2 * Copyright 2014 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.
23 #include <linux/firmware.h>
26 #include "amdgpu_ih.h"
27 #include "amdgpu_gfx.h"
31 #include "amdgpu_ucode.h"
32 #include "clearstate_ci.h"
34 #include "dce/dce_8_0_d.h"
35 #include "dce/dce_8_0_sh_mask.h"
37 #include "bif/bif_4_1_d.h"
38 #include "bif/bif_4_1_sh_mask.h"
40 #include "gca/gfx_7_0_d.h"
41 #include "gca/gfx_7_2_enum.h"
42 #include "gca/gfx_7_2_sh_mask.h"
44 #include "gmc/gmc_7_0_d.h"
45 #include "gmc/gmc_7_0_sh_mask.h"
47 #include "oss/oss_2_0_d.h"
48 #include "oss/oss_2_0_sh_mask.h"
50 #define GFX7_NUM_GFX_RINGS 1
51 #define GFX7_NUM_COMPUTE_RINGS 8
53 static void gfx_v7_0_set_ring_funcs(struct amdgpu_device *adev);
54 static void gfx_v7_0_set_irq_funcs(struct amdgpu_device *adev);
55 static void gfx_v7_0_set_gds_init(struct amdgpu_device *adev);
57 MODULE_FIRMWARE("radeon/bonaire_pfp.bin");
58 MODULE_FIRMWARE("radeon/bonaire_me.bin");
59 MODULE_FIRMWARE("radeon/bonaire_ce.bin");
60 MODULE_FIRMWARE("radeon/bonaire_rlc.bin");
61 MODULE_FIRMWARE("radeon/bonaire_mec.bin");
63 MODULE_FIRMWARE("radeon/hawaii_pfp.bin");
64 MODULE_FIRMWARE("radeon/hawaii_me.bin");
65 MODULE_FIRMWARE("radeon/hawaii_ce.bin");
66 MODULE_FIRMWARE("radeon/hawaii_rlc.bin");
67 MODULE_FIRMWARE("radeon/hawaii_mec.bin");
69 MODULE_FIRMWARE("radeon/kaveri_pfp.bin");
70 MODULE_FIRMWARE("radeon/kaveri_me.bin");
71 MODULE_FIRMWARE("radeon/kaveri_ce.bin");
72 MODULE_FIRMWARE("radeon/kaveri_rlc.bin");
73 MODULE_FIRMWARE("radeon/kaveri_mec.bin");
74 MODULE_FIRMWARE("radeon/kaveri_mec2.bin");
76 MODULE_FIRMWARE("radeon/kabini_pfp.bin");
77 MODULE_FIRMWARE("radeon/kabini_me.bin");
78 MODULE_FIRMWARE("radeon/kabini_ce.bin");
79 MODULE_FIRMWARE("radeon/kabini_rlc.bin");
80 MODULE_FIRMWARE("radeon/kabini_mec.bin");
82 MODULE_FIRMWARE("radeon/mullins_pfp.bin");
83 MODULE_FIRMWARE("radeon/mullins_me.bin");
84 MODULE_FIRMWARE("radeon/mullins_ce.bin");
85 MODULE_FIRMWARE("radeon/mullins_rlc.bin");
86 MODULE_FIRMWARE("radeon/mullins_mec.bin");
88 static const struct amdgpu_gds_reg_offset amdgpu_gds_reg_offset[] =
90 {mmGDS_VMID0_BASE, mmGDS_VMID0_SIZE, mmGDS_GWS_VMID0, mmGDS_OA_VMID0},
91 {mmGDS_VMID1_BASE, mmGDS_VMID1_SIZE, mmGDS_GWS_VMID1, mmGDS_OA_VMID1},
92 {mmGDS_VMID2_BASE, mmGDS_VMID2_SIZE, mmGDS_GWS_VMID2, mmGDS_OA_VMID2},
93 {mmGDS_VMID3_BASE, mmGDS_VMID3_SIZE, mmGDS_GWS_VMID3, mmGDS_OA_VMID3},
94 {mmGDS_VMID4_BASE, mmGDS_VMID4_SIZE, mmGDS_GWS_VMID4, mmGDS_OA_VMID4},
95 {mmGDS_VMID5_BASE, mmGDS_VMID5_SIZE, mmGDS_GWS_VMID5, mmGDS_OA_VMID5},
96 {mmGDS_VMID6_BASE, mmGDS_VMID6_SIZE, mmGDS_GWS_VMID6, mmGDS_OA_VMID6},
97 {mmGDS_VMID7_BASE, mmGDS_VMID7_SIZE, mmGDS_GWS_VMID7, mmGDS_OA_VMID7},
98 {mmGDS_VMID8_BASE, mmGDS_VMID8_SIZE, mmGDS_GWS_VMID8, mmGDS_OA_VMID8},
99 {mmGDS_VMID9_BASE, mmGDS_VMID9_SIZE, mmGDS_GWS_VMID9, mmGDS_OA_VMID9},
100 {mmGDS_VMID10_BASE, mmGDS_VMID10_SIZE, mmGDS_GWS_VMID10, mmGDS_OA_VMID10},
101 {mmGDS_VMID11_BASE, mmGDS_VMID11_SIZE, mmGDS_GWS_VMID11, mmGDS_OA_VMID11},
102 {mmGDS_VMID12_BASE, mmGDS_VMID12_SIZE, mmGDS_GWS_VMID12, mmGDS_OA_VMID12},
103 {mmGDS_VMID13_BASE, mmGDS_VMID13_SIZE, mmGDS_GWS_VMID13, mmGDS_OA_VMID13},
104 {mmGDS_VMID14_BASE, mmGDS_VMID14_SIZE, mmGDS_GWS_VMID14, mmGDS_OA_VMID14},
105 {mmGDS_VMID15_BASE, mmGDS_VMID15_SIZE, mmGDS_GWS_VMID15, mmGDS_OA_VMID15}
108 static const u32 spectre_rlc_save_restore_register_list[] =
110 (0x0e00 << 16) | (0xc12c >> 2),
112 (0x0e00 << 16) | (0xc140 >> 2),
114 (0x0e00 << 16) | (0xc150 >> 2),
116 (0x0e00 << 16) | (0xc15c >> 2),
118 (0x0e00 << 16) | (0xc168 >> 2),
120 (0x0e00 << 16) | (0xc170 >> 2),
122 (0x0e00 << 16) | (0xc178 >> 2),
124 (0x0e00 << 16) | (0xc204 >> 2),
126 (0x0e00 << 16) | (0xc2b4 >> 2),
128 (0x0e00 << 16) | (0xc2b8 >> 2),
130 (0x0e00 << 16) | (0xc2bc >> 2),
132 (0x0e00 << 16) | (0xc2c0 >> 2),
134 (0x0e00 << 16) | (0x8228 >> 2),
136 (0x0e00 << 16) | (0x829c >> 2),
138 (0x0e00 << 16) | (0x869c >> 2),
140 (0x0600 << 16) | (0x98f4 >> 2),
142 (0x0e00 << 16) | (0x98f8 >> 2),
144 (0x0e00 << 16) | (0x9900 >> 2),
146 (0x0e00 << 16) | (0xc260 >> 2),
148 (0x0e00 << 16) | (0x90e8 >> 2),
150 (0x0e00 << 16) | (0x3c000 >> 2),
152 (0x0e00 << 16) | (0x3c00c >> 2),
154 (0x0e00 << 16) | (0x8c1c >> 2),
156 (0x0e00 << 16) | (0x9700 >> 2),
158 (0x0e00 << 16) | (0xcd20 >> 2),
160 (0x4e00 << 16) | (0xcd20 >> 2),
162 (0x5e00 << 16) | (0xcd20 >> 2),
164 (0x6e00 << 16) | (0xcd20 >> 2),
166 (0x7e00 << 16) | (0xcd20 >> 2),
168 (0x8e00 << 16) | (0xcd20 >> 2),
170 (0x9e00 << 16) | (0xcd20 >> 2),
172 (0xae00 << 16) | (0xcd20 >> 2),
174 (0xbe00 << 16) | (0xcd20 >> 2),
176 (0x0e00 << 16) | (0x89bc >> 2),
178 (0x0e00 << 16) | (0x8900 >> 2),
181 (0x0e00 << 16) | (0xc130 >> 2),
183 (0x0e00 << 16) | (0xc134 >> 2),
185 (0x0e00 << 16) | (0xc1fc >> 2),
187 (0x0e00 << 16) | (0xc208 >> 2),
189 (0x0e00 << 16) | (0xc264 >> 2),
191 (0x0e00 << 16) | (0xc268 >> 2),
193 (0x0e00 << 16) | (0xc26c >> 2),
195 (0x0e00 << 16) | (0xc270 >> 2),
197 (0x0e00 << 16) | (0xc274 >> 2),
199 (0x0e00 << 16) | (0xc278 >> 2),
201 (0x0e00 << 16) | (0xc27c >> 2),
203 (0x0e00 << 16) | (0xc280 >> 2),
205 (0x0e00 << 16) | (0xc284 >> 2),
207 (0x0e00 << 16) | (0xc288 >> 2),
209 (0x0e00 << 16) | (0xc28c >> 2),
211 (0x0e00 << 16) | (0xc290 >> 2),
213 (0x0e00 << 16) | (0xc294 >> 2),
215 (0x0e00 << 16) | (0xc298 >> 2),
217 (0x0e00 << 16) | (0xc29c >> 2),
219 (0x0e00 << 16) | (0xc2a0 >> 2),
221 (0x0e00 << 16) | (0xc2a4 >> 2),
223 (0x0e00 << 16) | (0xc2a8 >> 2),
225 (0x0e00 << 16) | (0xc2ac >> 2),
227 (0x0e00 << 16) | (0xc2b0 >> 2),
229 (0x0e00 << 16) | (0x301d0 >> 2),
231 (0x0e00 << 16) | (0x30238 >> 2),
233 (0x0e00 << 16) | (0x30250 >> 2),
235 (0x0e00 << 16) | (0x30254 >> 2),
237 (0x0e00 << 16) | (0x30258 >> 2),
239 (0x0e00 << 16) | (0x3025c >> 2),
241 (0x4e00 << 16) | (0xc900 >> 2),
243 (0x5e00 << 16) | (0xc900 >> 2),
245 (0x6e00 << 16) | (0xc900 >> 2),
247 (0x7e00 << 16) | (0xc900 >> 2),
249 (0x8e00 << 16) | (0xc900 >> 2),
251 (0x9e00 << 16) | (0xc900 >> 2),
253 (0xae00 << 16) | (0xc900 >> 2),
255 (0xbe00 << 16) | (0xc900 >> 2),
257 (0x4e00 << 16) | (0xc904 >> 2),
259 (0x5e00 << 16) | (0xc904 >> 2),
261 (0x6e00 << 16) | (0xc904 >> 2),
263 (0x7e00 << 16) | (0xc904 >> 2),
265 (0x8e00 << 16) | (0xc904 >> 2),
267 (0x9e00 << 16) | (0xc904 >> 2),
269 (0xae00 << 16) | (0xc904 >> 2),
271 (0xbe00 << 16) | (0xc904 >> 2),
273 (0x4e00 << 16) | (0xc908 >> 2),
275 (0x5e00 << 16) | (0xc908 >> 2),
277 (0x6e00 << 16) | (0xc908 >> 2),
279 (0x7e00 << 16) | (0xc908 >> 2),
281 (0x8e00 << 16) | (0xc908 >> 2),
283 (0x9e00 << 16) | (0xc908 >> 2),
285 (0xae00 << 16) | (0xc908 >> 2),
287 (0xbe00 << 16) | (0xc908 >> 2),
289 (0x4e00 << 16) | (0xc90c >> 2),
291 (0x5e00 << 16) | (0xc90c >> 2),
293 (0x6e00 << 16) | (0xc90c >> 2),
295 (0x7e00 << 16) | (0xc90c >> 2),
297 (0x8e00 << 16) | (0xc90c >> 2),
299 (0x9e00 << 16) | (0xc90c >> 2),
301 (0xae00 << 16) | (0xc90c >> 2),
303 (0xbe00 << 16) | (0xc90c >> 2),
305 (0x4e00 << 16) | (0xc910 >> 2),
307 (0x5e00 << 16) | (0xc910 >> 2),
309 (0x6e00 << 16) | (0xc910 >> 2),
311 (0x7e00 << 16) | (0xc910 >> 2),
313 (0x8e00 << 16) | (0xc910 >> 2),
315 (0x9e00 << 16) | (0xc910 >> 2),
317 (0xae00 << 16) | (0xc910 >> 2),
319 (0xbe00 << 16) | (0xc910 >> 2),
321 (0x0e00 << 16) | (0xc99c >> 2),
323 (0x0e00 << 16) | (0x9834 >> 2),
325 (0x0000 << 16) | (0x30f00 >> 2),
327 (0x0001 << 16) | (0x30f00 >> 2),
329 (0x0000 << 16) | (0x30f04 >> 2),
331 (0x0001 << 16) | (0x30f04 >> 2),
333 (0x0000 << 16) | (0x30f08 >> 2),
335 (0x0001 << 16) | (0x30f08 >> 2),
337 (0x0000 << 16) | (0x30f0c >> 2),
339 (0x0001 << 16) | (0x30f0c >> 2),
341 (0x0600 << 16) | (0x9b7c >> 2),
343 (0x0e00 << 16) | (0x8a14 >> 2),
345 (0x0e00 << 16) | (0x8a18 >> 2),
347 (0x0600 << 16) | (0x30a00 >> 2),
349 (0x0e00 << 16) | (0x8bf0 >> 2),
351 (0x0e00 << 16) | (0x8bcc >> 2),
353 (0x0e00 << 16) | (0x8b24 >> 2),
355 (0x0e00 << 16) | (0x30a04 >> 2),
357 (0x0600 << 16) | (0x30a10 >> 2),
359 (0x0600 << 16) | (0x30a14 >> 2),
361 (0x0600 << 16) | (0x30a18 >> 2),
363 (0x0600 << 16) | (0x30a2c >> 2),
365 (0x0e00 << 16) | (0xc700 >> 2),
367 (0x0e00 << 16) | (0xc704 >> 2),
369 (0x0e00 << 16) | (0xc708 >> 2),
371 (0x0e00 << 16) | (0xc768 >> 2),
373 (0x0400 << 16) | (0xc770 >> 2),
375 (0x0400 << 16) | (0xc774 >> 2),
377 (0x0400 << 16) | (0xc778 >> 2),
379 (0x0400 << 16) | (0xc77c >> 2),
381 (0x0400 << 16) | (0xc780 >> 2),
383 (0x0400 << 16) | (0xc784 >> 2),
385 (0x0400 << 16) | (0xc788 >> 2),
387 (0x0400 << 16) | (0xc78c >> 2),
389 (0x0400 << 16) | (0xc798 >> 2),
391 (0x0400 << 16) | (0xc79c >> 2),
393 (0x0400 << 16) | (0xc7a0 >> 2),
395 (0x0400 << 16) | (0xc7a4 >> 2),
397 (0x0400 << 16) | (0xc7a8 >> 2),
399 (0x0400 << 16) | (0xc7ac >> 2),
401 (0x0400 << 16) | (0xc7b0 >> 2),
403 (0x0400 << 16) | (0xc7b4 >> 2),
405 (0x0e00 << 16) | (0x9100 >> 2),
407 (0x0e00 << 16) | (0x3c010 >> 2),
409 (0x0e00 << 16) | (0x92a8 >> 2),
411 (0x0e00 << 16) | (0x92ac >> 2),
413 (0x0e00 << 16) | (0x92b4 >> 2),
415 (0x0e00 << 16) | (0x92b8 >> 2),
417 (0x0e00 << 16) | (0x92bc >> 2),
419 (0x0e00 << 16) | (0x92c0 >> 2),
421 (0x0e00 << 16) | (0x92c4 >> 2),
423 (0x0e00 << 16) | (0x92c8 >> 2),
425 (0x0e00 << 16) | (0x92cc >> 2),
427 (0x0e00 << 16) | (0x92d0 >> 2),
429 (0x0e00 << 16) | (0x8c00 >> 2),
431 (0x0e00 << 16) | (0x8c04 >> 2),
433 (0x0e00 << 16) | (0x8c20 >> 2),
435 (0x0e00 << 16) | (0x8c38 >> 2),
437 (0x0e00 << 16) | (0x8c3c >> 2),
439 (0x0e00 << 16) | (0xae00 >> 2),
441 (0x0e00 << 16) | (0x9604 >> 2),
443 (0x0e00 << 16) | (0xac08 >> 2),
445 (0x0e00 << 16) | (0xac0c >> 2),
447 (0x0e00 << 16) | (0xac10 >> 2),
449 (0x0e00 << 16) | (0xac14 >> 2),
451 (0x0e00 << 16) | (0xac58 >> 2),
453 (0x0e00 << 16) | (0xac68 >> 2),
455 (0x0e00 << 16) | (0xac6c >> 2),
457 (0x0e00 << 16) | (0xac70 >> 2),
459 (0x0e00 << 16) | (0xac74 >> 2),
461 (0x0e00 << 16) | (0xac78 >> 2),
463 (0x0e00 << 16) | (0xac7c >> 2),
465 (0x0e00 << 16) | (0xac80 >> 2),
467 (0x0e00 << 16) | (0xac84 >> 2),
469 (0x0e00 << 16) | (0xac88 >> 2),
471 (0x0e00 << 16) | (0xac8c >> 2),
473 (0x0e00 << 16) | (0x970c >> 2),
475 (0x0e00 << 16) | (0x9714 >> 2),
477 (0x0e00 << 16) | (0x9718 >> 2),
479 (0x0e00 << 16) | (0x971c >> 2),
481 (0x0e00 << 16) | (0x31068 >> 2),
483 (0x4e00 << 16) | (0x31068 >> 2),
485 (0x5e00 << 16) | (0x31068 >> 2),
487 (0x6e00 << 16) | (0x31068 >> 2),
489 (0x7e00 << 16) | (0x31068 >> 2),
491 (0x8e00 << 16) | (0x31068 >> 2),
493 (0x9e00 << 16) | (0x31068 >> 2),
495 (0xae00 << 16) | (0x31068 >> 2),
497 (0xbe00 << 16) | (0x31068 >> 2),
499 (0x0e00 << 16) | (0xcd10 >> 2),
501 (0x0e00 << 16) | (0xcd14 >> 2),
503 (0x0e00 << 16) | (0x88b0 >> 2),
505 (0x0e00 << 16) | (0x88b4 >> 2),
507 (0x0e00 << 16) | (0x88b8 >> 2),
509 (0x0e00 << 16) | (0x88bc >> 2),
511 (0x0400 << 16) | (0x89c0 >> 2),
513 (0x0e00 << 16) | (0x88c4 >> 2),
515 (0x0e00 << 16) | (0x88c8 >> 2),
517 (0x0e00 << 16) | (0x88d0 >> 2),
519 (0x0e00 << 16) | (0x88d4 >> 2),
521 (0x0e00 << 16) | (0x88d8 >> 2),
523 (0x0e00 << 16) | (0x8980 >> 2),
525 (0x0e00 << 16) | (0x30938 >> 2),
527 (0x0e00 << 16) | (0x3093c >> 2),
529 (0x0e00 << 16) | (0x30940 >> 2),
531 (0x0e00 << 16) | (0x89a0 >> 2),
533 (0x0e00 << 16) | (0x30900 >> 2),
535 (0x0e00 << 16) | (0x30904 >> 2),
537 (0x0e00 << 16) | (0x89b4 >> 2),
539 (0x0e00 << 16) | (0x3c210 >> 2),
541 (0x0e00 << 16) | (0x3c214 >> 2),
543 (0x0e00 << 16) | (0x3c218 >> 2),
545 (0x0e00 << 16) | (0x8904 >> 2),
548 (0x0e00 << 16) | (0x8c28 >> 2),
549 (0x0e00 << 16) | (0x8c2c >> 2),
550 (0x0e00 << 16) | (0x8c30 >> 2),
551 (0x0e00 << 16) | (0x8c34 >> 2),
552 (0x0e00 << 16) | (0x9600 >> 2),
555 static const u32 kalindi_rlc_save_restore_register_list[] =
557 (0x0e00 << 16) | (0xc12c >> 2),
559 (0x0e00 << 16) | (0xc140 >> 2),
561 (0x0e00 << 16) | (0xc150 >> 2),
563 (0x0e00 << 16) | (0xc15c >> 2),
565 (0x0e00 << 16) | (0xc168 >> 2),
567 (0x0e00 << 16) | (0xc170 >> 2),
569 (0x0e00 << 16) | (0xc204 >> 2),
571 (0x0e00 << 16) | (0xc2b4 >> 2),
573 (0x0e00 << 16) | (0xc2b8 >> 2),
575 (0x0e00 << 16) | (0xc2bc >> 2),
577 (0x0e00 << 16) | (0xc2c0 >> 2),
579 (0x0e00 << 16) | (0x8228 >> 2),
581 (0x0e00 << 16) | (0x829c >> 2),
583 (0x0e00 << 16) | (0x869c >> 2),
585 (0x0600 << 16) | (0x98f4 >> 2),
587 (0x0e00 << 16) | (0x98f8 >> 2),
589 (0x0e00 << 16) | (0x9900 >> 2),
591 (0x0e00 << 16) | (0xc260 >> 2),
593 (0x0e00 << 16) | (0x90e8 >> 2),
595 (0x0e00 << 16) | (0x3c000 >> 2),
597 (0x0e00 << 16) | (0x3c00c >> 2),
599 (0x0e00 << 16) | (0x8c1c >> 2),
601 (0x0e00 << 16) | (0x9700 >> 2),
603 (0x0e00 << 16) | (0xcd20 >> 2),
605 (0x4e00 << 16) | (0xcd20 >> 2),
607 (0x5e00 << 16) | (0xcd20 >> 2),
609 (0x6e00 << 16) | (0xcd20 >> 2),
611 (0x7e00 << 16) | (0xcd20 >> 2),
613 (0x0e00 << 16) | (0x89bc >> 2),
615 (0x0e00 << 16) | (0x8900 >> 2),
618 (0x0e00 << 16) | (0xc130 >> 2),
620 (0x0e00 << 16) | (0xc134 >> 2),
622 (0x0e00 << 16) | (0xc1fc >> 2),
624 (0x0e00 << 16) | (0xc208 >> 2),
626 (0x0e00 << 16) | (0xc264 >> 2),
628 (0x0e00 << 16) | (0xc268 >> 2),
630 (0x0e00 << 16) | (0xc26c >> 2),
632 (0x0e00 << 16) | (0xc270 >> 2),
634 (0x0e00 << 16) | (0xc274 >> 2),
636 (0x0e00 << 16) | (0xc28c >> 2),
638 (0x0e00 << 16) | (0xc290 >> 2),
640 (0x0e00 << 16) | (0xc294 >> 2),
642 (0x0e00 << 16) | (0xc298 >> 2),
644 (0x0e00 << 16) | (0xc2a0 >> 2),
646 (0x0e00 << 16) | (0xc2a4 >> 2),
648 (0x0e00 << 16) | (0xc2a8 >> 2),
650 (0x0e00 << 16) | (0xc2ac >> 2),
652 (0x0e00 << 16) | (0x301d0 >> 2),
654 (0x0e00 << 16) | (0x30238 >> 2),
656 (0x0e00 << 16) | (0x30250 >> 2),
658 (0x0e00 << 16) | (0x30254 >> 2),
660 (0x0e00 << 16) | (0x30258 >> 2),
662 (0x0e00 << 16) | (0x3025c >> 2),
664 (0x4e00 << 16) | (0xc900 >> 2),
666 (0x5e00 << 16) | (0xc900 >> 2),
668 (0x6e00 << 16) | (0xc900 >> 2),
670 (0x7e00 << 16) | (0xc900 >> 2),
672 (0x4e00 << 16) | (0xc904 >> 2),
674 (0x5e00 << 16) | (0xc904 >> 2),
676 (0x6e00 << 16) | (0xc904 >> 2),
678 (0x7e00 << 16) | (0xc904 >> 2),
680 (0x4e00 << 16) | (0xc908 >> 2),
682 (0x5e00 << 16) | (0xc908 >> 2),
684 (0x6e00 << 16) | (0xc908 >> 2),
686 (0x7e00 << 16) | (0xc908 >> 2),
688 (0x4e00 << 16) | (0xc90c >> 2),
690 (0x5e00 << 16) | (0xc90c >> 2),
692 (0x6e00 << 16) | (0xc90c >> 2),
694 (0x7e00 << 16) | (0xc90c >> 2),
696 (0x4e00 << 16) | (0xc910 >> 2),
698 (0x5e00 << 16) | (0xc910 >> 2),
700 (0x6e00 << 16) | (0xc910 >> 2),
702 (0x7e00 << 16) | (0xc910 >> 2),
704 (0x0e00 << 16) | (0xc99c >> 2),
706 (0x0e00 << 16) | (0x9834 >> 2),
708 (0x0000 << 16) | (0x30f00 >> 2),
710 (0x0000 << 16) | (0x30f04 >> 2),
712 (0x0000 << 16) | (0x30f08 >> 2),
714 (0x0000 << 16) | (0x30f0c >> 2),
716 (0x0600 << 16) | (0x9b7c >> 2),
718 (0x0e00 << 16) | (0x8a14 >> 2),
720 (0x0e00 << 16) | (0x8a18 >> 2),
722 (0x0600 << 16) | (0x30a00 >> 2),
724 (0x0e00 << 16) | (0x8bf0 >> 2),
726 (0x0e00 << 16) | (0x8bcc >> 2),
728 (0x0e00 << 16) | (0x8b24 >> 2),
730 (0x0e00 << 16) | (0x30a04 >> 2),
732 (0x0600 << 16) | (0x30a10 >> 2),
734 (0x0600 << 16) | (0x30a14 >> 2),
736 (0x0600 << 16) | (0x30a18 >> 2),
738 (0x0600 << 16) | (0x30a2c >> 2),
740 (0x0e00 << 16) | (0xc700 >> 2),
742 (0x0e00 << 16) | (0xc704 >> 2),
744 (0x0e00 << 16) | (0xc708 >> 2),
746 (0x0e00 << 16) | (0xc768 >> 2),
748 (0x0400 << 16) | (0xc770 >> 2),
750 (0x0400 << 16) | (0xc774 >> 2),
752 (0x0400 << 16) | (0xc798 >> 2),
754 (0x0400 << 16) | (0xc79c >> 2),
756 (0x0e00 << 16) | (0x9100 >> 2),
758 (0x0e00 << 16) | (0x3c010 >> 2),
760 (0x0e00 << 16) | (0x8c00 >> 2),
762 (0x0e00 << 16) | (0x8c04 >> 2),
764 (0x0e00 << 16) | (0x8c20 >> 2),
766 (0x0e00 << 16) | (0x8c38 >> 2),
768 (0x0e00 << 16) | (0x8c3c >> 2),
770 (0x0e00 << 16) | (0xae00 >> 2),
772 (0x0e00 << 16) | (0x9604 >> 2),
774 (0x0e00 << 16) | (0xac08 >> 2),
776 (0x0e00 << 16) | (0xac0c >> 2),
778 (0x0e00 << 16) | (0xac10 >> 2),
780 (0x0e00 << 16) | (0xac14 >> 2),
782 (0x0e00 << 16) | (0xac58 >> 2),
784 (0x0e00 << 16) | (0xac68 >> 2),
786 (0x0e00 << 16) | (0xac6c >> 2),
788 (0x0e00 << 16) | (0xac70 >> 2),
790 (0x0e00 << 16) | (0xac74 >> 2),
792 (0x0e00 << 16) | (0xac78 >> 2),
794 (0x0e00 << 16) | (0xac7c >> 2),
796 (0x0e00 << 16) | (0xac80 >> 2),
798 (0x0e00 << 16) | (0xac84 >> 2),
800 (0x0e00 << 16) | (0xac88 >> 2),
802 (0x0e00 << 16) | (0xac8c >> 2),
804 (0x0e00 << 16) | (0x970c >> 2),
806 (0x0e00 << 16) | (0x9714 >> 2),
808 (0x0e00 << 16) | (0x9718 >> 2),
810 (0x0e00 << 16) | (0x971c >> 2),
812 (0x0e00 << 16) | (0x31068 >> 2),
814 (0x4e00 << 16) | (0x31068 >> 2),
816 (0x5e00 << 16) | (0x31068 >> 2),
818 (0x6e00 << 16) | (0x31068 >> 2),
820 (0x7e00 << 16) | (0x31068 >> 2),
822 (0x0e00 << 16) | (0xcd10 >> 2),
824 (0x0e00 << 16) | (0xcd14 >> 2),
826 (0x0e00 << 16) | (0x88b0 >> 2),
828 (0x0e00 << 16) | (0x88b4 >> 2),
830 (0x0e00 << 16) | (0x88b8 >> 2),
832 (0x0e00 << 16) | (0x88bc >> 2),
834 (0x0400 << 16) | (0x89c0 >> 2),
836 (0x0e00 << 16) | (0x88c4 >> 2),
838 (0x0e00 << 16) | (0x88c8 >> 2),
840 (0x0e00 << 16) | (0x88d0 >> 2),
842 (0x0e00 << 16) | (0x88d4 >> 2),
844 (0x0e00 << 16) | (0x88d8 >> 2),
846 (0x0e00 << 16) | (0x8980 >> 2),
848 (0x0e00 << 16) | (0x30938 >> 2),
850 (0x0e00 << 16) | (0x3093c >> 2),
852 (0x0e00 << 16) | (0x30940 >> 2),
854 (0x0e00 << 16) | (0x89a0 >> 2),
856 (0x0e00 << 16) | (0x30900 >> 2),
858 (0x0e00 << 16) | (0x30904 >> 2),
860 (0x0e00 << 16) | (0x89b4 >> 2),
862 (0x0e00 << 16) | (0x3e1fc >> 2),
864 (0x0e00 << 16) | (0x3c210 >> 2),
866 (0x0e00 << 16) | (0x3c214 >> 2),
868 (0x0e00 << 16) | (0x3c218 >> 2),
870 (0x0e00 << 16) | (0x8904 >> 2),
873 (0x0e00 << 16) | (0x8c28 >> 2),
874 (0x0e00 << 16) | (0x8c2c >> 2),
875 (0x0e00 << 16) | (0x8c30 >> 2),
876 (0x0e00 << 16) | (0x8c34 >> 2),
877 (0x0e00 << 16) | (0x9600 >> 2),
880 static u32 gfx_v7_0_get_csb_size(struct amdgpu_device *adev);
881 static void gfx_v7_0_get_csb_buffer(struct amdgpu_device *adev, volatile u32 *buffer);
882 static void gfx_v7_0_init_cp_pg_table(struct amdgpu_device *adev);
883 static void gfx_v7_0_init_pg(struct amdgpu_device *adev);
884 static void gfx_v7_0_get_cu_info(struct amdgpu_device *adev);
890 * gfx_v7_0_init_microcode - load ucode images from disk
892 * @adev: amdgpu_device pointer
894 * Use the firmware interface to load the ucode images into
895 * the driver (not loaded into hw).
896 * Returns 0 on success, error on failure.
898 static int gfx_v7_0_init_microcode(struct amdgpu_device *adev)
900 const char *chip_name;
906 switch (adev->asic_type) {
908 chip_name = "bonaire";
911 chip_name = "hawaii";
914 chip_name = "kaveri";
917 chip_name = "kabini";
920 chip_name = "mullins";
925 snprintf(fw_name, sizeof(fw_name), "radeon/%s_pfp.bin", chip_name);
926 err = request_firmware(&adev->gfx.pfp_fw, fw_name, adev->dev);
929 err = amdgpu_ucode_validate(adev->gfx.pfp_fw);
933 snprintf(fw_name, sizeof(fw_name), "radeon/%s_me.bin", chip_name);
934 err = request_firmware(&adev->gfx.me_fw, fw_name, adev->dev);
937 err = amdgpu_ucode_validate(adev->gfx.me_fw);
941 snprintf(fw_name, sizeof(fw_name), "radeon/%s_ce.bin", chip_name);
942 err = request_firmware(&adev->gfx.ce_fw, fw_name, adev->dev);
945 err = amdgpu_ucode_validate(adev->gfx.ce_fw);
949 snprintf(fw_name, sizeof(fw_name), "radeon/%s_mec.bin", chip_name);
950 err = request_firmware(&adev->gfx.mec_fw, fw_name, adev->dev);
953 err = amdgpu_ucode_validate(adev->gfx.mec_fw);
957 if (adev->asic_type == CHIP_KAVERI) {
958 snprintf(fw_name, sizeof(fw_name), "radeon/%s_mec2.bin", chip_name);
959 err = request_firmware(&adev->gfx.mec2_fw, fw_name, adev->dev);
962 err = amdgpu_ucode_validate(adev->gfx.mec2_fw);
967 snprintf(fw_name, sizeof(fw_name), "radeon/%s_rlc.bin", chip_name);
968 err = request_firmware(&adev->gfx.rlc_fw, fw_name, adev->dev);
971 err = amdgpu_ucode_validate(adev->gfx.rlc_fw);
976 "gfx7: Failed to load firmware \"%s\"\n",
978 release_firmware(adev->gfx.pfp_fw);
979 adev->gfx.pfp_fw = NULL;
980 release_firmware(adev->gfx.me_fw);
981 adev->gfx.me_fw = NULL;
982 release_firmware(adev->gfx.ce_fw);
983 adev->gfx.ce_fw = NULL;
984 release_firmware(adev->gfx.mec_fw);
985 adev->gfx.mec_fw = NULL;
986 release_firmware(adev->gfx.mec2_fw);
987 adev->gfx.mec2_fw = NULL;
988 release_firmware(adev->gfx.rlc_fw);
989 adev->gfx.rlc_fw = NULL;
994 static void gfx_v7_0_free_microcode(struct amdgpu_device *adev)
996 release_firmware(adev->gfx.pfp_fw);
997 adev->gfx.pfp_fw = NULL;
998 release_firmware(adev->gfx.me_fw);
999 adev->gfx.me_fw = NULL;
1000 release_firmware(adev->gfx.ce_fw);
1001 adev->gfx.ce_fw = NULL;
1002 release_firmware(adev->gfx.mec_fw);
1003 adev->gfx.mec_fw = NULL;
1004 release_firmware(adev->gfx.mec2_fw);
1005 adev->gfx.mec2_fw = NULL;
1006 release_firmware(adev->gfx.rlc_fw);
1007 adev->gfx.rlc_fw = NULL;
1011 * gfx_v7_0_tiling_mode_table_init - init the hw tiling table
1013 * @adev: amdgpu_device pointer
1015 * Starting with SI, the tiling setup is done globally in a
1016 * set of 32 tiling modes. Rather than selecting each set of
1017 * parameters per surface as on older asics, we just select
1018 * which index in the tiling table we want to use, and the
1019 * surface uses those parameters (CIK).
1021 static void gfx_v7_0_tiling_mode_table_init(struct amdgpu_device *adev)
1023 const u32 num_tile_mode_states =
1024 ARRAY_SIZE(adev->gfx.config.tile_mode_array);
1025 const u32 num_secondary_tile_mode_states =
1026 ARRAY_SIZE(adev->gfx.config.macrotile_mode_array);
1027 u32 reg_offset, split_equal_to_row_size;
1028 uint32_t *tile, *macrotile;
1030 tile = adev->gfx.config.tile_mode_array;
1031 macrotile = adev->gfx.config.macrotile_mode_array;
1033 switch (adev->gfx.config.mem_row_size_in_kb) {
1035 split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_1KB;
1039 split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_2KB;
1042 split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_4KB;
1046 for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
1047 tile[reg_offset] = 0;
1048 for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
1049 macrotile[reg_offset] = 0;
1051 switch (adev->asic_type) {
1053 tile[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1054 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1055 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
1056 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1057 tile[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1058 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1059 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
1060 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1061 tile[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1062 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1063 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1064 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1065 tile[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1066 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1067 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
1068 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1069 tile[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1070 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1071 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1072 TILE_SPLIT(split_equal_to_row_size));
1073 tile[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1074 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1075 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1076 tile[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1077 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1078 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1079 TILE_SPLIT(split_equal_to_row_size));
1080 tile[7] = (TILE_SPLIT(split_equal_to_row_size));
1081 tile[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
1082 PIPE_CONFIG(ADDR_SURF_P4_16x16));
1083 tile[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1084 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1085 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING));
1086 tile[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1087 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1088 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
1089 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1090 tile[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1091 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1092 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
1093 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1094 tile[12] = (TILE_SPLIT(split_equal_to_row_size));
1095 tile[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1096 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1097 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING));
1098 tile[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1099 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1100 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1101 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1102 tile[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
1103 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1104 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1105 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1106 tile[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1107 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1108 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1109 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1110 tile[17] = (TILE_SPLIT(split_equal_to_row_size));
1111 tile[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
1112 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1113 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1114 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1115 tile[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
1116 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1117 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING));
1118 tile[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
1119 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1120 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1121 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1122 tile[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
1123 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1124 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1125 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1126 tile[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
1127 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1128 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1129 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1130 tile[23] = (TILE_SPLIT(split_equal_to_row_size));
1131 tile[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
1132 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1133 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1134 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1135 tile[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
1136 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1137 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1138 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1139 tile[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
1140 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1141 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1142 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1143 tile[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1144 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1145 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING));
1146 tile[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1147 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1148 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
1149 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1150 tile[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1151 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1152 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
1153 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1154 tile[30] = (TILE_SPLIT(split_equal_to_row_size));
1156 macrotile[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1157 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1158 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1159 NUM_BANKS(ADDR_SURF_16_BANK));
1160 macrotile[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1161 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1162 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1163 NUM_BANKS(ADDR_SURF_16_BANK));
1164 macrotile[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1165 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1166 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1167 NUM_BANKS(ADDR_SURF_16_BANK));
1168 macrotile[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1169 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1170 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1171 NUM_BANKS(ADDR_SURF_16_BANK));
1172 macrotile[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1173 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1174 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1175 NUM_BANKS(ADDR_SURF_16_BANK));
1176 macrotile[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1177 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1178 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1179 NUM_BANKS(ADDR_SURF_8_BANK));
1180 macrotile[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1181 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1182 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1183 NUM_BANKS(ADDR_SURF_4_BANK));
1184 macrotile[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
1185 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
1186 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1187 NUM_BANKS(ADDR_SURF_16_BANK));
1188 macrotile[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
1189 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1190 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1191 NUM_BANKS(ADDR_SURF_16_BANK));
1192 macrotile[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1193 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1194 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1195 NUM_BANKS(ADDR_SURF_16_BANK));
1196 macrotile[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1197 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1198 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1199 NUM_BANKS(ADDR_SURF_16_BANK));
1200 macrotile[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1201 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1202 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1203 NUM_BANKS(ADDR_SURF_16_BANK));
1204 macrotile[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1205 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1206 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1207 NUM_BANKS(ADDR_SURF_8_BANK));
1208 macrotile[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1209 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1210 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1211 NUM_BANKS(ADDR_SURF_4_BANK));
1213 for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
1214 WREG32(mmGB_TILE_MODE0 + reg_offset, tile[reg_offset]);
1215 for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
1216 if (reg_offset != 7)
1217 WREG32(mmGB_MACROTILE_MODE0 + reg_offset, macrotile[reg_offset]);
1220 tile[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1221 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1222 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
1223 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1224 tile[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1225 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1226 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
1227 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1228 tile[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1229 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1230 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1231 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1232 tile[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1233 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1234 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
1235 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1236 tile[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1237 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1238 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1239 TILE_SPLIT(split_equal_to_row_size));
1240 tile[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1241 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1242 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1243 TILE_SPLIT(split_equal_to_row_size));
1244 tile[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1245 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1246 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1247 TILE_SPLIT(split_equal_to_row_size));
1248 tile[7] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1249 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1250 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1251 TILE_SPLIT(split_equal_to_row_size));
1252 tile[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
1253 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16));
1254 tile[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1255 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1256 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING));
1257 tile[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1258 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1259 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
1260 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1261 tile[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1262 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1263 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
1264 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1265 tile[12] = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
1266 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1267 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
1268 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1269 tile[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1270 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1271 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING));
1272 tile[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1273 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1274 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1275 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1276 tile[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
1277 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1278 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1279 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1280 tile[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1281 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1282 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1283 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1284 tile[17] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1285 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1286 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1287 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1288 tile[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
1289 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1290 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1291 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1292 tile[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
1293 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1294 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING));
1295 tile[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
1296 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1297 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1298 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1299 tile[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
1300 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1301 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1302 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1303 tile[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
1304 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1305 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1306 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1307 tile[23] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
1308 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1309 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1310 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1311 tile[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
1312 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1313 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1314 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1315 tile[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
1316 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1317 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1318 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1319 tile[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
1320 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1321 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1322 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1323 tile[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1324 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1325 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING));
1326 tile[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1327 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1328 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
1329 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1330 tile[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1331 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1332 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
1333 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1334 tile[30] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1335 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1336 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
1337 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1339 macrotile[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1340 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1341 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1342 NUM_BANKS(ADDR_SURF_16_BANK));
1343 macrotile[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1344 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1345 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1346 NUM_BANKS(ADDR_SURF_16_BANK));
1347 macrotile[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1348 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1349 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1350 NUM_BANKS(ADDR_SURF_16_BANK));
1351 macrotile[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1352 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1353 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1354 NUM_BANKS(ADDR_SURF_16_BANK));
1355 macrotile[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1356 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1357 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1358 NUM_BANKS(ADDR_SURF_8_BANK));
1359 macrotile[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1360 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1361 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1362 NUM_BANKS(ADDR_SURF_4_BANK));
1363 macrotile[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1364 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1365 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1366 NUM_BANKS(ADDR_SURF_4_BANK));
1367 macrotile[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1368 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1369 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1370 NUM_BANKS(ADDR_SURF_16_BANK));
1371 macrotile[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1372 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1373 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1374 NUM_BANKS(ADDR_SURF_16_BANK));
1375 macrotile[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1376 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1377 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1378 NUM_BANKS(ADDR_SURF_16_BANK));
1379 macrotile[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1380 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1381 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1382 NUM_BANKS(ADDR_SURF_8_BANK));
1383 macrotile[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1384 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1385 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1386 NUM_BANKS(ADDR_SURF_16_BANK));
1387 macrotile[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1388 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1389 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1390 NUM_BANKS(ADDR_SURF_8_BANK));
1391 macrotile[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1392 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1393 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1394 NUM_BANKS(ADDR_SURF_4_BANK));
1396 for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
1397 WREG32(mmGB_TILE_MODE0 + reg_offset, tile[reg_offset]);
1398 for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
1399 if (reg_offset != 7)
1400 WREG32(mmGB_MACROTILE_MODE0 + reg_offset, macrotile[reg_offset]);
1406 tile[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1407 PIPE_CONFIG(ADDR_SURF_P2) |
1408 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
1409 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1410 tile[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1411 PIPE_CONFIG(ADDR_SURF_P2) |
1412 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
1413 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1414 tile[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1415 PIPE_CONFIG(ADDR_SURF_P2) |
1416 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1417 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1418 tile[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1419 PIPE_CONFIG(ADDR_SURF_P2) |
1420 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
1421 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1422 tile[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1423 PIPE_CONFIG(ADDR_SURF_P2) |
1424 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1425 TILE_SPLIT(split_equal_to_row_size));
1426 tile[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1427 PIPE_CONFIG(ADDR_SURF_P2) |
1428 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1429 tile[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1430 PIPE_CONFIG(ADDR_SURF_P2) |
1431 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1432 TILE_SPLIT(split_equal_to_row_size));
1433 tile[7] = (TILE_SPLIT(split_equal_to_row_size));
1434 tile[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
1435 PIPE_CONFIG(ADDR_SURF_P2));
1436 tile[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1437 PIPE_CONFIG(ADDR_SURF_P2) |
1438 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING));
1439 tile[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1440 PIPE_CONFIG(ADDR_SURF_P2) |
1441 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
1442 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1443 tile[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1444 PIPE_CONFIG(ADDR_SURF_P2) |
1445 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
1446 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1447 tile[12] = (TILE_SPLIT(split_equal_to_row_size));
1448 tile[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1449 PIPE_CONFIG(ADDR_SURF_P2) |
1450 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING));
1451 tile[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1452 PIPE_CONFIG(ADDR_SURF_P2) |
1453 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1454 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1455 tile[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
1456 PIPE_CONFIG(ADDR_SURF_P2) |
1457 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1458 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1459 tile[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1460 PIPE_CONFIG(ADDR_SURF_P2) |
1461 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1462 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1463 tile[17] = (TILE_SPLIT(split_equal_to_row_size));
1464 tile[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
1465 PIPE_CONFIG(ADDR_SURF_P2) |
1466 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1467 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1468 tile[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
1469 PIPE_CONFIG(ADDR_SURF_P2) |
1470 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING));
1471 tile[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
1472 PIPE_CONFIG(ADDR_SURF_P2) |
1473 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1474 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1475 tile[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
1476 PIPE_CONFIG(ADDR_SURF_P2) |
1477 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1478 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1479 tile[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
1480 PIPE_CONFIG(ADDR_SURF_P2) |
1481 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1482 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1483 tile[23] = (TILE_SPLIT(split_equal_to_row_size));
1484 tile[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
1485 PIPE_CONFIG(ADDR_SURF_P2) |
1486 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1487 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1488 tile[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
1489 PIPE_CONFIG(ADDR_SURF_P2) |
1490 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1491 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1492 tile[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
1493 PIPE_CONFIG(ADDR_SURF_P2) |
1494 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1495 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1496 tile[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1497 PIPE_CONFIG(ADDR_SURF_P2) |
1498 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING));
1499 tile[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1500 PIPE_CONFIG(ADDR_SURF_P2) |
1501 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
1502 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1503 tile[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1504 PIPE_CONFIG(ADDR_SURF_P2) |
1505 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
1506 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1507 tile[30] = (TILE_SPLIT(split_equal_to_row_size));
1509 macrotile[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1510 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1511 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1512 NUM_BANKS(ADDR_SURF_8_BANK));
1513 macrotile[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1514 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1515 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1516 NUM_BANKS(ADDR_SURF_8_BANK));
1517 macrotile[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1518 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1519 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1520 NUM_BANKS(ADDR_SURF_8_BANK));
1521 macrotile[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1522 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1523 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1524 NUM_BANKS(ADDR_SURF_8_BANK));
1525 macrotile[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1526 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1527 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1528 NUM_BANKS(ADDR_SURF_8_BANK));
1529 macrotile[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1530 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1531 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1532 NUM_BANKS(ADDR_SURF_8_BANK));
1533 macrotile[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1534 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1535 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1536 NUM_BANKS(ADDR_SURF_8_BANK));
1537 macrotile[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
1538 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
1539 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1540 NUM_BANKS(ADDR_SURF_16_BANK));
1541 macrotile[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
1542 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1543 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1544 NUM_BANKS(ADDR_SURF_16_BANK));
1545 macrotile[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
1546 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1547 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1548 NUM_BANKS(ADDR_SURF_16_BANK));
1549 macrotile[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
1550 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1551 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1552 NUM_BANKS(ADDR_SURF_16_BANK));
1553 macrotile[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1554 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1555 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1556 NUM_BANKS(ADDR_SURF_16_BANK));
1557 macrotile[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1558 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1559 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1560 NUM_BANKS(ADDR_SURF_16_BANK));
1561 macrotile[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1562 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1563 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1564 NUM_BANKS(ADDR_SURF_8_BANK));
1566 for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
1567 WREG32(mmGB_TILE_MODE0 + reg_offset, tile[reg_offset]);
1568 for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
1569 if (reg_offset != 7)
1570 WREG32(mmGB_MACROTILE_MODE0 + reg_offset, macrotile[reg_offset]);
1576 * gfx_v7_0_select_se_sh - select which SE, SH to address
1578 * @adev: amdgpu_device pointer
1579 * @se_num: shader engine to address
1580 * @sh_num: sh block to address
1582 * Select which SE, SH combinations to address. Certain
1583 * registers are instanced per SE or SH. 0xffffffff means
1584 * broadcast to all SEs or SHs (CIK).
1586 static void gfx_v7_0_select_se_sh(struct amdgpu_device *adev,
1587 u32 se_num, u32 sh_num, u32 instance)
1591 if (instance == 0xffffffff)
1592 data = REG_SET_FIELD(0, GRBM_GFX_INDEX, INSTANCE_BROADCAST_WRITES, 1);
1594 data = REG_SET_FIELD(0, GRBM_GFX_INDEX, INSTANCE_INDEX, instance);
1596 if ((se_num == 0xffffffff) && (sh_num == 0xffffffff))
1597 data |= GRBM_GFX_INDEX__SH_BROADCAST_WRITES_MASK |
1598 GRBM_GFX_INDEX__SE_BROADCAST_WRITES_MASK;
1599 else if (se_num == 0xffffffff)
1600 data |= GRBM_GFX_INDEX__SE_BROADCAST_WRITES_MASK |
1601 (sh_num << GRBM_GFX_INDEX__SH_INDEX__SHIFT);
1602 else if (sh_num == 0xffffffff)
1603 data |= GRBM_GFX_INDEX__SH_BROADCAST_WRITES_MASK |
1604 (se_num << GRBM_GFX_INDEX__SE_INDEX__SHIFT);
1606 data |= (sh_num << GRBM_GFX_INDEX__SH_INDEX__SHIFT) |
1607 (se_num << GRBM_GFX_INDEX__SE_INDEX__SHIFT);
1608 WREG32(mmGRBM_GFX_INDEX, data);
1612 * gfx_v7_0_create_bitmask - create a bitmask
1614 * @bit_width: length of the mask
1616 * create a variable length bit mask (CIK).
1617 * Returns the bitmask.
1619 static u32 gfx_v7_0_create_bitmask(u32 bit_width)
1621 return (u32)((1ULL << bit_width) - 1);
1625 * gfx_v7_0_get_rb_active_bitmap - computes the mask of enabled RBs
1627 * @adev: amdgpu_device pointer
1629 * Calculates the bitmask of enabled RBs (CIK).
1630 * Returns the enabled RB bitmask.
1632 static u32 gfx_v7_0_get_rb_active_bitmap(struct amdgpu_device *adev)
1636 data = RREG32(mmCC_RB_BACKEND_DISABLE);
1637 data |= RREG32(mmGC_USER_RB_BACKEND_DISABLE);
1639 data &= CC_RB_BACKEND_DISABLE__BACKEND_DISABLE_MASK;
1640 data >>= GC_USER_RB_BACKEND_DISABLE__BACKEND_DISABLE__SHIFT;
1642 mask = gfx_v7_0_create_bitmask(adev->gfx.config.max_backends_per_se /
1643 adev->gfx.config.max_sh_per_se);
1645 return (~data) & mask;
1649 gfx_v7_0_raster_config(struct amdgpu_device *adev, u32 *rconf, u32 *rconf1)
1651 switch (adev->asic_type) {
1653 *rconf |= RB_MAP_PKR0(2) | RB_XSEL2(1) | SE_MAP(2) |
1654 SE_XSEL(1) | SE_YSEL(1);
1658 *rconf |= RB_MAP_PKR0(2) | RB_MAP_PKR1(2) |
1659 RB_XSEL2(1) | PKR_MAP(2) | PKR_XSEL(1) |
1660 PKR_YSEL(1) | SE_MAP(2) | SE_XSEL(2) |
1662 *rconf1 |= SE_PAIR_MAP(2) | SE_PAIR_XSEL(3) |
1666 *rconf |= RB_MAP_PKR0(2);
1675 DRM_ERROR("unknown asic: 0x%x\n", adev->asic_type);
1681 gfx_v7_0_write_harvested_raster_configs(struct amdgpu_device *adev,
1682 u32 raster_config, u32 raster_config_1,
1683 unsigned rb_mask, unsigned num_rb)
1685 unsigned sh_per_se = max_t(unsigned, adev->gfx.config.max_sh_per_se, 1);
1686 unsigned num_se = max_t(unsigned, adev->gfx.config.max_shader_engines, 1);
1687 unsigned rb_per_pkr = min_t(unsigned, num_rb / num_se / sh_per_se, 2);
1688 unsigned rb_per_se = num_rb / num_se;
1689 unsigned se_mask[4];
1692 se_mask[0] = ((1 << rb_per_se) - 1) & rb_mask;
1693 se_mask[1] = (se_mask[0] << rb_per_se) & rb_mask;
1694 se_mask[2] = (se_mask[1] << rb_per_se) & rb_mask;
1695 se_mask[3] = (se_mask[2] << rb_per_se) & rb_mask;
1697 WARN_ON(!(num_se == 1 || num_se == 2 || num_se == 4));
1698 WARN_ON(!(sh_per_se == 1 || sh_per_se == 2));
1699 WARN_ON(!(rb_per_pkr == 1 || rb_per_pkr == 2));
1701 if ((num_se > 2) && ((!se_mask[0] && !se_mask[1]) ||
1702 (!se_mask[2] && !se_mask[3]))) {
1703 raster_config_1 &= ~SE_PAIR_MAP_MASK;
1705 if (!se_mask[0] && !se_mask[1]) {
1707 SE_PAIR_MAP(RASTER_CONFIG_SE_PAIR_MAP_3);
1710 SE_PAIR_MAP(RASTER_CONFIG_SE_PAIR_MAP_0);
1714 for (se = 0; se < num_se; se++) {
1715 unsigned raster_config_se = raster_config;
1716 unsigned pkr0_mask = ((1 << rb_per_pkr) - 1) << (se * rb_per_se);
1717 unsigned pkr1_mask = pkr0_mask << rb_per_pkr;
1718 int idx = (se / 2) * 2;
1720 if ((num_se > 1) && (!se_mask[idx] || !se_mask[idx + 1])) {
1721 raster_config_se &= ~SE_MAP_MASK;
1723 if (!se_mask[idx]) {
1724 raster_config_se |= SE_MAP(RASTER_CONFIG_SE_MAP_3);
1726 raster_config_se |= SE_MAP(RASTER_CONFIG_SE_MAP_0);
1730 pkr0_mask &= rb_mask;
1731 pkr1_mask &= rb_mask;
1732 if (rb_per_se > 2 && (!pkr0_mask || !pkr1_mask)) {
1733 raster_config_se &= ~PKR_MAP_MASK;
1736 raster_config_se |= PKR_MAP(RASTER_CONFIG_PKR_MAP_3);
1738 raster_config_se |= PKR_MAP(RASTER_CONFIG_PKR_MAP_0);
1742 if (rb_per_se >= 2) {
1743 unsigned rb0_mask = 1 << (se * rb_per_se);
1744 unsigned rb1_mask = rb0_mask << 1;
1746 rb0_mask &= rb_mask;
1747 rb1_mask &= rb_mask;
1748 if (!rb0_mask || !rb1_mask) {
1749 raster_config_se &= ~RB_MAP_PKR0_MASK;
1753 RB_MAP_PKR0(RASTER_CONFIG_RB_MAP_3);
1756 RB_MAP_PKR0(RASTER_CONFIG_RB_MAP_0);
1760 if (rb_per_se > 2) {
1761 rb0_mask = 1 << (se * rb_per_se + rb_per_pkr);
1762 rb1_mask = rb0_mask << 1;
1763 rb0_mask &= rb_mask;
1764 rb1_mask &= rb_mask;
1765 if (!rb0_mask || !rb1_mask) {
1766 raster_config_se &= ~RB_MAP_PKR1_MASK;
1770 RB_MAP_PKR1(RASTER_CONFIG_RB_MAP_3);
1773 RB_MAP_PKR1(RASTER_CONFIG_RB_MAP_0);
1779 /* GRBM_GFX_INDEX has a different offset on CI+ */
1780 gfx_v7_0_select_se_sh(adev, se, 0xffffffff, 0xffffffff);
1781 WREG32(mmPA_SC_RASTER_CONFIG, raster_config_se);
1782 WREG32(mmPA_SC_RASTER_CONFIG_1, raster_config_1);
1785 /* GRBM_GFX_INDEX has a different offset on CI+ */
1786 gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
1790 * gfx_v7_0_setup_rb - setup the RBs on the asic
1792 * @adev: amdgpu_device pointer
1793 * @se_num: number of SEs (shader engines) for the asic
1794 * @sh_per_se: number of SH blocks per SE for the asic
1796 * Configures per-SE/SH RB registers (CIK).
1798 static void gfx_v7_0_setup_rb(struct amdgpu_device *adev)
1802 u32 raster_config = 0, raster_config_1 = 0;
1804 u32 rb_bitmap_width_per_sh = adev->gfx.config.max_backends_per_se /
1805 adev->gfx.config.max_sh_per_se;
1806 unsigned num_rb_pipes;
1808 mutex_lock(&adev->grbm_idx_mutex);
1809 for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
1810 for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
1811 gfx_v7_0_select_se_sh(adev, i, j, 0xffffffff);
1812 data = gfx_v7_0_get_rb_active_bitmap(adev);
1813 active_rbs |= data << ((i * adev->gfx.config.max_sh_per_se + j) *
1814 rb_bitmap_width_per_sh);
1817 gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
1819 adev->gfx.config.backend_enable_mask = active_rbs;
1820 adev->gfx.config.num_rbs = hweight32(active_rbs);
1822 num_rb_pipes = min_t(unsigned, adev->gfx.config.max_backends_per_se *
1823 adev->gfx.config.max_shader_engines, 16);
1825 gfx_v7_0_raster_config(adev, &raster_config, &raster_config_1);
1827 if (!adev->gfx.config.backend_enable_mask ||
1828 adev->gfx.config.num_rbs >= num_rb_pipes) {
1829 WREG32(mmPA_SC_RASTER_CONFIG, raster_config);
1830 WREG32(mmPA_SC_RASTER_CONFIG_1, raster_config_1);
1832 gfx_v7_0_write_harvested_raster_configs(adev, raster_config, raster_config_1,
1833 adev->gfx.config.backend_enable_mask,
1836 mutex_unlock(&adev->grbm_idx_mutex);
1840 * gmc_v7_0_init_compute_vmid - gart enable
1842 * @rdev: amdgpu_device pointer
1844 * Initialize compute vmid sh_mem registers
1847 #define DEFAULT_SH_MEM_BASES (0x6000)
1848 #define FIRST_COMPUTE_VMID (8)
1849 #define LAST_COMPUTE_VMID (16)
1850 static void gmc_v7_0_init_compute_vmid(struct amdgpu_device *adev)
1853 uint32_t sh_mem_config;
1854 uint32_t sh_mem_bases;
1857 * Configure apertures:
1858 * LDS: 0x60000000'00000000 - 0x60000001'00000000 (4GB)
1859 * Scratch: 0x60000001'00000000 - 0x60000002'00000000 (4GB)
1860 * GPUVM: 0x60010000'00000000 - 0x60020000'00000000 (1TB)
1862 sh_mem_bases = DEFAULT_SH_MEM_BASES | (DEFAULT_SH_MEM_BASES << 16);
1863 sh_mem_config = SH_MEM_ALIGNMENT_MODE_UNALIGNED <<
1864 SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT;
1865 sh_mem_config |= MTYPE_NONCACHED << SH_MEM_CONFIG__DEFAULT_MTYPE__SHIFT;
1866 mutex_lock(&adev->srbm_mutex);
1867 for (i = FIRST_COMPUTE_VMID; i < LAST_COMPUTE_VMID; i++) {
1868 cik_srbm_select(adev, 0, 0, 0, i);
1869 /* CP and shaders */
1870 WREG32(mmSH_MEM_CONFIG, sh_mem_config);
1871 WREG32(mmSH_MEM_APE1_BASE, 1);
1872 WREG32(mmSH_MEM_APE1_LIMIT, 0);
1873 WREG32(mmSH_MEM_BASES, sh_mem_bases);
1875 cik_srbm_select(adev, 0, 0, 0, 0);
1876 mutex_unlock(&adev->srbm_mutex);
1880 * gfx_v7_0_gpu_init - setup the 3D engine
1882 * @adev: amdgpu_device pointer
1884 * Configures the 3D engine and tiling configuration
1885 * registers so that the 3D engine is usable.
1887 static void gfx_v7_0_gpu_init(struct amdgpu_device *adev)
1889 u32 tmp, sh_mem_cfg;
1892 WREG32(mmGRBM_CNTL, (0xff << GRBM_CNTL__READ_TIMEOUT__SHIFT));
1894 WREG32(mmGB_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
1895 WREG32(mmHDP_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
1896 WREG32(mmDMIF_ADDR_CALC, adev->gfx.config.gb_addr_config);
1898 gfx_v7_0_tiling_mode_table_init(adev);
1900 gfx_v7_0_setup_rb(adev);
1901 gfx_v7_0_get_cu_info(adev);
1903 /* set HW defaults for 3D engine */
1904 WREG32(mmCP_MEQ_THRESHOLDS,
1905 (0x30 << CP_MEQ_THRESHOLDS__MEQ1_START__SHIFT) |
1906 (0x60 << CP_MEQ_THRESHOLDS__MEQ2_START__SHIFT));
1908 mutex_lock(&adev->grbm_idx_mutex);
1910 * making sure that the following register writes will be broadcasted
1911 * to all the shaders
1913 gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
1915 /* XXX SH_MEM regs */
1916 /* where to put LDS, scratch, GPUVM in FSA64 space */
1917 sh_mem_cfg = REG_SET_FIELD(0, SH_MEM_CONFIG, ALIGNMENT_MODE,
1918 SH_MEM_ALIGNMENT_MODE_UNALIGNED);
1920 mutex_lock(&adev->srbm_mutex);
1921 for (i = 0; i < 16; i++) {
1922 cik_srbm_select(adev, 0, 0, 0, i);
1923 /* CP and shaders */
1924 WREG32(mmSH_MEM_CONFIG, sh_mem_cfg);
1925 WREG32(mmSH_MEM_APE1_BASE, 1);
1926 WREG32(mmSH_MEM_APE1_LIMIT, 0);
1927 WREG32(mmSH_MEM_BASES, 0);
1929 cik_srbm_select(adev, 0, 0, 0, 0);
1930 mutex_unlock(&adev->srbm_mutex);
1932 gmc_v7_0_init_compute_vmid(adev);
1934 WREG32(mmSX_DEBUG_1, 0x20);
1936 WREG32(mmTA_CNTL_AUX, 0x00010000);
1938 tmp = RREG32(mmSPI_CONFIG_CNTL);
1940 WREG32(mmSPI_CONFIG_CNTL, tmp);
1942 WREG32(mmSQ_CONFIG, 1);
1944 WREG32(mmDB_DEBUG, 0);
1946 tmp = RREG32(mmDB_DEBUG2) & ~0xf00fffff;
1948 WREG32(mmDB_DEBUG2, tmp);
1950 tmp = RREG32(mmDB_DEBUG3) & ~0x0002021c;
1952 WREG32(mmDB_DEBUG3, tmp);
1954 tmp = RREG32(mmCB_HW_CONTROL) & ~0x00010000;
1956 WREG32(mmCB_HW_CONTROL, tmp);
1958 WREG32(mmSPI_CONFIG_CNTL_1, (4 << SPI_CONFIG_CNTL_1__VTX_DONE_DELAY__SHIFT));
1960 WREG32(mmPA_SC_FIFO_SIZE,
1961 ((adev->gfx.config.sc_prim_fifo_size_frontend << PA_SC_FIFO_SIZE__SC_FRONTEND_PRIM_FIFO_SIZE__SHIFT) |
1962 (adev->gfx.config.sc_prim_fifo_size_backend << PA_SC_FIFO_SIZE__SC_BACKEND_PRIM_FIFO_SIZE__SHIFT) |
1963 (adev->gfx.config.sc_hiz_tile_fifo_size << PA_SC_FIFO_SIZE__SC_HIZ_TILE_FIFO_SIZE__SHIFT) |
1964 (adev->gfx.config.sc_earlyz_tile_fifo_size << PA_SC_FIFO_SIZE__SC_EARLYZ_TILE_FIFO_SIZE__SHIFT)));
1966 WREG32(mmVGT_NUM_INSTANCES, 1);
1968 WREG32(mmCP_PERFMON_CNTL, 0);
1970 WREG32(mmSQ_CONFIG, 0);
1972 WREG32(mmPA_SC_FORCE_EOV_MAX_CNTS,
1973 ((4095 << PA_SC_FORCE_EOV_MAX_CNTS__FORCE_EOV_MAX_CLK_CNT__SHIFT) |
1974 (255 << PA_SC_FORCE_EOV_MAX_CNTS__FORCE_EOV_MAX_REZ_CNT__SHIFT)));
1976 WREG32(mmVGT_CACHE_INVALIDATION,
1977 (VC_AND_TC << VGT_CACHE_INVALIDATION__CACHE_INVALIDATION__SHIFT) |
1978 (ES_AND_GS_AUTO << VGT_CACHE_INVALIDATION__AUTO_INVLD_EN__SHIFT));
1980 WREG32(mmVGT_GS_VERTEX_REUSE, 16);
1981 WREG32(mmPA_SC_LINE_STIPPLE_STATE, 0);
1983 WREG32(mmPA_CL_ENHANCE, PA_CL_ENHANCE__CLIP_VTX_REORDER_ENA_MASK |
1984 (3 << PA_CL_ENHANCE__NUM_CLIP_SEQ__SHIFT));
1985 WREG32(mmPA_SC_ENHANCE, PA_SC_ENHANCE__ENABLE_PA_SC_OUT_OF_ORDER_MASK);
1987 tmp = RREG32(mmSPI_ARB_PRIORITY);
1988 tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS0, 2);
1989 tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS1, 2);
1990 tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS2, 2);
1991 tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS3, 2);
1992 WREG32(mmSPI_ARB_PRIORITY, tmp);
1994 mutex_unlock(&adev->grbm_idx_mutex);
2000 * GPU scratch registers helpers function.
2003 * gfx_v7_0_scratch_init - setup driver info for CP scratch regs
2005 * @adev: amdgpu_device pointer
2007 * Set up the number and offset of the CP scratch registers.
2008 * NOTE: use of CP scratch registers is a legacy inferface and
2009 * is not used by default on newer asics (r6xx+). On newer asics,
2010 * memory buffers are used for fences rather than scratch regs.
2012 static void gfx_v7_0_scratch_init(struct amdgpu_device *adev)
2014 adev->gfx.scratch.num_reg = 7;
2015 adev->gfx.scratch.reg_base = mmSCRATCH_REG0;
2016 adev->gfx.scratch.free_mask = (1u << adev->gfx.scratch.num_reg) - 1;
2020 * gfx_v7_0_ring_test_ring - basic gfx ring test
2022 * @adev: amdgpu_device pointer
2023 * @ring: amdgpu_ring structure holding ring information
2025 * Allocate a scratch register and write to it using the gfx ring (CIK).
2026 * Provides a basic gfx ring test to verify that the ring is working.
2027 * Used by gfx_v7_0_cp_gfx_resume();
2028 * Returns 0 on success, error on failure.
2030 static int gfx_v7_0_ring_test_ring(struct amdgpu_ring *ring)
2032 struct amdgpu_device *adev = ring->adev;
2038 r = amdgpu_gfx_scratch_get(adev, &scratch);
2040 DRM_ERROR("amdgpu: cp failed to get scratch reg (%d).\n", r);
2043 WREG32(scratch, 0xCAFEDEAD);
2044 r = amdgpu_ring_alloc(ring, 3);
2046 DRM_ERROR("amdgpu: cp failed to lock ring %d (%d).\n", ring->idx, r);
2047 amdgpu_gfx_scratch_free(adev, scratch);
2050 amdgpu_ring_write(ring, PACKET3(PACKET3_SET_UCONFIG_REG, 1));
2051 amdgpu_ring_write(ring, (scratch - PACKET3_SET_UCONFIG_REG_START));
2052 amdgpu_ring_write(ring, 0xDEADBEEF);
2053 amdgpu_ring_commit(ring);
2055 for (i = 0; i < adev->usec_timeout; i++) {
2056 tmp = RREG32(scratch);
2057 if (tmp == 0xDEADBEEF)
2061 if (i < adev->usec_timeout) {
2062 DRM_INFO("ring test on %d succeeded in %d usecs\n", ring->idx, i);
2064 DRM_ERROR("amdgpu: ring %d test failed (scratch(0x%04X)=0x%08X)\n",
2065 ring->idx, scratch, tmp);
2068 amdgpu_gfx_scratch_free(adev, scratch);
2073 * gfx_v7_0_ring_emit_hdp - emit an hdp flush on the cp
2075 * @adev: amdgpu_device pointer
2076 * @ridx: amdgpu ring index
2078 * Emits an hdp flush on the cp.
2080 static void gfx_v7_0_ring_emit_hdp_flush(struct amdgpu_ring *ring)
2083 int usepfp = ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE ? 0 : 1;
2085 if (ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) {
2088 ref_and_mask = GPU_HDP_FLUSH_DONE__CP2_MASK << ring->pipe;
2091 ref_and_mask = GPU_HDP_FLUSH_DONE__CP6_MASK << ring->pipe;
2097 ref_and_mask = GPU_HDP_FLUSH_DONE__CP0_MASK;
2100 amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
2101 amdgpu_ring_write(ring, (WAIT_REG_MEM_OPERATION(1) | /* write, wait, write */
2102 WAIT_REG_MEM_FUNCTION(3) | /* == */
2103 WAIT_REG_MEM_ENGINE(usepfp))); /* pfp or me */
2104 amdgpu_ring_write(ring, mmGPU_HDP_FLUSH_REQ);
2105 amdgpu_ring_write(ring, mmGPU_HDP_FLUSH_DONE);
2106 amdgpu_ring_write(ring, ref_and_mask);
2107 amdgpu_ring_write(ring, ref_and_mask);
2108 amdgpu_ring_write(ring, 0x20); /* poll interval */
2111 static void gfx_v7_0_ring_emit_vgt_flush(struct amdgpu_ring *ring)
2113 amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE, 0));
2114 amdgpu_ring_write(ring, EVENT_TYPE(VS_PARTIAL_FLUSH) |
2117 amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE, 0));
2118 amdgpu_ring_write(ring, EVENT_TYPE(VGT_FLUSH) |
2124 * gfx_v7_0_ring_emit_hdp_invalidate - emit an hdp invalidate on the cp
2126 * @adev: amdgpu_device pointer
2127 * @ridx: amdgpu ring index
2129 * Emits an hdp invalidate on the cp.
2131 static void gfx_v7_0_ring_emit_hdp_invalidate(struct amdgpu_ring *ring)
2133 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
2134 amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
2135 WRITE_DATA_DST_SEL(0) |
2137 amdgpu_ring_write(ring, mmHDP_DEBUG0);
2138 amdgpu_ring_write(ring, 0);
2139 amdgpu_ring_write(ring, 1);
2143 * gfx_v7_0_ring_emit_fence_gfx - emit a fence on the gfx ring
2145 * @adev: amdgpu_device pointer
2146 * @fence: amdgpu fence object
2148 * Emits a fence sequnce number on the gfx ring and flushes
2151 static void gfx_v7_0_ring_emit_fence_gfx(struct amdgpu_ring *ring, u64 addr,
2152 u64 seq, unsigned flags)
2154 bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT;
2155 bool int_sel = flags & AMDGPU_FENCE_FLAG_INT;
2156 /* Workaround for cache flush problems. First send a dummy EOP
2157 * event down the pipe with seq one below.
2159 amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE_EOP, 4));
2160 amdgpu_ring_write(ring, (EOP_TCL1_ACTION_EN |
2162 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
2164 amdgpu_ring_write(ring, addr & 0xfffffffc);
2165 amdgpu_ring_write(ring, (upper_32_bits(addr) & 0xffff) |
2166 DATA_SEL(1) | INT_SEL(0));
2167 amdgpu_ring_write(ring, lower_32_bits(seq - 1));
2168 amdgpu_ring_write(ring, upper_32_bits(seq - 1));
2170 /* Then send the real EOP event down the pipe. */
2171 amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE_EOP, 4));
2172 amdgpu_ring_write(ring, (EOP_TCL1_ACTION_EN |
2174 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
2176 amdgpu_ring_write(ring, addr & 0xfffffffc);
2177 amdgpu_ring_write(ring, (upper_32_bits(addr) & 0xffff) |
2178 DATA_SEL(write64bit ? 2 : 1) | INT_SEL(int_sel ? 2 : 0));
2179 amdgpu_ring_write(ring, lower_32_bits(seq));
2180 amdgpu_ring_write(ring, upper_32_bits(seq));
2184 * gfx_v7_0_ring_emit_fence_compute - emit a fence on the compute ring
2186 * @adev: amdgpu_device pointer
2187 * @fence: amdgpu fence object
2189 * Emits a fence sequnce number on the compute ring and flushes
2192 static void gfx_v7_0_ring_emit_fence_compute(struct amdgpu_ring *ring,
2196 bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT;
2197 bool int_sel = flags & AMDGPU_FENCE_FLAG_INT;
2199 /* RELEASE_MEM - flush caches, send int */
2200 amdgpu_ring_write(ring, PACKET3(PACKET3_RELEASE_MEM, 5));
2201 amdgpu_ring_write(ring, (EOP_TCL1_ACTION_EN |
2203 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
2205 amdgpu_ring_write(ring, DATA_SEL(write64bit ? 2 : 1) | INT_SEL(int_sel ? 2 : 0));
2206 amdgpu_ring_write(ring, addr & 0xfffffffc);
2207 amdgpu_ring_write(ring, upper_32_bits(addr));
2208 amdgpu_ring_write(ring, lower_32_bits(seq));
2209 amdgpu_ring_write(ring, upper_32_bits(seq));
2216 * gfx_v7_0_ring_emit_ib - emit an IB (Indirect Buffer) on the ring
2218 * @ring: amdgpu_ring structure holding ring information
2219 * @ib: amdgpu indirect buffer object
2221 * Emits an DE (drawing engine) or CE (constant engine) IB
2222 * on the gfx ring. IBs are usually generated by userspace
2223 * acceleration drivers and submitted to the kernel for
2224 * sheduling on the ring. This function schedules the IB
2225 * on the gfx ring for execution by the GPU.
2227 static void gfx_v7_0_ring_emit_ib_gfx(struct amdgpu_ring *ring,
2228 struct amdgpu_ib *ib,
2229 unsigned vm_id, bool ctx_switch)
2231 u32 header, control = 0;
2233 /* insert SWITCH_BUFFER packet before first IB in the ring frame */
2235 amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
2236 amdgpu_ring_write(ring, 0);
2239 if (ib->flags & AMDGPU_IB_FLAG_CE)
2240 header = PACKET3(PACKET3_INDIRECT_BUFFER_CONST, 2);
2242 header = PACKET3(PACKET3_INDIRECT_BUFFER, 2);
2244 control |= ib->length_dw | (vm_id << 24);
2246 amdgpu_ring_write(ring, header);
2247 amdgpu_ring_write(ring,
2251 (ib->gpu_addr & 0xFFFFFFFC));
2252 amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xFFFF);
2253 amdgpu_ring_write(ring, control);
2256 static void gfx_v7_0_ring_emit_ib_compute(struct amdgpu_ring *ring,
2257 struct amdgpu_ib *ib,
2258 unsigned vm_id, bool ctx_switch)
2260 u32 control = INDIRECT_BUFFER_VALID | ib->length_dw | (vm_id << 24);
2262 amdgpu_ring_write(ring, PACKET3(PACKET3_INDIRECT_BUFFER, 2));
2263 amdgpu_ring_write(ring,
2267 (ib->gpu_addr & 0xFFFFFFFC));
2268 amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xFFFF);
2269 amdgpu_ring_write(ring, control);
2272 static void gfx_v7_ring_emit_cntxcntl(struct amdgpu_ring *ring, uint32_t flags)
2276 dw2 |= 0x80000000; /* set load_enable otherwise this package is just NOPs */
2277 if (flags & AMDGPU_HAVE_CTX_SWITCH) {
2278 gfx_v7_0_ring_emit_vgt_flush(ring);
2279 /* set load_global_config & load_global_uconfig */
2281 /* set load_cs_sh_regs */
2283 /* set load_per_context_state & load_gfx_sh_regs */
2287 amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1));
2288 amdgpu_ring_write(ring, dw2);
2289 amdgpu_ring_write(ring, 0);
2293 * gfx_v7_0_ring_test_ib - basic ring IB test
2295 * @ring: amdgpu_ring structure holding ring information
2297 * Allocate an IB and execute it on the gfx ring (CIK).
2298 * Provides a basic gfx ring test to verify that IBs are working.
2299 * Returns 0 on success, error on failure.
2301 static int gfx_v7_0_ring_test_ib(struct amdgpu_ring *ring, long timeout)
2303 struct amdgpu_device *adev = ring->adev;
2304 struct amdgpu_ib ib;
2305 struct dma_fence *f = NULL;
2310 r = amdgpu_gfx_scratch_get(adev, &scratch);
2312 DRM_ERROR("amdgpu: failed to get scratch reg (%ld).\n", r);
2315 WREG32(scratch, 0xCAFEDEAD);
2316 memset(&ib, 0, sizeof(ib));
2317 r = amdgpu_ib_get(adev, NULL, 256, &ib);
2319 DRM_ERROR("amdgpu: failed to get ib (%ld).\n", r);
2322 ib.ptr[0] = PACKET3(PACKET3_SET_UCONFIG_REG, 1);
2323 ib.ptr[1] = ((scratch - PACKET3_SET_UCONFIG_REG_START));
2324 ib.ptr[2] = 0xDEADBEEF;
2327 r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f);
2331 r = dma_fence_wait_timeout(f, false, timeout);
2333 DRM_ERROR("amdgpu: IB test timed out\n");
2337 DRM_ERROR("amdgpu: fence wait failed (%ld).\n", r);
2340 tmp = RREG32(scratch);
2341 if (tmp == 0xDEADBEEF) {
2342 DRM_INFO("ib test on ring %d succeeded\n", ring->idx);
2345 DRM_ERROR("amdgpu: ib test failed (scratch(0x%04X)=0x%08X)\n",
2351 amdgpu_ib_free(adev, &ib, NULL);
2354 amdgpu_gfx_scratch_free(adev, scratch);
2360 * On CIK, gfx and compute now have independant command processors.
2363 * Gfx consists of a single ring and can process both gfx jobs and
2364 * compute jobs. The gfx CP consists of three microengines (ME):
2365 * PFP - Pre-Fetch Parser
2367 * CE - Constant Engine
2368 * The PFP and ME make up what is considered the Drawing Engine (DE).
2369 * The CE is an asynchronous engine used for updating buffer desciptors
2370 * used by the DE so that they can be loaded into cache in parallel
2371 * while the DE is processing state update packets.
2374 * The compute CP consists of two microengines (ME):
2375 * MEC1 - Compute MicroEngine 1
2376 * MEC2 - Compute MicroEngine 2
2377 * Each MEC supports 4 compute pipes and each pipe supports 8 queues.
2378 * The queues are exposed to userspace and are programmed directly
2379 * by the compute runtime.
2382 * gfx_v7_0_cp_gfx_enable - enable/disable the gfx CP MEs
2384 * @adev: amdgpu_device pointer
2385 * @enable: enable or disable the MEs
2387 * Halts or unhalts the gfx MEs.
2389 static void gfx_v7_0_cp_gfx_enable(struct amdgpu_device *adev, bool enable)
2394 WREG32(mmCP_ME_CNTL, 0);
2396 WREG32(mmCP_ME_CNTL, (CP_ME_CNTL__ME_HALT_MASK | CP_ME_CNTL__PFP_HALT_MASK | CP_ME_CNTL__CE_HALT_MASK));
2397 for (i = 0; i < adev->gfx.num_gfx_rings; i++)
2398 adev->gfx.gfx_ring[i].ready = false;
2404 * gfx_v7_0_cp_gfx_load_microcode - load the gfx CP ME ucode
2406 * @adev: amdgpu_device pointer
2408 * Loads the gfx PFP, ME, and CE ucode.
2409 * Returns 0 for success, -EINVAL if the ucode is not available.
2411 static int gfx_v7_0_cp_gfx_load_microcode(struct amdgpu_device *adev)
2413 const struct gfx_firmware_header_v1_0 *pfp_hdr;
2414 const struct gfx_firmware_header_v1_0 *ce_hdr;
2415 const struct gfx_firmware_header_v1_0 *me_hdr;
2416 const __le32 *fw_data;
2417 unsigned i, fw_size;
2419 if (!adev->gfx.me_fw || !adev->gfx.pfp_fw || !adev->gfx.ce_fw)
2422 pfp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.pfp_fw->data;
2423 ce_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.ce_fw->data;
2424 me_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.me_fw->data;
2426 amdgpu_ucode_print_gfx_hdr(&pfp_hdr->header);
2427 amdgpu_ucode_print_gfx_hdr(&ce_hdr->header);
2428 amdgpu_ucode_print_gfx_hdr(&me_hdr->header);
2429 adev->gfx.pfp_fw_version = le32_to_cpu(pfp_hdr->header.ucode_version);
2430 adev->gfx.ce_fw_version = le32_to_cpu(ce_hdr->header.ucode_version);
2431 adev->gfx.me_fw_version = le32_to_cpu(me_hdr->header.ucode_version);
2432 adev->gfx.me_feature_version = le32_to_cpu(me_hdr->ucode_feature_version);
2433 adev->gfx.ce_feature_version = le32_to_cpu(ce_hdr->ucode_feature_version);
2434 adev->gfx.pfp_feature_version = le32_to_cpu(pfp_hdr->ucode_feature_version);
2436 gfx_v7_0_cp_gfx_enable(adev, false);
2439 fw_data = (const __le32 *)
2440 (adev->gfx.pfp_fw->data +
2441 le32_to_cpu(pfp_hdr->header.ucode_array_offset_bytes));
2442 fw_size = le32_to_cpu(pfp_hdr->header.ucode_size_bytes) / 4;
2443 WREG32(mmCP_PFP_UCODE_ADDR, 0);
2444 for (i = 0; i < fw_size; i++)
2445 WREG32(mmCP_PFP_UCODE_DATA, le32_to_cpup(fw_data++));
2446 WREG32(mmCP_PFP_UCODE_ADDR, adev->gfx.pfp_fw_version);
2449 fw_data = (const __le32 *)
2450 (adev->gfx.ce_fw->data +
2451 le32_to_cpu(ce_hdr->header.ucode_array_offset_bytes));
2452 fw_size = le32_to_cpu(ce_hdr->header.ucode_size_bytes) / 4;
2453 WREG32(mmCP_CE_UCODE_ADDR, 0);
2454 for (i = 0; i < fw_size; i++)
2455 WREG32(mmCP_CE_UCODE_DATA, le32_to_cpup(fw_data++));
2456 WREG32(mmCP_CE_UCODE_ADDR, adev->gfx.ce_fw_version);
2459 fw_data = (const __le32 *)
2460 (adev->gfx.me_fw->data +
2461 le32_to_cpu(me_hdr->header.ucode_array_offset_bytes));
2462 fw_size = le32_to_cpu(me_hdr->header.ucode_size_bytes) / 4;
2463 WREG32(mmCP_ME_RAM_WADDR, 0);
2464 for (i = 0; i < fw_size; i++)
2465 WREG32(mmCP_ME_RAM_DATA, le32_to_cpup(fw_data++));
2466 WREG32(mmCP_ME_RAM_WADDR, adev->gfx.me_fw_version);
2472 * gfx_v7_0_cp_gfx_start - start the gfx ring
2474 * @adev: amdgpu_device pointer
2476 * Enables the ring and loads the clear state context and other
2477 * packets required to init the ring.
2478 * Returns 0 for success, error for failure.
2480 static int gfx_v7_0_cp_gfx_start(struct amdgpu_device *adev)
2482 struct amdgpu_ring *ring = &adev->gfx.gfx_ring[0];
2483 const struct cs_section_def *sect = NULL;
2484 const struct cs_extent_def *ext = NULL;
2488 WREG32(mmCP_MAX_CONTEXT, adev->gfx.config.max_hw_contexts - 1);
2489 WREG32(mmCP_ENDIAN_SWAP, 0);
2490 WREG32(mmCP_DEVICE_ID, 1);
2492 gfx_v7_0_cp_gfx_enable(adev, true);
2494 r = amdgpu_ring_alloc(ring, gfx_v7_0_get_csb_size(adev) + 8);
2496 DRM_ERROR("amdgpu: cp failed to lock ring (%d).\n", r);
2500 /* init the CE partitions. CE only used for gfx on CIK */
2501 amdgpu_ring_write(ring, PACKET3(PACKET3_SET_BASE, 2));
2502 amdgpu_ring_write(ring, PACKET3_BASE_INDEX(CE_PARTITION_BASE));
2503 amdgpu_ring_write(ring, 0x8000);
2504 amdgpu_ring_write(ring, 0x8000);
2506 /* clear state buffer */
2507 amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
2508 amdgpu_ring_write(ring, PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
2510 amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1));
2511 amdgpu_ring_write(ring, 0x80000000);
2512 amdgpu_ring_write(ring, 0x80000000);
2514 for (sect = adev->gfx.rlc.cs_data; sect->section != NULL; ++sect) {
2515 for (ext = sect->section; ext->extent != NULL; ++ext) {
2516 if (sect->id == SECT_CONTEXT) {
2517 amdgpu_ring_write(ring,
2518 PACKET3(PACKET3_SET_CONTEXT_REG, ext->reg_count));
2519 amdgpu_ring_write(ring, ext->reg_index - PACKET3_SET_CONTEXT_REG_START);
2520 for (i = 0; i < ext->reg_count; i++)
2521 amdgpu_ring_write(ring, ext->extent[i]);
2526 amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 2));
2527 amdgpu_ring_write(ring, mmPA_SC_RASTER_CONFIG - PACKET3_SET_CONTEXT_REG_START);
2528 switch (adev->asic_type) {
2530 amdgpu_ring_write(ring, 0x16000012);
2531 amdgpu_ring_write(ring, 0x00000000);
2534 amdgpu_ring_write(ring, 0x00000000); /* XXX */
2535 amdgpu_ring_write(ring, 0x00000000);
2539 amdgpu_ring_write(ring, 0x00000000); /* XXX */
2540 amdgpu_ring_write(ring, 0x00000000);
2543 amdgpu_ring_write(ring, 0x3a00161a);
2544 amdgpu_ring_write(ring, 0x0000002e);
2547 amdgpu_ring_write(ring, 0x00000000);
2548 amdgpu_ring_write(ring, 0x00000000);
2552 amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
2553 amdgpu_ring_write(ring, PACKET3_PREAMBLE_END_CLEAR_STATE);
2555 amdgpu_ring_write(ring, PACKET3(PACKET3_CLEAR_STATE, 0));
2556 amdgpu_ring_write(ring, 0);
2558 amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 2));
2559 amdgpu_ring_write(ring, 0x00000316);
2560 amdgpu_ring_write(ring, 0x0000000e); /* VGT_VERTEX_REUSE_BLOCK_CNTL */
2561 amdgpu_ring_write(ring, 0x00000010); /* VGT_OUT_DEALLOC_CNTL */
2563 amdgpu_ring_commit(ring);
2569 * gfx_v7_0_cp_gfx_resume - setup the gfx ring buffer registers
2571 * @adev: amdgpu_device pointer
2573 * Program the location and size of the gfx ring buffer
2574 * and test it to make sure it's working.
2575 * Returns 0 for success, error for failure.
2577 static int gfx_v7_0_cp_gfx_resume(struct amdgpu_device *adev)
2579 struct amdgpu_ring *ring;
2582 u64 rb_addr, rptr_addr;
2585 WREG32(mmCP_SEM_WAIT_TIMER, 0x0);
2586 if (adev->asic_type != CHIP_HAWAII)
2587 WREG32(mmCP_SEM_INCOMPLETE_TIMER_CNTL, 0x0);
2589 /* Set the write pointer delay */
2590 WREG32(mmCP_RB_WPTR_DELAY, 0);
2592 /* set the RB to use vmid 0 */
2593 WREG32(mmCP_RB_VMID, 0);
2595 WREG32(mmSCRATCH_ADDR, 0);
2597 /* ring 0 - compute and gfx */
2598 /* Set ring buffer size */
2599 ring = &adev->gfx.gfx_ring[0];
2600 rb_bufsz = order_base_2(ring->ring_size / 8);
2601 tmp = (order_base_2(AMDGPU_GPU_PAGE_SIZE/8) << 8) | rb_bufsz;
2603 tmp |= 2 << CP_RB0_CNTL__BUF_SWAP__SHIFT;
2605 WREG32(mmCP_RB0_CNTL, tmp);
2607 /* Initialize the ring buffer's read and write pointers */
2608 WREG32(mmCP_RB0_CNTL, tmp | CP_RB0_CNTL__RB_RPTR_WR_ENA_MASK);
2610 WREG32(mmCP_RB0_WPTR, ring->wptr);
2612 /* set the wb address wether it's enabled or not */
2613 rptr_addr = adev->wb.gpu_addr + (ring->rptr_offs * 4);
2614 WREG32(mmCP_RB0_RPTR_ADDR, lower_32_bits(rptr_addr));
2615 WREG32(mmCP_RB0_RPTR_ADDR_HI, upper_32_bits(rptr_addr) & 0xFF);
2617 /* scratch register shadowing is no longer supported */
2618 WREG32(mmSCRATCH_UMSK, 0);
2621 WREG32(mmCP_RB0_CNTL, tmp);
2623 rb_addr = ring->gpu_addr >> 8;
2624 WREG32(mmCP_RB0_BASE, rb_addr);
2625 WREG32(mmCP_RB0_BASE_HI, upper_32_bits(rb_addr));
2627 /* start the ring */
2628 gfx_v7_0_cp_gfx_start(adev);
2630 r = amdgpu_ring_test_ring(ring);
2632 ring->ready = false;
2639 static u32 gfx_v7_0_ring_get_rptr(struct amdgpu_ring *ring)
2641 return ring->adev->wb.wb[ring->rptr_offs];
2644 static u32 gfx_v7_0_ring_get_wptr_gfx(struct amdgpu_ring *ring)
2646 struct amdgpu_device *adev = ring->adev;
2648 return RREG32(mmCP_RB0_WPTR);
2651 static void gfx_v7_0_ring_set_wptr_gfx(struct amdgpu_ring *ring)
2653 struct amdgpu_device *adev = ring->adev;
2655 WREG32(mmCP_RB0_WPTR, ring->wptr);
2656 (void)RREG32(mmCP_RB0_WPTR);
2659 static u32 gfx_v7_0_ring_get_wptr_compute(struct amdgpu_ring *ring)
2661 /* XXX check if swapping is necessary on BE */
2662 return ring->adev->wb.wb[ring->wptr_offs];
2665 static void gfx_v7_0_ring_set_wptr_compute(struct amdgpu_ring *ring)
2667 struct amdgpu_device *adev = ring->adev;
2669 /* XXX check if swapping is necessary on BE */
2670 adev->wb.wb[ring->wptr_offs] = ring->wptr;
2671 WDOORBELL32(ring->doorbell_index, ring->wptr);
2675 * gfx_v7_0_cp_compute_enable - enable/disable the compute CP MEs
2677 * @adev: amdgpu_device pointer
2678 * @enable: enable or disable the MEs
2680 * Halts or unhalts the compute MEs.
2682 static void gfx_v7_0_cp_compute_enable(struct amdgpu_device *adev, bool enable)
2687 WREG32(mmCP_MEC_CNTL, 0);
2689 WREG32(mmCP_MEC_CNTL, (CP_MEC_CNTL__MEC_ME1_HALT_MASK | CP_MEC_CNTL__MEC_ME2_HALT_MASK));
2690 for (i = 0; i < adev->gfx.num_compute_rings; i++)
2691 adev->gfx.compute_ring[i].ready = false;
2697 * gfx_v7_0_cp_compute_load_microcode - load the compute CP ME ucode
2699 * @adev: amdgpu_device pointer
2701 * Loads the compute MEC1&2 ucode.
2702 * Returns 0 for success, -EINVAL if the ucode is not available.
2704 static int gfx_v7_0_cp_compute_load_microcode(struct amdgpu_device *adev)
2706 const struct gfx_firmware_header_v1_0 *mec_hdr;
2707 const __le32 *fw_data;
2708 unsigned i, fw_size;
2710 if (!adev->gfx.mec_fw)
2713 mec_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data;
2714 amdgpu_ucode_print_gfx_hdr(&mec_hdr->header);
2715 adev->gfx.mec_fw_version = le32_to_cpu(mec_hdr->header.ucode_version);
2716 adev->gfx.mec_feature_version = le32_to_cpu(
2717 mec_hdr->ucode_feature_version);
2719 gfx_v7_0_cp_compute_enable(adev, false);
2722 fw_data = (const __le32 *)
2723 (adev->gfx.mec_fw->data +
2724 le32_to_cpu(mec_hdr->header.ucode_array_offset_bytes));
2725 fw_size = le32_to_cpu(mec_hdr->header.ucode_size_bytes) / 4;
2726 WREG32(mmCP_MEC_ME1_UCODE_ADDR, 0);
2727 for (i = 0; i < fw_size; i++)
2728 WREG32(mmCP_MEC_ME1_UCODE_DATA, le32_to_cpup(fw_data++));
2729 WREG32(mmCP_MEC_ME1_UCODE_ADDR, 0);
2731 if (adev->asic_type == CHIP_KAVERI) {
2732 const struct gfx_firmware_header_v1_0 *mec2_hdr;
2734 if (!adev->gfx.mec2_fw)
2737 mec2_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec2_fw->data;
2738 amdgpu_ucode_print_gfx_hdr(&mec2_hdr->header);
2739 adev->gfx.mec2_fw_version = le32_to_cpu(mec2_hdr->header.ucode_version);
2740 adev->gfx.mec2_feature_version = le32_to_cpu(
2741 mec2_hdr->ucode_feature_version);
2744 fw_data = (const __le32 *)
2745 (adev->gfx.mec2_fw->data +
2746 le32_to_cpu(mec2_hdr->header.ucode_array_offset_bytes));
2747 fw_size = le32_to_cpu(mec2_hdr->header.ucode_size_bytes) / 4;
2748 WREG32(mmCP_MEC_ME2_UCODE_ADDR, 0);
2749 for (i = 0; i < fw_size; i++)
2750 WREG32(mmCP_MEC_ME2_UCODE_DATA, le32_to_cpup(fw_data++));
2751 WREG32(mmCP_MEC_ME2_UCODE_ADDR, 0);
2758 * gfx_v7_0_cp_compute_fini - stop the compute queues
2760 * @adev: amdgpu_device pointer
2762 * Stop the compute queues and tear down the driver queue
2765 static void gfx_v7_0_cp_compute_fini(struct amdgpu_device *adev)
2769 for (i = 0; i < adev->gfx.num_compute_rings; i++) {
2770 struct amdgpu_ring *ring = &adev->gfx.compute_ring[i];
2772 if (ring->mqd_obj) {
2773 r = amdgpu_bo_reserve(ring->mqd_obj, false);
2774 if (unlikely(r != 0))
2775 dev_warn(adev->dev, "(%d) reserve MQD bo failed\n", r);
2777 amdgpu_bo_unpin(ring->mqd_obj);
2778 amdgpu_bo_unreserve(ring->mqd_obj);
2780 amdgpu_bo_unref(&ring->mqd_obj);
2781 ring->mqd_obj = NULL;
2786 static void gfx_v7_0_mec_fini(struct amdgpu_device *adev)
2790 if (adev->gfx.mec.hpd_eop_obj) {
2791 r = amdgpu_bo_reserve(adev->gfx.mec.hpd_eop_obj, false);
2792 if (unlikely(r != 0))
2793 dev_warn(adev->dev, "(%d) reserve HPD EOP bo failed\n", r);
2794 amdgpu_bo_unpin(adev->gfx.mec.hpd_eop_obj);
2795 amdgpu_bo_unreserve(adev->gfx.mec.hpd_eop_obj);
2797 amdgpu_bo_unref(&adev->gfx.mec.hpd_eop_obj);
2798 adev->gfx.mec.hpd_eop_obj = NULL;
2802 #define MEC_HPD_SIZE 2048
2804 static int gfx_v7_0_mec_init(struct amdgpu_device *adev)
2810 * KV: 2 MEC, 4 Pipes/MEC, 8 Queues/Pipe - 64 Queues total
2811 * CI/KB: 1 MEC, 4 Pipes/MEC, 8 Queues/Pipe - 32 Queues total
2812 * Nonetheless, we assign only 1 pipe because all other pipes will
2815 adev->gfx.mec.num_mec = 1;
2816 adev->gfx.mec.num_pipe = 1;
2817 adev->gfx.mec.num_queue = adev->gfx.mec.num_mec * adev->gfx.mec.num_pipe * 8;
2819 if (adev->gfx.mec.hpd_eop_obj == NULL) {
2820 r = amdgpu_bo_create(adev,
2821 adev->gfx.mec.num_mec *adev->gfx.mec.num_pipe * MEC_HPD_SIZE * 2,
2823 AMDGPU_GEM_DOMAIN_GTT, 0, NULL, NULL,
2824 &adev->gfx.mec.hpd_eop_obj);
2826 dev_warn(adev->dev, "(%d) create HDP EOP bo failed\n", r);
2831 r = amdgpu_bo_reserve(adev->gfx.mec.hpd_eop_obj, false);
2832 if (unlikely(r != 0)) {
2833 gfx_v7_0_mec_fini(adev);
2836 r = amdgpu_bo_pin(adev->gfx.mec.hpd_eop_obj, AMDGPU_GEM_DOMAIN_GTT,
2837 &adev->gfx.mec.hpd_eop_gpu_addr);
2839 dev_warn(adev->dev, "(%d) pin HDP EOP bo failed\n", r);
2840 gfx_v7_0_mec_fini(adev);
2843 r = amdgpu_bo_kmap(adev->gfx.mec.hpd_eop_obj, (void **)&hpd);
2845 dev_warn(adev->dev, "(%d) map HDP EOP bo failed\n", r);
2846 gfx_v7_0_mec_fini(adev);
2850 /* clear memory. Not sure if this is required or not */
2851 memset(hpd, 0, adev->gfx.mec.num_mec *adev->gfx.mec.num_pipe * MEC_HPD_SIZE * 2);
2853 amdgpu_bo_kunmap(adev->gfx.mec.hpd_eop_obj);
2854 amdgpu_bo_unreserve(adev->gfx.mec.hpd_eop_obj);
2859 struct hqd_registers
2861 u32 cp_mqd_base_addr;
2862 u32 cp_mqd_base_addr_hi;
2865 u32 cp_hqd_persistent_state;
2866 u32 cp_hqd_pipe_priority;
2867 u32 cp_hqd_queue_priority;
2870 u32 cp_hqd_pq_base_hi;
2872 u32 cp_hqd_pq_rptr_report_addr;
2873 u32 cp_hqd_pq_rptr_report_addr_hi;
2874 u32 cp_hqd_pq_wptr_poll_addr;
2875 u32 cp_hqd_pq_wptr_poll_addr_hi;
2876 u32 cp_hqd_pq_doorbell_control;
2878 u32 cp_hqd_pq_control;
2879 u32 cp_hqd_ib_base_addr;
2880 u32 cp_hqd_ib_base_addr_hi;
2882 u32 cp_hqd_ib_control;
2883 u32 cp_hqd_iq_timer;
2885 u32 cp_hqd_dequeue_request;
2886 u32 cp_hqd_dma_offload;
2887 u32 cp_hqd_sema_cmd;
2888 u32 cp_hqd_msg_type;
2889 u32 cp_hqd_atomic0_preop_lo;
2890 u32 cp_hqd_atomic0_preop_hi;
2891 u32 cp_hqd_atomic1_preop_lo;
2892 u32 cp_hqd_atomic1_preop_hi;
2893 u32 cp_hqd_hq_scheduler0;
2894 u32 cp_hqd_hq_scheduler1;
2901 u32 dispatch_initiator;
2905 u32 pipeline_stat_enable;
2906 u32 perf_counter_enable;
2912 u32 resource_limits;
2913 u32 static_thread_mgmt01[2];
2915 u32 static_thread_mgmt23[2];
2917 u32 thread_trace_enable;
2920 u32 vgtcs_invoke_count[2];
2921 struct hqd_registers queue_state;
2923 u32 interrupt_queue[64];
2927 * gfx_v7_0_cp_compute_resume - setup the compute queue registers
2929 * @adev: amdgpu_device pointer
2931 * Program the compute queues and test them to make sure they
2933 * Returns 0 for success, error for failure.
2935 static int gfx_v7_0_cp_compute_resume(struct amdgpu_device *adev)
2939 bool use_doorbell = true;
2945 struct bonaire_mqd *mqd;
2946 struct amdgpu_ring *ring;
2948 /* fix up chicken bits */
2949 tmp = RREG32(mmCP_CPF_DEBUG);
2951 WREG32(mmCP_CPF_DEBUG, tmp);
2953 /* init the pipes */
2954 mutex_lock(&adev->srbm_mutex);
2955 for (i = 0; i < (adev->gfx.mec.num_pipe * adev->gfx.mec.num_mec); i++) {
2956 int me = (i < 4) ? 1 : 2;
2957 int pipe = (i < 4) ? i : (i - 4);
2959 eop_gpu_addr = adev->gfx.mec.hpd_eop_gpu_addr + (i * MEC_HPD_SIZE * 2);
2961 cik_srbm_select(adev, me, pipe, 0, 0);
2963 /* write the EOP addr */
2964 WREG32(mmCP_HPD_EOP_BASE_ADDR, eop_gpu_addr >> 8);
2965 WREG32(mmCP_HPD_EOP_BASE_ADDR_HI, upper_32_bits(eop_gpu_addr) >> 8);
2967 /* set the VMID assigned */
2968 WREG32(mmCP_HPD_EOP_VMID, 0);
2970 /* set the EOP size, register value is 2^(EOP_SIZE+1) dwords */
2971 tmp = RREG32(mmCP_HPD_EOP_CONTROL);
2972 tmp &= ~CP_HPD_EOP_CONTROL__EOP_SIZE_MASK;
2973 tmp |= order_base_2(MEC_HPD_SIZE / 8);
2974 WREG32(mmCP_HPD_EOP_CONTROL, tmp);
2976 cik_srbm_select(adev, 0, 0, 0, 0);
2977 mutex_unlock(&adev->srbm_mutex);
2979 /* init the queues. Just two for now. */
2980 for (i = 0; i < adev->gfx.num_compute_rings; i++) {
2981 ring = &adev->gfx.compute_ring[i];
2983 if (ring->mqd_obj == NULL) {
2984 r = amdgpu_bo_create(adev,
2985 sizeof(struct bonaire_mqd),
2987 AMDGPU_GEM_DOMAIN_GTT, 0, NULL, NULL,
2990 dev_warn(adev->dev, "(%d) create MQD bo failed\n", r);
2995 r = amdgpu_bo_reserve(ring->mqd_obj, false);
2996 if (unlikely(r != 0)) {
2997 gfx_v7_0_cp_compute_fini(adev);
3000 r = amdgpu_bo_pin(ring->mqd_obj, AMDGPU_GEM_DOMAIN_GTT,
3003 dev_warn(adev->dev, "(%d) pin MQD bo failed\n", r);
3004 gfx_v7_0_cp_compute_fini(adev);
3007 r = amdgpu_bo_kmap(ring->mqd_obj, (void **)&buf);
3009 dev_warn(adev->dev, "(%d) map MQD bo failed\n", r);
3010 gfx_v7_0_cp_compute_fini(adev);
3014 /* init the mqd struct */
3015 memset(buf, 0, sizeof(struct bonaire_mqd));
3017 mqd = (struct bonaire_mqd *)buf;
3018 mqd->header = 0xC0310800;
3019 mqd->static_thread_mgmt01[0] = 0xffffffff;
3020 mqd->static_thread_mgmt01[1] = 0xffffffff;
3021 mqd->static_thread_mgmt23[0] = 0xffffffff;
3022 mqd->static_thread_mgmt23[1] = 0xffffffff;
3024 mutex_lock(&adev->srbm_mutex);
3025 cik_srbm_select(adev, ring->me,
3029 /* disable wptr polling */
3030 tmp = RREG32(mmCP_PQ_WPTR_POLL_CNTL);
3031 tmp &= ~CP_PQ_WPTR_POLL_CNTL__EN_MASK;
3032 WREG32(mmCP_PQ_WPTR_POLL_CNTL, tmp);
3034 /* enable doorbell? */
3035 mqd->queue_state.cp_hqd_pq_doorbell_control =
3036 RREG32(mmCP_HQD_PQ_DOORBELL_CONTROL);
3038 mqd->queue_state.cp_hqd_pq_doorbell_control |= CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_EN_MASK;
3040 mqd->queue_state.cp_hqd_pq_doorbell_control &= ~CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_EN_MASK;
3041 WREG32(mmCP_HQD_PQ_DOORBELL_CONTROL,
3042 mqd->queue_state.cp_hqd_pq_doorbell_control);
3044 /* disable the queue if it's active */
3045 mqd->queue_state.cp_hqd_dequeue_request = 0;
3046 mqd->queue_state.cp_hqd_pq_rptr = 0;
3047 mqd->queue_state.cp_hqd_pq_wptr= 0;
3048 if (RREG32(mmCP_HQD_ACTIVE) & 1) {
3049 WREG32(mmCP_HQD_DEQUEUE_REQUEST, 1);
3050 for (j = 0; j < adev->usec_timeout; j++) {
3051 if (!(RREG32(mmCP_HQD_ACTIVE) & 1))
3055 WREG32(mmCP_HQD_DEQUEUE_REQUEST, mqd->queue_state.cp_hqd_dequeue_request);
3056 WREG32(mmCP_HQD_PQ_RPTR, mqd->queue_state.cp_hqd_pq_rptr);
3057 WREG32(mmCP_HQD_PQ_WPTR, mqd->queue_state.cp_hqd_pq_wptr);
3060 /* set the pointer to the MQD */
3061 mqd->queue_state.cp_mqd_base_addr = mqd_gpu_addr & 0xfffffffc;
3062 mqd->queue_state.cp_mqd_base_addr_hi = upper_32_bits(mqd_gpu_addr);
3063 WREG32(mmCP_MQD_BASE_ADDR, mqd->queue_state.cp_mqd_base_addr);
3064 WREG32(mmCP_MQD_BASE_ADDR_HI, mqd->queue_state.cp_mqd_base_addr_hi);
3065 /* set MQD vmid to 0 */
3066 mqd->queue_state.cp_mqd_control = RREG32(mmCP_MQD_CONTROL);
3067 mqd->queue_state.cp_mqd_control &= ~CP_MQD_CONTROL__VMID_MASK;
3068 WREG32(mmCP_MQD_CONTROL, mqd->queue_state.cp_mqd_control);
3070 /* set the pointer to the HQD, this is similar CP_RB0_BASE/_HI */
3071 hqd_gpu_addr = ring->gpu_addr >> 8;
3072 mqd->queue_state.cp_hqd_pq_base = hqd_gpu_addr;
3073 mqd->queue_state.cp_hqd_pq_base_hi = upper_32_bits(hqd_gpu_addr);
3074 WREG32(mmCP_HQD_PQ_BASE, mqd->queue_state.cp_hqd_pq_base);
3075 WREG32(mmCP_HQD_PQ_BASE_HI, mqd->queue_state.cp_hqd_pq_base_hi);
3077 /* set up the HQD, this is similar to CP_RB0_CNTL */
3078 mqd->queue_state.cp_hqd_pq_control = RREG32(mmCP_HQD_PQ_CONTROL);
3079 mqd->queue_state.cp_hqd_pq_control &=
3080 ~(CP_HQD_PQ_CONTROL__QUEUE_SIZE_MASK |
3081 CP_HQD_PQ_CONTROL__RPTR_BLOCK_SIZE_MASK);
3083 mqd->queue_state.cp_hqd_pq_control |=
3084 order_base_2(ring->ring_size / 8);
3085 mqd->queue_state.cp_hqd_pq_control |=
3086 (order_base_2(AMDGPU_GPU_PAGE_SIZE/8) << 8);
3088 mqd->queue_state.cp_hqd_pq_control |=
3089 2 << CP_HQD_PQ_CONTROL__ENDIAN_SWAP__SHIFT;
3091 mqd->queue_state.cp_hqd_pq_control &=
3092 ~(CP_HQD_PQ_CONTROL__UNORD_DISPATCH_MASK |
3093 CP_HQD_PQ_CONTROL__ROQ_PQ_IB_FLIP_MASK |
3094 CP_HQD_PQ_CONTROL__PQ_VOLATILE_MASK);
3095 mqd->queue_state.cp_hqd_pq_control |=
3096 CP_HQD_PQ_CONTROL__PRIV_STATE_MASK |
3097 CP_HQD_PQ_CONTROL__KMD_QUEUE_MASK; /* assuming kernel queue control */
3098 WREG32(mmCP_HQD_PQ_CONTROL, mqd->queue_state.cp_hqd_pq_control);
3100 /* only used if CP_PQ_WPTR_POLL_CNTL.CP_PQ_WPTR_POLL_CNTL__EN_MASK=1 */
3101 wb_gpu_addr = adev->wb.gpu_addr + (ring->wptr_offs * 4);
3102 mqd->queue_state.cp_hqd_pq_wptr_poll_addr = wb_gpu_addr & 0xfffffffc;
3103 mqd->queue_state.cp_hqd_pq_wptr_poll_addr_hi = upper_32_bits(wb_gpu_addr) & 0xffff;
3104 WREG32(mmCP_HQD_PQ_WPTR_POLL_ADDR, mqd->queue_state.cp_hqd_pq_wptr_poll_addr);
3105 WREG32(mmCP_HQD_PQ_WPTR_POLL_ADDR_HI,
3106 mqd->queue_state.cp_hqd_pq_wptr_poll_addr_hi);
3108 /* set the wb address wether it's enabled or not */
3109 wb_gpu_addr = adev->wb.gpu_addr + (ring->rptr_offs * 4);
3110 mqd->queue_state.cp_hqd_pq_rptr_report_addr = wb_gpu_addr & 0xfffffffc;
3111 mqd->queue_state.cp_hqd_pq_rptr_report_addr_hi =
3112 upper_32_bits(wb_gpu_addr) & 0xffff;
3113 WREG32(mmCP_HQD_PQ_RPTR_REPORT_ADDR,
3114 mqd->queue_state.cp_hqd_pq_rptr_report_addr);
3115 WREG32(mmCP_HQD_PQ_RPTR_REPORT_ADDR_HI,
3116 mqd->queue_state.cp_hqd_pq_rptr_report_addr_hi);
3118 /* enable the doorbell if requested */
3120 mqd->queue_state.cp_hqd_pq_doorbell_control =
3121 RREG32(mmCP_HQD_PQ_DOORBELL_CONTROL);
3122 mqd->queue_state.cp_hqd_pq_doorbell_control &=
3123 ~CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_OFFSET_MASK;
3124 mqd->queue_state.cp_hqd_pq_doorbell_control |=
3125 (ring->doorbell_index <<
3126 CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_OFFSET__SHIFT);
3127 mqd->queue_state.cp_hqd_pq_doorbell_control |=
3128 CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_EN_MASK;
3129 mqd->queue_state.cp_hqd_pq_doorbell_control &=
3130 ~(CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_SOURCE_MASK |
3131 CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_HIT_MASK);
3134 mqd->queue_state.cp_hqd_pq_doorbell_control = 0;
3136 WREG32(mmCP_HQD_PQ_DOORBELL_CONTROL,
3137 mqd->queue_state.cp_hqd_pq_doorbell_control);
3139 /* read and write pointers, similar to CP_RB0_WPTR/_RPTR */
3141 mqd->queue_state.cp_hqd_pq_wptr = ring->wptr;
3142 WREG32(mmCP_HQD_PQ_WPTR, mqd->queue_state.cp_hqd_pq_wptr);
3143 mqd->queue_state.cp_hqd_pq_rptr = RREG32(mmCP_HQD_PQ_RPTR);
3145 /* set the vmid for the queue */
3146 mqd->queue_state.cp_hqd_vmid = 0;
3147 WREG32(mmCP_HQD_VMID, mqd->queue_state.cp_hqd_vmid);
3149 /* activate the queue */
3150 mqd->queue_state.cp_hqd_active = 1;
3151 WREG32(mmCP_HQD_ACTIVE, mqd->queue_state.cp_hqd_active);
3153 cik_srbm_select(adev, 0, 0, 0, 0);
3154 mutex_unlock(&adev->srbm_mutex);
3156 amdgpu_bo_kunmap(ring->mqd_obj);
3157 amdgpu_bo_unreserve(ring->mqd_obj);
3162 gfx_v7_0_cp_compute_enable(adev, true);
3164 for (i = 0; i < adev->gfx.num_compute_rings; i++) {
3165 ring = &adev->gfx.compute_ring[i];
3167 r = amdgpu_ring_test_ring(ring);
3169 ring->ready = false;
3175 static void gfx_v7_0_cp_enable(struct amdgpu_device *adev, bool enable)
3177 gfx_v7_0_cp_gfx_enable(adev, enable);
3178 gfx_v7_0_cp_compute_enable(adev, enable);
3181 static int gfx_v7_0_cp_load_microcode(struct amdgpu_device *adev)
3185 r = gfx_v7_0_cp_gfx_load_microcode(adev);
3188 r = gfx_v7_0_cp_compute_load_microcode(adev);
3195 static void gfx_v7_0_enable_gui_idle_interrupt(struct amdgpu_device *adev,
3198 u32 tmp = RREG32(mmCP_INT_CNTL_RING0);
3201 tmp |= (CP_INT_CNTL_RING0__CNTX_BUSY_INT_ENABLE_MASK |
3202 CP_INT_CNTL_RING0__CNTX_EMPTY_INT_ENABLE_MASK);
3204 tmp &= ~(CP_INT_CNTL_RING0__CNTX_BUSY_INT_ENABLE_MASK |
3205 CP_INT_CNTL_RING0__CNTX_EMPTY_INT_ENABLE_MASK);
3206 WREG32(mmCP_INT_CNTL_RING0, tmp);
3209 static int gfx_v7_0_cp_resume(struct amdgpu_device *adev)
3213 gfx_v7_0_enable_gui_idle_interrupt(adev, false);
3215 r = gfx_v7_0_cp_load_microcode(adev);
3219 r = gfx_v7_0_cp_gfx_resume(adev);
3222 r = gfx_v7_0_cp_compute_resume(adev);
3226 gfx_v7_0_enable_gui_idle_interrupt(adev, true);
3232 * gfx_v7_0_ring_emit_vm_flush - cik vm flush using the CP
3234 * @ring: the ring to emmit the commands to
3236 * Sync the command pipeline with the PFP. E.g. wait for everything
3239 static void gfx_v7_0_ring_emit_pipeline_sync(struct amdgpu_ring *ring)
3241 int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
3242 uint32_t seq = ring->fence_drv.sync_seq;
3243 uint64_t addr = ring->fence_drv.gpu_addr;
3245 amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
3246 amdgpu_ring_write(ring, (WAIT_REG_MEM_MEM_SPACE(1) | /* memory */
3247 WAIT_REG_MEM_FUNCTION(3) | /* equal */
3248 WAIT_REG_MEM_ENGINE(usepfp))); /* pfp or me */
3249 amdgpu_ring_write(ring, addr & 0xfffffffc);
3250 amdgpu_ring_write(ring, upper_32_bits(addr) & 0xffffffff);
3251 amdgpu_ring_write(ring, seq);
3252 amdgpu_ring_write(ring, 0xffffffff);
3253 amdgpu_ring_write(ring, 4); /* poll interval */
3256 /* synce CE with ME to prevent CE fetch CEIB before context switch done */
3257 amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
3258 amdgpu_ring_write(ring, 0);
3259 amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
3260 amdgpu_ring_write(ring, 0);
3266 * VMID 0 is the physical GPU addresses as used by the kernel.
3267 * VMIDs 1-15 are used for userspace clients and are handled
3268 * by the amdgpu vm/hsa code.
3271 * gfx_v7_0_ring_emit_vm_flush - cik vm flush using the CP
3273 * @adev: amdgpu_device pointer
3275 * Update the page table base and flush the VM TLB
3276 * using the CP (CIK).
3278 static void gfx_v7_0_ring_emit_vm_flush(struct amdgpu_ring *ring,
3279 unsigned vm_id, uint64_t pd_addr)
3281 int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
3283 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
3284 amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(usepfp) |
3285 WRITE_DATA_DST_SEL(0)));
3287 amdgpu_ring_write(ring,
3288 (mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR + vm_id));
3290 amdgpu_ring_write(ring,
3291 (mmVM_CONTEXT8_PAGE_TABLE_BASE_ADDR + vm_id - 8));
3293 amdgpu_ring_write(ring, 0);
3294 amdgpu_ring_write(ring, pd_addr >> 12);
3296 /* bits 0-15 are the VM contexts0-15 */
3297 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
3298 amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
3299 WRITE_DATA_DST_SEL(0)));
3300 amdgpu_ring_write(ring, mmVM_INVALIDATE_REQUEST);
3301 amdgpu_ring_write(ring, 0);
3302 amdgpu_ring_write(ring, 1 << vm_id);
3304 /* wait for the invalidate to complete */
3305 amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
3306 amdgpu_ring_write(ring, (WAIT_REG_MEM_OPERATION(0) | /* wait */
3307 WAIT_REG_MEM_FUNCTION(0) | /* always */
3308 WAIT_REG_MEM_ENGINE(0))); /* me */
3309 amdgpu_ring_write(ring, mmVM_INVALIDATE_REQUEST);
3310 amdgpu_ring_write(ring, 0);
3311 amdgpu_ring_write(ring, 0); /* ref */
3312 amdgpu_ring_write(ring, 0); /* mask */
3313 amdgpu_ring_write(ring, 0x20); /* poll interval */
3315 /* compute doesn't have PFP */
3317 /* sync PFP to ME, otherwise we might get invalid PFP reads */
3318 amdgpu_ring_write(ring, PACKET3(PACKET3_PFP_SYNC_ME, 0));
3319 amdgpu_ring_write(ring, 0x0);
3321 /* synce CE with ME to prevent CE fetch CEIB before context switch done */
3322 amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
3323 amdgpu_ring_write(ring, 0);
3324 amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
3325 amdgpu_ring_write(ring, 0);
3331 * The RLC is a multi-purpose microengine that handles a
3332 * variety of functions.
3334 static void gfx_v7_0_rlc_fini(struct amdgpu_device *adev)
3338 /* save restore block */
3339 if (adev->gfx.rlc.save_restore_obj) {
3340 r = amdgpu_bo_reserve(adev->gfx.rlc.save_restore_obj, false);
3341 if (unlikely(r != 0))
3342 dev_warn(adev->dev, "(%d) reserve RLC sr bo failed\n", r);
3343 amdgpu_bo_unpin(adev->gfx.rlc.save_restore_obj);
3344 amdgpu_bo_unreserve(adev->gfx.rlc.save_restore_obj);
3346 amdgpu_bo_unref(&adev->gfx.rlc.save_restore_obj);
3347 adev->gfx.rlc.save_restore_obj = NULL;
3350 /* clear state block */
3351 if (adev->gfx.rlc.clear_state_obj) {
3352 r = amdgpu_bo_reserve(adev->gfx.rlc.clear_state_obj, false);
3353 if (unlikely(r != 0))
3354 dev_warn(adev->dev, "(%d) reserve RLC c bo failed\n", r);
3355 amdgpu_bo_unpin(adev->gfx.rlc.clear_state_obj);
3356 amdgpu_bo_unreserve(adev->gfx.rlc.clear_state_obj);
3358 amdgpu_bo_unref(&adev->gfx.rlc.clear_state_obj);
3359 adev->gfx.rlc.clear_state_obj = NULL;
3362 /* clear state block */
3363 if (adev->gfx.rlc.cp_table_obj) {
3364 r = amdgpu_bo_reserve(adev->gfx.rlc.cp_table_obj, false);
3365 if (unlikely(r != 0))
3366 dev_warn(adev->dev, "(%d) reserve RLC cp table bo failed\n", r);
3367 amdgpu_bo_unpin(adev->gfx.rlc.cp_table_obj);
3368 amdgpu_bo_unreserve(adev->gfx.rlc.cp_table_obj);
3370 amdgpu_bo_unref(&adev->gfx.rlc.cp_table_obj);
3371 adev->gfx.rlc.cp_table_obj = NULL;
3375 static int gfx_v7_0_rlc_init(struct amdgpu_device *adev)
3378 volatile u32 *dst_ptr;
3380 const struct cs_section_def *cs_data;
3383 /* allocate rlc buffers */
3384 if (adev->flags & AMD_IS_APU) {
3385 if (adev->asic_type == CHIP_KAVERI) {
3386 adev->gfx.rlc.reg_list = spectre_rlc_save_restore_register_list;
3387 adev->gfx.rlc.reg_list_size =
3388 (u32)ARRAY_SIZE(spectre_rlc_save_restore_register_list);
3390 adev->gfx.rlc.reg_list = kalindi_rlc_save_restore_register_list;
3391 adev->gfx.rlc.reg_list_size =
3392 (u32)ARRAY_SIZE(kalindi_rlc_save_restore_register_list);
3395 adev->gfx.rlc.cs_data = ci_cs_data;
3396 adev->gfx.rlc.cp_table_size = ALIGN(CP_ME_TABLE_SIZE * 5 * 4, 2048); /* CP JT */
3397 adev->gfx.rlc.cp_table_size += 64 * 1024; /* GDS */
3399 src_ptr = adev->gfx.rlc.reg_list;
3400 dws = adev->gfx.rlc.reg_list_size;
3401 dws += (5 * 16) + 48 + 48 + 64;
3403 cs_data = adev->gfx.rlc.cs_data;
3406 /* save restore block */
3407 if (adev->gfx.rlc.save_restore_obj == NULL) {
3408 r = amdgpu_bo_create(adev, dws * 4, PAGE_SIZE, true,
3409 AMDGPU_GEM_DOMAIN_VRAM,
3410 AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
3411 AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS,
3413 &adev->gfx.rlc.save_restore_obj);
3415 dev_warn(adev->dev, "(%d) create RLC sr bo failed\n", r);
3420 r = amdgpu_bo_reserve(adev->gfx.rlc.save_restore_obj, false);
3421 if (unlikely(r != 0)) {
3422 gfx_v7_0_rlc_fini(adev);
3425 r = amdgpu_bo_pin(adev->gfx.rlc.save_restore_obj, AMDGPU_GEM_DOMAIN_VRAM,
3426 &adev->gfx.rlc.save_restore_gpu_addr);
3428 amdgpu_bo_unreserve(adev->gfx.rlc.save_restore_obj);
3429 dev_warn(adev->dev, "(%d) pin RLC sr bo failed\n", r);
3430 gfx_v7_0_rlc_fini(adev);
3434 r = amdgpu_bo_kmap(adev->gfx.rlc.save_restore_obj, (void **)&adev->gfx.rlc.sr_ptr);
3436 dev_warn(adev->dev, "(%d) map RLC sr bo failed\n", r);
3437 gfx_v7_0_rlc_fini(adev);
3440 /* write the sr buffer */
3441 dst_ptr = adev->gfx.rlc.sr_ptr;
3442 for (i = 0; i < adev->gfx.rlc.reg_list_size; i++)
3443 dst_ptr[i] = cpu_to_le32(src_ptr[i]);
3444 amdgpu_bo_kunmap(adev->gfx.rlc.save_restore_obj);
3445 amdgpu_bo_unreserve(adev->gfx.rlc.save_restore_obj);
3449 /* clear state block */
3450 adev->gfx.rlc.clear_state_size = dws = gfx_v7_0_get_csb_size(adev);
3452 if (adev->gfx.rlc.clear_state_obj == NULL) {
3453 r = amdgpu_bo_create(adev, dws * 4, PAGE_SIZE, true,
3454 AMDGPU_GEM_DOMAIN_VRAM,
3455 AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
3456 AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS,
3458 &adev->gfx.rlc.clear_state_obj);
3460 dev_warn(adev->dev, "(%d) create RLC c bo failed\n", r);
3461 gfx_v7_0_rlc_fini(adev);
3465 r = amdgpu_bo_reserve(adev->gfx.rlc.clear_state_obj, false);
3466 if (unlikely(r != 0)) {
3467 gfx_v7_0_rlc_fini(adev);
3470 r = amdgpu_bo_pin(adev->gfx.rlc.clear_state_obj, AMDGPU_GEM_DOMAIN_VRAM,
3471 &adev->gfx.rlc.clear_state_gpu_addr);
3473 amdgpu_bo_unreserve(adev->gfx.rlc.clear_state_obj);
3474 dev_warn(adev->dev, "(%d) pin RLC c bo failed\n", r);
3475 gfx_v7_0_rlc_fini(adev);
3479 r = amdgpu_bo_kmap(adev->gfx.rlc.clear_state_obj, (void **)&adev->gfx.rlc.cs_ptr);
3481 dev_warn(adev->dev, "(%d) map RLC c bo failed\n", r);
3482 gfx_v7_0_rlc_fini(adev);
3485 /* set up the cs buffer */
3486 dst_ptr = adev->gfx.rlc.cs_ptr;
3487 gfx_v7_0_get_csb_buffer(adev, dst_ptr);
3488 amdgpu_bo_kunmap(adev->gfx.rlc.clear_state_obj);
3489 amdgpu_bo_unreserve(adev->gfx.rlc.clear_state_obj);
3492 if (adev->gfx.rlc.cp_table_size) {
3493 if (adev->gfx.rlc.cp_table_obj == NULL) {
3494 r = amdgpu_bo_create(adev, adev->gfx.rlc.cp_table_size, PAGE_SIZE, true,
3495 AMDGPU_GEM_DOMAIN_VRAM,
3496 AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
3497 AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS,
3499 &adev->gfx.rlc.cp_table_obj);
3501 dev_warn(adev->dev, "(%d) create RLC cp table bo failed\n", r);
3502 gfx_v7_0_rlc_fini(adev);
3507 r = amdgpu_bo_reserve(adev->gfx.rlc.cp_table_obj, false);
3508 if (unlikely(r != 0)) {
3509 dev_warn(adev->dev, "(%d) reserve RLC cp table bo failed\n", r);
3510 gfx_v7_0_rlc_fini(adev);
3513 r = amdgpu_bo_pin(adev->gfx.rlc.cp_table_obj, AMDGPU_GEM_DOMAIN_VRAM,
3514 &adev->gfx.rlc.cp_table_gpu_addr);
3516 amdgpu_bo_unreserve(adev->gfx.rlc.cp_table_obj);
3517 dev_warn(adev->dev, "(%d) pin RLC cp_table bo failed\n", r);
3518 gfx_v7_0_rlc_fini(adev);
3521 r = amdgpu_bo_kmap(adev->gfx.rlc.cp_table_obj, (void **)&adev->gfx.rlc.cp_table_ptr);
3523 dev_warn(adev->dev, "(%d) map RLC cp table bo failed\n", r);
3524 gfx_v7_0_rlc_fini(adev);
3528 gfx_v7_0_init_cp_pg_table(adev);
3530 amdgpu_bo_kunmap(adev->gfx.rlc.cp_table_obj);
3531 amdgpu_bo_unreserve(adev->gfx.rlc.cp_table_obj);
3538 static void gfx_v7_0_enable_lbpw(struct amdgpu_device *adev, bool enable)
3542 tmp = RREG32(mmRLC_LB_CNTL);
3544 tmp |= RLC_LB_CNTL__LOAD_BALANCE_ENABLE_MASK;
3546 tmp &= ~RLC_LB_CNTL__LOAD_BALANCE_ENABLE_MASK;
3547 WREG32(mmRLC_LB_CNTL, tmp);
3550 static void gfx_v7_0_wait_for_rlc_serdes(struct amdgpu_device *adev)
3555 mutex_lock(&adev->grbm_idx_mutex);
3556 for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
3557 for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
3558 gfx_v7_0_select_se_sh(adev, i, j, 0xffffffff);
3559 for (k = 0; k < adev->usec_timeout; k++) {
3560 if (RREG32(mmRLC_SERDES_CU_MASTER_BUSY) == 0)
3566 gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3567 mutex_unlock(&adev->grbm_idx_mutex);
3569 mask = RLC_SERDES_NONCU_MASTER_BUSY__SE_MASTER_BUSY_MASK |
3570 RLC_SERDES_NONCU_MASTER_BUSY__GC_MASTER_BUSY_MASK |
3571 RLC_SERDES_NONCU_MASTER_BUSY__TC0_MASTER_BUSY_MASK |
3572 RLC_SERDES_NONCU_MASTER_BUSY__TC1_MASTER_BUSY_MASK;
3573 for (k = 0; k < adev->usec_timeout; k++) {
3574 if ((RREG32(mmRLC_SERDES_NONCU_MASTER_BUSY) & mask) == 0)
3580 static void gfx_v7_0_update_rlc(struct amdgpu_device *adev, u32 rlc)
3584 tmp = RREG32(mmRLC_CNTL);
3586 WREG32(mmRLC_CNTL, rlc);
3589 static u32 gfx_v7_0_halt_rlc(struct amdgpu_device *adev)
3593 orig = data = RREG32(mmRLC_CNTL);
3595 if (data & RLC_CNTL__RLC_ENABLE_F32_MASK) {
3598 data &= ~RLC_CNTL__RLC_ENABLE_F32_MASK;
3599 WREG32(mmRLC_CNTL, data);
3601 for (i = 0; i < adev->usec_timeout; i++) {
3602 if ((RREG32(mmRLC_GPM_STAT) & RLC_GPM_STAT__RLC_BUSY_MASK) == 0)
3607 gfx_v7_0_wait_for_rlc_serdes(adev);
3613 static void gfx_v7_0_enter_rlc_safe_mode(struct amdgpu_device *adev)
3617 tmp = 0x1 | (1 << 1);
3618 WREG32(mmRLC_GPR_REG2, tmp);
3620 mask = RLC_GPM_STAT__GFX_POWER_STATUS_MASK |
3621 RLC_GPM_STAT__GFX_CLOCK_STATUS_MASK;
3622 for (i = 0; i < adev->usec_timeout; i++) {
3623 if ((RREG32(mmRLC_GPM_STAT) & mask) == mask)
3628 for (i = 0; i < adev->usec_timeout; i++) {
3629 if ((RREG32(mmRLC_GPR_REG2) & 0x1) == 0)
3635 static void gfx_v7_0_exit_rlc_safe_mode(struct amdgpu_device *adev)
3639 tmp = 0x1 | (0 << 1);
3640 WREG32(mmRLC_GPR_REG2, tmp);
3644 * gfx_v7_0_rlc_stop - stop the RLC ME
3646 * @adev: amdgpu_device pointer
3648 * Halt the RLC ME (MicroEngine) (CIK).
3650 static void gfx_v7_0_rlc_stop(struct amdgpu_device *adev)
3652 WREG32(mmRLC_CNTL, 0);
3654 gfx_v7_0_enable_gui_idle_interrupt(adev, false);
3656 gfx_v7_0_wait_for_rlc_serdes(adev);
3660 * gfx_v7_0_rlc_start - start the RLC ME
3662 * @adev: amdgpu_device pointer
3664 * Unhalt the RLC ME (MicroEngine) (CIK).
3666 static void gfx_v7_0_rlc_start(struct amdgpu_device *adev)
3668 WREG32(mmRLC_CNTL, RLC_CNTL__RLC_ENABLE_F32_MASK);
3670 gfx_v7_0_enable_gui_idle_interrupt(adev, true);
3675 static void gfx_v7_0_rlc_reset(struct amdgpu_device *adev)
3677 u32 tmp = RREG32(mmGRBM_SOFT_RESET);
3679 tmp |= GRBM_SOFT_RESET__SOFT_RESET_RLC_MASK;
3680 WREG32(mmGRBM_SOFT_RESET, tmp);
3682 tmp &= ~GRBM_SOFT_RESET__SOFT_RESET_RLC_MASK;
3683 WREG32(mmGRBM_SOFT_RESET, tmp);
3688 * gfx_v7_0_rlc_resume - setup the RLC hw
3690 * @adev: amdgpu_device pointer
3692 * Initialize the RLC registers, load the ucode,
3693 * and start the RLC (CIK).
3694 * Returns 0 for success, -EINVAL if the ucode is not available.
3696 static int gfx_v7_0_rlc_resume(struct amdgpu_device *adev)
3698 const struct rlc_firmware_header_v1_0 *hdr;
3699 const __le32 *fw_data;
3700 unsigned i, fw_size;
3703 if (!adev->gfx.rlc_fw)
3706 hdr = (const struct rlc_firmware_header_v1_0 *)adev->gfx.rlc_fw->data;
3707 amdgpu_ucode_print_rlc_hdr(&hdr->header);
3708 adev->gfx.rlc_fw_version = le32_to_cpu(hdr->header.ucode_version);
3709 adev->gfx.rlc_feature_version = le32_to_cpu(
3710 hdr->ucode_feature_version);
3712 gfx_v7_0_rlc_stop(adev);
3715 tmp = RREG32(mmRLC_CGCG_CGLS_CTRL) & 0xfffffffc;
3716 WREG32(mmRLC_CGCG_CGLS_CTRL, tmp);
3718 gfx_v7_0_rlc_reset(adev);
3720 gfx_v7_0_init_pg(adev);
3722 WREG32(mmRLC_LB_CNTR_INIT, 0);
3723 WREG32(mmRLC_LB_CNTR_MAX, 0x00008000);
3725 mutex_lock(&adev->grbm_idx_mutex);
3726 gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3727 WREG32(mmRLC_LB_INIT_CU_MASK, 0xffffffff);
3728 WREG32(mmRLC_LB_PARAMS, 0x00600408);
3729 WREG32(mmRLC_LB_CNTL, 0x80000004);
3730 mutex_unlock(&adev->grbm_idx_mutex);
3732 WREG32(mmRLC_MC_CNTL, 0);
3733 WREG32(mmRLC_UCODE_CNTL, 0);
3735 fw_data = (const __le32 *)
3736 (adev->gfx.rlc_fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes));
3737 fw_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
3738 WREG32(mmRLC_GPM_UCODE_ADDR, 0);
3739 for (i = 0; i < fw_size; i++)
3740 WREG32(mmRLC_GPM_UCODE_DATA, le32_to_cpup(fw_data++));
3741 WREG32(mmRLC_GPM_UCODE_ADDR, adev->gfx.rlc_fw_version);
3743 /* XXX - find out what chips support lbpw */
3744 gfx_v7_0_enable_lbpw(adev, false);
3746 if (adev->asic_type == CHIP_BONAIRE)
3747 WREG32(mmRLC_DRIVER_CPDMA_STATUS, 0);
3749 gfx_v7_0_rlc_start(adev);
3754 static void gfx_v7_0_enable_cgcg(struct amdgpu_device *adev, bool enable)
3756 u32 data, orig, tmp, tmp2;
3758 orig = data = RREG32(mmRLC_CGCG_CGLS_CTRL);
3760 if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG)) {
3761 gfx_v7_0_enable_gui_idle_interrupt(adev, true);
3763 tmp = gfx_v7_0_halt_rlc(adev);
3765 mutex_lock(&adev->grbm_idx_mutex);
3766 gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3767 WREG32(mmRLC_SERDES_WR_CU_MASTER_MASK, 0xffffffff);
3768 WREG32(mmRLC_SERDES_WR_NONCU_MASTER_MASK, 0xffffffff);
3769 tmp2 = RLC_SERDES_WR_CTRL__BPM_ADDR_MASK |
3770 RLC_SERDES_WR_CTRL__CGCG_OVERRIDE_0_MASK |
3771 RLC_SERDES_WR_CTRL__CGLS_ENABLE_MASK;
3772 WREG32(mmRLC_SERDES_WR_CTRL, tmp2);
3773 mutex_unlock(&adev->grbm_idx_mutex);
3775 gfx_v7_0_update_rlc(adev, tmp);
3777 data |= RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK | RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK;
3779 gfx_v7_0_enable_gui_idle_interrupt(adev, false);
3781 RREG32(mmCB_CGTT_SCLK_CTRL);
3782 RREG32(mmCB_CGTT_SCLK_CTRL);
3783 RREG32(mmCB_CGTT_SCLK_CTRL);
3784 RREG32(mmCB_CGTT_SCLK_CTRL);
3786 data &= ~(RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK | RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK);
3790 WREG32(mmRLC_CGCG_CGLS_CTRL, data);
3794 static void gfx_v7_0_enable_mgcg(struct amdgpu_device *adev, bool enable)
3796 u32 data, orig, tmp = 0;
3798 if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG)) {
3799 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) {
3800 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CP_LS) {
3801 orig = data = RREG32(mmCP_MEM_SLP_CNTL);
3802 data |= CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK;
3804 WREG32(mmCP_MEM_SLP_CNTL, data);
3808 orig = data = RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
3812 WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data);
3814 tmp = gfx_v7_0_halt_rlc(adev);
3816 mutex_lock(&adev->grbm_idx_mutex);
3817 gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3818 WREG32(mmRLC_SERDES_WR_CU_MASTER_MASK, 0xffffffff);
3819 WREG32(mmRLC_SERDES_WR_NONCU_MASTER_MASK, 0xffffffff);
3820 data = RLC_SERDES_WR_CTRL__BPM_ADDR_MASK |
3821 RLC_SERDES_WR_CTRL__MGCG_OVERRIDE_0_MASK;
3822 WREG32(mmRLC_SERDES_WR_CTRL, data);
3823 mutex_unlock(&adev->grbm_idx_mutex);
3825 gfx_v7_0_update_rlc(adev, tmp);
3827 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGTS) {
3828 orig = data = RREG32(mmCGTS_SM_CTRL_REG);
3829 data &= ~CGTS_SM_CTRL_REG__SM_MODE_MASK;
3830 data |= (0x2 << CGTS_SM_CTRL_REG__SM_MODE__SHIFT);
3831 data |= CGTS_SM_CTRL_REG__SM_MODE_ENABLE_MASK;
3832 data &= ~CGTS_SM_CTRL_REG__OVERRIDE_MASK;
3833 if ((adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) &&
3834 (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGTS_LS))
3835 data &= ~CGTS_SM_CTRL_REG__LS_OVERRIDE_MASK;
3836 data &= ~CGTS_SM_CTRL_REG__ON_MONITOR_ADD_MASK;
3837 data |= CGTS_SM_CTRL_REG__ON_MONITOR_ADD_EN_MASK;
3838 data |= (0x96 << CGTS_SM_CTRL_REG__ON_MONITOR_ADD__SHIFT);
3840 WREG32(mmCGTS_SM_CTRL_REG, data);
3843 orig = data = RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
3846 WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data);
3848 data = RREG32(mmRLC_MEM_SLP_CNTL);
3849 if (data & RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK) {
3850 data &= ~RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK;
3851 WREG32(mmRLC_MEM_SLP_CNTL, data);
3854 data = RREG32(mmCP_MEM_SLP_CNTL);
3855 if (data & CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK) {
3856 data &= ~CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK;
3857 WREG32(mmCP_MEM_SLP_CNTL, data);
3860 orig = data = RREG32(mmCGTS_SM_CTRL_REG);
3861 data |= CGTS_SM_CTRL_REG__OVERRIDE_MASK | CGTS_SM_CTRL_REG__LS_OVERRIDE_MASK;
3863 WREG32(mmCGTS_SM_CTRL_REG, data);
3865 tmp = gfx_v7_0_halt_rlc(adev);
3867 mutex_lock(&adev->grbm_idx_mutex);
3868 gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3869 WREG32(mmRLC_SERDES_WR_CU_MASTER_MASK, 0xffffffff);
3870 WREG32(mmRLC_SERDES_WR_NONCU_MASTER_MASK, 0xffffffff);
3871 data = RLC_SERDES_WR_CTRL__BPM_ADDR_MASK | RLC_SERDES_WR_CTRL__MGCG_OVERRIDE_1_MASK;
3872 WREG32(mmRLC_SERDES_WR_CTRL, data);
3873 mutex_unlock(&adev->grbm_idx_mutex);
3875 gfx_v7_0_update_rlc(adev, tmp);
3879 static void gfx_v7_0_update_cg(struct amdgpu_device *adev,
3882 gfx_v7_0_enable_gui_idle_interrupt(adev, false);
3883 /* order matters! */
3885 gfx_v7_0_enable_mgcg(adev, true);
3886 gfx_v7_0_enable_cgcg(adev, true);
3888 gfx_v7_0_enable_cgcg(adev, false);
3889 gfx_v7_0_enable_mgcg(adev, false);
3891 gfx_v7_0_enable_gui_idle_interrupt(adev, true);
3894 static void gfx_v7_0_enable_sclk_slowdown_on_pu(struct amdgpu_device *adev,
3899 orig = data = RREG32(mmRLC_PG_CNTL);
3900 if (enable && (adev->pg_flags & AMD_PG_SUPPORT_RLC_SMU_HS))
3901 data |= RLC_PG_CNTL__SMU_CLK_SLOWDOWN_ON_PU_ENABLE_MASK;
3903 data &= ~RLC_PG_CNTL__SMU_CLK_SLOWDOWN_ON_PU_ENABLE_MASK;
3905 WREG32(mmRLC_PG_CNTL, data);
3908 static void gfx_v7_0_enable_sclk_slowdown_on_pd(struct amdgpu_device *adev,
3913 orig = data = RREG32(mmRLC_PG_CNTL);
3914 if (enable && (adev->pg_flags & AMD_PG_SUPPORT_RLC_SMU_HS))
3915 data |= RLC_PG_CNTL__SMU_CLK_SLOWDOWN_ON_PD_ENABLE_MASK;
3917 data &= ~RLC_PG_CNTL__SMU_CLK_SLOWDOWN_ON_PD_ENABLE_MASK;
3919 WREG32(mmRLC_PG_CNTL, data);
3922 static void gfx_v7_0_enable_cp_pg(struct amdgpu_device *adev, bool enable)
3926 orig = data = RREG32(mmRLC_PG_CNTL);
3927 if (enable && (adev->pg_flags & AMD_PG_SUPPORT_CP))
3932 WREG32(mmRLC_PG_CNTL, data);
3935 static void gfx_v7_0_enable_gds_pg(struct amdgpu_device *adev, bool enable)
3939 orig = data = RREG32(mmRLC_PG_CNTL);
3940 if (enable && (adev->pg_flags & AMD_PG_SUPPORT_GDS))
3945 WREG32(mmRLC_PG_CNTL, data);
3948 static void gfx_v7_0_init_cp_pg_table(struct amdgpu_device *adev)
3950 const __le32 *fw_data;
3951 volatile u32 *dst_ptr;
3952 int me, i, max_me = 4;
3954 u32 table_offset, table_size;
3956 if (adev->asic_type == CHIP_KAVERI)
3959 if (adev->gfx.rlc.cp_table_ptr == NULL)
3962 /* write the cp table buffer */
3963 dst_ptr = adev->gfx.rlc.cp_table_ptr;
3964 for (me = 0; me < max_me; me++) {
3966 const struct gfx_firmware_header_v1_0 *hdr =
3967 (const struct gfx_firmware_header_v1_0 *)adev->gfx.ce_fw->data;
3968 fw_data = (const __le32 *)
3969 (adev->gfx.ce_fw->data +
3970 le32_to_cpu(hdr->header.ucode_array_offset_bytes));
3971 table_offset = le32_to_cpu(hdr->jt_offset);
3972 table_size = le32_to_cpu(hdr->jt_size);
3973 } else if (me == 1) {
3974 const struct gfx_firmware_header_v1_0 *hdr =
3975 (const struct gfx_firmware_header_v1_0 *)adev->gfx.pfp_fw->data;
3976 fw_data = (const __le32 *)
3977 (adev->gfx.pfp_fw->data +
3978 le32_to_cpu(hdr->header.ucode_array_offset_bytes));
3979 table_offset = le32_to_cpu(hdr->jt_offset);
3980 table_size = le32_to_cpu(hdr->jt_size);
3981 } else if (me == 2) {
3982 const struct gfx_firmware_header_v1_0 *hdr =
3983 (const struct gfx_firmware_header_v1_0 *)adev->gfx.me_fw->data;
3984 fw_data = (const __le32 *)
3985 (adev->gfx.me_fw->data +
3986 le32_to_cpu(hdr->header.ucode_array_offset_bytes));
3987 table_offset = le32_to_cpu(hdr->jt_offset);
3988 table_size = le32_to_cpu(hdr->jt_size);
3989 } else if (me == 3) {
3990 const struct gfx_firmware_header_v1_0 *hdr =
3991 (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data;
3992 fw_data = (const __le32 *)
3993 (adev->gfx.mec_fw->data +
3994 le32_to_cpu(hdr->header.ucode_array_offset_bytes));
3995 table_offset = le32_to_cpu(hdr->jt_offset);
3996 table_size = le32_to_cpu(hdr->jt_size);
3998 const struct gfx_firmware_header_v1_0 *hdr =
3999 (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec2_fw->data;
4000 fw_data = (const __le32 *)
4001 (adev->gfx.mec2_fw->data +
4002 le32_to_cpu(hdr->header.ucode_array_offset_bytes));
4003 table_offset = le32_to_cpu(hdr->jt_offset);
4004 table_size = le32_to_cpu(hdr->jt_size);
4007 for (i = 0; i < table_size; i ++) {
4008 dst_ptr[bo_offset + i] =
4009 cpu_to_le32(le32_to_cpu(fw_data[table_offset + i]));
4012 bo_offset += table_size;
4016 static void gfx_v7_0_enable_gfx_cgpg(struct amdgpu_device *adev,
4021 if (enable && (adev->pg_flags & AMD_PG_SUPPORT_GFX_PG)) {
4022 orig = data = RREG32(mmRLC_PG_CNTL);
4023 data |= RLC_PG_CNTL__GFX_POWER_GATING_ENABLE_MASK;
4025 WREG32(mmRLC_PG_CNTL, data);
4027 orig = data = RREG32(mmRLC_AUTO_PG_CTRL);
4028 data |= RLC_AUTO_PG_CTRL__AUTO_PG_EN_MASK;
4030 WREG32(mmRLC_AUTO_PG_CTRL, data);
4032 orig = data = RREG32(mmRLC_PG_CNTL);
4033 data &= ~RLC_PG_CNTL__GFX_POWER_GATING_ENABLE_MASK;
4035 WREG32(mmRLC_PG_CNTL, data);
4037 orig = data = RREG32(mmRLC_AUTO_PG_CTRL);
4038 data &= ~RLC_AUTO_PG_CTRL__AUTO_PG_EN_MASK;
4040 WREG32(mmRLC_AUTO_PG_CTRL, data);
4042 data = RREG32(mmDB_RENDER_CONTROL);
4046 static void gfx_v7_0_set_user_cu_inactive_bitmap(struct amdgpu_device *adev,
4054 data = bitmap << GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS__SHIFT;
4055 data &= GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS_MASK;
4057 WREG32(mmGC_USER_SHADER_ARRAY_CONFIG, data);
4060 static u32 gfx_v7_0_get_cu_active_bitmap(struct amdgpu_device *adev)
4064 data = RREG32(mmCC_GC_SHADER_ARRAY_CONFIG);
4065 data |= RREG32(mmGC_USER_SHADER_ARRAY_CONFIG);
4067 data &= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_CUS_MASK;
4068 data >>= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_CUS__SHIFT;
4070 mask = gfx_v7_0_create_bitmask(adev->gfx.config.max_cu_per_sh);
4072 return (~data) & mask;
4075 static void gfx_v7_0_init_ao_cu_mask(struct amdgpu_device *adev)
4079 WREG32(mmRLC_PG_ALWAYS_ON_CU_MASK, adev->gfx.cu_info.ao_cu_mask);
4081 tmp = RREG32(mmRLC_MAX_PG_CU);
4082 tmp &= ~RLC_MAX_PG_CU__MAX_POWERED_UP_CU_MASK;
4083 tmp |= (adev->gfx.cu_info.number << RLC_MAX_PG_CU__MAX_POWERED_UP_CU__SHIFT);
4084 WREG32(mmRLC_MAX_PG_CU, tmp);
4087 static void gfx_v7_0_enable_gfx_static_mgpg(struct amdgpu_device *adev,
4092 orig = data = RREG32(mmRLC_PG_CNTL);
4093 if (enable && (adev->pg_flags & AMD_PG_SUPPORT_GFX_SMG))
4094 data |= RLC_PG_CNTL__STATIC_PER_CU_PG_ENABLE_MASK;
4096 data &= ~RLC_PG_CNTL__STATIC_PER_CU_PG_ENABLE_MASK;
4098 WREG32(mmRLC_PG_CNTL, data);
4101 static void gfx_v7_0_enable_gfx_dynamic_mgpg(struct amdgpu_device *adev,
4106 orig = data = RREG32(mmRLC_PG_CNTL);
4107 if (enable && (adev->pg_flags & AMD_PG_SUPPORT_GFX_DMG))
4108 data |= RLC_PG_CNTL__DYN_PER_CU_PG_ENABLE_MASK;
4110 data &= ~RLC_PG_CNTL__DYN_PER_CU_PG_ENABLE_MASK;
4112 WREG32(mmRLC_PG_CNTL, data);
4115 #define RLC_SAVE_AND_RESTORE_STARTING_OFFSET 0x90
4116 #define RLC_CLEAR_STATE_DESCRIPTOR_OFFSET 0x3D
4118 static void gfx_v7_0_init_gfx_cgpg(struct amdgpu_device *adev)
4123 if (adev->gfx.rlc.cs_data) {
4124 WREG32(mmRLC_GPM_SCRATCH_ADDR, RLC_CLEAR_STATE_DESCRIPTOR_OFFSET);
4125 WREG32(mmRLC_GPM_SCRATCH_DATA, upper_32_bits(adev->gfx.rlc.clear_state_gpu_addr));
4126 WREG32(mmRLC_GPM_SCRATCH_DATA, lower_32_bits(adev->gfx.rlc.clear_state_gpu_addr));
4127 WREG32(mmRLC_GPM_SCRATCH_DATA, adev->gfx.rlc.clear_state_size);
4129 WREG32(mmRLC_GPM_SCRATCH_ADDR, RLC_CLEAR_STATE_DESCRIPTOR_OFFSET);
4130 for (i = 0; i < 3; i++)
4131 WREG32(mmRLC_GPM_SCRATCH_DATA, 0);
4133 if (adev->gfx.rlc.reg_list) {
4134 WREG32(mmRLC_GPM_SCRATCH_ADDR, RLC_SAVE_AND_RESTORE_STARTING_OFFSET);
4135 for (i = 0; i < adev->gfx.rlc.reg_list_size; i++)
4136 WREG32(mmRLC_GPM_SCRATCH_DATA, adev->gfx.rlc.reg_list[i]);
4139 orig = data = RREG32(mmRLC_PG_CNTL);
4140 data |= RLC_PG_CNTL__GFX_POWER_GATING_SRC_MASK;
4142 WREG32(mmRLC_PG_CNTL, data);
4144 WREG32(mmRLC_SAVE_AND_RESTORE_BASE, adev->gfx.rlc.save_restore_gpu_addr >> 8);
4145 WREG32(mmRLC_JUMP_TABLE_RESTORE, adev->gfx.rlc.cp_table_gpu_addr >> 8);
4147 data = RREG32(mmCP_RB_WPTR_POLL_CNTL);
4148 data &= ~CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT_MASK;
4149 data |= (0x60 << CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT__SHIFT);
4150 WREG32(mmCP_RB_WPTR_POLL_CNTL, data);
4153 WREG32(mmRLC_PG_DELAY, data);
4155 data = RREG32(mmRLC_PG_DELAY_2);
4158 WREG32(mmRLC_PG_DELAY_2, data);
4160 data = RREG32(mmRLC_AUTO_PG_CTRL);
4161 data &= ~RLC_AUTO_PG_CTRL__GRBM_REG_SAVE_GFX_IDLE_THRESHOLD_MASK;
4162 data |= (0x700 << RLC_AUTO_PG_CTRL__GRBM_REG_SAVE_GFX_IDLE_THRESHOLD__SHIFT);
4163 WREG32(mmRLC_AUTO_PG_CTRL, data);
4167 static void gfx_v7_0_update_gfx_pg(struct amdgpu_device *adev, bool enable)
4169 gfx_v7_0_enable_gfx_cgpg(adev, enable);
4170 gfx_v7_0_enable_gfx_static_mgpg(adev, enable);
4171 gfx_v7_0_enable_gfx_dynamic_mgpg(adev, enable);
4174 static u32 gfx_v7_0_get_csb_size(struct amdgpu_device *adev)
4177 const struct cs_section_def *sect = NULL;
4178 const struct cs_extent_def *ext = NULL;
4180 if (adev->gfx.rlc.cs_data == NULL)
4183 /* begin clear state */
4185 /* context control state */
4188 for (sect = adev->gfx.rlc.cs_data; sect->section != NULL; ++sect) {
4189 for (ext = sect->section; ext->extent != NULL; ++ext) {
4190 if (sect->id == SECT_CONTEXT)
4191 count += 2 + ext->reg_count;
4196 /* pa_sc_raster_config/pa_sc_raster_config1 */
4198 /* end clear state */
4206 static void gfx_v7_0_get_csb_buffer(struct amdgpu_device *adev,
4207 volatile u32 *buffer)
4210 const struct cs_section_def *sect = NULL;
4211 const struct cs_extent_def *ext = NULL;
4213 if (adev->gfx.rlc.cs_data == NULL)
4218 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0));
4219 buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
4221 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CONTEXT_CONTROL, 1));
4222 buffer[count++] = cpu_to_le32(0x80000000);
4223 buffer[count++] = cpu_to_le32(0x80000000);
4225 for (sect = adev->gfx.rlc.cs_data; sect->section != NULL; ++sect) {
4226 for (ext = sect->section; ext->extent != NULL; ++ext) {
4227 if (sect->id == SECT_CONTEXT) {
4229 cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, ext->reg_count));
4230 buffer[count++] = cpu_to_le32(ext->reg_index - PACKET3_SET_CONTEXT_REG_START);
4231 for (i = 0; i < ext->reg_count; i++)
4232 buffer[count++] = cpu_to_le32(ext->extent[i]);
4239 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, 2));
4240 buffer[count++] = cpu_to_le32(mmPA_SC_RASTER_CONFIG - PACKET3_SET_CONTEXT_REG_START);
4241 switch (adev->asic_type) {
4243 buffer[count++] = cpu_to_le32(0x16000012);
4244 buffer[count++] = cpu_to_le32(0x00000000);
4247 buffer[count++] = cpu_to_le32(0x00000000); /* XXX */
4248 buffer[count++] = cpu_to_le32(0x00000000);
4252 buffer[count++] = cpu_to_le32(0x00000000); /* XXX */
4253 buffer[count++] = cpu_to_le32(0x00000000);
4256 buffer[count++] = cpu_to_le32(0x3a00161a);
4257 buffer[count++] = cpu_to_le32(0x0000002e);
4260 buffer[count++] = cpu_to_le32(0x00000000);
4261 buffer[count++] = cpu_to_le32(0x00000000);
4265 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0));
4266 buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_END_CLEAR_STATE);
4268 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CLEAR_STATE, 0));
4269 buffer[count++] = cpu_to_le32(0);
4272 static void gfx_v7_0_init_pg(struct amdgpu_device *adev)
4274 if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_PG |
4275 AMD_PG_SUPPORT_GFX_SMG |
4276 AMD_PG_SUPPORT_GFX_DMG |
4278 AMD_PG_SUPPORT_GDS |
4279 AMD_PG_SUPPORT_RLC_SMU_HS)) {
4280 gfx_v7_0_enable_sclk_slowdown_on_pu(adev, true);
4281 gfx_v7_0_enable_sclk_slowdown_on_pd(adev, true);
4282 if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PG) {
4283 gfx_v7_0_init_gfx_cgpg(adev);
4284 gfx_v7_0_enable_cp_pg(adev, true);
4285 gfx_v7_0_enable_gds_pg(adev, true);
4287 gfx_v7_0_init_ao_cu_mask(adev);
4288 gfx_v7_0_update_gfx_pg(adev, true);
4292 static void gfx_v7_0_fini_pg(struct amdgpu_device *adev)
4294 if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_PG |
4295 AMD_PG_SUPPORT_GFX_SMG |
4296 AMD_PG_SUPPORT_GFX_DMG |
4298 AMD_PG_SUPPORT_GDS |
4299 AMD_PG_SUPPORT_RLC_SMU_HS)) {
4300 gfx_v7_0_update_gfx_pg(adev, false);
4301 if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PG) {
4302 gfx_v7_0_enable_cp_pg(adev, false);
4303 gfx_v7_0_enable_gds_pg(adev, false);
4309 * gfx_v7_0_get_gpu_clock_counter - return GPU clock counter snapshot
4311 * @adev: amdgpu_device pointer
4313 * Fetches a GPU clock counter snapshot (SI).
4314 * Returns the 64 bit clock counter snapshot.
4316 static uint64_t gfx_v7_0_get_gpu_clock_counter(struct amdgpu_device *adev)
4320 mutex_lock(&adev->gfx.gpu_clock_mutex);
4321 WREG32(mmRLC_CAPTURE_GPU_CLOCK_COUNT, 1);
4322 clock = (uint64_t)RREG32(mmRLC_GPU_CLOCK_COUNT_LSB) |
4323 ((uint64_t)RREG32(mmRLC_GPU_CLOCK_COUNT_MSB) << 32ULL);
4324 mutex_unlock(&adev->gfx.gpu_clock_mutex);
4328 static void gfx_v7_0_ring_emit_gds_switch(struct amdgpu_ring *ring,
4330 uint32_t gds_base, uint32_t gds_size,
4331 uint32_t gws_base, uint32_t gws_size,
4332 uint32_t oa_base, uint32_t oa_size)
4334 gds_base = gds_base >> AMDGPU_GDS_SHIFT;
4335 gds_size = gds_size >> AMDGPU_GDS_SHIFT;
4337 gws_base = gws_base >> AMDGPU_GWS_SHIFT;
4338 gws_size = gws_size >> AMDGPU_GWS_SHIFT;
4340 oa_base = oa_base >> AMDGPU_OA_SHIFT;
4341 oa_size = oa_size >> AMDGPU_OA_SHIFT;
4344 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
4345 amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
4346 WRITE_DATA_DST_SEL(0)));
4347 amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].mem_base);
4348 amdgpu_ring_write(ring, 0);
4349 amdgpu_ring_write(ring, gds_base);
4352 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
4353 amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
4354 WRITE_DATA_DST_SEL(0)));
4355 amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].mem_size);
4356 amdgpu_ring_write(ring, 0);
4357 amdgpu_ring_write(ring, gds_size);
4360 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
4361 amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
4362 WRITE_DATA_DST_SEL(0)));
4363 amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].gws);
4364 amdgpu_ring_write(ring, 0);
4365 amdgpu_ring_write(ring, gws_size << GDS_GWS_VMID0__SIZE__SHIFT | gws_base);
4368 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
4369 amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
4370 WRITE_DATA_DST_SEL(0)));
4371 amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].oa);
4372 amdgpu_ring_write(ring, 0);
4373 amdgpu_ring_write(ring, (1 << (oa_size + oa_base)) - (1 << oa_base));
4376 static uint32_t wave_read_ind(struct amdgpu_device *adev, uint32_t simd, uint32_t wave, uint32_t address)
4378 WREG32(mmSQ_IND_INDEX,
4379 (wave << SQ_IND_INDEX__WAVE_ID__SHIFT) |
4380 (simd << SQ_IND_INDEX__SIMD_ID__SHIFT) |
4381 (address << SQ_IND_INDEX__INDEX__SHIFT) |
4382 (SQ_IND_INDEX__FORCE_READ_MASK));
4383 return RREG32(mmSQ_IND_DATA);
4386 static void wave_read_regs(struct amdgpu_device *adev, uint32_t simd,
4387 uint32_t wave, uint32_t thread,
4388 uint32_t regno, uint32_t num, uint32_t *out)
4390 WREG32(mmSQ_IND_INDEX,
4391 (wave << SQ_IND_INDEX__WAVE_ID__SHIFT) |
4392 (simd << SQ_IND_INDEX__SIMD_ID__SHIFT) |
4393 (regno << SQ_IND_INDEX__INDEX__SHIFT) |
4394 (thread << SQ_IND_INDEX__THREAD_ID__SHIFT) |
4395 (SQ_IND_INDEX__FORCE_READ_MASK) |
4396 (SQ_IND_INDEX__AUTO_INCR_MASK));
4398 *(out++) = RREG32(mmSQ_IND_DATA);
4401 static void gfx_v7_0_read_wave_data(struct amdgpu_device *adev, uint32_t simd, uint32_t wave, uint32_t *dst, int *no_fields)
4403 /* type 0 wave data */
4404 dst[(*no_fields)++] = 0;
4405 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_STATUS);
4406 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_LO);
4407 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_HI);
4408 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_LO);
4409 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_HI);
4410 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_HW_ID);
4411 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW0);
4412 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW1);
4413 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_GPR_ALLOC);
4414 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_LDS_ALLOC);
4415 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TRAPSTS);
4416 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_IB_STS);
4417 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TBA_LO);
4418 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TBA_HI);
4419 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TMA_LO);
4420 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TMA_HI);
4421 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_IB_DBG0);
4422 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_M0);
4425 static void gfx_v7_0_read_wave_sgprs(struct amdgpu_device *adev, uint32_t simd,
4426 uint32_t wave, uint32_t start,
4427 uint32_t size, uint32_t *dst)
4430 adev, simd, wave, 0,
4431 start + SQIND_WAVE_SGPRS_OFFSET, size, dst);
4434 static const struct amdgpu_gfx_funcs gfx_v7_0_gfx_funcs = {
4435 .get_gpu_clock_counter = &gfx_v7_0_get_gpu_clock_counter,
4436 .select_se_sh = &gfx_v7_0_select_se_sh,
4437 .read_wave_data = &gfx_v7_0_read_wave_data,
4438 .read_wave_sgprs = &gfx_v7_0_read_wave_sgprs,
4441 static const struct amdgpu_rlc_funcs gfx_v7_0_rlc_funcs = {
4442 .enter_safe_mode = gfx_v7_0_enter_rlc_safe_mode,
4443 .exit_safe_mode = gfx_v7_0_exit_rlc_safe_mode
4446 static int gfx_v7_0_early_init(void *handle)
4448 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4450 adev->gfx.num_gfx_rings = GFX7_NUM_GFX_RINGS;
4451 adev->gfx.num_compute_rings = GFX7_NUM_COMPUTE_RINGS;
4452 adev->gfx.funcs = &gfx_v7_0_gfx_funcs;
4453 adev->gfx.rlc.funcs = &gfx_v7_0_rlc_funcs;
4454 gfx_v7_0_set_ring_funcs(adev);
4455 gfx_v7_0_set_irq_funcs(adev);
4456 gfx_v7_0_set_gds_init(adev);
4461 static int gfx_v7_0_late_init(void *handle)
4463 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4466 r = amdgpu_irq_get(adev, &adev->gfx.priv_reg_irq, 0);
4470 r = amdgpu_irq_get(adev, &adev->gfx.priv_inst_irq, 0);
4477 static void gfx_v7_0_gpu_early_init(struct amdgpu_device *adev)
4480 u32 mc_shared_chmap, mc_arb_ramcfg;
4481 u32 dimm00_addr_map, dimm01_addr_map, dimm10_addr_map, dimm11_addr_map;
4484 switch (adev->asic_type) {
4486 adev->gfx.config.max_shader_engines = 2;
4487 adev->gfx.config.max_tile_pipes = 4;
4488 adev->gfx.config.max_cu_per_sh = 7;
4489 adev->gfx.config.max_sh_per_se = 1;
4490 adev->gfx.config.max_backends_per_se = 2;
4491 adev->gfx.config.max_texture_channel_caches = 4;
4492 adev->gfx.config.max_gprs = 256;
4493 adev->gfx.config.max_gs_threads = 32;
4494 adev->gfx.config.max_hw_contexts = 8;
4496 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
4497 adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
4498 adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
4499 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
4500 gb_addr_config = BONAIRE_GB_ADDR_CONFIG_GOLDEN;
4503 adev->gfx.config.max_shader_engines = 4;
4504 adev->gfx.config.max_tile_pipes = 16;
4505 adev->gfx.config.max_cu_per_sh = 11;
4506 adev->gfx.config.max_sh_per_se = 1;
4507 adev->gfx.config.max_backends_per_se = 4;
4508 adev->gfx.config.max_texture_channel_caches = 16;
4509 adev->gfx.config.max_gprs = 256;
4510 adev->gfx.config.max_gs_threads = 32;
4511 adev->gfx.config.max_hw_contexts = 8;
4513 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
4514 adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
4515 adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
4516 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
4517 gb_addr_config = HAWAII_GB_ADDR_CONFIG_GOLDEN;
4520 adev->gfx.config.max_shader_engines = 1;
4521 adev->gfx.config.max_tile_pipes = 4;
4522 if ((adev->pdev->device == 0x1304) ||
4523 (adev->pdev->device == 0x1305) ||
4524 (adev->pdev->device == 0x130C) ||
4525 (adev->pdev->device == 0x130F) ||
4526 (adev->pdev->device == 0x1310) ||
4527 (adev->pdev->device == 0x1311) ||
4528 (adev->pdev->device == 0x131C)) {
4529 adev->gfx.config.max_cu_per_sh = 8;
4530 adev->gfx.config.max_backends_per_se = 2;
4531 } else if ((adev->pdev->device == 0x1309) ||
4532 (adev->pdev->device == 0x130A) ||
4533 (adev->pdev->device == 0x130D) ||
4534 (adev->pdev->device == 0x1313) ||
4535 (adev->pdev->device == 0x131D)) {
4536 adev->gfx.config.max_cu_per_sh = 6;
4537 adev->gfx.config.max_backends_per_se = 2;
4538 } else if ((adev->pdev->device == 0x1306) ||
4539 (adev->pdev->device == 0x1307) ||
4540 (adev->pdev->device == 0x130B) ||
4541 (adev->pdev->device == 0x130E) ||
4542 (adev->pdev->device == 0x1315) ||
4543 (adev->pdev->device == 0x131B)) {
4544 adev->gfx.config.max_cu_per_sh = 4;
4545 adev->gfx.config.max_backends_per_se = 1;
4547 adev->gfx.config.max_cu_per_sh = 3;
4548 adev->gfx.config.max_backends_per_se = 1;
4550 adev->gfx.config.max_sh_per_se = 1;
4551 adev->gfx.config.max_texture_channel_caches = 4;
4552 adev->gfx.config.max_gprs = 256;
4553 adev->gfx.config.max_gs_threads = 16;
4554 adev->gfx.config.max_hw_contexts = 8;
4556 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
4557 adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
4558 adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
4559 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
4560 gb_addr_config = BONAIRE_GB_ADDR_CONFIG_GOLDEN;
4565 adev->gfx.config.max_shader_engines = 1;
4566 adev->gfx.config.max_tile_pipes = 2;
4567 adev->gfx.config.max_cu_per_sh = 2;
4568 adev->gfx.config.max_sh_per_se = 1;
4569 adev->gfx.config.max_backends_per_se = 1;
4570 adev->gfx.config.max_texture_channel_caches = 2;
4571 adev->gfx.config.max_gprs = 256;
4572 adev->gfx.config.max_gs_threads = 16;
4573 adev->gfx.config.max_hw_contexts = 8;
4575 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
4576 adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
4577 adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
4578 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
4579 gb_addr_config = BONAIRE_GB_ADDR_CONFIG_GOLDEN;
4583 mc_shared_chmap = RREG32(mmMC_SHARED_CHMAP);
4584 adev->gfx.config.mc_arb_ramcfg = RREG32(mmMC_ARB_RAMCFG);
4585 mc_arb_ramcfg = adev->gfx.config.mc_arb_ramcfg;
4587 adev->gfx.config.num_tile_pipes = adev->gfx.config.max_tile_pipes;
4588 adev->gfx.config.mem_max_burst_length_bytes = 256;
4589 if (adev->flags & AMD_IS_APU) {
4590 /* Get memory bank mapping mode. */
4591 tmp = RREG32(mmMC_FUS_DRAM0_BANK_ADDR_MAPPING);
4592 dimm00_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM0_BANK_ADDR_MAPPING, DIMM0ADDRMAP);
4593 dimm01_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM0_BANK_ADDR_MAPPING, DIMM1ADDRMAP);
4595 tmp = RREG32(mmMC_FUS_DRAM1_BANK_ADDR_MAPPING);
4596 dimm10_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM1_BANK_ADDR_MAPPING, DIMM0ADDRMAP);
4597 dimm11_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM1_BANK_ADDR_MAPPING, DIMM1ADDRMAP);
4599 /* Validate settings in case only one DIMM installed. */
4600 if ((dimm00_addr_map == 0) || (dimm00_addr_map == 3) || (dimm00_addr_map == 4) || (dimm00_addr_map > 12))
4601 dimm00_addr_map = 0;
4602 if ((dimm01_addr_map == 0) || (dimm01_addr_map == 3) || (dimm01_addr_map == 4) || (dimm01_addr_map > 12))
4603 dimm01_addr_map = 0;
4604 if ((dimm10_addr_map == 0) || (dimm10_addr_map == 3) || (dimm10_addr_map == 4) || (dimm10_addr_map > 12))
4605 dimm10_addr_map = 0;
4606 if ((dimm11_addr_map == 0) || (dimm11_addr_map == 3) || (dimm11_addr_map == 4) || (dimm11_addr_map > 12))
4607 dimm11_addr_map = 0;
4609 /* If DIMM Addr map is 8GB, ROW size should be 2KB. Otherwise 1KB. */
4610 /* If ROW size(DIMM1) != ROW size(DMIMM0), ROW size should be larger one. */
4611 if ((dimm00_addr_map == 11) || (dimm01_addr_map == 11) || (dimm10_addr_map == 11) || (dimm11_addr_map == 11))
4612 adev->gfx.config.mem_row_size_in_kb = 2;
4614 adev->gfx.config.mem_row_size_in_kb = 1;
4616 tmp = (mc_arb_ramcfg & MC_ARB_RAMCFG__NOOFCOLS_MASK) >> MC_ARB_RAMCFG__NOOFCOLS__SHIFT;
4617 adev->gfx.config.mem_row_size_in_kb = (4 * (1 << (8 + tmp))) / 1024;
4618 if (adev->gfx.config.mem_row_size_in_kb > 4)
4619 adev->gfx.config.mem_row_size_in_kb = 4;
4621 /* XXX use MC settings? */
4622 adev->gfx.config.shader_engine_tile_size = 32;
4623 adev->gfx.config.num_gpus = 1;
4624 adev->gfx.config.multi_gpu_tile_size = 64;
4626 /* fix up row size */
4627 gb_addr_config &= ~GB_ADDR_CONFIG__ROW_SIZE_MASK;
4628 switch (adev->gfx.config.mem_row_size_in_kb) {
4631 gb_addr_config |= (0 << GB_ADDR_CONFIG__ROW_SIZE__SHIFT);
4634 gb_addr_config |= (1 << GB_ADDR_CONFIG__ROW_SIZE__SHIFT);
4637 gb_addr_config |= (2 << GB_ADDR_CONFIG__ROW_SIZE__SHIFT);
4640 adev->gfx.config.gb_addr_config = gb_addr_config;
4643 static int gfx_v7_0_sw_init(void *handle)
4645 struct amdgpu_ring *ring;
4646 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4650 r = amdgpu_irq_add_id(adev, 181, &adev->gfx.eop_irq);
4654 /* Privileged reg */
4655 r = amdgpu_irq_add_id(adev, 184, &adev->gfx.priv_reg_irq);
4659 /* Privileged inst */
4660 r = amdgpu_irq_add_id(adev, 185, &adev->gfx.priv_inst_irq);
4664 gfx_v7_0_scratch_init(adev);
4666 r = gfx_v7_0_init_microcode(adev);
4668 DRM_ERROR("Failed to load gfx firmware!\n");
4672 r = gfx_v7_0_rlc_init(adev);
4674 DRM_ERROR("Failed to init rlc BOs!\n");
4678 /* allocate mec buffers */
4679 r = gfx_v7_0_mec_init(adev);
4681 DRM_ERROR("Failed to init MEC BOs!\n");
4685 for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
4686 ring = &adev->gfx.gfx_ring[i];
4687 ring->ring_obj = NULL;
4688 sprintf(ring->name, "gfx");
4689 r = amdgpu_ring_init(adev, ring, 1024,
4690 &adev->gfx.eop_irq, AMDGPU_CP_IRQ_GFX_EOP);
4695 /* set up the compute queues */
4696 for (i = 0; i < adev->gfx.num_compute_rings; i++) {
4699 /* max 32 queues per MEC */
4700 if ((i >= 32) || (i >= AMDGPU_MAX_COMPUTE_RINGS)) {
4701 DRM_ERROR("Too many (%d) compute rings!\n", i);
4704 ring = &adev->gfx.compute_ring[i];
4705 ring->ring_obj = NULL;
4706 ring->use_doorbell = true;
4707 ring->doorbell_index = AMDGPU_DOORBELL_MEC_RING0 + i;
4708 ring->me = 1; /* first MEC */
4710 ring->queue = i % 8;
4711 sprintf(ring->name, "comp_%d.%d.%d", ring->me, ring->pipe, ring->queue);
4712 irq_type = AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP + ring->pipe;
4713 /* type-2 packets are deprecated on MEC, use type-3 instead */
4714 r = amdgpu_ring_init(adev, ring, 1024,
4715 &adev->gfx.eop_irq, irq_type);
4720 /* reserve GDS, GWS and OA resource for gfx */
4721 r = amdgpu_bo_create_kernel(adev, adev->gds.mem.gfx_partition_size,
4722 PAGE_SIZE, AMDGPU_GEM_DOMAIN_GDS,
4723 &adev->gds.gds_gfx_bo, NULL, NULL);
4727 r = amdgpu_bo_create_kernel(adev, adev->gds.gws.gfx_partition_size,
4728 PAGE_SIZE, AMDGPU_GEM_DOMAIN_GWS,
4729 &adev->gds.gws_gfx_bo, NULL, NULL);
4733 r = amdgpu_bo_create_kernel(adev, adev->gds.oa.gfx_partition_size,
4734 PAGE_SIZE, AMDGPU_GEM_DOMAIN_OA,
4735 &adev->gds.oa_gfx_bo, NULL, NULL);
4739 adev->gfx.ce_ram_size = 0x8000;
4741 gfx_v7_0_gpu_early_init(adev);
4746 static int gfx_v7_0_sw_fini(void *handle)
4749 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4751 amdgpu_bo_free_kernel(&adev->gds.oa_gfx_bo, NULL, NULL);
4752 amdgpu_bo_free_kernel(&adev->gds.gws_gfx_bo, NULL, NULL);
4753 amdgpu_bo_free_kernel(&adev->gds.gds_gfx_bo, NULL, NULL);
4755 for (i = 0; i < adev->gfx.num_gfx_rings; i++)
4756 amdgpu_ring_fini(&adev->gfx.gfx_ring[i]);
4757 for (i = 0; i < adev->gfx.num_compute_rings; i++)
4758 amdgpu_ring_fini(&adev->gfx.compute_ring[i]);
4760 gfx_v7_0_cp_compute_fini(adev);
4761 gfx_v7_0_rlc_fini(adev);
4762 gfx_v7_0_mec_fini(adev);
4763 gfx_v7_0_free_microcode(adev);
4768 static int gfx_v7_0_hw_init(void *handle)
4771 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4773 gfx_v7_0_gpu_init(adev);
4776 r = gfx_v7_0_rlc_resume(adev);
4780 r = gfx_v7_0_cp_resume(adev);
4787 static int gfx_v7_0_hw_fini(void *handle)
4789 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4791 amdgpu_irq_put(adev, &adev->gfx.priv_reg_irq, 0);
4792 amdgpu_irq_put(adev, &adev->gfx.priv_inst_irq, 0);
4793 gfx_v7_0_cp_enable(adev, false);
4794 gfx_v7_0_rlc_stop(adev);
4795 gfx_v7_0_fini_pg(adev);
4800 static int gfx_v7_0_suspend(void *handle)
4802 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4804 return gfx_v7_0_hw_fini(adev);
4807 static int gfx_v7_0_resume(void *handle)
4809 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4811 return gfx_v7_0_hw_init(adev);
4814 static bool gfx_v7_0_is_idle(void *handle)
4816 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4818 if (RREG32(mmGRBM_STATUS) & GRBM_STATUS__GUI_ACTIVE_MASK)
4824 static int gfx_v7_0_wait_for_idle(void *handle)
4828 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4830 for (i = 0; i < adev->usec_timeout; i++) {
4831 /* read MC_STATUS */
4832 tmp = RREG32(mmGRBM_STATUS) & GRBM_STATUS__GUI_ACTIVE_MASK;
4841 static int gfx_v7_0_soft_reset(void *handle)
4843 u32 grbm_soft_reset = 0, srbm_soft_reset = 0;
4845 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4848 tmp = RREG32(mmGRBM_STATUS);
4849 if (tmp & (GRBM_STATUS__PA_BUSY_MASK | GRBM_STATUS__SC_BUSY_MASK |
4850 GRBM_STATUS__BCI_BUSY_MASK | GRBM_STATUS__SX_BUSY_MASK |
4851 GRBM_STATUS__TA_BUSY_MASK | GRBM_STATUS__VGT_BUSY_MASK |
4852 GRBM_STATUS__DB_BUSY_MASK | GRBM_STATUS__CB_BUSY_MASK |
4853 GRBM_STATUS__GDS_BUSY_MASK | GRBM_STATUS__SPI_BUSY_MASK |
4854 GRBM_STATUS__IA_BUSY_MASK | GRBM_STATUS__IA_BUSY_NO_DMA_MASK))
4855 grbm_soft_reset |= GRBM_SOFT_RESET__SOFT_RESET_CP_MASK |
4856 GRBM_SOFT_RESET__SOFT_RESET_GFX_MASK;
4858 if (tmp & (GRBM_STATUS__CP_BUSY_MASK | GRBM_STATUS__CP_COHERENCY_BUSY_MASK)) {
4859 grbm_soft_reset |= GRBM_SOFT_RESET__SOFT_RESET_CP_MASK;
4860 srbm_soft_reset |= SRBM_SOFT_RESET__SOFT_RESET_GRBM_MASK;
4864 tmp = RREG32(mmGRBM_STATUS2);
4865 if (tmp & GRBM_STATUS2__RLC_BUSY_MASK)
4866 grbm_soft_reset |= GRBM_SOFT_RESET__SOFT_RESET_RLC_MASK;
4869 tmp = RREG32(mmSRBM_STATUS);
4870 if (tmp & SRBM_STATUS__GRBM_RQ_PENDING_MASK)
4871 srbm_soft_reset |= SRBM_SOFT_RESET__SOFT_RESET_GRBM_MASK;
4873 if (grbm_soft_reset || srbm_soft_reset) {
4875 gfx_v7_0_fini_pg(adev);
4876 gfx_v7_0_update_cg(adev, false);
4879 gfx_v7_0_rlc_stop(adev);
4881 /* Disable GFX parsing/prefetching */
4882 WREG32(mmCP_ME_CNTL, CP_ME_CNTL__ME_HALT_MASK | CP_ME_CNTL__PFP_HALT_MASK | CP_ME_CNTL__CE_HALT_MASK);
4884 /* Disable MEC parsing/prefetching */
4885 WREG32(mmCP_MEC_CNTL, CP_MEC_CNTL__MEC_ME1_HALT_MASK | CP_MEC_CNTL__MEC_ME2_HALT_MASK);
4887 if (grbm_soft_reset) {
4888 tmp = RREG32(mmGRBM_SOFT_RESET);
4889 tmp |= grbm_soft_reset;
4890 dev_info(adev->dev, "GRBM_SOFT_RESET=0x%08X\n", tmp);
4891 WREG32(mmGRBM_SOFT_RESET, tmp);
4892 tmp = RREG32(mmGRBM_SOFT_RESET);
4896 tmp &= ~grbm_soft_reset;
4897 WREG32(mmGRBM_SOFT_RESET, tmp);
4898 tmp = RREG32(mmGRBM_SOFT_RESET);
4901 if (srbm_soft_reset) {
4902 tmp = RREG32(mmSRBM_SOFT_RESET);
4903 tmp |= srbm_soft_reset;
4904 dev_info(adev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp);
4905 WREG32(mmSRBM_SOFT_RESET, tmp);
4906 tmp = RREG32(mmSRBM_SOFT_RESET);
4910 tmp &= ~srbm_soft_reset;
4911 WREG32(mmSRBM_SOFT_RESET, tmp);
4912 tmp = RREG32(mmSRBM_SOFT_RESET);
4914 /* Wait a little for things to settle down */
4920 static void gfx_v7_0_set_gfx_eop_interrupt_state(struct amdgpu_device *adev,
4921 enum amdgpu_interrupt_state state)
4926 case AMDGPU_IRQ_STATE_DISABLE:
4927 cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0);
4928 cp_int_cntl &= ~CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK;
4929 WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl);
4931 case AMDGPU_IRQ_STATE_ENABLE:
4932 cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0);
4933 cp_int_cntl |= CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK;
4934 WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl);
4941 static void gfx_v7_0_set_compute_eop_interrupt_state(struct amdgpu_device *adev,
4943 enum amdgpu_interrupt_state state)
4945 u32 mec_int_cntl, mec_int_cntl_reg;
4948 * amdgpu controls only pipe 0 of MEC1. That's why this function only
4949 * handles the setting of interrupts for this specific pipe. All other
4950 * pipes' interrupts are set by amdkfd.
4956 mec_int_cntl_reg = mmCP_ME1_PIPE0_INT_CNTL;
4959 DRM_DEBUG("invalid pipe %d\n", pipe);
4963 DRM_DEBUG("invalid me %d\n", me);
4968 case AMDGPU_IRQ_STATE_DISABLE:
4969 mec_int_cntl = RREG32(mec_int_cntl_reg);
4970 mec_int_cntl &= ~CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK;
4971 WREG32(mec_int_cntl_reg, mec_int_cntl);
4973 case AMDGPU_IRQ_STATE_ENABLE:
4974 mec_int_cntl = RREG32(mec_int_cntl_reg);
4975 mec_int_cntl |= CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK;
4976 WREG32(mec_int_cntl_reg, mec_int_cntl);
4983 static int gfx_v7_0_set_priv_reg_fault_state(struct amdgpu_device *adev,
4984 struct amdgpu_irq_src *src,
4986 enum amdgpu_interrupt_state state)
4991 case AMDGPU_IRQ_STATE_DISABLE:
4992 cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0);
4993 cp_int_cntl &= ~CP_INT_CNTL_RING0__PRIV_REG_INT_ENABLE_MASK;
4994 WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl);
4996 case AMDGPU_IRQ_STATE_ENABLE:
4997 cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0);
4998 cp_int_cntl |= CP_INT_CNTL_RING0__PRIV_REG_INT_ENABLE_MASK;
4999 WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl);
5008 static int gfx_v7_0_set_priv_inst_fault_state(struct amdgpu_device *adev,
5009 struct amdgpu_irq_src *src,
5011 enum amdgpu_interrupt_state state)
5016 case AMDGPU_IRQ_STATE_DISABLE:
5017 cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0);
5018 cp_int_cntl &= ~CP_INT_CNTL_RING0__PRIV_INSTR_INT_ENABLE_MASK;
5019 WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl);
5021 case AMDGPU_IRQ_STATE_ENABLE:
5022 cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0);
5023 cp_int_cntl |= CP_INT_CNTL_RING0__PRIV_INSTR_INT_ENABLE_MASK;
5024 WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl);
5033 static int gfx_v7_0_set_eop_interrupt_state(struct amdgpu_device *adev,
5034 struct amdgpu_irq_src *src,
5036 enum amdgpu_interrupt_state state)
5039 case AMDGPU_CP_IRQ_GFX_EOP:
5040 gfx_v7_0_set_gfx_eop_interrupt_state(adev, state);
5042 case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP:
5043 gfx_v7_0_set_compute_eop_interrupt_state(adev, 1, 0, state);
5045 case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE1_EOP:
5046 gfx_v7_0_set_compute_eop_interrupt_state(adev, 1, 1, state);
5048 case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE2_EOP:
5049 gfx_v7_0_set_compute_eop_interrupt_state(adev, 1, 2, state);
5051 case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE3_EOP:
5052 gfx_v7_0_set_compute_eop_interrupt_state(adev, 1, 3, state);
5054 case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE0_EOP:
5055 gfx_v7_0_set_compute_eop_interrupt_state(adev, 2, 0, state);
5057 case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE1_EOP:
5058 gfx_v7_0_set_compute_eop_interrupt_state(adev, 2, 1, state);
5060 case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE2_EOP:
5061 gfx_v7_0_set_compute_eop_interrupt_state(adev, 2, 2, state);
5063 case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE3_EOP:
5064 gfx_v7_0_set_compute_eop_interrupt_state(adev, 2, 3, state);
5072 static int gfx_v7_0_eop_irq(struct amdgpu_device *adev,
5073 struct amdgpu_irq_src *source,
5074 struct amdgpu_iv_entry *entry)
5077 struct amdgpu_ring *ring;
5080 DRM_DEBUG("IH: CP EOP\n");
5081 me_id = (entry->ring_id & 0x0c) >> 2;
5082 pipe_id = (entry->ring_id & 0x03) >> 0;
5085 amdgpu_fence_process(&adev->gfx.gfx_ring[0]);
5089 for (i = 0; i < adev->gfx.num_compute_rings; i++) {
5090 ring = &adev->gfx.compute_ring[i];
5091 if ((ring->me == me_id) && (ring->pipe == pipe_id))
5092 amdgpu_fence_process(ring);
5099 static int gfx_v7_0_priv_reg_irq(struct amdgpu_device *adev,
5100 struct amdgpu_irq_src *source,
5101 struct amdgpu_iv_entry *entry)
5103 DRM_ERROR("Illegal register access in command stream\n");
5104 schedule_work(&adev->reset_work);
5108 static int gfx_v7_0_priv_inst_irq(struct amdgpu_device *adev,
5109 struct amdgpu_irq_src *source,
5110 struct amdgpu_iv_entry *entry)
5112 DRM_ERROR("Illegal instruction in command stream\n");
5113 // XXX soft reset the gfx block only
5114 schedule_work(&adev->reset_work);
5118 static int gfx_v7_0_set_clockgating_state(void *handle,
5119 enum amd_clockgating_state state)
5122 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5124 if (state == AMD_CG_STATE_GATE)
5127 gfx_v7_0_enable_gui_idle_interrupt(adev, false);
5128 /* order matters! */
5130 gfx_v7_0_enable_mgcg(adev, true);
5131 gfx_v7_0_enable_cgcg(adev, true);
5133 gfx_v7_0_enable_cgcg(adev, false);
5134 gfx_v7_0_enable_mgcg(adev, false);
5136 gfx_v7_0_enable_gui_idle_interrupt(adev, true);
5141 static int gfx_v7_0_set_powergating_state(void *handle,
5142 enum amd_powergating_state state)
5145 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5147 if (state == AMD_PG_STATE_GATE)
5150 if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_PG |
5151 AMD_PG_SUPPORT_GFX_SMG |
5152 AMD_PG_SUPPORT_GFX_DMG |
5154 AMD_PG_SUPPORT_GDS |
5155 AMD_PG_SUPPORT_RLC_SMU_HS)) {
5156 gfx_v7_0_update_gfx_pg(adev, gate);
5157 if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PG) {
5158 gfx_v7_0_enable_cp_pg(adev, gate);
5159 gfx_v7_0_enable_gds_pg(adev, gate);
5166 static const struct amd_ip_funcs gfx_v7_0_ip_funcs = {
5168 .early_init = gfx_v7_0_early_init,
5169 .late_init = gfx_v7_0_late_init,
5170 .sw_init = gfx_v7_0_sw_init,
5171 .sw_fini = gfx_v7_0_sw_fini,
5172 .hw_init = gfx_v7_0_hw_init,
5173 .hw_fini = gfx_v7_0_hw_fini,
5174 .suspend = gfx_v7_0_suspend,
5175 .resume = gfx_v7_0_resume,
5176 .is_idle = gfx_v7_0_is_idle,
5177 .wait_for_idle = gfx_v7_0_wait_for_idle,
5178 .soft_reset = gfx_v7_0_soft_reset,
5179 .set_clockgating_state = gfx_v7_0_set_clockgating_state,
5180 .set_powergating_state = gfx_v7_0_set_powergating_state,
5183 static const struct amdgpu_ring_funcs gfx_v7_0_ring_funcs_gfx = {
5184 .type = AMDGPU_RING_TYPE_GFX,
5186 .nop = PACKET3(PACKET3_NOP, 0x3FFF),
5187 .get_rptr = gfx_v7_0_ring_get_rptr,
5188 .get_wptr = gfx_v7_0_ring_get_wptr_gfx,
5189 .set_wptr = gfx_v7_0_ring_set_wptr_gfx,
5191 20 + /* gfx_v7_0_ring_emit_gds_switch */
5192 7 + /* gfx_v7_0_ring_emit_hdp_flush */
5193 5 + /* gfx_v7_0_ring_emit_hdp_invalidate */
5194 12 + 12 + 12 + /* gfx_v7_0_ring_emit_fence_gfx x3 for user fence, vm fence */
5195 7 + 4 + /* gfx_v7_0_ring_emit_pipeline_sync */
5196 17 + 6 + /* gfx_v7_0_ring_emit_vm_flush */
5197 3 + 4, /* gfx_v7_ring_emit_cntxcntl including vgt flush*/
5198 .emit_ib_size = 4, /* gfx_v7_0_ring_emit_ib_gfx */
5199 .emit_ib = gfx_v7_0_ring_emit_ib_gfx,
5200 .emit_fence = gfx_v7_0_ring_emit_fence_gfx,
5201 .emit_pipeline_sync = gfx_v7_0_ring_emit_pipeline_sync,
5202 .emit_vm_flush = gfx_v7_0_ring_emit_vm_flush,
5203 .emit_gds_switch = gfx_v7_0_ring_emit_gds_switch,
5204 .emit_hdp_flush = gfx_v7_0_ring_emit_hdp_flush,
5205 .emit_hdp_invalidate = gfx_v7_0_ring_emit_hdp_invalidate,
5206 .test_ring = gfx_v7_0_ring_test_ring,
5207 .test_ib = gfx_v7_0_ring_test_ib,
5208 .insert_nop = amdgpu_ring_insert_nop,
5209 .pad_ib = amdgpu_ring_generic_pad_ib,
5210 .emit_cntxcntl = gfx_v7_ring_emit_cntxcntl,
5213 static const struct amdgpu_ring_funcs gfx_v7_0_ring_funcs_compute = {
5214 .type = AMDGPU_RING_TYPE_COMPUTE,
5216 .nop = PACKET3(PACKET3_NOP, 0x3FFF),
5217 .get_rptr = gfx_v7_0_ring_get_rptr,
5218 .get_wptr = gfx_v7_0_ring_get_wptr_compute,
5219 .set_wptr = gfx_v7_0_ring_set_wptr_compute,
5221 20 + /* gfx_v7_0_ring_emit_gds_switch */
5222 7 + /* gfx_v7_0_ring_emit_hdp_flush */
5223 5 + /* gfx_v7_0_ring_emit_hdp_invalidate */
5224 7 + /* gfx_v7_0_ring_emit_pipeline_sync */
5225 17 + /* gfx_v7_0_ring_emit_vm_flush */
5226 7 + 7 + 7, /* gfx_v7_0_ring_emit_fence_compute x3 for user fence, vm fence */
5227 .emit_ib_size = 4, /* gfx_v7_0_ring_emit_ib_compute */
5228 .emit_ib = gfx_v7_0_ring_emit_ib_compute,
5229 .emit_fence = gfx_v7_0_ring_emit_fence_compute,
5230 .emit_pipeline_sync = gfx_v7_0_ring_emit_pipeline_sync,
5231 .emit_vm_flush = gfx_v7_0_ring_emit_vm_flush,
5232 .emit_gds_switch = gfx_v7_0_ring_emit_gds_switch,
5233 .emit_hdp_flush = gfx_v7_0_ring_emit_hdp_flush,
5234 .emit_hdp_invalidate = gfx_v7_0_ring_emit_hdp_invalidate,
5235 .test_ring = gfx_v7_0_ring_test_ring,
5236 .test_ib = gfx_v7_0_ring_test_ib,
5237 .insert_nop = amdgpu_ring_insert_nop,
5238 .pad_ib = amdgpu_ring_generic_pad_ib,
5241 static void gfx_v7_0_set_ring_funcs(struct amdgpu_device *adev)
5245 for (i = 0; i < adev->gfx.num_gfx_rings; i++)
5246 adev->gfx.gfx_ring[i].funcs = &gfx_v7_0_ring_funcs_gfx;
5247 for (i = 0; i < adev->gfx.num_compute_rings; i++)
5248 adev->gfx.compute_ring[i].funcs = &gfx_v7_0_ring_funcs_compute;
5251 static const struct amdgpu_irq_src_funcs gfx_v7_0_eop_irq_funcs = {
5252 .set = gfx_v7_0_set_eop_interrupt_state,
5253 .process = gfx_v7_0_eop_irq,
5256 static const struct amdgpu_irq_src_funcs gfx_v7_0_priv_reg_irq_funcs = {
5257 .set = gfx_v7_0_set_priv_reg_fault_state,
5258 .process = gfx_v7_0_priv_reg_irq,
5261 static const struct amdgpu_irq_src_funcs gfx_v7_0_priv_inst_irq_funcs = {
5262 .set = gfx_v7_0_set_priv_inst_fault_state,
5263 .process = gfx_v7_0_priv_inst_irq,
5266 static void gfx_v7_0_set_irq_funcs(struct amdgpu_device *adev)
5268 adev->gfx.eop_irq.num_types = AMDGPU_CP_IRQ_LAST;
5269 adev->gfx.eop_irq.funcs = &gfx_v7_0_eop_irq_funcs;
5271 adev->gfx.priv_reg_irq.num_types = 1;
5272 adev->gfx.priv_reg_irq.funcs = &gfx_v7_0_priv_reg_irq_funcs;
5274 adev->gfx.priv_inst_irq.num_types = 1;
5275 adev->gfx.priv_inst_irq.funcs = &gfx_v7_0_priv_inst_irq_funcs;
5278 static void gfx_v7_0_set_gds_init(struct amdgpu_device *adev)
5280 /* init asci gds info */
5281 adev->gds.mem.total_size = RREG32(mmGDS_VMID0_SIZE);
5282 adev->gds.gws.total_size = 64;
5283 adev->gds.oa.total_size = 16;
5285 if (adev->gds.mem.total_size == 64 * 1024) {
5286 adev->gds.mem.gfx_partition_size = 4096;
5287 adev->gds.mem.cs_partition_size = 4096;
5289 adev->gds.gws.gfx_partition_size = 4;
5290 adev->gds.gws.cs_partition_size = 4;
5292 adev->gds.oa.gfx_partition_size = 4;
5293 adev->gds.oa.cs_partition_size = 1;
5295 adev->gds.mem.gfx_partition_size = 1024;
5296 adev->gds.mem.cs_partition_size = 1024;
5298 adev->gds.gws.gfx_partition_size = 16;
5299 adev->gds.gws.cs_partition_size = 16;
5301 adev->gds.oa.gfx_partition_size = 4;
5302 adev->gds.oa.cs_partition_size = 4;
5307 static void gfx_v7_0_get_cu_info(struct amdgpu_device *adev)
5309 int i, j, k, counter, active_cu_number = 0;
5310 u32 mask, bitmap, ao_bitmap, ao_cu_mask = 0;
5311 struct amdgpu_cu_info *cu_info = &adev->gfx.cu_info;
5312 unsigned disable_masks[4 * 2];
5314 memset(cu_info, 0, sizeof(*cu_info));
5316 amdgpu_gfx_parse_disable_cu(disable_masks, 4, 2);
5318 mutex_lock(&adev->grbm_idx_mutex);
5319 for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
5320 for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
5324 gfx_v7_0_select_se_sh(adev, i, j, 0xffffffff);
5326 gfx_v7_0_set_user_cu_inactive_bitmap(
5327 adev, disable_masks[i * 2 + j]);
5328 bitmap = gfx_v7_0_get_cu_active_bitmap(adev);
5329 cu_info->bitmap[i][j] = bitmap;
5331 for (k = 0; k < 16; k ++) {
5332 if (bitmap & mask) {
5339 active_cu_number += counter;
5340 ao_cu_mask |= (ao_bitmap << (i * 16 + j * 8));
5343 gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
5344 mutex_unlock(&adev->grbm_idx_mutex);
5346 cu_info->number = active_cu_number;
5347 cu_info->ao_cu_mask = ao_cu_mask;
5350 const struct amdgpu_ip_block_version gfx_v7_0_ip_block =
5352 .type = AMD_IP_BLOCK_TYPE_GFX,
5356 .funcs = &gfx_v7_0_ip_funcs,
5359 const struct amdgpu_ip_block_version gfx_v7_1_ip_block =
5361 .type = AMD_IP_BLOCK_TYPE_GFX,
5365 .funcs = &gfx_v7_0_ip_funcs,
5368 const struct amdgpu_ip_block_version gfx_v7_2_ip_block =
5370 .type = AMD_IP_BLOCK_TYPE_GFX,
5374 .funcs = &gfx_v7_0_ip_funcs,
5377 const struct amdgpu_ip_block_version gfx_v7_3_ip_block =
5379 .type = AMD_IP_BLOCK_TYPE_GFX,
5383 .funcs = &gfx_v7_0_ip_funcs,
projects / linux.git / blob