xref: /linux/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c (revision 7d1f84f1b94e79cb95d974ab3ec54cb3a4d64f5b)
1 /*
2  * Copyright 2014 Advanced Micro Devices, Inc.
3  *
4  * Permission is hereby granted, free of charge, to any person obtaining a
5  * copy of this software and associated documentation files (the "Software"),
6  * to deal in the Software without restriction, including without limitation
7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8  * and/or sell copies of the Software, and to permit persons to whom the
9  * Software is furnished to do so, subject to the following conditions:
10  *
11  * The above copyright notice and this permission notice shall be included in
12  * all copies or substantial portions of the Software.
13  *
14  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
17  * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20  * OTHER DEALINGS IN THE SOFTWARE.
21  *
22  */
23 
24 #include <linux/firmware.h>
25 #include <linux/module.h>
26 
27 #include "amdgpu.h"
28 #include "amdgpu_ih.h"
29 #include "amdgpu_gfx.h"
30 #include "cikd.h"
31 #include "cik.h"
32 #include "cik_structs.h"
33 #include "atom.h"
34 #include "amdgpu_ucode.h"
35 #include "clearstate_ci.h"
36 
37 #include "dce/dce_8_0_d.h"
38 #include "dce/dce_8_0_sh_mask.h"
39 
40 #include "bif/bif_4_1_d.h"
41 #include "bif/bif_4_1_sh_mask.h"
42 
43 #include "gca/gfx_7_0_d.h"
44 #include "gca/gfx_7_2_enum.h"
45 #include "gca/gfx_7_2_sh_mask.h"
46 
47 #include "gmc/gmc_7_0_d.h"
48 #include "gmc/gmc_7_0_sh_mask.h"
49 
50 #include "oss/oss_2_0_d.h"
51 #include "oss/oss_2_0_sh_mask.h"
52 
53 #define NUM_SIMD_PER_CU 0x4 /* missing from the gfx_7 IP headers */
54 
55 #define GFX7_NUM_GFX_RINGS     1
56 #define GFX7_MEC_HPD_SIZE      2048
57 
58 #define BONAIRE_GB_ADDR_CONFIG_GOLDEN        0x12010001
59 #define HAWAII_GB_ADDR_CONFIG_GOLDEN         0x12011003
60 
61 static void gfx_v7_0_set_ring_funcs(struct amdgpu_device *adev);
62 static void gfx_v7_0_set_irq_funcs(struct amdgpu_device *adev);
63 static void gfx_v7_0_set_gds_init(struct amdgpu_device *adev);
64 
65 MODULE_FIRMWARE("amdgpu/bonaire_pfp.bin");
66 MODULE_FIRMWARE("amdgpu/bonaire_me.bin");
67 MODULE_FIRMWARE("amdgpu/bonaire_ce.bin");
68 MODULE_FIRMWARE("amdgpu/bonaire_rlc.bin");
69 MODULE_FIRMWARE("amdgpu/bonaire_mec.bin");
70 
71 MODULE_FIRMWARE("amdgpu/hawaii_pfp.bin");
72 MODULE_FIRMWARE("amdgpu/hawaii_me.bin");
73 MODULE_FIRMWARE("amdgpu/hawaii_ce.bin");
74 MODULE_FIRMWARE("amdgpu/hawaii_rlc.bin");
75 MODULE_FIRMWARE("amdgpu/hawaii_mec.bin");
76 
77 MODULE_FIRMWARE("amdgpu/kaveri_pfp.bin");
78 MODULE_FIRMWARE("amdgpu/kaveri_me.bin");
79 MODULE_FIRMWARE("amdgpu/kaveri_ce.bin");
80 MODULE_FIRMWARE("amdgpu/kaveri_rlc.bin");
81 MODULE_FIRMWARE("amdgpu/kaveri_mec.bin");
82 MODULE_FIRMWARE("amdgpu/kaveri_mec2.bin");
83 
84 MODULE_FIRMWARE("amdgpu/kabini_pfp.bin");
85 MODULE_FIRMWARE("amdgpu/kabini_me.bin");
86 MODULE_FIRMWARE("amdgpu/kabini_ce.bin");
87 MODULE_FIRMWARE("amdgpu/kabini_rlc.bin");
88 MODULE_FIRMWARE("amdgpu/kabini_mec.bin");
89 
90 MODULE_FIRMWARE("amdgpu/mullins_pfp.bin");
91 MODULE_FIRMWARE("amdgpu/mullins_me.bin");
92 MODULE_FIRMWARE("amdgpu/mullins_ce.bin");
93 MODULE_FIRMWARE("amdgpu/mullins_rlc.bin");
94 MODULE_FIRMWARE("amdgpu/mullins_mec.bin");
95 
96 static const struct amdgpu_gds_reg_offset amdgpu_gds_reg_offset[] = {
97 	{mmGDS_VMID0_BASE, mmGDS_VMID0_SIZE, mmGDS_GWS_VMID0, mmGDS_OA_VMID0},
98 	{mmGDS_VMID1_BASE, mmGDS_VMID1_SIZE, mmGDS_GWS_VMID1, mmGDS_OA_VMID1},
99 	{mmGDS_VMID2_BASE, mmGDS_VMID2_SIZE, mmGDS_GWS_VMID2, mmGDS_OA_VMID2},
100 	{mmGDS_VMID3_BASE, mmGDS_VMID3_SIZE, mmGDS_GWS_VMID3, mmGDS_OA_VMID3},
101 	{mmGDS_VMID4_BASE, mmGDS_VMID4_SIZE, mmGDS_GWS_VMID4, mmGDS_OA_VMID4},
102 	{mmGDS_VMID5_BASE, mmGDS_VMID5_SIZE, mmGDS_GWS_VMID5, mmGDS_OA_VMID5},
103 	{mmGDS_VMID6_BASE, mmGDS_VMID6_SIZE, mmGDS_GWS_VMID6, mmGDS_OA_VMID6},
104 	{mmGDS_VMID7_BASE, mmGDS_VMID7_SIZE, mmGDS_GWS_VMID7, mmGDS_OA_VMID7},
105 	{mmGDS_VMID8_BASE, mmGDS_VMID8_SIZE, mmGDS_GWS_VMID8, mmGDS_OA_VMID8},
106 	{mmGDS_VMID9_BASE, mmGDS_VMID9_SIZE, mmGDS_GWS_VMID9, mmGDS_OA_VMID9},
107 	{mmGDS_VMID10_BASE, mmGDS_VMID10_SIZE, mmGDS_GWS_VMID10, mmGDS_OA_VMID10},
108 	{mmGDS_VMID11_BASE, mmGDS_VMID11_SIZE, mmGDS_GWS_VMID11, mmGDS_OA_VMID11},
109 	{mmGDS_VMID12_BASE, mmGDS_VMID12_SIZE, mmGDS_GWS_VMID12, mmGDS_OA_VMID12},
110 	{mmGDS_VMID13_BASE, mmGDS_VMID13_SIZE, mmGDS_GWS_VMID13, mmGDS_OA_VMID13},
111 	{mmGDS_VMID14_BASE, mmGDS_VMID14_SIZE, mmGDS_GWS_VMID14, mmGDS_OA_VMID14},
112 	{mmGDS_VMID15_BASE, mmGDS_VMID15_SIZE, mmGDS_GWS_VMID15, mmGDS_OA_VMID15}
113 };
114 
115 static const u32 spectre_rlc_save_restore_register_list[] = {
116 	(0x0e00 << 16) | (0xc12c >> 2),
117 	0x00000000,
118 	(0x0e00 << 16) | (0xc140 >> 2),
119 	0x00000000,
120 	(0x0e00 << 16) | (0xc150 >> 2),
121 	0x00000000,
122 	(0x0e00 << 16) | (0xc15c >> 2),
123 	0x00000000,
124 	(0x0e00 << 16) | (0xc168 >> 2),
125 	0x00000000,
126 	(0x0e00 << 16) | (0xc170 >> 2),
127 	0x00000000,
128 	(0x0e00 << 16) | (0xc178 >> 2),
129 	0x00000000,
130 	(0x0e00 << 16) | (0xc204 >> 2),
131 	0x00000000,
132 	(0x0e00 << 16) | (0xc2b4 >> 2),
133 	0x00000000,
134 	(0x0e00 << 16) | (0xc2b8 >> 2),
135 	0x00000000,
136 	(0x0e00 << 16) | (0xc2bc >> 2),
137 	0x00000000,
138 	(0x0e00 << 16) | (0xc2c0 >> 2),
139 	0x00000000,
140 	(0x0e00 << 16) | (0x8228 >> 2),
141 	0x00000000,
142 	(0x0e00 << 16) | (0x829c >> 2),
143 	0x00000000,
144 	(0x0e00 << 16) | (0x869c >> 2),
145 	0x00000000,
146 	(0x0600 << 16) | (0x98f4 >> 2),
147 	0x00000000,
148 	(0x0e00 << 16) | (0x98f8 >> 2),
149 	0x00000000,
150 	(0x0e00 << 16) | (0x9900 >> 2),
151 	0x00000000,
152 	(0x0e00 << 16) | (0xc260 >> 2),
153 	0x00000000,
154 	(0x0e00 << 16) | (0x90e8 >> 2),
155 	0x00000000,
156 	(0x0e00 << 16) | (0x3c000 >> 2),
157 	0x00000000,
158 	(0x0e00 << 16) | (0x3c00c >> 2),
159 	0x00000000,
160 	(0x0e00 << 16) | (0x8c1c >> 2),
161 	0x00000000,
162 	(0x0e00 << 16) | (0x9700 >> 2),
163 	0x00000000,
164 	(0x0e00 << 16) | (0xcd20 >> 2),
165 	0x00000000,
166 	(0x4e00 << 16) | (0xcd20 >> 2),
167 	0x00000000,
168 	(0x5e00 << 16) | (0xcd20 >> 2),
169 	0x00000000,
170 	(0x6e00 << 16) | (0xcd20 >> 2),
171 	0x00000000,
172 	(0x7e00 << 16) | (0xcd20 >> 2),
173 	0x00000000,
174 	(0x8e00 << 16) | (0xcd20 >> 2),
175 	0x00000000,
176 	(0x9e00 << 16) | (0xcd20 >> 2),
177 	0x00000000,
178 	(0xae00 << 16) | (0xcd20 >> 2),
179 	0x00000000,
180 	(0xbe00 << 16) | (0xcd20 >> 2),
181 	0x00000000,
182 	(0x0e00 << 16) | (0x89bc >> 2),
183 	0x00000000,
184 	(0x0e00 << 16) | (0x8900 >> 2),
185 	0x00000000,
186 	0x3,
187 	(0x0e00 << 16) | (0xc130 >> 2),
188 	0x00000000,
189 	(0x0e00 << 16) | (0xc134 >> 2),
190 	0x00000000,
191 	(0x0e00 << 16) | (0xc1fc >> 2),
192 	0x00000000,
193 	(0x0e00 << 16) | (0xc208 >> 2),
194 	0x00000000,
195 	(0x0e00 << 16) | (0xc264 >> 2),
196 	0x00000000,
197 	(0x0e00 << 16) | (0xc268 >> 2),
198 	0x00000000,
199 	(0x0e00 << 16) | (0xc26c >> 2),
200 	0x00000000,
201 	(0x0e00 << 16) | (0xc270 >> 2),
202 	0x00000000,
203 	(0x0e00 << 16) | (0xc274 >> 2),
204 	0x00000000,
205 	(0x0e00 << 16) | (0xc278 >> 2),
206 	0x00000000,
207 	(0x0e00 << 16) | (0xc27c >> 2),
208 	0x00000000,
209 	(0x0e00 << 16) | (0xc280 >> 2),
210 	0x00000000,
211 	(0x0e00 << 16) | (0xc284 >> 2),
212 	0x00000000,
213 	(0x0e00 << 16) | (0xc288 >> 2),
214 	0x00000000,
215 	(0x0e00 << 16) | (0xc28c >> 2),
216 	0x00000000,
217 	(0x0e00 << 16) | (0xc290 >> 2),
218 	0x00000000,
219 	(0x0e00 << 16) | (0xc294 >> 2),
220 	0x00000000,
221 	(0x0e00 << 16) | (0xc298 >> 2),
222 	0x00000000,
223 	(0x0e00 << 16) | (0xc29c >> 2),
224 	0x00000000,
225 	(0x0e00 << 16) | (0xc2a0 >> 2),
226 	0x00000000,
227 	(0x0e00 << 16) | (0xc2a4 >> 2),
228 	0x00000000,
229 	(0x0e00 << 16) | (0xc2a8 >> 2),
230 	0x00000000,
231 	(0x0e00 << 16) | (0xc2ac  >> 2),
232 	0x00000000,
233 	(0x0e00 << 16) | (0xc2b0 >> 2),
234 	0x00000000,
235 	(0x0e00 << 16) | (0x301d0 >> 2),
236 	0x00000000,
237 	(0x0e00 << 16) | (0x30238 >> 2),
238 	0x00000000,
239 	(0x0e00 << 16) | (0x30250 >> 2),
240 	0x00000000,
241 	(0x0e00 << 16) | (0x30254 >> 2),
242 	0x00000000,
243 	(0x0e00 << 16) | (0x30258 >> 2),
244 	0x00000000,
245 	(0x0e00 << 16) | (0x3025c >> 2),
246 	0x00000000,
247 	(0x4e00 << 16) | (0xc900 >> 2),
248 	0x00000000,
249 	(0x5e00 << 16) | (0xc900 >> 2),
250 	0x00000000,
251 	(0x6e00 << 16) | (0xc900 >> 2),
252 	0x00000000,
253 	(0x7e00 << 16) | (0xc900 >> 2),
254 	0x00000000,
255 	(0x8e00 << 16) | (0xc900 >> 2),
256 	0x00000000,
257 	(0x9e00 << 16) | (0xc900 >> 2),
258 	0x00000000,
259 	(0xae00 << 16) | (0xc900 >> 2),
260 	0x00000000,
261 	(0xbe00 << 16) | (0xc900 >> 2),
262 	0x00000000,
263 	(0x4e00 << 16) | (0xc904 >> 2),
264 	0x00000000,
265 	(0x5e00 << 16) | (0xc904 >> 2),
266 	0x00000000,
267 	(0x6e00 << 16) | (0xc904 >> 2),
268 	0x00000000,
269 	(0x7e00 << 16) | (0xc904 >> 2),
270 	0x00000000,
271 	(0x8e00 << 16) | (0xc904 >> 2),
272 	0x00000000,
273 	(0x9e00 << 16) | (0xc904 >> 2),
274 	0x00000000,
275 	(0xae00 << 16) | (0xc904 >> 2),
276 	0x00000000,
277 	(0xbe00 << 16) | (0xc904 >> 2),
278 	0x00000000,
279 	(0x4e00 << 16) | (0xc908 >> 2),
280 	0x00000000,
281 	(0x5e00 << 16) | (0xc908 >> 2),
282 	0x00000000,
283 	(0x6e00 << 16) | (0xc908 >> 2),
284 	0x00000000,
285 	(0x7e00 << 16) | (0xc908 >> 2),
286 	0x00000000,
287 	(0x8e00 << 16) | (0xc908 >> 2),
288 	0x00000000,
289 	(0x9e00 << 16) | (0xc908 >> 2),
290 	0x00000000,
291 	(0xae00 << 16) | (0xc908 >> 2),
292 	0x00000000,
293 	(0xbe00 << 16) | (0xc908 >> 2),
294 	0x00000000,
295 	(0x4e00 << 16) | (0xc90c >> 2),
296 	0x00000000,
297 	(0x5e00 << 16) | (0xc90c >> 2),
298 	0x00000000,
299 	(0x6e00 << 16) | (0xc90c >> 2),
300 	0x00000000,
301 	(0x7e00 << 16) | (0xc90c >> 2),
302 	0x00000000,
303 	(0x8e00 << 16) | (0xc90c >> 2),
304 	0x00000000,
305 	(0x9e00 << 16) | (0xc90c >> 2),
306 	0x00000000,
307 	(0xae00 << 16) | (0xc90c >> 2),
308 	0x00000000,
309 	(0xbe00 << 16) | (0xc90c >> 2),
310 	0x00000000,
311 	(0x4e00 << 16) | (0xc910 >> 2),
312 	0x00000000,
313 	(0x5e00 << 16) | (0xc910 >> 2),
314 	0x00000000,
315 	(0x6e00 << 16) | (0xc910 >> 2),
316 	0x00000000,
317 	(0x7e00 << 16) | (0xc910 >> 2),
318 	0x00000000,
319 	(0x8e00 << 16) | (0xc910 >> 2),
320 	0x00000000,
321 	(0x9e00 << 16) | (0xc910 >> 2),
322 	0x00000000,
323 	(0xae00 << 16) | (0xc910 >> 2),
324 	0x00000000,
325 	(0xbe00 << 16) | (0xc910 >> 2),
326 	0x00000000,
327 	(0x0e00 << 16) | (0xc99c >> 2),
328 	0x00000000,
329 	(0x0e00 << 16) | (0x9834 >> 2),
330 	0x00000000,
331 	(0x0000 << 16) | (0x30f00 >> 2),
332 	0x00000000,
333 	(0x0001 << 16) | (0x30f00 >> 2),
334 	0x00000000,
335 	(0x0000 << 16) | (0x30f04 >> 2),
336 	0x00000000,
337 	(0x0001 << 16) | (0x30f04 >> 2),
338 	0x00000000,
339 	(0x0000 << 16) | (0x30f08 >> 2),
340 	0x00000000,
341 	(0x0001 << 16) | (0x30f08 >> 2),
342 	0x00000000,
343 	(0x0000 << 16) | (0x30f0c >> 2),
344 	0x00000000,
345 	(0x0001 << 16) | (0x30f0c >> 2),
346 	0x00000000,
347 	(0x0600 << 16) | (0x9b7c >> 2),
348 	0x00000000,
349 	(0x0e00 << 16) | (0x8a14 >> 2),
350 	0x00000000,
351 	(0x0e00 << 16) | (0x8a18 >> 2),
352 	0x00000000,
353 	(0x0600 << 16) | (0x30a00 >> 2),
354 	0x00000000,
355 	(0x0e00 << 16) | (0x8bf0 >> 2),
356 	0x00000000,
357 	(0x0e00 << 16) | (0x8bcc >> 2),
358 	0x00000000,
359 	(0x0e00 << 16) | (0x8b24 >> 2),
360 	0x00000000,
361 	(0x0e00 << 16) | (0x30a04 >> 2),
362 	0x00000000,
363 	(0x0600 << 16) | (0x30a10 >> 2),
364 	0x00000000,
365 	(0x0600 << 16) | (0x30a14 >> 2),
366 	0x00000000,
367 	(0x0600 << 16) | (0x30a18 >> 2),
368 	0x00000000,
369 	(0x0600 << 16) | (0x30a2c >> 2),
370 	0x00000000,
371 	(0x0e00 << 16) | (0xc700 >> 2),
372 	0x00000000,
373 	(0x0e00 << 16) | (0xc704 >> 2),
374 	0x00000000,
375 	(0x0e00 << 16) | (0xc708 >> 2),
376 	0x00000000,
377 	(0x0e00 << 16) | (0xc768 >> 2),
378 	0x00000000,
379 	(0x0400 << 16) | (0xc770 >> 2),
380 	0x00000000,
381 	(0x0400 << 16) | (0xc774 >> 2),
382 	0x00000000,
383 	(0x0400 << 16) | (0xc778 >> 2),
384 	0x00000000,
385 	(0x0400 << 16) | (0xc77c >> 2),
386 	0x00000000,
387 	(0x0400 << 16) | (0xc780 >> 2),
388 	0x00000000,
389 	(0x0400 << 16) | (0xc784 >> 2),
390 	0x00000000,
391 	(0x0400 << 16) | (0xc788 >> 2),
392 	0x00000000,
393 	(0x0400 << 16) | (0xc78c >> 2),
394 	0x00000000,
395 	(0x0400 << 16) | (0xc798 >> 2),
396 	0x00000000,
397 	(0x0400 << 16) | (0xc79c >> 2),
398 	0x00000000,
399 	(0x0400 << 16) | (0xc7a0 >> 2),
400 	0x00000000,
401 	(0x0400 << 16) | (0xc7a4 >> 2),
402 	0x00000000,
403 	(0x0400 << 16) | (0xc7a8 >> 2),
404 	0x00000000,
405 	(0x0400 << 16) | (0xc7ac >> 2),
406 	0x00000000,
407 	(0x0400 << 16) | (0xc7b0 >> 2),
408 	0x00000000,
409 	(0x0400 << 16) | (0xc7b4 >> 2),
410 	0x00000000,
411 	(0x0e00 << 16) | (0x9100 >> 2),
412 	0x00000000,
413 	(0x0e00 << 16) | (0x3c010 >> 2),
414 	0x00000000,
415 	(0x0e00 << 16) | (0x92a8 >> 2),
416 	0x00000000,
417 	(0x0e00 << 16) | (0x92ac >> 2),
418 	0x00000000,
419 	(0x0e00 << 16) | (0x92b4 >> 2),
420 	0x00000000,
421 	(0x0e00 << 16) | (0x92b8 >> 2),
422 	0x00000000,
423 	(0x0e00 << 16) | (0x92bc >> 2),
424 	0x00000000,
425 	(0x0e00 << 16) | (0x92c0 >> 2),
426 	0x00000000,
427 	(0x0e00 << 16) | (0x92c4 >> 2),
428 	0x00000000,
429 	(0x0e00 << 16) | (0x92c8 >> 2),
430 	0x00000000,
431 	(0x0e00 << 16) | (0x92cc >> 2),
432 	0x00000000,
433 	(0x0e00 << 16) | (0x92d0 >> 2),
434 	0x00000000,
435 	(0x0e00 << 16) | (0x8c00 >> 2),
436 	0x00000000,
437 	(0x0e00 << 16) | (0x8c04 >> 2),
438 	0x00000000,
439 	(0x0e00 << 16) | (0x8c20 >> 2),
440 	0x00000000,
441 	(0x0e00 << 16) | (0x8c38 >> 2),
442 	0x00000000,
443 	(0x0e00 << 16) | (0x8c3c >> 2),
444 	0x00000000,
445 	(0x0e00 << 16) | (0xae00 >> 2),
446 	0x00000000,
447 	(0x0e00 << 16) | (0x9604 >> 2),
448 	0x00000000,
449 	(0x0e00 << 16) | (0xac08 >> 2),
450 	0x00000000,
451 	(0x0e00 << 16) | (0xac0c >> 2),
452 	0x00000000,
453 	(0x0e00 << 16) | (0xac10 >> 2),
454 	0x00000000,
455 	(0x0e00 << 16) | (0xac14 >> 2),
456 	0x00000000,
457 	(0x0e00 << 16) | (0xac58 >> 2),
458 	0x00000000,
459 	(0x0e00 << 16) | (0xac68 >> 2),
460 	0x00000000,
461 	(0x0e00 << 16) | (0xac6c >> 2),
462 	0x00000000,
463 	(0x0e00 << 16) | (0xac70 >> 2),
464 	0x00000000,
465 	(0x0e00 << 16) | (0xac74 >> 2),
466 	0x00000000,
467 	(0x0e00 << 16) | (0xac78 >> 2),
468 	0x00000000,
469 	(0x0e00 << 16) | (0xac7c >> 2),
470 	0x00000000,
471 	(0x0e00 << 16) | (0xac80 >> 2),
472 	0x00000000,
473 	(0x0e00 << 16) | (0xac84 >> 2),
474 	0x00000000,
475 	(0x0e00 << 16) | (0xac88 >> 2),
476 	0x00000000,
477 	(0x0e00 << 16) | (0xac8c >> 2),
478 	0x00000000,
479 	(0x0e00 << 16) | (0x970c >> 2),
480 	0x00000000,
481 	(0x0e00 << 16) | (0x9714 >> 2),
482 	0x00000000,
483 	(0x0e00 << 16) | (0x9718 >> 2),
484 	0x00000000,
485 	(0x0e00 << 16) | (0x971c >> 2),
486 	0x00000000,
487 	(0x0e00 << 16) | (0x31068 >> 2),
488 	0x00000000,
489 	(0x4e00 << 16) | (0x31068 >> 2),
490 	0x00000000,
491 	(0x5e00 << 16) | (0x31068 >> 2),
492 	0x00000000,
493 	(0x6e00 << 16) | (0x31068 >> 2),
494 	0x00000000,
495 	(0x7e00 << 16) | (0x31068 >> 2),
496 	0x00000000,
497 	(0x8e00 << 16) | (0x31068 >> 2),
498 	0x00000000,
499 	(0x9e00 << 16) | (0x31068 >> 2),
500 	0x00000000,
501 	(0xae00 << 16) | (0x31068 >> 2),
502 	0x00000000,
503 	(0xbe00 << 16) | (0x31068 >> 2),
504 	0x00000000,
505 	(0x0e00 << 16) | (0xcd10 >> 2),
506 	0x00000000,
507 	(0x0e00 << 16) | (0xcd14 >> 2),
508 	0x00000000,
509 	(0x0e00 << 16) | (0x88b0 >> 2),
510 	0x00000000,
511 	(0x0e00 << 16) | (0x88b4 >> 2),
512 	0x00000000,
513 	(0x0e00 << 16) | (0x88b8 >> 2),
514 	0x00000000,
515 	(0x0e00 << 16) | (0x88bc >> 2),
516 	0x00000000,
517 	(0x0400 << 16) | (0x89c0 >> 2),
518 	0x00000000,
519 	(0x0e00 << 16) | (0x88c4 >> 2),
520 	0x00000000,
521 	(0x0e00 << 16) | (0x88c8 >> 2),
522 	0x00000000,
523 	(0x0e00 << 16) | (0x88d0 >> 2),
524 	0x00000000,
525 	(0x0e00 << 16) | (0x88d4 >> 2),
526 	0x00000000,
527 	(0x0e00 << 16) | (0x88d8 >> 2),
528 	0x00000000,
529 	(0x0e00 << 16) | (0x8980 >> 2),
530 	0x00000000,
531 	(0x0e00 << 16) | (0x30938 >> 2),
532 	0x00000000,
533 	(0x0e00 << 16) | (0x3093c >> 2),
534 	0x00000000,
535 	(0x0e00 << 16) | (0x30940 >> 2),
536 	0x00000000,
537 	(0x0e00 << 16) | (0x89a0 >> 2),
538 	0x00000000,
539 	(0x0e00 << 16) | (0x30900 >> 2),
540 	0x00000000,
541 	(0x0e00 << 16) | (0x30904 >> 2),
542 	0x00000000,
543 	(0x0e00 << 16) | (0x89b4 >> 2),
544 	0x00000000,
545 	(0x0e00 << 16) | (0x3c210 >> 2),
546 	0x00000000,
547 	(0x0e00 << 16) | (0x3c214 >> 2),
548 	0x00000000,
549 	(0x0e00 << 16) | (0x3c218 >> 2),
550 	0x00000000,
551 	(0x0e00 << 16) | (0x8904 >> 2),
552 	0x00000000,
553 	0x5,
554 	(0x0e00 << 16) | (0x8c28 >> 2),
555 	(0x0e00 << 16) | (0x8c2c >> 2),
556 	(0x0e00 << 16) | (0x8c30 >> 2),
557 	(0x0e00 << 16) | (0x8c34 >> 2),
558 	(0x0e00 << 16) | (0x9600 >> 2),
559 };
560 
561 static const u32 kalindi_rlc_save_restore_register_list[] = {
562 	(0x0e00 << 16) | (0xc12c >> 2),
563 	0x00000000,
564 	(0x0e00 << 16) | (0xc140 >> 2),
565 	0x00000000,
566 	(0x0e00 << 16) | (0xc150 >> 2),
567 	0x00000000,
568 	(0x0e00 << 16) | (0xc15c >> 2),
569 	0x00000000,
570 	(0x0e00 << 16) | (0xc168 >> 2),
571 	0x00000000,
572 	(0x0e00 << 16) | (0xc170 >> 2),
573 	0x00000000,
574 	(0x0e00 << 16) | (0xc204 >> 2),
575 	0x00000000,
576 	(0x0e00 << 16) | (0xc2b4 >> 2),
577 	0x00000000,
578 	(0x0e00 << 16) | (0xc2b8 >> 2),
579 	0x00000000,
580 	(0x0e00 << 16) | (0xc2bc >> 2),
581 	0x00000000,
582 	(0x0e00 << 16) | (0xc2c0 >> 2),
583 	0x00000000,
584 	(0x0e00 << 16) | (0x8228 >> 2),
585 	0x00000000,
586 	(0x0e00 << 16) | (0x829c >> 2),
587 	0x00000000,
588 	(0x0e00 << 16) | (0x869c >> 2),
589 	0x00000000,
590 	(0x0600 << 16) | (0x98f4 >> 2),
591 	0x00000000,
592 	(0x0e00 << 16) | (0x98f8 >> 2),
593 	0x00000000,
594 	(0x0e00 << 16) | (0x9900 >> 2),
595 	0x00000000,
596 	(0x0e00 << 16) | (0xc260 >> 2),
597 	0x00000000,
598 	(0x0e00 << 16) | (0x90e8 >> 2),
599 	0x00000000,
600 	(0x0e00 << 16) | (0x3c000 >> 2),
601 	0x00000000,
602 	(0x0e00 << 16) | (0x3c00c >> 2),
603 	0x00000000,
604 	(0x0e00 << 16) | (0x8c1c >> 2),
605 	0x00000000,
606 	(0x0e00 << 16) | (0x9700 >> 2),
607 	0x00000000,
608 	(0x0e00 << 16) | (0xcd20 >> 2),
609 	0x00000000,
610 	(0x4e00 << 16) | (0xcd20 >> 2),
611 	0x00000000,
612 	(0x5e00 << 16) | (0xcd20 >> 2),
613 	0x00000000,
614 	(0x6e00 << 16) | (0xcd20 >> 2),
615 	0x00000000,
616 	(0x7e00 << 16) | (0xcd20 >> 2),
617 	0x00000000,
618 	(0x0e00 << 16) | (0x89bc >> 2),
619 	0x00000000,
620 	(0x0e00 << 16) | (0x8900 >> 2),
621 	0x00000000,
622 	0x3,
623 	(0x0e00 << 16) | (0xc130 >> 2),
624 	0x00000000,
625 	(0x0e00 << 16) | (0xc134 >> 2),
626 	0x00000000,
627 	(0x0e00 << 16) | (0xc1fc >> 2),
628 	0x00000000,
629 	(0x0e00 << 16) | (0xc208 >> 2),
630 	0x00000000,
631 	(0x0e00 << 16) | (0xc264 >> 2),
632 	0x00000000,
633 	(0x0e00 << 16) | (0xc268 >> 2),
634 	0x00000000,
635 	(0x0e00 << 16) | (0xc26c >> 2),
636 	0x00000000,
637 	(0x0e00 << 16) | (0xc270 >> 2),
638 	0x00000000,
639 	(0x0e00 << 16) | (0xc274 >> 2),
640 	0x00000000,
641 	(0x0e00 << 16) | (0xc28c >> 2),
642 	0x00000000,
643 	(0x0e00 << 16) | (0xc290 >> 2),
644 	0x00000000,
645 	(0x0e00 << 16) | (0xc294 >> 2),
646 	0x00000000,
647 	(0x0e00 << 16) | (0xc298 >> 2),
648 	0x00000000,
649 	(0x0e00 << 16) | (0xc2a0 >> 2),
650 	0x00000000,
651 	(0x0e00 << 16) | (0xc2a4 >> 2),
652 	0x00000000,
653 	(0x0e00 << 16) | (0xc2a8 >> 2),
654 	0x00000000,
655 	(0x0e00 << 16) | (0xc2ac >> 2),
656 	0x00000000,
657 	(0x0e00 << 16) | (0x301d0 >> 2),
658 	0x00000000,
659 	(0x0e00 << 16) | (0x30238 >> 2),
660 	0x00000000,
661 	(0x0e00 << 16) | (0x30250 >> 2),
662 	0x00000000,
663 	(0x0e00 << 16) | (0x30254 >> 2),
664 	0x00000000,
665 	(0x0e00 << 16) | (0x30258 >> 2),
666 	0x00000000,
667 	(0x0e00 << 16) | (0x3025c >> 2),
668 	0x00000000,
669 	(0x4e00 << 16) | (0xc900 >> 2),
670 	0x00000000,
671 	(0x5e00 << 16) | (0xc900 >> 2),
672 	0x00000000,
673 	(0x6e00 << 16) | (0xc900 >> 2),
674 	0x00000000,
675 	(0x7e00 << 16) | (0xc900 >> 2),
676 	0x00000000,
677 	(0x4e00 << 16) | (0xc904 >> 2),
678 	0x00000000,
679 	(0x5e00 << 16) | (0xc904 >> 2),
680 	0x00000000,
681 	(0x6e00 << 16) | (0xc904 >> 2),
682 	0x00000000,
683 	(0x7e00 << 16) | (0xc904 >> 2),
684 	0x00000000,
685 	(0x4e00 << 16) | (0xc908 >> 2),
686 	0x00000000,
687 	(0x5e00 << 16) | (0xc908 >> 2),
688 	0x00000000,
689 	(0x6e00 << 16) | (0xc908 >> 2),
690 	0x00000000,
691 	(0x7e00 << 16) | (0xc908 >> 2),
692 	0x00000000,
693 	(0x4e00 << 16) | (0xc90c >> 2),
694 	0x00000000,
695 	(0x5e00 << 16) | (0xc90c >> 2),
696 	0x00000000,
697 	(0x6e00 << 16) | (0xc90c >> 2),
698 	0x00000000,
699 	(0x7e00 << 16) | (0xc90c >> 2),
700 	0x00000000,
701 	(0x4e00 << 16) | (0xc910 >> 2),
702 	0x00000000,
703 	(0x5e00 << 16) | (0xc910 >> 2),
704 	0x00000000,
705 	(0x6e00 << 16) | (0xc910 >> 2),
706 	0x00000000,
707 	(0x7e00 << 16) | (0xc910 >> 2),
708 	0x00000000,
709 	(0x0e00 << 16) | (0xc99c >> 2),
710 	0x00000000,
711 	(0x0e00 << 16) | (0x9834 >> 2),
712 	0x00000000,
713 	(0x0000 << 16) | (0x30f00 >> 2),
714 	0x00000000,
715 	(0x0000 << 16) | (0x30f04 >> 2),
716 	0x00000000,
717 	(0x0000 << 16) | (0x30f08 >> 2),
718 	0x00000000,
719 	(0x0000 << 16) | (0x30f0c >> 2),
720 	0x00000000,
721 	(0x0600 << 16) | (0x9b7c >> 2),
722 	0x00000000,
723 	(0x0e00 << 16) | (0x8a14 >> 2),
724 	0x00000000,
725 	(0x0e00 << 16) | (0x8a18 >> 2),
726 	0x00000000,
727 	(0x0600 << 16) | (0x30a00 >> 2),
728 	0x00000000,
729 	(0x0e00 << 16) | (0x8bf0 >> 2),
730 	0x00000000,
731 	(0x0e00 << 16) | (0x8bcc >> 2),
732 	0x00000000,
733 	(0x0e00 << 16) | (0x8b24 >> 2),
734 	0x00000000,
735 	(0x0e00 << 16) | (0x30a04 >> 2),
736 	0x00000000,
737 	(0x0600 << 16) | (0x30a10 >> 2),
738 	0x00000000,
739 	(0x0600 << 16) | (0x30a14 >> 2),
740 	0x00000000,
741 	(0x0600 << 16) | (0x30a18 >> 2),
742 	0x00000000,
743 	(0x0600 << 16) | (0x30a2c >> 2),
744 	0x00000000,
745 	(0x0e00 << 16) | (0xc700 >> 2),
746 	0x00000000,
747 	(0x0e00 << 16) | (0xc704 >> 2),
748 	0x00000000,
749 	(0x0e00 << 16) | (0xc708 >> 2),
750 	0x00000000,
751 	(0x0e00 << 16) | (0xc768 >> 2),
752 	0x00000000,
753 	(0x0400 << 16) | (0xc770 >> 2),
754 	0x00000000,
755 	(0x0400 << 16) | (0xc774 >> 2),
756 	0x00000000,
757 	(0x0400 << 16) | (0xc798 >> 2),
758 	0x00000000,
759 	(0x0400 << 16) | (0xc79c >> 2),
760 	0x00000000,
761 	(0x0e00 << 16) | (0x9100 >> 2),
762 	0x00000000,
763 	(0x0e00 << 16) | (0x3c010 >> 2),
764 	0x00000000,
765 	(0x0e00 << 16) | (0x8c00 >> 2),
766 	0x00000000,
767 	(0x0e00 << 16) | (0x8c04 >> 2),
768 	0x00000000,
769 	(0x0e00 << 16) | (0x8c20 >> 2),
770 	0x00000000,
771 	(0x0e00 << 16) | (0x8c38 >> 2),
772 	0x00000000,
773 	(0x0e00 << 16) | (0x8c3c >> 2),
774 	0x00000000,
775 	(0x0e00 << 16) | (0xae00 >> 2),
776 	0x00000000,
777 	(0x0e00 << 16) | (0x9604 >> 2),
778 	0x00000000,
779 	(0x0e00 << 16) | (0xac08 >> 2),
780 	0x00000000,
781 	(0x0e00 << 16) | (0xac0c >> 2),
782 	0x00000000,
783 	(0x0e00 << 16) | (0xac10 >> 2),
784 	0x00000000,
785 	(0x0e00 << 16) | (0xac14 >> 2),
786 	0x00000000,
787 	(0x0e00 << 16) | (0xac58 >> 2),
788 	0x00000000,
789 	(0x0e00 << 16) | (0xac68 >> 2),
790 	0x00000000,
791 	(0x0e00 << 16) | (0xac6c >> 2),
792 	0x00000000,
793 	(0x0e00 << 16) | (0xac70 >> 2),
794 	0x00000000,
795 	(0x0e00 << 16) | (0xac74 >> 2),
796 	0x00000000,
797 	(0x0e00 << 16) | (0xac78 >> 2),
798 	0x00000000,
799 	(0x0e00 << 16) | (0xac7c >> 2),
800 	0x00000000,
801 	(0x0e00 << 16) | (0xac80 >> 2),
802 	0x00000000,
803 	(0x0e00 << 16) | (0xac84 >> 2),
804 	0x00000000,
805 	(0x0e00 << 16) | (0xac88 >> 2),
806 	0x00000000,
807 	(0x0e00 << 16) | (0xac8c >> 2),
808 	0x00000000,
809 	(0x0e00 << 16) | (0x970c >> 2),
810 	0x00000000,
811 	(0x0e00 << 16) | (0x9714 >> 2),
812 	0x00000000,
813 	(0x0e00 << 16) | (0x9718 >> 2),
814 	0x00000000,
815 	(0x0e00 << 16) | (0x971c >> 2),
816 	0x00000000,
817 	(0x0e00 << 16) | (0x31068 >> 2),
818 	0x00000000,
819 	(0x4e00 << 16) | (0x31068 >> 2),
820 	0x00000000,
821 	(0x5e00 << 16) | (0x31068 >> 2),
822 	0x00000000,
823 	(0x6e00 << 16) | (0x31068 >> 2),
824 	0x00000000,
825 	(0x7e00 << 16) | (0x31068 >> 2),
826 	0x00000000,
827 	(0x0e00 << 16) | (0xcd10 >> 2),
828 	0x00000000,
829 	(0x0e00 << 16) | (0xcd14 >> 2),
830 	0x00000000,
831 	(0x0e00 << 16) | (0x88b0 >> 2),
832 	0x00000000,
833 	(0x0e00 << 16) | (0x88b4 >> 2),
834 	0x00000000,
835 	(0x0e00 << 16) | (0x88b8 >> 2),
836 	0x00000000,
837 	(0x0e00 << 16) | (0x88bc >> 2),
838 	0x00000000,
839 	(0x0400 << 16) | (0x89c0 >> 2),
840 	0x00000000,
841 	(0x0e00 << 16) | (0x88c4 >> 2),
842 	0x00000000,
843 	(0x0e00 << 16) | (0x88c8 >> 2),
844 	0x00000000,
845 	(0x0e00 << 16) | (0x88d0 >> 2),
846 	0x00000000,
847 	(0x0e00 << 16) | (0x88d4 >> 2),
848 	0x00000000,
849 	(0x0e00 << 16) | (0x88d8 >> 2),
850 	0x00000000,
851 	(0x0e00 << 16) | (0x8980 >> 2),
852 	0x00000000,
853 	(0x0e00 << 16) | (0x30938 >> 2),
854 	0x00000000,
855 	(0x0e00 << 16) | (0x3093c >> 2),
856 	0x00000000,
857 	(0x0e00 << 16) | (0x30940 >> 2),
858 	0x00000000,
859 	(0x0e00 << 16) | (0x89a0 >> 2),
860 	0x00000000,
861 	(0x0e00 << 16) | (0x30900 >> 2),
862 	0x00000000,
863 	(0x0e00 << 16) | (0x30904 >> 2),
864 	0x00000000,
865 	(0x0e00 << 16) | (0x89b4 >> 2),
866 	0x00000000,
867 	(0x0e00 << 16) | (0x3e1fc >> 2),
868 	0x00000000,
869 	(0x0e00 << 16) | (0x3c210 >> 2),
870 	0x00000000,
871 	(0x0e00 << 16) | (0x3c214 >> 2),
872 	0x00000000,
873 	(0x0e00 << 16) | (0x3c218 >> 2),
874 	0x00000000,
875 	(0x0e00 << 16) | (0x8904 >> 2),
876 	0x00000000,
877 	0x5,
878 	(0x0e00 << 16) | (0x8c28 >> 2),
879 	(0x0e00 << 16) | (0x8c2c >> 2),
880 	(0x0e00 << 16) | (0x8c30 >> 2),
881 	(0x0e00 << 16) | (0x8c34 >> 2),
882 	(0x0e00 << 16) | (0x9600 >> 2),
883 };
884 
885 static u32 gfx_v7_0_get_csb_size(struct amdgpu_device *adev);
886 static void gfx_v7_0_get_csb_buffer(struct amdgpu_device *adev, u32 *buffer);
887 static void gfx_v7_0_init_pg(struct amdgpu_device *adev);
888 static void gfx_v7_0_get_cu_info(struct amdgpu_device *adev);
889 
890 static void gfx_v7_0_free_microcode(struct amdgpu_device *adev)
891 {
892 	amdgpu_ucode_release(&adev->gfx.pfp_fw);
893 	amdgpu_ucode_release(&adev->gfx.me_fw);
894 	amdgpu_ucode_release(&adev->gfx.ce_fw);
895 	amdgpu_ucode_release(&adev->gfx.mec_fw);
896 	amdgpu_ucode_release(&adev->gfx.mec2_fw);
897 	amdgpu_ucode_release(&adev->gfx.rlc_fw);
898 }
899 
900 /*
901  * Core functions
902  */
903 /**
904  * gfx_v7_0_init_microcode - load ucode images from disk
905  *
906  * @adev: amdgpu_device pointer
907  *
908  * Use the firmware interface to load the ucode images into
909  * the driver (not loaded into hw).
910  * Returns 0 on success, error on failure.
911  */
912 static int gfx_v7_0_init_microcode(struct amdgpu_device *adev)
913 {
914 	const char *chip_name;
915 	int err;
916 
917 	DRM_DEBUG("\n");
918 
919 	switch (adev->asic_type) {
920 	case CHIP_BONAIRE:
921 		chip_name = "bonaire";
922 		break;
923 	case CHIP_HAWAII:
924 		chip_name = "hawaii";
925 		break;
926 	case CHIP_KAVERI:
927 		chip_name = "kaveri";
928 		break;
929 	case CHIP_KABINI:
930 		chip_name = "kabini";
931 		break;
932 	case CHIP_MULLINS:
933 		chip_name = "mullins";
934 		break;
935 	default:
936 		dev_warn(adev->dev, "Unsupported asic_type 0x%08x\n", adev->asic_type);
937 		return -EINVAL;
938 	}
939 
940 	err = amdgpu_ucode_request(adev, &adev->gfx.pfp_fw,
941 				   AMDGPU_UCODE_REQUIRED,
942 				   "amdgpu/%s_pfp.bin", chip_name);
943 	if (err)
944 		goto out;
945 
946 	err = amdgpu_ucode_request(adev, &adev->gfx.me_fw,
947 				   AMDGPU_UCODE_REQUIRED,
948 				   "amdgpu/%s_me.bin", chip_name);
949 	if (err)
950 		goto out;
951 
952 	err = amdgpu_ucode_request(adev, &adev->gfx.ce_fw,
953 				   AMDGPU_UCODE_REQUIRED,
954 				   "amdgpu/%s_ce.bin", chip_name);
955 	if (err)
956 		goto out;
957 
958 	err = amdgpu_ucode_request(adev, &adev->gfx.mec_fw,
959 				   AMDGPU_UCODE_REQUIRED,
960 				   "amdgpu/%s_mec.bin", chip_name);
961 	if (err)
962 		goto out;
963 
964 	if (adev->asic_type == CHIP_KAVERI) {
965 		err = amdgpu_ucode_request(adev, &adev->gfx.mec2_fw,
966 					   AMDGPU_UCODE_REQUIRED,
967 					   "amdgpu/%s_mec2.bin", chip_name);
968 		if (err)
969 			goto out;
970 	}
971 
972 	err = amdgpu_ucode_request(adev, &adev->gfx.rlc_fw,
973 				   AMDGPU_UCODE_REQUIRED,
974 				   "amdgpu/%s_rlc.bin", chip_name);
975 out:
976 	if (err) {
977 		pr_err("gfx7: Failed to load firmware %s gfx firmware\n", chip_name);
978 		gfx_v7_0_free_microcode(adev);
979 	}
980 	return err;
981 }
982 
983 /**
984  * gfx_v7_0_tiling_mode_table_init - init the hw tiling table
985  *
986  * @adev: amdgpu_device pointer
987  *
988  * Starting with SI, the tiling setup is done globally in a
989  * set of 32 tiling modes.  Rather than selecting each set of
990  * parameters per surface as on older asics, we just select
991  * which index in the tiling table we want to use, and the
992  * surface uses those parameters (CIK).
993  */
994 static void gfx_v7_0_tiling_mode_table_init(struct amdgpu_device *adev)
995 {
996 	const u32 num_tile_mode_states =
997 			ARRAY_SIZE(adev->gfx.config.tile_mode_array);
998 	const u32 num_secondary_tile_mode_states =
999 			ARRAY_SIZE(adev->gfx.config.macrotile_mode_array);
1000 	u32 reg_offset, split_equal_to_row_size;
1001 	uint32_t *tile, *macrotile;
1002 
1003 	tile = adev->gfx.config.tile_mode_array;
1004 	macrotile = adev->gfx.config.macrotile_mode_array;
1005 
1006 	switch (adev->gfx.config.mem_row_size_in_kb) {
1007 	case 1:
1008 		split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_1KB;
1009 		break;
1010 	case 2:
1011 	default:
1012 		split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_2KB;
1013 		break;
1014 	case 4:
1015 		split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_4KB;
1016 		break;
1017 	}
1018 
1019 	for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
1020 		tile[reg_offset] = 0;
1021 	for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
1022 		macrotile[reg_offset] = 0;
1023 
1024 	switch (adev->asic_type) {
1025 	case CHIP_BONAIRE:
1026 		tile[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1027 			   PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1028 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
1029 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1030 		tile[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1031 			   PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1032 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
1033 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1034 		tile[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1035 			   PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1036 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1037 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1038 		tile[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1039 			   PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1040 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
1041 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1042 		tile[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1043 			   PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1044 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1045 			   TILE_SPLIT(split_equal_to_row_size));
1046 		tile[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1047 			   PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1048 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1049 		tile[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1050 			   PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1051 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1052 			   TILE_SPLIT(split_equal_to_row_size));
1053 		tile[7] = (TILE_SPLIT(split_equal_to_row_size));
1054 		tile[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
1055 			   PIPE_CONFIG(ADDR_SURF_P4_16x16));
1056 		tile[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1057 			   PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1058 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING));
1059 		tile[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1060 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1061 			    MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
1062 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1063 		tile[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1064 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1065 			    MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
1066 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1067 		tile[12] = (TILE_SPLIT(split_equal_to_row_size));
1068 		tile[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1069 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1070 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING));
1071 		tile[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1072 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1073 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1074 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1075 		tile[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
1076 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1077 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1078 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1079 		tile[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1080 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1081 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1082 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1083 		tile[17] = (TILE_SPLIT(split_equal_to_row_size));
1084 		tile[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
1085 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1086 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1087 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1088 		tile[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
1089 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1090 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING));
1091 		tile[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
1092 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1093 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1094 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1095 		tile[21] =  (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
1096 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1097 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1098 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1099 		tile[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
1100 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1101 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1102 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1103 		tile[23] = (TILE_SPLIT(split_equal_to_row_size));
1104 		tile[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
1105 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1106 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1107 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1108 		tile[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
1109 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1110 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1111 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1112 		tile[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
1113 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1114 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1115 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1116 		tile[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1117 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1118 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING));
1119 		tile[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1120 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1121 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
1122 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1123 		tile[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1124 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1125 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
1126 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1127 		tile[30] = (TILE_SPLIT(split_equal_to_row_size));
1128 
1129 		macrotile[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1130 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1131 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1132 				NUM_BANKS(ADDR_SURF_16_BANK));
1133 		macrotile[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1134 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1135 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1136 				NUM_BANKS(ADDR_SURF_16_BANK));
1137 		macrotile[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1138 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1139 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1140 				NUM_BANKS(ADDR_SURF_16_BANK));
1141 		macrotile[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1142 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1143 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1144 				NUM_BANKS(ADDR_SURF_16_BANK));
1145 		macrotile[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1146 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1147 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1148 				NUM_BANKS(ADDR_SURF_16_BANK));
1149 		macrotile[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1150 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1151 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1152 				NUM_BANKS(ADDR_SURF_8_BANK));
1153 		macrotile[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1154 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1155 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1156 				NUM_BANKS(ADDR_SURF_4_BANK));
1157 		macrotile[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
1158 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
1159 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1160 				NUM_BANKS(ADDR_SURF_16_BANK));
1161 		macrotile[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
1162 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1163 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1164 				NUM_BANKS(ADDR_SURF_16_BANK));
1165 		macrotile[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1166 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1167 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1168 				NUM_BANKS(ADDR_SURF_16_BANK));
1169 		macrotile[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1170 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1171 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1172 				NUM_BANKS(ADDR_SURF_16_BANK));
1173 		macrotile[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1174 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1175 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1176 				NUM_BANKS(ADDR_SURF_16_BANK));
1177 		macrotile[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1178 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1179 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1180 				NUM_BANKS(ADDR_SURF_8_BANK));
1181 		macrotile[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1182 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1183 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1184 				NUM_BANKS(ADDR_SURF_4_BANK));
1185 
1186 		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
1187 			WREG32(mmGB_TILE_MODE0 + reg_offset, tile[reg_offset]);
1188 		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
1189 			if (reg_offset != 7)
1190 				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, macrotile[reg_offset]);
1191 		break;
1192 	case CHIP_HAWAII:
1193 		tile[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1194 			   PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1195 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
1196 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1197 		tile[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1198 			   PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1199 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
1200 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1201 		tile[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1202 			   PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1203 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1204 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1205 		tile[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1206 			   PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1207 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
1208 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1209 		tile[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1210 			   PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1211 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1212 			   TILE_SPLIT(split_equal_to_row_size));
1213 		tile[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1214 			   PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1215 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1216 			   TILE_SPLIT(split_equal_to_row_size));
1217 		tile[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1218 			   PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1219 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1220 			   TILE_SPLIT(split_equal_to_row_size));
1221 		tile[7] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1222 			   PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1223 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1224 			   TILE_SPLIT(split_equal_to_row_size));
1225 		tile[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
1226 			   PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16));
1227 		tile[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1228 			   PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1229 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING));
1230 		tile[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1231 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1232 			    MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
1233 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1234 		tile[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1235 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1236 			    MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
1237 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1238 		tile[12] = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
1239 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1240 			    MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
1241 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1242 		tile[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1243 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1244 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING));
1245 		tile[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1246 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1247 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1248 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1249 		tile[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
1250 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1251 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1252 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1253 		tile[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1254 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1255 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1256 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1257 		tile[17] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1258 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1259 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1260 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1261 		tile[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
1262 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1263 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1264 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1265 		tile[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
1266 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1267 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING));
1268 		tile[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
1269 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1270 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1271 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1272 		tile[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
1273 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1274 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1275 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1276 		tile[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
1277 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1278 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1279 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1280 		tile[23] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
1281 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1282 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1283 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1284 		tile[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
1285 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1286 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1287 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1288 		tile[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
1289 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1290 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1291 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1292 		tile[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
1293 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1294 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1295 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1296 		tile[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1297 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1298 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING));
1299 		tile[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1300 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1301 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
1302 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1303 		tile[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1304 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1305 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
1306 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1307 		tile[30] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1308 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1309 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
1310 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1311 
1312 		macrotile[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1313 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1314 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1315 				NUM_BANKS(ADDR_SURF_16_BANK));
1316 		macrotile[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1317 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1318 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1319 				NUM_BANKS(ADDR_SURF_16_BANK));
1320 		macrotile[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1321 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1322 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1323 				NUM_BANKS(ADDR_SURF_16_BANK));
1324 		macrotile[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1325 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1326 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1327 				NUM_BANKS(ADDR_SURF_16_BANK));
1328 		macrotile[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1329 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1330 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1331 				NUM_BANKS(ADDR_SURF_8_BANK));
1332 		macrotile[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1333 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1334 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1335 				NUM_BANKS(ADDR_SURF_4_BANK));
1336 		macrotile[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1337 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1338 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1339 				NUM_BANKS(ADDR_SURF_4_BANK));
1340 		macrotile[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1341 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1342 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1343 				NUM_BANKS(ADDR_SURF_16_BANK));
1344 		macrotile[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1345 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1346 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1347 				NUM_BANKS(ADDR_SURF_16_BANK));
1348 		macrotile[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1349 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1350 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1351 				NUM_BANKS(ADDR_SURF_16_BANK));
1352 		macrotile[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1353 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1354 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1355 				NUM_BANKS(ADDR_SURF_8_BANK));
1356 		macrotile[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1357 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1358 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1359 				NUM_BANKS(ADDR_SURF_16_BANK));
1360 		macrotile[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1361 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1362 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1363 				NUM_BANKS(ADDR_SURF_8_BANK));
1364 		macrotile[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1365 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1366 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1367 				NUM_BANKS(ADDR_SURF_4_BANK));
1368 
1369 		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
1370 			WREG32(mmGB_TILE_MODE0 + reg_offset, tile[reg_offset]);
1371 		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
1372 			if (reg_offset != 7)
1373 				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, macrotile[reg_offset]);
1374 		break;
1375 	case CHIP_KABINI:
1376 	case CHIP_KAVERI:
1377 	case CHIP_MULLINS:
1378 	default:
1379 		tile[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1380 			   PIPE_CONFIG(ADDR_SURF_P2) |
1381 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
1382 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1383 		tile[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1384 			   PIPE_CONFIG(ADDR_SURF_P2) |
1385 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
1386 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1387 		tile[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1388 			   PIPE_CONFIG(ADDR_SURF_P2) |
1389 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1390 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1391 		tile[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1392 			   PIPE_CONFIG(ADDR_SURF_P2) |
1393 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
1394 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1395 		tile[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1396 			   PIPE_CONFIG(ADDR_SURF_P2) |
1397 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1398 			   TILE_SPLIT(split_equal_to_row_size));
1399 		tile[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1400 			   PIPE_CONFIG(ADDR_SURF_P2) |
1401 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1402 		tile[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1403 			   PIPE_CONFIG(ADDR_SURF_P2) |
1404 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1405 			   TILE_SPLIT(split_equal_to_row_size));
1406 		tile[7] = (TILE_SPLIT(split_equal_to_row_size));
1407 		tile[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
1408 			   PIPE_CONFIG(ADDR_SURF_P2));
1409 		tile[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1410 			   PIPE_CONFIG(ADDR_SURF_P2) |
1411 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING));
1412 		tile[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1413 			    PIPE_CONFIG(ADDR_SURF_P2) |
1414 			    MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
1415 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1416 		tile[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1417 			    PIPE_CONFIG(ADDR_SURF_P2) |
1418 			    MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
1419 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1420 		tile[12] = (TILE_SPLIT(split_equal_to_row_size));
1421 		tile[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1422 			    PIPE_CONFIG(ADDR_SURF_P2) |
1423 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING));
1424 		tile[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1425 			    PIPE_CONFIG(ADDR_SURF_P2) |
1426 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1427 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1428 		tile[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
1429 			    PIPE_CONFIG(ADDR_SURF_P2) |
1430 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1431 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1432 		tile[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1433 			    PIPE_CONFIG(ADDR_SURF_P2) |
1434 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1435 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1436 		tile[17] = (TILE_SPLIT(split_equal_to_row_size));
1437 		tile[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
1438 			    PIPE_CONFIG(ADDR_SURF_P2) |
1439 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1440 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1441 		tile[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
1442 			    PIPE_CONFIG(ADDR_SURF_P2) |
1443 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING));
1444 		tile[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
1445 			    PIPE_CONFIG(ADDR_SURF_P2) |
1446 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1447 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1448 		tile[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
1449 			    PIPE_CONFIG(ADDR_SURF_P2) |
1450 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1451 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1452 		tile[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
1453 			    PIPE_CONFIG(ADDR_SURF_P2) |
1454 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1455 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1456 		tile[23] = (TILE_SPLIT(split_equal_to_row_size));
1457 		tile[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
1458 			    PIPE_CONFIG(ADDR_SURF_P2) |
1459 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1460 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1461 		tile[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
1462 			    PIPE_CONFIG(ADDR_SURF_P2) |
1463 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1464 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1465 		tile[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
1466 			    PIPE_CONFIG(ADDR_SURF_P2) |
1467 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1468 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1469 		tile[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1470 			    PIPE_CONFIG(ADDR_SURF_P2) |
1471 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING));
1472 		tile[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1473 			    PIPE_CONFIG(ADDR_SURF_P2) |
1474 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
1475 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1476 		tile[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1477 			    PIPE_CONFIG(ADDR_SURF_P2) |
1478 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
1479 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1480 		tile[30] = (TILE_SPLIT(split_equal_to_row_size));
1481 
1482 		macrotile[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1483 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1484 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1485 				NUM_BANKS(ADDR_SURF_8_BANK));
1486 		macrotile[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1487 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1488 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1489 				NUM_BANKS(ADDR_SURF_8_BANK));
1490 		macrotile[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1491 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1492 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1493 				NUM_BANKS(ADDR_SURF_8_BANK));
1494 		macrotile[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1495 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1496 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1497 				NUM_BANKS(ADDR_SURF_8_BANK));
1498 		macrotile[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1499 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1500 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1501 				NUM_BANKS(ADDR_SURF_8_BANK));
1502 		macrotile[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1503 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1504 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1505 				NUM_BANKS(ADDR_SURF_8_BANK));
1506 		macrotile[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1507 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1508 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1509 				NUM_BANKS(ADDR_SURF_8_BANK));
1510 		macrotile[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
1511 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
1512 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1513 				NUM_BANKS(ADDR_SURF_16_BANK));
1514 		macrotile[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
1515 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1516 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1517 				NUM_BANKS(ADDR_SURF_16_BANK));
1518 		macrotile[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
1519 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1520 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1521 				NUM_BANKS(ADDR_SURF_16_BANK));
1522 		macrotile[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
1523 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1524 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1525 				NUM_BANKS(ADDR_SURF_16_BANK));
1526 		macrotile[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1527 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1528 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1529 				NUM_BANKS(ADDR_SURF_16_BANK));
1530 		macrotile[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1531 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1532 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1533 				NUM_BANKS(ADDR_SURF_16_BANK));
1534 		macrotile[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1535 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1536 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1537 				NUM_BANKS(ADDR_SURF_8_BANK));
1538 
1539 		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
1540 			WREG32(mmGB_TILE_MODE0 + reg_offset, tile[reg_offset]);
1541 		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
1542 			if (reg_offset != 7)
1543 				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, macrotile[reg_offset]);
1544 		break;
1545 	}
1546 }
1547 
1548 /**
1549  * gfx_v7_0_select_se_sh - select which SE, SH to address
1550  *
1551  * @adev: amdgpu_device pointer
1552  * @se_num: shader engine to address
1553  * @sh_num: sh block to address
1554  * @instance: Certain registers are instanced per SE or SH.
1555  *            0xffffffff means broadcast to all SEs or SHs (CIK).
1556  * @xcc_id: xcc accelerated compute core id
1557  * Select which SE, SH combinations to address.
1558  */
1559 static void gfx_v7_0_select_se_sh(struct amdgpu_device *adev,
1560 				  u32 se_num, u32 sh_num, u32 instance,
1561 				  int xcc_id)
1562 {
1563 	u32 data;
1564 
1565 	if (instance == 0xffffffff)
1566 		data = REG_SET_FIELD(0, GRBM_GFX_INDEX, INSTANCE_BROADCAST_WRITES, 1);
1567 	else
1568 		data = REG_SET_FIELD(0, GRBM_GFX_INDEX, INSTANCE_INDEX, instance);
1569 
1570 	if ((se_num == 0xffffffff) && (sh_num == 0xffffffff))
1571 		data |= GRBM_GFX_INDEX__SH_BROADCAST_WRITES_MASK |
1572 			GRBM_GFX_INDEX__SE_BROADCAST_WRITES_MASK;
1573 	else if (se_num == 0xffffffff)
1574 		data |= GRBM_GFX_INDEX__SE_BROADCAST_WRITES_MASK |
1575 			(sh_num << GRBM_GFX_INDEX__SH_INDEX__SHIFT);
1576 	else if (sh_num == 0xffffffff)
1577 		data |= GRBM_GFX_INDEX__SH_BROADCAST_WRITES_MASK |
1578 			(se_num << GRBM_GFX_INDEX__SE_INDEX__SHIFT);
1579 	else
1580 		data |= (sh_num << GRBM_GFX_INDEX__SH_INDEX__SHIFT) |
1581 			(se_num << GRBM_GFX_INDEX__SE_INDEX__SHIFT);
1582 	WREG32(mmGRBM_GFX_INDEX, data);
1583 }
1584 
1585 /**
1586  * gfx_v7_0_get_rb_active_bitmap - computes the mask of enabled RBs
1587  *
1588  * @adev: amdgpu_device pointer
1589  *
1590  * Calculates the bitmask of enabled RBs (CIK).
1591  * Returns the enabled RB bitmask.
1592  */
1593 static u32 gfx_v7_0_get_rb_active_bitmap(struct amdgpu_device *adev)
1594 {
1595 	u32 data, mask;
1596 
1597 	data = RREG32(mmCC_RB_BACKEND_DISABLE);
1598 	data |= RREG32(mmGC_USER_RB_BACKEND_DISABLE);
1599 
1600 	data &= CC_RB_BACKEND_DISABLE__BACKEND_DISABLE_MASK;
1601 	data >>= GC_USER_RB_BACKEND_DISABLE__BACKEND_DISABLE__SHIFT;
1602 
1603 	mask = amdgpu_gfx_create_bitmask(adev->gfx.config.max_backends_per_se /
1604 					 adev->gfx.config.max_sh_per_se);
1605 
1606 	return (~data) & mask;
1607 }
1608 
1609 static void
1610 gfx_v7_0_raster_config(struct amdgpu_device *adev, u32 *rconf, u32 *rconf1)
1611 {
1612 	switch (adev->asic_type) {
1613 	case CHIP_BONAIRE:
1614 		*rconf |= RB_MAP_PKR0(2) | RB_XSEL2(1) | SE_MAP(2) |
1615 			  SE_XSEL(1) | SE_YSEL(1);
1616 		*rconf1 |= 0x0;
1617 		break;
1618 	case CHIP_HAWAII:
1619 		*rconf |= RB_MAP_PKR0(2) | RB_MAP_PKR1(2) |
1620 			  RB_XSEL2(1) | PKR_MAP(2) | PKR_XSEL(1) |
1621 			  PKR_YSEL(1) | SE_MAP(2) | SE_XSEL(2) |
1622 			  SE_YSEL(3);
1623 		*rconf1 |= SE_PAIR_MAP(2) | SE_PAIR_XSEL(3) |
1624 			   SE_PAIR_YSEL(2);
1625 		break;
1626 	case CHIP_KAVERI:
1627 		*rconf |= RB_MAP_PKR0(2);
1628 		*rconf1 |= 0x0;
1629 		break;
1630 	case CHIP_KABINI:
1631 	case CHIP_MULLINS:
1632 		*rconf |= 0x0;
1633 		*rconf1 |= 0x0;
1634 		break;
1635 	default:
1636 		DRM_ERROR("unknown asic: 0x%x\n", adev->asic_type);
1637 		break;
1638 	}
1639 }
1640 
1641 static void
1642 gfx_v7_0_write_harvested_raster_configs(struct amdgpu_device *adev,
1643 					u32 raster_config, u32 raster_config_1,
1644 					unsigned rb_mask, unsigned num_rb)
1645 {
1646 	unsigned sh_per_se = max_t(unsigned, adev->gfx.config.max_sh_per_se, 1);
1647 	unsigned num_se = max_t(unsigned, adev->gfx.config.max_shader_engines, 1);
1648 	unsigned rb_per_pkr = min_t(unsigned, num_rb / num_se / sh_per_se, 2);
1649 	unsigned rb_per_se = num_rb / num_se;
1650 	unsigned se_mask[4];
1651 	unsigned se;
1652 
1653 	se_mask[0] = ((1 << rb_per_se) - 1) & rb_mask;
1654 	se_mask[1] = (se_mask[0] << rb_per_se) & rb_mask;
1655 	se_mask[2] = (se_mask[1] << rb_per_se) & rb_mask;
1656 	se_mask[3] = (se_mask[2] << rb_per_se) & rb_mask;
1657 
1658 	WARN_ON(!(num_se == 1 || num_se == 2 || num_se == 4));
1659 	WARN_ON(!(sh_per_se == 1 || sh_per_se == 2));
1660 	WARN_ON(!(rb_per_pkr == 1 || rb_per_pkr == 2));
1661 
1662 	if ((num_se > 2) && ((!se_mask[0] && !se_mask[1]) ||
1663 			     (!se_mask[2] && !se_mask[3]))) {
1664 		raster_config_1 &= ~SE_PAIR_MAP_MASK;
1665 
1666 		if (!se_mask[0] && !se_mask[1]) {
1667 			raster_config_1 |=
1668 				SE_PAIR_MAP(RASTER_CONFIG_SE_PAIR_MAP_3);
1669 		} else {
1670 			raster_config_1 |=
1671 				SE_PAIR_MAP(RASTER_CONFIG_SE_PAIR_MAP_0);
1672 		}
1673 	}
1674 
1675 	for (se = 0; se < num_se; se++) {
1676 		unsigned raster_config_se = raster_config;
1677 		unsigned pkr0_mask = ((1 << rb_per_pkr) - 1) << (se * rb_per_se);
1678 		unsigned pkr1_mask = pkr0_mask << rb_per_pkr;
1679 		int idx = (se / 2) * 2;
1680 
1681 		if ((num_se > 1) && (!se_mask[idx] || !se_mask[idx + 1])) {
1682 			raster_config_se &= ~SE_MAP_MASK;
1683 
1684 			if (!se_mask[idx]) {
1685 				raster_config_se |= SE_MAP(RASTER_CONFIG_SE_MAP_3);
1686 			} else {
1687 				raster_config_se |= SE_MAP(RASTER_CONFIG_SE_MAP_0);
1688 			}
1689 		}
1690 
1691 		pkr0_mask &= rb_mask;
1692 		pkr1_mask &= rb_mask;
1693 		if (rb_per_se > 2 && (!pkr0_mask || !pkr1_mask)) {
1694 			raster_config_se &= ~PKR_MAP_MASK;
1695 
1696 			if (!pkr0_mask) {
1697 				raster_config_se |= PKR_MAP(RASTER_CONFIG_PKR_MAP_3);
1698 			} else {
1699 				raster_config_se |= PKR_MAP(RASTER_CONFIG_PKR_MAP_0);
1700 			}
1701 		}
1702 
1703 		if (rb_per_se >= 2) {
1704 			unsigned rb0_mask = 1 << (se * rb_per_se);
1705 			unsigned rb1_mask = rb0_mask << 1;
1706 
1707 			rb0_mask &= rb_mask;
1708 			rb1_mask &= rb_mask;
1709 			if (!rb0_mask || !rb1_mask) {
1710 				raster_config_se &= ~RB_MAP_PKR0_MASK;
1711 
1712 				if (!rb0_mask) {
1713 					raster_config_se |=
1714 						RB_MAP_PKR0(RASTER_CONFIG_RB_MAP_3);
1715 				} else {
1716 					raster_config_se |=
1717 						RB_MAP_PKR0(RASTER_CONFIG_RB_MAP_0);
1718 				}
1719 			}
1720 
1721 			if (rb_per_se > 2) {
1722 				rb0_mask = 1 << (se * rb_per_se + rb_per_pkr);
1723 				rb1_mask = rb0_mask << 1;
1724 				rb0_mask &= rb_mask;
1725 				rb1_mask &= rb_mask;
1726 				if (!rb0_mask || !rb1_mask) {
1727 					raster_config_se &= ~RB_MAP_PKR1_MASK;
1728 
1729 					if (!rb0_mask) {
1730 						raster_config_se |=
1731 							RB_MAP_PKR1(RASTER_CONFIG_RB_MAP_3);
1732 					} else {
1733 						raster_config_se |=
1734 							RB_MAP_PKR1(RASTER_CONFIG_RB_MAP_0);
1735 					}
1736 				}
1737 			}
1738 		}
1739 
1740 		/* GRBM_GFX_INDEX has a different offset on CI+ */
1741 		gfx_v7_0_select_se_sh(adev, se, 0xffffffff, 0xffffffff, 0);
1742 		WREG32(mmPA_SC_RASTER_CONFIG, raster_config_se);
1743 		WREG32(mmPA_SC_RASTER_CONFIG_1, raster_config_1);
1744 	}
1745 
1746 	/* GRBM_GFX_INDEX has a different offset on CI+ */
1747 	gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0);
1748 }
1749 
1750 /**
1751  * gfx_v7_0_setup_rb - setup the RBs on the asic
1752  *
1753  * @adev: amdgpu_device pointer
1754  *
1755  * Configures per-SE/SH RB registers (CIK).
1756  */
1757 static void gfx_v7_0_setup_rb(struct amdgpu_device *adev)
1758 {
1759 	int i, j;
1760 	u32 data;
1761 	u32 raster_config = 0, raster_config_1 = 0;
1762 	u32 active_rbs = 0;
1763 	u32 rb_bitmap_width_per_sh = adev->gfx.config.max_backends_per_se /
1764 					adev->gfx.config.max_sh_per_se;
1765 	unsigned num_rb_pipes;
1766 
1767 	mutex_lock(&adev->grbm_idx_mutex);
1768 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
1769 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
1770 			gfx_v7_0_select_se_sh(adev, i, j, 0xffffffff, 0);
1771 			data = gfx_v7_0_get_rb_active_bitmap(adev);
1772 			active_rbs |= data << ((i * adev->gfx.config.max_sh_per_se + j) *
1773 					       rb_bitmap_width_per_sh);
1774 		}
1775 	}
1776 	gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0);
1777 
1778 	adev->gfx.config.backend_enable_mask = active_rbs;
1779 	adev->gfx.config.num_rbs = hweight32(active_rbs);
1780 
1781 	num_rb_pipes = min_t(unsigned, adev->gfx.config.max_backends_per_se *
1782 			     adev->gfx.config.max_shader_engines, 16);
1783 
1784 	gfx_v7_0_raster_config(adev, &raster_config, &raster_config_1);
1785 
1786 	if (!adev->gfx.config.backend_enable_mask ||
1787 			adev->gfx.config.num_rbs >= num_rb_pipes) {
1788 		WREG32(mmPA_SC_RASTER_CONFIG, raster_config);
1789 		WREG32(mmPA_SC_RASTER_CONFIG_1, raster_config_1);
1790 	} else {
1791 		gfx_v7_0_write_harvested_raster_configs(adev, raster_config, raster_config_1,
1792 							adev->gfx.config.backend_enable_mask,
1793 							num_rb_pipes);
1794 	}
1795 
1796 	/* cache the values for userspace */
1797 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
1798 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
1799 			gfx_v7_0_select_se_sh(adev, i, j, 0xffffffff, 0);
1800 			adev->gfx.config.rb_config[i][j].rb_backend_disable =
1801 				RREG32(mmCC_RB_BACKEND_DISABLE);
1802 			adev->gfx.config.rb_config[i][j].user_rb_backend_disable =
1803 				RREG32(mmGC_USER_RB_BACKEND_DISABLE);
1804 			adev->gfx.config.rb_config[i][j].raster_config =
1805 				RREG32(mmPA_SC_RASTER_CONFIG);
1806 			adev->gfx.config.rb_config[i][j].raster_config_1 =
1807 				RREG32(mmPA_SC_RASTER_CONFIG_1);
1808 		}
1809 	}
1810 	gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0);
1811 	mutex_unlock(&adev->grbm_idx_mutex);
1812 }
1813 
1814 #define DEFAULT_SH_MEM_BASES	(0x6000)
1815 /**
1816  * gfx_v7_0_init_compute_vmid - gart enable
1817  *
1818  * @adev: amdgpu_device pointer
1819  *
1820  * Initialize compute vmid sh_mem registers
1821  *
1822  */
1823 static void gfx_v7_0_init_compute_vmid(struct amdgpu_device *adev)
1824 {
1825 	int i;
1826 	uint32_t sh_mem_config;
1827 	uint32_t sh_mem_bases;
1828 
1829 	/*
1830 	 * Configure apertures:
1831 	 * LDS:         0x60000000'00000000 - 0x60000001'00000000 (4GB)
1832 	 * Scratch:     0x60000001'00000000 - 0x60000002'00000000 (4GB)
1833 	 * GPUVM:       0x60010000'00000000 - 0x60020000'00000000 (1TB)
1834 	*/
1835 	sh_mem_bases = DEFAULT_SH_MEM_BASES | (DEFAULT_SH_MEM_BASES << 16);
1836 	sh_mem_config = SH_MEM_ALIGNMENT_MODE_UNALIGNED <<
1837 			SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT;
1838 	sh_mem_config |= MTYPE_NONCACHED << SH_MEM_CONFIG__DEFAULT_MTYPE__SHIFT;
1839 	mutex_lock(&adev->srbm_mutex);
1840 	for (i = adev->vm_manager.first_kfd_vmid; i < AMDGPU_NUM_VMID; i++) {
1841 		cik_srbm_select(adev, 0, 0, 0, i);
1842 		/* CP and shaders */
1843 		WREG32(mmSH_MEM_CONFIG, sh_mem_config);
1844 		WREG32(mmSH_MEM_APE1_BASE, 1);
1845 		WREG32(mmSH_MEM_APE1_LIMIT, 0);
1846 		WREG32(mmSH_MEM_BASES, sh_mem_bases);
1847 	}
1848 	cik_srbm_select(adev, 0, 0, 0, 0);
1849 	mutex_unlock(&adev->srbm_mutex);
1850 
1851 	/* Initialize all compute VMIDs to have no GDS, GWS, or OA
1852 	   access. These should be enabled by FW for target VMIDs. */
1853 	for (i = adev->vm_manager.first_kfd_vmid; i < AMDGPU_NUM_VMID; i++) {
1854 		WREG32(amdgpu_gds_reg_offset[i].mem_base, 0);
1855 		WREG32(amdgpu_gds_reg_offset[i].mem_size, 0);
1856 		WREG32(amdgpu_gds_reg_offset[i].gws, 0);
1857 		WREG32(amdgpu_gds_reg_offset[i].oa, 0);
1858 	}
1859 }
1860 
1861 static void gfx_v7_0_init_gds_vmid(struct amdgpu_device *adev)
1862 {
1863 	int vmid;
1864 
1865 	/*
1866 	 * Initialize all compute and user-gfx VMIDs to have no GDS, GWS, or OA
1867 	 * access. Compute VMIDs should be enabled by FW for target VMIDs,
1868 	 * the driver can enable them for graphics. VMID0 should maintain
1869 	 * access so that HWS firmware can save/restore entries.
1870 	 */
1871 	for (vmid = 1; vmid < AMDGPU_NUM_VMID; vmid++) {
1872 		WREG32(amdgpu_gds_reg_offset[vmid].mem_base, 0);
1873 		WREG32(amdgpu_gds_reg_offset[vmid].mem_size, 0);
1874 		WREG32(amdgpu_gds_reg_offset[vmid].gws, 0);
1875 		WREG32(amdgpu_gds_reg_offset[vmid].oa, 0);
1876 	}
1877 }
1878 
1879 static void gfx_v7_0_config_init(struct amdgpu_device *adev)
1880 {
1881 	adev->gfx.config.double_offchip_lds_buf = 1;
1882 }
1883 
1884 /**
1885  * gfx_v7_0_constants_init - setup the 3D engine
1886  *
1887  * @adev: amdgpu_device pointer
1888  *
1889  * init the gfx constants such as the 3D engine, tiling configuration
1890  * registers, maximum number of quad pipes, render backends...
1891  */
1892 static void gfx_v7_0_constants_init(struct amdgpu_device *adev)
1893 {
1894 	u32 sh_mem_cfg, sh_static_mem_cfg, sh_mem_base;
1895 	u32 tmp;
1896 	int i;
1897 
1898 	WREG32(mmGRBM_CNTL, (0xff << GRBM_CNTL__READ_TIMEOUT__SHIFT));
1899 
1900 	WREG32(mmGB_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
1901 	WREG32(mmHDP_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
1902 	WREG32(mmDMIF_ADDR_CALC, adev->gfx.config.gb_addr_config);
1903 
1904 	gfx_v7_0_tiling_mode_table_init(adev);
1905 
1906 	gfx_v7_0_setup_rb(adev);
1907 	gfx_v7_0_get_cu_info(adev);
1908 	gfx_v7_0_config_init(adev);
1909 
1910 	/* set HW defaults for 3D engine */
1911 	WREG32(mmCP_MEQ_THRESHOLDS,
1912 	       (0x30 << CP_MEQ_THRESHOLDS__MEQ1_START__SHIFT) |
1913 	       (0x60 << CP_MEQ_THRESHOLDS__MEQ2_START__SHIFT));
1914 
1915 	mutex_lock(&adev->grbm_idx_mutex);
1916 	/*
1917 	 * making sure that the following register writes will be broadcasted
1918 	 * to all the shaders
1919 	 */
1920 	gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0);
1921 
1922 	/* XXX SH_MEM regs */
1923 	/* where to put LDS, scratch, GPUVM in FSA64 space */
1924 	sh_mem_cfg = REG_SET_FIELD(0, SH_MEM_CONFIG, ALIGNMENT_MODE,
1925 				   SH_MEM_ALIGNMENT_MODE_UNALIGNED);
1926 	sh_mem_cfg = REG_SET_FIELD(sh_mem_cfg, SH_MEM_CONFIG, DEFAULT_MTYPE,
1927 				   MTYPE_NC);
1928 	sh_mem_cfg = REG_SET_FIELD(sh_mem_cfg, SH_MEM_CONFIG, APE1_MTYPE,
1929 				   MTYPE_UC);
1930 	sh_mem_cfg = REG_SET_FIELD(sh_mem_cfg, SH_MEM_CONFIG, PRIVATE_ATC, 0);
1931 
1932 	sh_static_mem_cfg = REG_SET_FIELD(0, SH_STATIC_MEM_CONFIG,
1933 				   SWIZZLE_ENABLE, 1);
1934 	sh_static_mem_cfg = REG_SET_FIELD(sh_static_mem_cfg, SH_STATIC_MEM_CONFIG,
1935 				   ELEMENT_SIZE, 1);
1936 	sh_static_mem_cfg = REG_SET_FIELD(sh_static_mem_cfg, SH_STATIC_MEM_CONFIG,
1937 				   INDEX_STRIDE, 3);
1938 	WREG32(mmSH_STATIC_MEM_CONFIG, sh_static_mem_cfg);
1939 
1940 	mutex_lock(&adev->srbm_mutex);
1941 	for_each_vmid_and_zero(i, adev, 0) {
1942 		if (i == 0)
1943 			sh_mem_base = 0;
1944 		else
1945 			sh_mem_base = adev->gmc.shared_aperture_start >> 48;
1946 		cik_srbm_select(adev, 0, 0, 0, i);
1947 		/* CP and shaders */
1948 		WREG32(mmSH_MEM_CONFIG, sh_mem_cfg);
1949 		WREG32(mmSH_MEM_APE1_BASE, 1);
1950 		WREG32(mmSH_MEM_APE1_LIMIT, 0);
1951 		WREG32(mmSH_MEM_BASES, sh_mem_base);
1952 	}
1953 	cik_srbm_select(adev, 0, 0, 0, 0);
1954 	mutex_unlock(&adev->srbm_mutex);
1955 
1956 	gfx_v7_0_init_compute_vmid(adev);
1957 	gfx_v7_0_init_gds_vmid(adev);
1958 
1959 	WREG32(mmSX_DEBUG_1, 0x20);
1960 
1961 	WREG32(mmTA_CNTL_AUX, 0x00010000);
1962 
1963 	tmp = RREG32(mmSPI_CONFIG_CNTL);
1964 	tmp |= 0x03000000;
1965 	WREG32(mmSPI_CONFIG_CNTL, tmp);
1966 
1967 	WREG32(mmSQ_CONFIG, 1);
1968 
1969 	WREG32(mmDB_DEBUG, 0);
1970 
1971 	tmp = RREG32(mmDB_DEBUG2) & ~0xf00fffff;
1972 	tmp |= 0x00000400;
1973 	WREG32(mmDB_DEBUG2, tmp);
1974 
1975 	tmp = RREG32(mmDB_DEBUG3) & ~0x0002021c;
1976 	tmp |= 0x00020200;
1977 	WREG32(mmDB_DEBUG3, tmp);
1978 
1979 	tmp = RREG32(mmCB_HW_CONTROL) & ~0x00010000;
1980 	tmp |= 0x00018208;
1981 	WREG32(mmCB_HW_CONTROL, tmp);
1982 
1983 	WREG32(mmSPI_CONFIG_CNTL_1, (4 << SPI_CONFIG_CNTL_1__VTX_DONE_DELAY__SHIFT));
1984 
1985 	WREG32(mmPA_SC_FIFO_SIZE,
1986 		((adev->gfx.config.sc_prim_fifo_size_frontend << PA_SC_FIFO_SIZE__SC_FRONTEND_PRIM_FIFO_SIZE__SHIFT) |
1987 		(adev->gfx.config.sc_prim_fifo_size_backend << PA_SC_FIFO_SIZE__SC_BACKEND_PRIM_FIFO_SIZE__SHIFT) |
1988 		(adev->gfx.config.sc_hiz_tile_fifo_size << PA_SC_FIFO_SIZE__SC_HIZ_TILE_FIFO_SIZE__SHIFT) |
1989 		(adev->gfx.config.sc_earlyz_tile_fifo_size << PA_SC_FIFO_SIZE__SC_EARLYZ_TILE_FIFO_SIZE__SHIFT)));
1990 
1991 	WREG32(mmVGT_NUM_INSTANCES, 1);
1992 
1993 	WREG32(mmCP_PERFMON_CNTL, 0);
1994 
1995 	WREG32(mmSQ_CONFIG, 0);
1996 
1997 	WREG32(mmPA_SC_FORCE_EOV_MAX_CNTS,
1998 		((4095 << PA_SC_FORCE_EOV_MAX_CNTS__FORCE_EOV_MAX_CLK_CNT__SHIFT) |
1999 		(255 << PA_SC_FORCE_EOV_MAX_CNTS__FORCE_EOV_MAX_REZ_CNT__SHIFT)));
2000 
2001 	WREG32(mmVGT_CACHE_INVALIDATION,
2002 		(VC_AND_TC << VGT_CACHE_INVALIDATION__CACHE_INVALIDATION__SHIFT) |
2003 		(ES_AND_GS_AUTO << VGT_CACHE_INVALIDATION__AUTO_INVLD_EN__SHIFT));
2004 
2005 	WREG32(mmVGT_GS_VERTEX_REUSE, 16);
2006 	WREG32(mmPA_SC_LINE_STIPPLE_STATE, 0);
2007 
2008 	WREG32(mmPA_CL_ENHANCE, PA_CL_ENHANCE__CLIP_VTX_REORDER_ENA_MASK |
2009 			(3 << PA_CL_ENHANCE__NUM_CLIP_SEQ__SHIFT));
2010 	WREG32(mmPA_SC_ENHANCE, PA_SC_ENHANCE__ENABLE_PA_SC_OUT_OF_ORDER_MASK);
2011 
2012 	tmp = RREG32(mmSPI_ARB_PRIORITY);
2013 	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS0, 2);
2014 	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS1, 2);
2015 	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS2, 2);
2016 	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS3, 2);
2017 	WREG32(mmSPI_ARB_PRIORITY, tmp);
2018 
2019 	mutex_unlock(&adev->grbm_idx_mutex);
2020 
2021 	udelay(50);
2022 }
2023 
2024 /**
2025  * gfx_v7_0_ring_test_ring - basic gfx ring test
2026  *
2027  * @ring: amdgpu_ring structure holding ring information
2028  *
2029  * Allocate a scratch register and write to it using the gfx ring (CIK).
2030  * Provides a basic gfx ring test to verify that the ring is working.
2031  * Used by gfx_v7_0_cp_gfx_resume();
2032  * Returns 0 on success, error on failure.
2033  */
2034 static int gfx_v7_0_ring_test_ring(struct amdgpu_ring *ring)
2035 {
2036 	struct amdgpu_device *adev = ring->adev;
2037 	uint32_t tmp = 0;
2038 	unsigned i;
2039 	int r;
2040 
2041 	WREG32(mmSCRATCH_REG0, 0xCAFEDEAD);
2042 	r = amdgpu_ring_alloc(ring, 3);
2043 	if (r)
2044 		return r;
2045 
2046 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_UCONFIG_REG, 1));
2047 	amdgpu_ring_write(ring, mmSCRATCH_REG0 - PACKET3_SET_UCONFIG_REG_START);
2048 	amdgpu_ring_write(ring, 0xDEADBEEF);
2049 	amdgpu_ring_commit(ring);
2050 
2051 	for (i = 0; i < adev->usec_timeout; i++) {
2052 		tmp = RREG32(mmSCRATCH_REG0);
2053 		if (tmp == 0xDEADBEEF)
2054 			break;
2055 		udelay(1);
2056 	}
2057 	if (i >= adev->usec_timeout)
2058 		r = -ETIMEDOUT;
2059 	return r;
2060 }
2061 
2062 /**
2063  * gfx_v7_0_ring_emit_hdp_flush - emit an hdp flush on the cp
2064  *
2065  * @ring: amdgpu_ring structure holding ring information
2066  *
2067  * Emits an hdp flush on the cp.
2068  */
2069 static void gfx_v7_0_ring_emit_hdp_flush(struct amdgpu_ring *ring)
2070 {
2071 	u32 ref_and_mask;
2072 	int usepfp;
2073 	struct amdgpu_device *adev = ring->adev;
2074 
2075 	if (!adev->gfx.funcs->get_hdp_flush_mask) {
2076 		dev_err(adev->dev, "%s: gfx hdp flush is not supported.\n", __func__);
2077 		return;
2078 	}
2079 
2080 	adev->gfx.funcs->get_hdp_flush_mask(ring, &ref_and_mask, &usepfp);
2081 	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
2082 	amdgpu_ring_write(ring, (WAIT_REG_MEM_OPERATION(1) | /* write, wait, write */
2083 				 WAIT_REG_MEM_FUNCTION(3) |  /* == */
2084 				 WAIT_REG_MEM_ENGINE(usepfp)));   /* pfp or me */
2085 	amdgpu_ring_write(ring, mmGPU_HDP_FLUSH_REQ);
2086 	amdgpu_ring_write(ring, mmGPU_HDP_FLUSH_DONE);
2087 	amdgpu_ring_write(ring, ref_and_mask);
2088 	amdgpu_ring_write(ring, ref_and_mask);
2089 	amdgpu_ring_write(ring, 0x20); /* poll interval */
2090 }
2091 
2092 static void gfx_v7_0_ring_emit_vgt_flush(struct amdgpu_ring *ring)
2093 {
2094 	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE, 0));
2095 	amdgpu_ring_write(ring, EVENT_TYPE(VS_PARTIAL_FLUSH) |
2096 		EVENT_INDEX(4));
2097 
2098 	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE, 0));
2099 	amdgpu_ring_write(ring, EVENT_TYPE(VGT_FLUSH) |
2100 		EVENT_INDEX(0));
2101 }
2102 
2103 /**
2104  * gfx_v7_0_ring_emit_fence_gfx - emit a fence on the gfx ring
2105  *
2106  * @ring: amdgpu_ring structure holding ring information
2107  * @addr: address
2108  * @seq: sequence number
2109  * @flags: fence related flags
2110  *
2111  * Emits a fence sequence number on the gfx ring and flushes
2112  * GPU caches.
2113  */
2114 static void gfx_v7_0_ring_emit_fence_gfx(struct amdgpu_ring *ring, u64 addr,
2115 					 u64 seq, unsigned flags)
2116 {
2117 	bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT;
2118 	bool int_sel = flags & AMDGPU_FENCE_FLAG_INT;
2119 	bool exec = flags & AMDGPU_FENCE_FLAG_EXEC;
2120 
2121 	/* Workaround for cache flush problems. First send a dummy EOP
2122 	 * event down the pipe with seq one below.
2123 	 */
2124 	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE_EOP, 4));
2125 	amdgpu_ring_write(ring, (EOP_TCL1_ACTION_EN |
2126 				 EOP_TC_ACTION_EN |
2127 				 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
2128 				 EVENT_INDEX(5)));
2129 	amdgpu_ring_write(ring, addr & 0xfffffffc);
2130 	amdgpu_ring_write(ring, (upper_32_bits(addr) & 0xffff) |
2131 				DATA_SEL(1) | INT_SEL(0));
2132 	amdgpu_ring_write(ring, lower_32_bits(seq - 1));
2133 	amdgpu_ring_write(ring, upper_32_bits(seq - 1));
2134 
2135 	/* Then send the real EOP event down the pipe. */
2136 	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE_EOP, 4));
2137 	amdgpu_ring_write(ring, (EOP_TCL1_ACTION_EN |
2138 				 EOP_TC_ACTION_EN |
2139 				 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
2140 				 EVENT_INDEX(5) |
2141 				 (exec ? EOP_EXEC : 0)));
2142 	amdgpu_ring_write(ring, addr & 0xfffffffc);
2143 	amdgpu_ring_write(ring, (upper_32_bits(addr) & 0xffff) |
2144 				DATA_SEL(write64bit ? 2 : 1) | INT_SEL(int_sel ? 2 : 0));
2145 	amdgpu_ring_write(ring, lower_32_bits(seq));
2146 	amdgpu_ring_write(ring, upper_32_bits(seq));
2147 }
2148 
2149 /**
2150  * gfx_v7_0_ring_emit_fence_compute - emit a fence on the compute ring
2151  *
2152  * @ring: amdgpu_ring structure holding ring information
2153  * @addr: address
2154  * @seq: sequence number
2155  * @flags: fence related flags
2156  *
2157  * Emits a fence sequence number on the compute ring and flushes
2158  * GPU caches.
2159  */
2160 static void gfx_v7_0_ring_emit_fence_compute(struct amdgpu_ring *ring,
2161 					     u64 addr, u64 seq,
2162 					     unsigned flags)
2163 {
2164 	bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT;
2165 	bool int_sel = flags & AMDGPU_FENCE_FLAG_INT;
2166 
2167 	/* RELEASE_MEM - flush caches, send int */
2168 	amdgpu_ring_write(ring, PACKET3(PACKET3_RELEASE_MEM, 5));
2169 	amdgpu_ring_write(ring, (EOP_TCL1_ACTION_EN |
2170 				 EOP_TC_ACTION_EN |
2171 				 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
2172 				 EVENT_INDEX(5)));
2173 	amdgpu_ring_write(ring, DATA_SEL(write64bit ? 2 : 1) | INT_SEL(int_sel ? 2 : 0));
2174 	amdgpu_ring_write(ring, addr & 0xfffffffc);
2175 	amdgpu_ring_write(ring, upper_32_bits(addr));
2176 	amdgpu_ring_write(ring, lower_32_bits(seq));
2177 	amdgpu_ring_write(ring, upper_32_bits(seq));
2178 }
2179 
2180 /*
2181  * IB stuff
2182  */
2183 /**
2184  * gfx_v7_0_ring_emit_ib_gfx - emit an IB (Indirect Buffer) on the ring
2185  *
2186  * @ring: amdgpu_ring structure holding ring information
2187  * @job: job to retrieve vmid from
2188  * @ib: amdgpu indirect buffer object
2189  * @flags: options (AMDGPU_HAVE_CTX_SWITCH)
2190  *
2191  * Emits an DE (drawing engine) or CE (constant engine) IB
2192  * on the gfx ring.  IBs are usually generated by userspace
2193  * acceleration drivers and submitted to the kernel for
2194  * scheduling on the ring.  This function schedules the IB
2195  * on the gfx ring for execution by the GPU.
2196  */
2197 static void gfx_v7_0_ring_emit_ib_gfx(struct amdgpu_ring *ring,
2198 					struct amdgpu_job *job,
2199 					struct amdgpu_ib *ib,
2200 					uint32_t flags)
2201 {
2202 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
2203 	u32 header, control = 0;
2204 
2205 	if (ib->flags & AMDGPU_IB_FLAG_CE)
2206 		header = PACKET3(PACKET3_INDIRECT_BUFFER_CONST, 2);
2207 	else
2208 		header = PACKET3(PACKET3_INDIRECT_BUFFER, 2);
2209 
2210 	control |= ib->length_dw | (vmid << 24);
2211 
2212 	amdgpu_ring_write(ring, header);
2213 	amdgpu_ring_write(ring,
2214 #ifdef __BIG_ENDIAN
2215 			  (2 << 0) |
2216 #endif
2217 			  (ib->gpu_addr & 0xFFFFFFFC));
2218 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xFFFF);
2219 	amdgpu_ring_write(ring, control);
2220 }
2221 
2222 static void gfx_v7_0_ring_emit_ib_compute(struct amdgpu_ring *ring,
2223 					  struct amdgpu_job *job,
2224 					  struct amdgpu_ib *ib,
2225 					  uint32_t flags)
2226 {
2227 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
2228 	u32 control = INDIRECT_BUFFER_VALID | ib->length_dw | (vmid << 24);
2229 
2230 	/* Currently, there is a high possibility to get wave ID mismatch
2231 	 * between ME and GDS, leading to a hw deadlock, because ME generates
2232 	 * different wave IDs than the GDS expects. This situation happens
2233 	 * randomly when at least 5 compute pipes use GDS ordered append.
2234 	 * The wave IDs generated by ME are also wrong after suspend/resume.
2235 	 * Those are probably bugs somewhere else in the kernel driver.
2236 	 *
2237 	 * Writing GDS_COMPUTE_MAX_WAVE_ID resets wave ID counters in ME and
2238 	 * GDS to 0 for this ring (me/pipe).
2239 	 */
2240 	if (ib->flags & AMDGPU_IB_FLAG_RESET_GDS_MAX_WAVE_ID) {
2241 		amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1));
2242 		amdgpu_ring_write(ring, mmGDS_COMPUTE_MAX_WAVE_ID - PACKET3_SET_CONFIG_REG_START);
2243 		amdgpu_ring_write(ring, ring->adev->gds.gds_compute_max_wave_id);
2244 	}
2245 
2246 	amdgpu_ring_write(ring, PACKET3(PACKET3_INDIRECT_BUFFER, 2));
2247 	amdgpu_ring_write(ring,
2248 #ifdef __BIG_ENDIAN
2249 					  (2 << 0) |
2250 #endif
2251 					  (ib->gpu_addr & 0xFFFFFFFC));
2252 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xFFFF);
2253 	amdgpu_ring_write(ring, control);
2254 }
2255 
2256 static void gfx_v7_0_ring_emit_sb(struct amdgpu_ring *ring)
2257 {
2258 	amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
2259 	amdgpu_ring_write(ring, 0);
2260 }
2261 
2262 static void gfx_v7_ring_emit_cntxcntl(struct amdgpu_ring *ring, uint32_t flags)
2263 {
2264 	uint32_t dw2 = 0;
2265 
2266 	dw2 |= 0x80000000; /* set load_enable otherwise this package is just NOPs */
2267 	if (flags & AMDGPU_HAVE_CTX_SWITCH) {
2268 		gfx_v7_0_ring_emit_vgt_flush(ring);
2269 		/* set load_global_config & load_global_uconfig */
2270 		dw2 |= 0x8001;
2271 		/* set load_cs_sh_regs */
2272 		dw2 |= 0x01000000;
2273 		/* set load_per_context_state & load_gfx_sh_regs */
2274 		dw2 |= 0x10002;
2275 	}
2276 
2277 	amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1));
2278 	amdgpu_ring_write(ring, dw2);
2279 	amdgpu_ring_write(ring, 0);
2280 }
2281 
2282 /**
2283  * gfx_v7_0_ring_test_ib - basic ring IB test
2284  *
2285  * @ring: amdgpu_ring structure holding ring information
2286  * @timeout: timeout value in jiffies, or MAX_SCHEDULE_TIMEOUT
2287  *
2288  * Allocate an IB and execute it on the gfx ring (CIK).
2289  * Provides a basic gfx ring test to verify that IBs are working.
2290  * Returns 0 on success, error on failure.
2291  */
2292 static int gfx_v7_0_ring_test_ib(struct amdgpu_ring *ring, long timeout)
2293 {
2294 	struct amdgpu_device *adev = ring->adev;
2295 	struct amdgpu_job *job;
2296 	struct amdgpu_ib *ib;
2297 	struct dma_fence *f = NULL;
2298 	uint32_t tmp = 0;
2299 	long r;
2300 
2301 	WREG32(mmSCRATCH_REG0, 0xCAFEDEAD);
2302 
2303 	r = amdgpu_job_alloc_with_ib(ring->adev, NULL, NULL, 256,
2304 				     AMDGPU_IB_POOL_DIRECT,
2305 				     AMDGPU_KERNEL_JOB_ID_GFX_RING_TEST,
2306 				     &job);
2307 	if (r)
2308 		return r;
2309 
2310 	ib = &job->ibs[0];
2311 	ib->ptr[0] = PACKET3(PACKET3_SET_UCONFIG_REG, 1);
2312 	ib->ptr[1] = mmSCRATCH_REG0 - PACKET3_SET_UCONFIG_REG_START;
2313 	ib->ptr[2] = 0xDEADBEEF;
2314 	ib->length_dw = 3;
2315 
2316 	r = amdgpu_job_submit_direct(job, ring, &f);
2317 	if (r) {
2318 		amdgpu_job_free(job);
2319 		goto error;
2320 	}
2321 
2322 	r = dma_fence_wait_timeout(f, false, timeout);
2323 	if (r == 0) {
2324 		r = -ETIMEDOUT;
2325 		goto error;
2326 	} else if (r < 0) {
2327 		goto error;
2328 	}
2329 	tmp = RREG32(mmSCRATCH_REG0);
2330 	if (tmp == 0xDEADBEEF)
2331 		r = 0;
2332 	else
2333 		r = -EINVAL;
2334 
2335 error:
2336 	dma_fence_put(f);
2337 	return r;
2338 }
2339 
2340 /*
2341  * CP.
2342  * On CIK, gfx and compute now have independent command processors.
2343  *
2344  * GFX
2345  * Gfx consists of a single ring and can process both gfx jobs and
2346  * compute jobs.  The gfx CP consists of three microengines (ME):
2347  * PFP - Pre-Fetch Parser
2348  * ME - Micro Engine
2349  * CE - Constant Engine
2350  * The PFP and ME make up what is considered the Drawing Engine (DE).
2351  * The CE is an asynchronous engine used for updating buffer desciptors
2352  * used by the DE so that they can be loaded into cache in parallel
2353  * while the DE is processing state update packets.
2354  *
2355  * Compute
2356  * The compute CP consists of two microengines (ME):
2357  * MEC1 - Compute MicroEngine 1
2358  * MEC2 - Compute MicroEngine 2
2359  * Each MEC supports 4 compute pipes and each pipe supports 8 queues.
2360  * The queues are exposed to userspace and are programmed directly
2361  * by the compute runtime.
2362  */
2363 /**
2364  * gfx_v7_0_cp_gfx_enable - enable/disable the gfx CP MEs
2365  *
2366  * @adev: amdgpu_device pointer
2367  * @enable: enable or disable the MEs
2368  *
2369  * Halts or unhalts the gfx MEs.
2370  */
2371 static void gfx_v7_0_cp_gfx_enable(struct amdgpu_device *adev, bool enable)
2372 {
2373 	if (enable)
2374 		WREG32(mmCP_ME_CNTL, 0);
2375 	else
2376 		WREG32(mmCP_ME_CNTL, (CP_ME_CNTL__ME_HALT_MASK |
2377 				      CP_ME_CNTL__PFP_HALT_MASK |
2378 				      CP_ME_CNTL__CE_HALT_MASK));
2379 	udelay(50);
2380 }
2381 
2382 /**
2383  * gfx_v7_0_cp_gfx_load_microcode - load the gfx CP ME ucode
2384  *
2385  * @adev: amdgpu_device pointer
2386  *
2387  * Loads the gfx PFP, ME, and CE ucode.
2388  * Returns 0 for success, -EINVAL if the ucode is not available.
2389  */
2390 static int gfx_v7_0_cp_gfx_load_microcode(struct amdgpu_device *adev)
2391 {
2392 	const struct gfx_firmware_header_v1_0 *pfp_hdr;
2393 	const struct gfx_firmware_header_v1_0 *ce_hdr;
2394 	const struct gfx_firmware_header_v1_0 *me_hdr;
2395 	const __le32 *fw_data;
2396 	unsigned i, fw_size;
2397 
2398 	if (!adev->gfx.me_fw || !adev->gfx.pfp_fw || !adev->gfx.ce_fw)
2399 		return -EINVAL;
2400 
2401 	pfp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.pfp_fw->data;
2402 	ce_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.ce_fw->data;
2403 	me_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.me_fw->data;
2404 
2405 	amdgpu_ucode_print_gfx_hdr(&pfp_hdr->header);
2406 	amdgpu_ucode_print_gfx_hdr(&ce_hdr->header);
2407 	amdgpu_ucode_print_gfx_hdr(&me_hdr->header);
2408 	adev->gfx.pfp_fw_version = le32_to_cpu(pfp_hdr->header.ucode_version);
2409 	adev->gfx.ce_fw_version = le32_to_cpu(ce_hdr->header.ucode_version);
2410 	adev->gfx.me_fw_version = le32_to_cpu(me_hdr->header.ucode_version);
2411 	adev->gfx.me_feature_version = le32_to_cpu(me_hdr->ucode_feature_version);
2412 	adev->gfx.ce_feature_version = le32_to_cpu(ce_hdr->ucode_feature_version);
2413 	adev->gfx.pfp_feature_version = le32_to_cpu(pfp_hdr->ucode_feature_version);
2414 
2415 	gfx_v7_0_cp_gfx_enable(adev, false);
2416 
2417 	/* PFP */
2418 	fw_data = (const __le32 *)
2419 		(adev->gfx.pfp_fw->data +
2420 		 le32_to_cpu(pfp_hdr->header.ucode_array_offset_bytes));
2421 	fw_size = le32_to_cpu(pfp_hdr->header.ucode_size_bytes) / 4;
2422 	WREG32(mmCP_PFP_UCODE_ADDR, 0);
2423 	for (i = 0; i < fw_size; i++)
2424 		WREG32(mmCP_PFP_UCODE_DATA, le32_to_cpup(fw_data++));
2425 	WREG32(mmCP_PFP_UCODE_ADDR, adev->gfx.pfp_fw_version);
2426 
2427 	/* CE */
2428 	fw_data = (const __le32 *)
2429 		(adev->gfx.ce_fw->data +
2430 		 le32_to_cpu(ce_hdr->header.ucode_array_offset_bytes));
2431 	fw_size = le32_to_cpu(ce_hdr->header.ucode_size_bytes) / 4;
2432 	WREG32(mmCP_CE_UCODE_ADDR, 0);
2433 	for (i = 0; i < fw_size; i++)
2434 		WREG32(mmCP_CE_UCODE_DATA, le32_to_cpup(fw_data++));
2435 	WREG32(mmCP_CE_UCODE_ADDR, adev->gfx.ce_fw_version);
2436 
2437 	/* ME */
2438 	fw_data = (const __le32 *)
2439 		(adev->gfx.me_fw->data +
2440 		 le32_to_cpu(me_hdr->header.ucode_array_offset_bytes));
2441 	fw_size = le32_to_cpu(me_hdr->header.ucode_size_bytes) / 4;
2442 	WREG32(mmCP_ME_RAM_WADDR, 0);
2443 	for (i = 0; i < fw_size; i++)
2444 		WREG32(mmCP_ME_RAM_DATA, le32_to_cpup(fw_data++));
2445 	WREG32(mmCP_ME_RAM_WADDR, adev->gfx.me_fw_version);
2446 
2447 	return 0;
2448 }
2449 
2450 /**
2451  * gfx_v7_0_cp_gfx_start - start the gfx ring
2452  *
2453  * @adev: amdgpu_device pointer
2454  *
2455  * Enables the ring and loads the clear state context and other
2456  * packets required to init the ring.
2457  * Returns 0 for success, error for failure.
2458  */
2459 static int gfx_v7_0_cp_gfx_start(struct amdgpu_device *adev)
2460 {
2461 	struct amdgpu_ring *ring = &adev->gfx.gfx_ring[0];
2462 	const struct cs_section_def *sect = NULL;
2463 	const struct cs_extent_def *ext = NULL;
2464 	int r, i;
2465 
2466 	/* init the CP */
2467 	WREG32(mmCP_MAX_CONTEXT, adev->gfx.config.max_hw_contexts - 1);
2468 	WREG32(mmCP_ENDIAN_SWAP, 0);
2469 	WREG32(mmCP_DEVICE_ID, 1);
2470 
2471 	gfx_v7_0_cp_gfx_enable(adev, true);
2472 
2473 	r = amdgpu_ring_alloc(ring, gfx_v7_0_get_csb_size(adev) + 8);
2474 	if (r) {
2475 		drm_err(adev_to_drm(adev), "cp failed to lock ring (%d).\n", r);
2476 		return r;
2477 	}
2478 
2479 	/* init the CE partitions.  CE only used for gfx on CIK */
2480 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_BASE, 2));
2481 	amdgpu_ring_write(ring, PACKET3_BASE_INDEX(CE_PARTITION_BASE));
2482 	amdgpu_ring_write(ring, 0x8000);
2483 	amdgpu_ring_write(ring, 0x8000);
2484 
2485 	/* clear state buffer */
2486 	amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
2487 	amdgpu_ring_write(ring, PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
2488 
2489 	amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1));
2490 	amdgpu_ring_write(ring, 0x80000000);
2491 	amdgpu_ring_write(ring, 0x80000000);
2492 
2493 	for (sect = adev->gfx.rlc.cs_data; sect->section != NULL; ++sect) {
2494 		for (ext = sect->section; ext->extent != NULL; ++ext) {
2495 			if (sect->id == SECT_CONTEXT) {
2496 				amdgpu_ring_write(ring,
2497 						  PACKET3(PACKET3_SET_CONTEXT_REG, ext->reg_count));
2498 				amdgpu_ring_write(ring, ext->reg_index - PACKET3_SET_CONTEXT_REG_START);
2499 				for (i = 0; i < ext->reg_count; i++)
2500 					amdgpu_ring_write(ring, ext->extent[i]);
2501 			}
2502 		}
2503 	}
2504 
2505 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 2));
2506 	amdgpu_ring_write(ring, mmPA_SC_RASTER_CONFIG - PACKET3_SET_CONTEXT_REG_START);
2507 	amdgpu_ring_write(ring, adev->gfx.config.rb_config[0][0].raster_config);
2508 	amdgpu_ring_write(ring, adev->gfx.config.rb_config[0][0].raster_config_1);
2509 
2510 	amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
2511 	amdgpu_ring_write(ring, PACKET3_PREAMBLE_END_CLEAR_STATE);
2512 
2513 	amdgpu_ring_write(ring, PACKET3(PACKET3_CLEAR_STATE, 0));
2514 	amdgpu_ring_write(ring, 0);
2515 
2516 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 2));
2517 	amdgpu_ring_write(ring, 0x00000316);
2518 	amdgpu_ring_write(ring, 0x0000000e); /* VGT_VERTEX_REUSE_BLOCK_CNTL */
2519 	amdgpu_ring_write(ring, 0x00000010); /* VGT_OUT_DEALLOC_CNTL */
2520 
2521 	amdgpu_ring_commit(ring);
2522 
2523 	return 0;
2524 }
2525 
2526 /**
2527  * gfx_v7_0_cp_gfx_resume - setup the gfx ring buffer registers
2528  *
2529  * @adev: amdgpu_device pointer
2530  *
2531  * Program the location and size of the gfx ring buffer
2532  * and test it to make sure it's working.
2533  * Returns 0 for success, error for failure.
2534  */
2535 static int gfx_v7_0_cp_gfx_resume(struct amdgpu_device *adev)
2536 {
2537 	struct amdgpu_ring *ring;
2538 	u32 tmp;
2539 	u32 rb_bufsz;
2540 	u64 rb_addr, rptr_addr;
2541 	int r;
2542 
2543 	WREG32(mmCP_SEM_WAIT_TIMER, 0x0);
2544 	if (adev->asic_type != CHIP_HAWAII)
2545 		WREG32(mmCP_SEM_INCOMPLETE_TIMER_CNTL, 0x0);
2546 
2547 	/* Set the write pointer delay */
2548 	WREG32(mmCP_RB_WPTR_DELAY, 0);
2549 
2550 	/* set the RB to use vmid 0 */
2551 	WREG32(mmCP_RB_VMID, 0);
2552 
2553 	WREG32(mmSCRATCH_ADDR, 0);
2554 
2555 	/* ring 0 - compute and gfx */
2556 	ring = &adev->gfx.gfx_ring[0];
2557 	*ring->wptr_cpu_addr = 0;
2558 	*ring->rptr_cpu_addr = 0;
2559 	amdgpu_ring_clear_ring(ring);
2560 
2561 	/* Set ring buffer size */
2562 	rb_bufsz = order_base_2(ring->ring_size / 8);
2563 	tmp = (order_base_2(AMDGPU_GPU_PAGE_SIZE/8) << 8) | rb_bufsz;
2564 #ifdef __BIG_ENDIAN
2565 	tmp |= 2 << CP_RB0_CNTL__BUF_SWAP__SHIFT;
2566 #endif
2567 	WREG32(mmCP_RB0_CNTL, tmp);
2568 
2569 	/* Initialize the ring buffer's read and write pointers */
2570 	WREG32(mmCP_RB0_CNTL, tmp | CP_RB0_CNTL__RB_RPTR_WR_ENA_MASK);
2571 	ring->wptr = 0;
2572 	WREG32(mmCP_RB0_WPTR, lower_32_bits(ring->wptr));
2573 	WREG32(mmCP_RB0_RPTR, lower_32_bits(ring->wptr));
2574 
2575 	/* set the wb address whether it's enabled or not */
2576 	rptr_addr = ring->rptr_gpu_addr;
2577 	WREG32(mmCP_RB0_RPTR_ADDR, lower_32_bits(rptr_addr));
2578 	WREG32(mmCP_RB0_RPTR_ADDR_HI, upper_32_bits(rptr_addr) & 0xFF);
2579 
2580 	/* scratch register shadowing is no longer supported */
2581 	WREG32(mmSCRATCH_UMSK, 0);
2582 
2583 	mdelay(1);
2584 	WREG32(mmCP_RB0_CNTL, tmp);
2585 
2586 	rb_addr = ring->gpu_addr >> 8;
2587 	WREG32(mmCP_RB0_BASE, rb_addr);
2588 	WREG32(mmCP_RB0_BASE_HI, upper_32_bits(rb_addr));
2589 
2590 	/* start the ring */
2591 	r = gfx_v7_0_cp_gfx_start(adev);
2592 	if (r)
2593 		return r;
2594 
2595 	r = amdgpu_ring_test_helper(ring);
2596 	if (r)
2597 		return r;
2598 
2599 	return 0;
2600 }
2601 
2602 static u64 gfx_v7_0_ring_get_rptr(struct amdgpu_ring *ring)
2603 {
2604 	return *ring->rptr_cpu_addr;
2605 }
2606 
2607 static u64 gfx_v7_0_ring_get_wptr_gfx(struct amdgpu_ring *ring)
2608 {
2609 	struct amdgpu_device *adev = ring->adev;
2610 
2611 	return RREG32(mmCP_RB0_WPTR);
2612 }
2613 
2614 static void gfx_v7_0_ring_set_wptr_gfx(struct amdgpu_ring *ring)
2615 {
2616 	struct amdgpu_device *adev = ring->adev;
2617 
2618 	WREG32(mmCP_RB0_WPTR, lower_32_bits(ring->wptr));
2619 	(void)RREG32(mmCP_RB0_WPTR);
2620 }
2621 
2622 static u64 gfx_v7_0_ring_get_wptr_compute(struct amdgpu_ring *ring)
2623 {
2624 	/* XXX check if swapping is necessary on BE */
2625 	return *ring->wptr_cpu_addr;
2626 }
2627 
2628 static void gfx_v7_0_ring_set_wptr_compute(struct amdgpu_ring *ring)
2629 {
2630 	struct amdgpu_device *adev = ring->adev;
2631 
2632 	/* XXX check if swapping is necessary on BE */
2633 	*ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
2634 	WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
2635 }
2636 
2637 /**
2638  * gfx_v7_0_cp_compute_enable - enable/disable the compute CP MEs
2639  *
2640  * @adev: amdgpu_device pointer
2641  * @enable: enable or disable the MEs
2642  *
2643  * Halts or unhalts the compute MEs.
2644  */
2645 static void gfx_v7_0_cp_compute_enable(struct amdgpu_device *adev, bool enable)
2646 {
2647 	if (enable)
2648 		WREG32(mmCP_MEC_CNTL, 0);
2649 	else
2650 		WREG32(mmCP_MEC_CNTL, (CP_MEC_CNTL__MEC_ME1_HALT_MASK |
2651 				       CP_MEC_CNTL__MEC_ME2_HALT_MASK));
2652 	udelay(50);
2653 }
2654 
2655 /**
2656  * gfx_v7_0_cp_compute_load_microcode - load the compute CP ME ucode
2657  *
2658  * @adev: amdgpu_device pointer
2659  *
2660  * Loads the compute MEC1&2 ucode.
2661  * Returns 0 for success, -EINVAL if the ucode is not available.
2662  */
2663 static int gfx_v7_0_cp_compute_load_microcode(struct amdgpu_device *adev)
2664 {
2665 	const struct gfx_firmware_header_v1_0 *mec_hdr;
2666 	const __le32 *fw_data;
2667 	unsigned i, fw_size;
2668 
2669 	if (!adev->gfx.mec_fw)
2670 		return -EINVAL;
2671 
2672 	mec_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data;
2673 	amdgpu_ucode_print_gfx_hdr(&mec_hdr->header);
2674 	adev->gfx.mec_fw_version = le32_to_cpu(mec_hdr->header.ucode_version);
2675 	adev->gfx.mec_feature_version = le32_to_cpu(
2676 					mec_hdr->ucode_feature_version);
2677 
2678 	gfx_v7_0_cp_compute_enable(adev, false);
2679 
2680 	/* MEC1 */
2681 	fw_data = (const __le32 *)
2682 		(adev->gfx.mec_fw->data +
2683 		 le32_to_cpu(mec_hdr->header.ucode_array_offset_bytes));
2684 	fw_size = le32_to_cpu(mec_hdr->header.ucode_size_bytes) / 4;
2685 	WREG32(mmCP_MEC_ME1_UCODE_ADDR, 0);
2686 	for (i = 0; i < fw_size; i++)
2687 		WREG32(mmCP_MEC_ME1_UCODE_DATA, le32_to_cpup(fw_data++));
2688 	WREG32(mmCP_MEC_ME1_UCODE_ADDR, 0);
2689 
2690 	if (adev->asic_type == CHIP_KAVERI) {
2691 		const struct gfx_firmware_header_v1_0 *mec2_hdr;
2692 
2693 		if (!adev->gfx.mec2_fw)
2694 			return -EINVAL;
2695 
2696 		mec2_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec2_fw->data;
2697 		amdgpu_ucode_print_gfx_hdr(&mec2_hdr->header);
2698 		adev->gfx.mec2_fw_version = le32_to_cpu(mec2_hdr->header.ucode_version);
2699 		adev->gfx.mec2_feature_version = le32_to_cpu(
2700 				mec2_hdr->ucode_feature_version);
2701 
2702 		/* MEC2 */
2703 		fw_data = (const __le32 *)
2704 			(adev->gfx.mec2_fw->data +
2705 			 le32_to_cpu(mec2_hdr->header.ucode_array_offset_bytes));
2706 		fw_size = le32_to_cpu(mec2_hdr->header.ucode_size_bytes) / 4;
2707 		WREG32(mmCP_MEC_ME2_UCODE_ADDR, 0);
2708 		for (i = 0; i < fw_size; i++)
2709 			WREG32(mmCP_MEC_ME2_UCODE_DATA, le32_to_cpup(fw_data++));
2710 		WREG32(mmCP_MEC_ME2_UCODE_ADDR, 0);
2711 	}
2712 
2713 	return 0;
2714 }
2715 
2716 static void gfx_v7_0_mec_fini(struct amdgpu_device *adev)
2717 {
2718 	amdgpu_bo_free_kernel(&adev->gfx.mec.hpd_eop_obj, NULL, NULL);
2719 }
2720 
2721 static int gfx_v7_0_mec_init(struct amdgpu_device *adev)
2722 {
2723 	int r;
2724 	u32 *hpd;
2725 	size_t mec_hpd_size;
2726 
2727 	bitmap_zero(adev->gfx.mec_bitmap[0].queue_bitmap, AMDGPU_MAX_COMPUTE_QUEUES);
2728 
2729 	/* take ownership of the relevant compute queues */
2730 	amdgpu_gfx_compute_queue_acquire(adev);
2731 
2732 	/* allocate space for ALL pipes (even the ones we don't own) */
2733 	mec_hpd_size = adev->gfx.mec.num_mec * adev->gfx.mec.num_pipe_per_mec
2734 		* GFX7_MEC_HPD_SIZE * 2;
2735 
2736 	r = amdgpu_bo_create_reserved(adev, mec_hpd_size, PAGE_SIZE,
2737 				      AMDGPU_GEM_DOMAIN_VRAM |
2738 				      AMDGPU_GEM_DOMAIN_GTT,
2739 				      &adev->gfx.mec.hpd_eop_obj,
2740 				      &adev->gfx.mec.hpd_eop_gpu_addr,
2741 				      (void **)&hpd);
2742 	if (r) {
2743 		dev_warn(adev->dev, "(%d) create, pin or map of HDP EOP bo failed\n", r);
2744 		gfx_v7_0_mec_fini(adev);
2745 		return r;
2746 	}
2747 
2748 	/* clear memory.  Not sure if this is required or not */
2749 	memset(hpd, 0, mec_hpd_size);
2750 
2751 	amdgpu_bo_kunmap(adev->gfx.mec.hpd_eop_obj);
2752 	amdgpu_bo_unreserve(adev->gfx.mec.hpd_eop_obj);
2753 
2754 	return 0;
2755 }
2756 
2757 static void gfx_v7_0_compute_pipe_init(struct amdgpu_device *adev,
2758 				       int mec, int pipe)
2759 {
2760 	u64 eop_gpu_addr;
2761 	u32 tmp;
2762 	size_t eop_offset = (mec * adev->gfx.mec.num_pipe_per_mec + pipe)
2763 			    * GFX7_MEC_HPD_SIZE * 2;
2764 
2765 	mutex_lock(&adev->srbm_mutex);
2766 	eop_gpu_addr = adev->gfx.mec.hpd_eop_gpu_addr + eop_offset;
2767 
2768 	cik_srbm_select(adev, mec + 1, pipe, 0, 0);
2769 
2770 	/* write the EOP addr */
2771 	WREG32(mmCP_HPD_EOP_BASE_ADDR, eop_gpu_addr >> 8);
2772 	WREG32(mmCP_HPD_EOP_BASE_ADDR_HI, upper_32_bits(eop_gpu_addr) >> 8);
2773 
2774 	/* set the VMID assigned */
2775 	WREG32(mmCP_HPD_EOP_VMID, 0);
2776 
2777 	/* set the EOP size, register value is 2^(EOP_SIZE+1) dwords */
2778 	tmp = RREG32(mmCP_HPD_EOP_CONTROL);
2779 	tmp &= ~CP_HPD_EOP_CONTROL__EOP_SIZE_MASK;
2780 	tmp |= order_base_2(GFX7_MEC_HPD_SIZE / 8);
2781 	WREG32(mmCP_HPD_EOP_CONTROL, tmp);
2782 
2783 	cik_srbm_select(adev, 0, 0, 0, 0);
2784 	mutex_unlock(&adev->srbm_mutex);
2785 }
2786 
2787 static int gfx_v7_0_deactivate_hqd(struct amdgpu_device *adev, u32 req)
2788 {
2789 	int i, r = 0;
2790 
2791 	/* disable the queue if it's active */
2792 	if (RREG32(mmCP_HQD_ACTIVE) & CP_HQD_ACTIVE__ACTIVE_MASK) {
2793 		WREG32_FIELD(CP_HQD_DEQUEUE_REQUEST, DEQUEUE_REQ, req);
2794 		for (i = 0; i < adev->usec_timeout; i++) {
2795 			if (!(RREG32(mmCP_HQD_ACTIVE) & CP_HQD_ACTIVE__ACTIVE_MASK))
2796 				break;
2797 			udelay(1);
2798 		}
2799 
2800 		if (i == adev->usec_timeout)
2801 			r = -ETIMEDOUT;
2802 
2803 	}
2804 
2805 	WREG32(mmCP_HQD_DEQUEUE_REQUEST, 0);
2806 	WREG32(mmCP_HQD_PQ_RPTR, 0);
2807 	WREG32(mmCP_HQD_PQ_WPTR, 0);
2808 
2809 	return r;
2810 }
2811 
2812 static void gfx_v7_0_mqd_init(struct amdgpu_device *adev,
2813 			     struct cik_mqd *mqd,
2814 			     uint64_t mqd_gpu_addr,
2815 			     struct amdgpu_ring *ring)
2816 {
2817 	u64 hqd_gpu_addr;
2818 	u64 wb_gpu_addr;
2819 	u32 tmp;
2820 
2821 	/* init the mqd struct */
2822 	memset(mqd, 0, sizeof(struct cik_mqd));
2823 
2824 	mqd->header = 0xC0310800;
2825 	mqd->compute_static_thread_mgmt_se0 = 0xffffffff;
2826 	mqd->compute_static_thread_mgmt_se1 = 0xffffffff;
2827 	mqd->compute_static_thread_mgmt_se2 = 0xffffffff;
2828 	mqd->compute_static_thread_mgmt_se3 = 0xffffffff;
2829 
2830 	/* enable doorbell? */
2831 	mqd->cp_hqd_pq_doorbell_control =
2832 		RREG32(mmCP_HQD_PQ_DOORBELL_CONTROL);
2833 	if (ring->use_doorbell)
2834 		mqd->cp_hqd_pq_doorbell_control |= CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_EN_MASK;
2835 	else
2836 		mqd->cp_hqd_pq_doorbell_control &= ~CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_EN_MASK;
2837 
2838 	/* set the pointer to the MQD */
2839 	mqd->cp_mqd_base_addr_lo = mqd_gpu_addr & 0xfffffffc;
2840 	mqd->cp_mqd_base_addr_hi = upper_32_bits(mqd_gpu_addr);
2841 
2842 	/* set MQD vmid to 0 */
2843 	mqd->cp_mqd_control = RREG32(mmCP_MQD_CONTROL);
2844 	mqd->cp_mqd_control &= ~CP_MQD_CONTROL__VMID_MASK;
2845 
2846 	/* set the pointer to the HQD, this is similar CP_RB0_BASE/_HI */
2847 	hqd_gpu_addr = ring->gpu_addr >> 8;
2848 	mqd->cp_hqd_pq_base_lo = hqd_gpu_addr;
2849 	mqd->cp_hqd_pq_base_hi = upper_32_bits(hqd_gpu_addr);
2850 
2851 	/* set up the HQD, this is similar to CP_RB0_CNTL */
2852 	mqd->cp_hqd_pq_control = RREG32(mmCP_HQD_PQ_CONTROL);
2853 	mqd->cp_hqd_pq_control &=
2854 		~(CP_HQD_PQ_CONTROL__QUEUE_SIZE_MASK |
2855 				CP_HQD_PQ_CONTROL__RPTR_BLOCK_SIZE_MASK);
2856 
2857 	mqd->cp_hqd_pq_control |=
2858 		order_base_2(ring->ring_size / 8);
2859 	mqd->cp_hqd_pq_control |=
2860 		(order_base_2(AMDGPU_GPU_PAGE_SIZE/8) << 8);
2861 #ifdef __BIG_ENDIAN
2862 	mqd->cp_hqd_pq_control |=
2863 		2 << CP_HQD_PQ_CONTROL__ENDIAN_SWAP__SHIFT;
2864 #endif
2865 	mqd->cp_hqd_pq_control &=
2866 		~(CP_HQD_PQ_CONTROL__UNORD_DISPATCH_MASK |
2867 				CP_HQD_PQ_CONTROL__ROQ_PQ_IB_FLIP_MASK |
2868 				CP_HQD_PQ_CONTROL__PQ_VOLATILE_MASK);
2869 	mqd->cp_hqd_pq_control |=
2870 		CP_HQD_PQ_CONTROL__PRIV_STATE_MASK |
2871 		CP_HQD_PQ_CONTROL__KMD_QUEUE_MASK; /* assuming kernel queue control */
2872 
2873 	/* only used if CP_PQ_WPTR_POLL_CNTL.CP_PQ_WPTR_POLL_CNTL__EN_MASK=1 */
2874 	wb_gpu_addr = ring->wptr_gpu_addr;
2875 	mqd->cp_hqd_pq_wptr_poll_addr_lo = wb_gpu_addr & 0xfffffffc;
2876 	mqd->cp_hqd_pq_wptr_poll_addr_hi = upper_32_bits(wb_gpu_addr) & 0xffff;
2877 
2878 	/* set the wb address whether it's enabled or not */
2879 	wb_gpu_addr = ring->rptr_gpu_addr;
2880 	mqd->cp_hqd_pq_rptr_report_addr_lo = wb_gpu_addr & 0xfffffffc;
2881 	mqd->cp_hqd_pq_rptr_report_addr_hi =
2882 		upper_32_bits(wb_gpu_addr) & 0xffff;
2883 
2884 	/* enable the doorbell if requested */
2885 	if (ring->use_doorbell) {
2886 		mqd->cp_hqd_pq_doorbell_control =
2887 			RREG32(mmCP_HQD_PQ_DOORBELL_CONTROL);
2888 		mqd->cp_hqd_pq_doorbell_control &=
2889 			~CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_OFFSET_MASK;
2890 		mqd->cp_hqd_pq_doorbell_control |=
2891 			(ring->doorbell_index <<
2892 			 CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_OFFSET__SHIFT);
2893 		mqd->cp_hqd_pq_doorbell_control |=
2894 			CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_EN_MASK;
2895 		mqd->cp_hqd_pq_doorbell_control &=
2896 			~(CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_SOURCE_MASK |
2897 					CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_HIT_MASK);
2898 
2899 	} else {
2900 		mqd->cp_hqd_pq_doorbell_control = 0;
2901 	}
2902 
2903 	/* read and write pointers, similar to CP_RB0_WPTR/_RPTR */
2904 	ring->wptr = 0;
2905 	mqd->cp_hqd_pq_wptr = lower_32_bits(ring->wptr);
2906 	mqd->cp_hqd_pq_rptr = RREG32(mmCP_HQD_PQ_RPTR);
2907 
2908 	/* set the vmid for the queue */
2909 	mqd->cp_hqd_vmid = 0;
2910 
2911 	/* defaults */
2912 	mqd->cp_hqd_ib_control = RREG32(mmCP_HQD_IB_CONTROL);
2913 	mqd->cp_hqd_ib_base_addr_lo = RREG32(mmCP_HQD_IB_BASE_ADDR);
2914 	mqd->cp_hqd_ib_base_addr_hi = RREG32(mmCP_HQD_IB_BASE_ADDR_HI);
2915 	mqd->cp_hqd_ib_rptr = RREG32(mmCP_HQD_IB_RPTR);
2916 	mqd->cp_hqd_persistent_state = RREG32(mmCP_HQD_PERSISTENT_STATE);
2917 	mqd->cp_hqd_sema_cmd = RREG32(mmCP_HQD_SEMA_CMD);
2918 	mqd->cp_hqd_msg_type = RREG32(mmCP_HQD_MSG_TYPE);
2919 	mqd->cp_hqd_atomic0_preop_lo = RREG32(mmCP_HQD_ATOMIC0_PREOP_LO);
2920 	mqd->cp_hqd_atomic0_preop_hi = RREG32(mmCP_HQD_ATOMIC0_PREOP_HI);
2921 	mqd->cp_hqd_atomic1_preop_lo = RREG32(mmCP_HQD_ATOMIC1_PREOP_LO);
2922 	mqd->cp_hqd_atomic1_preop_hi = RREG32(mmCP_HQD_ATOMIC1_PREOP_HI);
2923 	mqd->cp_hqd_pq_rptr = RREG32(mmCP_HQD_PQ_RPTR);
2924 	tmp = RREG32(mmCP_HQD_QUANTUM);
2925 	tmp = REG_SET_FIELD(tmp, CP_HQD_QUANTUM, QUANTUM_EN, 1);
2926 	tmp = REG_SET_FIELD(tmp, CP_HQD_QUANTUM, QUANTUM_SCALE, 1);
2927 	tmp = REG_SET_FIELD(tmp, CP_HQD_QUANTUM, QUANTUM_DURATION, 10);
2928 	mqd->cp_hqd_quantum = tmp;
2929 	mqd->cp_hqd_pipe_priority = RREG32(mmCP_HQD_PIPE_PRIORITY);
2930 	mqd->cp_hqd_queue_priority = RREG32(mmCP_HQD_QUEUE_PRIORITY);
2931 	mqd->cp_hqd_iq_rptr = RREG32(mmCP_HQD_IQ_RPTR);
2932 
2933 	/* activate the queue */
2934 	mqd->cp_hqd_active = 1;
2935 }
2936 
2937 static int gfx_v7_0_mqd_commit(struct amdgpu_device *adev, struct cik_mqd *mqd)
2938 {
2939 	uint32_t tmp;
2940 	uint32_t mqd_reg;
2941 	uint32_t *mqd_data;
2942 
2943 	/* HQD registers extend from mmCP_MQD_BASE_ADDR to mmCP_MQD_CONTROL */
2944 	mqd_data = &mqd->cp_mqd_base_addr_lo;
2945 
2946 	/* disable wptr polling */
2947 	tmp = RREG32(mmCP_PQ_WPTR_POLL_CNTL);
2948 	tmp = REG_SET_FIELD(tmp, CP_PQ_WPTR_POLL_CNTL, EN, 0);
2949 	WREG32(mmCP_PQ_WPTR_POLL_CNTL, tmp);
2950 
2951 	/* program all HQD registers */
2952 	for (mqd_reg = mmCP_HQD_VMID; mqd_reg <= mmCP_MQD_CONTROL; mqd_reg++)
2953 		WREG32(mqd_reg, mqd_data[mqd_reg - mmCP_MQD_BASE_ADDR]);
2954 
2955 	/* activate the HQD */
2956 	for (mqd_reg = mmCP_MQD_BASE_ADDR; mqd_reg <= mmCP_HQD_ACTIVE; mqd_reg++)
2957 		WREG32(mqd_reg, mqd_data[mqd_reg - mmCP_MQD_BASE_ADDR]);
2958 
2959 	return 0;
2960 }
2961 
2962 static int gfx_v7_0_compute_queue_init(struct amdgpu_device *adev, int ring_id)
2963 {
2964 	struct amdgpu_ring *ring = &adev->gfx.compute_ring[ring_id];
2965 	struct cik_mqd *mqd = ring->mqd_ptr;
2966 	int mqd_idx = ring - &adev->gfx.compute_ring[0];
2967 
2968 	if (!amdgpu_in_reset(adev) && !adev->in_suspend) {
2969 		memset((void *)mqd, 0, ring->mqd_size);
2970 		mutex_lock(&adev->srbm_mutex);
2971 		cik_srbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
2972 		gfx_v7_0_mqd_init(adev, mqd, ring->mqd_gpu_addr, ring);
2973 		gfx_v7_0_deactivate_hqd(adev, 1);
2974 		gfx_v7_0_mqd_commit(adev, mqd);
2975 		cik_srbm_select(adev, 0, 0, 0, 0);
2976 		mutex_unlock(&adev->srbm_mutex);
2977 
2978 		if (adev->gfx.mec.mqd_backup[mqd_idx])
2979 			memcpy(adev->gfx.mec.mqd_backup[mqd_idx], mqd, ring->mqd_size);
2980 	} else {
2981 		/* restore MQD to a clean status */
2982 		if (adev->gfx.mec.mqd_backup[mqd_idx])
2983 			memcpy(mqd, adev->gfx.mec.mqd_backup[mqd_idx], ring->mqd_size);
2984 
2985 		/* Re-commit the restored backup */
2986 		mutex_lock(&adev->srbm_mutex);
2987 		cik_srbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
2988 		gfx_v7_0_deactivate_hqd(adev, 2);
2989 		gfx_v7_0_mqd_commit(adev, mqd);
2990 		cik_srbm_select(adev, 0, 0, 0, 0);
2991 		mutex_unlock(&adev->srbm_mutex);
2992 
2993 		/* reset ring buffer */
2994 		ring->wptr = 0;
2995 		atomic64_set((atomic64_t *)ring->wptr_cpu_addr, 0);
2996 		atomic64_set((atomic64_t *)ring->rptr_cpu_addr, 0);
2997 		amdgpu_ring_clear_ring(ring);
2998 	}
2999 
3000 	return 0;
3001 }
3002 
3003 /**
3004  * gfx_v7_0_cp_compute_resume - setup the compute queue registers
3005  *
3006  * @adev: amdgpu_device pointer
3007  *
3008  * Program the compute queues and test them to make sure they
3009  * are working.
3010  * Returns 0 for success, error for failure.
3011  */
3012 static int gfx_v7_0_cp_compute_resume(struct amdgpu_device *adev)
3013 {
3014 	int r, i, j;
3015 	u32 tmp;
3016 	struct amdgpu_ring *ring;
3017 
3018 	/* fix up chicken bits */
3019 	tmp = RREG32(mmCP_CPF_DEBUG);
3020 	tmp |= (1 << 23);
3021 	WREG32(mmCP_CPF_DEBUG, tmp);
3022 
3023 	/* init all pipes (even the ones we don't own) */
3024 	for (i = 0; i < adev->gfx.mec.num_mec; i++)
3025 		for (j = 0; j < adev->gfx.mec.num_pipe_per_mec; j++)
3026 			gfx_v7_0_compute_pipe_init(adev, i, j);
3027 
3028 	/* init the queues */
3029 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
3030 		r = gfx_v7_0_compute_queue_init(adev, i);
3031 		if (r)
3032 			return r;
3033 	}
3034 
3035 	gfx_v7_0_cp_compute_enable(adev, true);
3036 
3037 	r = 0;
3038 
3039 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
3040 		ring = &adev->gfx.compute_ring[i];
3041 		r |= amdgpu_ring_test_helper(ring);
3042 	}
3043 
3044 	return r;
3045 }
3046 
3047 static void gfx_v7_0_cp_enable(struct amdgpu_device *adev, bool enable)
3048 {
3049 	gfx_v7_0_cp_gfx_enable(adev, enable);
3050 	gfx_v7_0_cp_compute_enable(adev, enable);
3051 }
3052 
3053 static int gfx_v7_0_cp_load_microcode(struct amdgpu_device *adev)
3054 {
3055 	int r;
3056 
3057 	r = gfx_v7_0_cp_gfx_load_microcode(adev);
3058 	if (r)
3059 		return r;
3060 	r = gfx_v7_0_cp_compute_load_microcode(adev);
3061 	if (r)
3062 		return r;
3063 
3064 	return 0;
3065 }
3066 
3067 static void gfx_v7_0_enable_gui_idle_interrupt(struct amdgpu_device *adev,
3068 					       bool enable)
3069 {
3070 	u32 tmp = RREG32(mmCP_INT_CNTL_RING0);
3071 
3072 	if (enable)
3073 		tmp |= (CP_INT_CNTL_RING0__CNTX_BUSY_INT_ENABLE_MASK |
3074 				CP_INT_CNTL_RING0__CNTX_EMPTY_INT_ENABLE_MASK);
3075 	else
3076 		tmp &= ~(CP_INT_CNTL_RING0__CNTX_BUSY_INT_ENABLE_MASK |
3077 				CP_INT_CNTL_RING0__CNTX_EMPTY_INT_ENABLE_MASK);
3078 	WREG32(mmCP_INT_CNTL_RING0, tmp);
3079 }
3080 
3081 static int gfx_v7_0_cp_resume(struct amdgpu_device *adev)
3082 {
3083 	int r;
3084 
3085 	gfx_v7_0_enable_gui_idle_interrupt(adev, false);
3086 
3087 	r = gfx_v7_0_cp_load_microcode(adev);
3088 	if (r)
3089 		return r;
3090 
3091 	r = gfx_v7_0_cp_gfx_resume(adev);
3092 	if (r)
3093 		return r;
3094 	r = gfx_v7_0_cp_compute_resume(adev);
3095 	if (r)
3096 		return r;
3097 
3098 	gfx_v7_0_enable_gui_idle_interrupt(adev, true);
3099 
3100 	return 0;
3101 }
3102 
3103 /**
3104  * gfx_v7_0_ring_emit_pipeline_sync - cik vm flush using the CP
3105  *
3106  * @ring: the ring to emit the commands to
3107  *
3108  * Sync the command pipeline with the PFP. E.g. wait for everything
3109  * to be completed.
3110  */
3111 static void gfx_v7_0_ring_emit_pipeline_sync(struct amdgpu_ring *ring)
3112 {
3113 	int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
3114 	uint32_t seq = ring->fence_drv.sync_seq;
3115 	uint64_t addr = ring->fence_drv.gpu_addr;
3116 
3117 	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
3118 	amdgpu_ring_write(ring, (WAIT_REG_MEM_MEM_SPACE(1) | /* memory */
3119 				 WAIT_REG_MEM_FUNCTION(3) | /* equal */
3120 				 WAIT_REG_MEM_ENGINE(usepfp)));   /* pfp or me */
3121 	amdgpu_ring_write(ring, addr & 0xfffffffc);
3122 	amdgpu_ring_write(ring, upper_32_bits(addr) & 0xffffffff);
3123 	amdgpu_ring_write(ring, seq);
3124 	amdgpu_ring_write(ring, 0xffffffff);
3125 	amdgpu_ring_write(ring, 4); /* poll interval */
3126 }
3127 
3128 /*
3129  * vm
3130  * VMID 0 is the physical GPU addresses as used by the kernel.
3131  * VMIDs 1-15 are used for userspace clients and are handled
3132  * by the amdgpu vm/hsa code.
3133  */
3134 /**
3135  * gfx_v7_0_ring_emit_vm_flush - cik vm flush using the CP
3136  *
3137  * @ring: amdgpu_ring pointer
3138  * @vmid: vmid number to use
3139  * @pd_addr: address
3140  *
3141  * Update the page table base and flush the VM TLB
3142  * using the CP (CIK).
3143  */
3144 static void gfx_v7_0_ring_emit_vm_flush(struct amdgpu_ring *ring,
3145 					unsigned vmid, uint64_t pd_addr)
3146 {
3147 	int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
3148 
3149 	amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr);
3150 
3151 	/* wait for the invalidate to complete */
3152 	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
3153 	amdgpu_ring_write(ring, (WAIT_REG_MEM_OPERATION(0) | /* wait */
3154 				 WAIT_REG_MEM_FUNCTION(0) |  /* always */
3155 				 WAIT_REG_MEM_ENGINE(0))); /* me */
3156 	amdgpu_ring_write(ring, mmVM_INVALIDATE_REQUEST);
3157 	amdgpu_ring_write(ring, 0);
3158 	amdgpu_ring_write(ring, 0); /* ref */
3159 	amdgpu_ring_write(ring, 0); /* mask */
3160 	amdgpu_ring_write(ring, 0x20); /* poll interval */
3161 
3162 	/* compute doesn't have PFP */
3163 	if (usepfp) {
3164 		/* sync PFP to ME, otherwise we might get invalid PFP reads */
3165 		amdgpu_ring_write(ring, PACKET3(PACKET3_PFP_SYNC_ME, 0));
3166 		amdgpu_ring_write(ring, 0x0);
3167 	}
3168 }
3169 
3170 static unsigned int gfx_v7_0_ring_emit_init_cond_exec(struct amdgpu_ring *ring,
3171 						      uint64_t gpu_addr)
3172 {
3173 	unsigned int ret;
3174 
3175 	/* Discard following DWs after this packet when gpu_addr==0 */
3176 	amdgpu_ring_write(ring, PACKET3(PACKET3_COND_EXEC, 3));
3177 	amdgpu_ring_write(ring, lower_32_bits(gpu_addr));
3178 	amdgpu_ring_write(ring, upper_32_bits(gpu_addr));
3179 	amdgpu_ring_write(ring, 0);
3180 	ret = ring->wptr & ring->buf_mask;
3181 	/* patch dummy value later */
3182 	amdgpu_ring_write(ring, 0);
3183 	return ret;
3184 }
3185 
3186 static void gfx_v7_0_ring_emit_wreg(struct amdgpu_ring *ring,
3187 				    uint32_t reg, uint32_t val)
3188 {
3189 	int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
3190 
3191 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
3192 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(usepfp) |
3193 				 WRITE_DATA_DST_SEL(0)));
3194 	amdgpu_ring_write(ring, reg);
3195 	amdgpu_ring_write(ring, 0);
3196 	amdgpu_ring_write(ring, val);
3197 }
3198 
3199 /*
3200  * RLC
3201  * The RLC is a multi-purpose microengine that handles a
3202  * variety of functions.
3203  */
3204 static int gfx_v7_0_rlc_init(struct amdgpu_device *adev)
3205 {
3206 	const u32 *src_ptr;
3207 	u32 dws;
3208 	const struct cs_section_def *cs_data;
3209 	int r;
3210 
3211 	/* allocate rlc buffers */
3212 	if (adev->flags & AMD_IS_APU) {
3213 		if (adev->asic_type == CHIP_KAVERI) {
3214 			adev->gfx.rlc.reg_list = spectre_rlc_save_restore_register_list;
3215 			adev->gfx.rlc.reg_list_size =
3216 				(u32)ARRAY_SIZE(spectre_rlc_save_restore_register_list);
3217 		} else {
3218 			adev->gfx.rlc.reg_list = kalindi_rlc_save_restore_register_list;
3219 			adev->gfx.rlc.reg_list_size =
3220 				(u32)ARRAY_SIZE(kalindi_rlc_save_restore_register_list);
3221 		}
3222 	}
3223 	adev->gfx.rlc.cs_data = ci_cs_data;
3224 	adev->gfx.rlc.cp_table_size = ALIGN(CP_ME_TABLE_SIZE * 5 * 4, 2048); /* CP JT */
3225 	adev->gfx.rlc.cp_table_size += 64 * 1024; /* GDS */
3226 
3227 	src_ptr = adev->gfx.rlc.reg_list;
3228 	dws = adev->gfx.rlc.reg_list_size;
3229 	dws += (5 * 16) + 48 + 48 + 64;
3230 
3231 	cs_data = adev->gfx.rlc.cs_data;
3232 
3233 	if (src_ptr) {
3234 		/* init save restore block */
3235 		r = amdgpu_gfx_rlc_init_sr(adev, dws);
3236 		if (r)
3237 			return r;
3238 	}
3239 
3240 	if (cs_data) {
3241 		/* init clear state block */
3242 		r = amdgpu_gfx_rlc_init_csb(adev);
3243 		if (r)
3244 			return r;
3245 	}
3246 
3247 	if (adev->gfx.rlc.cp_table_size) {
3248 		r = amdgpu_gfx_rlc_init_cpt(adev);
3249 		if (r)
3250 			return r;
3251 	}
3252 
3253 	/* init spm vmid with 0xf */
3254 	if (adev->gfx.rlc.funcs->update_spm_vmid)
3255 		adev->gfx.rlc.funcs->update_spm_vmid(adev, 0, NULL, 0xf);
3256 
3257 	return 0;
3258 }
3259 
3260 static void gfx_v7_0_enable_lbpw(struct amdgpu_device *adev, bool enable)
3261 {
3262 	u32 tmp;
3263 
3264 	tmp = RREG32(mmRLC_LB_CNTL);
3265 	if (enable)
3266 		tmp |= RLC_LB_CNTL__LOAD_BALANCE_ENABLE_MASK;
3267 	else
3268 		tmp &= ~RLC_LB_CNTL__LOAD_BALANCE_ENABLE_MASK;
3269 	WREG32(mmRLC_LB_CNTL, tmp);
3270 }
3271 
3272 static void gfx_v7_0_wait_for_rlc_serdes(struct amdgpu_device *adev)
3273 {
3274 	u32 i, j, k;
3275 	u32 mask;
3276 
3277 	mutex_lock(&adev->grbm_idx_mutex);
3278 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
3279 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
3280 			gfx_v7_0_select_se_sh(adev, i, j, 0xffffffff, 0);
3281 			for (k = 0; k < adev->usec_timeout; k++) {
3282 				if (RREG32(mmRLC_SERDES_CU_MASTER_BUSY) == 0)
3283 					break;
3284 				udelay(1);
3285 			}
3286 		}
3287 	}
3288 	gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0);
3289 	mutex_unlock(&adev->grbm_idx_mutex);
3290 
3291 	mask = RLC_SERDES_NONCU_MASTER_BUSY__SE_MASTER_BUSY_MASK |
3292 		RLC_SERDES_NONCU_MASTER_BUSY__GC_MASTER_BUSY_MASK |
3293 		RLC_SERDES_NONCU_MASTER_BUSY__TC0_MASTER_BUSY_MASK |
3294 		RLC_SERDES_NONCU_MASTER_BUSY__TC1_MASTER_BUSY_MASK;
3295 	for (k = 0; k < adev->usec_timeout; k++) {
3296 		if ((RREG32(mmRLC_SERDES_NONCU_MASTER_BUSY) & mask) == 0)
3297 			break;
3298 		udelay(1);
3299 	}
3300 }
3301 
3302 static void gfx_v7_0_update_rlc(struct amdgpu_device *adev, u32 rlc)
3303 {
3304 	u32 tmp;
3305 
3306 	tmp = RREG32(mmRLC_CNTL);
3307 	if (tmp != rlc)
3308 		WREG32(mmRLC_CNTL, rlc);
3309 }
3310 
3311 static u32 gfx_v7_0_halt_rlc(struct amdgpu_device *adev)
3312 {
3313 	u32 data, orig;
3314 
3315 	orig = data = RREG32(mmRLC_CNTL);
3316 
3317 	if (data & RLC_CNTL__RLC_ENABLE_F32_MASK) {
3318 		u32 i;
3319 
3320 		data &= ~RLC_CNTL__RLC_ENABLE_F32_MASK;
3321 		WREG32(mmRLC_CNTL, data);
3322 
3323 		for (i = 0; i < adev->usec_timeout; i++) {
3324 			if ((RREG32(mmRLC_GPM_STAT) & RLC_GPM_STAT__RLC_BUSY_MASK) == 0)
3325 				break;
3326 			udelay(1);
3327 		}
3328 
3329 		gfx_v7_0_wait_for_rlc_serdes(adev);
3330 	}
3331 
3332 	return orig;
3333 }
3334 
3335 static bool gfx_v7_0_is_rlc_enabled(struct amdgpu_device *adev)
3336 {
3337 	return true;
3338 }
3339 
3340 static void gfx_v7_0_set_safe_mode(struct amdgpu_device *adev, int xcc_id)
3341 {
3342 	u32 tmp, i, mask;
3343 
3344 	tmp = 0x1 | (1 << 1);
3345 	WREG32(mmRLC_GPR_REG2, tmp);
3346 
3347 	mask = RLC_GPM_STAT__GFX_POWER_STATUS_MASK |
3348 		RLC_GPM_STAT__GFX_CLOCK_STATUS_MASK;
3349 	for (i = 0; i < adev->usec_timeout; i++) {
3350 		if ((RREG32(mmRLC_GPM_STAT) & mask) == mask)
3351 			break;
3352 		udelay(1);
3353 	}
3354 
3355 	for (i = 0; i < adev->usec_timeout; i++) {
3356 		if ((RREG32(mmRLC_GPR_REG2) & 0x1) == 0)
3357 			break;
3358 		udelay(1);
3359 	}
3360 }
3361 
3362 static void gfx_v7_0_unset_safe_mode(struct amdgpu_device *adev, int xcc_id)
3363 {
3364 	u32 tmp;
3365 
3366 	tmp = 0x1 | (0 << 1);
3367 	WREG32(mmRLC_GPR_REG2, tmp);
3368 }
3369 
3370 /**
3371  * gfx_v7_0_rlc_stop - stop the RLC ME
3372  *
3373  * @adev: amdgpu_device pointer
3374  *
3375  * Halt the RLC ME (MicroEngine) (CIK).
3376  */
3377 static void gfx_v7_0_rlc_stop(struct amdgpu_device *adev)
3378 {
3379 	WREG32(mmRLC_CNTL, 0);
3380 
3381 	gfx_v7_0_enable_gui_idle_interrupt(adev, false);
3382 
3383 	gfx_v7_0_wait_for_rlc_serdes(adev);
3384 }
3385 
3386 /**
3387  * gfx_v7_0_rlc_start - start the RLC ME
3388  *
3389  * @adev: amdgpu_device pointer
3390  *
3391  * Unhalt the RLC ME (MicroEngine) (CIK).
3392  */
3393 static void gfx_v7_0_rlc_start(struct amdgpu_device *adev)
3394 {
3395 	WREG32(mmRLC_CNTL, RLC_CNTL__RLC_ENABLE_F32_MASK);
3396 
3397 	gfx_v7_0_enable_gui_idle_interrupt(adev, true);
3398 
3399 	udelay(50);
3400 }
3401 
3402 static void gfx_v7_0_rlc_reset(struct amdgpu_device *adev)
3403 {
3404 	u32 tmp = RREG32(mmGRBM_SOFT_RESET);
3405 
3406 	tmp |= GRBM_SOFT_RESET__SOFT_RESET_RLC_MASK;
3407 	WREG32(mmGRBM_SOFT_RESET, tmp);
3408 	udelay(50);
3409 	tmp &= ~GRBM_SOFT_RESET__SOFT_RESET_RLC_MASK;
3410 	WREG32(mmGRBM_SOFT_RESET, tmp);
3411 	udelay(50);
3412 }
3413 
3414 /**
3415  * gfx_v7_0_rlc_resume - setup the RLC hw
3416  *
3417  * @adev: amdgpu_device pointer
3418  *
3419  * Initialize the RLC registers, load the ucode,
3420  * and start the RLC (CIK).
3421  * Returns 0 for success, -EINVAL if the ucode is not available.
3422  */
3423 static int gfx_v7_0_rlc_resume(struct amdgpu_device *adev)
3424 {
3425 	const struct rlc_firmware_header_v1_0 *hdr;
3426 	const __le32 *fw_data;
3427 	unsigned i, fw_size;
3428 	u32 tmp;
3429 
3430 	if (!adev->gfx.rlc_fw)
3431 		return -EINVAL;
3432 
3433 	hdr = (const struct rlc_firmware_header_v1_0 *)adev->gfx.rlc_fw->data;
3434 	amdgpu_ucode_print_rlc_hdr(&hdr->header);
3435 	adev->gfx.rlc_fw_version = le32_to_cpu(hdr->header.ucode_version);
3436 	adev->gfx.rlc_feature_version = le32_to_cpu(
3437 					hdr->ucode_feature_version);
3438 
3439 	adev->gfx.rlc.funcs->stop(adev);
3440 
3441 	/* disable CG */
3442 	tmp = RREG32(mmRLC_CGCG_CGLS_CTRL) & 0xfffffffc;
3443 	WREG32(mmRLC_CGCG_CGLS_CTRL, tmp);
3444 
3445 	adev->gfx.rlc.funcs->reset(adev);
3446 
3447 	gfx_v7_0_init_pg(adev);
3448 
3449 	WREG32(mmRLC_LB_CNTR_INIT, 0);
3450 	WREG32(mmRLC_LB_CNTR_MAX, 0x00008000);
3451 
3452 	mutex_lock(&adev->grbm_idx_mutex);
3453 	gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0);
3454 	WREG32(mmRLC_LB_INIT_CU_MASK, 0xffffffff);
3455 	WREG32(mmRLC_LB_PARAMS, 0x00600408);
3456 	WREG32(mmRLC_LB_CNTL, 0x80000004);
3457 	mutex_unlock(&adev->grbm_idx_mutex);
3458 
3459 	WREG32(mmRLC_MC_CNTL, 0);
3460 	WREG32(mmRLC_UCODE_CNTL, 0);
3461 
3462 	fw_data = (const __le32 *)
3463 		(adev->gfx.rlc_fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes));
3464 	fw_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
3465 	WREG32(mmRLC_GPM_UCODE_ADDR, 0);
3466 	for (i = 0; i < fw_size; i++)
3467 		WREG32(mmRLC_GPM_UCODE_DATA, le32_to_cpup(fw_data++));
3468 	WREG32(mmRLC_GPM_UCODE_ADDR, adev->gfx.rlc_fw_version);
3469 
3470 	/* XXX - find out what chips support lbpw */
3471 	gfx_v7_0_enable_lbpw(adev, false);
3472 
3473 	if (adev->asic_type == CHIP_BONAIRE)
3474 		WREG32(mmRLC_DRIVER_CPDMA_STATUS, 0);
3475 
3476 	adev->gfx.rlc.funcs->start(adev);
3477 
3478 	return 0;
3479 }
3480 
3481 static void gfx_v7_0_update_spm_vmid(struct amdgpu_device *adev, int xcc_id,
3482 		struct amdgpu_ring *ring, unsigned vmid)
3483 {
3484 	u32 data;
3485 
3486 	amdgpu_gfx_off_ctrl(adev, false);
3487 
3488 	data = RREG32(mmRLC_SPM_VMID);
3489 
3490 	data &= ~RLC_SPM_VMID__RLC_SPM_VMID_MASK;
3491 	data |= (vmid & RLC_SPM_VMID__RLC_SPM_VMID_MASK) << RLC_SPM_VMID__RLC_SPM_VMID__SHIFT;
3492 
3493 	WREG32(mmRLC_SPM_VMID, data);
3494 
3495 	amdgpu_gfx_off_ctrl(adev, true);
3496 }
3497 
3498 static void gfx_v7_0_enable_cgcg(struct amdgpu_device *adev, bool enable)
3499 {
3500 	u32 data, orig, tmp, tmp2;
3501 
3502 	orig = data = RREG32(mmRLC_CGCG_CGLS_CTRL);
3503 
3504 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG)) {
3505 		gfx_v7_0_enable_gui_idle_interrupt(adev, true);
3506 
3507 		tmp = gfx_v7_0_halt_rlc(adev);
3508 
3509 		mutex_lock(&adev->grbm_idx_mutex);
3510 		gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0);
3511 		WREG32(mmRLC_SERDES_WR_CU_MASTER_MASK, 0xffffffff);
3512 		WREG32(mmRLC_SERDES_WR_NONCU_MASTER_MASK, 0xffffffff);
3513 		tmp2 = RLC_SERDES_WR_CTRL__BPM_ADDR_MASK |
3514 			RLC_SERDES_WR_CTRL__CGCG_OVERRIDE_0_MASK |
3515 			RLC_SERDES_WR_CTRL__CGLS_ENABLE_MASK;
3516 		WREG32(mmRLC_SERDES_WR_CTRL, tmp2);
3517 		mutex_unlock(&adev->grbm_idx_mutex);
3518 
3519 		gfx_v7_0_update_rlc(adev, tmp);
3520 
3521 		data |= RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK | RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK;
3522 		if (orig != data)
3523 			WREG32(mmRLC_CGCG_CGLS_CTRL, data);
3524 
3525 	} else {
3526 		gfx_v7_0_enable_gui_idle_interrupt(adev, false);
3527 
3528 		RREG32(mmCB_CGTT_SCLK_CTRL);
3529 		RREG32(mmCB_CGTT_SCLK_CTRL);
3530 		RREG32(mmCB_CGTT_SCLK_CTRL);
3531 		RREG32(mmCB_CGTT_SCLK_CTRL);
3532 
3533 		data &= ~(RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK | RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK);
3534 		if (orig != data)
3535 			WREG32(mmRLC_CGCG_CGLS_CTRL, data);
3536 
3537 		gfx_v7_0_enable_gui_idle_interrupt(adev, true);
3538 	}
3539 }
3540 
3541 static void gfx_v7_0_enable_mgcg(struct amdgpu_device *adev, bool enable)
3542 {
3543 	u32 data, orig, tmp = 0;
3544 
3545 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG)) {
3546 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) {
3547 			if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CP_LS) {
3548 				orig = data = RREG32(mmCP_MEM_SLP_CNTL);
3549 				data |= CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK;
3550 				if (orig != data)
3551 					WREG32(mmCP_MEM_SLP_CNTL, data);
3552 			}
3553 		}
3554 
3555 		orig = data = RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
3556 		data |= 0x00000001;
3557 		data &= 0xfffffffd;
3558 		if (orig != data)
3559 			WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data);
3560 
3561 		tmp = gfx_v7_0_halt_rlc(adev);
3562 
3563 		mutex_lock(&adev->grbm_idx_mutex);
3564 		gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0);
3565 		WREG32(mmRLC_SERDES_WR_CU_MASTER_MASK, 0xffffffff);
3566 		WREG32(mmRLC_SERDES_WR_NONCU_MASTER_MASK, 0xffffffff);
3567 		data = RLC_SERDES_WR_CTRL__BPM_ADDR_MASK |
3568 			RLC_SERDES_WR_CTRL__MGCG_OVERRIDE_0_MASK;
3569 		WREG32(mmRLC_SERDES_WR_CTRL, data);
3570 		mutex_unlock(&adev->grbm_idx_mutex);
3571 
3572 		gfx_v7_0_update_rlc(adev, tmp);
3573 
3574 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGTS) {
3575 			orig = data = RREG32(mmCGTS_SM_CTRL_REG);
3576 			data &= ~CGTS_SM_CTRL_REG__SM_MODE_MASK;
3577 			data |= (0x2 << CGTS_SM_CTRL_REG__SM_MODE__SHIFT);
3578 			data |= CGTS_SM_CTRL_REG__SM_MODE_ENABLE_MASK;
3579 			data &= ~CGTS_SM_CTRL_REG__OVERRIDE_MASK;
3580 			if ((adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) &&
3581 			    (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGTS_LS))
3582 				data &= ~CGTS_SM_CTRL_REG__LS_OVERRIDE_MASK;
3583 			data &= ~CGTS_SM_CTRL_REG__ON_MONITOR_ADD_MASK;
3584 			data |= CGTS_SM_CTRL_REG__ON_MONITOR_ADD_EN_MASK;
3585 			data |= (0x96 << CGTS_SM_CTRL_REG__ON_MONITOR_ADD__SHIFT);
3586 			if (orig != data)
3587 				WREG32(mmCGTS_SM_CTRL_REG, data);
3588 		}
3589 	} else {
3590 		orig = data = RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
3591 		data |= 0x00000003;
3592 		if (orig != data)
3593 			WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data);
3594 
3595 		data = RREG32(mmRLC_MEM_SLP_CNTL);
3596 		if (data & RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK) {
3597 			data &= ~RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK;
3598 			WREG32(mmRLC_MEM_SLP_CNTL, data);
3599 		}
3600 
3601 		data = RREG32(mmCP_MEM_SLP_CNTL);
3602 		if (data & CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK) {
3603 			data &= ~CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK;
3604 			WREG32(mmCP_MEM_SLP_CNTL, data);
3605 		}
3606 
3607 		orig = data = RREG32(mmCGTS_SM_CTRL_REG);
3608 		data |= CGTS_SM_CTRL_REG__OVERRIDE_MASK | CGTS_SM_CTRL_REG__LS_OVERRIDE_MASK;
3609 		if (orig != data)
3610 			WREG32(mmCGTS_SM_CTRL_REG, data);
3611 
3612 		tmp = gfx_v7_0_halt_rlc(adev);
3613 
3614 		mutex_lock(&adev->grbm_idx_mutex);
3615 		gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0);
3616 		WREG32(mmRLC_SERDES_WR_CU_MASTER_MASK, 0xffffffff);
3617 		WREG32(mmRLC_SERDES_WR_NONCU_MASTER_MASK, 0xffffffff);
3618 		data = RLC_SERDES_WR_CTRL__BPM_ADDR_MASK | RLC_SERDES_WR_CTRL__MGCG_OVERRIDE_1_MASK;
3619 		WREG32(mmRLC_SERDES_WR_CTRL, data);
3620 		mutex_unlock(&adev->grbm_idx_mutex);
3621 
3622 		gfx_v7_0_update_rlc(adev, tmp);
3623 	}
3624 }
3625 
3626 static void gfx_v7_0_enable_sclk_slowdown_on_pu(struct amdgpu_device *adev,
3627 						bool enable)
3628 {
3629 	u32 data, orig;
3630 
3631 	orig = data = RREG32(mmRLC_PG_CNTL);
3632 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_RLC_SMU_HS))
3633 		data |= RLC_PG_CNTL__SMU_CLK_SLOWDOWN_ON_PU_ENABLE_MASK;
3634 	else
3635 		data &= ~RLC_PG_CNTL__SMU_CLK_SLOWDOWN_ON_PU_ENABLE_MASK;
3636 	if (orig != data)
3637 		WREG32(mmRLC_PG_CNTL, data);
3638 }
3639 
3640 static void gfx_v7_0_enable_sclk_slowdown_on_pd(struct amdgpu_device *adev,
3641 						bool enable)
3642 {
3643 	u32 data, orig;
3644 
3645 	orig = data = RREG32(mmRLC_PG_CNTL);
3646 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_RLC_SMU_HS))
3647 		data |= RLC_PG_CNTL__SMU_CLK_SLOWDOWN_ON_PD_ENABLE_MASK;
3648 	else
3649 		data &= ~RLC_PG_CNTL__SMU_CLK_SLOWDOWN_ON_PD_ENABLE_MASK;
3650 	if (orig != data)
3651 		WREG32(mmRLC_PG_CNTL, data);
3652 }
3653 
3654 static void gfx_v7_0_enable_cp_pg(struct amdgpu_device *adev, bool enable)
3655 {
3656 	u32 data, orig;
3657 
3658 	orig = data = RREG32(mmRLC_PG_CNTL);
3659 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_CP))
3660 		data &= ~0x8000;
3661 	else
3662 		data |= 0x8000;
3663 	if (orig != data)
3664 		WREG32(mmRLC_PG_CNTL, data);
3665 }
3666 
3667 static void gfx_v7_0_enable_gds_pg(struct amdgpu_device *adev, bool enable)
3668 {
3669 	u32 data, orig;
3670 
3671 	orig = data = RREG32(mmRLC_PG_CNTL);
3672 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_GDS))
3673 		data &= ~0x2000;
3674 	else
3675 		data |= 0x2000;
3676 	if (orig != data)
3677 		WREG32(mmRLC_PG_CNTL, data);
3678 }
3679 
3680 static int gfx_v7_0_cp_pg_table_num(struct amdgpu_device *adev)
3681 {
3682 	if (adev->asic_type == CHIP_KAVERI)
3683 		return 5;
3684 	else
3685 		return 4;
3686 }
3687 
3688 static void gfx_v7_0_enable_gfx_cgpg(struct amdgpu_device *adev,
3689 				     bool enable)
3690 {
3691 	u32 data, orig;
3692 
3693 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_GFX_PG)) {
3694 		orig = data = RREG32(mmRLC_PG_CNTL);
3695 		data |= RLC_PG_CNTL__GFX_POWER_GATING_ENABLE_MASK;
3696 		if (orig != data)
3697 			WREG32(mmRLC_PG_CNTL, data);
3698 
3699 		orig = data = RREG32(mmRLC_AUTO_PG_CTRL);
3700 		data |= RLC_AUTO_PG_CTRL__AUTO_PG_EN_MASK;
3701 		if (orig != data)
3702 			WREG32(mmRLC_AUTO_PG_CTRL, data);
3703 	} else {
3704 		orig = data = RREG32(mmRLC_PG_CNTL);
3705 		data &= ~RLC_PG_CNTL__GFX_POWER_GATING_ENABLE_MASK;
3706 		if (orig != data)
3707 			WREG32(mmRLC_PG_CNTL, data);
3708 
3709 		orig = data = RREG32(mmRLC_AUTO_PG_CTRL);
3710 		data &= ~RLC_AUTO_PG_CTRL__AUTO_PG_EN_MASK;
3711 		if (orig != data)
3712 			WREG32(mmRLC_AUTO_PG_CTRL, data);
3713 
3714 		data = RREG32(mmDB_RENDER_CONTROL);
3715 	}
3716 }
3717 
3718 static void gfx_v7_0_set_user_cu_inactive_bitmap(struct amdgpu_device *adev,
3719 						 u32 bitmap)
3720 {
3721 	u32 data;
3722 
3723 	if (!bitmap)
3724 		return;
3725 
3726 	data = bitmap << GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS__SHIFT;
3727 	data &= GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS_MASK;
3728 
3729 	WREG32(mmGC_USER_SHADER_ARRAY_CONFIG, data);
3730 }
3731 
3732 static u32 gfx_v7_0_get_cu_active_bitmap(struct amdgpu_device *adev)
3733 {
3734 	u32 data, mask;
3735 
3736 	data = RREG32(mmCC_GC_SHADER_ARRAY_CONFIG);
3737 	data |= RREG32(mmGC_USER_SHADER_ARRAY_CONFIG);
3738 
3739 	data &= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_CUS_MASK;
3740 	data >>= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_CUS__SHIFT;
3741 
3742 	mask = amdgpu_gfx_create_bitmask(adev->gfx.config.max_cu_per_sh);
3743 
3744 	return (~data) & mask;
3745 }
3746 
3747 static void gfx_v7_0_init_ao_cu_mask(struct amdgpu_device *adev)
3748 {
3749 	u32 tmp;
3750 
3751 	WREG32(mmRLC_PG_ALWAYS_ON_CU_MASK, adev->gfx.cu_info.ao_cu_mask);
3752 
3753 	tmp = RREG32(mmRLC_MAX_PG_CU);
3754 	tmp &= ~RLC_MAX_PG_CU__MAX_POWERED_UP_CU_MASK;
3755 	tmp |= (adev->gfx.cu_info.number << RLC_MAX_PG_CU__MAX_POWERED_UP_CU__SHIFT);
3756 	WREG32(mmRLC_MAX_PG_CU, tmp);
3757 }
3758 
3759 static void gfx_v7_0_enable_gfx_static_mgpg(struct amdgpu_device *adev,
3760 					    bool enable)
3761 {
3762 	u32 data, orig;
3763 
3764 	orig = data = RREG32(mmRLC_PG_CNTL);
3765 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_GFX_SMG))
3766 		data |= RLC_PG_CNTL__STATIC_PER_CU_PG_ENABLE_MASK;
3767 	else
3768 		data &= ~RLC_PG_CNTL__STATIC_PER_CU_PG_ENABLE_MASK;
3769 	if (orig != data)
3770 		WREG32(mmRLC_PG_CNTL, data);
3771 }
3772 
3773 static void gfx_v7_0_enable_gfx_dynamic_mgpg(struct amdgpu_device *adev,
3774 					     bool enable)
3775 {
3776 	u32 data, orig;
3777 
3778 	orig = data = RREG32(mmRLC_PG_CNTL);
3779 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_GFX_DMG))
3780 		data |= RLC_PG_CNTL__DYN_PER_CU_PG_ENABLE_MASK;
3781 	else
3782 		data &= ~RLC_PG_CNTL__DYN_PER_CU_PG_ENABLE_MASK;
3783 	if (orig != data)
3784 		WREG32(mmRLC_PG_CNTL, data);
3785 }
3786 
3787 #define RLC_SAVE_AND_RESTORE_STARTING_OFFSET 0x90
3788 #define RLC_CLEAR_STATE_DESCRIPTOR_OFFSET    0x3D
3789 
3790 static void gfx_v7_0_init_gfx_cgpg(struct amdgpu_device *adev)
3791 {
3792 	u32 data, orig;
3793 	u32 i;
3794 
3795 	if (adev->gfx.rlc.cs_data) {
3796 		WREG32(mmRLC_GPM_SCRATCH_ADDR, RLC_CLEAR_STATE_DESCRIPTOR_OFFSET);
3797 		WREG32(mmRLC_GPM_SCRATCH_DATA, upper_32_bits(adev->gfx.rlc.clear_state_gpu_addr));
3798 		WREG32(mmRLC_GPM_SCRATCH_DATA, lower_32_bits(adev->gfx.rlc.clear_state_gpu_addr));
3799 		WREG32(mmRLC_GPM_SCRATCH_DATA, adev->gfx.rlc.clear_state_size);
3800 	} else {
3801 		WREG32(mmRLC_GPM_SCRATCH_ADDR, RLC_CLEAR_STATE_DESCRIPTOR_OFFSET);
3802 		for (i = 0; i < 3; i++)
3803 			WREG32(mmRLC_GPM_SCRATCH_DATA, 0);
3804 	}
3805 	if (adev->gfx.rlc.reg_list) {
3806 		WREG32(mmRLC_GPM_SCRATCH_ADDR, RLC_SAVE_AND_RESTORE_STARTING_OFFSET);
3807 		for (i = 0; i < adev->gfx.rlc.reg_list_size; i++)
3808 			WREG32(mmRLC_GPM_SCRATCH_DATA, adev->gfx.rlc.reg_list[i]);
3809 	}
3810 
3811 	orig = data = RREG32(mmRLC_PG_CNTL);
3812 	data |= RLC_PG_CNTL__GFX_POWER_GATING_SRC_MASK;
3813 	if (orig != data)
3814 		WREG32(mmRLC_PG_CNTL, data);
3815 
3816 	WREG32(mmRLC_SAVE_AND_RESTORE_BASE, adev->gfx.rlc.save_restore_gpu_addr >> 8);
3817 	WREG32(mmRLC_JUMP_TABLE_RESTORE, adev->gfx.rlc.cp_table_gpu_addr >> 8);
3818 
3819 	data = RREG32(mmCP_RB_WPTR_POLL_CNTL);
3820 	data &= ~CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT_MASK;
3821 	data |= (0x60 << CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT__SHIFT);
3822 	WREG32(mmCP_RB_WPTR_POLL_CNTL, data);
3823 
3824 	data = 0x10101010;
3825 	WREG32(mmRLC_PG_DELAY, data);
3826 
3827 	data = RREG32(mmRLC_PG_DELAY_2);
3828 	data &= ~0xff;
3829 	data |= 0x3;
3830 	WREG32(mmRLC_PG_DELAY_2, data);
3831 
3832 	data = RREG32(mmRLC_AUTO_PG_CTRL);
3833 	data &= ~RLC_AUTO_PG_CTRL__GRBM_REG_SAVE_GFX_IDLE_THRESHOLD_MASK;
3834 	data |= (0x700 << RLC_AUTO_PG_CTRL__GRBM_REG_SAVE_GFX_IDLE_THRESHOLD__SHIFT);
3835 	WREG32(mmRLC_AUTO_PG_CTRL, data);
3836 
3837 }
3838 
3839 static void gfx_v7_0_update_gfx_pg(struct amdgpu_device *adev, bool enable)
3840 {
3841 	gfx_v7_0_enable_gfx_cgpg(adev, enable);
3842 	gfx_v7_0_enable_gfx_static_mgpg(adev, enable);
3843 	gfx_v7_0_enable_gfx_dynamic_mgpg(adev, enable);
3844 }
3845 
3846 static u32 gfx_v7_0_get_csb_size(struct amdgpu_device *adev)
3847 {
3848 	u32 count = 0;
3849 	const struct cs_section_def *sect = NULL;
3850 	const struct cs_extent_def *ext = NULL;
3851 
3852 	if (adev->gfx.rlc.cs_data == NULL)
3853 		return 0;
3854 
3855 	/* begin clear state */
3856 	count += 2;
3857 	/* context control state */
3858 	count += 3;
3859 
3860 	for (sect = adev->gfx.rlc.cs_data; sect->section != NULL; ++sect) {
3861 		for (ext = sect->section; ext->extent != NULL; ++ext) {
3862 			if (sect->id == SECT_CONTEXT)
3863 				count += 2 + ext->reg_count;
3864 			else
3865 				return 0;
3866 		}
3867 	}
3868 	/* pa_sc_raster_config/pa_sc_raster_config1 */
3869 	count += 4;
3870 	/* end clear state */
3871 	count += 2;
3872 	/* clear state */
3873 	count += 2;
3874 
3875 	return count;
3876 }
3877 
3878 static void gfx_v7_0_get_csb_buffer(struct amdgpu_device *adev, u32 *buffer)
3879 {
3880 	u32 count = 0;
3881 
3882 	if (adev->gfx.rlc.cs_data == NULL)
3883 		return;
3884 	if (buffer == NULL)
3885 		return;
3886 
3887 	count = amdgpu_gfx_csb_preamble_start(buffer);
3888 	count = amdgpu_gfx_csb_data_parser(adev, buffer, count);
3889 
3890 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, 2));
3891 	buffer[count++] = cpu_to_le32(mmPA_SC_RASTER_CONFIG - PACKET3_SET_CONTEXT_REG_START);
3892 	buffer[count++] = cpu_to_le32(adev->gfx.config.rb_config[0][0].raster_config);
3893 	buffer[count++] = cpu_to_le32(adev->gfx.config.rb_config[0][0].raster_config_1);
3894 
3895 	amdgpu_gfx_csb_preamble_end(buffer, count);
3896 }
3897 
3898 static void gfx_v7_0_init_pg(struct amdgpu_device *adev)
3899 {
3900 	if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_PG |
3901 			      AMD_PG_SUPPORT_GFX_SMG |
3902 			      AMD_PG_SUPPORT_GFX_DMG |
3903 			      AMD_PG_SUPPORT_CP |
3904 			      AMD_PG_SUPPORT_GDS |
3905 			      AMD_PG_SUPPORT_RLC_SMU_HS)) {
3906 		gfx_v7_0_enable_sclk_slowdown_on_pu(adev, true);
3907 		gfx_v7_0_enable_sclk_slowdown_on_pd(adev, true);
3908 		if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PG) {
3909 			gfx_v7_0_init_gfx_cgpg(adev);
3910 			gfx_v7_0_enable_cp_pg(adev, true);
3911 			gfx_v7_0_enable_gds_pg(adev, true);
3912 		}
3913 		gfx_v7_0_init_ao_cu_mask(adev);
3914 		gfx_v7_0_update_gfx_pg(adev, true);
3915 	}
3916 }
3917 
3918 static void gfx_v7_0_fini_pg(struct amdgpu_device *adev)
3919 {
3920 	if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_PG |
3921 			      AMD_PG_SUPPORT_GFX_SMG |
3922 			      AMD_PG_SUPPORT_GFX_DMG |
3923 			      AMD_PG_SUPPORT_CP |
3924 			      AMD_PG_SUPPORT_GDS |
3925 			      AMD_PG_SUPPORT_RLC_SMU_HS)) {
3926 		gfx_v7_0_update_gfx_pg(adev, false);
3927 		if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PG) {
3928 			gfx_v7_0_enable_cp_pg(adev, false);
3929 			gfx_v7_0_enable_gds_pg(adev, false);
3930 		}
3931 	}
3932 }
3933 
3934 /**
3935  * gfx_v7_0_get_gpu_clock_counter - return GPU clock counter snapshot
3936  *
3937  * @adev: amdgpu_device pointer
3938  *
3939  * Fetches a GPU clock counter snapshot (SI).
3940  * Returns the 64 bit clock counter snapshot.
3941  */
3942 static uint64_t gfx_v7_0_get_gpu_clock_counter(struct amdgpu_device *adev)
3943 {
3944 	uint64_t clock;
3945 
3946 	mutex_lock(&adev->gfx.gpu_clock_mutex);
3947 	WREG32(mmRLC_CAPTURE_GPU_CLOCK_COUNT, 1);
3948 	clock = (uint64_t)RREG32(mmRLC_GPU_CLOCK_COUNT_LSB) |
3949 		((uint64_t)RREG32(mmRLC_GPU_CLOCK_COUNT_MSB) << 32ULL);
3950 	mutex_unlock(&adev->gfx.gpu_clock_mutex);
3951 	return clock;
3952 }
3953 
3954 static void gfx_v7_0_ring_emit_gds_switch(struct amdgpu_ring *ring,
3955 					  uint32_t vmid,
3956 					  uint32_t gds_base, uint32_t gds_size,
3957 					  uint32_t gws_base, uint32_t gws_size,
3958 					  uint32_t oa_base, uint32_t oa_size)
3959 {
3960 	/* GDS Base */
3961 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
3962 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
3963 				WRITE_DATA_DST_SEL(0)));
3964 	amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].mem_base);
3965 	amdgpu_ring_write(ring, 0);
3966 	amdgpu_ring_write(ring, gds_base);
3967 
3968 	/* GDS Size */
3969 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
3970 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
3971 				WRITE_DATA_DST_SEL(0)));
3972 	amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].mem_size);
3973 	amdgpu_ring_write(ring, 0);
3974 	amdgpu_ring_write(ring, gds_size);
3975 
3976 	/* GWS */
3977 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
3978 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
3979 				WRITE_DATA_DST_SEL(0)));
3980 	amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].gws);
3981 	amdgpu_ring_write(ring, 0);
3982 	amdgpu_ring_write(ring, gws_size << GDS_GWS_VMID0__SIZE__SHIFT | gws_base);
3983 
3984 	/* OA */
3985 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
3986 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
3987 				WRITE_DATA_DST_SEL(0)));
3988 	amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].oa);
3989 	amdgpu_ring_write(ring, 0);
3990 	amdgpu_ring_write(ring, (1 << (oa_size + oa_base)) - (1 << oa_base));
3991 }
3992 
3993 static void gfx_v7_0_ring_soft_recovery(struct amdgpu_ring *ring, unsigned vmid)
3994 {
3995 	struct amdgpu_device *adev = ring->adev;
3996 	uint32_t value = 0;
3997 
3998 	value = REG_SET_FIELD(value, SQ_CMD, CMD, 0x03);
3999 	value = REG_SET_FIELD(value, SQ_CMD, MODE, 0x01);
4000 	value = REG_SET_FIELD(value, SQ_CMD, CHECK_VMID, 1);
4001 	value = REG_SET_FIELD(value, SQ_CMD, VM_ID, vmid);
4002 	WREG32(mmSQ_CMD, value);
4003 }
4004 
4005 static uint32_t wave_read_ind(struct amdgpu_device *adev, uint32_t simd, uint32_t wave, uint32_t address)
4006 {
4007 	WREG32(mmSQ_IND_INDEX,
4008 		(wave << SQ_IND_INDEX__WAVE_ID__SHIFT) |
4009 		(simd << SQ_IND_INDEX__SIMD_ID__SHIFT) |
4010 		(address << SQ_IND_INDEX__INDEX__SHIFT) |
4011 		(SQ_IND_INDEX__FORCE_READ_MASK));
4012 	return RREG32(mmSQ_IND_DATA);
4013 }
4014 
4015 static void wave_read_regs(struct amdgpu_device *adev, uint32_t simd,
4016 			   uint32_t wave, uint32_t thread,
4017 			   uint32_t regno, uint32_t num, uint32_t *out)
4018 {
4019 	WREG32(mmSQ_IND_INDEX,
4020 		(wave << SQ_IND_INDEX__WAVE_ID__SHIFT) |
4021 		(simd << SQ_IND_INDEX__SIMD_ID__SHIFT) |
4022 		(regno << SQ_IND_INDEX__INDEX__SHIFT) |
4023 		(thread << SQ_IND_INDEX__THREAD_ID__SHIFT) |
4024 		(SQ_IND_INDEX__FORCE_READ_MASK) |
4025 		(SQ_IND_INDEX__AUTO_INCR_MASK));
4026 	while (num--)
4027 		*(out++) = RREG32(mmSQ_IND_DATA);
4028 }
4029 
4030 static void gfx_v7_0_read_wave_data(struct amdgpu_device *adev, uint32_t xcc_id, uint32_t simd, uint32_t wave, uint32_t *dst, int *no_fields)
4031 {
4032 	/* type 0 wave data */
4033 	dst[(*no_fields)++] = 0;
4034 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_STATUS);
4035 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_LO);
4036 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_HI);
4037 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_LO);
4038 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_HI);
4039 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_HW_ID);
4040 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW0);
4041 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW1);
4042 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_GPR_ALLOC);
4043 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_LDS_ALLOC);
4044 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TRAPSTS);
4045 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_IB_STS);
4046 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TBA_LO);
4047 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TBA_HI);
4048 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TMA_LO);
4049 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TMA_HI);
4050 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_IB_DBG0);
4051 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_M0);
4052 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_MODE);
4053 }
4054 
4055 static void gfx_v7_0_read_wave_sgprs(struct amdgpu_device *adev, uint32_t xcc_id, uint32_t simd,
4056 				     uint32_t wave, uint32_t start,
4057 				     uint32_t size, uint32_t *dst)
4058 {
4059 	wave_read_regs(
4060 		adev, simd, wave, 0,
4061 		start + SQIND_WAVE_SGPRS_OFFSET, size, dst);
4062 }
4063 
4064 static void gfx_v7_0_select_me_pipe_q(struct amdgpu_device *adev,
4065 				  u32 me, u32 pipe, u32 q, u32 vm, u32 xcc_id)
4066 {
4067 	cik_srbm_select(adev, me, pipe, q, vm);
4068 }
4069 
4070 /**
4071  * gfx_v7_0_get_hdp_flush_mask - get the reference and mask for HDP flush
4072  *
4073  * @ring: amdgpu_ring structure holding ring information
4074  * @ref_and_mask: pointer to store the reference and mask
4075  * @reg_mem_engine: pointer to store the register memory engine
4076  *
4077  * Calculates the reference and mask for HDP flush based on the ring type and me.
4078  */
4079 static void gfx_v7_0_get_hdp_flush_mask(struct amdgpu_ring *ring,
4080 					uint32_t *ref_and_mask, uint32_t *reg_mem_engine)
4081 {
4082 	if (!ring || !ref_and_mask || !reg_mem_engine) {
4083 		DRM_INFO("%s:invalid params\n", __func__);
4084 		return;
4085 	}
4086 
4087 	if (ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE ||
4088 		ring->funcs->type == AMDGPU_RING_TYPE_KIQ) {
4089 		switch (ring->me) {
4090 		case 1:
4091 			*ref_and_mask = GPU_HDP_FLUSH_DONE__CP2_MASK << ring->pipe;
4092 			break;
4093 		case 2:
4094 			*ref_and_mask = GPU_HDP_FLUSH_DONE__CP6_MASK << ring->pipe;
4095 			break;
4096 		default:
4097 			return;
4098 		}
4099 		*reg_mem_engine = 0;
4100 	} else {
4101 		*ref_and_mask = GPU_HDP_FLUSH_DONE__CP0_MASK;
4102 		*reg_mem_engine = 1;
4103 	}
4104 }
4105 
4106 static const struct amdgpu_gfx_funcs gfx_v7_0_gfx_funcs = {
4107 	.get_gpu_clock_counter = &gfx_v7_0_get_gpu_clock_counter,
4108 	.select_se_sh = &gfx_v7_0_select_se_sh,
4109 	.read_wave_data = &gfx_v7_0_read_wave_data,
4110 	.read_wave_sgprs = &gfx_v7_0_read_wave_sgprs,
4111 	.select_me_pipe_q = &gfx_v7_0_select_me_pipe_q,
4112 	.get_hdp_flush_mask = &gfx_v7_0_get_hdp_flush_mask,
4113 };
4114 
4115 static const struct amdgpu_rlc_funcs gfx_v7_0_rlc_funcs = {
4116 	.is_rlc_enabled = gfx_v7_0_is_rlc_enabled,
4117 	.set_safe_mode = gfx_v7_0_set_safe_mode,
4118 	.unset_safe_mode = gfx_v7_0_unset_safe_mode,
4119 	.init = gfx_v7_0_rlc_init,
4120 	.get_csb_size = gfx_v7_0_get_csb_size,
4121 	.get_csb_buffer = gfx_v7_0_get_csb_buffer,
4122 	.get_cp_table_num = gfx_v7_0_cp_pg_table_num,
4123 	.resume = gfx_v7_0_rlc_resume,
4124 	.stop = gfx_v7_0_rlc_stop,
4125 	.reset = gfx_v7_0_rlc_reset,
4126 	.start = gfx_v7_0_rlc_start,
4127 	.update_spm_vmid = gfx_v7_0_update_spm_vmid
4128 };
4129 
4130 static int gfx_v7_0_early_init(struct amdgpu_ip_block *ip_block)
4131 {
4132 	struct amdgpu_device *adev = ip_block->adev;
4133 
4134 	adev->gfx.xcc_mask = 1;
4135 	adev->gfx.num_gfx_rings = GFX7_NUM_GFX_RINGS;
4136 	adev->gfx.num_compute_rings = min(amdgpu_gfx_get_num_kcq(adev),
4137 					  AMDGPU_MAX_COMPUTE_RINGS);
4138 	adev->gfx.funcs = &gfx_v7_0_gfx_funcs;
4139 	adev->gfx.rlc.funcs = &gfx_v7_0_rlc_funcs;
4140 	gfx_v7_0_set_ring_funcs(adev);
4141 	gfx_v7_0_set_irq_funcs(adev);
4142 	gfx_v7_0_set_gds_init(adev);
4143 
4144 	return 0;
4145 }
4146 
4147 static int gfx_v7_0_late_init(struct amdgpu_ip_block *ip_block)
4148 {
4149 	struct amdgpu_device *adev = ip_block->adev;
4150 	int r;
4151 
4152 	r = amdgpu_irq_get(adev, &adev->gfx.priv_reg_irq, 0);
4153 	if (r)
4154 		return r;
4155 
4156 	r = amdgpu_irq_get(adev, &adev->gfx.priv_inst_irq, 0);
4157 	if (r)
4158 		return r;
4159 
4160 	return 0;
4161 }
4162 
4163 static void gfx_v7_0_gpu_early_init(struct amdgpu_device *adev)
4164 {
4165 	u32 gb_addr_config;
4166 	u32 mc_arb_ramcfg;
4167 	u32 dimm00_addr_map, dimm01_addr_map, dimm10_addr_map, dimm11_addr_map;
4168 	u32 tmp;
4169 
4170 	switch (adev->asic_type) {
4171 	case CHIP_BONAIRE:
4172 		adev->gfx.config.max_shader_engines = 2;
4173 		adev->gfx.config.max_tile_pipes = 4;
4174 		adev->gfx.config.max_cu_per_sh = 7;
4175 		adev->gfx.config.max_sh_per_se = 1;
4176 		adev->gfx.config.max_backends_per_se = 2;
4177 		adev->gfx.config.max_texture_channel_caches = 4;
4178 		adev->gfx.config.max_gprs = 256;
4179 		adev->gfx.config.max_gs_threads = 32;
4180 		adev->gfx.config.max_hw_contexts = 8;
4181 
4182 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
4183 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
4184 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
4185 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
4186 		gb_addr_config = BONAIRE_GB_ADDR_CONFIG_GOLDEN;
4187 		break;
4188 	case CHIP_HAWAII:
4189 		adev->gfx.config.max_shader_engines = 4;
4190 		adev->gfx.config.max_tile_pipes = 16;
4191 		adev->gfx.config.max_cu_per_sh = 11;
4192 		adev->gfx.config.max_sh_per_se = 1;
4193 		adev->gfx.config.max_backends_per_se = 4;
4194 		adev->gfx.config.max_texture_channel_caches = 16;
4195 		adev->gfx.config.max_gprs = 256;
4196 		adev->gfx.config.max_gs_threads = 32;
4197 		adev->gfx.config.max_hw_contexts = 8;
4198 
4199 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
4200 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
4201 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
4202 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
4203 		gb_addr_config = HAWAII_GB_ADDR_CONFIG_GOLDEN;
4204 		break;
4205 	case CHIP_KAVERI:
4206 		adev->gfx.config.max_shader_engines = 1;
4207 		adev->gfx.config.max_tile_pipes = 4;
4208 		adev->gfx.config.max_cu_per_sh = 8;
4209 		adev->gfx.config.max_backends_per_se = 2;
4210 		adev->gfx.config.max_sh_per_se = 1;
4211 		adev->gfx.config.max_texture_channel_caches = 4;
4212 		adev->gfx.config.max_gprs = 256;
4213 		adev->gfx.config.max_gs_threads = 16;
4214 		adev->gfx.config.max_hw_contexts = 8;
4215 
4216 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
4217 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
4218 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
4219 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
4220 		gb_addr_config = BONAIRE_GB_ADDR_CONFIG_GOLDEN;
4221 		break;
4222 	case CHIP_KABINI:
4223 	case CHIP_MULLINS:
4224 	default:
4225 		adev->gfx.config.max_shader_engines = 1;
4226 		adev->gfx.config.max_tile_pipes = 2;
4227 		adev->gfx.config.max_cu_per_sh = 2;
4228 		adev->gfx.config.max_sh_per_se = 1;
4229 		adev->gfx.config.max_backends_per_se = 1;
4230 		adev->gfx.config.max_texture_channel_caches = 2;
4231 		adev->gfx.config.max_gprs = 256;
4232 		adev->gfx.config.max_gs_threads = 16;
4233 		adev->gfx.config.max_hw_contexts = 8;
4234 
4235 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
4236 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
4237 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
4238 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
4239 		gb_addr_config = BONAIRE_GB_ADDR_CONFIG_GOLDEN;
4240 		break;
4241 	}
4242 
4243 	adev->gfx.config.mc_arb_ramcfg = RREG32(mmMC_ARB_RAMCFG);
4244 	mc_arb_ramcfg = adev->gfx.config.mc_arb_ramcfg;
4245 
4246 	adev->gfx.config.num_banks = REG_GET_FIELD(mc_arb_ramcfg,
4247 				MC_ARB_RAMCFG, NOOFBANK);
4248 	adev->gfx.config.num_ranks = REG_GET_FIELD(mc_arb_ramcfg,
4249 				MC_ARB_RAMCFG, NOOFRANKS);
4250 
4251 	adev->gfx.config.num_tile_pipes = adev->gfx.config.max_tile_pipes;
4252 	adev->gfx.config.mem_max_burst_length_bytes = 256;
4253 	if (adev->flags & AMD_IS_APU) {
4254 		/* Get memory bank mapping mode. */
4255 		tmp = RREG32(mmMC_FUS_DRAM0_BANK_ADDR_MAPPING);
4256 		dimm00_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM0_BANK_ADDR_MAPPING, DIMM0ADDRMAP);
4257 		dimm01_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM0_BANK_ADDR_MAPPING, DIMM1ADDRMAP);
4258 
4259 		tmp = RREG32(mmMC_FUS_DRAM1_BANK_ADDR_MAPPING);
4260 		dimm10_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM1_BANK_ADDR_MAPPING, DIMM0ADDRMAP);
4261 		dimm11_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM1_BANK_ADDR_MAPPING, DIMM1ADDRMAP);
4262 
4263 		/* Validate settings in case only one DIMM installed. */
4264 		if ((dimm00_addr_map == 0) || (dimm00_addr_map == 3) || (dimm00_addr_map == 4) || (dimm00_addr_map > 12))
4265 			dimm00_addr_map = 0;
4266 		if ((dimm01_addr_map == 0) || (dimm01_addr_map == 3) || (dimm01_addr_map == 4) || (dimm01_addr_map > 12))
4267 			dimm01_addr_map = 0;
4268 		if ((dimm10_addr_map == 0) || (dimm10_addr_map == 3) || (dimm10_addr_map == 4) || (dimm10_addr_map > 12))
4269 			dimm10_addr_map = 0;
4270 		if ((dimm11_addr_map == 0) || (dimm11_addr_map == 3) || (dimm11_addr_map == 4) || (dimm11_addr_map > 12))
4271 			dimm11_addr_map = 0;
4272 
4273 		/* If DIMM Addr map is 8GB, ROW size should be 2KB. Otherwise 1KB. */
4274 		/* If ROW size(DIMM1) != ROW size(DMIMM0), ROW size should be larger one. */
4275 		if ((dimm00_addr_map == 11) || (dimm01_addr_map == 11) || (dimm10_addr_map == 11) || (dimm11_addr_map == 11))
4276 			adev->gfx.config.mem_row_size_in_kb = 2;
4277 		else
4278 			adev->gfx.config.mem_row_size_in_kb = 1;
4279 	} else {
4280 		tmp = (mc_arb_ramcfg & MC_ARB_RAMCFG__NOOFCOLS_MASK) >> MC_ARB_RAMCFG__NOOFCOLS__SHIFT;
4281 		adev->gfx.config.mem_row_size_in_kb = (4 * (1 << (8 + tmp))) / 1024;
4282 		if (adev->gfx.config.mem_row_size_in_kb > 4)
4283 			adev->gfx.config.mem_row_size_in_kb = 4;
4284 	}
4285 	/* XXX use MC settings? */
4286 	adev->gfx.config.shader_engine_tile_size = 32;
4287 	adev->gfx.config.num_gpus = 1;
4288 	adev->gfx.config.multi_gpu_tile_size = 64;
4289 
4290 	/* fix up row size */
4291 	gb_addr_config &= ~GB_ADDR_CONFIG__ROW_SIZE_MASK;
4292 	switch (adev->gfx.config.mem_row_size_in_kb) {
4293 	case 1:
4294 	default:
4295 		gb_addr_config |= (0 << GB_ADDR_CONFIG__ROW_SIZE__SHIFT);
4296 		break;
4297 	case 2:
4298 		gb_addr_config |= (1 << GB_ADDR_CONFIG__ROW_SIZE__SHIFT);
4299 		break;
4300 	case 4:
4301 		gb_addr_config |= (2 << GB_ADDR_CONFIG__ROW_SIZE__SHIFT);
4302 		break;
4303 	}
4304 	adev->gfx.config.gb_addr_config = gb_addr_config;
4305 }
4306 
4307 static int gfx_v7_0_compute_ring_init(struct amdgpu_device *adev, int ring_id,
4308 					int mec, int pipe, int queue)
4309 {
4310 	int r;
4311 	unsigned irq_type;
4312 	struct amdgpu_ring *ring = &adev->gfx.compute_ring[ring_id];
4313 
4314 	/* mec0 is me1 */
4315 	ring->me = mec + 1;
4316 	ring->pipe = pipe;
4317 	ring->queue = queue;
4318 
4319 	ring->ring_obj = NULL;
4320 	ring->use_doorbell = true;
4321 	ring->doorbell_index = adev->doorbell_index.mec_ring0 + ring_id;
4322 	sprintf(ring->name, "comp_%d.%d.%d", ring->me, ring->pipe, ring->queue);
4323 
4324 	irq_type = AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP
4325 		+ ((ring->me - 1) * adev->gfx.mec.num_pipe_per_mec)
4326 		+ ring->pipe;
4327 
4328 	/* type-2 packets are deprecated on MEC, use type-3 instead */
4329 	r = amdgpu_ring_init(adev, ring, 1024,
4330 			     &adev->gfx.eop_irq, irq_type,
4331 			     AMDGPU_RING_PRIO_DEFAULT, NULL);
4332 	if (r)
4333 		return r;
4334 
4335 
4336 	return 0;
4337 }
4338 
4339 static int gfx_v7_0_sw_init(struct amdgpu_ip_block *ip_block)
4340 {
4341 	struct amdgpu_ring *ring;
4342 	struct amdgpu_device *adev = ip_block->adev;
4343 	int i, j, k, r, ring_id;
4344 
4345 	switch (adev->asic_type) {
4346 	case CHIP_KAVERI:
4347 		adev->gfx.mec.num_mec = 2;
4348 		break;
4349 	case CHIP_BONAIRE:
4350 	case CHIP_HAWAII:
4351 	case CHIP_KABINI:
4352 	case CHIP_MULLINS:
4353 	default:
4354 		adev->gfx.mec.num_mec = 1;
4355 		break;
4356 	}
4357 	adev->gfx.mec.num_pipe_per_mec = 4;
4358 	adev->gfx.mec.num_queue_per_pipe = 8;
4359 
4360 	/* EOP Event */
4361 	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, 181, &adev->gfx.eop_irq);
4362 	if (r)
4363 		return r;
4364 
4365 	/* Privileged reg */
4366 	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, 184,
4367 			      &adev->gfx.priv_reg_irq);
4368 	if (r)
4369 		return r;
4370 
4371 	/* Privileged inst */
4372 	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, 185,
4373 			      &adev->gfx.priv_inst_irq);
4374 	if (r)
4375 		return r;
4376 
4377 	r = gfx_v7_0_init_microcode(adev);
4378 	if (r) {
4379 		DRM_ERROR("Failed to load gfx firmware!\n");
4380 		return r;
4381 	}
4382 
4383 	r = adev->gfx.rlc.funcs->init(adev);
4384 	if (r) {
4385 		DRM_ERROR("Failed to init rlc BOs!\n");
4386 		return r;
4387 	}
4388 
4389 	/* allocate mec buffers */
4390 	r = gfx_v7_0_mec_init(adev);
4391 	if (r) {
4392 		DRM_ERROR("Failed to init MEC BOs!\n");
4393 		return r;
4394 	}
4395 
4396 	for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
4397 		ring = &adev->gfx.gfx_ring[i];
4398 		ring->ring_obj = NULL;
4399 		sprintf(ring->name, "gfx");
4400 		r = amdgpu_ring_init(adev, ring, 1024,
4401 				     &adev->gfx.eop_irq,
4402 				     AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP,
4403 				     AMDGPU_RING_PRIO_DEFAULT, NULL);
4404 		if (r)
4405 			return r;
4406 	}
4407 
4408 	/* set up the compute queues - allocate horizontally across pipes */
4409 	ring_id = 0;
4410 	for (i = 0; i < adev->gfx.mec.num_mec; ++i) {
4411 		for (j = 0; j < adev->gfx.mec.num_queue_per_pipe; j++) {
4412 			for (k = 0; k < adev->gfx.mec.num_pipe_per_mec; k++) {
4413 				if (!amdgpu_gfx_is_mec_queue_enabled(adev, 0, i,
4414 								     k, j))
4415 					continue;
4416 
4417 				r = gfx_v7_0_compute_ring_init(adev,
4418 								ring_id,
4419 								i, k, j);
4420 				if (r)
4421 					return r;
4422 
4423 				ring_id++;
4424 			}
4425 		}
4426 	}
4427 
4428 	/* create MQD for all compute queues */
4429 	r = amdgpu_gfx_mqd_sw_init(adev, sizeof(struct cik_mqd), 0);
4430 	if (r)
4431 		return r;
4432 
4433 	adev->gfx.ce_ram_size = 0x8000;
4434 
4435 	gfx_v7_0_gpu_early_init(adev);
4436 
4437 	adev->gfx.gfx_supported_reset =
4438 		amdgpu_get_soft_full_reset_mask(&adev->gfx.gfx_ring[0]);
4439 	adev->gfx.compute_supported_reset =
4440 		amdgpu_get_soft_full_reset_mask(&adev->gfx.compute_ring[0]);
4441 
4442 	if (!amdgpu_sriov_vf(adev) && !adev->debug_disable_ip_block_soft_reset) {
4443 		adev->gfx.compute_supported_reset |= AMDGPU_RESET_TYPE_IP_BLOCK_SOFT_RESET;
4444 		adev->gfx.gfx_supported_reset |= AMDGPU_RESET_TYPE_IP_BLOCK_SOFT_RESET;
4445 	}
4446 
4447 	return r;
4448 }
4449 
4450 static int gfx_v7_0_sw_fini(struct amdgpu_ip_block *ip_block)
4451 {
4452 	struct amdgpu_device *adev = ip_block->adev;
4453 	int i;
4454 
4455 	for (i = 0; i < adev->gfx.num_gfx_rings; i++)
4456 		amdgpu_ring_fini(&adev->gfx.gfx_ring[i]);
4457 	for (i = 0; i < adev->gfx.num_compute_rings; i++)
4458 		amdgpu_ring_fini(&adev->gfx.compute_ring[i]);
4459 
4460 	amdgpu_gfx_mqd_sw_fini(adev, 0);
4461 	amdgpu_gfx_rlc_fini(adev);
4462 	gfx_v7_0_mec_fini(adev);
4463 	amdgpu_bo_free_kernel(&adev->gfx.rlc.clear_state_obj,
4464 				&adev->gfx.rlc.clear_state_gpu_addr,
4465 				(void **)&adev->gfx.rlc.cs_ptr);
4466 	if (adev->gfx.rlc.cp_table_size) {
4467 		amdgpu_bo_free_kernel(&adev->gfx.rlc.cp_table_obj,
4468 				&adev->gfx.rlc.cp_table_gpu_addr,
4469 				(void **)&adev->gfx.rlc.cp_table_ptr);
4470 	}
4471 	gfx_v7_0_free_microcode(adev);
4472 
4473 	return 0;
4474 }
4475 
4476 static int gfx_v7_0_hw_init(struct amdgpu_ip_block *ip_block)
4477 {
4478 	int r;
4479 	struct amdgpu_device *adev = ip_block->adev;
4480 
4481 	gfx_v7_0_constants_init(adev);
4482 
4483 	/* init CSB */
4484 	adev->gfx.rlc.funcs->get_csb_buffer(adev, adev->gfx.rlc.cs_ptr);
4485 	/* init rlc */
4486 	r = adev->gfx.rlc.funcs->resume(adev);
4487 	if (r)
4488 		return r;
4489 
4490 	r = gfx_v7_0_cp_resume(adev);
4491 	if (r)
4492 		return r;
4493 
4494 	return r;
4495 }
4496 
4497 static int gfx_v7_0_hw_fini(struct amdgpu_ip_block *ip_block)
4498 {
4499 	struct amdgpu_device *adev = ip_block->adev;
4500 
4501 	amdgpu_irq_put(adev, &adev->gfx.priv_reg_irq, 0);
4502 	amdgpu_irq_put(adev, &adev->gfx.priv_inst_irq, 0);
4503 	gfx_v7_0_cp_enable(adev, false);
4504 	adev->gfx.rlc.funcs->stop(adev);
4505 	gfx_v7_0_fini_pg(adev);
4506 
4507 	return 0;
4508 }
4509 
4510 static int gfx_v7_0_suspend(struct amdgpu_ip_block *ip_block)
4511 {
4512 	return gfx_v7_0_hw_fini(ip_block);
4513 }
4514 
4515 static int gfx_v7_0_resume(struct amdgpu_ip_block *ip_block)
4516 {
4517 	return gfx_v7_0_hw_init(ip_block);
4518 }
4519 
4520 static int gfx_v7_0_wait_for_idle(struct amdgpu_ip_block *ip_block)
4521 {
4522 	unsigned i;
4523 	u32 tmp;
4524 	struct amdgpu_device *adev = ip_block->adev;
4525 
4526 	for (i = 0; i < adev->usec_timeout; i++) {
4527 		/* read MC_STATUS */
4528 		tmp = RREG32(mmGRBM_STATUS) & GRBM_STATUS__GUI_ACTIVE_MASK;
4529 
4530 		if (!tmp)
4531 			return 0;
4532 		udelay(1);
4533 	}
4534 	return -ETIMEDOUT;
4535 }
4536 
4537 static int gfx_v7_0_soft_reset(struct amdgpu_ip_block *ip_block)
4538 {
4539 	struct amdgpu_device *adev = ip_block->adev;
4540 	u32 grbm_soft_reset = 0, srbm_soft_reset = 0;
4541 	u32 tmp;
4542 	int i;
4543 	int r;
4544 
4545 	grbm_soft_reset =
4546 		REG_SET_FIELD(0, GRBM_SOFT_RESET, SOFT_RESET_RLC, 1) |
4547 		REG_SET_FIELD(0, GRBM_SOFT_RESET, SOFT_RESET_GFX, 1) |
4548 		REG_SET_FIELD(0, GRBM_SOFT_RESET, SOFT_RESET_CP, 1) |
4549 		REG_SET_FIELD(0, GRBM_SOFT_RESET, SOFT_RESET_CPF, 1) |
4550 		REG_SET_FIELD(0, GRBM_SOFT_RESET, SOFT_RESET_CPC, 1) |
4551 		REG_SET_FIELD(0, GRBM_SOFT_RESET, SOFT_RESET_CPG, 1);
4552 
4553 	srbm_soft_reset =
4554 		REG_SET_FIELD(0, SRBM_SOFT_RESET, SOFT_RESET_GRBM, 1) |
4555 		REG_SET_FIELD(0, SRBM_SOFT_RESET, SOFT_RESET_SEM, 1);
4556 
4557 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
4558 		struct amdgpu_ring *ring = &adev->gfx.compute_ring[i];
4559 
4560 		mutex_lock(&adev->srbm_mutex);
4561 		cik_srbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
4562 		gfx_v7_0_deactivate_hqd(adev, 2);
4563 		cik_srbm_select(adev, 0, 0, 0, 0);
4564 		mutex_unlock(&adev->srbm_mutex);
4565 
4566 		udelay(50);
4567 	}
4568 
4569 	ip_block->version->funcs->set_clockgating_state(ip_block, AMD_CG_STATE_UNGATE);
4570 	ip_block->version->funcs->set_powergating_state(ip_block, AMD_PG_STATE_UNGATE);
4571 	ip_block->version->funcs->suspend(ip_block);
4572 
4573 	if (grbm_soft_reset || srbm_soft_reset) {
4574 		tmp = RREG32(mmGMCON_DEBUG);
4575 		tmp = REG_SET_FIELD(tmp, GMCON_DEBUG, GFX_STALL, 1);
4576 		tmp = REG_SET_FIELD(tmp, GMCON_DEBUG, GFX_CLEAR, 1);
4577 		WREG32(mmGMCON_DEBUG, tmp);
4578 
4579 		udelay(100);
4580 	}
4581 
4582 	if (grbm_soft_reset) {
4583 		tmp = RREG32(mmGRBM_SOFT_RESET);
4584 		tmp |= grbm_soft_reset;
4585 		dev_info(adev->dev, "GRBM_SOFT_RESET=0x%08X\n", tmp);
4586 		WREG32(mmGRBM_SOFT_RESET, tmp);
4587 		tmp = RREG32(mmGRBM_SOFT_RESET);
4588 
4589 		udelay(100);
4590 
4591 		tmp &= ~grbm_soft_reset;
4592 		WREG32(mmGRBM_SOFT_RESET, tmp);
4593 		tmp = RREG32(mmGRBM_SOFT_RESET);
4594 
4595 		udelay(100);
4596 	}
4597 
4598 	if (srbm_soft_reset) {
4599 		tmp = RREG32(mmSRBM_SOFT_RESET);
4600 		tmp |= srbm_soft_reset;
4601 		dev_info(adev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp);
4602 		WREG32(mmSRBM_SOFT_RESET, tmp);
4603 		tmp = RREG32(mmSRBM_SOFT_RESET);
4604 
4605 		udelay(100);
4606 
4607 		tmp &= ~srbm_soft_reset;
4608 		WREG32(mmSRBM_SOFT_RESET, tmp);
4609 		tmp = RREG32(mmSRBM_SOFT_RESET);
4610 
4611 		udelay(100);
4612 	}
4613 
4614 	if (grbm_soft_reset || srbm_soft_reset) {
4615 		tmp = RREG32(mmGMCON_DEBUG);
4616 		tmp = REG_SET_FIELD(tmp, GMCON_DEBUG, GFX_STALL, 0);
4617 		tmp = REG_SET_FIELD(tmp, GMCON_DEBUG, GFX_CLEAR, 0);
4618 		WREG32(mmGMCON_DEBUG, tmp);
4619 	}
4620 
4621 	/* Wait a little for things to settle down */
4622 	udelay(100);
4623 
4624 	r = ip_block->version->funcs->resume(ip_block);
4625 	r |= ip_block->version->funcs->late_init(ip_block);
4626 	if (r)
4627 		return r;
4628 
4629 	ip_block->version->funcs->set_clockgating_state(ip_block, AMD_CG_STATE_GATE);
4630 	ip_block->version->funcs->set_powergating_state(ip_block, AMD_PG_STATE_GATE);
4631 
4632 	return 0;
4633 }
4634 
4635 static void gfx_v7_0_set_gfx_eop_interrupt_state(struct amdgpu_device *adev,
4636 						 enum amdgpu_interrupt_state state)
4637 {
4638 	u32 cp_int_cntl;
4639 
4640 	switch (state) {
4641 	case AMDGPU_IRQ_STATE_DISABLE:
4642 		cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0);
4643 		cp_int_cntl &= ~CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK;
4644 		WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl);
4645 		break;
4646 	case AMDGPU_IRQ_STATE_ENABLE:
4647 		cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0);
4648 		cp_int_cntl |= CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK;
4649 		WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl);
4650 		break;
4651 	default:
4652 		break;
4653 	}
4654 }
4655 
4656 static void gfx_v7_0_set_compute_eop_interrupt_state(struct amdgpu_device *adev,
4657 						     int me, int pipe,
4658 						     enum amdgpu_interrupt_state state)
4659 {
4660 	u32 mec_int_cntl, mec_int_cntl_reg;
4661 
4662 	/*
4663 	 * amdgpu controls only the first MEC. That's why this function only
4664 	 * handles the setting of interrupts for this specific MEC. All other
4665 	 * pipes' interrupts are set by amdkfd.
4666 	 */
4667 
4668 	if (me == 1) {
4669 		switch (pipe) {
4670 		case 0:
4671 			mec_int_cntl_reg = mmCP_ME1_PIPE0_INT_CNTL;
4672 			break;
4673 		case 1:
4674 			mec_int_cntl_reg = mmCP_ME1_PIPE1_INT_CNTL;
4675 			break;
4676 		case 2:
4677 			mec_int_cntl_reg = mmCP_ME1_PIPE2_INT_CNTL;
4678 			break;
4679 		case 3:
4680 			mec_int_cntl_reg = mmCP_ME1_PIPE3_INT_CNTL;
4681 			break;
4682 		default:
4683 			DRM_DEBUG("invalid pipe %d\n", pipe);
4684 			return;
4685 		}
4686 	} else {
4687 		DRM_DEBUG("invalid me %d\n", me);
4688 		return;
4689 	}
4690 
4691 	switch (state) {
4692 	case AMDGPU_IRQ_STATE_DISABLE:
4693 		mec_int_cntl = RREG32(mec_int_cntl_reg);
4694 		mec_int_cntl &= ~CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK;
4695 		WREG32(mec_int_cntl_reg, mec_int_cntl);
4696 		break;
4697 	case AMDGPU_IRQ_STATE_ENABLE:
4698 		mec_int_cntl = RREG32(mec_int_cntl_reg);
4699 		mec_int_cntl |= CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK;
4700 		WREG32(mec_int_cntl_reg, mec_int_cntl);
4701 		break;
4702 	default:
4703 		break;
4704 	}
4705 }
4706 
4707 static int gfx_v7_0_set_priv_reg_fault_state(struct amdgpu_device *adev,
4708 					     struct amdgpu_irq_src *src,
4709 					     unsigned type,
4710 					     enum amdgpu_interrupt_state state)
4711 {
4712 	u32 cp_int_cntl;
4713 
4714 	switch (state) {
4715 	case AMDGPU_IRQ_STATE_DISABLE:
4716 		cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0);
4717 		cp_int_cntl &= ~CP_INT_CNTL_RING0__PRIV_REG_INT_ENABLE_MASK;
4718 		WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl);
4719 		break;
4720 	case AMDGPU_IRQ_STATE_ENABLE:
4721 		cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0);
4722 		cp_int_cntl |= CP_INT_CNTL_RING0__PRIV_REG_INT_ENABLE_MASK;
4723 		WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl);
4724 		break;
4725 	default:
4726 		break;
4727 	}
4728 
4729 	return 0;
4730 }
4731 
4732 static int gfx_v7_0_set_priv_inst_fault_state(struct amdgpu_device *adev,
4733 					      struct amdgpu_irq_src *src,
4734 					      unsigned type,
4735 					      enum amdgpu_interrupt_state state)
4736 {
4737 	u32 cp_int_cntl;
4738 
4739 	switch (state) {
4740 	case AMDGPU_IRQ_STATE_DISABLE:
4741 		cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0);
4742 		cp_int_cntl &= ~CP_INT_CNTL_RING0__PRIV_INSTR_INT_ENABLE_MASK;
4743 		WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl);
4744 		break;
4745 	case AMDGPU_IRQ_STATE_ENABLE:
4746 		cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0);
4747 		cp_int_cntl |= CP_INT_CNTL_RING0__PRIV_INSTR_INT_ENABLE_MASK;
4748 		WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl);
4749 		break;
4750 	default:
4751 		break;
4752 	}
4753 
4754 	return 0;
4755 }
4756 
4757 static int gfx_v7_0_set_eop_interrupt_state(struct amdgpu_device *adev,
4758 					    struct amdgpu_irq_src *src,
4759 					    unsigned type,
4760 					    enum amdgpu_interrupt_state state)
4761 {
4762 	switch (type) {
4763 	case AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP:
4764 		gfx_v7_0_set_gfx_eop_interrupt_state(adev, state);
4765 		break;
4766 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP:
4767 		gfx_v7_0_set_compute_eop_interrupt_state(adev, 1, 0, state);
4768 		break;
4769 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE1_EOP:
4770 		gfx_v7_0_set_compute_eop_interrupt_state(adev, 1, 1, state);
4771 		break;
4772 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE2_EOP:
4773 		gfx_v7_0_set_compute_eop_interrupt_state(adev, 1, 2, state);
4774 		break;
4775 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE3_EOP:
4776 		gfx_v7_0_set_compute_eop_interrupt_state(adev, 1, 3, state);
4777 		break;
4778 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE0_EOP:
4779 		gfx_v7_0_set_compute_eop_interrupt_state(adev, 2, 0, state);
4780 		break;
4781 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE1_EOP:
4782 		gfx_v7_0_set_compute_eop_interrupt_state(adev, 2, 1, state);
4783 		break;
4784 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE2_EOP:
4785 		gfx_v7_0_set_compute_eop_interrupt_state(adev, 2, 2, state);
4786 		break;
4787 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE3_EOP:
4788 		gfx_v7_0_set_compute_eop_interrupt_state(adev, 2, 3, state);
4789 		break;
4790 	default:
4791 		break;
4792 	}
4793 	return 0;
4794 }
4795 
4796 static int gfx_v7_0_eop_irq(struct amdgpu_device *adev,
4797 			    struct amdgpu_irq_src *source,
4798 			    struct amdgpu_iv_entry *entry)
4799 {
4800 	u8 me_id, pipe_id;
4801 	struct amdgpu_ring *ring;
4802 	int i;
4803 
4804 	DRM_DEBUG("IH: CP EOP\n");
4805 	me_id = (entry->ring_id & 0x0c) >> 2;
4806 	pipe_id = (entry->ring_id & 0x03) >> 0;
4807 	switch (me_id) {
4808 	case 0:
4809 		amdgpu_fence_process(&adev->gfx.gfx_ring[0]);
4810 		break;
4811 	case 1:
4812 	case 2:
4813 		for (i = 0; i < adev->gfx.num_compute_rings; i++) {
4814 			ring = &adev->gfx.compute_ring[i];
4815 			if ((ring->me == me_id) && (ring->pipe == pipe_id))
4816 				amdgpu_fence_process(ring);
4817 		}
4818 		break;
4819 	}
4820 	return 0;
4821 }
4822 
4823 static void gfx_v7_0_fault(struct amdgpu_device *adev,
4824 			   struct amdgpu_iv_entry *entry)
4825 {
4826 	struct amdgpu_ring *ring;
4827 	u8 me_id, pipe_id;
4828 	int i;
4829 
4830 	me_id = (entry->ring_id & 0x0c) >> 2;
4831 	pipe_id = (entry->ring_id & 0x03) >> 0;
4832 	switch (me_id) {
4833 	case 0:
4834 		drm_sched_fault(&adev->gfx.gfx_ring[0].sched);
4835 		break;
4836 	case 1:
4837 	case 2:
4838 		for (i = 0; i < adev->gfx.num_compute_rings; i++) {
4839 			ring = &adev->gfx.compute_ring[i];
4840 			if ((ring->me == me_id) && (ring->pipe == pipe_id))
4841 				drm_sched_fault(&ring->sched);
4842 		}
4843 		break;
4844 	}
4845 }
4846 
4847 static int gfx_v7_0_priv_reg_irq(struct amdgpu_device *adev,
4848 				 struct amdgpu_irq_src *source,
4849 				 struct amdgpu_iv_entry *entry)
4850 {
4851 	DRM_ERROR("Illegal register access in command stream\n");
4852 	gfx_v7_0_fault(adev, entry);
4853 	return 0;
4854 }
4855 
4856 static int gfx_v7_0_priv_inst_irq(struct amdgpu_device *adev,
4857 				  struct amdgpu_irq_src *source,
4858 				  struct amdgpu_iv_entry *entry)
4859 {
4860 	DRM_ERROR("Illegal instruction in command stream\n");
4861 	// XXX soft reset the gfx block only
4862 	gfx_v7_0_fault(adev, entry);
4863 	return 0;
4864 }
4865 
4866 static int gfx_v7_0_set_clockgating_state(struct amdgpu_ip_block *ip_block,
4867 					  enum amd_clockgating_state state)
4868 {
4869 	bool gate = false;
4870 	struct amdgpu_device *adev = ip_block->adev;
4871 
4872 	if (state == AMD_CG_STATE_GATE)
4873 		gate = true;
4874 
4875 	gfx_v7_0_enable_gui_idle_interrupt(adev, false);
4876 	/* order matters! */
4877 	if (gate) {
4878 		gfx_v7_0_enable_mgcg(adev, true);
4879 		gfx_v7_0_enable_cgcg(adev, true);
4880 	} else {
4881 		gfx_v7_0_enable_cgcg(adev, false);
4882 		gfx_v7_0_enable_mgcg(adev, false);
4883 	}
4884 	gfx_v7_0_enable_gui_idle_interrupt(adev, true);
4885 
4886 	return 0;
4887 }
4888 
4889 static int gfx_v7_0_set_powergating_state(struct amdgpu_ip_block *ip_block,
4890 					  enum amd_powergating_state state)
4891 {
4892 	bool gate = false;
4893 	struct amdgpu_device *adev = ip_block->adev;
4894 
4895 	if (state == AMD_PG_STATE_GATE)
4896 		gate = true;
4897 
4898 	if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_PG |
4899 			      AMD_PG_SUPPORT_GFX_SMG |
4900 			      AMD_PG_SUPPORT_GFX_DMG |
4901 			      AMD_PG_SUPPORT_CP |
4902 			      AMD_PG_SUPPORT_GDS |
4903 			      AMD_PG_SUPPORT_RLC_SMU_HS)) {
4904 		gfx_v7_0_update_gfx_pg(adev, gate);
4905 		if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PG) {
4906 			gfx_v7_0_enable_cp_pg(adev, gate);
4907 			gfx_v7_0_enable_gds_pg(adev, gate);
4908 		}
4909 	}
4910 
4911 	return 0;
4912 }
4913 
4914 static void gfx_v7_0_emit_mem_sync(struct amdgpu_ring *ring)
4915 {
4916 	amdgpu_ring_write(ring, PACKET3(PACKET3_SURFACE_SYNC, 3));
4917 	amdgpu_ring_write(ring, PACKET3_TCL1_ACTION_ENA |
4918 			  PACKET3_TC_ACTION_ENA |
4919 			  PACKET3_SH_KCACHE_ACTION_ENA |
4920 			  PACKET3_SH_ICACHE_ACTION_ENA);  /* CP_COHER_CNTL */
4921 	amdgpu_ring_write(ring, 0xffffffff);  /* CP_COHER_SIZE */
4922 	amdgpu_ring_write(ring, 0);  /* CP_COHER_BASE */
4923 	amdgpu_ring_write(ring, 0x0000000A); /* poll interval */
4924 }
4925 
4926 static void gfx_v7_0_emit_mem_sync_compute(struct amdgpu_ring *ring)
4927 {
4928 	amdgpu_ring_write(ring, PACKET3(PACKET3_ACQUIRE_MEM, 5));
4929 	amdgpu_ring_write(ring, PACKET3_TCL1_ACTION_ENA |
4930 			  PACKET3_TC_ACTION_ENA |
4931 			  PACKET3_SH_KCACHE_ACTION_ENA |
4932 			  PACKET3_SH_ICACHE_ACTION_ENA);  /* CP_COHER_CNTL */
4933 	amdgpu_ring_write(ring, 0xffffffff);	/* CP_COHER_SIZE */
4934 	amdgpu_ring_write(ring, 0xff);		/* CP_COHER_SIZE_HI */
4935 	amdgpu_ring_write(ring, 0);		/* CP_COHER_BASE */
4936 	amdgpu_ring_write(ring, 0);		/* CP_COHER_BASE_HI */
4937 	amdgpu_ring_write(ring, 0x0000000A);	/* poll interval */
4938 }
4939 
4940 static const struct amd_ip_funcs gfx_v7_0_ip_funcs = {
4941 	.name = "gfx_v7_0",
4942 	.early_init = gfx_v7_0_early_init,
4943 	.late_init = gfx_v7_0_late_init,
4944 	.sw_init = gfx_v7_0_sw_init,
4945 	.sw_fini = gfx_v7_0_sw_fini,
4946 	.hw_init = gfx_v7_0_hw_init,
4947 	.hw_fini = gfx_v7_0_hw_fini,
4948 	.suspend = gfx_v7_0_suspend,
4949 	.resume = gfx_v7_0_resume,
4950 	.wait_for_idle = gfx_v7_0_wait_for_idle,
4951 	.soft_reset = gfx_v7_0_soft_reset,
4952 	.set_clockgating_state = gfx_v7_0_set_clockgating_state,
4953 	.set_powergating_state = gfx_v7_0_set_powergating_state,
4954 };
4955 
4956 static const struct amdgpu_ring_funcs gfx_v7_0_ring_funcs_gfx = {
4957 	.type = AMDGPU_RING_TYPE_GFX,
4958 	.align_mask = 0xff,
4959 	.nop = PACKET3(PACKET3_NOP, 0x3FFF),
4960 	.support_64bit_ptrs = false,
4961 	.get_rptr = gfx_v7_0_ring_get_rptr,
4962 	.get_wptr = gfx_v7_0_ring_get_wptr_gfx,
4963 	.set_wptr = gfx_v7_0_ring_set_wptr_gfx,
4964 	.emit_frame_size =
4965 		5 + /* gfx_v7_0_ring_emit_init_cond_exec (from amdgpu_ib_schedule) */
4966 		5 + /* gfx_v7_0_ring_emit_init_cond_exec (from amdgpu_vm_flush) */
4967 		20 + /* gfx_v7_0_ring_emit_gds_switch */
4968 		7 + /* gfx_v7_0_ring_emit_hdp_flush */
4969 		5 + /* hdp invalidate */
4970 		12 + 12 + 12 + /* gfx_v7_0_ring_emit_fence_gfx x3 for user fence, vm fence */
4971 		7 + /* gfx_v7_0_ring_emit_pipeline_sync */
4972 		CIK_FLUSH_GPU_TLB_NUM_WREG * 5 + 7 + 2 + /* gfx_v7_0_ring_emit_vm_flush */
4973 		3 * 2 + /* gfx_v7_0_ring_emit_sb x3 (from amdgpu_vm_flush, amdgpu_ib_schedule) */
4974 		3 + 4 + /* gfx_v7_ring_emit_cntxcntl including vgt flush*/
4975 		5, /* SURFACE_SYNC */
4976 	.emit_ib_size = 4, /* gfx_v7_0_ring_emit_ib_gfx */
4977 	.emit_ib = gfx_v7_0_ring_emit_ib_gfx,
4978 	.emit_fence = gfx_v7_0_ring_emit_fence_gfx,
4979 	.emit_pipeline_sync = gfx_v7_0_ring_emit_pipeline_sync,
4980 	.emit_vm_flush = gfx_v7_0_ring_emit_vm_flush,
4981 	.emit_gds_switch = gfx_v7_0_ring_emit_gds_switch,
4982 	.emit_hdp_flush = gfx_v7_0_ring_emit_hdp_flush,
4983 	.test_ring = gfx_v7_0_ring_test_ring,
4984 	.test_ib = gfx_v7_0_ring_test_ib,
4985 	.insert_nop = amdgpu_ring_insert_nop,
4986 	.pad_ib = amdgpu_ring_generic_pad_ib,
4987 	.emit_switch_buffer = gfx_v7_0_ring_emit_sb,
4988 	.emit_cntxcntl = gfx_v7_ring_emit_cntxcntl,
4989 	.init_cond_exec = gfx_v7_0_ring_emit_init_cond_exec,
4990 	.emit_wreg = gfx_v7_0_ring_emit_wreg,
4991 	.soft_recovery = gfx_v7_0_ring_soft_recovery,
4992 	.emit_mem_sync = gfx_v7_0_emit_mem_sync,
4993 };
4994 
4995 static const struct amdgpu_ring_funcs gfx_v7_0_ring_funcs_compute = {
4996 	.type = AMDGPU_RING_TYPE_COMPUTE,
4997 	.align_mask = 0xff,
4998 	.nop = PACKET3(PACKET3_NOP, 0x3FFF),
4999 	.support_64bit_ptrs = false,
5000 	.get_rptr = gfx_v7_0_ring_get_rptr,
5001 	.get_wptr = gfx_v7_0_ring_get_wptr_compute,
5002 	.set_wptr = gfx_v7_0_ring_set_wptr_compute,
5003 	.emit_frame_size =
5004 		5 + /* gfx_v7_0_ring_emit_init_cond_exec (from amdgpu_ib_schedule) */
5005 		5 + /* gfx_v7_0_ring_emit_init_cond_exec (from amdgpu_vm_flush) */
5006 		20 + /* gfx_v7_0_ring_emit_gds_switch */
5007 		7 + /* gfx_v7_0_ring_emit_hdp_flush */
5008 		5 + /* hdp invalidate */
5009 		7 + /* gfx_v7_0_ring_emit_pipeline_sync */
5010 		CIK_FLUSH_GPU_TLB_NUM_WREG * 5 + 7 + /* gfx_v7_0_ring_emit_vm_flush */
5011 		7 + 7 + 7 + /* gfx_v7_0_ring_emit_fence_compute x3 for user fence, vm fence */
5012 		7, /* gfx_v7_0_emit_mem_sync_compute */
5013 	.emit_ib_size =	7, /* gfx_v7_0_ring_emit_ib_compute */
5014 	.emit_ib = gfx_v7_0_ring_emit_ib_compute,
5015 	.emit_fence = gfx_v7_0_ring_emit_fence_compute,
5016 	.emit_pipeline_sync = gfx_v7_0_ring_emit_pipeline_sync,
5017 	.emit_vm_flush = gfx_v7_0_ring_emit_vm_flush,
5018 	.emit_gds_switch = gfx_v7_0_ring_emit_gds_switch,
5019 	.emit_hdp_flush = gfx_v7_0_ring_emit_hdp_flush,
5020 	.test_ring = gfx_v7_0_ring_test_ring,
5021 	.test_ib = gfx_v7_0_ring_test_ib,
5022 	.insert_nop = amdgpu_ring_insert_nop,
5023 	.pad_ib = amdgpu_ring_generic_pad_ib,
5024 	.init_cond_exec = gfx_v7_0_ring_emit_init_cond_exec,
5025 	.emit_wreg = gfx_v7_0_ring_emit_wreg,
5026 	.soft_recovery = gfx_v7_0_ring_soft_recovery,
5027 	.emit_mem_sync = gfx_v7_0_emit_mem_sync_compute,
5028 };
5029 
5030 static void gfx_v7_0_set_ring_funcs(struct amdgpu_device *adev)
5031 {
5032 	int i;
5033 
5034 	for (i = 0; i < adev->gfx.num_gfx_rings; i++)
5035 		adev->gfx.gfx_ring[i].funcs = &gfx_v7_0_ring_funcs_gfx;
5036 	for (i = 0; i < adev->gfx.num_compute_rings; i++)
5037 		adev->gfx.compute_ring[i].funcs = &gfx_v7_0_ring_funcs_compute;
5038 }
5039 
5040 static const struct amdgpu_irq_src_funcs gfx_v7_0_eop_irq_funcs = {
5041 	.set = gfx_v7_0_set_eop_interrupt_state,
5042 	.process = gfx_v7_0_eop_irq,
5043 };
5044 
5045 static const struct amdgpu_irq_src_funcs gfx_v7_0_priv_reg_irq_funcs = {
5046 	.set = gfx_v7_0_set_priv_reg_fault_state,
5047 	.process = gfx_v7_0_priv_reg_irq,
5048 };
5049 
5050 static const struct amdgpu_irq_src_funcs gfx_v7_0_priv_inst_irq_funcs = {
5051 	.set = gfx_v7_0_set_priv_inst_fault_state,
5052 	.process = gfx_v7_0_priv_inst_irq,
5053 };
5054 
5055 static void gfx_v7_0_set_irq_funcs(struct amdgpu_device *adev)
5056 {
5057 	adev->gfx.eop_irq.num_types = AMDGPU_CP_IRQ_LAST;
5058 	adev->gfx.eop_irq.funcs = &gfx_v7_0_eop_irq_funcs;
5059 
5060 	adev->gfx.priv_reg_irq.num_types = 1;
5061 	adev->gfx.priv_reg_irq.funcs = &gfx_v7_0_priv_reg_irq_funcs;
5062 
5063 	adev->gfx.priv_inst_irq.num_types = 1;
5064 	adev->gfx.priv_inst_irq.funcs = &gfx_v7_0_priv_inst_irq_funcs;
5065 }
5066 
5067 static void gfx_v7_0_set_gds_init(struct amdgpu_device *adev)
5068 {
5069 	/* init asci gds info */
5070 	adev->gds.gds_size = RREG32(mmGDS_VMID0_SIZE);
5071 	adev->gds.gws_size = 64;
5072 	adev->gds.oa_size = 16;
5073 	adev->gds.gds_compute_max_wave_id = RREG32(mmGDS_COMPUTE_MAX_WAVE_ID);
5074 }
5075 
5076 
5077 static void gfx_v7_0_get_cu_info(struct amdgpu_device *adev)
5078 {
5079 	int i, j, k, counter, active_cu_number = 0;
5080 	u32 mask, bitmap, ao_bitmap, ao_cu_mask = 0;
5081 	struct amdgpu_cu_info *cu_info = &adev->gfx.cu_info;
5082 	unsigned disable_masks[4 * 2];
5083 	u32 ao_cu_num;
5084 
5085 	if (adev->flags & AMD_IS_APU)
5086 		ao_cu_num = 2;
5087 	else
5088 		ao_cu_num = adev->gfx.config.max_cu_per_sh;
5089 
5090 	memset(cu_info, 0, sizeof(*cu_info));
5091 
5092 	amdgpu_gfx_parse_disable_cu(adev, disable_masks, 4, 2);
5093 
5094 	mutex_lock(&adev->grbm_idx_mutex);
5095 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
5096 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
5097 			mask = 1;
5098 			ao_bitmap = 0;
5099 			counter = 0;
5100 			gfx_v7_0_select_se_sh(adev, i, j, 0xffffffff, 0);
5101 			if (i < 4 && j < 2)
5102 				gfx_v7_0_set_user_cu_inactive_bitmap(
5103 					adev, disable_masks[i * 2 + j]);
5104 			bitmap = gfx_v7_0_get_cu_active_bitmap(adev);
5105 			cu_info->bitmap[0][i][j] = bitmap;
5106 
5107 			for (k = 0; k < adev->gfx.config.max_cu_per_sh; k++) {
5108 				if (bitmap & mask) {
5109 					if (counter < ao_cu_num)
5110 						ao_bitmap |= mask;
5111 					counter++;
5112 				}
5113 				mask <<= 1;
5114 			}
5115 			active_cu_number += counter;
5116 			if (i < 2 && j < 2)
5117 				ao_cu_mask |= (ao_bitmap << (i * 16 + j * 8));
5118 			cu_info->ao_cu_bitmap[i][j] = ao_bitmap;
5119 		}
5120 	}
5121 	gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0);
5122 	mutex_unlock(&adev->grbm_idx_mutex);
5123 
5124 	cu_info->number = active_cu_number;
5125 	cu_info->ao_cu_mask = ao_cu_mask;
5126 	cu_info->simd_per_cu = NUM_SIMD_PER_CU;
5127 	cu_info->max_waves_per_simd = 10;
5128 	cu_info->max_scratch_slots_per_cu = 32;
5129 	cu_info->wave_front_size = 64;
5130 	cu_info->lds_size = 64;
5131 }
5132 
5133 const struct amdgpu_ip_block_version gfx_v7_1_ip_block = {
5134 	.type = AMD_IP_BLOCK_TYPE_GFX,
5135 	.major = 7,
5136 	.minor = 1,
5137 	.rev = 0,
5138 	.funcs = &gfx_v7_0_ip_funcs,
5139 };
5140 
5141 const struct amdgpu_ip_block_version gfx_v7_2_ip_block = {
5142 	.type = AMD_IP_BLOCK_TYPE_GFX,
5143 	.major = 7,
5144 	.minor = 2,
5145 	.rev = 0,
5146 	.funcs = &gfx_v7_0_ip_funcs,
5147 };
5148 
5149 const struct amdgpu_ip_block_version gfx_v7_3_ip_block = {
5150 	.type = AMD_IP_BLOCK_TYPE_GFX,
5151 	.major = 7,
5152 	.minor = 3,
5153 	.rev = 0,
5154 	.funcs = &gfx_v7_0_ip_funcs,
5155 };
5156