device pointer
4082 * Loads the compute MEC1&2 ucode.
4083 * Returns 0 for success, -EINVAL if the ucode is not available.
4085 static int cik_cp_compute_load_microcode(struct radeon_device *rdev)
4087 const __be32 *fw_data;
4093 cik_cp_compute_enable(rdev, false);
4096 fw_data = (const __be32 *)rdev->mec_fw->data;
4097 WREG32(CP_MEC_ME1_UCODE_ADDR, 0);
4098 for (i = 0; i < CIK_MEC_UCODE_SIZE; i++)
4099 WREG32(CP_MEC_ME1_UCODE_DATA, be32_to_cpup(fw_data++));
4100 WREG32(CP_MEC_ME1_UCODE_ADDR, 0);
4102 if (rdev->family == CHIP_KAVERI) {
4104 fw_data = (const __be32 *)rdev->mec_fw->data;
4105 WREG32(CP_MEC_ME2_UCODE_ADDR, 0);
4106 for (i = 0; i < CIK_MEC_UCODE_SIZE; i++)
4107 WREG32(CP_MEC_ME2_UCODE_DATA, be32_to_cpup(fw_data++));
4108 WREG32(CP_MEC_ME2_UCODE_ADDR, 0);
4115 * cik_cp_compute_start - start the compute queues
4117 * @rdev: radeon_device pointer
4119 * Enable the compute queues.
4120 * Returns 0 for success, error for failure.
4122 static int cik_cp_compute_start(struct radeon_device *rdev)
4124 cik_cp_compute_enable(rdev, true);
4130 * cik_cp_compute_fini - stop the compute queues
4132 * @rdev: radeon_device pointer
4134 * Stop the compute queues and tear down the driver queue
4137 static void cik_cp_compute_fini(struct radeon_device *rdev)
4141 cik_cp_compute_enable(rdev, false);
4143 for (i = 0; i < 2; i++) {
4145 idx = CAYMAN_RING_TYPE_CP1_INDEX;
4147 idx = CAYMAN_RING_TYPE_CP2_INDEX;
4149 if (rdev->ring[idx].mqd_obj) {
4150 r = radeon_bo_reserve(rdev->ring[idx].mqd_obj, false);
4151 if (unlikely(r != 0))
4152 dev_warn(rdev->dev, "(%d) reserve MQD bo failed\n", r);
4154 radeon_bo_unpin(rdev->ring[idx].mqd_obj);
4155 radeon_bo_unreserve(rdev->ring[idx].mqd_obj);
4157 radeon_bo_unref(&rdev->ring[idx].mqd_obj);
4158 rdev->ring[idx].mqd_obj = NULL;
4163 static void cik_mec_fini(struct radeon_device *rdev)
4167 if (rdev->mec.hpd_eop_obj) {
4168 r = radeon_bo_reserve(rdev->mec.hpd_eop_obj, false);
4169 if (unlikely(r != 0))
4170 dev_warn(rdev->dev, "(%d) reserve HPD EOP bo failed\n", r);
4171 radeon_bo_unpin(rdev->mec.hpd_eop_obj);
4172 radeon_bo_unreserve(rdev->mec.hpd_eop_obj);
4174 radeon_bo_unref(&rdev->mec.hpd_eop_obj);
4175 rdev->mec.hpd_eop_obj = NULL;
4179 #define MEC_HPD_SIZE 2048
4181 static int cik_mec_init(struct radeon_device *rdev)
4187 * KV: 2 MEC, 4 Pipes/MEC, 8 Queues/Pipe - 64 Queues total
4188 * CI/KB: 1 MEC, 4 Pipes/MEC, 8 Queues/Pipe - 32 Queues total
4190 if (rdev->family == CHIP_KAVERI)
4191 rdev->mec.num_mec = 2;
4193 rdev->mec.num_mec = 1;
4194 rdev->mec.num_pipe = 4;
4195 rdev->mec.num_queue = rdev->mec.num_mec * rdev->mec.num_pipe * 8;
4197 if (rdev->mec.hpd_eop_obj == NULL) {
4198 r = radeon_bo_create(rdev,
4199 rdev->mec.num_mec *rdev->mec.num_pipe * MEC_HPD_SIZE * 2,
4201 RADEON_GEM_DOMAIN_GTT, NULL,
4202 &rdev->mec.hpd_eop_obj);
4204 dev_warn(rdev->dev, "(%d) create HDP EOP bo failed\n", r);
4209 r = radeon_bo_reserve(rdev->mec.hpd_eop_obj, false);
4210 if (unlikely(r != 0)) {
4214 r = radeon_bo_pin(rdev->mec.hpd_eop_obj, RADEON_GEM_DOMAIN_GTT,
4215 &rdev->mec.hpd_eop_gpu_addr);
4217 dev_warn(rdev->dev, "(%d) pin HDP EOP bo failed\n", r);
4221 r = radeon_bo_kmap(rdev->mec.hpd_eop_obj, (void **)&hpd);
4223 dev_warn(rdev->dev, "(%d) map HDP EOP bo failed\n", r);
4228 /* clear memory. Not sure if this is required or not */
4229 memset(hpd, 0, rdev->mec.num_mec *rdev->mec.num_pipe * MEC_HPD_SIZE * 2);
4231 radeon_bo_kunmap(rdev->mec.hpd_eop_obj);
4232 radeon_bo_unreserve(rdev->mec.hpd_eop_obj);
4237 struct hqd_registers
4239 u32 cp_mqd_base_addr;
4240 u32 cp_mqd_base_addr_hi;
4243 u32 cp_hqd_persistent_state;
4244 u32 cp_hqd_pipe_priority;
4245 u32 cp_hqd_queue_priority;
4248 u32 cp_hqd_pq_base_hi;
4250 u32 cp_hqd_pq_rptr_report_addr;
4251 u32 cp_hqd_pq_rptr_report_addr_hi;
4252 u32 cp_hqd_pq_wptr_poll_addr;
4253 u32 cp_hqd_pq_wptr_poll_addr_hi;
4254 u32 cp_hqd_pq_doorbell_control;
4256 u32 cp_hqd_pq_control;
4257 u32 cp_hqd_ib_base_addr;
4258 u32 cp_hqd_ib_base_addr_hi;
4260 u32 cp_hqd_ib_control;
4261 u32 cp_hqd_iq_timer;
4263 u32 cp_hqd_dequeue_request;
4264 u32 cp_hqd_dma_offload;
4265 u32 cp_hqd_sema_cmd;
4266 u32 cp_hqd_msg_type;
4267 u32 cp_hqd_atomic0_preop_lo;
4268 u32 cp_hqd_atomic0_preop_hi;
4269 u32 cp_hqd_atomic1_preop_lo;
4270 u32 cp_hqd_atomic1_preop_hi;
4271 u32 cp_hqd_hq_scheduler0;
4272 u32 cp_hqd_hq_scheduler1;
4279 u32 dispatch_initiator;
4283 u32 pipeline_stat_enable;
4284 u32 perf_counter_enable;
4290 u32 resource_limits;
4291 u32 static_thread_mgmt01[2];
4293 u32 static_thread_mgmt23[2];
4295 u32 thread_trace_enable;
4298 u32 vgtcs_invoke_count[2];
4299 struct hqd_registers queue_state;
4301 u32 interrupt_queue[64];
4305 * cik_cp_compute_resume - setup the compute queue registers
4307 * @rdev: radeon_device pointer
4309 * Program the compute queues and test them to make sure they
4311 * Returns 0 for success, error for failure.
4313 static int cik_cp_compute_resume(struct radeon_device *rdev)
4317 bool use_doorbell = true;
4323 struct bonaire_mqd *mqd;
4325 r = cik_cp_compute_start(rdev);
4329 /* fix up chicken bits */
4330 tmp = RREG32(CP_CPF_DEBUG);
4332 WREG32(CP_CPF_DEBUG, tmp);
4334 /* init the pipes */
4335 mutex_lock(&rdev->srbm_mutex);
4336 for (i = 0; i < (rdev->mec.num_pipe * rdev->mec.num_mec); i++) {
4337 int me = (i < 4) ? 1 : 2;
4338 int pipe = (i < 4) ? i : (i - 4);
4340 eop_gpu_addr = rdev->mec.hpd_eop_gpu_addr + (i * MEC_HPD_SIZE * 2);
4342 cik_srbm_select(rdev, me, pipe, 0, 0);
4344 /* write the EOP addr */
4345 WREG32(CP_HPD_EOP_BASE_ADDR, eop_gpu_addr >> 8);
4346 WREG32(CP_HPD_EOP_BASE_ADDR_HI, upper_32_bits(eop_gpu_addr) >> 8);
4348 /* set the VMID assigned */
4349 WREG32(CP_HPD_EOP_VMID, 0);
4351 /* set the EOP size, register value is 2^(EOP_SIZE+1) dwords */
4352 tmp = RREG32(CP_HPD_EOP_CONTROL);
4353 tmp &= ~EOP_SIZE_MASK;
4354 tmp |= order_base_2(MEC_HPD_SIZE / 8);
4355 WREG32(CP_HPD_EOP_CONTROL, tmp);
4357 cik_srbm_select(rdev, 0, 0, 0, 0);
4358 mutex_unlock(&rdev->srbm_mutex);
4360 /* init the queues. Just two for now. */
4361 for (i = 0; i < 2; i++) {
4363 idx = CAYMAN_RING_TYPE_CP1_INDEX;
4365 idx = CAYMAN_RING_TYPE_CP2_INDEX;
4367 if (rdev->ring[idx].mqd_obj == NULL) {
4368 r = radeon_bo_create(rdev,
4369 sizeof(struct bonaire_mqd),
4371 RADEON_GEM_DOMAIN_GTT, NULL,
4372 &rdev->ring[idx].mqd_obj);
4374 dev_warn(rdev->dev, "(%d) create MQD bo failed\n", r);
4379 r = radeon_bo_reserve(rdev->ring[idx].mqd_obj, false);
4380 if (unlikely(r != 0)) {
4381 cik_cp_compute_fini(rdev);
4384 r = radeon_bo_pin(rdev->ring[idx].mqd_obj, RADEON_GEM_DOMAIN_GTT,
4387 dev_warn(rdev->dev, "(%d) pin MQD bo failed\n", r);
4388 cik_cp_compute_fini(rdev);
4391 r = radeon_bo_kmap(rdev->ring[idx].mqd_obj, (void **)&buf);
4393 dev_warn(rdev->dev, "(%d) map MQD bo failed\n", r);
4394 cik_cp_compute_fini(rdev);
4398 /* init the mqd struct */
4399 memset(buf, 0, sizeof(struct bonaire_mqd));
4401 mqd = (struct bonaire_mqd *)buf;
4402 mqd->header = 0xC0310800;
4403 mqd->static_thread_mgmt01[0] = 0xffffffff;
4404 mqd->static_thread_mgmt01[1] = 0xffffffff;
4405 mqd->static_thread_mgmt23[0] = 0xffffffff;
4406 mqd->static_thread_mgmt23[1] = 0xffffffff;
4408 mutex_lock(&rdev->srbm_mutex);
4409 cik_srbm_select(rdev, rdev->ring[idx].me,
4410 rdev->ring[idx].pipe,
4411 rdev->ring[idx].queue, 0);
4413 /* disable wptr polling */
4414 tmp = RREG32(CP_PQ_WPTR_POLL_CNTL);
4415 tmp &= ~WPTR_POLL_EN;
4416 WREG32(CP_PQ_WPTR_POLL_CNTL, tmp);
4418 /* enable doorbell? */
4419 mqd->queue_state.cp_hqd_pq_doorbell_control =
4420 RREG32(CP_HQD_PQ_DOORBELL_CONTROL);
4422 mqd->queue_state.cp_hqd_pq_doorbell_control |= DOORBELL_EN;
4424 mqd->queue_state.cp_hqd_pq_doorbell_control &= ~DOORBELL_EN;
4425 WREG32(CP_HQD_PQ_DOORBELL_CONTROL,
4426 mqd->queue_state.cp_hqd_pq_doorbell_control);
4428 /* disable the queue if it's active */
4429 mqd->queue_state.cp_hqd_dequeue_request = 0;
4430 mqd->queue_state.cp_hqd_pq_rptr = 0;
4431 mqd->queue_state.cp_hqd_pq_wptr= 0;
4432 if (RREG32(CP_HQD_ACTIVE) & 1) {
4433 WREG32(CP_HQD_DEQUEUE_REQUEST, 1);
4434 for (i = 0; i < rdev->usec_timeout; i++) {
4435 if (!(RREG32(CP_HQD_ACTIVE) & 1))
4439 WREG32(CP_HQD_DEQUEUE_REQUEST, mqd->queue_state.cp_hqd_dequeue_request);
4440 WREG32(CP_HQD_PQ_RPTR, mqd->queue_state.cp_hqd_pq_rptr);
4441 WREG32(CP_HQD_PQ_WPTR, mqd->queue_state.cp_hqd_pq_wptr);
4444 /* set the pointer to the MQD */
4445 mqd->queue_state.cp_mqd_base_addr = mqd_gpu_addr & 0xfffffffc;
4446 mqd->queue_state.cp_mqd_base_addr_hi = upper_32_bits(mqd_gpu_addr);
4447 WREG32(CP_MQD_BASE_ADDR, mqd->queue_state.cp_mqd_base_addr);
4448 WREG32(CP_MQD_BASE_ADDR_HI, mqd->queue_state.cp_mqd_base_addr_hi);
4449 /* set MQD vmid to 0 */
4450 mqd->queue_state.cp_mqd_control = RREG32(CP_MQD_CONTROL);
4451 mqd->queue_state.cp_mqd_control &= ~MQD_VMID_MASK;
4452 WREG32(CP_MQD_CONTROL, mqd->queue_state.cp_mqd_control);
4454 /* set the pointer to the HQD, this is similar CP_RB0_BASE/_HI */
4455 hqd_gpu_addr = rdev->ring[idx].gpu_addr >> 8;
4456 mqd->queue_state.cp_hqd_pq_base = hqd_gpu_addr;
4457 mqd->queue_state.cp_hqd_pq_base_hi = upper_32_bits(hqd_gpu_addr);
4458 WREG32(CP_HQD_PQ_BASE, mqd->queue_state.cp_hqd_pq_base);
4459 WREG32(CP_HQD_PQ_BASE_HI, mqd->queue_state.cp_hqd_pq_base_hi);
4461 /* set up the HQD, this is similar to CP_RB0_CNTL */
4462 mqd->queue_state.cp_hqd_pq_control = RREG32(CP_HQD_PQ_CONTROL);
4463 mqd->queue_state.cp_hqd_pq_control &=
4464 ~(QUEUE_SIZE_MASK | RPTR_BLOCK_SIZE_MASK);
4466 mqd->queue_state.cp_hqd_pq_control |=
4467 order_base_2(rdev->ring[idx].ring_size / 8);
4468 mqd->queue_state.cp_hqd_pq_control |=
4469 (order_base_2(RADEON_GPU_PAGE_SIZE/8) << 8);
4471 mqd->queue_state.cp_hqd_pq_control |= BUF_SWAP_32BIT;
4473 mqd->queue_state.cp_hqd_pq_control &=
4474 ~(UNORD_DISPATCH | ROQ_PQ_IB_FLIP | PQ_VOLATILE);
4475 mqd->queue_state.cp_hqd_pq_control |=
4476 PRIV_STATE | KMD_QUEUE; /* assuming kernel queue control */
4477 WREG32(CP_HQD_PQ_CONTROL, mqd->queue_state.cp_hqd_pq_control);
4479 /* only used if CP_PQ_WPTR_POLL_CNTL.WPTR_POLL_EN=1 */
4481 wb_gpu_addr = rdev->wb.gpu_addr + CIK_WB_CP1_WPTR_OFFSET;
4483 wb_gpu_addr = rdev->wb.gpu_addr + CIK_WB_CP2_WPTR_OFFSET;
4484 mqd->queue_state.cp_hqd_pq_wptr_poll_addr = wb_gpu_addr & 0xfffffffc;
4485 mqd->queue_state.cp_hqd_pq_wptr_poll_addr_hi = upper_32_bits(wb_gpu_addr) & 0xffff;
4486 WREG32(CP_HQD_PQ_WPTR_POLL_ADDR, mqd->queue_state.cp_hqd_pq_wptr_poll_addr);
4487 WREG32(CP_HQD_PQ_WPTR_POLL_ADDR_HI,
4488 mqd->queue_state.cp_hqd_pq_wptr_poll_addr_hi);
4490 /* set the wb address wether it's enabled or not */
4492 wb_gpu_addr = rdev->wb.gpu_addr + RADEON_WB_CP1_RPTR_OFFSET;
4494 wb_gpu_addr = rdev->wb.gpu_addr + RADEON_WB_CP2_RPTR_OFFSET;
4495 mqd->queue_state.cp_hqd_pq_rptr_report_addr = wb_gpu_addr & 0xfffffffc;
4496 mqd->queue_state.cp_hqd_pq_rptr_report_addr_hi =
4497 upper_32_bits(wb_gpu_addr) & 0xffff;
4498 WREG32(CP_HQD_PQ_RPTR_REPORT_ADDR,
4499 mqd->queue_state.cp_hqd_pq_rptr_report_addr);
4500 WREG32(CP_HQD_PQ_RPTR_REPORT_ADDR_HI,
4501 mqd->queue_state.cp_hqd_pq_rptr_report_addr_hi);
4503 /* enable the doorbell if requested */
4505 mqd->queue_state.cp_hqd_pq_doorbell_control =
4506 RREG32(CP_HQD_PQ_DOORBELL_CONTROL);
4507 mqd->queue_state.cp_hqd_pq_doorbell_control &= ~DOORBELL_OFFSET_MASK;
4508 mqd->queue_state.cp_hqd_pq_doorbell_control |=
4509 DOORBELL_OFFSET(rdev->ring[idx].doorbell_index);
4510 mqd->queue_state.cp_hqd_pq_doorbell_control |= DOORBELL_EN;
4511 mqd->queue_state.cp_hqd_pq_doorbell_control &=
4512 ~(DOORBELL_SOURCE | DOORBELL_HIT);
4515 mqd->queue_state.cp_hqd_pq_doorbell_control = 0;
4517 WREG32(CP_HQD_PQ_DOORBELL_CONTROL,
4518 mqd->queue_state.cp_hqd_pq_doorbell_control);
4520 /* read and write pointers, similar to CP_RB0_WPTR/_RPTR */
4521 rdev->ring[idx].wptr = 0;
4522 mqd->queue_state.cp_hqd_pq_wptr = rdev->ring[idx].wptr;
4523 WREG32(CP_HQD_PQ_WPTR, mqd->queue_state.cp_hqd_pq_wptr);
4524 rdev->ring[idx].rptr = RREG32(CP_HQD_PQ_RPTR);
4525 mqd->queue_state.cp_hqd_pq_rptr = rdev->ring[idx].rptr;
4527 /* set the vmid for the queue */
4528 mqd->queue_state.cp_hqd_vmid = 0;
4529 WREG32(CP_HQD_VMID, mqd->queue_state.cp_hqd_vmid);
4531 /* activate the queue */
4532 mqd->queue_state.cp_hqd_active = 1;
4533 WREG32(CP_HQD_ACTIVE, mqd->queue_state.cp_hqd_active);
4535 cik_srbm_select(rdev, 0, 0, 0, 0);
4536 mutex_unlock(&rdev->srbm_mutex);
4538 radeon_bo_kunmap(rdev->ring[idx].mqd_obj);
4539 radeon_bo_unreserve(rdev->ring[idx].mqd_obj);
4541 rdev->ring[idx].ready = true;
4542 r = radeon_ring_test(rdev, idx, &rdev->ring[idx]);
4544 rdev->ring[idx].ready = false;
4550 static void cik_cp_enable(struct radeon_device *rdev, bool enable)
4552 cik_cp_gfx_enable(rdev, enable);
4553 cik_cp_compute_enable(rdev, enable);
4556 static int cik_cp_load_microcode(struct radeon_device *rdev)
4560 r = cik_cp_gfx_load_microcode(rdev);
4563 r = cik_cp_compute_load_microcode(rdev);
4570 static void cik_cp_fini(struct radeon_device *rdev)
4572 cik_cp_gfx_fini(rdev);
4573 cik_cp_compute_fini(rdev);
4576 static int cik_cp_resume(struct radeon_device *rdev)
4580 cik_enable_gui_idle_interrupt(rdev, false);
4582 r = cik_cp_load_microcode(rdev);
4586 r = cik_cp_gfx_resume(rdev);
4589 r = cik_cp_compute_resume(rdev);
4593 cik_enable_gui_idle_interrupt(rdev, true);
4598 static void cik_print_gpu_status_regs(struct radeon_device *rdev)
4600 dev_info(rdev->dev, " GRBM_STATUS=0x%08X\n",
4601 RREG32(GRBM_STATUS));
4602 dev_info(rdev->dev, " GRBM_STATUS2=0x%08X\n",
4603 RREG32(GRBM_STATUS2));
4604 dev_info(rdev->dev, " GRBM_STATUS_SE0=0x%08X\n",
4605 RREG32(GRBM_STATUS_SE0));
4606 dev_info(rdev->dev, " GRBM_STATUS_SE1=0x%08X\n",
4607 RREG32(GRBM_STATUS_SE1));
4608 dev_info(rdev->dev, " GRBM_STATUS_SE2=0x%08X\n",
4609 RREG32(GRBM_STATUS_SE2));
4610 dev_info(rdev->dev, " GRBM_STATUS_SE3=0x%08X\n",
4611 RREG32(GRBM_STATUS_SE3));
4612 dev_info(rdev->dev, " SRBM_STATUS=0x%08X\n",
4613 RREG32(SRBM_STATUS));
4614 dev_info(rdev->dev, " SRBM_STATUS2=0x%08X\n",
4615 RREG32(SRBM_STATUS2));
4616 dev_info(rdev->dev, " SDMA0_STATUS_REG = 0x%08X\n",
4617 RREG32(SDMA0_STATUS_REG + SDMA0_REGISTER_OFFSET));
4618 dev_info(rdev->dev, " SDMA1_STATUS_REG = 0x%08X\n",
4619 RREG32(SDMA0_STATUS_REG + SDMA1_REGISTER_OFFSET));
4620 dev_info(rdev->dev, " CP_STAT = 0x%08x\n", RREG32(CP_STAT));
4621 dev_info(rdev->dev, " CP_STALLED_STAT1 = 0x%08x\n",
4622 RREG32(CP_STALLED_STAT1));
4623 dev_info(rdev->dev, " CP_STALLED_STAT2 = 0x%08x\n",
4624 RREG32(CP_STALLED_STAT2));
4625 dev_info(rdev->dev, " CP_STALLED_STAT3 = 0x%08x\n",
4626 RREG32(CP_STALLED_STAT3));
4627 dev_info(rdev->dev, " CP_CPF_BUSY_STAT = 0x%08x\n",
4628 RREG32(CP_CPF_BUSY_STAT));
4629 dev_info(rdev->dev, " CP_CPF_STALLED_STAT1 = 0x%08x\n",
4630 RREG32(CP_CPF_STALLED_STAT1));
4631 dev_info(rdev->dev, " CP_CPF_STATUS = 0x%08x\n", RREG32(CP_CPF_STATUS));
4632 dev_info(rdev->dev, " CP_CPC_BUSY_STAT = 0x%08x\n", RREG32(CP_CPC_BUSY_STAT));
4633 dev_info(rdev->dev, " CP_CPC_STALLED_STAT1 = 0x%08x\n",
4634 RREG32(CP_CPC_STALLED_STAT1));
4635 dev_info(rdev->dev, " CP_CPC_STATUS = 0x%08x\n", RREG32(CP_CPC_STATUS));
4639 * cik_gpu_check_soft_reset - check which blocks are busy
4641 * @rdev: radeon_device pointer
4643 * Check which blocks are busy and return the relevant reset
4644 * mask to be used by cik_gpu_soft_reset().
4645 * Returns a mask of the blocks to be reset.
4647 u32 cik_gpu_check_soft_reset(struct radeon_device *rdev)
4653 tmp = RREG32(GRBM_STATUS);
4654 if (tmp & (PA_BUSY | SC_BUSY |
4655 BCI_BUSY | SX_BUSY |
4656 TA_BUSY | VGT_BUSY |
4658 GDS_BUSY | SPI_BUSY |
4659 IA_BUSY | IA_BUSY_NO_DMA))
4660 reset_mask |= RADEON_RESET_GFX;
4662 if (tmp & (CP_BUSY | CP_COHERENCY_BUSY))
4663 reset_mask |= RADEON_RESET_CP;
4666 tmp = RREG32(GRBM_STATUS2);
4668 reset_mask |= RADEON_RESET_RLC;
4670 /* SDMA0_STATUS_REG */
4671 tmp = RREG32(SDMA0_STATUS_REG + SDMA0_REGISTER_OFFSET);
4672 if (!(tmp & SDMA_IDLE))
4673 reset_mask |= RADEON_RESET_DMA;
4675 /* SDMA1_STATUS_REG */
4676 tmp = RREG32(SDMA0_STATUS_REG + SDMA1_REGISTER_OFFSET);
4677 if (!(tmp & SDMA_IDLE))
4678 reset_mask |= RADEON_RESET_DMA1;
4681 tmp = RREG32(SRBM_STATUS2);
4682 if (tmp & SDMA_BUSY)
4683 reset_mask |= RADEON_RESET_DMA;
4685 if (tmp & SDMA1_BUSY)
4686 reset_mask |= RADEON_RESET_DMA1;
4689 tmp = RREG32(SRBM_STATUS);
4692 reset_mask |= RADEON_RESET_IH;
4695 reset_mask |= RADEON_RESET_SEM;
4697 if (tmp & GRBM_RQ_PENDING)
4698 reset_mask |= RADEON_RESET_GRBM;
4701 reset_mask |= RADEON_RESET_VMC;
4703 if (tmp & (MCB_BUSY | MCB_NON_DISPLAY_BUSY |
4704 MCC_BUSY | MCD_BUSY))
4705 reset_mask |= RADEON_RESET_MC;
4707 if (evergreen_is_display_hung(rdev))
4708 reset_mask |= RADEON_RESET_DISPLAY;
4710 /* Skip MC reset as it's mostly likely not hung, just busy */
4711 if (reset_mask & RADEON_RESET_MC) {
4712 DRM_DEBUG("MC busy: 0x%08X, clearing.\n", reset_mask);
4713 reset_mask &= ~RADEON_RESET_MC;
4720 * cik_gpu_soft_reset - soft reset GPU
4722 * @rdev: radeon_device pointer
4723 * @reset_mask: mask of which blocks to reset
4725 * Soft reset the blocks specified in @reset_mask.
4727 static void cik_gpu_soft_reset(struct radeon_device *rdev, u32 reset_mask)
4729 struct evergreen_mc_save save;
4730 u32 grbm_soft_reset = 0, srbm_soft_reset = 0;
4733 if (reset_mask == 0)
4736 dev_info(rdev->dev, "GPU softreset: 0x%08X\n", reset_mask);
4738 cik_print_gpu_status_regs(rdev);
4739 dev_info(rdev->dev, " VM_CONTEXT1_PROTECTION_FAULT_ADDR 0x%08X\n",
4740 RREG32(VM_CONTEXT1_PROTECTION_FAULT_ADDR));
4741 dev_info(rdev->dev, " VM_CONTEXT1_PROTECTION_FAULT_STATUS 0x%08X\n",
4742 RREG32(VM_CONTEXT1_PROTECTION_FAULT_STATUS));
4751 /* Disable GFX parsing/prefetching */
4752 WREG32(CP_ME_CNTL, CP_ME_HALT | CP_PFP_HALT | CP_CE_HALT);
4754 /* Disable MEC parsing/prefetching */
4755 WREG32(CP_MEC_CNTL, MEC_ME1_HALT | MEC_ME2_HALT);
4757 if (reset_mask & RADEON_RESET_DMA) {
4759 tmp = RREG32(SDMA0_ME_CNTL + SDMA0_REGISTER_OFFSET);
4761 WREG32(SDMA0_ME_CNTL + SDMA0_REGISTER_OFFSET, tmp);
4763 if (reset_mask & RADEON_RESET_DMA1) {
4765 tmp = RREG32(SDMA0_ME_CNTL + SDMA1_REGISTER_OFFSET);
4767 WREG32(SDMA0_ME_CNTL + SDMA1_REGISTER_OFFSET, tmp);
4770 evergreen_mc_stop(rdev, &save);
4771 if (evergreen_mc_wait_for_idle(rdev)) {
4772 dev_warn(rdev->dev, "Wait for MC idle timedout !\n");
4775 if (reset_mask & (RADEON_RESET_GFX | RADEON_RESET_COMPUTE | RADEON_RESET_CP))
4776 grbm_soft_reset = SOFT_RESET_CP | SOFT_RESET_GFX;
4778 if (reset_mask & RADEON_RESET_CP) {
4779 grbm_soft_reset |= SOFT_RESET_CP;
4781 srbm_soft_reset |= SOFT_RESET_GRBM;
4784 if (reset_mask & RADEON_RESET_DMA)
4785 srbm_soft_reset |= SOFT_RESET_SDMA;
4787 if (reset_mask & RADEON_RESET_DMA1)
4788 srbm_soft_reset |= SOFT_RESET_SDMA1;
4790 if (reset_mask & RADEON_RESET_DISPLAY)
4791 srbm_soft_reset |= SOFT_RESET_DC;
4793 if (reset_mask & RADEON_RESET_RLC)
4794 grbm_soft_reset |= SOFT_RESET_RLC;
4796 if (reset_mask & RADEON_RESET_SEM)
4797 srbm_soft_reset |= SOFT_RESET_SEM;
4799 if (reset_mask & RADEON_RESET_IH)
4800 srbm_soft_reset |= SOFT_RESET_IH;
4802 if (reset_mask & RADEON_RESET_GRBM)
4803 srbm_soft_reset |= SOFT_RESET_GRBM;
4805 if (reset_mask & RADEON_RESET_VMC)
4806 srbm_soft_reset |= SOFT_RESET_VMC;
4808 if (!(rdev->flags & RADEON_IS_IGP)) {
4809 if (reset_mask & RADEON_RESET_MC)
4810 srbm_soft_reset |= SOFT_RESET_MC;
4813 if (grbm_soft_reset) {
4814 tmp = RREG32(GRBM_SOFT_RESET);
4815 tmp |= grbm_soft_reset;
4816 dev_info(rdev->dev, "GRBM_SOFT_RESET=0x%08X\n", tmp);
4817 WREG32(GRBM_SOFT_RESET, tmp);
4818 tmp = RREG32(GRBM_SOFT_RESET);
4822 tmp &= ~grbm_soft_reset;
4823 WREG32(GRBM_SOFT_RESET, tmp);
4824 tmp = RREG32(GRBM_SOFT_RESET);
4827 if (srbm_soft_reset) {
4828 tmp = RREG32(SRBM_SOFT_RESET);
4829 tmp |= srbm_soft_reset;
4830 dev_info(rdev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp);
4831 WREG32(SRBM_SOFT_RESET, tmp);
4832 tmp = RREG32(SRBM_SOFT_RESET);
4836 tmp &= ~srbm_soft_reset;
4837 WREG32(SRBM_SOFT_RESET, tmp);
4838 tmp = RREG32(SRBM_SOFT_RESET);
4841 /* Wait a little for things to settle down */
4844 evergreen_mc_resume(rdev, &save);
4847 cik_print_gpu_status_regs(rdev);
4851 * cik_asic_reset - soft reset GPU
4853 * @rdev: radeon_device pointer
4855 * Look up which blocks are hung and attempt
4857 * Returns 0 for success.
4859 int cik_asic_reset(struct radeon_device *rdev)
4863 reset_mask = cik_gpu_check_soft_reset(rdev);
4866 r600_set_bios_scratch_engine_hung(rdev, true);
4868 cik_gpu_soft_reset(rdev, reset_mask);
4870 reset_mask = cik_gpu_check_soft_reset(rdev);
4873 r600_set_bios_scratch_engine_hung(rdev, false);
4879 * cik_gfx_is_lockup - check if the 3D engine is locked up
4881 * @rdev: radeon_device pointer
4882 * @ring: radeon_ring structure holding ring information
4884 * Check if the 3D engine is locked up (CIK).
4885 * Returns true if the engine is locked, false if not.
4887 bool cik_gfx_is_lockup(struct radeon_device *rdev, struct radeon_ring *ring)
4889 u32 reset_mask = cik_gpu_check_soft_reset(rdev);
4891 if (!(reset_mask & (RADEON_RESET_GFX |
4892 RADEON_RESET_COMPUTE |
4893 RADEON_RESET_CP))) {
4894 radeon_ring_lockup_update(ring);
4897 /* force CP activities */
4898 radeon_ring_force_activity(rdev, ring);
4899 return radeon_ring_test_lockup(rdev, ring);
4904 * cik_mc_program - program the GPU memory controller
4906 * @rdev: radeon_device pointer
4908 * Set the location of vram, gart, and AGP in the GPU's
4909 * physical address space (CIK).
4911 static void cik_mc_program(struct radeon_device *rdev)
4913 struct evergreen_mc_save save;
4917 /* Initialize HDP */
4918 for (i = 0, j = 0; i < 32; i++, j += 0x18) {
4919 WREG32((0x2c14 + j), 0x00000000);
4920 WREG32((0x2c18 + j), 0x00000000);
4921 WREG32((0x2c1c + j), 0x00000000);
4922 WREG32((0x2c20 + j), 0x00000000);
4923 WREG32((0x2c24 + j), 0x00000000);
4925 WREG32(HDP_REG_COHERENCY_FLUSH_CNTL, 0);
4927 evergreen_mc_stop(rdev, &save);
4928 if (radeon_mc_wait_for_idle(rdev)) {
4929 dev_warn(rdev->dev, "Wait for MC idle timedout !\n");
4931 /* Lockout access through VGA aperture*/
4932 WREG32(VGA_HDP_CONTROL, VGA_MEMORY_DISABLE);
4933 /* Update configuration */
4934 WREG32(MC_VM_SYSTEM_APERTURE_LOW_ADDR,
4935 rdev->mc.vram_start >> 12);
4936 WREG32(MC_VM_SYSTEM_APERTURE_HIGH_ADDR,
4937 rdev->mc.vram_end >> 12);
4938 WREG32(MC_VM_SYSTEM_APERTURE_DEFAULT_ADDR,
4939 rdev->vram_scratch.gpu_addr >> 12);
4940 tmp = ((rdev->mc.vram_end >> 24) & 0xFFFF) << 16;
4941 tmp |= ((rdev->mc.vram_start >> 24) & 0xFFFF);
4942 WREG32(MC_VM_FB_LOCATION, tmp);
4943 /* XXX double check these! */
4944 WREG32(HDP_NONSURFACE_BASE, (rdev->mc.vram_start >> 8));
4945 WREG32(HDP_NONSURFACE_INFO, (2 << 7) | (1 << 30));
4946 WREG32(HDP_NONSURFACE_SIZE, 0x3FFFFFFF);
4947 WREG32(MC_VM_AGP_BASE, 0);
4948 WREG32(MC_VM_AGP_TOP, 0x0FFFFFFF);
4949 WREG32(MC_VM_AGP_BOT, 0x0FFFFFFF);
4950 if (radeon_mc_wait_for_idle(rdev)) {
4951 dev_warn(rdev->dev, "Wait for MC idle timedout !\n");
4953 evergreen_mc_resume(rdev, &save);
4954 /* we need to own VRAM, so turn off the VGA renderer here
4955 * to stop it overwriting our objects */
4956 rv515_vga_render_disable(rdev);
4960 * cik_mc_init - initialize the memory controller driver params
4962 * @rdev: radeon_device pointer
4964 * Look up the amount of vram, vram width, and decide how to place
4965 * vram and gart within the GPU's physical address space (CIK).
4966 * Returns 0 for success.
4968 static int cik_mc_init(struct radeon_device *rdev)
4971 int chansize, numchan;
4973 /* Get VRAM informations */
4974 rdev->mc.vram_is_ddr = true;
4975 tmp = RREG32(MC_ARB_RAMCFG);
4976 if (tmp & CHANSIZE_MASK) {
4981 tmp = RREG32(MC_SHARED_CHMAP);
4982 switch ((tmp & NOOFCHAN_MASK) >> NOOFCHAN_SHIFT) {
5012 rdev->mc.vram_width = numchan * chansize;
5013 /* Could aper size report 0 ? */
5014 rdev->mc.aper_base = pci_resource_start(rdev->pdev, 0);
5015 rdev->mc.aper_size = pci_resource_len(rdev->pdev, 0);
5016 /* size in MB on si */
5017 rdev->mc.mc_vram_size = RREG32(CONFIG_MEMSIZE) * 1024ULL * 1024ULL;
5018 rdev->mc.real_vram_size = RREG32(CONFIG_MEMSIZE) * 1024ULL * 1024ULL;
5019 rdev->mc.visible_vram_size = rdev->mc.aper_size;
5020 si_vram_gtt_location(rdev, &rdev->mc);
5021 radeon_update_bandwidth_info(rdev);
5028 * VMID 0 is the physical GPU addresses as used by the kernel.
5029 * VMIDs 1-15 are used for userspace clients and are handled
5030 * by the radeon vm/hsa code.
5033 * cik_pcie_gart_tlb_flush - gart tlb flush callback
5035 * @rdev: radeon_device pointer
5037 * Flush the TLB for the VMID 0 page table (CIK).
5039 void cik_pcie_gart_tlb_flush(struct radeon_device *rdev)
5041 /* flush hdp cache */
5042 WREG32(HDP_MEM_COHERENCY_FLUSH_CNTL, 0);
5044 /* bits 0-15 are the VM contexts0-15 */
5045 WREG32(VM_INVALIDATE_REQUEST, 0x1);
5049 * cik_pcie_gart_enable - gart enable
5051 * @rdev: radeon_device pointer
5053 * This sets up the TLBs, programs the page tables for VMID0,
5054 * sets up the hw for VMIDs 1-15 which are allocated on
5055 * demand, and sets up the global locations for the LDS, GDS,
5056 * and GPUVM for FSA64 clients (CIK).
5057 * Returns 0 for success, errors for failure.
5059 static int cik_pcie_gart_enable(struct radeon_device *rdev)
5063 if (rdev->gart.robj == NULL) {
5064 dev_err(rdev->dev, "No VRAM object for PCIE GART.\n");
5067 r = radeon_gart_table_vram_pin(rdev);
5070 radeon_gart_restore(rdev);
5071 /* Setup TLB control */
5072 WREG32(MC_VM_MX_L1_TLB_CNTL,
5075 SYSTEM_ACCESS_MODE_NOT_IN_SYS |
5076 ENABLE_ADVANCED_DRIVER_MODEL |
5077 SYSTEM_APERTURE_UNMAPPED_ACCESS_PASS_THRU);
5078 /* Setup L2 cache */
5079 WREG32(VM_L2_CNTL, ENABLE_L2_CACHE |
5080 ENABLE_L2_FRAGMENT_PROCESSING |
5081 ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE |
5082 ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE |
5083 EFFECTIVE_L2_QUEUE_SIZE(7) |
5084 CONTEXT1_IDENTITY_ACCESS_MODE(1));
5085 WREG32(VM_L2_CNTL2, INVALIDATE_ALL_L1_TLBS | INVALIDATE_L2_CACHE);
5086 WREG32(VM_L2_CNTL3, L2_CACHE_BIGK_ASSOCIATIVITY |
5087 L2_CACHE_BIGK_FRAGMENT_SIZE(6));
5088 /* setup context0 */
5089 WREG32(VM_CONTEXT0_PAGE_TABLE_START_ADDR, rdev->mc.gtt_start >> 12);
5090 WREG32(VM_CONTEXT0_PAGE_TABLE_END_ADDR, rdev->mc.gtt_end >> 12);
5091 WREG32(VM_CONTEXT0_PAGE_TABLE_BASE_ADDR, rdev->gart.table_addr >> 12);
5092 WREG32(VM_CONTEXT0_PROTECTION_FAULT_DEFAULT_ADDR,
5093 (u32)(rdev->dummy_page.addr >> 12));
5094 WREG32(VM_CONTEXT0_CNTL2, 0);
5095 WREG32(VM_CONTEXT0_CNTL, (ENABLE_CONTEXT | PAGE_TABLE_DEPTH(0) |
5096 RANGE_PROTECTION_FAULT_ENABLE_DEFAULT));
5102 /* empty context1-15 */
5103 /* FIXME start with 4G, once using 2 level pt switch to full
5106 /* set vm size, must be a multiple of 4 */
5107 WREG32(VM_CONTEXT1_PAGE_TABLE_START_ADDR, 0);
5108 WREG32(VM_CONTEXT1_PAGE_TABLE_END_ADDR, rdev->vm_manager.max_pfn);
5109 for (i = 1; i < 16; i++) {
5111 WREG32(VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + (i << 2),
5112 rdev->gart.table_addr >> 12);
5114 WREG32(VM_CONTEXT8_PAGE_TABLE_BASE_ADDR + ((i - 8) << 2),
5115 rdev->gart.table_addr >> 12);
5118 /* enable context1-15 */
5119 WREG32(VM_CONTEXT1_PROTECTION_FAULT_DEFAULT_ADDR,
5120 (u32)(rdev->dummy_page.addr >> 12));
5121 WREG32(VM_CONTEXT1_CNTL2, 4);
5122 WREG32(VM_CONTEXT1_CNTL, ENABLE_CONTEXT | PAGE_TABLE_DEPTH(1) |
5123 RANGE_PROTECTION_FAULT_ENABLE_INTERRUPT |
5124 RANGE_PROTECTION_FAULT_ENABLE_DEFAULT |
5125 DUMMY_PAGE_PROTECTION_FAULT_ENABLE_INTERRUPT |
5126 DUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT |
5127 PDE0_PROTECTION_FAULT_ENABLE_INTERRUPT |
5128 PDE0_PROTECTION_FAULT_ENABLE_DEFAULT |
5129 VALID_PROTECTION_FAULT_ENABLE_INTERRUPT |
5130 VALID_PROTECTION_FAULT_ENABLE_DEFAULT |
5131 READ_PROTECTION_FAULT_ENABLE_INTERRUPT |
5132 READ_PROTECTION_FAULT_ENABLE_DEFAULT |
5133 WRITE_PROTECTION_FAULT_ENABLE_INTERRUPT |
5134 WRITE_PROTECTION_FAULT_ENABLE_DEFAULT);
5136 /* TC cache setup ??? */
5137 WREG32(TC_CFG_L1_LOAD_POLICY0, 0);
5138 WREG32(TC_CFG_L1_LOAD_POLICY1, 0);
5139 WREG32(TC_CFG_L1_STORE_POLICY, 0);
5141 WREG32(TC_CFG_L2_LOAD_POLICY0, 0);
5142 WREG32(TC_CFG_L2_LOAD_POLICY1, 0);
5143 WREG32(TC_CFG_L2_STORE_POLICY0, 0);
5144 WREG32(TC_CFG_L2_STORE_POLICY1, 0);
5145 WREG32(TC_CFG_L2_ATOMIC_POLICY, 0);
5147 WREG32(TC_CFG_L1_VOLATILE, 0);
5148 WREG32(TC_CFG_L2_VOLATILE, 0);
5150 if (rdev->family == CHIP_KAVERI) {
5151 u32 tmp = RREG32(CHUB_CONTROL);
5153 WREG32(CHUB_CONTROL, tmp);
5156 /* XXX SH_MEM regs */
5157 /* where to put LDS, scratch, GPUVM in FSA64 space */
5158 mutex_lock(&rdev->srbm_mutex);
5159 for (i = 0; i < 16; i++) {
5160 cik_srbm_select(rdev, 0, 0, 0, i);
5161 /* CP and shaders */
5162 WREG32(SH_MEM_CONFIG, 0);
5163 WREG32(SH_MEM_APE1_BASE, 1);
5164 WREG32(SH_MEM_APE1_LIMIT, 0);
5165 WREG32(SH_MEM_BASES, 0);
5167 WREG32(SDMA0_GFX_VIRTUAL_ADDR + SDMA0_REGISTER_OFFSET, 0);
5168 WREG32(SDMA0_GFX_APE1_CNTL + SDMA0_REGISTER_OFFSET, 0);
5169 WREG32(SDMA0_GFX_VIRTUAL_ADDR + SDMA1_REGISTER_OFFSET, 0);
5170 WREG32(SDMA0_GFX_APE1_CNTL + SDMA1_REGISTER_OFFSET, 0);
5171 /* XXX SDMA RLC - todo */
5173 cik_srbm_select(rdev, 0, 0, 0, 0);
5174 mutex_unlock(&rdev->srbm_mutex);
5176 cik_pcie_gart_tlb_flush(rdev);
5177 DRM_INFO("PCIE GART of %uM enabled (table at 0x%016llX).\n",
5178 (unsigned)(rdev->mc.gtt_size >> 20),
5179 (unsigned long long)rdev->gart.table_addr);
5180 rdev->gart.ready = true;
5185 * cik_pcie_gart_disable - gart disable
5187 * @rdev: radeon_device pointer
5189 * This disables all VM page table (CIK).
5191 static void cik_pcie_gart_disable(struct radeon_device *rdev)
5193 /* Disable all tables */
5194 WREG32(VM_CONTEXT0_CNTL, 0);
5195 WREG32(VM_CONTEXT1_CNTL, 0);
5196 /* Setup TLB control */
5197 WREG32(MC_VM_MX_L1_TLB_CNTL, SYSTEM_ACCESS_MODE_NOT_IN_SYS |
5198 SYSTEM_APERTURE_UNMAPPED_ACCESS_PASS_THRU);
5199 /* Setup L2 cache */
5201 ENABLE_L2_FRAGMENT_PROCESSING |
5202 ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE |
5203 ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE |
5204 EFFECTIVE_L2_QUEUE_SIZE(7) |
5205 CONTEXT1_IDENTITY_ACCESS_MODE(1));
5206 WREG32(VM_L2_CNTL2, 0);
5207 WREG32(VM_L2_CNTL3, L2_CACHE_BIGK_ASSOCIATIVITY |
5208 L2_CACHE_BIGK_FRAGMENT_SIZE(6));
5209 radeon_gart_table_vram_unpin(rdev);
5213 * cik_pcie_gart_fini - vm fini callback
5215 * @rdev: radeon_device pointer
5217 * Tears down the driver GART/VM setup (CIK).
5219 static void cik_pcie_gart_fini(struct radeon_device *rdev)
5221 cik_pcie_gart_disable(rdev);
5222 radeon_gart_table_vram_free(rdev);
5223 radeon_gart_fini(rdev);
5228 * cik_ib_parse - vm ib_parse callback
5230 * @rdev: radeon_device pointer
5231 * @ib: indirect buffer pointer
5233 * CIK uses hw IB checking so this is a nop (CIK).
5235 int cik_ib_parse(struct radeon_device *rdev, struct radeon_ib *ib)
5242 * VMID 0 is the physical GPU addresses as used by the kernel.
5243 * VMIDs 1-15 are used for userspace clients and are handled
5244 * by the radeon vm/hsa code.
5247 * cik_vm_init - cik vm init callback
5249 * @rdev: radeon_device pointer
5251 * Inits cik specific vm parameters (number of VMs, base of vram for
5252 * VMIDs 1-15) (CIK).
5253 * Returns 0 for success.
5255 int cik_vm_init(struct radeon_device *rdev)
5258 rdev->vm_manager.nvm = 16;
5259 /* base offset of vram pages */
5260 if (rdev->flags & RADEON_IS_IGP) {
5261 u64 tmp = RREG32(MC_VM_FB_OFFSET);
5263 rdev->vm_manager.vram_base_offset = tmp;
5265 rdev->vm_manager.vram_base_offset = 0;
5271 * cik_vm_fini - cik vm fini callback
5273 * @rdev: radeon_device pointer
5275 * Tear down any asic specific VM setup (CIK).
5277 void cik_vm_fini(struct radeon_device *rdev)
5282 * cik_vm_decode_fault - print human readable fault info
5284 * @rdev: radeon_device pointer
5285 * @status: VM_CONTEXT1_PROTECTION_FAULT_STATUS register value
5286 * @addr: VM_CONTEXT1_PROTECTION_FAULT_ADDR register value
5288 * Print human readable fault information (CIK).
5290 static void cik_vm_decode_fault(struct radeon_device *rdev,
5291 u32 status, u32 addr, u32 mc_client)
5294 u32 vmid = (status & FAULT_VMID_MASK) >> FAULT_VMID_SHIFT;
5295 u32 protections = (status & PROTECTIONS_MASK) >> PROTECTIONS_SHIFT;
5296 char block[5] = { mc_client >> 24, (mc_client >> 16) & 0xff,
5297 (mc_client >> 8) & 0xff, mc_client & 0xff, 0 };
5299 if (rdev->family == CHIP_HAWAII)
5300 mc_id = (status & HAWAII_MEMORY_CLIENT_ID_MASK) >> MEMORY_CLIENT_ID_SHIFT;
5302 mc_id = (status & MEMORY_CLIENT_ID_MASK) >> MEMORY_CLIENT_ID_SHIFT;
5304 printk("VM fault (0x%02x, vmid %d) at page %u, %s from '%s' (0x%08x) (%d)\n",
5305 protections, vmid, addr,
5306 (status & MEMORY_CLIENT_RW_MASK) ? "write" : "read",
5307 block, mc_client, mc_id);
5311 * cik_vm_flush - cik vm flush using the CP
5313 * @rdev: radeon_device pointer
5315 * Update the page table base and flush the VM TLB
5316 * using the CP (CIK).
5318 void cik_vm_flush(struct radeon_device *rdev, int ridx, struct radeon_vm *vm)
5320 struct radeon_ring *ring = &rdev->ring[ridx];
5325 radeon_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5326 radeon_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
5327 WRITE_DATA_DST_SEL(0)));
5329 radeon_ring_write(ring,
5330 (VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + (vm->id << 2)) >> 2);
5332 radeon_ring_write(ring,
5333 (VM_CONTEXT8_PAGE_TABLE_BASE_ADDR + ((vm->id - 8) << 2)) >> 2);
5335 radeon_ring_write(ring, 0);
5336 radeon_ring_write(ring, vm->pd_gpu_addr >> 12);
5338 /* update SH_MEM_* regs */
5339 radeon_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5340 radeon_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
5341 WRITE_DATA_DST_SEL(0)));
5342 radeon_ring_write(ring, SRBM_GFX_CNTL >> 2);
5343 radeon_ring_write(ring, 0);
5344 radeon_ring_write(ring, VMID(vm->id));
5346 radeon_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 6));
5347 radeon_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
5348 WRITE_DATA_DST_SEL(0)));
5349 radeon_ring_write(ring, SH_MEM_BASES >> 2);
5350 radeon_ring_write(ring, 0);
5352 radeon_ring_write(ring, 0); /* SH_MEM_BASES */
5353 radeon_ring_write(ring, 0); /* SH_MEM_CONFIG */
5354 radeon_ring_write(ring, 1); /* SH_MEM_APE1_BASE */
5355 radeon_ring_write(ring, 0); /* SH_MEM_APE1_LIMIT */
5357 radeon_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5358 radeon_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
5359 WRITE_DATA_DST_SEL(0)));
5360 radeon_ring_write(ring, SRBM_GFX_CNTL >> 2);
5361 radeon_ring_write(ring, 0);
5362 radeon_ring_write(ring, VMID(0));
5365 /* We should be using the WAIT_REG_MEM packet here like in
5366 * cik_fence_ring_emit(), but it causes the CP to hang in this
5369 radeon_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5370 radeon_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
5371 WRITE_DATA_DST_SEL(0)));
5372 radeon_ring_write(ring, HDP_MEM_COHERENCY_FLUSH_CNTL >> 2);
5373 radeon_ring_write(ring, 0);
5374 radeon_ring_write(ring, 0);
5376 /* bits 0-15 are the VM contexts0-15 */
5377 radeon_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5378 radeon_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
5379 WRITE_DATA_DST_SEL(0)));
5380 radeon_ring_write(ring, VM_INVALIDATE_REQUEST >> 2);
5381 radeon_ring_write(ring, 0);
5382 radeon_ring_write(ring, 1 << vm->id);
5384 /* compute doesn't have PFP */
5385 if (ridx == RADEON_RING_TYPE_GFX_INDEX) {
5386 /* sync PFP to ME, otherwise we might get invalid PFP reads */
5387 radeon_ring_write(ring, PACKET3(PACKET3_PFP_SYNC_ME, 0));
5388 radeon_ring_write(ring, 0x0);
5394 * The RLC is a multi-purpose microengine that handles a
5395 * variety of functions, the most important of which is
5396 * the interrupt controller.
5398 static void cik_enable_gui_idle_interrupt(struct radeon_device *rdev,
5401 u32 tmp = RREG32(CP_INT_CNTL_RING0);
5404 tmp |= (CNTX_BUSY_INT_ENABLE | CNTX_EMPTY_INT_ENABLE);
5406 tmp &= ~(CNTX_BUSY_INT_ENABLE | CNTX_EMPTY_INT_ENABLE);
5407 WREG32(CP_INT_CNTL_RING0, tmp);
5410 static void cik_enable_lbpw(struct radeon_device *rdev, bool enable)
5414 tmp = RREG32(RLC_LB_CNTL);
5416 tmp |= LOAD_BALANCE_ENABLE;
5418 tmp &= ~LOAD_BALANCE_ENABLE;
5419 WREG32(RLC_LB_CNTL, tmp);
5422 static void cik_wait_for_rlc_serdes(struct radeon_device *rdev)
5427 for (i = 0; i < rdev->config.cik.max_shader_engines; i++) {
5428 for (j = 0; j < rdev->config.cik.max_sh_per_se; j++) {
5429 cik_select_se_sh(rdev, i, j);
5430 for (k = 0; k < rdev->usec_timeout; k++) {
5431 if (RREG32(RLC_SERDES_CU_MASTER_BUSY) == 0)
5437 cik_select_se_sh(rdev, 0xffffffff, 0xffffffff);
5439 mask = SE_MASTER_BUSY_MASK | GC_MASTER_BUSY | TC0_MASTER_BUSY | TC1_MASTER_BUSY;
5440 for (k = 0; k < rdev->usec_timeout; k++) {
5441 if ((RREG32(RLC_SERDES_NONCU_MASTER_BUSY) & mask) == 0)
5447 static void cik_update_rlc(struct radeon_device *rdev, u32 rlc)
5451 tmp = RREG32(RLC_CNTL);
5453 WREG32(RLC_CNTL, rlc);
5456 static u32 cik_halt_rlc(struct radeon_device *rdev)
5460 orig = data = RREG32(RLC_CNTL);
5462 if (data & RLC_ENABLE) {
5465 data &= ~RLC_ENABLE;
5466 WREG32(RLC_CNTL, data);
5468 for (i = 0; i < rdev->usec_timeout; i++) {
5469 if ((RREG32(RLC_GPM_STAT) & RLC_GPM_BUSY) == 0)
5474 cik_wait_for_rlc_serdes(rdev);
5480 void cik_enter_rlc_safe_mode(struct radeon_device *rdev)
5484 tmp = REQ | MESSAGE(MSG_ENTER_RLC_SAFE_MODE);
5485 WREG32(RLC_GPR_REG2, tmp);
5487 mask = GFX_POWER_STATUS | GFX_CLOCK_STATUS;
5488 for (i = 0; i < rdev->usec_timeout; i++) {
5489 if ((RREG32(RLC_GPM_STAT) & mask) == mask)
5494 for (i = 0; i < rdev->usec_timeout; i++) {
5495 if ((RREG32(RLC_GPR_REG2) & REQ) == 0)
5501 void cik_exit_rlc_safe_mode(struct radeon_device *rdev)
5505 tmp = REQ | MESSAGE(MSG_EXIT_RLC_SAFE_MODE);
5506 WREG32(RLC_GPR_REG2, tmp);
5510 * cik_rlc_stop - stop the RLC ME
5512 * @rdev: radeon_device pointer
5514 * Halt the RLC ME (MicroEngine) (CIK).
5516 static void cik_rlc_stop(struct radeon_device *rdev)
5518 WREG32(RLC_CNTL, 0);
5520 cik_enable_gui_idle_interrupt(rdev, false);
5522 cik_wait_for_rlc_serdes(rdev);
5526 * cik_rlc_start - start the RLC ME
5528 * @rdev: radeon_device pointer
5530 * Unhalt the RLC ME (MicroEngine) (CIK).
5532 static void cik_rlc_start(struct radeon_device *rdev)
5534 WREG32(RLC_CNTL, RLC_ENABLE);
5536 cik_enable_gui_idle_interrupt(rdev, true);
5542 * cik_rlc_resume - setup the RLC hw
5544 * @rdev: radeon_device pointer
5546 * Initialize the RLC registers, load the ucode,
5547 * and start the RLC (CIK).
5548 * Returns 0 for success, -EINVAL if the ucode is not available.
5550 static int cik_rlc_resume(struct radeon_device *rdev)
5553 const __be32 *fw_data;
5558 switch (rdev->family) {
5562 size = BONAIRE_RLC_UCODE_SIZE;
5565 size = KV_RLC_UCODE_SIZE;
5568 size = KB_RLC_UCODE_SIZE;
5575 tmp = RREG32(RLC_CGCG_CGLS_CTRL) & 0xfffffffc;
5576 WREG32(RLC_CGCG_CGLS_CTRL, tmp);
5584 WREG32(RLC_LB_CNTR_INIT, 0);
5585 WREG32(RLC_LB_CNTR_MAX, 0x00008000);
5587 cik_select_se_sh(rdev, 0xffffffff, 0xffffffff);
5588 WREG32(RLC_LB_INIT_CU_MASK, 0xffffffff);
5589 WREG32(RLC_LB_PARAMS, 0x00600408);
5590 WREG32(RLC_LB_CNTL, 0x80000004);
5592 WREG32(RLC_MC_CNTL, 0);
5593 WREG32(RLC_UCODE_CNTL, 0);
5595 fw_data = (const __be32 *)rdev->rlc_fw->data;
5596 WREG32(RLC_GPM_UCODE_ADDR, 0);
5597 for (i = 0; i < size; i++)
5598 WREG32(RLC_GPM_UCODE_DATA, be32_to_cpup(fw_data++));
5599 WREG32(RLC_GPM_UCODE_ADDR, 0);
5601 /* XXX - find out what chips support lbpw */
5602 cik_enable_lbpw(rdev, false);
5604 if (rdev->family == CHIP_BONAIRE)
5605 WREG32(RLC_DRIVER_DMA_STATUS, 0);
5607 cik_rlc_start(rdev);
5612 static void cik_enable_cgcg(struct radeon_device *rdev, bool enable)
5614 u32 data, orig, tmp, tmp2;
5616 orig = data = RREG32(RLC_CGCG_CGLS_CTRL);
5618 if (enable && (rdev->cg_flags & RADEON_CG_SUPPORT_GFX_CGCG)) {
5619 cik_enable_gui_idle_interrupt(rdev, true);
5621 tmp = cik_halt_rlc(rdev);
5623 cik_select_se_sh(rdev, 0xffffffff, 0xffffffff);
5624 WREG32(RLC_SERDES_WR_CU_MASTER_MASK, 0xffffffff);
5625 WREG32(RLC_SERDES_WR_NONCU_MASTER_MASK, 0xffffffff);
5626 tmp2 = BPM_ADDR_MASK | CGCG_OVERRIDE_0 | CGLS_ENABLE;
5627 WREG32(RLC_SERDES_WR_CTRL, tmp2);
5629 cik_update_rlc(rdev, tmp);
5631 data |= CGCG_EN | CGLS_EN;
5633 cik_enable_gui_idle_interrupt(rdev, false);
5635 RREG32(CB_CGTT_SCLK_CTRL);
5636 RREG32(CB_CGTT_SCLK_CTRL);
5637 RREG32(CB_CGTT_SCLK_CTRL);
5638 RREG32(CB_CGTT_SCLK_CTRL);
5640 data &= ~(CGCG_EN | CGLS_EN);
5644 WREG32(RLC_CGCG_CGLS_CTRL, data);
5648 static void cik_enable_mgcg(struct radeon_device *rdev, bool enable)
5650 u32 data, orig, tmp = 0;
5652 if (enable && (rdev->cg_flags & RADEON_CG_SUPPORT_GFX_MGCG)) {
5653 if (rdev->cg_flags & RADEON_CG_SUPPORT_GFX_MGLS) {
5654 if (rdev->cg_flags & RADEON_CG_SUPPORT_GFX_CP_LS) {
5655 orig = data = RREG32(CP_MEM_SLP_CNTL);
5656 data |= CP_MEM_LS_EN;
5658 WREG32(CP_MEM_SLP_CNTL, data);
5662 orig = data = RREG32(RLC_CGTT_MGCG_OVERRIDE);
5665 WREG32(RLC_CGTT_MGCG_OVERRIDE, data);
5667 tmp = cik_halt_rlc(rdev);
5669 cik_select_se_sh(rdev, 0xffffffff, 0xffffffff);
5670 WREG32(RLC_SERDES_WR_CU_MASTER_MASK, 0xffffffff);
5671 WREG32(RLC_SERDES_WR_NONCU_MASTER_MASK, 0xffffffff);
5672 data = BPM_ADDR_MASK | MGCG_OVERRIDE_0;
5673 WREG32(RLC_SERDES_WR_CTRL, data);
5675 cik_update_rlc(rdev, tmp);
5677 if (rdev->cg_flags & RADEON_CG_SUPPORT_GFX_CGTS) {
5678 orig = data = RREG32(CGTS_SM_CTRL_REG);
5679 data &= ~SM_MODE_MASK;
5680 data |= SM_MODE(0x2);
5681 data |= SM_MODE_ENABLE;
5682 data &= ~CGTS_OVERRIDE;
5683 if ((rdev->cg_flags & RADEON_CG_SUPPORT_GFX_MGLS) &&
5684 (rdev->cg_flags & RADEON_CG_SUPPORT_GFX_CGTS_LS))
5685 data &= ~CGTS_LS_OVERRIDE;
5686 data &= ~ON_MONITOR_ADD_MASK;
5687 data |= ON_MONITOR_ADD_EN;
5688 data |= ON_MONITOR_ADD(0x96);
5690 WREG32(CGTS_SM_CTRL_REG, data);
5693 orig = data = RREG32(RLC_CGTT_MGCG_OVERRIDE);
5696 WREG32(RLC_CGTT_MGCG_OVERRIDE, data);
5698 data = RREG32(RLC_MEM_SLP_CNTL);
5699 if (data & RLC_MEM_LS_EN) {
5700 data &= ~RLC_MEM_LS_EN;
5701 WREG32(RLC_MEM_SLP_CNTL, data);
5704 data = RREG32(CP_MEM_SLP_CNTL);
5705 if (data & CP_MEM_LS_EN) {
5706 data &= ~CP_MEM_LS_EN;
5707 WREG32(CP_MEM_SLP_CNTL, data);
5710 orig = data = RREG32(CGTS_SM_CTRL_REG);
5711 data |= CGTS_OVERRIDE | CGTS_LS_OVERRIDE;
5713 WREG32(CGTS_SM_CTRL_REG, data);
5715 tmp = cik_halt_rlc(rdev);
5717 cik_select_se_sh(rdev, 0xffffffff, 0xffffffff);
5718 WREG32(RLC_SERDES_WR_CU_MASTER_MASK, 0xffffffff);
5719 WREG32(RLC_SERDES_WR_NONCU_MASTER_MASK, 0xffffffff);
5720 data = BPM_ADDR_MASK | MGCG_OVERRIDE_1;
5721 WREG32(RLC_SERDES_WR_CTRL, data);
5723 cik_update_rlc(rdev, tmp);
5727 static const u32 mc_cg_registers[] =
5740 static void cik_enable_mc_ls(struct radeon_device *rdev,
5746 for (i = 0; i < ARRAY_SIZE(mc_cg_registers); i++) {
5747 orig = data = RREG32(mc_cg_registers[i]);
5748 if (enable && (rdev->cg_flags & RADEON_CG_SUPPORT_MC_LS))
5749 data |= MC_LS_ENABLE;
5751 data &= ~MC_LS_ENABLE;
5753 WREG32(mc_cg_registers[i], data);
5757 static void cik_enable_mc_mgcg(struct radeon_device *rdev,
5763 for (i = 0; i < ARRAY_SIZE(mc_cg_registers); i++) {
5764 orig = data = RREG32(mc_cg_registers[i]);
5765 if (enable && (rdev->cg_flags & RADEON_CG_SUPPORT_MC_MGCG))
5766 data |= MC_CG_ENABLE;
5768 data &= ~MC_CG_ENABLE;
5770 WREG32(mc_cg_registers[i], data);
5774 static void cik_enable_sdma_mgcg(struct radeon_device *rdev,
5779 if (enable && (rdev->cg_flags & RADEON_CG_SUPPORT_SDMA_MGCG)) {
5780 WREG32(SDMA0_CLK_CTRL + SDMA0_REGISTER_OFFSET, 0x00000100);
5781 WREG32(SDMA0_CLK_CTRL + SDMA1_REGISTER_OFFSET, 0x00000100);
5783 orig = data = RREG32(SDMA0_CLK_CTRL + SDMA0_REGISTER_OFFSET);
5786 WREG32(SDMA0_CLK_CTRL + SDMA0_REGISTER_OFFSET, data);
5788 orig = data = RREG32(SDMA0_CLK_CTRL + SDMA1_REGISTER_OFFSET);
5791 WREG32(SDMA0_CLK_CTRL + SDMA1_REGISTER_OFFSET, data);
5795 static void cik_enable_sdma_mgls(struct radeon_device *rdev,
5800 if (enable && (rdev->cg_flags & RADEON_CG_SUPPORT_SDMA_LS)) {
5801 orig = data = RREG32(SDMA0_POWER_CNTL + SDMA0_REGISTER_OFFSET);
5804 WREG32(SDMA0_POWER_CNTL + SDMA0_REGISTER_OFFSET, data);
5806 orig = data = RREG32(SDMA0_POWER_CNTL + SDMA1_REGISTER_OFFSET);
5809 WREG32(SDMA0_POWER_CNTL + SDMA1_REGISTER_OFFSET, data);
5811 orig = data = RREG32(SDMA0_POWER_CNTL + SDMA0_REGISTER_OFFSET);
5814 WREG32(SDMA0_POWER_CNTL + SDMA0_REGISTER_OFFSET, data);
5816 orig = data = RREG32(SDMA0_POWER_CNTL + SDMA1_REGISTER_OFFSET);
5819 WREG32(SDMA0_POWER_CNTL + SDMA1_REGISTER_OFFSET, data);
5823 static void cik_enable_uvd_mgcg(struct radeon_device *rdev,
5828 if (enable && (rdev->cg_flags & RADEON_CG_SUPPORT_UVD_MGCG)) {
5829 data = RREG32_UVD_CTX(UVD_CGC_MEM_CTRL);
5831 WREG32_UVD_CTX(UVD_CGC_MEM_CTRL, data);
5833 orig = data = RREG32(UVD_CGC_CTRL);
5836 WREG32(UVD_CGC_CTRL, data);
5838 data = RREG32_UVD_CTX(UVD_CGC_MEM_CTRL);
5840 WREG32_UVD_CTX(UVD_CGC_MEM_CTRL, data);
5842 orig = data = RREG32(UVD_CGC_CTRL);
5845 WREG32(UVD_CGC_CTRL, data);
5849 static void cik_enable_bif_mgls(struct radeon_device *rdev,
5854 orig = data = RREG32_PCIE_PORT(PCIE_CNTL2);
5856 if (enable && (rdev->cg_flags & RADEON_CG_SUPPORT_BIF_LS))
5857 data |= SLV_MEM_LS_EN | MST_MEM_LS_EN |
5858 REPLAY_MEM_LS_EN | SLV_MEM_AGGRESSIVE_LS_EN;
5860 data &= ~(SLV_MEM_LS_EN | MST_MEM_LS_EN |
5861 REPLAY_MEM_LS_EN | SLV_MEM_AGGRESSIVE_LS_EN);
5864 WREG32_PCIE_PORT(PCIE_CNTL2, data);
5867 static void cik_enable_hdp_mgcg(struct radeon_device *rdev,
5872 orig = data = RREG32(HDP_HOST_PATH_CNTL);
5874 if (enable && (rdev->cg_flags & RADEON_CG_SUPPORT_HDP_MGCG))
5875 data &= ~CLOCK_GATING_DIS;
5877 data |= CLOCK_GATING_DIS;
5880 WREG32(HDP_HOST_PATH_CNTL, data);
5883 static void cik_enable_hdp_ls(struct radeon_device *rdev,
5888 orig = data = RREG32(HDP_MEM_POWER_LS);
5890 if (enable && (rdev->cg_flags & RADEON_CG_SUPPORT_HDP_LS))
5891 data |= HDP_LS_ENABLE;
5893 data &= ~HDP_LS_ENABLE;
5896 WREG32(HDP_MEM_POWER_LS, data);
5899 void cik_update_cg(struct radeon_device *rdev,
5900 u32 block, bool enable)
5903 if (block & RADEON_CG_BLOCK_GFX) {
5904 cik_enable_gui_idle_interrupt(rdev, false);
5905 /* order matters! */
5907 cik_enable_mgcg(rdev, true);
5908 cik_enable_cgcg(rdev, true);
5910 cik_enable_cgcg(rdev, false);
5911 cik_enable_mgcg(rdev, false);
5913 cik_enable_gui_idle_interrupt(rdev, true);
5916 if (block & RADEON_CG_BLOCK_MC) {
5917 if (!(rdev->flags & RADEON_IS_IGP)) {
5918 cik_enable_mc_mgcg(rdev, enable);
5919 cik_enable_mc_ls(rdev, enable);
5923 if (block & RADEON_CG_BLOCK_SDMA) {
5924 cik_enable_sdma_mgcg(rdev, enable);
5925 cik_enable_sdma_mgls(rdev, enable);
5928 if (block & RADEON_CG_BLOCK_BIF) {
5929 cik_enable_bif_mgls(rdev, enable);
5932 if (block & RADEON_CG_BLOCK_UVD) {
5934 cik_enable_uvd_mgcg(rdev, enable);
5937 if (block & RADEON_CG_BLOCK_HDP) {
5938 cik_enable_hdp_mgcg(rdev, enable);
5939 cik_enable_hdp_ls(rdev, enable);
5943 static void cik_init_cg(struct radeon_device *rdev)
5946 cik_update_cg(rdev, RADEON_CG_BLOCK_GFX, true);
5949 si_init_uvd_internal_cg(rdev);
5951 cik_update_cg(rdev, (RADEON_CG_BLOCK_MC |
5952 RADEON_CG_BLOCK_SDMA |
5953 RADEON_CG_BLOCK_BIF |
5954 RADEON_CG_BLOCK_UVD |
5955 RADEON_CG_BLOCK_HDP), true);
5958 static void cik_fini_cg(struct radeon_device *rdev)
5960 cik_update_cg(rdev, (RADEON_CG_BLOCK_MC |
5961 RADEON_CG_BLOCK_SDMA |
5962 RADEON_CG_BLOCK_BIF |
5963 RADEON_CG_BLOCK_UVD |
5964 RADEON_CG_BLOCK_HDP), false);
5966 cik_update_cg(rdev, RADEON_CG_BLOCK_GFX, false);
5969 static void cik_enable_sck_slowdown_on_pu(struct radeon_device *rdev,
5974 orig = data = RREG32(RLC_PG_CNTL);
5975 if (enable && (rdev->pg_flags & RADEON_PG_SUPPORT_RLC_SMU_HS))
5976 data |= SMU_CLK_SLOWDOWN_ON_PU_ENABLE;
5978 data &= ~SMU_CLK_SLOWDOWN_ON_PU_ENABLE;
5980 WREG32(RLC_PG_CNTL, data);
5983 static void cik_enable_sck_slowdown_on_pd(struct radeon_device *rdev,
5988 orig = data = RREG32(RLC_PG_CNTL);
5989 if (enable && (rdev->pg_flags & RADEON_PG_SUPPORT_RLC_SMU_HS))
5990 data |= SMU_CLK_SLOWDOWN_ON_PD_ENABLE;
5992 data &= ~SMU_CLK_SLOWDOWN_ON_PD_ENABLE;
5994 WREG32(RLC_PG_CNTL, data);
5997 static void cik_enable_cp_pg(struct radeon_device *rdev, bool enable)
6001 orig = data = RREG32(RLC_PG_CNTL);
6002 if (enable && (rdev->pg_flags & RADEON_PG_SUPPORT_CP))
6003 data &= ~DISABLE_CP_PG;
6005 data |= DISABLE_CP_PG;
6007 WREG32(RLC_PG_CNTL, data);
6010 static void cik_enable_gds_pg(struct radeon_device *rdev, bool enable)
6014 orig = data = RREG32(RLC_PG_CNTL);
6015 if (enable && (rdev->pg_flags & RADEON_PG_SUPPORT_GDS))
6016 data &= ~DISABLE_GDS_PG;
6018 data |= DISABLE_GDS_PG;
6020 WREG32(RLC_PG_CNTL, data);
6023 #define CP_ME_TABLE_SIZE 96
6024 #define CP_ME_TABLE_OFFSET 2048
6025 #define CP_MEC_TABLE_OFFSET 4096
6027 void cik_init_cp_pg_table(struct radeon_device *rdev)
6029 const __be32 *fw_data;
6030 volatile u32 *dst_ptr;
6031 int me, i, max_me = 4;
6035 if (rdev->family == CHIP_KAVERI)
6038 if (rdev->rlc.cp_table_ptr == NULL)
6041 /* write the cp table buffer */
6042 dst_ptr = rdev->rlc.cp_table_ptr;
6043 for (me = 0; me < max_me; me++) {
6045 fw_data = (const __be32 *)rdev->ce_fw->data;
6046 table_offset = CP_ME_TABLE_OFFSET;
6047 } else if (me == 1) {
6048 fw_data = (const __be32 *)rdev->pfp_fw->data;
6049 table_offset = CP_ME_TABLE_OFFSET;
6050 } else if (me == 2) {
6051 fw_data = (const __be32 *)rdev->me_fw->data;
6052 table_offset = CP_ME_TABLE_OFFSET;
6054 fw_data = (const __be32 *)rdev->mec_fw->data;
6055 table_offset = CP_MEC_TABLE_OFFSET;
6058 for (i = 0; i < CP_ME_TABLE_SIZE; i ++) {
6059 dst_ptr[bo_offset + i] = cpu_to_le32(be32_to_cpu(fw_data[table_offset + i]));
6061 bo_offset += CP_ME_TABLE_SIZE;
6065 static void cik_enable_gfx_cgpg(struct radeon_device *rdev,
6070 if (enable && (rdev->pg_flags & RADEON_PG_SUPPORT_GFX_PG)) {
6071 orig = data = RREG32(RLC_PG_CNTL);
6072 data |= GFX_PG_ENABLE;
6074 WREG32(RLC_PG_CNTL, data);
6076 orig = data = RREG32(RLC_AUTO_PG_CTRL);
6079 WREG32(RLC_AUTO_PG_CTRL, data);
6081 orig = data = RREG32(RLC_PG_CNTL);
6082 data &= ~GFX_PG_ENABLE;
6084 WREG32(RLC_PG_CNTL, data);
6086 orig = data = RREG32(RLC_AUTO_PG_CTRL);
6087 data &= ~AUTO_PG_EN;
6089 WREG32(RLC_AUTO_PG_CTRL, data);
6091 data = RREG32(DB_RENDER_CONTROL);
6095 static u32 cik_get_cu_active_bitmap(struct radeon_device *rdev, u32 se, u32 sh)
6097 u32 mask = 0, tmp, tmp1;
6100 cik_select_se_sh(rdev, se, sh);
6101 tmp = RREG32(CC_GC_SHADER_ARRAY_CONFIG);
6102 tmp1 = RREG32(GC_USER_SHADER_ARRAY_CONFIG);
6103 cik_select_se_sh(rdev, 0xffffffff, 0xffffffff);
6110 for (i = 0; i < rdev->config.cik.max_cu_per_sh; i ++) {
6115 return (~tmp) & mask;
6118 static void cik_init_ao_cu_mask(struct radeon_device *rdev)
6120 u32 i, j, k, active_cu_number = 0;
6121 u32 mask, counter, cu_bitmap;
6124 for (i = 0; i < rdev->config.cik.max_shader_engines; i++) {
6125 for (j = 0; j < rdev->config.cik.max_sh_per_se; j++) {
6129 for (k = 0; k < rdev->config.cik.max_cu_per_sh; k ++) {
6130 if (cik_get_cu_active_bitmap(rdev, i, j) & mask) {
6138 active_cu_number += counter;
6139 tmp |= (cu_bitmap << (i * 16 + j * 8));
6143 WREG32(RLC_PG_AO_CU_MASK, tmp);
6145 tmp = RREG32(RLC_MAX_PG_CU);
6146 tmp &= ~MAX_PU_CU_MASK;
6147 tmp |= MAX_PU_CU(active_cu_number);
6148 WREG32(RLC_MAX_PG_CU, tmp);
6151 static void cik_enable_gfx_static_mgpg(struct radeon_device *rdev,
6156 orig = data = RREG32(RLC_PG_CNTL);
6157 if (enable && (rdev->pg_flags & RADEON_PG_SUPPORT_GFX_SMG))
6158 data |= STATIC_PER_CU_PG_ENABLE;
6160 data &= ~STATIC_PER_CU_PG_ENABLE;
6162 WREG32(RLC_PG_CNTL, data);
6165 static void cik_enable_gfx_dynamic_mgpg(struct radeon_device *rdev,
6170 orig = data = RREG32(RLC_PG_CNTL);
6171 if (enable && (rdev->pg_flags & RADEON_PG_SUPPORT_GFX_DMG))
6172 data |= DYN_PER_CU_PG_ENABLE;
6174 data &= ~DYN_PER_CU_PG_ENABLE;
6176 WREG32(RLC_PG_CNTL, data);
6179 #define RLC_SAVE_AND_RESTORE_STARTING_OFFSET 0x90
6180 #define RLC_CLEAR_STATE_DESCRIPTOR_OFFSET 0x3D
6182 static void cik_init_gfx_cgpg(struct radeon_device *rdev)
6187 if (rdev->rlc.cs_data) {
6188 WREG32(RLC_GPM_SCRATCH_ADDR, RLC_CLEAR_STATE_DESCRIPTOR_OFFSET);
6189 WREG32(RLC_GPM_SCRATCH_DATA, upper_32_bits(rdev->rlc.clear_state_gpu_addr));
6190 WREG32(RLC_GPM_SCRATCH_DATA, lower_32_bits(rdev->rlc.clear_state_gpu_addr));
6191 WREG32(RLC_GPM_SCRATCH_DATA, rdev->rlc.clear_state_size);
6193 WREG32(RLC_GPM_SCRATCH_ADDR, RLC_CLEAR_STATE_DESCRIPTOR_OFFSET);
6194 for (i = 0; i < 3; i++)
6195 WREG32(RLC_GPM_SCRATCH_DATA, 0);
6197 if (rdev->rlc.reg_list) {
6198 WREG32(RLC_GPM_SCRATCH_ADDR, RLC_SAVE_AND_RESTORE_STARTING_OFFSET);
6199 for (i = 0; i < rdev->rlc.reg_list_size; i++)
6200 WREG32(RLC_GPM_SCRATCH_DATA, rdev->rlc.reg_list[i]);
6203 orig = data = RREG32(RLC_PG_CNTL);
6206 WREG32(RLC_PG_CNTL, data);
6208 WREG32(RLC_SAVE_AND_RESTORE_BASE, rdev->rlc.save_restore_gpu_addr >> 8);
6209 WREG32(RLC_CP_TABLE_RESTORE, rdev->rlc.cp_table_gpu_addr >> 8);
6211 data = RREG32(CP_RB_WPTR_POLL_CNTL);
6212 data &= ~IDLE_POLL_COUNT_MASK;
6213 data |= IDLE_POLL_COUNT(0x60);
6214 WREG32(CP_RB_WPTR_POLL_CNTL, data);
6217 WREG32(RLC_PG_DELAY, data);
6219 data = RREG32(RLC_PG_DELAY_2);
6222 WREG32(RLC_PG_DELAY_2, data);
6224 data = RREG32(RLC_AUTO_PG_CTRL);
6225 data &= ~GRBM_REG_SGIT_MASK;
6226 data |= GRBM_REG_SGIT(0x700);
6227 WREG32(RLC_AUTO_PG_CTRL, data);
6231 static void cik_update_gfx_pg(struct radeon_device *rdev, bool enable)
6233 cik_enable_gfx_cgpg(rdev, enable);
6234 cik_enable_gfx_static_mgpg(rdev, enable);
6235 cik_enable_gfx_dynamic_mgpg(rdev, enable);
6238 u32 cik_get_csb_size(struct radeon_device *rdev)
6241 const struct cs_section_def *sect = NULL;
6242 const struct cs_extent_def *ext = NULL;
6244 if (rdev->rlc.cs_data == NULL)
6247 /* begin clear state */
6249 /* context control state */
6252 for (sect = rdev->rlc.cs_data; sect->section != NULL; ++sect) {
6253 for (ext = sect->section; ext->extent != NULL; ++ext) {
6254 if (sect->id == SECT_CONTEXT)
6255 count += 2 + ext->reg_count;
6260 /* pa_sc_raster_config/pa_sc_raster_config1 */
6262 /* end clear state */
6270 void cik_get_csb_buffer(struct radeon_device *rdev, volatile u32 *buffer)
6273 const struct cs_section_def *sect = NULL;
6274 const struct cs_extent_def *ext = NULL;
6276 if (rdev->rlc.cs_data == NULL)
6281 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0));
6282 buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
6284 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CONTEXT_CONTROL, 1));
6285 buffer[count++] = cpu_to_le32(0x80000000);
6286 buffer[count++] = cpu_to_le32(0x80000000);
6288 for (sect = rdev->rlc.cs_data; sect->section != NULL; ++sect) {
6289 for (ext = sect->section; ext->extent != NULL; ++ext) {
6290 if (sect->id == SECT_CONTEXT) {
6292 cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, ext->reg_count));
6293 buffer[count++] = cpu_to_le32(ext->reg_index - 0xa000);
6294 for (i = 0; i < ext->reg_count; i++)
6295 buffer[count++] = cpu_to_le32(ext->extent[i]);
6302 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, 2));
6303 buffer[count++] = cpu_to_le32(PA_SC_RASTER_CONFIG - PACKET3_SET_CONTEXT_REG_START);
6304 switch (rdev->family) {
6306 buffer[count++] = cpu_to_le32(0x16000012);
6307 buffer[count++] = cpu_to_le32(0x00000000);
6310 buffer[count++] = cpu_to_le32(0x00000000); /* XXX */
6311 buffer[count++] = cpu_to_le32(0x00000000);
6314 buffer[count++] = cpu_to_le32(0x00000000); /* XXX */
6315 buffer[count++] = cpu_to_le32(0x00000000);
6318 buffer[count++] = 0x3a00161a;
6319 buffer[count++] = 0x0000002e;
6322 buffer[count++] = cpu_to_le32(0x00000000);
6323 buffer[count++] = cpu_to_le32(0x00000000);
6327 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0));
6328 buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_END_CLEAR_STATE);
6330 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CLEAR_STATE, 0));
6331 buffer[count++] = cpu_to_le32(0);
6334 static void cik_init_pg(struct radeon_device *rdev)
6336 if (rdev->pg_flags) {
6337 cik_enable_sck_slowdown_on_pu(rdev, true);
6338 cik_enable_sck_slowdown_on_pd(rdev, true);
6339 if (rdev->pg_flags & RADEON_PG_SUPPORT_GFX_PG) {
6340 cik_init_gfx_cgpg(rdev);
6341 cik_enable_cp_pg(rdev, true);
6342 cik_enable_gds_pg(rdev, true);
6344 cik_init_ao_cu_mask(rdev);
6345 cik_update_gfx_pg(rdev, true);
6349 static void cik_fini_pg(struct radeon_device *rdev)
6351 if (rdev->pg_flags) {
6352 cik_update_gfx_pg(rdev, false);
6353 if (rdev->pg_flags & RADEON_PG_SUPPORT_GFX_PG) {
6354 cik_enable_cp_pg(rdev, false);
6355 cik_enable_gds_pg(rdev, false);
6362 * Starting with r6xx, interrupts are handled via a ring buffer.
6363 * Ring buffers are areas of GPU accessible memory that the GPU
6364 * writes interrupt vectors into and the host reads vectors out of.
6365 * There is a rptr (read pointer) that determines where the
6366 * host is currently reading, and a wptr (write pointer)
6367 * which determines where the GPU has written. When the
6368 * pointers are equal, the ring is idle. When the GPU
6369 * writes vectors to the ring buffer, it increments the
6370 * wptr. When there is an interrupt, the host then starts
6371 * fetching commands and processing them until the pointers are
6372 * equal again at which point it updates the rptr.
6376 * cik_enable_interrupts - Enable the interrupt ring buffer
6378 * @rdev: radeon_device pointer
6380 * Enable the interrupt ring buffer (CIK).
6382 static void cik_enable_interrupts(struct radeon_device *rdev)
6384 u32 ih_cntl = RREG32(IH_CNTL);
6385 u32 ih_rb_cntl = RREG32(IH_RB_CNTL);
6387 ih_cntl |= ENABLE_INTR;
6388 ih_rb_cntl |= IH_RB_ENABLE;
6389 WREG32(IH_CNTL, ih_cntl);
6390 WREG32(IH_RB_CNTL, ih_rb_cntl);
6391 rdev->ih.enabled = true;
6395 * cik_disable_interrupts - Disable the interrupt ring buffer
6397 * @rdev: radeon_device pointer
6399 * Disable the interrupt ring buffer (CIK).
6401 static void cik_disable_interrupts(struct radeon_device *rdev)
6403 u32 ih_rb_cntl = RREG32(IH_RB_CNTL);
6404 u32 ih_cntl = RREG32(IH_CNTL);
6406 ih_rb_cntl &= ~IH_RB_ENABLE;
6407 ih_cntl &= ~ENABLE_INTR;
6408 WREG32(IH_RB_CNTL, ih_rb_cntl);
6409 WREG32(IH_CNTL, ih_cntl);
6410 /* set rptr, wptr to 0 */
6411 WREG32(IH_RB_RPTR, 0);
6412 WREG32(IH_RB_WPTR, 0);
6413 rdev->ih.enabled = false;
6418 * cik_disable_interrupt_state - Disable all interrupt sources
6420 * @rdev: radeon_device pointer
6422 * Clear all interrupt enable bits used by the driver (CIK).
6424 static void cik_disable_interrupt_state(struct radeon_device *rdev)
6429 tmp = RREG32(CP_INT_CNTL_RING0) &
6430 (CNTX_BUSY_INT_ENABLE | CNTX_EMPTY_INT_ENABLE);
6431 WREG32(CP_INT_CNTL_RING0, tmp);
6433 tmp = RREG32(SDMA0_CNTL + SDMA0_REGISTER_OFFSET) & ~TRAP_ENABLE;
6434 WREG32(SDMA0_CNTL + SDMA0_REGISTER_OFFSET, tmp);
6435 tmp = RREG32(SDMA0_CNTL + SDMA1_REGISTER_OFFSET) & ~TRAP_ENABLE;
6436 WREG32(SDMA0_CNTL + SDMA1_REGISTER_OFFSET, tmp);
6437 /* compute queues */
6438 WREG32(CP_ME1_PIPE0_INT_CNTL, 0);
6439 WREG32(CP_ME1_PIPE1_INT_CNTL, 0);
6440 WREG32(CP_ME1_PIPE2_INT_CNTL, 0);
6441 WREG32(CP_ME1_PIPE3_INT_CNTL, 0);
6442 WREG32(CP_ME2_PIPE0_INT_CNTL, 0);
6443 WREG32(CP_ME2_PIPE1_INT_CNTL, 0);
6444 WREG32(CP_ME2_PIPE2_INT_CNTL, 0);
6445 WREG32(CP_ME2_PIPE3_INT_CNTL, 0);
6447 WREG32(GRBM_INT_CNTL, 0);
6448 /* vline/vblank, etc. */
6449 WREG32(LB_INTERRUPT_MASK + EVERGREEN_CRTC0_REGISTER_OFFSET, 0);
6450 WREG32(LB_INTERRUPT_MASK + EVERGREEN_CRTC1_REGISTER_OFFSET, 0);
6451 if (rdev->num_crtc >= 4) {
6452 WREG32(LB_INTERRUPT_MASK + EVERGREEN_CRTC2_REGISTER_OFFSET, 0);
6453 WREG32(LB_INTERRUPT_MASK + EVERGREEN_CRTC3_REGISTER_OFFSET, 0);
6455 if (rdev->num_crtc >= 6) {
6456 WREG32(LB_INTERRUPT_MASK + EVERGREEN_CRTC4_REGISTER_OFFSET, 0);
6457 WREG32(LB_INTERRUPT_MASK + EVERGREEN_CRTC5_REGISTER_OFFSET, 0);
6461 WREG32(DAC_AUTODETECT_INT_CONTROL, 0);
6463 /* digital hotplug */
6464 tmp = RREG32(DC_HPD1_INT_CONTROL) & DC_HPDx_INT_POLARITY;
6465 WREG32(DC_HPD1_INT_CONTROL, tmp);
6466 tmp = RREG32(DC_HPD2_INT_CONTROL) & DC_HPDx_INT_POLARITY;
6467 WREG32(DC_HPD2_INT_CONTROL, tmp);
6468 tmp = RREG32(DC_HPD3_INT_CONTROL) & DC_HPDx_INT_POLARITY;
6469 WREG32(DC_HPD3_INT_CONTROL, tmp);
6470 tmp = RREG32(DC_HPD4_INT_CONTROL) & DC_HPDx_INT_POLARITY;
6471 WREG32(DC_HPD4_INT_CONTROL, tmp);
6472 tmp = RREG32(DC_HPD5_INT_CONTROL) & DC_HPDx_INT_POLARITY;
6473 WREG32(DC_HPD5_INT_CONTROL, tmp);
6474 tmp = RREG32(DC_HPD6_INT_CONTROL) & DC_HPDx_INT_POLARITY;
6475 WREG32(DC_HPD6_INT_CONTROL, tmp);
6480 * cik_irq_init - init and enable the interrupt ring
6482 * @rdev: radeon_device pointer
6484 * Allocate a ring buffer for the interrupt controller,
6485 * enable the RLC, disable interrupts, enable the IH
6486 * ring buffer and enable it (CIK).
6487 * Called at device load and reume.
6488 * Returns 0 for success, errors for failure.
6490 static int cik_irq_init(struct radeon_device *rdev)
6494 u32 interrupt_cntl, ih_cntl, ih_rb_cntl;
6497 ret = r600_ih_ring_alloc(rdev);
6502 cik_disable_interrupts(rdev);
6505 ret = cik_rlc_resume(rdev);
6507 r600_ih_ring_fini(rdev);
6511 /* setup interrupt control */
6512 /* XXX this should actually be a bus address, not an MC address. same on older asics */
6513 WREG32(INTERRUPT_CNTL2, rdev->ih.gpu_addr >> 8);
6514 interrupt_cntl = RREG32(INTERRUPT_CNTL);
6515 /* IH_DUMMY_RD_OVERRIDE=0 - dummy read disabled with msi, enabled without msi
6516 * IH_DUMMY_RD_OVERRIDE=1 - dummy read controlled by IH_DUMMY_RD_EN
6518 interrupt_cntl &= ~IH_DUMMY_RD_OVERRIDE;
6519 /* IH_REQ_NONSNOOP_EN=1 if ring is in non-cacheable memory, e.g., vram */
6520 interrupt_cntl &= ~IH_REQ_NONSNOOP_EN;
6521 WREG32(INTERRUPT_CNTL, interrupt_cntl);
6523 WREG32(IH_RB_BASE, rdev->ih.gpu_addr >> 8);
6524 rb_bufsz = order_base_2(rdev->ih.ring_size / 4);
6526 ih_rb_cntl = (IH_WPTR_OVERFLOW_ENABLE |
6527 IH_WPTR_OVERFLOW_CLEAR |
6530 if (rdev->wb.enabled)
6531 ih_rb_cntl |= IH_WPTR_WRITEBACK_ENABLE;
6533 /* set the writeback address whether it's enabled or not */
6534 WREG32(IH_RB_WPTR_ADDR_LO, (rdev->wb.gpu_addr + R600_WB_IH_WPTR_OFFSET) & 0xFFFFFFFC);
6535 WREG32(IH_RB_WPTR_ADDR_HI, upper_32_bits(rdev->wb.gpu_addr + R600_WB_IH_WPTR_OFFSET) & 0xFF);
6537 WREG32(IH_RB_CNTL, ih_rb_cntl);
6539 /* set rptr, wptr to 0 */
6540 WREG32(IH_RB_RPTR, 0);
6541 WREG32(IH_RB_WPTR, 0);
6543 /* Default settings for IH_CNTL (disabled at first) */
6544 ih_cntl = MC_WRREQ_CREDIT(0x10) | MC_WR_CLEAN_CNT(0x10) | MC_VMID(0);
6545 /* RPTR_REARM only works if msi's are enabled */
6546 if (rdev->msi_enabled)
6547 ih_cntl |= RPTR_REARM;
6548 WREG32(IH_CNTL, ih_cntl);
6550 /* force the active interrupt state to all disabled */
6551 cik_disable_interrupt_state(rdev);
6553 pci_set_master(rdev->pdev);
6556 cik_enable_interrupts(rdev);
6562 * cik_irq_set - enable/disable interrupt sources
6564 * @rdev: radeon_device pointer
6566 * Enable interrupt sources on the GPU (vblanks, hpd,
6568 * Returns 0 for success, errors for failure.
6570 int cik_irq_set(struct radeon_device *rdev)
6573 u32 cp_m1p0, cp_m1p1, cp_m1p2, cp_m1p3;
6574 u32 cp_m2p0, cp_m2p1, cp_m2p2, cp_m2p3;
6575 u32 crtc1 = 0, crtc2 = 0, crtc3 = 0, crtc4 = 0, crtc5 = 0, crtc6 = 0;
6576 u32 hpd1, hpd2, hpd3, hpd4, hpd5, hpd6;
6577 u32 grbm_int_cntl = 0;
6578 u32 dma_cntl, dma_cntl1;
6581 if (!rdev->irq.installed) {
6582 WARN(1, "Can't enable IRQ/MSI because no handler is installed\n");
6585 /* don't enable anything if the ih is disabled */
6586 if (!rdev->ih.enabled) {
6587 cik_disable_interrupts(rdev);
6588 /* force the active interrupt state to all disabled */
6589 cik_disable_interrupt_state(rdev);
6593 cp_int_cntl = RREG32(CP_INT_CNTL_RING0) &
6594 (CNTX_BUSY_INT_ENABLE | CNTX_EMPTY_INT_ENABLE);
6595 cp_int_cntl |= PRIV_INSTR_INT_ENABLE | PRIV_REG_INT_ENABLE;
6597 hpd1 = RREG32(DC_HPD1_INT_CONTROL) & ~DC_HPDx_INT_EN;
6598 hpd2 = RREG32(DC_HPD2_INT_CONTROL) & ~DC_HPDx_INT_EN;
6599 hpd3 = RREG32(DC_HPD3_INT_CONTROL) & ~DC_HPDx_INT_EN;
6600 hpd4 = RREG32(DC_HPD4_INT_CONTROL) & ~DC_HPDx_INT_EN;
6601 hpd5 = RREG32(DC_HPD5_INT_CONTROL) & ~DC_HPDx_INT_EN;
6602 hpd6 = RREG32(DC_HPD6_INT_CONTROL) & ~DC_HPDx_INT_EN;
6604 dma_cntl = RREG32(SDMA0_CNTL + SDMA0_REGISTER_OFFSET) & ~TRAP_ENABLE;
6605 dma_cntl1 = RREG32(SDMA0_CNTL + SDMA1_REGISTER_OFFSET) & ~TRAP_ENABLE;
6607 cp_m1p0 = RREG32(CP_ME1_PIPE0_INT_CNTL) & ~TIME_STAMP_INT_ENABLE;
6608 cp_m1p1 = RREG32(CP_ME1_PIPE1_INT_CNTL) & ~TIME_STAMP_INT_ENABLE;
6609 cp_m1p2 = RREG32(CP_ME1_PIPE2_INT_CNTL) & ~TIME_STAMP_INT_ENABLE;
6610 cp_m1p3 = RREG32(CP_ME1_PIPE3_INT_CNTL) & ~TIME_STAMP_INT_ENABLE;
6611 cp_m2p0 = RREG32(CP_ME2_PIPE0_INT_CNTL) & ~TIME_STAMP_INT_ENABLE;
6612 cp_m2p1 = RREG32(CP_ME2_PIPE1_INT_CNTL) & ~TIME_STAMP_INT_ENABLE;
6613 cp_m2p2 = RREG32(CP_ME2_PIPE2_INT_CNTL) & ~TIME_STAMP_INT_ENABLE;
6614 cp_m2p3 = RREG32(CP_ME2_PIPE3_INT_CNTL) & ~TIME_STAMP_INT_ENABLE;
6616 if (rdev->flags & RADEON_IS_IGP)
6617 thermal_int = RREG32_SMC(CG_THERMAL_INT_CTRL) &
6618 ~(THERM_INTH_MASK | THERM_INTL_MASK);
6620 thermal_int = RREG32_SMC(CG_THERMAL_INT) &
6621 ~(THERM_INT_MASK_HIGH | THERM_INT_MASK_LOW);
6623 /* enable CP interrupts on all rings */
6624 if (atomic_read(&rdev->irq.ring_int[RADEON_RING_TYPE_GFX_INDEX])) {
6625 DRM_DEBUG("cik_irq_set: sw int gfx\n");
6626 cp_int_cntl |= TIME_STAMP_INT_ENABLE;
6628 if (atomic_read(&rdev->irq.ring_int[CAYMAN_RING_TYPE_CP1_INDEX])) {
6629 struct radeon_ring *ring = &rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX];
6630 DRM_DEBUG("si_irq_set: sw int cp1\n");
6631 if (ring->me == 1) {
6632 switch (ring->pipe) {
6634 cp_m1p0 |= TIME_STAMP_INT_ENABLE;
6637 cp_m1p1 |= TIME_STAMP_INT_ENABLE;
6640 cp_m1p2 |= TIME_STAMP_INT_ENABLE;
6643 cp_m1p2 |= TIME_STAMP_INT_ENABLE;
6646 DRM_DEBUG("si_irq_set: sw int cp1 invalid pipe %d\n", ring->pipe);
6649 } else if (ring->me == 2) {
6650 switch (ring->pipe) {
6652 cp_m2p0 |= TIME_STAMP_INT_ENABLE;
6655 cp_m2p1 |= TIME_STAMP_INT_ENABLE;
6658 cp_m2p2 |= TIME_STAMP_INT_ENABLE;
6661 cp_m2p2 |= TIME_STAMP_INT_ENABLE;
6664 DRM_DEBUG("si_irq_set: sw int cp1 invalid pipe %d\n", ring->pipe);
6668 DRM_DEBUG("si_irq_set: sw int cp1 invalid me %d\n", ring->me);
6671 if (atomic_read(&rdev->irq.ring_int[CAYMAN_RING_TYPE_CP2_INDEX])) {
6672 struct radeon_ring *ring = &rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX];
6673 DRM_DEBUG("si_irq_set: sw int cp2\n");
6674 if (ring->me == 1) {
6675 switch (ring->pipe) {
6677 cp_m1p0 |= TIME_STAMP_INT_ENABLE;
6680 cp_m1p1 |= TIME_STAMP_INT_ENABLE;
6683 cp_m1p2 |= TIME_STAMP_INT_ENABLE;
6686 cp_m1p2 |= TIME_STAMP_INT_ENABLE;
6689 DRM_DEBUG("si_irq_set: sw int cp2 invalid pipe %d\n", ring->pipe);
6692 } else if (ring->me == 2) {
6693 switch (ring->pipe) {
6695 cp_m2p0 |= TIME_STAMP_INT_ENABLE;
6698 cp_m2p1 |= TIME_STAMP_INT_ENABLE;
6701 cp_m2p2 |= TIME_STAMP_INT_ENABLE;
6704 cp_m2p2 |= TIME_STAMP_INT_ENABLE;
6707 DRM_DEBUG("si_irq_set: sw int cp2 invalid pipe %d\n", ring->pipe);
6711 DRM_DEBUG("si_irq_set: sw int cp2 invalid me %d\n", ring->me);
6715 if (atomic_read(&rdev->irq.ring_int[R600_RING_TYPE_DMA_INDEX])) {
6716 DRM_DEBUG("cik_irq_set: sw int dma\n");
6717 dma_cntl |= TRAP_ENABLE;
6720 if (atomic_read(&rdev->irq.ring_int[CAYMAN_RING_TYPE_DMA1_INDEX])) {
6721 DRM_DEBUG("cik_irq_set: sw int dma1\n");
6722 dma_cntl1 |= TRAP_ENABLE;
6725 if (rdev->irq.crtc_vblank_int[0] ||
6726 atomic_read(&rdev->irq.pflip[0])) {
6727 DRM_DEBUG("cik_irq_set: vblank 0\n");
6728 crtc1 |= VBLANK_INTERRUPT_MASK;
6730 if (rdev->irq.crtc_vblank_int[1] ||
6731 atomic_read(&rdev->irq.pflip[1])) {
6732 DRM_DEBUG("cik_irq_set: vblank 1\n");
6733 crtc2 |= VBLANK_INTERRUPT_MASK;
6735 if (rdev->irq.crtc_vblank_int[2] ||
6736 atomic_read(&rdev->irq.pflip[2])) {
6737 DRM_DEBUG("cik_irq_set: vblank 2\n");
6738 crtc3 |= VBLANK_INTERRUPT_MASK;
6740 if (rdev->irq.crtc_vblank_int[3] ||
6741 atomic_read(&rdev->irq.pflip[3])) {
6742 DRM_DEBUG("cik_irq_set: vblank 3\n");
6743 crtc4 |= VBLANK_INTERRUPT_MASK;
6745 if (rdev->irq.crtc_vblank_int[4] ||
6746 atomic_read(&rdev->irq.pflip[4])) {
6747 DRM_DEBUG("cik_irq_set: vblank 4\n");
6748 crtc5 |= VBLANK_INTERRUPT_MASK;
6750 if (rdev->irq.crtc_vblank_int[5] ||
6751 atomic_read(&rdev->irq.pflip[5])) {
6752 DRM_DEBUG("cik_irq_set: vblank 5\n");
6753 crtc6 |= VBLANK_INTERRUPT_MASK;
6755 if (rdev->irq.hpd[0]) {
6756 DRM_DEBUG("cik_irq_set: hpd 1\n");
6757 hpd1 |= DC_HPDx_INT_EN;
6759 if (rdev->irq.hpd[1]) {
6760 DRM_DEBUG("cik_irq_set: hpd 2\n");
6761 hpd2 |= DC_HPDx_INT_EN;
6763 if (rdev->irq.hpd[2]) {
6764 DRM_DEBUG("cik_irq_set: hpd 3\n");
6765 hpd3 |= DC_HPDx_INT_EN;
6767 if (rdev->irq.hpd[3]) {
6768 DRM_DEBUG("cik_irq_set: hpd 4\n");
6769 hpd4 |= DC_HPDx_INT_EN;
6771 if (rdev->irq.hpd[4]) {
6772 DRM_DEBUG("cik_irq_set: hpd 5\n");
6773 hpd5 |= DC_HPDx_INT_EN;
6775 if (rdev->irq.hpd[5]) {
6776 DRM_DEBUG("cik_irq_set: hpd 6\n");
6777 hpd6 |= DC_HPDx_INT_EN;
6780 if (rdev->irq.dpm_thermal) {
6781 DRM_DEBUG("dpm thermal\n");
6782 if (rdev->flags & RADEON_IS_IGP)
6783 thermal_int |= THERM_INTH_MASK | THERM_INTL_MASK;
6785 thermal_int |= THERM_INT_MASK_HIGH | THERM_INT_MASK_LOW;
6788 WREG32(CP_INT_CNTL_RING0, cp_int_cntl);
6790 WREG32(SDMA0_CNTL + SDMA0_REGISTER_OFFSET, dma_cntl);
6791 WREG32(SDMA0_CNTL + SDMA1_REGISTER_OFFSET, dma_cntl1);
6793 WREG32(CP_ME1_PIPE0_INT_CNTL, cp_m1p0);
6794 WREG32(CP_ME1_PIPE1_INT_CNTL, cp_m1p1);
6795 WREG32(CP_ME1_PIPE2_INT_CNTL, cp_m1p2);
6796 WREG32(CP_ME1_PIPE3_INT_CNTL, cp_m1p3);
6797 WREG32(CP_ME2_PIPE0_INT_CNTL, cp_m2p0);
6798 WREG32(CP_ME2_PIPE1_INT_CNTL, cp_m2p1);
6799 WREG32(CP_ME2_PIPE2_INT_CNTL, cp_m2p2);
6800 WREG32(CP_ME2_PIPE3_INT_CNTL, cp_m2p3);
6802 WREG32(GRBM_INT_CNTL, grbm_int_cntl);
6804 WREG32(LB_INTERRUPT_MASK + EVERGREEN_CRTC0_REGISTER_OFFSET, crtc1);
6805 WREG32(LB_INTERRUPT_MASK + EVERGREEN_CRTC1_REGISTER_OFFSET, crtc2);
6806 if (rdev->num_crtc >= 4) {
6807 WREG32(LB_INTERRUPT_MASK + EVERGREEN_CRTC2_REGISTER_OFFSET, crtc3);
6808 WREG32(LB_INTERRUPT_MASK + EVERGREEN_CRTC3_REGISTER_OFFSET, crtc4);
6810 if (rdev->num_crtc >= 6) {
6811 WREG32(LB_INTERRUPT_MASK + EVERGREEN_CRTC4_REGISTER_OFFSET, crtc5);
6812 WREG32(LB_INTERRUPT_MASK + EVERGREEN_CRTC5_REGISTER_OFFSET, crtc6);
6815 WREG32(DC_HPD1_INT_CONTROL, hpd1);
6816 WREG32(DC_HPD2_INT_CONTROL, hpd2);
6817 WREG32(DC_HPD3_INT_CONTROL, hpd3);
6818 WREG32(DC_HPD4_INT_CONTROL, hpd4);
6819 WREG32(DC_HPD5_INT_CONTROL, hpd5);
6820 WREG32(DC_HPD6_INT_CONTROL, hpd6);
6822 if (rdev->flags & RADEON_IS_IGP)
6823 WREG32_SMC(CG_THERMAL_INT_CTRL, thermal_int);
6825 WREG32_SMC(CG_THERMAL_INT, thermal_int);
6831 * cik_irq_ack - ack interrupt sources
6833 * @rdev: radeon_device pointer
6835 * Ack interrupt sources on the GPU (vblanks, hpd,
6836 * etc.) (CIK). Certain interrupts sources are sw
6837 * generated and do not require an explicit ack.
6839 static inline void cik_irq_ack(struct radeon_device *rdev)
6843 rdev->irq.stat_regs.cik.disp_int = RREG32(DISP_INTERRUPT_STATUS);
6844 rdev->irq.stat_regs.cik.disp_int_cont = RREG32(DISP_INTERRUPT_STATUS_CONTINUE);
6845 rdev->irq.stat_regs.cik.disp_int_cont2 = RREG32(DISP_INTERRUPT_STATUS_CONTINUE2);
6846 rdev->irq.stat_regs.cik.disp_int_cont3 = RREG32(DISP_INTERRUPT_STATUS_CONTINUE3);
6847 rdev->irq.stat_regs.cik.disp_int_cont4 = RREG32(DISP_INTERRUPT_STATUS_CONTINUE4);
6848 rdev->irq.stat_regs.cik.disp_int_cont5 = RREG32(DISP_INTERRUPT_STATUS_CONTINUE5);
6849 rdev->irq.stat_regs.cik.disp_int_cont6 = RREG32(DISP_INTERRUPT_STATUS_CONTINUE6);
6851 if (rdev->irq.stat_regs.cik.disp_int & LB_D1_VBLANK_INTERRUPT)
6852 WREG32(LB_VBLANK_STATUS + EVERGREEN_CRTC0_REGISTER_OFFSET, VBLANK_ACK);
6853 if (rdev->irq.stat_regs.cik.disp_int & LB_D1_VLINE_INTERRUPT)
6854 WREG32(LB_VLINE_STATUS + EVERGREEN_CRTC0_REGISTER_OFFSET, VLINE_ACK);
6855 if (rdev->irq.stat_regs.cik.disp_int_cont & LB_D2_VBLANK_INTERRUPT)
6856 WREG32(LB_VBLANK_STATUS + EVERGREEN_CRTC1_REGISTER_OFFSET, VBLANK_ACK);
6857 if (rdev->irq.stat_regs.cik.disp_int_cont & LB_D2_VLINE_INTERRUPT)
6858 WREG32(LB_VLINE_STATUS + EVERGREEN_CRTC1_REGISTER_OFFSET, VLINE_ACK);
6860 if (rdev->num_crtc >= 4) {
6861 if (rdev->irq.stat_regs.cik.disp_int_cont2 & LB_D3_VBLANK_INTERRUPT)
6862 WREG32(LB_VBLANK_STATUS + EVERGREEN_CRTC2_REGISTER_OFFSET, VBLANK_ACK);
6863 if (rdev->irq.stat_regs.cik.disp_int_cont2 & LB_D3_VLINE_INTERRUPT)
6864 WREG32(LB_VLINE_STATUS + EVERGREEN_CRTC2_REGISTER_OFFSET, VLINE_ACK);
6865 if (rdev->irq.stat_regs.cik.disp_int_cont3 & LB_D4_VBLANK_INTERRUPT)
6866 WREG32(LB_VBLANK_STATUS + EVERGREEN_CRTC3_REGISTER_OFFSET, VBLANK_ACK);
6867 if (rdev->irq.stat_regs.cik.disp_int_cont3 & LB_D4_VLINE_INTERRUPT)
6868 WREG32(LB_VLINE_STATUS + EVERGREEN_CRTC3_REGISTER_OFFSET, VLINE_ACK);
6871 if (rdev->num_crtc >= 6) {
6872 if (rdev->irq.stat_regs.cik.disp_int_cont4 & LB_D5_VBLANK_INTERRUPT)
6873 WREG32(LB_VBLANK_STATUS + EVERGREEN_CRTC4_REGISTER_OFFSET, VBLANK_ACK);
6874 if (rdev->irq.stat_regs.cik.disp_int_cont4 & LB_D5_VLINE_INTERRUPT)
6875 WREG32(LB_VLINE_STATUS + EVERGREEN_CRTC4_REGISTER_OFFSET, VLINE_ACK);
6876 if (rdev->irq.stat_regs.cik.disp_int_cont5 & LB_D6_VBLANK_INTERRUPT)
6877 WREG32(LB_VBLANK_STATUS + EVERGREEN_CRTC5_REGISTER_OFFSET, VBLANK_ACK);
6878 if (rdev->irq.stat_regs.cik.disp_int_cont5 & LB_D6_VLINE_INTERRUPT)
6879 WREG32(LB_VLINE_STATUS + EVERGREEN_CRTC5_REGISTER_OFFSET, VLINE_ACK);
6882 if (rdev->irq.stat_regs.cik.disp_int & DC_HPD1_INTERRUPT) {
6883 tmp = RREG32(DC_HPD1_INT_CONTROL);
6884 tmp |= DC_HPDx_INT_ACK;
6885 WREG32(DC_HPD1_INT_CONTROL, tmp);
6887 if (rdev->irq.stat_regs.cik.disp_int_cont & DC_HPD2_INTERRUPT) {
6888 tmp = RREG32(DC_HPD2_INT_CONTROL);
6889 tmp |= DC_HPDx_INT_ACK;
6890 WREG32(DC_HPD2_INT_CONTROL, tmp);
6892 if (rdev->irq.stat_regs.cik.disp_int_cont2 & DC_HPD3_INTERRUPT) {
6893 tmp = RREG32(DC_HPD3_INT_CONTROL);
6894 tmp |= DC_HPDx_INT_ACK;
6895 WREG32(DC_HPD3_INT_CONTROL, tmp);
6897 if (rdev->irq.stat_regs.cik.disp_int_cont3 & DC_HPD4_INTERRUPT) {
6898 tmp = RREG32(DC_HPD4_INT_CONTROL);
6899 tmp |= DC_HPDx_INT_ACK;
6900 WREG32(DC_HPD4_INT_CONTROL, tmp);
6902 if (rdev->irq.stat_regs.cik.disp_int_cont4 & DC_HPD5_INTERRUPT) {
6903 tmp = RREG32(DC_HPD5_INT_CONTROL);
6904 tmp |= DC_HPDx_INT_ACK;
6905 WREG32(DC_HPD5_INT_CONTROL, tmp);
6907 if (rdev->irq.stat_regs.cik.disp_int_cont5 & DC_HPD6_INTERRUPT) {
6908 tmp = RREG32(DC_HPD5_INT_CONTROL);
6909 tmp |= DC_HPDx_INT_ACK;
6910 WREG32(DC_HPD6_INT_CONTROL, tmp);
6915 * cik_irq_disable - disable interrupts
6917 * @rdev: radeon_device pointer
6919 * Disable interrupts on the hw (CIK).
6921 static void cik_irq_disable(struct radeon_device *rdev)
6923 cik_disable_interrupts(rdev);
6924 /* Wait and acknowledge irq */
6927 cik_disable_interrupt_state(rdev);
6931 * cik_irq_disable - disable interrupts for suspend
6933 * @rdev: radeon_device pointer
6935 * Disable interrupts and stop the RLC (CIK).
6938 static void cik_irq_suspend(struct radeon_device *rdev)
6940 cik_irq_disable(rdev);
6945 * cik_irq_fini - tear down interrupt support
6947 * @rdev: radeon_device pointer
6949 * Disable interrupts on the hw and free the IH ring
6951 * Used for driver unload.
6953 static void cik_irq_fini(struct radeon_device *rdev)
6955 cik_irq_suspend(rdev);
6956 r600_ih_ring_fini(rdev);
6960 * cik_get_ih_wptr - get the IH ring buffer wptr
6962 * @rdev: radeon_device pointer
6964 * Get the IH ring buffer wptr from either the register
6965 * or the writeback memory buffer (CIK). Also check for
6966 * ring buffer overflow and deal with it.
6967 * Used by cik_irq_process().
6968 * Returns the value of the wptr.
6970 static inline u32 cik_get_ih_wptr(struct radeon_device *rdev)
6974 if (rdev->wb.enabled)
6975 wptr = le32_to_cpu(rdev->wb.wb[R600_WB_IH_WPTR_OFFSET/4]);
6977 wptr = RREG32(IH_RB_WPTR);
6979 if (wptr & RB_OVERFLOW) {
6980 /* When a ring buffer overflow happen start parsing interrupt
6981 * from the last not overwritten vector (wptr + 16). Hopefully
6982 * this should allow us to catchup.
6984 dev_warn(rdev->dev, "IH ring buffer overflow (0x%08X, %d, %d)\n",
6985 wptr, rdev->ih.rptr, (wptr + 16) + rdev->ih.ptr_mask);
6986 rdev->ih.rptr = (wptr + 16) & rdev->ih.ptr_mask;
6987 tmp = RREG32(IH_RB_CNTL);
6988 tmp |= IH_WPTR_OVERFLOW_CLEAR;
6989 WREG32(IH_RB_CNTL, tmp);
6991 return (wptr & rdev->ih.ptr_mask);
6995 * Each IV ring entry is 128 bits:
6996 * [7:0] - interrupt source id
6998 * [59:32] - interrupt source data
6999 * [63:60] - reserved
7002 * ME_ID [1:0], PIPE_ID[1:0], QUEUE_ID[2:0]
7003 * QUEUE_ID - for compute, which of the 8 queues owned by the dispatcher
7004 * - for gfx, hw shader state (0=PS...5=LS, 6=CS)
7005 * ME_ID - 0 = gfx, 1 = first 4 CS pipes, 2 = second 4 CS pipes
7006 * PIPE_ID - ME0 0=3D
7007 * - ME1&2 compute dispatcher (4 pipes each)
7009 * INSTANCE_ID [1:0], QUEUE_ID[1:0]
7010 * INSTANCE_ID - 0 = sdma0, 1 = sdma1
7011 * QUEUE_ID - 0 = gfx, 1 = rlc0, 2 = rlc1
7014 * [127:96] - reserved
7017 * cik_irq_process - interrupt handler
7019 * @rdev: radeon_device pointer
7021 * Interrupt hander (CIK). Walk the IH ring,
7022 * ack interrupts and schedule work to handle
7024 * Returns irq process return code.
7026 int cik_irq_process(struct radeon_device *rdev)
7028 struct radeon_ring *cp1_ring = &rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX];
7029 struct radeon_ring *cp2_ring = &rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX];
7032 u32 src_id, src_data, ring_id;
7033 u8 me_id, pipe_id, queue_id;
7035 bool queue_hotplug = false;
7036 bool queue_reset = false;
7037 u32 addr, status, mc_client;
7038 bool queue_thermal = false;
7040 if (!rdev->ih.enabled || rdev->shutdown)
7043 wptr = cik_get_ih_wptr(rdev);
7046 /* is somebody else already processing irqs? */
7047 if (atomic_xchg(&rdev->ih.lock, 1))
7050 rptr = rdev->ih.rptr;
7051 DRM_DEBUG("cik_irq_process start: rptr %d, wptr %d\n", rptr, wptr);
7053 /* Order reading of wptr vs. reading of IH ring data */
7056 /* display interrupts */
7059 while (rptr != wptr) {
7060 /* wptr/rptr are in bytes! */
7061 ring_index = rptr / 4;
7062 src_id = le32_to_cpu(rdev->ih.ring[ring_index]) & 0xff;
7063 src_data = le32_to_cpu(rdev->ih.ring[ring_index + 1]) & 0xfffffff;
7064 ring_id = le32_to_cpu(rdev->ih.ring[ring_index + 2]) & 0xff;
7067 case 1: /* D1 vblank/vline */
7069 case 0: /* D1 vblank */
7070 if (rdev->irq.stat_regs.cik.disp_int & LB_D1_VBLANK_INTERRUPT) {
7071 if (rdev->irq.crtc_vblank_int[0]) {
7072 drm_handle_vblank(rdev->ddev, 0);
7073 rdev->pm.vblank_sync = true;
7074 wake_up(&rdev->irq.vblank_queue);
7076 if (atomic_read(&rdev->irq.pflip[0]))
7077 radeon_crtc_handle_flip(rdev, 0);
7078 rdev->irq.stat_regs.cik.disp_int &= ~LB_D1_VBLANK_INTERRUPT;
7079 DRM_DEBUG("IH: D1 vblank\n");
7082 case 1: /* D1 vline */
7083 if (rdev->irq.stat_regs.cik.disp_int & LB_D1_VLINE_INTERRUPT) {
7084 rdev->irq.stat_regs.cik.disp_int &= ~LB_D1_VLINE_INTERRUPT;
7085 DRM_DEBUG("IH: D1 vline\n");
7089 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
7093 case 2: /* D2 vblank/vline */
7095 case 0: /* D2 vblank */
7096 if (rdev->irq.stat_regs.cik.disp_int_cont & LB_D2_VBLANK_INTERRUPT) {
7097 if (rdev->irq.crtc_vblank_int[1]) {
7098 drm_handle_vblank(rdev->ddev, 1);
7099 rdev->pm.vblank_sync = true;
7100 wake_up(&rdev->irq.vblank_queue);
7102 if (atomic_read(&rdev->irq.pflip[1]))
7103 radeon_crtc_handle_flip(rdev, 1);
7104 rdev->irq.stat_regs.cik.disp_int_cont &= ~LB_D2_VBLANK_INTERRUPT;
7105 DRM_DEBUG("IH: D2 vblank\n");
7108 case 1: /* D2 vline */
7109 if (rdev->irq.stat_regs.cik.disp_int_cont & LB_D2_VLINE_INTERRUPT) {
7110 rdev->irq.stat_regs.cik.disp_int_cont &= ~LB_D2_VLINE_INTERRUPT;
7111 DRM_DEBUG("IH: D2 vline\n");
7115 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
7119 case 3: /* D3 vblank/vline */
7121 case 0: /* D3 vblank */
7122 if (rdev->irq.stat_regs.cik.disp_int_cont2 & LB_D3_VBLANK_INTERRUPT) {
7123 if (rdev->irq.crtc_vblank_int[2]) {
7124 drm_handle_vblank(rdev->ddev, 2);
7125 rdev->pm.vblank_sync = true;
7126 wake_up(&rdev->irq.vblank_queue);
7128 if (atomic_read(&rdev->irq.pflip[2]))
7129 radeon_crtc_handle_flip(rdev, 2);
7130 rdev->irq.stat_regs.cik.disp_int_cont2 &= ~LB_D3_VBLANK_INTERRUPT;
7131 DRM_DEBUG("IH: D3 vblank\n");
7134 case 1: /* D3 vline */
7135 if (rdev->irq.stat_regs.cik.disp_int_cont2 & LB_D3_VLINE_INTERRUPT) {
7136 rdev->irq.stat_regs.cik.disp_int_cont2 &= ~LB_D3_VLINE_INTERRUPT;
7137 DRM_DEBUG("IH: D3 vline\n");
7141 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
7145 case 4: /* D4 vblank/vline */
7147 case 0: /* D4 vblank */
7148 if (rdev->irq.stat_regs.cik.disp_int_cont3 & LB_D4_VBLANK_INTERRUPT) {
7149 if (rdev->irq.crtc_vblank_int[3]) {
7150 drm_handle_vblank(rdev->ddev, 3);
7151 rdev->pm.vblank_sync = true;
7152 wake_up(&rdev->irq.vblank_queue);
7154 if (atomic_read(&rdev->irq.pflip[3]))
7155 radeon_crtc_handle_flip(rdev, 3);
7156 rdev->irq.stat_regs.cik.disp_int_cont3 &= ~LB_D4_VBLANK_INTERRUPT;
7157 DRM_DEBUG("IH: D4 vblank\n");
7160 case 1: /* D4 vline */
7161 if (rdev->irq.stat_regs.cik.disp_int_cont3 & LB_D4_VLINE_INTERRUPT) {
7162 rdev->irq.stat_regs.cik.disp_int_cont3 &= ~LB_D4_VLINE_INTERRUPT;
7163 DRM_DEBUG("IH: D4 vline\n");
7167 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
7171 case 5: /* D5 vblank/vline */
7173 case 0: /* D5 vblank */
7174 if (rdev->irq.stat_regs.cik.disp_int_cont4 & LB_D5_VBLANK_INTERRUPT) {
7175 if (rdev->irq.crtc_vblank_int[4]) {
7176 drm_handle_vblank(rdev->ddev, 4);
7177 rdev->pm.vblank_sync = true;
7178 wake_up(&rdev->irq.vblank_queue);
7180 if (atomic_read(&rdev->irq.pflip[4]))
7181 radeon_crtc_handle_flip(rdev, 4);
7182 rdev->irq.stat_regs.cik.disp_int_cont4 &= ~LB_D5_VBLANK_INTERRUPT;
7183 DRM_DEBUG("IH: D5 vblank\n");
7186 case 1: /* D5 vline */
7187 if (rdev->irq.stat_regs.cik.disp_int_cont4 & LB_D5_VLINE_INTERRUPT) {
7188 rdev->irq.stat_regs.cik.disp_int_cont4 &= ~LB_D5_VLINE_INTERRUPT;
7189 DRM_DEBUG("IH: D5 vline\n");
7193 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
7197 case 6: /* D6 vblank/vline */
7199 case 0: /* D6 vblank */
7200 if (rdev->irq.stat_regs.cik.disp_int_cont5 & LB_D6_VBLANK_INTERRUPT) {
7201 if (rdev->irq.crtc_vblank_int[5]) {
7202 drm_handle_vblank(rdev->ddev, 5);
7203 rdev->pm.vblank_sync = true;
7204 wake_up(&rdev->irq.vblank_queue);
7206 if (atomic_read(&rdev->irq.pflip[5]))
7207 radeon_crtc_handle_flip(rdev, 5);
7208 rdev->irq.stat_regs.cik.disp_int_cont5 &= ~LB_D6_VBLANK_INTERRUPT;
7209 DRM_DEBUG("IH: D6 vblank\n");
7212 case 1: /* D6 vline */
7213 if (rdev->irq.stat_regs.cik.disp_int_cont5 & LB_D6_VLINE_INTERRUPT) {
7214 rdev->irq.stat_regs.cik.disp_int_cont5 &= ~LB_D6_VLINE_INTERRUPT;
7215 DRM_DEBUG("IH: D6 vline\n");
7219 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
7223 case 42: /* HPD hotplug */
7226 if (rdev->irq.stat_regs.cik.disp_int & DC_HPD1_INTERRUPT) {
7227 rdev->irq.stat_regs.cik.disp_int &= ~DC_HPD1_INTERRUPT;
7228 queue_hotplug = true;
7229 DRM_DEBUG("IH: HPD1\n");
7233 if (rdev->irq.stat_regs.cik.disp_int_cont & DC_HPD2_INTERRUPT) {
7234 rdev->irq.stat_regs.cik.disp_int_cont &= ~DC_HPD2_INTERRUPT;
7235 queue_hotplug = true;
7236 DRM_DEBUG("IH: HPD2\n");
7240 if (rdev->irq.stat_regs.cik.disp_int_cont2 & DC_HPD3_INTERRUPT) {
7241 rdev->irq.stat_regs.cik.disp_int_cont2 &= ~DC_HPD3_INTERRUPT;
7242 queue_hotplug = true;
7243 DRM_DEBUG("IH: HPD3\n");
7247 if (rdev->irq.stat_regs.cik.disp_int_cont3 & DC_HPD4_INTERRUPT) {
7248 rdev->irq.stat_regs.cik.disp_int_cont3 &= ~DC_HPD4_INTERRUPT;
7249 queue_hotplug = true;
7250 DRM_DEBUG("IH: HPD4\n");
7254 if (rdev->irq.stat_regs.cik.disp_int_cont4 & DC_HPD5_INTERRUPT) {
7255 rdev->irq.stat_regs.cik.disp_int_cont4 &= ~DC_HPD5_INTERRUPT;
7256 queue_hotplug = true;
7257 DRM_DEBUG("IH: HPD5\n");
7261 if (rdev->irq.stat_regs.cik.disp_int_cont5 & DC_HPD6_INTERRUPT) {
7262 rdev->irq.stat_regs.cik.disp_int_cont5 &= ~DC_HPD6_INTERRUPT;
7263 queue_hotplug = true;
7264 DRM_DEBUG("IH: HPD6\n");
7268 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
7273 DRM_DEBUG("IH: UVD int: 0x%08x\n", src_data);
7274 radeon_fence_process(rdev, R600_RING_TYPE_UVD_INDEX);
7278 addr = RREG32(VM_CONTEXT1_PROTECTION_FAULT_ADDR);
7279 status = RREG32(VM_CONTEXT1_PROTECTION_FAULT_STATUS);
7280 mc_client = RREG32(VM_CONTEXT1_PROTECTION_FAULT_MCCLIENT);
7281 dev_err(rdev->dev, "GPU fault detected: %d 0x%08x\n", src_id, src_data);
7282 dev_err(rdev->dev, " VM_CONTEXT1_PROTECTION_FAULT_ADDR 0x%08X\n",
7284 dev_err(rdev->dev, " VM_CONTEXT1_PROTECTION_FAULT_STATUS 0x%08X\n",
7286 cik_vm_decode_fault(rdev, status, addr, mc_client);
7287 /* reset addr and status */
7288 WREG32_P(VM_CONTEXT1_CNTL2, 1, ~1);
7290 case 176: /* GFX RB CP_INT */
7291 case 177: /* GFX IB CP_INT */
7292 radeon_fence_process(rdev, RADEON_RING_TYPE_GFX_INDEX);
7294 case 181: /* CP EOP event */
7295 DRM_DEBUG("IH: CP EOP\n");
7296 /* XXX check the bitfield order! */
7297 me_id = (ring_id & 0x60) >> 5;
7298 pipe_id = (ring_id & 0x18) >> 3;
7299 queue_id = (ring_id & 0x7) >> 0;
7302 radeon_fence_process(rdev, RADEON_RING_TYPE_GFX_INDEX);
7306 if ((cp1_ring->me == me_id) & (cp1_ring->pipe == pipe_id))
7307 radeon_fence_process(rdev, CAYMAN_RING_TYPE_CP1_INDEX);
7308 if ((cp2_ring->me == me_id) & (cp2_ring->pipe == pipe_id))
7309 radeon_fence_process(rdev, CAYMAN_RING_TYPE_CP2_INDEX);
7313 case 184: /* CP Privileged reg access */
7314 DRM_ERROR("Illegal register access in command stream\n");
7315 /* XXX check the bitfield order! */
7316 me_id = (ring_id & 0x60) >> 5;
7317 pipe_id = (ring_id & 0x18) >> 3;
7318 queue_id = (ring_id & 0x7) >> 0;
7321 /* This results in a full GPU reset, but all we need to do is soft
7322 * reset the CP for gfx
7336 case 185: /* CP Privileged inst */
7337 DRM_ERROR("Illegal instruction in command stream\n");
7338 /* XXX check the bitfield order! */
7339 me_id = (ring_id & 0x60) >> 5;
7340 pipe_id = (ring_id & 0x18) >> 3;
7341 queue_id = (ring_id & 0x7) >> 0;
7344 /* This results in a full GPU reset, but all we need to do is soft
7345 * reset the CP for gfx
7359 case 224: /* SDMA trap event */
7360 /* XXX check the bitfield order! */
7361 me_id = (ring_id & 0x3) >> 0;
7362 queue_id = (ring_id & 0xc) >> 2;
7363 DRM_DEBUG("IH: SDMA trap\n");
7368 radeon_fence_process(rdev, R600_RING_TYPE_DMA_INDEX);
7381 radeon_fence_process(rdev, CAYMAN_RING_TYPE_DMA1_INDEX);
7393 case 230: /* thermal low to high */
7394 DRM_DEBUG("IH: thermal low to high\n");
7395 rdev->pm.dpm.thermal.high_to_low = false;
7396 queue_thermal = true;
7398 case 231: /* thermal high to low */
7399 DRM_DEBUG("IH: thermal high to low\n");
7400 rdev->pm.dpm.thermal.high_to_low = true;
7401 queue_thermal = true;
7403 case 233: /* GUI IDLE */
7404 DRM_DEBUG("IH: GUI idle\n");
7406 case 241: /* SDMA Privileged inst */
7407 case 247: /* SDMA Privileged inst */
7408 DRM_ERROR("Illegal instruction in SDMA command stream\n");
7409 /* XXX check the bitfield order! */
7410 me_id = (ring_id & 0x3) >> 0;
7411 queue_id = (ring_id & 0xc) >> 2;
7446 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
7450 /* wptr/rptr are in bytes! */
7452 rptr &= rdev->ih.ptr_mask;
7455 schedule_work(&rdev->hotplug_work);
7457 schedule_work(&rdev->reset_work);
7459 schedule_work(&rdev->pm.dpm.thermal.work);
7460 rdev->ih.rptr = rptr;
7461 WREG32(IH_RB_RPTR, rdev->ih.rptr);
7462 atomic_set(&rdev->ih.lock, 0);
7464 /* make sure wptr hasn't changed while processing */
7465 wptr = cik_get_ih_wptr(rdev);
7473 * startup/shutdown callbacks
7476 * cik_startup - program the asic to a functional state
7478 * @rdev: radeon_device pointer
7480 * Programs the asic to a functional state (CIK).
7481 * Called by cik_init() and cik_resume().
7482 * Returns 0 for success, error for failure.
7484 static int cik_startup(struct radeon_device *rdev)
7486 struct radeon_ring *ring;
7489 /* enable pcie gen2/3 link */
7490 cik_pcie_gen3_enable(rdev);
7492 cik_program_aspm(rdev);
7494 /* scratch needs to be initialized before MC */
7495 r = r600_vram_scratch_init(rdev);
7499 cik_mc_program(rdev);
7501 if (rdev->flags & RADEON_IS_IGP) {
7502 if (!rdev->me_fw || !rdev->pfp_fw || !rdev->ce_fw ||
7503 !rdev->mec_fw || !rdev->sdma_fw || !rdev->rlc_fw) {
7504 r = cik_init_microcode(rdev);
7506 DRM_ERROR("Failed to load firmware!\n");
7511 if (!rdev->me_fw || !rdev->pfp_fw || !rdev->ce_fw ||
7512 !rdev->mec_fw || !rdev->sdma_fw || !rdev->rlc_fw ||
7514 r = cik_init_microcode(rdev);
7516 DRM_ERROR("Failed to load firmware!\n");
7521 r = ci_mc_load_microcode(rdev);
7523 DRM_ERROR("Failed to load MC firmware!\n");
7528 r = cik_pcie_gart_enable(rdev);
7533 /* allocate rlc buffers */
7534 if (rdev->flags & RADEON_IS_IGP) {
7535 if (rdev->family == CHIP_KAVERI) {
7536 rdev->rlc.reg_list = spectre_rlc_save_restore_register_list;
7537 rdev->rlc.reg_list_size =
7538 (u32)ARRAY_SIZE(spectre_rlc_save_restore_register_list);
7540 rdev->rlc.reg_list = kalindi_rlc_save_restore_register_list;
7541 rdev->rlc.reg_list_size =
7542 (u32)ARRAY_SIZE(kalindi_rlc_save_restore_register_list);
7545 rdev->rlc.cs_data = ci_cs_data;
7546 rdev->rlc.cp_table_size = CP_ME_TABLE_SIZE * 5 * 4;
7547 r = sumo_rlc_init(rdev);
7549 DRM_ERROR("Failed to init rlc BOs!\n");
7553 /* allocate wb buffer */
7554 r = radeon_wb_init(rdev);
7558 /* allocate mec buffers */
7559 r = cik_mec_init(rdev);
7561 DRM_ERROR("Failed to init MEC BOs!\n");
7565 r = radeon_fence_driver_start_ring(rdev, RADEON_RING_TYPE_GFX_INDEX);
7567 dev_err(rdev->dev, "failed initializing CP fences (%d).\n", r);
7571 r = radeon_fence_driver_start_ring(rdev, CAYMAN_RING_TYPE_CP1_INDEX);
7573 dev_err(rdev->dev, "failed initializing CP fences (%d).\n", r);
7577 r = radeon_fence_driver_start_ring(rdev, CAYMAN_RING_TYPE_CP2_INDEX);
7579 dev_err(rdev->dev, "failed initializing CP fences (%d).\n", r);
7583 r = radeon_fence_driver_start_ring(rdev, R600_RING_TYPE_DMA_INDEX);
7585 dev_err(rdev->dev, "failed initializing DMA fences (%d).\n", r);
7589 r = radeon_fence_driver_start_ring(rdev, CAYMAN_RING_TYPE_DMA1_INDEX);
7591 dev_err(rdev->dev, "failed initializing DMA fences (%d).\n", r);
7595 r = radeon_uvd_resume(rdev);
7597 r = uvd_v4_2_resume(rdev);
7599 r = radeon_fence_driver_start_ring(rdev,
7600 R600_RING_TYPE_UVD_INDEX);
7602 dev_err(rdev->dev, "UVD fences init error (%d).\n", r);
7606 rdev->ring[R600_RING_TYPE_UVD_INDEX].ring_size = 0;
7609 if (!rdev->irq.installed) {
7610 r = radeon_irq_kms_init(rdev);
7615 r = cik_irq_init(rdev);
7617 DRM_ERROR("radeon: IH init failed (%d).\n", r);
7618 radeon_irq_kms_fini(rdev);
7623 ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
7624 r = radeon_ring_init(rdev, ring, ring->ring_size, RADEON_WB_CP_RPTR_OFFSET,
7625 CP_RB0_RPTR, CP_RB0_WPTR,
7626 PACKET3(PACKET3_NOP, 0x3FFF));
7630 /* set up the compute queues */
7631 /* type-2 packets are deprecated on MEC, use type-3 instead */
7632 ring = &rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX];
7633 r = radeon_ring_init(rdev, ring, ring->ring_size, RADEON_WB_CP1_RPTR_OFFSET,
7634 CP_HQD_PQ_RPTR, CP_HQD_PQ_WPTR,
7635 PACKET3(PACKET3_NOP, 0x3FFF));
7638 ring->me = 1; /* first MEC */
7639 ring->pipe = 0; /* first pipe */
7640 ring->queue = 0; /* first queue */
7641 ring->wptr_offs = CIK_WB_CP1_WPTR_OFFSET;
7643 /* type-2 packets are deprecated on MEC, use type-3 instead */
7644 ring = &rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX];
7645 r = radeon_ring_init(rdev, ring, ring->ring_size, RADEON_WB_CP2_RPTR_OFFSET,
7646 CP_HQD_PQ_RPTR, CP_HQD_PQ_WPTR,
7647 PACKET3(PACKET3_NOP, 0x3FFF));
7650 /* dGPU only have 1 MEC */
7651 ring->me = 1; /* first MEC */
7652 ring->pipe = 0; /* first pipe */
7653 ring->queue = 1; /* second queue */
7654 ring->wptr_offs = CIK_WB_CP2_WPTR_OFFSET;
7656 ring = &rdev->ring[R600_RING_TYPE_DMA_INDEX];
7657 r = radeon_ring_init(rdev, ring, ring->ring_size, R600_WB_DMA_RPTR_OFFSET,
7658 SDMA0_GFX_RB_RPTR + SDMA0_REGISTER_OFFSET,
7659 SDMA0_GFX_RB_WPTR + SDMA0_REGISTER_OFFSET,
7660 SDMA_PACKET(SDMA_OPCODE_NOP, 0, 0));
7664 ring = &rdev->ring[CAYMAN_RING_TYPE_DMA1_INDEX];
7665 r = radeon_ring_init(rdev, ring, ring->ring_size, CAYMAN_WB_DMA1_RPTR_OFFSET,
7666 SDMA0_GFX_RB_RPTR + SDMA1_REGISTER_OFFSET,
7667 SDMA0_GFX_RB_WPTR + SDMA1_REGISTER_OFFSET,
7668 SDMA_PACKET(SDMA_OPCODE_NOP, 0, 0));
7672 r = cik_cp_resume(rdev);
7676 r = cik_sdma_resume(rdev);
7680 ring = &rdev->ring[R600_RING_TYPE_UVD_INDEX];
7681 if (ring->ring_size) {
7682 r = radeon_ring_init(rdev, ring, ring->ring_size, 0,
7683 UVD_RBC_RB_RPTR, UVD_RBC_RB_WPTR,
7686 r = uvd_v1_0_init(rdev);
7688 DRM_ERROR("radeon: failed initializing UVD (%d).\n", r);
7691 r = radeon_ib_pool_init(rdev);
7693 dev_err(rdev->dev, "IB initialization failed (%d).\n", r);
7697 r = radeon_vm_manager_init(rdev);
7699 dev_err(rdev->dev, "vm manager initialization failed (%d).\n", r);
7703 r = dce6_audio_init(rdev);
7711 * cik_resume - resume the asic to a functional state
7713 * @rdev: radeon_device pointer
7715 * Programs the asic to a functional state (CIK).
7717 * Returns 0 for success, error for failure.
7719 int cik_resume(struct radeon_device *rdev)
7724 atom_asic_init(rdev->mode_info.atom_context);
7726 /* init golden registers */
7727 cik_init_golden_registers(rdev);
7729 rdev->accel_working = true;
7730 r = cik_startup(rdev);
7732 DRM_ERROR("cik startup failed on resume\n");
7733 rdev->accel_working = false;
7742 * cik_suspend - suspend the asic
7744 * @rdev: radeon_device pointer
7746 * Bring the chip into a state suitable for suspend (CIK).
7747 * Called at suspend.
7748 * Returns 0 for success.
7750 int cik_suspend(struct radeon_device *rdev)
7752 dce6_audio_fini(rdev);
7753 radeon_vm_manager_fini(rdev);
7754 cik_cp_enable(rdev, false);
7755 cik_sdma_enable(rdev, false);
7756 uvd_v1_0_fini(rdev);
7757 radeon_uvd_suspend(rdev);
7760 cik_irq_suspend(rdev);
7761 radeon_wb_disable(rdev);
7762 cik_pcie_gart_disable(rdev);
7766 /* Plan is to move initialization in that function and use
7767 * helper function so that radeon_device_init pretty much
7768 * do nothing more than calling asic specific function. This
7769 * should also allow to remove a bunch of callback function
7773 * cik_init - asic specific driver and hw init
7775 * @rdev: radeon_device pointer
7777 * Setup asic specific driver variables and program the hw
7778 * to a functional state (CIK).
7779 * Called at driver startup.
7780 * Returns 0 for success, errors for failure.
7782 int cik_init(struct radeon_device *rdev)
7784 struct radeon_ring *ring;
7788 if (!radeon_get_bios(rdev)) {
7789 if (ASIC_IS_AVIVO(rdev))
7792 /* Must be an ATOMBIOS */
7793 if (!rdev->is_atom_bios) {
7794 dev_err(rdev->dev, "Expecting atombios for cayman GPU\n");
7797 r = radeon_atombios_init(rdev);
7801 /* Post card if necessary */
7802 if (!radeon_card_posted(rdev)) {
7804 dev_err(rdev->dev, "Card not posted and no BIOS - ignoring\n");
7807 DRM_INFO("GPU not posted. posting now...\n");
7808 atom_asic_init(rdev->mode_info.atom_context);
7810 /* init golden registers */
7811 cik_init_golden_registers(rdev);
7812 /* Initialize scratch registers */
7813 cik_scratch_init(rdev);
7814 /* Initialize surface registers */
7815 radeon_surface_init(rdev);
7816 /* Initialize clocks */
7817 radeon_get_clock_info(rdev->ddev);
7820 r = radeon_fence_driver_init(rdev);
7824 /* initialize memory controller */
7825 r = cik_mc_init(rdev);
7828 /* Memory manager */
7829 r = radeon_bo_init(rdev);
7833 ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
7834 ring->ring_obj = NULL;
7835 r600_ring_init(rdev, ring, 1024 * 1024);
7837 ring = &rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX];
7838 ring->ring_obj = NULL;
7839 r600_ring_init(rdev, ring, 1024 * 1024);
7840 r = radeon_doorbell_get(rdev, &ring->doorbell_index);
7844 ring = &rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX];
7845 ring->ring_obj = NULL;
7846 r600_ring_init(rdev, ring, 1024 * 1024);
7847 r = radeon_doorbell_get(rdev, &ring->doorbell_index);
7851 ring = &rdev->ring[R600_RING_TYPE_DMA_INDEX];
7852 ring->ring_obj = NULL;
7853 r600_ring_init(rdev, ring, 256 * 1024);
7855 ring = &rdev->ring[CAYMAN_RING_TYPE_DMA1_INDEX];
7856 ring->ring_obj = NULL;
7857 r600_ring_init(rdev, ring, 256 * 1024);
7859 r = radeon_uvd_init(rdev);
7861 ring = &rdev->ring[R600_RING_TYPE_UVD_INDEX];
7862 ring->ring_obj = NULL;
7863 r600_ring_init(rdev, ring, 4096);
7866 rdev->ih.ring_obj = NULL;
7867 r600_ih_ring_init(rdev, 64 * 1024);
7869 r = r600_pcie_gart_init(rdev);
7873 rdev->accel_working = true;
7874 r = cik_startup(rdev);
7876 dev_err(rdev->dev, "disabling GPU acceleration\n");
7878 cik_sdma_fini(rdev);
7880 sumo_rlc_fini(rdev);
7882 radeon_wb_fini(rdev);
7883 radeon_ib_pool_fini(rdev);
7884 radeon_vm_manager_fini(rdev);
7885 radeon_irq_kms_fini(rdev);
7886 cik_pcie_gart_fini(rdev);
7887 rdev->accel_working = false;
7890 /* Don't start up if the MC ucode is missing.
7891 * The default clocks and voltages before the MC ucode
7892 * is loaded are not suffient for advanced operations.
7894 if (!rdev->mc_fw && !(rdev->flags & RADEON_IS_IGP)) {
7895 DRM_ERROR("radeon: MC ucode required for NI+.\n");
7903 * cik_fini - asic specific driver and hw fini
7905 * @rdev: radeon_device pointer
7907 * Tear down the asic specific driver variables and program the hw
7908 * to an idle state (CIK).
7909 * Called at driver unload.
7911 void cik_fini(struct radeon_device *rdev)
7914 cik_sdma_fini(rdev);
7918 sumo_rlc_fini(rdev);
7920 radeon_wb_fini(rdev);
7921 radeon_vm_manager_fini(rdev);
7922 radeon_ib_pool_fini(rdev);
7923 radeon_irq_kms_fini(rdev);
7924 uvd_v1_0_fini(rdev);
7925 radeon_uvd_fini(rdev);
7926 cik_pcie_gart_fini(rdev);
7927 r600_vram_scratch_fini(rdev);
7928 radeon_gem_fini(rdev);
7929 radeon_fence_driver_fini(rdev);
7930 radeon_bo_fini(rdev);
7931 radeon_atombios_fini(rdev);
7936 void dce8_program_fmt(struct drm_encoder *encoder)
7938 struct drm_device *dev = encoder->dev;
7939 struct radeon_device *rdev = dev->dev_private;
7940 struct radeon_encoder *radeon_encoder = to_radeon_encoder(encoder);
7941 struct radeon_crtc *radeon_crtc = to_radeon_crtc(encoder->crtc);
7942 struct drm_connector *connector = radeon_get_connector_for_encoder(encoder);
7945 enum radeon_connector_dither dither = RADEON_FMT_DITHER_DISABLE;
7948 struct radeon_connector *radeon_connector = to_radeon_connector(connector);
7949 bpc = radeon_get_monitor_bpc(connector);
7950 dither = radeon_connector->dither;
7953 /* LVDS/eDP FMT is set up by atom */
7954 if (radeon_encoder->devices & ATOM_DEVICE_LCD_SUPPORT)
7957 /* not needed for analog */
7958 if ((radeon_encoder->encoder_id == ENCODER_OBJECT_ID_INTERNAL_KLDSCP_DAC1) ||
7959 (radeon_encoder->encoder_id == ENCODER_OBJECT_ID_INTERNAL_KLDSCP_DAC2))
7967 if (dither == RADEON_FMT_DITHER_ENABLE)
7968 /* XXX sort out optimal dither settings */
7969 tmp |= (FMT_FRAME_RANDOM_ENABLE | FMT_HIGHPASS_RANDOM_ENABLE |
7970 FMT_SPATIAL_DITHER_EN | FMT_SPATIAL_DITHER_DEPTH(0));
7972 tmp |= (FMT_TRUNCATE_EN | FMT_TRUNCATE_DEPTH(0));
7975 if (dither == RADEON_FMT_DITHER_ENABLE)
7976 /* XXX sort out optimal dither settings */
7977 tmp |= (FMT_FRAME_RANDOM_ENABLE | FMT_HIGHPASS_RANDOM_ENABLE |
7978 FMT_RGB_RANDOM_ENABLE |
7979 FMT_SPATIAL_DITHER_EN | FMT_SPATIAL_DITHER_DEPTH(1));
7981 tmp |= (FMT_TRUNCATE_EN | FMT_TRUNCATE_DEPTH(1));
7984 if (dither == RADEON_FMT_DITHER_ENABLE)
7985 /* XXX sort out optimal dither settings */
7986 tmp |= (FMT_FRAME_RANDOM_ENABLE | FMT_HIGHPASS_RANDOM_ENABLE |
7987 FMT_RGB_RANDOM_ENABLE |
7988 FMT_SPATIAL_DITHER_EN | FMT_SPATIAL_DITHER_DEPTH(2));
7990 tmp |= (FMT_TRUNCATE_EN | FMT_TRUNCATE_DEPTH(2));
7997 WREG32(FMT_BIT_DEPTH_CONTROL + radeon_crtc->crtc_offset, tmp);
8000 /* display watermark setup */
8002 * dce8_line_buffer_adjust - Set up the line buffer
8004 * @rdev: radeon_device pointer
8005 * @radeon_crtc: the selected display controller
8006 * @mode: the current display mode on the selected display
8009 * Setup up the line buffer allocation for
8010 * the selected display controller (CIK).
8011 * Returns the line buffer size in pixels.
8013 static u32 dce8_line_buffer_adjust(struct radeon_device *rdev,
8014 struct radeon_crtc *radeon_crtc,
8015 struct drm_display_mode *mode)
8017 u32 tmp, buffer_alloc, i;
8018 u32 pipe_offset = radeon_crtc->crtc_id * 0x20;
8021 * There are 6 line buffers, one for each display controllers.
8022 * There are 3 partitions per LB. Select the number of partitions
8023 * to enable based on the display width. For display widths larger
8024 * than 4096, you need use to use 2 display controllers and combine
8025 * them using the stereo blender.
8027 if (radeon_crtc->base.enabled && mode) {
8028 if (mode->crtc_hdisplay < 1920) {
8031 } else if (mode->crtc_hdisplay < 2560) {
8034 } else if (mode->crtc_hdisplay < 4096) {
8036 buffer_alloc = (rdev->flags & RADEON_IS_IGP) ? 2 : 4;
8038 DRM_DEBUG_KMS("Mode too big for LB!\n");
8040 buffer_alloc = (rdev->flags & RADEON_IS_IGP) ? 2 : 4;
8047 WREG32(LB_MEMORY_CTRL + radeon_crtc->crtc_offset,
8048 LB_MEMORY_CONFIG(tmp) | LB_MEMORY_SIZE(0x6B0));
8050 WREG32(PIPE0_DMIF_BUFFER_CONTROL + pipe_offset,
8051 DMIF_BUFFERS_ALLOCATED(buffer_alloc));
8052 for (i = 0; i < rdev->usec_timeout; i++) {
8053 if (RREG32(PIPE0_DMIF_BUFFER_CONTROL + pipe_offset) &
8054 DMIF_BUFFERS_ALLOCATED_COMPLETED)
8059 if (radeon_crtc->base.enabled && mode) {
8071 /* controller not enabled, so no lb used */
8076 * cik_get_number_of_dram_channels - get the number of dram channels
8078 * @rdev: radeon_device pointer
8080 * Look up the number of video ram channels (CIK).
8081 * Used for display watermark bandwidth calculations
8082 * Returns the number of dram channels
8084 static u32 cik_get_number_of_dram_channels(struct radeon_device *rdev)
8086 u32 tmp = RREG32(MC_SHARED_CHMAP);
8088 switch ((tmp & NOOFCHAN_MASK) >> NOOFCHAN_SHIFT) {
8111 struct dce8_wm_params {
8112 u32 dram_channels; /* number of dram channels */
8113 u32 yclk; /* bandwidth per dram data pin in kHz */
8114 u32 sclk; /* engine clock in kHz */
8115 u32 disp_clk; /* display clock in kHz */
8116 u32 src_width; /* viewport width */
8117 u32 active_time; /* active display time in ns */
8118 u32 blank_time; /* blank time in ns */
8119 bool interlaced; /* mode is interlaced */
8120 fixed20_12 vsc; /* vertical scale ratio */
8121 u32 num_heads; /* number of active crtcs */
8122 u32 bytes_per_pixel; /* bytes per pixel display + overlay */
8123 u32 lb_size; /* line buffer allocated to pipe */
8124 u32 vtaps; /* vertical scaler taps */
8128 * dce8_dram_bandwidth - get the dram bandwidth
8130 * @wm: watermark calculation data
8132 * Calculate the raw dram bandwidth (CIK).
8133 * Used for display watermark bandwidth calculations
8134 * Returns the dram bandwidth in MBytes/s
8136 static u32 dce8_dram_bandwidth(struct dce8_wm_params *wm)
8138 /* Calculate raw DRAM Bandwidth */
8139 fixed20_12 dram_efficiency; /* 0.7 */
8140 fixed20_12 yclk, dram_channels, bandwidth;
8143 a.full = dfixed_const(1000);
8144 yclk.full = dfixed_const(wm->yclk);
8145 yclk.full = dfixed_div(yclk, a);
8146 dram_channels.full = dfixed_const(wm->dram_channels * 4);
8147 a.full = dfixed_const(10);
8148 dram_efficiency.full = dfixed_const(7);
8149 dram_efficiency.full = dfixed_div(dram_efficiency, a);
8150 bandwidth.full = dfixed_mul(dram_channels, yclk);
8151 bandwidth.full = dfixed_mul(bandwidth, dram_efficiency);
8153 return dfixed_trunc(bandwidth);
8157 * dce8_dram_bandwidth_for_display - get the dram bandwidth for display
8159 * @wm: watermark calculation data
8161 * Calculate the dram bandwidth used for display (CIK).
8162 * Used for display watermark bandwidth calculations
8163 * Returns the dram bandwidth for display in MBytes/s
8165 static u32 dce8_dram_bandwidth_for_display(struct dce8_wm_params *wm)
8167 /* Calculate DRAM Bandwidth and the part allocated to display. */
8168 fixed20_12 disp_dram_allocation; /* 0.3 to 0.7 */
8169 fixed20_12 yclk, dram_channels, bandwidth;
8172 a.full = dfixed_const(1000);
8173 yclk.full = dfixed_const(wm->yclk);
8174 yclk.full = dfixed_div(yclk, a);
8175 dram_channels.full = dfixed_const(wm->dram_channels * 4);
8176 a.full = dfixed_const(10);
8177 disp_dram_allocation.full = dfixed_const(3); /* XXX worse case value 0.3 */
8178 disp_dram_allocation.full = dfixed_div(disp_dram_allocation, a);
8179 bandwidth.full = dfixed_mul(dram_channels, yclk);
8180 bandwidth.full = dfixed_mul(bandwidth, disp_dram_allocation);
8182 return dfixed_trunc(bandwidth);
8186 * dce8_data_return_bandwidth - get the data return bandwidth
8188 * @wm: watermark calculation data
8190 * Calculate the data return bandwidth used for display (CIK).
8191 * Used for display watermark bandwidth calculations
8192 * Returns the data return bandwidth in MBytes/s
8194 static u32 dce8_data_return_bandwidth(struct dce8_wm_params *wm)
8196 /* Calculate the display Data return Bandwidth */
8197 fixed20_12 return_efficiency; /* 0.8 */
8198 fixed20_12 sclk, bandwidth;
8201 a.full = dfixed_const(1000);
8202 sclk.full = dfixed_const(wm->sclk);
8203 sclk.full = dfixed_div(sclk, a);
8204 a.full = dfixed_const(10);
8205 return_efficiency.full = dfixed_const(8);
8206 return_efficiency.full = dfixed_div(return_efficiency, a);
8207 a.full = dfixed_const(32);
8208 bandwidth.full = dfixed_mul(a, sclk);
8209 bandwidth.full = dfixed_mul(bandwidth, return_efficiency);
8211 return dfixed_trunc(bandwidth);
8215 * dce8_dmif_request_bandwidth - get the dmif bandwidth
8217 * @wm: watermark calculation data
8219 * Calculate the dmif bandwidth used for display (CIK).
8220 * Used for display watermark bandwidth calculations
8221 * Returns the dmif bandwidth in MBytes/s
8223 static u32 dce8_dmif_request_bandwidth(struct dce8_wm_params *wm)
8225 /* Calculate the DMIF Request Bandwidth */
8226 fixed20_12 disp_clk_request_efficiency; /* 0.8 */
8227 fixed20_12 disp_clk, bandwidth;
8230 a.full = dfixed_const(1000);
8231 disp_clk.full = dfixed_const(wm->disp_clk);
8232 disp_clk.full = dfixed_div(disp_clk, a);
8233 a.full = dfixed_const(32);
8234 b.full = dfixed_mul(a, disp_clk);
8236 a.full = dfixed_const(10);
8237 disp_clk_request_efficiency.full = dfixed_const(8);
8238 disp_clk_request_efficiency.full = dfixed_div(disp_clk_request_efficiency, a);
8240 bandwidth.full = dfixed_mul(b, disp_clk_request_efficiency);
8242 return dfixed_trunc(bandwidth);
8246 * dce8_available_bandwidth - get the min available bandwidth
8248 * @wm: watermark calculation data
8250 * Calculate the min available bandwidth used for display (CIK).
8251 * Used for display watermark bandwidth calculations
8252 * Returns the min available bandwidth in MBytes/s
8254 static u32 dce8_available_bandwidth(struct dce8_wm_params *wm)
8256 /* Calculate the Available bandwidth. Display can use this temporarily but not in average. */
8257 u32 dram_bandwidth = dce8_dram_bandwidth(wm);
8258 u32 data_return_bandwidth = dce8_data_return_bandwidth(wm);
8259 u32 dmif_req_bandwidth = dce8_dmif_request_bandwidth(wm);
8261 return min(dram_bandwidth, min(data_return_bandwidth, dmif_req_bandwidth));
8265 * dce8_average_bandwidth - get the average available bandwidth
8267 * @wm: watermark calculation data
8269 * Calculate the average available bandwidth used for display (CIK).
8270 * Used for display watermark bandwidth calculations
8271 * Returns the average available bandwidth in MBytes/s
8273 static u32 dce8_average_bandwidth(struct dce8_wm_params *wm)
8275 /* Calculate the display mode Average Bandwidth
8276 * DisplayMode should contain the source and destination dimensions,
8280 fixed20_12 line_time;
8281 fixed20_12 src_width;
8282 fixed20_12 bandwidth;
8285 a.full = dfixed_const(1000);
8286 line_time.full = dfixed_const(wm->active_time + wm->blank_time);
8287 line_time.full = dfixed_div(line_time, a);
8288 bpp.full = dfixed_const(wm->bytes_per_pixel);
8289 src_width.full = dfixed_const(wm->src_width);
8290 bandwidth.full = dfixed_mul(src_width, bpp);
8291 bandwidth.full = dfixed_mul(bandwidth, wm->vsc);
8292 bandwidth.full = dfixed_div(bandwidth, line_time);
8294 return dfixed_trunc(bandwidth);
8298 * dce8_latency_watermark - get the latency watermark
8300 * @wm: watermark calculation data
8302 * Calculate the latency watermark (CIK).
8303 * Used for display watermark bandwidth calculations
8304 * Returns the latency watermark in ns
8306 static u32 dce8_latency_watermark(struct dce8_wm_params *wm)
8308 /* First calculate the latency in ns */
8309 u32 mc_latency = 2000; /* 2000 ns. */
8310 u32 available_bandwidth = dce8_available_bandwidth(wm);
8311 u32 worst_chunk_return_time = (512 * 8 * 1000) / available_bandwidth;
8312 u32 cursor_line_pair_return_time = (128 * 4 * 1000) / available_bandwidth;
8313 u32 dc_latency = 40000000 / wm->disp_clk; /* dc pipe latency */
8314 u32 other_heads_data_return_time = ((wm->num_heads + 1) * worst_chunk_return_time) +
8315 (wm->num_heads * cursor_line_pair_return_time);
8316 u32 latency = mc_latency + other_heads_data_return_time + dc_latency;
8317 u32 max_src_lines_per_dst_line, lb_fill_bw, line_fill_time;
8318 u32 tmp, dmif_size = 12288;
8321 if (wm->num_heads == 0)
8324 a.full = dfixed_const(2);
8325 b.full = dfixed_const(1);
8326 if ((wm->vsc.full > a.full) ||
8327 ((wm->vsc.full > b.full) && (wm->vtaps >= 3)) ||
8329 ((wm->vsc.full >= a.full) && wm->interlaced))
8330 max_src_lines_per_dst_line = 4;
8332 max_src_lines_per_dst_line = 2;
8334 a.full = dfixed_const(available_bandwidth);
8335 b.full = dfixed_const(wm->num_heads);
8336 a.full = dfixed_div(a, b);
8338 b.full = dfixed_const(mc_latency + 512);
8339 c.full = dfixed_const(wm->disp_clk);
8340 b.full = dfixed_div(b, c);
8342 c.full = dfixed_const(dmif_size);
8343 b.full = dfixed_div(c, b);
8345 tmp = min(dfixed_trunc(a), dfixed_trunc(b));
8347 b.full = dfixed_const(1000);
8348 c.full = dfixed_const(wm->disp_clk);
8349 b.full = dfixed_div(c, b);
8350 c.full = dfixed_const(wm->bytes_per_pixel);
8351 b.full = dfixed_mul(b, c);
8353 lb_fill_bw = min(tmp, dfixed_trunc(b));
8355 a.full = dfixed_const(max_src_lines_per_dst_line * wm->src_width * wm->bytes_per_pixel);
8356 b.full = dfixed_const(1000);
8357 c.full = dfixed_const(lb_fill_bw);
8358 b.full = dfixed_div(c, b);
8359 a.full = dfixed_div(a, b);
8360 line_fill_time = dfixed_trunc(a);
8362 if (line_fill_time < wm->active_time)
8365 return latency + (line_fill_time - wm->active_time);
8370 * dce8_average_bandwidth_vs_dram_bandwidth_for_display - check
8371 * average and available dram bandwidth
8373 * @wm: watermark calculation data
8375 * Check if the display average bandwidth fits in the display
8376 * dram bandwidth (CIK).
8377 * Used for display watermark bandwidth calculations
8378 * Returns true if the display fits, false if not.
8380 static bool dce8_average_bandwidth_vs_dram_bandwidth_for_display(struct dce8_wm_params *wm)
8382 if (dce8_average_bandwidth(wm) <=
8383 (dce8_dram_bandwidth_for_display(wm) / wm->num_heads))
8390 * dce8_average_bandwidth_vs_available_bandwidth - check
8391 * average and available bandwidth
8393 * @wm: watermark calculation data
8395 * Check if the display average bandwidth fits in the display
8396 * available bandwidth (CIK).
8397 * Used for display watermark bandwidth calculations
8398 * Returns true if the display fits, false if not.
8400 static bool dce8_average_bandwidth_vs_available_bandwidth(struct dce8_wm_params *wm)
8402 if (dce8_average_bandwidth(wm) <=
8403 (dce8_available_bandwidth(wm) / wm->num_heads))
8410 * dce8_check_latency_hiding - check latency hiding
8412 * @wm: watermark calculation data
8414 * Check latency hiding (CIK).
8415 * Used for display watermark bandwidth calculations
8416 * Returns true if the display fits, false if not.
8418 static bool dce8_check_latency_hiding(struct dce8_wm_params *wm)
8420 u32 lb_partitions = wm->lb_size / wm->src_width;
8421 u32 line_time = wm->active_time + wm->blank_time;
8422 u32 latency_tolerant_lines;
8426 a.full = dfixed_const(1);
8427 if (wm->vsc.full > a.full)
8428 latency_tolerant_lines = 1;
8430 if (lb_partitions <= (wm->vtaps + 1))
8431 latency_tolerant_lines = 1;
8433 latency_tolerant_lines = 2;
8436 latency_hiding = (latency_tolerant_lines * line_time + wm->blank_time);
8438 if (dce8_latency_watermark(wm) <= latency_hiding)
8445 * dce8_program_watermarks - program display watermarks
8447 * @rdev: radeon_device pointer
8448 * @radeon_crtc: the selected display controller
8449 * @lb_size: line buffer size
8450 * @num_heads: number of display controllers in use
8452 * Calculate and program the display watermarks for the
8453 * selected display controller (CIK).
8455 static void dce8_program_watermarks(struct radeon_device *rdev,
8456 struct radeon_crtc *radeon_crtc,
8457 u32 lb_size, u32 num_heads)
8459 struct drm_display_mode *mode = &radeon_crtc->base.mode;
8460 struct dce8_wm_params wm_low, wm_high;
8463 u32 latency_watermark_a = 0, latency_watermark_b = 0;
8466 if (radeon_crtc->base.enabled && num_heads && mode) {
8467 pixel_period = 1000000 / (u32)mode->clock;
8468 line_time = min((u32)mode->crtc_htotal * pixel_period, (u32)65535);
8470 /* watermark for high clocks */
8471 if ((rdev->pm.pm_method == PM_METHOD_DPM) &&
8472 rdev->pm.dpm_enabled) {
8474 radeon_dpm_get_mclk(rdev, false) * 10;
8476 radeon_dpm_get_sclk(rdev, false) * 10;
8478 wm_high.yclk = rdev->pm.current_mclk * 10;
8479 wm_high.sclk = rdev->pm.current_sclk * 10;
8482 wm_high.disp_clk = mode->clock;
8483 wm_high.src_width = mode->crtc_hdisplay;
8484 wm_high.active_time = mode->crtc_hdisplay * pixel_period;
8485 wm_high.blank_time = line_time - wm_high.active_time;
8486 wm_high.interlaced = false;
8487 if (mode->flags & DRM_MODE_FLAG_INTERLACE)
8488 wm_high.interlaced = true;
8489 wm_high.vsc = radeon_crtc->vsc;
8491 if (radeon_crtc->rmx_type != RMX_OFF)
8493 wm_high.bytes_per_pixel = 4; /* XXX: get this from fb config */
8494 wm_high.lb_size = lb_size;
8495 wm_high.dram_channels = cik_get_number_of_dram_channels(rdev);
8496 wm_high.num_heads = num_heads;
8498 /* set for high clocks */
8499 latency_watermark_a = min(dce8_latency_watermark(&wm_high), (u32)65535);
8501 /* possibly force display priority to high */
8502 /* should really do this at mode validation time... */
8503 if (!dce8_average_bandwidth_vs_dram_bandwidth_for_display(&wm_high) ||
8504 !dce8_average_bandwidth_vs_available_bandwidth(&wm_high) ||
8505 !dce8_check_latency_hiding(&wm_high) ||
8506 (rdev->disp_priority == 2)) {
8507 DRM_DEBUG_KMS("force priority to high\n");
8510 /* watermark for low clocks */
8511 if ((rdev->pm.pm_method == PM_METHOD_DPM) &&
8512 rdev->pm.dpm_enabled) {
8514 radeon_dpm_get_mclk(rdev, true) * 10;
8516 radeon_dpm_get_sclk(rdev, true) * 10;
8518 wm_low.yclk = rdev->pm.current_mclk * 10;
8519 wm_low.sclk = rdev->pm.current_sclk * 10;
8522 wm_low.disp_clk = mode->clock;
8523 wm_low.src_width = mode->crtc_hdisplay;
8524 wm_low.active_time = mode->crtc_hdisplay * pixel_period;
8525 wm_low.blank_time = line_time - wm_low.active_time;
8526 wm_low.interlaced = false;
8527 if (mode->flags & DRM_MODE_FLAG_INTERLACE)
8528 wm_low.interlaced = true;
8529 wm_low.vsc = radeon_crtc->vsc;
8531 if (radeon_crtc->rmx_type != RMX_OFF)
8533 wm_low.bytes_per_pixel = 4; /* XXX: get this from fb config */
8534 wm_low.lb_size = lb_size;
8535 wm_low.dram_channels = cik_get_number_of_dram_channels(rdev);
8536 wm_low.num_heads = num_heads;
8538 /* set for low clocks */
8539 latency_watermark_b = min(dce8_latency_watermark(&wm_low), (u32)65535);
8541 /* possibly force display priority to high */
8542 /* should really do this at mode validation time... */
8543 if (!dce8_average_bandwidth_vs_dram_bandwidth_for_display(&wm_low) ||
8544 !dce8_average_bandwidth_vs_available_bandwidth(&wm_low) ||
8545 !dce8_check_latency_hiding(&wm_low) ||
8546 (rdev->disp_priority == 2)) {
8547 DRM_DEBUG_KMS("force priority to high\n");
8552 wm_mask = RREG32(DPG_WATERMARK_MASK_CONTROL + radeon_crtc->crtc_offset);
8554 tmp &= ~LATENCY_WATERMARK_MASK(3);
8555 tmp |= LATENCY_WATERMARK_MASK(1);
8556 WREG32(DPG_WATERMARK_MASK_CONTROL + radeon_crtc->crtc_offset, tmp);
8557 WREG32(DPG_PIPE_LATENCY_CONTROL + radeon_crtc->crtc_offset,
8558 (LATENCY_LOW_WATERMARK(latency_watermark_a) |
8559 LATENCY_HIGH_WATERMARK(line_time)));
8561 tmp = RREG32(DPG_WATERMARK_MASK_CONTROL + radeon_crtc->crtc_offset);
8562 tmp &= ~LATENCY_WATERMARK_MASK(3);
8563 tmp |= LATENCY_WATERMARK_MASK(2);
8564 WREG32(DPG_WATERMARK_MASK_CONTROL + radeon_crtc->crtc_offset, tmp);
8565 WREG32(DPG_PIPE_LATENCY_CONTROL + radeon_crtc->crtc_offset,
8566 (LATENCY_LOW_WATERMARK(latency_watermark_b) |
8567 LATENCY_HIGH_WATERMARK(line_time)));
8568 /* restore original selection */
8569 WREG32(DPG_WATERMARK_MASK_CONTROL + radeon_crtc->crtc_offset, wm_mask);
8571 /* save values for DPM */
8572 radeon_crtc->line_time = line_time;
8573 radeon_crtc->wm_high = latency_watermark_a;
8574 radeon_crtc->wm_low = latency_watermark_b;
8578 * dce8_bandwidth_update - program display watermarks
8580 * @rdev: radeon_device pointer
8582 * Calculate and program the display watermarks and line
8583 * buffer allocation (CIK).
8585 void dce8_bandwidth_update(struct radeon_device *rdev)
8587 struct drm_display_mode *mode = NULL;
8588 u32 num_heads = 0, lb_size;
8591 radeon_update_display_priority(rdev);
8593 for (i = 0; i < rdev->num_crtc; i++) {
8594 if (rdev->mode_info.crtcs[i]->base.enabled)
8597 for (i = 0; i < rdev->num_crtc; i++) {
8598 mode = &rdev->mode_info.crtcs[i]->base.mode;
8599 lb_size = dce8_line_buffer_adjust(rdev, rdev->mode_info.crtcs[i], mode);
8600 dce8_program_watermarks(rdev, rdev->mode_info.crtcs[i], lb_size, num_heads);
8605 * cik_get_gpu_clock_counter - return GPU clock counter snapshot
8607 * @rdev: radeon_device pointer
8609 * Fetches a GPU clock counter snapshot (SI).
8610 * Returns the 64 bit clock counter snapshot.
8612 uint64_t cik_get_gpu_clock_counter(struct radeon_device *rdev)
8616 mutex_lock(&rdev->gpu_clock_mutex);
8617 WREG32(RLC_CAPTURE_GPU_CLOCK_COUNT, 1);
8618 clock = (uint64_t)RREG32(RLC_GPU_CLOCK_COUNT_LSB) |
8619 ((uint64_t)RREG32(RLC_GPU_CLOCK_COUNT_MSB) << 32ULL);
8620 mutex_unlock(&rdev->gpu_clock_mutex);
8624 static int cik_set_uvd_clock(struct radeon_device *rdev, u32 clock,
8625 u32 cntl_reg, u32 status_reg)
8628 struct atom_clock_dividers dividers;
8631 r = radeon_atom_get_clock_dividers(rdev, COMPUTE_GPUCLK_INPUT_FLAG_DEFAULT_GPUCLK,
8632 clock, false, ÷rs);
8636 tmp = RREG32_SMC(cntl_reg);
8637 tmp &= ~(DCLK_DIR_CNTL_EN|DCLK_DIVIDER_MASK);
8638 tmp |= dividers.post_divider;
8639 WREG32_SMC(cntl_reg, tmp);
8641 for (i = 0; i < 100; i++) {
8642 if (RREG32_SMC(status_reg) & DCLK_STATUS)
8652 int cik_set_uvd_clocks(struct radeon_device *rdev, u32 vclk, u32 dclk)
8656 r = cik_set_uvd_clock(rdev, vclk, CG_VCLK_CNTL, CG_VCLK_STATUS);
8660 r = cik_set_uvd_clock(rdev, dclk, CG_DCLK_CNTL, CG_DCLK_STATUS);
8664 static void cik_pcie_gen3_enable(struct radeon_device *rdev)
8666 struct pci_dev *root = rdev->pdev->bus->self;
8667 int bridge_pos, gpu_pos;
8668 u32 speed_cntl, mask, current_data_rate;
8672 if (radeon_pcie_gen2 == 0)
8675 if (rdev->flags & RADEON_IS_IGP)
8678 if (!(rdev->flags & RADEON_IS_PCIE))
8681 ret = drm_pcie_get_speed_cap_mask(rdev->ddev, &mask);
8685 if (!(mask & (DRM_PCIE_SPEED_50 | DRM_PCIE_SPEED_80)))
8688 speed_cntl = RREG32_PCIE_PORT(PCIE_LC_SPEED_CNTL);
8689 current_data_rate = (speed_cntl & LC_CURRENT_DATA_RATE_MASK) >>
8690 LC_CURRENT_DATA_RATE_SHIFT;
8691 if (mask & DRM_PCIE_SPEED_80) {
8692 if (current_data_rate == 2) {
8693 DRM_INFO("PCIE gen 3 link speeds already enabled\n");
8696 DRM_INFO("enabling PCIE gen 3 link speeds, disable with radeon.pcie_gen2=0\n");
8697 } else if (mask & DRM_PCIE_SPEED_50) {
8698 if (current_data_rate == 1) {
8699 DRM_INFO("PCIE gen 2 link speeds already enabled\n");
8702 DRM_INFO("enabling PCIE gen 2 link speeds, disable with radeon.pcie_gen2=0\n");
8705 bridge_pos = pci_pcie_cap(root);
8709 gpu_pos = pci_pcie_cap(rdev->pdev);
8713 if (mask & DRM_PCIE_SPEED_80) {
8714 /* re-try equalization if gen3 is not already enabled */
8715 if (current_data_rate != 2) {
8716 u16 bridge_cfg, gpu_cfg;
8717 u16 bridge_cfg2, gpu_cfg2;
8718 u32 max_lw, current_lw, tmp;
8720 pci_read_config_word(root, bridge_pos + PCI_EXP_LNKCTL, &bridge_cfg);
8721 pci_read_config_word(rdev->pdev, gpu_pos + PCI_EXP_LNKCTL, &gpu_cfg);
8723 tmp16 = bridge_cfg | PCI_EXP_LNKCTL_HAWD;
8724 pci_write_config_word(root, bridge_pos + PCI_EXP_LNKCTL, tmp16);
8726 tmp16 = gpu_cfg | PCI_EXP_LNKCTL_HAWD;
8727 pci_write_config_word(rdev->pdev, gpu_pos + PCI_EXP_LNKCTL, tmp16);
8729 tmp = RREG32_PCIE_PORT(PCIE_LC_STATUS1);
8730 max_lw = (tmp & LC_DETECTED_LINK_WIDTH_MASK) >> LC_DETECTED_LINK_WIDTH_SHIFT;
8731 current_lw = (tmp & LC_OPERATING_LINK_WIDTH_MASK) >> LC_OPERATING_LINK_WIDTH_SHIFT;
8733 if (current_lw < max_lw) {
8734 tmp = RREG32_PCIE_PORT(PCIE_LC_LINK_WIDTH_CNTL);
8735 if (tmp & LC_RENEGOTIATION_SUPPORT) {
8736 tmp &= ~(LC_LINK_WIDTH_MASK | LC_UPCONFIGURE_DIS);
8737 tmp |= (max_lw << LC_LINK_WIDTH_SHIFT);
8738 tmp |= LC_UPCONFIGURE_SUPPORT | LC_RENEGOTIATE_EN | LC_RECONFIG_NOW;
8739 WREG32_PCIE_PORT(PCIE_LC_LINK_WIDTH_CNTL, tmp);
8743 for (i = 0; i < 10; i++) {
8745 pci_read_config_word(rdev->pdev, gpu_pos + PCI_EXP_DEVSTA, &tmp16);
8746 if (tmp16 & PCI_EXP_DEVSTA_TRPND)
8749 pci_read_config_word(root, bridge_pos + PCI_EXP_LNKCTL, &bridge_cfg);
8750 pci_read_config_word(rdev->pdev, gpu_pos + PCI_EXP_LNKCTL, &gpu_cfg);
8752 pci_read_config_word(root, bridge_pos + PCI_EXP_LNKCTL2, &bridge_cfg2);
8753 pci_read_config_word(rdev->pdev, gpu_pos + PCI_EXP_LNKCTL2, &gpu_cfg2);
8755 tmp = RREG32_PCIE_PORT(PCIE_LC_CNTL4);
8756 tmp |= LC_SET_QUIESCE;
8757 WREG32_PCIE_PORT(PCIE_LC_CNTL4, tmp);
8759 tmp = RREG32_PCIE_PORT(PCIE_LC_CNTL4);
8761 WREG32_PCIE_PORT(PCIE_LC_CNTL4, tmp);
8766 pci_read_config_word(root, bridge_pos + PCI_EXP_LNKCTL, &tmp16);
8767 tmp16 &= ~PCI_EXP_LNKCTL_HAWD;
8768 tmp16 |= (bridge_cfg & PCI_EXP_LNKCTL_HAWD);
8769 pci_write_config_word(root, bridge_pos + PCI_EXP_LNKCTL, tmp16);
8771 pci_read_config_word(rdev->pdev, gpu_pos + PCI_EXP_LNKCTL, &tmp16);
8772 tmp16 &= ~PCI_EXP_LNKCTL_HAWD;
8773 tmp16 |= (gpu_cfg & PCI_EXP_LNKCTL_HAWD);
8774 pci_write_config_word(rdev->pdev, gpu_pos + PCI_EXP_LNKCTL, tmp16);
8777 pci_read_config_word(root, bridge_pos + PCI_EXP_LNKCTL2, &tmp16);
8778 tmp16 &= ~((1 << 4) | (7 << 9));
8779 tmp16 |= (bridge_cfg2 & ((1 << 4) | (7 << 9)));
8780 pci_write_config_word(root, bridge_pos + PCI_EXP_LNKCTL2, tmp16);
8782 pci_read_config_word(rdev->pdev, gpu_pos + PCI_EXP_LNKCTL2, &tmp16);
8783 tmp16 &= ~((1 << 4) | (7 << 9));
8784 tmp16 |= (gpu_cfg2 & ((1 << 4) | (7 << 9)));
8785 pci_write_config_word(rdev->pdev, gpu_pos + PCI_EXP_LNKCTL2, tmp16);
8787 tmp = RREG32_PCIE_PORT(PCIE_LC_CNTL4);
8788 tmp &= ~LC_SET_QUIESCE;
8789 WREG32_PCIE_PORT(PCIE_LC_CNTL4, tmp);
8794 /* set the link speed */
8795 speed_cntl |= LC_FORCE_EN_SW_SPEED_CHANGE | LC_FORCE_DIS_HW_SPEED_CHANGE;
8796 speed_cntl &= ~LC_FORCE_DIS_SW_SPEED_CHANGE;
8797 WREG32_PCIE_PORT(PCIE_LC_SPEED_CNTL, speed_cntl);
8799 pci_read_config_word(rdev->pdev, gpu_pos + PCI_EXP_LNKCTL2, &tmp16);
8801 if (mask & DRM_PCIE_SPEED_80)
8802 tmp16 |= 3; /* gen3 */
8803 else if (mask & DRM_PCIE_SPEED_50)
8804 tmp16 |= 2; /* gen2 */
8806 tmp16 |= 1; /* gen1 */
8807 pci_write_config_word(rdev->pdev, gpu_pos + PCI_EXP_LNKCTL2, tmp16);
8809 speed_cntl = RREG32_PCIE_PORT(PCIE_LC_SPEED_CNTL);
8810 speed_cntl |= LC_INITIATE_LINK_SPEED_CHANGE;
8811 WREG32_PCIE_PORT(PCIE_LC_SPEED_CNTL, speed_cntl);
8813 for (i = 0; i < rdev->usec_timeout; i++) {
8814 speed_cntl = RREG32_PCIE_PORT(PCIE_LC_SPEED_CNTL);
8815 if ((speed_cntl & LC_INITIATE_LINK_SPEED_CHANGE) == 0)
8821 static void cik_program_aspm(struct radeon_device *rdev)
8824 bool disable_l0s = false, disable_l1 = false, disable_plloff_in_l1 = false;
8825 bool disable_clkreq = false;
8827 if (radeon_aspm == 0)
8830 /* XXX double check IGPs */
8831 if (rdev->flags & RADEON_IS_IGP)
8834 if (!(rdev->flags & RADEON_IS_PCIE))
8837 orig = data = RREG32_PCIE_PORT(PCIE_LC_N_FTS_CNTL);
8838 data &= ~LC_XMIT_N_FTS_MASK;
8839 data |= LC_XMIT_N_FTS(0x24) | LC_XMIT_N_FTS_OVERRIDE_EN;
8841 WREG32_PCIE_PORT(PCIE_LC_N_FTS_CNTL, data);
8843 orig = data = RREG32_PCIE_PORT(PCIE_LC_CNTL3);
8844 data |= LC_GO_TO_RECOVERY;
8846 WREG32_PCIE_PORT(PCIE_LC_CNTL3, data);
8848 orig = data = RREG32_PCIE_PORT(PCIE_P_CNTL);
8849 data |= P_IGNORE_EDB_ERR;
8851 WREG32_PCIE_PORT(PCIE_P_CNTL, data);
8853 orig = data = RREG32_PCIE_PORT(PCIE_LC_CNTL);
8854 data &= ~(LC_L0S_INACTIVITY_MASK | LC_L1_INACTIVITY_MASK);
8855 data |= LC_PMI_TO_L1_DIS;
8857 data |= LC_L0S_INACTIVITY(7);
8860 data |= LC_L1_INACTIVITY(7);
8861 data &= ~LC_PMI_TO_L1_DIS;
8863 WREG32_PCIE_PORT(PCIE_LC_CNTL, data);
8865 if (!disable_plloff_in_l1) {
8866 bool clk_req_support;
8868 orig = data = RREG32_PCIE_PORT(PB0_PIF_PWRDOWN_0);
8869 data &= ~(PLL_POWER_STATE_IN_OFF_0_MASK | PLL_POWER_STATE_IN_TXS2_0_MASK);
8870 data |= PLL_POWER_STATE_IN_OFF_0(7) | PLL_POWER_STATE_IN_TXS2_0(7);
8872 WREG32_PCIE_PORT(PB0_PIF_PWRDOWN_0, data);
8874 orig = data = RREG32_PCIE_PORT(PB0_PIF_PWRDOWN_1);
8875 data &= ~(PLL_POWER_STATE_IN_OFF_1_MASK | PLL_POWER_STATE_IN_TXS2_1_MASK);
8876 data |= PLL_POWER_STATE_IN_OFF_1(7) | PLL_POWER_STATE_IN_TXS2_1(7);
8878 WREG32_PCIE_PORT(PB0_PIF_PWRDOWN_1, data);
8880 orig = data = RREG32_PCIE_PORT(PB1_PIF_PWRDOWN_0);
8881 data &= ~(PLL_POWER_STATE_IN_OFF_0_MASK | PLL_POWER_STATE_IN_TXS2_0_MASK);
8882 data |= PLL_POWER_STATE_IN_OFF_0(7) | PLL_POWER_STATE_IN_TXS2_0(7);
8884 WREG32_PCIE_PORT(PB1_PIF_PWRDOWN_0, data);
8886 orig = data = RREG32_PCIE_PORT(PB1_PIF_PWRDOWN_1);
8887 data &= ~(PLL_POWER_STATE_IN_OFF_1_MASK | PLL_POWER_STATE_IN_TXS2_1_MASK);
8888 data |= PLL_POWER_STATE_IN_OFF_1(7) | PLL_POWER_STATE_IN_TXS2_1(7);
8890 WREG32_PCIE_PORT(PB1_PIF_PWRDOWN_1, data);
8892 orig = data = RREG32_PCIE_PORT(PCIE_LC_LINK_WIDTH_CNTL);
8893 data &= ~LC_DYN_LANES_PWR_STATE_MASK;
8894 data |= LC_DYN_LANES_PWR_STATE(3);
8896 WREG32_PCIE_PORT(PCIE_LC_LINK_WIDTH_CNTL, data);
8898 if (!disable_clkreq) {
8899 struct pci_dev *root = rdev->pdev->bus->self;
8902 clk_req_support = false;
8903 pcie_capability_read_dword(root, PCI_EXP_LNKCAP, &lnkcap);
8904 if (lnkcap & PCI_EXP_LNKCAP_CLKPM)
8905 clk_req_support = true;
8907 clk_req_support = false;
8910 if (clk_req_support) {
8911 orig = data = RREG32_PCIE_PORT(PCIE_LC_CNTL2);
8912 data |= LC_ALLOW_PDWN_IN_L1 | LC_ALLOW_PDWN_IN_L23;
8914 WREG32_PCIE_PORT(PCIE_LC_CNTL2, data);
8916 orig = data = RREG32_SMC(THM_CLK_CNTL);
8917 data &= ~(CMON_CLK_SEL_MASK | TMON_CLK_SEL_MASK);
8918 data |= CMON_CLK_SEL(1) | TMON_CLK_SEL(1);
8920 WREG32_SMC(THM_CLK_CNTL, data);
8922 orig = data = RREG32_SMC(MISC_CLK_CTRL);
8923 data &= ~(DEEP_SLEEP_CLK_SEL_MASK | ZCLK_SEL_MASK);
8924 data |= DEEP_SLEEP_CLK_SEL(1) | ZCLK_SEL(1);
8926 WREG32_SMC(MISC_CLK_CTRL, data);
8928 orig = data = RREG32_SMC(CG_CLKPIN_CNTL);
8929 data &= ~BCLK_AS_XCLK;
8931 WREG32_SMC(CG_CLKPIN_CNTL, data);
8933 orig = data = RREG32_SMC(CG_CLKPIN_CNTL_2);
8934 data &= ~FORCE_BIF_REFCLK_EN;
8936 WREG32_SMC(CG_CLKPIN_CNTL_2, data);
8938 orig = data = RREG32_SMC(MPLL_BYPASSCLK_SEL);
8939 data &= ~MPLL_CLKOUT_SEL_MASK;
8940 data |= MPLL_CLKOUT_SEL(4);
8942 WREG32_SMC(MPLL_BYPASSCLK_SEL, data);
8947 WREG32_PCIE_PORT(PCIE_LC_CNTL, data);
8950 orig = data = RREG32_PCIE_PORT(PCIE_CNTL2);
8951 data |= SLV_MEM_LS_EN | MST_MEM_LS_EN | REPLAY_MEM_LS_EN;
8953 WREG32_PCIE_PORT(PCIE_CNTL2, data);
8956 data = RREG32_PCIE_PORT(PCIE_LC_N_FTS_CNTL);
8957 if((data & LC_N_FTS_MASK) == LC_N_FTS_MASK) {
8958 data = RREG32_PCIE_PORT(PCIE_LC_STATUS1);
8959 if ((data & LC_REVERSE_XMIT) && (data & LC_REVERSE_RCVR)) {
8960 orig = data = RREG32_PCIE_PORT(PCIE_LC_CNTL);
8961 data &= ~LC_L0S_INACTIVITY_MASK;
8963 WREG32_PCIE_PORT(PCIE_LC_CNTL, data);