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"
35 #include "radeon_kfd.h"
37 MODULE_FIRMWARE("radeon/BONAIRE_pfp.bin");
38 MODULE_FIRMWARE("radeon/BONAIRE_me.bin");
39 MODULE_FIRMWARE("radeon/BONAIRE_ce.bin");
40 MODULE_FIRMWARE("radeon/BONAIRE_mec.bin");
41 MODULE_FIRMWARE("radeon/BONAIRE_mc.bin");
42 MODULE_FIRMWARE("radeon/BONAIRE_mc2.bin");
43 MODULE_FIRMWARE("radeon/BONAIRE_rlc.bin");
44 MODULE_FIRMWARE("radeon/BONAIRE_sdma.bin");
45 MODULE_FIRMWARE("radeon/BONAIRE_smc.bin");
47 MODULE_FIRMWARE("radeon/bonaire_pfp.bin");
48 MODULE_FIRMWARE("radeon/bonaire_me.bin");
49 MODULE_FIRMWARE("radeon/bonaire_ce.bin");
50 MODULE_FIRMWARE("radeon/bonaire_mec.bin");
51 MODULE_FIRMWARE("radeon/bonaire_mc.bin");
52 MODULE_FIRMWARE("radeon/bonaire_rlc.bin");
53 MODULE_FIRMWARE("radeon/bonaire_sdma.bin");
54 MODULE_FIRMWARE("radeon/bonaire_smc.bin");
56 MODULE_FIRMWARE("radeon/HAWAII_pfp.bin");
57 MODULE_FIRMWARE("radeon/HAWAII_me.bin");
58 MODULE_FIRMWARE("radeon/HAWAII_ce.bin");
59 MODULE_FIRMWARE("radeon/HAWAII_mec.bin");
60 MODULE_FIRMWARE("radeon/HAWAII_mc.bin");
61 MODULE_FIRMWARE("radeon/HAWAII_mc2.bin");
62 MODULE_FIRMWARE("radeon/HAWAII_rlc.bin");
63 MODULE_FIRMWARE("radeon/HAWAII_sdma.bin");
64 MODULE_FIRMWARE("radeon/HAWAII_smc.bin");
66 MODULE_FIRMWARE("radeon/hawaii_pfp.bin");
67 MODULE_FIRMWARE("radeon/hawaii_me.bin");
68 MODULE_FIRMWARE("radeon/hawaii_ce.bin");
69 MODULE_FIRMWARE("radeon/hawaii_mec.bin");
70 MODULE_FIRMWARE("radeon/hawaii_mc.bin");
71 MODULE_FIRMWARE("radeon/hawaii_rlc.bin");
72 MODULE_FIRMWARE("radeon/hawaii_sdma.bin");
73 MODULE_FIRMWARE("radeon/hawaii_smc.bin");
75 MODULE_FIRMWARE("radeon/KAVERI_pfp.bin");
76 MODULE_FIRMWARE("radeon/KAVERI_me.bin");
77 MODULE_FIRMWARE("radeon/KAVERI_ce.bin");
78 MODULE_FIRMWARE("radeon/KAVERI_mec.bin");
79 MODULE_FIRMWARE("radeon/KAVERI_rlc.bin");
80 MODULE_FIRMWARE("radeon/KAVERI_sdma.bin");
82 MODULE_FIRMWARE("radeon/kaveri_pfp.bin");
83 MODULE_FIRMWARE("radeon/kaveri_me.bin");
84 MODULE_FIRMWARE("radeon/kaveri_ce.bin");
85 MODULE_FIRMWARE("radeon/kaveri_mec.bin");
86 MODULE_FIRMWARE("radeon/kaveri_mec2.bin");
87 MODULE_FIRMWARE("radeon/kaveri_rlc.bin");
88 MODULE_FIRMWARE("radeon/kaveri_sdma.bin");
90 MODULE_FIRMWARE("radeon/KABINI_pfp.bin");
91 MODULE_FIRMWARE("radeon/KABINI_me.bin");
92 MODULE_FIRMWARE("radeon/KABINI_ce.bin");
93 MODULE_FIRMWARE("radeon/KABINI_mec.bin");
94 MODULE_FIRMWARE("radeon/KABINI_rlc.bin");
95 MODULE_FIRMWARE("radeon/KABINI_sdma.bin");
97 MODULE_FIRMWARE("radeon/kabini_pfp.bin");
98 MODULE_FIRMWARE("radeon/kabini_me.bin");
99 MODULE_FIRMWARE("radeon/kabini_ce.bin");
100 MODULE_FIRMWARE("radeon/kabini_mec.bin");
101 MODULE_FIRMWARE("radeon/kabini_rlc.bin");
102 MODULE_FIRMWARE("radeon/kabini_sdma.bin");
104 MODULE_FIRMWARE("radeon/MULLINS_pfp.bin");
105 MODULE_FIRMWARE("radeon/MULLINS_me.bin");
106 MODULE_FIRMWARE("radeon/MULLINS_ce.bin");
107 MODULE_FIRMWARE("radeon/MULLINS_mec.bin");
108 MODULE_FIRMWARE("radeon/MULLINS_rlc.bin");
109 MODULE_FIRMWARE("radeon/MULLINS_sdma.bin");
111 MODULE_FIRMWARE("radeon/mullins_pfp.bin");
112 MODULE_FIRMWARE("radeon/mullins_me.bin");
113 MODULE_FIRMWARE("radeon/mullins_ce.bin");
114 MODULE_FIRMWARE("radeon/mullins_mec.bin");
115 MODULE_FIRMWARE("radeon/mullins_rlc.bin");
116 MODULE_FIRMWARE("radeon/mullins_sdma.bin");
118 extern int r600_ih_ring_alloc(struct radeon_device *rdev);
119 extern void r600_ih_ring_fini(struct radeon_device *rdev);
120 extern void evergreen_mc_stop(struct radeon_device *rdev, struct evergreen_mc_save *save);
121 extern void evergreen_mc_resume(struct radeon_device *rdev, struct evergreen_mc_save *save);
122 extern bool evergreen_is_display_hung(struct radeon_device *rdev);
123 extern void sumo_rlc_fini(struct radeon_device *rdev);
124 extern int sumo_rlc_init(struct radeon_device *rdev);
125 extern void si_vram_gtt_location(struct radeon_device *rdev, struct radeon_mc *mc);
126 extern void si_rlc_reset(struct radeon_device *rdev);
127 extern void si_init_uvd_internal_cg(struct radeon_device *rdev);
128 static u32 cik_get_cu_active_bitmap(struct radeon_device *rdev, u32 se, u32 sh);
129 extern int cik_sdma_resume(struct radeon_device *rdev);
130 extern void cik_sdma_enable(struct radeon_device *rdev, bool enable);
131 extern void cik_sdma_fini(struct radeon_device *rdev);
132 extern void vce_v2_0_enable_mgcg(struct radeon_device *rdev, bool enable);
133 static void cik_rlc_stop(struct radeon_device *rdev);
134 static void cik_pcie_gen3_enable(struct radeon_device *rdev);
135 static void cik_program_aspm(struct radeon_device *rdev);
136 static void cik_init_pg(struct radeon_device *rdev);
137 static void cik_init_cg(struct radeon_device *rdev);
138 static void cik_fini_pg(struct radeon_device *rdev);
139 static void cik_fini_cg(struct radeon_device *rdev);
140 static void cik_enable_gui_idle_interrupt(struct radeon_device *rdev,
143 /* get temperature in millidegrees */
144 int ci_get_temp(struct radeon_device *rdev)
149 temp = (RREG32_SMC(CG_MULT_THERMAL_STATUS) & CTF_TEMP_MASK) >>
155 actual_temp = temp & 0x1ff;
157 actual_temp = actual_temp * 1000;
162 /* get temperature in millidegrees */
163 int kv_get_temp(struct radeon_device *rdev)
168 temp = RREG32_SMC(0xC0300E0C);
171 actual_temp = (temp / 8) - 49;
175 actual_temp = actual_temp * 1000;
181 * Indirect registers accessor
183 u32 cik_pciep_rreg(struct radeon_device *rdev, u32 reg)
188 spin_lock_irqsave(&rdev->pciep_idx_lock, flags);
189 WREG32(PCIE_INDEX, reg);
190 (void)RREG32(PCIE_INDEX);
191 r = RREG32(PCIE_DATA);
192 spin_unlock_irqrestore(&rdev->pciep_idx_lock, flags);
196 void cik_pciep_wreg(struct radeon_device *rdev, u32 reg, u32 v)
200 spin_lock_irqsave(&rdev->pciep_idx_lock, flags);
201 WREG32(PCIE_INDEX, reg);
202 (void)RREG32(PCIE_INDEX);
203 WREG32(PCIE_DATA, v);
204 (void)RREG32(PCIE_DATA);
205 spin_unlock_irqrestore(&rdev->pciep_idx_lock, flags);
208 static const u32 spectre_rlc_save_restore_register_list[] =
210 (0x0e00 << 16) | (0xc12c >> 2),
212 (0x0e00 << 16) | (0xc140 >> 2),
214 (0x0e00 << 16) | (0xc150 >> 2),
216 (0x0e00 << 16) | (0xc15c >> 2),
218 (0x0e00 << 16) | (0xc168 >> 2),
220 (0x0e00 << 16) | (0xc170 >> 2),
222 (0x0e00 << 16) | (0xc178 >> 2),
224 (0x0e00 << 16) | (0xc204 >> 2),
226 (0x0e00 << 16) | (0xc2b4 >> 2),
228 (0x0e00 << 16) | (0xc2b8 >> 2),
230 (0x0e00 << 16) | (0xc2bc >> 2),
232 (0x0e00 << 16) | (0xc2c0 >> 2),
234 (0x0e00 << 16) | (0x8228 >> 2),
236 (0x0e00 << 16) | (0x829c >> 2),
238 (0x0e00 << 16) | (0x869c >> 2),
240 (0x0600 << 16) | (0x98f4 >> 2),
242 (0x0e00 << 16) | (0x98f8 >> 2),
244 (0x0e00 << 16) | (0x9900 >> 2),
246 (0x0e00 << 16) | (0xc260 >> 2),
248 (0x0e00 << 16) | (0x90e8 >> 2),
250 (0x0e00 << 16) | (0x3c000 >> 2),
252 (0x0e00 << 16) | (0x3c00c >> 2),
254 (0x0e00 << 16) | (0x8c1c >> 2),
256 (0x0e00 << 16) | (0x9700 >> 2),
258 (0x0e00 << 16) | (0xcd20 >> 2),
260 (0x4e00 << 16) | (0xcd20 >> 2),
262 (0x5e00 << 16) | (0xcd20 >> 2),
264 (0x6e00 << 16) | (0xcd20 >> 2),
266 (0x7e00 << 16) | (0xcd20 >> 2),
268 (0x8e00 << 16) | (0xcd20 >> 2),
270 (0x9e00 << 16) | (0xcd20 >> 2),
272 (0xae00 << 16) | (0xcd20 >> 2),
274 (0xbe00 << 16) | (0xcd20 >> 2),
276 (0x0e00 << 16) | (0x89bc >> 2),
278 (0x0e00 << 16) | (0x8900 >> 2),
281 (0x0e00 << 16) | (0xc130 >> 2),
283 (0x0e00 << 16) | (0xc134 >> 2),
285 (0x0e00 << 16) | (0xc1fc >> 2),
287 (0x0e00 << 16) | (0xc208 >> 2),
289 (0x0e00 << 16) | (0xc264 >> 2),
291 (0x0e00 << 16) | (0xc268 >> 2),
293 (0x0e00 << 16) | (0xc26c >> 2),
295 (0x0e00 << 16) | (0xc270 >> 2),
297 (0x0e00 << 16) | (0xc274 >> 2),
299 (0x0e00 << 16) | (0xc278 >> 2),
301 (0x0e00 << 16) | (0xc27c >> 2),
303 (0x0e00 << 16) | (0xc280 >> 2),
305 (0x0e00 << 16) | (0xc284 >> 2),
307 (0x0e00 << 16) | (0xc288 >> 2),
309 (0x0e00 << 16) | (0xc28c >> 2),
311 (0x0e00 << 16) | (0xc290 >> 2),
313 (0x0e00 << 16) | (0xc294 >> 2),
315 (0x0e00 << 16) | (0xc298 >> 2),
317 (0x0e00 << 16) | (0xc29c >> 2),
319 (0x0e00 << 16) | (0xc2a0 >> 2),
321 (0x0e00 << 16) | (0xc2a4 >> 2),
323 (0x0e00 << 16) | (0xc2a8 >> 2),
325 (0x0e00 << 16) | (0xc2ac >> 2),
327 (0x0e00 << 16) | (0xc2b0 >> 2),
329 (0x0e00 << 16) | (0x301d0 >> 2),
331 (0x0e00 << 16) | (0x30238 >> 2),
333 (0x0e00 << 16) | (0x30250 >> 2),
335 (0x0e00 << 16) | (0x30254 >> 2),
337 (0x0e00 << 16) | (0x30258 >> 2),
339 (0x0e00 << 16) | (0x3025c >> 2),
341 (0x4e00 << 16) | (0xc900 >> 2),
343 (0x5e00 << 16) | (0xc900 >> 2),
345 (0x6e00 << 16) | (0xc900 >> 2),
347 (0x7e00 << 16) | (0xc900 >> 2),
349 (0x8e00 << 16) | (0xc900 >> 2),
351 (0x9e00 << 16) | (0xc900 >> 2),
353 (0xae00 << 16) | (0xc900 >> 2),
355 (0xbe00 << 16) | (0xc900 >> 2),
357 (0x4e00 << 16) | (0xc904 >> 2),
359 (0x5e00 << 16) | (0xc904 >> 2),
361 (0x6e00 << 16) | (0xc904 >> 2),
363 (0x7e00 << 16) | (0xc904 >> 2),
365 (0x8e00 << 16) | (0xc904 >> 2),
367 (0x9e00 << 16) | (0xc904 >> 2),
369 (0xae00 << 16) | (0xc904 >> 2),
371 (0xbe00 << 16) | (0xc904 >> 2),
373 (0x4e00 << 16) | (0xc908 >> 2),
375 (0x5e00 << 16) | (0xc908 >> 2),
377 (0x6e00 << 16) | (0xc908 >> 2),
379 (0x7e00 << 16) | (0xc908 >> 2),
381 (0x8e00 << 16) | (0xc908 >> 2),
383 (0x9e00 << 16) | (0xc908 >> 2),
385 (0xae00 << 16) | (0xc908 >> 2),
387 (0xbe00 << 16) | (0xc908 >> 2),
389 (0x4e00 << 16) | (0xc90c >> 2),
391 (0x5e00 << 16) | (0xc90c >> 2),
393 (0x6e00 << 16) | (0xc90c >> 2),
395 (0x7e00 << 16) | (0xc90c >> 2),
397 (0x8e00 << 16) | (0xc90c >> 2),
399 (0x9e00 << 16) | (0xc90c >> 2),
401 (0xae00 << 16) | (0xc90c >> 2),
403 (0xbe00 << 16) | (0xc90c >> 2),
405 (0x4e00 << 16) | (0xc910 >> 2),
407 (0x5e00 << 16) | (0xc910 >> 2),
409 (0x6e00 << 16) | (0xc910 >> 2),
411 (0x7e00 << 16) | (0xc910 >> 2),
413 (0x8e00 << 16) | (0xc910 >> 2),
415 (0x9e00 << 16) | (0xc910 >> 2),
417 (0xae00 << 16) | (0xc910 >> 2),
419 (0xbe00 << 16) | (0xc910 >> 2),
421 (0x0e00 << 16) | (0xc99c >> 2),
423 (0x0e00 << 16) | (0x9834 >> 2),
425 (0x0000 << 16) | (0x30f00 >> 2),
427 (0x0001 << 16) | (0x30f00 >> 2),
429 (0x0000 << 16) | (0x30f04 >> 2),
431 (0x0001 << 16) | (0x30f04 >> 2),
433 (0x0000 << 16) | (0x30f08 >> 2),
435 (0x0001 << 16) | (0x30f08 >> 2),
437 (0x0000 << 16) | (0x30f0c >> 2),
439 (0x0001 << 16) | (0x30f0c >> 2),
441 (0x0600 << 16) | (0x9b7c >> 2),
443 (0x0e00 << 16) | (0x8a14 >> 2),
445 (0x0e00 << 16) | (0x8a18 >> 2),
447 (0x0600 << 16) | (0x30a00 >> 2),
449 (0x0e00 << 16) | (0x8bf0 >> 2),
451 (0x0e00 << 16) | (0x8bcc >> 2),
453 (0x0e00 << 16) | (0x8b24 >> 2),
455 (0x0e00 << 16) | (0x30a04 >> 2),
457 (0x0600 << 16) | (0x30a10 >> 2),
459 (0x0600 << 16) | (0x30a14 >> 2),
461 (0x0600 << 16) | (0x30a18 >> 2),
463 (0x0600 << 16) | (0x30a2c >> 2),
465 (0x0e00 << 16) | (0xc700 >> 2),
467 (0x0e00 << 16) | (0xc704 >> 2),
469 (0x0e00 << 16) | (0xc708 >> 2),
471 (0x0e00 << 16) | (0xc768 >> 2),
473 (0x0400 << 16) | (0xc770 >> 2),
475 (0x0400 << 16) | (0xc774 >> 2),
477 (0x0400 << 16) | (0xc778 >> 2),
479 (0x0400 << 16) | (0xc77c >> 2),
481 (0x0400 << 16) | (0xc780 >> 2),
483 (0x0400 << 16) | (0xc784 >> 2),
485 (0x0400 << 16) | (0xc788 >> 2),
487 (0x0400 << 16) | (0xc78c >> 2),
489 (0x0400 << 16) | (0xc798 >> 2),
491 (0x0400 << 16) | (0xc79c >> 2),
493 (0x0400 << 16) | (0xc7a0 >> 2),
495 (0x0400 << 16) | (0xc7a4 >> 2),
497 (0x0400 << 16) | (0xc7a8 >> 2),
499 (0x0400 << 16) | (0xc7ac >> 2),
501 (0x0400 << 16) | (0xc7b0 >> 2),
503 (0x0400 << 16) | (0xc7b4 >> 2),
505 (0x0e00 << 16) | (0x9100 >> 2),
507 (0x0e00 << 16) | (0x3c010 >> 2),
509 (0x0e00 << 16) | (0x92a8 >> 2),
511 (0x0e00 << 16) | (0x92ac >> 2),
513 (0x0e00 << 16) | (0x92b4 >> 2),
515 (0x0e00 << 16) | (0x92b8 >> 2),
517 (0x0e00 << 16) | (0x92bc >> 2),
519 (0x0e00 << 16) | (0x92c0 >> 2),
521 (0x0e00 << 16) | (0x92c4 >> 2),
523 (0x0e00 << 16) | (0x92c8 >> 2),
525 (0x0e00 << 16) | (0x92cc >> 2),
527 (0x0e00 << 16) | (0x92d0 >> 2),
529 (0x0e00 << 16) | (0x8c00 >> 2),
531 (0x0e00 << 16) | (0x8c04 >> 2),
533 (0x0e00 << 16) | (0x8c20 >> 2),
535 (0x0e00 << 16) | (0x8c38 >> 2),
537 (0x0e00 << 16) | (0x8c3c >> 2),
539 (0x0e00 << 16) | (0xae00 >> 2),
541 (0x0e00 << 16) | (0x9604 >> 2),
543 (0x0e00 << 16) | (0xac08 >> 2),
545 (0x0e00 << 16) | (0xac0c >> 2),
547 (0x0e00 << 16) | (0xac10 >> 2),
549 (0x0e00 << 16) | (0xac14 >> 2),
551 (0x0e00 << 16) | (0xac58 >> 2),
553 (0x0e00 << 16) | (0xac68 >> 2),
555 (0x0e00 << 16) | (0xac6c >> 2),
557 (0x0e00 << 16) | (0xac70 >> 2),
559 (0x0e00 << 16) | (0xac74 >> 2),
561 (0x0e00 << 16) | (0xac78 >> 2),
563 (0x0e00 << 16) | (0xac7c >> 2),
565 (0x0e00 << 16) | (0xac80 >> 2),
567 (0x0e00 << 16) | (0xac84 >> 2),
569 (0x0e00 << 16) | (0xac88 >> 2),
571 (0x0e00 << 16) | (0xac8c >> 2),
573 (0x0e00 << 16) | (0x970c >> 2),
575 (0x0e00 << 16) | (0x9714 >> 2),
577 (0x0e00 << 16) | (0x9718 >> 2),
579 (0x0e00 << 16) | (0x971c >> 2),
581 (0x0e00 << 16) | (0x31068 >> 2),
583 (0x4e00 << 16) | (0x31068 >> 2),
585 (0x5e00 << 16) | (0x31068 >> 2),
587 (0x6e00 << 16) | (0x31068 >> 2),
589 (0x7e00 << 16) | (0x31068 >> 2),
591 (0x8e00 << 16) | (0x31068 >> 2),
593 (0x9e00 << 16) | (0x31068 >> 2),
595 (0xae00 << 16) | (0x31068 >> 2),
597 (0xbe00 << 16) | (0x31068 >> 2),
599 (0x0e00 << 16) | (0xcd10 >> 2),
601 (0x0e00 << 16) | (0xcd14 >> 2),
603 (0x0e00 << 16) | (0x88b0 >> 2),
605 (0x0e00 << 16) | (0x88b4 >> 2),
607 (0x0e00 << 16) | (0x88b8 >> 2),
609 (0x0e00 << 16) | (0x88bc >> 2),
611 (0x0400 << 16) | (0x89c0 >> 2),
613 (0x0e00 << 16) | (0x88c4 >> 2),
615 (0x0e00 << 16) | (0x88c8 >> 2),
617 (0x0e00 << 16) | (0x88d0 >> 2),
619 (0x0e00 << 16) | (0x88d4 >> 2),
621 (0x0e00 << 16) | (0x88d8 >> 2),
623 (0x0e00 << 16) | (0x8980 >> 2),
625 (0x0e00 << 16) | (0x30938 >> 2),
627 (0x0e00 << 16) | (0x3093c >> 2),
629 (0x0e00 << 16) | (0x30940 >> 2),
631 (0x0e00 << 16) | (0x89a0 >> 2),
633 (0x0e00 << 16) | (0x30900 >> 2),
635 (0x0e00 << 16) | (0x30904 >> 2),
637 (0x0e00 << 16) | (0x89b4 >> 2),
639 (0x0e00 << 16) | (0x3c210 >> 2),
641 (0x0e00 << 16) | (0x3c214 >> 2),
643 (0x0e00 << 16) | (0x3c218 >> 2),
645 (0x0e00 << 16) | (0x8904 >> 2),
648 (0x0e00 << 16) | (0x8c28 >> 2),
649 (0x0e00 << 16) | (0x8c2c >> 2),
650 (0x0e00 << 16) | (0x8c30 >> 2),
651 (0x0e00 << 16) | (0x8c34 >> 2),
652 (0x0e00 << 16) | (0x9600 >> 2),
655 static const u32 kalindi_rlc_save_restore_register_list[] =
657 (0x0e00 << 16) | (0xc12c >> 2),
659 (0x0e00 << 16) | (0xc140 >> 2),
661 (0x0e00 << 16) | (0xc150 >> 2),
663 (0x0e00 << 16) | (0xc15c >> 2),
665 (0x0e00 << 16) | (0xc168 >> 2),
667 (0x0e00 << 16) | (0xc170 >> 2),
669 (0x0e00 << 16) | (0xc204 >> 2),
671 (0x0e00 << 16) | (0xc2b4 >> 2),
673 (0x0e00 << 16) | (0xc2b8 >> 2),
675 (0x0e00 << 16) | (0xc2bc >> 2),
677 (0x0e00 << 16) | (0xc2c0 >> 2),
679 (0x0e00 << 16) | (0x8228 >> 2),
681 (0x0e00 << 16) | (0x829c >> 2),
683 (0x0e00 << 16) | (0x869c >> 2),
685 (0x0600 << 16) | (0x98f4 >> 2),
687 (0x0e00 << 16) | (0x98f8 >> 2),
689 (0x0e00 << 16) | (0x9900 >> 2),
691 (0x0e00 << 16) | (0xc260 >> 2),
693 (0x0e00 << 16) | (0x90e8 >> 2),
695 (0x0e00 << 16) | (0x3c000 >> 2),
697 (0x0e00 << 16) | (0x3c00c >> 2),
699 (0x0e00 << 16) | (0x8c1c >> 2),
701 (0x0e00 << 16) | (0x9700 >> 2),
703 (0x0e00 << 16) | (0xcd20 >> 2),
705 (0x4e00 << 16) | (0xcd20 >> 2),
707 (0x5e00 << 16) | (0xcd20 >> 2),
709 (0x6e00 << 16) | (0xcd20 >> 2),
711 (0x7e00 << 16) | (0xcd20 >> 2),
713 (0x0e00 << 16) | (0x89bc >> 2),
715 (0x0e00 << 16) | (0x8900 >> 2),
718 (0x0e00 << 16) | (0xc130 >> 2),
720 (0x0e00 << 16) | (0xc134 >> 2),
722 (0x0e00 << 16) | (0xc1fc >> 2),
724 (0x0e00 << 16) | (0xc208 >> 2),
726 (0x0e00 << 16) | (0xc264 >> 2),
728 (0x0e00 << 16) | (0xc268 >> 2),
730 (0x0e00 << 16) | (0xc26c >> 2),
732 (0x0e00 << 16) | (0xc270 >> 2),
734 (0x0e00 << 16) | (0xc274 >> 2),
736 (0x0e00 << 16) | (0xc28c >> 2),
738 (0x0e00 << 16) | (0xc290 >> 2),
740 (0x0e00 << 16) | (0xc294 >> 2),
742 (0x0e00 << 16) | (0xc298 >> 2),
744 (0x0e00 << 16) | (0xc2a0 >> 2),
746 (0x0e00 << 16) | (0xc2a4 >> 2),
748 (0x0e00 << 16) | (0xc2a8 >> 2),
750 (0x0e00 << 16) | (0xc2ac >> 2),
752 (0x0e00 << 16) | (0x301d0 >> 2),
754 (0x0e00 << 16) | (0x30238 >> 2),
756 (0x0e00 << 16) | (0x30250 >> 2),
758 (0x0e00 << 16) | (0x30254 >> 2),
760 (0x0e00 << 16) | (0x30258 >> 2),
762 (0x0e00 << 16) | (0x3025c >> 2),
764 (0x4e00 << 16) | (0xc900 >> 2),
766 (0x5e00 << 16) | (0xc900 >> 2),
768 (0x6e00 << 16) | (0xc900 >> 2),
770 (0x7e00 << 16) | (0xc900 >> 2),
772 (0x4e00 << 16) | (0xc904 >> 2),
774 (0x5e00 << 16) | (0xc904 >> 2),
776 (0x6e00 << 16) | (0xc904 >> 2),
778 (0x7e00 << 16) | (0xc904 >> 2),
780 (0x4e00 << 16) | (0xc908 >> 2),
782 (0x5e00 << 16) | (0xc908 >> 2),
784 (0x6e00 << 16) | (0xc908 >> 2),
786 (0x7e00 << 16) | (0xc908 >> 2),
788 (0x4e00 << 16) | (0xc90c >> 2),
790 (0x5e00 << 16) | (0xc90c >> 2),
792 (0x6e00 << 16) | (0xc90c >> 2),
794 (0x7e00 << 16) | (0xc90c >> 2),
796 (0x4e00 << 16) | (0xc910 >> 2),
798 (0x5e00 << 16) | (0xc910 >> 2),
800 (0x6e00 << 16) | (0xc910 >> 2),
802 (0x7e00 << 16) | (0xc910 >> 2),
804 (0x0e00 << 16) | (0xc99c >> 2),
806 (0x0e00 << 16) | (0x9834 >> 2),
808 (0x0000 << 16) | (0x30f00 >> 2),
810 (0x0000 << 16) | (0x30f04 >> 2),
812 (0x0000 << 16) | (0x30f08 >> 2),
814 (0x0000 << 16) | (0x30f0c >> 2),
816 (0x0600 << 16) | (0x9b7c >> 2),
818 (0x0e00 << 16) | (0x8a14 >> 2),
820 (0x0e00 << 16) | (0x8a18 >> 2),
822 (0x0600 << 16) | (0x30a00 >> 2),
824 (0x0e00 << 16) | (0x8bf0 >> 2),
826 (0x0e00 << 16) | (0x8bcc >> 2),
828 (0x0e00 << 16) | (0x8b24 >> 2),
830 (0x0e00 << 16) | (0x30a04 >> 2),
832 (0x0600 << 16) | (0x30a10 >> 2),
834 (0x0600 << 16) | (0x30a14 >> 2),
836 (0x0600 << 16) | (0x30a18 >> 2),
838 (0x0600 << 16) | (0x30a2c >> 2),
840 (0x0e00 << 16) | (0xc700 >> 2),
842 (0x0e00 << 16) | (0xc704 >> 2),
844 (0x0e00 << 16) | (0xc708 >> 2),
846 (0x0e00 << 16) | (0xc768 >> 2),
848 (0x0400 << 16) | (0xc770 >> 2),
850 (0x0400 << 16) | (0xc774 >> 2),
852 (0x0400 << 16) | (0xc798 >> 2),
854 (0x0400 << 16) | (0xc79c >> 2),
856 (0x0e00 << 16) | (0x9100 >> 2),
858 (0x0e00 << 16) | (0x3c010 >> 2),
860 (0x0e00 << 16) | (0x8c00 >> 2),
862 (0x0e00 << 16) | (0x8c04 >> 2),
864 (0x0e00 << 16) | (0x8c20 >> 2),
866 (0x0e00 << 16) | (0x8c38 >> 2),
868 (0x0e00 << 16) | (0x8c3c >> 2),
870 (0x0e00 << 16) | (0xae00 >> 2),
872 (0x0e00 << 16) | (0x9604 >> 2),
874 (0x0e00 << 16) | (0xac08 >> 2),
876 (0x0e00 << 16) | (0xac0c >> 2),
878 (0x0e00 << 16) | (0xac10 >> 2),
880 (0x0e00 << 16) | (0xac14 >> 2),
882 (0x0e00 << 16) | (0xac58 >> 2),
884 (0x0e00 << 16) | (0xac68 >> 2),
886 (0x0e00 << 16) | (0xac6c >> 2),
888 (0x0e00 << 16) | (0xac70 >> 2),
890 (0x0e00 << 16) | (0xac74 >> 2),
892 (0x0e00 << 16) | (0xac78 >> 2),
894 (0x0e00 << 16) | (0xac7c >> 2),
896 (0x0e00 << 16) | (0xac80 >> 2),
898 (0x0e00 << 16) | (0xac84 >> 2),
900 (0x0e00 << 16) | (0xac88 >> 2),
902 (0x0e00 << 16) | (0xac8c >> 2),
904 (0x0e00 << 16) | (0x970c >> 2),
906 (0x0e00 << 16) | (0x9714 >> 2),
908 (0x0e00 << 16) | (0x9718 >> 2),
910 (0x0e00 << 16) | (0x971c >> 2),
912 (0x0e00 << 16) | (0x31068 >> 2),
914 (0x4e00 << 16) | (0x31068 >> 2),
916 (0x5e00 << 16) | (0x31068 >> 2),
918 (0x6e00 << 16) | (0x31068 >> 2),
920 (0x7e00 << 16) | (0x31068 >> 2),
922 (0x0e00 << 16) | (0xcd10 >> 2),
924 (0x0e00 << 16) | (0xcd14 >> 2),
926 (0x0e00 << 16) | (0x88b0 >> 2),
928 (0x0e00 << 16) | (0x88b4 >> 2),
930 (0x0e00 << 16) | (0x88b8 >> 2),
932 (0x0e00 << 16) | (0x88bc >> 2),
934 (0x0400 << 16) | (0x89c0 >> 2),
936 (0x0e00 << 16) | (0x88c4 >> 2),
938 (0x0e00 << 16) | (0x88c8 >> 2),
940 (0x0e00 << 16) | (0x88d0 >> 2),
942 (0x0e00 << 16) | (0x88d4 >> 2),
944 (0x0e00 << 16) | (0x88d8 >> 2),
946 (0x0e00 << 16) | (0x8980 >> 2),
948 (0x0e00 << 16) | (0x30938 >> 2),
950 (0x0e00 << 16) | (0x3093c >> 2),
952 (0x0e00 << 16) | (0x30940 >> 2),
954 (0x0e00 << 16) | (0x89a0 >> 2),
956 (0x0e00 << 16) | (0x30900 >> 2),
958 (0x0e00 << 16) | (0x30904 >> 2),
960 (0x0e00 << 16) | (0x89b4 >> 2),
962 (0x0e00 << 16) | (0x3e1fc >> 2),
964 (0x0e00 << 16) | (0x3c210 >> 2),
966 (0x0e00 << 16) | (0x3c214 >> 2),
968 (0x0e00 << 16) | (0x3c218 >> 2),
970 (0x0e00 << 16) | (0x8904 >> 2),
973 (0x0e00 << 16) | (0x8c28 >> 2),
974 (0x0e00 << 16) | (0x8c2c >> 2),
975 (0x0e00 << 16) | (0x8c30 >> 2),
976 (0x0e00 << 16) | (0x8c34 >> 2),
977 (0x0e00 << 16) | (0x9600 >> 2),
980 static const u32 bonaire_golden_spm_registers[] =
982 0x30800, 0xe0ffffff, 0xe0000000
985 static const u32 bonaire_golden_common_registers[] =
987 0xc770, 0xffffffff, 0x00000800,
988 0xc774, 0xffffffff, 0x00000800,
989 0xc798, 0xffffffff, 0x00007fbf,
990 0xc79c, 0xffffffff, 0x00007faf
993 static const u32 bonaire_golden_registers[] =
995 0x3354, 0x00000333, 0x00000333,
996 0x3350, 0x000c0fc0, 0x00040200,
997 0x9a10, 0x00010000, 0x00058208,
998 0x3c000, 0xffff1fff, 0x00140000,
999 0x3c200, 0xfdfc0fff, 0x00000100,
1000 0x3c234, 0x40000000, 0x40000200,
1001 0x9830, 0xffffffff, 0x00000000,
1002 0x9834, 0xf00fffff, 0x00000400,
1003 0x9838, 0x0002021c, 0x00020200,
1004 0xc78, 0x00000080, 0x00000000,
1005 0x5bb0, 0x000000f0, 0x00000070,
1006 0x5bc0, 0xf0311fff, 0x80300000,
1007 0x98f8, 0x73773777, 0x12010001,
1008 0x350c, 0x00810000, 0x408af000,
1009 0x7030, 0x31000111, 0x00000011,
1010 0x2f48, 0x73773777, 0x12010001,
1011 0x220c, 0x00007fb6, 0x0021a1b1,
1012 0x2210, 0x00007fb6, 0x002021b1,
1013 0x2180, 0x00007fb6, 0x00002191,
1014 0x2218, 0x00007fb6, 0x002121b1,
1015 0x221c, 0x00007fb6, 0x002021b1,
1016 0x21dc, 0x00007fb6, 0x00002191,
1017 0x21e0, 0x00007fb6, 0x00002191,
1018 0x3628, 0x0000003f, 0x0000000a,
1019 0x362c, 0x0000003f, 0x0000000a,
1020 0x2ae4, 0x00073ffe, 0x000022a2,
1021 0x240c, 0x000007ff, 0x00000000,
1022 0x8a14, 0xf000003f, 0x00000007,
1023 0x8bf0, 0x00002001, 0x00000001,
1024 0x8b24, 0xffffffff, 0x00ffffff,
1025 0x30a04, 0x0000ff0f, 0x00000000,
1026 0x28a4c, 0x07ffffff, 0x06000000,
1027 0x4d8, 0x00000fff, 0x00000100,
1028 0x3e78, 0x00000001, 0x00000002,
1029 0x9100, 0x03000000, 0x0362c688,
1030 0x8c00, 0x000000ff, 0x00000001,
1031 0xe40, 0x00001fff, 0x00001fff,
1032 0x9060, 0x0000007f, 0x00000020,
1033 0x9508, 0x00010000, 0x00010000,
1034 0xac14, 0x000003ff, 0x000000f3,
1035 0xac0c, 0xffffffff, 0x00001032
1038 static const u32 bonaire_mgcg_cgcg_init[] =
1040 0xc420, 0xffffffff, 0xfffffffc,
1041 0x30800, 0xffffffff, 0xe0000000,
1042 0x3c2a0, 0xffffffff, 0x00000100,
1043 0x3c208, 0xffffffff, 0x00000100,
1044 0x3c2c0, 0xffffffff, 0xc0000100,
1045 0x3c2c8, 0xffffffff, 0xc0000100,
1046 0x3c2c4, 0xffffffff, 0xc0000100,
1047 0x55e4, 0xffffffff, 0x00600100,
1048 0x3c280, 0xffffffff, 0x00000100,
1049 0x3c214, 0xffffffff, 0x06000100,
1050 0x3c220, 0xffffffff, 0x00000100,
1051 0x3c218, 0xffffffff, 0x06000100,
1052 0x3c204, 0xffffffff, 0x00000100,
1053 0x3c2e0, 0xffffffff, 0x00000100,
1054 0x3c224, 0xffffffff, 0x00000100,
1055 0x3c200, 0xffffffff, 0x00000100,
1056 0x3c230, 0xffffffff, 0x00000100,
1057 0x3c234, 0xffffffff, 0x00000100,
1058 0x3c250, 0xffffffff, 0x00000100,
1059 0x3c254, 0xffffffff, 0x00000100,
1060 0x3c258, 0xffffffff, 0x00000100,
1061 0x3c25c, 0xffffffff, 0x00000100,
1062 0x3c260, 0xffffffff, 0x00000100,
1063 0x3c27c, 0xffffffff, 0x00000100,
1064 0x3c278, 0xffffffff, 0x00000100,
1065 0x3c210, 0xffffffff, 0x06000100,
1066 0x3c290, 0xffffffff, 0x00000100,
1067 0x3c274, 0xffffffff, 0x00000100,
1068 0x3c2b4, 0xffffffff, 0x00000100,
1069 0x3c2b0, 0xffffffff, 0x00000100,
1070 0x3c270, 0xffffffff, 0x00000100,
1071 0x30800, 0xffffffff, 0xe0000000,
1072 0x3c020, 0xffffffff, 0x00010000,
1073 0x3c024, 0xffffffff, 0x00030002,
1074 0x3c028, 0xffffffff, 0x00040007,
1075 0x3c02c, 0xffffffff, 0x00060005,
1076 0x3c030, 0xffffffff, 0x00090008,
1077 0x3c034, 0xffffffff, 0x00010000,
1078 0x3c038, 0xffffffff, 0x00030002,
1079 0x3c03c, 0xffffffff, 0x00040007,
1080 0x3c040, 0xffffffff, 0x00060005,
1081 0x3c044, 0xffffffff, 0x00090008,
1082 0x3c048, 0xffffffff, 0x00010000,
1083 0x3c04c, 0xffffffff, 0x00030002,
1084 0x3c050, 0xffffffff, 0x00040007,
1085 0x3c054, 0xffffffff, 0x00060005,
1086 0x3c058, 0xffffffff, 0x00090008,
1087 0x3c05c, 0xffffffff, 0x00010000,
1088 0x3c060, 0xffffffff, 0x00030002,
1089 0x3c064, 0xffffffff, 0x00040007,
1090 0x3c068, 0xffffffff, 0x00060005,
1091 0x3c06c, 0xffffffff, 0x00090008,
1092 0x3c070, 0xffffffff, 0x00010000,
1093 0x3c074, 0xffffffff, 0x00030002,
1094 0x3c078, 0xffffffff, 0x00040007,
1095 0x3c07c, 0xffffffff, 0x00060005,
1096 0x3c080, 0xffffffff, 0x00090008,
1097 0x3c084, 0xffffffff, 0x00010000,
1098 0x3c088, 0xffffffff, 0x00030002,
1099 0x3c08c, 0xffffffff, 0x00040007,
1100 0x3c090, 0xffffffff, 0x00060005,
1101 0x3c094, 0xffffffff, 0x00090008,
1102 0x3c098, 0xffffffff, 0x00010000,
1103 0x3c09c, 0xffffffff, 0x00030002,
1104 0x3c0a0, 0xffffffff, 0x00040007,
1105 0x3c0a4, 0xffffffff, 0x00060005,
1106 0x3c0a8, 0xffffffff, 0x00090008,
1107 0x3c000, 0xffffffff, 0x96e00200,
1108 0x8708, 0xffffffff, 0x00900100,
1109 0xc424, 0xffffffff, 0x0020003f,
1110 0x38, 0xffffffff, 0x0140001c,
1111 0x3c, 0x000f0000, 0x000f0000,
1112 0x220, 0xffffffff, 0xC060000C,
1113 0x224, 0xc0000fff, 0x00000100,
1114 0xf90, 0xffffffff, 0x00000100,
1115 0xf98, 0x00000101, 0x00000000,
1116 0x20a8, 0xffffffff, 0x00000104,
1117 0x55e4, 0xff000fff, 0x00000100,
1118 0x30cc, 0xc0000fff, 0x00000104,
1119 0xc1e4, 0x00000001, 0x00000001,
1120 0xd00c, 0xff000ff0, 0x00000100,
1121 0xd80c, 0xff000ff0, 0x00000100
1124 static const u32 spectre_golden_spm_registers[] =
1126 0x30800, 0xe0ffffff, 0xe0000000
1129 static const u32 spectre_golden_common_registers[] =
1131 0xc770, 0xffffffff, 0x00000800,
1132 0xc774, 0xffffffff, 0x00000800,
1133 0xc798, 0xffffffff, 0x00007fbf,
1134 0xc79c, 0xffffffff, 0x00007faf
1137 static const u32 spectre_golden_registers[] =
1139 0x3c000, 0xffff1fff, 0x96940200,
1140 0x3c00c, 0xffff0001, 0xff000000,
1141 0x3c200, 0xfffc0fff, 0x00000100,
1142 0x6ed8, 0x00010101, 0x00010000,
1143 0x9834, 0xf00fffff, 0x00000400,
1144 0x9838, 0xfffffffc, 0x00020200,
1145 0x5bb0, 0x000000f0, 0x00000070,
1146 0x5bc0, 0xf0311fff, 0x80300000,
1147 0x98f8, 0x73773777, 0x12010001,
1148 0x9b7c, 0x00ff0000, 0x00fc0000,
1149 0x2f48, 0x73773777, 0x12010001,
1150 0x8a14, 0xf000003f, 0x00000007,
1151 0x8b24, 0xffffffff, 0x00ffffff,
1152 0x28350, 0x3f3f3fff, 0x00000082,
1153 0x28354, 0x0000003f, 0x00000000,
1154 0x3e78, 0x00000001, 0x00000002,
1155 0x913c, 0xffff03df, 0x00000004,
1156 0xc768, 0x00000008, 0x00000008,
1157 0x8c00, 0x000008ff, 0x00000800,
1158 0x9508, 0x00010000, 0x00010000,
1159 0xac0c, 0xffffffff, 0x54763210,
1160 0x214f8, 0x01ff01ff, 0x00000002,
1161 0x21498, 0x007ff800, 0x00200000,
1162 0x2015c, 0xffffffff, 0x00000f40,
1163 0x30934, 0xffffffff, 0x00000001
1166 static const u32 spectre_mgcg_cgcg_init[] =
1168 0xc420, 0xffffffff, 0xfffffffc,
1169 0x30800, 0xffffffff, 0xe0000000,
1170 0x3c2a0, 0xffffffff, 0x00000100,
1171 0x3c208, 0xffffffff, 0x00000100,
1172 0x3c2c0, 0xffffffff, 0x00000100,
1173 0x3c2c8, 0xffffffff, 0x00000100,
1174 0x3c2c4, 0xffffffff, 0x00000100,
1175 0x55e4, 0xffffffff, 0x00600100,
1176 0x3c280, 0xffffffff, 0x00000100,
1177 0x3c214, 0xffffffff, 0x06000100,
1178 0x3c220, 0xffffffff, 0x00000100,
1179 0x3c218, 0xffffffff, 0x06000100,
1180 0x3c204, 0xffffffff, 0x00000100,
1181 0x3c2e0, 0xffffffff, 0x00000100,
1182 0x3c224, 0xffffffff, 0x00000100,
1183 0x3c200, 0xffffffff, 0x00000100,
1184 0x3c230, 0xffffffff, 0x00000100,
1185 0x3c234, 0xffffffff, 0x00000100,
1186 0x3c250, 0xffffffff, 0x00000100,
1187 0x3c254, 0xffffffff, 0x00000100,
1188 0x3c258, 0xffffffff, 0x00000100,
1189 0x3c25c, 0xffffffff, 0x00000100,
1190 0x3c260, 0xffffffff, 0x00000100,
1191 0x3c27c, 0xffffffff, 0x00000100,
1192 0x3c278, 0xffffffff, 0x00000100,
1193 0x3c210, 0xffffffff, 0x06000100,
1194 0x3c290, 0xffffffff, 0x00000100,
1195 0x3c274, 0xffffffff, 0x00000100,
1196 0x3c2b4, 0xffffffff, 0x00000100,
1197 0x3c2b0, 0xffffffff, 0x00000100,
1198 0x3c270, 0xffffffff, 0x00000100,
1199 0x30800, 0xffffffff, 0xe0000000,
1200 0x3c020, 0xffffffff, 0x00010000,
1201 0x3c024, 0xffffffff, 0x00030002,
1202 0x3c028, 0xffffffff, 0x00040007,
1203 0x3c02c, 0xffffffff, 0x00060005,
1204 0x3c030, 0xffffffff, 0x00090008,
1205 0x3c034, 0xffffffff, 0x00010000,
1206 0x3c038, 0xffffffff, 0x00030002,
1207 0x3c03c, 0xffffffff, 0x00040007,
1208 0x3c040, 0xffffffff, 0x00060005,
1209 0x3c044, 0xffffffff, 0x00090008,
1210 0x3c048, 0xffffffff, 0x00010000,
1211 0x3c04c, 0xffffffff, 0x00030002,
1212 0x3c050, 0xffffffff, 0x00040007,
1213 0x3c054, 0xffffffff, 0x00060005,
1214 0x3c058, 0xffffffff, 0x00090008,
1215 0x3c05c, 0xffffffff, 0x00010000,
1216 0x3c060, 0xffffffff, 0x00030002,
1217 0x3c064, 0xffffffff, 0x00040007,
1218 0x3c068, 0xffffffff, 0x00060005,
1219 0x3c06c, 0xffffffff, 0x00090008,
1220 0x3c070, 0xffffffff, 0x00010000,
1221 0x3c074, 0xffffffff, 0x00030002,
1222 0x3c078, 0xffffffff, 0x00040007,
1223 0x3c07c, 0xffffffff, 0x00060005,
1224 0x3c080, 0xffffffff, 0x00090008,
1225 0x3c084, 0xffffffff, 0x00010000,
1226 0x3c088, 0xffffffff, 0x00030002,
1227 0x3c08c, 0xffffffff, 0x00040007,
1228 0x3c090, 0xffffffff, 0x00060005,
1229 0x3c094, 0xffffffff, 0x00090008,
1230 0x3c098, 0xffffffff, 0x00010000,
1231 0x3c09c, 0xffffffff, 0x00030002,
1232 0x3c0a0, 0xffffffff, 0x00040007,
1233 0x3c0a4, 0xffffffff, 0x00060005,
1234 0x3c0a8, 0xffffffff, 0x00090008,
1235 0x3c0ac, 0xffffffff, 0x00010000,
1236 0x3c0b0, 0xffffffff, 0x00030002,
1237 0x3c0b4, 0xffffffff, 0x00040007,
1238 0x3c0b8, 0xffffffff, 0x00060005,
1239 0x3c0bc, 0xffffffff, 0x00090008,
1240 0x3c000, 0xffffffff, 0x96e00200,
1241 0x8708, 0xffffffff, 0x00900100,
1242 0xc424, 0xffffffff, 0x0020003f,
1243 0x38, 0xffffffff, 0x0140001c,
1244 0x3c, 0x000f0000, 0x000f0000,
1245 0x220, 0xffffffff, 0xC060000C,
1246 0x224, 0xc0000fff, 0x00000100,
1247 0xf90, 0xffffffff, 0x00000100,
1248 0xf98, 0x00000101, 0x00000000,
1249 0x20a8, 0xffffffff, 0x00000104,
1250 0x55e4, 0xff000fff, 0x00000100,
1251 0x30cc, 0xc0000fff, 0x00000104,
1252 0xc1e4, 0x00000001, 0x00000001,
1253 0xd00c, 0xff000ff0, 0x00000100,
1254 0xd80c, 0xff000ff0, 0x00000100
1257 static const u32 kalindi_golden_spm_registers[] =
1259 0x30800, 0xe0ffffff, 0xe0000000
1262 static const u32 kalindi_golden_common_registers[] =
1264 0xc770, 0xffffffff, 0x00000800,
1265 0xc774, 0xffffffff, 0x00000800,
1266 0xc798, 0xffffffff, 0x00007fbf,
1267 0xc79c, 0xffffffff, 0x00007faf
1270 static const u32 kalindi_golden_registers[] =
1272 0x3c000, 0xffffdfff, 0x6e944040,
1273 0x55e4, 0xff607fff, 0xfc000100,
1274 0x3c220, 0xff000fff, 0x00000100,
1275 0x3c224, 0xff000fff, 0x00000100,
1276 0x3c200, 0xfffc0fff, 0x00000100,
1277 0x6ed8, 0x00010101, 0x00010000,
1278 0x9830, 0xffffffff, 0x00000000,
1279 0x9834, 0xf00fffff, 0x00000400,
1280 0x5bb0, 0x000000f0, 0x00000070,
1281 0x5bc0, 0xf0311fff, 0x80300000,
1282 0x98f8, 0x73773777, 0x12010001,
1283 0x98fc, 0xffffffff, 0x00000010,
1284 0x9b7c, 0x00ff0000, 0x00fc0000,
1285 0x8030, 0x00001f0f, 0x0000100a,
1286 0x2f48, 0x73773777, 0x12010001,
1287 0x2408, 0x000fffff, 0x000c007f,
1288 0x8a14, 0xf000003f, 0x00000007,
1289 0x8b24, 0x3fff3fff, 0x00ffcfff,
1290 0x30a04, 0x0000ff0f, 0x00000000,
1291 0x28a4c, 0x07ffffff, 0x06000000,
1292 0x4d8, 0x00000fff, 0x00000100,
1293 0x3e78, 0x00000001, 0x00000002,
1294 0xc768, 0x00000008, 0x00000008,
1295 0x8c00, 0x000000ff, 0x00000003,
1296 0x214f8, 0x01ff01ff, 0x00000002,
1297 0x21498, 0x007ff800, 0x00200000,
1298 0x2015c, 0xffffffff, 0x00000f40,
1299 0x88c4, 0x001f3ae3, 0x00000082,
1300 0x88d4, 0x0000001f, 0x00000010,
1301 0x30934, 0xffffffff, 0x00000000
1304 static const u32 kalindi_mgcg_cgcg_init[] =
1306 0xc420, 0xffffffff, 0xfffffffc,
1307 0x30800, 0xffffffff, 0xe0000000,
1308 0x3c2a0, 0xffffffff, 0x00000100,
1309 0x3c208, 0xffffffff, 0x00000100,
1310 0x3c2c0, 0xffffffff, 0x00000100,
1311 0x3c2c8, 0xffffffff, 0x00000100,
1312 0x3c2c4, 0xffffffff, 0x00000100,
1313 0x55e4, 0xffffffff, 0x00600100,
1314 0x3c280, 0xffffffff, 0x00000100,
1315 0x3c214, 0xffffffff, 0x06000100,
1316 0x3c220, 0xffffffff, 0x00000100,
1317 0x3c218, 0xffffffff, 0x06000100,
1318 0x3c204, 0xffffffff, 0x00000100,
1319 0x3c2e0, 0xffffffff, 0x00000100,
1320 0x3c224, 0xffffffff, 0x00000100,
1321 0x3c200, 0xffffffff, 0x00000100,
1322 0x3c230, 0xffffffff, 0x00000100,
1323 0x3c234, 0xffffffff, 0x00000100,
1324 0x3c250, 0xffffffff, 0x00000100,
1325 0x3c254, 0xffffffff, 0x00000100,
1326 0x3c258, 0xffffffff, 0x00000100,
1327 0x3c25c, 0xffffffff, 0x00000100,
1328 0x3c260, 0xffffffff, 0x00000100,
1329 0x3c27c, 0xffffffff, 0x00000100,
1330 0x3c278, 0xffffffff, 0x00000100,
1331 0x3c210, 0xffffffff, 0x06000100,
1332 0x3c290, 0xffffffff, 0x00000100,
1333 0x3c274, 0xffffffff, 0x00000100,
1334 0x3c2b4, 0xffffffff, 0x00000100,
1335 0x3c2b0, 0xffffffff, 0x00000100,
1336 0x3c270, 0xffffffff, 0x00000100,
1337 0x30800, 0xffffffff, 0xe0000000,
1338 0x3c020, 0xffffffff, 0x00010000,
1339 0x3c024, 0xffffffff, 0x00030002,
1340 0x3c028, 0xffffffff, 0x00040007,
1341 0x3c02c, 0xffffffff, 0x00060005,
1342 0x3c030, 0xffffffff, 0x00090008,
1343 0x3c034, 0xffffffff, 0x00010000,
1344 0x3c038, 0xffffffff, 0x00030002,
1345 0x3c03c, 0xffffffff, 0x00040007,
1346 0x3c040, 0xffffffff, 0x00060005,
1347 0x3c044, 0xffffffff, 0x00090008,
1348 0x3c000, 0xffffffff, 0x96e00200,
1349 0x8708, 0xffffffff, 0x00900100,
1350 0xc424, 0xffffffff, 0x0020003f,
1351 0x38, 0xffffffff, 0x0140001c,
1352 0x3c, 0x000f0000, 0x000f0000,
1353 0x220, 0xffffffff, 0xC060000C,
1354 0x224, 0xc0000fff, 0x00000100,
1355 0x20a8, 0xffffffff, 0x00000104,
1356 0x55e4, 0xff000fff, 0x00000100,
1357 0x30cc, 0xc0000fff, 0x00000104,
1358 0xc1e4, 0x00000001, 0x00000001,
1359 0xd00c, 0xff000ff0, 0x00000100,
1360 0xd80c, 0xff000ff0, 0x00000100
1363 static const u32 hawaii_golden_spm_registers[] =
1365 0x30800, 0xe0ffffff, 0xe0000000
1368 static const u32 hawaii_golden_common_registers[] =
1370 0x30800, 0xffffffff, 0xe0000000,
1371 0x28350, 0xffffffff, 0x3a00161a,
1372 0x28354, 0xffffffff, 0x0000002e,
1373 0x9a10, 0xffffffff, 0x00018208,
1374 0x98f8, 0xffffffff, 0x12011003
1377 static const u32 hawaii_golden_registers[] =
1379 0x3354, 0x00000333, 0x00000333,
1380 0x9a10, 0x00010000, 0x00058208,
1381 0x9830, 0xffffffff, 0x00000000,
1382 0x9834, 0xf00fffff, 0x00000400,
1383 0x9838, 0x0002021c, 0x00020200,
1384 0xc78, 0x00000080, 0x00000000,
1385 0x5bb0, 0x000000f0, 0x00000070,
1386 0x5bc0, 0xf0311fff, 0x80300000,
1387 0x350c, 0x00810000, 0x408af000,
1388 0x7030, 0x31000111, 0x00000011,
1389 0x2f48, 0x73773777, 0x12010001,
1390 0x2120, 0x0000007f, 0x0000001b,
1391 0x21dc, 0x00007fb6, 0x00002191,
1392 0x3628, 0x0000003f, 0x0000000a,
1393 0x362c, 0x0000003f, 0x0000000a,
1394 0x2ae4, 0x00073ffe, 0x000022a2,
1395 0x240c, 0x000007ff, 0x00000000,
1396 0x8bf0, 0x00002001, 0x00000001,
1397 0x8b24, 0xffffffff, 0x00ffffff,
1398 0x30a04, 0x0000ff0f, 0x00000000,
1399 0x28a4c, 0x07ffffff, 0x06000000,
1400 0x3e78, 0x00000001, 0x00000002,
1401 0xc768, 0x00000008, 0x00000008,
1402 0xc770, 0x00000f00, 0x00000800,
1403 0xc774, 0x00000f00, 0x00000800,
1404 0xc798, 0x00ffffff, 0x00ff7fbf,
1405 0xc79c, 0x00ffffff, 0x00ff7faf,
1406 0x8c00, 0x000000ff, 0x00000800,
1407 0xe40, 0x00001fff, 0x00001fff,
1408 0x9060, 0x0000007f, 0x00000020,
1409 0x9508, 0x00010000, 0x00010000,
1410 0xae00, 0x00100000, 0x000ff07c,
1411 0xac14, 0x000003ff, 0x0000000f,
1412 0xac10, 0xffffffff, 0x7564fdec,
1413 0xac0c, 0xffffffff, 0x3120b9a8,
1414 0xac08, 0x20000000, 0x0f9c0000
1417 static const u32 hawaii_mgcg_cgcg_init[] =
1419 0xc420, 0xffffffff, 0xfffffffd,
1420 0x30800, 0xffffffff, 0xe0000000,
1421 0x3c2a0, 0xffffffff, 0x00000100,
1422 0x3c208, 0xffffffff, 0x00000100,
1423 0x3c2c0, 0xffffffff, 0x00000100,
1424 0x3c2c8, 0xffffffff, 0x00000100,
1425 0x3c2c4, 0xffffffff, 0x00000100,
1426 0x55e4, 0xffffffff, 0x00200100,
1427 0x3c280, 0xffffffff, 0x00000100,
1428 0x3c214, 0xffffffff, 0x06000100,
1429 0x3c220, 0xffffffff, 0x00000100,
1430 0x3c218, 0xffffffff, 0x06000100,
1431 0x3c204, 0xffffffff, 0x00000100,
1432 0x3c2e0, 0xffffffff, 0x00000100,
1433 0x3c224, 0xffffffff, 0x00000100,
1434 0x3c200, 0xffffffff, 0x00000100,
1435 0x3c230, 0xffffffff, 0x00000100,
1436 0x3c234, 0xffffffff, 0x00000100,
1437 0x3c250, 0xffffffff, 0x00000100,
1438 0x3c254, 0xffffffff, 0x00000100,
1439 0x3c258, 0xffffffff, 0x00000100,
1440 0x3c25c, 0xffffffff, 0x00000100,
1441 0x3c260, 0xffffffff, 0x00000100,
1442 0x3c27c, 0xffffffff, 0x00000100,
1443 0x3c278, 0xffffffff, 0x00000100,
1444 0x3c210, 0xffffffff, 0x06000100,
1445 0x3c290, 0xffffffff, 0x00000100,
1446 0x3c274, 0xffffffff, 0x00000100,
1447 0x3c2b4, 0xffffffff, 0x00000100,
1448 0x3c2b0, 0xffffffff, 0x00000100,
1449 0x3c270, 0xffffffff, 0x00000100,
1450 0x30800, 0xffffffff, 0xe0000000,
1451 0x3c020, 0xffffffff, 0x00010000,
1452 0x3c024, 0xffffffff, 0x00030002,
1453 0x3c028, 0xffffffff, 0x00040007,
1454 0x3c02c, 0xffffffff, 0x00060005,
1455 0x3c030, 0xffffffff, 0x00090008,
1456 0x3c034, 0xffffffff, 0x00010000,
1457 0x3c038, 0xffffffff, 0x00030002,
1458 0x3c03c, 0xffffffff, 0x00040007,
1459 0x3c040, 0xffffffff, 0x00060005,
1460 0x3c044, 0xffffffff, 0x00090008,
1461 0x3c048, 0xffffffff, 0x00010000,
1462 0x3c04c, 0xffffffff, 0x00030002,
1463 0x3c050, 0xffffffff, 0x00040007,
1464 0x3c054, 0xffffffff, 0x00060005,
1465 0x3c058, 0xffffffff, 0x00090008,
1466 0x3c05c, 0xffffffff, 0x00010000,
1467 0x3c060, 0xffffffff, 0x00030002,
1468 0x3c064, 0xffffffff, 0x00040007,
1469 0x3c068, 0xffffffff, 0x00060005,
1470 0x3c06c, 0xffffffff, 0x00090008,
1471 0x3c070, 0xffffffff, 0x00010000,
1472 0x3c074, 0xffffffff, 0x00030002,
1473 0x3c078, 0xffffffff, 0x00040007,
1474 0x3c07c, 0xffffffff, 0x00060005,
1475 0x3c080, 0xffffffff, 0x00090008,
1476 0x3c084, 0xffffffff, 0x00010000,
1477 0x3c088, 0xffffffff, 0x00030002,
1478 0x3c08c, 0xffffffff, 0x00040007,
1479 0x3c090, 0xffffffff, 0x00060005,
1480 0x3c094, 0xffffffff, 0x00090008,
1481 0x3c098, 0xffffffff, 0x00010000,
1482 0x3c09c, 0xffffffff, 0x00030002,
1483 0x3c0a0, 0xffffffff, 0x00040007,
1484 0x3c0a4, 0xffffffff, 0x00060005,
1485 0x3c0a8, 0xffffffff, 0x00090008,
1486 0x3c0ac, 0xffffffff, 0x00010000,
1487 0x3c0b0, 0xffffffff, 0x00030002,
1488 0x3c0b4, 0xffffffff, 0x00040007,
1489 0x3c0b8, 0xffffffff, 0x00060005,
1490 0x3c0bc, 0xffffffff, 0x00090008,
1491 0x3c0c0, 0xffffffff, 0x00010000,
1492 0x3c0c4, 0xffffffff, 0x00030002,
1493 0x3c0c8, 0xffffffff, 0x00040007,
1494 0x3c0cc, 0xffffffff, 0x00060005,
1495 0x3c0d0, 0xffffffff, 0x00090008,
1496 0x3c0d4, 0xffffffff, 0x00010000,
1497 0x3c0d8, 0xffffffff, 0x00030002,
1498 0x3c0dc, 0xffffffff, 0x00040007,
1499 0x3c0e0, 0xffffffff, 0x00060005,
1500 0x3c0e4, 0xffffffff, 0x00090008,
1501 0x3c0e8, 0xffffffff, 0x00010000,
1502 0x3c0ec, 0xffffffff, 0x00030002,
1503 0x3c0f0, 0xffffffff, 0x00040007,
1504 0x3c0f4, 0xffffffff, 0x00060005,
1505 0x3c0f8, 0xffffffff, 0x00090008,
1506 0xc318, 0xffffffff, 0x00020200,
1507 0x3350, 0xffffffff, 0x00000200,
1508 0x15c0, 0xffffffff, 0x00000400,
1509 0x55e8, 0xffffffff, 0x00000000,
1510 0x2f50, 0xffffffff, 0x00000902,
1511 0x3c000, 0xffffffff, 0x96940200,
1512 0x8708, 0xffffffff, 0x00900100,
1513 0xc424, 0xffffffff, 0x0020003f,
1514 0x38, 0xffffffff, 0x0140001c,
1515 0x3c, 0x000f0000, 0x000f0000,
1516 0x220, 0xffffffff, 0xc060000c,
1517 0x224, 0xc0000fff, 0x00000100,
1518 0xf90, 0xffffffff, 0x00000100,
1519 0xf98, 0x00000101, 0x00000000,
1520 0x20a8, 0xffffffff, 0x00000104,
1521 0x55e4, 0xff000fff, 0x00000100,
1522 0x30cc, 0xc0000fff, 0x00000104,
1523 0xc1e4, 0x00000001, 0x00000001,
1524 0xd00c, 0xff000ff0, 0x00000100,
1525 0xd80c, 0xff000ff0, 0x00000100
1528 static const u32 godavari_golden_registers[] =
1530 0x55e4, 0xff607fff, 0xfc000100,
1531 0x6ed8, 0x00010101, 0x00010000,
1532 0x9830, 0xffffffff, 0x00000000,
1533 0x98302, 0xf00fffff, 0x00000400,
1534 0x6130, 0xffffffff, 0x00010000,
1535 0x5bb0, 0x000000f0, 0x00000070,
1536 0x5bc0, 0xf0311fff, 0x80300000,
1537 0x98f8, 0x73773777, 0x12010001,
1538 0x98fc, 0xffffffff, 0x00000010,
1539 0x8030, 0x00001f0f, 0x0000100a,
1540 0x2f48, 0x73773777, 0x12010001,
1541 0x2408, 0x000fffff, 0x000c007f,
1542 0x8a14, 0xf000003f, 0x00000007,
1543 0x8b24, 0xffffffff, 0x00ff0fff,
1544 0x30a04, 0x0000ff0f, 0x00000000,
1545 0x28a4c, 0x07ffffff, 0x06000000,
1546 0x4d8, 0x00000fff, 0x00000100,
1547 0xd014, 0x00010000, 0x00810001,
1548 0xd814, 0x00010000, 0x00810001,
1549 0x3e78, 0x00000001, 0x00000002,
1550 0xc768, 0x00000008, 0x00000008,
1551 0xc770, 0x00000f00, 0x00000800,
1552 0xc774, 0x00000f00, 0x00000800,
1553 0xc798, 0x00ffffff, 0x00ff7fbf,
1554 0xc79c, 0x00ffffff, 0x00ff7faf,
1555 0x8c00, 0x000000ff, 0x00000001,
1556 0x214f8, 0x01ff01ff, 0x00000002,
1557 0x21498, 0x007ff800, 0x00200000,
1558 0x2015c, 0xffffffff, 0x00000f40,
1559 0x88c4, 0x001f3ae3, 0x00000082,
1560 0x88d4, 0x0000001f, 0x00000010,
1561 0x30934, 0xffffffff, 0x00000000
1565 static void cik_init_golden_registers(struct radeon_device *rdev)
1567 /* Some of the registers might be dependent on GRBM_GFX_INDEX */
1568 mutex_lock(&rdev->grbm_idx_mutex);
1569 switch (rdev->family) {
1571 radeon_program_register_sequence(rdev,
1572 bonaire_mgcg_cgcg_init,
1573 (const u32)ARRAY_SIZE(bonaire_mgcg_cgcg_init));
1574 radeon_program_register_sequence(rdev,
1575 bonaire_golden_registers,
1576 (const u32)ARRAY_SIZE(bonaire_golden_registers));
1577 radeon_program_register_sequence(rdev,
1578 bonaire_golden_common_registers,
1579 (const u32)ARRAY_SIZE(bonaire_golden_common_registers));
1580 radeon_program_register_sequence(rdev,
1581 bonaire_golden_spm_registers,
1582 (const u32)ARRAY_SIZE(bonaire_golden_spm_registers));
1585 radeon_program_register_sequence(rdev,
1586 kalindi_mgcg_cgcg_init,
1587 (const u32)ARRAY_SIZE(kalindi_mgcg_cgcg_init));
1588 radeon_program_register_sequence(rdev,
1589 kalindi_golden_registers,
1590 (const u32)ARRAY_SIZE(kalindi_golden_registers));
1591 radeon_program_register_sequence(rdev,
1592 kalindi_golden_common_registers,
1593 (const u32)ARRAY_SIZE(kalindi_golden_common_registers));
1594 radeon_program_register_sequence(rdev,
1595 kalindi_golden_spm_registers,
1596 (const u32)ARRAY_SIZE(kalindi_golden_spm_registers));
1599 radeon_program_register_sequence(rdev,
1600 kalindi_mgcg_cgcg_init,
1601 (const u32)ARRAY_SIZE(kalindi_mgcg_cgcg_init));
1602 radeon_program_register_sequence(rdev,
1603 godavari_golden_registers,
1604 (const u32)ARRAY_SIZE(godavari_golden_registers));
1605 radeon_program_register_sequence(rdev,
1606 kalindi_golden_common_registers,
1607 (const u32)ARRAY_SIZE(kalindi_golden_common_registers));
1608 radeon_program_register_sequence(rdev,
1609 kalindi_golden_spm_registers,
1610 (const u32)ARRAY_SIZE(kalindi_golden_spm_registers));
1613 radeon_program_register_sequence(rdev,
1614 spectre_mgcg_cgcg_init,
1615 (const u32)ARRAY_SIZE(spectre_mgcg_cgcg_init));
1616 radeon_program_register_sequence(rdev,
1617 spectre_golden_registers,
1618 (const u32)ARRAY_SIZE(spectre_golden_registers));
1619 radeon_program_register_sequence(rdev,
1620 spectre_golden_common_registers,
1621 (const u32)ARRAY_SIZE(spectre_golden_common_registers));
1622 radeon_program_register_sequence(rdev,
1623 spectre_golden_spm_registers,
1624 (const u32)ARRAY_SIZE(spectre_golden_spm_registers));
1627 radeon_program_register_sequence(rdev,
1628 hawaii_mgcg_cgcg_init,
1629 (const u32)ARRAY_SIZE(hawaii_mgcg_cgcg_init));
1630 radeon_program_register_sequence(rdev,
1631 hawaii_golden_registers,
1632 (const u32)ARRAY_SIZE(hawaii_golden_registers));
1633 radeon_program_register_sequence(rdev,
1634 hawaii_golden_common_registers,
1635 (const u32)ARRAY_SIZE(hawaii_golden_common_registers));
1636 radeon_program_register_sequence(rdev,
1637 hawaii_golden_spm_registers,
1638 (const u32)ARRAY_SIZE(hawaii_golden_spm_registers));
1643 mutex_unlock(&rdev->grbm_idx_mutex);
1647 * cik_get_xclk - get the xclk
1649 * @rdev: radeon_device pointer
1651 * Returns the reference clock used by the gfx engine
1654 u32 cik_get_xclk(struct radeon_device *rdev)
1656 u32 reference_clock = rdev->clock.spll.reference_freq;
1658 if (rdev->flags & RADEON_IS_IGP) {
1659 if (RREG32_SMC(GENERAL_PWRMGT) & GPU_COUNTER_CLK)
1660 return reference_clock / 2;
1662 if (RREG32_SMC(CG_CLKPIN_CNTL) & XTALIN_DIVIDE)
1663 return reference_clock / 4;
1665 return reference_clock;
1669 * cik_mm_rdoorbell - read a doorbell dword
1671 * @rdev: radeon_device pointer
1672 * @index: doorbell index
1674 * Returns the value in the doorbell aperture at the
1675 * requested doorbell index (CIK).
1677 u32 cik_mm_rdoorbell(struct radeon_device *rdev, u32 index)
1679 if (index < rdev->doorbell.num_doorbells) {
1680 return readl(rdev->doorbell.ptr + index);
1682 DRM_ERROR("reading beyond doorbell aperture: 0x%08x!\n", index);
1688 * cik_mm_wdoorbell - write a doorbell dword
1690 * @rdev: radeon_device pointer
1691 * @index: doorbell index
1692 * @v: value to write
1694 * Writes @v to the doorbell aperture at the
1695 * requested doorbell index (CIK).
1697 void cik_mm_wdoorbell(struct radeon_device *rdev, u32 index, u32 v)
1699 if (index < rdev->doorbell.num_doorbells) {
1700 writel(v, rdev->doorbell.ptr + index);
1702 DRM_ERROR("writing beyond doorbell aperture: 0x%08x!\n", index);
1706 #define BONAIRE_IO_MC_REGS_SIZE 36
1708 static const u32 bonaire_io_mc_regs[BONAIRE_IO_MC_REGS_SIZE][2] =
1710 {0x00000070, 0x04400000},
1711 {0x00000071, 0x80c01803},
1712 {0x00000072, 0x00004004},
1713 {0x00000073, 0x00000100},
1714 {0x00000074, 0x00ff0000},
1715 {0x00000075, 0x34000000},
1716 {0x00000076, 0x08000014},
1717 {0x00000077, 0x00cc08ec},
1718 {0x00000078, 0x00000400},
1719 {0x00000079, 0x00000000},
1720 {0x0000007a, 0x04090000},
1721 {0x0000007c, 0x00000000},
1722 {0x0000007e, 0x4408a8e8},
1723 {0x0000007f, 0x00000304},
1724 {0x00000080, 0x00000000},
1725 {0x00000082, 0x00000001},
1726 {0x00000083, 0x00000002},
1727 {0x00000084, 0xf3e4f400},
1728 {0x00000085, 0x052024e3},
1729 {0x00000087, 0x00000000},
1730 {0x00000088, 0x01000000},
1731 {0x0000008a, 0x1c0a0000},
1732 {0x0000008b, 0xff010000},
1733 {0x0000008d, 0xffffefff},
1734 {0x0000008e, 0xfff3efff},
1735 {0x0000008f, 0xfff3efbf},
1736 {0x00000092, 0xf7ffffff},
1737 {0x00000093, 0xffffff7f},
1738 {0x00000095, 0x00101101},
1739 {0x00000096, 0x00000fff},
1740 {0x00000097, 0x00116fff},
1741 {0x00000098, 0x60010000},
1742 {0x00000099, 0x10010000},
1743 {0x0000009a, 0x00006000},
1744 {0x0000009b, 0x00001000},
1745 {0x0000009f, 0x00b48000}
1748 #define HAWAII_IO_MC_REGS_SIZE 22
1750 static const u32 hawaii_io_mc_regs[HAWAII_IO_MC_REGS_SIZE][2] =
1752 {0x0000007d, 0x40000000},
1753 {0x0000007e, 0x40180304},
1754 {0x0000007f, 0x0000ff00},
1755 {0x00000081, 0x00000000},
1756 {0x00000083, 0x00000800},
1757 {0x00000086, 0x00000000},
1758 {0x00000087, 0x00000100},
1759 {0x00000088, 0x00020100},
1760 {0x00000089, 0x00000000},
1761 {0x0000008b, 0x00040000},
1762 {0x0000008c, 0x00000100},
1763 {0x0000008e, 0xff010000},
1764 {0x00000090, 0xffffefff},
1765 {0x00000091, 0xfff3efff},
1766 {0x00000092, 0xfff3efbf},
1767 {0x00000093, 0xf7ffffff},
1768 {0x00000094, 0xffffff7f},
1769 {0x00000095, 0x00000fff},
1770 {0x00000096, 0x00116fff},
1771 {0x00000097, 0x60010000},
1772 {0x00000098, 0x10010000},
1773 {0x0000009f, 0x00c79000}
1778 * cik_srbm_select - select specific register instances
1780 * @rdev: radeon_device pointer
1781 * @me: selected ME (micro engine)
1786 * Switches the currently active registers instances. Some
1787 * registers are instanced per VMID, others are instanced per
1788 * me/pipe/queue combination.
1790 static void cik_srbm_select(struct radeon_device *rdev,
1791 u32 me, u32 pipe, u32 queue, u32 vmid)
1793 u32 srbm_gfx_cntl = (PIPEID(pipe & 0x3) |
1796 QUEUEID(queue & 0x7));
1797 WREG32(SRBM_GFX_CNTL, srbm_gfx_cntl);
1802 * ci_mc_load_microcode - load MC ucode into the hw
1804 * @rdev: radeon_device pointer
1806 * Load the GDDR MC ucode into the hw (CIK).
1807 * Returns 0 on success, error on failure.
1809 int ci_mc_load_microcode(struct radeon_device *rdev)
1811 const __be32 *fw_data = NULL;
1812 const __le32 *new_fw_data = NULL;
1813 u32 running, blackout = 0, tmp;
1814 u32 *io_mc_regs = NULL;
1815 const __le32 *new_io_mc_regs = NULL;
1816 int i, regs_size, ucode_size;
1822 const struct mc_firmware_header_v1_0 *hdr =
1823 (const struct mc_firmware_header_v1_0 *)rdev->mc_fw->data;
1825 radeon_ucode_print_mc_hdr(&hdr->header);
1827 regs_size = le32_to_cpu(hdr->io_debug_size_bytes) / (4 * 2);
1828 new_io_mc_regs = (const __le32 *)
1829 (rdev->mc_fw->data + le32_to_cpu(hdr->io_debug_array_offset_bytes));
1830 ucode_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
1831 new_fw_data = (const __le32 *)
1832 (rdev->mc_fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes));
1834 ucode_size = rdev->mc_fw->size / 4;
1836 switch (rdev->family) {
1838 io_mc_regs = (u32 *)&bonaire_io_mc_regs;
1839 regs_size = BONAIRE_IO_MC_REGS_SIZE;
1842 io_mc_regs = (u32 *)&hawaii_io_mc_regs;
1843 regs_size = HAWAII_IO_MC_REGS_SIZE;
1848 fw_data = (const __be32 *)rdev->mc_fw->data;
1851 running = RREG32(MC_SEQ_SUP_CNTL) & RUN_MASK;
1855 blackout = RREG32(MC_SHARED_BLACKOUT_CNTL);
1856 WREG32(MC_SHARED_BLACKOUT_CNTL, blackout | 1);
1859 /* reset the engine and set to writable */
1860 WREG32(MC_SEQ_SUP_CNTL, 0x00000008);
1861 WREG32(MC_SEQ_SUP_CNTL, 0x00000010);
1863 /* load mc io regs */
1864 for (i = 0; i < regs_size; i++) {
1866 WREG32(MC_SEQ_IO_DEBUG_INDEX, le32_to_cpup(new_io_mc_regs++));
1867 WREG32(MC_SEQ_IO_DEBUG_DATA, le32_to_cpup(new_io_mc_regs++));
1869 WREG32(MC_SEQ_IO_DEBUG_INDEX, io_mc_regs[(i << 1)]);
1870 WREG32(MC_SEQ_IO_DEBUG_DATA, io_mc_regs[(i << 1) + 1]);
1874 tmp = RREG32(MC_SEQ_MISC0);
1875 if ((rdev->pdev->device == 0x6649) && ((tmp & 0xff00) == 0x5600)) {
1876 WREG32(MC_SEQ_IO_DEBUG_INDEX, 5);
1877 WREG32(MC_SEQ_IO_DEBUG_DATA, 0x00000023);
1878 WREG32(MC_SEQ_IO_DEBUG_INDEX, 9);
1879 WREG32(MC_SEQ_IO_DEBUG_DATA, 0x000001f0);
1882 /* load the MC ucode */
1883 for (i = 0; i < ucode_size; i++) {
1885 WREG32(MC_SEQ_SUP_PGM, le32_to_cpup(new_fw_data++));
1887 WREG32(MC_SEQ_SUP_PGM, be32_to_cpup(fw_data++));
1890 /* put the engine back into the active state */
1891 WREG32(MC_SEQ_SUP_CNTL, 0x00000008);
1892 WREG32(MC_SEQ_SUP_CNTL, 0x00000004);
1893 WREG32(MC_SEQ_SUP_CNTL, 0x00000001);
1895 /* wait for training to complete */
1896 for (i = 0; i < rdev->usec_timeout; i++) {
1897 if (RREG32(MC_SEQ_TRAIN_WAKEUP_CNTL) & TRAIN_DONE_D0)
1901 for (i = 0; i < rdev->usec_timeout; i++) {
1902 if (RREG32(MC_SEQ_TRAIN_WAKEUP_CNTL) & TRAIN_DONE_D1)
1908 WREG32(MC_SHARED_BLACKOUT_CNTL, blackout);
1915 * cik_init_microcode - load ucode images from disk
1917 * @rdev: radeon_device pointer
1919 * Use the firmware interface to load the ucode images into
1920 * the driver (not loaded into hw).
1921 * Returns 0 on success, error on failure.
1923 static int cik_init_microcode(struct radeon_device *rdev)
1925 const char *chip_name;
1926 const char *new_chip_name;
1927 size_t pfp_req_size, me_req_size, ce_req_size,
1928 mec_req_size, rlc_req_size, mc_req_size = 0,
1929 sdma_req_size, smc_req_size = 0, mc2_req_size = 0;
1937 switch (rdev->family) {
1939 chip_name = "BONAIRE";
1940 new_chip_name = "bonaire";
1941 pfp_req_size = CIK_PFP_UCODE_SIZE * 4;
1942 me_req_size = CIK_ME_UCODE_SIZE * 4;
1943 ce_req_size = CIK_CE_UCODE_SIZE * 4;
1944 mec_req_size = CIK_MEC_UCODE_SIZE * 4;
1945 rlc_req_size = BONAIRE_RLC_UCODE_SIZE * 4;
1946 mc_req_size = BONAIRE_MC_UCODE_SIZE * 4;
1947 mc2_req_size = BONAIRE_MC2_UCODE_SIZE * 4;
1948 sdma_req_size = CIK_SDMA_UCODE_SIZE * 4;
1949 smc_req_size = ALIGN(BONAIRE_SMC_UCODE_SIZE, 4);
1953 chip_name = "HAWAII";
1954 new_chip_name = "hawaii";
1955 pfp_req_size = CIK_PFP_UCODE_SIZE * 4;
1956 me_req_size = CIK_ME_UCODE_SIZE * 4;
1957 ce_req_size = CIK_CE_UCODE_SIZE * 4;
1958 mec_req_size = CIK_MEC_UCODE_SIZE * 4;
1959 rlc_req_size = BONAIRE_RLC_UCODE_SIZE * 4;
1960 mc_req_size = HAWAII_MC_UCODE_SIZE * 4;
1961 mc2_req_size = HAWAII_MC2_UCODE_SIZE * 4;
1962 sdma_req_size = CIK_SDMA_UCODE_SIZE * 4;
1963 smc_req_size = ALIGN(HAWAII_SMC_UCODE_SIZE, 4);
1967 chip_name = "KAVERI";
1968 new_chip_name = "kaveri";
1969 pfp_req_size = CIK_PFP_UCODE_SIZE * 4;
1970 me_req_size = CIK_ME_UCODE_SIZE * 4;
1971 ce_req_size = CIK_CE_UCODE_SIZE * 4;
1972 mec_req_size = CIK_MEC_UCODE_SIZE * 4;
1973 rlc_req_size = KV_RLC_UCODE_SIZE * 4;
1974 sdma_req_size = CIK_SDMA_UCODE_SIZE * 4;
1978 chip_name = "KABINI";
1979 new_chip_name = "kabini";
1980 pfp_req_size = CIK_PFP_UCODE_SIZE * 4;
1981 me_req_size = CIK_ME_UCODE_SIZE * 4;
1982 ce_req_size = CIK_CE_UCODE_SIZE * 4;
1983 mec_req_size = CIK_MEC_UCODE_SIZE * 4;
1984 rlc_req_size = KB_RLC_UCODE_SIZE * 4;
1985 sdma_req_size = CIK_SDMA_UCODE_SIZE * 4;
1989 chip_name = "MULLINS";
1990 new_chip_name = "mullins";
1991 pfp_req_size = CIK_PFP_UCODE_SIZE * 4;
1992 me_req_size = CIK_ME_UCODE_SIZE * 4;
1993 ce_req_size = CIK_CE_UCODE_SIZE * 4;
1994 mec_req_size = CIK_MEC_UCODE_SIZE * 4;
1995 rlc_req_size = ML_RLC_UCODE_SIZE * 4;
1996 sdma_req_size = CIK_SDMA_UCODE_SIZE * 4;
2002 DRM_INFO("Loading %s Microcode\n", new_chip_name);
2004 snprintf(fw_name, sizeof(fw_name), "radeon/%s_pfp.bin", new_chip_name);
2005 err = request_firmware(&rdev->pfp_fw, fw_name, rdev->dev);
2007 snprintf(fw_name, sizeof(fw_name), "radeon/%s_pfp.bin", chip_name);
2008 err = request_firmware(&rdev->pfp_fw, fw_name, rdev->dev);
2011 if (rdev->pfp_fw->size != pfp_req_size) {
2013 "cik_cp: Bogus length %zu in firmware \"%s\"\n",
2014 rdev->pfp_fw->size, fw_name);
2019 err = radeon_ucode_validate(rdev->pfp_fw);
2022 "cik_fw: validation failed for firmware \"%s\"\n",
2030 snprintf(fw_name, sizeof(fw_name), "radeon/%s_me.bin", new_chip_name);
2031 err = request_firmware(&rdev->me_fw, fw_name, rdev->dev);
2033 snprintf(fw_name, sizeof(fw_name), "radeon/%s_me.bin", chip_name);
2034 err = request_firmware(&rdev->me_fw, fw_name, rdev->dev);
2037 if (rdev->me_fw->size != me_req_size) {
2039 "cik_cp: Bogus length %zu in firmware \"%s\"\n",
2040 rdev->me_fw->size, fw_name);
2044 err = radeon_ucode_validate(rdev->me_fw);
2047 "cik_fw: validation failed for firmware \"%s\"\n",
2055 snprintf(fw_name, sizeof(fw_name), "radeon/%s_ce.bin", new_chip_name);
2056 err = request_firmware(&rdev->ce_fw, fw_name, rdev->dev);
2058 snprintf(fw_name, sizeof(fw_name), "radeon/%s_ce.bin", chip_name);
2059 err = request_firmware(&rdev->ce_fw, fw_name, rdev->dev);
2062 if (rdev->ce_fw->size != ce_req_size) {
2064 "cik_cp: Bogus length %zu in firmware \"%s\"\n",
2065 rdev->ce_fw->size, fw_name);
2069 err = radeon_ucode_validate(rdev->ce_fw);
2072 "cik_fw: validation failed for firmware \"%s\"\n",
2080 snprintf(fw_name, sizeof(fw_name), "radeon/%s_mec.bin", new_chip_name);
2081 err = request_firmware(&rdev->mec_fw, fw_name, rdev->dev);
2083 snprintf(fw_name, sizeof(fw_name), "radeon/%s_mec.bin", chip_name);
2084 err = request_firmware(&rdev->mec_fw, fw_name, rdev->dev);
2087 if (rdev->mec_fw->size != mec_req_size) {
2089 "cik_cp: Bogus length %zu in firmware \"%s\"\n",
2090 rdev->mec_fw->size, fw_name);
2094 err = radeon_ucode_validate(rdev->mec_fw);
2097 "cik_fw: validation failed for firmware \"%s\"\n",
2105 if (rdev->family == CHIP_KAVERI) {
2106 snprintf(fw_name, sizeof(fw_name), "radeon/%s_mec2.bin", new_chip_name);
2107 err = request_firmware(&rdev->mec2_fw, fw_name, rdev->dev);
2111 err = radeon_ucode_validate(rdev->mec2_fw);
2120 snprintf(fw_name, sizeof(fw_name), "radeon/%s_rlc.bin", new_chip_name);
2121 err = request_firmware(&rdev->rlc_fw, fw_name, rdev->dev);
2123 snprintf(fw_name, sizeof(fw_name), "radeon/%s_rlc.bin", chip_name);
2124 err = request_firmware(&rdev->rlc_fw, fw_name, rdev->dev);
2127 if (rdev->rlc_fw->size != rlc_req_size) {
2129 "cik_rlc: Bogus length %zu in firmware \"%s\"\n",
2130 rdev->rlc_fw->size, fw_name);
2134 err = radeon_ucode_validate(rdev->rlc_fw);
2137 "cik_fw: validation failed for firmware \"%s\"\n",
2145 snprintf(fw_name, sizeof(fw_name), "radeon/%s_sdma.bin", new_chip_name);
2146 err = request_firmware(&rdev->sdma_fw, fw_name, rdev->dev);
2148 snprintf(fw_name, sizeof(fw_name), "radeon/%s_sdma.bin", chip_name);
2149 err = request_firmware(&rdev->sdma_fw, fw_name, rdev->dev);
2152 if (rdev->sdma_fw->size != sdma_req_size) {
2154 "cik_sdma: Bogus length %zu in firmware \"%s\"\n",
2155 rdev->sdma_fw->size, fw_name);
2159 err = radeon_ucode_validate(rdev->sdma_fw);
2162 "cik_fw: validation failed for firmware \"%s\"\n",
2170 /* No SMC, MC ucode on APUs */
2171 if (!(rdev->flags & RADEON_IS_IGP)) {
2172 snprintf(fw_name, sizeof(fw_name), "radeon/%s_mc.bin", new_chip_name);
2173 err = request_firmware(&rdev->mc_fw, fw_name, rdev->dev);
2175 snprintf(fw_name, sizeof(fw_name), "radeon/%s_mc2.bin", chip_name);
2176 err = request_firmware(&rdev->mc_fw, fw_name, rdev->dev);
2178 snprintf(fw_name, sizeof(fw_name), "radeon/%s_mc.bin", chip_name);
2179 err = request_firmware(&rdev->mc_fw, fw_name, rdev->dev);
2183 if ((rdev->mc_fw->size != mc_req_size) &&
2184 (rdev->mc_fw->size != mc2_req_size)){
2186 "cik_mc: Bogus length %zu in firmware \"%s\"\n",
2187 rdev->mc_fw->size, fw_name);
2190 DRM_INFO("%s: %zu bytes\n", fw_name, rdev->mc_fw->size);
2192 err = radeon_ucode_validate(rdev->mc_fw);
2195 "cik_fw: validation failed for firmware \"%s\"\n",
2203 snprintf(fw_name, sizeof(fw_name), "radeon/%s_smc.bin", new_chip_name);
2204 err = request_firmware(&rdev->smc_fw, fw_name, rdev->dev);
2206 snprintf(fw_name, sizeof(fw_name), "radeon/%s_smc.bin", chip_name);
2207 err = request_firmware(&rdev->smc_fw, fw_name, rdev->dev);
2210 "smc: error loading firmware \"%s\"\n",
2212 release_firmware(rdev->smc_fw);
2213 rdev->smc_fw = NULL;
2215 } else if (rdev->smc_fw->size != smc_req_size) {
2217 "cik_smc: Bogus length %zu in firmware \"%s\"\n",
2218 rdev->smc_fw->size, fw_name);
2222 err = radeon_ucode_validate(rdev->smc_fw);
2225 "cik_fw: validation failed for firmware \"%s\"\n",
2235 rdev->new_fw = false;
2236 } else if (new_fw < num_fw) {
2237 printk(KERN_ERR "ci_fw: mixing new and old firmware!\n");
2240 rdev->new_fw = true;
2247 "cik_cp: Failed to load firmware \"%s\"\n",
2249 release_firmware(rdev->pfp_fw);
2250 rdev->pfp_fw = NULL;
2251 release_firmware(rdev->me_fw);
2253 release_firmware(rdev->ce_fw);
2255 release_firmware(rdev->mec_fw);
2256 rdev->mec_fw = NULL;
2257 release_firmware(rdev->mec2_fw);
2258 rdev->mec2_fw = NULL;
2259 release_firmware(rdev->rlc_fw);
2260 rdev->rlc_fw = NULL;
2261 release_firmware(rdev->sdma_fw);
2262 rdev->sdma_fw = NULL;
2263 release_firmware(rdev->mc_fw);
2265 release_firmware(rdev->smc_fw);
2266 rdev->smc_fw = NULL;
2275 * cik_tiling_mode_table_init - init the hw tiling table
2277 * @rdev: radeon_device pointer
2279 * Starting with SI, the tiling setup is done globally in a
2280 * set of 32 tiling modes. Rather than selecting each set of
2281 * parameters per surface as on older asics, we just select
2282 * which index in the tiling table we want to use, and the
2283 * surface uses those parameters (CIK).
2285 static void cik_tiling_mode_table_init(struct radeon_device *rdev)
2287 const u32 num_tile_mode_states = 32;
2288 const u32 num_secondary_tile_mode_states = 16;
2289 u32 reg_offset, gb_tile_moden, split_equal_to_row_size;
2290 u32 num_pipe_configs;
2291 u32 num_rbs = rdev->config.cik.max_backends_per_se *
2292 rdev->config.cik.max_shader_engines;
2294 switch (rdev->config.cik.mem_row_size_in_kb) {
2296 split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_1KB;
2300 split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_2KB;
2303 split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_4KB;
2307 num_pipe_configs = rdev->config.cik.max_tile_pipes;
2308 if (num_pipe_configs > 8)
2309 num_pipe_configs = 16;
2311 if (num_pipe_configs == 16) {
2312 for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) {
2313 switch (reg_offset) {
2315 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2316 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2317 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2318 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B));
2321 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2322 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2323 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2324 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B));
2327 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2328 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2329 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2330 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B));
2333 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2334 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2335 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2336 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B));
2339 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2340 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2341 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2342 TILE_SPLIT(split_equal_to_row_size));
2345 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2346 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2347 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2350 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2351 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2352 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2353 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B));
2356 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2357 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2358 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2359 TILE_SPLIT(split_equal_to_row_size));
2362 gb_tile_moden = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
2363 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16));
2366 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2367 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2368 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING));
2371 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2372 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2373 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2374 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2377 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2378 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2379 PIPE_CONFIG(ADDR_SURF_P16_32x32_8x16) |
2380 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2383 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2384 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2385 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2386 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2389 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2390 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2391 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING));
2394 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2395 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2396 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2397 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2400 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2401 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2402 PIPE_CONFIG(ADDR_SURF_P16_32x32_8x16) |
2403 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2406 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2407 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2408 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2409 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2412 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2413 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2414 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING));
2417 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2418 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2419 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2420 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2423 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2424 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2425 PIPE_CONFIG(ADDR_SURF_P16_32x32_8x16) |
2426 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2429 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2430 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2431 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2432 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2438 rdev->config.cik.tile_mode_array[reg_offset] = gb_tile_moden;
2439 WREG32(GB_TILE_MODE0 + (reg_offset * 4), gb_tile_moden);
2441 for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++) {
2442 switch (reg_offset) {
2444 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2445 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2446 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2447 NUM_BANKS(ADDR_SURF_16_BANK));
2450 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2451 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2452 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2453 NUM_BANKS(ADDR_SURF_16_BANK));
2456 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2457 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2458 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2459 NUM_BANKS(ADDR_SURF_16_BANK));
2462 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2463 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2464 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2465 NUM_BANKS(ADDR_SURF_16_BANK));
2468 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2469 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2470 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2471 NUM_BANKS(ADDR_SURF_8_BANK));
2474 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2475 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2476 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2477 NUM_BANKS(ADDR_SURF_4_BANK));
2480 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2481 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2482 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2483 NUM_BANKS(ADDR_SURF_2_BANK));
2486 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2487 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2488 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2489 NUM_BANKS(ADDR_SURF_16_BANK));
2492 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2493 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2494 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2495 NUM_BANKS(ADDR_SURF_16_BANK));
2498 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2499 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2500 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2501 NUM_BANKS(ADDR_SURF_16_BANK));
2504 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2505 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2506 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2507 NUM_BANKS(ADDR_SURF_8_BANK));
2510 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2511 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2512 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2513 NUM_BANKS(ADDR_SURF_4_BANK));
2516 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2517 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2518 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2519 NUM_BANKS(ADDR_SURF_2_BANK));
2522 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2523 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2524 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2525 NUM_BANKS(ADDR_SURF_2_BANK));
2531 rdev->config.cik.macrotile_mode_array[reg_offset] = gb_tile_moden;
2532 WREG32(GB_MACROTILE_MODE0 + (reg_offset * 4), gb_tile_moden);
2534 } else if (num_pipe_configs == 8) {
2535 for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) {
2536 switch (reg_offset) {
2538 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2539 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2540 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2541 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B));
2544 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2545 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2546 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2547 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B));
2550 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2551 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2552 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2553 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B));
2556 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2557 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2558 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2559 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B));
2562 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2563 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2564 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2565 TILE_SPLIT(split_equal_to_row_size));
2568 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2569 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2570 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2573 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2574 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2575 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2576 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B));
2579 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2580 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2581 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2582 TILE_SPLIT(split_equal_to_row_size));
2585 gb_tile_moden = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
2586 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16));
2589 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2590 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2591 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING));
2594 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2595 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2596 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2597 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2600 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2601 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2602 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
2603 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2606 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2607 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2608 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2609 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2612 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2613 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2614 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING));
2617 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2618 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2619 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
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_THIN_MICRO_TILING) |
2625 PIPE_CONFIG(ADDR_SURF_P8_32x32_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_THIN_MICRO_TILING) |
2631 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2632 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2635 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2636 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2637 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING));
2640 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2641 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2642 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2643 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2646 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2647 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2648 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
2649 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2652 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2653 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2654 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2655 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2661 rdev->config.cik.tile_mode_array[reg_offset] = gb_tile_moden;
2662 WREG32(GB_TILE_MODE0 + (reg_offset * 4), gb_tile_moden);
2664 for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++) {
2665 switch (reg_offset) {
2667 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2668 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2669 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2670 NUM_BANKS(ADDR_SURF_16_BANK));
2673 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2674 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2675 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2676 NUM_BANKS(ADDR_SURF_16_BANK));
2679 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2680 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2681 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2682 NUM_BANKS(ADDR_SURF_16_BANK));
2685 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2686 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2687 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2688 NUM_BANKS(ADDR_SURF_16_BANK));
2691 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2692 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2693 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2694 NUM_BANKS(ADDR_SURF_8_BANK));
2697 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2698 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2699 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2700 NUM_BANKS(ADDR_SURF_4_BANK));
2703 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2704 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2705 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2706 NUM_BANKS(ADDR_SURF_2_BANK));
2709 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2710 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
2711 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2712 NUM_BANKS(ADDR_SURF_16_BANK));
2715 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2716 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2717 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2718 NUM_BANKS(ADDR_SURF_16_BANK));
2721 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2722 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2723 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2724 NUM_BANKS(ADDR_SURF_16_BANK));
2727 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2728 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2729 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2730 NUM_BANKS(ADDR_SURF_16_BANK));
2733 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2734 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2735 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2736 NUM_BANKS(ADDR_SURF_8_BANK));
2739 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2740 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2741 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2742 NUM_BANKS(ADDR_SURF_4_BANK));
2745 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2746 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2747 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2748 NUM_BANKS(ADDR_SURF_2_BANK));
2754 rdev->config.cik.macrotile_mode_array[reg_offset] = gb_tile_moden;
2755 WREG32(GB_MACROTILE_MODE0 + (reg_offset * 4), gb_tile_moden);
2757 } else if (num_pipe_configs == 4) {
2759 for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) {
2760 switch (reg_offset) {
2762 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2763 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2764 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2765 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B));
2768 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2769 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2770 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2771 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B));
2774 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2775 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2776 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2777 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B));
2780 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2781 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2782 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2783 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B));
2786 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2787 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2788 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2789 TILE_SPLIT(split_equal_to_row_size));
2792 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2793 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2794 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2797 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2798 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2799 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2800 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B));
2803 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2804 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2805 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2806 TILE_SPLIT(split_equal_to_row_size));
2809 gb_tile_moden = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
2810 PIPE_CONFIG(ADDR_SURF_P4_16x16));
2813 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2814 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2815 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING));
2818 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2819 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2820 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2821 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2824 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2825 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2826 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
2827 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2830 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2831 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2832 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2833 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2836 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2837 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2838 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING));
2841 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2842 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2843 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2844 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2847 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2848 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2849 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
2850 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2853 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2854 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2855 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2856 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2859 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2860 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2861 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING));
2864 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2865 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2866 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2867 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2870 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2871 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2872 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
2873 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2876 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2877 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2878 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2879 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2885 rdev->config.cik.tile_mode_array[reg_offset] = gb_tile_moden;
2886 WREG32(GB_TILE_MODE0 + (reg_offset * 4), gb_tile_moden);
2888 } else if (num_rbs < 4) {
2889 for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) {
2890 switch (reg_offset) {
2892 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2893 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2894 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
2895 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B));
2898 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2899 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2900 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
2901 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B));
2904 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2905 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2906 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
2907 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B));
2910 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2911 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2912 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
2913 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B));
2916 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2917 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2918 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
2919 TILE_SPLIT(split_equal_to_row_size));
2922 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2923 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
2924 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2927 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2928 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2929 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
2930 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B));
2933 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2934 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
2935 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
2936 TILE_SPLIT(split_equal_to_row_size));
2939 gb_tile_moden = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
2940 PIPE_CONFIG(ADDR_SURF_P4_8x16));
2943 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2944 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
2945 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING));
2948 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2949 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2950 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
2951 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2954 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2955 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2956 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
2957 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2960 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2961 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2962 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
2963 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2966 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2967 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
2968 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING));
2971 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2972 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2973 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
2974 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2977 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2978 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2979 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
2980 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2983 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2984 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2985 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
2986 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2989 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2990 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
2991 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING));
2994 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
2995 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2996 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
2997 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3000 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3001 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
3002 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
3003 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3006 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
3007 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
3008 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
3009 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3015 rdev->config.cik.tile_mode_array[reg_offset] = gb_tile_moden;
3016 WREG32(GB_TILE_MODE0 + (reg_offset * 4), gb_tile_moden);
3019 for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++) {
3020 switch (reg_offset) {
3022 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3023 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3024 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3025 NUM_BANKS(ADDR_SURF_16_BANK));
3028 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3029 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
3030 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3031 NUM_BANKS(ADDR_SURF_16_BANK));
3034 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3035 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3036 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3037 NUM_BANKS(ADDR_SURF_16_BANK));
3040 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3041 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3042 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3043 NUM_BANKS(ADDR_SURF_16_BANK));
3046 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3047 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3048 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3049 NUM_BANKS(ADDR_SURF_16_BANK));
3052 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3053 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3054 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3055 NUM_BANKS(ADDR_SURF_8_BANK));
3058 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3059 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3060 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
3061 NUM_BANKS(ADDR_SURF_4_BANK));
3064 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
3065 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
3066 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3067 NUM_BANKS(ADDR_SURF_16_BANK));
3070 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
3071 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3072 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3073 NUM_BANKS(ADDR_SURF_16_BANK));
3076 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3077 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3078 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3079 NUM_BANKS(ADDR_SURF_16_BANK));
3082 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3083 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
3084 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3085 NUM_BANKS(ADDR_SURF_16_BANK));
3088 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3089 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3090 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3091 NUM_BANKS(ADDR_SURF_16_BANK));
3094 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3095 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3096 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3097 NUM_BANKS(ADDR_SURF_8_BANK));
3100 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3101 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3102 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
3103 NUM_BANKS(ADDR_SURF_4_BANK));
3109 rdev->config.cik.macrotile_mode_array[reg_offset] = gb_tile_moden;
3110 WREG32(GB_MACROTILE_MODE0 + (reg_offset * 4), gb_tile_moden);
3112 } else if (num_pipe_configs == 2) {
3113 for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) {
3114 switch (reg_offset) {
3116 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3117 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
3118 PIPE_CONFIG(ADDR_SURF_P2) |
3119 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B));
3122 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3123 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
3124 PIPE_CONFIG(ADDR_SURF_P2) |
3125 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B));
3128 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3129 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
3130 PIPE_CONFIG(ADDR_SURF_P2) |
3131 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B));
3134 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3135 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
3136 PIPE_CONFIG(ADDR_SURF_P2) |
3137 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B));
3140 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3141 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
3142 PIPE_CONFIG(ADDR_SURF_P2) |
3143 TILE_SPLIT(split_equal_to_row_size));
3146 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3147 PIPE_CONFIG(ADDR_SURF_P2) |
3148 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3151 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
3152 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
3153 PIPE_CONFIG(ADDR_SURF_P2) |
3154 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B));
3157 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
3158 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
3159 PIPE_CONFIG(ADDR_SURF_P2) |
3160 TILE_SPLIT(split_equal_to_row_size));
3163 gb_tile_moden = ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
3164 PIPE_CONFIG(ADDR_SURF_P2);
3167 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3168 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
3169 PIPE_CONFIG(ADDR_SURF_P2));
3172 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3173 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
3174 PIPE_CONFIG(ADDR_SURF_P2) |
3175 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3178 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3179 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
3180 PIPE_CONFIG(ADDR_SURF_P2) |
3181 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3184 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
3185 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
3186 PIPE_CONFIG(ADDR_SURF_P2) |
3187 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3190 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3191 PIPE_CONFIG(ADDR_SURF_P2) |
3192 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING));
3195 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3196 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3197 PIPE_CONFIG(ADDR_SURF_P2) |
3198 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3201 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3202 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3203 PIPE_CONFIG(ADDR_SURF_P2) |
3204 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3207 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
3208 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3209 PIPE_CONFIG(ADDR_SURF_P2) |
3210 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3213 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3214 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
3215 PIPE_CONFIG(ADDR_SURF_P2));
3218 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
3219 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
3220 PIPE_CONFIG(ADDR_SURF_P2) |
3221 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3224 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3225 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
3226 PIPE_CONFIG(ADDR_SURF_P2) |
3227 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3230 gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
3231 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
3232 PIPE_CONFIG(ADDR_SURF_P2) |
3233 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3239 rdev->config.cik.tile_mode_array[reg_offset] = gb_tile_moden;
3240 WREG32(GB_TILE_MODE0 + (reg_offset * 4), gb_tile_moden);
3242 for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++) {
3243 switch (reg_offset) {
3245 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
3246 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3247 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3248 NUM_BANKS(ADDR_SURF_16_BANK));
3251 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
3252 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
3253 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3254 NUM_BANKS(ADDR_SURF_16_BANK));
3257 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3258 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
3259 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3260 NUM_BANKS(ADDR_SURF_16_BANK));
3263 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3264 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3265 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3266 NUM_BANKS(ADDR_SURF_16_BANK));
3269 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3270 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3271 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3272 NUM_BANKS(ADDR_SURF_16_BANK));
3275 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3276 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3277 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3278 NUM_BANKS(ADDR_SURF_16_BANK));
3281 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3282 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3283 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3284 NUM_BANKS(ADDR_SURF_8_BANK));
3287 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
3288 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
3289 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3290 NUM_BANKS(ADDR_SURF_16_BANK));
3293 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
3294 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3295 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3296 NUM_BANKS(ADDR_SURF_16_BANK));
3299 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
3300 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3301 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3302 NUM_BANKS(ADDR_SURF_16_BANK));
3305 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
3306 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
3307 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3308 NUM_BANKS(ADDR_SURF_16_BANK));
3311 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3312 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
3313 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3314 NUM_BANKS(ADDR_SURF_16_BANK));
3317 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3318 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3319 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3320 NUM_BANKS(ADDR_SURF_16_BANK));
3323 gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3324 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3325 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3326 NUM_BANKS(ADDR_SURF_8_BANK));
3332 rdev->config.cik.macrotile_mode_array[reg_offset] = gb_tile_moden;
3333 WREG32(GB_MACROTILE_MODE0 + (reg_offset * 4), gb_tile_moden);
3336 DRM_ERROR("unknown num pipe config: 0x%x\n", num_pipe_configs);
3340 * cik_select_se_sh - select which SE, SH to address
3342 * @rdev: radeon_device pointer
3343 * @se_num: shader engine to address
3344 * @sh_num: sh block to address
3346 * Select which SE, SH combinations to address. Certain
3347 * registers are instanced per SE or SH. 0xffffffff means
3348 * broadcast to all SEs or SHs (CIK).
3350 static void cik_select_se_sh(struct radeon_device *rdev,
3351 u32 se_num, u32 sh_num)
3353 u32 data = INSTANCE_BROADCAST_WRITES;
3355 if ((se_num == 0xffffffff) && (sh_num == 0xffffffff))
3356 data |= SH_BROADCAST_WRITES | SE_BROADCAST_WRITES;
3357 else if (se_num == 0xffffffff)
3358 data |= SE_BROADCAST_WRITES | SH_INDEX(sh_num);
3359 else if (sh_num == 0xffffffff)
3360 data |= SH_BROADCAST_WRITES | SE_INDEX(se_num);
3362 data |= SH_INDEX(sh_num) | SE_INDEX(se_num);
3363 WREG32(GRBM_GFX_INDEX, data);
3367 * cik_create_bitmask - create a bitmask
3369 * @bit_width: length of the mask
3371 * create a variable length bit mask (CIK).
3372 * Returns the bitmask.
3374 static u32 cik_create_bitmask(u32 bit_width)
3378 for (i = 0; i < bit_width; i++) {
3386 * cik_get_rb_disabled - computes the mask of disabled RBs
3388 * @rdev: radeon_device pointer
3389 * @max_rb_num: max RBs (render backends) for the asic
3390 * @se_num: number of SEs (shader engines) for the asic
3391 * @sh_per_se: number of SH blocks per SE for the asic
3393 * Calculates the bitmask of disabled RBs (CIK).
3394 * Returns the disabled RB bitmask.
3396 static u32 cik_get_rb_disabled(struct radeon_device *rdev,
3397 u32 max_rb_num_per_se,
3402 data = RREG32(CC_RB_BACKEND_DISABLE);
3404 data &= BACKEND_DISABLE_MASK;
3407 data |= RREG32(GC_USER_RB_BACKEND_DISABLE);
3409 data >>= BACKEND_DISABLE_SHIFT;
3411 mask = cik_create_bitmask(max_rb_num_per_se / sh_per_se);
3417 * cik_setup_rb - setup the RBs on the asic
3419 * @rdev: radeon_device pointer
3420 * @se_num: number of SEs (shader engines) for the asic
3421 * @sh_per_se: number of SH blocks per SE for the asic
3422 * @max_rb_num: max RBs (render backends) for the asic
3424 * Configures per-SE/SH RB registers (CIK).
3426 static void cik_setup_rb(struct radeon_device *rdev,
3427 u32 se_num, u32 sh_per_se,
3428 u32 max_rb_num_per_se)
3432 u32 disabled_rbs = 0;
3433 u32 enabled_rbs = 0;
3435 mutex_lock(&rdev->grbm_idx_mutex);
3436 for (i = 0; i < se_num; i++) {
3437 for (j = 0; j < sh_per_se; j++) {
3438 cik_select_se_sh(rdev, i, j);
3439 data = cik_get_rb_disabled(rdev, max_rb_num_per_se, sh_per_se);
3440 if (rdev->family == CHIP_HAWAII)
3441 disabled_rbs |= data << ((i * sh_per_se + j) * HAWAII_RB_BITMAP_WIDTH_PER_SH);
3443 disabled_rbs |= data << ((i * sh_per_se + j) * CIK_RB_BITMAP_WIDTH_PER_SH);
3446 cik_select_se_sh(rdev, 0xffffffff, 0xffffffff);
3447 mutex_unlock(&rdev->grbm_idx_mutex);
3450 for (i = 0; i < max_rb_num_per_se * se_num; i++) {
3451 if (!(disabled_rbs & mask))
3452 enabled_rbs |= mask;
3456 rdev->config.cik.backend_enable_mask = enabled_rbs;
3458 mutex_lock(&rdev->grbm_idx_mutex);
3459 for (i = 0; i < se_num; i++) {
3460 cik_select_se_sh(rdev, i, 0xffffffff);
3462 for (j = 0; j < sh_per_se; j++) {
3463 switch (enabled_rbs & 3) {
3466 data |= PKR_MAP(RASTER_CONFIG_RB_MAP_3);
3468 data |= PKR_MAP(RASTER_CONFIG_RB_MAP_0);
3471 data |= (RASTER_CONFIG_RB_MAP_0 << (i * sh_per_se + j) * 2);
3474 data |= (RASTER_CONFIG_RB_MAP_3 << (i * sh_per_se + j) * 2);
3478 data |= (RASTER_CONFIG_RB_MAP_2 << (i * sh_per_se + j) * 2);
3483 WREG32(PA_SC_RASTER_CONFIG, data);
3485 cik_select_se_sh(rdev, 0xffffffff, 0xffffffff);
3486 mutex_unlock(&rdev->grbm_idx_mutex);
3490 * cik_gpu_init - setup the 3D engine
3492 * @rdev: radeon_device pointer
3494 * Configures the 3D engine and tiling configuration
3495 * registers so that the 3D engine is usable.
3497 static void cik_gpu_init(struct radeon_device *rdev)
3499 u32 gb_addr_config = RREG32(GB_ADDR_CONFIG);
3500 u32 mc_shared_chmap, mc_arb_ramcfg;
3501 u32 hdp_host_path_cntl;
3505 switch (rdev->family) {
3507 rdev->config.cik.max_shader_engines = 2;
3508 rdev->config.cik.max_tile_pipes = 4;
3509 rdev->config.cik.max_cu_per_sh = 7;
3510 rdev->config.cik.max_sh_per_se = 1;
3511 rdev->config.cik.max_backends_per_se = 2;
3512 rdev->config.cik.max_texture_channel_caches = 4;
3513 rdev->config.cik.max_gprs = 256;
3514 rdev->config.cik.max_gs_threads = 32;
3515 rdev->config.cik.max_hw_contexts = 8;
3517 rdev->config.cik.sc_prim_fifo_size_frontend = 0x20;
3518 rdev->config.cik.sc_prim_fifo_size_backend = 0x100;
3519 rdev->config.cik.sc_hiz_tile_fifo_size = 0x30;
3520 rdev->config.cik.sc_earlyz_tile_fifo_size = 0x130;
3521 gb_addr_config = BONAIRE_GB_ADDR_CONFIG_GOLDEN;
3524 rdev->config.cik.max_shader_engines = 4;
3525 rdev->config.cik.max_tile_pipes = 16;
3526 rdev->config.cik.max_cu_per_sh = 11;
3527 rdev->config.cik.max_sh_per_se = 1;
3528 rdev->config.cik.max_backends_per_se = 4;
3529 rdev->config.cik.max_texture_channel_caches = 16;
3530 rdev->config.cik.max_gprs = 256;
3531 rdev->config.cik.max_gs_threads = 32;
3532 rdev->config.cik.max_hw_contexts = 8;
3534 rdev->config.cik.sc_prim_fifo_size_frontend = 0x20;
3535 rdev->config.cik.sc_prim_fifo_size_backend = 0x100;
3536 rdev->config.cik.sc_hiz_tile_fifo_size = 0x30;
3537 rdev->config.cik.sc_earlyz_tile_fifo_size = 0x130;
3538 gb_addr_config = HAWAII_GB_ADDR_CONFIG_GOLDEN;
3541 rdev->config.cik.max_shader_engines = 1;
3542 rdev->config.cik.max_tile_pipes = 4;
3543 if ((rdev->pdev->device == 0x1304) ||
3544 (rdev->pdev->device == 0x1305) ||
3545 (rdev->pdev->device == 0x130C) ||
3546 (rdev->pdev->device == 0x130F) ||
3547 (rdev->pdev->device == 0x1310) ||
3548 (rdev->pdev->device == 0x1311) ||
3549 (rdev->pdev->device == 0x131C)) {
3550 rdev->config.cik.max_cu_per_sh = 8;
3551 rdev->config.cik.max_backends_per_se = 2;
3552 } else if ((rdev->pdev->device == 0x1309) ||
3553 (rdev->pdev->device == 0x130A) ||
3554 (rdev->pdev->device == 0x130D) ||
3555 (rdev->pdev->device == 0x1313) ||
3556 (rdev->pdev->device == 0x131D)) {
3557 rdev->config.cik.max_cu_per_sh = 6;
3558 rdev->config.cik.max_backends_per_se = 2;
3559 } else if ((rdev->pdev->device == 0x1306) ||
3560 (rdev->pdev->device == 0x1307) ||
3561 (rdev->pdev->device == 0x130B) ||
3562 (rdev->pdev->device == 0x130E) ||
3563 (rdev->pdev->device == 0x1315) ||
3564 (rdev->pdev->device == 0x1318) ||
3565 (rdev->pdev->device == 0x131B)) {
3566 rdev->config.cik.max_cu_per_sh = 4;
3567 rdev->config.cik.max_backends_per_se = 1;
3569 rdev->config.cik.max_cu_per_sh = 3;
3570 rdev->config.cik.max_backends_per_se = 1;
3572 rdev->config.cik.max_sh_per_se = 1;
3573 rdev->config.cik.max_texture_channel_caches = 4;
3574 rdev->config.cik.max_gprs = 256;
3575 rdev->config.cik.max_gs_threads = 16;
3576 rdev->config.cik.max_hw_contexts = 8;
3578 rdev->config.cik.sc_prim_fifo_size_frontend = 0x20;
3579 rdev->config.cik.sc_prim_fifo_size_backend = 0x100;
3580 rdev->config.cik.sc_hiz_tile_fifo_size = 0x30;
3581 rdev->config.cik.sc_earlyz_tile_fifo_size = 0x130;
3582 gb_addr_config = BONAIRE_GB_ADDR_CONFIG_GOLDEN;
3587 rdev->config.cik.max_shader_engines = 1;
3588 rdev->config.cik.max_tile_pipes = 2;
3589 rdev->config.cik.max_cu_per_sh = 2;
3590 rdev->config.cik.max_sh_per_se = 1;
3591 rdev->config.cik.max_backends_per_se = 1;
3592 rdev->config.cik.max_texture_channel_caches = 2;
3593 rdev->config.cik.max_gprs = 256;
3594 rdev->config.cik.max_gs_threads = 16;
3595 rdev->config.cik.max_hw_contexts = 8;
3597 rdev->config.cik.sc_prim_fifo_size_frontend = 0x20;
3598 rdev->config.cik.sc_prim_fifo_size_backend = 0x100;
3599 rdev->config.cik.sc_hiz_tile_fifo_size = 0x30;
3600 rdev->config.cik.sc_earlyz_tile_fifo_size = 0x130;
3601 gb_addr_config = BONAIRE_GB_ADDR_CONFIG_GOLDEN;
3605 /* Initialize HDP */
3606 for (i = 0, j = 0; i < 32; i++, j += 0x18) {
3607 WREG32((0x2c14 + j), 0x00000000);
3608 WREG32((0x2c18 + j), 0x00000000);
3609 WREG32((0x2c1c + j), 0x00000000);
3610 WREG32((0x2c20 + j), 0x00000000);
3611 WREG32((0x2c24 + j), 0x00000000);
3614 WREG32(GRBM_CNTL, GRBM_READ_TIMEOUT(0xff));
3616 WREG32(BIF_FB_EN, FB_READ_EN | FB_WRITE_EN);
3618 mc_shared_chmap = RREG32(MC_SHARED_CHMAP);
3619 mc_arb_ramcfg = RREG32(MC_ARB_RAMCFG);
3621 rdev->config.cik.num_tile_pipes = rdev->config.cik.max_tile_pipes;
3622 rdev->config.cik.mem_max_burst_length_bytes = 256;
3623 tmp = (mc_arb_ramcfg & NOOFCOLS_MASK) >> NOOFCOLS_SHIFT;
3624 rdev->config.cik.mem_row_size_in_kb = (4 * (1 << (8 + tmp))) / 1024;
3625 if (rdev->config.cik.mem_row_size_in_kb > 4)
3626 rdev->config.cik.mem_row_size_in_kb = 4;
3627 /* XXX use MC settings? */
3628 rdev->config.cik.shader_engine_tile_size = 32;
3629 rdev->config.cik.num_gpus = 1;
3630 rdev->config.cik.multi_gpu_tile_size = 64;
3632 /* fix up row size */
3633 gb_addr_config &= ~ROW_SIZE_MASK;
3634 switch (rdev->config.cik.mem_row_size_in_kb) {
3637 gb_addr_config |= ROW_SIZE(0);
3640 gb_addr_config |= ROW_SIZE(1);
3643 gb_addr_config |= ROW_SIZE(2);
3647 /* setup tiling info dword. gb_addr_config is not adequate since it does
3648 * not have bank info, so create a custom tiling dword.
3649 * bits 3:0 num_pipes
3650 * bits 7:4 num_banks
3651 * bits 11:8 group_size
3652 * bits 15:12 row_size
3654 rdev->config.cik.tile_config = 0;
3655 switch (rdev->config.cik.num_tile_pipes) {
3657 rdev->config.cik.tile_config |= (0 << 0);
3660 rdev->config.cik.tile_config |= (1 << 0);
3663 rdev->config.cik.tile_config |= (2 << 0);
3667 /* XXX what about 12? */
3668 rdev->config.cik.tile_config |= (3 << 0);
3671 rdev->config.cik.tile_config |=
3672 ((mc_arb_ramcfg & NOOFBANK_MASK) >> NOOFBANK_SHIFT) << 4;
3673 rdev->config.cik.tile_config |=
3674 ((gb_addr_config & PIPE_INTERLEAVE_SIZE_MASK) >> PIPE_INTERLEAVE_SIZE_SHIFT) << 8;
3675 rdev->config.cik.tile_config |=
3676 ((gb_addr_config & ROW_SIZE_MASK) >> ROW_SIZE_SHIFT) << 12;
3678 WREG32(GB_ADDR_CONFIG, gb_addr_config);
3679 WREG32(HDP_ADDR_CONFIG, gb_addr_config);
3680 WREG32(DMIF_ADDR_CALC, gb_addr_config);
3681 WREG32(SDMA0_TILING_CONFIG + SDMA0_REGISTER_OFFSET, gb_addr_config & 0x70);
3682 WREG32(SDMA0_TILING_CONFIG + SDMA1_REGISTER_OFFSET, gb_addr_config & 0x70);
3683 WREG32(UVD_UDEC_ADDR_CONFIG, gb_addr_config);
3684 WREG32(UVD_UDEC_DB_ADDR_CONFIG, gb_addr_config);
3685 WREG32(UVD_UDEC_DBW_ADDR_CONFIG, gb_addr_config);
3687 cik_tiling_mode_table_init(rdev);
3689 cik_setup_rb(rdev, rdev->config.cik.max_shader_engines,
3690 rdev->config.cik.max_sh_per_se,
3691 rdev->config.cik.max_backends_per_se);
3693 rdev->config.cik.active_cus = 0;
3694 for (i = 0; i < rdev->config.cik.max_shader_engines; i++) {
3695 for (j = 0; j < rdev->config.cik.max_sh_per_se; j++) {
3696 rdev->config.cik.active_cus +=
3697 hweight32(cik_get_cu_active_bitmap(rdev, i, j));
3701 /* set HW defaults for 3D engine */
3702 WREG32(CP_MEQ_THRESHOLDS, MEQ1_START(0x30) | MEQ2_START(0x60));
3704 mutex_lock(&rdev->grbm_idx_mutex);
3706 * making sure that the following register writes will be broadcasted
3707 * to all the shaders
3709 cik_select_se_sh(rdev, 0xffffffff, 0xffffffff);
3710 WREG32(SX_DEBUG_1, 0x20);
3712 WREG32(TA_CNTL_AUX, 0x00010000);
3714 tmp = RREG32(SPI_CONFIG_CNTL);
3716 WREG32(SPI_CONFIG_CNTL, tmp);
3718 WREG32(SQ_CONFIG, 1);
3720 WREG32(DB_DEBUG, 0);
3722 tmp = RREG32(DB_DEBUG2) & ~0xf00fffff;
3724 WREG32(DB_DEBUG2, tmp);
3726 tmp = RREG32(DB_DEBUG3) & ~0x0002021c;
3728 WREG32(DB_DEBUG3, tmp);
3730 tmp = RREG32(CB_HW_CONTROL) & ~0x00010000;
3732 WREG32(CB_HW_CONTROL, tmp);
3734 WREG32(SPI_CONFIG_CNTL_1, VTX_DONE_DELAY(4));
3736 WREG32(PA_SC_FIFO_SIZE, (SC_FRONTEND_PRIM_FIFO_SIZE(rdev->config.cik.sc_prim_fifo_size_frontend) |
3737 SC_BACKEND_PRIM_FIFO_SIZE(rdev->config.cik.sc_prim_fifo_size_backend) |
3738 SC_HIZ_TILE_FIFO_SIZE(rdev->config.cik.sc_hiz_tile_fifo_size) |
3739 SC_EARLYZ_TILE_FIFO_SIZE(rdev->config.cik.sc_earlyz_tile_fifo_size)));
3741 WREG32(VGT_NUM_INSTANCES, 1);
3743 WREG32(CP_PERFMON_CNTL, 0);
3745 WREG32(SQ_CONFIG, 0);
3747 WREG32(PA_SC_FORCE_EOV_MAX_CNTS, (FORCE_EOV_MAX_CLK_CNT(4095) |
3748 FORCE_EOV_MAX_REZ_CNT(255)));
3750 WREG32(VGT_CACHE_INVALIDATION, CACHE_INVALIDATION(VC_AND_TC) |
3751 AUTO_INVLD_EN(ES_AND_GS_AUTO));
3753 WREG32(VGT_GS_VERTEX_REUSE, 16);
3754 WREG32(PA_SC_LINE_STIPPLE_STATE, 0);
3756 tmp = RREG32(HDP_MISC_CNTL);
3757 tmp |= HDP_FLUSH_INVALIDATE_CACHE;
3758 WREG32(HDP_MISC_CNTL, tmp);
3760 hdp_host_path_cntl = RREG32(HDP_HOST_PATH_CNTL);
3761 WREG32(HDP_HOST_PATH_CNTL, hdp_host_path_cntl);
3763 WREG32(PA_CL_ENHANCE, CLIP_VTX_REORDER_ENA | NUM_CLIP_SEQ(3));
3764 WREG32(PA_SC_ENHANCE, ENABLE_PA_SC_OUT_OF_ORDER);
3765 mutex_unlock(&rdev->grbm_idx_mutex);
3771 * GPU scratch registers helpers function.
3774 * cik_scratch_init - setup driver info for CP scratch regs
3776 * @rdev: radeon_device pointer
3778 * Set up the number and offset of the CP scratch registers.
3779 * NOTE: use of CP scratch registers is a legacy inferface and
3780 * is not used by default on newer asics (r6xx+). On newer asics,
3781 * memory buffers are used for fences rather than scratch regs.
3783 static void cik_scratch_init(struct radeon_device *rdev)
3787 rdev->scratch.num_reg = 7;
3788 rdev->scratch.reg_base = SCRATCH_REG0;
3789 for (i = 0; i < rdev->scratch.num_reg; i++) {
3790 rdev->scratch.free[i] = true;
3791 rdev->scratch.reg[i] = rdev->scratch.reg_base + (i * 4);
3796 * cik_ring_test - basic gfx ring test
3798 * @rdev: radeon_device pointer
3799 * @ring: radeon_ring structure holding ring information
3801 * Allocate a scratch register and write to it using the gfx ring (CIK).
3802 * Provides a basic gfx ring test to verify that the ring is working.
3803 * Used by cik_cp_gfx_resume();
3804 * Returns 0 on success, error on failure.
3806 int cik_ring_test(struct radeon_device *rdev, struct radeon_ring *ring)
3813 r = radeon_scratch_get(rdev, &scratch);
3815 DRM_ERROR("radeon: cp failed to get scratch reg (%d).\n", r);
3818 WREG32(scratch, 0xCAFEDEAD);
3819 r = radeon_ring_lock(rdev, ring, 3);
3821 DRM_ERROR("radeon: cp failed to lock ring %d (%d).\n", ring->idx, r);
3822 radeon_scratch_free(rdev, scratch);
3825 radeon_ring_write(ring, PACKET3(PACKET3_SET_UCONFIG_REG, 1));
3826 radeon_ring_write(ring, ((scratch - PACKET3_SET_UCONFIG_REG_START) >> 2));
3827 radeon_ring_write(ring, 0xDEADBEEF);
3828 radeon_ring_unlock_commit(rdev, ring, false);
3830 for (i = 0; i < rdev->usec_timeout; i++) {
3831 tmp = RREG32(scratch);
3832 if (tmp == 0xDEADBEEF)
3836 if (i < rdev->usec_timeout) {
3837 DRM_INFO("ring test on %d succeeded in %d usecs\n", ring->idx, i);
3839 DRM_ERROR("radeon: ring %d test failed (scratch(0x%04X)=0x%08X)\n",
3840 ring->idx, scratch, tmp);
3843 radeon_scratch_free(rdev, scratch);
3848 * cik_hdp_flush_cp_ring_emit - emit an hdp flush on the cp
3850 * @rdev: radeon_device pointer
3851 * @ridx: radeon ring index
3853 * Emits an hdp flush on the cp.
3855 static void cik_hdp_flush_cp_ring_emit(struct radeon_device *rdev,
3858 struct radeon_ring *ring = &rdev->ring[ridx];
3861 switch (ring->idx) {
3862 case CAYMAN_RING_TYPE_CP1_INDEX:
3863 case CAYMAN_RING_TYPE_CP2_INDEX:
3867 ref_and_mask = CP2 << ring->pipe;
3870 ref_and_mask = CP6 << ring->pipe;
3876 case RADEON_RING_TYPE_GFX_INDEX:
3881 radeon_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
3882 radeon_ring_write(ring, (WAIT_REG_MEM_OPERATION(1) | /* write, wait, write */
3883 WAIT_REG_MEM_FUNCTION(3) | /* == */
3884 WAIT_REG_MEM_ENGINE(1))); /* pfp */
3885 radeon_ring_write(ring, GPU_HDP_FLUSH_REQ >> 2);
3886 radeon_ring_write(ring, GPU_HDP_FLUSH_DONE >> 2);
3887 radeon_ring_write(ring, ref_and_mask);
3888 radeon_ring_write(ring, ref_and_mask);
3889 radeon_ring_write(ring, 0x20); /* poll interval */
3893 * cik_fence_gfx_ring_emit - emit a fence on the gfx ring
3895 * @rdev: radeon_device pointer
3896 * @fence: radeon fence object
3898 * Emits a fence sequnce number on the gfx ring and flushes
3901 void cik_fence_gfx_ring_emit(struct radeon_device *rdev,
3902 struct radeon_fence *fence)
3904 struct radeon_ring *ring = &rdev->ring[fence->ring];
3905 u64 addr = rdev->fence_drv[fence->ring].gpu_addr;
3907 /* EVENT_WRITE_EOP - flush caches, send int */
3908 radeon_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE_EOP, 4));
3909 radeon_ring_write(ring, (EOP_TCL1_ACTION_EN |
3911 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
3913 radeon_ring_write(ring, addr & 0xfffffffc);
3914 radeon_ring_write(ring, (upper_32_bits(addr) & 0xffff) | DATA_SEL(1) | INT_SEL(2));
3915 radeon_ring_write(ring, fence->seq);
3916 radeon_ring_write(ring, 0);
3920 * cik_fence_compute_ring_emit - emit a fence on the compute ring
3922 * @rdev: radeon_device pointer
3923 * @fence: radeon fence object
3925 * Emits a fence sequnce number on the compute ring and flushes
3928 void cik_fence_compute_ring_emit(struct radeon_device *rdev,
3929 struct radeon_fence *fence)
3931 struct radeon_ring *ring = &rdev->ring[fence->ring];
3932 u64 addr = rdev->fence_drv[fence->ring].gpu_addr;
3934 /* RELEASE_MEM - flush caches, send int */
3935 radeon_ring_write(ring, PACKET3(PACKET3_RELEASE_MEM, 5));
3936 radeon_ring_write(ring, (EOP_TCL1_ACTION_EN |
3938 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
3940 radeon_ring_write(ring, DATA_SEL(1) | INT_SEL(2));
3941 radeon_ring_write(ring, addr & 0xfffffffc);
3942 radeon_ring_write(ring, upper_32_bits(addr));
3943 radeon_ring_write(ring, fence->seq);
3944 radeon_ring_write(ring, 0);
3948 * cik_semaphore_ring_emit - emit a semaphore on the CP ring
3950 * @rdev: radeon_device pointer
3951 * @ring: radeon ring buffer object
3952 * @semaphore: radeon semaphore object
3953 * @emit_wait: Is this a sempahore wait?
3955 * Emits a semaphore signal/wait packet to the CP ring and prevents the PFP
3956 * from running ahead of semaphore waits.
3958 bool cik_semaphore_ring_emit(struct radeon_device *rdev,
3959 struct radeon_ring *ring,
3960 struct radeon_semaphore *semaphore,
3963 uint64_t addr = semaphore->gpu_addr;
3964 unsigned sel = emit_wait ? PACKET3_SEM_SEL_WAIT : PACKET3_SEM_SEL_SIGNAL;
3966 radeon_ring_write(ring, PACKET3(PACKET3_MEM_SEMAPHORE, 1));
3967 radeon_ring_write(ring, lower_32_bits(addr));
3968 radeon_ring_write(ring, (upper_32_bits(addr) & 0xffff) | sel);
3970 if (emit_wait && ring->idx == RADEON_RING_TYPE_GFX_INDEX) {
3971 /* Prevent the PFP from running ahead of the semaphore wait */
3972 radeon_ring_write(ring, PACKET3(PACKET3_PFP_SYNC_ME, 0));
3973 radeon_ring_write(ring, 0x0);
3980 * cik_copy_cpdma - copy pages using the CP DMA engine
3982 * @rdev: radeon_device pointer
3983 * @src_offset: src GPU address
3984 * @dst_offset: dst GPU address
3985 * @num_gpu_pages: number of GPU pages to xfer
3986 * @resv: reservation object to sync to
3988 * Copy GPU paging using the CP DMA engine (CIK+).
3989 * Used by the radeon ttm implementation to move pages if
3990 * registered as the asic copy callback.
3992 struct radeon_fence *cik_copy_cpdma(struct radeon_device *rdev,
3993 uint64_t src_offset, uint64_t dst_offset,
3994 unsigned num_gpu_pages,
3995 struct reservation_object *resv)
3997 struct radeon_fence *fence;
3998 struct radeon_sync sync;
3999 int ring_index = rdev->asic->copy.blit_ring_index;
4000 struct radeon_ring *ring = &rdev->ring[ring_index];
4001 u32 size_in_bytes, cur_size_in_bytes, control;
4005 radeon_sync_create(&sync);
4007 size_in_bytes = (num_gpu_pages << RADEON_GPU_PAGE_SHIFT);
4008 num_loops = DIV_ROUND_UP(size_in_bytes, 0x1fffff);
4009 r = radeon_ring_lock(rdev, ring, num_loops * 7 + 18);
4011 DRM_ERROR("radeon: moving bo (%d).\n", r);
4012 radeon_sync_free(rdev, &sync, NULL);
4016 radeon_sync_resv(rdev, &sync, resv, false);
4017 radeon_sync_rings(rdev, &sync, ring->idx);
4019 for (i = 0; i < num_loops; i++) {
4020 cur_size_in_bytes = size_in_bytes;
4021 if (cur_size_in_bytes > 0x1fffff)
4022 cur_size_in_bytes = 0x1fffff;
4023 size_in_bytes -= cur_size_in_bytes;
4025 if (size_in_bytes == 0)
4026 control |= PACKET3_DMA_DATA_CP_SYNC;
4027 radeon_ring_write(ring, PACKET3(PACKET3_DMA_DATA, 5));
4028 radeon_ring_write(ring, control);
4029 radeon_ring_write(ring, lower_32_bits(src_offset));
4030 radeon_ring_write(ring, upper_32_bits(src_offset));
4031 radeon_ring_write(ring, lower_32_bits(dst_offset));
4032 radeon_ring_write(ring, upper_32_bits(dst_offset));
4033 radeon_ring_write(ring, cur_size_in_bytes);
4034 src_offset += cur_size_in_bytes;
4035 dst_offset += cur_size_in_bytes;
4038 r = radeon_fence_emit(rdev, &fence, ring->idx);
4040 radeon_ring_unlock_undo(rdev, ring);
4041 radeon_sync_free(rdev, &sync, NULL);
4045 radeon_ring_unlock_commit(rdev, ring, false);
4046 radeon_sync_free(rdev, &sync, fence);
4055 * cik_ring_ib_execute - emit an IB (Indirect Buffer) on the gfx ring
4057 * @rdev: radeon_device pointer
4058 * @ib: radeon indirect buffer object
4060 * Emits an DE (drawing engine) or CE (constant engine) IB
4061 * on the gfx ring. IBs are usually generated by userspace
4062 * acceleration drivers and submitted to the kernel for
4063 * sheduling on the ring. This function schedules the IB
4064 * on the gfx ring for execution by the GPU.
4066 void cik_ring_ib_execute(struct radeon_device *rdev, struct radeon_ib *ib)
4068 struct radeon_ring *ring = &rdev->ring[ib->ring];
4069 unsigned vm_id = ib->vm ? ib->vm->ids[ib->ring].id : 0;
4070 u32 header, control = INDIRECT_BUFFER_VALID;
4072 if (ib->is_const_ib) {
4073 /* set switch buffer packet before const IB */
4074 radeon_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
4075 radeon_ring_write(ring, 0);
4077 header = PACKET3(PACKET3_INDIRECT_BUFFER_CONST, 2);
4080 if (ring->rptr_save_reg) {
4081 next_rptr = ring->wptr + 3 + 4;
4082 radeon_ring_write(ring, PACKET3(PACKET3_SET_UCONFIG_REG, 1));
4083 radeon_ring_write(ring, ((ring->rptr_save_reg -
4084 PACKET3_SET_UCONFIG_REG_START) >> 2));
4085 radeon_ring_write(ring, next_rptr);
4086 } else if (rdev->wb.enabled) {
4087 next_rptr = ring->wptr + 5 + 4;
4088 radeon_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
4089 radeon_ring_write(ring, WRITE_DATA_DST_SEL(1));
4090 radeon_ring_write(ring, ring->next_rptr_gpu_addr & 0xfffffffc);
4091 radeon_ring_write(ring, upper_32_bits(ring->next_rptr_gpu_addr));
4092 radeon_ring_write(ring, next_rptr);
4095 header = PACKET3(PACKET3_INDIRECT_BUFFER, 2);
4098 control |= ib->length_dw | (vm_id << 24);
4100 radeon_ring_write(ring, header);
4101 radeon_ring_write(ring,
4105 (ib->gpu_addr & 0xFFFFFFFC));
4106 radeon_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xFFFF);
4107 radeon_ring_write(ring, control);
4111 * cik_ib_test - basic gfx ring IB test
4113 * @rdev: radeon_device pointer
4114 * @ring: radeon_ring structure holding ring information
4116 * Allocate an IB and execute it on the gfx ring (CIK).
4117 * Provides a basic gfx ring test to verify that IBs are working.
4118 * Returns 0 on success, error on failure.
4120 int cik_ib_test(struct radeon_device *rdev, struct radeon_ring *ring)
4122 struct radeon_ib ib;
4128 r = radeon_scratch_get(rdev, &scratch);
4130 DRM_ERROR("radeon: failed to get scratch reg (%d).\n", r);
4133 WREG32(scratch, 0xCAFEDEAD);
4134 r = radeon_ib_get(rdev, ring->idx, &ib, NULL, 256);
4136 DRM_ERROR("radeon: failed to get ib (%d).\n", r);
4137 radeon_scratch_free(rdev, scratch);
4140 ib.ptr[0] = PACKET3(PACKET3_SET_UCONFIG_REG, 1);
4141 ib.ptr[1] = ((scratch - PACKET3_SET_UCONFIG_REG_START) >> 2);
4142 ib.ptr[2] = 0xDEADBEEF;
4144 r = radeon_ib_schedule(rdev, &ib, NULL, false);
4146 radeon_scratch_free(rdev, scratch);
4147 radeon_ib_free(rdev, &ib);
4148 DRM_ERROR("radeon: failed to schedule ib (%d).\n", r);
4151 r = radeon_fence_wait(ib.fence, false);
4153 DRM_ERROR("radeon: fence wait failed (%d).\n", r);
4154 radeon_scratch_free(rdev, scratch);
4155 radeon_ib_free(rdev, &ib);
4158 for (i = 0; i < rdev->usec_timeout; i++) {
4159 tmp = RREG32(scratch);
4160 if (tmp == 0xDEADBEEF)
4164 if (i < rdev->usec_timeout) {
4165 DRM_INFO("ib test on ring %d succeeded in %u usecs\n", ib.fence->ring, i);
4167 DRM_ERROR("radeon: ib test failed (scratch(0x%04X)=0x%08X)\n",
4171 radeon_scratch_free(rdev, scratch);
4172 radeon_ib_free(rdev, &ib);
4178 * On CIK, gfx and compute now have independant command processors.
4181 * Gfx consists of a single ring and can process both gfx jobs and
4182 * compute jobs. The gfx CP consists of three microengines (ME):
4183 * PFP - Pre-Fetch Parser
4185 * CE - Constant Engine
4186 * The PFP and ME make up what is considered the Drawing Engine (DE).
4187 * The CE is an asynchronous engine used for updating buffer desciptors
4188 * used by the DE so that they can be loaded into cache in parallel
4189 * while the DE is processing state update packets.
4192 * The compute CP consists of two microengines (ME):
4193 * MEC1 - Compute MicroEngine 1
4194 * MEC2 - Compute MicroEngine 2
4195 * Each MEC supports 4 compute pipes and each pipe supports 8 queues.
4196 * The queues are exposed to userspace and are programmed directly
4197 * by the compute runtime.
4200 * cik_cp_gfx_enable - enable/disable the gfx CP MEs
4202 * @rdev: radeon_device pointer
4203 * @enable: enable or disable the MEs
4205 * Halts or unhalts the gfx MEs.
4207 static void cik_cp_gfx_enable(struct radeon_device *rdev, bool enable)
4210 WREG32(CP_ME_CNTL, 0);
4212 if (rdev->asic->copy.copy_ring_index == RADEON_RING_TYPE_GFX_INDEX)
4213 radeon_ttm_set_active_vram_size(rdev, rdev->mc.visible_vram_size);
4214 WREG32(CP_ME_CNTL, (CP_ME_HALT | CP_PFP_HALT | CP_CE_HALT));
4215 rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = false;
4221 * cik_cp_gfx_load_microcode - load the gfx CP ME ucode
4223 * @rdev: radeon_device pointer
4225 * Loads the gfx PFP, ME, and CE ucode.
4226 * Returns 0 for success, -EINVAL if the ucode is not available.
4228 static int cik_cp_gfx_load_microcode(struct radeon_device *rdev)
4232 if (!rdev->me_fw || !rdev->pfp_fw || !rdev->ce_fw)
4235 cik_cp_gfx_enable(rdev, false);
4238 const struct gfx_firmware_header_v1_0 *pfp_hdr =
4239 (const struct gfx_firmware_header_v1_0 *)rdev->pfp_fw->data;
4240 const struct gfx_firmware_header_v1_0 *ce_hdr =
4241 (const struct gfx_firmware_header_v1_0 *)rdev->ce_fw->data;
4242 const struct gfx_firmware_header_v1_0 *me_hdr =
4243 (const struct gfx_firmware_header_v1_0 *)rdev->me_fw->data;
4244 const __le32 *fw_data;
4247 radeon_ucode_print_gfx_hdr(&pfp_hdr->header);
4248 radeon_ucode_print_gfx_hdr(&ce_hdr->header);
4249 radeon_ucode_print_gfx_hdr(&me_hdr->header);
4252 fw_data = (const __le32 *)
4253 (rdev->pfp_fw->data + le32_to_cpu(pfp_hdr->header.ucode_array_offset_bytes));
4254 fw_size = le32_to_cpu(pfp_hdr->header.ucode_size_bytes) / 4;
4255 WREG32(CP_PFP_UCODE_ADDR, 0);
4256 for (i = 0; i < fw_size; i++)
4257 WREG32(CP_PFP_UCODE_DATA, le32_to_cpup(fw_data++));
4258 WREG32(CP_PFP_UCODE_ADDR, le32_to_cpu(pfp_hdr->header.ucode_version));
4261 fw_data = (const __le32 *)
4262 (rdev->ce_fw->data + le32_to_cpu(ce_hdr->header.ucode_array_offset_bytes));
4263 fw_size = le32_to_cpu(ce_hdr->header.ucode_size_bytes) / 4;
4264 WREG32(CP_CE_UCODE_ADDR, 0);
4265 for (i = 0; i < fw_size; i++)
4266 WREG32(CP_CE_UCODE_DATA, le32_to_cpup(fw_data++));
4267 WREG32(CP_CE_UCODE_ADDR, le32_to_cpu(ce_hdr->header.ucode_version));
4270 fw_data = (const __be32 *)
4271 (rdev->me_fw->data + le32_to_cpu(me_hdr->header.ucode_array_offset_bytes));
4272 fw_size = le32_to_cpu(me_hdr->header.ucode_size_bytes) / 4;
4273 WREG32(CP_ME_RAM_WADDR, 0);
4274 for (i = 0; i < fw_size; i++)
4275 WREG32(CP_ME_RAM_DATA, le32_to_cpup(fw_data++));
4276 WREG32(CP_ME_RAM_WADDR, le32_to_cpu(me_hdr->header.ucode_version));
4277 WREG32(CP_ME_RAM_RADDR, le32_to_cpu(me_hdr->header.ucode_version));
4279 const __be32 *fw_data;
4282 fw_data = (const __be32 *)rdev->pfp_fw->data;
4283 WREG32(CP_PFP_UCODE_ADDR, 0);
4284 for (i = 0; i < CIK_PFP_UCODE_SIZE; i++)
4285 WREG32(CP_PFP_UCODE_DATA, be32_to_cpup(fw_data++));
4286 WREG32(CP_PFP_UCODE_ADDR, 0);
4289 fw_data = (const __be32 *)rdev->ce_fw->data;
4290 WREG32(CP_CE_UCODE_ADDR, 0);
4291 for (i = 0; i < CIK_CE_UCODE_SIZE; i++)
4292 WREG32(CP_CE_UCODE_DATA, be32_to_cpup(fw_data++));
4293 WREG32(CP_CE_UCODE_ADDR, 0);
4296 fw_data = (const __be32 *)rdev->me_fw->data;
4297 WREG32(CP_ME_RAM_WADDR, 0);
4298 for (i = 0; i < CIK_ME_UCODE_SIZE; i++)
4299 WREG32(CP_ME_RAM_DATA, be32_to_cpup(fw_data++));
4300 WREG32(CP_ME_RAM_WADDR, 0);
4307 * cik_cp_gfx_start - start the gfx ring
4309 * @rdev: radeon_device pointer
4311 * Enables the ring and loads the clear state context and other
4312 * packets required to init the ring.
4313 * Returns 0 for success, error for failure.
4315 static int cik_cp_gfx_start(struct radeon_device *rdev)
4317 struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
4321 WREG32(CP_MAX_CONTEXT, rdev->config.cik.max_hw_contexts - 1);
4322 WREG32(CP_ENDIAN_SWAP, 0);
4323 WREG32(CP_DEVICE_ID, 1);
4325 cik_cp_gfx_enable(rdev, true);
4327 r = radeon_ring_lock(rdev, ring, cik_default_size + 17);
4329 DRM_ERROR("radeon: cp failed to lock ring (%d).\n", r);
4333 /* init the CE partitions. CE only used for gfx on CIK */
4334 radeon_ring_write(ring, PACKET3(PACKET3_SET_BASE, 2));
4335 radeon_ring_write(ring, PACKET3_BASE_INDEX(CE_PARTITION_BASE));
4336 radeon_ring_write(ring, 0x8000);
4337 radeon_ring_write(ring, 0x8000);
4339 /* setup clear context state */
4340 radeon_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
4341 radeon_ring_write(ring, PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
4343 radeon_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1));
4344 radeon_ring_write(ring, 0x80000000);
4345 radeon_ring_write(ring, 0x80000000);
4347 for (i = 0; i < cik_default_size; i++)
4348 radeon_ring_write(ring, cik_default_state[i]);
4350 radeon_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
4351 radeon_ring_write(ring, PACKET3_PREAMBLE_END_CLEAR_STATE);
4353 /* set clear context state */
4354 radeon_ring_write(ring, PACKET3(PACKET3_CLEAR_STATE, 0));
4355 radeon_ring_write(ring, 0);
4357 radeon_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 2));
4358 radeon_ring_write(ring, 0x00000316);
4359 radeon_ring_write(ring, 0x0000000e); /* VGT_VERTEX_REUSE_BLOCK_CNTL */
4360 radeon_ring_write(ring, 0x00000010); /* VGT_OUT_DEALLOC_CNTL */
4362 radeon_ring_unlock_commit(rdev, ring, false);
4368 * cik_cp_gfx_fini - stop the gfx ring
4370 * @rdev: radeon_device pointer
4372 * Stop the gfx ring and tear down the driver ring
4375 static void cik_cp_gfx_fini(struct radeon_device *rdev)
4377 cik_cp_gfx_enable(rdev, false);
4378 radeon_ring_fini(rdev, &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]);
4382 * cik_cp_gfx_resume - setup the gfx ring buffer registers
4384 * @rdev: radeon_device pointer
4386 * Program the location and size of the gfx ring buffer
4387 * and test it to make sure it's working.
4388 * Returns 0 for success, error for failure.
4390 static int cik_cp_gfx_resume(struct radeon_device *rdev)
4392 struct radeon_ring *ring;
4398 WREG32(CP_SEM_WAIT_TIMER, 0x0);
4399 if (rdev->family != CHIP_HAWAII)
4400 WREG32(CP_SEM_INCOMPLETE_TIMER_CNTL, 0x0);
4402 /* Set the write pointer delay */
4403 WREG32(CP_RB_WPTR_DELAY, 0);
4405 /* set the RB to use vmid 0 */
4406 WREG32(CP_RB_VMID, 0);
4408 WREG32(SCRATCH_ADDR, ((rdev->wb.gpu_addr + RADEON_WB_SCRATCH_OFFSET) >> 8) & 0xFFFFFFFF);
4410 /* ring 0 - compute and gfx */
4411 /* Set ring buffer size */
4412 ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
4413 rb_bufsz = order_base_2(ring->ring_size / 8);
4414 tmp = (order_base_2(RADEON_GPU_PAGE_SIZE/8) << 8) | rb_bufsz;
4416 tmp |= BUF_SWAP_32BIT;
4418 WREG32(CP_RB0_CNTL, tmp);
4420 /* Initialize the ring buffer's read and write pointers */
4421 WREG32(CP_RB0_CNTL, tmp | RB_RPTR_WR_ENA);
4423 WREG32(CP_RB0_WPTR, ring->wptr);
4425 /* set the wb address wether it's enabled or not */
4426 WREG32(CP_RB0_RPTR_ADDR, (rdev->wb.gpu_addr + RADEON_WB_CP_RPTR_OFFSET) & 0xFFFFFFFC);
4427 WREG32(CP_RB0_RPTR_ADDR_HI, upper_32_bits(rdev->wb.gpu_addr + RADEON_WB_CP_RPTR_OFFSET) & 0xFF);
4429 /* scratch register shadowing is no longer supported */
4430 WREG32(SCRATCH_UMSK, 0);
4432 if (!rdev->wb.enabled)
4433 tmp |= RB_NO_UPDATE;
4436 WREG32(CP_RB0_CNTL, tmp);
4438 rb_addr = ring->gpu_addr >> 8;
4439 WREG32(CP_RB0_BASE, rb_addr);
4440 WREG32(CP_RB0_BASE_HI, upper_32_bits(rb_addr));
4442 /* start the ring */
4443 cik_cp_gfx_start(rdev);
4444 rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = true;
4445 r = radeon_ring_test(rdev, RADEON_RING_TYPE_GFX_INDEX, &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]);
4447 rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = false;
4451 if (rdev->asic->copy.copy_ring_index == RADEON_RING_TYPE_GFX_INDEX)
4452 radeon_ttm_set_active_vram_size(rdev, rdev->mc.real_vram_size);
4457 u32 cik_gfx_get_rptr(struct radeon_device *rdev,
4458 struct radeon_ring *ring)
4462 if (rdev->wb.enabled)
4463 rptr = rdev->wb.wb[ring->rptr_offs/4];
4465 rptr = RREG32(CP_RB0_RPTR);
4470 u32 cik_gfx_get_wptr(struct radeon_device *rdev,
4471 struct radeon_ring *ring)
4475 wptr = RREG32(CP_RB0_WPTR);
4480 void cik_gfx_set_wptr(struct radeon_device *rdev,
4481 struct radeon_ring *ring)
4483 WREG32(CP_RB0_WPTR, ring->wptr);
4484 (void)RREG32(CP_RB0_WPTR);
4487 u32 cik_compute_get_rptr(struct radeon_device *rdev,
4488 struct radeon_ring *ring)
4492 if (rdev->wb.enabled) {
4493 rptr = rdev->wb.wb[ring->rptr_offs/4];
4495 mutex_lock(&rdev->srbm_mutex);
4496 cik_srbm_select(rdev, ring->me, ring->pipe, ring->queue, 0);
4497 rptr = RREG32(CP_HQD_PQ_RPTR);
4498 cik_srbm_select(rdev, 0, 0, 0, 0);
4499 mutex_unlock(&rdev->srbm_mutex);
4505 u32 cik_compute_get_wptr(struct radeon_device *rdev,
4506 struct radeon_ring *ring)
4510 if (rdev->wb.enabled) {
4511 /* XXX check if swapping is necessary on BE */
4512 wptr = rdev->wb.wb[ring->wptr_offs/4];
4514 mutex_lock(&rdev->srbm_mutex);
4515 cik_srbm_select(rdev, ring->me, ring->pipe, ring->queue, 0);
4516 wptr = RREG32(CP_HQD_PQ_WPTR);
4517 cik_srbm_select(rdev, 0, 0, 0, 0);
4518 mutex_unlock(&rdev->srbm_mutex);
4524 void cik_compute_set_wptr(struct radeon_device *rdev,
4525 struct radeon_ring *ring)
4527 /* XXX check if swapping is necessary on BE */
4528 rdev->wb.wb[ring->wptr_offs/4] = ring->wptr;
4529 WDOORBELL32(ring->doorbell_index, ring->wptr);
4533 * cik_cp_compute_enable - enable/disable the compute CP MEs
4535 * @rdev: radeon_device pointer
4536 * @enable: enable or disable the MEs
4538 * Halts or unhalts the compute MEs.
4540 static void cik_cp_compute_enable(struct radeon_device *rdev, bool enable)
4543 WREG32(CP_MEC_CNTL, 0);
4545 WREG32(CP_MEC_CNTL, (MEC_ME1_HALT | MEC_ME2_HALT));
4546 rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX].ready = false;
4547 rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX].ready = false;
4553 * cik_cp_compute_load_microcode - load the compute CP ME ucode
4555 * @rdev: radeon_device pointer
4557 * Loads the compute MEC1&2 ucode.
4558 * Returns 0 for success, -EINVAL if the ucode is not available.
4560 static int cik_cp_compute_load_microcode(struct radeon_device *rdev)
4567 cik_cp_compute_enable(rdev, false);
4570 const struct gfx_firmware_header_v1_0 *mec_hdr =
4571 (const struct gfx_firmware_header_v1_0 *)rdev->mec_fw->data;
4572 const __le32 *fw_data;
4575 radeon_ucode_print_gfx_hdr(&mec_hdr->header);
4578 fw_data = (const __le32 *)
4579 (rdev->mec_fw->data + le32_to_cpu(mec_hdr->header.ucode_array_offset_bytes));
4580 fw_size = le32_to_cpu(mec_hdr->header.ucode_size_bytes) / 4;
4581 WREG32(CP_MEC_ME1_UCODE_ADDR, 0);
4582 for (i = 0; i < fw_size; i++)
4583 WREG32(CP_MEC_ME1_UCODE_DATA, le32_to_cpup(fw_data++));
4584 WREG32(CP_MEC_ME1_UCODE_ADDR, le32_to_cpu(mec_hdr->header.ucode_version));
4587 if (rdev->family == CHIP_KAVERI) {
4588 const struct gfx_firmware_header_v1_0 *mec2_hdr =
4589 (const struct gfx_firmware_header_v1_0 *)rdev->mec2_fw->data;
4591 fw_data = (const __le32 *)
4592 (rdev->mec2_fw->data +
4593 le32_to_cpu(mec2_hdr->header.ucode_array_offset_bytes));
4594 fw_size = le32_to_cpu(mec2_hdr->header.ucode_size_bytes) / 4;
4595 WREG32(CP_MEC_ME2_UCODE_ADDR, 0);
4596 for (i = 0; i < fw_size; i++)
4597 WREG32(CP_MEC_ME2_UCODE_DATA, le32_to_cpup(fw_data++));
4598 WREG32(CP_MEC_ME2_UCODE_ADDR, le32_to_cpu(mec2_hdr->header.ucode_version));
4601 const __be32 *fw_data;
4604 fw_data = (const __be32 *)rdev->mec_fw->data;
4605 WREG32(CP_MEC_ME1_UCODE_ADDR, 0);
4606 for (i = 0; i < CIK_MEC_UCODE_SIZE; i++)
4607 WREG32(CP_MEC_ME1_UCODE_DATA, be32_to_cpup(fw_data++));
4608 WREG32(CP_MEC_ME1_UCODE_ADDR, 0);
4610 if (rdev->family == CHIP_KAVERI) {
4612 fw_data = (const __be32 *)rdev->mec_fw->data;
4613 WREG32(CP_MEC_ME2_UCODE_ADDR, 0);
4614 for (i = 0; i < CIK_MEC_UCODE_SIZE; i++)
4615 WREG32(CP_MEC_ME2_UCODE_DATA, be32_to_cpup(fw_data++));
4616 WREG32(CP_MEC_ME2_UCODE_ADDR, 0);
4624 * cik_cp_compute_start - start the compute queues
4626 * @rdev: radeon_device pointer
4628 * Enable the compute queues.
4629 * Returns 0 for success, error for failure.
4631 static int cik_cp_compute_start(struct radeon_device *rdev)
4633 cik_cp_compute_enable(rdev, true);
4639 * cik_cp_compute_fini - stop the compute queues
4641 * @rdev: radeon_device pointer
4643 * Stop the compute queues and tear down the driver queue
4646 static void cik_cp_compute_fini(struct radeon_device *rdev)
4650 cik_cp_compute_enable(rdev, false);
4652 for (i = 0; i < 2; i++) {
4654 idx = CAYMAN_RING_TYPE_CP1_INDEX;
4656 idx = CAYMAN_RING_TYPE_CP2_INDEX;
4658 if (rdev->ring[idx].mqd_obj) {
4659 r = radeon_bo_reserve(rdev->ring[idx].mqd_obj, false);
4660 if (unlikely(r != 0))
4661 dev_warn(rdev->dev, "(%d) reserve MQD bo failed\n", r);
4663 radeon_bo_unpin(rdev->ring[idx].mqd_obj);
4664 radeon_bo_unreserve(rdev->ring[idx].mqd_obj);
4666 radeon_bo_unref(&rdev->ring[idx].mqd_obj);
4667 rdev->ring[idx].mqd_obj = NULL;
4672 static void cik_mec_fini(struct radeon_device *rdev)
4676 if (rdev->mec.hpd_eop_obj) {
4677 r = radeon_bo_reserve(rdev->mec.hpd_eop_obj, false);
4678 if (unlikely(r != 0))
4679 dev_warn(rdev->dev, "(%d) reserve HPD EOP bo failed\n", r);
4680 radeon_bo_unpin(rdev->mec.hpd_eop_obj);
4681 radeon_bo_unreserve(rdev->mec.hpd_eop_obj);
4683 radeon_bo_unref(&rdev->mec.hpd_eop_obj);
4684 rdev->mec.hpd_eop_obj = NULL;
4688 #define MEC_HPD_SIZE 2048
4690 static int cik_mec_init(struct radeon_device *rdev)
4696 * KV: 2 MEC, 4 Pipes/MEC, 8 Queues/Pipe - 64 Queues total
4697 * CI/KB: 1 MEC, 4 Pipes/MEC, 8 Queues/Pipe - 32 Queues total
4698 * Nonetheless, we assign only 1 pipe because all other pipes will
4701 rdev->mec.num_mec = 1;
4702 rdev->mec.num_pipe = 1;
4703 rdev->mec.num_queue = rdev->mec.num_mec * rdev->mec.num_pipe * 8;
4705 if (rdev->mec.hpd_eop_obj == NULL) {
4706 r = radeon_bo_create(rdev,
4707 rdev->mec.num_mec *rdev->mec.num_pipe * MEC_HPD_SIZE * 2,
4709 RADEON_GEM_DOMAIN_GTT, 0, NULL, NULL,
4710 &rdev->mec.hpd_eop_obj);
4712 dev_warn(rdev->dev, "(%d) create HDP EOP bo failed\n", r);
4717 r = radeon_bo_reserve(rdev->mec.hpd_eop_obj, false);
4718 if (unlikely(r != 0)) {
4722 r = radeon_bo_pin(rdev->mec.hpd_eop_obj, RADEON_GEM_DOMAIN_GTT,
4723 &rdev->mec.hpd_eop_gpu_addr);
4725 dev_warn(rdev->dev, "(%d) pin HDP EOP bo failed\n", r);
4729 r = radeon_bo_kmap(rdev->mec.hpd_eop_obj, (void **)&hpd);
4731 dev_warn(rdev->dev, "(%d) map HDP EOP bo failed\n", r);
4736 /* clear memory. Not sure if this is required or not */
4737 memset(hpd, 0, rdev->mec.num_mec *rdev->mec.num_pipe * MEC_HPD_SIZE * 2);
4739 radeon_bo_kunmap(rdev->mec.hpd_eop_obj);
4740 radeon_bo_unreserve(rdev->mec.hpd_eop_obj);
4745 struct hqd_registers
4747 u32 cp_mqd_base_addr;
4748 u32 cp_mqd_base_addr_hi;
4751 u32 cp_hqd_persistent_state;
4752 u32 cp_hqd_pipe_priority;
4753 u32 cp_hqd_queue_priority;
4756 u32 cp_hqd_pq_base_hi;
4758 u32 cp_hqd_pq_rptr_report_addr;
4759 u32 cp_hqd_pq_rptr_report_addr_hi;
4760 u32 cp_hqd_pq_wptr_poll_addr;
4761 u32 cp_hqd_pq_wptr_poll_addr_hi;
4762 u32 cp_hqd_pq_doorbell_control;
4764 u32 cp_hqd_pq_control;
4765 u32 cp_hqd_ib_base_addr;
4766 u32 cp_hqd_ib_base_addr_hi;
4768 u32 cp_hqd_ib_control;
4769 u32 cp_hqd_iq_timer;
4771 u32 cp_hqd_dequeue_request;
4772 u32 cp_hqd_dma_offload;
4773 u32 cp_hqd_sema_cmd;
4774 u32 cp_hqd_msg_type;
4775 u32 cp_hqd_atomic0_preop_lo;
4776 u32 cp_hqd_atomic0_preop_hi;
4777 u32 cp_hqd_atomic1_preop_lo;
4778 u32 cp_hqd_atomic1_preop_hi;
4779 u32 cp_hqd_hq_scheduler0;
4780 u32 cp_hqd_hq_scheduler1;
4787 u32 dispatch_initiator;
4791 u32 pipeline_stat_enable;
4792 u32 perf_counter_enable;
4798 u32 resource_limits;
4799 u32 static_thread_mgmt01[2];
4801 u32 static_thread_mgmt23[2];
4803 u32 thread_trace_enable;
4806 u32 vgtcs_invoke_count[2];
4807 struct hqd_registers queue_state;
4809 u32 interrupt_queue[64];
4813 * cik_cp_compute_resume - setup the compute queue registers
4815 * @rdev: radeon_device pointer
4817 * Program the compute queues and test them to make sure they
4819 * Returns 0 for success, error for failure.
4821 static int cik_cp_compute_resume(struct radeon_device *rdev)
4825 bool use_doorbell = true;
4831 struct bonaire_mqd *mqd;
4833 r = cik_cp_compute_start(rdev);
4837 /* fix up chicken bits */
4838 tmp = RREG32(CP_CPF_DEBUG);
4840 WREG32(CP_CPF_DEBUG, tmp);
4842 /* init the pipes */
4843 mutex_lock(&rdev->srbm_mutex);
4845 eop_gpu_addr = rdev->mec.hpd_eop_gpu_addr;
4847 cik_srbm_select(rdev, 0, 0, 0, 0);
4849 /* write the EOP addr */
4850 WREG32(CP_HPD_EOP_BASE_ADDR, eop_gpu_addr >> 8);
4851 WREG32(CP_HPD_EOP_BASE_ADDR_HI, upper_32_bits(eop_gpu_addr) >> 8);
4853 /* set the VMID assigned */
4854 WREG32(CP_HPD_EOP_VMID, 0);
4856 /* set the EOP size, register value is 2^(EOP_SIZE+1) dwords */
4857 tmp = RREG32(CP_HPD_EOP_CONTROL);
4858 tmp &= ~EOP_SIZE_MASK;
4859 tmp |= order_base_2(MEC_HPD_SIZE / 8);
4860 WREG32(CP_HPD_EOP_CONTROL, tmp);
4862 mutex_unlock(&rdev->srbm_mutex);
4864 /* init the queues. Just two for now. */
4865 for (i = 0; i < 2; i++) {
4867 idx = CAYMAN_RING_TYPE_CP1_INDEX;
4869 idx = CAYMAN_RING_TYPE_CP2_INDEX;
4871 if (rdev->ring[idx].mqd_obj == NULL) {
4872 r = radeon_bo_create(rdev,
4873 sizeof(struct bonaire_mqd),
4875 RADEON_GEM_DOMAIN_GTT, 0, NULL,
4876 NULL, &rdev->ring[idx].mqd_obj);
4878 dev_warn(rdev->dev, "(%d) create MQD bo failed\n", r);
4883 r = radeon_bo_reserve(rdev->ring[idx].mqd_obj, false);
4884 if (unlikely(r != 0)) {
4885 cik_cp_compute_fini(rdev);
4888 r = radeon_bo_pin(rdev->ring[idx].mqd_obj, RADEON_GEM_DOMAIN_GTT,
4891 dev_warn(rdev->dev, "(%d) pin MQD bo failed\n", r);
4892 cik_cp_compute_fini(rdev);
4895 r = radeon_bo_kmap(rdev->ring[idx].mqd_obj, (void **)&buf);
4897 dev_warn(rdev->dev, "(%d) map MQD bo failed\n", r);
4898 cik_cp_compute_fini(rdev);
4902 /* init the mqd struct */
4903 memset(buf, 0, sizeof(struct bonaire_mqd));
4905 mqd = (struct bonaire_mqd *)buf;
4906 mqd->header = 0xC0310800;
4907 mqd->static_thread_mgmt01[0] = 0xffffffff;
4908 mqd->static_thread_mgmt01[1] = 0xffffffff;
4909 mqd->static_thread_mgmt23[0] = 0xffffffff;
4910 mqd->static_thread_mgmt23[1] = 0xffffffff;
4912 mutex_lock(&rdev->srbm_mutex);
4913 cik_srbm_select(rdev, rdev->ring[idx].me,
4914 rdev->ring[idx].pipe,
4915 rdev->ring[idx].queue, 0);
4917 /* disable wptr polling */
4918 tmp = RREG32(CP_PQ_WPTR_POLL_CNTL);
4919 tmp &= ~WPTR_POLL_EN;
4920 WREG32(CP_PQ_WPTR_POLL_CNTL, tmp);
4922 /* enable doorbell? */
4923 mqd->queue_state.cp_hqd_pq_doorbell_control =
4924 RREG32(CP_HQD_PQ_DOORBELL_CONTROL);
4926 mqd->queue_state.cp_hqd_pq_doorbell_control |= DOORBELL_EN;
4928 mqd->queue_state.cp_hqd_pq_doorbell_control &= ~DOORBELL_EN;
4929 WREG32(CP_HQD_PQ_DOORBELL_CONTROL,
4930 mqd->queue_state.cp_hqd_pq_doorbell_control);
4932 /* disable the queue if it's active */
4933 mqd->queue_state.cp_hqd_dequeue_request = 0;
4934 mqd->queue_state.cp_hqd_pq_rptr = 0;
4935 mqd->queue_state.cp_hqd_pq_wptr= 0;
4936 if (RREG32(CP_HQD_ACTIVE) & 1) {
4937 WREG32(CP_HQD_DEQUEUE_REQUEST, 1);
4938 for (j = 0; j < rdev->usec_timeout; j++) {
4939 if (!(RREG32(CP_HQD_ACTIVE) & 1))
4943 WREG32(CP_HQD_DEQUEUE_REQUEST, mqd->queue_state.cp_hqd_dequeue_request);
4944 WREG32(CP_HQD_PQ_RPTR, mqd->queue_state.cp_hqd_pq_rptr);
4945 WREG32(CP_HQD_PQ_WPTR, mqd->queue_state.cp_hqd_pq_wptr);
4948 /* set the pointer to the MQD */
4949 mqd->queue_state.cp_mqd_base_addr = mqd_gpu_addr & 0xfffffffc;
4950 mqd->queue_state.cp_mqd_base_addr_hi = upper_32_bits(mqd_gpu_addr);
4951 WREG32(CP_MQD_BASE_ADDR, mqd->queue_state.cp_mqd_base_addr);
4952 WREG32(CP_MQD_BASE_ADDR_HI, mqd->queue_state.cp_mqd_base_addr_hi);
4953 /* set MQD vmid to 0 */
4954 mqd->queue_state.cp_mqd_control = RREG32(CP_MQD_CONTROL);
4955 mqd->queue_state.cp_mqd_control &= ~MQD_VMID_MASK;
4956 WREG32(CP_MQD_CONTROL, mqd->queue_state.cp_mqd_control);
4958 /* set the pointer to the HQD, this is similar CP_RB0_BASE/_HI */
4959 hqd_gpu_addr = rdev->ring[idx].gpu_addr >> 8;
4960 mqd->queue_state.cp_hqd_pq_base = hqd_gpu_addr;
4961 mqd->queue_state.cp_hqd_pq_base_hi = upper_32_bits(hqd_gpu_addr);
4962 WREG32(CP_HQD_PQ_BASE, mqd->queue_state.cp_hqd_pq_base);
4963 WREG32(CP_HQD_PQ_BASE_HI, mqd->queue_state.cp_hqd_pq_base_hi);
4965 /* set up the HQD, this is similar to CP_RB0_CNTL */
4966 mqd->queue_state.cp_hqd_pq_control = RREG32(CP_HQD_PQ_CONTROL);
4967 mqd->queue_state.cp_hqd_pq_control &=
4968 ~(QUEUE_SIZE_MASK | RPTR_BLOCK_SIZE_MASK);
4970 mqd->queue_state.cp_hqd_pq_control |=
4971 order_base_2(rdev->ring[idx].ring_size / 8);
4972 mqd->queue_state.cp_hqd_pq_control |=
4973 (order_base_2(RADEON_GPU_PAGE_SIZE/8) << 8);
4975 mqd->queue_state.cp_hqd_pq_control |= BUF_SWAP_32BIT;
4977 mqd->queue_state.cp_hqd_pq_control &=
4978 ~(UNORD_DISPATCH | ROQ_PQ_IB_FLIP | PQ_VOLATILE);
4979 mqd->queue_state.cp_hqd_pq_control |=
4980 PRIV_STATE | KMD_QUEUE; /* assuming kernel queue control */
4981 WREG32(CP_HQD_PQ_CONTROL, mqd->queue_state.cp_hqd_pq_control);
4983 /* only used if CP_PQ_WPTR_POLL_CNTL.WPTR_POLL_EN=1 */
4985 wb_gpu_addr = rdev->wb.gpu_addr + CIK_WB_CP1_WPTR_OFFSET;
4987 wb_gpu_addr = rdev->wb.gpu_addr + CIK_WB_CP2_WPTR_OFFSET;
4988 mqd->queue_state.cp_hqd_pq_wptr_poll_addr = wb_gpu_addr & 0xfffffffc;
4989 mqd->queue_state.cp_hqd_pq_wptr_poll_addr_hi = upper_32_bits(wb_gpu_addr) & 0xffff;
4990 WREG32(CP_HQD_PQ_WPTR_POLL_ADDR, mqd->queue_state.cp_hqd_pq_wptr_poll_addr);
4991 WREG32(CP_HQD_PQ_WPTR_POLL_ADDR_HI,
4992 mqd->queue_state.cp_hqd_pq_wptr_poll_addr_hi);
4994 /* set the wb address wether it's enabled or not */
4996 wb_gpu_addr = rdev->wb.gpu_addr + RADEON_WB_CP1_RPTR_OFFSET;
4998 wb_gpu_addr = rdev->wb.gpu_addr + RADEON_WB_CP2_RPTR_OFFSET;
4999 mqd->queue_state.cp_hqd_pq_rptr_report_addr = wb_gpu_addr & 0xfffffffc;
5000 mqd->queue_state.cp_hqd_pq_rptr_report_addr_hi =
5001 upper_32_bits(wb_gpu_addr) & 0xffff;
5002 WREG32(CP_HQD_PQ_RPTR_REPORT_ADDR,
5003 mqd->queue_state.cp_hqd_pq_rptr_report_addr);
5004 WREG32(CP_HQD_PQ_RPTR_REPORT_ADDR_HI,
5005 mqd->queue_state.cp_hqd_pq_rptr_report_addr_hi);
5007 /* enable the doorbell if requested */
5009 mqd->queue_state.cp_hqd_pq_doorbell_control =
5010 RREG32(CP_HQD_PQ_DOORBELL_CONTROL);
5011 mqd->queue_state.cp_hqd_pq_doorbell_control &= ~DOORBELL_OFFSET_MASK;
5012 mqd->queue_state.cp_hqd_pq_doorbell_control |=
5013 DOORBELL_OFFSET(rdev->ring[idx].doorbell_index);
5014 mqd->queue_state.cp_hqd_pq_doorbell_control |= DOORBELL_EN;
5015 mqd->queue_state.cp_hqd_pq_doorbell_control &=
5016 ~(DOORBELL_SOURCE | DOORBELL_HIT);
5019 mqd->queue_state.cp_hqd_pq_doorbell_control = 0;
5021 WREG32(CP_HQD_PQ_DOORBELL_CONTROL,
5022 mqd->queue_state.cp_hqd_pq_doorbell_control);
5024 /* read and write pointers, similar to CP_RB0_WPTR/_RPTR */
5025 rdev->ring[idx].wptr = 0;
5026 mqd->queue_state.cp_hqd_pq_wptr = rdev->ring[idx].wptr;
5027 WREG32(CP_HQD_PQ_WPTR, mqd->queue_state.cp_hqd_pq_wptr);
5028 mqd->queue_state.cp_hqd_pq_rptr = RREG32(CP_HQD_PQ_RPTR);
5030 /* set the vmid for the queue */
5031 mqd->queue_state.cp_hqd_vmid = 0;
5032 WREG32(CP_HQD_VMID, mqd->queue_state.cp_hqd_vmid);
5034 /* activate the queue */
5035 mqd->queue_state.cp_hqd_active = 1;
5036 WREG32(CP_HQD_ACTIVE, mqd->queue_state.cp_hqd_active);
5038 cik_srbm_select(rdev, 0, 0, 0, 0);
5039 mutex_unlock(&rdev->srbm_mutex);
5041 radeon_bo_kunmap(rdev->ring[idx].mqd_obj);
5042 radeon_bo_unreserve(rdev->ring[idx].mqd_obj);
5044 rdev->ring[idx].ready = true;
5045 r = radeon_ring_test(rdev, idx, &rdev->ring[idx]);
5047 rdev->ring[idx].ready = false;
5053 static void cik_cp_enable(struct radeon_device *rdev, bool enable)
5055 cik_cp_gfx_enable(rdev, enable);
5056 cik_cp_compute_enable(rdev, enable);
5059 static int cik_cp_load_microcode(struct radeon_device *rdev)
5063 r = cik_cp_gfx_load_microcode(rdev);
5066 r = cik_cp_compute_load_microcode(rdev);
5073 static void cik_cp_fini(struct radeon_device *rdev)
5075 cik_cp_gfx_fini(rdev);
5076 cik_cp_compute_fini(rdev);
5079 static int cik_cp_resume(struct radeon_device *rdev)
5083 cik_enable_gui_idle_interrupt(rdev, false);
5085 r = cik_cp_load_microcode(rdev);
5089 r = cik_cp_gfx_resume(rdev);
5092 r = cik_cp_compute_resume(rdev);
5096 cik_enable_gui_idle_interrupt(rdev, true);
5101 static void cik_print_gpu_status_regs(struct radeon_device *rdev)
5103 dev_info(rdev->dev, " GRBM_STATUS=0x%08X\n",
5104 RREG32(GRBM_STATUS));
5105 dev_info(rdev->dev, " GRBM_STATUS2=0x%08X\n",
5106 RREG32(GRBM_STATUS2));
5107 dev_info(rdev->dev, " GRBM_STATUS_SE0=0x%08X\n",
5108 RREG32(GRBM_STATUS_SE0));
5109 dev_info(rdev->dev, " GRBM_STATUS_SE1=0x%08X\n",
5110 RREG32(GRBM_STATUS_SE1));
5111 dev_info(rdev->dev, " GRBM_STATUS_SE2=0x%08X\n",
5112 RREG32(GRBM_STATUS_SE2));
5113 dev_info(rdev->dev, " GRBM_STATUS_SE3=0x%08X\n",
5114 RREG32(GRBM_STATUS_SE3));
5115 dev_info(rdev->dev, " SRBM_STATUS=0x%08X\n",
5116 RREG32(SRBM_STATUS));
5117 dev_info(rdev->dev, " SRBM_STATUS2=0x%08X\n",
5118 RREG32(SRBM_STATUS2));
5119 dev_info(rdev->dev, " SDMA0_STATUS_REG = 0x%08X\n",
5120 RREG32(SDMA0_STATUS_REG + SDMA0_REGISTER_OFFSET));
5121 dev_info(rdev->dev, " SDMA1_STATUS_REG = 0x%08X\n",
5122 RREG32(SDMA0_STATUS_REG + SDMA1_REGISTER_OFFSET));
5123 dev_info(rdev->dev, " CP_STAT = 0x%08x\n", RREG32(CP_STAT));
5124 dev_info(rdev->dev, " CP_STALLED_STAT1 = 0x%08x\n",
5125 RREG32(CP_STALLED_STAT1));
5126 dev_info(rdev->dev, " CP_STALLED_STAT2 = 0x%08x\n",
5127 RREG32(CP_STALLED_STAT2));
5128 dev_info(rdev->dev, " CP_STALLED_STAT3 = 0x%08x\n",
5129 RREG32(CP_STALLED_STAT3));
5130 dev_info(rdev->dev, " CP_CPF_BUSY_STAT = 0x%08x\n",
5131 RREG32(CP_CPF_BUSY_STAT));
5132 dev_info(rdev->dev, " CP_CPF_STALLED_STAT1 = 0x%08x\n",
5133 RREG32(CP_CPF_STALLED_STAT1));
5134 dev_info(rdev->dev, " CP_CPF_STATUS = 0x%08x\n", RREG32(CP_CPF_STATUS));
5135 dev_info(rdev->dev, " CP_CPC_BUSY_STAT = 0x%08x\n", RREG32(CP_CPC_BUSY_STAT));
5136 dev_info(rdev->dev, " CP_CPC_STALLED_STAT1 = 0x%08x\n",
5137 RREG32(CP_CPC_STALLED_STAT1));
5138 dev_info(rdev->dev, " CP_CPC_STATUS = 0x%08x\n", RREG32(CP_CPC_STATUS));
5142 * cik_gpu_check_soft_reset - check which blocks are busy
5144 * @rdev: radeon_device pointer
5146 * Check which blocks are busy and return the relevant reset
5147 * mask to be used by cik_gpu_soft_reset().
5148 * Returns a mask of the blocks to be reset.
5150 u32 cik_gpu_check_soft_reset(struct radeon_device *rdev)
5156 tmp = RREG32(GRBM_STATUS);
5157 if (tmp & (PA_BUSY | SC_BUSY |
5158 BCI_BUSY | SX_BUSY |
5159 TA_BUSY | VGT_BUSY |
5161 GDS_BUSY | SPI_BUSY |
5162 IA_BUSY | IA_BUSY_NO_DMA))
5163 reset_mask |= RADEON_RESET_GFX;
5165 if (tmp & (CP_BUSY | CP_COHERENCY_BUSY))
5166 reset_mask |= RADEON_RESET_CP;
5169 tmp = RREG32(GRBM_STATUS2);
5171 reset_mask |= RADEON_RESET_RLC;
5173 /* SDMA0_STATUS_REG */
5174 tmp = RREG32(SDMA0_STATUS_REG + SDMA0_REGISTER_OFFSET);
5175 if (!(tmp & SDMA_IDLE))
5176 reset_mask |= RADEON_RESET_DMA;
5178 /* SDMA1_STATUS_REG */
5179 tmp = RREG32(SDMA0_STATUS_REG + SDMA1_REGISTER_OFFSET);
5180 if (!(tmp & SDMA_IDLE))
5181 reset_mask |= RADEON_RESET_DMA1;
5184 tmp = RREG32(SRBM_STATUS2);
5185 if (tmp & SDMA_BUSY)
5186 reset_mask |= RADEON_RESET_DMA;
5188 if (tmp & SDMA1_BUSY)
5189 reset_mask |= RADEON_RESET_DMA1;
5192 tmp = RREG32(SRBM_STATUS);
5195 reset_mask |= RADEON_RESET_IH;
5198 reset_mask |= RADEON_RESET_SEM;
5200 if (tmp & GRBM_RQ_PENDING)
5201 reset_mask |= RADEON_RESET_GRBM;
5204 reset_mask |= RADEON_RESET_VMC;
5206 if (tmp & (MCB_BUSY | MCB_NON_DISPLAY_BUSY |
5207 MCC_BUSY | MCD_BUSY))
5208 reset_mask |= RADEON_RESET_MC;
5210 if (evergreen_is_display_hung(rdev))
5211 reset_mask |= RADEON_RESET_DISPLAY;
5213 /* Skip MC reset as it's mostly likely not hung, just busy */
5214 if (reset_mask & RADEON_RESET_MC) {
5215 DRM_DEBUG("MC busy: 0x%08X, clearing.\n", reset_mask);
5216 reset_mask &= ~RADEON_RESET_MC;
5223 * cik_gpu_soft_reset - soft reset GPU
5225 * @rdev: radeon_device pointer
5226 * @reset_mask: mask of which blocks to reset
5228 * Soft reset the blocks specified in @reset_mask.
5230 static void cik_gpu_soft_reset(struct radeon_device *rdev, u32 reset_mask)
5232 struct evergreen_mc_save save;
5233 u32 grbm_soft_reset = 0, srbm_soft_reset = 0;
5236 if (reset_mask == 0)
5239 dev_info(rdev->dev, "GPU softreset: 0x%08X\n", reset_mask);
5241 cik_print_gpu_status_regs(rdev);
5242 dev_info(rdev->dev, " VM_CONTEXT1_PROTECTION_FAULT_ADDR 0x%08X\n",
5243 RREG32(VM_CONTEXT1_PROTECTION_FAULT_ADDR));
5244 dev_info(rdev->dev, " VM_CONTEXT1_PROTECTION_FAULT_STATUS 0x%08X\n",
5245 RREG32(VM_CONTEXT1_PROTECTION_FAULT_STATUS));
5254 /* Disable GFX parsing/prefetching */
5255 WREG32(CP_ME_CNTL, CP_ME_HALT | CP_PFP_HALT | CP_CE_HALT);
5257 /* Disable MEC parsing/prefetching */
5258 WREG32(CP_MEC_CNTL, MEC_ME1_HALT | MEC_ME2_HALT);
5260 if (reset_mask & RADEON_RESET_DMA) {
5262 tmp = RREG32(SDMA0_ME_CNTL + SDMA0_REGISTER_OFFSET);
5264 WREG32(SDMA0_ME_CNTL + SDMA0_REGISTER_OFFSET, tmp);
5266 if (reset_mask & RADEON_RESET_DMA1) {
5268 tmp = RREG32(SDMA0_ME_CNTL + SDMA1_REGISTER_OFFSET);
5270 WREG32(SDMA0_ME_CNTL + SDMA1_REGISTER_OFFSET, tmp);
5273 evergreen_mc_stop(rdev, &save);
5274 if (evergreen_mc_wait_for_idle(rdev)) {
5275 dev_warn(rdev->dev, "Wait for MC idle timedout !\n");
5278 if (reset_mask & (RADEON_RESET_GFX | RADEON_RESET_COMPUTE | RADEON_RESET_CP))
5279 grbm_soft_reset = SOFT_RESET_CP | SOFT_RESET_GFX;
5281 if (reset_mask & RADEON_RESET_CP) {
5282 grbm_soft_reset |= SOFT_RESET_CP;
5284 srbm_soft_reset |= SOFT_RESET_GRBM;
5287 if (reset_mask & RADEON_RESET_DMA)
5288 srbm_soft_reset |= SOFT_RESET_SDMA;
5290 if (reset_mask & RADEON_RESET_DMA1)
5291 srbm_soft_reset |= SOFT_RESET_SDMA1;
5293 if (reset_mask & RADEON_RESET_DISPLAY)
5294 srbm_soft_reset |= SOFT_RESET_DC;
5296 if (reset_mask & RADEON_RESET_RLC)
5297 grbm_soft_reset |= SOFT_RESET_RLC;
5299 if (reset_mask & RADEON_RESET_SEM)
5300 srbm_soft_reset |= SOFT_RESET_SEM;
5302 if (reset_mask & RADEON_RESET_IH)
5303 srbm_soft_reset |= SOFT_RESET_IH;
5305 if (reset_mask & RADEON_RESET_GRBM)
5306 srbm_soft_reset |= SOFT_RESET_GRBM;
5308 if (reset_mask & RADEON_RESET_VMC)
5309 srbm_soft_reset |= SOFT_RESET_VMC;
5311 if (!(rdev->flags & RADEON_IS_IGP)) {
5312 if (reset_mask & RADEON_RESET_MC)
5313 srbm_soft_reset |= SOFT_RESET_MC;
5316 if (grbm_soft_reset) {
5317 tmp = RREG32(GRBM_SOFT_RESET);
5318 tmp |= grbm_soft_reset;
5319 dev_info(rdev->dev, "GRBM_SOFT_RESET=0x%08X\n", tmp);
5320 WREG32(GRBM_SOFT_RESET, tmp);
5321 tmp = RREG32(GRBM_SOFT_RESET);
5325 tmp &= ~grbm_soft_reset;
5326 WREG32(GRBM_SOFT_RESET, tmp);
5327 tmp = RREG32(GRBM_SOFT_RESET);
5330 if (srbm_soft_reset) {
5331 tmp = RREG32(SRBM_SOFT_RESET);
5332 tmp |= srbm_soft_reset;
5333 dev_info(rdev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp);
5334 WREG32(SRBM_SOFT_RESET, tmp);
5335 tmp = RREG32(SRBM_SOFT_RESET);
5339 tmp &= ~srbm_soft_reset;
5340 WREG32(SRBM_SOFT_RESET, tmp);
5341 tmp = RREG32(SRBM_SOFT_RESET);
5344 /* Wait a little for things to settle down */
5347 evergreen_mc_resume(rdev, &save);
5350 cik_print_gpu_status_regs(rdev);
5353 struct kv_reset_save_regs {
5354 u32 gmcon_reng_execute;
5359 static void kv_save_regs_for_reset(struct radeon_device *rdev,
5360 struct kv_reset_save_regs *save)
5362 save->gmcon_reng_execute = RREG32(GMCON_RENG_EXECUTE);
5363 save->gmcon_misc = RREG32(GMCON_MISC);
5364 save->gmcon_misc3 = RREG32(GMCON_MISC3);
5366 WREG32(GMCON_RENG_EXECUTE, save->gmcon_reng_execute & ~RENG_EXECUTE_ON_PWR_UP);
5367 WREG32(GMCON_MISC, save->gmcon_misc & ~(RENG_EXECUTE_ON_REG_UPDATE |
5368 STCTRL_STUTTER_EN));
5371 static void kv_restore_regs_for_reset(struct radeon_device *rdev,
5372 struct kv_reset_save_regs *save)
5376 WREG32(GMCON_PGFSM_WRITE, 0);
5377 WREG32(GMCON_PGFSM_CONFIG, 0x200010ff);
5379 for (i = 0; i < 5; i++)
5380 WREG32(GMCON_PGFSM_WRITE, 0);
5382 WREG32(GMCON_PGFSM_WRITE, 0);
5383 WREG32(GMCON_PGFSM_CONFIG, 0x300010ff);
5385 for (i = 0; i < 5; i++)
5386 WREG32(GMCON_PGFSM_WRITE, 0);
5388 WREG32(GMCON_PGFSM_WRITE, 0x210000);
5389 WREG32(GMCON_PGFSM_CONFIG, 0xa00010ff);
5391 for (i = 0; i < 5; i++)
5392 WREG32(GMCON_PGFSM_WRITE, 0);
5394 WREG32(GMCON_PGFSM_WRITE, 0x21003);
5395 WREG32(GMCON_PGFSM_CONFIG, 0xb00010ff);
5397 for (i = 0; i < 5; i++)
5398 WREG32(GMCON_PGFSM_WRITE, 0);
5400 WREG32(GMCON_PGFSM_WRITE, 0x2b00);
5401 WREG32(GMCON_PGFSM_CONFIG, 0xc00010ff);
5403 for (i = 0; i < 5; i++)
5404 WREG32(GMCON_PGFSM_WRITE, 0);
5406 WREG32(GMCON_PGFSM_WRITE, 0);
5407 WREG32(GMCON_PGFSM_CONFIG, 0xd00010ff);
5409 for (i = 0; i < 5; i++)
5410 WREG32(GMCON_PGFSM_WRITE, 0);
5412 WREG32(GMCON_PGFSM_WRITE, 0x420000);
5413 WREG32(GMCON_PGFSM_CONFIG, 0x100010ff);
5415 for (i = 0; i < 5; i++)
5416 WREG32(GMCON_PGFSM_WRITE, 0);
5418 WREG32(GMCON_PGFSM_WRITE, 0x120202);
5419 WREG32(GMCON_PGFSM_CONFIG, 0x500010ff);
5421 for (i = 0; i < 5; i++)
5422 WREG32(GMCON_PGFSM_WRITE, 0);
5424 WREG32(GMCON_PGFSM_WRITE, 0x3e3e36);
5425 WREG32(GMCON_PGFSM_CONFIG, 0x600010ff);
5427 for (i = 0; i < 5; i++)
5428 WREG32(GMCON_PGFSM_WRITE, 0);
5430 WREG32(GMCON_PGFSM_WRITE, 0x373f3e);
5431 WREG32(GMCON_PGFSM_CONFIG, 0x700010ff);
5433 for (i = 0; i < 5; i++)
5434 WREG32(GMCON_PGFSM_WRITE, 0);
5436 WREG32(GMCON_PGFSM_WRITE, 0x3e1332);
5437 WREG32(GMCON_PGFSM_CONFIG, 0xe00010ff);
5439 WREG32(GMCON_MISC3, save->gmcon_misc3);
5440 WREG32(GMCON_MISC, save->gmcon_misc);
5441 WREG32(GMCON_RENG_EXECUTE, save->gmcon_reng_execute);
5444 static void cik_gpu_pci_config_reset(struct radeon_device *rdev)
5446 struct evergreen_mc_save save;
5447 struct kv_reset_save_regs kv_save = { 0 };
5450 dev_info(rdev->dev, "GPU pci config reset\n");
5458 /* Disable GFX parsing/prefetching */
5459 WREG32(CP_ME_CNTL, CP_ME_HALT | CP_PFP_HALT | CP_CE_HALT);
5461 /* Disable MEC parsing/prefetching */
5462 WREG32(CP_MEC_CNTL, MEC_ME1_HALT | MEC_ME2_HALT);
5465 tmp = RREG32(SDMA0_ME_CNTL + SDMA0_REGISTER_OFFSET);
5467 WREG32(SDMA0_ME_CNTL + SDMA0_REGISTER_OFFSET, tmp);
5469 tmp = RREG32(SDMA0_ME_CNTL + SDMA1_REGISTER_OFFSET);
5471 WREG32(SDMA0_ME_CNTL + SDMA1_REGISTER_OFFSET, tmp);
5472 /* XXX other engines? */
5474 /* halt the rlc, disable cp internal ints */
5479 /* disable mem access */
5480 evergreen_mc_stop(rdev, &save);
5481 if (evergreen_mc_wait_for_idle(rdev)) {
5482 dev_warn(rdev->dev, "Wait for MC idle timed out !\n");
5485 if (rdev->flags & RADEON_IS_IGP)
5486 kv_save_regs_for_reset(rdev, &kv_save);
5489 pci_clear_master(rdev->pdev);
5491 radeon_pci_config_reset(rdev);
5495 /* wait for asic to come out of reset */
5496 for (i = 0; i < rdev->usec_timeout; i++) {
5497 if (RREG32(CONFIG_MEMSIZE) != 0xffffffff)
5502 /* does asic init need to be run first??? */
5503 if (rdev->flags & RADEON_IS_IGP)
5504 kv_restore_regs_for_reset(rdev, &kv_save);
5508 * cik_asic_reset - soft reset GPU
5510 * @rdev: radeon_device pointer
5512 * Look up which blocks are hung and attempt
5514 * Returns 0 for success.
5516 int cik_asic_reset(struct radeon_device *rdev)
5520 reset_mask = cik_gpu_check_soft_reset(rdev);
5523 r600_set_bios_scratch_engine_hung(rdev, true);
5525 /* try soft reset */
5526 cik_gpu_soft_reset(rdev, reset_mask);
5528 reset_mask = cik_gpu_check_soft_reset(rdev);
5530 /* try pci config reset */
5531 if (reset_mask && radeon_hard_reset)
5532 cik_gpu_pci_config_reset(rdev);
5534 reset_mask = cik_gpu_check_soft_reset(rdev);
5537 r600_set_bios_scratch_engine_hung(rdev, false);
5543 * cik_gfx_is_lockup - check if the 3D engine is locked up
5545 * @rdev: radeon_device pointer
5546 * @ring: radeon_ring structure holding ring information
5548 * Check if the 3D engine is locked up (CIK).
5549 * Returns true if the engine is locked, false if not.
5551 bool cik_gfx_is_lockup(struct radeon_device *rdev, struct radeon_ring *ring)
5553 u32 reset_mask = cik_gpu_check_soft_reset(rdev);
5555 if (!(reset_mask & (RADEON_RESET_GFX |
5556 RADEON_RESET_COMPUTE |
5557 RADEON_RESET_CP))) {
5558 radeon_ring_lockup_update(rdev, ring);
5561 return radeon_ring_test_lockup(rdev, ring);
5566 * cik_mc_program - program the GPU memory controller
5568 * @rdev: radeon_device pointer
5570 * Set the location of vram, gart, and AGP in the GPU's
5571 * physical address space (CIK).
5573 static void cik_mc_program(struct radeon_device *rdev)
5575 struct evergreen_mc_save save;
5579 /* Initialize HDP */
5580 for (i = 0, j = 0; i < 32; i++, j += 0x18) {
5581 WREG32((0x2c14 + j), 0x00000000);
5582 WREG32((0x2c18 + j), 0x00000000);
5583 WREG32((0x2c1c + j), 0x00000000);
5584 WREG32((0x2c20 + j), 0x00000000);
5585 WREG32((0x2c24 + j), 0x00000000);
5587 WREG32(HDP_REG_COHERENCY_FLUSH_CNTL, 0);
5589 evergreen_mc_stop(rdev, &save);
5590 if (radeon_mc_wait_for_idle(rdev)) {
5591 dev_warn(rdev->dev, "Wait for MC idle timedout !\n");
5593 /* Lockout access through VGA aperture*/
5594 WREG32(VGA_HDP_CONTROL, VGA_MEMORY_DISABLE);
5595 /* Update configuration */
5596 WREG32(MC_VM_SYSTEM_APERTURE_LOW_ADDR,
5597 rdev->mc.vram_start >> 12);
5598 WREG32(MC_VM_SYSTEM_APERTURE_HIGH_ADDR,
5599 rdev->mc.vram_end >> 12);
5600 WREG32(MC_VM_SYSTEM_APERTURE_DEFAULT_ADDR,
5601 rdev->vram_scratch.gpu_addr >> 12);
5602 tmp = ((rdev->mc.vram_end >> 24) & 0xFFFF) << 16;
5603 tmp |= ((rdev->mc.vram_start >> 24) & 0xFFFF);
5604 WREG32(MC_VM_FB_LOCATION, tmp);
5605 /* XXX double check these! */
5606 WREG32(HDP_NONSURFACE_BASE, (rdev->mc.vram_start >> 8));
5607 WREG32(HDP_NONSURFACE_INFO, (2 << 7) | (1 << 30));
5608 WREG32(HDP_NONSURFACE_SIZE, 0x3FFFFFFF);
5609 WREG32(MC_VM_AGP_BASE, 0);
5610 WREG32(MC_VM_AGP_TOP, 0x0FFFFFFF);
5611 WREG32(MC_VM_AGP_BOT, 0x0FFFFFFF);
5612 if (radeon_mc_wait_for_idle(rdev)) {
5613 dev_warn(rdev->dev, "Wait for MC idle timedout !\n");
5615 evergreen_mc_resume(rdev, &save);
5616 /* we need to own VRAM, so turn off the VGA renderer here
5617 * to stop it overwriting our objects */
5618 rv515_vga_render_disable(rdev);
5622 * cik_mc_init - initialize the memory controller driver params
5624 * @rdev: radeon_device pointer
5626 * Look up the amount of vram, vram width, and decide how to place
5627 * vram and gart within the GPU's physical address space (CIK).
5628 * Returns 0 for success.
5630 static int cik_mc_init(struct radeon_device *rdev)
5633 int chansize, numchan;
5635 /* Get VRAM informations */
5636 rdev->mc.vram_is_ddr = true;
5637 tmp = RREG32(MC_ARB_RAMCFG);
5638 if (tmp & CHANSIZE_MASK) {
5643 tmp = RREG32(MC_SHARED_CHMAP);
5644 switch ((tmp & NOOFCHAN_MASK) >> NOOFCHAN_SHIFT) {
5674 rdev->mc.vram_width = numchan * chansize;
5675 /* Could aper size report 0 ? */
5676 rdev->mc.aper_base = pci_resource_start(rdev->pdev, 0);
5677 rdev->mc.aper_size = pci_resource_len(rdev->pdev, 0);
5678 /* size in MB on si */
5679 rdev->mc.mc_vram_size = RREG32(CONFIG_MEMSIZE) * 1024ULL * 1024ULL;
5680 rdev->mc.real_vram_size = RREG32(CONFIG_MEMSIZE) * 1024ULL * 1024ULL;
5681 rdev->mc.visible_vram_size = rdev->mc.aper_size;
5682 si_vram_gtt_location(rdev, &rdev->mc);
5683 radeon_update_bandwidth_info(rdev);
5690 * VMID 0 is the physical GPU addresses as used by the kernel.
5691 * VMIDs 1-15 are used for userspace clients and are handled
5692 * by the radeon vm/hsa code.
5695 * cik_pcie_gart_tlb_flush - gart tlb flush callback
5697 * @rdev: radeon_device pointer
5699 * Flush the TLB for the VMID 0 page table (CIK).
5701 void cik_pcie_gart_tlb_flush(struct radeon_device *rdev)
5703 /* flush hdp cache */
5704 WREG32(HDP_MEM_COHERENCY_FLUSH_CNTL, 0);
5706 /* bits 0-15 are the VM contexts0-15 */
5707 WREG32(VM_INVALIDATE_REQUEST, 0x1);
5711 * cik_pcie_gart_enable - gart enable
5713 * @rdev: radeon_device pointer
5715 * This sets up the TLBs, programs the page tables for VMID0,
5716 * sets up the hw for VMIDs 1-15 which are allocated on
5717 * demand, and sets up the global locations for the LDS, GDS,
5718 * and GPUVM for FSA64 clients (CIK).
5719 * Returns 0 for success, errors for failure.
5721 static int cik_pcie_gart_enable(struct radeon_device *rdev)
5725 if (rdev->gart.robj == NULL) {
5726 dev_err(rdev->dev, "No VRAM object for PCIE GART.\n");
5729 r = radeon_gart_table_vram_pin(rdev);
5732 /* Setup TLB control */
5733 WREG32(MC_VM_MX_L1_TLB_CNTL,
5736 ENABLE_L1_FRAGMENT_PROCESSING |
5737 SYSTEM_ACCESS_MODE_NOT_IN_SYS |
5738 ENABLE_ADVANCED_DRIVER_MODEL |
5739 SYSTEM_APERTURE_UNMAPPED_ACCESS_PASS_THRU);
5740 /* Setup L2 cache */
5741 WREG32(VM_L2_CNTL, ENABLE_L2_CACHE |
5742 ENABLE_L2_FRAGMENT_PROCESSING |
5743 ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE |
5744 ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE |
5745 EFFECTIVE_L2_QUEUE_SIZE(7) |
5746 CONTEXT1_IDENTITY_ACCESS_MODE(1));
5747 WREG32(VM_L2_CNTL2, INVALIDATE_ALL_L1_TLBS | INVALIDATE_L2_CACHE);
5748 WREG32(VM_L2_CNTL3, L2_CACHE_BIGK_ASSOCIATIVITY |
5750 L2_CACHE_BIGK_FRAGMENT_SIZE(4));
5751 /* setup context0 */
5752 WREG32(VM_CONTEXT0_PAGE_TABLE_START_ADDR, rdev->mc.gtt_start >> 12);
5753 WREG32(VM_CONTEXT0_PAGE_TABLE_END_ADDR, rdev->mc.gtt_end >> 12);
5754 WREG32(VM_CONTEXT0_PAGE_TABLE_BASE_ADDR, rdev->gart.table_addr >> 12);
5755 WREG32(VM_CONTEXT0_PROTECTION_FAULT_DEFAULT_ADDR,
5756 (u32)(rdev->dummy_page.addr >> 12));
5757 WREG32(VM_CONTEXT0_CNTL2, 0);
5758 WREG32(VM_CONTEXT0_CNTL, (ENABLE_CONTEXT | PAGE_TABLE_DEPTH(0) |
5759 RANGE_PROTECTION_FAULT_ENABLE_DEFAULT));
5765 /* restore context1-15 */
5766 /* set vm size, must be a multiple of 4 */
5767 WREG32(VM_CONTEXT1_PAGE_TABLE_START_ADDR, 0);
5768 WREG32(VM_CONTEXT1_PAGE_TABLE_END_ADDR, rdev->vm_manager.max_pfn);
5769 for (i = 1; i < 16; i++) {
5771 WREG32(VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + (i << 2),
5772 rdev->vm_manager.saved_table_addr[i]);
5774 WREG32(VM_CONTEXT8_PAGE_TABLE_BASE_ADDR + ((i - 8) << 2),
5775 rdev->vm_manager.saved_table_addr[i]);
5778 /* enable context1-15 */
5779 WREG32(VM_CONTEXT1_PROTECTION_FAULT_DEFAULT_ADDR,
5780 (u32)(rdev->dummy_page.addr >> 12));
5781 WREG32(VM_CONTEXT1_CNTL2, 4);
5782 WREG32(VM_CONTEXT1_CNTL, ENABLE_CONTEXT | PAGE_TABLE_DEPTH(1) |
5783 PAGE_TABLE_BLOCK_SIZE(radeon_vm_block_size - 9) |
5784 RANGE_PROTECTION_FAULT_ENABLE_INTERRUPT |
5785 RANGE_PROTECTION_FAULT_ENABLE_DEFAULT |
5786 DUMMY_PAGE_PROTECTION_FAULT_ENABLE_INTERRUPT |
5787 DUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT |
5788 PDE0_PROTECTION_FAULT_ENABLE_INTERRUPT |
5789 PDE0_PROTECTION_FAULT_ENABLE_DEFAULT |
5790 VALID_PROTECTION_FAULT_ENABLE_INTERRUPT |
5791 VALID_PROTECTION_FAULT_ENABLE_DEFAULT |
5792 READ_PROTECTION_FAULT_ENABLE_INTERRUPT |
5793 READ_PROTECTION_FAULT_ENABLE_DEFAULT |
5794 WRITE_PROTECTION_FAULT_ENABLE_INTERRUPT |
5795 WRITE_PROTECTION_FAULT_ENABLE_DEFAULT);
5797 if (rdev->family == CHIP_KAVERI) {
5798 u32 tmp = RREG32(CHUB_CONTROL);
5800 WREG32(CHUB_CONTROL, tmp);
5803 /* XXX SH_MEM regs */
5804 /* where to put LDS, scratch, GPUVM in FSA64 space */
5805 mutex_lock(&rdev->srbm_mutex);
5806 for (i = 0; i < 16; i++) {
5807 cik_srbm_select(rdev, 0, 0, 0, i);
5808 /* CP and shaders */
5809 WREG32(SH_MEM_CONFIG, 0);
5810 WREG32(SH_MEM_APE1_BASE, 1);
5811 WREG32(SH_MEM_APE1_LIMIT, 0);
5812 WREG32(SH_MEM_BASES, 0);
5814 WREG32(SDMA0_GFX_VIRTUAL_ADDR + SDMA0_REGISTER_OFFSET, 0);
5815 WREG32(SDMA0_GFX_APE1_CNTL + SDMA0_REGISTER_OFFSET, 0);
5816 WREG32(SDMA0_GFX_VIRTUAL_ADDR + SDMA1_REGISTER_OFFSET, 0);
5817 WREG32(SDMA0_GFX_APE1_CNTL + SDMA1_REGISTER_OFFSET, 0);
5818 /* XXX SDMA RLC - todo */
5820 cik_srbm_select(rdev, 0, 0, 0, 0);
5821 mutex_unlock(&rdev->srbm_mutex);
5823 cik_pcie_gart_tlb_flush(rdev);
5824 DRM_INFO("PCIE GART of %uM enabled (table at 0x%016llX).\n",
5825 (unsigned)(rdev->mc.gtt_size >> 20),
5826 (unsigned long long)rdev->gart.table_addr);
5827 rdev->gart.ready = true;
5832 * cik_pcie_gart_disable - gart disable
5834 * @rdev: radeon_device pointer
5836 * This disables all VM page table (CIK).
5838 static void cik_pcie_gart_disable(struct radeon_device *rdev)
5842 for (i = 1; i < 16; ++i) {
5845 reg = VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + (i << 2);
5847 reg = VM_CONTEXT8_PAGE_TABLE_BASE_ADDR + ((i - 8) << 2);
5848 rdev->vm_manager.saved_table_addr[i] = RREG32(reg);
5851 /* Disable all tables */
5852 WREG32(VM_CONTEXT0_CNTL, 0);
5853 WREG32(VM_CONTEXT1_CNTL, 0);
5854 /* Setup TLB control */
5855 WREG32(MC_VM_MX_L1_TLB_CNTL, SYSTEM_ACCESS_MODE_NOT_IN_SYS |
5856 SYSTEM_APERTURE_UNMAPPED_ACCESS_PASS_THRU);
5857 /* Setup L2 cache */
5859 ENABLE_L2_FRAGMENT_PROCESSING |
5860 ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE |
5861 ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE |
5862 EFFECTIVE_L2_QUEUE_SIZE(7) |
5863 CONTEXT1_IDENTITY_ACCESS_MODE(1));
5864 WREG32(VM_L2_CNTL2, 0);
5865 WREG32(VM_L2_CNTL3, L2_CACHE_BIGK_ASSOCIATIVITY |
5866 L2_CACHE_BIGK_FRAGMENT_SIZE(6));
5867 radeon_gart_table_vram_unpin(rdev);
5871 * cik_pcie_gart_fini - vm fini callback
5873 * @rdev: radeon_device pointer
5875 * Tears down the driver GART/VM setup (CIK).
5877 static void cik_pcie_gart_fini(struct radeon_device *rdev)
5879 cik_pcie_gart_disable(rdev);
5880 radeon_gart_table_vram_free(rdev);
5881 radeon_gart_fini(rdev);
5886 * cik_ib_parse - vm ib_parse callback
5888 * @rdev: radeon_device pointer
5889 * @ib: indirect buffer pointer
5891 * CIK uses hw IB checking so this is a nop (CIK).
5893 int cik_ib_parse(struct radeon_device *rdev, struct radeon_ib *ib)
5900 * VMID 0 is the physical GPU addresses as used by the kernel.
5901 * VMIDs 1-15 are used for userspace clients and are handled
5902 * by the radeon vm/hsa code.
5905 * cik_vm_init - cik vm init callback
5907 * @rdev: radeon_device pointer
5909 * Inits cik specific vm parameters (number of VMs, base of vram for
5910 * VMIDs 1-15) (CIK).
5911 * Returns 0 for success.
5913 int cik_vm_init(struct radeon_device *rdev)
5917 * VMID 0 is reserved for System
5918 * radeon graphics/compute will use VMIDs 1-7
5919 * amdkfd will use VMIDs 8-15
5921 rdev->vm_manager.nvm = RADEON_NUM_OF_VMIDS;
5922 /* base offset of vram pages */
5923 if (rdev->flags & RADEON_IS_IGP) {
5924 u64 tmp = RREG32(MC_VM_FB_OFFSET);
5926 rdev->vm_manager.vram_base_offset = tmp;
5928 rdev->vm_manager.vram_base_offset = 0;
5934 * cik_vm_fini - cik vm fini callback
5936 * @rdev: radeon_device pointer
5938 * Tear down any asic specific VM setup (CIK).
5940 void cik_vm_fini(struct radeon_device *rdev)
5945 * cik_vm_decode_fault - print human readable fault info
5947 * @rdev: radeon_device pointer
5948 * @status: VM_CONTEXT1_PROTECTION_FAULT_STATUS register value
5949 * @addr: VM_CONTEXT1_PROTECTION_FAULT_ADDR register value
5951 * Print human readable fault information (CIK).
5953 static void cik_vm_decode_fault(struct radeon_device *rdev,
5954 u32 status, u32 addr, u32 mc_client)
5957 u32 vmid = (status & FAULT_VMID_MASK) >> FAULT_VMID_SHIFT;
5958 u32 protections = (status & PROTECTIONS_MASK) >> PROTECTIONS_SHIFT;
5959 char block[5] = { mc_client >> 24, (mc_client >> 16) & 0xff,
5960 (mc_client >> 8) & 0xff, mc_client & 0xff, 0 };
5962 if (rdev->family == CHIP_HAWAII)
5963 mc_id = (status & HAWAII_MEMORY_CLIENT_ID_MASK) >> MEMORY_CLIENT_ID_SHIFT;
5965 mc_id = (status & MEMORY_CLIENT_ID_MASK) >> MEMORY_CLIENT_ID_SHIFT;
5967 printk("VM fault (0x%02x, vmid %d) at page %u, %s from '%s' (0x%08x) (%d)\n",
5968 protections, vmid, addr,
5969 (status & MEMORY_CLIENT_RW_MASK) ? "write" : "read",
5970 block, mc_client, mc_id);
5974 * cik_vm_flush - cik vm flush using the CP
5976 * @rdev: radeon_device pointer
5978 * Update the page table base and flush the VM TLB
5979 * using the CP (CIK).
5981 void cik_vm_flush(struct radeon_device *rdev, struct radeon_ring *ring,
5982 unsigned vm_id, uint64_t pd_addr)
5984 int usepfp = (ring->idx == RADEON_RING_TYPE_GFX_INDEX);
5986 radeon_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5987 radeon_ring_write(ring, (WRITE_DATA_ENGINE_SEL(usepfp) |
5988 WRITE_DATA_DST_SEL(0)));
5990 radeon_ring_write(ring,
5991 (VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + (vm_id << 2)) >> 2);
5993 radeon_ring_write(ring,
5994 (VM_CONTEXT8_PAGE_TABLE_BASE_ADDR + ((vm_id - 8) << 2)) >> 2);
5996 radeon_ring_write(ring, 0);
5997 radeon_ring_write(ring, pd_addr >> 12);
5999 /* update SH_MEM_* regs */
6000 radeon_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
6001 radeon_ring_write(ring, (WRITE_DATA_ENGINE_SEL(usepfp) |
6002 WRITE_DATA_DST_SEL(0)));
6003 radeon_ring_write(ring, SRBM_GFX_CNTL >> 2);
6004 radeon_ring_write(ring, 0);
6005 radeon_ring_write(ring, VMID(vm_id));
6007 radeon_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 6));
6008 radeon_ring_write(ring, (WRITE_DATA_ENGINE_SEL(usepfp) |
6009 WRITE_DATA_DST_SEL(0)));
6010 radeon_ring_write(ring, SH_MEM_BASES >> 2);
6011 radeon_ring_write(ring, 0);
6013 radeon_ring_write(ring, 0); /* SH_MEM_BASES */
6014 radeon_ring_write(ring, 0); /* SH_MEM_CONFIG */
6015 radeon_ring_write(ring, 1); /* SH_MEM_APE1_BASE */
6016 radeon_ring_write(ring, 0); /* SH_MEM_APE1_LIMIT */
6018 radeon_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
6019 radeon_ring_write(ring, (WRITE_DATA_ENGINE_SEL(usepfp) |
6020 WRITE_DATA_DST_SEL(0)));
6021 radeon_ring_write(ring, SRBM_GFX_CNTL >> 2);
6022 radeon_ring_write(ring, 0);
6023 radeon_ring_write(ring, VMID(0));
6026 cik_hdp_flush_cp_ring_emit(rdev, ring->idx);
6028 /* bits 0-15 are the VM contexts0-15 */
6029 radeon_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
6030 radeon_ring_write(ring, (WRITE_DATA_ENGINE_SEL(usepfp) |
6031 WRITE_DATA_DST_SEL(0)));
6032 radeon_ring_write(ring, VM_INVALIDATE_REQUEST >> 2);
6033 radeon_ring_write(ring, 0);
6034 radeon_ring_write(ring, 1 << vm_id);
6036 /* wait for the invalidate to complete */
6037 radeon_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
6038 radeon_ring_write(ring, (WAIT_REG_MEM_OPERATION(0) | /* wait */
6039 WAIT_REG_MEM_FUNCTION(0) | /* always */
6040 WAIT_REG_MEM_ENGINE(0))); /* me */
6041 radeon_ring_write(ring, VM_INVALIDATE_REQUEST >> 2);
6042 radeon_ring_write(ring, 0);
6043 radeon_ring_write(ring, 0); /* ref */
6044 radeon_ring_write(ring, 0); /* mask */
6045 radeon_ring_write(ring, 0x20); /* poll interval */
6047 /* compute doesn't have PFP */
6049 /* sync PFP to ME, otherwise we might get invalid PFP reads */
6050 radeon_ring_write(ring, PACKET3(PACKET3_PFP_SYNC_ME, 0));
6051 radeon_ring_write(ring, 0x0);
6057 * The RLC is a multi-purpose microengine that handles a
6058 * variety of functions, the most important of which is
6059 * the interrupt controller.
6061 static void cik_enable_gui_idle_interrupt(struct radeon_device *rdev,
6064 u32 tmp = RREG32(CP_INT_CNTL_RING0);
6067 tmp |= (CNTX_BUSY_INT_ENABLE | CNTX_EMPTY_INT_ENABLE);
6069 tmp &= ~(CNTX_BUSY_INT_ENABLE | CNTX_EMPTY_INT_ENABLE);
6070 WREG32(CP_INT_CNTL_RING0, tmp);
6073 static void cik_enable_lbpw(struct radeon_device *rdev, bool enable)
6077 tmp = RREG32(RLC_LB_CNTL);
6079 tmp |= LOAD_BALANCE_ENABLE;
6081 tmp &= ~LOAD_BALANCE_ENABLE;
6082 WREG32(RLC_LB_CNTL, tmp);
6085 static void cik_wait_for_rlc_serdes(struct radeon_device *rdev)
6090 mutex_lock(&rdev->grbm_idx_mutex);
6091 for (i = 0; i < rdev->config.cik.max_shader_engines; i++) {
6092 for (j = 0; j < rdev->config.cik.max_sh_per_se; j++) {
6093 cik_select_se_sh(rdev, i, j);
6094 for (k = 0; k < rdev->usec_timeout; k++) {
6095 if (RREG32(RLC_SERDES_CU_MASTER_BUSY) == 0)
6101 cik_select_se_sh(rdev, 0xffffffff, 0xffffffff);
6102 mutex_unlock(&rdev->grbm_idx_mutex);
6104 mask = SE_MASTER_BUSY_MASK | GC_MASTER_BUSY | TC0_MASTER_BUSY | TC1_MASTER_BUSY;
6105 for (k = 0; k < rdev->usec_timeout; k++) {
6106 if ((RREG32(RLC_SERDES_NONCU_MASTER_BUSY) & mask) == 0)
6112 static void cik_update_rlc(struct radeon_device *rdev, u32 rlc)
6116 tmp = RREG32(RLC_CNTL);
6118 WREG32(RLC_CNTL, rlc);
6121 static u32 cik_halt_rlc(struct radeon_device *rdev)
6125 orig = data = RREG32(RLC_CNTL);
6127 if (data & RLC_ENABLE) {
6130 data &= ~RLC_ENABLE;
6131 WREG32(RLC_CNTL, data);
6133 for (i = 0; i < rdev->usec_timeout; i++) {
6134 if ((RREG32(RLC_GPM_STAT) & RLC_GPM_BUSY) == 0)
6139 cik_wait_for_rlc_serdes(rdev);
6145 void cik_enter_rlc_safe_mode(struct radeon_device *rdev)
6149 tmp = REQ | MESSAGE(MSG_ENTER_RLC_SAFE_MODE);
6150 WREG32(RLC_GPR_REG2, tmp);
6152 mask = GFX_POWER_STATUS | GFX_CLOCK_STATUS;
6153 for (i = 0; i < rdev->usec_timeout; i++) {
6154 if ((RREG32(RLC_GPM_STAT) & mask) == mask)
6159 for (i = 0; i < rdev->usec_timeout; i++) {
6160 if ((RREG32(RLC_GPR_REG2) & REQ) == 0)
6166 void cik_exit_rlc_safe_mode(struct radeon_device *rdev)
6170 tmp = REQ | MESSAGE(MSG_EXIT_RLC_SAFE_MODE);
6171 WREG32(RLC_GPR_REG2, tmp);
6175 * cik_rlc_stop - stop the RLC ME
6177 * @rdev: radeon_device pointer
6179 * Halt the RLC ME (MicroEngine) (CIK).
6181 static void cik_rlc_stop(struct radeon_device *rdev)
6183 WREG32(RLC_CNTL, 0);
6185 cik_enable_gui_idle_interrupt(rdev, false);
6187 cik_wait_for_rlc_serdes(rdev);
6191 * cik_rlc_start - start the RLC ME
6193 * @rdev: radeon_device pointer
6195 * Unhalt the RLC ME (MicroEngine) (CIK).
6197 static void cik_rlc_start(struct radeon_device *rdev)
6199 WREG32(RLC_CNTL, RLC_ENABLE);
6201 cik_enable_gui_idle_interrupt(rdev, true);
6207 * cik_rlc_resume - setup the RLC hw
6209 * @rdev: radeon_device pointer
6211 * Initialize the RLC registers, load the ucode,
6212 * and start the RLC (CIK).
6213 * Returns 0 for success, -EINVAL if the ucode is not available.
6215 static int cik_rlc_resume(struct radeon_device *rdev)
6225 tmp = RREG32(RLC_CGCG_CGLS_CTRL) & 0xfffffffc;
6226 WREG32(RLC_CGCG_CGLS_CTRL, tmp);
6234 WREG32(RLC_LB_CNTR_INIT, 0);
6235 WREG32(RLC_LB_CNTR_MAX, 0x00008000);
6237 mutex_lock(&rdev->grbm_idx_mutex);
6238 cik_select_se_sh(rdev, 0xffffffff, 0xffffffff);
6239 WREG32(RLC_LB_INIT_CU_MASK, 0xffffffff);
6240 WREG32(RLC_LB_PARAMS, 0x00600408);
6241 WREG32(RLC_LB_CNTL, 0x80000004);
6242 mutex_unlock(&rdev->grbm_idx_mutex);
6244 WREG32(RLC_MC_CNTL, 0);
6245 WREG32(RLC_UCODE_CNTL, 0);
6248 const struct rlc_firmware_header_v1_0 *hdr =
6249 (const struct rlc_firmware_header_v1_0 *)rdev->rlc_fw->data;
6250 const __le32 *fw_data = (const __le32 *)
6251 (rdev->rlc_fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes));
6253 radeon_ucode_print_rlc_hdr(&hdr->header);
6255 size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
6256 WREG32(RLC_GPM_UCODE_ADDR, 0);
6257 for (i = 0; i < size; i++)
6258 WREG32(RLC_GPM_UCODE_DATA, le32_to_cpup(fw_data++));
6259 WREG32(RLC_GPM_UCODE_ADDR, le32_to_cpu(hdr->header.ucode_version));
6261 const __be32 *fw_data;
6263 switch (rdev->family) {
6267 size = BONAIRE_RLC_UCODE_SIZE;
6270 size = KV_RLC_UCODE_SIZE;
6273 size = KB_RLC_UCODE_SIZE;
6276 size = ML_RLC_UCODE_SIZE;
6280 fw_data = (const __be32 *)rdev->rlc_fw->data;
6281 WREG32(RLC_GPM_UCODE_ADDR, 0);
6282 for (i = 0; i < size; i++)
6283 WREG32(RLC_GPM_UCODE_DATA, be32_to_cpup(fw_data++));
6284 WREG32(RLC_GPM_UCODE_ADDR, 0);
6287 /* XXX - find out what chips support lbpw */
6288 cik_enable_lbpw(rdev, false);
6290 if (rdev->family == CHIP_BONAIRE)
6291 WREG32(RLC_DRIVER_DMA_STATUS, 0);
6293 cik_rlc_start(rdev);
6298 static void cik_enable_cgcg(struct radeon_device *rdev, bool enable)
6300 u32 data, orig, tmp, tmp2;
6302 orig = data = RREG32(RLC_CGCG_CGLS_CTRL);
6304 if (enable && (rdev->cg_flags & RADEON_CG_SUPPORT_GFX_CGCG)) {
6305 cik_enable_gui_idle_interrupt(rdev, true);
6307 tmp = cik_halt_rlc(rdev);
6309 mutex_lock(&rdev->grbm_idx_mutex);
6310 cik_select_se_sh(rdev, 0xffffffff, 0xffffffff);
6311 WREG32(RLC_SERDES_WR_CU_MASTER_MASK, 0xffffffff);
6312 WREG32(RLC_SERDES_WR_NONCU_MASTER_MASK, 0xffffffff);
6313 tmp2 = BPM_ADDR_MASK | CGCG_OVERRIDE_0 | CGLS_ENABLE;
6314 WREG32(RLC_SERDES_WR_CTRL, tmp2);
6315 mutex_unlock(&rdev->grbm_idx_mutex);
6317 cik_update_rlc(rdev, tmp);
6319 data |= CGCG_EN | CGLS_EN;
6321 cik_enable_gui_idle_interrupt(rdev, false);
6323 RREG32(CB_CGTT_SCLK_CTRL);
6324 RREG32(CB_CGTT_SCLK_CTRL);
6325 RREG32(CB_CGTT_SCLK_CTRL);
6326 RREG32(CB_CGTT_SCLK_CTRL);
6328 data &= ~(CGCG_EN | CGLS_EN);
6332 WREG32(RLC_CGCG_CGLS_CTRL, data);
6336 static void cik_enable_mgcg(struct radeon_device *rdev, bool enable)
6338 u32 data, orig, tmp = 0;
6340 if (enable && (rdev->cg_flags & RADEON_CG_SUPPORT_GFX_MGCG)) {
6341 if (rdev->cg_flags & RADEON_CG_SUPPORT_GFX_MGLS) {
6342 if (rdev->cg_flags & RADEON_CG_SUPPORT_GFX_CP_LS) {
6343 orig = data = RREG32(CP_MEM_SLP_CNTL);
6344 data |= CP_MEM_LS_EN;
6346 WREG32(CP_MEM_SLP_CNTL, data);
6350 orig = data = RREG32(RLC_CGTT_MGCG_OVERRIDE);
6354 WREG32(RLC_CGTT_MGCG_OVERRIDE, data);
6356 tmp = cik_halt_rlc(rdev);
6358 mutex_lock(&rdev->grbm_idx_mutex);
6359 cik_select_se_sh(rdev, 0xffffffff, 0xffffffff);
6360 WREG32(RLC_SERDES_WR_CU_MASTER_MASK, 0xffffffff);
6361 WREG32(RLC_SERDES_WR_NONCU_MASTER_MASK, 0xffffffff);
6362 data = BPM_ADDR_MASK | MGCG_OVERRIDE_0;
6363 WREG32(RLC_SERDES_WR_CTRL, data);
6364 mutex_unlock(&rdev->grbm_idx_mutex);
6366 cik_update_rlc(rdev, tmp);
6368 if (rdev->cg_flags & RADEON_CG_SUPPORT_GFX_CGTS) {
6369 orig = data = RREG32(CGTS_SM_CTRL_REG);
6370 data &= ~SM_MODE_MASK;
6371 data |= SM_MODE(0x2);
6372 data |= SM_MODE_ENABLE;
6373 data &= ~CGTS_OVERRIDE;
6374 if ((rdev->cg_flags & RADEON_CG_SUPPORT_GFX_MGLS) &&
6375 (rdev->cg_flags & RADEON_CG_SUPPORT_GFX_CGTS_LS))
6376 data &= ~CGTS_LS_OVERRIDE;
6377 data &= ~ON_MONITOR_ADD_MASK;
6378 data |= ON_MONITOR_ADD_EN;
6379 data |= ON_MONITOR_ADD(0x96);
6381 WREG32(CGTS_SM_CTRL_REG, data);
6384 orig = data = RREG32(RLC_CGTT_MGCG_OVERRIDE);
6387 WREG32(RLC_CGTT_MGCG_OVERRIDE, data);
6389 data = RREG32(RLC_MEM_SLP_CNTL);
6390 if (data & RLC_MEM_LS_EN) {
6391 data &= ~RLC_MEM_LS_EN;
6392 WREG32(RLC_MEM_SLP_CNTL, data);
6395 data = RREG32(CP_MEM_SLP_CNTL);
6396 if (data & CP_MEM_LS_EN) {
6397 data &= ~CP_MEM_LS_EN;
6398 WREG32(CP_MEM_SLP_CNTL, data);
6401 orig = data = RREG32(CGTS_SM_CTRL_REG);
6402 data |= CGTS_OVERRIDE | CGTS_LS_OVERRIDE;
6404 WREG32(CGTS_SM_CTRL_REG, data);
6406 tmp = cik_halt_rlc(rdev);
6408 mutex_lock(&rdev->grbm_idx_mutex);
6409 cik_select_se_sh(rdev, 0xffffffff, 0xffffffff);
6410 WREG32(RLC_SERDES_WR_CU_MASTER_MASK, 0xffffffff);
6411 WREG32(RLC_SERDES_WR_NONCU_MASTER_MASK, 0xffffffff);
6412 data = BPM_ADDR_MASK | MGCG_OVERRIDE_1;
6413 WREG32(RLC_SERDES_WR_CTRL, data);
6414 mutex_unlock(&rdev->grbm_idx_mutex);
6416 cik_update_rlc(rdev, tmp);
6420 static const u32 mc_cg_registers[] =
6433 static void cik_enable_mc_ls(struct radeon_device *rdev,
6439 for (i = 0; i < ARRAY_SIZE(mc_cg_registers); i++) {
6440 orig = data = RREG32(mc_cg_registers[i]);
6441 if (enable && (rdev->cg_flags & RADEON_CG_SUPPORT_MC_LS))
6442 data |= MC_LS_ENABLE;
6444 data &= ~MC_LS_ENABLE;
6446 WREG32(mc_cg_registers[i], data);
6450 static void cik_enable_mc_mgcg(struct radeon_device *rdev,
6456 for (i = 0; i < ARRAY_SIZE(mc_cg_registers); i++) {
6457 orig = data = RREG32(mc_cg_registers[i]);
6458 if (enable && (rdev->cg_flags & RADEON_CG_SUPPORT_MC_MGCG))
6459 data |= MC_CG_ENABLE;
6461 data &= ~MC_CG_ENABLE;
6463 WREG32(mc_cg_registers[i], data);
6467 static void cik_enable_sdma_mgcg(struct radeon_device *rdev,
6472 if (enable && (rdev->cg_flags & RADEON_CG_SUPPORT_SDMA_MGCG)) {
6473 WREG32(SDMA0_CLK_CTRL + SDMA0_REGISTER_OFFSET, 0x00000100);
6474 WREG32(SDMA0_CLK_CTRL + SDMA1_REGISTER_OFFSET, 0x00000100);
6476 orig = data = RREG32(SDMA0_CLK_CTRL + SDMA0_REGISTER_OFFSET);
6479 WREG32(SDMA0_CLK_CTRL + SDMA0_REGISTER_OFFSET, data);
6481 orig = data = RREG32(SDMA0_CLK_CTRL + SDMA1_REGISTER_OFFSET);
6484 WREG32(SDMA0_CLK_CTRL + SDMA1_REGISTER_OFFSET, data);
6488 static void cik_enable_sdma_mgls(struct radeon_device *rdev,
6493 if (enable && (rdev->cg_flags & RADEON_CG_SUPPORT_SDMA_LS)) {
6494 orig = data = RREG32(SDMA0_POWER_CNTL + SDMA0_REGISTER_OFFSET);
6497 WREG32(SDMA0_POWER_CNTL + SDMA0_REGISTER_OFFSET, data);
6499 orig = data = RREG32(SDMA0_POWER_CNTL + SDMA1_REGISTER_OFFSET);
6502 WREG32(SDMA0_POWER_CNTL + SDMA1_REGISTER_OFFSET, data);
6504 orig = data = RREG32(SDMA0_POWER_CNTL + SDMA0_REGISTER_OFFSET);
6507 WREG32(SDMA0_POWER_CNTL + SDMA0_REGISTER_OFFSET, data);
6509 orig = data = RREG32(SDMA0_POWER_CNTL + SDMA1_REGISTER_OFFSET);
6512 WREG32(SDMA0_POWER_CNTL + SDMA1_REGISTER_OFFSET, data);
6516 static void cik_enable_uvd_mgcg(struct radeon_device *rdev,
6521 if (enable && (rdev->cg_flags & RADEON_CG_SUPPORT_UVD_MGCG)) {
6522 data = RREG32_UVD_CTX(UVD_CGC_MEM_CTRL);
6524 WREG32_UVD_CTX(UVD_CGC_MEM_CTRL, data);
6526 orig = data = RREG32(UVD_CGC_CTRL);
6529 WREG32(UVD_CGC_CTRL, data);
6531 data = RREG32_UVD_CTX(UVD_CGC_MEM_CTRL);
6533 WREG32_UVD_CTX(UVD_CGC_MEM_CTRL, data);
6535 orig = data = RREG32(UVD_CGC_CTRL);
6538 WREG32(UVD_CGC_CTRL, data);
6542 static void cik_enable_bif_mgls(struct radeon_device *rdev,
6547 orig = data = RREG32_PCIE_PORT(PCIE_CNTL2);
6549 if (enable && (rdev->cg_flags & RADEON_CG_SUPPORT_BIF_LS))
6550 data |= SLV_MEM_LS_EN | MST_MEM_LS_EN |
6551 REPLAY_MEM_LS_EN | SLV_MEM_AGGRESSIVE_LS_EN;
6553 data &= ~(SLV_MEM_LS_EN | MST_MEM_LS_EN |
6554 REPLAY_MEM_LS_EN | SLV_MEM_AGGRESSIVE_LS_EN);
6557 WREG32_PCIE_PORT(PCIE_CNTL2, data);
6560 static void cik_enable_hdp_mgcg(struct radeon_device *rdev,
6565 orig = data = RREG32(HDP_HOST_PATH_CNTL);
6567 if (enable && (rdev->cg_flags & RADEON_CG_SUPPORT_HDP_MGCG))
6568 data &= ~CLOCK_GATING_DIS;
6570 data |= CLOCK_GATING_DIS;
6573 WREG32(HDP_HOST_PATH_CNTL, data);
6576 static void cik_enable_hdp_ls(struct radeon_device *rdev,
6581 orig = data = RREG32(HDP_MEM_POWER_LS);
6583 if (enable && (rdev->cg_flags & RADEON_CG_SUPPORT_HDP_LS))
6584 data |= HDP_LS_ENABLE;
6586 data &= ~HDP_LS_ENABLE;
6589 WREG32(HDP_MEM_POWER_LS, data);
6592 void cik_update_cg(struct radeon_device *rdev,
6593 u32 block, bool enable)
6596 if (block & RADEON_CG_BLOCK_GFX) {
6597 cik_enable_gui_idle_interrupt(rdev, false);
6598 /* order matters! */
6600 cik_enable_mgcg(rdev, true);
6601 cik_enable_cgcg(rdev, true);
6603 cik_enable_cgcg(rdev, false);
6604 cik_enable_mgcg(rdev, false);
6606 cik_enable_gui_idle_interrupt(rdev, true);
6609 if (block & RADEON_CG_BLOCK_MC) {
6610 if (!(rdev->flags & RADEON_IS_IGP)) {
6611 cik_enable_mc_mgcg(rdev, enable);
6612 cik_enable_mc_ls(rdev, enable);
6616 if (block & RADEON_CG_BLOCK_SDMA) {
6617 cik_enable_sdma_mgcg(rdev, enable);
6618 cik_enable_sdma_mgls(rdev, enable);
6621 if (block & RADEON_CG_BLOCK_BIF) {
6622 cik_enable_bif_mgls(rdev, enable);
6625 if (block & RADEON_CG_BLOCK_UVD) {
6627 cik_enable_uvd_mgcg(rdev, enable);
6630 if (block & RADEON_CG_BLOCK_HDP) {
6631 cik_enable_hdp_mgcg(rdev, enable);
6632 cik_enable_hdp_ls(rdev, enable);
6635 if (block & RADEON_CG_BLOCK_VCE) {
6636 vce_v2_0_enable_mgcg(rdev, enable);
6640 static void cik_init_cg(struct radeon_device *rdev)
6643 cik_update_cg(rdev, RADEON_CG_BLOCK_GFX, true);
6646 si_init_uvd_internal_cg(rdev);
6648 cik_update_cg(rdev, (RADEON_CG_BLOCK_MC |
6649 RADEON_CG_BLOCK_SDMA |
6650 RADEON_CG_BLOCK_BIF |
6651 RADEON_CG_BLOCK_UVD |
6652 RADEON_CG_BLOCK_HDP), true);
6655 static void cik_fini_cg(struct radeon_device *rdev)
6657 cik_update_cg(rdev, (RADEON_CG_BLOCK_MC |
6658 RADEON_CG_BLOCK_SDMA |
6659 RADEON_CG_BLOCK_BIF |
6660 RADEON_CG_BLOCK_UVD |
6661 RADEON_CG_BLOCK_HDP), false);
6663 cik_update_cg(rdev, RADEON_CG_BLOCK_GFX, false);
6666 static void cik_enable_sck_slowdown_on_pu(struct radeon_device *rdev,
6671 orig = data = RREG32(RLC_PG_CNTL);
6672 if (enable && (rdev->pg_flags & RADEON_PG_SUPPORT_RLC_SMU_HS))
6673 data |= SMU_CLK_SLOWDOWN_ON_PU_ENABLE;
6675 data &= ~SMU_CLK_SLOWDOWN_ON_PU_ENABLE;
6677 WREG32(RLC_PG_CNTL, data);
6680 static void cik_enable_sck_slowdown_on_pd(struct radeon_device *rdev,
6685 orig = data = RREG32(RLC_PG_CNTL);
6686 if (enable && (rdev->pg_flags & RADEON_PG_SUPPORT_RLC_SMU_HS))
6687 data |= SMU_CLK_SLOWDOWN_ON_PD_ENABLE;
6689 data &= ~SMU_CLK_SLOWDOWN_ON_PD_ENABLE;
6691 WREG32(RLC_PG_CNTL, data);
6694 static void cik_enable_cp_pg(struct radeon_device *rdev, bool enable)
6698 orig = data = RREG32(RLC_PG_CNTL);
6699 if (enable && (rdev->pg_flags & RADEON_PG_SUPPORT_CP))
6700 data &= ~DISABLE_CP_PG;
6702 data |= DISABLE_CP_PG;
6704 WREG32(RLC_PG_CNTL, data);
6707 static void cik_enable_gds_pg(struct radeon_device *rdev, bool enable)
6711 orig = data = RREG32(RLC_PG_CNTL);
6712 if (enable && (rdev->pg_flags & RADEON_PG_SUPPORT_GDS))
6713 data &= ~DISABLE_GDS_PG;
6715 data |= DISABLE_GDS_PG;
6717 WREG32(RLC_PG_CNTL, data);
6720 #define CP_ME_TABLE_SIZE 96
6721 #define CP_ME_TABLE_OFFSET 2048
6722 #define CP_MEC_TABLE_OFFSET 4096
6724 void cik_init_cp_pg_table(struct radeon_device *rdev)
6726 volatile u32 *dst_ptr;
6727 int me, i, max_me = 4;
6729 u32 table_offset, table_size;
6731 if (rdev->family == CHIP_KAVERI)
6734 if (rdev->rlc.cp_table_ptr == NULL)
6737 /* write the cp table buffer */
6738 dst_ptr = rdev->rlc.cp_table_ptr;
6739 for (me = 0; me < max_me; me++) {
6741 const __le32 *fw_data;
6742 const struct gfx_firmware_header_v1_0 *hdr;
6745 hdr = (const struct gfx_firmware_header_v1_0 *)rdev->ce_fw->data;
6746 fw_data = (const __le32 *)
6747 (rdev->ce_fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes));
6748 table_offset = le32_to_cpu(hdr->jt_offset);
6749 table_size = le32_to_cpu(hdr->jt_size);
6750 } else if (me == 1) {
6751 hdr = (const struct gfx_firmware_header_v1_0 *)rdev->pfp_fw->data;
6752 fw_data = (const __le32 *)
6753 (rdev->pfp_fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes));
6754 table_offset = le32_to_cpu(hdr->jt_offset);
6755 table_size = le32_to_cpu(hdr->jt_size);
6756 } else if (me == 2) {
6757 hdr = (const struct gfx_firmware_header_v1_0 *)rdev->me_fw->data;
6758 fw_data = (const __le32 *)
6759 (rdev->me_fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes));
6760 table_offset = le32_to_cpu(hdr->jt_offset);
6761 table_size = le32_to_cpu(hdr->jt_size);
6762 } else if (me == 3) {
6763 hdr = (const struct gfx_firmware_header_v1_0 *)rdev->mec_fw->data;
6764 fw_data = (const __le32 *)
6765 (rdev->mec_fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes));
6766 table_offset = le32_to_cpu(hdr->jt_offset);
6767 table_size = le32_to_cpu(hdr->jt_size);
6769 hdr = (const struct gfx_firmware_header_v1_0 *)rdev->mec2_fw->data;
6770 fw_data = (const __le32 *)
6771 (rdev->mec2_fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes));
6772 table_offset = le32_to_cpu(hdr->jt_offset);
6773 table_size = le32_to_cpu(hdr->jt_size);
6776 for (i = 0; i < table_size; i ++) {
6777 dst_ptr[bo_offset + i] =
6778 cpu_to_le32(le32_to_cpu(fw_data[table_offset + i]));
6780 bo_offset += table_size;
6782 const __be32 *fw_data;
6783 table_size = CP_ME_TABLE_SIZE;
6786 fw_data = (const __be32 *)rdev->ce_fw->data;
6787 table_offset = CP_ME_TABLE_OFFSET;
6788 } else if (me == 1) {
6789 fw_data = (const __be32 *)rdev->pfp_fw->data;
6790 table_offset = CP_ME_TABLE_OFFSET;
6791 } else if (me == 2) {
6792 fw_data = (const __be32 *)rdev->me_fw->data;
6793 table_offset = CP_ME_TABLE_OFFSET;
6795 fw_data = (const __be32 *)rdev->mec_fw->data;
6796 table_offset = CP_MEC_TABLE_OFFSET;
6799 for (i = 0; i < table_size; i ++) {
6800 dst_ptr[bo_offset + i] =
6801 cpu_to_le32(be32_to_cpu(fw_data[table_offset + i]));
6803 bo_offset += table_size;
6808 static void cik_enable_gfx_cgpg(struct radeon_device *rdev,
6813 if (enable && (rdev->pg_flags & RADEON_PG_SUPPORT_GFX_PG)) {
6814 orig = data = RREG32(RLC_PG_CNTL);
6815 data |= GFX_PG_ENABLE;
6817 WREG32(RLC_PG_CNTL, data);
6819 orig = data = RREG32(RLC_AUTO_PG_CTRL);
6822 WREG32(RLC_AUTO_PG_CTRL, data);
6824 orig = data = RREG32(RLC_PG_CNTL);
6825 data &= ~GFX_PG_ENABLE;
6827 WREG32(RLC_PG_CNTL, data);
6829 orig = data = RREG32(RLC_AUTO_PG_CTRL);
6830 data &= ~AUTO_PG_EN;
6832 WREG32(RLC_AUTO_PG_CTRL, data);
6834 data = RREG32(DB_RENDER_CONTROL);
6838 static u32 cik_get_cu_active_bitmap(struct radeon_device *rdev, u32 se, u32 sh)
6840 u32 mask = 0, tmp, tmp1;
6843 mutex_lock(&rdev->grbm_idx_mutex);
6844 cik_select_se_sh(rdev, se, sh);
6845 tmp = RREG32(CC_GC_SHADER_ARRAY_CONFIG);
6846 tmp1 = RREG32(GC_USER_SHADER_ARRAY_CONFIG);
6847 cik_select_se_sh(rdev, 0xffffffff, 0xffffffff);
6848 mutex_unlock(&rdev->grbm_idx_mutex);
6855 for (i = 0; i < rdev->config.cik.max_cu_per_sh; i ++) {
6860 return (~tmp) & mask;
6863 static void cik_init_ao_cu_mask(struct radeon_device *rdev)
6865 u32 i, j, k, active_cu_number = 0;
6866 u32 mask, counter, cu_bitmap;
6869 for (i = 0; i < rdev->config.cik.max_shader_engines; i++) {
6870 for (j = 0; j < rdev->config.cik.max_sh_per_se; j++) {
6874 for (k = 0; k < rdev->config.cik.max_cu_per_sh; k ++) {
6875 if (cik_get_cu_active_bitmap(rdev, i, j) & mask) {
6883 active_cu_number += counter;
6884 tmp |= (cu_bitmap << (i * 16 + j * 8));
6888 WREG32(RLC_PG_AO_CU_MASK, tmp);
6890 tmp = RREG32(RLC_MAX_PG_CU);
6891 tmp &= ~MAX_PU_CU_MASK;
6892 tmp |= MAX_PU_CU(active_cu_number);
6893 WREG32(RLC_MAX_PG_CU, tmp);
6896 static void cik_enable_gfx_static_mgpg(struct radeon_device *rdev,
6901 orig = data = RREG32(RLC_PG_CNTL);
6902 if (enable && (rdev->pg_flags & RADEON_PG_SUPPORT_GFX_SMG))
6903 data |= STATIC_PER_CU_PG_ENABLE;
6905 data &= ~STATIC_PER_CU_PG_ENABLE;
6907 WREG32(RLC_PG_CNTL, data);
6910 static void cik_enable_gfx_dynamic_mgpg(struct radeon_device *rdev,
6915 orig = data = RREG32(RLC_PG_CNTL);
6916 if (enable && (rdev->pg_flags & RADEON_PG_SUPPORT_GFX_DMG))
6917 data |= DYN_PER_CU_PG_ENABLE;
6919 data &= ~DYN_PER_CU_PG_ENABLE;
6921 WREG32(RLC_PG_CNTL, data);
6924 #define RLC_SAVE_AND_RESTORE_STARTING_OFFSET 0x90
6925 #define RLC_CLEAR_STATE_DESCRIPTOR_OFFSET 0x3D
6927 static void cik_init_gfx_cgpg(struct radeon_device *rdev)
6932 if (rdev->rlc.cs_data) {
6933 WREG32(RLC_GPM_SCRATCH_ADDR, RLC_CLEAR_STATE_DESCRIPTOR_OFFSET);
6934 WREG32(RLC_GPM_SCRATCH_DATA, upper_32_bits(rdev->rlc.clear_state_gpu_addr));
6935 WREG32(RLC_GPM_SCRATCH_DATA, lower_32_bits(rdev->rlc.clear_state_gpu_addr));
6936 WREG32(RLC_GPM_SCRATCH_DATA, rdev->rlc.clear_state_size);
6938 WREG32(RLC_GPM_SCRATCH_ADDR, RLC_CLEAR_STATE_DESCRIPTOR_OFFSET);
6939 for (i = 0; i < 3; i++)
6940 WREG32(RLC_GPM_SCRATCH_DATA, 0);
6942 if (rdev->rlc.reg_list) {
6943 WREG32(RLC_GPM_SCRATCH_ADDR, RLC_SAVE_AND_RESTORE_STARTING_OFFSET);
6944 for (i = 0; i < rdev->rlc.reg_list_size; i++)
6945 WREG32(RLC_GPM_SCRATCH_DATA, rdev->rlc.reg_list[i]);
6948 orig = data = RREG32(RLC_PG_CNTL);
6951 WREG32(RLC_PG_CNTL, data);
6953 WREG32(RLC_SAVE_AND_RESTORE_BASE, rdev->rlc.save_restore_gpu_addr >> 8);
6954 WREG32(RLC_CP_TABLE_RESTORE, rdev->rlc.cp_table_gpu_addr >> 8);
6956 data = RREG32(CP_RB_WPTR_POLL_CNTL);
6957 data &= ~IDLE_POLL_COUNT_MASK;
6958 data |= IDLE_POLL_COUNT(0x60);
6959 WREG32(CP_RB_WPTR_POLL_CNTL, data);
6962 WREG32(RLC_PG_DELAY, data);
6964 data = RREG32(RLC_PG_DELAY_2);
6967 WREG32(RLC_PG_DELAY_2, data);
6969 data = RREG32(RLC_AUTO_PG_CTRL);
6970 data &= ~GRBM_REG_SGIT_MASK;
6971 data |= GRBM_REG_SGIT(0x700);
6972 WREG32(RLC_AUTO_PG_CTRL, data);
6976 static void cik_update_gfx_pg(struct radeon_device *rdev, bool enable)
6978 cik_enable_gfx_cgpg(rdev, enable);
6979 cik_enable_gfx_static_mgpg(rdev, enable);
6980 cik_enable_gfx_dynamic_mgpg(rdev, enable);
6983 u32 cik_get_csb_size(struct radeon_device *rdev)
6986 const struct cs_section_def *sect = NULL;
6987 const struct cs_extent_def *ext = NULL;
6989 if (rdev->rlc.cs_data == NULL)
6992 /* begin clear state */
6994 /* context control state */
6997 for (sect = rdev->rlc.cs_data; sect->section != NULL; ++sect) {
6998 for (ext = sect->section; ext->extent != NULL; ++ext) {
6999 if (sect->id == SECT_CONTEXT)
7000 count += 2 + ext->reg_count;
7005 /* pa_sc_raster_config/pa_sc_raster_config1 */
7007 /* end clear state */
7015 void cik_get_csb_buffer(struct radeon_device *rdev, volatile u32 *buffer)
7018 const struct cs_section_def *sect = NULL;
7019 const struct cs_extent_def *ext = NULL;
7021 if (rdev->rlc.cs_data == NULL)
7026 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0));
7027 buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
7029 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CONTEXT_CONTROL, 1));
7030 buffer[count++] = cpu_to_le32(0x80000000);
7031 buffer[count++] = cpu_to_le32(0x80000000);
7033 for (sect = rdev->rlc.cs_data; sect->section != NULL; ++sect) {
7034 for (ext = sect->section; ext->extent != NULL; ++ext) {
7035 if (sect->id == SECT_CONTEXT) {
7037 cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, ext->reg_count));
7038 buffer[count++] = cpu_to_le32(ext->reg_index - 0xa000);
7039 for (i = 0; i < ext->reg_count; i++)
7040 buffer[count++] = cpu_to_le32(ext->extent[i]);
7047 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, 2));
7048 buffer[count++] = cpu_to_le32(PA_SC_RASTER_CONFIG - PACKET3_SET_CONTEXT_REG_START);
7049 switch (rdev->family) {
7051 buffer[count++] = cpu_to_le32(0x16000012);
7052 buffer[count++] = cpu_to_le32(0x00000000);
7055 buffer[count++] = cpu_to_le32(0x00000000); /* XXX */
7056 buffer[count++] = cpu_to_le32(0x00000000);
7060 buffer[count++] = cpu_to_le32(0x00000000); /* XXX */
7061 buffer[count++] = cpu_to_le32(0x00000000);
7064 buffer[count++] = cpu_to_le32(0x3a00161a);
7065 buffer[count++] = cpu_to_le32(0x0000002e);
7068 buffer[count++] = cpu_to_le32(0x00000000);
7069 buffer[count++] = cpu_to_le32(0x00000000);
7073 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0));
7074 buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_END_CLEAR_STATE);
7076 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CLEAR_STATE, 0));
7077 buffer[count++] = cpu_to_le32(0);
7080 static void cik_init_pg(struct radeon_device *rdev)
7082 if (rdev->pg_flags) {
7083 cik_enable_sck_slowdown_on_pu(rdev, true);
7084 cik_enable_sck_slowdown_on_pd(rdev, true);
7085 if (rdev->pg_flags & RADEON_PG_SUPPORT_GFX_PG) {
7086 cik_init_gfx_cgpg(rdev);
7087 cik_enable_cp_pg(rdev, true);
7088 cik_enable_gds_pg(rdev, true);
7090 cik_init_ao_cu_mask(rdev);
7091 cik_update_gfx_pg(rdev, true);
7095 static void cik_fini_pg(struct radeon_device *rdev)
7097 if (rdev->pg_flags) {
7098 cik_update_gfx_pg(rdev, false);
7099 if (rdev->pg_flags & RADEON_PG_SUPPORT_GFX_PG) {
7100 cik_enable_cp_pg(rdev, false);
7101 cik_enable_gds_pg(rdev, false);
7108 * Starting with r6xx, interrupts are handled via a ring buffer.
7109 * Ring buffers are areas of GPU accessible memory that the GPU
7110 * writes interrupt vectors into and the host reads vectors out of.
7111 * There is a rptr (read pointer) that determines where the
7112 * host is currently reading, and a wptr (write pointer)
7113 * which determines where the GPU has written. When the
7114 * pointers are equal, the ring is idle. When the GPU
7115 * writes vectors to the ring buffer, it increments the
7116 * wptr. When there is an interrupt, the host then starts
7117 * fetching commands and processing them until the pointers are
7118 * equal again at which point it updates the rptr.
7122 * cik_enable_interrupts - Enable the interrupt ring buffer
7124 * @rdev: radeon_device pointer
7126 * Enable the interrupt ring buffer (CIK).
7128 static void cik_enable_interrupts(struct radeon_device *rdev)
7130 u32 ih_cntl = RREG32(IH_CNTL);
7131 u32 ih_rb_cntl = RREG32(IH_RB_CNTL);
7133 ih_cntl |= ENABLE_INTR;
7134 ih_rb_cntl |= IH_RB_ENABLE;
7135 WREG32(IH_CNTL, ih_cntl);
7136 WREG32(IH_RB_CNTL, ih_rb_cntl);
7137 rdev->ih.enabled = true;
7141 * cik_disable_interrupts - Disable the interrupt ring buffer
7143 * @rdev: radeon_device pointer
7145 * Disable the interrupt ring buffer (CIK).
7147 static void cik_disable_interrupts(struct radeon_device *rdev)
7149 u32 ih_rb_cntl = RREG32(IH_RB_CNTL);
7150 u32 ih_cntl = RREG32(IH_CNTL);
7152 ih_rb_cntl &= ~IH_RB_ENABLE;
7153 ih_cntl &= ~ENABLE_INTR;
7154 WREG32(IH_RB_CNTL, ih_rb_cntl);
7155 WREG32(IH_CNTL, ih_cntl);
7156 /* set rptr, wptr to 0 */
7157 WREG32(IH_RB_RPTR, 0);
7158 WREG32(IH_RB_WPTR, 0);
7159 rdev->ih.enabled = false;
7164 * cik_disable_interrupt_state - Disable all interrupt sources
7166 * @rdev: radeon_device pointer
7168 * Clear all interrupt enable bits used by the driver (CIK).
7170 static void cik_disable_interrupt_state(struct radeon_device *rdev)
7175 tmp = RREG32(CP_INT_CNTL_RING0) &
7176 (CNTX_BUSY_INT_ENABLE | CNTX_EMPTY_INT_ENABLE);
7177 WREG32(CP_INT_CNTL_RING0, tmp);
7179 tmp = RREG32(SDMA0_CNTL + SDMA0_REGISTER_OFFSET) & ~TRAP_ENABLE;
7180 WREG32(SDMA0_CNTL + SDMA0_REGISTER_OFFSET, tmp);
7181 tmp = RREG32(SDMA0_CNTL + SDMA1_REGISTER_OFFSET) & ~TRAP_ENABLE;
7182 WREG32(SDMA0_CNTL + SDMA1_REGISTER_OFFSET, tmp);
7183 /* compute queues */
7184 WREG32(CP_ME1_PIPE0_INT_CNTL, 0);
7185 WREG32(CP_ME1_PIPE1_INT_CNTL, 0);
7186 WREG32(CP_ME1_PIPE2_INT_CNTL, 0);
7187 WREG32(CP_ME1_PIPE3_INT_CNTL, 0);
7188 WREG32(CP_ME2_PIPE0_INT_CNTL, 0);
7189 WREG32(CP_ME2_PIPE1_INT_CNTL, 0);
7190 WREG32(CP_ME2_PIPE2_INT_CNTL, 0);
7191 WREG32(CP_ME2_PIPE3_INT_CNTL, 0);
7193 WREG32(GRBM_INT_CNTL, 0);
7194 /* vline/vblank, etc. */
7195 WREG32(LB_INTERRUPT_MASK + EVERGREEN_CRTC0_REGISTER_OFFSET, 0);
7196 WREG32(LB_INTERRUPT_MASK + EVERGREEN_CRTC1_REGISTER_OFFSET, 0);
7197 if (rdev->num_crtc >= 4) {
7198 WREG32(LB_INTERRUPT_MASK + EVERGREEN_CRTC2_REGISTER_OFFSET, 0);
7199 WREG32(LB_INTERRUPT_MASK + EVERGREEN_CRTC3_REGISTER_OFFSET, 0);
7201 if (rdev->num_crtc >= 6) {
7202 WREG32(LB_INTERRUPT_MASK + EVERGREEN_CRTC4_REGISTER_OFFSET, 0);
7203 WREG32(LB_INTERRUPT_MASK + EVERGREEN_CRTC5_REGISTER_OFFSET, 0);
7206 if (rdev->num_crtc >= 2) {
7207 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC0_REGISTER_OFFSET, 0);
7208 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC1_REGISTER_OFFSET, 0);
7210 if (rdev->num_crtc >= 4) {
7211 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC2_REGISTER_OFFSET, 0);
7212 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC3_REGISTER_OFFSET, 0);
7214 if (rdev->num_crtc >= 6) {
7215 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC4_REGISTER_OFFSET, 0);
7216 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC5_REGISTER_OFFSET, 0);
7220 WREG32(DAC_AUTODETECT_INT_CONTROL, 0);
7222 /* digital hotplug */
7223 tmp = RREG32(DC_HPD1_INT_CONTROL) & DC_HPDx_INT_POLARITY;
7224 WREG32(DC_HPD1_INT_CONTROL, tmp);
7225 tmp = RREG32(DC_HPD2_INT_CONTROL) & DC_HPDx_INT_POLARITY;
7226 WREG32(DC_HPD2_INT_CONTROL, tmp);
7227 tmp = RREG32(DC_HPD3_INT_CONTROL) & DC_HPDx_INT_POLARITY;
7228 WREG32(DC_HPD3_INT_CONTROL, tmp);
7229 tmp = RREG32(DC_HPD4_INT_CONTROL) & DC_HPDx_INT_POLARITY;
7230 WREG32(DC_HPD4_INT_CONTROL, tmp);
7231 tmp = RREG32(DC_HPD5_INT_CONTROL) & DC_HPDx_INT_POLARITY;
7232 WREG32(DC_HPD5_INT_CONTROL, tmp);
7233 tmp = RREG32(DC_HPD6_INT_CONTROL) & DC_HPDx_INT_POLARITY;
7234 WREG32(DC_HPD6_INT_CONTROL, tmp);
7239 * cik_irq_init - init and enable the interrupt ring
7241 * @rdev: radeon_device pointer
7243 * Allocate a ring buffer for the interrupt controller,
7244 * enable the RLC, disable interrupts, enable the IH
7245 * ring buffer and enable it (CIK).
7246 * Called at device load and reume.
7247 * Returns 0 for success, errors for failure.
7249 static int cik_irq_init(struct radeon_device *rdev)
7253 u32 interrupt_cntl, ih_cntl, ih_rb_cntl;
7256 ret = r600_ih_ring_alloc(rdev);
7261 cik_disable_interrupts(rdev);
7264 ret = cik_rlc_resume(rdev);
7266 r600_ih_ring_fini(rdev);
7270 /* setup interrupt control */
7271 /* XXX this should actually be a bus address, not an MC address. same on older asics */
7272 WREG32(INTERRUPT_CNTL2, rdev->ih.gpu_addr >> 8);
7273 interrupt_cntl = RREG32(INTERRUPT_CNTL);
7274 /* IH_DUMMY_RD_OVERRIDE=0 - dummy read disabled with msi, enabled without msi
7275 * IH_DUMMY_RD_OVERRIDE=1 - dummy read controlled by IH_DUMMY_RD_EN
7277 interrupt_cntl &= ~IH_DUMMY_RD_OVERRIDE;
7278 /* IH_REQ_NONSNOOP_EN=1 if ring is in non-cacheable memory, e.g., vram */
7279 interrupt_cntl &= ~IH_REQ_NONSNOOP_EN;
7280 WREG32(INTERRUPT_CNTL, interrupt_cntl);
7282 WREG32(IH_RB_BASE, rdev->ih.gpu_addr >> 8);
7283 rb_bufsz = order_base_2(rdev->ih.ring_size / 4);
7285 ih_rb_cntl = (IH_WPTR_OVERFLOW_ENABLE |
7286 IH_WPTR_OVERFLOW_CLEAR |
7289 if (rdev->wb.enabled)
7290 ih_rb_cntl |= IH_WPTR_WRITEBACK_ENABLE;
7292 /* set the writeback address whether it's enabled or not */
7293 WREG32(IH_RB_WPTR_ADDR_LO, (rdev->wb.gpu_addr + R600_WB_IH_WPTR_OFFSET) & 0xFFFFFFFC);
7294 WREG32(IH_RB_WPTR_ADDR_HI, upper_32_bits(rdev->wb.gpu_addr + R600_WB_IH_WPTR_OFFSET) & 0xFF);
7296 WREG32(IH_RB_CNTL, ih_rb_cntl);
7298 /* set rptr, wptr to 0 */
7299 WREG32(IH_RB_RPTR, 0);
7300 WREG32(IH_RB_WPTR, 0);
7302 /* Default settings for IH_CNTL (disabled at first) */
7303 ih_cntl = MC_WRREQ_CREDIT(0x10) | MC_WR_CLEAN_CNT(0x10) | MC_VMID(0);
7304 /* RPTR_REARM only works if msi's are enabled */
7305 if (rdev->msi_enabled)
7306 ih_cntl |= RPTR_REARM;
7307 WREG32(IH_CNTL, ih_cntl);
7309 /* force the active interrupt state to all disabled */
7310 cik_disable_interrupt_state(rdev);
7312 pci_set_master(rdev->pdev);
7315 cik_enable_interrupts(rdev);
7321 * cik_irq_set - enable/disable interrupt sources
7323 * @rdev: radeon_device pointer
7325 * Enable interrupt sources on the GPU (vblanks, hpd,
7327 * Returns 0 for success, errors for failure.
7329 int cik_irq_set(struct radeon_device *rdev)
7333 u32 crtc1 = 0, crtc2 = 0, crtc3 = 0, crtc4 = 0, crtc5 = 0, crtc6 = 0;
7334 u32 hpd1, hpd2, hpd3, hpd4, hpd5, hpd6;
7335 u32 grbm_int_cntl = 0;
7336 u32 dma_cntl, dma_cntl1;
7339 if (!rdev->irq.installed) {
7340 WARN(1, "Can't enable IRQ/MSI because no handler is installed\n");
7343 /* don't enable anything if the ih is disabled */
7344 if (!rdev->ih.enabled) {
7345 cik_disable_interrupts(rdev);
7346 /* force the active interrupt state to all disabled */
7347 cik_disable_interrupt_state(rdev);
7351 cp_int_cntl = RREG32(CP_INT_CNTL_RING0) &
7352 (CNTX_BUSY_INT_ENABLE | CNTX_EMPTY_INT_ENABLE);
7353 cp_int_cntl |= PRIV_INSTR_INT_ENABLE | PRIV_REG_INT_ENABLE;
7355 hpd1 = RREG32(DC_HPD1_INT_CONTROL) & ~DC_HPDx_INT_EN;
7356 hpd2 = RREG32(DC_HPD2_INT_CONTROL) & ~DC_HPDx_INT_EN;
7357 hpd3 = RREG32(DC_HPD3_INT_CONTROL) & ~DC_HPDx_INT_EN;
7358 hpd4 = RREG32(DC_HPD4_INT_CONTROL) & ~DC_HPDx_INT_EN;
7359 hpd5 = RREG32(DC_HPD5_INT_CONTROL) & ~DC_HPDx_INT_EN;
7360 hpd6 = RREG32(DC_HPD6_INT_CONTROL) & ~DC_HPDx_INT_EN;
7362 dma_cntl = RREG32(SDMA0_CNTL + SDMA0_REGISTER_OFFSET) & ~TRAP_ENABLE;
7363 dma_cntl1 = RREG32(SDMA0_CNTL + SDMA1_REGISTER_OFFSET) & ~TRAP_ENABLE;
7365 cp_m1p0 = RREG32(CP_ME1_PIPE0_INT_CNTL) & ~TIME_STAMP_INT_ENABLE;
7367 if (rdev->flags & RADEON_IS_IGP)
7368 thermal_int = RREG32_SMC(CG_THERMAL_INT_CTRL) &
7369 ~(THERM_INTH_MASK | THERM_INTL_MASK);
7371 thermal_int = RREG32_SMC(CG_THERMAL_INT) &
7372 ~(THERM_INT_MASK_HIGH | THERM_INT_MASK_LOW);
7374 /* enable CP interrupts on all rings */
7375 if (atomic_read(&rdev->irq.ring_int[RADEON_RING_TYPE_GFX_INDEX])) {
7376 DRM_DEBUG("cik_irq_set: sw int gfx\n");
7377 cp_int_cntl |= TIME_STAMP_INT_ENABLE;
7379 if (atomic_read(&rdev->irq.ring_int[CAYMAN_RING_TYPE_CP1_INDEX])) {
7380 struct radeon_ring *ring = &rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX];
7381 DRM_DEBUG("si_irq_set: sw int cp1\n");
7382 if (ring->me == 1) {
7383 switch (ring->pipe) {
7385 cp_m1p0 |= TIME_STAMP_INT_ENABLE;
7388 DRM_DEBUG("si_irq_set: sw int cp1 invalid pipe %d\n", ring->pipe);
7392 DRM_DEBUG("si_irq_set: sw int cp1 invalid me %d\n", ring->me);
7395 if (atomic_read(&rdev->irq.ring_int[CAYMAN_RING_TYPE_CP2_INDEX])) {
7396 struct radeon_ring *ring = &rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX];
7397 DRM_DEBUG("si_irq_set: sw int cp2\n");
7398 if (ring->me == 1) {
7399 switch (ring->pipe) {
7401 cp_m1p0 |= TIME_STAMP_INT_ENABLE;
7404 DRM_DEBUG("si_irq_set: sw int cp2 invalid pipe %d\n", ring->pipe);
7408 DRM_DEBUG("si_irq_set: sw int cp2 invalid me %d\n", ring->me);
7412 if (atomic_read(&rdev->irq.ring_int[R600_RING_TYPE_DMA_INDEX])) {
7413 DRM_DEBUG("cik_irq_set: sw int dma\n");
7414 dma_cntl |= TRAP_ENABLE;
7417 if (atomic_read(&rdev->irq.ring_int[CAYMAN_RING_TYPE_DMA1_INDEX])) {
7418 DRM_DEBUG("cik_irq_set: sw int dma1\n");
7419 dma_cntl1 |= TRAP_ENABLE;
7422 if (rdev->irq.crtc_vblank_int[0] ||
7423 atomic_read(&rdev->irq.pflip[0])) {
7424 DRM_DEBUG("cik_irq_set: vblank 0\n");
7425 crtc1 |= VBLANK_INTERRUPT_MASK;
7427 if (rdev->irq.crtc_vblank_int[1] ||
7428 atomic_read(&rdev->irq.pflip[1])) {
7429 DRM_DEBUG("cik_irq_set: vblank 1\n");
7430 crtc2 |= VBLANK_INTERRUPT_MASK;
7432 if (rdev->irq.crtc_vblank_int[2] ||
7433 atomic_read(&rdev->irq.pflip[2])) {
7434 DRM_DEBUG("cik_irq_set: vblank 2\n");
7435 crtc3 |= VBLANK_INTERRUPT_MASK;
7437 if (rdev->irq.crtc_vblank_int[3] ||
7438 atomic_read(&rdev->irq.pflip[3])) {
7439 DRM_DEBUG("cik_irq_set: vblank 3\n");
7440 crtc4 |= VBLANK_INTERRUPT_MASK;
7442 if (rdev->irq.crtc_vblank_int[4] ||
7443 atomic_read(&rdev->irq.pflip[4])) {
7444 DRM_DEBUG("cik_irq_set: vblank 4\n");
7445 crtc5 |= VBLANK_INTERRUPT_MASK;
7447 if (rdev->irq.crtc_vblank_int[5] ||
7448 atomic_read(&rdev->irq.pflip[5])) {
7449 DRM_DEBUG("cik_irq_set: vblank 5\n");
7450 crtc6 |= VBLANK_INTERRUPT_MASK;
7452 if (rdev->irq.hpd[0]) {
7453 DRM_DEBUG("cik_irq_set: hpd 1\n");
7454 hpd1 |= DC_HPDx_INT_EN;
7456 if (rdev->irq.hpd[1]) {
7457 DRM_DEBUG("cik_irq_set: hpd 2\n");
7458 hpd2 |= DC_HPDx_INT_EN;
7460 if (rdev->irq.hpd[2]) {
7461 DRM_DEBUG("cik_irq_set: hpd 3\n");
7462 hpd3 |= DC_HPDx_INT_EN;
7464 if (rdev->irq.hpd[3]) {
7465 DRM_DEBUG("cik_irq_set: hpd 4\n");
7466 hpd4 |= DC_HPDx_INT_EN;
7468 if (rdev->irq.hpd[4]) {
7469 DRM_DEBUG("cik_irq_set: hpd 5\n");
7470 hpd5 |= DC_HPDx_INT_EN;
7472 if (rdev->irq.hpd[5]) {
7473 DRM_DEBUG("cik_irq_set: hpd 6\n");
7474 hpd6 |= DC_HPDx_INT_EN;
7477 if (rdev->irq.dpm_thermal) {
7478 DRM_DEBUG("dpm thermal\n");
7479 if (rdev->flags & RADEON_IS_IGP)
7480 thermal_int |= THERM_INTH_MASK | THERM_INTL_MASK;
7482 thermal_int |= THERM_INT_MASK_HIGH | THERM_INT_MASK_LOW;
7485 WREG32(CP_INT_CNTL_RING0, cp_int_cntl);
7487 WREG32(SDMA0_CNTL + SDMA0_REGISTER_OFFSET, dma_cntl);
7488 WREG32(SDMA0_CNTL + SDMA1_REGISTER_OFFSET, dma_cntl1);
7490 WREG32(CP_ME1_PIPE0_INT_CNTL, cp_m1p0);
7492 WREG32(GRBM_INT_CNTL, grbm_int_cntl);
7494 WREG32(LB_INTERRUPT_MASK + EVERGREEN_CRTC0_REGISTER_OFFSET, crtc1);
7495 WREG32(LB_INTERRUPT_MASK + EVERGREEN_CRTC1_REGISTER_OFFSET, crtc2);
7496 if (rdev->num_crtc >= 4) {
7497 WREG32(LB_INTERRUPT_MASK + EVERGREEN_CRTC2_REGISTER_OFFSET, crtc3);
7498 WREG32(LB_INTERRUPT_MASK + EVERGREEN_CRTC3_REGISTER_OFFSET, crtc4);
7500 if (rdev->num_crtc >= 6) {
7501 WREG32(LB_INTERRUPT_MASK + EVERGREEN_CRTC4_REGISTER_OFFSET, crtc5);
7502 WREG32(LB_INTERRUPT_MASK + EVERGREEN_CRTC5_REGISTER_OFFSET, crtc6);
7505 if (rdev->num_crtc >= 2) {
7506 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC0_REGISTER_OFFSET,
7507 GRPH_PFLIP_INT_MASK);
7508 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC1_REGISTER_OFFSET,
7509 GRPH_PFLIP_INT_MASK);
7511 if (rdev->num_crtc >= 4) {
7512 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC2_REGISTER_OFFSET,
7513 GRPH_PFLIP_INT_MASK);
7514 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC3_REGISTER_OFFSET,
7515 GRPH_PFLIP_INT_MASK);
7517 if (rdev->num_crtc >= 6) {
7518 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC4_REGISTER_OFFSET,
7519 GRPH_PFLIP_INT_MASK);
7520 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC5_REGISTER_OFFSET,
7521 GRPH_PFLIP_INT_MASK);
7524 WREG32(DC_HPD1_INT_CONTROL, hpd1);
7525 WREG32(DC_HPD2_INT_CONTROL, hpd2);
7526 WREG32(DC_HPD3_INT_CONTROL, hpd3);
7527 WREG32(DC_HPD4_INT_CONTROL, hpd4);
7528 WREG32(DC_HPD5_INT_CONTROL, hpd5);
7529 WREG32(DC_HPD6_INT_CONTROL, hpd6);
7531 if (rdev->flags & RADEON_IS_IGP)
7532 WREG32_SMC(CG_THERMAL_INT_CTRL, thermal_int);
7534 WREG32_SMC(CG_THERMAL_INT, thermal_int);
7540 * cik_irq_ack - ack interrupt sources
7542 * @rdev: radeon_device pointer
7544 * Ack interrupt sources on the GPU (vblanks, hpd,
7545 * etc.) (CIK). Certain interrupts sources are sw
7546 * generated and do not require an explicit ack.
7548 static inline void cik_irq_ack(struct radeon_device *rdev)
7552 rdev->irq.stat_regs.cik.disp_int = RREG32(DISP_INTERRUPT_STATUS);
7553 rdev->irq.stat_regs.cik.disp_int_cont = RREG32(DISP_INTERRUPT_STATUS_CONTINUE);
7554 rdev->irq.stat_regs.cik.disp_int_cont2 = RREG32(DISP_INTERRUPT_STATUS_CONTINUE2);
7555 rdev->irq.stat_regs.cik.disp_int_cont3 = RREG32(DISP_INTERRUPT_STATUS_CONTINUE3);
7556 rdev->irq.stat_regs.cik.disp_int_cont4 = RREG32(DISP_INTERRUPT_STATUS_CONTINUE4);
7557 rdev->irq.stat_regs.cik.disp_int_cont5 = RREG32(DISP_INTERRUPT_STATUS_CONTINUE5);
7558 rdev->irq.stat_regs.cik.disp_int_cont6 = RREG32(DISP_INTERRUPT_STATUS_CONTINUE6);
7560 rdev->irq.stat_regs.cik.d1grph_int = RREG32(GRPH_INT_STATUS +
7561 EVERGREEN_CRTC0_REGISTER_OFFSET);
7562 rdev->irq.stat_regs.cik.d2grph_int = RREG32(GRPH_INT_STATUS +
7563 EVERGREEN_CRTC1_REGISTER_OFFSET);
7564 if (rdev->num_crtc >= 4) {
7565 rdev->irq.stat_regs.cik.d3grph_int = RREG32(GRPH_INT_STATUS +
7566 EVERGREEN_CRTC2_REGISTER_OFFSET);
7567 rdev->irq.stat_regs.cik.d4grph_int = RREG32(GRPH_INT_STATUS +
7568 EVERGREEN_CRTC3_REGISTER_OFFSET);
7570 if (rdev->num_crtc >= 6) {
7571 rdev->irq.stat_regs.cik.d5grph_int = RREG32(GRPH_INT_STATUS +
7572 EVERGREEN_CRTC4_REGISTER_OFFSET);
7573 rdev->irq.stat_regs.cik.d6grph_int = RREG32(GRPH_INT_STATUS +
7574 EVERGREEN_CRTC5_REGISTER_OFFSET);
7577 if (rdev->irq.stat_regs.cik.d1grph_int & GRPH_PFLIP_INT_OCCURRED)
7578 WREG32(GRPH_INT_STATUS + EVERGREEN_CRTC0_REGISTER_OFFSET,
7579 GRPH_PFLIP_INT_CLEAR);
7580 if (rdev->irq.stat_regs.cik.d2grph_int & GRPH_PFLIP_INT_OCCURRED)
7581 WREG32(GRPH_INT_STATUS + EVERGREEN_CRTC1_REGISTER_OFFSET,
7582 GRPH_PFLIP_INT_CLEAR);
7583 if (rdev->irq.stat_regs.cik.disp_int & LB_D1_VBLANK_INTERRUPT)
7584 WREG32(LB_VBLANK_STATUS + EVERGREEN_CRTC0_REGISTER_OFFSET, VBLANK_ACK);
7585 if (rdev->irq.stat_regs.cik.disp_int & LB_D1_VLINE_INTERRUPT)
7586 WREG32(LB_VLINE_STATUS + EVERGREEN_CRTC0_REGISTER_OFFSET, VLINE_ACK);
7587 if (rdev->irq.stat_regs.cik.disp_int_cont & LB_D2_VBLANK_INTERRUPT)
7588 WREG32(LB_VBLANK_STATUS + EVERGREEN_CRTC1_REGISTER_OFFSET, VBLANK_ACK);
7589 if (rdev->irq.stat_regs.cik.disp_int_cont & LB_D2_VLINE_INTERRUPT)
7590 WREG32(LB_VLINE_STATUS + EVERGREEN_CRTC1_REGISTER_OFFSET, VLINE_ACK);
7592 if (rdev->num_crtc >= 4) {
7593 if (rdev->irq.stat_regs.cik.d3grph_int & GRPH_PFLIP_INT_OCCURRED)
7594 WREG32(GRPH_INT_STATUS + EVERGREEN_CRTC2_REGISTER_OFFSET,
7595 GRPH_PFLIP_INT_CLEAR);
7596 if (rdev->irq.stat_regs.cik.d4grph_int & GRPH_PFLIP_INT_OCCURRED)
7597 WREG32(GRPH_INT_STATUS + EVERGREEN_CRTC3_REGISTER_OFFSET,
7598 GRPH_PFLIP_INT_CLEAR);
7599 if (rdev->irq.stat_regs.cik.disp_int_cont2 & LB_D3_VBLANK_INTERRUPT)
7600 WREG32(LB_VBLANK_STATUS + EVERGREEN_CRTC2_REGISTER_OFFSET, VBLANK_ACK);
7601 if (rdev->irq.stat_regs.cik.disp_int_cont2 & LB_D3_VLINE_INTERRUPT)
7602 WREG32(LB_VLINE_STATUS + EVERGREEN_CRTC2_REGISTER_OFFSET, VLINE_ACK);
7603 if (rdev->irq.stat_regs.cik.disp_int_cont3 & LB_D4_VBLANK_INTERRUPT)
7604 WREG32(LB_VBLANK_STATUS + EVERGREEN_CRTC3_REGISTER_OFFSET, VBLANK_ACK);
7605 if (rdev->irq.stat_regs.cik.disp_int_cont3 & LB_D4_VLINE_INTERRUPT)
7606 WREG32(LB_VLINE_STATUS + EVERGREEN_CRTC3_REGISTER_OFFSET, VLINE_ACK);
7609 if (rdev->num_crtc >= 6) {
7610 if (rdev->irq.stat_regs.cik.d5grph_int & GRPH_PFLIP_INT_OCCURRED)
7611 WREG32(GRPH_INT_STATUS + EVERGREEN_CRTC4_REGISTER_OFFSET,
7612 GRPH_PFLIP_INT_CLEAR);
7613 if (rdev->irq.stat_regs.cik.d6grph_int & GRPH_PFLIP_INT_OCCURRED)
7614 WREG32(GRPH_INT_STATUS + EVERGREEN_CRTC5_REGISTER_OFFSET,
7615 GRPH_PFLIP_INT_CLEAR);
7616 if (rdev->irq.stat_regs.cik.disp_int_cont4 & LB_D5_VBLANK_INTERRUPT)
7617 WREG32(LB_VBLANK_STATUS + EVERGREEN_CRTC4_REGISTER_OFFSET, VBLANK_ACK);
7618 if (rdev->irq.stat_regs.cik.disp_int_cont4 & LB_D5_VLINE_INTERRUPT)
7619 WREG32(LB_VLINE_STATUS + EVERGREEN_CRTC4_REGISTER_OFFSET, VLINE_ACK);
7620 if (rdev->irq.stat_regs.cik.disp_int_cont5 & LB_D6_VBLANK_INTERRUPT)
7621 WREG32(LB_VBLANK_STATUS + EVERGREEN_CRTC5_REGISTER_OFFSET, VBLANK_ACK);
7622 if (rdev->irq.stat_regs.cik.disp_int_cont5 & LB_D6_VLINE_INTERRUPT)
7623 WREG32(LB_VLINE_STATUS + EVERGREEN_CRTC5_REGISTER_OFFSET, VLINE_ACK);
7626 if (rdev->irq.stat_regs.cik.disp_int & DC_HPD1_INTERRUPT) {
7627 tmp = RREG32(DC_HPD1_INT_CONTROL);
7628 tmp |= DC_HPDx_INT_ACK;
7629 WREG32(DC_HPD1_INT_CONTROL, tmp);
7631 if (rdev->irq.stat_regs.cik.disp_int_cont & DC_HPD2_INTERRUPT) {
7632 tmp = RREG32(DC_HPD2_INT_CONTROL);
7633 tmp |= DC_HPDx_INT_ACK;
7634 WREG32(DC_HPD2_INT_CONTROL, tmp);
7636 if (rdev->irq.stat_regs.cik.disp_int_cont2 & DC_HPD3_INTERRUPT) {
7637 tmp = RREG32(DC_HPD3_INT_CONTROL);
7638 tmp |= DC_HPDx_INT_ACK;
7639 WREG32(DC_HPD3_INT_CONTROL, tmp);
7641 if (rdev->irq.stat_regs.cik.disp_int_cont3 & DC_HPD4_INTERRUPT) {
7642 tmp = RREG32(DC_HPD4_INT_CONTROL);
7643 tmp |= DC_HPDx_INT_ACK;
7644 WREG32(DC_HPD4_INT_CONTROL, tmp);
7646 if (rdev->irq.stat_regs.cik.disp_int_cont4 & DC_HPD5_INTERRUPT) {
7647 tmp = RREG32(DC_HPD5_INT_CONTROL);
7648 tmp |= DC_HPDx_INT_ACK;
7649 WREG32(DC_HPD5_INT_CONTROL, tmp);
7651 if (rdev->irq.stat_regs.cik.disp_int_cont5 & DC_HPD6_INTERRUPT) {
7652 tmp = RREG32(DC_HPD5_INT_CONTROL);
7653 tmp |= DC_HPDx_INT_ACK;
7654 WREG32(DC_HPD6_INT_CONTROL, tmp);
7659 * cik_irq_disable - disable interrupts
7661 * @rdev: radeon_device pointer
7663 * Disable interrupts on the hw (CIK).
7665 static void cik_irq_disable(struct radeon_device *rdev)
7667 cik_disable_interrupts(rdev);
7668 /* Wait and acknowledge irq */
7671 cik_disable_interrupt_state(rdev);
7675 * cik_irq_disable - disable interrupts for suspend
7677 * @rdev: radeon_device pointer
7679 * Disable interrupts and stop the RLC (CIK).
7682 static void cik_irq_suspend(struct radeon_device *rdev)
7684 cik_irq_disable(rdev);
7689 * cik_irq_fini - tear down interrupt support
7691 * @rdev: radeon_device pointer
7693 * Disable interrupts on the hw and free the IH ring
7695 * Used for driver unload.
7697 static void cik_irq_fini(struct radeon_device *rdev)
7699 cik_irq_suspend(rdev);
7700 r600_ih_ring_fini(rdev);
7704 * cik_get_ih_wptr - get the IH ring buffer wptr
7706 * @rdev: radeon_device pointer
7708 * Get the IH ring buffer wptr from either the register
7709 * or the writeback memory buffer (CIK). Also check for
7710 * ring buffer overflow and deal with it.
7711 * Used by cik_irq_process().
7712 * Returns the value of the wptr.
7714 static inline u32 cik_get_ih_wptr(struct radeon_device *rdev)
7718 if (rdev->wb.enabled)
7719 wptr = le32_to_cpu(rdev->wb.wb[R600_WB_IH_WPTR_OFFSET/4]);
7721 wptr = RREG32(IH_RB_WPTR);
7723 if (wptr & RB_OVERFLOW) {
7724 wptr &= ~RB_OVERFLOW;
7725 /* When a ring buffer overflow happen start parsing interrupt
7726 * from the last not overwritten vector (wptr + 16). Hopefully
7727 * this should allow us to catchup.
7729 dev_warn(rdev->dev, "IH ring buffer overflow (0x%08X, 0x%08X, 0x%08X)\n",
7730 wptr, rdev->ih.rptr, (wptr + 16) & rdev->ih.ptr_mask);
7731 rdev->ih.rptr = (wptr + 16) & rdev->ih.ptr_mask;
7732 tmp = RREG32(IH_RB_CNTL);
7733 tmp |= IH_WPTR_OVERFLOW_CLEAR;
7734 WREG32(IH_RB_CNTL, tmp);
7736 return (wptr & rdev->ih.ptr_mask);
7740 * Each IV ring entry is 128 bits:
7741 * [7:0] - interrupt source id
7743 * [59:32] - interrupt source data
7744 * [63:60] - reserved
7747 * ME_ID [1:0], PIPE_ID[1:0], QUEUE_ID[2:0]
7748 * QUEUE_ID - for compute, which of the 8 queues owned by the dispatcher
7749 * - for gfx, hw shader state (0=PS...5=LS, 6=CS)
7750 * ME_ID - 0 = gfx, 1 = first 4 CS pipes, 2 = second 4 CS pipes
7751 * PIPE_ID - ME0 0=3D
7752 * - ME1&2 compute dispatcher (4 pipes each)
7754 * INSTANCE_ID [1:0], QUEUE_ID[1:0]
7755 * INSTANCE_ID - 0 = sdma0, 1 = sdma1
7756 * QUEUE_ID - 0 = gfx, 1 = rlc0, 2 = rlc1
7759 * [127:96] - reserved
7762 * cik_irq_process - interrupt handler
7764 * @rdev: radeon_device pointer
7766 * Interrupt hander (CIK). Walk the IH ring,
7767 * ack interrupts and schedule work to handle
7769 * Returns irq process return code.
7771 int cik_irq_process(struct radeon_device *rdev)
7773 struct radeon_ring *cp1_ring = &rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX];
7774 struct radeon_ring *cp2_ring = &rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX];
7777 u32 src_id, src_data, ring_id;
7778 u8 me_id, pipe_id, queue_id;
7780 bool queue_hotplug = false;
7781 bool queue_reset = false;
7782 u32 addr, status, mc_client;
7783 bool queue_thermal = false;
7785 if (!rdev->ih.enabled || rdev->shutdown)
7788 wptr = cik_get_ih_wptr(rdev);
7791 /* is somebody else already processing irqs? */
7792 if (atomic_xchg(&rdev->ih.lock, 1))
7795 rptr = rdev->ih.rptr;
7796 DRM_DEBUG("cik_irq_process start: rptr %d, wptr %d\n", rptr, wptr);
7798 /* Order reading of wptr vs. reading of IH ring data */
7801 /* display interrupts */
7804 while (rptr != wptr) {
7805 /* wptr/rptr are in bytes! */
7806 ring_index = rptr / 4;
7808 radeon_kfd_interrupt(rdev,
7809 (const void *) &rdev->ih.ring[ring_index]);
7811 src_id = le32_to_cpu(rdev->ih.ring[ring_index]) & 0xff;
7812 src_data = le32_to_cpu(rdev->ih.ring[ring_index + 1]) & 0xfffffff;
7813 ring_id = le32_to_cpu(rdev->ih.ring[ring_index + 2]) & 0xff;
7816 case 1: /* D1 vblank/vline */
7818 case 0: /* D1 vblank */
7819 if (rdev->irq.stat_regs.cik.disp_int & LB_D1_VBLANK_INTERRUPT) {
7820 if (rdev->irq.crtc_vblank_int[0]) {
7821 drm_handle_vblank(rdev->ddev, 0);
7822 rdev->pm.vblank_sync = true;
7823 wake_up(&rdev->irq.vblank_queue);
7825 if (atomic_read(&rdev->irq.pflip[0]))
7826 radeon_crtc_handle_vblank(rdev, 0);
7827 rdev->irq.stat_regs.cik.disp_int &= ~LB_D1_VBLANK_INTERRUPT;
7828 DRM_DEBUG("IH: D1 vblank\n");
7831 case 1: /* D1 vline */
7832 if (rdev->irq.stat_regs.cik.disp_int & LB_D1_VLINE_INTERRUPT) {
7833 rdev->irq.stat_regs.cik.disp_int &= ~LB_D1_VLINE_INTERRUPT;
7834 DRM_DEBUG("IH: D1 vline\n");
7838 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
7842 case 2: /* D2 vblank/vline */
7844 case 0: /* D2 vblank */
7845 if (rdev->irq.stat_regs.cik.disp_int_cont & LB_D2_VBLANK_INTERRUPT) {
7846 if (rdev->irq.crtc_vblank_int[1]) {
7847 drm_handle_vblank(rdev->ddev, 1);
7848 rdev->pm.vblank_sync = true;
7849 wake_up(&rdev->irq.vblank_queue);
7851 if (atomic_read(&rdev->irq.pflip[1]))
7852 radeon_crtc_handle_vblank(rdev, 1);
7853 rdev->irq.stat_regs.cik.disp_int_cont &= ~LB_D2_VBLANK_INTERRUPT;
7854 DRM_DEBUG("IH: D2 vblank\n");
7857 case 1: /* D2 vline */
7858 if (rdev->irq.stat_regs.cik.disp_int_cont & LB_D2_VLINE_INTERRUPT) {
7859 rdev->irq.stat_regs.cik.disp_int_cont &= ~LB_D2_VLINE_INTERRUPT;
7860 DRM_DEBUG("IH: D2 vline\n");
7864 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
7868 case 3: /* D3 vblank/vline */
7870 case 0: /* D3 vblank */
7871 if (rdev->irq.stat_regs.cik.disp_int_cont2 & LB_D3_VBLANK_INTERRUPT) {
7872 if (rdev->irq.crtc_vblank_int[2]) {
7873 drm_handle_vblank(rdev->ddev, 2);
7874 rdev->pm.vblank_sync = true;
7875 wake_up(&rdev->irq.vblank_queue);
7877 if (atomic_read(&rdev->irq.pflip[2]))
7878 radeon_crtc_handle_vblank(rdev, 2);
7879 rdev->irq.stat_regs.cik.disp_int_cont2 &= ~LB_D3_VBLANK_INTERRUPT;
7880 DRM_DEBUG("IH: D3 vblank\n");
7883 case 1: /* D3 vline */
7884 if (rdev->irq.stat_regs.cik.disp_int_cont2 & LB_D3_VLINE_INTERRUPT) {
7885 rdev->irq.stat_regs.cik.disp_int_cont2 &= ~LB_D3_VLINE_INTERRUPT;
7886 DRM_DEBUG("IH: D3 vline\n");
7890 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
7894 case 4: /* D4 vblank/vline */
7896 case 0: /* D4 vblank */
7897 if (rdev->irq.stat_regs.cik.disp_int_cont3 & LB_D4_VBLANK_INTERRUPT) {
7898 if (rdev->irq.crtc_vblank_int[3]) {
7899 drm_handle_vblank(rdev->ddev, 3);
7900 rdev->pm.vblank_sync = true;
7901 wake_up(&rdev->irq.vblank_queue);
7903 if (atomic_read(&rdev->irq.pflip[3]))
7904 radeon_crtc_handle_vblank(rdev, 3);
7905 rdev->irq.stat_regs.cik.disp_int_cont3 &= ~LB_D4_VBLANK_INTERRUPT;
7906 DRM_DEBUG("IH: D4 vblank\n");
7909 case 1: /* D4 vline */
7910 if (rdev->irq.stat_regs.cik.disp_int_cont3 & LB_D4_VLINE_INTERRUPT) {
7911 rdev->irq.stat_regs.cik.disp_int_cont3 &= ~LB_D4_VLINE_INTERRUPT;
7912 DRM_DEBUG("IH: D4 vline\n");
7916 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
7920 case 5: /* D5 vblank/vline */
7922 case 0: /* D5 vblank */
7923 if (rdev->irq.stat_regs.cik.disp_int_cont4 & LB_D5_VBLANK_INTERRUPT) {
7924 if (rdev->irq.crtc_vblank_int[4]) {
7925 drm_handle_vblank(rdev->ddev, 4);
7926 rdev->pm.vblank_sync = true;
7927 wake_up(&rdev->irq.vblank_queue);
7929 if (atomic_read(&rdev->irq.pflip[4]))
7930 radeon_crtc_handle_vblank(rdev, 4);
7931 rdev->irq.stat_regs.cik.disp_int_cont4 &= ~LB_D5_VBLANK_INTERRUPT;
7932 DRM_DEBUG("IH: D5 vblank\n");
7935 case 1: /* D5 vline */
7936 if (rdev->irq.stat_regs.cik.disp_int_cont4 & LB_D5_VLINE_INTERRUPT) {
7937 rdev->irq.stat_regs.cik.disp_int_cont4 &= ~LB_D5_VLINE_INTERRUPT;
7938 DRM_DEBUG("IH: D5 vline\n");
7942 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
7946 case 6: /* D6 vblank/vline */
7948 case 0: /* D6 vblank */
7949 if (rdev->irq.stat_regs.cik.disp_int_cont5 & LB_D6_VBLANK_INTERRUPT) {
7950 if (rdev->irq.crtc_vblank_int[5]) {
7951 drm_handle_vblank(rdev->ddev, 5);
7952 rdev->pm.vblank_sync = true;
7953 wake_up(&rdev->irq.vblank_queue);
7955 if (atomic_read(&rdev->irq.pflip[5]))
7956 radeon_crtc_handle_vblank(rdev, 5);
7957 rdev->irq.stat_regs.cik.disp_int_cont5 &= ~LB_D6_VBLANK_INTERRUPT;
7958 DRM_DEBUG("IH: D6 vblank\n");
7961 case 1: /* D6 vline */
7962 if (rdev->irq.stat_regs.cik.disp_int_cont5 & LB_D6_VLINE_INTERRUPT) {
7963 rdev->irq.stat_regs.cik.disp_int_cont5 &= ~LB_D6_VLINE_INTERRUPT;
7964 DRM_DEBUG("IH: D6 vline\n");
7968 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
7972 case 8: /* D1 page flip */
7973 case 10: /* D2 page flip */
7974 case 12: /* D3 page flip */
7975 case 14: /* D4 page flip */
7976 case 16: /* D5 page flip */
7977 case 18: /* D6 page flip */
7978 DRM_DEBUG("IH: D%d flip\n", ((src_id - 8) >> 1) + 1);
7979 if (radeon_use_pflipirq > 0)
7980 radeon_crtc_handle_flip(rdev, (src_id - 8) >> 1);
7982 case 42: /* HPD hotplug */
7985 if (rdev->irq.stat_regs.cik.disp_int & DC_HPD1_INTERRUPT) {
7986 rdev->irq.stat_regs.cik.disp_int &= ~DC_HPD1_INTERRUPT;
7987 queue_hotplug = true;
7988 DRM_DEBUG("IH: HPD1\n");
7992 if (rdev->irq.stat_regs.cik.disp_int_cont & DC_HPD2_INTERRUPT) {
7993 rdev->irq.stat_regs.cik.disp_int_cont &= ~DC_HPD2_INTERRUPT;
7994 queue_hotplug = true;
7995 DRM_DEBUG("IH: HPD2\n");
7999 if (rdev->irq.stat_regs.cik.disp_int_cont2 & DC_HPD3_INTERRUPT) {
8000 rdev->irq.stat_regs.cik.disp_int_cont2 &= ~DC_HPD3_INTERRUPT;
8001 queue_hotplug = true;
8002 DRM_DEBUG("IH: HPD3\n");
8006 if (rdev->irq.stat_regs.cik.disp_int_cont3 & DC_HPD4_INTERRUPT) {
8007 rdev->irq.stat_regs.cik.disp_int_cont3 &= ~DC_HPD4_INTERRUPT;
8008 queue_hotplug = true;
8009 DRM_DEBUG("IH: HPD4\n");
8013 if (rdev->irq.stat_regs.cik.disp_int_cont4 & DC_HPD5_INTERRUPT) {
8014 rdev->irq.stat_regs.cik.disp_int_cont4 &= ~DC_HPD5_INTERRUPT;
8015 queue_hotplug = true;
8016 DRM_DEBUG("IH: HPD5\n");
8020 if (rdev->irq.stat_regs.cik.disp_int_cont5 & DC_HPD6_INTERRUPT) {
8021 rdev->irq.stat_regs.cik.disp_int_cont5 &= ~DC_HPD6_INTERRUPT;
8022 queue_hotplug = true;
8023 DRM_DEBUG("IH: HPD6\n");
8027 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
8032 DRM_DEBUG("IH: UVD int: 0x%08x\n", src_data);
8033 radeon_fence_process(rdev, R600_RING_TYPE_UVD_INDEX);
8037 addr = RREG32(VM_CONTEXT1_PROTECTION_FAULT_ADDR);
8038 status = RREG32(VM_CONTEXT1_PROTECTION_FAULT_STATUS);
8039 mc_client = RREG32(VM_CONTEXT1_PROTECTION_FAULT_MCCLIENT);
8040 /* reset addr and status */
8041 WREG32_P(VM_CONTEXT1_CNTL2, 1, ~1);
8042 if (addr == 0x0 && status == 0x0)
8044 dev_err(rdev->dev, "GPU fault detected: %d 0x%08x\n", src_id, src_data);
8045 dev_err(rdev->dev, " VM_CONTEXT1_PROTECTION_FAULT_ADDR 0x%08X\n",
8047 dev_err(rdev->dev, " VM_CONTEXT1_PROTECTION_FAULT_STATUS 0x%08X\n",
8049 cik_vm_decode_fault(rdev, status, addr, mc_client);
8052 DRM_DEBUG("IH: VCE int: 0x%08x\n", src_data);
8055 radeon_fence_process(rdev, TN_RING_TYPE_VCE1_INDEX);
8058 radeon_fence_process(rdev, TN_RING_TYPE_VCE2_INDEX);
8061 DRM_ERROR("Unhandled interrupt: %d %d\n", src_id, src_data);
8065 case 176: /* GFX RB CP_INT */
8066 case 177: /* GFX IB CP_INT */
8067 radeon_fence_process(rdev, RADEON_RING_TYPE_GFX_INDEX);
8069 case 181: /* CP EOP event */
8070 DRM_DEBUG("IH: CP EOP\n");
8071 /* XXX check the bitfield order! */
8072 me_id = (ring_id & 0x60) >> 5;
8073 pipe_id = (ring_id & 0x18) >> 3;
8074 queue_id = (ring_id & 0x7) >> 0;
8077 radeon_fence_process(rdev, RADEON_RING_TYPE_GFX_INDEX);
8081 if ((cp1_ring->me == me_id) & (cp1_ring->pipe == pipe_id))
8082 radeon_fence_process(rdev, CAYMAN_RING_TYPE_CP1_INDEX);
8083 if ((cp2_ring->me == me_id) & (cp2_ring->pipe == pipe_id))
8084 radeon_fence_process(rdev, CAYMAN_RING_TYPE_CP2_INDEX);
8088 case 184: /* CP Privileged reg access */
8089 DRM_ERROR("Illegal register access in command stream\n");
8090 /* XXX check the bitfield order! */
8091 me_id = (ring_id & 0x60) >> 5;
8092 pipe_id = (ring_id & 0x18) >> 3;
8093 queue_id = (ring_id & 0x7) >> 0;
8096 /* This results in a full GPU reset, but all we need to do is soft
8097 * reset the CP for gfx
8111 case 185: /* CP Privileged inst */
8112 DRM_ERROR("Illegal instruction in command stream\n");
8113 /* XXX check the bitfield order! */
8114 me_id = (ring_id & 0x60) >> 5;
8115 pipe_id = (ring_id & 0x18) >> 3;
8116 queue_id = (ring_id & 0x7) >> 0;
8119 /* This results in a full GPU reset, but all we need to do is soft
8120 * reset the CP for gfx
8134 case 224: /* SDMA trap event */
8135 /* XXX check the bitfield order! */
8136 me_id = (ring_id & 0x3) >> 0;
8137 queue_id = (ring_id & 0xc) >> 2;
8138 DRM_DEBUG("IH: SDMA trap\n");
8143 radeon_fence_process(rdev, R600_RING_TYPE_DMA_INDEX);
8156 radeon_fence_process(rdev, CAYMAN_RING_TYPE_DMA1_INDEX);
8168 case 230: /* thermal low to high */
8169 DRM_DEBUG("IH: thermal low to high\n");
8170 rdev->pm.dpm.thermal.high_to_low = false;
8171 queue_thermal = true;
8173 case 231: /* thermal high to low */
8174 DRM_DEBUG("IH: thermal high to low\n");
8175 rdev->pm.dpm.thermal.high_to_low = true;
8176 queue_thermal = true;
8178 case 233: /* GUI IDLE */
8179 DRM_DEBUG("IH: GUI idle\n");
8181 case 241: /* SDMA Privileged inst */
8182 case 247: /* SDMA Privileged inst */
8183 DRM_ERROR("Illegal instruction in SDMA command stream\n");
8184 /* XXX check the bitfield order! */
8185 me_id = (ring_id & 0x3) >> 0;
8186 queue_id = (ring_id & 0xc) >> 2;
8221 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
8225 /* wptr/rptr are in bytes! */
8227 rptr &= rdev->ih.ptr_mask;
8228 WREG32(IH_RB_RPTR, rptr);
8231 schedule_work(&rdev->hotplug_work);
8233 rdev->needs_reset = true;
8234 wake_up_all(&rdev->fence_queue);
8237 schedule_work(&rdev->pm.dpm.thermal.work);
8238 rdev->ih.rptr = rptr;
8239 atomic_set(&rdev->ih.lock, 0);
8241 /* make sure wptr hasn't changed while processing */
8242 wptr = cik_get_ih_wptr(rdev);
8250 * startup/shutdown callbacks
8253 * cik_startup - program the asic to a functional state
8255 * @rdev: radeon_device pointer
8257 * Programs the asic to a functional state (CIK).
8258 * Called by cik_init() and cik_resume().
8259 * Returns 0 for success, error for failure.
8261 static int cik_startup(struct radeon_device *rdev)
8263 struct radeon_ring *ring;
8267 /* enable pcie gen2/3 link */
8268 cik_pcie_gen3_enable(rdev);
8270 cik_program_aspm(rdev);
8272 /* scratch needs to be initialized before MC */
8273 r = r600_vram_scratch_init(rdev);
8277 cik_mc_program(rdev);
8279 if (!(rdev->flags & RADEON_IS_IGP) && !rdev->pm.dpm_enabled) {
8280 r = ci_mc_load_microcode(rdev);
8282 DRM_ERROR("Failed to load MC firmware!\n");
8287 r = cik_pcie_gart_enable(rdev);
8292 /* allocate rlc buffers */
8293 if (rdev->flags & RADEON_IS_IGP) {
8294 if (rdev->family == CHIP_KAVERI) {
8295 rdev->rlc.reg_list = spectre_rlc_save_restore_register_list;
8296 rdev->rlc.reg_list_size =
8297 (u32)ARRAY_SIZE(spectre_rlc_save_restore_register_list);
8299 rdev->rlc.reg_list = kalindi_rlc_save_restore_register_list;
8300 rdev->rlc.reg_list_size =
8301 (u32)ARRAY_SIZE(kalindi_rlc_save_restore_register_list);
8304 rdev->rlc.cs_data = ci_cs_data;
8305 rdev->rlc.cp_table_size = CP_ME_TABLE_SIZE * 5 * 4;
8306 r = sumo_rlc_init(rdev);
8308 DRM_ERROR("Failed to init rlc BOs!\n");
8312 /* allocate wb buffer */
8313 r = radeon_wb_init(rdev);
8317 /* allocate mec buffers */
8318 r = cik_mec_init(rdev);
8320 DRM_ERROR("Failed to init MEC BOs!\n");
8324 r = radeon_fence_driver_start_ring(rdev, RADEON_RING_TYPE_GFX_INDEX);
8326 dev_err(rdev->dev, "failed initializing CP fences (%d).\n", r);
8330 r = radeon_fence_driver_start_ring(rdev, CAYMAN_RING_TYPE_CP1_INDEX);
8332 dev_err(rdev->dev, "failed initializing CP fences (%d).\n", r);
8336 r = radeon_fence_driver_start_ring(rdev, CAYMAN_RING_TYPE_CP2_INDEX);
8338 dev_err(rdev->dev, "failed initializing CP fences (%d).\n", r);
8342 r = radeon_fence_driver_start_ring(rdev, R600_RING_TYPE_DMA_INDEX);
8344 dev_err(rdev->dev, "failed initializing DMA fences (%d).\n", r);
8348 r = radeon_fence_driver_start_ring(rdev, CAYMAN_RING_TYPE_DMA1_INDEX);
8350 dev_err(rdev->dev, "failed initializing DMA fences (%d).\n", r);
8354 r = radeon_uvd_resume(rdev);
8356 r = uvd_v4_2_resume(rdev);
8358 r = radeon_fence_driver_start_ring(rdev,
8359 R600_RING_TYPE_UVD_INDEX);
8361 dev_err(rdev->dev, "UVD fences init error (%d).\n", r);
8365 rdev->ring[R600_RING_TYPE_UVD_INDEX].ring_size = 0;
8367 r = radeon_vce_resume(rdev);
8369 r = vce_v2_0_resume(rdev);
8371 r = radeon_fence_driver_start_ring(rdev,
8372 TN_RING_TYPE_VCE1_INDEX);
8374 r = radeon_fence_driver_start_ring(rdev,
8375 TN_RING_TYPE_VCE2_INDEX);
8378 dev_err(rdev->dev, "VCE init error (%d).\n", r);
8379 rdev->ring[TN_RING_TYPE_VCE1_INDEX].ring_size = 0;
8380 rdev->ring[TN_RING_TYPE_VCE2_INDEX].ring_size = 0;
8384 if (!rdev->irq.installed) {
8385 r = radeon_irq_kms_init(rdev);
8390 r = cik_irq_init(rdev);
8392 DRM_ERROR("radeon: IH init failed (%d).\n", r);
8393 radeon_irq_kms_fini(rdev);
8398 if (rdev->family == CHIP_HAWAII) {
8400 nop = PACKET3(PACKET3_NOP, 0x3FFF);
8402 nop = RADEON_CP_PACKET2;
8404 nop = PACKET3(PACKET3_NOP, 0x3FFF);
8407 ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
8408 r = radeon_ring_init(rdev, ring, ring->ring_size, RADEON_WB_CP_RPTR_OFFSET,
8413 /* set up the compute queues */
8414 /* type-2 packets are deprecated on MEC, use type-3 instead */
8415 ring = &rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX];
8416 r = radeon_ring_init(rdev, ring, ring->ring_size, RADEON_WB_CP1_RPTR_OFFSET,
8420 ring->me = 1; /* first MEC */
8421 ring->pipe = 0; /* first pipe */
8422 ring->queue = 0; /* first queue */
8423 ring->wptr_offs = CIK_WB_CP1_WPTR_OFFSET;
8425 /* type-2 packets are deprecated on MEC, use type-3 instead */
8426 ring = &rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX];
8427 r = radeon_ring_init(rdev, ring, ring->ring_size, RADEON_WB_CP2_RPTR_OFFSET,
8431 /* dGPU only have 1 MEC */
8432 ring->me = 1; /* first MEC */
8433 ring->pipe = 0; /* first pipe */
8434 ring->queue = 1; /* second queue */
8435 ring->wptr_offs = CIK_WB_CP2_WPTR_OFFSET;
8437 ring = &rdev->ring[R600_RING_TYPE_DMA_INDEX];
8438 r = radeon_ring_init(rdev, ring, ring->ring_size, R600_WB_DMA_RPTR_OFFSET,
8439 SDMA_PACKET(SDMA_OPCODE_NOP, 0, 0));
8443 ring = &rdev->ring[CAYMAN_RING_TYPE_DMA1_INDEX];
8444 r = radeon_ring_init(rdev, ring, ring->ring_size, CAYMAN_WB_DMA1_RPTR_OFFSET,
8445 SDMA_PACKET(SDMA_OPCODE_NOP, 0, 0));
8449 r = cik_cp_resume(rdev);
8453 r = cik_sdma_resume(rdev);
8457 ring = &rdev->ring[R600_RING_TYPE_UVD_INDEX];
8458 if (ring->ring_size) {
8459 r = radeon_ring_init(rdev, ring, ring->ring_size, 0,
8462 r = uvd_v1_0_init(rdev);
8464 DRM_ERROR("radeon: failed initializing UVD (%d).\n", r);
8469 ring = &rdev->ring[TN_RING_TYPE_VCE1_INDEX];
8470 if (ring->ring_size)
8471 r = radeon_ring_init(rdev, ring, ring->ring_size, 0,
8474 ring = &rdev->ring[TN_RING_TYPE_VCE2_INDEX];
8475 if (ring->ring_size)
8476 r = radeon_ring_init(rdev, ring, ring->ring_size, 0,
8480 r = vce_v1_0_init(rdev);
8481 else if (r != -ENOENT)
8482 DRM_ERROR("radeon: failed initializing VCE (%d).\n", r);
8484 r = radeon_ib_pool_init(rdev);
8486 dev_err(rdev->dev, "IB initialization failed (%d).\n", r);
8490 r = radeon_vm_manager_init(rdev);
8492 dev_err(rdev->dev, "vm manager initialization failed (%d).\n", r);
8496 r = dce6_audio_init(rdev);
8500 r = radeon_kfd_resume(rdev);
8508 * cik_resume - resume the asic to a functional state
8510 * @rdev: radeon_device pointer
8512 * Programs the asic to a functional state (CIK).
8514 * Returns 0 for success, error for failure.
8516 int cik_resume(struct radeon_device *rdev)
8521 atom_asic_init(rdev->mode_info.atom_context);
8523 /* init golden registers */
8524 cik_init_golden_registers(rdev);
8526 if (rdev->pm.pm_method == PM_METHOD_DPM)
8527 radeon_pm_resume(rdev);
8529 rdev->accel_working = true;
8530 r = cik_startup(rdev);
8532 DRM_ERROR("cik startup failed on resume\n");
8533 rdev->accel_working = false;
8542 * cik_suspend - suspend the asic
8544 * @rdev: radeon_device pointer
8546 * Bring the chip into a state suitable for suspend (CIK).
8547 * Called at suspend.
8548 * Returns 0 for success.
8550 int cik_suspend(struct radeon_device *rdev)
8552 radeon_kfd_suspend(rdev);
8553 radeon_pm_suspend(rdev);
8554 dce6_audio_fini(rdev);
8555 radeon_vm_manager_fini(rdev);
8556 cik_cp_enable(rdev, false);
8557 cik_sdma_enable(rdev, false);
8558 uvd_v1_0_fini(rdev);
8559 radeon_uvd_suspend(rdev);
8560 radeon_vce_suspend(rdev);
8563 cik_irq_suspend(rdev);
8564 radeon_wb_disable(rdev);
8565 cik_pcie_gart_disable(rdev);
8569 /* Plan is to move initialization in that function and use
8570 * helper function so that radeon_device_init pretty much
8571 * do nothing more than calling asic specific function. This
8572 * should also allow to remove a bunch of callback function
8576 * cik_init - asic specific driver and hw init
8578 * @rdev: radeon_device pointer
8580 * Setup asic specific driver variables and program the hw
8581 * to a functional state (CIK).
8582 * Called at driver startup.
8583 * Returns 0 for success, errors for failure.
8585 int cik_init(struct radeon_device *rdev)
8587 struct radeon_ring *ring;
8591 if (!radeon_get_bios(rdev)) {
8592 if (ASIC_IS_AVIVO(rdev))
8595 /* Must be an ATOMBIOS */
8596 if (!rdev->is_atom_bios) {
8597 dev_err(rdev->dev, "Expecting atombios for cayman GPU\n");
8600 r = radeon_atombios_init(rdev);
8604 /* Post card if necessary */
8605 if (!radeon_card_posted(rdev)) {
8607 dev_err(rdev->dev, "Card not posted and no BIOS - ignoring\n");
8610 DRM_INFO("GPU not posted. posting now...\n");
8611 atom_asic_init(rdev->mode_info.atom_context);
8613 /* init golden registers */
8614 cik_init_golden_registers(rdev);
8615 /* Initialize scratch registers */
8616 cik_scratch_init(rdev);
8617 /* Initialize surface registers */
8618 radeon_surface_init(rdev);
8619 /* Initialize clocks */
8620 radeon_get_clock_info(rdev->ddev);
8623 r = radeon_fence_driver_init(rdev);
8627 /* initialize memory controller */
8628 r = cik_mc_init(rdev);
8631 /* Memory manager */
8632 r = radeon_bo_init(rdev);
8636 if (rdev->flags & RADEON_IS_IGP) {
8637 if (!rdev->me_fw || !rdev->pfp_fw || !rdev->ce_fw ||
8638 !rdev->mec_fw || !rdev->sdma_fw || !rdev->rlc_fw) {
8639 r = cik_init_microcode(rdev);
8641 DRM_ERROR("Failed to load firmware!\n");
8646 if (!rdev->me_fw || !rdev->pfp_fw || !rdev->ce_fw ||
8647 !rdev->mec_fw || !rdev->sdma_fw || !rdev->rlc_fw ||
8649 r = cik_init_microcode(rdev);
8651 DRM_ERROR("Failed to load firmware!\n");
8657 /* Initialize power management */
8658 radeon_pm_init(rdev);
8660 ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
8661 ring->ring_obj = NULL;
8662 r600_ring_init(rdev, ring, 1024 * 1024);
8664 ring = &rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX];
8665 ring->ring_obj = NULL;
8666 r600_ring_init(rdev, ring, 1024 * 1024);
8667 r = radeon_doorbell_get(rdev, &ring->doorbell_index);
8671 ring = &rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX];
8672 ring->ring_obj = NULL;
8673 r600_ring_init(rdev, ring, 1024 * 1024);
8674 r = radeon_doorbell_get(rdev, &ring->doorbell_index);
8678 ring = &rdev->ring[R600_RING_TYPE_DMA_INDEX];
8679 ring->ring_obj = NULL;
8680 r600_ring_init(rdev, ring, 256 * 1024);
8682 ring = &rdev->ring[CAYMAN_RING_TYPE_DMA1_INDEX];
8683 ring->ring_obj = NULL;
8684 r600_ring_init(rdev, ring, 256 * 1024);
8686 r = radeon_uvd_init(rdev);
8688 ring = &rdev->ring[R600_RING_TYPE_UVD_INDEX];
8689 ring->ring_obj = NULL;
8690 r600_ring_init(rdev, ring, 4096);
8693 r = radeon_vce_init(rdev);
8695 ring = &rdev->ring[TN_RING_TYPE_VCE1_INDEX];
8696 ring->ring_obj = NULL;
8697 r600_ring_init(rdev, ring, 4096);
8699 ring = &rdev->ring[TN_RING_TYPE_VCE2_INDEX];
8700 ring->ring_obj = NULL;
8701 r600_ring_init(rdev, ring, 4096);
8704 rdev->ih.ring_obj = NULL;
8705 r600_ih_ring_init(rdev, 64 * 1024);
8707 r = r600_pcie_gart_init(rdev);
8711 rdev->accel_working = true;
8712 r = cik_startup(rdev);
8714 dev_err(rdev->dev, "disabling GPU acceleration\n");
8716 cik_sdma_fini(rdev);
8718 sumo_rlc_fini(rdev);
8720 radeon_wb_fini(rdev);
8721 radeon_ib_pool_fini(rdev);
8722 radeon_vm_manager_fini(rdev);
8723 radeon_irq_kms_fini(rdev);
8724 cik_pcie_gart_fini(rdev);
8725 rdev->accel_working = false;
8728 /* Don't start up if the MC ucode is missing.
8729 * The default clocks and voltages before the MC ucode
8730 * is loaded are not suffient for advanced operations.
8732 if (!rdev->mc_fw && !(rdev->flags & RADEON_IS_IGP)) {
8733 DRM_ERROR("radeon: MC ucode required for NI+.\n");
8741 * cik_fini - asic specific driver and hw fini
8743 * @rdev: radeon_device pointer
8745 * Tear down the asic specific driver variables and program the hw
8746 * to an idle state (CIK).
8747 * Called at driver unload.
8749 void cik_fini(struct radeon_device *rdev)
8751 radeon_pm_fini(rdev);
8753 cik_sdma_fini(rdev);
8757 sumo_rlc_fini(rdev);
8759 radeon_wb_fini(rdev);
8760 radeon_vm_manager_fini(rdev);
8761 radeon_ib_pool_fini(rdev);
8762 radeon_irq_kms_fini(rdev);
8763 uvd_v1_0_fini(rdev);
8764 radeon_uvd_fini(rdev);
8765 radeon_vce_fini(rdev);
8766 cik_pcie_gart_fini(rdev);
8767 r600_vram_scratch_fini(rdev);
8768 radeon_gem_fini(rdev);
8769 radeon_fence_driver_fini(rdev);
8770 radeon_bo_fini(rdev);
8771 radeon_atombios_fini(rdev);
8776 void dce8_program_fmt(struct drm_encoder *encoder)
8778 struct drm_device *dev = encoder->dev;
8779 struct radeon_device *rdev = dev->dev_private;
8780 struct radeon_encoder *radeon_encoder = to_radeon_encoder(encoder);
8781 struct radeon_crtc *radeon_crtc = to_radeon_crtc(encoder->crtc);
8782 struct drm_connector *connector = radeon_get_connector_for_encoder(encoder);
8785 enum radeon_connector_dither dither = RADEON_FMT_DITHER_DISABLE;
8788 struct radeon_connector *radeon_connector = to_radeon_connector(connector);
8789 bpc = radeon_get_monitor_bpc(connector);
8790 dither = radeon_connector->dither;
8793 /* LVDS/eDP FMT is set up by atom */
8794 if (radeon_encoder->devices & ATOM_DEVICE_LCD_SUPPORT)
8797 /* not needed for analog */
8798 if ((radeon_encoder->encoder_id == ENCODER_OBJECT_ID_INTERNAL_KLDSCP_DAC1) ||
8799 (radeon_encoder->encoder_id == ENCODER_OBJECT_ID_INTERNAL_KLDSCP_DAC2))
8807 if (dither == RADEON_FMT_DITHER_ENABLE)
8808 /* XXX sort out optimal dither settings */
8809 tmp |= (FMT_FRAME_RANDOM_ENABLE | FMT_HIGHPASS_RANDOM_ENABLE |
8810 FMT_SPATIAL_DITHER_EN | FMT_SPATIAL_DITHER_DEPTH(0));
8812 tmp |= (FMT_TRUNCATE_EN | FMT_TRUNCATE_DEPTH(0));
8815 if (dither == RADEON_FMT_DITHER_ENABLE)
8816 /* XXX sort out optimal dither settings */
8817 tmp |= (FMT_FRAME_RANDOM_ENABLE | FMT_HIGHPASS_RANDOM_ENABLE |
8818 FMT_RGB_RANDOM_ENABLE |
8819 FMT_SPATIAL_DITHER_EN | FMT_SPATIAL_DITHER_DEPTH(1));
8821 tmp |= (FMT_TRUNCATE_EN | FMT_TRUNCATE_DEPTH(1));
8824 if (dither == RADEON_FMT_DITHER_ENABLE)
8825 /* XXX sort out optimal dither settings */
8826 tmp |= (FMT_FRAME_RANDOM_ENABLE | FMT_HIGHPASS_RANDOM_ENABLE |
8827 FMT_RGB_RANDOM_ENABLE |
8828 FMT_SPATIAL_DITHER_EN | FMT_SPATIAL_DITHER_DEPTH(2));
8830 tmp |= (FMT_TRUNCATE_EN | FMT_TRUNCATE_DEPTH(2));
8837 WREG32(FMT_BIT_DEPTH_CONTROL + radeon_crtc->crtc_offset, tmp);
8840 /* display watermark setup */
8842 * dce8_line_buffer_adjust - Set up the line buffer
8844 * @rdev: radeon_device pointer
8845 * @radeon_crtc: the selected display controller
8846 * @mode: the current display mode on the selected display
8849 * Setup up the line buffer allocation for
8850 * the selected display controller (CIK).
8851 * Returns the line buffer size in pixels.
8853 static u32 dce8_line_buffer_adjust(struct radeon_device *rdev,
8854 struct radeon_crtc *radeon_crtc,
8855 struct drm_display_mode *mode)
8857 u32 tmp, buffer_alloc, i;
8858 u32 pipe_offset = radeon_crtc->crtc_id * 0x20;
8861 * There are 6 line buffers, one for each display controllers.
8862 * There are 3 partitions per LB. Select the number of partitions
8863 * to enable based on the display width. For display widths larger
8864 * than 4096, you need use to use 2 display controllers and combine
8865 * them using the stereo blender.
8867 if (radeon_crtc->base.enabled && mode) {
8868 if (mode->crtc_hdisplay < 1920) {
8871 } else if (mode->crtc_hdisplay < 2560) {
8874 } else if (mode->crtc_hdisplay < 4096) {
8876 buffer_alloc = (rdev->flags & RADEON_IS_IGP) ? 2 : 4;
8878 DRM_DEBUG_KMS("Mode too big for LB!\n");
8880 buffer_alloc = (rdev->flags & RADEON_IS_IGP) ? 2 : 4;
8887 WREG32(LB_MEMORY_CTRL + radeon_crtc->crtc_offset,
8888 LB_MEMORY_CONFIG(tmp) | LB_MEMORY_SIZE(0x6B0));
8890 WREG32(PIPE0_DMIF_BUFFER_CONTROL + pipe_offset,
8891 DMIF_BUFFERS_ALLOCATED(buffer_alloc));
8892 for (i = 0; i < rdev->usec_timeout; i++) {
8893 if (RREG32(PIPE0_DMIF_BUFFER_CONTROL + pipe_offset) &
8894 DMIF_BUFFERS_ALLOCATED_COMPLETED)
8899 if (radeon_crtc->base.enabled && mode) {
8911 /* controller not enabled, so no lb used */
8916 * cik_get_number_of_dram_channels - get the number of dram channels
8918 * @rdev: radeon_device pointer
8920 * Look up the number of video ram channels (CIK).
8921 * Used for display watermark bandwidth calculations
8922 * Returns the number of dram channels
8924 static u32 cik_get_number_of_dram_channels(struct radeon_device *rdev)
8926 u32 tmp = RREG32(MC_SHARED_CHMAP);
8928 switch ((tmp & NOOFCHAN_MASK) >> NOOFCHAN_SHIFT) {
8951 struct dce8_wm_params {
8952 u32 dram_channels; /* number of dram channels */
8953 u32 yclk; /* bandwidth per dram data pin in kHz */
8954 u32 sclk; /* engine clock in kHz */
8955 u32 disp_clk; /* display clock in kHz */
8956 u32 src_width; /* viewport width */
8957 u32 active_time; /* active display time in ns */
8958 u32 blank_time; /* blank time in ns */
8959 bool interlaced; /* mode is interlaced */
8960 fixed20_12 vsc; /* vertical scale ratio */
8961 u32 num_heads; /* number of active crtcs */
8962 u32 bytes_per_pixel; /* bytes per pixel display + overlay */
8963 u32 lb_size; /* line buffer allocated to pipe */
8964 u32 vtaps; /* vertical scaler taps */
8968 * dce8_dram_bandwidth - get the dram bandwidth
8970 * @wm: watermark calculation data
8972 * Calculate the raw dram bandwidth (CIK).
8973 * Used for display watermark bandwidth calculations
8974 * Returns the dram bandwidth in MBytes/s
8976 static u32 dce8_dram_bandwidth(struct dce8_wm_params *wm)
8978 /* Calculate raw DRAM Bandwidth */
8979 fixed20_12 dram_efficiency; /* 0.7 */
8980 fixed20_12 yclk, dram_channels, bandwidth;
8983 a.full = dfixed_const(1000);
8984 yclk.full = dfixed_const(wm->yclk);
8985 yclk.full = dfixed_div(yclk, a);
8986 dram_channels.full = dfixed_const(wm->dram_channels * 4);
8987 a.full = dfixed_const(10);
8988 dram_efficiency.full = dfixed_const(7);
8989 dram_efficiency.full = dfixed_div(dram_efficiency, a);
8990 bandwidth.full = dfixed_mul(dram_channels, yclk);
8991 bandwidth.full = dfixed_mul(bandwidth, dram_efficiency);
8993 return dfixed_trunc(bandwidth);
8997 * dce8_dram_bandwidth_for_display - get the dram bandwidth for display
8999 * @wm: watermark calculation data
9001 * Calculate the dram bandwidth used for display (CIK).
9002 * Used for display watermark bandwidth calculations
9003 * Returns the dram bandwidth for display in MBytes/s
9005 static u32 dce8_dram_bandwidth_for_display(struct dce8_wm_params *wm)
9007 /* Calculate DRAM Bandwidth and the part allocated to display. */
9008 fixed20_12 disp_dram_allocation; /* 0.3 to 0.7 */
9009 fixed20_12 yclk, dram_channels, bandwidth;
9012 a.full = dfixed_const(1000);
9013 yclk.full = dfixed_const(wm->yclk);
9014 yclk.full = dfixed_div(yclk, a);
9015 dram_channels.full = dfixed_const(wm->dram_channels * 4);
9016 a.full = dfixed_const(10);
9017 disp_dram_allocation.full = dfixed_const(3); /* XXX worse case value 0.3 */
9018 disp_dram_allocation.full = dfixed_div(disp_dram_allocation, a);
9019 bandwidth.full = dfixed_mul(dram_channels, yclk);
9020 bandwidth.full = dfixed_mul(bandwidth, disp_dram_allocation);
9022 return dfixed_trunc(bandwidth);
9026 * dce8_data_return_bandwidth - get the data return bandwidth
9028 * @wm: watermark calculation data
9030 * Calculate the data return bandwidth used for display (CIK).
9031 * Used for display watermark bandwidth calculations
9032 * Returns the data return bandwidth in MBytes/s
9034 static u32 dce8_data_return_bandwidth(struct dce8_wm_params *wm)
9036 /* Calculate the display Data return Bandwidth */
9037 fixed20_12 return_efficiency; /* 0.8 */
9038 fixed20_12 sclk, bandwidth;
9041 a.full = dfixed_const(1000);
9042 sclk.full = dfixed_const(wm->sclk);
9043 sclk.full = dfixed_div(sclk, a);
9044 a.full = dfixed_const(10);
9045 return_efficiency.full = dfixed_const(8);
9046 return_efficiency.full = dfixed_div(return_efficiency, a);
9047 a.full = dfixed_const(32);
9048 bandwidth.full = dfixed_mul(a, sclk);
9049 bandwidth.full = dfixed_mul(bandwidth, return_efficiency);
9051 return dfixed_trunc(bandwidth);
9055 * dce8_dmif_request_bandwidth - get the dmif bandwidth
9057 * @wm: watermark calculation data
9059 * Calculate the dmif bandwidth used for display (CIK).
9060 * Used for display watermark bandwidth calculations
9061 * Returns the dmif bandwidth in MBytes/s
9063 static u32 dce8_dmif_request_bandwidth(struct dce8_wm_params *wm)
9065 /* Calculate the DMIF Request Bandwidth */
9066 fixed20_12 disp_clk_request_efficiency; /* 0.8 */
9067 fixed20_12 disp_clk, bandwidth;
9070 a.full = dfixed_const(1000);
9071 disp_clk.full = dfixed_const(wm->disp_clk);
9072 disp_clk.full = dfixed_div(disp_clk, a);
9073 a.full = dfixed_const(32);
9074 b.full = dfixed_mul(a, disp_clk);
9076 a.full = dfixed_const(10);
9077 disp_clk_request_efficiency.full = dfixed_const(8);
9078 disp_clk_request_efficiency.full = dfixed_div(disp_clk_request_efficiency, a);
9080 bandwidth.full = dfixed_mul(b, disp_clk_request_efficiency);
9082 return dfixed_trunc(bandwidth);
9086 * dce8_available_bandwidth - get the min available bandwidth
9088 * @wm: watermark calculation data
9090 * Calculate the min available bandwidth used for display (CIK).
9091 * Used for display watermark bandwidth calculations
9092 * Returns the min available bandwidth in MBytes/s
9094 static u32 dce8_available_bandwidth(struct dce8_wm_params *wm)
9096 /* Calculate the Available bandwidth. Display can use this temporarily but not in average. */
9097 u32 dram_bandwidth = dce8_dram_bandwidth(wm);
9098 u32 data_return_bandwidth = dce8_data_return_bandwidth(wm);
9099 u32 dmif_req_bandwidth = dce8_dmif_request_bandwidth(wm);
9101 return min(dram_bandwidth, min(data_return_bandwidth, dmif_req_bandwidth));
9105 * dce8_average_bandwidth - get the average available bandwidth
9107 * @wm: watermark calculation data
9109 * Calculate the average available bandwidth used for display (CIK).
9110 * Used for display watermark bandwidth calculations
9111 * Returns the average available bandwidth in MBytes/s
9113 static u32 dce8_average_bandwidth(struct dce8_wm_params *wm)
9115 /* Calculate the display mode Average Bandwidth
9116 * DisplayMode should contain the source and destination dimensions,
9120 fixed20_12 line_time;
9121 fixed20_12 src_width;
9122 fixed20_12 bandwidth;
9125 a.full = dfixed_const(1000);
9126 line_time.full = dfixed_const(wm->active_time + wm->blank_time);
9127 line_time.full = dfixed_div(line_time, a);
9128 bpp.full = dfixed_const(wm->bytes_per_pixel);
9129 src_width.full = dfixed_const(wm->src_width);
9130 bandwidth.full = dfixed_mul(src_width, bpp);
9131 bandwidth.full = dfixed_mul(bandwidth, wm->vsc);
9132 bandwidth.full = dfixed_div(bandwidth, line_time);
9134 return dfixed_trunc(bandwidth);
9138 * dce8_latency_watermark - get the latency watermark
9140 * @wm: watermark calculation data
9142 * Calculate the latency watermark (CIK).
9143 * Used for display watermark bandwidth calculations
9144 * Returns the latency watermark in ns
9146 static u32 dce8_latency_watermark(struct dce8_wm_params *wm)
9148 /* First calculate the latency in ns */
9149 u32 mc_latency = 2000; /* 2000 ns. */
9150 u32 available_bandwidth = dce8_available_bandwidth(wm);
9151 u32 worst_chunk_return_time = (512 * 8 * 1000) / available_bandwidth;
9152 u32 cursor_line_pair_return_time = (128 * 4 * 1000) / available_bandwidth;
9153 u32 dc_latency = 40000000 / wm->disp_clk; /* dc pipe latency */
9154 u32 other_heads_data_return_time = ((wm->num_heads + 1) * worst_chunk_return_time) +
9155 (wm->num_heads * cursor_line_pair_return_time);
9156 u32 latency = mc_latency + other_heads_data_return_time + dc_latency;
9157 u32 max_src_lines_per_dst_line, lb_fill_bw, line_fill_time;
9158 u32 tmp, dmif_size = 12288;
9161 if (wm->num_heads == 0)
9164 a.full = dfixed_const(2);
9165 b.full = dfixed_const(1);
9166 if ((wm->vsc.full > a.full) ||
9167 ((wm->vsc.full > b.full) && (wm->vtaps >= 3)) ||
9169 ((wm->vsc.full >= a.full) && wm->interlaced))
9170 max_src_lines_per_dst_line = 4;
9172 max_src_lines_per_dst_line = 2;
9174 a.full = dfixed_const(available_bandwidth);
9175 b.full = dfixed_const(wm->num_heads);
9176 a.full = dfixed_div(a, b);
9178 b.full = dfixed_const(mc_latency + 512);
9179 c.full = dfixed_const(wm->disp_clk);
9180 b.full = dfixed_div(b, c);
9182 c.full = dfixed_const(dmif_size);
9183 b.full = dfixed_div(c, b);
9185 tmp = min(dfixed_trunc(a), dfixed_trunc(b));
9187 b.full = dfixed_const(1000);
9188 c.full = dfixed_const(wm->disp_clk);
9189 b.full = dfixed_div(c, b);
9190 c.full = dfixed_const(wm->bytes_per_pixel);
9191 b.full = dfixed_mul(b, c);
9193 lb_fill_bw = min(tmp, dfixed_trunc(b));
9195 a.full = dfixed_const(max_src_lines_per_dst_line * wm->src_width * wm->bytes_per_pixel);
9196 b.full = dfixed_const(1000);
9197 c.full = dfixed_const(lb_fill_bw);
9198 b.full = dfixed_div(c, b);
9199 a.full = dfixed_div(a, b);
9200 line_fill_time = dfixed_trunc(a);
9202 if (line_fill_time < wm->active_time)
9205 return latency + (line_fill_time - wm->active_time);
9210 * dce8_average_bandwidth_vs_dram_bandwidth_for_display - check
9211 * average and available dram bandwidth
9213 * @wm: watermark calculation data
9215 * Check if the display average bandwidth fits in the display
9216 * dram bandwidth (CIK).
9217 * Used for display watermark bandwidth calculations
9218 * Returns true if the display fits, false if not.
9220 static bool dce8_average_bandwidth_vs_dram_bandwidth_for_display(struct dce8_wm_params *wm)
9222 if (dce8_average_bandwidth(wm) <=
9223 (dce8_dram_bandwidth_for_display(wm) / wm->num_heads))
9230 * dce8_average_bandwidth_vs_available_bandwidth - check
9231 * average and available bandwidth
9233 * @wm: watermark calculation data
9235 * Check if the display average bandwidth fits in the display
9236 * available bandwidth (CIK).
9237 * Used for display watermark bandwidth calculations
9238 * Returns true if the display fits, false if not.
9240 static bool dce8_average_bandwidth_vs_available_bandwidth(struct dce8_wm_params *wm)
9242 if (dce8_average_bandwidth(wm) <=
9243 (dce8_available_bandwidth(wm) / wm->num_heads))
9250 * dce8_check_latency_hiding - check latency hiding
9252 * @wm: watermark calculation data
9254 * Check latency hiding (CIK).
9255 * Used for display watermark bandwidth calculations
9256 * Returns true if the display fits, false if not.
9258 static bool dce8_check_latency_hiding(struct dce8_wm_params *wm)
9260 u32 lb_partitions = wm->lb_size / wm->src_width;
9261 u32 line_time = wm->active_time + wm->blank_time;
9262 u32 latency_tolerant_lines;
9266 a.full = dfixed_const(1);
9267 if (wm->vsc.full > a.full)
9268 latency_tolerant_lines = 1;
9270 if (lb_partitions <= (wm->vtaps + 1))
9271 latency_tolerant_lines = 1;
9273 latency_tolerant_lines = 2;
9276 latency_hiding = (latency_tolerant_lines * line_time + wm->blank_time);
9278 if (dce8_latency_watermark(wm) <= latency_hiding)
9285 * dce8_program_watermarks - program display watermarks
9287 * @rdev: radeon_device pointer
9288 * @radeon_crtc: the selected display controller
9289 * @lb_size: line buffer size
9290 * @num_heads: number of display controllers in use
9292 * Calculate and program the display watermarks for the
9293 * selected display controller (CIK).
9295 static void dce8_program_watermarks(struct radeon_device *rdev,
9296 struct radeon_crtc *radeon_crtc,
9297 u32 lb_size, u32 num_heads)
9299 struct drm_display_mode *mode = &radeon_crtc->base.mode;
9300 struct dce8_wm_params wm_low, wm_high;
9303 u32 latency_watermark_a = 0, latency_watermark_b = 0;
9306 if (radeon_crtc->base.enabled && num_heads && mode) {
9307 pixel_period = 1000000 / (u32)mode->clock;
9308 line_time = min((u32)mode->crtc_htotal * pixel_period, (u32)65535);
9310 /* watermark for high clocks */
9311 if ((rdev->pm.pm_method == PM_METHOD_DPM) &&
9312 rdev->pm.dpm_enabled) {
9314 radeon_dpm_get_mclk(rdev, false) * 10;
9316 radeon_dpm_get_sclk(rdev, false) * 10;
9318 wm_high.yclk = rdev->pm.current_mclk * 10;
9319 wm_high.sclk = rdev->pm.current_sclk * 10;
9322 wm_high.disp_clk = mode->clock;
9323 wm_high.src_width = mode->crtc_hdisplay;
9324 wm_high.active_time = mode->crtc_hdisplay * pixel_period;
9325 wm_high.blank_time = line_time - wm_high.active_time;
9326 wm_high.interlaced = false;
9327 if (mode->flags & DRM_MODE_FLAG_INTERLACE)
9328 wm_high.interlaced = true;
9329 wm_high.vsc = radeon_crtc->vsc;
9331 if (radeon_crtc->rmx_type != RMX_OFF)
9333 wm_high.bytes_per_pixel = 4; /* XXX: get this from fb config */
9334 wm_high.lb_size = lb_size;
9335 wm_high.dram_channels = cik_get_number_of_dram_channels(rdev);
9336 wm_high.num_heads = num_heads;
9338 /* set for high clocks */
9339 latency_watermark_a = min(dce8_latency_watermark(&wm_high), (u32)65535);
9341 /* possibly force display priority to high */
9342 /* should really do this at mode validation time... */
9343 if (!dce8_average_bandwidth_vs_dram_bandwidth_for_display(&wm_high) ||
9344 !dce8_average_bandwidth_vs_available_bandwidth(&wm_high) ||
9345 !dce8_check_latency_hiding(&wm_high) ||
9346 (rdev->disp_priority == 2)) {
9347 DRM_DEBUG_KMS("force priority to high\n");
9350 /* watermark for low clocks */
9351 if ((rdev->pm.pm_method == PM_METHOD_DPM) &&
9352 rdev->pm.dpm_enabled) {
9354 radeon_dpm_get_mclk(rdev, true) * 10;
9356 radeon_dpm_get_sclk(rdev, true) * 10;
9358 wm_low.yclk = rdev->pm.current_mclk * 10;
9359 wm_low.sclk = rdev->pm.current_sclk * 10;
9362 wm_low.disp_clk = mode->clock;
9363 wm_low.src_width = mode->crtc_hdisplay;
9364 wm_low.active_time = mode->crtc_hdisplay * pixel_period;
9365 wm_low.blank_time = line_time - wm_low.active_time;
9366 wm_low.interlaced = false;
9367 if (mode->flags & DRM_MODE_FLAG_INTERLACE)
9368 wm_low.interlaced = true;
9369 wm_low.vsc = radeon_crtc->vsc;
9371 if (radeon_crtc->rmx_type != RMX_OFF)
9373 wm_low.bytes_per_pixel = 4; /* XXX: get this from fb config */
9374 wm_low.lb_size = lb_size;
9375 wm_low.dram_channels = cik_get_number_of_dram_channels(rdev);
9376 wm_low.num_heads = num_heads;
9378 /* set for low clocks */
9379 latency_watermark_b = min(dce8_latency_watermark(&wm_low), (u32)65535);
9381 /* possibly force display priority to high */
9382 /* should really do this at mode validation time... */
9383 if (!dce8_average_bandwidth_vs_dram_bandwidth_for_display(&wm_low) ||
9384 !dce8_average_bandwidth_vs_available_bandwidth(&wm_low) ||
9385 !dce8_check_latency_hiding(&wm_low) ||
9386 (rdev->disp_priority == 2)) {
9387 DRM_DEBUG_KMS("force priority to high\n");
9392 wm_mask = RREG32(DPG_WATERMARK_MASK_CONTROL + radeon_crtc->crtc_offset);
9394 tmp &= ~LATENCY_WATERMARK_MASK(3);
9395 tmp |= LATENCY_WATERMARK_MASK(1);
9396 WREG32(DPG_WATERMARK_MASK_CONTROL + radeon_crtc->crtc_offset, tmp);
9397 WREG32(DPG_PIPE_LATENCY_CONTROL + radeon_crtc->crtc_offset,
9398 (LATENCY_LOW_WATERMARK(latency_watermark_a) |
9399 LATENCY_HIGH_WATERMARK(line_time)));
9401 tmp = RREG32(DPG_WATERMARK_MASK_CONTROL + radeon_crtc->crtc_offset);
9402 tmp &= ~LATENCY_WATERMARK_MASK(3);
9403 tmp |= LATENCY_WATERMARK_MASK(2);
9404 WREG32(DPG_WATERMARK_MASK_CONTROL + radeon_crtc->crtc_offset, tmp);
9405 WREG32(DPG_PIPE_LATENCY_CONTROL + radeon_crtc->crtc_offset,
9406 (LATENCY_LOW_WATERMARK(latency_watermark_b) |
9407 LATENCY_HIGH_WATERMARK(line_time)));
9408 /* restore original selection */
9409 WREG32(DPG_WATERMARK_MASK_CONTROL + radeon_crtc->crtc_offset, wm_mask);
9411 /* save values for DPM */
9412 radeon_crtc->line_time = line_time;
9413 radeon_crtc->wm_high = latency_watermark_a;
9414 radeon_crtc->wm_low = latency_watermark_b;
9418 * dce8_bandwidth_update - program display watermarks
9420 * @rdev: radeon_device pointer
9422 * Calculate and program the display watermarks and line
9423 * buffer allocation (CIK).
9425 void dce8_bandwidth_update(struct radeon_device *rdev)
9427 struct drm_display_mode *mode = NULL;
9428 u32 num_heads = 0, lb_size;
9431 if (!rdev->mode_info.mode_config_initialized)
9434 radeon_update_display_priority(rdev);
9436 for (i = 0; i < rdev->num_crtc; i++) {
9437 if (rdev->mode_info.crtcs[i]->base.enabled)
9440 for (i = 0; i < rdev->num_crtc; i++) {
9441 mode = &rdev->mode_info.crtcs[i]->base.mode;
9442 lb_size = dce8_line_buffer_adjust(rdev, rdev->mode_info.crtcs[i], mode);
9443 dce8_program_watermarks(rdev, rdev->mode_info.crtcs[i], lb_size, num_heads);
9448 * cik_get_gpu_clock_counter - return GPU clock counter snapshot
9450 * @rdev: radeon_device pointer
9452 * Fetches a GPU clock counter snapshot (SI).
9453 * Returns the 64 bit clock counter snapshot.
9455 uint64_t cik_get_gpu_clock_counter(struct radeon_device *rdev)
9459 mutex_lock(&rdev->gpu_clock_mutex);
9460 WREG32(RLC_CAPTURE_GPU_CLOCK_COUNT, 1);
9461 clock = (uint64_t)RREG32(RLC_GPU_CLOCK_COUNT_LSB) |
9462 ((uint64_t)RREG32(RLC_GPU_CLOCK_COUNT_MSB) << 32ULL);
9463 mutex_unlock(&rdev->gpu_clock_mutex);
9467 static int cik_set_uvd_clock(struct radeon_device *rdev, u32 clock,
9468 u32 cntl_reg, u32 status_reg)
9471 struct atom_clock_dividers dividers;
9474 r = radeon_atom_get_clock_dividers(rdev, COMPUTE_GPUCLK_INPUT_FLAG_DEFAULT_GPUCLK,
9475 clock, false, ÷rs);
9479 tmp = RREG32_SMC(cntl_reg);
9480 tmp &= ~(DCLK_DIR_CNTL_EN|DCLK_DIVIDER_MASK);
9481 tmp |= dividers.post_divider;
9482 WREG32_SMC(cntl_reg, tmp);
9484 for (i = 0; i < 100; i++) {
9485 if (RREG32_SMC(status_reg) & DCLK_STATUS)
9495 int cik_set_uvd_clocks(struct radeon_device *rdev, u32 vclk, u32 dclk)
9499 r = cik_set_uvd_clock(rdev, vclk, CG_VCLK_CNTL, CG_VCLK_STATUS);
9503 r = cik_set_uvd_clock(rdev, dclk, CG_DCLK_CNTL, CG_DCLK_STATUS);
9507 int cik_set_vce_clocks(struct radeon_device *rdev, u32 evclk, u32 ecclk)
9510 struct atom_clock_dividers dividers;
9513 r = radeon_atom_get_clock_dividers(rdev, COMPUTE_GPUCLK_INPUT_FLAG_DEFAULT_GPUCLK,
9514 ecclk, false, ÷rs);
9518 for (i = 0; i < 100; i++) {
9519 if (RREG32_SMC(CG_ECLK_STATUS) & ECLK_STATUS)
9526 tmp = RREG32_SMC(CG_ECLK_CNTL);
9527 tmp &= ~(ECLK_DIR_CNTL_EN|ECLK_DIVIDER_MASK);
9528 tmp |= dividers.post_divider;
9529 WREG32_SMC(CG_ECLK_CNTL, tmp);
9531 for (i = 0; i < 100; i++) {
9532 if (RREG32_SMC(CG_ECLK_STATUS) & ECLK_STATUS)
9542 static void cik_pcie_gen3_enable(struct radeon_device *rdev)
9544 struct pci_dev *root = rdev->pdev->bus->self;
9545 int bridge_pos, gpu_pos;
9546 u32 speed_cntl, mask, current_data_rate;
9550 if (pci_is_root_bus(rdev->pdev->bus))
9553 if (radeon_pcie_gen2 == 0)
9556 if (rdev->flags & RADEON_IS_IGP)
9559 if (!(rdev->flags & RADEON_IS_PCIE))
9562 ret = drm_pcie_get_speed_cap_mask(rdev->ddev, &mask);
9566 if (!(mask & (DRM_PCIE_SPEED_50 | DRM_PCIE_SPEED_80)))
9569 speed_cntl = RREG32_PCIE_PORT(PCIE_LC_SPEED_CNTL);
9570 current_data_rate = (speed_cntl & LC_CURRENT_DATA_RATE_MASK) >>
9571 LC_CURRENT_DATA_RATE_SHIFT;
9572 if (mask & DRM_PCIE_SPEED_80) {
9573 if (current_data_rate == 2) {
9574 DRM_INFO("PCIE gen 3 link speeds already enabled\n");
9577 DRM_INFO("enabling PCIE gen 3 link speeds, disable with radeon.pcie_gen2=0\n");
9578 } else if (mask & DRM_PCIE_SPEED_50) {
9579 if (current_data_rate == 1) {
9580 DRM_INFO("PCIE gen 2 link speeds already enabled\n");
9583 DRM_INFO("enabling PCIE gen 2 link speeds, disable with radeon.pcie_gen2=0\n");
9586 bridge_pos = pci_pcie_cap(root);
9590 gpu_pos = pci_pcie_cap(rdev->pdev);
9594 if (mask & DRM_PCIE_SPEED_80) {
9595 /* re-try equalization if gen3 is not already enabled */
9596 if (current_data_rate != 2) {
9597 u16 bridge_cfg, gpu_cfg;
9598 u16 bridge_cfg2, gpu_cfg2;
9599 u32 max_lw, current_lw, tmp;
9601 pci_read_config_word(root, bridge_pos + PCI_EXP_LNKCTL, &bridge_cfg);
9602 pci_read_config_word(rdev->pdev, gpu_pos + PCI_EXP_LNKCTL, &gpu_cfg);
9604 tmp16 = bridge_cfg | PCI_EXP_LNKCTL_HAWD;
9605 pci_write_config_word(root, bridge_pos + PCI_EXP_LNKCTL, tmp16);
9607 tmp16 = gpu_cfg | PCI_EXP_LNKCTL_HAWD;
9608 pci_write_config_word(rdev->pdev, gpu_pos + PCI_EXP_LNKCTL, tmp16);
9610 tmp = RREG32_PCIE_PORT(PCIE_LC_STATUS1);
9611 max_lw = (tmp & LC_DETECTED_LINK_WIDTH_MASK) >> LC_DETECTED_LINK_WIDTH_SHIFT;
9612 current_lw = (tmp & LC_OPERATING_LINK_WIDTH_MASK) >> LC_OPERATING_LINK_WIDTH_SHIFT;
9614 if (current_lw < max_lw) {
9615 tmp = RREG32_PCIE_PORT(PCIE_LC_LINK_WIDTH_CNTL);
9616 if (tmp & LC_RENEGOTIATION_SUPPORT) {
9617 tmp &= ~(LC_LINK_WIDTH_MASK | LC_UPCONFIGURE_DIS);
9618 tmp |= (max_lw << LC_LINK_WIDTH_SHIFT);
9619 tmp |= LC_UPCONFIGURE_SUPPORT | LC_RENEGOTIATE_EN | LC_RECONFIG_NOW;
9620 WREG32_PCIE_PORT(PCIE_LC_LINK_WIDTH_CNTL, tmp);
9624 for (i = 0; i < 10; i++) {
9626 pci_read_config_word(rdev->pdev, gpu_pos + PCI_EXP_DEVSTA, &tmp16);
9627 if (tmp16 & PCI_EXP_DEVSTA_TRPND)
9630 pci_read_config_word(root, bridge_pos + PCI_EXP_LNKCTL, &bridge_cfg);
9631 pci_read_config_word(rdev->pdev, gpu_pos + PCI_EXP_LNKCTL, &gpu_cfg);
9633 pci_read_config_word(root, bridge_pos + PCI_EXP_LNKCTL2, &bridge_cfg2);
9634 pci_read_config_word(rdev->pdev, gpu_pos + PCI_EXP_LNKCTL2, &gpu_cfg2);
9636 tmp = RREG32_PCIE_PORT(PCIE_LC_CNTL4);
9637 tmp |= LC_SET_QUIESCE;
9638 WREG32_PCIE_PORT(PCIE_LC_CNTL4, tmp);
9640 tmp = RREG32_PCIE_PORT(PCIE_LC_CNTL4);
9642 WREG32_PCIE_PORT(PCIE_LC_CNTL4, tmp);
9647 pci_read_config_word(root, bridge_pos + PCI_EXP_LNKCTL, &tmp16);
9648 tmp16 &= ~PCI_EXP_LNKCTL_HAWD;
9649 tmp16 |= (bridge_cfg & PCI_EXP_LNKCTL_HAWD);
9650 pci_write_config_word(root, bridge_pos + PCI_EXP_LNKCTL, tmp16);
9652 pci_read_config_word(rdev->pdev, gpu_pos + PCI_EXP_LNKCTL, &tmp16);
9653 tmp16 &= ~PCI_EXP_LNKCTL_HAWD;
9654 tmp16 |= (gpu_cfg & PCI_EXP_LNKCTL_HAWD);
9655 pci_write_config_word(rdev->pdev, gpu_pos + PCI_EXP_LNKCTL, tmp16);
9658 pci_read_config_word(root, bridge_pos + PCI_EXP_LNKCTL2, &tmp16);
9659 tmp16 &= ~((1 << 4) | (7 << 9));
9660 tmp16 |= (bridge_cfg2 & ((1 << 4) | (7 << 9)));
9661 pci_write_config_word(root, bridge_pos + PCI_EXP_LNKCTL2, tmp16);
9663 pci_read_config_word(rdev->pdev, gpu_pos + PCI_EXP_LNKCTL2, &tmp16);
9664 tmp16 &= ~((1 << 4) | (7 << 9));
9665 tmp16 |= (gpu_cfg2 & ((1 << 4) | (7 << 9)));
9666 pci_write_config_word(rdev->pdev, gpu_pos + PCI_EXP_LNKCTL2, tmp16);
9668 tmp = RREG32_PCIE_PORT(PCIE_LC_CNTL4);
9669 tmp &= ~LC_SET_QUIESCE;
9670 WREG32_PCIE_PORT(PCIE_LC_CNTL4, tmp);
9675 /* set the link speed */
9676 speed_cntl |= LC_FORCE_EN_SW_SPEED_CHANGE | LC_FORCE_DIS_HW_SPEED_CHANGE;
9677 speed_cntl &= ~LC_FORCE_DIS_SW_SPEED_CHANGE;
9678 WREG32_PCIE_PORT(PCIE_LC_SPEED_CNTL, speed_cntl);
9680 pci_read_config_word(rdev->pdev, gpu_pos + PCI_EXP_LNKCTL2, &tmp16);
9682 if (mask & DRM_PCIE_SPEED_80)
9683 tmp16 |= 3; /* gen3 */
9684 else if (mask & DRM_PCIE_SPEED_50)
9685 tmp16 |= 2; /* gen2 */
9687 tmp16 |= 1; /* gen1 */
9688 pci_write_config_word(rdev->pdev, gpu_pos + PCI_EXP_LNKCTL2, tmp16);
9690 speed_cntl = RREG32_PCIE_PORT(PCIE_LC_SPEED_CNTL);
9691 speed_cntl |= LC_INITIATE_LINK_SPEED_CHANGE;
9692 WREG32_PCIE_PORT(PCIE_LC_SPEED_CNTL, speed_cntl);
9694 for (i = 0; i < rdev->usec_timeout; i++) {
9695 speed_cntl = RREG32_PCIE_PORT(PCIE_LC_SPEED_CNTL);
9696 if ((speed_cntl & LC_INITIATE_LINK_SPEED_CHANGE) == 0)
9702 static void cik_program_aspm(struct radeon_device *rdev)
9705 bool disable_l0s = false, disable_l1 = false, disable_plloff_in_l1 = false;
9706 bool disable_clkreq = false;
9708 if (radeon_aspm == 0)
9711 /* XXX double check IGPs */
9712 if (rdev->flags & RADEON_IS_IGP)
9715 if (!(rdev->flags & RADEON_IS_PCIE))
9718 orig = data = RREG32_PCIE_PORT(PCIE_LC_N_FTS_CNTL);
9719 data &= ~LC_XMIT_N_FTS_MASK;
9720 data |= LC_XMIT_N_FTS(0x24) | LC_XMIT_N_FTS_OVERRIDE_EN;
9722 WREG32_PCIE_PORT(PCIE_LC_N_FTS_CNTL, data);
9724 orig = data = RREG32_PCIE_PORT(PCIE_LC_CNTL3);
9725 data |= LC_GO_TO_RECOVERY;
9727 WREG32_PCIE_PORT(PCIE_LC_CNTL3, data);
9729 orig = data = RREG32_PCIE_PORT(PCIE_P_CNTL);
9730 data |= P_IGNORE_EDB_ERR;
9732 WREG32_PCIE_PORT(PCIE_P_CNTL, data);
9734 orig = data = RREG32_PCIE_PORT(PCIE_LC_CNTL);
9735 data &= ~(LC_L0S_INACTIVITY_MASK | LC_L1_INACTIVITY_MASK);
9736 data |= LC_PMI_TO_L1_DIS;
9738 data |= LC_L0S_INACTIVITY(7);
9741 data |= LC_L1_INACTIVITY(7);
9742 data &= ~LC_PMI_TO_L1_DIS;
9744 WREG32_PCIE_PORT(PCIE_LC_CNTL, data);
9746 if (!disable_plloff_in_l1) {
9747 bool clk_req_support;
9749 orig = data = RREG32_PCIE_PORT(PB0_PIF_PWRDOWN_0);
9750 data &= ~(PLL_POWER_STATE_IN_OFF_0_MASK | PLL_POWER_STATE_IN_TXS2_0_MASK);
9751 data |= PLL_POWER_STATE_IN_OFF_0(7) | PLL_POWER_STATE_IN_TXS2_0(7);
9753 WREG32_PCIE_PORT(PB0_PIF_PWRDOWN_0, data);
9755 orig = data = RREG32_PCIE_PORT(PB0_PIF_PWRDOWN_1);
9756 data &= ~(PLL_POWER_STATE_IN_OFF_1_MASK | PLL_POWER_STATE_IN_TXS2_1_MASK);
9757 data |= PLL_POWER_STATE_IN_OFF_1(7) | PLL_POWER_STATE_IN_TXS2_1(7);
9759 WREG32_PCIE_PORT(PB0_PIF_PWRDOWN_1, data);
9761 orig = data = RREG32_PCIE_PORT(PB1_PIF_PWRDOWN_0);
9762 data &= ~(PLL_POWER_STATE_IN_OFF_0_MASK | PLL_POWER_STATE_IN_TXS2_0_MASK);
9763 data |= PLL_POWER_STATE_IN_OFF_0(7) | PLL_POWER_STATE_IN_TXS2_0(7);
9765 WREG32_PCIE_PORT(PB1_PIF_PWRDOWN_0, data);
9767 orig = data = RREG32_PCIE_PORT(PB1_PIF_PWRDOWN_1);
9768 data &= ~(PLL_POWER_STATE_IN_OFF_1_MASK | PLL_POWER_STATE_IN_TXS2_1_MASK);
9769 data |= PLL_POWER_STATE_IN_OFF_1(7) | PLL_POWER_STATE_IN_TXS2_1(7);
9771 WREG32_PCIE_PORT(PB1_PIF_PWRDOWN_1, data);
9773 orig = data = RREG32_PCIE_PORT(PCIE_LC_LINK_WIDTH_CNTL);
9774 data &= ~LC_DYN_LANES_PWR_STATE_MASK;
9775 data |= LC_DYN_LANES_PWR_STATE(3);
9777 WREG32_PCIE_PORT(PCIE_LC_LINK_WIDTH_CNTL, data);
9779 if (!disable_clkreq &&
9780 !pci_is_root_bus(rdev->pdev->bus)) {
9781 struct pci_dev *root = rdev->pdev->bus->self;
9784 clk_req_support = false;
9785 pcie_capability_read_dword(root, PCI_EXP_LNKCAP, &lnkcap);
9786 if (lnkcap & PCI_EXP_LNKCAP_CLKPM)
9787 clk_req_support = true;
9789 clk_req_support = false;
9792 if (clk_req_support) {
9793 orig = data = RREG32_PCIE_PORT(PCIE_LC_CNTL2);
9794 data |= LC_ALLOW_PDWN_IN_L1 | LC_ALLOW_PDWN_IN_L23;
9796 WREG32_PCIE_PORT(PCIE_LC_CNTL2, data);
9798 orig = data = RREG32_SMC(THM_CLK_CNTL);
9799 data &= ~(CMON_CLK_SEL_MASK | TMON_CLK_SEL_MASK);
9800 data |= CMON_CLK_SEL(1) | TMON_CLK_SEL(1);
9802 WREG32_SMC(THM_CLK_CNTL, data);
9804 orig = data = RREG32_SMC(MISC_CLK_CTRL);
9805 data &= ~(DEEP_SLEEP_CLK_SEL_MASK | ZCLK_SEL_MASK);
9806 data |= DEEP_SLEEP_CLK_SEL(1) | ZCLK_SEL(1);
9808 WREG32_SMC(MISC_CLK_CTRL, data);
9810 orig = data = RREG32_SMC(CG_CLKPIN_CNTL);
9811 data &= ~BCLK_AS_XCLK;
9813 WREG32_SMC(CG_CLKPIN_CNTL, data);
9815 orig = data = RREG32_SMC(CG_CLKPIN_CNTL_2);
9816 data &= ~FORCE_BIF_REFCLK_EN;
9818 WREG32_SMC(CG_CLKPIN_CNTL_2, data);
9820 orig = data = RREG32_SMC(MPLL_BYPASSCLK_SEL);
9821 data &= ~MPLL_CLKOUT_SEL_MASK;
9822 data |= MPLL_CLKOUT_SEL(4);
9824 WREG32_SMC(MPLL_BYPASSCLK_SEL, data);
9829 WREG32_PCIE_PORT(PCIE_LC_CNTL, data);
9832 orig = data = RREG32_PCIE_PORT(PCIE_CNTL2);
9833 data |= SLV_MEM_LS_EN | MST_MEM_LS_EN | REPLAY_MEM_LS_EN;
9835 WREG32_PCIE_PORT(PCIE_CNTL2, data);
9838 data = RREG32_PCIE_PORT(PCIE_LC_N_FTS_CNTL);
9839 if((data & LC_N_FTS_MASK) == LC_N_FTS_MASK) {
9840 data = RREG32_PCIE_PORT(PCIE_LC_STATUS1);
9841 if ((data & LC_REVERSE_XMIT) && (data & LC_REVERSE_RCVR)) {
9842 orig = data = RREG32_PCIE_PORT(PCIE_LC_CNTL);
9843 data &= ~LC_L0S_INACTIVITY_MASK;
9845 WREG32_PCIE_PORT(PCIE_LC_CNTL, data);